vm_kern.c revision 15722
155839Sasmodai/* 269626Sru * Copyright (c) 1991, 1993 369626Sru * The Regents of the University of California. All rights reserved. 4114402Sru * 5114402Sru * This code is derived from software contributed to Berkeley by 669626Sru * The Mach Operating System project at Carnegie-Mellon University. 7114402Sru * 8104862Sru * Redistribution and use in source and binary forms, with or without 969626Sru * modification, are permitted provided that the following conditions 1055839Sasmodai * are met: 11104862Sru * 1. Redistributions of source code must retain the above copyright 1255839Sasmodai * notice, this list of conditions and the following disclaimer. 1355839Sasmodai * 2. Redistributions in binary form must reproduce the above copyright 1455839Sasmodai * notice, this list of conditions and the following disclaimer in the 1555839Sasmodai * documentation and/or other materials provided with the distribution. 1655839Sasmodai * 3. All advertising materials mentioning features or use of this software 1755839Sasmodai * must display the following acknowledgement: 18104862Sru * This product includes software developed by the University of 19104862Sru * California, Berkeley and its contributors. 20104862Sru * 4. Neither the name of the University nor the names of its contributors 21104862Sru * may be used to endorse or promote products derived from this software 22104862Sru * without specific prior written permission. 23104862Sru * 24114402Sru * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25104862Sru * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26114402Sru * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27114402Sru * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28104862Sru * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29104862Sru * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30104862Sru * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31104862Sru * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32104862Sru * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33104862Sru * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34104862Sru * SUCH DAMAGE. 35104862Sru * 36104862Sru * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 37104862Sru * 38104862Sru * 39104862Sru * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40104862Sru * All rights reserved. 41104862Sru * 42104862Sru * Authors: Avadis Tevanian, Jr., Michael Wayne Young 43104862Sru * 44104862Sru * Permission to use, copy, modify and distribute this software and 4569626Sru * its documentation is hereby granted, provided that both the copyright 4669626Sru * notice and this permission notice appear in all copies of the 4769626Sru * software, derivative works or modified versions, and any portions 4875584Sru * thereof, and that both notices appear in supporting documentation. 4975584Sru * 5069626Sru * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 5169626Sru * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 5269626Sru * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 5369626Sru * 5475584Sru * Carnegie Mellon requests users of this software to return to 5575584Sru * 5669626Sru * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 5769626Sru * School of Computer Science 5869626Sru * Carnegie Mellon University 5969626Sru * Pittsburgh PA 15213-3890 6075584Sru * 6175584Sru * any improvements or extensions that they make and grant Carnegie the 6269626Sru * rights to redistribute these changes. 6375584Sru * 6469626Sru * $Id: vm_kern.c,v 1.23 1996/04/24 04:16:44 dyson Exp $ 6569626Sru */ 6669626Sru 6769626Sru/* 68114402Sru * Kernel memory management. 