]>
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 | ||
056c6241 RT |
473 | #define BAD_ALIEN_MAGIC 0x01020304ul |
474 | ||
f1aaee53 AV |
475 | #ifdef CONFIG_LOCKDEP |
476 | ||
477 | /* | |
478 | * Slab sometimes uses the kmalloc slabs to store the slab headers | |
479 | * for other slabs "off slab". | |
480 | * The locking for this is tricky in that it nests within the locks | |
481 | * of all other slabs in a few places; to deal with this special | |
482 | * locking we put on-slab caches into a separate lock-class. | |
056c6241 RT |
483 | * |
484 | * We set lock class for alien array caches which are up during init. | |
485 | * The lock annotation will be lost if all cpus of a node goes down and | |
486 | * then comes back up during hotplug | |
f1aaee53 | 487 | */ |
056c6241 RT |
488 | static struct lock_class_key on_slab_l3_key; |
489 | static struct lock_class_key on_slab_alc_key; | |
490 | ||
83835b3d PZ |
491 | static struct lock_class_key debugobj_l3_key; |
492 | static struct lock_class_key debugobj_alc_key; | |
493 | ||
494 | static void slab_set_lock_classes(struct kmem_cache *cachep, | |
495 | struct lock_class_key *l3_key, struct lock_class_key *alc_key, | |
18bf8541 | 496 | struct kmem_cache_node *n) |
83835b3d | 497 | { |
c8522a3a | 498 | struct alien_cache **alc; |
83835b3d PZ |
499 | int r; |
500 | ||
ce8eb6c4 CL |
501 | lockdep_set_class(&n->list_lock, l3_key); |
502 | alc = n->alien; | |
83835b3d PZ |
503 | /* |
504 | * FIXME: This check for BAD_ALIEN_MAGIC | |
505 | * should go away when common slab code is taught to | |
506 | * work even without alien caches. | |
507 | * Currently, non NUMA code returns BAD_ALIEN_MAGIC | |
508 | * for alloc_alien_cache, | |
509 | */ | |
510 | if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) | |
511 | return; | |
512 | for_each_node(r) { | |
513 | if (alc[r]) | |
49dfc304 | 514 | lockdep_set_class(&(alc[r]->lock), alc_key); |
83835b3d PZ |
515 | } |
516 | } | |
517 | ||
18bf8541 CL |
518 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, |
519 | struct kmem_cache_node *n) | |
83835b3d | 520 | { |
18bf8541 | 521 | slab_set_lock_classes(cachep, &debugobj_l3_key, &debugobj_alc_key, n); |
83835b3d PZ |
522 | } |
523 | ||
524 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
525 | { | |
526 | int node; | |
18bf8541 | 527 | struct kmem_cache_node *n; |
83835b3d | 528 | |
18bf8541 CL |
529 | for_each_kmem_cache_node(cachep, node, n) |
530 | slab_set_debugobj_lock_classes_node(cachep, n); | |
83835b3d PZ |
531 | } |
532 | ||
ce79ddc8 | 533 | static void init_node_lock_keys(int q) |
f1aaee53 | 534 | { |
e3366016 | 535 | int i; |
056c6241 | 536 | |
97d06609 | 537 | if (slab_state < UP) |
ce79ddc8 PE |
538 | return; |
539 | ||
0f8f8094 | 540 | for (i = 1; i <= KMALLOC_SHIFT_HIGH; i++) { |
ce8eb6c4 | 541 | struct kmem_cache_node *n; |
e3366016 CL |
542 | struct kmem_cache *cache = kmalloc_caches[i]; |
543 | ||
544 | if (!cache) | |
545 | continue; | |
ce79ddc8 | 546 | |
18bf8541 | 547 | n = get_node(cache, q); |
ce8eb6c4 | 548 | if (!n || OFF_SLAB(cache)) |
00afa758 | 549 | continue; |
83835b3d | 550 | |
e3366016 | 551 | slab_set_lock_classes(cache, &on_slab_l3_key, |
18bf8541 | 552 | &on_slab_alc_key, n); |
f1aaee53 AV |
553 | } |
554 | } | |
ce79ddc8 | 555 | |
18bf8541 CL |
556 | static void on_slab_lock_classes_node(struct kmem_cache *cachep, |
557 | struct kmem_cache_node *n) | |
6ccfb5bc | 558 | { |
6ccfb5bc | 559 | slab_set_lock_classes(cachep, &on_slab_l3_key, |
18bf8541 | 560 | &on_slab_alc_key, n); |
6ccfb5bc GC |
561 | } |
562 | ||
563 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) | |
564 | { | |
565 | int node; | |
18bf8541 | 566 | struct kmem_cache_node *n; |
6ccfb5bc GC |
567 | |
568 | VM_BUG_ON(OFF_SLAB(cachep)); | |
18bf8541 CL |
569 | for_each_kmem_cache_node(cachep, node, n) |
570 | on_slab_lock_classes_node(cachep, n); | |
6ccfb5bc GC |
571 | } |
572 | ||
1536cb39 | 573 | static inline void __init init_lock_keys(void) |
ce79ddc8 PE |
574 | { |
575 | int node; | |
576 | ||
577 | for_each_node(node) | |
578 | init_node_lock_keys(node); | |
579 | } | |
f1aaee53 | 580 | #else |
1536cb39 | 581 | static void __init init_node_lock_keys(int q) |
ce79ddc8 PE |
582 | { |
583 | } | |
584 | ||
056c6241 | 585 | static inline void init_lock_keys(void) |
f1aaee53 AV |
586 | { |
587 | } | |
83835b3d | 588 | |
6ccfb5bc GC |
589 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) |
590 | { | |
591 | } | |
592 | ||
18bf8541 CL |
593 | static inline void on_slab_lock_classes_node(struct kmem_cache *cachep, |
594 | struct kmem_cache_node *n) | |
6ccfb5bc GC |
595 | { |
596 | } | |
597 | ||
18bf8541 CL |
598 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, |
599 | struct kmem_cache_node *n) | |
83835b3d PZ |
600 | { |
601 | } | |
602 | ||
603 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
604 | { | |
605 | } | |
f1aaee53 AV |
606 | #endif |
607 | ||
1871e52c | 608 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 609 | |
343e0d7a | 610 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
611 | { |
612 | return cachep->array[smp_processor_id()]; | |
613 | } | |
614 | ||
03787301 JK |
615 | static size_t calculate_freelist_size(int nr_objs, size_t align) |
616 | { | |
617 | size_t freelist_size; | |
618 | ||
619 | freelist_size = nr_objs * sizeof(freelist_idx_t); | |
620 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) | |
621 | freelist_size += nr_objs * sizeof(char); | |
622 | ||
623 | if (align) | |
624 | freelist_size = ALIGN(freelist_size, align); | |
625 | ||
626 | return freelist_size; | |
627 | } | |
628 | ||
9cef2e2b JK |
629 | static int calculate_nr_objs(size_t slab_size, size_t buffer_size, |
630 | size_t idx_size, size_t align) | |
1da177e4 | 631 | { |
9cef2e2b | 632 | int nr_objs; |
03787301 | 633 | size_t remained_size; |
9cef2e2b | 634 | size_t freelist_size; |
03787301 | 635 | int extra_space = 0; |
9cef2e2b | 636 | |
03787301 JK |
637 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
638 | extra_space = sizeof(char); | |
9cef2e2b JK |
639 | /* |
640 | * Ignore padding for the initial guess. The padding | |
641 | * is at most @align-1 bytes, and @buffer_size is at | |
642 | * least @align. In the worst case, this result will | |
643 | * be one greater than the number of objects that fit | |
644 | * into the memory allocation when taking the padding | |
645 | * into account. | |
646 | */ | |
03787301 | 647 | nr_objs = slab_size / (buffer_size + idx_size + extra_space); |
9cef2e2b JK |
648 | |
649 | /* | |
650 | * This calculated number will be either the right | |
651 | * amount, or one greater than what we want. | |
652 | */ | |
03787301 JK |
653 | remained_size = slab_size - nr_objs * buffer_size; |
654 | freelist_size = calculate_freelist_size(nr_objs, align); | |
655 | if (remained_size < freelist_size) | |
9cef2e2b JK |
656 | nr_objs--; |
657 | ||
658 | return nr_objs; | |
fbaccacf | 659 | } |
1da177e4 | 660 | |
a737b3e2 AM |
661 | /* |
662 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
663 | */ | |
fbaccacf SR |
664 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
665 | size_t align, int flags, size_t *left_over, | |
666 | unsigned int *num) | |
667 | { | |
668 | int nr_objs; | |
669 | size_t mgmt_size; | |
670 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 671 | |
fbaccacf SR |
672 | /* |
673 | * The slab management structure can be either off the slab or | |
674 | * on it. For the latter case, the memory allocated for a | |
675 | * slab is used for: | |
676 | * | |
16025177 | 677 | * - One unsigned int for each object |
fbaccacf SR |
678 | * - Padding to respect alignment of @align |
679 | * - @buffer_size bytes for each object | |
680 | * | |
681 | * If the slab management structure is off the slab, then the | |
682 | * alignment will already be calculated into the size. Because | |
683 | * the slabs are all pages aligned, the objects will be at the | |
684 | * correct alignment when allocated. | |
685 | */ | |
686 | if (flags & CFLGS_OFF_SLAB) { | |
687 | mgmt_size = 0; | |
688 | nr_objs = slab_size / buffer_size; | |
689 | ||
fbaccacf | 690 | } else { |
9cef2e2b | 691 | nr_objs = calculate_nr_objs(slab_size, buffer_size, |
a41adfaa | 692 | sizeof(freelist_idx_t), align); |
03787301 | 693 | mgmt_size = calculate_freelist_size(nr_objs, align); |
fbaccacf SR |
694 | } |
695 | *num = nr_objs; | |
696 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
697 | } |
698 | ||
f28510d3 | 699 | #if DEBUG |
d40cee24 | 700 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 701 | |
a737b3e2 AM |
702 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
703 | char *msg) | |
1da177e4 LT |
704 | { |
705 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 706 | function, cachep->name, msg); |
1da177e4 | 707 | dump_stack(); |
373d4d09 | 708 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 709 | } |
f28510d3 | 710 | #endif |
1da177e4 | 711 | |
3395ee05 PM |
712 | /* |
713 | * By default on NUMA we use alien caches to stage the freeing of | |
714 | * objects allocated from other nodes. This causes massive memory | |
715 | * inefficiencies when using fake NUMA setup to split memory into a | |
716 | * large number of small nodes, so it can be disabled on the command | |
717 | * line | |
718 | */ | |
719 | ||
720 | static int use_alien_caches __read_mostly = 1; | |
721 | static int __init noaliencache_setup(char *s) | |
722 | { | |
723 | use_alien_caches = 0; | |
724 | return 1; | |
725 | } | |
726 | __setup("noaliencache", noaliencache_setup); | |
727 | ||
3df1cccd DR |
728 | static int __init slab_max_order_setup(char *str) |
729 | { | |
730 | get_option(&str, &slab_max_order); | |
731 | slab_max_order = slab_max_order < 0 ? 0 : | |
732 | min(slab_max_order, MAX_ORDER - 1); | |
733 | slab_max_order_set = true; | |
734 | ||
735 | return 1; | |
736 | } | |
737 | __setup("slab_max_order=", slab_max_order_setup); | |
738 | ||
8fce4d8e CL |
739 | #ifdef CONFIG_NUMA |
740 | /* | |
741 | * Special reaping functions for NUMA systems called from cache_reap(). | |
742 | * These take care of doing round robin flushing of alien caches (containing | |
743 | * objects freed on different nodes from which they were allocated) and the | |
744 | * flushing of remote pcps by calling drain_node_pages. | |
745 | */ | |
1871e52c | 746 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
747 | |
748 | static void init_reap_node(int cpu) | |
749 | { | |
750 | int node; | |
751 | ||
7d6e6d09 | 752 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 753 | if (node == MAX_NUMNODES) |
442295c9 | 754 | node = first_node(node_online_map); |
8fce4d8e | 755 | |
1871e52c | 756 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
757 | } |
758 | ||
759 | static void next_reap_node(void) | |
760 | { | |
909ea964 | 761 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 762 | |
8fce4d8e CL |
763 | node = next_node(node, node_online_map); |
764 | if (unlikely(node >= MAX_NUMNODES)) | |
765 | node = first_node(node_online_map); | |
909ea964 | 766 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
767 | } |
768 | ||
769 | #else | |
770 | #define init_reap_node(cpu) do { } while (0) | |
771 | #define next_reap_node(void) do { } while (0) | |
772 | #endif | |
773 | ||
1da177e4 LT |
774 | /* |
775 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
776 | * via the workqueue/eventd. | |
777 | * Add the CPU number into the expiration time to minimize the possibility of | |
778 | * the CPUs getting into lockstep and contending for the global cache chain | |
779 | * lock. | |
780 | */ | |
0db0628d | 781 | static void start_cpu_timer(int cpu) |
1da177e4 | 782 | { |
1871e52c | 783 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
784 | |
785 | /* | |
786 | * When this gets called from do_initcalls via cpucache_init(), | |
787 | * init_workqueues() has already run, so keventd will be setup | |
788 | * at that time. | |
789 | */ | |
52bad64d | 790 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 791 | init_reap_node(cpu); |
203b42f7 | 792 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
793 | schedule_delayed_work_on(cpu, reap_work, |
794 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
795 | } |
796 | } | |
797 | ||
1fe00d50 | 798 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 799 | { |
d5cff635 CM |
800 | /* |
801 | * The array_cache structures contain pointers to free object. | |
25985edc | 802 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
803 | * cache the pointers are not cleared and they could be counted as |
804 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
805 | * not scan such objects. | |
806 | */ | |
1fe00d50 JK |
807 | kmemleak_no_scan(ac); |
808 | if (ac) { | |
809 | ac->avail = 0; | |
810 | ac->limit = limit; | |
811 | ac->batchcount = batch; | |
812 | ac->touched = 0; | |
1da177e4 | 813 | } |
1fe00d50 JK |
814 | } |
815 | ||
816 | static struct array_cache *alloc_arraycache(int node, int entries, | |
817 | int batchcount, gfp_t gfp) | |
818 | { | |
819 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); | |
820 | struct array_cache *ac = NULL; | |
821 | ||
822 | ac = kmalloc_node(memsize, gfp, node); | |
823 | init_arraycache(ac, entries, batchcount); | |
824 | return ac; | |
1da177e4 LT |
825 | } |
826 | ||
8456a648 | 827 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 828 | { |
072bb0aa MG |
829 | return PageSlabPfmemalloc(page); |
830 | } | |
831 | ||
832 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
833 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
834 | struct array_cache *ac) | |
835 | { | |
18bf8541 | 836 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 837 | struct page *page; |
072bb0aa MG |
838 | unsigned long flags; |
839 | ||
840 | if (!pfmemalloc_active) | |
841 | return; | |
842 | ||
ce8eb6c4 | 843 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
844 | list_for_each_entry(page, &n->slabs_full, lru) |
845 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
846 | goto out; |
847 | ||
8456a648 JK |
848 | list_for_each_entry(page, &n->slabs_partial, lru) |
849 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
850 | goto out; |
851 | ||
8456a648 JK |
852 | list_for_each_entry(page, &n->slabs_free, lru) |
853 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
854 | goto out; |
855 | ||
856 | pfmemalloc_active = false; | |
857 | out: | |
ce8eb6c4 | 858 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
859 | } |
860 | ||
381760ea | 861 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
862 | gfp_t flags, bool force_refill) |
863 | { | |
864 | int i; | |
865 | void *objp = ac->entry[--ac->avail]; | |
866 | ||
867 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
868 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 869 | struct kmem_cache_node *n; |
072bb0aa MG |
870 | |
871 | if (gfp_pfmemalloc_allowed(flags)) { | |
872 | clear_obj_pfmemalloc(&objp); | |
873 | return objp; | |
874 | } | |
875 | ||
876 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 877 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
878 | /* If a !PFMEMALLOC object is found, swap them */ |
879 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
880 | objp = ac->entry[i]; | |
881 | ac->entry[i] = ac->entry[ac->avail]; | |
882 | ac->entry[ac->avail] = objp; | |
883 | return objp; | |
884 | } | |
885 | } | |
886 | ||
887 | /* | |
888 | * If there are empty slabs on the slabs_free list and we are | |
889 | * being forced to refill the cache, mark this one !pfmemalloc. | |
890 | */ | |
18bf8541 | 891 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 892 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 893 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 894 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
895 | clear_obj_pfmemalloc(&objp); |
896 | recheck_pfmemalloc_active(cachep, ac); | |
897 | return objp; | |
898 | } | |
899 | ||
900 | /* No !PFMEMALLOC objects available */ | |
901 | ac->avail++; | |
902 | objp = NULL; | |
903 | } | |
904 | ||
905 | return objp; | |
906 | } | |
907 | ||
381760ea MG |
908 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
909 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
910 | { | |
911 | void *objp; | |
912 | ||
913 | if (unlikely(sk_memalloc_socks())) | |
914 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
915 | else | |
916 | objp = ac->entry[--ac->avail]; | |
917 | ||
918 | return objp; | |
919 | } | |
920 | ||
921 | static void *__ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
072bb0aa MG |
922 | void *objp) |
923 | { | |
924 | if (unlikely(pfmemalloc_active)) { | |
925 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 926 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
927 | if (PageSlabPfmemalloc(page)) |
928 | set_obj_pfmemalloc(&objp); | |
929 | } | |
930 | ||
381760ea MG |
931 | return objp; |
932 | } | |
933 | ||
934 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
935 | void *objp) | |
936 | { | |
937 | if (unlikely(sk_memalloc_socks())) | |
938 | objp = __ac_put_obj(cachep, ac, objp); | |
939 | ||
072bb0aa MG |
940 | ac->entry[ac->avail++] = objp; |
941 | } | |
942 | ||
3ded175a CL |
943 | /* |
944 | * Transfer objects in one arraycache to another. | |
945 | * Locking must be handled by the caller. | |
946 | * | |
947 | * Return the number of entries transferred. | |
948 | */ | |
949 | static int transfer_objects(struct array_cache *to, | |
950 | struct array_cache *from, unsigned int max) | |
951 | { | |
952 | /* Figure out how many entries to transfer */ | |
732eacc0 | 953 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
954 | |
955 | if (!nr) | |
956 | return 0; | |
957 | ||
958 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
959 | sizeof(void *) *nr); | |
960 | ||
961 | from->avail -= nr; | |
962 | to->avail += nr; | |
3ded175a CL |
963 | return nr; |
964 | } | |
965 | ||
765c4507 CL |
966 | #ifndef CONFIG_NUMA |
967 | ||
968 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 969 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 970 | |
c8522a3a JK |
971 | static inline struct alien_cache **alloc_alien_cache(int node, |
972 | int limit, gfp_t gfp) | |
765c4507 | 973 | { |
c8522a3a | 974 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
975 | } |
976 | ||
c8522a3a | 977 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
978 | { |
979 | } | |
980 | ||
981 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
982 | { | |
983 | return 0; | |
984 | } | |
985 | ||
986 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
987 | gfp_t flags) | |
988 | { | |
989 | return NULL; | |
990 | } | |
991 | ||
8b98c169 | 992 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
993 | gfp_t flags, int nodeid) |
994 | { | |
