6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice unmodified, this list of conditions, and the following 12 * disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 *
|
29 * 30 */ 31 32/* 33 * This file includes definitions, structures, prototypes, and inlines that 34 * should not be used outside of the actual implementation of UMA. 35 */ 36 37/* 38 * Here's a quick description of the relationship between the objects: 39 * 40 * Kegs contain lists of slabs which are stored in either the full bin, empty 41 * bin, or partially allocated bin, to reduce fragmentation. They also contain 42 * the user supplied value for size, which is adjusted for alignment purposes 43 * and rsize is the result of that. The Keg also stores information for 44 * managing a hash of page addresses that maps pages to uma_slab_t structures 45 * for pages that don't have embedded uma_slab_t's. 46 * 47 * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may 48 * be allocated off the page from a special slab zone. The free list within a 49 * slab is managed with a linked list of indexes, which are 8 bit values. If 50 * UMA_SLAB_SIZE is defined to be too large I will have to switch to 16bit 51 * values. Currently on alpha you can get 250 or so 32 byte items and on x86 52 * you can get 250 or so 16byte items. For item sizes that would yield more 53 * than 10% memory waste we potentially allocate a separate uma_slab_t if this 54 * will improve the number of items per slab that will fit. 55 * 56 * Other potential space optimizations are storing the 8bit of linkage in space 57 * wasted between items due to alignment problems. This may yield a much better 58 * memory footprint for certain sizes of objects. Another alternative is to 59 * increase the UMA_SLAB_SIZE, or allow for dynamic slab sizes. I prefer 60 * dynamic slab sizes because we could stick with 8 bit indexes and only use 61 * large slab sizes for zones with a lot of waste per slab. This may create 62 * ineffeciencies in the vm subsystem due to fragmentation in the address space. 63 * 64 * The only really gross cases, with regards to memory waste, are for those 65 * items that are just over half the page size. You can get nearly 50% waste, 66 * so you fall back to the memory footprint of the power of two allocator. I 67 * have looked at memory allocation sizes on many of the machines available to 68 * me, and there does not seem to be an abundance of allocations at this range 69 * so at this time it may not make sense to optimize for it. This can, of 70 * course, be solved with dynamic slab sizes. 71 * 72 * Kegs may serve multiple Zones but by far most of the time they only serve 73 * one. When a Zone is created, a Keg is allocated and setup for it. While 74 * the backing Keg stores slabs, the Zone caches Buckets of items allocated 75 * from the slabs. Each Zone is equipped with an init/fini and ctor/dtor 76 * pair, as well as with its own set of small per-CPU caches, layered above 77 * the Zone's general Bucket cache. 78 * 79 * The PCPU caches are protected by their own locks, while the Zones backed 80 * by the same Keg all share a common Keg lock (to coalesce contention on 81 * the backing slabs). The backing Keg typically only serves one Zone but 82 * in the case of multiple Zones, one of the Zones is considered the 83 * Master Zone and all Zone-related stats from the Keg are done in the 84 * Master Zone. For an example of a Multi-Zone setup, refer to the 85 * Mbuf allocation code. 86 */ 87 88/* 89 * This is the representation for normal (Non OFFPAGE slab) 90 * 91 * i == item 92 * s == slab pointer 93 * 94 * <---------------- Page (UMA_SLAB_SIZE) ------------------> 95 * ___________________________________________________________ 96 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ | 97 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header|| 98 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________|| 99 * |___________________________________________________________| 100 * 101 * 102 * This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE. 103 * 104 * ___________________________________________________________ 105 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ | 106 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| | 107 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| | 108 * |___________________________________________________________| 109 * ___________ ^ 110 * |slab header| | 111 * |___________|---* 112 * 113 */ 114 115#ifndef VM_UMA_INT_H 116#define VM_UMA_INT_H 117 118#define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */ 119#define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */ 120#define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */ 121 122#define UMA_BOOT_PAGES 40 /* Pages allocated for startup */ 123 124/* Max waste before going to off page slab management */ 125#define UMA_MAX_WASTE (UMA_SLAB_SIZE / 10) 126 127/* 128 * I doubt there will be many cases where this is exceeded. This is the initial 129 * size of the hash table for uma_slabs that are managed off page. This hash 130 * does expand by powers of two. Currently it doesn't get smaller. 131 */ 132#define UMA_HASH_SIZE_INIT 32 133 134/* 135 * I should investigate other hashing algorithms. This should yield a low 136 * number of collisions if the pages are relatively contiguous. 137 * 138 * This is the same algorithm that most processor caches use. 139 * 140 * I'm shifting and masking instead of % because it should be faster. 141 */ 142 143#define UMA_HASH(h, s) ((((unsigned long)s) >> UMA_SLAB_SHIFT) & \ 144 (h)->uh_hashmask) 145 146#define UMA_HASH_INSERT(h, s, mem) \ 147 SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \ 148 (mem))], (s), us_hlink); 149#define UMA_HASH_REMOVE(h, s, mem) \ 150 SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \ 151 (mem))], (s), uma_slab, us_hlink); 152 153/* Hash table for freed address -> slab translation */ 154 155SLIST_HEAD(slabhead, uma_slab); 156 157struct uma_hash { 158 struct slabhead *uh_slab_hash; /* Hash table for slabs */ 159 int uh_hashsize; /* Current size of the hash table */ 160 int uh_hashmask; /* Mask used during hashing */ 161}; 162 163/* 164 * Structures for per cpu queues. 165 */ 166 167struct uma_bucket { 168 LIST_ENTRY(uma_bucket) ub_link; /* Link into the zone */ 169 int16_t ub_cnt; /* Count of free items. */ 170 int16_t ub_entries; /* Max items. */ 171 void *ub_bucket[]; /* actual allocation storage */ 172}; 173 174typedef struct uma_bucket * uma_bucket_t; 175 176struct uma_cache { 177 uma_bucket_t uc_freebucket; /* Bucket we're freeing to */ 178 uma_bucket_t uc_allocbucket; /* Bucket to allocate from */ 179 u_int64_t uc_allocs; /* Count of allocations */ 180}; 181 182typedef struct uma_cache * uma_cache_t; 183 184/* 185 * Keg management structure 186 * 187 * TODO: Optimize for cache line size 188 * 189 */ 190struct uma_keg { 191 LIST_ENTRY(uma_keg) uk_link; /* List of all kegs */ 192 193 struct mtx uk_lock; /* Lock for the keg */ 194 struct uma_hash uk_hash; 195 196 LIST_HEAD(,uma_zone) uk_zones; /* Keg's zones */ 197 LIST_HEAD(,uma_slab) uk_part_slab; /* partially allocated slabs */ 198 