vm_kern.c revision 2112
11541Srgrimes/* 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 431541Srgrimes * 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. 491541Srgrimes * 501541Srgrimes * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 511541Srgrimes * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 521541Srgrimes * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 531541Srgrimes * 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 * 642112Swollman * $Id: vm_kern.c,v 1.6 1994/08/07 14:53:26 davidg Exp $ 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> 751541Srgrimes 761541Srgrimes#include <vm/vm.h> 771541Srgrimes#include <vm/vm_page.h> 781541Srgrimes#include <vm/vm_pageout.h> 791541Srgrimes#include <vm/vm_kern.h> 801541Srgrimes 812112Swollmanvm_map_t buffer_map; 822112Swollmanvm_map_t kernel_map; 832112Swollmanvm_map_t kmem_map; 842112Swollmanvm_map_t mb_map; 852112Swollmanvm_map_t io_map; 862112Swollmanvm_map_t clean_map; 872112Swollmanvm_map_t pager_map; 882112Swollmanvm_map_t phys_map; 892112Swollman 901541Srgrimes/* 911541Srgrimes * kmem_alloc_pageable: 921541Srgrimes * 931541Srgrimes * Allocate pageable memory to the kernel's address map. 941541Srgrimes * map must be "kernel_map" below. 951541Srgrimes */ 961541Srgrimes 971541Srgrimesvm_offset_t kmem_alloc_pageable(map, size) 981541Srgrimes vm_map_t map; 991541Srgrimes register vm_size_t size; 1001541Srgrimes{ 1011541Srgrimes vm_offset_t addr; 1021541Srgrimes register int result; 1031541Srgrimes 1041541Srgrimes#if 0 1051541Srgrimes if (map != kernel_map) 1061541Srgrimes panic("kmem_alloc_pageable: not called with kernel_map"); 1071541Srgrimes#endif 1081541Srgrimes 1091541Srgrimes size = round_page(size); 1101541Srgrimes 1111541Srgrimes addr = vm_map_min(map); 1121541Srgrimes result = vm_map_find(map, NULL, (vm_offset_t) 0, 1131541Srgrimes &addr, size, TRUE); 1141541Srgrimes if (result != KERN_SUCCESS) { 1151541Srgrimes return(0); 1161541Srgrimes } 1171541Srgrimes 1181541Srgrimes return(addr); 1191541Srgrimes} 1201541Srgrimes 1211541Srgrimes/* 1221541Srgrimes * Allocate wired-down memory in the kernel's address map 1231541Srgrimes * or a submap. 1241541Srgrimes */ 1251541Srgrimesvm_offset_t kmem_alloc(map, size) 1261541Srgrimes register vm_map_t map; 1271541Srgrimes register vm_size_t size; 1281541Srgrimes{ 1291541Srgrimes vm_offset_t addr; 1301541Srgrimes register vm_offset_t offset; 1311541Srgrimes vm_offset_t i; 1321541Srgrimes 1331541Srgrimes size = round_page(size); 1341541Srgrimes 1351541Srgrimes /* 1361541Srgrimes * Use the kernel object for wired-down kernel pages. 1371541Srgrimes * Assume that no region of the kernel object is 1381541Srgrimes * referenced more than once. 1391541Srgrimes */ 1401541Srgrimes 1411541Srgrimes /* 1421541Srgrimes * Locate sufficient space in the map. This will give us the 1431541Srgrimes * final virtual address for the new memory, and thus will tell 1441541Srgrimes * us the offset within the kernel map. 1451541Srgrimes */ 1461541Srgrimes vm_map_lock(map); 1471541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 1481541Srgrimes vm_map_unlock(map); 1491541Srgrimes return (0); 1501541Srgrimes } 1511541Srgrimes offset = addr - VM_MIN_KERNEL_ADDRESS; 1521541Srgrimes vm_object_reference(kernel_object); 1531541Srgrimes vm_map_insert(map, kernel_object, offset, addr, addr + size); 1541541Srgrimes vm_map_unlock(map); 1551541Srgrimes 1561541Srgrimes /* 1571541Srgrimes * Guarantee that there are pages already in this object 1581541Srgrimes * before calling vm_map_pageable. This is to prevent the 1591541Srgrimes * following scenario: 1601541Srgrimes * 1611541Srgrimes * 1) Threads have swapped out, so that there is a 1621541Srgrimes * pager for the kernel_object. 1631541Srgrimes * 2) The kmsg zone is empty, and so we are kmem_allocing 1641541Srgrimes * a new page for it. 1651541Srgrimes * 3) vm_map_pageable calls vm_fault; there is no page, 1661541Srgrimes * but there is a pager, so we call 1671541Srgrimes * pager_data_request. But the kmsg zone is empty, 1681541Srgrimes * so we must kmem_alloc. 1691541Srgrimes * 4) goto 1 1701541Srgrimes * 5) Even if the kmsg zone is not empty: when we get 1711541Srgrimes * the data back from the pager, it will be (very 1721541Srgrimes * stale) non-zero data. kmem_alloc is defined to 1731541Srgrimes * return zero-filled memory. 1741541Srgrimes * 1751541Srgrimes * We're intentionally not activating the pages we allocate 1761541Srgrimes * to prevent a race with page-out. vm_map_pageable will wire 1771541Srgrimes * the pages. 1781541Srgrimes */ 1791541Srgrimes 1801541Srgrimes vm_object_lock(kernel_object); 1811541Srgrimes for (i = 0 ; i < size; i+= PAGE_SIZE) { 1821541Srgrimes vm_page_t mem; 1831541Srgrimes 1841541Srgrimes while ((mem = vm_page_alloc(kernel_object, offset+i)) == NULL) { 1851541Srgrimes vm_object_unlock(kernel_object); 1861541Srgrimes VM_WAIT; 1871541Srgrimes vm_object_lock(kernel_object); 1881541Srgrimes } 1891541Srgrimes vm_page_zero_fill(mem); 1901541Srgrimes mem->flags &= ~PG_BUSY; 1911541Srgrimes } 1921541Srgrimes vm_object_unlock(kernel_object); 1931541Srgrimes 1941541Srgrimes /* 1951541Srgrimes * And finally, mark the data as non-pageable. 1961541Srgrimes */ 1971541Srgrimes 1981541Srgrimes (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 1991541Srgrimes 2001541Srgrimes /* 2011541Srgrimes * Try to coalesce the map 2021541Srgrimes */ 2031541Srgrimes 2041541Srgrimes vm_map_simplify(map, addr); 2051541Srgrimes 2061541Srgrimes return(addr); 2071541Srgrimes} 2081541Srgrimes 2091541Srgrimes/* 2101541Srgrimes * kmem_free: 2111541Srgrimes * 2121541Srgrimes * Release a region of kernel virtual memory allocated 2131541Srgrimes * with kmem_alloc, and return the physical pages 2141541Srgrimes * associated with that region. 2151541Srgrimes */ 2161541Srgrimesvoid kmem_free(map, addr, size) 2171541Srgrimes vm_map_t map; 2181541Srgrimes register vm_offset_t addr; 2191541Srgrimes vm_size_t size; 2201541Srgrimes{ 2211541Srgrimes (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 2221541Srgrimes} 2231541Srgrimes 2241541Srgrimes/* 2251541Srgrimes * kmem_suballoc: 2261541Srgrimes * 2271541Srgrimes * Allocates a map to manage a subrange 2281541Srgrimes * of the kernel virtual address space. 