vm_kern.c revision 38799
168349Sobrien/* 2133359Sobrien * Copyright (c) 1991, 1993 3133359Sobrien * The Regents of the University of California. All rights reserved. 4133359Sobrien * 5133359Sobrien * This code is derived from software contributed to Berkeley by 6133359Sobrien * The Mach Operating System project at Carnegie-Mellon University. 7133359Sobrien * 8133359Sobrien * Redistribution and use in source and binary forms, with or without 9133359Sobrien * modification, are permitted provided that the following conditions 10133359Sobrien * are met: 11133359Sobrien * 1. Redistributions of source code must retain the above copyright 12133359Sobrien * notice, this list of conditions and the following disclaimer. 13133359Sobrien * 2. Redistributions in binary form must reproduce the above copyright 14133359Sobrien * notice, this list of conditions and the following disclaimer in the 15133359Sobrien * documentation and/or other materials provided with the distribution. 16133359Sobrien * 3. All advertising materials mentioning features or use of this software 17133359Sobrien * must display the following acknowledgement: 18133359Sobrien * This product includes software developed by the University of 19133359Sobrien * California, Berkeley and its contributors. 20133359Sobrien * 4. Neither the name of the University nor the names of its contributors 21133359Sobrien * may be used to endorse or promote products derived from this software 22133359Sobrien * without specific prior written permission. 23133359Sobrien * 24133359Sobrien * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25133359Sobrien * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26133359Sobrien * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27133359Sobrien * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28133359Sobrien * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2968349Sobrien * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 3068349Sobrien * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 3168349Sobrien * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 3268349Sobrien * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3368349Sobrien * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3468349Sobrien * SUCH DAMAGE. 35226048Sobrien * 3668349Sobrien * from: @(#)vm_kern.c 8.3 (Berkeley) 1/12/94 37103373Sobrien * 38103373Sobrien * 39103373Sobrien * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40103373Sobrien * All rights reserved. 4168349Sobrien * 4268349Sobrien * Authors: Avadis Tevanian, Jr., Michael Wayne Young 4368349Sobrien * 4468349Sobrien * Permission to use, copy, modify and distribute this software and 4568349Sobrien * its documentation is hereby granted, provided that both the copyright 4668349Sobrien * notice and this permission notice appear in all copies of the 4768349Sobrien * software, derivative works or modified versions, and any portions 4868349Sobrien * thereof, and that both notices appear in supporting documentation. 4968349Sobrien * 5068349Sobrien * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 5168349Sobrien * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 5268349Sobrien * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 5368349Sobrien * 54226048Sobrien * Carnegie Mellon requests users of this software to return to 5568349Sobrien * 5668349Sobrien * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57330569Sgordon * School of Computer Science 58330569Sgordon * Carnegie Mellon University 59330569Sgordon * Pittsburgh PA 15213-3890 60330569Sgordon * 61330569Sgordon * any improvements or extensions that they make and grant Carnegie the 62330569Sgordon * rights to redistribute these changes. 63330569Sgordon * 64330569Sgordon * $Id: vm_kern.c,v 1.49 1998/08/24 08:39:37 dfr Exp $ 65330569Sgordon */ 66330569Sgordon 67330569Sgordon/* 68330569Sgordon * Kernel memory management. 