1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2002-2019 Jeffrey Roberson <jeff@FreeBSD.org> 5 * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org> 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice unmodified, this list of conditions, and the following 13 * disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 * 29 * $FreeBSD$ 30 * 31 */ 32 33 #include <sys/counter.h> 34 #include <sys/_bitset.h> 35 #include <sys/_domainset.h> 36 #include <sys/_task.h> 37 38 /* 39 * This file includes definitions, structures, prototypes, and inlines that 40 * should not be used outside of the actual implementation of UMA. 41 */ 42 43 /* 44 * The brief summary; Zones describe unique allocation types. Zones are 45 * organized into per-CPU caches which are filled by buckets. Buckets are 46 * organized according to memory domains. Buckets are filled from kegs which 47 * are also organized according to memory domains. Kegs describe a unique 48 * allocation type, backend memory provider, and layout. Kegs are associated 49 * with one or more zones and zones reference one or more kegs. Kegs provide 50 * slabs which are virtually contiguous collections of pages. Each slab is 51 * broken down int one or more items that will satisfy an individual allocation. 52 * 53 * Allocation is satisfied in the following order: 54 * 1) Per-CPU cache 55 * 2) Per-domain cache of buckets 56 * 3) Slab from any of N kegs 57 * 4) Backend page provider 58 * 59 * More detail on individual objects is contained below: 60 * 61 * Kegs contain lists of slabs which are stored in either the full bin, empty 62 * bin, or partially allocated bin, to reduce fragmentation. They also contain 63 * the user supplied value for size, which is adjusted for alignment purposes 64 * and rsize is the result of that. The Keg also stores information for 65 * managing a hash of page addresses that maps pages to uma_slab_t structures 66 * for pages that don't have embedded uma_slab_t's. 67 * 68 * Keg slab lists are organized by memory domain to support NUMA allocation 69 * policies. By default allocations are spread across domains to reduce the 70 * potential for hotspots. Special keg creation flags may be specified to 71 * prefer location allocation. However there is no strict enforcement as frees 72 * may happen on any CPU and these are returned to the CPU-local cache 73 * regardless of the originating domain. 74 * 75 * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may 76 * be allocated off the page from a special slab zone. The free list within a 77 * slab is managed with a bitmask. For item sizes that would yield more than 78 * 10% memory waste we potentially allocate a separate uma_slab_t if this will 79 * improve the number of items per slab that will fit. 80 * 81 * The only really gross cases, with regards to memory waste, are for those 82 * items that are just over half the page size. You can get nearly 50% waste, 83 * so you fall back to the memory footprint of the power of two allocator. I 84 * have looked at memory allocation sizes on many of the machines available to 85 * me, and there does not seem to be an abundance of allocations at this range 86 * so at this time it may not make sense to optimize for it. This can, of 87 * course, be solved with dynamic slab sizes. 