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 | |
| 792 | /* |
| 793 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz |
| 794 | * via the workqueue/eventd. |
| 795 | * Add the CPU number into the expiration time to minimize the possibility of |
| 796 | * the CPUs getting into lockstep and contending for the global cache chain |
| 797 | * lock. |
| 798 | */ |
| 799 | static void __devinit start_cpu_timer(int cpu) |
| 800 | { |
| 801 | struct work_struct *reap_work = &per_cpu(reap_work, cpu); |
| 802 | |
| 803 | /* |
| 804 | * When this gets called from do_initcalls via cpucache_init(), |
| 805 | * init_workqueues() has already run, so keventd will be setup |
| 806 | * at that time. |
| 807 | */ |
| 808 | if (keventd_up() && reap_work->func == NULL) { |
| 809 | INIT_WORK(reap_work, cache_reap, NULL); |
| 810 | schedule_delayed_work_on(cpu, reap_work, HZ + 3 * cpu); |
| 811 | } |
| 812 | } |
| 813 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 814 | static struct array_cache *alloc_arraycache(int node, int entries, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 815 | int batchcount) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 816 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 817 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 818 | struct array_cache *nc = NULL; |
| 819 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 820 | nc = kmalloc_node(memsize, GFP_KERNEL, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 821 | if (nc) { |
| 822 | nc->avail = 0; |
| 823 | nc->limit = entries; |
| 824 | nc->batchcount = batchcount; |
| 825 | nc->touched = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 826 | spin_lock_init(&nc->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | } |
| 828 | return nc; |
| 829 | } |
| 830 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 831 | #ifdef CONFIG_NUMA |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 832 | static void *__cache_alloc_node(struct kmem_cache *, gfp_t, int); |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 833 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 834 | static struct array_cache **alloc_alien_cache(int node, int limit) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 835 | { |
| 836 | struct array_cache **ac_ptr; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 837 | int memsize = sizeof(void *) * MAX_NUMNODES; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 838 | int i; |
| 839 | |
| 840 | if (limit > 1) |
| 841 | limit = 12; |
| 842 | ac_ptr = kmalloc_node(memsize, GFP_KERNEL, node); |
| 843 | if (ac_ptr) { |
| 844 | for_each_node(i) { |
| 845 | if (i == node || !node_online(i)) { |
| 846 | ac_ptr[i] = NULL; |
| 847 | continue; |
| 848 | } |
| 849 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d); |
| 850 | if (!ac_ptr[i]) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 851 | for (i--; i <= 0; i--) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 852 | kfree(ac_ptr[i]); |
| 853 | kfree(ac_ptr); |
| 854 | return NULL; |
| 855 | } |
| 856 | } |
| 857 | } |
| 858 | return ac_ptr; |
| 859 | } |
| 860 | |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 861 | static void free_alien_cache(struct array_cache **ac_ptr) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 862 | { |
| 863 | int i; |
| 864 | |
| 865 | if (!ac_ptr) |
| 866 | return; |
| 867 | |
| 868 | for_each_node(i) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 869 | kfree(ac_ptr[i]); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 870 | |
| 871 | kfree(ac_ptr); |
| 872 | } |
| 873 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 874 | static void __drain_alien_cache(struct kmem_cache *cachep, |
Pekka Enberg | 5295a74 | 2006-02-01 03:05:48 -0800 | [diff] [blame] | 875 | struct array_cache *ac, int node) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 876 | { |
| 877 | struct kmem_list3 *rl3 = cachep->nodelists[node]; |
| 878 | |
| 879 | if (ac->avail) { |
| 880 | spin_lock(&rl3->list_lock); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 881 | free_block(cachep, ac->entry, ac->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 882 | ac->avail = 0; |
| 883 | spin_unlock(&rl3->list_lock); |
| 884 | } |
| 885 | } |
| 886 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 887 | 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] | 888 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 889 | int i = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 890 | struct array_cache *ac; |
| 891 | unsigned long flags; |
| 892 | |
| 893 | for_each_online_node(i) { |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 894 | ac = alien[i]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 895 | if (ac) { |
| 896 | spin_lock_irqsave(&ac->lock, flags); |
| 897 | __drain_alien_cache(cachep, ac, i); |
| 898 | spin_unlock_irqrestore(&ac->lock, flags); |
| 899 | } |
| 900 | } |
| 901 | } |
| 902 | #else |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 903 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 904 | #define drain_alien_cache(cachep, alien) do { } while (0) |
| 905 | |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 906 | static inline struct array_cache **alloc_alien_cache(int node, int limit) |
| 907 | { |
| 908 | return (struct array_cache **) 0x01020304ul; |
| 909 | } |
| 910 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 911 | static inline void free_alien_cache(struct array_cache **ac_ptr) |
| 912 | { |
| 913 | } |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 914 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 915 | #endif |
| 916 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 917 | static int __devinit cpuup_callback(struct notifier_block *nfb, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 918 | unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | { |
| 920 | long cpu = (long)hcpu; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 921 | struct kmem_cache *cachep; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 922 | struct kmem_list3 *l3 = NULL; |
| 923 | int node = cpu_to_node(cpu); |
| 924 | int memsize = sizeof(struct kmem_list3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | |
| 926 | switch (action) { |
| 927 | case CPU_UP_PREPARE: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 928 | mutex_lock(&cache_chain_mutex); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 929 | /* we need to do this right in the beginning since |
| 930 | * alloc_arraycache's are going to use this list. |
| 931 | * kmalloc_node allows us to add the slab to the right |
| 932 | * kmem_list3 and not this cpu's kmem_list3 |
| 933 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 935 | list_for_each_entry(cachep, &cache_chain, next) { |
| 936 | /* setup the size64 kmemlist for cpu before we can |
| 937 | * begin anything. Make sure some other cpu on this |
| 938 | * node has not already allocated this |
| 939 | */ |
| 940 | if (!cachep->nodelists[node]) { |
| 941 | if (!(l3 = kmalloc_node(memsize, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 942 | GFP_KERNEL, node))) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 943 | goto bad; |
| 944 | kmem_list3_init(l3); |
| 945 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 946 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 947 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 948 | /* |
| 949 | * The l3s don't come and go as CPUs come and |
| 950 | * go. cache_chain_mutex is sufficient |
| 951 | * protection here. |
| 952 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 953 | cachep->nodelists[node] = l3; |
| 954 | } |
| 955 | |
| 956 | spin_lock_irq(&cachep->nodelists[node]->list_lock); |
| 957 | cachep->nodelists[node]->free_limit = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 958 | (1 + nr_cpus_node(node)) * |
| 959 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 960 | spin_unlock_irq(&cachep->nodelists[node]->list_lock); |
| 961 | } |
| 962 | |
| 963 | /* Now we can go ahead with allocating the shared array's |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 964 | & array cache's */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 965 | list_for_each_entry(cachep, &cache_chain, next) { |
Tobias Klauser | cd105df | 2006-01-08 01:00:59 -0800 | [diff] [blame] | 966 | struct array_cache *nc; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 967 | struct array_cache *shared; |
| 968 | struct array_cache **alien; |
Tobias Klauser | cd105df | 2006-01-08 01:00:59 -0800 | [diff] [blame] | 969 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 970 | nc = alloc_arraycache(node, cachep->limit, |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 971 | cachep->batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | if (!nc) |
| 973 | goto bad; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 974 | shared = alloc_arraycache(node, |
| 975 | cachep->shared * cachep->batchcount, |
| 976 | 0xbaadf00d); |
| 977 | if (!shared) |
| 978 | goto bad; |
Linus Torvalds | 7a21ef6 | 2006-02-05 11:26:38 -0800 | [diff] [blame] | 979 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 980 | alien = alloc_alien_cache(node, cachep->limit); |
| 981 | if (!alien) |
| 982 | goto bad; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 983 | cachep->array[cpu] = nc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 984 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 985 | l3 = cachep->nodelists[node]; |
| 986 | BUG_ON(!l3); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 987 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 988 | spin_lock_irq(&l3->list_lock); |
| 989 | if (!l3->shared) { |
| 990 | /* |
| 991 | * We are serialised from CPU_DEAD or |
| 992 | * CPU_UP_CANCELLED by the cpucontrol lock |
| 993 | */ |
| 994 | l3->shared = shared; |
| 995 | shared = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 996 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 997 | #ifdef CONFIG_NUMA |
| 998 | if (!l3->alien) { |
| 999 | l3->alien = alien; |
| 1000 | alien = NULL; |
| 1001 | } |
| 1002 | #endif |
| 1003 | spin_unlock_irq(&l3->list_lock); |
| 1004 | |
| 1005 | kfree(shared); |
| 1006 | free_alien_cache(alien); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | } |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1008 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | break; |
| 1010 | case CPU_ONLINE: |
| 1011 | start_cpu_timer(cpu); |
| 1012 | break; |
| 1013 | #ifdef CONFIG_HOTPLUG_CPU |
| 1014 | case CPU_DEAD: |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1015 | /* |
| 1016 | * Even if all the cpus of a node are down, we don't free the |
| 1017 | * kmem_list3 of any cache. This to avoid a race between |
| 1018 | * cpu_down, and a kmalloc allocation from another cpu for |
| 1019 | * memory from the node of the cpu going down. The list3 |
| 1020 | * structure is usually allocated from kmem_cache_create() and |
| 1021 | * gets destroyed at kmem_cache_destroy(). |
| 1022 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1023 | /* fall thru */ |
| 1024 | case CPU_UP_CANCELED: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1025 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | |
| 1027 | list_for_each_entry(cachep, &cache_chain, next) { |
| 1028 | struct array_cache *nc; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1029 | struct array_cache *shared; |
| 1030 | struct array_cache **alien; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1031 | cpumask_t mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1033 | mask = node_to_cpumask(node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | /* cpu is dead; no one can alloc from it. */ |
| 1035 | nc = cachep->array[cpu]; |
| 1036 | cachep->array[cpu] = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1037 | l3 = cachep->nodelists[node]; |
| 1038 | |
| 1039 | if (!l3) |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1040 | goto free_array_cache; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1041 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 1042 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1043 | |
| 1044 | /* Free limit for this kmem_list3 */ |
| 1045 | l3->free_limit -= cachep->batchcount; |
| 1046 | if (nc) |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 1047 | free_block(cachep, nc->entry, nc->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1048 | |
| 1049 | if (!cpus_empty(mask)) { |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 1050 | spin_unlock_irq(&l3->list_lock); |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1051 | goto free_array_cache; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1052 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1053 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1054 | shared = l3->shared; |
| 1055 | if (shared) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1056 | free_block(cachep, l3->shared->entry, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1057 | l3->shared->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1058 | l3->shared = NULL; |
| 1059 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1060 | |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1061 | alien = l3->alien; |
| 1062 | l3->alien = NULL; |
| 1063 | |
| 1064 | spin_unlock_irq(&l3->list_lock); |
| 1065 | |
| 1066 | kfree(shared); |
| 1067 | if (alien) { |
| 1068 | drain_alien_cache(cachep, alien); |
| 1069 | free_alien_cache(alien); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1070 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1071 | free_array_cache: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | kfree(nc); |
| 1073 | } |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 1074 | /* |
| 1075 | * In the previous loop, all the objects were freed to |
| 1076 | * the respective cache's slabs, now we can go ahead and |
| 1077 | * shrink each nodelist to its limit. |
| 1078 | */ |
| 1079 | list_for_each_entry(cachep, &cache_chain, next) { |
| 1080 | l3 = cachep->nodelists[node]; |
| 1081 | if (!l3) |
| 1082 | continue; |
| 1083 | spin_lock_irq(&l3->list_lock); |
| 1084 | /* free slabs belonging to this node */ |
| 1085 | __node_shrink(cachep, node); |
| 1086 | spin_unlock_irq(&l3->list_lock); |
| 1087 | } |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1088 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1089 | break; |
| 1090 | #endif |
| 1091 | } |
| 1092 | return NOTIFY_OK; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1093 | bad: |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1094 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | return NOTIFY_BAD; |
| 1096 | } |
| 1097 | |
| 1098 | static struct notifier_block cpucache_notifier = { &cpuup_callback, NULL, 0 }; |
| 1099 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1100 | /* |
| 1101 | * swap the static kmem_list3 with kmalloced memory |
| 1102 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1103 | 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] | 1104 | { |
| 1105 | struct kmem_list3 *ptr; |
| 1106 | |
| 1107 | BUG_ON(cachep->nodelists[nodeid] != list); |
| 1108 | ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_KERNEL, nodeid); |
| 1109 | BUG_ON(!ptr); |
| 1110 | |
| 1111 | local_irq_disable(); |
| 1112 | memcpy(ptr, list, sizeof(struct kmem_list3)); |
| 1113 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
| 1114 | cachep->nodelists[nodeid] = ptr; |
| 1115 | local_irq_enable(); |
| 1116 | } |
| 1117 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | /* Initialisation. |
| 1119 | * Called after the gfp() functions have been enabled, and before smp_init(). |
| 1120 | */ |
| 1121 | void __init kmem_cache_init(void) |
| 1122 | { |
| 1123 | size_t left_over; |
| 1124 | struct cache_sizes *sizes; |
| 1125 | struct cache_names *names; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1126 | int i; |
| 1127 | |
| 1128 | for (i = 0; i < NUM_INIT_LISTS; i++) { |
| 1129 | kmem_list3_init(&initkmem_list3[i]); |
| 1130 | if (i < MAX_NUMNODES) |
| 1131 | cache_cache.nodelists[i] = NULL; |
| 1132 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | |
| 1134 | /* |
| 1135 | * Fragmentation resistance on low memory - only use bigger |
| 1136 | * page orders on machines with more than 32MB of memory. |
| 1137 | */ |
| 1138 | if (num_physpages > (32 << 20) >> PAGE_SHIFT) |
| 1139 | slab_break_gfp_order = BREAK_GFP_ORDER_HI; |
| 1140 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1141 | /* Bootstrap is tricky, because several objects are allocated |
| 1142 | * from caches that do not exist yet: |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1143 | * 1) initialize the cache_cache cache: it contains the struct kmem_cache |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1144 | * structures of all caches, except cache_cache itself: cache_cache |
| 1145 | * is statically allocated. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1146 | * Initially an __init data area is used for the head array and the |
| 1147 | * kmem_list3 structures, it's replaced with a kmalloc allocated |
| 1148 | * array at the end of the bootstrap. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | * 2) Create the first kmalloc cache. |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1150 | * The struct kmem_cache for the new cache is allocated normally. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1151 | * An __init data area is used for the head array. |
| 1152 | * 3) Create the remaining kmalloc caches, with minimally sized |
| 1153 | * head arrays. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | * 4) Replace the __init data head arrays for cache_cache and the first |
| 1155 | * kmalloc cache with kmalloc allocated arrays. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1156 | * 5) Replace the __init data for kmem_list3 for cache_cache and |
| 1157 | * the other cache's with kmalloc allocated memory. |
| 1158 | * 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] | 1159 | */ |
| 1160 | |
| 1161 | /* 1) create the cache_cache */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | INIT_LIST_HEAD(&cache_chain); |
| 1163 | list_add(&cache_cache.next, &cache_chain); |
| 1164 | cache_cache.colour_off = cache_line_size(); |
