vm_kern.c revision 22521
15455Sdg/* 21541Srgrimes * Copyright (c) 1991, 1993 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * 51541Srgrimes * This code is derived from software contributed to Berkeley by 61541Srgrimes * The Mach Operating System project at Carnegie-Mellon University. 71541Srgrimes * 81541Srgrimes * Redistribution and use in source and binary forms, with or without 91541Srgrimes * modification, are permitted provided that the following conditions 101541Srgrimes * are met: 111541Srgrimes * 1. Redistributions of source code must retain the above copyright 121541Srgrimes * notice, this list of conditions and the following disclaimer. 131541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 141541Srgrimes * notice, this list of conditions and the following disclaimer in the 151541Srgrimes * documentation and/or other materials provided with the distribution. 161541Srgrimes * 3. All advertising materials mentioning features or use of this software 171541Srgrimes * must display the following acknowledgement: 181541Srgrimes * This product includes software developed by the University of 191541Srgrimes * California, Berkeley and its contributors. 201541Srgrimes * 4. Neither the name of the University nor the names of its contributors 211541Srgrimes * may be used to endorse or promote products derived from this software 221541Srgrimes * without specific prior written permission. 231541Srgrimes * 241541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 251541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 261541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 271541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 281541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 291541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 301541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 311541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 321541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 331541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 341541Srgrimes * SUCH DAMAGE. 351541Srgrimes * 361817Sdg * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 371541Srgrimes * 381541Srgrimes * 391541Srgrimes * Copyright (c) 1987, 1990 Carnegie-Mellon University. 401541Srgrimes * All rights reserved. 411541Srgrimes * 421541Srgrimes * Authors: Avadis Tevanian, Jr., Michael Wayne Young 435455Sdg * 441541Srgrimes * Permission to use, copy, modify and distribute this software and 451541Srgrimes * its documentation is hereby granted, provided that both the copyright 461541Srgrimes * notice and this permission notice appear in all copies of the 471541Srgrimes * software, derivative works or modified versions, and any portions 481541Srgrimes * thereof, and that both notices appear in supporting documentation. 495455Sdg * 505455Sdg * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 515455Sdg * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 521541Srgrimes * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 535455Sdg * 541541Srgrimes * Carnegie Mellon requests users of this software to return to 551541Srgrimes * 561541Srgrimes * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 571541Srgrimes * School of Computer Science 581541Srgrimes * Carnegie Mellon University 591541Srgrimes * Pittsburgh PA 15213-3890 601541Srgrimes * 611541Srgrimes * any improvements or extensions that they make and grant Carnegie the 621541Srgrimes * rights to redistribute these changes. 631817Sdg * 6421673Sjkh * $FreeBSD: head/sys/vm/vm_kern.c 22521 1997-02-10 02:22:35Z dyson $ 651541Srgrimes */ 661541Srgrimes 671541Srgrimes/* 681541Srgrimes * Kernel memory management. 