1 /*- 2 * Copyright (c) 2002, 2003, 2004, 2005 Jeffrey Roberson <jeff@FreeBSD.org> 3 * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice unmodified, this list of conditions, and the following 11 * disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 * 27 * $FreeBSD$ 28 * 29 */ 30 31 /* 32 * uma.h - External definitions for the Universal Memory Allocator 33 * 34 */ 35 36 #ifndef VM_UMA_H 37 #define VM_UMA_H 38 39 #include <sys/param.h> /* For NULL */ 40 #include <sys/malloc.h> /* For M_* */ 41 42 /* User visible parameters */ 43 #define UMA_SMALLEST_UNIT (PAGE_SIZE / 256) /* Smallest item allocated */ 44 45 /* Types and type defs */ 46 47 struct uma_zone; 48 /* Opaque type used as a handle to the zone */ 49 typedef struct uma_zone * uma_zone_t; 50 51 void zone_drain(uma_zone_t); 52 53 /* 54 * Item constructor 55 * 56 * Arguments: 57 * item A pointer to the memory which has been allocated. 58 * arg The arg field passed to uma_zalloc_arg 59 * size The size of the allocated item 60 * flags See zalloc flags 61 * 62 * Returns: 63 * 0 on success 64 * errno on failure 65 * 66 * Discussion: 67 * The constructor is called just before the memory is returned 68 * to the user. It may block if necessary. 69 */ 70 typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags); 71 72 /* 73 * Item destructor 74 * 75 * Arguments: 76 * item A pointer to the memory which has been allocated. 77 * size The size of the item being destructed. 78 * arg Argument passed through uma_zfree_arg 79 * 80 * Returns: 81 * Nothing 82 * 83 * Discussion: 84 * The destructor may perform operations that differ from those performed 85 * by the initializer, but it must leave the object in the same state. 86 * This IS type stable storage. This is called after EVERY zfree call. 87 */ 88 typedef void (*uma_dtor)(void *mem, int size, void *arg); 89 90 /* 91 * Item initializer 92 * 93 * Arguments: 94 * item A pointer to the memory which has been allocated. 95 * size The size of the item being initialized. 96 * flags See zalloc flags 97 * 98 * Returns: 99 * 0 on success 100 * errno on failure 101 * 102 * Discussion: 103 * The initializer is called when the memory is cached in the uma zone. 104 * The initializer and the destructor should leave the object in the same 105 * state. 106 */ 107 typedef int (*uma_init)(void *mem, int size, int flags); 108 109 /* 110 * Item discard function 111 * 112 * Arguments: 113 * item A pointer to memory which has been 'freed' but has not left the 114 * zone's cache. 115 * size The size of the item being discarded. 116 * 117 * Returns: 118 * Nothing 119 * 120 * Discussion: 121 * This routine is called when memory leaves a zone and is returned to the 122 * system for other uses. It is the counter-part to the init function. 123 */ 124 typedef void (*uma_fini)(void *mem, int size); 125 126 /* 127 * What's the difference between initializing and constructing? 128 * 129 * The item is initialized when it is cached, and this is the state that the 130 * object should be in when returned to the allocator. The purpose of this is 131 * to remove some code which would otherwise be called on each allocation by 132 * utilizing a known, stable state. This differs from the constructor which 133 * will be called on EVERY allocation. 134 * 135 * For example, in the initializer you may want to initialize embedded locks, 136 * NULL list pointers, set up initial states, magic numbers, etc. This way if 137 * the object is held in the allocator and re-used it won't be necessary to 138 * re-initialize it. 139 * 140 * The constructor may be used to lock a data structure, link it on to lists, 141 * bump reference counts or total counts of outstanding structures, etc. 142 * 143 */ 144 145 146 /* Function proto types */ 147 148 /* 149 * Create a new uma zone 150 * 151 * Arguments: 152 * name The text name of the zone for debugging and stats. This memory 153 * should not be freed until the zone has been deallocated. 154 * size The size of the object that is being created. 155 * ctor The constructor that is called when the object is allocated. 156 * dtor The destructor that is called when the object is freed. 