2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2002, 2003, 2004, 2005 Jeffrey Roberson <jeff@FreeBSD.org>
5 * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice unmodified, this list of conditions, and the following
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.
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.
34 * uma.h - External definitions for the Universal Memory Allocator
41 #include <sys/param.h> /* For NULL */
42 #include <sys/malloc.h> /* For M_* */
44 /* User visible parameters */
45 #define UMA_SMALLEST_UNIT (PAGE_SIZE / 256) /* Smallest item allocated */
47 /* Types and type defs */
50 /* Opaque type used as a handle to the zone */
51 typedef struct uma_zone * uma_zone_t;
53 void zone_drain(uma_zone_t);
59 * item A pointer to the memory which has been allocated.
60 * arg The arg field passed to uma_zalloc_arg
61 * size The size of the allocated item
62 * flags See zalloc flags
69 * The constructor is called just before the memory is returned
70 * to the user. It may block if necessary.
72 typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags);
78 * item A pointer to the memory which has been allocated.
79 * size The size of the item being destructed.
80 * arg Argument passed through uma_zfree_arg
86 * The destructor may perform operations that differ from those performed
87 * by the initializer, but it must leave the object in the same state.
88 * This IS type stable storage. This is called after EVERY zfree call.
90 typedef void (*uma_dtor)(void *mem, int size, void *arg);
96 * item A pointer to the memory which has been allocated.
97 * size The size of the item being initialized.
98 * flags See zalloc flags
105 * The initializer is called when the memory is cached in the uma zone.
106 * The initializer and the destructor should leave the object in the same
109 typedef int (*uma_init)(void *mem, int size, int flags);
112 * Item discard function
115 * item A pointer to memory which has been 'freed' but has not left the
117 * size The size of the item being discarded.
123 * This routine is called when memory leaves a zone and is returned to the
124 * system for other uses. It is the counter-part to the init function.
126 typedef void (*uma_fini)(void *mem, int size);
129 * Import new memory into a cache zone.
131 typedef int (*uma_import)(void *arg, void **store, int count, int domain,
135 * Free memory from a cache zone.
137 typedef void (*uma_release)(void *arg, void **store, int count);
140 * What's the difference between initializing and constructing?
142 * The item is initialized when it is cached, and this is the state that the
143 * object should be in when returned to the allocator. The purpose of this is
144 * to remove some code which would otherwise be called on each allocation by
145 * utilizing a known, stable state. This differs from the constructor which
146 * will be called on EVERY allocation.
148 * For example, in the initializer you may want to initialize embedded locks,
149 * NULL list pointers, set up initial states, magic numbers, etc. This way if
150 * the object is held in the allocator and re-used it won't be necessary to
153 * The constructor may be used to lock a data structure, link it on to lists,
154 * bump reference counts or total counts of outstanding structures, etc.
159 /* Function proto types */
162 * Create a new uma zone
165 * name The text name of the zone for debugging and stats. This memory
166 * should not be freed until the zone has been deallocated.
167 * size The size of the object that is being created.
168 * ctor The constructor that is called when the object is allocated.
169 * dtor The destructor that is called when the object is freed.
170 * init An initializer that sets up the initial state of the memory.
171 * fini A discard function that undoes initialization done by init.
172 * ctor/dtor/init/fini may all be null, see notes above.
173 * align A bitmask that corresponds to the requested alignment
175 * flags A set of parameters that control the behavior of the zone.
178 * A pointer to a structure which is intended to be opaque to users of
179 * the interface. The value may be null if the wait flag is not set.
181 uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor,
182 uma_dtor dtor, uma_init uminit, uma_fini fini,
183 int align, uint32_t flags);
186 * Create a secondary uma zone
189 * name The text name of the zone for debugging and stats. This memory
190 * should not be freed until the zone has been deallocated.
191 * ctor The constructor that is called when the object is allocated.
192 * dtor The destructor that is called when the object is freed.
193 * zinit An initializer that sets up the initial state of the memory
194 * as the object passes from the Keg's slab to the Zone's cache.
195 * zfini A discard function that undoes initialization done by init
196 * as the object passes from the Zone's cache to the Keg's slab.
198 * ctor/dtor/zinit/zfini may all be null, see notes above.
199 * Note that the zinit and zfini specified here are NOT
200 * exactly the same as the init/fini specified to uma_zcreate()
201 * when creating a master zone. These zinit/zfini are called
202 * on the TRANSITION from keg to zone (and vice-versa). Once
203 * these are set, the primary zone may alter its init/fini
204 * (which are called when the object passes from VM to keg)
205 * using uma_zone_set_init/fini()) as well as its own
206 * zinit/zfini (unset by default for master zone) with
207 * uma_zone_set_zinit/zfini() (note subtle 'z' prefix).
