Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/slab.c | |
3 | * Written by Mark Hemment, 1996/97. | |
4 | * (markhe@nextd.demon.co.uk) | |
5 | * | |
6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
7 | * | |
8 | * Major cleanup, different bufctl logic, per-cpu arrays | |
9 | * (c) 2000 Manfred Spraul | |
10 | * | |
11 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
12 | * (c) 2002 Manfred Spraul | |
13 | * | |
14 | * An implementation of the Slab Allocator as described in outline in; | |
15 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
16 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
17 | * or with a little more detail in; | |
18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
19 | * Jeff Bonwick (Sun Microsystems). | |
20 | * Presented at: USENIX Summer 1994 Technical Conference | |
21 | * | |
22 | * The memory is organized in caches, one cache for each object type. | |
23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
24 | * Each cache consists out of many slabs (they are small (usually one | |
25 | * page long) and always contiguous), and each slab contains multiple | |
26 | * initialized objects. | |
27 | * | |
28 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 29 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
30 | * kmem_cache_free. |
31 | * | |
32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
33 | * normal). If you need a special memory type, then must create a new | |
34 | * cache for that memory type. | |
35 | * | |
36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
37 | * full slabs with 0 free objects | |
38 | * partial slabs | |
39 | * empty slabs with no allocated objects | |
40 | * | |
41 | * If partial slabs exist, then new allocations come from these slabs, | |
42 | * otherwise from empty slabs or new slabs are allocated. | |
43 | * | |
44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
46 | * | |
47 | * Each cache has a short per-cpu head array, most allocs | |
48 | * and frees go into that array, and if that array overflows, then 1/2 | |
49 | * of the entries in the array are given back into the global cache. | |
50 | * The head array is strictly LIFO and should improve the cache hit rates. | |
51 | * On SMP, it additionally reduces the spinlock operations. | |
52 | * | |
a737b3e2 | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
54 | * it's changed with a smp_call_function(). |
55 | * | |
56 | * SMP synchronization: | |
57 | * constructors and destructors are called without any locking. | |
343e0d7a | 58 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
61 | * and local interrupts are disabled so slab code is preempt-safe. | |
62 | * The non-constant members are protected with a per-cache irq spinlock. | |
63 | * | |
64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
65 | * in 2000 - many ideas in the current implementation are derived from | |
66 | * his patch. | |
67 | * | |
68 | * Further notes from the original documentation: | |
69 | * | |
70 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 71 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
72 | * The sem is only needed when accessing/extending the cache-chain, which |
73 | * can never happen inside an interrupt (kmem_cache_create(), | |
74 | * kmem_cache_shrink() and kmem_cache_reap()). | |
75 | * | |
76 | * At present, each engine can be growing a cache. This should be blocked. | |
77 | * | |
e498be7d CL |
78 | * 15 March 2005. NUMA slab allocator. |
79 | * Shai Fultheim <shai@scalex86.org>. | |
80 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
81 | * Alok N Kataria <alokk@calsoftinc.com> | |
82 | * Christoph Lameter <christoph@lameter.com> | |
83 | * | |
84 | * Modified the slab allocator to be node aware on NUMA systems. | |
85 | * Each node has its own list of partial, free and full slabs. | |
86 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
87 | */ |
88 | ||
1da177e4 LT |
89 | #include <linux/slab.h> |
90 | #include <linux/mm.h> | |
c9cf5528 | 91 | #include <linux/poison.h> |
1da177e4 LT |
92 | #include <linux/swap.h> |
93 | #include <linux/cache.h> | |
94 | #include <linux/interrupt.h> | |
95 | #include <linux/init.h> | |
96 | #include <linux/compiler.h> | |
101a5001 | 97 | #include <linux/cpuset.h> |
a0ec95a8 | 98 | #include <linux/proc_fs.h> |
1da177e4 LT |
99 | #include <linux/seq_file.h> |
100 | #include <linux/notifier.h> | |
101 | #include <linux/kallsyms.h> | |
102 | #include <linux/cpu.h> | |
103 | #include <linux/sysctl.h> | |
104 | #include <linux/module.h> | |
105 | #include <linux/rcupdate.h> | |
543537bd | 106 | #include <linux/string.h> |
138ae663 | 107 | #include <linux/uaccess.h> |
e498be7d | 108 | #include <linux/nodemask.h> |
d5cff635 | 109 | #include <linux/kmemleak.h> |
dc85da15 | 110 | #include <linux/mempolicy.h> |
fc0abb14 | 111 | #include <linux/mutex.h> |
8a8b6502 | 112 | #include <linux/fault-inject.h> |
e7eebaf6 | 113 | #include <linux/rtmutex.h> |
6a2d7a95 | 114 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 115 | #include <linux/debugobjects.h> |
c175eea4 | 116 | #include <linux/kmemcheck.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
1da177e4 | 119 | |
381760ea MG |
120 | #include <net/sock.h> |
121 | ||
1da177e4 LT |
122 | #include <asm/cacheflush.h> |
123 | #include <asm/tlbflush.h> | |
124 | #include <asm/page.h> | |
125 | ||
4dee6b64 SR |
126 | #include <trace/events/kmem.h> |
127 | ||
072bb0aa MG |
128 | #include "internal.h" |
129 | ||
b9ce5ef4 GC |
130 | #include "slab.h" |
131 | ||
1da177e4 | 132 | /* |
50953fe9 | 133 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
134 | * 0 for faster, smaller code (especially in the critical paths). |
135 | * | |
136 | * STATS - 1 to collect stats for /proc/slabinfo. | |
137 | * 0 for faster, smaller code (especially in the critical paths). | |
138 | * | |
139 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
140 | */ | |
141 | ||
142 | #ifdef CONFIG_DEBUG_SLAB | |
143 | #define DEBUG 1 | |
144 | #define STATS 1 | |
145 | #define FORCED_DEBUG 1 | |
146 | #else | |
147 | #define DEBUG 0 | |
148 | #define STATS 0 | |
149 | #define FORCED_DEBUG 0 | |
150 | #endif | |
151 | ||
1da177e4 LT |
152 | /* Shouldn't this be in a header file somewhere? */ |
153 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 154 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 155 | |
1da177e4 LT |
156 | #ifndef ARCH_KMALLOC_FLAGS |
157 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
158 | #endif | |
159 | ||
f315e3fa JK |
160 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
161 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
162 | ||
163 | #if FREELIST_BYTE_INDEX | |
164 | typedef unsigned char freelist_idx_t; | |
165 | #else | |
166 | typedef unsigned short freelist_idx_t; | |
167 | #endif | |
168 | ||
30321c7b | 169 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 170 | |
072bb0aa MG |
171 | /* |
172 | * true if a page was allocated from pfmemalloc reserves for network-based | |
173 | * swap | |
174 | */ | |
175 | static bool pfmemalloc_active __read_mostly; | |
176 | ||
1da177e4 LT |
177 | /* |
178 | * struct array_cache | |
179 | * | |
1da177e4 LT |
180 | * Purpose: |
181 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
182 | * - reduce the number of linked list operations | |
183 | * - reduce spinlock operations | |
184 | * | |
185 | * The limit is stored in the per-cpu structure to reduce the data cache | |
186 | * footprint. | |
187 | * | |
188 | */ | |
189 | struct array_cache { | |
190 | unsigned int avail; | |
191 | unsigned int limit; | |
192 | unsigned int batchcount; | |
193 | unsigned int touched; | |
bda5b655 | 194 | void *entry[]; /* |
a737b3e2 AM |
195 | * Must have this definition in here for the proper |
196 | * alignment of array_cache. Also simplifies accessing | |
197 | * the entries. | |
072bb0aa MG |
198 | * |
199 | * Entries should not be directly dereferenced as | |
200 | * entries belonging to slabs marked pfmemalloc will | |
201 | * have the lower bits set SLAB_OBJ_PFMEMALLOC | |
a737b3e2 | 202 | */ |
1da177e4 LT |
203 | }; |
204 | ||
c8522a3a JK |
205 | struct alien_cache { |
206 | spinlock_t lock; | |
207 | struct array_cache ac; | |
208 | }; | |
209 | ||
072bb0aa MG |
210 | #define SLAB_OBJ_PFMEMALLOC 1 |
211 | static inline bool is_obj_pfmemalloc(void *objp) | |
212 | { | |
213 | return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC; | |
214 | } | |
215 | ||
216 | static inline void set_obj_pfmemalloc(void **objp) | |
217 | { | |
218 | *objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC); | |
219 | return; | |
220 | } | |
221 | ||
222 | static inline void clear_obj_pfmemalloc(void **objp) | |
223 | { | |
224 | *objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC); | |
225 | } | |
226 | ||
a737b3e2 AM |
227 | /* |
228 | * bootstrap: The caches do not work without cpuarrays anymore, but the | |
229 | * cpuarrays are allocated from the generic caches... | |
1da177e4 LT |
230 | */ |
231 | #define BOOT_CPUCACHE_ENTRIES 1 | |
232 | struct arraycache_init { | |
233 | struct array_cache cache; | |
b28a02de | 234 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
1da177e4 LT |
235 | }; |
236 | ||
e498be7d CL |
237 | /* |
238 | * Need this for bootstrapping a per node allocator. | |
239 | */ | |
556a169d | 240 | #define NUM_INIT_LISTS (3 * MAX_NUMNODES) |
ce8eb6c4 | 241 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 242 | #define CACHE_CACHE 0 |
556a169d | 243 | #define SIZE_AC MAX_NUMNODES |
ce8eb6c4 | 244 | #define SIZE_NODE (2 * MAX_NUMNODES) |
e498be7d | 245 | |
ed11d9eb | 246 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 247 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 248 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
249 | int node, struct list_head *list); |
250 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 251 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 252 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 253 | |
e0a42726 IM |
254 | static int slab_early_init = 1; |
255 | ||
e3366016 | 256 | #define INDEX_AC kmalloc_index(sizeof(struct arraycache_init)) |
ce8eb6c4 | 257 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 258 | |
ce8eb6c4 | 259 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
260 | { |
261 | INIT_LIST_HEAD(&parent->slabs_full); | |
262 | INIT_LIST_HEAD(&parent->slabs_partial); | |
263 | INIT_LIST_HEAD(&parent->slabs_free); | |
264 | parent->shared = NULL; | |
265 | parent->alien = NULL; | |
2e1217cf | 266 | parent->colour_next = 0; |
e498be7d CL |
267 | spin_lock_init(&parent->list_lock); |
268 | parent->free_objects = 0; | |
269 | parent->free_touched = 0; | |
270 | } | |
271 | ||
a737b3e2 AM |
272 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
273 | do { \ | |
274 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 275 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
276 | } while (0) |
277 | ||
a737b3e2 AM |
278 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
279 | do { \ | |
e498be7d CL |
280 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
281 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
282 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
283 | } while (0) | |
1da177e4 | 284 | |
1da177e4 LT |
285 | #define CFLGS_OFF_SLAB (0x80000000UL) |
286 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
287 | ||
288 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
289 | /* |
290 | * Optimization question: fewer reaps means less probability for unnessary | |
291 | * cpucache drain/refill cycles. | |
1da177e4 | 292 | * |
dc6f3f27 | 293 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
294 | * which could lock up otherwise freeable slabs. |
295 | */ | |
5f0985bb JZ |
296 | #define REAPTIMEOUT_AC (2*HZ) |
297 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
298 | |
299 | #if STATS | |
300 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
301 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
302 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
303 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 304 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
305 | #define STATS_SET_HIGH(x) \ |
306 | do { \ | |
307 | if ((x)->num_active > (x)->high_mark) \ | |
308 | (x)->high_mark = (x)->num_active; \ | |
309 | } while (0) | |
1da177e4 LT |
310 | #define STATS_INC_ERR(x) ((x)->errors++) |
311 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 312 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 313 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
314 | #define STATS_SET_FREEABLE(x, i) \ |
315 | do { \ | |
316 | if ((x)->max_freeable < i) \ | |
317 | (x)->max_freeable = i; \ | |
318 | } while (0) | |
1da177e4 LT |
319 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
320 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
321 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
322 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
323 | #else | |
324 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
325 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
326 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
327 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 328 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
329 | #define STATS_SET_HIGH(x) do { } while (0) |
330 | #define STATS_INC_ERR(x) do { } while (0) | |
331 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 332 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 333 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 334 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
335 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
336 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
337 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
338 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
339 | #endif | |
340 | ||
341 | #if DEBUG | |
1da177e4 | 342 | |
a737b3e2 AM |
343 | /* |
344 | * memory layout of objects: | |
1da177e4 | 345 | * 0 : objp |
3dafccf2 | 346 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
347 | * the end of an object is aligned with the end of the real |
348 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 349 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 350 | * redzone word. |
3dafccf2 | 351 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
352 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
353 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 354 | * [BYTES_PER_WORD long] |
1da177e4 | 355 | */ |
343e0d7a | 356 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 357 | { |
3dafccf2 | 358 | return cachep->obj_offset; |
1da177e4 LT |
359 | } |
360 | ||
b46b8f19 | 361 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
362 | { |
363 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
364 | return (unsigned long long*) (objp + obj_offset(cachep) - |
365 | sizeof(unsigned long long)); | |
1da177e4 LT |
366 | } |
367 | ||
b46b8f19 | 368 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
369 | { |
370 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
371 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 372 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 373 | sizeof(unsigned long long) - |
87a927c7 | 374 | REDZONE_ALIGN); |
3b0efdfa | 375 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 376 | sizeof(unsigned long long)); |
1da177e4 LT |
377 | } |
378 | ||
343e0d7a | 379 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
380 | { |
381 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 382 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
383 | } |
384 | ||
385 | #else | |
386 | ||
3dafccf2 | 387 | #define obj_offset(x) 0 |
b46b8f19 DW |
388 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
389 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
390 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
391 | ||
392 | #endif | |
393 | ||
03787301 JK |
394 | #define OBJECT_FREE (0) |
395 | #define OBJECT_ACTIVE (1) | |
396 | ||
397 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
398 | ||
399 | static void set_obj_status(struct page *page, int idx, int val) | |
400 | { | |
401 | int freelist_size; | |
402 | char *status; | |
403 | struct kmem_cache *cachep = page->slab_cache; | |
404 | ||
405 | freelist_size = cachep->num * sizeof(freelist_idx_t); | |
406 | status = (char *)page->freelist + freelist_size; | |
407 | status[idx] = val; | |
408 | } | |
409 | ||
410 | static inline unsigned int get_obj_status(struct page *page, int idx) | |
411 | { | |
412 | int freelist_size; | |
413 | char *status; | |
414 | struct kmem_cache *cachep = page->slab_cache; | |
415 | ||
416 | freelist_size = cachep->num * sizeof(freelist_idx_t); | |
417 | status = (char *)page->freelist + freelist_size; | |
418 | ||
419 | return status[idx]; | |
420 | } | |
421 | ||
422 | #else | |
423 | static inline void set_obj_status(struct page *page, int idx, int val) {} | |
424 | ||
425 | #endif | |
426 | ||
1da177e4 | 427 | /* |
3df1cccd DR |
428 | * Do not go above this order unless 0 objects fit into the slab or |
429 | * overridden on the command line. | |
1da177e4 | 430 | */ |
543585cc DR |
431 | #define SLAB_MAX_ORDER_HI 1 |
432 | #define SLAB_MAX_ORDER_LO 0 | |
433 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 434 | static bool slab_max_order_set __initdata; |
1da177e4 | 435 | |
6ed5eb22 PE |
436 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
437 | { | |
b49af68f | 438 | struct page *page = virt_to_head_page(obj); |
35026088 | 439 | return page->slab_cache; |
6ed5eb22 PE |
440 | } |
441 | ||
8456a648 | 442 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
443 | unsigned int idx) |
444 | { | |
8456a648 | 445 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
446 | } |
447 | ||
6a2d7a95 | 448 | /* |
3b0efdfa CL |
449 | * We want to avoid an expensive divide : (offset / cache->size) |
450 | * Using the fact that size is a constant for a particular cache, | |
451 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
452 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
453 | */ | |
454 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 455 | const struct page *page, void *obj) |
8fea4e96 | 456 | { |
8456a648 | 457 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 458 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
459 | } |
460 | ||
1da177e4 | 461 | static struct arraycache_init initarray_generic = |
b28a02de | 462 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 LT |
463 | |
464 | /* internal cache of cache description objs */ | |
9b030cb8 | 465 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
466 | .batchcount = 1, |
467 | .limit = BOOT_CPUCACHE_ENTRIES, | |
468 | .shared = 1, | |
3b0efdfa | 469 | .size = sizeof(struct kmem_cache), |
b28a02de | 470 | .name = "kmem_cache", |
1da177e4 LT |
471 | }; |
472 | ||
1871e52c | 473 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 474 | |
343e0d7a | 475 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
476 | { |
477 | return cachep->array[smp_processor_id()]; | |
478 | } | |
479 | ||
03787301 JK |
480 | static size_t calculate_freelist_size(int nr_objs, size_t align) |
481 | { | |
482 | size_t freelist_size; | |
483 | ||
484 | freelist_size = nr_objs * sizeof(freelist_idx_t); | |
485 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) | |
486 | freelist_size += nr_objs * sizeof(char); | |
487 | ||
488 | if (align) | |
489 | freelist_size = ALIGN(freelist_size, align); | |
490 | ||
491 | return freelist_size; | |
492 | } | |
493 | ||
9cef2e2b JK |
494 | static int calculate_nr_objs(size_t slab_size, size_t buffer_size, |
495 | size_t idx_size, size_t align) | |
1da177e4 | 496 | { |
9cef2e2b | 497 | int nr_objs; |
03787301 | 498 | size_t remained_size; |
9cef2e2b | 499 | size_t freelist_size; |
03787301 | 500 | int extra_space = 0; |
9cef2e2b | 501 | |
03787301 JK |
502 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
503 | extra_space = sizeof(char); | |
9cef2e2b JK |
504 | /* |
505 | * Ignore padding for the initial guess. The padding | |
506 | * is at most @align-1 bytes, and @buffer_size is at | |
507 | * least @align. In the worst case, this result will | |
508 | * be one greater than the number of objects that fit | |
509 | * into the memory allocation when taking the padding | |
510 | * into account. | |
511 | */ | |
