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  * $DragonFly: src/sys/vm/vm_kern.c,v 1.29 2007/06/07 23:14:29 dillon Exp $
  68  */
  69
  70 /*
  71  *      Kernel memory management.
  72  */
  73
  74 #include <sys/param.h>
  75 #include <sys/systm.h>
  76 #include <sys/proc.h>
  77 #include <sys/malloc.h>
  78 #include <sys/kernel.h>
  79 #include <sys/sysctl.h>
  80
  81 #include <vm/vm.h>
  82 #include <vm/vm_param.h>
  83 #include <sys/lock.h>
  84 #include <vm/pmap.h>
  85 #include <vm/vm_map.h>
  86 #include <vm/vm_object.h>
  87 #include <vm/vm_page.h>
  88 #include <vm/vm_pageout.h>
  89 #include <vm/vm_kern.h>
  90 #include <vm/vm_extern.h>
  91
  92 struct vm_map kernel_map;
  93 struct vm_map clean_map;
  94 struct vm_map buffer_map;
  95
  96 /*
  97  * Allocate pageable memory to the kernel's address map.  "map" must
  98  * be kernel_map or a submap of kernel_map.
  99  *
 100  * No requirements.
 101  */
 102 vm_offset_t
 103 kmem_alloc_pageable(vm_map_t map, vm_size_t size)
 104 {
 105         vm_offset_t addr;
 106         int result;
 107
 108         size = round_page(size);
 109         addr = vm_map_min(map);
 110         result = vm_map_find(map, NULL, (vm_offset_t) 0,
 111                              &addr, size, PAGE_SIZE,
 112                              TRUE, VM_MAPTYPE_NORMAL,
 113                              VM_PROT_ALL, VM_PROT_ALL,
 114                              0);
 115         if (result != KERN_SUCCESS)
 116                 return (0);
 117         return (addr);
 118 }
 119
 120 /*
 121  * Same as kmem_alloc_pageable, except that it create a nofault entry.
 122  *
 123  * No requirements.
 124  */
 125 vm_offset_t
 126 kmem_alloc_nofault(vm_map_t map, vm_size_t size, vm_size_t align)
 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, align,
 135                              TRUE, VM_MAPTYPE_NORMAL,
 136                              VM_PROT_ALL, VM_PROT_ALL,
 137                              MAP_NOFAULT);
 138         if (result != KERN_SUCCESS)
 139                 return (0);
 140         return (addr);
 141 }
 142
 143 /*
 144  * Allocate wired-down memory in the kernel's address map or a submap.
 145  *
 146  * No requirements.
 147  */
 148 vm_offset_t
 149 kmem_alloc3(vm_map_t map, vm_size_t size, int kmflags)
 150 {
 151         vm_offset_t addr;
 152         vm_offset_t i;
 153         int count;
 154
 155         size = round_page(size);
 156
 157         if (kmflags & KM_KRESERVE)
 158                 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
 159         else
 160                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 161
 162         /*
 163          * Use the kernel object for wired-down kernel pages. Assume that no
 164          * region of the kernel object is referenced more than once.
 165          *
 166          * Locate sufficient space in the map.  This will give us the final
 167          * virtual address for the new memory, and thus will tell us the
 168          * offset within the kernel map.
 169          */
 170         vm_map_lock(map);
 171         if (vm_map_findspace(map, vm_map_min(map), size, PAGE_SIZE, 0, &addr)) {
 172                 vm_map_unlock(map);
 173                 if (kmflags & KM_KRESERVE)
 174                         vm_map_entry_krelease(count);
 175                 else
 176                         vm_map_entry_release(count);
 177                 return (0);
 178         }
 179         vm_object_reference(&kernel_object);
 180         vm_map_insert(map, &count,
 181                       &kernel_object, addr, addr, addr + size,
 182                       VM_MAPTYPE_NORMAL,
 183                       VM_PROT_ALL, VM_PROT_ALL,
 184                       0);
 185         vm_map_unlock(map);
 186         if (kmflags & KM_KRESERVE)
 187                 vm_map_entry_krelease(count);
 188         else
 189                 vm_map_entry_release(count);
 190
 191         /*
 192          * Guarantee that there are pages already in this object before
 193          * calling vm_map_wire.  This is to prevent the following
 194          * scenario:
 195          *
 196          * 1) Threads have swapped out, so that there is a pager for the
 197          * kernel_object. 2) The kmsg zone is empty, and so we are
 198          * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
 199          * there is no page, but there is a pager, so we call
 200          * pager_data_request.  But the kmsg zone is empty, so we must
 201          * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
 202          * we get the data back from the pager, it will be (very stale)
 203          * non-zero data.  kmem_alloc is defined to return zero-filled memory.
