18355f576SJeff Roberson /* 2f461cf22SJeff Roberson * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org> 38355f576SJeff Roberson * All rights reserved. 48355f576SJeff Roberson * 58355f576SJeff Roberson * Redistribution and use in source and binary forms, with or without 68355f576SJeff Roberson * modification, are permitted provided that the following conditions 78355f576SJeff Roberson * are met: 88355f576SJeff Roberson * 1. Redistributions of source code must retain the above copyright 98355f576SJeff Roberson * notice unmodified, this list of conditions, and the following 108355f576SJeff Roberson * disclaimer. 118355f576SJeff Roberson * 2. Redistributions in binary form must reproduce the above copyright 128355f576SJeff Roberson * notice, this list of conditions and the following disclaimer in the 138355f576SJeff Roberson * documentation and/or other materials provided with the distribution. 148355f576SJeff Roberson * 158355f576SJeff Roberson * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 168355f576SJeff Roberson * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 178355f576SJeff Roberson * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 188355f576SJeff Roberson * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 198355f576SJeff Roberson * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 208355f576SJeff Roberson * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 218355f576SJeff Roberson * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 228355f576SJeff Roberson * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 238355f576SJeff Roberson * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 248355f576SJeff Roberson * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 258355f576SJeff Roberson */ 268355f576SJeff Roberson 278355f576SJeff Roberson /* 288355f576SJeff Roberson * uma_core.c Implementation of the Universal Memory allocator 298355f576SJeff Roberson * 308355f576SJeff Roberson * This allocator is intended to replace the multitude of similar object caches 318355f576SJeff Roberson * in the standard FreeBSD kernel. The intent is to be flexible as well as 328355f576SJeff Roberson * effecient. A primary design goal is to return unused memory to the rest of 338355f576SJeff Roberson * the system. This will make the system as a whole more flexible due to the 348355f576SJeff Roberson * ability to move memory to subsystems which most need it instead of leaving 358355f576SJeff Roberson * pools of reserved memory unused. 368355f576SJeff Roberson * 378355f576SJeff Roberson * The basic ideas stem from similar slab/zone based allocators whose algorithms 388355f576SJeff Roberson * are well known. 398355f576SJeff Roberson * 408355f576SJeff Roberson */ 418355f576SJeff Roberson 428355f576SJeff Roberson /* 438355f576SJeff Roberson * TODO: 448355f576SJeff Roberson * - Improve memory usage for large allocations 458355f576SJeff Roberson * - Investigate cache size adjustments 468355f576SJeff Roberson */ 478355f576SJeff Roberson 48874651b1SDavid E. O'Brien #include <sys/cdefs.h> 49874651b1SDavid E. O'Brien __FBSDID("$FreeBSD$"); 50874651b1SDavid E. O'Brien 518355f576SJeff Roberson /* I should really use ktr.. */ 528355f576SJeff Roberson /* 538355f576SJeff Roberson #define UMA_DEBUG 1 548355f576SJeff Roberson #define UMA_DEBUG_ALLOC 1 558355f576SJeff Roberson #define UMA_DEBUG_ALLOC_1 1 568355f576SJeff Roberson */ 578355f576SJeff Roberson 588355f576SJeff Roberson #include "opt_param.h" 598355f576SJeff Roberson #include <sys/param.h> 608355f576SJeff Roberson #include <sys/systm.h> 618355f576SJeff Roberson #include <sys/kernel.h> 628355f576SJeff Roberson #include <sys/types.h> 638355f576SJeff Roberson #include <sys/queue.h> 648355f576SJeff Roberson #include <sys/malloc.h> 658355f576SJeff Roberson #include <sys/lock.h> 668355f576SJeff Roberson #include <sys/sysctl.h> 678355f576SJeff Roberson #include <sys/mutex.h> 684c1cc01cSJohn Baldwin #include <sys/proc.h> 698355f576SJeff Roberson #include <sys/smp.h> 7086bbae32SJeff Roberson #include <sys/vmmeter.h> 718522511bSHartmut Brandt #include <sys/mbuf.h> 7286bbae32SJeff Roberson 738355f576SJeff Roberson #include <vm/vm.h> 748355f576SJeff Roberson #include <vm/vm_object.h> 758355f576SJeff Roberson #include <vm/vm_page.h> 768355f576SJeff Roberson #include <vm/vm_param.h> 778355f576SJeff Roberson #include <vm/vm_map.h> 788355f576SJeff Roberson #include <vm/vm_kern.h> 798355f576SJeff Roberson #include <vm/vm_extern.h> 808355f576SJeff Roberson #include <vm/uma.h> 818355f576SJeff Roberson #include <vm/uma_int.h> 82639c9550SJeff Roberson #include <vm/uma_dbg.h> 838355f576SJeff Roberson 8448eea375SJeff Roberson #include <machine/vmparam.h> 8548eea375SJeff Roberson 868355f576SJeff Roberson /* 878355f576SJeff Roberson * This is the zone from which all zones are spawned. The idea is that even 888355f576SJeff Roberson * the zone heads are allocated from the allocator, so we use the bss section 898355f576SJeff Roberson * to bootstrap us. 908355f576SJeff Roberson */ 9186bbae32SJeff Roberson static struct uma_zone masterzone; 9286bbae32SJeff Roberson static uma_zone_t zones = &masterzone; 938355f576SJeff Roberson 948355f576SJeff Roberson /* This is the zone from which all of uma_slab_t's are allocated. */ 958355f576SJeff Roberson static uma_zone_t slabzone; 968355f576SJeff Roberson 978355f576SJeff Roberson /* 988355f576SJeff Roberson * The initial hash tables come out of this zone so they can be allocated 998355f576SJeff Roberson * prior to malloc coming up. 1008355f576SJeff Roberson */ 1018355f576SJeff Roberson static uma_zone_t hashzone; 1028355f576SJeff Roberson 103961647dfSJeff Roberson static MALLOC_DEFINE(M_UMAHASH, "UMAHash", "UMA Hash Buckets"); 104961647dfSJeff Roberson 1058355f576SJeff Roberson /* 10686bbae32SJeff Roberson * Are we allowed to allocate buckets? 10786bbae32SJeff Roberson */ 10886bbae32SJeff Roberson static int bucketdisable = 1; 10986bbae32SJeff Roberson 1108355f576SJeff Roberson /* Linked list of all zones in the system */ 1118355f576SJeff Roberson static LIST_HEAD(,uma_zone) uma_zones = LIST_HEAD_INITIALIZER(&uma_zones); 1128355f576SJeff Roberson 1138355f576SJeff Roberson /* This mutex protects the zone list */ 1148355f576SJeff Roberson static struct mtx uma_mtx; 1158355f576SJeff Roberson 116d88797c2SBosko Milekic /* These are the pcpu cache locks */ 117d88797c2SBosko Milekic static struct mtx uma_pcpu_mtx[MAXCPU]; 118d88797c2SBosko Milekic 1198355f576SJeff Roberson /* Linked list of boot time pages */ 1208355f576SJeff Roberson static LIST_HEAD(,uma_slab) uma_boot_pages = 1218355f576SJeff Roberson LIST_HEAD_INITIALIZER(&uma_boot_pages); 1228355f576SJeff Roberson 1238355f576SJeff Roberson /* Count of free boottime pages */ 1248355f576SJeff Roberson static int uma_boot_free = 0; 1258355f576SJeff Roberson 1268355f576SJeff Roberson /* Is the VM done starting up? */ 1278355f576SJeff Roberson static int booted = 0; 1288355f576SJeff Roberson 1299643769aSJeff Roberson /* 1309643769aSJeff Roberson * This is the handle used to schedule events that need to happen 1319643769aSJeff Roberson * outside of the allocation fast path. 1329643769aSJeff Roberson */ 1338355f576SJeff Roberson static struct callout uma_callout; 1349643769aSJeff Roberson #define UMA_TIMEOUT 20 /* Seconds for callout interval. */ 1358355f576SJeff Roberson 1368355f576SJeff Roberson /* This is mp_maxid + 1, for use while looping over each cpu */ 1378355f576SJeff Roberson static int maxcpu; 1388355f576SJeff Roberson 1398355f576SJeff Roberson /* 1408355f576SJeff Roberson * This structure is passed as the zone ctor arg so that I don't have to create 1418355f576SJeff Roberson * a special allocation function just for zones. 1428355f576SJeff Roberson */ 1438355f576SJeff Roberson struct uma_zctor_args { 1448355f576SJeff Roberson char *name; 145c3bdc05fSAndrew R. Reiter size_t size; 1468355f576SJeff Roberson uma_ctor ctor; 1478355f576SJeff Roberson uma_dtor dtor; 1488355f576SJeff Roberson uma_init uminit; 1498355f576SJeff Roberson uma_fini fini; 1508355f576SJeff Roberson int align; 1518355f576SJeff Roberson u_int16_t flags; 1528355f576SJeff Roberson }; 1538355f576SJeff Roberson 154cae33c14SJeff Roberson struct uma_bucket_zone { 155cae33c14SJeff Roberson uma_zone_t ubz_zone; 156cae33c14SJeff Roberson char *ubz_name; 157cae33c14SJeff Roberson int ubz_entries; 158cae33c14SJeff Roberson }; 159cae33c14SJeff Roberson 160cae33c14SJeff Roberson #define BUCKET_MAX 128 161cae33c14SJeff Roberson 162cae33c14SJeff Roberson struct uma_bucket_zone bucket_zones[] = { 163cae33c14SJeff Roberson { NULL, "16 Bucket", 16 }, 164cae33c14SJeff Roberson { NULL, "32 Bucket", 32 }, 165cae33c14SJeff Roberson { NULL, "64 Bucket", 64 }, 166cae33c14SJeff Roberson { NULL, "128 Bucket", 128 }, 167cae33c14SJeff Roberson { NULL, NULL, 0} 168cae33c14SJeff Roberson }; 169cae33c14SJeff Roberson 170cae33c14SJeff Roberson #define BUCKET_SHIFT 4 171cae33c14SJeff Roberson #define BUCKET_ZONES ((BUCKET_MAX >> BUCKET_SHIFT) + 1) 172cae33c14SJeff Roberson 173cae33c14SJeff Roberson uint8_t bucket_size[BUCKET_ZONES]; 174cae33c14SJeff Roberson 1758355f576SJeff Roberson /* Prototypes.. */ 1768355f576SJeff Roberson 1778355f576SJeff Roberson static void *obj_alloc(uma_zone_t, int, u_int8_t *, int); 1788355f576SJeff Roberson static void *page_alloc(uma_zone_t, int, u_int8_t *, int); 179009b6fcbSJeff Roberson static void *startup_alloc(uma_zone_t, int, u_int8_t *, int); 1808355f576SJeff Roberson static void page_free(void *, int, u_int8_t); 1818355f576SJeff Roberson static uma_slab_t slab_zalloc(uma_zone_t, int); 1829643769aSJeff Roberson static void cache_drain(uma_zone_t); 1838355f576SJeff Roberson static void bucket_drain(uma_zone_t, uma_bucket_t); 1848355f576SJeff Roberson static void zone_ctor(void *, int, void *); 1859c2cd7e5SJeff Roberson static void zone_dtor(void *, int, void *); 1868355f576SJeff Roberson static void zero_init(void *, int); 1878355f576SJeff Roberson static void zone_small_init(uma_zone_t zone); 1888355f576SJeff Roberson static void zone_large_init(uma_zone_t zone); 1898355f576SJeff Roberson static void zone_foreach(void (*zfunc)(uma_zone_t)); 1908355f576SJeff Roberson static void zone_timeout(uma_zone_t zone); 1910aef6126SJeff Roberson static int hash_alloc(struct uma_hash *); 1920aef6126SJeff Roberson static int hash_expand(struct uma_hash *, struct uma_hash *); 1930aef6126SJeff Roberson static void hash_free(struct uma_hash *hash); 1948355f576SJeff Roberson static void uma_timeout(void *); 1958355f576SJeff Roberson static void uma_startup3(void); 196bbee39c6SJeff Roberson static void *uma_zalloc_internal(uma_zone_t, void *, int); 19786bbae32SJeff Roberson static void uma_zfree_internal(uma_zone_t, void *, void *, int); 19886bbae32SJeff Roberson static void bucket_enable(void); 199cae33c14SJeff Roberson static void bucket_init(void); 200cae33c14SJeff Roberson static uma_bucket_t bucket_alloc(int, int); 201cae33c14SJeff Roberson static void bucket_free(uma_bucket_t); 202cae33c14SJeff Roberson static void bucket_zone_drain(void); 203bbee39c6SJeff Roberson static int uma_zalloc_bucket(uma_zone_t zone, int flags); 204bbee39c6SJeff Roberson static uma_slab_t uma_zone_slab(uma_zone_t zone, int flags); 205bbee39c6SJeff Roberson static void *uma_slab_alloc(uma_zone_t zone, uma_slab_t slab); 2069643769aSJeff Roberson static void zone_drain(uma_zone_t); 207bbee39c6SJeff Roberson 2088355f576SJeff Roberson void uma_print_zone(uma_zone_t); 2098355f576SJeff Roberson void uma_print_stats(void); 2108355f576SJeff Roberson static int sysctl_vm_zone(SYSCTL_HANDLER_ARGS); 2118355f576SJeff Roberson 2128355f576SJeff Roberson SYSCTL_OID(_vm, OID_AUTO, zone, CTLTYPE_STRING|CTLFLAG_RD, 2138355f576SJeff Roberson NULL, 0, sysctl_vm_zone, "A", "Zone Info"); 2148355f576SJeff Roberson SYSINIT(uma_startup3, SI_SUB_VM_CONF, SI_ORDER_SECOND, uma_startup3, NULL); 2158355f576SJeff Roberson 21686bbae32SJeff Roberson /* 21786bbae32SJeff Roberson * This routine checks to see whether or not it's safe to enable buckets. 21886bbae32SJeff Roberson */ 21986bbae32SJeff Roberson 22086bbae32SJeff Roberson static void 22186bbae32SJeff Roberson bucket_enable(void) 22286bbae32SJeff Roberson { 22386bbae32SJeff Roberson if (cnt.v_free_count < cnt.v_free_min) 22486bbae32SJeff Roberson bucketdisable = 1; 22586bbae32SJeff Roberson else 22686bbae32SJeff Roberson bucketdisable = 0; 22786bbae32SJeff Roberson } 22886bbae32SJeff Roberson 229cae33c14SJeff Roberson static void 230cae33c14SJeff Roberson bucket_init(void) 231cae33c14SJeff Roberson { 232cae33c14SJeff Roberson struct uma_bucket_zone *ubz; 233cae33c14SJeff Roberson int i; 234cae33c14SJeff Roberson int j; 235cae33c14SJeff Roberson 236cae33c14SJeff Roberson for (i = 0, j = 0; bucket_zones[j].ubz_entries != 0; j++) { 237cae33c14SJeff Roberson int size; 238cae33c14SJeff Roberson 239cae33c14SJeff Roberson ubz = &bucket_zones[j]; 240cae33c14SJeff Roberson size = roundup(sizeof(struct uma_bucket), sizeof(void *)); 241cae33c14SJeff Roberson size += sizeof(void *) * ubz->ubz_entries; 242cae33c14SJeff Roberson ubz->ubz_zone = uma_zcreate(ubz->ubz_name, size, 243b60f5b79SJeff Roberson NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); 244cae33c14SJeff Roberson for (; i <= ubz->ubz_entries; i += (1 << BUCKET_SHIFT)) 245cae33c14SJeff Roberson bucket_size[i >> BUCKET_SHIFT] = j; 246cae33c14SJeff Roberson } 247cae33c14SJeff Roberson } 248cae33c14SJeff Roberson 249cae33c14SJeff Roberson static uma_bucket_t 250cae33c14SJeff Roberson bucket_alloc(int entries, int bflags) 251cae33c14SJeff Roberson { 252cae33c14SJeff Roberson struct uma_bucket_zone *ubz; 253cae33c14SJeff Roberson uma_bucket_t bucket; 254cae33c14SJeff Roberson int idx; 255cae33c14SJeff Roberson 256cae33c14SJeff Roberson /* 257cae33c14SJeff Roberson * This is to stop us from allocating per cpu buckets while we're 258cae33c14SJeff Roberson * running out of UMA_BOOT_PAGES. Otherwise, we would exhaust the 259cae33c14SJeff Roberson * boot pages. This also prevents us from allocating buckets in 260cae33c14SJeff Roberson * low memory situations. 