]>
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 | ||
edcad250 JK |
473 | #define BAD_ALIEN_MAGIC 0x01020304ul |
474 | ||
1871e52c | 475 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 476 | |
343e0d7a | 477 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
478 | { |
479 | return cachep->array[smp_processor_id()]; | |
480 | } | |
481 | ||
03787301 JK |
482 | static size_t calculate_freelist_size(int nr_objs, size_t align) |
483 | { | |
484 | size_t freelist_size; | |
485 | ||
486 | freelist_size = nr_objs * sizeof(freelist_idx_t); | |
487 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) | |
488 | freelist_size += nr_objs * sizeof(char); | |
489 | ||
490 | if (align) | |
491 | freelist_size = ALIGN(freelist_size, align); | |
492 | ||
493 | return freelist_size; | |
494 | } | |
495 | ||
9cef2e2b JK |
496 | static int calculate_nr_objs(size_t slab_size, size_t buffer_size, |
497 | size_t idx_size, size_t align) | |
1da177e4 | 498 | { |
9cef2e2b | 499 | int nr_objs; |
03787301 | 500 | size_t remained_size; |
9cef2e2b | 501 | size_t freelist_size; |
03787301 | 502 | int extra_space = 0; |
9cef2e2b | 503 | |
03787301 JK |
504 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
505 | extra_space = sizeof(char); | |
9cef2e2b JK |
506 | /* |
507 | * Ignore padding for the initial guess. The padding | |
508 | * is at most @align-1 bytes, and @buffer_size is at | |
509 | * least @align. In the worst case, this result will | |
510 | * be one greater than the number of objects that fit | |
511 | * into the memory allocation when taking the padding | |
512 | * into account. | |
513 | */ | |
03787301 | 514 | nr_objs = slab_size / (buffer_size + idx_size + extra_space); |
9cef2e2b JK |
515 | |
516 | /* | |
517 | * This calculated number will be either the right | |
518 | * amount, or one greater than what we want. | |
519 | */ | |
03787301 JK |
520 | remained_size = slab_size - nr_objs * buffer_size; |
521 | freelist_size = calculate_freelist_size(nr_objs, align); | |
522 | if (remained_size < freelist_size) | |
9cef2e2b JK |
523 | nr_objs--; |
524 | ||
525 | return nr_objs; | |
fbaccacf | 526 | } |
1da177e4 | 527 | |
a737b3e2 AM |
528 | /* |
529 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
530 | */ | |
fbaccacf SR |
531 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
532 | size_t align, int flags, size_t *left_over, | |
533 | unsigned int *num) | |
534 | { | |
535 | int nr_objs; | |
536 | size_t mgmt_size; | |
537 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 538 | |
fbaccacf SR |
539 | /* |
540 | * The slab management structure can be either off the slab or | |
541 | * on it. For the latter case, the memory allocated for a | |
542 | * slab is used for: | |
543 | * | |
16025177 | 544 | * - One unsigned int for each object |
fbaccacf SR |
545 | * - Padding to respect alignment of @align |
546 | * - @buffer_size bytes for each object | |
547 | * | |
548 | * If the slab management structure is off the slab, then the | |
549 | * alignment will already be calculated into the size. Because | |
550 | * the slabs are all pages aligned, the objects will be at the | |
551 | * correct alignment when allocated. | |
552 | */ | |
553 | if (flags & CFLGS_OFF_SLAB) { | |
554 | mgmt_size = 0; | |
555 | nr_objs = slab_size / buffer_size; | |
556 | ||
fbaccacf | 557 | } else { |
9cef2e2b | 558 | nr_objs = calculate_nr_objs(slab_size, buffer_size, |
a41adfaa | 559 | sizeof(freelist_idx_t), align); |
03787301 | 560 | mgmt_size = calculate_freelist_size(nr_objs, align); |
fbaccacf SR |
561 | } |
562 | *num = nr_objs; | |
563 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
564 | } |
565 | ||
f28510d3 | 566 | #if DEBUG |
d40cee24 | 567 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 568 | |
a737b3e2 AM |
569 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
570 | char *msg) | |
1da177e4 LT |
571 | { |
572 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 573 | function, cachep->name, msg); |
1da177e4 | 574 | dump_stack(); |
373d4d09 | 575 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 576 | } |
f28510d3 | 577 | #endif |
1da177e4 | 578 | |
3395ee05 PM |
579 | /* |
580 | * By default on NUMA we use alien caches to stage the freeing of | |
581 | * objects allocated from other nodes. This causes massive memory | |
582 | * inefficiencies when using fake NUMA setup to split memory into a | |
583 | * large number of small nodes, so it can be disabled on the command | |
584 | * line | |
585 | */ | |
586 | ||
587 | static int use_alien_caches __read_mostly = 1; | |
588 | static int __init noaliencache_setup(char *s) | |
589 | { | |
590 | use_alien_caches = 0; | |
591 | return 1; | |
592 | } | |
593 | __setup("noaliencache", noaliencache_setup); | |
594 | ||
3df1cccd DR |
595 | static int __init slab_max_order_setup(char *str) |
596 | { | |
597 | get_option(&str, &slab_max_order); | |
598 | slab_max_order = slab_max_order < 0 ? 0 : | |
599 | min(slab_max_order, MAX_ORDER - 1); | |
600 | slab_max_order_set = true; | |
601 | ||
602 | return 1; | |
603 | } | |
604 | __setup("slab_max_order=", slab_max_order_setup); | |
605 | ||
8fce4d8e CL |
606 | #ifdef CONFIG_NUMA |
607 | /* | |
608 | * Special reaping functions for NUMA systems called from cache_reap(). | |
609 | * These take care of doing round robin flushing of alien caches (containing | |
610 | * objects freed on different nodes from which they were allocated) and the | |
611 | * flushing of remote pcps by calling drain_node_pages. | |
612 | */ | |
1871e52c | 613 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
614 | |
615 | static void init_reap_node(int cpu) | |
616 | { | |
617 | int node; | |
618 | ||
7d6e6d09 | 619 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 620 | if (node == MAX_NUMNODES) |
442295c9 | 621 | node = first_node(node_online_map); |
8fce4d8e | 622 | |
1871e52c | 623 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
624 | } |
625 | ||
626 | static void next_reap_node(void) | |
627 | { | |
909ea964 | 628 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 629 | |
8fce4d8e CL |
630 | node = next_node(node, node_online_map); |
631 | if (unlikely(node >= MAX_NUMNODES)) | |
632 | node = first_node(node_online_map); | |
909ea964 | 633 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
634 | } |
635 | ||
636 | #else | |
637 | #define init_reap_node(cpu) do { } while (0) | |
638 | #define next_reap_node(void) do { } while (0) | |
639 | #endif | |
640 | ||
1da177e4 LT |
641 | /* |
642 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
643 | * via the workqueue/eventd. | |
644 | * Add the CPU number into the expiration time to minimize the possibility of | |
645 | * the CPUs getting into lockstep and contending for the global cache chain | |
646 | * lock. | |
647 | */ | |
0db0628d | 648 | static void start_cpu_timer(int cpu) |
1da177e4 | 649 | { |
1871e52c | 650 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
651 | |
652 | /* | |
653 | * When this gets called from do_initcalls via cpucache_init(), | |
654 | * init_workqueues() has already run, so keventd will be setup | |
655 | * at that time. | |
656 | */ | |
52bad64d | 657 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 658 | init_reap_node(cpu); |
203b42f7 | 659 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
660 | schedule_delayed_work_on(cpu, reap_work, |
661 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
662 | } |
663 | } | |
664 | ||
1fe00d50 | 665 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 666 | { |
d5cff635 CM |
667 | /* |
668 | * The array_cache structures contain pointers to free object. | |
25985edc | 669 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
670 | * cache the pointers are not cleared and they could be counted as |
671 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
672 | * not scan such objects. | |
673 | */ | |
1fe00d50 JK |
674 | kmemleak_no_scan(ac); |
675 | if (ac) { | |
676 | ac->avail = 0; | |
677 | ac->limit = limit; | |
678 | ac->batchcount = batch; | |
679 | ac->touched = 0; | |
1da177e4 | 680 | } |
1fe00d50 JK |
681 | } |
682 | ||
683 | static struct array_cache *alloc_arraycache(int node, int entries, | |
684 | int batchcount, gfp_t gfp) | |
685 | { | |
5e804789 | 686 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
687 | struct array_cache *ac = NULL; |
688 | ||
689 | ac = kmalloc_node(memsize, gfp, node); | |
690 | init_arraycache(ac, entries, batchcount); | |
691 | return ac; | |
1da177e4 LT |
692 | } |
693 | ||
8456a648 | 694 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 695 | { |
072bb0aa MG |
696 | return PageSlabPfmemalloc(page); |
697 | } | |
698 | ||
699 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
700 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
701 | struct array_cache *ac) | |
702 | { | |
18bf8541 | 703 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 704 | struct page *page; |
072bb0aa MG |
705 | unsigned long flags; |
706 | ||
707 | if (!pfmemalloc_active) | |
708 | return; | |
709 | ||
ce8eb6c4 | 710 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
711 | list_for_each_entry(page, &n->slabs_full, lru) |
712 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
713 | goto out; |
714 | ||
8456a648 JK |
715 | list_for_each_entry(page, &n->slabs_partial, lru) |
716 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
717 | goto out; |
718 | ||
8456a648 JK |
719 | list_for_each_entry(page, &n->slabs_free, lru) |
720 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
721 | goto out; |
722 | ||
723 | pfmemalloc_active = false; | |
724 | out: | |
ce8eb6c4 | 725 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
726 | } |
727 | ||
381760ea | 728 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
729 | gfp_t flags, bool force_refill) |
730 | { | |
731 | int i; | |
732 | void *objp = ac->entry[--ac->avail]; | |
733 | ||
734 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
735 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 736 | struct kmem_cache_node *n; |
072bb0aa MG |
737 | |
738 | if (gfp_pfmemalloc_allowed(flags)) { | |
739 | clear_obj_pfmemalloc(&objp); | |
740 | return objp; | |
741 | } | |
742 | ||
743 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 744 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
745 | /* If a !PFMEMALLOC object is found, swap them */ |
746 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
747 | objp = ac->entry[i]; | |
748 | ac->entry[i] = ac->entry[ac->avail]; | |
749 | ac->entry[ac->avail] = objp; | |
750 | return objp; | |
751 | } | |
752 | } | |
753 | ||
754 | /* | |
755 | * If there are empty slabs on the slabs_free list and we are | |
756 | * being forced to refill the cache, mark this one !pfmemalloc. | |
757 | */ | |
18bf8541 | 758 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 759 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 760 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 761 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
762 | clear_obj_pfmemalloc(&objp); |
763 | recheck_pfmemalloc_active(cachep, ac); | |
764 | return objp; | |
765 | } | |
766 | ||
767 | /* No !PFMEMALLOC objects available */ | |
768 | ac->avail++; | |
769 | objp = NULL; | |
770 | } | |
771 | ||
772 | return objp; | |
773 | } | |
774 | ||
381760ea MG |
775 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
776 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
777 | { | |
778 | void *objp; | |
779 | ||
780 | if (unlikely(sk_memalloc_socks())) | |
781 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
782 | else | |
783 | objp = ac->entry[--ac->avail]; | |
784 | ||
785 | return objp; | |
786 | } | |
787 | ||
788 | static void *__ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
072bb0aa MG |
789 | void *objp) |
790 | { | |
791 | if (unlikely(pfmemalloc_active)) { | |
792 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 793 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
794 | if (PageSlabPfmemalloc(page)) |
795 | set_obj_pfmemalloc(&objp); | |
796 | } | |
797 | ||
381760ea MG |
798 | return objp; |
799 | } | |
800 | ||
801 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
802 | void *objp) | |
803 | { | |
804 | if (unlikely(sk_memalloc_socks())) | |
805 | objp = __ac_put_obj(cachep, ac, objp); | |
806 | ||
072bb0aa MG |
807 | ac->entry[ac->avail++] = objp; |
808 | } | |
809 | ||
3ded175a CL |
810 | /* |
811 | * Transfer objects in one arraycache to another. | |
812 | * Locking must be handled by the caller. | |
813 | * | |
814 | * Return the number of entries transferred. | |
815 | */ | |
816 | static int transfer_objects(struct array_cache *to, | |
817 | struct array_cache *from, unsigned int max) | |
818 | { | |
819 | /* Figure out how many entries to transfer */ | |
732eacc0 | 820 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
821 | |
822 | if (!nr) | |
823 | return 0; | |
824 | ||
825 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
826 | sizeof(void *) *nr); | |
827 | ||
828 | from->avail -= nr; | |
829 | to->avail += nr; | |
3ded175a CL |
830 | return nr; |
831 | } | |
832 | ||
765c4507 CL |
833 | #ifndef CONFIG_NUMA |
834 | ||
835 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 836 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 837 | |
c8522a3a JK |
838 | static inline struct alien_cache **alloc_alien_cache(int node, |
839 | int limit, gfp_t gfp) | |
765c4507 | 840 | { |
edcad250 | 841 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
842 | } |
843 | ||
c8522a3a | 844 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
845 | { |
846 | } | |
847 | ||
848 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
849 | { | |
850 | return 0; | |
851 | } | |
852 | ||
853 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
854 | gfp_t flags) | |
855 | { | |
856 | return NULL; | |
857 | } | |
858 | ||
8b98c169 | 859 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
860 | gfp_t flags, int nodeid) |
861 | { | |
862 | return NULL; | |
863 | } | |
864 | ||
865 | #else /* CONFIG_NUMA */ | |
866 | ||
8b98c169 | 867 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 868 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 869 | |
c8522a3a JK |
870 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
871 | int batch, gfp_t gfp) | |
872 | { | |
5e804789 | 873 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
874 | struct alien_cache *alc = NULL; |
875 | ||
876 | alc = kmalloc_node(memsize, gfp, node); | |
877 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 878 | spin_lock_init(&alc->lock); |
c8522a3a JK |
879 | return alc; |
880 | } | |
881 | ||
882 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 883 | { |
c8522a3a | 884 | struct alien_cache **alc_ptr; |
5e804789 | 885 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
886 | int i; |
887 | ||
888 | if (limit > 1) | |
889 | limit = 12; | |
c8522a3a JK |
890 | alc_ptr = kzalloc_node(memsize, gfp, node); |
891 | if (!alc_ptr) | |
892 | return NULL; | |
893 | ||
894 | for_each_node(i) { | |
895 | if (i == node || !node_online(i)) | |
896 | continue; | |
897 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
898 | if (!alc_ptr[i]) { | |
899 | for (i--; i >= 0; i--) | |
900 | kfree(alc_ptr[i]); | |
901 | kfree(alc_ptr); | |
902 | return NULL; | |
e498be7d CL |
903 | } |
904 | } | |
c8522a3a | 905 | return alc_ptr; |
e498be7d CL |
906 | } |
907 | ||
c8522a3a | 908 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
909 | { |
910 | int i; | |
911 | ||
c8522a3a | 912 | if (!alc_ptr) |
e498be7d | 913 | return; |
e498be7d | 914 | for_each_node(i) |
c8522a3a JK |
915 | kfree(alc_ptr[i]); |
916 | kfree(alc_ptr); | |
e498be7d CL |
917 | } |
918 | ||
343e0d7a | 919 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
920 | struct array_cache *ac, int node, |
921 | struct list_head *list) | |
e498be7d | 922 | { |
18bf8541 | 923 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
924 | |
925 | if (ac->avail) { | |
ce8eb6c4 | 926 | spin_lock(&n->list_lock); |
e00946fe CL |
927 | /* |
928 | * Stuff objects into the remote nodes shared array first. | |
929 | * That way we could avoid the overhead of putting the objects | |
930 | * into the free lists and getting them back later. | |
931 | */ | |
ce8eb6c4 CL |
932 | if (n->shared) |
933 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 934 | |
833b706c | 935 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 936 | ac->avail = 0; |
ce8eb6c4 | 937 | spin_unlock(&n->list_lock); |
e498be7d CL |
938 | } |
939 | } | |
940 | ||
8fce4d8e CL |
941 | /* |
942 | * Called from cache_reap() to regularly drain alien caches round robin. | |
943 | */ | |
ce8eb6c4 | 944 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 945 | { |
909ea964 | 946 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 947 | |
ce8eb6c4 | 948 | if (n->alien) { |
c8522a3a JK |
949 | struct alien_cache *alc = n->alien[node]; |
950 | struct array_cache *ac; | |
951 | ||
952 | if (alc) { | |
953 | ac = &alc->ac; | |
49dfc304 | 954 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
955 | LIST_HEAD(list); |
956 | ||
957 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 958 | spin_unlock_irq(&alc->lock); |
833b706c | 959 | slabs_destroy(cachep, &list); |
c8522a3a | 960 | } |
8fce4d8e CL |
961 | } |
962 | } | |
963 | } | |
964 | ||
a737b3e2 | 965 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 966 | struct alien_cache **alien) |
e498be7d | 967 | { |
b28a02de | 968 | int i = 0; |
c8522a3a | 969 | struct alien_cache *alc; |
e498be7d CL |
970 | struct array_cache *ac; |
971 | unsigned long flags; | |
972 | ||
973 | for_each_online_node(i) { | |
c8522a3a JK |
974 | alc = alien[i]; |
975 | if (alc) { | |
833b706c JK |
976 | LIST_HEAD(list); |
977 | ||
c8522a3a | 978 | ac = &alc->ac; |
