2 * SPDX-License-Identifier: BSD-2-Clause
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.
32 * uma.h - External definitions for the Universal Memory Allocator
39 #include <sys/param.h> /* For NULL */
40 #include <sys/malloc.h> /* For M_* */
43 /* User visible parameters */
44 #define UMA_SMALLEST_UNIT 8 /* Smallest item allocated */
46 /* Types and type defs */
49 /* Opaque type used as a handle to the zone */
50 typedef struct uma_zone * uma_zone_t;
56 * item A pointer to the memory which has been allocated.
57 * arg The arg field passed to uma_zalloc_arg
58 * size The size of the allocated item
59 * flags See zalloc flags
66 * The constructor is called just before the memory is returned
67 * to the user. It may block if necessary.
69 typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags);
75 * item A pointer to the memory which has been allocated.
76 * size The size of the item being destructed.
77 * arg Argument passed through uma_zfree_arg
83 * The destructor may perform operations that differ from those performed
84 * by the initializer, but it must leave the object in the same state.
85 * This IS type stable storage. This is called after EVERY zfree call.
87 typedef void (*uma_dtor)(void *mem, int size, void *arg);
93 * item A pointer to the memory which has been allocated.
94 * size The size of the item being initialized.
95 * flags See zalloc flags
102 * The initializer is called when the memory is cached in the uma zone.
103 * The initializer and the destructor should leave the object in the same
106 typedef int (*uma_init)(void *mem, int size, int flags);
109 * Item discard function
112 * item A pointer to memory which has been 'freed' but has not left the
114 * size The size of the item being discarded.
120 * This routine is called when memory leaves a zone and is returned to the
121 * system for other uses. It is the counter-part to the init function.
123 typedef void (*uma_fini)(void *mem, int size);
126 * Import new memory into a cache zone.
128 typedef int (*uma_import)(void *arg, void **store, int count, int domain,
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.
155 /* Function proto types */
158 * Create a new uma zone
161 * name The text name of the zone for debugging and stats. This memory
162 * should not be freed until the zone has been deallocated.
163 * size The size of the object that is being created.
164 * ctor The constructor that is called when the object is allocated.
165 * dtor The destructor that is called when the object is freed.
166 * init An initializer that sets up the initial state of the memory.
167 * fini A discard function that undoes initialization done by init.
168 * ctor/dtor/init/fini may all be null, see notes above.
169 * align A bitmask that corresponds to the requested alignment
171 * flags A set of parameters that control the behavior of the zone.
174 * A pointer to a structure which is intended to be opaque to users of
175 * the interface. The value may be null if the wait flag is not set.
177 uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor,
178 uma_dtor dtor, uma_init uminit, uma_fini fini,
179 int align, uint32_t flags);
182 * Create a secondary uma zone
185 * name The text name of the zone for debugging and stats. This memory
186 * should not be freed until the zone has been deallocated.
187 * ctor The constructor that is called when the object is allocated.
188 * dtor The destructor that is called when the object is freed.
189 * zinit An initializer that sets up the initial state of the memory
190 * as the object passes from the Keg's slab to the Zone's cache.
191 * zfini A discard function that undoes initialization done by init
192 * as the object passes from the Zone's cache to the Keg's slab.
194 * ctor/dtor/zinit/zfini may all be null, see notes above.
195 * Note that the zinit and zfini specified here are NOT
196 * exactly the same as the init/fini specified to uma_zcreate()
197 * when creating a primary zone. These zinit/zfini are called
198 * on the TRANSITION from keg to zone (and vice-versa). Once
199 * these are set, the primary zone may alter its init/fini
200 * (which are called when the object passes from VM to keg)
201 * using uma_zone_set_init/fini()) as well as its own
202 * zinit/zfini (unset by default for primary zone) with
203 * uma_zone_set_zinit/zfini() (note subtle 'z' prefix).
205 * primary A reference to this zone's Primary Zone which contains the
206 * backing Keg for the Secondary Zone being added.
209 * A pointer to a structure which is intended to be opaque to users of
210 * the interface. The value may be null if the wait flag is not set.
212 uma_zone_t uma_zsecond_create(const char *name, uma_ctor ctor, uma_dtor dtor,
213 uma_init zinit, uma_fini zfini, uma_zone_t primary);
216 * Create cache-only zones.
218 * This allows uma's per-cpu cache facilities to handle arbitrary
219 * pointers. Consumers must specify the import and release functions to
220 * fill and destroy caches. UMA does not allocate any memory for these
221 * zones. The 'arg' parameter is passed to import/release and is caller
224 uma_zone_t uma_zcache_create(const char *name, int size, uma_ctor ctor,
225 uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_import zimport,
226 uma_release zrelease, void *arg, int flags);
229 * Definitions for uma_zcreate flags
231 * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to
232 * overlap when adding new features.
