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 flags);
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 * Add a second master to a secondary zone. This provides multiple data
221 * backends for objects with the same size. Both masters must have
222 * compatible allocation flags. Presently, UMA_ZONE_MALLOC type zones are
223 * the only supported.
226 * Error on failure, 0 on success.
228 int uma_zsecond_add(uma_zone_t zone, uma_zone_t master);
231 * Create cache-only zones.
233 * This allows uma's per-cpu cache facilities to handle arbitrary
234 * pointers. Consumers must specify the import and release functions to
235 * fill and destroy caches. UMA does not allocate any memory for these
236 * zones. The 'arg' parameter is passed to import/release and is caller
239 uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor,
240 uma_init zinit, uma_fini zfini, uma_import zimport,
241 uma_release zrelease, void *arg, int flags);
244 * Definitions for uma_zcreate flags
246 * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to
247 * overlap when adding new features. 0xff000000 is in use by uma_int.h.
249 #define UMA_ZONE_PAGEABLE 0x0001 /* Return items not fully backed by
250 physical memory XXX Not yet */
251 #define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */
252 #define UMA_ZONE_STATIC 0x0004 /* Statically sized zone */
253 #define UMA_ZONE_OFFPAGE 0x0008 /* Force the slab structure allocation
254 off of the real memory */
255 #define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */
256 #define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */
257 #define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */
258 #define UMA_ZONE_VM 0x0080 /*
259 * Used for internal vm datastructures
262 #define UMA_ZONE_HASH 0x0100 /*
263 * Use a hash table instead of caching
264 * information in the vm_page.
266 #define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */
268 #define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets */
269 #define UMA_ZONE_CACHESPREAD 0x1000 /*
270 * Spread memory start locations across
271 * all possible cache lines. May
272 * require many virtually contiguous
273 * backend pages and can fail early.
275 #define UMA_ZONE_VTOSLAB 0x2000 /* Zone uses vtoslab for lookup. */
276 #define UMA_ZONE_NODUMP 0x4000 /*
277 * Zone's pages will not be included in
280 #define UMA_ZONE_PCPU 0x8000 /*
281 * Allocates mp_maxid + 1 slabs sized to
282 * sizeof(struct pcpu).
286 * These flags are shared between the keg and zone. In zones wishing to add
287 * new kegs these flags must be compatible. Some are determined based on
288 * physical parameters of the request and may not be provided by the consumer.
290 #define UMA_ZONE_INHERIT \
291 (UMA_ZONE_OFFPAGE | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \
292 UMA_ZONE_HASH | UMA_ZONE_VTOSLAB | UMA_ZONE_PCPU)
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' */
304 * Destroys an empty uma zone. If the zone is not empty uma complains loudly.
307 * zone The zone we want to destroy.
310 void uma_zdestroy(uma_zone_t zone);
313 * Allocates an item out of a zone
316 * zone The zone we are allocating from
317 * arg This data is passed to the ctor function
318 * flags See sys/malloc.h for available flags.
321 * A non-null pointer to an initialized element from the zone is
322 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer
323 * may be returned if the zone is empty or the ctor failed.
326 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
329 * Allocates an item out of a zone without supplying an argument
331 * This is just a wrapper for uma_zalloc_arg for convenience.
334 static __inline void *uma_zalloc(uma_zone_t zone, int flags);
336 static __inline void *
337 uma_zalloc(uma_zone_t zone, int flags)
339 return uma_zalloc_arg(zone, NULL, flags);
343 * Frees an item back into the specified zone.
346 * zone The zone the item was originally allocated out of.
347 * item The memory to be freed.
348 * arg Argument passed to the destructor
354 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
357 * Frees an item back to a zone without supplying an argument
359 * This is just a wrapper for uma_zfree_arg for convenience.
362 static __inline void uma_zfree(uma_zone_t zone, void *item);
365 uma_zfree(uma_zone_t zone, void *item)
367 uma_zfree_arg(zone, item, NULL);
371 * Wait until the specified zone can allocate an item.
