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/uvm/uvm_map.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 /*      $NetBSD: uvm_map.c,v 1.164.2.3 2004/05/09 09:01:32 jdc Exp $    */
    2 
    3 /*
    4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
    5  * Copyright (c) 1991, 1993, The Regents of the University of California.
    6  *
    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 Charles D. Cranor,
   23  *      Washington University, the University of California, Berkeley and
   24  *      its contributors.
   25  * 4. Neither the name of the University nor the names of its contributors
   26  *    may be used to endorse or promote products derived from this software
   27  *    without specific prior written permission.
   28  *
   29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   39  * SUCH DAMAGE.
   40  *
   41  *      @(#)vm_map.c    8.3 (Berkeley) 1/12/94
   42  * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
   43  *
   44  *
   45  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
   46  * All rights reserved.
   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 
   69 /*
   70  * uvm_map.c: uvm map operations
   71  */
   72 
   73 #include <sys/cdefs.h>
   74 __KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.164.2.3 2004/05/09 09:01:32 jdc Exp $");
   75 
   76 #include "opt_ddb.h"
   77 #include "opt_uvmhist.h"
   78 #include "opt_uvm.h"
   79 #include "opt_sysv.h"
   80 
   81 #include <sys/param.h>
   82 #include <sys/systm.h>
   83 #include <sys/mman.h>
   84 #include <sys/proc.h>
   85 #include <sys/malloc.h>
   86 #include <sys/pool.h>
   87 #include <sys/kernel.h>
   88 #include <sys/mount.h>
   89 #include <sys/vnode.h>
   90 
   91 #ifdef SYSVSHM
   92 #include <sys/shm.h>
   93 #endif
   94 
   95 #define UVM_MAP
   96 #include <uvm/uvm.h>
   97 #undef RB_AUGMENT
   98 #define RB_AUGMENT(x)   uvm_rb_augment(x)
   99 
  100 #ifdef DDB
  101 #include <uvm/uvm_ddb.h>
  102 #endif
  103 
  104 extern struct vm_map *pager_map;
  105 
  106 #ifndef UVMMAP_NOCOUNTERS
  107 #include <sys/device.h>
  108 struct evcnt map_ubackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  109     "uvmmap", "ubackmerge");
  110 struct evcnt map_uforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  111     "uvmmap", "uforwmerge");
  112 struct evcnt map_ubimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  113     "uvmmap", "ubimerge");
  114 struct evcnt map_unomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  115     "uvmmap", "unomerge");
  116 struct evcnt map_kbackmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  117     "uvmmap", "kbackmerge");
  118 struct evcnt map_kforwmerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  119     "uvmmap", "kforwmerge");
  120 struct evcnt map_kbimerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  121     "uvmmap", "kbimerge");
  122 struct evcnt map_knomerge = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  123     "uvmmap", "knomerge");
  124 struct evcnt uvm_map_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  125     "uvmmap", "map_call");
  126 struct evcnt uvm_mlk_call = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  127     "uvmmap", "mlk_call");
  128 struct evcnt uvm_mlk_hint = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
  129     "uvmmap", "mlk_hint");
  130 
  131 EVCNT_ATTACH_STATIC(map_ubackmerge);
  132 EVCNT_ATTACH_STATIC(map_uforwmerge);
  133 EVCNT_ATTACH_STATIC(map_ubimerge);
  134 EVCNT_ATTACH_STATIC(map_unomerge);
  135 EVCNT_ATTACH_STATIC(map_kbackmerge);
  136 EVCNT_ATTACH_STATIC(map_kforwmerge);
  137 EVCNT_ATTACH_STATIC(map_kbimerge);
  138 EVCNT_ATTACH_STATIC(map_knomerge);
  139 EVCNT_ATTACH_STATIC(uvm_map_call);
  140 EVCNT_ATTACH_STATIC(uvm_mlk_call);
  141 EVCNT_ATTACH_STATIC(uvm_mlk_hint);
  142 
  143 #define UVMCNT_INCR(ev)         ev.ev_count++
  144 #define UVMCNT_DECR(ev)         ev.ev_count--
  145 #else
  146 #define UVMCNT_INCR(ev)
  147 #define UVMCNT_DECR(ev)
  148 #endif
  149 
  150 const char vmmapbsy[] = "vmmapbsy";
  151 
  152 /*
  153  * pool for vmspace structures.
  154  */
  155 
  156 struct pool uvm_vmspace_pool;
  157 
  158 /*
  159  * pool for dynamically-allocated map entries.
  160  */
  161 
  162 struct pool uvm_map_entry_pool;
  163 struct pool uvm_map_entry_kmem_pool;
  164 
  165 MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures");
  166 MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap");
  167 
  168 #ifdef PMAP_GROWKERNEL
  169 /*
  170  * This global represents the end of the kernel virtual address
  171  * space.  If we want to exceed this, we must grow the kernel
  172  * virtual address space dynamically.
  173  *
  174  * Note, this variable is locked by kernel_map's lock.
  175  */
  176 vaddr_t uvm_maxkaddr;
  177 #endif
  178 
  179 /*
  180  * macros
  181  */
  182 
  183 /*
  184  * uvm_map_entry_link: insert entry into a map
  185  *
  186  * => map must be locked
  187  */
  188 #define uvm_map_entry_link(map, after_where, entry) do { \
  189         KASSERT(entry->start < entry->end); \
  190         (map)->nentries++; \
  191         (entry)->prev = (after_where); \
  192         (entry)->next = (after_where)->next; \
  193         (entry)->prev->next = (entry); \
  194         (entry)->next->prev = (entry); \
  195         uvm_rb_insert((map), (entry)); \
  196 } while (/*CONSTCOND*/ 0)
  197 
  198 /*
  199  * uvm_map_entry_unlink: remove entry from a map
  200  *
  201  * => map must be locked
  202  */
  203 #define uvm_map_entry_unlink(map, entry) do { \
  204         (map)->nentries--; \
  205         (entry)->next->prev = (entry)->prev; \
  206         (entry)->prev->next = (entry)->next; \
  207         uvm_rb_remove((map), (entry)); \
  208 } while (/*CONSTCOND*/ 0)
  209 
  210 /*
  211  * SAVE_HINT: saves the specified entry as the hint for future lookups.
  212  *
  213  * => map need not be locked (protected by hint_lock).