234 #define UMA_ZONE_UNMANAGED 0x0001 /*
235 * Don't regulate the cache size, even
236 * under memory pressure.
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_CACHESPREAD 0x2000 /*
256 * Spread memory start locations across
257 * all possible cache lines. May
258 * require many virtually contiguous
259 * backend pages and can fail early.
261 #define UMA_ZONE_NODUMP 0x4000 /*
262 * Zone's pages will not be included in
265 #define UMA_ZONE_PCPU 0x8000 /*
266 * Allocates mp_maxid + 1 slabs of
269 #define UMA_ZONE_FIRSTTOUCH 0x10000 /* First touch NUMA policy */
270 #define UMA_ZONE_ROUNDROBIN 0x20000 /* Round-robin NUMA policy. */
271 #define UMA_ZONE_SMR 0x40000 /*
272 * Safe memory reclamation defers
273 * frees until all read sections
274 * have exited. This flag creates
275 * a unique SMR context for this
276 * zone. To share contexts see
277 * uma_zone_set_smr() below.
279 * See sys/smr.h for more details.
281 #define UMA_ZONE_NOKASAN 0x80000 /*
282 * Disable KASAN verification. This is
283 * implied by NOFREE. Cache zones are
284 * not verified by default.
286 /* In use by UMA_ZFLAGs: 0xffe00000 */
289 * These flags are shared between the keg and zone. Some are determined
290 * based on physical parameters of the request and may not be provided by
293 #define UMA_ZONE_INHERIT \
294 (UMA_ZONE_NOTOUCH | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \
295 UMA_ZONE_VM | UMA_ZONE_NOTPAGE | UMA_ZONE_PCPU | \
296 UMA_ZONE_FIRSTTOUCH | UMA_ZONE_ROUNDROBIN | UMA_ZONE_NOKASAN)
298 /* Definitions for align */
299 #define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */
300 #define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */
301 #define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */
302 #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */
303 #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */
304 #define UMA_ALIGN_CACHE (uma_get_cache_align_mask()) /* Cache line size align */
305 /* Align both to cache line size and an explicit alignment (through mask). */
306 #define UMA_ALIGN_CACHE_AND_MASK(mask) (uma_get_cache_align_mask() | (mask))
307 #define UMA_ALIGNOF(type) (_Alignof(type) - 1) /* Alignment fit for 'type' */
309 #define UMA_ANYDOMAIN -1 /* Special value for domain search. */
312 * Destroys an empty uma zone. If the zone is not empty uma complains loudly.
315 * zone The zone we want to destroy.
318 void uma_zdestroy(uma_zone_t zone);
321 * Allocates an item out of a zone
324 * zone The zone we are allocating from
325 * arg This data is passed to the ctor function
326 * flags See sys/malloc.h for available flags.
329 * A non-null pointer to an initialized element from the zone is
330 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer
331 * may be returned if the zone is empty or the ctor failed.
334 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
336 /* Allocate per-cpu data. Access the correct data with zpcpu_get(). */
337 void *uma_zalloc_pcpu_arg(uma_zone_t zone, void *arg, int flags);
339 /* Use with SMR zones. */
340 void *uma_zalloc_smr(uma_zone_t zone, int flags);
343 * Allocate an item from a specific NUMA domain. This uses a slow path in
344 * the allocator but is guaranteed to allocate memory from the requested
345 * domain if M_WAITOK is set.
348 * zone The zone we are allocating from
349 * arg This data is passed to the ctor function
350 * domain The domain to allocate from.
351 * flags See sys/malloc.h for available flags.
353 void *uma_zalloc_domain(uma_zone_t zone, void *arg, int domain, int flags);
356 * Allocates an item out of a zone without supplying an argument
358 * This is just a wrapper for uma_zalloc_arg for convenience.
361 static __inline void *uma_zalloc(uma_zone_t zone, int flags);
362 static __inline void *uma_zalloc_pcpu(uma_zone_t zone, int flags);
364 static __inline void *
365 uma_zalloc(uma_zone_t zone, int flags)
367 return uma_zalloc_arg(zone, NULL, flags);
370 static __inline void *
371 uma_zalloc_pcpu(uma_zone_t zone, int flags)
373 return uma_zalloc_pcpu_arg(zone, NULL, flags);
377 * Frees an item back into the specified zone.
380 * zone The zone the item was originally allocated out of.
381 * item The memory to be freed.
382 * arg Argument passed to the destructor
388 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
390 /* Use with PCPU zones. */
391 void uma_zfree_pcpu_arg(uma_zone_t zone, void *item, void *arg);
393 /* Use with SMR zones. */
394 void uma_zfree_smr(uma_zone_t zone, void *item);
397 * Frees an item back to a zone without supplying an argument
399 * This is just a wrapper for uma_zfree_arg for convenience.
