Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/slab.c |
| 3 | * Written by Mark Hemment, 1996/97. |
| 4 | * (markhe@nextd.demon.co.uk) |
| 5 | * |
| 6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli |
| 7 | * |
| 8 | * Major cleanup, different bufctl logic, per-cpu arrays |
| 9 | * (c) 2000 Manfred Spraul |
| 10 | * |
| 11 | * Cleanup, make the head arrays unconditional, preparation for NUMA |
| 12 | * (c) 2002 Manfred Spraul |
| 13 | * |
| 14 | * An implementation of the Slab Allocator as described in outline in; |
| 15 | * UNIX Internals: The New Frontiers by Uresh Vahalia |
| 16 | * Pub: Prentice Hall ISBN 0-13-101908-2 |
| 17 | * or with a little more detail in; |
| 18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator |
| 19 | * Jeff Bonwick (Sun Microsystems). |
| 20 | * Presented at: USENIX Summer 1994 Technical Conference |
| 21 | * |
| 22 | * The memory is organized in caches, one cache for each object type. |
| 23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) |
| 24 | * Each cache consists out of many slabs (they are small (usually one |
| 25 | * page long) and always contiguous), and each slab contains multiple |
| 26 | * initialized objects. |
| 27 | * |
| 28 | * This means, that your constructor is used only for newly allocated |
| 29 | * slabs and you must pass objects with the same intializations to |
| 30 | * kmem_cache_free. |
| 31 | * |
| 32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, |
| 33 | * normal). If you need a special memory type, then must create a new |
| 34 | * cache for that memory type. |
| 35 | * |
| 36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: |
| 37 | * full slabs with 0 free objects |
| 38 | * partial slabs |
| 39 | * empty slabs with no allocated objects |
| 40 | * |
| 41 | * If partial slabs exist, then new allocations come from these slabs, |
| 42 | * otherwise from empty slabs or new slabs are allocated. |
| 43 | * |
| 44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache |
| 45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. |
| 46 | * |
| 47 | * Each cache has a short per-cpu head array, most allocs |
| 48 | * and frees go into that array, and if that array overflows, then 1/2 |
| 49 | * of the entries in the array are given back into the global cache. |
| 50 | * The head array is strictly LIFO and should improve the cache hit rates. |
| 51 | * On SMP, it additionally reduces the spinlock operations. |
| 52 | * |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | * it's changed with a smp_call_function(). |
| 55 | * |
| 56 | * SMP synchronization: |
| 57 | * constructors and destructors are called without any locking. |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 58 | * Several members in struct kmem_cache and struct slab never change, they |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | * are accessed without any locking. |
| 60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, |
| 61 | * and local interrupts are disabled so slab code is preempt-safe. |
| 62 | * The non-constant members are protected with a per-cache irq spinlock. |
| 63 | * |
| 64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch |
| 65 | * in 2000 - many ideas in the current implementation are derived from |
| 66 | * his patch. |
| 67 | * |
| 68 | * Further notes from the original documentation: |
| 69 | * |
| 70 | * 11 April '97. Started multi-threading - markhe |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 71 | * The global cache-chain is protected by the mutex 'cache_chain_mutex'. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | * The sem is only needed when accessing/extending the cache-chain, which |
| 73 | * can never happen inside an interrupt (kmem_cache_create(), |
| 74 | * kmem_cache_shrink() and kmem_cache_reap()). |
| 75 | * |
| 76 | * At present, each engine can be growing a cache. This should be blocked. |
| 77 | * |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 78 | * 15 March 2005. NUMA slab allocator. |
| 79 | * Shai Fultheim <shai@scalex86.org>. |
| 80 | * Shobhit Dayal <shobhit@calsoftinc.com> |
| 81 | * Alok N Kataria <alokk@calsoftinc.com> |
| 82 | * Christoph Lameter <christoph@lameter.com> |
| 83 | * |
| 84 | * Modified the slab allocator to be node aware on NUMA systems. |
| 85 | * Each node has its own list of partial, free and full slabs. |
| 86 | * All object allocations for a node occur from node specific slab lists. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 87 | */ |
| 88 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | #include <linux/slab.h> |
| 90 | #include <linux/mm.h> |
Randy Dunlap | c9cf552 | 2006-06-27 02:53:52 -0700 | [diff] [blame] | 91 | #include <linux/poison.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | #include <linux/swap.h> |
| 93 | #include <linux/cache.h> |
| 94 | #include <linux/interrupt.h> |
| 95 | #include <linux/init.h> |
| 96 | #include <linux/compiler.h> |
Paul Jackson | 101a500 | 2006-03-24 03:16:07 -0800 | [diff] [blame] | 97 | #include <linux/cpuset.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | #include <linux/seq_file.h> |
| 99 | #include <linux/notifier.h> |
| 100 | #include <linux/kallsyms.h> |
| 101 | #include <linux/cpu.h> |
| 102 | #include <linux/sysctl.h> |
| 103 | #include <linux/module.h> |
| 104 | #include <linux/rcupdate.h> |
Paulo Marques | 543537b | 2005-06-23 00:09:02 -0700 | [diff] [blame] | 105 | #include <linux/string.h> |
Andrew Morton | 138ae66 | 2006-12-06 20:36:41 -0800 | [diff] [blame] | 106 | #include <linux/uaccess.h> |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 107 | #include <linux/nodemask.h> |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 108 | #include <linux/mempolicy.h> |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 109 | #include <linux/mutex.h> |
Akinobu Mita | 8a8b650 | 2006-12-08 02:39:44 -0800 | [diff] [blame] | 110 | #include <linux/fault-inject.h> |
Ingo Molnar | e7eebaf | 2006-06-27 02:54:55 -0700 | [diff] [blame] | 111 | #include <linux/rtmutex.h> |
Eric Dumazet | 6a2d7a9 | 2006-12-13 00:34:27 -0800 | [diff] [blame] | 112 | #include <linux/reciprocal_div.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | #include <asm/cacheflush.h> |
| 115 | #include <asm/tlbflush.h> |
| 116 | #include <asm/page.h> |
| 117 | |
| 118 | /* |
Christoph Lameter | 50953fe | 2007-05-06 14:50:16 -0700 | [diff] [blame] | 119 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | * 0 for faster, smaller code (especially in the critical paths). |
| 121 | * |
| 122 | * STATS - 1 to collect stats for /proc/slabinfo. |
| 123 | * 0 for faster, smaller code (especially in the critical paths). |
| 124 | * |
| 125 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) |
| 126 | */ |
| 127 | |
| 128 | #ifdef CONFIG_DEBUG_SLAB |
| 129 | #define DEBUG 1 |
| 130 | #define STATS 1 |
| 131 | #define FORCED_DEBUG 1 |
| 132 | #else |
| 133 | #define DEBUG 0 |
| 134 | #define STATS 0 |
| 135 | #define FORCED_DEBUG 0 |
| 136 | #endif |
| 137 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | /* Shouldn't this be in a header file somewhere? */ |
| 139 | #define BYTES_PER_WORD sizeof(void *) |
David Woodhouse | 87a927c | 2007-07-04 21:26:44 -0400 | [diff] [blame] | 140 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | |
| 142 | #ifndef cache_line_size |
| 143 | #define cache_line_size() L1_CACHE_BYTES |
| 144 | #endif |
| 145 | |
| 146 | #ifndef ARCH_KMALLOC_MINALIGN |
| 147 | /* |
| 148 | * Enforce a minimum alignment for the kmalloc caches. |
| 149 | * Usually, the kmalloc caches are cache_line_size() aligned, except when |
| 150 | * DEBUG and FORCED_DEBUG are enabled, then they are BYTES_PER_WORD aligned. |
| 151 | * Some archs want to perform DMA into kmalloc caches and need a guaranteed |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 152 | * alignment larger than the alignment of a 64-bit integer. |
| 153 | * ARCH_KMALLOC_MINALIGN allows that. |
| 154 | * Note that increasing this value may disable some debug features. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | */ |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 156 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | #endif |
| 158 | |
| 159 | #ifndef ARCH_SLAB_MINALIGN |
| 160 | /* |
| 161 | * Enforce a minimum alignment for all caches. |
| 162 | * Intended for archs that get misalignment faults even for BYTES_PER_WORD |
| 163 | * aligned buffers. Includes ARCH_KMALLOC_MINALIGN. |
| 164 | * If possible: Do not enable this flag for CONFIG_DEBUG_SLAB, it disables |
| 165 | * some debug features. |
| 166 | */ |
| 167 | #define ARCH_SLAB_MINALIGN 0 |
| 168 | #endif |
| 169 | |
| 170 | #ifndef ARCH_KMALLOC_FLAGS |
| 171 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN |
| 172 | #endif |
| 173 | |
| 174 | /* Legal flag mask for kmem_cache_create(). */ |
| 175 | #if DEBUG |
Christoph Lameter | 50953fe | 2007-05-06 14:50:16 -0700 | [diff] [blame] | 176 | # define CREATE_MASK (SLAB_RED_ZONE | \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | SLAB_POISON | SLAB_HWCACHE_ALIGN | \ |
Christoph Lameter | ac2b898 | 2006-03-22 00:08:15 -0800 | [diff] [blame] | 178 | SLAB_CACHE_DMA | \ |
Christoph Lameter | 5af6083 | 2007-05-06 14:49:56 -0700 | [diff] [blame] | 179 | SLAB_STORE_USER | \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
Paul Jackson | 101a500 | 2006-03-24 03:16:07 -0800 | [diff] [blame] | 181 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | #else |
Christoph Lameter | ac2b898 | 2006-03-22 00:08:15 -0800 | [diff] [blame] | 183 | # define CREATE_MASK (SLAB_HWCACHE_ALIGN | \ |
Christoph Lameter | 5af6083 | 2007-05-06 14:49:56 -0700 | [diff] [blame] | 184 | SLAB_CACHE_DMA | \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
Paul Jackson | 101a500 | 2006-03-24 03:16:07 -0800 | [diff] [blame] | 186 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | #endif |
| 188 | |
| 189 | /* |
| 190 | * kmem_bufctl_t: |
| 191 | * |
| 192 | * Bufctl's are used for linking objs within a slab |
| 193 | * linked offsets. |
| 194 | * |
| 195 | * This implementation relies on "struct page" for locating the cache & |
| 196 | * slab an object belongs to. |
| 197 | * This allows the bufctl structure to be small (one int), but limits |
| 198 | * the number of objects a slab (not a cache) can contain when off-slab |
| 199 | * bufctls are used. The limit is the size of the largest general cache |
| 200 | * that does not use off-slab slabs. |
| 201 | * For 32bit archs with 4 kB pages, is this 56. |
| 202 | * This is not serious, as it is only for large objects, when it is unwise |
| 203 | * to have too many per slab. |
| 204 | * Note: This limit can be raised by introducing a general cache whose size |
| 205 | * is less than 512 (PAGE_SIZE<<3), but greater than 256. |
| 206 | */ |
| 207 | |
Kyle Moffett | fa5b08d | 2005-09-03 15:55:03 -0700 | [diff] [blame] | 208 | typedef unsigned int kmem_bufctl_t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0) |
| 210 | #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1) |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 211 | #define BUFCTL_ACTIVE (((kmem_bufctl_t)(~0U))-2) |
| 212 | #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-3) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 214 | /* |
| 215 | * struct slab |
| 216 | * |
| 217 | * Manages the objs in a slab. Placed either at the beginning of mem allocated |
| 218 | * for a slab, or allocated from an general cache. |
| 219 | * Slabs are chained into three list: fully used, partial, fully free slabs. |
| 220 | */ |
| 221 | struct slab { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 222 | struct list_head list; |
| 223 | unsigned long colouroff; |
| 224 | void *s_mem; /* including colour offset */ |
| 225 | unsigned int inuse; /* num of objs active in slab */ |
| 226 | kmem_bufctl_t free; |
| 227 | unsigned short nodeid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | }; |
| 229 | |
| 230 | /* |
| 231 | * struct slab_rcu |
| 232 | * |
| 233 | * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to |
| 234 | * arrange for kmem_freepages to be called via RCU. This is useful if |
| 235 | * we need to approach a kernel structure obliquely, from its address |
| 236 | * obtained without the usual locking. We can lock the structure to |
| 237 | * stabilize it and check it's still at the given address, only if we |
| 238 | * can be sure that the memory has not been meanwhile reused for some |
| 239 | * other kind of object (which our subsystem's lock might corrupt). |
| 240 | * |
| 241 | * rcu_read_lock before reading the address, then rcu_read_unlock after |
| 242 | * taking the spinlock within the structure expected at that address. |
| 243 | * |
| 244 | * We assume struct slab_rcu can overlay struct slab when destroying. |
| 245 | */ |
| 246 | struct slab_rcu { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 247 | struct rcu_head head; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 248 | struct kmem_cache *cachep; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 249 | void *addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 250 | }; |
| 251 | |
| 252 | /* |
| 253 | * struct array_cache |
| 254 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | * Purpose: |
| 256 | * - LIFO ordering, to hand out cache-warm objects from _alloc |
| 257 | * - reduce the number of linked list operations |
| 258 | * - reduce spinlock operations |
| 259 | * |
| 260 | * The limit is stored in the per-cpu structure to reduce the data cache |
| 261 | * footprint. |
| 262 | * |
| 263 | */ |
| 264 | struct array_cache { |
| 265 | unsigned int avail; |
| 266 | unsigned int limit; |
| 267 | unsigned int batchcount; |
| 268 | unsigned int touched; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 269 | spinlock_t lock; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 270 | void *entry[0]; /* |
| 271 | * Must have this definition in here for the proper |
| 272 | * alignment of array_cache. Also simplifies accessing |
| 273 | * the entries. |
| 274 | * [0] is for gcc 2.95. It should really be []. |
| 275 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | }; |
| 277 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 278 | /* |
| 279 | * bootstrap: The caches do not work without cpuarrays anymore, but the |
| 280 | * cpuarrays are allocated from the generic caches... |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | */ |
| 282 | #define BOOT_CPUCACHE_ENTRIES 1 |
| 283 | struct arraycache_init { |
| 284 | struct array_cache cache; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 285 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | }; |
| 287 | |
| 288 | /* |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 289 | * The slab lists for all objects. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | */ |
| 291 | struct kmem_list3 { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 292 | struct list_head slabs_partial; /* partial list first, better asm code */ |
| 293 | struct list_head slabs_full; |
| 294 | struct list_head slabs_free; |
| 295 | unsigned long free_objects; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 296 | unsigned int free_limit; |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 297 | unsigned int colour_next; /* Per-node cache coloring */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 298 | spinlock_t list_lock; |
| 299 | struct array_cache *shared; /* shared per node */ |
| 300 | struct array_cache **alien; /* on other nodes */ |
Christoph Lameter | 35386e3 | 2006-03-22 00:09:05 -0800 | [diff] [blame] | 301 | unsigned long next_reap; /* updated without locking */ |
| 302 | int free_touched; /* updated without locking */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | }; |
| 304 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 305 | /* |
| 306 | * Need this for bootstrapping a per node allocator. |
| 307 | */ |
| 308 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES + 1) |
| 309 | struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS]; |
| 310 | #define CACHE_CACHE 0 |
| 311 | #define SIZE_AC 1 |
| 312 | #define SIZE_L3 (1 + MAX_NUMNODES) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 313 | |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 314 | static int drain_freelist(struct kmem_cache *cache, |
| 315 | struct kmem_list3 *l3, int tofree); |
| 316 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
| 317 | int node); |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 318 | static int enable_cpucache(struct kmem_cache *cachep); |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 319 | static void cache_reap(struct work_struct *unused); |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 320 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 321 | /* |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 322 | * This function must be completely optimized away if a constant is passed to |
| 323 | * it. Mostly the same as what is in linux/slab.h except it returns an index. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 324 | */ |
Ivan Kokshaysky | 7243cc0 | 2005-09-22 21:43:58 -0700 | [diff] [blame] | 325 | static __always_inline int index_of(const size_t size) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 326 | { |
Steven Rostedt | 5ec8a84 | 2006-02-01 03:05:44 -0800 | [diff] [blame] | 327 | extern void __bad_size(void); |
| 328 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 329 | if (__builtin_constant_p(size)) { |
| 330 | int i = 0; |
| 331 | |
| 332 | #define CACHE(x) \ |
| 333 | if (size <=x) \ |
| 334 | return i; \ |
| 335 | else \ |
| 336 | i++; |
| 337 | #include "linux/kmalloc_sizes.h" |
| 338 | #undef CACHE |
Steven Rostedt | 5ec8a84 | 2006-02-01 03:05:44 -0800 | [diff] [blame] | 339 | __bad_size(); |
Ivan Kokshaysky | 7243cc0 | 2005-09-22 21:43:58 -0700 | [diff] [blame] | 340 | } else |
Steven Rostedt | 5ec8a84 | 2006-02-01 03:05:44 -0800 | [diff] [blame] | 341 | __bad_size(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 342 | return 0; |
| 343 | } |
| 344 | |
Ingo Molnar | e0a4272 | 2006-06-23 02:03:46 -0700 | [diff] [blame] | 345 | static int slab_early_init = 1; |
| 346 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 347 | #define INDEX_AC index_of(sizeof(struct arraycache_init)) |
| 348 | #define INDEX_L3 index_of(sizeof(struct kmem_list3)) |
| 349 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 350 | static void kmem_list3_init(struct kmem_list3 *parent) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 351 | { |
| 352 | INIT_LIST_HEAD(&parent->slabs_full); |
| 353 | INIT_LIST_HEAD(&parent->slabs_partial); |
| 354 | INIT_LIST_HEAD(&parent->slabs_free); |
| 355 | parent->shared = NULL; |
| 356 | parent->alien = NULL; |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 357 | parent->colour_next = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 358 | spin_lock_init(&parent->list_lock); |
| 359 | parent->free_objects = 0; |
| 360 | parent->free_touched = 0; |
| 361 | } |
| 362 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 363 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
| 364 | do { \ |
| 365 | INIT_LIST_HEAD(listp); \ |
| 366 | list_splice(&(cachep->nodelists[nodeid]->slab), listp); \ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 367 | } while (0) |
| 368 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 369 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
| 370 | do { \ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 371 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
| 372 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ |
| 373 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ |
| 374 | } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | |
| 376 | /* |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 377 | * struct kmem_cache |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | * |
| 379 | * manages a cache. |
| 380 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 381 | |
Pekka J Enberg | 2109a2d | 2005-11-07 00:58:01 -0800 | [diff] [blame] | 382 | struct kmem_cache { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | /* 1) per-cpu data, touched during every alloc/free */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 384 | struct array_cache *array[NR_CPUS]; |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 385 | /* 2) Cache tunables. Protected by cache_chain_mutex */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 386 | unsigned int batchcount; |
| 387 | unsigned int limit; |
| 388 | unsigned int shared; |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 389 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 390 | unsigned int buffer_size; |
Eric Dumazet | 6a2d7a9 | 2006-12-13 00:34:27 -0800 | [diff] [blame] | 391 | u32 reciprocal_buffer_size; |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 392 | /* 3) touched by every alloc & free from the backend */ |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 393 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 394 | unsigned int flags; /* constant flags */ |
| 395 | unsigned int num; /* # of objs per slab */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 397 | /* 4) cache_grow/shrink */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | /* order of pgs per slab (2^n) */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 399 | unsigned int gfporder; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | |
| 401 | /* force GFP flags, e.g. GFP_DMA */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 402 | gfp_t gfpflags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 404 | size_t colour; /* cache colouring range */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 405 | unsigned int colour_off; /* colour offset */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 406 | struct kmem_cache *slabp_cache; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 407 | unsigned int slab_size; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 408 | unsigned int dflags; /* dynamic flags */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | |
| 410 | /* constructor func */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 411 | void (*ctor) (void *, struct kmem_cache *, unsigned long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 413 | /* 5) cache creation/removal */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 414 | const char *name; |
| 415 | struct list_head next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 417 | /* 6) statistics */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | #if STATS |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 419 | unsigned long num_active; |
| 420 | unsigned long num_allocations; |
| 421 | unsigned long high_mark; |
| 422 | unsigned long grown; |
| 423 | unsigned long reaped; |
| 424 | unsigned long errors; |
| 425 | unsigned long max_freeable; |
| 426 | unsigned long node_allocs; |
| 427 | unsigned long node_frees; |
Ravikiran G Thirumalai | fb7faf3 | 2006-04-10 22:52:54 -0700 | [diff] [blame] | 428 | unsigned long node_overflow; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 429 | atomic_t allochit; |
| 430 | atomic_t allocmiss; |
| 431 | atomic_t freehit; |
| 432 | atomic_t freemiss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | #endif |
| 434 | #if DEBUG |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 435 | /* |
| 436 | * If debugging is enabled, then the allocator can add additional |
| 437 | * fields and/or padding to every object. buffer_size contains the total |
| 438 | * object size including these internal fields, the following two |
| 439 | * variables contain the offset to the user object and its size. |
| 440 | */ |
| 441 | int obj_offset; |
| 442 | int obj_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 443 | #endif |
Eric Dumazet | 8da3430 | 2007-05-06 14:49:29 -0700 | [diff] [blame] | 444 | /* |
| 445 | * We put nodelists[] at the end of kmem_cache, because we want to size |
| 446 | * this array to nr_node_ids slots instead of MAX_NUMNODES |
| 447 | * (see kmem_cache_init()) |
| 448 | * We still use [MAX_NUMNODES] and not [1] or [0] because cache_cache |
| 449 | * is statically defined, so we reserve the max number of nodes. |
| 450 | */ |
| 451 | struct kmem_list3 *nodelists[MAX_NUMNODES]; |
| 452 | /* |
| 453 | * Do not add fields after nodelists[] |
| 454 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 455 | }; |
| 456 | |
| 457 | #define CFLGS_OFF_SLAB (0x80000000UL) |
| 458 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
| 459 | |
| 460 | #define BATCHREFILL_LIMIT 16 |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 461 | /* |
| 462 | * Optimization question: fewer reaps means less probability for unnessary |
| 463 | * cpucache drain/refill cycles. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | * |
Adrian Bunk | dc6f3f2 | 2005-11-08 16:44:08 +0100 | [diff] [blame] | 465 | * OTOH the cpuarrays can contain lots of objects, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | * which could lock up otherwise freeable slabs. |
| 467 | */ |
| 468 | #define REAPTIMEOUT_CPUC (2*HZ) |
| 469 | #define REAPTIMEOUT_LIST3 (4*HZ) |
| 470 | |
| 471 | #if STATS |
| 472 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) |
| 473 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) |
| 474 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) |
| 475 | #define STATS_INC_GROWN(x) ((x)->grown++) |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 476 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 477 | #define STATS_SET_HIGH(x) \ |
| 478 | do { \ |
| 479 | if ((x)->num_active > (x)->high_mark) \ |
| 480 | (x)->high_mark = (x)->num_active; \ |
| 481 | } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | #define STATS_INC_ERR(x) ((x)->errors++) |
| 483 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 484 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
Ravikiran G Thirumalai | fb7faf3 | 2006-04-10 22:52:54 -0700 | [diff] [blame] | 485 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 486 | #define STATS_SET_FREEABLE(x, i) \ |
| 487 | do { \ |
| 488 | if ((x)->max_freeable < i) \ |
| 489 | (x)->max_freeable = i; \ |
| 490 | } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
| 492 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) |
| 493 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) |
| 494 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) |
| 495 | #else |
| 496 | #define STATS_INC_ACTIVE(x) do { } while (0) |
| 497 | #define STATS_DEC_ACTIVE(x) do { } while (0) |
| 498 | #define STATS_INC_ALLOCED(x) do { } while (0) |
| 499 | #define STATS_INC_GROWN(x) do { } while (0) |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 500 | #define STATS_ADD_REAPED(x,y) do { } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | #define STATS_SET_HIGH(x) do { } while (0) |
| 502 | #define STATS_INC_ERR(x) do { } while (0) |
| 503 | #define STATS_INC_NODEALLOCS(x) do { } while (0) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 504 | #define STATS_INC_NODEFREES(x) do { } while (0) |
Ravikiran G Thirumalai | fb7faf3 | 2006-04-10 22:52:54 -0700 | [diff] [blame] | 505 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 506 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
| 508 | #define STATS_INC_ALLOCMISS(x) do { } while (0) |
| 509 | #define STATS_INC_FREEHIT(x) do { } while (0) |
| 510 | #define STATS_INC_FREEMISS(x) do { } while (0) |
| 511 | #endif |
| 512 | |
| 513 | #if DEBUG |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 515 | /* |
| 516 | * memory layout of objects: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | * 0 : objp |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 518 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | * the end of an object is aligned with the end of the real |
| 520 | * allocation. Catches writes behind the end of the allocation. |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 521 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | * redzone word. |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 523 | * cachep->obj_offset: The real object. |
| 524 | * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 525 | * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address |
| 526 | * [BYTES_PER_WORD long] |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 528 | static int obj_offset(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 530 | return cachep->obj_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | } |
| 532 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 533 | static int obj_size(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 535 | return cachep->obj_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | } |
| 537 | |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 538 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | { |
| 540 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 541 | return (unsigned long long*) (objp + obj_offset(cachep) - |
| 542 | sizeof(unsigned long long)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | } |
| 544 | |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 545 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | { |
| 547 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
| 548 | if (cachep->flags & SLAB_STORE_USER) |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 549 | return (unsigned long long *)(objp + cachep->buffer_size - |