  214  */
  215 #define SAVE_HINT(map,check,value) do { \
  216         simple_lock(&(map)->hint_lock); \
  217         if ((map)->hint == (check)) \
  218                 (map)->hint = (value); \
  219         simple_unlock(&(map)->hint_lock); \
  220 } while (/*CONSTCOND*/ 0)
  221 
  222 /*
  223  * VM_MAP_RANGE_CHECK: check and correct range
  224  *
  225  * => map must at least be read locked
  226  */
  227 
  228 #define VM_MAP_RANGE_CHECK(map, start, end) do { \
  229         if (start < vm_map_min(map))            \
  230                 start = vm_map_min(map);        \
  231         if (end > vm_map_max(map))              \
  232                 end = vm_map_max(map);          \
  233         if (start > end)                        \
  234                 start = end;                    \
  235 } while (/*CONSTCOND*/ 0)
  236 
  237 /*
  238  * local prototypes
  239  */
  240 
  241 static struct vm_map_entry *
  242                 uvm_mapent_alloc(struct vm_map *, int);
  243 static void     uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *);
  244 static void     uvm_mapent_free(struct vm_map_entry *);
  245 static void     uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *);
  246 static void     uvm_map_reference_amap(struct vm_map_entry *, int);
  247 static int      uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int,
  248                     struct vm_map_entry *);
  249 static void     uvm_map_unreference_amap(struct vm_map_entry *, int);
  250 
  251 int _uvm_tree_sanity(struct vm_map *, const char *);
  252 static vsize_t uvm_rb_subtree_space(const struct vm_map_entry *);
  253 
  254 static __inline int
  255 uvm_compare(const struct vm_map_entry *a, const struct vm_map_entry *b)
  256 {
  257 
  258         if (a->start < b->start)
  259                 return (-1);
  260         else if (a->start > b->start)
  261                 return (1);
  262 
  263         return (0);
  264 }
  265 
  266 static __inline void
  267 uvm_rb_augment(struct vm_map_entry *entry)
  268 {
  269 
  270         entry->space = uvm_rb_subtree_space(entry);
  271 }
  272 
  273 RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
  274 
  275 RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
  276 
  277 static __inline vsize_t
  278 uvm_rb_space(const struct vm_map *map, const struct vm_map_entry *entry)
  279 {
  280         /* XXX map is not used */
  281 
  282         KASSERT(entry->next != NULL);
  283         return entry->next->start - entry->end;
  284 }
  285 
  286 static vsize_t
  287 uvm_rb_subtree_space(const struct vm_map_entry *entry)
  288 {
  289         vaddr_t space, tmp;
  290 
  291         space = entry->ownspace;
  292         if (RB_LEFT(entry, rb_entry)) {
  293                 tmp = RB_LEFT(entry, rb_entry)->space;
  294                 if (tmp > space)
  295                         space = tmp;
  296         }
  297 
  298         if (RB_RIGHT(entry, rb_entry)) {
  299                 tmp = RB_RIGHT(entry, rb_entry)->space;
  300                 if (tmp > space)
  301                         space = tmp;
  302         }
  303 
  304         return (space);
  305 }
  306 
  307 static __inline void
  308 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry)
  309 {
  310         /* We need to traverse to the very top */
  311         do {
  312                 entry->ownspace = uvm_rb_space(map, entry);
  313                 entry->space = uvm_rb_subtree_space(entry);
  314         } while ((entry = RB_PARENT(entry, rb_entry)) != NULL);
  315 }
  316 
  317 static __inline void
  318 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry)
  319 {
  320         vaddr_t space = uvm_rb_space(map, entry);
  321         struct vm_map_entry *tmp;
  322 
  323         entry->ownspace = entry->space = space;
  324         tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry);
  325 #ifdef DIAGNOSTIC
  326         if (tmp != NULL)
  327                 panic("uvm_rb_insert: duplicate entry?");
  328 #endif
  329         uvm_rb_fixup(map, entry);
  330         if (entry->prev != &map->header)
  331                 uvm_rb_fixup(map, entry->prev);
  332 }
  333 
  334 static __inline void
  335 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry)
  336 {
  337         struct vm_map_entry *parent;
  338 
  339         parent = RB_PARENT(entry, rb_entry);
  340         RB_REMOVE(uvm_tree, &(map)->rbhead, entry);
  341         if (entry->prev != &map->header)
  342                 uvm_rb_fixup(map, entry->prev);
  343         if (parent)
  344                 uvm_rb_fixup(map, parent);
  345 }
  346 
  347 #ifdef DEBUG
  348 int uvm_debug_check_rbtree = 0;
  349 #define uvm_tree_sanity(x,y)            \
  350         if (uvm_debug_check_rbtree)     \
  351                 _uvm_tree_sanity(x,y)
  352 #else
  353 #define uvm_tree_sanity(x,y)
  354 #endif
  355 
  356 int
  357 _uvm_tree_sanity(struct vm_map *map, const char *name)
  358 {
  359         struct vm_map_entry *tmp, *trtmp;
  360         int n = 0, i = 1;
  361 
  362         RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
  363                 if (tmp->ownspace != uvm_rb_space(map, tmp)) {
  364                         printf("%s: %d/%d ownspace %lx != %lx %s\n",
  365                             name, n + 1, map->nentries,
  366                             (ulong)tmp->ownspace, (ulong)uvm_rb_space(map, tmp),
  367                             tmp->next == &map->header ? "(last)" : "");
  368                         goto error;
  369                 }
  370         }
  371         trtmp = NULL;
  372         RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
  373                 if (tmp->space != uvm_rb_subtree_space(tmp)) {
  374                         printf("%s: space %lx != %lx\n",
  375                             name, (ulong)tmp->space,
  376                             (ulong)uvm_rb_subtree_space(tmp));
  377                         goto error;
  378                 }
  379                 if (trtmp != NULL && trtmp->start >= tmp->start) {
  380                         printf("%s: corrupt: 0x%lx >= 0x%lx\n",
  381                             name, trtmp->start, tmp->start);
  382                         goto error;
  383                 }
  384                 n++;
  385 
  386                 trtmp = tmp;
  387         }
  388 
  389         if (n != map->nentries) {
  390                 printf("%s: nentries: %d vs %d\n",
  391                     name, n, map->nentries);
  392                 goto error;
  393         }
  394 
  395         for (tmp = map->header.next; tmp && tmp != &map->header;
  396             tmp = tmp->next, i++) {
  397                 trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp);
  398                 if (trtmp != tmp) {
  399                         printf("%s: lookup: %d: %p - %p: %p\n",
  400                             name, i, tmp, trtmp,
  401                             RB_PARENT(tmp, rb_entry));
  402                         goto error;