691541Srgrimes */ 701541Srgrimes 711541Srgrimes#include <sys/param.h> 721541Srgrimes#include <sys/systm.h> 732112Swollman#include <sys/kernel.h> 742112Swollman#include <sys/proc.h> 756129Sdg#include <sys/malloc.h> 767066Sdg#include <sys/syslog.h> 7712662Sdg#include <sys/queue.h> 7812662Sdg#include <sys/vmmeter.h> 791541Srgrimes 801541Srgrimes#include <vm/vm.h> 8112662Sdg#include <vm/vm_param.h> 8212662Sdg#include <vm/vm_prot.h> 8322521Sdyson#include <sys/lock.h> 8412662Sdg#include <vm/pmap.h> 8512662Sdg#include <vm/vm_map.h> 8612662Sdg#include <vm/vm_object.h> 871541Srgrimes#include <vm/vm_page.h> 881541Srgrimes#include <vm/vm_pageout.h> 891541Srgrimes#include <vm/vm_kern.h> 9012726Sbde#include <vm/vm_extern.h> 911541Srgrimes 9219830Sdysonvm_map_t kernel_map=0; 9319830Sdysonvm_map_t kmem_map=0; 9419830Sdysonvm_map_t exec_map=0; 9519830Sdysonvm_map_t exech_map=0; 9619830Sdysonvm_map_t clean_map=0; 9719830Sdysonvm_map_t u_map=0; 9819830Sdysonvm_map_t buffer_map=0; 9919830Sdysonvm_map_t mb_map=0; 10019830Sdysonint mb_map_full=0; 10119830Sdysonvm_map_t io_map=0; 10219830Sdysonvm_map_t phys_map=0; 1032112Swollman 1041541Srgrimes/* 1051541Srgrimes * kmem_alloc_pageable: 1061541Srgrimes * 1071541Srgrimes * Allocate pageable memory to the kernel's address map. 10812259Sdg * "map" must be kernel_map or a submap of kernel_map. 1091541Srgrimes */ 1101541Srgrimes 1118876Srgrimesvm_offset_t 1125455Sdgkmem_alloc_pageable(map, size) 1135455Sdg vm_map_t map; 1145455Sdg register vm_size_t size; 1151541Srgrimes{ 1165455Sdg vm_offset_t addr; 1175455Sdg register int result; 1181541Srgrimes 1191541Srgrimes size = round_page(size); 1201541Srgrimes addr = vm_map_min(map); 1211541Srgrimes result = vm_map_find(map, NULL, (vm_offset_t) 0, 12213490Sdyson &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 1231541Srgrimes if (result != KERN_SUCCESS) { 1245455Sdg return (0); 1251541Srgrimes } 1265455Sdg return (addr); 1271541Srgrimes} 1281541Srgrimes 1291541Srgrimes/* 1301541Srgrimes * Allocate wired-down memory in the kernel's address map 1311541Srgrimes * or a submap. 1321541Srgrimes */ 1338876Srgrimesvm_offset_t 1345455Sdgkmem_alloc(map, size) 1355455Sdg register vm_map_t map; 1365455Sdg register vm_size_t size; 1371541Srgrimes{ 1385455Sdg vm_offset_t addr; 1395455Sdg register vm_offset_t offset; 1405455Sdg vm_offset_t i; 1411541Srgrimes 1421541Srgrimes size = round_page(size); 1431541Srgrimes 1441541Srgrimes /* 1455455Sdg * Use the kernel object for wired-down kernel pages. Assume that no 1465455Sdg * region of the kernel object is referenced more than once. 1471541Srgrimes */ 1481541Srgrimes 1491541Srgrimes /* 1505455Sdg * Locate sufficient space in the map. This will give us the final 1515455Sdg * virtual address for the new memory, and thus will tell us the 1525455Sdg * offset within the kernel map. 1531541Srgrimes */ 1541541Srgrimes vm_map_lock(map); 1551541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 1561541Srgrimes vm_map_unlock(map); 1571541Srgrimes return (0); 1581541Srgrimes } 1591541Srgrimes offset = addr - VM_MIN_KERNEL_ADDRESS; 1601541Srgrimes vm_object_reference(kernel_object); 16113490Sdyson vm_map_insert(map, kernel_object, offset, addr, addr + size, 16213490Sdyson VM_PROT_ALL, VM_PROT_ALL, 0); 1631541Srgrimes vm_map_unlock(map); 1641541Srgrimes 1651541Srgrimes /* 1665455Sdg * Guarantee that there are pages already in this object before 1675455Sdg * calling vm_map_pageable. This is to prevent the following 1685455Sdg * scenario: 1698876Srgrimes * 1705455Sdg * 1) Threads have swapped out, so that there is a pager for the 1715455Sdg * kernel_object. 