49dfc304 | 979 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 980 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 981 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 982 | slabs_destroy(cachep, &list); |
e498be7d CL |
983 | } |
984 | } | |
985 | } | |
729bd0b7 | 986 | |
873623df | 987 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 | 988 | { |
1ea991b0 | 989 | int nodeid = page_to_nid(virt_to_page(objp)); |
ce8eb6c4 | 990 | struct kmem_cache_node *n; |
c8522a3a JK |
991 | struct alien_cache *alien = NULL; |
992 | struct array_cache *ac; | |
1ca4cb24 | 993 | int node; |
97654dfa | 994 | LIST_HEAD(list); |
1ca4cb24 | 995 | |
7d6e6d09 | 996 | node = numa_mem_id(); |
729bd0b7 PE |
997 | |
998 | /* | |
999 | * Make sure we are not freeing a object from another node to the array | |
1000 | * cache on this cpu. | |
1001 | */ | |
1ea991b0 | 1002 | if (likely(nodeid == node)) |
729bd0b7 PE |
1003 | return 0; |
1004 | ||
18bf8541 | 1005 | n = get_node(cachep, node); |
729bd0b7 | 1006 | STATS_INC_NODEFREES(cachep); |
ce8eb6c4 CL |
1007 | if (n->alien && n->alien[nodeid]) { |
1008 | alien = n->alien[nodeid]; | |
c8522a3a | 1009 | ac = &alien->ac; |
49dfc304 | 1010 | spin_lock(&alien->lock); |
c8522a3a | 1011 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 1012 | STATS_INC_ACOVERFLOW(cachep); |
833b706c | 1013 | __drain_alien_cache(cachep, ac, nodeid, &list); |
729bd0b7 | 1014 | } |
c8522a3a | 1015 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 1016 | spin_unlock(&alien->lock); |
833b706c | 1017 | slabs_destroy(cachep, &list); |
729bd0b7 | 1018 | } else { |
18bf8541 CL |
1019 | n = get_node(cachep, nodeid); |
1020 | spin_lock(&n->list_lock); | |
97654dfa | 1021 | free_block(cachep, &objp, 1, nodeid, &list); |
18bf8541 | 1022 | spin_unlock(&n->list_lock); |
97654dfa | 1023 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
1024 | } |
1025 | return 1; | |
1026 | } | |
e498be7d CL |
1027 | #endif |
1028 | ||
8f9f8d9e | 1029 | /* |
6a67368c | 1030 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1031 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1032 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1033 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1034 | * already in use. |
1035 | * | |
18004c5d | 1036 | * Must hold slab_mutex. |
8f9f8d9e | 1037 | */ |
6a67368c | 1038 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1039 | { |
1040 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1041 | struct kmem_cache_node *n; |
5e804789 | 1042 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1043 | |
18004c5d | 1044 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 1045 | /* |
5f0985bb | 1046 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
1047 | * begin anything. Make sure some other cpu on this |
1048 | * node has not already allocated this | |
1049 | */ | |
18bf8541 CL |
1050 | n = get_node(cachep, node); |
1051 | if (!n) { | |
ce8eb6c4 CL |
1052 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1053 | if (!n) | |
8f9f8d9e | 1054 | return -ENOMEM; |
ce8eb6c4 | 1055 | kmem_cache_node_init(n); |
5f0985bb JZ |
1056 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1057 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1058 | |
1059 | /* | |
5f0985bb JZ |
1060 | * The kmem_cache_nodes don't come and go as CPUs |
1061 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1062 | * protection here. |
1063 | */ | |
ce8eb6c4 | 1064 | cachep->node[node] = n; |
8f9f8d9e DR |
1065 | } |
1066 | ||
18bf8541 CL |
1067 | spin_lock_irq(&n->list_lock); |
1068 | n->free_limit = | |
8f9f8d9e DR |
1069 | (1 + nr_cpus_node(node)) * |
1070 | cachep->batchcount + cachep->num; | |
18bf8541 | 1071 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1072 | } |
1073 | return 0; | |
1074 | } | |
1075 | ||
0fa8103b WL |
1076 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1077 | struct kmem_cache_node *n) | |
1078 | { | |
1079 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1080 | } | |
1081 | ||
0db0628d | 1082 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1083 | { |
1084 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1085 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1086 | int node = cpu_to_mem(cpu); |
a70f7302 | 1087 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1088 | |
18004c5d | 1089 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1090 | struct array_cache *nc; |
1091 | struct array_cache *shared; | |
c8522a3a | 1092 | struct alien_cache **alien; |
97654dfa | 1093 | LIST_HEAD(list); |
fbf1e473 | 1094 | |
fbf1e473 AM |
1095 | /* cpu is dead; no one can alloc from it. */ |
1096 | nc = cachep->array[cpu]; | |
1097 | cachep->array[cpu] = NULL; | |
18bf8541 | 1098 | n = get_node(cachep, node); |
fbf1e473 | 1099 | |
ce8eb6c4 | 1100 | if (!n) |
fbf1e473 AM |
1101 | goto free_array_cache; |
1102 | ||
ce8eb6c4 | 1103 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1104 | |
ce8eb6c4 CL |
1105 | /* Free limit for this kmem_cache_node */ |
1106 | n->free_limit -= cachep->batchcount; | |
fbf1e473 | 1107 | if (nc) |
97654dfa | 1108 | free_block(cachep, nc->entry, nc->avail, node, &list); |
fbf1e473 | 1109 | |
58463c1f | 1110 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1111 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1112 | goto free_array_cache; |
1113 | } | |
1114 | ||
ce8eb6c4 | 1115 | shared = n->shared; |
fbf1e473 AM |
1116 | if (shared) { |
1117 | free_block(cachep, shared->entry, | |
97654dfa | 1118 | shared->avail, node, &list); |
ce8eb6c4 | 1119 | n->shared = NULL; |
fbf1e473 AM |
1120 | } |
1121 | ||
ce8eb6c4 CL |
1122 | alien = n->alien; |
1123 | n->alien = NULL; | |
fbf1e473 | 1124 | |
ce8eb6c4 | 1125 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1126 | |
1127 | kfree(shared); | |
1128 | if (alien) { | |
1129 | drain_alien_cache(cachep, alien); | |
1130 | free_alien_cache(alien); | |
1131 | } | |
1132 | free_array_cache: | |
97654dfa | 1133 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1134 | kfree(nc); |
1135 | } | |
1136 | /* | |
1137 | * In the previous loop, all the objects were freed to | |
1138 | * the respective cache's slabs, now we can go ahead and | |
1139 | * shrink each nodelist to its limit. | |
1140 | */ | |
18004c5d | 1141 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1142 | n = get_node(cachep, node); |
ce8eb6c4 | 1143 | if (!n) |
fbf1e473 | 1144 | continue; |
0fa8103b | 1145 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1146 | } |
1147 | } | |
1148 | ||
0db0628d | 1149 | static int cpuup_prepare(long cpu) |
1da177e4 | 1150 | { |
343e0d7a | 1151 | struct kmem_cache *cachep; |
ce8eb6c4 | 1152 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1153 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1154 | int err; |
1da177e4 | 1155 | |
fbf1e473 AM |
1156 | /* |
1157 | * We need to do this right in the beginning since | |
1158 | * alloc_arraycache's are going to use this list. | |
1159 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1160 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1161 | */ |
6a67368c | 1162 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1163 | if (err < 0) |
1164 | goto bad; | |
fbf1e473 AM |
1165 | |
1166 | /* | |
1167 | * Now we can go ahead with allocating the shared arrays and | |
1168 | * array caches | |
1169 | */ | |
18004c5d | 1170 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1171 | struct array_cache *nc; |
1172 | struct array_cache *shared = NULL; | |
c8522a3a | 1173 | struct alien_cache **alien = NULL; |
fbf1e473 AM |
1174 | |
1175 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1176 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1177 | if (!nc) |
1178 | goto bad; | |
1179 | if (cachep->shared) { | |
1180 | shared = alloc_arraycache(node, | |
1181 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1182 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1183 | if (!shared) { |
1184 | kfree(nc); | |
1da177e4 | 1185 | goto bad; |
12d00f6a | 1186 | } |
fbf1e473 AM |
1187 | } |
1188 | if (use_alien_caches) { | |
83b519e8 | 1189 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1190 | if (!alien) { |
1191 | kfree(shared); | |
1192 | kfree(nc); | |
fbf1e473 | 1193 | goto bad; |
12d00f6a | 1194 | } |
fbf1e473 AM |
1195 | } |
1196 | cachep->array[cpu] = nc; | |
18bf8541 | 1197 | n = get_node(cachep, node); |
ce8eb6c4 | 1198 | BUG_ON(!n); |
fbf1e473 | 1199 | |
ce8eb6c4 CL |
1200 | spin_lock_irq(&n->list_lock); |
1201 | if (!n->shared) { | |
fbf1e473 AM |
1202 | /* |
1203 | * We are serialised from CPU_DEAD or | |
1204 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1205 | */ | |
ce8eb6c4 | 1206 | n->shared = shared; |
fbf1e473 AM |
1207 | shared = NULL; |
1208 | } | |
4484ebf1 | 1209 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1210 | if (!n->alien) { |
1211 | n->alien = alien; | |
fbf1e473 | 1212 | alien = NULL; |
1da177e4 | 1213 | } |
fbf1e473 | 1214 | #endif |
ce8eb6c4 | 1215 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1216 | kfree(shared); |
1217 | free_alien_cache(alien); | |
1218 | } | |
ce79ddc8 | 1219 | |
fbf1e473 AM |
1220 | return 0; |
1221 | bad: | |
12d00f6a | 1222 | cpuup_canceled(cpu); |
fbf1e473 AM |
1223 | return -ENOMEM; |
1224 | } | |
1225 | ||
0db0628d | 1226 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1227 | unsigned long action, void *hcpu) |
1228 | { | |
1229 | long cpu = (long)hcpu; | |
1230 | int err = 0; | |
1231 | ||
1232 | switch (action) { | |
fbf1e473 AM |
1233 | case CPU_UP_PREPARE: |
1234 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1235 | mutex_lock(&slab_mutex); |
fbf1e473 | 1236 | err = cpuup_prepare(cpu); |
18004c5d | 1237 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1238 | break; |
1239 | case CPU_ONLINE: | |
8bb78442 | 1240 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1241 | start_cpu_timer(cpu); |
1242 | break; | |
1243 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1244 | case CPU_DOWN_PREPARE: |
8bb78442 | 1245 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1246 | /* |
18004c5d | 1247 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1248 | * held so that if cache_reap() is invoked it cannot do |
1249 | * anything expensive but will only modify reap_work | |
1250 | * and reschedule the timer. | |
1251 | */ | |
afe2c511 | 1252 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1253 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1254 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1255 | break; |
1256 | case CPU_DOWN_FAILED: | |
8bb78442 | 1257 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1258 | start_cpu_timer(cpu); |
1259 | break; | |
1da177e4 | 1260 | case CPU_DEAD: |
8bb78442 | 1261 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1262 | /* |
1263 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1264 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1265 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1266 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1267 | * structure is usually allocated from kmem_cache_create() and |
1268 | * gets destroyed at kmem_cache_destroy(). | |
1269 | */ | |
183ff22b | 1270 | /* fall through */ |
8f5be20b | 1271 | #endif |
1da177e4 | 1272 | case CPU_UP_CANCELED: |
8bb78442 | 1273 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1274 | mutex_lock(&slab_mutex); |
fbf1e473 | 1275 | cpuup_canceled(cpu); |
18004c5d | 1276 | mutex_unlock(&slab_mutex); |
1da177e4 | 1277 | break; |
1da177e4 | 1278 | } |
eac40680 | 1279 | return notifier_from_errno(err); |
1da177e4 LT |
1280 | } |
1281 | ||
0db0628d | 1282 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1283 | &cpuup_callback, NULL, 0 |
1284 | }; | |
1da177e4 | 1285 | |
8f9f8d9e DR |
1286 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1287 | /* | |
1288 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1289 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1290 | * removed. | |
1291 | * | |
18004c5d | 1292 | * Must hold slab_mutex. |
8f9f8d9e | 1293 | */ |
6a67368c | 1294 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1295 | { |
1296 | struct kmem_cache *cachep; | |
1297 | int ret = 0; | |
1298 | ||
18004c5d | 1299 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1300 | struct kmem_cache_node *n; |
8f9f8d9e | 1301 | |
18bf8541 | 1302 | n = get_node(cachep, node); |
ce8eb6c4 | 1303 | if (!n) |
8f9f8d9e DR |
1304 | continue; |
1305 | ||
0fa8103b | 1306 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1307 | |
ce8eb6c4 CL |
1308 | if (!list_empty(&n->slabs_full) || |
1309 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1310 | ret = -EBUSY; |
1311 | break; | |
1312 | } | |
1313 | } | |
1314 | return ret; | |
1315 | } | |
1316 | ||
1317 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1318 | unsigned long action, void *arg) | |
1319 | { | |
1320 | struct memory_notify *mnb = arg; | |
1321 | int ret = 0; | |
1322 | int nid; | |
1323 | ||
1324 | nid = mnb->status_change_nid; | |
1325 | if (nid < 0) | |
1326 | goto out; | |
1327 | ||
1328 | switch (action) { | |
1329 | case MEM_GOING_ONLINE: | |
18004c5d | 1330 | mutex_lock(&slab_mutex); |
6a67368c | 1331 | ret = init_cache_node_node(nid); |
18004c5d | 1332 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1333 | break; |
1334 | case MEM_GOING_OFFLINE: | |
18004c5d | 1335 | mutex_lock(&slab_mutex); |
6a67368c | 1336 | ret = drain_cache_node_node(nid); |
18004c5d | 1337 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1338 | break; |
1339 | case MEM_ONLINE: | |
1340 | case MEM_OFFLINE: | |
1341 | case MEM_CANCEL_ONLINE: | |
1342 | case MEM_CANCEL_OFFLINE: | |
1343 | break; | |
1344 | } | |
1345 | out: | |
5fda1bd5 | 1346 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1347 | } |
1348 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1349 | ||
e498be7d | 1350 | /* |
ce8eb6c4 | 1351 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1352 | */ |
6744f087 | 1353 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1354 | int nodeid) |
e498be7d | 1355 | { |
6744f087 | 1356 | struct kmem_cache_node *ptr; |
e498be7d | 1357 | |
6744f087 | 1358 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1359 | BUG_ON(!ptr); |
1360 | ||
6744f087 | 1361 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1362 | /* |
1363 | * Do not assume that spinlocks can be initialized via memcpy: | |
1364 | */ | |
1365 | spin_lock_init(&ptr->list_lock); | |
1366 | ||
e498be7d | 1367 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1368 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1369 | } |
1370 | ||
556a169d | 1371 | /* |
ce8eb6c4 CL |
1372 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1373 | * size of kmem_cache_node. | |
556a169d | 1374 | */ |
ce8eb6c4 | 1375 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1376 | { |
1377 | int node; | |
1378 | ||
1379 | for_each_online_node(node) { | |
ce8eb6c4 | 1380 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1381 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1382 | REAPTIMEOUT_NODE + |
1383 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1384 | } |
1385 | } | |
1386 | ||
3c583465 CL |
1387 | /* |
1388 | * The memory after the last cpu cache pointer is used for the | |
6a67368c | 1389 | * the node pointer. |
3c583465 | 1390 | */ |
6a67368c | 1391 | static void setup_node_pointer(struct kmem_cache *cachep) |
3c583465 | 1392 | { |
6a67368c | 1393 | cachep->node = (struct kmem_cache_node **)&cachep->array[nr_cpu_ids]; |
3c583465 CL |
1394 | } |
1395 | ||
a737b3e2 AM |
1396 | /* |
1397 | * Initialisation. Called after the page allocator have been initialised and | |
1398 | * before smp_init(). | |
1da177e4 LT |
1399 | */ |
1400 | void __init kmem_cache_init(void) | |
1401 | { | |
e498be7d CL |
1402 | int i; |
1403 | ||
68126702 JK |
1404 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1405 | sizeof(struct rcu_head)); | |
9b030cb8 | 1406 | kmem_cache = &kmem_cache_boot; |
6a67368c | 1407 | setup_node_pointer(kmem_cache); |
9b030cb8 | 1408 | |
b6e68bc1 | 1409 | if (num_possible_nodes() == 1) |
62918a03 SS |
1410 | use_alien_caches = 0; |
1411 | ||
3c583465 | 1412 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1413 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1414 | |
ce8eb6c4 | 1415 | set_up_node(kmem_cache, CACHE_CACHE); |
1da177e4 LT |
1416 | |
1417 | /* | |
1418 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1419 | * page orders on machines with more than 32MB of memory if |
1420 | * not overridden on the command line. | |
1da177e4 | 1421 | */ |
3df1cccd | 1422 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1423 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1424 | |
1da177e4 LT |
1425 | /* Bootstrap is tricky, because several objects are allocated |
1426 | * from caches that do not exist yet: | |
9b030cb8 CL |
1427 | * 1) initialize the kmem_cache cache: it contains the struct |
1428 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1429 | * kmem_cache is statically allocated. | |
e498be7d | 1430 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1431 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1432 | * array at the end of the bootstrap. |
1da177e4 | 1433 | * 2) Create the first kmalloc cache. |
343e0d7a | 1434 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1435 | * An __init data area is used for the head array. |
1436 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1437 | * head arrays. | |
9b030cb8 | 1438 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1439 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1440 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1441 | * the other cache's with kmalloc allocated memory. |
1442 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1443 | */ |
1444 | ||
9b030cb8 | 1445 | /* 1) create the kmem_cache */ |
1da177e4 | 1446 | |
8da3430d | 1447 | /* |
b56efcf0 | 1448 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1449 | */ |
2f9baa9f CL |
1450 | create_boot_cache(kmem_cache, "kmem_cache", |
