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_pageout.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 Regents of the University of California.
    3  * All rights reserved.
    4  * Copyright (c) 1994 John S. Dyson
    5  * All rights reserved.
    6  * Copyright (c) 1994 David Greenman
    7  * All rights reserved.
    8  *
    9  * This code is derived from software contributed to Berkeley by
   10  * The Mach Operating System project at Carnegie-Mellon University.
   11  *
   12  * Redistribution and use in source and binary forms, with or without
   13  * modification, are permitted provided that the following conditions
   14  * are met:
   15  * 1. Redistributions of source code must retain the above copyright
   16  *    notice, this list of conditions and the following disclaimer.
   17  * 2. Redistributions in binary form must reproduce the above copyright
   18  *    notice, this list of conditions and the following disclaimer in the
   19  *    documentation and/or other materials provided with the distribution.
   20  * 3. All advertising materials mentioning features or use of this software
   21  *    must display the following acknowledgement:
   22  *      This product includes software developed by the University of
   23  *      California, Berkeley and its contributors.
   24  * 4. Neither the name of the University nor the names of its contributors
   25  *    may be used to endorse or promote products derived from this software
   26  *    without specific prior written permission.
   27  *
   28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   38  * SUCH DAMAGE.
   39  *
   40  *      from: @(#)vm_pageout.c  7.4 (Berkeley) 5/7/91
   41  *
   42  *
   43  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
   44  * All rights reserved.
   45  *
   46  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
   47  *
   48  * Permission to use, copy, modify and distribute this software and
   49  * its documentation is hereby granted, provided that both the copyright
   50  * notice and this permission notice appear in all copies of the
   51  * software, derivative works or modified versions, and any portions
   52  * thereof, and that both notices appear in supporting documentation.
   53  *
   54  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
   55  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
   56  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
   57  *
   58  * Carnegie Mellon requests users of this software to return to
   59  *
   60  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
   61  *  School of Computer Science
   62  *  Carnegie Mellon University
   63  *  Pittsburgh PA 15213-3890
   64  *
   65  * any improvements or extensions that they make and grant Carnegie the
   66  * rights to redistribute these changes.
   67  *
   68  * $FreeBSD$
   69  */
   70 
   71 /*
   72  *      The proverbial page-out daemon.
   73  */
   74 
   75 #include "opt_vm.h"
   76 #include <sys/param.h>
   77 #include <sys/systm.h>
   78 #include <sys/kernel.h>
   79 #include <sys/proc.h>
   80 #include <sys/resourcevar.h>
   81 #include <sys/signalvar.h>
   82 #include <sys/vnode.h>
   83 #include <sys/vmmeter.h>
   84 #include <sys/sysctl.h>
   85 
   86 #include <vm/vm.h>
   87 #include <vm/vm_param.h>
   88 #include <vm/vm_prot.h>
   89 #include <sys/lock.h>
   90 #include <vm/vm_object.h>
   91 #include <vm/vm_page.h>
   92 #include <vm/vm_map.h>
   93 #include <vm/vm_pageout.h>
   94 #include <vm/vm_pager.h>
   95 #include <vm/swap_pager.h>
   96 #include <vm/vm_extern.h>
   97 
   98 /*
   99  * System initialization
  100  */
  101 
  102 /* the kernel process "vm_pageout"*/
  103 static void vm_pageout __P((void));
  104 static int vm_pageout_clean __P((vm_page_t));
  105 static int vm_pageout_scan __P((void));
  106 static int vm_pageout_free_page_calc __P((vm_size_t count));
  107 struct proc *pageproc;
  108 
  109 static struct kproc_desc page_kp = {
  110         "pagedaemon",
  111         vm_pageout,
  112         &pageproc
  113 };
  114 SYSINIT_KT(pagedaemon, SI_SUB_KTHREAD_PAGE, SI_ORDER_FIRST, kproc_start, &page_kp)
  115 
  116 #if !defined(NO_SWAPPING)
  117 /* the kernel process "vm_daemon"*/
  118 static void vm_daemon __P((void));
  119 static struct   proc *vmproc;
  120 
  121 static struct kproc_desc vm_kp = {
  122         "vmdaemon",
  123         vm_daemon,
  124         &vmproc
  125 };
  126 SYSINIT_KT(vmdaemon, SI_SUB_KTHREAD_VM, SI_ORDER_FIRST, kproc_start, &vm_kp)
  127 #endif
  128 
  129 
  130 int vm_pages_needed=0;          /* Event on which pageout daemon sleeps */
  131 int vm_pageout_deficit=0;       /* Estimated number of pages deficit */
  132 int vm_pageout_pages_needed=0;  /* flag saying that the pageout daemon needs pages */
  133 
  134 extern int npendingio;
  135 #if !defined(NO_SWAPPING)
  136 static int vm_pageout_req_swapout;      /* XXX */
  137 static int vm_daemon_needed;
  138 #endif
  139 extern int nswiodone;
  140 extern int vm_swap_size;
  141 extern int vfs_update_wakeup;
  142 static int vm_pageout_stats_max=0, vm_pageout_stats_interval = 0;
  143 static int vm_pageout_full_stats_interval = 0;
  144 static int vm_pageout_stats_free_max=0, vm_pageout_algorithm_lru=0;
  145 static int defer_swap_pageouts=0;
  146 static int disable_swap_pageouts=0;
  147 
  148 static int max_page_launder=100;
  149 #if defined(NO_SWAPPING)
  150 static int vm_swap_enabled=0;
  151 static int vm_swap_idle_enabled=0;
  152 #else
  153 static int vm_swap_enabled=1;
  154 static int vm_swap_idle_enabled=0;
  155 #endif
  156 
  157 SYSCTL_INT(_vm, VM_PAGEOUT_ALGORITHM, pageout_algorithm,
  158         CTLFLAG_RW, &vm_pageout_algorithm_lru, 0, "LRU page mgmt");
  159 
  160 SYSCTL_INT(_vm, OID_AUTO, pageout_stats_max,
  161         CTLFLAG_RW, &vm_pageout_stats_max, 0, "Max pageout stats scan length");
  162 
  163 SYSCTL_INT(_vm, OID_AUTO, pageout_full_stats_interval,
  164         CTLFLAG_RW, &vm_pageout_full_stats_interval, 0, "Interval for full stats scan");
  165 
  166 SYSCTL_INT(_vm, OID_AUTO, pageout_stats_interval,
  167         CTLFLAG_RW, &vm_pageout_stats_interval, 0, "Interval for partial stats scan");
  168 
  169 SYSCTL_INT(_vm, OID_AUTO, pageout_stats_free_max,
  170         CTLFLAG_RW, &vm_pageout_stats_free_max, 0, "Not implemented");
  171 
  172 #if defined(NO_SWAPPING)
  173 SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled,
  174         CTLFLAG_RD, &vm_swap_enabled, 0, "");
  175 SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled,
  176         CTLFLAG_RD, &vm_swap_idle_enabled, 0, "");
  177 #else
  178 SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled,
  179         CTLFLAG_RW, &vm_swap_enabled, 0, "Enable entire process swapout");
  180 SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled,
  181         CTLFLAG_RW, &vm_swap_idle_enabled, 0, "Allow swapout on idle criteria");
  182 #endif
  183 
  184 SYSCTL_INT(_vm, OID_AUTO, defer_swapspace_pageouts,
  185         CTLFLAG_RW, &defer_swap_pageouts, 0, "Give preference to dirty pages in mem");
  186 
  187 SYSCTL_INT(_vm, OID_AUTO, disable_swapspace_pageouts,
  188         CTLFLAG_RW, &disable_swap_pageouts, 0, "Disallow swapout of dirty pages");
  189 
  190 SYSCTL_INT(_vm, OID_AUTO, max_page_launder,
  191         CTLFLAG_RW, &max_page_launder, 0, "Maximum number of pages to clean per pass");
  192 
  193 
  194 #define VM_PAGEOUT_PAGE_COUNT 16
  195 int vm_pageout_page_count = VM_PAGEOUT_PAGE_COUNT;
  196 
  197 int vm_page_max_wired;          /* XXX max # of wired pages system-wide */
  198 
  199 #if !defined(NO_SWAPPING)
  200 typedef void freeer_fcn_t __P((vm_map_t, vm_object_t, vm_pindex_t, int));
  201 static void vm_pageout_map_deactivate_pages __P((vm_map_t, vm_pindex_t));
  202 static freeer_fcn_t vm_pageout_object_deactivate_pages;
  203 static void vm_req_vmdaemon __P((void));
  204 #endif
  205 static void vm_pageout_page_stats(void);
  206 void pmap_collect(void);
  207 
  208 /*
  209  * vm_pageout_clean:
  210  *
  211  * Clean the page and remove it from the laundry.
