1 /* 2 * Copyright (c) 2002, Jeffrey Roberson <jroberson@chesapeake.net> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $FreeBSD$ 27 * 28 */ 29 30 /* 31 * 32 * Jeff Roberson <jroberson@chesapeake.net> 33 * 34 * This file includes definitions, structures, prototypes, and inlines that 35 * should not be used outside of the actual implementation of UMA. 36 * 37 */ 38 39 /* 40 * Here's a quick description of the relationship between the objects: 41 * 42 * Zones contain lists of slabs which are stored in either the full bin, empty 43 * bin, or partially allocated bin, to reduce fragmentation. They also contain 44 * the user supplied value for size, which is adjusted for alignment purposes 45 * and rsize is the result of that. The zone also stores information for 46 * managing a hash of page addresses that maps pages to uma_slab_t structures 47 * for pages that don't have embedded uma_slab_t's. 48 * 49 * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may 50 * be allocated off the page from a special slab zone. The free list within a 51 * slab is managed with a linked list of indexes, which are 8 bit values. If 52 * UMA_SLAB_SIZE is defined to be too large I will have to switch to 16bit 53 * values. Currently on alpha you can get 250 or so 32 byte items and on x86 54 * you can get 250 or so 16byte items. For item sizes that would yield more 55 * than 10% memory waste we potentially allocate a separate uma_slab_t if this 56 * will improve the number of items per slab that will fit. 57 * 58 * Other potential space optimizations are storing the 8bit of linkage in space 59 * wasted between items due to alignment problems. This may yield a much better 60 * memory footprint for certain sizes of objects. Another alternative is to 61 * increase the UMA_SLAB_SIZE, or allow for dynamic slab sizes. I prefer 62 * dynamic slab sizes because we could stick with 8 bit indexes and only use 63 * large slab sizes for zones with a lot of waste per slab. This may create 64 * ineffeciencies in the vm subsystem due to fragmentation in the address space. 65 * 66 * The only really gross cases, with regards to memory waste, are for those 67 * items that are just over half the page size. You can get nearly 50% waste, 68 * so you fall back to the memory footprint of the power of two allocator. I 69 * have looked at memory allocation sizes on many of the machines available to 70 * me, and there does not seem to be an abundance of allocations at this range 71 * so at this time it may not make sense to optimize for it. This can, of 72 * course, be solved with dynamic slab sizes. 73 * 74 */ 75 76 /* 77 * This is the representation for normal (Non OFFPAGE slab) 78 * 79 * i == item 80 * s == slab pointer 81 * 82 * <---------------- Page (UMA_SLAB_SIZE) ------------------> 83 * ___________________________________________________________ 84 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ | 85 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header|| 86 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________|| 87 * |___________________________________________________________| 88 * 89 * 90 * This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE. 91 * 92 * ___________________________________________________________ 93 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ | 94 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| | 95 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| | 96 * |___________________________________________________________| 97 * ___________ ^ 98 * |slab header| | 99 * |___________|---* 100 * 101 */ 102 103 #ifndef VM_UMA_INT_H 104 #define VM_UMA_INT_H 105 106 #include <sys/mutex.h> 107 108 #define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */ 109 #define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */ 110 #define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */ 111 112 #define UMA_BOOT_PAGES 15 /* Number of pages allocated for startup */ 113 #define UMA_WORKING_TIME 20 /* Seconds worth of items to keep */ 114 115 116 /* Max waste before going to off page slab management */ 117 #define UMA_MAX_WASTE (UMA_SLAB_SIZE / 10) 118 119 /* 120 * I doubt there will be many cases where this is exceeded. This is the initial 121 * size of the hash table for uma_slabs that are managed off page. This hash 122 * does expand by powers of two. Currently it doesn't get smaller. 