sys/vm/vm_kern.c

   1 /*
   2  * (MPSAFE)
   3  *
   4  * Copyright (c) 1991, 1993
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * This code is derived from software contributed to Berkeley by
   8  * The Mach Operating System project at Carnegie-Mellon University.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  * 3. All advertising materials mentioning features or use of this software
  19  *    must display the following acknowledgement:
  20  *      This product includes software developed by the University of
  21  *      California, Berkeley and its contributors.
  22  * 4. Neither the name of the University nor the names of its contributors
  23  *    may be used to endorse or promote products derived from this software
  24  *    without specific prior written permission.
  25  *
  26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  36  * SUCH DAMAGE.
  37  *
  38  *      from: @(#)vm_kern.c     8.3 (Berkeley) 1/12/94
  39  *
  40  *
  41  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
  42  * All rights reserved.
  43  *
  44  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
  45  *
  46  * Permission to use, copy, modify and distribute this software and
  47  * its documentation is hereby granted, provided that both the copyright
  48  * notice and this permission notice appear in all copies of the
  49  * software, derivative works or modified versions, and any portions
  50  * thereof, and that both notices appear in supporting documentation.
  51  *
  52  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  53  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
  54  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  55  *
  56  * Carnegie Mellon requests users of this software to return to
  57  *
  58  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  59  *  School of Computer Science
  60  *  Carnegie Mellon University
  61  *  Pittsburgh PA 15213-3890
  62  *
  63  * any improvements or extensions that they make and grant Carnegie the
  64  * rights to redistribute these changes.
  65  *
  66  * $FreeBSD: src/sys/vm/vm_kern.c,v 1.61.2.2 2002/03/12 18:25:26 tegge Exp $
  67  */
  68
  69 /*
  70  *      Kernel memory management.
  71  */
  72
  73 #include <sys/param.h>
  74 #include <sys/systm.h>
  75 #include <sys/proc.h>
  76 #include <sys/malloc.h>
  77 #include <sys/kernel.h>
  78 #include <sys/sysctl.h>
  79
  80 #include <vm/vm.h>
  81 #include <vm/vm_param.h>
  82 #include <sys/lock.h>
  83 #include <vm/pmap.h>
  84 #include <vm/vm_map.h>
  85 #include <vm/vm_object.h>
  86 #include <vm/vm_page.h>
  87 #include <vm/vm_pageout.h>
  88 #include <vm/vm_kern.h>
  89 #include <vm/vm_extern.h>
  90
  91 struct vm_map kernel_map;
  92 struct vm_map clean_map;
  93 struct vm_map buffer_map;
  94
  95 /*
  96  * Allocate pageable memory to the kernel's address map.  "map" must
  97  * be kernel_map or a submap of kernel_map.
  98  *
  99  * No requirements.
 100  */
 101 vm_offset_t
 102 kmem_alloc_pageable(vm_map_t map, vm_size_t size)
 103 {
 104         vm_offset_t addr;
 105         int result;
 106
 107         size = round_page(size);
 108         addr = vm_map_min(map);
 109         result = vm_map_find(map, NULL, (vm_offset_t) 0,
 110                              &addr, size, PAGE_SIZE,
 111                              TRUE, VM_MAPTYPE_NORMAL,
 112                              VM_PROT_ALL, VM_PROT_ALL,
 113                              0);
 114         if (result != KERN_SUCCESS)
 115                 return (0);
 116         return (addr);
 117 }
 118
 119 /*
 120  * Same as kmem_alloc_pageable, except that it create a nofault entry.
 121  *
 122  * No requirements.
 123  */
 124 vm_offset_t
 125 kmem_alloc_nofault(vm_map_t map, vm_size_t size, vm_size_t align)
 126 {
 127         vm_offset_t addr;
 128         int result;
 129
 130         size = round_page(size);
 131         addr = vm_map_min(map);
 132         result = vm_map_find(map, NULL, (vm_offset_t) 0,
 133                              &addr, size, align,
 134                              TRUE, VM_MAPTYPE_NORMAL,
 135                              VM_PROT_ALL, VM_PROT_ALL,
 136                              MAP_NOFAULT);
 137         if (result != KERN_SUCCESS)
 138                 return (0);
 139         return (addr);
 140 }
 141
 142 /*
 143  * Allocate wired-down memory in the kernel's address map or a submap.
 144  *
 145  * No requirements.
 146  */
 147 vm_offset_t
 148 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags)
 149 {
 150         vm_offset_t addr;
 151         vm_offset_t gstart;
 152         vm_offset_t i;
 153         int count;
 154         int cow;
 155
 156         size = round_page(size);
 157
 158         if (kmflags & KM_KRESERVE)
 159                 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
 160         else
 161                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 162
 163         if (kmflags & KM_STACK) {
 164                 cow = MAP_IS_KSTACK;
 165                 gstart = PAGE_SIZE;
 166         } else {
 167                 cow = 0;
 168                 gstart = 0;
 169         }
 170
 171         /*
 172          * Use the kernel object for wired-down kernel pages. Assume that no
 173          * region of the kernel object is referenced more than once.
