2 .\" Copyright (c) 2001 Dag-Erling Coïdan Smørgrav
3 .\" All rights reserved.
5 .\" Redistribution and use in source and binary forms, with or without
6 .\" modification, are permitted provided that the following conditions
8 .\" 1. Redistributions of source code must retain the above copyright
9 .\" notice, this list of conditions and the following disclaimer.
10 .\" 2. Redistributions in binary form must reproduce the above copyright
11 .\" notice, this list of conditions and the following disclaimer in the
12 .\" documentation and/or other materials provided with the distribution.
14 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 .Nd general-purpose kernel object allocator
38 .Cd "options UMA_FIRSTTOUCH"
39 .Cd "options UMA_XDOMAIN"
41 typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags);
42 typedef void (*uma_dtor)(void *mem, int size, void *arg);
43 typedef int (*uma_init)(void *mem, int size, int flags);
44 typedef void (*uma_fini)(void *mem, int size);
45 typedef int (*uma_import)(void *arg, void **store, int count, int domain,
47 typedef void (*uma_release)(void *arg, void **store, int count);
48 typedef void *(*uma_alloc)(uma_zone_t zone, vm_size_t size, int domain,
49 uint8_t *pflag, int wait);
50 typedef void (*uma_free)(void *item, vm_size_t size, uint8_t pflag);
55 .Fa "char *name" "int size"
56 .Fa "uma_ctor ctor" "uma_dtor dtor" "uma_init zinit" "uma_fini zfini"
57 .Fa "int align" "uint16_t flags"
61 .Fa "char *name" "int size"
62 .Fa "uma_ctor ctor" "uma_dtor dtor" "uma_init zinit" "uma_fini zfini"
63 .Fa "uma_import zimport" "uma_release zrelease"
64 .Fa "void *arg" "int flags"
67 .Fo uma_zsecond_create
69 .Fa "uma_ctor ctor" "uma_dtor dtor" "uma_init zinit" "uma_fini zfini"
70 .Fa "uma_zone_t master"
73 .Fn uma_zdestroy "uma_zone_t zone"
75 .Fn uma_zalloc "uma_zone_t zone" "int flags"
77 .Fn uma_zalloc_arg "uma_zone_t zone" "void *arg" "int flags"
79 .Fn uma_zalloc_domain "uma_zone_t zone" "void *arg" "int domain" "int flags"
81 .Fn uma_zalloc_pcpu "uma_zone_t zone" "int flags"
83 .Fn uma_zalloc_pcpu_arg "uma_zone_t zone" "void *arg" "int flags"
85 .Fn uma_zfree "uma_zone_t zone" "void *item"
87 .Fn uma_zfree_arg "uma_zone_t zone" "void *item" "void *arg"
89 .Fn uma_zfree_pcpu "uma_zone_t zone" "void *item"
91 .Fn uma_zfree_pcpu_arg "uma_zone_t zone" "void *item" "void *arg"
93 .Fn uma_prealloc "uma_zone_t zone" "int nitems"
95 .Fn uma_zone_reserve "uma_zone_t zone" "int nitems"
97 .Fn uma_zone_reserve_kva "uma_zone_t zone" "int nitems"
99 .Fn uma_reclaim "int req"
101 .Fn uma_zone_reclaim "uma_zone_t zone" "int req"
103 .Fn uma_zone_set_allocf "uma_zone_t zone" "uma_alloc allocf"
105 .Fn uma_zone_set_freef "uma_zone_t zone" "uma_free freef"
107 .Fn uma_zone_set_max "uma_zone_t zone" "int nitems"
109 .Fn uma_zone_set_maxcache "uma_zone_t zone" "int nitems"
111 .Fn uma_zone_get_max "uma_zone_t zone"
113 .Fn uma_zone_get_cur "uma_zone_t zone"
115 .Fn uma_zone_set_warning "uma_zone_t zone" "const char *warning"
117 .Fn uma_zone_set_maxaction "uma_zone_t zone" "void (*maxaction)(uma_zone_t)"
121 .Fn SYSCTL_UMA_MAX parent nbr name access zone descr
122 .Fn SYSCTL_ADD_UMA_MAX ctx parent nbr name access zone descr
123 .Fn SYSCTL_UMA_CUR parent nbr name access zone descr
124 .Fn SYSCTL_ADD_UMA_CUR ctx parent nbr name access zone descr
126 UMA (Universal Memory Allocator) provides an efficient interface for managing
127 dynamically-sized collections of items of identical size, referred to as zones.
128 Zones keep track of which items are in use and which
129 are not, and UMA provides functions for allocating items from a zone and
130 for releasing them back, making them available for subsequent allocation requests.
131 Zones maintain per-CPU caches with linear scalability on SMP
132 systems as well as round-robin and first-touch policies for NUMA
134 The number of items cached per CPU is bounded, and each zone additionally
135 maintains an unbounded cache of items that is used to quickly satisfy
136 per-CPU cache allocation misses.
