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_reserv.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) 2002-2006 Rice University
    3  * Copyright (c) 2007-2008 Alan L. Cox <alc@cs.rice.edu>
    4  * All rights reserved.
    5  *
    6  * This software was developed for the FreeBSD Project by Alan L. Cox,
    7  * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
    8  *
    9  * Redistribution and use in source and binary forms, with or without
   10  * modification, are permitted provided that the following conditions
   11  * are met:
   12  * 1. Redistributions of source code must retain the above copyright
   13  *    notice, this list of conditions and the following disclaimer.
   14  * 2. Redistributions in binary form must reproduce the above copyright
   15  *    notice, this list of conditions and the following disclaimer in the
   16  *    documentation and/or other materials provided with the distribution.
   17  *
   18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
   19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
   20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
   21  * A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT
   22  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
   23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
   24  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
   25  * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
   26  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
   28  * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
   29  * POSSIBILITY OF SUCH DAMAGE.
   30  */
   31 
   32 /*
   33  *      Superpage reservation management module
   34  */
   35 
   36 #include <sys/cdefs.h>
   37 __FBSDID("$FreeBSD: releng/8.0/sys/vm/vm_reserv.c 190912 2009-04-11 09:09:00Z alc $");
   38 
   39 #include "opt_vm.h"
   40 
   41 #include <sys/param.h>
   42 #include <sys/kernel.h>
   43 #include <sys/lock.h>
   44 #include <sys/malloc.h>
   45 #include <sys/mutex.h>
   46 #include <sys/queue.h>
   47 #include <sys/sbuf.h>
   48 #include <sys/sysctl.h>
   49 #include <sys/systm.h>
   50 
   51 #include <vm/vm.h>
   52 #include <vm/vm_param.h>
   53 #include <vm/vm_object.h>
   54 #include <vm/vm_page.h>
   55 #include <vm/vm_phys.h>
   56 #include <vm/vm_reserv.h>
   57 
   58 /*
   59  * The reservation system supports the speculative allocation of large physical
   60  * pages ("superpages").  Speculative allocation enables the fully-automatic
   61  * utilization of superpages by the virtual memory system.  In other words, no
   62  * programmatic directives are required to use superpages.
   63  */
   64 
   65 #if VM_NRESERVLEVEL > 0
   66 
   67 /*
   68  * The number of small pages that are contained in a level 0 reservation
   69  */
   70 #define VM_LEVEL_0_NPAGES       (1 << VM_LEVEL_0_ORDER)
   71 
   72 /*
   73  * The number of bits by which a physical address is shifted to obtain the
   74  * reservation number
   75  */
   76 #define VM_LEVEL_0_SHIFT        (VM_LEVEL_0_ORDER + PAGE_SHIFT)
   77 
   78 /*
   79  * The size of a level 0 reservation in bytes
   80  */
   81 #define VM_LEVEL_0_SIZE         (1 << VM_LEVEL_0_SHIFT)
   82 
   83 /*
   84  * Computes the index of the small page underlying the given (object, pindex)
   85  * within the reservation's array of small pages.
   86  */
   87 #define VM_RESERV_INDEX(object, pindex) \
   88     (((object)->pg_color + (pindex)) & (VM_LEVEL_0_NPAGES - 1))
   89 
   90 /*
   91  * The reservation structure
   92  *
   93  * A reservation structure is constructed whenever a large physical page is
   94  * speculatively allocated to an object.  The reservation provides the small
   95  * physical pages for the range [pindex, pindex + VM_LEVEL_0_NPAGES) of offsets
   96  * within that object.  The reservation's "popcnt" tracks the number of these
   97  * small physical pages that are in use at any given time.  When and if the
   98  * reservation is not fully utilized, it appears in the queue of partially-
   99  * populated reservations.  The reservation always appears on the containing
  100  * object's list of reservations.
  101  *
  102  * A partially-populated reservation can be broken and reclaimed at any time.
