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 | * |
| 53 | * The c_cpuarray may not be read with enabled local interrupts - |
| 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 | |
| 89 | #include <linux/config.h> |
| 90 | #include <linux/slab.h> |
| 91 | #include <linux/mm.h> |
| 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> |
| 97 | #include <linux/seq_file.h> |
| 98 | #include <linux/notifier.h> |
| 99 | #include <linux/kallsyms.h> |
| 100 | #include <linux/cpu.h> |
| 101 | #include <linux/sysctl.h> |
| 102 | #include <linux/module.h> |
| 103 | #include <linux/rcupdate.h> |
Paulo Marques | 543537b | 2005-06-23 00:09:02 -0700 | [diff] [blame] | 104 | #include <linux/string.h> |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 105 | #include <linux/nodemask.h> |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 106 | #include <linux/mempolicy.h> |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 107 | #include <linux/mutex.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | |
| 109 | #include <asm/uaccess.h> |
| 110 | #include <asm/cacheflush.h> |
| 111 | #include <asm/tlbflush.h> |
| 112 | #include <asm/page.h> |
| 113 | |
| 114 | /* |
| 115 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_DEBUG_INITIAL, |
| 116 | * SLAB_RED_ZONE & SLAB_POISON. |
| 117 | * 0 for faster, smaller code (especially in the critical paths). |
| 118 | * |
| 119 | * STATS - 1 to collect stats for /proc/slabinfo. |
| 120 | * 0 for faster, smaller code (especially in the critical paths). |
| 121 | * |
| 122 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) |
| 123 | */ |
| 124 | |
| 125 | #ifdef CONFIG_DEBUG_SLAB |
| 126 | #define DEBUG 1 |
| 127 | #define STATS 1 |
| 128 | #define FORCED_DEBUG 1 |
| 129 | #else |
| 130 | #define DEBUG 0 |
| 131 | #define STATS 0 |
| 132 | #define FORCED_DEBUG 0 |
| 133 | #endif |
| 134 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | /* Shouldn't this be in a header file somewhere? */ |
| 136 | #define BYTES_PER_WORD sizeof(void *) |
| 137 | |
| 138 | #ifndef cache_line_size |
| 139 | #define cache_line_size() L1_CACHE_BYTES |
| 140 | #endif |
| 141 | |
| 142 | #ifndef ARCH_KMALLOC_MINALIGN |
| 143 | /* |
| 144 | * Enforce a minimum alignment for the kmalloc caches. |
| 145 | * Usually, the kmalloc caches are cache_line_size() aligned, except when |
| 146 | * DEBUG and FORCED_DEBUG are enabled, then they are BYTES_PER_WORD aligned. |
| 147 | * Some archs want to perform DMA into kmalloc caches and need a guaranteed |
| 148 | * alignment larger than BYTES_PER_WORD. ARCH_KMALLOC_MINALIGN allows that. |
| 149 | * Note that this flag disables some debug features. |
| 150 | */ |
| 151 | #define ARCH_KMALLOC_MINALIGN 0 |
| 152 | #endif |
| 153 | |
| 154 | #ifndef ARCH_SLAB_MINALIGN |
| 155 | /* |
| 156 | * Enforce a minimum alignment for all caches. |
| 157 | * Intended for archs that get misalignment faults even for BYTES_PER_WORD |
| 158 | * aligned buffers. Includes ARCH_KMALLOC_MINALIGN. |
| 159 | * If possible: Do not enable this flag for CONFIG_DEBUG_SLAB, it disables |
| 160 | * some debug features. |
| 161 | */ |
| 162 | #define ARCH_SLAB_MINALIGN 0 |
| 163 | #endif |
| 164 | |
| 165 | #ifndef ARCH_KMALLOC_FLAGS |
| 166 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN |
| 167 | #endif |
| 168 | |
| 169 | /* Legal flag mask for kmem_cache_create(). */ |
| 170 | #if DEBUG |
| 171 | # define CREATE_MASK (SLAB_DEBUG_INITIAL | SLAB_RED_ZONE | \ |
| 172 | SLAB_POISON | SLAB_HWCACHE_ALIGN | \ |
| 173 | SLAB_NO_REAP | SLAB_CACHE_DMA | \ |
| 174 | SLAB_MUST_HWCACHE_ALIGN | SLAB_STORE_USER | \ |
| 175 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
| 176 | SLAB_DESTROY_BY_RCU) |
| 177 | #else |
| 178 | # define CREATE_MASK (SLAB_HWCACHE_ALIGN | SLAB_NO_REAP | \ |
| 179 | SLAB_CACHE_DMA | SLAB_MUST_HWCACHE_ALIGN | \ |
| 180 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
| 181 | SLAB_DESTROY_BY_RCU) |
| 182 | #endif |
| 183 | |
| 184 | /* |
| 185 | * kmem_bufctl_t: |
| 186 | * |
| 187 | * Bufctl's are used for linking objs within a slab |
| 188 | * linked offsets. |
| 189 | * |
| 190 | * This implementation relies on "struct page" for locating the cache & |
| 191 | * slab an object belongs to. |
| 192 | * This allows the bufctl structure to be small (one int), but limits |
| 193 | * the number of objects a slab (not a cache) can contain when off-slab |
| 194 | * bufctls are used. The limit is the size of the largest general cache |
| 195 | * that does not use off-slab slabs. |
| 196 | * For 32bit archs with 4 kB pages, is this 56. |
| 197 | * This is not serious, as it is only for large objects, when it is unwise |
| 198 | * to have too many per slab. |
| 199 | * Note: This limit can be raised by introducing a general cache whose size |
| 200 | * is less than 512 (PAGE_SIZE<<3), but greater than 256. |
| 201 | */ |
| 202 | |
Kyle Moffett | fa5b08d | 2005-09-03 15:55:03 -0700 | [diff] [blame] | 203 | typedef unsigned int kmem_bufctl_t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0) |
| 205 | #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1) |
| 206 | #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-2) |
| 207 | |
| 208 | /* Max number of objs-per-slab for caches which use off-slab slabs. |
| 209 | * Needed to avoid a possible looping condition in cache_grow(). |
| 210 | */ |
| 211 | static unsigned long offslab_limit; |
| 212 | |
| 213 | /* |
| 214 | * struct slab |
| 215 | * |
| 216 | * Manages the objs in a slab. Placed either at the beginning of mem allocated |
| 217 | * for a slab, or allocated from an general cache. |
| 218 | * Slabs are chained into three list: fully used, partial, fully free slabs. |
| 219 | */ |
| 220 | struct slab { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 221 | struct list_head list; |
| 222 | unsigned long colouroff; |
| 223 | void *s_mem; /* including colour offset */ |
| 224 | unsigned int inuse; /* num of objs active in slab */ |
| 225 | kmem_bufctl_t free; |
| 226 | unsigned short nodeid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | }; |
| 228 | |
| 229 | /* |
| 230 | * struct slab_rcu |
| 231 | * |
| 232 | * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to |
| 233 | * arrange for kmem_freepages to be called via RCU. This is useful if |
| 234 | * we need to approach a kernel structure obliquely, from its address |
| 235 | * obtained without the usual locking. We can lock the structure to |
| 236 | * stabilize it and check it's still at the given address, only if we |
| 237 | * can be sure that the memory has not been meanwhile reused for some |
| 238 | * other kind of object (which our subsystem's lock might corrupt). |
| 239 | * |
| 240 | * rcu_read_lock before reading the address, then rcu_read_unlock after |
| 241 | * taking the spinlock within the structure expected at that address. |
| 242 | * |
| 243 | * We assume struct slab_rcu can overlay struct slab when destroying. |
| 244 | */ |
| 245 | struct slab_rcu { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 246 | struct rcu_head head; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 247 | struct kmem_cache *cachep; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 248 | void *addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 249 | }; |
| 250 | |
| 251 | /* |
| 252 | * struct array_cache |
| 253 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | * Purpose: |
| 255 | * - LIFO ordering, to hand out cache-warm objects from _alloc |
| 256 | * - reduce the number of linked list operations |
| 257 | * - reduce spinlock operations |
| 258 | * |
| 259 | * The limit is stored in the per-cpu structure to reduce the data cache |
| 260 | * footprint. |
| 261 | * |
| 262 | */ |
| 263 | struct array_cache { |
| 264 | unsigned int avail; |
| 265 | unsigned int limit; |
| 266 | unsigned int batchcount; |
| 267 | unsigned int touched; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 268 | spinlock_t lock; |
| 269 | void *entry[0]; /* |
| 270 | * Must have this definition in here for the proper |
| 271 | * alignment of array_cache. Also simplifies accessing |
| 272 | * the entries. |
| 273 | * [0] is for gcc 2.95. It should really be []. |
| 274 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | }; |
| 276 | |
| 277 | /* bootstrap: The caches do not work without cpuarrays anymore, |
| 278 | * but the cpuarrays are allocated from the generic caches... |
| 279 | */ |
| 280 | #define BOOT_CPUCACHE_ENTRIES 1 |
| 281 | struct arraycache_init { |
| 282 | struct array_cache cache; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 283 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | }; |
| 285 | |
| 286 | /* |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 287 | * The slab lists for all objects. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | */ |
| 289 | struct kmem_list3 { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 290 | struct list_head slabs_partial; /* partial list first, better asm code */ |
| 291 | struct list_head slabs_full; |
| 292 | struct list_head slabs_free; |
| 293 | unsigned long free_objects; |
| 294 | unsigned long next_reap; |
| 295 | int free_touched; |
| 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 */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | }; |
| 302 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 303 | /* |
| 304 | * Need this for bootstrapping a per node allocator. |
| 305 | */ |
| 306 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES + 1) |
| 307 | struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS]; |
| 308 | #define CACHE_CACHE 0 |
| 309 | #define SIZE_AC 1 |
| 310 | #define SIZE_L3 (1 + MAX_NUMNODES) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 312 | /* |
Ivan Kokshaysky | 7243cc0 | 2005-09-22 21:43:58 -0700 | [diff] [blame] | 313 | * This function must be completely optimized away if |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 314 | * a constant is passed to it. Mostly the same as |
| 315 | * what is in linux/slab.h except it returns an |
| 316 | * index. |
| 317 | */ |
Ivan Kokshaysky | 7243cc0 | 2005-09-22 21:43:58 -0700 | [diff] [blame] | 318 | static __always_inline int index_of(const size_t size) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 319 | { |
Steven Rostedt | 5ec8a84 | 2006-02-01 03:05:44 -0800 | [diff] [blame] | 320 | extern void __bad_size(void); |
| 321 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 322 | if (__builtin_constant_p(size)) { |
| 323 | int i = 0; |
| 324 | |
| 325 | #define CACHE(x) \ |
| 326 | if (size <=x) \ |
| 327 | return i; \ |
| 328 | else \ |
| 329 | i++; |
| 330 | #include "linux/kmalloc_sizes.h" |
| 331 | #undef CACHE |
Steven Rostedt | 5ec8a84 | 2006-02-01 03:05:44 -0800 | [diff] [blame] | 332 | __bad_size(); |
Ivan Kokshaysky | 7243cc0 | 2005-09-22 21:43:58 -0700 | [diff] [blame] | 333 | } else |
Steven Rostedt | 5ec8a84 | 2006-02-01 03:05:44 -0800 | [diff] [blame] | 334 | __bad_size(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 335 | return 0; |
| 336 | } |
| 337 | |
| 338 | #define INDEX_AC index_of(sizeof(struct arraycache_init)) |
| 339 | #define INDEX_L3 index_of(sizeof(struct kmem_list3)) |
| 340 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 341 | static void kmem_list3_init(struct kmem_list3 *parent) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 342 | { |
| 343 | INIT_LIST_HEAD(&parent->slabs_full); |
| 344 | INIT_LIST_HEAD(&parent->slabs_partial); |
| 345 | INIT_LIST_HEAD(&parent->slabs_free); |
| 346 | parent->shared = NULL; |
| 347 | parent->alien = NULL; |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 348 | parent->colour_next = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 349 | spin_lock_init(&parent->list_lock); |
| 350 | parent->free_objects = 0; |
| 351 | parent->free_touched = 0; |
| 352 | } |
| 353 | |
| 354 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
| 355 | do { \ |
| 356 | INIT_LIST_HEAD(listp); \ |
| 357 | list_splice(&(cachep->nodelists[nodeid]->slab), listp); \ |
| 358 | } while (0) |
| 359 | |
| 360 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
| 361 | do { \ |
| 362 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
| 363 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ |
| 364 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ |
| 365 | } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | |
| 367 | /* |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 368 | * struct kmem_cache |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | * |
| 370 | * manages a cache. |
| 371 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 372 | |
Pekka J Enberg | 2109a2d | 2005-11-07 00:58:01 -0800 | [diff] [blame] | 373 | struct kmem_cache { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | /* 1) per-cpu data, touched during every alloc/free */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 375 | struct array_cache *array[NR_CPUS]; |
| 376 | unsigned int batchcount; |
| 377 | unsigned int limit; |
| 378 | unsigned int shared; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 379 | unsigned int buffer_size; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 380 | /* 2) touched by every alloc & free from the backend */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 381 | struct kmem_list3 *nodelists[MAX_NUMNODES]; |
| 382 | unsigned int flags; /* constant flags */ |
| 383 | unsigned int num; /* # of objs per slab */ |
| 384 | spinlock_t spinlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | |
| 386 | /* 3) cache_grow/shrink */ |
| 387 | /* order of pgs per slab (2^n) */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 388 | unsigned int gfporder; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | |
| 390 | /* force GFP flags, e.g. GFP_DMA */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 391 | gfp_t gfpflags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 392 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 393 | size_t colour; /* cache colouring range */ |
| 394 | unsigned int colour_off; /* colour offset */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 395 | struct kmem_cache *slabp_cache; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 396 | unsigned int slab_size; |
| 397 | unsigned int dflags; /* dynamic flags */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | |
| 399 | /* constructor func */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 400 | void (*ctor) (void *, struct kmem_cache *, unsigned long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | |
| 402 | /* de-constructor func */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 403 | void (*dtor) (void *, struct kmem_cache *, unsigned long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | |
| 405 | /* 4) cache creation/removal */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 406 | const char *name; |
| 407 | struct list_head next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | |
| 409 | /* 5) statistics */ |
| 410 | #if STATS |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 411 | unsigned long num_active; |
| 412 | unsigned long num_allocations; |
| 413 | unsigned long high_mark; |
| 414 | unsigned long grown; |
| 415 | unsigned long reaped; |
| 416 | unsigned long errors; |
| 417 | unsigned long max_freeable; |
| 418 | unsigned long node_allocs; |
| 419 | unsigned long node_frees; |
| 420 | atomic_t allochit; |
| 421 | atomic_t allocmiss; |
| 422 | atomic_t freehit; |
| 423 | atomic_t freemiss; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | #endif |
| 425 | #if DEBUG |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 426 | /* |
| 427 | * If debugging is enabled, then the allocator can add additional |
| 428 | * fields and/or padding to every object. buffer_size contains the total |
| 429 | * object size including these internal fields, the following two |
| 430 | * variables contain the offset to the user object and its size. |
| 431 | */ |
| 432 | int obj_offset; |
| 433 | int obj_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | #endif |
| 435 | }; |
| 436 | |
| 437 | #define CFLGS_OFF_SLAB (0x80000000UL) |
| 438 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
| 439 | |
| 440 | #define BATCHREFILL_LIMIT 16 |
| 441 | /* Optimization question: fewer reaps means less |
| 442 | * probability for unnessary cpucache drain/refill cycles. |
| 443 | * |
Adrian Bunk | dc6f3f2 | 2005-11-08 16:44:08 +0100 | [diff] [blame] | 444 | * OTOH the cpuarrays can contain lots of objects, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | * which could lock up otherwise freeable slabs. |
| 446 | */ |
| 447 | #define REAPTIMEOUT_CPUC (2*HZ) |
| 448 | #define REAPTIMEOUT_LIST3 (4*HZ) |
| 449 | |
| 450 | #if STATS |
| 451 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) |
| 452 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) |
| 453 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) |
| 454 | #define STATS_INC_GROWN(x) ((x)->grown++) |
| 455 | #define STATS_INC_REAPED(x) ((x)->reaped++) |
| 456 | #define STATS_SET_HIGH(x) do { if ((x)->num_active > (x)->high_mark) \ |
| 457 | (x)->high_mark = (x)->num_active; \ |
| 458 | } while (0) |
| 459 | #define STATS_INC_ERR(x) ((x)->errors++) |
| 460 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 461 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | #define STATS_SET_FREEABLE(x, i) \ |
| 463 | do { if ((x)->max_freeable < i) \ |
| 464 | (x)->max_freeable = i; \ |
| 465 | } while (0) |
| 466 | |
| 467 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
| 468 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) |
| 469 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) |
| 470 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) |
| 471 | #else |
| 472 | #define STATS_INC_ACTIVE(x) do { } while (0) |
| 473 | #define STATS_DEC_ACTIVE(x) do { } while (0) |
| 474 | #define STATS_INC_ALLOCED(x) do { } while (0) |
| 475 | #define STATS_INC_GROWN(x) do { } while (0) |
| 476 | #define STATS_INC_REAPED(x) do { } while (0) |
| 477 | #define STATS_SET_HIGH(x) do { } while (0) |
| 478 | #define STATS_INC_ERR(x) do { } while (0) |
| 479 | #define STATS_INC_NODEALLOCS(x) do { } while (0) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 480 | #define STATS_INC_NODEFREES(x) do { } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | #define STATS_SET_FREEABLE(x, i) \ |
| 482 | do { } while (0) |
| 483 | |
| 484 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
| 485 | #define STATS_INC_ALLOCMISS(x) do { } while (0) |
| 486 | #define STATS_INC_FREEHIT(x) do { } while (0) |
| 487 | #define STATS_INC_FREEMISS(x) do { } while (0) |
| 488 | #endif |
| 489 | |
| 490 | #if DEBUG |
| 491 | /* Magic nums for obj red zoning. |
| 492 | * Placed in the first word before and the first word after an obj. |
| 493 | */ |
| 494 | #define RED_INACTIVE 0x5A2CF071UL /* when obj is inactive */ |
| 495 | #define RED_ACTIVE 0x170FC2A5UL /* when obj is active */ |
| 496 | |
| 497 | /* ...and for poisoning */ |
| 498 | #define POISON_INUSE 0x5a /* for use-uninitialised poisoning */ |
| 499 | #define POISON_FREE 0x6b /* for use-after-free poisoning */ |
| 500 | #define POISON_END 0xa5 /* end-byte of poisoning */ |
| 501 | |
| 502 | /* memory layout of objects: |
| 503 | * 0 : objp |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 504 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | * the end of an object is aligned with the end of the real |
| 506 | * allocation. Catches writes behind the end of the allocation. |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 507 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 508 | * redzone word. |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 509 | * cachep->obj_offset: The real object. |
| 510 | * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
| 511 | * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address [BYTES_PER_WORD long] |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 513 | static int obj_offset(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 515 | return cachep->obj_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | } |
| 517 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 518 | static int obj_size(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 520 | return cachep->obj_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | } |
| 522 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 523 | static unsigned long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | { |
| 525 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 526 | return (unsigned long*) (objp+obj_offset(cachep)-BYTES_PER_WORD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | } |
| 528 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 529 | static unsigned long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | { |
| 531 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
| 532 | if (cachep->flags & SLAB_STORE_USER) |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 533 | return (unsigned long *)(objp + cachep->buffer_size - |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 534 | 2 * BYTES_PER_WORD); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 535 | return (unsigned long *)(objp + cachep->buffer_size - BYTES_PER_WORD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | } |
| 537 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 538 | static void **dbg_userword(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_STORE_USER)); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 541 | return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | } |
| 543 | |
| 544 | #else |
| 545 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 546 | #define obj_offset(x) 0 |
| 547 | #define obj_size(cachep) (cachep->buffer_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long *)NULL;}) |
| 549 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long *)NULL;}) |
| 550 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
| 551 | |
| 552 | #endif |
| 553 | |
| 554 | /* |
| 555 | * Maximum size of an obj (in 2^order pages) |
| 556 | * and absolute limit for the gfp order. |
| 557 | */ |
| 558 | #if defined(CONFIG_LARGE_ALLOCS) |
| 559 | #define MAX_OBJ_ORDER 13 /* up to 32Mb */ |
| 560 | #define MAX_GFP_ORDER 13 /* up to 32Mb */ |
| 561 | #elif defined(CONFIG_MMU) |
| 562 | #define MAX_OBJ_ORDER 5 /* 32 pages */ |
| 563 | #define MAX_GFP_ORDER 5 /* 32 pages */ |
| 564 | #else |
| 565 | #define MAX_OBJ_ORDER 8 /* up to 1Mb */ |
| 566 | #define MAX_GFP_ORDER 8 /* up to 1Mb */ |
| 567 | #endif |
| 568 | |
| 569 | /* |
| 570 | * Do not go above this order unless 0 objects fit into the slab. |
| 571 | */ |
| 572 | #define BREAK_GFP_ORDER_HI 1 |
| 573 | #define BREAK_GFP_ORDER_LO 0 |
| 574 | static int slab_break_gfp_order = BREAK_GFP_ORDER_LO; |
| 575 | |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 576 | /* Functions for storing/retrieving the cachep and or slab from the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | * global 'mem_map'. These are used to find the slab an obj belongs to. |