| 550 | sizeof(unsigned long long) - |
David Woodhouse | 87a927c | 2007-07-04 21:26:44 -0400 | [diff] [blame] | 551 | REDZONE_ALIGN); |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 552 | return (unsigned long long *) (objp + cachep->buffer_size - |
| 553 | sizeof(unsigned long long)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | } |
| 555 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 556 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | { |
| 558 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 559 | return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | } |
| 561 | |
| 562 | #else |
| 563 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 564 | #define obj_offset(x) 0 |
| 565 | #define obj_size(cachep) (cachep->buffer_size) |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 566 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
| 567 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
| 569 | |
| 570 | #endif |
| 571 | |
| 572 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | * Do not go above this order unless 0 objects fit into the slab. |
| 574 | */ |
| 575 | #define BREAK_GFP_ORDER_HI 1 |
| 576 | #define BREAK_GFP_ORDER_LO 0 |
| 577 | static int slab_break_gfp_order = BREAK_GFP_ORDER_LO; |
| 578 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 579 | /* |
| 580 | * Functions for storing/retrieving the cachep and or slab from the page |
| 581 | * allocator. These are used to find the slab an obj belongs to. With kfree(), |
| 582 | * these are used to find the cache which an obj belongs to. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 583 | */ |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 584 | static inline void page_set_cache(struct page *page, struct kmem_cache *cache) |
| 585 | { |
| 586 | page->lru.next = (struct list_head *)cache; |
| 587 | } |
| 588 | |
| 589 | static inline struct kmem_cache *page_get_cache(struct page *page) |
| 590 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 591 | page = compound_head(page); |
Pekka Enberg | ddc2e81 | 2006-06-23 02:03:40 -0700 | [diff] [blame] | 592 | BUG_ON(!PageSlab(page)); |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 593 | return (struct kmem_cache *)page->lru.next; |
| 594 | } |
| 595 | |
| 596 | static inline void page_set_slab(struct page *page, struct slab *slab) |
| 597 | { |
| 598 | page->lru.prev = (struct list_head *)slab; |
| 599 | } |
| 600 | |
| 601 | static inline struct slab *page_get_slab(struct page *page) |
| 602 | { |
Pekka Enberg | ddc2e81 | 2006-06-23 02:03:40 -0700 | [diff] [blame] | 603 | BUG_ON(!PageSlab(page)); |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 604 | return (struct slab *)page->lru.prev; |
| 605 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 606 | |
Pekka Enberg | 6ed5eb2 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 607 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
| 608 | { |
Christoph Lameter | b49af68 | 2007-05-06 14:49:41 -0700 | [diff] [blame] | 609 | struct page *page = virt_to_head_page(obj); |
Pekka Enberg | 6ed5eb2 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 610 | return page_get_cache(page); |
| 611 | } |
| 612 | |
| 613 | static inline struct slab *virt_to_slab(const void *obj) |
| 614 | { |
Christoph Lameter | b49af68 | 2007-05-06 14:49:41 -0700 | [diff] [blame] | 615 | struct page *page = virt_to_head_page(obj); |
Pekka Enberg | 6ed5eb2 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 616 | return page_get_slab(page); |
| 617 | } |
| 618 | |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 619 | static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab, |
| 620 | unsigned int idx) |
| 621 | { |
| 622 | return slab->s_mem + cache->buffer_size * idx; |
| 623 | } |
| 624 | |
Eric Dumazet | 6a2d7a9 | 2006-12-13 00:34:27 -0800 | [diff] [blame] | 625 | /* |
| 626 | * We want to avoid an expensive divide : (offset / cache->buffer_size) |
| 627 | * Using the fact that buffer_size is a constant for a particular cache, |
| 628 | * we can replace (offset / cache->buffer_size) by |
| 629 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
| 630 | */ |
| 631 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, |
| 632 | const struct slab *slab, void *obj) |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 633 | { |
Eric Dumazet | 6a2d7a9 | 2006-12-13 00:34:27 -0800 | [diff] [blame] | 634 | u32 offset = (obj - slab->s_mem); |
| 635 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 636 | } |
| 637 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 638 | /* |
| 639 | * These are the default caches for kmalloc. Custom caches can have other sizes. |
| 640 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | struct cache_sizes malloc_sizes[] = { |
| 642 | #define CACHE(x) { .cs_size = (x) }, |
| 643 | #include <linux/kmalloc_sizes.h> |
| 644 | CACHE(ULONG_MAX) |
| 645 | #undef CACHE |
| 646 | }; |
| 647 | EXPORT_SYMBOL(malloc_sizes); |
| 648 | |
| 649 | /* Must match cache_sizes above. Out of line to keep cache footprint low. */ |
| 650 | struct cache_names { |
| 651 | char *name; |
| 652 | char *name_dma; |
| 653 | }; |
| 654 | |
| 655 | static struct cache_names __initdata cache_names[] = { |
| 656 | #define CACHE(x) { .name = "size-" #x, .name_dma = "size-" #x "(DMA)" }, |
| 657 | #include <linux/kmalloc_sizes.h> |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 658 | {NULL,} |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 659 | #undef CACHE |
| 660 | }; |
| 661 | |
| 662 | static struct arraycache_init initarray_cache __initdata = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 663 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | static struct arraycache_init initarray_generic = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 665 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | |
| 667 | /* internal cache of cache description objs */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 668 | static struct kmem_cache cache_cache = { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 669 | .batchcount = 1, |
| 670 | .limit = BOOT_CPUCACHE_ENTRIES, |
| 671 | .shared = 1, |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 672 | .buffer_size = sizeof(struct kmem_cache), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 673 | .name = "kmem_cache", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | }; |
| 675 | |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 676 | #define BAD_ALIEN_MAGIC 0x01020304ul |
| 677 | |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 678 | #ifdef CONFIG_LOCKDEP |
| 679 | |
| 680 | /* |
| 681 | * Slab sometimes uses the kmalloc slabs to store the slab headers |
| 682 | * for other slabs "off slab". |
| 683 | * The locking for this is tricky in that it nests within the locks |
| 684 | * of all other slabs in a few places; to deal with this special |
| 685 | * locking we put on-slab caches into a separate lock-class. |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 686 | * |
| 687 | * We set lock class for alien array caches which are up during init. |
| 688 | * The lock annotation will be lost if all cpus of a node goes down and |
| 689 | * then comes back up during hotplug |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 690 | */ |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 691 | static struct lock_class_key on_slab_l3_key; |
| 692 | static struct lock_class_key on_slab_alc_key; |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 693 | |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 694 | static inline void init_lock_keys(void) |
| 695 | |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 696 | { |
| 697 | int q; |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 698 | struct cache_sizes *s = malloc_sizes; |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 699 | |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 700 | while (s->cs_size != ULONG_MAX) { |
| 701 | for_each_node(q) { |
| 702 | struct array_cache **alc; |
| 703 | int r; |
| 704 | struct kmem_list3 *l3 = s->cs_cachep->nodelists[q]; |
| 705 | if (!l3 || OFF_SLAB(s->cs_cachep)) |
| 706 | continue; |
| 707 | lockdep_set_class(&l3->list_lock, &on_slab_l3_key); |
| 708 | alc = l3->alien; |
| 709 | /* |
| 710 | * FIXME: This check for BAD_ALIEN_MAGIC |
| 711 | * should go away when common slab code is taught to |
| 712 | * work even without alien caches. |
| 713 | * Currently, non NUMA code returns BAD_ALIEN_MAGIC |
| 714 | * for alloc_alien_cache, |
| 715 | */ |
| 716 | if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) |
| 717 | continue; |
| 718 | for_each_node(r) { |
| 719 | if (alc[r]) |
| 720 | lockdep_set_class(&alc[r]->lock, |
| 721 | &on_slab_alc_key); |
| 722 | } |
| 723 | } |
| 724 | s++; |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 725 | } |
| 726 | } |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 727 | #else |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 728 | static inline void init_lock_keys(void) |
Arjan van de Ven | f1aaee5 | 2006-07-13 14:46:03 +0200 | [diff] [blame] | 729 | { |
| 730 | } |
| 731 | #endif |
| 732 | |
Ravikiran G Thirumalai | 8f5be20 | 2006-12-06 20:32:14 -0800 | [diff] [blame] | 733 | /* |
| 734 | * 1. Guard access to the cache-chain. |
| 735 | * 2. Protect sanity of cpu_online_map against cpu hotplug events |
| 736 | */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 737 | static DEFINE_MUTEX(cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 738 | static struct list_head cache_chain; |
| 739 | |
| 740 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | * chicken and egg problem: delay the per-cpu array allocation |
| 742 | * until the general caches are up. |
| 743 | */ |
| 744 | static enum { |
| 745 | NONE, |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 746 | PARTIAL_AC, |
| 747 | PARTIAL_L3, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 748 | FULL |
| 749 | } g_cpucache_up; |
| 750 | |
Mike Kravetz | 39d24e6 | 2006-05-15 09:44:13 -0700 | [diff] [blame] | 751 | /* |
| 752 | * used by boot code to determine if it can use slab based allocator |
| 753 | */ |
| 754 | int slab_is_available(void) |
| 755 | { |
| 756 | return g_cpucache_up == FULL; |
| 757 | } |
| 758 | |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 759 | static DEFINE_PER_CPU(struct delayed_work, reap_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 761 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | { |
| 763 | return cachep->array[smp_processor_id()]; |
| 764 | } |
| 765 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 766 | static inline struct kmem_cache *__find_general_cachep(size_t size, |
| 767 | gfp_t gfpflags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | { |
| 769 | struct cache_sizes *csizep = malloc_sizes; |
| 770 | |
| 771 | #if DEBUG |
| 772 | /* This happens if someone tries to call |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 773 | * kmem_cache_create(), or __kmalloc(), before |
| 774 | * the generic caches are initialized. |
| 775 | */ |
Alok Kataria | c7e43c7 | 2005-09-14 12:17:53 -0700 | [diff] [blame] | 776 | BUG_ON(malloc_sizes[INDEX_AC].cs_cachep == NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 777 | #endif |
| 778 | while (size > csizep->cs_size) |
| 779 | csizep++; |
| 780 | |
| 781 | /* |
Martin Hicks | 0abf40c | 2005-09-03 15:54:54 -0700 | [diff] [blame] | 782 | * Really subtle: The last entry with cs->cs_size==ULONG_MAX |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | * has cs_{dma,}cachep==NULL. Thus no special case |
| 784 | * for large kmalloc calls required. |
| 785 | */ |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 786 | #ifdef CONFIG_ZONE_DMA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | if (unlikely(gfpflags & GFP_DMA)) |
| 788 | return csizep->cs_dmacachep; |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 789 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 790 | return csizep->cs_cachep; |
| 791 | } |
| 792 | |
Adrian Bunk | b221385 | 2006-09-25 23:31:02 -0700 | [diff] [blame] | 793 | static struct kmem_cache *kmem_find_general_cachep(size_t size, gfp_t gfpflags) |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 794 | { |
| 795 | return __find_general_cachep(size, gfpflags); |
| 796 | } |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 797 | |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 798 | static size_t slab_mgmt_size(size_t nr_objs, size_t align) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | { |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 800 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); |
| 801 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 803 | /* |
| 804 | * Calculate the number of objects and left-over bytes for a given buffer size. |
| 805 | */ |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 806 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
| 807 | size_t align, int flags, size_t *left_over, |
| 808 | unsigned int *num) |
| 809 | { |
| 810 | int nr_objs; |
| 811 | size_t mgmt_size; |
| 812 | size_t slab_size = PAGE_SIZE << gfporder; |
| 813 | |
| 814 | /* |
| 815 | * The slab management structure can be either off the slab or |
| 816 | * on it. For the latter case, the memory allocated for a |
| 817 | * slab is used for: |
| 818 | * |
| 819 | * - The struct slab |
| 820 | * - One kmem_bufctl_t for each object |
| 821 | * - Padding to respect alignment of @align |
| 822 | * - @buffer_size bytes for each object |
| 823 | * |
| 824 | * If the slab management structure is off the slab, then the |
| 825 | * alignment will already be calculated into the size. Because |
| 826 | * the slabs are all pages aligned, the objects will be at the |
| 827 | * correct alignment when allocated. |
| 828 | */ |
| 829 | if (flags & CFLGS_OFF_SLAB) { |
| 830 | mgmt_size = 0; |
| 831 | nr_objs = slab_size / buffer_size; |
| 832 | |
| 833 | if (nr_objs > SLAB_LIMIT) |
| 834 | nr_objs = SLAB_LIMIT; |
| 835 | } else { |
| 836 | /* |
| 837 | * Ignore padding for the initial guess. The padding |
| 838 | * is at most @align-1 bytes, and @buffer_size is at |
| 839 | * least @align. In the worst case, this result will |
| 840 | * be one greater than the number of objects that fit |
| 841 | * into the memory allocation when taking the padding |
| 842 | * into account. |
| 843 | */ |
| 844 | nr_objs = (slab_size - sizeof(struct slab)) / |
| 845 | (buffer_size + sizeof(kmem_bufctl_t)); |
| 846 | |
| 847 | /* |
| 848 | * This calculated number will be either the right |
| 849 | * amount, or one greater than what we want. |
| 850 | */ |
| 851 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size |
| 852 | > slab_size) |
| 853 | nr_objs--; |
| 854 | |
| 855 | if (nr_objs > SLAB_LIMIT) |
| 856 | nr_objs = SLAB_LIMIT; |
| 857 | |
| 858 | mgmt_size = slab_mgmt_size(nr_objs, align); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 859 | } |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 860 | *num = nr_objs; |
| 861 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 862 | } |
| 863 | |
| 864 | #define slab_error(cachep, msg) __slab_error(__FUNCTION__, cachep, msg) |
| 865 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 866 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
| 867 | char *msg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 868 | { |
| 869 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 870 | function, cachep->name, msg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | dump_stack(); |
| 872 | } |
| 873 | |
Paul Menage | 3395ee0 | 2006-12-06 20:32:16 -0800 | [diff] [blame] | 874 | /* |
| 875 | * By default on NUMA we use alien caches to stage the freeing of |
| 876 | * objects allocated from other nodes. This causes massive memory |
| 877 | * inefficiencies when using fake NUMA setup to split memory into a |
| 878 | * large number of small nodes, so it can be disabled on the command |
| 879 | * line |
| 880 | */ |
| 881 | |
| 882 | static int use_alien_caches __read_mostly = 1; |
| 883 | static int __init noaliencache_setup(char *s) |
| 884 | { |
| 885 | use_alien_caches = 0; |
| 886 | return 1; |
| 887 | } |
| 888 | __setup("noaliencache", noaliencache_setup); |
| 889 | |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 890 | #ifdef CONFIG_NUMA |
| 891 | /* |
| 892 | * Special reaping functions for NUMA systems called from cache_reap(). |
| 893 | * These take care of doing round robin flushing of alien caches (containing |
| 894 | * objects freed on different nodes from which they were allocated) and the |
| 895 | * flushing of remote pcps by calling drain_node_pages. |
| 896 | */ |
| 897 | static DEFINE_PER_CPU(unsigned long, reap_node); |
| 898 | |
| 899 | static void init_reap_node(int cpu) |
| 900 | { |
| 901 | int node; |
| 902 | |
| 903 | node = next_node(cpu_to_node(cpu), node_online_map); |
| 904 | if (node == MAX_NUMNODES) |
Paul Jackson | 442295c | 2006-03-22 00:09:11 -0800 | [diff] [blame] | 905 | node = first_node(node_online_map); |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 906 | |
Daniel Yeisley | 7f6b887 | 2006-11-02 22:07:14 -0800 | [diff] [blame] | 907 | per_cpu(reap_node, cpu) = node; |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 908 | } |
| 909 | |
| 910 | static void next_reap_node(void) |
| 911 | { |
| 912 | int node = __get_cpu_var(reap_node); |
| 913 | |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 914 | node = next_node(node, node_online_map); |
| 915 | if (unlikely(node >= MAX_NUMNODES)) |
| 916 | node = first_node(node_online_map); |
| 917 | __get_cpu_var(reap_node) = node; |
| 918 | } |
| 919 | |
| 920 | #else |
| 921 | #define init_reap_node(cpu) do { } while (0) |
| 922 | #define next_reap_node(void) do { } while (0) |
| 923 | #endif |
| 924 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | /* |
| 926 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz |
| 927 | * via the workqueue/eventd. |
| 928 | * Add the CPU number into the expiration time to minimize the possibility of |
| 929 | * the CPUs getting into lockstep and contending for the global cache chain |
| 930 | * lock. |
| 931 | */ |
| 932 | static void __devinit start_cpu_timer(int cpu) |
| 933 | { |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 934 | struct delayed_work *reap_work = &per_cpu(reap_work, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | |
| 936 | /* |
| 937 | * When this gets called from do_initcalls via cpucache_init(), |
| 938 | * init_workqueues() has already run, so keventd will be setup |
| 939 | * at that time. |
| 940 | */ |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 941 | if (keventd_up() && reap_work->work.func == NULL) { |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 942 | init_reap_node(cpu); |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 943 | INIT_DELAYED_WORK(reap_work, cache_reap); |
Arjan van de Ven | 2b28421 | 2006-12-10 02:21:28 -0800 | [diff] [blame] | 944 | schedule_delayed_work_on(cpu, reap_work, |
| 945 | __round_jiffies_relative(HZ, cpu)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | } |
| 947 | } |
| 948 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 949 | static struct array_cache *alloc_arraycache(int node, int entries, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 950 | int batchcount) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 952 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 953 | struct array_cache *nc = NULL; |
| 954 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 955 | nc = kmalloc_node(memsize, GFP_KERNEL, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | if (nc) { |
| 957 | nc->avail = 0; |
| 958 | nc->limit = entries; |
| 959 | nc->batchcount = batchcount; |
| 960 | nc->touched = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 961 | spin_lock_init(&nc->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 962 | } |
| 963 | return nc; |
| 964 | } |
| 965 | |
Christoph Lameter | 3ded175 | 2006-03-25 03:06:44 -0800 | [diff] [blame] | 966 | /* |
| 967 | * Transfer objects in one arraycache to another. |
| 968 | * Locking must be handled by the caller. |
| 969 | * |
| 970 | * Return the number of entries transferred. |
| 971 | */ |
| 972 | static int transfer_objects(struct array_cache *to, |
| 973 | struct array_cache *from, unsigned int max) |
| 974 | { |
| 975 | /* Figure out how many entries to transfer */ |
| 976 | int nr = min(min(from->avail, max), to->limit - to->avail); |
| 977 | |
| 978 | if (!nr) |
| 979 | return 0; |
| 980 | |
| 981 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, |
| 982 | sizeof(void *) *nr); |
| 983 | |
| 984 | from->avail -= nr; |
| 985 | to->avail += nr; |
| 986 | to->touched = 1; |
| 987 | return nr; |
| 988 | } |
| 989 | |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 990 | #ifndef CONFIG_NUMA |
| 991 | |
| 992 | #define drain_alien_cache(cachep, alien) do { } while (0) |
| 993 | #define reap_alien(cachep, l3) do { } while (0) |
| 994 | |
| 995 | static inline struct array_cache **alloc_alien_cache(int node, int limit) |
| 996 | { |
| 997 | return (struct array_cache **)BAD_ALIEN_MAGIC; |
| 998 | } |
| 999 | |
| 1000 | static inline void free_alien_cache(struct array_cache **ac_ptr) |
| 1001 | { |
| 1002 | } |
| 1003 | |
| 1004 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
| 1005 | { |
| 1006 | return 0; |
| 1007 | } |
| 1008 | |
| 1009 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, |
| 1010 | gfp_t flags) |
| 1011 | { |
| 1012 | return NULL; |
| 1013 | } |
| 1014 | |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 1015 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 1016 | gfp_t flags, int nodeid) |
| 1017 | { |
| 1018 | return NULL; |
| 1019 | } |
| 1020 | |
| 1021 | #else /* CONFIG_NUMA */ |
| 1022 | |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 1023 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 1024 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1025 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 1026 | static struct array_cache **alloc_alien_cache(int node, int limit) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1027 | { |
| 1028 | struct array_cache **ac_ptr; |
Christoph Lameter | 8ef8286 | 2007-02-20 13:57:52 -0800 | [diff] [blame] | 1029 | int memsize = sizeof(void *) * nr_node_ids; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1030 | int i; |
| 1031 | |
| 1032 | if (limit > 1) |
| 1033 | limit = 12; |
| 1034 | ac_ptr = kmalloc_node(memsize, GFP_KERNEL, node); |
| 1035 | if (ac_ptr) { |
| 1036 | for_each_node(i) { |
| 1037 | if (i == node || !node_online(i)) { |
| 1038 | ac_ptr[i] = NULL; |
| 1039 | continue; |
| 1040 | } |
| 1041 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d); |
| 1042 | if (!ac_ptr[i]) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1043 | for (i--; i <= 0; i--) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1044 | kfree(ac_ptr[i]); |
| 1045 | kfree(ac_ptr); |
| 1046 | return NULL; |
| 1047 | } |
| 1048 | } |
| 1049 | } |
| 1050 | return ac_ptr; |
| 1051 | } |
| 1052 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 1053 | static void free_alien_cache(struct array_cache **ac_ptr) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1054 | { |
| 1055 | int i; |
| 1056 | |
| 1057 | if (!ac_ptr) |
| 1058 | return; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1059 | for_each_node(i) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1060 | kfree(ac_ptr[i]); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1061 | kfree(ac_ptr); |
| 1062 | } |
| 1063 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1064 | static void __drain_alien_cache(struct kmem_cache *cachep, |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 1065 | struct array_cache *ac, int node) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1066 | { |
| 1067 | struct kmem_list3 *rl3 = cachep->nodelists[node]; |
| 1068 | |
| 1069 | if (ac->avail) { |
| 1070 | spin_lock(&rl3->list_lock); |
Christoph Lameter | e00946f | 2006-03-25 03:06:45 -0800 | [diff] [blame] | 1071 | /* |
| 1072 | * Stuff objects into the remote nodes shared array first. |
| 1073 | * That way we could avoid the overhead of putting the objects |
| 1074 | * into the free lists and getting them back later. |
| 1075 | */ |
shin, jacob | 693f7d3 | 2006-04-28 10:54:37 -0500 | [diff] [blame] | 1076 | if (rl3->shared) |
| 1077 | transfer_objects(rl3->shared, ac, ac->limit); |
Christoph Lameter | e00946f | 2006-03-25 03:06:45 -0800 | [diff] [blame] | 1078 | |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 1079 | free_block(cachep, ac->entry, ac->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1080 | ac->avail = 0; |
| 1081 | spin_unlock(&rl3->list_lock); |
| 1082 | } |
| 1083 | } |
| 1084 | |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 1085 | /* |
| 1086 | * Called from cache_reap() to regularly drain alien caches round robin. |
| 1087 | */ |
| 1088 | static void reap_alien(struct kmem_cache *cachep, struct kmem_list3 *l3) |
| 1089 | { |
| 1090 | int node = __get_cpu_var(reap_node); |
| 1091 | |
| 1092 | if (l3->alien) { |
| 1093 | struct array_cache *ac = l3->alien[node]; |
Christoph Lameter | e00946f | 2006-03-25 03:06:45 -0800 | [diff] [blame] | 1094 | |
| 1095 | if (ac && ac->avail && spin_trylock_irq(&ac->lock)) { |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 1096 | __drain_alien_cache(cachep, ac, node); |
| 1097 | spin_unlock_irq(&ac->lock); |
| 1098 | } |
| 1099 | } |
| 1100 | } |
| 1101 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1102 | static void drain_alien_cache(struct kmem_cache *cachep, |
| 1103 | struct array_cache **alien) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1104 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1105 | int i = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1106 | struct array_cache *ac; |
| 1107 | unsigned long flags; |
| 1108 | |
| 1109 | for_each_online_node(i) { |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1110 | ac = alien[i]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1111 | if (ac) { |
| 1112 | spin_lock_irqsave(&ac->lock, flags); |
| 1113 | __drain_alien_cache(cachep, ac, i); |
| 1114 | spin_unlock_irqrestore(&ac->lock, flags); |
| 1115 | } |
| 1116 | } |
| 1117 | } |
Pekka Enberg | 729bd0b | 2006-06-23 02:03:05 -0700 | [diff] [blame] | 1118 | |
Ingo Molnar | 873623d | 2006-07-13 14:44:38 +0200 | [diff] [blame] | 1119 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
Pekka Enberg | 729bd0b | 2006-06-23 02:03:05 -0700 | [diff] [blame] | 1120 | { |
| 1121 | struct slab *slabp = virt_to_slab(objp); |
| 1122 | int nodeid = slabp->nodeid; |
| 1123 | struct kmem_list3 *l3; |
| 1124 | struct array_cache *alien = NULL; |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1125 | int node; |
| 1126 | |
| 1127 | node = numa_node_id(); |
Pekka Enberg | 729bd0b | 2006-06-23 02:03:05 -0700 | [diff] [blame] | 1128 | |
| 1129 | /* |
| 1130 | * Make sure we are not freeing a object from another node to the array |
| 1131 | * cache on this cpu. |
| 1132 | */ |
Siddha, Suresh B | 62918a0 | 2007-05-02 19:27:18 +0200 | [diff] [blame] | 1133 | if (likely(slabp->nodeid == node)) |
Pekka Enberg | 729bd0b | 2006-06-23 02:03:05 -0700 | [diff] [blame] | 1134 | return 0; |
| 1135 | |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1136 | l3 = cachep->nodelists[node]; |
Pekka Enberg | 729bd0b | 2006-06-23 02:03:05 -0700 | [diff] [blame] | 1137 | STATS_INC_NODEFREES(cachep); |
| 1138 | if (l3->alien && l3->alien[nodeid]) { |
| 1139 | alien = l3->alien[nodeid]; |
Ingo Molnar | 873623d | 2006-07-13 14:44:38 +0200 | [diff] [blame] | 1140 | spin_lock(&alien->lock); |
Pekka Enberg | 729bd0b | 2006-06-23 02:03:05 -0700 | [diff] [blame] | 1141 | if (unlikely(alien->avail == alien->limit)) { |
| 1142 | STATS_INC_ACOVERFLOW(cachep); |
| 1143 | __drain_alien_cache(cachep, alien, nodeid); |
| 1144 | } |
| 1145 | alien->entry[alien->avail++] = objp; |
| 1146 | spin_unlock(&alien->lock); |
| 1147 | } else { |
| 1148 | spin_lock(&(cachep->nodelists[nodeid])->list_lock); |
| 1149 | free_block(cachep, &objp, 1, nodeid); |
| 1150 | spin_unlock(&(cachep->nodelists[nodeid])->list_lock); |
| 1151 | } |
| 1152 | return 1; |
| 1153 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1154 | #endif |
| 1155 | |
Chandra Seetharaman | 8c78f30 | 2006-07-30 03:03:35 -0700 | [diff] [blame] | 1156 | static int __cpuinit cpuup_callback(struct notifier_block *nfb, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1157 | unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1158 | { |
| 1159 | long cpu = (long)hcpu; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1160 | struct kmem_cache *cachep; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1161 | struct kmem_list3 *l3 = NULL; |
| 1162 | int node = cpu_to_node(cpu); |
| 1163 | int memsize = sizeof(struct kmem_list3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1164 | |
| 1165 | switch (action) { |
Heiko Carstens | 38c3bd9 | 2007-05-09 02:34:05 -0700 | [diff] [blame] | 1166 | case CPU_LOCK_ACQUIRE: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1167 | mutex_lock(&cache_chain_mutex); |
Heiko Carstens | 38c3bd9 | 2007-05-09 02:34:05 -0700 | [diff] [blame] | 1168 | break; |
| 1169 | case CPU_UP_PREPARE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1170 | case CPU_UP_PREPARE_FROZEN: |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1171 | /* |