69330569Sgordon */ 70330569Sgordon 71330569Sgordon#include <sys/param.h> 72330569Sgordon#include <sys/systm.h> 73330569Sgordon#include <sys/proc.h> 74330569Sgordon#include <sys/malloc.h> 75330569Sgordon 76330569Sgordon#include <vm/vm.h> 77330569Sgordon#include <vm/vm_param.h> 78330569Sgordon#include <vm/vm_prot.h> 79330569Sgordon#include <sys/lock.h> 80330569Sgordon#include <vm/pmap.h> 81330569Sgordon#include <vm/vm_map.h> 82330569Sgordon#include <vm/vm_object.h> 83330569Sgordon#include <vm/vm_page.h> 84330569Sgordon#include <vm/vm_pageout.h> 85330569Sgordon#include <vm/vm_extern.h> 86330569Sgordon 87330569Sgordonvm_map_t kernel_map=0; 88330569Sgordonvm_map_t kmem_map=0; 89330569Sgordonvm_map_t exec_map=0; 90330569Sgordonvm_map_t clean_map=0; 91330569Sgordonvm_map_t u_map=0; 92330569Sgordonvm_map_t buffer_map=0; 9368349Sobrienvm_map_t mb_map=0; 9468349Sobrienint mb_map_full=0; 9568349Sobrienvm_map_t io_map=0; 9668349Sobrienvm_map_t phys_map=0; 9768349Sobrien 9868349Sobrien/* 9968349Sobrien * kmem_alloc_pageable: 10068349Sobrien * 10168349Sobrien * Allocate pageable memory to the kernel's address map. 10268349Sobrien * "map" must be kernel_map or a submap of kernel_map. 10368349Sobrien */ 10468349Sobrien 10568349Sobrienvm_offset_t 10668349Sobrienkmem_alloc_pageable(map, size) 10768349Sobrien vm_map_t map; 10868349Sobrien register vm_size_t size; 10968349Sobrien{ 11068349Sobrien vm_offset_t addr; 11168349Sobrien register int result; 11268349Sobrien 11368349Sobrien size = round_page(size); 11468349Sobrien addr = vm_map_min(map); 11568349Sobrien result = vm_map_find(map, NULL, (vm_offset_t) 0, 11668349Sobrien &addr, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 11768349Sobrien if (result != KERN_SUCCESS) { 11868349Sobrien return (0); 11968349Sobrien } 12068349Sobrien return (addr); 12168349Sobrien} 12268349Sobrien 12368349Sobrien/* 12468349Sobrien * Allocate wired-down memory in the kernel's address map 12568349Sobrien * or a submap. 12668349Sobrien */ 127226048Sobrienvm_offset_t 128226048Sobrienkmem_alloc(map, size) 129226048Sobrien register vm_map_t map; 130226048Sobrien register vm_size_t size; 13168349Sobrien{ 132186690Sobrien vm_offset_t addr; 133186690Sobrien register vm_offset_t offset; 134186690Sobrien vm_offset_t i; 135186690Sobrien 136186690Sobrien size = round_page(size); 137186690Sobrien 138186690Sobrien /* 139186690Sobrien * Use the kernel object for wired-down kernel pages. Assume that no 14068349Sobrien * region of the kernel object is referenced more than once. 141226048Sobrien */ 142226048Sobrien 143226048Sobrien /* 144330569Sgordon * Locate sufficient space in the map. This will give us the final 14568349Sobrien * virtual address for the new memory, and thus will tell us the 14668349Sobrien * offset within the kernel map. 147226048Sobrien */ 148226048Sobrien vm_map_lock(map); 149226048Sobrien if (vm_map_findspace(map, vm_map_min(map), size, &addr)) { 15068349Sobrien vm_map_unlock(map); 15168349Sobrien return (0); 152226048Sobrien } 153226048Sobrien offset = addr - VM_MIN_KERNEL_ADDRESS; 154226048Sobrien vm_object_reference(kernel_object); 15568349Sobrien vm_map_insert(map, kernel_object, offset, addr, addr + size, 15668349Sobrien VM_PROT_ALL, VM_PROT_ALL, 0); 15768349Sobrien vm_map_unlock(map); 15868349Sobrien 15968349Sobrien /* 16068349Sobrien * Guarantee that there are pages already in this object before 16168349Sobrien * calling vm_map_pageable. This is to prevent the following 16268349Sobrien * scenario: 16368349Sobrien * 16468349Sobrien * 1) Threads have swapped out, so that there is a pager for the 16568349Sobrien * kernel_object. 2) The kmsg zone is empty, and so we are 16668349Sobrien * kmem_allocing a new page for it. 3) vm_map_pageable calls vm_fault; 16768349Sobrien * there is no page, but there is a pager, so we call 16868349Sobrien * pager_data_request. But the kmsg zone is empty, so we must 16968349Sobrien * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when 17068349Sobrien * we get the data back from the pager, it will be (very stale) 17168349Sobrien * non-zero data. kmem_alloc is defined to return zero-filled memory. 