]>
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 | */ | |
bf0dea23 | 240 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 241 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 242 | #define CACHE_CACHE 0 |
bf0dea23 | 243 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 244 | |
ed11d9eb | 245 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 246 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 247 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
248 | int node, struct list_head *list); |
249 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 250 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 251 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 252 | |
e0a42726 IM |
253 | static int slab_early_init = 1; |
254 | ||
ce8eb6c4 | 255 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 256 | |
ce8eb6c4 | 257 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
258 | { |
259 | INIT_LIST_HEAD(&parent->slabs_full); | |
260 | INIT_LIST_HEAD(&parent->slabs_partial); | |
261 | INIT_LIST_HEAD(&parent->slabs_free); | |
262 | parent->shared = NULL; | |
263 | parent->alien = NULL; | |
2e1217cf | 264 | parent->colour_next = 0; |
e498be7d CL |
265 | spin_lock_init(&parent->list_lock); |
266 | parent->free_objects = 0; | |
267 | parent->free_touched = 0; | |
268 | } | |
269 | ||
a737b3e2 AM |
270 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
271 | do { \ | |
272 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 273 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
274 | } while (0) |
275 | ||
a737b3e2 AM |
276 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
277 | do { \ | |
e498be7d CL |
278 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
279 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
280 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
281 | } while (0) | |
1da177e4 | 282 | |
1da177e4 LT |
283 | #define CFLGS_OFF_SLAB (0x80000000UL) |
284 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
d4322d88 | 285 | #define OFF_SLAB_MIN_SIZE (max_t(size_t, PAGE_SIZE >> 5, KMALLOC_MIN_SIZE + 1)) |
1da177e4 LT |
286 | |
287 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
288 | /* |
289 | * Optimization question: fewer reaps means less probability for unnessary | |
290 | * cpucache drain/refill cycles. | |
1da177e4 | 291 | * |
dc6f3f27 | 292 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
293 | * which could lock up otherwise freeable slabs. |
294 | */ | |
5f0985bb JZ |
295 | #define REAPTIMEOUT_AC (2*HZ) |
296 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
297 | |
298 | #if STATS | |
299 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
300 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
301 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
302 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 303 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
304 | #define STATS_SET_HIGH(x) \ |
305 | do { \ | |
306 | if ((x)->num_active > (x)->high_mark) \ | |
307 | (x)->high_mark = (x)->num_active; \ | |
308 | } while (0) | |
1da177e4 LT |
309 | #define STATS_INC_ERR(x) ((x)->errors++) |
310 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 311 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 312 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
313 | #define STATS_SET_FREEABLE(x, i) \ |
314 | do { \ | |
315 | if ((x)->max_freeable < i) \ | |
316 | (x)->max_freeable = i; \ | |
317 | } while (0) | |
1da177e4 LT |
318 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
319 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
320 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
321 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
322 | #else | |
323 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
324 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
325 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
326 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 327 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
328 | #define STATS_SET_HIGH(x) do { } while (0) |
329 | #define STATS_INC_ERR(x) do { } while (0) | |
330 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 331 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 332 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 333 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
334 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
335 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
336 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
337 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
338 | #endif | |
339 | ||
340 | #if DEBUG | |
1da177e4 | 341 | |
a737b3e2 AM |
342 | /* |
343 | * memory layout of objects: | |
1da177e4 | 344 | * 0 : objp |
3dafccf2 | 345 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
346 | * the end of an object is aligned with the end of the real |
347 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 348 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 349 | * redzone word. |
3dafccf2 | 350 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
351 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
352 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 353 | * [BYTES_PER_WORD long] |
1da177e4 | 354 | */ |
343e0d7a | 355 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 356 | { |
3dafccf2 | 357 | return cachep->obj_offset; |
1da177e4 LT |
358 | } |
359 | ||
b46b8f19 | 360 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
361 | { |
362 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
363 | return (unsigned long long*) (objp + obj_offset(cachep) - |
364 | sizeof(unsigned long long)); | |
1da177e4 LT |
365 | } |
366 | ||
b46b8f19 | 367 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
368 | { |
369 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
370 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 371 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 372 | sizeof(unsigned long long) - |
87a927c7 | 373 | REDZONE_ALIGN); |
3b0efdfa | 374 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 375 | sizeof(unsigned long long)); |
1da177e4 LT |
376 | } |
377 | ||
343e0d7a | 378 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
379 | { |
380 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 381 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
382 | } |
383 | ||
384 | #else | |
385 | ||
3dafccf2 | 386 | #define obj_offset(x) 0 |
b46b8f19 DW |
387 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
388 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
389 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
390 | ||
391 | #endif | |
392 | ||
03787301 JK |
393 | #define OBJECT_FREE (0) |
394 | #define OBJECT_ACTIVE (1) | |
395 | ||
396 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
397 | ||
398 | static void set_obj_status(struct page *page, int idx, int val) | |
399 | { | |
400 | int freelist_size; | |
401 | char *status; | |
402 | struct kmem_cache *cachep = page->slab_cache; | |
403 | ||
404 | freelist_size = cachep->num * sizeof(freelist_idx_t); | |
405 | status = (char *)page->freelist + freelist_size; | |
406 | status[idx] = val; | |
407 | } | |
408 | ||
409 | static inline unsigned int get_obj_status(struct page *page, int idx) | |
410 | { | |
411 | int freelist_size; | |
412 | char *status; | |
413 | struct kmem_cache *cachep = page->slab_cache; | |
414 | ||
415 | freelist_size = cachep->num * sizeof(freelist_idx_t); | |
416 | status = (char *)page->freelist + freelist_size; | |
417 | ||
418 | return status[idx]; | |
419 | } | |
420 | ||
421 | #else | |
422 | static inline void set_obj_status(struct page *page, int idx, int val) {} | |
423 | ||
424 | #endif | |
425 | ||
1da177e4 | 426 | /* |
3df1cccd DR |
427 | * Do not go above this order unless 0 objects fit into the slab or |
428 | * overridden on the command line. | |
1da177e4 | 429 | */ |
543585cc DR |
430 | #define SLAB_MAX_ORDER_HI 1 |
431 | #define SLAB_MAX_ORDER_LO 0 | |
432 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 433 | static bool slab_max_order_set __initdata; |
1da177e4 | 434 | |
6ed5eb22 PE |
435 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
436 | { | |
b49af68f | 437 | struct page *page = virt_to_head_page(obj); |
35026088 | 438 | return page->slab_cache; |
6ed5eb22 PE |
439 | } |
440 | ||
8456a648 | 441 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
442 | unsigned int idx) |
443 | { | |
8456a648 | 444 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
445 | } |
446 | ||
6a2d7a95 | 447 | /* |
3b0efdfa CL |
448 | * We want to avoid an expensive divide : (offset / cache->size) |
449 | * Using the fact that size is a constant for a particular cache, | |
450 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
451 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
452 | */ | |
453 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 454 | const struct page *page, void *obj) |
8fea4e96 | 455 | { |
8456a648 | 456 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 457 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
458 | } |
459 | ||
1da177e4 | 460 | /* internal cache of cache description objs */ |
9b030cb8 | 461 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
462 | .batchcount = 1, |
463 | .limit = BOOT_CPUCACHE_ENTRIES, | |
464 | .shared = 1, | |
3b0efdfa | 465 | .size = sizeof(struct kmem_cache), |
b28a02de | 466 | .name = "kmem_cache", |
1da177e4 LT |
467 | }; |
468 | ||
edcad250 JK |
469 | #define BAD_ALIEN_MAGIC 0x01020304ul |
470 | ||
1871e52c | 471 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 472 | |
343e0d7a | 473 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 474 | { |
bf0dea23 | 475 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
476 | } |
477 | ||
03787301 JK |
478 | static size_t calculate_freelist_size(int nr_objs, size_t align) |
479 | { | |
480 | size_t freelist_size; | |
481 | ||
482 | freelist_size = nr_objs * sizeof(freelist_idx_t); | |
483 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) | |
484 | freelist_size += nr_objs * sizeof(char); | |
485 | ||
486 | if (align) | |
487 | freelist_size = ALIGN(freelist_size, align); | |
488 | ||
489 | return freelist_size; | |
490 | } | |
491 | ||
9cef2e2b JK |
492 | static int calculate_nr_objs(size_t slab_size, size_t buffer_size, |
493 | size_t idx_size, size_t align) | |
1da177e4 | 494 | { |
9cef2e2b | 495 | int nr_objs; |
03787301 | 496 | size_t remained_size; |
9cef2e2b | 497 | size_t freelist_size; |
03787301 | 498 | int extra_space = 0; |
9cef2e2b | 499 | |
03787301 JK |
500 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
501 | extra_space = sizeof(char); | |
9cef2e2b JK |
502 | /* |
503 | * Ignore padding for the initial guess. The padding | |
504 | * is at most @align-1 bytes, and @buffer_size is at | |
505 | * least @align. In the worst case, this result will | |
506 | * be one greater than the number of objects that fit | |
507 | * into the memory allocation when taking the padding | |
508 | * into account. | |
509 | */ | |
03787301 | 510 | nr_objs = slab_size / (buffer_size + idx_size + extra_space); |
9cef2e2b JK |
511 | |
512 | /* | |
513 | * This calculated number will be either the right | |
514 | * amount, or one greater than what we want. | |
515 | */ | |
03787301 JK |
516 | remained_size = slab_size - nr_objs * buffer_size; |
517 | freelist_size = calculate_freelist_size(nr_objs, align); | |
518 | if (remained_size < freelist_size) | |
9cef2e2b JK |
519 | nr_objs--; |
520 | ||
521 | return nr_objs; | |
fbaccacf | 522 | } |
1da177e4 | 523 | |
a737b3e2 AM |
524 | /* |
525 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
526 | */ | |
fbaccacf SR |
527 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
528 | size_t align, int flags, size_t *left_over, | |
529 | unsigned int *num) | |
530 | { | |
531 | int nr_objs; | |
532 | size_t mgmt_size; | |
533 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 534 | |
fbaccacf SR |
535 | /* |
536 | * The slab management structure can be either off the slab or | |
537 | * on it. For the latter case, the memory allocated for a | |
538 | * slab is used for: | |
539 | * | |
16025177 | 540 | * - One unsigned int for each object |
fbaccacf SR |
541 | * - Padding to respect alignment of @align |
542 | * - @buffer_size bytes for each object | |
543 | * | |
544 | * If the slab management structure is off the slab, then the | |
545 | * alignment will already be calculated into the size. Because | |
546 | * the slabs are all pages aligned, the objects will be at the | |
547 | * correct alignment when allocated. | |
548 | */ | |
549 | if (flags & CFLGS_OFF_SLAB) { | |
550 | mgmt_size = 0; | |
551 | nr_objs = slab_size / buffer_size; | |
552 | ||
fbaccacf | 553 | } else { |
9cef2e2b | 554 | nr_objs = calculate_nr_objs(slab_size, buffer_size, |
a41adfaa | 555 | sizeof(freelist_idx_t), align); |
03787301 | 556 | mgmt_size = calculate_freelist_size(nr_objs, align); |
fbaccacf SR |
557 | } |
558 | *num = nr_objs; | |
559 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
560 | } |
561 | ||
f28510d3 | 562 | #if DEBUG |
d40cee24 | 563 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 564 | |
a737b3e2 AM |
565 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
566 | char *msg) | |
1da177e4 LT |
567 | { |
568 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 569 | function, cachep->name, msg); |
1da177e4 | 570 | dump_stack(); |
373d4d09 | 571 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 572 | } |
f28510d3 | 573 | #endif |
1da177e4 | 574 | |
3395ee05 PM |
575 | /* |
576 | * By default on NUMA we use alien caches to stage the freeing of | |
577 | * objects allocated from other nodes. This causes massive memory | |
578 | * inefficiencies when using fake NUMA setup to split memory into a | |
579 | * large number of small nodes, so it can be disabled on the command | |
580 | * line | |
581 | */ | |
582 | ||
583 | static int use_alien_caches __read_mostly = 1; | |
584 | static int __init noaliencache_setup(char *s) | |
585 | { | |
586 | use_alien_caches = 0; | |
587 | return 1; | |
588 | } | |
589 | __setup("noaliencache", noaliencache_setup); | |
590 | ||
3df1cccd DR |
591 | static int __init slab_max_order_setup(char *str) |
592 | { | |
593 | get_option(&str, &slab_max_order); | |
594 | slab_max_order = slab_max_order < 0 ? 0 : | |
595 | min(slab_max_order, MAX_ORDER - 1); | |
596 | slab_max_order_set = true; | |
597 | ||
598 | return 1; | |
599 | } | |
600 | __setup("slab_max_order=", slab_max_order_setup); | |
601 | ||
8fce4d8e CL |
602 | #ifdef CONFIG_NUMA |
603 | /* | |
604 | * Special reaping functions for NUMA systems called from cache_reap(). | |
605 | * These take care of doing round robin flushing of alien caches (containing | |
606 | * objects freed on different nodes from which they were allocated) and the | |
607 | * flushing of remote pcps by calling drain_node_pages. | |
608 | */ | |
1871e52c | 609 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
610 | |
611 | static void init_reap_node(int cpu) | |
612 | { | |
613 | int node; | |
614 | ||
7d6e6d09 | 615 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 616 | if (node == MAX_NUMNODES) |
442295c9 | 617 | node = first_node(node_online_map); |
8fce4d8e | 618 | |
1871e52c | 619 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
620 | } |
621 | ||
622 | static void next_reap_node(void) | |
623 | { | |
909ea964 | 624 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 625 | |
8fce4d8e CL |
626 | node = next_node(node, node_online_map); |
627 | if (unlikely(node >= MAX_NUMNODES)) | |
628 | node = first_node(node_online_map); | |
909ea964 | 629 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
630 | } |
631 | ||
632 | #else | |
633 | #define init_reap_node(cpu) do { } while (0) | |
634 | #define next_reap_node(void) do { } while (0) | |
635 | #endif | |
636 | ||
1da177e4 LT |
637 | /* |
638 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
639 | * via the workqueue/eventd. | |
640 | * Add the CPU number into the expiration time to minimize the possibility of | |
641 | * the CPUs getting into lockstep and contending for the global cache chain | |
642 | * lock. | |
643 | */ | |
0db0628d | 644 | static void start_cpu_timer(int cpu) |
1da177e4 | 645 | { |
1871e52c | 646 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
647 | |
648 | /* | |
649 | * When this gets called from do_initcalls via cpucache_init(), | |
650 | * init_workqueues() has already run, so keventd will be setup | |
651 | * at that time. | |
652 | */ | |
52bad64d | 653 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 654 | init_reap_node(cpu); |
203b42f7 | 655 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
656 | schedule_delayed_work_on(cpu, reap_work, |
657 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
658 | } |
659 | } | |
660 | ||
1fe00d50 | 661 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 662 | { |
d5cff635 CM |
663 | /* |
664 | * The array_cache structures contain pointers to free object. | |
25985edc | 665 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
666 | * cache the pointers are not cleared and they could be counted as |
667 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
668 | * not scan such objects. | |
669 | */ | |
1fe00d50 JK |
670 | kmemleak_no_scan(ac); |
671 | if (ac) { | |
672 | ac->avail = 0; | |
673 | ac->limit = limit; | |
674 | ac->batchcount = batch; | |
675 | ac->touched = 0; | |
1da177e4 | 676 | } |
1fe00d50 JK |
677 | } |
678 | ||
679 | static struct array_cache *alloc_arraycache(int node, int entries, | |
680 | int batchcount, gfp_t gfp) | |
681 | { | |
5e804789 | 682 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
683 | struct array_cache *ac = NULL; |
684 | ||
685 | ac = kmalloc_node(memsize, gfp, node); | |
686 | init_arraycache(ac, entries, batchcount); | |
687 | return ac; | |
1da177e4 LT |
688 | } |
689 | ||
8456a648 | 690 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 691 | { |
072bb0aa MG |
692 | return PageSlabPfmemalloc(page); |
693 | } | |
694 | ||
695 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
696 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
697 | struct array_cache *ac) | |
698 | { | |
18bf8541 | 699 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 700 | struct page *page; |
072bb0aa MG |
701 | unsigned long flags; |
702 | ||
703 | if (!pfmemalloc_active) | |
704 | return; | |
705 | ||
ce8eb6c4 | 706 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
707 | list_for_each_entry(page, &n->slabs_full, lru) |
708 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
709 | goto out; |
710 | ||
8456a648 JK |
711 | list_for_each_entry(page, &n->slabs_partial, lru) |
712 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
713 | goto out; |
714 | ||
8456a648 JK |
715 | list_for_each_entry(page, &n->slabs_free, lru) |
716 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