2291541Srgrimes * 2301541Srgrimes * Arguments are as follows: 2311541Srgrimes * 2321541Srgrimes * parent Map to take range from 2331541Srgrimes * size Size of range to find 2341541Srgrimes * min, max Returned endpoints of map 2351541Srgrimes * pageable Can the region be paged 2361541Srgrimes */ 2371541Srgrimesvm_map_t kmem_suballoc(parent, min, max, size, pageable) 2381541Srgrimes register vm_map_t parent; 2391541Srgrimes vm_offset_t *min, *max; 2401541Srgrimes register vm_size_t size; 2411541Srgrimes boolean_t pageable; 2421541Srgrimes{ 2431541Srgrimes register int ret; 2441541Srgrimes vm_map_t result; 2451541Srgrimes 2461541Srgrimes size = round_page(size); 2471541Srgrimes 2481541Srgrimes *min = (vm_offset_t) vm_map_min(parent); 2491541Srgrimes ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 2501541Srgrimes min, size, TRUE); 2511541Srgrimes if (ret != KERN_SUCCESS) { 2521541Srgrimes printf("kmem_suballoc: bad status return of %d.\n", ret); 2531541Srgrimes panic("kmem_suballoc"); 2541541Srgrimes } 2551541Srgrimes *max = *min + size; 2561541Srgrimes pmap_reference(vm_map_pmap(parent)); 2571541Srgrimes result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 2581541Srgrimes if (result == NULL) 2591541Srgrimes panic("kmem_suballoc: cannot create submap"); 2601541Srgrimes if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 2611541Srgrimes panic("kmem_suballoc: unable to change range to submap"); 2621541Srgrimes return(result); 2631541Srgrimes} 2641541Srgrimes 2651541Srgrimes/* 2661541Srgrimes * Allocate wired-down memory in the kernel's address map for the higher 2671541Srgrimes * level kernel memory allocator (kern/kern_malloc.c). We cannot use 2681541Srgrimes * kmem_alloc() because we may need to allocate memory at interrupt 2691541Srgrimes * level where we cannot block (canwait == FALSE). 2701541Srgrimes * 2711541Srgrimes * This routine has its own private kernel submap (kmem_map) and object 2721541Srgrimes * (kmem_object). This, combined with the fact that only malloc uses 2731541Srgrimes * this routine, ensures that we will never block in map or object waits. 2741541Srgrimes * 2751541Srgrimes * Note that this still only works in a uni-processor environment and 2761541Srgrimes * when called at splhigh(). 2771541Srgrimes * 2781541Srgrimes * We don't worry about expanding the map (adding entries) since entries 2791541Srgrimes * for wired maps are statically allocated. 2801541Srgrimes */ 2811541Srgrimesvm_offset_t 2821541Srgrimeskmem_malloc(map, size, canwait) 2831541Srgrimes register vm_map_t map; 2841541Srgrimes register vm_size_t size; 2851541Srgrimes boolean_t canwait; 2861541Srgrimes{ 2871541Srgrimes register vm_offset_t offset, i; 2881541Srgrimes vm_map_entry_t entry; 2891541Srgrimes vm_offset_t addr; 2901541Srgrimes vm_page_t m; 2911541Srgrimes 2921541Srgrimes if (map != kmem_map && map != mb_map) 2931541Srgrimes panic("kern_malloc_alloc: map != {kmem,mb}_map"); 2941541Srgrimes 2951541Srgrimes size = round_page(size); 2961541Srgrimes addr = vm_map_min(map); 2971541Srgrimes 2981541Srgrimes /* 2991541Srgrimes * Locate sufficient space in the map. This will give us the 3001541Srgrimes * final virtual address for the new memory, and thus will tell 3011541Srgrimes * us the offset within the kernel map. 3021541Srgrimes */ 3031541Srgrimes vm_map_lock(map); 