vm_kern.c revision 6129
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 * 646129Sdg * $Id: vm_kern.c,v 1.9 1995/01/24 10:12:51 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> 756129Sdg#include <sys/malloc.h> 761541Srgrimes 771541Srgrimes#include <vm/vm.h> 781541Srgrimes#include <vm/vm_page.h> 791541Srgrimes#include <vm/vm_pageout.h> 801541Srgrimes#include <vm/vm_kern.h> 811541Srgrimes 825455Sdgvm_map_t buffer_map; 835455Sdgvm_map_t kernel_map; 845455Sdgvm_map_t kmem_map; 855455Sdgvm_map_t mb_map; 865455Sdgvm_map_t io_map; 875455Sdgvm_map_t clean_map; 885455Sdgvm_map_t pager_map; 895455Sdgvm_map_t phys_map; 905455Sdgvm_map_t exec_map; 915455Sdgvm_map_t u_map; 922112Swollman 931541Srgrimes/* 941541Srgrimes * kmem_alloc_pageable: 951541Srgrimes * 961541Srgrimes * Allocate pageable memory to the kernel's address map. 971541Srgrimes * map must be "kernel_map" below. 981541Srgrimes */ 991541Srgrimes 1005455Sdgvm_offset_t 1015455Sdgkmem_alloc_pageable(map, size) 1025455Sdg vm_map_t map; 1035455Sdg register vm_size_t size; 1041541Srgrimes{ 1055455Sdg vm_offset_t addr; 1065455Sdg register int result; 1071541Srgrimes 1081541Srgrimes#if 0 1091541Srgrimes if (map != kernel_map) 1101541Srgrimes panic("kmem_alloc_pageable: not called with kernel_map"); 1111541Srgrimes#endif 1121541Srgrimes 1131541Srgrimes size = round_page(size); 1141541Srgrimes 1151541Srgrimes addr = vm_map_min(map); 1161541Srgrimes result = vm_map_find(map, NULL, (vm_offset_t) 0, 1175455Sdg &addr, size, TRUE); 1181541Srgrimes if (result != KERN_SUCCESS) { 1195455Sdg return (0); 1201541Srgrimes } 1215455Sdg return (addr); 1221541Srgrimes} 1231541Srgrimes 1241541Srgrimes/* 1251541Srgrimes * Allocate wired-down memory in the kernel's address map 1261541Srgrimes * or a submap. 1271541Srgrimes */ 1285455Sdgvm_offset_t 1295455Sdgkmem_alloc(map, size) 1305455Sdg register vm_map_t map; 1315455Sdg register vm_size_t size; 1321541Srgrimes{ 1335455Sdg vm_offset_t addr; 1345455Sdg register vm_offset_t offset; 1355455Sdg vm_offset_t i; 1361541Srgrimes 1371541Srgrimes size = round_page(size); 1381541Srgrimes 1391541Srgrimes /* 1405455Sdg * Use the kernel object for wired-down kernel pages. Assume that no 1415455Sdg * region of the kernel object is referenced more than once. 1421541Srgrimes */ 1431541Srgrimes 1441541Srgrimes /* 1455455Sdg * Locate sufficient space in the map. This will give us the final 1465455Sdg * virtual address for the new memory, and thus will tell us the 1475455Sdg * offset within the kernel map. 1481541Srgrimes */ 1491541Srgrimes vm_map_lock(map); 1501541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 1511541Srgrimes vm_map_unlock(map); 1521541Srgrimes return (0); 1531541Srgrimes } 1541541Srgrimes offset = addr - VM_MIN_KERNEL_ADDRESS; 1551541Srgrimes vm_object_reference(kernel_object); 1561541Srgrimes vm_map_insert(map, kernel_object, offset, addr, addr + size); 1571541Srgrimes vm_map_unlock(map); 1581541Srgrimes 1591541Srgrimes /* 1605455Sdg * Guarantee that there are pages already in this object before 1615455Sdg * calling vm_map_pageable. This is to prevent the following 1625455Sdg * scenario: 1635455Sdg * 1645455Sdg * 1) Threads have swapped out, so that there is a pager for the 1655455Sdg * kernel_object. 2) The kmsg zone is empty, and so we are 1665455Sdg * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 1675455Sdg * there is no page, but there is a pager, so we call 1685455Sdg * pager_data_request. But the kmsg zone is empty, so we must 1695455Sdg * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 1705455Sdg * we get the data back from the pager, it will be (very stale) 1715455Sdg * non-zero data. kmem_alloc is defined to return zero-filled memory. 