sys/vm/vm_kern.c

   1 /*
   2  * Copyright (c) 1991, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * The Mach Operating System project at Carnegie-Mellon University.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      from: @(#)vm_kern.c     8.3 (Berkeley) 1/12/94
  37  *
  38  *
  39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
  40  * All rights reserved.
  41  *
  42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
  43  *
  44  * Permission to use, copy, modify and distribute this software and
  45  * its documentation is hereby granted, provided that both the copyright
  46  * notice and this permission notice appear in all copies of the
  47  * software, derivative works or modified versions, and any portions
  48  * thereof, and that both notices appear in supporting documentation.
  49  *
  50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
  52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  53  *
  54  * Carnegie Mellon requests users of this software to return to
  55  *
  56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  57  *  School of Computer Science
  58  *  Carnegie Mellon University
  59  *  Pittsburgh PA 15213-3890
  60  *
  61  * any improvements or extensions that they make and grant Carnegie the
  62  * rights to redistribute these changes.
  63  *
  64  * $FreeBSD: src/sys/vm/vm_kern.c,v 1.61.2.2 2002/03/12 18:25:26 tegge Exp $
  65  * $DragonFly: src/sys/vm/vm_kern.c,v 1.29 2007/06/07 23:14:29 dillon Exp $
  66  */
  67
  68 /*
  69  *      Kernel memory management.
  70  */
  71
  72 #include <sys/param.h>
  73 #include <sys/systm.h>
  74 #include <sys/proc.h>
  75 #include <sys/malloc.h>
  76 #include <sys/kernel.h>
  77 #include <sys/sysctl.h>
  78
  79 #include <vm/vm.h>
  80 #include <vm/vm_param.h>
  81 #include <sys/lock.h>
  82 #include <vm/pmap.h>
  83 #include <vm/vm_map.h>
  84 #include <vm/vm_object.h>
  85 #include <vm/vm_page.h>
  86 #include <vm/vm_pageout.h>
  87 #include <vm/vm_kern.h>
  88 #include <vm/vm_extern.h>
  89
  90 struct vm_map kernel_map;
  91 struct vm_map clean_map;
  92 struct vm_map buffer_map;
  93
  94 /*
  95  *      kmem_alloc_pageable:
  96  *
  97  *      Allocate pageable memory to the kernel's address map.
  98  *      "map" must be kernel_map or a submap of kernel_map.
  99  */
 100 vm_offset_t
 101 kmem_alloc_pageable(vm_map_t map, vm_size_t size)
 102 {
 103         vm_offset_t addr;
 104         int result;
 105
 106         size = round_page(size);
 107         addr = vm_map_min(map);
 108         result = vm_map_find(map, NULL, (vm_offset_t) 0,
 109                              &addr, size,
 110                              TRUE,
 111                              VM_MAPTYPE_NORMAL,
 112                              VM_PROT_ALL, VM_PROT_ALL,
 113                              0);
 114         if (result != KERN_SUCCESS) {
 115                 return (0);
 116         }
 117         return (addr);
 118 }
 119
 120 /*
 121  *      kmem_alloc_nofault:
 122  *
 123  *      Same as kmem_alloc_pageable, except that it create a nofault entry.
 124  */
 125 vm_offset_t
 126 kmem_alloc_nofault(vm_map_t map, vm_size_t size)
 127 {
 128         vm_offset_t addr;
 129         int result;
 130
 131         size = round_page(size);
 132         addr = vm_map_min(map);
 133         result = vm_map_find(map, NULL, (vm_offset_t) 0,
 134                              &addr, size,
 135                              TRUE,
 136                              VM_MAPTYPE_NORMAL,
 137                              VM_PROT_ALL, VM_PROT_ALL,
 138                              MAP_NOFAULT);
 139         if (result != KERN_SUCCESS) {
 140                 return (0);
 141         }
 142         return (addr);
 143 }
 144
 145 /*
 146  *      Allocate wired-down memory in the kernel's address map
 147  *      or a submap.
 148  */
 149 vm_offset_t
 150 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags)
 151 {
 152         vm_offset_t addr;
 153         vm_offset_t i;
 154         int count;
 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         /*
 164          * Use the kernel object for wired-down kernel pages. Assume that no
 165          * region of the kernel object is referenced more than once.
 166          *
 167          * Locate sufficient space in the map.  This will give us the final
 168          * virtual address for the new memory, and thus will tell us the
 169          * offset within the kernel map.