209 * master A reference to this zone's Master Zone (Primary Zone),
210 * which contains the backing Keg for the Secondary Zone
214 * A pointer to a structure which is intended to be opaque to users of
215 * the interface. The value may be null if the wait flag is not set.
217 uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor,
218 uma_init zinit, uma_fini zfini, uma_zone_t master);
221 * Create cache-only zones.
223 * This allows uma's per-cpu cache facilities to handle arbitrary
224 * pointers. Consumers must specify the import and release functions to
225 * fill and destroy caches. UMA does not allocate any memory for these
226 * zones. The 'arg' parameter is passed to import/release and is caller
229 uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor,
230 uma_init zinit, uma_fini zfini, uma_import zimport,
231 uma_release zrelease, void *arg, int flags);
234 * Definitions for uma_zcreate flags
236 * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to
237 * overlap when adding new features. 0xff000000 is in use by uma_int.h.
239 #define UMA_ZONE_PAGEABLE 0x0001 /* Return items not fully backed by
240 physical memory XXX Not yet */
241 #define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */
242 #define UMA_ZONE_STATIC 0x0004 /* Statically sized zone */
243 #define UMA_ZONE_OFFPAGE 0x0008 /* Force the slab structure allocation
244 off of the real memory */
245 #define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */
246 #define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */
247 #define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */
248 #define UMA_ZONE_VM 0x0080 /*
249 * Used for internal vm datastructures
252 #define UMA_ZONE_HASH 0x0100 /*
253 * Use a hash table instead of caching
254 * information in the vm_page.
256 #define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */
257 #define UMA_ZONE_NOBUCKET 0x0400 /* Do not use buckets. */
258 #define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets. */
259 #define UMA_ZONE_CACHESPREAD 0x1000 /*
260 * Spread memory start locations across
261 * all possible cache lines. May
262 * require many virtually contiguous
263 * backend pages and can fail early.
265 #define UMA_ZONE_VTOSLAB 0x2000 /* Zone uses vtoslab for lookup. */
266 #define UMA_ZONE_NODUMP 0x4000 /*
267 * Zone's pages will not be included in
270 #define UMA_ZONE_PCPU 0x8000 /*
271 * Allocates mp_maxid + 1 slabs of PAGE_SIZE
273 #define UMA_ZONE_NUMA 0x10000 /*
274 * NUMA aware Zone. Implements a best
275 * effort first-touch policy.
277 #define UMA_ZONE_MINBUCKET 0x20000 /* Use smallest buckets. */
280 * These flags are shared between the keg and zone. In zones wishing to add
281 * new kegs these flags must be compatible. Some are determined based on
282 * physical parameters of the request and may not be provided by the consumer.
284 #define UMA_ZONE_INHERIT \
285 (UMA_ZONE_OFFPAGE | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \
286 UMA_ZONE_HASH | UMA_ZONE_VTOSLAB | UMA_ZONE_PCPU | UMA_ZONE_NUMA)
288 /* Definitions for align */
289 #define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */
290 #define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */
291 #define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */
292 #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */
293 #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */
294 #define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */
295 #define UMA_ALIGNOF(type) (_Alignof(type) - 1) /* Alignment fit for 'type' */
298 * Destroys an empty uma zone. If the zone is not empty uma complains loudly.
301 * zone The zone we want to destroy.
304 void uma_zdestroy(uma_zone_t zone);
307 * Allocates an item out of a zone
310 * zone The zone we are allocating from
311 * arg This data is passed to the ctor function
312 * flags See sys/malloc.h for available flags.
315 * A non-null pointer to an initialized element from the zone is
316 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer
317 * may be returned if the zone is empty or the ctor failed.
320 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
321 void *uma_zalloc_pcpu_arg(uma_zone_t zone, void *arg, int flags);
324 * Allocate an item from a specific NUMA domain. This uses a slow path in
325 * the allocator but is guaranteed to allocate memory from the requested
326 * domain if M_WAITOK is set.
329 * zone The zone we are allocating from
330 * arg This data is passed to the ctor function
331 * domain The domain to allocate from.
332 * flags See sys/malloc.h for available flags.
334 void *uma_zalloc_domain(uma_zone_t zone, void *arg, int domain, int flags);
337 * Allocates an item out of a zone without supplying an argument
339 * This is just a wrapper for uma_zalloc_arg for convenience.