| 1172 | * We need to do this right in the beginning since |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1173 | * alloc_arraycache's are going to use this list. |
| 1174 | * kmalloc_node allows us to add the slab to the right |
| 1175 | * kmem_list3 and not this cpu's kmem_list3 |
| 1176 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1178 | list_for_each_entry(cachep, &cache_chain, next) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1179 | /* |
| 1180 | * Set up the size64 kmemlist for cpu before we can |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1181 | * begin anything. Make sure some other cpu on this |
| 1182 | * node has not already allocated this |
| 1183 | */ |
| 1184 | if (!cachep->nodelists[node]) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1185 | l3 = kmalloc_node(memsize, GFP_KERNEL, node); |
| 1186 | if (!l3) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1187 | goto bad; |
| 1188 | kmem_list3_init(l3); |
| 1189 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1190 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1191 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1192 | /* |
| 1193 | * The l3s don't come and go as CPUs come and |
| 1194 | * go. cache_chain_mutex is sufficient |
| 1195 | * protection here. |
| 1196 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1197 | cachep->nodelists[node] = l3; |
| 1198 | } |
| 1199 | |
| 1200 | spin_lock_irq(&cachep->nodelists[node]->list_lock); |
| 1201 | cachep->nodelists[node]->free_limit = |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1202 | (1 + nr_cpus_node(node)) * |
| 1203 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1204 | spin_unlock_irq(&cachep->nodelists[node]->list_lock); |
| 1205 | } |
| 1206 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1207 | /* |
| 1208 | * Now we can go ahead with allocating the shared arrays and |
| 1209 | * array caches |
| 1210 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1211 | list_for_each_entry(cachep, &cache_chain, next) { |
Tobias Klauser | cd105df | 2006-01-08 01:00:59 -0800 | [diff] [blame] | 1212 | struct array_cache *nc; |
Eric Dumazet | 6310984 | 2007-05-06 14:49:28 -0700 | [diff] [blame] | 1213 | struct array_cache *shared = NULL; |
Paul Menage | 3395ee0 | 2006-12-06 20:32:16 -0800 | [diff] [blame] | 1214 | struct array_cache **alien = NULL; |
Tobias Klauser | cd105df | 2006-01-08 01:00:59 -0800 | [diff] [blame] | 1215 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1216 | nc = alloc_arraycache(node, cachep->limit, |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1217 | cachep->batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1218 | if (!nc) |
| 1219 | goto bad; |
Eric Dumazet | 6310984 | 2007-05-06 14:49:28 -0700 | [diff] [blame] | 1220 | if (cachep->shared) { |
| 1221 | shared = alloc_arraycache(node, |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1222 | cachep->shared * cachep->batchcount, |
| 1223 | 0xbaadf00d); |
Eric Dumazet | 6310984 | 2007-05-06 14:49:28 -0700 | [diff] [blame] | 1224 | if (!shared) |
| 1225 | goto bad; |
| 1226 | } |
Paul Menage | 3395ee0 | 2006-12-06 20:32:16 -0800 | [diff] [blame] | 1227 | if (use_alien_caches) { |
| 1228 | alien = alloc_alien_cache(node, cachep->limit); |
| 1229 | if (!alien) |
| 1230 | goto bad; |
| 1231 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1232 | cachep->array[cpu] = nc; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1233 | l3 = cachep->nodelists[node]; |
| 1234 | BUG_ON(!l3); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1235 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1236 | spin_lock_irq(&l3->list_lock); |
| 1237 | if (!l3->shared) { |
| 1238 | /* |
| 1239 | * We are serialised from CPU_DEAD or |
| 1240 | * CPU_UP_CANCELLED by the cpucontrol lock |
| 1241 | */ |
| 1242 | l3->shared = shared; |
| 1243 | shared = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1244 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1245 | #ifdef CONFIG_NUMA |
| 1246 | if (!l3->alien) { |
| 1247 | l3->alien = alien; |
| 1248 | alien = NULL; |
| 1249 | } |
| 1250 | #endif |
| 1251 | spin_unlock_irq(&l3->list_lock); |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1252 | kfree(shared); |
| 1253 | free_alien_cache(alien); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1254 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | break; |
| 1256 | case CPU_ONLINE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1257 | case CPU_ONLINE_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1258 | start_cpu_timer(cpu); |
| 1259 | break; |
| 1260 | #ifdef CONFIG_HOTPLUG_CPU |
Christoph Lameter | 5830c59 | 2007-05-09 02:34:22 -0700 | [diff] [blame] | 1261 | case CPU_DOWN_PREPARE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1262 | case CPU_DOWN_PREPARE_FROZEN: |
Christoph Lameter | 5830c59 | 2007-05-09 02:34:22 -0700 | [diff] [blame] | 1263 | /* |
| 1264 | * Shutdown cache reaper. Note that the cache_chain_mutex is |
| 1265 | * held so that if cache_reap() is invoked it cannot do |
| 1266 | * anything expensive but will only modify reap_work |
| 1267 | * and reschedule the timer. |
| 1268 | */ |
| 1269 | cancel_rearming_delayed_work(&per_cpu(reap_work, cpu)); |
| 1270 | /* Now the cache_reaper is guaranteed to be not running. */ |
| 1271 | per_cpu(reap_work, cpu).work.func = NULL; |
| 1272 | break; |
| 1273 | case CPU_DOWN_FAILED: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1274 | case CPU_DOWN_FAILED_FROZEN: |
Christoph Lameter | 5830c59 | 2007-05-09 02:34:22 -0700 | [diff] [blame] | 1275 | start_cpu_timer(cpu); |
| 1276 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | case CPU_DEAD: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1278 | case CPU_DEAD_FROZEN: |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1279 | /* |
| 1280 | * Even if all the cpus of a node are down, we don't free the |
| 1281 | * kmem_list3 of any cache. This to avoid a race between |
| 1282 | * cpu_down, and a kmalloc allocation from another cpu for |
| 1283 | * memory from the node of the cpu going down. The list3 |
| 1284 | * structure is usually allocated from kmem_cache_create() and |
| 1285 | * gets destroyed at kmem_cache_destroy(). |
| 1286 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | /* fall thru */ |
Ravikiran G Thirumalai | 8f5be20 | 2006-12-06 20:32:14 -0800 | [diff] [blame] | 1288 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1289 | case CPU_UP_CANCELED: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 1290 | case CPU_UP_CANCELED_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1291 | list_for_each_entry(cachep, &cache_chain, next) { |
| 1292 | struct array_cache *nc; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1293 | struct array_cache *shared; |
| 1294 | struct array_cache **alien; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1295 | cpumask_t mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1296 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1297 | mask = node_to_cpumask(node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1298 | /* cpu is dead; no one can alloc from it. */ |
| 1299 | nc = cachep->array[cpu]; |
| 1300 | cachep->array[cpu] = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1301 | l3 = cachep->nodelists[node]; |
| 1302 | |
| 1303 | if (!l3) |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1304 | goto free_array_cache; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1305 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 1306 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1307 | |
| 1308 | /* Free limit for this kmem_list3 */ |
| 1309 | l3->free_limit -= cachep->batchcount; |
| 1310 | if (nc) |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 1311 | free_block(cachep, nc->entry, nc->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1312 | |
| 1313 | if (!cpus_empty(mask)) { |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 1314 | spin_unlock_irq(&l3->list_lock); |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1315 | goto free_array_cache; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1316 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1317 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1318 | shared = l3->shared; |
| 1319 | if (shared) { |
Eric Dumazet | 6310984 | 2007-05-06 14:49:28 -0700 | [diff] [blame] | 1320 | free_block(cachep, shared->entry, |
| 1321 | shared->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1322 | l3->shared = NULL; |
| 1323 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1324 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1325 | alien = l3->alien; |
| 1326 | l3->alien = NULL; |
| 1327 | |
| 1328 | spin_unlock_irq(&l3->list_lock); |
| 1329 | |
| 1330 | kfree(shared); |
| 1331 | if (alien) { |
| 1332 | drain_alien_cache(cachep, alien); |
| 1333 | free_alien_cache(alien); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1334 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1335 | free_array_cache: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1336 | kfree(nc); |
| 1337 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1338 | /* |
| 1339 | * In the previous loop, all the objects were freed to |
| 1340 | * the respective cache's slabs, now we can go ahead and |
| 1341 | * shrink each nodelist to its limit. |
| 1342 | */ |
| 1343 | list_for_each_entry(cachep, &cache_chain, next) { |
| 1344 | l3 = cachep->nodelists[node]; |
| 1345 | if (!l3) |
| 1346 | continue; |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1347 | drain_freelist(cachep, l3, l3->free_objects); |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1348 | } |
Heiko Carstens | 38c3bd9 | 2007-05-09 02:34:05 -0700 | [diff] [blame] | 1349 | break; |
| 1350 | case CPU_LOCK_RELEASE: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1351 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1353 | } |
| 1354 | return NOTIFY_OK; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1355 | bad: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | return NOTIFY_BAD; |
| 1357 | } |
| 1358 | |
Chandra Seetharaman | 74b85f3 | 2006-06-27 02:54:09 -0700 | [diff] [blame] | 1359 | static struct notifier_block __cpuinitdata cpucache_notifier = { |
| 1360 | &cpuup_callback, NULL, 0 |
| 1361 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1362 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1363 | /* |
| 1364 | * swap the static kmem_list3 with kmalloced memory |
| 1365 | */ |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1366 | static void init_list(struct kmem_cache *cachep, struct kmem_list3 *list, |
| 1367 | int nodeid) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1368 | { |
| 1369 | struct kmem_list3 *ptr; |
| 1370 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1371 | ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_KERNEL, nodeid); |
| 1372 | BUG_ON(!ptr); |
| 1373 | |
| 1374 | local_irq_disable(); |
| 1375 | memcpy(ptr, list, sizeof(struct kmem_list3)); |
Ingo Molnar | 2b2d549 | 2006-07-03 00:25:28 -0700 | [diff] [blame] | 1376 | /* |
| 1377 | * Do not assume that spinlocks can be initialized via memcpy: |
| 1378 | */ |
| 1379 | spin_lock_init(&ptr->list_lock); |
| 1380 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1381 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
| 1382 | cachep->nodelists[nodeid] = ptr; |
| 1383 | local_irq_enable(); |
| 1384 | } |
| 1385 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1386 | /* |
| 1387 | * Initialisation. Called after the page allocator have been initialised and |
| 1388 | * before smp_init(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1389 | */ |
| 1390 | void __init kmem_cache_init(void) |
| 1391 | { |
| 1392 | size_t left_over; |
| 1393 | struct cache_sizes *sizes; |
| 1394 | struct cache_names *names; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1395 | int i; |
Jack Steiner | 07ed76b | 2006-03-07 21:55:46 -0800 | [diff] [blame] | 1396 | int order; |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1397 | int node; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1398 | |
Siddha, Suresh B | 62918a0 | 2007-05-02 19:27:18 +0200 | [diff] [blame] | 1399 | if (num_possible_nodes() == 1) |
| 1400 | use_alien_caches = 0; |
| 1401 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1402 | for (i = 0; i < NUM_INIT_LISTS; i++) { |
| 1403 | kmem_list3_init(&initkmem_list3[i]); |
| 1404 | if (i < MAX_NUMNODES) |
| 1405 | cache_cache.nodelists[i] = NULL; |
| 1406 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | |
| 1408 | /* |
| 1409 | * Fragmentation resistance on low memory - only use bigger |
| 1410 | * page orders on machines with more than 32MB of memory. |
| 1411 | */ |
| 1412 | if (num_physpages > (32 << 20) >> PAGE_SHIFT) |
| 1413 | slab_break_gfp_order = BREAK_GFP_ORDER_HI; |
| 1414 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1415 | /* Bootstrap is tricky, because several objects are allocated |
| 1416 | * from caches that do not exist yet: |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1417 | * 1) initialize the cache_cache cache: it contains the struct |
| 1418 | * kmem_cache structures of all caches, except cache_cache itself: |
| 1419 | * cache_cache is statically allocated. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1420 | * Initially an __init data area is used for the head array and the |
| 1421 | * kmem_list3 structures, it's replaced with a kmalloc allocated |
| 1422 | * array at the end of the bootstrap. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1423 | * 2) Create the first kmalloc cache. |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1424 | * The struct kmem_cache for the new cache is allocated normally. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1425 | * An __init data area is used for the head array. |
| 1426 | * 3) Create the remaining kmalloc caches, with minimally sized |
| 1427 | * head arrays. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1428 | * 4) Replace the __init data head arrays for cache_cache and the first |
| 1429 | * kmalloc cache with kmalloc allocated arrays. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1430 | * 5) Replace the __init data for kmem_list3 for cache_cache and |
| 1431 | * the other cache's with kmalloc allocated memory. |
| 1432 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | */ |
| 1434 | |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1435 | node = numa_node_id(); |
| 1436 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | /* 1) create the cache_cache */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1438 | INIT_LIST_HEAD(&cache_chain); |
| 1439 | list_add(&cache_cache.next, &cache_chain); |
| 1440 | cache_cache.colour_off = cache_line_size(); |
| 1441 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1442 | cache_cache.nodelists[node] = &initkmem_list3[CACHE_CACHE]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | |
Eric Dumazet | 8da3430 | 2007-05-06 14:49:29 -0700 | [diff] [blame] | 1444 | /* |
| 1445 | * struct kmem_cache size depends on nr_node_ids, which |
| 1446 | * can be less than MAX_NUMNODES. |
| 1447 | */ |
| 1448 | cache_cache.buffer_size = offsetof(struct kmem_cache, nodelists) + |
| 1449 | nr_node_ids * sizeof(struct kmem_list3 *); |
| 1450 | #if DEBUG |
| 1451 | cache_cache.obj_size = cache_cache.buffer_size; |
| 1452 | #endif |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1453 | cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, |
| 1454 | cache_line_size()); |
Eric Dumazet | 6a2d7a9 | 2006-12-13 00:34:27 -0800 | [diff] [blame] | 1455 | cache_cache.reciprocal_buffer_size = |
| 1456 | reciprocal_value(cache_cache.buffer_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | |
Jack Steiner | 07ed76b | 2006-03-07 21:55:46 -0800 | [diff] [blame] | 1458 | for (order = 0; order < MAX_ORDER; order++) { |
| 1459 | cache_estimate(order, cache_cache.buffer_size, |
| 1460 | cache_line_size(), 0, &left_over, &cache_cache.num); |
| 1461 | if (cache_cache.num) |
| 1462 | break; |
| 1463 | } |
Eric Sesterhenn | 40094fa | 2006-04-02 13:49:25 +0200 | [diff] [blame] | 1464 | BUG_ON(!cache_cache.num); |
Jack Steiner | 07ed76b | 2006-03-07 21:55:46 -0800 | [diff] [blame] | 1465 | cache_cache.gfporder = order; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1466 | cache_cache.colour = left_over / cache_cache.colour_off; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1467 | cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) + |
| 1468 | sizeof(struct slab), cache_line_size()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1469 | |
| 1470 | /* 2+3) create the kmalloc caches */ |
| 1471 | sizes = malloc_sizes; |
| 1472 | names = cache_names; |
| 1473 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1474 | /* |
| 1475 | * Initialize the caches that provide memory for the array cache and the |
| 1476 | * kmem_list3 structures first. Without this, further allocations will |
| 1477 | * bug. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1478 | */ |
| 1479 | |
| 1480 | sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1481 | sizes[INDEX_AC].cs_size, |
| 1482 | ARCH_KMALLOC_MINALIGN, |
| 1483 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, |
| 1484 | NULL, NULL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1485 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1486 | if (INDEX_AC != INDEX_L3) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1487 | sizes[INDEX_L3].cs_cachep = |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1488 | kmem_cache_create(names[INDEX_L3].name, |
| 1489 | sizes[INDEX_L3].cs_size, |
| 1490 | ARCH_KMALLOC_MINALIGN, |
| 1491 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, |
| 1492 | NULL, NULL); |
| 1493 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1494 | |
Ingo Molnar | e0a4272 | 2006-06-23 02:03:46 -0700 | [diff] [blame] | 1495 | slab_early_init = 0; |
| 1496 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1497 | while (sizes->cs_size != ULONG_MAX) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1498 | /* |
| 1499 | * For performance, all the general caches are L1 aligned. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | * This should be particularly beneficial on SMP boxes, as it |
| 1501 | * eliminates "false sharing". |
| 1502 | * Note for systems short on memory removing the alignment will |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1503 | * allow tighter packing of the smaller caches. |
| 1504 | */ |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1505 | if (!sizes->cs_cachep) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1506 | sizes->cs_cachep = kmem_cache_create(names->name, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1507 | sizes->cs_size, |
| 1508 | ARCH_KMALLOC_MINALIGN, |
| 1509 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, |
| 1510 | NULL, NULL); |
| 1511 | } |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 1512 | #ifdef CONFIG_ZONE_DMA |
| 1513 | sizes->cs_dmacachep = kmem_cache_create( |
| 1514 | names->name_dma, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1515 | sizes->cs_size, |
| 1516 | ARCH_KMALLOC_MINALIGN, |
| 1517 | ARCH_KMALLOC_FLAGS|SLAB_CACHE_DMA| |
| 1518 | SLAB_PANIC, |
| 1519 | NULL, NULL); |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 1520 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1521 | sizes++; |
| 1522 | names++; |
| 1523 | } |
| 1524 | /* 4) Replace the bootstrap head arrays */ |
| 1525 | { |
Ingo Molnar | 2b2d549 | 2006-07-03 00:25:28 -0700 | [diff] [blame] | 1526 | struct array_cache *ptr; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1527 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1528 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1529 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1530 | local_irq_disable(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1531 | BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); |
| 1532 | memcpy(ptr, cpu_cache_get(&cache_cache), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1533 | sizeof(struct arraycache_init)); |
Ingo Molnar | 2b2d549 | 2006-07-03 00:25:28 -0700 | [diff] [blame] | 1534 | /* |
| 1535 | * Do not assume that spinlocks can be initialized via memcpy: |
| 1536 | */ |
| 1537 | spin_lock_init(&ptr->lock); |
| 1538 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1539 | cache_cache.array[smp_processor_id()] = ptr; |
| 1540 | local_irq_enable(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1541 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1542 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1543 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1544 | local_irq_disable(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1545 | BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1546 | != &initarray_generic.cache); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1547 | memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1548 | sizeof(struct arraycache_init)); |
Ingo Molnar | 2b2d549 | 2006-07-03 00:25:28 -0700 | [diff] [blame] | 1549 | /* |
| 1550 | * Do not assume that spinlocks can be initialized via memcpy: |
| 1551 | */ |
| 1552 | spin_lock_init(&ptr->lock); |
| 1553 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1554 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1555 | ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1556 | local_irq_enable(); |
| 1557 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1558 | /* 5) Replace the bootstrap kmem_list3's */ |
| 1559 | { |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1560 | int nid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1561 | |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1562 | /* Replace the static kmem_list3 structures for the boot cpu */ |
| 1563 | init_list(&cache_cache, &initkmem_list3[CACHE_CACHE], node); |
| 1564 | |
| 1565 | for_each_online_node(nid) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1566 | init_list(malloc_sizes[INDEX_AC].cs_cachep, |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1567 | &initkmem_list3[SIZE_AC + nid], nid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1568 | |
| 1569 | if (INDEX_AC != INDEX_L3) { |
| 1570 | init_list(malloc_sizes[INDEX_L3].cs_cachep, |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 1571 | &initkmem_list3[SIZE_L3 + nid], nid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1572 | } |
| 1573 | } |
| 1574 | } |
| 1575 | |
| 1576 | /* 6) resize the head arrays to their final sizes */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1578 | struct kmem_cache *cachep; |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1579 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1580 | list_for_each_entry(cachep, &cache_chain, next) |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 1581 | if (enable_cpucache(cachep)) |
| 1582 | BUG(); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1583 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1584 | } |
| 1585 | |
Ravikiran G Thirumalai | 056c624 | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 1586 | /* Annotate slab for lockdep -- annotate the malloc caches */ |
| 1587 | init_lock_keys(); |
| 1588 | |
| 1589 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1590 | /* Done! */ |
| 1591 | g_cpucache_up = FULL; |
| 1592 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1593 | /* |
| 1594 | * Register a cpu startup notifier callback that initializes |
| 1595 | * cpu_cache_get for all new cpus |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1596 | */ |
| 1597 | register_cpu_notifier(&cpucache_notifier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1598 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1599 | /* |
| 1600 | * The reap timers are started later, with a module init call: That part |
| 1601 | * of the kernel is not yet operational. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1602 | */ |
| 1603 | } |
| 1604 | |
| 1605 | static int __init cpucache_init(void) |
| 1606 | { |
| 1607 | int cpu; |
| 1608 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1609 | /* |
| 1610 | * Register the timers that return unneeded pages to the page allocator |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1611 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1612 | for_each_online_cpu(cpu) |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1613 | start_cpu_timer(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | return 0; |
| 1615 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1616 | __initcall(cpucache_init); |
| 1617 | |
| 1618 | /* |
| 1619 | * Interface to system's page allocator. No need to hold the cache-lock. |
| 1620 | * |
| 1621 | * If we requested dmaable memory, we will get it. Even if we |
| 1622 | * did not request dmaable memory, we might get it, but that |
| 1623 | * would be relatively rare and ignorable. |
| 1624 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1625 | static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | { |
| 1627 | struct page *page; |
Christoph Hellwig | e1b6aa6 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 1628 | int nr_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1629 | int i; |
| 1630 | |
Luke Yang | d6fef9d | 2006-04-10 22:52:56 -0700 | [diff] [blame] | 1631 | #ifndef CONFIG_MMU |
Christoph Hellwig | e1b6aa6 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 1632 | /* |
| 1633 | * Nommu uses slab's for process anonymous memory allocations, and thus |
| 1634 | * requires __GFP_COMP to properly refcount higher order allocations |
Luke Yang | d6fef9d | 2006-04-10 22:52:56 -0700 | [diff] [blame] | 1635 | */ |
Christoph Hellwig | e1b6aa6 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 1636 | flags |= __GFP_COMP; |
Luke Yang | d6fef9d | 2006-04-10 22:52:56 -0700 | [diff] [blame] | 1637 | #endif |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 1638 | |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 1639 | flags |= cachep->gfpflags; |
Christoph Hellwig | e1b6aa6 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 1640 | |
| 1641 | page = alloc_pages_node(nodeid, flags, cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | if (!page) |
| 1643 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1644 | |
Christoph Hellwig | e1b6aa6 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 1645 | nr_pages = (1 << cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1646 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
Christoph Lameter | 972d1a7 | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 1647 | add_zone_page_state(page_zone(page), |
| 1648 | NR_SLAB_RECLAIMABLE, nr_pages); |
| 1649 | else |
| 1650 | add_zone_page_state(page_zone(page), |
| 1651 | NR_SLAB_UNRECLAIMABLE, nr_pages); |
Christoph Hellwig | e1b6aa6 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 1652 | for (i = 0; i < nr_pages; i++) |
| 1653 | __SetPageSlab(page + i); |
| 1654 | return page_address(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1655 | } |
| 1656 | |
| 1657 | /* |
| 1658 | * Interface to system's page release. |
| 1659 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1660 | static void kmem_freepages(struct kmem_cache *cachep, void *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1661 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1662 | unsigned long i = (1 << cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1663 | struct page *page = virt_to_page(addr); |
| 1664 | const unsigned long nr_freed = i; |
| 1665 | |
Christoph Lameter | 972d1a7 | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 1666 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1667 | sub_zone_page_state(page_zone(page), |
| 1668 | NR_SLAB_RECLAIMABLE, nr_freed); |
| 1669 | else |
| 1670 | sub_zone_page_state(page_zone(page), |
| 1671 | NR_SLAB_UNRECLAIMABLE, nr_freed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1672 | while (i--) { |
Nick Piggin | f205b2f | 2006-03-22 00:08:02 -0800 | [diff] [blame] | 1673 | BUG_ON(!PageSlab(page)); |
| 1674 | __ClearPageSlab(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1675 | page++; |
| 1676 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1677 | if (current->reclaim_state) |
| 1678 | current->reclaim_state->reclaimed_slab += nr_freed; |
| 1679 | free_pages((unsigned long)addr, cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1680 | } |
| 1681 | |
| 1682 | static void kmem_rcu_free(struct rcu_head *head) |
| 1683 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1684 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1685 | struct kmem_cache *cachep = slab_rcu->cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1686 | |
| 1687 | kmem_freepages(cachep, slab_rcu->addr); |
| 1688 | if (OFF_SLAB(cachep)) |
| 1689 | kmem_cache_free(cachep->slabp_cache, slab_rcu); |
| 1690 | } |
| 1691 | |
| 1692 | #if DEBUG |
| 1693 | |
| 1694 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1695 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1696 | unsigned long caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1697 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1698 | int size = obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1700 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1701 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1702 | if (size < 5 * sizeof(unsigned long)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1703 | return; |