 170          */
 171         vm_map_lock(map);
 172         if (vm_map_findspace(map, vm_map_min(map), size, 1, 0, &addr)) {
 173                 vm_map_unlock(map);
 174                 if (kmflags & KM_KRESERVE)
 175                         vm_map_entry_krelease(count);
 176                 else
 177                         vm_map_entry_release(count);
 178                 return (0);
 179         }
 180         vm_object_reference(&kernel_object);
 181         vm_map_insert(map, &count,
 182                       &kernel_object, addr, addr, addr + size,
 183                       VM_MAPTYPE_NORMAL,
 184                       VM_PROT_ALL, VM_PROT_ALL,
 185                       0);
 186         vm_map_unlock(map);
 187         if (kmflags & KM_KRESERVE)
 188                 vm_map_entry_krelease(count);
 189         else
 190                 vm_map_entry_release(count);
 191
 192         /*
 193          * Guarantee that there are pages already in this object before
 194          * calling vm_map_wire.  This is to prevent the following
 195          * scenario:
 196          *
 197          * 1) Threads have swapped out, so that there is a pager for the
 198          * kernel_object. 2) The kmsg zone is empty, and so we are
 199          * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
 200          * there is no page, but there is a pager, so we call
 201          * pager_data_request.  But the kmsg zone is empty, so we must
 202          * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
 203          * we get the data back from the pager, it will be (very stale)
 204          * non-zero data.  kmem_alloc is defined to return zero-filled memory.
 205          *
 206          * We're intentionally not activating the pages we allocate to prevent a
 207          * race with page-out.  vm_map_wire will wire the pages.
 208          */
 209
 210         for (i = 0; i < size; i += PAGE_SIZE) {
 211                 vm_page_t mem;
 212
 213                 mem = vm_page_grab(&kernel_object, OFF_TO_IDX(addr + i),
 214                             VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
 215                 if ((mem->flags & PG_ZERO) == 0)
 216                         vm_page_zero_fill(mem);
 217                 mem->valid = VM_PAGE_BITS_ALL;
 218                 vm_page_flag_clear(mem, PG_ZERO);
 219                 vm_page_wakeup(mem);
 220         }
 221
 222         /*
 223          * And finally, mark the data as non-pageable.
 224          */
 225
 226         vm_map_wire(map, (vm_offset_t) addr, addr + size, kmflags);
 227
 228         return (addr);
 229 }
 230
 231 /*
 232  *      kmem_free:
 233  *
 234  *      Release a region of kernel virtual memory allocated
 235  *      with kmem_alloc, and return the physical pages
 236  *      associated with that region.
 237  *
 238  *      This routine may not block on kernel maps.
 239  */
 240 void
 241 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size)
 242 {
 243         vm_map_remove(map, trunc_page(addr), round_page(addr + size));
 244 }
 245
 246 /*
 247  *      kmem_suballoc:
 248  *
 249  *      Used to break a system map into smaller maps, usually to reduce
 250  *      contention and to provide large KVA spaces for subsystems like the
 251  *      buffer cache.
 252  *
 253  *      parent          Map to take range from
 254  *      result
 255  *      size            Size of range to find
 256  *      min, max        Returned endpoints of map
 257  *      pageable        Can the region be paged
 258  */
 259 void
 260 kmem_suballoc(vm_map_t parent, vm_map_t result,
 261               vm_offset_t *min, vm_offset_t *max, vm_size_t size)
 262 {
 263         int ret;
 264
 265         size = round_page(size);
 266
 267         *min = (vm_offset_t) vm_map_min(parent);
 268         ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
 269                           min, size,
 270                           TRUE,
 271                           VM_MAPTYPE_UNSPECIFIED,
 272                           VM_PROT_ALL, VM_PROT_ALL,
 273                           0);
 274         if (ret != KERN_SUCCESS) {
 275                 kprintf("kmem_suballoc: bad status return of %d.\n", ret);
 276                 panic("kmem_suballoc");
 277         }
 278         *max = *min + size;
 279         pmap_reference(vm_map_pmap(parent));
 280         vm_map_init(result, *min, *max, vm_map_pmap(parent));
 281         if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
 282                 panic("kmem_suballoc: unable to change range to submap");
 283 }
 284
 285 /*
 286  *      kmem_alloc_wait:
 287  *
 288  *      Allocates pageable memory from a sub-map of the kernel.  If the submap
 289  *      has no room, the caller sleeps waiting for more memory in the submap.
 290  *
 291  *      This routine may block.