  403                 }
  404         }
  405 
  406         return (0);
  407  error:
  408 #ifdef  DDB
  409         /* handy breakpoint location for error case */
  410         __asm(".globl treesanity_label\ntreesanity_label:");
  411 #endif
  412         return (-1);
  413 }
  414 
  415 /*
  416  * local inlines
  417  */
  418 
  419 /*
  420  * uvm_mapent_alloc: allocate a map entry
  421  */
  422 
  423 static __inline struct vm_map_entry *
  424 uvm_mapent_alloc(struct vm_map *map, int flags)
  425 {
  426         struct vm_map_entry *me;
  427         int s;
  428         int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
  429         UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
  430 
  431         if (map->flags & VM_MAP_INTRSAFE || cold) {
  432                 s = splvm();
  433                 simple_lock(&uvm.kentry_lock);
  434                 me = uvm.kentry_free;
  435                 if (me)
  436                         uvm.kentry_free = me->next;
  437                 simple_unlock(&uvm.kentry_lock);
  438                 splx(s);
  439                 if (__predict_false(me == NULL)) {
  440                         panic("uvm_mapent_alloc: out of static map entries, "
  441                             "check MAX_KMAPENT (currently %d)",
  442                             MAX_KMAPENT);
  443                 }
  444                 me->flags = UVM_MAP_STATIC;
  445         } else if (map == kernel_map) {
  446                 me = pool_get(&uvm_map_entry_kmem_pool, pflags);
  447                 if (__predict_false(me == NULL))
  448                         return NULL;
  449                 me->flags = UVM_MAP_KMEM;
  450         } else {
  451                 me = pool_get(&uvm_map_entry_pool, pflags);
  452                 if (__predict_false(me == NULL))
  453                         return NULL;
  454                 me->flags = 0;
  455         }
  456 
  457         UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me,
  458             ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0);
  459         return (me);
  460 }
  461 
  462 /*
  463  * uvm_mapent_free: free map entry
  464  */
  465 
  466 static __inline void
  467 uvm_mapent_free(struct vm_map_entry *me)
  468 {
  469         int s;
  470         UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
  471 
  472         UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
  473                 me, me->flags, 0, 0);
  474         if (me->flags & UVM_MAP_STATIC) {
  475                 s = splvm();
  476                 simple_lock(&uvm.kentry_lock);
  477                 me->next = uvm.kentry_free;
  478                 uvm.kentry_free = me;
  479                 simple_unlock(&uvm.kentry_lock);
  480                 splx(s);
  481         } else if (me->flags & UVM_MAP_KMEM) {
  482                 pool_put(&uvm_map_entry_kmem_pool, me);
  483         } else {
  484                 pool_put(&uvm_map_entry_pool, me);
  485         }
  486 }
  487 
  488 /*
  489  * uvm_mapent_copy: copy a map entry, preserving flags
  490  */
  491 
  492 static __inline void
  493 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst)
  494 {
  495 
  496         memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
  497             ((char *)src));
  498 }
  499 
  500 /*
  501  * uvm_map_entry_unwire: unwire a map entry
  502  *
  503  * => map should be locked by caller
  504  */
  505 
  506 static __inline void
  507 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry)
  508 {
  509 
  510         entry->wired_count = 0;
  511         uvm_fault_unwire_locked(map, entry->start, entry->end);
  512 }
  513 
  514 
  515 /*
  516  * wrapper for calling amap_ref()
  517  */
  518 static __inline void
  519 uvm_map_reference_amap(struct vm_map_entry *entry, int flags)
  520 {
  521 
  522         amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
  523             (entry->end - entry->start) >> PAGE_SHIFT, flags);
  524 }
  525 
  526 
  527 /*
  528  * wrapper for calling amap_unref()
  529  */
  530 static __inline void
  531 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags)
  532 {
  533 
  534         amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
  535             (entry->end - entry->start) >> PAGE_SHIFT, flags);
  536 }
  537 
  538 
  539 /*
  540  * uvm_map_init: init mapping system at boot time.   note that we allocate
  541  * and init the static pool of struct vm_map_entry *'s for the kernel here.
  542  */
  543 
  544 void
  545 uvm_map_init(void)
  546 {
  547         static struct vm_map_entry kernel_map_entry[MAX_KMAPENT];
  548 #if defined(UVMHIST)
  549         static struct uvm_history_ent maphistbuf[100];
  550         static struct uvm_history_ent pdhistbuf[100];
  551 #endif
  552         int lcv;
  553 
  554         /*
  555          * first, init logging system.
  556          */
  557 
  558         UVMHIST_FUNC("uvm_map_init");
  559         UVMHIST_INIT_STATIC(maphist, maphistbuf);
  560         UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
  561         UVMHIST_CALLED(maphist);
  562         UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
  563 
  564         /*
  565          * now set up static pool of kernel map entrys ...
  566          */
  567 
  568         simple_lock_init(&uvm.kentry_lock);
  569         uvm.kentry_free = NULL;
  570         for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) {
  571                 kernel_map_entry[lcv].next = uvm.kentry_free;
  572                 uvm.kentry_free = &kernel_map_entry[lcv];
  573         }
  574 
  575         /*
  576          * initialize the map-related pools.
  577          */
  578         pool_init(&uvm_vmspace_pool, sizeof(struct vmspace),
  579             0, 0, 0, "vmsppl", &pool_allocator_nointr);
  580         pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry),
  581             0, 0, 0, "vmmpepl", &pool_allocator_nointr);
  582         pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry),
  583             0, 0, 0, "vmmpekpl", NULL);
  584 }
  585 
  586 /*
  587  * clippers
  588  */
  589 
  590 /*
  591  * uvm_map_clip_start: ensure that the entry begins at or after
  592  *      the starting address, if it doesn't we split the entry.
  593  *
  594  * => caller should use UVM_MAP_CLIP_START macro rather than calling
  595  *    this directly
  596  * => map must be locked by caller
  597  */
  598 
  599 void
  600 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry,
  601     vaddr_t start)
  602 {
  603         struct vm_map_entry *new_entry;
  604         vaddr_t new_adj;
  605 
  606         /* uvm_map_simplify_entry(map, entry); */ /* XXX */
  607 
  608         uvm_tree_sanity(map, "clip_start entry");
  609 
  610         /*
  611          * Split off the front portion.  note that we must insert the new
  612          * entry BEFORE this one, so that this entry has the specified
  613          * starting address.