3041541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 3051541Srgrimes vm_map_unlock(map); 3061549Srgrimes#if 0 3071541Srgrimes if (canwait) /* XXX should wait */ 3081541Srgrimes panic("kmem_malloc: %s too small", 3091541Srgrimes map == kmem_map ? "kmem_map" : "mb_map"); 3101549Srgrimes#endif 3111549Srgrimes if (canwait) 3121549Srgrimes panic("kmem_malloc: map too small"); 3131541Srgrimes return (0); 3141541Srgrimes } 3151541Srgrimes offset = addr - vm_map_min(kmem_map); 3161541Srgrimes vm_object_reference(kmem_object); 3171541Srgrimes vm_map_insert(map, kmem_object, offset, addr, addr + size); 3181541Srgrimes 3191541Srgrimes /* 3201541Srgrimes * If we can wait, just mark the range as wired 3211541Srgrimes * (will fault pages as necessary). 3221541Srgrimes */ 3231541Srgrimes if (canwait) { 3241541Srgrimes vm_map_unlock(map); 3251541Srgrimes (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 3261541Srgrimes FALSE); 3271541Srgrimes vm_map_simplify(map, addr); 3281541Srgrimes return(addr); 3291541Srgrimes } 3301541Srgrimes 3311541Srgrimes /* 3321541Srgrimes * If we cannot wait then we must allocate all memory up front, 3331541Srgrimes * pulling it off the active queue to prevent pageout. 3341541Srgrimes */ 3351541Srgrimes vm_object_lock(kmem_object); 3361541Srgrimes for (i = 0; i < size; i += PAGE_SIZE) { 3371541Srgrimes m = vm_page_alloc(kmem_object, offset + i); 3381541Srgrimes 3391541Srgrimes /* 3401541Srgrimes * Ran out of space, free everything up and return. 3411541Srgrimes * Don't need to lock page queues here as we know 3421541Srgrimes * that the pages we got aren't on any queues. 3431541Srgrimes */ 3441541Srgrimes if (m == NULL) { 3451541Srgrimes while (i != 0) { 3461541Srgrimes i -= PAGE_SIZE; 3471541Srgrimes m = vm_page_lookup(kmem_object, offset + i); 3481541Srgrimes vm_page_free(m); 3491541Srgrimes } 3501541Srgrimes vm_object_unlock(kmem_object); 3511541Srgrimes vm_map_delete(map, addr, addr + size); 3521541Srgrimes vm_map_unlock(map); 3531541Srgrimes return(0); 3541541Srgrimes } 3551541Srgrimes#if 0 3561541Srgrimes vm_page_zero_fill(m); 3571541Srgrimes#endif 3581541Srgrimes m->flags &= ~PG_BUSY; 3591541Srgrimes } 3601541Srgrimes vm_object_unlock(kmem_object); 3611541Srgrimes 3621541Srgrimes /* 3631541Srgrimes * Mark map entry as non-pageable. 3641541Srgrimes * Assert: vm_map_insert() will never be able to extend the previous 3651541Srgrimes * entry so there will be a new entry exactly corresponding to this 3661541Srgrimes * address range and it will have wired_count == 0. 3671541Srgrimes */ 3681541Srgrimes if (!vm_map_lookup_entry(map, addr, &entry) || 3691541Srgrimes entry->start != addr || entry->end != addr + size || 3701541Srgrimes entry->wired_count) 3711541Srgrimes panic("kmem_malloc: entry not found or misaligned"); 3721541Srgrimes entry->wired_count++; 3731541Srgrimes 3741541Srgrimes /* 3751541Srgrimes * Loop thru pages, entering them in the pmap. 3761541Srgrimes * (We cannot add them to the wired count without 3771541Srgrimes * wrapping the vm_page_queue_lock in splimp...) 