| 1165 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1166 | cache_cache.nodelists[numa_node_id()] = &initkmem_list3[CACHE_CACHE]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1167 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1168 | cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, cache_line_size()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1170 | cache_estimate(0, cache_cache.buffer_size, cache_line_size(), 0, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1171 | &left_over, &cache_cache.num); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | if (!cache_cache.num) |
| 1173 | BUG(); |
| 1174 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1175 | cache_cache.colour = left_over / cache_cache.colour_off; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1176 | cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) + |
| 1177 | sizeof(struct slab), cache_line_size()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | |
| 1179 | /* 2+3) create the kmalloc caches */ |
| 1180 | sizes = malloc_sizes; |
| 1181 | names = cache_names; |
| 1182 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1183 | /* Initialize the caches that provide memory for the array cache |
| 1184 | * and the kmem_list3 structures first. |
| 1185 | * Without this, further allocations will bug |
| 1186 | */ |
| 1187 | |
| 1188 | sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1189 | sizes[INDEX_AC].cs_size, |
| 1190 | ARCH_KMALLOC_MINALIGN, |
| 1191 | (ARCH_KMALLOC_FLAGS | |
| 1192 | SLAB_PANIC), NULL, NULL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1193 | |
| 1194 | if (INDEX_AC != INDEX_L3) |
| 1195 | sizes[INDEX_L3].cs_cachep = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1196 | kmem_cache_create(names[INDEX_L3].name, |
| 1197 | sizes[INDEX_L3].cs_size, |
| 1198 | ARCH_KMALLOC_MINALIGN, |
| 1199 | (ARCH_KMALLOC_FLAGS | SLAB_PANIC), NULL, |
| 1200 | NULL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1201 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1202 | while (sizes->cs_size != ULONG_MAX) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1203 | /* |
| 1204 | * For performance, all the general caches are L1 aligned. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1205 | * This should be particularly beneficial on SMP boxes, as it |
| 1206 | * eliminates "false sharing". |
| 1207 | * Note for systems short on memory removing the alignment will |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1208 | * allow tighter packing of the smaller caches. |
| 1209 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1210 | if (!sizes->cs_cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1211 | sizes->cs_cachep = kmem_cache_create(names->name, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1212 | sizes->cs_size, |
| 1213 | ARCH_KMALLOC_MINALIGN, |
| 1214 | (ARCH_KMALLOC_FLAGS |
| 1215 | | SLAB_PANIC), |
| 1216 | NULL, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1217 | |
| 1218 | /* Inc off-slab bufctl limit until the ceiling is hit. */ |
| 1219 | if (!(OFF_SLAB(sizes->cs_cachep))) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1220 | offslab_limit = sizes->cs_size - sizeof(struct slab); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1221 | offslab_limit /= sizeof(kmem_bufctl_t); |
| 1222 | } |
| 1223 | |
| 1224 | sizes->cs_dmacachep = kmem_cache_create(names->name_dma, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1225 | sizes->cs_size, |
| 1226 | ARCH_KMALLOC_MINALIGN, |
| 1227 | (ARCH_KMALLOC_FLAGS | |
| 1228 | SLAB_CACHE_DMA | |
| 1229 | SLAB_PANIC), NULL, |
| 1230 | NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1231 | |
| 1232 | sizes++; |
| 1233 | names++; |
| 1234 | } |
| 1235 | /* 4) Replace the bootstrap head arrays */ |
| 1236 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1237 | void *ptr; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1238 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1239 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1240 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | local_irq_disable(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1242 | BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); |
| 1243 | memcpy(ptr, cpu_cache_get(&cache_cache), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1244 | sizeof(struct arraycache_init)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | cache_cache.array[smp_processor_id()] = ptr; |
| 1246 | local_irq_enable(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1247 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1248 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1249 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1250 | local_irq_disable(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1251 | BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1252 | != &initarray_generic.cache); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1253 | memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1254 | sizeof(struct arraycache_init)); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1255 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1256 | ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1257 | local_irq_enable(); |
| 1258 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1259 | /* 5) Replace the bootstrap kmem_list3's */ |
| 1260 | { |
| 1261 | int node; |
| 1262 | /* Replace the static kmem_list3 structures for the boot cpu */ |
| 1263 | init_list(&cache_cache, &initkmem_list3[CACHE_CACHE], |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1264 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1266 | for_each_online_node(node) { |
| 1267 | init_list(malloc_sizes[INDEX_AC].cs_cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1268 | &initkmem_list3[SIZE_AC + node], node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1269 | |
| 1270 | if (INDEX_AC != INDEX_L3) { |
| 1271 | init_list(malloc_sizes[INDEX_L3].cs_cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1272 | &initkmem_list3[SIZE_L3 + node], |
| 1273 | node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1274 | } |
| 1275 | } |
| 1276 | } |
| 1277 | |
| 1278 | /* 6) resize the head arrays to their final sizes */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1279 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1280 | struct kmem_cache *cachep; |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1281 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | list_for_each_entry(cachep, &cache_chain, next) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1283 | enable_cpucache(cachep); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1284 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1285 | } |
| 1286 | |
| 1287 | /* Done! */ |
| 1288 | g_cpucache_up = FULL; |
| 1289 | |
| 1290 | /* Register a cpu startup notifier callback |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1291 | * that initializes cpu_cache_get for all new cpus |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | */ |
| 1293 | register_cpu_notifier(&cpucache_notifier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1294 | |
| 1295 | /* The reap timers are started later, with a module init call: |
| 1296 | * That part of the kernel is not yet operational. |
| 1297 | */ |
| 1298 | } |
| 1299 | |
| 1300 | static int __init cpucache_init(void) |
| 1301 | { |
| 1302 | int cpu; |
| 1303 | |
| 1304 | /* |
| 1305 | * Register the timers that return unneeded |
| 1306 | * pages to gfp. |
| 1307 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1308 | for_each_online_cpu(cpu) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1309 | start_cpu_timer(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1310 | |
| 1311 | return 0; |
| 1312 | } |
| 1313 | |
| 1314 | __initcall(cpucache_init); |
| 1315 | |
| 1316 | /* |
| 1317 | * Interface to system's page allocator. No need to hold the cache-lock. |
| 1318 | * |
| 1319 | * If we requested dmaable memory, we will get it. Even if we |
| 1320 | * did not request dmaable memory, we might get it, but that |
| 1321 | * would be relatively rare and ignorable. |
| 1322 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1323 | 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] | 1324 | { |
| 1325 | struct page *page; |
| 1326 | void *addr; |
| 1327 | int i; |
| 1328 | |
| 1329 | flags |= cachep->gfpflags; |
Christoph Lameter | 50c85a1 | 2005-11-13 16:06:47 -0800 | [diff] [blame] | 1330 | page = alloc_pages_node(nodeid, flags, cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1331 | if (!page) |
| 1332 | return NULL; |
| 1333 | addr = page_address(page); |
| 1334 | |
| 1335 | i = (1 << cachep->gfporder); |
| 1336 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1337 | atomic_add(i, &slab_reclaim_pages); |
| 1338 | add_page_state(nr_slab, i); |
| 1339 | while (i--) { |
| 1340 | SetPageSlab(page); |
| 1341 | page++; |
| 1342 | } |
| 1343 | return addr; |
| 1344 | } |
| 1345 | |
| 1346 | /* |
| 1347 | * Interface to system's page release. |
| 1348 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1349 | static void kmem_freepages(struct kmem_cache *cachep, void *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1351 | unsigned long i = (1 << cachep->gfporder); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | struct page *page = virt_to_page(addr); |
| 1353 | const unsigned long nr_freed = i; |
| 1354 | |
| 1355 | while (i--) { |
| 1356 | if (!TestClearPageSlab(page)) |
| 1357 | BUG(); |
| 1358 | page++; |
| 1359 | } |
| 1360 | sub_page_state(nr_slab, nr_freed); |
| 1361 | if (current->reclaim_state) |
| 1362 | current->reclaim_state->reclaimed_slab += nr_freed; |
| 1363 | free_pages((unsigned long)addr, cachep->gfporder); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1364 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
| 1365 | atomic_sub(1 << cachep->gfporder, &slab_reclaim_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1366 | } |
| 1367 | |
| 1368 | static void kmem_rcu_free(struct rcu_head *head) |
| 1369 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1370 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1371 | struct kmem_cache *cachep = slab_rcu->cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | |
| 1373 | kmem_freepages(cachep, slab_rcu->addr); |
| 1374 | if (OFF_SLAB(cachep)) |
| 1375 | kmem_cache_free(cachep->slabp_cache, slab_rcu); |
| 1376 | } |
| 1377 | |
| 1378 | #if DEBUG |
| 1379 | |
| 1380 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1381 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1382 | unsigned long caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1383 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1384 | int size = obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1385 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1386 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1387 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1388 | if (size < 5 * sizeof(unsigned long)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1389 | return; |
| 1390 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1391 | *addr++ = 0x12345678; |
| 1392 | *addr++ = caller; |
| 1393 | *addr++ = smp_processor_id(); |
| 1394 | size -= 3 * sizeof(unsigned long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1395 | { |
| 1396 | unsigned long *sptr = &caller; |
| 1397 | unsigned long svalue; |
| 1398 | |
| 1399 | while (!kstack_end(sptr)) { |
| 1400 | svalue = *sptr++; |
| 1401 | if (kernel_text_address(svalue)) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1402 | *addr++ = svalue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | size -= sizeof(unsigned long); |
| 1404 | if (size <= sizeof(unsigned long)) |
| 1405 | break; |
| 1406 | } |
| 1407 | } |
| 1408 | |
| 1409 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1410 | *addr++ = 0x87654321; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | } |
| 1412 | #endif |
| 1413 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1414 | 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] | 1415 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1416 | int size = obj_size(cachep); |
| 1417 | addr = &((char *)addr)[obj_offset(cachep)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1418 | |
| 1419 | memset(addr, val, size); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1420 | *(unsigned char *)(addr + size - 1) = POISON_END; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1421 | } |
| 1422 | |
| 1423 | static void dump_line(char *data, int offset, int limit) |
| 1424 | { |
| 1425 | int i; |
| 1426 | printk(KERN_ERR "%03x:", offset); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1427 | for (i = 0; i < limit; i++) { |
| 1428 | printk(" %02x", (unsigned char)data[offset + i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1429 | } |
| 1430 | printk("\n"); |
| 1431 | } |
| 1432 | #endif |
| 1433 | |
| 1434 | #if DEBUG |
| 1435 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1436 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | { |
| 1438 | int i, size; |
| 1439 | char *realobj; |
| 1440 | |
| 1441 | if (cachep->flags & SLAB_RED_ZONE) { |
| 1442 | printk(KERN_ERR "Redzone: 0x%lx/0x%lx.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1443 | *dbg_redzone1(cachep, objp), |
| 1444 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1445 | } |
| 1446 | |
| 1447 | if (cachep->flags & SLAB_STORE_USER) { |
| 1448 | printk(KERN_ERR "Last user: [<%p>]", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1449 | *dbg_userword(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | print_symbol("(%s)", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1451 | (unsigned long)*dbg_userword(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | printk("\n"); |
| 1453 | } |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1454 | realobj = (char *)objp + obj_offset(cachep); |
| 1455 | size = obj_size(cachep); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1456 | for (i = 0; i < size && lines; i += 16, lines--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | int limit; |
| 1458 | limit = 16; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1459 | if (i + limit > size) |
| 1460 | limit = size - i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1461 | dump_line(realobj, i, limit); |
| 1462 | } |
| 1463 | } |
| 1464 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1465 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | { |
| 1467 | char *realobj; |
| 1468 | int size, i; |
| 1469 | int lines = 0; |
| 1470 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1471 | realobj = (char *)objp + obj_offset(cachep); |
| 1472 | size = obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1474 | for (i = 0; i < size; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1475 | char exp = POISON_FREE; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1476 | if (i == size - 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | exp = POISON_END; |
| 1478 | if (realobj[i] != exp) { |
| 1479 | int limit; |
| 1480 | /* Mismatch ! */ |
| 1481 | /* Print header */ |
| 1482 | if (lines == 0) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1483 | printk(KERN_ERR |
| 1484 | "Slab corruption: start=%p, len=%d\n", |
| 1485 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | print_objinfo(cachep, objp, 0); |
| 1487 | } |
| 1488 | /* Hexdump the affected line */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1489 | i = (i / 16) * 16; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | limit = 16; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1491 | if (i + limit > size) |
| 1492 | limit = size - i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1493 | dump_line(realobj, i, limit); |
| 1494 | i += 16; |
| 1495 | lines++; |
| 1496 | /* Limit to 5 lines */ |
| 1497 | if (lines > 5) |
| 1498 | break; |
| 1499 | } |
| 1500 | } |
| 1501 | if (lines != 0) { |
| 1502 | /* Print some data about the neighboring objects, if they |
| 1503 | * exist: |
| 1504 | */ |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1505 | struct slab *slabp = virt_to_slab(objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | int objnr; |
| 1507 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1508 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1509 | if (objnr) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1510 | objp = slabp->s_mem + (objnr - 1) * cachep->buffer_size; |
| 1511 | realobj = (char *)objp + obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1513 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | print_objinfo(cachep, objp, 2); |
| 1515 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1516 | if (objnr + 1 < cachep->num) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1517 | objp = slabp->s_mem + (objnr + 1) * cachep->buffer_size; |
| 1518 | realobj = (char *)objp + obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1520 | realobj, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1521 | print_objinfo(cachep, objp, 2); |
| 1522 | } |
| 1523 | } |
| 1524 | } |
| 1525 | #endif |
| 1526 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | #if DEBUG |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1528 | /** |
| 1529 | * slab_destroy_objs - call the registered destructor for each object in |
| 1530 | * a slab that is to be destroyed. |
| 1531 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1532 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1533 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1534 | int i; |
| 1535 | for (i = 0; i < cachep->num; i++) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1536 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1537 | |
| 1538 | if (cachep->flags & SLAB_POISON) { |
| 1539 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1540 | if ((cachep->buffer_size % PAGE_SIZE) == 0 |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1541 | && OFF_SLAB(cachep)) |
| 1542 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1543 | cachep->buffer_size / PAGE_SIZE, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1544 | 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1545 | else |
| 1546 | check_poison_obj(cachep, objp); |
| 1547 | #else |
| 1548 | check_poison_obj(cachep, objp); |
| 1549 | #endif |
| 1550 | } |
| 1551 | if (cachep->flags & SLAB_RED_ZONE) { |
| 1552 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 1553 | slab_error(cachep, "start of a freed object " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1554 | "was overwritten"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1555 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 1556 | slab_error(cachep, "end of a freed object " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1557 | "was overwritten"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1558 | } |