  614          */
  615 
  616         new_entry = uvm_mapent_alloc(map, 0);
  617         uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
  618 
  619         new_entry->end = start;
  620         new_adj = start - new_entry->start;
  621         if (entry->object.uvm_obj)
  622                 entry->offset += new_adj;       /* shift start over */
  623 
  624         /* Does not change order for the RB tree */
  625         entry->start = start;
  626 
  627         if (new_entry->aref.ar_amap) {
  628                 amap_splitref(&new_entry->aref, &entry->aref, new_adj);
  629         }
  630 
  631         uvm_map_entry_link(map, entry->prev, new_entry);
  632 
  633         if (UVM_ET_ISSUBMAP(entry)) {
  634                 /* ... unlikely to happen, but play it safe */
  635                  uvm_map_reference(new_entry->object.sub_map);
  636         } else {
  637                 if (UVM_ET_ISOBJ(entry) &&
  638                     entry->object.uvm_obj->pgops &&
  639                     entry->object.uvm_obj->pgops->pgo_reference)
  640                         entry->object.uvm_obj->pgops->pgo_reference(
  641                             entry->object.uvm_obj);
  642         }
  643 
  644         uvm_tree_sanity(map, "clip_start leave");
  645 }
  646 
  647 /*
  648  * uvm_map_clip_end: ensure that the entry ends at or before
  649  *      the ending address, if it does't we split the reference
  650  *
  651  * => caller should use UVM_MAP_CLIP_END macro rather than calling
  652  *    this directly
  653  * => map must be locked by caller
  654  */
  655 
  656 void
  657 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end)
  658 {
  659         struct vm_map_entry *   new_entry;
  660         vaddr_t new_adj; /* #bytes we move start forward */
  661 
  662         uvm_tree_sanity(map, "clip_end entry");
  663         /*
  664          *      Create a new entry and insert it
  665          *      AFTER the specified entry
  666          */
  667 
  668         new_entry = uvm_mapent_alloc(map, 0);
  669         uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
  670 
  671         new_entry->start = entry->end = end;
  672         new_adj = end - entry->start;
  673         if (new_entry->object.uvm_obj)
  674                 new_entry->offset += new_adj;
  675 
  676         if (entry->aref.ar_amap)
  677                 amap_splitref(&entry->aref, &new_entry->aref, new_adj);
  678 
  679         uvm_rb_fixup(map, entry);
  680 
  681         uvm_map_entry_link(map, entry, new_entry);
  682 
  683         if (UVM_ET_ISSUBMAP(entry)) {
  684                 /* ... unlikely to happen, but play it safe */
  685                 uvm_map_reference(new_entry->object.sub_map);
  686         } else {
  687                 if (UVM_ET_ISOBJ(entry) &&
  688                     entry->object.uvm_obj->pgops &&
  689                     entry->object.uvm_obj->pgops->pgo_reference)
  690                         entry->object.uvm_obj->pgops->pgo_reference(
  691                             entry->object.uvm_obj);
  692         }
  693 
  694         uvm_tree_sanity(map, "clip_end leave");
  695 }
  696 
  697 
  698 /*
  699  *   M A P   -   m a i n   e n t r y   p o i n t
  700  */
  701 /*
  702  * uvm_map: establish a valid mapping in a map
  703  *
  704  * => assume startp is page aligned.
  705  * => assume size is a multiple of PAGE_SIZE.
  706  * => assume sys_mmap provides enough of a "hint" to have us skip
  707  *      over text/data/bss area.
  708  * => map must be unlocked (we will lock it)
  709  * => <uobj,uoffset> value meanings (4 cases):
  710  *       [1] <NULL,uoffset>             == uoffset is a hint for PMAP_PREFER
  711  *       [2] <NULL,UVM_UNKNOWN_OFFSET>  == don't PMAP_PREFER
  712  *       [3] <uobj,uoffset>             == normal mapping
  713  *       [4] <uobj,UVM_UNKNOWN_OFFSET>  == uvm_map finds offset based on VA
  714  *
  715  *    case [4] is for kernel mappings where we don't know the offset until
  716  *    we've found a virtual address.   note that kernel object offsets are
  717  *    always relative to vm_map_min(kernel_map).
  718  *
  719  * => if `align' is non-zero, we align the virtual address to the specified
  720  *      alignment.
  721  *      this is provided as a mechanism for large pages.
  722  *
  723  * => XXXCDC: need way to map in external amap?
  724  */
  725 
  726 int
  727 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size,
  728     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags)
  729 {
  730         struct vm_map_entry *prev_entry, *new_entry;
  731         const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
  732             AMAP_EXTEND_NOWAIT : 0;
  733         vm_prot_t prot = UVM_PROTECTION(flags), maxprot =
  734             UVM_MAXPROTECTION(flags);
  735         vm_inherit_t inherit = UVM_INHERIT(flags);
  736         int advice = UVM_ADVICE(flags);
  737         int error, merged = 0, kmap = (vm_map_pmap(map) == pmap_kernel());
  738         UVMHIST_FUNC("uvm_map");
  739         UVMHIST_CALLED(maphist);
  740 
  741         UVMHIST_LOG(maphist, "(map=0x%x, *startp=0x%x, size=%d, flags=0x%x)",
  742             map, *startp, size, flags);
  743         UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
  744 
  745         /*
  746          * detect a popular device driver bug.
  747          */
  748 
  749         KASSERT(doing_shutdown || curlwp != NULL ||
  750             (map->flags & VM_MAP_INTRSAFE));
  751 
  752         /*
  753          * zero-sized mapping doesn't make any sense.
  754          */
  755         KASSERT(size > 0);
  756 
  757         uvm_tree_sanity(map, "map entry");
  758 
  759         /*
  760          * check sanity of protection code
  761          */
  762 
  763         if ((prot & maxprot) != prot) {
  764                 UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
  765                 prot, maxprot,0,0);
  766                 return EACCES;
  767         }
  768 
  769         /*
  770          * for pager_map, allocate the new entry first to avoid sleeping
  771          * for memory while we have the map locked.
  772          */
  773 
  774         new_entry = NULL;
  775         if (map == pager_map) {
  776                 new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
  777                 if (__predict_false(new_entry == NULL))
  778                         return ENOMEM;
  779         }
  780 
  781         /*
  782          * figure out where to put new VM range
  783          */
  784 
  785         if (vm_map_lock_try(map) == FALSE) {
  786                 if (flags & UVM_FLAG_TRYLOCK) {
  787                         if (new_entry) {
  788                                 uvm_mapent_free(new_entry);
  789                         }
  790                         return EAGAIN;
  791                 }
  792                 vm_map_lock(map); /* could sleep here */
  793         }
  794         if ((prev_entry = uvm_map_findspace(map, *startp, size, startp,
  795             uobj, uoffset, align, flags)) == NULL) {
  796                 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0);
  797                 vm_map_unlock(map);
  798                 if (new_entry) {
  799                         uvm_mapent_free(new_entry);
  800                 }
  801                 return ENOMEM;
  802         }
  803 
  804 #ifdef PMAP_GROWKERNEL
  805         /*
  806          * If the kernel pmap can't map the requested space,
  807          * then allocate more resources for it.
  808          */
  809         if (map == kernel_map && uvm_maxkaddr < (*startp + size))
  810                 uvm_maxkaddr = pmap_growkernel(*startp + size);
  811 #endif
  812 
  813         UVMCNT_INCR(uvm_map_call);
  814 
  815         /*
  816          * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
  817          * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
  818          * either case we want to zero it  before storing it in the map entry
  819          * (because it looks strange and confusing when debugging...)
  820          *
  821          * if uobj is not null
  822          *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
  823          *      and we do not need to change uoffset.
  824          *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
  825          *      now (based on the starting address of the map).   this case is
  826          *      for kernel object mappings where we don't know the offset until
  827          *      the virtual address is found (with uvm_map_findspace).   the
  828          *      offset is the distance we are from the start of the map.
  829          */
  830 
  831         if (uobj == NULL) {
  832                 uoffset = 0;
  833         } else {
  834                 if (uoffset == UVM_UNKNOWN_OFFSET) {
  835                         KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
  836                         uoffset = *startp - vm_map_min(kernel_map);
  837                 }
  838         }
  839 
  840         /*
  841          * try and insert in map by extending previous entry, if possible.
  842          * XXX: we don't try and pull back the next entry.   might be useful
  843          * for a stack, but we are currently allocating our stack in advance.