261cae33c14SJeff Roberson */ 262cae33c14SJeff Roberson 263cae33c14SJeff Roberson if (bucketdisable) 264cae33c14SJeff Roberson return (NULL); 265cae33c14SJeff Roberson idx = howmany(entries, 1 << BUCKET_SHIFT); 266cae33c14SJeff Roberson ubz = &bucket_zones[bucket_size[idx]]; 267cae33c14SJeff Roberson bucket = uma_zalloc_internal(ubz->ubz_zone, NULL, bflags); 268cae33c14SJeff Roberson if (bucket) { 269cae33c14SJeff Roberson #ifdef INVARIANTS 270cae33c14SJeff Roberson bzero(bucket->ub_bucket, sizeof(void *) * ubz->ubz_entries); 271cae33c14SJeff Roberson #endif 272cae33c14SJeff Roberson bucket->ub_cnt = 0; 273cae33c14SJeff Roberson bucket->ub_entries = ubz->ubz_entries; 274cae33c14SJeff Roberson } 275cae33c14SJeff Roberson 276cae33c14SJeff Roberson return (bucket); 277cae33c14SJeff Roberson } 278cae33c14SJeff Roberson 279cae33c14SJeff Roberson static void 280cae33c14SJeff Roberson bucket_free(uma_bucket_t bucket) 281cae33c14SJeff Roberson { 282cae33c14SJeff Roberson struct uma_bucket_zone *ubz; 283cae33c14SJeff Roberson int idx; 284cae33c14SJeff Roberson 285cae33c14SJeff Roberson idx = howmany(bucket->ub_entries, 1 << BUCKET_SHIFT); 286cae33c14SJeff Roberson ubz = &bucket_zones[bucket_size[idx]]; 287cae33c14SJeff Roberson uma_zfree_internal(ubz->ubz_zone, bucket, NULL, 0); 288cae33c14SJeff Roberson } 289cae33c14SJeff Roberson 290cae33c14SJeff Roberson static void 291cae33c14SJeff Roberson bucket_zone_drain(void) 292cae33c14SJeff Roberson { 293cae33c14SJeff Roberson struct uma_bucket_zone *ubz; 294cae33c14SJeff Roberson 295cae33c14SJeff Roberson for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) 296cae33c14SJeff Roberson zone_drain(ubz->ubz_zone); 297cae33c14SJeff Roberson } 298cae33c14SJeff Roberson 2998355f576SJeff Roberson 3008355f576SJeff Roberson /* 3018355f576SJeff Roberson * Routine called by timeout which is used to fire off some time interval 3029643769aSJeff Roberson * based calculations. (stats, hash size, etc.) 3038355f576SJeff Roberson * 3048355f576SJeff Roberson * Arguments: 3058355f576SJeff Roberson * arg Unused 3068355f576SJeff Roberson * 3078355f576SJeff Roberson * Returns: 3088355f576SJeff Roberson * Nothing 3098355f576SJeff Roberson */ 3108355f576SJeff Roberson static void 3118355f576SJeff Roberson uma_timeout(void *unused) 3128355f576SJeff Roberson { 31386bbae32SJeff Roberson bucket_enable(); 3148355f576SJeff Roberson zone_foreach(zone_timeout); 3158355f576SJeff Roberson 3168355f576SJeff Roberson /* Reschedule this event */ 3179643769aSJeff Roberson callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); 3188355f576SJeff Roberson } 3198355f576SJeff Roberson 3208355f576SJeff Roberson /* 3219643769aSJeff Roberson * Routine to perform timeout driven calculations. This expands the 3229643769aSJeff Roberson * hashes and does per cpu statistics aggregation. 3238355f576SJeff Roberson * 3248355f576SJeff Roberson * Arguments: 3258355f576SJeff Roberson * zone The zone to operate on 3268355f576SJeff Roberson * 3278355f576SJeff Roberson * Returns: 3288355f576SJeff Roberson * Nothing 3298355f576SJeff Roberson */ 3308355f576SJeff Roberson static void 3318355f576SJeff Roberson zone_timeout(uma_zone_t zone) 3328355f576SJeff Roberson { 3338355f576SJeff Roberson uma_cache_t cache; 3348355f576SJeff Roberson u_int64_t alloc; 3358355f576SJeff Roberson int cpu; 3368355f576SJeff Roberson 3378355f576SJeff Roberson alloc = 0; 3388355f576SJeff Roberson 3398355f576SJeff Roberson /* 3408355f576SJeff Roberson * Aggregate per cpu cache statistics back to the zone. 3418355f576SJeff Roberson * 3428355f576SJeff Roberson * I may rewrite this to set a flag in the per cpu cache instead of 3438355f576SJeff Roberson * locking. If the flag is not cleared on the next round I will have 3448355f576SJeff Roberson * to lock and do it here instead so that the statistics don't get too 3458355f576SJeff Roberson * far out of sync. 3468355f576SJeff Roberson */ 3478355f576SJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) { 3488355f576SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 3498355f576SJeff Roberson if (CPU_ABSENT(cpu)) 3508355f576SJeff Roberson continue; 351d88797c2SBosko Milekic CPU_LOCK(cpu); 3528355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 3538355f576SJeff Roberson /* Add them up, and reset */ 3548355f576SJeff Roberson alloc += cache->uc_allocs; 3558355f576SJeff Roberson cache->uc_allocs = 0; 356d88797c2SBosko Milekic CPU_UNLOCK(cpu); 3578355f576SJeff Roberson } 3588355f576SJeff Roberson } 3598355f576SJeff Roberson 3608355f576SJeff Roberson /* Now push these stats back into the zone.. */ 3618355f576SJeff Roberson ZONE_LOCK(zone); 3628355f576SJeff Roberson zone->uz_allocs += alloc; 3638355f576SJeff Roberson 3648355f576SJeff Roberson /* 3658355f576SJeff Roberson * Expand the zone hash table. 3668355f576SJeff Roberson * 3678355f576SJeff Roberson * This is done if the number of slabs is larger than the hash size. 3688355f576SJeff Roberson * What I'm trying to do here is completely reduce collisions. This 3698355f576SJeff Roberson * may be a little aggressive. Should I allow for two collisions max? 3708355f576SJeff Roberson */ 3718355f576SJeff Roberson 372b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_HASH && 37399571dc3SJeff Roberson zone->uz_pages / zone->uz_ppera >= zone->uz_hash.uh_hashsize) { 3740aef6126SJeff Roberson struct uma_hash newhash; 3750aef6126SJeff Roberson struct uma_hash oldhash; 3760aef6126SJeff Roberson int ret; 3775300d9ddSJeff Roberson 3780aef6126SJeff Roberson /* 3790aef6126SJeff Roberson * This is so involved because allocating and freeing 3800aef6126SJeff Roberson * while the zone lock is held will lead to deadlock. 3810aef6126SJeff Roberson * I have to do everything in stages and check for 3820aef6126SJeff Roberson * races. 3830aef6126SJeff Roberson */ 3840aef6126SJeff Roberson newhash = zone->uz_hash; 3855300d9ddSJeff Roberson ZONE_UNLOCK(zone); 3860aef6126SJeff Roberson ret = hash_alloc(&newhash); 3875300d9ddSJeff Roberson ZONE_LOCK(zone); 3880aef6126SJeff Roberson if (ret) { 3890aef6126SJeff Roberson if (hash_expand(&zone->uz_hash, &newhash)) { 3900aef6126SJeff Roberson oldhash = zone->uz_hash; 3910aef6126SJeff Roberson zone->uz_hash = newhash; 3920aef6126SJeff Roberson } else 3930aef6126SJeff Roberson oldhash = newhash; 3940aef6126SJeff Roberson 3950aef6126SJeff Roberson ZONE_UNLOCK(zone); 3960aef6126SJeff Roberson hash_free(&oldhash); 3970aef6126SJeff Roberson ZONE_LOCK(zone); 3980aef6126SJeff Roberson } 3995300d9ddSJeff Roberson } 4008355f576SJeff Roberson ZONE_UNLOCK(zone); 4018355f576SJeff Roberson } 4028355f576SJeff Roberson 4038355f576SJeff Roberson /* 4045300d9ddSJeff Roberson * Allocate and zero fill the next sized hash table from the appropriate 4055300d9ddSJeff Roberson * backing store. 4065300d9ddSJeff Roberson * 4075300d9ddSJeff Roberson * Arguments: 4080aef6126SJeff Roberson * hash A new hash structure with the old hash size in uh_hashsize 4095300d9ddSJeff Roberson * 4105300d9ddSJeff Roberson * Returns: 4110aef6126SJeff Roberson * 1 on sucess and 0 on failure. 4125300d9ddSJeff Roberson */ 41337c84183SPoul-Henning Kamp static int 4140aef6126SJeff Roberson hash_alloc(struct uma_hash *hash) 4155300d9ddSJeff Roberson { 4160aef6126SJeff Roberson int oldsize; 4175300d9ddSJeff Roberson int alloc; 4185300d9ddSJeff Roberson 4190aef6126SJeff Roberson oldsize = hash->uh_hashsize; 4200aef6126SJeff Roberson 4215300d9ddSJeff Roberson /* We're just going to go to a power of two greater */ 4220aef6126SJeff Roberson if (oldsize) { 4230aef6126SJeff Roberson hash->uh_hashsize = oldsize * 2; 4240aef6126SJeff Roberson alloc = sizeof(hash->uh_slab_hash[0]) * hash->uh_hashsize; 4250aef6126SJeff Roberson hash->uh_slab_hash = (struct slabhead *)malloc(alloc, 426961647dfSJeff Roberson M_UMAHASH, M_NOWAIT); 4275300d9ddSJeff Roberson } else { 4280aef6126SJeff Roberson alloc = sizeof(hash->uh_slab_hash[0]) * UMA_HASH_SIZE_INIT; 4290aef6126SJeff Roberson hash->uh_slab_hash = uma_zalloc_internal(hashzone, NULL, 430a163d034SWarner Losh M_WAITOK); 4310aef6126SJeff Roberson hash->uh_hashsize = UMA_HASH_SIZE_INIT; 4325300d9ddSJeff Roberson } 4330aef6126SJeff Roberson if (hash->uh_slab_hash) { 4340aef6126SJeff Roberson bzero(hash->uh_slab_hash, alloc); 4350aef6126SJeff Roberson hash->uh_hashmask = hash->uh_hashsize - 1; 4360aef6126SJeff Roberson return (1); 4370aef6126SJeff Roberson } 4385300d9ddSJeff Roberson 4390aef6126SJeff Roberson return (0); 4405300d9ddSJeff Roberson } 4415300d9ddSJeff Roberson 4425300d9ddSJeff Roberson /* 44364f051e9SJeff Roberson * Expands the hash table for HASH zones. This is done from zone_timeout 44464f051e9SJeff Roberson * to reduce collisions. This must not be done in the regular allocation 44564f051e9SJeff Roberson * path, otherwise, we can recurse on the vm while allocating pages. 4468355f576SJeff Roberson * 4478355f576SJeff Roberson * Arguments: 4480aef6126SJeff Roberson * oldhash The hash you want to expand 4490aef6126SJeff Roberson * newhash The hash structure for the new table 4508355f576SJeff Roberson * 4518355f576SJeff Roberson * Returns: 4528355f576SJeff Roberson * Nothing 4538355f576SJeff Roberson * 4548355f576SJeff Roberson * Discussion: 4558355f576SJeff Roberson */ 4560aef6126SJeff Roberson static int 4570aef6126SJeff Roberson hash_expand(struct uma_hash *oldhash, struct uma_hash *newhash) 4588355f576SJeff Roberson { 4598355f576SJeff Roberson uma_slab_t slab; 4608355f576SJeff Roberson int hval; 4618355f576SJeff Roberson int i; 4628355f576SJeff Roberson 4630aef6126SJeff Roberson if (!newhash->uh_slab_hash) 4640aef6126SJeff Roberson return (0); 4658355f576SJeff Roberson 4660aef6126SJeff Roberson if (oldhash->uh_hashsize >= newhash->uh_hashsize) 4670aef6126SJeff Roberson return (0); 4688355f576SJeff Roberson 4698355f576SJeff Roberson /* 4708355f576SJeff Roberson * I need to investigate hash algorithms for resizing without a 4718355f576SJeff Roberson * full rehash. 4728355f576SJeff Roberson */ 4738355f576SJeff Roberson 4740aef6126SJeff Roberson for (i = 0; i < oldhash->uh_hashsize; i++) 4750aef6126SJeff Roberson while (!SLIST_EMPTY(&oldhash->uh_slab_hash[i])) { 4760aef6126SJeff Roberson slab = SLIST_FIRST(&oldhash->uh_slab_hash[i]); 4770aef6126SJeff Roberson SLIST_REMOVE_HEAD(&oldhash->uh_slab_hash[i], us_hlink); 4780aef6126SJeff Roberson hval = UMA_HASH(newhash, slab->us_data); 4790aef6126SJeff Roberson SLIST_INSERT_HEAD(&newhash->uh_slab_hash[hval], 4800aef6126SJeff Roberson slab, us_hlink); 4818355f576SJeff Roberson } 4828355f576SJeff Roberson 4830aef6126SJeff Roberson return (1); 4849c2cd7e5SJeff Roberson } 4859c2cd7e5SJeff Roberson 4865300d9ddSJeff Roberson /* 4875300d9ddSJeff Roberson * Free the hash bucket to the appropriate backing store. 4885300d9ddSJeff Roberson * 4895300d9ddSJeff Roberson * Arguments: 4905300d9ddSJeff Roberson * slab_hash The hash bucket we're freeing 4915300d9ddSJeff Roberson * hashsize The number of entries in that hash bucket 4925300d9ddSJeff Roberson * 4935300d9ddSJeff Roberson * Returns: 4945300d9ddSJeff Roberson * Nothing 4955300d9ddSJeff Roberson */ 4969c2cd7e5SJeff Roberson static void 4970aef6126SJeff Roberson hash_free(struct uma_hash *hash) 4989c2cd7e5SJeff Roberson { 4990aef6126SJeff Roberson if (hash->uh_slab_hash == NULL) 5000aef6126SJeff Roberson return; 5010aef6126SJeff Roberson if (hash->uh_hashsize == UMA_HASH_SIZE_INIT) 5028355f576SJeff Roberson uma_zfree_internal(hashzone, 5030aef6126SJeff Roberson hash->uh_slab_hash, NULL, 0); 5048355f576SJeff Roberson else 505961647dfSJeff Roberson free(hash->uh_slab_hash, M_UMAHASH); 5068355f576SJeff Roberson } 5078355f576SJeff Roberson 5088355f576SJeff Roberson /* 5098355f576SJeff Roberson * Frees all outstanding items in a bucket 5108355f576SJeff Roberson * 5118355f576SJeff Roberson * Arguments: 5128355f576SJeff Roberson * zone The zone to free to, must be unlocked. 5138355f576SJeff Roberson * bucket The free/alloc bucket with items, cpu queue must be locked. 