402 static __inline void uma_zfree(uma_zone_t zone, void *item);
403 static __inline void uma_zfree_pcpu(uma_zone_t zone, void *item);
406 uma_zfree(uma_zone_t zone, void *item)
408 uma_zfree_arg(zone, item, NULL);
412 uma_zfree_pcpu(uma_zone_t zone, void *item)
414 uma_zfree_pcpu_arg(zone, item, NULL);
418 * Wait until the specified zone can allocate an item.
420 void uma_zwait(uma_zone_t zone);
423 * Backend page supplier routines
426 * zone The zone that is requesting pages.
427 * size The number of bytes being requested.
428 * pflag Flags for these memory pages, see below.
429 * domain The NUMA domain that we prefer for this allocation.
430 * wait Indicates our willingness to block.
433 * A pointer to the allocated memory or NULL on failure.
436 typedef void *(*uma_alloc)(uma_zone_t zone, vm_size_t size, int domain,
437 uint8_t *pflag, int wait);
440 * Backend page free routines
443 * item A pointer to the previously allocated pages.
444 * size The original size of the allocation.
445 * pflag The flags for the slab. See UMA_SLAB_* below.
450 typedef void (*uma_free)(void *item, vm_size_t size, uint8_t pflag);
453 * Reclaims unused memory. If no NUMA domain is specified, memory from all
454 * domains is reclaimed.
457 * req Reclamation request type.
458 * domain The target NUMA domain.
462 #define UMA_RECLAIM_DRAIN 1 /* release bucket cache */
463 #define UMA_RECLAIM_DRAIN_CPU 2 /* release bucket and per-CPU caches */
464 #define UMA_RECLAIM_TRIM 3 /* trim bucket cache to WSS */
465 void uma_reclaim(int req);
466 void uma_reclaim_domain(int req, int domain);
467 void uma_zone_reclaim(uma_zone_t, int req);
468 void uma_zone_reclaim_domain(uma_zone_t, int req, int domain);
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 * mask The alignment mask
480 void uma_set_cache_align_mask(unsigned int mask);
482 #include <vm/uma_align_mask.h>
485 * Set a reserved number of items to hold for M_USE_RESERVE allocations. All
486 * other requests must allocate new backing pages.
488 void uma_zone_reserve(uma_zone_t zone, int nitems);
491 * Reserves the maximum KVA space required by the zone and configures the zone
492 * to use a backend that allocates physical memory and maps it using the
496 * zone The zone to update.
497 * nitems The upper limit on the number of items that can be allocated.
500 * 0 if KVA space can not be allocated
504 * When the machine supports a direct map and the zone's items are smaller
505 * than a page, the zone will use the direct map instead of allocating KVA
508 int uma_zone_reserve_kva(uma_zone_t zone, int nitems);
511 * Sets an upper limit on the number of items allocated from a zone
514 * zone The zone to limit
515 * nitems The requested upper limit on the number of items allowed
518 * int The effective value of nitems
520 int uma_zone_set_max(uma_zone_t zone, int nitems);
523 * Sets an upper limit on the number of items allowed in zone's caches
526 * zone The zone to limit
527 * nitems The requested upper limit on the number of items allowed
529 void uma_zone_set_maxcache(uma_zone_t zone, int nitems);
532 * Obtains the effective limit on the number of items in a zone
535 * zone The zone to obtain the effective limit from
539 * int The effective limit of the zone
541 int uma_zone_get_max(uma_zone_t zone);
544 * Sets a warning to be printed when limit is reached
547 * zone The zone we will warn about
548 * warning Warning content
553 void uma_zone_set_warning(uma_zone_t zone, const char *warning);
556 * Sets a function to run when limit is reached
559 * zone The zone to which this applies
560 * fx The function ro run
565 typedef void (*uma_maxaction_t)(uma_zone_t, int);
566 void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t);
569 * Obtains the approximate current number of items allocated from a zone
572 * zone The zone to obtain the current allocation count from
575 * int The approximate current number of items allocated from the zone
577 int uma_zone_get_cur(uma_zone_t zone);
580 * The following two routines (uma_zone_set_init/fini)
581 * are used to set the backend init/fini pair which acts on an
582 * object as it becomes allocated and is placed in a slab within
583 * the specified zone's backing keg. These should probably not
584 * be changed once allocations have already begun, but only be set
585 * immediately upon zone creation.