17268349Sobrien * 17368349Sobrien * We're intentionally not activating the pages we allocate to prevent a 17468349Sobrien * race with page-out. vm_map_pageable will wire the pages. 17568349Sobrien */ 17668349Sobrien 17768349Sobrien for (i = 0; i < size; i += PAGE_SIZE) { 17868349Sobrien vm_page_t mem; 17968349Sobrien 18068349Sobrien mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i), 18168349Sobrien VM_ALLOC_ZERO | VM_ALLOC_RETRY); 18268349Sobrien if ((mem->flags & PG_ZERO) == 0) 18368349Sobrien vm_page_zero_fill(mem); 18468349Sobrien vm_page_flag_clear(mem, (PG_BUSY | PG_ZERO)); 18568349Sobrien mem->valid = VM_PAGE_BITS_ALL; 18668349Sobrien } 18768349Sobrien 18868349Sobrien /* 18968349Sobrien * And finally, mark the data as non-pageable. 19068349Sobrien */ 19168349Sobrien 19268349Sobrien (void) vm_map_pageable(map, (vm_offset_t) addr, addr + size, FALSE); 19368349Sobrien 19468349Sobrien return (addr); 19568349Sobrien} 19668349Sobrien 19768349Sobrien/* 19868349Sobrien * kmem_free: 19968349Sobrien * 20068349Sobrien * Release a region of kernel virtual memory allocated 20168349Sobrien * with kmem_alloc, and return the physical pages 20268349Sobrien * associated with that region. 20368349Sobrien */ 20468349Sobrienvoid 20568349Sobrienkmem_free(map, addr, size) 20668349Sobrien vm_map_t map; 20768349Sobrien register vm_offset_t addr; 20868349Sobrien vm_size_t size; 20968349Sobrien{ 21068349Sobrien (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size)); 21168349Sobrien} 21268349Sobrien 21368349Sobrien/* 21468349Sobrien * kmem_suballoc: 21568349Sobrien * 21668349Sobrien * Allocates a map to manage a subrange 21768349Sobrien * of the kernel virtual address space. 21868349Sobrien * 21968349Sobrien * Arguments are as follows: 22068349Sobrien * 22168349Sobrien * parent Map to take range from 22268349Sobrien * size Size of range to find 22368349Sobrien * min, max Returned endpoints of map 22468349Sobrien * pageable Can the region be paged 22568349Sobrien */ 22668349Sobrienvm_map_t 22768349Sobrienkmem_suballoc(parent, min, max, size) 22868349Sobrien register vm_map_t parent; 22968349Sobrien vm_offset_t *min, *max; 23068349Sobrien register vm_size_t size; 23168349Sobrien{ 232103373Sobrien register int ret; 233330569Sgordon vm_map_t result; 234103373Sobrien 235330569Sgordon size = round_page(size); 236330569Sgordon 237330569Sgordon *min = (vm_offset_t) vm_map_min(parent); 238330569Sgordon ret = vm_map_find(parent, NULL, (vm_offset_t) 0, 239330569Sgordon min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0); 240330569Sgordon if (ret != KERN_SUCCESS) { 241330569Sgordon printf("kmem_suballoc: bad status return of %d.\n", ret); 242330569Sgordon panic("kmem_suballoc"); 243330569Sgordon } 244330569Sgordon *max = *min + size; 245330569Sgordon pmap_reference(vm_map_pmap(parent)); 246330569Sgordon result = vm_map_create(vm_map_pmap(parent), *min, *max); 247330569Sgordon if (result == NULL) 248330569Sgordon panic("kmem_suballoc: cannot create submap"); 249330569Sgordon if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS) 250330569Sgordon panic("kmem_suballoc: unable to change range to submap"); 251330569Sgordon return (result); 252330569Sgordon} 253330569Sgordon 254330569Sgordon/* 255330569Sgordon * Allocate wired-down memory in the kernel's address map for the higher 256330569Sgordon * level kernel memory allocator (kern/kern_malloc.c). We cannot use 257330569Sgordon * kmem_alloc() because we may need to allocate memory at interrupt 258330569Sgordon * level where we cannot block (canwait == FALSE). 