LIST_HEAD(,uma_slab) uk_free_slab; /* empty slab list */ 199 LIST_HEAD(,uma_slab) uk_full_slab; /* full slabs */ 200 201 u_int32_t uk_recurse; /* Allocation recursion count */ 202 u_int32_t uk_align; /* Alignment mask */ 203 u_int32_t uk_pages; /* Total page count */ 204 u_int32_t uk_free; /* Count of items free in slabs */ 205 u_int32_t uk_size; /* Requested size of each item */ 206 u_int32_t uk_rsize; /* Real size of each item */ 207 u_int32_t uk_maxpages; /* Maximum number of pages to alloc */ 208 209 uma_init uk_init; /* Keg's init routine */ 210 uma_fini uk_fini; /* Keg's fini routine */ 211 uma_alloc uk_allocf; /* Allocation function */ 212 uma_free uk_freef; /* Free routine */ 213 214 struct vm_object *uk_obj; /* Zone specific object */ 215 vm_offset_t uk_kva; /* Base kva for zones with objs */ 216 uma_zone_t uk_slabzone; /* Slab zone backing us, if OFFPAGE */ 217 218 u_int16_t uk_pgoff; /* Offset to uma_slab struct */ 219 u_int16_t uk_ppera; /* pages per allocation from backend */ 220 u_int16_t uk_ipers; /* Items per slab */ 221 u_int16_t uk_flags; /* Internal flags */ 222}; 223 224/* Simpler reference to uma_keg for internal use. */ 225typedef struct uma_keg * uma_keg_t; 226 227/* Page management structure */ 228 229/* Sorry for the union, but space efficiency is important */ 230struct uma_slab_head { 231 uma_keg_t us_keg; /* Keg we live in */ 232 union { 233 LIST_ENTRY(uma_slab) _us_link; /* slabs in zone */ 234 unsigned long _us_size; /* Size of allocation */ 235 } us_type; 236 SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */ 237 u_int8_t *us_data; /* First item */ 238 u_int8_t us_flags; /* Page flags see uma.h */ 239 u_int8_t us_freecount; /* How many are free? */ 240 u_int8_t us_firstfree; /* First free item index */ 241}; 242 243/* The standard slab structure */ 244struct uma_slab { 245 struct uma_slab_head us_head; /* slab header data */ 246 struct { 247 u_int8_t us_item; 248 } us_freelist[1]; /* actual number bigger */ 249}; 250 251/* 252 * The slab structure for UMA_ZONE_REFCNT zones for whose items we 253 * maintain reference counters in the slab for. 254 */ 255struct uma_slab_refcnt { 256 struct uma_slab_head us_head; /* slab header data */ 257 struct { 258 u_int8_t us_item; 259 u_int32_t us_refcnt; 260 } us_freelist[1]; /* actual number bigger */ 261}; 262 263#define us_keg us_head.us_keg 264#define us_link us_head.us_type._us_link 265#define us_size us_head.us_type._us_size 266#define us_hlink us_head.us_hlink 267#define us_data us_head.us_data 268#define us_flags us_head.us_flags 269#define us_freecount us_head.us_freecount 270#define us_firstfree us_head.us_firstfree 271 272typedef struct uma_slab * uma_slab_t; 273typedef struct uma_slab_refcnt * uma_slabrefcnt_t; 274 275/* 276 * These give us the size of one free item reference within our corresponding 277 * uma_slab structures, so that our calculations during zone setup are correct 278 * regardless of what the compiler decides to do with padding the structure 279 * arrays within uma_slab. 280 */ 281#define UMA_FRITM_SZ (sizeof(struct uma_slab) - sizeof(struct uma_slab_head)) 282#define UMA_FRITMREF_SZ (sizeof(struct uma_slab_refcnt) - \ 283 sizeof(struct uma_slab_head)) 284 285/* 286 * Zone management structure 287 * 288 * TODO: Optimize for cache line size 289 * 290 */ 291struct uma_zone { 292 char *uz_name; /* Text name of the zone */ 293 struct mtx *uz_lock; /* Lock for the zone (keg's lock) */ 294 uma_keg_t uz_keg; /* Our underlying Keg */ 295 296 LIST_ENTRY(uma_zone) uz_link; /* List of all zones in keg */ 297 LIST_HEAD(,uma_bucket) uz_full_bucket; /* full