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 | ||
072bb0aa MG |
160 | /* |
161 | * true if a page was allocated from pfmemalloc reserves for network-based | |
162 | * swap | |
163 | */ | |
164 | static bool pfmemalloc_active __read_mostly; | |
165 | ||
1da177e4 LT |
166 | /* |
167 | * kmem_bufctl_t: | |
168 | * | |
169 | * Bufctl's are used for linking objs within a slab | |
170 | * linked offsets. | |
171 | * | |
172 | * This implementation relies on "struct page" for locating the cache & | |
173 | * slab an object belongs to. | |
174 | * This allows the bufctl structure to be small (one int), but limits | |
175 | * the number of objects a slab (not a cache) can contain when off-slab | |
176 | * bufctls are used. The limit is the size of the largest general cache | |
177 | * that does not use off-slab slabs. | |
178 | * For 32bit archs with 4 kB pages, is this 56. | |
179 | * This is not serious, as it is only for large objects, when it is unwise | |
180 | * to have too many per slab. | |
181 | * Note: This limit can be raised by introducing a general cache whose size | |
182 | * is less than 512 (PAGE_SIZE<<3), but greater than 256. | |
183 | */ | |
184 | ||
fa5b08d5 | 185 | typedef unsigned int kmem_bufctl_t; |
1da177e4 LT |
186 | #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0) |
187 | #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1) | |
871751e2 AV |
188 | #define BUFCTL_ACTIVE (((kmem_bufctl_t)(~0U))-2) |
189 | #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-3) | |
1da177e4 | 190 | |
1da177e4 LT |
191 | /* |
192 | * struct slab_rcu | |
193 | * | |
194 | * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to | |
195 | * arrange for kmem_freepages to be called via RCU. This is useful if | |
196 | * we need to approach a kernel structure obliquely, from its address | |
197 | * obtained without the usual locking. We can lock the structure to | |
198 | * stabilize it and check it's still at the given address, only if we | |
199 | * can be sure that the memory has not been meanwhile reused for some | |
200 | * other kind of object (which our subsystem's lock might corrupt). | |
201 | * | |
202 | * rcu_read_lock before reading the address, then rcu_read_unlock after | |
203 | * taking the spinlock within the structure expected at that address. | |
1da177e4 LT |
204 | */ |
205 | struct slab_rcu { | |
b28a02de | 206 | struct rcu_head head; |
0c3aa83e | 207 | struct page *page; |
1da177e4 LT |
208 | }; |
209 | ||
5bfe53a7 LJ |
210 | /* |
211 | * struct slab | |
212 | * | |
213 | * Manages the objs in a slab. Placed either at the beginning of mem allocated | |
214 | * for a slab, or allocated from an general cache. | |
215 | * Slabs are chained into three list: fully used, partial, fully free slabs. | |
216 | */ | |
217 | struct slab { | |
218 | union { | |
219 | struct { | |
220 | struct list_head list; | |
5bfe53a7 LJ |
221 | void *s_mem; /* including colour offset */ |
222 | unsigned int inuse; /* num of objs active in slab */ | |
223 | kmem_bufctl_t free; | |
5bfe53a7 LJ |
224 | }; |
225 | struct slab_rcu __slab_cover_slab_rcu; | |
226 | }; | |
227 | }; | |
228 | ||
1da177e4 LT |
229 | /* |
230 | * struct array_cache | |
231 | * | |
1da177e4 LT |
232 | * Purpose: |
233 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
234 | * - reduce the number of linked list operations | |
235 | * - reduce spinlock operations | |
236 | * | |
237 | * The limit is stored in the per-cpu structure to reduce the data cache | |
238 | * footprint. | |
239 | * | |
240 | */ | |
241 | struct array_cache { | |
242 | unsigned int avail; | |
243 | unsigned int limit; | |
244 | unsigned int batchcount; | |
245 | unsigned int touched; | |
e498be7d | 246 | spinlock_t lock; |
bda5b655 | 247 | void *entry[]; /* |
a737b3e2 AM |
248 | * Must have this definition in here for the proper |
249 | * alignment of array_cache. Also simplifies accessing | |
250 | * the entries. | |
072bb0aa MG |
251 | * |
252 | * Entries should not be directly dereferenced as | |
253 | * entries belonging to slabs marked pfmemalloc will | |
254 | * have the lower bits set SLAB_OBJ_PFMEMALLOC | |
a737b3e2 | 255 | */ |
1da177e4 LT |
256 | }; |
257 | ||
072bb0aa MG |
258 | #define SLAB_OBJ_PFMEMALLOC 1 |
259 | static inline bool is_obj_pfmemalloc(void *objp) | |
260 | { | |
261 | return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC; | |
262 | } | |
263 | ||
264 | static inline void set_obj_pfmemalloc(void **objp) | |
265 | { | |
266 | *objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC); | |
267 | return; | |
268 | } | |
269 | ||
270 | static inline void clear_obj_pfmemalloc(void **objp) | |
271 | { | |
272 | *objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC); | |
273 | } | |
274 | ||
a737b3e2 AM |
275 | /* |
276 | * bootstrap: The caches do not work without cpuarrays anymore, but the | |
277 | * cpuarrays are allocated from the generic caches... | |
1da177e4 LT |
278 | */ |
279 | #define BOOT_CPUCACHE_ENTRIES 1 | |
280 | struct arraycache_init { | |
281 | struct array_cache cache; | |
b28a02de | 282 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
1da177e4 LT |
283 | }; |
284 | ||
e498be7d CL |
285 | /* |
286 | * Need this for bootstrapping a per node allocator. | |
287 | */ | |
556a169d | 288 | #define NUM_INIT_LISTS (3 * MAX_NUMNODES) |
ce8eb6c4 | 289 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 290 | #define CACHE_CACHE 0 |
556a169d | 291 | #define SIZE_AC MAX_NUMNODES |
ce8eb6c4 | 292 | #define SIZE_NODE (2 * MAX_NUMNODES) |
e498be7d | 293 | |
ed11d9eb | 294 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 295 | struct kmem_cache_node *n, int tofree); |
ed11d9eb CL |
296 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
297 | int node); | |
83b519e8 | 298 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 299 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 300 | |
e0a42726 IM |
301 | static int slab_early_init = 1; |
302 | ||
e3366016 | 303 | #define INDEX_AC kmalloc_index(sizeof(struct arraycache_init)) |
ce8eb6c4 | 304 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 305 | |
ce8eb6c4 | 306 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
307 | { |
308 | INIT_LIST_HEAD(&parent->slabs_full); | |
309 | INIT_LIST_HEAD(&parent->slabs_partial); | |
310 | INIT_LIST_HEAD(&parent->slabs_free); | |
311 | parent->shared = NULL; | |
312 | parent->alien = NULL; | |
2e1217cf | 313 | parent->colour_next = 0; |
e498be7d CL |
314 | spin_lock_init(&parent->list_lock); |
315 | parent->free_objects = 0; | |
316 | parent->free_touched = 0; | |
317 | } | |
318 | ||
a737b3e2 AM |
319 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
320 | do { \ | |
321 | INIT_LIST_HEAD(listp); \ | |
6a67368c | 322 | list_splice(&(cachep->node[nodeid]->slab), listp); \ |
e498be7d CL |
323 | } while (0) |
324 | ||
a737b3e2 AM |
325 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
326 | do { \ | |
e498be7d CL |
327 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
328 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
329 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
330 | } while (0) | |
1da177e4 | 331 | |
1da177e4 LT |
332 | #define CFLGS_OFF_SLAB (0x80000000UL) |
333 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
334 | ||
335 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
336 | /* |
337 | * Optimization question: fewer reaps means less probability for unnessary | |
338 | * cpucache drain/refill cycles. | |
1da177e4 | 339 | * |
dc6f3f27 | 340 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
341 | * which could lock up otherwise freeable slabs. |
342 | */ | |
343 | #define REAPTIMEOUT_CPUC (2*HZ) | |
344 | #define REAPTIMEOUT_LIST3 (4*HZ) | |
345 | ||
346 | #if STATS | |
347 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
348 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
349 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
350 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 351 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
352 | #define STATS_SET_HIGH(x) \ |
353 | do { \ | |
354 | if ((x)->num_active > (x)->high_mark) \ | |
355 | (x)->high_mark = (x)->num_active; \ | |
356 | } while (0) | |
1da177e4 LT |
357 | #define STATS_INC_ERR(x) ((x)->errors++) |
358 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 359 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 360 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
361 | #define STATS_SET_FREEABLE(x, i) \ |
362 | do { \ | |
363 | if ((x)->max_freeable < i) \ | |
364 | (x)->max_freeable = i; \ | |
365 | } while (0) | |
1da177e4 LT |
366 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
367 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
368 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
369 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
370 | #else | |
371 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
372 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
373 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
374 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 375 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
376 | #define STATS_SET_HIGH(x) do { } while (0) |
377 | #define STATS_INC_ERR(x) do { } while (0) | |
378 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 379 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 380 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 381 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
382 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
383 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
384 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
385 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
386 | #endif | |
387 | ||
388 | #if DEBUG | |
1da177e4 | 389 | |
a737b3e2 AM |
390 | /* |
391 | * memory layout of objects: | |
1da177e4 | 392 | * 0 : objp |
3dafccf2 | 393 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
394 | * the end of an object is aligned with the end of the real |
395 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 396 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 397 | * redzone word. |
3dafccf2 | 398 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
399 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
400 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 401 | * [BYTES_PER_WORD long] |
1da177e4 | 402 | */ |
343e0d7a | 403 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 404 | { |
3dafccf2 | 405 | return cachep->obj_offset; |
1da177e4 LT |
406 | } |
407 | ||
b46b8f19 | 408 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
409 | { |
410 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
411 | return (unsigned long long*) (objp + obj_offset(cachep) - |
412 | sizeof(unsigned long long)); | |
1da177e4 LT |
413 | } |
414 | ||
b46b8f19 | 415 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
416 | { |
417 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
418 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 419 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 420 | sizeof(unsigned long long) - |
87a927c7 | 421 | REDZONE_ALIGN); |
3b0efdfa | 422 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 423 | sizeof(unsigned long long)); |
1da177e4 LT |
424 | } |
425 | ||
343e0d7a | 426 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
427 | { |
428 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 429 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
430 | } |
431 | ||
432 | #else | |
433 | ||
3dafccf2 | 434 | #define obj_offset(x) 0 |
b46b8f19 DW |
435 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
436 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
437 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
438 | ||
439 | #endif | |
440 | ||
1da177e4 | 441 | /* |
3df1cccd DR |
442 | * Do not go above this order unless 0 objects fit into the slab or |
443 | * overridden on the command line. | |
1da177e4 | 444 | */ |
543585cc DR |
445 | #define SLAB_MAX_ORDER_HI 1 |
446 | #define SLAB_MAX_ORDER_LO 0 | |
447 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 448 | static bool slab_max_order_set __initdata; |
1da177e4 | 449 | |
6ed5eb22 PE |
450 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
451 | { | |
b49af68f | 452 | struct page *page = virt_to_head_page(obj); |
35026088 | 453 | return page->slab_cache; |
6ed5eb22 PE |
454 | } |
455 | ||
456 | static inline struct slab *virt_to_slab(const void *obj) | |
457 | { | |
b49af68f | 458 | struct page *page = virt_to_head_page(obj); |
35026088 CL |
459 | |
460 | VM_BUG_ON(!PageSlab(page)); | |
461 | return page->slab_page; | |
6ed5eb22 PE |
462 | } |
463 | ||
8fea4e96 PE |
464 | static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab, |
465 | unsigned int idx) | |
466 | { | |
3b0efdfa | 467 | return slab->s_mem + cache->size * idx; |
8fea4e96 PE |
468 | } |
469 | ||
6a2d7a95 | 470 | /* |
3b0efdfa CL |
471 | * We want to avoid an expensive divide : (offset / cache->size) |
472 | * Using the fact that size is a constant for a particular cache, | |
473 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
474 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
475 | */ | |
476 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
477 | const struct slab *slab, void *obj) | |
8fea4e96 | 478 | { |
6a2d7a95 ED |
479 | u32 offset = (obj - slab->s_mem); |
480 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); | |
8fea4e96 PE |
481 | } |
482 | ||
1da177e4 | 483 | static struct arraycache_init initarray_generic = |
b28a02de | 484 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 LT |
485 | |
486 | /* internal cache of cache description objs */ | |
9b030cb8 | 487 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
488 | .batchcount = 1, |
489 | .limit = BOOT_CPUCACHE_ENTRIES, | |
490 | .shared = 1, | |
3b0efdfa | 491 | .size = sizeof(struct kmem_cache), |
b28a02de | 492 | .name = "kmem_cache", |
1da177e4 LT |
493 | }; |
494 | ||
056c6241 RT |
495 | #define BAD_ALIEN_MAGIC 0x01020304ul |
496 | ||
f1aaee53 AV |
497 | #ifdef CONFIG_LOCKDEP |
498 | ||
499 | /* | |
500 | * Slab sometimes uses the kmalloc slabs to store the slab headers | |
501 | * for other slabs "off slab". | |
502 | * The locking for this is tricky in that it nests within the locks | |
503 | * of all other slabs in a few places; to deal with this special | |
504 | * locking we put on-slab caches into a separate lock-class. | |
056c6241 RT |
505 | * |
506 | * We set lock class for alien array caches which are up during init. | |
507 | * The lock annotation will be lost if all cpus of a node goes down and | |
508 | * then comes back up during hotplug | |
f1aaee53 | 509 | */ |
056c6241 RT |
510 | static struct lock_class_key on_slab_l3_key; |
511 | static struct lock_class_key on_slab_alc_key; | |
512 | ||
83835b3d PZ |
513 | static struct lock_class_key debugobj_l3_key; |
514 | static struct lock_class_key debugobj_alc_key; | |
515 | ||
516 | static void slab_set_lock_classes(struct kmem_cache *cachep, | |
517 | struct lock_class_key *l3_key, struct lock_class_key *alc_key, | |
518 | int q) | |
519 | { | |
520 | struct array_cache **alc; | |
ce8eb6c4 | 521 | struct kmem_cache_node *n; |
83835b3d PZ |
522 | int r; |
523 | ||
ce8eb6c4 CL |
524 | n = cachep->node[q]; |
525 | if (!n) | |
83835b3d PZ |
526 | return; |
527 | ||
ce8eb6c4 CL |
528 | lockdep_set_class(&n->list_lock, l3_key); |
529 | alc = n->alien; | |
83835b3d PZ |
530 | /* |
531 | * FIXME: This check for BAD_ALIEN_MAGIC | |
532 | * should go away when common slab code is taught to | |
533 | * work even without alien caches. | |
534 | * Currently, non NUMA code returns BAD_ALIEN_MAGIC | |
535 | * for alloc_alien_cache, | |
536 | */ | |
537 | if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) | |
538 | return; | |
539 | for_each_node(r) { | |
540 | if (alc[r]) | |
541 | lockdep_set_class(&alc[r]->lock, alc_key); | |
542 | } | |
543 | } | |
544 | ||
545 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node) | |
546 | { | |
547 | slab_set_lock_classes(cachep, &debugobj_l3_key, &debugobj_alc_key, node); | |
548 | } | |
549 | ||
550 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
551 | { | |
552 | int node; | |
553 | ||
554 | for_each_online_node(node) | |
555 | slab_set_debugobj_lock_classes_node(cachep, node); | |
556 | } | |
557 | ||
ce79ddc8 | 558 | static void init_node_lock_keys(int q) |
f1aaee53 | 559 | { |
e3366016 | 560 | int i; |
056c6241 | 561 | |
97d06609 | 562 | if (slab_state < UP) |
ce79ddc8 PE |
563 | return; |
564 | ||
0f8f8094 | 565 | for (i = 1; i <= KMALLOC_SHIFT_HIGH; i++) { |
ce8eb6c4 | 566 | struct kmem_cache_node *n; |
e3366016 CL |
567 | struct kmem_cache *cache = kmalloc_caches[i]; |
568 | ||
569 | if (!cache) | |
570 | continue; | |
ce79ddc8 | 571 | |
ce8eb6c4 CL |
572 | n = cache->node[q]; |
573 | if (!n || OFF_SLAB(cache)) | |
00afa758 | 574 | continue; |
83835b3d | 575 | |
e3366016 | 576 | slab_set_lock_classes(cache, &on_slab_l3_key, |
83835b3d | 577 | &on_slab_alc_key, q); |
f1aaee53 AV |
578 | } |
579 | } | |
ce79ddc8 | 580 | |
6ccfb5bc GC |
581 | static void on_slab_lock_classes_node(struct kmem_cache *cachep, int q) |
582 | { | |
6a67368c | 583 | if (!cachep->node[q]) |
6ccfb5bc GC |
584 | return; |
585 | ||
586 | slab_set_lock_classes(cachep, &on_slab_l3_key, | |
587 | &on_slab_alc_key, q); | |
588 | } | |
589 | ||
590 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) | |
591 | { | |
592 | int node; | |
593 | ||
594 | VM_BUG_ON(OFF_SLAB(cachep)); | |
595 | for_each_node(node) | |
596 | on_slab_lock_classes_node(cachep, node); | |
597 | } | |
598 | ||
ce79ddc8 PE |
599 | static inline void init_lock_keys(void) |
600 | { | |
601 | int node; | |
602 | ||
603 | for_each_node(node) | |
604 | init_node_lock_keys(node); | |
605 | } | |
f1aaee53 | 606 | #else |
ce79ddc8 PE |
607 | static void init_node_lock_keys(int q) |
608 | { | |
609 | } | |
610 | ||
056c6241 | 611 | static inline void init_lock_keys(void) |
f1aaee53 AV |
612 | { |
613 | } | |
83835b3d | 614 | |
6ccfb5bc GC |
615 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) |
616 | { | |
617 | } | |
618 | ||
619 | static inline void on_slab_lock_classes_node(struct kmem_cache *cachep, int node) | |
620 | { | |
621 | } | |
622 | ||
83835b3d PZ |
623 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node) |
624 | { | |
625 | } | |
626 | ||
627 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
628 | { | |
629 | } | |
f1aaee53 AV |
630 | #endif |
631 | ||
1871e52c | 632 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 633 | |
343e0d7a | 634 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
635 | { |
636 | return cachep->array[smp_processor_id()]; | |
637 | } | |
638 | ||
fbaccacf | 639 | static size_t slab_mgmt_size(size_t nr_objs, size_t align) |
1da177e4 | 640 | { |
fbaccacf SR |
641 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); |
642 | } | |
1da177e4 | 643 | |
a737b3e2 AM |
644 | /* |
645 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
646 | */ | |
fbaccacf SR |
647 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
648 | size_t align, int flags, size_t *left_over, | |
649 | unsigned int *num) | |
650 | { | |
651 | int nr_objs; | |
652 | size_t mgmt_size; | |
653 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 654 | |
fbaccacf SR |
655 | /* |
656 | * The slab management structure can be either off the slab or | |
657 | * on it. For the latter case, the memory allocated for a | |
658 | * slab is used for: | |
659 | * | |
660 | * - The struct slab | |
661 | * - One kmem_bufctl_t for each object | |
662 | * - Padding to respect alignment of @align | |
663 | * - @buffer_size bytes for each object | |
664 | * | |
665 | * If the slab management structure is off the slab, then the | |
666 | * alignment will already be calculated into the size. Because | |
667 | * the slabs are all pages aligned, the objects will be at the | |
668 | * correct alignment when allocated. | |
669 | */ | |
670 | if (flags & CFLGS_OFF_SLAB) { | |
671 | mgmt_size = 0; | |
672 | nr_objs = slab_size / buffer_size; | |
673 | ||
674 | if (nr_objs > SLAB_LIMIT) | |
675 | nr_objs = SLAB_LIMIT; | |
676 | } else { | |
677 | /* | |
678 | * Ignore padding for the initial guess. The padding | |
679 | * is at most @align-1 bytes, and @buffer_size is at | |
680 | * least @align. In the worst case, this result will | |
681 | * be one greater than the number of objects that fit | |
682 | * into the memory allocation when taking the padding | |