88 * 89 * Kegs may serve multiple Zones but by far most of the time they only serve 90 * one. When a Zone is created, a Keg is allocated and setup for it. While 91 * the backing Keg stores slabs, the Zone caches Buckets of items allocated 92 * from the slabs. Each Zone is equipped with an init/fini and ctor/dtor 93 * pair, as well as with its own set of small per-CPU caches, layered above 94 * the Zone's general Bucket cache. 95 * 96 * The PCPU caches are protected by critical sections, and may be accessed 97 * safely only from their associated CPU, while the Zones backed by the same 98 * Keg all share a common Keg lock (to coalesce contention on the backing 99 * slabs). The backing Keg typically only serves one Zone but in the case of 100 * multiple Zones, one of the Zones is considered the Master Zone and all 101 * Zone-related stats from the Keg are done in the Master Zone. For an 102 * example of a Multi-Zone setup, refer to the Mbuf allocation code. 103 */ 104 105 /* 106 * This is the representation for normal (Non OFFPAGE slab) 107 * 108 * i == item 109 * s == slab pointer 110 * 111 * <---------------- Page (UMA_SLAB_SIZE) ------------------> 112 * ___________________________________________________________ 113 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ | 114 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header|| 115 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________|| 116 * |___________________________________________________________| 117 * 118 * 119 * This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE. 120 * 121 * ___________________________________________________________ 122 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ | 123 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| | 124 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| | 125 * |___________________________________________________________| 126 * ___________ ^ 127 * |slab header| | 128 * |___________|---* 129 * 130 */ 131 132 #ifndef VM_UMA_INT_H 133 #define VM_UMA_INT_H 134 135 #define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */ 136 #define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */ 137 #define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */ 138 139 /* Max waste percentage before going to off page slab management */ 140 #define UMA_MAX_WASTE 10 141 142 /* Max size of a CACHESPREAD slab. */ 143 #define UMA_CACHESPREAD_MAX_SIZE (128 * 1024) 144 145 /* 146 * These flags must not overlap with the UMA_ZONE flags specified in uma.h. 147 */ 148 #define UMA_ZFLAG_OFFPAGE 0x00200000 /* 149 * Force the slab structure 150 * allocation off of the real 151 * memory. 152 */ 153 #define UMA_ZFLAG_HASH 0x00400000 /* 154 * Use a hash table instead of 155 * caching information in the 156 * vm_page. 157 */ 158 #define UMA_ZFLAG_VTOSLAB 0x00800000 /* 159 * Zone uses vtoslab for 160 * lookup. 161 */ 162 #define UMA_ZFLAG_CTORDTOR 0x01000000 /* Zone has ctor/dtor set. */ 163 #define UMA_ZFLAG_LIMIT 0x02000000 /* Zone has limit set. */ 164 #define UMA_ZFLAG_CACHE 0x04000000 /* uma_zcache_create()d it */ 165 #define UMA_ZFLAG_RECLAIMING 0x08000000 /* Running zone_reclaim(). */ 166 #define UMA_ZFLAG_BUCKET 0x10000000 /* Bucket zone. */ 167 #define UMA_ZFLAG_INTERNAL 0x20000000 /* No offpage no PCPU. */ 168 #define UMA_ZFLAG_TRASH 0x40000000 /* Add trash ctor/dtor. */ 169 #define UMA_ZFLAG_CACHEONLY 0x80000000 /* Don't ask VM for buckets. */ 170 