1451 | offsetof(struct kmem_cache, array[nr_cpu_ids]) + | |
6744f087 | 1452 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1453 | SLAB_HWCACHE_ALIGN); |
1454 | list_add(&kmem_cache->list, &slab_caches); | |
1da177e4 LT |
1455 | |
1456 | /* 2+3) create the kmalloc caches */ | |
1da177e4 | 1457 | |
a737b3e2 AM |
1458 | /* |
1459 | * Initialize the caches that provide memory for the array cache and the | |
ce8eb6c4 | 1460 | * kmem_cache_node structures first. Without this, further allocations will |
a737b3e2 | 1461 | * bug. |
e498be7d CL |
1462 | */ |
1463 | ||
e3366016 CL |
1464 | kmalloc_caches[INDEX_AC] = create_kmalloc_cache("kmalloc-ac", |
1465 | kmalloc_size(INDEX_AC), ARCH_KMALLOC_FLAGS); | |
45530c44 | 1466 | |
ce8eb6c4 CL |
1467 | if (INDEX_AC != INDEX_NODE) |
1468 | kmalloc_caches[INDEX_NODE] = | |
1469 | create_kmalloc_cache("kmalloc-node", | |
1470 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); | |
e498be7d | 1471 | |
e0a42726 IM |
1472 | slab_early_init = 0; |
1473 | ||
1da177e4 LT |
1474 | /* 4) Replace the bootstrap head arrays */ |
1475 | { | |
2b2d5493 | 1476 | struct array_cache *ptr; |
e498be7d | 1477 | |
83b519e8 | 1478 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1479 | |
9b030cb8 | 1480 | memcpy(ptr, cpu_cache_get(kmem_cache), |
b28a02de | 1481 | sizeof(struct arraycache_init)); |
2b2d5493 | 1482 | |
9b030cb8 | 1483 | kmem_cache->array[smp_processor_id()] = ptr; |
e498be7d | 1484 | |
83b519e8 | 1485 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1486 | |
e3366016 | 1487 | BUG_ON(cpu_cache_get(kmalloc_caches[INDEX_AC]) |
b28a02de | 1488 | != &initarray_generic.cache); |
e3366016 | 1489 | memcpy(ptr, cpu_cache_get(kmalloc_caches[INDEX_AC]), |
b28a02de | 1490 | sizeof(struct arraycache_init)); |
2b2d5493 | 1491 | |
e3366016 | 1492 | kmalloc_caches[INDEX_AC]->array[smp_processor_id()] = ptr; |
1da177e4 | 1493 | } |
ce8eb6c4 | 1494 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1495 | { |
1ca4cb24 PE |
1496 | int nid; |
1497 | ||
9c09a95c | 1498 | for_each_online_node(nid) { |
ce8eb6c4 | 1499 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1500 | |
e3366016 | 1501 | init_list(kmalloc_caches[INDEX_AC], |
ce8eb6c4 | 1502 | &init_kmem_cache_node[SIZE_AC + nid], nid); |
e498be7d | 1503 | |
ce8eb6c4 CL |
1504 | if (INDEX_AC != INDEX_NODE) { |
1505 | init_list(kmalloc_caches[INDEX_NODE], | |
1506 | &init_kmem_cache_node[SIZE_NODE + nid], nid); | |
e498be7d CL |
1507 | } |
1508 | } | |
1509 | } | |
1da177e4 | 1510 | |
f97d5f63 | 1511 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1512 | } |
1513 | ||
1514 | void __init kmem_cache_init_late(void) | |
1515 | { | |
1516 | struct kmem_cache *cachep; | |
1517 | ||
97d06609 | 1518 | slab_state = UP; |
52cef189 | 1519 | |
8429db5c | 1520 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1521 | mutex_lock(&slab_mutex); |
1522 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1523 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1524 | BUG(); | |
18004c5d | 1525 | mutex_unlock(&slab_mutex); |
056c6241 | 1526 | |
97d06609 CL |
1527 | /* Done! */ |
1528 | slab_state = FULL; | |
1529 | ||
a737b3e2 AM |
1530 | /* |
1531 | * Register a cpu startup notifier callback that initializes | |
1532 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1533 | */ |
1534 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1535 | |
8f9f8d9e DR |
1536 | #ifdef CONFIG_NUMA |
1537 | /* | |
1538 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1539 | * node. |
8f9f8d9e DR |
1540 | */ |
1541 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1542 | #endif | |
1543 | ||
a737b3e2 AM |
1544 | /* |
1545 | * The reap timers are started later, with a module init call: That part | |
1546 | * of the kernel is not yet operational. | |
1da177e4 LT |
1547 | */ |
1548 | } | |
1549 | ||
1550 | static int __init cpucache_init(void) | |
1551 | { | |
1552 | int cpu; | |
1553 | ||
a737b3e2 AM |
1554 | /* |
1555 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1556 | */ |
e498be7d | 1557 | for_each_online_cpu(cpu) |
a737b3e2 | 1558 | start_cpu_timer(cpu); |
a164f896 GC |
1559 | |
1560 | /* Done! */ | |
97d06609 | 1561 | slab_state = FULL; |
1da177e4 LT |
1562 | return 0; |
1563 | } | |
1da177e4 LT |
1564 | __initcall(cpucache_init); |
1565 | ||
8bdec192 RA |
1566 | static noinline void |
1567 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1568 | { | |
9a02d699 | 1569 | #if DEBUG |
ce8eb6c4 | 1570 | struct kmem_cache_node *n; |
8456a648 | 1571 | struct page *page; |
8bdec192 RA |
1572 | unsigned long flags; |
1573 | int node; | |
9a02d699 DR |
1574 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1575 | DEFAULT_RATELIMIT_BURST); | |
1576 | ||
1577 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1578 | return; | |
8bdec192 RA |
1579 | |
1580 | printk(KERN_WARNING | |
1581 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1582 | nodeid, gfpflags); | |
1583 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1584 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1585 | |
18bf8541 | 1586 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1587 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1588 | unsigned long active_slabs = 0, num_slabs = 0; | |
1589 | ||
ce8eb6c4 | 1590 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1591 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1592 | active_objs += cachep->num; |
1593 | active_slabs++; | |
1594 | } | |
8456a648 JK |
1595 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1596 | active_objs += page->active; | |
8bdec192 RA |
1597 | active_slabs++; |
1598 | } | |
8456a648 | 1599 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1600 | num_slabs++; |
1601 | ||
ce8eb6c4 CL |
1602 | free_objects += n->free_objects; |
1603 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1604 | |
1605 | num_slabs += active_slabs; | |
1606 | num_objs = num_slabs * cachep->num; | |
1607 | printk(KERN_WARNING | |
1608 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1609 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1610 | free_objects); | |
1611 | } | |
9a02d699 | 1612 | #endif |
8bdec192 RA |
1613 | } |
1614 | ||
1da177e4 | 1615 | /* |
8a7d9b43 WSH |
1616 | * Interface to system's page allocator. No need to hold the |
1617 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1618 | * |
1619 | * If we requested dmaable memory, we will get it. Even if we | |
1620 | * did not request dmaable memory, we might get it, but that | |
1621 | * would be relatively rare and ignorable. | |
1622 | */ | |
0c3aa83e JK |
1623 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1624 | int nodeid) | |
1da177e4 LT |
1625 | { |
1626 | struct page *page; | |
e1b6aa6f | 1627 | int nr_pages; |
765c4507 | 1628 | |
a618e89f | 1629 | flags |= cachep->allocflags; |
e12ba74d MG |
1630 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1631 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1632 | |
5dfb4175 VD |
1633 | if (memcg_charge_slab(cachep, flags, cachep->gfporder)) |
1634 | return NULL; | |
1635 | ||
517d0869 | 1636 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1637 | if (!page) { |
5dfb4175 | 1638 | memcg_uncharge_slab(cachep, cachep->gfporder); |
9a02d699 | 1639 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1640 | return NULL; |
8bdec192 | 1641 | } |
1da177e4 | 1642 | |
b37f1dd0 | 1643 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
072bb0aa MG |
1644 | if (unlikely(page->pfmemalloc)) |
1645 | pfmemalloc_active = true; | |
1646 | ||
e1b6aa6f | 1647 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1648 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1649 | add_zone_page_state(page_zone(page), |
1650 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1651 | else | |
1652 | add_zone_page_state(page_zone(page), | |
1653 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 JK |
1654 | __SetPageSlab(page); |
1655 | if (page->pfmemalloc) | |
1656 | SetPageSlabPfmemalloc(page); | |
072bb0aa | 1657 | |
b1eeab67 VN |
1658 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1659 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1660 | ||
1661 | if (cachep->ctor) | |
1662 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1663 | else | |
1664 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1665 | } | |
c175eea4 | 1666 | |
0c3aa83e | 1667 | return page; |
1da177e4 LT |
1668 | } |
1669 | ||
1670 | /* | |
1671 | * Interface to system's page release. | |
1672 | */ | |
0c3aa83e | 1673 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1674 | { |
a57a4988 | 1675 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1676 | |
b1eeab67 | 1677 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1678 | |
972d1a7b CL |
1679 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1680 | sub_zone_page_state(page_zone(page), | |
1681 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1682 | else | |
1683 | sub_zone_page_state(page_zone(page), | |
1684 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1685 | |
a57a4988 | 1686 | BUG_ON(!PageSlab(page)); |
73293c2f | 1687 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1688 | __ClearPageSlab(page); |
8456a648 JK |
1689 | page_mapcount_reset(page); |
1690 | page->mapping = NULL; | |
1f458cbf | 1691 | |
1da177e4 LT |
1692 | if (current->reclaim_state) |
1693 | current->reclaim_state->reclaimed_slab += nr_freed; | |
5dfb4175 VD |
1694 | __free_pages(page, cachep->gfporder); |
1695 | memcg_uncharge_slab(cachep, cachep->gfporder); | |
1da177e4 LT |
1696 | } |
1697 | ||
1698 | static void kmem_rcu_free(struct rcu_head *head) | |
1699 | { | |
68126702 JK |
1700 | struct kmem_cache *cachep; |
1701 | struct page *page; | |
1da177e4 | 1702 | |
68126702 JK |
1703 | page = container_of(head, struct page, rcu_head); |
1704 | cachep = page->slab_cache; | |
1705 | ||
1706 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1707 | } |
1708 | ||
1709 | #if DEBUG | |
1710 | ||
1711 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1712 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1713 | unsigned long caller) |
1da177e4 | 1714 | { |
8c138bc0 | 1715 | int size = cachep->object_size; |
1da177e4 | 1716 | |
3dafccf2 | 1717 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1718 | |
b28a02de | 1719 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1720 | return; |
1721 | ||
b28a02de PE |
1722 | *addr++ = 0x12345678; |
1723 | *addr++ = caller; | |
1724 | *addr++ = smp_processor_id(); | |
1725 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1726 | { |
1727 | unsigned long *sptr = &caller; | |
1728 | unsigned long svalue; | |
1729 | ||
1730 | while (!kstack_end(sptr)) { | |
1731 | svalue = *sptr++; | |
1732 | if (kernel_text_address(svalue)) { | |
b28a02de | 1733 | *addr++ = svalue; |
1da177e4 LT |
1734 | size -= sizeof(unsigned long); |
1735 | if (size <= sizeof(unsigned long)) | |
1736 | break; | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | } | |
b28a02de | 1741 | *addr++ = 0x87654321; |
1da177e4 LT |
1742 | } |
1743 | #endif | |
1744 | ||
343e0d7a | 1745 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1746 | { |
8c138bc0 | 1747 | int size = cachep->object_size; |
3dafccf2 | 1748 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1749 | |
1750 | memset(addr, val, size); | |
b28a02de | 1751 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1752 | } |
1753 | ||
1754 | static void dump_line(char *data, int offset, int limit) | |
1755 | { | |
1756 | int i; | |
aa83aa40 DJ |
1757 | unsigned char error = 0; |
1758 | int bad_count = 0; | |
1759 | ||
fdde6abb | 1760 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1761 | for (i = 0; i < limit; i++) { |
1762 | if (data[offset + i] != POISON_FREE) { | |
1763 | error = data[offset + i]; | |
1764 | bad_count++; | |
1765 | } | |
aa83aa40 | 1766 | } |
fdde6abb SAS |
1767 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1768 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1769 | |
1770 | if (bad_count == 1) { | |
1771 | error ^= POISON_FREE; | |
1772 | if (!(error & (error - 1))) { | |
1773 | printk(KERN_ERR "Single bit error detected. Probably " | |
1774 | "bad RAM.\n"); | |
1775 | #ifdef CONFIG_X86 | |
1776 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1777 | "test tool.\n"); | |
1778 | #else | |
1779 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1780 | #endif | |
1781 | } | |
1782 | } | |
1da177e4 LT |
1783 | } |
1784 | #endif | |
1785 | ||
1786 | #if DEBUG | |
1787 | ||
343e0d7a | 1788 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1789 | { |
1790 | int i, size; | |
1791 | char *realobj; | |
1792 | ||
1793 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1794 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1795 | *dbg_redzone1(cachep, objp), |
1796 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1797 | } |
1798 | ||
1799 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1800 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1801 | *dbg_userword(cachep, objp), | |
1802 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1803 | } |
3dafccf2 | 1804 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1805 | size = cachep->object_size; |
b28a02de | 1806 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1807 | int limit; |
1808 | limit = 16; | |
b28a02de PE |
1809 | if (i + limit > size) |
1810 | limit = size - i; | |
1da177e4 LT |
1811 | dump_line(realobj, i, limit); |
1812 | } | |
1813 | } | |
1814 | ||
343e0d7a | 1815 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1816 | { |
1817 | char *realobj; | |
1818 | int size, i; | |
1819 | int lines = 0; | |
1820 | ||
3dafccf2 | 1821 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1822 | size = cachep->object_size; |
1da177e4 | 1823 | |
b28a02de | 1824 | for (i = 0; i < size; i++) { |
1da177e4 | 1825 | char exp = POISON_FREE; |
b28a02de | 1826 | if (i == size - 1) |
1da177e4 LT |
1827 | exp = POISON_END; |
1828 | if (realobj[i] != exp) { | |
1829 | int limit; | |
1830 | /* Mismatch ! */ | |
1831 | /* Print header */ | |
1832 | if (lines == 0) { | |
b28a02de | 1833 | printk(KERN_ERR |
face37f5 DJ |
1834 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1835 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1836 | print_objinfo(cachep, objp, 0); |
1837 | } | |
1838 | /* Hexdump the affected line */ | |
b28a02de | 1839 | i = (i / 16) * 16; |
1da177e4 | 1840 | limit = 16; |
b28a02de PE |
1841 | if (i + limit > size) |
1842 | limit = size - i; | |
1da177e4 LT |
1843 | dump_line(realobj, i, limit); |
1844 | i += 16; | |
1845 | lines++; | |
1846 | /* Limit to 5 lines */ | |
1847 | if (lines > 5) | |
1848 | break; | |
1849 | } | |
1850 | } | |
1851 | if (lines != 0) { | |
1852 | /* Print some data about the neighboring objects, if they | |
1853 | * exist: | |
1854 | */ | |
8456a648 | 1855 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1856 | unsigned int objnr; |
1da177e4 | 1857 | |
8456a648 | 1858 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1859 | if (objnr) { |
8456a648 | 1860 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1861 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1862 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1863 | realobj, size); |
1da177e4 LT |
1864 | print_objinfo(cachep, objp, 2); |
1865 | } | |
b28a02de | 1866 | if (objnr + 1 < cachep->num) { |
8456a648 | 1867 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1868 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1869 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1870 | realobj, size); |
1da177e4 LT |
1871 | print_objinfo(cachep, objp, 2); |
1872 | } | |
1873 | } | |
1874 | } | |
1875 | #endif | |
1876 | ||
12dd36fa | 1877 | #if DEBUG |
8456a648 JK |
1878 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1879 | struct page *page) | |
1da177e4 | 1880 | { |
1da177e4 LT |
1881 | int i; |
1882 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 1883 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1884 | |
1885 | if (cachep->flags & SLAB_POISON) { | |
1886 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 1887 | if (cachep->size % PAGE_SIZE == 0 && |
a737b3e2 | 1888 | OFF_SLAB(cachep)) |
b28a02de | 1889 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 1890 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
1891 | else |
1892 | check_poison_obj(cachep, objp); | |
1893 | #else | |
1894 | check_poison_obj(cachep, objp); | |
1895 | #endif | |
1896 | } | |
1897 | if (cachep->flags & SLAB_RED_ZONE) { | |
1898 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1899 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1900 | "was overwritten"); |
1da177e4 LT |
1901 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1902 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1903 | "was overwritten"); |
1da177e4 | 1904 | } |
1da177e4 | 1905 | } |
12dd36fa | 1906 | } |
1da177e4 | 1907 | #else |
8456a648 JK |
1908 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1909 | struct page *page) | |
12dd36fa | 1910 | { |
12dd36fa | 1911 | } |
1da177e4 LT |
1912 | #endif |
1913 | ||
911851e6 RD |
1914 | /** |
1915 | * slab_destroy - destroy and release all objects in a slab | |
1916 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1917 | * @page: page pointer being destroyed |
911851e6 | 1918 | * |
8a7d9b43 WSH |
1919 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1920 | * Before calling the slab page must have been unlinked from the cache. The | |
1921 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1922 | */ |
8456a648 | 1923 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1924 | { |
7e007355 | 1925 | void *freelist; |
12dd36fa | 1926 | |
8456a648 JK |
1927 | freelist = page->freelist; |
1928 | slab_destroy_debugcheck(cachep, page); | |
1da177e4 | 1929 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
68126702 JK |
1930 | struct rcu_head *head; |
1931 | ||
1932 | /* | |
