Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit 00afa758067ac1c947149ef766adcdfe30c44d7d

Authored by Pekka Enberg
1 parent 6b7b284958
Exists in master and in 7 other branches imx_3.0.35_4.1.0, imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-l5.0.0_1.0.0-ga

SLAB: Fix lockdep annotation breakage 

Commit ce79ddc8e2376a9a93c7d42daf89bfcbb9187e62 ("SLAB: Fix lockdep annotations
for CPU hotplug") broke init_node_lock_keys() off-slab logic which causes
lockdep false positives.

Fix that up by reverting the logic back to original while keeping CPU hotplug
fixes intact.

Reported-and-tested-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reported-and-tested-by: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>

Showing 1 changed file with 2 additions and 2 deletions Inline Diff

1 /* 1 /*
2 * linux/mm/slab.c 2 * linux/mm/slab.c
3 * Written by Mark Hemment, 1996/97. 3 * Written by Mark Hemment, 1996/97.
4 * (markhe@nextd.demon.co.uk) 4 * (markhe@nextd.demon.co.uk)
5 * 5 *
6 * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli 6 * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli
7 * 7 *
8 * Major cleanup, different bufctl logic, per-cpu arrays 8 * Major cleanup, different bufctl logic, per-cpu arrays
9 * (c) 2000 Manfred Spraul 9 * (c) 2000 Manfred Spraul
10 * 10 *
11 * Cleanup, make the head arrays unconditional, preparation for NUMA 11 * Cleanup, make the head arrays unconditional, preparation for NUMA
12 * (c) 2002 Manfred Spraul 12 * (c) 2002 Manfred Spraul
13 * 13 *
14 * An implementation of the Slab Allocator as described in outline in; 14 * An implementation of the Slab Allocator as described in outline in;
15 * UNIX Internals: The New Frontiers by Uresh Vahalia 15 * UNIX Internals: The New Frontiers by Uresh Vahalia
16 * Pub: Prentice Hall ISBN 0-13-101908-2 16 * Pub: Prentice Hall ISBN 0-13-101908-2
17 * or with a little more detail in; 17 * or with a little more detail in;
18 * The Slab Allocator: An Object-Caching Kernel Memory Allocator 18 * The Slab Allocator: An Object-Caching Kernel Memory Allocator
19 * Jeff Bonwick (Sun Microsystems). 19 * Jeff Bonwick (Sun Microsystems).
20 * Presented at: USENIX Summer 1994 Technical Conference 20 * Presented at: USENIX Summer 1994 Technical Conference
21 * 21 *
22 * The memory is organized in caches, one cache for each object type. 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) 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 24 * Each cache consists out of many slabs (they are small (usually one
25 * page long) and always contiguous), and each slab contains multiple 25 * page long) and always contiguous), and each slab contains multiple
26 * initialized objects. 26 * initialized objects.
27 * 27 *
28 * This means, that your constructor is used only for newly allocated 28 * This means, that your constructor is used only for newly allocated
29 * slabs and you must pass objects with the same initializations to 29 * slabs and you must pass objects with the same initializations to
30 * kmem_cache_free. 30 * kmem_cache_free.
31 * 31 *
32 * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, 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 33 * normal). If you need a special memory type, then must create a new
34 * cache for that memory type. 34 * cache for that memory type.
35 * 35 *
36 * In order to reduce fragmentation, the slabs are sorted in 3 groups: 36 * In order to reduce fragmentation, the slabs are sorted in 3 groups:
37 * full slabs with 0 free objects 37 * full slabs with 0 free objects
38 * partial slabs 38 * partial slabs
39 * empty slabs with no allocated objects 39 * empty slabs with no allocated objects
40 * 40 *
41 * If partial slabs exist, then new allocations come from these slabs, 41 * If partial slabs exist, then new allocations come from these slabs,
42 * otherwise from empty slabs or new slabs are allocated. 42 * otherwise from empty slabs or new slabs are allocated.
43 * 43 *
44 * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache 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. 45 * during kmem_cache_destroy(). The caller must prevent concurrent allocs.
46 * 46 *
47 * Each cache has a short per-cpu head array, most allocs 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 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. 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. 50 * The head array is strictly LIFO and should improve the cache hit rates.
51 * On SMP, it additionally reduces the spinlock operations. 51 * On SMP, it additionally reduces the spinlock operations.
52 * 52 *
53 * The c_cpuarray may not be read with enabled local interrupts - 53 * The c_cpuarray may not be read with enabled local interrupts -
54 * it's changed with a smp_call_function(). 54 * it's changed with a smp_call_function().
55 * 55 *
56 * SMP synchronization: 56 * SMP synchronization:
57 * constructors and destructors are called without any locking. 57 * constructors and destructors are called without any locking.
58 * Several members in struct kmem_cache and struct slab never change, they 58 * Several members in struct kmem_cache and struct slab never change, they
59 * are accessed without any locking. 59 * are accessed without any locking.
60 * The per-cpu arrays are never accessed from the wrong cpu, no 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. 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. 62 * The non-constant members are protected with a per-cache irq spinlock.
63 * 63 *
64 * Many thanks to Mark Hemment, who wrote another per-cpu slab patch 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 65 * in 2000 - many ideas in the current implementation are derived from
66 * his patch. 66 * his patch.
67 * 67 *
68 * Further notes from the original documentation: 68 * Further notes from the original documentation:
69 * 69 *
70 * 11 April '97. Started multi-threading - markhe 70 * 11 April '97. Started multi-threading - markhe
71 * The global cache-chain is protected by the mutex 'cache_chain_mutex'. 71 * The global cache-chain is protected by the mutex 'cache_chain_mutex'.
72 * The sem is only needed when accessing/extending the cache-chain, which 72 * The sem is only needed when accessing/extending the cache-chain, which
73 * can never happen inside an interrupt (kmem_cache_create(), 73 * can never happen inside an interrupt (kmem_cache_create(),
74 * kmem_cache_shrink() and kmem_cache_reap()). 74 * kmem_cache_shrink() and kmem_cache_reap()).
75 * 75 *
76 * At present, each engine can be growing a cache. This should be blocked. 76 * At present, each engine can be growing a cache. This should be blocked.
77 * 77 *
78 * 15 March 2005. NUMA slab allocator. 78 * 15 March 2005. NUMA slab allocator.
79 * Shai Fultheim <shai@scalex86.org>. 79 * Shai Fultheim <shai@scalex86.org>.
80 * Shobhit Dayal <shobhit@calsoftinc.com> 80 * Shobhit Dayal <shobhit@calsoftinc.com>
81 * Alok N Kataria <alokk@calsoftinc.com> 81 * Alok N Kataria <alokk@calsoftinc.com>
82 * Christoph Lameter <christoph@lameter.com> 82 * Christoph Lameter <christoph@lameter.com>
83 * 83 *
84 * Modified the slab allocator to be node aware on NUMA systems. 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. 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. 86 * All object allocations for a node occur from node specific slab lists.
87 */ 87 */
88 88
89 #include <linux/slab.h> 89 #include <linux/slab.h>
90 #include <linux/mm.h> 90 #include <linux/mm.h>
91 #include <linux/poison.h> 91 #include <linux/poison.h>
92 #include <linux/swap.h> 92 #include <linux/swap.h>
93 #include <linux/cache.h> 93 #include <linux/cache.h>
94 #include <linux/interrupt.h> 94 #include <linux/interrupt.h>
95 #include <linux/init.h> 95 #include <linux/init.h>
96 #include <linux/compiler.h> 96 #include <linux/compiler.h>
97 #include <linux/cpuset.h> 97 #include <linux/cpuset.h>
98 #include <linux/proc_fs.h> 98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h> 99 #include <linux/seq_file.h>
100 #include <linux/notifier.h> 100 #include <linux/notifier.h>
101 #include <linux/kallsyms.h> 101 #include <linux/kallsyms.h>
102 #include <linux/cpu.h> 102 #include <linux/cpu.h>
103 #include <linux/sysctl.h> 103 #include <linux/sysctl.h>
104 #include <linux/module.h> 104 #include <linux/module.h>
105 #include <linux/kmemtrace.h> 105 #include <linux/kmemtrace.h>
106 #include <linux/rcupdate.h> 106 #include <linux/rcupdate.h>
107 #include <linux/string.h> 107 #include <linux/string.h>
108 #include <linux/uaccess.h> 108 #include <linux/uaccess.h>
109 #include <linux/nodemask.h> 109 #include <linux/nodemask.h>
110 #include <linux/kmemleak.h> 110 #include <linux/kmemleak.h>
111 #include <linux/mempolicy.h> 111 #include <linux/mempolicy.h>
112 #include <linux/mutex.h> 112 #include <linux/mutex.h>
113 #include <linux/fault-inject.h> 113 #include <linux/fault-inject.h>
114 #include <linux/rtmutex.h> 114 #include <linux/rtmutex.h>
115 #include <linux/reciprocal_div.h> 115 #include <linux/reciprocal_div.h>
116 #include <linux/debugobjects.h> 116 #include <linux/debugobjects.h>
117 #include <linux/kmemcheck.h> 117 #include <linux/kmemcheck.h>
118 118
119 #include <asm/cacheflush.h> 119 #include <asm/cacheflush.h>
120 #include <asm/tlbflush.h> 120 #include <asm/tlbflush.h>
121 #include <asm/page.h> 121 #include <asm/page.h>
122 122
123 /* 123 /*
124 * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. 124 * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
125 * 0 for faster, smaller code (especially in the critical paths). 125 * 0 for faster, smaller code (especially in the critical paths).
126 * 126 *
127 * STATS - 1 to collect stats for /proc/slabinfo. 127 * STATS - 1 to collect stats for /proc/slabinfo.
128 * 0 for faster, smaller code (especially in the critical paths). 128 * 0 for faster, smaller code (especially in the critical paths).
129 * 129 *
130 * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) 130 * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible)
131 */ 131 */
132 132
133 #ifdef CONFIG_DEBUG_SLAB 133 #ifdef CONFIG_DEBUG_SLAB
134 #define DEBUG 1 134 #define DEBUG 1
135 #define STATS 1 135 #define STATS 1
136 #define FORCED_DEBUG 1 136 #define FORCED_DEBUG 1
137 #else 137 #else
138 #define DEBUG 0 138 #define DEBUG 0
139 #define STATS 0 139 #define STATS 0
140 #define FORCED_DEBUG 0 140 #define FORCED_DEBUG 0
141 #endif 141 #endif
142 142
143 /* Shouldn't this be in a header file somewhere? */ 143 /* Shouldn't this be in a header file somewhere? */
144 #define BYTES_PER_WORD sizeof(void *) 144 #define BYTES_PER_WORD sizeof(void *)
145 #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) 145 #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long))
146 146
147 #ifndef ARCH_KMALLOC_MINALIGN 147 #ifndef ARCH_KMALLOC_MINALIGN
148 /* 148 /*
149 * Enforce a minimum alignment for the kmalloc caches. 149 * Enforce a minimum alignment for the kmalloc caches.
150 * Usually, the kmalloc caches are cache_line_size() aligned, except when 150 * Usually, the kmalloc caches are cache_line_size() aligned, except when
151 * DEBUG and FORCED_DEBUG are enabled, then they are BYTES_PER_WORD aligned. 151 * DEBUG and FORCED_DEBUG are enabled, then they are BYTES_PER_WORD aligned.
152 * Some archs want to perform DMA into kmalloc caches and need a guaranteed 152 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
153 * alignment larger than the alignment of a 64-bit integer. 153 * alignment larger than the alignment of a 64-bit integer.
154 * ARCH_KMALLOC_MINALIGN allows that. 154 * ARCH_KMALLOC_MINALIGN allows that.
155 * Note that increasing this value may disable some debug features. 155 * Note that increasing this value may disable some debug features.
156 */ 156 */
157 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) 157 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
158 #endif 158 #endif
159 159
160 #ifndef ARCH_SLAB_MINALIGN 160 #ifndef ARCH_SLAB_MINALIGN
161 /* 161 /*
162 * Enforce a minimum alignment for all caches. 162 * Enforce a minimum alignment for all caches.
163 * Intended for archs that get misalignment faults even for BYTES_PER_WORD 163 * Intended for archs that get misalignment faults even for BYTES_PER_WORD
164 * aligned buffers. Includes ARCH_KMALLOC_MINALIGN. 164 * aligned buffers. Includes ARCH_KMALLOC_MINALIGN.
165 * If possible: Do not enable this flag for CONFIG_DEBUG_SLAB, it disables 165 * If possible: Do not enable this flag for CONFIG_DEBUG_SLAB, it disables
166 * some debug features. 166 * some debug features.
167 */ 167 */
168 #define ARCH_SLAB_MINALIGN 0 168 #define ARCH_SLAB_MINALIGN 0
169 #endif 169 #endif
170 170
171 #ifndef ARCH_KMALLOC_FLAGS 171 #ifndef ARCH_KMALLOC_FLAGS
172 #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN 172 #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN
173 #endif 173 #endif
174 174
175 /* Legal flag mask for kmem_cache_create(). */ 175 /* Legal flag mask for kmem_cache_create(). */
176 #if DEBUG 176 #if DEBUG
177 # define CREATE_MASK (SLAB_RED_ZONE | \ 177 # define CREATE_MASK (SLAB_RED_ZONE | \
178 SLAB_POISON | SLAB_HWCACHE_ALIGN | \ 178 SLAB_POISON | SLAB_HWCACHE_ALIGN | \
179 SLAB_CACHE_DMA | \ 179 SLAB_CACHE_DMA | \
180 SLAB_STORE_USER | \ 180 SLAB_STORE_USER | \
181 SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ 181 SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \
182 SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \ 182 SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \
183 SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK) 183 SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK)
184 #else 184 #else
185 # define CREATE_MASK (SLAB_HWCACHE_ALIGN | \ 185 # define CREATE_MASK (SLAB_HWCACHE_ALIGN | \
186 SLAB_CACHE_DMA | \ 186 SLAB_CACHE_DMA | \
187 SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ 187 SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \
188 SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \ 188 SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \
189 SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK) 189 SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK)
190 #endif 190 #endif
191 191
192 /* 192 /*
193 * kmem_bufctl_t: 193 * kmem_bufctl_t:
194 * 194 *
195 * Bufctl's are used for linking objs within a slab 195 * Bufctl's are used for linking objs within a slab
196 * linked offsets. 196 * linked offsets.
197 * 197 *
198 * This implementation relies on "struct page" for locating the cache & 198 * This implementation relies on "struct page" for locating the cache &
199 * slab an object belongs to. 199 * slab an object belongs to.
200 * This allows the bufctl structure to be small (one int), but limits 200 * This allows the bufctl structure to be small (one int), but limits
201 * the number of objects a slab (not a cache) can contain when off-slab 201 * the number of objects a slab (not a cache) can contain when off-slab
202 * bufctls are used. The limit is the size of the largest general cache 202 * bufctls are used. The limit is the size of the largest general cache
203 * that does not use off-slab slabs. 203 * that does not use off-slab slabs.
204 * For 32bit archs with 4 kB pages, is this 56. 204 * For 32bit archs with 4 kB pages, is this 56.
205 * This is not serious, as it is only for large objects, when it is unwise 205 * This is not serious, as it is only for large objects, when it is unwise
206 * to have too many per slab. 206 * to have too many per slab.
207 * Note: This limit can be raised by introducing a general cache whose size 207 * Note: This limit can be raised by introducing a general cache whose size
208 * is less than 512 (PAGE_SIZE<<3), but greater than 256. 208 * is less than 512 (PAGE_SIZE<<3), but greater than 256.
209 */ 209 */
210 210
211 typedef unsigned int kmem_bufctl_t; 211 typedef unsigned int kmem_bufctl_t;
212 #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0) 212 #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0)
213 #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1) 213 #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1)
214 #define BUFCTL_ACTIVE (((kmem_bufctl_t)(~0U))-2) 214 #define BUFCTL_ACTIVE (((kmem_bufctl_t)(~0U))-2)
215 #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-3) 215 #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-3)
216 216
217 /* 217 /*
218 * struct slab 218 * struct slab
219 * 219 *
220 * Manages the objs in a slab. Placed either at the beginning of mem allocated 220 * Manages the objs in a slab. Placed either at the beginning of mem allocated
221 * for a slab, or allocated from an general cache. 221 * for a slab, or allocated from an general cache.
222 * Slabs are chained into three list: fully used, partial, fully free slabs. 222 * Slabs are chained into three list: fully used, partial, fully free slabs.
223 */ 223 */
224 struct slab { 224 struct slab {
225 struct list_head list; 225 struct list_head list;
226 unsigned long colouroff; 226 unsigned long colouroff;
227 void *s_mem; /* including colour offset */ 227 void *s_mem; /* including colour offset */
228 unsigned int inuse; /* num of objs active in slab */ 228 unsigned int inuse; /* num of objs active in slab */
229 kmem_bufctl_t free; 229 kmem_bufctl_t free;
230 unsigned short nodeid; 230 unsigned short nodeid;
231 }; 231 };
232 232
233 /* 233 /*
234 * struct slab_rcu 234 * struct slab_rcu
235 * 235 *
236 * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to 236 * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to
237 * arrange for kmem_freepages to be called via RCU. This is useful if 237 * arrange for kmem_freepages to be called via RCU. This is useful if
238 * we need to approach a kernel structure obliquely, from its address 238 * we need to approach a kernel structure obliquely, from its address
239 * obtained without the usual locking. We can lock the structure to 239 * obtained without the usual locking. We can lock the structure to
240 * stabilize it and check it's still at the given address, only if we 240 * stabilize it and check it's still at the given address, only if we
241 * can be sure that the memory has not been meanwhile reused for some 241 * can be sure that the memory has not been meanwhile reused for some
242 * other kind of object (which our subsystem's lock might corrupt). 242 * other kind of object (which our subsystem's lock might corrupt).
243 * 243 *
244 * rcu_read_lock before reading the address, then rcu_read_unlock after 244 * rcu_read_lock before reading the address, then rcu_read_unlock after
245 * taking the spinlock within the structure expected at that address. 245 * taking the spinlock within the structure expected at that address.
246 * 246 *
247 * We assume struct slab_rcu can overlay struct slab when destroying. 247 * We assume struct slab_rcu can overlay struct slab when destroying.
248 */ 248 */
249 struct slab_rcu { 249 struct slab_rcu {
250 struct rcu_head head; 250 struct rcu_head head;
251 struct kmem_cache *cachep; 251 struct kmem_cache *cachep;
252 void *addr; 252 void *addr;
253 }; 253 };
254 254
255 /* 255 /*
256 * struct array_cache 256 * struct array_cache
257 * 257 *
258 * Purpose: 258 * Purpose:
259 * - LIFO ordering, to hand out cache-warm objects from _alloc 259 * - LIFO ordering, to hand out cache-warm objects from _alloc
260 * - reduce the number of linked list operations 260 * - reduce the number of linked list operations
261 * - reduce spinlock operations 261 * - reduce spinlock operations
262 * 262 *
263 * The limit is stored in the per-cpu structure to reduce the data cache 263 * The limit is stored in the per-cpu structure to reduce the data cache
264 * footprint. 264 * footprint.
265 * 265 *
266 */ 266 */
267 struct array_cache { 267 struct array_cache {
268 unsigned int avail; 268 unsigned int avail;
269 unsigned int limit; 269 unsigned int limit;
270 unsigned int batchcount; 270 unsigned int batchcount;
271 unsigned int touched; 271 unsigned int touched;
272 spinlock_t lock; 272 spinlock_t lock;
273 void *entry[]; /* 273 void *entry[]; /*
274 * Must have this definition in here for the proper 274 * Must have this definition in here for the proper
275 * alignment of array_cache. Also simplifies accessing 275 * alignment of array_cache. Also simplifies accessing
276 * the entries. 276 * the entries.
277 */ 277 */
278 }; 278 };
279 279
280 /* 280 /*
281 * bootstrap: The caches do not work without cpuarrays anymore, but the 281 * bootstrap: The caches do not work without cpuarrays anymore, but the
282 * cpuarrays are allocated from the generic caches... 282 * cpuarrays are allocated from the generic caches...
283 */ 283 */
284 #define BOOT_CPUCACHE_ENTRIES 1 284 #define BOOT_CPUCACHE_ENTRIES 1
285 struct arraycache_init { 285 struct arraycache_init {
286 struct array_cache cache; 286 struct array_cache cache;
287 void *entries[BOOT_CPUCACHE_ENTRIES]; 287 void *entries[BOOT_CPUCACHE_ENTRIES];
288 }; 288 };
289 289
290 /* 290 /*
291 * The slab lists for all objects. 291 * The slab lists for all objects.
292 */ 292 */
293 struct kmem_list3 { 293 struct kmem_list3 {
294 struct list_head slabs_partial; /* partial list first, better asm code */ 294 struct list_head slabs_partial; /* partial list first, better asm code */
295 struct list_head slabs_full; 295 struct list_head slabs_full;
296 struct list_head slabs_free; 296 struct list_head slabs_free;
297 unsigned long free_objects; 297 unsigned long free_objects;
298 unsigned int free_limit; 298 unsigned int free_limit;
299 unsigned int colour_next; /* Per-node cache coloring */ 299 unsigned int colour_next; /* Per-node cache coloring */
300 spinlock_t list_lock; 300 spinlock_t list_lock;
301 struct array_cache *shared; /* shared per node */ 301 struct array_cache *shared; /* shared per node */
302 struct array_cache **alien; /* on other nodes */ 302 struct array_cache **alien; /* on other nodes */
303 unsigned long next_reap; /* updated without locking */ 303 unsigned long next_reap; /* updated without locking */
304 int free_touched; /* updated without locking */ 304 int free_touched; /* updated without locking */
305 }; 305 };
306 306
307 /* 307 /*
308 * Need this for bootstrapping a per node allocator. 308 * Need this for bootstrapping a per node allocator.
309 */ 309 */
310 #define NUM_INIT_LISTS (3 * MAX_NUMNODES) 310 #define NUM_INIT_LISTS (3 * MAX_NUMNODES)
311 struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS]; 311 struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS];
312 #define CACHE_CACHE 0 312 #define CACHE_CACHE 0
313 #define SIZE_AC MAX_NUMNODES 313 #define SIZE_AC MAX_NUMNODES
314 #define SIZE_L3 (2 * MAX_NUMNODES) 314 #define SIZE_L3 (2 * MAX_NUMNODES)
315 315
316 static int drain_freelist(struct kmem_cache *cache, 316 static int drain_freelist(struct kmem_cache *cache,
317 struct kmem_list3 *l3, int tofree); 317 struct kmem_list3 *l3, int tofree);
318 static void free_block(struct kmem_cache *cachep, void **objpp, int len, 318 static void free_block(struct kmem_cache *cachep, void **objpp, int len,
319 int node); 319 int node);
320 static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); 320 static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
321 static void cache_reap(struct work_struct *unused); 321 static void cache_reap(struct work_struct *unused);
322 322
323 /* 323 /*
324 * This function must be completely optimized away if a constant is passed to 324 * This function must be completely optimized away if a constant is passed to
325 * it. Mostly the same as what is in linux/slab.h except it returns an index. 325 * it. Mostly the same as what is in linux/slab.h except it returns an index.
326 */ 326 */
327 static __always_inline int index_of(const size_t size) 327 static __always_inline int index_of(const size_t size)
328 { 328 {
329 extern void __bad_size(void); 329 extern void __bad_size(void);
330 330
331 if (__builtin_constant_p(size)) { 331 if (__builtin_constant_p(size)) {
332 int i = 0; 332 int i = 0;
333 333
334 #define CACHE(x) \ 334 #define CACHE(x) \
335 if (size <=x) \ 335 if (size <=x) \
336 return i; \ 336 return i; \
337 else \ 337 else \
338 i++; 338 i++;
339 #include <linux/kmalloc_sizes.h> 339 #include <linux/kmalloc_sizes.h>
340 #undef CACHE 340 #undef CACHE
341 __bad_size(); 341 __bad_size();
342 } else 342 } else
343 __bad_size(); 343 __bad_size();
344 return 0; 344 return 0;
345 } 345 }
346 346
347 static int slab_early_init = 1; 347 static int slab_early_init = 1;
348 348
349 #define INDEX_AC index_of(sizeof(struct arraycache_init)) 349 #define INDEX_AC index_of(sizeof(struct arraycache_init))
350 #define INDEX_L3 index_of(sizeof(struct kmem_list3)) 350 #define INDEX_L3 index_of(sizeof(struct kmem_list3))
351 351
352 static void kmem_list3_init(struct kmem_list3 *parent) 352 static void kmem_list3_init(struct kmem_list3 *parent)
353 { 353 {
354 INIT_LIST_HEAD(&parent->slabs_full); 354 INIT_LIST_HEAD(&parent->slabs_full);
355 INIT_LIST_HEAD(&parent->slabs_partial); 355 INIT_LIST_HEAD(&parent->slabs_partial);
356 INIT_LIST_HEAD(&parent->slabs_free); 356 INIT_LIST_HEAD(&parent->slabs_free);
357 parent->shared = NULL; 357 parent->shared = NULL;
358 parent->alien = NULL; 358 parent->alien = NULL;
359 parent->colour_next = 0; 359 parent->colour_next = 0;
360 spin_lock_init(&parent->list_lock); 360 spin_lock_init(&parent->list_lock);
361 parent->free_objects = 0; 361 parent->free_objects = 0;
362 parent->free_touched = 0; 362 parent->free_touched = 0;
363 } 363 }
364 364
365 #define MAKE_LIST(cachep, listp, slab, nodeid) \ 365 #define MAKE_LIST(cachep, listp, slab, nodeid) \
366 do { \ 366 do { \
367 INIT_LIST_HEAD(listp); \ 367 INIT_LIST_HEAD(listp); \
368 list_splice(&(cachep->nodelists[nodeid]->slab), listp); \ 368 list_splice(&(cachep->nodelists[nodeid]->slab), listp); \
369 } while (0) 369 } while (0)
370 370
371 #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ 371 #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \
372 do { \ 372 do { \
373 MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ 373 MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \
374 MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ 374 MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \
375 MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ 375 MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \
376 } while (0) 376 } while (0)
377 377
378 #define CFLGS_OFF_SLAB (0x80000000UL) 378 #define CFLGS_OFF_SLAB (0x80000000UL)
379 #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) 379 #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB)
380 380
381 #define BATCHREFILL_LIMIT 16 381 #define BATCHREFILL_LIMIT 16
382 /* 382 /*
383 * Optimization question: fewer reaps means less probability for unnessary 383 * Optimization question: fewer reaps means less probability for unnessary
384 * cpucache drain/refill cycles. 384 * cpucache drain/refill cycles.
385 * 385 *
386 * OTOH the cpuarrays can contain lots of objects, 386 * OTOH the cpuarrays can contain lots of objects,
387 * which could lock up otherwise freeable slabs. 387 * which could lock up otherwise freeable slabs.
388 */ 388 */
389 #define REAPTIMEOUT_CPUC (2*HZ) 389 #define REAPTIMEOUT_CPUC (2*HZ)
390 #define REAPTIMEOUT_LIST3 (4*HZ) 390 #define REAPTIMEOUT_LIST3 (4*HZ)
391 391
392 #if STATS 392 #if STATS
393 #define STATS_INC_ACTIVE(x) ((x)->num_active++) 393 #define STATS_INC_ACTIVE(x) ((x)->num_active++)
394 #define STATS_DEC_ACTIVE(x) ((x)->num_active--) 394 #define STATS_DEC_ACTIVE(x) ((x)->num_active--)
395 #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) 395 #define STATS_INC_ALLOCED(x) ((x)->num_allocations++)
396 #define STATS_INC_GROWN(x) ((x)->grown++) 396 #define STATS_INC_GROWN(x) ((x)->grown++)
397 #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) 397 #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y))
398 #define STATS_SET_HIGH(x) \ 398 #define STATS_SET_HIGH(x) \
399 do { \ 399 do { \
400 if ((x)->num_active > (x)->high_mark) \ 400 if ((x)->num_active > (x)->high_mark) \
401 (x)->high_mark = (x)->num_active; \ 401 (x)->high_mark = (x)->num_active; \
402 } while (0) 402 } while (0)
403 #define STATS_INC_ERR(x) ((x)->errors++) 403 #define STATS_INC_ERR(x) ((x)->errors++)
404 #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) 404 #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++)
405 #define STATS_INC_NODEFREES(x) ((x)->node_frees++) 405 #define STATS_INC_NODEFREES(x) ((x)->node_frees++)
406 #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) 406 #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++)
407 #define STATS_SET_FREEABLE(x, i) \ 407 #define STATS_SET_FREEABLE(x, i) \
408 do { \ 408 do { \
409 if ((x)->max_freeable < i) \ 409 if ((x)->max_freeable < i) \
410 (x)->max_freeable = i; \ 410 (x)->max_freeable = i; \
411 } while (0) 411 } while (0)
412 #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) 412 #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit)
413 #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) 413 #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss)
414 #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) 414 #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit)
415 #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) 415 #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss)
416 #else 416 #else
417 #define STATS_INC_ACTIVE(x) do { } while (0) 417 #define STATS_INC_ACTIVE(x) do { } while (0)
418 #define STATS_DEC_ACTIVE(x) do { } while (0) 418 #define STATS_DEC_ACTIVE(x) do { } while (0)
419 #define STATS_INC_ALLOCED(x) do { } while (0) 419 #define STATS_INC_ALLOCED(x) do { } while (0)
420 #define STATS_INC_GROWN(x) do { } while (0) 420 #define STATS_INC_GROWN(x) do { } while (0)
421 #define STATS_ADD_REAPED(x,y) do { } while (0) 421 #define STATS_ADD_REAPED(x,y) do { } while (0)
422 #define STATS_SET_HIGH(x) do { } while (0) 422 #define STATS_SET_HIGH(x) do { } while (0)
423 #define STATS_INC_ERR(x) do { } while (0) 423 #define STATS_INC_ERR(x) do { } while (0)
424 #define STATS_INC_NODEALLOCS(x) do { } while (0) 424 #define STATS_INC_NODEALLOCS(x) do { } while (0)
425 #define STATS_INC_NODEFREES(x) do { } while (0) 425 #define STATS_INC_NODEFREES(x) do { } while (0)
426 #define STATS_INC_ACOVERFLOW(x) do { } while (0) 426 #define STATS_INC_ACOVERFLOW(x) do { } while (0)
427 #define STATS_SET_FREEABLE(x, i) do { } while (0) 427 #define STATS_SET_FREEABLE(x, i) do { } while (0)
428 #define STATS_INC_ALLOCHIT(x) do { } while (0) 428 #define STATS_INC_ALLOCHIT(x) do { } while (0)
429 #define STATS_INC_ALLOCMISS(x) do { } while (0) 429 #define STATS_INC_ALLOCMISS(x) do { } while (0)
430 #define STATS_INC_FREEHIT(x) do { } while (0) 430 #define STATS_INC_FREEHIT(x) do { } while (0)
431 #define STATS_INC_FREEMISS(x) do { } while (0) 431 #define STATS_INC_FREEMISS(x) do { } while (0)
432 #endif 432 #endif
433 433
434 #if DEBUG 434 #if DEBUG
435 435
436 /* 436 /*
437 * memory layout of objects: 437 * memory layout of objects:
438 * 0 : objp 438 * 0 : objp
439 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that 439 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
440 * the end of an object is aligned with the end of the real 440 * the end of an object is aligned with the end of the real
441 * allocation. Catches writes behind the end of the allocation. 441 * allocation. Catches writes behind the end of the allocation.
442 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: 442 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
443 * redzone word. 443 * redzone word.
444 * cachep->obj_offset: The real object. 444 * cachep->obj_offset: The real object.
445 * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] 445 * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long]
446 * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address 446 * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address
447 * [BYTES_PER_WORD long] 447 * [BYTES_PER_WORD long]
448 */ 448 */
449 static int obj_offset(struct kmem_cache *cachep) 449 static int obj_offset(struct kmem_cache *cachep)
450 { 450 {
451 return cachep->obj_offset; 451 return cachep->obj_offset;
452 } 452 }
453 453
454 static int obj_size(struct kmem_cache *cachep) 454 static int obj_size(struct kmem_cache *cachep)
455 { 455 {
456 return cachep->obj_size; 456 return cachep->obj_size;
457 } 457 }
458 458
459 static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) 459 static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
460 { 460 {
461 BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); 461 BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
462 return (unsigned long long*) (objp + obj_offset(cachep) - 462 return (unsigned long long*) (objp + obj_offset(cachep) -
463 sizeof(unsigned long long)); 463 sizeof(unsigned long long));
464 } 464 }
465 465
466 static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) 466 static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
467 { 467 {
468 BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); 468 BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
469 if (cachep->flags & SLAB_STORE_USER) 469 if (cachep->flags & SLAB_STORE_USER)
470 return (unsigned long long *)(objp + cachep->buffer_size - 470 return (unsigned long long *)(objp + cachep->buffer_size -
471 sizeof(unsigned long long) - 471 sizeof(unsigned long long) -
472 REDZONE_ALIGN); 472 REDZONE_ALIGN);
473 return (unsigned long long *) (objp + cachep->buffer_size - 473 return (unsigned long long *) (objp + cachep->buffer_size -
474 sizeof(unsigned long long)); 474 sizeof(unsigned long long));
475 } 475 }
476 476
477 static void **dbg_userword(struct kmem_cache *cachep, void *objp) 477 static void **dbg_userword(struct kmem_cache *cachep, void *objp)
478 { 478 {
479 BUG_ON(!(cachep->flags & SLAB_STORE_USER)); 479 BUG_ON(!(cachep->flags & SLAB_STORE_USER));
480 return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD); 480 return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD);
481 } 481 }
482 482
483 #else 483 #else
484 484
485 #define obj_offset(x) 0 485 #define obj_offset(x) 0
486 #define obj_size(cachep) (cachep->buffer_size) 486 #define obj_size(cachep) (cachep->buffer_size)
487 #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) 487 #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;})
488 #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) 488 #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;})
489 #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) 489 #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;})
490 490
491 #endif 491 #endif
492 492
493 #ifdef CONFIG_TRACING 493 #ifdef CONFIG_TRACING
494 size_t slab_buffer_size(struct kmem_cache *cachep) 494 size_t slab_buffer_size(struct kmem_cache *cachep)
495 { 495 {
496 return cachep->buffer_size; 496 return cachep->buffer_size;
497 } 497 }
498 EXPORT_SYMBOL(slab_buffer_size); 498 EXPORT_SYMBOL(slab_buffer_size);
499 #endif 499 #endif
500 500
501 /* 501 /*
502 * Do not go above this order unless 0 objects fit into the slab. 502 * Do not go above this order unless 0 objects fit into the slab.
503 */ 503 */
504 #define BREAK_GFP_ORDER_HI 1 504 #define BREAK_GFP_ORDER_HI 1
505 #define BREAK_GFP_ORDER_LO 0 505 #define BREAK_GFP_ORDER_LO 0
506 static int slab_break_gfp_order = BREAK_GFP_ORDER_LO; 506 static int slab_break_gfp_order = BREAK_GFP_ORDER_LO;
507 507
508 /* 508 /*
509 * Functions for storing/retrieving the cachep and or slab from the page 509 * Functions for storing/retrieving the cachep and or slab from the page
510 * allocator. These are used to find the slab an obj belongs to. With kfree(), 510 * allocator. These are used to find the slab an obj belongs to. With kfree(),
511 * these are used to find the cache which an obj belongs to. 511 * these are used to find the cache which an obj belongs to.
512 */ 512 */
513 static inline void page_set_cache(struct page *page, struct kmem_cache *cache) 513 static inline void page_set_cache(struct page *page, struct kmem_cache *cache)
514 { 514 {
515 page->lru.next = (struct list_head *)cache; 515 page->lru.next = (struct list_head *)cache;
516 } 516 }
517 517
518 static inline struct kmem_cache *page_get_cache(struct page *page) 518 static inline struct kmem_cache *page_get_cache(struct page *page)
519 { 519 {
520 page = compound_head(page); 520 page = compound_head(page);
521 BUG_ON(!PageSlab(page)); 521 BUG_ON(!PageSlab(page));
522 return (struct kmem_cache *)page->lru.next; 522 return (struct kmem_cache *)page->lru.next;
523 } 523 }
524 524
525 static inline void page_set_slab(struct page *page, struct slab *slab) 525 static inline void page_set_slab(struct page *page, struct slab *slab)
526 { 526 {
527 page->lru.prev = (struct list_head *)slab; 527 page->lru.prev = (struct list_head *)slab;
528 } 528 }
529 529
530 static inline struct slab *page_get_slab(struct page *page) 530 static inline struct slab *page_get_slab(struct page *page)
531 { 531 {
532 BUG_ON(!PageSlab(page)); 532 BUG_ON(!PageSlab(page));
533 return (struct slab *)page->lru.prev; 533 return (struct slab *)page->lru.prev;
534 } 534 }
535 535
536 static inline struct kmem_cache *virt_to_cache(const void *obj) 536 static inline struct kmem_cache *virt_to_cache(const void *obj)
537 { 537 {
538 struct page *page = virt_to_head_page(obj); 538 struct page *page = virt_to_head_page(obj);
539 return page_get_cache(page); 539 return page_get_cache(page);
540 } 540 }
541 541
542 static inline struct slab *virt_to_slab(const void *obj) 542 static inline struct slab *virt_to_slab(const void *obj)
543 { 543 {
544 struct page *page = virt_to_head_page(obj); 544 struct page *page = virt_to_head_page(obj);
545 return page_get_slab(page); 545 return page_get_slab(page);
546 } 546 }
547 547
548 static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab, 548 static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab,
549 unsigned int idx) 549 unsigned int idx)
550 { 550 {
551 return slab->s_mem + cache->buffer_size * idx; 551 return slab->s_mem + cache->buffer_size * idx;
552 } 552 }
553 553
554 /* 554 /*
555 * We want to avoid an expensive divide : (offset / cache->buffer_size) 555 * We want to avoid an expensive divide : (offset / cache->buffer_size)
556 * Using the fact that buffer_size is a constant for a particular cache, 556 * Using the fact that buffer_size is a constant for a particular cache,
557 * we can replace (offset / cache->buffer_size) by 557 * we can replace (offset / cache->buffer_size) by
558 * reciprocal_divide(offset, cache->reciprocal_buffer_size) 558 * reciprocal_divide(offset, cache->reciprocal_buffer_size)
559 */ 559 */
560 static inline unsigned int obj_to_index(const struct kmem_cache *cache, 560 static inline unsigned int obj_to_index(const struct kmem_cache *cache,
561 const struct slab *slab, void *obj) 561 const struct slab *slab, void *obj)
562 { 562 {
563 u32 offset = (obj - slab->s_mem); 563 u32 offset = (obj - slab->s_mem);
564 return reciprocal_divide(offset, cache->reciprocal_buffer_size); 564 return reciprocal_divide(offset, cache->reciprocal_buffer_size);
565 } 565 }
566 566
567 /* 567 /*
568 * These are the default caches for kmalloc. Custom caches can have other sizes. 568 * These are the default caches for kmalloc. Custom caches can have other sizes.