717 | goto out; |
718 | ||
719 | pfmemalloc_active = false; | |
720 | out: | |
ce8eb6c4 | 721 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
722 | } |
723 | ||
381760ea | 724 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
725 | gfp_t flags, bool force_refill) |
726 | { | |
727 | int i; | |
728 | void *objp = ac->entry[--ac->avail]; | |
729 | ||
730 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
731 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 732 | struct kmem_cache_node *n; |
072bb0aa MG |
733 | |
734 | if (gfp_pfmemalloc_allowed(flags)) { | |
735 | clear_obj_pfmemalloc(&objp); | |
736 | return objp; | |
737 | } | |
738 | ||
739 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 740 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
741 | /* If a !PFMEMALLOC object is found, swap them */ |
742 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
743 | objp = ac->entry[i]; | |
744 | ac->entry[i] = ac->entry[ac->avail]; | |
745 | ac->entry[ac->avail] = objp; | |
746 | return objp; | |
747 | } | |
748 | } | |
749 | ||
750 | /* | |
751 | * If there are empty slabs on the slabs_free list and we are | |
752 | * being forced to refill the cache, mark this one !pfmemalloc. | |
753 | */ | |
18bf8541 | 754 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 755 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 756 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 757 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
758 | clear_obj_pfmemalloc(&objp); |
759 | recheck_pfmemalloc_active(cachep, ac); | |
760 | return objp; | |
761 | } | |
762 | ||
763 | /* No !PFMEMALLOC objects available */ | |
764 | ac->avail++; | |
765 | objp = NULL; | |
766 | } | |
767 | ||
768 | return objp; | |
769 | } | |
770 | ||
381760ea MG |
771 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
772 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
773 | { | |
774 | void *objp; | |
775 | ||
776 | if (unlikely(sk_memalloc_socks())) | |
777 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
778 | else | |
779 | objp = ac->entry[--ac->avail]; | |
780 | ||
781 | return objp; | |
782 | } | |
783 | ||
d3aec344 JK |
784 | static noinline void *__ac_put_obj(struct kmem_cache *cachep, |
785 | struct array_cache *ac, void *objp) | |
072bb0aa MG |
786 | { |
787 | if (unlikely(pfmemalloc_active)) { | |
788 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 789 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
790 | if (PageSlabPfmemalloc(page)) |
791 | set_obj_pfmemalloc(&objp); | |
792 | } | |
793 | ||
381760ea MG |
794 | return objp; |
795 | } | |
796 | ||
797 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
798 | void *objp) | |
799 | { | |
800 | if (unlikely(sk_memalloc_socks())) | |
801 | objp = __ac_put_obj(cachep, ac, objp); | |
802 | ||
072bb0aa MG |
803 | ac->entry[ac->avail++] = objp; |
804 | } | |
805 | ||
3ded175a CL |
806 | /* |
807 | * Transfer objects in one arraycache to another. | |
808 | * Locking must be handled by the caller. | |
809 | * | |
810 | * Return the number of entries transferred. | |
811 | */ | |
812 | static int transfer_objects(struct array_cache *to, | |
813 | struct array_cache *from, unsigned int max) | |
814 | { | |
815 | /* Figure out how many entries to transfer */ | |
732eacc0 | 816 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
817 | |
818 | if (!nr) | |
819 | return 0; | |
820 | ||
821 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
822 | sizeof(void *) *nr); | |
823 | ||
824 | from->avail -= nr; | |
825 | to->avail += nr; | |
3ded175a CL |
826 | return nr; |
827 | } | |
828 | ||
765c4507 CL |
829 | #ifndef CONFIG_NUMA |
830 | ||
831 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 832 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 833 | |
c8522a3a JK |
834 | static inline struct alien_cache **alloc_alien_cache(int node, |
835 | int limit, gfp_t gfp) | |
765c4507 | 836 | { |
edcad250 | 837 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
838 | } |
839 | ||
c8522a3a | 840 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
841 | { |
842 | } | |
843 | ||
844 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
845 | { | |
846 | return 0; | |
847 | } | |
848 | ||
849 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
850 | gfp_t flags) | |
851 | { | |
852 | return NULL; | |
853 | } | |
854 | ||
8b98c169 | 855 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
856 | gfp_t flags, int nodeid) |
857 | { | |
858 | return NULL; | |
859 | } | |
860 | ||
4167e9b2 DR |
861 | static inline gfp_t gfp_exact_node(gfp_t flags) |
862 | { | |
863 | return flags; | |
864 | } | |
865 | ||
765c4507 CL |
866 | #else /* CONFIG_NUMA */ |
867 | ||
8b98c169 | 868 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 869 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 870 | |
c8522a3a JK |
871 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
872 | int batch, gfp_t gfp) | |
873 | { | |
5e804789 | 874 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
875 | struct alien_cache *alc = NULL; |
876 | ||
877 | alc = kmalloc_node(memsize, gfp, node); | |
878 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 879 | spin_lock_init(&alc->lock); |
c8522a3a JK |
880 | return alc; |
881 | } | |
882 | ||
883 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 884 | { |
c8522a3a | 885 | struct alien_cache **alc_ptr; |
5e804789 | 886 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
887 | int i; |
888 | ||
889 | if (limit > 1) | |
890 | limit = 12; | |
c8522a3a JK |
891 | alc_ptr = kzalloc_node(memsize, gfp, node); |
892 | if (!alc_ptr) | |
893 | return NULL; | |
894 | ||
895 | for_each_node(i) { | |
896 | if (i == node || !node_online(i)) | |
897 | continue; | |
898 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
899 | if (!alc_ptr[i]) { | |
900 | for (i--; i >= 0; i--) | |
901 | kfree(alc_ptr[i]); | |
902 | kfree(alc_ptr); | |
903 | return NULL; | |
e498be7d CL |
904 | } |
905 | } | |
c8522a3a | 906 | return alc_ptr; |
e498be7d CL |
907 | } |
908 | ||
c8522a3a | 909 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
910 | { |
911 | int i; | |
912 | ||
c8522a3a | 913 | if (!alc_ptr) |
e498be7d | 914 | return; |
e498be7d | 915 | for_each_node(i) |
c8522a3a JK |
916 | kfree(alc_ptr[i]); |
917 | kfree(alc_ptr); | |
e498be7d CL |
918 | } |
919 | ||
343e0d7a | 920 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
921 | struct array_cache *ac, int node, |
922 | struct list_head *list) | |
e498be7d | 923 | { |
18bf8541 | 924 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
925 | |
926 | if (ac->avail) { | |
ce8eb6c4 | 927 | spin_lock(&n->list_lock); |
e00946fe CL |
928 | /* |
929 | * Stuff objects into the remote nodes shared array first. | |
930 | * That way we could avoid the overhead of putting the objects | |
931 | * into the free lists and getting them back later. | |
932 | */ | |
ce8eb6c4 CL |
933 | if (n->shared) |
934 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 935 | |
833b706c | 936 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 937 | ac->avail = 0; |
ce8eb6c4 | 938 | spin_unlock(&n->list_lock); |
e498be7d CL |
939 | } |
940 | } | |
941 | ||
8fce4d8e CL |
942 | /* |
943 | * Called from cache_reap() to regularly drain alien caches round robin. | |
944 | */ | |
ce8eb6c4 | 945 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 946 | { |
909ea964 | 947 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 948 | |
ce8eb6c4 | 949 | if (n->alien) { |
c8522a3a JK |
950 | struct alien_cache *alc = n->alien[node]; |
951 | struct array_cache *ac; | |
952 | ||
953 | if (alc) { | |
954 | ac = &alc->ac; | |
49dfc304 | 955 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
956 | LIST_HEAD(list); |
957 | ||
958 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 959 | spin_unlock_irq(&alc->lock); |
833b706c | 960 | slabs_destroy(cachep, &list); |
c8522a3a | 961 | } |
8fce4d8e CL |
962 | } |
963 | } | |
964 | } | |
965 | ||
a737b3e2 | 966 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 967 | struct alien_cache **alien) |
e498be7d | 968 | { |
b28a02de | 969 | int i = 0; |
c8522a3a | 970 | struct alien_cache *alc; |
e498be7d CL |
971 | struct array_cache *ac; |
972 | unsigned long flags; | |
973 | ||
974 | for_each_online_node(i) { | |
c8522a3a JK |
975 | alc = alien[i]; |
976 | if (alc) { | |
833b706c JK |
977 | LIST_HEAD(list); |
978 | ||
c8522a3a | 979 | ac = &alc->ac; |
49dfc304 | 980 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 981 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 982 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 983 | slabs_destroy(cachep, &list); |
e498be7d CL |
984 | } |
985 | } | |
986 | } | |
729bd0b7 | 987 | |
25c4f304 JK |
988 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
989 | int node, int page_node) | |
729bd0b7 | 990 | { |
ce8eb6c4 | 991 | struct kmem_cache_node *n; |
c8522a3a JK |
992 | struct alien_cache *alien = NULL; |
993 | struct array_cache *ac; | |
97654dfa | 994 | LIST_HEAD(list); |
1ca4cb24 | 995 | |
18bf8541 | 996 | n = get_node(cachep, node); |
729bd0b7 | 997 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
998 | if (n->alien && n->alien[page_node]) { |
999 | alien = n->alien[page_node]; | |
c8522a3a | 1000 | ac = &alien->ac; |
49dfc304 | 1001 | spin_lock(&alien->lock); |
c8522a3a | 1002 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 1003 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 1004 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 1005 | } |
c8522a3a | 1006 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 1007 | spin_unlock(&alien->lock); |
833b706c | 1008 | slabs_destroy(cachep, &list); |
729bd0b7 | 1009 | } else { |
25c4f304 | 1010 | n = get_node(cachep, page_node); |
18bf8541 | 1011 | spin_lock(&n->list_lock); |
25c4f304 | 1012 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 1013 | spin_unlock(&n->list_lock); |
97654dfa | 1014 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
1015 | } |
1016 | return 1; | |
1017 | } | |
25c4f304 JK |
1018 | |
1019 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
1020 | { | |
1021 | int page_node = page_to_nid(virt_to_page(objp)); | |
1022 | int node = numa_mem_id(); | |
1023 | /* | |
1024 | * Make sure we are not freeing a object from another node to the array | |
1025 | * cache on this cpu. | |
1026 | */ | |
1027 | if (likely(node == page_node)) | |
1028 | return 0; | |
1029 | ||
1030 | return __cache_free_alien(cachep, objp, node, page_node); | |
1031 | } | |
4167e9b2 DR |
1032 | |
1033 | /* | |
d0164adc MG |
1034 | * Construct gfp mask to allocate from a specific node but do not direct reclaim |
1035 | * or warn about failures. kswapd may still wake to reclaim in the background. | |
4167e9b2 DR |
1036 | */ |
1037 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
1038 | { | |
d0164adc | 1039 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~__GFP_DIRECT_RECLAIM; |
4167e9b2 | 1040 | } |
e498be7d CL |
1041 | #endif |
1042 | ||
8f9f8d9e | 1043 | /* |
6a67368c | 1044 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1045 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1046 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1047 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1048 | * already in use. |
1049 | * | |
18004c5d | 1050 | * Must hold slab_mutex. |
8f9f8d9e | 1051 | */ |
6a67368c | 1052 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1053 | { |
1054 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1055 | struct kmem_cache_node *n; |
5e804789 | 1056 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1057 | |
18004c5d | 1058 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 1059 | /* |
5f0985bb | 1060 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
1061 | * begin anything. Make sure some other cpu on this |
1062 | * node has not already allocated this | |
1063 | */ | |
18bf8541 CL |
1064 | n = get_node(cachep, node); |
1065 | if (!n) { | |
ce8eb6c4 CL |
1066 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1067 | if (!n) | |
8f9f8d9e | 1068 | return -ENOMEM; |
ce8eb6c4 | 1069 | kmem_cache_node_init(n); |
5f0985bb JZ |
1070 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1071 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1072 | |
1073 | /* | |
5f0985bb JZ |
1074 | * The kmem_cache_nodes don't come and go as CPUs |
1075 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1076 | * protection here. |
1077 | */ | |
ce8eb6c4 | 1078 | cachep->node[node] = n; |
8f9f8d9e DR |
1079 | } |
1080 | ||
18bf8541 CL |
1081 | spin_lock_irq(&n->list_lock); |
1082 | n->free_limit = | |
8f9f8d9e DR |
1083 | (1 + nr_cpus_node(node)) * |
1084 | cachep->batchcount + cachep->num; | |
18bf8541 | 1085 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1086 | } |
1087 | return 0; | |
1088 | } | |
1089 | ||
0fa8103b WL |
1090 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1091 | struct kmem_cache_node *n) | |
1092 | { | |
1093 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1094 | } | |
1095 | ||
0db0628d | 1096 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1097 | { |
1098 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1099 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1100 | int node = cpu_to_mem(cpu); |
a70f7302 | 1101 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1102 | |
18004c5d | 1103 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1104 | struct array_cache *nc; |
1105 | struct array_cache *shared; | |
c8522a3a | 1106 | struct alien_cache **alien; |
97654dfa | 1107 | LIST_HEAD(list); |
fbf1e473 | 1108 | |
18bf8541 | 1109 | n = get_node(cachep, node); |
ce8eb6c4 | 1110 | if (!n) |
bf0dea23 | 1111 | continue; |
fbf1e473 | 1112 | |
ce8eb6c4 | 1113 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1114 | |
ce8eb6c4 CL |
1115 | /* Free limit for this kmem_cache_node */ |
1116 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1117 | |
1118 | /* cpu is dead; no one can alloc from it. */ | |
1119 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1120 | if (nc) { | |
97654dfa | 1121 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1122 | nc->avail = 0; |
1123 | } | |
fbf1e473 | 1124 | |
58463c1f | 1125 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1126 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1127 | goto free_slab; |
fbf1e473 AM |
1128 | } |
1129 | ||
ce8eb6c4 | 1130 | shared = n->shared; |
fbf1e473 AM |
1131 | if (shared) { |
1132 | free_block(cachep, shared->entry, | |
97654dfa | 1133 | shared->avail, node, &list); |
ce8eb6c4 | 1134 | n->shared = NULL; |
fbf1e473 AM |
1135 | } |
1136 | ||
ce8eb6c4 CL |
1137 | alien = n->alien; |
1138 | n->alien = NULL; | |
fbf1e473 | 1139 | |
ce8eb6c4 | 1140 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1141 | |
1142 | kfree(shared); | |
1143 | if (alien) { | |
1144 | drain_alien_cache(cachep, alien); | |
1145 | free_alien_cache(alien); | |
1146 | } | |
bf0dea23 JK |
1147 | |
1148 | free_slab: | |
97654dfa | 1149 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1150 | } |
1151 | /* | |
1152 | * In the previous loop, all the objects were freed to | |
1153 | * the respective cache's slabs, now we can go ahead and | |
1154 | * shrink each nodelist to its limit. | |
1155 | */ | |
18004c5d | 1156 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1157 | n = get_node(cachep, node); |
ce8eb6c4 | 1158 | if (!n) |
fbf1e473 | 1159 | continue; |
0fa8103b | 1160 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1161 | } |
1162 | } | |
1163 | ||
0db0628d | 1164 | static int cpuup_prepare(long cpu) |
1da177e4 | 1165 | { |
343e0d7a | 1166 | struct kmem_cache *cachep; |
ce8eb6c4 | 1167 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1168 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1169 | int err; |
1da177e4 | 1170 | |
fbf1e473 AM |
1171 | /* |
1172 | * We need to do this right in the beginning since | |
1173 | * alloc_arraycache's are going to use this list. | |
1174 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1175 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1176 | */ |
6a67368c | 1177 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1178 | if (err < 0) |
1179 | goto bad; | |
fbf1e473 AM |
1180 | |
1181 | /* | |
1182 | * Now we can go ahead with allocating the shared arrays and | |
1183 | * array caches | |
1184 | */ | |
18004c5d | 1185 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 1186 | struct array_cache *shared = NULL; |
c8522a3a | 1187 | struct alien_cache **alien = NULL; |
fbf1e473 | 1188 | |
fbf1e473 AM |
1189 | if (cachep->shared) { |
1190 | shared = alloc_arraycache(node, | |
1191 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1192 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 1193 | if (!shared) |
1da177e4 | 1194 | goto bad; |
fbf1e473 AM |
1195 | } |
1196 | if (use_alien_caches) { | |
83b519e8 | 1197 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1198 | if (!alien) { |
1199 | kfree(shared); | |
fbf1e473 | 1200 | goto bad; |
12d00f6a | 1201 | } |
fbf1e473 | 1202 | } |
18bf8541 | 1203 | n = get_node(cachep, node); |
ce8eb6c4 | 1204 | BUG_ON(!n); |
fbf1e473 | 1205 | |
ce8eb6c4 CL |
1206 | spin_lock_irq(&n->list_lock); |
1207 | if (!n->shared) { | |
fbf1e473 AM |
1208 | /* |
1209 | * We are serialised from CPU_DEAD or | |
1210 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1211 | */ | |
ce8eb6c4 | 1212 | n->shared = shared; |
fbf1e473 AM |
1213 | shared = NULL; |
1214 | } | |
4484ebf1 | 1215 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1216 | if (!n->alien) { |
1217 | n->alien = alien; | |
fbf1e473 | 1218 | alien = NULL; |
1da177e4 | 1219 | } |
fbf1e473 | 1220 | #endif |
ce8eb6c4 | 1221 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1222 | kfree(shared); |
1223 | free_alien_cache(alien); | |
1224 | } | |
ce79ddc8 | 1225 | |
fbf1e473 AM |
1226 | return 0; |
1227 | bad: | |
12d00f6a | 1228 | cpuup_canceled(cpu); |
fbf1e473 AM |
1229 | return -ENOMEM; |
1230 | } | |
1231 | ||
0db0628d | 1232 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1233 | unsigned long action, void *hcpu) |
1234 | { | |
1235 | long cpu = (long)hcpu; | |
1236 | int err = 0; | |
1237 | ||
1238 | switch (action) { | |
fbf1e473 AM |
1239 | case CPU_UP_PREPARE: |
1240 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1241 | mutex_lock(&slab_mutex); |
fbf1e473 | 1242 | err = cpuup_prepare(cpu); |
18004c5d | 1243 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1244 | break; |
1245 | case CPU_ONLINE: | |
8bb78442 | 1246 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1247 | start_cpu_timer(cpu); |
1248 | break; | |
1249 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1250 | case CPU_DOWN_PREPARE: |
8bb78442 | 1251 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1252 | /* |
18004c5d | 1253 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1254 | * held so that if cache_reap() is invoked it cannot do |