 204          *
 205          * We're intentionally not activating the pages we allocate to prevent a
 206          * race with page-out.  vm_map_wire will wire the pages.
 207          */
 208
 209         lwkt_gettoken(&vm_token);
 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         lwkt_reltoken(&vm_token);
 222
 223         /*
 224          * And finally, mark the data as non-pageable.
 225          */
 226         vm_map_wire(map, (vm_offset_t)addr, addr + size, kmflags);
 227
 228         return (addr);
 229 }
 230
 231 /*
 232  * Release a region of kernel virtual memory allocated with kmem_alloc,
 233  * and return the physical pages associated with that region.
 234  *
 235  * WARNING!  If the caller entered pages into the region using pmap_kenter()
 236  * it must remove the pages using pmap_kremove[_quick]() before freeing the
 237  * underlying kmem, otherwise resident_count will be mistabulated.
 238  *
 239  * No requirements.
 240  */
 241 void
 242 kmem_free(vm_map_t map, vm_offset_t addr, vm_size_t size)
 243 {
 244         vm_map_remove(map, trunc_page(addr), round_page(addr + size));
 245 }
 246
 247 /*
 248  * Used to break a system map into smaller maps, usually to reduce
 249  * contention and to provide large KVA spaces for subsystems like the
 250  * buffer cache.
 251  *
 252  *      parent          Map to take range from
 253  *      result
 254  *      size            Size of range to find
 255  *      min, max        Returned endpoints of map
 256  *      pageable        Can the region be paged
 257  *
 258  * No requirements.
 259  */
 260 void
 261 kmem_suballoc(vm_map_t parent, vm_map_t result,
 262               vm_offset_t *min, vm_offset_t *max, vm_size_t size)
 263 {
 264         int ret;
 265
 266         size = round_page(size);
 267
 268         lwkt_gettoken(&vm_token);
 269         *min = (vm_offset_t) vm_map_min(parent);
 270         ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
 271                           min, size, PAGE_SIZE,
 272                           TRUE, VM_MAPTYPE_UNSPECIFIED,
 273                           VM_PROT_ALL, VM_PROT_ALL,
 274                           0);
 275         if (ret != KERN_SUCCESS) {
 276                 kprintf("kmem_suballoc: bad status return of %d.\n", ret);
 277                 panic("kmem_suballoc");
 278         }
 279         *max = *min + size;
 280         pmap_reference(vm_map_pmap(parent));
 281         vm_map_init(result, *min, *max, vm_map_pmap(parent));
 282         if ((ret = vm_map_submap(parent, *min, *max, result)) != KERN_SUCCESS)
 283                 panic("kmem_suballoc: unable to change range to submap");
 284         lwkt_reltoken(&vm_token);
 285 }
 286
 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  * No requirements.
 292  */
 293 vm_offset_t
 294 kmem_alloc_wait(vm_map_t map, vm_size_t size)
 295 {
 296         vm_offset_t addr;
 297         int count;
 298
 299         size = round_page(size);
 300
 301         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 302
 303         for (;;) {
 304                 /*
 305                  * To make this work for more than one map, use the map's lock
 306                  * to lock out sleepers/wakers.