123 */ 124 #define UMA_HASH_SIZE_INIT 32 125 126 127 /* 128 * I should investigate other hashing algorithms. This should yield a low 129 * number of collisions if the pages are relatively contiguous. 130 * 131 * This is the same algorithm that most processor caches use. 132 * 133 * I'm shifting and masking instead of % because it should be faster. 134 */ 135 136 #define UMA_HASH(h, s) ((((unsigned long)s) >> UMA_SLAB_SHIFT) & \ 137 (h)->uh_hashmask) 138 139 #define UMA_HASH_INSERT(h, s, mem) \ 140 SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \ 141 (mem))], (s), us_hlink); 142 #define UMA_HASH_REMOVE(h, s, mem) \ 143 SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \ 144 (mem))], (s), uma_slab, us_hlink); 145 146 /* Page management structure */ 147 148 /* Sorry for the union, but space efficiency is important */ 149 struct uma_slab { 150 uma_zone_t us_zone; /* Zone we live in */ 151 union { 152 LIST_ENTRY(uma_slab) us_link; /* slabs in zone */ 153 unsigned long us_size; /* Size of allocation */ 154 } us_type; 155 SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */ 156 u_int8_t *us_data; /* First item */ 157 u_int8_t us_flags; /* Page flags see uma.h */ 158 u_int8_t us_freecount; /* How many are free? */ 159 u_int8_t us_firstfree; /* First free item index */ 160 u_int8_t us_freelist[1]; /* Free List (actually larger) */ 161 }; 162 163 #define us_link us_type.us_link 164 #define us_size us_type.us_size 165 166 typedef struct uma_slab * uma_slab_t; 167 168 /* Hash table for freed address -> slab translation */ 169 170 SLIST_HEAD(slabhead, uma_slab); 171 172 struct uma_hash { 173 struct slabhead *uh_slab_hash; /* Hash table for slabs */ 174 int uh_hashsize; /* Current size of the hash table */ 175 int uh_hashmask; /* Mask used during hashing */ 176 }; 177 178 extern struct uma_hash *mallochash; 179 180 /* 181 * Structures for per cpu queues. 182 */ 183 184 /* 185 * This size was chosen so that the struct bucket size is roughly 186 * 128 * sizeof(void *). This is exactly true for x86, and for alpha 187 * it will would be 32bits smaller if it didn't have alignment adjustments. 188 */ 189 190 #define UMA_BUCKET_SIZE 125 191 192 struct uma_bucket { 193 LIST_ENTRY(uma_bucket) ub_link; /* Link into the zone */ 194 int16_t ub_ptr; /* Pointer to current item */ 195 void *ub_bucket[UMA_BUCKET_SIZE]; /* actual allocation storage */ 196 }; 197 198 typedef struct uma_bucket * uma_bucket_t; 199 200 struct uma_cache { 201 struct mtx uc_lock; /* Spin lock on this cpu's bucket */ 202 uma_bucket_t uc_freebucket; /* Bucket we're freeing to */ 203 uma_bucket_t uc_allocbucket; /* Bucket to allocate from */ 204 u_int64_t uc_allocs; /* Count of allocations */ 205 }; 206 207 typedef struct uma_cache * uma_cache_t; 208 209 #define LOCKNAME_LEN 16 /* Length of the name for cpu locks */ 210 211 /* 212 * Zone management structure 213 * 214 * TODO: Optimize for cache line size 215 * 216 */ 217 struct uma_zone { 218 char uz_lname[LOCKNAME_LEN]; /* Text name for the cpu lock */ 219 char *uz_name; /* Text name of the zone */ 220 LIST_ENTRY(uma_zone) uz_link; /* List of all zones */ 221 u_int32_t uz_align; /* Alignment mask */ 222 u_int32_t uz_pages; /* Total page count */ 223 224 /* Used during alloc / free */ 225 struct mtx uz_lock; /* Lock for the zone */ 226 u_int32_t uz_free; /* Count of items free in slabs */ 227 u_int16_t uz_ipers; /* Items per slab */ 228 u_int16_t uz_flags; /* Internal flags */ 229 230 LIST_HEAD(,uma_slab) uz_part_slab; /* partially allocated slabs */ 231 LIST_HEAD(,uma_slab) uz_free_slab; /* empty slab list */ 232 LIST_HEAD(,uma_slab) uz_full_slab; /* full slabs */ 233 