1255 | * anything expensive but will only modify reap_work | |
1256 | * and reschedule the timer. | |
1257 | */ | |
afe2c511 | 1258 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1259 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1260 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1261 | break; |
1262 | case CPU_DOWN_FAILED: | |
8bb78442 | 1263 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1264 | start_cpu_timer(cpu); |
1265 | break; | |
1da177e4 | 1266 | case CPU_DEAD: |
8bb78442 | 1267 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1268 | /* |
1269 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1270 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1271 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1272 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1273 | * structure is usually allocated from kmem_cache_create() and |
1274 | * gets destroyed at kmem_cache_destroy(). | |
1275 | */ | |
183ff22b | 1276 | /* fall through */ |
8f5be20b | 1277 | #endif |
1da177e4 | 1278 | case CPU_UP_CANCELED: |
8bb78442 | 1279 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1280 | mutex_lock(&slab_mutex); |
fbf1e473 | 1281 | cpuup_canceled(cpu); |
18004c5d | 1282 | mutex_unlock(&slab_mutex); |
1da177e4 | 1283 | break; |
1da177e4 | 1284 | } |
eac40680 | 1285 | return notifier_from_errno(err); |
1da177e4 LT |
1286 | } |
1287 | ||
0db0628d | 1288 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1289 | &cpuup_callback, NULL, 0 |
1290 | }; | |
1da177e4 | 1291 | |
8f9f8d9e DR |
1292 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1293 | /* | |
1294 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1295 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1296 | * removed. | |
1297 | * | |
18004c5d | 1298 | * Must hold slab_mutex. |
8f9f8d9e | 1299 | */ |
6a67368c | 1300 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1301 | { |
1302 | struct kmem_cache *cachep; | |
1303 | int ret = 0; | |
1304 | ||
18004c5d | 1305 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1306 | struct kmem_cache_node *n; |
8f9f8d9e | 1307 | |
18bf8541 | 1308 | n = get_node(cachep, node); |
ce8eb6c4 | 1309 | if (!n) |
8f9f8d9e DR |
1310 | continue; |
1311 | ||
0fa8103b | 1312 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1313 | |
ce8eb6c4 CL |
1314 | if (!list_empty(&n->slabs_full) || |
1315 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1316 | ret = -EBUSY; |
1317 | break; | |
1318 | } | |
1319 | } | |
1320 | return ret; | |
1321 | } | |
1322 | ||
1323 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1324 | unsigned long action, void *arg) | |
1325 | { | |
1326 | struct memory_notify *mnb = arg; | |
1327 | int ret = 0; | |
1328 | int nid; | |
1329 | ||
1330 | nid = mnb->status_change_nid; | |
1331 | if (nid < 0) | |
1332 | goto out; | |
1333 | ||
1334 | switch (action) { | |
1335 | case MEM_GOING_ONLINE: | |
18004c5d | 1336 | mutex_lock(&slab_mutex); |
6a67368c | 1337 | ret = init_cache_node_node(nid); |
18004c5d | 1338 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1339 | break; |
1340 | case MEM_GOING_OFFLINE: | |
18004c5d | 1341 | mutex_lock(&slab_mutex); |
6a67368c | 1342 | ret = drain_cache_node_node(nid); |
18004c5d | 1343 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1344 | break; |
1345 | case MEM_ONLINE: | |
1346 | case MEM_OFFLINE: | |
1347 | case MEM_CANCEL_ONLINE: | |
1348 | case MEM_CANCEL_OFFLINE: | |
1349 | break; | |
1350 | } | |
1351 | out: | |
5fda1bd5 | 1352 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1353 | } |
1354 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1355 | ||
e498be7d | 1356 | /* |
ce8eb6c4 | 1357 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1358 | */ |
6744f087 | 1359 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1360 | int nodeid) |
e498be7d | 1361 | { |
6744f087 | 1362 | struct kmem_cache_node *ptr; |
e498be7d | 1363 | |
6744f087 | 1364 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1365 | BUG_ON(!ptr); |
1366 | ||
6744f087 | 1367 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1368 | /* |
1369 | * Do not assume that spinlocks can be initialized via memcpy: | |
1370 | */ | |
1371 | spin_lock_init(&ptr->list_lock); | |
1372 | ||
e498be7d | 1373 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1374 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1375 | } |
1376 | ||
556a169d | 1377 | /* |
ce8eb6c4 CL |
1378 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1379 | * size of kmem_cache_node. | |
556a169d | 1380 | */ |
ce8eb6c4 | 1381 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1382 | { |
1383 | int node; | |
1384 | ||
1385 | for_each_online_node(node) { | |
ce8eb6c4 | 1386 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1387 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1388 | REAPTIMEOUT_NODE + |
1389 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1390 | } |
1391 | } | |
1392 | ||
a737b3e2 AM |
1393 | /* |
1394 | * Initialisation. Called after the page allocator have been initialised and | |
1395 | * before smp_init(). | |
1da177e4 LT |
1396 | */ |
1397 | void __init kmem_cache_init(void) | |
1398 | { | |
e498be7d CL |
1399 | int i; |
1400 | ||
68126702 JK |
1401 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1402 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1403 | kmem_cache = &kmem_cache_boot; |
1404 | ||
b6e68bc1 | 1405 | if (num_possible_nodes() == 1) |
62918a03 SS |
1406 | use_alien_caches = 0; |
1407 | ||
3c583465 | 1408 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1409 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1410 | |
1da177e4 LT |
1411 | /* |
1412 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1413 | * page orders on machines with more than 32MB of memory if |
1414 | * not overridden on the command line. | |
1da177e4 | 1415 | */ |
3df1cccd | 1416 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1417 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1418 | |
1da177e4 LT |
1419 | /* Bootstrap is tricky, because several objects are allocated |
1420 | * from caches that do not exist yet: | |
9b030cb8 CL |
1421 | * 1) initialize the kmem_cache cache: it contains the struct |
1422 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1423 | * kmem_cache is statically allocated. | |
e498be7d | 1424 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1425 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1426 | * array at the end of the bootstrap. |
1da177e4 | 1427 | * 2) Create the first kmalloc cache. |
343e0d7a | 1428 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1429 | * An __init data area is used for the head array. |
1430 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1431 | * head arrays. | |
9b030cb8 | 1432 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1433 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1434 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1435 | * the other cache's with kmalloc allocated memory. |
1436 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1437 | */ |
1438 | ||
9b030cb8 | 1439 | /* 1) create the kmem_cache */ |
1da177e4 | 1440 | |
8da3430d | 1441 | /* |
b56efcf0 | 1442 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1443 | */ |
2f9baa9f | 1444 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1445 | offsetof(struct kmem_cache, node) + |
6744f087 | 1446 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1447 | SLAB_HWCACHE_ALIGN); |
1448 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1449 | slab_state = PARTIAL; |
1da177e4 | 1450 | |
a737b3e2 | 1451 | /* |
bf0dea23 JK |
1452 | * Initialize the caches that provide memory for the kmem_cache_node |
1453 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1454 | */ |
bf0dea23 | 1455 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1456 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1457 | slab_state = PARTIAL_NODE; |
34cc6990 | 1458 | setup_kmalloc_cache_index_table(); |
e498be7d | 1459 | |
e0a42726 IM |
1460 | slab_early_init = 0; |
1461 | ||
ce8eb6c4 | 1462 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1463 | { |
1ca4cb24 PE |
1464 | int nid; |
1465 | ||
9c09a95c | 1466 | for_each_online_node(nid) { |
ce8eb6c4 | 1467 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1468 | |
bf0dea23 | 1469 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1470 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1471 | } |
1472 | } | |
1da177e4 | 1473 | |
f97d5f63 | 1474 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1475 | } |
1476 | ||
1477 | void __init kmem_cache_init_late(void) | |
1478 | { | |
1479 | struct kmem_cache *cachep; | |
1480 | ||
97d06609 | 1481 | slab_state = UP; |
52cef189 | 1482 | |
8429db5c | 1483 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1484 | mutex_lock(&slab_mutex); |
1485 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1486 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1487 | BUG(); | |
18004c5d | 1488 | mutex_unlock(&slab_mutex); |
056c6241 | 1489 | |
97d06609 CL |
1490 | /* Done! */ |
1491 | slab_state = FULL; | |
1492 | ||
a737b3e2 AM |
1493 | /* |
1494 | * Register a cpu startup notifier callback that initializes | |
1495 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1496 | */ |
1497 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1498 | |
8f9f8d9e DR |
1499 | #ifdef CONFIG_NUMA |
1500 | /* | |
1501 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1502 | * node. |
8f9f8d9e DR |
1503 | */ |
1504 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1505 | #endif | |
1506 | ||
a737b3e2 AM |
1507 | /* |
1508 | * The reap timers are started later, with a module init call: That part | |
1509 | * of the kernel is not yet operational. | |
1da177e4 LT |
1510 | */ |
1511 | } | |
1512 | ||
1513 | static int __init cpucache_init(void) | |
1514 | { | |
1515 | int cpu; | |
1516 | ||
a737b3e2 AM |
1517 | /* |
1518 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1519 | */ |
e498be7d | 1520 | for_each_online_cpu(cpu) |
a737b3e2 | 1521 | start_cpu_timer(cpu); |
a164f896 GC |
1522 | |
1523 | /* Done! */ | |
97d06609 | 1524 | slab_state = FULL; |
1da177e4 LT |
1525 | return 0; |
1526 | } | |
1da177e4 LT |
1527 | __initcall(cpucache_init); |
1528 | ||
8bdec192 RA |
1529 | static noinline void |
1530 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1531 | { | |
9a02d699 | 1532 | #if DEBUG |
ce8eb6c4 | 1533 | struct kmem_cache_node *n; |
8456a648 | 1534 | struct page *page; |
8bdec192 RA |
1535 | unsigned long flags; |
1536 | int node; | |
9a02d699 DR |
1537 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1538 | DEFAULT_RATELIMIT_BURST); | |
1539 | ||
1540 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1541 | return; | |
8bdec192 RA |
1542 | |
1543 | printk(KERN_WARNING | |
1544 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1545 | nodeid, gfpflags); | |
1546 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1547 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1548 | |
18bf8541 | 1549 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1550 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1551 | unsigned long active_slabs = 0, num_slabs = 0; | |
1552 | ||
ce8eb6c4 | 1553 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1554 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1555 | active_objs += cachep->num; |
1556 | active_slabs++; | |
1557 | } | |
8456a648 JK |
1558 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1559 | active_objs += page->active; | |
8bdec192 RA |
1560 | active_slabs++; |
1561 | } | |
8456a648 | 1562 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1563 | num_slabs++; |
1564 | ||
ce8eb6c4 CL |
1565 | free_objects += n->free_objects; |
1566 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1567 | |
1568 | num_slabs += active_slabs; | |
1569 | num_objs = num_slabs * cachep->num; | |
1570 | printk(KERN_WARNING | |
1571 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1572 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1573 | free_objects); | |
1574 | } | |
9a02d699 | 1575 | #endif |
8bdec192 RA |
1576 | } |
1577 | ||
1da177e4 | 1578 | /* |
8a7d9b43 WSH |
1579 | * Interface to system's page allocator. No need to hold the |
1580 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1581 | * |
1582 | * If we requested dmaable memory, we will get it. Even if we | |
1583 | * did not request dmaable memory, we might get it, but that | |
1584 | * would be relatively rare and ignorable. | |
1585 | */ | |
0c3aa83e JK |
1586 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1587 | int nodeid) | |
1da177e4 LT |
1588 | { |
1589 | struct page *page; | |
e1b6aa6f | 1590 | int nr_pages; |
765c4507 | 1591 | |
a618e89f | 1592 | flags |= cachep->allocflags; |
e12ba74d MG |
1593 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1594 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1595 | |
96db800f | 1596 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1597 | if (!page) { |
9a02d699 | 1598 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1599 | return NULL; |
8bdec192 | 1600 | } |
1da177e4 | 1601 | |
f3ccb2c4 VD |
1602 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1603 | __free_pages(page, cachep->gfporder); | |
1604 | return NULL; | |
1605 | } | |
1606 | ||
b37f1dd0 | 1607 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
2f064f34 | 1608 | if (page_is_pfmemalloc(page)) |
072bb0aa MG |
1609 | pfmemalloc_active = true; |
1610 | ||
e1b6aa6f | 1611 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1612 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1613 | add_zone_page_state(page_zone(page), |
1614 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1615 | else | |
1616 | add_zone_page_state(page_zone(page), | |
1617 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 | 1618 | __SetPageSlab(page); |
2f064f34 | 1619 | if (page_is_pfmemalloc(page)) |
a57a4988 | 1620 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1621 | |
b1eeab67 VN |
1622 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1623 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1624 | ||
1625 | if (cachep->ctor) | |
1626 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1627 | else | |
1628 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1629 | } | |
c175eea4 | 1630 | |
0c3aa83e | 1631 | return page; |
1da177e4 LT |
1632 | } |
1633 | ||
1634 | /* | |
1635 | * Interface to system's page release. | |
1636 | */ | |
0c3aa83e | 1637 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1638 | { |
a57a4988 | 1639 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1640 | |
b1eeab67 | 1641 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1642 | |
972d1a7b CL |
1643 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1644 | sub_zone_page_state(page_zone(page), | |
1645 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1646 | else | |
1647 | sub_zone_page_state(page_zone(page), | |
1648 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1649 | |
a57a4988 | 1650 | BUG_ON(!PageSlab(page)); |
73293c2f | 1651 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1652 | __ClearPageSlab(page); |
8456a648 JK |
1653 | page_mapcount_reset(page); |
1654 | page->mapping = NULL; | |
1f458cbf | 1655 | |
1da177e4 LT |
1656 | if (current->reclaim_state) |
1657 | current->reclaim_state->reclaimed_slab += nr_freed; | |
f3ccb2c4 | 1658 | __free_kmem_pages(page, cachep->gfporder); |
1da177e4 LT |
1659 | } |
1660 | ||
1661 | static void kmem_rcu_free(struct rcu_head *head) | |
1662 | { | |
68126702 JK |
1663 | struct kmem_cache *cachep; |
1664 | struct page *page; | |
1da177e4 | 1665 | |
68126702 JK |
1666 | page = container_of(head, struct page, rcu_head); |
1667 | cachep = page->slab_cache; | |
1668 | ||
1669 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1670 | } |
1671 | ||
1672 | #if DEBUG | |
1673 | ||
1674 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1675 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1676 | unsigned long caller) |
1da177e4 | 1677 | { |
8c138bc0 | 1678 | int size = cachep->object_size; |
1da177e4 | 1679 | |
3dafccf2 | 1680 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1681 | |
b28a02de | 1682 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1683 | return; |
1684 | ||
b28a02de PE |
1685 | *addr++ = 0x12345678; |
1686 | *addr++ = caller; | |
1687 | *addr++ = smp_processor_id(); | |
1688 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1689 | { |
1690 | unsigned long *sptr = &caller; | |
1691 | unsigned long svalue; | |
1692 | ||
1693 | while (!kstack_end(sptr)) { | |
1694 | svalue = *sptr++; | |
1695 | if (kernel_text_address(svalue)) { | |
b28a02de | 1696 | *addr++ = svalue; |
1da177e4 LT |
1697 | size -= sizeof(unsigned long); |
1698 | if (size <= sizeof(unsigned long)) | |
1699 | break; | |
1700 | } | |
1701 | } | |
1702 | ||
1703 | } | |
b28a02de | 1704 | *addr++ = 0x87654321; |
1da177e4 LT |
1705 | } |
1706 | #endif | |
1707 | ||
343e0d7a | 1708 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1709 | { |
8c138bc0 | 1710 | int size = cachep->object_size; |
3dafccf2 | 1711 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1712 | |
1713 | memset(addr, val, size); | |
b28a02de | 1714 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1715 | } |
1716 | ||
1717 | static void dump_line(char *data, int offset, int limit) | |
1718 | { | |
1719 | int i; | |
aa83aa40 DJ |
1720 | unsigned char error = 0; |
1721 | int bad_count = 0; | |
1722 | ||
fdde6abb | 1723 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1724 | for (i = 0; i < limit; i++) { |
1725 | if (data[offset + i] != POISON_FREE) { | |
1726 | error = data[offset + i]; | |
1727 | bad_count++; | |
1728 | } | |
aa83aa40 | 1729 | } |
fdde6abb SAS |
1730 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1731 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1732 | |
1733 | if (bad_count == 1) { | |
1734 | error ^= POISON_FREE; | |
1735 | if (!(error & (error - 1))) { | |
1736 | printk(KERN_ERR "Single bit error detected. Probably " | |
1737 | "bad RAM.\n"); | |
1738 | #ifdef CONFIG_X86 | |
1739 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1740 | "test tool.\n"); | |
1741 | #else | |
1742 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1743 | #endif | |
1744 | } | |
1745 | } | |
1da177e4 LT |
1746 | } |
1747 | #endif | |
1748 | ||
1749 | #if DEBUG | |
1750 | ||
343e0d7a | 1751 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1752 | { |
1753 | int i, size; | |
1754 | char *realobj; | |
1755 | ||
1756 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1757 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1758 | *dbg_redzone1(cachep, objp), |
1759 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1760 | } |
1761 | ||
1762 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1763 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1764 | *dbg_userword(cachep, objp), | |