683 | * into account. | |
684 | */ | |
685 | nr_objs = (slab_size - sizeof(struct slab)) / | |
686 | (buffer_size + sizeof(kmem_bufctl_t)); | |
687 | ||
688 | /* | |
689 | * This calculated number will be either the right | |
690 | * amount, or one greater than what we want. | |
691 | */ | |
692 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size | |
693 | > slab_size) | |
694 | nr_objs--; | |
695 | ||
696 | if (nr_objs > SLAB_LIMIT) | |
697 | nr_objs = SLAB_LIMIT; | |
698 | ||
699 | mgmt_size = slab_mgmt_size(nr_objs, align); | |
700 | } | |
701 | *num = nr_objs; | |
702 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
703 | } |
704 | ||
f28510d3 | 705 | #if DEBUG |
d40cee24 | 706 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 707 | |
a737b3e2 AM |
708 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
709 | char *msg) | |
1da177e4 LT |
710 | { |
711 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 712 | function, cachep->name, msg); |
1da177e4 | 713 | dump_stack(); |
373d4d09 | 714 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 715 | } |
f28510d3 | 716 | #endif |
1da177e4 | 717 | |
3395ee05 PM |
718 | /* |
719 | * By default on NUMA we use alien caches to stage the freeing of | |
720 | * objects allocated from other nodes. This causes massive memory | |
721 | * inefficiencies when using fake NUMA setup to split memory into a | |
722 | * large number of small nodes, so it can be disabled on the command | |
723 | * line | |
724 | */ | |
725 | ||
726 | static int use_alien_caches __read_mostly = 1; | |
727 | static int __init noaliencache_setup(char *s) | |
728 | { | |
729 | use_alien_caches = 0; | |
730 | return 1; | |
731 | } | |
732 | __setup("noaliencache", noaliencache_setup); | |
733 | ||
3df1cccd DR |
734 | static int __init slab_max_order_setup(char *str) |
735 | { | |
736 | get_option(&str, &slab_max_order); | |
737 | slab_max_order = slab_max_order < 0 ? 0 : | |
738 | min(slab_max_order, MAX_ORDER - 1); | |
739 | slab_max_order_set = true; | |
740 | ||
741 | return 1; | |
742 | } | |
743 | __setup("slab_max_order=", slab_max_order_setup); | |
744 | ||
8fce4d8e CL |
745 | #ifdef CONFIG_NUMA |
746 | /* | |
747 | * Special reaping functions for NUMA systems called from cache_reap(). | |
748 | * These take care of doing round robin flushing of alien caches (containing | |
749 | * objects freed on different nodes from which they were allocated) and the | |
750 | * flushing of remote pcps by calling drain_node_pages. | |
751 | */ | |
1871e52c | 752 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
753 | |
754 | static void init_reap_node(int cpu) | |
755 | { | |
756 | int node; | |
757 | ||
7d6e6d09 | 758 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 759 | if (node == MAX_NUMNODES) |
442295c9 | 760 | node = first_node(node_online_map); |
8fce4d8e | 761 | |
1871e52c | 762 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
763 | } |
764 | ||
765 | static void next_reap_node(void) | |
766 | { | |
909ea964 | 767 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 768 | |
8fce4d8e CL |
769 | node = next_node(node, node_online_map); |
770 | if (unlikely(node >= MAX_NUMNODES)) | |
771 | node = first_node(node_online_map); | |
909ea964 | 772 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
773 | } |
774 | ||
775 | #else | |
776 | #define init_reap_node(cpu) do { } while (0) | |
777 | #define next_reap_node(void) do { } while (0) | |
778 | #endif | |
779 | ||
1da177e4 LT |
780 | /* |
781 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
782 | * via the workqueue/eventd. | |
783 | * Add the CPU number into the expiration time to minimize the possibility of | |
784 | * the CPUs getting into lockstep and contending for the global cache chain | |
785 | * lock. | |
786 | */ | |
0db0628d | 787 | static void start_cpu_timer(int cpu) |
1da177e4 | 788 | { |
1871e52c | 789 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
790 | |
791 | /* | |
792 | * When this gets called from do_initcalls via cpucache_init(), | |
793 | * init_workqueues() has already run, so keventd will be setup | |
794 | * at that time. | |
795 | */ | |
52bad64d | 796 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 797 | init_reap_node(cpu); |
203b42f7 | 798 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
799 | schedule_delayed_work_on(cpu, reap_work, |
800 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
801 | } |
802 | } | |
803 | ||
e498be7d | 804 | static struct array_cache *alloc_arraycache(int node, int entries, |
83b519e8 | 805 | int batchcount, gfp_t gfp) |
1da177e4 | 806 | { |
b28a02de | 807 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1da177e4 LT |
808 | struct array_cache *nc = NULL; |
809 | ||
83b519e8 | 810 | nc = kmalloc_node(memsize, gfp, node); |
d5cff635 CM |
811 | /* |
812 | * The array_cache structures contain pointers to free object. | |
25985edc | 813 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
814 | * cache the pointers are not cleared and they could be counted as |
815 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
816 | * not scan such objects. | |
817 | */ | |
818 | kmemleak_no_scan(nc); | |
1da177e4 LT |
819 | if (nc) { |
820 | nc->avail = 0; | |
821 | nc->limit = entries; | |
822 | nc->batchcount = batchcount; | |
823 | nc->touched = 0; | |
e498be7d | 824 | spin_lock_init(&nc->lock); |
1da177e4 LT |
825 | } |
826 | return nc; | |
827 | } | |
828 | ||
072bb0aa MG |
829 | static inline bool is_slab_pfmemalloc(struct slab *slabp) |
830 | { | |
831 | struct page *page = virt_to_page(slabp->s_mem); | |
832 | ||
833 | return PageSlabPfmemalloc(page); | |
834 | } | |
835 | ||
836 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
837 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
838 | struct array_cache *ac) | |
839 | { | |
ce8eb6c4 | 840 | struct kmem_cache_node *n = cachep->node[numa_mem_id()]; |
072bb0aa MG |
841 | struct slab *slabp; |
842 | unsigned long flags; | |
843 | ||
844 | if (!pfmemalloc_active) | |
845 | return; | |
846 | ||
ce8eb6c4 CL |
847 | spin_lock_irqsave(&n->list_lock, flags); |
848 | list_for_each_entry(slabp, &n->slabs_full, list) | |
072bb0aa MG |
849 | if (is_slab_pfmemalloc(slabp)) |
850 | goto out; | |
851 | ||
ce8eb6c4 | 852 | list_for_each_entry(slabp, &n->slabs_partial, list) |
072bb0aa MG |
853 | if (is_slab_pfmemalloc(slabp)) |
854 | goto out; | |
855 | ||
ce8eb6c4 | 856 | list_for_each_entry(slabp, &n->slabs_free, list) |
072bb0aa MG |
857 | if (is_slab_pfmemalloc(slabp)) |
858 | goto out; | |
859 | ||
860 | pfmemalloc_active = false; | |
861 | out: | |
ce8eb6c4 | 862 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
863 | } |
864 | ||
381760ea | 865 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
866 | gfp_t flags, bool force_refill) |
867 | { | |
868 | int i; | |
869 | void *objp = ac->entry[--ac->avail]; | |
870 | ||
871 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
872 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 873 | struct kmem_cache_node *n; |
072bb0aa MG |
874 | |
875 | if (gfp_pfmemalloc_allowed(flags)) { | |
876 | clear_obj_pfmemalloc(&objp); | |
877 | return objp; | |
878 | } | |
879 | ||
880 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 881 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
882 | /* If a !PFMEMALLOC object is found, swap them */ |
883 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
884 | objp = ac->entry[i]; | |
885 | ac->entry[i] = ac->entry[ac->avail]; | |
886 | ac->entry[ac->avail] = objp; | |
887 | return objp; | |
888 | } | |
889 | } | |
890 | ||
891 | /* | |
892 | * If there are empty slabs on the slabs_free list and we are | |
893 | * being forced to refill the cache, mark this one !pfmemalloc. | |
894 | */ | |
ce8eb6c4 CL |
895 | n = cachep->node[numa_mem_id()]; |
896 | if (!list_empty(&n->slabs_free) && force_refill) { | |
072bb0aa | 897 | struct slab *slabp = virt_to_slab(objp); |
30c29bea | 898 | ClearPageSlabPfmemalloc(virt_to_head_page(slabp->s_mem)); |
072bb0aa MG |
899 | clear_obj_pfmemalloc(&objp); |
900 | recheck_pfmemalloc_active(cachep, ac); | |
901 | return objp; | |
902 | } | |
903 | ||
904 | /* No !PFMEMALLOC objects available */ | |
905 | ac->avail++; | |
906 | objp = NULL; | |
907 | } | |
908 | ||
909 | return objp; | |
910 | } | |
911 | ||
381760ea MG |
912 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
913 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
914 | { | |
915 | void *objp; | |
916 | ||
917 | if (unlikely(sk_memalloc_socks())) | |
918 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
919 | else | |
920 | objp = ac->entry[--ac->avail]; | |
921 | ||
922 | return objp; | |
923 | } | |
924 | ||
925 | static void *__ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
072bb0aa MG |
926 | void *objp) |
927 | { | |
928 | if (unlikely(pfmemalloc_active)) { | |
929 | /* Some pfmemalloc slabs exist, check if this is one */ | |
73293c2f JK |
930 | struct slab *slabp = virt_to_slab(objp); |
931 | struct page *page = virt_to_head_page(slabp->s_mem); | |
072bb0aa MG |
932 | if (PageSlabPfmemalloc(page)) |
933 | set_obj_pfmemalloc(&objp); | |
934 | } | |
935 | ||
381760ea MG |
936 | return objp; |
937 | } | |
938 | ||
939 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
940 | void *objp) | |
941 | { | |
942 | if (unlikely(sk_memalloc_socks())) | |
943 | objp = __ac_put_obj(cachep, ac, objp); | |
944 | ||
072bb0aa MG |
945 | ac->entry[ac->avail++] = objp; |
946 | } | |
947 | ||
3ded175a CL |
948 | /* |
949 | * Transfer objects in one arraycache to another. | |
950 | * Locking must be handled by the caller. | |
951 | * | |
952 | * Return the number of entries transferred. | |
953 | */ | |
954 | static int transfer_objects(struct array_cache *to, | |
955 | struct array_cache *from, unsigned int max) | |
956 | { | |
957 | /* Figure out how many entries to transfer */ | |
732eacc0 | 958 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
959 | |
960 | if (!nr) | |
961 | return 0; | |
962 | ||
963 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
964 | sizeof(void *) *nr); | |
965 | ||
966 | from->avail -= nr; | |
967 | to->avail += nr; | |
3ded175a CL |
968 | return nr; |
969 | } | |
970 | ||
765c4507 CL |
971 | #ifndef CONFIG_NUMA |
972 | ||
973 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 974 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 975 | |
83b519e8 | 976 | static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
765c4507 CL |
977 | { |
978 | return (struct array_cache **)BAD_ALIEN_MAGIC; | |
979 | } | |
980 | ||
981 | static inline void free_alien_cache(struct array_cache **ac_ptr) | |
982 | { | |
983 | } | |
984 | ||
985 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
986 | { | |
987 | return 0; | |
988 | } | |
989 | ||
990 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
991 | gfp_t flags) | |
992 | { | |
993 | return NULL; | |
994 | } | |
995 | ||
8b98c169 | 996 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
997 | gfp_t flags, int nodeid) |
998 | { | |
999 | return NULL; | |
1000 | } | |
1001 | ||
1002 | #else /* CONFIG_NUMA */ | |
1003 | ||
8b98c169 | 1004 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 1005 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 1006 | |
83b519e8 | 1007 | static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
e498be7d CL |
1008 | { |
1009 | struct array_cache **ac_ptr; | |
8ef82866 | 1010 | int memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
1011 | int i; |
1012 | ||
1013 | if (limit > 1) | |
1014 | limit = 12; | |
f3186a9c | 1015 | ac_ptr = kzalloc_node(memsize, gfp, node); |
e498be7d CL |
1016 | if (ac_ptr) { |
1017 | for_each_node(i) { | |
f3186a9c | 1018 | if (i == node || !node_online(i)) |
e498be7d | 1019 | continue; |
83b519e8 | 1020 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp); |
e498be7d | 1021 | if (!ac_ptr[i]) { |
cc550def | 1022 | for (i--; i >= 0; i--) |
e498be7d CL |
1023 | kfree(ac_ptr[i]); |
1024 | kfree(ac_ptr); | |
1025 | return NULL; | |
1026 | } | |
1027 | } | |
1028 | } | |
1029 | return ac_ptr; | |
1030 | } | |
1031 | ||
5295a74c | 1032 | static void free_alien_cache(struct array_cache **ac_ptr) |
e498be7d CL |
1033 | { |
1034 | int i; | |
1035 | ||
1036 | if (!ac_ptr) | |
1037 | return; | |
e498be7d | 1038 | for_each_node(i) |
b28a02de | 1039 | kfree(ac_ptr[i]); |
e498be7d CL |
1040 | kfree(ac_ptr); |
1041 | } | |
1042 | ||
343e0d7a | 1043 | static void __drain_alien_cache(struct kmem_cache *cachep, |
5295a74c | 1044 | struct array_cache *ac, int node) |
e498be7d | 1045 | { |
ce8eb6c4 | 1046 | struct kmem_cache_node *n = cachep->node[node]; |
e498be7d CL |
1047 | |
1048 | if (ac->avail) { | |
ce8eb6c4 | 1049 | spin_lock(&n->list_lock); |
e00946fe CL |
1050 | /* |
1051 | * Stuff objects into the remote nodes shared array first. | |
1052 | * That way we could avoid the overhead of putting the objects | |
1053 | * into the free lists and getting them back later. | |
1054 | */ | |
ce8eb6c4 CL |
1055 | if (n->shared) |
1056 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 1057 | |
ff69416e | 1058 | free_block(cachep, ac->entry, ac->avail, node); |
e498be7d | 1059 | ac->avail = 0; |
ce8eb6c4 | 1060 | spin_unlock(&n->list_lock); |
e498be7d CL |
1061 | } |
1062 | } | |
1063 | ||
8fce4d8e CL |
1064 | /* |
1065 | * Called from cache_reap() to regularly drain alien caches round robin. | |
1066 | */ | |
ce8eb6c4 | 1067 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 1068 | { |
909ea964 | 1069 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 1070 | |
ce8eb6c4 CL |
1071 | if (n->alien) { |
1072 | struct array_cache *ac = n->alien[node]; | |
e00946fe CL |
1073 | |
1074 | if (ac && ac->avail && spin_trylock_irq(&ac->lock)) { | |
8fce4d8e CL |
1075 | __drain_alien_cache(cachep, ac, node); |
1076 | spin_unlock_irq(&ac->lock); | |
1077 | } | |
1078 | } | |
1079 | } | |
1080 | ||
a737b3e2 AM |
1081 | static void drain_alien_cache(struct kmem_cache *cachep, |
1082 | struct array_cache **alien) | |
e498be7d | 1083 | { |
b28a02de | 1084 | int i = 0; |
e498be7d CL |
1085 | struct array_cache *ac; |
1086 | unsigned long flags; | |
1087 | ||
1088 | for_each_online_node(i) { | |
4484ebf1 | 1089 | ac = alien[i]; |
e498be7d CL |
1090 | if (ac) { |
1091 | spin_lock_irqsave(&ac->lock, flags); | |
1092 | __drain_alien_cache(cachep, ac, i); | |
1093 | spin_unlock_irqrestore(&ac->lock, flags); | |
1094 | } | |
1095 | } | |
1096 | } | |
729bd0b7 | 1097 | |
873623df | 1098 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 | 1099 | { |
1ea991b0 | 1100 | int nodeid = page_to_nid(virt_to_page(objp)); |
ce8eb6c4 | 1101 | struct kmem_cache_node *n; |
729bd0b7 | 1102 | struct array_cache *alien = NULL; |
1ca4cb24 PE |
1103 | int node; |
1104 | ||
7d6e6d09 | 1105 | node = numa_mem_id(); |
729bd0b7 PE |
1106 | |
1107 | /* | |
1108 | * Make sure we are not freeing a object from another node to the array | |
1109 | * cache on this cpu. | |
1110 | */ | |
1ea991b0 | 1111 | if (likely(nodeid == node)) |
729bd0b7 PE |
1112 | return 0; |
1113 | ||
ce8eb6c4 | 1114 | n = cachep->node[node]; |
729bd0b7 | 1115 | STATS_INC_NODEFREES(cachep); |
ce8eb6c4 CL |
1116 | if (n->alien && n->alien[nodeid]) { |
1117 | alien = n->alien[nodeid]; | |
873623df | 1118 | spin_lock(&alien->lock); |
729bd0b7 PE |
1119 | if (unlikely(alien->avail == alien->limit)) { |
1120 | STATS_INC_ACOVERFLOW(cachep); | |
1121 | __drain_alien_cache(cachep, alien, nodeid); | |
1122 | } | |
072bb0aa | 1123 | ac_put_obj(cachep, alien, objp); |
729bd0b7 PE |
1124 | spin_unlock(&alien->lock); |
1125 | } else { | |
6a67368c | 1126 | spin_lock(&(cachep->node[nodeid])->list_lock); |
729bd0b7 | 1127 | free_block(cachep, &objp, 1, nodeid); |
6a67368c | 1128 | spin_unlock(&(cachep->node[nodeid])->list_lock); |
729bd0b7 PE |
1129 | } |
1130 | return 1; | |
1131 | } | |
e498be7d CL |
1132 | #endif |
1133 | ||
8f9f8d9e | 1134 | /* |
6a67368c | 1135 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1136 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1137 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1138 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1139 | * already in use. |
1140 | * | |
18004c5d | 1141 | * Must hold slab_mutex. |
8f9f8d9e | 1142 | */ |
6a67368c | 1143 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1144 | { |
1145 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1146 | struct kmem_cache_node *n; |
6744f087 | 1147 | const int memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1148 | |
18004c5d | 1149 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e DR |
1150 | /* |
1151 | * Set up the size64 kmemlist for cpu before we can | |
1152 | * begin anything. Make sure some other cpu on this | |
1153 | * node has not already allocated this | |
1154 | */ | |
6a67368c | 1155 | if (!cachep->node[node]) { |
ce8eb6c4 CL |
1156 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1157 | if (!n) | |
8f9f8d9e | 1158 | return -ENOMEM; |
ce8eb6c4 CL |
1159 | kmem_cache_node_init(n); |
1160 | n->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
8f9f8d9e DR |
1161 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
1162 | ||
1163 | /* | |
1164 | * The l3s don't come and go as CPUs come and | |
18004c5d | 1165 | * go. slab_mutex is sufficient |
8f9f8d9e DR |
1166 | * protection here. |
1167 | */ | |
ce8eb6c4 | 1168 | cachep->node[node] = n; |
8f9f8d9e DR |
1169 | } |
1170 | ||
6a67368c CL |
1171 | spin_lock_irq(&cachep->node[node]->list_lock); |
1172 | cachep->node[node]->free_limit = | |
8f9f8d9e DR |
1173 | (1 + nr_cpus_node(node)) * |
1174 | cachep->batchcount + cachep->num; | |
6a67368c | 1175 | spin_unlock_irq(&cachep->node[node]->list_lock); |
8f9f8d9e DR |
1176 | } |
1177 | return 0; | |
1178 | } | |
1179 | ||
0fa8103b WL |
1180 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1181 | struct kmem_cache_node *n) | |
1182 | { | |
1183 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1184 | } | |
1185 | ||
0db0628d | 1186 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1187 | { |
1188 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1189 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1190 | int node = cpu_to_mem(cpu); |
a70f7302 | 1191 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1192 | |
18004c5d | 1193 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1194 | struct array_cache *nc; |
1195 | struct array_cache *shared; | |
1196 | struct array_cache **alien; | |
fbf1e473 | 1197 | |
fbf1e473 AM |
1198 | /* cpu is dead; no one can alloc from it. */ |
1199 | nc = cachep->array[cpu]; | |
1200 | cachep->array[cpu] = NULL; | |
ce8eb6c4 | 1201 | n = cachep->node[node]; |
fbf1e473 | 1202 | |
ce8eb6c4 | 1203 | if (!n) |
fbf1e473 AM |
1204 | goto free_array_cache; |
1205 | ||
ce8eb6c4 | 1206 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1207 | |
ce8eb6c4 CL |
1208 | /* Free limit for this kmem_cache_node */ |
1209 | n->free_limit -= cachep->batchcount; | |
fbf1e473 AM |
1210 | if (nc) |
1211 | free_block(cachep, nc->entry, nc->avail, node); | |
1212 | ||
58463c1f | 1213 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1214 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1215 | goto free_array_cache; |
1216 | } | |
1217 | ||
ce8eb6c4 | 1218 | shared = n->shared; |
fbf1e473 AM |
1219 | if (shared) { |
1220 | free_block(cachep, shared->entry, | |
1221 | shared->avail, node); | |
ce8eb6c4 | 1222 | n->shared = NULL; |
fbf1e473 AM |
1223 | } |
1224 | ||
ce8eb6c4 CL |
1225 | alien = n->alien; |
1226 | n->alien = NULL; | |
fbf1e473 | 1227 | |
ce8eb6c4 | 1228 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1229 | |
1230 | kfree(shared); | |
1231 | if (alien) { | |
1232 | drain_alien_cache(cachep, alien); | |
1233 | free_alien_cache(alien); | |
1234 | } | |
1235 | free_array_cache: | |
1236 | kfree(nc); | |
1237 | } | |
1238 | /* | |
1239 | * In the previous loop, all the objects were freed to | |
1240 | * the respective cache's slabs, now we can go ahead and | |
1241 | * shrink each nodelist to its limit. | |
1242 | */ | |
18004c5d | 1243 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 CL |
1244 | n = cachep->node[node]; |
1245 | if (!n) | |
fbf1e473 | 1246 | continue; |
0fa8103b | 1247 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1248 | } |
1249 | } | |
1250 | ||
0db0628d | 1251 | static int cpuup_prepare(long cpu) |
1da177e4 | 1252 | { |
343e0d7a | 1253 | struct kmem_cache *cachep; |