 174          *
 175          * Locate sufficient space in the map.  This will give us the final
 176          * virtual address for the new memory, and thus will tell us the
 177          * offset within the kernel map.
 178          */
 179         vm_map_lock(map);
 180         if (vm_map_findspace(map, vm_map_min(map), size, PAGE_SIZE, 0, &addr)) {
 181                 vm_map_unlock(map);
 182                 if (kmflags & KM_KRESERVE)
 183                         vm_map_entry_krelease(count);
 184                 else
 185                         vm_map_entry_release(count);
 186                 return (0);
 187         }
 188         vm_object_hold(&kernel_object);
 189         vm_object_reference_locked(&kernel_object);
 190         vm_map_insert(map, &count,
 191                       &kernel_object, addr, addr, addr + size,
 192                       VM_MAPTYPE_NORMAL,
 193                       VM_PROT_ALL, VM_PROT_ALL,
 194                       cow);
 195         vm_object_drop(&kernel_object);
 196
 197         vm_map_unlock(map);
 198         if (kmflags & KM_KRESERVE)
 199                 vm_map_entry_krelease(count);
 200         else
 201                 vm_map_entry_release(count);
 202
 203         /*
 204          * Guarantee that there are pages already in this object before
 205          * calling vm_map_wire.  This is to prevent the following
 206          * scenario:
 207          *
 208          * 1) Threads have swapped out, so that there is a pager for the
 209          * kernel_object. 2) The kmsg zone is empty, and so we are
 210          * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
 211          * there is no page, but there is a pager, so we call
 212          * pager_data_request.  But the kmsg zone is empty, so we must
 213          * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
 214          * we get the data back from the pager, it will be (very stale)
 215          * non-zero data.  kmem_alloc is defined to return zero-filled memory.
 216          *
 217          * We're intentionally not activating the pages we allocate to prevent a
 218          * race with page-out.  vm_map_wire will wire the pages.
 219          */
 220         vm_object_hold(&kernel_object);
 221         for (i = gstart; i < size; i += PAGE_SIZE) {
 222                 vm_page_t mem;
 223
 224                 mem = vm_page_grab(&kernel_object, OFF_TO_IDX(addr + i),
 225                                    VM_ALLOC_FORCE_ZERO | VM_ALLOC_NORMAL |
 226                                    VM_ALLOC_RETRY);
 227                 vm_page_unqueue_nowakeup(mem);
 228                 vm_page_wakeup(mem);
 229         }
 230         vm_object_drop(&kernel_object);
 231
 232         /*
 233          * And finally, mark the data as non-pageable.
 234          *
 235          * NOTE: vm_map_wire() handles any kstack guard.
 236          */
 237         vm_map_wire(map, addr, addr + size, kmflags);
 238
 239         return (addr);
 240 }
 241
 242 /*
 243  * Release a region of kernel virtual memory allocated with kmem_alloc,
 244  * and return the physical pages associated with that region.
 245  *
 246  * WARNING!  If the caller entered pages into the region using pmap_kenter()
 247  * it must remove the pages using pmap_kremove[_quick]() before freeing the
 248  * underlying kmem, otherwise resident_count will be mistabulated.
 249  *
 250  * No requirements.
 251  */
 252 void
 253 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size)
 254 {
 255         vm_map_remove(map, trunc_page(addr), round_page(addr + size));
 256 }
 257
 258 /*
 259  * Used to break a system map into smaller maps, usually to reduce
 260  * contention and to provide large KVA spaces for subsystems like the
 261  * buffer cache.
 262  *
 263  *      parent          Map to take range from
 264  *      result
 265  *      size            Size of range to find
 266  *      min, max        Returned endpoints of map
 267  *      pageable        Can the region be paged
 268  *
 269  * No requirements.
 270  */
 271 void
 272 kmem_suballoc(vm_map_t parent, vm_map_t result,
 273               vm_offset_t *min, vm_offset_t *max, vm_size_t size)
 274 {
 275         int ret;
 276
 277         size = round_page(size);
 278
 279         *min = (vm_offset_t) vm_map_min(parent);
 280         ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
 281                           min, size, PAGE_SIZE,
 282                           TRUE, VM_MAPTYPE_UNSPECIFIED,
 283                           VM_PROT_ALL, VM_PROT_ALL,
 284                           0);
 285         if (ret != KERN_SUCCESS) {
 286                 kprintf("kmem_suballoc: bad status return of %d.\n", ret);
 287                 panic("kmem_suballoc");
 288         }
 289         *max = *min + size;
 290         pmap_reference(vm_map_pmap(parent));
 291         vm_map_init(result, *min, *max, vm_map_pmap(parent));
 292         if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
 293                 panic("kmem_suballoc: unable to change range to submap");
 294 }
 295
 296 /*
 297  * Allocates pageable memory from a sub-map of the kernel.  If the submap
 298  * has no room, the caller sleeps waiting for more memory in the submap.
