2 * Copyright (c) 2002, 2003, 2004, 2005 Jeffrey Roberson <jeff@FreeBSD.org>
3 * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice unmodified, this list of conditions, and the following
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.
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.
32 * uma.h - External definitions for the Universal Memory Allocator
39 #include <sys/param.h> /* For NULL */
40 #include <sys/malloc.h> /* For M_* */
42 /* User visible parameters */
43 #define UMA_SMALLEST_UNIT (PAGE_SIZE / 256) /* Smallest item allocated */
45 /* Types and type defs */
48 /* Opaque type used as a handle to the zone */
49 typedef struct uma_zone * uma_zone_t;
51 void zone_drain(uma_zone_t);
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
67 * The constructor is called just before the memory is returned
68 * to the user. It may block if necessary.
70 typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags);
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
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.
88 typedef void (*uma_dtor)(void *mem, int size, void *arg);
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
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
107 typedef int (*uma_init)(void *mem, int size, int flags);
110 * Item discard function
113 * item A pointer to memory which has been 'freed' but has not left the
115 * size The size of the item being discarded.
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.
124 typedef void (*uma_fini)(void *mem, int size);
127 * Import new memory into a cache zone.
129 typedef int (*uma_import)(void *arg, void **store, int count, int flags);
132 * Free memory from a cache zone.
134 typedef void (*uma_release)(void *arg, void **store, int count);
137 * What's the difference between initializing and constructing?
139 * The item is initialized when it is cached, and this is the state that the
140 * object should be in when returned to the allocator. The purpose of this is
141 * to remove some code which would otherwise be called on each allocation by
142 * utilizing a known, stable state. This differs from the constructor which
143 * will be called on EVERY allocation.
145 * For example, in the initializer you may want to initialize embedded locks,
146 * NULL list pointers, set up initial states, magic numbers, etc. This way if
147 * the object is held in the allocator and re-used it won't be necessary to
150 * The constructor may be used to lock a data structure, link it on to lists,
151 * bump reference counts or total counts of outstanding structures, etc.
156 /* Function proto types */
159 * Create a new uma zone
162 * name The text name of the zone for debugging and stats. This memory
163 * should not be freed until the zone has been deallocated.
164 * size The size of the object that is being created.
165 * ctor The constructor that is called when the object is allocated.
166 * dtor The destructor that is called when the object is freed.
167 * init An initializer that sets up the initial state of the memory.
168 * fini A discard function that undoes initialization done by init.
169 * ctor/dtor/init/fini may all be null, see notes above.
170 * align A bitmask that corresponds to the requested alignment
172 * flags A set of parameters that control the behavior of the zone.
175 * A pointer to a structure which is intended to be opaque to users of
176 * the interface. The value may be null if the wait flag is not set.
178 uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor,
179 uma_dtor dtor, uma_init uminit, uma_fini fini,
180 int align, uint32_t flags);
183 * Create a secondary uma zone
186 * name The text name of the zone for debugging and stats. This memory
187 * should not be freed until the zone has been deallocated.
188 * ctor The constructor that is called when the object is allocated.
189 * dtor The destructor that is called when the object is freed.
190 * zinit An initializer that sets up the initial state of the memory
191 * as the object passes from the Keg's slab to the Zone's cache.
192 * zfini A discard function that undoes initialization done by init
193 * as the object passes from the Zone's cache to the Keg's slab.
195 * ctor/dtor/zinit/zfini may all be null, see notes above.
196 * Note that the zinit and zfini specified here are NOT
197 * exactly the same as the init/fini specified to uma_zcreate()
198 * when creating a master zone. These zinit/zfini are called
199 * on the TRANSITION from keg to zone (and vice-versa). Once
200 * these are set, the primary zone may alter its init/fini
201 * (which are called when the object passes from VM to keg)
202 * using uma_zone_set_init/fini()) as well as its own
203 * zinit/zfini (unset by default for master zone) with
204 * uma_zone_set_zinit/zfini() (note subtle 'z' prefix).
206 * master A reference to this zone's Master Zone (Primary Zone),
207 * which contains the backing Keg for the Secondary Zone
211 * A pointer to a structure which is intended to be opaque to users of
212 * the interface. The value may be null if the wait flag is not set.