69114402Sru */ 7075584Sru 71114402Sru#include <sys/param.h> 7275584Sru#include <sys/systm.h> 73114402Sru#include <sys/kernel.h> 7475584Sru#include <sys/proc.h> 7575584Sru#include <sys/malloc.h> 7675584Sru#include <sys/syslog.h> 7769626Sru#include <sys/queue.h> 7875584Sru#include <sys/vmmeter.h> 7975584Sru 8075584Sru#include <vm/vm.h> 8169626Sru#include <vm/vm_param.h> 8275584Sru#include <vm/vm_prot.h> 83104862Sru#include <vm/lock.h> 84104862Sru#include <vm/pmap.h> 85104862Sru#include <vm/vm_map.h> 86104862Sru#include <vm/vm_object.h> 87104862Sru#include <vm/vm_page.h> 88104862Sru#include <vm/vm_pageout.h> 89104862Sru#include <vm/vm_kern.h> 90104862Sru#include <vm/vm_extern.h> 91104862Sru 92104862Sruvm_map_t buffer_map; 93104862Sruvm_map_t kernel_map; 94104862Sruvm_map_t kmem_map; 95104862Sruvm_map_t mb_map; 96104862Sruint mb_map_full; 97104862Sruvm_map_t mcl_map; 98104862Sruint mcl_map_full; 99104862Sruvm_map_t io_map; 100104862Sruvm_map_t clean_map; 101104862Sruvm_map_t phys_map; 102104862Sruvm_map_t exec_map; 103104862Sruvm_map_t u_map; 104104862Sru 105104862Sru/* 106104862Sru * kmem_alloc_pageable: 107104862Sru * 108104862Sru * Allocate pageable memory to the kernel's address map. 109104862Sru * "map" must be kernel_map or a submap of kernel_map. 110104862Sru */ 111104862Sru 112114402Sruvm_offset_t 113104862Srukmem_alloc_pageable(map, size) 114104862Sru vm_map_t map; 115104862Sru register vm_size_t size; 116104862Sru{ 117104862Sru vm_offset_t addr; 118104862Sru register int result; 119104862Sru 120104862Sru size = round_page(size); 12175584Sru addr = vm_map_min(map); 12275584Sru result = vm_map_find(map, NULL, (vm_offset_t) 0, 12375584Sru &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 124104862Sru if (result != KERN_SUCCESS) { 12575584Sru return (0); 126104862Sru } 127104862Sru return (addr); 128104862Sru} 12975584Sru 130104862Sru/* 131104862Sru * Allocate wired-down memory in the kernel's address map 13275584Sru * or a submap. 13375584Sru */ 134104862Sruvm_offset_t 135104862Srukmem_alloc(map, size) 136104862Sru register vm_map_t map; 13775584Sru register vm_size_t size; 138104862Sru{ 139104862Sru vm_offset_t addr; 140104862Sru register vm_offset_t offset; 14175584Sru vm_offset_t i; 142104862Sru 14375584Sru size = round_page(size); 14475584Sru 14575584Sru /* 14669626Sru * Use the kernel object for wired-down kernel pages. Assume that no 14769626Sru * region of the kernel object is referenced more than once. 14869626Sru */ 14975584Sru 15075584Sru /* 15175584Sru * Locate sufficient space in the map. This will give us the final 15275584Sru * virtual address for the new memory, and thus will tell us the 153114402Sru * offset within the kernel map. 15475584Sru */ 155104862Sru vm_map_lock(map); 156104862Sru if (vm_map_findspace(map, 0, size, &addr)) { 157104862Sru vm_map_unlock(map); 158114402Sru return (0); 159104862Sru } 160104862Sru offset = addr - VM_MIN_KERNEL_ADDRESS; 16169626Sru vm_object_reference(kernel_object); 16269626Sru vm_map_insert(map, kernel_object, offset, addr, addr + size, 163104862Sru VM_PROT_ALL, VM_PROT_ALL, 0); 164114402Sru vm_map_unlock(map); 165104862Sru 16675584Sru /* 167104862Sru * Guarantee that there are pages already in this object before 168104862Sru * calling vm_map_pageable. This is to prevent the following 169104862Sru * scenario: 170114402Sru * 171104862Sru * 1) Threads have swapped out, so that there is a pager for the 17269626Sru * kernel_object. 2) The kmsg zone is empty, and so we are 17369626Sru * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 17475584Sru * there is no page, but there is a pager, so we call 17575584Sru * pager_data_request. But the kmsg zone is empty, so we must 17675584Sru * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 17775584Sru * we get the data back from the pager, it will be (very stale) 17875584Sru * non-zero data. kmem_alloc is defined to return zero-filled memory. 