  212  * 
  213  * We set the busy bit to cause potential page faults on this page to
  214  * block.  Note the careful timing, however, the busy bit isn't set till
  215  * late and we cannot do anything that will mess with the page.
  216  */
  217 
  218 static int
  219 vm_pageout_clean(m)
  220         vm_page_t m;
  221 {
  222         register vm_object_t object;
  223         vm_page_t mc[2*vm_pageout_page_count];
  224         int pageout_count;
  225         int i, forward_okay, backward_okay, page_base;
  226         vm_pindex_t pindex = m->pindex;
  227 
  228         object = m->object;
  229 
  230         /*
  231          * It doesn't cost us anything to pageout OBJT_DEFAULT or OBJT_SWAP
  232          * with the new swapper, but we could have serious problems paging
  233          * out other object types if there is insufficient memory.  
  234          *
  235          * Unfortunately, checking free memory here is far too late, so the
  236          * check has been moved up a procedural level.
  237          */
  238 
  239 #if 1
  240         /*
  241          * If not OBJT_SWAP, additional memory may be needed to do the pageout.
  242          * Try to avoid the deadlock.
  243          */
  244         if ((object->type == OBJT_DEFAULT) &&
  245             ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min))
  246                 return 0;
  247 #endif
  248 
  249         /*
  250          * Don't mess with the page if it's busy.
  251          */
  252         if ((m->hold_count != 0) ||
  253             ((m->busy != 0) || (m->flags & PG_BUSY)))
  254                 return 0;
  255 
  256 #if 0
  257         /*
  258          * XXX REMOVED XXX.  vm_object_collapse() can block, which can
  259          * change the page state.  Calling vm_object_collapse() might also
  260          * destroy or rename the page because we have not busied it yet!!!
  261          * So this code segment is removed.
  262          */
  263         /*
  264          * Try collapsing before it's too late.   XXX huh?  Why are we doing
  265          * this here?
  266          */
  267         if (object->backing_object) {
  268                 vm_object_collapse(object);
  269         }
  270 #endif
  271 
  272         mc[vm_pageout_page_count] = m;
  273         pageout_count = 1;
  274         page_base = vm_pageout_page_count;
  275         forward_okay = TRUE;
  276         if (pindex != 0)
  277                 backward_okay = TRUE;
  278         else
  279                 backward_okay = FALSE;
  280         /*
  281          * Scan object for clusterable pages.
  282          *
  283          * We can cluster ONLY if: ->> the page is NOT
  284          * clean, wired, busy, held, or mapped into a
  285          * buffer, and one of the following:
  286          * 1) The page is inactive, or a seldom used
  287          *    active page.
  288          * -or-
  289          * 2) we force the issue.
  290          */
  291         for (i = 1; (i < vm_pageout_page_count) && (forward_okay || backward_okay); i++) {
  292                 vm_page_t p;
  293 
  294                 /*
  295                  * See if forward page is clusterable.
  296                  */
  297                 if (forward_okay) {
  298                         /*
  299                          * Stop forward scan at end of object.
  300                          */
  301                         if ((pindex + i) > object->size) {
  302                                 forward_okay = FALSE;
  303                                 goto do_backward;
  304                         }
  305                         p = vm_page_lookup(object, pindex + i);
  306                         if (p) {
  307                                 if (((p->queue - p->pc) == PQ_CACHE) ||
  308                                         (p->flags & PG_BUSY) || p->busy) {
  309                                         forward_okay = FALSE;
  310                                         goto do_backward;
  311                                 }
  312                                 vm_page_test_dirty(p);
  313                                 if ((p->dirty & p->valid) != 0 &&
  314                                     (p->queue == PQ_INACTIVE) &&
  315                                     (p->wire_count == 0) &&
  316                                     (p->hold_count == 0)) {
  317                                         mc[vm_pageout_page_count + i] = p;
  318                                         pageout_count++;
  319                                         if (pageout_count == vm_pageout_page_count)
  320                                                 break;
  321                                 } else {
  322                                         forward_okay = FALSE;
  323                                 }
  324                         } else {
  325                                 forward_okay = FALSE;
  326                         }
  327                 }
  328 do_backward:
  329                 /*
  330                  * See if backward page is clusterable.
  331                  */
  332                 if (backward_okay) {
  333                         /*
  334                          * Stop backward scan at beginning of object.
  335                          */
  336                         if ((pindex - i) == 0) {
  337                                 backward_okay = FALSE;
  338                         }
  339                         p = vm_page_lookup(object, pindex - i);
  340                         if (p) {
  341                                 if (((p->queue - p->pc) == PQ_CACHE) ||
  342                                         (p->flags & PG_BUSY) || p->busy) {
  343                                         backward_okay = FALSE;
  344                                         continue;
  345                                 }
  346                                 vm_page_test_dirty(p);
  347                                 if ((p->dirty & p->valid) != 0 &&
  348                                     (p->queue == PQ_INACTIVE) &&
  349                                     (p->wire_count == 0) &&
  350                                     (p->hold_count == 0)) {
  351                                         mc[vm_pageout_page_count - i] = p;
  352                                         pageout_count++;
  353                                         page_base--;
  354                                         if (pageout_count == vm_pageout_page_count)
  355                                                 break;
  356                                 } else {
  357                                         backward_okay = FALSE;
  358                                 }
  359                         } else {
  360                                 backward_okay = FALSE;
  361                         }
  362                 }
  363         }
  364 
  365         /*
  366          * we allow reads during pageouts...