  844          */
  845 
  846         if (flags & UVM_FLAG_NOMERGE)
  847                 goto nomerge;
  848 
  849         if (prev_entry->end == *startp &&
  850             prev_entry != &map->header &&
  851             prev_entry->object.uvm_obj == uobj) {
  852 
  853                 if (prev_entry->flags & UVM_MAP_NOMERGE)
  854                         goto forwardmerge;
  855 
  856                 if (uobj && prev_entry->offset +
  857                     (prev_entry->end - prev_entry->start) != uoffset)
  858                         goto forwardmerge;
  859 
  860                 if (UVM_ET_ISSUBMAP(prev_entry))
  861                         goto forwardmerge;
  862 
  863                 if (prev_entry->protection != prot ||
  864                     prev_entry->max_protection != maxprot)
  865                         goto forwardmerge;
  866 
  867                 if (prev_entry->inheritance != inherit ||
  868                     prev_entry->advice != advice)
  869                         goto forwardmerge;
  870 
  871                 /* wiring status must match (new area is unwired) */
  872                 if (VM_MAPENT_ISWIRED(prev_entry))
  873                         goto forwardmerge;
  874 
  875                 /*
  876                  * can't extend a shared amap.  note: no need to lock amap to
  877                  * look at refs since we don't care about its exact value.
  878                  * if it is one (i.e. we have only reference) it will stay there
  879                  */
  880 
  881                 if (prev_entry->aref.ar_amap &&
  882                     amap_refs(prev_entry->aref.ar_amap) != 1) {
  883                         goto forwardmerge;
  884                 }
  885 
  886                 if (prev_entry->aref.ar_amap) {
  887                         error = amap_extend(prev_entry, size,
  888                             amapwaitflag | AMAP_EXTEND_FORWARDS);
  889                         if (error) {
  890                                 vm_map_unlock(map);
  891                                 if (new_entry) {
  892                                         uvm_mapent_free(new_entry);
  893                                 }
  894                                 return error;
  895                         }
  896                 }
  897 
  898                 if (kmap)
  899                         UVMCNT_INCR(map_kbackmerge);
  900                 else
  901                         UVMCNT_INCR(map_ubackmerge);
  902                 UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
  903 
  904                 /*
  905                  * drop our reference to uobj since we are extending a reference
  906                  * that we already have (the ref count can not drop to zero).
  907                  */
  908 
  909                 if (uobj && uobj->pgops->pgo_detach)
  910                         uobj->pgops->pgo_detach(uobj);
  911 
  912                 prev_entry->end += size;
  913                 uvm_rb_fixup(map, prev_entry);
  914 
  915                 uvm_tree_sanity(map, "map backmerged");
  916 
  917                 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
  918                 if (new_entry) {
  919                         uvm_mapent_free(new_entry);
  920                         new_entry = NULL;
  921                 }
  922                 merged++;
  923         }
  924 
  925 forwardmerge:
  926         if (prev_entry->next->start == (*startp + size) &&
  927             prev_entry->next != &map->header &&
  928             prev_entry->next->object.uvm_obj == uobj) {
  929 
  930                 if (prev_entry->next->flags & UVM_MAP_NOMERGE)
  931                         goto nomerge;
  932 
  933                 if (uobj && prev_entry->next->offset != uoffset + size)
  934                         goto nomerge;
  935 
  936                 if (UVM_ET_ISSUBMAP(prev_entry->next))
  937                         goto nomerge;
  938 
  939                 if (prev_entry->next->protection != prot ||
  940                     prev_entry->next->max_protection != maxprot)
  941                         goto nomerge;
  942 
  943                 if (prev_entry->next->inheritance != inherit ||
  944                     prev_entry->next->advice != advice)
  945                         goto nomerge;
  946 
  947                 /* wiring status must match (new area is unwired) */
  948                 if (VM_MAPENT_ISWIRED(prev_entry->next))
  949                         goto nomerge;
  950 
  951                 /*
  952                  * can't extend a shared amap.  note: no need to lock amap to
  953                  * look at refs since we don't care about its exact value.
  954                  * if it is one (i.e. we have only reference) it will stay there.
  955                  *
  956                  * note that we also can't merge two amaps, so if we
  957                  * merged with the previous entry which has an amap,
  958                  * and the next entry also has an amap, we give up.
  959                  *
  960                  * Interesting cases:
  961                  * amap, new, amap -> give up second merge (single fwd extend)
  962                  * amap, new, none -> double forward extend (extend again here)
  963                  * none, new, amap -> double backward extend (done here)
  964                  * uobj, new, amap -> single backward extend (done here)
  965                  *
  966                  * XXX should we attempt to deal with someone refilling
  967                  * the deallocated region between two entries that are
  968                  * backed by the same amap (ie, arefs is 2, "prev" and
  969                  * "next" refer to it, and adding this allocation will
  970                  * close the hole, thus restoring arefs to 1 and
  971                  * deallocating the "next" vm_map_entry)?  -- @@@
  972                  */
  973 
  974                 if (prev_entry->next->aref.ar_amap &&
  975                     (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
  976                      (merged && prev_entry->aref.ar_amap))) {
  977                         goto nomerge;
  978                 }
  979 
  980                 if (merged) {
  981                         /*
  982                          * Try to extend the amap of the previous entry to
  983                          * cover the next entry as well.  If it doesn't work
  984                          * just skip on, don't actually give up, since we've
  985                          * already completed the back merge.
  986                          */
  987                         if (prev_entry->aref.ar_amap) {
  988                                 if (amap_extend(prev_entry,
  989                                     prev_entry->next->end -
  990                                     prev_entry->next->start,
  991                                     amapwaitflag | AMAP_EXTEND_FORWARDS))
  992                                         goto nomerge;
  993                         }
  994 
  995                         /*
  996                          * Try to extend the amap of the *next* entry
  997                          * back to cover the new allocation *and* the
  998                          * previous entry as well (the previous merge
  999                          * didn't have an amap already otherwise we
 1000                          * wouldn't be checking here for an amap).  If
 1001                          * it doesn't work just skip on, again, don't
 1002                          * actually give up, since we've already
 1003                          * completed the back merge.
 1004                          */
 1005                         else if (prev_entry->next->aref.ar_amap) {
 1006                                 if (amap_extend(prev_entry->next,
 1007                                     prev_entry->end -
 1008                                     prev_entry->start,
 1009                                     amapwaitflag | AMAP_EXTEND_BACKWARDS))
 1010                                         goto nomerge;
 1011                         }
 1012                 } else {
 1013                         /*
 1014                          * Pull the next entry's amap backwards to cover this
 1015                          * new allocation.
