18355f576SJeff Roberson /* 28355f576SJeff Roberson * Copyright (c) 2002, Jeffrey Roberson <jroberson@chesapeake.net> 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 * $FreeBSD$ 278355f576SJeff Roberson * 288355f576SJeff Roberson */ 298355f576SJeff Roberson 308355f576SJeff Roberson /* 318355f576SJeff Roberson * uma_core.c Implementation of the Universal Memory allocator 328355f576SJeff Roberson * 338355f576SJeff Roberson * This allocator is intended to replace the multitude of similar object caches 348355f576SJeff Roberson * in the standard FreeBSD kernel. The intent is to be flexible as well as 358355f576SJeff Roberson * effecient. A primary design goal is to return unused memory to the rest of 368355f576SJeff Roberson * the system. This will make the system as a whole more flexible due to the 378355f576SJeff Roberson * ability to move memory to subsystems which most need it instead of leaving 388355f576SJeff Roberson * pools of reserved memory unused. 398355f576SJeff Roberson * 408355f576SJeff Roberson * The basic ideas stem from similar slab/zone based allocators whose algorithms 418355f576SJeff Roberson * are well known. 428355f576SJeff Roberson * 438355f576SJeff Roberson */ 448355f576SJeff Roberson 458355f576SJeff Roberson /* 468355f576SJeff Roberson * TODO: 478355f576SJeff Roberson * - Improve memory usage for large allocations 488355f576SJeff Roberson * - Improve INVARIANTS (0xdeadc0de write out) 498355f576SJeff Roberson * - Investigate cache size adjustments 508355f576SJeff Roberson */ 518355f576SJeff Roberson 528355f576SJeff Roberson /* I should really use ktr.. */ 538355f576SJeff Roberson /* 548355f576SJeff Roberson #define UMA_DEBUG 1 558355f576SJeff Roberson #define UMA_DEBUG_ALLOC 1 568355f576SJeff Roberson #define UMA_DEBUG_ALLOC_1 1 578355f576SJeff Roberson */ 588355f576SJeff Roberson 598355f576SJeff Roberson 608355f576SJeff Roberson #include "opt_param.h" 618355f576SJeff Roberson #include <sys/param.h> 628355f576SJeff Roberson #include <sys/systm.h> 638355f576SJeff Roberson #include <sys/kernel.h> 648355f576SJeff Roberson #include <sys/types.h> 658355f576SJeff Roberson #include <sys/queue.h> 668355f576SJeff Roberson #include <sys/malloc.h> 678355f576SJeff Roberson #include <sys/lock.h> 688355f576SJeff Roberson #include <sys/sysctl.h> 698355f576SJeff Roberson #include <sys/mutex.h> 708355f576SJeff Roberson #include <sys/smp.h> 7186bbae32SJeff Roberson #include <sys/vmmeter.h> 7286bbae32SJeff Roberson 7386bbae32SJeff Roberson #include <machine/types.h> 748355f576SJeff Roberson 758355f576SJeff Roberson #include <vm/vm.h> 768355f576SJeff Roberson #include <vm/vm_object.h> 778355f576SJeff Roberson #include <vm/vm_page.h> 788355f576SJeff Roberson #include <vm/vm_param.h> 798355f576SJeff Roberson #include <vm/vm_map.h> 808355f576SJeff Roberson #include <vm/vm_kern.h> 818355f576SJeff Roberson #include <vm/vm_extern.h> 828355f576SJeff Roberson #include <vm/uma.h> 838355f576SJeff Roberson #include <vm/uma_int.h> 848355f576SJeff Roberson 858355f576SJeff Roberson /* 868355f576SJeff Roberson * This is the zone from which all zones are spawned. The idea is that even 878355f576SJeff Roberson * the zone heads are allocated from the allocator, so we use the bss section 888355f576SJeff Roberson * to bootstrap us. 898355f576SJeff Roberson */ 9086bbae32SJeff Roberson static struct uma_zone masterzone; 9186bbae32SJeff Roberson static uma_zone_t zones = &masterzone; 928355f576SJeff Roberson 938355f576SJeff Roberson /* This is the zone from which all of uma_slab_t's are allocated. */ 948355f576SJeff Roberson static uma_zone_t slabzone; 958355f576SJeff Roberson 968355f576SJeff Roberson /* 978355f576SJeff Roberson * The initial hash tables come out of this zone so they can be allocated 988355f576SJeff Roberson * prior to malloc coming up. 998355f576SJeff Roberson */ 1008355f576SJeff Roberson static uma_zone_t hashzone; 1018355f576SJeff Roberson 1028355f576SJeff Roberson /* 1038355f576SJeff Roberson * Zone that buckets come from. 1048355f576SJeff Roberson */ 1058355f576SJeff Roberson static uma_zone_t bucketzone; 1068355f576SJeff Roberson 10786bbae32SJeff Roberson /* 10886bbae32SJeff Roberson * Are we allowed to allocate buckets? 10986bbae32SJeff Roberson */ 11086bbae32SJeff Roberson static int bucketdisable = 1; 11186bbae32SJeff Roberson 1128355f576SJeff Roberson /* Linked list of all zones in the system */ 1138355f576SJeff Roberson static LIST_HEAD(,uma_zone) uma_zones = LIST_HEAD_INITIALIZER(&uma_zones); 1148355f576SJeff Roberson 1158355f576SJeff Roberson /* This mutex protects the zone list */ 1168355f576SJeff Roberson static struct mtx uma_mtx; 1178355f576SJeff Roberson 1188355f576SJeff Roberson /* Linked list of boot time pages */ 1198355f576SJeff Roberson static LIST_HEAD(,uma_slab) uma_boot_pages = 1208355f576SJeff Roberson LIST_HEAD_INITIALIZER(&uma_boot_pages); 1218355f576SJeff Roberson 1228355f576SJeff Roberson /* Count of free boottime pages */ 1238355f576SJeff Roberson static int uma_boot_free = 0; 1248355f576SJeff Roberson 1258355f576SJeff Roberson /* Is the VM done starting up? */ 1268355f576SJeff Roberson static int booted = 0; 1278355f576SJeff Roberson 1288355f576SJeff Roberson /* This is the handle used to schedule our working set calculator */ 1298355f576SJeff Roberson static struct callout uma_callout; 1308355f576SJeff Roberson 1318355f576SJeff Roberson /* This is mp_maxid + 1, for use while looping over each cpu */ 1328355f576SJeff Roberson static int maxcpu; 1338355f576SJeff Roberson 1348355f576SJeff Roberson /* 1358355f576SJeff Roberson * This structure is passed as the zone ctor arg so that I don't have to create 1368355f576SJeff Roberson * a special allocation function just for zones. 1378355f576SJeff Roberson */ 1388355f576SJeff Roberson struct uma_zctor_args { 1398355f576SJeff Roberson char *name; 1408355f576SJeff Roberson int size; 1418355f576SJeff Roberson uma_ctor ctor; 1428355f576SJeff Roberson uma_dtor dtor; 1438355f576SJeff Roberson uma_init uminit; 1448355f576SJeff Roberson uma_fini fini; 1458355f576SJeff Roberson int align; 1468355f576SJeff Roberson u_int16_t flags; 1478355f576SJeff Roberson }; 1488355f576SJeff Roberson 1498355f576SJeff Roberson /* 1508355f576SJeff Roberson * This is the malloc hash table which is used to find the zone that a 1518355f576SJeff Roberson * malloc allocation came from. It is not currently resizeable. The 1528355f576SJeff Roberson * memory for the actual hash bucket is allocated in kmeminit. 1538355f576SJeff Roberson */ 1548355f576SJeff Roberson struct uma_hash mhash; 1558355f576SJeff Roberson struct uma_hash *mallochash = &mhash; 1568355f576SJeff Roberson 1578355f576SJeff Roberson /* Prototypes.. */ 1588355f576SJeff Roberson 1598355f576SJeff Roberson static void *obj_alloc(uma_zone_t, int, u_int8_t *, int); 1608355f576SJeff Roberson static void *page_alloc(uma_zone_t, int, u_int8_t *, int); 1618355f576SJeff Roberson static void page_free(void *, int, u_int8_t); 1628355f576SJeff Roberson static uma_slab_t slab_zalloc(uma_zone_t, int); 1638355f576SJeff Roberson static void cache_drain(uma_zone_t); 1648355f576SJeff Roberson static void bucket_drain(uma_zone_t, uma_bucket_t); 1658355f576SJeff Roberson static void zone_drain(uma_zone_t); 1668355f576SJeff Roberson static void zone_ctor(void *, int, void *); 1679c2cd7e5SJeff Roberson static void zone_dtor(void *, int, void *); 1688355f576SJeff Roberson static void zero_init(void *, int); 1698355f576SJeff Roberson static void zone_small_init(uma_zone_t zone); 1708355f576SJeff Roberson static void zone_large_init(uma_zone_t zone); 1718355f576SJeff Roberson static void zone_foreach(void (*zfunc)(uma_zone_t)); 1728355f576SJeff Roberson static void zone_timeout(uma_zone_t zone); 1735300d9ddSJeff Roberson static struct slabhead *hash_alloc(int *); 1745300d9ddSJeff Roberson static void hash_expand(struct uma_hash *, struct slabhead *, int); 1755300d9ddSJeff Roberson static void hash_free(struct slabhead *hash, int hashsize); 1768355f576SJeff Roberson static void uma_timeout(void *); 1778355f576SJeff Roberson static void uma_startup3(void); 178a553d4b8SJeff Roberson static void *uma_zalloc_internal(uma_zone_t, void *, int, uma_bucket_t); 17986bbae32SJeff Roberson static void uma_zfree_internal(uma_zone_t, void *, void *, int); 18086bbae32SJeff Roberson static void bucket_enable(void); 1818355f576SJeff Roberson void uma_print_zone(uma_zone_t); 1828355f576SJeff Roberson void uma_print_stats(void); 1838355f576SJeff Roberson static int sysctl_vm_zone(SYSCTL_HANDLER_ARGS); 1848355f576SJeff Roberson 1858355f576SJeff Roberson SYSCTL_OID(_vm, OID_AUTO, zone, CTLTYPE_STRING|CTLFLAG_RD, 1868355f576SJeff Roberson NULL, 0, sysctl_vm_zone, "A", "Zone Info"); 1878355f576SJeff Roberson SYSINIT(uma_startup3, SI_SUB_VM_CONF, SI_ORDER_SECOND, uma_startup3, NULL); 1888355f576SJeff Roberson 18986bbae32SJeff Roberson /* 19086bbae32SJeff Roberson * This routine checks to see whether or not it's safe to enable buckets. 19186bbae32SJeff Roberson */ 19286bbae32SJeff Roberson 19386bbae32SJeff Roberson static void 19486bbae32SJeff Roberson bucket_enable(void) 19586bbae32SJeff Roberson { 19686bbae32SJeff Roberson if (cnt.v_free_count < cnt.v_free_min) 19786bbae32SJeff Roberson bucketdisable = 1; 19886bbae32SJeff Roberson else 19986bbae32SJeff Roberson bucketdisable = 0; 20086bbae32SJeff Roberson } 20186bbae32SJeff Roberson 2028355f576SJeff Roberson 2038355f576SJeff Roberson /* 2048355f576SJeff Roberson * Routine called by timeout which is used to fire off some time interval 2058355f576SJeff Roberson * based calculations. (working set, stats, etc.) 2068355f576SJeff Roberson * 2078355f576SJeff Roberson * Arguments: 2088355f576SJeff Roberson * arg Unused 2098355f576SJeff Roberson * 2108355f576SJeff Roberson * Returns: 2118355f576SJeff Roberson * Nothing 2128355f576SJeff Roberson */ 2138355f576SJeff Roberson static void 2148355f576SJeff Roberson uma_timeout(void *unused) 2158355f576SJeff Roberson { 21686bbae32SJeff Roberson bucket_enable(); 2178355f576SJeff Roberson zone_foreach(zone_timeout); 2188355f576SJeff Roberson 2198355f576SJeff Roberson /* Reschedule this event */ 2208355f576SJeff Roberson callout_reset(&uma_callout, UMA_WORKING_TIME * hz, uma_timeout, NULL); 2218355f576SJeff Roberson } 2228355f576SJeff Roberson 2238355f576SJeff Roberson /* 2248355f576SJeff Roberson * Routine to perform timeout driven calculations. This does the working set 2258355f576SJeff Roberson * as well as hash expanding, and per cpu statistics aggregation. 2268355f576SJeff Roberson * 2278355f576SJeff Roberson * Arguments: 2288355f576SJeff Roberson * zone The zone to operate on 2298355f576SJeff Roberson * 2308355f576SJeff Roberson * Returns: 2318355f576SJeff Roberson * Nothing 2328355f576SJeff Roberson */ 2338355f576SJeff Roberson static void 2348355f576SJeff Roberson zone_timeout(uma_zone_t zone) 2358355f576SJeff Roberson { 2368355f576SJeff Roberson uma_cache_t cache; 2378355f576SJeff Roberson u_int64_t alloc; 2388355f576SJeff Roberson int free; 2398355f576SJeff Roberson int cpu; 2408355f576SJeff Roberson 2418355f576SJeff Roberson alloc = 0; 2428355f576SJeff Roberson free = 0; 2438355f576SJeff Roberson 2448355f576SJeff Roberson /* 2458355f576SJeff Roberson * Aggregate per cpu cache statistics back to the zone. 2468355f576SJeff Roberson * 2478355f576SJeff Roberson * I may rewrite this to set a flag in the per cpu cache instead of 2488355f576SJeff Roberson * locking. If the flag is not cleared on the next round I will have 2498355f576SJeff Roberson * to lock and do it here instead so that the statistics don't get too 2508355f576SJeff Roberson * far out of sync. 2518355f576SJeff Roberson */ 2528355f576SJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) { 2538355f576SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 2548355f576SJeff Roberson if (CPU_ABSENT(cpu)) 2558355f576SJeff Roberson continue; 2568355f576SJeff Roberson CPU_LOCK(zone, cpu); 2578355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 2588355f576SJeff Roberson /* Add them up, and reset */ 2598355f576SJeff Roberson alloc += cache->uc_allocs; 2608355f576SJeff Roberson cache->uc_allocs = 0; 2618355f576SJeff Roberson if (cache->uc_allocbucket) 2628355f576SJeff Roberson free += cache->uc_allocbucket->ub_ptr + 1; 2638355f576SJeff Roberson if (cache->uc_freebucket) 2648355f576SJeff Roberson free += cache->uc_freebucket->ub_ptr + 1; 2658355f576SJeff Roberson CPU_UNLOCK(zone, cpu); 2668355f576SJeff Roberson } 2678355f576SJeff Roberson } 2688355f576SJeff Roberson 2698355f576SJeff Roberson /* Now push these stats back into the zone.. */ 2708355f576SJeff Roberson ZONE_LOCK(zone); 2718355f576SJeff Roberson zone->uz_allocs += alloc; 2728355f576SJeff Roberson 2738355f576SJeff Roberson /* 2748355f576SJeff Roberson * cachefree is an instantanious snapshot of what is in the per cpu 2758355f576SJeff Roberson * caches, not an accurate counter 2768355f576SJeff Roberson */ 2778355f576SJeff Roberson zone->uz_cachefree = free; 2788355f576SJeff Roberson 2798355f576SJeff Roberson /* 2808355f576SJeff Roberson * Expand the zone hash table. 