  367          */
  368         return vm_pageout_flush(&mc[page_base], pageout_count, 0);
  369 }
  370 
  371 /*
  372  * vm_pageout_flush() - launder the given pages
  373  *
  374  *      The given pages are laundered.  Note that we setup for the start of
  375  *      I/O ( i.e. busy the page ), mark it read-only, and bump the object
  376  *      reference count all in here rather then in the parent.  If we want
  377  *      the parent to do more sophisticated things we may have to change
  378  *      the ordering.
  379  */
  380 
  381 int
  382 vm_pageout_flush(mc, count, flags)
  383         vm_page_t *mc;
  384         int count;
  385         int flags;
  386 {
  387         register vm_object_t object;
  388         int pageout_status[count];
  389         int numpagedout = 0;
  390         int i;
  391 
  392         /*
  393          * Initiate I/O.  Bump the vm_page_t->busy counter and
  394          * mark the pages read-only.
  395          *
  396          * We do not have to fixup the clean/dirty bits here... we can
  397          * allow the pager to do it after the I/O completes.
  398          */
  399 
  400         for (i = 0; i < count; i++) {
  401                 vm_page_io_start(mc[i]);
  402                 vm_page_protect(mc[i], VM_PROT_READ);
  403         }
  404 
  405         object = mc[0]->object;
  406         vm_object_pip_add(object, count);
  407 
  408         vm_pager_put_pages(object, mc, count,
  409             (flags | ((object == kernel_object) ? OBJPC_SYNC : 0)),
  410             pageout_status);
  411 
  412         for (i = 0; i < count; i++) {
  413                 vm_page_t mt = mc[i];
  414 
  415                 switch (pageout_status[i]) {
  416                 case VM_PAGER_OK:
  417                         numpagedout++;
  418                         break;
  419                 case VM_PAGER_PEND:
  420                         numpagedout++;
  421                         break;
  422                 case VM_PAGER_BAD:
  423                         /*
  424                          * Page outside of range of object. Right now we
  425                          * essentially lose the changes by pretending it
  426                          * worked.
  427                          */
  428                         pmap_clear_modify(VM_PAGE_TO_PHYS(mt));
  429                         mt->dirty = 0;
  430                         break;
  431                 case VM_PAGER_ERROR:
  432                 case VM_PAGER_FAIL:
  433                         /*
  434                          * If page couldn't be paged out, then reactivate the
  435                          * page so it doesn't clog the inactive list.  (We
  436                          * will try paging out it again later).
  437                          */
  438                         vm_page_activate(mt);
  439                         break;
  440                 case VM_PAGER_AGAIN:
  441                         break;
  442                 }
  443 
  444                 /*
  445                  * If the operation is still going, leave the page busy to
  446                  * block all other accesses. Also, leave the paging in
  447                  * progress indicator set so that we don't attempt an object
  448                  * collapse.
  449                  */
  450                 if (pageout_status[i] != VM_PAGER_PEND) {
  451                         vm_object_pip_wakeup(object);
  452                         vm_page_io_finish(mt);
  453                 }
  454         }
  455         return numpagedout;
  456 }
  457 
  458 #if !defined(NO_SWAPPING)
  459 /*
  460  *      vm_pageout_object_deactivate_pages
  461  *
  462  *      deactivate enough pages to satisfy the inactive target
  463  *      requirements or if vm_page_proc_limit is set, then
  464  *      deactivate all of the pages in the object and its
  465  *      backing_objects.
  466  *
  467  *      The object and map must be locked.
  468  */
  469 static void
  470 vm_pageout_object_deactivate_pages(map, object, desired, map_remove_only)
  471         vm_map_t map;
  472         vm_object_t object;
  473         vm_pindex_t desired;
  474         int map_remove_only;
  475 {
  476         register vm_page_t p, next;
  477         int rcount;
  478         int remove_mode;
  479         int s;
  480 
  481         if (object->type == OBJT_DEVICE)
  482                 return;
  483 
  484         while (object) {
  485                 if (vm_map_pmap(map)->pm_stats.resident_count <= desired)
  486                         return;
  487                 if (object->paging_in_progress)
  488                         return;
  489 
  490                 remove_mode = map_remove_only;
  491                 if (object->shadow_count > 1)
  492                         remove_mode = 1;
  493         /*
  494          * scan the objects entire memory queue
  495          */
  496                 rcount = object->resident_page_count;
  497                 p = TAILQ_FIRST(&object->memq);
  498                 while (p && (rcount-- > 0)) {
  499                         int actcount;
  500                         if (vm_map_pmap(map)->pm_stats.resident_count <= desired)
  501                                 return;
  502                         next = TAILQ_NEXT(p, listq);
  503                         cnt.v_pdpages++;
  504                         if (p->wire_count != 0 ||
  505                             p->hold_count != 0 ||
  506                             p->busy != 0 ||
  507                             (p->flags & PG_BUSY) ||
  508                             !pmap_page_exists(vm_map_pmap(map), VM_PAGE_TO_PHYS(p))) {
  509                                 p = next;
  510                                 continue;
  511                         }
  512 
  513                         actcount = pmap_ts_referenced(VM_PAGE_TO_PHYS(p));
  514                         if (actcount) {
  515                                 vm_page_flag_set(p, PG_REFERENCED);
  516                         } else if (p->flags & PG_REFERENCED) {
  517                                 actcount = 1;
  518                         }
  519 
  520                         if ((p->queue != PQ_ACTIVE) &&
  521                                 (p->flags & PG_REFERENCED)) {
  522                                 vm_page_activate(p);
  523                                 p->act_count += actcount;
  524                                 vm_page_flag_clear(p, PG_REFERENCED);
  525                         } else if (p->queue == PQ_ACTIVE) {
  526                                 if ((p->flags & PG_REFERENCED) == 0) {
  527                                         p->act_count -= min(p->act_count, ACT_DECLINE);
  528                                         if (!remove_mode && (vm_pageout_algorithm_lru || (p->act_count == 0))) {
  529                                                 vm_page_protect(p, VM_PROT_NONE);
  530                                                 vm_page_deactivate(p);
  531                                         } else {
  532                                                 s = splvm();
  533                                                 TAILQ_REMOVE(&vm_page_queue_active, p, pageq);
  534                                                 TAILQ_INSERT_TAIL(&vm_page_queue_active, p, pageq);
  535                                                 splx(s);
  536                                         }
  537                                 } else {
  538                                         vm_page_activate(p);
  539                                         vm_page_flag_clear(p, PG_REFERENCED);
  540                                         if (p->act_count < (ACT_MAX - ACT_ADVANCE))
  541                                                 p->act_count += ACT_ADVANCE;
  542                                         s = splvm();
  543                                         TAILQ_REMOVE(&vm_page_queue_active, p, pageq);
  544                                         TAILQ_INSERT_TAIL(&vm_page_queue_active, p, pageq);
  545                                         splx(s);
  546                                 }
  547                         } else if (p->queue == PQ_INACTIVE) {
  548                                 vm_page_protect(p, VM_PROT_NONE);
  549                         }
  550                         p = next;
  551                 }
  552                 object = object->backing_object;
  553         }
  554         return;
  555 }
  556 
  557 /*
  558  * deactivate some number of pages in a map, try to do it fairly, but
  559  * that is really hard to do.