1765 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1766 | } |
3dafccf2 | 1767 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1768 | size = cachep->object_size; |
b28a02de | 1769 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1770 | int limit; |
1771 | limit = 16; | |
b28a02de PE |
1772 | if (i + limit > size) |
1773 | limit = size - i; | |
1da177e4 LT |
1774 | dump_line(realobj, i, limit); |
1775 | } | |
1776 | } | |
1777 | ||
343e0d7a | 1778 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1779 | { |
1780 | char *realobj; | |
1781 | int size, i; | |
1782 | int lines = 0; | |
1783 | ||
3dafccf2 | 1784 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1785 | size = cachep->object_size; |
1da177e4 | 1786 | |
b28a02de | 1787 | for (i = 0; i < size; i++) { |
1da177e4 | 1788 | char exp = POISON_FREE; |
b28a02de | 1789 | if (i == size - 1) |
1da177e4 LT |
1790 | exp = POISON_END; |
1791 | if (realobj[i] != exp) { | |
1792 | int limit; | |
1793 | /* Mismatch ! */ | |
1794 | /* Print header */ | |
1795 | if (lines == 0) { | |
b28a02de | 1796 | printk(KERN_ERR |
face37f5 DJ |
1797 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1798 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1799 | print_objinfo(cachep, objp, 0); |
1800 | } | |
1801 | /* Hexdump the affected line */ | |
b28a02de | 1802 | i = (i / 16) * 16; |
1da177e4 | 1803 | limit = 16; |
b28a02de PE |
1804 | if (i + limit > size) |
1805 | limit = size - i; | |
1da177e4 LT |
1806 | dump_line(realobj, i, limit); |
1807 | i += 16; | |
1808 | lines++; | |
1809 | /* Limit to 5 lines */ | |
1810 | if (lines > 5) | |
1811 | break; | |
1812 | } | |
1813 | } | |
1814 | if (lines != 0) { | |
1815 | /* Print some data about the neighboring objects, if they | |
1816 | * exist: | |
1817 | */ | |
8456a648 | 1818 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1819 | unsigned int objnr; |
1da177e4 | 1820 | |
8456a648 | 1821 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1822 | if (objnr) { |
8456a648 | 1823 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1824 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1825 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1826 | realobj, size); |
1da177e4 LT |
1827 | print_objinfo(cachep, objp, 2); |
1828 | } | |
b28a02de | 1829 | if (objnr + 1 < cachep->num) { |
8456a648 | 1830 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1831 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1832 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1833 | realobj, size); |
1da177e4 LT |
1834 | print_objinfo(cachep, objp, 2); |
1835 | } | |
1836 | } | |
1837 | } | |
1838 | #endif | |
1839 | ||
12dd36fa | 1840 | #if DEBUG |
8456a648 JK |
1841 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1842 | struct page *page) | |
1da177e4 | 1843 | { |
1da177e4 LT |
1844 | int i; |
1845 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 1846 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1847 | |
1848 | if (cachep->flags & SLAB_POISON) { | |
1849 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 1850 | if (cachep->size % PAGE_SIZE == 0 && |
a737b3e2 | 1851 | OFF_SLAB(cachep)) |
b28a02de | 1852 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 1853 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
1854 | else |
1855 | check_poison_obj(cachep, objp); | |
1856 | #else | |
1857 | check_poison_obj(cachep, objp); | |
1858 | #endif | |
1859 | } | |
1860 | if (cachep->flags & SLAB_RED_ZONE) { | |
1861 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1862 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1863 | "was overwritten"); |
1da177e4 LT |
1864 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1865 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1866 | "was overwritten"); |
1da177e4 | 1867 | } |
1da177e4 | 1868 | } |
12dd36fa | 1869 | } |
1da177e4 | 1870 | #else |
8456a648 JK |
1871 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1872 | struct page *page) | |
12dd36fa | 1873 | { |
12dd36fa | 1874 | } |
1da177e4 LT |
1875 | #endif |
1876 | ||
911851e6 RD |
1877 | /** |
1878 | * slab_destroy - destroy and release all objects in a slab | |
1879 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1880 | * @page: page pointer being destroyed |
911851e6 | 1881 | * |
8a7d9b43 WSH |
1882 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1883 | * Before calling the slab page must have been unlinked from the cache. The | |
1884 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1885 | */ |
8456a648 | 1886 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1887 | { |
7e007355 | 1888 | void *freelist; |
12dd36fa | 1889 | |
8456a648 JK |
1890 | freelist = page->freelist; |
1891 | slab_destroy_debugcheck(cachep, page); | |
1da177e4 | 1892 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
68126702 JK |
1893 | struct rcu_head *head; |
1894 | ||
1895 | /* | |
1896 | * RCU free overloads the RCU head over the LRU. | |
1897 | * slab_page has been overloeaded over the LRU, | |
1898 | * however it is not used from now on so that | |
1899 | * we can use it safely. | |
1900 | */ | |
1901 | head = (void *)&page->rcu_head; | |
1902 | call_rcu(head, kmem_rcu_free); | |
1da177e4 | 1903 | |
1da177e4 | 1904 | } else { |
0c3aa83e | 1905 | kmem_freepages(cachep, page); |
1da177e4 | 1906 | } |
68126702 JK |
1907 | |
1908 | /* | |
8456a648 | 1909 | * From now on, we don't use freelist |
68126702 JK |
1910 | * although actual page can be freed in rcu context |
1911 | */ | |
1912 | if (OFF_SLAB(cachep)) | |
8456a648 | 1913 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1914 | } |
1915 | ||
97654dfa JK |
1916 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1917 | { | |
1918 | struct page *page, *n; | |
1919 | ||
1920 | list_for_each_entry_safe(page, n, list, lru) { | |
1921 | list_del(&page->lru); | |
1922 | slab_destroy(cachep, page); | |
1923 | } | |
1924 | } | |
1925 | ||
4d268eba | 1926 | /** |
a70773dd RD |
1927 | * calculate_slab_order - calculate size (page order) of slabs |
1928 | * @cachep: pointer to the cache that is being created | |
1929 | * @size: size of objects to be created in this cache. | |
1930 | * @align: required alignment for the objects. | |
1931 | * @flags: slab allocation flags | |
1932 | * | |
1933 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1934 | * |
1935 | * This could be made much more intelligent. For now, try to avoid using | |
1936 | * high order pages for slabs. When the gfp() functions are more friendly | |
1937 | * towards high-order requests, this should be changed. | |
1938 | */ | |
a737b3e2 | 1939 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 1940 | size_t size, size_t align, unsigned long flags) |
4d268eba | 1941 | { |
b1ab41c4 | 1942 | unsigned long offslab_limit; |
4d268eba | 1943 | size_t left_over = 0; |
9888e6fa | 1944 | int gfporder; |
4d268eba | 1945 | |
0aa817f0 | 1946 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1947 | unsigned int num; |
1948 | size_t remainder; | |
1949 | ||
9888e6fa | 1950 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
1951 | if (!num) |
1952 | continue; | |
9888e6fa | 1953 | |
f315e3fa JK |
1954 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1955 | if (num > SLAB_OBJ_MAX_NUM) | |
1956 | break; | |
1957 | ||
b1ab41c4 | 1958 | if (flags & CFLGS_OFF_SLAB) { |
03787301 | 1959 | size_t freelist_size_per_obj = sizeof(freelist_idx_t); |
b1ab41c4 IM |
1960 | /* |
1961 | * Max number of objs-per-slab for caches which | |
1962 | * use off-slab slabs. Needed to avoid a possible | |
1963 | * looping condition in cache_grow(). | |
1964 | */ | |
03787301 JK |
1965 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
1966 | freelist_size_per_obj += sizeof(char); | |
8456a648 | 1967 | offslab_limit = size; |
03787301 | 1968 | offslab_limit /= freelist_size_per_obj; |
b1ab41c4 IM |
1969 | |
1970 | if (num > offslab_limit) | |
1971 | break; | |
1972 | } | |
4d268eba | 1973 | |
9888e6fa | 1974 | /* Found something acceptable - save it away */ |
4d268eba | 1975 | cachep->num = num; |
9888e6fa | 1976 | cachep->gfporder = gfporder; |
4d268eba PE |
1977 | left_over = remainder; |
1978 | ||
f78bb8ad LT |
1979 | /* |
1980 | * A VFS-reclaimable slab tends to have most allocations | |
1981 | * as GFP_NOFS and we really don't want to have to be allocating | |
1982 | * higher-order pages when we are unable to shrink dcache. | |
1983 | */ | |
1984 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1985 | break; | |
1986 | ||
4d268eba PE |
1987 | /* |
1988 | * Large number of objects is good, but very large slabs are | |
1989 | * currently bad for the gfp()s. | |
1990 | */ | |
543585cc | 1991 | if (gfporder >= slab_max_order) |
4d268eba PE |
1992 | break; |
1993 | ||
9888e6fa LT |
1994 | /* |
1995 | * Acceptable internal fragmentation? | |
1996 | */ | |
a737b3e2 | 1997 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1998 | break; |
1999 | } | |
2000 | return left_over; | |
2001 | } | |
2002 | ||
bf0dea23 JK |
2003 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
2004 | struct kmem_cache *cachep, int entries, int batchcount) | |
2005 | { | |
2006 | int cpu; | |
2007 | size_t size; | |
2008 | struct array_cache __percpu *cpu_cache; | |
2009 | ||
2010 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 2011 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
2012 | |
2013 | if (!cpu_cache) | |
2014 | return NULL; | |
2015 | ||
2016 | for_each_possible_cpu(cpu) { | |
2017 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
2018 | entries, batchcount); | |
2019 | } | |
2020 | ||
2021 | return cpu_cache; | |
2022 | } | |
2023 | ||
83b519e8 | 2024 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2025 | { |
97d06609 | 2026 | if (slab_state >= FULL) |
83b519e8 | 2027 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2028 | |
bf0dea23 JK |
2029 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
2030 | if (!cachep->cpu_cache) | |
2031 | return 1; | |
2032 | ||
97d06609 | 2033 | if (slab_state == DOWN) { |
bf0dea23 JK |
2034 | /* Creation of first cache (kmem_cache). */ |
2035 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 2036 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
2037 | /* For kmem_cache_node */ |
2038 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 2039 | } else { |
bf0dea23 | 2040 | int node; |
f30cf7d1 | 2041 | |
bf0dea23 JK |
2042 | for_each_online_node(node) { |
2043 | cachep->node[node] = kmalloc_node( | |
2044 | sizeof(struct kmem_cache_node), gfp, node); | |
2045 | BUG_ON(!cachep->node[node]); | |
2046 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
2047 | } |
2048 | } | |
bf0dea23 | 2049 | |
6a67368c | 2050 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
2051 | jiffies + REAPTIMEOUT_NODE + |
2052 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
2053 | |
2054 | cpu_cache_get(cachep)->avail = 0; | |
2055 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2056 | cpu_cache_get(cachep)->batchcount = 1; | |
2057 | cpu_cache_get(cachep)->touched = 0; | |
2058 | cachep->batchcount = 1; | |
2059 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2060 | return 0; |
f30cf7d1 PE |
2061 | } |
2062 | ||
12220dea JK |
2063 | unsigned long kmem_cache_flags(unsigned long object_size, |
2064 | unsigned long flags, const char *name, | |
2065 | void (*ctor)(void *)) | |
2066 | { | |
2067 | return flags; | |
2068 | } | |
2069 | ||
2070 | struct kmem_cache * | |
2071 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
2072 | unsigned long flags, void (*ctor)(void *)) | |
2073 | { | |
2074 | struct kmem_cache *cachep; | |
2075 | ||
2076 | cachep = find_mergeable(size, align, flags, name, ctor); | |
2077 | if (cachep) { | |
2078 | cachep->refcount++; | |
2079 | ||
2080 | /* | |
2081 | * Adjust the object sizes so that we clear | |
2082 | * the complete object on kzalloc. | |
2083 | */ | |
2084 | cachep->object_size = max_t(int, cachep->object_size, size); | |
2085 | } | |
2086 | return cachep; | |
2087 | } | |
2088 | ||
1da177e4 | 2089 | /** |
039363f3 | 2090 | * __kmem_cache_create - Create a cache. |
a755b76a | 2091 | * @cachep: cache management descriptor |
1da177e4 | 2092 | * @flags: SLAB flags |
1da177e4 LT |
2093 | * |
2094 | * Returns a ptr to the cache on success, NULL on failure. | |
2095 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2096 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2097 | * |
1da177e4 LT |
2098 | * The flags are |
2099 | * | |
2100 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2101 | * to catch references to uninitialised memory. | |
2102 | * | |
2103 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2104 | * for buffer overruns. | |
2105 | * | |
1da177e4 LT |
2106 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2107 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2108 | * as davem. | |
2109 | */ | |
278b1bb1 | 2110 | int |
8a13a4cc | 2111 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2112 | { |
d4a5fca5 DR |
2113 | size_t left_over, freelist_size; |
2114 | size_t ralign = BYTES_PER_WORD; | |
83b519e8 | 2115 | gfp_t gfp; |
278b1bb1 | 2116 | int err; |
8a13a4cc | 2117 | size_t size = cachep->size; |
1da177e4 | 2118 | |
1da177e4 | 2119 | #if DEBUG |
1da177e4 LT |
2120 | #if FORCED_DEBUG |
2121 | /* | |
2122 | * Enable redzoning and last user accounting, except for caches with | |
2123 | * large objects, if the increased size would increase the object size | |
2124 | * above the next power of two: caches with object sizes just above a | |
2125 | * power of two have a significant amount of internal fragmentation. | |
2126 | */ | |
87a927c7 DW |
2127 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2128 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2129 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2130 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2131 | flags |= SLAB_POISON; | |
2132 | #endif | |
2133 | if (flags & SLAB_DESTROY_BY_RCU) | |
2134 | BUG_ON(flags & SLAB_POISON); | |
2135 | #endif | |
1da177e4 | 2136 | |
a737b3e2 AM |
2137 | /* |
2138 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2139 | * unaligned accesses for some archs when redzoning is used, and makes |
2140 | * sure any on-slab bufctl's are also correctly aligned. | |
2141 | */ | |
b28a02de PE |
2142 | if (size & (BYTES_PER_WORD - 1)) { |
2143 | size += (BYTES_PER_WORD - 1); | |
2144 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2145 | } |
2146 | ||
87a927c7 DW |
2147 | if (flags & SLAB_RED_ZONE) { |
2148 | ralign = REDZONE_ALIGN; | |
2149 | /* If redzoning, ensure that the second redzone is suitably | |
2150 | * aligned, by adjusting the object size accordingly. */ | |
2151 | size += REDZONE_ALIGN - 1; | |
2152 | size &= ~(REDZONE_ALIGN - 1); | |
2153 | } | |
ca5f9703 | 2154 | |
a44b56d3 | 2155 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2156 | if (ralign < cachep->align) { |
2157 | ralign = cachep->align; | |
1da177e4 | 2158 | } |
3ff84a7f PE |
2159 | /* disable debug if necessary */ |
2160 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2161 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2162 | /* |
ca5f9703 | 2163 | * 4) Store it. |
1da177e4 | 2164 | */ |
8a13a4cc | 2165 | cachep->align = ralign; |
1da177e4 | 2166 | |
83b519e8 PE |
2167 | if (slab_is_available()) |
2168 | gfp = GFP_KERNEL; | |
2169 | else | |
2170 | gfp = GFP_NOWAIT; | |
2171 | ||
1da177e4 | 2172 | #if DEBUG |
1da177e4 | 2173 | |
ca5f9703 PE |
2174 | /* |
2175 | * Both debugging options require word-alignment which is calculated | |
2176 | * into align above. | |
2177 | */ | |
1da177e4 | 2178 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2179 | /* add space for red zone words */ |
3ff84a7f PE |
2180 | cachep->obj_offset += sizeof(unsigned long long); |
2181 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2182 | } |
2183 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2184 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2185 | * the real object. But if the second red zone needs to be |
2186 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2187 | */ |
87a927c7 DW |
2188 | if (flags & SLAB_RED_ZONE) |
2189 | size += REDZONE_ALIGN; | |
2190 | else | |
2191 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2192 | } |
2193 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
03a2d2a3 JK |
2194 | /* |
2195 | * To activate debug pagealloc, off-slab management is necessary | |
2196 | * requirement. In early phase of initialization, small sized slab | |
2197 | * doesn't get initialized so it would not be possible. So, we need | |
2198 | * to check size >= 256. It guarantees that all necessary small | |
2199 | * sized slab is initialized in current slab initialization sequence. | |
2200 | */ | |
2201 | if (!slab_early_init && size >= kmalloc_size(INDEX_NODE) && | |
2202 | size >= 256 && cachep->object_size > cache_line_size() && | |
2203 | ALIGN(size, cachep->align) < PAGE_SIZE) { | |
608da7e3 | 2204 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); |
1da177e4 LT |
2205 | size = PAGE_SIZE; |
2206 | } | |
2207 | #endif | |
2208 | #endif | |
2209 | ||
e0a42726 IM |
2210 | /* |
2211 | * Determine if the slab management is 'on' or 'off' slab. | |
2212 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2213 | * it too early on. Always use on-slab management when |
2214 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2215 | */ |
d4322d88 | 2216 | if (size >= OFF_SLAB_MIN_SIZE && !slab_early_init && |
e7cb55b9 | 2217 | !(flags & SLAB_NOLEAKTRACE)) |
1da177e4 LT |
2218 | /* |
2219 | * Size is large, assume best to place the slab management obj | |
2220 | * off-slab (should allow better packing of objs). | |
2221 | */ | |
2222 | flags |= CFLGS_OFF_SLAB; | |
2223 | ||
8a13a4cc | 2224 | size = ALIGN(size, cachep->align); |
f315e3fa JK |
2225 | /* |
2226 | * We should restrict the number of objects in a slab to implement | |
2227 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2228 | */ | |
2229 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2230 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
1da177e4 | 2231 | |
8a13a4cc | 2232 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2233 | |
8a13a4cc | 2234 | if (!cachep->num) |
278b1bb1 | 2235 | return -E2BIG; |
1da177e4 | 2236 | |
03787301 | 2237 | freelist_size = calculate_freelist_size(cachep->num, cachep->align); |
1da177e4 LT |
2238 | |
2239 | /* | |
2240 | * If the slab has been placed off-slab, and we have enough space then | |
2241 | * move it on-slab. This is at the expense of any extra colouring. | |
2242 | */ | |
8456a648 | 2243 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
1da177e4 | 2244 | flags &= ~CFLGS_OFF_SLAB; |
8456a648 | 2245 | left_over -= freelist_size; |
1da177e4 LT |
2246 | } |
2247 | ||
2248 | if (flags & CFLGS_OFF_SLAB) { | |
2249 | /* really off slab. No need for manual alignment */ | |