03787301 | 512 | nr_objs = slab_size / (buffer_size + idx_size + extra_space); |
9cef2e2b JK |
513 | |
514 | /* | |
515 | * This calculated number will be either the right | |
516 | * amount, or one greater than what we want. | |
517 | */ | |
03787301 JK |
518 | remained_size = slab_size - nr_objs * buffer_size; |
519 | freelist_size = calculate_freelist_size(nr_objs, align); | |
520 | if (remained_size < freelist_size) | |
9cef2e2b JK |
521 | nr_objs--; |
522 | ||
523 | return nr_objs; | |
fbaccacf | 524 | } |
1da177e4 | 525 | |
a737b3e2 AM |
526 | /* |
527 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
528 | */ | |
fbaccacf SR |
529 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
530 | size_t align, int flags, size_t *left_over, | |
531 | unsigned int *num) | |
532 | { | |
533 | int nr_objs; | |
534 | size_t mgmt_size; | |
535 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 536 | |
fbaccacf SR |
537 | /* |
538 | * The slab management structure can be either off the slab or | |
539 | * on it. For the latter case, the memory allocated for a | |
540 | * slab is used for: | |
541 | * | |
16025177 | 542 | * - One unsigned int for each object |
fbaccacf SR |
543 | * - Padding to respect alignment of @align |
544 | * - @buffer_size bytes for each object | |
545 | * | |
546 | * If the slab management structure is off the slab, then the | |
547 | * alignment will already be calculated into the size. Because | |
548 | * the slabs are all pages aligned, the objects will be at the | |
549 | * correct alignment when allocated. | |
550 | */ | |
551 | if (flags & CFLGS_OFF_SLAB) { | |
552 | mgmt_size = 0; | |
553 | nr_objs = slab_size / buffer_size; | |
554 | ||
fbaccacf | 555 | } else { |
9cef2e2b | 556 | nr_objs = calculate_nr_objs(slab_size, buffer_size, |
a41adfaa | 557 | sizeof(freelist_idx_t), align); |
03787301 | 558 | mgmt_size = calculate_freelist_size(nr_objs, align); |
fbaccacf SR |
559 | } |
560 | *num = nr_objs; | |
561 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
562 | } |
563 | ||
f28510d3 | 564 | #if DEBUG |
d40cee24 | 565 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 566 | |
a737b3e2 AM |
567 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
568 | char *msg) | |
1da177e4 LT |
569 | { |
570 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 571 | function, cachep->name, msg); |
1da177e4 | 572 | dump_stack(); |
373d4d09 | 573 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 574 | } |
f28510d3 | 575 | #endif |
1da177e4 | 576 | |
3395ee05 PM |
577 | /* |
578 | * By default on NUMA we use alien caches to stage the freeing of | |
579 | * objects allocated from other nodes. This causes massive memory | |
580 | * inefficiencies when using fake NUMA setup to split memory into a | |
581 | * large number of small nodes, so it can be disabled on the command | |
582 | * line | |
583 | */ | |
584 | ||
585 | static int use_alien_caches __read_mostly = 1; | |
586 | static int __init noaliencache_setup(char *s) | |
587 | { | |
588 | use_alien_caches = 0; | |
589 | return 1; | |
590 | } | |
591 | __setup("noaliencache", noaliencache_setup); | |
592 | ||
3df1cccd DR |
593 | static int __init slab_max_order_setup(char *str) |
594 | { | |
595 | get_option(&str, &slab_max_order); | |
596 | slab_max_order = slab_max_order < 0 ? 0 : | |
597 | min(slab_max_order, MAX_ORDER - 1); | |
598 | slab_max_order_set = true; | |
599 | ||
600 | return 1; | |
601 | } | |
602 | __setup("slab_max_order=", slab_max_order_setup); | |
603 | ||
8fce4d8e CL |
604 | #ifdef CONFIG_NUMA |
605 | /* | |
606 | * Special reaping functions for NUMA systems called from cache_reap(). | |
607 | * These take care of doing round robin flushing of alien caches (containing | |
608 | * objects freed on different nodes from which they were allocated) and the | |
609 | * flushing of remote pcps by calling drain_node_pages. | |
610 | */ | |
1871e52c | 611 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
612 | |
613 | static void init_reap_node(int cpu) | |
614 | { | |
615 | int node; | |
616 | ||
7d6e6d09 | 617 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 618 | if (node == MAX_NUMNODES) |
442295c9 | 619 | node = first_node(node_online_map); |
8fce4d8e | 620 | |
1871e52c | 621 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
622 | } |
623 | ||
624 | static void next_reap_node(void) | |
625 | { | |
909ea964 | 626 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 627 | |
8fce4d8e CL |
628 | node = next_node(node, node_online_map); |
629 | if (unlikely(node >= MAX_NUMNODES)) | |
630 | node = first_node(node_online_map); | |
909ea964 | 631 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
632 | } |
633 | ||
634 | #else | |
635 | #define init_reap_node(cpu) do { } while (0) | |
636 | #define next_reap_node(void) do { } while (0) | |
637 | #endif | |
638 | ||
1da177e4 LT |
639 | /* |
640 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
641 | * via the workqueue/eventd. | |
642 | * Add the CPU number into the expiration time to minimize the possibility of | |
643 | * the CPUs getting into lockstep and contending for the global cache chain | |
644 | * lock. | |
645 | */ | |
0db0628d | 646 | static void start_cpu_timer(int cpu) |
1da177e4 | 647 | { |
1871e52c | 648 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
649 | |
650 | /* | |
651 | * When this gets called from do_initcalls via cpucache_init(), | |
652 | * init_workqueues() has already run, so keventd will be setup | |
653 | * at that time. | |
654 | */ | |
52bad64d | 655 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 656 | init_reap_node(cpu); |
203b42f7 | 657 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
658 | schedule_delayed_work_on(cpu, reap_work, |
659 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
660 | } |
661 | } | |
662 | ||
1fe00d50 | 663 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 664 | { |
d5cff635 CM |
665 | /* |
666 | * The array_cache structures contain pointers to free object. | |
25985edc | 667 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
668 | * cache the pointers are not cleared and they could be counted as |
669 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
670 | * not scan such objects. | |
671 | */ | |
1fe00d50 JK |
672 | kmemleak_no_scan(ac); |
673 | if (ac) { | |
674 | ac->avail = 0; | |
675 | ac->limit = limit; | |
676 | ac->batchcount = batch; | |
677 | ac->touched = 0; | |
1da177e4 | 678 | } |
1fe00d50 JK |
679 | } |
680 | ||
681 | static struct array_cache *alloc_arraycache(int node, int entries, | |
682 | int batchcount, gfp_t gfp) | |
683 | { | |
5e804789 | 684 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
685 | struct array_cache *ac = NULL; |
686 | ||
687 | ac = kmalloc_node(memsize, gfp, node); | |
688 | init_arraycache(ac, entries, batchcount); | |
689 | return ac; | |
1da177e4 LT |
690 | } |
691 | ||
8456a648 | 692 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 693 | { |
072bb0aa MG |
694 | return PageSlabPfmemalloc(page); |
695 | } | |
696 | ||
697 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
698 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
699 | struct array_cache *ac) | |
700 | { | |
18bf8541 | 701 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 702 | struct page *page; |
072bb0aa MG |
703 | unsigned long flags; |
704 | ||
705 | if (!pfmemalloc_active) | |
706 | return; | |
707 | ||
ce8eb6c4 | 708 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
709 | list_for_each_entry(page, &n->slabs_full, lru) |
710 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
711 | goto out; |
712 | ||
8456a648 JK |
713 | list_for_each_entry(page, &n->slabs_partial, lru) |
714 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
715 | goto out; |
716 | ||
8456a648 JK |
717 | list_for_each_entry(page, &n->slabs_free, lru) |
718 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
719 | goto out; |
720 | ||
721 | pfmemalloc_active = false; | |
722 | out: | |
ce8eb6c4 | 723 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
724 | } |
725 | ||
381760ea | 726 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
727 | gfp_t flags, bool force_refill) |
728 | { | |
729 | int i; | |
730 | void *objp = ac->entry[--ac->avail]; | |
731 | ||
732 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
733 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 734 | struct kmem_cache_node *n; |
072bb0aa MG |
735 | |
736 | if (gfp_pfmemalloc_allowed(flags)) { | |
737 | clear_obj_pfmemalloc(&objp); | |
738 | return objp; | |
739 | } | |
740 | ||
741 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 742 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
743 | /* If a !PFMEMALLOC object is found, swap them */ |
744 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
745 | objp = ac->entry[i]; | |
746 | ac->entry[i] = ac->entry[ac->avail]; | |
747 | ac->entry[ac->avail] = objp; | |
748 | return objp; | |
749 | } | |
750 | } | |
751 | ||
752 | /* | |
753 | * If there are empty slabs on the slabs_free list and we are | |
754 | * being forced to refill the cache, mark this one !pfmemalloc. | |
755 | */ | |
18bf8541 | 756 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 757 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 758 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 759 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
760 | clear_obj_pfmemalloc(&objp); |
761 | recheck_pfmemalloc_active(cachep, ac); | |
762 | return objp; | |
763 | } | |
764 | ||
765 | /* No !PFMEMALLOC objects available */ | |
766 | ac->avail++; | |
767 | objp = NULL; | |
768 | } | |
769 | ||
770 | return objp; | |
771 | } | |
772 | ||
381760ea MG |
773 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
774 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
775 | { | |
776 | void *objp; | |
777 | ||
778 | if (unlikely(sk_memalloc_socks())) | |
779 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
780 | else | |
781 | objp = ac->entry[--ac->avail]; | |
782 | ||
783 | return objp; | |
784 | } | |
785 | ||
786 | static void *__ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
072bb0aa MG |
787 | void *objp) |
788 | { | |
789 | if (unlikely(pfmemalloc_active)) { | |
790 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 791 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
792 | if (PageSlabPfmemalloc(page)) |
793 | set_obj_pfmemalloc(&objp); | |
794 | } | |
795 | ||
381760ea MG |
796 | return objp; |
797 | } | |
798 | ||
799 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
800 | void *objp) | |
801 | { | |
802 | if (unlikely(sk_memalloc_socks())) | |
803 | objp = __ac_put_obj(cachep, ac, objp); | |
804 | ||
072bb0aa MG |
805 | ac->entry[ac->avail++] = objp; |
806 | } | |
807 | ||
3ded175a CL |
808 | /* |
809 | * Transfer objects in one arraycache to another. | |
810 | * Locking must be handled by the caller. | |
811 | * | |
812 | * Return the number of entries transferred. | |
813 | */ | |
814 | static int transfer_objects(struct array_cache *to, | |
815 | struct array_cache *from, unsigned int max) | |
816 | { | |
817 | /* Figure out how many entries to transfer */ | |
732eacc0 | 818 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
819 | |
820 | if (!nr) | |
821 | return 0; | |
822 | ||
823 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
824 | sizeof(void *) *nr); | |
825 | ||
826 | from->avail -= nr; | |
827 | to->avail += nr; | |
3ded175a CL |
828 | return nr; |
829 | } | |
830 | ||
765c4507 CL |
831 | #ifndef CONFIG_NUMA |
832 | ||
833 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 834 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 835 | |
c8522a3a JK |
836 | static inline struct alien_cache **alloc_alien_cache(int node, |
837 | int limit, gfp_t gfp) | |
765c4507 | 838 | { |
a6406168 | 839 | return NULL; |
765c4507 CL |
840 | } |
841 | ||
c8522a3a | 842 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
843 | { |
844 | } | |
845 | ||
846 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
847 | { | |
848 | return 0; | |
849 | } | |
850 | ||
851 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
852 | gfp_t flags) | |
853 | { | |
854 | return NULL; | |
855 | } | |
856 | ||
8b98c169 | 857 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
858 | gfp_t flags, int nodeid) |
859 | { | |
860 | return NULL; | |
861 | } | |
862 | ||
863 | #else /* CONFIG_NUMA */ | |
864 | ||
8b98c169 | 865 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 866 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 867 | |
c8522a3a JK |
868 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
869 | int batch, gfp_t gfp) | |
870 | { | |
5e804789 | 871 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
872 | struct alien_cache *alc = NULL; |
873 | ||
874 | alc = kmalloc_node(memsize, gfp, node); | |
875 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 876 | spin_lock_init(&alc->lock); |
c8522a3a JK |
877 | return alc; |
878 | } | |
879 | ||
880 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 881 | { |
c8522a3a | 882 | struct alien_cache **alc_ptr; |
5e804789 | 883 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
884 | int i; |
885 | ||
886 | if (limit > 1) | |
887 | limit = 12; | |
c8522a3a JK |
888 | alc_ptr = kzalloc_node(memsize, gfp, node); |
889 | if (!alc_ptr) | |
890 | return NULL; | |
891 | ||
892 | for_each_node(i) { | |
893 | if (i == node || !node_online(i)) | |
894 | continue; | |
895 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
896 | if (!alc_ptr[i]) { | |
897 | for (i--; i >= 0; i--) | |
898 | kfree(alc_ptr[i]); | |
899 | kfree(alc_ptr); | |
900 | return NULL; | |
e498be7d CL |
901 | } |
902 | } | |
c8522a3a | 903 | return alc_ptr; |
e498be7d CL |
904 | } |
905 | ||
c8522a3a | 906 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
907 | { |
908 | int i; | |
909 | ||
c8522a3a | 910 | if (!alc_ptr) |
e498be7d | 911 | return; |
e498be7d | 912 | for_each_node(i) |
c8522a3a JK |
913 | kfree(alc_ptr[i]); |
914 | kfree(alc_ptr); | |
e498be7d CL |
915 | } |
916 | ||
343e0d7a | 917 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
918 | struct array_cache *ac, int node, |
919 | struct list_head *list) | |
e498be7d | 920 | { |
18bf8541 | 921 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
922 | |
923 | if (ac->avail) { | |
ce8eb6c4 | 924 | spin_lock(&n->list_lock); |
e00946fe CL |
925 | /* |
926 | * Stuff objects into the remote nodes shared array first. | |
927 | * That way we could avoid the overhead of putting the objects | |
928 | * into the free lists and getting them back later. | |
929 | */ | |
ce8eb6c4 CL |
930 | if (n->shared) |
931 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 932 | |
833b706c | 933 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 934 | ac->avail = 0; |
ce8eb6c4 | 935 | spin_unlock(&n->list_lock); |
e498be7d CL |
936 | } |
937 | } | |
938 | ||
8fce4d8e CL |
939 | /* |
940 | * Called from cache_reap() to regularly drain alien caches round robin. | |
941 | */ | |
ce8eb6c4 | 942 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 943 | { |
909ea964 | 944 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 945 | |
ce8eb6c4 | 946 | if (n->alien) { |
c8522a3a JK |
947 | struct alien_cache *alc = n->alien[node]; |
948 | struct array_cache *ac; | |
949 | ||
950 | if (alc) { | |
951 | ac = &alc->ac; | |
49dfc304 | 952 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
953 | LIST_HEAD(list); |
954 | ||
955 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 956 | spin_unlock_irq(&alc->lock); |
833b706c | 957 | slabs_destroy(cachep, &list); |
c8522a3a | 958 | } |
8fce4d8e CL |
959 | } |
960 | } | |
961 | } | |
962 | ||
a737b3e2 | 963 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 964 | struct alien_cache **alien) |
e498be7d | 965 | { |
b28a02de | 966 | int i = 0; |
c8522a3a | 967 | struct alien_cache *alc; |
e498be7d CL |
968 | struct array_cache *ac; |
969 | unsigned long flags; | |
970 | ||
971 | for_each_online_node(i) { | |
c8522a3a JK |
972 | alc = alien[i]; |
973 | if (alc) { | |
833b706c JK |
974 | LIST_HEAD(list); |
975 | ||
c8522a3a | 976 | ac = &alc->ac; |
49dfc304 | 977 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 978 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 979 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 980 | slabs_destroy(cachep, &list); |
e498be7d CL |
981 | } |
982 | } | |
983 | } | |
729bd0b7 | 984 | |
873623df | 985 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 | 986 | { |
1ea991b0 | 987 | int nodeid = page_to_nid(virt_to_page(objp)); |
ce8eb6c4 | 988 | struct kmem_cache_node *n; |
c8522a3a JK |
989 | struct alien_cache *alien = NULL; |
990 | struct array_cache *ac; | |
1ca4cb24 | 991 | int node; |
97654dfa | 992 | LIST_HEAD(list); |
1ca4cb24 | 993 | |
7d6e6d09 | 994 | node = numa_mem_id(); |
729bd0b7 PE |
995 | |
996 | /* | |
997 | * Make sure we are not freeing a object from another node to the array | |
998 | * cache on this cpu. | |
999 | */ | |
1ea991b0 | 1000 | if (likely(nodeid == node)) |
729bd0b7 PE |
1001 | return 0; |
1002 | ||
18bf8541 | 1003 | n = get_node(cachep, node); |
729bd0b7 | 1004 | STATS_INC_NODEFREES(cachep); |
ce8eb6c4 CL |
1005 | if (n->alien && n->alien[nodeid]) { |
1006 | alien = n->alien[nodeid]; | |
c8522a3a | 1007 | ac = &alien->ac; |
49dfc304 | 1008 | spin_lock(&alien->lock); |
c8522a3a | 1009 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 1010 | STATS_INC_ACOVERFLOW(cachep); |
833b706c | 1011 | __drain_alien_cache(cachep, ac, nodeid, &list); |
729bd0b7 | 1012 | } |
c8522a3a | 1013 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 1014 | spin_unlock(&alien->lock); |
833b706c | 1015 | slabs_destroy(cachep, &list); |
729bd0b7 | 1016 | } else { |
18bf8541 CL |
1017 | n = get_node(cachep, nodeid); |
1018 | spin_lock(&n->list_lock); | |
97654dfa | 1019 | free_block(cachep, &objp, 1, nodeid, &list); |
18bf8541 | 1020 | spin_unlock(&n->list_lock); |
97654dfa | 1021 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
1022 | } |
1023 | return 1; | |
1024 | } | |
e498be7d CL |
1025 | #endif |
1026 | ||
8f9f8d9e | 1027 | /* |
6a67368c | 1028 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1029 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1030 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1031 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1032 | * already in use. |
1033 | * | |
18004c5d | 1034 | * Must hold slab_mutex. |
8f9f8d9e | 1035 | */ |
6a67368c | 1036 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1037 | { |
1038 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1039 | struct kmem_cache_node *n; |
5e804789 | 1040 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1041 | |
18004c5d | 1042 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 1043 | /* |
5f0985bb | 1044 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
1045 | * begin anything. Make sure some other cpu on this |
1046 | * node has not already allocated this | |
1047 | */ | |
18bf8541 CL |
1048 | n = get_node(cachep, node); |
1049 | if (!n) { | |
ce8eb6c4 CL |
1050 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1051 | if (!n) | |
8f9f8d9e | 1052 | return -ENOMEM; |
ce8eb6c4 | 1053 | kmem_cache_node_init(n); |
5f0985bb JZ |
1054 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1055 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1056 | |
1057 | /* | |
5f0985bb JZ |
1058 | * The kmem_cache_nodes don't come and go as CPUs |
1059 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1060 | * protection here. |
1061 | */ | |
ce8eb6c4 | 1062 | cachep->node[node] = n; |
8f9f8d9e DR |
1063 | } |
1064 | ||
18bf8541 CL |
1065 | spin_lock_irq(&n->list_lock); |
1066 | n->free_limit = | |
8f9f8d9e DR |
1067 | (1 + nr_cpus_node(node)) * |
1068 | cachep->batchcount + cachep->num; | |
18bf8541 | 1069 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1070 | } |
1071 | return 0; | |
1072 | } | |
1073 | ||
0fa8103b WL |
1074 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1075 | struct kmem_cache_node *n) | |
1076 | { | |
1077 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1078 | } | |
1079 | ||
0db0628d | 1080 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1081 | { |
1082 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1083 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1084 | int node = cpu_to_mem(cpu); |