214 uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor,
215 uma_init zinit, uma_fini zfini, uma_zone_t master);
218 * Add a second master to a secondary zone. This provides multiple data
219 * backends for objects with the same size. Both masters must have
220 * compatible allocation flags. Presently, UMA_ZONE_MALLOC type zones are
221 * the only supported.
224 * Error on failure, 0 on success.
226 int uma_zsecond_add(uma_zone_t zone, uma_zone_t master);
229 * Create cache-only zones.
231 * This allows uma's per-cpu cache facilities to handle arbitrary
232 * pointers. Consumers must specify the import and release functions to
233 * fill and destroy caches. UMA does not allocate any memory for these
234 * zones. The 'arg' parameter is passed to import/release and is caller
237 uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor,
238 uma_init zinit, uma_fini zfini, uma_import zimport,
239 uma_release zrelease, void *arg, int flags);
242 * Definitions for uma_zcreate flags
244 * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to
245 * overlap when adding new features. 0xf0000000 is in use by uma_int.h.
247 #define UMA_ZONE_PAGEABLE 0x0001 /* Return items not fully backed by
248 physical memory XXX Not yet */
249 #define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */
250 #define UMA_ZONE_STATIC 0x0004 /* Statically sized zone */
251 #define UMA_ZONE_OFFPAGE 0x0008 /* Force the slab structure allocation
252 off of the real memory */
253 #define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */
254 #define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */
255 #define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */
256 #define UMA_ZONE_VM 0x0080 /*
257 * Used for internal vm datastructures
260 #define UMA_ZONE_HASH 0x0100 /*
261 * Use a hash table instead of caching
262 * information in the vm_page.
264 #define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */
265 #define UMA_ZONE_REFCNT 0x0400 /* Allocate refcnts in slabs */
266 #define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets */
267 #define UMA_ZONE_CACHESPREAD 0x1000 /*
268 * Spread memory start locations across
269 * all possible cache lines. May
270 * require many virtually contiguous
271 * backend pages and can fail early.
273 #define UMA_ZONE_VTOSLAB 0x2000 /* Zone uses vtoslab for lookup. */
274 #define UMA_ZONE_NODUMP 0x4000 /*
275 * Zone's pages will not be included in
278 #define UMA_ZONE_PCPU 0x8000 /*
279 * Allocates mp_ncpus slabs sized to
280 * sizeof(struct pcpu).
284 * These flags are shared between the keg and zone. In zones wishing to add
285 * new kegs these flags must be compatible. Some are determined based on
286 * physical parameters of the request and may not be provided by the consumer.
288 #define UMA_ZONE_INHERIT \
289 (UMA_ZONE_OFFPAGE | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \
290 UMA_ZONE_HASH | UMA_ZONE_REFCNT | UMA_ZONE_VTOSLAB | UMA_ZONE_PCPU)
292 /* Definitions for align */
293 #define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */
294 #define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */
295 #define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */
296 #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */
297 #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */
298 #define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */
301 * Destroys an empty uma zone. If the zone is not empty uma complains loudly.
304 * zone The zone we want to destroy.
307 void uma_zdestroy(uma_zone_t zone);
310 * Allocates an item out of a zone
313 * zone The zone we are allocating from
314 * arg This data is passed to the ctor function
315 * flags See sys/malloc.h for available flags.
318 * A non-null pointer to an initialized element from the zone is
319 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer
320 * may be returned if the zone is empty or the ctor failed.
323 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
326 * Allocates an item out of a zone without supplying an argument
328 * This is just a wrapper for uma_zalloc_arg for convenience.
331 static __inline void *uma_zalloc(uma_zone_t zone, int flags);
333 static __inline void *
334 uma_zalloc(uma_zone_t zone, int flags)
336 return uma_zalloc_arg(zone, NULL, flags);
340 * Frees an item back into the specified zone.
343 * zone The zone the item was originally allocated out of.
344 * item The memory to be freed.
345 * arg Argument passed to the destructor
351 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
354 * Frees an item back to a zone without supplying an argument
356 * This is just a wrapper for uma_zfree_arg for convenience.