| 578 | * With kfree(), these are used to find the cache which an obj belongs to. |
| 579 | */ |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 580 | static inline void page_set_cache(struct page *page, struct kmem_cache *cache) |
| 581 | { |
| 582 | page->lru.next = (struct list_head *)cache; |
| 583 | } |
| 584 | |
| 585 | static inline struct kmem_cache *page_get_cache(struct page *page) |
| 586 | { |
| 587 | return (struct kmem_cache *)page->lru.next; |
| 588 | } |
| 589 | |
| 590 | static inline void page_set_slab(struct page *page, struct slab *slab) |
| 591 | { |
| 592 | page->lru.prev = (struct list_head *)slab; |
| 593 | } |
| 594 | |
| 595 | static inline struct slab *page_get_slab(struct page *page) |
| 596 | { |
| 597 | return (struct slab *)page->lru.prev; |
| 598 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 599 | |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 600 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
| 601 | { |
| 602 | struct page *page = virt_to_page(obj); |
| 603 | return page_get_cache(page); |
| 604 | } |
| 605 | |
| 606 | static inline struct slab *virt_to_slab(const void *obj) |
| 607 | { |
| 608 | struct page *page = virt_to_page(obj); |
| 609 | return page_get_slab(page); |
| 610 | } |
| 611 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 612 | /* These are the default caches for kmalloc. Custom caches can have other sizes. */ |
| 613 | struct cache_sizes malloc_sizes[] = { |
| 614 | #define CACHE(x) { .cs_size = (x) }, |
| 615 | #include <linux/kmalloc_sizes.h> |
| 616 | CACHE(ULONG_MAX) |
| 617 | #undef CACHE |
| 618 | }; |
| 619 | EXPORT_SYMBOL(malloc_sizes); |
| 620 | |
| 621 | /* Must match cache_sizes above. Out of line to keep cache footprint low. */ |
| 622 | struct cache_names { |
| 623 | char *name; |
| 624 | char *name_dma; |
| 625 | }; |
| 626 | |
| 627 | static struct cache_names __initdata cache_names[] = { |
| 628 | #define CACHE(x) { .name = "size-" #x, .name_dma = "size-" #x "(DMA)" }, |
| 629 | #include <linux/kmalloc_sizes.h> |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 630 | {NULL,} |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | #undef CACHE |
| 632 | }; |
| 633 | |
| 634 | static struct arraycache_init initarray_cache __initdata = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 635 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | static struct arraycache_init initarray_generic = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 637 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | |
| 639 | /* internal cache of cache description objs */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 640 | static struct kmem_cache cache_cache = { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 641 | .batchcount = 1, |
| 642 | .limit = BOOT_CPUCACHE_ENTRIES, |
| 643 | .shared = 1, |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 644 | .buffer_size = sizeof(struct kmem_cache), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 645 | .flags = SLAB_NO_REAP, |
| 646 | .spinlock = SPIN_LOCK_UNLOCKED, |
| 647 | .name = "kmem_cache", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | #if DEBUG |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 649 | .obj_size = sizeof(struct kmem_cache), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | #endif |
| 651 | }; |
| 652 | |
| 653 | /* Guard access to the cache-chain. */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 654 | static DEFINE_MUTEX(cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | static struct list_head cache_chain; |
| 656 | |
| 657 | /* |
| 658 | * vm_enough_memory() looks at this to determine how many |
| 659 | * slab-allocated pages are possibly freeable under pressure |
| 660 | * |
| 661 | * SLAB_RECLAIM_ACCOUNT turns this on per-slab |
| 662 | */ |
| 663 | atomic_t slab_reclaim_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | |
| 665 | /* |
| 666 | * chicken and egg problem: delay the per-cpu array allocation |
| 667 | * until the general caches are up. |
| 668 | */ |
| 669 | static enum { |
| 670 | NONE, |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 671 | PARTIAL_AC, |
| 672 | PARTIAL_L3, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | FULL |
| 674 | } g_cpucache_up; |
| 675 | |
| 676 | static DEFINE_PER_CPU(struct work_struct, reap_work); |
| 677 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 678 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, int node); |
| 679 | static void enable_cpucache(struct kmem_cache *cachep); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 680 | static void cache_reap(void *unused); |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 681 | static int __node_shrink(struct kmem_cache *cachep, int node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 683 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | { |
| 685 | return cachep->array[smp_processor_id()]; |
| 686 | } |
| 687 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 688 | static inline struct kmem_cache *__find_general_cachep(size_t size, gfp_t gfpflags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | { |
| 690 | struct cache_sizes *csizep = malloc_sizes; |
| 691 | |
| 692 | #if DEBUG |
| 693 | /* This happens if someone tries to call |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 694 | * kmem_cache_create(), or __kmalloc(), before |
| 695 | * the generic caches are initialized. |
| 696 | */ |
Alok Kataria | c7e43c7 | 2005-09-14 12:17:53 -0700 | [diff] [blame] | 697 | BUG_ON(malloc_sizes[INDEX_AC].cs_cachep == NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | #endif |
| 699 | while (size > csizep->cs_size) |
| 700 | csizep++; |
| 701 | |
| 702 | /* |
Martin Hicks | 0abf40c | 2005-09-03 15:54:54 -0700 | [diff] [blame] | 703 | * Really subtle: The last entry with cs->cs_size==ULONG_MAX |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | * has cs_{dma,}cachep==NULL. Thus no special case |
| 705 | * for large kmalloc calls required. |
| 706 | */ |
| 707 | if (unlikely(gfpflags & GFP_DMA)) |
| 708 | return csizep->cs_dmacachep; |
| 709 | return csizep->cs_cachep; |
| 710 | } |
| 711 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 712 | 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] | 713 | { |
| 714 | return __find_general_cachep(size, gfpflags); |
| 715 | } |
| 716 | EXPORT_SYMBOL(kmem_find_general_cachep); |
| 717 | |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 718 | 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] | 719 | { |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 720 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); |
| 721 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 722 | |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 723 | /* Calculate the number of objects and left-over bytes for a given |
| 724 | buffer size. */ |
| 725 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
| 726 | size_t align, int flags, size_t *left_over, |
| 727 | unsigned int *num) |
| 728 | { |
| 729 | int nr_objs; |
| 730 | size_t mgmt_size; |
| 731 | size_t slab_size = PAGE_SIZE << gfporder; |
| 732 | |
| 733 | /* |
| 734 | * The slab management structure can be either off the slab or |
| 735 | * on it. For the latter case, the memory allocated for a |
| 736 | * slab is used for: |
| 737 | * |
| 738 | * - The struct slab |
| 739 | * - One kmem_bufctl_t for each object |
| 740 | * - Padding to respect alignment of @align |
| 741 | * - @buffer_size bytes for each object |
| 742 | * |
| 743 | * If the slab management structure is off the slab, then the |
| 744 | * alignment will already be calculated into the size. Because |
| 745 | * the slabs are all pages aligned, the objects will be at the |
| 746 | * correct alignment when allocated. |
| 747 | */ |
| 748 | if (flags & CFLGS_OFF_SLAB) { |
| 749 | mgmt_size = 0; |
| 750 | nr_objs = slab_size / buffer_size; |
| 751 | |
| 752 | if (nr_objs > SLAB_LIMIT) |
| 753 | nr_objs = SLAB_LIMIT; |
| 754 | } else { |
| 755 | /* |
| 756 | * Ignore padding for the initial guess. The padding |
| 757 | * is at most @align-1 bytes, and @buffer_size is at |
| 758 | * least @align. In the worst case, this result will |
| 759 | * be one greater than the number of objects that fit |
| 760 | * into the memory allocation when taking the padding |
| 761 | * into account. |
| 762 | */ |
| 763 | nr_objs = (slab_size - sizeof(struct slab)) / |
| 764 | (buffer_size + sizeof(kmem_bufctl_t)); |
| 765 | |
| 766 | /* |
| 767 | * This calculated number will be either the right |
| 768 | * amount, or one greater than what we want. |
| 769 | */ |
| 770 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size |
| 771 | > slab_size) |
| 772 | nr_objs--; |
| 773 | |
| 774 | if (nr_objs > SLAB_LIMIT) |
| 775 | nr_objs = SLAB_LIMIT; |
| 776 | |
| 777 | mgmt_size = slab_mgmt_size(nr_objs, align); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 778 | } |
Steven Rostedt | fbaccac | 2006-02-01 03:05:45 -0800 | [diff] [blame] | 779 | *num = nr_objs; |
| 780 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 781 | } |
| 782 | |
| 783 | #define slab_error(cachep, msg) __slab_error(__FUNCTION__, cachep, msg) |
| 784 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 785 | static void __slab_error(const char *function, struct kmem_cache *cachep, char *msg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 786 | { |
| 787 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 788 | function, cachep->name, msg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | dump_stack(); |
| 790 | } |
| 791 | |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 792 | #ifdef CONFIG_NUMA |
| 793 | /* |
| 794 | * Special reaping functions for NUMA systems called from cache_reap(). |
| 795 | * These take care of doing round robin flushing of alien caches (containing |
| 796 | * objects freed on different nodes from which they were allocated) and the |
| 797 | * flushing of remote pcps by calling drain_node_pages. |
| 798 | */ |
| 799 | static DEFINE_PER_CPU(unsigned long, reap_node); |
| 800 | |
| 801 | static void init_reap_node(int cpu) |
| 802 | { |
| 803 | int node; |
| 804 | |
| 805 | node = next_node(cpu_to_node(cpu), node_online_map); |
| 806 | if (node == MAX_NUMNODES) |
| 807 | node = 0; |
| 808 | |
| 809 | __get_cpu_var(reap_node) = node; |
| 810 | } |
| 811 | |
| 812 | static void next_reap_node(void) |
| 813 | { |
| 814 | int node = __get_cpu_var(reap_node); |
| 815 | |
| 816 | /* |
| 817 | * Also drain per cpu pages on remote zones |
| 818 | */ |
| 819 | if (node != numa_node_id()) |
| 820 | drain_node_pages(node); |
| 821 | |
| 822 | node = next_node(node, node_online_map); |
| 823 | if (unlikely(node >= MAX_NUMNODES)) |
| 824 | node = first_node(node_online_map); |
| 825 | __get_cpu_var(reap_node) = node; |
| 826 | } |
| 827 | |
| 828 | #else |
| 829 | #define init_reap_node(cpu) do { } while (0) |
| 830 | #define next_reap_node(void) do { } while (0) |
| 831 | #endif |
| 832 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 833 | /* |
| 834 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz |
| 835 | * via the workqueue/eventd. |
| 836 | * Add the CPU number into the expiration time to minimize the possibility of |
| 837 | * the CPUs getting into lockstep and contending for the global cache chain |
| 838 | * lock. |
| 839 | */ |
| 840 | static void __devinit start_cpu_timer(int cpu) |
| 841 | { |
| 842 | struct work_struct *reap_work = &per_cpu(reap_work, cpu); |
| 843 | |
| 844 | /* |
| 845 | * When this gets called from do_initcalls via cpucache_init(), |
| 846 | * init_workqueues() has already run, so keventd will be setup |
| 847 | * at that time. |
| 848 | */ |
| 849 | if (keventd_up() && reap_work->func == NULL) { |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 850 | init_reap_node(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 851 | INIT_WORK(reap_work, cache_reap, NULL); |
| 852 | schedule_delayed_work_on(cpu, reap_work, HZ + 3 * cpu); |
| 853 | } |
| 854 | } |
| 855 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 856 | static struct array_cache *alloc_arraycache(int node, int entries, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 857 | int batchcount) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 859 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 860 | struct array_cache *nc = NULL; |
| 861 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 862 | nc = kmalloc_node(memsize, GFP_KERNEL, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | if (nc) { |
| 864 | nc->avail = 0; |
| 865 | nc->limit = entries; |
| 866 | nc->batchcount = batchcount; |
| 867 | nc->touched = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 868 | spin_lock_init(&nc->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 869 | } |
| 870 | return nc; |
| 871 | } |
| 872 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 873 | #ifdef CONFIG_NUMA |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 874 | static void *__cache_alloc_node(struct kmem_cache *, gfp_t, int); |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 875 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 876 | static struct array_cache **alloc_alien_cache(int node, int limit) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 877 | { |
| 878 | struct array_cache **ac_ptr; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 879 | int memsize = sizeof(void *) * MAX_NUMNODES; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 880 | int i; |
| 881 | |
| 882 | if (limit > 1) |
| 883 | limit = 12; |
| 884 | ac_ptr = kmalloc_node(memsize, GFP_KERNEL, node); |
| 885 | if (ac_ptr) { |
| 886 | for_each_node(i) { |
| 887 | if (i == node || !node_online(i)) { |
| 888 | ac_ptr[i] = NULL; |
| 889 | continue; |
| 890 | } |
| 891 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d); |
| 892 | if (!ac_ptr[i]) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 893 | for (i--; i <= 0; i--) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 894 | kfree(ac_ptr[i]); |
| 895 | kfree(ac_ptr); |
| 896 | return NULL; |
| 897 | } |
| 898 | } |
| 899 | } |
| 900 | return ac_ptr; |
| 901 | } |
| 902 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 903 | static void free_alien_cache(struct array_cache **ac_ptr) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 904 | { |
| 905 | int i; |
| 906 | |
| 907 | if (!ac_ptr) |
| 908 | return; |
| 909 | |
| 910 | for_each_node(i) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 911 | kfree(ac_ptr[i]); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 912 | |
| 913 | kfree(ac_ptr); |
| 914 | } |
| 915 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 916 | static void __drain_alien_cache(struct kmem_cache *cachep, |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 917 | struct array_cache *ac, int node) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 918 | { |
| 919 | struct kmem_list3 *rl3 = cachep->nodelists[node]; |
| 920 | |
| 921 | if (ac->avail) { |
| 922 | spin_lock(&rl3->list_lock); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 923 | free_block(cachep, ac->entry, ac->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 924 | ac->avail = 0; |
| 925 | spin_unlock(&rl3->list_lock); |
| 926 | } |
| 927 | } |
| 928 | |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 929 | /* |
| 930 | * Called from cache_reap() to regularly drain alien caches round robin. |
| 931 | */ |
| 932 | static void reap_alien(struct kmem_cache *cachep, struct kmem_list3 *l3) |
| 933 | { |
| 934 | int node = __get_cpu_var(reap_node); |
| 935 | |
| 936 | if (l3->alien) { |
| 937 | struct array_cache *ac = l3->alien[node]; |
| 938 | if (ac && ac->avail) { |
| 939 | spin_lock_irq(&ac->lock); |
| 940 | __drain_alien_cache(cachep, ac, node); |
| 941 | spin_unlock_irq(&ac->lock); |
| 942 | } |
| 943 | } |
| 944 | } |
| 945 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 946 | static void drain_alien_cache(struct kmem_cache *cachep, struct array_cache **alien) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 947 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 948 | int i = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 949 | struct array_cache *ac; |
| 950 | unsigned long flags; |
| 951 | |
| 952 | for_each_online_node(i) { |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 953 | ac = alien[i]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 954 | if (ac) { |
| 955 | spin_lock_irqsave(&ac->lock, flags); |
| 956 | __drain_alien_cache(cachep, ac, i); |
| 957 | spin_unlock_irqrestore(&ac->lock, flags); |
| 958 | } |
| 959 | } |
| 960 | } |
| 961 | #else |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 962 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 963 | #define drain_alien_cache(cachep, alien) do { } while (0) |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 964 | #define reap_alien(cachep, l3) do { } while (0) |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 965 | |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 966 | static inline struct array_cache **alloc_alien_cache(int node, int limit) |
| 967 | { |
| 968 | return (struct array_cache **) 0x01020304ul; |
| 969 | } |
| 970 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 971 | static inline void free_alien_cache(struct array_cache **ac_ptr) |
| 972 | { |
| 973 | } |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 974 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 975 | #endif |
| 976 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 977 | static int __devinit cpuup_callback(struct notifier_block *nfb, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 978 | unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 979 | { |
| 980 | long cpu = (long)hcpu; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 981 | struct kmem_cache *cachep; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 982 | struct kmem_list3 *l3 = NULL; |
| 983 | int node = cpu_to_node(cpu); |
| 984 | int memsize = sizeof(struct kmem_list3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 985 | |
| 986 | switch (action) { |
| 987 | case CPU_UP_PREPARE: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 988 | mutex_lock(&cache_chain_mutex); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 989 | /* we need to do this right in the beginning since |
| 990 | * alloc_arraycache's are going to use this list. |
| 991 | * kmalloc_node allows us to add the slab to the right |
| 992 | * kmem_list3 and not this cpu's kmem_list3 |
| 993 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 994 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 995 | list_for_each_entry(cachep, &cache_chain, next) { |
| 996 | /* setup the size64 kmemlist for cpu before we can |
| 997 | * begin anything. Make sure some other cpu on this |
| 998 | * node has not already allocated this |
| 999 | */ |
| 1000 | if (!cachep->nodelists[node]) { |
| 1001 | if (!(l3 = kmalloc_node(memsize, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1002 | GFP_KERNEL, node))) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1003 | goto bad; |
| 1004 | kmem_list3_init(l3); |
| 1005 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1006 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1007 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1008 | /* |
| 1009 | * The l3s don't come and go as CPUs come and |
| 1010 | * go. cache_chain_mutex is sufficient |
| 1011 | * protection here. |
| 1012 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1013 | cachep->nodelists[node] = l3; |
| 1014 | } |
| 1015 | |
| 1016 | spin_lock_irq(&cachep->nodelists[node]->list_lock); |
| 1017 | cachep->nodelists[node]->free_limit = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1018 | (1 + nr_cpus_node(node)) * |
| 1019 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1020 | spin_unlock_irq(&cachep->nodelists[node]->list_lock); |
| 1021 | } |
| 1022 | |
| 1023 | /* Now we can go ahead with allocating the shared array's |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1024 | & array cache's */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1025 | list_for_each_entry(cachep, &cache_chain, next) { |
Tobias Klauser | cd105df | 2006-01-08 01:00:59 -0800 | [diff] [blame] | 1026 | struct array_cache *nc; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1027 | struct array_cache *shared; |
| 1028 | struct array_cache **alien; |
Tobias Klauser | cd105df | 2006-01-08 01:00:59 -0800 | [diff] [blame] | 1029 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1030 | nc = alloc_arraycache(node, cachep->limit, |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1031 | cachep->batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | if (!nc) |
| 1033 | goto bad; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1034 | shared = alloc_arraycache(node, |
| 1035 | cachep->shared * cachep->batchcount, |
| 1036 | 0xbaadf00d); |
| 1037 | if (!shared) |
| 1038 | goto bad; |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 1039 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1040 | alien = alloc_alien_cache(node, cachep->limit); |
| 1041 | if (!alien) |
| 1042 | goto bad; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1043 | cachep->array[cpu] = nc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1044 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1045 | l3 = cachep->nodelists[node]; |
| 1046 | BUG_ON(!l3); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1047 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1048 | spin_lock_irq(&l3->list_lock); |
| 1049 | if (!l3->shared) { |
| 1050 | /* |
| 1051 | * We are serialised from CPU_DEAD or |
| 1052 | * CPU_UP_CANCELLED by the cpucontrol lock |
| 1053 | */ |
| 1054 | l3->shared = shared; |
| 1055 | shared = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1056 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1057 | #ifdef CONFIG_NUMA |
| 1058 | if (!l3->alien) { |
| 1059 | l3->alien = alien; |
| 1060 | alien = NULL; |
| 1061 | } |
| 1062 | #endif |
| 1063 | spin_unlock_irq(&l3->list_lock); |
| 1064 | |
| 1065 | kfree(shared); |
| 1066 | free_alien_cache(alien); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1067 | } |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1068 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | break; |
| 1070 | case CPU_ONLINE: |
| 1071 | start_cpu_timer(cpu); |
| 1072 | break; |
| 1073 | #ifdef CONFIG_HOTPLUG_CPU |
| 1074 | case CPU_DEAD: |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1075 | /* |
| 1076 | * Even if all the cpus of a node are down, we don't free the |
| 1077 | * kmem_list3 of any cache. This to avoid a race between |
| 1078 | * cpu_down, and a kmalloc allocation from another cpu for |
| 1079 | * memory from the node of the cpu going down. The list3 |