| 1559 | if (cachep->dtor && !(cachep->flags & SLAB_POISON)) |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1560 | (cachep->dtor) (objp + obj_offset(cachep), cachep, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1561 | } |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1562 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1563 | #else |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1564 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1565 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | if (cachep->dtor) { |
| 1567 | int i; |
| 1568 | for (i = 0; i < cachep->num; i++) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1569 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1570 | (cachep->dtor) (objp, cachep, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1571 | } |
| 1572 | } |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1573 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | #endif |
| 1575 | |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1576 | /** |
| 1577 | * Destroy all the objs in a slab, and release the mem back to the system. |
| 1578 | * Before calling the slab must have been unlinked from the cache. |
| 1579 | * The cache-lock is not held/needed. |
| 1580 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1581 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
Matthew Dobson | 12dd36f | 2006-02-01 03:05:46 -0800 | [diff] [blame] | 1582 | { |
| 1583 | void *addr = slabp->s_mem - slabp->colouroff; |
| 1584 | |
| 1585 | slab_destroy_objs(cachep, slabp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
| 1587 | struct slab_rcu *slab_rcu; |
| 1588 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1589 | slab_rcu = (struct slab_rcu *)slabp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1590 | slab_rcu->cachep = cachep; |
| 1591 | slab_rcu->addr = addr; |
| 1592 | call_rcu(&slab_rcu->head, kmem_rcu_free); |
| 1593 | } else { |
| 1594 | kmem_freepages(cachep, addr); |
| 1595 | if (OFF_SLAB(cachep)) |
| 1596 | kmem_cache_free(cachep->slabp_cache, slabp); |
| 1597 | } |
| 1598 | } |
| 1599 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1600 | /* 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] | 1601 | as size of kmem_list3. */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1602 | static void set_up_list3s(struct kmem_cache *cachep, int index) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1603 | { |
| 1604 | int node; |
| 1605 | |
| 1606 | for_each_online_node(node) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1607 | cachep->nodelists[node] = &initkmem_list3[index + node]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1608 | cachep->nodelists[node]->next_reap = jiffies + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1609 | REAPTIMEOUT_LIST3 + |
| 1610 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1611 | } |
| 1612 | } |
| 1613 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | /** |
Randy.Dunlap | a70773d | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 1615 | * calculate_slab_order - calculate size (page order) of slabs |
| 1616 | * @cachep: pointer to the cache that is being created |
| 1617 | * @size: size of objects to be created in this cache. |
| 1618 | * @align: required alignment for the objects. |
| 1619 | * @flags: slab allocation flags |
| 1620 | * |
| 1621 | * Also calculates the number of objects per slab. |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1622 | * |
| 1623 | * This could be made much more intelligent. For now, try to avoid using |
| 1624 | * high order pages for slabs. When the gfp() functions are more friendly |
| 1625 | * towards high-order requests, this should be changed. |
| 1626 | */ |
Randy Dunlap | ee13d78 | 2006-02-01 03:05:53 -0800 | [diff] [blame] | 1627 | static inline size_t calculate_slab_order(struct kmem_cache *cachep, |
| 1628 | size_t size, size_t align, unsigned long flags) |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1629 | { |
| 1630 | size_t left_over = 0; |
| 1631 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1632 | for (;; cachep->gfporder++) { |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1633 | unsigned int num; |
| 1634 | size_t remainder; |
| 1635 | |
| 1636 | if (cachep->gfporder > MAX_GFP_ORDER) { |
| 1637 | cachep->num = 0; |
| 1638 | break; |
| 1639 | } |
| 1640 | |
| 1641 | cache_estimate(cachep->gfporder, size, align, flags, |
| 1642 | &remainder, &num); |
| 1643 | if (!num) |
| 1644 | continue; |
| 1645 | /* More than offslab_limit objects will cause problems */ |
| 1646 | if (flags & CFLGS_OFF_SLAB && cachep->num > offslab_limit) |
| 1647 | break; |
| 1648 | |
| 1649 | cachep->num = num; |
| 1650 | left_over = remainder; |
| 1651 | |
| 1652 | /* |
| 1653 | * Large number of objects is good, but very large slabs are |
| 1654 | * currently bad for the gfp()s. |
| 1655 | */ |
| 1656 | if (cachep->gfporder >= slab_break_gfp_order) |
| 1657 | break; |
| 1658 | |
| 1659 | if ((left_over * 8) <= (PAGE_SIZE << cachep->gfporder)) |
| 1660 | /* Acceptable internal fragmentation */ |
| 1661 | break; |
| 1662 | } |
| 1663 | return left_over; |
| 1664 | } |
| 1665 | |
| 1666 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1667 | * kmem_cache_create - Create a cache. |
| 1668 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 1669 | * @size: The size of objects to be created in this cache. |
| 1670 | * @align: The required alignment for the objects. |
| 1671 | * @flags: SLAB flags |
| 1672 | * @ctor: A constructor for the objects. |
| 1673 | * @dtor: A destructor for the objects. |
| 1674 | * |
| 1675 | * Returns a ptr to the cache on success, NULL on failure. |
| 1676 | * Cannot be called within a int, but can be interrupted. |
| 1677 | * The @ctor is run when new pages are allocated by the cache |
| 1678 | * and the @dtor is run before the pages are handed back. |
| 1679 | * |
| 1680 | * @name must be valid until the cache is destroyed. This implies that |
| 1681 | * the module calling this has to destroy the cache before getting |
| 1682 | * unloaded. |
| 1683 | * |
| 1684 | * The flags are |
| 1685 | * |
| 1686 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 1687 | * to catch references to uninitialised memory. |
| 1688 | * |
| 1689 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check |
| 1690 | * for buffer overruns. |
| 1691 | * |
| 1692 | * %SLAB_NO_REAP - Don't automatically reap this cache when we're under |
| 1693 | * memory pressure. |
| 1694 | * |
| 1695 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 1696 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 1697 | * as davem. |
| 1698 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1699 | struct kmem_cache * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1700 | kmem_cache_create (const char *name, size_t size, size_t align, |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1701 | unsigned long flags, void (*ctor)(void*, struct kmem_cache *, unsigned long), |
| 1702 | void (*dtor)(void*, struct kmem_cache *, unsigned long)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1703 | { |
| 1704 | size_t left_over, slab_size, ralign; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1705 | struct kmem_cache *cachep = NULL; |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1706 | struct list_head *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1707 | |
| 1708 | /* |
| 1709 | * Sanity checks... these are all serious usage bugs. |
| 1710 | */ |
| 1711 | if ((!name) || |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1712 | in_interrupt() || |
| 1713 | (size < BYTES_PER_WORD) || |
| 1714 | (size > (1 << MAX_OBJ_ORDER) * PAGE_SIZE) || (dtor && !ctor)) { |
| 1715 | printk(KERN_ERR "%s: Early error in slab %s\n", |
| 1716 | __FUNCTION__, name); |
| 1717 | BUG(); |
| 1718 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1719 | |
Ravikiran G Thirumalai | f0188f4 | 2006-02-10 01:51:13 -0800 | [diff] [blame] | 1720 | /* |
| 1721 | * Prevent CPUs from coming and going. |
| 1722 | * lock_cpu_hotplug() nests outside cache_chain_mutex |
| 1723 | */ |
| 1724 | lock_cpu_hotplug(); |
| 1725 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1726 | mutex_lock(&cache_chain_mutex); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1727 | |
| 1728 | list_for_each(p, &cache_chain) { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1729 | struct kmem_cache *pc = list_entry(p, struct kmem_cache, next); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1730 | mm_segment_t old_fs = get_fs(); |
| 1731 | char tmp; |
| 1732 | int res; |
| 1733 | |
| 1734 | /* |
| 1735 | * This happens when the module gets unloaded and doesn't |
| 1736 | * destroy its slab cache and no-one else reuses the vmalloc |
| 1737 | * area of the module. Print a warning. |
| 1738 | */ |
| 1739 | set_fs(KERNEL_DS); |
| 1740 | res = __get_user(tmp, pc->name); |
| 1741 | set_fs(old_fs); |
| 1742 | if (res) { |
| 1743 | printk("SLAB: cache with size %d has lost its name\n", |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1744 | pc->buffer_size); |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1745 | continue; |
| 1746 | } |
| 1747 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1748 | if (!strcmp(pc->name, name)) { |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1749 | printk("kmem_cache_create: duplicate cache %s\n", name); |
| 1750 | dump_stack(); |
| 1751 | goto oops; |
| 1752 | } |
| 1753 | } |
| 1754 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1755 | #if DEBUG |
| 1756 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
| 1757 | if ((flags & SLAB_DEBUG_INITIAL) && !ctor) { |
| 1758 | /* No constructor, but inital state check requested */ |
| 1759 | printk(KERN_ERR "%s: No con, but init state check " |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1760 | "requested - %s\n", __FUNCTION__, name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1761 | flags &= ~SLAB_DEBUG_INITIAL; |
| 1762 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1763 | #if FORCED_DEBUG |
| 1764 | /* |
| 1765 | * Enable redzoning and last user accounting, except for caches with |
| 1766 | * large objects, if the increased size would increase the object size |
| 1767 | * above the next power of two: caches with object sizes just above a |
| 1768 | * power of two have a significant amount of internal fragmentation. |
| 1769 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1770 | if ((size < 4096 |
| 1771 | || fls(size - 1) == fls(size - 1 + 3 * BYTES_PER_WORD))) |
| 1772 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
| 1774 | flags |= SLAB_POISON; |
| 1775 | #endif |
| 1776 | if (flags & SLAB_DESTROY_BY_RCU) |
| 1777 | BUG_ON(flags & SLAB_POISON); |
| 1778 | #endif |
| 1779 | if (flags & SLAB_DESTROY_BY_RCU) |
| 1780 | BUG_ON(dtor); |
| 1781 | |
| 1782 | /* |
| 1783 | * Always checks flags, a caller might be expecting debug |
| 1784 | * support which isn't available. |
| 1785 | */ |
| 1786 | if (flags & ~CREATE_MASK) |
| 1787 | BUG(); |
| 1788 | |
| 1789 | /* Check that size is in terms of words. This is needed to avoid |
| 1790 | * unaligned accesses for some archs when redzoning is used, and makes |
| 1791 | * sure any on-slab bufctl's are also correctly aligned. |
| 1792 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1793 | if (size & (BYTES_PER_WORD - 1)) { |
| 1794 | size += (BYTES_PER_WORD - 1); |
| 1795 | size &= ~(BYTES_PER_WORD - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1796 | } |
| 1797 | |
| 1798 | /* calculate out the final buffer alignment: */ |
| 1799 | /* 1) arch recommendation: can be overridden for debug */ |
| 1800 | if (flags & SLAB_HWCACHE_ALIGN) { |
| 1801 | /* Default alignment: as specified by the arch code. |
| 1802 | * Except if an object is really small, then squeeze multiple |
| 1803 | * objects into one cacheline. |
| 1804 | */ |
| 1805 | ralign = cache_line_size(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1806 | while (size <= ralign / 2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1807 | ralign /= 2; |
| 1808 | } else { |
| 1809 | ralign = BYTES_PER_WORD; |
| 1810 | } |
| 1811 | /* 2) arch mandated alignment: disables debug if necessary */ |
| 1812 | if (ralign < ARCH_SLAB_MINALIGN) { |
| 1813 | ralign = ARCH_SLAB_MINALIGN; |
| 1814 | if (ralign > BYTES_PER_WORD) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1815 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1816 | } |
| 1817 | /* 3) caller mandated alignment: disables debug if necessary */ |
| 1818 | if (ralign < align) { |
| 1819 | ralign = align; |
| 1820 | if (ralign > BYTES_PER_WORD) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1821 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1822 | } |
| 1823 | /* 4) Store it. Note that the debug code below can reduce |
| 1824 | * the alignment to BYTES_PER_WORD. |
| 1825 | */ |
| 1826 | align = ralign; |
| 1827 | |
| 1828 | /* Get cache's description obj. */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1829 | cachep = kmem_cache_alloc(&cache_cache, SLAB_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1830 | if (!cachep) |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1831 | goto oops; |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 1832 | memset(cachep, 0, sizeof(struct kmem_cache)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | |
| 1834 | #if DEBUG |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1835 | cachep->obj_size = size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1836 | |
| 1837 | if (flags & SLAB_RED_ZONE) { |
| 1838 | /* redzoning only works with word aligned caches */ |
| 1839 | align = BYTES_PER_WORD; |
| 1840 | |
| 1841 | /* add space for red zone words */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1842 | cachep->obj_offset += BYTES_PER_WORD; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1843 | size += 2 * BYTES_PER_WORD; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1844 | } |
| 1845 | if (flags & SLAB_STORE_USER) { |
| 1846 | /* user store requires word alignment and |
| 1847 | * one word storage behind the end of the real |
| 1848 | * object. |
| 1849 | */ |
| 1850 | align = BYTES_PER_WORD; |
| 1851 | size += BYTES_PER_WORD; |
| 1852 | } |
| 1853 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1854 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1855 | && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { |
| 1856 | cachep->obj_offset += PAGE_SIZE - size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1857 | size = PAGE_SIZE; |
| 1858 | } |
| 1859 | #endif |
| 1860 | #endif |
| 1861 | |
| 1862 | /* Determine if the slab management is 'on' or 'off' slab. */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1863 | if (size >= (PAGE_SIZE >> 3)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1864 | /* |
| 1865 | * Size is large, assume best to place the slab management obj |
| 1866 | * off-slab (should allow better packing of objs). |
| 1867 | */ |
| 1868 | flags |= CFLGS_OFF_SLAB; |
| 1869 | |
| 1870 | size = ALIGN(size, align); |
| 1871 | |
| 1872 | if ((flags & SLAB_RECLAIM_ACCOUNT) && size <= PAGE_SIZE) { |
| 1873 | /* |
| 1874 | * A VFS-reclaimable slab tends to have most allocations |
| 1875 | * as GFP_NOFS and we really don't want to have to be allocating |
| 1876 | * higher-order pages when we are unable to shrink dcache. |
| 1877 | */ |
| 1878 | cachep->gfporder = 0; |
| 1879 | cache_estimate(cachep->gfporder, size, align, flags, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1880 | &left_over, &cachep->num); |
Pekka Enberg | 4d268eb | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 1881 | } else |
| 1882 | left_over = calculate_slab_order(cachep, size, align, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1883 | |
| 1884 | if (!cachep->num) { |
| 1885 | printk("kmem_cache_create: couldn't create cache %s.\n", name); |
| 1886 | kmem_cache_free(&cache_cache, cachep); |
| 1887 | cachep = NULL; |
Andrew Morton | 4f12bb4 | 2005-11-07 00:58:00 -0800 | [diff] [blame] | 1888 | goto oops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1889 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1890 | slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) |
| 1891 | + sizeof(struct slab), align); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | |
| 1893 | /* |
| 1894 | * If the slab has been placed off-slab, and we have enough space then |
| 1895 | * move it on-slab. This is at the expense of any extra colouring. |
| 1896 | */ |
| 1897 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { |
| 1898 | flags &= ~CFLGS_OFF_SLAB; |
| 1899 | left_over -= slab_size; |
| 1900 | } |
| 1901 | |
| 1902 | if (flags & CFLGS_OFF_SLAB) { |
| 1903 | /* really off slab. No need for manual alignment */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1904 | slab_size = |
| 1905 | cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1906 | } |
| 1907 | |
| 1908 | cachep->colour_off = cache_line_size(); |
| 1909 | /* Offset must be a multiple of the alignment. */ |
| 1910 | if (cachep->colour_off < align) |
| 1911 | cachep->colour_off = align; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1912 | cachep->colour = left_over / cachep->colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1913 | cachep->slab_size = slab_size; |
| 1914 | cachep->flags = flags; |
| 1915 | cachep->gfpflags = 0; |
| 1916 | if (flags & SLAB_CACHE_DMA) |
| 1917 | cachep->gfpflags |= GFP_DMA; |
| 1918 | spin_lock_init(&cachep->spinlock); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 1919 | cachep->buffer_size = size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1920 | |
| 1921 | if (flags & CFLGS_OFF_SLAB) |
Victor Fusco | b2d5507 | 2005-09-10 00:26:36 -0700 | [diff] [blame] | 1922 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1923 | cachep->ctor = ctor; |
| 1924 | cachep->dtor = dtor; |
| 1925 | cachep->name = name; |
| 1926 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1927 | |
| 1928 | if (g_cpucache_up == FULL) { |
| 1929 | enable_cpucache(cachep); |
| 1930 | } else { |
| 1931 | if (g_cpucache_up == NONE) { |
| 1932 | /* Note: the first kmem_cache_create must create |
| 1933 | * the cache that's used by kmalloc(24), otherwise |
