1 /* 2 * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org> 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 * This file includes definitions, structures, prototypes, and inlines that 32 * should not be used outside of the actual implementation of UMA. 33 */ 34 35 /* 36 * Here's a quick description of the relationship between the objects: 37 * 38 * Zones contain lists of slabs which are stored in either the full bin, empty 39 * bin, or partially allocated bin, to reduce fragmentation. They also contain 40 * the user supplied value for size, which is adjusted for alignment purposes 41 * and rsize is the result of that. The zone also stores information for 42 * managing a hash of page addresses that maps pages to uma_slab_t structures 43 * for pages that don't have embedded uma_slab_t's. 44 * 45 * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may 46 * be allocated off the page from a special slab zone. The free list within a 47 * slab is managed with a linked list of indexes, which are 8 bit values. If 48 * UMA_SLAB_SIZE is defined to be too large I will have to switch to 16bit 49 * values. Currently on alpha you can get 250 or so 32 byte items and on x86 50 * you can get 250 or so 16byte items. For item sizes that would yield more 51 * than 10% memory waste we potentially allocate a separate uma_slab_t if this 52 * will improve the number of items per slab that will fit. 53 * 54 * Other potential space optimizations are storing the 8bit of linkage in space 55 * wasted between items due to alignment problems. This may yield a much better 56 * memory footprint for certain sizes of objects. Another alternative is to 57 * increase the UMA_SLAB_SIZE, or allow for dynamic slab sizes. I prefer 58 * dynamic slab sizes because we could stick with 8 bit indexes and only use 59 * large slab sizes for zones with a lot of waste per slab. This may create 60 * ineffeciencies in the vm subsystem due to fragmentation in the address space. 61 * 62 * The only really gross cases, with regards to memory waste, are for those 63 * items that are just over half the page size. You can get nearly 50% waste, 64 * so you fall back to the memory footprint of the power of two allocator. I 65 * have looked at memory allocation sizes on many of the machines available to 66 * me, and there does not seem to be an abundance of allocations at this range 67 * so at this time it may not make sense to optimize for it. This can, of 68 * course, be solved with dynamic slab sizes. 69 * 70 */ 71 72 /* 73 * This is the representation for normal (Non OFFPAGE slab) 74 * 75 * i == item 76 * s == slab pointer 77 * 78 * <---------------- Page (UMA_SLAB_SIZE) ------------------> 79 * ___________________________________________________________ 80 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ | 81 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header|| 82 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________|| 83 * |___________________________________________________________| 84 * 85 * 86 * This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE. 87 * 88 * ___________________________________________________________ 89 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ | 90 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| | 91 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| | 92 * |___________________________________________________________| 93 * ___________ ^ 94 * |slab header| | 95 * |___________|---* 96 * 97 */ 98 99 #ifndef VM_UMA_INT_H 100 #define VM_UMA_INT_H 101 102 #define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */ 103 #define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */ 104 #define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */ 105 106 #define UMA_BOOT_PAGES 30 /* Number of pages allocated for startup */ 107 #define UMA_WORKING_TIME 20 /* Seconds worth of items to keep */ 108 109 110 /* Max waste before going to off page slab management */ 111 #define UMA_MAX_WASTE (UMA_SLAB_SIZE / 10) 112 113 /* 114 * I doubt there will be many cases where this is exceeded. This is the initial 115 * size of the hash table for uma_slabs that are managed off page. This hash 116 * does expand by powers of two. Currently it doesn't get smaller. 