569 */ 569 */
570 struct cache_sizes malloc_sizes[] = { 570 struct cache_sizes malloc_sizes[] = {
571 #define CACHE(x) { .cs_size = (x) }, 571 #define CACHE(x) { .cs_size = (x) },
572 #include <linux/kmalloc_sizes.h> 572 #include <linux/kmalloc_sizes.h>
573 CACHE(ULONG_MAX) 573 CACHE(ULONG_MAX)
574 #undef CACHE 574 #undef CACHE
575 }; 575 };
576 EXPORT_SYMBOL(malloc_sizes); 576 EXPORT_SYMBOL(malloc_sizes);
577 577
578 /* Must match cache_sizes above. Out of line to keep cache footprint low. */ 578 /* Must match cache_sizes above. Out of line to keep cache footprint low. */
579 struct cache_names { 579 struct cache_names {
580 char *name; 580 char *name;
581 char *name_dma; 581 char *name_dma;
582 }; 582 };
583 583
584 static struct cache_names __initdata cache_names[] = { 584 static struct cache_names __initdata cache_names[] = {
585 #define CACHE(x) { .name = "size-" #x, .name_dma = "size-" #x "(DMA)" }, 585 #define CACHE(x) { .name = "size-" #x, .name_dma = "size-" #x "(DMA)" },
586 #include <linux/kmalloc_sizes.h> 586 #include <linux/kmalloc_sizes.h>
587 {NULL,} 587 {NULL,}
588 #undef CACHE 588 #undef CACHE
589 }; 589 };
590 590
591 static struct arraycache_init initarray_cache __initdata = 591 static struct arraycache_init initarray_cache __initdata =
592 { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; 592 { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} };
593 static struct arraycache_init initarray_generic = 593 static struct arraycache_init initarray_generic =
594 { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; 594 { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} };
595 595
596 /* internal cache of cache description objs */ 596 /* internal cache of cache description objs */
597 static struct kmem_cache cache_cache = { 597 static struct kmem_cache cache_cache = {
598 .batchcount = 1, 598 .batchcount = 1,
599 .limit = BOOT_CPUCACHE_ENTRIES, 599 .limit = BOOT_CPUCACHE_ENTRIES,
600 .shared = 1, 600 .shared = 1,
601 .buffer_size = sizeof(struct kmem_cache), 601 .buffer_size = sizeof(struct kmem_cache),
602 .name = "kmem_cache", 602 .name = "kmem_cache",
603 }; 603 };
604 604
605 #define BAD_ALIEN_MAGIC 0x01020304ul 605 #define BAD_ALIEN_MAGIC 0x01020304ul
606 606
607 /* 607 /*
608 * chicken and egg problem: delay the per-cpu array allocation 608 * chicken and egg problem: delay the per-cpu array allocation
609 * until the general caches are up. 609 * until the general caches are up.
610 */ 610 */
611 static enum { 611 static enum {
612 NONE, 612 NONE,
613 PARTIAL_AC, 613 PARTIAL_AC,
614 PARTIAL_L3, 614 PARTIAL_L3,
615 EARLY, 615 EARLY,
616 FULL 616 FULL
617 } g_cpucache_up; 617 } g_cpucache_up;
618 618
619 /* 619 /*
620 * used by boot code to determine if it can use slab based allocator 620 * used by boot code to determine if it can use slab based allocator
621 */ 621 */
622 int slab_is_available(void) 622 int slab_is_available(void)
623 { 623 {
624 return g_cpucache_up >= EARLY; 624 return g_cpucache_up >= EARLY;
625 } 625 }
626 626
627 #ifdef CONFIG_LOCKDEP 627 #ifdef CONFIG_LOCKDEP
628 628
629 /* 629 /*
630 * Slab sometimes uses the kmalloc slabs to store the slab headers 630 * Slab sometimes uses the kmalloc slabs to store the slab headers
631 * for other slabs "off slab". 631 * for other slabs "off slab".
632 * The locking for this is tricky in that it nests within the locks 632 * The locking for this is tricky in that it nests within the locks
633 * of all other slabs in a few places; to deal with this special 633 * of all other slabs in a few places; to deal with this special
634 * locking we put on-slab caches into a separate lock-class. 634 * locking we put on-slab caches into a separate lock-class.
635 * 635 *
636 * We set lock class for alien array caches which are up during init. 636 * We set lock class for alien array caches which are up during init.
637 * The lock annotation will be lost if all cpus of a node goes down and 637 * The lock annotation will be lost if all cpus of a node goes down and
638 * then comes back up during hotplug 638 * then comes back up during hotplug
639 */ 639 */
640 static struct lock_class_key on_slab_l3_key; 640 static struct lock_class_key on_slab_l3_key;
641 static struct lock_class_key on_slab_alc_key; 641 static struct lock_class_key on_slab_alc_key;
642 642
643 static void init_node_lock_keys(int q) 643 static void init_node_lock_keys(int q)
644 { 644 {
645 struct cache_sizes *s = malloc_sizes; 645 struct cache_sizes *s = malloc_sizes;
646 646
647 if (g_cpucache_up != FULL) 647 if (g_cpucache_up != FULL)
648 return; 648 return;
649 649
650 for (s = malloc_sizes; s->cs_size != ULONG_MAX; s++) { 650 for (s = malloc_sizes; s->cs_size != ULONG_MAX; s++) {
651 struct array_cache **alc; 651 struct array_cache **alc;
652 struct kmem_list3 *l3; 652 struct kmem_list3 *l3;
653 int r; 653 int r;
654 654
655 l3 = s->cs_cachep->nodelists[q]; 655 l3 = s->cs_cachep->nodelists[q];
656 if (!l3 || OFF_SLAB(s->cs_cachep)) 656 if (!l3 || OFF_SLAB(s->cs_cachep))
657 return; 657 continue;
658 lockdep_set_class(&l3->list_lock, &on_slab_l3_key); 658 lockdep_set_class(&l3->list_lock, &on_slab_l3_key);
659 alc = l3->alien; 659 alc = l3->alien;
660 /* 660 /*
661 * FIXME: This check for BAD_ALIEN_MAGIC 661 * FIXME: This check for BAD_ALIEN_MAGIC
662 * should go away when common slab code is taught to 662 * should go away when common slab code is taught to
663 * work even without alien caches. 663 * work even without alien caches.
664 * Currently, non NUMA code returns BAD_ALIEN_MAGIC 664 * Currently, non NUMA code returns BAD_ALIEN_MAGIC
665 * for alloc_alien_cache, 665 * for alloc_alien_cache,
666 */ 666 */
667 if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) 667 if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC)
668 return; 668 continue;
669 for_each_node(r) { 669 for_each_node(r) {
670 if (alc[r]) 670 if (alc[r])
671 lockdep_set_class(&alc[r]->lock, 671 lockdep_set_class(&alc[r]->lock,
672 &on_slab_alc_key); 672 &on_slab_alc_key);
673 } 673 }
674 } 674 }
675 } 675 }
676 676
677 static inline void init_lock_keys(void) 677 static inline void init_lock_keys(void)
678 { 678 {
679 int node; 679 int node;
680 680
681 for_each_node(node) 681 for_each_node(node)
682 init_node_lock_keys(node); 682 init_node_lock_keys(node);
683 } 683 }
684 #else 684 #else
685 static void init_node_lock_keys(int q) 685 static void init_node_lock_keys(int q)
686 { 686 {
687 } 687 }
688 688
689 static inline void init_lock_keys(void) 689 static inline void init_lock_keys(void)
690 { 690 {
691 } 691 }
692 #endif 692 #endif
693 693
694 /* 694 /*
695 * Guard access to the cache-chain. 695 * Guard access to the cache-chain.
696 */ 696 */
697 static DEFINE_MUTEX(cache_chain_mutex); 697 static DEFINE_MUTEX(cache_chain_mutex);
698 static struct list_head cache_chain; 698 static struct list_head cache_chain;
699 699
700 static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); 700 static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
701 701
702 static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) 702 static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
703 { 703 {
704 return cachep->array[smp_processor_id()]; 704 return cachep->array[smp_processor_id()];
705 } 705 }
706 706
707 static inline struct kmem_cache *__find_general_cachep(size_t size, 707 static inline struct kmem_cache *__find_general_cachep(size_t size,
708 gfp_t gfpflags) 708 gfp_t gfpflags)
709 { 709 {
710 struct cache_sizes *csizep = malloc_sizes; 710 struct cache_sizes *csizep = malloc_sizes;
711 711
712 #if DEBUG 712 #if DEBUG
713 /* This happens if someone tries to call 713 /* This happens if someone tries to call
714 * kmem_cache_create(), or __kmalloc(), before 714 * kmem_cache_create(), or __kmalloc(), before
715 * the generic caches are initialized. 715 * the generic caches are initialized.
716 */ 716 */
717 BUG_ON(malloc_sizes[INDEX_AC].cs_cachep == NULL); 717 BUG_ON(malloc_sizes[INDEX_AC].cs_cachep == NULL);
718 #endif 718 #endif
719 if (!size) 719 if (!size)
720 return ZERO_SIZE_PTR; 720 return ZERO_SIZE_PTR;
721 721
722 while (size > csizep->cs_size) 722 while (size > csizep->cs_size)
723 csizep++; 723 csizep++;
724 724
725 /* 725 /*
726 * Really subtle: The last entry with cs->cs_size==ULONG_MAX 726 * Really subtle: The last entry with cs->cs_size==ULONG_MAX
727 * has cs_{dma,}cachep==NULL. Thus no special case 727 * has cs_{dma,}cachep==NULL. Thus no special case
728 * for large kmalloc calls required. 728 * for large kmalloc calls required.
729 */ 729 */
730 #ifdef CONFIG_ZONE_DMA 730 #ifdef CONFIG_ZONE_DMA
731 if (unlikely(gfpflags & GFP_DMA)) 731 if (unlikely(gfpflags & GFP_DMA))
732 return csizep->cs_dmacachep; 732 return csizep->cs_dmacachep;
733 #endif 733 #endif
734 return csizep->cs_cachep; 734 return csizep->cs_cachep;
735 } 735 }
736 736
737 static struct kmem_cache *kmem_find_general_cachep(size_t size, gfp_t gfpflags) 737 static struct kmem_cache *kmem_find_general_cachep(size_t size, gfp_t gfpflags)
738 { 738 {
739 return __find_general_cachep(size, gfpflags); 739 return __find_general_cachep(size, gfpflags);
740 } 740 }
741 741
742 static size_t slab_mgmt_size(size_t nr_objs, size_t align) 742 static size_t slab_mgmt_size(size_t nr_objs, size_t align)
743 { 743 {
744 return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); 744 return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align);
745 } 745 }
746 746
747 /* 747 /*
748 * Calculate the number of objects and left-over bytes for a given buffer size. 748 * Calculate the number of objects and left-over bytes for a given buffer size.
749 */ 749 */
750 static void cache_estimate(unsigned long gfporder, size_t buffer_size, 750 static void cache_estimate(unsigned long gfporder, size_t buffer_size,
751 size_t align, int flags, size_t *left_over, 751 size_t align, int flags, size_t *left_over,
752 unsigned int *num) 752 unsigned int *num)
753 { 753 {
754 int nr_objs; 754 int nr_objs;
755 size_t mgmt_size; 755 size_t mgmt_size;
756 size_t slab_size = PAGE_SIZE << gfporder; 756 size_t slab_size = PAGE_SIZE << gfporder;
757 757
758 /* 758 /*
759 * The slab management structure can be either off the slab or 759 * The slab management structure can be either off the slab or
760 * on it. For the latter case, the memory allocated for a 760 * on it. For the latter case, the memory allocated for a
761 * slab is used for: 761 * slab is used for:
762 * 762 *
763 * - The struct slab 763 * - The struct slab
764 * - One kmem_bufctl_t for each object 764 * - One kmem_bufctl_t for each object
765 * - Padding to respect alignment of @align 765 * - Padding to respect alignment of @align
766 * - @buffer_size bytes for each object 766 * - @buffer_size bytes for each object
767 * 767 *
768 * If the slab management structure is off the slab, then the 768 * If the slab management structure is off the slab, then the
769 * alignment will already be calculated into the size. Because 769 * alignment will already be calculated into the size. Because
770 * the slabs are all pages aligned, the objects will be at the 770 * the slabs are all pages aligned, the objects will be at the
771 * correct alignment when allocated. 771 * correct alignment when allocated.
772 */ 772 */
773 if (flags & CFLGS_OFF_SLAB) { 773 if (flags & CFLGS_OFF_SLAB) {
774 mgmt_size = 0; 774 mgmt_size = 0;
775 nr_objs = slab_size / buffer_size; 775 nr_objs = slab_size / buffer_size;
776 776
777 if (nr_objs > SLAB_LIMIT) 777 if (nr_objs > SLAB_LIMIT)
778 nr_objs = SLAB_LIMIT; 778 nr_objs = SLAB_LIMIT;
779 } else { 779 } else {
780 /* 780 /*
781 * Ignore padding for the initial guess. The padding 781 * Ignore padding for the initial guess. The padding
782 * is at most @align-1 bytes, and @buffer_size is at 782 * is at most @align-1 bytes, and @buffer_size is at
783 * least @align. In the worst case, this result will 783 * least @align. In the worst case, this result will
784 * be one greater than the number of objects that fit 784 * be one greater than the number of objects that fit
785 * into the memory allocation when taking the padding 785 * into the memory allocation when taking the padding
786 * into account. 786 * into account.
787 */ 787 */
788 nr_objs = (slab_size - sizeof(struct slab)) / 788 nr_objs = (slab_size - sizeof(struct slab)) /
789 (buffer_size + sizeof(kmem_bufctl_t)); 789 (buffer_size + sizeof(kmem_bufctl_t));
790 790
791 /* 791 /*
792 * This calculated number will be either the right 792 * This calculated number will be either the right
793 * amount, or one greater than what we want. 793 * amount, or one greater than what we want.
794 */ 794 */
795 if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size 795 if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size
796 > slab_size) 796 > slab_size)
797 nr_objs--; 797 nr_objs--;
798 798
799 if (nr_objs > SLAB_LIMIT) 799 if (nr_objs > SLAB_LIMIT)
800 nr_objs = SLAB_LIMIT; 800 nr_objs = SLAB_LIMIT;
801 801
802 mgmt_size = slab_mgmt_size(nr_objs, align); 802 mgmt_size = slab_mgmt_size(nr_objs, align);
803 } 803 }
804 *num = nr_objs; 804 *num = nr_objs;
805 *left_over = slab_size - nr_objs*buffer_size - mgmt_size; 805 *left_over = slab_size - nr_objs*buffer_size - mgmt_size;
806 } 806 }
807 807
808 #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) 808 #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
809 809
810 static void __slab_error(const char *function, struct kmem_cache *cachep, 810 static void __slab_error(const char *function, struct kmem_cache *cachep,
811 char *msg) 811 char *msg)
812 { 812 {
813 printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", 813 printk(KERN_ERR "slab error in %s(): cache `%s': %s\n",
814 function, cachep->name, msg); 814 function, cachep->name, msg);
815 dump_stack(); 815 dump_stack();
816 } 816 }
817 817
818 /* 818 /*
819 * By default on NUMA we use alien caches to stage the freeing of 819 * By default on NUMA we use alien caches to stage the freeing of
820 * objects allocated from other nodes. This causes massive memory 820 * objects allocated from other nodes. This causes massive memory
821 * inefficiencies when using fake NUMA setup to split memory into a 821 * inefficiencies when using fake NUMA setup to split memory into a
822 * large number of small nodes, so it can be disabled on the command 822 * large number of small nodes, so it can be disabled on the command
823 * line 823 * line
824 */ 824 */
825 825
826 static int use_alien_caches __read_mostly = 1; 826 static int use_alien_caches __read_mostly = 1;
827 static int __init noaliencache_setup(char *s) 827 static int __init noaliencache_setup(char *s)
828 { 828 {
829 use_alien_caches = 0; 829 use_alien_caches = 0;
830 return 1; 830 return 1;
831 } 831 }
832 __setup("noaliencache", noaliencache_setup); 832 __setup("noaliencache", noaliencache_setup);
833 833
834 #ifdef CONFIG_NUMA 834 #ifdef CONFIG_NUMA
835 /* 835 /*
836 * Special reaping functions for NUMA systems called from cache_reap(). 836 * Special reaping functions for NUMA systems called from cache_reap().
837 * These take care of doing round robin flushing of alien caches (containing 837 * These take care of doing round robin flushing of alien caches (containing
838 * objects freed on different nodes from which they were allocated) and the 838 * objects freed on different nodes from which they were allocated) and the
839 * flushing of remote pcps by calling drain_node_pages. 839 * flushing of remote pcps by calling drain_node_pages.
840 */ 840 */
841 static DEFINE_PER_CPU(unsigned long, slab_reap_node); 841 static DEFINE_PER_CPU(unsigned long, slab_reap_node);
842 842
843 static void init_reap_node(int cpu) 843 static void init_reap_node(int cpu)
844 { 844 {
845 int node; 845 int node;
846 846
847 node = next_node(cpu_to_node(cpu), node_online_map); 847 node = next_node(cpu_to_node(cpu), node_online_map);
848 if (node == MAX_NUMNODES) 848 if (node == MAX_NUMNODES)
849 node = first_node(node_online_map); 849 node = first_node(node_online_map);
850 850
851 per_cpu(slab_reap_node, cpu) = node; 851 per_cpu(slab_reap_node, cpu) = node;
852 } 852 }
853 853
854 static void next_reap_node(void) 854 static void next_reap_node(void)
855 { 855 {
856 int node = __get_cpu_var(slab_reap_node); 856 int node = __get_cpu_var(slab_reap_node);
857 857
858 node = next_node(node, node_online_map); 858 node = next_node(node, node_online_map);
859 if (unlikely(node >= MAX_NUMNODES)) 859 if (unlikely(node >= MAX_NUMNODES))
860 node = first_node(node_online_map); 860 node = first_node(node_online_map);
861 __get_cpu_var(slab_reap_node) = node; 861 __get_cpu_var(slab_reap_node) = node;
862 } 862 }
863 863
864 #else 864 #else
865 #define init_reap_node(cpu) do { } while (0) 865 #define init_reap_node(cpu) do { } while (0)
866 #define next_reap_node(void) do { } while (0) 866 #define next_reap_node(void) do { } while (0)
867 #endif 867 #endif
868 868
869 /* 869 /*
870 * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz 870 * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz
871 * via the workqueue/eventd. 871 * via the workqueue/eventd.
872 * Add the CPU number into the expiration time to minimize the possibility of 872 * Add the CPU number into the expiration time to minimize the possibility of
873 * the CPUs getting into lockstep and contending for the global cache chain 873 * the CPUs getting into lockstep and contending for the global cache chain
874 * lock. 874 * lock.
875 */ 875 */
876 static void __cpuinit start_cpu_timer(int cpu) 876 static void __cpuinit start_cpu_timer(int cpu)
877 { 877 {
878 struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); 878 struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
879 879
880 /* 880 /*
881 * When this gets called from do_initcalls via cpucache_init(), 881 * When this gets called from do_initcalls via cpucache_init(),
882 * init_workqueues() has already run, so keventd will be setup 882 * init_workqueues() has already run, so keventd will be setup
883 * at that time. 883 * at that time.
884 */ 884 */
885 if (keventd_up() && reap_work->work.func == NULL) { 885 if (keventd_up() && reap_work->work.func == NULL) {
886 init_reap_node(cpu); 886 init_reap_node(cpu);
887 INIT_DELAYED_WORK(reap_work, cache_reap); 887 INIT_DELAYED_WORK(reap_work, cache_reap);
888 schedule_delayed_work_on(cpu, reap_work, 888 schedule_delayed_work_on(cpu, reap_work,
889 __round_jiffies_relative(HZ, cpu)); 889 __round_jiffies_relative(HZ, cpu));
890 } 890 }
891 } 891 }
892 892
893 static struct array_cache *alloc_arraycache(int node, int entries, 893 static struct array_cache *alloc_arraycache(int node, int entries,
894 int batchcount, gfp_t gfp) 894 int batchcount, gfp_t gfp)
895 { 895 {
896 int memsize = sizeof(void *) * entries + sizeof(struct array_cache); 896 int memsize = sizeof(void *) * entries + sizeof(struct array_cache);
897 struct array_cache *nc = NULL; 897 struct array_cache *nc = NULL;
898 898
899 nc = kmalloc_node(memsize, gfp, node); 899 nc = kmalloc_node(memsize, gfp, node);
900 /* 900 /*
901 * The array_cache structures contain pointers to free object. 901 * The array_cache structures contain pointers to free object.
902 * However, when such objects are allocated or transfered to another 902 * However, when such objects are allocated or transfered to another
903 * cache the pointers are not cleared and they could be counted as 903 * cache the pointers are not cleared and they could be counted as
904 * valid references during a kmemleak scan. Therefore, kmemleak must 904 * valid references during a kmemleak scan. Therefore, kmemleak must
905 * not scan such objects. 905 * not scan such objects.
906 */ 906 */
907 kmemleak_no_scan(nc); 907 kmemleak_no_scan(nc);
908 if (nc) { 908 if (nc) {
909 nc->avail = 0; 909 nc->avail = 0;
910 nc->limit = entries; 910 nc->limit = entries;
911 nc->batchcount = batchcount; 911 nc->batchcount = batchcount;
912 nc->touched = 0; 912 nc->touched = 0;
913 spin_lock_init(&nc->lock); 913 spin_lock_init(&nc->lock);
914 } 914 }
915 return nc; 915 return nc;
916 } 916 }
917 917
918 /* 918 /*
919 * Transfer objects in one arraycache to another. 919 * Transfer objects in one arraycache to another.
920 * Locking must be handled by the caller. 920 * Locking must be handled by the caller.
921 * 921 *
922 * Return the number of entries transferred. 922 * Return the number of entries transferred.
923 */ 923 */
924 static int transfer_objects(struct array_cache *to, 924 static int transfer_objects(struct array_cache *to,
925 struct array_cache *from, unsigned int max) 925 struct array_cache *from, unsigned int max)
926 { 926 {
927 /* Figure out how many entries to transfer */ 927 /* Figure out how many entries to transfer */
928 int nr = min(min(from->avail, max), to->limit - to->avail); 928 int nr = min(min(from->avail, max), to->limit - to->avail);
929 929
930 if (!nr) 930 if (!nr)
931 return 0; 931 return 0;
932 932
933 memcpy(to->entry + to->avail, from->entry + from->avail -nr, 933 memcpy(to->entry + to->avail, from->entry + from->avail -nr,
934 sizeof(void *) *nr); 934 sizeof(void *) *nr);
935 935
936 from->avail -= nr; 936 from->avail -= nr;
937 to->avail += nr; 937 to->avail += nr;
938 to->touched = 1; 938 to->touched = 1;
939 return nr; 939 return nr;
940 } 940 }
941 941
942 #ifndef CONFIG_NUMA 942 #ifndef CONFIG_NUMA
943 943
944 #define drain_alien_cache(cachep, alien) do { } while (0) 944 #define drain_alien_cache(cachep, alien) do { } while (0)
945 #define reap_alien(cachep, l3) do { } while (0) 945 #define reap_alien(cachep, l3) do { } while (0)
946 946
947 static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) 947 static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
948 { 948 {
949 return (struct array_cache **)BAD_ALIEN_MAGIC; 949 return (struct array_cache **)BAD_ALIEN_MAGIC;
950 } 950 }
951 951
952 static inline void free_alien_cache(struct array_cache **ac_ptr) 952 static inline void free_alien_cache(struct array_cache **ac_ptr)
953 { 953 {
954 } 954 }
955 955
956 static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) 956 static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
957 { 957 {
958 return 0; 958 return 0;
959 } 959 }
960 960
961 static inline void *alternate_node_alloc(struct kmem_cache *cachep, 961 static inline void *alternate_node_alloc(struct kmem_cache *cachep,
962 gfp_t flags) 962 gfp_t flags)
963 { 963 {
964 return NULL; 964 return NULL;
965 } 965 }
966 966
967 static inline void *____cache_alloc_node(struct kmem_cache *cachep, 967 static inline void *____cache_alloc_node(struct kmem_cache *cachep,
968 gfp_t flags, int nodeid) 968 gfp_t flags, int nodeid)
969 { 969 {
970 return NULL; 970 return NULL;
971 } 971 }
972 972
973 #else /* CONFIG_NUMA */ 973 #else /* CONFIG_NUMA */
974 974
975 static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); 975 static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
976 static void *alternate_node_alloc(struct kmem_cache *, gfp_t); 976 static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
977 977
978 static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) 978 static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
979 { 979 {
980 struct array_cache **ac_ptr; 980 struct array_cache **ac_ptr;
981 int memsize = sizeof(void *) * nr_node_ids; 981 int memsize = sizeof(void *) * nr_node_ids;
982 int i; 982 int i;
983 983
984 if (limit > 1) 984 if (limit > 1)
985 limit = 12; 985 limit = 12;
986 ac_ptr = kmalloc_node(memsize, gfp, node); 986 ac_ptr = kmalloc_node(memsize, gfp, node);
987 if (ac_ptr) { 987 if (ac_ptr) {
988 for_each_node(i) { 988 for_each_node(i) {
989 if (i == node || !node_online(i)) { 989 if (i == node || !node_online(i)) {
990 ac_ptr[i] = NULL; 990 ac_ptr[i] = NULL;
991 continue; 991 continue;
992 } 992 }
993 ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp); 993 ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp);
994 if (!ac_ptr[i]) { 994 if (!ac_ptr[i]) {
995 for (i--; i >= 0; i--) 995 for (i--; i >= 0; i--)
996 kfree(ac_ptr[i]); 996 kfree(ac_ptr[i]);
997 kfree(ac_ptr); 997 kfree(ac_ptr);
998 return NULL; 998 return NULL;
999 } 999 }
1000 } 1000 }
1001 } 1001 }
1002 return ac_ptr; 1002 return ac_ptr;
1003 } 1003 }
1004 1004
1005 static void free_alien_cache(struct array_cache **ac_ptr) 1005 static void free_alien_cache(struct array_cache **ac_ptr)
1006 { 1006 {
1007 int i; 1007 int i;
1008 1008
1009 if (!ac_ptr) 1009 if (!ac_ptr)
1010 return; 1010 return;
1011 for_each_node(i) 1011 for_each_node(i)
1012 kfree(ac_ptr[i]); 1012 kfree(ac_ptr[i]);
1013 kfree(ac_ptr); 1013 kfree(ac_ptr);
1014 } 1014 }
1015 1015
1016 static void __drain_alien_cache(struct kmem_cache *cachep, 1016 static void __drain_alien_cache(struct kmem_cache *cachep,
1017 struct array_cache *ac, int node) 1017 struct array_cache *ac, int node)
1018 { 1018 {
1019 struct kmem_list3 *rl3 = cachep->nodelists[node]; 1019 struct kmem_list3 *rl3 = cachep->nodelists[node];
1020 1020
1021 if (ac->avail) { 1021 if (ac->avail) {
1022 spin_lock(&rl3->list_lock); 1022 spin_lock(&rl3->list_lock);
1023 /* 1023 /*
1024 * Stuff objects into the remote nodes shared array first. 1024 * Stuff objects into the remote nodes shared array first.
1025 * That way we could avoid the overhead of putting the objects 1025 * That way we could avoid the overhead of putting the objects
1026 * into the free lists and getting them back later. 1026 * into the free lists and getting them back later.
1027 */ 1027 */
1028 if (rl3->shared) 1028 if (rl3->shared)
1029 transfer_objects(rl3->shared, ac, ac->limit); 1029 transfer_objects(rl3->shared, ac, ac->limit);
1030 1030
1031 free_block(cachep, ac->entry, ac->avail, node); 1031 free_block(cachep, ac->entry, ac->avail, node);
1032 ac->avail = 0; 1032 ac->avail = 0;
1033 spin_unlock(&rl3->list_lock); 1033 spin_unlock(&rl3->list_lock);
1034 } 1034 }
1035 } 1035 }
1036 1036
1037 /* 1037 /*
1038 * Called from cache_reap() to regularly drain alien caches round robin. 1038 * Called from cache_reap() to regularly drain alien caches round robin.
1039 */ 1039 */
1040 static void reap_alien(struct kmem_cache *cachep, struct kmem_list3 *l3) 1040 static void reap_alien(struct kmem_cache *cachep, struct kmem_list3 *l3)
1041 { 1041 {
1042 int node = __get_cpu_var(slab_reap_node); 1042 int node = __get_cpu_var(slab_reap_node);
1043 1043
1044 if (l3->alien) { 1044 if (l3->alien) {
1045 struct array_cache *ac = l3->alien[node]; 1045 struct array_cache *ac = l3->alien[node];
1046 1046
1047 if (ac && ac->avail && spin_trylock_irq(&ac->lock)) { 1047 if (ac && ac->avail && spin_trylock_irq(&ac->lock)) {
1048 __drain_alien_cache(cachep, ac, node); 1048 __drain_alien_cache(cachep, ac, node);
1049 spin_unlock_irq(&ac->lock); 1049 spin_unlock_irq(&ac->lock);
1050 } 1050 }
1051 } 1051 }
1052 } 1052 }
1053 1053
1054 static void drain_alien_cache(struct kmem_cache *cachep, 1054 static void drain_alien_cache(struct kmem_cache *cachep,
1055 struct array_cache **alien) 1055 struct array_cache **alien)
1056 { 1056 {
1057 int i = 0; 1057 int i = 0;
1058 struct array_cache *ac; 1058 struct array_cache *ac;
1059 unsigned long flags; 1059 unsigned long flags;
1060 1060
1061 for_each_online_node(i) { 1061 for_each_online_node(i) {
1062 ac = alien[i]; 1062 ac = alien[i];
1063 if (ac) { 1063 if (ac) {
1064 spin_lock_irqsave(&ac->lock, flags); 1064 spin_lock_irqsave(&ac->lock, flags);
1065 __drain_alien_cache(cachep, ac, i); 1065 __drain_alien_cache(cachep, ac, i);
1066 spin_unlock_irqrestore(&ac->lock, flags); 1066 spin_unlock_irqrestore(&ac->lock, flags);
1067 } 1067 }
1068 } 1068 }
1069 } 1069 }
1070 1070
1071 static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) 1071 static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
1072 { 1072 {
1073 struct slab *slabp = virt_to_slab(objp); 1073 struct slab *slabp = virt_to_slab(objp);
1074 int nodeid = slabp->nodeid; 1074 int nodeid = slabp->nodeid;
1075 struct kmem_list3 *l3; 1075 struct kmem_list3 *l3;
1076 struct array_cache *alien = NULL; 1076 struct array_cache *alien = NULL;
1077 int node; 1077 int node;
1078 1078
1079 node = numa_node_id(); 1079 node = numa_node_id();
1080 1080
1081 /* 1081 /*
1082 * Make sure we are not freeing a object from another node to the array 1082 * Make sure we are not freeing a object from another node to the array
1083 * cache on this cpu. 1083 * cache on this cpu.
1084 */ 1084 */
1085 if (likely(slabp->nodeid == node)) 1085 if (likely(slabp->nodeid == node))
1086 return 0; 1086 return 0;
1087 1087
1088 l3 = cachep->nodelists[node]; 1088 l3 = cachep->nodelists[node];
1089 STATS_INC_NODEFREES(cachep); 1089 STATS_INC_NODEFREES(cachep);
1090 if (l3->alien && l3->alien[nodeid]) { 1090 if (l3->alien && l3->alien[nodeid]) {
1091 alien = l3->alien[nodeid]; 1091 alien = l3->alien[nodeid];
1092 spin_lock(&alien->lock); 1092 spin_lock(&alien->lock);
1093 if (unlikely(alien->avail == alien->limit)) { 1093 if (unlikely(alien->avail == alien->limit)) {
1094 STATS_INC_ACOVERFLOW(cachep); 1094 STATS_INC_ACOVERFLOW(cachep);
1095 __drain_alien_cache(cachep, alien, nodeid); 1095 __drain_alien_cache(cachep, alien, nodeid);
1096 } 1096 }
1097 alien->entry[alien->avail++] = objp; 1097 alien->entry[alien->avail++] = objp;
1098 spin_unlock(&alien->lock); 1098 spin_unlock(&alien->lock);
1099 } else { 1099 } else {
1100 spin_lock(&(cachep->nodelists[nodeid])->list_lock); 1100 spin_lock(&(cachep->nodelists[nodeid])->list_lock);
1101 free_block(cachep, &objp, 1, nodeid); 1101 free_block(cachep, &objp, 1, nodeid);
1102 spin_unlock(&(cachep->nodelists[nodeid])->list_lock); 1102 spin_unlock(&(cachep->nodelists[nodeid])->list_lock);
1103 } 1103 }
1104 return 1; 1104 return 1;
1105 } 1105 }
1106 #endif 1106 #endif
1107 1107
1108 static void __cpuinit cpuup_canceled(long cpu) 1108 static void __cpuinit cpuup_canceled(long cpu)
1109 { 1109 {
1110 struct kmem_cache *cachep; 1110 struct kmem_cache *cachep;
1111 struct kmem_list3 *l3 = NULL; 1111 struct kmem_list3 *l3 = NULL;
1112 int node = cpu_to_node(cpu); 1112 int node = cpu_to_node(cpu);
1113 const struct cpumask *mask = cpumask_of_node(node); 1113 const struct cpumask *mask = cpumask_of_node(node);
1114 1114
1115 list_for_each_entry(cachep, &cache_chain, next) { 1115 list_for_each_entry(cachep, &cache_chain, next) {
1116 struct array_cache *nc; 1116 struct array_cache *nc;
1117 struct array_cache *shared; 1117 struct array_cache *shared;
1118 struct array_cache **alien; 1118 struct array_cache **alien;
1119 1119
1120 /* cpu is dead; no one can alloc from it. */ 1120 /* cpu is dead; no one can alloc from it. */
1121 nc = cachep->array[cpu]; 1121 nc = cachep->array[cpu];
1122 cachep->array[cpu] = NULL; 1122 cachep->array[cpu] = NULL;
1123 l3 = cachep->nodelists[node]; 1123 l3 = cachep->nodelists[node];
1124 1124
1125 if (!l3) 1125 if (!l3)
1126 goto free_array_cache; 1126 goto free_array_cache;
1127 1127
1128 spin_lock_irq(&l3->list_lock); 1128 spin_lock_irq(&l3->list_lock);
1129 1129
1130 /* Free limit for this kmem_list3 */ 1130 /* Free limit for this kmem_list3 */
1131 l3->free_limit -= cachep->batchcount; 1131 l3->free_limit -= cachep->batchcount;
1132 if (nc) 1132 if (nc)
1133 free_block(cachep, nc->entry, nc->avail, node); 1133 free_block(cachep, nc->entry, nc->avail, node);
1134 1134
1135 if (!cpumask_empty(mask)) { 1135 if (!cpumask_empty(mask)) {
1136 spin_unlock_irq(&l3->list_lock); 1136 spin_unlock_irq(&l3->list_lock);
1137 goto free_array_cache; 1137 goto free_array_cache;
1138 } 1138 }
1139 1139
1140 shared = l3->shared; 1140 shared = l3->shared;
1141 if (shared) { 1141 if (shared) {
1142 free_block(cachep, shared->entry, 1142 free_block(cachep, shared->entry,
1143 shared->avail, node); 1143 shared->avail, node);
1144 l3->shared = NULL; 1144 l3->shared = NULL;
1145 } 1145 }
1146 1146
1147 alien = l3->alien; 1147 alien = l3->alien;
1148 l3->alien = NULL; 1148 l3->alien = NULL;
1149 1149
1150 spin_unlock_irq(&l3->list_lock); 1150 spin_unlock_irq(&l3->list_lock);
1151 1151
1152 kfree(shared); 1152 kfree(shared);
1153 if (alien) { 1153 if (alien) {
1154 drain_alien_cache(cachep, alien); 1154 drain_alien_cache(cachep, alien);
1155 free_alien_cache(alien); 1155 free_alien_cache(alien);
1156 } 1156 }
1157 free_array_cache: 1157 free_array_cache:
1158 kfree(nc); 1158 kfree(nc);
1159 } 1159 }
1160 /* 1160 /*
1161 * In the previous loop, all the objects were freed to 1161 * In the previous loop, all the objects were freed to
1162 * the respective cache's slabs, now we can go ahead and 1162 * the respective cache's slabs, now we can go ahead and
1163 * shrink each nodelist to its limit. 1163 * shrink each nodelist to its limit.