  103  */
  104 struct vm_reserv {
  105         TAILQ_ENTRY(vm_reserv) partpopq;
  106         LIST_ENTRY(vm_reserv) objq;
  107         vm_object_t     object;                 /* containing object */
  108         vm_pindex_t     pindex;                 /* offset within object */
  109         vm_page_t       pages;                  /* first page of a superpage */
  110         int             popcnt;                 /* # of pages in use */
  111         char            inpartpopq;
  112 };
  113 
  114 /*
  115  * The reservation array
  116  *
  117  * This array is analoguous in function to vm_page_array.  It differs in the
  118  * respect that it may contain a greater number of useful reservation
  119  * structures than there are (physical) superpages.  These "invalid"
  120  * reservation structures exist to trade-off space for time in the
  121  * implementation of vm_reserv_from_page().  Invalid reservation structures are
  122  * distinguishable from "valid" reservation structures by inspecting the
  123  * reservation's "pages" field.  Invalid reservation structures have a NULL
  124  * "pages" field.
  125  *
  126  * vm_reserv_from_page() maps a small (physical) page to an element of this
  127  * array by computing a physical reservation number from the page's physical
  128  * address.  The physical reservation number is used as the array index.
  129  *
  130  * An "active" reservation is a valid reservation structure that has a non-NULL
  131  * "object" field and a non-zero "popcnt" field.  In other words, every active
  132  * reservation belongs to a particular object.  Moreover, every active
  133  * reservation has an entry in the containing object's list of reservations.  
  134  */
  135 static vm_reserv_t vm_reserv_array;
  136 
  137 /*
  138  * The partially-populated reservation queue
  139  *
  140  * This queue enables the fast recovery of an unused cached or free small page
  141  * from a partially-populated reservation.  The reservation at the head of
  142  * this queue is the least-recently-changed, partially-populated reservation.
  143  *
  144  * Access to this queue is synchronized by the free page queue lock.
  145  */
  146 static TAILQ_HEAD(, vm_reserv) vm_rvq_partpop =
  147                             TAILQ_HEAD_INITIALIZER(vm_rvq_partpop);
  148 
  149 static SYSCTL_NODE(_vm, OID_AUTO, reserv, CTLFLAG_RD, 0, "Reservation Info");
  150 
  151 static long vm_reserv_broken;
  152 SYSCTL_LONG(_vm_reserv, OID_AUTO, broken, CTLFLAG_RD,
  153     &vm_reserv_broken, 0, "Cumulative number of broken reservations");
  154 
  155 static long vm_reserv_freed;
  156 SYSCTL_LONG(_vm_reserv, OID_AUTO, freed, CTLFLAG_RD,
  157     &vm_reserv_freed, 0, "Cumulative number of freed reservations");
  158 
  159 static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS);
  160 
  161 SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
  162     sysctl_vm_reserv_partpopq, "A", "Partially-populated reservation queues");
  163 
  164 static long vm_reserv_reclaimed;
  165 SYSCTL_LONG(_vm_reserv, OID_AUTO, reclaimed, CTLFLAG_RD,
  166     &vm_reserv_reclaimed, 0, "Cumulative number of reclaimed reservations");
  167 
  168 static void             vm_reserv_depopulate(vm_reserv_t rv);
  169 static vm_reserv_t      vm_reserv_from_page(vm_page_t m);
  170 static boolean_t        vm_reserv_has_pindex(vm_reserv_t rv,
  171                             vm_pindex_t pindex);
  172 static void             vm_reserv_populate(vm_reserv_t rv);
  173 static void             vm_reserv_reclaim(vm_reserv_t rv);
  174 
  175 /*
  176  * Describes the current state of the partially-populated reservation queue.
