The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_kern.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    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  * 4. Neither the name of the University nor the names of its contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  *      from: @(#)vm_kern.c     8.3 (Berkeley) 1/12/94
   33  *
   34  *
   35  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
   36  * All rights reserved.
   37  *
   38  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
   39  *
   40  * Permission to use, copy, modify and distribute this software and
   41  * its documentation is hereby granted, provided that both the copyright
   42  * notice and this permission notice appear in all copies of the
   43  * software, derivative works or modified versions, and any portions
   44  * thereof, and that both notices appear in supporting documentation.
   45  *
   46  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
   47  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
   48  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
   49  *
   50  * Carnegie Mellon requests users of this software to return to
   51  *
   52  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
   53  *  School of Computer Science
   54  *  Carnegie Mellon University
   55  *  Pittsburgh PA 15213-3890
   56  *
   57  * any improvements or extensions that they make and grant Carnegie the
   58  * rights to redistribute these changes.
   59  */
   60 
   61 /*
   62  *      Kernel memory management.
   63  */
   64 
   65 #include <sys/cdefs.h>
   66 __FBSDID("$FreeBSD: releng/7.4/sys/vm/vm_kern.c 212437 2010-09-10 21:53:47Z mdf $");
   67 
   68 #include <sys/param.h>
   69 #include <sys/systm.h>
   70 #include <sys/kernel.h>         /* for ticks and hz */
   71 #include <sys/eventhandler.h>
   72 #include <sys/lock.h>
   73 #include <sys/mutex.h>
   74 #include <sys/proc.h>
   75 #include <sys/malloc.h>
   76 
   77 #include <vm/vm.h>
   78 #include <vm/vm_param.h>
   79 #include <vm/pmap.h>
   80 #include <vm/vm_map.h>
   81 #include <vm/vm_object.h>
   82 #include <vm/vm_page.h>
   83 #include <vm/vm_pageout.h>
   84 #include <vm/vm_extern.h>
   85 #include <vm/uma.h>
   86 
   87 vm_map_t kernel_map=0;
   88 vm_map_t kmem_map=0;
   89 vm_map_t exec_map=0;
   90 vm_map_t pipe_map;
   91 vm_map_t buffer_map=0;
   92 
   93 /*
   94  *      kmem_alloc_nofault:
   95  *
   96  *      Allocate a virtual address range with no underlying object and
   97  *      no initial mapping to physical memory.  Any mapping from this
   98  *      range to physical memory must be explicitly created prior to
   99  *      its use, typically with pmap_qenter().  Any attempt to create
  100  *      a mapping on demand through vm_fault() will result in a panic. 
  101  */
  102 vm_offset_t
  103 kmem_alloc_nofault(map, size)
  104         vm_map_t map;
  105         vm_size_t size;
  106 {
  107         vm_offset_t addr;
  108         int result;
  109 
  110         size = round_page(size);
  111         addr = vm_map_min(map);
  112         result = vm_map_find(map, NULL, 0, &addr, size, VMFS_ANY_SPACE,
  113             VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
  114         if (result != KERN_SUCCESS) {
  115                 return (0);
  116         }
  117         return (addr);
  118 }
  119 
  120 /*
  121  *      Allocate wired-down memory in the kernel's address map
  122  *      or a submap.
  123  */
  124 vm_offset_t
  125 kmem_alloc(map, size)
  126         vm_map_t map;
  127         vm_size_t size;
  128 {
  129         vm_offset_t addr;
  130         vm_offset_t offset;
  131         vm_offset_t i;
  132 
  133         size = round_page(size);
  134 
  135         /*
  136          * Use the kernel object for wired-down kernel pages. Assume that no
  137          * region of the kernel object is referenced more than once.
  138          */
  139 
  140         /*
  141          * Locate sufficient space in the map.  This will give us the final
  142          * virtual address for the new memory, and thus will tell us the
  143          * offset within the kernel map.