171 #define UMA_ZFLAG_INHERIT \ 172 (UMA_ZFLAG_OFFPAGE | UMA_ZFLAG_HASH | UMA_ZFLAG_VTOSLAB | \ 173 UMA_ZFLAG_BUCKET | UMA_ZFLAG_INTERNAL | UMA_ZFLAG_CACHEONLY) 174 175 #define PRINT_UMA_ZFLAGS "\20" \ 176 "\40CACHEONLY" \ 177 "\37TRASH" \ 178 "\36INTERNAL" \ 179 "\35BUCKET" \ 180 "\34RECLAIMING" \ 181 "\33CACHE" \ 182 "\32LIMIT" \ 183 "\31CTORDTOR" \ 184 "\30VTOSLAB" \ 185 "\27HASH" \ 186 "\26OFFPAGE" \ 187 "\22ROUNDROBIN" \ 188 "\21FIRSTTOUCH" \ 189 "\20PCPU" \ 190 "\17NODUMP" \ 191 "\16CACHESPREAD" \ 192 "\15MINBUCKET" \ 193 "\14MAXBUCKET" \ 194 "\13NOBUCKET" \ 195 "\12SECONDARY" \ 196 "\11NOTPAGE" \ 197 "\10VM" \ 198 "\7MTXCLASS" \ 199 "\6NOFREE" \ 200 "\5MALLOC" \ 201 "\4NOTOUCH" \ 202 "\2ZINIT" 203 204 /* 205 * Hash table for freed address -> slab translation. 206 * 207 * Only zones with memory not touchable by the allocator use the 208 * hash table. Otherwise slabs are found with vtoslab(). 209 */ 210 #define UMA_HASH_SIZE_INIT 32 211 212 #define UMA_HASH(h, s) ((((uintptr_t)s) >> UMA_SLAB_SHIFT) & (h)->uh_hashmask) 213 214 #define UMA_HASH_INSERT(h, s, mem) \ 215 LIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \ 216 (mem))], slab_tohashslab(s), uhs_hlink) 217 218 #define UMA_HASH_REMOVE(h, s) \ 219 LIST_REMOVE(slab_tohashslab(s), uhs_hlink) 220 221 LIST_HEAD(slabhashhead, uma_hash_slab); 222 223 struct uma_hash { 224 struct slabhashhead *uh_slab_hash; /* Hash table for slabs */ 225 u_int uh_hashsize; /* Current size of the hash table */ 226 u_int uh_hashmask; /* Mask used during hashing */ 227 }; 228 229 /* 230 * Align field or structure to cache 'sector' in intel terminology. This 231 * is more efficient with adjacent line prefetch. 232 */ 233 #if defined(__amd64__) || defined(__powerpc64__) 234 #define UMA_SUPER_ALIGN (CACHE_LINE_SIZE * 2) 235 #else 236 #define UMA_SUPER_ALIGN CACHE_LINE_SIZE 237 #endif 238 239 #define UMA_ALIGN __aligned(UMA_SUPER_ALIGN) 240 241 /* 242 * The uma_bucket structure is used to queue and manage buckets divorced 243 * from per-cpu caches. They are loaded into uma_cache_bucket structures 244 * for use. 245 */ 246 struct uma_bucket { 247 TAILQ_ENTRY(uma_bucket) ub_link; /* Link into the zone */ 248 int16_t ub_cnt; /* Count of items in bucket. */ 249 int16_t ub_entries; /* Max items. */ 250 void *ub_bucket[]; /* actual allocation storage */ 251 }; 252 253 typedef struct uma_bucket * uma_bucket_t; 254 255 /* 256 * The uma_cache_bucket structure is statically allocated on each per-cpu 257 * cache. Its use reduces branches and cache misses in the fast path. 258 */ 259 struct uma_cache_bucket { 260 uma_bucket_t ucb_bucket; 261 int16_t ucb_cnt; 262 int16_t ucb_entries; 263 uint32_t ucb_spare; 264 }; 265 266 typedef struct uma_cache_bucket * uma_cache_bucket_t; 267 268 /* 269 * The uma_cache structure is allocated for each cpu for every zone 270 * type. This optimizes synchronization out of the allocator fast path. 271 */ 272 struct uma_cache { 273 struct uma_cache_bucket uc_freebucket; /* Bucket we're freeing to */ 274 struct uma_cache_bucket uc_allocbucket; /* Bucket to allocate from */ 275 struct uma_cache_bucket uc_crossbucket; /* cross domain bucket */ 276 uint64_t uc_allocs; /* Count of allocations */ 277 uint64_t uc_frees; /* Count of frees */ 278 } UMA_ALIGN; 279 280 typedef struct uma_cache * uma_cache_t; 281 282 LIST_HEAD(slabhead, uma_slab); 283 284 /* 285 * The cache structure pads perfectly into 64 bytes so we use spare 286 * bits from the embedded cache buckets to store information from the zone 287 * and keep all fast-path allocations accessing a single per-cpu line. 288 */ 289 static inline void 290 cache_set_uz_flags(uma_cache_t cache, uint32_t flags) 291 { 292 293 cache->uc_freebucket.ucb_spare = flags; 294 } 295 296 static inline void 297 cache_set_uz_size(uma_cache_t cache, uint32_t size) 298 { 299 300 cache->uc_allocbucket.ucb_spare = size; 301 } 302 303 static inline uint32_t 304 cache_uz_flags(uma_cache_t cache) 305 { 306 307 return (cache->uc_freebucket.ucb_spare); 308 } 309 310 static inline uint32_t 311 cache_uz_size(uma_cache_t cache) 312 { 313 314 return (cache->uc_allocbucket.ucb_spare); 315 } 316 317 /* 318 * Per-domain slab lists. Embedded in the kegs. 319 */ 320 struct uma_domain { 321 struct mtx_padalign ud_lock; /* Lock for the domain lists. */ 322 struct slabhead ud_part_slab; /* partially allocated slabs */ 323 struct slabhead ud_free_slab; /* completely unallocated slabs */ 324 struct slabhead ud_full_slab; /* fully allocated slabs */ 325 uint32_t ud_pages; /* Total page count */ 326 uint32_t ud_free; /* Count of items free in slabs */ 327 } __aligned(CACHE_LINE_SIZE); 328 329 typedef struct uma_domain * uma_domain_t; 330 331 /* 332 * Keg management structure 333 * 334 * TODO: Optimize for cache line size 335 * 336 */ 337 struct uma_keg { 338 struct uma_hash uk_hash; 339 LIST_HEAD(,uma_zone) uk_zones; /* Keg's zones */ 340 341 struct domainset_ref uk_dr; /* Domain selection policy. */ 342 uint32_t uk_align; /* Alignment mask */ 343 uint32_t uk_reserve; /* Number of reserved items. */ 344 uint32_t uk_size; /* Requested size of each item */ 345 uint32_t uk_rsize; /* Real size of each item */ 346 347 uma_init uk_init; /* Keg's init routine */ 348 uma_fini uk_fini; /* Keg's fini routine */ 349 uma_alloc uk_allocf; /* Allocation function */ 350 uma_free uk_freef; /* Free routine */ 351 352 u_long uk_offset; /* Next free offset from base KVA */ 353 vm_offset_t uk_kva; /* Zone base KVA */ 354 355 uint32_t uk_pgoff; /* Offset to uma_slab struct */ 356 uint16_t uk_ppera; /* pages per allocation from backend */ 357 uint16_t uk_ipers; /* Items per slab */ 358 uint32_t uk_flags; /* Internal flags */ 359 360 /* Least used fields go to the last cache line. */ 361 const char *uk_name; /* Name of creating zone. */ 362 LIST_ENTRY(uma_keg) uk_link; /* List of all kegs */ 363 364 /* Must be last, variable sized. */ 365 struct uma_domain uk_domain[]; /* Keg's slab lists. */ 366 }; 367 typedef struct uma_keg * uma_keg_t; 368 369 #ifdef _KERNEL 370 #define KEG_ASSERT_COLD(k) \ 371 KASSERT(uma_keg_get_allocs((k)) == 0, \ 372 ("keg %s initialization after use.", (k)->uk_name)) 373 374 /* 375 * Free bits per-slab. 376 */ 377 #define SLAB_MAX_SETSIZE (PAGE_SIZE / UMA_SMALLEST_UNIT) 378 #define SLAB_MIN_SETSIZE _BITSET_BITS 379 BITSET_DEFINE(noslabbits, 0); 380 381 /* 382 * The slab structure manages a single contiguous allocation from backing 383 * store and subdivides it into individually allocatable items. 