995 | return NULL; | |
996 | } | |
997 | ||
998 | #else /* CONFIG_NUMA */ | |
999 | ||
8b98c169 | 1000 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 1001 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 1002 | |
c8522a3a JK |
1003 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
1004 | int batch, gfp_t gfp) | |
1005 | { | |
1006 | int memsize = sizeof(void *) * entries + sizeof(struct alien_cache); | |
1007 | struct alien_cache *alc = NULL; | |
1008 | ||
1009 | alc = kmalloc_node(memsize, gfp, node); | |
1010 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 1011 | spin_lock_init(&alc->lock); |
c8522a3a JK |
1012 | return alc; |
1013 | } | |
1014 | ||
1015 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 1016 | { |
c8522a3a | 1017 | struct alien_cache **alc_ptr; |
8ef82866 | 1018 | int memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
1019 | int i; |
1020 | ||
1021 | if (limit > 1) | |
1022 | limit = 12; | |
c8522a3a JK |
1023 | alc_ptr = kzalloc_node(memsize, gfp, node); |
1024 | if (!alc_ptr) | |
1025 | return NULL; | |
1026 | ||
1027 | for_each_node(i) { | |
1028 | if (i == node || !node_online(i)) | |
1029 | continue; | |
1030 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
1031 | if (!alc_ptr[i]) { | |
1032 | for (i--; i >= 0; i--) | |
1033 | kfree(alc_ptr[i]); | |
1034 | kfree(alc_ptr); | |
1035 | return NULL; | |
e498be7d CL |
1036 | } |
1037 | } | |
c8522a3a | 1038 | return alc_ptr; |
e498be7d CL |
1039 | } |
1040 | ||
c8522a3a | 1041 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
1042 | { |
1043 | int i; | |
1044 | ||
c8522a3a | 1045 | if (!alc_ptr) |
e498be7d | 1046 | return; |
e498be7d | 1047 | for_each_node(i) |
c8522a3a JK |
1048 | kfree(alc_ptr[i]); |
1049 | kfree(alc_ptr); | |
e498be7d CL |
1050 | } |
1051 | ||
343e0d7a | 1052 | static void __drain_alien_cache(struct kmem_cache *cachep, |
5295a74c | 1053 | struct array_cache *ac, int node) |
e498be7d | 1054 | { |
18bf8541 | 1055 | struct kmem_cache_node *n = get_node(cachep, node); |
97654dfa | 1056 | LIST_HEAD(list); |
e498be7d CL |
1057 | |
1058 | if (ac->avail) { | |
ce8eb6c4 | 1059 | spin_lock(&n->list_lock); |
e00946fe CL |
1060 | /* |
1061 | * Stuff objects into the remote nodes shared array first. | |
1062 | * That way we could avoid the overhead of putting the objects | |
1063 | * into the free lists and getting them back later. | |
1064 | */ | |
ce8eb6c4 CL |
1065 | if (n->shared) |
1066 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 1067 | |
97654dfa | 1068 | free_block(cachep, ac->entry, ac->avail, node, &list); |
e498be7d | 1069 | ac->avail = 0; |
ce8eb6c4 | 1070 | spin_unlock(&n->list_lock); |
97654dfa | 1071 | slabs_destroy(cachep, &list); |
e498be7d CL |
1072 | } |
1073 | } | |
1074 | ||
8fce4d8e CL |
1075 | /* |
1076 | * Called from cache_reap() to regularly drain alien caches round robin. | |
1077 | */ | |
ce8eb6c4 | 1078 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 1079 | { |
909ea964 | 1080 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 1081 | |
ce8eb6c4 | 1082 | if (n->alien) { |
c8522a3a JK |
1083 | struct alien_cache *alc = n->alien[node]; |
1084 | struct array_cache *ac; | |
1085 | ||
1086 | if (alc) { | |
1087 | ac = &alc->ac; | |
49dfc304 | 1088 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
c8522a3a | 1089 | __drain_alien_cache(cachep, ac, node); |
49dfc304 | 1090 | spin_unlock_irq(&alc->lock); |
c8522a3a | 1091 | } |
8fce4d8e CL |
1092 | } |
1093 | } | |
1094 | } | |
1095 | ||
a737b3e2 | 1096 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 1097 | struct alien_cache **alien) |
e498be7d | 1098 | { |
b28a02de | 1099 | int i = 0; |
c8522a3a | 1100 | struct alien_cache *alc; |
e498be7d CL |
1101 | struct array_cache *ac; |
1102 | unsigned long flags; | |
1103 | ||
1104 | for_each_online_node(i) { | |
c8522a3a JK |
1105 | alc = alien[i]; |
1106 | if (alc) { | |
1107 | ac = &alc->ac; | |
49dfc304 | 1108 | spin_lock_irqsave(&alc->lock, flags); |
e498be7d | 1109 | __drain_alien_cache(cachep, ac, i); |
49dfc304 | 1110 | spin_unlock_irqrestore(&alc->lock, flags); |
e498be7d CL |
1111 | } |
1112 | } | |
1113 | } | |
729bd0b7 | 1114 | |
873623df | 1115 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 | 1116 | { |
1ea991b0 | 1117 | int nodeid = page_to_nid(virt_to_page(objp)); |
ce8eb6c4 | 1118 | struct kmem_cache_node *n; |
c8522a3a JK |
1119 | struct alien_cache *alien = NULL; |
1120 | struct array_cache *ac; | |
1ca4cb24 | 1121 | int node; |
97654dfa | 1122 | LIST_HEAD(list); |
1ca4cb24 | 1123 | |
7d6e6d09 | 1124 | node = numa_mem_id(); |
729bd0b7 PE |
1125 | |
1126 | /* | |
1127 | * Make sure we are not freeing a object from another node to the array | |
1128 | * cache on this cpu. | |
1129 | */ | |
1ea991b0 | 1130 | if (likely(nodeid == node)) |
729bd0b7 PE |
1131 | return 0; |
1132 | ||
18bf8541 | 1133 | n = get_node(cachep, node); |
729bd0b7 | 1134 | STATS_INC_NODEFREES(cachep); |
ce8eb6c4 CL |
1135 | if (n->alien && n->alien[nodeid]) { |
1136 | alien = n->alien[nodeid]; | |
c8522a3a | 1137 | ac = &alien->ac; |
49dfc304 | 1138 | spin_lock(&alien->lock); |
c8522a3a | 1139 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 1140 | STATS_INC_ACOVERFLOW(cachep); |
c8522a3a | 1141 | __drain_alien_cache(cachep, ac, nodeid); |
729bd0b7 | 1142 | } |
c8522a3a | 1143 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 1144 | spin_unlock(&alien->lock); |
729bd0b7 | 1145 | } else { |
18bf8541 CL |
1146 | n = get_node(cachep, nodeid); |
1147 | spin_lock(&n->list_lock); | |
97654dfa | 1148 | free_block(cachep, &objp, 1, nodeid, &list); |
18bf8541 | 1149 | spin_unlock(&n->list_lock); |
97654dfa | 1150 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
1151 | } |
1152 | return 1; | |
1153 | } | |
e498be7d CL |
1154 | #endif |
1155 | ||
8f9f8d9e | 1156 | /* |
6a67368c | 1157 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1158 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1159 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1160 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1161 | * already in use. |
1162 | * | |
18004c5d | 1163 | * Must hold slab_mutex. |
8f9f8d9e | 1164 | */ |
6a67368c | 1165 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1166 | { |
1167 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1168 | struct kmem_cache_node *n; |
6744f087 | 1169 | const int memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1170 | |
18004c5d | 1171 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 1172 | /* |
5f0985bb | 1173 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
1174 | * begin anything. Make sure some other cpu on this |
1175 | * node has not already allocated this | |
1176 | */ | |
18bf8541 CL |
1177 | n = get_node(cachep, node); |
1178 | if (!n) { | |
ce8eb6c4 CL |
1179 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1180 | if (!n) | |
8f9f8d9e | 1181 | return -ENOMEM; |
ce8eb6c4 | 1182 | kmem_cache_node_init(n); |
5f0985bb JZ |
1183 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1184 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1185 | |
1186 | /* | |
5f0985bb JZ |
1187 | * The kmem_cache_nodes don't come and go as CPUs |
1188 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1189 | * protection here. |
1190 | */ | |
ce8eb6c4 | 1191 | cachep->node[node] = n; |
8f9f8d9e DR |
1192 | } |
1193 | ||
18bf8541 CL |
1194 | spin_lock_irq(&n->list_lock); |
1195 | n->free_limit = | |
8f9f8d9e DR |
1196 | (1 + nr_cpus_node(node)) * |
1197 | cachep->batchcount + cachep->num; | |
18bf8541 | 1198 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1199 | } |
1200 | return 0; | |
1201 | } | |
1202 | ||
0fa8103b WL |
1203 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1204 | struct kmem_cache_node *n) | |
1205 | { | |
1206 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1207 | } | |
1208 | ||
0db0628d | 1209 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1210 | { |
1211 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1212 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1213 | int node = cpu_to_mem(cpu); |
a70f7302 | 1214 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1215 | |
18004c5d | 1216 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1217 | struct array_cache *nc; |
1218 | struct array_cache *shared; | |
c8522a3a | 1219 | struct alien_cache **alien; |
97654dfa | 1220 | LIST_HEAD(list); |
fbf1e473 | 1221 | |
fbf1e473 AM |
1222 | /* cpu is dead; no one can alloc from it. */ |
1223 | nc = cachep->array[cpu]; | |
1224 | cachep->array[cpu] = NULL; | |
18bf8541 | 1225 | n = get_node(cachep, node); |
fbf1e473 | 1226 | |
ce8eb6c4 | 1227 | if (!n) |
fbf1e473 AM |
1228 | goto free_array_cache; |
1229 | ||
ce8eb6c4 | 1230 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1231 | |
ce8eb6c4 CL |
1232 | /* Free limit for this kmem_cache_node */ |
1233 | n->free_limit -= cachep->batchcount; | |
fbf1e473 | 1234 | if (nc) |
97654dfa | 1235 | free_block(cachep, nc->entry, nc->avail, node, &list); |
fbf1e473 | 1236 | |
58463c1f | 1237 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1238 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1239 | goto free_array_cache; |
1240 | } | |
1241 | ||
ce8eb6c4 | 1242 | shared = n->shared; |
fbf1e473 AM |
1243 | if (shared) { |
1244 | free_block(cachep, shared->entry, | |
97654dfa | 1245 | shared->avail, node, &list); |
ce8eb6c4 | 1246 | n->shared = NULL; |
fbf1e473 AM |
1247 | } |
1248 | ||
ce8eb6c4 CL |
1249 | alien = n->alien; |
1250 | n->alien = NULL; | |
fbf1e473 | 1251 | |
ce8eb6c4 | 1252 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1253 | |
1254 | kfree(shared); | |
1255 | if (alien) { | |
1256 | drain_alien_cache(cachep, alien); | |
1257 | free_alien_cache(alien); | |
1258 | } | |
1259 | free_array_cache: | |
97654dfa | 1260 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1261 | kfree(nc); |
1262 | } | |
1263 | /* | |
1264 | * In the previous loop, all the objects were freed to | |
1265 | * the respective cache's slabs, now we can go ahead and | |
1266 | * shrink each nodelist to its limit. | |
1267 | */ | |
18004c5d | 1268 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1269 | n = get_node(cachep, node); |
ce8eb6c4 | 1270 | if (!n) |
fbf1e473 | 1271 | continue; |
0fa8103b | 1272 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1273 | } |
1274 | } | |
1275 | ||
0db0628d | 1276 | static int cpuup_prepare(long cpu) |
1da177e4 | 1277 | { |
343e0d7a | 1278 | struct kmem_cache *cachep; |
ce8eb6c4 | 1279 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1280 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1281 | int err; |
1da177e4 | 1282 | |
fbf1e473 AM |
1283 | /* |
1284 | * We need to do this right in the beginning since | |
1285 | * alloc_arraycache's are going to use this list. | |
1286 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1287 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1288 | */ |
6a67368c | 1289 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1290 | if (err < 0) |
1291 | goto bad; | |
fbf1e473 AM |
1292 | |
1293 | /* | |
1294 | * Now we can go ahead with allocating the shared arrays and | |
1295 | * array caches | |
1296 | */ | |
18004c5d | 1297 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1298 | struct array_cache *nc; |
1299 | struct array_cache *shared = NULL; | |
c8522a3a | 1300 | struct alien_cache **alien = NULL; |
fbf1e473 AM |
1301 | |
1302 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1303 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1304 | if (!nc) |
1305 | goto bad; | |
1306 | if (cachep->shared) { | |
1307 | shared = alloc_arraycache(node, | |
1308 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1309 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1310 | if (!shared) { |
1311 | kfree(nc); | |
1da177e4 | 1312 | goto bad; |
12d00f6a | 1313 | } |
fbf1e473 AM |
1314 | } |
1315 | if (use_alien_caches) { | |
83b519e8 | 1316 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1317 | if (!alien) { |
1318 | kfree(shared); | |
1319 | kfree(nc); | |
fbf1e473 | 1320 | goto bad; |
12d00f6a | 1321 | } |
fbf1e473 AM |
1322 | } |
1323 | cachep->array[cpu] = nc; | |
18bf8541 | 1324 | n = get_node(cachep, node); |
ce8eb6c4 | 1325 | BUG_ON(!n); |
fbf1e473 | 1326 | |
ce8eb6c4 CL |
1327 | spin_lock_irq(&n->list_lock); |
1328 | if (!n->shared) { | |
fbf1e473 AM |
1329 | /* |
1330 | * We are serialised from CPU_DEAD or | |
1331 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1332 | */ | |
ce8eb6c4 | 1333 | n->shared = shared; |
fbf1e473 AM |
1334 | shared = NULL; |
1335 | } | |
4484ebf1 | 1336 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1337 | if (!n->alien) { |
1338 | n->alien = alien; | |
fbf1e473 | 1339 | alien = NULL; |
1da177e4 | 1340 | } |
fbf1e473 | 1341 | #endif |
ce8eb6c4 | 1342 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1343 | kfree(shared); |
1344 | free_alien_cache(alien); | |
83835b3d | 1345 | if (cachep->flags & SLAB_DEBUG_OBJECTS) |
18bf8541 | 1346 | slab_set_debugobj_lock_classes_node(cachep, n); |
6ccfb5bc GC |
1347 | else if (!OFF_SLAB(cachep) && |
1348 | !(cachep->flags & SLAB_DESTROY_BY_RCU)) | |
18bf8541 | 1349 | on_slab_lock_classes_node(cachep, n); |
fbf1e473 | 1350 | } |
ce79ddc8 PE |
1351 | init_node_lock_keys(node); |
1352 | ||
fbf1e473 AM |
1353 | return 0; |
1354 | bad: | |
12d00f6a | 1355 | cpuup_canceled(cpu); |
fbf1e473 AM |
1356 | return -ENOMEM; |
1357 | } | |
1358 | ||
0db0628d | 1359 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1360 | unsigned long action, void *hcpu) |
1361 | { | |
1362 | long cpu = (long)hcpu; | |
1363 | int err = 0; | |
1364 | ||
1365 | switch (action) { | |
fbf1e473 AM |
1366 | case CPU_UP_PREPARE: |
1367 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1368 | mutex_lock(&slab_mutex); |
fbf1e473 | 1369 | err = cpuup_prepare(cpu); |
18004c5d | 1370 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1371 | break; |
1372 | case CPU_ONLINE: | |
8bb78442 | 1373 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1374 | start_cpu_timer(cpu); |
1375 | break; | |
1376 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1377 | case CPU_DOWN_PREPARE: |
8bb78442 | 1378 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1379 | /* |
18004c5d | 1380 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1381 | * held so that if cache_reap() is invoked it cannot do |
1382 | * anything expensive but will only modify reap_work | |
1383 | * and reschedule the timer. | |
1384 | */ | |
afe2c511 | 1385 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1386 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1387 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1388 | break; |
1389 | case CPU_DOWN_FAILED: | |
8bb78442 | 1390 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1391 | start_cpu_timer(cpu); |
1392 | break; | |
1da177e4 | 1393 | case CPU_DEAD: |
8bb78442 | 1394 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1395 | /* |
1396 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1397 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1398 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1399 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1400 | * structure is usually allocated from kmem_cache_create() and |
1401 | * gets destroyed at kmem_cache_destroy(). | |
1402 | */ | |
183ff22b | 1403 | /* fall through */ |
8f5be20b | 1404 | #endif |
1da177e4 | 1405 | case CPU_UP_CANCELED: |
8bb78442 | 1406 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1407 | mutex_lock(&slab_mutex); |
fbf1e473 | 1408 | cpuup_canceled(cpu); |
18004c5d | 1409 | mutex_unlock(&slab_mutex); |
1da177e4 | 1410 | break; |
1da177e4 | 1411 | } |
eac40680 | 1412 | return notifier_from_errno(err); |
1da177e4 LT |
1413 | } |
1414 | ||
0db0628d | 1415 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1416 | &cpuup_callback, NULL, 0 |
1417 | }; | |
1da177e4 | 1418 | |
8f9f8d9e DR |
1419 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1420 | /* | |
1421 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1422 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1423 | * removed. | |
1424 | * | |
18004c5d | 1425 | * Must hold slab_mutex. |
8f9f8d9e | 1426 | */ |
6a67368c | 1427 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1428 | { |
1429 | struct kmem_cache *cachep; | |
1430 | int ret = 0; | |
1431 | ||
18004c5d | 1432 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1433 | struct kmem_cache_node *n; |
8f9f8d9e | 1434 | |
18bf8541 | 1435 | n = get_node(cachep, node); |
ce8eb6c4 | 1436 | if (!n) |
8f9f8d9e DR |
1437 | continue; |
1438 | ||
0fa8103b | 1439 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1440 | |
ce8eb6c4 CL |
1441 | if (!list_empty(&n->slabs_full) || |
1442 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1443 | ret = -EBUSY; |
1444 | break; | |
1445 | } | |
1446 | } | |
1447 | return ret; | |
1448 | } | |
1449 | ||
1450 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1451 | unsigned long action, void *arg) | |
1452 | { | |
1453 | struct memory_notify *mnb = arg; | |
1454 | int ret = 0; | |
1455 | int nid; | |
1456 | ||
1457 | nid = mnb->status_change_nid; | |
1458 | if (nid < 0) | |
1459 | goto out; | |
1460 | ||
1461 | switch (action) { | |
1462 | case MEM_GOING_ONLINE: | |
18004c5d | 1463 | mutex_lock(&slab_mutex); |
6a67368c | 1464 | ret = init_cache_node_node(nid); |
18004c5d | 1465 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1466 | break; |
1467 | case MEM_GOING_OFFLINE: | |
18004c5d | 1468 | mutex_lock(&slab_mutex); |
6a67368c | 1469 | ret = drain_cache_node_node(nid); |
18004c5d | 1470 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1471 | break; |
1472 | case MEM_ONLINE: | |
1473 | case MEM_OFFLINE: | |
1474 | case MEM_CANCEL_ONLINE: | |
1475 | case MEM_CANCEL_OFFLINE: | |
1476 | break; | |
1477 | } | |
1478 | out: | |
5fda1bd5 | 1479 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1480 | } |
1481 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1482 | ||
e498be7d | 1483 | /* |
ce8eb6c4 | 1484 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1485 | */ |
6744f087 | 1486 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1487 | int nodeid) |
e498be7d | 1488 | { |
6744f087 | 1489 | struct kmem_cache_node *ptr; |
e498be7d | 1490 | |
6744f087 | 1491 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1492 | BUG_ON(!ptr); |
1493 | ||
6744f087 | 1494 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1495 | /* |
1496 | * Do not assume that spinlocks can be initialized via memcpy: | |
1497 | */ | |
1498 | spin_lock_init(&ptr->list_lock); | |
1499 | ||
e498be7d | 1500 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1501 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1502 | } |
1503 | ||
