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 47struct uma_zone; 48/* Opaque type used as a handle to the zone */ 49typedef struct uma_zone * uma_zone_t; 50 51void 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 */ 70typedef 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 */ 88typedef 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 */ 107typedef 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 */ 124typedef 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 */ 168uma_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, u_int32_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 */ 204uma_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 */ 216int 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 */ 255 256/* 257 * These flags are shared between the keg and zone. In zones wishing to add 258 * new kegs these flags must be compatible. Some are determined based on 259 * physical parameters of the request and may not be provided by the consumer. 260 */ 261#define UMA_ZONE_INHERIT \ 262 (UMA_ZONE_OFFPAGE | UMA_ZONE_MALLOC | UMA_ZONE_HASH | \ 263 UMA_ZONE_REFCNT | UMA_ZONE_VTOSLAB) 264 265/* Definitions for align */ 266#define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */ 267#define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */ 268#define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */ 269#define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */ 270#define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */ 271#define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */ 272 273/* 274 * Destroys an empty uma zone. If the zone is not empty uma complains loudly. 275 * 276 * Arguments: 277 * zone The zone we want to destroy. 278 * 279 */ 280void uma_zdestroy(uma_zone_t zone); 281 282/* 283 * Allocates an item out of a zone 284 * 285 * Arguments: 286 * zone The zone we are allocating from 287 * arg This data is passed to the ctor function 288 * flags See sys/malloc.h for available flags. 289 * 290 * Returns: 291 * A non-null pointer to an initialized element from the zone is 292 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer 293 * may be returned if the zone is empty or the ctor failed. 294 */ 295 296void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags); 297 298/* 299 * Allocates an item out of a zone without supplying an argument 300 * 301 * This is just a wrapper for uma_zalloc_arg for convenience. 302 * 303 */ 304static __inline void *uma_zalloc(uma_zone_t zone, int flags); 305 306static __inline void * 307uma_zalloc(uma_zone_t zone, int flags) 308{ 309 return uma_zalloc_arg(zone, NULL, flags); 310} 311 312/* 313 * Frees an item back into the specified zone. 314 * 315 * Arguments: 316 * zone The zone the item was originally allocated out of. 317 * item The memory to be freed. 318 * arg Argument passed to the destructor 319 * 320 * Returns: 321 * Nothing. 322 */ 323 324void uma_zfree_arg(uma_zone_t zone, void *item, void *arg); 325 326/* 327 * Frees an item back to a zone without supplying an argument 328 * 329 * This is just a wrapper for uma_zfree_arg for convenience. 330 * 331 */ 332static __inline void uma_zfree(uma_zone_t zone, void *item); 333 334static __inline void 335uma_zfree(uma_zone_t zone, void *item) 336{ 337 uma_zfree_arg(zone, item, NULL); 338} 339 340/* 341 * XXX The rest of the prototypes in this header are h0h0 magic for the VM. 342 * If you think you need to use it for a normal zone you're probably incorrect. 343 */ 344 345/* 346 * Backend page supplier routines 347 * 348 * Arguments: 349 * zone The zone that is requesting pages. 350 * size The number of bytes being requested. 351 * pflag Flags for these memory pages, see below. 352 * wait Indicates our willingness to block. 353 * 354 * Returns: 355 * A pointer to the allocated memory or NULL on failure. 356 */ 357 358typedef void *(*uma_alloc)(uma_zone_t zone, int size, u_int8_t *pflag, int wait); 359 360/* 361 * Backend page free routines 362 * 363 * Arguments: 364 * item A pointer to the previously allocated pages. 365 * size The original size of the allocation. 366 * pflag The flags for the slab. See UMA_SLAB_* below. 