157 * init An initializer that sets up the initial state of the memory. 158 * fini A discard function that undoes initialization done by init. 159 * ctor/dtor/init/fini may all be null, see notes above. 160 * align A bitmask that corresponds to the requested alignment 161 * eg 4 would be 0x3 162 * flags A set of parameters that control the behavior of the zone. 163 * 164 * Returns: 165 * A pointer to a structure which is intended to be opaque to users of 166 * the interface. The value may be null if the wait flag is not set. 167 */ 168 uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor, 169 uma_dtor dtor, uma_init uminit, uma_fini fini, 170 int align, uint32_t flags); 171 172 /* 173 * Create a secondary uma zone 174 * 175 * Arguments: 176 * name The text name of the zone for debugging and stats. This memory 177 * should not be freed until the zone has been deallocated. 178 * ctor The constructor that is called when the object is allocated. 179 * dtor The destructor that is called when the object is freed. 180 * zinit An initializer that sets up the initial state of the memory 181 * as the object passes from the Keg's slab to the Zone's cache. 182 * zfini A discard function that undoes initialization done by init 183 * as the object passes from the Zone's cache to the Keg's slab. 184 * 185 * ctor/dtor/zinit/zfini may all be null, see notes above. 186 * Note that the zinit and zfini specified here are NOT 187 * exactly the same as the init/fini specified to uma_zcreate() 188 * when creating a master zone. These zinit/zfini are called 189 * on the TRANSITION from keg to zone (and vice-versa). Once 190 * these are set, the primary zone may alter its init/fini 191 * (which are called when the object passes from VM to keg) 192 * using uma_zone_set_init/fini()) as well as its own 193 * zinit/zfini (unset by default for master zone) with 194 * uma_zone_set_zinit/zfini() (note subtle 'z' prefix). 195 * 196 * master A reference to this zone's Master Zone (Primary Zone), 197 * which contains the backing Keg for the Secondary Zone 198 * being added. 199 * 200 * Returns: 201 * A pointer to a structure which is intended to be opaque to users of 202 * the interface. The value may be null if the wait flag is not set. 203 */ 204 uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor, 205 uma_init zinit, uma_fini zfini, uma_zone_t master); 206 207 /* 208 * Add a second master to a secondary zone. This provides multiple data 209 * backends for objects with the same size. Both masters must have 210 * compatible allocation flags. Presently, UMA_ZONE_MALLOC type zones are 211 * the only supported. 212 * 213 * Returns: 214 * Error on failure, 0 on success. 215 */ 216 int uma_zsecond_add(uma_zone_t zone, uma_zone_t master); 217 218 /* 219 * Definitions for uma_zcreate flags 220 * 221 * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to 222 * overlap when adding new features. 0xf0000000 is in use by uma_int.h. 223 */ 224 #define UMA_ZONE_PAGEABLE 0x0001 /* Return items not fully backed by 225 physical memory XXX Not yet */ 226 #define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */ 227 #define UMA_ZONE_STATIC 0x0004 /* Statically sized zone */ 228 #define UMA_ZONE_OFFPAGE 0x0008 /* Force the slab structure allocation 229 off of the real memory */ 230 #define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */ 231 #define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */ 232 #define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */ 233 #define UMA_ZONE_VM 0x0080 /* 234 * Used for internal vm datastructures 235 * only. 236 */ 237 #define UMA_ZONE_HASH 0x0100 /* 238 * Use a hash table instead of caching 239 * information in the vm_page. 240 */ 241 #define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */ 242 #define UMA_ZONE_REFCNT 0x0400 /* Allocate refcnts in slabs */ 243 #define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets */ 244 #define UMA_ZONE_CACHESPREAD 0x1000 /* 245 * Spread memory start locations across 246 * all possible cache lines. May 247 * require many virtually contiguous 248 * backend pages and can fail early. 