1164 */ 1164 */
1165 list_for_each_entry(cachep, &cache_chain, next) { 1165 list_for_each_entry(cachep, &cache_chain, next) {
1166 l3 = cachep->nodelists[node]; 1166 l3 = cachep->nodelists[node];
1167 if (!l3) 1167 if (!l3)
1168 continue; 1168 continue;
1169 drain_freelist(cachep, l3, l3->free_objects); 1169 drain_freelist(cachep, l3, l3->free_objects);
1170 } 1170 }
1171 } 1171 }
1172 1172
1173 static int __cpuinit cpuup_prepare(long cpu) 1173 static int __cpuinit cpuup_prepare(long cpu)
1174 { 1174 {
1175 struct kmem_cache *cachep; 1175 struct kmem_cache *cachep;
1176 struct kmem_list3 *l3 = NULL; 1176 struct kmem_list3 *l3 = NULL;
1177 int node = cpu_to_node(cpu); 1177 int node = cpu_to_node(cpu);
1178 const int memsize = sizeof(struct kmem_list3); 1178 const int memsize = sizeof(struct kmem_list3);
1179 1179
1180 /* 1180 /*
1181 * We need to do this right in the beginning since 1181 * We need to do this right in the beginning since
1182 * alloc_arraycache's are going to use this list. 1182 * alloc_arraycache's are going to use this list.
1183 * kmalloc_node allows us to add the slab to the right 1183 * kmalloc_node allows us to add the slab to the right
1184 * kmem_list3 and not this cpu's kmem_list3 1184 * kmem_list3 and not this cpu's kmem_list3
1185 */ 1185 */
1186 1186
1187 list_for_each_entry(cachep, &cache_chain, next) { 1187 list_for_each_entry(cachep, &cache_chain, next) {
1188 /* 1188 /*
1189 * Set up the size64 kmemlist for cpu before we can 1189 * Set up the size64 kmemlist for cpu before we can
1190 * begin anything. Make sure some other cpu on this 1190 * begin anything. Make sure some other cpu on this
1191 * node has not already allocated this 1191 * node has not already allocated this
1192 */ 1192 */
1193 if (!cachep->nodelists[node]) { 1193 if (!cachep->nodelists[node]) {
1194 l3 = kmalloc_node(memsize, GFP_KERNEL, node); 1194 l3 = kmalloc_node(memsize, GFP_KERNEL, node);
1195 if (!l3) 1195 if (!l3)
1196 goto bad; 1196 goto bad;
1197 kmem_list3_init(l3); 1197 kmem_list3_init(l3);
1198 l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + 1198 l3->next_reap = jiffies + REAPTIMEOUT_LIST3 +
1199 ((unsigned long)cachep) % REAPTIMEOUT_LIST3; 1199 ((unsigned long)cachep) % REAPTIMEOUT_LIST3;
1200 1200
1201 /* 1201 /*
1202 * The l3s don't come and go as CPUs come and 1202 * The l3s don't come and go as CPUs come and
1203 * go. cache_chain_mutex is sufficient 1203 * go. cache_chain_mutex is sufficient
1204 * protection here. 1204 * protection here.
1205 */ 1205 */
1206 cachep->nodelists[node] = l3; 1206 cachep->nodelists[node] = l3;
1207 } 1207 }
1208 1208
1209 spin_lock_irq(&cachep->nodelists[node]->list_lock); 1209 spin_lock_irq(&cachep->nodelists[node]->list_lock);
1210 cachep->nodelists[node]->free_limit = 1210 cachep->nodelists[node]->free_limit =
1211 (1 + nr_cpus_node(node)) * 1211 (1 + nr_cpus_node(node)) *
1212 cachep->batchcount + cachep->num; 1212 cachep->batchcount + cachep->num;
1213 spin_unlock_irq(&cachep->nodelists[node]->list_lock); 1213 spin_unlock_irq(&cachep->nodelists[node]->list_lock);
1214 } 1214 }
1215 1215
1216 /* 1216 /*
1217 * Now we can go ahead with allocating the shared arrays and 1217 * Now we can go ahead with allocating the shared arrays and
1218 * array caches 1218 * array caches
1219 */ 1219 */
1220 list_for_each_entry(cachep, &cache_chain, next) { 1220 list_for_each_entry(cachep, &cache_chain, next) {
1221 struct array_cache *nc; 1221 struct array_cache *nc;
1222 struct array_cache *shared = NULL; 1222 struct array_cache *shared = NULL;
1223 struct array_cache **alien = NULL; 1223 struct array_cache **alien = NULL;
1224 1224
1225 nc = alloc_arraycache(node, cachep->limit, 1225 nc = alloc_arraycache(node, cachep->limit,
1226 cachep->batchcount, GFP_KERNEL); 1226 cachep->batchcount, GFP_KERNEL);
1227 if (!nc) 1227 if (!nc)
1228 goto bad; 1228 goto bad;
1229 if (cachep->shared) { 1229 if (cachep->shared) {
1230 shared = alloc_arraycache(node, 1230 shared = alloc_arraycache(node,
1231 cachep->shared * cachep->batchcount, 1231 cachep->shared * cachep->batchcount,
1232 0xbaadf00d, GFP_KERNEL); 1232 0xbaadf00d, GFP_KERNEL);
1233 if (!shared) { 1233 if (!shared) {
1234 kfree(nc); 1234 kfree(nc);
1235 goto bad; 1235 goto bad;
1236 } 1236 }
1237 } 1237 }
1238 if (use_alien_caches) { 1238 if (use_alien_caches) {
1239 alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); 1239 alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL);
1240 if (!alien) { 1240 if (!alien) {
1241 kfree(shared); 1241 kfree(shared);
1242 kfree(nc); 1242 kfree(nc);
1243 goto bad; 1243 goto bad;
1244 } 1244 }
1245 } 1245 }
1246 cachep->array[cpu] = nc; 1246 cachep->array[cpu] = nc;
1247 l3 = cachep->nodelists[node]; 1247 l3 = cachep->nodelists[node];
1248 BUG_ON(!l3); 1248 BUG_ON(!l3);
1249 1249
1250 spin_lock_irq(&l3->list_lock); 1250 spin_lock_irq(&l3->list_lock);
1251 if (!l3->shared) { 1251 if (!l3->shared) {
1252 /* 1252 /*
1253 * We are serialised from CPU_DEAD or 1253 * We are serialised from CPU_DEAD or
1254 * CPU_UP_CANCELLED by the cpucontrol lock 1254 * CPU_UP_CANCELLED by the cpucontrol lock
1255 */ 1255 */
1256 l3->shared = shared; 1256 l3->shared = shared;
1257 shared = NULL; 1257 shared = NULL;
1258 } 1258 }
1259 #ifdef CONFIG_NUMA 1259 #ifdef CONFIG_NUMA
1260 if (!l3->alien) { 1260 if (!l3->alien) {
1261 l3->alien = alien; 1261 l3->alien = alien;
1262 alien = NULL; 1262 alien = NULL;
1263 } 1263 }
1264 #endif 1264 #endif
1265 spin_unlock_irq(&l3->list_lock); 1265 spin_unlock_irq(&l3->list_lock);
1266 kfree(shared); 1266 kfree(shared);
1267 free_alien_cache(alien); 1267 free_alien_cache(alien);
1268 } 1268 }
1269 init_node_lock_keys(node); 1269 init_node_lock_keys(node);
1270 1270
1271 return 0; 1271 return 0;
1272 bad: 1272 bad:
1273 cpuup_canceled(cpu); 1273 cpuup_canceled(cpu);
1274 return -ENOMEM; 1274 return -ENOMEM;
1275 } 1275 }
1276 1276
1277 static int __cpuinit cpuup_callback(struct notifier_block *nfb, 1277 static int __cpuinit cpuup_callback(struct notifier_block *nfb,
1278 unsigned long action, void *hcpu) 1278 unsigned long action, void *hcpu)
1279 { 1279 {
1280 long cpu = (long)hcpu; 1280 long cpu = (long)hcpu;
1281 int err = 0; 1281 int err = 0;
1282 1282
1283 switch (action) { 1283 switch (action) {
1284 case CPU_UP_PREPARE: 1284 case CPU_UP_PREPARE:
1285 case CPU_UP_PREPARE_FROZEN: 1285 case CPU_UP_PREPARE_FROZEN:
1286 mutex_lock(&cache_chain_mutex); 1286 mutex_lock(&cache_chain_mutex);
1287 err = cpuup_prepare(cpu); 1287 err = cpuup_prepare(cpu);
1288 mutex_unlock(&cache_chain_mutex); 1288 mutex_unlock(&cache_chain_mutex);
1289 break; 1289 break;
1290 case CPU_ONLINE: 1290 case CPU_ONLINE:
1291 case CPU_ONLINE_FROZEN: 1291 case CPU_ONLINE_FROZEN:
1292 start_cpu_timer(cpu); 1292 start_cpu_timer(cpu);
1293 break; 1293 break;
1294 #ifdef CONFIG_HOTPLUG_CPU 1294 #ifdef CONFIG_HOTPLUG_CPU
1295 case CPU_DOWN_PREPARE: 1295 case CPU_DOWN_PREPARE:
1296 case CPU_DOWN_PREPARE_FROZEN: 1296 case CPU_DOWN_PREPARE_FROZEN:
1297 /* 1297 /*
1298 * Shutdown cache reaper. Note that the cache_chain_mutex is 1298 * Shutdown cache reaper. Note that the cache_chain_mutex is
1299 * held so that if cache_reap() is invoked it cannot do 1299 * held so that if cache_reap() is invoked it cannot do
1300 * anything expensive but will only modify reap_work 1300 * anything expensive but will only modify reap_work
1301 * and reschedule the timer. 1301 * and reschedule the timer.
1302 */ 1302 */
1303 cancel_rearming_delayed_work(&per_cpu(slab_reap_work, cpu)); 1303 cancel_rearming_delayed_work(&per_cpu(slab_reap_work, cpu));
1304 /* Now the cache_reaper is guaranteed to be not running. */ 1304 /* Now the cache_reaper is guaranteed to be not running. */
1305 per_cpu(slab_reap_work, cpu).work.func = NULL; 1305 per_cpu(slab_reap_work, cpu).work.func = NULL;
1306 break; 1306 break;
1307 case CPU_DOWN_FAILED: 1307 case CPU_DOWN_FAILED:
1308 case CPU_DOWN_FAILED_FROZEN: 1308 case CPU_DOWN_FAILED_FROZEN:
1309 start_cpu_timer(cpu); 1309 start_cpu_timer(cpu);
1310 break; 1310 break;
1311 case CPU_DEAD: 1311 case CPU_DEAD:
1312 case CPU_DEAD_FROZEN: 1312 case CPU_DEAD_FROZEN:
1313 /* 1313 /*
1314 * Even if all the cpus of a node are down, we don't free the 1314 * Even if all the cpus of a node are down, we don't free the
1315 * kmem_list3 of any cache. This to avoid a race between 1315 * kmem_list3 of any cache. This to avoid a race between
1316 * cpu_down, and a kmalloc allocation from another cpu for 1316 * cpu_down, and a kmalloc allocation from another cpu for
1317 * memory from the node of the cpu going down. The list3 1317 * memory from the node of the cpu going down. The list3
1318 * structure is usually allocated from kmem_cache_create() and 1318 * structure is usually allocated from kmem_cache_create() and
1319 * gets destroyed at kmem_cache_destroy(). 1319 * gets destroyed at kmem_cache_destroy().
1320 */ 1320 */
1321 /* fall through */ 1321 /* fall through */
1322 #endif 1322 #endif
1323 case CPU_UP_CANCELED: 1323 case CPU_UP_CANCELED:
1324 case CPU_UP_CANCELED_FROZEN: 1324 case CPU_UP_CANCELED_FROZEN:
1325 mutex_lock(&cache_chain_mutex); 1325 mutex_lock(&cache_chain_mutex);
1326 cpuup_canceled(cpu); 1326 cpuup_canceled(cpu);
1327 mutex_unlock(&cache_chain_mutex); 1327 mutex_unlock(&cache_chain_mutex);
1328 break; 1328 break;
1329 } 1329 }
1330 return err ? NOTIFY_BAD : NOTIFY_OK; 1330 return err ? NOTIFY_BAD : NOTIFY_OK;
1331 } 1331 }
1332 1332
1333 static struct notifier_block __cpuinitdata cpucache_notifier = { 1333 static struct notifier_block __cpuinitdata cpucache_notifier = {
1334 &cpuup_callback, NULL, 0 1334 &cpuup_callback, NULL, 0
1335 }; 1335 };
1336 1336
1337 /* 1337 /*
1338 * swap the static kmem_list3 with kmalloced memory 1338 * swap the static kmem_list3 with kmalloced memory
1339 */ 1339 */
1340 static void init_list(struct kmem_cache *cachep, struct kmem_list3 *list, 1340 static void init_list(struct kmem_cache *cachep, struct kmem_list3 *list,
1341 int nodeid) 1341 int nodeid)
1342 { 1342 {
1343 struct kmem_list3 *ptr; 1343 struct kmem_list3 *ptr;
1344 1344
1345 ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_NOWAIT, nodeid); 1345 ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_NOWAIT, nodeid);
1346 BUG_ON(!ptr); 1346 BUG_ON(!ptr);
1347 1347
1348 memcpy(ptr, list, sizeof(struct kmem_list3)); 1348 memcpy(ptr, list, sizeof(struct kmem_list3));
1349 /* 1349 /*
1350 * Do not assume that spinlocks can be initialized via memcpy: 1350 * Do not assume that spinlocks can be initialized via memcpy:
1351 */ 1351 */
1352 spin_lock_init(&ptr->list_lock); 1352 spin_lock_init(&ptr->list_lock);
1353 1353
1354 MAKE_ALL_LISTS(cachep, ptr, nodeid); 1354 MAKE_ALL_LISTS(cachep, ptr, nodeid);
1355 cachep->nodelists[nodeid] = ptr; 1355 cachep->nodelists[nodeid] = ptr;
1356 } 1356 }
1357 1357
1358 /* 1358 /*
1359 * For setting up all the kmem_list3s for cache whose buffer_size is same as 1359 * For setting up all the kmem_list3s for cache whose buffer_size is same as
1360 * size of kmem_list3. 1360 * size of kmem_list3.
1361 */ 1361 */
1362 static void __init set_up_list3s(struct kmem_cache *cachep, int index) 1362 static void __init set_up_list3s(struct kmem_cache *cachep, int index)
1363 { 1363 {
1364 int node; 1364 int node;
1365 1365
1366 for_each_online_node(node) { 1366 for_each_online_node(node) {
1367 cachep->nodelists[node] = &initkmem_list3[index + node]; 1367 cachep->nodelists[node] = &initkmem_list3[index + node];
1368 cachep->nodelists[node]->next_reap = jiffies + 1368 cachep->nodelists[node]->next_reap = jiffies +
1369 REAPTIMEOUT_LIST3 + 1369 REAPTIMEOUT_LIST3 +
1370 ((unsigned long)cachep) % REAPTIMEOUT_LIST3; 1370 ((unsigned long)cachep) % REAPTIMEOUT_LIST3;
1371 } 1371 }
1372 } 1372 }
1373 1373
1374 /* 1374 /*
1375 * Initialisation. Called after the page allocator have been initialised and 1375 * Initialisation. Called after the page allocator have been initialised and
1376 * before smp_init(). 1376 * before smp_init().
1377 */ 1377 */
1378 void __init kmem_cache_init(void) 1378 void __init kmem_cache_init(void)
1379 { 1379 {
1380 size_t left_over; 1380 size_t left_over;
1381 struct cache_sizes *sizes; 1381 struct cache_sizes *sizes;
1382 struct cache_names *names; 1382 struct cache_names *names;
1383 int i; 1383 int i;
1384 int order; 1384 int order;
1385 int node; 1385 int node;
1386 1386
1387 if (num_possible_nodes() == 1) 1387 if (num_possible_nodes() == 1)
1388 use_alien_caches = 0; 1388 use_alien_caches = 0;
1389 1389
1390 for (i = 0; i < NUM_INIT_LISTS; i++) { 1390 for (i = 0; i < NUM_INIT_LISTS; i++) {
1391 kmem_list3_init(&initkmem_list3[i]); 1391 kmem_list3_init(&initkmem_list3[i]);
1392 if (i < MAX_NUMNODES) 1392 if (i < MAX_NUMNODES)
1393 cache_cache.nodelists[i] = NULL; 1393 cache_cache.nodelists[i] = NULL;
1394 } 1394 }
1395 set_up_list3s(&cache_cache, CACHE_CACHE); 1395 set_up_list3s(&cache_cache, CACHE_CACHE);
1396 1396
1397 /* 1397 /*
1398 * Fragmentation resistance on low memory - only use bigger 1398 * Fragmentation resistance on low memory - only use bigger
1399 * page orders on machines with more than 32MB of memory. 1399 * page orders on machines with more than 32MB of memory.
1400 */ 1400 */
1401 if (totalram_pages > (32 << 20) >> PAGE_SHIFT) 1401 if (totalram_pages > (32 << 20) >> PAGE_SHIFT)
1402 slab_break_gfp_order = BREAK_GFP_ORDER_HI; 1402 slab_break_gfp_order = BREAK_GFP_ORDER_HI;
1403 1403
1404 /* Bootstrap is tricky, because several objects are allocated 1404 /* Bootstrap is tricky, because several objects are allocated
1405 * from caches that do not exist yet: 1405 * from caches that do not exist yet:
1406 * 1) initialize the cache_cache cache: it contains the struct 1406 * 1) initialize the cache_cache cache: it contains the struct
1407 * kmem_cache structures of all caches, except cache_cache itself: 1407 * kmem_cache structures of all caches, except cache_cache itself:
1408 * cache_cache is statically allocated. 1408 * cache_cache is statically allocated.
1409 * Initially an __init data area is used for the head array and the 1409 * Initially an __init data area is used for the head array and the
1410 * kmem_list3 structures, it's replaced with a kmalloc allocated 1410 * kmem_list3 structures, it's replaced with a kmalloc allocated
1411 * array at the end of the bootstrap. 1411 * array at the end of the bootstrap.
1412 * 2) Create the first kmalloc cache. 1412 * 2) Create the first kmalloc cache.
1413 * The struct kmem_cache for the new cache is allocated normally. 1413 * The struct kmem_cache for the new cache is allocated normally.
1414 * An __init data area is used for the head array. 1414 * An __init data area is used for the head array.
1415 * 3) Create the remaining kmalloc caches, with minimally sized 1415 * 3) Create the remaining kmalloc caches, with minimally sized
1416 * head arrays. 1416 * head arrays.
1417 * 4) Replace the __init data head arrays for cache_cache and the first 1417 * 4) Replace the __init data head arrays for cache_cache and the first
1418 * kmalloc cache with kmalloc allocated arrays. 1418 * kmalloc cache with kmalloc allocated arrays.
1419 * 5) Replace the __init data for kmem_list3 for cache_cache and 1419 * 5) Replace the __init data for kmem_list3 for cache_cache and
1420 * the other cache's with kmalloc allocated memory. 1420 * the other cache's with kmalloc allocated memory.
1421 * 6) Resize the head arrays of the kmalloc caches to their final sizes. 1421 * 6) Resize the head arrays of the kmalloc caches to their final sizes.
1422 */ 1422 */
1423 1423
1424 node = numa_node_id(); 1424 node = numa_node_id();
1425 1425
1426 /* 1) create the cache_cache */ 1426 /* 1) create the cache_cache */
1427 INIT_LIST_HEAD(&cache_chain); 1427 INIT_LIST_HEAD(&cache_chain);
1428 list_add(&cache_cache.next, &cache_chain); 1428 list_add(&cache_cache.next, &cache_chain);
1429 cache_cache.colour_off = cache_line_size(); 1429 cache_cache.colour_off = cache_line_size();
1430 cache_cache.array[smp_processor_id()] = &initarray_cache.cache; 1430 cache_cache.array[smp_processor_id()] = &initarray_cache.cache;
1431 cache_cache.nodelists[node] = &initkmem_list3[CACHE_CACHE + node]; 1431 cache_cache.nodelists[node] = &initkmem_list3[CACHE_CACHE + node];
1432 1432
1433 /* 1433 /*
1434 * struct kmem_cache size depends on nr_node_ids, which 1434 * struct kmem_cache size depends on nr_node_ids, which
1435 * can be less than MAX_NUMNODES. 1435 * can be less than MAX_NUMNODES.
1436 */ 1436 */
1437 cache_cache.buffer_size = offsetof(struct kmem_cache, nodelists) + 1437 cache_cache.buffer_size = offsetof(struct kmem_cache, nodelists) +
1438 nr_node_ids * sizeof(struct kmem_list3 *); 1438 nr_node_ids * sizeof(struct kmem_list3 *);
1439 #if DEBUG 1439 #if DEBUG
1440 cache_cache.obj_size = cache_cache.buffer_size; 1440 cache_cache.obj_size = cache_cache.buffer_size;
1441 #endif 1441 #endif
1442 cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, 1442 cache_cache.buffer_size = ALIGN(cache_cache.buffer_size,
1443 cache_line_size()); 1443 cache_line_size());
1444 cache_cache.reciprocal_buffer_size = 1444 cache_cache.reciprocal_buffer_size =
1445 reciprocal_value(cache_cache.buffer_size); 1445 reciprocal_value(cache_cache.buffer_size);
1446 1446
1447 for (order = 0; order < MAX_ORDER; order++) { 1447 for (order = 0; order < MAX_ORDER; order++) {
1448 cache_estimate(order, cache_cache.buffer_size, 1448 cache_estimate(order, cache_cache.buffer_size,
1449 cache_line_size(), 0, &left_over, &cache_cache.num); 1449 cache_line_size(), 0, &left_over, &cache_cache.num);
1450 if (cache_cache.num) 1450 if (cache_cache.num)
1451 break; 1451 break;
1452 } 1452 }
1453 BUG_ON(!cache_cache.num); 1453 BUG_ON(!cache_cache.num);
1454 cache_cache.gfporder = order; 1454 cache_cache.gfporder = order;
1455 cache_cache.colour = left_over / cache_cache.colour_off; 1455 cache_cache.colour = left_over / cache_cache.colour_off;
1456 cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) + 1456 cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) +
1457 sizeof(struct slab), cache_line_size()); 1457 sizeof(struct slab), cache_line_size());
1458 1458
1459 /* 2+3) create the kmalloc caches */ 1459 /* 2+3) create the kmalloc caches */
1460 sizes = malloc_sizes; 1460 sizes = malloc_sizes;
1461 names = cache_names; 1461 names = cache_names;
1462 1462
1463 /* 1463 /*
1464 * Initialize the caches that provide memory for the array cache and the 1464 * Initialize the caches that provide memory for the array cache and the
1465 * kmem_list3 structures first. Without this, further allocations will 1465 * kmem_list3 structures first. Without this, further allocations will
1466 * bug. 1466 * bug.
1467 */ 1467 */
1468 1468
1469 sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name, 1469 sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name,
1470 sizes[INDEX_AC].cs_size, 1470 sizes[INDEX_AC].cs_size,
1471 ARCH_KMALLOC_MINALIGN, 1471 ARCH_KMALLOC_MINALIGN,
1472 ARCH_KMALLOC_FLAGS|SLAB_PANIC, 1472 ARCH_KMALLOC_FLAGS|SLAB_PANIC,
1473 NULL); 1473 NULL);
1474 1474
1475 if (INDEX_AC != INDEX_L3) { 1475 if (INDEX_AC != INDEX_L3) {
1476 sizes[INDEX_L3].cs_cachep = 1476 sizes[INDEX_L3].cs_cachep =
1477 kmem_cache_create(names[INDEX_L3].name, 1477 kmem_cache_create(names[INDEX_L3].name,
1478 sizes[INDEX_L3].cs_size, 1478 sizes[INDEX_L3].cs_size,
1479 ARCH_KMALLOC_MINALIGN, 1479 ARCH_KMALLOC_MINALIGN,
1480 ARCH_KMALLOC_FLAGS|SLAB_PANIC, 1480 ARCH_KMALLOC_FLAGS|SLAB_PANIC,
1481 NULL); 1481 NULL);
1482 } 1482 }
1483 1483
1484 slab_early_init = 0; 1484 slab_early_init = 0;
1485 1485
1486 while (sizes->cs_size != ULONG_MAX) { 1486 while (sizes->cs_size != ULONG_MAX) {
1487 /* 1487 /*
1488 * For performance, all the general caches are L1 aligned. 1488 * For performance, all the general caches are L1 aligned.
1489 * This should be particularly beneficial on SMP boxes, as it 1489 * This should be particularly beneficial on SMP boxes, as it
1490 * eliminates "false sharing". 1490 * eliminates "false sharing".
1491 * Note for systems short on memory removing the alignment will 1491 * Note for systems short on memory removing the alignment will
1492 * allow tighter packing of the smaller caches. 1492 * allow tighter packing of the smaller caches.
1493 */ 1493 */
1494 if (!sizes->cs_cachep) { 1494 if (!sizes->cs_cachep) {
1495 sizes->cs_cachep = kmem_cache_create(names->name, 1495 sizes->cs_cachep = kmem_cache_create(names->name,
1496 sizes->cs_size, 1496 sizes->cs_size,
1497 ARCH_KMALLOC_MINALIGN, 1497 ARCH_KMALLOC_MINALIGN,
1498 ARCH_KMALLOC_FLAGS|SLAB_PANIC, 1498 ARCH_KMALLOC_FLAGS|SLAB_PANIC,
1499 NULL); 1499 NULL);
1500 } 1500 }
1501 #ifdef CONFIG_ZONE_DMA 1501 #ifdef CONFIG_ZONE_DMA
1502 sizes->cs_dmacachep = kmem_cache_create( 1502 sizes->cs_dmacachep = kmem_cache_create(
1503 names->name_dma, 1503 names->name_dma,
1504 sizes->cs_size, 1504 sizes->cs_size,
1505 ARCH_KMALLOC_MINALIGN, 1505 ARCH_KMALLOC_MINALIGN,
1506 ARCH_KMALLOC_FLAGS|SLAB_CACHE_DMA| 1506 ARCH_KMALLOC_FLAGS|SLAB_CACHE_DMA|
1507 SLAB_PANIC, 1507 SLAB_PANIC,
1508 NULL); 1508 NULL);
1509 #endif 1509 #endif
1510 sizes++; 1510 sizes++;
1511 names++; 1511 names++;
1512 } 1512 }
1513 /* 4) Replace the bootstrap head arrays */ 1513 /* 4) Replace the bootstrap head arrays */
1514 { 1514 {
1515 struct array_cache *ptr; 1515 struct array_cache *ptr;
1516 1516
1517 ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); 1517 ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT);
1518 1518
1519 BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); 1519 BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache);
1520 memcpy(ptr, cpu_cache_get(&cache_cache), 1520 memcpy(ptr, cpu_cache_get(&cache_cache),
1521 sizeof(struct arraycache_init)); 1521 sizeof(struct arraycache_init));
1522 /* 1522 /*
1523 * Do not assume that spinlocks can be initialized via memcpy: 1523 * Do not assume that spinlocks can be initialized via memcpy:
1524 */ 1524 */
1525 spin_lock_init(&ptr->lock); 1525 spin_lock_init(&ptr->lock);
1526 1526
1527 cache_cache.array[smp_processor_id()] = ptr; 1527 cache_cache.array[smp_processor_id()] = ptr;
1528 1528
1529 ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); 1529 ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT);
1530 1530
1531 BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) 1531 BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep)
1532 != &initarray_generic.cache); 1532 != &initarray_generic.cache);
1533 memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), 1533 memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep),
1534 sizeof(struct arraycache_init)); 1534 sizeof(struct arraycache_init));
1535 /* 1535 /*
1536 * Do not assume that spinlocks can be initialized via memcpy: 1536 * Do not assume that spinlocks can be initialized via memcpy:
1537 */ 1537 */
1538 spin_lock_init(&ptr->lock); 1538 spin_lock_init(&ptr->lock);
1539 1539
1540 malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = 1540 malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] =
1541 ptr; 1541 ptr;
1542 } 1542 }
1543 /* 5) Replace the bootstrap kmem_list3's */ 1543 /* 5) Replace the bootstrap kmem_list3's */
1544 { 1544 {
1545 int nid; 1545 int nid;
1546 1546
1547 for_each_online_node(nid) { 1547 for_each_online_node(nid) {
1548 init_list(&cache_cache, &initkmem_list3[CACHE_CACHE + nid], nid); 1548 init_list(&cache_cache, &initkmem_list3[CACHE_CACHE + nid], nid);
1549 1549
1550 init_list(malloc_sizes[INDEX_AC].cs_cachep, 1550 init_list(malloc_sizes[INDEX_AC].cs_cachep,
1551 &initkmem_list3[SIZE_AC + nid], nid); 1551 &initkmem_list3[SIZE_AC + nid], nid);
1552 1552
1553 if (INDEX_AC != INDEX_L3) { 1553 if (INDEX_AC != INDEX_L3) {
1554 init_list(malloc_sizes[INDEX_L3].cs_cachep, 1554 init_list(malloc_sizes[INDEX_L3].cs_cachep,
1555 &initkmem_list3[SIZE_L3 + nid], nid); 1555 &initkmem_list3[SIZE_L3 + nid], nid);
1556 } 1556 }
1557 } 1557 }
1558 } 1558 }
1559 1559
1560 g_cpucache_up = EARLY; 1560 g_cpucache_up = EARLY;
1561 } 1561 }
1562 1562
1563 void __init kmem_cache_init_late(void) 1563 void __init kmem_cache_init_late(void)
1564 { 1564 {
1565 struct kmem_cache *cachep; 1565 struct kmem_cache *cachep;
1566 1566
1567 /* 6) resize the head arrays to their final sizes */ 1567 /* 6) resize the head arrays to their final sizes */
1568 mutex_lock(&cache_chain_mutex); 1568 mutex_lock(&cache_chain_mutex);
1569 list_for_each_entry(cachep, &cache_chain, next) 1569 list_for_each_entry(cachep, &cache_chain, next)
1570 if (enable_cpucache(cachep, GFP_NOWAIT)) 1570 if (enable_cpucache(cachep, GFP_NOWAIT))
1571 BUG(); 1571 BUG();
1572 mutex_unlock(&cache_chain_mutex); 1572 mutex_unlock(&cache_chain_mutex);
1573 1573
1574 /* Done! */ 1574 /* Done! */
1575 g_cpucache_up = FULL; 1575 g_cpucache_up = FULL;
1576 1576
1577 /* Annotate slab for lockdep -- annotate the malloc caches */ 1577 /* Annotate slab for lockdep -- annotate the malloc caches */
1578 init_lock_keys(); 1578 init_lock_keys();
1579 1579
1580 /* 1580 /*
1581 * Register a cpu startup notifier callback that initializes 1581 * Register a cpu startup notifier callback that initializes
1582 * cpu_cache_get for all new cpus 1582 * cpu_cache_get for all new cpus
1583 */ 1583 */
1584 register_cpu_notifier(&cpucache_notifier); 1584 register_cpu_notifier(&cpucache_notifier);
1585 1585
1586 /* 1586 /*
1587 * The reap timers are started later, with a module init call: That part 1587 * The reap timers are started later, with a module init call: That part
1588 * of the kernel is not yet operational. 1588 * of the kernel is not yet operational.
1589 */ 1589 */
1590 } 1590 }
1591 1591
1592 static int __init cpucache_init(void) 1592 static int __init cpucache_init(void)
1593 { 1593 {
1594 int cpu; 1594 int cpu;
1595 1595
1596 /* 1596 /*
1597 * Register the timers that return unneeded pages to the page allocator 1597 * Register the timers that return unneeded pages to the page allocator
1598 */ 1598 */
1599 for_each_online_cpu(cpu) 1599 for_each_online_cpu(cpu)
1600 start_cpu_timer(cpu); 1600 start_cpu_timer(cpu);
1601 return 0; 1601 return 0;
1602 } 1602 }
1603 __initcall(cpucache_init); 1603 __initcall(cpucache_init);
1604 1604
1605 /* 1605 /*
1606 * Interface to system's page allocator. No need to hold the cache-lock. 1606 * Interface to system's page allocator. No need to hold the cache-lock.
1607 * 1607 *
1608 * If we requested dmaable memory, we will get it. Even if we 1608 * If we requested dmaable memory, we will get it. Even if we
1609 * did not request dmaable memory, we might get it, but that 1609 * did not request dmaable memory, we might get it, but that
1610 * would be relatively rare and ignorable. 1610 * would be relatively rare and ignorable.
1611 */ 1611 */
1612 static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid) 1612 static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid)
1613 { 1613 {
1614 struct page *page; 1614 struct page *page;
1615 int nr_pages; 1615 int nr_pages;
1616 int i; 1616 int i;
1617 1617
1618 #ifndef CONFIG_MMU 1618 #ifndef CONFIG_MMU
1619 /* 1619 /*
1620 * Nommu uses slab's for process anonymous memory allocations, and thus 1620 * Nommu uses slab's for process anonymous memory allocations, and thus
1621 * requires __GFP_COMP to properly refcount higher order allocations 1621 * requires __GFP_COMP to properly refcount higher order allocations
1622 */ 1622 */
1623 flags |= __GFP_COMP; 1623 flags |= __GFP_COMP;
1624 #endif 1624 #endif
1625 1625
1626 flags |= cachep->gfpflags; 1626 flags |= cachep->gfpflags;
1627 if (cachep->flags & SLAB_RECLAIM_ACCOUNT) 1627 if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
1628 flags |= __GFP_RECLAIMABLE; 1628 flags |= __GFP_RECLAIMABLE;
1629 1629
1630 page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); 1630 page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder);
1631 if (!page) 1631 if (!page)
1632 return NULL; 1632 return NULL;
1633 1633
1634 nr_pages = (1 << cachep->gfporder); 1634 nr_pages = (1 << cachep->gfporder);
1635 if (cachep->flags & SLAB_RECLAIM_ACCOUNT) 1635 if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
1636 add_zone_page_state(page_zone(page), 1636 add_zone_page_state(page_zone(page),
1637 NR_SLAB_RECLAIMABLE, nr_pages); 1637 NR_SLAB_RECLAIMABLE, nr_pages);
1638 else 1638 else
1639 add_zone_page_state(page_zone(page), 1639 add_zone_page_state(page_zone(page),
1640 NR_SLAB_UNRECLAIMABLE, nr_pages); 1640 NR_SLAB_UNRECLAIMABLE, nr_pages);
1641 for (i = 0; i < nr_pages; i++) 1641 for (i = 0; i < nr_pages; i++)
1642 __SetPageSlab(page + i); 1642 __SetPageSlab(page + i);
1643 1643
1644 if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { 1644 if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) {
1645 kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); 1645 kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid);
1646 1646
1647 if (cachep->ctor) 1647 if (cachep->ctor)
1648 kmemcheck_mark_uninitialized_pages(page, nr_pages); 1648 kmemcheck_mark_uninitialized_pages(page, nr_pages);
1649 else 1649 else
1650 kmemcheck_mark_unallocated_pages(page, nr_pages); 1650 kmemcheck_mark_unallocated_pages(page, nr_pages);
1651 } 1651 }
1652 1652
1653 return page_address(page); 1653 return page_address(page);
1654 } 1654 }
1655 1655
1656 /* 1656 /*
1657 * Interface to system's page release. 1657 * Interface to system's page release.