a70f7302 | 1085 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1086 | |
18004c5d | 1087 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1088 | struct array_cache *nc; |
1089 | struct array_cache *shared; | |
c8522a3a | 1090 | struct alien_cache **alien; |
97654dfa | 1091 | LIST_HEAD(list); |
fbf1e473 | 1092 | |
fbf1e473 AM |
1093 | /* cpu is dead; no one can alloc from it. */ |
1094 | nc = cachep->array[cpu]; | |
1095 | cachep->array[cpu] = NULL; | |
18bf8541 | 1096 | n = get_node(cachep, node); |
fbf1e473 | 1097 | |
ce8eb6c4 | 1098 | if (!n) |
fbf1e473 AM |
1099 | goto free_array_cache; |
1100 | ||
ce8eb6c4 | 1101 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1102 | |
ce8eb6c4 CL |
1103 | /* Free limit for this kmem_cache_node */ |
1104 | n->free_limit -= cachep->batchcount; | |
fbf1e473 | 1105 | if (nc) |
97654dfa | 1106 | free_block(cachep, nc->entry, nc->avail, node, &list); |
fbf1e473 | 1107 | |
58463c1f | 1108 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1109 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1110 | goto free_array_cache; |
1111 | } | |
1112 | ||
ce8eb6c4 | 1113 | shared = n->shared; |
fbf1e473 AM |
1114 | if (shared) { |
1115 | free_block(cachep, shared->entry, | |
97654dfa | 1116 | shared->avail, node, &list); |
ce8eb6c4 | 1117 | n->shared = NULL; |
fbf1e473 AM |
1118 | } |
1119 | ||
ce8eb6c4 CL |
1120 | alien = n->alien; |
1121 | n->alien = NULL; | |
fbf1e473 | 1122 | |
ce8eb6c4 | 1123 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1124 | |
1125 | kfree(shared); | |
1126 | if (alien) { | |
1127 | drain_alien_cache(cachep, alien); | |
1128 | free_alien_cache(alien); | |
1129 | } | |
1130 | free_array_cache: | |
97654dfa | 1131 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1132 | kfree(nc); |
1133 | } | |
1134 | /* | |
1135 | * In the previous loop, all the objects were freed to | |
1136 | * the respective cache's slabs, now we can go ahead and | |
1137 | * shrink each nodelist to its limit. | |
1138 | */ | |
18004c5d | 1139 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1140 | n = get_node(cachep, node); |
ce8eb6c4 | 1141 | if (!n) |
fbf1e473 | 1142 | continue; |
0fa8103b | 1143 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1144 | } |
1145 | } | |
1146 | ||
0db0628d | 1147 | static int cpuup_prepare(long cpu) |
1da177e4 | 1148 | { |
343e0d7a | 1149 | struct kmem_cache *cachep; |
ce8eb6c4 | 1150 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1151 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1152 | int err; |
1da177e4 | 1153 | |
fbf1e473 AM |
1154 | /* |
1155 | * We need to do this right in the beginning since | |
1156 | * alloc_arraycache's are going to use this list. | |
1157 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1158 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1159 | */ |
6a67368c | 1160 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1161 | if (err < 0) |
1162 | goto bad; | |
fbf1e473 AM |
1163 | |
1164 | /* | |
1165 | * Now we can go ahead with allocating the shared arrays and | |
1166 | * array caches | |
1167 | */ | |
18004c5d | 1168 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1169 | struct array_cache *nc; |
1170 | struct array_cache *shared = NULL; | |
c8522a3a | 1171 | struct alien_cache **alien = NULL; |
fbf1e473 AM |
1172 | |
1173 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1174 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1175 | if (!nc) |
1176 | goto bad; | |
1177 | if (cachep->shared) { | |
1178 | shared = alloc_arraycache(node, | |
1179 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1180 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1181 | if (!shared) { |
1182 | kfree(nc); | |
1da177e4 | 1183 | goto bad; |
12d00f6a | 1184 | } |
fbf1e473 AM |
1185 | } |
1186 | if (use_alien_caches) { | |
83b519e8 | 1187 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1188 | if (!alien) { |
1189 | kfree(shared); | |
1190 | kfree(nc); | |
fbf1e473 | 1191 | goto bad; |
12d00f6a | 1192 | } |
fbf1e473 AM |
1193 | } |
1194 | cachep->array[cpu] = nc; | |
18bf8541 | 1195 | n = get_node(cachep, node); |
ce8eb6c4 | 1196 | BUG_ON(!n); |
fbf1e473 | 1197 | |
ce8eb6c4 CL |
1198 | spin_lock_irq(&n->list_lock); |
1199 | if (!n->shared) { | |
fbf1e473 AM |
1200 | /* |
1201 | * We are serialised from CPU_DEAD or | |
1202 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1203 | */ | |
ce8eb6c4 | 1204 | n->shared = shared; |
fbf1e473 AM |
1205 | shared = NULL; |
1206 | } | |
4484ebf1 | 1207 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1208 | if (!n->alien) { |
1209 | n->alien = alien; | |
fbf1e473 | 1210 | alien = NULL; |
1da177e4 | 1211 | } |
fbf1e473 | 1212 | #endif |
ce8eb6c4 | 1213 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1214 | kfree(shared); |
1215 | free_alien_cache(alien); | |
1216 | } | |
ce79ddc8 | 1217 | |
fbf1e473 AM |
1218 | return 0; |
1219 | bad: | |
12d00f6a | 1220 | cpuup_canceled(cpu); |
fbf1e473 AM |
1221 | return -ENOMEM; |
1222 | } | |
1223 | ||
0db0628d | 1224 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1225 | unsigned long action, void *hcpu) |
1226 | { | |
1227 | long cpu = (long)hcpu; | |
1228 | int err = 0; | |
1229 | ||
1230 | switch (action) { | |
fbf1e473 AM |
1231 | case CPU_UP_PREPARE: |
1232 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1233 | mutex_lock(&slab_mutex); |
fbf1e473 | 1234 | err = cpuup_prepare(cpu); |
18004c5d | 1235 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1236 | break; |
1237 | case CPU_ONLINE: | |
8bb78442 | 1238 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1239 | start_cpu_timer(cpu); |
1240 | break; | |
1241 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1242 | case CPU_DOWN_PREPARE: |
8bb78442 | 1243 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1244 | /* |
18004c5d | 1245 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1246 | * held so that if cache_reap() is invoked it cannot do |
1247 | * anything expensive but will only modify reap_work | |
1248 | * and reschedule the timer. | |
1249 | */ | |
afe2c511 | 1250 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1251 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1252 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1253 | break; |
1254 | case CPU_DOWN_FAILED: | |
8bb78442 | 1255 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1256 | start_cpu_timer(cpu); |
1257 | break; | |
1da177e4 | 1258 | case CPU_DEAD: |
8bb78442 | 1259 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1260 | /* |
1261 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1262 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1263 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1264 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1265 | * structure is usually allocated from kmem_cache_create() and |
1266 | * gets destroyed at kmem_cache_destroy(). | |
1267 | */ | |
183ff22b | 1268 | /* fall through */ |
8f5be20b | 1269 | #endif |
1da177e4 | 1270 | case CPU_UP_CANCELED: |
8bb78442 | 1271 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1272 | mutex_lock(&slab_mutex); |
fbf1e473 | 1273 | cpuup_canceled(cpu); |
18004c5d | 1274 | mutex_unlock(&slab_mutex); |
1da177e4 | 1275 | break; |
1da177e4 | 1276 | } |
eac40680 | 1277 | return notifier_from_errno(err); |
1da177e4 LT |
1278 | } |
1279 | ||
0db0628d | 1280 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1281 | &cpuup_callback, NULL, 0 |
1282 | }; | |
1da177e4 | 1283 | |
8f9f8d9e DR |
1284 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1285 | /* | |
1286 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1287 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1288 | * removed. | |
1289 | * | |
18004c5d | 1290 | * Must hold slab_mutex. |
8f9f8d9e | 1291 | */ |
6a67368c | 1292 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1293 | { |
1294 | struct kmem_cache *cachep; | |
1295 | int ret = 0; | |
1296 | ||
18004c5d | 1297 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1298 | struct kmem_cache_node *n; |
8f9f8d9e | 1299 | |
18bf8541 | 1300 | n = get_node(cachep, node); |
ce8eb6c4 | 1301 | if (!n) |
8f9f8d9e DR |
1302 | continue; |
1303 | ||
0fa8103b | 1304 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1305 | |
ce8eb6c4 CL |
1306 | if (!list_empty(&n->slabs_full) || |
1307 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1308 | ret = -EBUSY; |
1309 | break; | |
1310 | } | |
1311 | } | |
1312 | return ret; | |
1313 | } | |
1314 | ||
1315 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1316 | unsigned long action, void *arg) | |
1317 | { | |
1318 | struct memory_notify *mnb = arg; | |
1319 | int ret = 0; | |
1320 | int nid; | |
1321 | ||
1322 | nid = mnb->status_change_nid; | |
1323 | if (nid < 0) | |
1324 | goto out; | |
1325 | ||
1326 | switch (action) { | |
1327 | case MEM_GOING_ONLINE: | |
18004c5d | 1328 | mutex_lock(&slab_mutex); |
6a67368c | 1329 | ret = init_cache_node_node(nid); |
18004c5d | 1330 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1331 | break; |
1332 | case MEM_GOING_OFFLINE: | |
18004c5d | 1333 | mutex_lock(&slab_mutex); |
6a67368c | 1334 | ret = drain_cache_node_node(nid); |
18004c5d | 1335 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1336 | break; |
1337 | case MEM_ONLINE: | |
1338 | case MEM_OFFLINE: | |
1339 | case MEM_CANCEL_ONLINE: | |
1340 | case MEM_CANCEL_OFFLINE: | |
1341 | break; | |
1342 | } | |
1343 | out: | |
5fda1bd5 | 1344 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1345 | } |
1346 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1347 | ||
e498be7d | 1348 | /* |
ce8eb6c4 | 1349 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1350 | */ |
6744f087 | 1351 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1352 | int nodeid) |
e498be7d | 1353 | { |
6744f087 | 1354 | struct kmem_cache_node *ptr; |
e498be7d | 1355 | |
6744f087 | 1356 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1357 | BUG_ON(!ptr); |
1358 | ||
6744f087 | 1359 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1360 | /* |
1361 | * Do not assume that spinlocks can be initialized via memcpy: | |
1362 | */ | |
1363 | spin_lock_init(&ptr->list_lock); | |
1364 | ||
e498be7d | 1365 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1366 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1367 | } |
1368 | ||
556a169d | 1369 | /* |
ce8eb6c4 CL |
1370 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1371 | * size of kmem_cache_node. | |
556a169d | 1372 | */ |
ce8eb6c4 | 1373 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1374 | { |
1375 | int node; | |
1376 | ||
1377 | for_each_online_node(node) { | |
ce8eb6c4 | 1378 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1379 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1380 | REAPTIMEOUT_NODE + |
1381 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1382 | } |
1383 | } | |
1384 | ||
3c583465 CL |
1385 | /* |
1386 | * The memory after the last cpu cache pointer is used for the | |
6a67368c | 1387 | * the node pointer. |
3c583465 | 1388 | */ |
6a67368c | 1389 | static void setup_node_pointer(struct kmem_cache *cachep) |
3c583465 | 1390 | { |
6a67368c | 1391 | cachep->node = (struct kmem_cache_node **)&cachep->array[nr_cpu_ids]; |
3c583465 CL |
1392 | } |
1393 | ||
a737b3e2 AM |
1394 | /* |
1395 | * Initialisation. Called after the page allocator have been initialised and | |
1396 | * before smp_init(). | |
1da177e4 LT |
1397 | */ |
1398 | void __init kmem_cache_init(void) | |
1399 | { | |
e498be7d CL |
1400 | int i; |
1401 | ||
68126702 JK |
1402 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1403 | sizeof(struct rcu_head)); | |
9b030cb8 | 1404 | kmem_cache = &kmem_cache_boot; |
6a67368c | 1405 | setup_node_pointer(kmem_cache); |
9b030cb8 | 1406 | |
b6e68bc1 | 1407 | if (num_possible_nodes() == 1) |
62918a03 SS |
1408 | use_alien_caches = 0; |
1409 | ||
3c583465 | 1410 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1411 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1412 | |
ce8eb6c4 | 1413 | set_up_node(kmem_cache, CACHE_CACHE); |
1da177e4 LT |
1414 | |
1415 | /* | |
1416 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1417 | * page orders on machines with more than 32MB of memory if |
1418 | * not overridden on the command line. | |
1da177e4 | 1419 | */ |
3df1cccd | 1420 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1421 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1422 | |
1da177e4 LT |
1423 | /* Bootstrap is tricky, because several objects are allocated |
1424 | * from caches that do not exist yet: | |
9b030cb8 CL |
1425 | * 1) initialize the kmem_cache cache: it contains the struct |
1426 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1427 | * kmem_cache is statically allocated. | |
e498be7d | 1428 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1429 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1430 | * array at the end of the bootstrap. |
1da177e4 | 1431 | * 2) Create the first kmalloc cache. |
343e0d7a | 1432 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1433 | * An __init data area is used for the head array. |
1434 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1435 | * head arrays. | |
9b030cb8 | 1436 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1437 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1438 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1439 | * the other cache's with kmalloc allocated memory. |
1440 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1441 | */ |
1442 | ||
9b030cb8 | 1443 | /* 1) create the kmem_cache */ |
1da177e4 | 1444 | |
8da3430d | 1445 | /* |
b56efcf0 | 1446 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1447 | */ |
2f9baa9f CL |
1448 | create_boot_cache(kmem_cache, "kmem_cache", |
1449 | offsetof(struct kmem_cache, array[nr_cpu_ids]) + | |
6744f087 | 1450 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1451 | SLAB_HWCACHE_ALIGN); |
1452 | list_add(&kmem_cache->list, &slab_caches); | |
1da177e4 LT |
1453 | |
1454 | /* 2+3) create the kmalloc caches */ | |
1da177e4 | 1455 | |
a737b3e2 AM |
1456 | /* |
1457 | * Initialize the caches that provide memory for the array cache and the | |
ce8eb6c4 | 1458 | * kmem_cache_node structures first. Without this, further allocations will |
a737b3e2 | 1459 | * bug. |
e498be7d CL |
1460 | */ |
1461 | ||
e3366016 CL |
1462 | kmalloc_caches[INDEX_AC] = create_kmalloc_cache("kmalloc-ac", |
1463 | kmalloc_size(INDEX_AC), ARCH_KMALLOC_FLAGS); | |
45530c44 | 1464 | |
ce8eb6c4 CL |
1465 | if (INDEX_AC != INDEX_NODE) |
1466 | kmalloc_caches[INDEX_NODE] = | |
1467 | create_kmalloc_cache("kmalloc-node", | |
1468 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); | |
e498be7d | 1469 | |
e0a42726 IM |
1470 | slab_early_init = 0; |
1471 | ||
1da177e4 LT |
1472 | /* 4) Replace the bootstrap head arrays */ |
1473 | { | |
2b2d5493 | 1474 | struct array_cache *ptr; |
e498be7d | 1475 | |
83b519e8 | 1476 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1477 | |
9b030cb8 | 1478 | memcpy(ptr, cpu_cache_get(kmem_cache), |
b28a02de | 1479 | sizeof(struct arraycache_init)); |
2b2d5493 | 1480 | |
9b030cb8 | 1481 | kmem_cache->array[smp_processor_id()] = ptr; |
e498be7d | 1482 | |
83b519e8 | 1483 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1484 | |
e3366016 | 1485 | BUG_ON(cpu_cache_get(kmalloc_caches[INDEX_AC]) |
b28a02de | 1486 | != &initarray_generic.cache); |
e3366016 | 1487 | memcpy(ptr, cpu_cache_get(kmalloc_caches[INDEX_AC]), |
b28a02de | 1488 | sizeof(struct arraycache_init)); |
2b2d5493 | 1489 | |
e3366016 | 1490 | kmalloc_caches[INDEX_AC]->array[smp_processor_id()] = ptr; |
1da177e4 | 1491 | } |
ce8eb6c4 | 1492 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1493 | { |
1ca4cb24 PE |
1494 | int nid; |
1495 | ||
9c09a95c | 1496 | for_each_online_node(nid) { |
ce8eb6c4 | 1497 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1498 | |
e3366016 | 1499 | init_list(kmalloc_caches[INDEX_AC], |
ce8eb6c4 | 1500 | &init_kmem_cache_node[SIZE_AC + nid], nid); |
e498be7d | 1501 | |
ce8eb6c4 CL |
1502 | if (INDEX_AC != INDEX_NODE) { |
1503 | init_list(kmalloc_caches[INDEX_NODE], | |
1504 | &init_kmem_cache_node[SIZE_NODE + nid], nid); | |
e498be7d CL |
1505 | } |
1506 | } | |
1507 | } | |
1da177e4 | 1508 | |
f97d5f63 | 1509 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1510 | } |
1511 | ||
1512 | void __init kmem_cache_init_late(void) | |
1513 | { | |
1514 | struct kmem_cache *cachep; | |
1515 | ||
97d06609 | 1516 | slab_state = UP; |
52cef189 | 1517 | |
8429db5c | 1518 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1519 | mutex_lock(&slab_mutex); |
1520 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1521 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1522 | BUG(); | |
18004c5d | 1523 | mutex_unlock(&slab_mutex); |
056c6241 | 1524 | |
97d06609 CL |
1525 | /* Done! */ |
1526 | slab_state = FULL; | |
1527 | ||
a737b3e2 AM |
1528 | /* |
1529 | * Register a cpu startup notifier callback that initializes | |
1530 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1531 | */ |
1532 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1533 | |
8f9f8d9e DR |
1534 | #ifdef CONFIG_NUMA |
1535 | /* | |
1536 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1537 | * node. |
8f9f8d9e DR |
1538 | */ |
1539 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1540 | #endif | |
1541 | ||
a737b3e2 AM |
1542 | /* |
1543 | * The reap timers are started later, with a module init call: That part | |
1544 | * of the kernel is not yet operational. | |
1da177e4 LT |
1545 | */ |
1546 | } | |
1547 | ||
1548 | static int __init cpucache_init(void) | |
1549 | { | |
1550 | int cpu; | |
1551 | ||
a737b3e2 AM |
1552 | /* |
1553 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1554 | */ |
e498be7d | 1555 | for_each_online_cpu(cpu) |
a737b3e2 | 1556 | start_cpu_timer(cpu); |
a164f896 GC |
1557 | |
1558 | /* Done! */ | |
97d06609 | 1559 | slab_state = FULL; |
1da177e4 LT |
1560 | return 0; |
1561 | } | |
1da177e4 LT |
1562 | __initcall(cpucache_init); |
1563 | ||
8bdec192 RA |
1564 | static noinline void |
1565 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1566 | { | |
9a02d699 | 1567 | #if DEBUG |
ce8eb6c4 | 1568 | struct kmem_cache_node *n; |
8456a648 | 1569 | struct page *page; |
8bdec192 RA |
1570 | unsigned long flags; |
1571 | int node; | |
9a02d699 DR |
1572 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1573 | DEFAULT_RATELIMIT_BURST); | |
1574 | ||
1575 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1576 | return; | |
8bdec192 RA |
1577 | |
1578 | printk(KERN_WARNING | |
1579 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1580 | nodeid, gfpflags); | |
1581 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1582 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1583 | |
18bf8541 | 1584 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1585 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1586 | unsigned long active_slabs = 0, num_slabs = 0; | |
1587 | ||
ce8eb6c4 | 1588 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1589 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1590 | active_objs += cachep->num; |
1591 | active_slabs++; | |
1592 | } | |
8456a648 JK |
1593 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1594 | active_objs += page->active; | |
8bdec192 RA |
1595 | active_slabs++; |
1596 | } | |
8456a648 | 1597 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1598 | num_slabs++; |
1599 | ||
ce8eb6c4 CL |
1600 | free_objects += n->free_objects; |
1601 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1602 | |
1603 | num_slabs += active_slabs; | |
1604 | num_objs = num_slabs * cachep->num; | |
1605 | printk(KERN_WARNING | |
1606 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1607 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1608 | free_objects); | |
1609 | } | |
9a02d699 | 1610 | #endif |
8bdec192 RA |
1611 | } |
1612 | ||
1da177e4 | 1613 | /* |
8a7d9b43 WSH |
1614 | * Interface to system's page allocator. No need to hold the |
1615 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1616 | * |
1617 | * If we requested dmaable memory, we will get it. Even if we | |
1618 | * did not request dmaable memory, we might get it, but that | |
1619 | * would be relatively rare and ignorable. | |
1620 | */ | |
0c3aa83e JK |
1621 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1622 | int nodeid) | |