1725455Sdg * 1735455Sdg * We're intentionally not activating the pages we allocate to prevent a 1745455Sdg * race with page-out. vm_map_pageable will wire the pages. 1751541Srgrimes */ 1761541Srgrimes 1771541Srgrimes vm_object_lock(kernel_object); 1785455Sdg for (i = 0; i < size; i += PAGE_SIZE) { 1795455Sdg vm_page_t mem; 1801541Srgrimes 1815841Sdg while ((mem = vm_page_alloc(kernel_object, offset + i, VM_ALLOC_NORMAL)) == NULL) { 1821541Srgrimes vm_object_unlock(kernel_object); 1831541Srgrimes VM_WAIT; 1841541Srgrimes vm_object_lock(kernel_object); 1851541Srgrimes } 1861541Srgrimes vm_page_zero_fill(mem); 1871541Srgrimes mem->flags &= ~PG_BUSY; 1885455Sdg mem->valid |= VM_PAGE_BITS_ALL; 1891541Srgrimes } 1901541Srgrimes vm_object_unlock(kernel_object); 1915455Sdg 1921541Srgrimes /* 1935455Sdg * And finally, mark the data as non-pageable. 1941541Srgrimes */ 1951541Srgrimes 1961541Srgrimes (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 1971541Srgrimes 1981541Srgrimes /* 1995455Sdg * Try to coalesce the map 2001541Srgrimes */ 2011541Srgrimes vm_map_simplify(map, addr); 2021541Srgrimes 2035455Sdg return (addr); 2041541Srgrimes} 2051541Srgrimes 2061541Srgrimes/* 2071541Srgrimes * kmem_free: 2081541Srgrimes * 2091541Srgrimes * Release a region of kernel virtual memory allocated 2101541Srgrimes * with kmem_alloc, and return the physical pages 2111541Srgrimes * associated with that region. 2121541Srgrimes */ 2135455Sdgvoid 2145455Sdgkmem_free(map, addr, size) 2155455Sdg vm_map_t map; 2165455Sdg register vm_offset_t addr; 2175455Sdg vm_size_t size; 2181541Srgrimes{ 2191541Srgrimes (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 2201541Srgrimes} 2211541Srgrimes 2221541Srgrimes/* 2231541Srgrimes * kmem_suballoc: 2241541Srgrimes * 2251541Srgrimes * Allocates a map to manage a subrange 2261541Srgrimes * of the kernel virtual address space. 2271541Srgrimes * 2281541Srgrimes * Arguments are as follows: 2291541Srgrimes * 2301541Srgrimes * parent Map to take range from 2311541Srgrimes * size Size of range to find 2321541Srgrimes * min, max Returned endpoints of map 2331541Srgrimes * pageable Can the region be paged 2341541Srgrimes */ 2355455Sdgvm_map_t 2365455Sdgkmem_suballoc(parent, min, max, size, pageable) 2375455Sdg register vm_map_t parent; 2385455Sdg vm_offset_t *min, *max; 2395455Sdg register vm_size_t size; 2405455Sdg boolean_t pageable; 2411541Srgrimes{ 2425455Sdg register int ret; 2435455Sdg vm_map_t result; 2441541Srgrimes 2451541Srgrimes size = round_page(size); 2461541Srgrimes 2471541Srgrimes *min = (vm_offset_t) vm_map_min(parent); 2481541Srgrimes ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 2495455Sdg min, size, TRUE); 2501541Srgrimes if (ret != KERN_SUCCESS) { 2511541Srgrimes printf("kmem_suballoc: bad status return of %d.\n", ret); 2521541Srgrimes panic("kmem_suballoc"); 2531541Srgrimes } 2541541Srgrimes *max = *min + size; 2551541Srgrimes pmap_reference(vm_map_pmap(parent)); 2561541Srgrimes result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 2571541Srgrimes if (result == NULL) 2581541Srgrimes panic("kmem_suballoc: cannot create submap"); 2591541Srgrimes if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 2601541Srgrimes panic("kmem_suballoc: unable to change range to submap"); 2615455Sdg return (result); 2621541Srgrimes} 2631541Srgrimes 2641541Srgrimes/* 2651541Srgrimes * Allocate wired-down memory in the kernel's address map for the higher 2661541Srgrimes * level kernel memory allocator (kern/kern_malloc.c). We cannot use 2671541Srgrimes * kmem_alloc() because we may need to allocate memory at interrupt 2681541Srgrimes * level where we cannot block (canwait == FALSE). 