| 1934 | * the creation of further caches will BUG(). |
| 1935 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1936 | cachep->array[smp_processor_id()] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1937 | &initarray_generic.cache; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1938 | |
| 1939 | /* If the cache that's used by |
| 1940 | * kmalloc(sizeof(kmem_list3)) is the first cache, |
| 1941 | * then we need to set up all its list3s, otherwise |
| 1942 | * the creation of further caches will BUG(). |
| 1943 | */ |
| 1944 | set_up_list3s(cachep, SIZE_AC); |
| 1945 | if (INDEX_AC == INDEX_L3) |
| 1946 | g_cpucache_up = PARTIAL_L3; |
| 1947 | else |
| 1948 | g_cpucache_up = PARTIAL_AC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1949 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1950 | cachep->array[smp_processor_id()] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1951 | kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1952 | |
| 1953 | if (g_cpucache_up == PARTIAL_AC) { |
| 1954 | set_up_list3s(cachep, SIZE_L3); |
| 1955 | g_cpucache_up = PARTIAL_L3; |
| 1956 | } else { |
| 1957 | int node; |
| 1958 | for_each_online_node(node) { |
| 1959 | |
| 1960 | cachep->nodelists[node] = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1961 | kmalloc_node(sizeof |
| 1962 | (struct kmem_list3), |
| 1963 | GFP_KERNEL, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1964 | BUG_ON(!cachep->nodelists[node]); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1965 | kmem_list3_init(cachep-> |
| 1966 | nodelists[node]); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1967 | } |
| 1968 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1969 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1970 | cachep->nodelists[numa_node_id()]->next_reap = |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1971 | jiffies + REAPTIMEOUT_LIST3 + |
| 1972 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 1973 | |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 1974 | BUG_ON(!cpu_cache_get(cachep)); |
| 1975 | cpu_cache_get(cachep)->avail = 0; |
| 1976 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; |
| 1977 | cpu_cache_get(cachep)->batchcount = 1; |
| 1978 | cpu_cache_get(cachep)->touched = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1979 | cachep->batchcount = 1; |
| 1980 | cachep->limit = BOOT_CPUCACHE_ENTRIES; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1981 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1982 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1983 | /* cache setup completed, link it into the list */ |
| 1984 | list_add(&cachep->next, &cache_chain); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1985 | oops: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1986 | if (!cachep && (flags & SLAB_PANIC)) |
| 1987 | panic("kmem_cache_create(): failed to create slab `%s'\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 1988 | name); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 1989 | mutex_unlock(&cache_chain_mutex); |
Ravikiran G Thirumalai | f0188f4 | 2006-02-10 01:51:13 -0800 | [diff] [blame] | 1990 | unlock_cpu_hotplug(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1991 | return cachep; |
| 1992 | } |
| 1993 | EXPORT_SYMBOL(kmem_cache_create); |
| 1994 | |
| 1995 | #if DEBUG |
| 1996 | static void check_irq_off(void) |
| 1997 | { |
| 1998 | BUG_ON(!irqs_disabled()); |
| 1999 | } |
| 2000 | |
| 2001 | static void check_irq_on(void) |
| 2002 | { |
| 2003 | BUG_ON(irqs_disabled()); |
| 2004 | } |
| 2005 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2006 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2007 | { |
| 2008 | #ifdef CONFIG_SMP |
| 2009 | check_irq_off(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2010 | assert_spin_locked(&cachep->nodelists[numa_node_id()]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2011 | #endif |
| 2012 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2013 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2014 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2015 | { |
| 2016 | #ifdef CONFIG_SMP |
| 2017 | check_irq_off(); |
| 2018 | assert_spin_locked(&cachep->nodelists[node]->list_lock); |
| 2019 | #endif |
| 2020 | } |
| 2021 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2022 | #else |
| 2023 | #define check_irq_off() do { } while(0) |
| 2024 | #define check_irq_on() do { } while(0) |
| 2025 | #define check_spinlock_acquired(x) do { } while(0) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2026 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2027 | #endif |
| 2028 | |
| 2029 | /* |
| 2030 | * Waits for all CPUs to execute func(). |
| 2031 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2032 | 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] | 2033 | { |
| 2034 | check_irq_on(); |
| 2035 | preempt_disable(); |
| 2036 | |
| 2037 | local_irq_disable(); |
| 2038 | func(arg); |
| 2039 | local_irq_enable(); |
| 2040 | |
| 2041 | if (smp_call_function(func, arg, 1, 1)) |
| 2042 | BUG(); |
| 2043 | |
| 2044 | preempt_enable(); |
| 2045 | } |
| 2046 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2047 | 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] | 2048 | int force, int node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2049 | |
| 2050 | static void do_drain(void *arg) |
| 2051 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2052 | struct kmem_cache *cachep = (struct kmem_cache *) arg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2053 | struct array_cache *ac; |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2054 | int node = numa_node_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2055 | |
| 2056 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2057 | ac = cpu_cache_get(cachep); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2058 | spin_lock(&cachep->nodelists[node]->list_lock); |
| 2059 | free_block(cachep, ac->entry, ac->avail, node); |
| 2060 | spin_unlock(&cachep->nodelists[node]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2061 | ac->avail = 0; |
| 2062 | } |
| 2063 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2064 | static void drain_cpu_caches(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2065 | { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2066 | struct kmem_list3 *l3; |
| 2067 | int node; |
| 2068 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2069 | smp_call_function_all_cpus(do_drain, cachep); |
| 2070 | check_irq_on(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2071 | for_each_online_node(node) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2072 | l3 = cachep->nodelists[node]; |
| 2073 | if (l3) { |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 2074 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2075 | drain_array_locked(cachep, l3->shared, 1, node); |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 2076 | spin_unlock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2077 | if (l3->alien) |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 2078 | drain_alien_cache(cachep, l3->alien); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2079 | } |
| 2080 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2081 | } |
| 2082 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2083 | static int __node_shrink(struct kmem_cache *cachep, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2084 | { |
| 2085 | struct slab *slabp; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2086 | struct kmem_list3 *l3 = cachep->nodelists[node]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2087 | int ret; |
| 2088 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2089 | for (;;) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2090 | struct list_head *p; |
| 2091 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2092 | p = l3->slabs_free.prev; |
| 2093 | if (p == &l3->slabs_free) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | break; |
| 2095 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2096 | slabp = list_entry(l3->slabs_free.prev, struct slab, list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2097 | #if DEBUG |
| 2098 | if (slabp->inuse) |
| 2099 | BUG(); |
| 2100 | #endif |
| 2101 | list_del(&slabp->list); |
| 2102 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2103 | l3->free_objects -= cachep->num; |
| 2104 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2105 | slab_destroy(cachep, slabp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2106 | spin_lock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2107 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2108 | ret = !list_empty(&l3->slabs_full) || !list_empty(&l3->slabs_partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2109 | return ret; |
| 2110 | } |
| 2111 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2112 | static int __cache_shrink(struct kmem_cache *cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2113 | { |
| 2114 | int ret = 0, i = 0; |
| 2115 | struct kmem_list3 *l3; |
| 2116 | |
| 2117 | drain_cpu_caches(cachep); |
| 2118 | |
| 2119 | check_irq_on(); |
| 2120 | for_each_online_node(i) { |
| 2121 | l3 = cachep->nodelists[i]; |
| 2122 | if (l3) { |
| 2123 | spin_lock_irq(&l3->list_lock); |
| 2124 | ret += __node_shrink(cachep, i); |
| 2125 | spin_unlock_irq(&l3->list_lock); |
| 2126 | } |
| 2127 | } |
| 2128 | return (ret ? 1 : 0); |
| 2129 | } |
| 2130 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | /** |
| 2132 | * kmem_cache_shrink - Shrink a cache. |
| 2133 | * @cachep: The cache to shrink. |
| 2134 | * |
| 2135 | * Releases as many slabs as possible for a cache. |
| 2136 | * To help debugging, a zero exit status indicates all slabs were released. |
| 2137 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2138 | int kmem_cache_shrink(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2139 | { |
| 2140 | if (!cachep || in_interrupt()) |
| 2141 | BUG(); |
| 2142 | |
| 2143 | return __cache_shrink(cachep); |
| 2144 | } |
| 2145 | EXPORT_SYMBOL(kmem_cache_shrink); |
| 2146 | |
| 2147 | /** |
| 2148 | * kmem_cache_destroy - delete a cache |
| 2149 | * @cachep: the cache to destroy |
| 2150 | * |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2151 | * Remove a struct kmem_cache object from the slab cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | * Returns 0 on success. |
| 2153 | * |
| 2154 | * It is expected this function will be called by a module when it is |
| 2155 | * unloaded. This will remove the cache completely, and avoid a duplicate |
| 2156 | * cache being allocated each time a module is loaded and unloaded, if the |
| 2157 | * module doesn't have persistent in-kernel storage across loads and unloads. |
| 2158 | * |
| 2159 | * The cache must be empty before calling this function. |
| 2160 | * |
| 2161 | * The caller must guarantee that noone will allocate memory from the cache |
| 2162 | * during the kmem_cache_destroy(). |
| 2163 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2164 | int kmem_cache_destroy(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2165 | { |
| 2166 | int i; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2167 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | |
| 2169 | if (!cachep || in_interrupt()) |
| 2170 | BUG(); |
| 2171 | |
| 2172 | /* Don't let CPUs to come and go */ |
| 2173 | lock_cpu_hotplug(); |
| 2174 | |
| 2175 | /* Find the cache in the chain of caches. */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2176 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2177 | /* |
| 2178 | * the chain is never empty, cache_cache is never destroyed |
| 2179 | */ |
| 2180 | list_del(&cachep->next); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2181 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2182 | |
| 2183 | if (__cache_shrink(cachep)) { |
| 2184 | slab_error(cachep, "Can't free all objects"); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2185 | mutex_lock(&cache_chain_mutex); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2186 | list_add(&cachep->next, &cache_chain); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 2187 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2188 | unlock_cpu_hotplug(); |
| 2189 | return 1; |
| 2190 | } |
| 2191 | |
| 2192 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
Paul E. McKenney | fbd568a3e | 2005-05-01 08:59:04 -0700 | [diff] [blame] | 2193 | synchronize_rcu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2195 | for_each_online_cpu(i) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2196 | kfree(cachep->array[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2197 | |
| 2198 | /* NUMA: free the list3 structures */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2199 | for_each_online_node(i) { |
| 2200 | if ((l3 = cachep->nodelists[i])) { |
| 2201 | kfree(l3->shared); |
| 2202 | free_alien_cache(l3->alien); |
| 2203 | kfree(l3); |
| 2204 | } |
| 2205 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2206 | kmem_cache_free(&cache_cache, cachep); |
| 2207 | |
| 2208 | unlock_cpu_hotplug(); |
| 2209 | |
| 2210 | return 0; |
| 2211 | } |
| 2212 | EXPORT_SYMBOL(kmem_cache_destroy); |
| 2213 | |
| 2214 | /* Get the memory for a slab management obj. */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2215 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2216 | int colour_off, gfp_t local_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2217 | { |
| 2218 | struct slab *slabp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2219 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2220 | if (OFF_SLAB(cachep)) { |
| 2221 | /* Slab management obj is off-slab. */ |
| 2222 | slabp = kmem_cache_alloc(cachep->slabp_cache, local_flags); |
| 2223 | if (!slabp) |
| 2224 | return NULL; |
| 2225 | } else { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2226 | slabp = objp + colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2227 | colour_off += cachep->slab_size; |
| 2228 | } |
| 2229 | slabp->inuse = 0; |
| 2230 | slabp->colouroff = colour_off; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2231 | slabp->s_mem = objp + colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2232 | |
| 2233 | return slabp; |
| 2234 | } |
| 2235 | |
| 2236 | static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) |
| 2237 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2238 | return (kmem_bufctl_t *) (slabp + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2239 | } |
| 2240 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2241 | static void cache_init_objs(struct kmem_cache *cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2242 | struct slab *slabp, unsigned long ctor_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2243 | { |
| 2244 | int i; |
| 2245 | |
| 2246 | for (i = 0; i < cachep->num; i++) { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2247 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2248 | #if DEBUG |
| 2249 | /* need to poison the objs? */ |
| 2250 | if (cachep->flags & SLAB_POISON) |
| 2251 | poison_obj(cachep, objp, POISON_FREE); |
| 2252 | if (cachep->flags & SLAB_STORE_USER) |
| 2253 | *dbg_userword(cachep, objp) = NULL; |
| 2254 | |
| 2255 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2256 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2257 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2258 | } |
| 2259 | /* |
| 2260 | * Constructors are not allowed to allocate memory from |
| 2261 | * the same cache which they are a constructor for. |
| 2262 | * Otherwise, deadlock. They must also be threaded. |
| 2263 | */ |
| 2264 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2265 | cachep->ctor(objp + obj_offset(cachep), cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2266 | ctor_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2267 | |
| 2268 | if (cachep->flags & SLAB_RED_ZONE) { |
| 2269 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
| 2270 | slab_error(cachep, "constructor overwrote the" |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2271 | " end of an object"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2272 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
| 2273 | slab_error(cachep, "constructor overwrote the" |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2274 | " start of an object"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2275 | } |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2276 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2277 | && cachep->flags & SLAB_POISON) |
| 2278 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2279 | cachep->buffer_size / PAGE_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2280 | #else |
| 2281 | if (cachep->ctor) |
| 2282 | cachep->ctor(objp, cachep, ctor_flags); |
| 2283 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2284 | slab_bufctl(slabp)[i] = i + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2285 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2286 | slab_bufctl(slabp)[i - 1] = BUFCTL_END; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2287 | slabp->free = 0; |
| 2288 | } |
| 2289 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2290 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2291 | { |
| 2292 | if (flags & SLAB_DMA) { |
| 2293 | if (!(cachep->gfpflags & GFP_DMA)) |
| 2294 | BUG(); |
| 2295 | } else { |
| 2296 | if (cachep->gfpflags & GFP_DMA) |
| 2297 | BUG(); |
| 2298 | } |
| 2299 | } |
| 2300 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2301 | 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] | 2302 | { |
| 2303 | void *objp = slabp->s_mem + (slabp->free * cachep->buffer_size); |
| 2304 | kmem_bufctl_t next; |
| 2305 | |
| 2306 | slabp->inuse++; |
| 2307 | next = slab_bufctl(slabp)[slabp->free]; |
| 2308 | #if DEBUG |
| 2309 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; |
| 2310 | WARN_ON(slabp->nodeid != nodeid); |
| 2311 | #endif |
| 2312 | slabp->free = next; |
| 2313 | |
| 2314 | return objp; |
| 2315 | } |
| 2316 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2317 | 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] | 2318 | int nodeid) |
| 2319 | { |
| 2320 | unsigned int objnr = (unsigned)(objp-slabp->s_mem) / cachep->buffer_size; |
| 2321 | |
| 2322 | #if DEBUG |
| 2323 | /* Verify that the slab belongs to the intended node */ |
| 2324 | WARN_ON(slabp->nodeid != nodeid); |
| 2325 | |