ce8eb6c4 | 1254 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1255 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1256 | int err; |
1da177e4 | 1257 | |
fbf1e473 AM |
1258 | /* |
1259 | * We need to do this right in the beginning since | |
1260 | * alloc_arraycache's are going to use this list. | |
1261 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1262 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1263 | */ |
6a67368c | 1264 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1265 | if (err < 0) |
1266 | goto bad; | |
fbf1e473 AM |
1267 | |
1268 | /* | |
1269 | * Now we can go ahead with allocating the shared arrays and | |
1270 | * array caches | |
1271 | */ | |
18004c5d | 1272 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1273 | struct array_cache *nc; |
1274 | struct array_cache *shared = NULL; | |
1275 | struct array_cache **alien = NULL; | |
1276 | ||
1277 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1278 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1279 | if (!nc) |
1280 | goto bad; | |
1281 | if (cachep->shared) { | |
1282 | shared = alloc_arraycache(node, | |
1283 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1284 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1285 | if (!shared) { |
1286 | kfree(nc); | |
1da177e4 | 1287 | goto bad; |
12d00f6a | 1288 | } |
fbf1e473 AM |
1289 | } |
1290 | if (use_alien_caches) { | |
83b519e8 | 1291 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1292 | if (!alien) { |
1293 | kfree(shared); | |
1294 | kfree(nc); | |
fbf1e473 | 1295 | goto bad; |
12d00f6a | 1296 | } |
fbf1e473 AM |
1297 | } |
1298 | cachep->array[cpu] = nc; | |
ce8eb6c4 CL |
1299 | n = cachep->node[node]; |
1300 | BUG_ON(!n); | |
fbf1e473 | 1301 | |
ce8eb6c4 CL |
1302 | spin_lock_irq(&n->list_lock); |
1303 | if (!n->shared) { | |
fbf1e473 AM |
1304 | /* |
1305 | * We are serialised from CPU_DEAD or | |
1306 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1307 | */ | |
ce8eb6c4 | 1308 | n->shared = shared; |
fbf1e473 AM |
1309 | shared = NULL; |
1310 | } | |
4484ebf1 | 1311 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1312 | if (!n->alien) { |
1313 | n->alien = alien; | |
fbf1e473 | 1314 | alien = NULL; |
1da177e4 | 1315 | } |
fbf1e473 | 1316 | #endif |
ce8eb6c4 | 1317 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1318 | kfree(shared); |
1319 | free_alien_cache(alien); | |
83835b3d PZ |
1320 | if (cachep->flags & SLAB_DEBUG_OBJECTS) |
1321 | slab_set_debugobj_lock_classes_node(cachep, node); | |
6ccfb5bc GC |
1322 | else if (!OFF_SLAB(cachep) && |
1323 | !(cachep->flags & SLAB_DESTROY_BY_RCU)) | |
1324 | on_slab_lock_classes_node(cachep, node); | |
fbf1e473 | 1325 | } |
ce79ddc8 PE |
1326 | init_node_lock_keys(node); |
1327 | ||
fbf1e473 AM |
1328 | return 0; |
1329 | bad: | |
12d00f6a | 1330 | cpuup_canceled(cpu); |
fbf1e473 AM |
1331 | return -ENOMEM; |
1332 | } | |
1333 | ||
0db0628d | 1334 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1335 | unsigned long action, void *hcpu) |
1336 | { | |
1337 | long cpu = (long)hcpu; | |
1338 | int err = 0; | |
1339 | ||
1340 | switch (action) { | |
fbf1e473 AM |
1341 | case CPU_UP_PREPARE: |
1342 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1343 | mutex_lock(&slab_mutex); |
fbf1e473 | 1344 | err = cpuup_prepare(cpu); |
18004c5d | 1345 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1346 | break; |
1347 | case CPU_ONLINE: | |
8bb78442 | 1348 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1349 | start_cpu_timer(cpu); |
1350 | break; | |
1351 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1352 | case CPU_DOWN_PREPARE: |
8bb78442 | 1353 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1354 | /* |
18004c5d | 1355 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1356 | * held so that if cache_reap() is invoked it cannot do |
1357 | * anything expensive but will only modify reap_work | |
1358 | * and reschedule the timer. | |
1359 | */ | |
afe2c511 | 1360 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1361 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1362 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1363 | break; |
1364 | case CPU_DOWN_FAILED: | |
8bb78442 | 1365 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1366 | start_cpu_timer(cpu); |
1367 | break; | |
1da177e4 | 1368 | case CPU_DEAD: |
8bb78442 | 1369 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1370 | /* |
1371 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1372 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1373 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1374 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1375 | * structure is usually allocated from kmem_cache_create() and |
1376 | * gets destroyed at kmem_cache_destroy(). | |
1377 | */ | |
183ff22b | 1378 | /* fall through */ |
8f5be20b | 1379 | #endif |
1da177e4 | 1380 | case CPU_UP_CANCELED: |
8bb78442 | 1381 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1382 | mutex_lock(&slab_mutex); |
fbf1e473 | 1383 | cpuup_canceled(cpu); |
18004c5d | 1384 | mutex_unlock(&slab_mutex); |
1da177e4 | 1385 | break; |
1da177e4 | 1386 | } |
eac40680 | 1387 | return notifier_from_errno(err); |
1da177e4 LT |
1388 | } |
1389 | ||
0db0628d | 1390 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1391 | &cpuup_callback, NULL, 0 |
1392 | }; | |
1da177e4 | 1393 | |
8f9f8d9e DR |
1394 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1395 | /* | |
1396 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1397 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1398 | * removed. | |
1399 | * | |
18004c5d | 1400 | * Must hold slab_mutex. |
8f9f8d9e | 1401 | */ |
6a67368c | 1402 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1403 | { |
1404 | struct kmem_cache *cachep; | |
1405 | int ret = 0; | |
1406 | ||
18004c5d | 1407 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1408 | struct kmem_cache_node *n; |
8f9f8d9e | 1409 | |
ce8eb6c4 CL |
1410 | n = cachep->node[node]; |
1411 | if (!n) | |
8f9f8d9e DR |
1412 | continue; |
1413 | ||
0fa8103b | 1414 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1415 | |
ce8eb6c4 CL |
1416 | if (!list_empty(&n->slabs_full) || |
1417 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1418 | ret = -EBUSY; |
1419 | break; | |
1420 | } | |
1421 | } | |
1422 | return ret; | |
1423 | } | |
1424 | ||
1425 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1426 | unsigned long action, void *arg) | |
1427 | { | |
1428 | struct memory_notify *mnb = arg; | |
1429 | int ret = 0; | |
1430 | int nid; | |
1431 | ||
1432 | nid = mnb->status_change_nid; | |
1433 | if (nid < 0) | |
1434 | goto out; | |
1435 | ||
1436 | switch (action) { | |
1437 | case MEM_GOING_ONLINE: | |
18004c5d | 1438 | mutex_lock(&slab_mutex); |
6a67368c | 1439 | ret = init_cache_node_node(nid); |
18004c5d | 1440 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1441 | break; |
1442 | case MEM_GOING_OFFLINE: | |
18004c5d | 1443 | mutex_lock(&slab_mutex); |
6a67368c | 1444 | ret = drain_cache_node_node(nid); |
18004c5d | 1445 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1446 | break; |
1447 | case MEM_ONLINE: | |
1448 | case MEM_OFFLINE: | |
1449 | case MEM_CANCEL_ONLINE: | |
1450 | case MEM_CANCEL_OFFLINE: | |
1451 | break; | |
1452 | } | |
1453 | out: | |
5fda1bd5 | 1454 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1455 | } |
1456 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1457 | ||
e498be7d | 1458 | /* |
ce8eb6c4 | 1459 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1460 | */ |
6744f087 | 1461 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1462 | int nodeid) |
e498be7d | 1463 | { |
6744f087 | 1464 | struct kmem_cache_node *ptr; |
e498be7d | 1465 | |
6744f087 | 1466 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1467 | BUG_ON(!ptr); |
1468 | ||
6744f087 | 1469 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1470 | /* |
1471 | * Do not assume that spinlocks can be initialized via memcpy: | |
1472 | */ | |
1473 | spin_lock_init(&ptr->list_lock); | |
1474 | ||
e498be7d | 1475 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1476 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1477 | } |
1478 | ||
556a169d | 1479 | /* |
ce8eb6c4 CL |
1480 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1481 | * size of kmem_cache_node. | |
556a169d | 1482 | */ |
ce8eb6c4 | 1483 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1484 | { |
1485 | int node; | |
1486 | ||
1487 | for_each_online_node(node) { | |
ce8eb6c4 | 1488 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1489 | cachep->node[node]->next_reap = jiffies + |
556a169d PE |
1490 | REAPTIMEOUT_LIST3 + |
1491 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
1492 | } | |
1493 | } | |
1494 | ||
3c583465 CL |
1495 | /* |
1496 | * The memory after the last cpu cache pointer is used for the | |
6a67368c | 1497 | * the node pointer. |
3c583465 | 1498 | */ |
6a67368c | 1499 | static void setup_node_pointer(struct kmem_cache *cachep) |
3c583465 | 1500 | { |
6a67368c | 1501 | cachep->node = (struct kmem_cache_node **)&cachep->array[nr_cpu_ids]; |
3c583465 CL |
1502 | } |
1503 | ||
a737b3e2 AM |
1504 | /* |
1505 | * Initialisation. Called after the page allocator have been initialised and | |
1506 | * before smp_init(). | |
1da177e4 LT |
1507 | */ |
1508 | void __init kmem_cache_init(void) | |
1509 | { | |
e498be7d CL |
1510 | int i; |
1511 | ||
9b030cb8 | 1512 | kmem_cache = &kmem_cache_boot; |
6a67368c | 1513 | setup_node_pointer(kmem_cache); |
9b030cb8 | 1514 | |
b6e68bc1 | 1515 | if (num_possible_nodes() == 1) |
62918a03 SS |
1516 | use_alien_caches = 0; |
1517 | ||
3c583465 | 1518 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1519 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1520 | |
ce8eb6c4 | 1521 | set_up_node(kmem_cache, CACHE_CACHE); |
1da177e4 LT |
1522 | |
1523 | /* | |
1524 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1525 | * page orders on machines with more than 32MB of memory if |
1526 | * not overridden on the command line. | |
1da177e4 | 1527 | */ |
3df1cccd | 1528 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1529 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1530 | |
1da177e4 LT |
1531 | /* Bootstrap is tricky, because several objects are allocated |
1532 | * from caches that do not exist yet: | |
9b030cb8 CL |
1533 | * 1) initialize the kmem_cache cache: it contains the struct |
1534 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1535 | * kmem_cache is statically allocated. | |
e498be7d | 1536 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1537 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1538 | * array at the end of the bootstrap. |
1da177e4 | 1539 | * 2) Create the first kmalloc cache. |
343e0d7a | 1540 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1541 | * An __init data area is used for the head array. |
1542 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1543 | * head arrays. | |
9b030cb8 | 1544 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1545 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1546 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1547 | * the other cache's with kmalloc allocated memory. |
1548 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1549 | */ |
1550 | ||
9b030cb8 | 1551 | /* 1) create the kmem_cache */ |
1da177e4 | 1552 | |
8da3430d | 1553 | /* |
b56efcf0 | 1554 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1555 | */ |
2f9baa9f CL |
1556 | create_boot_cache(kmem_cache, "kmem_cache", |
1557 | offsetof(struct kmem_cache, array[nr_cpu_ids]) + | |
6744f087 | 1558 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1559 | SLAB_HWCACHE_ALIGN); |
1560 | list_add(&kmem_cache->list, &slab_caches); | |
1da177e4 LT |
1561 | |
1562 | /* 2+3) create the kmalloc caches */ | |
1da177e4 | 1563 | |
a737b3e2 AM |
1564 | /* |
1565 | * Initialize the caches that provide memory for the array cache and the | |
ce8eb6c4 | 1566 | * kmem_cache_node structures first. Without this, further allocations will |
a737b3e2 | 1567 | * bug. |
e498be7d CL |
1568 | */ |
1569 | ||
e3366016 CL |
1570 | kmalloc_caches[INDEX_AC] = create_kmalloc_cache("kmalloc-ac", |
1571 | kmalloc_size(INDEX_AC), ARCH_KMALLOC_FLAGS); | |
45530c44 | 1572 | |
ce8eb6c4 CL |
1573 | if (INDEX_AC != INDEX_NODE) |
1574 | kmalloc_caches[INDEX_NODE] = | |
1575 | create_kmalloc_cache("kmalloc-node", | |
1576 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); | |
e498be7d | 1577 | |
e0a42726 IM |
1578 | slab_early_init = 0; |
1579 | ||
1da177e4 LT |
1580 | /* 4) Replace the bootstrap head arrays */ |
1581 | { | |
2b2d5493 | 1582 | struct array_cache *ptr; |
e498be7d | 1583 | |
83b519e8 | 1584 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1585 | |
9b030cb8 | 1586 | memcpy(ptr, cpu_cache_get(kmem_cache), |
b28a02de | 1587 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1588 | /* |
1589 | * Do not assume that spinlocks can be initialized via memcpy: | |
1590 | */ | |
1591 | spin_lock_init(&ptr->lock); | |
1592 | ||
9b030cb8 | 1593 | kmem_cache->array[smp_processor_id()] = ptr; |
e498be7d | 1594 | |
83b519e8 | 1595 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1596 | |
e3366016 | 1597 | BUG_ON(cpu_cache_get(kmalloc_caches[INDEX_AC]) |
b28a02de | 1598 | != &initarray_generic.cache); |
e3366016 | 1599 | memcpy(ptr, cpu_cache_get(kmalloc_caches[INDEX_AC]), |
b28a02de | 1600 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1601 | /* |
1602 | * Do not assume that spinlocks can be initialized via memcpy: | |
1603 | */ | |
1604 | spin_lock_init(&ptr->lock); | |
1605 | ||
e3366016 | 1606 | kmalloc_caches[INDEX_AC]->array[smp_processor_id()] = ptr; |
1da177e4 | 1607 | } |
ce8eb6c4 | 1608 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1609 | { |
1ca4cb24 PE |
1610 | int nid; |
1611 | ||
9c09a95c | 1612 | for_each_online_node(nid) { |
ce8eb6c4 | 1613 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1614 | |
e3366016 | 1615 | init_list(kmalloc_caches[INDEX_AC], |
ce8eb6c4 | 1616 | &init_kmem_cache_node[SIZE_AC + nid], nid); |
e498be7d | 1617 | |
ce8eb6c4 CL |
1618 | if (INDEX_AC != INDEX_NODE) { |
1619 | init_list(kmalloc_caches[INDEX_NODE], | |
1620 | &init_kmem_cache_node[SIZE_NODE + nid], nid); | |
e498be7d CL |
1621 | } |
1622 | } | |
1623 | } | |
1da177e4 | 1624 | |
f97d5f63 | 1625 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1626 | } |
1627 | ||
1628 | void __init kmem_cache_init_late(void) | |
1629 | { | |
1630 | struct kmem_cache *cachep; | |
1631 | ||
97d06609 | 1632 | slab_state = UP; |
52cef189 | 1633 | |
8429db5c | 1634 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1635 | mutex_lock(&slab_mutex); |
1636 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1637 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1638 | BUG(); | |
18004c5d | 1639 | mutex_unlock(&slab_mutex); |
056c6241 | 1640 | |
947ca185 MW |
1641 | /* Annotate slab for lockdep -- annotate the malloc caches */ |
1642 | init_lock_keys(); | |
1643 | ||
97d06609 CL |
1644 | /* Done! */ |
1645 | slab_state = FULL; | |
1646 | ||
a737b3e2 AM |
1647 | /* |
1648 | * Register a cpu startup notifier callback that initializes | |
1649 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1650 | */ |
1651 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1652 | |
8f9f8d9e DR |
1653 | #ifdef CONFIG_NUMA |
1654 | /* | |
1655 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1656 | * node. |
8f9f8d9e DR |
1657 | */ |
1658 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1659 | #endif | |
1660 | ||
a737b3e2 AM |
1661 | /* |
1662 | * The reap timers are started later, with a module init call: That part | |
1663 | * of the kernel is not yet operational. | |
1da177e4 LT |
1664 | */ |
1665 | } | |
1666 | ||
1667 | static int __init cpucache_init(void) | |
1668 | { | |
1669 | int cpu; | |
1670 | ||
a737b3e2 AM |
1671 | /* |
1672 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1673 | */ |
e498be7d | 1674 | for_each_online_cpu(cpu) |
a737b3e2 | 1675 | start_cpu_timer(cpu); |
a164f896 GC |
1676 | |
1677 | /* Done! */ | |
97d06609 | 1678 | slab_state = FULL; |
1da177e4 LT |
1679 | return 0; |
1680 | } | |
1da177e4 LT |
1681 | __initcall(cpucache_init); |
1682 | ||
8bdec192 RA |
1683 | static noinline void |
1684 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1685 | { | |
ce8eb6c4 | 1686 | struct kmem_cache_node *n; |
8bdec192 RA |
1687 | struct slab *slabp; |
1688 | unsigned long flags; | |
1689 | int node; | |
1690 | ||
1691 | printk(KERN_WARNING | |
1692 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1693 | nodeid, gfpflags); | |
1694 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1695 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 RA |
1696 | |
1697 | for_each_online_node(node) { | |
1698 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; | |
1699 | unsigned long active_slabs = 0, num_slabs = 0; | |
1700 | ||
ce8eb6c4 CL |
1701 | n = cachep->node[node]; |
1702 | if (!n) | |
8bdec192 RA |
1703 | continue; |
1704 | ||
ce8eb6c4 CL |
1705 | spin_lock_irqsave(&n->list_lock, flags); |
1706 | list_for_each_entry(slabp, &n->slabs_full, list) { | |
8bdec192 RA |
1707 | active_objs += cachep->num; |
1708 | active_slabs++; | |
1709 | } | |
ce8eb6c4 | 1710 | list_for_each_entry(slabp, &n->slabs_partial, list) { |
8bdec192 RA |
1711 | active_objs += slabp->inuse; |
1712 | active_slabs++; | |
1713 | } | |
ce8eb6c4 | 1714 | list_for_each_entry(slabp, &n->slabs_free, list) |
8bdec192 RA |
1715 | num_slabs++; |
1716 | ||
ce8eb6c4 CL |
1717 | free_objects += n->free_objects; |
1718 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1719 | |
1720 | num_slabs += active_slabs; | |
1721 | num_objs = num_slabs * cachep->num; | |
1722 | printk(KERN_WARNING | |
1723 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1724 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1725 | free_objects); | |
1726 | } | |
1727 | } | |
1728 | ||
1da177e4 LT |
1729 | /* |
1730 | * Interface to system's page allocator. No need to hold the cache-lock. | |
1731 | * | |
1732 | * If we requested dmaable memory, we will get it. Even if we | |
1733 | * did not request dmaable memory, we might get it, but that | |
1734 | * would be relatively rare and ignorable. | |
1735 | */ | |
0c3aa83e JK |
1736 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1737 | int nodeid) | |
1da177e4 LT |
1738 | { |
1739 | struct page *page; | |
e1b6aa6f | 1740 | int nr_pages; |
1da177e4 LT |
1741 | int i; |
1742 | ||
d6fef9da | 1743 | #ifndef CONFIG_MMU |
e1b6aa6f CH |
1744 | /* |
1745 | * Nommu uses slab's for process anonymous memory allocations, and thus | |
1746 | * requires __GFP_COMP to properly refcount higher order allocations | |
d6fef9da | 1747 | */ |
e1b6aa6f | 1748 | flags |= __GFP_COMP; |
d6fef9da | 1749 | #endif |
765c4507 | 1750 | |
a618e89f | 1751 | flags |= cachep->allocflags; |
e12ba74d MG |
1752 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1753 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1754 | |
517d0869 | 1755 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 RA |
1756 | if (!page) { |
1757 | if (!(flags & __GFP_NOWARN) && printk_ratelimit()) | |
1758 | slab_out_of_memory(cachep, flags, nodeid); | |
1da177e4 | 1759 | return NULL; |
8bdec192 | 1760 | } |
1da177e4 | 1761 | |
b37f1dd0 | 1762 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
072bb0aa MG |
1763 | if (unlikely(page->pfmemalloc)) |
1764 | pfmemalloc_active = true; | |
1765 | ||
e1b6aa6f | 1766 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1767 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1768 | add_zone_page_state(page_zone(page), |
1769 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1770 | else | |
1771 | add_zone_page_state(page_zone(page), | |
1772 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
072bb0aa | 1773 | for (i = 0; i < nr_pages; i++) { |
e1b6aa6f | 1774 | __SetPageSlab(page + i); |
c175eea4 | 1775 | |
072bb0aa | 1776 | if (page->pfmemalloc) |
73293c2f | 1777 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1778 | } |
1f458cbf | 1779 | memcg_bind_pages(cachep, cachep->gfporder); |
072bb0aa | 1780 | |
b1eeab67 VN |
1781 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1782 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1783 | ||
1784 | if (cachep->ctor) | |
1785 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1786 | else | |
1787 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1788 | } | |
c175eea4 | 1789 | |
0c3aa83e | 1790 | return page; |
1da177e4 LT |
1791 | } |