| 1704 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1705 | *addr++ = 0x12345678; |
| 1706 | *addr++ = caller; |
| 1707 | *addr++ = smp_processor_id(); |
| 1708 | size -= 3 * sizeof(unsigned long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1709 | { |
| 1710 | unsigned long *sptr = &caller; |
| 1711 | unsigned long svalue; |
| 1712 | |
| 1713 | while (!kstack_end(sptr)) { |
| 1714 | svalue = *sptr++; |
| 1715 | if (kernel_text_address(svalue)) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1716 | *addr++ = svalue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1717 | size -= sizeof(unsigned long); |
| 1718 | if (size <= sizeof(unsigned long)) |
| 1719 | break; |
| 1720 | } |
| 1721 | } |
| 1722 | |
| 1723 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1724 | *addr++ = 0x87654321; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1725 | } |
| 1726 | #endif |
| 1727 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1728 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1729 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1730 | int size = obj_size(cachep); |
| 1731 | addr = &((char *)addr)[obj_offset(cachep)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1732 | |
| 1733 | memset(addr, val, size); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1734 | *(unsigned char *)(addr + size - 1) = POISON_END; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1735 | } |
| 1736 | |
| 1737 | static void dump_line(char *data, int offset, int limit) |
| 1738 | { |
| 1739 | int i; |
Dave Jones | aa83aa4 | 2006-09-29 01:59:51 -0700 | [diff] [blame] | 1740 | unsigned char error = 0; |
| 1741 | int bad_count = 0; |
| 1742 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1743 | printk(KERN_ERR "%03x:", offset); |
Dave Jones | aa83aa4 | 2006-09-29 01:59:51 -0700 | [diff] [blame] | 1744 | for (i = 0; i < limit; i++) { |
| 1745 | if (data[offset + i] != POISON_FREE) { |
| 1746 | error = data[offset + i]; |
| 1747 | bad_count++; |
| 1748 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1749 | printk(" %02x", (unsigned char)data[offset + i]); |
Dave Jones | aa83aa4 | 2006-09-29 01:59:51 -0700 | [diff] [blame] | 1750 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1751 | printk("\n"); |
Dave Jones | aa83aa4 | 2006-09-29 01:59:51 -0700 | [diff] [blame] | 1752 | |
| 1753 | if (bad_count == 1) { |
| 1754 | error ^= POISON_FREE; |
| 1755 | if (!(error & (error - 1))) { |
| 1756 | printk(KERN_ERR "Single bit error detected. Probably " |
| 1757 | "bad RAM.\n"); |
| 1758 | #ifdef CONFIG_X86 |
| 1759 | printk(KERN_ERR "Run memtest86+ or a similar memory " |
| 1760 | "test tool.\n"); |
| 1761 | #else |
| 1762 | printk(KERN_ERR "Run a memory test tool.\n"); |
| 1763 | #endif |
| 1764 | } |
| 1765 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1766 | } |
| 1767 | #endif |
| 1768 | |
| 1769 | #if DEBUG |
| 1770 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1771 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1772 | { |
| 1773 | int i, size; |
| 1774 | char *realobj; |
| 1775 | |
| 1776 | if (cachep->flags & SLAB_RED_ZONE) { |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 1777 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1778 | *dbg_redzone1(cachep, objp), |
| 1779 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1780 | } |
| 1781 | |
| 1782 | if (cachep->flags & SLAB_STORE_USER) { |
| 1783 | printk(KERN_ERR "Last user: [<%p>]", |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1784 | *dbg_userword(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1785 | print_symbol("(%s)", |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1786 | (unsigned long)*dbg_userword(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1787 | printk("\n"); |
| 1788 | } |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1789 | realobj = (char *)objp + obj_offset(cachep); |
| 1790 | size = obj_size(cachep); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1791 | for (i = 0; i < size && lines; i += 16, lines--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1792 | int limit; |
| 1793 | limit = 16; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1794 | if (i + limit > size) |
| 1795 | limit = size - i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1796 | dump_line(realobj, i, limit); |
| 1797 | } |
| 1798 | } |
| 1799 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1800 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1801 | { |
| 1802 | char *realobj; |
| 1803 | int size, i; |
| 1804 | int lines = 0; |
| 1805 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1806 | realobj = (char *)objp + obj_offset(cachep); |
| 1807 | size = obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1808 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1809 | for (i = 0; i < size; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1810 | char exp = POISON_FREE; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1811 | if (i == size - 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1812 | exp = POISON_END; |
| 1813 | if (realobj[i] != exp) { |
| 1814 | int limit; |
| 1815 | /* Mismatch ! */ |
| 1816 | /* Print header */ |
| 1817 | if (lines == 0) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1818 | printk(KERN_ERR |
David Howells | e94a40c | 2007-04-02 23:46:28 +0100 | [diff] [blame] | 1819 | "Slab corruption: %s start=%p, len=%d\n", |
| 1820 | cachep->name, realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1821 | print_objinfo(cachep, objp, 0); |
| 1822 | } |
| 1823 | /* Hexdump the affected line */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1824 | i = (i / 16) * 16; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1825 | limit = 16; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1826 | if (i + limit > size) |
| 1827 | limit = size - i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1828 | dump_line(realobj, i, limit); |
| 1829 | i += 16; |
| 1830 | lines++; |
| 1831 | /* Limit to 5 lines */ |
| 1832 | if (lines > 5) |
| 1833 | break; |
| 1834 | } |
| 1835 | } |
| 1836 | if (lines != 0) { |
| 1837 | /* Print some data about the neighboring objects, if they |
| 1838 | * exist: |
| 1839 | */ |
Pekka Enberg | 6ed5eb2 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1840 | struct slab *slabp = virt_to_slab(objp); |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 1841 | unsigned int objnr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1842 | |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 1843 | objnr = obj_to_index(cachep, slabp, objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1844 | if (objnr) { |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 1845 | objp = index_to_obj(cachep, slabp, objnr - 1); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1846 | realobj = (char *)objp + obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1847 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1848 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1849 | print_objinfo(cachep, objp, 2); |
| 1850 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1851 | if (objnr + 1 < cachep->num) { |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 1852 | objp = index_to_obj(cachep, slabp, objnr + 1); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1853 | realobj = (char *)objp + obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1854 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1855 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1856 | print_objinfo(cachep, objp, 2); |
| 1857 | } |
| 1858 | } |
| 1859 | } |
| 1860 | #endif |
| 1861 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1862 | #if DEBUG |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1863 | /** |
Randy Dunlap | 911851e | 2006-03-22 00:08:14 -0800 | [diff] [blame] | 1864 | * slab_destroy_objs - destroy a slab and its objects |
| 1865 | * @cachep: cache pointer being destroyed |
| 1866 | * @slabp: slab pointer being destroyed |
| 1867 | * |
| 1868 | * Call the registered destructor for each object in a slab that is being |
| 1869 | * destroyed. |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1870 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1871 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1872 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1873 | int i; |
| 1874 | for (i = 0; i < cachep->num; i++) { |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 1875 | void *objp = index_to_obj(cachep, slabp, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1876 | |
| 1877 | if (cachep->flags & SLAB_POISON) { |
| 1878 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1879 | if (cachep->buffer_size % PAGE_SIZE == 0 && |
| 1880 | OFF_SLAB(cachep)) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1881 | kernel_map_pages(virt_to_page(objp), |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1882 | cachep->buffer_size / PAGE_SIZE, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1883 | else |
| 1884 | check_poison_obj(cachep, objp); |
| 1885 | #else |
| 1886 | check_poison_obj(cachep, objp); |
| 1887 | #endif |
| 1888 | } |
| 1889 | if (cachep->flags & SLAB_RED_ZONE) { |
| 1890 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 1891 | slab_error(cachep, "start of a freed object " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1892 | "was overwritten"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1893 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 1894 | slab_error(cachep, "end of a freed object " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1895 | "was overwritten"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1896 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1897 | } |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1898 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1899 | #else |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1900 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1901 | { |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1902 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1903 | #endif |
| 1904 | |
Randy Dunlap | 911851e | 2006-03-22 00:08:14 -0800 | [diff] [blame] | 1905 | /** |
| 1906 | * slab_destroy - destroy and release all objects in a slab |
| 1907 | * @cachep: cache pointer being destroyed |
| 1908 | * @slabp: slab pointer being destroyed |
| 1909 | * |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1910 | * Destroy all the objs in a slab, and release the mem back to the system. |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1911 | * Before calling the slab must have been unlinked from the cache. The |
| 1912 | * cache-lock is not held/needed. |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1913 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1914 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1915 | { |
| 1916 | void *addr = slabp->s_mem - slabp->colouroff; |
| 1917 | |
| 1918 | slab_destroy_objs(cachep, slabp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1919 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
| 1920 | struct slab_rcu *slab_rcu; |
| 1921 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1922 | slab_rcu = (struct slab_rcu *)slabp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1923 | slab_rcu->cachep = cachep; |
| 1924 | slab_rcu->addr = addr; |
| 1925 | call_rcu(&slab_rcu->head, kmem_rcu_free); |
| 1926 | } else { |
| 1927 | kmem_freepages(cachep, addr); |
Ingo Molnar | 873623d | 2006-07-13 14:44:38 +0200 | [diff] [blame] | 1928 | if (OFF_SLAB(cachep)) |
| 1929 | kmem_cache_free(cachep->slabp_cache, slabp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1930 | } |
| 1931 | } |
| 1932 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1933 | /* |
| 1934 | * For setting up all the kmem_list3s for cache whose buffer_size is same as |
| 1935 | * size of kmem_list3. |
| 1936 | */ |
Andrew Morton | a3a02be | 2007-05-06 14:49:31 -0700 | [diff] [blame] | 1937 | static void __init set_up_list3s(struct kmem_cache *cachep, int index) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1938 | { |
| 1939 | int node; |
| 1940 | |
| 1941 | for_each_online_node(node) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1942 | cachep->nodelists[node] = &initkmem_list3[index + node]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1943 | cachep->nodelists[node]->next_reap = jiffies + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1944 | REAPTIMEOUT_LIST3 + |
| 1945 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1946 | } |
| 1947 | } |
| 1948 | |
Christoph Lameter | 117f6eb | 2006-09-25 23:31:37 -0700 | [diff] [blame] | 1949 | static void __kmem_cache_destroy(struct kmem_cache *cachep) |
| 1950 | { |
| 1951 | int i; |
| 1952 | struct kmem_list3 *l3; |
| 1953 | |
| 1954 | for_each_online_cpu(i) |
| 1955 | kfree(cachep->array[i]); |
| 1956 | |
| 1957 | /* NUMA: free the list3 structures */ |
| 1958 | for_each_online_node(i) { |
| 1959 | l3 = cachep->nodelists[i]; |
| 1960 | if (l3) { |
| 1961 | kfree(l3->shared); |
| 1962 | free_alien_cache(l3->alien); |
| 1963 | kfree(l3); |
| 1964 | } |
| 1965 | } |
| 1966 | kmem_cache_free(&cache_cache, cachep); |
| 1967 | } |
| 1968 | |
| 1969 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1970 | /** |
Randy.Dunlap | a70773d | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 1971 | * calculate_slab_order - calculate size (page order) of slabs |
| 1972 | * @cachep: pointer to the cache that is being created |
| 1973 | * @size: size of objects to be created in this cache. |
| 1974 | * @align: required alignment for the objects. |
| 1975 | * @flags: slab allocation flags |
| 1976 | * |
| 1977 | * Also calculates the number of objects per slab. |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1978 | * |
| 1979 | * This could be made much more intelligent. For now, try to avoid using |
| 1980 | * high order pages for slabs. When the gfp() functions are more friendly |
| 1981 | * towards high-order requests, this should be changed. |
| 1982 | */ |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 1983 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
Randy Dunlap | ee13d78 | 2006-02-01 03:05:53 -0800 | [diff] [blame] | 1984 | size_t size, size_t align, unsigned long flags) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1985 | { |
Ingo Molnar | b1ab41c | 2006-06-02 15:44:58 +0200 | [diff] [blame] | 1986 | unsigned long offslab_limit; |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1987 | size_t left_over = 0; |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1988 | int gfporder; |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1989 | |
Christoph Lameter | 0aa817f | 2007-05-16 22:11:01 -0700 | [diff] [blame] | 1990 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1991 | unsigned int num; |
| 1992 | size_t remainder; |
| 1993 | |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1994 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1995 | if (!num) |
| 1996 | continue; |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1997 | |
Ingo Molnar | b1ab41c | 2006-06-02 15:44:58 +0200 | [diff] [blame] | 1998 | if (flags & CFLGS_OFF_SLAB) { |
| 1999 | /* |
| 2000 | * Max number of objs-per-slab for caches which |
| 2001 | * use off-slab slabs. Needed to avoid a possible |
| 2002 | * looping condition in cache_grow(). |
| 2003 | */ |
| 2004 | offslab_limit = size - sizeof(struct slab); |
| 2005 | offslab_limit /= sizeof(kmem_bufctl_t); |
| 2006 | |
| 2007 | if (num > offslab_limit) |
| 2008 | break; |
| 2009 | } |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 2010 | |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 2011 | /* Found something acceptable - save it away */ |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 2012 | cachep->num = num; |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 2013 | cachep->gfporder = gfporder; |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 2014 | left_over = remainder; |
| 2015 | |
| 2016 | /* |
Linus Torvalds | f78bb8a | 2006-03-08 10:33:05 -0800 | [diff] [blame] | 2017 | * A VFS-reclaimable slab tends to have most allocations |
| 2018 | * as GFP_NOFS and we really don't want to have to be allocating |
| 2019 | * higher-order pages when we are unable to shrink dcache. |
| 2020 | */ |
| 2021 | if (flags & SLAB_RECLAIM_ACCOUNT) |
| 2022 | break; |
| 2023 | |
| 2024 | /* |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 2025 | * Large number of objects is good, but very large slabs are |
| 2026 | * currently bad for the gfp()s. |
| 2027 | */ |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 2028 | if (gfporder >= slab_break_gfp_order) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 2029 | break; |
| 2030 | |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 2031 | /* |
| 2032 | * Acceptable internal fragmentation? |
| 2033 | */ |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2034 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 2035 | break; |
| 2036 | } |
| 2037 | return left_over; |
| 2038 | } |
| 2039 | |
Sam Ravnborg | 38bdc32 | 2007-05-17 23:48:19 +0200 | [diff] [blame] | 2040 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep) |
Pekka Enberg | f30cf7d | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2041 | { |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 2042 | if (g_cpucache_up == FULL) |
| 2043 | return enable_cpucache(cachep); |
| 2044 | |
Pekka Enberg | f30cf7d | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2045 | if (g_cpucache_up == NONE) { |
| 2046 | /* |
| 2047 | * Note: the first kmem_cache_create must create the cache |
| 2048 | * that's used by kmalloc(24), otherwise the creation of |
| 2049 | * further caches will BUG(). |
| 2050 | */ |
| 2051 | cachep->array[smp_processor_id()] = &initarray_generic.cache; |
| 2052 | |
| 2053 | /* |
| 2054 | * If the cache that's used by kmalloc(sizeof(kmem_list3)) is |
| 2055 | * the first cache, then we need to set up all its list3s, |
| 2056 | * otherwise the creation of further caches will BUG(). |
| 2057 | */ |
| 2058 | set_up_list3s(cachep, SIZE_AC); |
| 2059 | if (INDEX_AC == INDEX_L3) |
| 2060 | g_cpucache_up = PARTIAL_L3; |
| 2061 | else |
| 2062 | g_cpucache_up = PARTIAL_AC; |
| 2063 | } else { |
| 2064 | cachep->array[smp_processor_id()] = |
| 2065 | kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
| 2066 | |
| 2067 | if (g_cpucache_up == PARTIAL_AC) { |
| 2068 | set_up_list3s(cachep, SIZE_L3); |
| 2069 | g_cpucache_up = PARTIAL_L3; |
| 2070 | } else { |
| 2071 | int node; |
| 2072 | for_each_online_node(node) { |
| 2073 | cachep->nodelists[node] = |
| 2074 | kmalloc_node(sizeof(struct kmem_list3), |
| 2075 | GFP_KERNEL, node); |
| 2076 | BUG_ON(!cachep->nodelists[node]); |
| 2077 | kmem_list3_init(cachep->nodelists[node]); |
| 2078 | } |
| 2079 | } |
| 2080 | } |
| 2081 | cachep->nodelists[numa_node_id()]->next_reap = |
| 2082 | jiffies + REAPTIMEOUT_LIST3 + |
| 2083 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
| 2084 | |
| 2085 | cpu_cache_get(cachep)->avail = 0; |
| 2086 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; |
| 2087 | cpu_cache_get(cachep)->batchcount = 1; |
| 2088 | cpu_cache_get(cachep)->touched = 0; |
| 2089 | cachep->batchcount = 1; |
| 2090 | cachep->limit = BOOT_CPUCACHE_ENTRIES; |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 2091 | return 0; |
Pekka Enberg | f30cf7d | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2092 | } |
| 2093 | |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 2094 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | * kmem_cache_create - Create a cache. |
| 2096 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 2097 | * @size: The size of objects to be created in this cache. |
| 2098 | * @align: The required alignment for the objects. |
| 2099 | * @flags: SLAB flags |
| 2100 | * @ctor: A constructor for the objects. |
Christoph Lameter | c59def9f | 2007-05-16 22:10:50 -0700 | [diff] [blame] | 2101 | * @dtor: A destructor for the objects (not implemented anymore). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2102 | * |
| 2103 | * Returns a ptr to the cache on success, NULL on failure. |
| 2104 | * Cannot be called within a int, but can be interrupted. |
| 2105 | * The @ctor is run when new pages are allocated by the cache |
| 2106 | * and the @dtor is run before the pages are handed back. |
| 2107 | * |
| 2108 | * @name must be valid until the cache is destroyed. This implies that |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2109 | * the module calling this has to destroy the cache before getting unloaded. |
| 2110 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2111 | * The flags are |
| 2112 | * |
| 2113 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 2114 | * to catch references to uninitialised memory. |
| 2115 | * |
| 2116 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check |
| 2117 | * for buffer overruns. |
| 2118 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2119 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 2120 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 2121 | * as davem. |
| 2122 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2123 | struct kmem_cache * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2124 | kmem_cache_create (const char *name, size_t size, size_t align, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2125 | unsigned long flags, |
| 2126 | void (*ctor)(void*, struct kmem_cache *, unsigned long), |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2127 | void (*dtor)(void*, struct kmem_cache *, unsigned long)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2128 | { |
| 2129 | size_t left_over, slab_size, ralign; |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 2130 | struct kmem_cache *cachep = NULL, *pc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | |
| 2132 | /* |
| 2133 | * Sanity checks... these are all serious usage bugs. |
| 2134 | */ |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2135 | if (!name || in_interrupt() || (size < BYTES_PER_WORD) || |
Christoph Lameter | 0aa817f | 2007-05-16 22:11:01 -0700 | [diff] [blame] | 2136 | size > KMALLOC_MAX_SIZE || dtor) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2137 | printk(KERN_ERR "%s: Early error in slab %s\n", __FUNCTION__, |
| 2138 | name); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2139 | BUG(); |
| 2140 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2141 | |
Ravikiran G Thirumalai | f0188f4 | 2006-02-10 01:51:13 -0800 | [diff] [blame] | 2142 | /* |
Ravikiran G Thirumalai | 8f5be20 | 2006-12-06 20:32:14 -0800 | [diff] [blame] | 2143 | * We use cache_chain_mutex to ensure a consistent view of |
| 2144 | * cpu_online_map as well. Please see cpuup_callback |
Ravikiran G Thirumalai | f0188f4 | 2006-02-10 01:51:13 -0800 | [diff] [blame] | 2145 | */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2146 | mutex_lock(&cache_chain_mutex); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 2147 | |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 2148 | list_for_each_entry(pc, &cache_chain, next) { |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 2149 | char tmp; |
| 2150 | int res; |
| 2151 | |
| 2152 | /* |
| 2153 | * This happens when the module gets unloaded and doesn't |
| 2154 | * destroy its slab cache and no-one else reuses the vmalloc |
| 2155 | * area of the module. Print a warning. |
| 2156 | */ |
Andrew Morton | 138ae66 | 2006-12-06 20:36:41 -0800 | [diff] [blame] | 2157 | res = probe_kernel_address(pc->name, tmp); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 2158 | if (res) { |
matze | b416952 | 2007-05-06 14:49:52 -0700 | [diff] [blame] | 2159 | printk(KERN_ERR |
| 2160 | "SLAB: cache with size %d has lost its name\n", |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2161 | pc->buffer_size); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 2162 | continue; |
| 2163 | } |
| 2164 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2165 | if (!strcmp(pc->name, name)) { |
matze | b416952 | 2007-05-06 14:49:52 -0700 | [diff] [blame] | 2166 | printk(KERN_ERR |
| 2167 | "kmem_cache_create: duplicate cache %s\n", name); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 2168 | dump_stack(); |
| 2169 | goto oops; |
| 2170 | } |
| 2171 | } |
| 2172 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2173 | #if DEBUG |
| 2174 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | #if FORCED_DEBUG |
| 2176 | /* |
| 2177 | * Enable redzoning and last user accounting, except for caches with |
| 2178 | * large objects, if the increased size would increase the object size |
| 2179 | * above the next power of two: caches with object sizes just above a |
| 2180 | * power of two have a significant amount of internal fragmentation. |
| 2181 | */ |
David Woodhouse | 87a927c | 2007-07-04 21:26:44 -0400 | [diff] [blame] | 2182 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
| 2183 | 2 * sizeof(unsigned long long))) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2184 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2185 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
| 2186 | flags |= SLAB_POISON; |
| 2187 | #endif |
| 2188 | if (flags & SLAB_DESTROY_BY_RCU) |
| 2189 | BUG_ON(flags & SLAB_POISON); |
| 2190 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2191 | /* |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2192 | * Always checks flags, a caller might be expecting debug support which |
| 2193 | * isn't available. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | */ |
Eric Sesterhenn | 40094fa | 2006-04-02 13:49:25 +0200 | [diff] [blame] | 2195 | BUG_ON(flags & ~CREATE_MASK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2196 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2197 | /* |
| 2198 | * Check that size is in terms of words. This is needed to avoid |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2199 | * unaligned accesses for some archs when redzoning is used, and makes |
| 2200 | * sure any on-slab bufctl's are also correctly aligned. |
| 2201 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2202 | if (size & (BYTES_PER_WORD - 1)) { |
| 2203 | size += (BYTES_PER_WORD - 1); |
| 2204 | size &= ~(BYTES_PER_WORD - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2205 | } |
| 2206 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2207 | /* calculate the final buffer alignment: */ |
| 2208 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2209 | /* 1) arch recommendation: can be overridden for debug */ |
| 2210 | if (flags & SLAB_HWCACHE_ALIGN) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2211 | /* |
| 2212 | * Default alignment: as specified by the arch code. Except if |
| 2213 | * an object is really small, then squeeze multiple objects into |
| 2214 | * one cacheline. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2215 | */ |
| 2216 | ralign = cache_line_size(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2217 | while (size <= ralign / 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2218 | ralign /= 2; |
| 2219 | } else { |
| 2220 | ralign = BYTES_PER_WORD; |
| 2221 | } |
Pekka Enberg | ca5f970 | 2006-09-25 23:31:25 -0700 | [diff] [blame] | 2222 | |
| 2223 | /* |
David Woodhouse | 87a927c | 2007-07-04 21:26:44 -0400 | [diff] [blame] | 2224 | * Redzoning and user store require word alignment or possibly larger. |
| 2225 | * Note this will be overridden by architecture or caller mandated |
| 2226 | * alignment if either is greater than BYTES_PER_WORD. |
Pekka Enberg | ca5f970 | 2006-09-25 23:31:25 -0700 | [diff] [blame] | 2227 | */ |
David Woodhouse | 87a927c | 2007-07-04 21:26:44 -0400 | [diff] [blame] | 2228 | if (flags & SLAB_STORE_USER) |
| 2229 | ralign = BYTES_PER_WORD; |
| 2230 | |
| 2231 | if (flags & SLAB_RED_ZONE) { |
| 2232 | ralign = REDZONE_ALIGN; |
| 2233 | /* If redzoning, ensure that the second redzone is suitably |
| 2234 | * aligned, by adjusting the object size accordingly. */ |
| 2235 | size += REDZONE_ALIGN - 1; |
| 2236 | size &= ~(REDZONE_ALIGN - 1); |
| 2237 | } |
Pekka Enberg | ca5f970 | 2006-09-25 23:31:25 -0700 | [diff] [blame] | 2238 | |
Kevin Hilman | a44b56d | 2006-12-06 20:32:11 -0800 | [diff] [blame] | 2239 | /* 2) arch mandated alignment */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2240 | if (ralign < ARCH_SLAB_MINALIGN) { |
| 2241 | ralign = ARCH_SLAB_MINALIGN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2242 | } |
Kevin Hilman | a44b56d | 2006-12-06 20:32:11 -0800 | [diff] [blame] | 2243 | /* 3) caller mandated alignment */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2244 | if (ralign < align) { |
| 2245 | ralign = align; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2246 | } |
Kevin Hilman | a44b56d | 2006-12-06 20:32:11 -0800 | [diff] [blame] | 2247 | /* disable debug if necessary */ |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 2248 | if (ralign > __alignof__(unsigned long long)) |
Kevin Hilman | a44b56d | 2006-12-06 20:32:11 -0800 | [diff] [blame] | 2249 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2250 | /* |
Pekka Enberg | ca5f970 | 2006-09-25 23:31:25 -0700 | [diff] [blame] | 2251 | * 4) Store it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2252 | */ |