556a169d | 1504 | /* |
ce8eb6c4 CL |
1505 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1506 | * size of kmem_cache_node. | |
556a169d | 1507 | */ |
ce8eb6c4 | 1508 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1509 | { |
1510 | int node; | |
1511 | ||
1512 | for_each_online_node(node) { | |
ce8eb6c4 | 1513 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1514 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1515 | REAPTIMEOUT_NODE + |
1516 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1517 | } |
1518 | } | |
1519 | ||
3c583465 CL |
1520 | /* |
1521 | * The memory after the last cpu cache pointer is used for the | |
6a67368c | 1522 | * the node pointer. |
3c583465 | 1523 | */ |
6a67368c | 1524 | static void setup_node_pointer(struct kmem_cache *cachep) |
3c583465 | 1525 | { |
6a67368c | 1526 | cachep->node = (struct kmem_cache_node **)&cachep->array[nr_cpu_ids]; |
3c583465 CL |
1527 | } |
1528 | ||
a737b3e2 AM |
1529 | /* |
1530 | * Initialisation. Called after the page allocator have been initialised and | |
1531 | * before smp_init(). | |
1da177e4 LT |
1532 | */ |
1533 | void __init kmem_cache_init(void) | |
1534 | { | |
e498be7d CL |
1535 | int i; |
1536 | ||
68126702 JK |
1537 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1538 | sizeof(struct rcu_head)); | |
9b030cb8 | 1539 | kmem_cache = &kmem_cache_boot; |
6a67368c | 1540 | setup_node_pointer(kmem_cache); |
9b030cb8 | 1541 | |
b6e68bc1 | 1542 | if (num_possible_nodes() == 1) |
62918a03 SS |
1543 | use_alien_caches = 0; |
1544 | ||
3c583465 | 1545 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1546 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1547 | |
ce8eb6c4 | 1548 | set_up_node(kmem_cache, CACHE_CACHE); |
1da177e4 LT |
1549 | |
1550 | /* | |
1551 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1552 | * page orders on machines with more than 32MB of memory if |
1553 | * not overridden on the command line. | |
1da177e4 | 1554 | */ |
3df1cccd | 1555 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1556 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1557 | |
1da177e4 LT |
1558 | /* Bootstrap is tricky, because several objects are allocated |
1559 | * from caches that do not exist yet: | |
9b030cb8 CL |
1560 | * 1) initialize the kmem_cache cache: it contains the struct |
1561 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1562 | * kmem_cache is statically allocated. | |
e498be7d | 1563 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1564 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1565 | * array at the end of the bootstrap. |
1da177e4 | 1566 | * 2) Create the first kmalloc cache. |
343e0d7a | 1567 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1568 | * An __init data area is used for the head array. |
1569 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1570 | * head arrays. | |
9b030cb8 | 1571 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1572 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1573 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1574 | * the other cache's with kmalloc allocated memory. |
1575 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1576 | */ |
1577 | ||
9b030cb8 | 1578 | /* 1) create the kmem_cache */ |
1da177e4 | 1579 | |
8da3430d | 1580 | /* |
b56efcf0 | 1581 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1582 | */ |
2f9baa9f CL |
1583 | create_boot_cache(kmem_cache, "kmem_cache", |
1584 | offsetof(struct kmem_cache, array[nr_cpu_ids]) + | |
6744f087 | 1585 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1586 | SLAB_HWCACHE_ALIGN); |
1587 | list_add(&kmem_cache->list, &slab_caches); | |
1da177e4 LT |
1588 | |
1589 | /* 2+3) create the kmalloc caches */ | |
1da177e4 | 1590 | |
a737b3e2 AM |
1591 | /* |
1592 | * Initialize the caches that provide memory for the array cache and the | |
ce8eb6c4 | 1593 | * kmem_cache_node structures first. Without this, further allocations will |
a737b3e2 | 1594 | * bug. |
e498be7d CL |
1595 | */ |
1596 | ||
e3366016 CL |
1597 | kmalloc_caches[INDEX_AC] = create_kmalloc_cache("kmalloc-ac", |
1598 | kmalloc_size(INDEX_AC), ARCH_KMALLOC_FLAGS); | |
45530c44 | 1599 | |
ce8eb6c4 CL |
1600 | if (INDEX_AC != INDEX_NODE) |
1601 | kmalloc_caches[INDEX_NODE] = | |
1602 | create_kmalloc_cache("kmalloc-node", | |
1603 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); | |
e498be7d | 1604 | |
e0a42726 IM |
1605 | slab_early_init = 0; |
1606 | ||
1da177e4 LT |
1607 | /* 4) Replace the bootstrap head arrays */ |
1608 | { | |
2b2d5493 | 1609 | struct array_cache *ptr; |
e498be7d | 1610 | |
83b519e8 | 1611 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1612 | |
9b030cb8 | 1613 | memcpy(ptr, cpu_cache_get(kmem_cache), |
b28a02de | 1614 | sizeof(struct arraycache_init)); |
2b2d5493 | 1615 | |
9b030cb8 | 1616 | kmem_cache->array[smp_processor_id()] = ptr; |
e498be7d | 1617 | |
83b519e8 | 1618 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1619 | |
e3366016 | 1620 | BUG_ON(cpu_cache_get(kmalloc_caches[INDEX_AC]) |
b28a02de | 1621 | != &initarray_generic.cache); |
e3366016 | 1622 | memcpy(ptr, cpu_cache_get(kmalloc_caches[INDEX_AC]), |
b28a02de | 1623 | sizeof(struct arraycache_init)); |
2b2d5493 | 1624 | |
e3366016 | 1625 | kmalloc_caches[INDEX_AC]->array[smp_processor_id()] = ptr; |
1da177e4 | 1626 | } |
ce8eb6c4 | 1627 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1628 | { |
1ca4cb24 PE |
1629 | int nid; |
1630 | ||
9c09a95c | 1631 | for_each_online_node(nid) { |
ce8eb6c4 | 1632 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1633 | |
e3366016 | 1634 | init_list(kmalloc_caches[INDEX_AC], |
ce8eb6c4 | 1635 | &init_kmem_cache_node[SIZE_AC + nid], nid); |
e498be7d | 1636 | |
ce8eb6c4 CL |
1637 | if (INDEX_AC != INDEX_NODE) { |
1638 | init_list(kmalloc_caches[INDEX_NODE], | |
1639 | &init_kmem_cache_node[SIZE_NODE + nid], nid); | |
e498be7d CL |
1640 | } |
1641 | } | |
1642 | } | |
1da177e4 | 1643 | |
f97d5f63 | 1644 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1645 | } |
1646 | ||
1647 | void __init kmem_cache_init_late(void) | |
1648 | { | |
1649 | struct kmem_cache *cachep; | |
1650 | ||
97d06609 | 1651 | slab_state = UP; |
52cef189 | 1652 | |
8429db5c | 1653 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1654 | mutex_lock(&slab_mutex); |
1655 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1656 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1657 | BUG(); | |
18004c5d | 1658 | mutex_unlock(&slab_mutex); |
056c6241 | 1659 | |
947ca185 MW |
1660 | /* Annotate slab for lockdep -- annotate the malloc caches */ |
1661 | init_lock_keys(); | |
1662 | ||
97d06609 CL |
1663 | /* Done! */ |
1664 | slab_state = FULL; | |
1665 | ||
a737b3e2 AM |
1666 | /* |
1667 | * Register a cpu startup notifier callback that initializes | |
1668 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1669 | */ |
1670 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1671 | |
8f9f8d9e DR |
1672 | #ifdef CONFIG_NUMA |
1673 | /* | |
1674 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1675 | * node. |
8f9f8d9e DR |
1676 | */ |
1677 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1678 | #endif | |
1679 | ||
a737b3e2 AM |
1680 | /* |
1681 | * The reap timers are started later, with a module init call: That part | |
1682 | * of the kernel is not yet operational. | |
1da177e4 LT |
1683 | */ |
1684 | } | |
1685 | ||
1686 | static int __init cpucache_init(void) | |
1687 | { | |
1688 | int cpu; | |
1689 | ||
a737b3e2 AM |
1690 | /* |
1691 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1692 | */ |
e498be7d | 1693 | for_each_online_cpu(cpu) |
a737b3e2 | 1694 | start_cpu_timer(cpu); |
a164f896 GC |
1695 | |
1696 | /* Done! */ | |
97d06609 | 1697 | slab_state = FULL; |
1da177e4 LT |
1698 | return 0; |
1699 | } | |
1da177e4 LT |
1700 | __initcall(cpucache_init); |
1701 | ||
8bdec192 RA |
1702 | static noinline void |
1703 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1704 | { | |
9a02d699 | 1705 | #if DEBUG |
ce8eb6c4 | 1706 | struct kmem_cache_node *n; |
8456a648 | 1707 | struct page *page; |
8bdec192 RA |
1708 | unsigned long flags; |
1709 | int node; | |
9a02d699 DR |
1710 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1711 | DEFAULT_RATELIMIT_BURST); | |
1712 | ||
1713 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1714 | return; | |
8bdec192 RA |
1715 | |
1716 | printk(KERN_WARNING | |
1717 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1718 | nodeid, gfpflags); | |
1719 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1720 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1721 | |
18bf8541 | 1722 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1723 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1724 | unsigned long active_slabs = 0, num_slabs = 0; | |
1725 | ||
ce8eb6c4 | 1726 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1727 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1728 | active_objs += cachep->num; |
1729 | active_slabs++; | |
1730 | } | |
8456a648 JK |
1731 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1732 | active_objs += page->active; | |
8bdec192 RA |
1733 | active_slabs++; |
1734 | } | |
8456a648 | 1735 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1736 | num_slabs++; |
1737 | ||
ce8eb6c4 CL |
1738 | free_objects += n->free_objects; |
1739 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1740 | |
1741 | num_slabs += active_slabs; | |
1742 | num_objs = num_slabs * cachep->num; | |
1743 | printk(KERN_WARNING | |
1744 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1745 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1746 | free_objects); | |
1747 | } | |
9a02d699 | 1748 | #endif |
8bdec192 RA |
1749 | } |
1750 | ||
1da177e4 LT |
1751 | /* |
1752 | * Interface to system's page allocator. No need to hold the cache-lock. | |
1753 | * | |
1754 | * If we requested dmaable memory, we will get it. Even if we | |
1755 | * did not request dmaable memory, we might get it, but that | |
1756 | * would be relatively rare and ignorable. | |
1757 | */ | |
0c3aa83e JK |
1758 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1759 | int nodeid) | |
1da177e4 LT |
1760 | { |
1761 | struct page *page; | |
e1b6aa6f | 1762 | int nr_pages; |
765c4507 | 1763 | |
a618e89f | 1764 | flags |= cachep->allocflags; |
e12ba74d MG |
1765 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1766 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1767 | |
5dfb4175 VD |
1768 | if (memcg_charge_slab(cachep, flags, cachep->gfporder)) |
1769 | return NULL; | |
1770 | ||
517d0869 | 1771 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1772 | if (!page) { |
5dfb4175 | 1773 | memcg_uncharge_slab(cachep, cachep->gfporder); |
9a02d699 | 1774 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1775 | return NULL; |
8bdec192 | 1776 | } |
1da177e4 | 1777 | |
b37f1dd0 | 1778 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
072bb0aa MG |
1779 | if (unlikely(page->pfmemalloc)) |
1780 | pfmemalloc_active = true; | |
1781 | ||
e1b6aa6f | 1782 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1783 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1784 | add_zone_page_state(page_zone(page), |
1785 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1786 | else | |
1787 | add_zone_page_state(page_zone(page), | |
1788 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 JK |
1789 | __SetPageSlab(page); |
1790 | if (page->pfmemalloc) | |
1791 | SetPageSlabPfmemalloc(page); | |
072bb0aa | 1792 | |
b1eeab67 VN |
1793 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1794 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1795 | ||
1796 | if (cachep->ctor) | |
1797 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1798 | else | |
1799 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1800 | } | |
c175eea4 | 1801 | |
0c3aa83e | 1802 | return page; |
1da177e4 LT |
1803 | } |
1804 | ||
1805 | /* | |
1806 | * Interface to system's page release. | |
1807 | */ | |
0c3aa83e | 1808 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1809 | { |
a57a4988 | 1810 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1811 | |
b1eeab67 | 1812 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1813 | |
972d1a7b CL |
1814 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1815 | sub_zone_page_state(page_zone(page), | |
1816 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1817 | else | |
1818 | sub_zone_page_state(page_zone(page), | |
1819 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1820 | |
a57a4988 | 1821 | BUG_ON(!PageSlab(page)); |
73293c2f | 1822 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1823 | __ClearPageSlab(page); |
8456a648 JK |
1824 | page_mapcount_reset(page); |
1825 | page->mapping = NULL; | |
1f458cbf | 1826 | |
1da177e4 LT |
1827 | if (current->reclaim_state) |
1828 | current->reclaim_state->reclaimed_slab += nr_freed; | |
5dfb4175 VD |
1829 | __free_pages(page, cachep->gfporder); |
1830 | memcg_uncharge_slab(cachep, cachep->gfporder); | |
1da177e4 LT |
1831 | } |
1832 | ||
1833 | static void kmem_rcu_free(struct rcu_head *head) | |
1834 | { | |
68126702 JK |
1835 | struct kmem_cache *cachep; |
1836 | struct page *page; | |
1da177e4 | 1837 | |
68126702 JK |
1838 | page = container_of(head, struct page, rcu_head); |
1839 | cachep = page->slab_cache; | |
1840 | ||
1841 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1842 | } |
1843 | ||
1844 | #if DEBUG | |
1845 | ||
1846 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1847 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1848 | unsigned long caller) |
1da177e4 | 1849 | { |
8c138bc0 | 1850 | int size = cachep->object_size; |
1da177e4 | 1851 | |
3dafccf2 | 1852 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1853 | |
b28a02de | 1854 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1855 | return; |
1856 | ||
b28a02de PE |
1857 | *addr++ = 0x12345678; |
1858 | *addr++ = caller; | |
1859 | *addr++ = smp_processor_id(); | |
1860 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1861 | { |
1862 | unsigned long *sptr = &caller; | |
1863 | unsigned long svalue; | |
1864 | ||
1865 | while (!kstack_end(sptr)) { | |
1866 | svalue = *sptr++; | |
1867 | if (kernel_text_address(svalue)) { | |
b28a02de | 1868 | *addr++ = svalue; |
1da177e4 LT |
1869 | size -= sizeof(unsigned long); |
1870 | if (size <= sizeof(unsigned long)) | |
1871 | break; | |
1872 | } | |
1873 | } | |
1874 | ||
1875 | } | |
b28a02de | 1876 | *addr++ = 0x87654321; |
1da177e4 LT |
1877 | } |
1878 | #endif | |
1879 | ||
343e0d7a | 1880 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1881 | { |
8c138bc0 | 1882 | int size = cachep->object_size; |
3dafccf2 | 1883 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1884 | |
1885 | memset(addr, val, size); | |
b28a02de | 1886 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1887 | } |
1888 | ||
1889 | static void dump_line(char *data, int offset, int limit) | |
1890 | { | |
1891 | int i; | |
aa83aa40 DJ |
1892 | unsigned char error = 0; |
1893 | int bad_count = 0; | |
1894 | ||
fdde6abb | 1895 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1896 | for (i = 0; i < limit; i++) { |
1897 | if (data[offset + i] != POISON_FREE) { | |
1898 | error = data[offset + i]; | |
1899 | bad_count++; | |
1900 | } | |
aa83aa40 | 1901 | } |
fdde6abb SAS |
1902 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1903 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1904 | |
1905 | if (bad_count == 1) { | |
1906 | error ^= POISON_FREE; | |
1907 | if (!(error & (error - 1))) { | |
1908 | printk(KERN_ERR "Single bit error detected. Probably " | |
1909 | "bad RAM.\n"); | |
1910 | #ifdef CONFIG_X86 | |
1911 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1912 | "test tool.\n"); | |
1913 | #else | |
1914 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1915 | #endif | |
1916 | } | |
1917 | } | |
1da177e4 LT |
1918 | } |
1919 | #endif | |
1920 | ||
1921 | #if DEBUG | |
1922 | ||
343e0d7a | 1923 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1924 | { |
1925 | int i, size; | |
1926 | char *realobj; | |
1927 | ||
1928 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1929 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1930 | *dbg_redzone1(cachep, objp), |
1931 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1932 | } |
1933 | ||
1934 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1935 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1936 | *dbg_userword(cachep, objp), | |
1937 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1938 | } |
3dafccf2 | 1939 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1940 | size = cachep->object_size; |
b28a02de | 1941 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1942 | int limit; |
1943 | limit = 16; | |
b28a02de PE |
1944 | if (i + limit > size) |
1945 | limit = size - i; | |
1da177e4 LT |
1946 | dump_line(realobj, i, limit); |
1947 | } | |
1948 | } | |
1949 | ||
343e0d7a | 1950 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1951 | { |
1952 | char *realobj; | |
1953 | int size, i; | |
1954 | int lines = 0; | |
1955 | ||
3dafccf2 | 1956 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1957 | size = cachep->object_size; |
1da177e4 | 1958 | |
b28a02de | 1959 | for (i = 0; i < size; i++) { |
1da177e4 | 1960 | char exp = POISON_FREE; |
b28a02de | 1961 | if (i == size - 1) |
1da177e4 LT |
1962 | exp = POISON_END; |
1963 | if (realobj[i] != exp) { | |
1964 | int limit; | |
1965 | /* Mismatch ! */ | |
1966 | /* Print header */ | |
1967 | if (lines == 0) { | |
b28a02de | 1968 | printk(KERN_ERR |
face37f5 DJ |
1969 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1970 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1971 | print_objinfo(cachep, objp, 0); |
1972 | } | |
1973 | /* Hexdump the affected line */ | |
b28a02de | 1974 | i = (i / 16) * 16; |
1da177e4 | 1975 | limit = 16; |
b28a02de PE |
1976 | if (i + limit > size) |
1977 | limit = size - i; | |
1da177e4 LT |
1978 | dump_line(realobj, i, limit); |
1979 | i += 16; | |
1980 | lines++; | |
1981 | /* Limit to 5 lines */ | |
1982 | if (lines > 5) | |
1983 | break; | |
1984 | } | |
1985 | } | |
1986 | if (lines != 0) { | |
1987 | /* Print some data about the neighboring objects, if they | |
1988 | * exist: | |
1989 | */ | |
8456a648 | 1990 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1991 | unsigned int objnr; |
1da177e4 | 1992 | |
8456a648 | 1993 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1994 | if (objnr) { |
8456a648 | 1995 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1996 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1997 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1998 | realobj, size); |
1da177e4 LT |
1999 | print_objinfo(cachep, objp, 2); |