367 * 368 * Returns: 369 * None 370 */ 371typedef void (*uma_free)(void *item, int size, u_int8_t pflag); 372 373 374 375/* 376 * Sets up the uma allocator. (Called by vm_mem_init) 377 * 378 * Arguments: 379 * bootmem A pointer to memory used to bootstrap the system. 380 * 381 * Returns: 382 * Nothing 383 * 384 * Discussion: 385 * This memory is used for zones which allocate things before the 386 * backend page supplier can give us pages. It should be 387 * UMA_SLAB_SIZE * boot_pages bytes. (see uma_int.h) 388 * 389 */ 390 391void uma_startup(void *bootmem, int boot_pages); 392 393/* 394 * Finishes starting up the allocator. This should 395 * be called when kva is ready for normal allocs. 396 * 397 * Arguments: 398 * None 399 * 400 * Returns: 401 * Nothing 402 * 403 * Discussion: 404 * uma_startup2 is called by kmeminit() to enable us of uma for malloc. 405 */ 406 407void uma_startup2(void); 408 409/* 410 * Reclaims unused memory for all zones 411 * 412 * Arguments: 413 * None 414 * Returns: 415 * None 416 * 417 * This should only be called by the page out daemon. 418 */ 419 420void uma_reclaim(void); 421 422/* 423 * Sets the alignment mask to be used for all zones requesting cache 424 * alignment. Should be called by MD boot code prior to starting VM/UMA. 425 * 426 * Arguments: 427 * align The alignment mask 428 * 429 * Returns: 430 * Nothing 431 */ 432void uma_set_align(int align); 433 434/* 435 * Switches the backing object of a zone 436 * 437 * Arguments: 438 * zone The zone to update. 439 * obj The VM object to use for future allocations. 440 * size The size of the object to allocate. 441 * 442 * Returns: 443 * 0 if kva space can not be allocated 444 * 1 if successful 445 * 446 * Discussion: 447 * A NULL object can be used and uma will allocate one for you. Setting 448 * the size will limit the amount of memory allocated to this zone. 449 * 450 */ 451struct vm_object; 452int uma_zone_set_obj(uma_zone_t zone, struct vm_object *obj, int size); 453 454/* 455 * Sets a high limit on the number of items allowed in a zone 456 * 457 * Arguments: 458 * zone The zone to limit 459 * nitems The requested upper limit on the number of items allowed 460 * 461 * Returns: 462 * int The effective value of nitems after rounding up based on page size 463 */ 464int uma_zone_set_max(uma_zone_t zone, int nitems); 465 466/* 467 * Obtains the effective limit on the number of items in a zone 468 * 469 * Arguments: 470 * zone The zone to obtain the effective limit from 471 * 472 * Return: 473 * 0 No limit 474 * int The effective limit of the zone 475 */ 476int uma_zone_get_max(uma_zone_t zone); 477 478/* 479 * Obtains the approximate current number of items allocated from a zone 480 * 481 * Arguments: 482 * zone The zone to obtain the current allocation count from 483 * 484 * Return: 485 * int The approximate current number of items allocated from the zone 486 */ 487int uma_zone_get_cur(uma_zone_t zone); 488 489/* 490 * The following two routines (uma_zone_set_init/fini) 491 * are used to set the backend init/fini pair which acts on an 492 * object as it becomes allocated and is placed in a slab within 493 * the specified zone's backing keg. These should probably not 494 * be changed once allocations have already begun, but only be set 495 * immediately upon zone creation. 496 */ 497void uma_zone_set_init(uma_zone_t zone, uma_init uminit); 498void uma_zone_set_fini(uma_zone_t zone, uma_fini fini); 499 500/* 501 * The following two routines (uma_zone_set_zinit/zfini) are 502 * used to set the zinit/zfini pair which acts on an object as 503 * it passes from the backing Keg's slab cache to the 504 * specified Zone's bucket cache. These should probably not 505 * be changed once allocations have already begun, but only be set 506 * immediately upon zone creation. 507 */ 508void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit); 509void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini); 510 511/* 512 * Replaces the standard page_alloc or obj_alloc functions for this zone 513 * 514 * Arguments: 515 * zone The zone whose backend allocator is being changed. 