/*-1* SPDX-License-Identifier: BSD-2-Clause2*3* Copyright (c) 2002, 2003, 2004, 2005 Jeffrey Roberson <[email protected]>4* Copyright (c) 2004, 2005 Bosko Milekic <[email protected]>5* All rights reserved.6*7* Redistribution and use in source and binary forms, with or without8* modification, are permitted provided that the following conditions9* are met:10* 1. Redistributions of source code must retain the above copyright11* notice unmodified, this list of conditions, and the following12* disclaimer.13* 2. Redistributions in binary form must reproduce the above copyright14* notice, this list of conditions and the following disclaimer in the15* documentation and/or other materials provided with the distribution.16*17* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR18* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES19* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.20* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,21* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT22* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,23* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY24* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT25* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF26* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.27*28*/2930/*31* uma.h - External definitions for the Universal Memory Allocator32*33*/3435#ifndef _VM_UMA_H_36#define _VM_UMA_H_3738#include <sys/param.h> /* For NULL */39#include <sys/malloc.h> /* For M_* */40#include <sys/_smr.h>4142/* User visible parameters */43#define UMA_SMALLEST_UNIT 8 /* Smallest item allocated */4445/* Types and type defs */4647struct uma_zone;48/* Opaque type used as a handle to the zone */49typedef struct uma_zone * uma_zone_t;5051/*52* Item constructor53*54* Arguments:55* item A pointer to the memory which has been allocated.56* arg The arg field passed to uma_zalloc_arg57* size The size of the allocated item58* flags See zalloc flags59*60* Returns:61* 0 on success62* errno on failure63*64* Discussion:65* The constructor is called just before the memory is returned66* to the user. It may block if necessary.67*/68typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags);6970/*71* Item destructor72*73* Arguments:74* item A pointer to the memory which has been allocated.75* size The size of the item being destructed.76* arg Argument passed through uma_zfree_arg77*78* Returns:79* Nothing80*81* Discussion:82* The destructor may perform operations that differ from those performed83* by the initializer, but it must leave the object in the same state.84* This IS type stable storage. This is called after EVERY zfree call.85*/86typedef void (*uma_dtor)(void *mem, int size, void *arg);8788/*89* Item initializer90*91* Arguments:92* item A pointer to the memory which has been allocated.93* size The size of the item being initialized.94* flags See zalloc flags95*96* Returns:97* 0 on success98* errno on failure99*100* Discussion:101* The initializer is called when the memory is cached in the uma zone.102* The initializer and the destructor should leave the object in the same103* state.104*/105typedef int (*uma_init)(void *mem, int size, int flags);106107/*108* Item discard function109*110* Arguments:111* item A pointer to memory which has been 'freed' but has not left the112* zone's cache.113* size The size of the item being discarded.114*115* Returns:116* Nothing117*118* Discussion:119* This routine is called when memory leaves a zone and is returned to the120* system for other uses. It is the counter-part to the init function.121*/122typedef void (*uma_fini)(void *mem, int size);123124/*125* Import new memory into a cache zone.126*/127typedef int (*uma_import)(void *arg, void **store, int count, int domain,128int flags);129130/*131* Free memory from a cache zone.132*/133typedef void (*uma_release)(void *arg, void **store, int count);134135/*136* What's the difference between initializing and constructing?137*138* The item is initialized when it is cached, and this is the state that the139* object should be in when returned to the allocator. The purpose of this is140* to remove some code which would otherwise be called on each allocation by141* utilizing a known, stable state. This differs from the constructor which142* will be called on EVERY allocation.143*144* For example, in the initializer you may want to initialize embedded locks,145* NULL list pointers, set up initial states, magic numbers, etc. This