3781541Srgrimes */ 3791541Srgrimes for (i = 0; i < size; i += PAGE_SIZE) { 3801541Srgrimes vm_object_lock(kmem_object); 3811541Srgrimes m = vm_page_lookup(kmem_object, offset + i); 3821541Srgrimes vm_object_unlock(kmem_object); 3831887Sdg pmap_kenter( addr + i, VM_PAGE_TO_PHYS(m)); 3841541Srgrimes } 3851541Srgrimes vm_map_unlock(map); 3861541Srgrimes 3871541Srgrimes vm_map_simplify(map, addr); 3881541Srgrimes return(addr); 3891541Srgrimes} 3901541Srgrimes 3911541Srgrimes/* 3921541Srgrimes * kmem_alloc_wait 3931541Srgrimes * 3941541Srgrimes * Allocates pageable memory from a sub-map of the kernel. If the submap 3951541Srgrimes * has no room, the caller sleeps waiting for more memory in the submap. 3961541Srgrimes * 3971541Srgrimes */ 3981541Srgrimesvm_offset_t kmem_alloc_wait(map, size) 3991541Srgrimes vm_map_t map; 4001541Srgrimes vm_size_t size; 4011541Srgrimes{ 4021541Srgrimes vm_offset_t addr; 4031541Srgrimes 4041541Srgrimes size = round_page(size); 4051541Srgrimes 4061541Srgrimes for (;;) { 4071541Srgrimes /* 4081541Srgrimes * To make this work for more than one map, 4091541Srgrimes * use the map's lock to lock out sleepers/wakers. 4101541Srgrimes */ 4111541Srgrimes vm_map_lock(map); 4121541Srgrimes if (vm_map_findspace(map, 0, size, &addr) == 0) 4131541Srgrimes break; 4141541Srgrimes /* no space now; see if we can ever get space */ 4151541Srgrimes if (vm_map_max(map) - vm_map_min(map) < size) { 4161541Srgrimes vm_map_unlock(map); 4171541Srgrimes return (0); 4181541Srgrimes } 4191541Srgrimes assert_wait((int)map, TRUE); 4201541Srgrimes vm_map_unlock(map); 4211549Srgrimes thread_block("kmaw"); 4221541Srgrimes } 4231541Srgrimes vm_map_insert(map, NULL, (vm_offset_t)0, addr, addr + size); 4241541Srgrimes vm_map_unlock(map); 4251541Srgrimes return (addr); 4261541Srgrimes} 4271541Srgrimes 4281541Srgrimes/* 4291541Srgrimes * kmem_free_wakeup 4301541Srgrimes * 4311541Srgrimes * Returns memory to a submap of the kernel, and wakes up any threads 4321541Srgrimes * waiting for memory in that map. 4331541Srgrimes */ 4341541Srgrimesvoid kmem_free_wakeup(map, addr, size) 4351541Srgrimes vm_map_t map; 4361541Srgrimes vm_offset_t addr; 4371541Srgrimes vm_size_t size; 4381541Srgrimes{ 4391541Srgrimes vm_map_lock(map); 4401541Srgrimes (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 4411541Srgrimes thread_wakeup((int)map); 4421541Srgrimes vm_map_unlock(map); 4431541Srgrimes} 4441541Srgrimes 4451541Srgrimes/* 4461541Srgrimes * Create the kernel map; insert a mapping covering kernel text, data, bss, 4471541Srgrimes * and all space allocated thus far (`boostrap' data). The new map will thus 4481541Srgrimes * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 4491541Srgrimes * the range between `start' and `end' as free. 4501541Srgrimes */ 4511541Srgrimesvoid kmem_init(start, end) 4521541Srgrimes vm_offset_t start, end; 4531541Srgrimes{ 4541541Srgrimes register vm_map_t m; 4551541Srgrimes 4561541Srgrimes m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 4571541Srgrimes vm_map_lock(m); 4581541Srgrimes /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 4591541Srgrimes kernel_map = m; 4601541Srgrimes (void) vm_map_insert(m, NULL, (vm_offset_t)0, 4611541Srgrimes VM_MIN_KERNEL_ADDRESS, start); 4621541Srgrimes /* ... and ending with the completion of the above `insert' */ 4631541Srgrimes vm_map_unlock(m); 4641541Srgrimes} 465