2) The kmsg zone is empty, and so we are 1725455Sdg * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 1735455Sdg * there is no page, but there is a pager, so we call 1745455Sdg * pager_data_request. But the kmsg zone is empty, so we must 1755455Sdg * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 1765455Sdg * we get the data back from the pager, it will be (very stale) 1775455Sdg * non-zero data. kmem_alloc is defined to return zero-filled memory. 1788876Srgrimes * 1795455Sdg * We're intentionally not activating the pages we allocate to prevent a 1805455Sdg * race with page-out. vm_map_pageable will wire the pages. 1811541Srgrimes */ 1821541Srgrimes 1835455Sdg for (i = 0; i < size; i += PAGE_SIZE) { 1845455Sdg vm_page_t mem; 1851541Srgrimes 18612767Sdyson while ((mem = vm_page_alloc(kernel_object, 18713490Sdyson OFF_TO_IDX(offset + i), VM_ALLOC_ZERO)) == NULL) { 1881541Srgrimes VM_WAIT; 1891541Srgrimes } 19010548Sdyson if ((mem->flags & PG_ZERO) == 0) 19110548Sdyson vm_page_zero_fill(mem); 19210548Sdyson mem->flags &= ~(PG_BUSY|PG_ZERO); 1936585Sdg mem->valid = VM_PAGE_BITS_ALL; 1941541Srgrimes } 1955455Sdg 1961541Srgrimes /* 1975455Sdg * And finally, mark the data as non-pageable. 1981541Srgrimes */ 1991541Srgrimes 2001541Srgrimes (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 2011541Srgrimes 2025455Sdg return (addr); 2031541Srgrimes} 2041541Srgrimes 2051541Srgrimes/* 2061541Srgrimes * kmem_free: 2071541Srgrimes * 2081541Srgrimes * Release a region of kernel virtual memory allocated 2091541Srgrimes * with kmem_alloc, and return the physical pages 2101541Srgrimes * associated with that region. 2111541Srgrimes */ 2128876Srgrimesvoid 2135455Sdgkmem_free(map, addr, size) 2145455Sdg vm_map_t map; 2155455Sdg register vm_offset_t addr; 2165455Sdg vm_size_t size; 2171541Srgrimes{ 2181541Srgrimes (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 2191541Srgrimes} 2201541Srgrimes 2211541Srgrimes/* 2221541Srgrimes * kmem_suballoc: 2231541Srgrimes * 2241541Srgrimes * Allocates a map to manage a subrange 2251541Srgrimes * of the kernel virtual address space. 2261541Srgrimes * 2271541Srgrimes * Arguments are as follows: 2281541Srgrimes * 2291541Srgrimes * parent Map to take range from 2301541Srgrimes * size Size of range to find 2311541Srgrimes * min, max Returned endpoints of map 2321541Srgrimes * pageable Can the region be paged 2331541Srgrimes */ 2348876Srgrimesvm_map_t 2355455Sdgkmem_suballoc(parent, min, max, size, pageable) 2365455Sdg register vm_map_t parent; 2375455Sdg vm_offset_t *min, *max; 2385455Sdg register vm_size_t size; 2395455Sdg boolean_t pageable; 2401541Srgrimes{ 2415455Sdg register int ret; 2425455Sdg vm_map_t result; 2431541Srgrimes 2441541Srgrimes size = round_page(size); 2451541Srgrimes 2461541Srgrimes *min = (vm_offset_t) vm_map_min(parent); 2471541Srgrimes ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 24813490Sdyson min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 2491541Srgrimes if (ret != KERN_SUCCESS) { 2501541Srgrimes printf("kmem_suballoc: bad status return of %d.\n", ret); 2511541Srgrimes panic("kmem_suballoc"); 2521541Srgrimes } 2531541Srgrimes *max = *min + size; 2541541Srgrimes pmap_reference(vm_map_pmap(parent)); 2551541Srgrimes result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 2561541Srgrimes if (result == NULL) 2571541Srgrimes panic("kmem_suballoc: cannot create submap"); 2581541Srgrimes if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 2591541Srgrimes panic("kmem_suballoc: unable to change range to submap"); 2605455Sdg return (result); 2611541Srgrimes} 2621541Srgrimes 2631541Srgrimes/* 2641541Srgrimes * Allocate wired-down memory in the kernel's address map for the higher 2651541Srgrimes * level kernel memory allocator (kern/kern_malloc.c). We cannot use 2661541Srgrimes * kmem_alloc() because we may need to allocate memory at interrupt 2671541Srgrimes * level where we cannot block (canwait == FALSE). 