1658 */ 1658 */
1659 static void kmem_freepages(struct kmem_cache *cachep, void *addr) 1659 static void kmem_freepages(struct kmem_cache *cachep, void *addr)
1660 { 1660 {
1661 unsigned long i = (1 << cachep->gfporder); 1661 unsigned long i = (1 << cachep->gfporder);
1662 struct page *page = virt_to_page(addr); 1662 struct page *page = virt_to_page(addr);
1663 const unsigned long nr_freed = i; 1663 const unsigned long nr_freed = i;
1664 1664
1665 kmemcheck_free_shadow(page, cachep->gfporder); 1665 kmemcheck_free_shadow(page, cachep->gfporder);
1666 1666
1667 if (cachep->flags & SLAB_RECLAIM_ACCOUNT) 1667 if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
1668 sub_zone_page_state(page_zone(page), 1668 sub_zone_page_state(page_zone(page),
1669 NR_SLAB_RECLAIMABLE, nr_freed); 1669 NR_SLAB_RECLAIMABLE, nr_freed);
1670 else 1670 else
1671 sub_zone_page_state(page_zone(page), 1671 sub_zone_page_state(page_zone(page),
1672 NR_SLAB_UNRECLAIMABLE, nr_freed); 1672 NR_SLAB_UNRECLAIMABLE, nr_freed);
1673 while (i--) { 1673 while (i--) {
1674 BUG_ON(!PageSlab(page)); 1674 BUG_ON(!PageSlab(page));
1675 __ClearPageSlab(page); 1675 __ClearPageSlab(page);
1676 page++; 1676 page++;
1677 } 1677 }
1678 if (current->reclaim_state) 1678 if (current->reclaim_state)
1679 current->reclaim_state->reclaimed_slab += nr_freed; 1679 current->reclaim_state->reclaimed_slab += nr_freed;
1680 free_pages((unsigned long)addr, cachep->gfporder); 1680 free_pages((unsigned long)addr, cachep->gfporder);
1681 } 1681 }
1682 1682
1683 static void kmem_rcu_free(struct rcu_head *head) 1683 static void kmem_rcu_free(struct rcu_head *head)
1684 { 1684 {
1685 struct slab_rcu *slab_rcu = (struct slab_rcu *)head; 1685 struct slab_rcu *slab_rcu = (struct slab_rcu *)head;
1686 struct kmem_cache *cachep = slab_rcu->cachep; 1686 struct kmem_cache *cachep = slab_rcu->cachep;
1687 1687
1688 kmem_freepages(cachep, slab_rcu->addr); 1688 kmem_freepages(cachep, slab_rcu->addr);
1689 if (OFF_SLAB(cachep)) 1689 if (OFF_SLAB(cachep))
1690 kmem_cache_free(cachep->slabp_cache, slab_rcu); 1690 kmem_cache_free(cachep->slabp_cache, slab_rcu);
1691 } 1691 }
1692 1692
1693 #if DEBUG 1693 #if DEBUG
1694 1694
1695 #ifdef CONFIG_DEBUG_PAGEALLOC 1695 #ifdef CONFIG_DEBUG_PAGEALLOC
1696 static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, 1696 static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr,
1697 unsigned long caller) 1697 unsigned long caller)
1698 { 1698 {
1699 int size = obj_size(cachep); 1699 int size = obj_size(cachep);
1700 1700
1701 addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; 1701 addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)];
1702 1702
1703 if (size < 5 * sizeof(unsigned long)) 1703 if (size < 5 * sizeof(unsigned long))
1704 return; 1704 return;
1705 1705
1706 *addr++ = 0x12345678; 1706 *addr++ = 0x12345678;
1707 *addr++ = caller; 1707 *addr++ = caller;
1708 *addr++ = smp_processor_id(); 1708 *addr++ = smp_processor_id();
1709 size -= 3 * sizeof(unsigned long); 1709 size -= 3 * sizeof(unsigned long);
1710 { 1710 {
1711 unsigned long *sptr = &caller; 1711 unsigned long *sptr = &caller;
1712 unsigned long svalue; 1712 unsigned long svalue;
1713 1713
1714 while (!kstack_end(sptr)) { 1714 while (!kstack_end(sptr)) {
1715 svalue = *sptr++; 1715 svalue = *sptr++;
1716 if (kernel_text_address(svalue)) { 1716 if (kernel_text_address(svalue)) {
1717 *addr++ = svalue; 1717 *addr++ = svalue;
1718 size -= sizeof(unsigned long); 1718 size -= sizeof(unsigned long);
1719 if (size <= sizeof(unsigned long)) 1719 if (size <= sizeof(unsigned long))
1720 break; 1720 break;
1721 } 1721 }
1722 } 1722 }
1723 1723
1724 } 1724 }
1725 *addr++ = 0x87654321; 1725 *addr++ = 0x87654321;
1726 } 1726 }
1727 #endif 1727 #endif
1728 1728
1729 static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) 1729 static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val)
1730 { 1730 {
1731 int size = obj_size(cachep); 1731 int size = obj_size(cachep);
1732 addr = &((char *)addr)[obj_offset(cachep)]; 1732 addr = &((char *)addr)[obj_offset(cachep)];
1733 1733
1734 memset(addr, val, size); 1734 memset(addr, val, size);
1735 *(unsigned char *)(addr + size - 1) = POISON_END; 1735 *(unsigned char *)(addr + size - 1) = POISON_END;
1736 } 1736 }
1737 1737
1738 static void dump_line(char *data, int offset, int limit) 1738 static void dump_line(char *data, int offset, int limit)
1739 { 1739 {
1740 int i; 1740 int i;
1741 unsigned char error = 0; 1741 unsigned char error = 0;
1742 int bad_count = 0; 1742 int bad_count = 0;
1743 1743
1744 printk(KERN_ERR "%03x:", offset); 1744 printk(KERN_ERR "%03x:", offset);
1745 for (i = 0; i < limit; i++) { 1745 for (i = 0; i < limit; i++) {
1746 if (data[offset + i] != POISON_FREE) { 1746 if (data[offset + i] != POISON_FREE) {
1747 error = data[offset + i]; 1747 error = data[offset + i];
1748 bad_count++; 1748 bad_count++;
1749 } 1749 }
1750 printk(" %02x", (unsigned char)data[offset + i]); 1750 printk(" %02x", (unsigned char)data[offset + i]);
1751 } 1751 }
1752 printk("\n"); 1752 printk("\n");
1753 1753
1754 if (bad_count == 1) { 1754 if (bad_count == 1) {
1755 error ^= POISON_FREE; 1755 error ^= POISON_FREE;
1756 if (!(error & (error - 1))) { 1756 if (!(error & (error - 1))) {
1757 printk(KERN_ERR "Single bit error detected. Probably " 1757 printk(KERN_ERR "Single bit error detected. Probably "
1758 "bad RAM.\n"); 1758 "bad RAM.\n");
1759 #ifdef CONFIG_X86 1759 #ifdef CONFIG_X86
1760 printk(KERN_ERR "Run memtest86+ or a similar memory " 1760 printk(KERN_ERR "Run memtest86+ or a similar memory "
1761 "test tool.\n"); 1761 "test tool.\n");
1762 #else 1762 #else
1763 printk(KERN_ERR "Run a memory test tool.\n"); 1763 printk(KERN_ERR "Run a memory test tool.\n");
1764 #endif 1764 #endif
1765 } 1765 }
1766 } 1766 }
1767 } 1767 }
1768 #endif 1768 #endif
1769 1769
1770 #if DEBUG 1770 #if DEBUG
1771 1771
1772 static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) 1772 static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines)
1773 { 1773 {
1774 int i, size; 1774 int i, size;
1775 char *realobj; 1775 char *realobj;
1776 1776
1777 if (cachep->flags & SLAB_RED_ZONE) { 1777 if (cachep->flags & SLAB_RED_ZONE) {
1778 printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", 1778 printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n",
1779 *dbg_redzone1(cachep, objp), 1779 *dbg_redzone1(cachep, objp),
1780 *dbg_redzone2(cachep, objp)); 1780 *dbg_redzone2(cachep, objp));
1781 } 1781 }
1782 1782
1783 if (cachep->flags & SLAB_STORE_USER) { 1783 if (cachep->flags & SLAB_STORE_USER) {
1784 printk(KERN_ERR "Last user: [<%p>]", 1784 printk(KERN_ERR "Last user: [<%p>]",
1785 *dbg_userword(cachep, objp)); 1785 *dbg_userword(cachep, objp));
1786 print_symbol("(%s)", 1786 print_symbol("(%s)",
1787 (unsigned long)*dbg_userword(cachep, objp)); 1787 (unsigned long)*dbg_userword(cachep, objp));
1788 printk("\n"); 1788 printk("\n");
1789 } 1789 }
1790 realobj = (char *)objp + obj_offset(cachep); 1790 realobj = (char *)objp + obj_offset(cachep);
1791 size = obj_size(cachep); 1791 size = obj_size(cachep);
1792 for (i = 0; i < size && lines; i += 16, lines--) { 1792 for (i = 0; i < size && lines; i += 16, lines--) {
1793 int limit; 1793 int limit;
1794 limit = 16; 1794 limit = 16;
1795 if (i + limit > size) 1795 if (i + limit > size)
1796 limit = size - i; 1796 limit = size - i;
1797 dump_line(realobj, i, limit); 1797 dump_line(realobj, i, limit);
1798 } 1798 }
1799 } 1799 }
1800 1800
1801 static void check_poison_obj(struct kmem_cache *cachep, void *objp) 1801 static void check_poison_obj(struct kmem_cache *cachep, void *objp)
1802 { 1802 {
1803 char *realobj; 1803 char *realobj;
1804 int size, i; 1804 int size, i;
1805 int lines = 0; 1805 int lines = 0;
1806 1806
1807 realobj = (char *)objp + obj_offset(cachep); 1807 realobj = (char *)objp + obj_offset(cachep);
1808 size = obj_size(cachep); 1808 size = obj_size(cachep);
1809 1809
1810 for (i = 0; i < size; i++) { 1810 for (i = 0; i < size; i++) {
1811 char exp = POISON_FREE; 1811 char exp = POISON_FREE;
1812 if (i == size - 1) 1812 if (i == size - 1)
1813 exp = POISON_END; 1813 exp = POISON_END;
1814 if (realobj[i] != exp) { 1814 if (realobj[i] != exp) {
1815 int limit; 1815 int limit;
1816 /* Mismatch ! */ 1816 /* Mismatch ! */
1817 /* Print header */ 1817 /* Print header */
1818 if (lines == 0) { 1818 if (lines == 0) {
1819 printk(KERN_ERR 1819 printk(KERN_ERR
1820 "Slab corruption: %s start=%p, len=%d\n", 1820 "Slab corruption: %s start=%p, len=%d\n",
1821 cachep->name, realobj, size); 1821 cachep->name, realobj, size);
1822 print_objinfo(cachep, objp, 0); 1822 print_objinfo(cachep, objp, 0);
1823 } 1823 }
1824 /* Hexdump the affected line */ 1824 /* Hexdump the affected line */
1825 i = (i / 16) * 16; 1825 i = (i / 16) * 16;
1826 limit = 16; 1826 limit = 16;
1827 if (i + limit > size) 1827 if (i + limit > size)
1828 limit = size - i; 1828 limit = size - i;
1829 dump_line(realobj, i, limit); 1829 dump_line(realobj, i, limit);
1830 i += 16; 1830 i += 16;
1831 lines++; 1831 lines++;
1832 /* Limit to 5 lines */ 1832 /* Limit to 5 lines */
1833 if (lines > 5) 1833 if (lines > 5)
1834 break; 1834 break;
1835 } 1835 }
1836 } 1836 }
1837 if (lines != 0) { 1837 if (lines != 0) {
1838 /* Print some data about the neighboring objects, if they 1838 /* Print some data about the neighboring objects, if they
1839 * exist: 1839 * exist:
1840 */ 1840 */
1841 struct slab *slabp = virt_to_slab(objp); 1841 struct slab *slabp = virt_to_slab(objp);
1842 unsigned int objnr; 1842 unsigned int objnr;
1843 1843
1844 objnr = obj_to_index(cachep, slabp, objp); 1844 objnr = obj_to_index(cachep, slabp, objp);
1845 if (objnr) { 1845 if (objnr) {
1846 objp = index_to_obj(cachep, slabp, objnr - 1); 1846 objp = index_to_obj(cachep, slabp, objnr - 1);
1847 realobj = (char *)objp + obj_offset(cachep); 1847 realobj = (char *)objp + obj_offset(cachep);
1848 printk(KERN_ERR "Prev obj: start=%p, len=%d\n", 1848 printk(KERN_ERR "Prev obj: start=%p, len=%d\n",
1849 realobj, size); 1849 realobj, size);
1850 print_objinfo(cachep, objp, 2); 1850 print_objinfo(cachep, objp, 2);
1851 } 1851 }
1852 if (objnr + 1 < cachep->num) { 1852 if (objnr + 1 < cachep->num) {
1853 objp = index_to_obj(cachep, slabp, objnr + 1); 1853 objp = index_to_obj(cachep, slabp, objnr + 1);
1854 realobj = (char *)objp + obj_offset(cachep); 1854 realobj = (char *)objp + obj_offset(cachep);
1855 printk(KERN_ERR "Next obj: start=%p, len=%d\n", 1855 printk(KERN_ERR "Next obj: start=%p, len=%d\n",
1856 realobj, size); 1856 realobj, size);
1857 print_objinfo(cachep, objp, 2); 1857 print_objinfo(cachep, objp, 2);
1858 } 1858 }
1859 } 1859 }
1860 } 1860 }
1861 #endif 1861 #endif
1862 1862
1863 #if DEBUG 1863 #if DEBUG
1864 static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) 1864 static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp)
1865 { 1865 {
1866 int i; 1866 int i;
1867 for (i = 0; i < cachep->num; i++) { 1867 for (i = 0; i < cachep->num; i++) {
1868 void *objp = index_to_obj(cachep, slabp, i); 1868 void *objp = index_to_obj(cachep, slabp, i);
1869 1869
1870 if (cachep->flags & SLAB_POISON) { 1870 if (cachep->flags & SLAB_POISON) {
1871 #ifdef CONFIG_DEBUG_PAGEALLOC 1871 #ifdef CONFIG_DEBUG_PAGEALLOC
1872 if (cachep->buffer_size % PAGE_SIZE == 0 && 1872 if (cachep->buffer_size % PAGE_SIZE == 0 &&
1873 OFF_SLAB(cachep)) 1873 OFF_SLAB(cachep))
1874 kernel_map_pages(virt_to_page(objp), 1874 kernel_map_pages(virt_to_page(objp),
1875 cachep->buffer_size / PAGE_SIZE, 1); 1875 cachep->buffer_size / PAGE_SIZE, 1);
1876 else 1876 else
1877 check_poison_obj(cachep, objp); 1877 check_poison_obj(cachep, objp);
1878 #else 1878 #else
1879 check_poison_obj(cachep, objp); 1879 check_poison_obj(cachep, objp);
1880 #endif 1880 #endif
1881 } 1881 }
1882 if (cachep->flags & SLAB_RED_ZONE) { 1882 if (cachep->flags & SLAB_RED_ZONE) {
1883 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) 1883 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
1884 slab_error(cachep, "start of a freed object " 1884 slab_error(cachep, "start of a freed object "
1885 "was overwritten"); 1885 "was overwritten");
1886 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) 1886 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
1887 slab_error(cachep, "end of a freed object " 1887 slab_error(cachep, "end of a freed object "
1888 "was overwritten"); 1888 "was overwritten");
1889 } 1889 }
1890 } 1890 }
1891 } 1891 }
1892 #else 1892 #else
1893 static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) 1893 static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp)
1894 { 1894 {
1895 } 1895 }
1896 #endif 1896 #endif
1897 1897
1898 /** 1898 /**
1899 * slab_destroy - destroy and release all objects in a slab 1899 * slab_destroy - destroy and release all objects in a slab
1900 * @cachep: cache pointer being destroyed 1900 * @cachep: cache pointer being destroyed
1901 * @slabp: slab pointer being destroyed 1901 * @slabp: slab pointer being destroyed
1902 * 1902 *
1903 * Destroy all the objs in a slab, and release the mem back to the system. 1903 * Destroy all the objs in a slab, and release the mem back to the system.
1904 * Before calling the slab must have been unlinked from the cache. The 1904 * Before calling the slab must have been unlinked from the cache. The
1905 * cache-lock is not held/needed. 1905 * cache-lock is not held/needed.
1906 */ 1906 */
1907 static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) 1907 static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp)
1908 { 1908 {
1909 void *addr = slabp->s_mem - slabp->colouroff; 1909 void *addr = slabp->s_mem - slabp->colouroff;
1910 1910
1911 slab_destroy_debugcheck(cachep, slabp); 1911 slab_destroy_debugcheck(cachep, slabp);
1912 if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { 1912 if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) {
1913 struct slab_rcu *slab_rcu; 1913 struct slab_rcu *slab_rcu;
1914 1914
1915 slab_rcu = (struct slab_rcu *)slabp; 1915 slab_rcu = (struct slab_rcu *)slabp;
1916 slab_rcu->cachep = cachep; 1916 slab_rcu->cachep = cachep;
1917 slab_rcu->addr = addr; 1917 slab_rcu->addr = addr;
1918 call_rcu(&slab_rcu->head, kmem_rcu_free); 1918 call_rcu(&slab_rcu->head, kmem_rcu_free);
1919 } else { 1919 } else {
1920 kmem_freepages(cachep, addr); 1920 kmem_freepages(cachep, addr);
1921 if (OFF_SLAB(cachep)) 1921 if (OFF_SLAB(cachep))
1922 kmem_cache_free(cachep->slabp_cache, slabp); 1922 kmem_cache_free(cachep->slabp_cache, slabp);
1923 } 1923 }
1924 } 1924 }
1925 1925
1926 static void __kmem_cache_destroy(struct kmem_cache *cachep) 1926 static void __kmem_cache_destroy(struct kmem_cache *cachep)
1927 { 1927 {
1928 int i; 1928 int i;
1929 struct kmem_list3 *l3; 1929 struct kmem_list3 *l3;
1930 1930
1931 for_each_online_cpu(i) 1931 for_each_online_cpu(i)
1932 kfree(cachep->array[i]); 1932 kfree(cachep->array[i]);
1933 1933
1934 /* NUMA: free the list3 structures */ 1934 /* NUMA: free the list3 structures */
1935 for_each_online_node(i) { 1935 for_each_online_node(i) {
1936 l3 = cachep->nodelists[i]; 1936 l3 = cachep->nodelists[i];
1937 if (l3) { 1937 if (l3) {
1938 kfree(l3->shared); 1938 kfree(l3->shared);
1939 free_alien_cache(l3->alien); 1939 free_alien_cache(l3->alien);
1940 kfree(l3); 1940 kfree(l3);
1941 } 1941 }
1942 } 1942 }
1943 kmem_cache_free(&cache_cache, cachep); 1943 kmem_cache_free(&cache_cache, cachep);
1944 } 1944 }
1945 1945
1946 1946
1947 /** 1947 /**
1948 * calculate_slab_order - calculate size (page order) of slabs 1948 * calculate_slab_order - calculate size (page order) of slabs
1949 * @cachep: pointer to the cache that is being created 1949 * @cachep: pointer to the cache that is being created
1950 * @size: size of objects to be created in this cache. 1950 * @size: size of objects to be created in this cache.
1951 * @align: required alignment for the objects. 1951 * @align: required alignment for the objects.
1952 * @flags: slab allocation flags 1952 * @flags: slab allocation flags
1953 * 1953 *
1954 * Also calculates the number of objects per slab. 1954 * Also calculates the number of objects per slab.
1955 * 1955 *
1956 * This could be made much more intelligent. For now, try to avoid using 1956 * This could be made much more intelligent. For now, try to avoid using
1957 * high order pages for slabs. When the gfp() functions are more friendly 1957 * high order pages for slabs. When the gfp() functions are more friendly
1958 * towards high-order requests, this should be changed. 1958 * towards high-order requests, this should be changed.
1959 */ 1959 */
1960 static size_t calculate_slab_order(struct kmem_cache *cachep, 1960 static size_t calculate_slab_order(struct kmem_cache *cachep,
1961 size_t size, size_t align, unsigned long flags) 1961 size_t size, size_t align, unsigned long flags)
1962 { 1962 {
1963 unsigned long offslab_limit; 1963 unsigned long offslab_limit;
1964 size_t left_over = 0; 1964 size_t left_over = 0;
1965 int gfporder; 1965 int gfporder;
1966 1966
1967 for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { 1967 for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) {
1968 unsigned int num; 1968 unsigned int num;
1969 size_t remainder; 1969 size_t remainder;
1970 1970
1971 cache_estimate(gfporder, size, align, flags, &remainder, &num); 1971 cache_estimate(gfporder, size, align, flags, &remainder, &num);
1972 if (!num) 1972 if (!num)
1973 continue; 1973 continue;
1974 1974
1975 if (flags & CFLGS_OFF_SLAB) { 1975 if (flags & CFLGS_OFF_SLAB) {
1976 /* 1976 /*
1977 * Max number of objs-per-slab for caches which 1977 * Max number of objs-per-slab for caches which
1978 * use off-slab slabs. Needed to avoid a possible 1978 * use off-slab slabs. Needed to avoid a possible
1979 * looping condition in cache_grow(). 1979 * looping condition in cache_grow().
1980 */ 1980 */
1981 offslab_limit = size - sizeof(struct slab); 1981 offslab_limit = size - sizeof(struct slab);
1982 offslab_limit /= sizeof(kmem_bufctl_t); 1982 offslab_limit /= sizeof(kmem_bufctl_t);
1983 1983
1984 if (num > offslab_limit) 1984 if (num > offslab_limit)
1985 break; 1985 break;
1986 } 1986 }
1987 1987
1988 /* Found something acceptable - save it away */ 1988 /* Found something acceptable - save it away */
1989 cachep->num = num; 1989 cachep->num = num;
1990 cachep->gfporder = gfporder; 1990 cachep->gfporder = gfporder;
1991 left_over = remainder; 1991 left_over = remainder;
1992 1992
1993 /* 1993 /*
1994 * A VFS-reclaimable slab tends to have most allocations 1994 * A VFS-reclaimable slab tends to have most allocations
1995 * as GFP_NOFS and we really don't want to have to be allocating 1995 * as GFP_NOFS and we really don't want to have to be allocating
1996 * higher-order pages when we are unable to shrink dcache. 1996 * higher-order pages when we are unable to shrink dcache.
1997 */ 1997 */
1998 if (flags & SLAB_RECLAIM_ACCOUNT) 1998 if (flags & SLAB_RECLAIM_ACCOUNT)
1999 break; 1999 break;
2000 2000
2001 /* 2001 /*
2002 * Large number of objects is good, but very large slabs are 2002 * Large number of objects is good, but very large slabs are
2003 * currently bad for the gfp()s. 2003 * currently bad for the gfp()s.
2004 */ 2004 */
2005 if (gfporder >= slab_break_gfp_order) 2005 if (gfporder >= slab_break_gfp_order)
2006 break; 2006 break;
2007 2007
2008 /* 2008 /*
2009 * Acceptable internal fragmentation? 2009 * Acceptable internal fragmentation?
2010 */ 2010 */
2011 if (left_over * 8 <= (PAGE_SIZE << gfporder)) 2011 if (left_over * 8 <= (PAGE_SIZE << gfporder))
2012 break; 2012 break;
2013 } 2013 }
2014 return left_over; 2014 return left_over;
2015 } 2015 }
2016 2016
2017 static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) 2017 static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp)
2018 { 2018 {
2019 if (g_cpucache_up == FULL) 2019 if (g_cpucache_up == FULL)
2020 return enable_cpucache(cachep, gfp); 2020 return enable_cpucache(cachep, gfp);
2021 2021
2022 if (g_cpucache_up == NONE) { 2022 if (g_cpucache_up == NONE) {
2023 /* 2023 /*
2024 * Note: the first kmem_cache_create must create the cache 2024 * Note: the first kmem_cache_create must create the cache
2025 * that's used by kmalloc(24), otherwise the creation of 2025 * that's used by kmalloc(24), otherwise the creation of
2026 * further caches will BUG(). 2026 * further caches will BUG().
2027 */ 2027 */
2028 cachep->array[smp_processor_id()] = &initarray_generic.cache; 2028 cachep->array[smp_processor_id()] = &initarray_generic.cache;
2029 2029
2030 /* 2030 /*
2031 * If the cache that's used by kmalloc(sizeof(kmem_list3)) is 2031 * If the cache that's used by kmalloc(sizeof(kmem_list3)) is
2032 * the first cache, then we need to set up all its list3s, 2032 * the first cache, then we need to set up all its list3s,
2033 * otherwise the creation of further caches will BUG(). 2033 * otherwise the creation of further caches will BUG().
2034 */ 2034 */
2035 set_up_list3s(cachep, SIZE_AC); 2035 set_up_list3s(cachep, SIZE_AC);
2036 if (INDEX_AC == INDEX_L3) 2036 if (INDEX_AC == INDEX_L3)
2037 g_cpucache_up = PARTIAL_L3; 2037 g_cpucache_up = PARTIAL_L3;
2038 else 2038 else
2039 g_cpucache_up = PARTIAL_AC; 2039 g_cpucache_up = PARTIAL_AC;
2040 } else { 2040 } else {
2041 cachep->array[smp_processor_id()] = 2041 cachep->array[smp_processor_id()] =
2042 kmalloc(sizeof(struct arraycache_init), gfp); 2042 kmalloc(sizeof(struct arraycache_init), gfp);
2043 2043
2044 if (g_cpucache_up == PARTIAL_AC) { 2044 if (g_cpucache_up == PARTIAL_AC) {
2045 set_up_list3s(cachep, SIZE_L3); 2045 set_up_list3s(cachep, SIZE_L3);
2046 g_cpucache_up = PARTIAL_L3; 2046 g_cpucache_up = PARTIAL_L3;
2047 } else { 2047 } else {
2048 int node; 2048 int node;
2049 for_each_online_node(node) { 2049 for_each_online_node(node) {
2050 cachep->nodelists[node] = 2050 cachep->nodelists[node] =
2051 kmalloc_node(sizeof(struct kmem_list3), 2051 kmalloc_node(sizeof(struct kmem_list3),
2052 gfp, node); 2052 gfp, node);
2053 BUG_ON(!cachep->nodelists[node]); 2053 BUG_ON(!cachep->nodelists[node]);
2054 kmem_list3_init(cachep->nodelists[node]); 2054 kmem_list3_init(cachep->nodelists[node]);
2055 } 2055 }
2056 } 2056 }
2057 } 2057 }
2058 cachep->nodelists[numa_node_id()]->next_reap = 2058 cachep->nodelists[numa_node_id()]->next_reap =
2059 jiffies + REAPTIMEOUT_LIST3 + 2059 jiffies + REAPTIMEOUT_LIST3 +
2060 ((unsigned long)cachep) % REAPTIMEOUT_LIST3; 2060 ((unsigned long)cachep) % REAPTIMEOUT_LIST3;
2061 2061
2062 cpu_cache_get(cachep)->avail = 0; 2062 cpu_cache_get(cachep)->avail = 0;
2063 cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; 2063 cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES;
2064 cpu_cache_get(cachep)->batchcount = 1; 2064 cpu_cache_get(cachep)->batchcount = 1;
2065 cpu_cache_get(cachep)->touched = 0; 2065 cpu_cache_get(cachep)->touched = 0;
2066 cachep->batchcount = 1; 2066 cachep->batchcount = 1;
2067 cachep->limit = BOOT_CPUCACHE_ENTRIES; 2067 cachep->limit = BOOT_CPUCACHE_ENTRIES;
2068 return 0; 2068 return 0;
2069 } 2069 }
2070 2070
2071 /** 2071 /**
2072 * kmem_cache_create - Create a cache. 2072 * kmem_cache_create - Create a cache.
2073 * @name: A string which is used in /proc/slabinfo to identify this cache. 2073 * @name: A string which is used in /proc/slabinfo to identify this cache.
2074 * @size: The size of objects to be created in this cache. 2074 * @size: The size of objects to be created in this cache.
2075 * @align: The required alignment for the objects. 2075 * @align: The required alignment for the objects.
2076 * @flags: SLAB flags 2076 * @flags: SLAB flags
2077 * @ctor: A constructor for the objects. 2077 * @ctor: A constructor for the objects.
2078 * 2078 *
2079 * Returns a ptr to the cache on success, NULL on failure. 2079 * Returns a ptr to the cache on success, NULL on failure.
2080 * Cannot be called within a int, but can be interrupted. 2080 * Cannot be called within a int, but can be interrupted.
2081 * The @ctor is run when new pages are allocated by the cache. 2081 * The @ctor is run when new pages are allocated by the cache.
2082 * 2082 *
2083 * @name must be valid until the cache is destroyed. This implies that 2083 * @name must be valid until the cache is destroyed. This implies that
2084 * the module calling this has to destroy the cache before getting unloaded. 2084 * the module calling this has to destroy the cache before getting unloaded.
2085 * Note that kmem_cache_name() is not guaranteed to return the same pointer, 2085 * Note that kmem_cache_name() is not guaranteed to return the same pointer,
2086 * therefore applications must manage it themselves. 2086 * therefore applications must manage it themselves.
2087 * 2087 *
2088 * The flags are 2088 * The flags are
2089 * 2089 *
2090 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 2090 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
2091 * to catch references to uninitialised memory. 2091 * to catch references to uninitialised memory.
2092 * 2092 *
2093 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check 2093 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
2094 * for buffer overruns. 2094 * for buffer overruns.
2095 * 2095 *
2096 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 2096 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
2097 * cacheline. This can be beneficial if you're counting cycles as closely 2097 * cacheline. This can be beneficial if you're counting cycles as closely
2098 * as davem. 2098 * as davem.
2099 */ 2099 */
2100 struct kmem_cache * 2100 struct kmem_cache *
2101 kmem_cache_create (const char *name, size_t size, size_t align, 2101 kmem_cache_create (const char *name, size_t size, size_t align,
2102 unsigned long flags, void (*ctor)(void *)) 2102 unsigned long flags, void (*ctor)(void *))
2103 { 2103 {
2104 size_t left_over, slab_size, ralign; 2104 size_t left_over, slab_size, ralign;
2105 struct kmem_cache *cachep = NULL, *pc; 2105 struct kmem_cache *cachep = NULL, *pc;
2106 gfp_t gfp; 2106 gfp_t gfp;
2107 2107
2108 /* 2108 /*
2109 * Sanity checks... these are all serious usage bugs. 2109 * Sanity checks... these are all serious usage bugs.
2110 */ 2110 */
2111 if (!name || in_interrupt() || (size < BYTES_PER_WORD) || 2111 if (!name || in_interrupt() || (size < BYTES_PER_WORD) ||
2112 size > KMALLOC_MAX_SIZE) { 2112 size > KMALLOC_MAX_SIZE) {
2113 printk(KERN_ERR "%s: Early error in slab %s\n", __func__, 2113 printk(KERN_ERR "%s: Early error in slab %s\n", __func__,
2114 name); 2114 name);
2115 BUG(); 2115 BUG();
2116 } 2116 }
2117 2117
2118 /* 2118 /*
2119 * We use cache_chain_mutex to ensure a consistent view of 2119 * We use cache_chain_mutex to ensure a consistent view of
2120 * cpu_online_mask as well. Please see cpuup_callback 2120 * cpu_online_mask as well. Please see cpuup_callback
2121 */ 2121 */
2122 if (slab_is_available()) { 2122 if (slab_is_available()) {
2123 get_online_cpus(); 2123 get_online_cpus();
2124 mutex_lock(&cache_chain_mutex); 2124 mutex_lock(&cache_chain_mutex);
2125 } 2125 }
2126 2126
2127 list_for_each_entry(pc, &cache_chain, next) { 2127 list_for_each_entry(pc, &cache_chain, next) {
2128 char tmp; 2128 char tmp;
2129 int res; 2129 int res;
2130 2130
2131 /* 2131 /*
2132 * This happens when the module gets unloaded and doesn't 2132 * This happens when the module gets unloaded and doesn't
2133 * destroy its slab cache and no-one else reuses the vmalloc 2133 * destroy its slab cache and no-one else reuses the vmalloc
2134 * area of the module. Print a warning. 2134 * area of the module. Print a warning.
2135 */ 2135 */
2136 res = probe_kernel_address(pc->name, tmp); 2136 res = probe_kernel_address(pc->name, tmp);
2137 if (res) { 2137 if (res) {
2138 printk(KERN_ERR 2138 printk(KERN_ERR
2139 "SLAB: cache with size %d has lost its name\n", 2139 "SLAB: cache with size %d has lost its name\n",
2140 pc->buffer_size); 2140 pc->buffer_size);
2141 continue; 2141 continue;
2142 } 2142 }
2143 2143
2144 if (!strcmp(pc->name, name)) { 2144 if (!strcmp(pc->name, name)) {
2145 printk(KERN_ERR 2145 printk(KERN_ERR
2146 "kmem_cache_create: duplicate cache %s\n", name); 2146 "kmem_cache_create: duplicate cache %s\n", name);
2147 dump_stack(); 2147 dump_stack();
2148 goto oops; 2148 goto oops;
2149 } 2149 }
2150 } 2150 }
2151 2151
2152 #if DEBUG 2152 #if DEBUG
2153 WARN_ON(strchr(name, ' ')); /* It confuses parsers */ 2153 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
2154 #if FORCED_DEBUG 2154 #if FORCED_DEBUG
2155 /* 2155 /*
2156 * Enable redzoning and last user accounting, except for caches with 2156 * Enable redzoning and last user accounting, except for caches with
2157 * large objects, if the increased size would increase the object size 2157 * large objects, if the increased size would increase the object size
2158 * above the next power of two: caches with object sizes just above a 2158 * above the next power of two: caches with object sizes just above a
2159 * power of two have a significant amount of internal fragmentation. 2159 * power of two have a significant amount of internal fragmentation.
2160 */ 2160 */
2161 if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + 2161 if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN +
2162 2 * sizeof(unsigned long long))) 2162 2 * sizeof(unsigned long long)))
2163 flags |= SLAB_RED_ZONE | SLAB_STORE_USER; 2163 flags |= SLAB_RED_ZONE | SLAB_STORE_USER;
2164 if (!(flags & SLAB_DESTROY_BY_RCU)) 2164 if (!(flags & SLAB_DESTROY_BY_RCU))
2165 flags |= SLAB_POISON; 2165 flags |= SLAB_POISON;
2166 #endif 2166 #endif
2167 if (flags & SLAB_DESTROY_BY_RCU) 2167 if (flags & SLAB_DESTROY_BY_RCU)
2168 BUG_ON(flags & SLAB_POISON); 2168 BUG_ON(flags & SLAB_POISON);
2169 #endif 2169 #endif
2170 /* 2170 /*
2171 * Always checks flags, a caller might be expecting debug support which 2171 * Always checks flags, a caller might be expecting debug support which
2172 * isn't available. 2172 * isn't available.
2173 */ 2173 */
2174 BUG_ON(flags & ~CREATE_MASK); 2174 BUG_ON(flags & ~CREATE_MASK);
2175 2175
2176 /* 2176 /*
2177 * Check that size is in terms of words. This is needed to avoid 2177 * Check that size is in terms of words. This is needed to avoid
2178 * unaligned accesses for some archs when redzoning is used, and makes 2178 * unaligned accesses for some archs when redzoning is used, and makes
2179 * sure any on-slab bufctl's are also correctly aligned. 2179 * sure any on-slab bufctl's are also correctly aligned.
2180 */ 2180 */
2181 if (size & (BYTES_PER_WORD - 1)) { 2181 if (size & (BYTES_PER_WORD - 1)) {
2182 size += (BYTES_PER_WORD - 1); 2182 size += (BYTES_PER_WORD - 1);
2183 size &= ~(BYTES_PER_WORD - 1); 2183 size &= ~(BYTES_PER_WORD - 1);
2184 } 2184 }
2185 2185
2186 /* calculate the final buffer alignment: */ 2186 /* calculate the final buffer alignment: */
2187 2187
2188 /* 1) arch recommendation: can be overridden for debug */ 2188 /* 1) arch recommendation: can be overridden for debug */
2189 if (flags & SLAB_HWCACHE_ALIGN) { 2189 if (flags & SLAB_HWCACHE_ALIGN) {
2190 /* 2190 /*
2191 * Default alignment: as specified by the arch code. Except if 2191 * Default alignment: as specified by the arch code. Except if
2192 * an object is really small, then squeeze multiple objects into 2192 * an object is really small, then squeeze multiple objects into
2193 * one cacheline. 2193 * one cacheline.
2194 */ 2194 */
2195 ralign = cache_line_size(); 2195 ralign = cache_line_size();
2196 while (size <= ralign / 2) 2196 while (size <= ralign / 2)
2197 ralign /= 2; 2197 ralign /= 2;
2198 } else { 2198 } else {
2199 ralign = BYTES_PER_WORD; 2199 ralign = BYTES_PER_WORD;
2200 } 2200 }
2201 2201
2202 /* 2202 /*
2203 * Redzoning and user store require word alignment or possibly larger. 2203 * Redzoning and user store require word alignment or possibly larger.
2204 * Note this will be overridden by architecture or caller mandated 2204 * Note this will be overridden by architecture or caller mandated
2205 * alignment if either is greater than BYTES_PER_WORD. 2205 * alignment if either is greater than BYTES_PER_WORD.
2206 */ 2206 */
2207 if (flags & SLAB_STORE_USER) 2207 if (flags & SLAB_STORE_USER)
2208 ralign = BYTES_PER_WORD; 2208 ralign = BYTES_PER_WORD;
2209 2209
2210 if (flags & SLAB_RED_ZONE) { 2210 if (flags & SLAB_RED_ZONE) {
2211 ralign = REDZONE_ALIGN; 2211 ralign = REDZONE_ALIGN;
2212 /* If redzoning, ensure that the second redzone is suitably 2212 /* If redzoning, ensure that the second redzone is suitably
2213 * aligned, by adjusting the object size accordingly. */ 2213 * aligned, by adjusting the object size accordingly. */
2214 size += REDZONE_ALIGN - 1; 2214 size += REDZONE_ALIGN - 1;
2215 size &= ~(REDZONE_ALIGN - 1); 2215 size &= ~(REDZONE_ALIGN - 1);
2216 } 2216 }
2217 2217
2218 /* 2) arch mandated alignment */ 2218 /* 2) arch mandated alignment */
2219 if (ralign < ARCH_SLAB_MINALIGN) { 2219 if (ralign < ARCH_SLAB_MINALIGN) {
2220 ralign = ARCH_SLAB_MINALIGN; 2220 ralign = ARCH_SLAB_MINALIGN;
2221 } 2221 }
2222 /* 3) caller mandated alignment */ 2222 /* 3) caller mandated alignment */
2223 if (ralign < align) { 2223 if (ralign < align) {
2224 ralign = align; 2224 ralign = align;
2225 } 2225 }
2226 /* disable debug if necessary */ 2226 /* disable debug if necessary */
2227 if (ralign > __alignof__(unsigned long long)) 2227 if (ralign > __alignof__(unsigned long long))
2228 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); 2228 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
2229 /* 2229 /*
2230 * 4) Store it. 2230 * 4) Store it.
2231 */ 2231 */
2232 align = ralign; 2232 align = ralign;
2233 2233
2234 if (slab_is_available()) 2234 if (slab_is_available())
2235 gfp = GFP_KERNEL; 2235 gfp = GFP_KERNEL;
2236 else 2236 else
2237 gfp = GFP_NOWAIT; 2237 gfp = GFP_NOWAIT;
2238 2238
2239 /* Get cache's description obj. */ 2239 /* Get cache's description obj. */
2240 cachep = kmem_cache_zalloc(&cache_cache, gfp); 2240 cachep = kmem_cache_zalloc(&cache_cache, gfp);
2241 if (!cachep) 2241 if (!cachep)
2242 goto oops; 2242 goto oops;
2243 2243
2244 #if DEBUG 2244 #if DEBUG
2245 cachep->obj_size = size; 2245 cachep->obj_size = size;
2246 2246
2247 /* 2247 /*
2248 * Both debugging options require word-alignment which is calculated 2248 * Both debugging options require word-alignment which is calculated
2249 * into align above. 2249 * into align above.