1da177e4 LT |
1623 | { |
1624 | struct page *page; | |
e1b6aa6f | 1625 | int nr_pages; |
765c4507 | 1626 | |
a618e89f | 1627 | flags |= cachep->allocflags; |
e12ba74d MG |
1628 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1629 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1630 | |
5dfb4175 VD |
1631 | if (memcg_charge_slab(cachep, flags, cachep->gfporder)) |
1632 | return NULL; | |
1633 | ||
517d0869 | 1634 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1635 | if (!page) { |
5dfb4175 | 1636 | memcg_uncharge_slab(cachep, cachep->gfporder); |
9a02d699 | 1637 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1638 | return NULL; |
8bdec192 | 1639 | } |
1da177e4 | 1640 | |
b37f1dd0 | 1641 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
072bb0aa MG |
1642 | if (unlikely(page->pfmemalloc)) |
1643 | pfmemalloc_active = true; | |
1644 | ||
e1b6aa6f | 1645 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1646 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1647 | add_zone_page_state(page_zone(page), |
1648 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1649 | else | |
1650 | add_zone_page_state(page_zone(page), | |
1651 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 JK |
1652 | __SetPageSlab(page); |
1653 | if (page->pfmemalloc) | |
1654 | SetPageSlabPfmemalloc(page); | |
072bb0aa | 1655 | |
b1eeab67 VN |
1656 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1657 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1658 | ||
1659 | if (cachep->ctor) | |
1660 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1661 | else | |
1662 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1663 | } | |
c175eea4 | 1664 | |
0c3aa83e | 1665 | return page; |
1da177e4 LT |
1666 | } |
1667 | ||
1668 | /* | |
1669 | * Interface to system's page release. | |
1670 | */ | |
0c3aa83e | 1671 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1672 | { |
a57a4988 | 1673 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1674 | |
b1eeab67 | 1675 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1676 | |
972d1a7b CL |
1677 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1678 | sub_zone_page_state(page_zone(page), | |
1679 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1680 | else | |
1681 | sub_zone_page_state(page_zone(page), | |
1682 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1683 | |
a57a4988 | 1684 | BUG_ON(!PageSlab(page)); |
73293c2f | 1685 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1686 | __ClearPageSlab(page); |
8456a648 JK |
1687 | page_mapcount_reset(page); |
1688 | page->mapping = NULL; | |
1f458cbf | 1689 | |
1da177e4 LT |
1690 | if (current->reclaim_state) |
1691 | current->reclaim_state->reclaimed_slab += nr_freed; | |
5dfb4175 VD |
1692 | __free_pages(page, cachep->gfporder); |
1693 | memcg_uncharge_slab(cachep, cachep->gfporder); | |
1da177e4 LT |
1694 | } |
1695 | ||
1696 | static void kmem_rcu_free(struct rcu_head *head) | |
1697 | { | |
68126702 JK |
1698 | struct kmem_cache *cachep; |
1699 | struct page *page; | |
1da177e4 | 1700 | |
68126702 JK |
1701 | page = container_of(head, struct page, rcu_head); |
1702 | cachep = page->slab_cache; | |
1703 | ||
1704 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1705 | } |
1706 | ||
1707 | #if DEBUG | |
1708 | ||
1709 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1710 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1711 | unsigned long caller) |
1da177e4 | 1712 | { |
8c138bc0 | 1713 | int size = cachep->object_size; |
1da177e4 | 1714 | |
3dafccf2 | 1715 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1716 | |
b28a02de | 1717 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1718 | return; |
1719 | ||
b28a02de PE |
1720 | *addr++ = 0x12345678; |
1721 | *addr++ = caller; | |
1722 | *addr++ = smp_processor_id(); | |
1723 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1724 | { |
1725 | unsigned long *sptr = &caller; | |
1726 | unsigned long svalue; | |
1727 | ||
1728 | while (!kstack_end(sptr)) { | |
1729 | svalue = *sptr++; | |
1730 | if (kernel_text_address(svalue)) { | |
b28a02de | 1731 | *addr++ = svalue; |
1da177e4 LT |
1732 | size -= sizeof(unsigned long); |
1733 | if (size <= sizeof(unsigned long)) | |
1734 | break; | |
1735 | } | |
1736 | } | |
1737 | ||
1738 | } | |
b28a02de | 1739 | *addr++ = 0x87654321; |
1da177e4 LT |
1740 | } |
1741 | #endif | |
1742 | ||
343e0d7a | 1743 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1744 | { |
8c138bc0 | 1745 | int size = cachep->object_size; |
3dafccf2 | 1746 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1747 | |
1748 | memset(addr, val, size); | |
b28a02de | 1749 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1750 | } |
1751 | ||
1752 | static void dump_line(char *data, int offset, int limit) | |
1753 | { | |
1754 | int i; | |
aa83aa40 DJ |
1755 | unsigned char error = 0; |
1756 | int bad_count = 0; | |
1757 | ||
fdde6abb | 1758 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1759 | for (i = 0; i < limit; i++) { |
1760 | if (data[offset + i] != POISON_FREE) { | |
1761 | error = data[offset + i]; | |
1762 | bad_count++; | |
1763 | } | |
aa83aa40 | 1764 | } |
fdde6abb SAS |
1765 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1766 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1767 | |
1768 | if (bad_count == 1) { | |
1769 | error ^= POISON_FREE; | |
1770 | if (!(error & (error - 1))) { | |
1771 | printk(KERN_ERR "Single bit error detected. Probably " | |
1772 | "bad RAM.\n"); | |
1773 | #ifdef CONFIG_X86 | |
1774 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1775 | "test tool.\n"); | |
1776 | #else | |
1777 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1778 | #endif | |
1779 | } | |
1780 | } | |
1da177e4 LT |
1781 | } |
1782 | #endif | |
1783 | ||
1784 | #if DEBUG | |
1785 | ||
343e0d7a | 1786 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1787 | { |
1788 | int i, size; | |
1789 | char *realobj; | |
1790 | ||
1791 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1792 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1793 | *dbg_redzone1(cachep, objp), |
1794 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1795 | } |
1796 | ||
1797 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1798 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1799 | *dbg_userword(cachep, objp), | |
1800 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1801 | } |
3dafccf2 | 1802 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1803 | size = cachep->object_size; |
b28a02de | 1804 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1805 | int limit; |
1806 | limit = 16; | |
b28a02de PE |
1807 | if (i + limit > size) |
1808 | limit = size - i; | |
1da177e4 LT |
1809 | dump_line(realobj, i, limit); |
1810 | } | |
1811 | } | |
1812 | ||
343e0d7a | 1813 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1814 | { |
1815 | char *realobj; | |
1816 | int size, i; | |
1817 | int lines = 0; | |
1818 | ||
3dafccf2 | 1819 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1820 | size = cachep->object_size; |
1da177e4 | 1821 | |
b28a02de | 1822 | for (i = 0; i < size; i++) { |
1da177e4 | 1823 | char exp = POISON_FREE; |
b28a02de | 1824 | if (i == size - 1) |
1da177e4 LT |
1825 | exp = POISON_END; |
1826 | if (realobj[i] != exp) { | |
1827 | int limit; | |
1828 | /* Mismatch ! */ | |
1829 | /* Print header */ | |
1830 | if (lines == 0) { | |
b28a02de | 1831 | printk(KERN_ERR |
face37f5 DJ |
1832 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1833 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1834 | print_objinfo(cachep, objp, 0); |
1835 | } | |
1836 | /* Hexdump the affected line */ | |
b28a02de | 1837 | i = (i / 16) * 16; |
1da177e4 | 1838 | limit = 16; |
b28a02de PE |
1839 | if (i + limit > size) |
1840 | limit = size - i; | |
1da177e4 LT |
1841 | dump_line(realobj, i, limit); |
1842 | i += 16; | |
1843 | lines++; | |
1844 | /* Limit to 5 lines */ | |
1845 | if (lines > 5) | |
1846 | break; | |
1847 | } | |
1848 | } | |
1849 | if (lines != 0) { | |
1850 | /* Print some data about the neighboring objects, if they | |
1851 | * exist: | |
1852 | */ | |
8456a648 | 1853 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1854 | unsigned int objnr; |
1da177e4 | 1855 | |
8456a648 | 1856 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1857 | if (objnr) { |
8456a648 | 1858 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1859 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1860 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1861 | realobj, size); |
1da177e4 LT |
1862 | print_objinfo(cachep, objp, 2); |
1863 | } | |
b28a02de | 1864 | if (objnr + 1 < cachep->num) { |
8456a648 | 1865 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1866 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1867 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1868 | realobj, size); |
1da177e4 LT |
1869 | print_objinfo(cachep, objp, 2); |
1870 | } | |
1871 | } | |
1872 | } | |
1873 | #endif | |
1874 | ||
12dd36fa | 1875 | #if DEBUG |
8456a648 JK |
1876 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1877 | struct page *page) | |
1da177e4 | 1878 | { |
1da177e4 LT |
1879 | int i; |
1880 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 1881 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1882 | |
1883 | if (cachep->flags & SLAB_POISON) { | |
1884 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 1885 | if (cachep->size % PAGE_SIZE == 0 && |
a737b3e2 | 1886 | OFF_SLAB(cachep)) |
b28a02de | 1887 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 1888 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
1889 | else |
1890 | check_poison_obj(cachep, objp); | |
1891 | #else | |
1892 | check_poison_obj(cachep, objp); | |
1893 | #endif | |
1894 | } | |
1895 | if (cachep->flags & SLAB_RED_ZONE) { | |
1896 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1897 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1898 | "was overwritten"); |
1da177e4 LT |
1899 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1900 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1901 | "was overwritten"); |
1da177e4 | 1902 | } |
1da177e4 | 1903 | } |
12dd36fa | 1904 | } |
1da177e4 | 1905 | #else |
8456a648 JK |
1906 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1907 | struct page *page) | |
12dd36fa | 1908 | { |
12dd36fa | 1909 | } |
1da177e4 LT |
1910 | #endif |
1911 | ||
911851e6 RD |
1912 | /** |
1913 | * slab_destroy - destroy and release all objects in a slab | |
1914 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1915 | * @page: page pointer being destroyed |
911851e6 | 1916 | * |
8a7d9b43 WSH |
1917 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1918 | * Before calling the slab page must have been unlinked from the cache. The | |
1919 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1920 | */ |
8456a648 | 1921 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1922 | { |
7e007355 | 1923 | void *freelist; |
12dd36fa | 1924 | |
8456a648 JK |
1925 | freelist = page->freelist; |
1926 | slab_destroy_debugcheck(cachep, page); | |
1da177e4 | 1927 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
68126702 JK |
1928 | struct rcu_head *head; |
1929 | ||
1930 | /* | |
1931 | * RCU free overloads the RCU head over the LRU. | |
1932 | * slab_page has been overloeaded over the LRU, | |
1933 | * however it is not used from now on so that | |
1934 | * we can use it safely. | |
1935 | */ | |
1936 | head = (void *)&page->rcu_head; | |
1937 | call_rcu(head, kmem_rcu_free); | |
1da177e4 | 1938 | |
1da177e4 | 1939 | } else { |
0c3aa83e | 1940 | kmem_freepages(cachep, page); |
1da177e4 | 1941 | } |
68126702 JK |
1942 | |
1943 | /* | |
8456a648 | 1944 | * From now on, we don't use freelist |
68126702 JK |
1945 | * although actual page can be freed in rcu context |
1946 | */ | |
1947 | if (OFF_SLAB(cachep)) | |
8456a648 | 1948 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1949 | } |
1950 | ||
97654dfa JK |
1951 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1952 | { | |
1953 | struct page *page, *n; | |
1954 | ||
1955 | list_for_each_entry_safe(page, n, list, lru) { | |
1956 | list_del(&page->lru); | |
1957 | slab_destroy(cachep, page); | |
1958 | } | |
1959 | } | |
1960 | ||
4d268eba | 1961 | /** |
a70773dd RD |
1962 | * calculate_slab_order - calculate size (page order) of slabs |
1963 | * @cachep: pointer to the cache that is being created | |
1964 | * @size: size of objects to be created in this cache. | |
1965 | * @align: required alignment for the objects. | |
1966 | * @flags: slab allocation flags | |
1967 | * | |
1968 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1969 | * |
1970 | * This could be made much more intelligent. For now, try to avoid using | |
1971 | * high order pages for slabs. When the gfp() functions are more friendly | |
1972 | * towards high-order requests, this should be changed. | |
1973 | */ | |
a737b3e2 | 1974 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 1975 | size_t size, size_t align, unsigned long flags) |
4d268eba | 1976 | { |
b1ab41c4 | 1977 | unsigned long offslab_limit; |
4d268eba | 1978 | size_t left_over = 0; |
9888e6fa | 1979 | int gfporder; |
4d268eba | 1980 | |
0aa817f0 | 1981 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1982 | unsigned int num; |
1983 | size_t remainder; | |
1984 | ||
9888e6fa | 1985 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
1986 | if (!num) |
1987 | continue; | |
9888e6fa | 1988 | |
f315e3fa JK |
1989 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1990 | if (num > SLAB_OBJ_MAX_NUM) | |
1991 | break; | |
1992 | ||
b1ab41c4 | 1993 | if (flags & CFLGS_OFF_SLAB) { |
03787301 | 1994 | size_t freelist_size_per_obj = sizeof(freelist_idx_t); |
b1ab41c4 IM |
1995 | /* |
1996 | * Max number of objs-per-slab for caches which | |
1997 | * use off-slab slabs. Needed to avoid a possible | |
1998 | * looping condition in cache_grow(). | |
1999 | */ | |
03787301 JK |
2000 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
2001 | freelist_size_per_obj += sizeof(char); | |
8456a648 | 2002 | offslab_limit = size; |
03787301 | 2003 | offslab_limit /= freelist_size_per_obj; |
b1ab41c4 IM |
2004 | |
2005 | if (num > offslab_limit) | |
2006 | break; | |
2007 | } | |
4d268eba | 2008 | |
9888e6fa | 2009 | /* Found something acceptable - save it away */ |
4d268eba | 2010 | cachep->num = num; |
9888e6fa | 2011 | cachep->gfporder = gfporder; |
4d268eba PE |
2012 | left_over = remainder; |
2013 | ||
f78bb8ad LT |
2014 | /* |
2015 | * A VFS-reclaimable slab tends to have most allocations | |
2016 | * as GFP_NOFS and we really don't want to have to be allocating | |
2017 | * higher-order pages when we are unable to shrink dcache. | |
2018 | */ | |
2019 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
2020 | break; | |
2021 | ||
4d268eba PE |
2022 | /* |
2023 | * Large number of objects is good, but very large slabs are | |
2024 | * currently bad for the gfp()s. | |
2025 | */ | |
543585cc | 2026 | if (gfporder >= slab_max_order) |
4d268eba PE |
2027 | break; |
2028 | ||
9888e6fa LT |
2029 | /* |
2030 | * Acceptable internal fragmentation? | |
2031 | */ | |
a737b3e2 | 2032 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
2033 | break; |
2034 | } | |
2035 | return left_over; | |
2036 | } | |
2037 | ||
83b519e8 | 2038 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2039 | { |
97d06609 | 2040 | if (slab_state >= FULL) |
83b519e8 | 2041 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2042 | |
97d06609 | 2043 | if (slab_state == DOWN) { |
f30cf7d1 | 2044 | /* |
2f9baa9f | 2045 | * Note: Creation of first cache (kmem_cache). |
ce8eb6c4 | 2046 | * The setup_node is taken care |
2f9baa9f CL |
2047 | * of by the caller of __kmem_cache_create |
2048 | */ | |
2049 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2050 | slab_state = PARTIAL; | |
2051 | } else if (slab_state == PARTIAL) { | |
2052 | /* | |
2053 | * Note: the second kmem_cache_create must create the cache | |
f30cf7d1 PE |
2054 | * that's used by kmalloc(24), otherwise the creation of |
2055 | * further caches will BUG(). | |
2056 | */ | |
2057 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2058 | ||
2059 | /* | |
ce8eb6c4 CL |
2060 | * If the cache that's used by kmalloc(sizeof(kmem_cache_node)) is |
2061 | * the second cache, then we need to set up all its node/, | |
f30cf7d1 PE |
2062 | * otherwise the creation of further caches will BUG(). |
2063 | */ | |
ce8eb6c4 CL |
2064 | set_up_node(cachep, SIZE_AC); |
2065 | if (INDEX_AC == INDEX_NODE) | |
2066 | slab_state = PARTIAL_NODE; | |
f30cf7d1 | 2067 | else |
97d06609 | 2068 | slab_state = PARTIAL_ARRAYCACHE; |
f30cf7d1 | 2069 | } else { |
2f9baa9f | 2070 | /* Remaining boot caches */ |
f30cf7d1 | 2071 | cachep->array[smp_processor_id()] = |
83b519e8 | 2072 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 | 2073 | |
97d06609 | 2074 | if (slab_state == PARTIAL_ARRAYCACHE) { |
ce8eb6c4 CL |
2075 | set_up_node(cachep, SIZE_NODE); |
2076 | slab_state = PARTIAL_NODE; | |
f30cf7d1 PE |
2077 | } else { |
2078 | int node; | |
556a169d | 2079 | for_each_online_node(node) { |
6a67368c | 2080 | cachep->node[node] = |
6744f087 | 2081 | kmalloc_node(sizeof(struct kmem_cache_node), |
eb91f1d0 | 2082 | gfp, node); |
6a67368c | 2083 | BUG_ON(!cachep->node[node]); |
ce8eb6c4 | 2084 | kmem_cache_node_init(cachep->node[node]); |
f30cf7d1 PE |
2085 | } |
2086 | } | |
2087 | } | |
6a67368c | 2088 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
2089 | jiffies + REAPTIMEOUT_NODE + |
2090 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
2091 | |
2092 | cpu_cache_get(cachep)->avail = 0; | |
2093 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2094 | cpu_cache_get(cachep)->batchcount = 1; | |
2095 | cpu_cache_get(cachep)->touched = 0; | |
2096 | cachep->batchcount = 1; | |
2097 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2098 | return 0; |
f30cf7d1 PE |
2099 | } |
2100 | ||
1da177e4 | 2101 | /** |
039363f3 | 2102 | * __kmem_cache_create - Create a cache. |
a755b76a | 2103 | * @cachep: cache management descriptor |
1da177e4 | 2104 | * @flags: SLAB flags |
1da177e4 LT |
2105 | * |
2106 | * Returns a ptr to the cache on success, NULL on failure. | |
2107 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2108 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2109 | * |
1da177e4 LT |
2110 | * The flags are |
2111 | * | |
2112 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2113 | * to catch references to uninitialised memory. | |
2114 | * | |
2115 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2116 | * for buffer overruns. | |
2117 | * | |
1da177e4 LT |
2118 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2119 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2120 | * as davem. | |
2121 | */ | |
278b1bb1 | 2122 | int |
8a13a4cc | 2123 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2124 | { |
8456a648 | 2125 | size_t left_over, freelist_size, ralign; |
83b519e8 | 2126 | gfp_t gfp; |
278b1bb1 | 2127 | int err; |
8a13a4cc | 2128 | size_t size = cachep->size; |
1da177e4 | 2129 | |
1da177e4 | 2130 | #if DEBUG |
1da177e4 LT |
2131 | #if FORCED_DEBUG |
2132 | /* | |
2133 | * Enable redzoning and last user accounting, except for caches with | |
2134 | * large objects, if the increased size would increase the object size | |
2135 | * above the next power of two: caches with object sizes just above a | |
2136 | * power of two have a significant amount of internal fragmentation. | |
2137 | */ | |
87a927c7 DW |
2138 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2139 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2140 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2141 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2142 | flags |= SLAB_POISON; | |
2143 | #endif | |
2144 | if (flags & SLAB_DESTROY_BY_RCU) | |
2145 | BUG_ON(flags & SLAB_POISON); | |
2146 | #endif | |
1da177e4 | 2147 | |
a737b3e2 AM |
2148 | /* |
2149 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2150 | * unaligned accesses for some archs when redzoning is used, and makes |
2151 | * sure any on-slab bufctl's are also correctly aligned. | |
2152 | */ | |
b28a02de PE |
2153 | if (size & (BYTES_PER_WORD - 1)) { |
2154 | size += (BYTES_PER_WORD - 1); | |
2155 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2156 | } |