1933 | * RCU free overloads the RCU head over the LRU. | |
1934 | * slab_page has been overloeaded over the LRU, | |
1935 | * however it is not used from now on so that | |
1936 | * we can use it safely. | |
1937 | */ | |
1938 | head = (void *)&page->rcu_head; | |
1939 | call_rcu(head, kmem_rcu_free); | |
1da177e4 | 1940 | |
1da177e4 | 1941 | } else { |
0c3aa83e | 1942 | kmem_freepages(cachep, page); |
1da177e4 | 1943 | } |
68126702 JK |
1944 | |
1945 | /* | |
8456a648 | 1946 | * From now on, we don't use freelist |
68126702 JK |
1947 | * although actual page can be freed in rcu context |
1948 | */ | |
1949 | if (OFF_SLAB(cachep)) | |
8456a648 | 1950 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1951 | } |
1952 | ||
97654dfa JK |
1953 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1954 | { | |
1955 | struct page *page, *n; | |
1956 | ||
1957 | list_for_each_entry_safe(page, n, list, lru) { | |
1958 | list_del(&page->lru); | |
1959 | slab_destroy(cachep, page); | |
1960 | } | |
1961 | } | |
1962 | ||
4d268eba | 1963 | /** |
a70773dd RD |
1964 | * calculate_slab_order - calculate size (page order) of slabs |
1965 | * @cachep: pointer to the cache that is being created | |
1966 | * @size: size of objects to be created in this cache. | |
1967 | * @align: required alignment for the objects. | |
1968 | * @flags: slab allocation flags | |
1969 | * | |
1970 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1971 | * |
1972 | * This could be made much more intelligent. For now, try to avoid using | |
1973 | * high order pages for slabs. When the gfp() functions are more friendly | |
1974 | * towards high-order requests, this should be changed. | |
1975 | */ | |
a737b3e2 | 1976 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 1977 | size_t size, size_t align, unsigned long flags) |
4d268eba | 1978 | { |
b1ab41c4 | 1979 | unsigned long offslab_limit; |
4d268eba | 1980 | size_t left_over = 0; |
9888e6fa | 1981 | int gfporder; |
4d268eba | 1982 | |
0aa817f0 | 1983 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1984 | unsigned int num; |
1985 | size_t remainder; | |
1986 | ||
9888e6fa | 1987 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
1988 | if (!num) |
1989 | continue; | |
9888e6fa | 1990 | |
f315e3fa JK |
1991 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1992 | if (num > SLAB_OBJ_MAX_NUM) | |
1993 | break; | |
1994 | ||
b1ab41c4 | 1995 | if (flags & CFLGS_OFF_SLAB) { |
03787301 | 1996 | size_t freelist_size_per_obj = sizeof(freelist_idx_t); |
b1ab41c4 IM |
1997 | /* |
1998 | * Max number of objs-per-slab for caches which | |
1999 | * use off-slab slabs. Needed to avoid a possible | |
2000 | * looping condition in cache_grow(). | |
2001 | */ | |
03787301 JK |
2002 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
2003 | freelist_size_per_obj += sizeof(char); | |
8456a648 | 2004 | offslab_limit = size; |
03787301 | 2005 | offslab_limit /= freelist_size_per_obj; |
b1ab41c4 IM |
2006 | |
2007 | if (num > offslab_limit) | |
2008 | break; | |
2009 | } | |
4d268eba | 2010 | |
9888e6fa | 2011 | /* Found something acceptable - save it away */ |
4d268eba | 2012 | cachep->num = num; |
9888e6fa | 2013 | cachep->gfporder = gfporder; |
4d268eba PE |
2014 | left_over = remainder; |
2015 | ||
f78bb8ad LT |
2016 | /* |
2017 | * A VFS-reclaimable slab tends to have most allocations | |
2018 | * as GFP_NOFS and we really don't want to have to be allocating | |
2019 | * higher-order pages when we are unable to shrink dcache. | |
2020 | */ | |
2021 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
2022 | break; | |
2023 | ||
4d268eba PE |
2024 | /* |
2025 | * Large number of objects is good, but very large slabs are | |
2026 | * currently bad for the gfp()s. | |
2027 | */ | |
543585cc | 2028 | if (gfporder >= slab_max_order) |
4d268eba PE |
2029 | break; |
2030 | ||
9888e6fa LT |
2031 | /* |
2032 | * Acceptable internal fragmentation? | |
2033 | */ | |
a737b3e2 | 2034 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
2035 | break; |
2036 | } | |
2037 | return left_over; | |
2038 | } | |
2039 | ||
83b519e8 | 2040 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2041 | { |
97d06609 | 2042 | if (slab_state >= FULL) |
83b519e8 | 2043 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2044 | |
97d06609 | 2045 | if (slab_state == DOWN) { |
f30cf7d1 | 2046 | /* |
2f9baa9f | 2047 | * Note: Creation of first cache (kmem_cache). |
ce8eb6c4 | 2048 | * The setup_node is taken care |
2f9baa9f CL |
2049 | * of by the caller of __kmem_cache_create |
2050 | */ | |
2051 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2052 | slab_state = PARTIAL; | |
2053 | } else if (slab_state == PARTIAL) { | |
2054 | /* | |
2055 | * Note: the second kmem_cache_create must create the cache | |
f30cf7d1 PE |
2056 | * that's used by kmalloc(24), otherwise the creation of |
2057 | * further caches will BUG(). | |
2058 | */ | |
2059 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2060 | ||
2061 | /* | |
ce8eb6c4 CL |
2062 | * If the cache that's used by kmalloc(sizeof(kmem_cache_node)) is |
2063 | * the second cache, then we need to set up all its node/, | |
f30cf7d1 PE |
2064 | * otherwise the creation of further caches will BUG(). |
2065 | */ | |
ce8eb6c4 CL |
2066 | set_up_node(cachep, SIZE_AC); |
2067 | if (INDEX_AC == INDEX_NODE) | |
2068 | slab_state = PARTIAL_NODE; | |
f30cf7d1 | 2069 | else |
97d06609 | 2070 | slab_state = PARTIAL_ARRAYCACHE; |
f30cf7d1 | 2071 | } else { |
2f9baa9f | 2072 | /* Remaining boot caches */ |
f30cf7d1 | 2073 | cachep->array[smp_processor_id()] = |
83b519e8 | 2074 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 | 2075 | |
97d06609 | 2076 | if (slab_state == PARTIAL_ARRAYCACHE) { |
ce8eb6c4 CL |
2077 | set_up_node(cachep, SIZE_NODE); |
2078 | slab_state = PARTIAL_NODE; | |
f30cf7d1 PE |
2079 | } else { |
2080 | int node; | |
556a169d | 2081 | for_each_online_node(node) { |
6a67368c | 2082 | cachep->node[node] = |
6744f087 | 2083 | kmalloc_node(sizeof(struct kmem_cache_node), |
eb91f1d0 | 2084 | gfp, node); |
6a67368c | 2085 | BUG_ON(!cachep->node[node]); |
ce8eb6c4 | 2086 | kmem_cache_node_init(cachep->node[node]); |
f30cf7d1 PE |
2087 | } |
2088 | } | |
2089 | } | |
6a67368c | 2090 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
2091 | jiffies + REAPTIMEOUT_NODE + |
2092 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
2093 | |
2094 | cpu_cache_get(cachep)->avail = 0; | |
2095 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2096 | cpu_cache_get(cachep)->batchcount = 1; | |
2097 | cpu_cache_get(cachep)->touched = 0; | |
2098 | cachep->batchcount = 1; | |
2099 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2100 | return 0; |
f30cf7d1 PE |
2101 | } |
2102 | ||
1da177e4 | 2103 | /** |
039363f3 | 2104 | * __kmem_cache_create - Create a cache. |
a755b76a | 2105 | * @cachep: cache management descriptor |
1da177e4 | 2106 | * @flags: SLAB flags |
1da177e4 LT |
2107 | * |
2108 | * Returns a ptr to the cache on success, NULL on failure. | |
2109 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2110 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2111 | * |
1da177e4 LT |
2112 | * The flags are |
2113 | * | |
2114 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2115 | * to catch references to uninitialised memory. | |
2116 | * | |
2117 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2118 | * for buffer overruns. | |
2119 | * | |
1da177e4 LT |
2120 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2121 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2122 | * as davem. | |
2123 | */ | |
278b1bb1 | 2124 | int |
8a13a4cc | 2125 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2126 | { |
d4a5fca5 DR |
2127 | size_t left_over, freelist_size; |
2128 | size_t ralign = BYTES_PER_WORD; | |
83b519e8 | 2129 | gfp_t gfp; |
278b1bb1 | 2130 | int err; |
8a13a4cc | 2131 | size_t size = cachep->size; |
1da177e4 | 2132 | |
1da177e4 | 2133 | #if DEBUG |
1da177e4 LT |
2134 | #if FORCED_DEBUG |
2135 | /* | |
2136 | * Enable redzoning and last user accounting, except for caches with | |
2137 | * large objects, if the increased size would increase the object size | |
2138 | * above the next power of two: caches with object sizes just above a | |
2139 | * power of two have a significant amount of internal fragmentation. | |
2140 | */ | |
87a927c7 DW |
2141 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2142 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2143 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2144 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2145 | flags |= SLAB_POISON; | |
2146 | #endif | |
2147 | if (flags & SLAB_DESTROY_BY_RCU) | |
2148 | BUG_ON(flags & SLAB_POISON); | |
2149 | #endif | |
1da177e4 | 2150 | |
a737b3e2 AM |
2151 | /* |
2152 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2153 | * unaligned accesses for some archs when redzoning is used, and makes |
2154 | * sure any on-slab bufctl's are also correctly aligned. | |
2155 | */ | |
b28a02de PE |
2156 | if (size & (BYTES_PER_WORD - 1)) { |
2157 | size += (BYTES_PER_WORD - 1); | |
2158 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2159 | } |
2160 | ||
87a927c7 DW |
2161 | if (flags & SLAB_RED_ZONE) { |
2162 | ralign = REDZONE_ALIGN; | |
2163 | /* If redzoning, ensure that the second redzone is suitably | |
2164 | * aligned, by adjusting the object size accordingly. */ | |
2165 | size += REDZONE_ALIGN - 1; | |
2166 | size &= ~(REDZONE_ALIGN - 1); | |
2167 | } | |
ca5f9703 | 2168 | |
a44b56d3 | 2169 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2170 | if (ralign < cachep->align) { |
2171 | ralign = cachep->align; | |
1da177e4 | 2172 | } |
3ff84a7f PE |
2173 | /* disable debug if necessary */ |
2174 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2175 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2176 | /* |
ca5f9703 | 2177 | * 4) Store it. |
1da177e4 | 2178 | */ |
8a13a4cc | 2179 | cachep->align = ralign; |
1da177e4 | 2180 | |
83b519e8 PE |
2181 | if (slab_is_available()) |
2182 | gfp = GFP_KERNEL; | |
2183 | else | |
2184 | gfp = GFP_NOWAIT; | |
2185 | ||
6a67368c | 2186 | setup_node_pointer(cachep); |
1da177e4 | 2187 | #if DEBUG |
1da177e4 | 2188 | |
ca5f9703 PE |
2189 | /* |
2190 | * Both debugging options require word-alignment which is calculated | |
2191 | * into align above. | |
2192 | */ | |
1da177e4 | 2193 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2194 | /* add space for red zone words */ |
3ff84a7f PE |
2195 | cachep->obj_offset += sizeof(unsigned long long); |
2196 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2197 | } |
2198 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2199 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2200 | * the real object. But if the second red zone needs to be |
2201 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2202 | */ |
87a927c7 DW |
2203 | if (flags & SLAB_RED_ZONE) |
2204 | size += REDZONE_ALIGN; | |
2205 | else | |
2206 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2207 | } |
2208 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
ce8eb6c4 | 2209 | if (size >= kmalloc_size(INDEX_NODE + 1) |
608da7e3 TH |
2210 | && cachep->object_size > cache_line_size() |
2211 | && ALIGN(size, cachep->align) < PAGE_SIZE) { | |
2212 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); | |
1da177e4 LT |
2213 | size = PAGE_SIZE; |
2214 | } | |
2215 | #endif | |
2216 | #endif | |
2217 | ||
e0a42726 IM |
2218 | /* |
2219 | * Determine if the slab management is 'on' or 'off' slab. | |
2220 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2221 | * it too early on. Always use on-slab management when |
2222 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2223 | */ |
8fc9cf42 | 2224 | if ((size >= (PAGE_SIZE >> 5)) && !slab_early_init && |
e7cb55b9 | 2225 | !(flags & SLAB_NOLEAKTRACE)) |
1da177e4 LT |
2226 | /* |
2227 | * Size is large, assume best to place the slab management obj | |
2228 | * off-slab (should allow better packing of objs). | |
2229 | */ | |
2230 | flags |= CFLGS_OFF_SLAB; | |
2231 | ||
8a13a4cc | 2232 | size = ALIGN(size, cachep->align); |
f315e3fa JK |
2233 | /* |
2234 | * We should restrict the number of objects in a slab to implement | |
2235 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2236 | */ | |
2237 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2238 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
1da177e4 | 2239 | |
8a13a4cc | 2240 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2241 | |
8a13a4cc | 2242 | if (!cachep->num) |
278b1bb1 | 2243 | return -E2BIG; |
1da177e4 | 2244 | |
03787301 | 2245 | freelist_size = calculate_freelist_size(cachep->num, cachep->align); |
1da177e4 LT |
2246 | |
2247 | /* | |
2248 | * If the slab has been placed off-slab, and we have enough space then | |
2249 | * move it on-slab. This is at the expense of any extra colouring. | |
2250 | */ | |
8456a648 | 2251 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
1da177e4 | 2252 | flags &= ~CFLGS_OFF_SLAB; |
8456a648 | 2253 | left_over -= freelist_size; |
1da177e4 LT |
2254 | } |
2255 | ||
2256 | if (flags & CFLGS_OFF_SLAB) { | |
2257 | /* really off slab. No need for manual alignment */ | |
03787301 | 2258 | freelist_size = calculate_freelist_size(cachep->num, 0); |
67461365 RL |
2259 | |
2260 | #ifdef CONFIG_PAGE_POISONING | |
2261 | /* If we're going to use the generic kernel_map_pages() | |
2262 | * poisoning, then it's going to smash the contents of | |
2263 | * the redzone and userword anyhow, so switch them off. | |
2264 | */ | |
2265 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2266 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2267 | #endif | |
1da177e4 LT |
2268 | } |
2269 | ||
2270 | cachep->colour_off = cache_line_size(); | |
2271 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2272 | if (cachep->colour_off < cachep->align) |
2273 | cachep->colour_off = cachep->align; | |
b28a02de | 2274 | cachep->colour = left_over / cachep->colour_off; |
8456a648 | 2275 | cachep->freelist_size = freelist_size; |
1da177e4 | 2276 | cachep->flags = flags; |
a57a4988 | 2277 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2278 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2279 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2280 | cachep->size = size; |
6a2d7a95 | 2281 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2282 | |
e5ac9c5a | 2283 | if (flags & CFLGS_OFF_SLAB) { |
8456a648 | 2284 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
e5ac9c5a | 2285 | /* |
5f0985bb | 2286 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
e5ac9c5a | 2287 | * But since we go off slab only for object size greater than |
5f0985bb JZ |
2288 | * PAGE_SIZE/8, and kmalloc_{dma,}_caches get created |
2289 | * in ascending order,this should not happen at all. | |
e5ac9c5a RT |
2290 | * But leave a BUG_ON for some lucky dude. |
2291 | */ | |
8456a648 | 2292 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
e5ac9c5a | 2293 | } |
1da177e4 | 2294 | |
278b1bb1 CL |
2295 | err = setup_cpu_cache(cachep, gfp); |
2296 | if (err) { | |
12c3667f | 2297 | __kmem_cache_shutdown(cachep); |
278b1bb1 | 2298 | return err; |
2ed3a4ef | 2299 | } |
1da177e4 | 2300 | |
278b1bb1 | 2301 | return 0; |
1da177e4 | 2302 | } |
1da177e4 LT |
2303 | |
2304 | #if DEBUG | |
2305 | static void check_irq_off(void) | |
2306 | { | |
2307 | BUG_ON(!irqs_disabled()); | |
2308 | } | |
2309 | ||
2310 | static void check_irq_on(void) | |
2311 | { | |
2312 | BUG_ON(irqs_disabled()); | |
2313 | } | |
2314 | ||
343e0d7a | 2315 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2316 | { |
2317 | #ifdef CONFIG_SMP | |
2318 | check_irq_off(); | |
18bf8541 | 2319 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2320 | #endif |
2321 | } | |
e498be7d | 2322 | |
343e0d7a | 2323 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2324 | { |
2325 | #ifdef CONFIG_SMP | |
2326 | check_irq_off(); | |
18bf8541 | 2327 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2328 | #endif |
2329 | } | |
2330 | ||
1da177e4 LT |
2331 | #else |
2332 | #define check_irq_off() do { } while(0) | |
2333 | #define check_irq_on() do { } while(0) | |
2334 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2335 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2336 | #endif |
2337 | ||
ce8eb6c4 | 2338 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2339 | struct array_cache *ac, |
2340 | int force, int node); | |
2341 | ||
1da177e4 LT |
2342 | static void do_drain(void *arg) |
2343 | { | |
a737b3e2 | 2344 | struct kmem_cache *cachep = arg; |
1da177e4 | 2345 | struct array_cache *ac; |
7d6e6d09 | 2346 | int node = numa_mem_id(); |
18bf8541 | 2347 | struct kmem_cache_node *n; |
97654dfa | 2348 | LIST_HEAD(list); |
1da177e4 LT |
2349 | |
2350 | check_irq_off(); | |
9a2dba4b | 2351 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2352 | n = get_node(cachep, node); |
2353 | spin_lock(&n->list_lock); | |
97654dfa | 2354 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2355 | spin_unlock(&n->list_lock); |
97654dfa | 2356 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2357 | ac->avail = 0; |
2358 | } | |
2359 | ||
343e0d7a | 2360 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2361 | { |
ce8eb6c4 | 2362 | struct kmem_cache_node *n; |
e498be7d CL |
2363 | int node; |
2364 | ||
15c8b6c1 | 2365 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2366 | check_irq_on(); |
18bf8541 CL |
2367 | for_each_kmem_cache_node(cachep, node, n) |
2368 | if (n->alien) | |
ce8eb6c4 | 2369 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2370 | |
18bf8541 CL |
2371 | for_each_kmem_cache_node(cachep, node, n) |
2372 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2373 | } |
2374 | ||
ed11d9eb CL |
2375 | /* |
2376 | * Remove slabs from the list of free slabs. | |
2377 | * Specify the number of slabs to drain in tofree. | |