| 1080 | * structure is usually allocated from kmem_cache_create() and |
| 1081 | * gets destroyed at kmem_cache_destroy(). |
| 1082 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1083 | /* fall thru */ |
| 1084 | case CPU_UP_CANCELED: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1085 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | |
| 1087 | list_for_each_entry(cachep, &cache_chain, next) { |
| 1088 | struct array_cache *nc; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1089 | struct array_cache *shared; |
| 1090 | struct array_cache **alien; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1091 | cpumask_t mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1093 | mask = node_to_cpumask(node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1094 | /* cpu is dead; no one can alloc from it. */ |
| 1095 | nc = cachep->array[cpu]; |
| 1096 | cachep->array[cpu] = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1097 | l3 = cachep->nodelists[node]; |
| 1098 | |
| 1099 | if (!l3) |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1100 | goto free_array_cache; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1101 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 1102 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1103 | |
| 1104 | /* Free limit for this kmem_list3 */ |
| 1105 | l3->free_limit -= cachep->batchcount; |
| 1106 | if (nc) |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 1107 | free_block(cachep, nc->entry, nc->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1108 | |
| 1109 | if (!cpus_empty(mask)) { |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 1110 | spin_unlock_irq(&l3->list_lock); |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1111 | goto free_array_cache; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1112 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1113 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1114 | shared = l3->shared; |
| 1115 | if (shared) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1116 | free_block(cachep, l3->shared->entry, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1117 | l3->shared->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1118 | l3->shared = NULL; |
| 1119 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1120 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1121 | alien = l3->alien; |
| 1122 | l3->alien = NULL; |
| 1123 | |
| 1124 | spin_unlock_irq(&l3->list_lock); |
| 1125 | |
| 1126 | kfree(shared); |
| 1127 | if (alien) { |
| 1128 | drain_alien_cache(cachep, alien); |
| 1129 | free_alien_cache(alien); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1130 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1131 | free_array_cache: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1132 | kfree(nc); |
| 1133 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1134 | /* |
| 1135 | * In the previous loop, all the objects were freed to |
| 1136 | * the respective cache's slabs, now we can go ahead and |
| 1137 | * shrink each nodelist to its limit. |
| 1138 | */ |
| 1139 | list_for_each_entry(cachep, &cache_chain, next) { |
| 1140 | l3 = cachep->nodelists[node]; |
| 1141 | if (!l3) |
| 1142 | continue; |
| 1143 | spin_lock_irq(&l3->list_lock); |
| 1144 | /* free slabs belonging to this node */ |
| 1145 | __node_shrink(cachep, node); |
| 1146 | spin_unlock_irq(&l3->list_lock); |
| 1147 | } |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1148 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | break; |
| 1150 | #endif |
| 1151 | } |
| 1152 | return NOTIFY_OK; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1153 | bad: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1154 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1155 | return NOTIFY_BAD; |
| 1156 | } |
| 1157 | |
| 1158 | static struct notifier_block cpucache_notifier = { &cpuup_callback, NULL, 0 }; |
| 1159 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1160 | /* |
| 1161 | * swap the static kmem_list3 with kmalloced memory |
| 1162 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1163 | static void init_list(struct kmem_cache *cachep, struct kmem_list3 *list, int nodeid) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1164 | { |
| 1165 | struct kmem_list3 *ptr; |
| 1166 | |
| 1167 | BUG_ON(cachep->nodelists[nodeid] != list); |
| 1168 | ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_KERNEL, nodeid); |
| 1169 | BUG_ON(!ptr); |
| 1170 | |
| 1171 | local_irq_disable(); |
| 1172 | memcpy(ptr, list, sizeof(struct kmem_list3)); |
| 1173 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
| 1174 | cachep->nodelists[nodeid] = ptr; |
| 1175 | local_irq_enable(); |
| 1176 | } |
| 1177 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | /* Initialisation. |
| 1179 | * Called after the gfp() functions have been enabled, and before smp_init(). |
| 1180 | */ |
| 1181 | void __init kmem_cache_init(void) |
| 1182 | { |
| 1183 | size_t left_over; |
| 1184 | struct cache_sizes *sizes; |
| 1185 | struct cache_names *names; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1186 | int i; |
Jack Steiner | 07ed76b | 2006-03-07 21:55:46 -0800 | [diff] [blame] | 1187 | int order; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1188 | |
| 1189 | for (i = 0; i < NUM_INIT_LISTS; i++) { |
| 1190 | kmem_list3_init(&initkmem_list3[i]); |
| 1191 | if (i < MAX_NUMNODES) |
| 1192 | cache_cache.nodelists[i] = NULL; |
| 1193 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1194 | |
| 1195 | /* |
| 1196 | * Fragmentation resistance on low memory - only use bigger |
| 1197 | * page orders on machines with more than 32MB of memory. |
| 1198 | */ |
| 1199 | if (num_physpages > (32 << 20) >> PAGE_SHIFT) |
| 1200 | slab_break_gfp_order = BREAK_GFP_ORDER_HI; |
| 1201 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1202 | /* Bootstrap is tricky, because several objects are allocated |
| 1203 | * from caches that do not exist yet: |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1204 | * 1) initialize the cache_cache cache: it contains the struct kmem_cache |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1205 | * structures of all caches, except cache_cache itself: cache_cache |
| 1206 | * is statically allocated. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1207 | * Initially an __init data area is used for the head array and the |
| 1208 | * kmem_list3 structures, it's replaced with a kmalloc allocated |
| 1209 | * array at the end of the bootstrap. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1210 | * 2) Create the first kmalloc cache. |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1211 | * The struct kmem_cache for the new cache is allocated normally. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1212 | * An __init data area is used for the head array. |
| 1213 | * 3) Create the remaining kmalloc caches, with minimally sized |
| 1214 | * head arrays. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | * 4) Replace the __init data head arrays for cache_cache and the first |
| 1216 | * kmalloc cache with kmalloc allocated arrays. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1217 | * 5) Replace the __init data for kmem_list3 for cache_cache and |
| 1218 | * the other cache's with kmalloc allocated memory. |
| 1219 | * 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] | 1220 | */ |
| 1221 | |
| 1222 | /* 1) create the cache_cache */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1223 | INIT_LIST_HEAD(&cache_chain); |
| 1224 | list_add(&cache_cache.next, &cache_chain); |
| 1225 | cache_cache.colour_off = cache_line_size(); |
| 1226 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1227 | cache_cache.nodelists[numa_node_id()] = &initkmem_list3[CACHE_CACHE]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1228 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1229 | cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, cache_line_size()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1230 | |
Jack Steiner | 07ed76b | 2006-03-07 21:55:46 -0800 | [diff] [blame] | 1231 | for (order = 0; order < MAX_ORDER; order++) { |
| 1232 | cache_estimate(order, cache_cache.buffer_size, |
| 1233 | cache_line_size(), 0, &left_over, &cache_cache.num); |
| 1234 | if (cache_cache.num) |
| 1235 | break; |
| 1236 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1237 | if (!cache_cache.num) |
| 1238 | BUG(); |
Jack Steiner | 07ed76b | 2006-03-07 21:55:46 -0800 | [diff] [blame] | 1239 | cache_cache.gfporder = order; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1240 | cache_cache.colour = left_over / cache_cache.colour_off; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1241 | cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) + |
| 1242 | sizeof(struct slab), cache_line_size()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | |
| 1244 | /* 2+3) create the kmalloc caches */ |
| 1245 | sizes = malloc_sizes; |
| 1246 | names = cache_names; |
| 1247 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1248 | /* Initialize the caches that provide memory for the array cache |
| 1249 | * and the kmem_list3 structures first. |
| 1250 | * Without this, further allocations will bug |
| 1251 | */ |
| 1252 | |
| 1253 | sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1254 | sizes[INDEX_AC].cs_size, |
| 1255 | ARCH_KMALLOC_MINALIGN, |
| 1256 | (ARCH_KMALLOC_FLAGS | |
| 1257 | SLAB_PANIC), NULL, NULL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1258 | |
| 1259 | if (INDEX_AC != INDEX_L3) |
| 1260 | sizes[INDEX_L3].cs_cachep = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1261 | kmem_cache_create(names[INDEX_L3].name, |
| 1262 | sizes[INDEX_L3].cs_size, |
| 1263 | ARCH_KMALLOC_MINALIGN, |
| 1264 | (ARCH_KMALLOC_FLAGS | SLAB_PANIC), NULL, |
| 1265 | NULL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1266 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | while (sizes->cs_size != ULONG_MAX) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1268 | /* |
| 1269 | * For performance, all the general caches are L1 aligned. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | * This should be particularly beneficial on SMP boxes, as it |
| 1271 | * eliminates "false sharing". |
| 1272 | * Note for systems short on memory removing the alignment will |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1273 | * allow tighter packing of the smaller caches. |
| 1274 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1275 | if (!sizes->cs_cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1276 | sizes->cs_cachep = kmem_cache_create(names->name, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1277 | sizes->cs_size, |
| 1278 | ARCH_KMALLOC_MINALIGN, |
| 1279 | (ARCH_KMALLOC_FLAGS |
| 1280 | | SLAB_PANIC), |
| 1281 | NULL, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | |
| 1283 | /* Inc off-slab bufctl limit until the ceiling is hit. */ |
| 1284 | if (!(OFF_SLAB(sizes->cs_cachep))) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1285 | offslab_limit = sizes->cs_size - sizeof(struct slab); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1286 | offslab_limit /= sizeof(kmem_bufctl_t); |
| 1287 | } |
| 1288 | |
| 1289 | sizes->cs_dmacachep = kmem_cache_create(names->name_dma, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1290 | sizes->cs_size, |
| 1291 | ARCH_KMALLOC_MINALIGN, |
| 1292 | (ARCH_KMALLOC_FLAGS | |
| 1293 | SLAB_CACHE_DMA | |
| 1294 | SLAB_PANIC), NULL, |
| 1295 | NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1296 | |
| 1297 | sizes++; |
| 1298 | names++; |
| 1299 | } |
| 1300 | /* 4) Replace the bootstrap head arrays */ |
| 1301 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1302 | void *ptr; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1303 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1304 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1305 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1306 | local_irq_disable(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1307 | BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); |
| 1308 | memcpy(ptr, cpu_cache_get(&cache_cache), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1309 | sizeof(struct arraycache_init)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1310 | cache_cache.array[smp_processor_id()] = ptr; |
| 1311 | local_irq_enable(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1312 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1313 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1314 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1315 | local_irq_disable(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1316 | BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1317 | != &initarray_generic.cache); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1318 | memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1319 | sizeof(struct arraycache_init)); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1320 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1321 | ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1322 | local_irq_enable(); |
| 1323 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1324 | /* 5) Replace the bootstrap kmem_list3's */ |
| 1325 | { |
| 1326 | int node; |
| 1327 | /* Replace the static kmem_list3 structures for the boot cpu */ |
| 1328 | init_list(&cache_cache, &initkmem_list3[CACHE_CACHE], |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1329 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1330 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1331 | for_each_online_node(node) { |
| 1332 | init_list(malloc_sizes[INDEX_AC].cs_cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1333 | &initkmem_list3[SIZE_AC + node], node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1334 | |
| 1335 | if (INDEX_AC != INDEX_L3) { |
| 1336 | init_list(malloc_sizes[INDEX_L3].cs_cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1337 | &initkmem_list3[SIZE_L3 + node], |
| 1338 | node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1339 | } |
| 1340 | } |
| 1341 | } |
| 1342 | |
| 1343 | /* 6) resize the head arrays to their final sizes */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1345 | struct kmem_cache *cachep; |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1346 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1347 | list_for_each_entry(cachep, &cache_chain, next) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1348 | enable_cpucache(cachep); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1349 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | } |
| 1351 | |
| 1352 | /* Done! */ |
| 1353 | g_cpucache_up = FULL; |
| 1354 | |
| 1355 | /* Register a cpu startup notifier callback |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1356 | * that initializes cpu_cache_get for all new cpus |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | */ |
| 1358 | register_cpu_notifier(&cpucache_notifier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1359 | |
| 1360 | /* The reap timers are started later, with a module init call: |
| 1361 | * That part of the kernel is not yet operational. |
| 1362 | */ |
| 1363 | } |
| 1364 | |
| 1365 | static int __init cpucache_init(void) |
| 1366 | { |
| 1367 | int cpu; |
| 1368 | |
| 1369 | /* |
| 1370 | * Register the timers that return unneeded |
| 1371 | * pages to gfp. |
| 1372 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1373 | for_each_online_cpu(cpu) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1374 | start_cpu_timer(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1375 | |
| 1376 | return 0; |
| 1377 | } |
| 1378 | |
| 1379 | __initcall(cpucache_init); |
| 1380 | |
| 1381 | /* |
| 1382 | * Interface to system's page allocator. No need to hold the cache-lock. |
| 1383 | * |
| 1384 | * If we requested dmaable memory, we will get it. Even if we |
| 1385 | * did not request dmaable memory, we might get it, but that |
| 1386 | * would be relatively rare and ignorable. |
| 1387 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1388 | 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] | 1389 | { |
| 1390 | struct page *page; |
| 1391 | void *addr; |
| 1392 | int i; |
| 1393 | |
| 1394 | flags |= cachep->gfpflags; |
Christoph Lameter | 50c85a1 | 2005-11-13 16:06:47 -0800 | [diff] [blame] | 1395 | page = alloc_pages_node(nodeid, flags, cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1396 | if (!page) |
| 1397 | return NULL; |
| 1398 | addr = page_address(page); |
| 1399 | |
| 1400 | i = (1 << cachep->gfporder); |
| 1401 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1402 | atomic_add(i, &slab_reclaim_pages); |
| 1403 | add_page_state(nr_slab, i); |
| 1404 | while (i--) { |
| 1405 | SetPageSlab(page); |
| 1406 | page++; |
| 1407 | } |
| 1408 | return addr; |
| 1409 | } |
| 1410 | |
| 1411 | /* |
| 1412 | * Interface to system's page release. |
| 1413 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1414 | static void kmem_freepages(struct kmem_cache *cachep, void *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1415 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1416 | unsigned long i = (1 << cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1417 | struct page *page = virt_to_page(addr); |
| 1418 | const unsigned long nr_freed = i; |
| 1419 | |
| 1420 | while (i--) { |
| 1421 | if (!TestClearPageSlab(page)) |
| 1422 | BUG(); |
| 1423 | page++; |
| 1424 | } |
| 1425 | sub_page_state(nr_slab, nr_freed); |
| 1426 | if (current->reclaim_state) |
| 1427 | current->reclaim_state->reclaimed_slab += nr_freed; |
| 1428 | free_pages((unsigned long)addr, cachep->gfporder); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1429 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1430 | atomic_sub(1 << cachep->gfporder, &slab_reclaim_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1431 | } |
| 1432 | |
| 1433 | static void kmem_rcu_free(struct rcu_head *head) |
| 1434 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1435 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1436 | struct kmem_cache *cachep = slab_rcu->cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | |
| 1438 | kmem_freepages(cachep, slab_rcu->addr); |
| 1439 | if (OFF_SLAB(cachep)) |
| 1440 | kmem_cache_free(cachep->slabp_cache, slab_rcu); |
| 1441 | } |
| 1442 | |
| 1443 | #if DEBUG |
| 1444 | |
| 1445 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1446 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1447 | unsigned long caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1448 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1449 | int size = obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1451 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1453 | if (size < 5 * sizeof(unsigned long)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | return; |
| 1455 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1456 | *addr++ = 0x12345678; |
| 1457 | *addr++ = caller; |
| 1458 | *addr++ = smp_processor_id(); |
| 1459 | size -= 3 * sizeof(unsigned long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1460 | { |
| 1461 | unsigned long *sptr = &caller; |
| 1462 | unsigned long svalue; |
| 1463 | |
| 1464 | while (!kstack_end(sptr)) { |
| 1465 | svalue = *sptr++; |
| 1466 | if (kernel_text_address(svalue)) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1467 | *addr++ = svalue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1468 | size -= sizeof(unsigned long); |
| 1469 | if (size <= sizeof(unsigned long)) |
| 1470 | break; |
| 1471 | } |
| 1472 | } |
| 1473 | |
| 1474 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1475 | *addr++ = 0x87654321; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1476 | } |
| 1477 | #endif |
| 1478 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1479 | 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] | 1480 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1481 | int size = obj_size(cachep); |
| 1482 | addr = &((char *)addr)[obj_offset(cachep)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1483 | |
| 1484 | memset(addr, val, size); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1485 | *(unsigned char *)(addr + size - 1) = POISON_END; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | } |
| 1487 | |
| 1488 | static void dump_line(char *data, int offset, int limit) |
| 1489 | { |
| 1490 | int i; |
| 1491 | printk(KERN_ERR "%03x:", offset); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1492 | for (i = 0; i < limit; i++) { |
| 1493 | printk(" %02x", (unsigned char)data[offset + i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | } |
| 1495 | printk("\n"); |
| 1496 | } |
| 1497 | #endif |
| 1498 | |
| 1499 | #if DEBUG |
| 1500 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1501 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | { |
| 1503 | int i, size; |
| 1504 | char *realobj; |
| 1505 | |
| 1506 | if (cachep->flags & SLAB_RED_ZONE) { |
| 1507 | printk(KERN_ERR "Redzone: 0x%lx/0x%lx.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1508 | *dbg_redzone1(cachep, objp), |
| 1509 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | } |
| 1511 | |
| 1512 | if (cachep->flags & SLAB_STORE_USER) { |
| 1513 | printk(KERN_ERR "Last user: [<%p>]", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1514 | *dbg_userword(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1515 | print_symbol("(%s)", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1516 | (unsigned long)*dbg_userword(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1517 | printk("\n"); |
| 1518 | } |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1519 | realobj = (char *)objp + obj_offset(cachep); |
| 1520 | size = obj_size(cachep); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1521 | for (i = 0; i < size && lines; i += 16, lines--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | int limit; |
| 1523 | limit = 16; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1524 | if (i + limit > size) |
| 1525 | limit = size - i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1526 | dump_line(realobj, i, limit); |
| 1527 | } |
| 1528 | } |
| 1529 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1530 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | { |
| 1532 | char *realobj; |
| 1533 | int size, i; |
| 1534 | int lines = 0; |
| 1535 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1536 | realobj = (char *)objp + obj_offset(cachep); |
| 1537 | size = obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1538 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1539 | for (i = 0; i < size; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1540 | char exp = POISON_FREE; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1541 | if (i == size - 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1542 | exp = POISON_END; |
| 1543 | if (realobj[i] != exp) { |
| 1544 | int limit; |