117 */ 118 #define UMA_HASH_SIZE_INIT 32 119 120 121 /* 122 * I should investigate other hashing algorithms. This should yield a low 123 * number of collisions if the pages are relatively contiguous. 124 * 125 * This is the same algorithm that most processor caches use. 126 * 127 * I'm shifting and masking instead of % because it should be faster. 128 */ 129 130 #define UMA_HASH(h, s) ((((unsigned long)s) >> UMA_SLAB_SHIFT) & \ 131 (h)->uh_hashmask) 132 133 #define UMA_HASH_INSERT(h, s, mem) \ 134 SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \ 135 (mem))], (s), us_hlink); 136 #define UMA_HASH_REMOVE(h, s, mem) \ 137 SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \ 138 (mem))], (s), uma_slab, us_hlink); 139 140 /* Page management structure */ 141 142 /* Sorry for the union, but space efficiency is important */ 143 struct uma_slab { 144 uma_zone_t us_zone; /* Zone we live in */ 145 union { 146 LIST_ENTRY(uma_slab) _us_link; /* slabs in zone */ 147 unsigned long _us_size; /* Size of allocation */ 148 } us_type; 149 SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */ 150 u_int8_t *us_data; /* First item */ 151 u_int8_t us_flags; /* Page flags see uma.h */ 152 u_int8_t us_freecount; /* How many are free? */ 153 u_int8_t us_firstfree; /* First free item index */ 154 u_int8_t us_freelist[1]; /* Free List (actually larger) */ 155 }; 156 157 #define us_link us_type._us_link 158 #define us_size us_type._us_size 159 160 typedef struct uma_slab * uma_slab_t; 161 162 /* Hash table for freed address -> slab translation */ 163 164 SLIST_HEAD(slabhead, uma_slab); 165 166 struct uma_hash { 167 struct slabhead *uh_slab_hash; /* Hash table for slabs */ 168 int uh_hashsize; /* Current size of the hash table */ 169 int uh_hashmask; /* Mask used during hashing */ 170 }; 171 172 /* 173 * Structures for per cpu queues. 174 */ 175 176 /* 177 * This size was chosen so that the struct bucket size is roughly 178 * 128 * sizeof(void *). This is exactly true for x86, and for alpha 179 * it will would be 32bits smaller if it didn't have alignment adjustments. 180 */ 181 182 #define UMA_BUCKET_SIZE 125 183 184 struct uma_bucket { 185 LIST_ENTRY(uma_bucket) ub_link; /* Link into the zone */ 186 int16_t ub_ptr; /* Pointer to current item */ 187 void *ub_bucket[UMA_BUCKET_SIZE]; /* actual allocation storage */ 188 }; 189 190 typedef struct uma_bucket * uma_bucket_t; 191 192 struct uma_cache { 193 uma_bucket_t uc_freebucket; /* Bucket we're freeing to */ 194 uma_bucket_t uc_allocbucket; /* Bucket to allocate from */ 195 u_int64_t uc_allocs; /* Count of allocations */ 196 }; 197 198 typedef struct uma_cache * uma_cache_t; 199 200 /* 201 * Zone management structure 202 * 203 * TODO: Optimize for cache line size 204 * 205 */ 206 struct uma_zone { 207 char *uz_name; /* Text name of the zone */ 208 LIST_ENTRY(uma_zone) uz_link; /* List of all zones */ 209 u_int32_t uz_align; /* Alignment mask */ 210 u_int32_t uz_pages; /* Total page count */ 211 212 /* Used during alloc / free */ 213 struct mtx uz_lock; /* Lock for the zone */ 214 u_int32_t uz_free; /* Count of items free in slabs */ 215 u_int16_t uz_ipers; /* Items per slab */ 216 u_int16_t uz_flags; /* Internal flags */ 217 218 LIST_HEAD(,uma_slab) uz_part_slab; /* partially allocated slabs */ 219 LIST_HEAD(,uma_slab) uz_free_slab; /* empty slab list */ 220 LIST_HEAD(,uma_slab) uz_full_slab; /* full slabs */ 221 LIST_HEAD(,uma_bucket) uz_full_bucket; /* full buckets */ 222 LIST_HEAD(,uma_bucket) uz_free_bucket; /* Buckets for frees */ 223 u_int32_t uz_size; /* Requested size of each item */ 224 u_int32_t uz_rsize; /* Real size of each item */ 225 226 struct uma_hash uz_hash; 227 u_int16_t uz_pgoff; /* Offset to uma_slab struct */ 228 u_int16_t uz_ppera; /* pages per allocation from backend */ 229 u_int16_t uz_cacheoff; /* Next cache offset */ 230 u_int16_t uz_cachemax; /* Max cache offset */ 231 232 uma_ctor uz_ctor; /* Constructor for each allocation */ 233 uma_dtor uz_dtor; /* Destructor */ 234 u_int64_t uz_allocs; /* Total number of allocations */ 235 236 uma_init uz_init; /* Initializer for each item */ 237 uma_fini uz_fini; /* Discards memory */ 