587 void uma_zone_set_init(uma_zone_t zone, uma_init uminit);
588 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);
591 * The following two routines (uma_zone_set_zinit/zfini) are
592 * used to set the zinit/zfini pair which acts on an object as
593 * it passes from the backing Keg's slab cache to the
594 * specified Zone's bucket cache. These should probably not
595 * be changed once allocations have already begun, but only be set
596 * immediately upon zone creation.
598 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);
599 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);
602 * Replaces the standard backend allocator for this zone.
605 * zone The zone whose backend allocator is being changed.
606 * allocf A pointer to the allocation function
612 * This could be used to implement pageable allocation, or perhaps
613 * even DMA allocators if used in conjunction with the OFFPAGE
617 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);
620 * Used for freeing memory provided by the allocf above
623 * zone The zone that intends to use this free routine.
624 * freef The page freeing routine.
630 void uma_zone_set_freef(uma_zone_t zone, uma_free freef);
633 * Associate a zone with a smr context that is allocated after creation
634 * so that multiple zones may share the same context.
636 void uma_zone_set_smr(uma_zone_t zone, smr_t smr);
639 * Fetch the smr context that was set or made in uma_zcreate().
641 smr_t uma_zone_get_smr(uma_zone_t zone);
644 * These flags are settable in the allocf and visible in the freef.
646 #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */
647 #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kmem */
648 #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */
649 /* 0x02, 0x10, 0x40, and 0x80 are available */
652 * Used to pre-fill a zone with some number of items
655 * zone The zone to fill
656 * itemcnt The number of items to reserve
661 * NOTE: This is blocking and should only be done at startup
663 void uma_prealloc(uma_zone_t zone, int itemcnt);
666 * Used to determine if a fixed-size zone is exhausted.
669 * zone The zone to check
672 * Non-zero if zone is exhausted.
674 int uma_zone_exhausted(uma_zone_t zone);
677 * Returns the bytes of memory consumed by the zone.
679 size_t uma_zone_memory(uma_zone_t zone);
682 * Common UMA_ZONE_PCPU zones.
684 extern uma_zone_t pcpu_zone_4;
685 extern uma_zone_t pcpu_zone_8;
686 extern uma_zone_t pcpu_zone_16;
687 extern uma_zone_t pcpu_zone_32;
688 extern uma_zone_t pcpu_zone_64;
691 * Exported statistics structures to be used by user space monitoring tools.
692 * Statistics stream consists of a uma_stream_header, followed by a series of
693 * alternative uma_type_header and uma_type_stat structures.
695 #define UMA_STREAM_VERSION 0x00000001
696 struct uma_stream_header {
697 uint32_t ush_version; /* Stream format version. */
698 uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */
699 uint32_t ush_count; /* Number of records. */
700 uint32_t _ush_pad; /* Pad/reserved field. */
703 #define UTH_MAX_NAME 32
704 #define UTH_ZONE_SECONDARY 0x00000001
705 struct uma_type_header {
707 * Static per-zone data, some extracted from the supporting keg.
709 char uth_name[UTH_MAX_NAME];
710 uint32_t uth_align; /* Keg: alignment. */
711 uint32_t uth_size; /* Keg: requested size of item. */
712 uint32_t uth_rsize; /* Keg: real size of item. */
713 uint32_t uth_maxpages; /* Keg: maximum number of pages. */
714 uint32_t uth_limit; /* Keg: max items to allocate. */
717 * Current dynamic zone/keg-derived statistics.
719 uint32_t uth_pages; /* Keg: pages allocated. */
720 uint32_t uth_keg_free; /* Keg: items free. */
721 uint32_t uth_zone_free; /* Zone: items free. */
722 uint32_t uth_bucketsize; /* Zone: desired bucket size. */
723 uint32_t uth_zone_flags; /* Zone: flags. */
724 uint64_t uth_allocs; /* Zone: number of allocations. */
725 uint64_t uth_frees; /* Zone: number of frees. */
726 uint64_t uth_fails; /* Zone: number of alloc failures. */
727 uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */
728 uint64_t uth_xdomain; /* Zone: Number of cross domain frees. */
729 uint64_t _uth_reserved1[1]; /* Reserved. */
732 struct uma_percpu_stat {
733 uint64_t ups_allocs; /* Cache: number of allocations. */
734 uint64_t ups_frees; /* Cache: number of frees. */
735 uint64_t ups_cache_free; /* Cache: free items in cache. */
736 uint64_t _ups_reserved[5]; /* Reserved. */
739 void uma_reclaim_wakeup(void);
740 void uma_reclaim_worker(void *);
742 unsigned long uma_limit(void);
744 /* Return the amount of memory managed by UMA. */
745 unsigned long uma_size(void);
747 /* Return the amount of memory remaining. May be negative. */
748 long uma_avail(void);
750 #endif /* _VM_UMA_H_ */