359 static __inline void uma_zfree(uma_zone_t zone, void *item);
362 uma_zfree(uma_zone_t zone, void *item)
364 uma_zfree_arg(zone, item, NULL);
368 * XXX The rest of the prototypes in this header are h0h0 magic for the VM.
369 * If you think you need to use it for a normal zone you're probably incorrect.
373 * Backend page supplier routines
376 * zone The zone that is requesting pages.
377 * size The number of bytes being requested.
378 * pflag Flags for these memory pages, see below.
379 * wait Indicates our willingness to block.
382 * A pointer to the allocated memory or NULL on failure.
385 typedef void *(*uma_alloc)(uma_zone_t zone, int size, uint8_t *pflag, int wait);
388 * Backend page free routines
391 * item A pointer to the previously allocated pages.
392 * size The original size of the allocation.
393 * pflag The flags for the slab. See UMA_SLAB_* below.
398 typedef void (*uma_free)(void *item, int size, uint8_t pflag);
403 * Sets up the uma allocator. (Called by vm_mem_init)
406 * bootmem A pointer to memory used to bootstrap the system.
412 * This memory is used for zones which allocate things before the
413 * backend page supplier can give us pages. It should be
414 * UMA_SLAB_SIZE * boot_pages bytes. (see uma_int.h)
418 void uma_startup(void *bootmem, int boot_pages);
421 * Finishes starting up the allocator. This should
422 * be called when kva is ready for normal allocs.
431 * uma_startup2 is called by kmeminit() to enable us of uma for malloc.
434 void uma_startup2(void);
437 * Reclaims unused memory for all zones
444 * This should only be called by the page out daemon.
447 void uma_reclaim(void);
450 * Sets the alignment mask to be used for all zones requesting cache
451 * alignment. Should be called by MD boot code prior to starting VM/UMA.
454 * align The alignment mask
459 void uma_set_align(int align);
462 * Set a reserved number of items to hold for M_USE_RESERVE allocations. All
463 * other requests must allocate new backing pages.
465 void uma_zone_reserve(uma_zone_t zone, int nitems);
468 * Reserves the maximum KVA space required by the zone and configures the zone
469 * to use a VM_ALLOC_NOOBJ-based backend allocator.
472 * zone The zone to update.
473 * nitems The upper limit on the number of items that can be allocated.
476 * 0 if KVA space can not be allocated
480 * When the machine supports a direct map and the zone's items are smaller
481 * than a page, the zone will use the direct map instead of allocating KVA
484 int uma_zone_reserve_kva(uma_zone_t zone, int nitems);
487 * Sets a high limit on the number of items allowed in a zone
490 * zone The zone to limit
491 * nitems The requested upper limit on the number of items allowed
494 * int The effective value of nitems after rounding up based on page size
496 int uma_zone_set_max(uma_zone_t zone, int nitems);
499 * Obtains the effective limit on the number of items in a zone
502 * zone The zone to obtain the effective limit from
506 * int The effective limit of the zone
508 int uma_zone_get_max(uma_zone_t zone);
511 * Sets a warning to be printed when limit is reached
514 * zone The zone we will warn about
515 * warning Warning content
520 void uma_zone_set_warning(uma_zone_t zone, const char *warning);
523 * Obtains the approximate current number of items allocated from a zone
526 * zone The zone to obtain the current allocation count from
529 * int The approximate current number of items allocated from the zone
531 int uma_zone_get_cur(uma_zone_t zone);
534 * The following two routines (uma_zone_set_init/fini)
535 * are used to set the backend init/fini pair which acts on an
536 * object as it becomes allocated and is placed in a slab within
537 * the specified zone's backing keg. These should probably not
538 * be changed once allocations have already begun, but only be set
539 * immediately upon zone creation.
541 void uma_zone_set_init(uma_zone_t zone, uma_init uminit);
542 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);
545 * The following two routines (uma_zone_set_zinit/zfini) are
546 * used to set the zinit/zfini pair which acts on an object as
547 * it passes from the backing Keg's slab cache to the
548 * specified Zone's bucket cache. These should probably not
549 * be changed once allocations have already begun, but only be set
550 * immediately upon zone creation.
552 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);
553 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);
556 * Replaces the standard backend allocator for this zone.