| 2326 | if (slab_bufctl(slabp)[objnr] != BUFCTL_FREE) { |
| 2327 | printk(KERN_ERR "slab: double free detected in cache " |
| 2328 | "'%s', objp %p\n", cachep->name, objp); |
| 2329 | BUG(); |
| 2330 | } |
| 2331 | #endif |
| 2332 | slab_bufctl(slabp)[objnr] = slabp->free; |
| 2333 | slabp->free = objnr; |
| 2334 | slabp->inuse--; |
| 2335 | } |
| 2336 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2337 | 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] | 2338 | { |
| 2339 | int i; |
| 2340 | struct page *page; |
| 2341 | |
| 2342 | /* Nasty!!!!!! I hope this is OK. */ |
| 2343 | i = 1 << cachep->gfporder; |
| 2344 | page = virt_to_page(objp); |
| 2345 | do { |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 2346 | page_set_cache(page, cachep); |
| 2347 | page_set_slab(page, slabp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | page++; |
| 2349 | } while (--i); |
| 2350 | } |
| 2351 | |
| 2352 | /* |
| 2353 | * Grow (by 1) the number of slabs within a cache. This is called by |
| 2354 | * kmem_cache_alloc() when there are no active objs left in a cache. |
| 2355 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2356 | 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] | 2357 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2358 | struct slab *slabp; |
| 2359 | void *objp; |
| 2360 | size_t offset; |
| 2361 | gfp_t local_flags; |
| 2362 | unsigned long ctor_flags; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2363 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2364 | |
| 2365 | /* Be lazy and only check for valid flags here, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2366 | * keeping it out of the critical path in kmem_cache_alloc(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2367 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2368 | if (flags & ~(SLAB_DMA | SLAB_LEVEL_MASK | SLAB_NO_GROW)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2369 | BUG(); |
| 2370 | if (flags & SLAB_NO_GROW) |
| 2371 | return 0; |
| 2372 | |
| 2373 | ctor_flags = SLAB_CTOR_CONSTRUCTOR; |
| 2374 | local_flags = (flags & SLAB_LEVEL_MASK); |
| 2375 | if (!(local_flags & __GFP_WAIT)) |
| 2376 | /* |
| 2377 | * Not allowed to sleep. Need to tell a constructor about |
| 2378 | * this - it might need to know... |
| 2379 | */ |
| 2380 | ctor_flags |= SLAB_CTOR_ATOMIC; |
| 2381 | |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2382 | /* 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] | 2383 | check_irq_off(); |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2384 | l3 = cachep->nodelists[nodeid]; |
| 2385 | spin_lock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2386 | |
| 2387 | /* Get colour for the slab, and cal the next value. */ |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2388 | offset = l3->colour_next; |
| 2389 | l3->colour_next++; |
| 2390 | if (l3->colour_next >= cachep->colour) |
| 2391 | l3->colour_next = 0; |
| 2392 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2393 | |
Ravikiran G Thirumalai | 2e1217c | 2006-02-04 23:27:56 -0800 | [diff] [blame] | 2394 | offset *= cachep->colour_off; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2395 | |
| 2396 | if (local_flags & __GFP_WAIT) |
| 2397 | local_irq_enable(); |
| 2398 | |
| 2399 | /* |
| 2400 | * The test for missing atomic flag is performed here, rather than |
| 2401 | * the more obvious place, simply to reduce the critical path length |
| 2402 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they |
| 2403 | * will eventually be caught here (where it matters). |
| 2404 | */ |
| 2405 | kmem_flagcheck(cachep, flags); |
| 2406 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2407 | /* Get mem for the objs. |
| 2408 | * Attempt to allocate a physical page from 'nodeid', |
| 2409 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2410 | if (!(objp = kmem_getpages(cachep, flags, nodeid))) |
| 2411 | goto failed; |
| 2412 | |
| 2413 | /* Get slab management. */ |
| 2414 | if (!(slabp = alloc_slabmgmt(cachep, objp, offset, local_flags))) |
| 2415 | goto opps1; |
| 2416 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2417 | slabp->nodeid = nodeid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2418 | set_slab_attr(cachep, slabp, objp); |
| 2419 | |
| 2420 | cache_init_objs(cachep, slabp, ctor_flags); |
| 2421 | |
| 2422 | if (local_flags & __GFP_WAIT) |
| 2423 | local_irq_disable(); |
| 2424 | check_irq_off(); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2425 | spin_lock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2426 | |
| 2427 | /* Make slab active. */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2428 | list_add_tail(&slabp->list, &(l3->slabs_free)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2429 | STATS_INC_GROWN(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2430 | l3->free_objects += cachep->num; |
| 2431 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2432 | return 1; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2433 | opps1: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2434 | kmem_freepages(cachep, objp); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2435 | failed: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2436 | if (local_flags & __GFP_WAIT) |
| 2437 | local_irq_disable(); |
| 2438 | return 0; |
| 2439 | } |
| 2440 | |
| 2441 | #if DEBUG |
| 2442 | |
| 2443 | /* |
| 2444 | * Perform extra freeing checks: |
| 2445 | * - detect bad pointers. |
| 2446 | * - POISON/RED_ZONE checking |
| 2447 | * - destructor calls, for caches with POISON+dtor |
| 2448 | */ |
| 2449 | static void kfree_debugcheck(const void *objp) |
| 2450 | { |
| 2451 | struct page *page; |
| 2452 | |
| 2453 | if (!virt_addr_valid(objp)) { |
| 2454 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2455 | (unsigned long)objp); |
| 2456 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2457 | } |
| 2458 | page = virt_to_page(objp); |
| 2459 | if (!PageSlab(page)) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2460 | printk(KERN_ERR "kfree_debugcheck: bad ptr %lxh.\n", |
| 2461 | (unsigned long)objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2462 | BUG(); |
| 2463 | } |
| 2464 | } |
| 2465 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2466 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2467 | void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2468 | { |
| 2469 | struct page *page; |
| 2470 | unsigned int objnr; |
| 2471 | struct slab *slabp; |
| 2472 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2473 | objp -= obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2474 | kfree_debugcheck(objp); |
| 2475 | page = virt_to_page(objp); |
| 2476 | |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 2477 | if (page_get_cache(page) != cachep) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2478 | printk(KERN_ERR |
| 2479 | "mismatch in kmem_cache_free: expected cache %p, got %p\n", |
| 2480 | page_get_cache(page), cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2481 | printk(KERN_ERR "%p is %s.\n", cachep, cachep->name); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2482 | printk(KERN_ERR "%p is %s.\n", page_get_cache(page), |
| 2483 | page_get_cache(page)->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2484 | WARN_ON(1); |
| 2485 | } |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 2486 | slabp = page_get_slab(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2487 | |
| 2488 | if (cachep->flags & SLAB_RED_ZONE) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2489 | if (*dbg_redzone1(cachep, objp) != RED_ACTIVE |
| 2490 | || *dbg_redzone2(cachep, objp) != RED_ACTIVE) { |
| 2491 | slab_error(cachep, |
| 2492 | "double free, or memory outside" |
| 2493 | " object was overwritten"); |
| 2494 | printk(KERN_ERR |
| 2495 | "%p: redzone 1: 0x%lx, redzone 2: 0x%lx.\n", |
| 2496 | objp, *dbg_redzone1(cachep, objp), |
| 2497 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2498 | } |
| 2499 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
| 2500 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; |
| 2501 | } |
| 2502 | if (cachep->flags & SLAB_STORE_USER) |
| 2503 | *dbg_userword(cachep, objp) = caller; |
| 2504 | |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2505 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2506 | |
| 2507 | BUG_ON(objnr >= cachep->num); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2508 | BUG_ON(objp != slabp->s_mem + objnr * cachep->buffer_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2509 | |
| 2510 | if (cachep->flags & SLAB_DEBUG_INITIAL) { |
| 2511 | /* Need to call the slab's constructor so the |
| 2512 | * caller can perform a verify of its state (debugging). |
| 2513 | * Called without the cache-lock held. |
| 2514 | */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2515 | cachep->ctor(objp + obj_offset(cachep), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2516 | cachep, SLAB_CTOR_CONSTRUCTOR | SLAB_CTOR_VERIFY); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2517 | } |
| 2518 | if (cachep->flags & SLAB_POISON && cachep->dtor) { |
| 2519 | /* we want to cache poison the object, |
| 2520 | * call the destruction callback |
| 2521 | */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2522 | cachep->dtor(objp + obj_offset(cachep), cachep, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2523 | } |
| 2524 | if (cachep->flags & SLAB_POISON) { |
| 2525 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2526 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2527 | store_stackinfo(cachep, objp, (unsigned long)caller); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2528 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2529 | cachep->buffer_size / PAGE_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2530 | } else { |
| 2531 | poison_obj(cachep, objp, POISON_FREE); |
| 2532 | } |
| 2533 | #else |
| 2534 | poison_obj(cachep, objp, POISON_FREE); |
| 2535 | #endif |
| 2536 | } |
| 2537 | return objp; |
| 2538 | } |
| 2539 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2540 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2541 | { |
| 2542 | kmem_bufctl_t i; |
| 2543 | int entries = 0; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2544 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2545 | /* Check slab's freelist to see if this obj is there. */ |
| 2546 | for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { |
| 2547 | entries++; |
| 2548 | if (entries > cachep->num || i >= cachep->num) |
| 2549 | goto bad; |
| 2550 | } |
| 2551 | if (entries != cachep->num - slabp->inuse) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2552 | bad: |
| 2553 | printk(KERN_ERR |
| 2554 | "slab: Internal list corruption detected in cache '%s'(%d), slabp %p(%d). Hexdump:\n", |
| 2555 | cachep->name, cachep->num, slabp, slabp->inuse); |
| 2556 | for (i = 0; |
Linus Torvalds | 264132b | 2006-03-06 12:10:07 -0800 | [diff] [blame] | 2557 | i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2558 | i++) { |
| 2559 | if ((i % 16) == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2560 | printk("\n%03x:", i); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2561 | printk(" %02x", ((unsigned char *)slabp)[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2562 | } |
| 2563 | printk("\n"); |
| 2564 | BUG(); |
| 2565 | } |
| 2566 | } |
| 2567 | #else |
| 2568 | #define kfree_debugcheck(x) do { } while(0) |
| 2569 | #define cache_free_debugcheck(x,objp,z) (objp) |
| 2570 | #define check_slabp(x,y) do { } while(0) |
| 2571 | #endif |
| 2572 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2573 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2574 | { |
| 2575 | int batchcount; |
| 2576 | struct kmem_list3 *l3; |
| 2577 | struct array_cache *ac; |
| 2578 | |
| 2579 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2580 | ac = cpu_cache_get(cachep); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2581 | retry: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2582 | batchcount = ac->batchcount; |
| 2583 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { |
| 2584 | /* if there was little recent activity on this |
| 2585 | * cache, then perform only a partial refill. |
| 2586 | * Otherwise we could generate refill bouncing. |
| 2587 | */ |
| 2588 | batchcount = BATCHREFILL_LIMIT; |
| 2589 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2590 | l3 = cachep->nodelists[numa_node_id()]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2591 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2592 | BUG_ON(ac->avail > 0 || !l3); |
| 2593 | spin_lock(&l3->list_lock); |
| 2594 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | if (l3->shared) { |
| 2596 | struct array_cache *shared_array = l3->shared; |
| 2597 | if (shared_array->avail) { |
| 2598 | if (batchcount > shared_array->avail) |
| 2599 | batchcount = shared_array->avail; |
| 2600 | shared_array->avail -= batchcount; |
| 2601 | ac->avail = batchcount; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2602 | memcpy(ac->entry, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2603 | &(shared_array->entry[shared_array->avail]), |
| 2604 | sizeof(void *) * batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2605 | shared_array->touched = 1; |
| 2606 | goto alloc_done; |
| 2607 | } |
| 2608 | } |
| 2609 | while (batchcount > 0) { |
| 2610 | struct list_head *entry; |
| 2611 | struct slab *slabp; |
| 2612 | /* Get slab alloc is to come from. */ |
| 2613 | entry = l3->slabs_partial.next; |
| 2614 | if (entry == &l3->slabs_partial) { |
| 2615 | l3->free_touched = 1; |
| 2616 | entry = l3->slabs_free.next; |
| 2617 | if (entry == &l3->slabs_free) |
| 2618 | goto must_grow; |
| 2619 | } |
| 2620 | |
| 2621 | slabp = list_entry(entry, struct slab, list); |
| 2622 | check_slabp(cachep, slabp); |
| 2623 | check_spinlock_acquired(cachep); |
| 2624 | while (slabp->inuse < cachep->num && batchcount--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2625 | STATS_INC_ALLOCED(cachep); |
| 2626 | STATS_INC_ACTIVE(cachep); |
| 2627 | STATS_SET_HIGH(cachep); |
| 2628 | |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2629 | ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, |
| 2630 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | } |
| 2632 | check_slabp(cachep, slabp); |
| 2633 | |
| 2634 | /* move slabp to correct slabp list: */ |
| 2635 | list_del(&slabp->list); |
| 2636 | if (slabp->free == BUFCTL_END) |
| 2637 | list_add(&slabp->list, &l3->slabs_full); |
| 2638 | else |
| 2639 | list_add(&slabp->list, &l3->slabs_partial); |
| 2640 | } |
| 2641 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2642 | must_grow: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2643 | l3->free_objects -= ac->avail; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2644 | alloc_done: |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2645 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2646 | |
| 2647 | if (unlikely(!ac->avail)) { |
| 2648 | int x; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2649 | x = cache_grow(cachep, flags, numa_node_id()); |
| 2650 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2651 | // cache_grow can reenable interrupts, then ac could change. |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2652 | ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2653 | if (!x && ac->avail == 0) // no objects in sight? abort |
| 2654 | return NULL; |
| 2655 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2656 | if (!ac->avail) // objects refilled by interrupt? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2657 | goto retry; |
| 2658 | } |
| 2659 | ac->touched = 1; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2660 | return ac->entry[--ac->avail]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2661 | } |
| 2662 | |
| 2663 | static inline void |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2664 | cache_alloc_debugcheck_before(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2665 | { |
| 2666 | might_sleep_if(flags & __GFP_WAIT); |
| 2667 | #if DEBUG |
| 2668 | kmem_flagcheck(cachep, flags); |
| 2669 | #endif |
| 2670 | } |
| 2671 | |
| 2672 | #if DEBUG |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2673 | 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] | 2674 | void *objp, void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2675 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2676 | if (!objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2677 | return objp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2678 | if (cachep->flags & SLAB_POISON) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2679 | #ifdef CONFIG_DEBUG_PAGEALLOC |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2680 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2681 | kernel_map_pages(virt_to_page(objp), |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2682 | cachep->buffer_size / PAGE_SIZE, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2683 | else |
| 2684 | check_poison_obj(cachep, objp); |
| 2685 | #else |
| 2686 | check_poison_obj(cachep, objp); |
| 2687 | #endif |
| 2688 | poison_obj(cachep, objp, POISON_INUSE); |
| 2689 | } |
| 2690 | if (cachep->flags & SLAB_STORE_USER) |
| 2691 | *dbg_userword(cachep, objp) = caller; |
| 2692 | |
| 2693 | if (cachep->flags & SLAB_RED_ZONE) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2694 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE |
| 2695 | || *dbg_redzone2(cachep, objp) != RED_INACTIVE) { |
| 2696 | slab_error(cachep, |
| 2697 | "double free, or memory outside" |
| 2698 | " object was overwritten"); |
| 2699 | printk(KERN_ERR |
| 2700 | "%p: redzone 1: 0x%lx, redzone 2: 0x%lx.\n", |
| 2701 | objp, *dbg_redzone1(cachep, objp), |
| 2702 | *dbg_redzone2(cachep, objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2703 | } |
| 2704 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; |
| 2705 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; |
| 2706 | } |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 2707 | objp += obj_offset(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2708 | if (cachep->ctor && cachep->flags & SLAB_POISON) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2709 | unsigned long ctor_flags = SLAB_CTOR_CONSTRUCTOR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2710 | |
| 2711 | if (!(flags & __GFP_WAIT)) |
| 2712 | ctor_flags |= SLAB_CTOR_ATOMIC; |
| 2713 | |