 292  */
 293
 294 vm_offset_t
 295 kmem_alloc_wait(vm_map_t map, vm_size_t size)
 296 {
 297         vm_offset_t addr;
 298         int count;
 299
 300         size = round_page(size);
 301
 302         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 303
 304         for (;;) {
 305                 /*
 306                  * To make this work for more than one map, use the map's lock
 307                  * to lock out sleepers/wakers.
 308                  */
 309                 vm_map_lock(map);
 310                 if (vm_map_findspace(map, vm_map_min(map), size,
 311                                      1, 0, &addr) == 0) {
 312                         break;
 313                 }
 314                 /* no space now; see if we can ever get space */
 315                 if (vm_map_max(map) - vm_map_min(map) < size) {
 316                         vm_map_entry_release(count);
 317                         vm_map_unlock(map);
 318                         return (0);
 319                 }
 320                 vm_map_unlock(map);
 321                 tsleep(map, 0, "kmaw", 0);
 322         }
 323         vm_map_insert(map, &count,
 324                       NULL, (vm_offset_t) 0,
 325                       addr, addr + size,
 326                       VM_MAPTYPE_NORMAL,
 327                       VM_PROT_ALL, VM_PROT_ALL,
 328                       0);
 329         vm_map_unlock(map);
 330         vm_map_entry_release(count);
 331         return (addr);
 332 }
 333
 334 /*
 335  *      kmem_free_wakeup:
 336  *
 337  *      Returns memory to a submap of the kernel, and wakes up any processes
 338  *      waiting for memory in that map.
 339  */
 340 void
 341 kmem_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
 342 {
 343         int count;
 344
 345         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 346         vm_map_lock(map);
 347         vm_map_delete(map, trunc_page(addr), round_page(addr + size), &count);
 348         wakeup(map);
 349         vm_map_unlock(map);
 350         vm_map_entry_release(count);
 351 }
 352
 353 /*
 354  *      kmem_init:
 355  *
 356  *      Create the kernel_map and insert mappings to cover areas already
 357  *      allocated or reserved thus far.  That is, the area (KvaStart,start)
 358  *      and (end,KvaEnd) must be marked as allocated.
 359  *
 360  *      virtual2_start/end is a cutout Between KvaStart and start,
 361  *      for x86_64 due to the location of KERNBASE (at -2G).
 362  *
 363  *      We could use a min_offset of 0 instead of KvaStart, but since the
 364  *      min_offset is not used for any calculations other then a bounds check
 365  *      it does not effect readability.  KvaStart is more appropriate.
 366  *
 367  *      Depend on the zalloc bootstrap cache to get our vm_map_entry_t.
 368  */
 369 void
 370 kmem_init(vm_offset_t start, vm_offset_t end)
 371 {
 372         vm_offset_t addr;
 373         vm_map_t m;
 374         int count;
 375
 376         m = vm_map_create(&kernel_map, &kernel_pmap, KvaStart, KvaEnd);
 377         vm_map_lock(m);
 378         /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
 379         m->system_map = 1;
 380         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 381         addr = KvaStart;
 382         if (virtual2_start) {
 383                 if (addr < virtual2_start) {
 384                         vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 385                                       addr, virtual2_start,
 386                                       VM_MAPTYPE_NORMAL,
 387                                       VM_PROT_ALL, VM_PROT_ALL,
 388                                       0);
 389                 }
 390                 addr = virtual2_end;
 391         }
 392         if (addr < start) {
 393                 vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 394                               addr, start,
 395                               VM_MAPTYPE_NORMAL,
 396                               VM_PROT_ALL, VM_PROT_ALL,
 397                               0);
 398         }
 399         addr = end;
 400         if (addr < KvaEnd) {
 401                 vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 402                               addr, KvaEnd,
 403                               VM_MAPTYPE_NORMAL,
 404                               VM_PROT_ALL, VM_PROT_ALL,
 405                               0);
 406         }
 407         /* ... and ending with the completion of the above `insert' */
 408         vm_map_unlock(m);
 409         vm_map_entry_release(count);
 410 }
 411
 412 static int
 413 kvm_size(SYSCTL_HANDLER_ARGS)
 414 {
 415         unsigned long ksize = KvaSize;
 416
 417         return sysctl_handle_long(oidp, &ksize, 0, req);
 418 }
 419 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
 420     0, 0, kvm_size, "IU", "Size of KVM");
 421
 422 static int
 423 kvm_free(SYSCTL_HANDLER_ARGS)
 424 {
 425         unsigned long kfree = virtual_end - kernel_vm_end;
 426
 427         return sysctl_handle_long(oidp, &kfree, 0, req);
 428 }
 429 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
 430     0, 0, kvm_free, "IU", "Amount of KVM free");
 431