238 uma_alloc uz_allocf; /* Allocation function */ 239 uma_free uz_freef; /* Free routine */ 240 struct vm_object *uz_obj; /* Zone specific object */ 241 vm_offset_t uz_kva; /* Base kva for zones with objs */ 242 u_int32_t uz_maxpages; /* Maximum number of pages to alloc */ 243 u_int32_t uz_cachefree; /* Last count of items free in caches */ 244 u_int64_t uz_oallocs; /* old allocs count */ 245 u_int64_t uz_wssize; /* Working set size */ 246 int uz_recurse; /* Allocation recursion count */ 247 uint16_t uz_fills; /* Outstanding bucket fills */ 248 uint16_t uz_count; /* Highest value ub_ptr can have */ 249 /* 250 * This HAS to be the last item because we adjust the zone size 251 * based on NCPU and then allocate the space for the zones. 252 */ 253 struct uma_cache uz_cpu[1]; /* Per cpu caches */ 254 }; 255 256 #define UMA_CACHE_INC 16 /* How much will we move data */ 257 258 #define UMA_ZFLAG_OFFPAGE 0x0001 /* Struct slab/freelist off page */ 259 #define UMA_ZFLAG_PRIVALLOC 0x0002 /* Zone has supplied it's own alloc */ 260 #define UMA_ZFLAG_INTERNAL 0x0004 /* Internal zone, no offpage no PCPU */ 261 #define UMA_ZFLAG_MALLOC 0x0008 /* Zone created by malloc */ 262 #define UMA_ZFLAG_NOFREE 0x0010 /* Don't free data from this zone */ 263 #define UMA_ZFLAG_FULL 0x0020 /* This zone reached uz_maxpages */ 264 #define UMA_ZFLAG_BUCKETCACHE 0x0040 /* Only allocate buckets from cache */ 265 #define UMA_ZFLAG_HASH 0x0080 /* Look up slab via hash */ 266 267 /* This lives in uflags */ 268 #define UMA_ZONE_INTERNAL 0x1000 /* Internal zone for uflags */ 269 270 /* Internal prototypes */ 271 static __inline uma_slab_t hash_sfind(struct uma_hash *hash, u_int8_t *data); 272 void *uma_large_malloc(int size, int wait); 273 void uma_large_free(uma_slab_t slab); 274 275 /* Lock Macros */ 276 277 #define ZONE_LOCK_INIT(z, lc) \ 278 do { \ 279 if ((lc)) \ 280 mtx_init(&(z)->uz_lock, (z)->uz_name, \ 281 (z)->uz_name, MTX_DEF | MTX_DUPOK); \ 282 else \ 283 mtx_init(&(z)->uz_lock, (z)->uz_name, \ 284 "UMA zone", MTX_DEF | MTX_DUPOK); \ 285 } while (0) 286 287 #define ZONE_LOCK_FINI(z) mtx_destroy(&(z)->uz_lock) 288 #define ZONE_LOCK(z) mtx_lock(&(z)->uz_lock) 289 #define ZONE_UNLOCK(z) mtx_unlock(&(z)->uz_lock) 290 291 #define CPU_LOCK_INIT(cpu) \ 292 mtx_init(&uma_pcpu_mtx[(cpu)], "UMA pcpu", "UMA pcpu", \ 293 MTX_DEF | MTX_DUPOK) 294 295 #define CPU_LOCK(cpu) \ 296 mtx_lock(&uma_pcpu_mtx[(cpu)]) 297 298 #define CPU_UNLOCK(cpu) \ 299 mtx_unlock(&uma_pcpu_mtx[(cpu)]) 300 301 /* 302 * Find a slab within a hash table. This is used for OFFPAGE zones to lookup 303 * the slab structure. 304 * 305 * Arguments: 306 * hash The hash table to search. 307 * data The base page of the item. 308 * 309 * Returns: 310 * A pointer to a slab if successful, else NULL. 311 */ 312 static __inline uma_slab_t 313 hash_sfind(struct uma_hash *hash, u_int8_t *data) 314 { 315 uma_slab_t slab; 316 int hval; 317 318 hval = UMA_HASH(hash, data); 319 320 SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) { 321 if ((u_int8_t *)slab->us_data == data) 322 return (slab); 323 } 324 return (NULL); 325 } 326 327 static __inline uma_slab_t 328 vtoslab(vm_offset_t va) 329 { 330 vm_page_t p; 331 uma_slab_t slab; 332 333 p = PHYS_TO_VM_PAGE(pmap_kextract(va)); 334 slab = (uma_slab_t )p->object; 335 336 if (p->flags & PG_SLAB) 337 return (slab); 338 else 339 return (NULL); 340 } 341 342 static __inline void 343 vsetslab(vm_offset_t va, uma_slab_t slab) 344 { 345 vm_page_t p; 346 347 p = PHYS_TO_VM_PAGE(pmap_kextract((vm_offset_t)va)); 348 p->object = (vm_object_t)slab; 349 p->flags |= PG_SLAB; 350 } 351 352 static __inline void 353 vsetobj(vm_offset_t va, vm_object_t obj) 354 { 355 vm_page_t p; 356 357 p = PHYS_TO_VM_PAGE(pmap_kextract((vm_offset_t)va)); 358 p->object = obj; 359 p->flags &= ~PG_SLAB; 360 } 361 362 /* 363 * The following two functions may be defined by architecture specific code 364 * if they can provide more effecient allocation functions. This is useful 365 * for using direct mapped addresses. 366 */ 367 void *uma_small_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait); 368 void uma_small_free(void *mem, int size, u_int8_t flags); 369 370 #endif /* VM_UMA_INT_H */ 371