  144          */
  145         vm_map_lock(map);
  146         if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
  147                 vm_map_unlock(map);
  148                 return (0);
  149         }
  150         offset = addr - VM_MIN_KERNEL_ADDRESS;
  151         vm_object_reference(kernel_object);
  152         vm_map_insert(map, kernel_object, offset, addr, addr + size,
  153                 VM_PROT_ALL, VM_PROT_ALL, 0);
  154         vm_map_unlock(map);
  155 
  156         /*
  157          * Guarantee that there are pages already in this object before
  158          * calling vm_map_wire.  This is to prevent the following
  159          * scenario:
  160          *
  161          * 1) Threads have swapped out, so that there is a pager for the
  162          * kernel_object. 2) The kmsg zone is empty, and so we are
  163          * kmem_allocing a new page for it. 3) vm_map_wire calls vm_fault;
  164          * there is no page, but there is a pager, so we call
  165          * pager_data_request.  But the kmsg zone is empty, so we must
  166          * kmem_alloc. 4) goto 1 5) Even if the kmsg zone is not empty: when
  167          * we get the data back from the pager, it will be (very stale)
  168          * non-zero data.  kmem_alloc is defined to return zero-filled memory.
  169          *
  170          * We're intentionally not activating the pages we allocate to prevent a
  171          * race with page-out.  vm_map_wire will wire the pages.
  172          */
  173         VM_OBJECT_LOCK(kernel_object);
  174         for (i = 0; i < size; i += PAGE_SIZE) {
  175                 vm_page_t mem;
  176 
  177                 mem = vm_page_grab(kernel_object, OFF_TO_IDX(offset + i),
  178                     VM_ALLOC_NOBUSY | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
  179                 mem->valid = VM_PAGE_BITS_ALL;
  180                 KASSERT((mem->flags & PG_UNMANAGED) != 0,
  181                     ("kmem_alloc: page %p is managed", mem));
  182         }
  183         VM_OBJECT_UNLOCK(kernel_object);
  184 
  185         /*
  186          * And finally, mark the data as non-pageable.
  187          */
  188         (void) vm_map_wire(map, addr, addr + size,
  189             VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
  190 
  191         return (addr);
  192 }
  193 
  194 /*
  195  *      kmem_free:
  196  *
  197  *      Release a region of kernel virtual memory allocated
  198  *      with kmem_alloc, and return the physical pages
  199  *      associated with that region.
  200  *
  201  *      This routine may not block on kernel maps.
  202  */
  203 void
  204 kmem_free(map, addr, size)
  205         vm_map_t map;
  206         vm_offset_t addr;
  207         vm_size_t size;
  208 {
  209 
  210         (void) vm_map_remove(map, trunc_page(addr), round_page(addr + size));
  211 }
  212 
  213 /*
  214  *      kmem_suballoc:
  215  *
  216  *      Allocates a map to manage a subrange
  217  *      of the kernel virtual address space.
  218  *
  219  *      Arguments are as follows:
  220  *
  221  *      parent          Map to take range from
  222  *      min, max        Returned endpoints of map
  223  *      size            Size of range to find
  224  *      superpage_align Request that min is superpage aligned
  225  */
  226 vm_map_t
  227 kmem_suballoc(vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
  228     vm_size_t size, boolean_t superpage_align)
  229 {
  230         int ret;
  231         vm_map_t result;
  232 
  233         size = round_page(size);
  234 
  235         *min = vm_map_min(parent);
  236         ret = vm_map_find(parent, NULL, 0, min, size, superpage_align ?
  237             VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0);
  238         if (ret != KERN_SUCCESS) {
  239                 printf("kmem_suballoc: bad status return of %d.\n", ret);
  240                 panic("kmem_suballoc");
  241         }
  242         *max = *min + size;
  243         result = vm_map_create(vm_map_pmap(parent), *min, *max);
  244         if (result == NULL)
  245                 panic("kmem_suballoc: cannot create submap");
  246         if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
  247                 panic("kmem_suballoc: unable to change range to submap");
  248         return (result);
  249 }
  250 
  251 /*
  252  *      kmem_malloc:
  253  *
  254  *      Allocate wired-down memory in the kernel's address map for the higher
  255  *      level kernel memory allocator (kern/kern_malloc.c).  We cannot use
  256  *      kmem_alloc() because we may need to allocate memory at interrupt
  257  *      level where we cannot block (canwait == FALSE).
  258  *
  259  *      This routine has its own private kernel submap (kmem_map) and object
  260  *      (kmem_object).  This, combined with the fact that only malloc uses
  261  *      this routine, ensures that we will never block in map or object waits.
  262  *
  263  *      We don't worry about expanding the map (adding entries) since entries
  264  *      for wired maps are statically allocated.
  265  *
  266  *      `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
  267  *      which we never free.
