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 * What's the difference between initializing and constructing?
129 * The item is initialized when it is cached, and this is the state that the
130 * object should be in when returned to the allocator. The purpose of this is
131 * to remove some code which would otherwise be called on each allocation by
132 * utilizing a known, stable state. This differs from the constructor which
133 * will be called on EVERY allocation.
135 * For example, in the initializer you may want to initialize embedded locks,
136 * NULL list pointers, set up initial states, magic numbers, etc. This way if
137 * the object is held in the allocator and re-used it won't be necessary to
140 * The constructor may be used to lock a data structure, link it on to lists,
141 * bump reference counts or total counts of outstanding structures, etc.
146 /* Function proto types */
149 * Create a new uma zone
152 * name The text name of the zone for debugging and stats. This memory
153 * should not be freed until the zone has been deallocated.
154 * size The size of the object that is being created.
155 * ctor The constructor that is called when the object is allocated.
156 * dtor The destructor that is called when the object is freed.
157 * init An initializer that sets up the initial state of the memory.
158 * fini A discard function that undoes initialization done by init.
159 * ctor/dtor/init/fini may all be null, see notes above.
160 * align A bitmask that corresponds to the requested alignment
162 * flags A set of parameters that control the behavior of the zone.
165 * A pointer to a structure which is intended to be opaque to users of
166 * the interface. The value may be null if the wait flag is not set.
168 uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor,
169 uma_dtor dtor, uma_init uminit, uma_fini fini,
170 int align, u_int32_t flags);
173 * Create a secondary uma zone
176 * name The text name of the zone for debugging and stats. This memory
177 * should not be freed until the zone has been deallocated.
178 * ctor The constructor that is called when the object is allocated.
179 * dtor The destructor that is called when the object is freed.
180 * zinit An initializer that sets up the initial state of the memory
181 * as the object passes from the Keg's slab to the Zone's cache.
182 * zfini A discard function that undoes initialization done by init
183 * as the object passes from the Zone's cache to the Keg's slab.
185 * ctor/dtor/zinit/zfini may all be null, see notes above.
186 * Note that the zinit and zfini specified here are NOT
187 * exactly the same as the init/fini specified to uma_zcreate()
188 * when creating a master zone. These zinit/zfini are called
189 * on the TRANSITION from keg to zone (and vice-versa). Once
190 * these are set, the primary zone may alter its init/fini
191 * (which are called when the object passes from VM to keg)
192 * using uma_zone_set_init/fini()) as well as its own
193 * zinit/zfini (unset by default for master zone) with
194 * uma_zone_set_zinit/zfini() (note subtle 'z' prefix).
196 * master A reference to this zone's Master Zone (Primary Zone),
197 * which contains the backing Keg for the Secondary Zone
201 * A pointer to a structure which is intended to be opaque to users of
202 * the interface. The value may be null if the wait flag is not set.
204 uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor,
205 uma_init zinit, uma_fini zfini, uma_zone_t master);
208 * Add a second master to a secondary zone. This provides multiple data
209 * backends for objects with the same size. Both masters must have
210 * compatible allocation flags. Presently, UMA_ZONE_MALLOC type zones are
211 * the only supported.
214 * Error on failure, 0 on success.
216 int uma_zsecond_add(uma_zone_t zone, uma_zone_t master);
219 * Definitions for uma_zcreate flags
221 * These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to
222 * overlap when adding new features. 0xf0000000 is in use by uma_int.h.
224 #define UMA_ZONE_PAGEABLE 0x0001 /* Return items not fully backed by
225 physical memory XXX Not yet */
226 #define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */
227 #define UMA_ZONE_STATIC 0x0004 /* Statically sized zone */
228 #define UMA_ZONE_OFFPAGE 0x0008 /* Force the slab structure allocation
229 off of the real memory */
230 #define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */
231 #define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */
232 #define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */
233 #define UMA_ZONE_VM 0x0080 /*
234 * Used for internal vm datastructures
237 #define UMA_ZONE_HASH 0x0100 /*
238 * Use a hash table instead of caching
239 * information in the vm_page.
241 #define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */
242 #define UMA_ZONE_REFCNT 0x0400 /* Allocate refcnts in slabs */
243 #define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets */
244 #define UMA_ZONE_CACHESPREAD 0x1000 /*
245 * Spread memory start locations across
246 * all possible cache lines. May
247 * require many virtually contiguous
248 * backend pages and can fail early.
250 #define UMA_ZONE_VTOSLAB 0x2000 /* Zone uses vtoslab for lookup. */
251 #define UMA_ZONE_NODUMP 0x4000 /*
252 * Zone's pages will not be included in
257 * These flags are shared between the keg and zone. In zones wishing to add
258 * new kegs these flags must be compatible. Some are determined based on
259 * physical parameters of the request and may not be provided by the consumer.