 307                  */
 308                 vm_map_lock(map);
 309                 if (vm_map_findspace(map, vm_map_min(map),
 310                                      size, PAGE_SIZE, 0, &addr) == 0) {
 311                         break;
 312                 }
 313                 /* no space now; see if we can ever get space */
 314                 if (vm_map_max(map) - vm_map_min(map) < size) {
 315                         vm_map_entry_release(count);
 316                         vm_map_unlock(map);
 317                         return (0);
 318                 }
 319                 vm_map_unlock(map);
 320                 tsleep(map, 0, "kmaw", 0);
 321         }
 322         vm_map_insert(map, &count,
 323                       NULL, (vm_offset_t) 0,
 324                       addr, addr + size,
 325                       VM_MAPTYPE_NORMAL,
 326                       VM_PROT_ALL, VM_PROT_ALL,
 327                       0);
 328         vm_map_unlock(map);
 329         vm_map_entry_release(count);
 330
 331         return (addr);
 332 }
 333
 334 /*
 335  * Returns memory to a submap of the kernel, and wakes up any processes
 336  * waiting for memory in that map.
 337  *
 338  * No requirements.
 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  * Create the kernel_map and insert mappings to cover areas already
 355  * allocated or reserved thus far.  That is, the area (KvaStart,start)
 356  * and (end,KvaEnd) must be marked as allocated.
 357  *
 358  * virtual2_start/end is a cutout Between KvaStart and start,
 359  * for x86_64 due to the location of KERNBASE (at -2G).
 360  *
 361  * We could use a min_offset of 0 instead of KvaStart, but since the
 362  * min_offset is not used for any calculations other then a bounds check
 363  * it does not effect readability.  KvaStart is more appropriate.
 364  *
 365  * Depend on the zalloc bootstrap cache to get our vm_map_entry_t.
 366  * Called from the low level boot code only.
 367  */
 368 void
 369 kmem_init(vm_offset_t start, vm_offset_t end)
 370 {
 371         vm_offset_t addr;
 372         vm_map_t m;
 373         int count;
 374
 375         m = vm_map_create(&kernel_map, &kernel_pmap, KvaStart, KvaEnd);
 376         vm_map_lock(m);
 377         /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
 378         m->system_map = 1;
 379         count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 380         addr = KvaStart;
 381         if (virtual2_start) {
 382                 if (addr < virtual2_start) {
 383                         vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 384                                       addr, virtual2_start,
 385                                       VM_MAPTYPE_NORMAL,
 386                                       VM_PROT_ALL, VM_PROT_ALL,
 387                                       0);
 388                 }
 389                 addr = virtual2_end;
 390         }
 391         if (addr < start) {
 392                 vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 393                               addr, start,
 394                               VM_MAPTYPE_NORMAL,
 395                               VM_PROT_ALL, VM_PROT_ALL,
 396                               0);
 397         }
 398         addr = end;
 399         if (addr < KvaEnd) {
 400                 vm_map_insert(m, &count, NULL, (vm_offset_t) 0,
 401                               addr, KvaEnd,
 402                               VM_MAPTYPE_NORMAL,
 403                               VM_PROT_ALL, VM_PROT_ALL,
 404                               0);
 405         }
 406         /* ... and ending with the completion of the above `insert' */
 407         vm_map_unlock(m);
 408         vm_map_entry_release(count);
 409 }
 410
 411 /*
 412  * No requirements.
 413  */
 414 static int
 415 kvm_size(SYSCTL_HANDLER_ARGS)
 416 {
 417         unsigned long ksize = KvaSize;
 418
 419         return sysctl_handle_long(oidp, &ksize, 0, req);
 420 }
 421 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
 422     0, 0, kvm_size, "IU", "Size of KVM");
 423
 424 /*
 425  * No requirements.
 426  */
 427 static int
 428 kvm_free(SYSCTL_HANDLER_ARGS)
 429 {
 430         unsigned long kfree = virtual_end - kernel_vm_end;
 431
 432         return sysctl_handle_long(oidp, &kfree, 0, req);
 433 }
 434 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
 435     0, 0, kvm_free, "IU", "Amount of KVM free");
 436