| 1545 | /* Mismatch ! */ |
| 1546 | /* Print header */ |
| 1547 | if (lines == 0) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1548 | printk(KERN_ERR |
| 1549 | "Slab corruption: start=%p, len=%d\n", |
| 1550 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1551 | print_objinfo(cachep, objp, 0); |
| 1552 | } |
| 1553 | /* Hexdump the affected line */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1554 | i = (i / 16) * 16; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1555 | limit = 16; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1556 | if (i + limit > size) |
| 1557 | limit = size - i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1558 | dump_line(realobj, i, limit); |
| 1559 | i += 16; |
| 1560 | lines++; |
| 1561 | /* Limit to 5 lines */ |
| 1562 | if (lines > 5) |
| 1563 | break; |
| 1564 | } |
| 1565 | } |
| 1566 | if (lines != 0) { |
| 1567 | /* Print some data about the neighboring objects, if they |
| 1568 | * exist: |
| 1569 | */ |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1570 | struct slab *slabp = virt_to_slab(objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1571 | int objnr; |
| 1572 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1573 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | if (objnr) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1575 | objp = slabp->s_mem + (objnr - 1) * cachep->buffer_size; |
| 1576 | realobj = (char *)objp + obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1578 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1579 | print_objinfo(cachep, objp, 2); |
| 1580 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1581 | if (objnr + 1 < cachep->num) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1582 | objp = slabp->s_mem + (objnr + 1) * cachep->buffer_size; |
| 1583 | realobj = (char *)objp + obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1584 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1585 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | print_objinfo(cachep, objp, 2); |
| 1587 | } |
| 1588 | } |
| 1589 | } |
| 1590 | #endif |
| 1591 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1592 | #if DEBUG |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1593 | /** |
| 1594 | * slab_destroy_objs - call the registered destructor for each object in |
| 1595 | * a slab that is to be destroyed. |
| 1596 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1597 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1598 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1599 | int i; |
| 1600 | for (i = 0; i < cachep->num; i++) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1601 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1602 | |
| 1603 | if (cachep->flags & SLAB_POISON) { |
| 1604 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1605 | if ((cachep->buffer_size % PAGE_SIZE) == 0 |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1606 | && OFF_SLAB(cachep)) |
| 1607 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1608 | cachep->buffer_size / PAGE_SIZE, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1609 | 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1610 | else |
| 1611 | check_poison_obj(cachep, objp); |
| 1612 | #else |
| 1613 | check_poison_obj(cachep, objp); |
| 1614 | #endif |
| 1615 | } |
| 1616 | if (cachep->flags & SLAB_RED_ZONE) { |
| 1617 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 1618 | slab_error(cachep, "start of a freed object " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1619 | "was overwritten"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1620 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 1621 | slab_error(cachep, "end of a freed object " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1622 | "was overwritten"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1623 | } |
| 1624 | if (cachep->dtor && !(cachep->flags & SLAB_POISON)) |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1625 | (cachep->dtor) (objp + obj_offset(cachep), cachep, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | } |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1627 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | #else |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1629 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1630 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1631 | if (cachep->dtor) { |
| 1632 | int i; |
| 1633 | for (i = 0; i < cachep->num; i++) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1634 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1635 | (cachep->dtor) (objp, cachep, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | } |
| 1637 | } |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1638 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1639 | #endif |
| 1640 | |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1641 | /** |
| 1642 | * Destroy all the objs in a slab, and release the mem back to the system. |
| 1643 | * Before calling the slab must have been unlinked from the cache. |
| 1644 | * The cache-lock is not held/needed. |
| 1645 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1646 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1647 | { |
| 1648 | void *addr = slabp->s_mem - slabp->colouroff; |
| 1649 | |
| 1650 | slab_destroy_objs(cachep, slabp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1651 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
| 1652 | struct slab_rcu *slab_rcu; |
| 1653 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1654 | slab_rcu = (struct slab_rcu *)slabp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1655 | slab_rcu->cachep = cachep; |
| 1656 | slab_rcu->addr = addr; |
| 1657 | call_rcu(&slab_rcu->head, kmem_rcu_free); |
| 1658 | } else { |
| 1659 | kmem_freepages(cachep, addr); |
| 1660 | if (OFF_SLAB(cachep)) |
| 1661 | kmem_cache_free(cachep->slabp_cache, slabp); |
| 1662 | } |
| 1663 | } |
| 1664 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1665 | /* For setting up all the kmem_list3s for cache whose buffer_size is same |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1666 | as size of kmem_list3. */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1667 | static void set_up_list3s(struct kmem_cache *cachep, int index) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1668 | { |
| 1669 | int node; |
| 1670 | |
| 1671 | for_each_online_node(node) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1672 | cachep->nodelists[node] = &initkmem_list3[index + node]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1673 | cachep->nodelists[node]->next_reap = jiffies + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1674 | REAPTIMEOUT_LIST3 + |
| 1675 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1676 | } |
| 1677 | } |
| 1678 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1679 | /** |
Randy.Dunlap | a70773d | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 1680 | * calculate_slab_order - calculate size (page order) of slabs |
| 1681 | * @cachep: pointer to the cache that is being created |
| 1682 | * @size: size of objects to be created in this cache. |
| 1683 | * @align: required alignment for the objects. |
| 1684 | * @flags: slab allocation flags |
| 1685 | * |
| 1686 | * Also calculates the number of objects per slab. |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1687 | * |
| 1688 | * This could be made much more intelligent. For now, try to avoid using |
| 1689 | * high order pages for slabs. When the gfp() functions are more friendly |
| 1690 | * towards high-order requests, this should be changed. |
| 1691 | */ |
Randy Dunlap | ee13d78 | 2006-02-01 03:05:53 -0800 | [diff] [blame] | 1692 | static inline size_t calculate_slab_order(struct kmem_cache *cachep, |
| 1693 | size_t size, size_t align, unsigned long flags) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1694 | { |
| 1695 | size_t left_over = 0; |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1696 | int gfporder; |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1697 | |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1698 | for (gfporder = 0 ; gfporder <= MAX_GFP_ORDER; gfporder++) { |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1699 | unsigned int num; |
| 1700 | size_t remainder; |
| 1701 | |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1702 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1703 | if (!num) |
| 1704 | continue; |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1705 | |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1706 | /* More than offslab_limit objects will cause problems */ |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1707 | if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1708 | break; |
| 1709 | |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1710 | /* Found something acceptable - save it away */ |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1711 | cachep->num = num; |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1712 | cachep->gfporder = gfporder; |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1713 | left_over = remainder; |
| 1714 | |
| 1715 | /* |
Linus Torvalds | f78bb8a | 2006-03-08 10:33:05 -0800 | [diff] [blame] | 1716 | * A VFS-reclaimable slab tends to have most allocations |
| 1717 | * as GFP_NOFS and we really don't want to have to be allocating |
| 1718 | * higher-order pages when we are unable to shrink dcache. |
| 1719 | */ |
| 1720 | if (flags & SLAB_RECLAIM_ACCOUNT) |
| 1721 | break; |
| 1722 | |
| 1723 | /* |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1724 | * Large number of objects is good, but very large slabs are |
| 1725 | * currently bad for the gfp()s. |
| 1726 | */ |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1727 | if (gfporder >= slab_break_gfp_order) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1728 | break; |
| 1729 | |
Linus Torvalds | 9888e6f | 2006-03-06 17:44:43 -0800 | [diff] [blame] | 1730 | /* |
| 1731 | * Acceptable internal fragmentation? |
| 1732 | */ |
| 1733 | if ((left_over * 8) <= (PAGE_SIZE << gfporder)) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1734 | break; |
| 1735 | } |
| 1736 | return left_over; |
| 1737 | } |
| 1738 | |
| 1739 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1740 | * kmem_cache_create - Create a cache. |
| 1741 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 1742 | * @size: The size of objects to be created in this cache. |
| 1743 | * @align: The required alignment for the objects. |
| 1744 | * @flags: SLAB flags |
| 1745 | * @ctor: A constructor for the objects. |
| 1746 | * @dtor: A destructor for the objects. |
| 1747 | * |
| 1748 | * Returns a ptr to the cache on success, NULL on failure. |
| 1749 | * Cannot be called within a int, but can be interrupted. |
| 1750 | * The @ctor is run when new pages are allocated by the cache |
| 1751 | * and the @dtor is run before the pages are handed back. |
| 1752 | * |
| 1753 | * @name must be valid until the cache is destroyed. This implies that |
| 1754 | * the module calling this has to destroy the cache before getting |
| 1755 | * unloaded. |
| 1756 | * |
| 1757 | * The flags are |
| 1758 | * |
| 1759 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 1760 | * to catch references to uninitialised memory. |
| 1761 | * |
| 1762 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check |
| 1763 | * for buffer overruns. |
| 1764 | * |
| 1765 | * %SLAB_NO_REAP - Don't automatically reap this cache when we're under |
| 1766 | * memory pressure. |
| 1767 | * |
| 1768 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 1769 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 1770 | * as davem. |
| 1771 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1772 | struct kmem_cache * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | kmem_cache_create (const char *name, size_t size, size_t align, |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1774 | unsigned long flags, void (*ctor)(void*, struct kmem_cache *, unsigned long), |
| 1775 | void (*dtor)(void*, struct kmem_cache *, unsigned long)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1776 | { |
| 1777 | size_t left_over, slab_size, ralign; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1778 | struct kmem_cache *cachep = NULL; |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1779 | struct list_head *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1780 | |
| 1781 | /* |
| 1782 | * Sanity checks... these are all serious usage bugs. |
| 1783 | */ |
| 1784 | if ((!name) || |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1785 | in_interrupt() || |
| 1786 | (size < BYTES_PER_WORD) || |
| 1787 | (size > (1 << MAX_OBJ_ORDER) * PAGE_SIZE) || (dtor && !ctor)) { |
| 1788 | printk(KERN_ERR "%s: Early error in slab %s\n", |
| 1789 | __FUNCTION__, name); |
| 1790 | BUG(); |
| 1791 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1792 | |
Ravikiran G Thirumalai | f0188f4 | 2006-02-10 01:51:13 -0800 | [diff] [blame] | 1793 | /* |
| 1794 | * Prevent CPUs from coming and going. |
| 1795 | * lock_cpu_hotplug() nests outside cache_chain_mutex |
| 1796 | */ |
| 1797 | lock_cpu_hotplug(); |
| 1798 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1799 | mutex_lock(&cache_chain_mutex); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1800 | |
| 1801 | list_for_each(p, &cache_chain) { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1802 | struct kmem_cache *pc = list_entry(p, struct kmem_cache, next); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1803 | mm_segment_t old_fs = get_fs(); |
| 1804 | char tmp; |
| 1805 | int res; |
| 1806 | |
| 1807 | /* |
| 1808 | * This happens when the module gets unloaded and doesn't |
| 1809 | * destroy its slab cache and no-one else reuses the vmalloc |
| 1810 | * area of the module. Print a warning. |
| 1811 | */ |
| 1812 | set_fs(KERNEL_DS); |
| 1813 | res = __get_user(tmp, pc->name); |
| 1814 | set_fs(old_fs); |
| 1815 | if (res) { |
| 1816 | printk("SLAB: cache with size %d has lost its name\n", |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1817 | pc->buffer_size); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1818 | continue; |
| 1819 | } |
| 1820 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1821 | if (!strcmp(pc->name, name)) { |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1822 | printk("kmem_cache_create: duplicate cache %s\n", name); |
| 1823 | dump_stack(); |
| 1824 | goto oops; |
| 1825 | } |
| 1826 | } |
| 1827 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1828 | #if DEBUG |
| 1829 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
| 1830 | if ((flags & SLAB_DEBUG_INITIAL) && !ctor) { |
| 1831 | /* No constructor, but inital state check requested */ |
| 1832 | printk(KERN_ERR "%s: No con, but init state check " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1833 | "requested - %s\n", __FUNCTION__, name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1834 | flags &= ~SLAB_DEBUG_INITIAL; |
| 1835 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1836 | #if FORCED_DEBUG |
| 1837 | /* |
| 1838 | * Enable redzoning and last user accounting, except for caches with |
| 1839 | * large objects, if the increased size would increase the object size |
| 1840 | * above the next power of two: caches with object sizes just above a |
| 1841 | * power of two have a significant amount of internal fragmentation. |
| 1842 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1843 | if ((size < 4096 |
| 1844 | || fls(size - 1) == fls(size - 1 + 3 * BYTES_PER_WORD))) |
| 1845 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1846 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
| 1847 | flags |= SLAB_POISON; |
| 1848 | #endif |
| 1849 | if (flags & SLAB_DESTROY_BY_RCU) |
| 1850 | BUG_ON(flags & SLAB_POISON); |
| 1851 | #endif |
| 1852 | if (flags & SLAB_DESTROY_BY_RCU) |
| 1853 | BUG_ON(dtor); |
| 1854 | |
| 1855 | /* |
| 1856 | * Always checks flags, a caller might be expecting debug |
| 1857 | * support which isn't available. |
| 1858 | */ |
| 1859 | if (flags & ~CREATE_MASK) |
| 1860 | BUG(); |
| 1861 | |
| 1862 | /* Check that size is in terms of words. This is needed to avoid |
| 1863 | * unaligned accesses for some archs when redzoning is used, and makes |
| 1864 | * sure any on-slab bufctl's are also correctly aligned. |
| 1865 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1866 | if (size & (BYTES_PER_WORD - 1)) { |
| 1867 | size += (BYTES_PER_WORD - 1); |
| 1868 | size &= ~(BYTES_PER_WORD - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1869 | } |
| 1870 | |
| 1871 | /* calculate out the final buffer alignment: */ |
| 1872 | /* 1) arch recommendation: can be overridden for debug */ |
| 1873 | if (flags & SLAB_HWCACHE_ALIGN) { |
| 1874 | /* Default alignment: as specified by the arch code. |
| 1875 | * Except if an object is really small, then squeeze multiple |
| 1876 | * objects into one cacheline. |
| 1877 | */ |
| 1878 | ralign = cache_line_size(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1879 | while (size <= ralign / 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1880 | ralign /= 2; |
| 1881 | } else { |
| 1882 | ralign = BYTES_PER_WORD; |
| 1883 | } |
| 1884 | /* 2) arch mandated alignment: disables debug if necessary */ |
| 1885 | if (ralign < ARCH_SLAB_MINALIGN) { |
| 1886 | ralign = ARCH_SLAB_MINALIGN; |
| 1887 | if (ralign > BYTES_PER_WORD) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1888 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1889 | } |
| 1890 | /* 3) caller mandated alignment: disables debug if necessary */ |
| 1891 | if (ralign < align) { |
| 1892 | ralign = align; |
| 1893 | if (ralign > BYTES_PER_WORD) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1894 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1895 | } |
| 1896 | /* 4) Store it. Note that the debug code below can reduce |
| 1897 | * the alignment to BYTES_PER_WORD. |
| 1898 | */ |
| 1899 | align = ralign; |
| 1900 | |
| 1901 | /* Get cache's description obj. */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1902 | cachep = kmem_cache_alloc(&cache_cache, SLAB_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1903 | if (!cachep) |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1904 | goto oops; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1905 | memset(cachep, 0, sizeof(struct kmem_cache)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1906 | |
| 1907 | #if DEBUG |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1908 | cachep->obj_size = size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1909 | |
| 1910 | if (flags & SLAB_RED_ZONE) { |
| 1911 | /* redzoning only works with word aligned caches */ |
| 1912 | align = BYTES_PER_WORD; |
| 1913 | |
| 1914 | /* add space for red zone words */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1915 | cachep->obj_offset += BYTES_PER_WORD; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1916 | size += 2 * BYTES_PER_WORD; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1917 | } |
| 1918 | if (flags & SLAB_STORE_USER) { |
| 1919 | /* user store requires word alignment and |
| 1920 | * one word storage behind the end of the real |
| 1921 | * object. |
| 1922 | */ |
| 1923 | align = BYTES_PER_WORD; |
| 1924 | size += BYTES_PER_WORD; |
| 1925 | } |
| 1926 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1927 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1928 | && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { |
| 1929 | cachep->obj_offset += PAGE_SIZE - size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1930 | size = PAGE_SIZE; |
| 1931 | } |
| 1932 | #endif |
| 1933 | #endif |
| 1934 | |
| 1935 | /* Determine if the slab management is 'on' or 'off' slab. */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1936 | if (size >= (PAGE_SIZE >> 3)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1937 | /* |
| 1938 | * Size is large, assume best to place the slab management obj |
| 1939 | * off-slab (should allow better packing of objs). |
| 1940 | */ |
| 1941 | flags |= CFLGS_OFF_SLAB; |
| 1942 | |
| 1943 | size = ALIGN(size, align); |
| 1944 | |
Linus Torvalds | f78bb8a | 2006-03-08 10:33:05 -0800 | [diff] [blame] | 1945 | left_over = calculate_slab_order(cachep, size, align, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1946 | |
| 1947 | if (!cachep->num) { |
| 1948 | printk("kmem_cache_create: couldn't create cache %s.\n", name); |
| 1949 | kmem_cache_free(&cache_cache, cachep); |
| 1950 | cachep = NULL; |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1951 | goto oops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1952 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1953 | slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) |
| 1954 | + sizeof(struct slab), align); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1955 | |
| 1956 | /* |
| 1957 | * If the slab has been placed off-slab, and we have enough space then |
| 1958 | * move it on-slab. This is at the expense of any extra colouring. |
| 1959 | */ |
| 1960 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { |
| 1961 | flags &= ~CFLGS_OFF_SLAB; |
| 1962 | left_over -= slab_size; |
| 1963 | } |
| 1964 | |
| 1965 | if (flags & CFLGS_OFF_SLAB) { |
| 1966 | /* really off slab. No need for manual alignment */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1967 | slab_size = |
| 1968 | cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1969 | } |
| 1970 | |
| 1971 | cachep->colour_off = cache_line_size(); |
| 1972 | /* Offset must be a multiple of the alignment. */ |
| 1973 | if (cachep->colour_off < align) |
| 1974 | cachep->colour_off = align; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1975 | cachep->colour = left_over / cachep->colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1976 | cachep->slab_size = slab_size; |
| 1977 | cachep->flags = flags; |
| 1978 | cachep->gfpflags = 0; |
| 1979 | if (flags & SLAB_CACHE_DMA) |
| 1980 | cachep->gfpflags |= GFP_DMA; |
| 1981 | spin_lock_init(&cachep->spinlock); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1982 | cachep->buffer_size = size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1983 | |
| 1984 | if (flags & CFLGS_OFF_SLAB) |
Victor Fusco | b2d5507 | 2005-09-10 00:26:36 -0700 | [diff] [blame] | 1985 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1986 | cachep->ctor = ctor; |
| 1987 | cachep->dtor = dtor; |
| 1988 | cachep->name = name; |
| 1989 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1990 | |
| 1991 | if (g_cpucache_up == FULL) { |
| 1992 | enable_cpucache(cachep); |
| 1993 | } else { |
| 1994 | if (g_cpucache_up == NONE) { |
| 1995 | /* Note: the first kmem_cache_create must create |
| 1996 | * the cache that's used by kmalloc(24), otherwise |
| 1997 | * the creation of further caches will BUG(). |