1792 | ||
1793 | /* | |
1794 | * Interface to system's page release. | |
1795 | */ | |
0c3aa83e | 1796 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1797 | { |
b28a02de | 1798 | unsigned long i = (1 << cachep->gfporder); |
1da177e4 LT |
1799 | const unsigned long nr_freed = i; |
1800 | ||
b1eeab67 | 1801 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1802 | |
972d1a7b CL |
1803 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1804 | sub_zone_page_state(page_zone(page), | |
1805 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1806 | else | |
1807 | sub_zone_page_state(page_zone(page), | |
1808 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f JK |
1809 | |
1810 | __ClearPageSlabPfmemalloc(page); | |
1da177e4 | 1811 | while (i--) { |
f205b2fe NP |
1812 | BUG_ON(!PageSlab(page)); |
1813 | __ClearPageSlab(page); | |
1da177e4 LT |
1814 | page++; |
1815 | } | |
1f458cbf GC |
1816 | |
1817 | memcg_release_pages(cachep, cachep->gfporder); | |
1da177e4 LT |
1818 | if (current->reclaim_state) |
1819 | current->reclaim_state->reclaimed_slab += nr_freed; | |
0c3aa83e | 1820 | __free_memcg_kmem_pages(page, cachep->gfporder); |
1da177e4 LT |
1821 | } |
1822 | ||
1823 | static void kmem_rcu_free(struct rcu_head *head) | |
1824 | { | |
b28a02de | 1825 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
07d417a1 | 1826 | struct kmem_cache *cachep = slab_rcu->page->slab_cache; |
1da177e4 | 1827 | |
0c3aa83e | 1828 | kmem_freepages(cachep, slab_rcu->page); |
1da177e4 LT |
1829 | if (OFF_SLAB(cachep)) |
1830 | kmem_cache_free(cachep->slabp_cache, slab_rcu); | |
1831 | } | |
1832 | ||
1833 | #if DEBUG | |
1834 | ||
1835 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1836 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1837 | unsigned long caller) |
1da177e4 | 1838 | { |
8c138bc0 | 1839 | int size = cachep->object_size; |
1da177e4 | 1840 | |
3dafccf2 | 1841 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1842 | |
b28a02de | 1843 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1844 | return; |
1845 | ||
b28a02de PE |
1846 | *addr++ = 0x12345678; |
1847 | *addr++ = caller; | |
1848 | *addr++ = smp_processor_id(); | |
1849 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1850 | { |
1851 | unsigned long *sptr = &caller; | |
1852 | unsigned long svalue; | |
1853 | ||
1854 | while (!kstack_end(sptr)) { | |
1855 | svalue = *sptr++; | |
1856 | if (kernel_text_address(svalue)) { | |
b28a02de | 1857 | *addr++ = svalue; |
1da177e4 LT |
1858 | size -= sizeof(unsigned long); |
1859 | if (size <= sizeof(unsigned long)) | |
1860 | break; | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | } | |
b28a02de | 1865 | *addr++ = 0x87654321; |
1da177e4 LT |
1866 | } |
1867 | #endif | |
1868 | ||
343e0d7a | 1869 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1870 | { |
8c138bc0 | 1871 | int size = cachep->object_size; |
3dafccf2 | 1872 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1873 | |
1874 | memset(addr, val, size); | |
b28a02de | 1875 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1876 | } |
1877 | ||
1878 | static void dump_line(char *data, int offset, int limit) | |
1879 | { | |
1880 | int i; | |
aa83aa40 DJ |
1881 | unsigned char error = 0; |
1882 | int bad_count = 0; | |
1883 | ||
fdde6abb | 1884 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1885 | for (i = 0; i < limit; i++) { |
1886 | if (data[offset + i] != POISON_FREE) { | |
1887 | error = data[offset + i]; | |
1888 | bad_count++; | |
1889 | } | |
aa83aa40 | 1890 | } |
fdde6abb SAS |
1891 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1892 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1893 | |
1894 | if (bad_count == 1) { | |
1895 | error ^= POISON_FREE; | |
1896 | if (!(error & (error - 1))) { | |
1897 | printk(KERN_ERR "Single bit error detected. Probably " | |
1898 | "bad RAM.\n"); | |
1899 | #ifdef CONFIG_X86 | |
1900 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1901 | "test tool.\n"); | |
1902 | #else | |
1903 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1904 | #endif | |
1905 | } | |
1906 | } | |
1da177e4 LT |
1907 | } |
1908 | #endif | |
1909 | ||
1910 | #if DEBUG | |
1911 | ||
343e0d7a | 1912 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1913 | { |
1914 | int i, size; | |
1915 | char *realobj; | |
1916 | ||
1917 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1918 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1919 | *dbg_redzone1(cachep, objp), |
1920 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1921 | } |
1922 | ||
1923 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1924 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1925 | *dbg_userword(cachep, objp), | |
1926 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1927 | } |
3dafccf2 | 1928 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1929 | size = cachep->object_size; |
b28a02de | 1930 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1931 | int limit; |
1932 | limit = 16; | |
b28a02de PE |
1933 | if (i + limit > size) |
1934 | limit = size - i; | |
1da177e4 LT |
1935 | dump_line(realobj, i, limit); |
1936 | } | |
1937 | } | |
1938 | ||
343e0d7a | 1939 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1940 | { |
1941 | char *realobj; | |
1942 | int size, i; | |
1943 | int lines = 0; | |
1944 | ||
3dafccf2 | 1945 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1946 | size = cachep->object_size; |
1da177e4 | 1947 | |
b28a02de | 1948 | for (i = 0; i < size; i++) { |
1da177e4 | 1949 | char exp = POISON_FREE; |
b28a02de | 1950 | if (i == size - 1) |
1da177e4 LT |
1951 | exp = POISON_END; |
1952 | if (realobj[i] != exp) { | |
1953 | int limit; | |
1954 | /* Mismatch ! */ | |
1955 | /* Print header */ | |
1956 | if (lines == 0) { | |
b28a02de | 1957 | printk(KERN_ERR |
face37f5 DJ |
1958 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1959 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1960 | print_objinfo(cachep, objp, 0); |
1961 | } | |
1962 | /* Hexdump the affected line */ | |
b28a02de | 1963 | i = (i / 16) * 16; |
1da177e4 | 1964 | limit = 16; |
b28a02de PE |
1965 | if (i + limit > size) |
1966 | limit = size - i; | |
1da177e4 LT |
1967 | dump_line(realobj, i, limit); |
1968 | i += 16; | |
1969 | lines++; | |
1970 | /* Limit to 5 lines */ | |
1971 | if (lines > 5) | |
1972 | break; | |
1973 | } | |
1974 | } | |
1975 | if (lines != 0) { | |
1976 | /* Print some data about the neighboring objects, if they | |
1977 | * exist: | |
1978 | */ | |
6ed5eb22 | 1979 | struct slab *slabp = virt_to_slab(objp); |
8fea4e96 | 1980 | unsigned int objnr; |
1da177e4 | 1981 | |
8fea4e96 | 1982 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 | 1983 | if (objnr) { |
8fea4e96 | 1984 | objp = index_to_obj(cachep, slabp, objnr - 1); |
3dafccf2 | 1985 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1986 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1987 | realobj, size); |
1da177e4 LT |
1988 | print_objinfo(cachep, objp, 2); |
1989 | } | |
b28a02de | 1990 | if (objnr + 1 < cachep->num) { |
8fea4e96 | 1991 | objp = index_to_obj(cachep, slabp, objnr + 1); |
3dafccf2 | 1992 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1993 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1994 | realobj, size); |
1da177e4 LT |
1995 | print_objinfo(cachep, objp, 2); |
1996 | } | |
1997 | } | |
1998 | } | |
1999 | #endif | |
2000 | ||
12dd36fa | 2001 | #if DEBUG |
e79aec29 | 2002 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 | 2003 | { |
1da177e4 LT |
2004 | int i; |
2005 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 2006 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
2007 | |
2008 | if (cachep->flags & SLAB_POISON) { | |
2009 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2010 | if (cachep->size % PAGE_SIZE == 0 && |
a737b3e2 | 2011 | OFF_SLAB(cachep)) |
b28a02de | 2012 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2013 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
2014 | else |
2015 | check_poison_obj(cachep, objp); | |
2016 | #else | |
2017 | check_poison_obj(cachep, objp); | |
2018 | #endif | |
2019 | } | |
2020 | if (cachep->flags & SLAB_RED_ZONE) { | |
2021 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
2022 | slab_error(cachep, "start of a freed object " | |
b28a02de | 2023 | "was overwritten"); |
1da177e4 LT |
2024 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
2025 | slab_error(cachep, "end of a freed object " | |
b28a02de | 2026 | "was overwritten"); |
1da177e4 | 2027 | } |
1da177e4 | 2028 | } |
12dd36fa | 2029 | } |
1da177e4 | 2030 | #else |
e79aec29 | 2031 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa | 2032 | { |
12dd36fa | 2033 | } |
1da177e4 LT |
2034 | #endif |
2035 | ||
911851e6 RD |
2036 | /** |
2037 | * slab_destroy - destroy and release all objects in a slab | |
2038 | * @cachep: cache pointer being destroyed | |
2039 | * @slabp: slab pointer being destroyed | |
2040 | * | |
12dd36fa | 2041 | * Destroy all the objs in a slab, and release the mem back to the system. |
a737b3e2 AM |
2042 | * Before calling the slab must have been unlinked from the cache. The |
2043 | * cache-lock is not held/needed. | |
12dd36fa | 2044 | */ |
343e0d7a | 2045 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa | 2046 | { |
0c3aa83e | 2047 | struct page *page = virt_to_head_page(slabp->s_mem); |
12dd36fa | 2048 | |
e79aec29 | 2049 | slab_destroy_debugcheck(cachep, slabp); |
1da177e4 LT |
2050 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
2051 | struct slab_rcu *slab_rcu; | |
2052 | ||
b28a02de | 2053 | slab_rcu = (struct slab_rcu *)slabp; |
0c3aa83e | 2054 | slab_rcu->page = page; |
1da177e4 LT |
2055 | call_rcu(&slab_rcu->head, kmem_rcu_free); |
2056 | } else { | |
0c3aa83e | 2057 | kmem_freepages(cachep, page); |
873623df IM |
2058 | if (OFF_SLAB(cachep)) |
2059 | kmem_cache_free(cachep->slabp_cache, slabp); | |
1da177e4 LT |
2060 | } |
2061 | } | |
2062 | ||
4d268eba | 2063 | /** |
a70773dd RD |
2064 | * calculate_slab_order - calculate size (page order) of slabs |
2065 | * @cachep: pointer to the cache that is being created | |
2066 | * @size: size of objects to be created in this cache. | |
2067 | * @align: required alignment for the objects. | |
2068 | * @flags: slab allocation flags | |
2069 | * | |
2070 | * Also calculates the number of objects per slab. | |
4d268eba PE |
2071 | * |
2072 | * This could be made much more intelligent. For now, try to avoid using | |
2073 | * high order pages for slabs. When the gfp() functions are more friendly | |
2074 | * towards high-order requests, this should be changed. | |
2075 | */ | |
a737b3e2 | 2076 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 2077 | size_t size, size_t align, unsigned long flags) |
4d268eba | 2078 | { |
b1ab41c4 | 2079 | unsigned long offslab_limit; |
4d268eba | 2080 | size_t left_over = 0; |
9888e6fa | 2081 | int gfporder; |
4d268eba | 2082 | |
0aa817f0 | 2083 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
2084 | unsigned int num; |
2085 | size_t remainder; | |
2086 | ||
9888e6fa | 2087 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
2088 | if (!num) |
2089 | continue; | |
9888e6fa | 2090 | |
b1ab41c4 IM |
2091 | if (flags & CFLGS_OFF_SLAB) { |
2092 | /* | |
2093 | * Max number of objs-per-slab for caches which | |
2094 | * use off-slab slabs. Needed to avoid a possible | |
2095 | * looping condition in cache_grow(). | |
2096 | */ | |
2097 | offslab_limit = size - sizeof(struct slab); | |
2098 | offslab_limit /= sizeof(kmem_bufctl_t); | |
2099 | ||
2100 | if (num > offslab_limit) | |
2101 | break; | |
2102 | } | |
4d268eba | 2103 | |
9888e6fa | 2104 | /* Found something acceptable - save it away */ |
4d268eba | 2105 | cachep->num = num; |
9888e6fa | 2106 | cachep->gfporder = gfporder; |
4d268eba PE |
2107 | left_over = remainder; |
2108 | ||
f78bb8ad LT |
2109 | /* |
2110 | * A VFS-reclaimable slab tends to have most allocations | |
2111 | * as GFP_NOFS and we really don't want to have to be allocating | |
2112 | * higher-order pages when we are unable to shrink dcache. | |
2113 | */ | |
2114 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
2115 | break; | |
2116 | ||
4d268eba PE |
2117 | /* |
2118 | * Large number of objects is good, but very large slabs are | |
2119 | * currently bad for the gfp()s. | |
2120 | */ | |
543585cc | 2121 | if (gfporder >= slab_max_order) |
4d268eba PE |
2122 | break; |
2123 | ||
9888e6fa LT |
2124 | /* |
2125 | * Acceptable internal fragmentation? | |
2126 | */ | |
a737b3e2 | 2127 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
2128 | break; |
2129 | } | |
2130 | return left_over; | |
2131 | } | |
2132 | ||
83b519e8 | 2133 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2134 | { |
97d06609 | 2135 | if (slab_state >= FULL) |
83b519e8 | 2136 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2137 | |
97d06609 | 2138 | if (slab_state == DOWN) { |
f30cf7d1 | 2139 | /* |
2f9baa9f | 2140 | * Note: Creation of first cache (kmem_cache). |
ce8eb6c4 | 2141 | * The setup_node is taken care |
2f9baa9f CL |
2142 | * of by the caller of __kmem_cache_create |
2143 | */ | |
2144 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2145 | slab_state = PARTIAL; | |
2146 | } else if (slab_state == PARTIAL) { | |
2147 | /* | |
2148 | * Note: the second kmem_cache_create must create the cache | |
f30cf7d1 PE |
2149 | * that's used by kmalloc(24), otherwise the creation of |
2150 | * further caches will BUG(). | |
2151 | */ | |
2152 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2153 | ||
2154 | /* | |
ce8eb6c4 CL |
2155 | * If the cache that's used by kmalloc(sizeof(kmem_cache_node)) is |
2156 | * the second cache, then we need to set up all its node/, | |
f30cf7d1 PE |
2157 | * otherwise the creation of further caches will BUG(). |
2158 | */ | |
ce8eb6c4 CL |
2159 | set_up_node(cachep, SIZE_AC); |
2160 | if (INDEX_AC == INDEX_NODE) | |
2161 | slab_state = PARTIAL_NODE; | |
f30cf7d1 | 2162 | else |
97d06609 | 2163 | slab_state = PARTIAL_ARRAYCACHE; |
f30cf7d1 | 2164 | } else { |
2f9baa9f | 2165 | /* Remaining boot caches */ |
f30cf7d1 | 2166 | cachep->array[smp_processor_id()] = |
83b519e8 | 2167 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 | 2168 | |
97d06609 | 2169 | if (slab_state == PARTIAL_ARRAYCACHE) { |
ce8eb6c4 CL |
2170 | set_up_node(cachep, SIZE_NODE); |
2171 | slab_state = PARTIAL_NODE; | |
f30cf7d1 PE |
2172 | } else { |
2173 | int node; | |
556a169d | 2174 | for_each_online_node(node) { |
6a67368c | 2175 | cachep->node[node] = |
6744f087 | 2176 | kmalloc_node(sizeof(struct kmem_cache_node), |
eb91f1d0 | 2177 | gfp, node); |
6a67368c | 2178 | BUG_ON(!cachep->node[node]); |
ce8eb6c4 | 2179 | kmem_cache_node_init(cachep->node[node]); |
f30cf7d1 PE |
2180 | } |
2181 | } | |
2182 | } | |
6a67368c | 2183 | cachep->node[numa_mem_id()]->next_reap = |
f30cf7d1 PE |
2184 | jiffies + REAPTIMEOUT_LIST3 + |
2185 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
2186 | ||
2187 | cpu_cache_get(cachep)->avail = 0; | |
2188 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2189 | cpu_cache_get(cachep)->batchcount = 1; | |
2190 | cpu_cache_get(cachep)->touched = 0; | |
2191 | cachep->batchcount = 1; | |
2192 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2193 | return 0; |
f30cf7d1 PE |
2194 | } |
2195 | ||
1da177e4 | 2196 | /** |
039363f3 | 2197 | * __kmem_cache_create - Create a cache. |
a755b76a | 2198 | * @cachep: cache management descriptor |
1da177e4 | 2199 | * @flags: SLAB flags |
1da177e4 LT |
2200 | * |
2201 | * Returns a ptr to the cache on success, NULL on failure. | |
2202 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2203 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2204 | * |
1da177e4 LT |
2205 | * The flags are |
2206 | * | |
2207 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2208 | * to catch references to uninitialised memory. | |
2209 | * | |
2210 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2211 | * for buffer overruns. | |
2212 | * | |
1da177e4 LT |
2213 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2214 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2215 | * as davem. | |
2216 | */ | |
278b1bb1 | 2217 | int |
8a13a4cc | 2218 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 LT |
2219 | { |
2220 | size_t left_over, slab_size, ralign; | |
83b519e8 | 2221 | gfp_t gfp; |
278b1bb1 | 2222 | int err; |
8a13a4cc | 2223 | size_t size = cachep->size; |
1da177e4 | 2224 | |
1da177e4 | 2225 | #if DEBUG |
1da177e4 LT |
2226 | #if FORCED_DEBUG |
2227 | /* | |
2228 | * Enable redzoning and last user accounting, except for caches with | |
2229 | * large objects, if the increased size would increase the object size | |
2230 | * above the next power of two: caches with object sizes just above a | |
2231 | * power of two have a significant amount of internal fragmentation. | |
2232 | */ | |
87a927c7 DW |
2233 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2234 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2235 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2236 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2237 | flags |= SLAB_POISON; | |
2238 | #endif | |
2239 | if (flags & SLAB_DESTROY_BY_RCU) | |
2240 | BUG_ON(flags & SLAB_POISON); | |
2241 | #endif | |
1da177e4 | 2242 | |
a737b3e2 AM |
2243 | /* |
2244 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2245 | * unaligned accesses for some archs when redzoning is used, and makes |
2246 | * sure any on-slab bufctl's are also correctly aligned. | |
2247 | */ | |
b28a02de PE |
2248 | if (size & (BYTES_PER_WORD - 1)) { |
2249 | size += (BYTES_PER_WORD - 1); | |
2250 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2251 | } |
2252 | ||
ca5f9703 | 2253 | /* |
87a927c7 DW |
2254 | * Redzoning and user store require word alignment or possibly larger. |
2255 | * Note this will be overridden by architecture or caller mandated | |
2256 | * alignment if either is greater than BYTES_PER_WORD. | |
ca5f9703 | 2257 | */ |
87a927c7 DW |
2258 | if (flags & SLAB_STORE_USER) |
2259 | ralign = BYTES_PER_WORD; | |
2260 | ||
2261 | if (flags & SLAB_RED_ZONE) { | |
2262 | ralign = REDZONE_ALIGN; | |
2263 | /* If redzoning, ensure that the second redzone is suitably | |
2264 | * aligned, by adjusting the object size accordingly. */ | |
2265 | size += REDZONE_ALIGN - 1; | |
2266 | size &= ~(REDZONE_ALIGN - 1); | |
2267 | } | |
ca5f9703 | 2268 | |
a44b56d3 | 2269 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2270 | if (ralign < cachep->align) { |
2271 | ralign = cachep->align; | |
1da177e4 | 2272 | } |
3ff84a7f PE |
2273 | /* disable debug if necessary */ |
2274 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2275 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2276 | /* |
ca5f9703 | 2277 | * 4) Store it. |
1da177e4 | 2278 | */ |
8a13a4cc | 2279 | cachep->align = ralign; |
1da177e4 | 2280 | |
83b519e8 PE |
2281 | if (slab_is_available()) |
2282 | gfp = GFP_KERNEL; | |
2283 | else | |
2284 | gfp = GFP_NOWAIT; | |
2285 | ||
6a67368c | 2286 | setup_node_pointer(cachep); |
1da177e4 | 2287 | #if DEBUG |
1da177e4 | 2288 | |
ca5f9703 PE |
2289 | /* |
2290 | * Both debugging options require word-alignment which is calculated | |
2291 | * into align above. | |
2292 | */ | |
1da177e4 | 2293 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2294 | /* add space for red zone words */ |
3ff84a7f PE |
2295 | cachep->obj_offset += sizeof(unsigned long long); |
2296 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2297 | } |
2298 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2299 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2300 | * the real object. But if the second red zone needs to be |
2301 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2302 | */ |
87a927c7 DW |
2303 | if (flags & SLAB_RED_ZONE) |
2304 | size += REDZONE_ALIGN; | |
2305 | else | |
2306 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2307 | } |
2308 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
ce8eb6c4 | 2309 | if (size >= kmalloc_size(INDEX_NODE + 1) |
608da7e3 TH |
2310 | && cachep->object_size > cache_line_size() |
2311 | && ALIGN(size, cachep->align) < PAGE_SIZE) { | |
2312 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); | |
1da177e4 LT |
2313 | size = PAGE_SIZE; |
2314 | } | |
2315 | #endif | |
2316 | #endif | |
2317 | ||
e0a42726 IM |
2318 | /* |
2319 | * Determine if the slab management is 'on' or 'off' slab. | |