  177  */
  178 static int
  179 sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS)
  180 {
  181         struct sbuf sbuf;
  182         vm_reserv_t rv;
  183         char *cbuf;
  184         const int cbufsize = (VM_NRESERVLEVEL + 1) * 81;
  185         int counter, error, level, unused_pages;
  186 
  187         cbuf = malloc(cbufsize, M_TEMP, M_WAITOK | M_ZERO);
  188         sbuf_new(&sbuf, cbuf, cbufsize, SBUF_FIXEDLEN);
  189         sbuf_printf(&sbuf, "\nLEVEL     SIZE  NUMBER\n\n");
  190         for (level = -1; level <= VM_NRESERVLEVEL - 2; level++) {
  191                 counter = 0;
  192                 unused_pages = 0;
  193                 mtx_lock(&vm_page_queue_free_mtx);
  194                 TAILQ_FOREACH(rv, &vm_rvq_partpop/*[level]*/, partpopq) {
  195                         counter++;
  196                         unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt;
  197                 }
  198                 mtx_unlock(&vm_page_queue_free_mtx);
  199                 sbuf_printf(&sbuf, "%5.5d: %6.6dK, %6.6d\n", level,
  200                     unused_pages * (PAGE_SIZE / 1024), counter);
  201         }
  202         sbuf_finish(&sbuf);
  203         error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
  204         sbuf_delete(&sbuf);
  205         free(cbuf, M_TEMP);
  206         return (error);
  207 }
  208 
  209 /*
  210  * Reduces the given reservation's population count.  If the population count
  211  * becomes zero, the reservation is destroyed.  Additionally, moves the
  212  * reservation to the tail of the partially-populated reservations queue if the
  213  * population count is non-zero.
  214  *
  215  * The free page queue lock must be held.
  216  */
  217 static void
  218 vm_reserv_depopulate(vm_reserv_t rv)
  219 {
  220 
  221         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  222         KASSERT(rv->object != NULL,
  223             ("vm_reserv_depopulate: reserv %p is free", rv));
  224         KASSERT(rv->popcnt > 0,
  225             ("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv));
  226         if (rv->inpartpopq) {
  227                 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
  228                 rv->inpartpopq = FALSE;
  229         }
  230         rv->popcnt--;
  231         if (rv->popcnt == 0) {
  232                 LIST_REMOVE(rv, objq);
  233                 rv->object = NULL;
  234                 vm_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER);
  235                 vm_reserv_freed++;
  236         } else {
  237                 rv->inpartpopq = TRUE;
  238                 TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
  239         }
  240 }
  241 
  242 /*
  243  * Returns the reservation to which the given page might belong.
  244  */
  245 static __inline vm_reserv_t
  246 vm_reserv_from_page(vm_page_t m)
  247 {
  248 
  249         return (&vm_reserv_array[VM_PAGE_TO_PHYS(m) >> VM_LEVEL_0_SHIFT]);
  250 }
  251 
  252 /*
  253  * Returns TRUE if the given reservation contains the given page index and
  254  * FALSE otherwise.
  255  */
  256 static __inline boolean_t
  257 vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex)
  258 {
  259 
  260         return (((pindex - rv->pindex) & ~(VM_LEVEL_0_NPAGES - 1)) == 0);
  261 }
  262 
  263 /*
  264  * Increases the given reservation's population count.  Moves the reservation
  265  * to the tail of the partially-populated reservation queue.
  266  *
  267  * The free page queue must be locked.
  268  */
  269 static void
  270 vm_reserv_populate(vm_reserv_t rv)
  271 {
  272 
  273         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  274         KASSERT(rv->object != NULL,
  275             ("vm_reserv_populate: reserv %p is free", rv));
  276         KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES,
  277             ("vm_reserv_populate: reserv %p is already full", rv));
  278         if (rv->inpartpopq) {
  279                 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
  280                 rv->inpartpopq = FALSE;
  281         }
  282         rv->popcnt++;
  283         if (rv->popcnt < VM_LEVEL_0_NPAGES) {
  284                 rv->inpartpopq = TRUE;
  285                 TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
  286         }
  287 }
  288 
  289 /*
  290  * Allocates a page from an existing or newly-created reservation.
  291  *
  292  * The object and free page queue must be locked.
  293  */
  294 vm_page_t
  295 vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex)
  296 {
  297         vm_page_t m, mpred, msucc;
  298         vm_pindex_t first, leftcap, rightcap;
  299         vm_reserv_t rv;
  300 
  301         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  302 
  303         /*
  304          * Is a reservation fundamentally not possible?
  305          */
  306         VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
  307         if (pindex < VM_RESERV_INDEX(object, pindex) ||
  308             pindex >= object->size)
  309                 return (NULL);
  310 
  311         /*
  312          * Look for an existing reservation.