| 2253 | align = ralign; |
| 2254 | |
| 2255 | /* Get cache's description obj. */ |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 2256 | cachep = kmem_cache_zalloc(&cache_cache, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2257 | if (!cachep) |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 2258 | goto oops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2259 | |
| 2260 | #if DEBUG |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2261 | cachep->obj_size = size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2262 | |
Pekka Enberg | ca5f970 | 2006-09-25 23:31:25 -0700 | [diff] [blame] | 2263 | /* |
| 2264 | * Both debugging options require word-alignment which is calculated |
| 2265 | * into align above. |
| 2266 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2267 | if (flags & SLAB_RED_ZONE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2268 | /* add space for red zone words */ |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 2269 | cachep->obj_offset += sizeof(unsigned long long); |
| 2270 | size += 2 * sizeof(unsigned long long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2271 | } |
| 2272 | if (flags & SLAB_STORE_USER) { |
Pekka Enberg | ca5f970 | 2006-09-25 23:31:25 -0700 | [diff] [blame] | 2273 | /* user store requires one word storage behind the end of |
David Woodhouse | 87a927c | 2007-07-04 21:26:44 -0400 | [diff] [blame] | 2274 | * the real object. But if the second red zone needs to be |
| 2275 | * aligned to 64 bits, we must allow that much space. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2276 | */ |
David Woodhouse | 87a927c | 2007-07-04 21:26:44 -0400 | [diff] [blame] | 2277 | if (flags & SLAB_RED_ZONE) |
| 2278 | size += REDZONE_ALIGN; |
| 2279 | else |
| 2280 | size += BYTES_PER_WORD; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2281 | } |
| 2282 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2283 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2284 | && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { |
| 2285 | cachep->obj_offset += PAGE_SIZE - size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2286 | size = PAGE_SIZE; |
| 2287 | } |
| 2288 | #endif |
| 2289 | #endif |
| 2290 | |
Ingo Molnar | e0a4272 | 2006-06-23 02:03:46 -0700 | [diff] [blame] | 2291 | /* |
| 2292 | * Determine if the slab management is 'on' or 'off' slab. |
| 2293 | * (bootstrapping cannot cope with offslab caches so don't do |
| 2294 | * it too early on.) |
| 2295 | */ |
| 2296 | if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2297 | /* |
| 2298 | * Size is large, assume best to place the slab management obj |
| 2299 | * off-slab (should allow better packing of objs). |
| 2300 | */ |
| 2301 | flags |= CFLGS_OFF_SLAB; |
| 2302 | |
| 2303 | size = ALIGN(size, align); |
| 2304 | |
Linus Torvalds | f78bb8a | 2006-03-08 10:33:05 -0800 | [diff] [blame] | 2305 | left_over = calculate_slab_order(cachep, size, align, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2306 | |
| 2307 | if (!cachep->num) { |
matze | b416952 | 2007-05-06 14:49:52 -0700 | [diff] [blame] | 2308 | printk(KERN_ERR |
| 2309 | "kmem_cache_create: couldn't create cache %s.\n", name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2310 | kmem_cache_free(&cache_cache, cachep); |
| 2311 | cachep = NULL; |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 2312 | goto oops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2313 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2314 | slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) |
| 2315 | + sizeof(struct slab), align); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2316 | |
| 2317 | /* |
| 2318 | * If the slab has been placed off-slab, and we have enough space then |
| 2319 | * move it on-slab. This is at the expense of any extra colouring. |
| 2320 | */ |
| 2321 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { |
| 2322 | flags &= ~CFLGS_OFF_SLAB; |
| 2323 | left_over -= slab_size; |
| 2324 | } |
| 2325 | |
| 2326 | if (flags & CFLGS_OFF_SLAB) { |
| 2327 | /* really off slab. No need for manual alignment */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2328 | slab_size = |
| 2329 | cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2330 | } |
| 2331 | |
| 2332 | cachep->colour_off = cache_line_size(); |
| 2333 | /* Offset must be a multiple of the alignment. */ |
| 2334 | if (cachep->colour_off < align) |
| 2335 | cachep->colour_off = align; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2336 | cachep->colour = left_over / cachep->colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | cachep->slab_size = slab_size; |
| 2338 | cachep->flags = flags; |
| 2339 | cachep->gfpflags = 0; |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 2340 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2341 | cachep->gfpflags |= GFP_DMA; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2342 | cachep->buffer_size = size; |
Eric Dumazet | 6a2d7a9 | 2006-12-13 00:34:27 -0800 | [diff] [blame] | 2343 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2344 | |
Ravikiran G Thirumalai | e5ac9c5 | 2006-09-25 23:31:34 -0700 | [diff] [blame] | 2345 | if (flags & CFLGS_OFF_SLAB) { |
Victor Fusco | b2d5507 | 2005-09-10 00:26:36 -0700 | [diff] [blame] | 2346 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); |
Ravikiran G Thirumalai | e5ac9c5 | 2006-09-25 23:31:34 -0700 | [diff] [blame] | 2347 | /* |
| 2348 | * This is a possibility for one of the malloc_sizes caches. |
| 2349 | * But since we go off slab only for object size greater than |
| 2350 | * PAGE_SIZE/8, and malloc_sizes gets created in ascending order, |
| 2351 | * this should not happen at all. |
| 2352 | * But leave a BUG_ON for some lucky dude. |
| 2353 | */ |
| 2354 | BUG_ON(!cachep->slabp_cache); |
| 2355 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2356 | cachep->ctor = ctor; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2357 | cachep->name = name; |
| 2358 | |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 2359 | if (setup_cpu_cache(cachep)) { |
| 2360 | __kmem_cache_destroy(cachep); |
| 2361 | cachep = NULL; |
| 2362 | goto oops; |
| 2363 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2364 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2365 | /* cache setup completed, link it into the list */ |
| 2366 | list_add(&cachep->next, &cache_chain); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2367 | oops: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2368 | if (!cachep && (flags & SLAB_PANIC)) |
| 2369 | panic("kmem_cache_create(): failed to create slab `%s'\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2370 | name); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2371 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2372 | return cachep; |
| 2373 | } |
| 2374 | EXPORT_SYMBOL(kmem_cache_create); |
| 2375 | |
| 2376 | #if DEBUG |
| 2377 | static void check_irq_off(void) |
| 2378 | { |
| 2379 | BUG_ON(!irqs_disabled()); |
| 2380 | } |
| 2381 | |
| 2382 | static void check_irq_on(void) |
| 2383 | { |
| 2384 | BUG_ON(irqs_disabled()); |
| 2385 | } |
| 2386 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2387 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2388 | { |
| 2389 | #ifdef CONFIG_SMP |
| 2390 | check_irq_off(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2391 | assert_spin_locked(&cachep->nodelists[numa_node_id()]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2392 | #endif |
| 2393 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2394 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2395 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2396 | { |
| 2397 | #ifdef CONFIG_SMP |
| 2398 | check_irq_off(); |
| 2399 | assert_spin_locked(&cachep->nodelists[node]->list_lock); |
| 2400 | #endif |
| 2401 | } |
| 2402 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2403 | #else |
| 2404 | #define check_irq_off() do { } while(0) |
| 2405 | #define check_irq_on() do { } while(0) |
| 2406 | #define check_spinlock_acquired(x) do { } while(0) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2407 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2408 | #endif |
| 2409 | |
Christoph Lameter | aab2207 | 2006-03-22 00:09:06 -0800 | [diff] [blame] | 2410 | static void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, |
| 2411 | struct array_cache *ac, |
| 2412 | int force, int node); |
| 2413 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2414 | static void do_drain(void *arg) |
| 2415 | { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2416 | struct kmem_cache *cachep = arg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2417 | struct array_cache *ac; |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2418 | int node = numa_node_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2419 | |
| 2420 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2421 | ac = cpu_cache_get(cachep); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2422 | spin_lock(&cachep->nodelists[node]->list_lock); |
| 2423 | free_block(cachep, ac->entry, ac->avail, node); |
| 2424 | spin_unlock(&cachep->nodelists[node]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2425 | ac->avail = 0; |
| 2426 | } |
| 2427 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2428 | static void drain_cpu_caches(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2429 | { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2430 | struct kmem_list3 *l3; |
| 2431 | int node; |
| 2432 | |
Andrew Morton | a07fa39 | 2006-03-22 00:08:17 -0800 | [diff] [blame] | 2433 | on_each_cpu(do_drain, cachep, 1, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2434 | check_irq_on(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2435 | for_each_online_node(node) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2436 | l3 = cachep->nodelists[node]; |
Roland Dreier | a4523a8 | 2006-05-15 11:41:00 -0700 | [diff] [blame] | 2437 | if (l3 && l3->alien) |
| 2438 | drain_alien_cache(cachep, l3->alien); |
| 2439 | } |
| 2440 | |
| 2441 | for_each_online_node(node) { |
| 2442 | l3 = cachep->nodelists[node]; |
| 2443 | if (l3) |
Christoph Lameter | aab2207 | 2006-03-22 00:09:06 -0800 | [diff] [blame] | 2444 | drain_array(cachep, l3, l3->shared, 1, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2445 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2446 | } |
| 2447 | |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2448 | /* |
| 2449 | * Remove slabs from the list of free slabs. |
| 2450 | * Specify the number of slabs to drain in tofree. |
| 2451 | * |
| 2452 | * Returns the actual number of slabs released. |
| 2453 | */ |
| 2454 | static int drain_freelist(struct kmem_cache *cache, |
| 2455 | struct kmem_list3 *l3, int tofree) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2456 | { |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2457 | struct list_head *p; |
| 2458 | int nr_freed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2459 | struct slab *slabp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2460 | |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2461 | nr_freed = 0; |
| 2462 | while (nr_freed < tofree && !list_empty(&l3->slabs_free)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2463 | |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2464 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2465 | p = l3->slabs_free.prev; |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2466 | if (p == &l3->slabs_free) { |
| 2467 | spin_unlock_irq(&l3->list_lock); |
| 2468 | goto out; |
| 2469 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2470 | |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2471 | slabp = list_entry(p, struct slab, list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2472 | #if DEBUG |
Eric Sesterhenn | 40094fa | 2006-04-02 13:49:25 +0200 | [diff] [blame] | 2473 | BUG_ON(slabp->inuse); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2474 | #endif |
| 2475 | list_del(&slabp->list); |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2476 | /* |
| 2477 | * Safe to drop the lock. The slab is no longer linked |
| 2478 | * to the cache. |
| 2479 | */ |
| 2480 | l3->free_objects -= cache->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2481 | spin_unlock_irq(&l3->list_lock); |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2482 | slab_destroy(cache, slabp); |
| 2483 | nr_freed++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2484 | } |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2485 | out: |
| 2486 | return nr_freed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2487 | } |
| 2488 | |
Ravikiran G Thirumalai | 8f5be20 | 2006-12-06 20:32:14 -0800 | [diff] [blame] | 2489 | /* Called with cache_chain_mutex held to protect against cpu hotplug */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2490 | static int __cache_shrink(struct kmem_cache *cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2491 | { |
| 2492 | int ret = 0, i = 0; |
| 2493 | struct kmem_list3 *l3; |
| 2494 | |
| 2495 | drain_cpu_caches(cachep); |
| 2496 | |
| 2497 | check_irq_on(); |
| 2498 | for_each_online_node(i) { |
| 2499 | l3 = cachep->nodelists[i]; |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2500 | if (!l3) |
| 2501 | continue; |
| 2502 | |
| 2503 | drain_freelist(cachep, l3, l3->free_objects); |
| 2504 | |
| 2505 | ret += !list_empty(&l3->slabs_full) || |
| 2506 | !list_empty(&l3->slabs_partial); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2507 | } |
| 2508 | return (ret ? 1 : 0); |
| 2509 | } |
| 2510 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2511 | /** |
| 2512 | * kmem_cache_shrink - Shrink a cache. |
| 2513 | * @cachep: The cache to shrink. |
| 2514 | * |
| 2515 | * Releases as many slabs as possible for a cache. |
| 2516 | * To help debugging, a zero exit status indicates all slabs were released. |
| 2517 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2518 | int kmem_cache_shrink(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2519 | { |
Ravikiran G Thirumalai | 8f5be20 | 2006-12-06 20:32:14 -0800 | [diff] [blame] | 2520 | int ret; |
Eric Sesterhenn | 40094fa | 2006-04-02 13:49:25 +0200 | [diff] [blame] | 2521 | BUG_ON(!cachep || in_interrupt()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2522 | |
Ravikiran G Thirumalai | 8f5be20 | 2006-12-06 20:32:14 -0800 | [diff] [blame] | 2523 | mutex_lock(&cache_chain_mutex); |
| 2524 | ret = __cache_shrink(cachep); |
| 2525 | mutex_unlock(&cache_chain_mutex); |
| 2526 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2527 | } |
| 2528 | EXPORT_SYMBOL(kmem_cache_shrink); |
| 2529 | |
| 2530 | /** |
| 2531 | * kmem_cache_destroy - delete a cache |
| 2532 | * @cachep: the cache to destroy |
| 2533 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 2534 | * Remove a &struct kmem_cache object from the slab cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2535 | * |
| 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 |
| 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. |
| 2540 | * |
| 2541 | * The cache must be empty before calling this function. |
| 2542 | * |
| 2543 | * The caller must guarantee that noone will allocate memory from the cache |
| 2544 | * during the kmem_cache_destroy(). |
| 2545 | */ |
Alexey Dobriyan | 133d205 | 2006-09-27 01:49:41 -0700 | [diff] [blame] | 2546 | void kmem_cache_destroy(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2547 | { |
Eric Sesterhenn | 40094fa | 2006-04-02 13:49:25 +0200 | [diff] [blame] | 2548 | BUG_ON(!cachep || in_interrupt()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2549 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2550 | /* Find the cache in the chain of caches. */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2551 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2552 | /* |
| 2553 | * the chain is never empty, cache_cache is never destroyed |
| 2554 | */ |
| 2555 | list_del(&cachep->next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2556 | if (__cache_shrink(cachep)) { |
| 2557 | slab_error(cachep, "Can't free all objects"); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2558 | list_add(&cachep->next, &cache_chain); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2559 | mutex_unlock(&cache_chain_mutex); |
Alexey Dobriyan | 133d205 | 2006-09-27 01:49:41 -0700 | [diff] [blame] | 2560 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2561 | } |
| 2562 | |
| 2563 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
Paul E. McKenney | fbd568a3e | 2005-05-01 08:59:04 -0700 | [diff] [blame] | 2564 | synchronize_rcu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2565 | |
Christoph Lameter | 117f6eb | 2006-09-25 23:31:37 -0700 | [diff] [blame] | 2566 | __kmem_cache_destroy(cachep); |
Ravikiran G Thirumalai | 8f5be20 | 2006-12-06 20:32:14 -0800 | [diff] [blame] | 2567 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2568 | } |
| 2569 | EXPORT_SYMBOL(kmem_cache_destroy); |
| 2570 | |
Ravikiran G Thirumalai | e5ac9c5 | 2006-09-25 23:31:34 -0700 | [diff] [blame] | 2571 | /* |
| 2572 | * Get the memory for a slab management obj. |
| 2573 | * For a slab cache when the slab descriptor is off-slab, slab descriptors |
| 2574 | * always come from malloc_sizes caches. The slab descriptor cannot |
| 2575 | * come from the same cache which is getting created because, |
| 2576 | * when we are searching for an appropriate cache for these |
| 2577 | * descriptors in kmem_cache_create, we search through the malloc_sizes array. |
| 2578 | * If we are creating a malloc_sizes cache here it would not be visible to |
| 2579 | * kmem_find_general_cachep till the initialization is complete. |
| 2580 | * Hence we cannot have slabp_cache same as the original cache. |
| 2581 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2582 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, |
Ravikiran G Thirumalai | 5b74ada | 2006-04-10 22:52:53 -0700 | [diff] [blame] | 2583 | int colour_off, gfp_t local_flags, |
| 2584 | int nodeid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2585 | { |
| 2586 | struct slab *slabp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2587 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2588 | if (OFF_SLAB(cachep)) { |
| 2589 | /* Slab management obj is off-slab. */ |
Ravikiran G Thirumalai | 5b74ada | 2006-04-10 22:52:53 -0700 | [diff] [blame] | 2590 | slabp = kmem_cache_alloc_node(cachep->slabp_cache, |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 2591 | local_flags & ~GFP_THISNODE, nodeid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2592 | if (!slabp) |
| 2593 | return NULL; |
| 2594 | } else { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2595 | slabp = objp + colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2596 | colour_off += cachep->slab_size; |
| 2597 | } |
| 2598 | slabp->inuse = 0; |
| 2599 | slabp->colouroff = colour_off; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2600 | slabp->s_mem = objp + colour_off; |
Ravikiran G Thirumalai | 5b74ada | 2006-04-10 22:52:53 -0700 | [diff] [blame] | 2601 | slabp->nodeid = nodeid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2602 | return slabp; |
| 2603 | } |
| 2604 | |
| 2605 | static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) |
| 2606 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2607 | return (kmem_bufctl_t *) (slabp + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2608 | } |
| 2609 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2610 | static void cache_init_objs(struct kmem_cache *cachep, |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 2611 | struct slab *slabp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2612 | { |
| 2613 | int i; |
| 2614 | |
| 2615 | for (i = 0; i < cachep->num; i++) { |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 2616 | void *objp = index_to_obj(cachep, slabp, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2617 | #if DEBUG |
| 2618 | /* need to poison the objs? */ |
| 2619 | if (cachep->flags & SLAB_POISON) |
| 2620 | poison_obj(cachep, objp, POISON_FREE); |
| 2621 | if (cachep->flags & SLAB_STORE_USER) |
| 2622 | *dbg_userword(cachep, objp) = NULL; |
| 2623 | |
| 2624 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2625 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2626 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2627 | } |
| 2628 | /* |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2629 | * Constructors are not allowed to allocate memory from the same |
| 2630 | * cache which they are a constructor for. Otherwise, deadlock. |
| 2631 | * They must also be threaded. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2632 | */ |
| 2633 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2634 | cachep->ctor(objp + obj_offset(cachep), cachep, |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 2635 | 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2636 | |
| 2637 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2638 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 2639 | slab_error(cachep, "constructor overwrote the" |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2640 | " end of an object"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2641 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 2642 | slab_error(cachep, "constructor overwrote the" |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2643 | " start of an object"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2644 | } |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2645 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && |
| 2646 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2647 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2648 | cachep->buffer_size / PAGE_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2649 | #else |
| 2650 | if (cachep->ctor) |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 2651 | cachep->ctor(objp, cachep, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2652 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2653 | slab_bufctl(slabp)[i] = i + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2654 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2655 | slab_bufctl(slabp)[i - 1] = BUFCTL_END; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2656 | slabp->free = 0; |
| 2657 | } |
| 2658 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2659 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2660 | { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 2661 | if (CONFIG_ZONE_DMA_FLAG) { |
| 2662 | if (flags & GFP_DMA) |
| 2663 | BUG_ON(!(cachep->gfpflags & GFP_DMA)); |
| 2664 | else |
| 2665 | BUG_ON(cachep->gfpflags & GFP_DMA); |
| 2666 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2667 | } |
| 2668 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2669 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, |
| 2670 | int nodeid) |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2671 | { |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 2672 | void *objp = index_to_obj(cachep, slabp, slabp->free); |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2673 | kmem_bufctl_t next; |
| 2674 | |
| 2675 | slabp->inuse++; |
| 2676 | next = slab_bufctl(slabp)[slabp->free]; |
| 2677 | #if DEBUG |
| 2678 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; |
| 2679 | WARN_ON(slabp->nodeid != nodeid); |
| 2680 | #endif |
| 2681 | slabp->free = next; |
| 2682 | |
| 2683 | return objp; |
| 2684 | } |
| 2685 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2686 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, |
| 2687 | void *objp, int nodeid) |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2688 | { |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 2689 | unsigned int objnr = obj_to_index(cachep, slabp, objp); |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2690 | |
| 2691 | #if DEBUG |
| 2692 | /* Verify that the slab belongs to the intended node */ |
| 2693 | WARN_ON(slabp->nodeid != nodeid); |
| 2694 | |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 2695 | if (slab_bufctl(slabp)[objnr] + 1 <= SLAB_LIMIT + 1) { |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2696 | printk(KERN_ERR "slab: double free detected in cache " |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2697 | "'%s', objp %p\n", cachep->name, objp); |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2698 | BUG(); |
| 2699 | } |
| 2700 | #endif |
| 2701 | slab_bufctl(slabp)[objnr] = slabp->free; |
| 2702 | slabp->free = objnr; |
| 2703 | slabp->inuse--; |
| 2704 | } |
| 2705 | |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2706 | /* |
| 2707 | * Map pages beginning at addr to the given cache and slab. This is required |
| 2708 | * for the slab allocator to be able to lookup the cache and slab of a |
| 2709 | * virtual address for kfree, ksize, kmem_ptr_validate, and slab debugging. |
| 2710 | */ |
| 2711 | static void slab_map_pages(struct kmem_cache *cache, struct slab *slab, |
| 2712 | void *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2713 | { |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2714 | int nr_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | struct page *page; |
| 2716 | |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2717 | page = virt_to_page(addr); |
Nick Piggin | 8409751 | 2006-03-22 00:08:34 -0800 | [diff] [blame] | 2718 | |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2719 | nr_pages = 1; |
Nick Piggin | 8409751 | 2006-03-22 00:08:34 -0800 | [diff] [blame] | 2720 | if (likely(!PageCompound(page))) |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2721 | nr_pages <<= cache->gfporder; |
| 2722 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2723 | do { |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2724 | page_set_cache(page, cache); |
| 2725 | page_set_slab(page, slab); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2726 | page++; |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2727 | } while (--nr_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2728 | } |
| 2729 | |
| 2730 | /* |
| 2731 | * Grow (by 1) the number of slabs within a cache. This is called by |
| 2732 | * kmem_cache_alloc() when there are no active objs left in a cache. |
| 2733 | */ |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 2734 | static int cache_grow(struct kmem_cache *cachep, |
| 2735 | gfp_t flags, int nodeid, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2736 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2737 | struct slab *slabp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2738 | size_t offset; |
| 2739 | gfp_t local_flags; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2740 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2741 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2742 | /* |
| 2743 | * Be lazy and only check for valid flags here, keeping it out of the |
| 2744 | * critical path in kmem_cache_alloc(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2745 | */ |
Christoph Lameter | cfce660 | 2007-05-06 14:50:17 -0700 | [diff] [blame] | 2746 | BUG_ON(flags & ~(GFP_DMA | GFP_LEVEL_MASK)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2747 | |
Christoph Lameter | a06d72c | 2006-12-06 20:33:12 -0800 | [diff] [blame] | 2748 | local_flags = (flags & GFP_LEVEL_MASK); |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2749 | /* Take the l3 list lock to change the colour_next on this node */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2750 | check_irq_off(); |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2751 | l3 = cachep->nodelists[nodeid]; |
| 2752 | spin_lock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2753 | |
| 2754 | /* Get colour for the slab, and cal the next value. */ |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2755 | offset = l3->colour_next; |
| 2756 | l3->colour_next++; |
| 2757 | if (l3->colour_next >= cachep->colour) |
| 2758 | l3->colour_next = 0; |
| 2759 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2760 | |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2761 | offset *= cachep->colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2762 | |
| 2763 | if (local_flags & __GFP_WAIT) |
| 2764 | local_irq_enable(); |
| 2765 | |
| 2766 | /* |
| 2767 | * The test for missing atomic flag is performed here, rather than |
| 2768 | * the more obvious place, simply to reduce the critical path length |
| 2769 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they |
| 2770 | * will eventually be caught here (where it matters). |
| 2771 | */ |
| 2772 | kmem_flagcheck(cachep, flags); |
| 2773 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2774 | /* |
| 2775 | * Get mem for the objs. Attempt to allocate a physical page from |
| 2776 | * 'nodeid'. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2777 | */ |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 2778 | if (!objp) |
| 2779 | objp = kmem_getpages(cachep, flags, nodeid); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2780 | if (!objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2781 | goto failed; |
| 2782 | |
| 2783 | /* Get slab management. */ |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 2784 | slabp = alloc_slabmgmt(cachep, objp, offset, |