2320 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2321 | * it too early on. Always use on-slab management when |
2322 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2323 | */ |
e7cb55b9 CM |
2324 | if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init && |
2325 | !(flags & SLAB_NOLEAKTRACE)) | |
1da177e4 LT |
2326 | /* |
2327 | * Size is large, assume best to place the slab management obj | |
2328 | * off-slab (should allow better packing of objs). | |
2329 | */ | |
2330 | flags |= CFLGS_OFF_SLAB; | |
2331 | ||
8a13a4cc | 2332 | size = ALIGN(size, cachep->align); |
1da177e4 | 2333 | |
8a13a4cc | 2334 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2335 | |
8a13a4cc | 2336 | if (!cachep->num) |
278b1bb1 | 2337 | return -E2BIG; |
1da177e4 | 2338 | |
b28a02de | 2339 | slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) |
8a13a4cc | 2340 | + sizeof(struct slab), cachep->align); |
1da177e4 LT |
2341 | |
2342 | /* | |
2343 | * If the slab has been placed off-slab, and we have enough space then | |
2344 | * move it on-slab. This is at the expense of any extra colouring. | |
2345 | */ | |
2346 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { | |
2347 | flags &= ~CFLGS_OFF_SLAB; | |
2348 | left_over -= slab_size; | |
2349 | } | |
2350 | ||
2351 | if (flags & CFLGS_OFF_SLAB) { | |
2352 | /* really off slab. No need for manual alignment */ | |
b28a02de PE |
2353 | slab_size = |
2354 | cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); | |
67461365 RL |
2355 | |
2356 | #ifdef CONFIG_PAGE_POISONING | |
2357 | /* If we're going to use the generic kernel_map_pages() | |
2358 | * poisoning, then it's going to smash the contents of | |
2359 | * the redzone and userword anyhow, so switch them off. | |
2360 | */ | |
2361 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2362 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2363 | #endif | |
1da177e4 LT |
2364 | } |
2365 | ||
2366 | cachep->colour_off = cache_line_size(); | |
2367 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2368 | if (cachep->colour_off < cachep->align) |
2369 | cachep->colour_off = cachep->align; | |
b28a02de | 2370 | cachep->colour = left_over / cachep->colour_off; |
1da177e4 LT |
2371 | cachep->slab_size = slab_size; |
2372 | cachep->flags = flags; | |
a618e89f | 2373 | cachep->allocflags = 0; |
4b51d669 | 2374 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2375 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2376 | cachep->size = size; |
6a2d7a95 | 2377 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2378 | |
e5ac9c5a | 2379 | if (flags & CFLGS_OFF_SLAB) { |
2c59dd65 | 2380 | cachep->slabp_cache = kmalloc_slab(slab_size, 0u); |
e5ac9c5a RT |
2381 | /* |
2382 | * This is a possibility for one of the malloc_sizes caches. | |
2383 | * But since we go off slab only for object size greater than | |
2384 | * PAGE_SIZE/8, and malloc_sizes gets created in ascending order, | |
2385 | * this should not happen at all. | |
2386 | * But leave a BUG_ON for some lucky dude. | |
2387 | */ | |
6cb8f913 | 2388 | BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache)); |
e5ac9c5a | 2389 | } |
1da177e4 | 2390 | |
278b1bb1 CL |
2391 | err = setup_cpu_cache(cachep, gfp); |
2392 | if (err) { | |
12c3667f | 2393 | __kmem_cache_shutdown(cachep); |
278b1bb1 | 2394 | return err; |
2ed3a4ef | 2395 | } |
1da177e4 | 2396 | |
83835b3d PZ |
2397 | if (flags & SLAB_DEBUG_OBJECTS) { |
2398 | /* | |
2399 | * Would deadlock through slab_destroy()->call_rcu()-> | |
2400 | * debug_object_activate()->kmem_cache_alloc(). | |
2401 | */ | |
2402 | WARN_ON_ONCE(flags & SLAB_DESTROY_BY_RCU); | |
2403 | ||
2404 | slab_set_debugobj_lock_classes(cachep); | |
6ccfb5bc GC |
2405 | } else if (!OFF_SLAB(cachep) && !(flags & SLAB_DESTROY_BY_RCU)) |
2406 | on_slab_lock_classes(cachep); | |
83835b3d | 2407 | |
278b1bb1 | 2408 | return 0; |
1da177e4 | 2409 | } |
1da177e4 LT |
2410 | |
2411 | #if DEBUG | |
2412 | static void check_irq_off(void) | |
2413 | { | |
2414 | BUG_ON(!irqs_disabled()); | |
2415 | } | |
2416 | ||
2417 | static void check_irq_on(void) | |
2418 | { | |
2419 | BUG_ON(irqs_disabled()); | |
2420 | } | |
2421 | ||
343e0d7a | 2422 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2423 | { |
2424 | #ifdef CONFIG_SMP | |
2425 | check_irq_off(); | |
6a67368c | 2426 | assert_spin_locked(&cachep->node[numa_mem_id()]->list_lock); |
1da177e4 LT |
2427 | #endif |
2428 | } | |
e498be7d | 2429 | |
343e0d7a | 2430 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2431 | { |
2432 | #ifdef CONFIG_SMP | |
2433 | check_irq_off(); | |
6a67368c | 2434 | assert_spin_locked(&cachep->node[node]->list_lock); |
e498be7d CL |
2435 | #endif |
2436 | } | |
2437 | ||
1da177e4 LT |
2438 | #else |
2439 | #define check_irq_off() do { } while(0) | |
2440 | #define check_irq_on() do { } while(0) | |
2441 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2442 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2443 | #endif |
2444 | ||
ce8eb6c4 | 2445 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2446 | struct array_cache *ac, |
2447 | int force, int node); | |
2448 | ||
1da177e4 LT |
2449 | static void do_drain(void *arg) |
2450 | { | |
a737b3e2 | 2451 | struct kmem_cache *cachep = arg; |
1da177e4 | 2452 | struct array_cache *ac; |
7d6e6d09 | 2453 | int node = numa_mem_id(); |
1da177e4 LT |
2454 | |
2455 | check_irq_off(); | |
9a2dba4b | 2456 | ac = cpu_cache_get(cachep); |
6a67368c | 2457 | spin_lock(&cachep->node[node]->list_lock); |
ff69416e | 2458 | free_block(cachep, ac->entry, ac->avail, node); |
6a67368c | 2459 | spin_unlock(&cachep->node[node]->list_lock); |
1da177e4 LT |
2460 | ac->avail = 0; |
2461 | } | |
2462 | ||
343e0d7a | 2463 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2464 | { |
ce8eb6c4 | 2465 | struct kmem_cache_node *n; |
e498be7d CL |
2466 | int node; |
2467 | ||
15c8b6c1 | 2468 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2469 | check_irq_on(); |
b28a02de | 2470 | for_each_online_node(node) { |
ce8eb6c4 CL |
2471 | n = cachep->node[node]; |
2472 | if (n && n->alien) | |
2473 | drain_alien_cache(cachep, n->alien); | |
a4523a8b RD |
2474 | } |
2475 | ||
2476 | for_each_online_node(node) { | |
ce8eb6c4 CL |
2477 | n = cachep->node[node]; |
2478 | if (n) | |
2479 | drain_array(cachep, n, n->shared, 1, node); | |
e498be7d | 2480 | } |
1da177e4 LT |
2481 | } |
2482 | ||
ed11d9eb CL |
2483 | /* |
2484 | * Remove slabs from the list of free slabs. | |
2485 | * Specify the number of slabs to drain in tofree. | |
2486 | * | |
2487 | * Returns the actual number of slabs released. | |
2488 | */ | |
2489 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2490 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2491 | { |
ed11d9eb CL |
2492 | struct list_head *p; |
2493 | int nr_freed; | |
1da177e4 | 2494 | struct slab *slabp; |
1da177e4 | 2495 | |
ed11d9eb | 2496 | nr_freed = 0; |
ce8eb6c4 | 2497 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2498 | |
ce8eb6c4 CL |
2499 | spin_lock_irq(&n->list_lock); |
2500 | p = n->slabs_free.prev; | |
2501 | if (p == &n->slabs_free) { | |
2502 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2503 | goto out; |
2504 | } | |
1da177e4 | 2505 | |
ed11d9eb | 2506 | slabp = list_entry(p, struct slab, list); |
1da177e4 | 2507 | #if DEBUG |
40094fa6 | 2508 | BUG_ON(slabp->inuse); |
1da177e4 LT |
2509 | #endif |
2510 | list_del(&slabp->list); | |
ed11d9eb CL |
2511 | /* |
2512 | * Safe to drop the lock. The slab is no longer linked | |
2513 | * to the cache. | |
2514 | */ | |
ce8eb6c4 CL |
2515 | n->free_objects -= cache->num; |
2516 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2517 | slab_destroy(cache, slabp); |
2518 | nr_freed++; | |
1da177e4 | 2519 | } |
ed11d9eb CL |
2520 | out: |
2521 | return nr_freed; | |
1da177e4 LT |
2522 | } |
2523 | ||
18004c5d | 2524 | /* Called with slab_mutex held to protect against cpu hotplug */ |
343e0d7a | 2525 | static int __cache_shrink(struct kmem_cache *cachep) |
e498be7d CL |
2526 | { |
2527 | int ret = 0, i = 0; | |
ce8eb6c4 | 2528 | struct kmem_cache_node *n; |
e498be7d CL |
2529 | |
2530 | drain_cpu_caches(cachep); | |
2531 | ||
2532 | check_irq_on(); | |
2533 | for_each_online_node(i) { | |
ce8eb6c4 CL |
2534 | n = cachep->node[i]; |
2535 | if (!n) | |
ed11d9eb CL |
2536 | continue; |
2537 | ||
0fa8103b | 2538 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2539 | |
ce8eb6c4 CL |
2540 | ret += !list_empty(&n->slabs_full) || |
2541 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2542 | } |
2543 | return (ret ? 1 : 0); | |
2544 | } | |
2545 | ||
1da177e4 LT |
2546 | /** |
2547 | * kmem_cache_shrink - Shrink a cache. | |
2548 | * @cachep: The cache to shrink. | |
2549 | * | |
2550 | * Releases as many slabs as possible for a cache. | |
2551 | * To help debugging, a zero exit status indicates all slabs were released. | |
2552 | */ | |
343e0d7a | 2553 | int kmem_cache_shrink(struct kmem_cache *cachep) |
1da177e4 | 2554 | { |
8f5be20b | 2555 | int ret; |
40094fa6 | 2556 | BUG_ON(!cachep || in_interrupt()); |
1da177e4 | 2557 | |
95402b38 | 2558 | get_online_cpus(); |
18004c5d | 2559 | mutex_lock(&slab_mutex); |
8f5be20b | 2560 | ret = __cache_shrink(cachep); |
18004c5d | 2561 | mutex_unlock(&slab_mutex); |
95402b38 | 2562 | put_online_cpus(); |
8f5be20b | 2563 | return ret; |
1da177e4 LT |
2564 | } |
2565 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2566 | ||
945cf2b6 | 2567 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
1da177e4 | 2568 | { |
12c3667f | 2569 | int i; |
ce8eb6c4 | 2570 | struct kmem_cache_node *n; |
12c3667f | 2571 | int rc = __cache_shrink(cachep); |
1da177e4 | 2572 | |
12c3667f CL |
2573 | if (rc) |
2574 | return rc; | |
1da177e4 | 2575 | |
12c3667f CL |
2576 | for_each_online_cpu(i) |
2577 | kfree(cachep->array[i]); | |
1da177e4 | 2578 | |
ce8eb6c4 | 2579 | /* NUMA: free the node structures */ |
12c3667f | 2580 | for_each_online_node(i) { |
ce8eb6c4 CL |
2581 | n = cachep->node[i]; |
2582 | if (n) { | |
2583 | kfree(n->shared); | |
2584 | free_alien_cache(n->alien); | |
2585 | kfree(n); | |
12c3667f CL |
2586 | } |
2587 | } | |
2588 | return 0; | |
1da177e4 | 2589 | } |
1da177e4 | 2590 | |
e5ac9c5a RT |
2591 | /* |
2592 | * Get the memory for a slab management obj. | |
2593 | * For a slab cache when the slab descriptor is off-slab, slab descriptors | |
2594 | * always come from malloc_sizes caches. The slab descriptor cannot | |
2595 | * come from the same cache which is getting created because, | |
2596 | * when we are searching for an appropriate cache for these | |
2597 | * descriptors in kmem_cache_create, we search through the malloc_sizes array. | |
2598 | * If we are creating a malloc_sizes cache here it would not be visible to | |
2599 | * kmem_find_general_cachep till the initialization is complete. | |
2600 | * Hence we cannot have slabp_cache same as the original cache. | |
2601 | */ | |
0c3aa83e JK |
2602 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, |
2603 | struct page *page, int colour_off, | |
2604 | gfp_t local_flags, int nodeid) | |
1da177e4 LT |
2605 | { |
2606 | struct slab *slabp; | |
0c3aa83e | 2607 | void *addr = page_address(page); |
b28a02de | 2608 | |
1da177e4 LT |
2609 | if (OFF_SLAB(cachep)) { |
2610 | /* Slab management obj is off-slab. */ | |
5b74ada7 | 2611 | slabp = kmem_cache_alloc_node(cachep->slabp_cache, |
8759ec50 | 2612 | local_flags, nodeid); |
d5cff635 CM |
2613 | /* |
2614 | * If the first object in the slab is leaked (it's allocated | |
2615 | * but no one has a reference to it), we want to make sure | |
2616 | * kmemleak does not treat the ->s_mem pointer as a reference | |
2617 | * to the object. Otherwise we will not report the leak. | |
2618 | */ | |
c017b4be CM |
2619 | kmemleak_scan_area(&slabp->list, sizeof(struct list_head), |
2620 | local_flags); | |
1da177e4 LT |
2621 | if (!slabp) |
2622 | return NULL; | |
2623 | } else { | |
0c3aa83e | 2624 | slabp = addr + colour_off; |
1da177e4 LT |
2625 | colour_off += cachep->slab_size; |
2626 | } | |
2627 | slabp->inuse = 0; | |
0c3aa83e | 2628 | slabp->s_mem = addr + colour_off; |
e51bfd0a | 2629 | slabp->free = 0; |
1da177e4 LT |
2630 | return slabp; |
2631 | } | |
2632 | ||
2633 | static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) | |
2634 | { | |
b28a02de | 2635 | return (kmem_bufctl_t *) (slabp + 1); |
1da177e4 LT |
2636 | } |
2637 | ||
343e0d7a | 2638 | static void cache_init_objs(struct kmem_cache *cachep, |
a35afb83 | 2639 | struct slab *slabp) |
1da177e4 LT |
2640 | { |
2641 | int i; | |
2642 | ||
2643 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 2644 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
2645 | #if DEBUG |
2646 | /* need to poison the objs? */ | |
2647 | if (cachep->flags & SLAB_POISON) | |
2648 | poison_obj(cachep, objp, POISON_FREE); | |
2649 | if (cachep->flags & SLAB_STORE_USER) | |
2650 | *dbg_userword(cachep, objp) = NULL; | |
2651 | ||
2652 | if (cachep->flags & SLAB_RED_ZONE) { | |
2653 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2654 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2655 | } | |
2656 | /* | |
a737b3e2 AM |
2657 | * Constructors are not allowed to allocate memory from the same |
2658 | * cache which they are a constructor for. Otherwise, deadlock. | |
2659 | * They must also be threaded. | |
1da177e4 LT |
2660 | */ |
2661 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2662 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2663 | |
2664 | if (cachep->flags & SLAB_RED_ZONE) { | |
2665 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2666 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2667 | " end of an object"); |
1da177e4 LT |
2668 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2669 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2670 | " start of an object"); |
1da177e4 | 2671 | } |
3b0efdfa | 2672 | if ((cachep->size % PAGE_SIZE) == 0 && |
a737b3e2 | 2673 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
b28a02de | 2674 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2675 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2676 | #else |
2677 | if (cachep->ctor) | |
51cc5068 | 2678 | cachep->ctor(objp); |
1da177e4 | 2679 | #endif |
b28a02de | 2680 | slab_bufctl(slabp)[i] = i + 1; |
1da177e4 | 2681 | } |
b28a02de | 2682 | slab_bufctl(slabp)[i - 1] = BUFCTL_END; |
1da177e4 LT |
2683 | } |
2684 | ||
343e0d7a | 2685 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2686 | { |
4b51d669 CL |
2687 | if (CONFIG_ZONE_DMA_FLAG) { |
2688 | if (flags & GFP_DMA) | |
a618e89f | 2689 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2690 | else |
a618e89f | 2691 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2692 | } |
1da177e4 LT |
2693 | } |
2694 | ||
a737b3e2 AM |
2695 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, |
2696 | int nodeid) | |
78d382d7 | 2697 | { |
8fea4e96 | 2698 | void *objp = index_to_obj(cachep, slabp, slabp->free); |
78d382d7 MD |
2699 | kmem_bufctl_t next; |
2700 | ||
2701 | slabp->inuse++; | |
2702 | next = slab_bufctl(slabp)[slabp->free]; | |
2703 | #if DEBUG | |
2704 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; | |
1ea991b0 | 2705 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 MD |
2706 | #endif |
2707 | slabp->free = next; | |
2708 | ||
2709 | return objp; | |
2710 | } | |
2711 | ||
a737b3e2 AM |
2712 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, |
2713 | void *objp, int nodeid) | |
78d382d7 | 2714 | { |
8fea4e96 | 2715 | unsigned int objnr = obj_to_index(cachep, slabp, objp); |
78d382d7 MD |
2716 | |
2717 | #if DEBUG | |
2718 | /* Verify that the slab belongs to the intended node */ | |
1ea991b0 | 2719 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2720 | |
871751e2 | 2721 | if (slab_bufctl(slabp)[objnr] + 1 <= SLAB_LIMIT + 1) { |
78d382d7 | 2722 | printk(KERN_ERR "slab: double free detected in cache " |
a737b3e2 | 2723 | "'%s', objp %p\n", cachep->name, objp); |
78d382d7 MD |
2724 | BUG(); |
2725 | } | |
2726 | #endif | |
2727 | slab_bufctl(slabp)[objnr] = slabp->free; | |
2728 | slabp->free = objnr; | |
2729 | slabp->inuse--; | |
2730 | } | |
2731 | ||
4776874f PE |
2732 | /* |
2733 | * Map pages beginning at addr to the given cache and slab. This is required | |
2734 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2735 | * virtual address for kfree, ksize, and slab debugging. |
4776874f PE |
2736 | */ |
2737 | static void slab_map_pages(struct kmem_cache *cache, struct slab *slab, | |
0c3aa83e | 2738 | struct page *page) |
1da177e4 | 2739 | { |
4776874f | 2740 | int nr_pages; |
84097518 | 2741 | |
4776874f | 2742 | nr_pages = 1; |
84097518 | 2743 | if (likely(!PageCompound(page))) |
4776874f PE |
2744 | nr_pages <<= cache->gfporder; |
2745 | ||
1da177e4 | 2746 | do { |
35026088 CL |
2747 | page->slab_cache = cache; |
2748 | page->slab_page = slab; | |
1da177e4 | 2749 | page++; |
4776874f | 2750 | } while (--nr_pages); |
1da177e4 LT |
2751 | } |
2752 | ||
2753 | /* | |
2754 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2755 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2756 | */ | |
3c517a61 | 2757 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2758 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2759 | { |
b28a02de | 2760 | struct slab *slabp; |
b28a02de PE |
2761 | size_t offset; |
2762 | gfp_t local_flags; | |
ce8eb6c4 | 2763 | struct kmem_cache_node *n; |
1da177e4 | 2764 | |
a737b3e2 AM |
2765 | /* |
2766 | * Be lazy and only check for valid flags here, keeping it out of the | |
2767 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2768 | */ |
6cb06229 CL |
2769 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2770 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2771 | |
ce8eb6c4 | 2772 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2773 | check_irq_off(); |
ce8eb6c4 CL |
2774 | n = cachep->node[nodeid]; |
2775 | spin_lock(&n->list_lock); | |
1da177e4 LT |
2776 | |
2777 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2778 | offset = n->colour_next; |
2779 | n->colour_next++; | |
2780 | if (n->colour_next >= cachep->colour) | |
2781 | n->colour_next = 0; | |
2782 | spin_unlock(&n->list_lock); | |
1da177e4 | 2783 | |
2e1217cf | 2784 | offset *= cachep->colour_off; |
1da177e4 LT |
2785 | |
2786 | if (local_flags & __GFP_WAIT) | |
2787 | local_irq_enable(); | |
2788 | ||
2789 | /* | |
2790 | * The test for missing atomic flag is performed here, rather than | |
2791 | * the more obvious place, simply to reduce the critical path length | |
2792 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2793 | * will eventually be caught here (where it matters). | |
2794 | */ | |
2795 | kmem_flagcheck(cachep, flags); | |
2796 | ||
a737b3e2 AM |
2797 | /* |
2798 | * Get mem for the objs. Attempt to allocate a physical page from | |
2799 | * 'nodeid'. | |
e498be7d | 2800 | */ |
0c3aa83e JK |
2801 | if (!page) |
2802 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2803 | if (!page) | |
1da177e4 LT |
2804 | goto failed; |
2805 | ||
2806 | /* Get slab management. */ | |
0c3aa83e | 2807 | slabp = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2808 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
a737b3e2 | 2809 | if (!slabp) |
1da177e4 LT |
2810 | goto opps1; |
2811 | ||
0c3aa83e | 2812 | slab_map_pages(cachep, slabp, page); |
1da177e4 | 2813 | |
a35afb83 | 2814 | cache_init_objs(cachep, slabp); |
1da177e4 LT |
2815 | |
2816 | if (local_flags & __GFP_WAIT) | |
2817 | local_irq_disable(); | |
2818 | check_irq_off(); | |
ce8eb6c4 | 2819 | spin_lock(&n->list_lock); |
1da177e4 LT |
2820 | |
2821 | /* Make slab active. */ | |
ce8eb6c4 | 2822 | list_add_tail(&slabp->list, &(n->slabs_free)); |
1da177e4 | 2823 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2824 | n->free_objects += cachep->num; |
2825 | spin_unlock(&n->list_lock); | |
1da177e4 | 2826 | return 1; |
a737b3e2 | 2827 | opps1: |
0c3aa83e | 2828 | kmem_freepages(cachep, page); |
a737b3e2 | 2829 | failed: |
1da177e4 LT |
2830 | if (local_flags & __GFP_WAIT) |
2831 | local_irq_disable(); | |
2832 | return 0; | |
2833 | } | |
2834 | ||
2835 | #if DEBUG | |
2836 | ||
2837 | /* | |
2838 | * Perform extra freeing checks: | |
2839 | * - detect bad pointers. | |
2840 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2841 | */ |
2842 | static void kfree_debugcheck(const void *objp) | |
2843 | { | |
1da177e4 LT |
2844 | if (!virt_addr_valid(objp)) { |
2845 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2846 | (unsigned long)objp); |
2847 | BUG(); | |
1da177e4 | 2848 | } |
1da177e4 LT |