 299  *
 300  * No requirements.
 301  */
 302 vm_offset_t
 303 kmem_alloc_wait(vm_map_t map, vm_size_t size)
 304 {
 305         vm_offset_t addr;
 306         int count;
 307
 308         size = round_page(size);
 309
 310         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 311
 312         for (;;) {
 313                 /*
 314                  * To make this work for more than one map, use the map's lock
 315                  * to lock out sleepers/wakers.
 316                  */
 317                 vm_map_lock(map);
 318                 if (vm_map_findspace(map, vm_map_min(map),
 319                                      size, PAGE_SIZE, 0, &addr) == 0) {
 320                         break;
 321                 }
 322                 /* no space now; see if we can ever get space */
 323                 if (vm_map_max(map) - vm_map_min(map) < size) {
 324                         vm_map_entry_release(count);
 325                         vm_map_unlock(map);
 326                         return (0);
 327                 }
 328                 vm_map_unlock(map);
 329                 tsleep(map, 0, "kmaw", 0);
 330         }
 331         vm_map_insert(map, &count,
 332                       NULL, (vm_offset_t) 0,
 333                       addr, addr + size,
 334                       VM_MAPTYPE_NORMAL,
 335                       VM_PROT_ALL, VM_PROT_ALL,
 336                       0);
 337         vm_map_unlock(map);
 338         vm_map_entry_release(count);
 339
 340         return (addr);
 341 }
 342
 343 /*
 344  * Returns memory to a submap of the kernel, and wakes up any processes
 345  * waiting for memory in that map.
 346  *
 347  * No requirements.
 348  */
 349 void
 350 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
 351 {
 352         int count;
 353
 354         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 355         vm_map_lock(map);
 356         vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count);
 357         wakeup(map);
 358         vm_map_unlock(map);
 359         vm_map_entry_release(count);
 360 }
 361
 362 /*
 363  * Create the kernel_ma for (KvaStart,KvaEnd) and insert mappings to
 364  * cover areas already allocated or reserved thus far.
 365  *
 366  * The areas (virtual_start, virtual_end) and (virtual2_start, virtual2_end)
 367  * are available so the cutouts are the areas around these ranges between
 368  * KvaStart and KvaEnd.
 369  *
 370  * Depend on the zalloc bootstrap cache to get our vm_map_entry_t.
 371  * Called from the low level boot code only.
 372  */
 373 void
 374 kmem_init(void)
 375 {
 376         vm_offset_t addr;
 377         vm_map_t m;
 378         int count;
 379
 380         m = vm_map_create(&kernel_map, &kernel_pmap, KvaStart, KvaEnd);
 381         vm_map_lock(m);
 382         /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
 383         m->system_map = 1;
 384         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 385         addr = KvaStart;
 386         if (virtual2_start) {
 387                 if (addr < virtual2_start) {
 388                         vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 389                                       addr, virtual2_start,
 390                                       VM_MAPTYPE_NORMAL,
 391                                       VM_PROT_ALL, VM_PROT_ALL,
 392                                       0);
 393                 }
 394                 addr = virtual2_end;
 395         }
 396         if (addr < virtual_start) {
 397                 vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 398                               addr, virtual_start,
 399                               VM_MAPTYPE_NORMAL,
 400                               VM_PROT_ALL, VM_PROT_ALL,
 401                               0);
 402         }
 403         addr = virtual_end;
 404         if (addr < KvaEnd) {
 405                 vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 406                               addr, KvaEnd,
 407                               VM_MAPTYPE_NORMAL,
 408                               VM_PROT_ALL, VM_PROT_ALL,
 409                               0);
 410         }
 411         /* ... and ending with the completion of the above `insert' */
 412         vm_map_unlock(m);
 413         vm_map_entry_release(count);
 414 }
 415
 416 /*
 417  * No requirements.
 418  */
 419 static int
 420 kvm_size(SYSCTL_HANDLER_ARGS)
 421 {
 422         unsigned long ksize = KvaSize;
 423
 424         return sysctl_handle_long(oidp, &ksize, 0, req);
 425 }
 426 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_ULONG|CTLFLAG_RD,
 427     0, 0, kvm_size, "LU", "Size of KVM");
 428
 429 /*
 430  * No requirements.
 431  */
 432 static int
 433 kvm_free(SYSCTL_HANDLER_ARGS)
 434 {
 435         unsigned long kfree = virtual_end - kernel_vm_end;
 436
 437         return sysctl_handle_long(oidp, &kfree, 0, req);
 438 }
 439 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_ULONG|CTLFLAG_RD,
 440     0, 0, kvm_free, "LU", "Amount of KVM free");
 441