138 Two types of zones exist: regular zones and cache zones.
139 In a regular zone, items are allocated from a slab, which is one or more
140 virtually contiguous memory pages that have been allocated from the kernel's
142 Internally, slabs are managed by a UMA keg, which is responsible for allocating
143 slabs and keeping track of their usage by one or more zones.
144 In typical usage, there is one keg per zone, so slabs are not shared among
147 Normal zones import items from a keg, and release items back to that keg if
149 Cache zones do not have a keg, and instead use custom import and release
151 For example, some collections of kernel objects are statically allocated
152 at boot-time, and the size of the collection does not change.
153 A cache zone can be used to implement an efficient allocator for the objects in
159 .Fn uma_zcache_create
160 functions create a new regular zone and cache zone, respectively.
162 .Fn uma_zsecond_create
163 function creates a regular zone which shares the keg of the zone
169 argument is a text name of the zone for debugging and stats; this memory
170 should not be freed until the zone has been deallocated.
176 arguments are callback functions that are called by
177 the UMA subsystem at the time of the call to
182 Their purpose is to provide hooks for initializing or
183 destroying things that need to be done at the time of the allocation
184 or release of a resource.
189 callbacks might be to initialize a data structure embedded in the item,
198 arguments are used to optimize the allocation of items from the zone.
199 They are called by the UMA subsystem whenever
200 it needs to allocate or free items to satisfy requests or memory pressure.
205 callbacks might be to
206 initialize and destroy a mutex contained within an item.
207 This would allow one to avoid destroying and re-initializing the mutex
208 each time the item is freed and re-allocated.
209 They are not called on each call to
213 but rather when an item is imported into a zone's cache, and when a zone
214 releases an item to the slab allocator, typically as a response to memory
218 .Fn uma_zcache_create ,
223 functions are called to import items into the zone and to release items
224 from the zone, respectively.
227 function should store pointers to items in the
229 array, which contains a maximum of
232 The function must return the number of imported items, which may be less than
238 function contains an array of
244 .Fn uma_zcache_create
245 is provided to the import and release functions.
250 specifies the requested
252 domain for the allocation.
253 It is either a NUMA domain number or the special value
261 .Fn uma_zcache_create
262 is a subset of the following flags:
263 .Bl -tag -width "foo"
264 .It Dv UMA_ZONE_NOFREE
265 Slabs allocated to the zone's keg are never freed.
266 .It Dv UMA_ZONE_NODUMP
267 Pages belonging to the zone will not be included in minidumps.
269 An allocation from zone would have
271 shadow copies, that are privately assigned to CPUs.
272 A CPU can address its private copy using base the allocation address plus
273 a multiple of the current CPU ID and
274 .Fn sizeof "struct pcpu" :
275 .Bd -literal -offset indent
276 foo_zone = uma_zcreate(..., UMA_ZONE_PCPU);
278 foo_base = uma_zalloc(foo_zone, ...);
281 foo_pcpu = (foo_t *)zpcpu_get(foo_base);
282 /* do something with foo_pcpu */
288 cannot be used when allocating items from a PCPU zone.
289 To obtain zeroed memory from a PCPU zone, use the
291 function and its variants instead, and pass
293 .It Dv UMA_ZONE_NOTOUCH
294 The UMA subsystem may not directly touch (i.e. read or write) the slab memory.
295 Otherwise, by default, book-keeping of items within a slab may be done in the
296 slab page itself, and
298 kernels may also do use-after-free checking by accessing the slab memory.
299 .It Dv UMA_ZONE_ZINIT
300 The zone will have its
302 method set to internal method that initializes a new allocated slab
310 flag would not return zeroed memory on every
312 .It Dv UMA_ZONE_NOTPAGE
313 An allocator function will be supplied with
314 .Fn uma_zone_set_allocf
315 and the memory that it returns may not be kernel virtual memory backed by VM
316 pages in the page array.
317 .It Dv UMA_ZONE_MALLOC
322 The zone is for the VM subsystem.
324 The zone should use a first-touch NUMA policy rather than the round-robin
328 kernel option is configured, all zones implicitly use a first-touch policy,
334 kernel option, when configured, causes UMA to do the extra tracking to ensure
335 that allocations from first-touch zones are always local.
336 Otherwise, consumers that do not free memory on the same domain from which it
337 was allocated will cause mixing in per-CPU caches.
341 .It Dv UMA_ZONE_CONTIG
342 Items in this zone must be contiguous in physical address space.
343 Items will follow normal alignment constraints and may span page boundaries
344 between pages with contiguous physical addresses.
347 Zones can be destroyed using
349 freeing all memory that is cached in the zone.
350 All items allocated from the zone must be freed to the zone before the zone
351 may be safely destroyed.
353 To allocate an item from a zone, simply call
355 with a pointer to that zone and set the
357 argument to selected flags as documented in
359 It will return a pointer to an item if successful, or
361 in the rare case where all items in the zone are in use and the
362 allocator is unable to grow the zone and
366 Items are released back to the zone from which they were allocated by
369 with a pointer to the zone and a pointer to the item.
383 specify an argument for the
387 functions of the zone, respectively.
389 .Fn uma_zalloc_domain
390 function allows callers to specify a fixed
392 domain to allocate from.