2250 */ 2250 */
2251 if (flags & SLAB_RED_ZONE) { 2251 if (flags & SLAB_RED_ZONE) {
2252 /* add space for red zone words */ 2252 /* add space for red zone words */
2253 cachep->obj_offset += sizeof(unsigned long long); 2253 cachep->obj_offset += sizeof(unsigned long long);
2254 size += 2 * sizeof(unsigned long long); 2254 size += 2 * sizeof(unsigned long long);
2255 } 2255 }
2256 if (flags & SLAB_STORE_USER) { 2256 if (flags & SLAB_STORE_USER) {
2257 /* user store requires one word storage behind the end of 2257 /* user store requires one word storage behind the end of
2258 * the real object. But if the second red zone needs to be 2258 * the real object. But if the second red zone needs to be
2259 * aligned to 64 bits, we must allow that much space. 2259 * aligned to 64 bits, we must allow that much space.
2260 */ 2260 */
2261 if (flags & SLAB_RED_ZONE) 2261 if (flags & SLAB_RED_ZONE)
2262 size += REDZONE_ALIGN; 2262 size += REDZONE_ALIGN;
2263 else 2263 else
2264 size += BYTES_PER_WORD; 2264 size += BYTES_PER_WORD;
2265 } 2265 }
2266 #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) 2266 #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC)
2267 if (size >= malloc_sizes[INDEX_L3 + 1].cs_size 2267 if (size >= malloc_sizes[INDEX_L3 + 1].cs_size
2268 && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { 2268 && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) {
2269 cachep->obj_offset += PAGE_SIZE - size; 2269 cachep->obj_offset += PAGE_SIZE - size;
2270 size = PAGE_SIZE; 2270 size = PAGE_SIZE;
2271 } 2271 }
2272 #endif 2272 #endif
2273 #endif 2273 #endif
2274 2274
2275 /* 2275 /*
2276 * Determine if the slab management is 'on' or 'off' slab. 2276 * Determine if the slab management is 'on' or 'off' slab.
2277 * (bootstrapping cannot cope with offslab caches so don't do 2277 * (bootstrapping cannot cope with offslab caches so don't do
2278 * it too early on. Always use on-slab management when 2278 * it too early on. Always use on-slab management when
2279 * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) 2279 * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak)
2280 */ 2280 */
2281 if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init && 2281 if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init &&
2282 !(flags & SLAB_NOLEAKTRACE)) 2282 !(flags & SLAB_NOLEAKTRACE))
2283 /* 2283 /*
2284 * Size is large, assume best to place the slab management obj 2284 * Size is large, assume best to place the slab management obj
2285 * off-slab (should allow better packing of objs). 2285 * off-slab (should allow better packing of objs).
2286 */ 2286 */
2287 flags |= CFLGS_OFF_SLAB; 2287 flags |= CFLGS_OFF_SLAB;
2288 2288
2289 size = ALIGN(size, align); 2289 size = ALIGN(size, align);
2290 2290
2291 left_over = calculate_slab_order(cachep, size, align, flags); 2291 left_over = calculate_slab_order(cachep, size, align, flags);
2292 2292
2293 if (!cachep->num) { 2293 if (!cachep->num) {
2294 printk(KERN_ERR 2294 printk(KERN_ERR
2295 "kmem_cache_create: couldn't create cache %s.\n", name); 2295 "kmem_cache_create: couldn't create cache %s.\n", name);
2296 kmem_cache_free(&cache_cache, cachep); 2296 kmem_cache_free(&cache_cache, cachep);
2297 cachep = NULL; 2297 cachep = NULL;
2298 goto oops; 2298 goto oops;
2299 } 2299 }
2300 slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) 2300 slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t)
2301 + sizeof(struct slab), align); 2301 + sizeof(struct slab), align);
2302 2302
2303 /* 2303 /*
2304 * If the slab has been placed off-slab, and we have enough space then 2304 * If the slab has been placed off-slab, and we have enough space then
2305 * move it on-slab. This is at the expense of any extra colouring. 2305 * move it on-slab. This is at the expense of any extra colouring.
2306 */ 2306 */
2307 if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { 2307 if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) {
2308 flags &= ~CFLGS_OFF_SLAB; 2308 flags &= ~CFLGS_OFF_SLAB;
2309 left_over -= slab_size; 2309 left_over -= slab_size;
2310 } 2310 }
2311 2311
2312 if (flags & CFLGS_OFF_SLAB) { 2312 if (flags & CFLGS_OFF_SLAB) {
2313 /* really off slab. No need for manual alignment */ 2313 /* really off slab. No need for manual alignment */
2314 slab_size = 2314 slab_size =
2315 cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); 2315 cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab);
2316 2316
2317 #ifdef CONFIG_PAGE_POISONING 2317 #ifdef CONFIG_PAGE_POISONING
2318 /* If we're going to use the generic kernel_map_pages() 2318 /* If we're going to use the generic kernel_map_pages()
2319 * poisoning, then it's going to smash the contents of 2319 * poisoning, then it's going to smash the contents of
2320 * the redzone and userword anyhow, so switch them off. 2320 * the redzone and userword anyhow, so switch them off.
2321 */ 2321 */
2322 if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) 2322 if (size % PAGE_SIZE == 0 && flags & SLAB_POISON)
2323 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); 2323 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
2324 #endif 2324 #endif
2325 } 2325 }
2326 2326
2327 cachep->colour_off = cache_line_size(); 2327 cachep->colour_off = cache_line_size();
2328 /* Offset must be a multiple of the alignment. */ 2328 /* Offset must be a multiple of the alignment. */
2329 if (cachep->colour_off < align) 2329 if (cachep->colour_off < align)
2330 cachep->colour_off = align; 2330 cachep->colour_off = align;
2331 cachep->colour = left_over / cachep->colour_off; 2331 cachep->colour = left_over / cachep->colour_off;
2332 cachep->slab_size = slab_size; 2332 cachep->slab_size = slab_size;
2333 cachep->flags = flags; 2333 cachep->flags = flags;
2334 cachep->gfpflags = 0; 2334 cachep->gfpflags = 0;
2335 if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) 2335 if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA))
2336 cachep->gfpflags |= GFP_DMA; 2336 cachep->gfpflags |= GFP_DMA;
2337 cachep->buffer_size = size; 2337 cachep->buffer_size = size;
2338 cachep->reciprocal_buffer_size = reciprocal_value(size); 2338 cachep->reciprocal_buffer_size = reciprocal_value(size);
2339 2339
2340 if (flags & CFLGS_OFF_SLAB) { 2340 if (flags & CFLGS_OFF_SLAB) {
2341 cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); 2341 cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u);
2342 /* 2342 /*
2343 * This is a possibility for one of the malloc_sizes caches. 2343 * This is a possibility for one of the malloc_sizes caches.
2344 * But since we go off slab only for object size greater than 2344 * But since we go off slab only for object size greater than
2345 * PAGE_SIZE/8, and malloc_sizes gets created in ascending order, 2345 * PAGE_SIZE/8, and malloc_sizes gets created in ascending order,
2346 * this should not happen at all. 2346 * this should not happen at all.
2347 * But leave a BUG_ON for some lucky dude. 2347 * But leave a BUG_ON for some lucky dude.
2348 */ 2348 */
2349 BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache)); 2349 BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache));
2350 } 2350 }
2351 cachep->ctor = ctor; 2351 cachep->ctor = ctor;
2352 cachep->name = name; 2352 cachep->name = name;
2353 2353
2354 if (setup_cpu_cache(cachep, gfp)) { 2354 if (setup_cpu_cache(cachep, gfp)) {
2355 __kmem_cache_destroy(cachep); 2355 __kmem_cache_destroy(cachep);
2356 cachep = NULL; 2356 cachep = NULL;
2357 goto oops; 2357 goto oops;
2358 } 2358 }
2359 2359
2360 /* cache setup completed, link it into the list */ 2360 /* cache setup completed, link it into the list */
2361 list_add(&cachep->next, &cache_chain); 2361 list_add(&cachep->next, &cache_chain);
2362 oops: 2362 oops:
2363 if (!cachep && (flags & SLAB_PANIC)) 2363 if (!cachep && (flags & SLAB_PANIC))
2364 panic("kmem_cache_create(): failed to create slab `%s'\n", 2364 panic("kmem_cache_create(): failed to create slab `%s'\n",
2365 name); 2365 name);
2366 if (slab_is_available()) { 2366 if (slab_is_available()) {
2367 mutex_unlock(&cache_chain_mutex); 2367 mutex_unlock(&cache_chain_mutex);
2368 put_online_cpus(); 2368 put_online_cpus();
2369 } 2369 }
2370 return cachep; 2370 return cachep;
2371 } 2371 }
2372 EXPORT_SYMBOL(kmem_cache_create); 2372 EXPORT_SYMBOL(kmem_cache_create);
2373 2373
2374 #if DEBUG 2374 #if DEBUG
2375 static void check_irq_off(void) 2375 static void check_irq_off(void)
2376 { 2376 {
2377 BUG_ON(!irqs_disabled()); 2377 BUG_ON(!irqs_disabled());
2378 } 2378 }
2379 2379
2380 static void check_irq_on(void) 2380 static void check_irq_on(void)
2381 { 2381 {
2382 BUG_ON(irqs_disabled()); 2382 BUG_ON(irqs_disabled());
2383 } 2383 }
2384 2384
2385 static void check_spinlock_acquired(struct kmem_cache *cachep) 2385 static void check_spinlock_acquired(struct kmem_cache *cachep)
2386 { 2386 {
2387 #ifdef CONFIG_SMP 2387 #ifdef CONFIG_SMP
2388 check_irq_off(); 2388 check_irq_off();
2389 assert_spin_locked(&cachep->nodelists[numa_node_id()]->list_lock); 2389 assert_spin_locked(&cachep->nodelists[numa_node_id()]->list_lock);
2390 #endif 2390 #endif
2391 } 2391 }
2392 2392
2393 static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) 2393 static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node)
2394 { 2394 {
2395 #ifdef CONFIG_SMP 2395 #ifdef CONFIG_SMP
2396 check_irq_off(); 2396 check_irq_off();
2397 assert_spin_locked(&cachep->nodelists[node]->list_lock); 2397 assert_spin_locked(&cachep->nodelists[node]->list_lock);
2398 #endif 2398 #endif
2399 } 2399 }
2400 2400
2401 #else 2401 #else
2402 #define check_irq_off() do { } while(0) 2402 #define check_irq_off() do { } while(0)
2403 #define check_irq_on() do { } while(0) 2403 #define check_irq_on() do { } while(0)
2404 #define check_spinlock_acquired(x) do { } while(0) 2404 #define check_spinlock_acquired(x) do { } while(0)
2405 #define check_spinlock_acquired_node(x, y) do { } while(0) 2405 #define check_spinlock_acquired_node(x, y) do { } while(0)
2406 #endif 2406 #endif
2407 2407
2408 static void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, 2408 static void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3,
2409 struct array_cache *ac, 2409 struct array_cache *ac,
2410 int force, int node); 2410 int force, int node);
2411 2411
2412 static void do_drain(void *arg) 2412 static void do_drain(void *arg)
2413 { 2413 {
2414 struct kmem_cache *cachep = arg; 2414 struct kmem_cache *cachep = arg;
2415 struct array_cache *ac; 2415 struct array_cache *ac;
2416 int node = numa_node_id(); 2416 int node = numa_node_id();
2417 2417
2418 check_irq_off(); 2418 check_irq_off();
2419 ac = cpu_cache_get(cachep); 2419 ac = cpu_cache_get(cachep);
2420 spin_lock(&cachep->nodelists[node]->list_lock); 2420 spin_lock(&cachep->nodelists[node]->list_lock);
2421 free_block(cachep, ac->entry, ac->avail, node); 2421 free_block(cachep, ac->entry, ac->avail, node);
2422 spin_unlock(&cachep->nodelists[node]->list_lock); 2422 spin_unlock(&cachep->nodelists[node]->list_lock);
2423 ac->avail = 0; 2423 ac->avail = 0;
2424 } 2424 }
2425 2425
2426 static void drain_cpu_caches(struct kmem_cache *cachep) 2426 static void drain_cpu_caches(struct kmem_cache *cachep)
2427 { 2427 {
2428 struct kmem_list3 *l3; 2428 struct kmem_list3 *l3;
2429 int node; 2429 int node;
2430 2430
2431 on_each_cpu(do_drain, cachep, 1); 2431 on_each_cpu(do_drain, cachep, 1);
2432 check_irq_on(); 2432 check_irq_on();
2433 for_each_online_node(node) { 2433 for_each_online_node(node) {
2434 l3 = cachep->nodelists[node]; 2434 l3 = cachep->nodelists[node];
2435 if (l3 && l3->alien) 2435 if (l3 && l3->alien)
2436 drain_alien_cache(cachep, l3->alien); 2436 drain_alien_cache(cachep, l3->alien);
2437 } 2437 }
2438 2438
2439 for_each_online_node(node) { 2439 for_each_online_node(node) {
2440 l3 = cachep->nodelists[node]; 2440 l3 = cachep->nodelists[node];
2441 if (l3) 2441 if (l3)
2442 drain_array(cachep, l3, l3->shared, 1, node); 2442 drain_array(cachep, l3, l3->shared, 1, node);
2443 } 2443 }
2444 } 2444 }
2445 2445
2446 /* 2446 /*
2447 * Remove slabs from the list of free slabs. 2447 * Remove slabs from the list of free slabs.
2448 * Specify the number of slabs to drain in tofree. 2448 * Specify the number of slabs to drain in tofree.
2449 * 2449 *
2450 * Returns the actual number of slabs released. 2450 * Returns the actual number of slabs released.
2451 */ 2451 */
2452 static int drain_freelist(struct kmem_cache *cache, 2452 static int drain_freelist(struct kmem_cache *cache,
2453 struct kmem_list3 *l3, int tofree) 2453 struct kmem_list3 *l3, int tofree)
2454 { 2454 {
2455 struct list_head *p; 2455 struct list_head *p;
2456 int nr_freed; 2456 int nr_freed;
2457 struct slab *slabp; 2457 struct slab *slabp;
2458 2458
2459 nr_freed = 0; 2459 nr_freed = 0;
2460 while (nr_freed < tofree && !list_empty(&l3->slabs_free)) { 2460 while (nr_freed < tofree && !list_empty(&l3->slabs_free)) {
2461 2461
2462 spin_lock_irq(&l3->list_lock); 2462 spin_lock_irq(&l3->list_lock);
2463 p = l3->slabs_free.prev; 2463 p = l3->slabs_free.prev;
2464 if (p == &l3->slabs_free) { 2464 if (p == &l3->slabs_free) {
2465 spin_unlock_irq(&l3->list_lock); 2465 spin_unlock_irq(&l3->list_lock);
2466 goto out; 2466 goto out;
2467 } 2467 }
2468 2468
2469 slabp = list_entry(p, struct slab, list); 2469 slabp = list_entry(p, struct slab, list);
2470 #if DEBUG 2470 #if DEBUG
2471 BUG_ON(slabp->inuse); 2471 BUG_ON(slabp->inuse);
2472 #endif 2472 #endif
2473 list_del(&slabp->list); 2473 list_del(&slabp->list);
2474 /* 2474 /*
2475 * Safe to drop the lock. The slab is no longer linked 2475 * Safe to drop the lock. The slab is no longer linked
2476 * to the cache. 2476 * to the cache.
2477 */ 2477 */
2478 l3->free_objects -= cache->num; 2478 l3->free_objects -= cache->num;
2479 spin_unlock_irq(&l3->list_lock); 2479 spin_unlock_irq(&l3->list_lock);
2480 slab_destroy(cache, slabp); 2480 slab_destroy(cache, slabp);
2481 nr_freed++; 2481 nr_freed++;
2482 } 2482 }
2483 out: 2483 out:
2484 return nr_freed; 2484 return nr_freed;
2485 } 2485 }
2486 2486
2487 /* Called with cache_chain_mutex held to protect against cpu hotplug */ 2487 /* Called with cache_chain_mutex held to protect against cpu hotplug */
2488 static int __cache_shrink(struct kmem_cache *cachep) 2488 static int __cache_shrink(struct kmem_cache *cachep)
2489 { 2489 {
2490 int ret = 0, i = 0; 2490 int ret = 0, i = 0;
2491 struct kmem_list3 *l3; 2491 struct kmem_list3 *l3;
2492 2492
2493 drain_cpu_caches(cachep); 2493 drain_cpu_caches(cachep);
2494 2494
2495 check_irq_on(); 2495 check_irq_on();
2496 for_each_online_node(i) { 2496 for_each_online_node(i) {
2497 l3 = cachep->nodelists[i]; 2497 l3 = cachep->nodelists[i];
2498 if (!l3) 2498 if (!l3)
2499 continue; 2499 continue;
2500 2500
2501 drain_freelist(cachep, l3, l3->free_objects); 2501 drain_freelist(cachep, l3, l3->free_objects);
2502 2502
2503 ret += !list_empty(&l3->slabs_full) || 2503 ret += !list_empty(&l3->slabs_full) ||
2504 !list_empty(&l3->slabs_partial); 2504 !list_empty(&l3->slabs_partial);
2505 } 2505 }
2506 return (ret ? 1 : 0); 2506 return (ret ? 1 : 0);
2507 } 2507 }
2508 2508
2509 /** 2509 /**
2510 * kmem_cache_shrink - Shrink a cache. 2510 * kmem_cache_shrink - Shrink a cache.
2511 * @cachep: The cache to shrink. 2511 * @cachep: The cache to shrink.
2512 * 2512 *
2513 * Releases as many slabs as possible for a cache. 2513 * Releases as many slabs as possible for a cache.
2514 * To help debugging, a zero exit status indicates all slabs were released. 2514 * To help debugging, a zero exit status indicates all slabs were released.
2515 */ 2515 */
2516 int kmem_cache_shrink(struct kmem_cache *cachep) 2516 int kmem_cache_shrink(struct kmem_cache *cachep)
2517 { 2517 {
2518 int ret; 2518 int ret;
2519 BUG_ON(!cachep || in_interrupt()); 2519 BUG_ON(!cachep || in_interrupt());
2520 2520
2521 get_online_cpus(); 2521 get_online_cpus();
2522 mutex_lock(&cache_chain_mutex); 2522 mutex_lock(&cache_chain_mutex);
2523 ret = __cache_shrink(cachep); 2523 ret = __cache_shrink(cachep);
2524 mutex_unlock(&cache_chain_mutex); 2524 mutex_unlock(&cache_chain_mutex);
2525 put_online_cpus(); 2525 put_online_cpus();
2526 return ret; 2526 return ret;
2527 } 2527 }
2528 EXPORT_SYMBOL(kmem_cache_shrink); 2528 EXPORT_SYMBOL(kmem_cache_shrink);
2529 2529
2530 /** 2530 /**
2531 * kmem_cache_destroy - delete a cache 2531 * kmem_cache_destroy - delete a cache
2532 * @cachep: the cache to destroy 2532 * @cachep: the cache to destroy
2533 * 2533 *
2534 * Remove a &struct kmem_cache object from the slab cache. 2534 * Remove a &struct kmem_cache object from the slab cache.
2535 * 2535 *
2536 * It is expected this function will be called by a module when it is 2536 * It is expected this function will be called by a module when it is
2537 * unloaded. This will remove the cache completely, and avoid a duplicate 2537 * unloaded. This will remove the cache completely, and avoid a duplicate
2538 * cache being allocated each time a module is loaded and unloaded, if the 2538 * cache being allocated each time a module is loaded and unloaded, if the
2539 * module doesn't have persistent in-kernel storage across loads and unloads. 2539 * module doesn't have persistent in-kernel storage across loads and unloads.
2540 * 2540 *
2541 * The cache must be empty before calling this function. 2541 * The cache must be empty before calling this function.
2542 * 2542 *
2543 * The caller must guarantee that noone will allocate memory from the cache 2543 * The caller must guarantee that noone will allocate memory from the cache
2544 * during the kmem_cache_destroy(). 2544 * during the kmem_cache_destroy().
2545 */ 2545 */
2546 void kmem_cache_destroy(struct kmem_cache *cachep) 2546 void kmem_cache_destroy(struct kmem_cache *cachep)
2547 { 2547 {
2548 BUG_ON(!cachep || in_interrupt()); 2548 BUG_ON(!cachep || in_interrupt());
2549 2549
2550 /* Find the cache in the chain of caches. */ 2550 /* Find the cache in the chain of caches. */
2551 get_online_cpus(); 2551 get_online_cpus();
2552 mutex_lock(&cache_chain_mutex); 2552 mutex_lock(&cache_chain_mutex);
2553 /* 2553 /*
2554 * the chain is never empty, cache_cache is never destroyed 2554 * the chain is never empty, cache_cache is never destroyed
2555 */ 2555 */
2556 list_del(&cachep->next); 2556 list_del(&cachep->next);
2557 if (__cache_shrink(cachep)) { 2557 if (__cache_shrink(cachep)) {
2558 slab_error(cachep, "Can't free all objects"); 2558 slab_error(cachep, "Can't free all objects");
2559 list_add(&cachep->next, &cache_chain); 2559 list_add(&cachep->next, &cache_chain);
2560 mutex_unlock(&cache_chain_mutex); 2560 mutex_unlock(&cache_chain_mutex);
2561 put_online_cpus(); 2561 put_online_cpus();
2562 return; 2562 return;
2563 } 2563 }
2564 2564
2565 if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) 2565 if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU))
2566 rcu_barrier(); 2566 rcu_barrier();
2567 2567
2568 __kmem_cache_destroy(cachep); 2568 __kmem_cache_destroy(cachep);
2569 mutex_unlock(&cache_chain_mutex); 2569 mutex_unlock(&cache_chain_mutex);
2570 put_online_cpus(); 2570 put_online_cpus();
2571 } 2571 }
2572 EXPORT_SYMBOL(kmem_cache_destroy); 2572 EXPORT_SYMBOL(kmem_cache_destroy);
2573 2573
2574 /* 2574 /*
2575 * Get the memory for a slab management obj. 2575 * Get the memory for a slab management obj.
2576 * For a slab cache when the slab descriptor is off-slab, slab descriptors 2576 * For a slab cache when the slab descriptor is off-slab, slab descriptors
2577 * always come from malloc_sizes caches. The slab descriptor cannot 2577 * always come from malloc_sizes caches. The slab descriptor cannot
2578 * come from the same cache which is getting created because, 2578 * come from the same cache which is getting created because,
2579 * when we are searching for an appropriate cache for these 2579 * when we are searching for an appropriate cache for these
2580 * descriptors in kmem_cache_create, we search through the malloc_sizes array. 2580 * descriptors in kmem_cache_create, we search through the malloc_sizes array.
2581 * If we are creating a malloc_sizes cache here it would not be visible to 2581 * If we are creating a malloc_sizes cache here it would not be visible to
2582 * kmem_find_general_cachep till the initialization is complete. 2582 * kmem_find_general_cachep till the initialization is complete.
2583 * Hence we cannot have slabp_cache same as the original cache. 2583 * Hence we cannot have slabp_cache same as the original cache.
2584 */ 2584 */
2585 static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, 2585 static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp,
2586 int colour_off, gfp_t local_flags, 2586 int colour_off, gfp_t local_flags,
2587 int nodeid) 2587 int nodeid)
2588 { 2588 {
2589 struct slab *slabp; 2589 struct slab *slabp;
2590 2590
2591 if (OFF_SLAB(cachep)) { 2591 if (OFF_SLAB(cachep)) {
2592 /* Slab management obj is off-slab. */ 2592 /* Slab management obj is off-slab. */
2593 slabp = kmem_cache_alloc_node(cachep->slabp_cache, 2593 slabp = kmem_cache_alloc_node(cachep->slabp_cache,
2594 local_flags, nodeid); 2594 local_flags, nodeid);
2595 /* 2595 /*
2596 * If the first object in the slab is leaked (it's allocated 2596 * If the first object in the slab is leaked (it's allocated
2597 * but no one has a reference to it), we want to make sure 2597 * but no one has a reference to it), we want to make sure
2598 * kmemleak does not treat the ->s_mem pointer as a reference 2598 * kmemleak does not treat the ->s_mem pointer as a reference
2599 * to the object. Otherwise we will not report the leak. 2599 * to the object. Otherwise we will not report the leak.
2600 */ 2600 */
2601 kmemleak_scan_area(&slabp->list, sizeof(struct list_head), 2601 kmemleak_scan_area(&slabp->list, sizeof(struct list_head),
2602 local_flags); 2602 local_flags);
2603 if (!slabp) 2603 if (!slabp)
2604 return NULL; 2604 return NULL;
2605 } else { 2605 } else {
2606 slabp = objp + colour_off; 2606 slabp = objp + colour_off;
2607 colour_off += cachep->slab_size; 2607 colour_off += cachep->slab_size;
2608 } 2608 }
2609 slabp->inuse = 0; 2609 slabp->inuse = 0;
2610 slabp->colouroff = colour_off; 2610 slabp->colouroff = colour_off;
2611 slabp->s_mem = objp + colour_off; 2611 slabp->s_mem = objp + colour_off;
2612 slabp->nodeid = nodeid; 2612 slabp->nodeid = nodeid;
2613 slabp->free = 0; 2613 slabp->free = 0;
2614 return slabp; 2614 return slabp;
2615 } 2615 }
2616 2616
2617 static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) 2617 static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp)
2618 { 2618 {
2619 return (kmem_bufctl_t *) (slabp + 1); 2619 return (kmem_bufctl_t *) (slabp + 1);
2620 } 2620 }
2621 2621
2622 static void cache_init_objs(struct kmem_cache *cachep, 2622 static void cache_init_objs(struct kmem_cache *cachep,
2623 struct slab *slabp) 2623 struct slab *slabp)
2624 { 2624 {
2625 int i; 2625 int i;
2626 2626
2627 for (i = 0; i < cachep->num; i++) { 2627 for (i = 0; i < cachep->num; i++) {
2628 void *objp = index_to_obj(cachep, slabp, i); 2628 void *objp = index_to_obj(cachep, slabp, i);
2629 #if DEBUG 2629 #if DEBUG
2630 /* need to poison the objs? */ 2630 /* need to poison the objs? */
2631 if (cachep->flags & SLAB_POISON) 2631 if (cachep->flags & SLAB_POISON)
2632 poison_obj(cachep, objp, POISON_FREE); 2632 poison_obj(cachep, objp, POISON_FREE);
2633 if (cachep->flags & SLAB_STORE_USER) 2633 if (cachep->flags & SLAB_STORE_USER)
2634 *dbg_userword(cachep, objp) = NULL; 2634 *dbg_userword(cachep, objp) = NULL;
2635 2635
2636 if (cachep->flags & SLAB_RED_ZONE) { 2636 if (cachep->flags & SLAB_RED_ZONE) {
2637 *dbg_redzone1(cachep, objp) = RED_INACTIVE; 2637 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2638 *dbg_redzone2(cachep, objp) = RED_INACTIVE; 2638 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2639 } 2639 }
2640 /* 2640 /*
2641 * Constructors are not allowed to allocate memory from the same 2641 * Constructors are not allowed to allocate memory from the same
2642 * cache which they are a constructor for. Otherwise, deadlock. 2642 * cache which they are a constructor for. Otherwise, deadlock.
2643 * They must also be threaded. 2643 * They must also be threaded.
2644 */ 2644 */
2645 if (cachep->ctor && !(cachep->flags & SLAB_POISON)) 2645 if (cachep->ctor && !(cachep->flags & SLAB_POISON))
2646 cachep->ctor(objp + obj_offset(cachep)); 2646 cachep->ctor(objp + obj_offset(cachep));
2647 2647
2648 if (cachep->flags & SLAB_RED_ZONE) { 2648 if (cachep->flags & SLAB_RED_ZONE) {
2649 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) 2649 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
2650 slab_error(cachep, "constructor overwrote the" 2650 slab_error(cachep, "constructor overwrote the"
2651 " end of an object"); 2651 " end of an object");
2652 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) 2652 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
2653 slab_error(cachep, "constructor overwrote the" 2653 slab_error(cachep, "constructor overwrote the"
2654 " start of an object"); 2654 " start of an object");
2655 } 2655 }
2656 if ((cachep->buffer_size % PAGE_SIZE) == 0 && 2656 if ((cachep->buffer_size % PAGE_SIZE) == 0 &&
2657 OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) 2657 OFF_SLAB(cachep) && cachep->flags & SLAB_POISON)
2658 kernel_map_pages(virt_to_page(objp), 2658 kernel_map_pages(virt_to_page(objp),
2659 cachep->buffer_size / PAGE_SIZE, 0); 2659 cachep->buffer_size / PAGE_SIZE, 0);
2660 #else 2660 #else
2661 if (cachep->ctor) 2661 if (cachep->ctor)
2662 cachep->ctor(objp); 2662 cachep->ctor(objp);
2663 #endif 2663 #endif
2664 slab_bufctl(slabp)[i] = i + 1; 2664 slab_bufctl(slabp)[i] = i + 1;
2665 } 2665 }
2666 slab_bufctl(slabp)[i - 1] = BUFCTL_END; 2666 slab_bufctl(slabp)[i - 1] = BUFCTL_END;
2667 } 2667 }
2668 2668
2669 static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) 2669 static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags)
2670 { 2670 {
2671 if (CONFIG_ZONE_DMA_FLAG) { 2671 if (CONFIG_ZONE_DMA_FLAG) {
2672 if (flags & GFP_DMA) 2672 if (flags & GFP_DMA)
2673 BUG_ON(!(cachep->gfpflags & GFP_DMA)); 2673 BUG_ON(!(cachep->gfpflags & GFP_DMA));
2674 else 2674 else
2675 BUG_ON(cachep->gfpflags & GFP_DMA); 2675 BUG_ON(cachep->gfpflags & GFP_DMA);
2676 } 2676 }
2677 } 2677 }
2678 2678
2679 static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, 2679 static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp,
2680 int nodeid) 2680 int nodeid)
2681 { 2681 {
2682 void *objp = index_to_obj(cachep, slabp, slabp->free); 2682 void *objp = index_to_obj(cachep, slabp, slabp->free);
2683 kmem_bufctl_t next; 2683 kmem_bufctl_t next;
2684 2684
2685 slabp->inuse++; 2685 slabp->inuse++;
2686 next = slab_bufctl(slabp)[slabp->free]; 2686 next = slab_bufctl(slabp)[slabp->free];
2687 #if DEBUG 2687 #if DEBUG
2688 slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; 2688 slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE;
2689 WARN_ON(slabp->nodeid != nodeid); 2689 WARN_ON(slabp->nodeid != nodeid);
2690 #endif 2690 #endif
2691 slabp->free = next; 2691 slabp->free = next;
2692 2692
2693 return objp; 2693 return objp;
2694 } 2694 }
2695 2695
2696 static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, 2696 static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp,
2697 void *objp, int nodeid) 2697 void *objp, int nodeid)
2698 { 2698 {
2699 unsigned int objnr = obj_to_index(cachep, slabp, objp); 2699 unsigned int objnr = obj_to_index(cachep, slabp, objp);
2700 2700
2701 #if DEBUG 2701 #if DEBUG
2702 /* Verify that the slab belongs to the intended node */ 2702 /* Verify that the slab belongs to the intended node */
2703 WARN_ON(slabp->nodeid != nodeid); 2703 WARN_ON(slabp->nodeid != nodeid);
2704 2704
2705 if (slab_bufctl(slabp)[objnr] + 1 <= SLAB_LIMIT + 1) { 2705 if (slab_bufctl(slabp)[objnr] + 1 <= SLAB_LIMIT + 1) {
2706 printk(KERN_ERR "slab: double free detected in cache " 2706 printk(KERN_ERR "slab: double free detected in cache "
2707 "'%s', objp %p\n", cachep->name, objp); 2707 "'%s', objp %p\n", cachep->name, objp);
2708 BUG(); 2708 BUG();
2709 } 2709 }
2710 #endif 2710 #endif
2711 slab_bufctl(slabp)[objnr] = slabp->free; 2711 slab_bufctl(slabp)[objnr] = slabp->free;
2712 slabp->free = objnr; 2712 slabp->free = objnr;
2713 slabp->inuse--; 2713 slabp->inuse--;
2714 } 2714 }
2715 2715
2716 /* 2716 /*
2717 * Map pages beginning at addr to the given cache and slab. This is required 2717 * Map pages beginning at addr to the given cache and slab. This is required
2718 * for the slab allocator to be able to lookup the cache and slab of a 2718 * for the slab allocator to be able to lookup the cache and slab of a
2719 * virtual address for kfree, ksize, kmem_ptr_validate, and slab debugging. 2719 * virtual address for kfree, ksize, kmem_ptr_validate, and slab debugging.
2720 */ 2720 */
2721 static void slab_map_pages(struct kmem_cache *cache, struct slab *slab, 2721 static void slab_map_pages(struct kmem_cache *cache, struct slab *slab,
2722 void *addr) 2722 void *addr)
2723 { 2723 {
2724 int nr_pages; 2724 int nr_pages;
2725 struct page *page; 2725 struct page *page;
2726 2726
2727 page = virt_to_page(addr); 2727 page = virt_to_page(addr);
2728 2728
2729 nr_pages = 1; 2729 nr_pages = 1;
2730 if (likely(!PageCompound(page))) 2730 if (likely(!PageCompound(page)))
2731 nr_pages <<= cache->gfporder; 2731 nr_pages <<= cache->gfporder;
2732 2732
2733 do { 2733 do {
2734 page_set_cache(page, cache); 2734 page_set_cache(page, cache);
2735 page_set_slab(page, slab); 2735 page_set_slab(page, slab);
2736 page++; 2736 page++;
2737 } while (--nr_pages); 2737 } while (--nr_pages);
2738 } 2738 }
2739 2739
2740 /* 2740 /*
2741 * Grow (by 1) the number of slabs within a cache. This is called by 2741 * Grow (by 1) the number of slabs within a cache. This is called by
2742 * kmem_cache_alloc() when there are no active objs left in a cache. 2742 * kmem_cache_alloc() when there are no active objs left in a cache.
2743 */ 2743 */
2744 static int cache_grow(struct kmem_cache *cachep, 2744 static int cache_grow(struct kmem_cache *cachep,
2745 gfp_t flags, int nodeid, void *objp) 2745 gfp_t flags, int nodeid, void *objp)
2746 { 2746 {
2747 struct slab *slabp; 2747 struct slab *slabp;
2748 size_t offset; 2748 size_t offset;
2749 gfp_t local_flags; 2749 gfp_t local_flags;
2750 struct kmem_list3 *l3; 2750 struct kmem_list3 *l3;
2751 2751
2752 /* 2752 /*
2753 * Be lazy and only check for valid flags here, keeping it out of the 2753 * Be lazy and only check for valid flags here, keeping it out of the
2754 * critical path in kmem_cache_alloc(). 2754 * critical path in kmem_cache_alloc().
2755 */ 2755 */
2756 BUG_ON(flags & GFP_SLAB_BUG_MASK); 2756 BUG_ON(flags & GFP_SLAB_BUG_MASK);
2757 local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); 2757 local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK);
2758 2758
2759 /* Take the l3 list lock to change the colour_next on this node */ 2759 /* Take the l3 list lock to change the colour_next on this node */
2760 check_irq_off(); 2760 check_irq_off();
2761 l3 = cachep->nodelists[nodeid]; 2761 l3 = cachep->nodelists[nodeid];
2762 spin_lock(&l3->list_lock); 2762 spin_lock(&l3->list_lock);
2763 2763
2764 /* Get colour for the slab, and cal the next value. */ 2764 /* Get colour for the slab, and cal the next value. */
2765 offset = l3->colour_next; 2765 offset = l3->colour_next;
2766 l3->colour_next++; 2766 l3->colour_next++;
2767 if (l3->colour_next >= cachep->colour) 2767 if (l3->colour_next >= cachep->colour)
2768 l3->colour_next = 0; 2768 l3->colour_next = 0;
2769 spin_unlock(&l3->list_lock); 2769 spin_unlock(&l3->list_lock);
2770 2770
2771 offset *= cachep->colour_off; 2771 offset *= cachep->colour_off;
2772 2772
2773 if (local_flags & __GFP_WAIT) 2773 if (local_flags & __GFP_WAIT)
2774 local_irq_enable(); 2774 local_irq_enable();
2775 2775
2776 /* 2776 /*
2777 * The test for missing atomic flag is performed here, rather than 2777 * The test for missing atomic flag is performed here, rather than
2778 * the more obvious place, simply to reduce the critical path length 2778 * the more obvious place, simply to reduce the critical path length
2779 * in kmem_cache_alloc(). If a caller is seriously mis-behaving they 2779 * in kmem_cache_alloc(). If a caller is seriously mis-behaving they
2780 * will eventually be caught here (where it matters). 2780 * will eventually be caught here (where it matters).
2781 */ 2781 */
2782 kmem_flagcheck(cachep, flags); 2782 kmem_flagcheck(cachep, flags);
2783 2783
2784 /* 2784 /*
2785 * Get mem for the objs. Attempt to allocate a physical page from 2785 * Get mem for the objs. Attempt to allocate a physical page from
2786 * 'nodeid'. 2786 * 'nodeid'.