2000 | } | |
b28a02de | 2001 | if (objnr + 1 < cachep->num) { |
8456a648 | 2002 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 2003 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 2004 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 2005 | realobj, size); |
1da177e4 LT |
2006 | print_objinfo(cachep, objp, 2); |
2007 | } | |
2008 | } | |
2009 | } | |
2010 | #endif | |
2011 | ||
12dd36fa | 2012 | #if DEBUG |
8456a648 JK |
2013 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
2014 | struct page *page) | |
1da177e4 | 2015 | { |
1da177e4 LT |
2016 | int i; |
2017 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2018 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
2019 | |
2020 | if (cachep->flags & SLAB_POISON) { | |
2021 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2022 | if (cachep->size % PAGE_SIZE == 0 && |
a737b3e2 | 2023 | OFF_SLAB(cachep)) |
b28a02de | 2024 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2025 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
2026 | else |
2027 | check_poison_obj(cachep, objp); | |
2028 | #else | |
2029 | check_poison_obj(cachep, objp); | |
2030 | #endif | |
2031 | } | |
2032 | if (cachep->flags & SLAB_RED_ZONE) { | |
2033 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
2034 | slab_error(cachep, "start of a freed object " | |
b28a02de | 2035 | "was overwritten"); |
1da177e4 LT |
2036 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
2037 | slab_error(cachep, "end of a freed object " | |
b28a02de | 2038 | "was overwritten"); |
1da177e4 | 2039 | } |
1da177e4 | 2040 | } |
12dd36fa | 2041 | } |
1da177e4 | 2042 | #else |
8456a648 JK |
2043 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
2044 | struct page *page) | |
12dd36fa | 2045 | { |
12dd36fa | 2046 | } |
1da177e4 LT |
2047 | #endif |
2048 | ||
911851e6 RD |
2049 | /** |
2050 | * slab_destroy - destroy and release all objects in a slab | |
2051 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 2052 | * @page: page pointer being destroyed |
911851e6 | 2053 | * |
12dd36fa | 2054 | * Destroy all the objs in a slab, and release the mem back to the system. |
a737b3e2 AM |
2055 | * Before calling the slab must have been unlinked from the cache. The |
2056 | * cache-lock is not held/needed. | |
12dd36fa | 2057 | */ |
8456a648 | 2058 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 2059 | { |
7e007355 | 2060 | void *freelist; |
12dd36fa | 2061 | |
8456a648 JK |
2062 | freelist = page->freelist; |
2063 | slab_destroy_debugcheck(cachep, page); | |
1da177e4 | 2064 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
68126702 JK |
2065 | struct rcu_head *head; |
2066 | ||
2067 | /* | |
2068 | * RCU free overloads the RCU head over the LRU. | |
2069 | * slab_page has been overloeaded over the LRU, | |
2070 | * however it is not used from now on so that | |
2071 | * we can use it safely. | |
2072 | */ | |
2073 | head = (void *)&page->rcu_head; | |
2074 | call_rcu(head, kmem_rcu_free); | |
1da177e4 | 2075 | |
1da177e4 | 2076 | } else { |
0c3aa83e | 2077 | kmem_freepages(cachep, page); |
1da177e4 | 2078 | } |
68126702 JK |
2079 | |
2080 | /* | |
8456a648 | 2081 | * From now on, we don't use freelist |
68126702 JK |
2082 | * although actual page can be freed in rcu context |
2083 | */ | |
2084 | if (OFF_SLAB(cachep)) | |
8456a648 | 2085 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
2086 | } |
2087 | ||
97654dfa JK |
2088 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
2089 | { | |
2090 | struct page *page, *n; | |
2091 | ||
2092 | list_for_each_entry_safe(page, n, list, lru) { | |
2093 | list_del(&page->lru); | |
2094 | slab_destroy(cachep, page); | |
2095 | } | |
2096 | } | |
2097 | ||
4d268eba | 2098 | /** |
a70773dd RD |
2099 | * calculate_slab_order - calculate size (page order) of slabs |
2100 | * @cachep: pointer to the cache that is being created | |
2101 | * @size: size of objects to be created in this cache. | |
2102 | * @align: required alignment for the objects. | |
2103 | * @flags: slab allocation flags | |
2104 | * | |
2105 | * Also calculates the number of objects per slab. | |
4d268eba PE |
2106 | * |
2107 | * This could be made much more intelligent. For now, try to avoid using | |
2108 | * high order pages for slabs. When the gfp() functions are more friendly | |
2109 | * towards high-order requests, this should be changed. | |
2110 | */ | |
a737b3e2 | 2111 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 2112 | size_t size, size_t align, unsigned long flags) |
4d268eba | 2113 | { |
b1ab41c4 | 2114 | unsigned long offslab_limit; |
4d268eba | 2115 | size_t left_over = 0; |
9888e6fa | 2116 | int gfporder; |
4d268eba | 2117 | |
0aa817f0 | 2118 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
2119 | unsigned int num; |
2120 | size_t remainder; | |
2121 | ||
9888e6fa | 2122 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
2123 | if (!num) |
2124 | continue; | |
9888e6fa | 2125 | |
f315e3fa JK |
2126 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
2127 | if (num > SLAB_OBJ_MAX_NUM) | |
2128 | break; | |
2129 | ||
b1ab41c4 | 2130 | if (flags & CFLGS_OFF_SLAB) { |
03787301 | 2131 | size_t freelist_size_per_obj = sizeof(freelist_idx_t); |
b1ab41c4 IM |
2132 | /* |
2133 | * Max number of objs-per-slab for caches which | |
2134 | * use off-slab slabs. Needed to avoid a possible | |
2135 | * looping condition in cache_grow(). | |
2136 | */ | |
03787301 JK |
2137 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
2138 | freelist_size_per_obj += sizeof(char); | |
8456a648 | 2139 | offslab_limit = size; |
03787301 | 2140 | offslab_limit /= freelist_size_per_obj; |
b1ab41c4 IM |
2141 | |
2142 | if (num > offslab_limit) | |
2143 | break; | |
2144 | } | |
4d268eba | 2145 | |
9888e6fa | 2146 | /* Found something acceptable - save it away */ |
4d268eba | 2147 | cachep->num = num; |
9888e6fa | 2148 | cachep->gfporder = gfporder; |
4d268eba PE |
2149 | left_over = remainder; |
2150 | ||
f78bb8ad LT |
2151 | /* |
2152 | * A VFS-reclaimable slab tends to have most allocations | |
2153 | * as GFP_NOFS and we really don't want to have to be allocating | |
2154 | * higher-order pages when we are unable to shrink dcache. | |
2155 | */ | |
2156 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
2157 | break; | |
2158 | ||
4d268eba PE |
2159 | /* |
2160 | * Large number of objects is good, but very large slabs are | |
2161 | * currently bad for the gfp()s. | |
2162 | */ | |
543585cc | 2163 | if (gfporder >= slab_max_order) |
4d268eba PE |
2164 | break; |
2165 | ||
9888e6fa LT |
2166 | /* |
2167 | * Acceptable internal fragmentation? | |
2168 | */ | |
a737b3e2 | 2169 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
2170 | break; |
2171 | } | |
2172 | return left_over; | |
2173 | } | |
2174 | ||
83b519e8 | 2175 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2176 | { |
97d06609 | 2177 | if (slab_state >= FULL) |
83b519e8 | 2178 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2179 | |
97d06609 | 2180 | if (slab_state == DOWN) { |
f30cf7d1 | 2181 | /* |
2f9baa9f | 2182 | * Note: Creation of first cache (kmem_cache). |
ce8eb6c4 | 2183 | * The setup_node is taken care |
2f9baa9f CL |
2184 | * of by the caller of __kmem_cache_create |
2185 | */ | |
2186 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2187 | slab_state = PARTIAL; | |
2188 | } else if (slab_state == PARTIAL) { | |
2189 | /* | |
2190 | * Note: the second kmem_cache_create must create the cache | |
f30cf7d1 PE |
2191 | * that's used by kmalloc(24), otherwise the creation of |
2192 | * further caches will BUG(). | |
2193 | */ | |
2194 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2195 | ||
2196 | /* | |
ce8eb6c4 CL |
2197 | * If the cache that's used by kmalloc(sizeof(kmem_cache_node)) is |
2198 | * the second cache, then we need to set up all its node/, | |
f30cf7d1 PE |
2199 | * otherwise the creation of further caches will BUG(). |
2200 | */ | |
ce8eb6c4 CL |
2201 | set_up_node(cachep, SIZE_AC); |
2202 | if (INDEX_AC == INDEX_NODE) | |
2203 | slab_state = PARTIAL_NODE; | |
f30cf7d1 | 2204 | else |
97d06609 | 2205 | slab_state = PARTIAL_ARRAYCACHE; |
f30cf7d1 | 2206 | } else { |
2f9baa9f | 2207 | /* Remaining boot caches */ |
f30cf7d1 | 2208 | cachep->array[smp_processor_id()] = |
83b519e8 | 2209 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 | 2210 | |
97d06609 | 2211 | if (slab_state == PARTIAL_ARRAYCACHE) { |
ce8eb6c4 CL |
2212 | set_up_node(cachep, SIZE_NODE); |
2213 | slab_state = PARTIAL_NODE; | |
f30cf7d1 PE |
2214 | } else { |
2215 | int node; | |
556a169d | 2216 | for_each_online_node(node) { |
6a67368c | 2217 | cachep->node[node] = |
6744f087 | 2218 | kmalloc_node(sizeof(struct kmem_cache_node), |
eb91f1d0 | 2219 | gfp, node); |
6a67368c | 2220 | BUG_ON(!cachep->node[node]); |
ce8eb6c4 | 2221 | kmem_cache_node_init(cachep->node[node]); |
f30cf7d1 PE |
2222 | } |
2223 | } | |
2224 | } | |
6a67368c | 2225 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
2226 | jiffies + REAPTIMEOUT_NODE + |
2227 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
2228 | |
2229 | cpu_cache_get(cachep)->avail = 0; | |
2230 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2231 | cpu_cache_get(cachep)->batchcount = 1; | |
2232 | cpu_cache_get(cachep)->touched = 0; | |
2233 | cachep->batchcount = 1; | |
2234 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2235 | return 0; |
f30cf7d1 PE |
2236 | } |
2237 | ||
1da177e4 | 2238 | /** |
039363f3 | 2239 | * __kmem_cache_create - Create a cache. |
a755b76a | 2240 | * @cachep: cache management descriptor |
1da177e4 | 2241 | * @flags: SLAB flags |
1da177e4 LT |
2242 | * |
2243 | * Returns a ptr to the cache on success, NULL on failure. | |
2244 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2245 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2246 | * |
1da177e4 LT |
2247 | * The flags are |
2248 | * | |
2249 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2250 | * to catch references to uninitialised memory. | |
2251 | * | |
2252 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2253 | * for buffer overruns. | |
2254 | * | |
1da177e4 LT |
2255 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2256 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2257 | * as davem. | |
2258 | */ | |
278b1bb1 | 2259 | int |
8a13a4cc | 2260 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2261 | { |
8456a648 | 2262 | size_t left_over, freelist_size, ralign; |
83b519e8 | 2263 | gfp_t gfp; |
278b1bb1 | 2264 | int err; |
8a13a4cc | 2265 | size_t size = cachep->size; |
1da177e4 | 2266 | |
1da177e4 | 2267 | #if DEBUG |
1da177e4 LT |
2268 | #if FORCED_DEBUG |
2269 | /* | |
2270 | * Enable redzoning and last user accounting, except for caches with | |
2271 | * large objects, if the increased size would increase the object size | |
2272 | * above the next power of two: caches with object sizes just above a | |
2273 | * power of two have a significant amount of internal fragmentation. | |
2274 | */ | |
87a927c7 DW |
2275 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2276 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2277 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2278 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2279 | flags |= SLAB_POISON; | |
2280 | #endif | |
2281 | if (flags & SLAB_DESTROY_BY_RCU) | |
2282 | BUG_ON(flags & SLAB_POISON); | |
2283 | #endif | |
1da177e4 | 2284 | |
a737b3e2 AM |
2285 | /* |
2286 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2287 | * unaligned accesses for some archs when redzoning is used, and makes |
2288 | * sure any on-slab bufctl's are also correctly aligned. | |
2289 | */ | |
b28a02de PE |
2290 | if (size & (BYTES_PER_WORD - 1)) { |
2291 | size += (BYTES_PER_WORD - 1); | |
2292 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2293 | } |
2294 | ||
ca5f9703 | 2295 | /* |
87a927c7 DW |
2296 | * Redzoning and user store require word alignment or possibly larger. |
2297 | * Note this will be overridden by architecture or caller mandated | |
2298 | * alignment if either is greater than BYTES_PER_WORD. | |
ca5f9703 | 2299 | */ |
87a927c7 DW |
2300 | if (flags & SLAB_STORE_USER) |
2301 | ralign = BYTES_PER_WORD; | |
2302 | ||
2303 | if (flags & SLAB_RED_ZONE) { | |
2304 | ralign = REDZONE_ALIGN; | |
2305 | /* If redzoning, ensure that the second redzone is suitably | |
2306 | * aligned, by adjusting the object size accordingly. */ | |
2307 | size += REDZONE_ALIGN - 1; | |
2308 | size &= ~(REDZONE_ALIGN - 1); | |
2309 | } | |
ca5f9703 | 2310 | |
a44b56d3 | 2311 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2312 | if (ralign < cachep->align) { |
2313 | ralign = cachep->align; | |
1da177e4 | 2314 | } |
3ff84a7f PE |
2315 | /* disable debug if necessary */ |
2316 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2317 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2318 | /* |
ca5f9703 | 2319 | * 4) Store it. |
1da177e4 | 2320 | */ |
8a13a4cc | 2321 | cachep->align = ralign; |
1da177e4 | 2322 | |
83b519e8 PE |
2323 | if (slab_is_available()) |
2324 | gfp = GFP_KERNEL; | |
2325 | else | |
2326 | gfp = GFP_NOWAIT; | |
2327 | ||
6a67368c | 2328 | setup_node_pointer(cachep); |
1da177e4 | 2329 | #if DEBUG |
1da177e4 | 2330 | |
ca5f9703 PE |
2331 | /* |
2332 | * Both debugging options require word-alignment which is calculated | |
2333 | * into align above. | |
2334 | */ | |
1da177e4 | 2335 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2336 | /* add space for red zone words */ |
3ff84a7f PE |
2337 | cachep->obj_offset += sizeof(unsigned long long); |
2338 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2339 | } |
2340 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2341 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2342 | * the real object. But if the second red zone needs to be |
2343 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2344 | */ |
87a927c7 DW |
2345 | if (flags & SLAB_RED_ZONE) |
2346 | size += REDZONE_ALIGN; | |
2347 | else | |
2348 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2349 | } |
2350 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
ce8eb6c4 | 2351 | if (size >= kmalloc_size(INDEX_NODE + 1) |
608da7e3 TH |
2352 | && cachep->object_size > cache_line_size() |
2353 | && ALIGN(size, cachep->align) < PAGE_SIZE) { | |
2354 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); | |
1da177e4 LT |
2355 | size = PAGE_SIZE; |
2356 | } | |
2357 | #endif | |
2358 | #endif | |
2359 | ||
e0a42726 IM |
2360 | /* |
2361 | * Determine if the slab management is 'on' or 'off' slab. | |
2362 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2363 | * it too early on. Always use on-slab management when |
2364 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2365 | */ |
8fc9cf42 | 2366 | if ((size >= (PAGE_SIZE >> 5)) && !slab_early_init && |
e7cb55b9 | 2367 | !(flags & SLAB_NOLEAKTRACE)) |
1da177e4 LT |
2368 | /* |
2369 | * Size is large, assume best to place the slab management obj | |
2370 | * off-slab (should allow better packing of objs). | |
2371 | */ | |
2372 | flags |= CFLGS_OFF_SLAB; | |
2373 | ||
8a13a4cc | 2374 | size = ALIGN(size, cachep->align); |
f315e3fa JK |
2375 | /* |
2376 | * We should restrict the number of objects in a slab to implement | |
2377 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2378 | */ | |
2379 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2380 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
1da177e4 | 2381 | |
8a13a4cc | 2382 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2383 | |
8a13a4cc | 2384 | if (!cachep->num) |
278b1bb1 | 2385 | return -E2BIG; |
1da177e4 | 2386 | |
03787301 | 2387 | freelist_size = calculate_freelist_size(cachep->num, cachep->align); |
1da177e4 LT |
2388 | |
2389 | /* | |
2390 | * If the slab has been placed off-slab, and we have enough space then | |
2391 | * move it on-slab. This is at the expense of any extra colouring. | |
2392 | */ | |
8456a648 | 2393 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
1da177e4 | 2394 | flags &= ~CFLGS_OFF_SLAB; |
8456a648 | 2395 | left_over -= freelist_size; |
1da177e4 LT |
2396 | } |
2397 | ||
2398 | if (flags & CFLGS_OFF_SLAB) { | |
2399 | /* really off slab. No need for manual alignment */ | |
03787301 | 2400 | freelist_size = calculate_freelist_size(cachep->num, 0); |
67461365 RL |
2401 | |
2402 | #ifdef CONFIG_PAGE_POISONING | |
2403 | /* If we're going to use the generic kernel_map_pages() | |
2404 | * poisoning, then it's going to smash the contents of | |
2405 | * the redzone and userword anyhow, so switch them off. | |
2406 | */ | |
2407 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2408 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2409 | #endif | |
1da177e4 LT |
2410 | } |
2411 | ||
2412 | cachep->colour_off = cache_line_size(); | |
2413 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2414 | if (cachep->colour_off < cachep->align) |
2415 | cachep->colour_off = cachep->align; | |
b28a02de | 2416 | cachep->colour = left_over / cachep->colour_off; |
8456a648 | 2417 | cachep->freelist_size = freelist_size; |
1da177e4 | 2418 | cachep->flags = flags; |
a57a4988 | 2419 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2420 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2421 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2422 | cachep->size = size; |
6a2d7a95 | 2423 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2424 | |
e5ac9c5a | 2425 | if (flags & CFLGS_OFF_SLAB) { |
8456a648 | 2426 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
e5ac9c5a | 2427 | /* |
5f0985bb | 2428 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
e5ac9c5a | 2429 | * But since we go off slab only for object size greater than |
5f0985bb JZ |
2430 | * PAGE_SIZE/8, and kmalloc_{dma,}_caches get created |
2431 | * in ascending order,this should not happen at all. | |
e5ac9c5a RT |
2432 | * But leave a BUG_ON for some lucky dude. |
2433 | */ | |
8456a648 | 2434 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
e5ac9c5a | 2435 | } |
1da177e4 | 2436 | |
278b1bb1 CL |
2437 | err = setup_cpu_cache(cachep, gfp); |
2438 | if (err) { | |
12c3667f | 2439 | __kmem_cache_shutdown(cachep); |
278b1bb1 | 2440 | return err; |
2ed3a4ef | 2441 | } |
1da177e4 | 2442 | |
83835b3d PZ |
2443 | if (flags & SLAB_DEBUG_OBJECTS) { |
2444 | /* | |
2445 | * Would deadlock through slab_destroy()->call_rcu()-> | |
2446 | * debug_object_activate()->kmem_cache_alloc(). | |
2447 | */ | |
2448 | WARN_ON_ONCE(flags & SLAB_DESTROY_BY_RCU); | |
2449 | ||
2450 | slab_set_debugobj_lock_classes(cachep); | |
6ccfb5bc GC |
2451 | } else if (!OFF_SLAB(cachep) && !(flags & SLAB_DESTROY_BY_RCU)) |
2452 | on_slab_lock_classes(cachep); | |
83835b3d | 2453 | |
278b1bb1 | 2454 | return 0; |