5148355f576SJeff Roberson * 5158355f576SJeff Roberson * Returns: 5168355f576SJeff Roberson * Nothing 5178355f576SJeff Roberson */ 5188355f576SJeff Roberson 5198355f576SJeff Roberson static void 5208355f576SJeff Roberson bucket_drain(uma_zone_t zone, uma_bucket_t bucket) 5218355f576SJeff Roberson { 5228355f576SJeff Roberson uma_slab_t slab; 5238355f576SJeff Roberson int mzone; 5248355f576SJeff Roberson void *item; 5258355f576SJeff Roberson 5268355f576SJeff Roberson if (bucket == NULL) 5278355f576SJeff Roberson return; 5288355f576SJeff Roberson 5298355f576SJeff Roberson slab = NULL; 5308355f576SJeff Roberson mzone = 0; 5318355f576SJeff Roberson 5328355f576SJeff Roberson /* We have to lookup the slab again for malloc.. */ 533b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_MALLOC) 5348355f576SJeff Roberson mzone = 1; 5358355f576SJeff Roberson 536cae33c14SJeff Roberson while (bucket->ub_cnt > 0) { 537cae33c14SJeff Roberson bucket->ub_cnt--; 538cae33c14SJeff Roberson item = bucket->ub_bucket[bucket->ub_cnt]; 5398355f576SJeff Roberson #ifdef INVARIANTS 540cae33c14SJeff Roberson bucket->ub_bucket[bucket->ub_cnt] = NULL; 5418355f576SJeff Roberson KASSERT(item != NULL, 5428355f576SJeff Roberson ("bucket_drain: botched ptr, item is NULL")); 5438355f576SJeff Roberson #endif 5448355f576SJeff Roberson /* 5458355f576SJeff Roberson * This is extremely inefficient. The slab pointer was passed 5468355f576SJeff Roberson * to uma_zfree_arg, but we lost it because the buckets don't 5478355f576SJeff Roberson * hold them. This will go away when free() gets a size passed 5488355f576SJeff Roberson * to it. 5498355f576SJeff Roberson */ 55099571dc3SJeff Roberson if (mzone) 55199571dc3SJeff Roberson slab = vtoslab((vm_offset_t)item & (~UMA_SLAB_MASK)); 5528355f576SJeff Roberson uma_zfree_internal(zone, item, slab, 1); 5538355f576SJeff Roberson } 5548355f576SJeff Roberson } 5558355f576SJeff Roberson 5568355f576SJeff Roberson /* 5578355f576SJeff Roberson * Drains the per cpu caches for a zone. 5588355f576SJeff Roberson * 5598355f576SJeff Roberson * Arguments: 5608355f576SJeff Roberson * zone The zone to drain, must be unlocked. 5618355f576SJeff Roberson * 5628355f576SJeff Roberson * Returns: 5638355f576SJeff Roberson * Nothing 5648355f576SJeff Roberson */ 5658355f576SJeff Roberson static void 5669643769aSJeff Roberson cache_drain(uma_zone_t zone) 5678355f576SJeff Roberson { 5688355f576SJeff Roberson uma_bucket_t bucket; 5698355f576SJeff Roberson uma_cache_t cache; 5708355f576SJeff Roberson int cpu; 5718355f576SJeff Roberson 5728355f576SJeff Roberson /* 5738355f576SJeff Roberson * We have to lock each cpu cache before locking the zone 5748355f576SJeff Roberson */ 5758355f576SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 5768355f576SJeff Roberson if (CPU_ABSENT(cpu)) 5778355f576SJeff Roberson continue; 578d88797c2SBosko Milekic CPU_LOCK(cpu); 5798355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 5808355f576SJeff Roberson bucket_drain(zone, cache->uc_allocbucket); 5818355f576SJeff Roberson bucket_drain(zone, cache->uc_freebucket); 582174ab450SBosko Milekic if (cache->uc_allocbucket != NULL) 583cae33c14SJeff Roberson bucket_free(cache->uc_allocbucket); 584174ab450SBosko Milekic if (cache->uc_freebucket != NULL) 585cae33c14SJeff Roberson bucket_free(cache->uc_freebucket); 586d56368d7SBosko Milekic cache->uc_allocbucket = cache->uc_freebucket = NULL; 587d56368d7SBosko Milekic } 5888355f576SJeff Roberson 5898355f576SJeff Roberson /* 5908355f576SJeff Roberson * Drain the bucket queues and free the buckets, we just keep two per 5918355f576SJeff Roberson * cpu (alloc/free). 5928355f576SJeff Roberson */ 5938355f576SJeff Roberson ZONE_LOCK(zone); 5948355f576SJeff Roberson while ((bucket = LIST_FIRST(&zone->uz_full_bucket)) != NULL) { 5958355f576SJeff Roberson LIST_REMOVE(bucket, ub_link); 5968355f576SJeff Roberson ZONE_UNLOCK(zone); 5978355f576SJeff Roberson bucket_drain(zone, bucket); 598cae33c14SJeff Roberson bucket_free(bucket); 5998355f576SJeff Roberson ZONE_LOCK(zone); 6008355f576SJeff Roberson } 6018355f576SJeff Roberson 6028355f576SJeff Roberson /* Now we do the free queue.. */ 6038355f576SJeff Roberson while ((bucket = LIST_FIRST(&zone->uz_free_bucket)) != NULL) { 6048355f576SJeff Roberson LIST_REMOVE(bucket, ub_link); 605cae33c14SJeff Roberson bucket_free(bucket); 6068355f576SJeff Roberson } 6078355f576SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 6088355f576SJeff Roberson if (CPU_ABSENT(cpu)) 6098355f576SJeff Roberson continue; 610d88797c2SBosko Milekic CPU_UNLOCK(cpu); 6118355f576SJeff Roberson } 6129643769aSJeff Roberson ZONE_UNLOCK(zone); 6138355f576SJeff Roberson } 6148355f576SJeff Roberson 6158355f576SJeff Roberson /* 6168355f576SJeff Roberson * Frees pages from a zone back to the system. This is done on demand from 6178355f576SJeff Roberson * the pageout daemon. 6188355f576SJeff Roberson * 6198355f576SJeff Roberson * Arguments: 6208355f576SJeff Roberson * zone The zone to free pages from 6219c2cd7e5SJeff Roberson * all Should we drain all items? 6228355f576SJeff Roberson * 6238355f576SJeff Roberson * Returns: 6248355f576SJeff Roberson * Nothing. 6258355f576SJeff Roberson */ 6268355f576SJeff Roberson static void 6279643769aSJeff Roberson zone_drain(uma_zone_t zone) 6288355f576SJeff Roberson { 629713deb36SJeff Roberson struct slabhead freeslabs = {}; 6308355f576SJeff Roberson uma_slab_t slab; 6318355f576SJeff Roberson uma_slab_t n; 6328355f576SJeff Roberson u_int8_t flags; 6338355f576SJeff Roberson u_int8_t *mem; 6348355f576SJeff Roberson int i; 6358355f576SJeff Roberson 6368355f576SJeff Roberson /* 6378355f576SJeff Roberson * We don't want to take pages from staticly allocated zones at this 6388355f576SJeff Roberson * time 6398355f576SJeff Roberson */ 640b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_NOFREE || zone->uz_freef == NULL) 6418355f576SJeff Roberson return; 6428355f576SJeff Roberson 6438355f576SJeff Roberson ZONE_LOCK(zone); 6448355f576SJeff Roberson 6458355f576SJeff Roberson #ifdef UMA_DEBUG 6469643769aSJeff Roberson printf("%s free items: %u\n", zone->uz_name, zone->uz_free); 6478355f576SJeff Roberson #endif 6489643769aSJeff Roberson if (zone->uz_free == 0) 6498355f576SJeff Roberson goto finished; 6508355f576SJeff Roberson 6518355f576SJeff Roberson slab = LIST_FIRST(&zone->uz_free_slab); 6529643769aSJeff Roberson while (slab) { 6538355f576SJeff Roberson n = LIST_NEXT(slab, us_link); 6548355f576SJeff Roberson 6558355f576SJeff Roberson /* We have no where to free these to */ 6568355f576SJeff Roberson if (slab->us_flags & UMA_SLAB_BOOT) { 6578355f576SJeff Roberson slab = n; 6588355f576SJeff Roberson continue; 6598355f576SJeff Roberson } 6608355f576SJeff Roberson 6618355f576SJeff Roberson LIST_REMOVE(slab, us_link); 6628355f576SJeff Roberson zone->uz_pages -= zone->uz_ppera; 6638355f576SJeff Roberson zone->uz_free -= zone->uz_ipers; 664713deb36SJeff Roberson 665b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_HASH) 666713deb36SJeff Roberson UMA_HASH_REMOVE(&zone->uz_hash, slab, slab->us_data); 667713deb36SJeff Roberson 668713deb36SJeff Roberson SLIST_INSERT_HEAD(&freeslabs, slab, us_hlink); 669713deb36SJeff Roberson 670713deb36SJeff Roberson slab = n; 671713deb36SJeff Roberson } 672713deb36SJeff Roberson finished: 673713deb36SJeff Roberson ZONE_UNLOCK(zone); 674713deb36SJeff Roberson 675713deb36SJeff Roberson while ((slab = SLIST_FIRST(&freeslabs)) != NULL) { 676713deb36SJeff Roberson SLIST_REMOVE(&freeslabs, slab, uma_slab, us_hlink); 6778355f576SJeff Roberson if (zone->uz_fini) 6788355f576SJeff Roberson for (i = 0; i < zone->uz_ipers; i++) 6798355f576SJeff Roberson zone->uz_fini( 6808355f576SJeff Roberson slab->us_data + (zone->uz_rsize * i), 6818355f576SJeff Roberson zone->uz_size); 6828355f576SJeff Roberson flags = slab->us_flags; 6838355f576SJeff Roberson mem = slab->us_data; 68499571dc3SJeff Roberson 685b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_OFFPAGE) 6868355f576SJeff Roberson uma_zfree_internal(slabzone, slab, NULL, 0); 687b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_MALLOC) { 68848eea375SJeff Roberson vm_object_t obj; 68948eea375SJeff Roberson 69048eea375SJeff Roberson if (flags & UMA_SLAB_KMEM) 69148eea375SJeff Roberson obj = kmem_object; 69248eea375SJeff Roberson else 69348eea375SJeff Roberson obj = NULL; 69499571dc3SJeff Roberson for (i = 0; i < zone->uz_ppera; i++) 69599571dc3SJeff Roberson vsetobj((vm_offset_t)mem + (i * PAGE_SIZE), 69648eea375SJeff Roberson obj); 69748eea375SJeff Roberson } 6988355f576SJeff Roberson #ifdef UMA_DEBUG 6998355f576SJeff Roberson printf("%s: Returning %d bytes.\n", 7008355f576SJeff Roberson zone->uz_name, UMA_SLAB_SIZE * zone->uz_ppera); 7018355f576SJeff Roberson #endif 7028355f576SJeff Roberson zone->uz_freef(mem, UMA_SLAB_SIZE * zone->uz_ppera, flags); 7038355f576SJeff Roberson } 7048355f576SJeff Roberson 7058355f576SJeff Roberson } 7068355f576SJeff Roberson 7078355f576SJeff Roberson /* 7088355f576SJeff Roberson * Allocate a new slab for a zone. This does not insert the slab onto a list. 7098355f576SJeff Roberson * 7108355f576SJeff Roberson * Arguments: 7118355f576SJeff Roberson * zone The zone to allocate slabs for 7128355f576SJeff Roberson * wait Shall we wait? 7138355f576SJeff Roberson * 7148355f576SJeff Roberson * Returns: 7158355f576SJeff Roberson * The slab that was allocated or NULL if there is no memory and the 7168355f576SJeff Roberson * caller specified M_NOWAIT. 7178355f576SJeff Roberson */ 7188355f576SJeff Roberson static uma_slab_t 7198355f576SJeff Roberson slab_zalloc(uma_zone_t zone, int wait) 7208355f576SJeff Roberson { 7218355f576SJeff Roberson uma_slab_t slab; /* Starting slab */ 7228355f576SJeff Roberson u_int8_t *mem; 7238355f576SJeff Roberson u_int8_t flags; 7248355f576SJeff Roberson int i; 7258355f576SJeff Roberson 726a553d4b8SJeff Roberson slab = NULL; 727a553d4b8SJeff Roberson 7288355f576SJeff Roberson #ifdef UMA_DEBUG 7298355f576SJeff Roberson printf("slab_zalloc: Allocating a new slab for %s\n", zone->uz_name); 7308355f576SJeff Roberson #endif 7318355f576SJeff Roberson ZONE_UNLOCK(zone); 732a553d4b8SJeff Roberson 733b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_OFFPAGE) { 734bbee39c6SJeff Roberson slab = uma_zalloc_internal(slabzone, NULL, wait); 735a553d4b8SJeff Roberson if (slab == NULL) { 736a553d4b8SJeff Roberson ZONE_LOCK(zone); 737a553d4b8SJeff Roberson return NULL; 738a553d4b8SJeff Roberson } 739a553d4b8SJeff Roberson } 740a553d4b8SJeff Roberson 7413370c5bfSJeff Roberson /* 7423370c5bfSJeff Roberson * This reproduces the old vm_zone behavior of zero filling pages the 7433370c5bfSJeff Roberson * first time they are added to a zone. 7443370c5bfSJeff Roberson * 7453370c5bfSJeff Roberson * Malloced items are zeroed in uma_zalloc. 7463370c5bfSJeff Roberson */ 7473370c5bfSJeff Roberson 748b60f5b79SJeff Roberson if ((zone->uz_flags & UMA_ZONE_MALLOC) == 0) 7493370c5bfSJeff Roberson wait |= M_ZERO; 7503370c5bfSJeff Roberson else 7513370c5bfSJeff Roberson wait &= ~M_ZERO; 7523370c5bfSJeff Roberson 753234c7726SAlan Cox mem = zone->uz_allocf(zone, zone->uz_ppera * UMA_SLAB_SIZE, 754234c7726SAlan Cox &flags, wait); 755a553d4b8SJeff Roberson if (mem == NULL) { 7568355f576SJeff Roberson ZONE_LOCK(zone); 7578355f576SJeff Roberson return (NULL); 758a553d4b8SJeff Roberson } 7598355f576SJeff Roberson 7605c0e403bSJeff Roberson /* Point the slab into the allocated memory */ 761b60f5b79SJeff Roberson if (!(zone->uz_flags & UMA_ZONE_OFFPAGE)) 7628355f576SJeff Roberson slab = (uma_slab_t )(mem + zone->uz_pgoff); 7635c0e403bSJeff Roberson 764b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_MALLOC) 76599571dc3SJeff Roberson for (i = 0; i < zone->uz_ppera; i++) 76699571dc3SJeff Roberson vsetslab((vm_offset_t)mem + (i * PAGE_SIZE), slab); 7678355f576SJeff Roberson 7688355f576SJeff Roberson slab->us_zone = zone; 7698355f576SJeff Roberson slab->us_data = mem; 7708355f576SJeff Roberson slab->us_freecount = zone->uz_ipers; 7718355f576SJeff Roberson slab->us_firstfree = 0; 7728355f576SJeff Roberson slab->us_flags = flags; 7738355f576SJeff Roberson for (i = 0; i < zone->uz_ipers; i++) 7748355f576SJeff Roberson slab->us_freelist[i] = i+1; 7758355f576SJeff Roberson 7768355f576SJeff Roberson if (zone->uz_init) 7778355f576SJeff Roberson for (i = 0; i < zone->uz_ipers; i++) 7788355f576SJeff Roberson zone->uz_init(slab->us_data + (zone->uz_rsize * i), 7798355f576SJeff Roberson zone->uz_size); 7805c0e403bSJeff Roberson ZONE_LOCK(zone); 7815c0e403bSJeff Roberson 782b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_HASH) 7835c0e403bSJeff Roberson UMA_HASH_INSERT(&zone->uz_hash, slab, mem); 7848355f576SJeff Roberson 7858355f576SJeff Roberson zone->uz_pages += zone->uz_ppera; 7868355f576SJeff Roberson zone->uz_free += zone->uz_ipers; 7878355f576SJeff Roberson 7888355f576SJeff Roberson return (slab); 7898355f576SJeff Roberson } 7908355f576SJeff Roberson 7918355f576SJeff Roberson /* 792009b6fcbSJeff Roberson * This function is intended to be used early on in place of page_alloc() so 793009b6fcbSJeff Roberson * that we may use the boot time page cache to satisfy allocations before 794009b6fcbSJeff Roberson * the VM is ready. 795009b6fcbSJeff Roberson */ 796009b6fcbSJeff Roberson static void * 797009b6fcbSJeff Roberson startup_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait) 798009b6fcbSJeff Roberson { 799009b6fcbSJeff Roberson /* 800009b6fcbSJeff Roberson * Check our small startup cache to see if it has pages remaining. 