384 */ 385 struct uma_slab { 386 LIST_ENTRY(uma_slab) us_link; /* slabs in zone */ 387 uint16_t us_freecount; /* How many are free? */ 388 uint8_t us_flags; /* Page flags see uma.h */ 389 uint8_t us_domain; /* Backing NUMA domain. */ 390 struct noslabbits us_free; /* Free bitmask, flexible. */ 391 }; 392 _Static_assert(sizeof(struct uma_slab) == offsetof(struct uma_slab, us_free), 393 "us_free field must be last"); 394 #if MAXMEMDOM >= 255 395 #error "Slab domain type insufficient" 396 #endif 397 398 typedef struct uma_slab * uma_slab_t; 399 400 /* 401 * On INVARIANTS builds, the slab contains a second bitset of the same size, 402 * "dbg_bits", which is laid out immediately after us_free. 403 */ 404 #ifdef INVARIANTS 405 #define SLAB_BITSETS 2 406 #else 407 #define SLAB_BITSETS 1 408 #endif 409 410 /* These three functions are for embedded (!OFFPAGE) use only. */ 411 size_t slab_sizeof(int nitems); 412 size_t slab_space(int nitems); 413 int slab_ipers(size_t size, int align); 414 415 /* 416 * Slab structure with a full sized bitset and hash link for both 417 * HASH and OFFPAGE zones. 418 */ 419 struct uma_hash_slab { 420 LIST_ENTRY(uma_hash_slab) uhs_hlink; /* Link for hash table */ 421 uint8_t *uhs_data; /* First item */ 422 struct uma_slab uhs_slab; /* Must be last. */ 423 }; 424 425 typedef struct uma_hash_slab * uma_hash_slab_t; 426 427 static inline uma_hash_slab_t 428 slab_tohashslab(uma_slab_t slab) 429 { 430 431 return (__containerof(slab, struct uma_hash_slab, uhs_slab)); 432 } 433 434 static inline void * 435 slab_data(uma_slab_t slab, uma_keg_t keg) 436 { 437 438 if ((keg->uk_flags & UMA_ZFLAG_OFFPAGE) == 0) 439 return ((void *)((uintptr_t)slab - keg->uk_pgoff)); 440 else 441 return (slab_tohashslab(slab)->uhs_data); 442 } 443 444 static inline void * 445 slab_item(uma_slab_t slab, uma_keg_t keg, int index) 446 { 447 uintptr_t data; 448 449 data = (uintptr_t)slab_data(slab, keg); 450 return ((void *)(data + keg->uk_rsize * index)); 451 } 452 453 static inline int 454 slab_item_index(uma_slab_t slab, uma_keg_t keg, void *item) 455 { 456 uintptr_t data; 457 458 data = (uintptr_t)slab_data(slab, keg); 459 return (((uintptr_t)item - data) / keg->uk_rsize); 460 } 461 #endif /* _KERNEL */ 462 463 TAILQ_HEAD(uma_bucketlist, uma_bucket); 464 465 struct uma_zone_domain { 466 struct uma_bucketlist uzd_buckets; /* full buckets */ 467 uma_bucket_t uzd_cross; /* Fills from cross buckets. */ 468 long uzd_nitems; /* total item count */ 469 long uzd_imax; /* maximum item count this period */ 470 long uzd_imin; /* minimum item count this period */ 471 long uzd_wss; /* working set size estimate */ 472 } __aligned(CACHE_LINE_SIZE); 473 474 typedef struct uma_zone_domain * uma_zone_domain_t; 475 476 /* 477 * Zone structure - per memory type. 478 */ 479 struct uma_zone { 480 /* Offset 0, used in alloc/free fast/medium fast path and const. */ 481 uma_keg_t uz_keg; /* This zone's keg if !CACHE */ 482 struct uma_zone_domain *uz_domain; /* per-domain buckets */ 483 uint32_t uz_flags; /* Flags inherited from kegs */ 484 uint32_t uz_size; /* Size inherited from kegs */ 485 uma_ctor uz_ctor; /* Constructor for each allocation */ 486 uma_dtor uz_dtor; /* Destructor */ 487 uint64_t uz_spare0; 488 uint64_t uz_max_items; /* Maximum number of items to alloc */ 489 uint32_t uz_sleepers; /* Threads sleeping on limit */ 490 uint16_t uz_bucket_size; /* Number of items in full bucket */ 491 uint16_t uz_bucket_size_max; /* Maximum number of bucket items */ 492 493 /* Offset 64, used in bucket replenish. */ 494 uma_import uz_import; /* Import new memory to cache. */ 495 uma_release