 1016                          */
 1017                         if (prev_entry->next->aref.ar_amap) {
 1018                                 error = amap_extend(prev_entry->next, size,
 1019                                     amapwaitflag | AMAP_EXTEND_BACKWARDS);
 1020                                 if (error) {
 1021                                         vm_map_unlock(map);
 1022                                         if (new_entry) {
 1023                                                 uvm_mapent_free(new_entry);
 1024                                         }
 1025                                         return error;
 1026                                 }
 1027                         }
 1028                 }
 1029 
 1030                 if (merged) {
 1031                         if (kmap) {
 1032                                 UVMCNT_DECR(map_kbackmerge);
 1033                                 UVMCNT_INCR(map_kbimerge);
 1034                         } else {
 1035                                 UVMCNT_DECR(map_ubackmerge);
 1036                                 UVMCNT_INCR(map_ubimerge);
 1037                         }
 1038                 } else {
 1039                         if (kmap)
 1040                                 UVMCNT_INCR(map_kforwmerge);
 1041                         else
 1042                                 UVMCNT_INCR(map_uforwmerge);
 1043                 }
 1044                 UVMHIST_LOG(maphist,"  starting forward merge", 0, 0, 0, 0);
 1045 
 1046                 /*
 1047                  * drop our reference to uobj since we are extending a reference
 1048                  * that we already have (the ref count can not drop to zero).
 1049                  * (if merged, we've already detached)
 1050                  */
 1051                 if (uobj && uobj->pgops->pgo_detach && !merged)
 1052                         uobj->pgops->pgo_detach(uobj);
 1053 
 1054                 if (merged) {
 1055                         struct vm_map_entry *dead = prev_entry->next;
 1056                         prev_entry->end = dead->end;
 1057                         uvm_map_entry_unlink(map, dead);
 1058                         if (dead->aref.ar_amap != NULL) {
 1059                                 prev_entry->aref = dead->aref;
 1060                                 dead->aref.ar_amap = NULL;
 1061                         }
 1062                         uvm_mapent_free(dead);
 1063                 } else {
 1064                         prev_entry->next->start -= size;
 1065                         if (prev_entry != &map->header)
 1066                                 uvm_rb_fixup(map, prev_entry);
 1067                         if (uobj)
 1068                                 prev_entry->next->offset = uoffset;
 1069                 }
 1070 
 1071                 uvm_tree_sanity(map, "map forwardmerged");
 1072 
 1073                 UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
 1074                 if (new_entry) {
 1075                         uvm_mapent_free(new_entry);
 1076                         new_entry = NULL;
 1077                 }
 1078                 merged++;
 1079         }
 1080 
 1081 nomerge:
 1082         if (!merged) {
 1083                 UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
 1084                 if (kmap)
 1085                         UVMCNT_INCR(map_knomerge);
 1086                 else
 1087                         UVMCNT_INCR(map_unomerge);
 1088 
 1089                 /*
 1090                  * allocate new entry and link it in.
 1091                  */
 1092 
 1093                 if (new_entry == NULL) {
 1094                         new_entry = uvm_mapent_alloc(map,
 1095                                 (flags & UVM_FLAG_NOWAIT));
 1096                         if (__predict_false(new_entry == NULL)) {
 1097                                 vm_map_unlock(map);
 1098                                 return ENOMEM;
 1099                         }
 1100                 }
 1101                 new_entry->start = *startp;
 1102                 new_entry->end = new_entry->start + size;
 1103                 new_entry->object.uvm_obj = uobj;
 1104                 new_entry->offset = uoffset;
 1105 
 1106                 if (uobj)
 1107                         new_entry->etype = UVM_ET_OBJ;
 1108                 else
 1109                         new_entry->etype = 0;
 1110 
 1111                 if (flags & UVM_FLAG_COPYONW) {
 1112                         new_entry->etype |= UVM_ET_COPYONWRITE;
 1113                         if ((flags & UVM_FLAG_OVERLAY) == 0)
 1114                                 new_entry->etype |= UVM_ET_NEEDSCOPY;
 1115                 }
 1116                 if (flags & UVM_FLAG_NOMERGE) {
 1117                         new_entry->flags |= UVM_MAP_NOMERGE;
 1118                 }
 1119 
 1120                 new_entry->protection = prot;
 1121                 new_entry->max_protection = maxprot;
 1122                 new_entry->inheritance = inherit;
 1123                 new_entry->wired_count = 0;
 1124                 new_entry->advice = advice;
 1125                 if (flags & UVM_FLAG_OVERLAY) {
 1126 
 1127                         /*
 1128                          * to_add: for BSS we overallocate a little since we
 1129                          * are likely to extend
 1130                          */
 1131 
 1132                         vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
 1133                                 UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
 1134                         struct vm_amap *amap = amap_alloc(size, to_add,
 1135                             (flags & UVM_FLAG_NOWAIT) ? M_NOWAIT : M_WAITOK);
 1136                         if (__predict_false(amap == NULL)) {
 1137                                 vm_map_unlock(map);
 1138                                 uvm_mapent_free(new_entry);
 1139                                 return ENOMEM;
 1140                         }
 1141                         new_entry->aref.ar_pageoff = 0;
 1142                         new_entry->aref.ar_amap = amap;
 1143                 } else {
 1144                         new_entry->aref.ar_pageoff = 0;
 1145                         new_entry->aref.ar_amap = NULL;
 1146                 }
 1147                 uvm_map_entry_link(map, prev_entry, new_entry);
 1148 
 1149                 /*
 1150                  * Update the free space hint
 1151                  */
 1152 
 1153                 if ((map->first_free == prev_entry) &&
 1154                     (prev_entry->end >= new_entry->start))
 1155                         map->first_free = new_entry;
 1156         }
 1157 
 1158         map->size += size;
 1159 
 1160         UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
 1161         vm_map_unlock(map);
 1162         return 0;
 1163 }
 1164 
 1165 /*
 1166  * uvm_map_lookup_entry: find map entry at or before an address
 1167  *
 1168  * => map must at least be read-locked by caller
 1169  * => entry is returned in "entry"
 1170  * => return value is true if address is in the returned entry
 1171  */
 1172 
 1173 boolean_t
 1174 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address,
 1175     struct vm_map_entry **entry /* OUT */)
 1176 {
 1177         struct vm_map_entry *cur;
 1178         boolean_t use_tree = FALSE;
 1179         UVMHIST_FUNC("uvm_map_lookup_entry");
 1180         UVMHIST_CALLED(maphist);
 1181 
 1182         UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
 1183             map, address, entry, 0);
 1184 
 1185         /*
 1186          * start looking either from the head of the
 1187          * list, or from the hint.
 1188          */
 1189 
 1190         simple_lock(&map->hint_lock);
 1191         cur = map->hint;
 1192         simple_unlock(&map->hint_lock);
 1193 
 1194         if (cur == &map->header)
 1195                 cur = cur->next;
 1196 
 1197         UVMCNT_INCR(uvm_mlk_call);
 1198         if (address >= cur->start) {
 1199 
 1200                 /*
 1201                  * go from hint to end of list.
 1202                  *
 1203                  * but first, make a quick check to see if
 1204                  * we are already looking at the entry we
 1205                  * want (which is usually the case).
 1206                  * note also that we don't need to save the hint
 1207                  * here... it is the same hint (unless we are
 1208                  * at the header, in which case the hint didn't
 1209                  * buy us anything anyway).