801009b6fcbSJeff Roberson */ 802009b6fcbSJeff Roberson mtx_lock(&uma_mtx); 803009b6fcbSJeff Roberson if (uma_boot_free != 0) { 804009b6fcbSJeff Roberson uma_slab_t tmps; 805009b6fcbSJeff Roberson 806009b6fcbSJeff Roberson tmps = LIST_FIRST(&uma_boot_pages); 807009b6fcbSJeff Roberson LIST_REMOVE(tmps, us_link); 808009b6fcbSJeff Roberson uma_boot_free--; 809009b6fcbSJeff Roberson mtx_unlock(&uma_mtx); 810009b6fcbSJeff Roberson *pflag = tmps->us_flags; 811009b6fcbSJeff Roberson return (tmps->us_data); 812009b6fcbSJeff Roberson } 813009b6fcbSJeff Roberson mtx_unlock(&uma_mtx); 814009b6fcbSJeff Roberson if (booted == 0) 815009b6fcbSJeff Roberson panic("UMA: Increase UMA_BOOT_PAGES"); 816009b6fcbSJeff Roberson /* 817009b6fcbSJeff Roberson * Now that we've booted reset these users to their real allocator. 818009b6fcbSJeff Roberson */ 819009b6fcbSJeff Roberson #ifdef UMA_MD_SMALL_ALLOC 820009b6fcbSJeff Roberson zone->uz_allocf = uma_small_alloc; 821009b6fcbSJeff Roberson #else 822009b6fcbSJeff Roberson zone->uz_allocf = page_alloc; 823009b6fcbSJeff Roberson #endif 824009b6fcbSJeff Roberson return zone->uz_allocf(zone, bytes, pflag, wait); 825009b6fcbSJeff Roberson } 826009b6fcbSJeff Roberson 827009b6fcbSJeff Roberson /* 8288355f576SJeff Roberson * Allocates a number of pages from the system 8298355f576SJeff Roberson * 8308355f576SJeff Roberson * Arguments: 8318355f576SJeff Roberson * zone Unused 8328355f576SJeff Roberson * bytes The number of bytes requested 8338355f576SJeff Roberson * wait Shall we wait? 8348355f576SJeff Roberson * 8358355f576SJeff Roberson * Returns: 8368355f576SJeff Roberson * A pointer to the alloced memory or possibly 8378355f576SJeff Roberson * NULL if M_NOWAIT is set. 8388355f576SJeff Roberson */ 8398355f576SJeff Roberson static void * 8408355f576SJeff Roberson page_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait) 8418355f576SJeff Roberson { 8428355f576SJeff Roberson void *p; /* Returned page */ 8438355f576SJeff Roberson 8448355f576SJeff Roberson *pflag = UMA_SLAB_KMEM; 8458355f576SJeff Roberson p = (void *) kmem_malloc(kmem_map, bytes, wait); 8468355f576SJeff Roberson 8478355f576SJeff Roberson return (p); 8488355f576SJeff Roberson } 8498355f576SJeff Roberson 8508355f576SJeff Roberson /* 8518355f576SJeff Roberson * Allocates a number of pages from within an object 8528355f576SJeff Roberson * 8538355f576SJeff Roberson * Arguments: 8548355f576SJeff Roberson * zone Unused 8558355f576SJeff Roberson * bytes The number of bytes requested 8568355f576SJeff Roberson * wait Shall we wait? 8578355f576SJeff Roberson * 8588355f576SJeff Roberson * Returns: 8598355f576SJeff Roberson * A pointer to the alloced memory or possibly 8608355f576SJeff Roberson * NULL if M_NOWAIT is set. 8618355f576SJeff Roberson */ 8628355f576SJeff Roberson static void * 8638355f576SJeff Roberson obj_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait) 8648355f576SJeff Roberson { 865b245ac95SAlan Cox vm_object_t object; 866b245ac95SAlan Cox vm_offset_t retkva, zkva; 8678355f576SJeff Roberson vm_page_t p; 868b245ac95SAlan Cox int pages, startpages; 8698355f576SJeff Roberson 870b245ac95SAlan Cox object = zone->uz_obj; 87155f7c614SArchie Cobbs retkva = 0; 8728355f576SJeff Roberson 8738355f576SJeff Roberson /* 87464f051e9SJeff Roberson * This looks a little weird since we're getting one page at a time. 8758355f576SJeff Roberson */ 876b245ac95SAlan Cox VM_OBJECT_LOCK(object); 877b245ac95SAlan Cox p = TAILQ_LAST(&object->memq, pglist); 878b245ac95SAlan Cox pages = p != NULL ? p->pindex + 1 : 0; 879b245ac95SAlan Cox startpages = pages; 8808355f576SJeff Roberson zkva = zone->uz_kva + pages * PAGE_SIZE; 881b245ac95SAlan Cox for (; bytes > 0; bytes -= PAGE_SIZE) { 882b245ac95SAlan Cox p = vm_page_alloc(object, pages, 883b245ac95SAlan Cox VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED); 884b245ac95SAlan Cox if (p == NULL) { 885b245ac95SAlan Cox if (pages != startpages) 886b245ac95SAlan Cox pmap_qremove(retkva, pages - startpages); 887b245ac95SAlan Cox while (pages != startpages) { 888b245ac95SAlan Cox pages--; 889b245ac95SAlan Cox p = TAILQ_LAST(&object->memq, pglist); 890b245ac95SAlan Cox vm_page_lock_queues(); 891b245ac95SAlan Cox vm_page_unwire(p, 0); 892b245ac95SAlan Cox vm_page_free(p); 893b245ac95SAlan Cox vm_page_unlock_queues(); 894b245ac95SAlan Cox } 895b245ac95SAlan Cox retkva = 0; 896b245ac95SAlan Cox goto done; 897b245ac95SAlan Cox } 898b245ac95SAlan Cox pmap_qenter(zkva, &p, 1); 89955f7c614SArchie Cobbs if (retkva == 0) 9008355f576SJeff Roberson retkva = zkva; 901b245ac95SAlan Cox zkva += PAGE_SIZE; 9028355f576SJeff Roberson pages += 1; 9038355f576SJeff Roberson } 904b245ac95SAlan Cox done: 905b245ac95SAlan Cox VM_OBJECT_UNLOCK(object); 9068355f576SJeff Roberson *flags = UMA_SLAB_PRIV; 9078355f576SJeff Roberson 9088355f576SJeff Roberson return ((void *)retkva); 9098355f576SJeff Roberson } 9108355f576SJeff Roberson 9118355f576SJeff Roberson /* 9128355f576SJeff Roberson * Frees a number of pages to the system 9138355f576SJeff Roberson * 9148355f576SJeff Roberson * Arguments: 9158355f576SJeff Roberson * mem A pointer to the memory to be freed 9168355f576SJeff Roberson * size The size of the memory being freed 9178355f576SJeff Roberson * flags The original p->us_flags field 9188355f576SJeff Roberson * 9198355f576SJeff Roberson * Returns: 9208355f576SJeff Roberson * Nothing 9218355f576SJeff Roberson */ 9228355f576SJeff Roberson static void 9238355f576SJeff Roberson page_free(void *mem, int size, u_int8_t flags) 9248355f576SJeff Roberson { 9258355f576SJeff Roberson vm_map_t map; 9263370c5bfSJeff Roberson 9278355f576SJeff Roberson if (flags & UMA_SLAB_KMEM) 9288355f576SJeff Roberson map = kmem_map; 9298355f576SJeff Roberson else 9308355f576SJeff Roberson panic("UMA: page_free used with invalid flags %d\n", flags); 9318355f576SJeff Roberson 9328355f576SJeff Roberson kmem_free(map, (vm_offset_t)mem, size); 9338355f576SJeff Roberson } 9348355f576SJeff Roberson 9358355f576SJeff Roberson /* 9368355f576SJeff Roberson * Zero fill initializer 9378355f576SJeff Roberson * 9388355f576SJeff Roberson * Arguments/Returns follow uma_init specifications 9398355f576SJeff Roberson */ 9408355f576SJeff Roberson static void 9418355f576SJeff Roberson zero_init(void *mem, int size) 9428355f576SJeff Roberson { 9438355f576SJeff Roberson bzero(mem, size); 9448355f576SJeff Roberson } 9458355f576SJeff Roberson 9468355f576SJeff Roberson /* 9478355f576SJeff Roberson * Finish creating a small uma zone. This calculates ipers, and the zone size. 9488355f576SJeff Roberson * 9498355f576SJeff Roberson * Arguments 9508355f576SJeff Roberson * zone The zone we should initialize 9518355f576SJeff Roberson * 9528355f576SJeff Roberson * Returns 9538355f576SJeff Roberson * Nothing 9548355f576SJeff Roberson */ 9558355f576SJeff Roberson static void 9568355f576SJeff Roberson zone_small_init(uma_zone_t zone) 9578355f576SJeff Roberson { 9588355f576SJeff Roberson int rsize; 9598355f576SJeff Roberson int memused; 9608355f576SJeff Roberson int ipers; 9618355f576SJeff Roberson 9628355f576SJeff Roberson rsize = zone->uz_size; 9638355f576SJeff Roberson 9648355f576SJeff Roberson if (rsize < UMA_SMALLEST_UNIT) 9658355f576SJeff Roberson rsize = UMA_SMALLEST_UNIT; 9668355f576SJeff Roberson 9678355f576SJeff Roberson if (rsize & zone->uz_align) 9688355f576SJeff Roberson rsize = (rsize & ~zone->uz_align) + (zone->uz_align + 1); 9698355f576SJeff Roberson 9708355f576SJeff Roberson zone->uz_rsize = rsize; 9718355f576SJeff Roberson 9728355f576SJeff Roberson rsize += 1; /* Account for the byte of linkage */ 9738355f576SJeff Roberson zone->uz_ipers = (UMA_SLAB_SIZE - sizeof(struct uma_slab)) / rsize; 9748355f576SJeff Roberson zone->uz_ppera = 1; 9758355f576SJeff Roberson 97620e8e865SBosko Milekic KASSERT(zone->uz_ipers != 0, ("zone_small_init: ipers is 0, uh-oh!")); 9778355f576SJeff Roberson memused = zone->uz_ipers * zone->uz_rsize; 9788355f576SJeff Roberson 9798355f576SJeff Roberson /* Can we do any better? */ 9808355f576SJeff Roberson if ((UMA_SLAB_SIZE - memused) >= UMA_MAX_WASTE) { 98120e8e865SBosko Milekic /* 98220e8e865SBosko Milekic * We can't do this if we're internal or if we've been 98320e8e865SBosko Milekic * asked to not go to the VM for buckets. If we do this we 98420e8e865SBosko Milekic * may end up going to the VM (kmem_map) for slabs which we 98520e8e865SBosko Milekic * do not want to do if we're UMA_ZFLAG_CACHEONLY as a 98620e8e865SBosko Milekic * result of UMA_ZONE_VM, which clearly forbids it. 98720e8e865SBosko Milekic */ 98820e8e865SBosko Milekic if ((zone->uz_flags & UMA_ZFLAG_INTERNAL) || 98920e8e865SBosko Milekic (zone->uz_flags & UMA_ZFLAG_CACHEONLY)) 9908355f576SJeff Roberson return; 9918355f576SJeff Roberson ipers = UMA_SLAB_SIZE / zone->uz_rsize; 9928355f576SJeff Roberson if (ipers > zone->uz_ipers) { 993b60f5b79SJeff Roberson zone->uz_flags |= UMA_ZONE_OFFPAGE; 994b60f5b79SJeff Roberson if ((zone->uz_flags & UMA_ZONE_MALLOC) == 0) 995b60f5b79SJeff Roberson zone->uz_flags |= UMA_ZONE_HASH; 9968355f576SJeff Roberson zone->uz_ipers = ipers; 9978355f576SJeff Roberson } 9988355f576SJeff Roberson } 9998355f576SJeff Roberson } 10008355f576SJeff Roberson 10018355f576SJeff Roberson /* 10028355f576SJeff Roberson * Finish creating a large (> UMA_SLAB_SIZE) uma zone. Just give in and do 10038355f576SJeff Roberson * OFFPAGE for now. When I can allow for more dynamic slab sizes this will be 10048355f576SJeff Roberson * more complicated. 10058355f576SJeff Roberson * 10068355f576SJeff Roberson * Arguments 10078355f576SJeff Roberson * zone The zone we should initialize 10088355f576SJeff Roberson * 10098355f576SJeff Roberson * Returns 10108355f576SJeff Roberson * Nothing 10118355f576SJeff Roberson */ 10128355f576SJeff Roberson static void 10138355f576SJeff Roberson zone_large_init(uma_zone_t zone) 10148355f576SJeff Roberson { 10158355f576SJeff Roberson int pages; 10168355f576SJeff Roberson 101720e8e865SBosko Milekic KASSERT((zone->uz_flags & UMA_ZFLAG_CACHEONLY) == 0, 101820e8e865SBosko Milekic ("zone_large_init: Cannot large-init a UMA_ZFLAG_CACHEONLY zone")); 101920e8e865SBosko Milekic 10208355f576SJeff Roberson pages = zone->uz_size / UMA_SLAB_SIZE; 10218355f576SJeff Roberson 10228355f576SJeff Roberson /* Account for remainder */ 10238355f576SJeff Roberson if ((pages * UMA_SLAB_SIZE) < zone->uz_size) 10248355f576SJeff Roberson pages++; 10258355f576SJeff Roberson 10268355f576SJeff Roberson zone->uz_ppera = pages; 10278355f576SJeff Roberson zone->uz_ipers = 1; 10288355f576SJeff Roberson 1029b60f5b79SJeff Roberson zone->uz_flags |= UMA_ZONE_OFFPAGE; 1030b60f5b79SJeff Roberson if ((zone->uz_flags & UMA_ZONE_MALLOC) == 0) 1031b60f5b79SJeff Roberson zone->uz_flags |= UMA_ZONE_HASH; 103299571dc3SJeff Roberson 10338355f576SJeff Roberson zone->uz_rsize = zone->uz_size; 10348355f576SJeff Roberson } 10358355f576SJeff Roberson 10368355f576SJeff Roberson /* 10378355f576SJeff Roberson * Zone header ctor. This initializes all fields, locks, etc. And inserts 10388355f576SJeff Roberson * the zone onto the global zone list. 10398355f576SJeff Roberson * 10408355f576SJeff Roberson * Arguments/Returns follow uma_ctor specifications 10418355f576SJeff Roberson * udata Actually uma_zcreat_args 10428355f576SJeff Roberson */ 10438355f576SJeff Roberson 10448355f576SJeff Roberson static void 10458355f576SJeff Roberson zone_ctor(void *mem, int size, void *udata) 10468355f576SJeff Roberson { 10478355f576SJeff Roberson struct uma_zctor_args *arg = udata; 10488355f576SJeff Roberson uma_zone_t zone = mem; 104928bc4419SJeff Roberson int privlc; 10508355f576SJeff Roberson 10518355f576SJeff Roberson bzero(zone, size); 10528355f576SJeff Roberson zone->uz_name = arg->name; 10538355f576SJeff Roberson zone->uz_size = arg->size; 10548355f576SJeff Roberson zone->uz_ctor = arg->ctor; 10558355f576SJeff Roberson zone->uz_dtor = arg->dtor; 10568355f576SJeff Roberson zone->uz_init = arg->uminit; 1057e221e841SJeff Roberson zone->uz_fini = arg->fini; 10588355f576SJeff Roberson zone->uz_align = arg->align; 10598355f576SJeff Roberson zone->uz_free = 0; 10608355f576SJeff Roberson zone->uz_pages = 0; 1061b60f5b79SJeff Roberson zone->uz_flags = arg->flags; 10628355f576SJeff Roberson zone->uz_allocf = page_alloc; 10638355f576SJeff Roberson zone->uz_freef = page_free; 10648355f576SJeff Roberson 10658355f576SJeff Roberson if (arg->flags & UMA_ZONE_ZINIT) 10668355f576SJeff Roberson zone->uz_init = zero_init; 10678355f576SJeff Roberson 106818aa2de5SJeff Roberson if (arg->flags & UMA_ZONE_VM) 106920e8e865SBosko Milekic zone->uz_flags |= UMA_ZFLAG_CACHEONLY; 107018aa2de5SJeff Roberson 107120e8e865SBosko Milekic /* 107220e8e865SBosko Milekic * XXX: 107320e8e865SBosko Milekic * The +1 byte added to uz_size is to account for the byte of 107420e8e865SBosko Milekic * linkage that is added to the size in zone_small_init(). If 107520e8e865SBosko Milekic * we don't account for this here then we may end up in 107620e8e865SBosko Milekic * zone_small_init() with a calculated 'ipers' of 0. 107720e8e865SBosko Milekic */ 107820e8e865SBosko Milekic if ((zone->uz_size+1) > (UMA_SLAB_SIZE - sizeof(struct uma_slab))) 10798355f576SJeff Roberson zone_large_init(zone); 10808355f576SJeff Roberson else 10818355f576SJeff Roberson zone_small_init(zone); 1082009b6fcbSJeff Roberson /* 1083009b6fcbSJeff Roberson * If we haven't booted yet we need allocations to go through the 1084009b6fcbSJeff Roberson * startup cache until the vm is ready. 