03787301 | 2250 | freelist_size = calculate_freelist_size(cachep->num, 0); |
67461365 RL |
2251 | |
2252 | #ifdef CONFIG_PAGE_POISONING | |
2253 | /* If we're going to use the generic kernel_map_pages() | |
2254 | * poisoning, then it's going to smash the contents of | |
2255 | * the redzone and userword anyhow, so switch them off. | |
2256 | */ | |
2257 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2258 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2259 | #endif | |
1da177e4 LT |
2260 | } |
2261 | ||
2262 | cachep->colour_off = cache_line_size(); | |
2263 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2264 | if (cachep->colour_off < cachep->align) |
2265 | cachep->colour_off = cachep->align; | |
b28a02de | 2266 | cachep->colour = left_over / cachep->colour_off; |
8456a648 | 2267 | cachep->freelist_size = freelist_size; |
1da177e4 | 2268 | cachep->flags = flags; |
a57a4988 | 2269 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2270 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2271 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2272 | cachep->size = size; |
6a2d7a95 | 2273 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2274 | |
e5ac9c5a | 2275 | if (flags & CFLGS_OFF_SLAB) { |
8456a648 | 2276 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
e5ac9c5a | 2277 | /* |
5f0985bb | 2278 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
e5ac9c5a | 2279 | * But since we go off slab only for object size greater than |
d4322d88 | 2280 | * OFF_SLAB_MIN_SIZE, and kmalloc_{dma,}_caches get created |
5f0985bb | 2281 | * in ascending order,this should not happen at all. |
e5ac9c5a RT |
2282 | * But leave a BUG_ON for some lucky dude. |
2283 | */ | |
8456a648 | 2284 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
e5ac9c5a | 2285 | } |
1da177e4 | 2286 | |
278b1bb1 CL |
2287 | err = setup_cpu_cache(cachep, gfp); |
2288 | if (err) { | |
12c3667f | 2289 | __kmem_cache_shutdown(cachep); |
278b1bb1 | 2290 | return err; |
2ed3a4ef | 2291 | } |
1da177e4 | 2292 | |
278b1bb1 | 2293 | return 0; |
1da177e4 | 2294 | } |
1da177e4 LT |
2295 | |
2296 | #if DEBUG | |
2297 | static void check_irq_off(void) | |
2298 | { | |
2299 | BUG_ON(!irqs_disabled()); | |
2300 | } | |
2301 | ||
2302 | static void check_irq_on(void) | |
2303 | { | |
2304 | BUG_ON(irqs_disabled()); | |
2305 | } | |
2306 | ||
343e0d7a | 2307 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2308 | { |
2309 | #ifdef CONFIG_SMP | |
2310 | check_irq_off(); | |
18bf8541 | 2311 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2312 | #endif |
2313 | } | |
e498be7d | 2314 | |
343e0d7a | 2315 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2316 | { |
2317 | #ifdef CONFIG_SMP | |
2318 | check_irq_off(); | |
18bf8541 | 2319 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2320 | #endif |
2321 | } | |
2322 | ||
1da177e4 LT |
2323 | #else |
2324 | #define check_irq_off() do { } while(0) | |
2325 | #define check_irq_on() do { } while(0) | |
2326 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2327 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2328 | #endif |
2329 | ||
ce8eb6c4 | 2330 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2331 | struct array_cache *ac, |
2332 | int force, int node); | |
2333 | ||
1da177e4 LT |
2334 | static void do_drain(void *arg) |
2335 | { | |
a737b3e2 | 2336 | struct kmem_cache *cachep = arg; |
1da177e4 | 2337 | struct array_cache *ac; |
7d6e6d09 | 2338 | int node = numa_mem_id(); |
18bf8541 | 2339 | struct kmem_cache_node *n; |
97654dfa | 2340 | LIST_HEAD(list); |
1da177e4 LT |
2341 | |
2342 | check_irq_off(); | |
9a2dba4b | 2343 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2344 | n = get_node(cachep, node); |
2345 | spin_lock(&n->list_lock); | |
97654dfa | 2346 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2347 | spin_unlock(&n->list_lock); |
97654dfa | 2348 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2349 | ac->avail = 0; |
2350 | } | |
2351 | ||
343e0d7a | 2352 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2353 | { |
ce8eb6c4 | 2354 | struct kmem_cache_node *n; |
e498be7d CL |
2355 | int node; |
2356 | ||
15c8b6c1 | 2357 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2358 | check_irq_on(); |
18bf8541 CL |
2359 | for_each_kmem_cache_node(cachep, node, n) |
2360 | if (n->alien) | |
ce8eb6c4 | 2361 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2362 | |
18bf8541 CL |
2363 | for_each_kmem_cache_node(cachep, node, n) |
2364 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2365 | } |
2366 | ||
ed11d9eb CL |
2367 | /* |
2368 | * Remove slabs from the list of free slabs. | |
2369 | * Specify the number of slabs to drain in tofree. | |
2370 | * | |
2371 | * Returns the actual number of slabs released. | |
2372 | */ | |
2373 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2374 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2375 | { |
ed11d9eb CL |
2376 | struct list_head *p; |
2377 | int nr_freed; | |
8456a648 | 2378 | struct page *page; |
1da177e4 | 2379 | |
ed11d9eb | 2380 | nr_freed = 0; |
ce8eb6c4 | 2381 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2382 | |
ce8eb6c4 CL |
2383 | spin_lock_irq(&n->list_lock); |
2384 | p = n->slabs_free.prev; | |
2385 | if (p == &n->slabs_free) { | |
2386 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2387 | goto out; |
2388 | } | |
1da177e4 | 2389 | |
8456a648 | 2390 | page = list_entry(p, struct page, lru); |
1da177e4 | 2391 | #if DEBUG |
8456a648 | 2392 | BUG_ON(page->active); |
1da177e4 | 2393 | #endif |
8456a648 | 2394 | list_del(&page->lru); |
ed11d9eb CL |
2395 | /* |
2396 | * Safe to drop the lock. The slab is no longer linked | |
2397 | * to the cache. | |
2398 | */ | |
ce8eb6c4 CL |
2399 | n->free_objects -= cache->num; |
2400 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2401 | slab_destroy(cache, page); |
ed11d9eb | 2402 | nr_freed++; |
1da177e4 | 2403 | } |
ed11d9eb CL |
2404 | out: |
2405 | return nr_freed; | |
1da177e4 LT |
2406 | } |
2407 | ||
d6e0b7fa | 2408 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2409 | { |
18bf8541 CL |
2410 | int ret = 0; |
2411 | int node; | |
ce8eb6c4 | 2412 | struct kmem_cache_node *n; |
e498be7d CL |
2413 | |
2414 | drain_cpu_caches(cachep); | |
2415 | ||
2416 | check_irq_on(); | |
18bf8541 | 2417 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2418 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2419 | |
ce8eb6c4 CL |
2420 | ret += !list_empty(&n->slabs_full) || |
2421 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2422 | } |
2423 | return (ret ? 1 : 0); | |
2424 | } | |
2425 | ||
945cf2b6 | 2426 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
1da177e4 | 2427 | { |
12c3667f | 2428 | int i; |
ce8eb6c4 | 2429 | struct kmem_cache_node *n; |
d6e0b7fa | 2430 | int rc = __kmem_cache_shrink(cachep, false); |
1da177e4 | 2431 | |
12c3667f CL |
2432 | if (rc) |
2433 | return rc; | |
1da177e4 | 2434 | |
bf0dea23 | 2435 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2436 | |
ce8eb6c4 | 2437 | /* NUMA: free the node structures */ |
18bf8541 CL |
2438 | for_each_kmem_cache_node(cachep, i, n) { |
2439 | kfree(n->shared); | |
2440 | free_alien_cache(n->alien); | |
2441 | kfree(n); | |
2442 | cachep->node[i] = NULL; | |
12c3667f CL |
2443 | } |
2444 | return 0; | |
1da177e4 | 2445 | } |
1da177e4 | 2446 | |
e5ac9c5a RT |
2447 | /* |
2448 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2449 | * |
2450 | * For a slab cache when the slab descriptor is off-slab, the | |
2451 | * slab descriptor can't come from the same cache which is being created, | |
2452 | * Because if it is the case, that means we defer the creation of | |
2453 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2454 | * And we eventually call down to __kmem_cache_create(), which | |
2455 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2456 | * This is a "chicken-and-egg" problem. | |
2457 | * | |
2458 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2459 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2460 | */ |
7e007355 | 2461 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2462 | struct page *page, int colour_off, |
2463 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2464 | { |
7e007355 | 2465 | void *freelist; |
0c3aa83e | 2466 | void *addr = page_address(page); |
b28a02de | 2467 | |
1da177e4 LT |
2468 | if (OFF_SLAB(cachep)) { |
2469 | /* Slab management obj is off-slab. */ | |
8456a648 | 2470 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2471 | local_flags, nodeid); |
8456a648 | 2472 | if (!freelist) |
1da177e4 LT |
2473 | return NULL; |
2474 | } else { | |
8456a648 JK |
2475 | freelist = addr + colour_off; |
2476 | colour_off += cachep->freelist_size; | |
1da177e4 | 2477 | } |
8456a648 JK |
2478 | page->active = 0; |
2479 | page->s_mem = addr + colour_off; | |
2480 | return freelist; | |
1da177e4 LT |
2481 | } |
2482 | ||
7cc68973 | 2483 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2484 | { |
a41adfaa | 2485 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2486 | } |
2487 | ||
2488 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2489 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2490 | { |
a41adfaa | 2491 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2492 | } |
2493 | ||
343e0d7a | 2494 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2495 | struct page *page) |
1da177e4 LT |
2496 | { |
2497 | int i; | |
2498 | ||
2499 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2500 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
2501 | #if DEBUG |
2502 | /* need to poison the objs? */ | |
2503 | if (cachep->flags & SLAB_POISON) | |
2504 | poison_obj(cachep, objp, POISON_FREE); | |
2505 | if (cachep->flags & SLAB_STORE_USER) | |
2506 | *dbg_userword(cachep, objp) = NULL; | |
2507 | ||
2508 | if (cachep->flags & SLAB_RED_ZONE) { | |
2509 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2510 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2511 | } | |
2512 | /* | |
a737b3e2 AM |
2513 | * Constructors are not allowed to allocate memory from the same |
2514 | * cache which they are a constructor for. Otherwise, deadlock. | |
2515 | * They must also be threaded. | |
1da177e4 LT |
2516 | */ |
2517 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2518 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2519 | |
2520 | if (cachep->flags & SLAB_RED_ZONE) { | |
2521 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2522 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2523 | " end of an object"); |
1da177e4 LT |
2524 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2525 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2526 | " start of an object"); |
1da177e4 | 2527 | } |
3b0efdfa | 2528 | if ((cachep->size % PAGE_SIZE) == 0 && |
a737b3e2 | 2529 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
b28a02de | 2530 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2531 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2532 | #else |
2533 | if (cachep->ctor) | |
51cc5068 | 2534 | cachep->ctor(objp); |
1da177e4 | 2535 | #endif |
03787301 | 2536 | set_obj_status(page, i, OBJECT_FREE); |
e5c58dfd | 2537 | set_free_obj(page, i, i); |
1da177e4 | 2538 | } |
1da177e4 LT |
2539 | } |
2540 | ||
343e0d7a | 2541 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2542 | { |
4b51d669 CL |
2543 | if (CONFIG_ZONE_DMA_FLAG) { |
2544 | if (flags & GFP_DMA) | |
a618e89f | 2545 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2546 | else |
a618e89f | 2547 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2548 | } |
1da177e4 LT |
2549 | } |
2550 | ||
8456a648 | 2551 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2552 | int nodeid) |
78d382d7 | 2553 | { |
b1cb0982 | 2554 | void *objp; |
78d382d7 | 2555 | |
e5c58dfd | 2556 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2557 | page->active++; |
78d382d7 | 2558 | #if DEBUG |
1ea991b0 | 2559 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2560 | #endif |
78d382d7 MD |
2561 | |
2562 | return objp; | |
2563 | } | |
2564 | ||
8456a648 | 2565 | static void slab_put_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2566 | void *objp, int nodeid) |
78d382d7 | 2567 | { |
8456a648 | 2568 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2569 | #if DEBUG |
16025177 | 2570 | unsigned int i; |
b1cb0982 | 2571 | |
78d382d7 | 2572 | /* Verify that the slab belongs to the intended node */ |
1ea991b0 | 2573 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2574 | |
b1cb0982 | 2575 | /* Verify double free bug */ |
8456a648 | 2576 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2577 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2578 | printk(KERN_ERR "slab: double free detected in cache " |
2579 | "'%s', objp %p\n", cachep->name, objp); | |
2580 | BUG(); | |
2581 | } | |
78d382d7 MD |
2582 | } |
2583 | #endif | |
8456a648 | 2584 | page->active--; |
e5c58dfd | 2585 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2586 | } |
2587 | ||
4776874f PE |
2588 | /* |
2589 | * Map pages beginning at addr to the given cache and slab. This is required | |
2590 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2591 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2592 | */ |
8456a648 | 2593 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2594 | void *freelist) |
1da177e4 | 2595 | { |
a57a4988 | 2596 | page->slab_cache = cache; |
8456a648 | 2597 | page->freelist = freelist; |
1da177e4 LT |
2598 | } |
2599 | ||
2600 | /* | |
2601 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2602 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2603 | */ | |
3c517a61 | 2604 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2605 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2606 | { |
7e007355 | 2607 | void *freelist; |
b28a02de PE |
2608 | size_t offset; |
2609 | gfp_t local_flags; | |
ce8eb6c4 | 2610 | struct kmem_cache_node *n; |
1da177e4 | 2611 | |
a737b3e2 AM |
2612 | /* |
2613 | * Be lazy and only check for valid flags here, keeping it out of the | |
2614 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2615 | */ |
c871ac4e AM |
2616 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2617 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2618 | BUG(); | |
2619 | } | |
6cb06229 | 2620 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2621 | |
ce8eb6c4 | 2622 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2623 | check_irq_off(); |
18bf8541 | 2624 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2625 | spin_lock(&n->list_lock); |
1da177e4 LT |
2626 | |
2627 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2628 | offset = n->colour_next; |
2629 | n->colour_next++; | |
2630 | if (n->colour_next >= cachep->colour) | |
2631 | n->colour_next = 0; | |
2632 | spin_unlock(&n->list_lock); | |
1da177e4 | 2633 | |
2e1217cf | 2634 | offset *= cachep->colour_off; |
1da177e4 | 2635 | |
d0164adc | 2636 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2637 | local_irq_enable(); |
2638 | ||
2639 | /* | |
2640 | * The test for missing atomic flag is performed here, rather than | |
2641 | * the more obvious place, simply to reduce the critical path length | |
2642 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2643 | * will eventually be caught here (where it matters). | |
2644 | */ | |
2645 | kmem_flagcheck(cachep, flags); | |
2646 | ||
a737b3e2 AM |
2647 | /* |
2648 | * Get mem for the objs. Attempt to allocate a physical page from | |
2649 | * 'nodeid'. | |
e498be7d | 2650 | */ |
0c3aa83e JK |
2651 | if (!page) |
2652 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2653 | if (!page) | |
1da177e4 LT |
2654 | goto failed; |
2655 | ||
2656 | /* Get slab management. */ | |
8456a648 | 2657 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2658 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2659 | if (!freelist) |
1da177e4 LT |
2660 | goto opps1; |
2661 | ||
8456a648 | 2662 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2663 | |
8456a648 | 2664 | cache_init_objs(cachep, page); |
1da177e4 | 2665 | |
d0164adc | 2666 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2667 | local_irq_disable(); |
2668 | check_irq_off(); | |
ce8eb6c4 | 2669 | spin_lock(&n->list_lock); |
1da177e4 LT |
2670 | |
2671 | /* Make slab active. */ | |
8456a648 | 2672 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2673 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2674 | n->free_objects += cachep->num; |
2675 | spin_unlock(&n->list_lock); | |
1da177e4 | 2676 | return 1; |
a737b3e2 | 2677 | opps1: |
0c3aa83e | 2678 | kmem_freepages(cachep, page); |
a737b3e2 | 2679 | failed: |
d0164adc | 2680 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2681 | local_irq_disable(); |
2682 | return 0; | |
2683 | } | |
2684 | ||
2685 | #if DEBUG | |
2686 | ||
2687 | /* | |
2688 | * Perform extra freeing checks: | |
2689 | * - detect bad pointers. | |
2690 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2691 | */ |
2692 | static void kfree_debugcheck(const void *objp) | |
2693 | { | |
1da177e4 LT |
2694 | if (!virt_addr_valid(objp)) { |
2695 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2696 | (unsigned long)objp); |
2697 | BUG(); | |
1da177e4 | 2698 | } |
1da177e4 LT |
2699 | } |
2700 | ||
58ce1fd5 PE |
2701 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2702 | { | |
b46b8f19 | 2703 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2704 | |
2705 | redzone1 = *dbg_redzone1(cache, obj); | |
2706 | redzone2 = *dbg_redzone2(cache, obj); | |
2707 | ||
2708 | /* | |
2709 | * Redzone is ok. | |
2710 | */ | |
2711 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2712 | return; | |
2713 | ||
2714 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2715 | slab_error(cache, "double free detected"); | |
2716 | else | |
2717 | slab_error(cache, "memory outside object was overwritten"); | |
2718 | ||
b46b8f19 | 2719 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2720 | obj, redzone1, redzone2); |
2721 | } | |
2722 | ||
343e0d7a | 2723 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2724 | unsigned long caller) |
1da177e4 | 2725 | { |
1da177e4 | 2726 | unsigned int objnr; |
8456a648 | 2727 | struct page *page; |
1da177e4 | 2728 | |
80cbd911 MW |
2729 | BUG_ON(virt_to_cache(objp) != cachep); |
2730 | ||
3dafccf2 | 2731 | objp -= obj_offset(cachep); |
1da177e4 | 2732 | kfree_debugcheck(objp); |
b49af68f | 2733 | page = virt_to_head_page(objp); |
1da177e4 | 2734 | |
1da177e4 | 2735 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2736 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2737 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2738 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2739 | } | |
2740 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2741 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2742 | |
8456a648 | 2743 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2744 | |