2818355f576SJeff Roberson * 2828355f576SJeff Roberson * This is done if the number of slabs is larger than the hash size. 2838355f576SJeff Roberson * What I'm trying to do here is completely reduce collisions. This 2848355f576SJeff Roberson * may be a little aggressive. Should I allow for two collisions max? 2858355f576SJeff Roberson */ 2868355f576SJeff Roberson 2878355f576SJeff Roberson if ((zone->uz_flags & UMA_ZFLAG_OFFPAGE) && 2888355f576SJeff Roberson !(zone->uz_flags & UMA_ZFLAG_MALLOC)) { 2898355f576SJeff Roberson if (zone->uz_pages / zone->uz_ppera 2905300d9ddSJeff Roberson >= zone->uz_hash.uh_hashsize) { 2915300d9ddSJeff Roberson struct slabhead *newhash; 2925300d9ddSJeff Roberson int newsize; 2935300d9ddSJeff Roberson 2945300d9ddSJeff Roberson newsize = zone->uz_hash.uh_hashsize; 2955300d9ddSJeff Roberson ZONE_UNLOCK(zone); 2965300d9ddSJeff Roberson newhash = hash_alloc(&newsize); 2975300d9ddSJeff Roberson ZONE_LOCK(zone); 2985300d9ddSJeff Roberson hash_expand(&zone->uz_hash, newhash, newsize); 2995300d9ddSJeff Roberson } 3008355f576SJeff Roberson } 3018355f576SJeff Roberson 3028355f576SJeff Roberson /* 3038355f576SJeff Roberson * Here we compute the working set size as the total number of items 3048355f576SJeff Roberson * left outstanding since the last time interval. This is slightly 3058355f576SJeff Roberson * suboptimal. What we really want is the highest number of outstanding 3068355f576SJeff Roberson * items during the last time quantum. This should be close enough. 3078355f576SJeff Roberson * 3088355f576SJeff Roberson * The working set size is used to throttle the zone_drain function. 3098355f576SJeff Roberson * We don't want to return memory that we may need again immediately. 3108355f576SJeff Roberson */ 3118355f576SJeff Roberson alloc = zone->uz_allocs - zone->uz_oallocs; 3128355f576SJeff Roberson zone->uz_oallocs = zone->uz_allocs; 3138355f576SJeff Roberson zone->uz_wssize = alloc; 3148355f576SJeff Roberson 3158355f576SJeff Roberson ZONE_UNLOCK(zone); 3168355f576SJeff Roberson } 3178355f576SJeff Roberson 3188355f576SJeff Roberson /* 3195300d9ddSJeff Roberson * Allocate and zero fill the next sized hash table from the appropriate 3205300d9ddSJeff Roberson * backing store. 3215300d9ddSJeff Roberson * 3225300d9ddSJeff Roberson * Arguments: 3235300d9ddSJeff Roberson * oldsize On input it's the size we're currently at and on output 3245300d9ddSJeff Roberson * it is the expanded size. 3255300d9ddSJeff Roberson * 3265300d9ddSJeff Roberson * Returns: 3275300d9ddSJeff Roberson * slabhead The new hash bucket or NULL if the allocation failed. 3285300d9ddSJeff Roberson */ 3295300d9ddSJeff Roberson struct slabhead * 3305300d9ddSJeff Roberson hash_alloc(int *oldsize) 3315300d9ddSJeff Roberson { 3325300d9ddSJeff Roberson struct slabhead *newhash; 3335300d9ddSJeff Roberson int newsize; 3345300d9ddSJeff Roberson int alloc; 3355300d9ddSJeff Roberson 3365300d9ddSJeff Roberson /* We're just going to go to a power of two greater */ 3375300d9ddSJeff Roberson if (*oldsize) { 3385300d9ddSJeff Roberson newsize = (*oldsize) * 2; 3395300d9ddSJeff Roberson alloc = sizeof(newhash[0]) * newsize; 3405300d9ddSJeff Roberson /* XXX Shouldn't be abusing DEVBUF here */ 3415300d9ddSJeff Roberson newhash = (struct slabhead *)malloc(alloc, M_DEVBUF, M_NOWAIT); 3425300d9ddSJeff Roberson } else { 3435300d9ddSJeff Roberson alloc = sizeof(newhash[0]) * UMA_HASH_SIZE_INIT; 3445300d9ddSJeff Roberson newhash = uma_zalloc_internal(hashzone, NULL, M_WAITOK, NULL); 3455300d9ddSJeff Roberson newsize = UMA_HASH_SIZE_INIT; 3465300d9ddSJeff Roberson } 3475300d9ddSJeff Roberson if (newhash) 3485300d9ddSJeff Roberson bzero(newhash, alloc); 3495300d9ddSJeff Roberson 3505300d9ddSJeff Roberson *oldsize = newsize; 3515300d9ddSJeff Roberson 3525300d9ddSJeff Roberson return (newhash); 3535300d9ddSJeff Roberson } 3545300d9ddSJeff Roberson 3555300d9ddSJeff Roberson /* 3568355f576SJeff Roberson * Expands the hash table for OFFPAGE zones. This is done from zone_timeout 3578355f576SJeff Roberson * to reduce collisions. This must not be done in the regular allocation path, 3588355f576SJeff Roberson * otherwise, we can recurse on the vm while allocating pages. 3598355f576SJeff Roberson * 3608355f576SJeff Roberson * Arguments: 3618355f576SJeff Roberson * hash The hash you want to expand by a factor of two. 3628355f576SJeff Roberson * 3638355f576SJeff Roberson * Returns: 3648355f576SJeff Roberson * Nothing 3658355f576SJeff Roberson * 3668355f576SJeff Roberson * Discussion: 3678355f576SJeff Roberson */ 3688355f576SJeff Roberson static void 3695300d9ddSJeff Roberson hash_expand(struct uma_hash *hash, struct slabhead *newhash, int newsize) 3708355f576SJeff Roberson { 3718355f576SJeff Roberson struct slabhead *oldhash; 3728355f576SJeff Roberson uma_slab_t slab; 3739c2cd7e5SJeff Roberson int oldsize; 3748355f576SJeff Roberson int hval; 3758355f576SJeff Roberson int i; 3768355f576SJeff Roberson 3775300d9ddSJeff Roberson if (!newhash) 3785300d9ddSJeff Roberson return; 3798355f576SJeff Roberson 3809c2cd7e5SJeff Roberson oldsize = hash->uh_hashsize; 3818355f576SJeff Roberson oldhash = hash->uh_slab_hash; 3828355f576SJeff Roberson 3835300d9ddSJeff Roberson if (oldsize >= newsize) { 3845300d9ddSJeff Roberson hash_free(newhash, newsize); 3858355f576SJeff Roberson return; 3868355f576SJeff Roberson } 3878355f576SJeff Roberson 3889c2cd7e5SJeff Roberson hash->uh_hashmask = newsize - 1; 3898355f576SJeff Roberson 3908355f576SJeff Roberson /* 3918355f576SJeff Roberson * I need to investigate hash algorithms for resizing without a 3928355f576SJeff Roberson * full rehash. 3938355f576SJeff Roberson */ 3948355f576SJeff Roberson 3959c2cd7e5SJeff Roberson for (i = 0; i < oldsize; i++) 3968355f576SJeff Roberson while (!SLIST_EMPTY(&hash->uh_slab_hash[i])) { 3978355f576SJeff Roberson slab = SLIST_FIRST(&hash->uh_slab_hash[i]); 3988355f576SJeff Roberson SLIST_REMOVE_HEAD(&hash->uh_slab_hash[i], us_hlink); 3998355f576SJeff Roberson hval = UMA_HASH(hash, slab->us_data); 4008355f576SJeff Roberson SLIST_INSERT_HEAD(&newhash[hval], slab, us_hlink); 4018355f576SJeff Roberson } 4028355f576SJeff Roberson 4039c2cd7e5SJeff Roberson if (oldhash) 4045300d9ddSJeff Roberson hash_free(oldhash, oldsize); 4059c2cd7e5SJeff Roberson 4069c2cd7e5SJeff Roberson hash->uh_slab_hash = newhash; 4079c2cd7e5SJeff Roberson hash->uh_hashsize = newsize; 4089c2cd7e5SJeff Roberson 4099c2cd7e5SJeff Roberson return; 4109c2cd7e5SJeff Roberson } 4119c2cd7e5SJeff Roberson 4125300d9ddSJeff Roberson /* 4135300d9ddSJeff Roberson * Free the hash bucket to the appropriate backing store. 4145300d9ddSJeff Roberson * 4155300d9ddSJeff Roberson * Arguments: 4165300d9ddSJeff Roberson * slab_hash The hash bucket we're freeing 4175300d9ddSJeff Roberson * hashsize The number of entries in that hash bucket 4185300d9ddSJeff Roberson * 4195300d9ddSJeff Roberson * Returns: 4205300d9ddSJeff Roberson * Nothing 4215300d9ddSJeff Roberson */ 4229c2cd7e5SJeff Roberson static void 4235300d9ddSJeff Roberson hash_free(struct slabhead *slab_hash, int hashsize) 4249c2cd7e5SJeff Roberson { 4255300d9ddSJeff Roberson if (hashsize == UMA_HASH_SIZE_INIT) 4268355f576SJeff Roberson uma_zfree_internal(hashzone, 4275300d9ddSJeff Roberson slab_hash, NULL, 0); 4288355f576SJeff Roberson else 4295300d9ddSJeff Roberson free(slab_hash, M_DEVBUF); 4308355f576SJeff Roberson } 4318355f576SJeff Roberson 4328355f576SJeff Roberson /* 4338355f576SJeff Roberson * Frees all outstanding items in a bucket 4348355f576SJeff Roberson * 4358355f576SJeff Roberson * Arguments: 4368355f576SJeff Roberson * zone The zone to free to, must be unlocked. 4378355f576SJeff Roberson * bucket The free/alloc bucket with items, cpu queue must be locked. 4388355f576SJeff Roberson * 4398355f576SJeff Roberson * Returns: 4408355f576SJeff Roberson * Nothing 4418355f576SJeff Roberson */ 4428355f576SJeff Roberson 4438355f576SJeff Roberson static void 4448355f576SJeff Roberson bucket_drain(uma_zone_t zone, uma_bucket_t bucket) 4458355f576SJeff Roberson { 4468355f576SJeff Roberson uma_slab_t slab; 4478355f576SJeff Roberson int mzone; 4488355f576SJeff Roberson void *item; 4498355f576SJeff Roberson 4508355f576SJeff Roberson if (bucket == NULL) 4518355f576SJeff Roberson return; 4528355f576SJeff Roberson 4538355f576SJeff Roberson slab = NULL; 4548355f576SJeff Roberson mzone = 0; 4558355f576SJeff Roberson 4568355f576SJeff Roberson /* We have to lookup the slab again for malloc.. */ 4578355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_MALLOC) 4588355f576SJeff Roberson mzone = 1; 4598355f576SJeff Roberson 4608355f576SJeff Roberson while (bucket->ub_ptr > -1) { 4618355f576SJeff Roberson item = bucket->ub_bucket[bucket->ub_ptr]; 4628355f576SJeff Roberson #ifdef INVARIANTS 4638355f576SJeff Roberson bucket->ub_bucket[bucket->ub_ptr] = NULL; 4648355f576SJeff Roberson KASSERT(item != NULL, 4658355f576SJeff Roberson ("bucket_drain: botched ptr, item is NULL")); 4668355f576SJeff Roberson #endif 4678355f576SJeff Roberson bucket->ub_ptr--; 4688355f576SJeff Roberson /* 4698355f576SJeff Roberson * This is extremely inefficient. The slab pointer was passed 4708355f576SJeff Roberson * to uma_zfree_arg, but we lost it because the buckets don't 4718355f576SJeff Roberson * hold them. This will go away when free() gets a size passed 4728355f576SJeff Roberson * to it. 4738355f576SJeff Roberson */ 4748355f576SJeff Roberson if (mzone) 4758355f576SJeff Roberson slab = hash_sfind(mallochash, 4768355f576SJeff Roberson (u_int8_t *)((unsigned long)item & 4778355f576SJeff Roberson (~UMA_SLAB_MASK))); 4788355f576SJeff Roberson uma_zfree_internal(zone, item, slab, 1); 4798355f576SJeff Roberson } 4808355f576SJeff Roberson } 4818355f576SJeff Roberson 4828355f576SJeff Roberson /* 4838355f576SJeff Roberson * Drains the per cpu caches for a zone. 4848355f576SJeff Roberson * 4858355f576SJeff Roberson * Arguments: 4868355f576SJeff Roberson * zone The zone to drain, must be unlocked. 4878355f576SJeff Roberson * 4888355f576SJeff Roberson * Returns: 4898355f576SJeff Roberson * Nothing 4908355f576SJeff Roberson * 4918355f576SJeff Roberson * This function returns with the zone locked so that the per cpu queues can 4928355f576SJeff Roberson * not be filled until zone_drain is finished. 4938355f576SJeff Roberson * 4948355f576SJeff Roberson */ 4958355f576SJeff Roberson static void 4968355f576SJeff Roberson cache_drain(uma_zone_t zone) 4978355f576SJeff Roberson { 4988355f576SJeff Roberson uma_bucket_t bucket; 4998355f576SJeff Roberson uma_cache_t cache; 5008355f576SJeff Roberson int cpu; 5018355f576SJeff Roberson 5028355f576SJeff Roberson /* 5038355f576SJeff Roberson * Flush out the per cpu queues. 5048355f576SJeff Roberson * 505157d7b35SAlfred Perlstein * XXX This causes unnecessary thrashing due to immediately having 5068355f576SJeff Roberson * empty per cpu queues. I need to improve this. 5078355f576SJeff Roberson */ 5088355f576SJeff Roberson 5098355f576SJeff Roberson /* 5108355f576SJeff Roberson * We have to lock each cpu cache before locking the zone 5118355f576SJeff Roberson */ 5128355f576SJeff Roberson ZONE_UNLOCK(zone); 5138355f576SJeff Roberson 5148355f576SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 5158355f576SJeff Roberson if (CPU_ABSENT(cpu)) 5168355f576SJeff Roberson continue; 5178355f576SJeff Roberson CPU_LOCK(zone, cpu); 5188355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 5198355f576SJeff Roberson bucket_drain(zone, cache->uc_allocbucket); 5208355f576SJeff Roberson bucket_drain(zone, cache->uc_freebucket); 5218355f576SJeff Roberson } 5228355f576SJeff Roberson 5238355f576SJeff Roberson /* 5248355f576SJeff Roberson * Drain the bucket queues and free the buckets, we just keep two per 5258355f576SJeff Roberson * cpu (alloc/free). 