261 #define UMA_ZONE_INHERIT \
262 (UMA_ZONE_OFFPAGE | UMA_ZONE_MALLOC | UMA_ZONE_HASH | \
263 UMA_ZONE_REFCNT | UMA_ZONE_VTOSLAB)
265 /* Definitions for align */
266 #define UMA_ALIGN_PTR (sizeof(void *) - 1) /* Alignment fit for ptr */
267 #define UMA_ALIGN_LONG (sizeof(long) - 1) /* "" long */
268 #define UMA_ALIGN_INT (sizeof(int) - 1) /* "" int */
269 #define UMA_ALIGN_SHORT (sizeof(short) - 1) /* "" short */
270 #define UMA_ALIGN_CHAR (sizeof(char) - 1) /* "" char */
271 #define UMA_ALIGN_CACHE (0 - 1) /* Cache line size align */
274 * Destroys an empty uma zone. If the zone is not empty uma complains loudly.
277 * zone The zone we want to destroy.
280 void uma_zdestroy(uma_zone_t zone);
283 * Allocates an item out of a zone
286 * zone The zone we are allocating from
287 * arg This data is passed to the ctor function
288 * flags See sys/malloc.h for available flags.
291 * A non-null pointer to an initialized element from the zone is
292 * guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer
293 * may be returned if the zone is empty or the ctor failed.
296 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
299 * Allocates an item out of a zone without supplying an argument
301 * This is just a wrapper for uma_zalloc_arg for convenience.
304 static __inline void *uma_zalloc(uma_zone_t zone, int flags);
306 static __inline void *
307 uma_zalloc(uma_zone_t zone, int flags)
309 return uma_zalloc_arg(zone, NULL, flags);
313 * Frees an item back into the specified zone.
316 * zone The zone the item was originally allocated out of.
317 * item The memory to be freed.
318 * arg Argument passed to the destructor
324 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
327 * Frees an item back to a zone without supplying an argument
329 * This is just a wrapper for uma_zfree_arg for convenience.
332 static __inline void uma_zfree(uma_zone_t zone, void *item);
335 uma_zfree(uma_zone_t zone, void *item)
337 uma_zfree_arg(zone, item, NULL);
341 * XXX The rest of the prototypes in this header are h0h0 magic for the VM.
342 * If you think you need to use it for a normal zone you're probably incorrect.
346 * Backend page supplier routines
349 * zone The zone that is requesting pages.
350 * size The number of bytes being requested.
351 * pflag Flags for these memory pages, see below.
352 * wait Indicates our willingness to block.
355 * A pointer to the allocated memory or NULL on failure.
358 typedef void *(*uma_alloc)(uma_zone_t zone, int size, u_int8_t *pflag, int wait);
361 * Backend page free routines
364 * item A pointer to the previously allocated pages.
365 * size The original size of the allocation.
366 * pflag The flags for the slab. See UMA_SLAB_* below.
371 typedef void (*uma_free)(void *item, int size, u_int8_t pflag);
376 * Sets up the uma allocator. (Called by vm_mem_init)
379 * bootmem A pointer to memory used to bootstrap the system.
385 * This memory is used for zones which allocate things before the
386 * backend page supplier can give us pages. It should be
387 * UMA_SLAB_SIZE * boot_pages bytes. (see uma_int.h)
391 void uma_startup(void *bootmem, int boot_pages);
394 * Finishes starting up the allocator. This should
395 * be called when kva is ready for normal allocs.
404 * uma_startup2 is called by kmeminit() to enable us of uma for malloc.
407 void uma_startup2(void);
410 * Reclaims unused memory for all zones
417 * This should only be called by the page out daemon.
420 void uma_reclaim(void);
423 * Sets the alignment mask to be used for all zones requesting cache
424 * alignment. Should be called by MD boot code prior to starting VM/UMA.
427 * align The alignment mask
432 void uma_set_align(int align);
435 * Switches the backing object of a zone
438 * zone The zone to update.
439 * obj The VM object to use for future allocations.
440 * size The size of the object to allocate.
443 * 0 if kva space can not be allocated
447 * A NULL object can be used and uma will allocate one for you. Setting
448 * the size will limit the amount of memory allocated to this zone.
452 int uma_zone_set_obj(uma_zone_t zone, struct vm_object *obj, int size);
455 * Sets a high limit on the number of items allowed in a zone
458 * zone The zone to limit
459 * nitems The requested upper limit on the number of items allowed
462 * int The effective value of nitems after rounding up based on page size
464 int uma_zone_set_max(uma_zone_t zone, int nitems);
467 * Obtains the effective limit on the number of items in a zone
470 * zone The zone to obtain the effective limit from
474 * int The effective limit of the zone
476 int uma_zone_get_max(uma_zone_t zone);
479 * Obtains the approximate current number of items allocated from a zone
482 * zone The zone to obtain the current allocation count from
485 * int The approximate current number of items allocated from the zone
487 int uma_zone_get_cur(uma_zone_t zone);
490 * The following two routines (uma_zone_set_init/fini)
491 * are used to set the backend init/fini pair which acts on an
492 * object as it becomes allocated and is placed in a slab within
493 * the specified zone's backing keg. These should probably not
494 * be changed once allocations have already begun, but only be set
495 * immediately upon zone creation.