  313          */
  314         msucc = NULL;
  315         mpred = object->root;
  316         while (mpred != NULL) {
  317                 KASSERT(mpred->pindex != pindex,
  318                     ("vm_reserv_alloc_page: pindex already allocated"));
  319                 rv = vm_reserv_from_page(mpred);
  320                 if (rv->object == object && vm_reserv_has_pindex(rv, pindex)) {
  321                         m = &rv->pages[VM_RESERV_INDEX(object, pindex)];
  322                         /* Handle vm_page_rename(m, new_object, ...). */
  323                         if ((m->flags & (PG_CACHED | PG_FREE)) == 0)
  324                                 return (NULL);
  325                         vm_reserv_populate(rv);
  326                         return (m);
  327                 } else if (mpred->pindex < pindex) {
  328                         if (msucc != NULL ||
  329                             (msucc = TAILQ_NEXT(mpred, listq)) == NULL)
  330                                 break;
  331                         KASSERT(msucc->pindex != pindex,
  332                             ("vm_reserv_alloc_page: pindex already allocated"));
  333                         rv = vm_reserv_from_page(msucc);
  334                         if (rv->object == object &&
  335                             vm_reserv_has_pindex(rv, pindex)) {
  336                                 m = &rv->pages[VM_RESERV_INDEX(object, pindex)];
  337                                 /* Handle vm_page_rename(m, new_object, ...). */
  338                                 if ((m->flags & (PG_CACHED | PG_FREE)) == 0)
  339                                         return (NULL);
  340                                 vm_reserv_populate(rv);
  341                                 return (m);
  342                         } else if (pindex < msucc->pindex)
  343                                 break;
  344                 } else if (msucc == NULL) {
  345                         msucc = mpred;
  346                         mpred = TAILQ_PREV(msucc, pglist, listq);
  347                         continue;
  348                 }
  349                 msucc = NULL;
  350                 mpred = object->root = vm_page_splay(pindex, object->root);
  351         }
  352 
  353         /*
  354          * Determine the first index to the left that can be used.
  355          */
  356         if (mpred == NULL)
  357                 leftcap = 0;
  358         else if ((rv = vm_reserv_from_page(mpred))->object != object)
  359                 leftcap = mpred->pindex + 1;
  360         else
  361                 leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
  362 
  363         /*
  364          * Determine the first index to the right that cannot be used.
  365          */
  366         if (msucc == NULL)
  367                 rightcap = pindex + VM_LEVEL_0_NPAGES;
  368         else if ((rv = vm_reserv_from_page(msucc))->object != object)
  369                 rightcap = msucc->pindex;
  370         else
  371                 rightcap = rv->pindex;
  372 
  373         /*
  374          * Determine if a reservation fits between the first index to
  375          * the left that can be used and the first index to the right
  376          * that cannot be used. 
  377          */
  378         first = pindex - VM_RESERV_INDEX(object, pindex);
  379         if (first < leftcap || first + VM_LEVEL_0_NPAGES > rightcap)
  380                 return (NULL);
  381 
  382         /*
  383          * Would a new reservation extend past the end of the given object? 
  384          */
  385         if (object->size < first + VM_LEVEL_0_NPAGES) {
  386                 /*
  387                  * Don't allocate a new reservation if the object is a vnode or
  388                  * backed by another object that is a vnode. 
  389                  */
  390                 if (object->type == OBJT_VNODE ||
  391                     (object->backing_object != NULL &&
  392                     object->backing_object->type == OBJT_VNODE))
  393                         return (NULL);
  394                 /* Speculate that the object may grow. */
  395         }
  396 
  397         /*
  398          * Allocate a new reservation.
  399          */
  400         m = vm_phys_alloc_pages(VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER);
  401         if (m != NULL) {
  402                 rv = vm_reserv_from_page(m);
  403                 KASSERT(rv->pages == m,
  404                     ("vm_reserv_alloc_page: reserv %p's pages is corrupted",
  405                     rv));
  406                 KASSERT(rv->object == NULL,
  407                     ("vm_reserv_alloc_page: reserv %p isn't free", rv));
  408                 LIST_INSERT_HEAD(&object->rvq, rv, objq);
  409                 rv->object = object;
  410                 rv->pindex = first;
  411                 KASSERT(rv->popcnt == 0,
  412                     ("vm_reserv_alloc_page: reserv %p's popcnt is corrupted",
  413                     rv));
  414                 KASSERT(!rv->inpartpopq,
  415                     ("vm_reserv_alloc_page: reserv %p's inpartpopq is TRUE",
  416                     rv));
  417                 vm_reserv_populate(rv);
  418                 m = &rv->pages[VM_RESERV_INDEX(object, pindex)];
  419         }
  420         return (m);
  421 }
  422 
  423 /*
  424  * Breaks all reservations belonging to the given object.