  560  */
  561 static void
  562 vm_pageout_map_deactivate_pages(map, desired)
  563         vm_map_t map;
  564         vm_pindex_t desired;
  565 {
  566         vm_map_entry_t tmpe;
  567         vm_object_t obj, bigobj;
  568 
  569         if (lockmgr(&map->lock, LK_EXCLUSIVE | LK_NOWAIT, (void *)0, curproc)) {
  570                 return;
  571         }
  572 
  573         bigobj = NULL;
  574 
  575         /*
  576          * first, search out the biggest object, and try to free pages from
  577          * that.
  578          */
  579         tmpe = map->header.next;
  580         while (tmpe != &map->header) {
  581                 if ((tmpe->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) == 0) {
  582                         obj = tmpe->object.vm_object;
  583                         if ((obj != NULL) && (obj->shadow_count <= 1) &&
  584                                 ((bigobj == NULL) ||
  585                                  (bigobj->resident_page_count < obj->resident_page_count))) {
  586                                 bigobj = obj;
  587                         }
  588                 }
  589                 tmpe = tmpe->next;
  590         }
  591 
  592         if (bigobj)
  593                 vm_pageout_object_deactivate_pages(map, bigobj, desired, 0);
  594 
  595         /*
  596          * Next, hunt around for other pages to deactivate.  We actually
  597          * do this search sort of wrong -- .text first is not the best idea.
  598          */
  599         tmpe = map->header.next;
  600         while (tmpe != &map->header) {
  601                 if (vm_map_pmap(map)->pm_stats.resident_count <= desired)
  602                         break;
  603                 if ((tmpe->eflags & (MAP_ENTRY_IS_A_MAP|MAP_ENTRY_IS_SUB_MAP)) == 0) {
  604                         obj = tmpe->object.vm_object;
  605                         if (obj)
  606                                 vm_pageout_object_deactivate_pages(map, obj, desired, 0);
  607                 }
  608                 tmpe = tmpe->next;
  609         };
  610 
  611         /*
  612          * Remove all mappings if a process is swapped out, this will free page
  613          * table pages.
  614          */
  615         if (desired == 0)
  616                 pmap_remove(vm_map_pmap(map),
  617                         VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
  618         vm_map_unlock(map);
  619         return;
  620 }
  621 #endif
  622 
  623 /*
  624  * Don't try to be fancy - being fancy can lead to VOP_LOCK's and therefore
  625  * to vnode deadlocks.  We only do it for OBJT_DEFAULT and OBJT_SWAP objects
  626  * which we know can be trivially freed.
  627  */
  628 
  629 void
  630 vm_pageout_page_free(vm_page_t m) {
  631         vm_object_t object = m->object;
  632         int type = object->type;
  633 
  634         if (type == OBJT_SWAP || type == OBJT_DEFAULT)
  635                 vm_object_reference(object);
  636         vm_page_busy(m);
  637         vm_page_protect(m, VM_PROT_NONE);
  638         vm_page_free(m);
  639         if (type == OBJT_SWAP || type == OBJT_DEFAULT)
  640                 vm_object_deallocate(object);
  641 }
  642 
  643 /*
  644  *      vm_pageout_scan does the dirty work for the pageout daemon.
  645  */
  646 static int
  647 vm_pageout_scan()
  648 {
  649         vm_page_t m, next;
  650         int page_shortage, maxscan, pcount;
  651         int addl_page_shortage, addl_page_shortage_init;
  652         int maxlaunder;
  653         int launder_loop = 0;
  654         struct proc *p, *bigproc;
  655         vm_offset_t size, bigsize;
  656         vm_object_t object;
  657         int force_wakeup = 0;
  658         int actcount;
  659         int vnodes_skipped = 0;
  660         int s;
  661 
  662         /*
  663          * Do whatever cleanup that the pmap code can.
  664          */
  665         pmap_collect();
  666 
  667         addl_page_shortage_init = vm_pageout_deficit;
  668         vm_pageout_deficit = 0;
  669 
  670         if (max_page_launder == 0)
  671                 max_page_launder = 1;
  672 
  673         /*
  674          * Calculate the number of pages we want to either free or move
  675          * to the cache.
  676          */
  677 
  678         page_shortage = (cnt.v_free_target + cnt.v_cache_min) -
  679             (cnt.v_free_count + cnt.v_cache_count);
  680         page_shortage += addl_page_shortage_init;
  681 
  682         /*
  683          * Figure out what to do with dirty pages when they are encountered.
  684          * Assume that 1/3 of the pages on the inactive list are clean.  If
  685          * we think we can reach our target, disable laundering (do not
  686          * clean any dirty pages).  If we miss the target we will loop back
  687          * up and do a laundering run.
  688          */
  689 
  690         if (cnt.v_inactive_count / 3 > page_shortage) {
  691                 maxlaunder = 0;
  692                 launder_loop = 0;
  693         } else {
  694                 maxlaunder = 
  695                     (cnt.v_inactive_target > max_page_launder) ?
  696                     max_page_launder : cnt.v_inactive_target;
  697                 launder_loop = 1;
  698         }
  699 
  700         /*
  701          * Start scanning the inactive queue for pages we can move to the
  702          * cache or free.  The scan will stop when the target is reached or
  703          * we have scanned the entire inactive queue.
  704          */
  705 
  706 rescan0:
  707         addl_page_shortage = addl_page_shortage_init;
  708         maxscan = cnt.v_inactive_count;
  709         for (
  710             m = TAILQ_FIRST(&vm_page_queue_inactive);
  711             m != NULL && maxscan-- > 0 && page_shortage > 0;
  712             m = next
  713         ) {
  714 
  715                 cnt.v_pdpages++;
  716 
  717                 if (m->queue != PQ_INACTIVE) {
  718                         goto rescan0;
  719                 }
  720 
  721                 next = TAILQ_NEXT(m, pageq);
  722 
  723                 if (m->hold_count) {
  724                         s = splvm();
  725                         TAILQ_REMOVE(&vm_page_queue_inactive, m, pageq);
  726                         TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq);
  727                         splx(s);
  728                         addl_page_shortage++;
  729                         continue;
  730                 }
  731                 /*
  732                  * Dont mess with busy pages, keep in the front of the
  733                  * queue, most likely are being paged out.
  734                  */
  735                 if (m->busy || (m->flags & PG_BUSY)) {
  736                         addl_page_shortage++;
  737                         continue;
  738                 }
  739 
  740                 /*
  741                  * If the object is not being used, we ignore previous 
  742                  * references.
  743                  */
  744                 if (m->object->ref_count == 0) {
  745                         vm_page_flag_clear(m, PG_REFERENCED);
  746                         pmap_clear_reference(VM_PAGE_TO_PHYS(m));
  747 
  748                 /*
  749                  * Otherwise, if the page has been referenced while in the 
  750                  * inactive queue, we bump the "activation count" upwards, 
  751                  * making it less likely that the page will be added back to 
  752                  * the inactive queue prematurely again.  Here we check the 
  753                  * page tables (or emulated bits, if any), given the upper 
  754                  * level VM system not knowing anything about existing 
  755                  * references.