LIST_HEAD(,uma_bucket) uz_full_bucket; /* full buckets */ 234 LIST_HEAD(,uma_bucket) uz_free_bucket; /* Buckets for frees */ 235 u_int32_t uz_size; /* Requested size of each item */ 236 u_int32_t uz_rsize; /* Real size of each item */ 237 238 struct uma_hash uz_hash; 239 u_int16_t uz_pgoff; /* Offset to uma_slab struct */ 240 u_int16_t uz_ppera; /* pages per allocation from backend */ 241 u_int16_t uz_cacheoff; /* Next cache offset */ 242 u_int16_t uz_cachemax; /* Max cache offset */ 243 244 uma_ctor uz_ctor; /* Constructor for each allocation */ 245 uma_dtor uz_dtor; /* Destructor */ 246 u_int64_t uz_allocs; /* Total number of allocations */ 247 248 uma_init uz_init; /* Initializer for each item */ 249 uma_fini uz_fini; /* Discards memory */ 250 uma_alloc uz_allocf; /* Allocation function */ 251 uma_free uz_freef; /* Free routine */ 252 struct vm_object *uz_obj; /* Zone specific object */ 253 vm_offset_t uz_kva; /* Base kva for zones with objs */ 254 u_int32_t uz_maxpages; /* Maximum number of pages to alloc */ 255 u_int32_t uz_cachefree; /* Last count of items free in caches */ 256 u_int64_t uz_oallocs; /* old allocs count */ 257 u_int64_t uz_wssize; /* Working set size */ 258 int uz_recurse; /* Allocation recursion count */ 259 uint16_t uz_fills; /* Outstanding bucket fills */ 260 uint16_t uz_count; /* Highest value ub_ptr can have */ 261 /* 262 * This HAS to be the last item because we adjust the zone size 263 * based on NCPU and then allocate the space for the zones. 264 */ 265 struct uma_cache uz_cpu[1]; /* Per cpu caches */ 266 }; 267 268 #define UMA_CACHE_INC 16 /* How much will we move data */ 269 270 #define UMA_ZFLAG_OFFPAGE 0x0001 /* Struct slab/freelist off page */ 271 #define UMA_ZFLAG_PRIVALLOC 0x0002 /* Zone has supplied it's own alloc */ 272 #define UMA_ZFLAG_INTERNAL 0x0004 /* Internal zone, no offpage no PCPU */ 273 #define UMA_ZFLAG_MALLOC 0x0008 /* Zone created by malloc */ 274 #define UMA_ZFLAG_NOFREE 0x0010 /* Don't free data from this zone */ 275 #define UMA_ZFLAG_FULL 0x0020 /* This zone reached uz_maxpages */ 276 277 /* This lives in uflags */ 278 #define UMA_ZONE_INTERNAL 0x1000 /* Internal zone for uflags */ 279 280 /* Internal prototypes */ 281 static __inline uma_slab_t hash_sfind(struct uma_hash *hash, u_int8_t *data); 282 void *uma_large_malloc(int size, int wait); 283 void uma_large_free(uma_slab_t slab); 284 285 /* Lock Macros */ 286 287 #define ZONE_LOCK_INIT(z, lc) \ 288 do { \ 289 if ((lc)) \ 290 mtx_init(&(z)->uz_lock, (z)->uz_name, \ 291 (z)->uz_name, MTX_DEF | MTX_DUPOK); \ 292 else \ 293 mtx_init(&(z)->uz_lock, (z)->uz_name, \ 294 "UMA zone", MTX_DEF | MTX_DUPOK); \ 295 } while (0) 296 297 #define ZONE_LOCK_FINI(z) mtx_destroy(&(z)->uz_lock) 298 #define ZONE_LOCK(z) mtx_lock(&(z)->uz_lock) 299 #define ZONE_UNLOCK(z) mtx_unlock(&(z)->uz_lock) 300 301 #define CPU_LOCK_INIT(z, cpu, lc) \ 302 do { \ 303 if ((lc)) \ 304 mtx_init(&(z)->uz_cpu[(cpu)].uc_lock, \ 305 (z)->uz_lname, (z)->uz_lname, \ 306 MTX_DEF | MTX_DUPOK); \ 307 else \ 308 mtx_init(&(z)->uz_cpu[(cpu)].uc_lock, \ 309 (z)->uz_lname, "UMA cpu", \ 310 MTX_DEF | MTX_DUPOK); \ 311 } while (0) 312 313 #define CPU_LOCK_FINI(z, cpu) \ 314 mtx_destroy(&(z)->uz_cpu[(cpu)].uc_lock) 315 316 #define CPU_LOCK(z, cpu) \ 317 mtx_lock(&(z)->uz_cpu[(cpu)].uc_lock) 318 319 #define CPU_UNLOCK(z, cpu) \ 320 mtx_unlock(&(z)->uz_cpu[(cpu)].uc_lock) 321 322 /* 323 * Find a slab within a hash table. This is used for OFFPAGE zones to lookup 324 * the slab structure. 325 * 326 * Arguments: 327 * hash The hash table to search. 328 * data The base page of the item. 329 * 330 * Returns: 331 * A pointer to a slab if successful, else NULL. 332 */ 333 static __inline uma_slab_t 334 hash_sfind(struct uma_hash *hash, u_int8_t *data) 335 { 336 uma_slab_t slab; 337 int hval; 338 339 hval = UMA_HASH(hash, data); 340 341 SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) { 342 if ((u_int8_t *)slab->us_data == data) 343 return (slab); 344 } 345 return (NULL); 346 } 347 348 349 #endif /* VM_UMA_INT_H */ 350