249 */ 250 #define UMA_ZONE_VTOSLAB 0x2000 /* Zone uses vtoslab for lookup. */ 251 #define UMA_ZONE_NODUMP 0x4000 /* 252 * Zone's pages will not be included in 253 * mini-dumps. 254 */ 255 #define UMA_ZONE_PCPU 0x8000 /* 256 * Allocates mp_ncpus slabs sized to 257 * sizeof(struct pcpu). 258 */ 259 260 /* 261 * These flags are shared between the keg and zone. In zones wishing to add 262 * new kegs these flags must be compatible. Some are determined based on 263 * physical parameters of the request and may not be provided by the consumer. 264 */ 265 #define UMA_ZONE_INHERIT \ 266 (UMA_ZONE_OFFPAGE | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \ 267 UMA_ZONE_HASH | UMA_ZONE_REFCNT | UMA_ZONE_VTOSLAB | UMA_ZONE_PCPU) 268 269 /* Definitions for align */ 270 #define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */ 271 #define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */ 272 #define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */ 273 #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */ 274 #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */ 275 #define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */ 276 277 /* 278 * Destroys an empty uma zone. If the zone is not empty uma complains loudly. 279 * 280 * Arguments: 281 * zone The zone we want to destroy. 282 * 283 */ 284 void uma_zdestroy(uma_zone_t zone); 285 286 /* 287 * Allocates an item out of a zone 288 * 289 * Arguments: 290 * zone The zone we are allocating from 291 * arg This data is passed to the ctor function 292 * flags See sys/malloc.h for available flags. 293 * 294 * Returns: 295 * A non-null pointer to an initialized element from the zone is 296 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer 297 * may be returned if the zone is empty or the ctor failed. 298 */ 299 300 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags); 301 302 /* 303 * Allocates an item out of a zone without supplying an argument 304 * 305 * This is just a wrapper for uma_zalloc_arg for convenience. 306 * 307 */ 308 static __inline void *uma_zalloc(uma_zone_t zone, int flags); 309 310 static __inline void * 311 uma_zalloc(uma_zone_t zone, int flags) 312 { 313 return uma_zalloc_arg(zone, NULL, flags); 314 } 315 316 /* 317 * Frees an item back into the specified zone. 318 * 319 * Arguments: 320 * zone The zone the item was originally allocated out of. 321 * item The memory to be freed. 322 * arg Argument passed to the destructor 323 * 324 * Returns: 325 * Nothing. 326 */ 327 328 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg); 329 330 /* 331 * Frees an item back to a zone without supplying an argument 332 * 333 * This is just a wrapper for uma_zfree_arg for convenience. 334 * 335 */ 336 static __inline void uma_zfree(uma_zone_t zone, void *item); 337 338 static __inline void 339 uma_zfree(uma_zone_t zone, void *item) 340 { 341 uma_zfree_arg(zone, item, NULL); 342 } 343 344 /* 345 * XXX The rest of the prototypes in this header are h0h0 magic for the VM. 346 * If you think you need to use it for a normal zone you're probably incorrect. 347 */ 348 349 /* 350 * Backend page supplier routines 351 * 352 * Arguments: 353 * zone The zone that is requesting pages. 354 * size The number of bytes being requested. 355 * pflag Flags for these memory pages, see below. 356 * wait Indicates our willingness to block. 357 * 358 * Returns: 359 * A pointer to the allocated memory or NULL on failure. 360 */ 361 362 typedef void *(*uma_alloc)(uma_zone_t zone, int size, uint8_t *pflag, int wait); 363 364 /* 365 * Backend page free routines 366 * 367 * Arguments: 368 * item A pointer to the previously allocated pages. 369 * size The original size of the allocation. 370 * pflag The flags for the slab. See UMA_SLAB_* below. 371 * 372 * Returns: 373 * None 374 */ 375 typedef void (*uma_free)(void *item, int size, uint8_t pflag); 376 377 378 379 /* 380 * Sets up the uma allocator. (Called by vm_mem_init) 381 * 382 * Arguments: 383 * bootmem A pointer to memory used to bootstrap the system. 