| 1998 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1999 | cachep->array[smp_processor_id()] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2000 | &initarray_generic.cache; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2001 | |
| 2002 | /* If the cache that's used by |
| 2003 | * kmalloc(sizeof(kmem_list3)) is the first cache, |
| 2004 | * then we need to set up all its list3s, otherwise |
| 2005 | * the creation of further caches will BUG(). |
| 2006 | */ |
| 2007 | set_up_list3s(cachep, SIZE_AC); |
| 2008 | if (INDEX_AC == INDEX_L3) |
| 2009 | g_cpucache_up = PARTIAL_L3; |
| 2010 | else |
| 2011 | g_cpucache_up = PARTIAL_AC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2012 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2013 | cachep->array[smp_processor_id()] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2014 | kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2015 | |
| 2016 | if (g_cpucache_up == PARTIAL_AC) { |
| 2017 | set_up_list3s(cachep, SIZE_L3); |
| 2018 | g_cpucache_up = PARTIAL_L3; |
| 2019 | } else { |
| 2020 | int node; |
| 2021 | for_each_online_node(node) { |
| 2022 | |
| 2023 | cachep->nodelists[node] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2024 | kmalloc_node(sizeof |
| 2025 | (struct kmem_list3), |
| 2026 | GFP_KERNEL, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2027 | BUG_ON(!cachep->nodelists[node]); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2028 | kmem_list3_init(cachep-> |
| 2029 | nodelists[node]); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2030 | } |
| 2031 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2033 | cachep->nodelists[numa_node_id()]->next_reap = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2034 | jiffies + REAPTIMEOUT_LIST3 + |
| 2035 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2036 | |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2037 | BUG_ON(!cpu_cache_get(cachep)); |
| 2038 | cpu_cache_get(cachep)->avail = 0; |
| 2039 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; |
| 2040 | cpu_cache_get(cachep)->batchcount = 1; |
| 2041 | cpu_cache_get(cachep)->touched = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | cachep->batchcount = 1; |
| 2043 | cachep->limit = BOOT_CPUCACHE_ENTRIES; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2044 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2045 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2046 | /* cache setup completed, link it into the list */ |
| 2047 | list_add(&cachep->next, &cache_chain); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2048 | oops: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2049 | if (!cachep && (flags & SLAB_PANIC)) |
| 2050 | panic("kmem_cache_create(): failed to create slab `%s'\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2051 | name); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2052 | mutex_unlock(&cache_chain_mutex); |
Ravikiran G Thirumalai | f0188f4 | 2006-02-10 01:51:13 -0800 | [diff] [blame] | 2053 | unlock_cpu_hotplug(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2054 | return cachep; |
| 2055 | } |
| 2056 | EXPORT_SYMBOL(kmem_cache_create); |
| 2057 | |
| 2058 | #if DEBUG |
| 2059 | static void check_irq_off(void) |
| 2060 | { |
| 2061 | BUG_ON(!irqs_disabled()); |
| 2062 | } |
| 2063 | |
| 2064 | static void check_irq_on(void) |
| 2065 | { |
| 2066 | BUG_ON(irqs_disabled()); |
| 2067 | } |
| 2068 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2069 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2070 | { |
| 2071 | #ifdef CONFIG_SMP |
| 2072 | check_irq_off(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2073 | assert_spin_locked(&cachep->nodelists[numa_node_id()]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2074 | #endif |
| 2075 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2076 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2077 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2078 | { |
| 2079 | #ifdef CONFIG_SMP |
| 2080 | check_irq_off(); |
| 2081 | assert_spin_locked(&cachep->nodelists[node]->list_lock); |
| 2082 | #endif |
| 2083 | } |
| 2084 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2085 | #else |
| 2086 | #define check_irq_off() do { } while(0) |
| 2087 | #define check_irq_on() do { } while(0) |
| 2088 | #define check_spinlock_acquired(x) do { } while(0) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2089 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2090 | #endif |
| 2091 | |
| 2092 | /* |
| 2093 | * Waits for all CPUs to execute func(). |
| 2094 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2095 | static void smp_call_function_all_cpus(void (*func)(void *arg), void *arg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2096 | { |
| 2097 | check_irq_on(); |
| 2098 | preempt_disable(); |
| 2099 | |
| 2100 | local_irq_disable(); |
| 2101 | func(arg); |
| 2102 | local_irq_enable(); |
| 2103 | |
| 2104 | if (smp_call_function(func, arg, 1, 1)) |
| 2105 | BUG(); |
| 2106 | |
| 2107 | preempt_enable(); |
| 2108 | } |
| 2109 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2110 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2111 | int force, int node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2112 | |
| 2113 | static void do_drain(void *arg) |
| 2114 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2115 | struct kmem_cache *cachep = (struct kmem_cache *) arg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2116 | struct array_cache *ac; |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2117 | int node = numa_node_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2118 | |
| 2119 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2120 | ac = cpu_cache_get(cachep); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2121 | spin_lock(&cachep->nodelists[node]->list_lock); |
| 2122 | free_block(cachep, ac->entry, ac->avail, node); |
| 2123 | spin_unlock(&cachep->nodelists[node]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2124 | ac->avail = 0; |
| 2125 | } |
| 2126 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2127 | static void drain_cpu_caches(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2128 | { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2129 | struct kmem_list3 *l3; |
| 2130 | int node; |
| 2131 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2132 | smp_call_function_all_cpus(do_drain, cachep); |
| 2133 | check_irq_on(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2134 | for_each_online_node(node) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2135 | l3 = cachep->nodelists[node]; |
| 2136 | if (l3) { |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 2137 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2138 | drain_array_locked(cachep, l3->shared, 1, node); |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 2139 | spin_unlock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2140 | if (l3->alien) |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 2141 | drain_alien_cache(cachep, l3->alien); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2142 | } |
| 2143 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | } |
| 2145 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2146 | static int __node_shrink(struct kmem_cache *cachep, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2147 | { |
| 2148 | struct slab *slabp; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2149 | struct kmem_list3 *l3 = cachep->nodelists[node]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2150 | int ret; |
| 2151 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2152 | for (;;) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | struct list_head *p; |
| 2154 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2155 | p = l3->slabs_free.prev; |
| 2156 | if (p == &l3->slabs_free) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2157 | break; |
| 2158 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2159 | slabp = list_entry(l3->slabs_free.prev, struct slab, list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2160 | #if DEBUG |
| 2161 | if (slabp->inuse) |
| 2162 | BUG(); |
| 2163 | #endif |
| 2164 | list_del(&slabp->list); |
| 2165 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2166 | l3->free_objects -= cachep->num; |
| 2167 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | slab_destroy(cachep, slabp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2169 | spin_lock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2170 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2171 | ret = !list_empty(&l3->slabs_full) || !list_empty(&l3->slabs_partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2172 | return ret; |
| 2173 | } |
| 2174 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2175 | static int __cache_shrink(struct kmem_cache *cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2176 | { |
| 2177 | int ret = 0, i = 0; |
| 2178 | struct kmem_list3 *l3; |
| 2179 | |
| 2180 | drain_cpu_caches(cachep); |
| 2181 | |
| 2182 | check_irq_on(); |
| 2183 | for_each_online_node(i) { |
| 2184 | l3 = cachep->nodelists[i]; |
| 2185 | if (l3) { |
| 2186 | spin_lock_irq(&l3->list_lock); |
| 2187 | ret += __node_shrink(cachep, i); |
| 2188 | spin_unlock_irq(&l3->list_lock); |
| 2189 | } |
| 2190 | } |
| 2191 | return (ret ? 1 : 0); |
| 2192 | } |
| 2193 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | /** |
| 2195 | * kmem_cache_shrink - Shrink a cache. |
| 2196 | * @cachep: The cache to shrink. |
| 2197 | * |
| 2198 | * Releases as many slabs as possible for a cache. |
| 2199 | * To help debugging, a zero exit status indicates all slabs were released. |
| 2200 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2201 | int kmem_cache_shrink(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2202 | { |
| 2203 | if (!cachep || in_interrupt()) |
| 2204 | BUG(); |
| 2205 | |
| 2206 | return __cache_shrink(cachep); |
| 2207 | } |
| 2208 | EXPORT_SYMBOL(kmem_cache_shrink); |
| 2209 | |
| 2210 | /** |
| 2211 | * kmem_cache_destroy - delete a cache |
| 2212 | * @cachep: the cache to destroy |
| 2213 | * |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2214 | * Remove a struct kmem_cache object from the slab cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2215 | * Returns 0 on success. |
| 2216 | * |
| 2217 | * It is expected this function will be called by a module when it is |
| 2218 | * unloaded. This will remove the cache completely, and avoid a duplicate |
| 2219 | * cache being allocated each time a module is loaded and unloaded, if the |
| 2220 | * module doesn't have persistent in-kernel storage across loads and unloads. |
| 2221 | * |
| 2222 | * The cache must be empty before calling this function. |
| 2223 | * |
| 2224 | * The caller must guarantee that noone will allocate memory from the cache |
| 2225 | * during the kmem_cache_destroy(). |
| 2226 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2227 | int kmem_cache_destroy(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2228 | { |
| 2229 | int i; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2230 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2231 | |
| 2232 | if (!cachep || in_interrupt()) |
| 2233 | BUG(); |
| 2234 | |
| 2235 | /* Don't let CPUs to come and go */ |
| 2236 | lock_cpu_hotplug(); |
| 2237 | |
| 2238 | /* Find the cache in the chain of caches. */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2239 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2240 | /* |
| 2241 | * the chain is never empty, cache_cache is never destroyed |
| 2242 | */ |
| 2243 | list_del(&cachep->next); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2244 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2245 | |
| 2246 | if (__cache_shrink(cachep)) { |
| 2247 | slab_error(cachep, "Can't free all objects"); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2248 | mutex_lock(&cache_chain_mutex); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2249 | list_add(&cachep->next, &cache_chain); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2250 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2251 | unlock_cpu_hotplug(); |
| 2252 | return 1; |
| 2253 | } |
| 2254 | |
| 2255 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
Paul E. McKenney | fbd568a3e | 2005-05-01 08:59:04 -0700 | [diff] [blame] | 2256 | synchronize_rcu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2257 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2258 | for_each_online_cpu(i) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2259 | kfree(cachep->array[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2260 | |
| 2261 | /* NUMA: free the list3 structures */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2262 | for_each_online_node(i) { |
| 2263 | if ((l3 = cachep->nodelists[i])) { |
| 2264 | kfree(l3->shared); |
| 2265 | free_alien_cache(l3->alien); |
| 2266 | kfree(l3); |
| 2267 | } |
| 2268 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2269 | kmem_cache_free(&cache_cache, cachep); |
| 2270 | |
| 2271 | unlock_cpu_hotplug(); |
| 2272 | |
| 2273 | return 0; |
| 2274 | } |
| 2275 | EXPORT_SYMBOL(kmem_cache_destroy); |
| 2276 | |
| 2277 | /* Get the memory for a slab management obj. */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2278 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2279 | int colour_off, gfp_t local_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2280 | { |
| 2281 | struct slab *slabp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2282 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2283 | if (OFF_SLAB(cachep)) { |
| 2284 | /* Slab management obj is off-slab. */ |
| 2285 | slabp = kmem_cache_alloc(cachep->slabp_cache, local_flags); |
| 2286 | if (!slabp) |
| 2287 | return NULL; |
| 2288 | } else { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2289 | slabp = objp + colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2290 | colour_off += cachep->slab_size; |
| 2291 | } |
| 2292 | slabp->inuse = 0; |
| 2293 | slabp->colouroff = colour_off; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2294 | slabp->s_mem = objp + colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2295 | |
| 2296 | return slabp; |
| 2297 | } |
| 2298 | |
| 2299 | static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) |
| 2300 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2301 | return (kmem_bufctl_t *) (slabp + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2302 | } |
| 2303 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2304 | static void cache_init_objs(struct kmem_cache *cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2305 | struct slab *slabp, unsigned long ctor_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2306 | { |
| 2307 | int i; |
| 2308 | |
| 2309 | for (i = 0; i < cachep->num; i++) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2310 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2311 | #if DEBUG |
| 2312 | /* need to poison the objs? */ |
| 2313 | if (cachep->flags & SLAB_POISON) |
| 2314 | poison_obj(cachep, objp, POISON_FREE); |
| 2315 | if (cachep->flags & SLAB_STORE_USER) |
| 2316 | *dbg_userword(cachep, objp) = NULL; |
| 2317 | |
| 2318 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2319 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2320 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2321 | } |
| 2322 | /* |
| 2323 | * Constructors are not allowed to allocate memory from |
| 2324 | * the same cache which they are a constructor for. |
| 2325 | * Otherwise, deadlock. They must also be threaded. |
| 2326 | */ |
| 2327 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2328 | cachep->ctor(objp + obj_offset(cachep), cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2329 | ctor_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2330 | |
| 2331 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2332 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 2333 | slab_error(cachep, "constructor overwrote the" |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2334 | " end of an object"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2335 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 2336 | slab_error(cachep, "constructor overwrote the" |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2337 | " start of an object"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2338 | } |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2339 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2340 | && cachep->flags & SLAB_POISON) |
| 2341 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2342 | cachep->buffer_size / PAGE_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2343 | #else |
| 2344 | if (cachep->ctor) |
| 2345 | cachep->ctor(objp, cachep, ctor_flags); |
| 2346 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2347 | slab_bufctl(slabp)[i] = i + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2349 | slab_bufctl(slabp)[i - 1] = BUFCTL_END; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2350 | slabp->free = 0; |
| 2351 | } |
| 2352 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2353 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2354 | { |
| 2355 | if (flags & SLAB_DMA) { |
| 2356 | if (!(cachep->gfpflags & GFP_DMA)) |
| 2357 | BUG(); |
| 2358 | } else { |
| 2359 | if (cachep->gfpflags & GFP_DMA) |
| 2360 | BUG(); |
| 2361 | } |
| 2362 | } |
| 2363 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2364 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, int nodeid) |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2365 | { |
| 2366 | void *objp = slabp->s_mem + (slabp->free * cachep->buffer_size); |
| 2367 | kmem_bufctl_t next; |
| 2368 | |
| 2369 | slabp->inuse++; |
| 2370 | next = slab_bufctl(slabp)[slabp->free]; |
| 2371 | #if DEBUG |
| 2372 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; |
| 2373 | WARN_ON(slabp->nodeid != nodeid); |
| 2374 | #endif |
| 2375 | slabp->free = next; |
| 2376 | |
| 2377 | return objp; |
| 2378 | } |
| 2379 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2380 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, void *objp, |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2381 | int nodeid) |
| 2382 | { |
| 2383 | unsigned int objnr = (unsigned)(objp-slabp->s_mem) / cachep->buffer_size; |
| 2384 | |
| 2385 | #if DEBUG |
| 2386 | /* Verify that the slab belongs to the intended node */ |
| 2387 | WARN_ON(slabp->nodeid != nodeid); |
| 2388 | |
| 2389 | if (slab_bufctl(slabp)[objnr] != BUFCTL_FREE) { |
| 2390 | printk(KERN_ERR "slab: double free detected in cache " |
| 2391 | "'%s', objp %p\n", cachep->name, objp); |
| 2392 | BUG(); |
| 2393 | } |
| 2394 | #endif |
| 2395 | slab_bufctl(slabp)[objnr] = slabp->free; |
| 2396 | slabp->free = objnr; |
| 2397 | slabp->inuse--; |
| 2398 | } |
| 2399 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2400 | static void set_slab_attr(struct kmem_cache *cachep, struct slab *slabp, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2401 | { |
| 2402 | int i; |
| 2403 | struct page *page; |
| 2404 | |
| 2405 | /* Nasty!!!!!! I hope this is OK. */ |
| 2406 | i = 1 << cachep->gfporder; |
| 2407 | page = virt_to_page(objp); |
| 2408 | do { |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 2409 | page_set_cache(page, cachep); |
| 2410 | page_set_slab(page, slabp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2411 | page++; |
| 2412 | } while (--i); |
| 2413 | } |
| 2414 | |
| 2415 | /* |
| 2416 | * Grow (by 1) the number of slabs within a cache. This is called by |
| 2417 | * kmem_cache_alloc() when there are no active objs left in a cache. |
| 2418 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2419 | static int cache_grow(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2420 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2421 | struct slab *slabp; |
| 2422 | void *objp; |
| 2423 | size_t offset; |
| 2424 | gfp_t local_flags; |
| 2425 | unsigned long ctor_flags; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2426 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2427 | |
| 2428 | /* Be lazy and only check for valid flags here, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2429 | * keeping it out of the critical path in kmem_cache_alloc(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2430 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2431 | if (flags & ~(SLAB_DMA | SLAB_LEVEL_MASK | SLAB_NO_GROW)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2432 | BUG(); |
| 2433 | if (flags & SLAB_NO_GROW) |
| 2434 | return 0; |
| 2435 | |
| 2436 | ctor_flags = SLAB_CTOR_CONSTRUCTOR; |
| 2437 | local_flags = (flags & SLAB_LEVEL_MASK); |
| 2438 | if (!(local_flags & __GFP_WAIT)) |
| 2439 | /* |
| 2440 | * Not allowed to sleep. Need to tell a constructor about |
| 2441 | * this - it might need to know... |
| 2442 | */ |
| 2443 | ctor_flags |= SLAB_CTOR_ATOMIC; |
| 2444 | |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2445 | /* 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] | 2446 | check_irq_off(); |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2447 | l3 = cachep->nodelists[nodeid]; |
| 2448 | spin_lock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2449 | |
| 2450 | /* Get colour for the slab, and cal the next value. */ |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2451 | offset = l3->colour_next; |
| 2452 | l3->colour_next++; |
| 2453 | if (l3->colour_next >= cachep->colour) |
| 2454 | l3->colour_next = 0; |
| 2455 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2456 | |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2457 | offset *= cachep->colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2458 | |
| 2459 | if (local_flags & __GFP_WAIT) |
| 2460 | local_irq_enable(); |
| 2461 | |
| 2462 | /* |
| 2463 | * The test for missing atomic flag is performed here, rather than |
| 2464 | * the more obvious place, simply to reduce the critical path length |
| 2465 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they |
| 2466 | * will eventually be caught here (where it matters). |
| 2467 | */ |
| 2468 | kmem_flagcheck(cachep, flags); |
| 2469 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2470 | /* Get mem for the objs. |
| 2471 | * Attempt to allocate a physical page from 'nodeid', |
| 2472 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2473 | if (!(objp = kmem_getpages(cachep, flags, nodeid))) |
| 2474 | goto failed; |
| 2475 | |
| 2476 | /* Get slab management. */ |
| 2477 | if (!(slabp = alloc_slabmgmt(cachep, objp, offset, local_flags))) |
| 2478 | goto opps1; |
| 2479 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2480 | slabp->nodeid = nodeid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2481 | set_slab_attr(cachep, slabp, objp); |
| 2482 | |
| 2483 | cache_init_objs(cachep, slabp, ctor_flags); |
| 2484 | |
| 2485 | if (local_flags & __GFP_WAIT) |
| 2486 | local_irq_disable(); |