buckets */ 298 LIST_HEAD(,uma_bucket) uz_free_bucket; /* Buckets for frees */ 299 300 uma_ctor uz_ctor; /* Constructor for each allocation */ 301 uma_dtor uz_dtor; /* Destructor */ 302 uma_init uz_init; /* Initializer for each item */ 303 uma_fini uz_fini; /* Discards memory */ 304 305 u_int64_t uz_allocs; /* Total number of allocations */ 306 uint16_t uz_fills; /* Outstanding bucket fills */ 307 uint16_t uz_count; /* Highest value ub_ptr can have */ 308 309 /* 310 * This HAS to be the last item because we adjust the zone size 311 * based on NCPU and then allocate the space for the zones. 312 */ 313 struct uma_cache uz_cpu[1]; /* Per cpu caches */ 314}; 315 316/* 317 * These flags must not overlap with the UMA_ZONE flags specified in uma.h. 318 */ 319#define UMA_ZFLAG_PRIVALLOC 0x1000 /* Use uz_allocf. */ 320#define UMA_ZFLAG_INTERNAL 0x2000 /* No offpage no PCPU. */ 321#define UMA_ZFLAG_FULL 0x4000 /* Reached uz_maxpages */ 322#define UMA_ZFLAG_CACHEONLY 0x8000 /* Don't ask VM for buckets. */ 323 324/* Internal prototypes */ 325static __inline uma_slab_t hash_sfind(struct uma_hash *hash, u_int8_t *data); 326void *uma_large_malloc(int size, int wait); 327void uma_large_free(uma_slab_t slab); 328 329/* Lock Macros */ 330 331#define ZONE_LOCK_INIT(z, lc) \ 332 do { \ 333 if ((lc)) \ 334 mtx_init((z)->uz_lock, (z)->uz_name, \ 335 (z)->uz_name, MTX_DEF | MTX_DUPOK); \ 336 else \ 337 mtx_init((z)->uz_lock, (z)->uz_name, \ 338 "UMA zone", MTX_DEF | MTX_DUPOK); \ 339 } while (0) 340 341#define ZONE_LOCK_FINI(z) mtx_destroy((z)->uz_lock) 342#define ZONE_LOCK(z) mtx_lock((z)->uz_lock) 343#define ZONE_UNLOCK(z) mtx_unlock((z)->uz_lock) 344 345#define CPU_LOCK_INIT(cpu) \ 346 mtx_init(&uma_pcpu_mtx[(cpu)], "UMA pcpu", "UMA pcpu", \ 347 MTX_DEF | MTX_DUPOK) 348 349#define CPU_LOCK(cpu) \ 350 mtx_lock(&uma_pcpu_mtx[(cpu)]) 351 352#define CPU_UNLOCK(cpu) \ 353 mtx_unlock(&uma_pcpu_mtx[(cpu)]) 354 355/* 356 * Find a slab within a hash table. This is used for OFFPAGE zones to lookup 357 * the slab structure. 358 * 359 * Arguments: 360 * hash The hash table to search. 361 * data The base page of the item. 362 * 363 * Returns: 364 * A pointer to a slab if successful, else NULL. 365 */ 366static __inline uma_slab_t 367hash_sfind(struct uma_hash *hash, u_int8_t *data) 368{ 369 uma_slab_t slab; 370 int hval; 371 372 hval = UMA_HASH(hash, data); 373 374 SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) { 375 if ((u_int8_t *)slab->us_data == data) 376 return (slab); 377 } 378 return (NULL); 379} 380 381static __inline uma_slab_t 382vtoslab(vm_offset_t va) 383{ 384 vm_page_t p; 385 uma_slab_t slab; 386 387 p = PHYS_TO_VM_PAGE(pmap_kextract(va)); 388 slab = (uma_slab_t )p->object; 389 390 if (p->flags & PG_SLAB) 391 return (slab); 392 else 393 return (NULL); 394} 395 396static __inline void 397vsetslab(vm_offset_t va, uma_slab_t slab) 398{ 399 vm_page_t p; 400 401 p = PHYS_TO_VM_PAGE(pmap_kextract(va)); 402 p->object = (vm_object_t)slab; 403 p->flags |= PG_SLAB; 404} 405 406static __inline void 407vsetobj(vm_offset_t va, vm_object_t obj) 408{ 409 vm_page_t p; 410 411 p = PHYS_TO_VM_PAGE(pmap_kextract(va)); 412 p->object = obj; 413 p->flags &= ~PG_SLAB; 414} 415 416/* 417 * The following two functions may be defined by architecture specific code 418 * if they can provide more effecient allocation functions. This is useful 419 * for using direct mapped addresses. 420 */ 421void *uma_small_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait); 422void uma_small_free(void *mem, int size, u_int8_t flags); 423 424#endif /* VM_UMA_INT_H */
|