2691541Srgrimes * 2701541Srgrimes * This routine has its own private kernel submap (kmem_map) and object 2711541Srgrimes * (kmem_object). This, combined with the fact that only malloc uses 2721541Srgrimes * this routine, ensures that we will never block in map or object waits. 2731541Srgrimes * 2741541Srgrimes * Note that this still only works in a uni-processor environment and 2751541Srgrimes * when called at splhigh(). 2761541Srgrimes * 2771541Srgrimes * We don't worry about expanding the map (adding entries) since entries 2781541Srgrimes * for wired maps are statically allocated. 2791541Srgrimes */ 2801541Srgrimesvm_offset_t 2816129Sdgkmem_malloc(map, size, waitflag) 2825455Sdg register vm_map_t map; 2835455Sdg register vm_size_t size; 2846129Sdg boolean_t waitflag; 2851541Srgrimes{ 2865455Sdg register vm_offset_t offset, i; 2875455Sdg vm_map_entry_t entry; 2885455Sdg vm_offset_t addr; 2895455Sdg vm_page_t m; 2901541Srgrimes 2911541Srgrimes if (map != kmem_map && map != mb_map) 2921541Srgrimes panic("kern_malloc_alloc: map != {kmem,mb}_map"); 2931541Srgrimes 2941541Srgrimes size = round_page(size); 2951541Srgrimes addr = vm_map_min(map); 2961541Srgrimes 2971541Srgrimes /* 2985455Sdg * Locate sufficient space in the map. This will give us the final 2995455Sdg * virtual address for the new memory, and thus will tell us the 3005455Sdg * offset within the kernel map. 3011541Srgrimes */ 3021541Srgrimes vm_map_lock(map); 3031541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 3041541Srgrimes vm_map_unlock(map); 3051549Srgrimes#if 0 3065455Sdg if (canwait) /* XXX should wait */ 3071541Srgrimes panic("kmem_malloc: %s too small", 3081541Srgrimes map == kmem_map ? "kmem_map" : "mb_map"); 3091549Srgrimes#endif 3106129Sdg if (waitflag == M_WAITOK) 3111549Srgrimes panic("kmem_malloc: map too small"); 3121541Srgrimes return (0); 3131541Srgrimes } 3141541Srgrimes offset = addr - vm_map_min(kmem_map); 3151541Srgrimes vm_object_reference(kmem_object); 3161541Srgrimes vm_map_insert(map, kmem_object, offset, addr, addr + size); 3171541Srgrimes 3181541Srgrimes /* 3195455Sdg * If we can wait, just mark the range as wired (will fault pages as 3205455Sdg * necessary). 3211541Srgrimes */ 3226129Sdg if (waitflag == M_WAITOK) { 3231541Srgrimes vm_map_unlock(map); 3241541Srgrimes (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 3255455Sdg FALSE); 3261541Srgrimes vm_map_simplify(map, addr); 3275455Sdg return (addr); 3281541Srgrimes } 3291541Srgrimes /* 3301541Srgrimes * If we cannot wait then we must allocate all memory up front, 3311541Srgrimes * pulling it off the active queue to prevent pageout. 3321541Srgrimes */ 3331541Srgrimes vm_object_lock(kmem_object); 3341541Srgrimes for (i = 0; i < size; i += PAGE_SIZE) { 3356129Sdg m = vm_page_alloc(kmem_object, offset + i, 3366129Sdg (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 3371541Srgrimes 3381541Srgrimes /* 3395455Sdg * Ran out of space, free everything up and return. Don't need 3405455Sdg * to lock page queues here as we know that the pages we got 3415455Sdg * aren't on any queues. 