2157 | ||
ca5f9703 | 2158 | /* |
87a927c7 DW |
2159 | * Redzoning and user store require word alignment or possibly larger. |
2160 | * Note this will be overridden by architecture or caller mandated | |
2161 | * alignment if either is greater than BYTES_PER_WORD. | |
ca5f9703 | 2162 | */ |
87a927c7 DW |
2163 | if (flags & SLAB_STORE_USER) |
2164 | ralign = BYTES_PER_WORD; | |
2165 | ||
2166 | if (flags & SLAB_RED_ZONE) { | |
2167 | ralign = REDZONE_ALIGN; | |
2168 | /* If redzoning, ensure that the second redzone is suitably | |
2169 | * aligned, by adjusting the object size accordingly. */ | |
2170 | size += REDZONE_ALIGN - 1; | |
2171 | size &= ~(REDZONE_ALIGN - 1); | |
2172 | } | |
ca5f9703 | 2173 | |
a44b56d3 | 2174 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2175 | if (ralign < cachep->align) { |
2176 | ralign = cachep->align; | |
1da177e4 | 2177 | } |
3ff84a7f PE |
2178 | /* disable debug if necessary */ |
2179 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2180 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2181 | /* |
ca5f9703 | 2182 | * 4) Store it. |
1da177e4 | 2183 | */ |
8a13a4cc | 2184 | cachep->align = ralign; |
1da177e4 | 2185 | |
83b519e8 PE |
2186 | if (slab_is_available()) |
2187 | gfp = GFP_KERNEL; | |
2188 | else | |
2189 | gfp = GFP_NOWAIT; | |
2190 | ||
6a67368c | 2191 | setup_node_pointer(cachep); |
1da177e4 | 2192 | #if DEBUG |
1da177e4 | 2193 | |
ca5f9703 PE |
2194 | /* |
2195 | * Both debugging options require word-alignment which is calculated | |
2196 | * into align above. | |
2197 | */ | |
1da177e4 | 2198 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2199 | /* add space for red zone words */ |
3ff84a7f PE |
2200 | cachep->obj_offset += sizeof(unsigned long long); |
2201 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2202 | } |
2203 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2204 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2205 | * the real object. But if the second red zone needs to be |
2206 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2207 | */ |
87a927c7 DW |
2208 | if (flags & SLAB_RED_ZONE) |
2209 | size += REDZONE_ALIGN; | |
2210 | else | |
2211 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2212 | } |
2213 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
ce8eb6c4 | 2214 | if (size >= kmalloc_size(INDEX_NODE + 1) |
608da7e3 TH |
2215 | && cachep->object_size > cache_line_size() |
2216 | && ALIGN(size, cachep->align) < PAGE_SIZE) { | |
2217 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); | |
1da177e4 LT |
2218 | size = PAGE_SIZE; |
2219 | } | |
2220 | #endif | |
2221 | #endif | |
2222 | ||
e0a42726 IM |
2223 | /* |
2224 | * Determine if the slab management is 'on' or 'off' slab. | |
2225 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2226 | * it too early on. Always use on-slab management when |
2227 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2228 | */ |
8fc9cf42 | 2229 | if ((size >= (PAGE_SIZE >> 5)) && !slab_early_init && |
e7cb55b9 | 2230 | !(flags & SLAB_NOLEAKTRACE)) |
1da177e4 LT |
2231 | /* |
2232 | * Size is large, assume best to place the slab management obj | |
2233 | * off-slab (should allow better packing of objs). | |
2234 | */ | |
2235 | flags |= CFLGS_OFF_SLAB; | |
2236 | ||
8a13a4cc | 2237 | size = ALIGN(size, cachep->align); |
f315e3fa JK |
2238 | /* |
2239 | * We should restrict the number of objects in a slab to implement | |
2240 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2241 | */ | |
2242 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2243 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
1da177e4 | 2244 | |
8a13a4cc | 2245 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2246 | |
8a13a4cc | 2247 | if (!cachep->num) |
278b1bb1 | 2248 | return -E2BIG; |
1da177e4 | 2249 | |
03787301 | 2250 | freelist_size = calculate_freelist_size(cachep->num, cachep->align); |
1da177e4 LT |
2251 | |
2252 | /* | |
2253 | * If the slab has been placed off-slab, and we have enough space then | |
2254 | * move it on-slab. This is at the expense of any extra colouring. | |
2255 | */ | |
8456a648 | 2256 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
1da177e4 | 2257 | flags &= ~CFLGS_OFF_SLAB; |
8456a648 | 2258 | left_over -= freelist_size; |
1da177e4 LT |
2259 | } |
2260 | ||
2261 | if (flags & CFLGS_OFF_SLAB) { | |
2262 | /* really off slab. No need for manual alignment */ | |
03787301 | 2263 | freelist_size = calculate_freelist_size(cachep->num, 0); |
67461365 RL |
2264 | |
2265 | #ifdef CONFIG_PAGE_POISONING | |
2266 | /* If we're going to use the generic kernel_map_pages() | |
2267 | * poisoning, then it's going to smash the contents of | |
2268 | * the redzone and userword anyhow, so switch them off. | |
2269 | */ | |
2270 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2271 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2272 | #endif | |
1da177e4 LT |
2273 | } |
2274 | ||
2275 | cachep->colour_off = cache_line_size(); | |
2276 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2277 | if (cachep->colour_off < cachep->align) |
2278 | cachep->colour_off = cachep->align; | |
b28a02de | 2279 | cachep->colour = left_over / cachep->colour_off; |
8456a648 | 2280 | cachep->freelist_size = freelist_size; |
1da177e4 | 2281 | cachep->flags = flags; |
a57a4988 | 2282 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2283 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2284 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2285 | cachep->size = size; |
6a2d7a95 | 2286 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2287 | |
e5ac9c5a | 2288 | if (flags & CFLGS_OFF_SLAB) { |
8456a648 | 2289 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
e5ac9c5a | 2290 | /* |
5f0985bb | 2291 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
e5ac9c5a | 2292 | * But since we go off slab only for object size greater than |
5f0985bb JZ |
2293 | * PAGE_SIZE/8, and kmalloc_{dma,}_caches get created |
2294 | * in ascending order,this should not happen at all. | |
e5ac9c5a RT |
2295 | * But leave a BUG_ON for some lucky dude. |
2296 | */ | |
8456a648 | 2297 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
e5ac9c5a | 2298 | } |
1da177e4 | 2299 | |
278b1bb1 CL |
2300 | err = setup_cpu_cache(cachep, gfp); |
2301 | if (err) { | |
12c3667f | 2302 | __kmem_cache_shutdown(cachep); |
278b1bb1 | 2303 | return err; |
2ed3a4ef | 2304 | } |
1da177e4 | 2305 | |
278b1bb1 | 2306 | return 0; |
1da177e4 | 2307 | } |
1da177e4 LT |
2308 | |
2309 | #if DEBUG | |
2310 | static void check_irq_off(void) | |
2311 | { | |
2312 | BUG_ON(!irqs_disabled()); | |
2313 | } | |
2314 | ||
2315 | static void check_irq_on(void) | |
2316 | { | |
2317 | BUG_ON(irqs_disabled()); | |
2318 | } | |
2319 | ||
343e0d7a | 2320 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2321 | { |
2322 | #ifdef CONFIG_SMP | |
2323 | check_irq_off(); | |
18bf8541 | 2324 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2325 | #endif |
2326 | } | |
e498be7d | 2327 | |
343e0d7a | 2328 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2329 | { |
2330 | #ifdef CONFIG_SMP | |
2331 | check_irq_off(); | |
18bf8541 | 2332 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2333 | #endif |
2334 | } | |
2335 | ||
1da177e4 LT |
2336 | #else |
2337 | #define check_irq_off() do { } while(0) | |
2338 | #define check_irq_on() do { } while(0) | |
2339 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2340 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2341 | #endif |
2342 | ||
ce8eb6c4 | 2343 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2344 | struct array_cache *ac, |
2345 | int force, int node); | |
2346 | ||
1da177e4 LT |
2347 | static void do_drain(void *arg) |
2348 | { | |
a737b3e2 | 2349 | struct kmem_cache *cachep = arg; |
1da177e4 | 2350 | struct array_cache *ac; |
7d6e6d09 | 2351 | int node = numa_mem_id(); |
18bf8541 | 2352 | struct kmem_cache_node *n; |
97654dfa | 2353 | LIST_HEAD(list); |
1da177e4 LT |
2354 | |
2355 | check_irq_off(); | |
9a2dba4b | 2356 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2357 | n = get_node(cachep, node); |
2358 | spin_lock(&n->list_lock); | |
97654dfa | 2359 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2360 | spin_unlock(&n->list_lock); |
97654dfa | 2361 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2362 | ac->avail = 0; |
2363 | } | |
2364 | ||
343e0d7a | 2365 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2366 | { |
ce8eb6c4 | 2367 | struct kmem_cache_node *n; |
e498be7d CL |
2368 | int node; |
2369 | ||
15c8b6c1 | 2370 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2371 | check_irq_on(); |
18bf8541 CL |
2372 | for_each_kmem_cache_node(cachep, node, n) |
2373 | if (n->alien) | |
ce8eb6c4 | 2374 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2375 | |
18bf8541 CL |
2376 | for_each_kmem_cache_node(cachep, node, n) |
2377 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2378 | } |
2379 | ||
ed11d9eb CL |
2380 | /* |
2381 | * Remove slabs from the list of free slabs. | |
2382 | * Specify the number of slabs to drain in tofree. | |
2383 | * | |
2384 | * Returns the actual number of slabs released. | |
2385 | */ | |
2386 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2387 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2388 | { |
ed11d9eb CL |
2389 | struct list_head *p; |
2390 | int nr_freed; | |
8456a648 | 2391 | struct page *page; |
1da177e4 | 2392 | |
ed11d9eb | 2393 | nr_freed = 0; |
ce8eb6c4 | 2394 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2395 | |
ce8eb6c4 CL |
2396 | spin_lock_irq(&n->list_lock); |
2397 | p = n->slabs_free.prev; | |
2398 | if (p == &n->slabs_free) { | |
2399 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2400 | goto out; |
2401 | } | |
1da177e4 | 2402 | |
8456a648 | 2403 | page = list_entry(p, struct page, lru); |
1da177e4 | 2404 | #if DEBUG |
8456a648 | 2405 | BUG_ON(page->active); |
1da177e4 | 2406 | #endif |
8456a648 | 2407 | list_del(&page->lru); |
ed11d9eb CL |
2408 | /* |
2409 | * Safe to drop the lock. The slab is no longer linked | |
2410 | * to the cache. | |
2411 | */ | |
ce8eb6c4 CL |
2412 | n->free_objects -= cache->num; |
2413 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2414 | slab_destroy(cache, page); |
ed11d9eb | 2415 | nr_freed++; |
1da177e4 | 2416 | } |
ed11d9eb CL |
2417 | out: |
2418 | return nr_freed; | |
1da177e4 LT |
2419 | } |
2420 | ||
03afc0e2 | 2421 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d | 2422 | { |
18bf8541 CL |
2423 | int ret = 0; |
2424 | int node; | |
ce8eb6c4 | 2425 | struct kmem_cache_node *n; |
e498be7d CL |
2426 | |
2427 | drain_cpu_caches(cachep); | |
2428 | ||
2429 | check_irq_on(); | |
18bf8541 | 2430 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2431 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2432 | |
ce8eb6c4 CL |
2433 | ret += !list_empty(&n->slabs_full) || |
2434 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2435 | } |
2436 | return (ret ? 1 : 0); | |
2437 | } | |
2438 | ||
945cf2b6 | 2439 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
1da177e4 | 2440 | { |
12c3667f | 2441 | int i; |
ce8eb6c4 | 2442 | struct kmem_cache_node *n; |
03afc0e2 | 2443 | int rc = __kmem_cache_shrink(cachep); |
1da177e4 | 2444 | |
12c3667f CL |
2445 | if (rc) |
2446 | return rc; | |
1da177e4 | 2447 | |
12c3667f CL |
2448 | for_each_online_cpu(i) |
2449 | kfree(cachep->array[i]); | |
1da177e4 | 2450 | |
ce8eb6c4 | 2451 | /* NUMA: free the node structures */ |
18bf8541 CL |
2452 | for_each_kmem_cache_node(cachep, i, n) { |
2453 | kfree(n->shared); | |
2454 | free_alien_cache(n->alien); | |
2455 | kfree(n); | |
2456 | cachep->node[i] = NULL; | |
12c3667f CL |
2457 | } |
2458 | return 0; | |
1da177e4 | 2459 | } |
1da177e4 | 2460 | |
e5ac9c5a RT |
2461 | /* |
2462 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2463 | * |
2464 | * For a slab cache when the slab descriptor is off-slab, the | |
2465 | * slab descriptor can't come from the same cache which is being created, | |
2466 | * Because if it is the case, that means we defer the creation of | |
2467 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2468 | * And we eventually call down to __kmem_cache_create(), which | |
2469 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2470 | * This is a "chicken-and-egg" problem. | |
2471 | * | |
2472 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2473 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2474 | */ |
7e007355 | 2475 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2476 | struct page *page, int colour_off, |
2477 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2478 | { |
7e007355 | 2479 | void *freelist; |
0c3aa83e | 2480 | void *addr = page_address(page); |
b28a02de | 2481 | |
1da177e4 LT |
2482 | if (OFF_SLAB(cachep)) { |
2483 | /* Slab management obj is off-slab. */ | |
8456a648 | 2484 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2485 | local_flags, nodeid); |
8456a648 | 2486 | if (!freelist) |
1da177e4 LT |
2487 | return NULL; |
2488 | } else { | |
8456a648 JK |
2489 | freelist = addr + colour_off; |
2490 | colour_off += cachep->freelist_size; | |
1da177e4 | 2491 | } |
8456a648 JK |
2492 | page->active = 0; |
2493 | page->s_mem = addr + colour_off; | |
2494 | return freelist; | |
1da177e4 LT |
2495 | } |
2496 | ||
7cc68973 | 2497 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2498 | { |
a41adfaa | 2499 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2500 | } |
2501 | ||
2502 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2503 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2504 | { |
a41adfaa | 2505 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2506 | } |
2507 | ||
343e0d7a | 2508 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2509 | struct page *page) |
1da177e4 LT |
2510 | { |
2511 | int i; | |
2512 | ||
2513 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2514 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
2515 | #if DEBUG |
2516 | /* need to poison the objs? */ | |
2517 | if (cachep->flags & SLAB_POISON) | |
2518 | poison_obj(cachep, objp, POISON_FREE); | |
2519 | if (cachep->flags & SLAB_STORE_USER) | |
2520 | *dbg_userword(cachep, objp) = NULL; | |
2521 | ||
2522 | if (cachep->flags & SLAB_RED_ZONE) { | |
2523 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2524 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2525 | } | |
2526 | /* | |
a737b3e2 AM |
2527 | * Constructors are not allowed to allocate memory from the same |
2528 | * cache which they are a constructor for. Otherwise, deadlock. | |
2529 | * They must also be threaded. | |
1da177e4 LT |
2530 | */ |
2531 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2532 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2533 | |
2534 | if (cachep->flags & SLAB_RED_ZONE) { | |
2535 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2536 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2537 | " end of an object"); |
1da177e4 LT |
2538 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2539 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2540 | " start of an object"); |
1da177e4 | 2541 | } |
3b0efdfa | 2542 | if ((cachep->size % PAGE_SIZE) == 0 && |
a737b3e2 | 2543 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
b28a02de | 2544 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2545 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2546 | #else |
2547 | if (cachep->ctor) | |
51cc5068 | 2548 | cachep->ctor(objp); |
1da177e4 | 2549 | #endif |
03787301 | 2550 | set_obj_status(page, i, OBJECT_FREE); |
e5c58dfd | 2551 | set_free_obj(page, i, i); |
1da177e4 | 2552 | } |
1da177e4 LT |
2553 | } |
2554 | ||
343e0d7a | 2555 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2556 | { |
4b51d669 CL |
2557 | if (CONFIG_ZONE_DMA_FLAG) { |
2558 | if (flags & GFP_DMA) | |
a618e89f | 2559 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2560 | else |
a618e89f | 2561 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2562 | } |
1da177e4 LT |
2563 | } |
2564 | ||
8456a648 | 2565 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2566 | int nodeid) |
78d382d7 | 2567 | { |
b1cb0982 | 2568 | void *objp; |
78d382d7 | 2569 | |
e5c58dfd | 2570 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2571 | page->active++; |
78d382d7 | 2572 | #if DEBUG |
1ea991b0 | 2573 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2574 | #endif |
78d382d7 MD |
2575 | |
2576 | return objp; | |
2577 | } | |
2578 | ||
8456a648 | 2579 | static void slab_put_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2580 | void *objp, int nodeid) |
78d382d7 | 2581 | { |
8456a648 | 2582 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2583 | #if DEBUG |
16025177 | 2584 | unsigned int i; |
b1cb0982 | 2585 | |
78d382d7 | 2586 | /* Verify that the slab belongs to the intended node */ |
1ea991b0 | 2587 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2588 | |
b1cb0982 | 2589 | /* Verify double free bug */ |
8456a648 | 2590 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2591 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2592 | printk(KERN_ERR "slab: double free detected in cache " |
2593 | "'%s', objp %p\n", cachep->name, objp); | |
2594 | BUG(); | |
2595 | } | |
78d382d7 MD |
2596 | } |
2597 | #endif | |
8456a648 | 2598 | page->active--; |
e5c58dfd | 2599 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2600 | } |
2601 | ||
4776874f PE |
2602 | /* |
2603 | * Map pages beginning at addr to the given cache and slab. This is required | |
2604 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2605 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2606 | */ |
8456a648 | 2607 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2608 | void *freelist) |
1da177e4 | 2609 | { |
a57a4988 | 2610 | page->slab_cache = cache; |
8456a648 | 2611 | page->freelist = freelist; |
1da177e4 LT |
2612 | } |
2613 | ||
2614 | /* | |
2615 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2616 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2617 | */ | |
3c517a61 | 2618 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2619 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2620 | { |
7e007355 | 2621 | void *freelist; |
b28a02de PE |
2622 | size_t offset; |
2623 | gfp_t local_flags; | |
ce8eb6c4 | 2624 | struct kmem_cache_node *n; |
1da177e4 | 2625 | |
a737b3e2 AM |
2626 | /* |
2627 | * Be lazy and only check for valid flags here, keeping it out of the | |
2628 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2629 | */ |
6cb06229 CL |
2630 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2631 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2632 | |
ce8eb6c4 | 2633 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2634 | check_irq_off(); |
18bf8541 | 2635 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2636 | spin_lock(&n->list_lock); |
1da177e4 LT |
2637 | |
2638 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2639 | offset = n->colour_next; |
2640 | n->colour_next++; | |
2641 | if (n->colour_next >= cachep->colour) | |
2642 | n->colour_next = 0; | |
2643 | spin_unlock(&n->list_lock); | |
1da177e4 | 2644 | |
2e1217cf | 2645 | offset *= cachep->colour_off; |
1da177e4 LT |
2646 | |
2647 | if (local_flags & __GFP_WAIT) | |
2648 | local_irq_enable(); | |
2649 | ||
2650 | /* | |
2651 | * The test for missing atomic flag is performed here, rather than | |
2652 | * the more obvious place, simply to reduce the critical path length | |
2653 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2654 | * will eventually be caught here (where it matters). | |
2655 | */ | |
2656 | kmem_flagcheck(cachep, flags); | |
2657 | ||
a737b3e2 AM |
2658 | /* |
2659 | * Get mem for the objs. Attempt to allocate a physical page from | |
2660 | * 'nodeid'. | |
e498be7d | 2661 | */ |
0c3aa83e JK |
2662 | if (!page) |
2663 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2664 | if (!page) | |
1da177e4 LT |
2665 | goto failed; |
2666 | ||
2667 | /* Get slab management. */ | |
8456a648 | 2668 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2669 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2670 | if (!freelist) |
1da177e4 LT |
2671 | goto opps1; |
2672 | ||
8456a648 | 2673 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2674 | |
8456a648 | 2675 | cache_init_objs(cachep, page); |
1da177e4 LT |
2676 | |