way if146* the object is held in the allocator and re-used it won't be necessary to147* re-initialize it.148*149* The constructor may be used to lock a data structure, link it on to lists,150* bump reference counts or total counts of outstanding structures, etc.151*152*/153154/* Function proto types */155156/*157* Create a new uma zone158*159* Arguments:160* name The text name of the zone for debugging and stats. This memory161* should not be freed until the zone has been deallocated.162* size The size of the object that is being created.163* ctor The constructor that is called when the object is allocated.164* dtor The destructor that is called when the object is freed.165* init An initializer that sets up the initial state of the memory.166* fini A discard function that undoes initialization done by init.167* ctor/dtor/init/fini may all be null, see notes above.168* align A bitmask that corresponds to the requested alignment169* eg 4 would be 0x3170* flags A set of parameters that control the behavior of the zone.171*172* Returns:173* A pointer to a structure which is intended to be opaque to users of174* the interface. The value may be null if the wait flag is not set.175*/176uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor,177uma_dtor dtor, uma_init uminit, uma_fini fini,178int align, uint32_t flags);179180/*181* Create a secondary uma zone182*183* Arguments:184* name The text name of the zone for debugging and stats. This memory185* should not be freed until the zone has been deallocated.186* ctor The constructor that is called when the object is allocated.187* dtor The destructor that is called when the object is freed.188* zinit An initializer that sets up the initial state of the memory189* as the object passes from the Keg's slab to the Zone's cache.190* zfini A discard function that undoes initialization done by init191* as the object passes from the Zone's cache to the Keg's slab.192*193* ctor/dtor/zinit/zfini may all be null, see notes above.194* Note that the zinit and zfini specified here are NOT195* exactly the same as the init/fini specified to uma_zcreate()196* when creating a primary zone. These zinit/zfini are called197* on the TRANSITION from keg to zone (and vice-versa). Once198* these are set, the primary zone may alter its init/fini199* (which are called when the object passes from VM to keg)200* using uma_zone_set_init/fini()) as well as its own201* zinit/zfini (unset by default for primary zone) with202* uma_zone_set_zinit/zfini() (note subtle 'z' prefix).203*204* primary A reference to this zone's Primary Zone which contains the205* backing Keg for the Secondary Zone being added.206*207* Returns:208* A pointer to a structure which is intended to be opaque to users of209* the interface. The value may be null if the wait flag is not set.210*/211uma_zone_t uma_zsecond_create(const char *name, uma_ctor ctor, uma_dtor dtor,212uma_init zinit, uma_fini zfini, uma_zone_t primary);213214/*215* Create cache-only zones.216*217* This allows uma's per-cpu cache facilities to handle arbitrary218* pointers. Consumers must specify the import and release functions to219* fill and destroy caches. UMA does not allocate any memory for these220* zones. The 'arg' parameter is passed to import/release and is caller221* specific.222*/223uma_zone_t uma_zcache_create(const char *name, int size, uma_ctor ctor,224uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_import zimport,225uma_release zrelease, void *arg, int flags);226227/*228* Definitions for uma_zcreate flags229*230* These flags share space with UMA_ZFLAGs in uma_int.h. Be careful not to231* overlap when adding new features.232*/233#define UMA_ZONE_UNMANAGED 0x0001 /*234* Don't regulate the cache size, even235* under memory pressure.236*/237#define UMA_ZONE_ZINIT 0x0002 /* Initialize with zeros */238#define UMA_ZONE_CONTIG 0x0004 /*239* Physical memory underlying an object240* must be contiguous.241*/242#define UMA_ZONE_NOTOUCH 0x0008 /* UMA may not access the memory */243#define UMA_ZONE_MALLOC 0x0010 /* For use by malloc(9) only! */244#define UMA_ZONE_NOFREE 0x0020 /* Do not free slabs of this type! */245#define UMA_ZONE_MTXCLASS 0x0040 /* Create a new lock class */246#define UMA_ZONE_VM 0x0080 /*247* Used for internal vm datastructures248* only.249*/250#define UMA_ZONE_NOTPAGE 0x0100 /* allocf memory not vm pages */251#define UMA_ZONE_SECONDARY 0x0200 /* Zone is a Secondary Zone */252#define UMA_ZONE_NOBUCKET 0x0400 /* Do not use buckets. */253#define UMA_ZONE_MAXBUCKET 0x0800 /* Use largest buckets. */254#define UMA_ZONE_NOTRIM 0x1000 /* Don't trim this zone */255#define UMA_ZONE_CACHESPREAD 0x2000 /*256* Spread memory start locations across257* all possible cache lines. May258* require many virtually contiguous259* backend pages and can fail early.260*/261#define UMA_ZONE_NODUMP 0x4000 /*262* Zone's pages will not be included in263* mini-dumps.264*/265#define UMA_ZONE_PCPU 0x8000 /*266* Allocates mp_maxid + 1 slabs of267* PAGE_SIZE268*/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 defers273* frees until all read sections274* have exited. This flag creates275* a unique SMR context for this276* zone. To share contexts see277* uma_zone_set_smr() below.278*279* See sys/smr.h for more details.280*/281#define UMA_ZONE_NOKASAN 0x80000 /*282* Disable KASAN verification. This is283* implied by NOFREE. Cache zones are284* not verified by default.285*/286/* In use by UMA_ZFLAGs: 0xffe00000 */287288/*289* These flags are shared between the keg and zone. Some are determined290* based on physical parameters of the request and may not be provided by291* the consumer.292*/293#define UMA_ZONE_INHERIT \294(UMA_ZONE_NOTOUCH | UMA_ZONE_MALLOC | UMA_ZONE_NOFREE | \295UMA_ZONE_VM | UMA_ZONE_NOTPAGE | UMA_ZONE_PCPU | \296UMA_ZONE_FIRSTTOUCH | UMA_ZONE_ROUNDROBIN | UMA_ZONE_NOKASAN)297298/* 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' */308309#define UMA_ANYDOMAIN -1 /* Special value for domain search. */310311/*312* Destroys an empty uma zone. If the zone is not empty uma complains loudly.313*314* Arguments:315* zone The zone we want to destroy.316*317*/318void uma_zdestroy(uma_zone_t zone);319320/*321* Allocates an item out of a zone322*323* Arguments:324* zone The zone we are allocating from325* arg This data is passed to the ctor function326* flags See sys/malloc.h for available flags.327*328* Returns:329* A non-null pointer to an initialized element from the zone is330* guaranteed if the wait flag is M_WAITOK. Otherwise a null pointer331* may be returned if the zone is empty or the ctor failed.332*/333334void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);335336/* Allocate per-cpu data. Access the correct data with zpcpu_get(). */337void *uma_zalloc_pcpu_arg(uma_zone_t zone, void *arg, int flags);338339/* Use with SMR zones. */340void *uma_zalloc_smr(uma_zone_t zone, int flags);341342/*343* Allocate an item from a specific NUMA domain. This uses a slow path in344* the allocator but is guaranteed to allocate memory from the requested345* domain if M_WAITOK is set.346*347* Arguments:348* zone The zone we are allocating from349* arg This data is passed to the ctor function350* domain The domain to allocate from.351* flags See sys/malloc.h for available flags.352*/353void *uma_zalloc_domain(uma_zone_t zone, void *arg, int domain, int flags);354355/*356* Allocates an item out of a zone without supplying an argument357*358* This is just a wrapper for uma_zalloc_arg for convenience.359*360*/361static __inline void *uma_zalloc(uma_zone_t zone, int flags);362static __inline void *uma_zalloc_pcpu(uma_zone_t zone, int flags);363364static __inline void *365uma_zalloc(uma_zone_t zone, int flags)366{367return uma_zalloc_arg(zone, NULL, flags);368}369370static __inline void *371uma_zalloc_pcpu(uma_zone_t zone, int flags)372{373return uma_zalloc_pcpu_arg(zone, NULL, flags);374}375376/*377* Frees an item back into the specified zone.378*379* Arguments:380* zone The zone the item was originally allocated out of.381* item The memory to be freed.382* arg Argument passed to the destructor383*384* Returns:385* Nothing.386*/387388void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);389390/* Use with PCPU zones. */391void uma_zfree_pcpu_arg(uma_zone_t zone, void *item, void *arg);392393/* Use with SMR zones. */394void uma_zfree_smr(uma_zone_t zone, void *item);395396/*397* Frees an item back to a zone without supplying an argument398*399* This is just a wrapper for uma_zfree_arg for convenience.400*401*/402static __inline void uma_zfree(uma_zone_t zone, void *item);403static __inline void uma_zfree_pcpu(uma_zone_t zone, void *item);404405static __inline void406uma_zfree(uma_zone_t zone, void *item)407{408uma_zfree_arg(zone, item, NULL);409}410411static __inline void412uma_zfree_pcpu(uma_zone_t zone, void *item)413{414uma_zfree_pcpu_arg(zone, item, NULL);415}416417/*418* Wait until the specified zone can allocate