1085009b6fcbSJeff Roberson */ 108648eea375SJeff Roberson if (zone->uz_ppera == 1) { 1087009b6fcbSJeff Roberson #ifdef UMA_MD_SMALL_ALLOC 108848eea375SJeff Roberson zone->uz_allocf = uma_small_alloc; 108948eea375SJeff Roberson zone->uz_freef = uma_small_free; 1090009b6fcbSJeff Roberson #endif 1091009b6fcbSJeff Roberson if (booted == 0) 1092009b6fcbSJeff Roberson zone->uz_allocf = startup_alloc; 109348eea375SJeff Roberson } 109428bc4419SJeff Roberson if (arg->flags & UMA_ZONE_MTXCLASS) 109528bc4419SJeff Roberson privlc = 1; 109628bc4419SJeff Roberson else 109728bc4419SJeff Roberson privlc = 0; 109828bc4419SJeff Roberson 10998355f576SJeff Roberson /* 11008355f576SJeff Roberson * If we're putting the slab header in the actual page we need to 11018355f576SJeff Roberson * figure out where in each page it goes. This calculates a right 11029d5abbddSJens Schweikhardt * justified offset into the memory on an ALIGN_PTR boundary. 11038355f576SJeff Roberson */ 1104b60f5b79SJeff Roberson if (!(zone->uz_flags & UMA_ZONE_OFFPAGE)) { 11058355f576SJeff Roberson int totsize; 11068355f576SJeff Roberson 11078355f576SJeff Roberson /* Size of the slab struct and free list */ 11088355f576SJeff Roberson totsize = sizeof(struct uma_slab) + zone->uz_ipers; 11098355f576SJeff Roberson if (totsize & UMA_ALIGN_PTR) 11108355f576SJeff Roberson totsize = (totsize & ~UMA_ALIGN_PTR) + 11118355f576SJeff Roberson (UMA_ALIGN_PTR + 1); 11128355f576SJeff Roberson zone->uz_pgoff = UMA_SLAB_SIZE - totsize; 11138355f576SJeff Roberson totsize = zone->uz_pgoff + sizeof(struct uma_slab) 11148355f576SJeff Roberson + zone->uz_ipers; 11158355f576SJeff Roberson /* I don't think it's possible, but I'll make sure anyway */ 11168355f576SJeff Roberson if (totsize > UMA_SLAB_SIZE) { 11178355f576SJeff Roberson printf("zone %s ipers %d rsize %d size %d\n", 11188355f576SJeff Roberson zone->uz_name, zone->uz_ipers, zone->uz_rsize, 11198355f576SJeff Roberson zone->uz_size); 11208355f576SJeff Roberson panic("UMA slab won't fit.\n"); 11218355f576SJeff Roberson } 11228355f576SJeff Roberson } 11238355f576SJeff Roberson 1124b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_HASH) 112599571dc3SJeff Roberson hash_alloc(&zone->uz_hash); 112699571dc3SJeff Roberson 11278355f576SJeff Roberson #ifdef UMA_DEBUG 11288355f576SJeff Roberson printf("%s(%p) size = %d ipers = %d ppera = %d pgoff = %d\n", 11298355f576SJeff Roberson zone->uz_name, zone, 11308355f576SJeff Roberson zone->uz_size, zone->uz_ipers, 11318355f576SJeff Roberson zone->uz_ppera, zone->uz_pgoff); 11328355f576SJeff Roberson #endif 113328bc4419SJeff Roberson ZONE_LOCK_INIT(zone, privlc); 11348355f576SJeff Roberson 11358355f576SJeff Roberson mtx_lock(&uma_mtx); 11368355f576SJeff Roberson LIST_INSERT_HEAD(&uma_zones, zone, uz_link); 11378355f576SJeff Roberson mtx_unlock(&uma_mtx); 11388355f576SJeff Roberson 11398355f576SJeff Roberson /* 11408355f576SJeff Roberson * Some internal zones don't have room allocated for the per cpu 11418355f576SJeff Roberson * caches. If we're internal, bail out here. 11428355f576SJeff Roberson */ 11438355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_INTERNAL) 11448355f576SJeff Roberson return; 11458355f576SJeff Roberson 1146cae33c14SJeff Roberson if (zone->uz_ipers <= BUCKET_MAX) 1147cae33c14SJeff Roberson zone->uz_count = zone->uz_ipers; 11488355f576SJeff Roberson else 1149cae33c14SJeff Roberson zone->uz_count = BUCKET_MAX; 11508355f576SJeff Roberson } 11518355f576SJeff Roberson 11528355f576SJeff Roberson /* 11539c2cd7e5SJeff Roberson * Zone header dtor. This frees all data, destroys locks, frees the hash table 11549c2cd7e5SJeff Roberson * and removes the zone from the global list. 11559c2cd7e5SJeff Roberson * 11569c2cd7e5SJeff Roberson * Arguments/Returns follow uma_dtor specifications 11579c2cd7e5SJeff Roberson * udata unused 11589c2cd7e5SJeff Roberson */ 11599c2cd7e5SJeff Roberson 11609c2cd7e5SJeff Roberson static void 11619c2cd7e5SJeff Roberson zone_dtor(void *arg, int size, void *udata) 11629c2cd7e5SJeff Roberson { 11639c2cd7e5SJeff Roberson uma_zone_t zone; 11649c2cd7e5SJeff Roberson 11659c2cd7e5SJeff Roberson zone = (uma_zone_t)arg; 11669643769aSJeff Roberson 11679643769aSJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) 11689643769aSJeff Roberson cache_drain(zone); 116917b9cc49SJeff Roberson mtx_lock(&uma_mtx); 117017b9cc49SJeff Roberson LIST_REMOVE(zone, uz_link); 11719643769aSJeff Roberson zone_drain(zone); 117217b9cc49SJeff Roberson mtx_unlock(&uma_mtx); 117317b9cc49SJeff Roberson 11749c2cd7e5SJeff Roberson ZONE_LOCK(zone); 11759c2cd7e5SJeff Roberson if (zone->uz_free != 0) 117664f051e9SJeff Roberson printf("Zone %s was not empty (%d items). " 117764f051e9SJeff Roberson " Lost %d pages of memory.\n", 1178886eaaacSPoul-Henning Kamp zone->uz_name, zone->uz_free, zone->uz_pages); 11799c2cd7e5SJeff Roberson 11809c2cd7e5SJeff Roberson ZONE_UNLOCK(zone); 1181b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_HASH) 11820aef6126SJeff Roberson hash_free(&zone->uz_hash); 11830aef6126SJeff Roberson 11849c2cd7e5SJeff Roberson ZONE_LOCK_FINI(zone); 11859c2cd7e5SJeff Roberson } 11869c2cd7e5SJeff Roberson /* 11878355f576SJeff Roberson * Traverses every zone in the system and calls a callback 11888355f576SJeff Roberson * 11898355f576SJeff Roberson * Arguments: 11908355f576SJeff Roberson * zfunc A pointer to a function which accepts a zone 11918355f576SJeff Roberson * as an argument. 11928355f576SJeff Roberson * 11938355f576SJeff Roberson * Returns: 11948355f576SJeff Roberson * Nothing 11958355f576SJeff Roberson */ 11968355f576SJeff Roberson static void 11978355f576SJeff Roberson zone_foreach(void (*zfunc)(uma_zone_t)) 11988355f576SJeff Roberson { 11998355f576SJeff Roberson uma_zone_t zone; 12008355f576SJeff Roberson 12018355f576SJeff Roberson mtx_lock(&uma_mtx); 120264f051e9SJeff Roberson LIST_FOREACH(zone, &uma_zones, uz_link) 12038355f576SJeff Roberson zfunc(zone); 12048355f576SJeff Roberson mtx_unlock(&uma_mtx); 12058355f576SJeff Roberson } 12068355f576SJeff Roberson 12078355f576SJeff Roberson /* Public functions */ 12088355f576SJeff Roberson /* See uma.h */ 12098355f576SJeff Roberson void 12108355f576SJeff Roberson uma_startup(void *bootmem) 12118355f576SJeff Roberson { 12128355f576SJeff Roberson struct uma_zctor_args args; 12138355f576SJeff Roberson uma_slab_t slab; 12148355f576SJeff Roberson int slabsize; 12158355f576SJeff Roberson int i; 12168355f576SJeff Roberson 12178355f576SJeff Roberson #ifdef UMA_DEBUG 12188355f576SJeff Roberson printf("Creating uma zone headers zone.\n"); 12198355f576SJeff Roberson #endif 12208355f576SJeff Roberson #ifdef SMP 12218355f576SJeff Roberson maxcpu = mp_maxid + 1; 12228355f576SJeff Roberson #else 12238355f576SJeff Roberson maxcpu = 1; 12248355f576SJeff Roberson #endif 12258355f576SJeff Roberson #ifdef UMA_DEBUG 12268355f576SJeff Roberson printf("Max cpu = %d, mp_maxid = %d\n", maxcpu, mp_maxid); 12278355f576SJeff Roberson #endif 12286008862bSJohn Baldwin mtx_init(&uma_mtx, "UMA lock", NULL, MTX_DEF); 12298355f576SJeff Roberson /* "manually" Create the initial zone */ 12308355f576SJeff Roberson args.name = "UMA Zones"; 12318355f576SJeff Roberson args.size = sizeof(struct uma_zone) + 12328355f576SJeff Roberson (sizeof(struct uma_cache) * (maxcpu - 1)); 12338355f576SJeff Roberson args.ctor = zone_ctor; 12349c2cd7e5SJeff Roberson args.dtor = zone_dtor; 12358355f576SJeff Roberson args.uminit = zero_init; 12368355f576SJeff Roberson args.fini = NULL; 12378355f576SJeff Roberson args.align = 32 - 1; 1238b60f5b79SJeff Roberson args.flags = UMA_ZFLAG_INTERNAL; 12398355f576SJeff Roberson /* The initial zone has no Per cpu queues so it's smaller */ 12408355f576SJeff Roberson zone_ctor(zones, sizeof(struct uma_zone), &args); 12418355f576SJeff Roberson 1242d88797c2SBosko Milekic /* Initialize the pcpu cache lock set once and for all */ 1243d88797c2SBosko Milekic for (i = 0; i < maxcpu; i++) 1244d88797c2SBosko Milekic CPU_LOCK_INIT(i); 1245d88797c2SBosko Milekic 12468355f576SJeff Roberson #ifdef UMA_DEBUG 12478355f576SJeff Roberson printf("Filling boot free list.\n"); 12488355f576SJeff Roberson #endif 12498355f576SJeff Roberson for (i = 0; i < UMA_BOOT_PAGES; i++) { 12508355f576SJeff Roberson slab = (uma_slab_t)((u_int8_t *)bootmem + (i * UMA_SLAB_SIZE)); 12518355f576SJeff Roberson slab->us_data = (u_int8_t *)slab; 12528355f576SJeff Roberson slab->us_flags = UMA_SLAB_BOOT; 12538355f576SJeff Roberson LIST_INSERT_HEAD(&uma_boot_pages, slab, us_link); 12548355f576SJeff Roberson uma_boot_free++; 12558355f576SJeff Roberson } 12568355f576SJeff Roberson 12578355f576SJeff Roberson #ifdef UMA_DEBUG 12588355f576SJeff Roberson printf("Creating slab zone.\n"); 12598355f576SJeff Roberson #endif 12608355f576SJeff Roberson 12618355f576SJeff Roberson /* 12628355f576SJeff Roberson * This is the max number of free list items we'll have with 12638355f576SJeff Roberson * offpage slabs. 12648355f576SJeff Roberson */ 12658355f576SJeff Roberson slabsize = UMA_SLAB_SIZE - sizeof(struct uma_slab); 12668355f576SJeff Roberson slabsize /= UMA_MAX_WASTE; 12678355f576SJeff Roberson slabsize++; /* In case there it's rounded */ 12688355f576SJeff Roberson slabsize += sizeof(struct uma_slab); 12698355f576SJeff Roberson 12708355f576SJeff Roberson /* Now make a zone for slab headers */ 12718355f576SJeff Roberson slabzone = uma_zcreate("UMA Slabs", 12728355f576SJeff Roberson slabsize, 12738355f576SJeff Roberson NULL, NULL, NULL, NULL, 1274b60f5b79SJeff Roberson UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); 12758355f576SJeff Roberson 12768355f576SJeff Roberson hashzone = uma_zcreate("UMA Hash", 12778355f576SJeff Roberson sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT, 12788355f576SJeff Roberson NULL, NULL, NULL, NULL, 1279b60f5b79SJeff Roberson UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); 12808355f576SJeff Roberson 1281cae33c14SJeff Roberson bucket_init(); 12828355f576SJeff Roberson 1283009b6fcbSJeff Roberson #ifdef UMA_MD_SMALL_ALLOC 128448eea375SJeff Roberson booted = 1; 128548eea375SJeff Roberson #endif 12868355f576SJeff Roberson 12878355f576SJeff Roberson #ifdef UMA_DEBUG 12888355f576SJeff Roberson printf("UMA startup complete.\n"); 12898355f576SJeff Roberson #endif 12908355f576SJeff Roberson } 12918355f576SJeff Roberson 12928355f576SJeff Roberson /* see uma.h */ 12938355f576SJeff Roberson void 129499571dc3SJeff Roberson uma_startup2(void) 12958355f576SJeff Roberson { 12968355f576SJeff Roberson booted = 1; 129786bbae32SJeff Roberson bucket_enable(); 12988355f576SJeff Roberson #ifdef UMA_DEBUG 12998355f576SJeff Roberson printf("UMA startup2 complete.\n"); 13008355f576SJeff Roberson #endif 13018355f576SJeff Roberson } 13028355f576SJeff Roberson 13038355f576SJeff Roberson /* 13048355f576SJeff Roberson * Initialize our callout handle 13058355f576SJeff Roberson * 13068355f576SJeff Roberson */ 13078355f576SJeff Roberson 13088355f576SJeff Roberson static void 13098355f576SJeff Roberson uma_startup3(void) 13108355f576SJeff Roberson { 13118355f576SJeff Roberson #ifdef UMA_DEBUG 13128355f576SJeff Roberson printf("Starting callout.\n"); 13138355f576SJeff Roberson #endif 13148355f576SJeff Roberson callout_init(&uma_callout, 0); 13159643769aSJeff Roberson callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); 13168355f576SJeff Roberson #ifdef UMA_DEBUG 13178355f576SJeff Roberson printf("UMA startup3 complete.\n"); 13188355f576SJeff Roberson #endif 13198355f576SJeff Roberson } 13208355f576SJeff Roberson 13218355f576SJeff Roberson /* See uma.h */ 13228355f576SJeff Roberson uma_zone_t 1323c3bdc05fSAndrew R. Reiter uma_zcreate(char *name, size_t size, uma_ctor ctor, uma_dtor dtor, 1324c3bdc05fSAndrew R. Reiter uma_init uminit, uma_fini fini, int align, u_int16_t flags) 13258355f576SJeff Roberson 13268355f576SJeff Roberson { 13278355f576SJeff Roberson struct uma_zctor_args args; 13288355f576SJeff Roberson 13298355f576SJeff Roberson /* This stuff is essential for the zone ctor */ 13308355f576SJeff Roberson args.name = name; 13318355f576SJeff Roberson args.size = size; 13328355f576SJeff Roberson args.ctor = ctor; 13338355f576SJeff Roberson args.dtor = dtor; 13348355f576SJeff Roberson args.uminit = uminit; 13358355f576SJeff Roberson args.fini = fini; 13368355f576SJeff Roberson args.align = align; 13378355f576SJeff Roberson args.flags = flags; 13388355f576SJeff Roberson 1339a163d034SWarner Losh return (uma_zalloc_internal(zones, &args, M_WAITOK)); 13408355f576SJeff Roberson } 13418355f576SJeff Roberson 13428355f576SJeff Roberson /* See uma.h */ 13439c2cd7e5SJeff Roberson void 13449c2cd7e5SJeff Roberson uma_zdestroy(uma_zone_t zone) 13459c2cd7e5SJeff Roberson { 13469c2cd7e5SJeff Roberson uma_zfree_internal(zones, zone, NULL, 0); 13479c2cd7e5SJeff Roberson } 13489c2cd7e5SJeff Roberson 13499c2cd7e5SJeff Roberson /* See uma.h */ 13508355f576SJeff Roberson void * 13512cc35ff9SJeff Roberson uma_zalloc_arg(uma_zone_t zone, void *udata, int flags) 13528355f576SJeff Roberson { 13538355f576SJeff Roberson void *item; 13548355f576SJeff Roberson uma_cache_t cache; 13558355f576SJeff Roberson uma_bucket_t bucket; 13568355f576SJeff Roberson int cpu; 13578355f576SJeff Roberson 13588355f576SJeff Roberson /* This is the fast path allocation */ 13598355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC_1 13608355f576SJeff Roberson printf("Allocating one item from %s(%p)\n", zone->uz_name, zone); 13618355f576SJeff Roberson #endif 1362a553d4b8SJeff Roberson 13638522511bSHartmut Brandt #ifdef INVARIANTS 13648522511bSHartmut Brandt /* 13658522511bSHartmut Brandt * To make sure that WAITOK or NOWAIT is set, but not more than 13668522511bSHartmut Brandt * one, and check against the API botches that are common. 