2677 | if (local_flags & __GFP_WAIT) | |
2678 | local_irq_disable(); | |
2679 | check_irq_off(); | |
ce8eb6c4 | 2680 | spin_lock(&n->list_lock); |
1da177e4 LT |
2681 | |
2682 | /* Make slab active. */ | |
8456a648 | 2683 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2684 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2685 | n->free_objects += cachep->num; |
2686 | spin_unlock(&n->list_lock); | |
1da177e4 | 2687 | return 1; |
a737b3e2 | 2688 | opps1: |
0c3aa83e | 2689 | kmem_freepages(cachep, page); |
a737b3e2 | 2690 | failed: |
1da177e4 LT |
2691 | if (local_flags & __GFP_WAIT) |
2692 | local_irq_disable(); | |
2693 | return 0; | |
2694 | } | |
2695 | ||
2696 | #if DEBUG | |
2697 | ||
2698 | /* | |
2699 | * Perform extra freeing checks: | |
2700 | * - detect bad pointers. | |
2701 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2702 | */ |
2703 | static void kfree_debugcheck(const void *objp) | |
2704 | { | |
1da177e4 LT |
2705 | if (!virt_addr_valid(objp)) { |
2706 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2707 | (unsigned long)objp); |
2708 | BUG(); | |
1da177e4 | 2709 | } |
1da177e4 LT |
2710 | } |
2711 | ||
58ce1fd5 PE |
2712 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2713 | { | |
b46b8f19 | 2714 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2715 | |
2716 | redzone1 = *dbg_redzone1(cache, obj); | |
2717 | redzone2 = *dbg_redzone2(cache, obj); | |
2718 | ||
2719 | /* | |
2720 | * Redzone is ok. | |
2721 | */ | |
2722 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2723 | return; | |
2724 | ||
2725 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2726 | slab_error(cache, "double free detected"); | |
2727 | else | |
2728 | slab_error(cache, "memory outside object was overwritten"); | |
2729 | ||
b46b8f19 | 2730 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2731 | obj, redzone1, redzone2); |
2732 | } | |
2733 | ||
343e0d7a | 2734 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2735 | unsigned long caller) |
1da177e4 | 2736 | { |
1da177e4 | 2737 | unsigned int objnr; |
8456a648 | 2738 | struct page *page; |
1da177e4 | 2739 | |
80cbd911 MW |
2740 | BUG_ON(virt_to_cache(objp) != cachep); |
2741 | ||
3dafccf2 | 2742 | objp -= obj_offset(cachep); |
1da177e4 | 2743 | kfree_debugcheck(objp); |
b49af68f | 2744 | page = virt_to_head_page(objp); |
1da177e4 | 2745 | |
1da177e4 | 2746 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2747 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2748 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2749 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2750 | } | |
2751 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2752 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2753 | |
8456a648 | 2754 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2755 | |
2756 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2757 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2758 | |
03787301 | 2759 | set_obj_status(page, objnr, OBJECT_FREE); |
1da177e4 LT |
2760 | if (cachep->flags & SLAB_POISON) { |
2761 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2762 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
7c0cb9c6 | 2763 | store_stackinfo(cachep, objp, caller); |
b28a02de | 2764 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2765 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2766 | } else { |
2767 | poison_obj(cachep, objp, POISON_FREE); | |
2768 | } | |
2769 | #else | |
2770 | poison_obj(cachep, objp, POISON_FREE); | |
2771 | #endif | |
2772 | } | |
2773 | return objp; | |
2774 | } | |
2775 | ||
1da177e4 LT |
2776 | #else |
2777 | #define kfree_debugcheck(x) do { } while(0) | |
2778 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2779 | #endif |
2780 | ||
072bb0aa MG |
2781 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2782 | bool force_refill) | |
1da177e4 LT |
2783 | { |
2784 | int batchcount; | |
ce8eb6c4 | 2785 | struct kmem_cache_node *n; |
1da177e4 | 2786 | struct array_cache *ac; |
1ca4cb24 PE |
2787 | int node; |
2788 | ||
1da177e4 | 2789 | check_irq_off(); |
7d6e6d09 | 2790 | node = numa_mem_id(); |
072bb0aa MG |
2791 | if (unlikely(force_refill)) |
2792 | goto force_grow; | |
2793 | retry: | |
9a2dba4b | 2794 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2795 | batchcount = ac->batchcount; |
2796 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2797 | /* |
2798 | * If there was little recent activity on this cache, then | |
2799 | * perform only a partial refill. Otherwise we could generate | |
2800 | * refill bouncing. | |
1da177e4 LT |
2801 | */ |
2802 | batchcount = BATCHREFILL_LIMIT; | |
2803 | } | |
18bf8541 | 2804 | n = get_node(cachep, node); |
e498be7d | 2805 | |
ce8eb6c4 CL |
2806 | BUG_ON(ac->avail > 0 || !n); |
2807 | spin_lock(&n->list_lock); | |
1da177e4 | 2808 | |
3ded175a | 2809 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2810 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2811 | n->shared->touched = 1; | |
3ded175a | 2812 | goto alloc_done; |
44b57f1c | 2813 | } |
3ded175a | 2814 | |
1da177e4 LT |
2815 | while (batchcount > 0) { |
2816 | struct list_head *entry; | |
8456a648 | 2817 | struct page *page; |
1da177e4 | 2818 | /* Get slab alloc is to come from. */ |
ce8eb6c4 CL |
2819 | entry = n->slabs_partial.next; |
2820 | if (entry == &n->slabs_partial) { | |
2821 | n->free_touched = 1; | |
2822 | entry = n->slabs_free.next; | |
2823 | if (entry == &n->slabs_free) | |
1da177e4 LT |
2824 | goto must_grow; |
2825 | } | |
2826 | ||
8456a648 | 2827 | page = list_entry(entry, struct page, lru); |
1da177e4 | 2828 | check_spinlock_acquired(cachep); |
714b8171 PE |
2829 | |
2830 | /* | |
2831 | * The slab was either on partial or free list so | |
2832 | * there must be at least one object available for | |
2833 | * allocation. | |
2834 | */ | |
8456a648 | 2835 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2836 | |
8456a648 | 2837 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2838 | STATS_INC_ALLOCED(cachep); |
2839 | STATS_INC_ACTIVE(cachep); | |
2840 | STATS_SET_HIGH(cachep); | |
2841 | ||
8456a648 | 2842 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page, |
072bb0aa | 2843 | node)); |
1da177e4 | 2844 | } |
1da177e4 LT |
2845 | |
2846 | /* move slabp to correct slabp list: */ | |
8456a648 JK |
2847 | list_del(&page->lru); |
2848 | if (page->active == cachep->num) | |
34bf6ef9 | 2849 | list_add(&page->lru, &n->slabs_full); |
1da177e4 | 2850 | else |
34bf6ef9 | 2851 | list_add(&page->lru, &n->slabs_partial); |
1da177e4 LT |
2852 | } |
2853 | ||
a737b3e2 | 2854 | must_grow: |
ce8eb6c4 | 2855 | n->free_objects -= ac->avail; |
a737b3e2 | 2856 | alloc_done: |
ce8eb6c4 | 2857 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2858 | |
2859 | if (unlikely(!ac->avail)) { | |
2860 | int x; | |
072bb0aa | 2861 | force_grow: |
3c517a61 | 2862 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 2863 | |
a737b3e2 | 2864 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2865 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2866 | node = numa_mem_id(); |
072bb0aa MG |
2867 | |
2868 | /* no objects in sight? abort */ | |
2869 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2870 | return NULL; |
2871 | ||
a737b3e2 | 2872 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2873 | goto retry; |
2874 | } | |
2875 | ac->touched = 1; | |
072bb0aa MG |
2876 | |
2877 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2878 | } |
2879 | ||
a737b3e2 AM |
2880 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2881 | gfp_t flags) | |
1da177e4 LT |
2882 | { |
2883 | might_sleep_if(flags & __GFP_WAIT); | |
2884 | #if DEBUG | |
2885 | kmem_flagcheck(cachep, flags); | |
2886 | #endif | |
2887 | } | |
2888 | ||
2889 | #if DEBUG | |
a737b3e2 | 2890 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2891 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2892 | { |
03787301 JK |
2893 | struct page *page; |
2894 | ||
b28a02de | 2895 | if (!objp) |
1da177e4 | 2896 | return objp; |
b28a02de | 2897 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2898 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3b0efdfa | 2899 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 2900 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2901 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
2902 | else |
2903 | check_poison_obj(cachep, objp); | |
2904 | #else | |
2905 | check_poison_obj(cachep, objp); | |
2906 | #endif | |
2907 | poison_obj(cachep, objp, POISON_INUSE); | |
2908 | } | |
2909 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2910 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2911 | |
2912 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2913 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2914 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2915 | slab_error(cachep, "double free, or memory outside" | |
2916 | " object was overwritten"); | |
b28a02de | 2917 | printk(KERN_ERR |
b46b8f19 | 2918 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2919 | objp, *dbg_redzone1(cachep, objp), |
2920 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2921 | } |
2922 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2923 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2924 | } | |
03787301 JK |
2925 | |
2926 | page = virt_to_head_page(objp); | |
2927 | set_obj_status(page, obj_to_index(cachep, page, objp), OBJECT_ACTIVE); | |
3dafccf2 | 2928 | objp += obj_offset(cachep); |
4f104934 | 2929 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2930 | cachep->ctor(objp); |
7ea466f2 TH |
2931 | if (ARCH_SLAB_MINALIGN && |
2932 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 2933 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2934 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2935 | } |
1da177e4 LT |
2936 | return objp; |
2937 | } | |
2938 | #else | |
2939 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2940 | #endif | |
2941 | ||
773ff60e | 2942 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 | 2943 | { |
8a9c61d4 | 2944 | if (unlikely(cachep == kmem_cache)) |
773ff60e | 2945 | return false; |
8a8b6502 | 2946 | |
8c138bc0 | 2947 | return should_failslab(cachep->object_size, flags, cachep->flags); |
8a8b6502 AM |
2948 | } |
2949 | ||
343e0d7a | 2950 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2951 | { |
b28a02de | 2952 | void *objp; |
1da177e4 | 2953 | struct array_cache *ac; |
072bb0aa | 2954 | bool force_refill = false; |
1da177e4 | 2955 | |
5c382300 | 2956 | check_irq_off(); |
8a8b6502 | 2957 | |
9a2dba4b | 2958 | ac = cpu_cache_get(cachep); |
1da177e4 | 2959 | if (likely(ac->avail)) { |
1da177e4 | 2960 | ac->touched = 1; |
072bb0aa MG |
2961 | objp = ac_get_obj(cachep, ac, flags, false); |
2962 | ||
ddbf2e83 | 2963 | /* |
072bb0aa MG |
2964 | * Allow for the possibility all avail objects are not allowed |
2965 | * by the current flags | |
ddbf2e83 | 2966 | */ |
072bb0aa MG |
2967 | if (objp) { |
2968 | STATS_INC_ALLOCHIT(cachep); | |
2969 | goto out; | |
2970 | } | |
2971 | force_refill = true; | |
1da177e4 | 2972 | } |
072bb0aa MG |
2973 | |
2974 | STATS_INC_ALLOCMISS(cachep); | |
2975 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
2976 | /* | |
2977 | * the 'ac' may be updated by cache_alloc_refill(), | |
2978 | * and kmemleak_erase() requires its correct value. | |
2979 | */ | |
2980 | ac = cpu_cache_get(cachep); | |
2981 | ||
2982 | out: | |
d5cff635 CM |
2983 | /* |
2984 | * To avoid a false negative, if an object that is in one of the | |
2985 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2986 | * treat the array pointers as a reference to the object. | |
2987 | */ | |
f3d8b53a O |
2988 | if (objp) |
2989 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2990 | return objp; |
2991 | } | |
2992 | ||
e498be7d | 2993 | #ifdef CONFIG_NUMA |
c61afb18 | 2994 | /* |
f0432d15 | 2995 | * Try allocating on another node if PF_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2996 | * |
2997 | * If we are in_interrupt, then process context, including cpusets and | |
2998 | * mempolicy, may not apply and should not be used for allocation policy. | |
2999 | */ | |
3000 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3001 | { | |
3002 | int nid_alloc, nid_here; | |
3003 | ||
765c4507 | 3004 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3005 | return NULL; |
7d6e6d09 | 3006 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3007 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3008 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3009 | else if (current->mempolicy) |
2a389610 | 3010 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3011 | if (nid_alloc != nid_here) |
8b98c169 | 3012 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3013 | return NULL; |
3014 | } | |
3015 | ||
765c4507 CL |
3016 | /* |
3017 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3018 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3019 | * available node for available objects. If that fails then we |
3c517a61 CL |
3020 | * perform an allocation without specifying a node. This allows the page |
3021 | * allocator to do its reclaim / fallback magic. We then insert the | |
3022 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3023 | */ |
8c8cc2c1 | 3024 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3025 | { |
8c8cc2c1 PE |
3026 | struct zonelist *zonelist; |
3027 | gfp_t local_flags; | |
dd1a239f | 3028 | struct zoneref *z; |
54a6eb5c MG |
3029 | struct zone *zone; |
3030 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3031 | void *obj = NULL; |
3c517a61 | 3032 | int nid; |
cc9a6c87 | 3033 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3034 | |
3035 | if (flags & __GFP_THISNODE) | |
3036 | return NULL; | |
3037 | ||
6cb06229 | 3038 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3039 | |
cc9a6c87 | 3040 | retry_cpuset: |
d26914d1 | 3041 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3042 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3043 | |
3c517a61 CL |
3044 | retry: |
3045 | /* | |
3046 | * Look through allowed nodes for objects available | |
3047 | * from existing per node queues. | |
3048 | */ | |
54a6eb5c MG |
3049 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3050 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3051 | |
54a6eb5c | 3052 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
18bf8541 CL |
3053 | get_node(cache, nid) && |
3054 | get_node(cache, nid)->free_objects) { | |
3c517a61 CL |
3055 | obj = ____cache_alloc_node(cache, |
3056 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3057 | if (obj) |
3058 | break; | |
3059 | } | |
3c517a61 CL |
3060 | } |
3061 | ||
cfce6604 | 3062 | if (!obj) { |
3c517a61 CL |
3063 | /* |
3064 | * This allocation will be performed within the constraints | |
3065 | * of the current cpuset / memory policy requirements. | |
3066 | * We may trigger various forms of reclaim on the allowed | |
3067 | * set and go into memory reserves if necessary. | |
3068 | */ | |
0c3aa83e JK |
3069 | struct page *page; |
3070 | ||
dd47ea75 CL |
3071 | if (local_flags & __GFP_WAIT) |
3072 | local_irq_enable(); | |
3073 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3074 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
dd47ea75 CL |
3075 | if (local_flags & __GFP_WAIT) |
3076 | local_irq_disable(); | |
0c3aa83e | 3077 | if (page) { |
3c517a61 CL |
3078 | /* |
3079 | * Insert into the appropriate per node queues | |
3080 | */ | |
0c3aa83e JK |
3081 | nid = page_to_nid(page); |
3082 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 CL |
3083 | obj = ____cache_alloc_node(cache, |
3084 | flags | GFP_THISNODE, nid); | |
3085 | if (!obj) | |
3086 | /* | |
3087 | * Another processor may allocate the | |
3088 | * objects in the slab since we are | |
3089 | * not holding any locks. | |
3090 | */ | |
3091 | goto retry; | |
3092 | } else { | |
b6a60451 | 3093 | /* cache_grow already freed obj */ |
3c517a61 CL |
3094 | obj = NULL; |
3095 | } | |
3096 | } | |
aedb0eb1 | 3097 | } |
cc9a6c87 | 3098 | |
d26914d1 | 3099 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3100 | goto retry_cpuset; |
765c4507 CL |
3101 | return obj; |
3102 | } | |
3103 | ||
e498be7d CL |
3104 | /* |
3105 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3106 | */ |
8b98c169 | 3107 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3108 | int nodeid) |
e498be7d CL |
3109 | { |
3110 | struct list_head *entry; | |
8456a648 | 3111 | struct page *page; |
ce8eb6c4 | 3112 | struct kmem_cache_node *n; |
b28a02de | 3113 | void *obj; |
b28a02de PE |
3114 | int x; |
3115 | ||
14e50c6a | 3116 | VM_BUG_ON(nodeid > num_online_nodes()); |
18bf8541 | 3117 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3118 | BUG_ON(!n); |
b28a02de | 3119 | |
a737b3e2 | 3120 | retry: |
ca3b9b91 | 3121 | check_irq_off(); |
ce8eb6c4 CL |
3122 | spin_lock(&n->list_lock); |
3123 | entry = n->slabs_partial.next; | |
3124 | if (entry == &n->slabs_partial) { | |
3125 | n->free_touched = 1; | |
3126 | entry = n->slabs_free.next; | |
3127 | if (entry == &n->slabs_free) | |
b28a02de PE |
3128 | goto must_grow; |
3129 | } | |
3130 | ||
8456a648 | 3131 | page = list_entry(entry, struct page, lru); |
b28a02de | 3132 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3133 | |
3134 | STATS_INC_NODEALLOCS(cachep); | |
3135 | STATS_INC_ACTIVE(cachep); | |
3136 | STATS_SET_HIGH(cachep); | |
3137 | ||
8456a648 | 3138 | BUG_ON(page->active == cachep->num); |
b28a02de | 3139 | |
8456a648 | 3140 | obj = slab_get_obj(cachep, page, nodeid); |
ce8eb6c4 | 3141 | n->free_objects--; |
b28a02de | 3142 | /* move slabp to correct slabp list: */ |
8456a648 | 3143 | list_del(&page->lru); |
b28a02de | 3144 | |
8456a648 JK |
3145 | if (page->active == cachep->num) |
3146 | list_add(&page->lru, &n->slabs_full); | |
a737b3e2 | 3147 | else |
8456a648 | 3148 | list_add(&page->lru, &n->slabs_partial); |
e498be7d | 3149 | |
ce8eb6c4 | 3150 | spin_unlock(&n->list_lock); |
b28a02de | 3151 | goto done; |
e498be7d | 3152 | |
a737b3e2 | 3153 | must_grow: |
ce8eb6c4 | 3154 | spin_unlock(&n->list_lock); |
3c517a61 | 3155 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3156 | if (x) |
3157 | goto retry; | |
1da177e4 | 3158 | |
8c8cc2c1 | 3159 | return fallback_alloc(cachep, flags); |
e498be7d | 3160 | |
a737b3e2 | 3161 | done: |
b28a02de | 3162 | return obj; |
e498be7d | 3163 | } |
8c8cc2c1 | 3164 | |
8c8cc2c1 | 3165 | static __always_inline void * |
48356303 | 3166 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3167 | unsigned long caller) |
8c8cc2c1 PE |
3168 | { |
3169 | unsigned long save_flags; | |
3170 | void *ptr; | |
7d6e6d09 | 3171 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3172 | |
dcce284a | 3173 | flags &= gfp_allowed_mask; |
7e85ee0c | 3174 | |
cf40bd16 NP |
3175 | lockdep_trace_alloc(flags); |
3176 | ||
773ff60e | 3177 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3178 | return NULL; |
3179 | ||
d79923fa GC |
3180 | cachep = memcg_kmem_get_cache(cachep, flags); |
3181 | ||
8c8cc2c1 PE |
3182 | cache_alloc_debugcheck_before(cachep, flags); |
3183 | local_irq_save(save_flags); | |
3184 | ||
eacbbae3 | 3185 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3186 | nodeid = slab_node; |
8c8cc2c1 | 3187 | |
18bf8541 | 3188 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3189 | /* Node not bootstrapped yet */ |
3190 | ptr = fallback_alloc(cachep, flags); | |
3191 | goto out; | |
3192 | } | |
3193 | ||
7d6e6d09 | 3194 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3195 | /* |
3196 | * Use the locally cached objects if possible. | |
3197 | * However ____cache_alloc does not allow fallback | |
3198 | * to other nodes. It may fail while we still have | |
3199 | * objects on other nodes available. | |
3200 | */ | |
3201 | ptr = ____cache_alloc(cachep, flags); | |
3202 | if (ptr) | |
3203 | goto out; | |
3204 | } | |
3205 | /* ___cache_alloc_node can fall back to other nodes */ | |
3206 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3207 | out: | |
3208 | local_irq_restore(save_flags); | |
3209 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
8c138bc0 | 3210 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3211 | flags); |
8c8cc2c1 | 3212 | |
5087c822 | 3213 | if (likely(ptr)) { |
8c138bc0 | 3214 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
5087c822 JP |