  756                  */
  757                 } else if (((m->flags & PG_REFERENCED) == 0) &&
  758                         (actcount = pmap_ts_referenced(VM_PAGE_TO_PHYS(m)))) {
  759                         vm_page_activate(m);
  760                         m->act_count += (actcount + ACT_ADVANCE);
  761                         continue;
  762                 }
  763 
  764                 /*
  765                  * If the upper level VM system knows about any page 
  766                  * references, we activate the page.  We also set the 
  767                  * "activation count" higher than normal so that we will less 
  768                  * likely place pages back onto the inactive queue again.
  769                  */
  770                 if ((m->flags & PG_REFERENCED) != 0) {
  771                         vm_page_flag_clear(m, PG_REFERENCED);
  772                         actcount = pmap_ts_referenced(VM_PAGE_TO_PHYS(m));
  773                         vm_page_activate(m);
  774                         m->act_count += (actcount + ACT_ADVANCE + 1);
  775                         continue;
  776                 }
  777 
  778                 /*
  779                  * If the upper level VM system doesn't know anything about 
  780                  * the page being dirty, we have to check for it again.  As 
  781                  * far as the VM code knows, any partially dirty pages are 
  782                  * fully dirty.
  783                  */
  784                 if (m->dirty == 0) {
  785                         vm_page_test_dirty(m);
  786                 } else {
  787                         m->dirty = VM_PAGE_BITS_ALL;
  788                 }
  789 
  790                 /*
  791                  * Invalid pages can be easily freed
  792                  */
  793                 if (m->valid == 0) {
  794                         vm_pageout_page_free(m);
  795                         cnt.v_dfree++;
  796                         --page_shortage;
  797 
  798                 /*
  799                  * Clean pages can be placed onto the cache queue.
  800                  */
  801                 } else if (m->dirty == 0) {
  802                         vm_page_cache(m);
  803                         --page_shortage;
  804 
  805                 /*
  806                  * Dirty pages need to be paged out.  Note that we clean
  807                  * only a limited number of pages per pagedaemon pass.
  808                  */
  809                 } else if (maxlaunder > 0) {
  810                         int written;
  811                         int swap_pageouts_ok;
  812                         struct vnode *vp = NULL;
  813 
  814                         object = m->object;
  815 
  816                         if ((object->type != OBJT_SWAP) && (object->type != OBJT_DEFAULT)) {
  817                                 swap_pageouts_ok = 1;
  818                         } else {
  819                                 swap_pageouts_ok = !(defer_swap_pageouts || disable_swap_pageouts);
  820                                 swap_pageouts_ok |= (!disable_swap_pageouts && defer_swap_pageouts &&
  821                                         (cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min);
  822                                                                                 
  823                         }
  824 
  825                         /*
  826                          * We don't bother paging objects that are "dead".  
  827                          * Those objects are in a "rundown" state.
  828                          */
  829                         if (!swap_pageouts_ok || (object->flags & OBJ_DEAD)) {
  830                                 s = splvm();
  831                                 TAILQ_REMOVE(&vm_page_queue_inactive, m, pageq);
  832                                 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq);
  833                                 splx(s);
  834                                 continue;
  835                         }
  836 
  837                         /*
  838                          * For now we protect against potential memory
  839                          * deadlocks by requiring significant memory to be 
  840                          * free if the object is not OBJT_DEFAULT or OBJT_SWAP.
  841                          * We do not 'trust' any other object type to operate
  842                          * with low memory, not even OBJT_DEVICE.  The VM
  843                          * allocator will special case allocations done by
  844                          * the pageout daemon so the check below actually 
  845                          * does have some hysteresis in it.  It isn't the best
  846                          * solution, though.
  847                          */
  848 
  849                         if (
  850                             object->type != OBJT_DEFAULT &&
  851                             object->type != OBJT_SWAP &&
  852                             cnt.v_free_count < cnt.v_free_reserved
  853                         ) {
  854                                 s = splvm();
  855                                 TAILQ_REMOVE(&vm_page_queue_inactive, m, pageq);
  856                                 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq);
  857                                 splx(s);
  858                                 continue;
  859                         }
  860 
  861                         /*
  862                          * Presumably we have sufficient free memory to do
  863                          * the more sophisticated checks and locking required
  864                          * for vnodes.
  865                          *
  866                          * The object is already known NOT to be dead.  The
  867                          * vget() may still block, though, because 
  868                          * VOP_ISLOCKED() doesn't check to see if an inode
  869                          * (v_data) is associated with the vnode.  If it isn't,
  870                          * vget() will load in it from disk.  Worse, vget()
  871                          * may actually get stuck waiting on "inode" if another
  872                          * process is in the process of bringing the inode in.
  873                          * This is bad news for us either way.
  874                          *
  875                          * So for the moment we check v_data == NULL as a
  876                          * workaround.  This means that vnodes which do not
  877                          * use v_data in the way we expect probably will not
  878                          * wind up being paged out by the pager and it will be
  879                          * up to the syncer to get them.  That's better then
  880                          * us blocking here.
  881                          *
  882                          * This whole code section is bogus - we need to fix
  883                          * the vnode pager to handle vm_page_t's without us
  884                          * having to do any sophisticated VOP tests.
  885                          */
  886 
  887                         if (object->type == OBJT_VNODE) {
  888                                 vp = object->handle;
  889 
  890                                 if (VOP_ISLOCKED(vp) ||
  891                                     vp->v_data == NULL ||
  892                                     vget(vp, LK_EXCLUSIVE|LK_NOOBJ, curproc)) {
  893                                         if ((m->queue == PQ_INACTIVE) &&
  894                                                 (m->hold_count == 0) &&
  895                                                 (m->busy == 0) &&
  896                                                 (m->flags & PG_BUSY) == 0) {
  897                                                 s = splvm();
  898                                                 TAILQ_REMOVE(&vm_page_queue_inactive, m, pageq);
  899                                                 TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq);
  900                                                 splx(s);
  901                                         }
  902                                         if (object->flags & OBJ_MIGHTBEDIRTY)
  903                                                 vnodes_skipped++;
  904                                         continue;
  905                                 }
  906 
  907                                 /*
  908                                  * The page might have been moved to another queue
  909                                  * during potential blocking in vget() above.
  910                                  */
  911                                 if (m->queue != PQ_INACTIVE) {
  912                                         if (object->flags & OBJ_MIGHTBEDIRTY)
  913                                                 vnodes_skipped++;
  914                                         vput(vp);
  915                                         continue;
  916                                 }
  917         
  918                                 /*
  919                                  * The page may have been busied during the blocking in
  920                                  * vput();  We don't move the page back onto the end of
  921                                  * the queue so that statistics are more correct if we don't.