  268  */
  269 vm_offset_t
  270 kmem_malloc(map, size, flags)
  271         vm_map_t map;
  272         vm_size_t size;
  273         int flags;
  274 {
  275         vm_offset_t addr;
  276         int i, rv;
  277 
  278         size = round_page(size);
  279         addr = vm_map_min(map);
  280 
  281         /*
  282          * Locate sufficient space in the map.  This will give us the final
  283          * virtual address for the new memory, and thus will tell us the
  284          * offset within the kernel map.
  285          */
  286         vm_map_lock(map);
  287         if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
  288                 vm_map_unlock(map);
  289                 if ((flags & M_NOWAIT) == 0) {
  290                         for (i = 0; i < 8; i++) {
  291                                 EVENTHANDLER_INVOKE(vm_lowmem, 0);
  292                                 uma_reclaim();
  293                                 vm_map_lock(map);
  294                                 if (vm_map_findspace(map, vm_map_min(map),
  295                                     size, &addr) == 0) {
  296                                         break;
  297                                 }
  298                                 vm_map_unlock(map);
  299                                 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
  300                         }
  301                         if (i == 8) {
  302                                 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
  303                                     (long)size, (long)map->size);
  304                         }
  305                 } else {
  306                         return (0);
  307                 }
  308         }
  309 
  310         rv = kmem_back(map, addr, size, flags);
  311         vm_map_unlock(map);
  312         return (rv == KERN_SUCCESS ? addr : 0);
  313 }
  314 
  315 /*
  316  *      kmem_back:
  317  *
  318  *      Allocate physical pages for the specified virtual address range.
  319  */
  320 int
  321 kmem_back(vm_map_t map, vm_offset_t addr, vm_size_t size, int flags)
  322 {
  323         vm_offset_t offset, i;
  324         vm_map_entry_t entry;
  325         vm_page_t m;
  326         int pflags;
  327 
  328         /*
  329          * XXX the map must be locked for write on entry, but there's
  330          * no easy way to assert that.
  331          */
  332 
  333         offset = addr - VM_MIN_KERNEL_ADDRESS;
  334         vm_object_reference(kmem_object);
  335         vm_map_insert(map, kmem_object, offset, addr, addr + size,
  336                 VM_PROT_ALL, VM_PROT_ALL, 0);
  337 
  338         if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
  339                 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
  340         else
  341                 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
  342 
  343         if (flags & M_ZERO)
  344                 pflags |= VM_ALLOC_ZERO;
  345 
  346         VM_OBJECT_LOCK(kmem_object);
  347         for (i = 0; i < size; i += PAGE_SIZE) {
  348 retry:
  349                 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
  350 
  351                 /*
  352                  * Ran out of space, free everything up and return. Don't need
  353                  * to lock page queues here as we know that the pages we got
  354                  * aren't on any queues.
  355                  */
  356                 if (m == NULL) {
  357                         if ((flags & M_NOWAIT) == 0) {
  358                                 VM_OBJECT_UNLOCK(kmem_object);
  359                                 vm_map_unlock(map);
  360                                 VM_WAIT;
  361                                 vm_map_lock(map);
  362                                 VM_OBJECT_LOCK(kmem_object);
  363                                 goto retry;
  364                         }
  365                         /* 
  366                          * Free the pages before removing the map entry.
  367                          * They are already marked busy.  Calling
  368                          * vm_map_delete before the pages has been freed or
  369                          * unbusied will cause a deadlock.
  370                          */
  371                         while (i != 0) {
  372                                 i -= PAGE_SIZE;
  373                                 m = vm_page_lookup(kmem_object,
  374                                                    OFF_TO_IDX(offset + i));
  375                                 vm_page_lock_queues();
  376                                 vm_page_unwire(m, 0);
  377                                 vm_page_free(m);
  378                                 vm_page_unlock_queues();
  379                         }
  380                         VM_OBJECT_UNLOCK(kmem_object);
  381                         vm_map_delete(map, addr, addr + size);
  382                         return (KERN_NO_SPACE);
  383                 }
  384                 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
  385                         pmap_zero_page(m);
  386                 m->valid = VM_PAGE_BITS_ALL;
  387                 KASSERT((m->flags & PG_UNMANAGED) != 0,
  388                     ("kmem_malloc: page %p is managed", m));
  389         }
  390         VM_OBJECT_UNLOCK(kmem_object);
  391 
  392         /*
  393          * Mark map entry as non-pageable. Assert: vm_map_insert() will never
  394          * be able to extend the previous entry so there will be a new entry
  395          * exactly corresponding to this address range and it will have
  396          * wired_count == 0.
