vm_kern.c revision 8876
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 * 648876Srgrimes * $Id: vm_kern.c,v 1.12 1995/03/15 07:52:06 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> 767066Sdg#include <sys/syslog.h> 771541Srgrimes 781541Srgrimes#include <vm/vm.h> 791541Srgrimes#include <vm/vm_page.h> 801541Srgrimes#include <vm/vm_pageout.h> 811541Srgrimes#include <vm/vm_kern.h> 821541Srgrimes 835455Sdgvm_map_t buffer_map; 845455Sdgvm_map_t kernel_map; 855455Sdgvm_map_t kmem_map; 865455Sdgvm_map_t mb_map; 875455Sdgvm_map_t io_map; 885455Sdgvm_map_t clean_map; 895455Sdgvm_map_t pager_map; 905455Sdgvm_map_t phys_map; 915455Sdgvm_map_t exec_map; 925455Sdgvm_map_t u_map; 937066Sdgextern int mb_map_full; 942112Swollman 951541Srgrimes/* 961541Srgrimes * kmem_alloc_pageable: 971541Srgrimes * 981541Srgrimes * Allocate pageable memory to the kernel's address map. 991541Srgrimes * map must be "kernel_map" below. 1001541Srgrimes */ 1011541Srgrimes 1028876Srgrimesvm_offset_t 1035455Sdgkmem_alloc_pageable(map, size) 1045455Sdg vm_map_t map; 1055455Sdg register vm_size_t size; 1061541Srgrimes{ 1075455Sdg vm_offset_t addr; 1085455Sdg register int result; 1091541Srgrimes 1101541Srgrimes#if 0 1111541Srgrimes if (map != kernel_map) 1121541Srgrimes panic("kmem_alloc_pageable: not called with kernel_map"); 1131541Srgrimes#endif 1141541Srgrimes 1151541Srgrimes size = round_page(size); 1161541Srgrimes 1171541Srgrimes addr = vm_map_min(map); 1181541Srgrimes result = vm_map_find(map, NULL, (vm_offset_t) 0, 1195455Sdg &addr, size, TRUE); 1201541Srgrimes if (result != KERN_SUCCESS) { 1215455Sdg return (0); 1221541Srgrimes } 1235455Sdg return (addr); 1241541Srgrimes} 1251541Srgrimes 1261541Srgrimes/* 1271541Srgrimes * Allocate wired-down memory in the kernel's address map 1281541Srgrimes * or a submap. 1291541Srgrimes */ 1308876Srgrimesvm_offset_t 1315455Sdgkmem_alloc(map, size) 1325455Sdg register vm_map_t map; 1335455Sdg register vm_size_t size; 1341541Srgrimes{ 1355455Sdg vm_offset_t addr; 1365455Sdg register vm_offset_t offset; 1375455Sdg vm_offset_t i; 1381541Srgrimes 1391541Srgrimes size = round_page(size); 1401541Srgrimes 1411541Srgrimes /* 1425455Sdg * Use the kernel object for wired-down kernel pages. Assume that no 1435455Sdg * region of the kernel object is referenced more than once. 1441541Srgrimes */ 1451541Srgrimes 1461541Srgrimes /* 1475455Sdg * Locate sufficient space in the map. This will give us the final 1485455Sdg * virtual address for the new memory, and thus will tell us the 1495455Sdg * offset within the kernel map. 1501541Srgrimes */ 1511541Srgrimes vm_map_lock(map); 1521541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 1531541Srgrimes vm_map_unlock(map); 1541541Srgrimes return (0); 1551541Srgrimes } 1561541Srgrimes offset = addr - VM_MIN_KERNEL_ADDRESS; 1571541Srgrimes vm_object_reference(kernel_object); 1581541Srgrimes vm_map_insert(map, kernel_object, offset, addr, addr + size); 1591541Srgrimes vm_map_unlock(map); 1601541Srgrimes 1611541Srgrimes /* 1625455Sdg * Guarantee that there are pages already in this object before 1635455Sdg * calling vm_map_pageable. This is to prevent the following 1645455Sdg * scenario: 1658876Srgrimes * 1665455Sdg * 1) Threads have swapped out, so that there is a pager for the 1675455Sdg * kernel_object. 2) The kmsg zone is empty, and so we are 1685455Sdg * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 1695455Sdg * there is no page, but there is a pager, so we call 1705455Sdg * pager_data_request. But the kmsg zone is empty, so we must 1715455Sdg * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 1725455Sdg * we get the data back from the pager, it will be (very stale) 1735455Sdg * non-zero data. kmem_alloc is defined to return zero-filled memory. 