2745 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2746 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2747 | |
03787301 | 2748 | set_obj_status(page, objnr, OBJECT_FREE); |
1da177e4 LT |
2749 | if (cachep->flags & SLAB_POISON) { |
2750 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2751 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
7c0cb9c6 | 2752 | store_stackinfo(cachep, objp, caller); |
b28a02de | 2753 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2754 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2755 | } else { |
2756 | poison_obj(cachep, objp, POISON_FREE); | |
2757 | } | |
2758 | #else | |
2759 | poison_obj(cachep, objp, POISON_FREE); | |
2760 | #endif | |
2761 | } | |
2762 | return objp; | |
2763 | } | |
2764 | ||
1da177e4 LT |
2765 | #else |
2766 | #define kfree_debugcheck(x) do { } while(0) | |
2767 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2768 | #endif |
2769 | ||
072bb0aa MG |
2770 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2771 | bool force_refill) | |
1da177e4 LT |
2772 | { |
2773 | int batchcount; | |
ce8eb6c4 | 2774 | struct kmem_cache_node *n; |
1da177e4 | 2775 | struct array_cache *ac; |
1ca4cb24 PE |
2776 | int node; |
2777 | ||
1da177e4 | 2778 | check_irq_off(); |
7d6e6d09 | 2779 | node = numa_mem_id(); |
072bb0aa MG |
2780 | if (unlikely(force_refill)) |
2781 | goto force_grow; | |
2782 | retry: | |
9a2dba4b | 2783 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2784 | batchcount = ac->batchcount; |
2785 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2786 | /* |
2787 | * If there was little recent activity on this cache, then | |
2788 | * perform only a partial refill. Otherwise we could generate | |
2789 | * refill bouncing. | |
1da177e4 LT |
2790 | */ |
2791 | batchcount = BATCHREFILL_LIMIT; | |
2792 | } | |
18bf8541 | 2793 | n = get_node(cachep, node); |
e498be7d | 2794 | |
ce8eb6c4 CL |
2795 | BUG_ON(ac->avail > 0 || !n); |
2796 | spin_lock(&n->list_lock); | |
1da177e4 | 2797 | |
3ded175a | 2798 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2799 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2800 | n->shared->touched = 1; | |
3ded175a | 2801 | goto alloc_done; |
44b57f1c | 2802 | } |
3ded175a | 2803 | |
1da177e4 LT |
2804 | while (batchcount > 0) { |
2805 | struct list_head *entry; | |
8456a648 | 2806 | struct page *page; |
1da177e4 | 2807 | /* Get slab alloc is to come from. */ |
ce8eb6c4 CL |
2808 | entry = n->slabs_partial.next; |
2809 | if (entry == &n->slabs_partial) { | |
2810 | n->free_touched = 1; | |
2811 | entry = n->slabs_free.next; | |
2812 | if (entry == &n->slabs_free) | |
1da177e4 LT |
2813 | goto must_grow; |
2814 | } | |
2815 | ||
8456a648 | 2816 | page = list_entry(entry, struct page, lru); |
1da177e4 | 2817 | check_spinlock_acquired(cachep); |
714b8171 PE |
2818 | |
2819 | /* | |
2820 | * The slab was either on partial or free list so | |
2821 | * there must be at least one object available for | |
2822 | * allocation. | |
2823 | */ | |
8456a648 | 2824 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2825 | |
8456a648 | 2826 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2827 | STATS_INC_ALLOCED(cachep); |
2828 | STATS_INC_ACTIVE(cachep); | |
2829 | STATS_SET_HIGH(cachep); | |
2830 | ||
8456a648 | 2831 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page, |
072bb0aa | 2832 | node)); |
1da177e4 | 2833 | } |
1da177e4 LT |
2834 | |
2835 | /* move slabp to correct slabp list: */ | |
8456a648 JK |
2836 | list_del(&page->lru); |
2837 | if (page->active == cachep->num) | |
34bf6ef9 | 2838 | list_add(&page->lru, &n->slabs_full); |
1da177e4 | 2839 | else |
34bf6ef9 | 2840 | list_add(&page->lru, &n->slabs_partial); |
1da177e4 LT |
2841 | } |
2842 | ||
a737b3e2 | 2843 | must_grow: |
ce8eb6c4 | 2844 | n->free_objects -= ac->avail; |
a737b3e2 | 2845 | alloc_done: |
ce8eb6c4 | 2846 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2847 | |
2848 | if (unlikely(!ac->avail)) { | |
2849 | int x; | |
072bb0aa | 2850 | force_grow: |
4167e9b2 | 2851 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2852 | |
a737b3e2 | 2853 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2854 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2855 | node = numa_mem_id(); |
072bb0aa MG |
2856 | |
2857 | /* no objects in sight? abort */ | |
2858 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2859 | return NULL; |
2860 | ||
a737b3e2 | 2861 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2862 | goto retry; |
2863 | } | |
2864 | ac->touched = 1; | |
072bb0aa MG |
2865 | |
2866 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2867 | } |
2868 | ||
a737b3e2 AM |
2869 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2870 | gfp_t flags) | |
1da177e4 | 2871 | { |
d0164adc | 2872 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2873 | #if DEBUG |
2874 | kmem_flagcheck(cachep, flags); | |
2875 | #endif | |
2876 | } | |
2877 | ||
2878 | #if DEBUG | |
a737b3e2 | 2879 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2880 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2881 | { |
03787301 JK |
2882 | struct page *page; |
2883 | ||
b28a02de | 2884 | if (!objp) |
1da177e4 | 2885 | return objp; |
b28a02de | 2886 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2887 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3b0efdfa | 2888 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 2889 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2890 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
2891 | else |
2892 | check_poison_obj(cachep, objp); | |
2893 | #else | |
2894 | check_poison_obj(cachep, objp); | |
2895 | #endif | |
2896 | poison_obj(cachep, objp, POISON_INUSE); | |
2897 | } | |
2898 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2899 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2900 | |
2901 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2902 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2903 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2904 | slab_error(cachep, "double free, or memory outside" | |
2905 | " object was overwritten"); | |
b28a02de | 2906 | printk(KERN_ERR |
b46b8f19 | 2907 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2908 | objp, *dbg_redzone1(cachep, objp), |
2909 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2910 | } |
2911 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2912 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2913 | } | |
03787301 JK |
2914 | |
2915 | page = virt_to_head_page(objp); | |
2916 | set_obj_status(page, obj_to_index(cachep, page, objp), OBJECT_ACTIVE); | |
3dafccf2 | 2917 | objp += obj_offset(cachep); |
4f104934 | 2918 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2919 | cachep->ctor(objp); |
7ea466f2 TH |
2920 | if (ARCH_SLAB_MINALIGN && |
2921 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 2922 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2923 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2924 | } |
1da177e4 LT |
2925 | return objp; |
2926 | } | |
2927 | #else | |
2928 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2929 | #endif | |
2930 | ||
773ff60e | 2931 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 | 2932 | { |
8a9c61d4 | 2933 | if (unlikely(cachep == kmem_cache)) |
773ff60e | 2934 | return false; |
8a8b6502 | 2935 | |
8c138bc0 | 2936 | return should_failslab(cachep->object_size, flags, cachep->flags); |
8a8b6502 AM |
2937 | } |
2938 | ||
343e0d7a | 2939 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2940 | { |
b28a02de | 2941 | void *objp; |
1da177e4 | 2942 | struct array_cache *ac; |
072bb0aa | 2943 | bool force_refill = false; |
1da177e4 | 2944 | |
5c382300 | 2945 | check_irq_off(); |
8a8b6502 | 2946 | |
9a2dba4b | 2947 | ac = cpu_cache_get(cachep); |
1da177e4 | 2948 | if (likely(ac->avail)) { |
1da177e4 | 2949 | ac->touched = 1; |
072bb0aa MG |
2950 | objp = ac_get_obj(cachep, ac, flags, false); |
2951 | ||
ddbf2e83 | 2952 | /* |
072bb0aa MG |
2953 | * Allow for the possibility all avail objects are not allowed |
2954 | * by the current flags | |
ddbf2e83 | 2955 | */ |
072bb0aa MG |
2956 | if (objp) { |
2957 | STATS_INC_ALLOCHIT(cachep); | |
2958 | goto out; | |
2959 | } | |
2960 | force_refill = true; | |
1da177e4 | 2961 | } |
072bb0aa MG |
2962 | |
2963 | STATS_INC_ALLOCMISS(cachep); | |
2964 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
2965 | /* | |
2966 | * the 'ac' may be updated by cache_alloc_refill(), | |
2967 | * and kmemleak_erase() requires its correct value. | |
2968 | */ | |
2969 | ac = cpu_cache_get(cachep); | |
2970 | ||
2971 | out: | |
d5cff635 CM |
2972 | /* |
2973 | * To avoid a false negative, if an object that is in one of the | |
2974 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2975 | * treat the array pointers as a reference to the object. | |
2976 | */ | |
f3d8b53a O |
2977 | if (objp) |
2978 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2979 | return objp; |
2980 | } | |
2981 | ||
e498be7d | 2982 | #ifdef CONFIG_NUMA |
c61afb18 | 2983 | /* |
2ad654bc | 2984 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2985 | * |
2986 | * If we are in_interrupt, then process context, including cpusets and | |
2987 | * mempolicy, may not apply and should not be used for allocation policy. | |
2988 | */ | |
2989 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
2990 | { | |
2991 | int nid_alloc, nid_here; | |
2992 | ||
765c4507 | 2993 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 2994 | return NULL; |
7d6e6d09 | 2995 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 2996 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 2997 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 2998 | else if (current->mempolicy) |
2a389610 | 2999 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3000 | if (nid_alloc != nid_here) |
8b98c169 | 3001 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3002 | return NULL; |
3003 | } | |
3004 | ||
765c4507 CL |
3005 | /* |
3006 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3007 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3008 | * available node for available objects. If that fails then we |
3c517a61 CL |
3009 | * perform an allocation without specifying a node. This allows the page |
3010 | * allocator to do its reclaim / fallback magic. We then insert the | |
3011 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3012 | */ |
8c8cc2c1 | 3013 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3014 | { |
8c8cc2c1 PE |
3015 | struct zonelist *zonelist; |
3016 | gfp_t local_flags; | |
dd1a239f | 3017 | struct zoneref *z; |
54a6eb5c MG |
3018 | struct zone *zone; |
3019 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3020 | void *obj = NULL; |
3c517a61 | 3021 | int nid; |
cc9a6c87 | 3022 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3023 | |
3024 | if (flags & __GFP_THISNODE) | |
3025 | return NULL; | |
3026 | ||
6cb06229 | 3027 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3028 | |
cc9a6c87 | 3029 | retry_cpuset: |
d26914d1 | 3030 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3031 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3032 | |
3c517a61 CL |
3033 | retry: |
3034 | /* | |
3035 | * Look through allowed nodes for objects available | |
3036 | * from existing per node queues. | |
3037 | */ | |
54a6eb5c MG |
3038 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3039 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3040 | |
061d7074 | 3041 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3042 | get_node(cache, nid) && |
3043 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3044 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3045 | gfp_exact_node(flags), nid); |
481c5346 CL |
3046 | if (obj) |
3047 | break; | |
3048 | } | |
3c517a61 CL |
3049 | } |
3050 | ||
cfce6604 | 3051 | if (!obj) { |
3c517a61 CL |
3052 | /* |
3053 | * This allocation will be performed within the constraints | |
3054 | * of the current cpuset / memory policy requirements. | |
3055 | * We may trigger various forms of reclaim on the allowed | |
3056 | * set and go into memory reserves if necessary. | |
3057 | */ | |
0c3aa83e JK |
3058 | struct page *page; |
3059 | ||
d0164adc | 3060 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
3061 | local_irq_enable(); |
3062 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3063 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 3064 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 3065 | local_irq_disable(); |
0c3aa83e | 3066 | if (page) { |
3c517a61 CL |
3067 | /* |
3068 | * Insert into the appropriate per node queues | |
3069 | */ | |
0c3aa83e JK |
3070 | nid = page_to_nid(page); |
3071 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 3072 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3073 | gfp_exact_node(flags), nid); |
3c517a61 CL |
3074 | if (!obj) |
3075 | /* | |
3076 | * Another processor may allocate the | |
3077 | * objects in the slab since we are | |
3078 | * not holding any locks. | |
3079 | */ | |
3080 | goto retry; | |
3081 | } else { | |
b6a60451 | 3082 | /* cache_grow already freed obj */ |
3c517a61 CL |
3083 | obj = NULL; |
3084 | } | |
3085 | } | |
aedb0eb1 | 3086 | } |
cc9a6c87 | 3087 | |
d26914d1 | 3088 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3089 | goto retry_cpuset; |
765c4507 CL |
3090 | return obj; |
3091 | } | |
3092 | ||
e498be7d CL |
3093 | /* |
3094 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3095 | */ |
8b98c169 | 3096 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3097 | int nodeid) |
e498be7d CL |
3098 | { |
3099 | struct list_head *entry; | |
8456a648 | 3100 | struct page *page; |
ce8eb6c4 | 3101 | struct kmem_cache_node *n; |
b28a02de | 3102 | void *obj; |
b28a02de PE |
3103 | int x; |
3104 | ||
7c3fbbdd | 3105 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3106 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3107 | BUG_ON(!n); |
b28a02de | 3108 | |
a737b3e2 | 3109 | retry: |
ca3b9b91 | 3110 | check_irq_off(); |
ce8eb6c4 CL |
3111 | spin_lock(&n->list_lock); |
3112 | entry = n->slabs_partial.next; | |
3113 | if (entry == &n->slabs_partial) { | |
3114 | n->free_touched = 1; | |
3115 | entry = n->slabs_free.next; | |
3116 | if (entry == &n->slabs_free) | |
b28a02de PE |
3117 | goto must_grow; |
3118 | } | |
3119 | ||
8456a648 | 3120 | page = list_entry(entry, struct page, lru); |
b28a02de | 3121 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3122 | |
3123 | STATS_INC_NODEALLOCS(cachep); | |
3124 | STATS_INC_ACTIVE(cachep); | |
3125 | STATS_SET_HIGH(cachep); | |
3126 | ||
8456a648 | 3127 | BUG_ON(page->active == cachep->num); |
b28a02de | 3128 | |
8456a648 | 3129 | obj = slab_get_obj(cachep, page, nodeid); |
ce8eb6c4 | 3130 | n->free_objects--; |
b28a02de | 3131 | /* move slabp to correct slabp list: */ |
8456a648 | 3132 | list_del(&page->lru); |
b28a02de | 3133 | |
8456a648 JK |
3134 | if (page->active == cachep->num) |
3135 | list_add(&page->lru, &n->slabs_full); | |
a737b3e2 | 3136 | else |
8456a648 | 3137 | list_add(&page->lru, &n->slabs_partial); |
e498be7d | 3138 | |
ce8eb6c4 | 3139 | spin_unlock(&n->list_lock); |
b28a02de | 3140 | goto done; |
e498be7d | 3141 | |
a737b3e2 | 3142 | must_grow: |
ce8eb6c4 | 3143 | spin_unlock(&n->list_lock); |
4167e9b2 | 3144 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3145 | if (x) |
3146 | goto retry; | |
1da177e4 | 3147 | |
8c8cc2c1 | 3148 | return fallback_alloc(cachep, flags); |
e498be7d | 3149 | |
a737b3e2 | 3150 | done: |
b28a02de | 3151 | return obj; |
e498be7d | 3152 | } |
8c8cc2c1 | 3153 | |
8c8cc2c1 | 3154 | static __always_inline void * |
48356303 | 3155 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3156 | unsigned long caller) |
8c8cc2c1 PE |
3157 | { |
3158 | unsigned long save_flags; | |
3159 | void *ptr; | |
7d6e6d09 | 3160 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3161 | |
dcce284a | 3162 | flags &= gfp_allowed_mask; |
7e85ee0c | 3163 | |
cf40bd16 NP |
3164 | lockdep_trace_alloc(flags); |
3165 | ||
773ff60e | 3166 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3167 | return NULL; |
3168 | ||
d79923fa GC |
3169 | cachep = memcg_kmem_get_cache(cachep, flags); |
3170 | ||
8c8cc2c1 PE |
3171 | cache_alloc_debugcheck_before(cachep, flags); |
3172 | local_irq_save(save_flags); | |
3173 | ||
eacbbae3 | 3174 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3175 | nodeid = slab_node; |
8c8cc2c1 | 3176 | |
18bf8541 | 3177 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3178 | /* Node not bootstrapped yet */ |
3179 | ptr = fallback_alloc(cachep, flags); | |
3180 | goto out; | |
3181 | } | |
3182 | ||
7d6e6d09 | 3183 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3184 | /* |
3185 | * Use the locally cached objects if possible. | |
3186 | * However ____cache_alloc does not allow fallback | |
3187 | * to other nodes. It may fail while we still have | |
3188 | * objects on other nodes available. | |
3189 | */ | |
3190 | ptr = ____cache_alloc(cachep, flags); | |
3191 | if (ptr) | |
3192 | goto out; | |
3193 | } | |
3194 | /* ___cache_alloc_node can fall back to other nodes */ | |
3195 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3196 | out: | |
3197 | local_irq_restore(save_flags); | |
3198 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
8c138bc0 | 3199 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3200 | flags); |
8c8cc2c1 | 3201 | |
5087c822 | 3202 | if (likely(ptr)) { |
8c138bc0 | 3203 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
5087c822 JP |
3204 | if (unlikely(flags & __GFP_ZERO)) |
3205 | memset(ptr, 0, cachep->object_size); | |
3206 | } | |
d07dbea4 | 3207 | |
8135be5a | 3208 | memcg_kmem_put_cache(cachep); |
8c8cc2c1 PE |
3209 | return ptr; |
3210 | } | |
3211 | ||
3212 | static __always_inline void * | |
3213 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3214 | { | |
3215 | void *objp; | |
3216 | ||
2ad654bc | 3217 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3218 | objp = alternate_node_alloc(cache, flags); |
3219 | if (objp) | |
3220 | goto out; | |
3221 | } | |
3222 | objp = ____cache_alloc(cache, flags); | |
3223 | ||
3224 | /* | |
3225 | * We may just have run out of memory on the local node. | |
3226 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3227 | */ | |
7d6e6d09 LS |
3228 | if (!objp) |
3229 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3230 | |
3231 | out: | |
3232 | return objp; | |
3233 | } | |
3234 | #else | |
3235 | ||
3236 | static __always_inline void * | |
3237 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3238 | { | |
3239 | return ____cache_alloc(cachep, flags); | |
3240 | } | |
3241 | ||
3242 | #endif /* CONFIG_NUMA */ | |
3243 | ||
3244 | static __always_inline void * | |
48356303 | 3245 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3246 | { |
3247 | unsigned long save_flags; | |