| 2714 | cachep->ctor(objp, cachep, ctor_flags); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2715 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2716 | return objp; |
| 2717 | } |
| 2718 | #else |
| 2719 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) |
| 2720 | #endif |
| 2721 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2722 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2723 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2724 | void *objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2725 | struct array_cache *ac; |
| 2726 | |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 2727 | #ifdef CONFIG_NUMA |
Christoph Lameter | 86c562a | 2006-01-18 17:42:37 -0800 | [diff] [blame] | 2728 | if (unlikely(current->mempolicy && !in_interrupt())) { |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 2729 | int nid = slab_node(current->mempolicy); |
| 2730 | |
| 2731 | if (nid != numa_node_id()) |
| 2732 | return __cache_alloc_node(cachep, flags, nid); |
| 2733 | } |
| 2734 | #endif |
| 2735 | |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2736 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2737 | ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2738 | if (likely(ac->avail)) { |
| 2739 | STATS_INC_ALLOCHIT(cachep); |
| 2740 | ac->touched = 1; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2741 | objp = ac->entry[--ac->avail]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2742 | } else { |
| 2743 | STATS_INC_ALLOCMISS(cachep); |
| 2744 | objp = cache_alloc_refill(cachep, flags); |
| 2745 | } |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2746 | return objp; |
| 2747 | } |
| 2748 | |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 2749 | static __always_inline void * |
| 2750 | __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller) |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2751 | { |
| 2752 | unsigned long save_flags; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2753 | void *objp; |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 2754 | |
| 2755 | cache_alloc_debugcheck_before(cachep, flags); |
| 2756 | |
| 2757 | local_irq_save(save_flags); |
| 2758 | objp = ____cache_alloc(cachep, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2759 | local_irq_restore(save_flags); |
Eric Dumazet | 34342e8 | 2005-09-03 15:55:06 -0700 | [diff] [blame] | 2760 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 2761 | caller); |
Eric Dumazet | 34342e8 | 2005-09-03 15:55:06 -0700 | [diff] [blame] | 2762 | prefetchw(objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2763 | return objp; |
| 2764 | } |
| 2765 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2766 | #ifdef CONFIG_NUMA |
| 2767 | /* |
| 2768 | * A interface to enable slab creation on nodeid |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2769 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2770 | 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] | 2771 | { |
| 2772 | struct list_head *entry; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2773 | struct slab *slabp; |
| 2774 | struct kmem_list3 *l3; |
| 2775 | void *obj; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2776 | int x; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2777 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2778 | l3 = cachep->nodelists[nodeid]; |
| 2779 | BUG_ON(!l3); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2780 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2781 | retry: |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 2782 | check_irq_off(); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2783 | spin_lock(&l3->list_lock); |
| 2784 | entry = l3->slabs_partial.next; |
| 2785 | if (entry == &l3->slabs_partial) { |
| 2786 | l3->free_touched = 1; |
| 2787 | entry = l3->slabs_free.next; |
| 2788 | if (entry == &l3->slabs_free) |
| 2789 | goto must_grow; |
| 2790 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2791 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2792 | slabp = list_entry(entry, struct slab, list); |
| 2793 | check_spinlock_acquired_node(cachep, nodeid); |
| 2794 | check_slabp(cachep, slabp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2795 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2796 | STATS_INC_NODEALLOCS(cachep); |
| 2797 | STATS_INC_ACTIVE(cachep); |
| 2798 | STATS_SET_HIGH(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2799 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2800 | BUG_ON(slabp->inuse == cachep->num); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2801 | |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2802 | obj = slab_get_obj(cachep, slabp, nodeid); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2803 | check_slabp(cachep, slabp); |
| 2804 | l3->free_objects--; |
| 2805 | /* move slabp to correct slabp list: */ |
| 2806 | list_del(&slabp->list); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2807 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2808 | if (slabp->free == BUFCTL_END) { |
| 2809 | list_add(&slabp->list, &l3->slabs_full); |
| 2810 | } else { |
| 2811 | list_add(&slabp->list, &l3->slabs_partial); |
| 2812 | } |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2813 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2814 | spin_unlock(&l3->list_lock); |
| 2815 | goto done; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2816 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2817 | must_grow: |
| 2818 | spin_unlock(&l3->list_lock); |
| 2819 | x = cache_grow(cachep, flags, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2820 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2821 | if (!x) |
| 2822 | return NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2823 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2824 | goto retry; |
| 2825 | done: |
| 2826 | return obj; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2827 | } |
| 2828 | #endif |
| 2829 | |
| 2830 | /* |
| 2831 | * Caller needs to acquire correct kmem_list's list_lock |
| 2832 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2833 | 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] | 2834 | int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2835 | { |
| 2836 | int i; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2837 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2838 | |
| 2839 | for (i = 0; i < nr_objects; i++) { |
| 2840 | void *objp = objpp[i]; |
| 2841 | struct slab *slabp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2842 | |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2843 | slabp = virt_to_slab(objp); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2844 | l3 = cachep->nodelists[node]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2845 | list_del(&slabp->list); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2846 | check_spinlock_acquired_node(cachep, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2847 | check_slabp(cachep, slabp); |
Matthew Dobson | 78d382d | 2006-02-01 03:05:47 -0800 | [diff] [blame] | 2848 | slab_put_obj(cachep, slabp, objp, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2849 | STATS_DEC_ACTIVE(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2850 | l3->free_objects++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2851 | check_slabp(cachep, slabp); |
| 2852 | |
| 2853 | /* fixup slab chains */ |
| 2854 | if (slabp->inuse == 0) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2855 | if (l3->free_objects > l3->free_limit) { |
| 2856 | l3->free_objects -= cachep->num; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2857 | slab_destroy(cachep, slabp); |
| 2858 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2859 | list_add(&slabp->list, &l3->slabs_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2860 | } |
| 2861 | } else { |
| 2862 | /* Unconditionally move a slab to the end of the |
| 2863 | * partial list on free - maximum time for the |
| 2864 | * other objects to be freed, too. |
| 2865 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2866 | list_add_tail(&slabp->list, &l3->slabs_partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2867 | } |
| 2868 | } |
| 2869 | } |
| 2870 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2871 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2872 | { |
| 2873 | int batchcount; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2874 | struct kmem_list3 *l3; |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2875 | int node = numa_node_id(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2876 | |
| 2877 | batchcount = ac->batchcount; |
| 2878 | #if DEBUG |
| 2879 | BUG_ON(!batchcount || batchcount > ac->avail); |
| 2880 | #endif |
| 2881 | check_irq_off(); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2882 | l3 = cachep->nodelists[node]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2883 | spin_lock(&l3->list_lock); |
| 2884 | if (l3->shared) { |
| 2885 | struct array_cache *shared_array = l3->shared; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2886 | int max = shared_array->limit - shared_array->avail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2887 | if (max) { |
| 2888 | if (batchcount > max) |
| 2889 | batchcount = max; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2890 | memcpy(&(shared_array->entry[shared_array->avail]), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2891 | ac->entry, sizeof(void *) * batchcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2892 | shared_array->avail += batchcount; |
| 2893 | goto free_done; |
| 2894 | } |
| 2895 | } |
| 2896 | |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2897 | free_block(cachep, ac->entry, batchcount, node); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2898 | free_done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2899 | #if STATS |
| 2900 | { |
| 2901 | int i = 0; |
| 2902 | struct list_head *p; |
| 2903 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2904 | p = l3->slabs_free.next; |
| 2905 | while (p != &(l3->slabs_free)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2906 | struct slab *slabp; |
| 2907 | |
| 2908 | slabp = list_entry(p, struct slab, list); |
| 2909 | BUG_ON(slabp->inuse); |
| 2910 | |
| 2911 | i++; |
| 2912 | p = p->next; |
| 2913 | } |
| 2914 | STATS_SET_FREEABLE(cachep, i); |
| 2915 | } |
| 2916 | #endif |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2917 | spin_unlock(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2918 | ac->avail -= batchcount; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2919 | memmove(ac->entry, &(ac->entry[batchcount]), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2920 | sizeof(void *) * ac->avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2921 | } |
| 2922 | |
| 2923 | /* |
| 2924 | * __cache_free |
| 2925 | * Release an obj back to its cache. If the obj has a constructed |
| 2926 | * state, it must be in this state _before_ it is released. |
| 2927 | * |
| 2928 | * Called with disabled ints. |
| 2929 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2930 | static inline void __cache_free(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2931 | { |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2932 | struct array_cache *ac = cpu_cache_get(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2933 | |
| 2934 | check_irq_off(); |
| 2935 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); |
| 2936 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2937 | /* Make sure we are not freeing a object from another |
| 2938 | * node to the array cache on this cpu. |
| 2939 | */ |
| 2940 | #ifdef CONFIG_NUMA |
| 2941 | { |
| 2942 | struct slab *slabp; |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 2943 | slabp = virt_to_slab(objp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2944 | if (unlikely(slabp->nodeid != numa_node_id())) { |
| 2945 | struct array_cache *alien = NULL; |
| 2946 | int nodeid = slabp->nodeid; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2947 | struct kmem_list3 *l3 = |
| 2948 | cachep->nodelists[numa_node_id()]; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2949 | |
| 2950 | STATS_INC_NODEFREES(cachep); |
| 2951 | if (l3->alien && l3->alien[nodeid]) { |
| 2952 | alien = l3->alien[nodeid]; |
| 2953 | spin_lock(&alien->lock); |
| 2954 | if (unlikely(alien->avail == alien->limit)) |
| 2955 | __drain_alien_cache(cachep, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2956 | alien, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2957 | alien->entry[alien->avail++] = objp; |
| 2958 | spin_unlock(&alien->lock); |
| 2959 | } else { |
| 2960 | spin_lock(&(cachep->nodelists[nodeid])-> |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2961 | list_lock); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 2962 | free_block(cachep, &objp, 1, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2963 | spin_unlock(&(cachep->nodelists[nodeid])-> |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 2964 | list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2965 | } |
| 2966 | return; |
| 2967 | } |
| 2968 | } |
| 2969 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2970 | if (likely(ac->avail < ac->limit)) { |
| 2971 | STATS_INC_FREEHIT(cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2972 | ac->entry[ac->avail++] = objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2973 | return; |
| 2974 | } else { |
| 2975 | STATS_INC_FREEMISS(cachep); |
| 2976 | cache_flusharray(cachep, ac); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 2977 | ac->entry[ac->avail++] = objp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2978 | } |
| 2979 | } |
| 2980 | |
| 2981 | /** |
| 2982 | * kmem_cache_alloc - Allocate an object |
| 2983 | * @cachep: The cache to allocate from. |
| 2984 | * @flags: See kmalloc(). |
| 2985 | * |
| 2986 | * Allocate an object from this cache. The flags are only relevant |
| 2987 | * if the cache has no available objects. |
| 2988 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 2989 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2990 | { |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 2991 | return __cache_alloc(cachep, flags, __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2992 | } |
| 2993 | EXPORT_SYMBOL(kmem_cache_alloc); |
| 2994 | |
| 2995 | /** |
| 2996 | * kmem_ptr_validate - check if an untrusted pointer might |
| 2997 | * be a slab entry. |
| 2998 | * @cachep: the cache we're checking against |
| 2999 | * @ptr: pointer to validate |
| 3000 | * |
| 3001 | * This verifies that the untrusted pointer looks sane: |
| 3002 | * it is _not_ a guarantee that the pointer is actually |
| 3003 | * part of the slab cache in question, but it at least |
| 3004 | * validates that the pointer can be dereferenced and |
| 3005 | * looks half-way sane. |
| 3006 | * |
| 3007 | * Currently only used for dentry validation. |
| 3008 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3009 | int fastcall kmem_ptr_validate(struct kmem_cache *cachep, void *ptr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3010 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3011 | unsigned long addr = (unsigned long)ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3012 | unsigned long min_addr = PAGE_OFFSET; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3013 | unsigned long align_mask = BYTES_PER_WORD - 1; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3014 | unsigned long size = cachep->buffer_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3015 | struct page *page; |
| 3016 | |
| 3017 | if (unlikely(addr < min_addr)) |
| 3018 | goto out; |
| 3019 | if (unlikely(addr > (unsigned long)high_memory - size)) |
| 3020 | goto out; |
| 3021 | if (unlikely(addr & align_mask)) |
| 3022 | goto out; |
| 3023 | if (unlikely(!kern_addr_valid(addr))) |
| 3024 | goto out; |
| 3025 | if (unlikely(!kern_addr_valid(addr + size - 1))) |
| 3026 | goto out; |
| 3027 | page = virt_to_page(ptr); |
| 3028 | if (unlikely(!PageSlab(page))) |
| 3029 | goto out; |
Pekka Enberg | 065d41c | 2005-11-13 16:06:46 -0800 | [diff] [blame] | 3030 | if (unlikely(page_get_cache(page) != cachep)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | goto out; |
| 3032 | return 1; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3033 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3034 | return 0; |
| 3035 | } |
| 3036 | |
| 3037 | #ifdef CONFIG_NUMA |
| 3038 | /** |
| 3039 | * kmem_cache_alloc_node - Allocate an object on the specified node |
| 3040 | * @cachep: The cache to allocate from. |
| 3041 | * @flags: See kmalloc(). |
| 3042 | * @nodeid: node number of the target node. |
| 3043 | * |
| 3044 | * Identical to kmem_cache_alloc, except that this function is slow |
| 3045 | * and can sleep. And it will allocate memory on the given node, which |
| 3046 | * can improve the performance for cpu bound structures. |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3047 | * New and improved: it will now make sure that the object gets |
| 3048 | * 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] | 3049 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3050 | 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] | 3051 | { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3052 | unsigned long save_flags; |
| 3053 | void *ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3054 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3055 | cache_alloc_debugcheck_before(cachep, flags); |
| 3056 | local_irq_save(save_flags); |
Christoph Lameter | 18f820f | 2006-02-01 03:05:43 -0800 | [diff] [blame] | 3057 | |
| 3058 | if (nodeid == -1 || nodeid == numa_node_id() || |
| 3059 | !cachep->nodelists[nodeid]) |
Alok N Kataria | 5c38230 | 2005-09-27 21:45:46 -0700 | [diff] [blame] | 3060 | ptr = ____cache_alloc(cachep, flags); |
| 3061 | else |
| 3062 | ptr = __cache_alloc_node(cachep, flags, nodeid); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3063 | local_irq_restore(save_flags); |
Christoph Lameter | 18f820f | 2006-02-01 03:05:43 -0800 | [diff] [blame] | 3064 | |
| 3065 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, |
| 3066 | __builtin_return_address(0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3067 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3068 | return ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3069 | } |
| 3070 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