2378 | * | |
2379 | * Returns the actual number of slabs released. | |
2380 | */ | |
2381 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2382 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2383 | { |
ed11d9eb CL |
2384 | struct list_head *p; |
2385 | int nr_freed; | |
8456a648 | 2386 | struct page *page; |
1da177e4 | 2387 | |
ed11d9eb | 2388 | nr_freed = 0; |
ce8eb6c4 | 2389 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2390 | |
ce8eb6c4 CL |
2391 | spin_lock_irq(&n->list_lock); |
2392 | p = n->slabs_free.prev; | |
2393 | if (p == &n->slabs_free) { | |
2394 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2395 | goto out; |
2396 | } | |
1da177e4 | 2397 | |
8456a648 | 2398 | page = list_entry(p, struct page, lru); |
1da177e4 | 2399 | #if DEBUG |
8456a648 | 2400 | BUG_ON(page->active); |
1da177e4 | 2401 | #endif |
8456a648 | 2402 | list_del(&page->lru); |
ed11d9eb CL |
2403 | /* |
2404 | * Safe to drop the lock. The slab is no longer linked | |
2405 | * to the cache. | |
2406 | */ | |
ce8eb6c4 CL |
2407 | n->free_objects -= cache->num; |
2408 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2409 | slab_destroy(cache, page); |
ed11d9eb | 2410 | nr_freed++; |
1da177e4 | 2411 | } |
ed11d9eb CL |
2412 | out: |
2413 | return nr_freed; | |
1da177e4 LT |
2414 | } |
2415 | ||
03afc0e2 | 2416 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d | 2417 | { |
18bf8541 CL |
2418 | int ret = 0; |
2419 | int node; | |
ce8eb6c4 | 2420 | struct kmem_cache_node *n; |
e498be7d CL |
2421 | |
2422 | drain_cpu_caches(cachep); | |
2423 | ||
2424 | check_irq_on(); | |
18bf8541 | 2425 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2426 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2427 | |
ce8eb6c4 CL |
2428 | ret += !list_empty(&n->slabs_full) || |
2429 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2430 | } |
2431 | return (ret ? 1 : 0); | |
2432 | } | |
2433 | ||
945cf2b6 | 2434 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
1da177e4 | 2435 | { |
12c3667f | 2436 | int i; |
ce8eb6c4 | 2437 | struct kmem_cache_node *n; |
03afc0e2 | 2438 | int rc = __kmem_cache_shrink(cachep); |
1da177e4 | 2439 | |
12c3667f CL |
2440 | if (rc) |
2441 | return rc; | |
1da177e4 | 2442 | |
12c3667f CL |
2443 | for_each_online_cpu(i) |
2444 | kfree(cachep->array[i]); | |
1da177e4 | 2445 | |
ce8eb6c4 | 2446 | /* NUMA: free the node structures */ |
18bf8541 CL |
2447 | for_each_kmem_cache_node(cachep, i, n) { |
2448 | kfree(n->shared); | |
2449 | free_alien_cache(n->alien); | |
2450 | kfree(n); | |
2451 | cachep->node[i] = NULL; | |
12c3667f CL |
2452 | } |
2453 | return 0; | |
1da177e4 | 2454 | } |
1da177e4 | 2455 | |
e5ac9c5a RT |
2456 | /* |
2457 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2458 | * |
2459 | * For a slab cache when the slab descriptor is off-slab, the | |
2460 | * slab descriptor can't come from the same cache which is being created, | |
2461 | * Because if it is the case, that means we defer the creation of | |
2462 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2463 | * And we eventually call down to __kmem_cache_create(), which | |
2464 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2465 | * This is a "chicken-and-egg" problem. | |
2466 | * | |
2467 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2468 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2469 | */ |
7e007355 | 2470 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2471 | struct page *page, int colour_off, |
2472 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2473 | { |
7e007355 | 2474 | void *freelist; |
0c3aa83e | 2475 | void *addr = page_address(page); |
b28a02de | 2476 | |
1da177e4 LT |
2477 | if (OFF_SLAB(cachep)) { |
2478 | /* Slab management obj is off-slab. */ | |
8456a648 | 2479 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2480 | local_flags, nodeid); |
8456a648 | 2481 | if (!freelist) |
1da177e4 LT |
2482 | return NULL; |
2483 | } else { | |
8456a648 JK |
2484 | freelist = addr + colour_off; |
2485 | colour_off += cachep->freelist_size; | |
1da177e4 | 2486 | } |
8456a648 JK |
2487 | page->active = 0; |
2488 | page->s_mem = addr + colour_off; | |
2489 | return freelist; | |
1da177e4 LT |
2490 | } |
2491 | ||
7cc68973 | 2492 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2493 | { |
a41adfaa | 2494 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2495 | } |
2496 | ||
2497 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2498 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2499 | { |
a41adfaa | 2500 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2501 | } |
2502 | ||
343e0d7a | 2503 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2504 | struct page *page) |
1da177e4 LT |
2505 | { |
2506 | int i; | |
2507 | ||
2508 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2509 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
2510 | #if DEBUG |
2511 | /* need to poison the objs? */ | |
2512 | if (cachep->flags & SLAB_POISON) | |
2513 | poison_obj(cachep, objp, POISON_FREE); | |
2514 | if (cachep->flags & SLAB_STORE_USER) | |
2515 | *dbg_userword(cachep, objp) = NULL; | |
2516 | ||
2517 | if (cachep->flags & SLAB_RED_ZONE) { | |
2518 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2519 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2520 | } | |
2521 | /* | |
a737b3e2 AM |
2522 | * Constructors are not allowed to allocate memory from the same |
2523 | * cache which they are a constructor for. Otherwise, deadlock. | |
2524 | * They must also be threaded. | |
1da177e4 LT |
2525 | */ |
2526 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2527 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2528 | |
2529 | if (cachep->flags & SLAB_RED_ZONE) { | |
2530 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2531 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2532 | " end of an object"); |
1da177e4 LT |
2533 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2534 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2535 | " start of an object"); |
1da177e4 | 2536 | } |
3b0efdfa | 2537 | if ((cachep->size % PAGE_SIZE) == 0 && |
a737b3e2 | 2538 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
b28a02de | 2539 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2540 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2541 | #else |
2542 | if (cachep->ctor) | |
51cc5068 | 2543 | cachep->ctor(objp); |
1da177e4 | 2544 | #endif |
03787301 | 2545 | set_obj_status(page, i, OBJECT_FREE); |
e5c58dfd | 2546 | set_free_obj(page, i, i); |
1da177e4 | 2547 | } |
1da177e4 LT |
2548 | } |
2549 | ||
343e0d7a | 2550 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2551 | { |
4b51d669 CL |
2552 | if (CONFIG_ZONE_DMA_FLAG) { |
2553 | if (flags & GFP_DMA) | |
a618e89f | 2554 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2555 | else |
a618e89f | 2556 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2557 | } |
1da177e4 LT |
2558 | } |
2559 | ||
8456a648 | 2560 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2561 | int nodeid) |
78d382d7 | 2562 | { |
b1cb0982 | 2563 | void *objp; |
78d382d7 | 2564 | |
e5c58dfd | 2565 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2566 | page->active++; |
78d382d7 | 2567 | #if DEBUG |
1ea991b0 | 2568 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2569 | #endif |
78d382d7 MD |
2570 | |
2571 | return objp; | |
2572 | } | |
2573 | ||
8456a648 | 2574 | static void slab_put_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2575 | void *objp, int nodeid) |
78d382d7 | 2576 | { |
8456a648 | 2577 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2578 | #if DEBUG |
16025177 | 2579 | unsigned int i; |
b1cb0982 | 2580 | |
78d382d7 | 2581 | /* Verify that the slab belongs to the intended node */ |
1ea991b0 | 2582 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2583 | |
b1cb0982 | 2584 | /* Verify double free bug */ |
8456a648 | 2585 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2586 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2587 | printk(KERN_ERR "slab: double free detected in cache " |
2588 | "'%s', objp %p\n", cachep->name, objp); | |
2589 | BUG(); | |
2590 | } | |
78d382d7 MD |
2591 | } |
2592 | #endif | |
8456a648 | 2593 | page->active--; |
e5c58dfd | 2594 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2595 | } |
2596 | ||
4776874f PE |
2597 | /* |
2598 | * Map pages beginning at addr to the given cache and slab. This is required | |
2599 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2600 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2601 | */ |
8456a648 | 2602 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2603 | void *freelist) |
1da177e4 | 2604 | { |
a57a4988 | 2605 | page->slab_cache = cache; |
8456a648 | 2606 | page->freelist = freelist; |
1da177e4 LT |
2607 | } |
2608 | ||
2609 | /* | |
2610 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2611 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2612 | */ | |
3c517a61 | 2613 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2614 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2615 | { |
7e007355 | 2616 | void *freelist; |
b28a02de PE |
2617 | size_t offset; |
2618 | gfp_t local_flags; | |
ce8eb6c4 | 2619 | struct kmem_cache_node *n; |
1da177e4 | 2620 | |
a737b3e2 AM |
2621 | /* |
2622 | * Be lazy and only check for valid flags here, keeping it out of the | |
2623 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2624 | */ |
6cb06229 CL |
2625 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2626 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2627 | |
ce8eb6c4 | 2628 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2629 | check_irq_off(); |
18bf8541 | 2630 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2631 | spin_lock(&n->list_lock); |
1da177e4 LT |
2632 | |
2633 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2634 | offset = n->colour_next; |
2635 | n->colour_next++; | |
2636 | if (n->colour_next >= cachep->colour) | |
2637 | n->colour_next = 0; | |
2638 | spin_unlock(&n->list_lock); | |
1da177e4 | 2639 | |
2e1217cf | 2640 | offset *= cachep->colour_off; |
1da177e4 LT |
2641 | |
2642 | if (local_flags & __GFP_WAIT) | |
2643 | local_irq_enable(); | |
2644 | ||
2645 | /* | |
2646 | * The test for missing atomic flag is performed here, rather than | |
2647 | * the more obvious place, simply to reduce the critical path length | |
2648 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2649 | * will eventually be caught here (where it matters). | |
2650 | */ | |
2651 | kmem_flagcheck(cachep, flags); | |
2652 | ||
a737b3e2 AM |
2653 | /* |
2654 | * Get mem for the objs. Attempt to allocate a physical page from | |
2655 | * 'nodeid'. | |
e498be7d | 2656 | */ |
0c3aa83e JK |
2657 | if (!page) |
2658 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2659 | if (!page) | |
1da177e4 LT |
2660 | goto failed; |
2661 | ||
2662 | /* Get slab management. */ | |
8456a648 | 2663 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2664 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2665 | if (!freelist) |
1da177e4 LT |
2666 | goto opps1; |
2667 | ||
8456a648 | 2668 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2669 | |
8456a648 | 2670 | cache_init_objs(cachep, page); |
1da177e4 LT |
2671 | |
2672 | if (local_flags & __GFP_WAIT) | |
2673 | local_irq_disable(); | |
2674 | check_irq_off(); | |
ce8eb6c4 | 2675 | spin_lock(&n->list_lock); |
1da177e4 LT |
2676 | |
2677 | /* Make slab active. */ | |
8456a648 | 2678 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2679 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2680 | n->free_objects += cachep->num; |
2681 | spin_unlock(&n->list_lock); | |
1da177e4 | 2682 | return 1; |
a737b3e2 | 2683 | opps1: |
0c3aa83e | 2684 | kmem_freepages(cachep, page); |
a737b3e2 | 2685 | failed: |
1da177e4 LT |
2686 | if (local_flags & __GFP_WAIT) |
2687 | local_irq_disable(); | |
2688 | return 0; | |
2689 | } | |
2690 | ||
2691 | #if DEBUG | |
2692 | ||
2693 | /* | |
2694 | * Perform extra freeing checks: | |
2695 | * - detect bad pointers. | |
2696 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2697 | */ |
2698 | static void kfree_debugcheck(const void *objp) | |
2699 | { | |
1da177e4 LT |
2700 | if (!virt_addr_valid(objp)) { |
2701 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2702 | (unsigned long)objp); |
2703 | BUG(); | |
1da177e4 | 2704 | } |
1da177e4 LT |
2705 | } |
2706 | ||
58ce1fd5 PE |
2707 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2708 | { | |
b46b8f19 | 2709 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2710 | |
2711 | redzone1 = *dbg_redzone1(cache, obj); | |
2712 | redzone2 = *dbg_redzone2(cache, obj); | |
2713 | ||
2714 | /* | |
2715 | * Redzone is ok. | |
2716 | */ | |
2717 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2718 | return; | |
2719 | ||
2720 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2721 | slab_error(cache, "double free detected"); | |
2722 | else | |
2723 | slab_error(cache, "memory outside object was overwritten"); | |
2724 | ||
b46b8f19 | 2725 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2726 | obj, redzone1, redzone2); |
2727 | } | |
2728 | ||
343e0d7a | 2729 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2730 | unsigned long caller) |
1da177e4 | 2731 | { |
1da177e4 | 2732 | unsigned int objnr; |
8456a648 | 2733 | struct page *page; |
1da177e4 | 2734 | |
80cbd911 MW |
2735 | BUG_ON(virt_to_cache(objp) != cachep); |
2736 | ||
3dafccf2 | 2737 | objp -= obj_offset(cachep); |
1da177e4 | 2738 | kfree_debugcheck(objp); |
b49af68f | 2739 | page = virt_to_head_page(objp); |
1da177e4 | 2740 | |
1da177e4 | 2741 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2742 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2743 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2744 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2745 | } | |
2746 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2747 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2748 | |
8456a648 | 2749 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2750 | |
2751 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2752 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2753 | |
03787301 | 2754 | set_obj_status(page, objnr, OBJECT_FREE); |
1da177e4 LT |
2755 | if (cachep->flags & SLAB_POISON) { |
2756 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2757 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
7c0cb9c6 | 2758 | store_stackinfo(cachep, objp, caller); |
b28a02de | 2759 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2760 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2761 | } else { |
2762 | poison_obj(cachep, objp, POISON_FREE); | |
2763 | } | |
2764 | #else | |
2765 | poison_obj(cachep, objp, POISON_FREE); | |
2766 | #endif | |
2767 | } | |
2768 | return objp; | |
2769 | } | |
2770 | ||
1da177e4 LT |
2771 | #else |
2772 | #define kfree_debugcheck(x) do { } while(0) | |
2773 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2774 | #endif |
2775 | ||
072bb0aa MG |
2776 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2777 | bool force_refill) | |
1da177e4 LT |
2778 | { |
2779 | int batchcount; | |
ce8eb6c4 | 2780 | struct kmem_cache_node *n; |
1da177e4 | 2781 | struct array_cache *ac; |
1ca4cb24 PE |
2782 | int node; |
2783 | ||
1da177e4 | 2784 | check_irq_off(); |
7d6e6d09 | 2785 | node = numa_mem_id(); |
072bb0aa MG |
2786 | if (unlikely(force_refill)) |
2787 | goto force_grow; | |
2788 | retry: | |
9a2dba4b | 2789 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2790 | batchcount = ac->batchcount; |
2791 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2792 | /* |
2793 | * If there was little recent activity on this cache, then | |
2794 | * perform only a partial refill. Otherwise we could generate | |
2795 | * refill bouncing. | |
1da177e4 LT |
2796 | */ |
2797 | batchcount = BATCHREFILL_LIMIT; | |
2798 | } | |
18bf8541 | 2799 | n = get_node(cachep, node); |
e498be7d | 2800 | |
ce8eb6c4 CL |
2801 | BUG_ON(ac->avail > 0 || !n); |
2802 | spin_lock(&n->list_lock); | |
1da177e4 | 2803 | |
3ded175a | 2804 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2805 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2806 | n->shared->touched = 1; | |
3ded175a | 2807 | goto alloc_done; |
44b57f1c | 2808 | } |
3ded175a | 2809 | |
1da177e4 LT |
2810 | while (batchcount > 0) { |
2811 | struct list_head *entry; | |
8456a648 | 2812 | struct page *page; |
1da177e4 | 2813 | /* Get slab alloc is to come from. */ |
ce8eb6c4 CL |
2814 | entry = n->slabs_partial.next; |
2815 | if (entry == &n->slabs_partial) { | |
2816 | n->free_touched = 1; | |
2817 | entry = n->slabs_free.next; | |
2818 | if (entry == &n->slabs_free) | |
1da177e4 LT |
2819 | goto must_grow; |
2820 | } | |
2821 | ||
8456a648 | 2822 | page = list_entry(entry, struct page, lru); |
1da177e4 | 2823 | check_spinlock_acquired(cachep); |
714b8171 PE |
2824 | |
2825 | /* | |
2826 | * The slab was either on partial or free list so | |
2827 | * there must be at least one object available for | |
2828 | * allocation. | |
2829 | */ | |
8456a648 | 2830 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2831 | |
8456a648 | 2832 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2833 | STATS_INC_ALLOCED(cachep); |
2834 | STATS_INC_ACTIVE(cachep); | |
2835 | STATS_SET_HIGH(cachep); | |
2836 | ||
8456a648 | 2837 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page, |
072bb0aa | 2838 | node)); |
1da177e4 | 2839 | } |
1da177e4 LT |
2840 | |
2841 | /* move slabp to correct slabp list: */ | |
8456a648 JK |
2842 | list_del(&page->lru); |
2843 | if (page->active == cachep->num) | |
34bf6ef9 | 2844 | list_add(&page->lru, &n->slabs_full); |
1da177e4 | 2845 | else |
34bf6ef9 | 2846 | list_add(&page->lru, &n->slabs_partial); |
1da177e4 LT |
2847 | } |
2848 | ||
a737b3e2 | 2849 | must_grow: |
ce8eb6c4 | 2850 | n->free_objects -= ac->avail; |