559 * zone The zone whose backend allocator is being changed.
560 * allocf A pointer to the allocation function
566 * This could be used to implement pageable allocation, or perhaps
567 * even DMA allocators if used in conjunction with the OFFPAGE
571 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);
574 * Used for freeing memory provided by the allocf above
577 * zone The zone that intends to use this free routine.
578 * freef The page freeing routine.
584 void uma_zone_set_freef(uma_zone_t zone, uma_free freef);
587 * These flags are setable in the allocf and visible in the freef.
589 #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */
590 #define UMA_SLAB_KMEM 0x02 /* Slab alloced from kmem_map */
591 #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kernel_map */
592 #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */
593 #define UMA_SLAB_OFFP 0x10 /* Slab is managed separately */
594 #define UMA_SLAB_MALLOC 0x20 /* Slab is a large malloc slab */
595 /* 0x40 and 0x80 are available */
598 * Used to pre-fill a zone with some number of items
601 * zone The zone to fill
602 * itemcnt The number of items to reserve
607 * NOTE: This is blocking and should only be done at startup
609 void uma_prealloc(uma_zone_t zone, int itemcnt);
612 * Used to lookup the reference counter allocated for an item
613 * from a UMA_ZONE_REFCNT zone. For UMA_ZONE_REFCNT zones,
614 * reference counters are allocated for items and stored in
615 * the underlying slab header.
618 * zone The UMA_ZONE_REFCNT zone to which the item belongs.
619 * item The address of the item for which we want a refcnt.
622 * A pointer to a uint32_t reference counter.
624 uint32_t *uma_find_refcnt(uma_zone_t zone, void *item);
627 * Used to determine if a fixed-size zone is exhausted.
630 * zone The zone to check
633 * Non-zero if zone is exhausted.
635 int uma_zone_exhausted(uma_zone_t zone);
636 int uma_zone_exhausted_nolock(uma_zone_t zone);
639 * Common UMA_ZONE_PCPU zones.
641 extern uma_zone_t pcpu_zone_64;
642 extern uma_zone_t pcpu_zone_ptr;
645 * Exported statistics structures to be used by user space monitoring tools.
646 * Statistics stream consists of a uma_stream_header, followed by a series of
647 * alternative uma_type_header and uma_type_stat structures.
649 #define UMA_STREAM_VERSION 0x00000001
650 struct uma_stream_header {
651 uint32_t ush_version; /* Stream format version. */
652 uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */
653 uint32_t ush_count; /* Number of records. */
654 uint32_t _ush_pad; /* Pad/reserved field. */
657 #define UTH_MAX_NAME 32
658 #define UTH_ZONE_SECONDARY 0x00000001
659 struct uma_type_header {
661 * Static per-zone data, some extracted from the supporting keg.
663 char uth_name[UTH_MAX_NAME];
664 uint32_t uth_align; /* Keg: alignment. */
665 uint32_t uth_size; /* Keg: requested size of item. */
666 uint32_t uth_rsize; /* Keg: real size of item. */
667 uint32_t uth_maxpages; /* Keg: maximum number of pages. */
668 uint32_t uth_limit; /* Keg: max items to allocate. */
671 * Current dynamic zone/keg-derived statistics.
673 uint32_t uth_pages; /* Keg: pages allocated. */
674 uint32_t uth_keg_free; /* Keg: items free. */
675 uint32_t uth_zone_free; /* Zone: items free. */
676 uint32_t uth_bucketsize; /* Zone: desired bucket size. */
677 uint32_t uth_zone_flags; /* Zone: flags. */
678 uint64_t uth_allocs; /* Zone: number of allocations. */
679 uint64_t uth_frees; /* Zone: number of frees. */
680 uint64_t uth_fails; /* Zone: number of alloc failures. */
681 uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */
682 uint64_t _uth_reserved1[2]; /* Reserved. */
685 struct uma_percpu_stat {
686 uint64_t ups_allocs; /* Cache: number of allocations. */
687 uint64_t ups_frees; /* Cache: number of frees. */
688 uint64_t ups_cache_free; /* Cache: free items in cache. */
689 uint64_t _ups_reserved[5]; /* Reserved. */
692 #endif /* _VM_UMA_H_ */