13678522511bSHartmut Brandt * The uma code implies M_WAITOK if M_NOWAIT is not set, so 13688522511bSHartmut Brandt * we default to waiting if none of the flags is set. 13698522511bSHartmut Brandt */ 13708522511bSHartmut Brandt cpu = flags & (M_WAITOK | M_NOWAIT | M_DONTWAIT | M_TRYWAIT); 13718522511bSHartmut Brandt if (cpu != M_NOWAIT && cpu != M_WAITOK) { 13728522511bSHartmut Brandt static struct timeval lasterr; 13738522511bSHartmut Brandt static int curerr, once; 13748522511bSHartmut Brandt if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 13758522511bSHartmut Brandt printf("Bad uma_zalloc flags: %x\n", cpu); 13768522511bSHartmut Brandt backtrace(); 13778522511bSHartmut Brandt once++; 13788522511bSHartmut Brandt } 13798522511bSHartmut Brandt } 13808522511bSHartmut Brandt #endif 13814c1cc01cSJohn Baldwin if (!(flags & M_NOWAIT)) { 13824c1cc01cSJohn Baldwin KASSERT(curthread->td_intr_nesting_level == 0, 1383a163d034SWarner Losh ("malloc(M_WAITOK) in interrupt context")); 138426306795SJohn Baldwin WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 138526306795SJohn Baldwin "malloc() of \"%s\"", zone->uz_name); 13864c1cc01cSJohn Baldwin } 13874c1cc01cSJohn Baldwin 1388a553d4b8SJeff Roberson zalloc_restart: 13898355f576SJeff Roberson cpu = PCPU_GET(cpuid); 1390d88797c2SBosko Milekic CPU_LOCK(cpu); 13918355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 13928355f576SJeff Roberson 13938355f576SJeff Roberson zalloc_start: 13948355f576SJeff Roberson bucket = cache->uc_allocbucket; 13958355f576SJeff Roberson 13968355f576SJeff Roberson if (bucket) { 1397cae33c14SJeff Roberson if (bucket->ub_cnt > 0) { 1398cae33c14SJeff Roberson bucket->ub_cnt--; 1399cae33c14SJeff Roberson item = bucket->ub_bucket[bucket->ub_cnt]; 14008355f576SJeff Roberson #ifdef INVARIANTS 1401cae33c14SJeff Roberson bucket->ub_bucket[bucket->ub_cnt] = NULL; 14028355f576SJeff Roberson #endif 14038355f576SJeff Roberson KASSERT(item != NULL, 14048355f576SJeff Roberson ("uma_zalloc: Bucket pointer mangled.")); 14058355f576SJeff Roberson cache->uc_allocs++; 1406639c9550SJeff Roberson #ifdef INVARIANTS 140781f71edaSMatt Jacob ZONE_LOCK(zone); 1408639c9550SJeff Roberson uma_dbg_alloc(zone, NULL, item); 140981f71edaSMatt Jacob ZONE_UNLOCK(zone); 1410639c9550SJeff Roberson #endif 1411d88797c2SBosko Milekic CPU_UNLOCK(cpu); 14128355f576SJeff Roberson if (zone->uz_ctor) 14138355f576SJeff Roberson zone->uz_ctor(item, zone->uz_size, udata); 14142cc35ff9SJeff Roberson if (flags & M_ZERO) 14152cc35ff9SJeff Roberson bzero(item, zone->uz_size); 14168355f576SJeff Roberson return (item); 14178355f576SJeff Roberson } else if (cache->uc_freebucket) { 14188355f576SJeff Roberson /* 14198355f576SJeff Roberson * We have run out of items in our allocbucket. 14208355f576SJeff Roberson * See if we can switch with our free bucket. 14218355f576SJeff Roberson */ 1422cae33c14SJeff Roberson if (cache->uc_freebucket->ub_cnt > 0) { 14238355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 142464f051e9SJeff Roberson printf("uma_zalloc: Swapping empty with" 142564f051e9SJeff Roberson " alloc.\n"); 14268355f576SJeff Roberson #endif 1427b983089aSJeff Roberson bucket = cache->uc_freebucket; 14288355f576SJeff Roberson cache->uc_freebucket = cache->uc_allocbucket; 1429b983089aSJeff Roberson cache->uc_allocbucket = bucket; 14308355f576SJeff Roberson 14318355f576SJeff Roberson goto zalloc_start; 14328355f576SJeff Roberson } 14338355f576SJeff Roberson } 14348355f576SJeff Roberson } 1435a553d4b8SJeff Roberson ZONE_LOCK(zone); 1436a553d4b8SJeff Roberson /* Since we have locked the zone we may as well send back our stats */ 1437a553d4b8SJeff Roberson zone->uz_allocs += cache->uc_allocs; 1438a553d4b8SJeff Roberson cache->uc_allocs = 0; 14398355f576SJeff Roberson 1440a553d4b8SJeff Roberson /* Our old one is now a free bucket */ 1441a553d4b8SJeff Roberson if (cache->uc_allocbucket) { 1442cae33c14SJeff Roberson KASSERT(cache->uc_allocbucket->ub_cnt == 0, 1443a553d4b8SJeff Roberson ("uma_zalloc_arg: Freeing a non free bucket.")); 1444a553d4b8SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_bucket, 1445a553d4b8SJeff Roberson cache->uc_allocbucket, ub_link); 1446a553d4b8SJeff Roberson cache->uc_allocbucket = NULL; 1447a553d4b8SJeff Roberson } 14488355f576SJeff Roberson 1449a553d4b8SJeff Roberson /* Check the free list for a new alloc bucket */ 1450a553d4b8SJeff Roberson if ((bucket = LIST_FIRST(&zone->uz_full_bucket)) != NULL) { 1451cae33c14SJeff Roberson KASSERT(bucket->ub_cnt != 0, 1452a553d4b8SJeff Roberson ("uma_zalloc_arg: Returning an empty bucket.")); 14538355f576SJeff Roberson 1454a553d4b8SJeff Roberson LIST_REMOVE(bucket, ub_link); 1455a553d4b8SJeff Roberson cache->uc_allocbucket = bucket; 1456a553d4b8SJeff Roberson ZONE_UNLOCK(zone); 14578355f576SJeff Roberson goto zalloc_start; 1458a553d4b8SJeff Roberson } 1459bbee39c6SJeff Roberson /* We are no longer associated with this cpu!!! */ 1460d88797c2SBosko Milekic CPU_UNLOCK(cpu); 1461bbee39c6SJeff Roberson 1462a553d4b8SJeff Roberson /* Bump up our uz_count so we get here less */ 1463cae33c14SJeff Roberson if (zone->uz_count < BUCKET_MAX) 1464a553d4b8SJeff Roberson zone->uz_count++; 14658355f576SJeff Roberson /* 1466a553d4b8SJeff Roberson * Now lets just fill a bucket and put it on the free list. If that 1467a553d4b8SJeff Roberson * works we'll restart the allocation from the begining. 1468bbee39c6SJeff Roberson */ 1469bbee39c6SJeff Roberson if (uma_zalloc_bucket(zone, flags)) { 1470bbee39c6SJeff Roberson ZONE_UNLOCK(zone); 1471bbee39c6SJeff Roberson goto zalloc_restart; 1472bbee39c6SJeff Roberson } 1473bbee39c6SJeff Roberson ZONE_UNLOCK(zone); 1474bbee39c6SJeff Roberson /* 1475bbee39c6SJeff Roberson * We may not be able to get a bucket so return an actual item. 1476bbee39c6SJeff Roberson */ 1477bbee39c6SJeff Roberson #ifdef UMA_DEBUG 1478bbee39c6SJeff Roberson printf("uma_zalloc_arg: Bucketzone returned NULL\n"); 1479bbee39c6SJeff Roberson #endif 1480bbee39c6SJeff Roberson 1481bbee39c6SJeff Roberson return (uma_zalloc_internal(zone, udata, flags)); 1482bbee39c6SJeff Roberson } 1483bbee39c6SJeff Roberson 1484bbee39c6SJeff Roberson static uma_slab_t 1485bbee39c6SJeff Roberson uma_zone_slab(uma_zone_t zone, int flags) 1486bbee39c6SJeff Roberson { 1487bbee39c6SJeff Roberson uma_slab_t slab; 1488bbee39c6SJeff Roberson 1489bbee39c6SJeff Roberson /* 1490bbee39c6SJeff Roberson * This is to prevent us from recursively trying to allocate 1491bbee39c6SJeff Roberson * buckets. The problem is that if an allocation forces us to 1492bbee39c6SJeff Roberson * grab a new bucket we will call page_alloc, which will go off 1493bbee39c6SJeff Roberson * and cause the vm to allocate vm_map_entries. If we need new 1494bbee39c6SJeff Roberson * buckets there too we will recurse in kmem_alloc and bad 1495bbee39c6SJeff Roberson * things happen. So instead we return a NULL bucket, and make 1496bbee39c6SJeff Roberson * the code that allocates buckets smart enough to deal with it 1497bbee39c6SJeff Roberson */ 1498cae33c14SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_INTERNAL && zone->uz_recurse != 0) 1499bbee39c6SJeff Roberson return (NULL); 1500bbee39c6SJeff Roberson 1501bbee39c6SJeff Roberson slab = NULL; 1502bbee39c6SJeff Roberson 1503bbee39c6SJeff Roberson for (;;) { 1504bbee39c6SJeff Roberson /* 1505bbee39c6SJeff Roberson * Find a slab with some space. Prefer slabs that are partially 1506bbee39c6SJeff Roberson * used over those that are totally full. This helps to reduce 1507bbee39c6SJeff Roberson * fragmentation. 1508bbee39c6SJeff Roberson */ 1509bbee39c6SJeff Roberson if (zone->uz_free != 0) { 1510bbee39c6SJeff Roberson if (!LIST_EMPTY(&zone->uz_part_slab)) { 1511bbee39c6SJeff Roberson slab = LIST_FIRST(&zone->uz_part_slab); 1512bbee39c6SJeff Roberson } else { 1513bbee39c6SJeff Roberson slab = LIST_FIRST(&zone->uz_free_slab); 1514bbee39c6SJeff Roberson LIST_REMOVE(slab, us_link); 1515bbee39c6SJeff Roberson LIST_INSERT_HEAD(&zone->uz_part_slab, slab, 1516bbee39c6SJeff Roberson us_link); 1517bbee39c6SJeff Roberson } 1518bbee39c6SJeff Roberson return (slab); 1519bbee39c6SJeff Roberson } 1520bbee39c6SJeff Roberson 1521bbee39c6SJeff Roberson /* 1522bbee39c6SJeff Roberson * M_NOVM means don't ask at all! 1523bbee39c6SJeff Roberson */ 1524bbee39c6SJeff Roberson if (flags & M_NOVM) 1525bbee39c6SJeff Roberson break; 1526bbee39c6SJeff Roberson 1527bbee39c6SJeff Roberson if (zone->uz_maxpages && 1528bbee39c6SJeff Roberson zone->uz_pages >= zone->uz_maxpages) { 1529bbee39c6SJeff Roberson zone->uz_flags |= UMA_ZFLAG_FULL; 1530bbee39c6SJeff Roberson 1531ebc85edfSJeff Roberson if (flags & M_NOWAIT) 1532bbee39c6SJeff Roberson break; 1533ebc85edfSJeff Roberson else 153464f051e9SJeff Roberson msleep(zone, &zone->uz_lock, PVM, 153564f051e9SJeff Roberson "zonelimit", 0); 1536bbee39c6SJeff Roberson continue; 1537bbee39c6SJeff Roberson } 1538bbee39c6SJeff Roberson zone->uz_recurse++; 1539bbee39c6SJeff Roberson slab = slab_zalloc(zone, flags); 1540bbee39c6SJeff Roberson zone->uz_recurse--; 1541bbee39c6SJeff Roberson /* 1542bbee39c6SJeff Roberson * If we got a slab here it's safe to mark it partially used 1543bbee39c6SJeff Roberson * and return. We assume that the caller is going to remove 1544bbee39c6SJeff Roberson * at least one item. 1545bbee39c6SJeff Roberson */ 1546bbee39c6SJeff Roberson if (slab) { 1547bbee39c6SJeff Roberson LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link); 1548bbee39c6SJeff Roberson return (slab); 1549bbee39c6SJeff Roberson } 1550bbee39c6SJeff Roberson /* 1551bbee39c6SJeff Roberson * We might not have been able to get a slab but another cpu 1552bbee39c6SJeff Roberson * could have while we were unlocked. Check again before we 1553bbee39c6SJeff Roberson * fail. 1554bbee39c6SJeff Roberson */ 1555ebc85edfSJeff Roberson if (flags & M_NOWAIT) 1556bbee39c6SJeff Roberson flags |= M_NOVM; 1557bbee39c6SJeff Roberson } 1558bbee39c6SJeff Roberson return (slab); 1559bbee39c6SJeff Roberson } 1560bbee39c6SJeff Roberson 1561d56368d7SBosko Milekic static void * 1562bbee39c6SJeff Roberson uma_slab_alloc(uma_zone_t zone, uma_slab_t slab) 1563bbee39c6SJeff Roberson { 1564bbee39c6SJeff Roberson void *item; 1565bbee39c6SJeff Roberson u_int8_t freei; 1566bbee39c6SJeff Roberson 1567bbee39c6SJeff Roberson freei = slab->us_firstfree; 1568bbee39c6SJeff Roberson slab->us_firstfree = slab->us_freelist[freei]; 1569bbee39c6SJeff Roberson item = slab->us_data + (zone->uz_rsize * freei); 1570bbee39c6SJeff Roberson 1571bbee39c6SJeff Roberson slab->us_freecount--; 1572bbee39c6SJeff Roberson zone->uz_free--; 1573bbee39c6SJeff Roberson #ifdef INVARIANTS 1574bbee39c6SJeff Roberson uma_dbg_alloc(zone, slab, item); 1575bbee39c6SJeff Roberson #endif 1576bbee39c6SJeff Roberson /* Move this slab to the full list */ 1577bbee39c6SJeff Roberson if (slab->us_freecount == 0) { 1578bbee39c6SJeff Roberson LIST_REMOVE(slab, us_link); 1579bbee39c6SJeff Roberson LIST_INSERT_HEAD(&zone->uz_full_slab, slab, us_link); 1580bbee39c6SJeff Roberson } 1581bbee39c6SJeff Roberson 1582bbee39c6SJeff Roberson return (item); 1583bbee39c6SJeff Roberson } 1584bbee39c6SJeff Roberson 1585bbee39c6SJeff Roberson static int 1586bbee39c6SJeff Roberson uma_zalloc_bucket(uma_zone_t zone, int flags) 1587bbee39c6SJeff Roberson { 1588bbee39c6SJeff Roberson uma_bucket_t bucket; 1589bbee39c6SJeff Roberson uma_slab_t slab; 159044eca34aSJeff Roberson int max; 1591bbee39c6SJeff Roberson 1592bbee39c6SJeff Roberson /* 1593a553d4b8SJeff Roberson * Try this zone's free list first so we don't allocate extra buckets. 