a737b3e2 | 2851 | alloc_done: |
ce8eb6c4 | 2852 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2853 | |
2854 | if (unlikely(!ac->avail)) { | |
2855 | int x; | |
072bb0aa | 2856 | force_grow: |
3c517a61 | 2857 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 2858 | |
a737b3e2 | 2859 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2860 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2861 | node = numa_mem_id(); |
072bb0aa MG |
2862 | |
2863 | /* no objects in sight? abort */ | |
2864 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2865 | return NULL; |
2866 | ||
a737b3e2 | 2867 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2868 | goto retry; |
2869 | } | |
2870 | ac->touched = 1; | |
072bb0aa MG |
2871 | |
2872 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2873 | } |
2874 | ||
a737b3e2 AM |
2875 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2876 | gfp_t flags) | |
1da177e4 LT |
2877 | { |
2878 | might_sleep_if(flags & __GFP_WAIT); | |
2879 | #if DEBUG | |
2880 | kmem_flagcheck(cachep, flags); | |
2881 | #endif | |
2882 | } | |
2883 | ||
2884 | #if DEBUG | |
a737b3e2 | 2885 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2886 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2887 | { |
03787301 JK |
2888 | struct page *page; |
2889 | ||
b28a02de | 2890 | if (!objp) |
1da177e4 | 2891 | return objp; |
b28a02de | 2892 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2893 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3b0efdfa | 2894 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 2895 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2896 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
2897 | else |
2898 | check_poison_obj(cachep, objp); | |
2899 | #else | |
2900 | check_poison_obj(cachep, objp); | |
2901 | #endif | |
2902 | poison_obj(cachep, objp, POISON_INUSE); | |
2903 | } | |
2904 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2905 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2906 | |
2907 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2908 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2909 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2910 | slab_error(cachep, "double free, or memory outside" | |
2911 | " object was overwritten"); | |
b28a02de | 2912 | printk(KERN_ERR |
b46b8f19 | 2913 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2914 | objp, *dbg_redzone1(cachep, objp), |
2915 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2916 | } |
2917 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2918 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2919 | } | |
03787301 JK |
2920 | |
2921 | page = virt_to_head_page(objp); | |
2922 | set_obj_status(page, obj_to_index(cachep, page, objp), OBJECT_ACTIVE); | |
3dafccf2 | 2923 | objp += obj_offset(cachep); |
4f104934 | 2924 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2925 | cachep->ctor(objp); |
7ea466f2 TH |
2926 | if (ARCH_SLAB_MINALIGN && |
2927 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 2928 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2929 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2930 | } |
1da177e4 LT |
2931 | return objp; |
2932 | } | |
2933 | #else | |
2934 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2935 | #endif | |
2936 | ||
773ff60e | 2937 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 | 2938 | { |
8a9c61d4 | 2939 | if (unlikely(cachep == kmem_cache)) |
773ff60e | 2940 | return false; |
8a8b6502 | 2941 | |
8c138bc0 | 2942 | return should_failslab(cachep->object_size, flags, cachep->flags); |
8a8b6502 AM |
2943 | } |
2944 | ||
343e0d7a | 2945 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2946 | { |
b28a02de | 2947 | void *objp; |
1da177e4 | 2948 | struct array_cache *ac; |
072bb0aa | 2949 | bool force_refill = false; |
1da177e4 | 2950 | |
5c382300 | 2951 | check_irq_off(); |
8a8b6502 | 2952 | |
9a2dba4b | 2953 | ac = cpu_cache_get(cachep); |
1da177e4 | 2954 | if (likely(ac->avail)) { |
1da177e4 | 2955 | ac->touched = 1; |
072bb0aa MG |
2956 | objp = ac_get_obj(cachep, ac, flags, false); |
2957 | ||
ddbf2e83 | 2958 | /* |
072bb0aa MG |
2959 | * Allow for the possibility all avail objects are not allowed |
2960 | * by the current flags | |
ddbf2e83 | 2961 | */ |
072bb0aa MG |
2962 | if (objp) { |
2963 | STATS_INC_ALLOCHIT(cachep); | |
2964 | goto out; | |
2965 | } | |
2966 | force_refill = true; | |
1da177e4 | 2967 | } |
072bb0aa MG |
2968 | |
2969 | STATS_INC_ALLOCMISS(cachep); | |
2970 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
2971 | /* | |
2972 | * the 'ac' may be updated by cache_alloc_refill(), | |
2973 | * and kmemleak_erase() requires its correct value. | |
2974 | */ | |
2975 | ac = cpu_cache_get(cachep); | |
2976 | ||
2977 | out: | |
d5cff635 CM |
2978 | /* |
2979 | * To avoid a false negative, if an object that is in one of the | |
2980 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2981 | * treat the array pointers as a reference to the object. | |
2982 | */ | |
f3d8b53a O |
2983 | if (objp) |
2984 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2985 | return objp; |
2986 | } | |
2987 | ||
e498be7d | 2988 | #ifdef CONFIG_NUMA |
c61afb18 | 2989 | /* |
2ad654bc | 2990 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2991 | * |
2992 | * If we are in_interrupt, then process context, including cpusets and | |
2993 | * mempolicy, may not apply and should not be used for allocation policy. | |
2994 | */ | |
2995 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
2996 | { | |
2997 | int nid_alloc, nid_here; | |
2998 | ||
765c4507 | 2999 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3000 | return NULL; |
7d6e6d09 | 3001 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3002 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3003 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3004 | else if (current->mempolicy) |
2a389610 | 3005 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3006 | if (nid_alloc != nid_here) |
8b98c169 | 3007 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3008 | return NULL; |
3009 | } | |
3010 | ||
765c4507 CL |
3011 | /* |
3012 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3013 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3014 | * available node for available objects. If that fails then we |
3c517a61 CL |
3015 | * perform an allocation without specifying a node. This allows the page |
3016 | * allocator to do its reclaim / fallback magic. We then insert the | |
3017 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3018 | */ |
8c8cc2c1 | 3019 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3020 | { |
8c8cc2c1 PE |
3021 | struct zonelist *zonelist; |
3022 | gfp_t local_flags; | |
dd1a239f | 3023 | struct zoneref *z; |
54a6eb5c MG |
3024 | struct zone *zone; |
3025 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3026 | void *obj = NULL; |
3c517a61 | 3027 | int nid; |
cc9a6c87 | 3028 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3029 | |
3030 | if (flags & __GFP_THISNODE) | |
3031 | return NULL; | |
3032 | ||
6cb06229 | 3033 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3034 | |
cc9a6c87 | 3035 | retry_cpuset: |
d26914d1 | 3036 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3037 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3038 | |
3c517a61 CL |
3039 | retry: |
3040 | /* | |
3041 | * Look through allowed nodes for objects available | |
3042 | * from existing per node queues. | |
3043 | */ | |
54a6eb5c MG |
3044 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3045 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3046 | |
54a6eb5c | 3047 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
18bf8541 CL |
3048 | get_node(cache, nid) && |
3049 | get_node(cache, nid)->free_objects) { | |
3c517a61 CL |
3050 | obj = ____cache_alloc_node(cache, |
3051 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3052 | if (obj) |
3053 | break; | |
3054 | } | |
3c517a61 CL |
3055 | } |
3056 | ||
cfce6604 | 3057 | if (!obj) { |
3c517a61 CL |
3058 | /* |
3059 | * This allocation will be performed within the constraints | |
3060 | * of the current cpuset / memory policy requirements. | |
3061 | * We may trigger various forms of reclaim on the allowed | |
3062 | * set and go into memory reserves if necessary. | |
3063 | */ | |
0c3aa83e JK |
3064 | struct page *page; |
3065 | ||
dd47ea75 CL |
3066 | if (local_flags & __GFP_WAIT) |
3067 | local_irq_enable(); | |
3068 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3069 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
dd47ea75 CL |
3070 | if (local_flags & __GFP_WAIT) |
3071 | local_irq_disable(); | |
0c3aa83e | 3072 | if (page) { |
3c517a61 CL |
3073 | /* |
3074 | * Insert into the appropriate per node queues | |
3075 | */ | |
0c3aa83e JK |
3076 | nid = page_to_nid(page); |
3077 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 CL |
3078 | obj = ____cache_alloc_node(cache, |
3079 | flags | GFP_THISNODE, nid); | |
3080 | if (!obj) | |
3081 | /* | |
3082 | * Another processor may allocate the | |
3083 | * objects in the slab since we are | |
3084 | * not holding any locks. | |
3085 | */ | |
3086 | goto retry; | |
3087 | } else { | |
b6a60451 | 3088 | /* cache_grow already freed obj */ |
3c517a61 CL |
3089 | obj = NULL; |
3090 | } | |
3091 | } | |
aedb0eb1 | 3092 | } |
cc9a6c87 | 3093 | |
d26914d1 | 3094 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3095 | goto retry_cpuset; |
765c4507 CL |
3096 | return obj; |
3097 | } | |
3098 | ||
e498be7d CL |
3099 | /* |
3100 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3101 | */ |
8b98c169 | 3102 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3103 | int nodeid) |
e498be7d CL |
3104 | { |
3105 | struct list_head *entry; | |
8456a648 | 3106 | struct page *page; |
ce8eb6c4 | 3107 | struct kmem_cache_node *n; |
b28a02de | 3108 | void *obj; |
b28a02de PE |
3109 | int x; |
3110 | ||
14e50c6a | 3111 | VM_BUG_ON(nodeid > num_online_nodes()); |
18bf8541 | 3112 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3113 | BUG_ON(!n); |
b28a02de | 3114 | |
a737b3e2 | 3115 | retry: |
ca3b9b91 | 3116 | check_irq_off(); |
ce8eb6c4 CL |
3117 | spin_lock(&n->list_lock); |
3118 | entry = n->slabs_partial.next; | |
3119 | if (entry == &n->slabs_partial) { | |
3120 | n->free_touched = 1; | |
3121 | entry = n->slabs_free.next; | |
3122 | if (entry == &n->slabs_free) | |
b28a02de PE |
3123 | goto must_grow; |
3124 | } | |
3125 | ||
8456a648 | 3126 | page = list_entry(entry, struct page, lru); |
b28a02de | 3127 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3128 | |
3129 | STATS_INC_NODEALLOCS(cachep); | |
3130 | STATS_INC_ACTIVE(cachep); | |
3131 | STATS_SET_HIGH(cachep); | |
3132 | ||
8456a648 | 3133 | BUG_ON(page->active == cachep->num); |
b28a02de | 3134 | |
8456a648 | 3135 | obj = slab_get_obj(cachep, page, nodeid); |
ce8eb6c4 | 3136 | n->free_objects--; |
b28a02de | 3137 | /* move slabp to correct slabp list: */ |
8456a648 | 3138 | list_del(&page->lru); |
b28a02de | 3139 | |
8456a648 JK |
3140 | if (page->active == cachep->num) |
3141 | list_add(&page->lru, &n->slabs_full); | |
a737b3e2 | 3142 | else |
8456a648 | 3143 | list_add(&page->lru, &n->slabs_partial); |
e498be7d | 3144 | |
ce8eb6c4 | 3145 | spin_unlock(&n->list_lock); |
b28a02de | 3146 | goto done; |
e498be7d | 3147 | |
a737b3e2 | 3148 | must_grow: |
ce8eb6c4 | 3149 | spin_unlock(&n->list_lock); |
3c517a61 | 3150 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3151 | if (x) |
3152 | goto retry; | |
1da177e4 | 3153 | |
8c8cc2c1 | 3154 | return fallback_alloc(cachep, flags); |
e498be7d | 3155 | |
a737b3e2 | 3156 | done: |
b28a02de | 3157 | return obj; |
e498be7d | 3158 | } |
8c8cc2c1 | 3159 | |
8c8cc2c1 | 3160 | static __always_inline void * |
48356303 | 3161 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3162 | unsigned long caller) |
8c8cc2c1 PE |
3163 | { |
3164 | unsigned long save_flags; | |
3165 | void *ptr; | |
7d6e6d09 | 3166 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3167 | |
dcce284a | 3168 | flags &= gfp_allowed_mask; |
7e85ee0c | 3169 | |
cf40bd16 NP |
3170 | lockdep_trace_alloc(flags); |
3171 | ||
773ff60e | 3172 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3173 | return NULL; |
3174 | ||
d79923fa GC |
3175 | cachep = memcg_kmem_get_cache(cachep, flags); |
3176 | ||
8c8cc2c1 PE |
3177 | cache_alloc_debugcheck_before(cachep, flags); |
3178 | local_irq_save(save_flags); | |
3179 | ||
eacbbae3 | 3180 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3181 | nodeid = slab_node; |
8c8cc2c1 | 3182 | |
18bf8541 | 3183 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3184 | /* Node not bootstrapped yet */ |
3185 | ptr = fallback_alloc(cachep, flags); | |
3186 | goto out; | |
3187 | } | |
3188 | ||
7d6e6d09 | 3189 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3190 | /* |
3191 | * Use the locally cached objects if possible. | |
3192 | * However ____cache_alloc does not allow fallback | |
3193 | * to other nodes. It may fail while we still have | |
3194 | * objects on other nodes available. | |
3195 | */ | |
3196 | ptr = ____cache_alloc(cachep, flags); | |
3197 | if (ptr) | |
3198 | goto out; | |
3199 | } | |
3200 | /* ___cache_alloc_node can fall back to other nodes */ | |
3201 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3202 | out: | |
3203 | local_irq_restore(save_flags); | |
3204 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
8c138bc0 | 3205 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3206 | flags); |
8c8cc2c1 | 3207 | |
5087c822 | 3208 | if (likely(ptr)) { |
8c138bc0 | 3209 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
5087c822 JP |
3210 | if (unlikely(flags & __GFP_ZERO)) |
3211 | memset(ptr, 0, cachep->object_size); | |
3212 | } | |
d07dbea4 | 3213 | |
8c8cc2c1 PE |
3214 | return ptr; |
3215 | } | |
3216 | ||
3217 | static __always_inline void * | |
3218 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3219 | { | |
3220 | void *objp; | |
3221 | ||
2ad654bc | 3222 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3223 | objp = alternate_node_alloc(cache, flags); |
3224 | if (objp) | |
3225 | goto out; | |
3226 | } | |
3227 | objp = ____cache_alloc(cache, flags); | |
3228 | ||
3229 | /* | |
3230 | * We may just have run out of memory on the local node. | |
3231 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3232 | */ | |
7d6e6d09 LS |
3233 | if (!objp) |
3234 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3235 | |
3236 | out: | |
3237 | return objp; | |
3238 | } | |
3239 | #else | |
3240 | ||
3241 | static __always_inline void * | |
3242 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3243 | { | |
3244 | return ____cache_alloc(cachep, flags); | |
3245 | } | |
3246 | ||
3247 | #endif /* CONFIG_NUMA */ | |
3248 | ||
3249 | static __always_inline void * | |
48356303 | 3250 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3251 | { |
3252 | unsigned long save_flags; | |
3253 | void *objp; | |
3254 | ||
dcce284a | 3255 | flags &= gfp_allowed_mask; |
7e85ee0c | 3256 | |
cf40bd16 NP |
3257 | lockdep_trace_alloc(flags); |
3258 | ||
773ff60e | 3259 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3260 | return NULL; |
3261 | ||
d79923fa GC |
3262 | cachep = memcg_kmem_get_cache(cachep, flags); |
3263 | ||
8c8cc2c1 PE |
3264 | cache_alloc_debugcheck_before(cachep, flags); |
3265 | local_irq_save(save_flags); | |
3266 | objp = __do_cache_alloc(cachep, flags); | |
3267 | local_irq_restore(save_flags); | |
3268 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
8c138bc0 | 3269 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3270 | flags); |
8c8cc2c1 PE |
3271 | prefetchw(objp); |
3272 | ||
5087c822 | 3273 | if (likely(objp)) { |
8c138bc0 | 3274 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
5087c822 JP |
3275 | if (unlikely(flags & __GFP_ZERO)) |
3276 | memset(objp, 0, cachep->object_size); | |
3277 | } | |
d07dbea4 | 3278 | |
8c8cc2c1 PE |
3279 | return objp; |
3280 | } | |
e498be7d CL |
3281 | |
3282 | /* | |
5f0985bb | 3283 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3284 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3285 | */ |
97654dfa JK |
3286 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3287 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3288 | { |
3289 | int i; | |
25c063fb | 3290 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3291 | |
3292 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3293 | void *objp; |
8456a648 | 3294 | struct page *page; |
1da177e4 | 3295 | |
072bb0aa MG |
3296 | clear_obj_pfmemalloc(&objpp[i]); |
3297 | objp = objpp[i]; | |
3298 | ||
8456a648 | 3299 | page = virt_to_head_page(objp); |
8456a648 | 3300 | list_del(&page->lru); |
ff69416e | 3301 | check_spinlock_acquired_node(cachep, node); |
8456a648 | 3302 | slab_put_obj(cachep, page, objp, node); |
1da177e4 | 3303 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3304 | n->free_objects++; |
1da177e4 LT |
3305 | |
3306 | /* fixup slab chains */ | |
8456a648 | 3307 | if (page->active == 0) { |
ce8eb6c4 CL |
3308 | if (n->free_objects > n->free_limit) { |
3309 | n->free_objects -= cachep->num; | |
97654dfa | 3310 | list_add_tail(&page->lru, list); |
1da177e4 | 3311 | } else { |
8456a648 | 3312 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3313 | } |
3314 | } else { | |
3315 | /* Unconditionally move a slab to the end of the | |
3316 | * partial list on free - maximum time for the | |
3317 | * other objects to be freed, too. | |
3318 | */ | |
8456a648 | 3319 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3320 | } |
3321 | } | |
3322 | } | |
3323 | ||
343e0d7a | 3324 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3325 | { |
3326 | int batchcount; | |
ce8eb6c4 | 3327 | struct kmem_cache_node *n; |
7d6e6d09 | 3328 | int node = numa_mem_id(); |
97654dfa | 3329 | LIST_HEAD(list); |
1da177e4 LT |
3330 | |