2681541Srgrimes * 2691541Srgrimes * This routine has its own private kernel submap (kmem_map) and object 2701541Srgrimes * (kmem_object). This, combined with the fact that only malloc uses 2711541Srgrimes * this routine, ensures that we will never block in map or object waits. 2721541Srgrimes * 2731541Srgrimes * Note that this still only works in a uni-processor environment and 2741541Srgrimes * when called at splhigh(). 2751541Srgrimes * 2761541Srgrimes * We don't worry about expanding the map (adding entries) since entries 2771541Srgrimes * for wired maps are statically allocated. 2781541Srgrimes */ 2791541Srgrimesvm_offset_t 2806129Sdgkmem_malloc(map, size, waitflag) 2815455Sdg register vm_map_t map; 2825455Sdg register vm_size_t size; 2836129Sdg boolean_t waitflag; 2841541Srgrimes{ 2855455Sdg register vm_offset_t offset, i; 2865455Sdg vm_map_entry_t entry; 2875455Sdg vm_offset_t addr; 2885455Sdg vm_page_t m; 2891541Srgrimes 29021737Sdg if (map != kmem_map && map != mb_map) 29121737Sdg panic("kmem_malloc: map != {kmem,mb}_map"); 2921541Srgrimes 2931541Srgrimes size = round_page(size); 2941541Srgrimes addr = vm_map_min(map); 2951541Srgrimes 2961541Srgrimes /* 2975455Sdg * Locate sufficient space in the map. This will give us the final 2985455Sdg * virtual address for the new memory, and thus will tell us the 2995455Sdg * offset within the kernel map. 3001541Srgrimes */ 3011541Srgrimes vm_map_lock(map); 3021541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 3031541Srgrimes vm_map_unlock(map); 3047066Sdg if (map == mb_map) { 3057066Sdg mb_map_full = TRUE; 30621737Sdg log(LOG_ERR, "Out of mbuf clusters - increase maxusers!\n"); 3077066Sdg return (0); 3087066Sdg } 3096129Sdg if (waitflag == M_WAITOK) 3107066Sdg panic("kmem_malloc: kmem_map too small"); 3111541Srgrimes return (0); 3121541Srgrimes } 31315367Sdyson offset = addr - VM_MIN_KERNEL_ADDRESS; 3141541Srgrimes vm_object_reference(kmem_object); 31513490Sdyson vm_map_insert(map, kmem_object, offset, addr, addr + size, 31613490Sdyson VM_PROT_ALL, VM_PROT_ALL, 0); 3171541Srgrimes 3181541Srgrimes for (i = 0; i < size; i += PAGE_SIZE) { 31915809Sdysonretry: 32012767Sdyson m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), 3216129Sdg (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 3221541Srgrimes 3231541Srgrimes /* 3245455Sdg * Ran out of space, free everything up and return. Don't need 3255455Sdg * to lock page queues here as we know that the pages we got 3265455Sdg * aren't on any queues. 3271541Srgrimes */ 3281541Srgrimes if (m == NULL) { 32915809Sdyson if (waitflag == M_WAITOK) { 33015809Sdyson VM_WAIT; 33115809Sdyson goto retry; 33215809Sdyson } 3331541Srgrimes while (i != 0) { 3341541Srgrimes i -= PAGE_SIZE; 33512767Sdyson m = vm_page_lookup(kmem_object, 33612767Sdyson OFF_TO_IDX(offset + i)); 33715876Sdyson PAGE_WAKEUP(m); 3381541Srgrimes vm_page_free(m); 3391541Srgrimes } 3401541Srgrimes vm_map_delete(map, addr, addr + size); 3411541Srgrimes vm_map_unlock(map); 3425455Sdg return (0); 3431541Srgrimes } 34415809Sdyson m->flags &= ~PG_ZERO; 3456585Sdg m->valid = VM_PAGE_BITS_ALL; 3461541Srgrimes } 3471541Srgrimes 3481541Srgrimes /* 3495455Sdg * Mark map entry as non-pageable. Assert: vm_map_insert() will never 3505455Sdg * be able to extend the previous entry so there will be a new entry 3515455Sdg * exactly corresponding to this address range and it will have 3525455Sdg * wired_count == 0. 