3215 | if (unlikely(flags & __GFP_ZERO)) |
3216 | memset(ptr, 0, cachep->object_size); | |
3217 | } | |
d07dbea4 | 3218 | |
8c8cc2c1 PE |
3219 | return ptr; |
3220 | } | |
3221 | ||
3222 | static __always_inline void * | |
3223 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3224 | { | |
3225 | void *objp; | |
3226 | ||
f0432d15 | 3227 | if (current->mempolicy || unlikely(current->flags & PF_SPREAD_SLAB)) { |
8c8cc2c1 PE |
3228 | objp = alternate_node_alloc(cache, flags); |
3229 | if (objp) | |
3230 | goto out; | |
3231 | } | |
3232 | objp = ____cache_alloc(cache, flags); | |
3233 | ||
3234 | /* | |
3235 | * We may just have run out of memory on the local node. | |
3236 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3237 | */ | |
7d6e6d09 LS |
3238 | if (!objp) |
3239 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3240 | |
3241 | out: | |
3242 | return objp; | |
3243 | } | |
3244 | #else | |
3245 | ||
3246 | static __always_inline void * | |
3247 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3248 | { | |
3249 | return ____cache_alloc(cachep, flags); | |
3250 | } | |
3251 | ||
3252 | #endif /* CONFIG_NUMA */ | |
3253 | ||
3254 | static __always_inline void * | |
48356303 | 3255 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3256 | { |
3257 | unsigned long save_flags; | |
3258 | void *objp; | |
3259 | ||
dcce284a | 3260 | flags &= gfp_allowed_mask; |
7e85ee0c | 3261 | |
cf40bd16 NP |
3262 | lockdep_trace_alloc(flags); |
3263 | ||
773ff60e | 3264 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3265 | return NULL; |
3266 | ||
d79923fa GC |
3267 | cachep = memcg_kmem_get_cache(cachep, flags); |
3268 | ||
8c8cc2c1 PE |
3269 | cache_alloc_debugcheck_before(cachep, flags); |
3270 | local_irq_save(save_flags); | |
3271 | objp = __do_cache_alloc(cachep, flags); | |
3272 | local_irq_restore(save_flags); | |
3273 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
8c138bc0 | 3274 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3275 | flags); |
8c8cc2c1 PE |
3276 | prefetchw(objp); |
3277 | ||
5087c822 | 3278 | if (likely(objp)) { |
8c138bc0 | 3279 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
5087c822 JP |
3280 | if (unlikely(flags & __GFP_ZERO)) |
3281 | memset(objp, 0, cachep->object_size); | |
3282 | } | |
d07dbea4 | 3283 | |
8c8cc2c1 PE |
3284 | return objp; |
3285 | } | |
e498be7d CL |
3286 | |
3287 | /* | |
5f0985bb | 3288 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3289 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3290 | */ |
97654dfa JK |
3291 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3292 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3293 | { |
3294 | int i; | |
25c063fb | 3295 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3296 | |
3297 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3298 | void *objp; |
8456a648 | 3299 | struct page *page; |
1da177e4 | 3300 | |
072bb0aa MG |
3301 | clear_obj_pfmemalloc(&objpp[i]); |
3302 | objp = objpp[i]; | |
3303 | ||
8456a648 | 3304 | page = virt_to_head_page(objp); |
8456a648 | 3305 | list_del(&page->lru); |
ff69416e | 3306 | check_spinlock_acquired_node(cachep, node); |
8456a648 | 3307 | slab_put_obj(cachep, page, objp, node); |
1da177e4 | 3308 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3309 | n->free_objects++; |
1da177e4 LT |
3310 | |
3311 | /* fixup slab chains */ | |
8456a648 | 3312 | if (page->active == 0) { |
ce8eb6c4 CL |
3313 | if (n->free_objects > n->free_limit) { |
3314 | n->free_objects -= cachep->num; | |
97654dfa | 3315 | list_add_tail(&page->lru, list); |
1da177e4 | 3316 | } else { |
8456a648 | 3317 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3318 | } |
3319 | } else { | |
3320 | /* Unconditionally move a slab to the end of the | |
3321 | * partial list on free - maximum time for the | |
3322 | * other objects to be freed, too. | |
3323 | */ | |
8456a648 | 3324 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3325 | } |
3326 | } | |
3327 | } | |
3328 | ||
343e0d7a | 3329 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3330 | { |
3331 | int batchcount; | |
ce8eb6c4 | 3332 | struct kmem_cache_node *n; |
7d6e6d09 | 3333 | int node = numa_mem_id(); |
97654dfa | 3334 | LIST_HEAD(list); |
1da177e4 LT |
3335 | |
3336 | batchcount = ac->batchcount; | |
3337 | #if DEBUG | |
3338 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3339 | #endif | |
3340 | check_irq_off(); | |
18bf8541 | 3341 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3342 | spin_lock(&n->list_lock); |
3343 | if (n->shared) { | |
3344 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3345 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3346 | if (max) { |
3347 | if (batchcount > max) | |
3348 | batchcount = max; | |
e498be7d | 3349 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3350 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3351 | shared_array->avail += batchcount; |
3352 | goto free_done; | |
3353 | } | |
3354 | } | |
3355 | ||
97654dfa | 3356 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3357 | free_done: |
1da177e4 LT |
3358 | #if STATS |
3359 | { | |
3360 | int i = 0; | |
3361 | struct list_head *p; | |
3362 | ||
ce8eb6c4 CL |
3363 | p = n->slabs_free.next; |
3364 | while (p != &(n->slabs_free)) { | |
8456a648 | 3365 | struct page *page; |
1da177e4 | 3366 | |
8456a648 JK |
3367 | page = list_entry(p, struct page, lru); |
3368 | BUG_ON(page->active); | |
1da177e4 LT |
3369 | |
3370 | i++; | |
3371 | p = p->next; | |
3372 | } | |
3373 | STATS_SET_FREEABLE(cachep, i); | |
3374 | } | |
3375 | #endif | |
ce8eb6c4 | 3376 | spin_unlock(&n->list_lock); |
97654dfa | 3377 | slabs_destroy(cachep, &list); |
1da177e4 | 3378 | ac->avail -= batchcount; |
a737b3e2 | 3379 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3380 | } |
3381 | ||
3382 | /* | |
a737b3e2 AM |
3383 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3384 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3385 | */ |
a947eb95 | 3386 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3387 | unsigned long caller) |
1da177e4 | 3388 | { |
9a2dba4b | 3389 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3390 | |
3391 | check_irq_off(); | |
d5cff635 | 3392 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3393 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3394 | |
8c138bc0 | 3395 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3396 | |
1807a1aa SS |
3397 | /* |
3398 | * Skip calling cache_free_alien() when the platform is not numa. | |
3399 | * This will avoid cache misses that happen while accessing slabp (which | |
3400 | * is per page memory reference) to get nodeid. Instead use a global | |
3401 | * variable to skip the call, which is mostly likely to be present in | |
3402 | * the cache. | |
3403 | */ | |
b6e68bc1 | 3404 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3405 | return; |
3406 | ||
1da177e4 LT |
3407 | if (likely(ac->avail < ac->limit)) { |
3408 | STATS_INC_FREEHIT(cachep); | |
1da177e4 LT |
3409 | } else { |
3410 | STATS_INC_FREEMISS(cachep); | |
3411 | cache_flusharray(cachep, ac); | |
1da177e4 | 3412 | } |
42c8c99c | 3413 | |
072bb0aa | 3414 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3415 | } |
3416 | ||
3417 | /** | |
3418 | * kmem_cache_alloc - Allocate an object | |
3419 | * @cachep: The cache to allocate from. | |
3420 | * @flags: See kmalloc(). | |
3421 | * | |
3422 | * Allocate an object from this cache. The flags are only relevant | |
3423 | * if the cache has no available objects. | |
3424 | */ | |
343e0d7a | 3425 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3426 | { |
48356303 | 3427 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3428 | |
ca2b84cb | 3429 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3430 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3431 | |
3432 | return ret; | |
1da177e4 LT |
3433 | } |
3434 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3435 | ||
0f24f128 | 3436 | #ifdef CONFIG_TRACING |
85beb586 | 3437 | void * |
4052147c | 3438 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3439 | { |
85beb586 SR |
3440 | void *ret; |
3441 | ||
48356303 | 3442 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3443 | |
3444 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3445 | size, cachep->size, flags); |
85beb586 | 3446 | return ret; |
36555751 | 3447 | } |
85beb586 | 3448 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3449 | #endif |
3450 | ||
1da177e4 | 3451 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3452 | /** |
3453 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3454 | * @cachep: The cache to allocate from. | |
3455 | * @flags: See kmalloc(). | |
3456 | * @nodeid: node number of the target node. | |
3457 | * | |
3458 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3459 | * node, which can improve the performance for cpu bound structures. | |
3460 | * | |
3461 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3462 | */ | |
8b98c169 CH |
3463 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3464 | { | |
48356303 | 3465 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3466 | |
ca2b84cb | 3467 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3468 | cachep->object_size, cachep->size, |
ca2b84cb | 3469 | flags, nodeid); |
36555751 EGM |
3470 | |
3471 | return ret; | |
8b98c169 | 3472 | } |
1da177e4 LT |
3473 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3474 | ||
0f24f128 | 3475 | #ifdef CONFIG_TRACING |
4052147c | 3476 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3477 | gfp_t flags, |
4052147c EG |
3478 | int nodeid, |
3479 | size_t size) | |
36555751 | 3480 | { |
85beb586 SR |
3481 | void *ret; |
3482 | ||
592f4145 | 3483 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3484 | |
85beb586 | 3485 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3486 | size, cachep->size, |
85beb586 SR |
3487 | flags, nodeid); |
3488 | return ret; | |
36555751 | 3489 | } |
85beb586 | 3490 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3491 | #endif |
3492 | ||
8b98c169 | 3493 | static __always_inline void * |
7c0cb9c6 | 3494 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3495 | { |
343e0d7a | 3496 | struct kmem_cache *cachep; |
97e2bde4 | 3497 | |
2c59dd65 | 3498 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3499 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3500 | return cachep; | |
4052147c | 3501 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3502 | } |
8b98c169 | 3503 | |
0bb38a5c | 3504 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
8b98c169 CH |
3505 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3506 | { | |
7c0cb9c6 | 3507 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3508 | } |
dbe5e69d | 3509 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3510 | |
3511 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3512 | int node, unsigned long caller) |
8b98c169 | 3513 | { |
7c0cb9c6 | 3514 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3515 | } |
3516 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
3517 | #else | |
3518 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3519 | { | |
7c0cb9c6 | 3520 | return __do_kmalloc_node(size, flags, node, 0); |
8b98c169 CH |
3521 | } |
3522 | EXPORT_SYMBOL(__kmalloc_node); | |
0bb38a5c | 3523 | #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */ |
8b98c169 | 3524 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3525 | |
3526 | /** | |
800590f5 | 3527 | * __do_kmalloc - allocate memory |
1da177e4 | 3528 | * @size: how many bytes of memory are required. |
800590f5 | 3529 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3530 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3531 | */ |
7fd6b141 | 3532 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3533 | unsigned long caller) |
1da177e4 | 3534 | { |
343e0d7a | 3535 | struct kmem_cache *cachep; |
36555751 | 3536 | void *ret; |
1da177e4 | 3537 | |
2c59dd65 | 3538 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3539 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3540 | return cachep; | |
48356303 | 3541 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3542 | |
7c0cb9c6 | 3543 | trace_kmalloc(caller, ret, |
3b0efdfa | 3544 | size, cachep->size, flags); |
36555751 EGM |
3545 | |
3546 | return ret; | |
7fd6b141 PE |
3547 | } |
3548 | ||
7fd6b141 | 3549 | |
0bb38a5c | 3550 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
7fd6b141 PE |
3551 | void *__kmalloc(size_t size, gfp_t flags) |
3552 | { | |
7c0cb9c6 | 3553 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3554 | } |
3555 | EXPORT_SYMBOL(__kmalloc); | |
3556 | ||
ce71e27c | 3557 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3558 | { |
7c0cb9c6 | 3559 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3560 | } |
3561 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea CH |
3562 | |
3563 | #else | |
3564 | void *__kmalloc(size_t size, gfp_t flags) | |
3565 | { | |
7c0cb9c6 | 3566 | return __do_kmalloc(size, flags, 0); |
1d2c8eea CH |
3567 | } |
3568 | EXPORT_SYMBOL(__kmalloc); | |
7fd6b141 PE |
3569 | #endif |
3570 | ||
1da177e4 LT |
3571 | /** |
3572 | * kmem_cache_free - Deallocate an object | |
3573 | * @cachep: The cache the allocation was from. | |
3574 | * @objp: The previously allocated object. | |
3575 | * | |
3576 | * Free an object which was previously allocated from this | |
3577 | * cache. | |
3578 | */ | |
343e0d7a | 3579 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3580 | { |
3581 | unsigned long flags; | |
b9ce5ef4 GC |
3582 | cachep = cache_from_obj(cachep, objp); |
3583 | if (!cachep) | |
3584 | return; | |
1da177e4 LT |
3585 | |
3586 | local_irq_save(flags); | |
d97d476b | 3587 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3588 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3589 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3590 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3591 | local_irq_restore(flags); |
36555751 | 3592 | |
ca2b84cb | 3593 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3594 | } |
3595 | EXPORT_SYMBOL(kmem_cache_free); | |
3596 | ||
1da177e4 LT |
3597 | /** |
3598 | * kfree - free previously allocated memory | |
3599 | * @objp: pointer returned by kmalloc. | |
3600 | * | |
80e93eff PE |
3601 | * If @objp is NULL, no operation is performed. |
3602 | * | |
1da177e4 LT |
3603 | * Don't free memory not originally allocated by kmalloc() |
3604 | * or you will run into trouble. | |
3605 | */ | |
3606 | void kfree(const void *objp) | |
3607 | { | |
343e0d7a | 3608 | struct kmem_cache *c; |
1da177e4 LT |
3609 | unsigned long flags; |
3610 | ||
2121db74 PE |
3611 | trace_kfree(_RET_IP_, objp); |
3612 | ||
6cb8f913 | 3613 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3614 | return; |
3615 | local_irq_save(flags); | |
3616 | kfree_debugcheck(objp); | |
6ed5eb22 | 3617 | c = virt_to_cache(objp); |
8c138bc0 CL |
3618 | debug_check_no_locks_freed(objp, c->object_size); |
3619 | ||
3620 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3621 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3622 | local_irq_restore(flags); |
3623 | } | |
3624 | EXPORT_SYMBOL(kfree); | |
3625 | ||
e498be7d | 3626 | /* |
ce8eb6c4 | 3627 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3628 | */ |
5f0985bb | 3629 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3630 | { |
3631 | int node; | |
ce8eb6c4 | 3632 | struct kmem_cache_node *n; |
cafeb02e | 3633 | struct array_cache *new_shared; |
c8522a3a | 3634 | struct alien_cache **new_alien = NULL; |
e498be7d | 3635 | |
9c09a95c | 3636 | for_each_online_node(node) { |
cafeb02e | 3637 | |
3395ee05 | 3638 | if (use_alien_caches) { |
83b519e8 | 3639 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3640 | if (!new_alien) |
3641 | goto fail; | |
3642 | } | |
cafeb02e | 3643 | |
63109846 ED |
3644 | new_shared = NULL; |
3645 | if (cachep->shared) { | |
3646 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3647 | cachep->shared*cachep->batchcount, |
83b519e8 | 3648 | 0xbaadf00d, gfp); |
63109846 ED |
3649 | if (!new_shared) { |
3650 | free_alien_cache(new_alien); | |
3651 | goto fail; | |
3652 | } | |
0718dc2a | 3653 | } |
cafeb02e | 3654 | |
18bf8541 | 3655 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3656 | if (n) { |
3657 | struct array_cache *shared = n->shared; | |
97654dfa | 3658 | LIST_HEAD(list); |
cafeb02e | 3659 | |
ce8eb6c4 | 3660 | spin_lock_irq(&n->list_lock); |
e498be7d | 3661 | |
cafeb02e | 3662 | if (shared) |
0718dc2a | 3663 | free_block(cachep, shared->entry, |
97654dfa | 3664 | shared->avail, node, &list); |
e498be7d | 3665 | |
ce8eb6c4 CL |
3666 | n->shared = new_shared; |
3667 | if (!n->alien) { | |
3668 | n->alien = new_alien; | |
e498be7d CL |
3669 | new_alien = NULL; |
3670 | } | |
ce8eb6c4 | 3671 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3672 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3673 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3674 | slabs_destroy(cachep, &list); |
cafeb02e | 3675 | kfree(shared); |
e498be7d CL |
3676 | free_alien_cache(new_alien); |
3677 | continue; | |
3678 | } | |
ce8eb6c4 CL |
3679 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3680 | if (!n) { | |
0718dc2a CL |
3681 | free_alien_cache(new_alien); |
3682 | kfree(new_shared); | |
e498be7d | 3683 | goto fail; |
0718dc2a | 3684 | } |
e498be7d | 3685 | |
ce8eb6c4 | 3686 | kmem_cache_node_init(n); |
5f0985bb JZ |
3687 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3688 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3689 | n->shared = new_shared; |
3690 | n->alien = new_alien; | |
3691 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3692 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3693 | cachep->node[node] = n; |
e498be7d | 3694 | } |
cafeb02e | 3695 | return 0; |
0718dc2a | 3696 | |
a737b3e2 | 3697 | fail: |
3b0efdfa | 3698 | if (!cachep->list.next) { |
0718dc2a CL |
3699 | /* Cache is not active yet. Roll back what we did */ |
3700 | node--; | |
3701 | while (node >= 0) { | |
18bf8541 CL |
3702 | n = get_node(cachep, node); |
3703 | if (n) { | |
ce8eb6c4 CL |
3704 | kfree(n->shared); |
3705 | free_alien_cache(n->alien); | |
3706 | kfree(n); | |
6a67368c | 3707 | cachep->node[node] = NULL; |
0718dc2a CL |
3708 | } |
3709 | node--; | |
3710 | } | |
3711 | } | |
cafeb02e | 3712 | return -ENOMEM; |
e498be7d CL |
3713 | } |
3714 | ||
1da177e4 | 3715 | struct ccupdate_struct { |
343e0d7a | 3716 | struct kmem_cache *cachep; |
acfe7d74 | 3717 | struct array_cache *new[0]; |
1da177e4 LT |
3718 | }; |
3719 | ||
3720 | static void do_ccupdate_local(void *info) | |
3721 | { | |
a737b3e2 | 3722 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3723 | struct array_cache *old; |
3724 | ||
3725 | check_irq_off(); | |
9a2dba4b | 3726 | old = cpu_cache_get(new->cachep); |
e498be7d | 3727 | |
1da177e4 LT |
3728 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3729 | new->new[smp_processor_id()] = old; | |
3730 | } | |
3731 | ||
18004c5d | 3732 | /* Always called with the slab_mutex held */ |
943a451a | 3733 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3734 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3735 | { |
d2e7b7d0 | 3736 | struct ccupdate_struct *new; |
2ed3a4ef | 3737 | int i; |
1da177e4 | 3738 | |
acfe7d74 ED |
3739 | new = kzalloc(sizeof(*new) + nr_cpu_ids * sizeof(struct array_cache *), |
3740 | gfp); | |
d2e7b7d0 SS |
3741 | if (!new) |
3742 | return -ENOMEM; | |
3743 | ||
e498be7d | 3744 | for_each_online_cpu(i) { |
7d6e6d09 | 3745 | new->new[i] = alloc_arraycache(cpu_to_mem(i), limit, |
83b519e8 | 3746 | batchcount, gfp); |
d2e7b7d0 | 3747 | if (!new->new[i]) { |
b28a02de | 3748 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3749 | kfree(new->new[i]); |
3750 | kfree(new); | |
e498be7d | 3751 | return -ENOMEM; |
1da177e4 LT |
3752 | } |
3753 | } | |
d2e7b7d0 | 3754 | new->cachep = cachep; |
1da177e4 | 3755 | |
15c8b6c1 | 3756 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 3757 | |
1da177e4 | 3758 | check_irq_on(); |
1da177e4 LT |
3759 | cachep->batchcount = batchcount; |
3760 | cachep->limit = limit; | |
e498be7d | 3761 | cachep->shared = shared; |
1da177e4 | 3762 | |
e498be7d | 3763 | for_each_online_cpu(i) { |
97654dfa | 3764 | LIST_HEAD(list); |
d2e7b7d0 | 3765 | struct array_cache *ccold = new->new[i]; |
18bf8541 CL |
3766 | int node; |
3767 | struct kmem_cache_node *n; | |