3331 | batchcount = ac->batchcount; | |
3332 | #if DEBUG | |
3333 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3334 | #endif | |
3335 | check_irq_off(); | |
18bf8541 | 3336 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3337 | spin_lock(&n->list_lock); |
3338 | if (n->shared) { | |
3339 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3340 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3341 | if (max) { |
3342 | if (batchcount > max) | |
3343 | batchcount = max; | |
e498be7d | 3344 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3345 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3346 | shared_array->avail += batchcount; |
3347 | goto free_done; | |
3348 | } | |
3349 | } | |
3350 | ||
97654dfa | 3351 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3352 | free_done: |
1da177e4 LT |
3353 | #if STATS |
3354 | { | |
3355 | int i = 0; | |
3356 | struct list_head *p; | |
3357 | ||
ce8eb6c4 CL |
3358 | p = n->slabs_free.next; |
3359 | while (p != &(n->slabs_free)) { | |
8456a648 | 3360 | struct page *page; |
1da177e4 | 3361 | |
8456a648 JK |
3362 | page = list_entry(p, struct page, lru); |
3363 | BUG_ON(page->active); | |
1da177e4 LT |
3364 | |
3365 | i++; | |
3366 | p = p->next; | |
3367 | } | |
3368 | STATS_SET_FREEABLE(cachep, i); | |
3369 | } | |
3370 | #endif | |
ce8eb6c4 | 3371 | spin_unlock(&n->list_lock); |
97654dfa | 3372 | slabs_destroy(cachep, &list); |
1da177e4 | 3373 | ac->avail -= batchcount; |
a737b3e2 | 3374 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3375 | } |
3376 | ||
3377 | /* | |
a737b3e2 AM |
3378 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3379 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3380 | */ |
a947eb95 | 3381 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3382 | unsigned long caller) |
1da177e4 | 3383 | { |
9a2dba4b | 3384 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3385 | |
3386 | check_irq_off(); | |
d5cff635 | 3387 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3388 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3389 | |
8c138bc0 | 3390 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3391 | |
1807a1aa SS |
3392 | /* |
3393 | * Skip calling cache_free_alien() when the platform is not numa. | |
3394 | * This will avoid cache misses that happen while accessing slabp (which | |
3395 | * is per page memory reference) to get nodeid. Instead use a global | |
3396 | * variable to skip the call, which is mostly likely to be present in | |
3397 | * the cache. | |
3398 | */ | |
b6e68bc1 | 3399 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3400 | return; |
3401 | ||
1da177e4 LT |
3402 | if (likely(ac->avail < ac->limit)) { |
3403 | STATS_INC_FREEHIT(cachep); | |
1da177e4 LT |
3404 | } else { |
3405 | STATS_INC_FREEMISS(cachep); | |
3406 | cache_flusharray(cachep, ac); | |
1da177e4 | 3407 | } |
42c8c99c | 3408 | |
072bb0aa | 3409 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3410 | } |
3411 | ||
3412 | /** | |
3413 | * kmem_cache_alloc - Allocate an object | |
3414 | * @cachep: The cache to allocate from. | |
3415 | * @flags: See kmalloc(). | |
3416 | * | |
3417 | * Allocate an object from this cache. The flags are only relevant | |
3418 | * if the cache has no available objects. | |
3419 | */ | |
343e0d7a | 3420 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3421 | { |
48356303 | 3422 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3423 | |
ca2b84cb | 3424 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3425 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3426 | |
3427 | return ret; | |
1da177e4 LT |
3428 | } |
3429 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3430 | ||
0f24f128 | 3431 | #ifdef CONFIG_TRACING |
85beb586 | 3432 | void * |
4052147c | 3433 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3434 | { |
85beb586 SR |
3435 | void *ret; |
3436 | ||
48356303 | 3437 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3438 | |
3439 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3440 | size, cachep->size, flags); |
85beb586 | 3441 | return ret; |
36555751 | 3442 | } |
85beb586 | 3443 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3444 | #endif |
3445 | ||
1da177e4 | 3446 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3447 | /** |
3448 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3449 | * @cachep: The cache to allocate from. | |
3450 | * @flags: See kmalloc(). | |
3451 | * @nodeid: node number of the target node. | |
3452 | * | |
3453 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3454 | * node, which can improve the performance for cpu bound structures. | |
3455 | * | |
3456 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3457 | */ | |
8b98c169 CH |
3458 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3459 | { | |
48356303 | 3460 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3461 | |
ca2b84cb | 3462 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3463 | cachep->object_size, cachep->size, |
ca2b84cb | 3464 | flags, nodeid); |
36555751 EGM |
3465 | |
3466 | return ret; | |
8b98c169 | 3467 | } |
1da177e4 LT |
3468 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3469 | ||
0f24f128 | 3470 | #ifdef CONFIG_TRACING |
4052147c | 3471 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3472 | gfp_t flags, |
4052147c EG |
3473 | int nodeid, |
3474 | size_t size) | |
36555751 | 3475 | { |
85beb586 SR |
3476 | void *ret; |
3477 | ||
592f4145 | 3478 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3479 | |
85beb586 | 3480 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3481 | size, cachep->size, |
85beb586 SR |
3482 | flags, nodeid); |
3483 | return ret; | |
36555751 | 3484 | } |
85beb586 | 3485 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3486 | #endif |
3487 | ||
8b98c169 | 3488 | static __always_inline void * |
7c0cb9c6 | 3489 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3490 | { |
343e0d7a | 3491 | struct kmem_cache *cachep; |
97e2bde4 | 3492 | |
2c59dd65 | 3493 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3494 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3495 | return cachep; | |
4052147c | 3496 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3497 | } |
8b98c169 | 3498 | |
8b98c169 CH |
3499 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3500 | { | |
7c0cb9c6 | 3501 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3502 | } |
dbe5e69d | 3503 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3504 | |
3505 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3506 | int node, unsigned long caller) |
8b98c169 | 3507 | { |
7c0cb9c6 | 3508 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3509 | } |
3510 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3511 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3512 | |
3513 | /** | |
800590f5 | 3514 | * __do_kmalloc - allocate memory |
1da177e4 | 3515 | * @size: how many bytes of memory are required. |
800590f5 | 3516 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3517 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3518 | */ |
7fd6b141 | 3519 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3520 | unsigned long caller) |
1da177e4 | 3521 | { |
343e0d7a | 3522 | struct kmem_cache *cachep; |
36555751 | 3523 | void *ret; |
1da177e4 | 3524 | |
2c59dd65 | 3525 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3526 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3527 | return cachep; | |
48356303 | 3528 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3529 | |
7c0cb9c6 | 3530 | trace_kmalloc(caller, ret, |
3b0efdfa | 3531 | size, cachep->size, flags); |
36555751 EGM |
3532 | |
3533 | return ret; | |
7fd6b141 PE |
3534 | } |
3535 | ||
7fd6b141 PE |
3536 | void *__kmalloc(size_t size, gfp_t flags) |
3537 | { | |
7c0cb9c6 | 3538 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3539 | } |
3540 | EXPORT_SYMBOL(__kmalloc); | |
3541 | ||
ce71e27c | 3542 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3543 | { |
7c0cb9c6 | 3544 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3545 | } |
3546 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3547 | |
1da177e4 LT |
3548 | /** |
3549 | * kmem_cache_free - Deallocate an object | |
3550 | * @cachep: The cache the allocation was from. | |
3551 | * @objp: The previously allocated object. | |
3552 | * | |
3553 | * Free an object which was previously allocated from this | |
3554 | * cache. | |
3555 | */ | |
343e0d7a | 3556 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3557 | { |
3558 | unsigned long flags; | |
b9ce5ef4 GC |
3559 | cachep = cache_from_obj(cachep, objp); |
3560 | if (!cachep) | |
3561 | return; | |
1da177e4 LT |
3562 | |
3563 | local_irq_save(flags); | |
d97d476b | 3564 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3565 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3566 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3567 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3568 | local_irq_restore(flags); |
36555751 | 3569 | |
ca2b84cb | 3570 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3571 | } |
3572 | EXPORT_SYMBOL(kmem_cache_free); | |
3573 | ||
1da177e4 LT |
3574 | /** |
3575 | * kfree - free previously allocated memory | |
3576 | * @objp: pointer returned by kmalloc. | |
3577 | * | |
80e93eff PE |
3578 | * If @objp is NULL, no operation is performed. |
3579 | * | |
1da177e4 LT |
3580 | * Don't free memory not originally allocated by kmalloc() |
3581 | * or you will run into trouble. | |
3582 | */ | |
3583 | void kfree(const void *objp) | |
3584 | { | |
343e0d7a | 3585 | struct kmem_cache *c; |
1da177e4 LT |
3586 | unsigned long flags; |
3587 | ||
2121db74 PE |
3588 | trace_kfree(_RET_IP_, objp); |
3589 | ||
6cb8f913 | 3590 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3591 | return; |
3592 | local_irq_save(flags); | |
3593 | kfree_debugcheck(objp); | |
6ed5eb22 | 3594 | c = virt_to_cache(objp); |
8c138bc0 CL |
3595 | debug_check_no_locks_freed(objp, c->object_size); |
3596 | ||
3597 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3598 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3599 | local_irq_restore(flags); |
3600 | } | |
3601 | EXPORT_SYMBOL(kfree); | |
3602 | ||
e498be7d | 3603 | /* |
ce8eb6c4 | 3604 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3605 | */ |
5f0985bb | 3606 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3607 | { |
3608 | int node; | |
ce8eb6c4 | 3609 | struct kmem_cache_node *n; |
cafeb02e | 3610 | struct array_cache *new_shared; |
c8522a3a | 3611 | struct alien_cache **new_alien = NULL; |
e498be7d | 3612 | |
9c09a95c | 3613 | for_each_online_node(node) { |
cafeb02e | 3614 | |
3395ee05 | 3615 | if (use_alien_caches) { |
83b519e8 | 3616 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3617 | if (!new_alien) |
3618 | goto fail; | |
3619 | } | |
cafeb02e | 3620 | |
63109846 ED |
3621 | new_shared = NULL; |
3622 | if (cachep->shared) { | |
3623 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3624 | cachep->shared*cachep->batchcount, |
83b519e8 | 3625 | 0xbaadf00d, gfp); |
63109846 ED |
3626 | if (!new_shared) { |
3627 | free_alien_cache(new_alien); | |
3628 | goto fail; | |
3629 | } | |
0718dc2a | 3630 | } |
cafeb02e | 3631 | |
18bf8541 | 3632 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3633 | if (n) { |
3634 | struct array_cache *shared = n->shared; | |
97654dfa | 3635 | LIST_HEAD(list); |
cafeb02e | 3636 | |
ce8eb6c4 | 3637 | spin_lock_irq(&n->list_lock); |
e498be7d | 3638 | |
cafeb02e | 3639 | if (shared) |
0718dc2a | 3640 | free_block(cachep, shared->entry, |
97654dfa | 3641 | shared->avail, node, &list); |
e498be7d | 3642 | |
ce8eb6c4 CL |
3643 | n->shared = new_shared; |
3644 | if (!n->alien) { | |
3645 | n->alien = new_alien; | |
e498be7d CL |
3646 | new_alien = NULL; |
3647 | } | |
ce8eb6c4 | 3648 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3649 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3650 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3651 | slabs_destroy(cachep, &list); |
cafeb02e | 3652 | kfree(shared); |
e498be7d CL |
3653 | free_alien_cache(new_alien); |
3654 | continue; | |
3655 | } | |
ce8eb6c4 CL |
3656 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3657 | if (!n) { | |
0718dc2a CL |
3658 | free_alien_cache(new_alien); |
3659 | kfree(new_shared); | |
e498be7d | 3660 | goto fail; |
0718dc2a | 3661 | } |
e498be7d | 3662 | |
ce8eb6c4 | 3663 | kmem_cache_node_init(n); |
5f0985bb JZ |
3664 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3665 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3666 | n->shared = new_shared; |
3667 | n->alien = new_alien; | |
3668 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3669 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3670 | cachep->node[node] = n; |
e498be7d | 3671 | } |
cafeb02e | 3672 | return 0; |
0718dc2a | 3673 | |
a737b3e2 | 3674 | fail: |
3b0efdfa | 3675 | if (!cachep->list.next) { |
0718dc2a CL |
3676 | /* Cache is not active yet. Roll back what we did */ |
3677 | node--; | |
3678 | while (node >= 0) { | |
18bf8541 CL |
3679 | n = get_node(cachep, node); |
3680 | if (n) { | |
ce8eb6c4 CL |
3681 | kfree(n->shared); |
3682 | free_alien_cache(n->alien); | |
3683 | kfree(n); | |
6a67368c | 3684 | cachep->node[node] = NULL; |
0718dc2a CL |
3685 | } |
3686 | node--; | |
3687 | } | |
3688 | } | |
cafeb02e | 3689 | return -ENOMEM; |
e498be7d CL |
3690 | } |
3691 | ||
1da177e4 | 3692 | struct ccupdate_struct { |
343e0d7a | 3693 | struct kmem_cache *cachep; |
acfe7d74 | 3694 | struct array_cache *new[0]; |
1da177e4 LT |
3695 | }; |
3696 | ||
3697 | static void do_ccupdate_local(void *info) | |
3698 | { | |
a737b3e2 | 3699 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3700 | struct array_cache *old; |
3701 | ||
3702 | check_irq_off(); | |
9a2dba4b | 3703 | old = cpu_cache_get(new->cachep); |
e498be7d | 3704 | |
1da177e4 LT |
3705 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3706 | new->new[smp_processor_id()] = old; | |
3707 | } | |
3708 | ||
18004c5d | 3709 | /* Always called with the slab_mutex held */ |
943a451a | 3710 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3711 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3712 | { |
d2e7b7d0 | 3713 | struct ccupdate_struct *new; |
2ed3a4ef | 3714 | int i; |
1da177e4 | 3715 | |
acfe7d74 ED |
3716 | new = kzalloc(sizeof(*new) + nr_cpu_ids * sizeof(struct array_cache *), |
3717 | gfp); | |
d2e7b7d0 SS |
3718 | if (!new) |
3719 | return -ENOMEM; | |
3720 | ||
e498be7d | 3721 | for_each_online_cpu(i) { |
7d6e6d09 | 3722 | new->new[i] = alloc_arraycache(cpu_to_mem(i), limit, |
83b519e8 | 3723 | batchcount, gfp); |
d2e7b7d0 | 3724 | if (!new->new[i]) { |
b28a02de | 3725 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3726 | kfree(new->new[i]); |
3727 | kfree(new); | |
e498be7d | 3728 | return -ENOMEM; |
1da177e4 LT |
3729 | } |
3730 | } | |
d2e7b7d0 | 3731 | new->cachep = cachep; |
1da177e4 | 3732 | |
15c8b6c1 | 3733 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 3734 | |
1da177e4 | 3735 | check_irq_on(); |
1da177e4 LT |
3736 | cachep->batchcount = batchcount; |
3737 | cachep->limit = limit; | |
e498be7d | 3738 | cachep->shared = shared; |
1da177e4 | 3739 | |
e498be7d | 3740 | for_each_online_cpu(i) { |
97654dfa | 3741 | LIST_HEAD(list); |
d2e7b7d0 | 3742 | struct array_cache *ccold = new->new[i]; |
18bf8541 CL |
3743 | int node; |
3744 | struct kmem_cache_node *n; | |
3745 | ||
1da177e4 LT |
3746 | if (!ccold) |
3747 | continue; | |
18bf8541 CL |
3748 | |
3749 | node = cpu_to_mem(i); | |
3750 | n = get_node(cachep, node); | |
3751 | spin_lock_irq(&n->list_lock); | |
97654dfa | 3752 | free_block(cachep, ccold->entry, ccold->avail, node, &list); |
18bf8541 | 3753 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3754 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3755 | kfree(ccold); |
3756 | } | |
d2e7b7d0 | 3757 | kfree(new); |
5f0985bb | 3758 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3759 | } |
3760 | ||
943a451a GC |
3761 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3762 | int batchcount, int shared, gfp_t gfp) | |
3763 | { | |
3764 | int ret; | |
3765 | struct kmem_cache *c = NULL; | |
3766 | int i = 0; | |
3767 | ||
3768 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3769 | ||
3770 | if (slab_state < FULL) | |
3771 | return ret; | |
3772 | ||
3773 | if ((ret < 0) || !is_root_cache(cachep)) | |
3774 | return ret; | |
3775 | ||
ebe945c2 | 3776 | VM_BUG_ON(!mutex_is_locked(&slab_mutex)); |
943a451a | 3777 | for_each_memcg_cache_index(i) { |
2ade4de8 | 3778 | c = cache_from_memcg_idx(cachep, i); |
943a451a GC |
3779 | if (c) |
3780 | /* return value determined by the parent cache only */ | |
3781 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
3782 | } | |
3783 | ||
3784 | return ret; | |
3785 | } | |
3786 | ||
18004c5d | 3787 | /* Called with slab_mutex held always */ |
83b519e8 | 3788 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3789 | { |
3790 | int err; | |
943a451a GC |
3791 | int limit = 0; |
3792 | int shared = 0; | |
3793 | int batchcount = 0; | |
3794 | ||
3795 | if (!is_root_cache(cachep)) { | |
3796 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3797 | limit = root->limit; | |
3798 | shared = root->shared; | |
3799 | batchcount = root->batchcount; | |
3800 | } | |
1da177e4 | 3801 | |
943a451a GC |
3802 | if (limit && shared && batchcount) |
3803 | goto skip_setup; | |
a737b3e2 AM |
3804 | /* |
3805 | * The head array serves three purposes: | |
1da177e4 LT |
3806 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3807 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3808 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3809 | * bufctl chains: array operations are cheaper. |
3810 | * The numbers are guessed, we should auto-tune as described by | |
3811 | * Bonwick. | |
3812 | */ | |
3b0efdfa | 3813 | if (cachep->size > 131072) |