373 void uma_zwait(uma_zone_t zone);
376 * XXX The rest of the prototypes in this header are h0h0 magic for the VM.
377 * If you think you need to use it for a normal zone you're probably incorrect.
381 * Backend page supplier routines
384 * zone The zone that is requesting pages.
385 * size The number of bytes being requested.
386 * pflag Flags for these memory pages, see below.
387 * wait Indicates our willingness to block.
390 * A pointer to the allocated memory or NULL on failure.
393 typedef void *(*uma_alloc)(uma_zone_t zone, vm_size_t size, uint8_t *pflag,
397 * Backend page free routines
400 * item A pointer to the previously allocated pages.
401 * size The original size of the allocation.
402 * pflag The flags for the slab. See UMA_SLAB_* below.
407 typedef void (*uma_free)(void *item, vm_size_t size, uint8_t pflag);
412 * Sets up the uma allocator. (Called by vm_mem_init)
415 * bootmem A pointer to memory used to bootstrap the system.
421 * This memory is used for zones which allocate things before the
422 * backend page supplier can give us pages. It should be
423 * UMA_SLAB_SIZE * boot_pages bytes. (see uma_int.h)
427 void uma_startup(void *bootmem, int boot_pages);
430 * Finishes starting up the allocator. This should
431 * be called when kva is ready for normal allocs.
440 * uma_startup2 is called by kmeminit() to enable us of uma for malloc.
443 void uma_startup2(void);
446 * Reclaims unused memory for all zones
453 * This should only be called by the page out daemon.
456 void uma_reclaim(void);
459 * Sets the alignment mask to be used for all zones requesting cache
460 * alignment. Should be called by MD boot code prior to starting VM/UMA.
463 * align The alignment mask
468 void uma_set_align(int align);
471 * Set a reserved number of items to hold for M_USE_RESERVE allocations. All
472 * other requests must allocate new backing pages.
474 void uma_zone_reserve(uma_zone_t zone, int nitems);
477 * Reserves the maximum KVA space required by the zone and configures the zone
478 * to use a VM_ALLOC_NOOBJ-based backend allocator.
481 * zone The zone to update.
482 * nitems The upper limit on the number of items that can be allocated.
485 * 0 if KVA space can not be allocated
489 * When the machine supports a direct map and the zone's items are smaller
490 * than a page, the zone will use the direct map instead of allocating KVA
493 int uma_zone_reserve_kva(uma_zone_t zone, int nitems);
496 * Sets a high limit on the number of items allowed in a zone
499 * zone The zone to limit
500 * nitems The requested upper limit on the number of items allowed
503 * int The effective value of nitems after rounding up based on page size
505 int uma_zone_set_max(uma_zone_t zone, int nitems);
508 * Obtains the effective limit on the number of items in a zone
511 * zone The zone to obtain the effective limit from
515 * int The effective limit of the zone
517 int uma_zone_get_max(uma_zone_t zone);
520 * Sets a warning to be printed when limit is reached
523 * zone The zone we will warn about
524 * warning Warning content
529 void uma_zone_set_warning(uma_zone_t zone, const char *warning);
532 * Sets a function to run when limit is reached
535 * zone The zone to which this applies
536 * fx The function ro run
541 typedef void (*uma_maxaction_t)(uma_zone_t, int);
542 void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t);
545 * Obtains the approximate current number of items allocated from a zone
548 * zone The zone to obtain the current allocation count from
551 * int The approximate current number of items allocated from the zone
553 int uma_zone_get_cur(uma_zone_t zone);
556 * The following two routines (uma_zone_set_init/fini)
557 * are used to set the backend init/fini pair which acts on an
558 * object as it becomes allocated and is placed in a slab within
559 * the specified zone's backing keg. These should probably not
560 * be changed once allocations have already begun, but only be set
561 * immediately upon zone creation.