  922                                  */
  923                                 if (m->busy || (m->flags & PG_BUSY)) {
  924                                         vput(vp);
  925                                         continue;
  926                                 }
  927 
  928                                 /*
  929                                  * If the page has become held, then skip it
  930                                  */
  931                                 if (m->hold_count) {
  932                                         s = splvm();
  933                                         TAILQ_REMOVE(&vm_page_queue_inactive, m, pageq);
  934                                         TAILQ_INSERT_TAIL(&vm_page_queue_inactive, m, pageq);
  935                                         splx(s);
  936                                         if (object->flags & OBJ_MIGHTBEDIRTY)
  937                                                 vnodes_skipped++;
  938                                         vput(vp);
  939                                         continue;
  940                                 }
  941                         }
  942 
  943                         /*
  944                          * If a page is dirty, then it is either being washed
  945                          * (but not yet cleaned) or it is still in the
  946                          * laundry.  If it is still in the laundry, then we
  947                          * start the cleaning operation.
  948                          */
  949                         written = vm_pageout_clean(m);
  950                         if (vp)
  951                                 vput(vp);
  952 
  953                         maxlaunder -= written;
  954                 }
  955         }
  956 
  957         /*
  958          * If we still have a page shortage and we didn't launder anything,
  959          * run the inactive scan again and launder something this time.
  960          */
  961 
  962         if (launder_loop == 0 && page_shortage > 0) {
  963                 launder_loop = 1;
  964                 maxlaunder = 
  965                     (cnt.v_inactive_target > max_page_launder) ?
  966                     max_page_launder : cnt.v_inactive_target;
  967                 goto rescan0;
  968         }
  969 
  970         /*
  971          * Compute the page shortage from the point of view of having to
  972          * move pages from the active queue to the inactive queue.
  973          */
  974 
  975         page_shortage = (cnt.v_inactive_target + cnt.v_cache_min) -
  976             (cnt.v_free_count + cnt.v_inactive_count + cnt.v_cache_count);
  977         page_shortage += addl_page_shortage;
  978 
  979         /*
  980          * Scan the active queue for things we can deactivate
  981          */
  982 
  983         pcount = cnt.v_active_count;
  984         m = TAILQ_FIRST(&vm_page_queue_active);
  985 
  986         while ((m != NULL) && (pcount-- > 0) && (page_shortage > 0)) {
  987 
  988                 /*
  989                  * This is a consistancy check, and should likely be a panic
  990                  * or warning.
  991                  */
  992                 if (m->queue != PQ_ACTIVE) {
  993                         break;
  994                 }
  995 
  996                 next = TAILQ_NEXT(m, pageq);
  997                 /*
  998                  * Don't deactivate pages that are busy.
  999                  */
 1000                 if ((m->busy != 0) ||
 1001                     (m->flags & PG_BUSY) ||
 1002                     (m->hold_count != 0)) {
 1003                         s = splvm();
 1004                         TAILQ_REMOVE(&vm_page_queue_active, m, pageq);
 1005                         TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq);
 1006                         splx(s);
 1007                         m = next;
 1008                         continue;
 1009                 }
 1010 
 1011                 /*
 1012                  * The count for pagedaemon pages is done after checking the
 1013                  * page for eligbility...
 1014                  */
 1015                 cnt.v_pdpages++;
 1016 
 1017                 /*
 1018                  * Check to see "how much" the page has been used.
 1019                  */
 1020                 actcount = 0;
 1021                 if (m->object->ref_count != 0) {
 1022                         if (m->flags & PG_REFERENCED) {
 1023                                 actcount += 1;
 1024                         }
 1025                         actcount += pmap_ts_referenced(VM_PAGE_TO_PHYS(m));
 1026                         if (actcount) {
 1027                                 m->act_count += ACT_ADVANCE + actcount;
 1028                                 if (m->act_count > ACT_MAX)
 1029                                         m->act_count = ACT_MAX;
 1030                         }
 1031                 }
 1032 
 1033                 /*
 1034                  * Since we have "tested" this bit, we need to clear it now.
 1035                  */
 1036                 vm_page_flag_clear(m, PG_REFERENCED);
 1037 
 1038                 /*
 1039                  * Only if an object is currently being used, do we use the
 1040                  * page activation count stats.
 1041                  */
 1042                 if (actcount && (m->object->ref_count != 0)) {
 1043                         s = splvm();
 1044                         TAILQ_REMOVE(&vm_page_queue_active, m, pageq);
 1045                         TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq);
 1046                         splx(s);
 1047                 } else {
 1048                         m->act_count -= min(m->act_count, ACT_DECLINE);
 1049                         if (vm_pageout_algorithm_lru ||
 1050                                 (m->object->ref_count == 0) || (m->act_count == 0)) {
 1051                                 page_shortage--;
 1052                                 if (m->object->ref_count == 0) {
 1053                                         vm_page_protect(m, VM_PROT_NONE);
 1054                                         if (m->dirty == 0)
 1055                                                 vm_page_cache(m);
 1056                                         else
 1057                                                 vm_page_deactivate(m);
 1058                                 } else {
 1059                                         vm_page_deactivate(m);
 1060                                 }
 1061                         } else {
 1062                                 s = splvm();
 1063                                 TAILQ_REMOVE(&vm_page_queue_active, m, pageq);
 1064                                 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq);
 1065                                 splx(s);
 1066                         }
 1067                 }
 1068                 m = next;
 1069         }
 1070 
 1071         s = splvm();
 1072 
 1073         /*
 1074          * We try to maintain some *really* free pages, this allows interrupt
 1075          * code to be guaranteed space.  Since both cache and free queues 
 1076          * are considered basically 'free', moving pages from cache to free
 1077          * does not effect other calculations.
 1078          */
 1079 
 1080         while (cnt.v_free_count < cnt.v_free_reserved) {
 1081                 static int cache_rover = 0;
 1082                 m = vm_page_list_find(PQ_CACHE, cache_rover);
 1083                 if (!m)
 1084                         break;
 1085                 if ((m->flags & PG_BUSY) || m->busy || m->hold_count || m->wire_count) {
 1086 #ifdef INVARIANTS
 1087                         printf("Warning: busy page %p found in cache\n", m);
 1088 #endif
 1089                         vm_page_deactivate(m);
 1090                         continue;
 1091                 }
 1092                 cache_rover = (cache_rover + PQ_PRIME2) & PQ_L2_MASK;
 1093                 vm_pageout_page_free(m);
 1094                 cnt.v_dfree++;
 1095         }
 1096         splx(s);
 1097 
 1098 #if !defined(NO_SWAPPING)
 1099         /*
 1100          * Idle process swapout -- run once per second.
 1101          */
 1102         if (vm_swap_idle_enabled) {
 1103                 static long lsec;
 1104                 if (time_second != lsec) {
 1105                         vm_pageout_req_swapout |= VM_SWAP_IDLE;
 1106                         vm_req_vmdaemon();
 1107                         lsec = time_second;
 1108                 }
 1109         }
 1110 #endif
 1111                 
 1112         /*
 1113          * If we didn't get enough free pages, and we have skipped a vnode
 1114          * in a writeable object, wakeup the sync daemon.  And kick swapout
 1115          * if we did not get enough free pages.