  425  */
  426 void
  427 vm_reserv_break_all(vm_object_t object)
  428 {
  429         vm_reserv_t rv;
  430         int i;
  431 
  432         mtx_lock(&vm_page_queue_free_mtx);
  433         while ((rv = LIST_FIRST(&object->rvq)) != NULL) {
  434                 KASSERT(rv->object == object,
  435                     ("vm_reserv_break_all: reserv %p is corrupted", rv));
  436                 if (rv->inpartpopq) {
  437                         TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
  438                         rv->inpartpopq = FALSE;
  439                 }
  440                 LIST_REMOVE(rv, objq);
  441                 rv->object = NULL;
  442                 for (i = 0; i < VM_LEVEL_0_NPAGES; i++) {
  443                         if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
  444                                 vm_phys_free_pages(&rv->pages[i], 0);
  445                         else
  446                                 rv->popcnt--;
  447                 }
  448                 KASSERT(rv->popcnt == 0,
  449                     ("vm_reserv_break_all: reserv %p's popcnt is corrupted",
  450                     rv));
  451                 vm_reserv_broken++;
  452         }
  453         mtx_unlock(&vm_page_queue_free_mtx);
  454 }
  455 
  456 /*
  457  * Frees the given page if it belongs to a reservation.  Returns TRUE if the
  458  * page is freed and FALSE otherwise.
  459  *
  460  * The free page queue lock must be held.
  461  */
  462 boolean_t
  463 vm_reserv_free_page(vm_page_t m)
  464 {
  465         vm_reserv_t rv;
  466 
  467         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  468         rv = vm_reserv_from_page(m);
  469         if (rv->object != NULL) {
  470                 vm_reserv_depopulate(rv);
  471                 return (TRUE);
  472         }
  473         return (FALSE);
  474 }
  475 
  476 /*
  477  * Initializes the reservation management system.  Specifically, initializes
  478  * the reservation array.
  479  *
  480  * Requires that vm_page_array and first_page are initialized!
  481  */
  482 void
  483 vm_reserv_init(void)
  484 {
  485         vm_paddr_t paddr;
  486         int i;
  487 
  488         /*
  489          * Initialize the reservation array.  Specifically, initialize the
  490          * "pages" field for every element that has an underlying superpage.
  491          */
  492         for (i = 0; phys_avail[i + 1] != 0; i += 2) {
  493                 paddr = roundup2(phys_avail[i], VM_LEVEL_0_SIZE);
  494                 while (paddr + VM_LEVEL_0_SIZE <= phys_avail[i + 1]) {
  495                         vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT].pages =
  496                             PHYS_TO_VM_PAGE(paddr);
  497                         paddr += VM_LEVEL_0_SIZE;
  498                 }
  499         }
  500 }
  501 
  502 /*
  503  * Returns a reservation level if the given page belongs to a fully-populated
  504  * reservation and -1 otherwise.
  505  */
  506 int
  507 vm_reserv_level_iffullpop(vm_page_t m)
  508 {
  509         vm_reserv_t rv;
  510 
  511         rv = vm_reserv_from_page(m);
  512         return (rv->popcnt == VM_LEVEL_0_NPAGES ? 0 : -1);
  513 }
  514 
  515 /*
  516  * Prepare for the reactivation of a cached page.
  517  *
  518  * First, suppose that the given page "m" was allocated individually, i.e., not
  519  * as part of a reservation, and cached.  Then, suppose a reservation
  520  * containing "m" is allocated by the same object.  Although "m" and the
  521  * reservation belong to the same object, "m"'s pindex may not match the
  522  * reservation's.
  523  *
  524  * The free page queue must be locked.