| 2785 | local_flags & ~GFP_THISNODE, nodeid); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2786 | if (!slabp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2787 | goto opps1; |
| 2788 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2789 | slabp->nodeid = nodeid; |
Pekka Enberg | 4776874 | 2006-06-23 02:03:07 -0700 | [diff] [blame] | 2790 | slab_map_pages(cachep, slabp, objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2791 | |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 2792 | cache_init_objs(cachep, slabp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2793 | |
| 2794 | if (local_flags & __GFP_WAIT) |
| 2795 | local_irq_disable(); |
| 2796 | check_irq_off(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2797 | spin_lock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2798 | |
| 2799 | /* Make slab active. */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2800 | list_add_tail(&slabp->list, &(l3->slabs_free)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2801 | STATS_INC_GROWN(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2802 | l3->free_objects += cachep->num; |
| 2803 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2804 | return 1; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2805 | opps1: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2806 | kmem_freepages(cachep, objp); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2807 | failed: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2808 | if (local_flags & __GFP_WAIT) |
| 2809 | local_irq_disable(); |
| 2810 | return 0; |
| 2811 | } |
| 2812 | |
| 2813 | #if DEBUG |
| 2814 | |
| 2815 | /* |
| 2816 | * Perform extra freeing checks: |
| 2817 | * - detect bad pointers. |
| 2818 | * - POISON/RED_ZONE checking |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2819 | */ |
| 2820 | static void kfree_debugcheck(const void *objp) |
| 2821 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2822 | if (!virt_addr_valid(objp)) { |
| 2823 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2824 | (unsigned long)objp); |
| 2825 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2826 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2827 | } |
| 2828 | |
Pekka Enberg | 58ce1fd | 2006-06-23 02:03:24 -0700 | [diff] [blame] | 2829 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
| 2830 | { |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 2831 | unsigned long long redzone1, redzone2; |
Pekka Enberg | 58ce1fd | 2006-06-23 02:03:24 -0700 | [diff] [blame] | 2832 | |
| 2833 | redzone1 = *dbg_redzone1(cache, obj); |
| 2834 | redzone2 = *dbg_redzone2(cache, obj); |
| 2835 | |
| 2836 | /* |
| 2837 | * Redzone is ok. |
| 2838 | */ |
| 2839 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) |
| 2840 | return; |
| 2841 | |
| 2842 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) |
| 2843 | slab_error(cache, "double free detected"); |
| 2844 | else |
| 2845 | slab_error(cache, "memory outside object was overwritten"); |
| 2846 | |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 2847 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
Pekka Enberg | 58ce1fd | 2006-06-23 02:03:24 -0700 | [diff] [blame] | 2848 | obj, redzone1, redzone2); |
| 2849 | } |
| 2850 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2851 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2852 | void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2853 | { |
| 2854 | struct page *page; |
| 2855 | unsigned int objnr; |
| 2856 | struct slab *slabp; |
| 2857 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2858 | objp -= obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2859 | kfree_debugcheck(objp); |
Christoph Lameter | b49af68 | 2007-05-06 14:49:41 -0700 | [diff] [blame] | 2860 | page = virt_to_head_page(objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2861 | |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 2862 | slabp = page_get_slab(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2863 | |
| 2864 | if (cachep->flags & SLAB_RED_ZONE) { |
Pekka Enberg | 58ce1fd | 2006-06-23 02:03:24 -0700 | [diff] [blame] | 2865 | verify_redzone_free(cachep, objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2866 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2867 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2868 | } |
| 2869 | if (cachep->flags & SLAB_STORE_USER) |
| 2870 | *dbg_userword(cachep, objp) = caller; |
| 2871 | |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 2872 | objnr = obj_to_index(cachep, slabp, objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2873 | |
| 2874 | BUG_ON(objnr >= cachep->num); |
Pekka Enberg | 8fea4e9 | 2006-03-22 00:08:10 -0800 | [diff] [blame] | 2875 | BUG_ON(objp != index_to_obj(cachep, slabp, objnr)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2876 | |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 2877 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
| 2878 | slab_bufctl(slabp)[objnr] = BUFCTL_FREE; |
| 2879 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2880 | if (cachep->flags & SLAB_POISON) { |
| 2881 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2882 | if ((cachep->buffer_size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2883 | store_stackinfo(cachep, objp, (unsigned long)caller); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2884 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2885 | cachep->buffer_size / PAGE_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2886 | } else { |
| 2887 | poison_obj(cachep, objp, POISON_FREE); |
| 2888 | } |
| 2889 | #else |
| 2890 | poison_obj(cachep, objp, POISON_FREE); |
| 2891 | #endif |
| 2892 | } |
| 2893 | return objp; |
| 2894 | } |
| 2895 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2896 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2897 | { |
| 2898 | kmem_bufctl_t i; |
| 2899 | int entries = 0; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2900 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2901 | /* Check slab's freelist to see if this obj is there. */ |
| 2902 | for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { |
| 2903 | entries++; |
| 2904 | if (entries > cachep->num || i >= cachep->num) |
| 2905 | goto bad; |
| 2906 | } |
| 2907 | if (entries != cachep->num - slabp->inuse) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2908 | bad: |
| 2909 | printk(KERN_ERR "slab: Internal list corruption detected in " |
| 2910 | "cache '%s'(%d), slabp %p(%d). Hexdump:\n", |
| 2911 | cachep->name, cachep->num, slabp, slabp->inuse); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2912 | for (i = 0; |
Linus Torvalds | 264132b | 2006-03-06 12:10:07 -0800 | [diff] [blame] | 2913 | i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2914 | i++) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2915 | if (i % 16 == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2916 | printk("\n%03x:", i); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2917 | printk(" %02x", ((unsigned char *)slabp)[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2918 | } |
| 2919 | printk("\n"); |
| 2920 | BUG(); |
| 2921 | } |
| 2922 | } |
| 2923 | #else |
| 2924 | #define kfree_debugcheck(x) do { } while(0) |
| 2925 | #define cache_free_debugcheck(x,objp,z) (objp) |
| 2926 | #define check_slabp(x,y) do { } while(0) |
| 2927 | #endif |
| 2928 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2929 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2930 | { |
| 2931 | int batchcount; |
| 2932 | struct kmem_list3 *l3; |
| 2933 | struct array_cache *ac; |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 2934 | int node; |
| 2935 | |
| 2936 | node = numa_node_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2937 | |
| 2938 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2939 | ac = cpu_cache_get(cachep); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2940 | retry: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2941 | batchcount = ac->batchcount; |
| 2942 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 2943 | /* |
| 2944 | * If there was little recent activity on this cache, then |
| 2945 | * perform only a partial refill. Otherwise we could generate |
| 2946 | * refill bouncing. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2947 | */ |
| 2948 | batchcount = BATCHREFILL_LIMIT; |
| 2949 | } |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 2950 | l3 = cachep->nodelists[node]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2951 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2952 | BUG_ON(ac->avail > 0 || !l3); |
| 2953 | spin_lock(&l3->list_lock); |
| 2954 | |
Christoph Lameter | 3ded175 | 2006-03-25 03:06:44 -0800 | [diff] [blame] | 2955 | /* See if we can refill from the shared array */ |
| 2956 | if (l3->shared && transfer_objects(ac, l3->shared, batchcount)) |
| 2957 | goto alloc_done; |
| 2958 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2959 | while (batchcount > 0) { |
| 2960 | struct list_head *entry; |
| 2961 | struct slab *slabp; |
| 2962 | /* Get slab alloc is to come from. */ |
| 2963 | entry = l3->slabs_partial.next; |
| 2964 | if (entry == &l3->slabs_partial) { |
| 2965 | l3->free_touched = 1; |
| 2966 | entry = l3->slabs_free.next; |
| 2967 | if (entry == &l3->slabs_free) |
| 2968 | goto must_grow; |
| 2969 | } |
| 2970 | |
| 2971 | slabp = list_entry(entry, struct slab, list); |
| 2972 | check_slabp(cachep, slabp); |
| 2973 | check_spinlock_acquired(cachep); |
Pekka Enberg | 714b817 | 2007-05-06 14:49:03 -0700 | [diff] [blame] | 2974 | |
| 2975 | /* |
| 2976 | * The slab was either on partial or free list so |
| 2977 | * there must be at least one object available for |
| 2978 | * allocation. |
| 2979 | */ |
| 2980 | BUG_ON(slabp->inuse < 0 || slabp->inuse >= cachep->num); |
| 2981 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2982 | while (slabp->inuse < cachep->num && batchcount--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2983 | STATS_INC_ALLOCED(cachep); |
| 2984 | STATS_INC_ACTIVE(cachep); |
| 2985 | STATS_SET_HIGH(cachep); |
| 2986 | |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2987 | ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, |
Pekka Enberg | 1ca4cb2 | 2006-10-06 00:43:52 -0700 | [diff] [blame] | 2988 | node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2989 | } |
| 2990 | check_slabp(cachep, slabp); |
| 2991 | |
| 2992 | /* move slabp to correct slabp list: */ |
| 2993 | list_del(&slabp->list); |
| 2994 | if (slabp->free == BUFCTL_END) |
| 2995 | list_add(&slabp->list, &l3->slabs_full); |
| 2996 | else |
| 2997 | list_add(&slabp->list, &l3->slabs_partial); |
| 2998 | } |
| 2999 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3000 | must_grow: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3001 | l3->free_objects -= ac->avail; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3002 | alloc_done: |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3003 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3004 | |
| 3005 | if (unlikely(!ac->avail)) { |
| 3006 | int x; |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3007 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3008 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3009 | /* cache_grow can reenable interrupts, then ac could change. */ |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3010 | ac = cpu_cache_get(cachep); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3011 | if (!x && ac->avail == 0) /* no objects in sight? abort */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3012 | return NULL; |
| 3013 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3014 | if (!ac->avail) /* objects refilled by interrupt? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3015 | goto retry; |
| 3016 | } |
| 3017 | ac->touched = 1; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3018 | return ac->entry[--ac->avail]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3019 | } |
| 3020 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3021 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
| 3022 | gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3023 | { |
| 3024 | might_sleep_if(flags & __GFP_WAIT); |
| 3025 | #if DEBUG |
| 3026 | kmem_flagcheck(cachep, flags); |
| 3027 | #endif |
| 3028 | } |
| 3029 | |
| 3030 | #if DEBUG |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3031 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
| 3032 | gfp_t flags, void *objp, void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3033 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3034 | if (!objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3035 | return objp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3036 | if (cachep->flags & SLAB_POISON) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3037 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3038 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3039 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3040 | cachep->buffer_size / PAGE_SIZE, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3041 | else |
| 3042 | check_poison_obj(cachep, objp); |
| 3043 | #else |
| 3044 | check_poison_obj(cachep, objp); |
| 3045 | #endif |
| 3046 | poison_obj(cachep, objp, POISON_INUSE); |
| 3047 | } |
| 3048 | if (cachep->flags & SLAB_STORE_USER) |
| 3049 | *dbg_userword(cachep, objp) = caller; |
| 3050 | |
| 3051 | if (cachep->flags & SLAB_RED_ZONE) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3052 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
| 3053 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { |
| 3054 | slab_error(cachep, "double free, or memory outside" |
| 3055 | " object was overwritten"); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3056 | printk(KERN_ERR |
David Woodhouse | b46b8f1 | 2007-05-08 00:22:59 -0700 | [diff] [blame] | 3057 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3058 | objp, *dbg_redzone1(cachep, objp), |
| 3059 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3060 | } |
| 3061 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; |
| 3062 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; |
| 3063 | } |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 3064 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
| 3065 | { |
| 3066 | struct slab *slabp; |
| 3067 | unsigned objnr; |
| 3068 | |
Christoph Lameter | b49af68 | 2007-05-06 14:49:41 -0700 | [diff] [blame] | 3069 | slabp = page_get_slab(virt_to_head_page(objp)); |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 3070 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
| 3071 | slab_bufctl(slabp)[objnr] = BUFCTL_ACTIVE; |
| 3072 | } |
| 3073 | #endif |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3074 | objp += obj_offset(cachep); |
Christoph Lameter | 4f10493 | 2007-05-06 14:50:17 -0700 | [diff] [blame] | 3075 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 3076 | cachep->ctor(objp, cachep, 0); |
Kevin Hilman | a44b56d | 2006-12-06 20:32:11 -0800 | [diff] [blame] | 3077 | #if ARCH_SLAB_MINALIGN |
| 3078 | if ((u32)objp & (ARCH_SLAB_MINALIGN-1)) { |
| 3079 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
| 3080 | objp, ARCH_SLAB_MINALIGN); |
| 3081 | } |
| 3082 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3083 | return objp; |
| 3084 | } |
| 3085 | #else |
| 3086 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) |
| 3087 | #endif |
| 3088 | |
Akinobu Mita | 8a8b650 | 2006-12-08 02:39:44 -0800 | [diff] [blame] | 3089 | #ifdef CONFIG_FAILSLAB |
| 3090 | |
| 3091 | static struct failslab_attr { |
| 3092 | |
| 3093 | struct fault_attr attr; |
| 3094 | |
| 3095 | u32 ignore_gfp_wait; |
| 3096 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
| 3097 | struct dentry *ignore_gfp_wait_file; |
| 3098 | #endif |
| 3099 | |
| 3100 | } failslab = { |
| 3101 | .attr = FAULT_ATTR_INITIALIZER, |
Don Mullis | 6b1b60f | 2006-12-08 02:39:53 -0800 | [diff] [blame] | 3102 | .ignore_gfp_wait = 1, |
Akinobu Mita | 8a8b650 | 2006-12-08 02:39:44 -0800 | [diff] [blame] | 3103 | }; |
| 3104 | |
| 3105 | static int __init setup_failslab(char *str) |
| 3106 | { |
| 3107 | return setup_fault_attr(&failslab.attr, str); |
| 3108 | } |
| 3109 | __setup("failslab=", setup_failslab); |
| 3110 | |
| 3111 | static int should_failslab(struct kmem_cache *cachep, gfp_t flags) |
| 3112 | { |
| 3113 | if (cachep == &cache_cache) |
| 3114 | return 0; |
| 3115 | if (flags & __GFP_NOFAIL) |
| 3116 | return 0; |
| 3117 | if (failslab.ignore_gfp_wait && (flags & __GFP_WAIT)) |
| 3118 | return 0; |
| 3119 | |
| 3120 | return should_fail(&failslab.attr, obj_size(cachep)); |
| 3121 | } |
| 3122 | |
| 3123 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
| 3124 | |
| 3125 | static int __init failslab_debugfs(void) |
| 3126 | { |
| 3127 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
| 3128 | struct dentry *dir; |
| 3129 | int err; |
| 3130 | |
Akinobu Mita | 824ebef | 2007-05-06 14:49:58 -0700 | [diff] [blame] | 3131 | err = init_fault_attr_dentries(&failslab.attr, "failslab"); |
Akinobu Mita | 8a8b650 | 2006-12-08 02:39:44 -0800 | [diff] [blame] | 3132 | if (err) |
| 3133 | return err; |
| 3134 | dir = failslab.attr.dentries.dir; |
| 3135 | |
| 3136 | failslab.ignore_gfp_wait_file = |
| 3137 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
| 3138 | &failslab.ignore_gfp_wait); |
| 3139 | |
| 3140 | if (!failslab.ignore_gfp_wait_file) { |
| 3141 | err = -ENOMEM; |
| 3142 | debugfs_remove(failslab.ignore_gfp_wait_file); |
| 3143 | cleanup_fault_attr_dentries(&failslab.attr); |
| 3144 | } |
| 3145 | |
| 3146 | return err; |
| 3147 | } |
| 3148 | |
| 3149 | late_initcall(failslab_debugfs); |
| 3150 | |
| 3151 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ |
| 3152 | |
| 3153 | #else /* CONFIG_FAILSLAB */ |
| 3154 | |
| 3155 | static inline int should_failslab(struct kmem_cache *cachep, gfp_t flags) |
| 3156 | { |
| 3157 | return 0; |
| 3158 | } |
| 3159 | |
| 3160 | #endif /* CONFIG_FAILSLAB */ |
| 3161 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3162 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3163 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3164 | void *objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3165 | struct array_cache *ac; |
| 3166 | |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 3167 | check_irq_off(); |
Akinobu Mita | 8a8b650 | 2006-12-08 02:39:44 -0800 | [diff] [blame] | 3168 | |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3169 | ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3170 | if (likely(ac->avail)) { |
| 3171 | STATS_INC_ALLOCHIT(cachep); |
| 3172 | ac->touched = 1; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3173 | objp = ac->entry[--ac->avail]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3174 | } else { |
| 3175 | STATS_INC_ALLOCMISS(cachep); |
| 3176 | objp = cache_alloc_refill(cachep, flags); |
| 3177 | } |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 3178 | return objp; |
| 3179 | } |
| 3180 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3181 | #ifdef CONFIG_NUMA |
| 3182 | /* |
Paul Jackson | b245539 | 2006-03-24 03:16:12 -0800 | [diff] [blame] | 3183 | * Try allocating on another node if PF_SPREAD_SLAB|PF_MEMPOLICY. |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 3184 | * |
| 3185 | * If we are in_interrupt, then process context, including cpusets and |
| 3186 | * mempolicy, may not apply and should not be used for allocation policy. |
| 3187 | */ |
| 3188 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) |
| 3189 | { |
| 3190 | int nid_alloc, nid_here; |
| 3191 | |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3192 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 3193 | return NULL; |
| 3194 | nid_alloc = nid_here = numa_node_id(); |
| 3195 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
| 3196 | nid_alloc = cpuset_mem_spread_node(); |
| 3197 | else if (current->mempolicy) |
| 3198 | nid_alloc = slab_node(current->mempolicy); |
| 3199 | if (nid_alloc != nid_here) |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 3200 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 3201 | return NULL; |
| 3202 | } |
| 3203 | |
| 3204 | /* |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3205 | * Fallback function if there was no memory available and no objects on a |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3206 | * certain node and fall back is permitted. First we scan all the |
| 3207 | * available nodelists for available objects. If that fails then we |
| 3208 | * perform an allocation without specifying a node. This allows the page |
| 3209 | * allocator to do its reclaim / fallback magic. We then insert the |
| 3210 | * slab into the proper nodelist and then allocate from it. |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3211 | */ |
Pekka Enberg | 8c8cc2c | 2007-02-10 01:42:53 -0800 | [diff] [blame] | 3212 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3213 | { |
Pekka Enberg | 8c8cc2c | 2007-02-10 01:42:53 -0800 | [diff] [blame] | 3214 | struct zonelist *zonelist; |
| 3215 | gfp_t local_flags; |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3216 | struct zone **z; |
| 3217 | void *obj = NULL; |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3218 | int nid; |
Pekka Enberg | 8c8cc2c | 2007-02-10 01:42:53 -0800 | [diff] [blame] | 3219 | |
| 3220 | if (flags & __GFP_THISNODE) |
| 3221 | return NULL; |
| 3222 | |
| 3223 | zonelist = &NODE_DATA(slab_node(current->mempolicy)) |
| 3224 | ->node_zonelists[gfp_zone(flags)]; |
| 3225 | local_flags = (flags & GFP_LEVEL_MASK); |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3226 | |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3227 | retry: |
| 3228 | /* |
| 3229 | * Look through allowed nodes for objects available |
| 3230 | * from existing per node queues. |
| 3231 | */ |
Christoph Lameter | aedb0eb | 2006-10-21 10:24:16 -0700 | [diff] [blame] | 3232 | for (z = zonelist->zones; *z && !obj; z++) { |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3233 | nid = zone_to_nid(*z); |
Christoph Lameter | aedb0eb | 2006-10-21 10:24:16 -0700 | [diff] [blame] | 3234 | |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 3235 | if (cpuset_zone_allowed_hardwall(*z, flags) && |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3236 | cache->nodelists[nid] && |
| 3237 | cache->nodelists[nid]->free_objects) |
| 3238 | obj = ____cache_alloc_node(cache, |
| 3239 | flags | GFP_THISNODE, nid); |
| 3240 | } |
| 3241 | |
Christoph Lameter | cfce660 | 2007-05-06 14:50:17 -0700 | [diff] [blame] | 3242 | if (!obj) { |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3243 | /* |
| 3244 | * This allocation will be performed within the constraints |
| 3245 | * of the current cpuset / memory policy requirements. |
| 3246 | * We may trigger various forms of reclaim on the allowed |
| 3247 | * set and go into memory reserves if necessary. |
| 3248 | */ |
Christoph Lameter | dd47ea7 | 2006-12-13 00:34:11 -0800 | [diff] [blame] | 3249 | if (local_flags & __GFP_WAIT) |
| 3250 | local_irq_enable(); |
| 3251 | kmem_flagcheck(cache, flags); |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3252 | obj = kmem_getpages(cache, flags, -1); |
Christoph Lameter | dd47ea7 | 2006-12-13 00:34:11 -0800 | [diff] [blame] | 3253 | if (local_flags & __GFP_WAIT) |
| 3254 | local_irq_disable(); |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3255 | if (obj) { |
| 3256 | /* |
| 3257 | * Insert into the appropriate per node queues |
| 3258 | */ |
| 3259 | nid = page_to_nid(virt_to_page(obj)); |
| 3260 | if (cache_grow(cache, flags, nid, obj)) { |
| 3261 | obj = ____cache_alloc_node(cache, |
| 3262 | flags | GFP_THISNODE, nid); |
| 3263 | if (!obj) |
| 3264 | /* |
| 3265 | * Another processor may allocate the |
| 3266 | * objects in the slab since we are |
| 3267 | * not holding any locks. |
| 3268 | */ |
| 3269 | goto retry; |
| 3270 | } else { |
Hugh Dickins | b6a6045 | 2007-01-05 16:36:36 -0800 | [diff] [blame] | 3271 | /* cache_grow already freed obj */ |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3272 | obj = NULL; |
| 3273 | } |
| 3274 | } |
Christoph Lameter | aedb0eb | 2006-10-21 10:24:16 -0700 | [diff] [blame] | 3275 | } |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3276 | return obj; |
| 3277 | } |
| 3278 | |
| 3279 | /* |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3280 | * A interface to enable slab creation on nodeid |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3281 | */ |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 3282 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3283 | int nodeid) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3284 | { |
| 3285 | struct list_head *entry; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3286 | struct slab *slabp; |
| 3287 | struct kmem_list3 *l3; |
| 3288 | void *obj; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3289 | int x; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3290 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3291 | l3 = cachep->nodelists[nodeid]; |
| 3292 | BUG_ON(!l3); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3293 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3294 | retry: |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3295 | check_irq_off(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3296 | spin_lock(&l3->list_lock); |
| 3297 | entry = l3->slabs_partial.next; |
| 3298 | if (entry == &l3->slabs_partial) { |
| 3299 | l3->free_touched = 1; |
| 3300 | entry = l3->slabs_free.next; |
| 3301 | if (entry == &l3->slabs_free) |
| 3302 | goto must_grow; |
| 3303 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3304 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3305 | slabp = list_entry(entry, struct slab, list); |
| 3306 | check_spinlock_acquired_node(cachep, nodeid); |
| 3307 | check_slabp(cachep, slabp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3308 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3309 | STATS_INC_NODEALLOCS(cachep); |
| 3310 | STATS_INC_ACTIVE(cachep); |
| 3311 | STATS_SET_HIGH(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3312 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3313 | BUG_ON(slabp->inuse == cachep->num); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3314 | |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 3315 | obj = slab_get_obj(cachep, slabp, nodeid); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3316 | check_slabp(cachep, slabp); |
| 3317 | l3->free_objects--; |
| 3318 | /* move slabp to correct slabp list: */ |
| 3319 | list_del(&slabp->list); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3320 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3321 | if (slabp->free == BUFCTL_END) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3322 | list_add(&slabp->list, &l3->slabs_full); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3323 | else |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3324 | list_add(&slabp->list, &l3->slabs_partial); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3325 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3326 | spin_unlock(&l3->list_lock); |
| 3327 | goto done; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3328 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3329 | must_grow: |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3330 | spin_unlock(&l3->list_lock); |
Christoph Lameter | 3c517a6 | 2006-12-06 20:33:29 -0800 | [diff] [blame] | 3331 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3332 | if (x) |
| 3333 | goto retry; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3334 | |
Pekka Enberg | 8c8cc2c | 2007-02-10 01:42:53 -0800 | [diff] [blame] | 3335 | return fallback_alloc(cachep, flags); |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 3336 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3337 | done: |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3338 | return obj; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3339 | } |
Pekka Enberg | 8c8cc2c | 2007-02-10 01:42:53 -0800 | [diff] [blame] | 3340 | |
| 3341 | /** |
| 3342 | * kmem_cache_alloc_node - Allocate an object on the specified node |
| 3343 | * @cachep: The cache to allocate from. |
| 3344 | * @flags: See kmalloc(). |
| 3345 | * @nodeid: node number of the target node. |
| 3346 | * @caller: return address of caller, used for debug information |
| 3347 | * |
| 3348 | * Identical to kmem_cache_alloc but it will allocate memory on the given |
| 3349 | * node, which can improve the performance for cpu bound structures. |
| 3350 | * |
| 3351 | * Fallback to other node is possible if __GFP_THISNODE is not set. |
| 3352 | */ |
| 3353 | static __always_inline void * |
| 3354 | __cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
| 3355 | void *caller) |
| 3356 | { |
| 3357 | unsigned long save_flags; |
| 3358 | void *ptr; |
| 3359 | |
Akinobu Mita | 824ebef | 2007-05-06 14:49:58 -0700 | [diff] [blame] | 3360 | if (should_failslab(cachep, flags)) |
| 3361 | return NULL; |
| 3362 | |
Pekka Enberg | 8c8cc2c | 2007-02-10 01:42:53 -0800 | [diff] [blame] | 3363 | cache_alloc_debugcheck_before(cachep, flags); |
| 3364 | local_irq_save(save_flags); |
| 3365 | |
| 3366 | if (unlikely(nodeid == -1)) |
| 3367 | nodeid = numa_node_id(); |
| 3368 | |
| 3369 | if (unlikely(!cachep->nodelists[nodeid])) { |
| 3370 | /* Node not bootstrapped yet */ |
| 3371 | ptr = fallback_alloc(cachep, flags); |