| 2487 | check_irq_off(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2488 | spin_lock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2489 | |
| 2490 | /* Make slab active. */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2491 | list_add_tail(&slabp->list, &(l3->slabs_free)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2492 | STATS_INC_GROWN(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2493 | l3->free_objects += cachep->num; |
| 2494 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2495 | return 1; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2496 | opps1: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2497 | kmem_freepages(cachep, objp); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2498 | failed: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2499 | if (local_flags & __GFP_WAIT) |
| 2500 | local_irq_disable(); |
| 2501 | return 0; |
| 2502 | } |
| 2503 | |
| 2504 | #if DEBUG |
| 2505 | |
| 2506 | /* |
| 2507 | * Perform extra freeing checks: |
| 2508 | * - detect bad pointers. |
| 2509 | * - POISON/RED_ZONE checking |
| 2510 | * - destructor calls, for caches with POISON+dtor |
| 2511 | */ |
| 2512 | static void kfree_debugcheck(const void *objp) |
| 2513 | { |
| 2514 | struct page *page; |
| 2515 | |
| 2516 | if (!virt_addr_valid(objp)) { |
| 2517 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2518 | (unsigned long)objp); |
| 2519 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2520 | } |
| 2521 | page = virt_to_page(objp); |
| 2522 | if (!PageSlab(page)) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2523 | printk(KERN_ERR "kfree_debugcheck: bad ptr %lxh.\n", |
| 2524 | (unsigned long)objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2525 | BUG(); |
| 2526 | } |
| 2527 | } |
| 2528 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2529 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2530 | void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2531 | { |
| 2532 | struct page *page; |
| 2533 | unsigned int objnr; |
| 2534 | struct slab *slabp; |
| 2535 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2536 | objp -= obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2537 | kfree_debugcheck(objp); |
| 2538 | page = virt_to_page(objp); |
| 2539 | |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 2540 | if (page_get_cache(page) != cachep) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2541 | printk(KERN_ERR |
| 2542 | "mismatch in kmem_cache_free: expected cache %p, got %p\n", |
| 2543 | page_get_cache(page), cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2544 | printk(KERN_ERR "%p is %s.\n", cachep, cachep->name); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2545 | printk(KERN_ERR "%p is %s.\n", page_get_cache(page), |
| 2546 | page_get_cache(page)->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2547 | WARN_ON(1); |
| 2548 | } |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 2549 | slabp = page_get_slab(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2550 | |
| 2551 | if (cachep->flags & SLAB_RED_ZONE) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2552 | if (*dbg_redzone1(cachep, objp) != RED_ACTIVE |
| 2553 | || *dbg_redzone2(cachep, objp) != RED_ACTIVE) { |
| 2554 | slab_error(cachep, |
| 2555 | "double free, or memory outside" |
| 2556 | " object was overwritten"); |
| 2557 | printk(KERN_ERR |
| 2558 | "%p: redzone 1: 0x%lx, redzone 2: 0x%lx.\n", |
| 2559 | objp, *dbg_redzone1(cachep, objp), |
| 2560 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2561 | } |
| 2562 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2563 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2564 | } |
| 2565 | if (cachep->flags & SLAB_STORE_USER) |
| 2566 | *dbg_userword(cachep, objp) = caller; |
| 2567 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2568 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2569 | |
| 2570 | BUG_ON(objnr >= cachep->num); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2571 | BUG_ON(objp != slabp->s_mem + objnr * cachep->buffer_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2572 | |
| 2573 | if (cachep->flags & SLAB_DEBUG_INITIAL) { |
| 2574 | /* Need to call the slab's constructor so the |
| 2575 | * caller can perform a verify of its state (debugging). |
| 2576 | * Called without the cache-lock held. |
| 2577 | */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2578 | cachep->ctor(objp + obj_offset(cachep), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2579 | cachep, SLAB_CTOR_CONSTRUCTOR | SLAB_CTOR_VERIFY); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2580 | } |
| 2581 | if (cachep->flags & SLAB_POISON && cachep->dtor) { |
| 2582 | /* we want to cache poison the object, |
| 2583 | * call the destruction callback |
| 2584 | */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2585 | cachep->dtor(objp + obj_offset(cachep), cachep, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2586 | } |
| 2587 | if (cachep->flags & SLAB_POISON) { |
| 2588 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2589 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2590 | store_stackinfo(cachep, objp, (unsigned long)caller); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2591 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2592 | cachep->buffer_size / PAGE_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2593 | } else { |
| 2594 | poison_obj(cachep, objp, POISON_FREE); |
| 2595 | } |
| 2596 | #else |
| 2597 | poison_obj(cachep, objp, POISON_FREE); |
| 2598 | #endif |
| 2599 | } |
| 2600 | return objp; |
| 2601 | } |
| 2602 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2603 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2604 | { |
| 2605 | kmem_bufctl_t i; |
| 2606 | int entries = 0; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2607 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2608 | /* Check slab's freelist to see if this obj is there. */ |
| 2609 | for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { |
| 2610 | entries++; |
| 2611 | if (entries > cachep->num || i >= cachep->num) |
| 2612 | goto bad; |
| 2613 | } |
| 2614 | if (entries != cachep->num - slabp->inuse) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2615 | bad: |
| 2616 | printk(KERN_ERR |
| 2617 | "slab: Internal list corruption detected in cache '%s'(%d), slabp %p(%d). Hexdump:\n", |
| 2618 | cachep->name, cachep->num, slabp, slabp->inuse); |
| 2619 | for (i = 0; |
Linus Torvalds | 264132b | 2006-03-06 12:10:07 -0800 | [diff] [blame] | 2620 | i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2621 | i++) { |
| 2622 | if ((i % 16) == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2623 | printk("\n%03x:", i); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2624 | printk(" %02x", ((unsigned char *)slabp)[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2625 | } |
| 2626 | printk("\n"); |
| 2627 | BUG(); |
| 2628 | } |
| 2629 | } |
| 2630 | #else |
| 2631 | #define kfree_debugcheck(x) do { } while(0) |
| 2632 | #define cache_free_debugcheck(x,objp,z) (objp) |
| 2633 | #define check_slabp(x,y) do { } while(0) |
| 2634 | #endif |
| 2635 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2636 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2637 | { |
| 2638 | int batchcount; |
| 2639 | struct kmem_list3 *l3; |
| 2640 | struct array_cache *ac; |
| 2641 | |
| 2642 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2643 | ac = cpu_cache_get(cachep); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2644 | retry: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2645 | batchcount = ac->batchcount; |
| 2646 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { |
| 2647 | /* if there was little recent activity on this |
| 2648 | * cache, then perform only a partial refill. |
| 2649 | * Otherwise we could generate refill bouncing. |
| 2650 | */ |
| 2651 | batchcount = BATCHREFILL_LIMIT; |
| 2652 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2653 | l3 = cachep->nodelists[numa_node_id()]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2654 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2655 | BUG_ON(ac->avail > 0 || !l3); |
| 2656 | spin_lock(&l3->list_lock); |
| 2657 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2658 | if (l3->shared) { |
| 2659 | struct array_cache *shared_array = l3->shared; |
| 2660 | if (shared_array->avail) { |
| 2661 | if (batchcount > shared_array->avail) |
| 2662 | batchcount = shared_array->avail; |
| 2663 | shared_array->avail -= batchcount; |
| 2664 | ac->avail = batchcount; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2665 | memcpy(ac->entry, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2666 | &(shared_array->entry[shared_array->avail]), |
| 2667 | sizeof(void *) * batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2668 | shared_array->touched = 1; |
| 2669 | goto alloc_done; |
| 2670 | } |
| 2671 | } |
| 2672 | while (batchcount > 0) { |
| 2673 | struct list_head *entry; |
| 2674 | struct slab *slabp; |
| 2675 | /* Get slab alloc is to come from. */ |
| 2676 | entry = l3->slabs_partial.next; |
| 2677 | if (entry == &l3->slabs_partial) { |
| 2678 | l3->free_touched = 1; |
| 2679 | entry = l3->slabs_free.next; |
| 2680 | if (entry == &l3->slabs_free) |
| 2681 | goto must_grow; |
| 2682 | } |
| 2683 | |
| 2684 | slabp = list_entry(entry, struct slab, list); |
| 2685 | check_slabp(cachep, slabp); |
| 2686 | check_spinlock_acquired(cachep); |
| 2687 | while (slabp->inuse < cachep->num && batchcount--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2688 | STATS_INC_ALLOCED(cachep); |
| 2689 | STATS_INC_ACTIVE(cachep); |
| 2690 | STATS_SET_HIGH(cachep); |
| 2691 | |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2692 | ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, |
| 2693 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2694 | } |
| 2695 | check_slabp(cachep, slabp); |
| 2696 | |
| 2697 | /* move slabp to correct slabp list: */ |
| 2698 | list_del(&slabp->list); |
| 2699 | if (slabp->free == BUFCTL_END) |
| 2700 | list_add(&slabp->list, &l3->slabs_full); |
| 2701 | else |
| 2702 | list_add(&slabp->list, &l3->slabs_partial); |
| 2703 | } |
| 2704 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2705 | must_grow: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2706 | l3->free_objects -= ac->avail; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2707 | alloc_done: |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2708 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2709 | |
| 2710 | if (unlikely(!ac->avail)) { |
| 2711 | int x; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2712 | x = cache_grow(cachep, flags, numa_node_id()); |
| 2713 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2714 | // cache_grow can reenable interrupts, then ac could change. |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2715 | ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2716 | if (!x && ac->avail == 0) // no objects in sight? abort |
| 2717 | return NULL; |
| 2718 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2719 | if (!ac->avail) // objects refilled by interrupt? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2720 | goto retry; |
| 2721 | } |
| 2722 | ac->touched = 1; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2723 | return ac->entry[--ac->avail]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2724 | } |
| 2725 | |
| 2726 | static inline void |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2727 | cache_alloc_debugcheck_before(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2728 | { |
| 2729 | might_sleep_if(flags & __GFP_WAIT); |
| 2730 | #if DEBUG |
| 2731 | kmem_flagcheck(cachep, flags); |
| 2732 | #endif |
| 2733 | } |
| 2734 | |
| 2735 | #if DEBUG |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2736 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, gfp_t flags, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2737 | void *objp, void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2738 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2739 | if (!objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2740 | return objp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2741 | if (cachep->flags & SLAB_POISON) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2742 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2743 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2744 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2745 | cachep->buffer_size / PAGE_SIZE, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2746 | else |
| 2747 | check_poison_obj(cachep, objp); |
| 2748 | #else |
| 2749 | check_poison_obj(cachep, objp); |
| 2750 | #endif |
| 2751 | poison_obj(cachep, objp, POISON_INUSE); |
| 2752 | } |
| 2753 | if (cachep->flags & SLAB_STORE_USER) |
| 2754 | *dbg_userword(cachep, objp) = caller; |
| 2755 | |
| 2756 | if (cachep->flags & SLAB_RED_ZONE) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2757 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE |
| 2758 | || *dbg_redzone2(cachep, objp) != RED_INACTIVE) { |
| 2759 | slab_error(cachep, |
| 2760 | "double free, or memory outside" |
| 2761 | " object was overwritten"); |
| 2762 | printk(KERN_ERR |
| 2763 | "%p: redzone 1: 0x%lx, redzone 2: 0x%lx.\n", |
| 2764 | objp, *dbg_redzone1(cachep, objp), |
| 2765 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2766 | } |
| 2767 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; |
| 2768 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; |
| 2769 | } |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2770 | objp += obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2771 | if (cachep->ctor && cachep->flags & SLAB_POISON) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2772 | unsigned long ctor_flags = SLAB_CTOR_CONSTRUCTOR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2773 | |
| 2774 | if (!(flags & __GFP_WAIT)) |
| 2775 | ctor_flags |= SLAB_CTOR_ATOMIC; |
| 2776 | |
| 2777 | cachep->ctor(objp, cachep, ctor_flags); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2778 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2779 | return objp; |
| 2780 | } |
| 2781 | #else |
| 2782 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) |
| 2783 | #endif |
| 2784 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2785 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2786 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2787 | void *objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | struct array_cache *ac; |
| 2789 | |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 2790 | #ifdef CONFIG_NUMA |
Christoph Lameter | 86c562a | 2006-01-18 17:42:37 -0800 | [diff] [blame] | 2791 | if (unlikely(current->mempolicy && !in_interrupt())) { |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 2792 | int nid = slab_node(current->mempolicy); |
| 2793 | |
| 2794 | if (nid != numa_node_id()) |
| 2795 | return __cache_alloc_node(cachep, flags, nid); |
| 2796 | } |
| 2797 | #endif |
| 2798 | |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2799 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2800 | ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2801 | if (likely(ac->avail)) { |
| 2802 | STATS_INC_ALLOCHIT(cachep); |
| 2803 | ac->touched = 1; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2804 | objp = ac->entry[--ac->avail]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2805 | } else { |
| 2806 | STATS_INC_ALLOCMISS(cachep); |
| 2807 | objp = cache_alloc_refill(cachep, flags); |
| 2808 | } |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2809 | return objp; |
| 2810 | } |
| 2811 | |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 2812 | static __always_inline void * |
| 2813 | __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller) |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2814 | { |
| 2815 | unsigned long save_flags; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2816 | void *objp; |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2817 | |
| 2818 | cache_alloc_debugcheck_before(cachep, flags); |
| 2819 | |
| 2820 | local_irq_save(save_flags); |
| 2821 | objp = ____cache_alloc(cachep, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2822 | local_irq_restore(save_flags); |
Eric Dumazet | 34342e8 | 2005-09-03 15:55:06 -0700 | [diff] [blame] | 2823 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 2824 | caller); |
Eric Dumazet | 34342e8 | 2005-09-03 15:55:06 -0700 | [diff] [blame] | 2825 | prefetchw(objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2826 | return objp; |
| 2827 | } |
| 2828 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2829 | #ifdef CONFIG_NUMA |
| 2830 | /* |
| 2831 | * A interface to enable slab creation on nodeid |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2832 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2833 | static void *__cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2834 | { |
| 2835 | struct list_head *entry; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2836 | struct slab *slabp; |
| 2837 | struct kmem_list3 *l3; |
| 2838 | void *obj; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2839 | int x; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2840 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2841 | l3 = cachep->nodelists[nodeid]; |
| 2842 | BUG_ON(!l3); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2843 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2844 | retry: |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 2845 | check_irq_off(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2846 | spin_lock(&l3->list_lock); |
| 2847 | entry = l3->slabs_partial.next; |
| 2848 | if (entry == &l3->slabs_partial) { |
| 2849 | l3->free_touched = 1; |
| 2850 | entry = l3->slabs_free.next; |
| 2851 | if (entry == &l3->slabs_free) |
| 2852 | goto must_grow; |
| 2853 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2854 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2855 | slabp = list_entry(entry, struct slab, list); |
| 2856 | check_spinlock_acquired_node(cachep, nodeid); |
| 2857 | check_slabp(cachep, slabp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2858 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2859 | STATS_INC_NODEALLOCS(cachep); |
| 2860 | STATS_INC_ACTIVE(cachep); |
| 2861 | STATS_SET_HIGH(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2862 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2863 | BUG_ON(slabp->inuse == cachep->num); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2864 | |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2865 | obj = slab_get_obj(cachep, slabp, nodeid); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2866 | check_slabp(cachep, slabp); |
| 2867 | l3->free_objects--; |
| 2868 | /* move slabp to correct slabp list: */ |
| 2869 | list_del(&slabp->list); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2870 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2871 | if (slabp->free == BUFCTL_END) { |
| 2872 | list_add(&slabp->list, &l3->slabs_full); |
| 2873 | } else { |
| 2874 | list_add(&slabp->list, &l3->slabs_partial); |
| 2875 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2876 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2877 | spin_unlock(&l3->list_lock); |
| 2878 | goto done; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2879 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2880 | must_grow: |
| 2881 | spin_unlock(&l3->list_lock); |
| 2882 | x = cache_grow(cachep, flags, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2883 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2884 | if (!x) |
| 2885 | return NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2886 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2887 | goto retry; |
| 2888 | done: |
| 2889 | return obj; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2890 | } |
| 2891 | #endif |
| 2892 | |
| 2893 | /* |
| 2894 | * Caller needs to acquire correct kmem_list's list_lock |
| 2895 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2896 | 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] | 2897 | int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2898 | { |
| 2899 | int i; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2900 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2901 | |
| 2902 | for (i = 0; i < nr_objects; i++) { |
| 2903 | void *objp = objpp[i]; |
| 2904 | struct slab *slabp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2905 | |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2906 | slabp = virt_to_slab(objp); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2907 | l3 = cachep->nodelists[node]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2908 | list_del(&slabp->list); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2909 | check_spinlock_acquired_node(cachep, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2910 | check_slabp(cachep, slabp); |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2911 | slab_put_obj(cachep, slabp, objp, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2912 | STATS_DEC_ACTIVE(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2913 | l3->free_objects++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2914 | check_slabp(cachep, slabp); |
| 2915 | |
| 2916 | /* fixup slab chains */ |
| 2917 | if (slabp->inuse == 0) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2918 | if (l3->free_objects > l3->free_limit) { |
| 2919 | l3->free_objects -= cachep->num; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2920 | slab_destroy(cachep, slabp); |
| 2921 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2922 | list_add(&slabp->list, &l3->slabs_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2923 | } |
| 2924 | } else { |
| 2925 | /* Unconditionally move a slab to the end of the |
| 2926 | * partial list on free - maximum time for the |
| 2927 | * other objects to be freed, too. |
| 2928 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2929 | list_add_tail(&slabp->list, &l3->slabs_partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2930 | } |
| 2931 | } |
| 2932 | } |
| 2933 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2934 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2935 | { |
| 2936 | int batchcount; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2937 | struct kmem_list3 *l3; |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2938 | int node = numa_node_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2939 | |
| 2940 | batchcount = ac->batchcount; |
| 2941 | #if DEBUG |
| 2942 | BUG_ON(!batchcount || batchcount > ac->avail); |
| 2943 | #endif |
| 2944 | check_irq_off(); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2945 | l3 = cachep->nodelists[node]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2946 | spin_lock(&l3->list_lock); |