  525  */
  526 boolean_t
  527 vm_reserv_reactivate_page(vm_page_t m)
  528 {
  529         vm_reserv_t rv;
  530         int i, m_index;
  531 
  532         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  533         rv = vm_reserv_from_page(m);
  534         if (rv->object == NULL)
  535                 return (FALSE);
  536         KASSERT((m->flags & PG_CACHED) != 0,
  537             ("vm_reserv_uncache_page: page %p is not cached", m));
  538         if (m->object == rv->object &&
  539             m->pindex - rv->pindex == VM_RESERV_INDEX(m->object, m->pindex))
  540                 vm_reserv_populate(rv);
  541         else {
  542                 KASSERT(rv->inpartpopq,
  543                     ("vm_reserv_uncache_page: reserv %p's inpartpopq is FALSE",
  544                     rv));
  545                 TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
  546                 rv->inpartpopq = FALSE;
  547                 LIST_REMOVE(rv, objq);
  548                 rv->object = NULL;
  549                 /* Don't vm_phys_free_pages(m, 0). */
  550                 m_index = m - rv->pages;
  551                 for (i = 0; i < m_index; i++) {
  552                         if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
  553                                 vm_phys_free_pages(&rv->pages[i], 0);
  554                         else
  555                                 rv->popcnt--;
  556                 }
  557                 for (i++; i < VM_LEVEL_0_NPAGES; i++) {
  558                         if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
  559                                 vm_phys_free_pages(&rv->pages[i], 0);
  560                         else
  561                                 rv->popcnt--;
  562                 }
  563                 KASSERT(rv->popcnt == 0,
  564                     ("vm_reserv_uncache_page: reserv %p's popcnt is corrupted",
  565                     rv));
  566                 vm_reserv_broken++;
  567         }
  568         return (TRUE);
  569 }
  570 
  571 /*
  572  * Breaks the given partially-populated reservation, releasing its cached and
  573  * free pages to the physical memory allocator.
  574  *
  575  * The free page queue lock must be held.
  576  */
  577 static void
  578 vm_reserv_reclaim(vm_reserv_t rv)
  579 {
  580         int i;
  581 
  582         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  583         KASSERT(rv->inpartpopq,
  584             ("vm_reserv_reclaim: reserv %p's inpartpopq is corrupted", rv));
  585         TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
  586         rv->inpartpopq = FALSE;
  587         KASSERT(rv->object != NULL,
  588             ("vm_reserv_reclaim: reserv %p is free", rv));
  589         LIST_REMOVE(rv, objq);
  590         rv->object = NULL;
  591         for (i = 0; i < VM_LEVEL_0_NPAGES; i++) {
  592                 if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0)
  593                         vm_phys_free_pages(&rv->pages[i], 0);
  594                 else
  595                         rv->popcnt--;
  596         }
  597         KASSERT(rv->popcnt == 0,
  598             ("vm_reserv_reclaim: reserv %p's popcnt is corrupted", rv));
  599         vm_reserv_reclaimed++;
  600 }
  601 
  602 /*
  603  * Breaks the reservation at the head of the partially-populated reservation
  604  * queue, releasing its cached and free pages to the physical memory
  605  * allocator.  Returns TRUE if a reservation is broken and FALSE otherwise.
  606  *
  607  * The free page queue lock must be held.
  608  */
  609 boolean_t
  610 vm_reserv_reclaim_inactive(void)
  611 {
  612         vm_reserv_t rv;
  613 
  614         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  615         if ((rv = TAILQ_FIRST(&vm_rvq_partpop)) != NULL) {
  616                 vm_reserv_reclaim(rv);
  617                 return (TRUE);
  618         }
  619         return (FALSE);
  620 }
  621 
  622 /*
  623  * Searches the partially-populated reservation queue for the least recently
  624  * active reservation with unused pages, i.e., cached or free, that satisfy the
  625  * given request for contiguous physical memory.  If a satisfactory reservation
  626  * is found, it is broken.  Returns TRUE if a reservation is broken and FALSE
  627  * otherwise.
  628  *
  629  * The free page queue lock must be held.
