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_* */
45 /* User visible parameters */
46 #define UMA_SMALLEST_UNIT 8 /* Smallest item allocated */
48 /* Types and type defs */
51 /* Opaque type used as a handle to the zone */
52 typedef struct uma_zone * uma_zone_t;
58 * item A pointer to the memory which has been allocated.
59 * arg The arg field passed to uma_zalloc_arg
60 * size The size of the allocated item
61 * flags See zalloc flags
68 * The constructor is called just before the memory is returned
69 * to the user. It may block if necessary.
71 typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags);
77 * item A pointer to the memory which has been allocated.
78 * size The size of the item being destructed.
79 * arg Argument passed through uma_zfree_arg
85 * The destructor may perform operations that differ from those performed
86 * by the initializer, but it must leave the object in the same state.
87 * This IS type stable storage. This is called after EVERY zfree call.
89 typedef void (*uma_dtor)(void *mem, int size, void *arg);
95 * item A pointer to the memory which has been allocated.
96 * size The size of the item being initialized.
97 * flags See zalloc flags
104 * The initializer is called when the memory is cached in the uma zone.
105 * The initializer and the destructor should leave the object in the same
108 typedef int (*uma_init)(void *mem, int size, int flags);
111 * Item discard function
114 * item A pointer to memory which has been 'freed' but has not left the
116 * size The size of the item being discarded.
122 * This routine is called when memory leaves a zone and is returned to the
123 * system for other uses. It is the counter-part to the init function.
125 typedef void (*uma_fini)(void *mem, int size);
128 * Import new memory into a cache zone.
130 typedef int (*uma_import)(void *arg, void **store, int count, int domain,
134 * Free memory from a cache zone.
136 typedef void (*uma_release)(void *arg, void **store, int count);
139 * What's the difference between initializing and constructing?
141 * The item is initialized when it is cached, and this is the state that the
142 * object should be in when returned to the allocator. The purpose of this is
143 * to remove some code which would otherwise be called on each allocation by
144 * utilizing a known, stable state. This differs from the constructor which
145 * will be called on EVERY allocation.
147 * For example, in the initializer you may want to initialize embedded locks,
148 * NULL list pointers, set up initial states, magic numbers, etc. This way if
149 * the object is held in the allocator and re-used it won't be necessary to
152 * The constructor may be used to lock a data structure, link it on to lists,
153 * bump reference counts or total counts of outstanding structures, etc.
158 /* Function proto types */
161 * Create a new uma zone
164 * name The text name of the zone for debugging and stats. This memory
165 * should not be freed until the zone has been deallocated.
166 * size The size of the object that is being created.
167 * ctor The constructor that is called when the object is allocated.
168 * dtor The destructor that is called when the object is freed.
169 * init An initializer that sets up the initial state of the memory.
170 * fini A discard function that undoes initialization done by init.
171 * ctor/dtor/init/fini may all be null, see notes above.
172 * align A bitmask that corresponds to the requested alignment
174 * flags A set of parameters that control the behavior of the zone.
177 * A pointer to a structure which is intended to be opaque to users of
178 * the interface. The value may be null if the wait flag is not set.
180 uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor,
181 uma_dtor dtor, uma_init uminit, uma_fini fini,
182 int align, uint32_t flags);
185 * Create a secondary uma zone
188 * name The text name of the zone for debugging and stats. This memory
189 * should not be freed until the zone has been deallocated.
190 * ctor The constructor that is called when the object is allocated.
191 * dtor The destructor that is called when the object is freed.
192 * zinit An initializer that sets up the initial state of the memory
193 * as the object passes from the Keg's slab to the Zone's cache.
194 * zfini A discard function that undoes initialization done by init
195 * as the object passes from the Zone's cache to the Keg's slab.
197 * ctor/dtor/zinit/zfini may all be null, see notes above.
198 * Note that the zinit and zfini specified here are NOT
199 * exactly the same as the init/fini specified to uma_zcreate()
200 * when creating a master zone. These zinit/zfini are called
201 * on the TRANSITION from keg to zone (and vice-versa). Once
202 * these are set, the primary zone may alter its init/fini
203 * (which are called when the object passes from VM to keg)
204 * using uma_zone_set_init/fini()) as well as its own
205 * zinit/zfini (unset by default for master zone) with
206 * uma_zone_set_zinit/zfini() (note subtle 'z' prefix).
208 * master A reference to this zone's Master Zone (Primary Zone),
209 * which contains the backing Keg for the Secondary Zone
213 * A pointer to a structure which is intended to be opaque to users of
214 * the interface. The value may be null if the wait flag is not set.
216 uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor,
217 uma_init zinit, uma_fini zfini, uma_zone_t master);
220 * Create cache-only zones.