5268355f576SJeff Roberson */ 5278355f576SJeff Roberson ZONE_LOCK(zone); 5288355f576SJeff Roberson while ((bucket = LIST_FIRST(&zone->uz_full_bucket)) != NULL) { 5298355f576SJeff Roberson LIST_REMOVE(bucket, ub_link); 5308355f576SJeff Roberson ZONE_UNLOCK(zone); 5318355f576SJeff Roberson bucket_drain(zone, bucket); 5328355f576SJeff Roberson uma_zfree_internal(bucketzone, bucket, NULL, 0); 5338355f576SJeff Roberson ZONE_LOCK(zone); 5348355f576SJeff Roberson } 5358355f576SJeff Roberson 5368355f576SJeff Roberson /* Now we do the free queue.. */ 5378355f576SJeff Roberson while ((bucket = LIST_FIRST(&zone->uz_free_bucket)) != NULL) { 5388355f576SJeff Roberson LIST_REMOVE(bucket, ub_link); 5398355f576SJeff Roberson uma_zfree_internal(bucketzone, bucket, NULL, 0); 5408355f576SJeff Roberson } 5418355f576SJeff Roberson 5428355f576SJeff Roberson /* We unlock here, but they will all block until the zone is unlocked */ 5438355f576SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) { 5448355f576SJeff Roberson if (CPU_ABSENT(cpu)) 5458355f576SJeff Roberson continue; 5468355f576SJeff Roberson CPU_UNLOCK(zone, cpu); 5478355f576SJeff Roberson } 548f4af24d5SJeff Roberson 549f4af24d5SJeff Roberson zone->uz_cachefree = 0; 5508355f576SJeff Roberson } 5518355f576SJeff Roberson 5528355f576SJeff Roberson /* 5538355f576SJeff Roberson * Frees pages from a zone back to the system. This is done on demand from 5548355f576SJeff Roberson * the pageout daemon. 5558355f576SJeff Roberson * 5568355f576SJeff Roberson * Arguments: 5578355f576SJeff Roberson * zone The zone to free pages from 5589c2cd7e5SJeff Roberson * all Should we drain all items? 5598355f576SJeff Roberson * 5608355f576SJeff Roberson * Returns: 5618355f576SJeff Roberson * Nothing. 5628355f576SJeff Roberson */ 5638355f576SJeff Roberson static void 5648355f576SJeff Roberson zone_drain(uma_zone_t zone) 5658355f576SJeff Roberson { 5668355f576SJeff Roberson uma_slab_t slab; 5678355f576SJeff Roberson uma_slab_t n; 5688355f576SJeff Roberson u_int64_t extra; 5698355f576SJeff Roberson u_int8_t flags; 5708355f576SJeff Roberson u_int8_t *mem; 5718355f576SJeff Roberson int i; 5728355f576SJeff Roberson 5738355f576SJeff Roberson /* 5748355f576SJeff Roberson * We don't want to take pages from staticly allocated zones at this 5758355f576SJeff Roberson * time 5768355f576SJeff Roberson */ 5778355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_NOFREE || zone->uz_freef == NULL) 5788355f576SJeff Roberson return; 5798355f576SJeff Roberson 5808355f576SJeff Roberson ZONE_LOCK(zone); 5818355f576SJeff Roberson 5828355f576SJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) 5838355f576SJeff Roberson cache_drain(zone); 5848355f576SJeff Roberson 5858355f576SJeff Roberson if (zone->uz_free < zone->uz_wssize) 5868355f576SJeff Roberson goto finished; 5878355f576SJeff Roberson #ifdef UMA_DEBUG 5888355f576SJeff Roberson printf("%s working set size: %llu free items: %u\n", 5898355f576SJeff Roberson zone->uz_name, (unsigned long long)zone->uz_wssize, zone->uz_free); 5908355f576SJeff Roberson #endif 5919c2cd7e5SJeff Roberson extra = zone->uz_free - zone->uz_wssize; 5928355f576SJeff Roberson extra /= zone->uz_ipers; 5938355f576SJeff Roberson 5948355f576SJeff Roberson /* extra is now the number of extra slabs that we can free */ 5958355f576SJeff Roberson 5968355f576SJeff Roberson if (extra == 0) 5978355f576SJeff Roberson goto finished; 5988355f576SJeff Roberson 5998355f576SJeff Roberson slab = LIST_FIRST(&zone->uz_free_slab); 6008355f576SJeff Roberson while (slab && extra) { 6018355f576SJeff Roberson n = LIST_NEXT(slab, us_link); 6028355f576SJeff Roberson 6038355f576SJeff Roberson /* We have no where to free these to */ 6048355f576SJeff Roberson if (slab->us_flags & UMA_SLAB_BOOT) { 6058355f576SJeff Roberson slab = n; 6068355f576SJeff Roberson continue; 6078355f576SJeff Roberson } 6088355f576SJeff Roberson 6098355f576SJeff Roberson LIST_REMOVE(slab, us_link); 6108355f576SJeff Roberson zone->uz_pages -= zone->uz_ppera; 6118355f576SJeff Roberson zone->uz_free -= zone->uz_ipers; 6128355f576SJeff Roberson if (zone->uz_fini) 6138355f576SJeff Roberson for (i = 0; i < zone->uz_ipers; i++) 6148355f576SJeff Roberson zone->uz_fini( 6158355f576SJeff Roberson slab->us_data + (zone->uz_rsize * i), 6168355f576SJeff Roberson zone->uz_size); 6178355f576SJeff Roberson flags = slab->us_flags; 6188355f576SJeff Roberson mem = slab->us_data; 6198355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_OFFPAGE) { 6208355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_MALLOC) { 6218355f576SJeff Roberson UMA_HASH_REMOVE(mallochash, 6228355f576SJeff Roberson slab, slab->us_data); 6238355f576SJeff Roberson } else { 6248355f576SJeff Roberson UMA_HASH_REMOVE(&zone->uz_hash, 6258355f576SJeff Roberson slab, slab->us_data); 6268355f576SJeff Roberson } 6278355f576SJeff Roberson uma_zfree_internal(slabzone, slab, NULL, 0); 6288355f576SJeff Roberson } else if (zone->uz_flags & UMA_ZFLAG_MALLOC) 6298355f576SJeff Roberson UMA_HASH_REMOVE(mallochash, slab, slab->us_data); 6308355f576SJeff Roberson #ifdef UMA_DEBUG 6318355f576SJeff Roberson printf("%s: Returning %d bytes.\n", 6328355f576SJeff Roberson zone->uz_name, UMA_SLAB_SIZE * zone->uz_ppera); 6338355f576SJeff Roberson #endif 6348355f576SJeff Roberson zone->uz_freef(mem, UMA_SLAB_SIZE * zone->uz_ppera, flags); 6358355f576SJeff Roberson 6368355f576SJeff Roberson slab = n; 6378355f576SJeff Roberson extra--; 6388355f576SJeff Roberson } 6398355f576SJeff Roberson 6408355f576SJeff Roberson finished: 6418355f576SJeff Roberson ZONE_UNLOCK(zone); 6428355f576SJeff Roberson } 6438355f576SJeff Roberson 6448355f576SJeff Roberson /* 6458355f576SJeff Roberson * Allocate a new slab for a zone. This does not insert the slab onto a list. 6468355f576SJeff Roberson * 6478355f576SJeff Roberson * Arguments: 6488355f576SJeff Roberson * zone The zone to allocate slabs for 6498355f576SJeff Roberson * wait Shall we wait? 6508355f576SJeff Roberson * 6518355f576SJeff Roberson * Returns: 6528355f576SJeff Roberson * The slab that was allocated or NULL if there is no memory and the 6538355f576SJeff Roberson * caller specified M_NOWAIT. 6548355f576SJeff Roberson * 6558355f576SJeff Roberson */ 6568355f576SJeff Roberson static uma_slab_t 6578355f576SJeff Roberson slab_zalloc(uma_zone_t zone, int wait) 6588355f576SJeff Roberson { 6598355f576SJeff Roberson uma_slab_t slab; /* Starting slab */ 6608355f576SJeff Roberson u_int8_t *mem; 6618355f576SJeff Roberson u_int8_t flags; 6628355f576SJeff Roberson int i; 6638355f576SJeff Roberson 664a553d4b8SJeff Roberson slab = NULL; 665a553d4b8SJeff Roberson 6668355f576SJeff Roberson #ifdef UMA_DEBUG 6678355f576SJeff Roberson printf("slab_zalloc: Allocating a new slab for %s\n", zone->uz_name); 6688355f576SJeff Roberson #endif 6698355f576SJeff Roberson ZONE_UNLOCK(zone); 670a553d4b8SJeff Roberson 671a553d4b8SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_OFFPAGE) { 672a553d4b8SJeff Roberson slab = uma_zalloc_internal(slabzone, NULL, wait, NULL); 673a553d4b8SJeff Roberson if (slab == NULL) { 674a553d4b8SJeff Roberson ZONE_LOCK(zone); 675a553d4b8SJeff Roberson return NULL; 676a553d4b8SJeff Roberson } 677a553d4b8SJeff Roberson } 678a553d4b8SJeff Roberson 679a553d4b8SJeff Roberson if (booted || (zone->uz_flags & UMA_ZFLAG_PRIVALLOC)) { 6808355f576SJeff Roberson mtx_lock(&Giant); 681a553d4b8SJeff Roberson mem = zone->uz_allocf(zone, 6828355f576SJeff Roberson zone->uz_ppera * UMA_SLAB_SIZE, &flags, wait); 6838355f576SJeff Roberson mtx_unlock(&Giant); 684a553d4b8SJeff Roberson if (mem == NULL) { 6858355f576SJeff Roberson ZONE_LOCK(zone); 6868355f576SJeff Roberson return (NULL); 687a553d4b8SJeff Roberson } 6888355f576SJeff Roberson } else { 689a553d4b8SJeff Roberson uma_slab_t tmps; 6908355f576SJeff Roberson 6918355f576SJeff Roberson if (zone->uz_ppera > 1) 6928355f576SJeff Roberson panic("UMA: Attemping to allocate multiple pages before vm has started.\n"); 6938355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_MALLOC) 6948355f576SJeff Roberson panic("Mallocing before uma_startup2 has been called.\n"); 6958355f576SJeff Roberson if (uma_boot_free == 0) 6968355f576SJeff Roberson panic("UMA: Ran out of pre init pages, increase UMA_BOOT_PAGES\n"); 697a553d4b8SJeff Roberson tmps = LIST_FIRST(&uma_boot_pages); 698a553d4b8SJeff Roberson LIST_REMOVE(tmps, us_link); 6998355f576SJeff Roberson uma_boot_free--; 700a553d4b8SJeff Roberson mem = tmps->us_data; 7018355f576SJeff Roberson } 7028355f576SJeff Roberson 703a553d4b8SJeff Roberson ZONE_LOCK(zone); 7048355f576SJeff Roberson 7058355f576SJeff Roberson /* Alloc slab structure for offpage, otherwise adjust it's position */ 7068355f576SJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_OFFPAGE)) { 7078355f576SJeff Roberson slab = (uma_slab_t )(mem + zone->uz_pgoff); 7088355f576SJeff Roberson } else { 7098355f576SJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_MALLOC)) 7108355f576SJeff Roberson UMA_HASH_INSERT(&zone->uz_hash, slab, mem); 7118355f576SJeff Roberson } 7128355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_MALLOC) { 7138355f576SJeff Roberson #ifdef UMA_DEBUG 7148355f576SJeff Roberson printf("Inserting %p into malloc hash from slab %p\n", 7158355f576SJeff Roberson mem, slab); 7168355f576SJeff Roberson #endif 717a553d4b8SJeff Roberson /* XXX Yikes! No lock on the malloc hash! */ 7188355f576SJeff Roberson UMA_HASH_INSERT(mallochash, slab, mem); 7198355f576SJeff Roberson } 7208355f576SJeff Roberson 7218355f576SJeff Roberson slab->us_zone = zone; 7228355f576SJeff Roberson slab->us_data = mem; 7238355f576SJeff Roberson 7248355f576SJeff Roberson /* 7258355f576SJeff Roberson * This is intended to spread data out across cache lines. 7268355f576SJeff Roberson * 7278355f576SJeff Roberson * This code doesn't seem to work properly on x86, and on alpha 7288355f576SJeff Roberson * it makes absolutely no performance difference. I'm sure it could 7298355f576SJeff Roberson * use some tuning, but sun makes outrageous claims about it's 7308355f576SJeff Roberson * performance. 7318355f576SJeff Roberson */ 7328355f576SJeff Roberson #if 0 7338355f576SJeff Roberson if (zone->uz_cachemax) { 7348355f576SJeff Roberson slab->us_data += zone->uz_cacheoff; 7358355f576SJeff Roberson zone->uz_cacheoff += UMA_CACHE_INC; 7368355f576SJeff Roberson if (zone->uz_cacheoff > zone->uz_cachemax) 7378355f576SJeff Roberson zone->uz_cacheoff = 0; 7388355f576SJeff Roberson } 7398355f576SJeff Roberson #endif 7408355f576SJeff Roberson 7418355f576SJeff Roberson slab->us_freecount = zone->uz_ipers; 7428355f576SJeff Roberson slab->us_firstfree = 0; 7438355f576SJeff Roberson slab->us_flags = flags; 7448355f576SJeff Roberson for (i = 0; i < zone->uz_ipers; i++) 7458355f576SJeff Roberson slab->us_freelist[i] = i+1; 7468355f576SJeff Roberson 7478355f576SJeff Roberson if (zone->uz_init) 7488355f576SJeff Roberson for (i = 0; i < zone->uz_ipers; i++) 7498355f576SJeff Roberson zone->uz_init(slab->us_data + (zone->uz_rsize * i), 7508355f576SJeff Roberson zone->uz_size); 7518355f576SJeff Roberson 7528355f576SJeff Roberson zone->uz_pages += zone->uz_ppera; 7538355f576SJeff Roberson zone->uz_free += zone->uz_ipers; 7548355f576SJeff Roberson 7558355f576SJeff Roberson return (slab); 7568355f576SJeff Roberson } 7578355f576SJeff Roberson 7588355f576SJeff Roberson /* 7598355f576SJeff Roberson * Allocates a number of pages from the system 7608355f576SJeff Roberson * 7618355f576SJeff Roberson * Arguments: 7628355f576SJeff Roberson * zone Unused 7638355f576SJeff Roberson * bytes The number of bytes requested 7648355f576SJeff Roberson * wait Shall we wait? 7658355f576SJeff Roberson * 7668355f576SJeff Roberson * Returns: 7678355f576SJeff Roberson * A pointer to the alloced memory or possibly 7688355f576SJeff Roberson * NULL if M_NOWAIT is set. 7698355f576SJeff Roberson */ 7708355f576SJeff Roberson static void * 7718355f576SJeff Roberson page_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait) 7728355f576SJeff Roberson { 7738355f576SJeff Roberson void *p; /* Returned page */ 7748355f576SJeff Roberson 7758355f576SJeff Roberson /* 7768355f576SJeff Roberson * XXX The original zone allocator did this, but I don't think it's 777157d7b35SAlfred Perlstein * necessary in current. 