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("$FreeBSD$");
   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,
  113             &addr, size, TRUE, 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  */
  225 vm_map_t
  226 kmem_suballoc(parent, min, max, size)
  227         vm_map_t parent;
  228         vm_offset_t *min, *max;
  229         vm_size_t size;
  230 {
  231         int ret;
  232         vm_map_t result;
  233 
  234         size = round_page(size);
  235 
  236         *min = (vm_offset_t) vm_map_min(parent);
  237         ret = vm_map_find(parent, NULL, (vm_offset_t) 0,
  238             min, size, TRUE, VM_PROT_ALL, VM_PROT_ALL, 0);
  239         if (ret != KERN_SUCCESS) {
  240                 printf("kmem_suballoc: bad status return of %d.\n", ret);
  241                 panic("kmem_suballoc");
  242         }
  243         *max = *min + size;
  244         result = vm_map_create(vm_map_pmap(parent), *min, *max);
  245         if (result == NULL)
  246                 panic("kmem_suballoc: cannot create submap");
  247         if (vm_map_submap(parent, *min, *max, result) != KERN_SUCCESS)
  248                 panic("kmem_suballoc: unable to change range to submap");
  249         return (result);
  250 }
  251 
  252 /*
  253  *      kmem_malloc:
  254  *
  255  *      Allocate wired-down memory in the kernel's address map for the higher
  256  *      level kernel memory allocator (kern/kern_malloc.c).  We cannot use
  257  *      kmem_alloc() because we may need to allocate memory at interrupt
  258  *      level where we cannot block (canwait == FALSE).
  259  *
  260  *      This routine has its own private kernel submap (kmem_map) and object
  261  *      (kmem_object).  This, combined with the fact that only malloc uses
  262  *      this routine, ensures that we will never block in map or object waits.
  263  *
  264  *      Note that this still only works in a uni-processor environment and
  265  *      when called at splhigh().
  266  *
  267  *      We don't worry about expanding the map (adding entries) since entries
  268  *      for wired maps are statically allocated.
  269  *
  270  *      NOTE:  This routine is not supposed to block if M_NOWAIT is set, but
  271  *      I have not verified that it actually does not block.
  272  *
  273  *      `map' is ONLY allowed to be kmem_map or one of the mbuf submaps to
  274  *      which we never free.
  275  */
  276 vm_offset_t
  277 kmem_malloc(map, size, flags)
  278         vm_map_t map;
  279         vm_size_t size;
  280         int flags;
  281 {
  282         vm_offset_t offset, i;
  283         vm_map_entry_t entry;
  284         vm_offset_t addr;
  285         vm_page_t m;
  286         int pflags;
  287 
  288         size = round_page(size);
  289         addr = vm_map_min(map);
  290 
  291         /*
  292          * Locate sufficient space in the map.  This will give us the final
  293          * virtual address for the new memory, and thus will tell us the
  294          * offset within the kernel map.
  295          */
  296         vm_map_lock(map);
  297         if (vm_map_findspace(map, vm_map_min(map), size, &addr)) {
  298                 vm_map_unlock(map);
  299                 if ((flags & M_NOWAIT) == 0) {
  300                         for (i = 0; i < 8; i++) {
  301                                 EVENTHANDLER_INVOKE(vm_lowmem, 0);
  302                                 uma_reclaim();
  303                                 vm_map_lock(map);
  304                                 if (vm_map_findspace(map, vm_map_min(map),
  305                                     size, &addr) == 0) {
  306                                         break;
  307                                 }
  308                                 vm_map_unlock(map);
  309                                 tsleep(&i, 0, "nokva", (hz / 4) * (i + 1));
  310                         }
  311                         if (i == 8) {
  312                                 panic("kmem_malloc(%ld): kmem_map too small: %ld total allocated",
  313                                     (long)size, (long)map->size);
  314                         }
  315                 } else {
  316                         return (0);
  317                 }
  318         }
  319         offset = addr - VM_MIN_KERNEL_ADDRESS;
  320         vm_object_reference(kmem_object);
  321         vm_map_insert(map, kmem_object, offset, addr, addr + size,
  322                 VM_PROT_ALL, VM_PROT_ALL, 0);
  323 
  324         /*
  325          * Note: if M_NOWAIT specified alone, allocate from 
  326          * interrupt-safe queues only (just the free list).  If 
  327          * M_USE_RESERVE is also specified, we can also
  328          * allocate from the cache.  Neither of the latter two
  329          * flags may be specified from an interrupt since interrupts
  330          * are not allowed to mess with the cache queue.
  331          */
  332 
  333         if ((flags & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
  334                 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
  335         else
  336                 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
  337 
  338         if (flags & M_ZERO)
  339                 pflags |= VM_ALLOC_ZERO;
  340 
  341         VM_OBJECT_LOCK(kmem_object);
  342         for (i = 0; i < size; i += PAGE_SIZE) {
  343 retry:
  344                 m = vm_page_alloc(kmem_object, OFF_TO_IDX(offset + i), pflags);
  345 
  346                 /*
  347                  * Ran out of space, free everything up and return. Don't need
  348                  * to lock page queues here as we know that the pages we got
  349                  * aren't on any queues.
  350                  */
  351                 if (m == NULL) {
  352                         if ((flags & M_NOWAIT) == 0) {
  353                                 VM_OBJECT_UNLOCK(kmem_object);
  354                                 vm_map_unlock(map);
  355                                 VM_WAIT;
  356                                 vm_map_lock(map);
  357                                 VM_OBJECT_LOCK(kmem_object);
  358                                 goto retry;
  359                         }
  360                         /* 
  361                          * Free the pages before removing the map entry.