2787 */ 2787 */
2788 if (!objp) 2788 if (!objp)
2789 objp = kmem_getpages(cachep, local_flags, nodeid); 2789 objp = kmem_getpages(cachep, local_flags, nodeid);
2790 if (!objp) 2790 if (!objp)
2791 goto failed; 2791 goto failed;
2792 2792
2793 /* Get slab management. */ 2793 /* Get slab management. */
2794 slabp = alloc_slabmgmt(cachep, objp, offset, 2794 slabp = alloc_slabmgmt(cachep, objp, offset,
2795 local_flags & ~GFP_CONSTRAINT_MASK, nodeid); 2795 local_flags & ~GFP_CONSTRAINT_MASK, nodeid);
2796 if (!slabp) 2796 if (!slabp)
2797 goto opps1; 2797 goto opps1;
2798 2798
2799 slab_map_pages(cachep, slabp, objp); 2799 slab_map_pages(cachep, slabp, objp);
2800 2800
2801 cache_init_objs(cachep, slabp); 2801 cache_init_objs(cachep, slabp);
2802 2802
2803 if (local_flags & __GFP_WAIT) 2803 if (local_flags & __GFP_WAIT)
2804 local_irq_disable(); 2804 local_irq_disable();
2805 check_irq_off(); 2805 check_irq_off();
2806 spin_lock(&l3->list_lock); 2806 spin_lock(&l3->list_lock);
2807 2807
2808 /* Make slab active. */ 2808 /* Make slab active. */
2809 list_add_tail(&slabp->list, &(l3->slabs_free)); 2809 list_add_tail(&slabp->list, &(l3->slabs_free));
2810 STATS_INC_GROWN(cachep); 2810 STATS_INC_GROWN(cachep);
2811 l3->free_objects += cachep->num; 2811 l3->free_objects += cachep->num;
2812 spin_unlock(&l3->list_lock); 2812 spin_unlock(&l3->list_lock);
2813 return 1; 2813 return 1;
2814 opps1: 2814 opps1:
2815 kmem_freepages(cachep, objp); 2815 kmem_freepages(cachep, objp);
2816 failed: 2816 failed:
2817 if (local_flags & __GFP_WAIT) 2817 if (local_flags & __GFP_WAIT)
2818 local_irq_disable(); 2818 local_irq_disable();
2819 return 0; 2819 return 0;
2820 } 2820 }
2821 2821
2822 #if DEBUG 2822 #if DEBUG
2823 2823
2824 /* 2824 /*
2825 * Perform extra freeing checks: 2825 * Perform extra freeing checks:
2826 * - detect bad pointers. 2826 * - detect bad pointers.
2827 * - POISON/RED_ZONE checking 2827 * - POISON/RED_ZONE checking
2828 */ 2828 */
2829 static void kfree_debugcheck(const void *objp) 2829 static void kfree_debugcheck(const void *objp)
2830 { 2830 {
2831 if (!virt_addr_valid(objp)) { 2831 if (!virt_addr_valid(objp)) {
2832 printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", 2832 printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n",
2833 (unsigned long)objp); 2833 (unsigned long)objp);
2834 BUG(); 2834 BUG();
2835 } 2835 }
2836 } 2836 }
2837 2837
2838 static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) 2838 static inline void verify_redzone_free(struct kmem_cache *cache, void *obj)
2839 { 2839 {
2840 unsigned long long redzone1, redzone2; 2840 unsigned long long redzone1, redzone2;
2841 2841
2842 redzone1 = *dbg_redzone1(cache, obj); 2842 redzone1 = *dbg_redzone1(cache, obj);
2843 redzone2 = *dbg_redzone2(cache, obj); 2843 redzone2 = *dbg_redzone2(cache, obj);
2844 2844
2845 /* 2845 /*
2846 * Redzone is ok. 2846 * Redzone is ok.
2847 */ 2847 */
2848 if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) 2848 if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE)
2849 return; 2849 return;
2850 2850
2851 if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) 2851 if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE)
2852 slab_error(cache, "double free detected"); 2852 slab_error(cache, "double free detected");
2853 else 2853 else
2854 slab_error(cache, "memory outside object was overwritten"); 2854 slab_error(cache, "memory outside object was overwritten");
2855 2855
2856 printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", 2856 printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n",
2857 obj, redzone1, redzone2); 2857 obj, redzone1, redzone2);
2858 } 2858 }
2859 2859
2860 static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, 2860 static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp,
2861 void *caller) 2861 void *caller)
2862 { 2862 {
2863 struct page *page; 2863 struct page *page;
2864 unsigned int objnr; 2864 unsigned int objnr;
2865 struct slab *slabp; 2865 struct slab *slabp;
2866 2866
2867 BUG_ON(virt_to_cache(objp) != cachep); 2867 BUG_ON(virt_to_cache(objp) != cachep);
2868 2868
2869 objp -= obj_offset(cachep); 2869 objp -= obj_offset(cachep);
2870 kfree_debugcheck(objp); 2870 kfree_debugcheck(objp);
2871 page = virt_to_head_page(objp); 2871 page = virt_to_head_page(objp);
2872 2872
2873 slabp = page_get_slab(page); 2873 slabp = page_get_slab(page);
2874 2874
2875 if (cachep->flags & SLAB_RED_ZONE) { 2875 if (cachep->flags & SLAB_RED_ZONE) {
2876 verify_redzone_free(cachep, objp); 2876 verify_redzone_free(cachep, objp);
2877 *dbg_redzone1(cachep, objp) = RED_INACTIVE; 2877 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2878 *dbg_redzone2(cachep, objp) = RED_INACTIVE; 2878 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2879 } 2879 }
2880 if (cachep->flags & SLAB_STORE_USER) 2880 if (cachep->flags & SLAB_STORE_USER)
2881 *dbg_userword(cachep, objp) = caller; 2881 *dbg_userword(cachep, objp) = caller;
2882 2882
2883 objnr = obj_to_index(cachep, slabp, objp); 2883 objnr = obj_to_index(cachep, slabp, objp);
2884 2884
2885 BUG_ON(objnr >= cachep->num); 2885 BUG_ON(objnr >= cachep->num);
2886 BUG_ON(objp != index_to_obj(cachep, slabp, objnr)); 2886 BUG_ON(objp != index_to_obj(cachep, slabp, objnr));
2887 2887
2888 #ifdef CONFIG_DEBUG_SLAB_LEAK 2888 #ifdef CONFIG_DEBUG_SLAB_LEAK
2889 slab_bufctl(slabp)[objnr] = BUFCTL_FREE; 2889 slab_bufctl(slabp)[objnr] = BUFCTL_FREE;
2890 #endif 2890 #endif
2891 if (cachep->flags & SLAB_POISON) { 2891 if (cachep->flags & SLAB_POISON) {
2892 #ifdef CONFIG_DEBUG_PAGEALLOC 2892 #ifdef CONFIG_DEBUG_PAGEALLOC
2893 if ((cachep->buffer_size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { 2893 if ((cachep->buffer_size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) {
2894 store_stackinfo(cachep, objp, (unsigned long)caller); 2894 store_stackinfo(cachep, objp, (unsigned long)caller);
2895 kernel_map_pages(virt_to_page(objp), 2895 kernel_map_pages(virt_to_page(objp),
2896 cachep->buffer_size / PAGE_SIZE, 0); 2896 cachep->buffer_size / PAGE_SIZE, 0);
2897 } else { 2897 } else {
2898 poison_obj(cachep, objp, POISON_FREE); 2898 poison_obj(cachep, objp, POISON_FREE);
2899 } 2899 }
2900 #else 2900 #else
2901 poison_obj(cachep, objp, POISON_FREE); 2901 poison_obj(cachep, objp, POISON_FREE);
2902 #endif 2902 #endif
2903 } 2903 }
2904 return objp; 2904 return objp;
2905 } 2905 }
2906 2906
2907 static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) 2907 static void check_slabp(struct kmem_cache *cachep, struct slab *slabp)
2908 { 2908 {
2909 kmem_bufctl_t i; 2909 kmem_bufctl_t i;
2910 int entries = 0; 2910 int entries = 0;
2911 2911
2912 /* Check slab's freelist to see if this obj is there. */ 2912 /* Check slab's freelist to see if this obj is there. */
2913 for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { 2913 for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) {
2914 entries++; 2914 entries++;
2915 if (entries > cachep->num || i >= cachep->num) 2915 if (entries > cachep->num || i >= cachep->num)
2916 goto bad; 2916 goto bad;
2917 } 2917 }
2918 if (entries != cachep->num - slabp->inuse) { 2918 if (entries != cachep->num - slabp->inuse) {
2919 bad: 2919 bad:
2920 printk(KERN_ERR "slab: Internal list corruption detected in " 2920 printk(KERN_ERR "slab: Internal list corruption detected in "
2921 "cache '%s'(%d), slabp %p(%d). Hexdump:\n", 2921 "cache '%s'(%d), slabp %p(%d). Hexdump:\n",
2922 cachep->name, cachep->num, slabp, slabp->inuse); 2922 cachep->name, cachep->num, slabp, slabp->inuse);
2923 for (i = 0; 2923 for (i = 0;
2924 i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t); 2924 i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t);
2925 i++) { 2925 i++) {
2926 if (i % 16 == 0) 2926 if (i % 16 == 0)
2927 printk("\n%03x:", i); 2927 printk("\n%03x:", i);
2928 printk(" %02x", ((unsigned char *)slabp)[i]); 2928 printk(" %02x", ((unsigned char *)slabp)[i]);
2929 } 2929 }
2930 printk("\n"); 2930 printk("\n");
2931 BUG(); 2931 BUG();
2932 } 2932 }
2933 } 2933 }
2934 #else 2934 #else
2935 #define kfree_debugcheck(x) do { } while(0) 2935 #define kfree_debugcheck(x) do { } while(0)
2936 #define cache_free_debugcheck(x,objp,z) (objp) 2936 #define cache_free_debugcheck(x,objp,z) (objp)
2937 #define check_slabp(x,y) do { } while(0) 2937 #define check_slabp(x,y) do { } while(0)
2938 #endif 2938 #endif
2939 2939
2940 static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) 2940 static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags)
2941 { 2941 {
2942 int batchcount; 2942 int batchcount;
2943 struct kmem_list3 *l3; 2943 struct kmem_list3 *l3;
2944 struct array_cache *ac; 2944 struct array_cache *ac;
2945 int node; 2945 int node;
2946 2946
2947 retry: 2947 retry:
2948 check_irq_off(); 2948 check_irq_off();
2949 node = numa_node_id(); 2949 node = numa_node_id();
2950 ac = cpu_cache_get(cachep); 2950 ac = cpu_cache_get(cachep);
2951 batchcount = ac->batchcount; 2951 batchcount = ac->batchcount;
2952 if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { 2952 if (!ac->touched && batchcount > BATCHREFILL_LIMIT) {
2953 /* 2953 /*
2954 * If there was little recent activity on this cache, then 2954 * If there was little recent activity on this cache, then
2955 * perform only a partial refill. Otherwise we could generate 2955 * perform only a partial refill. Otherwise we could generate
2956 * refill bouncing. 2956 * refill bouncing.
2957 */ 2957 */
2958 batchcount = BATCHREFILL_LIMIT; 2958 batchcount = BATCHREFILL_LIMIT;
2959 } 2959 }
2960 l3 = cachep->nodelists[node]; 2960 l3 = cachep->nodelists[node];
2961 2961
2962 BUG_ON(ac->avail > 0 || !l3); 2962 BUG_ON(ac->avail > 0 || !l3);
2963 spin_lock(&l3->list_lock); 2963 spin_lock(&l3->list_lock);
2964 2964
2965 /* See if we can refill from the shared array */ 2965 /* See if we can refill from the shared array */
2966 if (l3->shared && transfer_objects(ac, l3->shared, batchcount)) 2966 if (l3->shared && transfer_objects(ac, l3->shared, batchcount))
2967 goto alloc_done; 2967 goto alloc_done;
2968 2968
2969 while (batchcount > 0) { 2969 while (batchcount > 0) {
2970 struct list_head *entry; 2970 struct list_head *entry;
2971 struct slab *slabp; 2971 struct slab *slabp;
2972 /* Get slab alloc is to come from. */ 2972 /* Get slab alloc is to come from. */
2973 entry = l3->slabs_partial.next; 2973 entry = l3->slabs_partial.next;
2974 if (entry == &l3->slabs_partial) { 2974 if (entry == &l3->slabs_partial) {
2975 l3->free_touched = 1; 2975 l3->free_touched = 1;
2976 entry = l3->slabs_free.next; 2976 entry = l3->slabs_free.next;
2977 if (entry == &l3->slabs_free) 2977 if (entry == &l3->slabs_free)
2978 goto must_grow; 2978 goto must_grow;
2979 } 2979 }
2980 2980
2981 slabp = list_entry(entry, struct slab, list); 2981 slabp = list_entry(entry, struct slab, list);
2982 check_slabp(cachep, slabp); 2982 check_slabp(cachep, slabp);
2983 check_spinlock_acquired(cachep); 2983 check_spinlock_acquired(cachep);
2984 2984
2985 /* 2985 /*
2986 * The slab was either on partial or free list so 2986 * The slab was either on partial or free list so
2987 * there must be at least one object available for 2987 * there must be at least one object available for
2988 * allocation. 2988 * allocation.
2989 */ 2989 */
2990 BUG_ON(slabp->inuse >= cachep->num); 2990 BUG_ON(slabp->inuse >= cachep->num);
2991 2991
2992 while (slabp->inuse < cachep->num && batchcount--) { 2992 while (slabp->inuse < cachep->num && batchcount--) {
2993 STATS_INC_ALLOCED(cachep); 2993 STATS_INC_ALLOCED(cachep);
2994 STATS_INC_ACTIVE(cachep); 2994 STATS_INC_ACTIVE(cachep);
2995 STATS_SET_HIGH(cachep); 2995 STATS_SET_HIGH(cachep);
2996 2996
2997 ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, 2997 ac->entry[ac->avail++] = slab_get_obj(cachep, slabp,
2998 node); 2998 node);
2999 } 2999 }
3000 check_slabp(cachep, slabp); 3000 check_slabp(cachep, slabp);
3001 3001
3002 /* move slabp to correct slabp list: */ 3002 /* move slabp to correct slabp list: */
3003 list_del(&slabp->list); 3003 list_del(&slabp->list);
3004 if (slabp->free == BUFCTL_END) 3004 if (slabp->free == BUFCTL_END)
3005 list_add(&slabp->list, &l3->slabs_full); 3005 list_add(&slabp->list, &l3->slabs_full);
3006 else 3006 else
3007 list_add(&slabp->list, &l3->slabs_partial); 3007 list_add(&slabp->list, &l3->slabs_partial);
3008 } 3008 }
3009 3009
3010 must_grow: 3010 must_grow:
3011 l3->free_objects -= ac->avail; 3011 l3->free_objects -= ac->avail;
3012 alloc_done: 3012 alloc_done:
3013 spin_unlock(&l3->list_lock); 3013 spin_unlock(&l3->list_lock);
3014 3014
3015 if (unlikely(!ac->avail)) { 3015 if (unlikely(!ac->avail)) {
3016 int x; 3016 int x;
3017 x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); 3017 x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL);
3018 3018
3019 /* cache_grow can reenable interrupts, then ac could change. */ 3019 /* cache_grow can reenable interrupts, then ac could change. */
3020 ac = cpu_cache_get(cachep); 3020 ac = cpu_cache_get(cachep);
3021 if (!x && ac->avail == 0) /* no objects in sight? abort */ 3021 if (!x && ac->avail == 0) /* no objects in sight? abort */
3022 return NULL; 3022 return NULL;
3023 3023
3024 if (!ac->avail) /* objects refilled by interrupt? */ 3024 if (!ac->avail) /* objects refilled by interrupt? */
3025 goto retry; 3025 goto retry;
3026 } 3026 }
3027 ac->touched = 1; 3027 ac->touched = 1;
3028 return ac->entry[--ac->avail]; 3028 return ac->entry[--ac->avail];
3029 } 3029 }
3030 3030
3031 static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, 3031 static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep,
3032 gfp_t flags) 3032 gfp_t flags)
3033 { 3033 {
3034 might_sleep_if(flags & __GFP_WAIT); 3034 might_sleep_if(flags & __GFP_WAIT);
3035 #if DEBUG 3035 #if DEBUG
3036 kmem_flagcheck(cachep, flags); 3036 kmem_flagcheck(cachep, flags);
3037 #endif 3037 #endif
3038 } 3038 }
3039 3039
3040 #if DEBUG 3040 #if DEBUG
3041 static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, 3041 static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep,
3042 gfp_t flags, void *objp, void *caller) 3042 gfp_t flags, void *objp, void *caller)
3043 { 3043 {
3044 if (!objp) 3044 if (!objp)
3045 return objp; 3045 return objp;
3046 if (cachep->flags & SLAB_POISON) { 3046 if (cachep->flags & SLAB_POISON) {
3047 #ifdef CONFIG_DEBUG_PAGEALLOC 3047 #ifdef CONFIG_DEBUG_PAGEALLOC
3048 if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) 3048 if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep))
3049 kernel_map_pages(virt_to_page(objp), 3049 kernel_map_pages(virt_to_page(objp),
3050 cachep->buffer_size / PAGE_SIZE, 1); 3050 cachep->buffer_size / PAGE_SIZE, 1);
3051 else 3051 else
3052 check_poison_obj(cachep, objp); 3052 check_poison_obj(cachep, objp);
3053 #else 3053 #else
3054 check_poison_obj(cachep, objp); 3054 check_poison_obj(cachep, objp);
3055 #endif 3055 #endif
3056 poison_obj(cachep, objp, POISON_INUSE); 3056 poison_obj(cachep, objp, POISON_INUSE);
3057 } 3057 }
3058 if (cachep->flags & SLAB_STORE_USER) 3058 if (cachep->flags & SLAB_STORE_USER)
3059 *dbg_userword(cachep, objp) = caller; 3059 *dbg_userword(cachep, objp) = caller;
3060 3060
3061 if (cachep->flags & SLAB_RED_ZONE) { 3061 if (cachep->flags & SLAB_RED_ZONE) {
3062 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || 3062 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE ||
3063 *dbg_redzone2(cachep, objp) != RED_INACTIVE) { 3063 *dbg_redzone2(cachep, objp) != RED_INACTIVE) {
3064 slab_error(cachep, "double free, or memory outside" 3064 slab_error(cachep, "double free, or memory outside"
3065 " object was overwritten"); 3065 " object was overwritten");
3066 printk(KERN_ERR 3066 printk(KERN_ERR
3067 "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", 3067 "%p: redzone 1:0x%llx, redzone 2:0x%llx\n",
3068 objp, *dbg_redzone1(cachep, objp), 3068 objp, *dbg_redzone1(cachep, objp),
3069 *dbg_redzone2(cachep, objp)); 3069 *dbg_redzone2(cachep, objp));
3070 } 3070 }
3071 *dbg_redzone1(cachep, objp) = RED_ACTIVE; 3071 *dbg_redzone1(cachep, objp) = RED_ACTIVE;
3072 *dbg_redzone2(cachep, objp) = RED_ACTIVE; 3072 *dbg_redzone2(cachep, objp) = RED_ACTIVE;
3073 } 3073 }
3074 #ifdef CONFIG_DEBUG_SLAB_LEAK 3074 #ifdef CONFIG_DEBUG_SLAB_LEAK
3075 { 3075 {
3076 struct slab *slabp; 3076 struct slab *slabp;
3077 unsigned objnr; 3077 unsigned objnr;
3078 3078
3079 slabp = page_get_slab(virt_to_head_page(objp)); 3079 slabp = page_get_slab(virt_to_head_page(objp));
3080 objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; 3080 objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size;
3081 slab_bufctl(slabp)[objnr] = BUFCTL_ACTIVE; 3081 slab_bufctl(slabp)[objnr] = BUFCTL_ACTIVE;
3082 } 3082 }
3083 #endif 3083 #endif
3084 objp += obj_offset(cachep); 3084 objp += obj_offset(cachep);
3085 if (cachep->ctor && cachep->flags & SLAB_POISON) 3085 if (cachep->ctor && cachep->flags & SLAB_POISON)
3086 cachep->ctor(objp); 3086 cachep->ctor(objp);
3087 #if ARCH_SLAB_MINALIGN 3087 #if ARCH_SLAB_MINALIGN
3088 if ((u32)objp & (ARCH_SLAB_MINALIGN-1)) { 3088 if ((u32)objp & (ARCH_SLAB_MINALIGN-1)) {
3089 printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", 3089 printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n",
3090 objp, ARCH_SLAB_MINALIGN); 3090 objp, ARCH_SLAB_MINALIGN);
3091 } 3091 }
3092 #endif 3092 #endif
3093 return objp; 3093 return objp;
3094 } 3094 }
3095 #else 3095 #else
3096 #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) 3096 #define cache_alloc_debugcheck_after(a,b,objp,d) (objp)
3097 #endif 3097 #endif
3098 3098
3099 static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) 3099 static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags)
3100 { 3100 {
3101 if (cachep == &cache_cache) 3101 if (cachep == &cache_cache)
3102 return false; 3102 return false;
3103 3103
3104 return should_failslab(obj_size(cachep), flags); 3104 return should_failslab(obj_size(cachep), flags);
3105 } 3105 }
3106 3106
3107 static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) 3107 static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3108 { 3108 {
3109 void *objp; 3109 void *objp;
3110 struct array_cache *ac; 3110 struct array_cache *ac;
3111 3111
3112 check_irq_off(); 3112 check_irq_off();
3113 3113
3114 ac = cpu_cache_get(cachep); 3114 ac = cpu_cache_get(cachep);
3115 if (likely(ac->avail)) { 3115 if (likely(ac->avail)) {
3116 STATS_INC_ALLOCHIT(cachep); 3116 STATS_INC_ALLOCHIT(cachep);
3117 ac->touched = 1; 3117 ac->touched = 1;
3118 objp = ac->entry[--ac->avail]; 3118 objp = ac->entry[--ac->avail];
3119 } else { 3119 } else {
3120 STATS_INC_ALLOCMISS(cachep); 3120 STATS_INC_ALLOCMISS(cachep);
3121 objp = cache_alloc_refill(cachep, flags); 3121 objp = cache_alloc_refill(cachep, flags);
3122 /* 3122 /*
3123 * the 'ac' may be updated by cache_alloc_refill(), 3123 * the 'ac' may be updated by cache_alloc_refill(),
3124 * and kmemleak_erase() requires its correct value. 3124 * and kmemleak_erase() requires its correct value.
3125 */ 3125 */
3126 ac = cpu_cache_get(cachep); 3126 ac = cpu_cache_get(cachep);
3127 } 3127 }
3128 /* 3128 /*
3129 * To avoid a false negative, if an object that is in one of the 3129 * To avoid a false negative, if an object that is in one of the
3130 * per-CPU caches is leaked, we need to make sure kmemleak doesn't 3130 * per-CPU caches is leaked, we need to make sure kmemleak doesn't
3131 * treat the array pointers as a reference to the object. 3131 * treat the array pointers as a reference to the object.
3132 */ 3132 */
3133 if (objp) 3133 if (objp)
3134 kmemleak_erase(&ac->entry[ac->avail]); 3134 kmemleak_erase(&ac->entry[ac->avail]);
3135 return objp; 3135 return objp;
3136 } 3136 }
3137 3137
3138 #ifdef CONFIG_NUMA 3138 #ifdef CONFIG_NUMA
3139 /* 3139 /*
3140 * Try allocating on another node if PF_SPREAD_SLAB|PF_MEMPOLICY. 3140 * Try allocating on another node if PF_SPREAD_SLAB|PF_MEMPOLICY.
3141 * 3141 *
3142 * If we are in_interrupt, then process context, including cpusets and 3142 * If we are in_interrupt, then process context, including cpusets and
3143 * mempolicy, may not apply and should not be used for allocation policy. 3143 * mempolicy, may not apply and should not be used for allocation policy.
3144 */ 3144 */
3145 static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) 3145 static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags)
3146 { 3146 {
3147 int nid_alloc, nid_here; 3147 int nid_alloc, nid_here;
3148 3148
3149 if (in_interrupt() || (flags & __GFP_THISNODE)) 3149 if (in_interrupt() || (flags & __GFP_THISNODE))
3150 return NULL; 3150 return NULL;
3151 nid_alloc = nid_here = numa_node_id(); 3151 nid_alloc = nid_here = numa_node_id();
3152 if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) 3152 if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD))
3153 nid_alloc = cpuset_mem_spread_node(); 3153 nid_alloc = cpuset_mem_spread_node();
3154 else if (current->mempolicy) 3154 else if (current->mempolicy)
3155 nid_alloc = slab_node(current->mempolicy); 3155 nid_alloc = slab_node(current->mempolicy);
3156 if (nid_alloc != nid_here) 3156 if (nid_alloc != nid_here)
3157 return ____cache_alloc_node(cachep, flags, nid_alloc); 3157 return ____cache_alloc_node(cachep, flags, nid_alloc);
3158 return NULL; 3158 return NULL;
3159 } 3159 }
3160 3160
3161 /* 3161 /*
3162 * Fallback function if there was no memory available and no objects on a 3162 * Fallback function if there was no memory available and no objects on a
3163 * certain node and fall back is permitted. First we scan all the 3163 * certain node and fall back is permitted. First we scan all the
3164 * available nodelists for available objects. If that fails then we 3164 * available nodelists for available objects. If that fails then we
3165 * perform an allocation without specifying a node. This allows the page 3165 * perform an allocation without specifying a node. This allows the page
3166 * allocator to do its reclaim / fallback magic. We then insert the 3166 * allocator to do its reclaim / fallback magic. We then insert the
3167 * slab into the proper nodelist and then allocate from it. 3167 * slab into the proper nodelist and then allocate from it.
3168 */ 3168 */
3169 static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) 3169 static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags)
3170 { 3170 {
3171 struct zonelist *zonelist; 3171 struct zonelist *zonelist;
3172 gfp_t local_flags; 3172 gfp_t local_flags;
3173 struct zoneref *z; 3173 struct zoneref *z;
3174 struct zone *zone; 3174 struct zone *zone;
3175 enum zone_type high_zoneidx = gfp_zone(flags); 3175 enum zone_type high_zoneidx = gfp_zone(flags);
3176 void *obj = NULL; 3176 void *obj = NULL;
3177 int nid; 3177 int nid;
3178 3178
3179 if (flags & __GFP_THISNODE) 3179 if (flags & __GFP_THISNODE)
3180 return NULL; 3180 return NULL;
3181 3181
3182 zonelist = node_zonelist(slab_node(current->mempolicy), flags); 3182 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
3183 local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); 3183 local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK);
3184 3184
3185 retry: 3185 retry:
3186 /* 3186 /*
3187 * Look through allowed nodes for objects available 3187 * Look through allowed nodes for objects available
3188 * from existing per node queues. 3188 * from existing per node queues.
3189 */ 3189 */
3190 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { 3190 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
3191 nid = zone_to_nid(zone); 3191 nid = zone_to_nid(zone);
3192 3192
3193 if (cpuset_zone_allowed_hardwall(zone, flags) && 3193 if (cpuset_zone_allowed_hardwall(zone, flags) &&
3194 cache->nodelists[nid] && 3194 cache->nodelists[nid] &&
3195 cache->nodelists[nid]->free_objects) { 3195 cache->nodelists[nid]->free_objects) {
3196 obj = ____cache_alloc_node(cache, 3196 obj = ____cache_alloc_node(cache,
3197 flags | GFP_THISNODE, nid); 3197 flags | GFP_THISNODE, nid);
3198 if (obj) 3198 if (obj)
3199 break; 3199 break;
3200 } 3200 }
3201 } 3201 }
3202 3202
3203 if (!obj) { 3203 if (!obj) {
3204 /* 3204 /*
3205 * This allocation will be performed within the constraints 3205 * This allocation will be performed within the constraints
3206 * of the current cpuset / memory policy requirements. 3206 * of the current cpuset / memory policy requirements.
3207 * We may trigger various forms of reclaim on the allowed 3207 * We may trigger various forms of reclaim on the allowed
3208 * set and go into memory reserves if necessary. 3208 * set and go into memory reserves if necessary.
3209 */ 3209 */
3210 if (local_flags & __GFP_WAIT) 3210 if (local_flags & __GFP_WAIT)
3211 local_irq_enable(); 3211 local_irq_enable();
3212 kmem_flagcheck(cache, flags); 3212 kmem_flagcheck(cache, flags);
3213 obj = kmem_getpages(cache, local_flags, numa_node_id()); 3213 obj = kmem_getpages(cache, local_flags, numa_node_id());
3214 if (local_flags & __GFP_WAIT) 3214 if (local_flags & __GFP_WAIT)
3215 local_irq_disable(); 3215 local_irq_disable();
3216 if (obj) { 3216 if (obj) {
3217 /* 3217 /*
3218 * Insert into the appropriate per node queues 3218 * Insert into the appropriate per node queues
3219 */ 3219 */
3220 nid = page_to_nid(virt_to_page(obj)); 3220 nid = page_to_nid(virt_to_page(obj));
3221 if (cache_grow(cache, flags, nid, obj)) { 3221 if (cache_grow(cache, flags, nid, obj)) {
3222 obj = ____cache_alloc_node(cache, 3222 obj = ____cache_alloc_node(cache,
3223 flags | GFP_THISNODE, nid); 3223 flags | GFP_THISNODE, nid);
3224 if (!obj) 3224 if (!obj)
3225 /* 3225 /*
3226 * Another processor may allocate the 3226 * Another processor may allocate the
3227 * objects in the slab since we are 3227 * objects in the slab since we are
3228 * not holding any locks. 3228 * not holding any locks.
3229 */ 3229 */
3230 goto retry; 3230 goto retry;
3231 } else { 3231 } else {
3232 /* cache_grow already freed obj */ 3232 /* cache_grow already freed obj */
3233 obj = NULL; 3233 obj = NULL;
3234 } 3234 }
3235 } 3235 }
3236 } 3236 }
3237 return obj; 3237 return obj;
3238 } 3238 }
3239 3239
3240 /* 3240 /*
3241 * A interface to enable slab creation on nodeid 3241 * A interface to enable slab creation on nodeid
3242 */ 3242 */
3243 static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, 3243 static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
3244 int nodeid) 3244 int nodeid)
3245 { 3245 {
3246 struct list_head *entry; 3246 struct list_head *entry;
3247 struct slab *slabp; 3247 struct slab *slabp;
3248 struct kmem_list3 *l3; 3248 struct kmem_list3 *l3;
3249 void *obj; 3249 void *obj;
3250 int x; 3250 int x;
3251 3251
3252 l3 = cachep->nodelists[nodeid]; 3252 l3 = cachep->nodelists[nodeid];
3253 BUG_ON(!l3); 3253 BUG_ON(!l3);
3254 3254
3255 retry: 3255 retry:
3256 check_irq_off(); 3256 check_irq_off();
3257 spin_lock(&l3->list_lock); 3257 spin_lock(&l3->list_lock);
3258 entry = l3->slabs_partial.next; 3258 entry = l3->slabs_partial.next;
3259 if (entry == &l3->slabs_partial) { 3259 if (entry == &l3->slabs_partial) {
3260 l3->free_touched = 1; 3260 l3->free_touched = 1;
3261 entry = l3->slabs_free.next; 3261 entry = l3->slabs_free.next;
3262 if (entry == &l3->slabs_free) 3262 if (entry == &l3->slabs_free)
3263 goto must_grow; 3263 goto must_grow;
3264 } 3264 }
3265 3265
3266 slabp = list_entry(entry, struct slab, list); 3266 slabp = list_entry(entry, struct slab, list);
3267 check_spinlock_acquired_node(cachep, nodeid); 3267 check_spinlock_acquired_node(cachep, nodeid);
3268 check_slabp(cachep, slabp); 3268 check_slabp(cachep, slabp);
3269 3269
3270 STATS_INC_NODEALLOCS(cachep); 3270 STATS_INC_NODEALLOCS(cachep);
3271 STATS_INC_ACTIVE(cachep); 3271 STATS_INC_ACTIVE(cachep);
3272 STATS_SET_HIGH(cachep); 3272 STATS_SET_HIGH(cachep);
3273 3273
3274 BUG_ON(slabp->inuse == cachep->num); 3274 BUG_ON(slabp->inuse == cachep->num);
3275 3275
3276 obj = slab_get_obj(cachep, slabp, nodeid); 3276 obj = slab_get_obj(cachep, slabp, nodeid);
3277 check_slabp(cachep, slabp); 3277 check_slabp(cachep, slabp);
3278 l3->free_objects--; 3278 l3->free_objects--;
3279 /* move slabp to correct slabp list: */ 3279 /* move slabp to correct slabp list: */
3280 list_del(&slabp->list); 3280 list_del(&slabp->list);
3281 3281
3282 if (slabp->free == BUFCTL_END) 3282 if (slabp->free == BUFCTL_END)
3283 list_add(&slabp->list, &l3->slabs_full); 3283 list_add(&slabp->list, &l3->slabs_full);
3284 else 3284 else
3285 list_add(&slabp->list, &l3->slabs_partial); 3285 list_add(&slabp->list, &l3->slabs_partial);
3286 3286
3287 spin_unlock(&l3->list_lock); 3287 spin_unlock(&l3->list_lock);
3288 goto done; 3288 goto done;
3289 3289
3290 must_grow: 3290 must_grow:
3291 spin_unlock(&l3->list_lock); 3291 spin_unlock(&l3->list_lock);
3292 x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); 3292 x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL);
3293 if (x) 3293 if (x)
3294 goto retry; 3294 goto retry;
3295 3295
3296 return fallback_alloc(cachep, flags); 3296 return fallback_alloc(cachep, flags);
3297 3297
3298 done: 3298 done:
3299 return obj; 3299 return obj;
3300 } 3300 }
3301 3301
3302 /** 3302 /**
3303 * kmem_cache_alloc_node - Allocate an object on the specified node 3303 * kmem_cache_alloc_node - Allocate an object on the specified node
3304 * @cachep: The cache to allocate from. 3304 * @cachep: The cache to allocate from.
3305 * @flags: See kmalloc(). 3305 * @flags: See kmalloc().
3306 * @nodeid: node number of the target node. 3306 * @nodeid: node number of the target node.
3307 * @caller: return address of caller, used for debug information 3307 * @caller: return address of caller, used for debug information
3308 * 3308 *
3309 * Identical to kmem_cache_alloc but it will allocate memory on the given 3309 * Identical to kmem_cache_alloc but it will allocate memory on the given
3310 * node, which can improve the performance for cpu bound structures. 3310 * node, which can improve the performance for cpu bound structures.
3311 * 3311 *
3312 * Fallback to other node is possible if __GFP_THISNODE is not set. 3312 * Fallback to other node is possible if __GFP_THISNODE is not set.
3313 */ 3313 */
3314 static __always_inline void * 3314 static __always_inline void *
3315 __cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, 3315 __cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid,
3316 void *caller) 3316 void *caller)
3317 { 3317 {
3318 unsigned long save_flags; 3318 unsigned long save_flags;
3319 void *ptr; 3319 void *ptr;
3320 3320
3321 flags &= gfp_allowed_mask; 3321 flags &= gfp_allowed_mask;
3322 3322
3323 lockdep_trace_alloc(flags); 3323 lockdep_trace_alloc(flags);
3324 3324
3325 if (slab_should_failslab(cachep, flags)) 3325 if (slab_should_failslab(cachep, flags))
3326 return NULL; 3326 return NULL;
3327 3327
3328 cache_alloc_debugcheck_before(cachep, flags); 3328 cache_alloc_debugcheck_before(cachep, flags);
3329 local_irq_save(save_flags); 3329 local_irq_save(save_flags);
3330 3330
3331 if (nodeid == -1) 3331 if (nodeid == -1)
3332 nodeid = numa_node_id(); 3332 nodeid = numa_node_id();
3333 3333
3334 if (unlikely(!cachep->nodelists[nodeid])) { 3334 if (unlikely(!cachep->nodelists[nodeid])) {
3335 /* Node not bootstrapped yet */ 3335 /* Node not bootstrapped yet */
3336 ptr = fallback_alloc(cachep, flags); 3336 ptr = fallback_alloc(cachep, flags);
3337 goto out; 3337 goto out;
3338 } 3338 }
3339 3339
3340 if (nodeid == numa_node_id()) { 3340 if (nodeid == numa_node_id()) {
3341 /* 3341 /*
3342 * Use the locally cached objects if possible. 3342 * Use the locally cached objects if possible.
3343 * However ____cache_alloc does not allow fallback 3343 * However ____cache_alloc does not allow fallback
3344 * to other nodes. It may fail while we still have 3344 * to other nodes. It may fail while we still have
3345 * objects on other nodes available. 3345 * objects on other nodes available.
3346 */ 3346 */
3347 ptr = ____cache_alloc(cachep, flags); 3347 ptr = ____cache_alloc(cachep, flags);
3348 if (ptr) 3348 if (ptr)
3349 goto out; 3349 goto out;
3350 } 3350 }
3351 /* ___cache_alloc_node can fall back to other nodes */ 3351 /* ___cache_alloc_node can fall back to other nodes */
3352 ptr = ____cache_alloc_node(cachep, flags, nodeid); 3352 ptr = ____cache_alloc_node(cachep, flags, nodeid);
3353 out: 3353 out:
3354 local_irq_restore(save_flags); 3354 local_irq_restore(save_flags);
3355 ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); 3355 ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller);
3356 kmemleak_alloc_recursive(ptr, obj_size(cachep), 1, cachep->flags, 3356 kmemleak_alloc_recursive(ptr, obj_size(cachep), 1, cachep->flags,
3357 flags); 3357 flags);
3358 3358
3359 if (likely(ptr)) 3359 if (likely(ptr))
3360 kmemcheck_slab_alloc(cachep, flags, ptr, obj_size(cachep)); 3360 kmemcheck_slab_alloc(cachep, flags, ptr, obj_size(cachep));
3361 3361
3362 if (unlikely((flags & __GFP_ZERO) && ptr)) 3362 if (unlikely((flags & __GFP_ZERO) && ptr))
3363 memset(ptr, 0, obj_size(cachep)); 3363 memset(ptr, 0, obj_size(cachep));
3364 3364
3365 return ptr; 3365 return ptr;
3366 } 3366 }
3367 3367
3368 static __always_inline void * 3368 static __always_inline void *
3369 __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) 3369 __do_cache_alloc(struct kmem_cache *cache, gfp_t flags)
3370 { 3370 {
3371 void *objp; 3371 void *objp;
3372 3372
3373 if (unlikely(current->flags & (PF_SPREAD_SLAB | PF_MEMPOLICY))) { 3373 if (unlikely(current->flags & (PF_SPREAD_SLAB | PF_MEMPOLICY))) {
3374 objp = alternate_node_alloc(cache, flags); 3374 objp = alternate_node_alloc(cache, flags);
3375 if (objp) 3375 if (objp)
3376 goto out; 3376 goto out;
3377 } 3377 }
3378 objp = ____cache_alloc(cache, flags); 3378 objp = ____cache_alloc(cache, flags);
3379 3379
3380 /* 3380 /*
3381 * We may just have run out of memory on the local node. 3381 * We may just have run out of memory on the local node.