1da177e4 | 3814 | limit = 1; |
3b0efdfa | 3815 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3816 | limit = 8; |
3b0efdfa | 3817 | else if (cachep->size > 1024) |
1da177e4 | 3818 | limit = 24; |
3b0efdfa | 3819 | else if (cachep->size > 256) |
1da177e4 LT |
3820 | limit = 54; |
3821 | else | |
3822 | limit = 120; | |
3823 | ||
a737b3e2 AM |
3824 | /* |
3825 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3826 | * allocation behaviour: Most allocs on one cpu, most free operations |
3827 | * on another cpu. For these cases, an efficient object passing between | |
3828 | * cpus is necessary. This is provided by a shared array. The array | |
3829 | * replaces Bonwick's magazine layer. | |
3830 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3831 | * to a larger limit. Thus disabled by default. | |
3832 | */ | |
3833 | shared = 0; | |
3b0efdfa | 3834 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3835 | shared = 8; |
1da177e4 LT |
3836 | |
3837 | #if DEBUG | |
a737b3e2 AM |
3838 | /* |
3839 | * With debugging enabled, large batchcount lead to excessively long | |
3840 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3841 | */ |
3842 | if (limit > 32) | |
3843 | limit = 32; | |
3844 | #endif | |
943a451a GC |
3845 | batchcount = (limit + 1) / 2; |
3846 | skip_setup: | |
3847 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3848 | if (err) |
3849 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3850 | cachep->name, -err); |
2ed3a4ef | 3851 | return err; |
1da177e4 LT |
3852 | } |
3853 | ||
1b55253a | 3854 | /* |
ce8eb6c4 CL |
3855 | * Drain an array if it contains any elements taking the node lock only if |
3856 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3857 | * if drain_array() is used on the shared array. |
1b55253a | 3858 | */ |
ce8eb6c4 | 3859 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3860 | struct array_cache *ac, int force, int node) |
1da177e4 | 3861 | { |
97654dfa | 3862 | LIST_HEAD(list); |
1da177e4 LT |
3863 | int tofree; |
3864 | ||
1b55253a CL |
3865 | if (!ac || !ac->avail) |
3866 | return; | |
1da177e4 LT |
3867 | if (ac->touched && !force) { |
3868 | ac->touched = 0; | |
b18e7e65 | 3869 | } else { |
ce8eb6c4 | 3870 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3871 | if (ac->avail) { |
3872 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3873 | if (tofree > ac->avail) | |
3874 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 3875 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
3876 | ac->avail -= tofree; |
3877 | memmove(ac->entry, &(ac->entry[tofree]), | |
3878 | sizeof(void *) * ac->avail); | |
3879 | } | |
ce8eb6c4 | 3880 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3881 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3882 | } |
3883 | } | |
3884 | ||
3885 | /** | |
3886 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3887 | * @w: work descriptor |
1da177e4 LT |
3888 | * |
3889 | * Called from workqueue/eventd every few seconds. | |
3890 | * Purpose: | |
3891 | * - clear the per-cpu caches for this CPU. | |
3892 | * - return freeable pages to the main free memory pool. | |
3893 | * | |
a737b3e2 AM |
3894 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3895 | * again on the next iteration. | |
1da177e4 | 3896 | */ |
7c5cae36 | 3897 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3898 | { |
7a7c381d | 3899 | struct kmem_cache *searchp; |
ce8eb6c4 | 3900 | struct kmem_cache_node *n; |
7d6e6d09 | 3901 | int node = numa_mem_id(); |
bf6aede7 | 3902 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3903 | |
18004c5d | 3904 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3905 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3906 | goto out; |
1da177e4 | 3907 | |
18004c5d | 3908 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3909 | check_irq_on(); |
3910 | ||
35386e3b | 3911 | /* |
ce8eb6c4 | 3912 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3913 | * have established with reasonable certainty that |
3914 | * we can do some work if the lock was obtained. | |
3915 | */ | |
18bf8541 | 3916 | n = get_node(searchp, node); |
35386e3b | 3917 | |
ce8eb6c4 | 3918 | reap_alien(searchp, n); |
1da177e4 | 3919 | |
ce8eb6c4 | 3920 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3921 | |
35386e3b CL |
3922 | /* |
3923 | * These are racy checks but it does not matter | |
3924 | * if we skip one check or scan twice. | |
3925 | */ | |
ce8eb6c4 | 3926 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3927 | goto next; |
1da177e4 | 3928 | |
5f0985bb | 3929 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3930 | |
ce8eb6c4 | 3931 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 3932 | |
ce8eb6c4 CL |
3933 | if (n->free_touched) |
3934 | n->free_touched = 0; | |
ed11d9eb CL |
3935 | else { |
3936 | int freed; | |
1da177e4 | 3937 | |
ce8eb6c4 | 3938 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3939 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3940 | STATS_ADD_REAPED(searchp, freed); | |
3941 | } | |
35386e3b | 3942 | next: |
1da177e4 LT |
3943 | cond_resched(); |
3944 | } | |
3945 | check_irq_on(); | |
18004c5d | 3946 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3947 | next_reap_node(); |
7c5cae36 | 3948 | out: |
a737b3e2 | 3949 | /* Set up the next iteration */ |
5f0985bb | 3950 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3951 | } |
3952 | ||
158a9624 | 3953 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3954 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3955 | { |
8456a648 | 3956 | struct page *page; |
b28a02de PE |
3957 | unsigned long active_objs; |
3958 | unsigned long num_objs; | |
3959 | unsigned long active_slabs = 0; | |
3960 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3961 | const char *name; |
1da177e4 | 3962 | char *error = NULL; |
e498be7d | 3963 | int node; |
ce8eb6c4 | 3964 | struct kmem_cache_node *n; |
1da177e4 | 3965 | |
1da177e4 LT |
3966 | active_objs = 0; |
3967 | num_slabs = 0; | |
18bf8541 | 3968 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 3969 | |
ca3b9b91 | 3970 | check_irq_on(); |
ce8eb6c4 | 3971 | spin_lock_irq(&n->list_lock); |
e498be7d | 3972 | |
8456a648 JK |
3973 | list_for_each_entry(page, &n->slabs_full, lru) { |
3974 | if (page->active != cachep->num && !error) | |
e498be7d CL |
3975 | error = "slabs_full accounting error"; |
3976 | active_objs += cachep->num; | |
3977 | active_slabs++; | |
3978 | } | |
8456a648 JK |
3979 | list_for_each_entry(page, &n->slabs_partial, lru) { |
3980 | if (page->active == cachep->num && !error) | |
106a74e1 | 3981 | error = "slabs_partial accounting error"; |
8456a648 | 3982 | if (!page->active && !error) |
106a74e1 | 3983 | error = "slabs_partial accounting error"; |
8456a648 | 3984 | active_objs += page->active; |
e498be7d CL |
3985 | active_slabs++; |
3986 | } | |
8456a648 JK |
3987 | list_for_each_entry(page, &n->slabs_free, lru) { |
3988 | if (page->active && !error) | |
106a74e1 | 3989 | error = "slabs_free accounting error"; |
e498be7d CL |
3990 | num_slabs++; |
3991 | } | |
ce8eb6c4 CL |
3992 | free_objects += n->free_objects; |
3993 | if (n->shared) | |
3994 | shared_avail += n->shared->avail; | |
e498be7d | 3995 | |
ce8eb6c4 | 3996 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 3997 | } |
b28a02de PE |
3998 | num_slabs += active_slabs; |
3999 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4000 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4001 | error = "free_objects accounting error"; |
4002 | ||
b28a02de | 4003 | name = cachep->name; |
1da177e4 LT |
4004 | if (error) |
4005 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4006 | ||
0d7561c6 GC |
4007 | sinfo->active_objs = active_objs; |
4008 | sinfo->num_objs = num_objs; | |
4009 | sinfo->active_slabs = active_slabs; | |
4010 | sinfo->num_slabs = num_slabs; | |
4011 | sinfo->shared_avail = shared_avail; | |
4012 | sinfo->limit = cachep->limit; | |
4013 | sinfo->batchcount = cachep->batchcount; | |
4014 | sinfo->shared = cachep->shared; | |
4015 | sinfo->objects_per_slab = cachep->num; | |
4016 | sinfo->cache_order = cachep->gfporder; | |
4017 | } | |
4018 | ||
4019 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4020 | { | |
1da177e4 | 4021 | #if STATS |
ce8eb6c4 | 4022 | { /* node stats */ |
1da177e4 LT |
4023 | unsigned long high = cachep->high_mark; |
4024 | unsigned long allocs = cachep->num_allocations; | |
4025 | unsigned long grown = cachep->grown; | |
4026 | unsigned long reaped = cachep->reaped; | |
4027 | unsigned long errors = cachep->errors; | |
4028 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4029 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4030 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4031 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4032 | |
e92dd4fd JP |
4033 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4034 | "%4lu %4lu %4lu %4lu %4lu", | |
4035 | allocs, high, grown, | |
4036 | reaped, errors, max_freeable, node_allocs, | |
4037 | node_frees, overflows); | |
1da177e4 LT |
4038 | } |
4039 | /* cpu stats */ | |
4040 | { | |
4041 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4042 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4043 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4044 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4045 | ||
4046 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4047 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4048 | } |
4049 | #endif | |
1da177e4 LT |
4050 | } |
4051 | ||
1da177e4 LT |
4052 | #define MAX_SLABINFO_WRITE 128 |
4053 | /** | |
4054 | * slabinfo_write - Tuning for the slab allocator | |
4055 | * @file: unused | |
4056 | * @buffer: user buffer | |
4057 | * @count: data length | |
4058 | * @ppos: unused | |
4059 | */ | |
b7454ad3 | 4060 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4061 | size_t count, loff_t *ppos) |
1da177e4 | 4062 | { |
b28a02de | 4063 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4064 | int limit, batchcount, shared, res; |
7a7c381d | 4065 | struct kmem_cache *cachep; |
b28a02de | 4066 | |
1da177e4 LT |
4067 | if (count > MAX_SLABINFO_WRITE) |
4068 | return -EINVAL; | |
4069 | if (copy_from_user(&kbuf, buffer, count)) | |
4070 | return -EFAULT; | |
b28a02de | 4071 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4072 | |
4073 | tmp = strchr(kbuf, ' '); | |
4074 | if (!tmp) | |
4075 | return -EINVAL; | |
4076 | *tmp = '\0'; | |
4077 | tmp++; | |
4078 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4079 | return -EINVAL; | |
4080 | ||
4081 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4082 | mutex_lock(&slab_mutex); |
1da177e4 | 4083 | res = -EINVAL; |
18004c5d | 4084 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4085 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4086 | if (limit < 1 || batchcount < 1 || |
4087 | batchcount > limit || shared < 0) { | |
e498be7d | 4088 | res = 0; |
1da177e4 | 4089 | } else { |
e498be7d | 4090 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4091 | batchcount, shared, |
4092 | GFP_KERNEL); | |
1da177e4 LT |
4093 | } |
4094 | break; | |
4095 | } | |
4096 | } | |
18004c5d | 4097 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4098 | if (res >= 0) |
4099 | res = count; | |
4100 | return res; | |
4101 | } | |
871751e2 AV |
4102 | |
4103 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4104 | ||
4105 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4106 | { | |
18004c5d CL |
4107 | mutex_lock(&slab_mutex); |
4108 | return seq_list_start(&slab_caches, *pos); | |
871751e2 AV |
4109 | } |
4110 | ||
4111 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4112 | { | |
4113 | unsigned long *p; | |
4114 | int l; | |
4115 | if (!v) | |
4116 | return 1; | |
4117 | l = n[1]; | |
4118 | p = n + 2; | |
4119 | while (l) { | |
4120 | int i = l/2; | |
4121 | unsigned long *q = p + 2 * i; | |
4122 | if (*q == v) { | |
4123 | q[1]++; | |
4124 | return 1; | |
4125 | } | |
4126 | if (*q > v) { | |
4127 | l = i; | |
4128 | } else { | |
4129 | p = q + 2; | |
4130 | l -= i + 1; | |
4131 | } | |
4132 | } | |
4133 | if (++n[1] == n[0]) | |
4134 | return 0; | |
4135 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4136 | p[0] = v; | |
4137 | p[1] = 1; | |
4138 | return 1; | |
4139 | } | |
4140 | ||
8456a648 JK |
4141 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4142 | struct page *page) | |
871751e2 AV |
4143 | { |
4144 | void *p; | |
03787301 | 4145 | int i; |
b1cb0982 | 4146 | |
871751e2 AV |
4147 | if (n[0] == n[1]) |
4148 | return; | |
8456a648 | 4149 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
03787301 | 4150 | if (get_obj_status(page, i) != OBJECT_ACTIVE) |
871751e2 | 4151 | continue; |
b1cb0982 | 4152 | |
871751e2 AV |
4153 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
4154 | return; | |
4155 | } | |
4156 | } | |
4157 | ||
4158 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4159 | { | |
4160 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4161 | unsigned long offset, size; |
9281acea | 4162 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4163 | |
a5c43dae | 4164 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4165 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4166 | if (modname[0]) |
871751e2 AV |
4167 | seq_printf(m, " [%s]", modname); |
4168 | return; | |
4169 | } | |
4170 | #endif | |
4171 | seq_printf(m, "%p", (void *)address); | |
4172 | } | |
4173 | ||
4174 | static int leaks_show(struct seq_file *m, void *p) | |
4175 | { | |
0672aa7c | 4176 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4177 | struct page *page; |
ce8eb6c4 | 4178 | struct kmem_cache_node *n; |
871751e2 | 4179 | const char *name; |
db845067 | 4180 | unsigned long *x = m->private; |
871751e2 AV |
4181 | int node; |
4182 | int i; | |
4183 | ||
4184 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4185 | return 0; | |
4186 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4187 | return 0; | |
4188 | ||
4189 | /* OK, we can do it */ | |
4190 | ||
db845067 | 4191 | x[1] = 0; |
871751e2 | 4192 | |
18bf8541 | 4193 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 AV |
4194 | |
4195 | check_irq_on(); | |
ce8eb6c4 | 4196 | spin_lock_irq(&n->list_lock); |
871751e2 | 4197 | |
8456a648 JK |
4198 | list_for_each_entry(page, &n->slabs_full, lru) |
4199 | handle_slab(x, cachep, page); | |
4200 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4201 | handle_slab(x, cachep, page); | |
ce8eb6c4 | 4202 | spin_unlock_irq(&n->list_lock); |
871751e2 AV |
4203 | } |
4204 | name = cachep->name; | |
db845067 | 4205 | if (x[0] == x[1]) { |
871751e2 | 4206 | /* Increase the buffer size */ |
18004c5d | 4207 | mutex_unlock(&slab_mutex); |
db845067 | 4208 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4209 | if (!m->private) { |
4210 | /* Too bad, we are really out */ | |
db845067 | 4211 | m->private = x; |
18004c5d | 4212 | mutex_lock(&slab_mutex); |
871751e2 AV |
4213 | return -ENOMEM; |
4214 | } | |
db845067 CL |
4215 | *(unsigned long *)m->private = x[0] * 2; |
4216 | kfree(x); | |
18004c5d | 4217 | mutex_lock(&slab_mutex); |
871751e2 AV |
4218 | /* Now make sure this entry will be retried */ |
4219 | m->count = m->size; | |
4220 | return 0; | |
4221 | } | |
db845067 CL |
4222 | for (i = 0; i < x[1]; i++) { |
4223 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4224 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4225 | seq_putc(m, '\n'); |
4226 | } | |
d2e7b7d0 | 4227 | |
871751e2 AV |
4228 | return 0; |
4229 | } | |
4230 | ||
a0ec95a8 | 4231 | static const struct seq_operations slabstats_op = { |
871751e2 | 4232 | .start = leaks_start, |
276a2439 WL |
4233 | .next = slab_next, |
4234 | .stop = slab_stop, | |
871751e2 AV |
4235 | .show = leaks_show, |
4236 | }; | |
a0ec95a8 AD |
4237 | |
4238 | static int slabstats_open(struct inode *inode, struct file *file) | |
4239 | { | |
4240 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4241 | int ret = -ENOMEM; | |
4242 | if (n) { | |
4243 | ret = seq_open(file, &slabstats_op); | |
4244 | if (!ret) { | |
4245 | struct seq_file *m = file->private_data; | |
4246 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4247 | m->private = n; | |
4248 | n = NULL; | |
4249 | } | |
4250 | kfree(n); | |
4251 | } | |
4252 | return ret; | |
4253 | } | |
4254 | ||
4255 | static const struct file_operations proc_slabstats_operations = { | |
4256 | .open = slabstats_open, | |
4257 | .read = seq_read, | |
4258 | .llseek = seq_lseek, | |
4259 | .release = seq_release_private, | |
4260 | }; | |
4261 | #endif | |
4262 | ||
4263 | static int __init slab_proc_init(void) | |
4264 | { | |
4265 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4266 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4267 | #endif |
a0ec95a8 AD |
4268 | return 0; |
4269 | } | |
4270 | module_init(slab_proc_init); | |
1da177e4 LT |
4271 | #endif |
4272 | ||
00e145b6 MS |
4273 | /** |
4274 | * ksize - get the actual amount of memory allocated for a given object | |
4275 | * @objp: Pointer to the object | |
4276 | * | |
4277 | * kmalloc may internally round up allocations and return more memory | |
4278 | * than requested. ksize() can be used to determine the actual amount of | |
4279 | * memory allocated. The caller may use this additional memory, even though | |
4280 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4281 | * The caller must guarantee that objp points to a valid object previously | |
4282 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4283 | * must not be freed during the duration of the call. | |
4284 | */ | |
fd76bab2 | 4285 | size_t ksize(const void *objp) |
1da177e4 | 4286 | { |
ef8b4520 CL |
4287 | BUG_ON(!objp); |
4288 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4289 | return 0; |
1da177e4 | 4290 | |
8c138bc0 | 4291 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4292 | } |
b1aabecd | 4293 | EXPORT_SYMBOL(ksize); |