342 static __inline void *uma_zalloc(uma_zone_t zone, int flags);
343 static __inline void *uma_zalloc_pcpu(uma_zone_t zone, int flags);
345 static __inline void *
346 uma_zalloc(uma_zone_t zone, int flags)
348 return uma_zalloc_arg(zone, NULL, flags);
351 static __inline void *
352 uma_zalloc_pcpu(uma_zone_t zone, int flags)
354 return uma_zalloc_pcpu_arg(zone, NULL, flags);
358 * Frees an item back into the specified zone.
361 * zone The zone the item was originally allocated out of.
362 * item The memory to be freed.
363 * arg Argument passed to the destructor
369 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
370 void uma_zfree_pcpu_arg(uma_zone_t zone, void *item, void *arg);
373 * Frees an item back to the specified zone's domain specific pool.
376 * zone The zone the item was originally allocated out of.
377 * item The memory to be freed.
378 * arg Argument passed to the destructor
380 void uma_zfree_domain(uma_zone_t zone, void *item, void *arg);
383 * Frees an item back to a zone without supplying an argument
385 * This is just a wrapper for uma_zfree_arg for convenience.
388 static __inline void uma_zfree(uma_zone_t zone, void *item);
389 static __inline void uma_zfree_pcpu(uma_zone_t zone, void *item);
392 uma_zfree(uma_zone_t zone, void *item)
394 uma_zfree_arg(zone, item, NULL);
398 uma_zfree_pcpu(uma_zone_t zone, void *item)
400 uma_zfree_pcpu_arg(zone, item, NULL);
404 * Wait until the specified zone can allocate an item.
406 void uma_zwait(uma_zone_t zone);
409 * Backend page supplier routines
412 * zone The zone that is requesting pages.
413 * size The number of bytes being requested.
414 * pflag Flags for these memory pages, see below.
415 * domain The NUMA domain that we prefer for this allocation.
416 * wait Indicates our willingness to block.
419 * A pointer to the allocated memory or NULL on failure.
422 typedef void *(*uma_alloc)(uma_zone_t zone, vm_size_t size, int domain,
423 uint8_t *pflag, int wait);
426 * Backend page free routines
429 * item A pointer to the previously allocated pages.
430 * size The original size of the allocation.
431 * pflag The flags for the slab. See UMA_SLAB_* below.
436 typedef void (*uma_free)(void *item, vm_size_t size, uint8_t pflag);
439 * Reclaims unused memory for all zones
446 * This should only be called by the page out daemon.
449 void uma_reclaim(void);
452 * Sets the alignment mask to be used for all zones requesting cache
453 * alignment. Should be called by MD boot code prior to starting VM/UMA.
456 * align The alignment mask
461 void uma_set_align(int align);
464 * Set a reserved number of items to hold for M_USE_RESERVE allocations. All
465 * other requests must allocate new backing pages.
467 void uma_zone_reserve(uma_zone_t zone, int nitems);
470 * Reserves the maximum KVA space required by the zone and configures the zone
471 * to use a VM_ALLOC_NOOBJ-based backend allocator.
474 * zone The zone to update.
475 * nitems The upper limit on the number of items that can be allocated.
478 * 0 if KVA space can not be allocated
482 * When the machine supports a direct map and the zone's items are smaller
483 * than a page, the zone will use the direct map instead of allocating KVA
486 int uma_zone_reserve_kva(uma_zone_t zone, int nitems);
489 * Sets a high limit on the number of items allowed in a zone
492 * zone The zone to limit
493 * nitems The requested upper limit on the number of items allowed
496 * int The effective value of nitems after rounding up based on page size
498 int uma_zone_set_max(uma_zone_t zone, int nitems);
501 * Sets a high limit on the number of items allowed in zone's bucket cache
504 * zone The zone to limit
505 * nitems The requested upper limit on the number of items allowed
508 * int The effective value of nitems set
510 int uma_zone_set_maxcache(uma_zone_t zone, int nitems);
513 * Obtains the effective limit on the number of items in a zone
516 * zone The zone to obtain the effective limit from
520 * int The effective limit of the zone
522 int uma_zone_get_max(uma_zone_t zone);
525 * Sets a warning to be printed when limit is reached
528 * zone The zone we will warn about
529 * warning Warning content
534 void uma_zone_set_warning(uma_zone_t zone, const char *warning);
537 * Sets a function to run when limit is reached
540 * zone The zone to which this applies
541 * fx The function ro run
546 typedef void (*uma_maxaction_t)(uma_zone_t, int);
547 void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t);
550 * Obtains the approximate current number of items allocated from a zone
553 * zone The zone to obtain the current allocation count from
556 * int The approximate current number of items allocated from the zone
558 int uma_zone_get_cur(uma_zone_t zone);
561 * The following two routines (uma_zone_set_init/fini)
562 * are used to set the backend init/fini pair which acts on an
563 * object as it becomes allocated and is placed in a slab within
564 * the specified zone's backing keg. These should probably not
565 * be changed once allocations have already begun, but only be set
566 * immediately upon zone creation.