7788355f576SJeff Roberson */ 7798355f576SJeff Roberson 7808355f576SJeff Roberson if (lockstatus(&kernel_map->lock, NULL)) { 7818355f576SJeff Roberson *pflag = UMA_SLAB_KMEM; 7828355f576SJeff Roberson p = (void *) kmem_malloc(kmem_map, bytes, wait); 7838355f576SJeff Roberson } else { 7848355f576SJeff Roberson *pflag = UMA_SLAB_KMAP; 7858355f576SJeff Roberson p = (void *) kmem_alloc(kernel_map, bytes); 7868355f576SJeff Roberson } 7878355f576SJeff Roberson 7888355f576SJeff Roberson return (p); 7898355f576SJeff Roberson } 7908355f576SJeff Roberson 7918355f576SJeff Roberson /* 7928355f576SJeff Roberson * Allocates a number of pages from within an object 7938355f576SJeff Roberson * 7948355f576SJeff Roberson * Arguments: 7958355f576SJeff Roberson * zone Unused 7968355f576SJeff Roberson * bytes The number of bytes requested 7978355f576SJeff Roberson * wait Shall we wait? 7988355f576SJeff Roberson * 7998355f576SJeff Roberson * Returns: 8008355f576SJeff Roberson * A pointer to the alloced memory or possibly 8018355f576SJeff Roberson * NULL if M_NOWAIT is set. 8028355f576SJeff Roberson */ 8038355f576SJeff Roberson static void * 8048355f576SJeff Roberson obj_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait) 8058355f576SJeff Roberson { 8068355f576SJeff Roberson vm_offset_t zkva; 8078355f576SJeff Roberson vm_offset_t retkva; 8088355f576SJeff Roberson vm_page_t p; 8098355f576SJeff Roberson int pages; 8108355f576SJeff Roberson 8118355f576SJeff Roberson retkva = NULL; 8128355f576SJeff Roberson pages = zone->uz_pages; 8138355f576SJeff Roberson 8148355f576SJeff Roberson /* 8158355f576SJeff Roberson * This looks a little weird since we're getting one page at a time 8168355f576SJeff Roberson */ 8178355f576SJeff Roberson while (bytes > 0) { 8188355f576SJeff Roberson p = vm_page_alloc(zone->uz_obj, pages, 8198355f576SJeff Roberson VM_ALLOC_INTERRUPT); 8208355f576SJeff Roberson if (p == NULL) 8218355f576SJeff Roberson return (NULL); 8228355f576SJeff Roberson 8238355f576SJeff Roberson zkva = zone->uz_kva + pages * PAGE_SIZE; 8248355f576SJeff Roberson if (retkva == NULL) 8258355f576SJeff Roberson retkva = zkva; 8268355f576SJeff Roberson pmap_qenter(zkva, &p, 1); 8278355f576SJeff Roberson bytes -= PAGE_SIZE; 8288355f576SJeff Roberson pages += 1; 8298355f576SJeff Roberson } 8308355f576SJeff Roberson 8318355f576SJeff Roberson *flags = UMA_SLAB_PRIV; 8328355f576SJeff Roberson 8338355f576SJeff Roberson return ((void *)retkva); 8348355f576SJeff Roberson } 8358355f576SJeff Roberson 8368355f576SJeff Roberson /* 8378355f576SJeff Roberson * Frees a number of pages to the system 8388355f576SJeff Roberson * 8398355f576SJeff Roberson * Arguments: 8408355f576SJeff Roberson * mem A pointer to the memory to be freed 8418355f576SJeff Roberson * size The size of the memory being freed 8428355f576SJeff Roberson * flags The original p->us_flags field 8438355f576SJeff Roberson * 8448355f576SJeff Roberson * Returns: 8458355f576SJeff Roberson * Nothing 8468355f576SJeff Roberson * 8478355f576SJeff Roberson */ 8488355f576SJeff Roberson static void 8498355f576SJeff Roberson page_free(void *mem, int size, u_int8_t flags) 8508355f576SJeff Roberson { 8518355f576SJeff Roberson vm_map_t map; 8528355f576SJeff Roberson if (flags & UMA_SLAB_KMEM) 8538355f576SJeff Roberson map = kmem_map; 8548355f576SJeff Roberson else if (flags & UMA_SLAB_KMAP) 8558355f576SJeff Roberson map = kernel_map; 8568355f576SJeff Roberson else 8578355f576SJeff Roberson panic("UMA: page_free used with invalid flags %d\n", flags); 8588355f576SJeff Roberson 8598355f576SJeff Roberson kmem_free(map, (vm_offset_t)mem, size); 8608355f576SJeff Roberson } 8618355f576SJeff Roberson 8628355f576SJeff Roberson /* 8638355f576SJeff Roberson * Zero fill initializer 8648355f576SJeff Roberson * 8658355f576SJeff Roberson * Arguments/Returns follow uma_init specifications 8668355f576SJeff Roberson * 8678355f576SJeff Roberson */ 8688355f576SJeff Roberson static void 8698355f576SJeff Roberson zero_init(void *mem, int size) 8708355f576SJeff Roberson { 8718355f576SJeff Roberson bzero(mem, size); 8728355f576SJeff Roberson } 8738355f576SJeff Roberson 8748355f576SJeff Roberson /* 8758355f576SJeff Roberson * Finish creating a small uma zone. This calculates ipers, and the zone size. 8768355f576SJeff Roberson * 8778355f576SJeff Roberson * Arguments 8788355f576SJeff Roberson * zone The zone we should initialize 8798355f576SJeff Roberson * 8808355f576SJeff Roberson * Returns 8818355f576SJeff Roberson * Nothing 8828355f576SJeff Roberson */ 8838355f576SJeff Roberson static void 8848355f576SJeff Roberson zone_small_init(uma_zone_t zone) 8858355f576SJeff Roberson { 8868355f576SJeff Roberson int rsize; 8878355f576SJeff Roberson int memused; 8888355f576SJeff Roberson int ipers; 8898355f576SJeff Roberson 8908355f576SJeff Roberson rsize = zone->uz_size; 8918355f576SJeff Roberson 8928355f576SJeff Roberson if (rsize < UMA_SMALLEST_UNIT) 8938355f576SJeff Roberson rsize = UMA_SMALLEST_UNIT; 8948355f576SJeff Roberson 8958355f576SJeff Roberson if (rsize & zone->uz_align) 8968355f576SJeff Roberson rsize = (rsize & ~zone->uz_align) + (zone->uz_align + 1); 8978355f576SJeff Roberson 8988355f576SJeff Roberson zone->uz_rsize = rsize; 8998355f576SJeff Roberson 9008355f576SJeff Roberson rsize += 1; /* Account for the byte of linkage */ 9018355f576SJeff Roberson zone->uz_ipers = (UMA_SLAB_SIZE - sizeof(struct uma_slab)) / rsize; 9028355f576SJeff Roberson zone->uz_ppera = 1; 9038355f576SJeff Roberson 9048355f576SJeff Roberson memused = zone->uz_ipers * zone->uz_rsize; 9058355f576SJeff Roberson 9068355f576SJeff Roberson /* Can we do any better? */ 9078355f576SJeff Roberson if ((UMA_SLAB_SIZE - memused) >= UMA_MAX_WASTE) { 9088355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_INTERNAL) 9098355f576SJeff Roberson return; 9108355f576SJeff Roberson ipers = UMA_SLAB_SIZE / zone->uz_rsize; 9118355f576SJeff Roberson if (ipers > zone->uz_ipers) { 9128355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_OFFPAGE; 9138355f576SJeff Roberson zone->uz_ipers = ipers; 9148355f576SJeff Roberson } 9158355f576SJeff Roberson } 9168355f576SJeff Roberson 9178355f576SJeff Roberson } 9188355f576SJeff Roberson 9198355f576SJeff Roberson /* 9208355f576SJeff Roberson * Finish creating a large (> UMA_SLAB_SIZE) uma zone. Just give in and do 9218355f576SJeff Roberson * OFFPAGE for now. When I can allow for more dynamic slab sizes this will be 9228355f576SJeff Roberson * more complicated. 9238355f576SJeff Roberson * 9248355f576SJeff Roberson * Arguments 9258355f576SJeff Roberson * zone The zone we should initialize 9268355f576SJeff Roberson * 9278355f576SJeff Roberson * Returns 9288355f576SJeff Roberson * Nothing 9298355f576SJeff Roberson */ 9308355f576SJeff Roberson static void 9318355f576SJeff Roberson zone_large_init(uma_zone_t zone) 9328355f576SJeff Roberson { 9338355f576SJeff Roberson int pages; 9348355f576SJeff Roberson 9358355f576SJeff Roberson pages = zone->uz_size / UMA_SLAB_SIZE; 9368355f576SJeff Roberson 9378355f576SJeff Roberson /* Account for remainder */ 9388355f576SJeff Roberson if ((pages * UMA_SLAB_SIZE) < zone->uz_size) 9398355f576SJeff Roberson pages++; 9408355f576SJeff Roberson 9418355f576SJeff Roberson zone->uz_ppera = pages; 9428355f576SJeff Roberson zone->uz_ipers = 1; 9438355f576SJeff Roberson 9448355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_OFFPAGE; 9458355f576SJeff Roberson zone->uz_rsize = zone->uz_size; 9468355f576SJeff Roberson } 9478355f576SJeff Roberson 9488355f576SJeff Roberson /* 9498355f576SJeff Roberson * Zone header ctor. This initializes all fields, locks, etc. And inserts 9508355f576SJeff Roberson * the zone onto the global zone list. 9518355f576SJeff Roberson * 9528355f576SJeff Roberson * Arguments/Returns follow uma_ctor specifications 9538355f576SJeff Roberson * udata Actually uma_zcreat_args 9548355f576SJeff Roberson * 9558355f576SJeff Roberson */ 9568355f576SJeff Roberson 9578355f576SJeff Roberson static void 9588355f576SJeff Roberson zone_ctor(void *mem, int size, void *udata) 9598355f576SJeff Roberson { 9608355f576SJeff Roberson struct uma_zctor_args *arg = udata; 9618355f576SJeff Roberson uma_zone_t zone = mem; 9628355f576SJeff Roberson int cplen; 9638355f576SJeff Roberson int cpu; 9648355f576SJeff Roberson 9658355f576SJeff Roberson bzero(zone, size); 9668355f576SJeff Roberson zone->uz_name = arg->name; 9678355f576SJeff Roberson zone->uz_size = arg->size; 9688355f576SJeff Roberson zone->uz_ctor = arg->ctor; 9698355f576SJeff Roberson zone->uz_dtor = arg->dtor; 9708355f576SJeff Roberson zone->uz_init = arg->uminit; 9718355f576SJeff Roberson zone->uz_align = arg->align; 9728355f576SJeff Roberson zone->uz_free = 0; 9738355f576SJeff Roberson zone->uz_pages = 0; 9748355f576SJeff Roberson zone->uz_flags = 0; 9758355f576SJeff Roberson zone->uz_allocf = page_alloc; 9768355f576SJeff Roberson zone->uz_freef = page_free; 9778355f576SJeff Roberson 9788355f576SJeff Roberson if (arg->flags & UMA_ZONE_ZINIT) 9798355f576SJeff Roberson zone->uz_init = zero_init; 9808355f576SJeff Roberson 9818355f576SJeff Roberson if (arg->flags & UMA_ZONE_INTERNAL) 9828355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_INTERNAL; 9838355f576SJeff Roberson 9848355f576SJeff Roberson if (arg->flags & UMA_ZONE_MALLOC) 9858355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_MALLOC; 9868355f576SJeff Roberson 9878355f576SJeff Roberson if (arg->flags & UMA_ZONE_NOFREE) 9888355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_NOFREE; 9898355f576SJeff Roberson 9908355f576SJeff Roberson if (zone->uz_size > UMA_SLAB_SIZE) 9918355f576SJeff Roberson zone_large_init(zone); 9928355f576SJeff Roberson else 9938355f576SJeff Roberson zone_small_init(zone); 9948355f576SJeff Roberson 9958355f576SJeff Roberson /* We do this so that the per cpu lock name is unique for each zone */ 9968355f576SJeff Roberson memcpy(zone->uz_lname, "PCPU ", 5); 9978355f576SJeff Roberson cplen = min(strlen(zone->uz_name) + 1, LOCKNAME_LEN - 6); 9988355f576SJeff Roberson memcpy(zone->uz_lname+5, zone->uz_name, cplen); 9998355f576SJeff Roberson zone->uz_lname[LOCKNAME_LEN - 1] = '\0'; 10008355f576SJeff Roberson 10018355f576SJeff Roberson /* 10028355f576SJeff Roberson * If we're putting the slab header in the actual page we need to 10038355f576SJeff Roberson * figure out where in each page it goes. This calculates a right 10048355f576SJeff Roberson * justified offset into the memory on a ALIGN_PTR boundary. 10058355f576SJeff Roberson */ 10068355f576SJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_OFFPAGE)) { 10078355f576SJeff Roberson int totsize; 10088355f576SJeff Roberson int waste; 10098355f576SJeff Roberson 10108355f576SJeff Roberson /* Size of the slab struct and free list */ 10118355f576SJeff Roberson totsize = sizeof(struct uma_slab) + zone->uz_ipers; 10128355f576SJeff Roberson if (totsize & UMA_ALIGN_PTR) 10138355f576SJeff Roberson totsize = (totsize & ~UMA_ALIGN_PTR) + 10148355f576SJeff Roberson (UMA_ALIGN_PTR + 1); 10158355f576SJeff Roberson zone->uz_pgoff = UMA_SLAB_SIZE - totsize; 10168355f576SJeff Roberson 10178355f576SJeff Roberson waste = zone->uz_pgoff; 10188355f576SJeff Roberson waste -= (zone->uz_ipers * zone->uz_rsize); 10198355f576SJeff Roberson 10208355f576SJeff Roberson /* 10218355f576SJeff Roberson * This calculates how much space we have for cache line size 10228355f576SJeff Roberson * optimizations. It works by offseting each slab slightly. 10238355f576SJeff Roberson * Currently it breaks on x86, and so it is disabled. 