2849 | } |
2850 | ||
58ce1fd5 PE |
2851 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2852 | { | |
b46b8f19 | 2853 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2854 | |
2855 | redzone1 = *dbg_redzone1(cache, obj); | |
2856 | redzone2 = *dbg_redzone2(cache, obj); | |
2857 | ||
2858 | /* | |
2859 | * Redzone is ok. | |
2860 | */ | |
2861 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2862 | return; | |
2863 | ||
2864 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2865 | slab_error(cache, "double free detected"); | |
2866 | else | |
2867 | slab_error(cache, "memory outside object was overwritten"); | |
2868 | ||
b46b8f19 | 2869 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2870 | obj, redzone1, redzone2); |
2871 | } | |
2872 | ||
343e0d7a | 2873 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2874 | unsigned long caller) |
1da177e4 LT |
2875 | { |
2876 | struct page *page; | |
2877 | unsigned int objnr; | |
2878 | struct slab *slabp; | |
2879 | ||
80cbd911 MW |
2880 | BUG_ON(virt_to_cache(objp) != cachep); |
2881 | ||
3dafccf2 | 2882 | objp -= obj_offset(cachep); |
1da177e4 | 2883 | kfree_debugcheck(objp); |
b49af68f | 2884 | page = virt_to_head_page(objp); |
1da177e4 | 2885 | |
35026088 | 2886 | slabp = page->slab_page; |
1da177e4 LT |
2887 | |
2888 | if (cachep->flags & SLAB_RED_ZONE) { | |
58ce1fd5 | 2889 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2890 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2891 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2892 | } | |
2893 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2894 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2895 | |
8fea4e96 | 2896 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 LT |
2897 | |
2898 | BUG_ON(objnr >= cachep->num); | |
8fea4e96 | 2899 | BUG_ON(objp != index_to_obj(cachep, slabp, objnr)); |
1da177e4 | 2900 | |
871751e2 AV |
2901 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
2902 | slab_bufctl(slabp)[objnr] = BUFCTL_FREE; | |
2903 | #endif | |
1da177e4 LT |
2904 | if (cachep->flags & SLAB_POISON) { |
2905 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2906 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
7c0cb9c6 | 2907 | store_stackinfo(cachep, objp, caller); |
b28a02de | 2908 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2909 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2910 | } else { |
2911 | poison_obj(cachep, objp, POISON_FREE); | |
2912 | } | |
2913 | #else | |
2914 | poison_obj(cachep, objp, POISON_FREE); | |
2915 | #endif | |
2916 | } | |
2917 | return objp; | |
2918 | } | |
2919 | ||
343e0d7a | 2920 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 LT |
2921 | { |
2922 | kmem_bufctl_t i; | |
2923 | int entries = 0; | |
b28a02de | 2924 | |
1da177e4 LT |
2925 | /* Check slab's freelist to see if this obj is there. */ |
2926 | for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { | |
2927 | entries++; | |
2928 | if (entries > cachep->num || i >= cachep->num) | |
2929 | goto bad; | |
2930 | } | |
2931 | if (entries != cachep->num - slabp->inuse) { | |
a737b3e2 AM |
2932 | bad: |
2933 | printk(KERN_ERR "slab: Internal list corruption detected in " | |
face37f5 DJ |
2934 | "cache '%s'(%d), slabp %p(%d). Tainted(%s). Hexdump:\n", |
2935 | cachep->name, cachep->num, slabp, slabp->inuse, | |
2936 | print_tainted()); | |
fdde6abb SAS |
2937 | print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, slabp, |
2938 | sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t), | |
2939 | 1); | |
1da177e4 LT |
2940 | BUG(); |
2941 | } | |
2942 | } | |
2943 | #else | |
2944 | #define kfree_debugcheck(x) do { } while(0) | |
2945 | #define cache_free_debugcheck(x,objp,z) (objp) | |
2946 | #define check_slabp(x,y) do { } while(0) | |
2947 | #endif | |
2948 | ||
072bb0aa MG |
2949 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2950 | bool force_refill) | |
1da177e4 LT |
2951 | { |
2952 | int batchcount; | |
ce8eb6c4 | 2953 | struct kmem_cache_node *n; |
1da177e4 | 2954 | struct array_cache *ac; |
1ca4cb24 PE |
2955 | int node; |
2956 | ||
1da177e4 | 2957 | check_irq_off(); |
7d6e6d09 | 2958 | node = numa_mem_id(); |
072bb0aa MG |
2959 | if (unlikely(force_refill)) |
2960 | goto force_grow; | |
2961 | retry: | |
9a2dba4b | 2962 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2963 | batchcount = ac->batchcount; |
2964 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2965 | /* |
2966 | * If there was little recent activity on this cache, then | |
2967 | * perform only a partial refill. Otherwise we could generate | |
2968 | * refill bouncing. | |
1da177e4 LT |
2969 | */ |
2970 | batchcount = BATCHREFILL_LIMIT; | |
2971 | } | |
ce8eb6c4 | 2972 | n = cachep->node[node]; |
e498be7d | 2973 | |
ce8eb6c4 CL |
2974 | BUG_ON(ac->avail > 0 || !n); |
2975 | spin_lock(&n->list_lock); | |
1da177e4 | 2976 | |
3ded175a | 2977 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2978 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2979 | n->shared->touched = 1; | |
3ded175a | 2980 | goto alloc_done; |
44b57f1c | 2981 | } |
3ded175a | 2982 | |
1da177e4 LT |
2983 | while (batchcount > 0) { |
2984 | struct list_head *entry; | |
2985 | struct slab *slabp; | |
2986 | /* Get slab alloc is to come from. */ | |
ce8eb6c4 CL |
2987 | entry = n->slabs_partial.next; |
2988 | if (entry == &n->slabs_partial) { | |
2989 | n->free_touched = 1; | |
2990 | entry = n->slabs_free.next; | |
2991 | if (entry == &n->slabs_free) | |
1da177e4 LT |
2992 | goto must_grow; |
2993 | } | |
2994 | ||
2995 | slabp = list_entry(entry, struct slab, list); | |
2996 | check_slabp(cachep, slabp); | |
2997 | check_spinlock_acquired(cachep); | |
714b8171 PE |
2998 | |
2999 | /* | |
3000 | * The slab was either on partial or free list so | |
3001 | * there must be at least one object available for | |
3002 | * allocation. | |
3003 | */ | |
249b9f33 | 3004 | BUG_ON(slabp->inuse >= cachep->num); |
714b8171 | 3005 | |
1da177e4 | 3006 | while (slabp->inuse < cachep->num && batchcount--) { |
1da177e4 LT |
3007 | STATS_INC_ALLOCED(cachep); |
3008 | STATS_INC_ACTIVE(cachep); | |
3009 | STATS_SET_HIGH(cachep); | |
3010 | ||
072bb0aa MG |
3011 | ac_put_obj(cachep, ac, slab_get_obj(cachep, slabp, |
3012 | node)); | |
1da177e4 LT |
3013 | } |
3014 | check_slabp(cachep, slabp); | |
3015 | ||
3016 | /* move slabp to correct slabp list: */ | |
3017 | list_del(&slabp->list); | |
3018 | if (slabp->free == BUFCTL_END) | |
ce8eb6c4 | 3019 | list_add(&slabp->list, &n->slabs_full); |
1da177e4 | 3020 | else |
ce8eb6c4 | 3021 | list_add(&slabp->list, &n->slabs_partial); |
1da177e4 LT |
3022 | } |
3023 | ||
a737b3e2 | 3024 | must_grow: |
ce8eb6c4 | 3025 | n->free_objects -= ac->avail; |
a737b3e2 | 3026 | alloc_done: |
ce8eb6c4 | 3027 | spin_unlock(&n->list_lock); |
1da177e4 LT |
3028 | |
3029 | if (unlikely(!ac->avail)) { | |
3030 | int x; | |
072bb0aa | 3031 | force_grow: |
3c517a61 | 3032 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 3033 | |
a737b3e2 | 3034 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 3035 | ac = cpu_cache_get(cachep); |
51cd8e6f | 3036 | node = numa_mem_id(); |
072bb0aa MG |
3037 | |
3038 | /* no objects in sight? abort */ | |
3039 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
3040 | return NULL; |
3041 | ||
a737b3e2 | 3042 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
3043 | goto retry; |
3044 | } | |
3045 | ac->touched = 1; | |
072bb0aa MG |
3046 | |
3047 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
3048 | } |
3049 | ||
a737b3e2 AM |
3050 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3051 | gfp_t flags) | |
1da177e4 LT |
3052 | { |
3053 | might_sleep_if(flags & __GFP_WAIT); | |
3054 | #if DEBUG | |
3055 | kmem_flagcheck(cachep, flags); | |
3056 | #endif | |
3057 | } | |
3058 | ||
3059 | #if DEBUG | |
a737b3e2 | 3060 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 3061 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 3062 | { |
b28a02de | 3063 | if (!objp) |
1da177e4 | 3064 | return objp; |
b28a02de | 3065 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3066 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3b0efdfa | 3067 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 3068 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 3069 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
3070 | else |
3071 | check_poison_obj(cachep, objp); | |
3072 | #else | |
3073 | check_poison_obj(cachep, objp); | |
3074 | #endif | |
3075 | poison_obj(cachep, objp, POISON_INUSE); | |
3076 | } | |
3077 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 3078 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
3079 | |
3080 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3081 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3082 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
3083 | slab_error(cachep, "double free, or memory outside" | |
3084 | " object was overwritten"); | |
b28a02de | 3085 | printk(KERN_ERR |
b46b8f19 | 3086 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
3087 | objp, *dbg_redzone1(cachep, objp), |
3088 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3089 | } |
3090 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3091 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3092 | } | |
871751e2 AV |
3093 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
3094 | { | |
3095 | struct slab *slabp; | |
3096 | unsigned objnr; | |
3097 | ||
35026088 | 3098 | slabp = virt_to_head_page(objp)->slab_page; |
3b0efdfa | 3099 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->size; |
871751e2 AV |
3100 | slab_bufctl(slabp)[objnr] = BUFCTL_ACTIVE; |
3101 | } | |
3102 | #endif | |
3dafccf2 | 3103 | objp += obj_offset(cachep); |
4f104934 | 3104 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3105 | cachep->ctor(objp); |
7ea466f2 TH |
3106 | if (ARCH_SLAB_MINALIGN && |
3107 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 3108 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3109 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3110 | } |
1da177e4 LT |
3111 | return objp; |
3112 | } | |
3113 | #else | |
3114 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3115 | #endif | |
3116 | ||
773ff60e | 3117 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 | 3118 | { |
9b030cb8 | 3119 | if (cachep == kmem_cache) |
773ff60e | 3120 | return false; |
8a8b6502 | 3121 | |
8c138bc0 | 3122 | return should_failslab(cachep->object_size, flags, cachep->flags); |
8a8b6502 AM |
3123 | } |
3124 | ||
343e0d7a | 3125 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3126 | { |
b28a02de | 3127 | void *objp; |
1da177e4 | 3128 | struct array_cache *ac; |
072bb0aa | 3129 | bool force_refill = false; |
1da177e4 | 3130 | |
5c382300 | 3131 | check_irq_off(); |
8a8b6502 | 3132 | |
9a2dba4b | 3133 | ac = cpu_cache_get(cachep); |
1da177e4 | 3134 | if (likely(ac->avail)) { |
1da177e4 | 3135 | ac->touched = 1; |
072bb0aa MG |
3136 | objp = ac_get_obj(cachep, ac, flags, false); |
3137 | ||
ddbf2e83 | 3138 | /* |
072bb0aa MG |
3139 | * Allow for the possibility all avail objects are not allowed |
3140 | * by the current flags | |
ddbf2e83 | 3141 | */ |
072bb0aa MG |
3142 | if (objp) { |
3143 | STATS_INC_ALLOCHIT(cachep); | |
3144 | goto out; | |
3145 | } | |
3146 | force_refill = true; | |
1da177e4 | 3147 | } |
072bb0aa MG |
3148 | |
3149 | STATS_INC_ALLOCMISS(cachep); | |
3150 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
3151 | /* | |
3152 | * the 'ac' may be updated by cache_alloc_refill(), | |
3153 | * and kmemleak_erase() requires its correct value. | |
3154 | */ | |
3155 | ac = cpu_cache_get(cachep); | |
3156 | ||
3157 | out: | |
d5cff635 CM |
3158 | /* |
3159 | * To avoid a false negative, if an object that is in one of the | |
3160 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3161 | * treat the array pointers as a reference to the object. | |
3162 | */ | |
f3d8b53a O |
3163 | if (objp) |
3164 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3165 | return objp; |
3166 | } | |
3167 | ||
e498be7d | 3168 | #ifdef CONFIG_NUMA |
c61afb18 | 3169 | /* |
b2455396 | 3170 | * Try allocating on another node if PF_SPREAD_SLAB|PF_MEMPOLICY. |
c61afb18 PJ |
3171 | * |
3172 | * If we are in_interrupt, then process context, including cpusets and | |
3173 | * mempolicy, may not apply and should not be used for allocation policy. | |
3174 | */ | |
3175 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3176 | { | |
3177 | int nid_alloc, nid_here; | |
3178 | ||
765c4507 | 3179 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3180 | return NULL; |
7d6e6d09 | 3181 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3182 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3183 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3184 | else if (current->mempolicy) |
e7b691b0 | 3185 | nid_alloc = slab_node(); |
c61afb18 | 3186 | if (nid_alloc != nid_here) |
8b98c169 | 3187 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3188 | return NULL; |
3189 | } | |
3190 | ||
765c4507 CL |
3191 | /* |
3192 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3193 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3194 | * available node for available objects. If that fails then we |
3c517a61 CL |
3195 | * perform an allocation without specifying a node. This allows the page |
3196 | * allocator to do its reclaim / fallback magic. We then insert the | |
3197 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3198 | */ |
8c8cc2c1 | 3199 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3200 | { |
8c8cc2c1 PE |
3201 | struct zonelist *zonelist; |
3202 | gfp_t local_flags; | |
dd1a239f | 3203 | struct zoneref *z; |
54a6eb5c MG |
3204 | struct zone *zone; |
3205 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3206 | void *obj = NULL; |
3c517a61 | 3207 | int nid; |
cc9a6c87 | 3208 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3209 | |
3210 | if (flags & __GFP_THISNODE) | |
3211 | return NULL; | |
3212 | ||
6cb06229 | 3213 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3214 | |
cc9a6c87 MG |
3215 | retry_cpuset: |
3216 | cpuset_mems_cookie = get_mems_allowed(); | |
e7b691b0 | 3217 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 | 3218 | |
3c517a61 CL |
3219 | retry: |
3220 | /* | |
3221 | * Look through allowed nodes for objects available | |
3222 | * from existing per node queues. | |
3223 | */ | |
54a6eb5c MG |
3224 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3225 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3226 | |
54a6eb5c | 3227 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
6a67368c CL |
3228 | cache->node[nid] && |
3229 | cache->node[nid]->free_objects) { | |
3c517a61 CL |
3230 | obj = ____cache_alloc_node(cache, |
3231 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3232 | if (obj) |
3233 | break; | |
3234 | } | |
3c517a61 CL |
3235 | } |
3236 | ||
cfce6604 | 3237 | if (!obj) { |
3c517a61 CL |
3238 | /* |
3239 | * This allocation will be performed within the constraints | |
3240 | * of the current cpuset / memory policy requirements. | |
3241 | * We may trigger various forms of reclaim on the allowed | |
3242 | * set and go into memory reserves if necessary. | |
3243 | */ | |
0c3aa83e JK |
3244 | struct page *page; |
3245 | ||
dd47ea75 CL |
3246 | if (local_flags & __GFP_WAIT) |
3247 | local_irq_enable(); | |
3248 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3249 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
dd47ea75 CL |
3250 | if (local_flags & __GFP_WAIT) |
3251 | local_irq_disable(); | |
0c3aa83e | 3252 | if (page) { |
3c517a61 CL |
3253 | /* |
3254 | * Insert into the appropriate per node queues | |
3255 | */ | |
0c3aa83e JK |
3256 | nid = page_to_nid(page); |
3257 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 CL |
3258 | obj = ____cache_alloc_node(cache, |
3259 | flags | GFP_THISNODE, nid); | |
3260 | if (!obj) | |
3261 | /* | |
3262 | * Another processor may allocate the | |
3263 | * objects in the slab since we are | |
3264 | * not holding any locks. | |
3265 | */ | |
3266 | goto retry; | |
3267 | } else { | |
b6a60451 | 3268 | /* cache_grow already freed obj */ |
3c517a61 CL |
3269 | obj = NULL; |
3270 | } | |
3271 | } | |
aedb0eb1 | 3272 | } |
cc9a6c87 MG |
3273 | |
3274 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !obj)) | |
3275 | goto retry_cpuset; | |
765c4507 CL |
3276 | return obj; |
3277 | } | |
3278 | ||
e498be7d CL |
3279 | /* |
3280 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3281 | */ |
8b98c169 | 3282 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3283 | int nodeid) |
e498be7d CL |
3284 | { |
3285 | struct list_head *entry; | |
b28a02de | 3286 | struct slab *slabp; |
ce8eb6c4 | 3287 | struct kmem_cache_node *n; |
b28a02de | 3288 | void *obj; |
b28a02de PE |
3289 | int x; |
3290 | ||
14e50c6a | 3291 | VM_BUG_ON(nodeid > num_online_nodes()); |
ce8eb6c4 CL |
3292 | n = cachep->node[nodeid]; |
3293 | BUG_ON(!n); | |
b28a02de | 3294 | |
a737b3e2 | 3295 | retry: |
ca3b9b91 | 3296 | check_irq_off(); |
ce8eb6c4 CL |
3297 | spin_lock(&n->list_lock); |
3298 | entry = n->slabs_partial.next; | |
3299 | if (entry == &n->slabs_partial) { | |
3300 | n->free_touched = 1; | |
3301 | entry = n->slabs_free.next; | |
3302 | if (entry == &n->slabs_free) | |
b28a02de PE |
3303 | goto must_grow; |
3304 | } | |
3305 | ||
3306 | slabp = list_entry(entry, struct slab, list); | |
3307 | check_spinlock_acquired_node(cachep, nodeid); | |
3308 | check_slabp(cachep, slabp); | |
3309 | ||
3310 | STATS_INC_NODEALLOCS(cachep); | |
3311 | STATS_INC_ACTIVE(cachep); | |
3312 | STATS_SET_HIGH(cachep); | |
3313 | ||
3314 | BUG_ON(slabp->inuse == cachep->num); | |
3315 | ||
78d382d7 | 3316 | obj = slab_get_obj(cachep, slabp, nodeid); |
b28a02de | 3317 | check_slabp(cachep, slabp); |
ce8eb6c4 | 3318 | n->free_objects--; |
b28a02de PE |
3319 | /* move slabp to correct slabp list: */ |
3320 | list_del(&slabp->list); | |
3321 | ||
a737b3e2 | 3322 | if (slabp->free == BUFCTL_END) |
ce8eb6c4 | 3323 | list_add(&slabp->list, &n->slabs_full); |
a737b3e2 | 3324 | else |
ce8eb6c4 | 3325 | list_add(&slabp->list, &n->slabs_partial); |
e498be7d | 3326 | |
ce8eb6c4 | 3327 | spin_unlock(&n->list_lock); |
b28a02de | 3328 | goto done; |
e498be7d | 3329 | |
a737b3e2 | 3330 | must_grow: |
ce8eb6c4 | 3331 | spin_unlock(&n->list_lock); |
3c517a61 | 3332 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3333 | if (x) |
3334 | goto retry; | |
1da177e4 | 3335 | |
8c8cc2c1 | 3336 | return fallback_alloc(cachep, flags); |
e498be7d | 3337 | |
a737b3e2 | 3338 | done: |
b28a02de | 3339 | return obj; |
e498be7d | 3340 | } |
8c8cc2c1 | 3341 | |
8c8cc2c1 | 3342 | static __always_inline void * |
48356303 | 3343 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3344 | unsigned long caller) |
8c8cc2c1 PE |
3345 | { |
3346 | unsigned long save_flags; | |
3347 | void *ptr; | |
7d6e6d09 | 3348 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3349 | |
dcce284a | 3350 | flags &= gfp_allowed_mask; |
7e85ee0c | 3351 | |
cf40bd16 NP |
3352 | lockdep_trace_alloc(flags); |
3353 | ||
773ff60e | 3354 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3355 | return NULL; |
3356 | ||
d79923fa GC |
3357 | cachep = memcg_kmem_get_cache(cachep, flags); |
3358 | ||
8c8cc2c1 PE |
3359 | cache_alloc_debugcheck_before(cachep, flags); |
3360 | local_irq_save(save_flags); | |
3361 | ||
eacbbae3 | 3362 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3363 | nodeid = slab_node; |
8c8cc2c1 | 3364 | |
6a67368c | 3365 | if (unlikely(!cachep->node[nodeid])) { |
8c8cc2c1 PE |
3366 | /* Node not bootstrapped yet */ |
3367 | ptr = fallback_alloc(cachep, flags); | |
3368 | goto out; | |
3369 | } | |
3370 | ||
7d6e6d09 | 3371 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3372 | /* |
3373 | * Use the locally cached objects if possible. | |
3374 | * However ____cache_alloc does not allow fallback | |
3375 | * to other nodes. It may fail while we still have | |
3376 | * objects on other nodes available. | |
3377 | */ | |
3378 | ptr = ____cache_alloc(cachep, flags); | |
3379 | if (ptr) | |
3380 | goto out; | |
3381 | } | |
3382 | /* ___cache_alloc_node can fall back to other nodes */ | |
3383 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3384 | out: | |
3385 | local_irq_restore(save_flags); | |
3386 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
8c138bc0 | 3387 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3388 | flags); |
8c8cc2c1 | 3389 | |