uz_release; /* Release memory from cache. */ 496 void *uz_arg; /* Import/release argument. */ 497 uma_init uz_init; /* Initializer for each item */ 498 uma_fini uz_fini; /* Finalizer for each item. */ 499 void *uz_spare1; 500 uint64_t uz_bkt_count; /* Items in bucket cache */ 501 uint64_t uz_bkt_max; /* Maximum bucket cache size */ 502 503 /* Offset 128 Rare. */ 504 /* 505 * The lock is placed here to avoid adjacent line prefetcher 506 * in fast paths and to take up space near infrequently accessed 507 * members to reduce alignment overhead. 508 */ 509 struct mtx uz_lock; /* Lock for the zone */ 510 LIST_ENTRY(uma_zone) uz_link; /* List of all zones in keg */ 511 const char *uz_name; /* Text name of the zone */ 512 /* The next two fields are used to print a rate-limited warnings. */ 513 const char *uz_warning; /* Warning to print on failure */ 514 struct timeval uz_ratecheck; /* Warnings rate-limiting */ 515 struct task uz_maxaction; /* Task to run when at limit */ 516 uint16_t uz_bucket_size_min; /* Min number of items in bucket */ 517 518 struct mtx_padalign uz_cross_lock; /* Cross domain free lock */ 519 520 /* Offset 256+, stats and misc. */ 521 counter_u64_t uz_allocs; /* Total number of allocations */ 522 counter_u64_t uz_frees; /* Total number of frees */ 523 counter_u64_t uz_fails; /* Total number of alloc failures */ 524 uint64_t uz_sleeps; /* Total number of alloc sleeps */ 525 uint64_t uz_xdomain; /* Total number of cross-domain frees */ 526 volatile uint64_t uz_items; /* Total items count & sleepers */ 527 528 char *uz_ctlname; /* sysctl safe name string. */ 529 struct sysctl_oid *uz_oid; /* sysctl oid pointer. */ 530 int uz_namecnt; /* duplicate name count. */ 531 532 /* 533 * This HAS to be the last item because we adjust the zone size 534 * based on NCPU and then allocate the space for the zones. 535 */ 536 struct uma_cache uz_cpu[]; /* Per cpu caches */ 537 538 /* uz_domain follows here. */ 539 }; 540 541 /* 542 * Macros for interpreting the uz_items field. 20 bits of sleeper count 543 * and 44 bit of item count. 544 */ 545 #define UZ_ITEMS_SLEEPER_SHIFT 44LL 546 #define UZ_ITEMS_SLEEPERS_MAX ((1 << (64 - UZ_ITEMS_SLEEPER_SHIFT)) - 1) 547 #define UZ_ITEMS_COUNT_MASK ((1LL << UZ_ITEMS_SLEEPER_SHIFT) - 1) 548 #define UZ_ITEMS_COUNT(x) ((x) & UZ_ITEMS_COUNT_MASK) 549 #define UZ_ITEMS_SLEEPERS(x) ((x) >> UZ_ITEMS_SLEEPER_SHIFT) 550 #define UZ_ITEMS_SLEEPER (1LL << UZ_ITEMS_SLEEPER_SHIFT) 551 552 #define ZONE_ASSERT_COLD(z) \ 553 KASSERT(uma_zone_get_allocs((z)) == 0, \ 554 ("zone %s initialization after use.", (z)->uz_name)) 555 556 #undef UMA_ALIGN 557 558 #ifdef _KERNEL 559 /* Internal prototypes */ 560 static __inline uma_slab_t hash_sfind(struct uma_hash *hash, uint8_t *data); 561 562 /* Lock Macros */ 563 564 #define KEG_LOCKPTR(k, d) (struct mtx *)&(k)->uk_domain[(d)].ud_lock 565 #define KEG_LOCK_INIT(k, d, lc) \ 566 do { \ 567 if ((lc)) \ 568 mtx_init(KEG_LOCKPTR(k, d), (k)->uk_name, \ 569 (k)->uk_name, MTX_DEF | MTX_DUPOK); \ 570 else \ 571 mtx_init(KEG_LOCKPTR(k, d), (k)->uk_name, \ 572 "UMA zone", MTX_DEF | MTX_DUPOK); \ 573 } while (0) 574 575 #define KEG_LOCK_FINI(k, d) mtx_destroy(KEG_LOCKPTR(k, d)) 576 #define KEG_LOCK(k, d) \ 577 ({ mtx_lock(KEG_LOCKPTR(k, d)); KEG_LOCKPTR(k, d); }) 578 #define KEG_UNLOCK(k, d) mtx_unlock(KEG_LOCKPTR(k, d)) 579 #define