10248355f576SJeff Roberson */ 10258355f576SJeff Roberson 10268355f576SJeff Roberson if (zone->uz_align < UMA_CACHE_INC && waste > UMA_CACHE_INC) { 10278355f576SJeff Roberson zone->uz_cachemax = waste - UMA_CACHE_INC; 10288355f576SJeff Roberson zone->uz_cacheoff = 0; 10298355f576SJeff Roberson } 10308355f576SJeff Roberson 10318355f576SJeff Roberson totsize = zone->uz_pgoff + sizeof(struct uma_slab) 10328355f576SJeff Roberson + zone->uz_ipers; 10338355f576SJeff Roberson /* I don't think it's possible, but I'll make sure anyway */ 10348355f576SJeff Roberson if (totsize > UMA_SLAB_SIZE) { 10358355f576SJeff Roberson printf("zone %s ipers %d rsize %d size %d\n", 10368355f576SJeff Roberson zone->uz_name, zone->uz_ipers, zone->uz_rsize, 10378355f576SJeff Roberson zone->uz_size); 10388355f576SJeff Roberson panic("UMA slab won't fit.\n"); 10398355f576SJeff Roberson } 10408355f576SJeff Roberson } else { 10415300d9ddSJeff Roberson struct slabhead *newhash; 10425300d9ddSJeff Roberson int hashsize; 10435300d9ddSJeff Roberson 10445300d9ddSJeff Roberson hashsize = 0; 10455300d9ddSJeff Roberson newhash = hash_alloc(&hashsize); 10465300d9ddSJeff Roberson hash_expand(&zone->uz_hash, newhash, hashsize); 10478355f576SJeff Roberson zone->uz_pgoff = 0; 10488355f576SJeff Roberson } 10498355f576SJeff Roberson 10508355f576SJeff Roberson #ifdef UMA_DEBUG 10518355f576SJeff Roberson printf("%s(%p) size = %d ipers = %d ppera = %d pgoff = %d\n", 10528355f576SJeff Roberson zone->uz_name, zone, 10538355f576SJeff Roberson zone->uz_size, zone->uz_ipers, 10548355f576SJeff Roberson zone->uz_ppera, zone->uz_pgoff); 10558355f576SJeff Roberson #endif 10568355f576SJeff Roberson ZONE_LOCK_INIT(zone); 10578355f576SJeff Roberson 10588355f576SJeff Roberson mtx_lock(&uma_mtx); 10598355f576SJeff Roberson LIST_INSERT_HEAD(&uma_zones, zone, uz_link); 10608355f576SJeff Roberson mtx_unlock(&uma_mtx); 10618355f576SJeff Roberson 10628355f576SJeff Roberson /* 10638355f576SJeff Roberson * Some internal zones don't have room allocated for the per cpu 10648355f576SJeff Roberson * caches. If we're internal, bail out here. 10658355f576SJeff Roberson */ 10668355f576SJeff Roberson 10678355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_INTERNAL) 10688355f576SJeff Roberson return; 10698355f576SJeff Roberson 10708355f576SJeff Roberson if (zone->uz_ipers < UMA_BUCKET_SIZE) 1071a553d4b8SJeff Roberson zone->uz_count = zone->uz_ipers - 1; 10728355f576SJeff Roberson else 1073a553d4b8SJeff Roberson zone->uz_count = UMA_BUCKET_SIZE - 1; 1074a553d4b8SJeff Roberson 1075a553d4b8SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) 10768355f576SJeff Roberson CPU_LOCK_INIT(zone, cpu); 10778355f576SJeff Roberson } 10788355f576SJeff Roberson 10798355f576SJeff Roberson /* 10809c2cd7e5SJeff Roberson * Zone header dtor. This frees all data, destroys locks, frees the hash table 10819c2cd7e5SJeff Roberson * and removes the zone from the global list. 10829c2cd7e5SJeff Roberson * 10839c2cd7e5SJeff Roberson * Arguments/Returns follow uma_dtor specifications 10849c2cd7e5SJeff Roberson * udata unused 10859c2cd7e5SJeff Roberson */ 10869c2cd7e5SJeff Roberson 10879c2cd7e5SJeff Roberson static void 10889c2cd7e5SJeff Roberson zone_dtor(void *arg, int size, void *udata) 10899c2cd7e5SJeff Roberson { 10909c2cd7e5SJeff Roberson uma_zone_t zone; 10919c2cd7e5SJeff Roberson int cpu; 10929c2cd7e5SJeff Roberson 10939c2cd7e5SJeff Roberson zone = (uma_zone_t)arg; 10949c2cd7e5SJeff Roberson 10959c2cd7e5SJeff Roberson mtx_lock(&uma_mtx); 10969c2cd7e5SJeff Roberson LIST_REMOVE(zone, uz_link); 10979c2cd7e5SJeff Roberson mtx_unlock(&uma_mtx); 10989c2cd7e5SJeff Roberson 10999c2cd7e5SJeff Roberson ZONE_LOCK(zone); 11009c2cd7e5SJeff Roberson zone->uz_wssize = 0; 11019c2cd7e5SJeff Roberson ZONE_UNLOCK(zone); 11029c2cd7e5SJeff Roberson 11039c2cd7e5SJeff Roberson zone_drain(zone); 11049c2cd7e5SJeff Roberson ZONE_LOCK(zone); 11059c2cd7e5SJeff Roberson if (zone->uz_free != 0) 11069c2cd7e5SJeff Roberson printf("Zone %s was not empty. Lost %d pages of memory.\n", 11079c2cd7e5SJeff Roberson zone->uz_name, zone->uz_pages); 11089c2cd7e5SJeff Roberson 11099c2cd7e5SJeff Roberson if ((zone->uz_flags & UMA_ZFLAG_INTERNAL) != 0) 11109c2cd7e5SJeff Roberson for (cpu = 0; cpu < maxcpu; cpu++) 11119c2cd7e5SJeff Roberson CPU_LOCK_FINI(zone, cpu); 11129c2cd7e5SJeff Roberson 11139c2cd7e5SJeff Roberson if ((zone->uz_flags & UMA_ZFLAG_OFFPAGE) != 0) 11145300d9ddSJeff Roberson hash_free(zone->uz_hash.uh_slab_hash, 11155300d9ddSJeff Roberson zone->uz_hash.uh_hashsize); 11169c2cd7e5SJeff Roberson 11179c2cd7e5SJeff Roberson ZONE_UNLOCK(zone); 11189c2cd7e5SJeff Roberson ZONE_LOCK_FINI(zone); 11199c2cd7e5SJeff Roberson } 11209c2cd7e5SJeff Roberson /* 11218355f576SJeff Roberson * Traverses every zone in the system and calls a callback 11228355f576SJeff Roberson * 11238355f576SJeff Roberson * Arguments: 11248355f576SJeff Roberson * zfunc A pointer to a function which accepts a zone 11258355f576SJeff Roberson * as an argument. 11268355f576SJeff Roberson * 11278355f576SJeff Roberson * Returns: 11288355f576SJeff Roberson * Nothing 11298355f576SJeff Roberson */ 11308355f576SJeff Roberson static void 11318355f576SJeff Roberson zone_foreach(void (*zfunc)(uma_zone_t)) 11328355f576SJeff Roberson { 11338355f576SJeff Roberson uma_zone_t zone; 11348355f576SJeff Roberson 11358355f576SJeff Roberson mtx_lock(&uma_mtx); 11368355f576SJeff Roberson LIST_FOREACH(zone, &uma_zones, uz_link) { 11378355f576SJeff Roberson zfunc(zone); 11388355f576SJeff Roberson } 11398355f576SJeff Roberson mtx_unlock(&uma_mtx); 11408355f576SJeff Roberson } 11418355f576SJeff Roberson 11428355f576SJeff Roberson /* Public functions */ 11438355f576SJeff Roberson /* See uma.h */ 11448355f576SJeff Roberson void 11458355f576SJeff Roberson uma_startup(void *bootmem) 11468355f576SJeff Roberson { 11478355f576SJeff Roberson struct uma_zctor_args args; 11488355f576SJeff Roberson uma_slab_t slab; 11498355f576SJeff Roberson int slabsize; 11508355f576SJeff Roberson int i; 11518355f576SJeff Roberson 11528355f576SJeff Roberson #ifdef UMA_DEBUG 11538355f576SJeff Roberson printf("Creating uma zone headers zone.\n"); 11548355f576SJeff Roberson #endif 11558355f576SJeff Roberson #ifdef SMP 11568355f576SJeff Roberson maxcpu = mp_maxid + 1; 11578355f576SJeff Roberson #else 11588355f576SJeff Roberson maxcpu = 1; 11598355f576SJeff Roberson #endif 11608355f576SJeff Roberson #ifdef UMA_DEBUG 11618355f576SJeff Roberson printf("Max cpu = %d, mp_maxid = %d\n", maxcpu, mp_maxid); 11628355f576SJeff Roberson Debugger("stop"); 11638355f576SJeff Roberson #endif 11646008862bSJohn Baldwin mtx_init(&uma_mtx, "UMA lock", NULL, MTX_DEF); 11658355f576SJeff Roberson /* "manually" Create the initial zone */ 11668355f576SJeff Roberson args.name = "UMA Zones"; 11678355f576SJeff Roberson args.size = sizeof(struct uma_zone) + 11688355f576SJeff Roberson (sizeof(struct uma_cache) * (maxcpu - 1)); 11698355f576SJeff Roberson args.ctor = zone_ctor; 11709c2cd7e5SJeff Roberson args.dtor = zone_dtor; 11718355f576SJeff Roberson args.uminit = zero_init; 11728355f576SJeff Roberson args.fini = NULL; 11738355f576SJeff Roberson args.align = 32 - 1; 11748355f576SJeff Roberson args.flags = UMA_ZONE_INTERNAL; 11758355f576SJeff Roberson /* The initial zone has no Per cpu queues so it's smaller */ 11768355f576SJeff Roberson zone_ctor(zones, sizeof(struct uma_zone), &args); 11778355f576SJeff Roberson 11788355f576SJeff Roberson #ifdef UMA_DEBUG 11798355f576SJeff Roberson printf("Filling boot free list.\n"); 11808355f576SJeff Roberson #endif 11818355f576SJeff Roberson for (i = 0; i < UMA_BOOT_PAGES; i++) { 11828355f576SJeff Roberson slab = (uma_slab_t)((u_int8_t *)bootmem + (i * UMA_SLAB_SIZE)); 11838355f576SJeff Roberson slab->us_data = (u_int8_t *)slab; 11848355f576SJeff Roberson slab->us_flags = UMA_SLAB_BOOT; 11858355f576SJeff Roberson LIST_INSERT_HEAD(&uma_boot_pages, slab, us_link); 11868355f576SJeff Roberson uma_boot_free++; 11878355f576SJeff Roberson } 11888355f576SJeff Roberson 11898355f576SJeff Roberson #ifdef UMA_DEBUG 11908355f576SJeff Roberson printf("Creating slab zone.\n"); 11918355f576SJeff Roberson #endif 11928355f576SJeff Roberson 11938355f576SJeff Roberson /* 11948355f576SJeff Roberson * This is the max number of free list items we'll have with 11958355f576SJeff Roberson * offpage slabs. 11968355f576SJeff Roberson */ 11978355f576SJeff Roberson 11988355f576SJeff Roberson slabsize = UMA_SLAB_SIZE - sizeof(struct uma_slab); 11998355f576SJeff Roberson slabsize /= UMA_MAX_WASTE; 12008355f576SJeff Roberson slabsize++; /* In case there it's rounded */ 12018355f576SJeff Roberson slabsize += sizeof(struct uma_slab); 12028355f576SJeff Roberson 12038355f576SJeff Roberson /* Now make a zone for slab headers */ 12048355f576SJeff Roberson slabzone = uma_zcreate("UMA Slabs", 12058355f576SJeff Roberson slabsize, 12068355f576SJeff Roberson NULL, NULL, NULL, NULL, 12078355f576SJeff Roberson UMA_ALIGN_PTR, UMA_ZONE_INTERNAL); 12088355f576SJeff Roberson 12098355f576SJeff Roberson hashzone = uma_zcreate("UMA Hash", 12108355f576SJeff Roberson sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT, 12118355f576SJeff Roberson NULL, NULL, NULL, NULL, 12128355f576SJeff Roberson UMA_ALIGN_PTR, UMA_ZONE_INTERNAL); 12138355f576SJeff Roberson 12148355f576SJeff Roberson bucketzone = uma_zcreate("UMA Buckets", sizeof(struct uma_bucket), 12158355f576SJeff Roberson NULL, NULL, NULL, NULL, 12168355f576SJeff Roberson UMA_ALIGN_PTR, UMA_ZONE_INTERNAL); 12178355f576SJeff Roberson 12188355f576SJeff Roberson 12198355f576SJeff Roberson #ifdef UMA_DEBUG 12208355f576SJeff Roberson printf("UMA startup complete.\n"); 12218355f576SJeff Roberson #endif 12228355f576SJeff Roberson } 12238355f576SJeff Roberson 12248355f576SJeff Roberson /* see uma.h */ 12258355f576SJeff Roberson void 12268355f576SJeff Roberson uma_startup2(void *hashmem, u_long elems) 12278355f576SJeff Roberson { 12288355f576SJeff Roberson bzero(hashmem, elems * sizeof(void *)); 12298355f576SJeff Roberson mallochash->uh_slab_hash = hashmem; 12308355f576SJeff Roberson mallochash->uh_hashsize = elems; 12318355f576SJeff Roberson mallochash->uh_hashmask = elems - 1; 12328355f576SJeff Roberson booted = 1; 123386bbae32SJeff Roberson bucket_enable(); 12348355f576SJeff Roberson #ifdef UMA_DEBUG 12358355f576SJeff Roberson printf("UMA startup2 complete.\n"); 12368355f576SJeff Roberson #endif 12378355f576SJeff Roberson } 12388355f576SJeff Roberson 12398355f576SJeff Roberson /* 12408355f576SJeff Roberson * Initialize our callout handle 12418355f576SJeff Roberson * 12428355f576SJeff Roberson */ 12438355f576SJeff Roberson 12448355f576SJeff Roberson static void 12458355f576SJeff Roberson uma_startup3(void) 12468355f576SJeff Roberson { 12478355f576SJeff Roberson #ifdef UMA_DEBUG 12488355f576SJeff Roberson printf("Starting callout.\n"); 12498355f576SJeff Roberson #endif 12508355f576SJeff Roberson callout_init(&uma_callout, 0); 12518355f576SJeff Roberson callout_reset(&uma_callout, UMA_WORKING_TIME * hz, uma_timeout, NULL); 12528355f576SJeff Roberson #ifdef UMA_DEBUG 12538355f576SJeff Roberson printf("UMA startup3 complete.