259330569Sgordon * 260330569Sgordon * This routine has its own private kernel submap (kmem_map) and object 26168349Sobrien * (kmem_object). This, combined with the fact that only malloc uses 26268349Sobrien * this routine, ensures that we will never block in map or object waits. 26368349Sobrien * 26468349Sobrien * Note that this still only works in a uni-processor environment and 26568349Sobrien * when called at splhigh(). 26668349Sobrien * 26768349Sobrien * We don't worry about expanding the map (adding entries) since entries 26868349Sobrien * for wired maps are statically allocated. 26968349Sobrien */ 27068349Sobrienvm_offset_t 27168349Sobrienkmem_malloc(map, size, waitflag) 27268349Sobrien register vm_map_t map; 27368349Sobrien register vm_size_t size; 274175296Sobrien boolean_t waitflag; 27568349Sobrien{ 27668349Sobrien register vm_offset_t offset, i; 27768349Sobrien vm_map_entry_t entry; 27868349Sobrien vm_offset_t addr; 279226048Sobrien vm_page_t m; 28068349Sobrien 28168349Sobrien if (map != kmem_map && map != mb_map) 28268349Sobrien panic("kmem_malloc: map != {kmem,mb}_map"); 283103373Sobrien 284103373Sobrien size = round_page(size); 285226048Sobrien addr = vm_map_min(map); 286226048Sobrien 287226048Sobrien /* 288226048Sobrien * Locate sufficient space in the map. This will give us the final 289226048Sobrien * virtual address for the new memory, and thus will tell us the 290226048Sobrien * offset within the kernel map. 291226048Sobrien */ 292226048Sobrien vm_map_lock(map); 293226048Sobrien if (vm_map_findspace(map, vm_map_min(map), size, &addr)) { 294226048Sobrien vm_map_unlock(map); 295226048Sobrien if (map == mb_map) { 296226048Sobrien mb_map_full = TRUE; 297226048Sobrien printf("Out of mbuf clusters - adjust NMBCLUSTERS or increase maxusers!\n"); 298103373Sobrien return (0); 299103373Sobrien } 300226048Sobrien if (waitflag == M_WAITOK) 301226048Sobrien panic("kmem_malloc(%d): kmem_map too small: %d total allocated", 302226048Sobrien size, map->size); 303226048Sobrien return (0); 304226048Sobrien } 305103373Sobrien offset = addr - VM_MIN_KERNEL_ADDRESS; 306226048Sobrien vm_object_reference(kmem_object); 307226048Sobrien vm_map_insert(map, kmem_object, offset, addr, addr + size, 308226048Sobrien VM_PROT_ALL, VM_PROT_ALL, 0); 309226048Sobrien 310226048Sobrien for (i = 0; i < size; i += PAGE_SIZE) { 311226048Sobrienretry: 312226048Sobrien m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), 313226048Sobrien (waitflag == M_NOWAIT) ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM); 314226048Sobrien 315226048Sobrien /* 316226048Sobrien * Ran out of space, free everything up and return. Don't need 317226048Sobrien * to lock page queues here as we know that the pages we got 318226048Sobrien * aren't on any queues. 319226048Sobrien */ 320226048Sobrien if (m == NULL) { 321267843Sdelphij if (waitflag == M_WAITOK) { 322267843Sdelphij VM_WAIT; 323267843Sdelphij goto retry; 324267843Sdelphij } 325267843Sdelphij while (i != 0) { 326267843Sdelphij i -= PAGE_SIZE; 327267843Sdelphij m = vm_page_lookup(kmem_object, 328267843Sdelphij OFF_TO_IDX(offset + i)); 329267843Sdelphij vm_page_free(m); 330267843Sdelphij } 331267843Sdelphij vm_map_delete(map, addr, addr + size); 332267843Sdelphij vm_map_unlock(map); 333267843Sdelphij return (0); 334267843Sdelphij } 335267843Sdelphij vm_page_flag_clear(m, PG_ZERO); 336267843Sdelphij m->valid = VM_PAGE_BITS_ALL; 337267843Sdelphij } 338267843Sdelphij 339267843Sdelphij /* 340267843Sdelphij * Mark map entry as non-pageable. Assert: vm_map_insert() will never 341267843Sdelphij * be able to extend the previous entry so there will be a new entry 342267843Sdelphij * exactly corresponding to this address range and it will have 343267843Sdelphij * wired_count == 0. 