1da177e4 | 2455 | } |
1da177e4 LT |
2456 | |
2457 | #if DEBUG | |
2458 | static void check_irq_off(void) | |
2459 | { | |
2460 | BUG_ON(!irqs_disabled()); | |
2461 | } | |
2462 | ||
2463 | static void check_irq_on(void) | |
2464 | { | |
2465 | BUG_ON(irqs_disabled()); | |
2466 | } | |
2467 | ||
343e0d7a | 2468 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2469 | { |
2470 | #ifdef CONFIG_SMP | |
2471 | check_irq_off(); | |
18bf8541 | 2472 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2473 | #endif |
2474 | } | |
e498be7d | 2475 | |
343e0d7a | 2476 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2477 | { |
2478 | #ifdef CONFIG_SMP | |
2479 | check_irq_off(); | |
18bf8541 | 2480 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2481 | #endif |
2482 | } | |
2483 | ||
1da177e4 LT |
2484 | #else |
2485 | #define check_irq_off() do { } while(0) | |
2486 | #define check_irq_on() do { } while(0) | |
2487 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2488 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2489 | #endif |
2490 | ||
ce8eb6c4 | 2491 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2492 | struct array_cache *ac, |
2493 | int force, int node); | |
2494 | ||
1da177e4 LT |
2495 | static void do_drain(void *arg) |
2496 | { | |
a737b3e2 | 2497 | struct kmem_cache *cachep = arg; |
1da177e4 | 2498 | struct array_cache *ac; |
7d6e6d09 | 2499 | int node = numa_mem_id(); |
18bf8541 | 2500 | struct kmem_cache_node *n; |
97654dfa | 2501 | LIST_HEAD(list); |
1da177e4 LT |
2502 | |
2503 | check_irq_off(); | |
9a2dba4b | 2504 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2505 | n = get_node(cachep, node); |
2506 | spin_lock(&n->list_lock); | |
97654dfa | 2507 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2508 | spin_unlock(&n->list_lock); |
97654dfa | 2509 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2510 | ac->avail = 0; |
2511 | } | |
2512 | ||
343e0d7a | 2513 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2514 | { |
ce8eb6c4 | 2515 | struct kmem_cache_node *n; |
e498be7d CL |
2516 | int node; |
2517 | ||
15c8b6c1 | 2518 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2519 | check_irq_on(); |
18bf8541 CL |
2520 | for_each_kmem_cache_node(cachep, node, n) |
2521 | if (n->alien) | |
ce8eb6c4 | 2522 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2523 | |
18bf8541 CL |
2524 | for_each_kmem_cache_node(cachep, node, n) |
2525 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2526 | } |
2527 | ||
ed11d9eb CL |
2528 | /* |
2529 | * Remove slabs from the list of free slabs. | |
2530 | * Specify the number of slabs to drain in tofree. | |
2531 | * | |
2532 | * Returns the actual number of slabs released. | |
2533 | */ | |
2534 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2535 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2536 | { |
ed11d9eb CL |
2537 | struct list_head *p; |
2538 | int nr_freed; | |
8456a648 | 2539 | struct page *page; |
1da177e4 | 2540 | |
ed11d9eb | 2541 | nr_freed = 0; |
ce8eb6c4 | 2542 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2543 | |
ce8eb6c4 CL |
2544 | spin_lock_irq(&n->list_lock); |
2545 | p = n->slabs_free.prev; | |
2546 | if (p == &n->slabs_free) { | |
2547 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2548 | goto out; |
2549 | } | |
1da177e4 | 2550 | |
8456a648 | 2551 | page = list_entry(p, struct page, lru); |
1da177e4 | 2552 | #if DEBUG |
8456a648 | 2553 | BUG_ON(page->active); |
1da177e4 | 2554 | #endif |
8456a648 | 2555 | list_del(&page->lru); |
ed11d9eb CL |
2556 | /* |
2557 | * Safe to drop the lock. The slab is no longer linked | |
2558 | * to the cache. | |
2559 | */ | |
ce8eb6c4 CL |
2560 | n->free_objects -= cache->num; |
2561 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2562 | slab_destroy(cache, page); |
ed11d9eb | 2563 | nr_freed++; |
1da177e4 | 2564 | } |
ed11d9eb CL |
2565 | out: |
2566 | return nr_freed; | |
1da177e4 LT |
2567 | } |
2568 | ||
03afc0e2 | 2569 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d | 2570 | { |
18bf8541 CL |
2571 | int ret = 0; |
2572 | int node; | |
ce8eb6c4 | 2573 | struct kmem_cache_node *n; |
e498be7d CL |
2574 | |
2575 | drain_cpu_caches(cachep); | |
2576 | ||
2577 | check_irq_on(); | |
18bf8541 | 2578 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2579 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2580 | |
ce8eb6c4 CL |
2581 | ret += !list_empty(&n->slabs_full) || |
2582 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2583 | } |
2584 | return (ret ? 1 : 0); | |
2585 | } | |
2586 | ||
945cf2b6 | 2587 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
1da177e4 | 2588 | { |
12c3667f | 2589 | int i; |
ce8eb6c4 | 2590 | struct kmem_cache_node *n; |
03afc0e2 | 2591 | int rc = __kmem_cache_shrink(cachep); |
1da177e4 | 2592 | |
12c3667f CL |
2593 | if (rc) |
2594 | return rc; | |
1da177e4 | 2595 | |
12c3667f CL |
2596 | for_each_online_cpu(i) |
2597 | kfree(cachep->array[i]); | |
1da177e4 | 2598 | |
ce8eb6c4 | 2599 | /* NUMA: free the node structures */ |
18bf8541 CL |
2600 | for_each_kmem_cache_node(cachep, i, n) { |
2601 | kfree(n->shared); | |
2602 | free_alien_cache(n->alien); | |
2603 | kfree(n); | |
2604 | cachep->node[i] = NULL; | |
12c3667f CL |
2605 | } |
2606 | return 0; | |
1da177e4 | 2607 | } |
1da177e4 | 2608 | |
e5ac9c5a RT |
2609 | /* |
2610 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2611 | * |
2612 | * For a slab cache when the slab descriptor is off-slab, the | |
2613 | * slab descriptor can't come from the same cache which is being created, | |
2614 | * Because if it is the case, that means we defer the creation of | |
2615 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2616 | * And we eventually call down to __kmem_cache_create(), which | |
2617 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2618 | * This is a "chicken-and-egg" problem. | |
2619 | * | |
2620 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2621 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2622 | */ |
7e007355 | 2623 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2624 | struct page *page, int colour_off, |
2625 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2626 | { |
7e007355 | 2627 | void *freelist; |
0c3aa83e | 2628 | void *addr = page_address(page); |
b28a02de | 2629 | |
1da177e4 LT |
2630 | if (OFF_SLAB(cachep)) { |
2631 | /* Slab management obj is off-slab. */ | |
8456a648 | 2632 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2633 | local_flags, nodeid); |
8456a648 | 2634 | if (!freelist) |
1da177e4 LT |
2635 | return NULL; |
2636 | } else { | |
8456a648 JK |
2637 | freelist = addr + colour_off; |
2638 | colour_off += cachep->freelist_size; | |
1da177e4 | 2639 | } |
8456a648 JK |
2640 | page->active = 0; |
2641 | page->s_mem = addr + colour_off; | |
2642 | return freelist; | |
1da177e4 LT |
2643 | } |
2644 | ||
7cc68973 | 2645 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2646 | { |
a41adfaa | 2647 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2648 | } |
2649 | ||
2650 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2651 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2652 | { |
a41adfaa | 2653 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2654 | } |
2655 | ||
343e0d7a | 2656 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2657 | struct page *page) |
1da177e4 LT |
2658 | { |
2659 | int i; | |
2660 | ||
2661 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2662 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
2663 | #if DEBUG |
2664 | /* need to poison the objs? */ | |
2665 | if (cachep->flags & SLAB_POISON) | |
2666 | poison_obj(cachep, objp, POISON_FREE); | |
2667 | if (cachep->flags & SLAB_STORE_USER) | |
2668 | *dbg_userword(cachep, objp) = NULL; | |
2669 | ||
2670 | if (cachep->flags & SLAB_RED_ZONE) { | |
2671 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2672 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2673 | } | |
2674 | /* | |
a737b3e2 AM |
2675 | * Constructors are not allowed to allocate memory from the same |
2676 | * cache which they are a constructor for. Otherwise, deadlock. | |
2677 | * They must also be threaded. | |
1da177e4 LT |
2678 | */ |
2679 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2680 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2681 | |
2682 | if (cachep->flags & SLAB_RED_ZONE) { | |
2683 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2684 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2685 | " end of an object"); |
1da177e4 LT |
2686 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2687 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2688 | " start of an object"); |
1da177e4 | 2689 | } |
3b0efdfa | 2690 | if ((cachep->size % PAGE_SIZE) == 0 && |
a737b3e2 | 2691 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
b28a02de | 2692 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2693 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2694 | #else |
2695 | if (cachep->ctor) | |
51cc5068 | 2696 | cachep->ctor(objp); |
1da177e4 | 2697 | #endif |
03787301 | 2698 | set_obj_status(page, i, OBJECT_FREE); |
e5c58dfd | 2699 | set_free_obj(page, i, i); |
1da177e4 | 2700 | } |
1da177e4 LT |
2701 | } |
2702 | ||
343e0d7a | 2703 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2704 | { |
4b51d669 CL |
2705 | if (CONFIG_ZONE_DMA_FLAG) { |
2706 | if (flags & GFP_DMA) | |
a618e89f | 2707 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2708 | else |
a618e89f | 2709 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2710 | } |
1da177e4 LT |
2711 | } |
2712 | ||
8456a648 | 2713 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2714 | int nodeid) |
78d382d7 | 2715 | { |
b1cb0982 | 2716 | void *objp; |
78d382d7 | 2717 | |
e5c58dfd | 2718 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2719 | page->active++; |
78d382d7 | 2720 | #if DEBUG |
1ea991b0 | 2721 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2722 | #endif |
78d382d7 MD |
2723 | |
2724 | return objp; | |
2725 | } | |
2726 | ||
8456a648 | 2727 | static void slab_put_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2728 | void *objp, int nodeid) |
78d382d7 | 2729 | { |
8456a648 | 2730 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2731 | #if DEBUG |
16025177 | 2732 | unsigned int i; |
b1cb0982 | 2733 | |
78d382d7 | 2734 | /* Verify that the slab belongs to the intended node */ |
1ea991b0 | 2735 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2736 | |
b1cb0982 | 2737 | /* Verify double free bug */ |
8456a648 | 2738 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2739 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2740 | printk(KERN_ERR "slab: double free detected in cache " |
2741 | "'%s', objp %p\n", cachep->name, objp); | |
2742 | BUG(); | |
2743 | } | |
78d382d7 MD |
2744 | } |
2745 | #endif | |
8456a648 | 2746 | page->active--; |
e5c58dfd | 2747 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2748 | } |
2749 | ||
4776874f PE |
2750 | /* |
2751 | * Map pages beginning at addr to the given cache and slab. This is required | |
2752 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2753 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2754 | */ |
8456a648 | 2755 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2756 | void *freelist) |
1da177e4 | 2757 | { |
a57a4988 | 2758 | page->slab_cache = cache; |
8456a648 | 2759 | page->freelist = freelist; |
1da177e4 LT |
2760 | } |
2761 | ||
2762 | /* | |
2763 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2764 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2765 | */ | |
3c517a61 | 2766 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2767 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2768 | { |
7e007355 | 2769 | void *freelist; |
b28a02de PE |
2770 | size_t offset; |
2771 | gfp_t local_flags; | |
ce8eb6c4 | 2772 | struct kmem_cache_node *n; |
1da177e4 | 2773 | |
a737b3e2 AM |
2774 | /* |
2775 | * Be lazy and only check for valid flags here, keeping it out of the | |
2776 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2777 | */ |
6cb06229 CL |
2778 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2779 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2780 | |
ce8eb6c4 | 2781 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2782 | check_irq_off(); |
18bf8541 | 2783 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2784 | spin_lock(&n->list_lock); |
1da177e4 LT |
2785 | |
2786 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2787 | offset = n->colour_next; |
2788 | n->colour_next++; | |
2789 | if (n->colour_next >= cachep->colour) | |
2790 | n->colour_next = 0; | |
2791 | spin_unlock(&n->list_lock); | |
1da177e4 | 2792 | |
2e1217cf | 2793 | offset *= cachep->colour_off; |
1da177e4 LT |
2794 | |
2795 | if (local_flags & __GFP_WAIT) | |
2796 | local_irq_enable(); | |
2797 | ||
2798 | /* | |
2799 | * The test for missing atomic flag is performed here, rather than | |
2800 | * the more obvious place, simply to reduce the critical path length | |
2801 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2802 | * will eventually be caught here (where it matters). | |
2803 | */ | |
2804 | kmem_flagcheck(cachep, flags); | |
2805 | ||
a737b3e2 AM |
2806 | /* |
2807 | * Get mem for the objs. Attempt to allocate a physical page from | |
2808 | * 'nodeid'. | |
e498be7d | 2809 | */ |
0c3aa83e JK |
2810 | if (!page) |
2811 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2812 | if (!page) | |
1da177e4 LT |
2813 | goto failed; |
2814 | ||
2815 | /* Get slab management. */ | |
8456a648 | 2816 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2817 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2818 | if (!freelist) |
1da177e4 LT |
2819 | goto opps1; |
2820 | ||
8456a648 | 2821 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2822 | |
8456a648 | 2823 | cache_init_objs(cachep, page); |
1da177e4 LT |
2824 | |
2825 | if (local_flags & __GFP_WAIT) | |
2826 | local_irq_disable(); | |
2827 | check_irq_off(); | |
ce8eb6c4 | 2828 | spin_lock(&n->list_lock); |
1da177e4 LT |
2829 | |
2830 | /* Make slab active. */ | |
8456a648 | 2831 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2832 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2833 | n->free_objects += cachep->num; |
2834 | spin_unlock(&n->list_lock); | |
1da177e4 | 2835 | return 1; |
a737b3e2 | 2836 | opps1: |
0c3aa83e | 2837 | kmem_freepages(cachep, page); |
a737b3e2 | 2838 | failed: |
1da177e4 LT |
2839 | if (local_flags & __GFP_WAIT) |
2840 | local_irq_disable(); | |
2841 | return 0; | |
2842 | } | |
2843 | ||
2844 | #if DEBUG | |
2845 | ||
2846 | /* | |
2847 | * Perform extra freeing checks: | |
2848 | * - detect bad pointers. | |
2849 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2850 | */ |
2851 | static void kfree_debugcheck(const void *objp) | |
2852 | { | |
1da177e4 LT |
2853 | if (!virt_addr_valid(objp)) { |
2854 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2855 | (unsigned long)objp); |
2856 | BUG(); | |
1da177e4 | 2857 | } |
1da177e4 LT |
2858 | } |
2859 | ||
58ce1fd5 PE |
2860 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2861 | { | |
b46b8f19 | 2862 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2863 | |
2864 | redzone1 = *dbg_redzone1(cache, obj); | |
2865 | redzone2 = *dbg_redzone2(cache, obj); | |
2866 | ||
2867 | /* | |
2868 | * Redzone is ok. | |
2869 | */ | |
2870 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2871 | return; | |
2872 | ||
2873 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2874 | slab_error(cache, "double free detected"); | |
2875 | else | |
2876 | slab_error(cache, "memory outside object was overwritten"); | |
2877 | ||
b46b8f19 | 2878 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2879 | obj, redzone1, redzone2); |
2880 | } | |
2881 | ||
343e0d7a | 2882 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2883 | unsigned long caller) |
1da177e4 | 2884 | { |
1da177e4 | 2885 | unsigned int objnr; |
8456a648 | 2886 | struct page *page; |
1da177e4 | 2887 | |
80cbd911 MW |
2888 | BUG_ON(virt_to_cache(objp) != cachep); |
2889 | ||
3dafccf2 | 2890 | objp -= obj_offset(cachep); |
1da177e4 | 2891 | kfree_debugcheck(objp); |
b49af68f | 2892 | page = virt_to_head_page(objp); |
1da177e4 | 2893 | |
1da177e4 | 2894 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2895 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2896 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2897 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2898 | } | |
2899 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2900 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2901 | |
8456a648 | 2902 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2903 | |
2904 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2905 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2906 | |
03787301 | 2907 | set_obj_status(page, objnr, OBJECT_FREE); |
1da177e4 LT |
2908 | if (cachep->flags & SLAB_POISON) { |
2909 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2910 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
7c0cb9c6 | 2911 | store_stackinfo(cachep, objp, caller); |
b28a02de | 2912 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2913 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2914 | } else { |
2915 | poison_obj(cachep, objp, POISON_FREE); | |
2916 | } | |
2917 | #else | |
2918 | poison_obj(cachep, objp, POISON_FREE); | |
2919 | #endif | |
2920 | } | |
2921 | return objp; | |
2922 | } | |
2923 | ||
1da177e4 LT |
2924 | #else |
2925 | #define kfree_debugcheck(x) do { } while(0) | |
2926 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2927 | #endif |
2928 | ||
072bb0aa MG |
2929 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2930 | bool force_refill) | |
1da177e4 LT |
2931 | { |
2932 | int batchcount; | |
ce8eb6c4 | 2933 | struct kmem_cache_node *n; |
1da177e4 | 2934 | struct array_cache *ac; |
1ca4cb24 PE |
2935 | int node; |
2936 | ||
1da177e4 | 2937 | check_irq_off(); |
7d6e6d09 | 2938 | node = numa_mem_id(); |
072bb0aa MG |
2939 | if (unlikely(force_refill)) |
2940 | goto force_grow; | |
2941 | retry: | |
9a2dba4b | 2942 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2943 | batchcount = ac->batchcount; |
2944 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2945 | /* |
2946 | * If there was little recent activity on this cache, then | |
2947 | * perform only a partial refill. Otherwise we could generate | |
2948 | * refill bouncing. | |
1da177e4 LT |
2949 | */ |
2950 | batchcount = BATCHREFILL_LIMIT; | |
2951 | } | |
18bf8541 | 2952 | n = get_node(cachep, node); |
e498be7d | 2953 | |
ce8eb6c4 CL |
2954 | BUG_ON(ac->avail > 0 || !n); |
2955 | spin_lock(&n->list_lock); | |
1da177e4 | 2956 | |