 1210                  */
 1211 
 1212                 if (cur != &map->header && cur->end > address) {
 1213                         UVMCNT_INCR(uvm_mlk_hint);
 1214                         *entry = cur;
 1215                         UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
 1216                             cur, 0, 0, 0);
 1217                         return (TRUE);
 1218                 }
 1219 
 1220                 if (map->nentries > 30)
 1221                         use_tree = TRUE;
 1222         } else {
 1223 
 1224                 /*
 1225                  * invalid hint.  use tree.
 1226                  */
 1227                 use_tree = TRUE;
 1228         }
 1229 
 1230         uvm_tree_sanity(map, __func__);
 1231 
 1232         if (use_tree) {
 1233                 struct vm_map_entry *prev = &map->header;
 1234                 cur = RB_ROOT(&map->rbhead);
 1235 
 1236                 /*
 1237                  * Simple lookup in the tree.  Happens when the hint is
 1238                  * invalid, or nentries reach a threshold.
 1239                  */
 1240                 while (cur) {
 1241                         if (address >= cur->start) {
 1242                                 if (address < cur->end) {
 1243                                         *entry = cur;
 1244                                         goto got;
 1245                                 }
 1246                                 prev = cur;
 1247                                 cur = RB_RIGHT(cur, rb_entry);
 1248                         } else
 1249                                 cur = RB_LEFT(cur, rb_entry);
 1250                 }
 1251                 *entry = prev;
 1252                 goto failed;
 1253         }
 1254 
 1255         /*
 1256          * search linearly
 1257          */
 1258 
 1259         while (cur != &map->header) {
 1260                 if (cur->end > address) {
 1261                         if (address >= cur->start) {
 1262                                 /*
 1263                                  * save this lookup for future
 1264                                  * hints, and return
 1265                                  */
 1266 
 1267                                 *entry = cur;
 1268 got:
 1269                                 SAVE_HINT(map, map->hint, *entry);
 1270                                 UVMHIST_LOG(maphist,"<- search got it (0x%x)",
 1271                                         cur, 0, 0, 0);
 1272                                 KDASSERT((*entry)->start <= address);
 1273                                 KDASSERT(address < (*entry)->end);
 1274                                 return (TRUE);
 1275                         }
 1276                         break;
 1277                 }
 1278                 cur = cur->next;
 1279         }
 1280         *entry = cur->prev;
 1281 failed:
 1282         SAVE_HINT(map, map->hint, *entry);
 1283         UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
 1284         KDASSERT((*entry) == &map->header || (*entry)->end <= address);
 1285         KDASSERT((*entry)->next == &map->header ||
 1286             address < (*entry)->next->start);
 1287         return (FALSE);
 1288 }
 1289 
 1290 /*
 1291  * See if the range between start and start + length fits in the gap
 1292  * entry->next->start and entry->end.  Returns 1 if fits, 0 if doesn't
 1293  * fit, and -1 address wraps around.
 1294  */
 1295 static __inline int
 1296 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset,
 1297     vsize_t align, int topdown, struct vm_map_entry *entry)
 1298 {
 1299         vaddr_t end;
 1300 
 1301 #ifdef PMAP_PREFER
 1302         /*
 1303          * push start address forward as needed to avoid VAC alias problems.
 1304          * we only do this if a valid offset is specified.
 1305          */
 1306 
 1307         if (uoffset != UVM_UNKNOWN_OFFSET)
 1308                 PMAP_PREFER(uoffset, start);
 1309 #endif
 1310         if (align != 0) {
 1311                 if ((*start & (align - 1)) != 0) {
 1312                         if (topdown)
 1313                                 *start &= ~(align - 1);
 1314                         else
 1315                                 *start = roundup(*start, align);
 1316                 }
 1317                 /*
 1318                  * XXX Should we PMAP_PREFER() here again?
 1319                  */
 1320         }
 1321 
 1322         /*
 1323          * Find the end of the proposed new region.  Be sure we didn't
 1324          * wrap around the address; if so, we lose.  Otherwise, if the
 1325          * proposed new region fits before the next entry, we win.
 1326          */
 1327 
 1328         end = *start + length;
 1329         if (end < *start)
 1330                 return (-1);
 1331 
 1332         if (entry->next->start >= end && *start >= entry->end)
 1333                 return (1);
 1334 
 1335         return (0);
 1336 }
 1337 
 1338 /*
 1339  * uvm_map_findspace: find "length" sized space in "map".
 1340  *
 1341  * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is
 1342  *      set (in which case we insist on using "hint").
 1343  * => "result" is VA returned
 1344  * => uobj/uoffset are to be used to handle VAC alignment, if required
 1345  * => if `align' is non-zero, we attempt to align to that value.
 1346  * => caller must at least have read-locked map
 1347  * => returns NULL on failure, or pointer to prev. map entry if success
 1348  * => note this is a cross between the old vm_map_findspace and vm_map_find
 1349  */
 1350 
 1351 struct vm_map_entry *
 1352 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length,
 1353     vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset,
 1354     vsize_t align, int flags)
 1355 {
 1356         struct vm_map_entry *entry;
 1357         struct vm_map_entry *child, *prev, *tmp;
 1358         vaddr_t orig_hint;
 1359         const int topdown = map->flags & VM_MAP_TOPDOWN;
 1360         UVMHIST_FUNC("uvm_map_findspace");
 1361         UVMHIST_CALLED(maphist);
 1362 
 1363         UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
 1364             map, hint, length, flags);
 1365         KASSERT((align & (align - 1)) == 0);
 1366         KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
 1367 
 1368         uvm_tree_sanity(map, "map_findspace entry");
 1369 
 1370         /*
 1371          * remember the original hint.  if we are aligning, then we
 1372          * may have to try again with no alignment constraint if
 1373          * we fail the first time.
 1374          */
 1375 
 1376         orig_hint = hint;
 1377         if (hint < map->min_offset) {   /* check ranges ... */
 1378                 if (flags & UVM_FLAG_FIXED) {
 1379                         UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
 1380                         return (NULL);
 1381                 }
 1382                 hint = map->min_offset;
 1383         }
 1384         if (hint > map->max_offset) {
 1385                 UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
 1386                     hint, map->min_offset, map->max_offset, 0);
 1387                 return (NULL);
 1388         }
 1389 
 1390         /*
 1391          * Look for the first possible address; if there's already
 1392          * something at this address, we have to start after it.
 1393          */
 1394 
 1395         /*
 1396          * @@@: there are four, no, eight cases to consider.
 1397          *
 1398          * 0: found,     fixed,     bottom up -> fail
 1399          * 1: found,     fixed,     top down  -> fail
 1400          * 2: found,     not fixed, bottom up -> start after entry->end,
 1401          *                                       loop up
 1402          * 3: found,     not fixed, top down  -> start before entry->start,
 1403          *                                       loop down
 1404          * 4: not found, fixed,     bottom up -> check entry->next->start, fail
 1405          * 5: not found, fixed,     top down  -> check entry->next->start, fail
 1406          * 6: not found, not fixed, bottom up -> check entry->next->start,
 1407          *                                       loop up
 1408          * 7: not found, not fixed, top down  -> check entry->next->start,
 1409          *                                       loop down
 1410          *
 1411          * as you can see, it reduces to roughly five cases, and that
 1412          * adding top down mapping only adds one unique case (without
 1413          * it, there would be four cases).