17975584Sru * 18075584Sru * We're intentionally not activating the pages we allocate to prevent a 18175584Sru * race with page-out. vm_map_pageable will wire the pages. 182104862Sru */ 18375584Sru 184104862Sru for (i = 0; i < size; i += PAGE_SIZE) { 185104862Sru vm_page_t mem; 186104862Sru 18775584Sru while ((mem = vm_page_alloc(kernel_object, 188104862Sru OFF_TO_IDX(offset + i), VM_ALLOC_ZERO)) == NULL) { 189104862Sru VM_WAIT; 19075584Sru } 19175584Sru if ((mem->flags & PG_ZERO) == 0) 192104862Sru vm_page_zero_fill(mem); 193104862Sru mem->flags &= ~(PG_BUSY|PG_ZERO); 194104862Sru mem->valid = VM_PAGE_BITS_ALL; 19575584Sru } 196104862Sru 197104862Sru /* 198104862Sru * And finally, mark the data as non-pageable. 19975584Sru */ 200104862Sru 20175584Sru (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 20275584Sru 20375584Sru /* 20475584Sru * Try to coalesce the map 20575584Sru */ 20675584Sru vm_map_simplify(map, addr); 20775584Sru 208104862Sru return (addr); 209104862Sru} 210104862Sru 211104862Sru/* 212104862Sru * kmem_free: 213104862Sru * 214104862Sru * Release a region of kernel virtual memory allocated 215104862Sru * with kmem_alloc, and return the physical pages 216104862Sru * associated with that region. 217104862Sru */ 218104862Sruvoid 219104862Srukmem_free(map, addr, size) 220104862Sru vm_map_t map; 221104862Sru register vm_offset_t addr; 222104862Sru vm_size_t size; 223104862Sru{ 224104862Sru (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 225104862Sru} 226104862Sru 227104862Sru/* 228104862Sru * kmem_suballoc: 229104862Sru * 230104862Sru * Allocates a map to manage a subrange 231104862Sru * of the kernel virtual address space. 232104862Sru * 233104862Sru * Arguments are as follows: 234104862Sru * 235104862Sru * parent Map to take range from 236104862Sru * size Size of range to find 23775584Sru * min, max Returned endpoints of map 23875584Sru * pageable Can the region be paged 239104862Sru */ 24075584Sruvm_map_t 241104862Srukmem_suballoc(parent, min, max, size, pageable) 24269626Sru register vm_map_t parent; 24369626Sru vm_offset_t *min, *max; 244104862Sru register vm_size_t size; 245104862Sru boolean_t pageable; 246104862Sru{ 247104862Sru register int ret; 248104862Sru vm_map_t result; 249104862Sru 250104862Sru size = round_page(size); 25175584Sru 252104862Sru *min = (vm_offset_t) vm_map_min(parent); 253104862Sru ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 25475584Sru min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 25575584Sru if (ret != KERN_SUCCESS) { 256104862Sru printf("kmem_suballoc: bad status return of %d.\n", ret); 257104862Sru panic("kmem_suballoc"); 258104862Sru } 259104862Sru *max = *min + size; 260104862Sru pmap_reference(vm_map_pmap(parent)); 26175584Sru result = vm_map_create(vm_map_pmap(parent), *min, *max, pageable); 26269626Sru if (result == NULL) 26369626Sru panic("kmem_suballoc: cannot create submap"); 26469626Sru if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 26575584Sru panic("kmem_suballoc: unable to change range to submap"); 26675584Sru return (result); 26775584Sru} 268104862Sru 269104862Sru/* 270104862Sru * Allocate wired-down memory in the kernel's address map for the higher 27169626Sru * level kernel memory allocator (kern/kern_malloc.c). We cannot use 27269626Sru * kmem_alloc() because we may need to allocate memory at interrupt 273104862Sru * level where we cannot block (canwait == FALSE). 274104862Sru * 275104862Sru * This routine has its own private kernel submap (kmem_map) and object 276104862Sru * (kmem_object). This, combined with the fact that only malloc uses 27769626Sru * this routine, ensures that we will never block in map or object waits. 