1748876Srgrimes * 1755455Sdg * We're intentionally not activating the pages we allocate to prevent a 1765455Sdg * race with page-out. vm_map_pageable will wire the pages. 1771541Srgrimes */ 1781541Srgrimes 1791541Srgrimes vm_object_lock(kernel_object); 1805455Sdg for (i = 0; i < size; i += PAGE_SIZE) { 1815455Sdg vm_page_t mem; 1821541Srgrimes 1835841Sdg while ((mem = vm_page_alloc(kernel_object, offset + i, VM_ALLOC_NORMAL)) == NULL) { 1841541Srgrimes vm_object_unlock(kernel_object); 1851541Srgrimes VM_WAIT; 1861541Srgrimes vm_object_lock(kernel_object); 1871541Srgrimes } 1881541Srgrimes vm_page_zero_fill(mem); 1891541Srgrimes mem->flags &= ~PG_BUSY; 1906585Sdg mem->valid = VM_PAGE_BITS_ALL; 1911541Srgrimes } 1921541Srgrimes vm_object_unlock(kernel_object); 1935455Sdg 1941541Srgrimes /* 1955455Sdg * 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 /* 2015455Sdg * Try to coalesce the map 2021541Srgrimes */ 2031541Srgrimes vm_map_simplify(map, addr); 2041541Srgrimes 2055455Sdg return (addr); 2061541Srgrimes} 2071541Srgrimes 2081541Srgrimes/* 2091541Srgrimes * kmem_free: 2101541Srgrimes * 2111541Srgrimes * Release a region of kernel virtual memory allocated 2121541Srgrimes * with kmem_alloc, and return the physical pages 2131541Srgrimes * associated with that region. 2141541Srgrimes */ 2158876Srgrimesvoid 2165455Sdgkmem_free(map, addr, size) 2175455Sdg vm_map_t map; 2185455Sdg register vm_offset_t addr; 2195455Sdg 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 */ 2378876Srgrimesvm_map_t 2385455Sdgkmem_suballoc(parent, min, max, size, pageable) 2395455Sdg register vm_map_t parent; 2405455Sdg vm_offset_t *min, *max; 2415455Sdg register vm_size_t size; 2425455Sdg boolean_t pageable; 2431541Srgrimes{ 2445455Sdg register int ret; 2455455Sdg vm_map_t result; 2461541Srgrimes 2471541Srgrimes size = round_page(size); 2481541Srgrimes 2491541Srgrimes *min = (vm_offset_t) vm_map_min(parent); 2501541Srgrimes ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 2515455Sdg min, size, TRUE); 2521541Srgrimes if (ret != KERN_SUCCESS) { 2531541Srgrimes printf("kmem_suballoc: bad status return of %d.\n", ret); 2541541Srgrimes panic("kmem_suballoc"); 2551541Srgrimes } 2561541Srgrimes *max = *min + size; 2571541Srgrimes pmap_reference(vm_map_pmap(parent)); 2581541Srgrimes result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 2591541Srgrimes if (result == NULL) 2601541Srgrimes panic("kmem_suballoc: cannot create submap"); 2611541Srgrimes if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 2621541Srgrimes panic("kmem_suballoc: unable to change range to submap"); 2635455Sdg return (result); 2641541Srgrimes} 2651541Srgrimes 2661541Srgrimes/* 2671541Srgrimes * Allocate wired-down memory in the kernel's address map for the higher 2681541Srgrimes * level kernel memory allocator (kern/kern_malloc.c). We cannot use 2691541Srgrimes * kmem_alloc() because we may need to allocate memory at interrupt 2701541Srgrimes * level where we cannot block (canwait == FALSE). 2711541Srgrimes * 2721541Srgrimes * This routine has its own private kernel submap (kmem_map) and object 2731541Srgrimes * (kmem_object). This, combined with the fact that only malloc uses 2741541Srgrimes * this routine, ensures that we will never block in map or object waits. 2751541Srgrimes * 2761541Srgrimes * Note that this still only works in a uni-processor environment and 2771541Srgrimes * when called at splhigh(). 