3248 | void *objp; | |
3249 | ||
dcce284a | 3250 | flags &= gfp_allowed_mask; |
7e85ee0c | 3251 | |
cf40bd16 NP |
3252 | lockdep_trace_alloc(flags); |
3253 | ||
773ff60e | 3254 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3255 | return NULL; |
3256 | ||
d79923fa GC |
3257 | cachep = memcg_kmem_get_cache(cachep, flags); |
3258 | ||
8c8cc2c1 PE |
3259 | cache_alloc_debugcheck_before(cachep, flags); |
3260 | local_irq_save(save_flags); | |
3261 | objp = __do_cache_alloc(cachep, flags); | |
3262 | local_irq_restore(save_flags); | |
3263 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
8c138bc0 | 3264 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3265 | flags); |
8c8cc2c1 PE |
3266 | prefetchw(objp); |
3267 | ||
5087c822 | 3268 | if (likely(objp)) { |
8c138bc0 | 3269 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
5087c822 JP |
3270 | if (unlikely(flags & __GFP_ZERO)) |
3271 | memset(objp, 0, cachep->object_size); | |
3272 | } | |
d07dbea4 | 3273 | |
8135be5a | 3274 | memcg_kmem_put_cache(cachep); |
8c8cc2c1 PE |
3275 | return objp; |
3276 | } | |
e498be7d CL |
3277 | |
3278 | /* | |
5f0985bb | 3279 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3280 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3281 | */ |
97654dfa JK |
3282 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3283 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3284 | { |
3285 | int i; | |
25c063fb | 3286 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3287 | |
3288 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3289 | void *objp; |
8456a648 | 3290 | struct page *page; |
1da177e4 | 3291 | |
072bb0aa MG |
3292 | clear_obj_pfmemalloc(&objpp[i]); |
3293 | objp = objpp[i]; | |
3294 | ||
8456a648 | 3295 | page = virt_to_head_page(objp); |
8456a648 | 3296 | list_del(&page->lru); |
ff69416e | 3297 | check_spinlock_acquired_node(cachep, node); |
8456a648 | 3298 | slab_put_obj(cachep, page, objp, node); |
1da177e4 | 3299 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3300 | n->free_objects++; |
1da177e4 LT |
3301 | |
3302 | /* fixup slab chains */ | |
8456a648 | 3303 | if (page->active == 0) { |
ce8eb6c4 CL |
3304 | if (n->free_objects > n->free_limit) { |
3305 | n->free_objects -= cachep->num; | |
97654dfa | 3306 | list_add_tail(&page->lru, list); |
1da177e4 | 3307 | } else { |
8456a648 | 3308 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3309 | } |
3310 | } else { | |
3311 | /* Unconditionally move a slab to the end of the | |
3312 | * partial list on free - maximum time for the | |
3313 | * other objects to be freed, too. | |
3314 | */ | |
8456a648 | 3315 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3316 | } |
3317 | } | |
3318 | } | |
3319 | ||
343e0d7a | 3320 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3321 | { |
3322 | int batchcount; | |
ce8eb6c4 | 3323 | struct kmem_cache_node *n; |
7d6e6d09 | 3324 | int node = numa_mem_id(); |
97654dfa | 3325 | LIST_HEAD(list); |
1da177e4 LT |
3326 | |
3327 | batchcount = ac->batchcount; | |
3328 | #if DEBUG | |
3329 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3330 | #endif | |
3331 | check_irq_off(); | |
18bf8541 | 3332 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3333 | spin_lock(&n->list_lock); |
3334 | if (n->shared) { | |
3335 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3336 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3337 | if (max) { |
3338 | if (batchcount > max) | |
3339 | batchcount = max; | |
e498be7d | 3340 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3341 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3342 | shared_array->avail += batchcount; |
3343 | goto free_done; | |
3344 | } | |
3345 | } | |
3346 | ||
97654dfa | 3347 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3348 | free_done: |
1da177e4 LT |
3349 | #if STATS |
3350 | { | |
3351 | int i = 0; | |
3352 | struct list_head *p; | |
3353 | ||
ce8eb6c4 CL |
3354 | p = n->slabs_free.next; |
3355 | while (p != &(n->slabs_free)) { | |
8456a648 | 3356 | struct page *page; |
1da177e4 | 3357 | |
8456a648 JK |
3358 | page = list_entry(p, struct page, lru); |
3359 | BUG_ON(page->active); | |
1da177e4 LT |
3360 | |
3361 | i++; | |
3362 | p = p->next; | |
3363 | } | |
3364 | STATS_SET_FREEABLE(cachep, i); | |
3365 | } | |
3366 | #endif | |
ce8eb6c4 | 3367 | spin_unlock(&n->list_lock); |
97654dfa | 3368 | slabs_destroy(cachep, &list); |
1da177e4 | 3369 | ac->avail -= batchcount; |
a737b3e2 | 3370 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3371 | } |
3372 | ||
3373 | /* | |
a737b3e2 AM |
3374 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3375 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3376 | */ |
a947eb95 | 3377 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3378 | unsigned long caller) |
1da177e4 | 3379 | { |
9a2dba4b | 3380 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3381 | |
3382 | check_irq_off(); | |
d5cff635 | 3383 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3384 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3385 | |
8c138bc0 | 3386 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3387 | |
1807a1aa SS |
3388 | /* |
3389 | * Skip calling cache_free_alien() when the platform is not numa. | |
3390 | * This will avoid cache misses that happen while accessing slabp (which | |
3391 | * is per page memory reference) to get nodeid. Instead use a global | |
3392 | * variable to skip the call, which is mostly likely to be present in | |
3393 | * the cache. | |
3394 | */ | |
b6e68bc1 | 3395 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3396 | return; |
3397 | ||
3d880194 | 3398 | if (ac->avail < ac->limit) { |
1da177e4 | 3399 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3400 | } else { |
3401 | STATS_INC_FREEMISS(cachep); | |
3402 | cache_flusharray(cachep, ac); | |
1da177e4 | 3403 | } |
42c8c99c | 3404 | |
072bb0aa | 3405 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3406 | } |
3407 | ||
3408 | /** | |
3409 | * kmem_cache_alloc - Allocate an object | |
3410 | * @cachep: The cache to allocate from. | |
3411 | * @flags: See kmalloc(). | |
3412 | * | |
3413 | * Allocate an object from this cache. The flags are only relevant | |
3414 | * if the cache has no available objects. | |
3415 | */ | |
343e0d7a | 3416 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3417 | { |
48356303 | 3418 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3419 | |
ca2b84cb | 3420 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3421 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3422 | |
3423 | return ret; | |
1da177e4 LT |
3424 | } |
3425 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3426 | ||
484748f0 CL |
3427 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
3428 | { | |
3429 | __kmem_cache_free_bulk(s, size, p); | |
3430 | } | |
3431 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3432 | ||
3433 | bool kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, | |
3434 | void **p) | |
3435 | { | |
3436 | return __kmem_cache_alloc_bulk(s, flags, size, p); | |
3437 | } | |
3438 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3439 | ||
0f24f128 | 3440 | #ifdef CONFIG_TRACING |
85beb586 | 3441 | void * |
4052147c | 3442 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3443 | { |
85beb586 SR |
3444 | void *ret; |
3445 | ||
48356303 | 3446 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3447 | |
3448 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3449 | size, cachep->size, flags); |
85beb586 | 3450 | return ret; |
36555751 | 3451 | } |
85beb586 | 3452 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3453 | #endif |
3454 | ||
1da177e4 | 3455 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3456 | /** |
3457 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3458 | * @cachep: The cache to allocate from. | |
3459 | * @flags: See kmalloc(). | |
3460 | * @nodeid: node number of the target node. | |
3461 | * | |
3462 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3463 | * node, which can improve the performance for cpu bound structures. | |
3464 | * | |
3465 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3466 | */ | |
8b98c169 CH |
3467 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3468 | { | |
48356303 | 3469 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3470 | |
ca2b84cb | 3471 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3472 | cachep->object_size, cachep->size, |
ca2b84cb | 3473 | flags, nodeid); |
36555751 EGM |
3474 | |
3475 | return ret; | |
8b98c169 | 3476 | } |
1da177e4 LT |
3477 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3478 | ||
0f24f128 | 3479 | #ifdef CONFIG_TRACING |
4052147c | 3480 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3481 | gfp_t flags, |
4052147c EG |
3482 | int nodeid, |
3483 | size_t size) | |
36555751 | 3484 | { |
85beb586 SR |
3485 | void *ret; |
3486 | ||
592f4145 | 3487 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3488 | |
85beb586 | 3489 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3490 | size, cachep->size, |
85beb586 SR |
3491 | flags, nodeid); |
3492 | return ret; | |
36555751 | 3493 | } |
85beb586 | 3494 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3495 | #endif |
3496 | ||
8b98c169 | 3497 | static __always_inline void * |
7c0cb9c6 | 3498 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3499 | { |
343e0d7a | 3500 | struct kmem_cache *cachep; |
97e2bde4 | 3501 | |
2c59dd65 | 3502 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3503 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3504 | return cachep; | |
4052147c | 3505 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3506 | } |
8b98c169 | 3507 | |
8b98c169 CH |
3508 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3509 | { | |
7c0cb9c6 | 3510 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3511 | } |
dbe5e69d | 3512 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3513 | |
3514 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3515 | int node, unsigned long caller) |
8b98c169 | 3516 | { |
7c0cb9c6 | 3517 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3518 | } |
3519 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3520 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3521 | |
3522 | /** | |
800590f5 | 3523 | * __do_kmalloc - allocate memory |
1da177e4 | 3524 | * @size: how many bytes of memory are required. |
800590f5 | 3525 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3526 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3527 | */ |
7fd6b141 | 3528 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3529 | unsigned long caller) |
1da177e4 | 3530 | { |
343e0d7a | 3531 | struct kmem_cache *cachep; |
36555751 | 3532 | void *ret; |
1da177e4 | 3533 | |
2c59dd65 | 3534 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3535 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3536 | return cachep; | |
48356303 | 3537 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3538 | |
7c0cb9c6 | 3539 | trace_kmalloc(caller, ret, |
3b0efdfa | 3540 | size, cachep->size, flags); |
36555751 EGM |
3541 | |
3542 | return ret; | |
7fd6b141 PE |
3543 | } |
3544 | ||
7fd6b141 PE |
3545 | void *__kmalloc(size_t size, gfp_t flags) |
3546 | { | |
7c0cb9c6 | 3547 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3548 | } |
3549 | EXPORT_SYMBOL(__kmalloc); | |
3550 | ||
ce71e27c | 3551 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3552 | { |
7c0cb9c6 | 3553 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3554 | } |
3555 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3556 | |
1da177e4 LT |
3557 | /** |
3558 | * kmem_cache_free - Deallocate an object | |
3559 | * @cachep: The cache the allocation was from. | |
3560 | * @objp: The previously allocated object. | |
3561 | * | |
3562 | * Free an object which was previously allocated from this | |
3563 | * cache. | |
3564 | */ | |
343e0d7a | 3565 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3566 | { |
3567 | unsigned long flags; | |
b9ce5ef4 GC |
3568 | cachep = cache_from_obj(cachep, objp); |
3569 | if (!cachep) | |
3570 | return; | |
1da177e4 LT |
3571 | |
3572 | local_irq_save(flags); | |
d97d476b | 3573 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3574 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3575 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3576 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3577 | local_irq_restore(flags); |
36555751 | 3578 | |
ca2b84cb | 3579 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3580 | } |
3581 | EXPORT_SYMBOL(kmem_cache_free); | |
3582 | ||
1da177e4 LT |
3583 | /** |
3584 | * kfree - free previously allocated memory | |
3585 | * @objp: pointer returned by kmalloc. | |
3586 | * | |
80e93eff PE |
3587 | * If @objp is NULL, no operation is performed. |
3588 | * | |
1da177e4 LT |
3589 | * Don't free memory not originally allocated by kmalloc() |
3590 | * or you will run into trouble. | |
3591 | */ | |
3592 | void kfree(const void *objp) | |
3593 | { | |
343e0d7a | 3594 | struct kmem_cache *c; |
1da177e4 LT |
3595 | unsigned long flags; |
3596 | ||
2121db74 PE |
3597 | trace_kfree(_RET_IP_, objp); |
3598 | ||
6cb8f913 | 3599 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3600 | return; |
3601 | local_irq_save(flags); | |
3602 | kfree_debugcheck(objp); | |
6ed5eb22 | 3603 | c = virt_to_cache(objp); |
8c138bc0 CL |
3604 | debug_check_no_locks_freed(objp, c->object_size); |
3605 | ||
3606 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3607 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3608 | local_irq_restore(flags); |
3609 | } | |
3610 | EXPORT_SYMBOL(kfree); | |
3611 | ||
e498be7d | 3612 | /* |
ce8eb6c4 | 3613 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3614 | */ |
5f0985bb | 3615 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3616 | { |
3617 | int node; | |
ce8eb6c4 | 3618 | struct kmem_cache_node *n; |
cafeb02e | 3619 | struct array_cache *new_shared; |
c8522a3a | 3620 | struct alien_cache **new_alien = NULL; |
e498be7d | 3621 | |
9c09a95c | 3622 | for_each_online_node(node) { |
cafeb02e | 3623 | |
b455def2 L |
3624 | if (use_alien_caches) { |
3625 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3626 | if (!new_alien) | |
3627 | goto fail; | |
3628 | } | |
cafeb02e | 3629 | |
63109846 ED |
3630 | new_shared = NULL; |
3631 | if (cachep->shared) { | |
3632 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3633 | cachep->shared*cachep->batchcount, |
83b519e8 | 3634 | 0xbaadf00d, gfp); |
63109846 ED |
3635 | if (!new_shared) { |
3636 | free_alien_cache(new_alien); | |
3637 | goto fail; | |
3638 | } | |
0718dc2a | 3639 | } |
cafeb02e | 3640 | |
18bf8541 | 3641 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3642 | if (n) { |
3643 | struct array_cache *shared = n->shared; | |
97654dfa | 3644 | LIST_HEAD(list); |
cafeb02e | 3645 | |
ce8eb6c4 | 3646 | spin_lock_irq(&n->list_lock); |
e498be7d | 3647 | |
cafeb02e | 3648 | if (shared) |
0718dc2a | 3649 | free_block(cachep, shared->entry, |
97654dfa | 3650 | shared->avail, node, &list); |
e498be7d | 3651 | |
ce8eb6c4 CL |
3652 | n->shared = new_shared; |
3653 | if (!n->alien) { | |
3654 | n->alien = new_alien; | |
e498be7d CL |
3655 | new_alien = NULL; |
3656 | } | |
ce8eb6c4 | 3657 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3658 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3659 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3660 | slabs_destroy(cachep, &list); |
cafeb02e | 3661 | kfree(shared); |
e498be7d CL |
3662 | free_alien_cache(new_alien); |
3663 | continue; | |
3664 | } | |
ce8eb6c4 CL |
3665 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3666 | if (!n) { | |
0718dc2a CL |
3667 | free_alien_cache(new_alien); |
3668 | kfree(new_shared); | |
e498be7d | 3669 | goto fail; |
0718dc2a | 3670 | } |
e498be7d | 3671 | |
ce8eb6c4 | 3672 | kmem_cache_node_init(n); |
5f0985bb JZ |
3673 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3674 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3675 | n->shared = new_shared; |
3676 | n->alien = new_alien; | |
3677 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3678 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3679 | cachep->node[node] = n; |
e498be7d | 3680 | } |
cafeb02e | 3681 | return 0; |
0718dc2a | 3682 | |
a737b3e2 | 3683 | fail: |
3b0efdfa | 3684 | if (!cachep->list.next) { |
0718dc2a CL |
3685 | /* Cache is not active yet. Roll back what we did */ |
3686 | node--; | |
3687 | while (node >= 0) { | |
18bf8541 CL |
3688 | n = get_node(cachep, node); |
3689 | if (n) { | |
ce8eb6c4 CL |
3690 | kfree(n->shared); |
3691 | free_alien_cache(n->alien); | |
3692 | kfree(n); | |
6a67368c | 3693 | cachep->node[node] = NULL; |
0718dc2a CL |
3694 | } |
3695 | node--; | |
3696 | } | |
3697 | } | |
cafeb02e | 3698 | return -ENOMEM; |
e498be7d CL |
3699 | } |
3700 | ||
18004c5d | 3701 | /* Always called with the slab_mutex held */ |
943a451a | 3702 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3703 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3704 | { |
bf0dea23 JK |
3705 | struct array_cache __percpu *cpu_cache, *prev; |
3706 | int cpu; | |
1da177e4 | 3707 | |
bf0dea23 JK |
3708 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3709 | if (!cpu_cache) | |
d2e7b7d0 SS |
3710 | return -ENOMEM; |
3711 | ||
bf0dea23 JK |
3712 | prev = cachep->cpu_cache; |
3713 | cachep->cpu_cache = cpu_cache; | |
3714 | kick_all_cpus_sync(); | |
e498be7d | 3715 | |
1da177e4 | 3716 | check_irq_on(); |
1da177e4 LT |
3717 | cachep->batchcount = batchcount; |
3718 | cachep->limit = limit; | |
e498be7d | 3719 | cachep->shared = shared; |
1da177e4 | 3720 | |
bf0dea23 JK |
3721 | if (!prev) |
3722 | goto alloc_node; | |
3723 | ||
3724 | for_each_online_cpu(cpu) { | |
97654dfa | 3725 | LIST_HEAD(list); |
18bf8541 CL |
3726 | int node; |
3727 | struct kmem_cache_node *n; | |
bf0dea23 | 3728 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3729 | |
bf0dea23 | 3730 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3731 | n = get_node(cachep, node); |
3732 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3733 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3734 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3735 | slabs_destroy(cachep, &list); |
1da177e4 | 3736 | } |
bf0dea23 JK |
3737 | free_percpu(prev); |
3738 | ||
3739 | alloc_node: | |
5f0985bb | 3740 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3741 | } |
3742 | ||
943a451a GC |
3743 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3744 | int batchcount, int shared, gfp_t gfp) | |
3745 | { | |
3746 | int ret; | |
426589f5 | 3747 | struct kmem_cache *c; |
943a451a GC |
3748 | |
3749 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3750 | ||
3751 | if (slab_state < FULL) | |
3752 | return ret; | |
3753 | ||
3754 | if ((ret < 0) || !is_root_cache(cachep)) | |
3755 | return ret; | |
3756 | ||
426589f5 VD |