568 void uma_zone_set_init(uma_zone_t zone, uma_init uminit);
569 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);
572 * The following two routines (uma_zone_set_zinit/zfini) are
573 * used to set the zinit/zfini pair which acts on an object as
574 * it passes from the backing Keg's slab cache to the
575 * specified Zone's bucket cache. These should probably not
576 * be changed once allocations have already begun, but only be set
577 * immediately upon zone creation.
579 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);
580 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);
583 * Replaces the standard backend allocator for this zone.
586 * zone The zone whose backend allocator is being changed.
587 * allocf A pointer to the allocation function
593 * This could be used to implement pageable allocation, or perhaps
594 * even DMA allocators if used in conjunction with the OFFPAGE
598 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);
601 * Used for freeing memory provided by the allocf above
604 * zone The zone that intends to use this free routine.
605 * freef The page freeing routine.
611 void uma_zone_set_freef(uma_zone_t zone, uma_free freef);
614 * These flags are setable in the allocf and visible in the freef.
616 #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */
617 #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kmem */
618 #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */
619 #define UMA_SLAB_OFFP 0x10 /* Slab is managed separately */
620 #define UMA_SLAB_MALLOC 0x20 /* Slab is a large malloc slab */
621 /* 0x02, 0x40, and 0x80 are available */
624 * Used to pre-fill a zone with some number of items
627 * zone The zone to fill
628 * itemcnt The number of items to reserve
633 * NOTE: This is blocking and should only be done at startup
635 void uma_prealloc(uma_zone_t zone, int itemcnt);
638 * Used to determine if a fixed-size zone is exhausted.
641 * zone The zone to check
644 * Non-zero if zone is exhausted.
646 int uma_zone_exhausted(uma_zone_t zone);
647 int uma_zone_exhausted_nolock(uma_zone_t zone);
650 * Common UMA_ZONE_PCPU zones.
652 extern uma_zone_t pcpu_zone_64;
653 extern uma_zone_t pcpu_zone_ptr;
656 * Exported statistics structures to be used by user space monitoring tools.
657 * Statistics stream consists of a uma_stream_header, followed by a series of
658 * alternative uma_type_header and uma_type_stat structures.
660 #define UMA_STREAM_VERSION 0x00000001
661 struct uma_stream_header {
662 uint32_t ush_version; /* Stream format version. */
663 uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */
664 uint32_t ush_count; /* Number of records. */
665 uint32_t _ush_pad; /* Pad/reserved field. */
668 #define UTH_MAX_NAME 32
669 #define UTH_ZONE_SECONDARY 0x00000001
670 struct uma_type_header {
672 * Static per-zone data, some extracted from the supporting keg.
674 char uth_name[UTH_MAX_NAME];
675 uint32_t uth_align; /* Keg: alignment. */
676 uint32_t uth_size; /* Keg: requested size of item. */
677 uint32_t uth_rsize; /* Keg: real size of item. */
678 uint32_t uth_maxpages; /* Keg: maximum number of pages. */
679 uint32_t uth_limit; /* Keg: max items to allocate. */
682 * Current dynamic zone/keg-derived statistics.
684 uint32_t uth_pages; /* Keg: pages allocated. */
685 uint32_t uth_keg_free; /* Keg: items free. */
686 uint32_t uth_zone_free; /* Zone: items free. */
687 uint32_t uth_bucketsize; /* Zone: desired bucket size. */
688 uint32_t uth_zone_flags; /* Zone: flags. */
689 uint64_t uth_allocs; /* Zone: number of allocations. */
690 uint64_t uth_frees; /* Zone: number of frees. */
691 uint64_t uth_fails; /* Zone: number of alloc failures. */
692 uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */
693 uint64_t uth_xdomain; /* Zone: Number of cross domain frees. */
694 uint64_t _uth_reserved1[1]; /* Reserved. */
697 struct uma_percpu_stat {
698 uint64_t ups_allocs; /* Cache: number of allocations. */
699 uint64_t ups_frees; /* Cache: number of frees. */
700 uint64_t ups_cache_free; /* Cache: free items in cache. */
701 uint64_t _ups_reserved[5]; /* Reserved. */
704 void uma_reclaim_wakeup(void);
705 void uma_reclaim_worker(void *);
707 unsigned long uma_limit(void);
709 /* Return the amount of memory managed by UMA. */
710 unsigned long uma_size(void);
712 /* Return the amount of memory remaining. May be negative. */
713 long uma_avail(void);
715 #endif /* _VM_UMA_H_ */