 1116          */
 1117         if ((cnt.v_cache_count + cnt.v_free_count) <
 1118                 (cnt.v_free_target + cnt.v_cache_min) ) {
 1119                 if (vnodes_skipped &&
 1120                     (cnt.v_cache_count + cnt.v_free_count) < cnt.v_free_min) {
 1121                         if (!vfs_update_wakeup) {
 1122                                 vfs_update_wakeup = 1;
 1123                                 wakeup(&vfs_update_wakeup);
 1124                         }
 1125                 }
 1126 #if !defined(NO_SWAPPING)
 1127                 if (vm_swap_enabled &&
 1128                         (cnt.v_free_count + cnt.v_cache_count < cnt.v_free_target)) {
 1129                         vm_req_vmdaemon();
 1130                         vm_pageout_req_swapout |= VM_SWAP_NORMAL;
 1131                 }
 1132 #endif
 1133         }
 1134 
 1135         /*
 1136          * make sure that we have swap space -- if we are low on memory and
 1137          * swap -- then kill the biggest process.
 1138          */
 1139         if ((vm_swap_size == 0 || swap_pager_full) &&
 1140             ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min)) {
 1141                 bigproc = NULL;
 1142                 bigsize = 0;
 1143                 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
 1144                         /*
 1145                          * if this is a system process, skip it
 1146                          */
 1147                         if ((p->p_flag & (P_SYSTEM|P_NOSWAP)) || (p->p_pid == 1) ||
 1148                             ((p->p_pid < 48) && (vm_swap_size != 0))) {
 1149                                 continue;
 1150                         }
 1151                         /*
 1152                          * if the process is in a non-running type state,
 1153                          * don't touch it.
 1154                          */
 1155                         if (p->p_stat != SRUN && p->p_stat != SSLEEP) {
 1156                                 continue;
 1157                         }
 1158                         /*
 1159                          * get the process size
 1160                          */
 1161                         size = p->p_vmspace->vm_pmap.pm_stats.resident_count;
 1162                         /*
 1163                          * if the this process is bigger than the biggest one
 1164                          * remember it.
 1165                          */
 1166                         if (size > bigsize) {
 1167                                 bigproc = p;
 1168                                 bigsize = size;
 1169                         }
 1170                 }
 1171                 if (bigproc != NULL) {
 1172                         killproc(bigproc, "out of swap space");
 1173                         bigproc->p_estcpu = 0;
 1174                         bigproc->p_nice = PRIO_MIN;
 1175                         resetpriority(bigproc);
 1176                         wakeup(&cnt.v_free_count);
 1177                 }
 1178         }
 1179         return force_wakeup;
 1180 }
 1181 
 1182 /*
 1183  * This routine tries to maintain the pseudo LRU active queue,
 1184  * so that during long periods of time where there is no paging,
 1185  * that some statistic accumlation still occurs.  This code
 1186  * helps the situation where paging just starts to occur.
 1187  */
 1188 static void
 1189 vm_pageout_page_stats()
 1190 {
 1191         int s;
 1192         vm_page_t m,next;
 1193         int pcount,tpcount;             /* Number of pages to check */
 1194         static int fullintervalcount = 0;
 1195         int page_shortage;
 1196 
 1197         page_shortage = (cnt.v_inactive_target + cnt.v_cache_max + cnt.v_free_min) -
 1198             (cnt.v_free_count + cnt.v_inactive_count + cnt.v_cache_count);
 1199         if (page_shortage <= 0)
 1200                 return;
 1201 
 1202         pcount = cnt.v_active_count;
 1203         fullintervalcount += vm_pageout_stats_interval;
 1204         if (fullintervalcount < vm_pageout_full_stats_interval) {
 1205                 tpcount = (vm_pageout_stats_max * cnt.v_active_count) / cnt.v_page_count;
 1206                 if (pcount > tpcount)
 1207                         pcount = tpcount;
 1208         }
 1209 
 1210         m = TAILQ_FIRST(&vm_page_queue_active);
 1211         while ((m != NULL) && (pcount-- > 0)) {
 1212                 int actcount;
 1213 
 1214                 if (m->queue != PQ_ACTIVE) {
 1215                         break;
 1216                 }
 1217 
 1218                 next = TAILQ_NEXT(m, pageq);
 1219                 /*
 1220                  * Don't deactivate pages that are busy.
 1221                  */
 1222                 if ((m->busy != 0) ||
 1223                     (m->flags & PG_BUSY) ||
 1224                     (m->hold_count != 0)) {
 1225                         s = splvm();
 1226                         TAILQ_REMOVE(&vm_page_queue_active, m, pageq);
 1227                         TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq);
 1228                         splx(s);
 1229                         m = next;
 1230                         continue;
 1231                 }
 1232 
 1233                 actcount = 0;
 1234                 if (m->flags & PG_REFERENCED) {
 1235                         vm_page_flag_clear(m, PG_REFERENCED);
 1236                         actcount += 1;
 1237                 }
 1238 
 1239                 actcount += pmap_ts_referenced(VM_PAGE_TO_PHYS(m));
 1240                 if (actcount) {
 1241                         m->act_count += ACT_ADVANCE + actcount;
 1242                         if (m->act_count > ACT_MAX)
 1243                                 m->act_count = ACT_MAX;
 1244                         s = splvm();
 1245                         TAILQ_REMOVE(&vm_page_queue_active, m, pageq);
 1246                         TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq);
 1247                         splx(s);
 1248                 } else {
 1249                         if (m->act_count == 0) {
 1250                                 /*
 1251                                  * We turn off page access, so that we have more accurate
 1252                                  * RSS stats.  We don't do this in the normal page deactivation
 1253                                  * when the system is loaded VM wise, because the cost of
 1254                                  * the large number of page protect operations would be higher
 1255                                  * than the value of doing the operation.
 1256                                  */
 1257                                 vm_page_protect(m, VM_PROT_NONE);
 1258                                 vm_page_deactivate(m);
 1259                         } else {
 1260                                 m->act_count -= min(m->act_count, ACT_DECLINE);
 1261                                 s = splvm();
 1262                                 TAILQ_REMOVE(&vm_page_queue_active, m, pageq);
 1263                                 TAILQ_INSERT_TAIL(&vm_page_queue_active, m, pageq);
 1264                                 splx(s);
 1265                         }
 1266                 }
 1267 
 1268                 m = next;
 1269         }
 1270 }
 1271 
 1272 static int
 1273 vm_pageout_free_page_calc(count)
 1274 vm_size_t count;
 1275 {
 1276         if (count < cnt.v_page_count)
 1277                  return 0;
 1278         /*
 1279          * free_reserved needs to include enough for the largest swap pager
 1280          * structures plus enough for any pv_entry structs when paging.
 1281          */
 1282         if (cnt.v_page_count > 1024)
 1283                 cnt.v_free_min = 4 + (cnt.v_page_count - 1024) / 200;
 1284         else
 1285                 cnt.v_free_min = 4;
 1286         cnt.v_pageout_free_min = (2*MAXBSIZE)/PAGE_SIZE +
 1287                 cnt.v_interrupt_free_min;
 1288         cnt.v_free_reserved = vm_pageout_page_count +
 1289                 cnt.v_pageout_free_min + (count / 768) + PQ_L2_SIZE;
 1290         cnt.v_free_min += cnt.v_free_reserved;
 1291         return 1;
 1292 }
 1293 
 1294 
 1295 /*
 1296  *      vm_pageout is the high level pageout daemon.