3531541Srgrimes */ 3541541Srgrimes if (!vm_map_lookup_entry(map, addr, &entry) || 3551541Srgrimes entry->start != addr || entry->end != addr + size || 3561541Srgrimes entry->wired_count) 3571541Srgrimes panic("kmem_malloc: entry not found or misaligned"); 3581541Srgrimes entry->wired_count++; 3591541Srgrimes 36020993Sdyson vm_map_simplify_entry(map, entry); 36120993Sdyson 3621541Srgrimes /* 3635455Sdg * Loop thru pages, entering them in the pmap. (We cannot add them to 3645455Sdg * the wired count without wrapping the vm_page_queue_lock in 3655455Sdg * splimp...) 3661541Srgrimes */ 3671541Srgrimes for (i = 0; i < size; i += PAGE_SIZE) { 36812767Sdyson m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 36913490Sdyson vm_page_wire(m); 37015809Sdyson PAGE_WAKEUP(m); 37115809Sdyson pmap_enter(kernel_pmap, addr + i, VM_PAGE_TO_PHYS(m), 37215809Sdyson VM_PROT_ALL, 1); 37316892Sdyson m->flags |= PG_MAPPED|PG_WRITEABLE; 3741541Srgrimes } 3751541Srgrimes vm_map_unlock(map); 3761541Srgrimes 3775455Sdg return (addr); 3781541Srgrimes} 3791541Srgrimes 3801541Srgrimes/* 3811541Srgrimes * kmem_alloc_wait 3821541Srgrimes * 3831541Srgrimes * Allocates pageable memory from a sub-map of the kernel. If the submap 3841541Srgrimes * has no room, the caller sleeps waiting for more memory in the submap. 3851541Srgrimes * 3861541Srgrimes */ 3878876Srgrimesvm_offset_t 3885455Sdgkmem_alloc_wait(map, size) 3895455Sdg vm_map_t map; 3905455Sdg vm_size_t size; 3911541Srgrimes{ 3925455Sdg vm_offset_t addr; 3931541Srgrimes 3941541Srgrimes size = round_page(size); 3951541Srgrimes 3961541Srgrimes for (;;) { 3971541Srgrimes /* 3985455Sdg * To make this work for more than one map, use the map's lock 3995455Sdg * to lock out sleepers/wakers. 4001541Srgrimes */ 4011541Srgrimes vm_map_lock(map); 4021541Srgrimes if (vm_map_findspace(map, 0, size, &addr) == 0) 4031541Srgrimes break; 4041541Srgrimes /* no space now; see if we can ever get space */ 4051541Srgrimes if (vm_map_max(map) - vm_map_min(map) < size) { 4061541Srgrimes vm_map_unlock(map); 4071541Srgrimes return (0); 4081541Srgrimes } 4091541Srgrimes vm_map_unlock(map); 4109507Sdg tsleep(map, PVM, "kmaw", 0); 4111541Srgrimes } 41213490Sdyson vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 4131541Srgrimes vm_map_unlock(map); 4141541Srgrimes return (addr); 4151541Srgrimes} 4161541Srgrimes 4171541Srgrimes/* 4181541Srgrimes * kmem_free_wakeup 4191541Srgrimes * 4209507Sdg * Returns memory to a submap of the kernel, and wakes up any processes 4211541Srgrimes * waiting for memory in that map. 4221541Srgrimes */ 4238876Srgrimesvoid 4245455Sdgkmem_free_wakeup(map, addr, size) 4255455Sdg vm_map_t map; 4265455Sdg vm_offset_t addr; 4275455Sdg vm_size_t size; 4281541Srgrimes{ 4291541Srgrimes vm_map_lock(map); 4301541Srgrimes (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 4319507Sdg wakeup(map); 4321541Srgrimes vm_map_unlock(map); 4331541Srgrimes} 4341541Srgrimes 4351541Srgrimes/* 4361541Srgrimes * Create the kernel map; insert a mapping covering kernel text, data, bss, 4371541Srgrimes * and all space allocated thus far (`boostrap' data). The new map will thus 4381541Srgrimes * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 4391541Srgrimes * the range between `start' and `end' as free. 4401541Srgrimes */ 4418876Srgrimesvoid 4425455Sdgkmem_init(start, end) 4431541Srgrimes vm_offset_t start, end; 4441541Srgrimes{ 4451541Srgrimes register vm_map_t m; 4461541Srgrimes 4471541Srgrimes m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 4481541Srgrimes vm_map_lock(m); 4491541Srgrimes /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 4501541Srgrimes kernel_map = m; 4515455Sdg (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 45213490Sdyson VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 4531541Srgrimes /* ... and ending with the completion of the above `insert' */ 4541541Srgrimes vm_map_unlock(m); 4551541Srgrimes} 456