2781541Srgrimes * 2791541Srgrimes * We don't worry about expanding the map (adding entries) since entries 2801541Srgrimes * for wired maps are statically allocated. 2811541Srgrimes */ 2821541Srgrimesvm_offset_t 2836129Sdgkmem_malloc(map, size, waitflag) 2845455Sdg register vm_map_t map; 2855455Sdg register vm_size_t size; 2866129Sdg boolean_t waitflag; 2871541Srgrimes{ 2885455Sdg register vm_offset_t offset, i; 2895455Sdg vm_map_entry_t entry; 2905455Sdg vm_offset_t addr; 2915455Sdg vm_page_t m; 2921541Srgrimes 2931541Srgrimes if (map != kmem_map && map != mb_map) 2947066Sdg panic("kmem_malloc: map != {kmem,mb}_map"); 2951541Srgrimes 2961541Srgrimes size = round_page(size); 2971541Srgrimes addr = vm_map_min(map); 2981541Srgrimes 2991541Srgrimes /* 3005455Sdg * Locate sufficient space in the map. This will give us the final 3015455Sdg * virtual address for the new memory, and thus will tell us the 3025455Sdg * offset within the kernel map. 3031541Srgrimes */ 3041541Srgrimes vm_map_lock(map); 3051541Srgrimes if (vm_map_findspace(map, 0, size, &addr)) { 3061541Srgrimes vm_map_unlock(map); 3077066Sdg if (map == mb_map) { 3087066Sdg mb_map_full = TRUE; 3097066Sdg log(LOG_ERR, "mb_map full\n"); 3107066Sdg return (0); 3117066Sdg } 3126129Sdg if (waitflag == M_WAITOK) 3137066Sdg panic("kmem_malloc: kmem_map too small"); 3141541Srgrimes return (0); 3151541Srgrimes } 3161541Srgrimes offset = addr - vm_map_min(kmem_map); 3171541Srgrimes vm_object_reference(kmem_object); 3181541Srgrimes vm_map_insert(map, kmem_object, offset, addr, addr + size); 3191541Srgrimes 3201541Srgrimes /* 3215455Sdg * If we can wait, just mark the range as wired (will fault pages as 3225455Sdg * necessary). 3231541Srgrimes */ 3246129Sdg if (waitflag == M_WAITOK) { 3251541Srgrimes vm_map_unlock(map); 3261541Srgrimes (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 3275455Sdg FALSE); 3281541Srgrimes vm_map_simplify(map, addr); 3295455Sdg return (addr); 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) { 3376129Sdg m = vm_page_alloc(kmem_object, offset + i, 3386129Sdg (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 3391541Srgrimes 3401541Srgrimes /* 3415455Sdg * Ran out of space, free everything up and return. Don't need 3425455Sdg * to lock page queues here as we know that the pages we got 3435455Sdg * aren't on any queues. 3441541Srgrimes */ 3451541Srgrimes if (m == NULL) { 3461541Srgrimes while (i != 0) { 3471541Srgrimes i -= PAGE_SIZE; 3481541Srgrimes m = vm_page_lookup(kmem_object, offset + i); 3491541Srgrimes vm_page_free(m); 3501541Srgrimes } 3511541Srgrimes vm_object_unlock(kmem_object); 3521541Srgrimes vm_map_delete(map, addr, addr + size); 3531541Srgrimes vm_map_unlock(map); 3545455Sdg return (0); 3551541Srgrimes } 3561541Srgrimes#if 0 3571541Srgrimes vm_page_zero_fill(m); 3581541Srgrimes#endif 3591541Srgrimes m->flags &= ~PG_BUSY; 3606585Sdg m->valid = VM_PAGE_BITS_ALL; 3611541Srgrimes } 3621541Srgrimes vm_object_unlock(kmem_object); 3631541Srgrimes 3641541Srgrimes /* 3655455Sdg * Mark map entry as non-pageable. Assert: vm_map_insert() will never 3665455Sdg * be able to extend the previous entry so there will be a new entry 3675455Sdg * exactly corresponding to this address range and it will have 3685455Sdg * wired_count == 0. 3691541Srgrimes */ 3701541Srgrimes if (!vm_map_lookup_entry(map, addr, &entry) || 3711541Srgrimes entry->start != addr || entry->end != addr + size || 3721541Srgrimes entry->wired_count) 3731541Srgrimes panic("kmem_malloc: entry not found or misaligned"); 3741541Srgrimes entry->wired_count++; 3751541Srgrimes 3761541Srgrimes /* 3775455Sdg * Loop thru pages, entering them in the pmap. (We cannot add them to 3785455Sdg * the wired count without wrapping the vm_page_queue_lock in 3795455Sdg * splimp...) 