\n"); 12548355f576SJeff Roberson #endif 12558355f576SJeff Roberson } 12568355f576SJeff Roberson 12578355f576SJeff Roberson /* See uma.h */ 12588355f576SJeff Roberson uma_zone_t 12598355f576SJeff Roberson uma_zcreate(char *name, int size, uma_ctor ctor, uma_dtor dtor, uma_init uminit, 12608355f576SJeff Roberson uma_fini fini, int align, u_int16_t flags) 12618355f576SJeff Roberson 12628355f576SJeff Roberson { 12638355f576SJeff Roberson struct uma_zctor_args args; 12648355f576SJeff Roberson 12658355f576SJeff Roberson /* This stuff is essential for the zone ctor */ 12668355f576SJeff Roberson args.name = name; 12678355f576SJeff Roberson args.size = size; 12688355f576SJeff Roberson args.ctor = ctor; 12698355f576SJeff Roberson args.dtor = dtor; 12708355f576SJeff Roberson args.uminit = uminit; 12718355f576SJeff Roberson args.fini = fini; 12728355f576SJeff Roberson args.align = align; 12738355f576SJeff Roberson args.flags = flags; 12748355f576SJeff Roberson 1275a553d4b8SJeff Roberson return (uma_zalloc_internal(zones, &args, M_WAITOK, NULL)); 12768355f576SJeff Roberson } 12778355f576SJeff Roberson 12788355f576SJeff Roberson /* See uma.h */ 12799c2cd7e5SJeff Roberson void 12809c2cd7e5SJeff Roberson uma_zdestroy(uma_zone_t zone) 12819c2cd7e5SJeff Roberson { 12829c2cd7e5SJeff Roberson uma_zfree_internal(zones, zone, NULL, 0); 12839c2cd7e5SJeff Roberson } 12849c2cd7e5SJeff Roberson 12859c2cd7e5SJeff Roberson /* See uma.h */ 12868355f576SJeff Roberson void * 12878355f576SJeff Roberson uma_zalloc_arg(uma_zone_t zone, void *udata, int wait) 12888355f576SJeff Roberson { 12898355f576SJeff Roberson void *item; 12908355f576SJeff Roberson uma_cache_t cache; 12918355f576SJeff Roberson uma_bucket_t bucket; 12928355f576SJeff Roberson int cpu; 12938355f576SJeff Roberson 12948355f576SJeff Roberson /* This is the fast path allocation */ 12958355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC_1 12968355f576SJeff Roberson printf("Allocating one item from %s(%p)\n", zone->uz_name, zone); 12978355f576SJeff Roberson #endif 1298a553d4b8SJeff Roberson 1299a553d4b8SJeff Roberson zalloc_restart: 13008355f576SJeff Roberson cpu = PCPU_GET(cpuid); 13018355f576SJeff Roberson CPU_LOCK(zone, cpu); 13028355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 13038355f576SJeff Roberson 13048355f576SJeff Roberson zalloc_start: 13058355f576SJeff Roberson bucket = cache->uc_allocbucket; 13068355f576SJeff Roberson 13078355f576SJeff Roberson if (bucket) { 13088355f576SJeff Roberson if (bucket->ub_ptr > -1) { 13098355f576SJeff Roberson item = bucket->ub_bucket[bucket->ub_ptr]; 13108355f576SJeff Roberson #ifdef INVARIANTS 13118355f576SJeff Roberson bucket->ub_bucket[bucket->ub_ptr] = NULL; 13128355f576SJeff Roberson #endif 13138355f576SJeff Roberson bucket->ub_ptr--; 13148355f576SJeff Roberson KASSERT(item != NULL, 13158355f576SJeff Roberson ("uma_zalloc: Bucket pointer mangled.")); 13168355f576SJeff Roberson cache->uc_allocs++; 13178355f576SJeff Roberson CPU_UNLOCK(zone, cpu); 13188355f576SJeff Roberson if (zone->uz_ctor) 13198355f576SJeff Roberson zone->uz_ctor(item, zone->uz_size, udata); 13208355f576SJeff Roberson return (item); 13218355f576SJeff Roberson } else if (cache->uc_freebucket) { 13228355f576SJeff Roberson /* 13238355f576SJeff Roberson * We have run out of items in our allocbucket. 13248355f576SJeff Roberson * See if we can switch with our free bucket. 13258355f576SJeff Roberson */ 13268355f576SJeff Roberson if (cache->uc_freebucket->ub_ptr > -1) { 13278355f576SJeff Roberson uma_bucket_t swap; 13288355f576SJeff Roberson 13298355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 13308355f576SJeff Roberson printf("uma_zalloc: Swapping empty with alloc.\n"); 13318355f576SJeff Roberson #endif 13328355f576SJeff Roberson swap = cache->uc_freebucket; 13338355f576SJeff Roberson cache->uc_freebucket = cache->uc_allocbucket; 13348355f576SJeff Roberson cache->uc_allocbucket = swap; 13358355f576SJeff Roberson 13368355f576SJeff Roberson goto zalloc_start; 13378355f576SJeff Roberson } 13388355f576SJeff Roberson } 13398355f576SJeff Roberson } 1340a553d4b8SJeff Roberson ZONE_LOCK(zone); 1341a553d4b8SJeff Roberson /* Since we have locked the zone we may as well send back our stats */ 1342a553d4b8SJeff Roberson zone->uz_allocs += cache->uc_allocs; 1343a553d4b8SJeff Roberson cache->uc_allocs = 0; 13448355f576SJeff Roberson 1345a553d4b8SJeff Roberson /* Our old one is now a free bucket */ 1346a553d4b8SJeff Roberson if (cache->uc_allocbucket) { 1347a553d4b8SJeff Roberson KASSERT(cache->uc_allocbucket->ub_ptr == -1, 1348a553d4b8SJeff Roberson ("uma_zalloc_arg: Freeing a non free bucket.")); 1349a553d4b8SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_bucket, 1350a553d4b8SJeff Roberson cache->uc_allocbucket, ub_link); 1351a553d4b8SJeff Roberson cache->uc_allocbucket = NULL; 1352a553d4b8SJeff Roberson } 13538355f576SJeff Roberson 1354a553d4b8SJeff Roberson /* Check the free list for a new alloc bucket */ 1355a553d4b8SJeff Roberson if ((bucket = LIST_FIRST(&zone->uz_full_bucket)) != NULL) { 1356a553d4b8SJeff Roberson KASSERT(bucket->ub_ptr != -1, 1357a553d4b8SJeff Roberson ("uma_zalloc_arg: Returning an empty bucket.")); 13588355f576SJeff Roberson 1359a553d4b8SJeff Roberson LIST_REMOVE(bucket, ub_link); 1360a553d4b8SJeff Roberson cache->uc_allocbucket = bucket; 1361a553d4b8SJeff Roberson ZONE_UNLOCK(zone); 13628355f576SJeff Roberson goto zalloc_start; 1363a553d4b8SJeff Roberson } 1364a553d4b8SJeff Roberson /* Bump up our uz_count so we get here less */ 1365a553d4b8SJeff Roberson if (zone->uz_count < UMA_BUCKET_SIZE - 1) 1366a553d4b8SJeff Roberson zone->uz_count++; 1367a553d4b8SJeff Roberson 1368a553d4b8SJeff Roberson /* We are no longer associated with this cpu!!! */ 1369a553d4b8SJeff Roberson CPU_UNLOCK(zone, cpu); 13708355f576SJeff Roberson 13718355f576SJeff Roberson /* 1372a553d4b8SJeff Roberson * Now lets just fill a bucket and put it on the free list. If that 1373a553d4b8SJeff Roberson * works we'll restart the allocation from the begining. 1374a553d4b8SJeff Roberson * 1375a553d4b8SJeff Roberson * Try this zone's free list first so we don't allocate extra buckets. 13768355f576SJeff Roberson */ 1377a553d4b8SJeff Roberson 1378a553d4b8SJeff Roberson if ((bucket = LIST_FIRST(&zone->uz_free_bucket)) != NULL) 1379a553d4b8SJeff Roberson LIST_REMOVE(bucket, ub_link); 1380a553d4b8SJeff Roberson 1381a553d4b8SJeff Roberson /* Now we no longer need the zone lock. */ 1382a553d4b8SJeff Roberson ZONE_UNLOCK(zone); 1383a553d4b8SJeff Roberson 1384a553d4b8SJeff Roberson if (bucket == NULL) 1385a553d4b8SJeff Roberson bucket = uma_zalloc_internal(bucketzone, 1386a553d4b8SJeff Roberson NULL, wait, NULL); 1387a553d4b8SJeff Roberson 1388a553d4b8SJeff Roberson if (bucket != NULL) { 1389a553d4b8SJeff Roberson #ifdef INVARIANTS 1390a553d4b8SJeff Roberson bzero(bucket, bucketzone->uz_size); 1391a553d4b8SJeff Roberson #endif 1392a553d4b8SJeff Roberson bucket->ub_ptr = -1; 1393a553d4b8SJeff Roberson 1394a553d4b8SJeff Roberson if (uma_zalloc_internal(zone, udata, wait, bucket)) 1395a553d4b8SJeff Roberson goto zalloc_restart; 1396a553d4b8SJeff Roberson else 1397a553d4b8SJeff Roberson uma_zfree_internal(bucketzone, bucket, NULL, 0); 1398a553d4b8SJeff Roberson } 1399a553d4b8SJeff Roberson /* 1400a553d4b8SJeff Roberson * We may not get a bucket if we recurse, so 1401a553d4b8SJeff Roberson * return an actual item. 1402a553d4b8SJeff Roberson */ 1403a553d4b8SJeff Roberson #ifdef UMA_DEBUG 1404a553d4b8SJeff Roberson printf("uma_zalloc_arg: Bucketzone returned NULL\n"); 1405a553d4b8SJeff Roberson #endif 1406a553d4b8SJeff Roberson 1407a553d4b8SJeff Roberson return (uma_zalloc_internal(zone, udata, wait, NULL)); 14088355f576SJeff Roberson } 14098355f576SJeff Roberson 14108355f576SJeff Roberson /* 14118355f576SJeff Roberson * Allocates an item for an internal zone OR fills a bucket 14128355f576SJeff Roberson * 14138355f576SJeff Roberson * Arguments 14148355f576SJeff Roberson * zone The zone to alloc for. 14158355f576SJeff Roberson * udata The data to be passed to the constructor. 14168355f576SJeff Roberson * wait M_WAITOK or M_NOWAIT. 1417a553d4b8SJeff Roberson * bucket The bucket to fill or NULL 14188355f576SJeff Roberson * 14198355f576SJeff Roberson * Returns 14208355f576SJeff Roberson * NULL if there is no memory and M_NOWAIT is set 14218355f576SJeff Roberson * An item if called on an interal zone 14228355f576SJeff Roberson * Non NULL if called to fill a bucket and it was successful. 14238355f576SJeff Roberson * 14248355f576SJeff Roberson * Discussion: 14258355f576SJeff Roberson * This was much cleaner before it had to do per cpu caches. It is 14268355f576SJeff Roberson * complicated now because it has to handle the simple internal case, and 1427a553d4b8SJeff Roberson * the more involved bucket filling and allocation. 14288355f576SJeff Roberson */ 14298355f576SJeff Roberson 14308355f576SJeff Roberson static void * 1431a553d4b8SJeff Roberson uma_zalloc_internal(uma_zone_t zone, void *udata, int wait, uma_bucket_t bucket) 14328355f576SJeff Roberson { 14338355f576SJeff Roberson uma_slab_t slab; 14348355f576SJeff Roberson u_int8_t freei; 14358355f576SJeff Roberson void *item; 14368355f576SJeff Roberson 14378355f576SJeff Roberson item = NULL; 14388355f576SJeff Roberson 14398355f576SJeff Roberson /* 1440a553d4b8SJeff Roberson * This is to stop us from allocating per cpu buckets while we're 1441a553d4b8SJeff Roberson * running out of UMA_BOOT_PAGES. Otherwise, we would exhaust the 1442a553d4b8SJeff Roberson * boot pages. 14438355f576SJeff Roberson */ 14448355f576SJeff Roberson 144586bbae32SJeff Roberson if (bucketdisable && zone == bucketzone) 14468355f576SJeff Roberson return (NULL); 14478355f576SJeff Roberson 14488355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 14498355f576SJeff Roberson printf("INTERNAL: Allocating one item from %s(%p)\n", zone->uz_name, zone); 14508355f576SJeff Roberson #endif 14518355f576SJeff Roberson ZONE_LOCK(zone); 14528355f576SJeff Roberson 14538355f576SJeff Roberson /* 1454a553d4b8SJeff Roberson * This code is here to limit the number of simultaneous bucket fills 1455a553d4b8SJeff Roberson * for any given zone to the number of per cpu caches in this zone. This 1456a553d4b8SJeff Roberson * is done so that we don't allocate more memory than we really need. 14578355f576SJeff Roberson */ 1458a553d4b8SJeff Roberson 1459a553d4b8SJeff Roberson if (bucket) { 1460a553d4b8SJeff Roberson #ifdef SMP 1461bce97791SJeff Roberson if (zone->uz_fills >= mp_ncpus) { 1462a553d4b8SJeff Roberson #else 1463bce97791SJeff Roberson if (zone->uz_fills > 1) { 14648355f576SJeff Roberson #endif 1465bce97791SJeff Roberson ZONE_UNLOCK(zone); 1466a553d4b8SJeff Roberson return (NULL); 1467bce97791SJeff Roberson } 1468a553d4b8SJeff Roberson 1469a553d4b8SJeff Roberson zone->uz_fills++; 14708355f576SJeff Roberson } 14718355f576SJeff Roberson 14728355f576SJeff Roberson new_slab: 14738355f576SJeff Roberson 14748355f576SJeff Roberson /* Find a slab with some space */ 14758355f576SJeff Roberson if (zone->uz_free) { 14768355f576SJeff Roberson if (!LIST_EMPTY(&zone->uz_part_slab)) { 14778355f576SJeff Roberson slab = LIST_FIRST(&zone->uz_part_slab); 14788355f576SJeff Roberson } else { 14798355f576SJeff Roberson slab = LIST_FIRST(&zone->uz_free_slab); 14808355f576SJeff Roberson LIST_REMOVE(slab, us_link); 14818355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link); 14828355f576SJeff Roberson } 14838355f576SJeff Roberson } else { 14848355f576SJeff Roberson /* 14858355f576SJeff Roberson * This is to prevent us from recursively trying to allocate 14868355f576SJeff Roberson * buckets. The problem is that if an allocation forces us to 14878355f576SJeff Roberson * grab a new bucket we will call page_alloc, which will go off 14888355f576SJeff Roberson * and cause the vm to allocate vm_map_entries. If we need new 14898355f576SJeff Roberson * buckets there too we will recurse in kmem_alloc and bad 14908355f576SJeff Roberson * things happen. So instead we return a NULL bucket, and make 14918355f576SJeff Roberson * the code that allocates buckets smart enough to deal with it */ 14928355f576SJeff Roberson if (zone == bucketzone && zone->uz_recurse != 0) { 14938355f576SJeff Roberson ZONE_UNLOCK(zone); 14948355f576SJeff Roberson return (NULL); 14958355f576SJeff Roberson } 1496af7f9b97SJeff Roberson while (zone->uz_maxpages && 1497af7f9b97SJeff Roberson zone->uz_pages >= zone->uz_maxpages) { 1498af7f9b97SJeff Roberson zone->uz_flags |= UMA_ZFLAG_FULL; 1499af7f9b97SJeff Roberson 1500af7f9b97SJeff Roberson if (wait & M_WAITOK) 1501af7f9b97SJeff Roberson msleep(zone, &zone->uz_lock, PVM, "zonelimit", 0); 1502af7f9b97SJeff Roberson else 1503af7f9b97SJeff Roberson goto alloc_fail; 1504af7f9b97SJeff Roberson 1505af7f9b97SJeff Roberson goto new_slab; 1506af7f9b97SJeff Roberson } 1507af7f9b97SJeff Roberson 15088355f576SJeff Roberson zone->uz_recurse++; 15098355f576SJeff Roberson slab = slab_zalloc(zone, wait); 15108355f576SJeff Roberson zone->uz_recurse--; 15118355f576SJeff Roberson /* 1512af7f9b97SJeff Roberson * We might not have been able to get a slab but another cpu 1513af7f9b97SJeff Roberson * could have while we were unlocked. If we did get a slab put 1514af7f9b97SJeff Roberson * it on the partially used slab list. If not check the free 1515af7f9b97SJeff Roberson * count and restart or fail accordingly. 