497 void uma_zone_set_init(uma_zone_t zone, uma_init uminit);
498 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);
501 * The following two routines (uma_zone_set_zinit/zfini) are
502 * used to set the zinit/zfini pair which acts on an object as
503 * it passes from the backing Keg's slab cache to the
504 * specified Zone's bucket cache. These should probably not
505 * be changed once allocations have already begun, but only be set
506 * immediately upon zone creation.
508 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);
509 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);
512 * Replaces the standard page_alloc or obj_alloc functions for this zone
515 * zone The zone whose backend allocator is being changed.
516 * allocf A pointer to the allocation function
522 * This could be used to implement pageable allocation, or perhaps
523 * even DMA allocators if used in conjunction with the OFFPAGE
527 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);
530 * Used for freeing memory provided by the allocf above
533 * zone The zone that intends to use this free routine.
534 * freef The page freeing routine.
540 void uma_zone_set_freef(uma_zone_t zone, uma_free freef);
543 * These flags are setable in the allocf and visible in the freef.
545 #define UMA_SLAB_BOOT 0x01 /* Slab alloced from boot pages */
546 #define UMA_SLAB_KMEM 0x02 /* Slab alloced from kmem_map */
547 #define UMA_SLAB_KERNEL 0x04 /* Slab alloced from kernel_map */
548 #define UMA_SLAB_PRIV 0x08 /* Slab alloced from priv allocator */
549 #define UMA_SLAB_OFFP 0x10 /* Slab is managed separately */
550 #define UMA_SLAB_MALLOC 0x20 /* Slab is a large malloc slab */
551 /* 0x40 and 0x80 are available */
554 * Used to pre-fill a zone with some number of items
557 * zone The zone to fill
558 * itemcnt The number of items to reserve
563 * NOTE: This is blocking and should only be done at startup
565 void uma_prealloc(uma_zone_t zone, int itemcnt);
568 * Used to lookup the reference counter allocated for an item
569 * from a UMA_ZONE_REFCNT zone. For UMA_ZONE_REFCNT zones,
570 * reference counters are allocated for items and stored in
571 * the underlying slab header.
574 * zone The UMA_ZONE_REFCNT zone to which the item belongs.
575 * item The address of the item for which we want a refcnt.
578 * A pointer to a u_int32_t reference counter.
580 u_int32_t *uma_find_refcnt(uma_zone_t zone, void *item);
583 * Used to determine if a fixed-size zone is exhausted.
586 * zone The zone to check
589 * Non-zero if zone is exhausted.
591 int uma_zone_exhausted(uma_zone_t zone);
592 int uma_zone_exhausted_nolock(uma_zone_t zone);
595 * Exported statistics structures to be used by user space monitoring tools.
596 * Statistics stream consists of a uma_stream_header, followed by a series of
597 * alternative uma_type_header and uma_type_stat structures.
599 #define UMA_STREAM_VERSION 0x00000001
600 struct uma_stream_header {
601 u_int32_t ush_version; /* Stream format version. */
602 u_int32_t ush_maxcpus; /* Value of MAXCPU for stream. */
603 u_int32_t ush_count; /* Number of records. */
604 u_int32_t _ush_pad; /* Pad/reserved field. */
607 #define UTH_MAX_NAME 32
608 #define UTH_ZONE_SECONDARY 0x00000001
609 struct uma_type_header {
611 * Static per-zone data, some extracted from the supporting keg.
613 char uth_name[UTH_MAX_NAME];
614 u_int32_t uth_align; /* Keg: alignment. */
615 u_int32_t uth_size; /* Keg: requested size of item. */
616 u_int32_t uth_rsize; /* Keg: real size of item. */
617 u_int32_t uth_maxpages; /* Keg: maximum number of pages. */
618 u_int32_t uth_limit; /* Keg: max items to allocate. */
621 * Current dynamic zone/keg-derived statistics.
623 u_int32_t uth_pages; /* Keg: pages allocated. */
624 u_int32_t uth_keg_free; /* Keg: items free. */
625 u_int32_t uth_zone_free; /* Zone: items free. */
626 u_int32_t uth_bucketsize; /* Zone: desired bucket size. */
627 u_int32_t uth_zone_flags; /* Zone: flags. */
628 u_int64_t uth_allocs; /* Zone: number of allocations. */
629 u_int64_t uth_frees; /* Zone: number of frees. */
630 u_int64_t uth_fails; /* Zone: number of alloc failures. */
631 u_int64_t uth_sleeps; /* Zone: number of alloc sleeps. */
632 u_int64_t _uth_reserved1[2]; /* Reserved. */
635 struct uma_percpu_stat {
636 u_int64_t ups_allocs; /* Cache: number of allocations. */
637 u_int64_t ups_frees; /* Cache: number of frees. */
638 u_int64_t ups_cache_free; /* Cache: free items in cache. */
639 u_int64_t _ups_reserved[5]; /* Reserved. */