15948355f576SJeff Roberson */ 1595bbee39c6SJeff Roberson if ((bucket = LIST_FIRST(&zone->uz_free_bucket)) != NULL) { 1596cae33c14SJeff Roberson KASSERT(bucket->ub_cnt == 0, 1597bbee39c6SJeff Roberson ("uma_zalloc_bucket: Bucket on free list is not empty.")); 1598a553d4b8SJeff Roberson LIST_REMOVE(bucket, ub_link); 1599bbee39c6SJeff Roberson } else { 160018aa2de5SJeff Roberson int bflags; 160118aa2de5SJeff Roberson 1602cae33c14SJeff Roberson bflags = (flags & ~M_ZERO); 160320e8e865SBosko Milekic if (zone->uz_flags & UMA_ZFLAG_CACHEONLY) 160418aa2de5SJeff Roberson bflags |= M_NOVM; 160518aa2de5SJeff Roberson 1606bbee39c6SJeff Roberson ZONE_UNLOCK(zone); 1607cae33c14SJeff Roberson bucket = bucket_alloc(zone->uz_count, bflags); 1608bbee39c6SJeff Roberson ZONE_LOCK(zone); 1609bbee39c6SJeff Roberson } 1610bbee39c6SJeff Roberson 1611bbee39c6SJeff Roberson if (bucket == NULL) 1612bbee39c6SJeff Roberson return (0); 1613bbee39c6SJeff Roberson 1614bbee39c6SJeff Roberson #ifdef SMP 1615a553d4b8SJeff Roberson /* 1616bbee39c6SJeff Roberson * This code is here to limit the number of simultaneous bucket fills 1617bbee39c6SJeff Roberson * for any given zone to the number of per cpu caches in this zone. This 1618bbee39c6SJeff Roberson * is done so that we don't allocate more memory than we really need. 1619a553d4b8SJeff Roberson */ 1620bbee39c6SJeff Roberson if (zone->uz_fills >= mp_ncpus) 1621bbee39c6SJeff Roberson goto done; 1622a553d4b8SJeff Roberson 1623bbee39c6SJeff Roberson #endif 1624bbee39c6SJeff Roberson zone->uz_fills++; 1625bbee39c6SJeff Roberson 162644eca34aSJeff Roberson max = MIN(bucket->ub_entries, zone->uz_count); 1627bbee39c6SJeff Roberson /* Try to keep the buckets totally full */ 162844eca34aSJeff Roberson while (bucket->ub_cnt < max && 1629d11e0ba5SJeff Roberson (slab = uma_zone_slab(zone, flags)) != NULL) { 163044eca34aSJeff Roberson while (slab->us_freecount && bucket->ub_cnt < max) { 1631cae33c14SJeff Roberson bucket->ub_bucket[bucket->ub_cnt++] = 1632bbee39c6SJeff Roberson uma_slab_alloc(zone, slab); 1633bbee39c6SJeff Roberson } 1634bbee39c6SJeff Roberson /* Don't block on the next fill */ 1635bbee39c6SJeff Roberson flags |= M_NOWAIT; 16368355f576SJeff Roberson } 16378355f576SJeff Roberson 1638bbee39c6SJeff Roberson zone->uz_fills--; 1639bbee39c6SJeff Roberson 1640cae33c14SJeff Roberson if (bucket->ub_cnt != 0) { 1641bbee39c6SJeff Roberson LIST_INSERT_HEAD(&zone->uz_full_bucket, 1642bbee39c6SJeff Roberson bucket, ub_link); 1643bbee39c6SJeff Roberson return (1); 1644bbee39c6SJeff Roberson } 1645bbee39c6SJeff Roberson #ifdef SMP 1646bbee39c6SJeff Roberson done: 1647bbee39c6SJeff Roberson #endif 1648cae33c14SJeff Roberson bucket_free(bucket); 1649bbee39c6SJeff Roberson 1650bbee39c6SJeff Roberson return (0); 1651bbee39c6SJeff Roberson } 16528355f576SJeff Roberson /* 1653bbee39c6SJeff Roberson * Allocates an item for an internal zone 16548355f576SJeff Roberson * 16558355f576SJeff Roberson * Arguments 16568355f576SJeff Roberson * zone The zone to alloc for. 16578355f576SJeff Roberson * udata The data to be passed to the constructor. 1658a163d034SWarner Losh * flags M_WAITOK, M_NOWAIT, M_ZERO. 16598355f576SJeff Roberson * 16608355f576SJeff Roberson * Returns 16618355f576SJeff Roberson * NULL if there is no memory and M_NOWAIT is set 1662bbee39c6SJeff Roberson * An item if successful 16638355f576SJeff Roberson */ 16648355f576SJeff Roberson 16658355f576SJeff Roberson static void * 1666bbee39c6SJeff Roberson uma_zalloc_internal(uma_zone_t zone, void *udata, int flags) 16678355f576SJeff Roberson { 16688355f576SJeff Roberson uma_slab_t slab; 16698355f576SJeff Roberson void *item; 16708355f576SJeff Roberson 16718355f576SJeff Roberson item = NULL; 16728355f576SJeff Roberson 16738355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 16748355f576SJeff Roberson printf("INTERNAL: Allocating one item from %s(%p)\n", zone->uz_name, zone); 16758355f576SJeff Roberson #endif 16768355f576SJeff Roberson ZONE_LOCK(zone); 16778355f576SJeff Roberson 1678bbee39c6SJeff Roberson slab = uma_zone_slab(zone, flags); 1679bbee39c6SJeff Roberson if (slab == NULL) { 1680bce97791SJeff Roberson ZONE_UNLOCK(zone); 1681a553d4b8SJeff Roberson return (NULL); 1682bce97791SJeff Roberson } 1683a553d4b8SJeff Roberson 1684bbee39c6SJeff Roberson item = uma_slab_alloc(zone, slab); 16858355f576SJeff Roberson 16868355f576SJeff Roberson ZONE_UNLOCK(zone); 16878355f576SJeff Roberson 16883370c5bfSJeff Roberson if (zone->uz_ctor != NULL) 16898355f576SJeff Roberson zone->uz_ctor(item, zone->uz_size, udata); 16902cc35ff9SJeff Roberson if (flags & M_ZERO) 16912cc35ff9SJeff Roberson bzero(item, zone->uz_size); 16928355f576SJeff Roberson 16938355f576SJeff Roberson return (item); 16948355f576SJeff Roberson } 16958355f576SJeff Roberson 16968355f576SJeff Roberson /* See uma.h */ 16978355f576SJeff Roberson void 16988355f576SJeff Roberson uma_zfree_arg(uma_zone_t zone, void *item, void *udata) 16998355f576SJeff Roberson { 17008355f576SJeff Roberson uma_cache_t cache; 17018355f576SJeff Roberson uma_bucket_t bucket; 17024741dcbfSJeff Roberson int bflags; 17038355f576SJeff Roberson int cpu; 17045c133dfaSBosko Milekic int skip; 17058355f576SJeff Roberson 17068355f576SJeff Roberson /* This is the fast path free */ 17075c133dfaSBosko Milekic skip = 0; 17088355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC_1 17098355f576SJeff Roberson printf("Freeing item %p to %s(%p)\n", item, zone->uz_name, zone); 17108355f576SJeff Roberson #endif 1711af7f9b97SJeff Roberson /* 1712af7f9b97SJeff Roberson * The race here is acceptable. If we miss it we'll just have to wait 1713af7f9b97SJeff Roberson * a little longer for the limits to be reset. 1714af7f9b97SJeff Roberson */ 1715af7f9b97SJeff Roberson 1716af7f9b97SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_FULL) 1717af7f9b97SJeff Roberson goto zfree_internal; 1718af7f9b97SJeff Roberson 17195c133dfaSBosko Milekic if (zone->uz_dtor) { 1720bba739abSJeff Roberson zone->uz_dtor(item, zone->uz_size, udata); 17215c133dfaSBosko Milekic skip = 1; 17225c133dfaSBosko Milekic } 1723bba739abSJeff Roberson 1724a553d4b8SJeff Roberson zfree_restart: 17258355f576SJeff Roberson cpu = PCPU_GET(cpuid); 1726d88797c2SBosko Milekic CPU_LOCK(cpu); 17278355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 17288355f576SJeff Roberson 17298355f576SJeff Roberson zfree_start: 17308355f576SJeff Roberson bucket = cache->uc_freebucket; 17318355f576SJeff Roberson 17328355f576SJeff Roberson if (bucket) { 1733a553d4b8SJeff Roberson /* 1734a553d4b8SJeff Roberson * Do we have room in our bucket? It is OK for this uz count 1735a553d4b8SJeff Roberson * check to be slightly out of sync. 1736a553d4b8SJeff Roberson */ 1737a553d4b8SJeff Roberson 1738cae33c14SJeff Roberson if (bucket->ub_cnt < bucket->ub_entries) { 1739cae33c14SJeff Roberson KASSERT(bucket->ub_bucket[bucket->ub_cnt] == NULL, 17408355f576SJeff Roberson ("uma_zfree: Freeing to non free bucket index.")); 1741cae33c14SJeff Roberson bucket->ub_bucket[bucket->ub_cnt] = item; 1742cae33c14SJeff Roberson bucket->ub_cnt++; 1743b9ba8931SJeff Roberson #ifdef INVARIANTS 174481f71edaSMatt Jacob ZONE_LOCK(zone); 1745b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_MALLOC) 1746b9ba8931SJeff Roberson uma_dbg_free(zone, udata, item); 1747b9ba8931SJeff Roberson else 1748b9ba8931SJeff Roberson uma_dbg_free(zone, NULL, item); 174981f71edaSMatt Jacob ZONE_UNLOCK(zone); 1750b9ba8931SJeff Roberson #endif 1751d88797c2SBosko Milekic CPU_UNLOCK(cpu); 17528355f576SJeff Roberson return; 17538355f576SJeff Roberson } else if (cache->uc_allocbucket) { 17548355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 17558355f576SJeff Roberson printf("uma_zfree: Swapping buckets.\n"); 17568355f576SJeff Roberson #endif 17578355f576SJeff Roberson /* 17588355f576SJeff Roberson * We have run out of space in our freebucket. 17598355f576SJeff Roberson * See if we can switch with our alloc bucket. 17608355f576SJeff Roberson */ 1761cae33c14SJeff Roberson if (cache->uc_allocbucket->ub_cnt < 1762cae33c14SJeff Roberson cache->uc_freebucket->ub_cnt) { 1763b983089aSJeff Roberson bucket = cache->uc_freebucket; 17648355f576SJeff Roberson cache->uc_freebucket = cache->uc_allocbucket; 1765b983089aSJeff Roberson cache->uc_allocbucket = bucket; 17668355f576SJeff Roberson goto zfree_start; 17678355f576SJeff Roberson } 17688355f576SJeff Roberson } 17698355f576SJeff Roberson } 17708355f576SJeff Roberson /* 1771a553d4b8SJeff Roberson * We can get here for two reasons: 17728355f576SJeff Roberson * 17738355f576SJeff Roberson * 1) The buckets are NULL 1774a553d4b8SJeff Roberson * 2) The alloc and free buckets are both somewhat full. 17758355f576SJeff Roberson */ 17768355f576SJeff Roberson 17778355f576SJeff Roberson ZONE_LOCK(zone); 17788355f576SJeff Roberson 17798355f576SJeff Roberson bucket = cache->uc_freebucket; 17808355f576SJeff Roberson cache->uc_freebucket = NULL; 17818355f576SJeff Roberson 17828355f576SJeff Roberson /* Can we throw this on the zone full list? */ 17838355f576SJeff Roberson if (bucket != NULL) { 17848355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 17858355f576SJeff Roberson printf("uma_zfree: Putting old bucket on the free list.\n"); 17868355f576SJeff Roberson #endif 1787cae33c14SJeff Roberson /* ub_cnt is pointing to the last free item */ 1788cae33c14SJeff Roberson KASSERT(bucket->ub_cnt != 0, 17898355f576SJeff Roberson ("uma_zfree: Attempting to insert an empty bucket onto the full list.\n")); 17908355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_full_bucket, 17918355f576SJeff Roberson bucket, ub_link); 17928355f576SJeff Roberson } 1793a553d4b8SJeff Roberson if ((bucket = LIST_FIRST(&zone->uz_free_bucket)) != NULL) { 1794a553d4b8SJeff Roberson LIST_REMOVE(bucket, ub_link); 1795a553d4b8SJeff Roberson ZONE_UNLOCK(zone); 1796a553d4b8SJeff Roberson cache->uc_freebucket = bucket; 1797a553d4b8SJeff Roberson goto zfree_start; 1798a553d4b8SJeff Roberson } 1799a553d4b8SJeff Roberson /* We're done with this CPU now */ 1800d88797c2SBosko Milekic CPU_UNLOCK(cpu); 1801a553d4b8SJeff Roberson 1802a553d4b8SJeff Roberson /* And the zone.. */ 1803a553d4b8SJeff Roberson ZONE_UNLOCK(zone); 1804a553d4b8SJeff Roberson 18058355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 18068355f576SJeff Roberson printf("uma_zfree: Allocating new free bucket.\n"); 18078355f576SJeff Roberson #endif 18084741dcbfSJeff Roberson bflags = M_NOWAIT; 18094741dcbfSJeff Roberson 181020e8e865SBosko Milekic if (zone->uz_flags & UMA_ZFLAG_CACHEONLY) 18114741dcbfSJeff Roberson bflags |= M_NOVM; 1812cae33c14SJeff Roberson bucket = bucket_alloc(zone->uz_count, bflags); 18134741dcbfSJeff Roberson if (bucket) { 1814a553d4b8SJeff Roberson ZONE_LOCK(zone); 1815a553d4b8SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_bucket, 1816a553d4b8SJeff Roberson bucket, ub_link); 18178355f576SJeff Roberson ZONE_UNLOCK(zone); 1818a553d4b8SJeff Roberson goto zfree_restart; 18198355f576SJeff Roberson } 18208355f576SJeff Roberson 1821a553d4b8SJeff Roberson /* 1822a553d4b8SJeff Roberson * If nothing else caught this, we'll just do an internal free. 1823a553d4b8SJeff Roberson */ 18248355f576SJeff Roberson 1825af7f9b97SJeff Roberson zfree_internal: 1826af7f9b97SJeff Roberson 182748bf8725SBosko Milekic #ifdef INVARIANTS 182848bf8725SBosko Milekic /* 182964f051e9SJeff Roberson * If we need to skip the dtor and the uma_dbg_free in 183064f051e9SJeff Roberson * uma_zfree_internal because we've already called the dtor 183164f051e9SJeff Roberson * above, but we ended up here, then we need to make sure 183264f051e9SJeff Roberson * that we take care of the uma_dbg_free immediately. 183348bf8725SBosko Milekic */ 183448bf8725SBosko Milekic if (skip) { 183548bf8725SBosko Milekic ZONE_LOCK(zone); 1836b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_MALLOC) 183748bf8725SBosko Milekic uma_dbg_free(zone, udata, item); 183848bf8725SBosko Milekic else 183948bf8725SBosko Milekic uma_dbg_free(zone, NULL, item); 184048bf8725SBosko Milekic ZONE_UNLOCK(zone); 184148bf8725SBosko Milekic } 184248bf8725SBosko Milekic #endif 18435c133dfaSBosko Milekic uma_zfree_internal(zone, item, udata, skip); 18448355f576SJeff Roberson 18458355f576SJeff Roberson return; 18468355f576SJeff Roberson 18478355f576SJeff Roberson } 18488355f576SJeff Roberson 18498355f576SJeff Roberson /* 18508355f576SJeff Roberson * Frees an item to an INTERNAL zone or allocates a free bucket 18518355f576SJeff Roberson * 18528355f576SJeff Roberson * Arguments: 18538355f576SJeff Roberson * zone The zone to free to 18548355f576SJeff Roberson * item The item we're freeing 18558355f576SJeff Roberson * udata User supplied data for the dtor 18568355f576SJeff Roberson * skip Skip the dtor, it was done in uma_zfree_arg 18578355f576SJeff Roberson */ 18588355f576SJeff Roberson static void 18598355f576SJeff Roberson uma_zfree_internal(uma_zone_t zone, void *item, void *udata, int skip) 18608355f576SJeff Roberson { 18618355f576SJeff Roberson uma_slab_t slab; 18628355f576SJeff Roberson u_int8_t *mem; 18638355f576SJeff Roberson u_int8_t freei; 18648355f576SJeff Roberson 1865bba739abSJeff Roberson if (!skip && zone->uz_dtor) 1866bba739abSJeff Roberson zone->uz_dtor(item, zone->uz_size, udata); 1867bba739abSJeff Roberson 18688355f576SJeff Roberson ZONE_LOCK(zone); 18698355f576SJeff Roberson 1870b60f5b79SJeff Roberson if (!