c175eea4 | 3390 | if (likely(ptr)) |
8c138bc0 | 3391 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
c175eea4 | 3392 | |
d07dbea4 | 3393 | if (unlikely((flags & __GFP_ZERO) && ptr)) |
8c138bc0 | 3394 | memset(ptr, 0, cachep->object_size); |
d07dbea4 | 3395 | |
8c8cc2c1 PE |
3396 | return ptr; |
3397 | } | |
3398 | ||
3399 | static __always_inline void * | |
3400 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3401 | { | |
3402 | void *objp; | |
3403 | ||
3404 | if (unlikely(current->flags & (PF_SPREAD_SLAB | PF_MEMPOLICY))) { | |
3405 | objp = alternate_node_alloc(cache, flags); | |
3406 | if (objp) | |
3407 | goto out; | |
3408 | } | |
3409 | objp = ____cache_alloc(cache, flags); | |
3410 | ||
3411 | /* | |
3412 | * We may just have run out of memory on the local node. | |
3413 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3414 | */ | |
7d6e6d09 LS |
3415 | if (!objp) |
3416 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3417 | |
3418 | out: | |
3419 | return objp; | |
3420 | } | |
3421 | #else | |
3422 | ||
3423 | static __always_inline void * | |
3424 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3425 | { | |
3426 | return ____cache_alloc(cachep, flags); | |
3427 | } | |
3428 | ||
3429 | #endif /* CONFIG_NUMA */ | |
3430 | ||
3431 | static __always_inline void * | |
48356303 | 3432 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3433 | { |
3434 | unsigned long save_flags; | |
3435 | void *objp; | |
3436 | ||
dcce284a | 3437 | flags &= gfp_allowed_mask; |
7e85ee0c | 3438 | |
cf40bd16 NP |
3439 | lockdep_trace_alloc(flags); |
3440 | ||
773ff60e | 3441 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3442 | return NULL; |
3443 | ||
d79923fa GC |
3444 | cachep = memcg_kmem_get_cache(cachep, flags); |
3445 | ||
8c8cc2c1 PE |
3446 | cache_alloc_debugcheck_before(cachep, flags); |
3447 | local_irq_save(save_flags); | |
3448 | objp = __do_cache_alloc(cachep, flags); | |
3449 | local_irq_restore(save_flags); | |
3450 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
8c138bc0 | 3451 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3452 | flags); |
8c8cc2c1 PE |
3453 | prefetchw(objp); |
3454 | ||
c175eea4 | 3455 | if (likely(objp)) |
8c138bc0 | 3456 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
c175eea4 | 3457 | |
d07dbea4 | 3458 | if (unlikely((flags & __GFP_ZERO) && objp)) |
8c138bc0 | 3459 | memset(objp, 0, cachep->object_size); |
d07dbea4 | 3460 | |
8c8cc2c1 PE |
3461 | return objp; |
3462 | } | |
e498be7d CL |
3463 | |
3464 | /* | |
3465 | * Caller needs to acquire correct kmem_list's list_lock | |
3466 | */ | |
343e0d7a | 3467 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
b28a02de | 3468 | int node) |
1da177e4 LT |
3469 | { |
3470 | int i; | |
ce8eb6c4 | 3471 | struct kmem_cache_node *n; |
1da177e4 LT |
3472 | |
3473 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3474 | void *objp; |
1da177e4 | 3475 | struct slab *slabp; |
1da177e4 | 3476 | |
072bb0aa MG |
3477 | clear_obj_pfmemalloc(&objpp[i]); |
3478 | objp = objpp[i]; | |
3479 | ||
6ed5eb22 | 3480 | slabp = virt_to_slab(objp); |
ce8eb6c4 | 3481 | n = cachep->node[node]; |
1da177e4 | 3482 | list_del(&slabp->list); |
ff69416e | 3483 | check_spinlock_acquired_node(cachep, node); |
1da177e4 | 3484 | check_slabp(cachep, slabp); |
78d382d7 | 3485 | slab_put_obj(cachep, slabp, objp, node); |
1da177e4 | 3486 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3487 | n->free_objects++; |
1da177e4 LT |
3488 | check_slabp(cachep, slabp); |
3489 | ||
3490 | /* fixup slab chains */ | |
3491 | if (slabp->inuse == 0) { | |
ce8eb6c4 CL |
3492 | if (n->free_objects > n->free_limit) { |
3493 | n->free_objects -= cachep->num; | |
e5ac9c5a RT |
3494 | /* No need to drop any previously held |
3495 | * lock here, even if we have a off-slab slab | |
3496 | * descriptor it is guaranteed to come from | |
3497 | * a different cache, refer to comments before | |
3498 | * alloc_slabmgmt. | |
3499 | */ | |
1da177e4 LT |
3500 | slab_destroy(cachep, slabp); |
3501 | } else { | |
ce8eb6c4 | 3502 | list_add(&slabp->list, &n->slabs_free); |
1da177e4 LT |
3503 | } |
3504 | } else { | |
3505 | /* Unconditionally move a slab to the end of the | |
3506 | * partial list on free - maximum time for the | |
3507 | * other objects to be freed, too. | |
3508 | */ | |
ce8eb6c4 | 3509 | list_add_tail(&slabp->list, &n->slabs_partial); |
1da177e4 LT |
3510 | } |
3511 | } | |
3512 | } | |
3513 | ||
343e0d7a | 3514 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3515 | { |
3516 | int batchcount; | |
ce8eb6c4 | 3517 | struct kmem_cache_node *n; |
7d6e6d09 | 3518 | int node = numa_mem_id(); |
1da177e4 LT |
3519 | |
3520 | batchcount = ac->batchcount; | |
3521 | #if DEBUG | |
3522 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3523 | #endif | |
3524 | check_irq_off(); | |
ce8eb6c4 CL |
3525 | n = cachep->node[node]; |
3526 | spin_lock(&n->list_lock); | |
3527 | if (n->shared) { | |
3528 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3529 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3530 | if (max) { |
3531 | if (batchcount > max) | |
3532 | batchcount = max; | |
e498be7d | 3533 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3534 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3535 | shared_array->avail += batchcount; |
3536 | goto free_done; | |
3537 | } | |
3538 | } | |
3539 | ||
ff69416e | 3540 | free_block(cachep, ac->entry, batchcount, node); |
a737b3e2 | 3541 | free_done: |
1da177e4 LT |
3542 | #if STATS |
3543 | { | |
3544 | int i = 0; | |
3545 | struct list_head *p; | |
3546 | ||
ce8eb6c4 CL |
3547 | p = n->slabs_free.next; |
3548 | while (p != &(n->slabs_free)) { | |
1da177e4 LT |
3549 | struct slab *slabp; |
3550 | ||
3551 | slabp = list_entry(p, struct slab, list); | |
3552 | BUG_ON(slabp->inuse); | |
3553 | ||
3554 | i++; | |
3555 | p = p->next; | |
3556 | } | |
3557 | STATS_SET_FREEABLE(cachep, i); | |
3558 | } | |
3559 | #endif | |
ce8eb6c4 | 3560 | spin_unlock(&n->list_lock); |
1da177e4 | 3561 | ac->avail -= batchcount; |
a737b3e2 | 3562 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3563 | } |
3564 | ||
3565 | /* | |
a737b3e2 AM |
3566 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3567 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3568 | */ |
a947eb95 | 3569 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3570 | unsigned long caller) |
1da177e4 | 3571 | { |
9a2dba4b | 3572 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3573 | |
3574 | check_irq_off(); | |
d5cff635 | 3575 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3576 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3577 | |
8c138bc0 | 3578 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3579 | |
1807a1aa SS |
3580 | /* |
3581 | * Skip calling cache_free_alien() when the platform is not numa. | |
3582 | * This will avoid cache misses that happen while accessing slabp (which | |
3583 | * is per page memory reference) to get nodeid. Instead use a global | |
3584 | * variable to skip the call, which is mostly likely to be present in | |
3585 | * the cache. | |
3586 | */ | |
b6e68bc1 | 3587 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3588 | return; |
3589 | ||
1da177e4 LT |
3590 | if (likely(ac->avail < ac->limit)) { |
3591 | STATS_INC_FREEHIT(cachep); | |
1da177e4 LT |
3592 | } else { |
3593 | STATS_INC_FREEMISS(cachep); | |
3594 | cache_flusharray(cachep, ac); | |
1da177e4 | 3595 | } |
42c8c99c | 3596 | |
072bb0aa | 3597 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3598 | } |
3599 | ||
3600 | /** | |
3601 | * kmem_cache_alloc - Allocate an object | |
3602 | * @cachep: The cache to allocate from. | |
3603 | * @flags: See kmalloc(). | |
3604 | * | |
3605 | * Allocate an object from this cache. The flags are only relevant | |
3606 | * if the cache has no available objects. | |
3607 | */ | |
343e0d7a | 3608 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3609 | { |
48356303 | 3610 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3611 | |
ca2b84cb | 3612 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3613 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3614 | |
3615 | return ret; | |
1da177e4 LT |
3616 | } |
3617 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3618 | ||
0f24f128 | 3619 | #ifdef CONFIG_TRACING |
85beb586 | 3620 | void * |
4052147c | 3621 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3622 | { |
85beb586 SR |
3623 | void *ret; |
3624 | ||
48356303 | 3625 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3626 | |
3627 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3628 | size, cachep->size, flags); |
85beb586 | 3629 | return ret; |
36555751 | 3630 | } |
85beb586 | 3631 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3632 | #endif |
3633 | ||
1da177e4 | 3634 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3635 | /** |
3636 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3637 | * @cachep: The cache to allocate from. | |
3638 | * @flags: See kmalloc(). | |
3639 | * @nodeid: node number of the target node. | |
3640 | * | |
3641 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3642 | * node, which can improve the performance for cpu bound structures. | |
3643 | * | |
3644 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3645 | */ | |
8b98c169 CH |
3646 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3647 | { | |
48356303 | 3648 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3649 | |
ca2b84cb | 3650 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3651 | cachep->object_size, cachep->size, |
ca2b84cb | 3652 | flags, nodeid); |
36555751 EGM |
3653 | |
3654 | return ret; | |
8b98c169 | 3655 | } |
1da177e4 LT |
3656 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3657 | ||
0f24f128 | 3658 | #ifdef CONFIG_TRACING |
4052147c | 3659 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3660 | gfp_t flags, |
4052147c EG |
3661 | int nodeid, |
3662 | size_t size) | |
36555751 | 3663 | { |
85beb586 SR |
3664 | void *ret; |
3665 | ||
592f4145 | 3666 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3667 | |
85beb586 | 3668 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3669 | size, cachep->size, |
85beb586 SR |
3670 | flags, nodeid); |
3671 | return ret; | |
36555751 | 3672 | } |
85beb586 | 3673 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3674 | #endif |
3675 | ||
8b98c169 | 3676 | static __always_inline void * |
7c0cb9c6 | 3677 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3678 | { |
343e0d7a | 3679 | struct kmem_cache *cachep; |
97e2bde4 | 3680 | |
2c59dd65 | 3681 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3682 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3683 | return cachep; | |
4052147c | 3684 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3685 | } |
8b98c169 | 3686 | |
0bb38a5c | 3687 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
8b98c169 CH |
3688 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3689 | { | |
7c0cb9c6 | 3690 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3691 | } |
dbe5e69d | 3692 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3693 | |
3694 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3695 | int node, unsigned long caller) |
8b98c169 | 3696 | { |
7c0cb9c6 | 3697 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3698 | } |
3699 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
3700 | #else | |
3701 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3702 | { | |
7c0cb9c6 | 3703 | return __do_kmalloc_node(size, flags, node, 0); |
8b98c169 CH |
3704 | } |
3705 | EXPORT_SYMBOL(__kmalloc_node); | |
0bb38a5c | 3706 | #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */ |
8b98c169 | 3707 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3708 | |
3709 | /** | |
800590f5 | 3710 | * __do_kmalloc - allocate memory |
1da177e4 | 3711 | * @size: how many bytes of memory are required. |
800590f5 | 3712 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3713 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3714 | */ |
7fd6b141 | 3715 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3716 | unsigned long caller) |
1da177e4 | 3717 | { |
343e0d7a | 3718 | struct kmem_cache *cachep; |
36555751 | 3719 | void *ret; |
1da177e4 | 3720 | |
97e2bde4 MS |
3721 | /* If you want to save a few bytes .text space: replace |
3722 | * __ with kmem_. | |
3723 | * Then kmalloc uses the uninlined functions instead of the inline | |
3724 | * functions. | |
3725 | */ | |
2c59dd65 | 3726 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3727 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3728 | return cachep; | |
48356303 | 3729 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3730 | |
7c0cb9c6 | 3731 | trace_kmalloc(caller, ret, |
3b0efdfa | 3732 | size, cachep->size, flags); |
36555751 EGM |
3733 | |
3734 | return ret; | |
7fd6b141 PE |
3735 | } |
3736 | ||
7fd6b141 | 3737 | |
0bb38a5c | 3738 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
7fd6b141 PE |
3739 | void *__kmalloc(size_t size, gfp_t flags) |
3740 | { | |
7c0cb9c6 | 3741 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3742 | } |
3743 | EXPORT_SYMBOL(__kmalloc); | |
3744 | ||
ce71e27c | 3745 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3746 | { |
7c0cb9c6 | 3747 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3748 | } |
3749 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea CH |
3750 | |
3751 | #else | |
3752 | void *__kmalloc(size_t size, gfp_t flags) | |
3753 | { | |
7c0cb9c6 | 3754 | return __do_kmalloc(size, flags, 0); |
1d2c8eea CH |
3755 | } |
3756 | EXPORT_SYMBOL(__kmalloc); | |
7fd6b141 PE |
3757 | #endif |
3758 | ||
1da177e4 LT |
3759 | /** |
3760 | * kmem_cache_free - Deallocate an object | |
3761 | * @cachep: The cache the allocation was from. | |
3762 | * @objp: The previously allocated object. | |
3763 | * | |
3764 | * Free an object which was previously allocated from this | |
3765 | * cache. | |
3766 | */ | |
343e0d7a | 3767 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3768 | { |
3769 | unsigned long flags; | |
b9ce5ef4 GC |
3770 | cachep = cache_from_obj(cachep, objp); |
3771 | if (!cachep) | |
3772 | return; | |
1da177e4 LT |
3773 | |
3774 | local_irq_save(flags); | |
d97d476b | 3775 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3776 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3777 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3778 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3779 | local_irq_restore(flags); |
36555751 | 3780 | |
ca2b84cb | 3781 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3782 | } |
3783 | EXPORT_SYMBOL(kmem_cache_free); | |
3784 | ||
1da177e4 LT |
3785 | /** |
3786 | * kfree - free previously allocated memory | |
3787 | * @objp: pointer returned by kmalloc. | |
3788 | * | |
80e93eff PE |
3789 | * If @objp is NULL, no operation is performed. |
3790 | * | |
1da177e4 LT |
3791 | * Don't free memory not originally allocated by kmalloc() |
3792 | * or you will run into trouble. | |
3793 | */ | |
3794 | void kfree(const void *objp) | |
3795 | { | |
343e0d7a | 3796 | struct kmem_cache *c; |
1da177e4 LT |
3797 | unsigned long flags; |
3798 | ||
2121db74 PE |
3799 | trace_kfree(_RET_IP_, objp); |
3800 | ||
6cb8f913 | 3801 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3802 | return; |
3803 | local_irq_save(flags); | |
3804 | kfree_debugcheck(objp); | |
6ed5eb22 | 3805 | c = virt_to_cache(objp); |
8c138bc0 CL |
3806 | debug_check_no_locks_freed(objp, c->object_size); |
3807 | ||
3808 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3809 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3810 | local_irq_restore(flags); |
3811 | } | |
3812 | EXPORT_SYMBOL(kfree); | |
3813 | ||
e498be7d | 3814 | /* |
ce8eb6c4 | 3815 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3816 | */ |
83b519e8 | 3817 | static int alloc_kmemlist(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3818 | { |
3819 | int node; | |
ce8eb6c4 | 3820 | struct kmem_cache_node *n; |
cafeb02e | 3821 | struct array_cache *new_shared; |
3395ee05 | 3822 | struct array_cache **new_alien = NULL; |
e498be7d | 3823 | |
9c09a95c | 3824 | for_each_online_node(node) { |
cafeb02e | 3825 | |
3395ee05 | 3826 | if (use_alien_caches) { |
83b519e8 | 3827 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3828 | if (!new_alien) |
3829 | goto fail; | |
3830 | } | |
cafeb02e | 3831 | |
63109846 ED |
3832 | new_shared = NULL; |
3833 | if (cachep->shared) { | |
3834 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3835 | cachep->shared*cachep->batchcount, |
83b519e8 | 3836 | 0xbaadf00d, gfp); |
63109846 ED |
3837 | if (!new_shared) { |
3838 | free_alien_cache(new_alien); | |
3839 | goto fail; | |
3840 | } | |
0718dc2a | 3841 | } |
cafeb02e | 3842 | |
ce8eb6c4 CL |
3843 | n = cachep->node[node]; |
3844 | if (n) { | |
3845 | struct array_cache *shared = n->shared; | |
cafeb02e | 3846 | |
ce8eb6c4 | 3847 | spin_lock_irq(&n->list_lock); |
e498be7d | 3848 | |
cafeb02e | 3849 | if (shared) |
0718dc2a CL |
3850 | free_block(cachep, shared->entry, |
3851 | shared->avail, node); | |
e498be7d | 3852 | |
ce8eb6c4 CL |
3853 | n->shared = new_shared; |
3854 | if (!n->alien) { | |
3855 | n->alien = new_alien; | |
e498be7d CL |
3856 | new_alien = NULL; |
3857 | } | |
ce8eb6c4 | 3858 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3859 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3860 | spin_unlock_irq(&n->list_lock); |
cafeb02e | 3861 | kfree(shared); |
e498be7d CL |
3862 | free_alien_cache(new_alien); |
3863 | continue; | |
3864 | } | |
ce8eb6c4 CL |
3865 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3866 | if (!n) { | |
0718dc2a CL |
3867 | free_alien_cache(new_alien); |
3868 | kfree(new_shared); | |
e498be7d | 3869 | goto fail; |
0718dc2a | 3870 | } |
e498be7d | 3871 | |
ce8eb6c4 CL |
3872 | kmem_cache_node_init(n); |
3873 | n->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
a737b3e2 | 3874 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
ce8eb6c4 CL |
3875 | n->shared = new_shared; |
3876 | n->alien = new_alien; | |
3877 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3878 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3879 | cachep->node[node] = n; |
e498be7d | 3880 | } |
cafeb02e | 3881 | return 0; |
0718dc2a | 3882 | |
a737b3e2 | 3883 | fail: |
3b0efdfa | 3884 | if (!cachep->list.next) { |
0718dc2a CL |
3885 | /* Cache is not active yet. Roll back what we did */ |
3886 | node--; | |
3887 | while (node >= 0) { | |
6a67368c | 3888 | if (cachep->node[node]) { |
ce8eb6c4 | 3889 | n = cachep->node[node]; |
0718dc2a | 3890 | |
ce8eb6c4 CL |
3891 | kfree(n->shared); |
3892 | free_alien_cache(n->alien); | |
3893 | kfree(n); | |
6a67368c | 3894 | cachep->node[node] = NULL; |
0718dc2a CL |
3895 | } |
3896 | node--; | |
3897 | } | |
3898 | } | |
cafeb02e | 3899 | return -ENOMEM; |
e498be7d CL |
3900 | } |
3901 | ||
1da177e4 | 3902 | struct ccupdate_struct { |
343e0d7a | 3903 | struct kmem_cache *cachep; |
acfe7d74 | 3904 | struct array_cache *new[0]; |
1da177e4 LT |
3905 | }; |
3906 | ||
3907 | static void do_ccupdate_local(void *info) | |
3908 | { | |
a737b3e2 | 3909 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3910 | struct array_cache *old; |
3911 | ||
3912 | check_irq_off(); | |
9a2dba4b | 3913 | old = cpu_cache_get(new->cachep); |
e498be7d | 3914 | |
1da177e4 LT |
3915 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3916 | new->new[smp_processor_id()] = old; | |
3917 | } | |
3918 | ||
18004c5d | 3919 | /* Always called with the slab_mutex held */ |
943a451a | 3920 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3921 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3922 | { |
d2e7b7d0 | 3923 | struct ccupdate_struct *new; |
2ed3a4ef | 3924 | int i; |
1da177e4 | 3925 | |
acfe7d74 ED |
3926 | new = kzalloc(sizeof(*new) + nr_cpu_ids * sizeof(struct array_cache *), |
3927 | gfp); | |
d2e7b7d0 SS |
3928 | if (!new) |
3929 | return -ENOMEM; | |
3930 | ||
e498be7d | 3931 | for_each_online_cpu(i) { |
7d6e6d09 | 3932 | new->new[i] = alloc_arraycache(cpu_to_mem(i), limit, |
83b519e8 | 3933 | batchcount, gfp); |
d2e7b7d0 | 3934 | if (!new->new[i]) { |
b28a02de | 3935 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3936 | kfree(new->new[i]); |
3937 | kfree(new); | |
e498be7d | 3938 | return -ENOMEM; |
1da177e4 LT |
3939 | } |
3940 | } | |
d2e7b7d0 | 3941 | new->cachep = cachep; |
1da177e4 | 3942 | |
15c8b6c1 | 3943 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 3944 | |