222 * This allows uma's per-cpu cache facilities to handle arbitrary
223 * pointers. Consumers must specify the import and release functions to
224 * fill and destroy caches. UMA does not allocate any memory for these
225 * zones. The 'arg' parameter is passed to import/release and is caller
228 uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor,
229 uma_init zinit, uma_fini zfini, uma_import zimport,
230 uma_release zrelease, void *arg, int flags);
233 * Definitions for uma_zcreate flags
235 * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to
236 * overlap when adding new features.
238 #define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */
239 #define UMA_ZONE_CONTIG 0x0004 /*
240 * Physical memory underlying an object
241 * must be contiguous.
243 #define UMA_ZONE_NOTOUCH 0x0008 /* UMA may not access the memory */
244 #define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */
245 #define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */
246 #define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */
247 #define UMA_ZONE_VM 0x0080 /*
248 * Used for internal vm datastructures
251 #define UMA_ZONE_NOTPAGE 0x0100 /* allocf memory not vm pages */
252 #define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */
253 #define UMA_ZONE_NOBUCKET 0x0400 /* Do not use buckets. */
254 #define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets. */
255 #define UMA_ZONE_MINBUCKET 0x1000 /* Use smallest buckets. */
256 #define UMA_ZONE_CACHESPREAD 0x2000 /*
257 * Spread memory start locations across
258 * all possible cache lines. May
259 * require many virtually contiguous
260 * backend pages and can fail early.
262 #define UMA_ZONE_NODUMP 0x4000 /*
263 * Zone's pages will not be included in
266 #define UMA_ZONE_PCPU 0x8000 /*
267 * Allocates mp_maxid + 1 slabs of
270 #define UMA_ZONE_FIRSTTOUCH 0x10000 /* First touch NUMA policy */
271 #define UMA_ZONE_ROUNDROBIN 0x20000 /* Round-robin NUMA policy. */
272 #define UMA_ZONE_SMR 0x40000 /*
273 * Safe memory reclamation defers
274 * frees until all read sections
275 * have exited. This flag creates
276 * a unique SMR context for this
277 * zone. To share contexts see
278 * uma_zone_set_smr() below.
280 * See sys/smr.h for more details.
282 /* In use by UMA_ZFLAGs: 0xffe00000 */
285 * These flags are shared between the keg and zone. Some are determined
286 * based on physical parameters of the request and may not be provided by
289 #define UMA_ZONE_INHERIT \
290 (UMA_ZONE_NOTOUCH | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \
291 UMA_ZONE_NOTPAGE | UMA_ZONE_PCPU | UMA_ZONE_FIRSTTOUCH | \
294 /* Definitions for align */
295 #define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */
296 #define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */
297 #define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */
298 #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */
299 #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */
300 #define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */
301 #define UMA_ALIGNOF(type) (_Alignof(type) - 1) /* Alignment fit for 'type' */
303 #define UMA_ANYDOMAIN -1 /* Special value for domain search. */
306 * Destroys an empty uma zone. If the zone is not empty uma complains loudly.
309 * zone The zone we want to destroy.
312 void uma_zdestroy(uma_zone_t zone);
315 * Allocates an item out of a zone
318 * zone The zone we are allocating from
319 * arg This data is passed to the ctor function
320 * flags See sys/malloc.h for available flags.
323 * A non-null pointer to an initialized element from the zone is
324 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer
325 * may be returned if the zone is empty or the ctor failed.
328 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
330 /* Allocate per-cpu data. Access the correct data with zpcpu_get(). */
331 void *uma_zalloc_pcpu_arg(uma_zone_t zone, void *arg, int flags);
333 /* Use with SMR zones. */
334 void *uma_zalloc_smr(uma_zone_t zone, int flags);
337 * Allocate an item from a specific NUMA domain. This uses a slow path in
338 * the allocator but is guaranteed to allocate memory from the requested
339 * domain if M_WAITOK is set.
342 * zone The zone we are allocating from
343 * arg This data is passed to the ctor function
344 * domain The domain to allocate from.
345 * flags See sys/malloc.h for available flags.
347 void *uma_zalloc_domain(uma_zone_t zone, void *arg, int domain, int flags);
350 * Allocates an item out of a zone without supplying an argument
352 * This is just a wrapper for uma_zalloc_arg for convenience.
355 static __inline void *uma_zalloc(uma_zone_t zone, int flags);
356 static __inline void *uma_zalloc_pcpu(uma_zone_t zone, int flags);
358 static __inline void *
359 uma_zalloc(uma_zone_t zone, int flags)
361 return uma_zalloc_arg(zone, NULL, flags);
364 static __inline void *
365 uma_zalloc_pcpu(uma_zone_t zone, int flags)
367 return uma_zalloc_pcpu_arg(zone, NULL, flags);
371 * Frees an item back into the specified zone.