  362                          * They are already marked busy.  Calling
  363                          * vm_map_delete before the pages has been freed or
  364                          * unbusied will cause a deadlock.
  365                          */
  366                         while (i != 0) {
  367                                 i -= PAGE_SIZE;
  368                                 m = vm_page_lookup(kmem_object,
  369                                                    OFF_TO_IDX(offset + i));
  370                                 vm_page_lock_queues();
  371                                 vm_page_unwire(m, 0);
  372                                 vm_page_free(m);
  373                                 vm_page_unlock_queues();
  374                         }
  375                         VM_OBJECT_UNLOCK(kmem_object);
  376                         vm_map_delete(map, addr, addr + size);
  377                         vm_map_unlock(map);
  378                         return (0);
  379                 }
  380                 if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
  381                         pmap_zero_page(m);
  382                 m->valid = VM_PAGE_BITS_ALL;
  383                 KASSERT((m->flags & PG_UNMANAGED) != 0,
  384                     ("kmem_malloc: page %p is managed", m));
  385         }
  386         VM_OBJECT_UNLOCK(kmem_object);
  387 
  388         /*
  389          * Mark map entry as non-pageable. Assert: vm_map_insert() will never
  390          * be able to extend the previous entry so there will be a new entry
  391          * exactly corresponding to this address range and it will have
  392          * wired_count == 0.
  393          */
  394         if (!vm_map_lookup_entry(map, addr, &entry) ||
  395             entry->start != addr || entry->end != addr + size ||
  396             entry->wired_count != 0)
  397                 panic("kmem_malloc: entry not found or misaligned");
  398         entry->wired_count = 1;
  399 
  400         /*
  401          * At this point, the kmem_object must be unlocked because
  402          * vm_map_simplify_entry() calls vm_object_deallocate(), which
  403          * locks the kmem_object.
  404          */
  405         vm_map_simplify_entry(map, entry);
  406 
  407         /*
  408          * Loop thru pages, entering them in the pmap.
  409          */
  410         VM_OBJECT_LOCK(kmem_object);
  411         for (i = 0; i < size; i += PAGE_SIZE) {
  412                 m = vm_page_lookup(kmem_object, OFF_TO_IDX(offset + i));
  413                 /*
  414                  * Because this is kernel_pmap, this call will not block.
  415                  */
  416                 pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL, 1);
  417                 vm_page_wakeup(m);
  418         }
  419         VM_OBJECT_UNLOCK(kmem_object);
  420         vm_map_unlock(map);
  421 
  422         return (addr);
  423 }
  424 
  425 /*
  426  *      kmem_alloc_wait:
  427  *
  428  *      Allocates pageable memory from a sub-map of the kernel.  If the submap
  429  *      has no room, the caller sleeps waiting for more memory in the submap.
  430  *
  431  *      This routine may block.
  432  */
  433 vm_offset_t
  434 kmem_alloc_wait(map, size)
  435         vm_map_t map;
  436         vm_size_t size;
  437 {
  438         vm_offset_t addr;
  439 
  440         size = round_page(size);
  441 
  442         for (;;) {
  443                 /*
  444                  * To make this work for more than one map, use the map's lock
  445                  * to lock out sleepers/wakers.
  446                  */
  447                 vm_map_lock(map);
  448                 if (vm_map_findspace(map, vm_map_min(map), size, &addr) == 0)
  449                         break;
  450                 /* no space now; see if we can ever get space */
  451                 if (vm_map_max(map) - vm_map_min(map) < size) {
  452                         vm_map_unlock(map);
  453                         return (0);
  454                 }
  455                 map->needs_wakeup = TRUE;
  456                 vm_map_unlock_and_wait(map, FALSE);
  457         }
  458         vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
  459         vm_map_unlock(map);
  460         return (addr);
  461 }
  462 
  463 /*
  464  *      kmem_free_wakeup:
  465  *
  466  *      Returns memory to a submap of the kernel, and wakes up any processes
  467  *      waiting for memory in that map.
  468  */
  469 void
  470 kmem_free_wakeup(map, addr, size)
  471         vm_map_t map;
  472         vm_offset_t addr;
  473         vm_size_t size;
  474 {
  475 
  476         vm_map_lock(map);
  477         (void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
  478         if (map->needs_wakeup) {
  479                 map->needs_wakeup = FALSE;
  480                 vm_map_wakeup(map);
  481         }
  482         vm_map_unlock(map);
  483 }
  484 
  485 /*
  486  *      kmem_init:
  487  *
  488  *      Create the kernel map; insert a mapping covering kernel text, 
  489  *      data, bss, and all space allocated thus far (`boostrap' data).  The 
  490  *      new map will thus map the range between VM_MIN_KERNEL_ADDRESS and 
  491  *      `start' as allocated, and the range between `start' and `end' as free.
  492  */
  493 void
  494 kmem_init(start, end)
  495         vm_offset_t start, end;
  496 {
  497         vm_map_t m;
  498 
  499         m = vm_map_create(kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
  500         m->system_map = 1;
  501         vm_map_lock(m);
  502         /* N.B.: cannot use kgdb to debug, starting with this assignment ... */
  503         kernel_map = m;
  504         (void) vm_map_insert(m, NULL, (vm_ooffset_t) 0,
  505             VM_MIN_KERNEL_ADDRESS, start, VM_PROT_ALL, VM_PROT_ALL,
  506             MAP_NOFAULT);
  507         /* ... and ending with the completion of the above `insert' */
  508         vm_map_unlock(m);
  509 }

Cache object: 4d3a04b043e3606872bed8daa5ee8d46


[ 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.