393 This uses a guaranteed but slow path in the allocator which reduces
398 function allocates slabs for the requested number of items, typically following
399 the initial creation of a zone.
400 Subsequent allocations from the zone will be satisfied using the pre-allocated
402 Note that slab allocation is performed with the
410 function sets the number of reserved items for the zone.
412 and variants will ensure that the zone contains at least the reserved number
414 Reserved items may be allocated by specifying
416 in the allocation request flags.
418 does not perform any pre-allocation by itself.
421 .Fn uma_zone_reserve_kva
422 function pre-allocates kernel virtual address space for the requested
424 Subsequent allocations from the zone will be satisfied using the pre-allocated
427 .Fn uma_zone_reserve ,
428 .Fn uma_zone_reserve_kva
429 does not restrict the use of the pre-allocation to
437 functions reclaim cached items from UMA zones, releasing unused memory.
440 function reclaims items from all regular zones, while
442 reclaims items only from the specified zone.
445 parameter must be one of three values which specify how aggressively
446 items are to be reclaimed:
447 .Bl -tag -width indent
448 .It Dv UMA_RECLAIM_TRIM
449 Reclaim items only in excess of the zone's estimated working set size.
450 The working set size is periodically updated and tracks the recent history
452 .It Dv UMA_RECLAIM_DRAIN
453 Reclaim all items from the unbounded cache.
454 Free items in the per-CPU caches are left alone.
455 .It Dv UMA_RECLAIM_DRAIN_CPU
456 Reclaim all cached items.
460 .Fn uma_zone_set_allocf
462 .Fn uma_zone_set_freef
463 functions allow a zone's default slab allocation and free functions to be
465 This is useful if memory with special constraints such as attributes,
466 alignment, or address ranges must be used.
470 function limits the number of items
471 .Pq and therefore memory
472 that can be allocated to
476 argument specifies the requested upper limit number of items.
477 The effective limit is returned to the caller, as it may end up being higher
478 than requested due to the implementation rounding up to ensure all memory pages
479 allocated to the zone are utilised to capacity.
480 The limit applies to the total number of items in the zone, which includes
481 allocated items, free items and free items in the per-cpu caches.
482 On systems with more than one CPU it may not be possible to allocate
483 the specified number of items even when there is no shortage of memory,
484 because all of the remaining free items may be in the caches of the
485 other CPUs when the limit is hit.
488 .Fn uma_zone_set_maxcache
489 function limits the number of free items which may be cached in the zone.
490 This limit applies to both the per-CPU caches and the cache of free buckets.
494 function returns the effective upper limit number of items for a zone.
498 function returns an approximation of the number of items currently allocated
500 The returned value is approximate because appropriate synchronisation to
501 determine an exact value is not performed by the implementation.
502 This ensures low overhead at the expense of potentially stale data being used
506 .Fn uma_zone_set_warning
507 function sets a warning that will be printed on the system console when the
508 given zone becomes full and fails to allocate an item.
509 The warning will be printed no more often than every five minutes.
510 Warnings can be turned off globally by setting the
516 .Fn uma_zone_set_maxaction
517 function sets a function that will be called when the given zone becomes full
518 and fails to allocate an item.
519 The function will be called with the zone locked.
521 that called the allocation function may have held additional locks.
523 this function should do very little work (similar to a signal handler).
526 .Fn SYSCTL_UMA_MAX parent nbr name access zone descr
527 macro declares a static
529 oid that exports the effective upper limit number of items for a zone.
532 argument should be a pointer to
534 A read of the oid returns value obtained through
535 .Fn uma_zone_get_max .
536 A write to the oid sets new value via
537 .Fn uma_zone_set_max .
539 .Fn SYSCTL_ADD_UMA_MAX ctx parent nbr name access zone descr
540 macro is provided to create this type of oid dynamically.
543 .Fn SYSCTL_UMA_CUR parent nbr name access zone descr
544 macro declares a static read-only
546 oid that exports the approximate current occupancy of the zone.
549 argument should be a pointer to
551 A read of the oid returns value obtained through
552 .Fn uma_zone_get_cur .
554 .Fn SYSCTL_ADD_UMA_CUR ctx parent nbr name zone descr
555 macro is provided to create this type of oid dynamically.
556 .Sh IMPLEMENTATION NOTES
557 The memory that these allocation calls return is not executable.
560 function does not support the
562 flag to allocate executable memory.
563 Not all platforms enforce a distinction between executable and
564 non-executable memory.
571 .%T "The Slab Allocator: An Object-Caching Kernel Memory Allocator"
575 The zone allocator first appeared in
577 It was radically changed in
579 to function as a slab allocator.
582 The zone allocator was written by
584 The zone allocator was rewritten in large parts by
585 .An Jeff Roberson Aq Mt jeff@FreeBSD.org
586 to function as a slab allocator.
588 This manual page was written by
589 .An Dag-Erling Sm\(/orgrav Aq Mt des@FreeBSD.org .
591 .An Jeroen Ruigrok van der Werven Aq Mt asmodai@FreeBSD.org .