(zone->uz_flags & UMA_ZONE_MALLOC)) { 18718355f576SJeff Roberson mem = (u_int8_t *)((unsigned long)item & (~UMA_SLAB_MASK)); 1872b60f5b79SJeff Roberson if (zone->uz_flags & UMA_ZONE_HASH) 18738355f576SJeff Roberson slab = hash_sfind(&zone->uz_hash, mem); 18748355f576SJeff Roberson else { 18758355f576SJeff Roberson mem += zone->uz_pgoff; 18768355f576SJeff Roberson slab = (uma_slab_t)mem; 18778355f576SJeff Roberson } 18788355f576SJeff Roberson } else { 18798355f576SJeff Roberson slab = (uma_slab_t)udata; 18808355f576SJeff Roberson } 18818355f576SJeff Roberson 18828355f576SJeff Roberson /* Do we need to remove from any lists? */ 18838355f576SJeff Roberson if (slab->us_freecount+1 == zone->uz_ipers) { 18848355f576SJeff Roberson LIST_REMOVE(slab, us_link); 18858355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_slab, slab, us_link); 18868355f576SJeff Roberson } else if (slab->us_freecount == 0) { 18878355f576SJeff Roberson LIST_REMOVE(slab, us_link); 18888355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link); 18898355f576SJeff Roberson } 18908355f576SJeff Roberson 18918355f576SJeff Roberson /* Slab management stuff */ 18928355f576SJeff Roberson freei = ((unsigned long)item - (unsigned long)slab->us_data) 18938355f576SJeff Roberson / zone->uz_rsize; 18948355f576SJeff Roberson 1895639c9550SJeff Roberson #ifdef INVARIANTS 1896639c9550SJeff Roberson if (!skip) 1897639c9550SJeff Roberson uma_dbg_free(zone, slab, item); 18988355f576SJeff Roberson #endif 1899639c9550SJeff Roberson 19008355f576SJeff Roberson slab->us_freelist[freei] = slab->us_firstfree; 19018355f576SJeff Roberson slab->us_firstfree = freei; 19028355f576SJeff Roberson slab->us_freecount++; 19038355f576SJeff Roberson 19048355f576SJeff Roberson /* Zone statistics */ 19058355f576SJeff Roberson zone->uz_free++; 19068355f576SJeff Roberson 1907af7f9b97SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_FULL) { 1908af7f9b97SJeff Roberson if (zone->uz_pages < zone->uz_maxpages) 1909af7f9b97SJeff Roberson zone->uz_flags &= ~UMA_ZFLAG_FULL; 1910af7f9b97SJeff Roberson 1911af7f9b97SJeff Roberson /* We can handle one more allocation */ 191274c924b5SJeff Roberson wakeup_one(zone); 1913af7f9b97SJeff Roberson } 1914af7f9b97SJeff Roberson 1915605cbd6aSJeff Roberson ZONE_UNLOCK(zone); 19168355f576SJeff Roberson } 19178355f576SJeff Roberson 19188355f576SJeff Roberson /* See uma.h */ 19198355f576SJeff Roberson void 1920736ee590SJeff Roberson uma_zone_set_max(uma_zone_t zone, int nitems) 1921736ee590SJeff Roberson { 1922736ee590SJeff Roberson ZONE_LOCK(zone); 1923736ee590SJeff Roberson if (zone->uz_ppera > 1) 1924af7f9b97SJeff Roberson zone->uz_maxpages = nitems * zone->uz_ppera; 1925736ee590SJeff Roberson else 1926736ee590SJeff Roberson zone->uz_maxpages = nitems / zone->uz_ipers; 192728bc4419SJeff Roberson 1928d4d6aee5SAndrew R. Reiter if (zone->uz_maxpages * zone->uz_ipers < nitems) 1929d4d6aee5SAndrew R. Reiter zone->uz_maxpages++; 193028bc4419SJeff Roberson 1931736ee590SJeff Roberson ZONE_UNLOCK(zone); 1932736ee590SJeff Roberson } 1933736ee590SJeff Roberson 1934736ee590SJeff Roberson /* See uma.h */ 1935736ee590SJeff Roberson void 19368355f576SJeff Roberson uma_zone_set_freef(uma_zone_t zone, uma_free freef) 19378355f576SJeff Roberson { 19388355f576SJeff Roberson ZONE_LOCK(zone); 19398355f576SJeff Roberson zone->uz_freef = freef; 19408355f576SJeff Roberson ZONE_UNLOCK(zone); 19418355f576SJeff Roberson } 19428355f576SJeff Roberson 19438355f576SJeff Roberson /* See uma.h */ 19448355f576SJeff Roberson void 19458355f576SJeff Roberson uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf) 19468355f576SJeff Roberson { 19478355f576SJeff Roberson ZONE_LOCK(zone); 19488355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_PRIVALLOC; 19498355f576SJeff Roberson zone->uz_allocf = allocf; 19508355f576SJeff Roberson ZONE_UNLOCK(zone); 19518355f576SJeff Roberson } 19528355f576SJeff Roberson 19538355f576SJeff Roberson /* See uma.h */ 19548355f576SJeff Roberson int 19558355f576SJeff Roberson uma_zone_set_obj(uma_zone_t zone, struct vm_object *obj, int count) 19568355f576SJeff Roberson { 19578355f576SJeff Roberson int pages; 19588355f576SJeff Roberson vm_offset_t kva; 19598355f576SJeff Roberson 19608355f576SJeff Roberson pages = count / zone->uz_ipers; 19618355f576SJeff Roberson 19628355f576SJeff Roberson if (pages * zone->uz_ipers < count) 19638355f576SJeff Roberson pages++; 1964a553d4b8SJeff Roberson 19658355f576SJeff Roberson kva = kmem_alloc_pageable(kernel_map, pages * UMA_SLAB_SIZE); 19668355f576SJeff Roberson 1967d1f42ac2SAlan Cox if (kva == 0) 19688355f576SJeff Roberson return (0); 196964f051e9SJeff Roberson if (obj == NULL) { 1970a553d4b8SJeff Roberson obj = vm_object_allocate(OBJT_DEFAULT, 1971c7173f58SJeff Roberson pages); 197264f051e9SJeff Roberson } else { 197382774d80SAlan Cox VM_OBJECT_LOCK_INIT(obj); 19748355f576SJeff Roberson _vm_object_allocate(OBJT_DEFAULT, 1975c7173f58SJeff Roberson pages, obj); 197682774d80SAlan Cox } 1977a553d4b8SJeff Roberson ZONE_LOCK(zone); 1978a553d4b8SJeff Roberson zone->uz_kva = kva; 1979a553d4b8SJeff Roberson zone->uz_obj = obj; 1980a553d4b8SJeff Roberson zone->uz_maxpages = pages; 19818355f576SJeff Roberson zone->uz_allocf = obj_alloc; 1982b60f5b79SJeff Roberson zone->uz_flags |= UMA_ZONE_NOFREE | UMA_ZFLAG_PRIVALLOC; 19838355f576SJeff Roberson ZONE_UNLOCK(zone); 19848355f576SJeff Roberson return (1); 19858355f576SJeff Roberson } 19868355f576SJeff Roberson 19878355f576SJeff Roberson /* See uma.h */ 19888355f576SJeff Roberson void 19898355f576SJeff Roberson uma_prealloc(uma_zone_t zone, int items) 19908355f576SJeff Roberson { 19918355f576SJeff Roberson int slabs; 19928355f576SJeff Roberson uma_slab_t slab; 19938355f576SJeff Roberson 19948355f576SJeff Roberson ZONE_LOCK(zone); 19958355f576SJeff Roberson slabs = items / zone->uz_ipers; 19968355f576SJeff Roberson if (slabs * zone->uz_ipers < items) 19978355f576SJeff Roberson slabs++; 19988355f576SJeff Roberson while (slabs > 0) { 1999a163d034SWarner Losh slab = slab_zalloc(zone, M_WAITOK); 20008355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_slab, slab, us_link); 20018355f576SJeff Roberson slabs--; 20028355f576SJeff Roberson } 20038355f576SJeff Roberson ZONE_UNLOCK(zone); 20048355f576SJeff Roberson } 20058355f576SJeff Roberson 20068355f576SJeff Roberson /* See uma.h */ 20078355f576SJeff Roberson void 20088355f576SJeff Roberson uma_reclaim(void) 20098355f576SJeff Roberson { 20108355f576SJeff Roberson #ifdef UMA_DEBUG 20118355f576SJeff Roberson printf("UMA: vm asked us to release pages!\n"); 20128355f576SJeff Roberson #endif 201386bbae32SJeff Roberson bucket_enable(); 20148355f576SJeff Roberson zone_foreach(zone_drain); 20158355f576SJeff Roberson /* 20168355f576SJeff Roberson * Some slabs may have been freed but this zone will be visited early 20178355f576SJeff Roberson * we visit again so that we can free pages that are empty once other 20188355f576SJeff Roberson * zones are drained. We have to do the same for buckets. 20198355f576SJeff Roberson */ 20209643769aSJeff Roberson zone_drain(slabzone); 2021cae33c14SJeff Roberson bucket_zone_drain(); 20228355f576SJeff Roberson } 20238355f576SJeff Roberson 20248355f576SJeff Roberson void * 20258355f576SJeff Roberson uma_large_malloc(int size, int wait) 20268355f576SJeff Roberson { 20278355f576SJeff Roberson void *mem; 20288355f576SJeff Roberson uma_slab_t slab; 20298355f576SJeff Roberson u_int8_t flags; 20308355f576SJeff Roberson 2031bbee39c6SJeff Roberson slab = uma_zalloc_internal(slabzone, NULL, wait); 20328355f576SJeff Roberson if (slab == NULL) 20338355f576SJeff Roberson return (NULL); 20348355f576SJeff Roberson mem = page_alloc(NULL, size, &flags, wait); 20358355f576SJeff Roberson if (mem) { 203699571dc3SJeff Roberson vsetslab((vm_offset_t)mem, slab); 20378355f576SJeff Roberson slab->us_data = mem; 20388355f576SJeff Roberson slab->us_flags = flags | UMA_SLAB_MALLOC; 20398355f576SJeff Roberson slab->us_size = size; 20408355f576SJeff Roberson } else { 20418355f576SJeff Roberson uma_zfree_internal(slabzone, slab, NULL, 0); 20428355f576SJeff Roberson } 20438355f576SJeff Roberson 20448355f576SJeff Roberson 20458355f576SJeff Roberson return (mem); 20468355f576SJeff Roberson } 20478355f576SJeff Roberson 20488355f576SJeff Roberson void 20498355f576SJeff Roberson uma_large_free(uma_slab_t slab) 20508355f576SJeff Roberson { 205199571dc3SJeff Roberson vsetobj((vm_offset_t)slab->us_data, kmem_object); 2052125ee0d1STor Egge /* 20535103186cSAlan Cox * XXX: We get a lock order reversal if we don't have Giant: 2054125ee0d1STor Egge * vm_map_remove (locks system map) -> vm_map_delete -> 2055125ee0d1STor Egge * vm_map_entry_unwire -> vm_fault_unwire -> mtx_lock(&Giant) 2056125ee0d1STor Egge */ 2057125ee0d1STor Egge if (!mtx_owned(&Giant)) { 2058125ee0d1STor Egge mtx_lock(&Giant); 2059125ee0d1STor Egge page_free(slab->us_data, slab->us_size, slab->us_flags); 2060125ee0d1STor Egge mtx_unlock(&Giant); 2061125ee0d1STor Egge } else 20628355f576SJeff Roberson page_free(slab->us_data, slab->us_size, slab->us_flags); 20638355f576SJeff Roberson uma_zfree_internal(slabzone, slab, NULL, 0); 20648355f576SJeff Roberson } 20658355f576SJeff Roberson 20668355f576SJeff Roberson void 20678355f576SJeff Roberson uma_print_stats(void) 20688355f576SJeff Roberson { 20698355f576SJeff Roberson zone_foreach(uma_print_zone); 20708355f576SJeff Roberson } 20718355f576SJeff Roberson 20728355f576SJeff Roberson void 20738355f576SJeff Roberson uma_print_zone(uma_zone_t zone) 20748355f576SJeff Roberson { 20758355f576SJeff Roberson printf("%s(%p) size %d(%d) flags %d ipers %d ppera %d out %d free %d\n", 20768355f576SJeff Roberson zone->uz_name, zone, zone->uz_size, zone->uz_rsize, zone->uz_flags, 20778355f576SJeff Roberson zone->uz_ipers, zone->uz_ppera, 20788355f576SJeff Roberson (zone->uz_ipers * zone->uz_pages) - zone->uz_free, zone->uz_free); 20798355f576SJeff Roberson } 20808355f576SJeff Roberson 20818355f576SJeff Roberson /* 20828355f576SJeff Roberson * Sysctl handler for vm.zone 20838355f576SJeff Roberson * 20848355f576SJeff Roberson * stolen from vm_zone.c 20858355f576SJeff Roberson */ 20868355f576SJeff Roberson static int 20878355f576SJeff Roberson sysctl_vm_zone(SYSCTL_HANDLER_ARGS) 20888355f576SJeff Roberson { 20898355f576SJeff Roberson int error, len, cnt; 20908355f576SJeff Roberson const int linesize = 128; /* conservative */ 20918355f576SJeff Roberson int totalfree; 20928355f576SJeff Roberson char *tmpbuf, *offset; 20938355f576SJeff Roberson uma_zone_t z; 20948355f576SJeff Roberson char *p; 2095f828e5beSJeff Roberson int cpu; 2096f828e5beSJeff Roberson int cachefree; 2097f828e5beSJeff Roberson uma_bucket_t bucket; 2098f828e5beSJeff Roberson uma_cache_t cache; 20998355f576SJeff Roberson 21008355f576SJeff Roberson cnt = 0; 21010da47b2fSJeff Roberson mtx_lock(&uma_mtx); 21028355f576SJeff Roberson LIST_FOREACH(z, &uma_zones, uz_link) 21038355f576SJeff Roberson cnt++; 21040da47b2fSJeff Roberson mtx_unlock(&uma_mtx); 21058355f576SJeff Roberson MALLOC(tmpbuf, char *, (cnt == 0 ? 1 : cnt) * linesize, 2106a163d034SWarner Losh M_TEMP, M_WAITOK); 21078355f576SJeff Roberson len = snprintf(tmpbuf, linesize, 21088355f576SJeff Roberson "\nITEM SIZE LIMIT USED FREE REQUESTS\n\n"); 21098355f576SJeff Roberson if (cnt == 0) 21108355f576SJeff Roberson tmpbuf[len - 1] = '\0'; 21118355f576SJeff Roberson error = SYSCTL_OUT(req, tmpbuf, cnt == 0 ? len-1 : len); 21128355f576SJeff Roberson if (error || cnt == 0) 21138355f576SJeff Roberson goto out; 21148355f576SJeff Roberson offset = tmpbuf; 2115f4af24d5SJeff Roberson mtx_lock(&uma_mtx); 21168355f576SJeff Roberson LIST_FOREACH(z, &uma_zones, uz_link) { 21178355f576SJeff Roberson if (cnt == 0) /* list may have changed size */ 21188355f576SJeff Roberson break; 21191c35e213SBosko Milekic if (!(z->uz_flags & UMA_ZFLAG_INTERNAL)) { 2120f828e5beSJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 2121f828e5beSJeff Roberson if (CPU_ABSENT(cpu)) 2122f828e5beSJeff Roberson continue; 2123f828e5beSJeff Roberson CPU_LOCK(cpu); 2124f828e5beSJeff Roberson } 21251c35e213SBosko Milekic } 21268355f576SJeff Roberson ZONE_LOCK(z); 2127f828e5beSJeff Roberson cachefree = 0; 21281c35e213SBosko Milekic if (!(z->uz_flags & UMA_ZFLAG_INTERNAL)) { 2129f828e5beSJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 2130f828e5beSJeff Roberson if (CPU_ABSENT(cpu)) 2131f828e5beSJeff Roberson continue; 2132f828e5beSJeff Roberson cache = &z->uz_cpu[cpu]; 2133f828e5beSJeff Roberson if (cache->uc_allocbucket != NULL) 2134cae33c14SJeff Roberson cachefree += cache->uc_allocbucket->ub_cnt; 2135f828e5beSJeff Roberson if (cache->uc_freebucket != NULL) 2136cae33c14SJeff Roberson cachefree += cache->uc_freebucket->ub_cnt; 2137f828e5beSJeff Roberson CPU_UNLOCK(cpu); 2138f828e5beSJeff Roberson } 21391c35e213SBosko Milekic } 2140f828e5beSJeff Roberson LIST_FOREACH(bucket, &z->uz_full_bucket, ub_link) { 2141cae33c14SJeff Roberson cachefree += bucket->ub_cnt; 2142f828e5beSJeff Roberson } 2143f828e5beSJeff Roberson totalfree = z->uz_free + cachefree; 21448355f576SJeff Roberson len = snprintf(offset, linesize, 21458355f576SJeff Roberson "%-12.12s %6.6u, %8.8u, %6.6u, %6.6u, %8.8llu\n", 21468355f576SJeff Roberson z->uz_name, z->uz_size, 21478355f576SJeff Roberson z->uz_maxpages * z->uz_ipers, 21488355f576SJeff Roberson (z->uz_ipers * (z->uz_pages / z->uz_ppera)) - totalfree, 21498355f576SJeff Roberson totalfree, 21508355f576SJeff Roberson (unsigned long long)z->uz_allocs); 21518355f576SJeff Roberson ZONE_UNLOCK(z); 21528355f576SJeff Roberson for (p = offset + 12; p > offset && *p == ' '; --p) 21538355f576SJeff Roberson /* nothing */ ; 21548355f576SJeff Roberson p[1] = ':'; 21558355f576SJeff Roberson cnt--; 21568355f576SJeff Roberson offset += len; 21578355f576SJeff Roberson } 2158f4af24d5SJeff Roberson mtx_unlock(&uma_mtx); 21598355f576SJeff Roberson *offset++ = '\0'; 21608355f576SJeff Roberson error = SYSCTL_OUT(req, tmpbuf, offset - tmpbuf); 21618355f576SJeff Roberson out: 21628355f576SJeff Roberson FREE(tmpbuf, M_TEMP); 21638355f576SJeff Roberson return (error); 21648355f576SJeff Roberson } 2165