3ded175a | 2957 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2958 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2959 | n->shared->touched = 1; | |
3ded175a | 2960 | goto alloc_done; |
44b57f1c | 2961 | } |
3ded175a | 2962 | |
1da177e4 LT |
2963 | while (batchcount > 0) { |
2964 | struct list_head *entry; | |
8456a648 | 2965 | struct page *page; |
1da177e4 | 2966 | /* Get slab alloc is to come from. */ |
ce8eb6c4 CL |
2967 | entry = n->slabs_partial.next; |
2968 | if (entry == &n->slabs_partial) { | |
2969 | n->free_touched = 1; | |
2970 | entry = n->slabs_free.next; | |
2971 | if (entry == &n->slabs_free) | |
1da177e4 LT |
2972 | goto must_grow; |
2973 | } | |
2974 | ||
8456a648 | 2975 | page = list_entry(entry, struct page, lru); |
1da177e4 | 2976 | check_spinlock_acquired(cachep); |
714b8171 PE |
2977 | |
2978 | /* | |
2979 | * The slab was either on partial or free list so | |
2980 | * there must be at least one object available for | |
2981 | * allocation. | |
2982 | */ | |
8456a648 | 2983 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2984 | |
8456a648 | 2985 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2986 | STATS_INC_ALLOCED(cachep); |
2987 | STATS_INC_ACTIVE(cachep); | |
2988 | STATS_SET_HIGH(cachep); | |
2989 | ||
8456a648 | 2990 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page, |
072bb0aa | 2991 | node)); |
1da177e4 | 2992 | } |
1da177e4 LT |
2993 | |
2994 | /* move slabp to correct slabp list: */ | |
8456a648 JK |
2995 | list_del(&page->lru); |
2996 | if (page->active == cachep->num) | |
34bf6ef9 | 2997 | list_add(&page->lru, &n->slabs_full); |
1da177e4 | 2998 | else |
34bf6ef9 | 2999 | list_add(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3000 | } |
3001 | ||
a737b3e2 | 3002 | must_grow: |
ce8eb6c4 | 3003 | n->free_objects -= ac->avail; |
a737b3e2 | 3004 | alloc_done: |
ce8eb6c4 | 3005 | spin_unlock(&n->list_lock); |
1da177e4 LT |
3006 | |
3007 | if (unlikely(!ac->avail)) { | |
3008 | int x; | |
072bb0aa | 3009 | force_grow: |
3c517a61 | 3010 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 3011 | |
a737b3e2 | 3012 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 3013 | ac = cpu_cache_get(cachep); |
51cd8e6f | 3014 | node = numa_mem_id(); |
072bb0aa MG |
3015 | |
3016 | /* no objects in sight? abort */ | |
3017 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
3018 | return NULL; |
3019 | ||
a737b3e2 | 3020 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
3021 | goto retry; |
3022 | } | |
3023 | ac->touched = 1; | |
072bb0aa MG |
3024 | |
3025 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
3026 | } |
3027 | ||
a737b3e2 AM |
3028 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3029 | gfp_t flags) | |
1da177e4 LT |
3030 | { |
3031 | might_sleep_if(flags & __GFP_WAIT); | |
3032 | #if DEBUG | |
3033 | kmem_flagcheck(cachep, flags); | |
3034 | #endif | |
3035 | } | |
3036 | ||
3037 | #if DEBUG | |
a737b3e2 | 3038 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 3039 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 3040 | { |
03787301 JK |
3041 | struct page *page; |
3042 | ||
b28a02de | 3043 | if (!objp) |
1da177e4 | 3044 | return objp; |
b28a02de | 3045 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3046 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3b0efdfa | 3047 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 3048 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 3049 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
3050 | else |
3051 | check_poison_obj(cachep, objp); | |
3052 | #else | |
3053 | check_poison_obj(cachep, objp); | |
3054 | #endif | |
3055 | poison_obj(cachep, objp, POISON_INUSE); | |
3056 | } | |
3057 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 3058 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
3059 | |
3060 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3061 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3062 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
3063 | slab_error(cachep, "double free, or memory outside" | |
3064 | " object was overwritten"); | |
b28a02de | 3065 | printk(KERN_ERR |
b46b8f19 | 3066 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
3067 | objp, *dbg_redzone1(cachep, objp), |
3068 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3069 | } |
3070 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3071 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3072 | } | |
03787301 JK |
3073 | |
3074 | page = virt_to_head_page(objp); | |
3075 | set_obj_status(page, obj_to_index(cachep, page, objp), OBJECT_ACTIVE); | |
3dafccf2 | 3076 | objp += obj_offset(cachep); |
4f104934 | 3077 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3078 | cachep->ctor(objp); |
7ea466f2 TH |
3079 | if (ARCH_SLAB_MINALIGN && |
3080 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 3081 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3082 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3083 | } |
1da177e4 LT |
3084 | return objp; |
3085 | } | |
3086 | #else | |
3087 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3088 | #endif | |
3089 | ||
773ff60e | 3090 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 | 3091 | { |
8a9c61d4 | 3092 | if (unlikely(cachep == kmem_cache)) |
773ff60e | 3093 | return false; |
8a8b6502 | 3094 | |
8c138bc0 | 3095 | return should_failslab(cachep->object_size, flags, cachep->flags); |
8a8b6502 AM |
3096 | } |
3097 | ||
343e0d7a | 3098 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3099 | { |
b28a02de | 3100 | void *objp; |
1da177e4 | 3101 | struct array_cache *ac; |
072bb0aa | 3102 | bool force_refill = false; |
1da177e4 | 3103 | |
5c382300 | 3104 | check_irq_off(); |
8a8b6502 | 3105 | |
9a2dba4b | 3106 | ac = cpu_cache_get(cachep); |
1da177e4 | 3107 | if (likely(ac->avail)) { |
1da177e4 | 3108 | ac->touched = 1; |
072bb0aa MG |
3109 | objp = ac_get_obj(cachep, ac, flags, false); |
3110 | ||
ddbf2e83 | 3111 | /* |
072bb0aa MG |
3112 | * Allow for the possibility all avail objects are not allowed |
3113 | * by the current flags | |
ddbf2e83 | 3114 | */ |
072bb0aa MG |
3115 | if (objp) { |
3116 | STATS_INC_ALLOCHIT(cachep); | |
3117 | goto out; | |
3118 | } | |
3119 | force_refill = true; | |
1da177e4 | 3120 | } |
072bb0aa MG |
3121 | |
3122 | STATS_INC_ALLOCMISS(cachep); | |
3123 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
3124 | /* | |
3125 | * the 'ac' may be updated by cache_alloc_refill(), | |
3126 | * and kmemleak_erase() requires its correct value. | |
3127 | */ | |
3128 | ac = cpu_cache_get(cachep); | |
3129 | ||
3130 | out: | |
d5cff635 CM |
3131 | /* |
3132 | * To avoid a false negative, if an object that is in one of the | |
3133 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3134 | * treat the array pointers as a reference to the object. | |
3135 | */ | |
f3d8b53a O |
3136 | if (objp) |
3137 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3138 | return objp; |
3139 | } | |
3140 | ||
e498be7d | 3141 | #ifdef CONFIG_NUMA |
c61afb18 | 3142 | /* |
f0432d15 | 3143 | * Try allocating on another node if PF_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3144 | * |
3145 | * If we are in_interrupt, then process context, including cpusets and | |
3146 | * mempolicy, may not apply and should not be used for allocation policy. | |
3147 | */ | |
3148 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3149 | { | |
3150 | int nid_alloc, nid_here; | |
3151 | ||
765c4507 | 3152 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3153 | return NULL; |
7d6e6d09 | 3154 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3155 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3156 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3157 | else if (current->mempolicy) |
2a389610 | 3158 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3159 | if (nid_alloc != nid_here) |
8b98c169 | 3160 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3161 | return NULL; |
3162 | } | |
3163 | ||
765c4507 CL |
3164 | /* |
3165 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3166 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3167 | * available node for available objects. If that fails then we |
3c517a61 CL |
3168 | * perform an allocation without specifying a node. This allows the page |
3169 | * allocator to do its reclaim / fallback magic. We then insert the | |
3170 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3171 | */ |
8c8cc2c1 | 3172 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3173 | { |
8c8cc2c1 PE |
3174 | struct zonelist *zonelist; |
3175 | gfp_t local_flags; | |
dd1a239f | 3176 | struct zoneref *z; |
54a6eb5c MG |
3177 | struct zone *zone; |
3178 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3179 | void *obj = NULL; |
3c517a61 | 3180 | int nid; |
cc9a6c87 | 3181 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3182 | |
3183 | if (flags & __GFP_THISNODE) | |
3184 | return NULL; | |
3185 | ||
6cb06229 | 3186 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3187 | |
cc9a6c87 | 3188 | retry_cpuset: |
d26914d1 | 3189 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3190 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3191 | |
3c517a61 CL |
3192 | retry: |
3193 | /* | |
3194 | * Look through allowed nodes for objects available | |
3195 | * from existing per node queues. | |
3196 | */ | |
54a6eb5c MG |
3197 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3198 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3199 | |
54a6eb5c | 3200 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
18bf8541 CL |
3201 | get_node(cache, nid) && |
3202 | get_node(cache, nid)->free_objects) { | |
3c517a61 CL |
3203 | obj = ____cache_alloc_node(cache, |
3204 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3205 | if (obj) |
3206 | break; | |
3207 | } | |
3c517a61 CL |
3208 | } |
3209 | ||
cfce6604 | 3210 | if (!obj) { |
3c517a61 CL |
3211 | /* |
3212 | * This allocation will be performed within the constraints | |
3213 | * of the current cpuset / memory policy requirements. | |
3214 | * We may trigger various forms of reclaim on the allowed | |
3215 | * set and go into memory reserves if necessary. | |
3216 | */ | |
0c3aa83e JK |
3217 | struct page *page; |
3218 | ||
dd47ea75 CL |
3219 | if (local_flags & __GFP_WAIT) |
3220 | local_irq_enable(); | |
3221 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3222 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
dd47ea75 CL |
3223 | if (local_flags & __GFP_WAIT) |
3224 | local_irq_disable(); | |
0c3aa83e | 3225 | if (page) { |
3c517a61 CL |
3226 | /* |
3227 | * Insert into the appropriate per node queues | |
3228 | */ | |
0c3aa83e JK |
3229 | nid = page_to_nid(page); |
3230 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 CL |
3231 | obj = ____cache_alloc_node(cache, |
3232 | flags | GFP_THISNODE, nid); | |
3233 | if (!obj) | |
3234 | /* | |
3235 | * Another processor may allocate the | |
3236 | * objects in the slab since we are | |
3237 | * not holding any locks. | |
3238 | */ | |
3239 | goto retry; | |
3240 | } else { | |
b6a60451 | 3241 | /* cache_grow already freed obj */ |
3c517a61 CL |
3242 | obj = NULL; |
3243 | } | |
3244 | } | |
aedb0eb1 | 3245 | } |
cc9a6c87 | 3246 | |
d26914d1 | 3247 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3248 | goto retry_cpuset; |
765c4507 CL |
3249 | return obj; |
3250 | } | |
3251 | ||
e498be7d CL |
3252 | /* |
3253 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3254 | */ |
8b98c169 | 3255 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3256 | int nodeid) |
e498be7d CL |
3257 | { |
3258 | struct list_head *entry; | |
8456a648 | 3259 | struct page *page; |
ce8eb6c4 | 3260 | struct kmem_cache_node *n; |
b28a02de | 3261 | void *obj; |
b28a02de PE |
3262 | int x; |
3263 | ||
14e50c6a | 3264 | VM_BUG_ON(nodeid > num_online_nodes()); |
18bf8541 | 3265 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3266 | BUG_ON(!n); |
b28a02de | 3267 | |
a737b3e2 | 3268 | retry: |
ca3b9b91 | 3269 | check_irq_off(); |
ce8eb6c4 CL |
3270 | spin_lock(&n->list_lock); |
3271 | entry = n->slabs_partial.next; | |
3272 | if (entry == &n->slabs_partial) { | |
3273 | n->free_touched = 1; | |
3274 | entry = n->slabs_free.next; | |
3275 | if (entry == &n->slabs_free) | |
b28a02de PE |
3276 | goto must_grow; |
3277 | } | |
3278 | ||
8456a648 | 3279 | page = list_entry(entry, struct page, lru); |
b28a02de | 3280 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3281 | |
3282 | STATS_INC_NODEALLOCS(cachep); | |
3283 | STATS_INC_ACTIVE(cachep); | |
3284 | STATS_SET_HIGH(cachep); | |
3285 | ||
8456a648 | 3286 | BUG_ON(page->active == cachep->num); |
b28a02de | 3287 | |
8456a648 | 3288 | obj = slab_get_obj(cachep, page, nodeid); |
ce8eb6c4 | 3289 | n->free_objects--; |
b28a02de | 3290 | /* move slabp to correct slabp list: */ |
8456a648 | 3291 | list_del(&page->lru); |
b28a02de | 3292 | |
8456a648 JK |
3293 | if (page->active == cachep->num) |
3294 | list_add(&page->lru, &n->slabs_full); | |
a737b3e2 | 3295 | else |
8456a648 | 3296 | list_add(&page->lru, &n->slabs_partial); |
e498be7d | 3297 | |
ce8eb6c4 | 3298 | spin_unlock(&n->list_lock); |
b28a02de | 3299 | goto done; |
e498be7d | 3300 | |
a737b3e2 | 3301 | must_grow: |
ce8eb6c4 | 3302 | spin_unlock(&n->list_lock); |
3c517a61 | 3303 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3304 | if (x) |
3305 | goto retry; | |
1da177e4 | 3306 | |
8c8cc2c1 | 3307 | return fallback_alloc(cachep, flags); |
e498be7d | 3308 | |
a737b3e2 | 3309 | done: |
b28a02de | 3310 | return obj; |
e498be7d | 3311 | } |
8c8cc2c1 | 3312 | |
8c8cc2c1 | 3313 | static __always_inline void * |
48356303 | 3314 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3315 | unsigned long caller) |
8c8cc2c1 PE |
3316 | { |
3317 | unsigned long save_flags; | |
3318 | void *ptr; | |
7d6e6d09 | 3319 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3320 | |
dcce284a | 3321 | flags &= gfp_allowed_mask; |
7e85ee0c | 3322 | |
cf40bd16 NP |
3323 | lockdep_trace_alloc(flags); |
3324 | ||
773ff60e | 3325 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3326 | return NULL; |
3327 | ||
d79923fa GC |
3328 | cachep = memcg_kmem_get_cache(cachep, flags); |
3329 | ||
8c8cc2c1 PE |
3330 | cache_alloc_debugcheck_before(cachep, flags); |
3331 | local_irq_save(save_flags); | |
3332 | ||
eacbbae3 | 3333 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3334 | nodeid = slab_node; |
8c8cc2c1 | 3335 | |
18bf8541 | 3336 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3337 | /* Node not bootstrapped yet */ |
3338 | ptr = fallback_alloc(cachep, flags); | |
3339 | goto out; | |
3340 | } | |
3341 | ||
7d6e6d09 | 3342 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3343 | /* |
3344 | * Use the locally cached objects if possible. | |
3345 | * However ____cache_alloc does not allow fallback | |
3346 | * to other nodes. It may fail while we still have | |
3347 | * objects on other nodes available. | |
3348 | */ | |
3349 | ptr = ____cache_alloc(cachep, flags); | |
3350 | if (ptr) | |
3351 | goto out; | |
3352 | } | |
3353 | /* ___cache_alloc_node can fall back to other nodes */ | |
3354 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3355 | out: | |
3356 | local_irq_restore(save_flags); | |
3357 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
8c138bc0 | 3358 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3359 | flags); |
8c8cc2c1 | 3360 | |
5087c822 | 3361 | if (likely(ptr)) { |
8c138bc0 | 3362 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
5087c822 JP |
3363 | if (unlikely(flags & __GFP_ZERO)) |
3364 | memset(ptr, 0, cachep->object_size); | |
3365 | } | |
d07dbea4 | 3366 | |
8c8cc2c1 PE |
3367 | return ptr; |
3368 | } | |
3369 | ||
3370 | static __always_inline void * | |
3371 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3372 | { | |
3373 | void *objp; | |
3374 | ||
f0432d15 | 3375 | if (current->mempolicy || unlikely(current->flags & PF_SPREAD_SLAB)) { |
8c8cc2c1 PE |
3376 | objp = alternate_node_alloc(cache, flags); |
3377 | if (objp) | |
3378 | goto out; | |
3379 | } | |
3380 | objp = ____cache_alloc(cache, flags); | |
3381 | ||
3382 | /* | |
3383 | * We may just have run out of memory on the local node. | |
3384 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3385 | */ | |
7d6e6d09 LS |
3386 | if (!objp) |
3387 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3388 | |
3389 | out: | |
3390 | return objp; | |
3391 | } | |
3392 | #else | |
3393 | ||
3394 | static __always_inline void * | |
3395 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3396 | { | |
3397 | return ____cache_alloc(cachep, flags); | |
3398 | } | |
3399 | ||
3400 | #endif /* CONFIG_NUMA */ | |
3401 | ||
3402 | static __always_inline void * | |
48356303 | 3403 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3404 | { |
3405 | unsigned long save_flags; | |
3406 | void *objp; | |
3407 | ||
dcce284a | 3408 | flags &= gfp_allowed_mask; |
7e85ee0c | 3409 | |
cf40bd16 NP |
3410 | lockdep_trace_alloc(flags); |
3411 | ||
773ff60e | 3412 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3413 | return NULL; |
3414 | ||
d79923fa GC |
3415 | cachep = memcg_kmem_get_cache(cachep, flags); |
3416 | ||
8c8cc2c1 PE |
3417 | cache_alloc_debugcheck_before(cachep, flags); |
3418 | local_irq_save(save_flags); | |
3419 | objp = __do_cache_alloc(cachep, flags); | |
3420 | local_irq_restore(save_flags); | |
3421 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
8c138bc0 | 3422 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3423 | flags); |
8c8cc2c1 PE |
3424 | prefetchw(objp); |
3425 | ||
5087c822 | 3426 | if (likely(objp)) { |
8c138bc0 | 3427 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
5087c822 JP |
3428 | if (unlikely(flags & __GFP_ZERO)) |
3429 | memset(objp, 0, cachep->object_size); | |
3430 | } | |
d07dbea4 | 3431 | |
8c8cc2c1 PE |
3432 | return objp; |
3433 | } | |
e498be7d CL |
3434 | |
3435 | /* | |
5f0985bb | 3436 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3437 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3438 | */ |
97654dfa JK |
3439 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3440 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3441 | { |
3442 | int i; | |
25c063fb | 3443 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3444 | |
3445 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3446 | void *objp; |
8456a648 | 3447 | struct page *page; |
1da177e4 | 3448 | |
072bb0aa MG |
3449 | clear_obj_pfmemalloc(&objpp[i]); |
3450 | objp = objpp[i]; | |
3451 | ||
8456a648 | 3452 | page = virt_to_head_page(objp); |
8456a648 | 3453 | list_del(&page->lru); |
ff69416e | 3454 | check_spinlock_acquired_node(cachep, node); |
8456a648 | 3455 | slab_put_obj(cachep, page, objp, node); |