374 * zone The zone the item was originally allocated out of.
375 * item The memory to be freed.
376 * arg Argument passed to the destructor
382 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
384 /* Use with PCPU zones. */
385 void uma_zfree_pcpu_arg(uma_zone_t zone, void *item, void *arg);
387 /* Use with SMR zones. */
388 void uma_zfree_smr(uma_zone_t zone, void *item);
391 * Frees an item back to the specified zone's domain specific pool.
394 * zone The zone the item was originally allocated out of.
395 * item The memory to be freed.
396 * arg Argument passed to the destructor
398 void uma_zfree_domain(uma_zone_t zone, void *item, void *arg);
401 * Frees an item back to a zone without supplying an argument
403 * This is just a wrapper for uma_zfree_arg for convenience.
406 static __inline void uma_zfree(uma_zone_t zone, void *item);
407 static __inline void uma_zfree_pcpu(uma_zone_t zone, void *item);
410 uma_zfree(uma_zone_t zone, void *item)
412 uma_zfree_arg(zone, item, NULL);
416 uma_zfree_pcpu(uma_zone_t zone, void *item)
418 uma_zfree_pcpu_arg(zone, item, NULL);
422 * Wait until the specified zone can allocate an item.
424 void uma_zwait(uma_zone_t zone);
427 * Backend page supplier routines
430 * zone The zone that is requesting pages.
431 * size The number of bytes being requested.
432 * pflag Flags for these memory pages, see below.
433 * domain The NUMA domain that we prefer for this allocation.
434 * wait Indicates our willingness to block.
437 * A pointer to the allocated memory or NULL on failure.
440 typedef void *(*uma_alloc)(uma_zone_t zone, vm_size_t size, int domain,
441 uint8_t *pflag, int wait);
444 * Backend page free routines
447 * item A pointer to the previously allocated pages.
448 * size The original size of the allocation.
449 * pflag The flags for the slab. See UMA_SLAB_* below.
454 typedef void (*uma_free)(void *item, vm_size_t size, uint8_t pflag);
457 * Reclaims unused memory
460 * req Reclamation request type.
464 #define UMA_RECLAIM_DRAIN 1 /* release bucket cache */
465 #define UMA_RECLAIM_DRAIN_CPU 2 /* release bucket and per-CPU caches */
466 #define UMA_RECLAIM_TRIM 3 /* trim bucket cache to WSS */
467 void uma_reclaim(int req);
468 void uma_zone_reclaim(uma_zone_t, int req);
471 * Sets the alignment mask to be used for all zones requesting cache
472 * alignment. Should be called by MD boot code prior to starting VM/UMA.
475 * align The alignment mask
480 void uma_set_align(int align);
483 * Set a reserved number of items to hold for M_USE_RESERVE allocations. All
484 * other requests must allocate new backing pages.
486 void uma_zone_reserve(uma_zone_t zone, int nitems);
489 * Reserves the maximum KVA space required by the zone and configures the zone
490 * to use a VM_ALLOC_NOOBJ-based backend allocator.
493 * zone The zone to update.
494 * nitems The upper limit on the number of items that can be allocated.
497 * 0 if KVA space can not be allocated
501 * When the machine supports a direct map and the zone's items are smaller
502 * than a page, the zone will use the direct map instead of allocating KVA
505 int uma_zone_reserve_kva(uma_zone_t zone, int nitems);
508 * Sets a high limit on the number of items allowed in a zone
511 * zone The zone to limit
512 * nitems The requested upper limit on the number of items allowed
515 * int The effective value of nitems
517 int uma_zone_set_max(uma_zone_t zone, int nitems);
520 * Sets a high limit on the number of items allowed in zone's bucket cache
523 * zone The zone to limit
524 * nitems The requested upper limit on the number of items allowed
526 void uma_zone_set_maxcache(uma_zone_t zone, int nitems);
529 * Obtains the effective limit on the number of items in a zone
532 * zone The zone to obtain the effective limit from
536 * int The effective limit of the zone
538 int uma_zone_get_max(uma_zone_t zone);
541 * Sets a warning to be printed when limit is reached
544 * zone The zone we will warn about
545 * warning Warning content
550 void uma_zone_set_warning(uma_zone_t zone, const char *warning);
553 * Sets a function to run when limit is reached
556 * zone The zone to which this applies
557 * fx The function ro run
562 typedef void (*uma_maxaction_t)(uma_zone_t, int);
563 void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t);
566 * Obtains the approximate current number of items allocated from a zone
569 * zone The zone to obtain the current allocation count from
572 * int The approximate current number of items allocated from the zone
574 int uma_zone_get_cur(uma_zone_t zone);
577 * The following two routines (uma_zone_set_init/fini)
578 * are used to set the backend init/fini pair which acts on an
579 * object as it becomes allocated and is placed in a slab within
580 * the specified zone's backing keg. These should probably not
581 * be changed once allocations have already begun, but only be set
582 * immediately upon zone creation.