563 void uma_zone_set_init(uma_zone_t zone, uma_init uminit);
564 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);
567 * The following two routines (uma_zone_set_zinit/zfini) are
568 * used to set the zinit/zfini pair which acts on an object as
569 * it passes from the backing Keg's slab cache to the
570 * specified Zone's bucket cache. These should probably not
571 * be changed once allocations have already begun, but only be set
572 * immediately upon zone creation.
574 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);
575 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);
578 * Replaces the standard backend allocator for this zone.
581 * zone The zone whose backend allocator is being changed.
582 * allocf A pointer to the allocation function
588 * This could be used to implement pageable allocation, or perhaps
589 * even DMA allocators if used in conjunction with the OFFPAGE
593 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);
596 * Used for freeing memory provided by the allocf above
599 * zone The zone that intends to use this free routine.
600 * freef The page freeing routine.
606 void uma_zone_set_freef(uma_zone_t zone, uma_free freef);
609 * These flags are setable in the allocf and visible in the freef.
611 #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */
612 #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kernel_map */
613 #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */
614 #define UMA_SLAB_OFFP 0x10 /* Slab is managed separately */
615 #define UMA_SLAB_MALLOC 0x20 /* Slab is a large malloc slab */
616 /* 0x02, 0x40 and 0x80 are available */
619 * Used to pre-fill a zone with some number of items
622 * zone The zone to fill
623 * itemcnt The number of items to reserve
628 * NOTE: This is blocking and should only be done at startup
630 void uma_prealloc(uma_zone_t zone, int itemcnt);
633 * Used to determine if a fixed-size zone is exhausted.
636 * zone The zone to check
639 * Non-zero if zone is exhausted.
641 int uma_zone_exhausted(uma_zone_t zone);
642 int uma_zone_exhausted_nolock(uma_zone_t zone);
645 * Common UMA_ZONE_PCPU zones.
647 extern uma_zone_t pcpu_zone_64;
648 extern uma_zone_t pcpu_zone_ptr;
651 * Exported statistics structures to be used by user space monitoring tools.
652 * Statistics stream consists of a uma_stream_header, followed by a series of
653 * alternative uma_type_header and uma_type_stat structures.
655 #define UMA_STREAM_VERSION 0x00000001
656 struct uma_stream_header {
657 uint32_t ush_version; /* Stream format version. */
658 uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */
659 uint32_t ush_count; /* Number of records. */
660 uint32_t _ush_pad; /* Pad/reserved field. */
663 #define UTH_MAX_NAME 32
664 #define UTH_ZONE_SECONDARY 0x00000001
665 struct uma_type_header {
667 * Static per-zone data, some extracted from the supporting keg.
669 char uth_name[UTH_MAX_NAME];
670 uint32_t uth_align; /* Keg: alignment. */
671 uint32_t uth_size; /* Keg: requested size of item. */
672 uint32_t uth_rsize; /* Keg: real size of item. */
673 uint32_t uth_maxpages; /* Keg: maximum number of pages. */
674 uint32_t uth_limit; /* Keg: max items to allocate. */
677 * Current dynamic zone/keg-derived statistics.
679 uint32_t uth_pages; /* Keg: pages allocated. */
680 uint32_t uth_keg_free; /* Keg: items free. */
681 uint32_t uth_zone_free; /* Zone: items free. */
682 uint32_t uth_bucketsize; /* Zone: desired bucket size. */
683 uint32_t uth_zone_flags; /* Zone: flags. */
684 uint64_t uth_allocs; /* Zone: number of allocations. */
685 uint64_t uth_frees; /* Zone: number of frees. */
686 uint64_t uth_fails; /* Zone: number of alloc failures. */
687 uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */
688 uint64_t _uth_reserved1[2]; /* Reserved. */
691 struct uma_percpu_stat {
692 uint64_t ups_allocs; /* Cache: number of allocations. */
693 uint64_t ups_frees; /* Cache: number of frees. */
694 uint64_t ups_cache_free; /* Cache: free items in cache. */
695 uint64_t _ups_reserved[5]; /* Reserved. */
698 void uma_reclaim_wakeup(void);
699 void uma_reclaim_worker(void *);
701 #endif /* _VM_UMA_H_ */