1da177e4 | 3456 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3457 | n->free_objects++; |
1da177e4 LT |
3458 | |
3459 | /* fixup slab chains */ | |
8456a648 | 3460 | if (page->active == 0) { |
ce8eb6c4 CL |
3461 | if (n->free_objects > n->free_limit) { |
3462 | n->free_objects -= cachep->num; | |
97654dfa | 3463 | list_add_tail(&page->lru, list); |
1da177e4 | 3464 | } else { |
8456a648 | 3465 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3466 | } |
3467 | } else { | |
3468 | /* Unconditionally move a slab to the end of the | |
3469 | * partial list on free - maximum time for the | |
3470 | * other objects to be freed, too. | |
3471 | */ | |
8456a648 | 3472 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3473 | } |
3474 | } | |
3475 | } | |
3476 | ||
343e0d7a | 3477 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3478 | { |
3479 | int batchcount; | |
ce8eb6c4 | 3480 | struct kmem_cache_node *n; |
7d6e6d09 | 3481 | int node = numa_mem_id(); |
97654dfa | 3482 | LIST_HEAD(list); |
1da177e4 LT |
3483 | |
3484 | batchcount = ac->batchcount; | |
3485 | #if DEBUG | |
3486 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3487 | #endif | |
3488 | check_irq_off(); | |
18bf8541 | 3489 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3490 | spin_lock(&n->list_lock); |
3491 | if (n->shared) { | |
3492 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3493 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3494 | if (max) { |
3495 | if (batchcount > max) | |
3496 | batchcount = max; | |
e498be7d | 3497 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3498 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3499 | shared_array->avail += batchcount; |
3500 | goto free_done; | |
3501 | } | |
3502 | } | |
3503 | ||
97654dfa | 3504 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3505 | free_done: |
1da177e4 LT |
3506 | #if STATS |
3507 | { | |
3508 | int i = 0; | |
3509 | struct list_head *p; | |
3510 | ||
ce8eb6c4 CL |
3511 | p = n->slabs_free.next; |
3512 | while (p != &(n->slabs_free)) { | |
8456a648 | 3513 | struct page *page; |
1da177e4 | 3514 | |
8456a648 JK |
3515 | page = list_entry(p, struct page, lru); |
3516 | BUG_ON(page->active); | |
1da177e4 LT |
3517 | |
3518 | i++; | |
3519 | p = p->next; | |
3520 | } | |
3521 | STATS_SET_FREEABLE(cachep, i); | |
3522 | } | |
3523 | #endif | |
ce8eb6c4 | 3524 | spin_unlock(&n->list_lock); |
97654dfa | 3525 | slabs_destroy(cachep, &list); |
1da177e4 | 3526 | ac->avail -= batchcount; |
a737b3e2 | 3527 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3528 | } |
3529 | ||
3530 | /* | |
a737b3e2 AM |
3531 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3532 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3533 | */ |
a947eb95 | 3534 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3535 | unsigned long caller) |
1da177e4 | 3536 | { |
9a2dba4b | 3537 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3538 | |
3539 | check_irq_off(); | |
d5cff635 | 3540 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3541 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3542 | |
8c138bc0 | 3543 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3544 | |
1807a1aa SS |
3545 | /* |
3546 | * Skip calling cache_free_alien() when the platform is not numa. | |
3547 | * This will avoid cache misses that happen while accessing slabp (which | |
3548 | * is per page memory reference) to get nodeid. Instead use a global | |
3549 | * variable to skip the call, which is mostly likely to be present in | |
3550 | * the cache. | |
3551 | */ | |
b6e68bc1 | 3552 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3553 | return; |
3554 | ||
1da177e4 LT |
3555 | if (likely(ac->avail < ac->limit)) { |
3556 | STATS_INC_FREEHIT(cachep); | |
1da177e4 LT |
3557 | } else { |
3558 | STATS_INC_FREEMISS(cachep); | |
3559 | cache_flusharray(cachep, ac); | |
1da177e4 | 3560 | } |
42c8c99c | 3561 | |
072bb0aa | 3562 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3563 | } |
3564 | ||
3565 | /** | |
3566 | * kmem_cache_alloc - Allocate an object | |
3567 | * @cachep: The cache to allocate from. | |
3568 | * @flags: See kmalloc(). | |
3569 | * | |
3570 | * Allocate an object from this cache. The flags are only relevant | |
3571 | * if the cache has no available objects. | |
3572 | */ | |
343e0d7a | 3573 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3574 | { |
48356303 | 3575 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3576 | |
ca2b84cb | 3577 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3578 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3579 | |
3580 | return ret; | |
1da177e4 LT |
3581 | } |
3582 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3583 | ||
0f24f128 | 3584 | #ifdef CONFIG_TRACING |
85beb586 | 3585 | void * |
4052147c | 3586 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3587 | { |
85beb586 SR |
3588 | void *ret; |
3589 | ||
48356303 | 3590 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3591 | |
3592 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3593 | size, cachep->size, flags); |
85beb586 | 3594 | return ret; |
36555751 | 3595 | } |
85beb586 | 3596 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3597 | #endif |
3598 | ||
1da177e4 | 3599 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3600 | /** |
3601 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3602 | * @cachep: The cache to allocate from. | |
3603 | * @flags: See kmalloc(). | |
3604 | * @nodeid: node number of the target node. | |
3605 | * | |
3606 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3607 | * node, which can improve the performance for cpu bound structures. | |
3608 | * | |
3609 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3610 | */ | |
8b98c169 CH |
3611 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3612 | { | |
48356303 | 3613 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3614 | |
ca2b84cb | 3615 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3616 | cachep->object_size, cachep->size, |
ca2b84cb | 3617 | flags, nodeid); |
36555751 EGM |
3618 | |
3619 | return ret; | |
8b98c169 | 3620 | } |
1da177e4 LT |
3621 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3622 | ||
0f24f128 | 3623 | #ifdef CONFIG_TRACING |
4052147c | 3624 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3625 | gfp_t flags, |
4052147c EG |
3626 | int nodeid, |
3627 | size_t size) | |
36555751 | 3628 | { |
85beb586 SR |
3629 | void *ret; |
3630 | ||
592f4145 | 3631 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3632 | |
85beb586 | 3633 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3634 | size, cachep->size, |
85beb586 SR |
3635 | flags, nodeid); |
3636 | return ret; | |
36555751 | 3637 | } |
85beb586 | 3638 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3639 | #endif |
3640 | ||
8b98c169 | 3641 | static __always_inline void * |
7c0cb9c6 | 3642 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3643 | { |
343e0d7a | 3644 | struct kmem_cache *cachep; |
97e2bde4 | 3645 | |
2c59dd65 | 3646 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3647 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3648 | return cachep; | |
4052147c | 3649 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3650 | } |
8b98c169 | 3651 | |
0bb38a5c | 3652 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
8b98c169 CH |
3653 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3654 | { | |
7c0cb9c6 | 3655 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3656 | } |
dbe5e69d | 3657 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3658 | |
3659 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3660 | int node, unsigned long caller) |
8b98c169 | 3661 | { |
7c0cb9c6 | 3662 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3663 | } |
3664 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
3665 | #else | |
3666 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3667 | { | |
7c0cb9c6 | 3668 | return __do_kmalloc_node(size, flags, node, 0); |
8b98c169 CH |
3669 | } |
3670 | EXPORT_SYMBOL(__kmalloc_node); | |
0bb38a5c | 3671 | #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */ |
8b98c169 | 3672 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3673 | |
3674 | /** | |
800590f5 | 3675 | * __do_kmalloc - allocate memory |
1da177e4 | 3676 | * @size: how many bytes of memory are required. |
800590f5 | 3677 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3678 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3679 | */ |
7fd6b141 | 3680 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3681 | unsigned long caller) |
1da177e4 | 3682 | { |
343e0d7a | 3683 | struct kmem_cache *cachep; |
36555751 | 3684 | void *ret; |
1da177e4 | 3685 | |
2c59dd65 | 3686 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3687 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3688 | return cachep; | |
48356303 | 3689 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3690 | |
7c0cb9c6 | 3691 | trace_kmalloc(caller, ret, |
3b0efdfa | 3692 | size, cachep->size, flags); |
36555751 EGM |
3693 | |
3694 | return ret; | |
7fd6b141 PE |
3695 | } |
3696 | ||
7fd6b141 | 3697 | |
0bb38a5c | 3698 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
7fd6b141 PE |
3699 | void *__kmalloc(size_t size, gfp_t flags) |
3700 | { | |
7c0cb9c6 | 3701 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3702 | } |
3703 | EXPORT_SYMBOL(__kmalloc); | |
3704 | ||
ce71e27c | 3705 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3706 | { |
7c0cb9c6 | 3707 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3708 | } |
3709 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea CH |
3710 | |
3711 | #else | |
3712 | void *__kmalloc(size_t size, gfp_t flags) | |
3713 | { | |
7c0cb9c6 | 3714 | return __do_kmalloc(size, flags, 0); |
1d2c8eea CH |
3715 | } |
3716 | EXPORT_SYMBOL(__kmalloc); | |
7fd6b141 PE |
3717 | #endif |
3718 | ||
1da177e4 LT |
3719 | /** |
3720 | * kmem_cache_free - Deallocate an object | |
3721 | * @cachep: The cache the allocation was from. | |
3722 | * @objp: The previously allocated object. | |
3723 | * | |
3724 | * Free an object which was previously allocated from this | |
3725 | * cache. | |
3726 | */ | |
343e0d7a | 3727 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3728 | { |
3729 | unsigned long flags; | |
b9ce5ef4 GC |
3730 | cachep = cache_from_obj(cachep, objp); |
3731 | if (!cachep) | |
3732 | return; | |
1da177e4 LT |
3733 | |
3734 | local_irq_save(flags); | |
d97d476b | 3735 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3736 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3737 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3738 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3739 | local_irq_restore(flags); |
36555751 | 3740 | |
ca2b84cb | 3741 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3742 | } |
3743 | EXPORT_SYMBOL(kmem_cache_free); | |
3744 | ||
1da177e4 LT |
3745 | /** |
3746 | * kfree - free previously allocated memory | |
3747 | * @objp: pointer returned by kmalloc. | |
3748 | * | |
80e93eff PE |
3749 | * If @objp is NULL, no operation is performed. |
3750 | * | |
1da177e4 LT |
3751 | * Don't free memory not originally allocated by kmalloc() |
3752 | * or you will run into trouble. | |
3753 | */ | |
3754 | void kfree(const void *objp) | |
3755 | { | |
343e0d7a | 3756 | struct kmem_cache *c; |
1da177e4 LT |
3757 | unsigned long flags; |
3758 | ||
2121db74 PE |
3759 | trace_kfree(_RET_IP_, objp); |
3760 | ||
6cb8f913 | 3761 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3762 | return; |
3763 | local_irq_save(flags); | |
3764 | kfree_debugcheck(objp); | |
6ed5eb22 | 3765 | c = virt_to_cache(objp); |
8c138bc0 CL |
3766 | debug_check_no_locks_freed(objp, c->object_size); |
3767 | ||
3768 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3769 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3770 | local_irq_restore(flags); |
3771 | } | |
3772 | EXPORT_SYMBOL(kfree); | |
3773 | ||
e498be7d | 3774 | /* |
ce8eb6c4 | 3775 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3776 | */ |
5f0985bb | 3777 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3778 | { |
3779 | int node; | |
ce8eb6c4 | 3780 | struct kmem_cache_node *n; |
cafeb02e | 3781 | struct array_cache *new_shared; |
c8522a3a | 3782 | struct alien_cache **new_alien = NULL; |
e498be7d | 3783 | |
9c09a95c | 3784 | for_each_online_node(node) { |
cafeb02e | 3785 | |
3395ee05 | 3786 | if (use_alien_caches) { |
83b519e8 | 3787 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3788 | if (!new_alien) |
3789 | goto fail; | |
3790 | } | |
cafeb02e | 3791 | |
63109846 ED |
3792 | new_shared = NULL; |
3793 | if (cachep->shared) { | |
3794 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3795 | cachep->shared*cachep->batchcount, |
83b519e8 | 3796 | 0xbaadf00d, gfp); |
63109846 ED |
3797 | if (!new_shared) { |
3798 | free_alien_cache(new_alien); | |
3799 | goto fail; | |
3800 | } | |
0718dc2a | 3801 | } |
cafeb02e | 3802 | |
18bf8541 | 3803 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3804 | if (n) { |
3805 | struct array_cache *shared = n->shared; | |
97654dfa | 3806 | LIST_HEAD(list); |
cafeb02e | 3807 | |
ce8eb6c4 | 3808 | spin_lock_irq(&n->list_lock); |
e498be7d | 3809 | |
cafeb02e | 3810 | if (shared) |
0718dc2a | 3811 | free_block(cachep, shared->entry, |
97654dfa | 3812 | shared->avail, node, &list); |
e498be7d | 3813 | |
ce8eb6c4 CL |
3814 | n->shared = new_shared; |
3815 | if (!n->alien) { | |
3816 | n->alien = new_alien; | |
e498be7d CL |
3817 | new_alien = NULL; |
3818 | } | |
ce8eb6c4 | 3819 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3820 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3821 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3822 | slabs_destroy(cachep, &list); |
cafeb02e | 3823 | kfree(shared); |
e498be7d CL |
3824 | free_alien_cache(new_alien); |
3825 | continue; | |
3826 | } | |
ce8eb6c4 CL |
3827 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3828 | if (!n) { | |
0718dc2a CL |
3829 | free_alien_cache(new_alien); |
3830 | kfree(new_shared); | |
e498be7d | 3831 | goto fail; |
0718dc2a | 3832 | } |
e498be7d | 3833 | |
ce8eb6c4 | 3834 | kmem_cache_node_init(n); |
5f0985bb JZ |
3835 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3836 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3837 | n->shared = new_shared; |
3838 | n->alien = new_alien; | |
3839 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3840 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3841 | cachep->node[node] = n; |
e498be7d | 3842 | } |
cafeb02e | 3843 | return 0; |
0718dc2a | 3844 | |
a737b3e2 | 3845 | fail: |
3b0efdfa | 3846 | if (!cachep->list.next) { |
0718dc2a CL |
3847 | /* Cache is not active yet. Roll back what we did */ |
3848 | node--; | |
3849 | while (node >= 0) { | |
18bf8541 CL |
3850 | n = get_node(cachep, node); |
3851 | if (n) { | |
ce8eb6c4 CL |
3852 | kfree(n->shared); |
3853 | free_alien_cache(n->alien); | |
3854 | kfree(n); | |
6a67368c | 3855 | cachep->node[node] = NULL; |
0718dc2a CL |
3856 | } |
3857 | node--; | |
3858 | } | |
3859 | } | |
cafeb02e | 3860 | return -ENOMEM; |
e498be7d CL |
3861 | } |
3862 | ||
1da177e4 | 3863 | struct ccupdate_struct { |
343e0d7a | 3864 | struct kmem_cache *cachep; |
acfe7d74 | 3865 | struct array_cache *new[0]; |
1da177e4 LT |
3866 | }; |
3867 | ||
3868 | static void do_ccupdate_local(void *info) | |
3869 | { | |
a737b3e2 | 3870 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3871 | struct array_cache *old; |
3872 | ||
3873 | check_irq_off(); | |
9a2dba4b | 3874 | old = cpu_cache_get(new->cachep); |
e498be7d | 3875 | |
1da177e4 LT |
3876 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3877 | new->new[smp_processor_id()] = old; | |
3878 | } | |
3879 | ||
18004c5d | 3880 | /* Always called with the slab_mutex held */ |
943a451a | 3881 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3882 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3883 | { |
d2e7b7d0 | 3884 | struct ccupdate_struct *new; |
2ed3a4ef | 3885 | int i; |
1da177e4 | 3886 | |
acfe7d74 ED |
3887 | new = kzalloc(sizeof(*new) + nr_cpu_ids * sizeof(struct array_cache *), |
3888 | gfp); | |
d2e7b7d0 SS |
3889 | if (!new) |
3890 | return -ENOMEM; | |
3891 | ||
e498be7d | 3892 | for_each_online_cpu(i) { |
7d6e6d09 | 3893 | new->new[i] = alloc_arraycache(cpu_to_mem(i), limit, |
83b519e8 | 3894 | batchcount, gfp); |
d2e7b7d0 | 3895 | if (!new->new[i]) { |
b28a02de | 3896 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3897 | kfree(new->new[i]); |
3898 | kfree(new); | |
e498be7d | 3899 | return -ENOMEM; |
1da177e4 LT |
3900 | } |
3901 | } | |
d2e7b7d0 | 3902 | new->cachep = cachep; |
1da177e4 | 3903 | |
15c8b6c1 | 3904 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 3905 | |
1da177e4 | 3906 | check_irq_on(); |
1da177e4 LT |
3907 | cachep->batchcount = batchcount; |
3908 | cachep->limit = limit; | |
e498be7d | 3909 | cachep->shared = shared; |
1da177e4 | 3910 | |
e498be7d | 3911 | for_each_online_cpu(i) { |
97654dfa | 3912 | LIST_HEAD(list); |
d2e7b7d0 | 3913 | struct array_cache *ccold = new->new[i]; |
18bf8541 CL |
3914 | int node; |
3915 | struct kmem_cache_node *n; | |
3916 | ||
1da177e4 LT |
3917 | if (!ccold) |
3918 | continue; | |
18bf8541 CL |
3919 | |
3920 | node = cpu_to_mem(i); | |
3921 | n = get_node(cachep, node); | |
3922 | spin_lock_irq(&n->list_lock); | |
97654dfa | 3923 | free_block(cachep, ccold->entry, ccold->avail, node, &list); |
18bf8541 | 3924 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3925 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3926 | kfree(ccold); |
3927 | } | |
d2e7b7d0 | 3928 | kfree(new); |
5f0985bb | 3929 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3930 | } |
3931 | ||
943a451a GC |
3932 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3933 | int batchcount, int shared, gfp_t gfp) | |
3934 | { | |
3935 | int ret; | |
3936 | struct kmem_cache *c = NULL; | |
3937 | int i = 0; | |
3938 | ||
3939 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3940 | ||
3941 | if (slab_state < FULL) | |
3942 | return ret; | |
3943 | ||
3944 | if ((ret < 0) || !is_root_cache(cachep)) | |
3945 | return ret; | |
3946 | ||
ebe945c2 | 3947 | VM_BUG_ON(!mutex_is_locked(&slab_mutex)); |
943a451a | 3948 | for_each_memcg_cache_index(i) { |
2ade4de8 | 3949 | c = cache_from_memcg_idx(cachep, i); |
943a451a GC |
3950 | if (c) |
3951 | /* return value determined by the parent cache only */ | |
3952 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
3953 | } | |
3954 | ||
3955 | return ret; | |
3956 | } | |
3957 | ||
18004c5d | 3958 | /* Called with slab_mutex held always */ |
83b519e8 | 3959 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3960 | { |
3961 | int err; | |
943a451a GC |
3962 | int limit = 0; |