3382 * ____cache_alloc_node() knows how to locate memory on other nodes 3382 * ____cache_alloc_node() knows how to locate memory on other nodes
3383 */ 3383 */
3384 if (!objp) 3384 if (!objp)
3385 objp = ____cache_alloc_node(cache, flags, numa_node_id()); 3385 objp = ____cache_alloc_node(cache, flags, numa_node_id());
3386 3386
3387 out: 3387 out:
3388 return objp; 3388 return objp;
3389 } 3389 }
3390 #else 3390 #else
3391 3391
3392 static __always_inline void * 3392 static __always_inline void *
3393 __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) 3393 __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3394 { 3394 {
3395 return ____cache_alloc(cachep, flags); 3395 return ____cache_alloc(cachep, flags);
3396 } 3396 }
3397 3397
3398 #endif /* CONFIG_NUMA */ 3398 #endif /* CONFIG_NUMA */
3399 3399
3400 static __always_inline void * 3400 static __always_inline void *
3401 __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller) 3401 __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller)
3402 { 3402 {
3403 unsigned long save_flags; 3403 unsigned long save_flags;
3404 void *objp; 3404 void *objp;
3405 3405
3406 flags &= gfp_allowed_mask; 3406 flags &= gfp_allowed_mask;
3407 3407
3408 lockdep_trace_alloc(flags); 3408 lockdep_trace_alloc(flags);
3409 3409
3410 if (slab_should_failslab(cachep, flags)) 3410 if (slab_should_failslab(cachep, flags))
3411 return NULL; 3411 return NULL;
3412 3412
3413 cache_alloc_debugcheck_before(cachep, flags); 3413 cache_alloc_debugcheck_before(cachep, flags);
3414 local_irq_save(save_flags); 3414 local_irq_save(save_flags);
3415 objp = __do_cache_alloc(cachep, flags); 3415 objp = __do_cache_alloc(cachep, flags);
3416 local_irq_restore(save_flags); 3416 local_irq_restore(save_flags);
3417 objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); 3417 objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
3418 kmemleak_alloc_recursive(objp, obj_size(cachep), 1, cachep->flags, 3418 kmemleak_alloc_recursive(objp, obj_size(cachep), 1, cachep->flags,
3419 flags); 3419 flags);
3420 prefetchw(objp); 3420 prefetchw(objp);
3421 3421
3422 if (likely(objp)) 3422 if (likely(objp))
3423 kmemcheck_slab_alloc(cachep, flags, objp, obj_size(cachep)); 3423 kmemcheck_slab_alloc(cachep, flags, objp, obj_size(cachep));
3424 3424
3425 if (unlikely((flags & __GFP_ZERO) && objp)) 3425 if (unlikely((flags & __GFP_ZERO) && objp))
3426 memset(objp, 0, obj_size(cachep)); 3426 memset(objp, 0, obj_size(cachep));
3427 3427
3428 return objp; 3428 return objp;
3429 } 3429 }
3430 3430
3431 /* 3431 /*
3432 * Caller needs to acquire correct kmem_list's list_lock 3432 * Caller needs to acquire correct kmem_list's list_lock
3433 */ 3433 */
3434 static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, 3434 static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects,
3435 int node) 3435 int node)
3436 { 3436 {
3437 int i; 3437 int i;
3438 struct kmem_list3 *l3; 3438 struct kmem_list3 *l3;
3439 3439
3440 for (i = 0; i < nr_objects; i++) { 3440 for (i = 0; i < nr_objects; i++) {
3441 void *objp = objpp[i]; 3441 void *objp = objpp[i];
3442 struct slab *slabp; 3442 struct slab *slabp;
3443 3443
3444 slabp = virt_to_slab(objp); 3444 slabp = virt_to_slab(objp);
3445 l3 = cachep->nodelists[node]; 3445 l3 = cachep->nodelists[node];
3446 list_del(&slabp->list); 3446 list_del(&slabp->list);
3447 check_spinlock_acquired_node(cachep, node); 3447 check_spinlock_acquired_node(cachep, node);
3448 check_slabp(cachep, slabp); 3448 check_slabp(cachep, slabp);
3449 slab_put_obj(cachep, slabp, objp, node); 3449 slab_put_obj(cachep, slabp, objp, node);
3450 STATS_DEC_ACTIVE(cachep); 3450 STATS_DEC_ACTIVE(cachep);
3451 l3->free_objects++; 3451 l3->free_objects++;
3452 check_slabp(cachep, slabp); 3452 check_slabp(cachep, slabp);
3453 3453
3454 /* fixup slab chains */ 3454 /* fixup slab chains */
3455 if (slabp->inuse == 0) { 3455 if (slabp->inuse == 0) {
3456 if (l3->free_objects > l3->free_limit) { 3456 if (l3->free_objects > l3->free_limit) {
3457 l3->free_objects -= cachep->num; 3457 l3->free_objects -= cachep->num;
3458 /* No need to drop any previously held 3458 /* No need to drop any previously held
3459 * lock here, even if we have a off-slab slab 3459 * lock here, even if we have a off-slab slab
3460 * descriptor it is guaranteed to come from 3460 * descriptor it is guaranteed to come from
3461 * a different cache, refer to comments before 3461 * a different cache, refer to comments before
3462 * alloc_slabmgmt. 3462 * alloc_slabmgmt.
3463 */ 3463 */
3464 slab_destroy(cachep, slabp); 3464 slab_destroy(cachep, slabp);
3465 } else { 3465 } else {
3466 list_add(&slabp->list, &l3->slabs_free); 3466 list_add(&slabp->list, &l3->slabs_free);
3467 } 3467 }
3468 } else { 3468 } else {
3469 /* Unconditionally move a slab to the end of the 3469 /* Unconditionally move a slab to the end of the
3470 * partial list on free - maximum time for the 3470 * partial list on free - maximum time for the
3471 * other objects to be freed, too. 3471 * other objects to be freed, too.
3472 */ 3472 */
3473 list_add_tail(&slabp->list, &l3->slabs_partial); 3473 list_add_tail(&slabp->list, &l3->slabs_partial);
3474 } 3474 }
3475 } 3475 }
3476 } 3476 }
3477 3477
3478 static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) 3478 static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac)
3479 { 3479 {
3480 int batchcount; 3480 int batchcount;
3481 struct kmem_list3 *l3; 3481 struct kmem_list3 *l3;
3482 int node = numa_node_id(); 3482 int node = numa_node_id();
3483 3483
3484 batchcount = ac->batchcount; 3484 batchcount = ac->batchcount;
3485 #if DEBUG 3485 #if DEBUG
3486 BUG_ON(!batchcount || batchcount > ac->avail); 3486 BUG_ON(!batchcount || batchcount > ac->avail);
3487 #endif 3487 #endif
3488 check_irq_off(); 3488 check_irq_off();
3489 l3 = cachep->nodelists[node]; 3489 l3 = cachep->nodelists[node];
3490 spin_lock(&l3->list_lock); 3490 spin_lock(&l3->list_lock);
3491 if (l3->shared) { 3491 if (l3->shared) {
3492 struct array_cache *shared_array = l3->shared; 3492 struct array_cache *shared_array = l3->shared;
3493 int max = shared_array->limit - shared_array->avail; 3493 int max = shared_array->limit - shared_array->avail;
3494 if (max) { 3494 if (max) {
3495 if (batchcount > max) 3495 if (batchcount > max)
3496 batchcount = max; 3496 batchcount = max;
3497 memcpy(&(shared_array->entry[shared_array->avail]), 3497 memcpy(&(shared_array->entry[shared_array->avail]),
3498 ac->entry, sizeof(void *) * batchcount); 3498 ac->entry, sizeof(void *) * batchcount);
3499 shared_array->avail += batchcount; 3499 shared_array->avail += batchcount;
3500 goto free_done; 3500 goto free_done;
3501 } 3501 }
3502 } 3502 }
3503 3503
3504 free_block(cachep, ac->entry, batchcount, node); 3504 free_block(cachep, ac->entry, batchcount, node);
3505 free_done: 3505 free_done:
3506 #if STATS 3506 #if STATS
3507 { 3507 {
3508 int i = 0; 3508 int i = 0;
3509 struct list_head *p; 3509 struct list_head *p;
3510 3510
3511 p = l3->slabs_free.next; 3511 p = l3->slabs_free.next;
3512 while (p != &(l3->slabs_free)) { 3512 while (p != &(l3->slabs_free)) {
3513 struct slab *slabp; 3513 struct slab *slabp;
3514 3514
3515 slabp = list_entry(p, struct slab, list); 3515 slabp = list_entry(p, struct slab, list);
3516 BUG_ON(slabp->inuse); 3516 BUG_ON(slabp->inuse);
3517 3517
3518 i++; 3518 i++;
3519 p = p->next; 3519 p = p->next;
3520 } 3520 }
3521 STATS_SET_FREEABLE(cachep, i); 3521 STATS_SET_FREEABLE(cachep, i);
3522 } 3522 }
3523 #endif 3523 #endif
3524 spin_unlock(&l3->list_lock); 3524 spin_unlock(&l3->list_lock);
3525 ac->avail -= batchcount; 3525 ac->avail -= batchcount;
3526 memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); 3526 memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail);
3527 } 3527 }
3528 3528
3529 /* 3529 /*
3530 * Release an obj back to its cache. If the obj has a constructed state, it must 3530 * Release an obj back to its cache. If the obj has a constructed state, it must
3531 * be in this state _before_ it is released. Called with disabled ints. 3531 * be in this state _before_ it is released. Called with disabled ints.
3532 */ 3532 */
3533 static inline void __cache_free(struct kmem_cache *cachep, void *objp) 3533 static inline void __cache_free(struct kmem_cache *cachep, void *objp)
3534 { 3534 {
3535 struct array_cache *ac = cpu_cache_get(cachep); 3535 struct array_cache *ac = cpu_cache_get(cachep);
3536 3536
3537 check_irq_off(); 3537 check_irq_off();
3538 kmemleak_free_recursive(objp, cachep->flags); 3538 kmemleak_free_recursive(objp, cachep->flags);
3539 objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); 3539 objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0));
3540 3540
3541 kmemcheck_slab_free(cachep, objp, obj_size(cachep)); 3541 kmemcheck_slab_free(cachep, objp, obj_size(cachep));
3542 3542
3543 /* 3543 /*
3544 * Skip calling cache_free_alien() when the platform is not numa. 3544 * Skip calling cache_free_alien() when the platform is not numa.
3545 * This will avoid cache misses that happen while accessing slabp (which 3545 * This will avoid cache misses that happen while accessing slabp (which
3546 * is per page memory reference) to get nodeid. Instead use a global 3546 * is per page memory reference) to get nodeid. Instead use a global
3547 * variable to skip the call, which is mostly likely to be present in 3547 * variable to skip the call, which is mostly likely to be present in
3548 * the cache. 3548 * the cache.
3549 */ 3549 */
3550 if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) 3550 if (nr_online_nodes > 1 && cache_free_alien(cachep, objp))
3551 return; 3551 return;
3552 3552
3553 if (likely(ac->avail < ac->limit)) { 3553 if (likely(ac->avail < ac->limit)) {
3554 STATS_INC_FREEHIT(cachep); 3554 STATS_INC_FREEHIT(cachep);
3555 ac->entry[ac->avail++] = objp; 3555 ac->entry[ac->avail++] = objp;
3556 return; 3556 return;
3557 } else { 3557 } else {
3558 STATS_INC_FREEMISS(cachep); 3558 STATS_INC_FREEMISS(cachep);
3559 cache_flusharray(cachep, ac); 3559 cache_flusharray(cachep, ac);
3560 ac->entry[ac->avail++] = objp; 3560 ac->entry[ac->avail++] = objp;
3561 } 3561 }
3562 } 3562 }
3563 3563
3564 /** 3564 /**
3565 * kmem_cache_alloc - Allocate an object 3565 * kmem_cache_alloc - Allocate an object
3566 * @cachep: The cache to allocate from. 3566 * @cachep: The cache to allocate from.
3567 * @flags: See kmalloc(). 3567 * @flags: See kmalloc().
3568 * 3568 *
3569 * Allocate an object from this cache. The flags are only relevant 3569 * Allocate an object from this cache. The flags are only relevant
3570 * if the cache has no available objects. 3570 * if the cache has no available objects.
3571 */ 3571 */
3572 void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) 3572 void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3573 { 3573 {
3574 void *ret = __cache_alloc(cachep, flags, __builtin_return_address(0)); 3574 void *ret = __cache_alloc(cachep, flags, __builtin_return_address(0));
3575 3575
3576 trace_kmem_cache_alloc(_RET_IP_, ret, 3576 trace_kmem_cache_alloc(_RET_IP_, ret,
3577 obj_size(cachep), cachep->buffer_size, flags); 3577 obj_size(cachep), cachep->buffer_size, flags);
3578 3578
3579 return ret; 3579 return ret;
3580 } 3580 }
3581 EXPORT_SYMBOL(kmem_cache_alloc); 3581 EXPORT_SYMBOL(kmem_cache_alloc);
3582 3582
3583 #ifdef CONFIG_TRACING 3583 #ifdef CONFIG_TRACING
3584 void *kmem_cache_alloc_notrace(struct kmem_cache *cachep, gfp_t flags) 3584 void *kmem_cache_alloc_notrace(struct kmem_cache *cachep, gfp_t flags)
3585 { 3585 {
3586 return __cache_alloc(cachep, flags, __builtin_return_address(0)); 3586 return __cache_alloc(cachep, flags, __builtin_return_address(0));
3587 } 3587 }
3588 EXPORT_SYMBOL(kmem_cache_alloc_notrace); 3588 EXPORT_SYMBOL(kmem_cache_alloc_notrace);
3589 #endif 3589 #endif
3590 3590
3591 /** 3591 /**
3592 * kmem_ptr_validate - check if an untrusted pointer might be a slab entry. 3592 * kmem_ptr_validate - check if an untrusted pointer might be a slab entry.
3593 * @cachep: the cache we're checking against 3593 * @cachep: the cache we're checking against
3594 * @ptr: pointer to validate 3594 * @ptr: pointer to validate
3595 * 3595 *
3596 * This verifies that the untrusted pointer looks sane; 3596 * This verifies that the untrusted pointer looks sane;
3597 * it is _not_ a guarantee that the pointer is actually 3597 * it is _not_ a guarantee that the pointer is actually
3598 * part of the slab cache in question, but it at least 3598 * part of the slab cache in question, but it at least
3599 * validates that the pointer can be dereferenced and 3599 * validates that the pointer can be dereferenced and
3600 * looks half-way sane. 3600 * looks half-way sane.
3601 * 3601 *
3602 * Currently only used for dentry validation. 3602 * Currently only used for dentry validation.
3603 */ 3603 */
3604 int kmem_ptr_validate(struct kmem_cache *cachep, const void *ptr) 3604 int kmem_ptr_validate(struct kmem_cache *cachep, const void *ptr)
3605 { 3605 {
3606 unsigned long addr = (unsigned long)ptr; 3606 unsigned long addr = (unsigned long)ptr;
3607 unsigned long min_addr = PAGE_OFFSET; 3607 unsigned long min_addr = PAGE_OFFSET;
3608 unsigned long align_mask = BYTES_PER_WORD - 1; 3608 unsigned long align_mask = BYTES_PER_WORD - 1;
3609 unsigned long size = cachep->buffer_size; 3609 unsigned long size = cachep->buffer_size;
3610 struct page *page; 3610 struct page *page;
3611 3611
3612 if (unlikely(addr < min_addr)) 3612 if (unlikely(addr < min_addr))
3613 goto out; 3613 goto out;
3614 if (unlikely(addr > (unsigned long)high_memory - size)) 3614 if (unlikely(addr > (unsigned long)high_memory - size))
3615 goto out; 3615 goto out;
3616 if (unlikely(addr & align_mask)) 3616 if (unlikely(addr & align_mask))
3617 goto out; 3617 goto out;
3618 if (unlikely(!kern_addr_valid(addr))) 3618 if (unlikely(!kern_addr_valid(addr)))
3619 goto out; 3619 goto out;
3620 if (unlikely(!kern_addr_valid(addr + size - 1))) 3620 if (unlikely(!kern_addr_valid(addr + size - 1)))
3621 goto out; 3621 goto out;
3622 page = virt_to_page(ptr); 3622 page = virt_to_page(ptr);
3623 if (unlikely(!PageSlab(page))) 3623 if (unlikely(!PageSlab(page)))
3624 goto out; 3624 goto out;
3625 if (unlikely(page_get_cache(page) != cachep)) 3625 if (unlikely(page_get_cache(page) != cachep))
3626 goto out; 3626 goto out;
3627 return 1; 3627 return 1;
3628 out: 3628 out:
3629 return 0; 3629 return 0;
3630 } 3630 }
3631 3631
3632 #ifdef CONFIG_NUMA 3632 #ifdef CONFIG_NUMA
3633 void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) 3633 void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid)
3634 { 3634 {
3635 void *ret = __cache_alloc_node(cachep, flags, nodeid, 3635 void *ret = __cache_alloc_node(cachep, flags, nodeid,
3636 __builtin_return_address(0)); 3636 __builtin_return_address(0));
3637 3637
3638 trace_kmem_cache_alloc_node(_RET_IP_, ret, 3638 trace_kmem_cache_alloc_node(_RET_IP_, ret,
3639 obj_size(cachep), cachep->buffer_size, 3639 obj_size(cachep), cachep->buffer_size,
3640 flags, nodeid); 3640 flags, nodeid);
3641 3641
3642 return ret; 3642 return ret;
3643 } 3643 }
3644 EXPORT_SYMBOL(kmem_cache_alloc_node); 3644 EXPORT_SYMBOL(kmem_cache_alloc_node);
3645 3645
3646 #ifdef CONFIG_TRACING 3646 #ifdef CONFIG_TRACING
3647 void *kmem_cache_alloc_node_notrace(struct kmem_cache *cachep, 3647 void *kmem_cache_alloc_node_notrace(struct kmem_cache *cachep,
3648 gfp_t flags, 3648 gfp_t flags,
3649 int nodeid) 3649 int nodeid)
3650 { 3650 {
3651 return __cache_alloc_node(cachep, flags, nodeid, 3651 return __cache_alloc_node(cachep, flags, nodeid,
3652 __builtin_return_address(0)); 3652 __builtin_return_address(0));
3653 } 3653 }
3654 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); 3654 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
3655 #endif 3655 #endif
3656 3656
3657 static __always_inline void * 3657 static __always_inline void *
3658 __do_kmalloc_node(size_t size, gfp_t flags, int node, void *caller) 3658 __do_kmalloc_node(size_t size, gfp_t flags, int node, void *caller)
3659 { 3659 {
3660 struct kmem_cache *cachep; 3660 struct kmem_cache *cachep;
3661 void *ret; 3661 void *ret;
3662 3662
3663 cachep = kmem_find_general_cachep(size, flags); 3663 cachep = kmem_find_general_cachep(size, flags);
3664 if (unlikely(ZERO_OR_NULL_PTR(cachep))) 3664 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3665 return cachep; 3665 return cachep;
3666 ret = kmem_cache_alloc_node_notrace(cachep, flags, node); 3666 ret = kmem_cache_alloc_node_notrace(cachep, flags, node);
3667 3667
3668 trace_kmalloc_node((unsigned long) caller, ret, 3668 trace_kmalloc_node((unsigned long) caller, ret,
3669 size, cachep->buffer_size, flags, node); 3669 size, cachep->buffer_size, flags, node);
3670 3670
3671 return ret; 3671 return ret;
3672 } 3672 }
3673 3673
3674 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) 3674 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING)
3675 void *__kmalloc_node(size_t size, gfp_t flags, int node) 3675 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3676 { 3676 {
3677 return __do_kmalloc_node(size, flags, node, 3677 return __do_kmalloc_node(size, flags, node,
3678 __builtin_return_address(0)); 3678 __builtin_return_address(0));
3679 } 3679 }
3680 EXPORT_SYMBOL(__kmalloc_node); 3680 EXPORT_SYMBOL(__kmalloc_node);
3681 3681
3682 void *__kmalloc_node_track_caller(size_t size, gfp_t flags, 3682 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
3683 int node, unsigned long caller) 3683 int node, unsigned long caller)
3684 { 3684 {
3685 return __do_kmalloc_node(size, flags, node, (void *)caller); 3685 return __do_kmalloc_node(size, flags, node, (void *)caller);
3686 } 3686 }
3687 EXPORT_SYMBOL(__kmalloc_node_track_caller); 3687 EXPORT_SYMBOL(__kmalloc_node_track_caller);
3688 #else 3688 #else
3689 void *__kmalloc_node(size_t size, gfp_t flags, int node) 3689 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3690 { 3690 {
3691 return __do_kmalloc_node(size, flags, node, NULL); 3691 return __do_kmalloc_node(size, flags, node, NULL);
3692 } 3692 }
3693 EXPORT_SYMBOL(__kmalloc_node); 3693 EXPORT_SYMBOL(__kmalloc_node);
3694 #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */ 3694 #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */
3695 #endif /* CONFIG_NUMA */ 3695 #endif /* CONFIG_NUMA */
3696 3696
3697 /** 3697 /**
3698 * __do_kmalloc - allocate memory 3698 * __do_kmalloc - allocate memory
3699 * @size: how many bytes of memory are required. 3699 * @size: how many bytes of memory are required.
3700 * @flags: the type of memory to allocate (see kmalloc). 3700 * @flags: the type of memory to allocate (see kmalloc).
3701 * @caller: function caller for debug tracking of the caller 3701 * @caller: function caller for debug tracking of the caller
3702 */ 3702 */
3703 static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, 3703 static __always_inline void *__do_kmalloc(size_t size, gfp_t flags,
3704 void *caller) 3704 void *caller)
3705 { 3705 {
3706 struct kmem_cache *cachep; 3706 struct kmem_cache *cachep;
3707 void *ret; 3707 void *ret;
3708 3708
3709 /* If you want to save a few bytes .text space: replace 3709 /* If you want to save a few bytes .text space: replace
3710 * __ with kmem_. 3710 * __ with kmem_.
3711 * Then kmalloc uses the uninlined functions instead of the inline 3711 * Then kmalloc uses the uninlined functions instead of the inline
3712 * functions. 3712 * functions.
3713 */ 3713 */
3714 cachep = __find_general_cachep(size, flags); 3714 cachep = __find_general_cachep(size, flags);
3715 if (unlikely(ZERO_OR_NULL_PTR(cachep))) 3715 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3716 return cachep; 3716 return cachep;
3717 ret = __cache_alloc(cachep, flags, caller); 3717 ret = __cache_alloc(cachep, flags, caller);
3718 3718
3719 trace_kmalloc((unsigned long) caller, ret, 3719 trace_kmalloc((unsigned long) caller, ret,
3720 size, cachep->buffer_size, flags); 3720 size, cachep->buffer_size, flags);
3721 3721
3722 return ret; 3722 return ret;
3723 } 3723 }
3724 3724
3725 3725
3726 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) 3726 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING)
3727 void *__kmalloc(size_t size, gfp_t flags) 3727 void *__kmalloc(size_t size, gfp_t flags)
3728 { 3728 {
3729 return __do_kmalloc(size, flags, __builtin_return_address(0)); 3729 return __do_kmalloc(size, flags, __builtin_return_address(0));
3730 } 3730 }
3731 EXPORT_SYMBOL(__kmalloc); 3731 EXPORT_SYMBOL(__kmalloc);
3732 3732
3733 void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) 3733 void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller)
3734 { 3734 {
3735 return __do_kmalloc(size, flags, (void *)caller); 3735 return __do_kmalloc(size, flags, (void *)caller);
3736 } 3736 }
3737 EXPORT_SYMBOL(__kmalloc_track_caller); 3737 EXPORT_SYMBOL(__kmalloc_track_caller);
3738 3738
3739 #else 3739 #else
3740 void *__kmalloc(size_t size, gfp_t flags) 3740 void *__kmalloc(size_t size, gfp_t flags)
3741 { 3741 {
3742 return __do_kmalloc(size, flags, NULL); 3742 return __do_kmalloc(size, flags, NULL);
3743 } 3743 }
3744 EXPORT_SYMBOL(__kmalloc); 3744 EXPORT_SYMBOL(__kmalloc);
3745 #endif 3745 #endif
3746 3746
3747 /** 3747 /**
3748 * kmem_cache_free - Deallocate an object 3748 * kmem_cache_free - Deallocate an object
3749 * @cachep: The cache the allocation was from. 3749 * @cachep: The cache the allocation was from.
3750 * @objp: The previously allocated object. 3750 * @objp: The previously allocated object.
3751 * 3751 *
3752 * Free an object which was previously allocated from this 3752 * Free an object which was previously allocated from this
3753 * cache. 3753 * cache.
3754 */ 3754 */
3755 void kmem_cache_free(struct kmem_cache *cachep, void *objp) 3755 void kmem_cache_free(struct kmem_cache *cachep, void *objp)
3756 { 3756 {
3757 unsigned long flags; 3757 unsigned long flags;
3758 3758
3759 local_irq_save(flags); 3759 local_irq_save(flags);
3760 debug_check_no_locks_freed(objp, obj_size(cachep)); 3760 debug_check_no_locks_freed(objp, obj_size(cachep));
3761 if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) 3761 if (!(cachep->flags & SLAB_DEBUG_OBJECTS))
3762 debug_check_no_obj_freed(objp, obj_size(cachep)); 3762 debug_check_no_obj_freed(objp, obj_size(cachep));
3763 __cache_free(cachep, objp); 3763 __cache_free(cachep, objp);
3764 local_irq_restore(flags); 3764 local_irq_restore(flags);
3765 3765
3766 trace_kmem_cache_free(_RET_IP_, objp); 3766 trace_kmem_cache_free(_RET_IP_, objp);
3767 } 3767 }
3768 EXPORT_SYMBOL(kmem_cache_free); 3768 EXPORT_SYMBOL(kmem_cache_free);
3769 3769
3770 /** 3770 /**
3771 * kfree - free previously allocated memory 3771 * kfree - free previously allocated memory
3772 * @objp: pointer returned by kmalloc. 3772 * @objp: pointer returned by kmalloc.
3773 * 3773 *
3774 * If @objp is NULL, no operation is performed. 3774 * If @objp is NULL, no operation is performed.
3775 * 3775 *
3776 * Don't free memory not originally allocated by kmalloc() 3776 * Don't free memory not originally allocated by kmalloc()
3777 * or you will run into trouble. 3777 * or you will run into trouble.
3778 */ 3778 */
3779 void kfree(const void *objp) 3779 void kfree(const void *objp)
3780 { 3780 {
3781 struct kmem_cache *c; 3781 struct kmem_cache *c;
3782 unsigned long flags; 3782 unsigned long flags;
3783 3783
3784 trace_kfree(_RET_IP_, objp); 3784 trace_kfree(_RET_IP_, objp);
3785 3785
3786 if (unlikely(ZERO_OR_NULL_PTR(objp))) 3786 if (unlikely(ZERO_OR_NULL_PTR(objp)))
3787 return; 3787 return;
3788 local_irq_save(flags); 3788 local_irq_save(flags);
3789 kfree_debugcheck(objp); 3789 kfree_debugcheck(objp);
3790 c = virt_to_cache(objp); 3790 c = virt_to_cache(objp);
3791 debug_check_no_locks_freed(objp, obj_size(c)); 3791 debug_check_no_locks_freed(objp, obj_size(c));
3792 debug_check_no_obj_freed(objp, obj_size(c)); 3792 debug_check_no_obj_freed(objp, obj_size(c));
3793 __cache_free(c, (void *)objp); 3793 __cache_free(c, (void *)objp);
3794 local_irq_restore(flags); 3794 local_irq_restore(flags);
3795 } 3795 }
3796 EXPORT_SYMBOL(kfree); 3796 EXPORT_SYMBOL(kfree);
3797 3797
3798 unsigned int kmem_cache_size(struct kmem_cache *cachep) 3798 unsigned int kmem_cache_size(struct kmem_cache *cachep)
3799 { 3799 {
3800 return obj_size(cachep); 3800 return obj_size(cachep);
3801 } 3801 }
3802 EXPORT_SYMBOL(kmem_cache_size); 3802 EXPORT_SYMBOL(kmem_cache_size);
3803 3803
3804 const char *kmem_cache_name(struct kmem_cache *cachep) 3804 const char *kmem_cache_name(struct kmem_cache *cachep)
3805 { 3805 {
3806 return cachep->name; 3806 return cachep->name;
3807 } 3807 }
3808 EXPORT_SYMBOL_GPL(kmem_cache_name); 3808 EXPORT_SYMBOL_GPL(kmem_cache_name);
3809 3809
3810 /* 3810 /*
3811 * This initializes kmem_list3 or resizes various caches for all nodes. 3811 * This initializes kmem_list3 or resizes various caches for all nodes.
3812 */ 3812 */
3813 static int alloc_kmemlist(struct kmem_cache *cachep, gfp_t gfp) 3813 static int alloc_kmemlist(struct kmem_cache *cachep, gfp_t gfp)
3814 { 3814 {
3815 int node; 3815 int node;
3816 struct kmem_list3 *l3; 3816 struct kmem_list3 *l3;
3817 struct array_cache *new_shared; 3817 struct array_cache *new_shared;
3818 struct array_cache **new_alien = NULL; 3818 struct array_cache **new_alien = NULL;
3819 3819
3820 for_each_online_node(node) { 3820 for_each_online_node(node) {
3821 3821
3822 if (use_alien_caches) { 3822 if (use_alien_caches) {
3823 new_alien = alloc_alien_cache(node, cachep->limit, gfp); 3823 new_alien = alloc_alien_cache(node, cachep->limit, gfp);
3824 if (!new_alien) 3824 if (!new_alien)
3825 goto fail; 3825 goto fail;
3826 } 3826 }
3827 3827
3828 new_shared = NULL; 3828 new_shared = NULL;
3829 if (cachep->shared) { 3829 if (cachep->shared) {
3830 new_shared = alloc_arraycache(node, 3830 new_shared = alloc_arraycache(node,
3831 cachep->shared*cachep->batchcount, 3831 cachep->shared*cachep->batchcount,
3832 0xbaadf00d, gfp); 3832 0xbaadf00d, gfp);
3833 if (!new_shared) { 3833 if (!new_shared) {
3834 free_alien_cache(new_alien); 3834 free_alien_cache(new_alien);
3835 goto fail; 3835 goto fail;
3836 } 3836 }
3837 } 3837 }
3838 3838
3839 l3 = cachep->nodelists[node]; 3839 l3 = cachep->nodelists[node];
3840 if (l3) { 3840 if (l3) {
3841 struct array_cache *shared = l3->shared; 3841 struct array_cache *shared = l3->shared;
3842 3842
3843 spin_lock_irq(&l3->list_lock); 3843 spin_lock_irq(&l3->list_lock);
3844 3844
3845 if (shared) 3845 if (shared)
3846 free_block(cachep, shared->entry, 3846 free_block(cachep, shared->entry,
3847 shared->avail, node); 3847 shared->avail, node);
3848 3848
3849 l3->shared = new_shared; 3849 l3->shared = new_shared;
3850 if (!l3->alien) { 3850 if (!l3->alien) {
3851 l3->alien = new_alien; 3851 l3->alien = new_alien;
3852 new_alien = NULL; 3852 new_alien = NULL;
3853 } 3853 }
3854 l3->free_limit = (1 + nr_cpus_node(node)) * 3854 l3->free_limit = (1 + nr_cpus_node(node)) *
3855 cachep->batchcount + cachep->num; 3855 cachep->batchcount + cachep->num;
3856 spin_unlock_irq(&l3->list_lock); 3856 spin_unlock_irq(&l3->list_lock);
3857 kfree(shared); 3857 kfree(shared);
3858 free_alien_cache(new_alien); 3858 free_alien_cache(new_alien);
3859 continue; 3859 continue;
3860 } 3860 }
3861 l3 = kmalloc_node(sizeof(struct kmem_list3), gfp, node); 3861 l3 = kmalloc_node(sizeof(struct kmem_list3), gfp, node);
3862 if (!l3) { 3862 if (!l3) {
3863 free_alien_cache(new_alien); 3863 free_alien_cache(new_alien);
3864 kfree(new_shared); 3864 kfree(new_shared);
3865 goto fail; 3865 goto fail;
3866 } 3866 }
3867 3867
3868 kmem_list3_init(l3); 3868 kmem_list3_init(l3);
3869 l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + 3869 l3->next_reap = jiffies + REAPTIMEOUT_LIST3 +
3870 ((unsigned long)cachep) % REAPTIMEOUT_LIST3; 3870 ((unsigned long)cachep) % REAPTIMEOUT_LIST3;
3871 l3->shared = new_shared; 3871 l3->shared = new_shared;
3872 l3->alien = new_alien; 3872 l3->alien = new_alien;
3873 l3->free_limit = (1 + nr_cpus_node(node)) * 3873 l3->free_limit = (1 + nr_cpus_node(node)) *
3874 cachep->batchcount + cachep->num; 3874 cachep->batchcount + cachep->num;
3875 cachep->nodelists[node] = l3; 3875 cachep->nodelists[node] = l3;
3876 } 3876 }
3877 return 0; 3877 return 0;
3878 3878
3879 fail: 3879 fail:
3880 if (!cachep->next.next) { 3880 if (!cachep->next.next) {
3881 /* Cache is not active yet. Roll back what we did */ 3881 /* Cache is not active yet. Roll back what we did */
3882 node--; 3882 node--;
3883 while (node >= 0) { 3883 while (node >= 0) {
3884 if (cachep->nodelists[node]) { 3884 if (cachep->nodelists[node]) {
3885 l3 = cachep->nodelists[node]; 3885 l3 = cachep->nodelists[node];
3886 3886
3887 kfree(l3->shared); 3887 kfree(l3->shared);
3888 free_alien_cache(l3->alien); 3888 free_alien_cache(l3->alien);
3889 kfree(l3); 3889 kfree(l3);
3890 cachep->nodelists[node] = NULL; 3890 cachep->nodelists[node] = NULL;
3891 } 3891 }
3892 node--; 3892 node--;
3893 } 3893 }
3894 } 3894 }
3895 return -ENOMEM; 3895 return -ENOMEM;
3896 } 3896 }
3897 3897
3898 struct ccupdate_struct { 3898 struct ccupdate_struct {
3899 struct kmem_cache *cachep; 3899 struct kmem_cache *cachep;
3900 struct array_cache *new[NR_CPUS]; 3900 struct array_cache *new[NR_CPUS];
3901 }; 3901 };
3902 3902
3903 static void do_ccupdate_local(void *info) 3903 static void do_ccupdate_local(void *info)
3904 { 3904 {
3905 struct ccupdate_struct *new = info; 3905 struct ccupdate_struct *new = info;
3906 struct array_cache *old; 3906 struct array_cache *old;
3907 3907
3908 check_irq_off(); 3908 check_irq_off();
3909 old = cpu_cache_get(new->cachep); 3909 old = cpu_cache_get(new->cachep);
3910 3910
3911 new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; 3911 new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()];
3912 new->new[smp_processor_id()] = old; 3912 new->new[smp_processor_id()] = old;
3913 } 3913 }
3914 3914
3915 /* Always called with the cache_chain_mutex held */ 3915 /* Always called with the cache_chain_mutex held */
3916 static int do_tune_cpucache(struct kmem_cache *cachep, int limit, 3916 static int do_tune_cpucache(struct kmem_cache *cachep, int limit,
3917 int batchcount, int shared, gfp_t gfp) 3917 int batchcount, int shared, gfp_t gfp)
3918 { 3918 {
3919 struct ccupdate_struct *new; 3919 struct ccupdate_struct *new;
3920 int i; 3920 int i;
3921 3921
3922 new = kzalloc(sizeof(*new), gfp); 3922 new = kzalloc(sizeof(*new), gfp);
3923 if (!new) 3923 if (!new)
3924 return -ENOMEM; 3924 return -ENOMEM;
3925 3925
3926 for_each_online_cpu(i) { 3926 for_each_online_cpu(i) {
3927 new->new[i] = alloc_arraycache(cpu_to_node(i), limit, 3927 new->new[i] = alloc_arraycache(cpu_to_node(i), limit,
3928 batchcount, gfp); 3928 batchcount, gfp);
3929 if (!new->new[i]) { 3929 if (!new->new[i]) {
3930 for (i--; i >= 0; i--) 3930 for (i--; i >= 0; i--)
3931 kfree(new->new[i]); 3931 kfree(new->new[i]);
3932 kfree(new); 3932 kfree(new);
3933 return -ENOMEM; 3933 return -ENOMEM;
3934 } 3934 }
3935 } 3935 }
3936 new->cachep = cachep; 3936 new->cachep = cachep;
3937 3937
3938 on_each_cpu(do_ccupdate_local, (void *)new, 1); 3938 on_each_cpu(do_ccupdate_local, (void *)new, 1);
3939 3939
3940 check_irq_on(); 3940 check_irq_on();
3941 cachep->batchcount = batchcount; 3941 cachep->batchcount = batchcount;
3942 cachep->limit = limit; 3942 cachep->limit = limit;
3943 cachep->shared = shared; 3943 cachep->shared = shared;
3944 3944
3945 for_each_online_cpu(i) { 3945 for_each_online_cpu(i) {
3946 struct array_cache *ccold = new->new[i]; 3946 struct array_cache *ccold = new->new[i];
3947 if (!ccold) 3947 if (!ccold)
3948 continue; 3948 continue;
3949 spin_lock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); 3949 spin_lock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock);
3950 free_block(cachep, ccold->entry, ccold->avail, cpu_to_node(i)); 3950 free_block(cachep, ccold->entry, ccold->avail, cpu_to_node(i));
3951 spin_unlock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); 3951 spin_unlock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock);
3952 kfree(ccold); 3952 kfree(ccold);
3953 } 3953 }
3954 kfree(new); 3954 kfree(new);
3955 return alloc_kmemlist(cachep, gfp); 3955 return alloc_kmemlist(cachep, gfp);
3956 } 3956 }
3957 3957
3958 /* Called with cache_chain_mutex held always */ 3958 /* Called with cache_chain_mutex held always */
3959 static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) 3959 static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp)
3960 { 3960 {
3961 int err; 3961 int err;
3962 int limit, shared; 3962 int limit, shared;
3963 3963
3964 /* 3964 /*
3965 * The head array serves three purposes: 3965 * The head array serves three purposes:
3966 * - create a LIFO ordering, i.e. return objects that are cache-warm 3966 * - create a LIFO ordering, i.e. return objects that are cache-warm
3967 * - reduce the number of spinlock operations. 3967 * - reduce the number of spinlock operations.