384 * 385 * Returns: 386 * Nothing 387 * 388 * Discussion: 389 * This memory is used for zones which allocate things before the 390 * backend page supplier can give us pages. It should be 391 * UMA_SLAB_SIZE * boot_pages bytes. (see uma_int.h) 392 * 393 */ 394 395 void uma_startup(void *bootmem, int boot_pages); 396 397 /* 398 * Finishes starting up the allocator. This should 399 * be called when kva is ready for normal allocs. 400 * 401 * Arguments: 402 * None 403 * 404 * Returns: 405 * Nothing 406 * 407 * Discussion: 408 * uma_startup2 is called by kmeminit() to enable us of uma for malloc. 409 */ 410 411 void uma_startup2(void); 412 413 /* 414 * Reclaims unused memory for all zones 415 * 416 * Arguments: 417 * None 418 * Returns: 419 * None 420 * 421 * This should only be called by the page out daemon. 422 */ 423 424 void uma_reclaim(void); 425 426 /* 427 * Sets the alignment mask to be used for all zones requesting cache 428 * alignment. Should be called by MD boot code prior to starting VM/UMA. 429 * 430 * Arguments: 431 * align The alignment mask 432 * 433 * Returns: 434 * Nothing 435 */ 436 void uma_set_align(int align); 437 438 /* 439 * Reserves the maximum KVA space required by the zone and configures the zone 440 * to use a VM_ALLOC_NOOBJ-based backend allocator. 441 * 442 * Arguments: 443 * zone The zone to update. 444 * nitems The upper limit on the number of items that can be allocated. 445 * 446 * Returns: 447 * 0 if KVA space can not be allocated 448 * 1 if successful 449 * 450 * Discussion: 451 * When the machine supports a direct map and the zone's items are smaller 452 * than a page, the zone will use the direct map instead of allocating KVA 453 * space. 454 */ 455 int uma_zone_reserve_kva(uma_zone_t zone, int nitems); 456 457 /* 458 * Sets a high limit on the number of items allowed in a zone 459 * 460 * Arguments: 461 * zone The zone to limit 462 * nitems The requested upper limit on the number of items allowed 463 * 464 * Returns: 465 * int The effective value of nitems after rounding up based on page size 466 */ 467 int uma_zone_set_max(uma_zone_t zone, int nitems); 468 469 /* 470 * Obtains the effective limit on the number of items in a zone 471 * 472 * Arguments: 473 * zone The zone to obtain the effective limit from 474 * 475 * Return: 476 * 0 No limit 477 * int The effective limit of the zone 478 */ 479 int uma_zone_get_max(uma_zone_t zone); 480 481 /* 482 * Sets a warning to be printed when limit is reached 483 * 484 * Arguments: 485 * zone The zone we will warn about 486 * warning Warning content 487 * 488 * Returns: 489 * Nothing 490 */ 491 void uma_zone_set_warning(uma_zone_t zone, const char *warning); 492 493 /* 494 * Obtains the approximate current number of items allocated from a zone 495 * 496 * Arguments: 497 * zone The zone to obtain the current allocation count from 498 * 499 * Return: 500 * int The approximate current number of items allocated from the zone 501 */ 502 int uma_zone_get_cur(uma_zone_t zone); 503 504 /* 505 * The following two routines (uma_zone_set_init/fini) 506 * are used to set the backend init/fini pair which acts on an 507 * object as it becomes allocated and is placed in a slab within 508 * the specified zone's backing keg. These should probably not 509 * be changed once allocations have already begun, but only be set 510 * immediately upon zone creation. 511 */ 512 void uma_zone_set_init(uma_zone_t zone, uma_init uminit); 513 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini); 514 515 /* 516 * The following two routines (uma_zone_set_zinit/zfini) are 517 * used to set the zinit/zfini pair which acts on an object as 518 * it passes from the backing Keg's slab cache to the 519 * specified Zone's bucket cache. These should probably not 520 * be changed once allocations have already begun, but only be set 521 * immediately upon zone creation. 522 */ 523 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit); 524 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini); 525 526 /* 527 * Replaces the standard backend allocator for this zone. 528 * 529 * Arguments: 530 * zone The zone whose backend allocator is being changed. 