| 3372 | goto out; |
| 3373 | } |
| 3374 | |
| 3375 | if (nodeid == numa_node_id()) { |
| 3376 | /* |
| 3377 | * Use the locally cached objects if possible. |
| 3378 | * However ____cache_alloc does not allow fallback |
| 3379 | * to other nodes. It may fail while we still have |
| 3380 | * objects on other nodes available. |
| 3381 | */ |
| 3382 | ptr = ____cache_alloc(cachep, flags); |
| 3383 | if (ptr) |
| 3384 | goto out; |
| 3385 | } |
| 3386 | /* ___cache_alloc_node can fall back to other nodes */ |
| 3387 | ptr = ____cache_alloc_node(cachep, flags, nodeid); |
| 3388 | out: |
| 3389 | local_irq_restore(save_flags); |
| 3390 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); |
| 3391 | |
| 3392 | return ptr; |
| 3393 | } |
| 3394 | |
| 3395 | static __always_inline void * |
| 3396 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) |
| 3397 | { |
| 3398 | void *objp; |
| 3399 | |
| 3400 | if (unlikely(current->flags & (PF_SPREAD_SLAB | PF_MEMPOLICY))) { |
| 3401 | objp = alternate_node_alloc(cache, flags); |
| 3402 | if (objp) |
| 3403 | goto out; |
| 3404 | } |
| 3405 | objp = ____cache_alloc(cache, flags); |
| 3406 | |
| 3407 | /* |
| 3408 | * We may just have run out of memory on the local node. |
| 3409 | * ____cache_alloc_node() knows how to locate memory on other nodes |
| 3410 | */ |
| 3411 | if (!objp) |
| 3412 | objp = ____cache_alloc_node(cache, flags, numa_node_id()); |
| 3413 | |
| 3414 | out: |
| 3415 | return objp; |
| 3416 | } |
| 3417 | #else |
| 3418 | |
| 3419 | static __always_inline void * |
| 3420 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
| 3421 | { |
| 3422 | return ____cache_alloc(cachep, flags); |
| 3423 | } |
| 3424 | |
| 3425 | #endif /* CONFIG_NUMA */ |
| 3426 | |
| 3427 | static __always_inline void * |
| 3428 | __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller) |
| 3429 | { |
| 3430 | unsigned long save_flags; |
| 3431 | void *objp; |
| 3432 | |
Akinobu Mita | 824ebef | 2007-05-06 14:49:58 -0700 | [diff] [blame] | 3433 | if (should_failslab(cachep, flags)) |
| 3434 | return NULL; |
| 3435 | |
Pekka Enberg | 8c8cc2c | 2007-02-10 01:42:53 -0800 | [diff] [blame] | 3436 | cache_alloc_debugcheck_before(cachep, flags); |
| 3437 | local_irq_save(save_flags); |
| 3438 | objp = __do_cache_alloc(cachep, flags); |
| 3439 | local_irq_restore(save_flags); |
| 3440 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); |
| 3441 | prefetchw(objp); |
| 3442 | |
| 3443 | return objp; |
| 3444 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3445 | |
| 3446 | /* |
| 3447 | * Caller needs to acquire correct kmem_list's list_lock |
| 3448 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3449 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3450 | int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3451 | { |
| 3452 | int i; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3453 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3454 | |
| 3455 | for (i = 0; i < nr_objects; i++) { |
| 3456 | void *objp = objpp[i]; |
| 3457 | struct slab *slabp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3458 | |
Pekka Enberg | 6ed5eb2 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3459 | slabp = virt_to_slab(objp); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3460 | l3 = cachep->nodelists[node]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3461 | list_del(&slabp->list); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3462 | check_spinlock_acquired_node(cachep, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3463 | check_slabp(cachep, slabp); |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 3464 | slab_put_obj(cachep, slabp, objp, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3465 | STATS_DEC_ACTIVE(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3466 | l3->free_objects++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3467 | check_slabp(cachep, slabp); |
| 3468 | |
| 3469 | /* fixup slab chains */ |
| 3470 | if (slabp->inuse == 0) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3471 | if (l3->free_objects > l3->free_limit) { |
| 3472 | l3->free_objects -= cachep->num; |
Ravikiran G Thirumalai | e5ac9c5 | 2006-09-25 23:31:34 -0700 | [diff] [blame] | 3473 | /* No need to drop any previously held |
| 3474 | * lock here, even if we have a off-slab slab |
| 3475 | * descriptor it is guaranteed to come from |
| 3476 | * a different cache, refer to comments before |
| 3477 | * alloc_slabmgmt. |
| 3478 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3479 | slab_destroy(cachep, slabp); |
| 3480 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3481 | list_add(&slabp->list, &l3->slabs_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3482 | } |
| 3483 | } else { |
| 3484 | /* Unconditionally move a slab to the end of the |
| 3485 | * partial list on free - maximum time for the |
| 3486 | * other objects to be freed, too. |
| 3487 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3488 | list_add_tail(&slabp->list, &l3->slabs_partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3489 | } |
| 3490 | } |
| 3491 | } |
| 3492 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3493 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3494 | { |
| 3495 | int batchcount; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3496 | struct kmem_list3 *l3; |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3497 | int node = numa_node_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3498 | |
| 3499 | batchcount = ac->batchcount; |
| 3500 | #if DEBUG |
| 3501 | BUG_ON(!batchcount || batchcount > ac->avail); |
| 3502 | #endif |
| 3503 | check_irq_off(); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3504 | l3 = cachep->nodelists[node]; |
Ingo Molnar | 873623d | 2006-07-13 14:44:38 +0200 | [diff] [blame] | 3505 | spin_lock(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3506 | if (l3->shared) { |
| 3507 | struct array_cache *shared_array = l3->shared; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3508 | int max = shared_array->limit - shared_array->avail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3509 | if (max) { |
| 3510 | if (batchcount > max) |
| 3511 | batchcount = max; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3512 | memcpy(&(shared_array->entry[shared_array->avail]), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3513 | ac->entry, sizeof(void *) * batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3514 | shared_array->avail += batchcount; |
| 3515 | goto free_done; |
| 3516 | } |
| 3517 | } |
| 3518 | |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3519 | free_block(cachep, ac->entry, batchcount, node); |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3520 | free_done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3521 | #if STATS |
| 3522 | { |
| 3523 | int i = 0; |
| 3524 | struct list_head *p; |
| 3525 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3526 | p = l3->slabs_free.next; |
| 3527 | while (p != &(l3->slabs_free)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3528 | struct slab *slabp; |
| 3529 | |
| 3530 | slabp = list_entry(p, struct slab, list); |
| 3531 | BUG_ON(slabp->inuse); |
| 3532 | |
| 3533 | i++; |
| 3534 | p = p->next; |
| 3535 | } |
| 3536 | STATS_SET_FREEABLE(cachep, i); |
| 3537 | } |
| 3538 | #endif |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3539 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3540 | ac->avail -= batchcount; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3541 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3542 | } |
| 3543 | |
| 3544 | /* |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3545 | * Release an obj back to its cache. If the obj has a constructed state, it must |
| 3546 | * be in this state _before_ it is released. Called with disabled ints. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3547 | */ |
Ingo Molnar | 873623d | 2006-07-13 14:44:38 +0200 | [diff] [blame] | 3548 | static inline void __cache_free(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3549 | { |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3550 | struct array_cache *ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3551 | |
| 3552 | check_irq_off(); |
| 3553 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); |
| 3554 | |
Christoph Lameter | 3cdc0ed | 2007-06-08 13:46:46 -0700 | [diff] [blame] | 3555 | if (cache_free_alien(cachep, objp)) |
Pekka Enberg | 729bd0b | 2006-06-23 02:03:05 -0700 | [diff] [blame] | 3556 | return; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3557 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3558 | if (likely(ac->avail < ac->limit)) { |
| 3559 | STATS_INC_FREEHIT(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3560 | ac->entry[ac->avail++] = objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3561 | return; |
| 3562 | } else { |
| 3563 | STATS_INC_FREEMISS(cachep); |
| 3564 | cache_flusharray(cachep, ac); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3565 | ac->entry[ac->avail++] = objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3566 | } |
| 3567 | } |
| 3568 | |
| 3569 | /** |
| 3570 | * kmem_cache_alloc - Allocate an object |
| 3571 | * @cachep: The cache to allocate from. |
| 3572 | * @flags: See kmalloc(). |
| 3573 | * |
| 3574 | * Allocate an object from this cache. The flags are only relevant |
| 3575 | * if the cache has no available objects. |
| 3576 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3577 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3578 | { |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3579 | return __cache_alloc(cachep, flags, __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3580 | } |
| 3581 | EXPORT_SYMBOL(kmem_cache_alloc); |
| 3582 | |
| 3583 | /** |
Rolf Eike Beer | b8008b2 | 2006-07-30 03:04:04 -0700 | [diff] [blame] | 3584 | * kmem_cache_zalloc - Allocate an object. The memory is set to zero. |
Pekka Enberg | a8c0f9a | 2006-03-25 03:06:42 -0800 | [diff] [blame] | 3585 | * @cache: The cache to allocate from. |
| 3586 | * @flags: See kmalloc(). |
| 3587 | * |
| 3588 | * Allocate an object from this cache and set the allocated memory to zero. |
| 3589 | * The flags are only relevant if the cache has no available objects. |
| 3590 | */ |
| 3591 | void *kmem_cache_zalloc(struct kmem_cache *cache, gfp_t flags) |
| 3592 | { |
| 3593 | void *ret = __cache_alloc(cache, flags, __builtin_return_address(0)); |
| 3594 | if (ret) |
| 3595 | memset(ret, 0, obj_size(cache)); |
| 3596 | return ret; |
| 3597 | } |
| 3598 | EXPORT_SYMBOL(kmem_cache_zalloc); |
| 3599 | |
| 3600 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3601 | * kmem_ptr_validate - check if an untrusted pointer might |
| 3602 | * be a slab entry. |
| 3603 | * @cachep: the cache we're checking against |
| 3604 | * @ptr: pointer to validate |
| 3605 | * |
| 3606 | * This verifies that the untrusted pointer looks sane: |
| 3607 | * it is _not_ a guarantee that the pointer is actually |
| 3608 | * part of the slab cache in question, but it at least |
| 3609 | * validates that the pointer can be dereferenced and |
| 3610 | * looks half-way sane. |
| 3611 | * |
| 3612 | * Currently only used for dentry validation. |
| 3613 | */ |
Christoph Lameter | b7f869a | 2006-12-22 01:06:44 -0800 | [diff] [blame] | 3614 | int kmem_ptr_validate(struct kmem_cache *cachep, const void *ptr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3615 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3616 | unsigned long addr = (unsigned long)ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3617 | unsigned long min_addr = PAGE_OFFSET; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3618 | unsigned long align_mask = BYTES_PER_WORD - 1; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3619 | unsigned long size = cachep->buffer_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3620 | struct page *page; |
| 3621 | |
| 3622 | if (unlikely(addr < min_addr)) |
| 3623 | goto out; |
| 3624 | if (unlikely(addr > (unsigned long)high_memory - size)) |
| 3625 | goto out; |
| 3626 | if (unlikely(addr & align_mask)) |
| 3627 | goto out; |
| 3628 | if (unlikely(!kern_addr_valid(addr))) |
| 3629 | goto out; |
| 3630 | if (unlikely(!kern_addr_valid(addr + size - 1))) |
| 3631 | goto out; |
| 3632 | page = virt_to_page(ptr); |
| 3633 | if (unlikely(!PageSlab(page))) |
| 3634 | goto out; |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 3635 | if (unlikely(page_get_cache(page) != cachep)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3636 | goto out; |
| 3637 | return 1; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3638 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3639 | return 0; |
| 3640 | } |
| 3641 | |
| 3642 | #ifdef CONFIG_NUMA |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 3643 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
| 3644 | { |
| 3645 | return __cache_alloc_node(cachep, flags, nodeid, |
| 3646 | __builtin_return_address(0)); |
| 3647 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3648 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
| 3649 | |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 3650 | static __always_inline void * |
| 3651 | __do_kmalloc_node(size_t size, gfp_t flags, int node, void *caller) |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3652 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3653 | struct kmem_cache *cachep; |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3654 | |
| 3655 | cachep = kmem_find_general_cachep(size, flags); |
| 3656 | if (unlikely(cachep == NULL)) |
| 3657 | return NULL; |
| 3658 | return kmem_cache_alloc_node(cachep, flags, node); |
| 3659 | } |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 3660 | |
| 3661 | #ifdef CONFIG_DEBUG_SLAB |
| 3662 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
| 3663 | { |
| 3664 | return __do_kmalloc_node(size, flags, node, |
| 3665 | __builtin_return_address(0)); |
| 3666 | } |
Christoph Hellwig | dbe5e69 | 2006-09-25 23:31:36 -0700 | [diff] [blame] | 3667 | EXPORT_SYMBOL(__kmalloc_node); |
Christoph Hellwig | 8b98c16 | 2006-12-06 20:32:30 -0800 | [diff] [blame] | 3668 | |
| 3669 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, |
| 3670 | int node, void *caller) |
| 3671 | { |
| 3672 | return __do_kmalloc_node(size, flags, node, caller); |
| 3673 | } |
| 3674 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
| 3675 | #else |
| 3676 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
| 3677 | { |
| 3678 | return __do_kmalloc_node(size, flags, node, NULL); |
| 3679 | } |
| 3680 | EXPORT_SYMBOL(__kmalloc_node); |
| 3681 | #endif /* CONFIG_DEBUG_SLAB */ |
| 3682 | #endif /* CONFIG_NUMA */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3683 | |
| 3684 | /** |
Paul Drynoff | 800590f | 2006-06-23 02:03:48 -0700 | [diff] [blame] | 3685 | * __do_kmalloc - allocate memory |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3686 | * @size: how many bytes of memory are required. |
Paul Drynoff | 800590f | 2006-06-23 02:03:48 -0700 | [diff] [blame] | 3687 | * @flags: the type of memory to allocate (see kmalloc). |
Randy Dunlap | 911851e | 2006-03-22 00:08:14 -0800 | [diff] [blame] | 3688 | * @caller: function caller for debug tracking of the caller |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3689 | */ |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3690 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
| 3691 | void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3692 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3693 | struct kmem_cache *cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3694 | |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3695 | /* If you want to save a few bytes .text space: replace |
| 3696 | * __ with kmem_. |
| 3697 | * Then kmalloc uses the uninlined functions instead of the inline |
| 3698 | * functions. |
| 3699 | */ |
| 3700 | cachep = __find_general_cachep(size, flags); |
Andrew Morton | dbdb904 | 2005-09-23 13:24:10 -0700 | [diff] [blame] | 3701 | if (unlikely(cachep == NULL)) |
| 3702 | return NULL; |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3703 | return __cache_alloc(cachep, flags, caller); |
| 3704 | } |
| 3705 | |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3706 | |
Christoph Hellwig | 1d2c8ee | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3707 | #ifdef CONFIG_DEBUG_SLAB |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3708 | void *__kmalloc(size_t size, gfp_t flags) |
| 3709 | { |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 3710 | return __do_kmalloc(size, flags, __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3711 | } |
| 3712 | EXPORT_SYMBOL(__kmalloc); |
| 3713 | |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3714 | void *__kmalloc_track_caller(size_t size, gfp_t flags, void *caller) |
| 3715 | { |
| 3716 | return __do_kmalloc(size, flags, caller); |
| 3717 | } |
| 3718 | EXPORT_SYMBOL(__kmalloc_track_caller); |
Christoph Hellwig | 1d2c8ee | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3719 | |
| 3720 | #else |
| 3721 | void *__kmalloc(size_t size, gfp_t flags) |
| 3722 | { |
| 3723 | return __do_kmalloc(size, flags, NULL); |
| 3724 | } |
| 3725 | EXPORT_SYMBOL(__kmalloc); |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3726 | #endif |
| 3727 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3728 | /** |
Pekka Enberg | fd76bab | 2007-05-06 14:48:40 -0700 | [diff] [blame] | 3729 | * krealloc - reallocate memory. The contents will remain unchanged. |
Pekka Enberg | fd76bab | 2007-05-06 14:48:40 -0700 | [diff] [blame] | 3730 | * @p: object to reallocate memory for. |
| 3731 | * @new_size: how many bytes of memory are required. |
| 3732 | * @flags: the type of memory to allocate. |
| 3733 | * |
| 3734 | * The contents of the object pointed to are preserved up to the |
| 3735 | * lesser of the new and old sizes. If @p is %NULL, krealloc() |
| 3736 | * behaves exactly like kmalloc(). If @size is 0 and @p is not a |
| 3737 | * %NULL pointer, the object pointed to is freed. |
| 3738 | */ |
| 3739 | void *krealloc(const void *p, size_t new_size, gfp_t flags) |
| 3740 | { |
| 3741 | struct kmem_cache *cache, *new_cache; |
| 3742 | void *ret; |
| 3743 | |
| 3744 | if (unlikely(!p)) |
| 3745 | return kmalloc_track_caller(new_size, flags); |
| 3746 | |
| 3747 | if (unlikely(!new_size)) { |
| 3748 | kfree(p); |
| 3749 | return NULL; |
| 3750 | } |
| 3751 | |
| 3752 | cache = virt_to_cache(p); |
| 3753 | new_cache = __find_general_cachep(new_size, flags); |
| 3754 | |
| 3755 | /* |
| 3756 | * If new size fits in the current cache, bail out. |
| 3757 | */ |
| 3758 | if (likely(cache == new_cache)) |
| 3759 | return (void *)p; |
| 3760 | |
| 3761 | /* |
| 3762 | * We are on the slow-path here so do not use __cache_alloc |
| 3763 | * because it bloats kernel text. |
| 3764 | */ |
| 3765 | ret = kmalloc_track_caller(new_size, flags); |
| 3766 | if (ret) { |
| 3767 | memcpy(ret, p, min(new_size, ksize(p))); |
| 3768 | kfree(p); |
| 3769 | } |
| 3770 | return ret; |
| 3771 | } |
| 3772 | EXPORT_SYMBOL(krealloc); |
| 3773 | |
| 3774 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3775 | * kmem_cache_free - Deallocate an object |
| 3776 | * @cachep: The cache the allocation was from. |
| 3777 | * @objp: The previously allocated object. |
| 3778 | * |
| 3779 | * Free an object which was previously allocated from this |
| 3780 | * cache. |
| 3781 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3782 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3783 | { |
| 3784 | unsigned long flags; |
| 3785 | |
Pekka Enberg | ddc2e81 | 2006-06-23 02:03:40 -0700 | [diff] [blame] | 3786 | BUG_ON(virt_to_cache(objp) != cachep); |
| 3787 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3788 | local_irq_save(flags); |
Ingo Molnar | 898552c | 2007-02-10 01:44:57 -0800 | [diff] [blame] | 3789 | debug_check_no_locks_freed(objp, obj_size(cachep)); |
Ingo Molnar | 873623d | 2006-07-13 14:44:38 +0200 | [diff] [blame] | 3790 | __cache_free(cachep, objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3791 | local_irq_restore(flags); |
| 3792 | } |
| 3793 | EXPORT_SYMBOL(kmem_cache_free); |
| 3794 | |
| 3795 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3796 | * kfree - free previously allocated memory |
| 3797 | * @objp: pointer returned by kmalloc. |
| 3798 | * |
Pekka Enberg | 80e93ef | 2005-09-09 13:10:16 -0700 | [diff] [blame] | 3799 | * If @objp is NULL, no operation is performed. |
| 3800 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3801 | * Don't free memory not originally allocated by kmalloc() |
| 3802 | * or you will run into trouble. |
| 3803 | */ |
| 3804 | void kfree(const void *objp) |
| 3805 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3806 | struct kmem_cache *c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3807 | unsigned long flags; |
| 3808 | |
| 3809 | if (unlikely(!objp)) |
| 3810 | return; |
| 3811 | local_irq_save(flags); |
| 3812 | kfree_debugcheck(objp); |
Pekka Enberg | 6ed5eb2 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3813 | c = virt_to_cache(objp); |
Ingo Molnar | f9b8404 | 2006-06-27 02:54:49 -0700 | [diff] [blame] | 3814 | debug_check_no_locks_freed(objp, obj_size(c)); |
Ingo Molnar | 873623d | 2006-07-13 14:44:38 +0200 | [diff] [blame] | 3815 | __cache_free(c, (void *)objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3816 | local_irq_restore(flags); |
| 3817 | } |
| 3818 | EXPORT_SYMBOL(kfree); |
| 3819 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3820 | unsigned int kmem_cache_size(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3821 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3822 | return obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3823 | } |
| 3824 | EXPORT_SYMBOL(kmem_cache_size); |
| 3825 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3826 | const char *kmem_cache_name(struct kmem_cache *cachep) |
Arnaldo Carvalho de Melo | 1944972 | 2005-06-18 22:46:19 -0700 | [diff] [blame] | 3827 | { |
| 3828 | return cachep->name; |
| 3829 | } |
| 3830 | EXPORT_SYMBOL_GPL(kmem_cache_name); |
| 3831 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3832 | /* |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3833 | * This initializes kmem_list3 or resizes varioius caches for all nodes. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3834 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3835 | static int alloc_kmemlist(struct kmem_cache *cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3836 | { |
| 3837 | int node; |
| 3838 | struct kmem_list3 *l3; |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3839 | struct array_cache *new_shared; |
Paul Menage | 3395ee0 | 2006-12-06 20:32:16 -0800 | [diff] [blame] | 3840 | struct array_cache **new_alien = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3841 | |
| 3842 | for_each_online_node(node) { |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3843 | |
Paul Menage | 3395ee0 | 2006-12-06 20:32:16 -0800 | [diff] [blame] | 3844 | if (use_alien_caches) { |
| 3845 | new_alien = alloc_alien_cache(node, cachep->limit); |
| 3846 | if (!new_alien) |
| 3847 | goto fail; |
| 3848 | } |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3849 | |
Eric Dumazet | 6310984 | 2007-05-06 14:49:28 -0700 | [diff] [blame] | 3850 | new_shared = NULL; |
| 3851 | if (cachep->shared) { |
| 3852 | new_shared = alloc_arraycache(node, |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3853 | cachep->shared*cachep->batchcount, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3854 | 0xbaadf00d); |
Eric Dumazet | 6310984 | 2007-05-06 14:49:28 -0700 | [diff] [blame] | 3855 | if (!new_shared) { |
| 3856 | free_alien_cache(new_alien); |
| 3857 | goto fail; |
| 3858 | } |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3859 | } |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3860 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3861 | l3 = cachep->nodelists[node]; |
| 3862 | if (l3) { |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3863 | struct array_cache *shared = l3->shared; |
| 3864 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3865 | spin_lock_irq(&l3->list_lock); |
| 3866 | |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3867 | if (shared) |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3868 | free_block(cachep, shared->entry, |
| 3869 | shared->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3870 | |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3871 | l3->shared = new_shared; |
| 3872 | if (!l3->alien) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3873 | l3->alien = new_alien; |
| 3874 | new_alien = NULL; |
| 3875 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3876 | l3->free_limit = (1 + nr_cpus_node(node)) * |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3877 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3878 | spin_unlock_irq(&l3->list_lock); |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3879 | kfree(shared); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3880 | free_alien_cache(new_alien); |
| 3881 | continue; |
| 3882 | } |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3883 | l3 = kmalloc_node(sizeof(struct kmem_list3), GFP_KERNEL, node); |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3884 | if (!l3) { |
| 3885 | free_alien_cache(new_alien); |
| 3886 | kfree(new_shared); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3887 | goto fail; |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3888 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3889 | |
| 3890 | kmem_list3_init(l3); |
| 3891 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3892 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3893 | l3->shared = new_shared; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3894 | l3->alien = new_alien; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3895 | l3->free_limit = (1 + nr_cpus_node(node)) * |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3896 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3897 | cachep->nodelists[node] = l3; |
| 3898 | } |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3899 | return 0; |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3900 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3901 | fail: |
Christoph Lameter | 0718dc2 | 2006-03-25 03:06:47 -0800 | [diff] [blame] | 3902 | if (!cachep->next.next) { |
| 3903 | /* Cache is not active yet. Roll back what we did */ |
| 3904 | node--; |
| 3905 | while (node >= 0) { |
| 3906 | if (cachep->nodelists[node]) { |
| 3907 | l3 = cachep->nodelists[node]; |
| 3908 | |
| 3909 | kfree(l3->shared); |
| 3910 | free_alien_cache(l3->alien); |
| 3911 | kfree(l3); |
| 3912 | cachep->nodelists[node] = NULL; |
| 3913 | } |
| 3914 | node--; |
| 3915 | } |
| 3916 | } |
Christoph Lameter | cafeb02 | 2006-03-25 03:06:46 -0800 | [diff] [blame] | 3917 | return -ENOMEM; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3918 | } |
| 3919 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3920 | struct ccupdate_struct { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3921 | struct kmem_cache *cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3922 | struct array_cache *new[NR_CPUS]; |
| 3923 | }; |
| 3924 | |
| 3925 | static void do_ccupdate_local(void *info) |
| 3926 | { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3927 | struct ccupdate_struct *new = info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3928 | struct array_cache *old; |
| 3929 | |
| 3930 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3931 | old = cpu_cache_get(new->cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3932 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3933 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
| 3934 | new->new[smp_processor_id()] = old; |
| 3935 | } |
| 3936 | |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 3937 | /* Always called with the cache_chain_mutex held */ |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3938 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
| 3939 | int batchcount, int shared) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3940 | { |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3941 | struct ccupdate_struct *new; |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 3942 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3943 | |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3944 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
| 3945 | if (!new) |
| 3946 | return -ENOMEM; |
| 3947 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3948 | for_each_online_cpu(i) { |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3949 | new->new[i] = alloc_arraycache(cpu_to_node(i), limit, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3950 | batchcount); |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3951 | if (!new->new[i]) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3952 | for (i--; i >= 0; i--) |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3953 | kfree(new->new[i]); |
| 3954 | kfree(new); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3955 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3956 | } |