an item.419*/420void uma_zwait(uma_zone_t zone);421422/*423* Backend page supplier routines424*425* Arguments: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.431*432* Returns:433* A pointer to the allocated memory or NULL on failure.434*/435436typedef void *(*uma_alloc)(uma_zone_t zone, vm_size_t size, int domain,437uint8_t *pflag, int wait);438439/*440* Backend page free routines441*442* Arguments: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.446*447* Returns:448* None449*/450typedef void (*uma_free)(void *item, vm_size_t size, uint8_t pflag);451452/*453* Reclaims unused memory. If no NUMA domain is specified, memory from all454* domains is reclaimed.455*456* Arguments:457* req Reclamation request type.458* domain The target NUMA domain.459* Returns:460* None461*/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 */465void uma_reclaim(int req);466void uma_reclaim_domain(int req, int domain);467void uma_zone_reclaim(uma_zone_t, int req);468void uma_zone_reclaim_domain(uma_zone_t, int req, int domain);469470/*471* Sets the alignment mask to be used for all zones requesting cache472* alignment. Should be called by MD boot code prior to starting VM/UMA.473*474* Arguments:475* mask The alignment mask476*477* Returns:478* Nothing479*/480void uma_set_cache_align_mask(unsigned int mask);481482#include <vm/uma_align_mask.h>483484/*485* Set a reserved number of items to hold for M_USE_RESERVE allocations. All486* other requests must allocate new backing pages.487*/488void uma_zone_reserve(uma_zone_t zone, int nitems);489490/*491* Reserves the maximum KVA space required by the zone and configures the zone492* to use a backend that allocates physical memory and maps it using the493* reserved KVA.494*495* Arguments:496* zone The zone to update.497* nitems The upper limit on the number of items that can be allocated.498*499* Returns:500* 0 if KVA space can not be allocated501* 1 if successful502*503* Discussion:504* When the machine supports a direct map and the zone's items are smaller505* than a page, the zone will use the direct map instead of allocating KVA506* space.507*/508int uma_zone_reserve_kva(uma_zone_t zone, int nitems);509510/*511* Sets an upper limit on the number of items allocated from a zone512*513* Arguments:514* zone The zone to limit515* nitems The requested upper limit on the number of items allowed516*517* Returns:518* int The effective value of nitems519*/520int uma_zone_set_max(uma_zone_t zone, int nitems);521522/*523* Sets an upper limit on the number of items allowed in zone's caches524*525* Arguments:526* zone The zone to limit527* nitems The requested upper limit on the number of items allowed528*/529void uma_zone_set_maxcache(uma_zone_t zone, int nitems);530531/*532* Obtains the effective limit on the number of items in a zone533*534* Arguments:535* zone The zone to obtain the effective limit from536*537* Return:538* 0 No limit539* int The effective limit of the zone540*/541int uma_zone_get_max(uma_zone_t zone);542543/*544* Sets a warning to be printed when limit is reached545*546* Arguments:547* zone The zone we will warn about548* warning Warning content549*550* Returns:551* Nothing552*/553void uma_zone_set_warning(uma_zone_t zone, const char *warning);554555/*556* Sets a function to run when limit is reached557*558* Arguments:559* zone The zone to which this applies560* fx The function ro run561*562* Returns:563* Nothing564*/565typedef void (*uma_maxaction_t)(uma_zone_t, int);566void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t);567568/*569* Obtains the approximate current number of items allocated from a zone570*571* Arguments:572* zone The zone to obtain the current allocation count from573*574* Return:575* int The approximate current number of items allocated from the zone576*/577int uma_zone_get_cur(uma_zone_t zone);578579/*580* The following two routines (uma_zone_set_init/fini)581* are used to set the backend init/fini pair which acts on an582* object as it becomes allocated and is placed in a slab within583* the specified zone's backing keg. These should probably not584* be changed once allocations have already begun, but only be set585* immediately upon zone creation.586*/587void uma_zone_set_init(uma_zone_t zone, uma_init uminit);588void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);589590/*591* The following two routines (uma_zone_set_zinit/zfini) are592* used to set the zinit/zfini pair which acts on an object as593* it passes