27869626Sru * 279104862Sru * Note that this still only works in a uni-processor environment and 280104862Sru * when called at splhigh(). 281104862Sru * 282104862Sru * We don't worry about expanding the map (adding entries) since entries 283104862Sru * for wired maps are statically allocated. 284104862Sru */ 285104862Sruvm_offset_t 286104862Srukmem_malloc(map, size, waitflag) 287104862Sru register vm_map_t map; 288104862Sru register vm_size_t size; 289104862Sru boolean_t waitflag; 29075584Sru{ 29175584Sru register vm_offset_t offset, i; 29275584Sru vm_map_entry_t entry; 29369626Sru vm_offset_t addr; 29475584Sru vm_page_t m; 29575584Sru 29675584Sru if (map != kmem_map && map != mb_map && map != mcl_map) 29775584Sru panic("kmem_malloc: map != {kmem,mb,mcl}_map"); 29875584Sru 29975584Sru size = round_page(size); 30075584Sru addr = vm_map_min(map); 30175584Sru 30275584Sru /* 30375584Sru * Locate sufficient space in the map. This will give us the final 30475584Sru * virtual address for the new memory, and thus will tell us the 30575584Sru * offset within the kernel map. 30675584Sru */ 30775584Sru vm_map_lock(map); 308104862Sru if (vm_map_findspace(map, 0, size, &addr)) { 309104862Sru vm_map_unlock(map); 310104862Sru if (map == mb_map) { 311104862Sru mb_map_full = TRUE; 312104862Sru log(LOG_ERR, "Out of mbufs - increase maxusers!\n"); 313104862Sru return (0); 314104862Sru } 315104862Sru if (map == mcl_map) { 316104862Sru mcl_map_full = TRUE; 317104862Sru log(LOG_ERR, 318104862Sru "Out of mbuf clusters - increase maxusers!\n"); 319104862Sru return (0); 320104862Sru } 321104862Sru if (waitflag == M_WAITOK) 322104862Sru panic("kmem_malloc: kmem_map too small"); 323104862Sru return (0); 324104862Sru } 325104862Sru offset = addr - VM_MIN_KERNEL_ADDRESS; 326104862Sru vm_object_reference(kmem_object); 327104862Sru vm_map_insert(map, kmem_object, offset, addr, addr + size, 328104862Sru VM_PROT_ALL, VM_PROT_ALL, 0); 329104862Sru 33075584Sru /* 33175584Sru * If we can wait, just mark the range as wired (will fault pages as 33275584Sru * necessary). 333114402Sru */ 33475584Sru if (waitflag == M_WAITOK) { 33575584Sru vm_map_unlock(map); 336104862Sru (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, 337104862Sru FALSE); 338104862Sru vm_map_simplify(map, addr); 33975584Sru return (addr); 34075584Sru } 34175584Sru /* 34275584Sru * If we cannot wait then we must allocate all memory up front, 34375584Sru * pulling it off the active queue to prevent pageout. 34475584Sru */ 34575584Sru for (i = 0; i < size; i += PAGE_SIZE) { 34675584Sru m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), 34775584Sru (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 34875584Sru 34975584Sru /* 35075584Sru * Ran out of space, free everything up and return. Don't need 35175584Sru * to lock page queues here as we know that the pages we got 35275584Sru * aren't on any queues. 35375584Sru */ 35475584Sru if (m == NULL) { 355104862Sru while (i != 0) { 356104862Sru i -= PAGE_SIZE; 357104862Sru m = vm_page_lookup(kmem_object, 358104862Sru OFF_TO_IDX(offset + i)); 35975584Sru vm_page_free(m); 360104862Sru } 36175584Sru vm_map_delete(map, addr, addr + size); 36275584Sru vm_map_unlock(map); 36375584Sru return (0); 364114402Sru } 36555839Sasmodai m->flags &= ~(PG_BUSY|PG_ZERO); 366104862Sru m->valid = VM_PAGE_BITS_ALL; 36755839Sasmodai } 36855839Sasmodai 36955839Sasmodai /* 37055839Sasmodai * Mark map entry as non-pageable. Assert: vm_map_insert() will never 37155839Sasmodai * be able to extend the previous entry so there will be a new entry 37269626Sru * exactly corresponding to this address range and it will have 373114402Sru * wired_count == 0. 