3421541Srgrimes */ 3431541Srgrimes if (m == NULL) { 3441541Srgrimes while (i != 0) { 3451541Srgrimes i -= PAGE_SIZE; 3461541Srgrimes m = vm_page_lookup(kmem_object, offset + i); 3471541Srgrimes vm_page_free(m); 3481541Srgrimes } 3491541Srgrimes vm_object_unlock(kmem_object); 3501541Srgrimes vm_map_delete(map, addr, addr + size); 3511541Srgrimes vm_map_unlock(map); 3525455Sdg return (0); 3531541Srgrimes } 3541541Srgrimes#if 0 3551541Srgrimes vm_page_zero_fill(m); 3561541Srgrimes#endif 3571541Srgrimes m->flags &= ~PG_BUSY; 3585455Sdg m->valid |= VM_PAGE_BITS_ALL; 3591541Srgrimes } 3601541Srgrimes vm_object_unlock(kmem_object); 3611541Srgrimes 3621541Srgrimes /* 3635455Sdg * Mark map entry as non-pageable. Assert: vm_map_insert() will never 3645455Sdg * be able to extend the previous entry so there will be a new entry 3655455Sdg * exactly corresponding to this address range and it will have 3665455Sdg * 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 /* 3755455Sdg * Loop thru pages, entering them in the pmap. (We cannot add them to 3765455Sdg * the wired count without wrapping the vm_page_queue_lock in 3775455Sdg * 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); 3835455Sdg pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 3841541Srgrimes } 3851541Srgrimes vm_map_unlock(map); 3861541Srgrimes 3871541Srgrimes vm_map_simplify(map, addr); 3885455Sdg 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 */ 3985455Sdgvm_offset_t 3995455Sdgkmem_alloc_wait(map, size) 4005455Sdg vm_map_t map; 4015455Sdg vm_size_t size; 4021541Srgrimes{ 4035455Sdg vm_offset_t addr; 4041541Srgrimes 4051541Srgrimes size = round_page(size); 4061541Srgrimes 4071541Srgrimes for (;;) { 4081541Srgrimes /* 4095455Sdg * To make this work for more than one map, use the map's lock 4105455Sdg * to lock out sleepers/wakers. 4111541Srgrimes */ 4121541Srgrimes vm_map_lock(map); 4131541Srgrimes if (vm_map_findspace(map, 0, size, &addr) == 0) 4141541Srgrimes break; 4151541Srgrimes /* no space now; see if we can ever get space */ 4161541Srgrimes if (vm_map_max(map) - vm_map_min(map) < size) { 4171541Srgrimes vm_map_unlock(map); 4181541Srgrimes return (0); 4191541Srgrimes } 4205455Sdg assert_wait((int) map, TRUE); 4211541Srgrimes vm_map_unlock(map); 4221549Srgrimes thread_block("kmaw"); 4231541Srgrimes } 4245455Sdg vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 4251541Srgrimes vm_map_unlock(map); 4261541Srgrimes return (addr); 4271541Srgrimes} 4281541Srgrimes 4291541Srgrimes/* 4301541Srgrimes * kmem_free_wakeup 4311541Srgrimes * 4321541Srgrimes * Returns memory to a submap of the kernel, and wakes up any threads 4331541Srgrimes * waiting for memory in that map. 4341541Srgrimes */ 4355455Sdgvoid 4365455Sdgkmem_free_wakeup(map, addr, size) 4375455Sdg vm_map_t map; 4385455Sdg vm_offset_t addr; 4395455Sdg vm_size_t size; 4401541Srgrimes{ 4411541Srgrimes vm_map_lock(map); 4421541Srgrimes (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 4435455Sdg thread_wakeup((int) map); 4441541Srgrimes vm_map_unlock(map); 4451541Srgrimes} 4461541Srgrimes 4471541Srgrimes/* 4481541Srgrimes * Create the kernel map; insert a mapping covering kernel text, data, bss, 4491541Srgrimes * and all space allocated thus far (`boostrap' data). The new map will thus 4501541Srgrimes * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 4511541Srgrimes * the range between `start' and `end' as free. 4521541Srgrimes */ 4535455Sdgvoid 4545455Sdgkmem_init(start, end) 4551541Srgrimes vm_offset_t start, end; 4561541Srgrimes{ 4571541Srgrimes register vm_map_t m; 4581541Srgrimes 4591541Srgrimes m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 4601541Srgrimes vm_map_lock(m); 4611541Srgrimes /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 4621541Srgrimes kernel_map = m; 4635455Sdg (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 4641541Srgrimes VM_MIN_KERNEL_ADDRESS, start); 4651541Srgrimes /* ... and ending with the completion of the above `insert' */ 4661541Srgrimes vm_map_unlock(m); 4671541Srgrimes} 468