| 3071 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 3072 | void *kmalloc_node(size_t size, gfp_t flags, int node) |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3073 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3074 | struct kmem_cache *cachep; |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3075 | |
| 3076 | cachep = kmem_find_general_cachep(size, flags); |
| 3077 | if (unlikely(cachep == NULL)) |
| 3078 | return NULL; |
| 3079 | return kmem_cache_alloc_node(cachep, flags, node); |
| 3080 | } |
| 3081 | EXPORT_SYMBOL(kmalloc_node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | #endif |
| 3083 | |
| 3084 | /** |
| 3085 | * kmalloc - allocate memory |
| 3086 | * @size: how many bytes of memory are required. |
| 3087 | * @flags: the type of memory to allocate. |
| 3088 | * |
| 3089 | * kmalloc is the normal method of allocating memory |
| 3090 | * in the kernel. |
| 3091 | * |
| 3092 | * The @flags argument may be one of: |
| 3093 | * |
| 3094 | * %GFP_USER - Allocate memory on behalf of user. May sleep. |
| 3095 | * |
| 3096 | * %GFP_KERNEL - Allocate normal kernel ram. May sleep. |
| 3097 | * |
| 3098 | * %GFP_ATOMIC - Allocation will not sleep. Use inside interrupt handlers. |
| 3099 | * |
| 3100 | * Additionally, the %GFP_DMA flag may be set to indicate the memory |
| 3101 | * must be suitable for DMA. This can mean different things on different |
| 3102 | * platforms. For example, on i386, it means that the memory must come |
| 3103 | * from the first 16MB. |
| 3104 | */ |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3105 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
| 3106 | void *caller) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3107 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3108 | struct kmem_cache *cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3109 | |
Manfred Spraul | 97e2bde | 2005-05-01 08:58:38 -0700 | [diff] [blame] | 3110 | /* If you want to save a few bytes .text space: replace |
| 3111 | * __ with kmem_. |
| 3112 | * Then kmalloc uses the uninlined functions instead of the inline |
| 3113 | * functions. |
| 3114 | */ |
| 3115 | cachep = __find_general_cachep(size, flags); |
Andrew Morton | dbdb904 | 2005-09-23 13:24:10 -0700 | [diff] [blame] | 3116 | if (unlikely(cachep == NULL)) |
| 3117 | return NULL; |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3118 | return __cache_alloc(cachep, flags, caller); |
| 3119 | } |
| 3120 | |
| 3121 | #ifndef CONFIG_DEBUG_SLAB |
| 3122 | |
| 3123 | void *__kmalloc(size_t size, gfp_t flags) |
| 3124 | { |
| 3125 | return __do_kmalloc(size, flags, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3126 | } |
| 3127 | EXPORT_SYMBOL(__kmalloc); |
| 3128 | |
Pekka Enberg | 7fd6b14 | 2006-02-01 03:05:52 -0800 | [diff] [blame] | 3129 | #else |
| 3130 | |
| 3131 | void *__kmalloc_track_caller(size_t size, gfp_t flags, void *caller) |
| 3132 | { |
| 3133 | return __do_kmalloc(size, flags, caller); |
| 3134 | } |
| 3135 | EXPORT_SYMBOL(__kmalloc_track_caller); |
| 3136 | |
| 3137 | #endif |
| 3138 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3139 | #ifdef CONFIG_SMP |
| 3140 | /** |
| 3141 | * __alloc_percpu - allocate one copy of the object for every present |
| 3142 | * cpu in the system, zeroing them. |
| 3143 | * Objects should be dereferenced using the per_cpu_ptr macro only. |
| 3144 | * |
| 3145 | * @size: how many bytes of memory are required. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3146 | */ |
Pekka Enberg | f9f7500 | 2006-01-08 01:00:33 -0800 | [diff] [blame] | 3147 | void *__alloc_percpu(size_t size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3148 | { |
| 3149 | int i; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3150 | struct percpu_data *pdata = kmalloc(sizeof(*pdata), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3151 | |
| 3152 | if (!pdata) |
| 3153 | return NULL; |
| 3154 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3155 | /* |
| 3156 | * Cannot use for_each_online_cpu since a cpu may come online |
| 3157 | * and we have no way of figuring out how to fix the array |
| 3158 | * that we have allocated then.... |
| 3159 | */ |
| 3160 | for_each_cpu(i) { |
| 3161 | int node = cpu_to_node(i); |
| 3162 | |
| 3163 | if (node_online(node)) |
| 3164 | pdata->ptrs[i] = kmalloc_node(size, GFP_KERNEL, node); |
| 3165 | else |
| 3166 | pdata->ptrs[i] = kmalloc(size, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3167 | |
| 3168 | if (!pdata->ptrs[i]) |
| 3169 | goto unwind_oom; |
| 3170 | memset(pdata->ptrs[i], 0, size); |
| 3171 | } |
| 3172 | |
| 3173 | /* Catch derefs w/o wrappers */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3174 | return (void *)(~(unsigned long)pdata); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3175 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3176 | unwind_oom: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3177 | while (--i >= 0) { |
| 3178 | if (!cpu_possible(i)) |
| 3179 | continue; |
| 3180 | kfree(pdata->ptrs[i]); |
| 3181 | } |
| 3182 | kfree(pdata); |
| 3183 | return NULL; |
| 3184 | } |
| 3185 | EXPORT_SYMBOL(__alloc_percpu); |
| 3186 | #endif |
| 3187 | |
| 3188 | /** |
| 3189 | * kmem_cache_free - Deallocate an object |
| 3190 | * @cachep: The cache the allocation was from. |
| 3191 | * @objp: The previously allocated object. |
| 3192 | * |
| 3193 | * Free an object which was previously allocated from this |
| 3194 | * cache. |
| 3195 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3196 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3197 | { |
| 3198 | unsigned long flags; |
| 3199 | |
| 3200 | local_irq_save(flags); |
| 3201 | __cache_free(cachep, objp); |
| 3202 | local_irq_restore(flags); |
| 3203 | } |
| 3204 | EXPORT_SYMBOL(kmem_cache_free); |
| 3205 | |
| 3206 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3207 | * kfree - free previously allocated memory |
| 3208 | * @objp: pointer returned by kmalloc. |
| 3209 | * |
Pekka Enberg | 80e93ef | 2005-09-09 13:10:16 -0700 | [diff] [blame] | 3210 | * If @objp is NULL, no operation is performed. |
| 3211 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3212 | * Don't free memory not originally allocated by kmalloc() |
| 3213 | * or you will run into trouble. |
| 3214 | */ |
| 3215 | void kfree(const void *objp) |
| 3216 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3217 | struct kmem_cache *c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3218 | unsigned long flags; |
| 3219 | |
| 3220 | if (unlikely(!objp)) |
| 3221 | return; |
| 3222 | local_irq_save(flags); |
| 3223 | kfree_debugcheck(objp); |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3224 | c = virt_to_cache(objp); |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3225 | mutex_debug_check_no_locks_freed(objp, obj_size(c)); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3226 | __cache_free(c, (void *)objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3227 | local_irq_restore(flags); |
| 3228 | } |
| 3229 | EXPORT_SYMBOL(kfree); |
| 3230 | |
| 3231 | #ifdef CONFIG_SMP |
| 3232 | /** |
| 3233 | * free_percpu - free previously allocated percpu memory |
| 3234 | * @objp: pointer returned by alloc_percpu. |
| 3235 | * |
| 3236 | * Don't free memory not originally allocated by alloc_percpu() |
| 3237 | * The complemented objp is to check for that. |
| 3238 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3239 | void free_percpu(const void *objp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3240 | { |
| 3241 | int i; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3242 | struct percpu_data *p = (struct percpu_data *)(~(unsigned long)objp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3243 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3244 | /* |
| 3245 | * We allocate for all cpus so we cannot use for online cpu here. |
| 3246 | */ |
| 3247 | for_each_cpu(i) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3248 | kfree(p->ptrs[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3249 | kfree(p); |
| 3250 | } |
| 3251 | EXPORT_SYMBOL(free_percpu); |
| 3252 | #endif |
| 3253 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3254 | unsigned int kmem_cache_size(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3255 | { |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3256 | return obj_size(cachep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3257 | } |
| 3258 | EXPORT_SYMBOL(kmem_cache_size); |
| 3259 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3260 | const char *kmem_cache_name(struct kmem_cache *cachep) |
Arnaldo Carvalho de Melo | 1944972 | 2005-06-18 22:46:19 -0700 | [diff] [blame] | 3261 | { |
| 3262 | return cachep->name; |
| 3263 | } |
| 3264 | EXPORT_SYMBOL_GPL(kmem_cache_name); |
| 3265 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3266 | /* |
| 3267 | * This initializes kmem_list3 for all nodes. |
| 3268 | */ |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3269 | static int alloc_kmemlist(struct kmem_cache *cachep) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3270 | { |
| 3271 | int node; |
| 3272 | struct kmem_list3 *l3; |
| 3273 | int err = 0; |
| 3274 | |
| 3275 | for_each_online_node(node) { |
| 3276 | struct array_cache *nc = NULL, *new; |
| 3277 | struct array_cache **new_alien = NULL; |
| 3278 | #ifdef CONFIG_NUMA |
| 3279 | if (!(new_alien = alloc_alien_cache(node, cachep->limit))) |
| 3280 | goto fail; |
| 3281 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3282 | if (!(new = alloc_arraycache(node, (cachep->shared * |
| 3283 | cachep->batchcount), |
| 3284 | 0xbaadf00d))) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3285 | goto fail; |
| 3286 | if ((l3 = cachep->nodelists[node])) { |
| 3287 | |
| 3288 | spin_lock_irq(&l3->list_lock); |
| 3289 | |
| 3290 | if ((nc = cachep->nodelists[node]->shared)) |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3291 | free_block(cachep, nc->entry, nc->avail, node); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3292 | |
| 3293 | l3->shared = new; |
| 3294 | if (!cachep->nodelists[node]->alien) { |
| 3295 | l3->alien = new_alien; |
| 3296 | new_alien = NULL; |
| 3297 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3298 | l3->free_limit = (1 + nr_cpus_node(node)) * |
| 3299 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3300 | spin_unlock_irq(&l3->list_lock); |
| 3301 | kfree(nc); |
| 3302 | free_alien_cache(new_alien); |
| 3303 | continue; |
| 3304 | } |
| 3305 | if (!(l3 = kmalloc_node(sizeof(struct kmem_list3), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3306 | GFP_KERNEL, node))) |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3307 | goto fail; |
| 3308 | |
| 3309 | kmem_list3_init(l3); |
| 3310 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3311 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3312 | l3->shared = new; |
| 3313 | l3->alien = new_alien; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3314 | l3->free_limit = (1 + nr_cpus_node(node)) * |
| 3315 | cachep->batchcount + cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3316 | cachep->nodelists[node] = l3; |
| 3317 | } |
| 3318 | return err; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3319 | fail: |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3320 | err = -ENOMEM; |
| 3321 | return err; |
| 3322 | } |
| 3323 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3324 | struct ccupdate_struct { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3325 | struct kmem_cache *cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3326 | struct array_cache *new[NR_CPUS]; |
| 3327 | }; |
| 3328 | |
| 3329 | static void do_ccupdate_local(void *info) |
| 3330 | { |
| 3331 | struct ccupdate_struct *new = (struct ccupdate_struct *)info; |
| 3332 | struct array_cache *old; |
| 3333 | |
| 3334 | check_irq_off(); |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3335 | old = cpu_cache_get(new->cachep); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3336 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3337 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
| 3338 | new->new[smp_processor_id()] = old; |
| 3339 | } |
| 3340 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3341 | 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] | 3342 | int shared) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3343 | { |
| 3344 | struct ccupdate_struct new; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3345 | int i, err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3346 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3347 | memset(&new.new, 0, sizeof(new.new)); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3348 | for_each_online_cpu(i) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3349 | new.new[i] = |
| 3350 | alloc_arraycache(cpu_to_node(i), limit, batchcount); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3351 | if (!new.new[i]) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3352 | for (i--; i >= 0; i--) |
| 3353 | kfree(new.new[i]); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3354 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3355 | } |
| 3356 | } |
| 3357 | new.cachep = cachep; |
| 3358 | |
| 3359 | smp_call_function_all_cpus(do_ccupdate_local, (void *)&new); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3360 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3361 | check_irq_on(); |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3362 | spin_lock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3363 | cachep->batchcount = batchcount; |
| 3364 | cachep->limit = limit; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3365 | cachep->shared = shared; |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3366 | spin_unlock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3367 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3368 | for_each_online_cpu(i) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3369 | struct array_cache *ccold = new.new[i]; |
| 3370 | if (!ccold) |
| 3371 | continue; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3372 | spin_lock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3373 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_node(i)); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3374 | spin_unlock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3375 | kfree(ccold); |
| 3376 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3377 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3378 | err = alloc_kmemlist(cachep); |
| 3379 | if (err) { |
| 3380 | printk(KERN_ERR "alloc_kmemlist failed for %s, error %d.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3381 | cachep->name, -err); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3382 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3383 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3384 | return 0; |
| 3385 | } |
| 3386 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3387 | static void enable_cpucache(struct kmem_cache *cachep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3388 | { |
| 3389 | int err; |
| 3390 | int limit, shared; |
| 3391 | |
| 3392 | /* The head array serves three purposes: |
| 3393 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
| 3394 | * - reduce the number of spinlock operations. |
| 3395 | * - reduce the number of linked list operations on the slab and |
| 3396 | * bufctl chains: array operations are cheaper. |
| 3397 | * The numbers are guessed, we should auto-tune as described by |
| 3398 | * Bonwick. |
| 3399 | */ |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3400 | if (cachep->buffer_size > 131072) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3401 | limit = 1; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3402 | else if (cachep->buffer_size > PAGE_SIZE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3403 | limit = 8; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3404 | else if (cachep->buffer_size > 1024) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3405 | limit = 24; |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3406 | else if (cachep->buffer_size > 256) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3407 | limit = 54; |
| 3408 | else |
| 3409 | limit = 120; |
| 3410 | |
| 3411 | /* Cpu bound tasks (e.g. network routing) can exhibit cpu bound |
| 3412 | * allocation behaviour: Most allocs on one cpu, most free operations |
| 3413 | * on another cpu. For these cases, an efficient object passing between |
| 3414 | * cpus is necessary. This is provided by a shared array. The array |
| 3415 | * replaces Bonwick's magazine layer. |
| 3416 | * On uniprocessor, it's functionally equivalent (but less efficient) |
| 3417 | * to a larger limit. Thus disabled by default. |
| 3418 | */ |
| 3419 | shared = 0; |
| 3420 | #ifdef CONFIG_SMP |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3421 | if (cachep->buffer_size <= PAGE_SIZE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3422 | shared = 8; |
| 3423 | #endif |
| 3424 | |
| 3425 | #if DEBUG |
| 3426 | /* With debugging enabled, large batchcount lead to excessively |
| 3427 | * long periods with disabled local interrupts. Limit the |
| 3428 | * batchcount |
| 3429 | */ |
| 3430 | if (limit > 32) |
| 3431 | limit = 32; |
| 3432 | #endif |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3433 | err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3434 | if (err) |
| 3435 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3436 | cachep->name, -err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3437 | } |
| 3438 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3439 | 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] | 3440 | int force, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3441 | { |
| 3442 | int tofree; |
| 3443 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3444 | check_spinlock_acquired_node(cachep, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3445 | if (ac->touched && !force) { |
| 3446 | ac->touched = 0; |
| 3447 | } else if (ac->avail) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3448 | tofree = force ? ac->avail : (ac->limit + 4) / 5; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3449 | if (tofree > ac->avail) { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3450 | tofree = (ac->avail + 1) / 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3451 | } |