3801541Srgrimes */ 3811541Srgrimes for (i = 0; i < size; i += PAGE_SIZE) { 3821541Srgrimes vm_object_lock(kmem_object); 3831541Srgrimes m = vm_page_lookup(kmem_object, offset + i); 3841541Srgrimes vm_object_unlock(kmem_object); 3855455Sdg pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 3861541Srgrimes } 3871541Srgrimes vm_map_unlock(map); 3881541Srgrimes 3891541Srgrimes vm_map_simplify(map, addr); 3905455Sdg return (addr); 3911541Srgrimes} 3921541Srgrimes 3931541Srgrimes/* 3941541Srgrimes * kmem_alloc_wait 3951541Srgrimes * 3961541Srgrimes * Allocates pageable memory from a sub-map of the kernel. If the submap 3971541Srgrimes * has no room, the caller sleeps waiting for more memory in the submap. 3981541Srgrimes * 3991541Srgrimes */ 4008876Srgrimesvm_offset_t 4015455Sdgkmem_alloc_wait(map, size) 4025455Sdg vm_map_t map; 4035455Sdg vm_size_t size; 4041541Srgrimes{ 4055455Sdg vm_offset_t addr; 4061541Srgrimes 4071541Srgrimes size = round_page(size); 4081541Srgrimes 4091541Srgrimes for (;;) { 4101541Srgrimes /* 4115455Sdg * To make this work for more than one map, use the map's lock 4125455Sdg * to lock out sleepers/wakers. 4131541Srgrimes */ 4141541Srgrimes vm_map_lock(map); 4151541Srgrimes if (vm_map_findspace(map, 0, size, &addr) == 0) 4161541Srgrimes break; 4171541Srgrimes /* no space now; see if we can ever get space */ 4181541Srgrimes if (vm_map_max(map) - vm_map_min(map) < size) { 4191541Srgrimes vm_map_unlock(map); 4201541Srgrimes return (0); 4211541Srgrimes } 4225455Sdg assert_wait((int) map, TRUE); 4231541Srgrimes vm_map_unlock(map); 4241549Srgrimes thread_block("kmaw"); 4251541Srgrimes } 4265455Sdg vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size); 4271541Srgrimes vm_map_unlock(map); 4281541Srgrimes return (addr); 4291541Srgrimes} 4301541Srgrimes 4311541Srgrimes/* 4321541Srgrimes * kmem_free_wakeup 4331541Srgrimes * 4341541Srgrimes * Returns memory to a submap of the kernel, and wakes up any threads 4351541Srgrimes * waiting for memory in that map. 4361541Srgrimes */ 4378876Srgrimesvoid 4385455Sdgkmem_free_wakeup(map, addr, size) 4395455Sdg vm_map_t map; 4405455Sdg vm_offset_t addr; 4415455Sdg vm_size_t size; 4421541Srgrimes{ 4431541Srgrimes vm_map_lock(map); 4441541Srgrimes (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 4455455Sdg thread_wakeup((int) map); 4461541Srgrimes vm_map_unlock(map); 4471541Srgrimes} 4481541Srgrimes 4491541Srgrimes/* 4501541Srgrimes * Create the kernel map; insert a mapping covering kernel text, data, bss, 4511541Srgrimes * and all space allocated thus far (`boostrap' data). The new map will thus 4521541Srgrimes * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 4531541Srgrimes * the range between `start' and `end' as free. 4541541Srgrimes */ 4558876Srgrimesvoid 4565455Sdgkmem_init(start, end) 4571541Srgrimes vm_offset_t start, end; 4581541Srgrimes{ 4591541Srgrimes register vm_map_t m; 4601541Srgrimes 4611541Srgrimes m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 4621541Srgrimes vm_map_lock(m); 4631541Srgrimes /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 4641541Srgrimes kernel_map = m; 4655455Sdg (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 4661541Srgrimes VM_MIN_KERNEL_ADDRESS, start); 4671541Srgrimes /* ... and ending with the completion of the above `insert' */ 4681541Srgrimes vm_map_unlock(m); 4691541Srgrimes} 470