15168355f576SJeff Roberson */ 1517af7f9b97SJeff Roberson if (slab) 1518af7f9b97SJeff Roberson LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link); 1519af7f9b97SJeff Roberson else if (zone->uz_free == 0) 1520af7f9b97SJeff Roberson goto alloc_fail; 1521a553d4b8SJeff Roberson else 1522af7f9b97SJeff Roberson goto new_slab; 15238355f576SJeff Roberson } 1524a553d4b8SJeff Roberson /* 1525a553d4b8SJeff Roberson * If this is our first time though put this guy on the list. 1526a553d4b8SJeff Roberson */ 1527a553d4b8SJeff Roberson if (bucket != NULL && bucket->ub_ptr == -1) 1528a553d4b8SJeff Roberson LIST_INSERT_HEAD(&zone->uz_full_bucket, 1529a553d4b8SJeff Roberson bucket, ub_link); 1530a553d4b8SJeff Roberson 15318355f576SJeff Roberson 15328355f576SJeff Roberson while (slab->us_freecount) { 15338355f576SJeff Roberson freei = slab->us_firstfree; 15348355f576SJeff Roberson slab->us_firstfree = slab->us_freelist[freei]; 15358355f576SJeff Roberson #ifdef INVARIANTS 15368355f576SJeff Roberson slab->us_freelist[freei] = 255; 15378355f576SJeff Roberson #endif 15388355f576SJeff Roberson slab->us_freecount--; 15398355f576SJeff Roberson zone->uz_free--; 15408355f576SJeff Roberson item = slab->us_data + (zone->uz_rsize * freei); 15418355f576SJeff Roberson 1542a553d4b8SJeff Roberson if (bucket == NULL) { 15438355f576SJeff Roberson zone->uz_allocs++; 15448355f576SJeff Roberson break; 15458355f576SJeff Roberson } 15468355f576SJeff Roberson bucket->ub_bucket[++bucket->ub_ptr] = item; 15478355f576SJeff Roberson 15488355f576SJeff Roberson /* Don't overfill the bucket! */ 1549a553d4b8SJeff Roberson if (bucket->ub_ptr == zone->uz_count) 15508355f576SJeff Roberson break; 15518355f576SJeff Roberson } 15528355f576SJeff Roberson 15538355f576SJeff Roberson /* Move this slab to the full list */ 15548355f576SJeff Roberson if (slab->us_freecount == 0) { 15558355f576SJeff Roberson LIST_REMOVE(slab, us_link); 15568355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_full_slab, slab, us_link); 15578355f576SJeff Roberson } 15588355f576SJeff Roberson 1559a553d4b8SJeff Roberson if (bucket != NULL) { 15608355f576SJeff Roberson /* Try to keep the buckets totally full, but don't block */ 1561a553d4b8SJeff Roberson if (bucket->ub_ptr < zone->uz_count) { 15628355f576SJeff Roberson wait = M_NOWAIT; 15638355f576SJeff Roberson goto new_slab; 1564a553d4b8SJeff Roberson } else 1565a553d4b8SJeff Roberson zone->uz_fills--; 15668355f576SJeff Roberson } 15678355f576SJeff Roberson 15688355f576SJeff Roberson ZONE_UNLOCK(zone); 15698355f576SJeff Roberson 15708355f576SJeff Roberson /* Only construct at this time if we're not filling a bucket */ 1571a553d4b8SJeff Roberson if (bucket == NULL && zone->uz_ctor != NULL) 15728355f576SJeff Roberson zone->uz_ctor(item, zone->uz_size, udata); 15738355f576SJeff Roberson 15748355f576SJeff Roberson return (item); 1575af7f9b97SJeff Roberson 1576af7f9b97SJeff Roberson alloc_fail: 1577af7f9b97SJeff Roberson if (bucket != NULL) 1578af7f9b97SJeff Roberson zone->uz_fills--; 1579af7f9b97SJeff Roberson ZONE_UNLOCK(zone); 1580af7f9b97SJeff Roberson 1581af7f9b97SJeff Roberson if (bucket != NULL && bucket->ub_ptr != -1) 1582af7f9b97SJeff Roberson return (bucket); 1583af7f9b97SJeff Roberson 1584af7f9b97SJeff Roberson return (NULL); 15858355f576SJeff Roberson } 15868355f576SJeff Roberson 15878355f576SJeff Roberson /* See uma.h */ 15888355f576SJeff Roberson void 15898355f576SJeff Roberson uma_zfree_arg(uma_zone_t zone, void *item, void *udata) 15908355f576SJeff Roberson { 15918355f576SJeff Roberson uma_cache_t cache; 15928355f576SJeff Roberson uma_bucket_t bucket; 15938355f576SJeff Roberson int cpu; 15948355f576SJeff Roberson 15958355f576SJeff Roberson /* This is the fast path free */ 15968355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC_1 15978355f576SJeff Roberson printf("Freeing item %p to %s(%p)\n", item, zone->uz_name, zone); 15988355f576SJeff Roberson #endif 1599af7f9b97SJeff Roberson /* 1600af7f9b97SJeff Roberson * The race here is acceptable. If we miss it we'll just have to wait 1601af7f9b97SJeff Roberson * a little longer for the limits to be reset. 1602af7f9b97SJeff Roberson */ 1603af7f9b97SJeff Roberson 1604af7f9b97SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_FULL) 1605af7f9b97SJeff Roberson goto zfree_internal; 1606af7f9b97SJeff Roberson 1607a553d4b8SJeff Roberson zfree_restart: 16088355f576SJeff Roberson cpu = PCPU_GET(cpuid); 16098355f576SJeff Roberson CPU_LOCK(zone, cpu); 16108355f576SJeff Roberson cache = &zone->uz_cpu[cpu]; 16118355f576SJeff Roberson 16128355f576SJeff Roberson zfree_start: 16138355f576SJeff Roberson bucket = cache->uc_freebucket; 16148355f576SJeff Roberson 16158355f576SJeff Roberson if (bucket) { 1616a553d4b8SJeff Roberson /* 1617a553d4b8SJeff Roberson * Do we have room in our bucket? It is OK for this uz count 1618a553d4b8SJeff Roberson * check to be slightly out of sync. 1619a553d4b8SJeff Roberson */ 1620a553d4b8SJeff Roberson 1621a553d4b8SJeff Roberson if (bucket->ub_ptr < zone->uz_count) { 16228355f576SJeff Roberson bucket->ub_ptr++; 16238355f576SJeff Roberson KASSERT(bucket->ub_bucket[bucket->ub_ptr] == NULL, 16248355f576SJeff Roberson ("uma_zfree: Freeing to non free bucket index.")); 16258355f576SJeff Roberson bucket->ub_bucket[bucket->ub_ptr] = item; 16268355f576SJeff Roberson if (zone->uz_dtor) 16278355f576SJeff Roberson zone->uz_dtor(item, zone->uz_size, udata); 1628605cbd6aSJeff Roberson CPU_UNLOCK(zone, cpu); 16298355f576SJeff Roberson return; 16308355f576SJeff Roberson } else if (cache->uc_allocbucket) { 16318355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 16328355f576SJeff Roberson printf("uma_zfree: Swapping buckets.\n"); 16338355f576SJeff Roberson #endif 16348355f576SJeff Roberson /* 16358355f576SJeff Roberson * We have run out of space in our freebucket. 16368355f576SJeff Roberson * See if we can switch with our alloc bucket. 16378355f576SJeff Roberson */ 16388355f576SJeff Roberson if (cache->uc_allocbucket->ub_ptr < 16398355f576SJeff Roberson cache->uc_freebucket->ub_ptr) { 16408355f576SJeff Roberson uma_bucket_t swap; 16418355f576SJeff Roberson 16428355f576SJeff Roberson swap = cache->uc_freebucket; 16438355f576SJeff Roberson cache->uc_freebucket = cache->uc_allocbucket; 16448355f576SJeff Roberson cache->uc_allocbucket = swap; 16458355f576SJeff Roberson 16468355f576SJeff Roberson goto zfree_start; 16478355f576SJeff Roberson } 16488355f576SJeff Roberson } 16498355f576SJeff Roberson } 16508355f576SJeff Roberson 16518355f576SJeff Roberson /* 1652a553d4b8SJeff Roberson * We can get here for two reasons: 16538355f576SJeff Roberson * 16548355f576SJeff Roberson * 1) The buckets are NULL 1655a553d4b8SJeff Roberson * 2) The alloc and free buckets are both somewhat full. 16568355f576SJeff Roberson * 16578355f576SJeff Roberson */ 16588355f576SJeff Roberson 16598355f576SJeff Roberson ZONE_LOCK(zone); 16608355f576SJeff Roberson 16618355f576SJeff Roberson bucket = cache->uc_freebucket; 16628355f576SJeff Roberson cache->uc_freebucket = NULL; 16638355f576SJeff Roberson 16648355f576SJeff Roberson /* Can we throw this on the zone full list? */ 16658355f576SJeff Roberson if (bucket != NULL) { 16668355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 16678355f576SJeff Roberson printf("uma_zfree: Putting old bucket on the free list.\n"); 16688355f576SJeff Roberson #endif 16698355f576SJeff Roberson /* ub_ptr is pointing to the last free item */ 16708355f576SJeff Roberson KASSERT(bucket->ub_ptr != -1, 16718355f576SJeff Roberson ("uma_zfree: Attempting to insert an empty bucket onto the full list.\n")); 16728355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_full_bucket, 16738355f576SJeff Roberson bucket, ub_link); 16748355f576SJeff Roberson } 1675a553d4b8SJeff Roberson if ((bucket = LIST_FIRST(&zone->uz_free_bucket)) != NULL) { 1676a553d4b8SJeff Roberson LIST_REMOVE(bucket, ub_link); 1677a553d4b8SJeff Roberson ZONE_UNLOCK(zone); 1678a553d4b8SJeff Roberson cache->uc_freebucket = bucket; 1679a553d4b8SJeff Roberson goto zfree_start; 1680a553d4b8SJeff Roberson } 1681a553d4b8SJeff Roberson /* We're done with this CPU now */ 1682a553d4b8SJeff Roberson CPU_UNLOCK(zone, cpu); 1683a553d4b8SJeff Roberson 1684a553d4b8SJeff Roberson /* And the zone.. */ 1685a553d4b8SJeff Roberson ZONE_UNLOCK(zone); 1686a553d4b8SJeff Roberson 16878355f576SJeff Roberson #ifdef UMA_DEBUG_ALLOC 16888355f576SJeff Roberson printf("uma_zfree: Allocating new free bucket.\n"); 16898355f576SJeff Roberson #endif 16908355f576SJeff Roberson bucket = uma_zalloc_internal(bucketzone, 1691a553d4b8SJeff Roberson NULL, M_NOWAIT, NULL); 16928355f576SJeff Roberson if (bucket) { 16938355f576SJeff Roberson #ifdef INVARIANTS 16948355f576SJeff Roberson bzero(bucket, bucketzone->uz_size); 16958355f576SJeff Roberson #endif 16968355f576SJeff Roberson bucket->ub_ptr = -1; 1697a553d4b8SJeff Roberson ZONE_LOCK(zone); 1698a553d4b8SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_bucket, 1699a553d4b8SJeff Roberson bucket, ub_link); 17008355f576SJeff Roberson ZONE_UNLOCK(zone); 1701a553d4b8SJeff Roberson goto zfree_restart; 17028355f576SJeff Roberson } 17038355f576SJeff Roberson 1704a553d4b8SJeff Roberson /* 1705a553d4b8SJeff Roberson * If nothing else caught this, we'll just do an internal free. 1706a553d4b8SJeff Roberson */ 17078355f576SJeff Roberson 1708af7f9b97SJeff Roberson zfree_internal: 1709af7f9b97SJeff Roberson 17108355f576SJeff Roberson uma_zfree_internal(zone, item, udata, 0); 17118355f576SJeff Roberson 17128355f576SJeff Roberson return; 17138355f576SJeff Roberson 17148355f576SJeff Roberson } 17158355f576SJeff Roberson 17168355f576SJeff Roberson /* 17178355f576SJeff Roberson * Frees an item to an INTERNAL zone or allocates a free bucket 17188355f576SJeff Roberson * 17198355f576SJeff Roberson * Arguments: 17208355f576SJeff Roberson * zone The zone to free to 17218355f576SJeff Roberson * item The item we're freeing 17228355f576SJeff Roberson * udata User supplied data for the dtor 17238355f576SJeff Roberson * skip Skip the dtor, it was done in uma_zfree_arg 17248355f576SJeff Roberson */ 17258355f576SJeff Roberson 17268355f576SJeff Roberson static void 17278355f576SJeff Roberson uma_zfree_internal(uma_zone_t zone, void *item, void *udata, int skip) 17288355f576SJeff Roberson { 17298355f576SJeff Roberson uma_slab_t slab; 17308355f576SJeff Roberson u_int8_t *mem; 17318355f576SJeff Roberson u_int8_t freei; 17328355f576SJeff Roberson 17338355f576SJeff Roberson ZONE_LOCK(zone); 17348355f576SJeff Roberson 17358355f576SJeff Roberson if (!(zone->uz_flags & UMA_ZFLAG_MALLOC)) { 17368355f576SJeff Roberson mem = (u_int8_t *)((unsigned long)item & (~UMA_SLAB_MASK)); 17378355f576SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_OFFPAGE) 17388355f576SJeff Roberson slab = hash_sfind(&zone->uz_hash, mem); 17398355f576SJeff Roberson else { 17408355f576SJeff Roberson mem += zone->uz_pgoff; 17418355f576SJeff Roberson slab = (uma_slab_t)mem; 17428355f576SJeff Roberson } 17438355f576SJeff Roberson } else { 17448355f576SJeff Roberson slab = (uma_slab_t)udata; 17458355f576SJeff Roberson } 17468355f576SJeff Roberson 17478355f576SJeff Roberson /* Do we need to remove from any lists? */ 17488355f576SJeff Roberson if (slab->us_freecount+1 == zone->uz_ipers) { 17498355f576SJeff Roberson LIST_REMOVE(slab, us_link); 17508355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_slab, slab, us_link); 17518355f576SJeff Roberson } else if (slab->us_freecount == 0) { 17528355f576SJeff Roberson LIST_REMOVE(slab, us_link); 17538355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link); 17548355f576SJeff Roberson } 17558355f576SJeff Roberson 17568355f576SJeff Roberson /* Slab management stuff */ 17578355f576SJeff Roberson freei = ((unsigned long)item - (unsigned long)slab->us_data) 17588355f576SJeff Roberson / zone->uz_rsize; 17598355f576SJeff Roberson #ifdef INVARIANTS 17608355f576SJeff Roberson if (((freei * zone->uz_rsize) + slab->us_data) != item) 17618355f576SJeff Roberson panic("zone: %s(%p) slab %p freed address %p unaligned.\n", 17628355f576SJeff Roberson zone->uz_name, zone, slab, item); 17638355f576SJeff Roberson if (freei >= zone->uz_ipers) 17648355f576SJeff Roberson panic("zone: %s(%p) slab %p freelist %i out of range 0-%d\n", 17658355f576SJeff Roberson zone->uz_name, zone, slab, freei, zone->uz_ipers-1); 17668355f576SJeff Roberson 17678355f576SJeff Roberson if (slab->us_freelist[freei] != 255) { 17688355f576SJeff Roberson printf("Slab at %p, freei %d = %d.