KEG_LOCK_ASSERT(k, d) mtx_assert(KEG_LOCKPTR(k, d), MA_OWNED) 580 581 #define KEG_GET(zone, keg) do { \ 582 (keg) = (zone)->uz_keg; \ 583 KASSERT((void *)(keg) != (void *)&(zone)->uz_lock, \ 584 ("%s: Invalid zone %p type", __func__, (zone))); \ 585 } while (0) 586 587 #define ZONE_LOCK_INIT(z, lc) \ 588 do { \ 589 if ((lc)) \ 590 mtx_init(&(z)->uz_lock, (z)->uz_name, \ 591 (z)->uz_name, MTX_DEF | MTX_DUPOK); \ 592 else \ 593 mtx_init(&(z)->uz_lock, (z)->uz_name, \ 594 "UMA zone", MTX_DEF | MTX_DUPOK); \ 595 } while (0) 596 597 #define ZONE_LOCK(z) mtx_lock(&(z)->uz_lock) 598 #define ZONE_TRYLOCK(z) mtx_trylock(&(z)->uz_lock) 599 #define ZONE_UNLOCK(z) mtx_unlock(&(z)->uz_lock) 600 #define ZONE_LOCK_FINI(z) mtx_destroy(&(z)->uz_lock) 601 #define ZONE_LOCK_ASSERT(z) mtx_assert(&(z)->uz_lock, MA_OWNED) 602 603 #define ZONE_CROSS_LOCK_INIT(z) \ 604 mtx_init(&(z)->uz_cross_lock, "UMA Cross", NULL, MTX_DEF) 605 #define ZONE_CROSS_LOCK(z) mtx_lock(&(z)->uz_cross_lock) 606 #define ZONE_CROSS_UNLOCK(z) mtx_unlock(&(z)->uz_cross_lock) 607 #define ZONE_CROSS_LOCK_FINI(z) mtx_destroy(&(z)->uz_cross_lock) 608 609 /* 610 * Find a slab within a hash table. This is used for OFFPAGE zones to lookup 611 * the slab structure. 612 * 613 * Arguments: 614 * hash The hash table to search. 615 * data The base page of the item. 616 * 617 * Returns: 618 * A pointer to a slab if successful, else NULL. 619 */ 620 static __inline uma_slab_t 621 hash_sfind(struct uma_hash *hash, uint8_t *data) 622 { 623 uma_hash_slab_t slab; 624 u_int hval; 625 626 hval = UMA_HASH(hash, data); 627 628 LIST_FOREACH(slab, &hash->uh_slab_hash[hval], uhs_hlink) { 629 if ((uint8_t *)slab->uhs_data == data) 630 return (&slab->uhs_slab); 631 } 632 return (NULL); 633 } 634 635 static __inline uma_slab_t 636 vtoslab(vm_offset_t va) 637 { 638 vm_page_t p; 639 640 p = PHYS_TO_VM_PAGE(pmap_kextract(va)); 641 return (p->plinks.uma.slab); 642 } 643 644 static __inline void 645 vtozoneslab(vm_offset_t va, uma_zone_t *zone, uma_slab_t *slab) 646 { 647 vm_page_t p; 648 649 p = PHYS_TO_VM_PAGE(pmap_kextract(va)); 650 *slab = p->plinks.uma.slab; 651 *zone = p->plinks.uma.zone; 652 } 653 654 static __inline void 655 vsetzoneslab(vm_offset_t va, uma_zone_t zone, uma_slab_t slab) 656 { 657 vm_page_t p; 658 659 p = PHYS_TO_VM_PAGE(pmap_kextract(va)); 660 p->plinks.uma.slab = slab; 661 p->plinks.uma.zone = zone; 662 } 663 664 extern unsigned long uma_kmem_limit; 665 extern unsigned long uma_kmem_total; 666 667 /* Adjust bytes under management by UMA. */ 668 static inline void 669 uma_total_dec(unsigned long size) 670 { 671 672 atomic_subtract_long(&uma_kmem_total, size); 673 } 674 675 static inline void 676 uma_total_inc(unsigned long size) 677 { 678 679 if (atomic_fetchadd_long(&uma_kmem_total, size) > uma_kmem_limit) 680 uma_reclaim_wakeup(); 681 } 682 683 /* 684 * The following two functions may be defined by architecture specific code 685 * if they can provide more efficient allocation functions. This is useful 686 * for using direct mapped addresses. 687 */ 688 void *uma_small_alloc(uma_zone_t zone, vm_size_t bytes, int domain, 689 uint8_t *pflag, int wait); 690 void uma_small_free(void *mem, vm_size_t size, uint8_t flags); 691 692 /* Set a global soft limit on UMA managed memory. */ 693 void uma_set_limit(unsigned long limit); 694 #endif /* _KERNEL */ 695 696 #endif /* VM_UMA_INT_H */ 697