  397          */
  398         if (!vm_map_lookup_entry(map, addr, &entry) ||
  399             entry->start != addr || entry->end != addr + size ||
  400             entry->wired_count != 0)
  401                 panic("kmem_malloc: entry not found or misaligned");
  402         entry->wired_count = 1;
  403 
  404         /*
  405          * At this point, the kmem_object must be unlocked because
  406          * vm_map_simplify_entry() calls vm_object_deallocate(), which
  407          * locks the kmem_object.
  408          */
  409         vm_map_simplify_entry(map, entry);
  410 
  411         /*
  412          * Loop thru pages, entering them in the pmap.
  413          */
  414         VM_OBJECT_LOCK(kmem_object);
  415         for (i = 0; i < size; i += PAGE_SIZE) {
  416                 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
  417                 /*
  418                  * Because this is kernel_pmap, this call will not block.
  419                  */
  420                 pmap_enter(kernel_pmap, addr + i, VM_PROT_ALL, m, VM_PROT_ALL,
  421                     TRUE);
  422                 vm_page_wakeup(m);
  423         }
  424         VM_OBJECT_UNLOCK(kmem_object);
  425 
  426         return (KERN_SUCCESS);
  427 }
  428 
  429 /*
  430  *      kmem_alloc_wait:
  431  *
  432  *      Allocates pageable memory from a sub-map of the kernel.  If the submap
  433  *      has no room, the caller sleeps waiting for more memory in the submap.
  434  *
  435  *      This routine may block.
  436  */
  437 vm_offset_t
  438 kmem_alloc_wait(map, size)
  439         vm_map_t map;
  440         vm_size_t size;
  441 {
  442         vm_offset_t addr;
  443 
  444         size = round_page(size);
  445 
  446         for (;;) {
  447                 /*
  448                  * To make this work for more than one map, use the map's lock
  449                  * to lock out sleepers/wakers.
  450                  */
  451                 vm_map_lock(map);
  452                 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
  453                         break;
  454                 /* no space now; see if we can ever get space */
  455                 if (vm_map_max(map) - vm_map_min(map) < size) {
  456                         vm_map_unlock(map);
  457                         return (0);
  458                 }
  459                 map->needs_wakeup = TRUE;
  460                 vm_map_unlock_and_wait(map, FALSE);
  461         }
  462         vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
  463         vm_map_unlock(map);
  464         return (addr);
  465 }
  466 
  467 /*
  468  *      kmem_free_wakeup:
  469  *
  470  *      Returns memory to a submap of the kernel, and wakes up any processes
  471  *      waiting for memory in that map.
  472  */
  473 void
  474 kmem_free_wakeup(map, addr, size)
  475         vm_map_t map;
  476         vm_offset_t addr;
  477         vm_size_t size;
  478 {
  479 
  480         vm_map_lock(map);
  481         (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
  482         if (map->needs_wakeup) {
  483                 map->needs_wakeup = FALSE;
  484                 vm_map_wakeup(map);
  485         }
  486         vm_map_unlock(map);
  487 }
  488 
  489 /*
  490  *      kmem_init:
  491  *
  492  *      Create the kernel map; insert a mapping covering kernel text, 
  493  *      data, bss, and all space allocated thus far (`boostrap' data).  The 
  494  *      new map will thus map the range between VM_MIN_KERNEL_ADDRESS and 
  495  *      `start' as allocated, and the range between `start' and `end' as free.
  496  */
  497 void
  498 kmem_init(start, end)
  499         vm_offset_t start, end;
  500 {
  501         vm_map_t m;
  502 
  503         m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
  504         m->system_map = 1;
  505         vm_map_lock(m);
  506         /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
  507         kernel_map = m;
  508         (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
  509 #ifdef __amd64__
  510             KERNBASE,
  511 #else                
  512             VM_MIN_KERNEL_ADDRESS,
  513 #endif
  514             start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
  515         /* ... and ending with the completion of the above `insert' */
  516         vm_map_unlock(m);
  517 }

Cache object: 78415fdecefa403198b416d286319f3d


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.