 1297  */
 1298 static void
 1299 vm_pageout()
 1300 {
 1301         /*
 1302          * Initialize some paging parameters.
 1303          */
 1304 
 1305         cnt.v_interrupt_free_min = 2;
 1306         if (cnt.v_page_count < 2000)
 1307                 vm_pageout_page_count = 8;
 1308 
 1309         vm_pageout_free_page_calc(cnt.v_page_count);
 1310         /*
 1311          * free_reserved needs to include enough for the largest swap pager
 1312          * structures plus enough for any pv_entry structs when paging.
 1313          */
 1314         if (cnt.v_free_count > 6144)
 1315                 cnt.v_free_target = 3 * cnt.v_free_min + cnt.v_free_reserved;
 1316         else
 1317                 cnt.v_free_target = 2 * cnt.v_free_min + cnt.v_free_reserved;
 1318 
 1319         if (cnt.v_free_count > 2048) {
 1320                 cnt.v_cache_min = cnt.v_free_target;
 1321                 cnt.v_cache_max = 2 * cnt.v_cache_min;
 1322                 cnt.v_inactive_target = (3 * cnt.v_free_target) / 2;
 1323         } else {
 1324                 cnt.v_cache_min = 0;
 1325                 cnt.v_cache_max = 0;
 1326                 cnt.v_inactive_target = cnt.v_free_count / 4;
 1327         }
 1328         if (cnt.v_inactive_target > cnt.v_free_count / 3)
 1329                 cnt.v_inactive_target = cnt.v_free_count / 3;
 1330 
 1331         /* XXX does not really belong here */
 1332         if (vm_page_max_wired == 0)
 1333                 vm_page_max_wired = cnt.v_free_count / 3;
 1334 
 1335         if (vm_pageout_stats_max == 0)
 1336                 vm_pageout_stats_max = cnt.v_free_target;
 1337 
 1338         /*
 1339          * Set interval in seconds for stats scan.
 1340          */
 1341         if (vm_pageout_stats_interval == 0)
 1342                 vm_pageout_stats_interval = 5;
 1343         if (vm_pageout_full_stats_interval == 0)
 1344                 vm_pageout_full_stats_interval = vm_pageout_stats_interval * 4;
 1345         
 1346 
 1347         /*
 1348          * Set maximum free per pass
 1349          */
 1350         if (vm_pageout_stats_free_max == 0)
 1351                 vm_pageout_stats_free_max = 5;
 1352 
 1353         max_page_launder = (cnt.v_page_count > 1800 ? 32 : 16);
 1354 
 1355         swap_pager_swap_init();
 1356         /*
 1357          * The pageout daemon is never done, so loop forever.
 1358          */
 1359         while (TRUE) {
 1360                 int error;
 1361                 int s = splvm();
 1362                 if (!vm_pages_needed ||
 1363                         ((cnt.v_free_count + cnt.v_cache_count) > cnt.v_free_min)) {
 1364                         vm_pages_needed = 0;
 1365                         error = tsleep(&vm_pages_needed,
 1366                                 PVM, "psleep", vm_pageout_stats_interval * hz);
 1367                         if (error && !vm_pages_needed) {
 1368                                 splx(s);
 1369                                 vm_pageout_page_stats();
 1370                                 continue;
 1371                         }
 1372                 } else if (vm_pages_needed) {
 1373                         vm_pages_needed = 0;
 1374                         tsleep(&vm_pages_needed, PVM, "psleep", hz/2);
 1375                 }
 1376 
 1377                 if (vm_pages_needed)
 1378                         cnt.v_pdwakeups++;
 1379                 vm_pages_needed = 0;
 1380                 splx(s);
 1381                 vm_pager_sync();
 1382                 vm_pageout_scan();
 1383                 vm_pageout_deficit = 0;
 1384                 vm_pager_sync();
 1385                 wakeup(&cnt.v_free_count);
 1386         }
 1387 }
 1388 
 1389 void
 1390 pagedaemon_wakeup()
 1391 {
 1392         if (!vm_pages_needed && curproc != pageproc) {
 1393                 vm_pages_needed++;
 1394                 wakeup(&vm_pages_needed);
 1395         }
 1396 }
 1397 
 1398 #if !defined(NO_SWAPPING)
 1399 static void
 1400 vm_req_vmdaemon()
 1401 {
 1402         static int lastrun = 0;
 1403 
 1404         if ((ticks > (lastrun + hz)) || (ticks < lastrun)) {
 1405                 wakeup(&vm_daemon_needed);
 1406                 lastrun = ticks;
 1407         }
 1408 }
 1409 
 1410 static void
 1411 vm_daemon()
 1412 {
 1413         struct proc *p;
 1414 
 1415         while (TRUE) {
 1416                 tsleep(&vm_daemon_needed, PPAUSE, "psleep", 0);
 1417                 if (vm_pageout_req_swapout) {
 1418                         swapout_procs(vm_pageout_req_swapout);
 1419                         vm_pageout_req_swapout = 0;
 1420                 }
 1421                 /*
 1422                  * scan the processes for exceeding their rlimits or if
 1423                  * process is swapped out -- deactivate pages
 1424                  */
 1425 
 1426                 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
 1427                         vm_pindex_t limit, size;
 1428 
 1429                         /*
 1430                          * if this is a system process or if we have already
 1431                          * looked at this process, skip it.
 1432                          */
 1433                         if (p->p_flag & (P_SYSTEM | P_WEXIT)) {
 1434                                 continue;
 1435                         }
 1436                         /*
 1437                          * if the process is in a non-running type state,
 1438                          * don't touch it.
 1439                          */
 1440                         if (p->p_stat != SRUN && p->p_stat != SSLEEP) {
 1441                                 continue;
 1442                         }
 1443                         /*
 1444                          * get a limit
 1445                          */
 1446                         limit = OFF_TO_IDX(
 1447                             qmin(p->p_rlimit[RLIMIT_RSS].rlim_cur,
 1448                                 p->p_rlimit[RLIMIT_RSS].rlim_max));
 1449 
 1450                         /*
 1451                          * let processes that are swapped out really be
 1452                          * swapped out set the limit to nothing (will force a
 1453                          * swap-out.)
 1454                          */
 1455                         if ((p->p_flag & P_INMEM) == 0)
 1456                                 limit = 0;      /* XXX */
 1457 
 1458                         size = p->p_vmspace->vm_pmap.pm_stats.resident_count;
 1459                         if (limit >= 0 && size >= limit) {
 1460                                 vm_pageout_map_deactivate_pages(
 1461                                     &p->p_vmspace->vm_map, limit);
 1462                         }
 1463                 }
 1464         }
 1465 }
 1466 #endif

Cache object: 99fcb0f5e4a770b80675700319b9a7ec


[ 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.