344267843Sdelphij */ 345267843Sdelphij if (!vm_map_lookup_entry(map, addr, &entry) || 346267843Sdelphij entry->start != addr || entry->end != addr + size || 347267843Sdelphij entry->wired_count) 348267843Sdelphij panic("kmem_malloc: entry not found or misaligned"); 349267843Sdelphij entry->wired_count++; 350267843Sdelphij 351267843Sdelphij vm_map_simplify_entry(map, entry); 352267843Sdelphij 353267843Sdelphij /* 354267843Sdelphij * Loop thru pages, entering them in the pmap. (We cannot add them to 355267843Sdelphij * the wired count without wrapping the vm_page_queue_lock in 356267843Sdelphij * splimp...) 357267843Sdelphij */ 358330569Sgordon for (i = 0; i < size; i += PAGE_SIZE) { 359330569Sgordon m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i)); 360330569Sgordon vm_page_wire(m); 361330569Sgordon vm_page_wakeup(m); 362330569Sgordon pmap_enter(kernel_pmap, addr + i, VM_PAGE_TO_PHYS(m), 363267843Sdelphij VM_PROT_ALL, 1); 364267843Sdelphij vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE | PG_REFERENCED); 365267843Sdelphij } 366186690Sobrien vm_map_unlock(map); 367186690Sobrien 368186690Sobrien return (addr); 369186690Sobrien} 370186690Sobrien 371186690Sobrien/* 372186690Sobrien * kmem_alloc_wait 373186690Sobrien * 374186690Sobrien * Allocates pageable memory from a sub-map of the kernel. If the submap 375186690Sobrien * has no room, the caller sleeps waiting for more memory in the submap. 376186690Sobrien * 377186690Sobrien */ 378186690Sobrienvm_offset_t 379186690Sobrienkmem_alloc_wait(map, size) 380186690Sobrien vm_map_t map; 381186690Sobrien vm_size_t size; 382186690Sobrien{ 383186690Sobrien vm_offset_t addr; 384186690Sobrien 385186690Sobrien size = round_page(size); 386186690Sobrien 387186690Sobrien for (;;) { 388186690Sobrien /* 389186690Sobrien * To make this work for more than one map, use the map's lock 390186690Sobrien * to lock out sleepers/wakers. 391186690Sobrien */ 392186690Sobrien vm_map_lock(map); 393186690Sobrien if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0) 394186690Sobrien break; 395186690Sobrien /* no space now; see if we can ever get space */ 396186690Sobrien if (vm_map_max(map) - vm_map_min(map) < size) { 397186690Sobrien vm_map_unlock(map); 398186690Sobrien return (0); 399186690Sobrien } 400186690Sobrien vm_map_unlock(map); 401186690Sobrien tsleep(map, PVM, "kmaw", 0); 402186690Sobrien } 403186690Sobrien vm_map_insert(map, NULL, (vm_offset_t) 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0); 404186690Sobrien vm_map_unlock(map); 405186690Sobrien return (addr); 406186690Sobrien} 407186690Sobrien 408186690Sobrien/* 409186690Sobrien * kmem_free_wakeup 410186690Sobrien * 411186690Sobrien * Returns memory to a submap of the kernel, and wakes up any processes 412186690Sobrien * waiting for memory in that map. 413186690Sobrien */ 414186690Sobrienvoid 415186690Sobrienkmem_free_wakeup(map, addr, size) 416186690Sobrien vm_map_t map; 417186690Sobrien vm_offset_t addr; 418186690Sobrien vm_size_t size; 419186690Sobrien{ 420186690Sobrien vm_map_lock(map); 421186690Sobrien (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size)); 422186690Sobrien wakeup(map); 423186690Sobrien vm_map_unlock(map); 424186690Sobrien} 425186690Sobrien 426186690Sobrien/* 427186690Sobrien * Create the kernel map; insert a mapping covering kernel text, data, bss, 428186690Sobrien * and all space allocated thus far (`boostrap' data). The new map will thus 429186690Sobrien * map the range between VM_MIN_KERNEL_ADDRESS and `start' as allocated, and 430186690Sobrien * the range between `start' and `end' as free. 431186690Sobrien */ 432186690Sobrienvoid 43368349Sobrienkmem_init(start, end) 434 vm_offset_t start, end; 435{ 436 register vm_map_t m; 437 438 m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end); 439 vm_map_lock(m); 440 /* N.B.: cannot use kgdb to debug, starting with this assignment ... */ 441 kernel_map = m; 442 kernel_map->system_map = 1; 443 (void) vm_map_insert(m, NULL, (vm_offset_t) 0, 444 VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL, 0); 445 /* ... and ending with the completion of the above `insert' */ 446 vm_map_unlock(m); 447} 448