Christoph Lameter | ff69416 | 2005-09-22 21:44:02 -0700 | [diff] [blame] | 3452 | free_block(cachep, ac->entry, tofree, node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3453 | ac->avail -= tofree; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3454 | memmove(ac->entry, &(ac->entry[tofree]), |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3455 | sizeof(void *) * ac->avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3456 | } |
| 3457 | } |
| 3458 | |
| 3459 | /** |
| 3460 | * cache_reap - Reclaim memory from caches. |
Randy Dunlap | 1e5d533 | 2005-11-07 01:01:06 -0800 | [diff] [blame] | 3461 | * @unused: unused parameter |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3462 | * |
| 3463 | * Called from workqueue/eventd every few seconds. |
| 3464 | * Purpose: |
| 3465 | * - clear the per-cpu caches for this CPU. |
| 3466 | * - return freeable pages to the main free memory pool. |
| 3467 | * |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3468 | * 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] | 3469 | * try again on the next iteration. |
| 3470 | */ |
| 3471 | static void cache_reap(void *unused) |
| 3472 | { |
| 3473 | struct list_head *walk; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3474 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3475 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3476 | if (!mutex_trylock(&cache_chain_mutex)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3477 | /* Give up. Setup the next iteration. */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3478 | schedule_delayed_work(&__get_cpu_var(reap_work), |
| 3479 | REAPTIMEOUT_CPUC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3480 | return; |
| 3481 | } |
| 3482 | |
| 3483 | list_for_each(walk, &cache_chain) { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3484 | struct kmem_cache *searchp; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3485 | struct list_head *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3486 | int tofree; |
| 3487 | struct slab *slabp; |
| 3488 | |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3489 | searchp = list_entry(walk, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3490 | |
| 3491 | if (searchp->flags & SLAB_NO_REAP) |
| 3492 | goto next; |
| 3493 | |
| 3494 | check_irq_on(); |
| 3495 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3496 | l3 = searchp->nodelists[numa_node_id()]; |
| 3497 | if (l3->alien) |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 3498 | drain_alien_cache(searchp, l3->alien); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3499 | spin_lock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3500 | |
Pekka Enberg | 9a2dba4 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3501 | drain_array_locked(searchp, cpu_cache_get(searchp), 0, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3502 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3503 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3504 | if (time_after(l3->next_reap, jiffies)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3505 | goto next_unlock; |
| 3506 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3507 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3508 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3509 | if (l3->shared) |
| 3510 | drain_array_locked(searchp, l3->shared, 0, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3511 | numa_node_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3512 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3513 | if (l3->free_touched) { |
| 3514 | l3->free_touched = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3515 | goto next_unlock; |
| 3516 | } |
| 3517 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3518 | tofree = |
| 3519 | (l3->free_limit + 5 * searchp->num - |
| 3520 | 1) / (5 * searchp->num); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3521 | do { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3522 | p = l3->slabs_free.next; |
| 3523 | if (p == &(l3->slabs_free)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3524 | break; |
| 3525 | |
| 3526 | slabp = list_entry(p, struct slab, list); |
| 3527 | BUG_ON(slabp->inuse); |
| 3528 | list_del(&slabp->list); |
| 3529 | STATS_INC_REAPED(searchp); |
| 3530 | |
| 3531 | /* Safe to drop the lock. The slab is no longer |
| 3532 | * linked to the cache. |
| 3533 | * searchp cannot disappear, we hold |
| 3534 | * cache_chain_lock |
| 3535 | */ |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3536 | l3->free_objects -= searchp->num; |
| 3537 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3538 | slab_destroy(searchp, slabp); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3539 | spin_lock_irq(&l3->list_lock); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3540 | } while (--tofree > 0); |
| 3541 | next_unlock: |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3542 | spin_unlock_irq(&l3->list_lock); |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3543 | next: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3544 | cond_resched(); |
| 3545 | } |
| 3546 | check_irq_on(); |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3547 | mutex_unlock(&cache_chain_mutex); |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 3548 | drain_remote_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3549 | /* Setup the next iteration */ |
Manfred Spraul | cd61ef6 | 2005-11-07 00:58:02 -0800 | [diff] [blame] | 3550 | schedule_delayed_work(&__get_cpu_var(reap_work), REAPTIMEOUT_CPUC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3551 | } |
| 3552 | |
| 3553 | #ifdef CONFIG_PROC_FS |
| 3554 | |
Pekka Enberg | 85289f9 | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 3555 | static void print_slabinfo_header(struct seq_file *m) |
| 3556 | { |
| 3557 | /* |
| 3558 | * Output format version, so at least we can change it |
| 3559 | * without _too_ many complaints. |
| 3560 | */ |
| 3561 | #if STATS |
| 3562 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
| 3563 | #else |
| 3564 | seq_puts(m, "slabinfo - version: 2.1\n"); |
| 3565 | #endif |
| 3566 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
| 3567 | "<objperslab> <pagesperslab>"); |
| 3568 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
| 3569 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); |
| 3570 | #if STATS |
| 3571 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
| 3572 | "<error> <maxfreeable> <nodeallocs> <remotefrees>"); |
| 3573 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
| 3574 | #endif |
| 3575 | seq_putc(m, '\n'); |
| 3576 | } |
| 3577 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3578 | static void *s_start(struct seq_file *m, loff_t *pos) |
| 3579 | { |
| 3580 | loff_t n = *pos; |
| 3581 | struct list_head *p; |
| 3582 | |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3583 | mutex_lock(&cache_chain_mutex); |
Pekka Enberg | 85289f9 | 2006-01-08 01:00:36 -0800 | [diff] [blame] | 3584 | if (!n) |
| 3585 | print_slabinfo_header(m); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3586 | p = cache_chain.next; |
| 3587 | while (n--) { |
| 3588 | p = p->next; |
| 3589 | if (p == &cache_chain) |
| 3590 | return NULL; |
| 3591 | } |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3592 | return list_entry(p, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3593 | } |
| 3594 | |
| 3595 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) |
| 3596 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3597 | struct kmem_cache *cachep = p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3598 | ++*pos; |
| 3599 | return cachep->next.next == &cache_chain ? NULL |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3600 | : list_entry(cachep->next.next, struct kmem_cache, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3601 | } |
| 3602 | |
| 3603 | static void s_stop(struct seq_file *m, void *p) |
| 3604 | { |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3605 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3606 | } |
| 3607 | |
| 3608 | static int s_show(struct seq_file *m, void *p) |
| 3609 | { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3610 | struct kmem_cache *cachep = p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3611 | struct list_head *q; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3612 | struct slab *slabp; |
| 3613 | unsigned long active_objs; |
| 3614 | unsigned long num_objs; |
| 3615 | unsigned long active_slabs = 0; |
| 3616 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3617 | const char *name; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3618 | char *error = NULL; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3619 | int node; |
| 3620 | struct kmem_list3 *l3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3621 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3622 | spin_lock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3623 | active_objs = 0; |
| 3624 | num_slabs = 0; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3625 | for_each_online_node(node) { |
| 3626 | l3 = cachep->nodelists[node]; |
| 3627 | if (!l3) |
| 3628 | continue; |
| 3629 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3630 | check_irq_on(); |
| 3631 | spin_lock_irq(&l3->list_lock); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3632 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3633 | list_for_each(q, &l3->slabs_full) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3634 | slabp = list_entry(q, struct slab, list); |
| 3635 | if (slabp->inuse != cachep->num && !error) |
| 3636 | error = "slabs_full accounting error"; |
| 3637 | active_objs += cachep->num; |
| 3638 | active_slabs++; |
| 3639 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3640 | list_for_each(q, &l3->slabs_partial) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3641 | slabp = list_entry(q, struct slab, list); |
| 3642 | if (slabp->inuse == cachep->num && !error) |
| 3643 | error = "slabs_partial inuse accounting error"; |
| 3644 | if (!slabp->inuse && !error) |
| 3645 | error = "slabs_partial/inuse accounting error"; |
| 3646 | active_objs += slabp->inuse; |
| 3647 | active_slabs++; |
| 3648 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3649 | list_for_each(q, &l3->slabs_free) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3650 | slabp = list_entry(q, struct slab, list); |
| 3651 | if (slabp->inuse && !error) |
| 3652 | error = "slabs_free/inuse accounting error"; |
| 3653 | num_slabs++; |
| 3654 | } |
| 3655 | free_objects += l3->free_objects; |
Ravikiran G Thirumalai | 4484ebf | 2006-02-04 23:27:59 -0800 | [diff] [blame] | 3656 | if (l3->shared) |
| 3657 | shared_avail += l3->shared->avail; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3658 | |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3659 | spin_unlock_irq(&l3->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3660 | } |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3661 | num_slabs += active_slabs; |
| 3662 | num_objs = num_slabs * cachep->num; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3663 | if (num_objs - active_objs != free_objects && !error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3664 | error = "free_objects accounting error"; |
| 3665 | |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3666 | name = cachep->name; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3667 | if (error) |
| 3668 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); |
| 3669 | |
| 3670 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
Manfred Spraul | 3dafccf | 2006-02-01 03:05:42 -0800 | [diff] [blame] | 3671 | name, active_objs, num_objs, cachep->buffer_size, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3672 | cachep->num, (1 << cachep->gfporder)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3673 | seq_printf(m, " : tunables %4u %4u %4u", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3674 | cachep->limit, cachep->batchcount, cachep->shared); |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3675 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3676 | active_slabs, num_slabs, shared_avail); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3677 | #if STATS |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3678 | { /* list3 stats */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3679 | unsigned long high = cachep->high_mark; |
| 3680 | unsigned long allocs = cachep->num_allocations; |
| 3681 | unsigned long grown = cachep->grown; |
| 3682 | unsigned long reaped = cachep->reaped; |
| 3683 | unsigned long errors = cachep->errors; |
| 3684 | unsigned long max_freeable = cachep->max_freeable; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3685 | unsigned long node_allocs = cachep->node_allocs; |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3686 | unsigned long node_frees = cachep->node_frees; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3687 | |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3688 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu \ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3689 | %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] | 3690 | } |
| 3691 | /* cpu stats */ |
| 3692 | { |
| 3693 | unsigned long allochit = atomic_read(&cachep->allochit); |
| 3694 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); |
| 3695 | unsigned long freehit = atomic_read(&cachep->freehit); |
| 3696 | unsigned long freemiss = atomic_read(&cachep->freemiss); |
| 3697 | |
| 3698 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3699 | allochit, allocmiss, freehit, freemiss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3700 | } |
| 3701 | #endif |
| 3702 | seq_putc(m, '\n'); |
Ravikiran G Thirumalai | ca3b9b9 | 2006-02-04 23:27:58 -0800 | [diff] [blame] | 3703 | spin_unlock(&cachep->spinlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3704 | return 0; |
| 3705 | } |
| 3706 | |
| 3707 | /* |
| 3708 | * slabinfo_op - iterator that generates /proc/slabinfo |
| 3709 | * |
| 3710 | * Output layout: |
| 3711 | * cache-name |
| 3712 | * num-active-objs |
| 3713 | * total-objs |
| 3714 | * object size |
| 3715 | * num-active-slabs |
| 3716 | * total-slabs |
| 3717 | * num-pages-per-slab |
| 3718 | * + further values on SMP and with statistics enabled |
| 3719 | */ |
| 3720 | |
| 3721 | struct seq_operations slabinfo_op = { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3722 | .start = s_start, |
| 3723 | .next = s_next, |
| 3724 | .stop = s_stop, |
| 3725 | .show = s_show, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3726 | }; |
| 3727 | |
| 3728 | #define MAX_SLABINFO_WRITE 128 |
| 3729 | /** |
| 3730 | * slabinfo_write - Tuning for the slab allocator |
| 3731 | * @file: unused |
| 3732 | * @buffer: user buffer |
| 3733 | * @count: data length |
| 3734 | * @ppos: unused |
| 3735 | */ |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3736 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, |
| 3737 | size_t count, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3738 | { |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3739 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3740 | int limit, batchcount, shared, res; |
| 3741 | struct list_head *p; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3742 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3743 | if (count > MAX_SLABINFO_WRITE) |
| 3744 | return -EINVAL; |
| 3745 | if (copy_from_user(&kbuf, buffer, count)) |
| 3746 | return -EFAULT; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3747 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3748 | |
| 3749 | tmp = strchr(kbuf, ' '); |
| 3750 | if (!tmp) |
| 3751 | return -EINVAL; |
| 3752 | *tmp = '\0'; |
| 3753 | tmp++; |
| 3754 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) |
| 3755 | return -EINVAL; |
| 3756 | |
| 3757 | /* Find the cache in the chain of caches. */ |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3758 | mutex_lock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3759 | res = -EINVAL; |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3760 | list_for_each(p, &cache_chain) { |
Pekka Enberg | 343e0d7 | 2006-02-01 03:05:50 -0800 | [diff] [blame] | 3761 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, |
| 3762 | next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3763 | |
| 3764 | if (!strcmp(cachep->name, kbuf)) { |
| 3765 | if (limit < 1 || |
| 3766 | batchcount < 1 || |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3767 | batchcount > limit || shared < 0) { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3768 | res = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3769 | } else { |
Christoph Lameter | e498be7 | 2005-09-09 13:03:32 -0700 | [diff] [blame] | 3770 | res = do_tune_cpucache(cachep, limit, |
Pekka Enberg | b28a02d | 2006-01-08 01:00:37 -0800 | [diff] [blame] | 3771 | batchcount, shared); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3772 | } |
| 3773 | break; |
| 3774 | } |
| 3775 | } |
Ingo Molnar | fc0abb1 | 2006-01-18 17:42:33 -0800 | [diff] [blame] | 3776 | mutex_unlock(&cache_chain_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3777 | if (res >= 0) |
| 3778 | res = count; |
| 3779 | return res; |
| 3780 | } |
| 3781 | #endif |
| 3782 | |
Manfred Spraul | 00e145b | 2005-09-03 15:55:07 -0700 | [diff] [blame] | 3783 | /** |
| 3784 | * ksize - get the actual amount of memory allocated for a given object |
| 3785 | * @objp: Pointer to the object |
| 3786 | * |
| 3787 | * kmalloc may internally round up allocations and return more memory |
| 3788 | * than requested. ksize() can be used to determine the actual amount of |
| 3789 | * memory allocated. The caller may use this additional memory, even though |
| 3790 | * a smaller amount of memory was initially specified with the kmalloc call. |
| 3791 | * The caller must guarantee that objp points to a valid object previously |
| 3792 | * allocated with either kmalloc() or kmem_cache_alloc(). The object |
| 3793 | * must not be freed during the duration of the call. |
| 3794 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3795 | unsigned int ksize(const void *objp) |
| 3796 | { |
Manfred Spraul | 00e145b | 2005-09-03 15:55:07 -0700 | [diff] [blame] | 3797 | if (unlikely(objp == NULL)) |
| 3798 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3799 | |
Pekka Enberg | 6ed5eb221 | 2006-02-01 03:05:49 -0800 | [diff] [blame] | 3800 | return obj_size(virt_to_cache(objp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3801 | } |