3768 | ||
1da177e4 LT |
3769 | if (!ccold) |
3770 | continue; | |
18bf8541 CL |
3771 | |
3772 | node = cpu_to_mem(i); | |
3773 | n = get_node(cachep, node); | |
3774 | spin_lock_irq(&n->list_lock); | |
97654dfa | 3775 | free_block(cachep, ccold->entry, ccold->avail, node, &list); |
18bf8541 | 3776 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3777 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3778 | kfree(ccold); |
3779 | } | |
d2e7b7d0 | 3780 | kfree(new); |
5f0985bb | 3781 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3782 | } |
3783 | ||
943a451a GC |
3784 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3785 | int batchcount, int shared, gfp_t gfp) | |
3786 | { | |
3787 | int ret; | |
3788 | struct kmem_cache *c = NULL; | |
3789 | int i = 0; | |
3790 | ||
3791 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3792 | ||
3793 | if (slab_state < FULL) | |
3794 | return ret; | |
3795 | ||
3796 | if ((ret < 0) || !is_root_cache(cachep)) | |
3797 | return ret; | |
3798 | ||
ebe945c2 | 3799 | VM_BUG_ON(!mutex_is_locked(&slab_mutex)); |
943a451a | 3800 | for_each_memcg_cache_index(i) { |
2ade4de8 | 3801 | c = cache_from_memcg_idx(cachep, i); |
943a451a GC |
3802 | if (c) |
3803 | /* return value determined by the parent cache only */ | |
3804 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
3805 | } | |
3806 | ||
3807 | return ret; | |
3808 | } | |
3809 | ||
18004c5d | 3810 | /* Called with slab_mutex held always */ |
83b519e8 | 3811 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3812 | { |
3813 | int err; | |
943a451a GC |
3814 | int limit = 0; |
3815 | int shared = 0; | |
3816 | int batchcount = 0; | |
3817 | ||
3818 | if (!is_root_cache(cachep)) { | |
3819 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3820 | limit = root->limit; | |
3821 | shared = root->shared; | |
3822 | batchcount = root->batchcount; | |
3823 | } | |
1da177e4 | 3824 | |
943a451a GC |
3825 | if (limit && shared && batchcount) |
3826 | goto skip_setup; | |
a737b3e2 AM |
3827 | /* |
3828 | * The head array serves three purposes: | |
1da177e4 LT |
3829 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3830 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3831 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3832 | * bufctl chains: array operations are cheaper. |
3833 | * The numbers are guessed, we should auto-tune as described by | |
3834 | * Bonwick. | |
3835 | */ | |
3b0efdfa | 3836 | if (cachep->size > 131072) |
1da177e4 | 3837 | limit = 1; |
3b0efdfa | 3838 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3839 | limit = 8; |
3b0efdfa | 3840 | else if (cachep->size > 1024) |
1da177e4 | 3841 | limit = 24; |
3b0efdfa | 3842 | else if (cachep->size > 256) |
1da177e4 LT |
3843 | limit = 54; |
3844 | else | |
3845 | limit = 120; | |
3846 | ||
a737b3e2 AM |
3847 | /* |
3848 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3849 | * allocation behaviour: Most allocs on one cpu, most free operations |
3850 | * on another cpu. For these cases, an efficient object passing between | |
3851 | * cpus is necessary. This is provided by a shared array. The array | |
3852 | * replaces Bonwick's magazine layer. | |
3853 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3854 | * to a larger limit. Thus disabled by default. | |
3855 | */ | |
3856 | shared = 0; | |
3b0efdfa | 3857 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3858 | shared = 8; |
1da177e4 LT |
3859 | |
3860 | #if DEBUG | |
a737b3e2 AM |
3861 | /* |
3862 | * With debugging enabled, large batchcount lead to excessively long | |
3863 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3864 | */ |
3865 | if (limit > 32) | |
3866 | limit = 32; | |
3867 | #endif | |
943a451a GC |
3868 | batchcount = (limit + 1) / 2; |
3869 | skip_setup: | |
3870 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3871 | if (err) |
3872 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3873 | cachep->name, -err); |
2ed3a4ef | 3874 | return err; |
1da177e4 LT |
3875 | } |
3876 | ||
1b55253a | 3877 | /* |
ce8eb6c4 CL |
3878 | * Drain an array if it contains any elements taking the node lock only if |
3879 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3880 | * if drain_array() is used on the shared array. |
1b55253a | 3881 | */ |
ce8eb6c4 | 3882 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3883 | struct array_cache *ac, int force, int node) |
1da177e4 | 3884 | { |
97654dfa | 3885 | LIST_HEAD(list); |
1da177e4 LT |
3886 | int tofree; |
3887 | ||
1b55253a CL |
3888 | if (!ac || !ac->avail) |
3889 | return; | |
1da177e4 LT |
3890 | if (ac->touched && !force) { |
3891 | ac->touched = 0; | |
b18e7e65 | 3892 | } else { |
ce8eb6c4 | 3893 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3894 | if (ac->avail) { |
3895 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3896 | if (tofree > ac->avail) | |
3897 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 3898 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
3899 | ac->avail -= tofree; |
3900 | memmove(ac->entry, &(ac->entry[tofree]), | |
3901 | sizeof(void *) * ac->avail); | |
3902 | } | |
ce8eb6c4 | 3903 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3904 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3905 | } |
3906 | } | |
3907 | ||
3908 | /** | |
3909 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3910 | * @w: work descriptor |
1da177e4 LT |
3911 | * |
3912 | * Called from workqueue/eventd every few seconds. | |
3913 | * Purpose: | |
3914 | * - clear the per-cpu caches for this CPU. | |
3915 | * - return freeable pages to the main free memory pool. | |
3916 | * | |
a737b3e2 AM |
3917 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3918 | * again on the next iteration. | |
1da177e4 | 3919 | */ |
7c5cae36 | 3920 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3921 | { |
7a7c381d | 3922 | struct kmem_cache *searchp; |
ce8eb6c4 | 3923 | struct kmem_cache_node *n; |
7d6e6d09 | 3924 | int node = numa_mem_id(); |
bf6aede7 | 3925 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3926 | |
18004c5d | 3927 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3928 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3929 | goto out; |
1da177e4 | 3930 | |
18004c5d | 3931 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3932 | check_irq_on(); |
3933 | ||
35386e3b | 3934 | /* |
ce8eb6c4 | 3935 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3936 | * have established with reasonable certainty that |
3937 | * we can do some work if the lock was obtained. | |
3938 | */ | |
18bf8541 | 3939 | n = get_node(searchp, node); |
35386e3b | 3940 | |
ce8eb6c4 | 3941 | reap_alien(searchp, n); |
1da177e4 | 3942 | |
ce8eb6c4 | 3943 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3944 | |
35386e3b CL |
3945 | /* |
3946 | * These are racy checks but it does not matter | |
3947 | * if we skip one check or scan twice. | |
3948 | */ | |
ce8eb6c4 | 3949 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3950 | goto next; |
1da177e4 | 3951 | |
5f0985bb | 3952 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3953 | |
ce8eb6c4 | 3954 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 3955 | |
ce8eb6c4 CL |
3956 | if (n->free_touched) |
3957 | n->free_touched = 0; | |
ed11d9eb CL |
3958 | else { |
3959 | int freed; | |
1da177e4 | 3960 | |
ce8eb6c4 | 3961 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3962 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3963 | STATS_ADD_REAPED(searchp, freed); | |
3964 | } | |
35386e3b | 3965 | next: |
1da177e4 LT |
3966 | cond_resched(); |
3967 | } | |
3968 | check_irq_on(); | |
18004c5d | 3969 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3970 | next_reap_node(); |
7c5cae36 | 3971 | out: |
a737b3e2 | 3972 | /* Set up the next iteration */ |
5f0985bb | 3973 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3974 | } |
3975 | ||
158a9624 | 3976 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3977 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3978 | { |
8456a648 | 3979 | struct page *page; |
b28a02de PE |
3980 | unsigned long active_objs; |
3981 | unsigned long num_objs; | |
3982 | unsigned long active_slabs = 0; | |
3983 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3984 | const char *name; |
1da177e4 | 3985 | char *error = NULL; |
e498be7d | 3986 | int node; |
ce8eb6c4 | 3987 | struct kmem_cache_node *n; |
1da177e4 | 3988 | |
1da177e4 LT |
3989 | active_objs = 0; |
3990 | num_slabs = 0; | |
18bf8541 | 3991 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 3992 | |
ca3b9b91 | 3993 | check_irq_on(); |
ce8eb6c4 | 3994 | spin_lock_irq(&n->list_lock); |
e498be7d | 3995 | |
8456a648 JK |
3996 | list_for_each_entry(page, &n->slabs_full, lru) { |
3997 | if (page->active != cachep->num && !error) | |
e498be7d CL |
3998 | error = "slabs_full accounting error"; |
3999 | active_objs += cachep->num; | |
4000 | active_slabs++; | |
4001 | } | |
8456a648 JK |
4002 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4003 | if (page->active == cachep->num && !error) | |
106a74e1 | 4004 | error = "slabs_partial accounting error"; |
8456a648 | 4005 | if (!page->active && !error) |
106a74e1 | 4006 | error = "slabs_partial accounting error"; |
8456a648 | 4007 | active_objs += page->active; |
e498be7d CL |
4008 | active_slabs++; |
4009 | } | |
8456a648 JK |
4010 | list_for_each_entry(page, &n->slabs_free, lru) { |
4011 | if (page->active && !error) | |
106a74e1 | 4012 | error = "slabs_free accounting error"; |
e498be7d CL |
4013 | num_slabs++; |
4014 | } | |
ce8eb6c4 CL |
4015 | free_objects += n->free_objects; |
4016 | if (n->shared) | |
4017 | shared_avail += n->shared->avail; | |
e498be7d | 4018 | |
ce8eb6c4 | 4019 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4020 | } |
b28a02de PE |
4021 | num_slabs += active_slabs; |
4022 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4023 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4024 | error = "free_objects accounting error"; |
4025 | ||
b28a02de | 4026 | name = cachep->name; |
1da177e4 LT |
4027 | if (error) |
4028 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4029 | ||
0d7561c6 GC |
4030 | sinfo->active_objs = active_objs; |
4031 | sinfo->num_objs = num_objs; | |
4032 | sinfo->active_slabs = active_slabs; | |
4033 | sinfo->num_slabs = num_slabs; | |
4034 | sinfo->shared_avail = shared_avail; | |
4035 | sinfo->limit = cachep->limit; | |
4036 | sinfo->batchcount = cachep->batchcount; | |
4037 | sinfo->shared = cachep->shared; | |
4038 | sinfo->objects_per_slab = cachep->num; | |
4039 | sinfo->cache_order = cachep->gfporder; | |
4040 | } | |
4041 | ||
4042 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4043 | { | |
1da177e4 | 4044 | #if STATS |
ce8eb6c4 | 4045 | { /* node stats */ |
1da177e4 LT |
4046 | unsigned long high = cachep->high_mark; |
4047 | unsigned long allocs = cachep->num_allocations; | |
4048 | unsigned long grown = cachep->grown; | |
4049 | unsigned long reaped = cachep->reaped; | |
4050 | unsigned long errors = cachep->errors; | |
4051 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4052 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4053 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4054 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4055 | |
e92dd4fd JP |
4056 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4057 | "%4lu %4lu %4lu %4lu %4lu", | |
4058 | allocs, high, grown, | |
4059 | reaped, errors, max_freeable, node_allocs, | |
4060 | node_frees, overflows); | |
1da177e4 LT |
4061 | } |
4062 | /* cpu stats */ | |
4063 | { | |
4064 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4065 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4066 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4067 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4068 | ||
4069 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4070 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4071 | } |
4072 | #endif | |
1da177e4 LT |
4073 | } |
4074 | ||
1da177e4 LT |
4075 | #define MAX_SLABINFO_WRITE 128 |
4076 | /** | |
4077 | * slabinfo_write - Tuning for the slab allocator | |
4078 | * @file: unused | |
4079 | * @buffer: user buffer | |
4080 | * @count: data length | |
4081 | * @ppos: unused | |
4082 | */ | |
b7454ad3 | 4083 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4084 | size_t count, loff_t *ppos) |
1da177e4 | 4085 | { |
b28a02de | 4086 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4087 | int limit, batchcount, shared, res; |
7a7c381d | 4088 | struct kmem_cache *cachep; |
b28a02de | 4089 | |
1da177e4 LT |
4090 | if (count > MAX_SLABINFO_WRITE) |
4091 | return -EINVAL; | |
4092 | if (copy_from_user(&kbuf, buffer, count)) | |
4093 | return -EFAULT; | |
b28a02de | 4094 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4095 | |
4096 | tmp = strchr(kbuf, ' '); | |
4097 | if (!tmp) | |
4098 | return -EINVAL; | |
4099 | *tmp = '\0'; | |
4100 | tmp++; | |
4101 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4102 | return -EINVAL; | |
4103 | ||
4104 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4105 | mutex_lock(&slab_mutex); |
1da177e4 | 4106 | res = -EINVAL; |
18004c5d | 4107 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4108 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4109 | if (limit < 1 || batchcount < 1 || |
4110 | batchcount > limit || shared < 0) { | |
e498be7d | 4111 | res = 0; |
1da177e4 | 4112 | } else { |
e498be7d | 4113 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4114 | batchcount, shared, |
4115 | GFP_KERNEL); | |
1da177e4 LT |
4116 | } |
4117 | break; | |
4118 | } | |
4119 | } | |
18004c5d | 4120 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4121 | if (res >= 0) |
4122 | res = count; | |
4123 | return res; | |
4124 | } | |
871751e2 AV |
4125 | |
4126 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4127 | ||
4128 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4129 | { | |
18004c5d CL |
4130 | mutex_lock(&slab_mutex); |
4131 | return seq_list_start(&slab_caches, *pos); | |
871751e2 AV |
4132 | } |
4133 | ||
4134 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4135 | { | |
4136 | unsigned long *p; | |
4137 | int l; | |
4138 | if (!v) | |
4139 | return 1; | |
4140 | l = n[1]; | |
4141 | p = n + 2; | |
4142 | while (l) { | |
4143 | int i = l/2; | |
4144 | unsigned long *q = p + 2 * i; | |
4145 | if (*q == v) { | |
4146 | q[1]++; | |
4147 | return 1; | |
4148 | } | |
4149 | if (*q > v) { | |
4150 | l = i; | |
4151 | } else { | |
4152 | p = q + 2; | |
4153 | l -= i + 1; | |
4154 | } | |
4155 | } | |
4156 | if (++n[1] == n[0]) | |
4157 | return 0; | |
4158 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4159 | p[0] = v; | |
4160 | p[1] = 1; | |
4161 | return 1; | |
4162 | } | |
4163 | ||
8456a648 JK |
4164 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4165 | struct page *page) | |
871751e2 AV |
4166 | { |
4167 | void *p; | |
03787301 | 4168 | int i; |
b1cb0982 | 4169 | |
871751e2 AV |
4170 | if (n[0] == n[1]) |
4171 | return; | |
8456a648 | 4172 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
03787301 | 4173 | if (get_obj_status(page, i) != OBJECT_ACTIVE) |
871751e2 | 4174 | continue; |
b1cb0982 | 4175 | |
871751e2 AV |
4176 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
4177 | return; | |
4178 | } | |
4179 | } | |
4180 | ||
4181 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4182 | { | |
4183 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4184 | unsigned long offset, size; |
9281acea | 4185 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4186 | |
a5c43dae | 4187 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4188 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4189 | if (modname[0]) |
871751e2 AV |
4190 | seq_printf(m, " [%s]", modname); |
4191 | return; | |
4192 | } | |
4193 | #endif | |
4194 | seq_printf(m, "%p", (void *)address); | |
4195 | } | |
4196 | ||
4197 | static int leaks_show(struct seq_file *m, void *p) | |
4198 | { | |
0672aa7c | 4199 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4200 | struct page *page; |
ce8eb6c4 | 4201 | struct kmem_cache_node *n; |
871751e2 | 4202 | const char *name; |
db845067 | 4203 | unsigned long *x = m->private; |
871751e2 AV |
4204 | int node; |
4205 | int i; | |
4206 | ||
4207 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4208 | return 0; | |
4209 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4210 | return 0; | |
4211 | ||
4212 | /* OK, we can do it */ | |
4213 | ||
db845067 | 4214 | x[1] = 0; |
871751e2 | 4215 | |
18bf8541 | 4216 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 AV |
4217 | |
4218 | check_irq_on(); | |
ce8eb6c4 | 4219 | spin_lock_irq(&n->list_lock); |
871751e2 | 4220 | |
8456a648 JK |
4221 | list_for_each_entry(page, &n->slabs_full, lru) |
4222 | handle_slab(x, cachep, page); | |
4223 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4224 | handle_slab(x, cachep, page); | |
ce8eb6c4 | 4225 | spin_unlock_irq(&n->list_lock); |
871751e2 AV |
4226 | } |
4227 | name = cachep->name; | |
db845067 | 4228 | if (x[0] == x[1]) { |
871751e2 | 4229 | /* Increase the buffer size */ |
18004c5d | 4230 | mutex_unlock(&slab_mutex); |
db845067 | 4231 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4232 | if (!m->private) { |
4233 | /* Too bad, we are really out */ | |
db845067 | 4234 | m->private = x; |
18004c5d | 4235 | mutex_lock(&slab_mutex); |
871751e2 AV |
4236 | return -ENOMEM; |
4237 | } | |
db845067 CL |
4238 | *(unsigned long *)m->private = x[0] * 2; |
4239 | kfree(x); | |
18004c5d | 4240 | mutex_lock(&slab_mutex); |
871751e2 AV |
4241 | /* Now make sure this entry will be retried */ |
4242 | m->count = m->size; | |
4243 | return 0; | |
4244 | } | |
db845067 CL |
4245 | for (i = 0; i < x[1]; i++) { |
4246 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4247 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4248 | seq_putc(m, '\n'); |
4249 | } | |
d2e7b7d0 | 4250 | |
871751e2 AV |
4251 | return 0; |
4252 | } | |
4253 | ||
a0ec95a8 | 4254 | static const struct seq_operations slabstats_op = { |
871751e2 | 4255 | .start = leaks_start, |
276a2439 WL |
4256 | .next = slab_next, |
4257 | .stop = slab_stop, | |
871751e2 AV |
4258 | .show = leaks_show, |
4259 | }; | |
a0ec95a8 AD |
4260 | |
4261 | static int slabstats_open(struct inode *inode, struct file *file) | |
4262 | { | |
4263 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4264 | int ret = -ENOMEM; | |
4265 | if (n) { | |
4266 | ret = seq_open(file, &slabstats_op); | |
4267 | if (!ret) { | |
4268 | struct seq_file *m = file->private_data; | |
4269 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4270 | m->private = n; | |
4271 | n = NULL; | |
4272 | } | |
4273 | kfree(n); | |
4274 | } | |
4275 | return ret; | |
4276 | } | |
4277 | ||
4278 | static const struct file_operations proc_slabstats_operations = { | |
4279 | .open = slabstats_open, | |
4280 | .read = seq_read, | |
4281 | .llseek = seq_lseek, | |
4282 | .release = seq_release_private, | |
4283 | }; | |
4284 | #endif | |
4285 | ||
4286 | static int __init slab_proc_init(void) | |
4287 | { | |
4288 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4289 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4290 | #endif |
a0ec95a8 AD |
4291 | return 0; |
4292 | } | |
4293 | module_init(slab_proc_init); | |
1da177e4 LT |
4294 | #endif |
4295 | ||
00e145b6 MS |
4296 | /** |
4297 | * ksize - get the actual amount of memory allocated for a given object | |
4298 | * @objp: Pointer to the object | |
4299 | * | |
4300 | * kmalloc may internally round up allocations and return more memory | |
4301 | * than requested. ksize() can be used to determine the actual amount of | |
4302 | * memory allocated. The caller may use this additional memory, even though | |
4303 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4304 | * The caller must guarantee that objp points to a valid object previously | |
4305 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4306 | * must not be freed during the duration of the call. | |
4307 | */ | |
fd76bab2 | 4308 | size_t ksize(const void *objp) |
1da177e4 | 4309 | { |
ef8b4520 CL |
4310 | BUG_ON(!objp); |
4311 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4312 | return 0; |
1da177e4 | 4313 | |
8c138bc0 | 4314 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4315 | } |
b1aabecd | 4316 | EXPORT_SYMBOL(ksize); |