516 * allocf A pointer to the allocation function 517 * 518 * Returns: 519 * Nothing 520 * 521 * Discussion: 522 * This could be used to implement pageable allocation, or perhaps 523 * even DMA allocators if used in conjunction with the OFFPAGE 524 * zone flag. 525 */ 526 527void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf); 528 529/* 530 * Used for freeing memory provided by the allocf above 531 * 532 * Arguments: 533 * zone The zone that intends to use this free routine. 534 * freef The page freeing routine. 535 * 536 * Returns: 537 * Nothing 538 */ 539 540void uma_zone_set_freef(uma_zone_t zone, uma_free freef); 541 542/* 543 * These flags are setable in the allocf and visible in the freef. 544 */ 545#define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */ 546#define UMA_SLAB_KMEM 0x02 /* Slab alloced from kmem_map */ 547#define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kernel_map */ 548#define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */ 549#define UMA_SLAB_OFFP 0x10 /* Slab is managed separately */ 550#define UMA_SLAB_MALLOC 0x20 /* Slab is a large malloc slab */ 551/* 0x40 and 0x80 are available */ 552 553/* 554 * Used to pre-fill a zone with some number of items 555 * 556 * Arguments: 557 * zone The zone to fill 558 * itemcnt The number of items to reserve 559 * 560 * Returns: 561 * Nothing 562 * 563 * NOTE: This is blocking and should only be done at startup 564 */ 565void uma_prealloc(uma_zone_t zone, int itemcnt); 566 567/* 568 * Used to lookup the reference counter allocated for an item 569 * from a UMA_ZONE_REFCNT zone. For UMA_ZONE_REFCNT zones, 570 * reference counters are allocated for items and stored in 571 * the underlying slab header. 572 * 573 * Arguments: 574 * zone The UMA_ZONE_REFCNT zone to which the item belongs. 575 * item The address of the item for which we want a refcnt. 576 * 577 * Returns: 578 * A pointer to a u_int32_t reference counter. 579 */ 580u_int32_t *uma_find_refcnt(uma_zone_t zone, void *item); 581 582/* 583 * Used to determine if a fixed-size zone is exhausted. 584 * 585 * Arguments: 586 * zone The zone to check 587 * 588 * Returns: 589 * Non-zero if zone is exhausted. 590 */ 591int uma_zone_exhausted(uma_zone_t zone); 592int uma_zone_exhausted_nolock(uma_zone_t zone); 593 594/* 595 * Exported statistics structures to be used by user space monitoring tools. 596 * Statistics stream consists of a uma_stream_header, followed by a series of 597 * alternative uma_type_header and uma_type_stat structures. 598 */ 599#define UMA_STREAM_VERSION 0x00000001 600struct uma_stream_header { 601 u_int32_t ush_version; /* Stream format version. */ 602 u_int32_t ush_maxcpus; /* Value of MAXCPU for stream. */ 603 u_int32_t ush_count; /* Number of records. */ 604 u_int32_t _ush_pad; /* Pad/reserved field. */ 605}; 606 607#define UTH_MAX_NAME 32 608#define UTH_ZONE_SECONDARY 0x00000001 609struct uma_type_header { 610 /* 611 * Static per-zone data, some extracted from the supporting keg. 612 */ 613 char uth_name[UTH_MAX_NAME]; 614 u_int32_t uth_align; /* Keg: alignment. */ 615 u_int32_t uth_size; /* Keg: requested size of item. */ 616 u_int32_t uth_rsize; /* Keg: real size of item. */ 617 u_int32_t uth_maxpages; /* Keg: maximum number of pages. */ 618 u_int32_t uth_limit; /* Keg: max items to allocate. */ 619 620 /* 621 * Current dynamic zone/keg-derived statistics. 622 */ 623 u_int32_t uth_pages; /* Keg: pages allocated. */ 624 u_int32_t uth_keg_free; /* Keg: items free. */ 625 u_int32_t uth_zone_free; /* Zone: items free. */ 626 u_int32_t uth_bucketsize; /* Zone: desired bucket size. */ 627 u_int32_t uth_zone_flags; /* Zone: flags. */ 628 u_int64_t uth_allocs; /* Zone: number of allocations. */ 629 u_int64_t uth_frees; /* Zone: number of frees. */ 630 u_int64_t uth_fails; /* Zone: number of alloc failures. */ 631 u_int64_t uth_sleeps; /* Zone: number of alloc sleeps. */ 632 u_int64_t _uth_reserved1[2]; /* Reserved. */ 633}; 634 635struct uma_percpu_stat { 636 u_int64_t ups_allocs; /* Cache: number of allocations. */ 637 u_int64_t ups_frees; /* Cache: number of frees. */ 638 u_int64_t ups_cache_free; /* Cache: free items in cache. */ 639 u_int64_t _ups_reserved[5]; /* Reserved. */ 640}; 641 642#endif 643