\n", 17698355f576SJeff Roberson slab, freei, slab->us_freelist[freei]); 17708355f576SJeff Roberson panic("Duplicate free of item %p from zone %p(%s)\n", 17718355f576SJeff Roberson item, zone, zone->uz_name); 17728355f576SJeff Roberson } 17738355f576SJeff Roberson #endif 17748355f576SJeff Roberson slab->us_freelist[freei] = slab->us_firstfree; 17758355f576SJeff Roberson slab->us_firstfree = freei; 17768355f576SJeff Roberson slab->us_freecount++; 17778355f576SJeff Roberson 17788355f576SJeff Roberson /* Zone statistics */ 17798355f576SJeff Roberson zone->uz_free++; 17808355f576SJeff Roberson 17818355f576SJeff Roberson if (!skip && zone->uz_dtor) 17828355f576SJeff Roberson zone->uz_dtor(item, zone->uz_size, udata); 1783605cbd6aSJeff Roberson 1784af7f9b97SJeff Roberson if (zone->uz_flags & UMA_ZFLAG_FULL) { 1785af7f9b97SJeff Roberson if (zone->uz_pages < zone->uz_maxpages) 1786af7f9b97SJeff Roberson zone->uz_flags &= ~UMA_ZFLAG_FULL; 1787af7f9b97SJeff Roberson 1788af7f9b97SJeff Roberson /* We can handle one more allocation */ 1789af7f9b97SJeff Roberson wakeup_one(&zone); 1790af7f9b97SJeff Roberson } 1791af7f9b97SJeff Roberson 1792605cbd6aSJeff Roberson ZONE_UNLOCK(zone); 17938355f576SJeff Roberson } 17948355f576SJeff Roberson 17958355f576SJeff Roberson /* See uma.h */ 17968355f576SJeff Roberson void 1797736ee590SJeff Roberson uma_zone_set_max(uma_zone_t zone, int nitems) 1798736ee590SJeff Roberson { 1799736ee590SJeff Roberson ZONE_LOCK(zone); 1800736ee590SJeff Roberson if (zone->uz_ppera > 1) 1801af7f9b97SJeff Roberson zone->uz_maxpages = nitems * zone->uz_ppera; 1802736ee590SJeff Roberson else 1803736ee590SJeff Roberson zone->uz_maxpages = nitems / zone->uz_ipers; 1804736ee590SJeff Roberson ZONE_UNLOCK(zone); 1805736ee590SJeff Roberson } 1806736ee590SJeff Roberson 1807736ee590SJeff Roberson /* See uma.h */ 1808736ee590SJeff Roberson void 18098355f576SJeff Roberson uma_zone_set_freef(uma_zone_t zone, uma_free freef) 18108355f576SJeff Roberson { 18118355f576SJeff Roberson ZONE_LOCK(zone); 18128355f576SJeff Roberson 18138355f576SJeff Roberson zone->uz_freef = freef; 18148355f576SJeff Roberson 18158355f576SJeff Roberson ZONE_UNLOCK(zone); 18168355f576SJeff Roberson } 18178355f576SJeff Roberson 18188355f576SJeff Roberson /* See uma.h */ 18198355f576SJeff Roberson void 18208355f576SJeff Roberson uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf) 18218355f576SJeff Roberson { 18228355f576SJeff Roberson ZONE_LOCK(zone); 18238355f576SJeff Roberson 18248355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_PRIVALLOC; 18258355f576SJeff Roberson zone->uz_allocf = allocf; 18268355f576SJeff Roberson 18278355f576SJeff Roberson ZONE_UNLOCK(zone); 18288355f576SJeff Roberson } 18298355f576SJeff Roberson 18308355f576SJeff Roberson /* See uma.h */ 18318355f576SJeff Roberson int 18328355f576SJeff Roberson uma_zone_set_obj(uma_zone_t zone, struct vm_object *obj, int count) 18338355f576SJeff Roberson { 18348355f576SJeff Roberson int pages; 18358355f576SJeff Roberson vm_offset_t kva; 18368355f576SJeff Roberson 18378355f576SJeff Roberson mtx_lock(&Giant); 18388355f576SJeff Roberson 18398355f576SJeff Roberson pages = count / zone->uz_ipers; 18408355f576SJeff Roberson 18418355f576SJeff Roberson if (pages * zone->uz_ipers < count) 18428355f576SJeff Roberson pages++; 1843a553d4b8SJeff Roberson 18448355f576SJeff Roberson kva = kmem_alloc_pageable(kernel_map, pages * UMA_SLAB_SIZE); 18458355f576SJeff Roberson 1846a553d4b8SJeff Roberson if (kva == 0) { 1847a553d4b8SJeff Roberson mtx_unlock(&Giant); 18488355f576SJeff Roberson return (0); 18498355f576SJeff Roberson } 18508355f576SJeff Roberson 18518355f576SJeff Roberson 1852a553d4b8SJeff Roberson if (obj == NULL) 1853a553d4b8SJeff Roberson obj = vm_object_allocate(OBJT_DEFAULT, 18548355f576SJeff Roberson zone->uz_maxpages); 18558355f576SJeff Roberson else 18568355f576SJeff Roberson _vm_object_allocate(OBJT_DEFAULT, 1857a553d4b8SJeff Roberson zone->uz_maxpages, obj); 1858a553d4b8SJeff Roberson 1859a553d4b8SJeff Roberson ZONE_LOCK(zone); 1860a553d4b8SJeff Roberson zone->uz_kva = kva; 1861a553d4b8SJeff Roberson zone->uz_obj = obj; 1862a553d4b8SJeff Roberson zone->uz_maxpages = pages; 18638355f576SJeff Roberson 18648355f576SJeff Roberson zone->uz_allocf = obj_alloc; 18658355f576SJeff Roberson zone->uz_flags |= UMA_ZFLAG_NOFREE | UMA_ZFLAG_PRIVALLOC; 18668355f576SJeff Roberson 18678355f576SJeff Roberson ZONE_UNLOCK(zone); 1868a553d4b8SJeff Roberson mtx_unlock(&Giant); 18698355f576SJeff Roberson 18708355f576SJeff Roberson return (1); 18718355f576SJeff Roberson } 18728355f576SJeff Roberson 18738355f576SJeff Roberson /* See uma.h */ 18748355f576SJeff Roberson void 18758355f576SJeff Roberson uma_prealloc(uma_zone_t zone, int items) 18768355f576SJeff Roberson { 18778355f576SJeff Roberson int slabs; 18788355f576SJeff Roberson uma_slab_t slab; 18798355f576SJeff Roberson 18808355f576SJeff Roberson ZONE_LOCK(zone); 18818355f576SJeff Roberson slabs = items / zone->uz_ipers; 18828355f576SJeff Roberson if (slabs * zone->uz_ipers < items) 18838355f576SJeff Roberson slabs++; 18848355f576SJeff Roberson 18858355f576SJeff Roberson while (slabs > 0) { 18868355f576SJeff Roberson slab = slab_zalloc(zone, M_WAITOK); 18878355f576SJeff Roberson LIST_INSERT_HEAD(&zone->uz_free_slab, slab, us_link); 18888355f576SJeff Roberson slabs--; 18898355f576SJeff Roberson } 18908355f576SJeff Roberson ZONE_UNLOCK(zone); 18918355f576SJeff Roberson } 18928355f576SJeff Roberson 18938355f576SJeff Roberson /* See uma.h */ 18948355f576SJeff Roberson void 18958355f576SJeff Roberson uma_reclaim(void) 18968355f576SJeff Roberson { 18978355f576SJeff Roberson /* 18988355f576SJeff Roberson * You might think that the delay below would improve performance since 18998355f576SJeff Roberson * the allocator will give away memory that it may ask for immediately. 19008355f576SJeff Roberson * Really, it makes things worse, since cpu cycles are so much cheaper 19018355f576SJeff Roberson * than disk activity. 19028355f576SJeff Roberson */ 19038355f576SJeff Roberson #if 0 19048355f576SJeff Roberson static struct timeval tv = {0}; 19058355f576SJeff Roberson struct timeval now; 19068355f576SJeff Roberson getmicrouptime(&now); 19078355f576SJeff Roberson if (now.tv_sec > tv.tv_sec + 30) 19088355f576SJeff Roberson tv = now; 19098355f576SJeff Roberson else 19108355f576SJeff Roberson return; 19118355f576SJeff Roberson #endif 19128355f576SJeff Roberson #ifdef UMA_DEBUG 19138355f576SJeff Roberson printf("UMA: vm asked us to release pages!\n"); 19148355f576SJeff Roberson #endif 191586bbae32SJeff Roberson bucket_enable(); 19168355f576SJeff Roberson zone_foreach(zone_drain); 19178355f576SJeff Roberson 19188355f576SJeff Roberson /* 19198355f576SJeff Roberson * Some slabs may have been freed but this zone will be visited early 19208355f576SJeff Roberson * we visit again so that we can free pages that are empty once other 19218355f576SJeff Roberson * zones are drained. We have to do the same for buckets. 19228355f576SJeff Roberson */ 19238355f576SJeff Roberson zone_drain(slabzone); 19248355f576SJeff Roberson zone_drain(bucketzone); 19258355f576SJeff Roberson } 19268355f576SJeff Roberson 19278355f576SJeff Roberson void * 19288355f576SJeff Roberson uma_large_malloc(int size, int wait) 19298355f576SJeff Roberson { 19308355f576SJeff Roberson void *mem; 19318355f576SJeff Roberson uma_slab_t slab; 19328355f576SJeff Roberson u_int8_t flags; 19338355f576SJeff Roberson 1934a553d4b8SJeff Roberson slab = uma_zalloc_internal(slabzone, NULL, wait, NULL); 19358355f576SJeff Roberson if (slab == NULL) 19368355f576SJeff Roberson return (NULL); 19378355f576SJeff Roberson 19388355f576SJeff Roberson mem = page_alloc(NULL, size, &flags, wait); 19398355f576SJeff Roberson if (mem) { 19408355f576SJeff Roberson slab->us_data = mem; 19418355f576SJeff Roberson slab->us_flags = flags | UMA_SLAB_MALLOC; 19428355f576SJeff Roberson slab->us_size = size; 19438355f576SJeff Roberson UMA_HASH_INSERT(mallochash, slab, mem); 19448355f576SJeff Roberson } else { 19458355f576SJeff Roberson uma_zfree_internal(slabzone, slab, NULL, 0); 19468355f576SJeff Roberson } 19478355f576SJeff Roberson 19488355f576SJeff Roberson 19498355f576SJeff Roberson return (mem); 19508355f576SJeff Roberson } 19518355f576SJeff Roberson 19528355f576SJeff Roberson void 19538355f576SJeff Roberson uma_large_free(uma_slab_t slab) 19548355f576SJeff Roberson { 19558355f576SJeff Roberson UMA_HASH_REMOVE(mallochash, slab, slab->us_data); 19568355f576SJeff Roberson page_free(slab->us_data, slab->us_size, slab->us_flags); 19578355f576SJeff Roberson uma_zfree_internal(slabzone, slab, NULL, 0); 19588355f576SJeff Roberson } 19598355f576SJeff Roberson 19608355f576SJeff Roberson void 19618355f576SJeff Roberson uma_print_stats(void) 19628355f576SJeff Roberson { 19638355f576SJeff Roberson zone_foreach(uma_print_zone); 19648355f576SJeff Roberson } 19658355f576SJeff Roberson 19668355f576SJeff Roberson void 19678355f576SJeff Roberson uma_print_zone(uma_zone_t zone) 19688355f576SJeff Roberson { 19698355f576SJeff Roberson printf("%s(%p) size %d(%d) flags %d ipers %d ppera %d out %d free %d\n", 19708355f576SJeff Roberson zone->uz_name, zone, zone->uz_size, zone->uz_rsize, zone->uz_flags, 19718355f576SJeff Roberson zone->uz_ipers, zone->uz_ppera, 19728355f576SJeff Roberson (zone->uz_ipers * zone->uz_pages) - zone->uz_free, zone->uz_free); 19738355f576SJeff Roberson } 19748355f576SJeff Roberson 19758355f576SJeff Roberson /* 19768355f576SJeff Roberson * Sysctl handler for vm.zone 19778355f576SJeff Roberson * 19788355f576SJeff Roberson * stolen from vm_zone.c 19798355f576SJeff Roberson */ 19808355f576SJeff Roberson static int 19818355f576SJeff Roberson sysctl_vm_zone(SYSCTL_HANDLER_ARGS) 19828355f576SJeff Roberson { 19838355f576SJeff Roberson int error, len, cnt; 19848355f576SJeff Roberson const int linesize = 128; /* conservative */ 19858355f576SJeff Roberson int totalfree; 19868355f576SJeff Roberson char *tmpbuf, *offset; 19878355f576SJeff Roberson uma_zone_t z; 19888355f576SJeff Roberson char *p; 19898355f576SJeff Roberson 19908355f576SJeff Roberson cnt = 0; 19910da47b2fSJeff Roberson mtx_lock(&uma_mtx); 19928355f576SJeff Roberson LIST_FOREACH(z, &uma_zones, uz_link) 19938355f576SJeff Roberson cnt++; 19940da47b2fSJeff Roberson mtx_unlock(&uma_mtx); 19958355f576SJeff Roberson MALLOC(tmpbuf, char *, (cnt == 0 ? 1 : cnt) * linesize, 19968355f576SJeff Roberson M_TEMP, M_WAITOK); 19978355f576SJeff Roberson len = snprintf(tmpbuf, linesize, 19988355f576SJeff Roberson "\nITEM SIZE LIMIT USED FREE REQUESTS\n\n"); 19998355f576SJeff Roberson if (cnt == 0) 20008355f576SJeff Roberson tmpbuf[len - 1] = '\0'; 20018355f576SJeff Roberson error = SYSCTL_OUT(req, tmpbuf, cnt == 0 ? len-1 : len); 20028355f576SJeff Roberson if (error || cnt == 0) 20038355f576SJeff Roberson goto out; 20048355f576SJeff Roberson offset = tmpbuf; 2005f4af24d5SJeff Roberson mtx_lock(&uma_mtx); 20068355f576SJeff Roberson LIST_FOREACH(z, &uma_zones, uz_link) { 20078355f576SJeff Roberson if (cnt == 0) /* list may have changed size */ 20088355f576SJeff Roberson break; 20098355f576SJeff Roberson ZONE_LOCK(z); 20108355f576SJeff Roberson totalfree = z->uz_free + z->uz_cachefree; 20118355f576SJeff Roberson len = snprintf(offset, linesize, 20128355f576SJeff Roberson "%-12.12s %6.6u, %8.8u, %6.6u, %6.6u, %8.8llu\n", 20138355f576SJeff Roberson z->uz_name, z->uz_size, 20148355f576SJeff Roberson z->uz_maxpages * z->uz_ipers, 20158355f576SJeff Roberson (z->uz_ipers * (z->uz_pages / z->uz_ppera)) - totalfree, 20168355f576SJeff Roberson totalfree, 20178355f576SJeff Roberson (unsigned long long)z->uz_allocs); 20188355f576SJeff Roberson ZONE_UNLOCK(z); 20198355f576SJeff Roberson for (p = offset + 12; p > offset && *p == ' '; --p) 20208355f576SJeff Roberson /* nothing */ ; 20218355f576SJeff Roberson p[1] = ':'; 20228355f576SJeff Roberson cnt--; 20238355f576SJeff Roberson offset += len; 20248355f576SJeff Roberson } 2025f4af24d5SJeff Roberson mtx_unlock(&uma_mtx); 20268355f576SJeff Roberson *offset++ = '\0'; 20278355f576SJeff Roberson error = SYSCTL_OUT(req, tmpbuf, offset - tmpbuf); 20288355f576SJeff Roberson out: 20298355f576SJeff Roberson FREE(tmpbuf, M_TEMP); 20308355f576SJeff Roberson return (error); 20318355f576SJeff Roberson } 2032