  630  */
  631 boolean_t
  632 vm_reserv_reclaim_contig(vm_paddr_t size, vm_paddr_t low, vm_paddr_t high,
  633     unsigned long alignment, unsigned long boundary)
  634 {
  635         vm_paddr_t pa, pa_length;
  636         vm_reserv_t rv;
  637         int i;
  638 
  639         mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
  640         if (size > VM_LEVEL_0_SIZE - PAGE_SIZE)
  641                 return (FALSE);
  642         TAILQ_FOREACH(rv, &vm_rvq_partpop, partpopq) {
  643                 pa = VM_PAGE_TO_PHYS(&rv->pages[VM_LEVEL_0_NPAGES - 1]);
  644                 if (pa + PAGE_SIZE - size < low) {
  645                         /* this entire reservation is too low; go to next */
  646                         continue;
  647                 }
  648                 pa_length = 0;
  649                 for (i = 0; i < VM_LEVEL_0_NPAGES; i++)
  650                         if ((rv->pages[i].flags & (PG_CACHED | PG_FREE)) != 0) {
  651                                 pa_length += PAGE_SIZE;
  652                                 if (pa_length == PAGE_SIZE) {
  653                                         pa = VM_PAGE_TO_PHYS(&rv->pages[i]);
  654                                         if (pa + size > high) {
  655                                                 /* skip to next reservation */
  656                                                 break;
  657                                         } else if (pa < low ||
  658                                             (pa & (alignment - 1)) != 0 ||
  659                                             ((pa ^ (pa + size - 1)) &
  660                                             ~(boundary - 1)) != 0)
  661                                                 pa_length = 0;
  662                                 } else if (pa_length >= size) {
  663                                         vm_reserv_reclaim(rv);
  664                                         return (TRUE);
  665                                 }
  666                         } else
  667                                 pa_length = 0;
  668         }
  669         return (FALSE);
  670 }
  671 
  672 /*
  673  * Transfers the reservation underlying the given page to a new object.
  674  *
  675  * The object must be locked.
  676  */
  677 void
  678 vm_reserv_rename(vm_page_t m, vm_object_t new_object, vm_object_t old_object,
  679     vm_pindex_t old_object_offset)
  680 {
  681         vm_reserv_t rv;
  682 
  683         VM_OBJECT_LOCK_ASSERT(new_object, MA_OWNED);
  684         rv = vm_reserv_from_page(m);
  685         if (rv->object == old_object) {
  686                 mtx_lock(&vm_page_queue_free_mtx);
  687                 if (rv->object == old_object) {
  688                         LIST_REMOVE(rv, objq);
  689                         LIST_INSERT_HEAD(&new_object->rvq, rv, objq);
  690                         rv->object = new_object;
  691                         rv->pindex -= old_object_offset;
  692                 }
  693                 mtx_unlock(&vm_page_queue_free_mtx);
  694         }
  695 }
  696 
  697 /*
  698  * Allocates the virtual and physical memory required by the reservation
  699  * management system's data structures, in particular, the reservation array.
  700  */
  701 vm_paddr_t
  702 vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t high_water)
  703 {
  704         vm_paddr_t new_end;
  705         size_t size;
  706 
  707         /*
  708          * Calculate the size (in bytes) of the reservation array.  Round up
  709          * from "high_water" because every small page is mapped to an element
  710          * in the reservation array based on its physical address.  Thus, the
  711          * number of elements in the reservation array can be greater than the
  712          * number of superpages. 
  713          */
  714         size = howmany(high_water, VM_LEVEL_0_SIZE) * sizeof(struct vm_reserv);
  715 
  716         /*
  717          * Allocate and map the physical memory for the reservation array.  The
  718          * next available virtual address is returned by reference.
  719          */
  720         new_end = end - round_page(size);
  721         vm_reserv_array = (void *)(uintptr_t)pmap_map(vaddr, new_end, end,
  722             VM_PROT_READ | VM_PROT_WRITE);
  723         bzero(vm_reserv_array, size);
  724 
  725         /*
  726          * Return the next available physical address.
  727          */
  728         return (new_end);
  729 }
  730 
  731 #endif  /* VM_NRESERVLEVEL > 0 */

Cache object: 21c6753aab5892239e5b1de2c5695ed3


[ 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.