| 2947 | if (l3->shared) { |
| 2948 | struct array_cache *shared_array = l3->shared; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2949 | int max = shared_array->limit - shared_array->avail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2950 | if (max) { |
| 2951 | if (batchcount > max) |
| 2952 | batchcount = max; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2953 | memcpy(&(shared_array->entry[shared_array->avail]), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2954 | ac->entry, sizeof(void *) * batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2955 | shared_array->avail += batchcount; |
| 2956 | goto free_done; |
| 2957 | } |
| 2958 | } |
| 2959 | |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2960 | free_block(cachep, ac->entry, batchcount, node); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2961 | free_done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2962 | #if STATS |
| 2963 | { |
| 2964 | int i = 0; |
| 2965 | struct list_head *p; |
| 2966 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2967 | p = l3->slabs_free.next; |
| 2968 | while (p != &(l3->slabs_free)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2969 | struct slab *slabp; |
| 2970 | |
| 2971 | slabp = list_entry(p, struct slab, list); |
| 2972 | BUG_ON(slabp->inuse); |
| 2973 | |
| 2974 | i++; |
| 2975 | p = p->next; |
| 2976 | } |
| 2977 | STATS_SET_FREEABLE(cachep, i); |
| 2978 | } |
| 2979 | #endif |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2980 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2981 | ac->avail -= batchcount; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2982 | memmove(ac->entry, &(ac->entry[batchcount]), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2983 | sizeof(void *) * ac->avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2984 | } |
| 2985 | |
| 2986 | /* |
| 2987 | * __cache_free |
| 2988 | * Release an obj back to its cache. If the obj has a constructed |
| 2989 | * state, it must be in this state _before_ it is released. |
| 2990 | * |
| 2991 | * Called with disabled ints. |
| 2992 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2993 | static inline void __cache_free(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2994 | { |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2995 | struct array_cache *ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2996 | |
| 2997 | check_irq_off(); |
| 2998 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); |
| 2999 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3000 | /* Make sure we are not freeing a object from another |
| 3001 | * node to the array cache on this cpu. |
| 3002 | */ |
| 3003 | #ifdef CONFIG_NUMA |
| 3004 | { |
| 3005 | struct slab *slabp; |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3006 | slabp = virt_to_slab(objp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3007 | if (unlikely(slabp->nodeid != numa_node_id())) { |
| 3008 | struct array_cache *alien = NULL; |
| 3009 | int nodeid = slabp->nodeid; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3010 | struct kmem_list3 *l3 = |
| 3011 | cachep->nodelists[numa_node_id()]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3012 | |
| 3013 | STATS_INC_NODEFREES(cachep); |
| 3014 | if (l3->alien && l3->alien[nodeid]) { |
| 3015 | alien = l3->alien[nodeid]; |
| 3016 | spin_lock(&alien->lock); |
| 3017 | if (unlikely(alien->avail == alien->limit)) |
| 3018 | __drain_alien_cache(cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3019 | alien, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3020 | alien->entry[alien->avail++] = objp; |
| 3021 | spin_unlock(&alien->lock); |
| 3022 | } else { |
| 3023 | spin_lock(&(cachep->nodelists[nodeid])-> |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3024 | list_lock); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3025 | free_block(cachep, &objp, 1, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3026 | spin_unlock(&(cachep->nodelists[nodeid])-> |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3027 | list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3028 | } |
| 3029 | return; |
| 3030 | } |
| 3031 | } |
| 3032 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3033 | if (likely(ac->avail < ac->limit)) { |
| 3034 | STATS_INC_FREEHIT(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3035 | ac->entry[ac->avail++] = objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3036 | return; |
| 3037 | } else { |
| 3038 | STATS_INC_FREEMISS(cachep); |
| 3039 | cache_flusharray(cachep, ac); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3040 | ac->entry[ac->avail++] = objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3041 | } |
| 3042 | } |
| 3043 | |
| 3044 | /** |
| 3045 | * kmem_cache_alloc - Allocate an object |
| 3046 | * @cachep: The cache to allocate from. |
| 3047 | * @flags: See kmalloc(). |
| 3048 | * |
| 3049 | * Allocate an object from this cache. The flags are only relevant |
| 3050 | * if the cache has no available objects. |
| 3051 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3052 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3053 | { |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3054 | return __cache_alloc(cachep, flags, __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3055 | } |
| 3056 | EXPORT_SYMBOL(kmem_cache_alloc); |
| 3057 | |
| 3058 | /** |
| 3059 | * kmem_ptr_validate - check if an untrusted pointer might |
| 3060 | * be a slab entry. |
| 3061 | * @cachep: the cache we're checking against |
| 3062 | * @ptr: pointer to validate |
| 3063 | * |
| 3064 | * This verifies that the untrusted pointer looks sane: |
| 3065 | * it is _not_ a guarantee that the pointer is actually |
| 3066 | * part of the slab cache in question, but it at least |
| 3067 | * validates that the pointer can be dereferenced and |
| 3068 | * looks half-way sane. |
| 3069 | * |
| 3070 | * Currently only used for dentry validation. |
| 3071 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3072 | int fastcall kmem_ptr_validate(struct kmem_cache *cachep, void *ptr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3073 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3074 | unsigned long addr = (unsigned long)ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3075 | unsigned long min_addr = PAGE_OFFSET; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3076 | unsigned long align_mask = BYTES_PER_WORD - 1; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3077 | unsigned long size = cachep->buffer_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3078 | struct page *page; |
| 3079 | |
| 3080 | if (unlikely(addr < min_addr)) |
| 3081 | goto out; |
| 3082 | if (unlikely(addr > (unsigned long)high_memory - size)) |
| 3083 | goto out; |
| 3084 | if (unlikely(addr & align_mask)) |
| 3085 | goto out; |
| 3086 | if (unlikely(!kern_addr_valid(addr))) |
| 3087 | goto out; |
| 3088 | if (unlikely(!kern_addr_valid(addr + size - 1))) |
| 3089 | goto out; |
| 3090 | page = virt_to_page(ptr); |
| 3091 | if (unlikely(!PageSlab(page))) |
| 3092 | goto out; |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 3093 | if (unlikely(page_get_cache(page) != cachep)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3094 | goto out; |
| 3095 | return 1; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3096 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3097 | return 0; |
| 3098 | } |
| 3099 | |
| 3100 | #ifdef CONFIG_NUMA |
| 3101 | /** |
| 3102 | * kmem_cache_alloc_node - Allocate an object on the specified node |
| 3103 | * @cachep: The cache to allocate from. |
| 3104 | * @flags: See kmalloc(). |
| 3105 | * @nodeid: node number of the target node. |
| 3106 | * |
| 3107 | * Identical to kmem_cache_alloc, except that this function is slow |
| 3108 | * and can sleep. And it will allocate memory on the given node, which |
| 3109 | * can improve the performance for cpu bound structures. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3110 | * New and improved: it will now make sure that the object gets |
| 3111 | * put on the correct node list so that there is no false sharing. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3112 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3113 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3114 | { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3115 | unsigned long save_flags; |
| 3116 | void *ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3117 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3118 | cache_alloc_debugcheck_before(cachep, flags); |
| 3119 | local_irq_save(save_flags); |
Christoph Lameter | 18f820f | 2006-02-01 03:05:43 -0800 | [diff] [blame] | 3120 | |
| 3121 | if (nodeid == -1 || nodeid == numa_node_id() || |
| 3122 | !cachep->nodelists[nodeid]) |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 3123 | ptr = ____cache_alloc(cachep, flags); |
| 3124 | else |
| 3125 | ptr = __cache_alloc_node(cachep, flags, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3126 | local_irq_restore(save_flags); |
Christoph Lameter | 18f820f | 2006-02-01 03:05:43 -0800 | [diff] [blame] | 3127 | |
| 3128 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, |
| 3129 | __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3130 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3131 | return ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3132 | } |
| 3133 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
| 3134 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 3135 | void *kmalloc_node(size_t size, gfp_t flags, int node) |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3136 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3137 | struct kmem_cache *cachep; |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3138 | |
| 3139 | cachep = kmem_find_general_cachep(size, flags); |
| 3140 | if (unlikely(cachep == NULL)) |
| 3141 | return NULL; |
| 3142 | return kmem_cache_alloc_node(cachep, flags, node); |
| 3143 | } |
| 3144 | EXPORT_SYMBOL(kmalloc_node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3145 | #endif |
| 3146 | |
| 3147 | /** |
| 3148 | * kmalloc - allocate memory |
| 3149 | * @size: how many bytes of memory are required. |
| 3150 | * @flags: the type of memory to allocate. |
| 3151 | * |
| 3152 | * kmalloc is the normal method of allocating memory |
| 3153 | * in the kernel. |
| 3154 | * |
| 3155 | * The @flags argument may be one of: |
| 3156 | * |
| 3157 | * %GFP_USER - Allocate memory on behalf of user. May sleep. |
| 3158 | * |
| 3159 | * %GFP_KERNEL - Allocate normal kernel ram. May sleep. |
| 3160 | * |
| 3161 | * %GFP_ATOMIC - Allocation will not sleep. Use inside interrupt handlers. |
| 3162 | * |
| 3163 | * Additionally, the %GFP_DMA flag may be set to indicate the memory |
| 3164 | * must be suitable for DMA. This can mean different things on different |
| 3165 | * platforms. For example, on i386, it means that the memory must come |
| 3166 | * from the first 16MB. |
| 3167 | */ |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3168 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
| 3169 | void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3170 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3171 | struct kmem_cache *cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3172 | |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3173 | /* If you want to save a few bytes .text space: replace |
| 3174 | * __ with kmem_. |
| 3175 | * Then kmalloc uses the uninlined functions instead of the inline |
| 3176 | * functions. |
| 3177 | */ |
| 3178 | cachep = __find_general_cachep(size, flags); |
Andrew Morton | dbdb904 | 2005-09-23 13:24:10 -0700 | [diff] [blame] | 3179 | if (unlikely(cachep == NULL)) |
| 3180 | return NULL; |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3181 | return __cache_alloc(cachep, flags, caller); |
| 3182 | } |
| 3183 | |
| 3184 | #ifndef CONFIG_DEBUG_SLAB |
| 3185 | |
| 3186 | void *__kmalloc(size_t size, gfp_t flags) |
| 3187 | { |
| 3188 | return __do_kmalloc(size, flags, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3189 | } |
| 3190 | EXPORT_SYMBOL(__kmalloc); |
| 3191 | |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3192 | #else |
| 3193 | |
| 3194 | void *__kmalloc_track_caller(size_t size, gfp_t flags, void *caller) |
| 3195 | { |
| 3196 | return __do_kmalloc(size, flags, caller); |
| 3197 | } |
| 3198 | EXPORT_SYMBOL(__kmalloc_track_caller); |
| 3199 | |
| 3200 | #endif |
| 3201 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3202 | #ifdef CONFIG_SMP |
| 3203 | /** |
| 3204 | * __alloc_percpu - allocate one copy of the object for every present |
| 3205 | * cpu in the system, zeroing them. |
| 3206 | * Objects should be dereferenced using the per_cpu_ptr macro only. |
| 3207 | * |
| 3208 | * @size: how many bytes of memory are required. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3209 | */ |
Pekka Enberg | f9f7500 | 2006-01-08 01:00:33 -0800 | [diff] [blame] | 3210 | void *__alloc_percpu(size_t size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3211 | { |
| 3212 | int i; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3213 | struct percpu_data *pdata = kmalloc(sizeof(*pdata), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3214 | |
| 3215 | if (!pdata) |
| 3216 | return NULL; |
| 3217 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3218 | /* |
| 3219 | * Cannot use for_each_online_cpu since a cpu may come online |
| 3220 | * and we have no way of figuring out how to fix the array |
| 3221 | * that we have allocated then.... |
| 3222 | */ |
| 3223 | for_each_cpu(i) { |
| 3224 | int node = cpu_to_node(i); |
| 3225 | |
| 3226 | if (node_online(node)) |
| 3227 | pdata->ptrs[i] = kmalloc_node(size, GFP_KERNEL, node); |
| 3228 | else |
| 3229 | pdata->ptrs[i] = kmalloc(size, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3230 | |
| 3231 | if (!pdata->ptrs[i]) |
| 3232 | goto unwind_oom; |
| 3233 | memset(pdata->ptrs[i], 0, size); |
| 3234 | } |
| 3235 | |
| 3236 | /* Catch derefs w/o wrappers */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3237 | return (void *)(~(unsigned long)pdata); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3238 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3239 | unwind_oom: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3240 | while (--i >= 0) { |
| 3241 | if (!cpu_possible(i)) |
| 3242 | continue; |
| 3243 | kfree(pdata->ptrs[i]); |
| 3244 | } |
| 3245 | kfree(pdata); |
| 3246 | return NULL; |
| 3247 | } |
| 3248 | EXPORT_SYMBOL(__alloc_percpu); |
| 3249 | #endif |
| 3250 | |
| 3251 | /** |
| 3252 | * kmem_cache_free - Deallocate an object |
| 3253 | * @cachep: The cache the allocation was from. |
| 3254 | * @objp: The previously allocated object. |
| 3255 | * |
| 3256 | * Free an object which was previously allocated from this |
| 3257 | * cache. |
| 3258 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3259 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3260 | { |
| 3261 | unsigned long flags; |
| 3262 | |
| 3263 | local_irq_save(flags); |
| 3264 | __cache_free(cachep, objp); |
| 3265 | local_irq_restore(flags); |
| 3266 | } |
| 3267 | EXPORT_SYMBOL(kmem_cache_free); |
| 3268 | |
| 3269 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3270 | * kfree - free previously allocated memory |
| 3271 | * @objp: pointer returned by kmalloc. |
| 3272 | * |
Pekka Enberg | 80e93ef | 2005-09-09 13:10:16 -0700 | [diff] [blame] | 3273 | * If @objp is NULL, no operation is performed. |
| 3274 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3275 | * Don't free memory not originally allocated by kmalloc() |
| 3276 | * or you will run into trouble. |
| 3277 | */ |
| 3278 | void kfree(const void *objp) |
| 3279 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3280 | struct kmem_cache *c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3281 | unsigned long flags; |
| 3282 | |
| 3283 | if (unlikely(!objp)) |
| 3284 | return; |
| 3285 | local_irq_save(flags); |
| 3286 | kfree_debugcheck(objp); |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3287 | c = virt_to_cache(objp); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3288 | mutex_debug_check_no_locks_freed(objp, obj_size(c)); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3289 | __cache_free(c, (void *)objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3290 | local_irq_restore(flags); |
| 3291 | } |
| 3292 | EXPORT_SYMBOL(kfree); |
| 3293 | |
| 3294 | #ifdef CONFIG_SMP |
| 3295 | /** |
| 3296 | * free_percpu - free previously allocated percpu memory |
| 3297 | * @objp: pointer returned by alloc_percpu. |
| 3298 | * |
| 3299 | * Don't free memory not originally allocated by alloc_percpu() |
| 3300 | * The complemented objp is to check for that. |
| 3301 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3302 | void free_percpu(const void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3303 | { |
| 3304 | int i; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3305 | struct percpu_data *p = (struct percpu_data *)(~(unsigned long)objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3306 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3307 | /* |
| 3308 | * We allocate for all cpus so we cannot use for online cpu here. |
| 3309 | */ |
| 3310 | for_each_cpu(i) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3311 | kfree(p->ptrs[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3312 | kfree(p); |
| 3313 | } |
| 3314 | EXPORT_SYMBOL(free_percpu); |
| 3315 | #endif |
| 3316 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3317 | unsigned int kmem_cache_size(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3318 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3319 | return obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3320 | } |
| 3321 | EXPORT_SYMBOL(kmem_cache_size); |
| 3322 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3323 | const char *kmem_cache_name(struct kmem_cache *cachep) |
Arnaldo Carvalho de Melo | 1944972 | 2005-06-18 22:46:19 -0700 | [diff] [blame] | 3324 | { |
| 3325 | return cachep->name; |
| 3326 | } |
| 3327 | EXPORT_SYMBOL_GPL(kmem_cache_name); |
| 3328 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3329 | /* |
| 3330 | * This initializes kmem_list3 for all nodes. |
| 3331 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3332 | static int alloc_kmemlist(struct kmem_cache *cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3333 | { |
| 3334 | int node; |
| 3335 | struct kmem_list3 *l3; |
| 3336 | int err = 0; |
| 3337 | |
| 3338 | for_each_online_node(node) { |
| 3339 | struct array_cache *nc = NULL, *new; |
| 3340 | struct array_cache **new_alien = NULL; |
| 3341 | #ifdef CONFIG_NUMA |
| 3342 | if (!(new_alien = alloc_alien_cache(node, cachep->limit))) |
| 3343 | goto fail; |
| 3344 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3345 | if (!(new = alloc_arraycache(node, (cachep->shared * |
| 3346 | cachep->batchcount), |
| 3347 | 0xbaadf00d))) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3348 | goto fail; |
| 3349 | if ((l3 = cachep->nodelists[node])) { |
| 3350 | |
| 3351 | spin_lock_irq(&l3->list_lock); |
| 3352 | |
| 3353 | if ((nc = cachep->nodelists[node]->shared)) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3354 | free_block(cachep, nc->entry, nc->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3355 | |
| 3356 | l3->shared = new; |
| 3357 | if (!cachep->nodelists[node]->alien) { |
| 3358 | l3->alien = new_alien; |
| 3359 | new_alien = NULL; |
| 3360 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3361 | l3->free_limit = (1 + nr_cpus_node(node)) * |
| 3362 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3363 | spin_unlock_irq(&l3->list_lock); |
| 3364 | kfree(nc); |
| 3365 | free_alien_cache(new_alien); |
| 3366 | continue; |
| 3367 | } |
| 3368 | if (!(l3 = kmalloc_node(sizeof(struct kmem_list3), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3369 | GFP_KERNEL, node))) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3370 | goto fail; |
| 3371 | |
| 3372 | kmem_list3_init(l3); |
| 3373 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3374 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3375 | l3->shared = new; |
| 3376 | l3->alien = new_alien; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3377 | l3->free_limit = (1 + nr_cpus_node(node)) * |
| 3378 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3379 | cachep->nodelists[node] = l3; |
| 3380 | } |
| 3381 | return err; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3382 | fail: |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3383 | err = -ENOMEM; |
| 3384 | return err; |
| 3385 | } |
| 3386 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3387 | struct ccupdate_struct { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3388 | struct kmem_cache *cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3389 | struct array_cache *new[NR_CPUS]; |
| 3390 | }; |
| 3391 | |
| 3392 | static void do_ccupdate_local(void *info) |
| 3393 | { |
| 3394 | struct ccupdate_struct *new = (struct ccupdate_struct *)info; |
| 3395 | struct array_cache *old; |
| 3396 | |
| 3397 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3398 | old = cpu_cache_get(new->cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3399 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3400 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
| 3401 | new->new[smp_processor_id()] = old; |
| 3402 | } |
| 3403 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3404 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, int batchcount, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3405 | int shared) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3406 | { |
| 3407 | struct ccupdate_struct new; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3408 | int i, err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3409 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3410 | memset(&new.new, 0, sizeof(new.new)); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3411 | for_each_online_cpu(i) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3412 | new.new[i] = |
| 3413 | alloc_arraycache(cpu_to_node(i), limit, batchcount); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3414 | if (!new.new[i]) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3415 | for (i--; i >= 0; i--) |
| 3416 | kfree(new.new[i]); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3417 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3418 | } |
| 3419 | } |
| 3420 | new.cachep = cachep; |
| 3421 | |
| 3422 | smp_call_function_all_cpus(do_ccupdate_local, (void *)&new); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3423 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3424 | check_irq_on(); |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3425 | spin_lock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3426 | cachep->batchcount = batchcount; |