1da177e4 | 3945 | check_irq_on(); |
1da177e4 LT |
3946 | cachep->batchcount = batchcount; |
3947 | cachep->limit = limit; | |
e498be7d | 3948 | cachep->shared = shared; |
1da177e4 | 3949 | |
e498be7d | 3950 | for_each_online_cpu(i) { |
d2e7b7d0 | 3951 | struct array_cache *ccold = new->new[i]; |
1da177e4 LT |
3952 | if (!ccold) |
3953 | continue; | |
6a67368c | 3954 | spin_lock_irq(&cachep->node[cpu_to_mem(i)]->list_lock); |
7d6e6d09 | 3955 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_mem(i)); |
6a67368c | 3956 | spin_unlock_irq(&cachep->node[cpu_to_mem(i)]->list_lock); |
1da177e4 LT |
3957 | kfree(ccold); |
3958 | } | |
d2e7b7d0 | 3959 | kfree(new); |
83b519e8 | 3960 | return alloc_kmemlist(cachep, gfp); |
1da177e4 LT |
3961 | } |
3962 | ||
943a451a GC |
3963 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3964 | int batchcount, int shared, gfp_t gfp) | |
3965 | { | |
3966 | int ret; | |
3967 | struct kmem_cache *c = NULL; | |
3968 | int i = 0; | |
3969 | ||
3970 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3971 | ||
3972 | if (slab_state < FULL) | |
3973 | return ret; | |
3974 | ||
3975 | if ((ret < 0) || !is_root_cache(cachep)) | |
3976 | return ret; | |
3977 | ||
ebe945c2 | 3978 | VM_BUG_ON(!mutex_is_locked(&slab_mutex)); |
943a451a GC |
3979 | for_each_memcg_cache_index(i) { |
3980 | c = cache_from_memcg(cachep, i); | |
3981 | if (c) | |
3982 | /* return value determined by the parent cache only */ | |
3983 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
3984 | } | |
3985 | ||
3986 | return ret; | |
3987 | } | |
3988 | ||
18004c5d | 3989 | /* Called with slab_mutex held always */ |
83b519e8 | 3990 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3991 | { |
3992 | int err; | |
943a451a GC |
3993 | int limit = 0; |
3994 | int shared = 0; | |
3995 | int batchcount = 0; | |
3996 | ||
3997 | if (!is_root_cache(cachep)) { | |
3998 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3999 | limit = root->limit; | |
4000 | shared = root->shared; | |
4001 | batchcount = root->batchcount; | |
4002 | } | |
1da177e4 | 4003 | |
943a451a GC |
4004 | if (limit && shared && batchcount) |
4005 | goto skip_setup; | |
a737b3e2 AM |
4006 | /* |
4007 | * The head array serves three purposes: | |
1da177e4 LT |
4008 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
4009 | * - reduce the number of spinlock operations. | |
a737b3e2 | 4010 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
4011 | * bufctl chains: array operations are cheaper. |
4012 | * The numbers are guessed, we should auto-tune as described by | |
4013 | * Bonwick. | |
4014 | */ | |
3b0efdfa | 4015 | if (cachep->size > 131072) |
1da177e4 | 4016 | limit = 1; |
3b0efdfa | 4017 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 4018 | limit = 8; |
3b0efdfa | 4019 | else if (cachep->size > 1024) |
1da177e4 | 4020 | limit = 24; |
3b0efdfa | 4021 | else if (cachep->size > 256) |
1da177e4 LT |
4022 | limit = 54; |
4023 | else | |
4024 | limit = 120; | |
4025 | ||
a737b3e2 AM |
4026 | /* |
4027 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
4028 | * allocation behaviour: Most allocs on one cpu, most free operations |
4029 | * on another cpu. For these cases, an efficient object passing between | |
4030 | * cpus is necessary. This is provided by a shared array. The array | |
4031 | * replaces Bonwick's magazine layer. | |
4032 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
4033 | * to a larger limit. Thus disabled by default. | |
4034 | */ | |
4035 | shared = 0; | |
3b0efdfa | 4036 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 4037 | shared = 8; |
1da177e4 LT |
4038 | |
4039 | #if DEBUG | |
a737b3e2 AM |
4040 | /* |
4041 | * With debugging enabled, large batchcount lead to excessively long | |
4042 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
4043 | */ |
4044 | if (limit > 32) | |
4045 | limit = 32; | |
4046 | #endif | |
943a451a GC |
4047 | batchcount = (limit + 1) / 2; |
4048 | skip_setup: | |
4049 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
4050 | if (err) |
4051 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 4052 | cachep->name, -err); |
2ed3a4ef | 4053 | return err; |
1da177e4 LT |
4054 | } |
4055 | ||
1b55253a | 4056 | /* |
ce8eb6c4 CL |
4057 | * Drain an array if it contains any elements taking the node lock only if |
4058 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 4059 | * if drain_array() is used on the shared array. |
1b55253a | 4060 | */ |
ce8eb6c4 | 4061 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 4062 | struct array_cache *ac, int force, int node) |
1da177e4 LT |
4063 | { |
4064 | int tofree; | |
4065 | ||
1b55253a CL |
4066 | if (!ac || !ac->avail) |
4067 | return; | |
1da177e4 LT |
4068 | if (ac->touched && !force) { |
4069 | ac->touched = 0; | |
b18e7e65 | 4070 | } else { |
ce8eb6c4 | 4071 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
4072 | if (ac->avail) { |
4073 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
4074 | if (tofree > ac->avail) | |
4075 | tofree = (ac->avail + 1) / 2; | |
4076 | free_block(cachep, ac->entry, tofree, node); | |
4077 | ac->avail -= tofree; | |
4078 | memmove(ac->entry, &(ac->entry[tofree]), | |
4079 | sizeof(void *) * ac->avail); | |
4080 | } | |
ce8eb6c4 | 4081 | spin_unlock_irq(&n->list_lock); |
1da177e4 LT |
4082 | } |
4083 | } | |
4084 | ||
4085 | /** | |
4086 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4087 | * @w: work descriptor |
1da177e4 LT |
4088 | * |
4089 | * Called from workqueue/eventd every few seconds. | |
4090 | * Purpose: | |
4091 | * - clear the per-cpu caches for this CPU. | |
4092 | * - return freeable pages to the main free memory pool. | |
4093 | * | |
a737b3e2 AM |
4094 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4095 | * again on the next iteration. | |
1da177e4 | 4096 | */ |
7c5cae36 | 4097 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4098 | { |
7a7c381d | 4099 | struct kmem_cache *searchp; |
ce8eb6c4 | 4100 | struct kmem_cache_node *n; |
7d6e6d09 | 4101 | int node = numa_mem_id(); |
bf6aede7 | 4102 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4103 | |
18004c5d | 4104 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 4105 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4106 | goto out; |
1da177e4 | 4107 | |
18004c5d | 4108 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
4109 | check_irq_on(); |
4110 | ||
35386e3b | 4111 | /* |
ce8eb6c4 | 4112 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4113 | * have established with reasonable certainty that |
4114 | * we can do some work if the lock was obtained. | |
4115 | */ | |
ce8eb6c4 | 4116 | n = searchp->node[node]; |
35386e3b | 4117 | |
ce8eb6c4 | 4118 | reap_alien(searchp, n); |
1da177e4 | 4119 | |
ce8eb6c4 | 4120 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 4121 | |
35386e3b CL |
4122 | /* |
4123 | * These are racy checks but it does not matter | |
4124 | * if we skip one check or scan twice. | |
4125 | */ | |
ce8eb6c4 | 4126 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4127 | goto next; |
1da177e4 | 4128 | |
ce8eb6c4 | 4129 | n->next_reap = jiffies + REAPTIMEOUT_LIST3; |
1da177e4 | 4130 | |
ce8eb6c4 | 4131 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 4132 | |
ce8eb6c4 CL |
4133 | if (n->free_touched) |
4134 | n->free_touched = 0; | |
ed11d9eb CL |
4135 | else { |
4136 | int freed; | |
1da177e4 | 4137 | |
ce8eb6c4 | 4138 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4139 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4140 | STATS_ADD_REAPED(searchp, freed); | |
4141 | } | |
35386e3b | 4142 | next: |
1da177e4 LT |
4143 | cond_resched(); |
4144 | } | |
4145 | check_irq_on(); | |
18004c5d | 4146 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4147 | next_reap_node(); |
7c5cae36 | 4148 | out: |
a737b3e2 | 4149 | /* Set up the next iteration */ |
7c5cae36 | 4150 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_CPUC)); |
1da177e4 LT |
4151 | } |
4152 | ||
158a9624 | 4153 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 4154 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4155 | { |
b28a02de PE |
4156 | struct slab *slabp; |
4157 | unsigned long active_objs; | |
4158 | unsigned long num_objs; | |
4159 | unsigned long active_slabs = 0; | |
4160 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 4161 | const char *name; |
1da177e4 | 4162 | char *error = NULL; |
e498be7d | 4163 | int node; |
ce8eb6c4 | 4164 | struct kmem_cache_node *n; |
1da177e4 | 4165 | |
1da177e4 LT |
4166 | active_objs = 0; |
4167 | num_slabs = 0; | |
e498be7d | 4168 | for_each_online_node(node) { |
ce8eb6c4 CL |
4169 | n = cachep->node[node]; |
4170 | if (!n) | |
e498be7d CL |
4171 | continue; |
4172 | ||
ca3b9b91 | 4173 | check_irq_on(); |
ce8eb6c4 | 4174 | spin_lock_irq(&n->list_lock); |
e498be7d | 4175 | |
ce8eb6c4 | 4176 | list_for_each_entry(slabp, &n->slabs_full, list) { |
e498be7d CL |
4177 | if (slabp->inuse != cachep->num && !error) |
4178 | error = "slabs_full accounting error"; | |
4179 | active_objs += cachep->num; | |
4180 | active_slabs++; | |
4181 | } | |
ce8eb6c4 | 4182 | list_for_each_entry(slabp, &n->slabs_partial, list) { |
e498be7d CL |
4183 | if (slabp->inuse == cachep->num && !error) |
4184 | error = "slabs_partial inuse accounting error"; | |
4185 | if (!slabp->inuse && !error) | |
4186 | error = "slabs_partial/inuse accounting error"; | |
4187 | active_objs += slabp->inuse; | |
4188 | active_slabs++; | |
4189 | } | |
ce8eb6c4 | 4190 | list_for_each_entry(slabp, &n->slabs_free, list) { |
e498be7d CL |
4191 | if (slabp->inuse && !error) |
4192 | error = "slabs_free/inuse accounting error"; | |
4193 | num_slabs++; | |
4194 | } | |
ce8eb6c4 CL |
4195 | free_objects += n->free_objects; |
4196 | if (n->shared) | |
4197 | shared_avail += n->shared->avail; | |
e498be7d | 4198 | |
ce8eb6c4 | 4199 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4200 | } |
b28a02de PE |
4201 | num_slabs += active_slabs; |
4202 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4203 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4204 | error = "free_objects accounting error"; |
4205 | ||
b28a02de | 4206 | name = cachep->name; |
1da177e4 LT |
4207 | if (error) |
4208 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4209 | ||
0d7561c6 GC |
4210 | sinfo->active_objs = active_objs; |
4211 | sinfo->num_objs = num_objs; | |
4212 | sinfo->active_slabs = active_slabs; | |
4213 | sinfo->num_slabs = num_slabs; | |
4214 | sinfo->shared_avail = shared_avail; | |
4215 | sinfo->limit = cachep->limit; | |
4216 | sinfo->batchcount = cachep->batchcount; | |
4217 | sinfo->shared = cachep->shared; | |
4218 | sinfo->objects_per_slab = cachep->num; | |
4219 | sinfo->cache_order = cachep->gfporder; | |
4220 | } | |
4221 | ||
4222 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4223 | { | |
1da177e4 | 4224 | #if STATS |
ce8eb6c4 | 4225 | { /* node stats */ |
1da177e4 LT |
4226 | unsigned long high = cachep->high_mark; |
4227 | unsigned long allocs = cachep->num_allocations; | |
4228 | unsigned long grown = cachep->grown; | |
4229 | unsigned long reaped = cachep->reaped; | |
4230 | unsigned long errors = cachep->errors; | |
4231 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4232 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4233 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4234 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4235 | |
e92dd4fd JP |
4236 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4237 | "%4lu %4lu %4lu %4lu %4lu", | |
4238 | allocs, high, grown, | |
4239 | reaped, errors, max_freeable, node_allocs, | |
4240 | node_frees, overflows); | |
1da177e4 LT |
4241 | } |
4242 | /* cpu stats */ | |
4243 | { | |
4244 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4245 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4246 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4247 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4248 | ||
4249 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4250 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4251 | } |
4252 | #endif | |
1da177e4 LT |
4253 | } |
4254 | ||
1da177e4 LT |
4255 | #define MAX_SLABINFO_WRITE 128 |
4256 | /** | |
4257 | * slabinfo_write - Tuning for the slab allocator | |
4258 | * @file: unused | |
4259 | * @buffer: user buffer | |
4260 | * @count: data length | |
4261 | * @ppos: unused | |
4262 | */ | |
b7454ad3 | 4263 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4264 | size_t count, loff_t *ppos) |
1da177e4 | 4265 | { |
b28a02de | 4266 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4267 | int limit, batchcount, shared, res; |
7a7c381d | 4268 | struct kmem_cache *cachep; |
b28a02de | 4269 | |
1da177e4 LT |
4270 | if (count > MAX_SLABINFO_WRITE) |
4271 | return -EINVAL; | |
4272 | if (copy_from_user(&kbuf, buffer, count)) | |
4273 | return -EFAULT; | |
b28a02de | 4274 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4275 | |
4276 | tmp = strchr(kbuf, ' '); | |
4277 | if (!tmp) | |
4278 | return -EINVAL; | |
4279 | *tmp = '\0'; | |
4280 | tmp++; | |
4281 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4282 | return -EINVAL; | |
4283 | ||
4284 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4285 | mutex_lock(&slab_mutex); |
1da177e4 | 4286 | res = -EINVAL; |
18004c5d | 4287 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4288 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4289 | if (limit < 1 || batchcount < 1 || |
4290 | batchcount > limit || shared < 0) { | |
e498be7d | 4291 | res = 0; |
1da177e4 | 4292 | } else { |
e498be7d | 4293 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4294 | batchcount, shared, |
4295 | GFP_KERNEL); | |
1da177e4 LT |
4296 | } |
4297 | break; | |
4298 | } | |
4299 | } | |
18004c5d | 4300 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4301 | if (res >= 0) |
4302 | res = count; | |
4303 | return res; | |
4304 | } | |
871751e2 AV |
4305 | |
4306 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4307 | ||
4308 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4309 | { | |
18004c5d CL |
4310 | mutex_lock(&slab_mutex); |
4311 | return seq_list_start(&slab_caches, *pos); | |
871751e2 AV |
4312 | } |
4313 | ||
4314 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4315 | { | |
4316 | unsigned long *p; | |
4317 | int l; | |
4318 | if (!v) | |
4319 | return 1; | |
4320 | l = n[1]; | |
4321 | p = n + 2; | |
4322 | while (l) { | |
4323 | int i = l/2; | |
4324 | unsigned long *q = p + 2 * i; | |
4325 | if (*q == v) { | |
4326 | q[1]++; | |
4327 | return 1; | |
4328 | } | |
4329 | if (*q > v) { | |
4330 | l = i; | |
4331 | } else { | |
4332 | p = q + 2; | |
4333 | l -= i + 1; | |
4334 | } | |
4335 | } | |
4336 | if (++n[1] == n[0]) | |
4337 | return 0; | |
4338 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4339 | p[0] = v; | |
4340 | p[1] = 1; | |
4341 | return 1; | |
4342 | } | |
4343 | ||
4344 | static void handle_slab(unsigned long *n, struct kmem_cache *c, struct slab *s) | |
4345 | { | |
4346 | void *p; | |
4347 | int i; | |
4348 | if (n[0] == n[1]) | |
4349 | return; | |
3b0efdfa | 4350 | for (i = 0, p = s->s_mem; i < c->num; i++, p += c->size) { |
871751e2 AV |
4351 | if (slab_bufctl(s)[i] != BUFCTL_ACTIVE) |
4352 | continue; | |
4353 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) | |
4354 | return; | |
4355 | } | |
4356 | } | |
4357 | ||
4358 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4359 | { | |
4360 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4361 | unsigned long offset, size; |
9281acea | 4362 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4363 | |
a5c43dae | 4364 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4365 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4366 | if (modname[0]) |
871751e2 AV |
4367 | seq_printf(m, " [%s]", modname); |
4368 | return; | |
4369 | } | |
4370 | #endif | |
4371 | seq_printf(m, "%p", (void *)address); | |
4372 | } | |
4373 | ||
4374 | static int leaks_show(struct seq_file *m, void *p) | |
4375 | { | |
0672aa7c | 4376 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
871751e2 | 4377 | struct slab *slabp; |
ce8eb6c4 | 4378 | struct kmem_cache_node *n; |
871751e2 | 4379 | const char *name; |
db845067 | 4380 | unsigned long *x = m->private; |
871751e2 AV |
4381 | int node; |
4382 | int i; | |
4383 | ||
4384 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4385 | return 0; | |
4386 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4387 | return 0; | |
4388 | ||
4389 | /* OK, we can do it */ | |
4390 | ||
db845067 | 4391 | x[1] = 0; |
871751e2 AV |
4392 | |
4393 | for_each_online_node(node) { | |
ce8eb6c4 CL |
4394 | n = cachep->node[node]; |
4395 | if (!n) | |
871751e2 AV |
4396 | continue; |
4397 | ||
4398 | check_irq_on(); | |
ce8eb6c4 | 4399 | spin_lock_irq(&n->list_lock); |
871751e2 | 4400 | |
ce8eb6c4 | 4401 | list_for_each_entry(slabp, &n->slabs_full, list) |
db845067 | 4402 | handle_slab(x, cachep, slabp); |
ce8eb6c4 | 4403 | list_for_each_entry(slabp, &n->slabs_partial, list) |
db845067 | 4404 | handle_slab(x, cachep, slabp); |
ce8eb6c4 | 4405 | spin_unlock_irq(&n->list_lock); |
871751e2 AV |
4406 | } |
4407 | name = cachep->name; | |
db845067 | 4408 | if (x[0] == x[1]) { |
871751e2 | 4409 | /* Increase the buffer size */ |
18004c5d | 4410 | mutex_unlock(&slab_mutex); |
db845067 | 4411 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4412 | if (!m->private) { |
4413 | /* Too bad, we are really out */ | |
db845067 | 4414 | m->private = x; |
18004c5d | 4415 | mutex_lock(&slab_mutex); |
871751e2 AV |
4416 | return -ENOMEM; |
4417 | } | |
db845067 CL |
4418 | *(unsigned long *)m->private = x[0] * 2; |
4419 | kfree(x); | |
18004c5d | 4420 | mutex_lock(&slab_mutex); |
871751e2 AV |
4421 | /* Now make sure this entry will be retried */ |
4422 | m->count = m->size; | |
4423 | return 0; | |
4424 | } | |
db845067 CL |
4425 | for (i = 0; i < x[1]; i++) { |
4426 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4427 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4428 | seq_putc(m, '\n'); |
4429 | } | |
d2e7b7d0 | 4430 | |
871751e2 AV |
4431 | return 0; |
4432 | } | |
4433 | ||
a0ec95a8 | 4434 | static const struct seq_operations slabstats_op = { |
871751e2 | 4435 | .start = leaks_start, |
276a2439 WL |
4436 | .next = slab_next, |
4437 | .stop = slab_stop, | |
871751e2 AV |
4438 | .show = leaks_show, |
4439 | }; | |
a0ec95a8 AD |
4440 | |
4441 | static int slabstats_open(struct inode *inode, struct file *file) | |
4442 | { | |
4443 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4444 | int ret = -ENOMEM; | |
4445 | if (n) { | |
4446 | ret = seq_open(file, &slabstats_op); | |
4447 | if (!ret) { | |
4448 | struct seq_file *m = file->private_data; | |
4449 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4450 | m->private = n; | |
4451 | n = NULL; | |
4452 | } | |
4453 | kfree(n); | |
4454 | } | |
4455 | return ret; | |
4456 | } | |
4457 | ||
4458 | static const struct file_operations proc_slabstats_operations = { | |
4459 | .open = slabstats_open, | |
4460 | .read = seq_read, | |
4461 | .llseek = seq_lseek, | |
4462 | .release = seq_release_private, | |
4463 | }; | |
4464 | #endif | |
4465 | ||
4466 | static int __init slab_proc_init(void) | |
4467 | { | |
4468 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4469 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4470 | #endif |
a0ec95a8 AD |
4471 | return 0; |
4472 | } | |
4473 | module_init(slab_proc_init); | |
1da177e4 LT |
4474 | #endif |
4475 | ||
00e145b6 MS |
4476 | /** |
4477 | * ksize - get the actual amount of memory allocated for a given object | |
4478 | * @objp: Pointer to the object | |
4479 | * | |
4480 | * kmalloc may internally round up allocations and return more memory | |
4481 | * than requested. ksize() can be used to determine the actual amount of | |
4482 | * memory allocated. The caller may use this additional memory, even though | |
4483 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4484 | * The caller must guarantee that objp points to a valid object previously | |
4485 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4486 | * must not be freed during the duration of the call. | |
4487 | */ | |
fd76bab2 | 4488 | size_t ksize(const void *objp) |
1da177e4 | 4489 | { |
ef8b4520 CL |
4490 | BUG_ON(!objp); |
4491 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4492 | return 0; |
1da177e4 | 4493 | |
8c138bc0 | 4494 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4495 | } |
b1aabecd | 4496 | EXPORT_SYMBOL(ksize); |