3963 | int shared = 0; | |
3964 | int batchcount = 0; | |
3965 | ||
3966 | if (!is_root_cache(cachep)) { | |
3967 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3968 | limit = root->limit; | |
3969 | shared = root->shared; | |
3970 | batchcount = root->batchcount; | |
3971 | } | |
1da177e4 | 3972 | |
943a451a GC |
3973 | if (limit && shared && batchcount) |
3974 | goto skip_setup; | |
a737b3e2 AM |
3975 | /* |
3976 | * The head array serves three purposes: | |
1da177e4 LT |
3977 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3978 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3979 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3980 | * bufctl chains: array operations are cheaper. |
3981 | * The numbers are guessed, we should auto-tune as described by | |
3982 | * Bonwick. | |
3983 | */ | |
3b0efdfa | 3984 | if (cachep->size > 131072) |
1da177e4 | 3985 | limit = 1; |
3b0efdfa | 3986 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3987 | limit = 8; |
3b0efdfa | 3988 | else if (cachep->size > 1024) |
1da177e4 | 3989 | limit = 24; |
3b0efdfa | 3990 | else if (cachep->size > 256) |
1da177e4 LT |
3991 | limit = 54; |
3992 | else | |
3993 | limit = 120; | |
3994 | ||
a737b3e2 AM |
3995 | /* |
3996 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3997 | * allocation behaviour: Most allocs on one cpu, most free operations |
3998 | * on another cpu. For these cases, an efficient object passing between | |
3999 | * cpus is necessary. This is provided by a shared array. The array | |
4000 | * replaces Bonwick's magazine layer. | |
4001 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
4002 | * to a larger limit. Thus disabled by default. | |
4003 | */ | |
4004 | shared = 0; | |
3b0efdfa | 4005 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 4006 | shared = 8; |
1da177e4 LT |
4007 | |
4008 | #if DEBUG | |
a737b3e2 AM |
4009 | /* |
4010 | * With debugging enabled, large batchcount lead to excessively long | |
4011 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
4012 | */ |
4013 | if (limit > 32) | |
4014 | limit = 32; | |
4015 | #endif | |
943a451a GC |
4016 | batchcount = (limit + 1) / 2; |
4017 | skip_setup: | |
4018 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
4019 | if (err) |
4020 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 4021 | cachep->name, -err); |
2ed3a4ef | 4022 | return err; |
1da177e4 LT |
4023 | } |
4024 | ||
1b55253a | 4025 | /* |
ce8eb6c4 CL |
4026 | * Drain an array if it contains any elements taking the node lock only if |
4027 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 4028 | * if drain_array() is used on the shared array. |
1b55253a | 4029 | */ |
ce8eb6c4 | 4030 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 4031 | struct array_cache *ac, int force, int node) |
1da177e4 | 4032 | { |
97654dfa | 4033 | LIST_HEAD(list); |
1da177e4 LT |
4034 | int tofree; |
4035 | ||
1b55253a CL |
4036 | if (!ac || !ac->avail) |
4037 | return; | |
1da177e4 LT |
4038 | if (ac->touched && !force) { |
4039 | ac->touched = 0; | |
b18e7e65 | 4040 | } else { |
ce8eb6c4 | 4041 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
4042 | if (ac->avail) { |
4043 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
4044 | if (tofree > ac->avail) | |
4045 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 4046 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
4047 | ac->avail -= tofree; |
4048 | memmove(ac->entry, &(ac->entry[tofree]), | |
4049 | sizeof(void *) * ac->avail); | |
4050 | } | |
ce8eb6c4 | 4051 | spin_unlock_irq(&n->list_lock); |
97654dfa | 4052 | slabs_destroy(cachep, &list); |
1da177e4 LT |
4053 | } |
4054 | } | |
4055 | ||
4056 | /** | |
4057 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4058 | * @w: work descriptor |
1da177e4 LT |
4059 | * |
4060 | * Called from workqueue/eventd every few seconds. | |
4061 | * Purpose: | |
4062 | * - clear the per-cpu caches for this CPU. | |
4063 | * - return freeable pages to the main free memory pool. | |
4064 | * | |
a737b3e2 AM |
4065 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4066 | * again on the next iteration. | |
1da177e4 | 4067 | */ |
7c5cae36 | 4068 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4069 | { |
7a7c381d | 4070 | struct kmem_cache *searchp; |
ce8eb6c4 | 4071 | struct kmem_cache_node *n; |
7d6e6d09 | 4072 | int node = numa_mem_id(); |
bf6aede7 | 4073 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4074 | |
18004c5d | 4075 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 4076 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4077 | goto out; |
1da177e4 | 4078 | |
18004c5d | 4079 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
4080 | check_irq_on(); |
4081 | ||
35386e3b | 4082 | /* |
ce8eb6c4 | 4083 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4084 | * have established with reasonable certainty that |
4085 | * we can do some work if the lock was obtained. | |
4086 | */ | |
18bf8541 | 4087 | n = get_node(searchp, node); |
35386e3b | 4088 | |
ce8eb6c4 | 4089 | reap_alien(searchp, n); |
1da177e4 | 4090 | |
ce8eb6c4 | 4091 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 4092 | |
35386e3b CL |
4093 | /* |
4094 | * These are racy checks but it does not matter | |
4095 | * if we skip one check or scan twice. | |
4096 | */ | |
ce8eb6c4 | 4097 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4098 | goto next; |
1da177e4 | 4099 | |
5f0985bb | 4100 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 4101 | |
ce8eb6c4 | 4102 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 4103 | |
ce8eb6c4 CL |
4104 | if (n->free_touched) |
4105 | n->free_touched = 0; | |
ed11d9eb CL |
4106 | else { |
4107 | int freed; | |
1da177e4 | 4108 | |
ce8eb6c4 | 4109 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4110 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4111 | STATS_ADD_REAPED(searchp, freed); | |
4112 | } | |
35386e3b | 4113 | next: |
1da177e4 LT |
4114 | cond_resched(); |
4115 | } | |
4116 | check_irq_on(); | |
18004c5d | 4117 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4118 | next_reap_node(); |
7c5cae36 | 4119 | out: |
a737b3e2 | 4120 | /* Set up the next iteration */ |
5f0985bb | 4121 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
4122 | } |
4123 | ||
158a9624 | 4124 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 4125 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4126 | { |
8456a648 | 4127 | struct page *page; |
b28a02de PE |
4128 | unsigned long active_objs; |
4129 | unsigned long num_objs; | |
4130 | unsigned long active_slabs = 0; | |
4131 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 4132 | const char *name; |
1da177e4 | 4133 | char *error = NULL; |
e498be7d | 4134 | int node; |
ce8eb6c4 | 4135 | struct kmem_cache_node *n; |
1da177e4 | 4136 | |
1da177e4 LT |
4137 | active_objs = 0; |
4138 | num_slabs = 0; | |
18bf8541 | 4139 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 4140 | |
ca3b9b91 | 4141 | check_irq_on(); |
ce8eb6c4 | 4142 | spin_lock_irq(&n->list_lock); |
e498be7d | 4143 | |
8456a648 JK |
4144 | list_for_each_entry(page, &n->slabs_full, lru) { |
4145 | if (page->active != cachep->num && !error) | |
e498be7d CL |
4146 | error = "slabs_full accounting error"; |
4147 | active_objs += cachep->num; | |
4148 | active_slabs++; | |
4149 | } | |
8456a648 JK |
4150 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4151 | if (page->active == cachep->num && !error) | |
106a74e1 | 4152 | error = "slabs_partial accounting error"; |
8456a648 | 4153 | if (!page->active && !error) |
106a74e1 | 4154 | error = "slabs_partial accounting error"; |
8456a648 | 4155 | active_objs += page->active; |
e498be7d CL |
4156 | active_slabs++; |
4157 | } | |
8456a648 JK |
4158 | list_for_each_entry(page, &n->slabs_free, lru) { |
4159 | if (page->active && !error) | |
106a74e1 | 4160 | error = "slabs_free accounting error"; |
e498be7d CL |
4161 | num_slabs++; |
4162 | } | |
ce8eb6c4 CL |
4163 | free_objects += n->free_objects; |
4164 | if (n->shared) | |
4165 | shared_avail += n->shared->avail; | |
e498be7d | 4166 | |
ce8eb6c4 | 4167 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4168 | } |
b28a02de PE |
4169 | num_slabs += active_slabs; |
4170 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4171 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4172 | error = "free_objects accounting error"; |
4173 | ||
b28a02de | 4174 | name = cachep->name; |
1da177e4 LT |
4175 | if (error) |
4176 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4177 | ||
0d7561c6 GC |
4178 | sinfo->active_objs = active_objs; |
4179 | sinfo->num_objs = num_objs; | |
4180 | sinfo->active_slabs = active_slabs; | |
4181 | sinfo->num_slabs = num_slabs; | |
4182 | sinfo->shared_avail = shared_avail; | |
4183 | sinfo->limit = cachep->limit; | |
4184 | sinfo->batchcount = cachep->batchcount; | |
4185 | sinfo->shared = cachep->shared; | |
4186 | sinfo->objects_per_slab = cachep->num; | |
4187 | sinfo->cache_order = cachep->gfporder; | |
4188 | } | |
4189 | ||
4190 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4191 | { | |
1da177e4 | 4192 | #if STATS |
ce8eb6c4 | 4193 | { /* node stats */ |
1da177e4 LT |
4194 | unsigned long high = cachep->high_mark; |
4195 | unsigned long allocs = cachep->num_allocations; | |
4196 | unsigned long grown = cachep->grown; | |
4197 | unsigned long reaped = cachep->reaped; | |
4198 | unsigned long errors = cachep->errors; | |
4199 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4200 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4201 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4202 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4203 | |
e92dd4fd JP |
4204 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4205 | "%4lu %4lu %4lu %4lu %4lu", | |
4206 | allocs, high, grown, | |
4207 | reaped, errors, max_freeable, node_allocs, | |
4208 | node_frees, overflows); | |
1da177e4 LT |
4209 | } |
4210 | /* cpu stats */ | |
4211 | { | |
4212 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4213 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4214 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4215 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4216 | ||
4217 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4218 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4219 | } |
4220 | #endif | |
1da177e4 LT |
4221 | } |
4222 | ||
1da177e4 LT |
4223 | #define MAX_SLABINFO_WRITE 128 |
4224 | /** | |
4225 | * slabinfo_write - Tuning for the slab allocator | |
4226 | * @file: unused | |
4227 | * @buffer: user buffer | |
4228 | * @count: data length | |
4229 | * @ppos: unused | |
4230 | */ | |
b7454ad3 | 4231 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4232 | size_t count, loff_t *ppos) |
1da177e4 | 4233 | { |
b28a02de | 4234 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4235 | int limit, batchcount, shared, res; |
7a7c381d | 4236 | struct kmem_cache *cachep; |
b28a02de | 4237 | |
1da177e4 LT |
4238 | if (count > MAX_SLABINFO_WRITE) |
4239 | return -EINVAL; | |
4240 | if (copy_from_user(&kbuf, buffer, count)) | |
4241 | return -EFAULT; | |
b28a02de | 4242 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4243 | |
4244 | tmp = strchr(kbuf, ' '); | |
4245 | if (!tmp) | |
4246 | return -EINVAL; | |
4247 | *tmp = '\0'; | |
4248 | tmp++; | |
4249 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4250 | return -EINVAL; | |
4251 | ||
4252 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4253 | mutex_lock(&slab_mutex); |
1da177e4 | 4254 | res = -EINVAL; |
18004c5d | 4255 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4256 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4257 | if (limit < 1 || batchcount < 1 || |
4258 | batchcount > limit || shared < 0) { | |
e498be7d | 4259 | res = 0; |
1da177e4 | 4260 | } else { |
e498be7d | 4261 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4262 | batchcount, shared, |
4263 | GFP_KERNEL); | |
1da177e4 LT |
4264 | } |
4265 | break; | |
4266 | } | |
4267 | } | |
18004c5d | 4268 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4269 | if (res >= 0) |
4270 | res = count; | |
4271 | return res; | |
4272 | } | |
871751e2 AV |
4273 | |
4274 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4275 | ||
4276 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4277 | { | |
18004c5d CL |
4278 | mutex_lock(&slab_mutex); |
4279 | return seq_list_start(&slab_caches, *pos); | |
871751e2 AV |
4280 | } |
4281 | ||
4282 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4283 | { | |
4284 | unsigned long *p; | |
4285 | int l; | |
4286 | if (!v) | |
4287 | return 1; | |
4288 | l = n[1]; | |
4289 | p = n + 2; | |
4290 | while (l) { | |
4291 | int i = l/2; | |
4292 | unsigned long *q = p + 2 * i; | |
4293 | if (*q == v) { | |
4294 | q[1]++; | |
4295 | return 1; | |
4296 | } | |
4297 | if (*q > v) { | |
4298 | l = i; | |
4299 | } else { | |
4300 | p = q + 2; | |
4301 | l -= i + 1; | |
4302 | } | |
4303 | } | |
4304 | if (++n[1] == n[0]) | |
4305 | return 0; | |
4306 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4307 | p[0] = v; | |
4308 | p[1] = 1; | |
4309 | return 1; | |
4310 | } | |
4311 | ||
8456a648 JK |
4312 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4313 | struct page *page) | |
871751e2 AV |
4314 | { |
4315 | void *p; | |
03787301 | 4316 | int i; |
b1cb0982 | 4317 | |
871751e2 AV |
4318 | if (n[0] == n[1]) |
4319 | return; | |
8456a648 | 4320 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
03787301 | 4321 | if (get_obj_status(page, i) != OBJECT_ACTIVE) |
871751e2 | 4322 | continue; |
b1cb0982 | 4323 | |
871751e2 AV |
4324 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
4325 | return; | |
4326 | } | |
4327 | } | |
4328 | ||
4329 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4330 | { | |
4331 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4332 | unsigned long offset, size; |
9281acea | 4333 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4334 | |
a5c43dae | 4335 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4336 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4337 | if (modname[0]) |
871751e2 AV |
4338 | seq_printf(m, " [%s]", modname); |
4339 | return; | |
4340 | } | |
4341 | #endif | |
4342 | seq_printf(m, "%p", (void *)address); | |
4343 | } | |
4344 | ||
4345 | static int leaks_show(struct seq_file *m, void *p) | |
4346 | { | |
0672aa7c | 4347 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4348 | struct page *page; |
ce8eb6c4 | 4349 | struct kmem_cache_node *n; |
871751e2 | 4350 | const char *name; |
db845067 | 4351 | unsigned long *x = m->private; |
871751e2 AV |
4352 | int node; |
4353 | int i; | |
4354 | ||
4355 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4356 | return 0; | |
4357 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4358 | return 0; | |
4359 | ||
4360 | /* OK, we can do it */ | |
4361 | ||
db845067 | 4362 | x[1] = 0; |
871751e2 | 4363 | |
18bf8541 | 4364 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 AV |
4365 | |
4366 | check_irq_on(); | |
ce8eb6c4 | 4367 | spin_lock_irq(&n->list_lock); |
871751e2 | 4368 | |
8456a648 JK |
4369 | list_for_each_entry(page, &n->slabs_full, lru) |
4370 | handle_slab(x, cachep, page); | |
4371 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4372 | handle_slab(x, cachep, page); | |
ce8eb6c4 | 4373 | spin_unlock_irq(&n->list_lock); |
871751e2 AV |
4374 | } |
4375 | name = cachep->name; | |
db845067 | 4376 | if (x[0] == x[1]) { |
871751e2 | 4377 | /* Increase the buffer size */ |
18004c5d | 4378 | mutex_unlock(&slab_mutex); |
db845067 | 4379 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4380 | if (!m->private) { |
4381 | /* Too bad, we are really out */ | |
db845067 | 4382 | m->private = x; |
18004c5d | 4383 | mutex_lock(&slab_mutex); |
871751e2 AV |
4384 | return -ENOMEM; |
4385 | } | |
db845067 CL |
4386 | *(unsigned long *)m->private = x[0] * 2; |
4387 | kfree(x); | |
18004c5d | 4388 | mutex_lock(&slab_mutex); |
871751e2 AV |
4389 | /* Now make sure this entry will be retried */ |
4390 | m->count = m->size; | |
4391 | return 0; | |
4392 | } | |
db845067 CL |
4393 | for (i = 0; i < x[1]; i++) { |
4394 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4395 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4396 | seq_putc(m, '\n'); |
4397 | } | |
d2e7b7d0 | 4398 | |
871751e2 AV |
4399 | return 0; |
4400 | } | |
4401 | ||
a0ec95a8 | 4402 | static const struct seq_operations slabstats_op = { |
871751e2 | 4403 | .start = leaks_start, |
276a2439 WL |
4404 | .next = slab_next, |
4405 | .stop = slab_stop, | |
871751e2 AV |
4406 | .show = leaks_show, |
4407 | }; | |
a0ec95a8 AD |
4408 | |
4409 | static int slabstats_open(struct inode *inode, struct file *file) | |
4410 | { | |
4411 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4412 | int ret = -ENOMEM; | |
4413 | if (n) { | |
4414 | ret = seq_open(file, &slabstats_op); | |
4415 | if (!ret) { | |
4416 | struct seq_file *m = file->private_data; | |
4417 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4418 | m->private = n; | |
4419 | n = NULL; | |
4420 | } | |
4421 | kfree(n); | |
4422 | } | |
4423 | return ret; | |
4424 | } | |
4425 | ||
4426 | static const struct file_operations proc_slabstats_operations = { | |
4427 | .open = slabstats_open, | |
4428 | .read = seq_read, | |
4429 | .llseek = seq_lseek, | |
4430 | .release = seq_release_private, | |
4431 | }; | |
4432 | #endif | |
4433 | ||
4434 | static int __init slab_proc_init(void) | |
4435 | { | |
4436 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4437 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4438 | #endif |
a0ec95a8 AD |
4439 | return 0; |
4440 | } | |
4441 | module_init(slab_proc_init); | |
1da177e4 LT |
4442 | #endif |
4443 | ||
00e145b6 MS |
4444 | /** |
4445 | * ksize - get the actual amount of memory allocated for a given object | |
4446 | * @objp: Pointer to the object | |
4447 | * | |
4448 | * kmalloc may internally round up allocations and return more memory | |
4449 | * than requested. ksize() can be used to determine the actual amount of | |
4450 | * memory allocated. The caller may use this additional memory, even though | |
4451 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4452 | * The caller must guarantee that objp points to a valid object previously | |
4453 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4454 | * must not be freed during the duration of the call. | |
4455 | */ | |
fd76bab2 | 4456 | size_t ksize(const void *objp) |
1da177e4 | 4457 | { |
ef8b4520 CL |
4458 | BUG_ON(!objp); |
4459 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4460 | return 0; |
1da177e4 | 4461 | |
8c138bc0 | 4462 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4463 | } |
b1aabecd | 4464 | EXPORT_SYMBOL(ksize); |