| 3427 | cachep->limit = limit; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3428 | cachep->shared = shared; |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3429 | spin_unlock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3430 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3431 | for_each_online_cpu(i) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3432 | struct array_cache *ccold = new.new[i]; |
| 3433 | if (!ccold) |
| 3434 | continue; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3435 | spin_lock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3436 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_node(i)); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3437 | spin_unlock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3438 | kfree(ccold); |
| 3439 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3440 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3441 | err = alloc_kmemlist(cachep); |
| 3442 | if (err) { |
| 3443 | printk(KERN_ERR "alloc_kmemlist failed for %s, error %d.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3444 | cachep->name, -err); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3445 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3446 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3447 | return 0; |
| 3448 | } |
| 3449 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3450 | static void enable_cpucache(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3451 | { |
| 3452 | int err; |
| 3453 | int limit, shared; |
| 3454 | |
| 3455 | /* The head array serves three purposes: |
| 3456 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
| 3457 | * - reduce the number of spinlock operations. |
| 3458 | * - reduce the number of linked list operations on the slab and |
| 3459 | * bufctl chains: array operations are cheaper. |
| 3460 | * The numbers are guessed, we should auto-tune as described by |
| 3461 | * Bonwick. |
| 3462 | */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3463 | if (cachep->buffer_size > 131072) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3464 | limit = 1; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3465 | else if (cachep->buffer_size > PAGE_SIZE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3466 | limit = 8; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3467 | else if (cachep->buffer_size > 1024) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3468 | limit = 24; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3469 | else if (cachep->buffer_size > 256) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3470 | limit = 54; |
| 3471 | else |
| 3472 | limit = 120; |
| 3473 | |
| 3474 | /* Cpu bound tasks (e.g. network routing) can exhibit cpu bound |
| 3475 | * allocation behaviour: Most allocs on one cpu, most free operations |
| 3476 | * on another cpu. For these cases, an efficient object passing between |
| 3477 | * cpus is necessary. This is provided by a shared array. The array |
| 3478 | * replaces Bonwick's magazine layer. |
| 3479 | * On uniprocessor, it's functionally equivalent (but less efficient) |
| 3480 | * to a larger limit. Thus disabled by default. |
| 3481 | */ |
| 3482 | shared = 0; |
| 3483 | #ifdef CONFIG_SMP |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3484 | if (cachep->buffer_size <= PAGE_SIZE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3485 | shared = 8; |
| 3486 | #endif |
| 3487 | |
| 3488 | #if DEBUG |
| 3489 | /* With debugging enabled, large batchcount lead to excessively |
| 3490 | * long periods with disabled local interrupts. Limit the |
| 3491 | * batchcount |
| 3492 | */ |
| 3493 | if (limit > 32) |
| 3494 | limit = 32; |
| 3495 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3496 | err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3497 | if (err) |
| 3498 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3499 | cachep->name, -err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3500 | } |
| 3501 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3502 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3503 | int force, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3504 | { |
| 3505 | int tofree; |
| 3506 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3507 | check_spinlock_acquired_node(cachep, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3508 | if (ac->touched && !force) { |
| 3509 | ac->touched = 0; |
| 3510 | } else if (ac->avail) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3511 | tofree = force ? ac->avail : (ac->limit + 4) / 5; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3512 | if (tofree > ac->avail) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3513 | tofree = (ac->avail + 1) / 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3514 | } |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3515 | free_block(cachep, ac->entry, tofree, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3516 | ac->avail -= tofree; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3517 | memmove(ac->entry, &(ac->entry[tofree]), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3518 | sizeof(void *) * ac->avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3519 | } |
| 3520 | } |
| 3521 | |
| 3522 | /** |
| 3523 | * cache_reap - Reclaim memory from caches. |
Randy Dunlap | 1e5d533 | 2005-11-07 01:01:06 -0800 | [diff] [blame] | 3524 | * @unused: unused parameter |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3525 | * |
| 3526 | * Called from workqueue/eventd every few seconds. |
| 3527 | * Purpose: |
| 3528 | * - clear the per-cpu caches for this CPU. |
| 3529 | * - return freeable pages to the main free memory pool. |
| 3530 | * |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3531 | * If we cannot acquire the cache chain mutex then just give up - we'll |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3532 | * try again on the next iteration. |
| 3533 | */ |
| 3534 | static void cache_reap(void *unused) |
| 3535 | { |
| 3536 | struct list_head *walk; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3537 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3538 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3539 | if (!mutex_trylock(&cache_chain_mutex)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3540 | /* Give up. Setup the next iteration. */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3541 | schedule_delayed_work(&__get_cpu_var(reap_work), |
| 3542 | REAPTIMEOUT_CPUC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3543 | return; |
| 3544 | } |
| 3545 | |
| 3546 | list_for_each(walk, &cache_chain) { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3547 | struct kmem_cache *searchp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3548 | struct list_head *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3549 | int tofree; |
| 3550 | struct slab *slabp; |
| 3551 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3552 | searchp = list_entry(walk, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3553 | |
| 3554 | if (searchp->flags & SLAB_NO_REAP) |
| 3555 | goto next; |
| 3556 | |
| 3557 | check_irq_on(); |
| 3558 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3559 | l3 = searchp->nodelists[numa_node_id()]; |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 3560 | reap_alien(searchp, l3); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3561 | spin_lock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3562 | |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3563 | drain_array_locked(searchp, cpu_cache_get(searchp), 0, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3564 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3565 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3566 | if (time_after(l3->next_reap, jiffies)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3567 | goto next_unlock; |
| 3568 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3569 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3570 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3571 | if (l3->shared) |
| 3572 | drain_array_locked(searchp, l3->shared, 0, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3573 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3574 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3575 | if (l3->free_touched) { |
| 3576 | l3->free_touched = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3577 | goto next_unlock; |
| 3578 | } |
| 3579 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3580 | tofree = |
| 3581 | (l3->free_limit + 5 * searchp->num - |
| 3582 | 1) / (5 * searchp->num); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3583 | do { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3584 | p = l3->slabs_free.next; |
| 3585 | if (p == &(l3->slabs_free)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3586 | break; |
| 3587 | |
| 3588 | slabp = list_entry(p, struct slab, list); |
| 3589 | BUG_ON(slabp->inuse); |
| 3590 | list_del(&slabp->list); |
| 3591 | STATS_INC_REAPED(searchp); |
| 3592 | |
| 3593 | /* Safe to drop the lock. The slab is no longer |
| 3594 | * linked to the cache. |
| 3595 | * searchp cannot disappear, we hold |
| 3596 | * cache_chain_lock |
| 3597 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3598 | l3->free_objects -= searchp->num; |
| 3599 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3600 | slab_destroy(searchp, slabp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3601 | spin_lock_irq(&l3->list_lock); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3602 | } while (--tofree > 0); |
| 3603 | next_unlock: |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3604 | spin_unlock_irq(&l3->list_lock); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3605 | next: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3606 | cond_resched(); |
| 3607 | } |
| 3608 | check_irq_on(); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3609 | mutex_unlock(&cache_chain_mutex); |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 3610 | next_reap_node(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3611 | /* Setup the next iteration */ |
Manfred Spraul | cd61ef6 | 2005-11-07 00:58:02 -0800 | [diff] [blame] | 3612 | schedule_delayed_work(&__get_cpu_var(reap_work), REAPTIMEOUT_CPUC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3613 | } |
| 3614 | |
| 3615 | #ifdef CONFIG_PROC_FS |
| 3616 | |
Pekka Enberg | 85289f9 | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 3617 | static void print_slabinfo_header(struct seq_file *m) |
| 3618 | { |
| 3619 | /* |
| 3620 | * Output format version, so at least we can change it |
| 3621 | * without _too_ many complaints. |
| 3622 | */ |
| 3623 | #if STATS |
| 3624 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
| 3625 | #else |
| 3626 | seq_puts(m, "slabinfo - version: 2.1\n"); |
| 3627 | #endif |
| 3628 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
| 3629 | "<objperslab> <pagesperslab>"); |
| 3630 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
| 3631 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); |
| 3632 | #if STATS |
| 3633 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
| 3634 | "<error> <maxfreeable> <nodeallocs> <remotefrees>"); |
| 3635 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
| 3636 | #endif |
| 3637 | seq_putc(m, '\n'); |
| 3638 | } |
| 3639 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3640 | static void *s_start(struct seq_file *m, loff_t *pos) |
| 3641 | { |
| 3642 | loff_t n = *pos; |
| 3643 | struct list_head *p; |
| 3644 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3645 | mutex_lock(&cache_chain_mutex); |
Pekka Enberg | 85289f9 | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 3646 | if (!n) |
| 3647 | print_slabinfo_header(m); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3648 | p = cache_chain.next; |
| 3649 | while (n--) { |
| 3650 | p = p->next; |
| 3651 | if (p == &cache_chain) |
| 3652 | return NULL; |
| 3653 | } |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3654 | return list_entry(p, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3655 | } |
| 3656 | |
| 3657 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) |
| 3658 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3659 | struct kmem_cache *cachep = p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3660 | ++*pos; |
| 3661 | return cachep->next.next == &cache_chain ? NULL |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3662 | : list_entry(cachep->next.next, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3663 | } |
| 3664 | |
| 3665 | static void s_stop(struct seq_file *m, void *p) |
| 3666 | { |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3667 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3668 | } |
| 3669 | |
| 3670 | static int s_show(struct seq_file *m, void *p) |
| 3671 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3672 | struct kmem_cache *cachep = p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3673 | struct list_head *q; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3674 | struct slab *slabp; |
| 3675 | unsigned long active_objs; |
| 3676 | unsigned long num_objs; |
| 3677 | unsigned long active_slabs = 0; |
| 3678 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3679 | const char *name; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3680 | char *error = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3681 | int node; |
| 3682 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3683 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3684 | spin_lock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3685 | active_objs = 0; |
| 3686 | num_slabs = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3687 | for_each_online_node(node) { |
| 3688 | l3 = cachep->nodelists[node]; |
| 3689 | if (!l3) |
| 3690 | continue; |
| 3691 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3692 | check_irq_on(); |
| 3693 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3694 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3695 | list_for_each(q, &l3->slabs_full) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3696 | slabp = list_entry(q, struct slab, list); |
| 3697 | if (slabp->inuse != cachep->num && !error) |
| 3698 | error = "slabs_full accounting error"; |
| 3699 | active_objs += cachep->num; |
| 3700 | active_slabs++; |
| 3701 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3702 | list_for_each(q, &l3->slabs_partial) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3703 | slabp = list_entry(q, struct slab, list); |
| 3704 | if (slabp->inuse == cachep->num && !error) |
| 3705 | error = "slabs_partial inuse accounting error"; |
| 3706 | if (!slabp->inuse && !error) |
| 3707 | error = "slabs_partial/inuse accounting error"; |
| 3708 | active_objs += slabp->inuse; |
| 3709 | active_slabs++; |
| 3710 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3711 | list_for_each(q, &l3->slabs_free) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3712 | slabp = list_entry(q, struct slab, list); |
| 3713 | if (slabp->inuse && !error) |
| 3714 | error = "slabs_free/inuse accounting error"; |
| 3715 | num_slabs++; |
| 3716 | } |
| 3717 | free_objects += l3->free_objects; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 3718 | if (l3->shared) |
| 3719 | shared_avail += l3->shared->avail; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3720 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3721 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3722 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3723 | num_slabs += active_slabs; |
| 3724 | num_objs = num_slabs * cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3725 | if (num_objs - active_objs != free_objects && !error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3726 | error = "free_objects accounting error"; |
| 3727 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3728 | name = cachep->name; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3729 | if (error) |
| 3730 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); |
| 3731 | |
| 3732 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3733 | name, active_objs, num_objs, cachep->buffer_size, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3734 | cachep->num, (1 << cachep->gfporder)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3735 | seq_printf(m, " : tunables %4u %4u %4u", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3736 | cachep->limit, cachep->batchcount, cachep->shared); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3737 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3738 | active_slabs, num_slabs, shared_avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3739 | #if STATS |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3740 | { /* list3 stats */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3741 | unsigned long high = cachep->high_mark; |
| 3742 | unsigned long allocs = cachep->num_allocations; |
| 3743 | unsigned long grown = cachep->grown; |
| 3744 | unsigned long reaped = cachep->reaped; |
| 3745 | unsigned long errors = cachep->errors; |
| 3746 | unsigned long max_freeable = cachep->max_freeable; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3747 | unsigned long node_allocs = cachep->node_allocs; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3748 | unsigned long node_frees = cachep->node_frees; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3749 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3750 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu \ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3751 | %4lu %4lu %4lu %4lu", allocs, high, grown, reaped, errors, max_freeable, node_allocs, node_frees); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3752 | } |
| 3753 | /* cpu stats */ |
| 3754 | { |
| 3755 | unsigned long allochit = atomic_read(&cachep->allochit); |
| 3756 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); |
| 3757 | unsigned long freehit = atomic_read(&cachep->freehit); |
| 3758 | unsigned long freemiss = atomic_read(&cachep->freemiss); |
| 3759 | |
| 3760 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3761 | allochit, allocmiss, freehit, freemiss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3762 | } |
| 3763 | #endif |
| 3764 | seq_putc(m, '\n'); |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3765 | spin_unlock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3766 | return 0; |
| 3767 | } |
| 3768 | |
| 3769 | /* |
| 3770 | * slabinfo_op - iterator that generates /proc/slabinfo |
| 3771 | * |
| 3772 | * Output layout: |
| 3773 | * cache-name |
| 3774 | * num-active-objs |
| 3775 | * total-objs |
| 3776 | * object size |
| 3777 | * num-active-slabs |
| 3778 | * total-slabs |
| 3779 | * num-pages-per-slab |
| 3780 | * + further values on SMP and with statistics enabled |
| 3781 | */ |
| 3782 | |
| 3783 | struct seq_operations slabinfo_op = { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3784 | .start = s_start, |
| 3785 | .next = s_next, |
| 3786 | .stop = s_stop, |
| 3787 | .show = s_show, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3788 | }; |
| 3789 | |
| 3790 | #define MAX_SLABINFO_WRITE 128 |
| 3791 | /** |
| 3792 | * slabinfo_write - Tuning for the slab allocator |
| 3793 | * @file: unused |
| 3794 | * @buffer: user buffer |
| 3795 | * @count: data length |
| 3796 | * @ppos: unused |
| 3797 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3798 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, |
| 3799 | size_t count, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3800 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3801 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3802 | int limit, batchcount, shared, res; |
| 3803 | struct list_head *p; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3804 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3805 | if (count > MAX_SLABINFO_WRITE) |
| 3806 | return -EINVAL; |
| 3807 | if (copy_from_user(&kbuf, buffer, count)) |
| 3808 | return -EFAULT; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3809 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3810 | |
| 3811 | tmp = strchr(kbuf, ' '); |
| 3812 | if (!tmp) |
| 3813 | return -EINVAL; |
| 3814 | *tmp = '\0'; |
| 3815 | tmp++; |
| 3816 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) |
| 3817 | return -EINVAL; |
| 3818 | |
| 3819 | /* Find the cache in the chain of caches. */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3820 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3821 | res = -EINVAL; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3822 | list_for_each(p, &cache_chain) { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3823 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, |
| 3824 | next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3825 | |
| 3826 | if (!strcmp(cachep->name, kbuf)) { |
| 3827 | if (limit < 1 || |
| 3828 | batchcount < 1 || |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3829 | batchcount > limit || shared < 0) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3830 | res = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3831 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3832 | res = do_tune_cpucache(cachep, limit, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3833 | batchcount, shared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3834 | } |
| 3835 | break; |
| 3836 | } |
| 3837 | } |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3838 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3839 | if (res >= 0) |
| 3840 | res = count; |
| 3841 | return res; |
| 3842 | } |
| 3843 | #endif |
| 3844 | |
Manfred Spraul | 00e145b | 2005-09-03 15:55:07 -0700 | [diff] [blame] | 3845 | /** |
| 3846 | * ksize - get the actual amount of memory allocated for a given object |
| 3847 | * @objp: Pointer to the object |
| 3848 | * |
| 3849 | * kmalloc may internally round up allocations and return more memory |
| 3850 | * than requested. ksize() can be used to determine the actual amount of |
| 3851 | * memory allocated. The caller may use this additional memory, even though |
| 3852 | * a smaller amount of memory was initially specified with the kmalloc call. |
| 3853 | * The caller must guarantee that objp points to a valid object previously |
| 3854 | * allocated with either kmalloc() or kmem_cache_alloc(). The object |
| 3855 | * must not be freed during the duration of the call. |
| 3856 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3857 | unsigned int ksize(const void *objp) |
| 3858 | { |
Manfred Spraul | 00e145b | 2005-09-03 15:55:07 -0700 | [diff] [blame] | 3859 | if (unlikely(objp == NULL)) |
| 3860 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3861 | |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3862 | return obj_size(virt_to_cache(objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3863 | } |