584 void uma_zone_set_init(uma_zone_t zone, uma_init uminit);
585 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);
588 * The following two routines (uma_zone_set_zinit/zfini) are
589 * used to set the zinit/zfini pair which acts on an object as
590 * it passes from the backing Keg's slab cache to the
591 * specified Zone's bucket cache. These should probably not
592 * be changed once allocations have already begun, but only be set
593 * immediately upon zone creation.
595 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);
596 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);
599 * Replaces the standard backend allocator for this zone.
602 * zone The zone whose backend allocator is being changed.
603 * allocf A pointer to the allocation function
609 * This could be used to implement pageable allocation, or perhaps
610 * even DMA allocators if used in conjunction with the OFFPAGE
614 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);
617 * Used for freeing memory provided by the allocf above
620 * zone The zone that intends to use this free routine.
621 * freef The page freeing routine.
627 void uma_zone_set_freef(uma_zone_t zone, uma_free freef);
630 * Associate a zone with a smr context that is allocated after creation
631 * so that multiple zones may share the same context.
633 void uma_zone_set_smr(uma_zone_t zone, smr_t smr);
636 * Fetch the smr context that was set or made in uma_zcreate().
638 smr_t uma_zone_get_smr(uma_zone_t zone);
641 * These flags are setable in the allocf and visible in the freef.
643 #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */
644 #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kmem */
645 #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */
646 /* 0x02, 0x10, 0x40, and 0x80 are available */
649 * Used to pre-fill a zone with some number of items
652 * zone The zone to fill
653 * itemcnt The number of items to reserve
658 * NOTE: This is blocking and should only be done at startup
660 void uma_prealloc(uma_zone_t zone, int itemcnt);
663 * Used to determine if a fixed-size zone is exhausted.
666 * zone The zone to check
669 * Non-zero if zone is exhausted.
671 int uma_zone_exhausted(uma_zone_t zone);
674 * Common UMA_ZONE_PCPU zones.
676 extern uma_zone_t pcpu_zone_int;
677 extern uma_zone_t pcpu_zone_64;
680 * Exported statistics structures to be used by user space monitoring tools.
681 * Statistics stream consists of a uma_stream_header, followed by a series of
682 * alternative uma_type_header and uma_type_stat structures.
684 #define UMA_STREAM_VERSION 0x00000001
685 struct uma_stream_header {
686 uint32_t ush_version; /* Stream format version. */
687 uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */
688 uint32_t ush_count; /* Number of records. */
689 uint32_t _ush_pad; /* Pad/reserved field. */
692 #define UTH_MAX_NAME 32
693 #define UTH_ZONE_SECONDARY 0x00000001
694 struct uma_type_header {
696 * Static per-zone data, some extracted from the supporting keg.
698 char uth_name[UTH_MAX_NAME];
699 uint32_t uth_align; /* Keg: alignment. */
700 uint32_t uth_size; /* Keg: requested size of item. */
701 uint32_t uth_rsize; /* Keg: real size of item. */
702 uint32_t uth_maxpages; /* Keg: maximum number of pages. */
703 uint32_t uth_limit; /* Keg: max items to allocate. */
706 * Current dynamic zone/keg-derived statistics.
708 uint32_t uth_pages; /* Keg: pages allocated. */
709 uint32_t uth_keg_free; /* Keg: items free. */
710 uint32_t uth_zone_free; /* Zone: items free. */
711 uint32_t uth_bucketsize; /* Zone: desired bucket size. */
712 uint32_t uth_zone_flags; /* Zone: flags. */
713 uint64_t uth_allocs; /* Zone: number of allocations. */
714 uint64_t uth_frees; /* Zone: number of frees. */
715 uint64_t uth_fails; /* Zone: number of alloc failures. */
716 uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */
717 uint64_t uth_xdomain; /* Zone: Number of cross domain frees. */
718 uint64_t _uth_reserved1[1]; /* Reserved. */
721 struct uma_percpu_stat {
722 uint64_t ups_allocs; /* Cache: number of allocations. */
723 uint64_t ups_frees; /* Cache: number of frees. */
724 uint64_t ups_cache_free; /* Cache: free items in cache. */
725 uint64_t _ups_reserved[5]; /* Reserved. */
728 void uma_reclaim_wakeup(void);
729 void uma_reclaim_worker(void *);
731 unsigned long uma_limit(void);
733 /* Return the amount of memory managed by UMA. */
734 unsigned long uma_size(void);
736 /* Return the amount of memory remaining. May be negative. */
737 long uma_avail(void);
739 #endif /* _VM_UMA_H_ */