531 * allocf A pointer to the allocation function 532 * 533 * Returns: 534 * Nothing 535 * 536 * Discussion: 537 * This could be used to implement pageable allocation, or perhaps 538 * even DMA allocators if used in conjunction with the OFFPAGE 539 * zone flag. 540 */ 541 542 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf); 543 544 /* 545 * Used for freeing memory provided by the allocf above 546 * 547 * Arguments: 548 * zone The zone that intends to use this free routine. 549 * freef The page freeing routine. 550 * 551 * Returns: 552 * Nothing 553 */ 554 555 void uma_zone_set_freef(uma_zone_t zone, uma_free freef); 556 557 /* 558 * These flags are setable in the allocf and visible in the freef. 559 */ 560 #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */ 561 #define UMA_SLAB_KMEM 0x02 /* Slab alloced from kmem_map */ 562 #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kernel_map */ 563 #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */ 564 #define UMA_SLAB_OFFP 0x10 /* Slab is managed separately */ 565 #define UMA_SLAB_MALLOC 0x20 /* Slab is a large malloc slab */ 566 /* 0x40 and 0x80 are available */ 567 568 /* 569 * Used to pre-fill a zone with some number of items 570 * 571 * Arguments: 572 * zone The zone to fill 573 * itemcnt The number of items to reserve 574 * 575 * Returns: 576 * Nothing 577 * 578 * NOTE: This is blocking and should only be done at startup 579 */ 580 void uma_prealloc(uma_zone_t zone, int itemcnt); 581 582 /* 583 * Used to lookup the reference counter allocated for an item 584 * from a UMA_ZONE_REFCNT zone. For UMA_ZONE_REFCNT zones, 585 * reference counters are allocated for items and stored in 586 * the underlying slab header. 587 * 588 * Arguments: 589 * zone The UMA_ZONE_REFCNT zone to which the item belongs. 590 * item The address of the item for which we want a refcnt. 591 * 592 * Returns: 593 * A pointer to a uint32_t reference counter. 594 */ 595 uint32_t *uma_find_refcnt(uma_zone_t zone, void *item); 596 597 /* 598 * Used to determine if a fixed-size zone is exhausted. 599 * 600 * Arguments: 601 * zone The zone to check 602 * 603 * Returns: 604 * Non-zero if zone is exhausted. 605 */ 606 int uma_zone_exhausted(uma_zone_t zone); 607 int uma_zone_exhausted_nolock(uma_zone_t zone); 608 609 /* 610 * Exported statistics structures to be used by user space monitoring tools. 611 * Statistics stream consists of a uma_stream_header, followed by a series of 612 * alternative uma_type_header and uma_type_stat structures. 613 */ 614 #define UMA_STREAM_VERSION 0x00000001 615 struct uma_stream_header { 616 uint32_t ush_version; /* Stream format version. */ 617 uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */ 618 uint32_t ush_count; /* Number of records. */ 619 uint32_t _ush_pad; /* Pad/reserved field. */ 620 }; 621 622 #define UTH_MAX_NAME 32 623 #define UTH_ZONE_SECONDARY 0x00000001 624 struct uma_type_header { 625 /* 626 * Static per-zone data, some extracted from the supporting keg. 627 */ 628 char uth_name[UTH_MAX_NAME]; 629 uint32_t uth_align; /* Keg: alignment. */ 630 uint32_t uth_size; /* Keg: requested size of item. */ 631 uint32_t uth_rsize; /* Keg: real size of item. */ 632 uint32_t uth_maxpages; /* Keg: maximum number of pages. */ 633 uint32_t uth_limit; /* Keg: max items to allocate. */ 634 635 /* 636 * Current dynamic zone/keg-derived statistics. 637 */ 638 uint32_t uth_pages; /* Keg: pages allocated. */ 639 uint32_t uth_keg_free; /* Keg: items free. */ 640 uint32_t uth_zone_free; /* Zone: items free. */ 641 uint32_t uth_bucketsize; /* Zone: desired bucket size. */ 642 uint32_t uth_zone_flags; /* Zone: flags. */ 643 uint64_t uth_allocs; /* Zone: number of allocations. */ 644 uint64_t uth_frees; /* Zone: number of frees. */ 645 uint64_t uth_fails; /* Zone: number of alloc failures. */ 646 uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */ 647 uint64_t _uth_reserved1[2]; /* Reserved. */ 648 }; 649 650 struct uma_percpu_stat { 651 uint64_t ups_allocs; /* Cache: number of allocations. */ 652 uint64_t ups_frees; /* Cache: number of frees. */ 653 uint64_t ups_cache_free; /* Cache: free items in cache. */ 654 uint64_t _ups_reserved[5]; /* Reserved. */ 655 }; 656 657 #endif 658