3968 * - reduce the number of linked list operations on the slab and 3968 * - reduce the number of linked list operations on the slab and
3969 * bufctl chains: array operations are cheaper. 3969 * bufctl chains: array operations are cheaper.
3970 * The numbers are guessed, we should auto-tune as described by 3970 * The numbers are guessed, we should auto-tune as described by
3971 * Bonwick. 3971 * Bonwick.
3972 */ 3972 */
3973 if (cachep->buffer_size > 131072) 3973 if (cachep->buffer_size > 131072)
3974 limit = 1; 3974 limit = 1;
3975 else if (cachep->buffer_size > PAGE_SIZE) 3975 else if (cachep->buffer_size > PAGE_SIZE)
3976 limit = 8; 3976 limit = 8;
3977 else if (cachep->buffer_size > 1024) 3977 else if (cachep->buffer_size > 1024)
3978 limit = 24; 3978 limit = 24;
3979 else if (cachep->buffer_size > 256) 3979 else if (cachep->buffer_size > 256)
3980 limit = 54; 3980 limit = 54;
3981 else 3981 else
3982 limit = 120; 3982 limit = 120;
3983 3983
3984 /* 3984 /*
3985 * CPU bound tasks (e.g. network routing) can exhibit cpu bound 3985 * CPU bound tasks (e.g. network routing) can exhibit cpu bound
3986 * allocation behaviour: Most allocs on one cpu, most free operations 3986 * allocation behaviour: Most allocs on one cpu, most free operations
3987 * on another cpu. For these cases, an efficient object passing between 3987 * on another cpu. For these cases, an efficient object passing between
3988 * cpus is necessary. This is provided by a shared array. The array 3988 * cpus is necessary. This is provided by a shared array. The array
3989 * replaces Bonwick's magazine layer. 3989 * replaces Bonwick's magazine layer.
3990 * On uniprocessor, it's functionally equivalent (but less efficient) 3990 * On uniprocessor, it's functionally equivalent (but less efficient)
3991 * to a larger limit. Thus disabled by default. 3991 * to a larger limit. Thus disabled by default.
3992 */ 3992 */
3993 shared = 0; 3993 shared = 0;
3994 if (cachep->buffer_size <= PAGE_SIZE && num_possible_cpus() > 1) 3994 if (cachep->buffer_size <= PAGE_SIZE && num_possible_cpus() > 1)
3995 shared = 8; 3995 shared = 8;
3996 3996
3997 #if DEBUG 3997 #if DEBUG
3998 /* 3998 /*
3999 * With debugging enabled, large batchcount lead to excessively long 3999 * With debugging enabled, large batchcount lead to excessively long
4000 * periods with disabled local interrupts. Limit the batchcount 4000 * periods with disabled local interrupts. Limit the batchcount
4001 */ 4001 */
4002 if (limit > 32) 4002 if (limit > 32)
4003 limit = 32; 4003 limit = 32;
4004 #endif 4004 #endif
4005 err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared, gfp); 4005 err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared, gfp);
4006 if (err) 4006 if (err)
4007 printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", 4007 printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n",
4008 cachep->name, -err); 4008 cachep->name, -err);
4009 return err; 4009 return err;
4010 } 4010 }
4011 4011
4012 /* 4012 /*
4013 * Drain an array if it contains any elements taking the l3 lock only if 4013 * Drain an array if it contains any elements taking the l3 lock only if
4014 * necessary. Note that the l3 listlock also protects the array_cache 4014 * necessary. Note that the l3 listlock also protects the array_cache
4015 * if drain_array() is used on the shared array. 4015 * if drain_array() is used on the shared array.
4016 */ 4016 */
4017 void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, 4017 void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3,
4018 struct array_cache *ac, int force, int node) 4018 struct array_cache *ac, int force, int node)
4019 { 4019 {
4020 int tofree; 4020 int tofree;
4021 4021
4022 if (!ac || !ac->avail) 4022 if (!ac || !ac->avail)
4023 return; 4023 return;
4024 if (ac->touched && !force) { 4024 if (ac->touched && !force) {
4025 ac->touched = 0; 4025 ac->touched = 0;
4026 } else { 4026 } else {
4027 spin_lock_irq(&l3->list_lock); 4027 spin_lock_irq(&l3->list_lock);
4028 if (ac->avail) { 4028 if (ac->avail) {
4029 tofree = force ? ac->avail : (ac->limit + 4) / 5; 4029 tofree = force ? ac->avail : (ac->limit + 4) / 5;
4030 if (tofree > ac->avail) 4030 if (tofree > ac->avail)
4031 tofree = (ac->avail + 1) / 2; 4031 tofree = (ac->avail + 1) / 2;
4032 free_block(cachep, ac->entry, tofree, node); 4032 free_block(cachep, ac->entry, tofree, node);
4033 ac->avail -= tofree; 4033 ac->avail -= tofree;
4034 memmove(ac->entry, &(ac->entry[tofree]), 4034 memmove(ac->entry, &(ac->entry[tofree]),
4035 sizeof(void *) * ac->avail); 4035 sizeof(void *) * ac->avail);
4036 } 4036 }
4037 spin_unlock_irq(&l3->list_lock); 4037 spin_unlock_irq(&l3->list_lock);
4038 } 4038 }
4039 } 4039 }
4040 4040
4041 /** 4041 /**
4042 * cache_reap - Reclaim memory from caches. 4042 * cache_reap - Reclaim memory from caches.
4043 * @w: work descriptor 4043 * @w: work descriptor
4044 * 4044 *
4045 * Called from workqueue/eventd every few seconds. 4045 * Called from workqueue/eventd every few seconds.
4046 * Purpose: 4046 * Purpose:
4047 * - clear the per-cpu caches for this CPU. 4047 * - clear the per-cpu caches for this CPU.
4048 * - return freeable pages to the main free memory pool. 4048 * - return freeable pages to the main free memory pool.
4049 * 4049 *
4050 * If we cannot acquire the cache chain mutex then just give up - we'll try 4050 * If we cannot acquire the cache chain mutex then just give up - we'll try
4051 * again on the next iteration. 4051 * again on the next iteration.
4052 */ 4052 */
4053 static void cache_reap(struct work_struct *w) 4053 static void cache_reap(struct work_struct *w)
4054 { 4054 {
4055 struct kmem_cache *searchp; 4055 struct kmem_cache *searchp;
4056 struct kmem_list3 *l3; 4056 struct kmem_list3 *l3;
4057 int node = numa_node_id(); 4057 int node = numa_node_id();
4058 struct delayed_work *work = to_delayed_work(w); 4058 struct delayed_work *work = to_delayed_work(w);
4059 4059
4060 if (!mutex_trylock(&cache_chain_mutex)) 4060 if (!mutex_trylock(&cache_chain_mutex))
4061 /* Give up. Setup the next iteration. */ 4061 /* Give up. Setup the next iteration. */
4062 goto out; 4062 goto out;
4063 4063
4064 list_for_each_entry(searchp, &cache_chain, next) { 4064 list_for_each_entry(searchp, &cache_chain, next) {
4065 check_irq_on(); 4065 check_irq_on();
4066 4066
4067 /* 4067 /*
4068 * We only take the l3 lock if absolutely necessary and we 4068 * We only take the l3 lock if absolutely necessary and we
4069 * have established with reasonable certainty that 4069 * have established with reasonable certainty that
4070 * we can do some work if the lock was obtained. 4070 * we can do some work if the lock was obtained.
4071 */ 4071 */
4072 l3 = searchp->nodelists[node]; 4072 l3 = searchp->nodelists[node];
4073 4073
4074 reap_alien(searchp, l3); 4074 reap_alien(searchp, l3);
4075 4075
4076 drain_array(searchp, l3, cpu_cache_get(searchp), 0, node); 4076 drain_array(searchp, l3, cpu_cache_get(searchp), 0, node);
4077 4077
4078 /* 4078 /*
4079 * These are racy checks but it does not matter 4079 * These are racy checks but it does not matter
4080 * if we skip one check or scan twice. 4080 * if we skip one check or scan twice.
4081 */ 4081 */
4082 if (time_after(l3->next_reap, jiffies)) 4082 if (time_after(l3->next_reap, jiffies))
4083 goto next; 4083 goto next;
4084 4084
4085 l3->next_reap = jiffies + REAPTIMEOUT_LIST3; 4085 l3->next_reap = jiffies + REAPTIMEOUT_LIST3;
4086 4086
4087 drain_array(searchp, l3, l3->shared, 0, node); 4087 drain_array(searchp, l3, l3->shared, 0, node);
4088 4088
4089 if (l3->free_touched) 4089 if (l3->free_touched)
4090 l3->free_touched = 0; 4090 l3->free_touched = 0;
4091 else { 4091 else {
4092 int freed; 4092 int freed;
4093 4093
4094 freed = drain_freelist(searchp, l3, (l3->free_limit + 4094 freed = drain_freelist(searchp, l3, (l3->free_limit +
4095 5 * searchp->num - 1) / (5 * searchp->num)); 4095 5 * searchp->num - 1) / (5 * searchp->num));
4096 STATS_ADD_REAPED(searchp, freed); 4096 STATS_ADD_REAPED(searchp, freed);
4097 } 4097 }
4098 next: 4098 next:
4099 cond_resched(); 4099 cond_resched();
4100 } 4100 }
4101 check_irq_on(); 4101 check_irq_on();
4102 mutex_unlock(&cache_chain_mutex); 4102 mutex_unlock(&cache_chain_mutex);
4103 next_reap_node(); 4103 next_reap_node();
4104 out: 4104 out:
4105 /* Set up the next iteration */ 4105 /* Set up the next iteration */
4106 schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_CPUC)); 4106 schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_CPUC));
4107 } 4107 }
4108 4108
4109 #ifdef CONFIG_SLABINFO 4109 #ifdef CONFIG_SLABINFO
4110 4110
4111 static void print_slabinfo_header(struct seq_file *m) 4111 static void print_slabinfo_header(struct seq_file *m)
4112 { 4112 {
4113 /* 4113 /*
4114 * Output format version, so at least we can change it 4114 * Output format version, so at least we can change it
4115 * without _too_ many complaints. 4115 * without _too_ many complaints.
4116 */ 4116 */
4117 #if STATS 4117 #if STATS
4118 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); 4118 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
4119 #else 4119 #else
4120 seq_puts(m, "slabinfo - version: 2.1\n"); 4120 seq_puts(m, "slabinfo - version: 2.1\n");
4121 #endif 4121 #endif
4122 seq_puts(m, "# name <active_objs> <num_objs> <objsize> " 4122 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4123 "<objperslab> <pagesperslab>"); 4123 "<objperslab> <pagesperslab>");
4124 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 4124 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4125 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 4125 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4126 #if STATS 4126 #if STATS
4127 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " 4127 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
4128 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); 4128 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
4129 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); 4129 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
4130 #endif 4130 #endif
4131 seq_putc(m, '\n'); 4131 seq_putc(m, '\n');
4132 } 4132 }
4133 4133
4134 static void *s_start(struct seq_file *m, loff_t *pos) 4134 static void *s_start(struct seq_file *m, loff_t *pos)
4135 { 4135 {
4136 loff_t n = *pos; 4136 loff_t n = *pos;
4137 4137
4138 mutex_lock(&cache_chain_mutex); 4138 mutex_lock(&cache_chain_mutex);
4139 if (!n) 4139 if (!n)
4140 print_slabinfo_header(m); 4140 print_slabinfo_header(m);
4141 4141
4142 return seq_list_start(&cache_chain, *pos); 4142 return seq_list_start(&cache_chain, *pos);
4143 } 4143 }
4144 4144
4145 static void *s_next(struct seq_file *m, void *p, loff_t *pos) 4145 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4146 { 4146 {
4147 return seq_list_next(p, &cache_chain, pos); 4147 return seq_list_next(p, &cache_chain, pos);
4148 } 4148 }
4149 4149
4150 static void s_stop(struct seq_file *m, void *p) 4150 static void s_stop(struct seq_file *m, void *p)
4151 { 4151 {
4152 mutex_unlock(&cache_chain_mutex); 4152 mutex_unlock(&cache_chain_mutex);
4153 } 4153 }
4154 4154
4155 static int s_show(struct seq_file *m, void *p) 4155 static int s_show(struct seq_file *m, void *p)
4156 { 4156 {
4157 struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next); 4157 struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next);
4158 struct slab *slabp; 4158 struct slab *slabp;
4159 unsigned long active_objs; 4159 unsigned long active_objs;
4160 unsigned long num_objs; 4160 unsigned long num_objs;
4161 unsigned long active_slabs = 0; 4161 unsigned long active_slabs = 0;
4162 unsigned long num_slabs, free_objects = 0, shared_avail = 0; 4162 unsigned long num_slabs, free_objects = 0, shared_avail = 0;
4163 const char *name; 4163 const char *name;
4164 char *error = NULL; 4164 char *error = NULL;
4165 int node; 4165 int node;
4166 struct kmem_list3 *l3; 4166 struct kmem_list3 *l3;
4167 4167
4168 active_objs = 0; 4168 active_objs = 0;
4169 num_slabs = 0; 4169 num_slabs = 0;
4170 for_each_online_node(node) { 4170 for_each_online_node(node) {
4171 l3 = cachep->nodelists[node]; 4171 l3 = cachep->nodelists[node];
4172 if (!l3) 4172 if (!l3)
4173 continue; 4173 continue;
4174 4174
4175 check_irq_on(); 4175 check_irq_on();
4176 spin_lock_irq(&l3->list_lock); 4176 spin_lock_irq(&l3->list_lock);
4177 4177
4178 list_for_each_entry(slabp, &l3->slabs_full, list) { 4178 list_for_each_entry(slabp, &l3->slabs_full, list) {
4179 if (slabp->inuse != cachep->num && !error) 4179 if (slabp->inuse != cachep->num && !error)
4180 error = "slabs_full accounting error"; 4180 error = "slabs_full accounting error";
4181 active_objs += cachep->num; 4181 active_objs += cachep->num;
4182 active_slabs++; 4182 active_slabs++;
4183 } 4183 }
4184 list_for_each_entry(slabp, &l3->slabs_partial, list) { 4184 list_for_each_entry(slabp, &l3->slabs_partial, list) {
4185 if (slabp->inuse == cachep->num && !error) 4185 if (slabp->inuse == cachep->num && !error)
4186 error = "slabs_partial inuse accounting error"; 4186 error = "slabs_partial inuse accounting error";
4187 if (!slabp->inuse && !error) 4187 if (!slabp->inuse && !error)
4188 error = "slabs_partial/inuse accounting error"; 4188 error = "slabs_partial/inuse accounting error";
4189 active_objs += slabp->inuse; 4189 active_objs += slabp->inuse;
4190 active_slabs++; 4190 active_slabs++;
4191 } 4191 }
4192 list_for_each_entry(slabp, &l3->slabs_free, list) { 4192 list_for_each_entry(slabp, &l3->slabs_free, list) {
4193 if (slabp->inuse && !error) 4193 if (slabp->inuse && !error)
4194 error = "slabs_free/inuse accounting error"; 4194 error = "slabs_free/inuse accounting error";
4195 num_slabs++; 4195 num_slabs++;
4196 } 4196 }
4197 free_objects += l3->free_objects; 4197 free_objects += l3->free_objects;
4198 if (l3->shared) 4198 if (l3->shared)
4199 shared_avail += l3->shared->avail; 4199 shared_avail += l3->shared->avail;
4200 4200
4201 spin_unlock_irq(&l3->list_lock); 4201 spin_unlock_irq(&l3->list_lock);
4202 } 4202 }
4203 num_slabs += active_slabs; 4203 num_slabs += active_slabs;
4204 num_objs = num_slabs * cachep->num; 4204 num_objs = num_slabs * cachep->num;
4205 if (num_objs - active_objs != free_objects && !error) 4205 if (num_objs - active_objs != free_objects && !error)
4206 error = "free_objects accounting error"; 4206 error = "free_objects accounting error";
4207 4207
4208 name = cachep->name; 4208 name = cachep->name;
4209 if (error) 4209 if (error)
4210 printk(KERN_ERR "slab: cache %s error: %s\n", name, error); 4210 printk(KERN_ERR "slab: cache %s error: %s\n", name, error);
4211 4211
4212 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 4212 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
4213 name, active_objs, num_objs, cachep->buffer_size, 4213 name, active_objs, num_objs, cachep->buffer_size,
4214 cachep->num, (1 << cachep->gfporder)); 4214 cachep->num, (1 << cachep->gfporder));
4215 seq_printf(m, " : tunables %4u %4u %4u", 4215 seq_printf(m, " : tunables %4u %4u %4u",
4216 cachep->limit, cachep->batchcount, cachep->shared); 4216 cachep->limit, cachep->batchcount, cachep->shared);
4217 seq_printf(m, " : slabdata %6lu %6lu %6lu", 4217 seq_printf(m, " : slabdata %6lu %6lu %6lu",
4218 active_slabs, num_slabs, shared_avail); 4218 active_slabs, num_slabs, shared_avail);
4219 #if STATS 4219 #if STATS
4220 { /* list3 stats */ 4220 { /* list3 stats */
4221 unsigned long high = cachep->high_mark; 4221 unsigned long high = cachep->high_mark;
4222 unsigned long allocs = cachep->num_allocations; 4222 unsigned long allocs = cachep->num_allocations;
4223 unsigned long grown = cachep->grown; 4223 unsigned long grown = cachep->grown;
4224 unsigned long reaped = cachep->reaped; 4224 unsigned long reaped = cachep->reaped;
4225 unsigned long errors = cachep->errors; 4225 unsigned long errors = cachep->errors;
4226 unsigned long max_freeable = cachep->max_freeable; 4226 unsigned long max_freeable = cachep->max_freeable;
4227 unsigned long node_allocs = cachep->node_allocs; 4227 unsigned long node_allocs = cachep->node_allocs;
4228 unsigned long node_frees = cachep->node_frees; 4228 unsigned long node_frees = cachep->node_frees;
4229 unsigned long overflows = cachep->node_overflow; 4229 unsigned long overflows = cachep->node_overflow;
4230 4230
4231 seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu \ 4231 seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu \
4232 %4lu %4lu %4lu %4lu %4lu", allocs, high, grown, 4232 %4lu %4lu %4lu %4lu %4lu", allocs, high, grown,
4233 reaped, errors, max_freeable, node_allocs, 4233 reaped, errors, max_freeable, node_allocs,
4234 node_frees, overflows); 4234 node_frees, overflows);
4235 } 4235 }
4236 /* cpu stats */ 4236 /* cpu stats */
4237 { 4237 {
4238 unsigned long allochit = atomic_read(&cachep->allochit); 4238 unsigned long allochit = atomic_read(&cachep->allochit);
4239 unsigned long allocmiss = atomic_read(&cachep->allocmiss); 4239 unsigned long allocmiss = atomic_read(&cachep->allocmiss);
4240 unsigned long freehit = atomic_read(&cachep->freehit); 4240 unsigned long freehit = atomic_read(&cachep->freehit);
4241 unsigned long freemiss = atomic_read(&cachep->freemiss); 4241 unsigned long freemiss = atomic_read(&cachep->freemiss);
4242 4242
4243 seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", 4243 seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu",
4244 allochit, allocmiss, freehit, freemiss); 4244 allochit, allocmiss, freehit, freemiss);
4245 } 4245 }
4246 #endif 4246 #endif
4247 seq_putc(m, '\n'); 4247 seq_putc(m, '\n');
4248 return 0; 4248 return 0;
4249 } 4249 }
4250 4250
4251 /* 4251 /*
4252 * slabinfo_op - iterator that generates /proc/slabinfo 4252 * slabinfo_op - iterator that generates /proc/slabinfo
4253 * 4253 *
4254 * Output layout: 4254 * Output layout:
4255 * cache-name 4255 * cache-name
4256 * num-active-objs 4256 * num-active-objs
4257 * total-objs 4257 * total-objs
4258 * object size 4258 * object size
4259 * num-active-slabs 4259 * num-active-slabs
4260 * total-slabs 4260 * total-slabs
4261 * num-pages-per-slab 4261 * num-pages-per-slab
4262 * + further values on SMP and with statistics enabled 4262 * + further values on SMP and with statistics enabled
4263 */ 4263 */
4264 4264
4265 static const struct seq_operations slabinfo_op = { 4265 static const struct seq_operations slabinfo_op = {
4266 .start = s_start, 4266 .start = s_start,
4267 .next = s_next, 4267 .next = s_next,
4268 .stop = s_stop, 4268 .stop = s_stop,
4269 .show = s_show, 4269 .show = s_show,
4270 }; 4270 };
4271 4271
4272 #define MAX_SLABINFO_WRITE 128 4272 #define MAX_SLABINFO_WRITE 128
4273 /** 4273 /**
4274 * slabinfo_write - Tuning for the slab allocator 4274 * slabinfo_write - Tuning for the slab allocator
4275 * @file: unused 4275 * @file: unused
4276 * @buffer: user buffer 4276 * @buffer: user buffer
4277 * @count: data length 4277 * @count: data length
4278 * @ppos: unused 4278 * @ppos: unused
4279 */ 4279 */
4280 ssize_t slabinfo_write(struct file *file, const char __user * buffer, 4280 ssize_t slabinfo_write(struct file *file, const char __user * buffer,
4281 size_t count, loff_t *ppos) 4281 size_t count, loff_t *ppos)
4282 { 4282 {
4283 char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; 4283 char kbuf[MAX_SLABINFO_WRITE + 1], *tmp;
4284 int limit, batchcount, shared, res; 4284 int limit, batchcount, shared, res;
4285 struct kmem_cache *cachep; 4285 struct kmem_cache *cachep;
4286 4286
4287 if (count > MAX_SLABINFO_WRITE) 4287 if (count > MAX_SLABINFO_WRITE)
4288 return -EINVAL; 4288 return -EINVAL;
4289 if (copy_from_user(&kbuf, buffer, count)) 4289 if (copy_from_user(&kbuf, buffer, count))
4290 return -EFAULT; 4290 return -EFAULT;
4291 kbuf[MAX_SLABINFO_WRITE] = '\0'; 4291 kbuf[MAX_SLABINFO_WRITE] = '\0';
4292 4292
4293 tmp = strchr(kbuf, ' '); 4293 tmp = strchr(kbuf, ' ');
4294 if (!tmp) 4294 if (!tmp)
4295 return -EINVAL; 4295 return -EINVAL;
4296 *tmp = '\0'; 4296 *tmp = '\0';
4297 tmp++; 4297 tmp++;
4298 if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) 4298 if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3)
4299 return -EINVAL; 4299 return -EINVAL;
4300 4300
4301 /* Find the cache in the chain of caches. */ 4301 /* Find the cache in the chain of caches. */
4302 mutex_lock(&cache_chain_mutex); 4302 mutex_lock(&cache_chain_mutex);
4303 res = -EINVAL; 4303 res = -EINVAL;
4304 list_for_each_entry(cachep, &cache_chain, next) { 4304 list_for_each_entry(cachep, &cache_chain, next) {
4305 if (!strcmp(cachep->name, kbuf)) { 4305 if (!strcmp(cachep->name, kbuf)) {
4306 if (limit < 1 || batchcount < 1 || 4306 if (limit < 1 || batchcount < 1 ||
4307 batchcount > limit || shared < 0) { 4307 batchcount > limit || shared < 0) {
4308 res = 0; 4308 res = 0;
4309 } else { 4309 } else {
4310 res = do_tune_cpucache(cachep, limit, 4310 res = do_tune_cpucache(cachep, limit,
4311 batchcount, shared, 4311 batchcount, shared,
4312 GFP_KERNEL); 4312 GFP_KERNEL);
4313 } 4313 }
4314 break; 4314 break;
4315 } 4315 }
4316 } 4316 }
4317 mutex_unlock(&cache_chain_mutex); 4317 mutex_unlock(&cache_chain_mutex);
4318 if (res >= 0) 4318 if (res >= 0)
4319 res = count; 4319 res = count;
4320 return res; 4320 return res;
4321 } 4321 }
4322 4322
4323 static int slabinfo_open(struct inode *inode, struct file *file) 4323 static int slabinfo_open(struct inode *inode, struct file *file)
4324 { 4324 {
4325 return seq_open(file, &slabinfo_op); 4325 return seq_open(file, &slabinfo_op);
4326 } 4326 }
4327 4327
4328 static const struct file_operations proc_slabinfo_operations = { 4328 static const struct file_operations proc_slabinfo_operations = {
4329 .open = slabinfo_open, 4329 .open = slabinfo_open,
4330 .read = seq_read, 4330 .read = seq_read,
4331 .write = slabinfo_write, 4331 .write = slabinfo_write,
4332 .llseek = seq_lseek, 4332 .llseek = seq_lseek,
4333 .release = seq_release, 4333 .release = seq_release,
4334 }; 4334 };
4335 4335
4336 #ifdef CONFIG_DEBUG_SLAB_LEAK 4336 #ifdef CONFIG_DEBUG_SLAB_LEAK
4337 4337
4338 static void *leaks_start(struct seq_file *m, loff_t *pos) 4338 static void *leaks_start(struct seq_file *m, loff_t *pos)
4339 { 4339 {
4340 mutex_lock(&cache_chain_mutex); 4340 mutex_lock(&cache_chain_mutex);
4341 return seq_list_start(&cache_chain, *pos); 4341 return seq_list_start(&cache_chain, *pos);
4342 } 4342 }
4343 4343
4344 static inline int add_caller(unsigned long *n, unsigned long v) 4344 static inline int add_caller(unsigned long *n, unsigned long v)
4345 { 4345 {
4346 unsigned long *p; 4346 unsigned long *p;
4347 int l; 4347 int l;
4348 if (!v) 4348 if (!v)
4349 return 1; 4349 return 1;
4350 l = n[1]; 4350 l = n[1];
4351 p = n + 2; 4351 p = n + 2;
4352 while (l) { 4352 while (l) {
4353 int i = l/2; 4353 int i = l/2;
4354 unsigned long *q = p + 2 * i; 4354 unsigned long *q = p + 2 * i;
4355 if (*q == v) { 4355 if (*q == v) {
4356 q[1]++; 4356 q[1]++;
4357 return 1; 4357 return 1;
4358 } 4358 }
4359 if (*q > v) { 4359 if (*q > v) {
4360 l = i; 4360 l = i;
4361 } else { 4361 } else {
4362 p = q + 2; 4362 p = q + 2;
4363 l -= i + 1; 4363 l -= i + 1;
4364 } 4364 }
4365 } 4365 }
4366 if (++n[1] == n[0]) 4366 if (++n[1] == n[0])
4367 return 0; 4367 return 0;
4368 memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); 4368 memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n));
4369 p[0] = v; 4369 p[0] = v;
4370 p[1] = 1; 4370 p[1] = 1;
4371 return 1; 4371 return 1;
4372 } 4372 }
4373 4373
4374 static void handle_slab(unsigned long *n, struct kmem_cache *c, struct slab *s) 4374 static void handle_slab(unsigned long *n, struct kmem_cache *c, struct slab *s)
4375 { 4375 {
4376 void *p; 4376 void *p;
4377 int i; 4377 int i;
4378 if (n[0] == n[1]) 4378 if (n[0] == n[1])
4379 return; 4379 return;
4380 for (i = 0, p = s->s_mem; i < c->num; i++, p += c->buffer_size) { 4380 for (i = 0, p = s->s_mem; i < c->num; i++, p += c->buffer_size) {
4381 if (slab_bufctl(s)[i] != BUFCTL_ACTIVE) 4381 if (slab_bufctl(s)[i] != BUFCTL_ACTIVE)
4382 continue; 4382 continue;
4383 if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) 4383 if (!add_caller(n, (unsigned long)*dbg_userword(c, p)))
4384 return; 4384 return;
4385 } 4385 }
4386 } 4386 }
4387 4387
4388 static void show_symbol(struct seq_file *m, unsigned long address) 4388 static void show_symbol(struct seq_file *m, unsigned long address)
4389 { 4389 {
4390 #ifdef CONFIG_KALLSYMS 4390 #ifdef CONFIG_KALLSYMS
4391 unsigned long offset, size; 4391 unsigned long offset, size;
4392 char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; 4392 char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN];
4393 4393
4394 if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { 4394 if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) {
4395 seq_printf(m, "%s+%#lx/%#lx", name, offset, size); 4395 seq_printf(m, "%s+%#lx/%#lx", name, offset, size);
4396 if (modname[0]) 4396 if (modname[0])
4397 seq_printf(m, " [%s]", modname); 4397 seq_printf(m, " [%s]", modname);
4398 return; 4398 return;
4399 } 4399 }
4400 #endif 4400 #endif
4401 seq_printf(m, "%p", (void *)address); 4401 seq_printf(m, "%p", (void *)address);
4402 } 4402 }
4403 4403
4404 static int leaks_show(struct seq_file *m, void *p) 4404 static int leaks_show(struct seq_file *m, void *p)
4405 { 4405 {
4406 struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next); 4406 struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next);
4407 struct slab *slabp; 4407 struct slab *slabp;
4408 struct kmem_list3 *l3; 4408 struct kmem_list3 *l3;
4409 const char *name; 4409 const char *name;
4410 unsigned long *n = m->private; 4410 unsigned long *n = m->private;
4411 int node; 4411 int node;
4412 int i; 4412 int i;
4413 4413
4414 if (!(cachep->flags & SLAB_STORE_USER)) 4414 if (!(cachep->flags & SLAB_STORE_USER))
4415 return 0; 4415 return 0;
4416 if (!(cachep->flags & SLAB_RED_ZONE)) 4416 if (!(cachep->flags & SLAB_RED_ZONE))
4417 return 0; 4417 return 0;
4418 4418
4419 /* OK, we can do it */ 4419 /* OK, we can do it */
4420 4420
4421 n[1] = 0; 4421 n[1] = 0;
4422 4422
4423 for_each_online_node(node) { 4423 for_each_online_node(node) {
4424 l3 = cachep->nodelists[node]; 4424 l3 = cachep->nodelists[node];
4425 if (!l3) 4425 if (!l3)
4426 continue; 4426 continue;
4427 4427
4428 check_irq_on(); 4428 check_irq_on();
4429 spin_lock_irq(&l3->list_lock); 4429 spin_lock_irq(&l3->list_lock);
4430 4430
4431 list_for_each_entry(slabp, &l3->slabs_full, list) 4431 list_for_each_entry(slabp, &l3->slabs_full, list)
4432 handle_slab(n, cachep, slabp); 4432 handle_slab(n, cachep, slabp);
4433 list_for_each_entry(slabp, &l3->slabs_partial, list) 4433 list_for_each_entry(slabp, &l3->slabs_partial, list)
4434 handle_slab(n, cachep, slabp); 4434 handle_slab(n, cachep, slabp);
4435 spin_unlock_irq(&l3->list_lock); 4435 spin_unlock_irq(&l3->list_lock);
4436 } 4436 }
4437 name = cachep->name; 4437 name = cachep->name;
4438 if (n[0] == n[1]) { 4438 if (n[0] == n[1]) {
4439 /* Increase the buffer size */ 4439 /* Increase the buffer size */
4440 mutex_unlock(&cache_chain_mutex); 4440 mutex_unlock(&cache_chain_mutex);
4441 m->private = kzalloc(n[0] * 4 * sizeof(unsigned long), GFP_KERNEL); 4441 m->private = kzalloc(n[0] * 4 * sizeof(unsigned long), GFP_KERNEL);
4442 if (!m->private) { 4442 if (!m->private) {
4443 /* Too bad, we are really out */ 4443 /* Too bad, we are really out */
4444 m->private = n; 4444 m->private = n;
4445 mutex_lock(&cache_chain_mutex); 4445 mutex_lock(&cache_chain_mutex);
4446 return -ENOMEM; 4446 return -ENOMEM;
4447 } 4447 }
4448 *(unsigned long *)m->private = n[0] * 2; 4448 *(unsigned long *)m->private = n[0] * 2;
4449 kfree(n); 4449 kfree(n);
4450 mutex_lock(&cache_chain_mutex); 4450 mutex_lock(&cache_chain_mutex);
4451 /* Now make sure this entry will be retried */ 4451 /* Now make sure this entry will be retried */
4452 m->count = m->size; 4452 m->count = m->size;
4453 return 0; 4453 return 0;
4454 } 4454 }
4455 for (i = 0; i < n[1]; i++) { 4455 for (i = 0; i < n[1]; i++) {
4456 seq_printf(m, "%s: %lu ", name, n[2*i+3]); 4456 seq_printf(m, "%s: %lu ", name, n[2*i+3]);
4457 show_symbol(m, n[2*i+2]); 4457 show_symbol(m, n[2*i+2]);
4458 seq_putc(m, '\n'); 4458 seq_putc(m, '\n');
4459 } 4459 }
4460 4460
4461 return 0; 4461 return 0;
4462 } 4462 }
4463 4463
4464 static const struct seq_operations slabstats_op = { 4464 static const struct seq_operations slabstats_op = {
4465 .start = leaks_start, 4465 .start = leaks_start,
4466 .next = s_next, 4466 .next = s_next,
4467 .stop = s_stop, 4467 .stop = s_stop,
4468 .show = leaks_show, 4468 .show = leaks_show,
4469 }; 4469 };
4470 4470
4471 static int slabstats_open(struct inode *inode, struct file *file) 4471 static int slabstats_open(struct inode *inode, struct file *file)
4472 { 4472 {
4473 unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); 4473 unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL);
4474 int ret = -ENOMEM; 4474 int ret = -ENOMEM;
4475 if (n) { 4475 if (n) {
4476 ret = seq_open(file, &slabstats_op); 4476 ret = seq_open(file, &slabstats_op);
4477 if (!ret) { 4477 if (!ret) {
4478 struct seq_file *m = file->private_data; 4478 struct seq_file *m = file->private_data;
4479 *n = PAGE_SIZE / (2 * sizeof(unsigned long)); 4479 *n = PAGE_SIZE / (2 * sizeof(unsigned long));
4480 m->private = n; 4480 m->private = n;
4481 n = NULL; 4481 n = NULL;
4482 } 4482 }
4483 kfree(n); 4483 kfree(n);
4484 } 4484 }
4485 return ret; 4485 return ret;
4486 } 4486 }
4487 4487
4488 static const struct file_operations proc_slabstats_operations = { 4488 static const struct file_operations proc_slabstats_operations = {
4489 .open = slabstats_open, 4489 .open = slabstats_open,
4490 .read = seq_read, 4490 .read = seq_read,
4491 .llseek = seq_lseek, 4491 .llseek = seq_lseek,
4492 .release = seq_release_private, 4492 .release = seq_release_private,
4493 }; 4493 };
4494 #endif 4494 #endif
4495 4495
4496 static int __init slab_proc_init(void) 4496 static int __init slab_proc_init(void)
4497 { 4497 {
4498 proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations); 4498 proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations);
4499 #ifdef CONFIG_DEBUG_SLAB_LEAK 4499 #ifdef CONFIG_DEBUG_SLAB_LEAK
4500 proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); 4500 proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations);
4501 #endif 4501 #endif
4502 return 0; 4502 return 0;
4503 } 4503 }
4504 module_init(slab_proc_init); 4504 module_init(slab_proc_init);
4505 #endif 4505 #endif
4506 4506
4507 /** 4507 /**
4508 * ksize - get the actual amount of memory allocated for a given object 4508 * ksize - get the actual amount of memory allocated for a given object
4509 * @objp: Pointer to the object 4509 * @objp: Pointer to the object
4510 * 4510 *
4511 * kmalloc may internally round up allocations and return more memory 4511 * kmalloc may internally round up allocations and return more memory
4512 * than requested. ksize() can be used to determine the actual amount of 4512 * than requested. ksize() can be used to determine the actual amount of
4513 * memory allocated. The caller may use this additional memory, even though 4513 * memory allocated. The caller may use this additional memory, even though
4514 * a smaller amount of memory was initially specified with the kmalloc call. 4514 * a smaller amount of memory was initially specified with the kmalloc call.
4515 * The caller must guarantee that objp points to a valid object previously 4515 * The caller must guarantee that objp points to a valid object previously
4516 * allocated with either kmalloc() or kmem_cache_alloc(). The object 4516 * allocated with either kmalloc() or kmem_cache_alloc(). The object
4517 * must not be freed during the duration of the call. 4517 * must not be freed during the duration of the call.
4518 */ 4518 */
4519 size_t ksize(const void *objp) 4519 size_t ksize(const void *objp)
4520 { 4520 {
4521 BUG_ON(!objp); 4521 BUG_ON(!objp);
4522 if (unlikely(objp == ZERO_SIZE_PTR)) 4522 if (unlikely(objp == ZERO_SIZE_PTR))
4523 return 0; 4523 return 0;
4524 4524
4525 return obj_size(virt_to_cache(objp)); 4525 return obj_size(virt_to_cache(objp));
4526 } 4526 }
4527 EXPORT_SYMBOL(ksize); 4527 EXPORT_SYMBOL(ksize);
4528 4528