| 3957 | } |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3958 | new->cachep = cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3959 | |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3960 | on_each_cpu(do_ccupdate_local, (void *)new, 1, 1); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3961 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3962 | check_irq_on(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3963 | cachep->batchcount = batchcount; |
| 3964 | cachep->limit = limit; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3965 | cachep->shared = shared; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3966 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3967 | for_each_online_cpu(i) { |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3968 | struct array_cache *ccold = new->new[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3969 | if (!ccold) |
| 3970 | continue; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3971 | spin_lock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3972 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_node(i)); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3973 | spin_unlock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3974 | kfree(ccold); |
| 3975 | } |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 3976 | kfree(new); |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 3977 | return alloc_kmemlist(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3978 | } |
| 3979 | |
Ravikiran G Thirumalai | b5d8ca7 | 2006-03-22 00:08:12 -0800 | [diff] [blame] | 3980 | /* Called with cache_chain_mutex held always */ |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 3981 | static int enable_cpucache(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3982 | { |
| 3983 | int err; |
| 3984 | int limit, shared; |
| 3985 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3986 | /* |
| 3987 | * The head array serves three purposes: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3988 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
| 3989 | * - reduce the number of spinlock operations. |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 3990 | * - reduce the number of linked list operations on the slab and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3991 | * bufctl chains: array operations are cheaper. |
| 3992 | * The numbers are guessed, we should auto-tune as described by |
| 3993 | * Bonwick. |
| 3994 | */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3995 | if (cachep->buffer_size > 131072) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3996 | limit = 1; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3997 | else if (cachep->buffer_size > PAGE_SIZE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3998 | limit = 8; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3999 | else if (cachep->buffer_size > 1024) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4000 | limit = 24; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 4001 | else if (cachep->buffer_size > 256) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4002 | limit = 54; |
| 4003 | else |
| 4004 | limit = 120; |
| 4005 | |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 4006 | /* |
| 4007 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4008 | * allocation behaviour: Most allocs on one cpu, most free operations |
| 4009 | * on another cpu. For these cases, an efficient object passing between |
| 4010 | * cpus is necessary. This is provided by a shared array. The array |
| 4011 | * replaces Bonwick's magazine layer. |
| 4012 | * On uniprocessor, it's functionally equivalent (but less efficient) |
| 4013 | * to a larger limit. Thus disabled by default. |
| 4014 | */ |
| 4015 | shared = 0; |
Eric Dumazet | 364fbb2 | 2007-05-06 14:49:27 -0700 | [diff] [blame] | 4016 | if (cachep->buffer_size <= PAGE_SIZE && num_possible_cpus() > 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4017 | shared = 8; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4018 | |
| 4019 | #if DEBUG |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 4020 | /* |
| 4021 | * With debugging enabled, large batchcount lead to excessively long |
| 4022 | * periods with disabled local interrupts. Limit the batchcount |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4023 | */ |
| 4024 | if (limit > 32) |
| 4025 | limit = 32; |
| 4026 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4027 | err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4028 | if (err) |
| 4029 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4030 | cachep->name, -err); |
Christoph Lameter | 2ed3a4e | 2006-09-25 23:31:38 -0700 | [diff] [blame] | 4031 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4032 | } |
| 4033 | |
Christoph Lameter | 1b55253 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4034 | /* |
| 4035 | * Drain an array if it contains any elements taking the l3 lock only if |
Christoph Lameter | b18e7e6 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4036 | * necessary. Note that the l3 listlock also protects the array_cache |
| 4037 | * if drain_array() is used on the shared array. |
Christoph Lameter | 1b55253 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4038 | */ |
| 4039 | void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, |
| 4040 | struct array_cache *ac, int force, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4041 | { |
| 4042 | int tofree; |
| 4043 | |
Christoph Lameter | 1b55253 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4044 | if (!ac || !ac->avail) |
| 4045 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4046 | if (ac->touched && !force) { |
| 4047 | ac->touched = 0; |
Christoph Lameter | b18e7e6 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4048 | } else { |
Christoph Lameter | 1b55253 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4049 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | b18e7e6 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4050 | if (ac->avail) { |
| 4051 | tofree = force ? ac->avail : (ac->limit + 4) / 5; |
| 4052 | if (tofree > ac->avail) |
| 4053 | tofree = (ac->avail + 1) / 2; |
| 4054 | free_block(cachep, ac->entry, tofree, node); |
| 4055 | ac->avail -= tofree; |
| 4056 | memmove(ac->entry, &(ac->entry[tofree]), |
| 4057 | sizeof(void *) * ac->avail); |
| 4058 | } |
Christoph Lameter | 1b55253 | 2006-03-22 00:09:07 -0800 | [diff] [blame] | 4059 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4060 | } |
| 4061 | } |
| 4062 | |
| 4063 | /** |
| 4064 | * cache_reap - Reclaim memory from caches. |
Randy Dunlap | 05fb6bf | 2007-02-28 20:12:13 -0800 | [diff] [blame] | 4065 | * @w: work descriptor |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4066 | * |
| 4067 | * Called from workqueue/eventd every few seconds. |
| 4068 | * Purpose: |
| 4069 | * - clear the per-cpu caches for this CPU. |
| 4070 | * - return freeable pages to the main free memory pool. |
| 4071 | * |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 4072 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
| 4073 | * again on the next iteration. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4074 | */ |
Christoph Lameter | 7c5cae3 | 2007-02-10 01:42:55 -0800 | [diff] [blame] | 4075 | static void cache_reap(struct work_struct *w) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4076 | { |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4077 | struct kmem_cache *searchp; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4078 | struct kmem_list3 *l3; |
Christoph Lameter | aab2207 | 2006-03-22 00:09:06 -0800 | [diff] [blame] | 4079 | int node = numa_node_id(); |
Christoph Lameter | 7c5cae3 | 2007-02-10 01:42:55 -0800 | [diff] [blame] | 4080 | struct delayed_work *work = |
| 4081 | container_of(w, struct delayed_work, work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4082 | |
Christoph Lameter | 7c5cae3 | 2007-02-10 01:42:55 -0800 | [diff] [blame] | 4083 | if (!mutex_trylock(&cache_chain_mutex)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4084 | /* Give up. Setup the next iteration. */ |
Christoph Lameter | 7c5cae3 | 2007-02-10 01:42:55 -0800 | [diff] [blame] | 4085 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4086 | |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4087 | list_for_each_entry(searchp, &cache_chain, next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4088 | check_irq_on(); |
| 4089 | |
Christoph Lameter | 35386e3 | 2006-03-22 00:09:05 -0800 | [diff] [blame] | 4090 | /* |
| 4091 | * We only take the l3 lock if absolutely necessary and we |
| 4092 | * have established with reasonable certainty that |
| 4093 | * we can do some work if the lock was obtained. |
| 4094 | */ |
Christoph Lameter | aab2207 | 2006-03-22 00:09:06 -0800 | [diff] [blame] | 4095 | l3 = searchp->nodelists[node]; |
Christoph Lameter | 35386e3 | 2006-03-22 00:09:05 -0800 | [diff] [blame] | 4096 | |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 4097 | reap_alien(searchp, l3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4098 | |
Christoph Lameter | aab2207 | 2006-03-22 00:09:06 -0800 | [diff] [blame] | 4099 | drain_array(searchp, l3, cpu_cache_get(searchp), 0, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4100 | |
Christoph Lameter | 35386e3 | 2006-03-22 00:09:05 -0800 | [diff] [blame] | 4101 | /* |
| 4102 | * These are racy checks but it does not matter |
| 4103 | * if we skip one check or scan twice. |
| 4104 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4105 | if (time_after(l3->next_reap, jiffies)) |
Christoph Lameter | 35386e3 | 2006-03-22 00:09:05 -0800 | [diff] [blame] | 4106 | goto next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4107 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4108 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4109 | |
Christoph Lameter | aab2207 | 2006-03-22 00:09:06 -0800 | [diff] [blame] | 4110 | drain_array(searchp, l3, l3->shared, 0, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4111 | |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 4112 | if (l3->free_touched) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4113 | l3->free_touched = 0; |
Christoph Lameter | ed11d9e | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 4114 | else { |
| 4115 | int freed; |
| 4116 | |
| 4117 | freed = drain_freelist(searchp, l3, (l3->free_limit + |
| 4118 | 5 * searchp->num - 1) / (5 * searchp->num)); |
| 4119 | STATS_ADD_REAPED(searchp, freed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4120 | } |
Christoph Lameter | 35386e3 | 2006-03-22 00:09:05 -0800 | [diff] [blame] | 4121 | next: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4122 | cond_resched(); |
| 4123 | } |
| 4124 | check_irq_on(); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 4125 | mutex_unlock(&cache_chain_mutex); |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 4126 | next_reap_node(); |
Christoph Lameter | 7c5cae3 | 2007-02-10 01:42:55 -0800 | [diff] [blame] | 4127 | out: |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 4128 | /* Set up the next iteration */ |
Christoph Lameter | 7c5cae3 | 2007-02-10 01:42:55 -0800 | [diff] [blame] | 4129 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_CPUC)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4130 | } |
| 4131 | |
| 4132 | #ifdef CONFIG_PROC_FS |
| 4133 | |
Pekka Enberg | 85289f9 | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 4134 | static void print_slabinfo_header(struct seq_file *m) |
| 4135 | { |
| 4136 | /* |
| 4137 | * Output format version, so at least we can change it |
| 4138 | * without _too_ many complaints. |
| 4139 | */ |
| 4140 | #if STATS |
| 4141 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
| 4142 | #else |
| 4143 | seq_puts(m, "slabinfo - version: 2.1\n"); |
| 4144 | #endif |
| 4145 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
| 4146 | "<objperslab> <pagesperslab>"); |
| 4147 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
| 4148 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); |
| 4149 | #if STATS |
| 4150 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
Ravikiran G Thirumalai | fb7faf3 | 2006-04-10 22:52:54 -0700 | [diff] [blame] | 4151 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
Pekka Enberg | 85289f9 | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 4152 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
| 4153 | #endif |
| 4154 | seq_putc(m, '\n'); |
| 4155 | } |
| 4156 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4157 | static void *s_start(struct seq_file *m, loff_t *pos) |
| 4158 | { |
| 4159 | loff_t n = *pos; |
| 4160 | struct list_head *p; |
| 4161 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 4162 | mutex_lock(&cache_chain_mutex); |
Pekka Enberg | 85289f9 | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 4163 | if (!n) |
| 4164 | print_slabinfo_header(m); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4165 | p = cache_chain.next; |
| 4166 | while (n--) { |
| 4167 | p = p->next; |
| 4168 | if (p == &cache_chain) |
| 4169 | return NULL; |
| 4170 | } |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 4171 | return list_entry(p, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4172 | } |
| 4173 | |
| 4174 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) |
| 4175 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 4176 | struct kmem_cache *cachep = p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4177 | ++*pos; |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 4178 | return cachep->next.next == &cache_chain ? |
| 4179 | NULL : list_entry(cachep->next.next, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4180 | } |
| 4181 | |
| 4182 | static void s_stop(struct seq_file *m, void *p) |
| 4183 | { |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 4184 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4185 | } |
| 4186 | |
| 4187 | static int s_show(struct seq_file *m, void *p) |
| 4188 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 4189 | struct kmem_cache *cachep = p; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4190 | struct slab *slabp; |
| 4191 | unsigned long active_objs; |
| 4192 | unsigned long num_objs; |
| 4193 | unsigned long active_slabs = 0; |
| 4194 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4195 | const char *name; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4196 | char *error = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4197 | int node; |
| 4198 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4199 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4200 | active_objs = 0; |
| 4201 | num_slabs = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4202 | for_each_online_node(node) { |
| 4203 | l3 = cachep->nodelists[node]; |
| 4204 | if (!l3) |
| 4205 | continue; |
| 4206 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 4207 | check_irq_on(); |
| 4208 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4209 | |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4210 | list_for_each_entry(slabp, &l3->slabs_full, list) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4211 | if (slabp->inuse != cachep->num && !error) |
| 4212 | error = "slabs_full accounting error"; |
| 4213 | active_objs += cachep->num; |
| 4214 | active_slabs++; |
| 4215 | } |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4216 | list_for_each_entry(slabp, &l3->slabs_partial, list) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4217 | if (slabp->inuse == cachep->num && !error) |
| 4218 | error = "slabs_partial inuse accounting error"; |
| 4219 | if (!slabp->inuse && !error) |
| 4220 | error = "slabs_partial/inuse accounting error"; |
| 4221 | active_objs += slabp->inuse; |
| 4222 | active_slabs++; |
| 4223 | } |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4224 | list_for_each_entry(slabp, &l3->slabs_free, list) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4225 | if (slabp->inuse && !error) |
| 4226 | error = "slabs_free/inuse accounting error"; |
| 4227 | num_slabs++; |
| 4228 | } |
| 4229 | free_objects += l3->free_objects; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 4230 | if (l3->shared) |
| 4231 | shared_avail += l3->shared->avail; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4232 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 4233 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4234 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4235 | num_slabs += active_slabs; |
| 4236 | num_objs = num_slabs * cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4237 | if (num_objs - active_objs != free_objects && !error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4238 | error = "free_objects accounting error"; |
| 4239 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4240 | name = cachep->name; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4241 | if (error) |
| 4242 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); |
| 4243 | |
| 4244 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 4245 | name, active_objs, num_objs, cachep->buffer_size, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4246 | cachep->num, (1 << cachep->gfporder)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4247 | seq_printf(m, " : tunables %4u %4u %4u", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4248 | cachep->limit, cachep->batchcount, cachep->shared); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4249 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4250 | active_slabs, num_slabs, shared_avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4251 | #if STATS |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4252 | { /* list3 stats */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4253 | unsigned long high = cachep->high_mark; |
| 4254 | unsigned long allocs = cachep->num_allocations; |
| 4255 | unsigned long grown = cachep->grown; |
| 4256 | unsigned long reaped = cachep->reaped; |
| 4257 | unsigned long errors = cachep->errors; |
| 4258 | unsigned long max_freeable = cachep->max_freeable; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4259 | unsigned long node_allocs = cachep->node_allocs; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4260 | unsigned long node_frees = cachep->node_frees; |
Ravikiran G Thirumalai | fb7faf3 | 2006-04-10 22:52:54 -0700 | [diff] [blame] | 4261 | unsigned long overflows = cachep->node_overflow; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4262 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4263 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu \ |
Ravikiran G Thirumalai | fb7faf3 | 2006-04-10 22:52:54 -0700 | [diff] [blame] | 4264 | %4lu %4lu %4lu %4lu %4lu", allocs, high, grown, |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 4265 | reaped, errors, max_freeable, node_allocs, |
Ravikiran G Thirumalai | fb7faf3 | 2006-04-10 22:52:54 -0700 | [diff] [blame] | 4266 | node_frees, overflows); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4267 | } |
| 4268 | /* cpu stats */ |
| 4269 | { |
| 4270 | unsigned long allochit = atomic_read(&cachep->allochit); |
| 4271 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); |
| 4272 | unsigned long freehit = atomic_read(&cachep->freehit); |
| 4273 | unsigned long freemiss = atomic_read(&cachep->freemiss); |
| 4274 | |
| 4275 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4276 | allochit, allocmiss, freehit, freemiss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4277 | } |
| 4278 | #endif |
| 4279 | seq_putc(m, '\n'); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4280 | return 0; |
| 4281 | } |
| 4282 | |
| 4283 | /* |
| 4284 | * slabinfo_op - iterator that generates /proc/slabinfo |
| 4285 | * |
| 4286 | * Output layout: |
| 4287 | * cache-name |
| 4288 | * num-active-objs |
| 4289 | * total-objs |
| 4290 | * object size |
| 4291 | * num-active-slabs |
| 4292 | * total-slabs |
| 4293 | * num-pages-per-slab |
| 4294 | * + further values on SMP and with statistics enabled |
| 4295 | */ |
| 4296 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 4297 | const struct seq_operations slabinfo_op = { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4298 | .start = s_start, |
| 4299 | .next = s_next, |
| 4300 | .stop = s_stop, |
| 4301 | .show = s_show, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4302 | }; |
| 4303 | |
| 4304 | #define MAX_SLABINFO_WRITE 128 |
| 4305 | /** |
| 4306 | * slabinfo_write - Tuning for the slab allocator |
| 4307 | * @file: unused |
| 4308 | * @buffer: user buffer |
| 4309 | * @count: data length |
| 4310 | * @ppos: unused |
| 4311 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4312 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, |
| 4313 | size_t count, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4314 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4315 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4316 | int limit, batchcount, shared, res; |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4317 | struct kmem_cache *cachep; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4318 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4319 | if (count > MAX_SLABINFO_WRITE) |
| 4320 | return -EINVAL; |
| 4321 | if (copy_from_user(&kbuf, buffer, count)) |
| 4322 | return -EFAULT; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4323 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4324 | |
| 4325 | tmp = strchr(kbuf, ' '); |
| 4326 | if (!tmp) |
| 4327 | return -EINVAL; |
| 4328 | *tmp = '\0'; |
| 4329 | tmp++; |
| 4330 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) |
| 4331 | return -EINVAL; |
| 4332 | |
| 4333 | /* Find the cache in the chain of caches. */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 4334 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4335 | res = -EINVAL; |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4336 | list_for_each_entry(cachep, &cache_chain, next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4337 | if (!strcmp(cachep->name, kbuf)) { |
Andrew Morton | a737b3e | 2006-03-22 00:08:11 -0800 | [diff] [blame] | 4338 | if (limit < 1 || batchcount < 1 || |
| 4339 | batchcount > limit || shared < 0) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4340 | res = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4341 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 4342 | res = do_tune_cpucache(cachep, limit, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 4343 | batchcount, shared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4344 | } |
| 4345 | break; |
| 4346 | } |
| 4347 | } |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 4348 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4349 | if (res >= 0) |
| 4350 | res = count; |
| 4351 | return res; |
| 4352 | } |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4353 | |
| 4354 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
| 4355 | |
| 4356 | static void *leaks_start(struct seq_file *m, loff_t *pos) |
| 4357 | { |
| 4358 | loff_t n = *pos; |
| 4359 | struct list_head *p; |
| 4360 | |
| 4361 | mutex_lock(&cache_chain_mutex); |
| 4362 | p = cache_chain.next; |
| 4363 | while (n--) { |
| 4364 | p = p->next; |
| 4365 | if (p == &cache_chain) |
| 4366 | return NULL; |
| 4367 | } |
| 4368 | return list_entry(p, struct kmem_cache, next); |
| 4369 | } |
| 4370 | |
| 4371 | static inline int add_caller(unsigned long *n, unsigned long v) |
| 4372 | { |
| 4373 | unsigned long *p; |
| 4374 | int l; |
| 4375 | if (!v) |
| 4376 | return 1; |
| 4377 | l = n[1]; |
| 4378 | p = n + 2; |
| 4379 | while (l) { |
| 4380 | int i = l/2; |
| 4381 | unsigned long *q = p + 2 * i; |
| 4382 | if (*q == v) { |
| 4383 | q[1]++; |
| 4384 | return 1; |
| 4385 | } |
| 4386 | if (*q > v) { |
| 4387 | l = i; |
| 4388 | } else { |
| 4389 | p = q + 2; |
| 4390 | l -= i + 1; |
| 4391 | } |
| 4392 | } |
| 4393 | if (++n[1] == n[0]) |
| 4394 | return 0; |
| 4395 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); |
| 4396 | p[0] = v; |
| 4397 | p[1] = 1; |
| 4398 | return 1; |
| 4399 | } |
| 4400 | |
| 4401 | static void handle_slab(unsigned long *n, struct kmem_cache *c, struct slab *s) |
| 4402 | { |
| 4403 | void *p; |
| 4404 | int i; |
| 4405 | if (n[0] == n[1]) |
| 4406 | return; |
| 4407 | for (i = 0, p = s->s_mem; i < c->num; i++, p += c->buffer_size) { |
| 4408 | if (slab_bufctl(s)[i] != BUFCTL_ACTIVE) |
| 4409 | continue; |
| 4410 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
| 4411 | return; |
| 4412 | } |
| 4413 | } |
| 4414 | |
| 4415 | static void show_symbol(struct seq_file *m, unsigned long address) |
| 4416 | { |
| 4417 | #ifdef CONFIG_KALLSYMS |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4418 | unsigned long offset, size; |
Alexey Dobriyan | a5c43da | 2007-05-08 00:28:47 -0700 | [diff] [blame] | 4419 | char modname[MODULE_NAME_LEN + 1], name[KSYM_NAME_LEN + 1]; |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4420 | |
Alexey Dobriyan | a5c43da | 2007-05-08 00:28:47 -0700 | [diff] [blame] | 4421 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4422 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
Alexey Dobriyan | a5c43da | 2007-05-08 00:28:47 -0700 | [diff] [blame] | 4423 | if (modname[0]) |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4424 | seq_printf(m, " [%s]", modname); |
| 4425 | return; |
| 4426 | } |
| 4427 | #endif |
| 4428 | seq_printf(m, "%p", (void *)address); |
| 4429 | } |
| 4430 | |
| 4431 | static int leaks_show(struct seq_file *m, void *p) |
| 4432 | { |
| 4433 | struct kmem_cache *cachep = p; |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4434 | struct slab *slabp; |
| 4435 | struct kmem_list3 *l3; |
| 4436 | const char *name; |
| 4437 | unsigned long *n = m->private; |
| 4438 | int node; |
| 4439 | int i; |
| 4440 | |
| 4441 | if (!(cachep->flags & SLAB_STORE_USER)) |
| 4442 | return 0; |
| 4443 | if (!(cachep->flags & SLAB_RED_ZONE)) |
| 4444 | return 0; |
| 4445 | |
| 4446 | /* OK, we can do it */ |
| 4447 | |
| 4448 | n[1] = 0; |
| 4449 | |
| 4450 | for_each_online_node(node) { |
| 4451 | l3 = cachep->nodelists[node]; |
| 4452 | if (!l3) |
| 4453 | continue; |
| 4454 | |
| 4455 | check_irq_on(); |
| 4456 | spin_lock_irq(&l3->list_lock); |
| 4457 | |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4458 | list_for_each_entry(slabp, &l3->slabs_full, list) |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4459 | handle_slab(n, cachep, slabp); |
Christoph Hellwig | 7a7c381 | 2006-06-23 02:03:17 -0700 | [diff] [blame] | 4460 | list_for_each_entry(slabp, &l3->slabs_partial, list) |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4461 | handle_slab(n, cachep, slabp); |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4462 | spin_unlock_irq(&l3->list_lock); |
| 4463 | } |
| 4464 | name = cachep->name; |
| 4465 | if (n[0] == n[1]) { |
| 4466 | /* Increase the buffer size */ |
| 4467 | mutex_unlock(&cache_chain_mutex); |
| 4468 | m->private = kzalloc(n[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
| 4469 | if (!m->private) { |
| 4470 | /* Too bad, we are really out */ |
| 4471 | m->private = n; |
| 4472 | mutex_lock(&cache_chain_mutex); |
| 4473 | return -ENOMEM; |
| 4474 | } |
| 4475 | *(unsigned long *)m->private = n[0] * 2; |
| 4476 | kfree(n); |
| 4477 | mutex_lock(&cache_chain_mutex); |
| 4478 | /* Now make sure this entry will be retried */ |
| 4479 | m->count = m->size; |
| 4480 | return 0; |
| 4481 | } |
| 4482 | for (i = 0; i < n[1]; i++) { |
| 4483 | seq_printf(m, "%s: %lu ", name, n[2*i+3]); |
| 4484 | show_symbol(m, n[2*i+2]); |
| 4485 | seq_putc(m, '\n'); |
| 4486 | } |
Siddha, Suresh B | d2e7b7d | 2006-09-25 23:31:47 -0700 | [diff] [blame] | 4487 | |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4488 | return 0; |
| 4489 | } |
| 4490 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 4491 | const struct seq_operations slabstats_op = { |
Al Viro | 871751e | 2006-03-25 03:06:39 -0800 | [diff] [blame] | 4492 | .start = leaks_start, |
| 4493 | .next = s_next, |
| 4494 | .stop = s_stop, |
| 4495 | .show = leaks_show, |
| 4496 | }; |
| 4497 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4498 | #endif |
| 4499 | |
Manfred Spraul | 00e145b | 2005-09-03 15:55:07 -0700 | [diff] [blame] | 4500 | /** |
| 4501 | * ksize - get the actual amount of memory allocated for a given object |
| 4502 | * @objp: Pointer to the object |
| 4503 | * |
| 4504 | * kmalloc may internally round up allocations and return more memory |
| 4505 | * than requested. ksize() can be used to determine the actual amount of |
| 4506 | * memory allocated. The caller may use this additional memory, even though |
| 4507 | * a smaller amount of memory was initially specified with the kmalloc call. |
| 4508 | * The caller must guarantee that objp points to a valid object previously |
| 4509 | * allocated with either kmalloc() or kmem_cache_alloc(). The object |
| 4510 | * must not be freed during the duration of the call. |
| 4511 | */ |
Pekka Enberg | fd76bab | 2007-05-06 14:48:40 -0700 | [diff] [blame] | 4512 | size_t ksize(const void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4513 | { |
Manfred Spraul | 00e145b | 2005-09-03 15:55:07 -0700 | [diff] [blame] | 4514 | if (unlikely(objp == NULL)) |
| 4515 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4516 | |
Pekka Enberg | 6ed5eb2 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 4517 | return obj_size(virt_to_cache(objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4518 | } |