3757 | lockdep_assert_held(&slab_mutex); |
3758 | for_each_memcg_cache(c, cachep) { | |
3759 | /* return value determined by the root cache only */ | |
3760 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3761 | } |
3762 | ||
3763 | return ret; | |
3764 | } | |
3765 | ||
18004c5d | 3766 | /* Called with slab_mutex held always */ |
83b519e8 | 3767 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3768 | { |
3769 | int err; | |
943a451a GC |
3770 | int limit = 0; |
3771 | int shared = 0; | |
3772 | int batchcount = 0; | |
3773 | ||
3774 | if (!is_root_cache(cachep)) { | |
3775 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3776 | limit = root->limit; | |
3777 | shared = root->shared; | |
3778 | batchcount = root->batchcount; | |
3779 | } | |
1da177e4 | 3780 | |
943a451a GC |
3781 | if (limit && shared && batchcount) |
3782 | goto skip_setup; | |
a737b3e2 AM |
3783 | /* |
3784 | * The head array serves three purposes: | |
1da177e4 LT |
3785 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3786 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3787 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3788 | * bufctl chains: array operations are cheaper. |
3789 | * The numbers are guessed, we should auto-tune as described by | |
3790 | * Bonwick. | |
3791 | */ | |
3b0efdfa | 3792 | if (cachep->size > 131072) |
1da177e4 | 3793 | limit = 1; |
3b0efdfa | 3794 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3795 | limit = 8; |
3b0efdfa | 3796 | else if (cachep->size > 1024) |
1da177e4 | 3797 | limit = 24; |
3b0efdfa | 3798 | else if (cachep->size > 256) |
1da177e4 LT |
3799 | limit = 54; |
3800 | else | |
3801 | limit = 120; | |
3802 | ||
a737b3e2 AM |
3803 | /* |
3804 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3805 | * allocation behaviour: Most allocs on one cpu, most free operations |
3806 | * on another cpu. For these cases, an efficient object passing between | |
3807 | * cpus is necessary. This is provided by a shared array. The array | |
3808 | * replaces Bonwick's magazine layer. | |
3809 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3810 | * to a larger limit. Thus disabled by default. | |
3811 | */ | |
3812 | shared = 0; | |
3b0efdfa | 3813 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3814 | shared = 8; |
1da177e4 LT |
3815 | |
3816 | #if DEBUG | |
a737b3e2 AM |
3817 | /* |
3818 | * With debugging enabled, large batchcount lead to excessively long | |
3819 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3820 | */ |
3821 | if (limit > 32) | |
3822 | limit = 32; | |
3823 | #endif | |
943a451a GC |
3824 | batchcount = (limit + 1) / 2; |
3825 | skip_setup: | |
3826 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3827 | if (err) |
3828 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3829 | cachep->name, -err); |
2ed3a4ef | 3830 | return err; |
1da177e4 LT |
3831 | } |
3832 | ||
1b55253a | 3833 | /* |
ce8eb6c4 CL |
3834 | * Drain an array if it contains any elements taking the node lock only if |
3835 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3836 | * if drain_array() is used on the shared array. |
1b55253a | 3837 | */ |
ce8eb6c4 | 3838 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3839 | struct array_cache *ac, int force, int node) |
1da177e4 | 3840 | { |
97654dfa | 3841 | LIST_HEAD(list); |
1da177e4 LT |
3842 | int tofree; |
3843 | ||
1b55253a CL |
3844 | if (!ac || !ac->avail) |
3845 | return; | |
1da177e4 LT |
3846 | if (ac->touched && !force) { |
3847 | ac->touched = 0; | |
b18e7e65 | 3848 | } else { |
ce8eb6c4 | 3849 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3850 | if (ac->avail) { |
3851 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3852 | if (tofree > ac->avail) | |
3853 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 3854 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
3855 | ac->avail -= tofree; |
3856 | memmove(ac->entry, &(ac->entry[tofree]), | |
3857 | sizeof(void *) * ac->avail); | |
3858 | } | |
ce8eb6c4 | 3859 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3860 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3861 | } |
3862 | } | |
3863 | ||
3864 | /** | |
3865 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3866 | * @w: work descriptor |
1da177e4 LT |
3867 | * |
3868 | * Called from workqueue/eventd every few seconds. | |
3869 | * Purpose: | |
3870 | * - clear the per-cpu caches for this CPU. | |
3871 | * - return freeable pages to the main free memory pool. | |
3872 | * | |
a737b3e2 AM |
3873 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3874 | * again on the next iteration. | |
1da177e4 | 3875 | */ |
7c5cae36 | 3876 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3877 | { |
7a7c381d | 3878 | struct kmem_cache *searchp; |
ce8eb6c4 | 3879 | struct kmem_cache_node *n; |
7d6e6d09 | 3880 | int node = numa_mem_id(); |
bf6aede7 | 3881 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3882 | |
18004c5d | 3883 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3884 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3885 | goto out; |
1da177e4 | 3886 | |
18004c5d | 3887 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3888 | check_irq_on(); |
3889 | ||
35386e3b | 3890 | /* |
ce8eb6c4 | 3891 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3892 | * have established with reasonable certainty that |
3893 | * we can do some work if the lock was obtained. | |
3894 | */ | |
18bf8541 | 3895 | n = get_node(searchp, node); |
35386e3b | 3896 | |
ce8eb6c4 | 3897 | reap_alien(searchp, n); |
1da177e4 | 3898 | |
ce8eb6c4 | 3899 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3900 | |
35386e3b CL |
3901 | /* |
3902 | * These are racy checks but it does not matter | |
3903 | * if we skip one check or scan twice. | |
3904 | */ | |
ce8eb6c4 | 3905 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3906 | goto next; |
1da177e4 | 3907 | |
5f0985bb | 3908 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3909 | |
ce8eb6c4 | 3910 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 3911 | |
ce8eb6c4 CL |
3912 | if (n->free_touched) |
3913 | n->free_touched = 0; | |
ed11d9eb CL |
3914 | else { |
3915 | int freed; | |
1da177e4 | 3916 | |
ce8eb6c4 | 3917 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3918 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3919 | STATS_ADD_REAPED(searchp, freed); | |
3920 | } | |
35386e3b | 3921 | next: |
1da177e4 LT |
3922 | cond_resched(); |
3923 | } | |
3924 | check_irq_on(); | |
18004c5d | 3925 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3926 | next_reap_node(); |
7c5cae36 | 3927 | out: |
a737b3e2 | 3928 | /* Set up the next iteration */ |
5f0985bb | 3929 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3930 | } |
3931 | ||
158a9624 | 3932 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3933 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3934 | { |
8456a648 | 3935 | struct page *page; |
b28a02de PE |
3936 | unsigned long active_objs; |
3937 | unsigned long num_objs; | |
3938 | unsigned long active_slabs = 0; | |
3939 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3940 | const char *name; |
1da177e4 | 3941 | char *error = NULL; |
e498be7d | 3942 | int node; |
ce8eb6c4 | 3943 | struct kmem_cache_node *n; |
1da177e4 | 3944 | |
1da177e4 LT |
3945 | active_objs = 0; |
3946 | num_slabs = 0; | |
18bf8541 | 3947 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 3948 | |
ca3b9b91 | 3949 | check_irq_on(); |
ce8eb6c4 | 3950 | spin_lock_irq(&n->list_lock); |
e498be7d | 3951 | |
8456a648 JK |
3952 | list_for_each_entry(page, &n->slabs_full, lru) { |
3953 | if (page->active != cachep->num && !error) | |
e498be7d CL |
3954 | error = "slabs_full accounting error"; |
3955 | active_objs += cachep->num; | |
3956 | active_slabs++; | |
3957 | } | |
8456a648 JK |
3958 | list_for_each_entry(page, &n->slabs_partial, lru) { |
3959 | if (page->active == cachep->num && !error) | |
106a74e1 | 3960 | error = "slabs_partial accounting error"; |
8456a648 | 3961 | if (!page->active && !error) |
106a74e1 | 3962 | error = "slabs_partial accounting error"; |
8456a648 | 3963 | active_objs += page->active; |
e498be7d CL |
3964 | active_slabs++; |
3965 | } | |
8456a648 JK |
3966 | list_for_each_entry(page, &n->slabs_free, lru) { |
3967 | if (page->active && !error) | |
106a74e1 | 3968 | error = "slabs_free accounting error"; |
e498be7d CL |
3969 | num_slabs++; |
3970 | } | |
ce8eb6c4 CL |
3971 | free_objects += n->free_objects; |
3972 | if (n->shared) | |
3973 | shared_avail += n->shared->avail; | |
e498be7d | 3974 | |
ce8eb6c4 | 3975 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 3976 | } |
b28a02de PE |
3977 | num_slabs += active_slabs; |
3978 | num_objs = num_slabs * cachep->num; | |
e498be7d | 3979 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
3980 | error = "free_objects accounting error"; |
3981 | ||
b28a02de | 3982 | name = cachep->name; |
1da177e4 LT |
3983 | if (error) |
3984 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
3985 | ||
0d7561c6 GC |
3986 | sinfo->active_objs = active_objs; |
3987 | sinfo->num_objs = num_objs; | |
3988 | sinfo->active_slabs = active_slabs; | |
3989 | sinfo->num_slabs = num_slabs; | |
3990 | sinfo->shared_avail = shared_avail; | |
3991 | sinfo->limit = cachep->limit; | |
3992 | sinfo->batchcount = cachep->batchcount; | |
3993 | sinfo->shared = cachep->shared; | |
3994 | sinfo->objects_per_slab = cachep->num; | |
3995 | sinfo->cache_order = cachep->gfporder; | |
3996 | } | |
3997 | ||
3998 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
3999 | { | |
1da177e4 | 4000 | #if STATS |
ce8eb6c4 | 4001 | { /* node stats */ |
1da177e4 LT |
4002 | unsigned long high = cachep->high_mark; |
4003 | unsigned long allocs = cachep->num_allocations; | |
4004 | unsigned long grown = cachep->grown; | |
4005 | unsigned long reaped = cachep->reaped; | |
4006 | unsigned long errors = cachep->errors; | |
4007 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4008 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4009 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4010 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4011 | |
e92dd4fd JP |
4012 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4013 | "%4lu %4lu %4lu %4lu %4lu", | |
4014 | allocs, high, grown, | |
4015 | reaped, errors, max_freeable, node_allocs, | |
4016 | node_frees, overflows); | |
1da177e4 LT |
4017 | } |
4018 | /* cpu stats */ | |
4019 | { | |
4020 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4021 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4022 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4023 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4024 | ||
4025 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4026 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4027 | } |
4028 | #endif | |
1da177e4 LT |
4029 | } |
4030 | ||
1da177e4 LT |
4031 | #define MAX_SLABINFO_WRITE 128 |
4032 | /** | |
4033 | * slabinfo_write - Tuning for the slab allocator | |
4034 | * @file: unused | |
4035 | * @buffer: user buffer | |
4036 | * @count: data length | |
4037 | * @ppos: unused | |
4038 | */ | |
b7454ad3 | 4039 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4040 | size_t count, loff_t *ppos) |
1da177e4 | 4041 | { |
b28a02de | 4042 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4043 | int limit, batchcount, shared, res; |
7a7c381d | 4044 | struct kmem_cache *cachep; |
b28a02de | 4045 | |
1da177e4 LT |
4046 | if (count > MAX_SLABINFO_WRITE) |
4047 | return -EINVAL; | |
4048 | if (copy_from_user(&kbuf, buffer, count)) | |
4049 | return -EFAULT; | |
b28a02de | 4050 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4051 | |
4052 | tmp = strchr(kbuf, ' '); | |
4053 | if (!tmp) | |
4054 | return -EINVAL; | |
4055 | *tmp = '\0'; | |
4056 | tmp++; | |
4057 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4058 | return -EINVAL; | |
4059 | ||
4060 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4061 | mutex_lock(&slab_mutex); |
1da177e4 | 4062 | res = -EINVAL; |
18004c5d | 4063 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4064 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4065 | if (limit < 1 || batchcount < 1 || |
4066 | batchcount > limit || shared < 0) { | |
e498be7d | 4067 | res = 0; |
1da177e4 | 4068 | } else { |
e498be7d | 4069 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4070 | batchcount, shared, |
4071 | GFP_KERNEL); | |
1da177e4 LT |
4072 | } |
4073 | break; | |
4074 | } | |
4075 | } | |
18004c5d | 4076 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4077 | if (res >= 0) |
4078 | res = count; | |
4079 | return res; | |
4080 | } | |
871751e2 AV |
4081 | |
4082 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4083 | ||
871751e2 AV |
4084 | static inline int add_caller(unsigned long *n, unsigned long v) |
4085 | { | |
4086 | unsigned long *p; | |
4087 | int l; | |
4088 | if (!v) | |
4089 | return 1; | |
4090 | l = n[1]; | |
4091 | p = n + 2; | |
4092 | while (l) { | |
4093 | int i = l/2; | |
4094 | unsigned long *q = p + 2 * i; | |
4095 | if (*q == v) { | |
4096 | q[1]++; | |
4097 | return 1; | |
4098 | } | |
4099 | if (*q > v) { | |
4100 | l = i; | |
4101 | } else { | |
4102 | p = q + 2; | |
4103 | l -= i + 1; | |
4104 | } | |
4105 | } | |
4106 | if (++n[1] == n[0]) | |
4107 | return 0; | |
4108 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4109 | p[0] = v; | |
4110 | p[1] = 1; | |
4111 | return 1; | |
4112 | } | |
4113 | ||
8456a648 JK |
4114 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4115 | struct page *page) | |
871751e2 AV |
4116 | { |
4117 | void *p; | |
03787301 | 4118 | int i; |
b1cb0982 | 4119 | |
871751e2 AV |
4120 | if (n[0] == n[1]) |
4121 | return; | |
8456a648 | 4122 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
03787301 | 4123 | if (get_obj_status(page, i) != OBJECT_ACTIVE) |
871751e2 | 4124 | continue; |
b1cb0982 | 4125 | |
871751e2 AV |
4126 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
4127 | return; | |
4128 | } | |
4129 | } | |
4130 | ||
4131 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4132 | { | |
4133 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4134 | unsigned long offset, size; |
9281acea | 4135 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4136 | |
a5c43dae | 4137 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4138 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4139 | if (modname[0]) |
871751e2 AV |
4140 | seq_printf(m, " [%s]", modname); |
4141 | return; | |
4142 | } | |
4143 | #endif | |
4144 | seq_printf(m, "%p", (void *)address); | |
4145 | } | |
4146 | ||
4147 | static int leaks_show(struct seq_file *m, void *p) | |
4148 | { | |
0672aa7c | 4149 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4150 | struct page *page; |
ce8eb6c4 | 4151 | struct kmem_cache_node *n; |
871751e2 | 4152 | const char *name; |
db845067 | 4153 | unsigned long *x = m->private; |
871751e2 AV |
4154 | int node; |
4155 | int i; | |
4156 | ||
4157 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4158 | return 0; | |
4159 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4160 | return 0; | |
4161 | ||
4162 | /* OK, we can do it */ | |
4163 | ||
db845067 | 4164 | x[1] = 0; |
871751e2 | 4165 | |
18bf8541 | 4166 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 AV |
4167 | |
4168 | check_irq_on(); | |
ce8eb6c4 | 4169 | spin_lock_irq(&n->list_lock); |
871751e2 | 4170 | |
8456a648 JK |
4171 | list_for_each_entry(page, &n->slabs_full, lru) |
4172 | handle_slab(x, cachep, page); | |
4173 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4174 | handle_slab(x, cachep, page); | |
ce8eb6c4 | 4175 | spin_unlock_irq(&n->list_lock); |
871751e2 AV |
4176 | } |
4177 | name = cachep->name; | |
db845067 | 4178 | if (x[0] == x[1]) { |
871751e2 | 4179 | /* Increase the buffer size */ |
18004c5d | 4180 | mutex_unlock(&slab_mutex); |
db845067 | 4181 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4182 | if (!m->private) { |
4183 | /* Too bad, we are really out */ | |
db845067 | 4184 | m->private = x; |
18004c5d | 4185 | mutex_lock(&slab_mutex); |
871751e2 AV |
4186 | return -ENOMEM; |
4187 | } | |
db845067 CL |
4188 | *(unsigned long *)m->private = x[0] * 2; |
4189 | kfree(x); | |
18004c5d | 4190 | mutex_lock(&slab_mutex); |
871751e2 AV |
4191 | /* Now make sure this entry will be retried */ |
4192 | m->count = m->size; | |
4193 | return 0; | |
4194 | } | |
db845067 CL |
4195 | for (i = 0; i < x[1]; i++) { |
4196 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4197 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4198 | seq_putc(m, '\n'); |
4199 | } | |
d2e7b7d0 | 4200 | |
871751e2 AV |
4201 | return 0; |
4202 | } | |
4203 | ||
a0ec95a8 | 4204 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4205 | .start = slab_start, |
276a2439 WL |
4206 | .next = slab_next, |
4207 | .stop = slab_stop, | |
871751e2 AV |
4208 | .show = leaks_show, |
4209 | }; | |
a0ec95a8 AD |
4210 | |
4211 | static int slabstats_open(struct inode *inode, struct file *file) | |
4212 | { | |
b208ce32 RJ |
4213 | unsigned long *n; |
4214 | ||
4215 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4216 | if (!n) | |
4217 | return -ENOMEM; | |
4218 | ||
4219 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4220 | ||
4221 | return 0; | |
a0ec95a8 AD |
4222 | } |
4223 | ||
4224 | static const struct file_operations proc_slabstats_operations = { | |
4225 | .open = slabstats_open, | |
4226 | .read = seq_read, | |
4227 | .llseek = seq_lseek, | |
4228 | .release = seq_release_private, | |
4229 | }; | |
4230 | #endif | |
4231 | ||
4232 | static int __init slab_proc_init(void) | |
4233 | { | |
4234 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4235 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4236 | #endif |
a0ec95a8 AD |
4237 | return 0; |
4238 | } | |
4239 | module_init(slab_proc_init); | |
1da177e4 LT |
4240 | #endif |
4241 | ||
00e145b6 MS |
4242 | /** |
4243 | * ksize - get the actual amount of memory allocated for a given object | |
4244 | * @objp: Pointer to the object | |
4245 | * | |
4246 | * kmalloc may internally round up allocations and return more memory | |
4247 | * than requested. ksize() can be used to determine the actual amount of | |
4248 | * memory allocated. The caller may use this additional memory, even though | |
4249 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4250 | * The caller must guarantee that objp points to a valid object previously | |
4251 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4252 | * must not be freed during the duration of the call. | |
4253 | */ | |
fd76bab2 | 4254 | size_t ksize(const void *objp) |
1da177e4 | 4255 | { |
ef8b4520 CL |
4256 | BUG_ON(!objp); |
4257 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4258 | return 0; |
1da177e4 | 4259 | |
8c138bc0 | 4260 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4261 | } |
b1aabecd | 4262 | EXPORT_SYMBOL(ksize); |