from the backing Keg's slab cache to the594* specified Zone's bucket cache. These should probably not595* be changed once allocations have already begun, but only be set596* immediately upon zone creation.597*/598void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);599void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);600601/*602* Replaces the standard backend allocator for this zone.603*604* Arguments:605* zone The zone whose backend allocator is being changed.606* allocf A pointer to the allocation function607*608* Returns:609* Nothing610*611* Discussion:612* This could be used to implement pageable allocation, or perhaps613* even DMA allocators if used in conjunction with the OFFPAGE614* zone flag.615*/616617void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);618619/*620* Used for freeing memory provided by the allocf above621*622* Arguments:623* zone The zone that intends to use this free routine.624* freef The page freeing routine.625*626* Returns:627* Nothing628*/629630void uma_zone_set_freef(uma_zone_t zone, uma_free freef);631632/*633* Associate a zone with a smr context that is allocated after creation634* so that multiple zones may share the same context.635*/636void uma_zone_set_smr(uma_zone_t zone, smr_t smr);637638/*639* Fetch the smr context that was set or made in uma_zcreate().640*/641smr_t uma_zone_get_smr(uma_zone_t zone);642643/*644* These flags are settable in the allocf and visible in the freef.645*/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 */650651/*652* Used to pre-fill a zone with some number of items653*654* Arguments:655* zone The zone to fill656* itemcnt The number of items to reserve657*658* Returns:659* Nothing660*661* NOTE: This is blocking and should only be done at startup662*/663void uma_prealloc(uma_zone_t zone, int itemcnt);664665/*666* Used to determine if a fixed-size zone is exhausted.667*668* Arguments:669* zone The zone to check670*671* Returns:672* Non-zero if zone is exhausted.673*/674int uma_zone_exhausted(uma_zone_t zone);675676/*677* Returns the bytes of memory consumed by the zone.678*/679size_t uma_zone_memory(uma_zone_t zone);680681/*682* Common UMA_ZONE_PCPU zones.683*/684extern uma_zone_t pcpu_zone_4;685extern uma_zone_t pcpu_zone_8;686extern uma_zone_t pcpu_zone_16;687extern uma_zone_t pcpu_zone_32;688extern uma_zone_t pcpu_zone_64;689690/*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 of693* alternative uma_type_header and uma_type_stat structures.694*/695#define UMA_STREAM_VERSION 0x00000001696struct uma_stream_header {697uint32_t ush_version; /* Stream format version. */698uint32_t ush_maxcpus; /* Value of MAXCPU for stream. */699uint32_t ush_count; /* Number of records. */700uint32_t _ush_pad; /* Pad/reserved field. */701};702703#define UTH_MAX_NAME 32704#define UTH_ZONE_SECONDARY 0x00000001705struct uma_type_header {706/*707* Static per-zone data, some extracted from the supporting keg.708*/709char uth_name[UTH_MAX_NAME];710uint32_t uth_align; /* Keg: alignment. */711uint32_t uth_size; /* Keg: requested size of item. */712uint32_t uth_rsize; /* Keg: real size of item. */713uint32_t uth_maxpages; /* Keg: maximum number of pages. */714uint32_t uth_limit; /* Keg: max items to allocate. */715716/*717* Current dynamic zone/keg-derived statistics.718*/719uint32_t uth_pages; /* Keg: pages allocated. */720uint32_t uth_keg_free; /* Keg: items free. */721uint32_t uth_zone_free; /* Zone: items free. */722uint32_t uth_bucketsize; /* Zone: desired bucket size. */723uint32_t uth_zone_flags; /* Zone: flags. */724uint64_t uth_allocs; /* Zone: number of allocations. */725uint64_t uth_frees; /* Zone: number of frees. */726uint64_t uth_fails; /* Zone: number of alloc failures. */727uint64_t uth_sleeps; /* Zone: number of alloc sleeps. */728uint64_t uth_xdomain; /* Zone: Number of cross domain frees. */729uint64_t _uth_reserved1[1]; /* Reserved. */730};731732struct uma_percpu_stat {733uint64_t ups_allocs; /* Cache: number of allocations. */734uint64_t ups_frees; /* Cache: number of frees. */735uint64_t ups_cache_free; /* Cache: free items in cache. */736uint64_t _ups_reserved[5]; /* Reserved. */737};738739void uma_reclaim_wakeup(void);740void uma_reclaim_worker(void *);741742unsigned long uma_limit(void);743744/* Return the amount of memory managed by UMA. */745unsigned long uma_size(void);746747/* Return the amount of memory remaining. May be negative. */748long uma_avail(void);749750#endif /* _VM_UMA_H_ */751752753