374114402Sru */ 375114402Sru if (!vm_map_lookup_entry(map, addr, &entry) || 376104862Sru entry->start != addr || entry->end != addr + size || 37755839Sasmodai entry->wired_count) 37855839Sasmodai panic("kmem_malloc: entry not found or misaligned"); 37955839Sasmodai entry->wired_count++; 380104862Sru 38155839Sasmodai /* 38255839Sasmodai * Loop thru pages, entering them in the pmap. (We cannot add them to 38355839Sasmodai * the wired count without wrapping the vm_page_queue_lock in 384104862Sru * splimp...) 38555839Sasmodai */ 38655839Sasmodai for (i = 0; i < size; i += PAGE_SIZE) { 387104862Sru m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 388104862Sru vm_page_wire(m); 38955839Sasmodai pmap_kenter(addr + i, VM_PAGE_TO_PHYS(m)); 390104862Sru } 39155839Sasmodai vm_map_unlock(map); 39255839Sasmodai 393104862Sru vm_map_simplify(map, addr); 394104862Sru return (addr); 395104862Sru} 396104862Sru 397104862Sru/* 398104862Sru * kmem_alloc_wait 399104862Sru * 40055839Sasmodai * Allocates pageable memory from a sub-map of the kernel. If the submap 40175584Sru * has no room, the caller sleeps waiting for more memory in the submap. 40275584Sru * 40375584Sru */ 40475584Sruvm_offset_t 40575584Srukmem_alloc_wait(map, size) 40675584Sru vm_map_t map; 40775584Sru vm_size_t size; 40875584Sru{ 40975584Sru vm_offset_t addr; 410104862Sru 41175584Sru size = round_page(size); 41275584Sru 41375584Sru for (;;) { 41475584Sru /* 41575584Sru * To make this work for more than one map, use the map's lock 41675584Sru * to lock out sleepers/wakers. 41775584Sru */ 41875584Sru vm_map_lock(map); 41975584Sru if (vm_map_findspace(map, 0, size, &addr) == 0) 42075584Sru break; 42155839Sasmodai /* no space now; see if we can ever get space */ 42255839Sasmodai if (vm_map_max(map) - vm_map_min(map) < size) { 42355839Sasmodai vm_map_unlock(map); 42455839Sasmodai return (0); 42569626Sru } 42669626Sru vm_map_unlock(map); 42769626Sru tsleep(map, PVM, "kmaw", 0); 428104862Sru } 42955839Sasmodai vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 43055839Sasmodai vm_map_unlock(map); 43155839Sasmodai return (addr); 43255839Sasmodai} 43355839Sasmodai 434114402Sru/* 43569626Sru * kmem_free_wakeup 43655839Sasmodai * 43755839Sasmodai * Returns memory to a submap of the kernel, and wakes up any processes 43875584Sru * waiting for memory in that map. 43975584Sru */ 44075584Sruvoid 44175584Srukmem_free_wakeup(map, addr, size) 44275584Sru vm_map_t map; 44375584Sru vm_offset_t addr; 44475584Sru vm_size_t size; 44555839Sasmodai{ 44655839Sasmodai vm_map_lock(map); 44769626Sru (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 44869626Sru wakeup(map); 44969626Sru vm_map_unlock(map); 45055839Sasmodai} 45155839Sasmodai 45255839Sasmodai/* 45355839Sasmodai * Create the kernel map; insert a mapping covering kernel text, data, bss, 45455839Sasmodai * and all space allocated thus far (`boostrap' data). The new map will thus 45569626Sru * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 45669626Sru * the range between `start' and `end' as free. 45769626Sru */ 45869626Sruvoid 45969626Srukmem_init(start, end) 46069626Sru vm_offset_t start, end; 46169626Sru{ 46269626Sru register vm_map_t m; 46355839Sasmodai 46455839Sasmodai m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end, FALSE); 46575584Sru vm_map_lock(m); 46655839Sasmodai /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 46775584Sru kernel_map = m; 46855839Sasmodai (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 46969626Sru VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 47069626Sru /* ... and ending with the completion of the above `insert' */ 47155839Sasmodai vm_map_unlock(m); 47269626Sru} 47355839Sasmodai