 1414          */
 1415 
 1416         if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) {
 1417                 entry = map->first_free;
 1418         } else {
 1419                 if (uvm_map_lookup_entry(map, hint, &entry)) {
 1420                         /* "hint" address already in use ... */
 1421                         if (flags & UVM_FLAG_FIXED) {
 1422                                 UVMHIST_LOG(maphist, "<- fixed & VA in use",
 1423                                     0, 0, 0, 0);
 1424                                 return (NULL);
 1425                         }
 1426                         if (topdown)
 1427                                 /* Start from lower gap. */
 1428                                 entry = entry->prev;
 1429                 } else if (flags & UVM_FLAG_FIXED) {
 1430                         if (entry->next->start >= hint + length &&
 1431                             hint + length > hint)
 1432                                 goto found;
 1433 
 1434                         /* "hint" address is gap but too small */
 1435                         UVMHIST_LOG(maphist, "<- fixed mapping failed",
 1436                             0, 0, 0, 0);
 1437                         return (NULL); /* only one shot at it ... */
 1438                 } else {
 1439                         /*
 1440                          * See if given hint fits in this gap.
 1441                          */
 1442                         switch (uvm_map_space_avail(&hint, length,
 1443                             uoffset, align, topdown, entry)) {
 1444                         case 1:
 1445                                 goto found;
 1446                         case -1:
 1447                                 goto wraparound;
 1448                         }
 1449 
 1450                         if (topdown) {
 1451                                 /*
 1452                                  * Still there is a chance to fit
 1453                                  * if hint > entry->end.
 1454                                  */
 1455                         } else {
 1456                                 /* Start from higer gap. */
 1457                                 entry = entry->next;
 1458                                 if (entry == &map->header)
 1459                                         goto notfound;
 1460                                 goto nextgap;
 1461                         }
 1462                 }
 1463         }
 1464 
 1465         /*
 1466          * Note that all UVM_FLAGS_FIXED case is already handled.
 1467          */
 1468         KDASSERT((flags & UVM_FLAG_FIXED) == 0);
 1469 
 1470         /* Try to find the space in the red-black tree */
 1471 
 1472         /* Check slot before any entry */
 1473         hint = topdown ? entry->next->start - length : entry->end;
 1474         switch (uvm_map_space_avail(&hint, length, uoffset, align,
 1475             topdown, entry)) {
 1476         case 1:
 1477                 goto found;
 1478         case -1:
 1479                 goto wraparound;
 1480         }
 1481 
 1482 nextgap:
 1483         KDASSERT((flags & UVM_FLAG_FIXED) == 0);
 1484         /* If there is not enough space in the whole tree, we fail */
 1485         tmp = RB_ROOT(&map->rbhead);
 1486         if (tmp == NULL || tmp->space < length)
 1487                 goto notfound;
 1488 
 1489         prev = NULL; /* previous candidate */
 1490 
 1491         /* Find an entry close to hint that has enough space */
 1492         for (; tmp;) {
 1493                 KASSERT(tmp->next->start == tmp->end + tmp->ownspace);
 1494                 if (topdown) {
 1495                         if (tmp->next->start < hint + length &&
 1496                             (prev == NULL || tmp->end > prev->end)) {
 1497                                 if (tmp->ownspace >= length)
 1498                                         prev = tmp;
 1499                                 else if ((child = RB_LEFT(tmp, rb_entry))
 1500                                     != NULL && child->space >= length)
 1501                                         prev = tmp;
 1502                         }
 1503                 } else {
 1504                         if (tmp->end >= hint &&
 1505                             (prev == NULL || tmp->end < prev->end)) {
 1506                                 if (tmp->ownspace >= length)
 1507                                         prev = tmp;
 1508                                 else if ((child = RB_RIGHT(tmp, rb_entry))
 1509                                     != NULL && child->space >= length)
 1510                                         prev = tmp;
 1511                         }
 1512                 }
 1513                 if (tmp->next->start < hint + length)
 1514                         child = RB_RIGHT(tmp, rb_entry);
 1515                 else if (tmp->end > hint)
 1516                         child = RB_LEFT(tmp, rb_entry);
 1517                 else {
 1518                         if (tmp->ownspace >= length)
 1519                                 break;
 1520                         if (topdown)
 1521                                 child = RB_LEFT(tmp, rb_entry);
 1522                         else
 1523                                 child = RB_RIGHT(tmp, rb_entry);
 1524                 }
 1525                 if (child == NULL || child->space < length)
 1526                         break;
 1527                 tmp = child;
 1528         }
 1529 
 1530         if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) {
 1531                 /*
 1532                  * Check if the entry that we found satifies the
 1533                  * space requirement
 1534                  */
 1535                 if (topdown) {
 1536                         if (hint > tmp->next->start - length)
 1537                                 hint = tmp->next->start - length;
 1538                 } else {
 1539                         if (hint < tmp->end)
 1540                                 hint = tmp->end;
 1541                 }
 1542                 switch (uvm_map_space_avail(&hint, length, uoffset, align,
 1543                     topdown, tmp)) {
 1544                 case 1:
 1545                         entry = tmp;
 1546                         goto found;
 1547                 case -1:
 1548                         goto wraparound;
 1549                 }
 1550                 if (tmp->ownspace >= length)
 1551                         goto listsearch;
 1552         }
 1553         if (prev == NULL)
 1554                 goto notfound;
 1555 
 1556         if (topdown) {
 1557                 KASSERT(orig_hint >= prev->next->start - length ||
 1558                     prev->next->start - length > prev->next->start);
 1559                 hint = prev->next->start - length;
 1560         } else {
 1561                 KASSERT(orig_hint <= prev->end);
 1562                 hint = prev->end;
 1563         }
 1564         switch (uvm_map_space_avail(&hint, length, uoffset, align,
 1565             topdown, prev)) {
 1566         case 1:
 1567                 entry = prev;
 1568                 goto found;
 1569         case -1:
 1570                 goto wraparound;
 1571         }
 1572         if (prev->ownspace >= length)
 1573                 goto listsearch;
 1574 
 1575         if (topdown)
 1576                 tmp = RB_LEFT(prev, rb_entry);
 1577         else
 1578                 tmp = RB_RIGHT(prev, rb_entry);
 1579         for (;;) {
 1580                 KASSERT(tmp && tmp->space >= length);
 1581                 if (topdown)
 1582                         child = RB_RIGHT(tmp, rb_entry);
 1583                 else
 1584                         child = RB_LEFT(tmp, rb_entry);
 1585                 if (child && child->space >= length) {
 1586                         tmp = child;
 1587                         continue;
 1588                 }
 1589                 if (tmp->ownspace >= length)
 1590                         break;
 1591                 if (topdown)
 1592                         tmp = RB_LEFT(tmp, rb_entry);
 1593                 else
 1594                         tmp = RB_RIGHT(tmp, rb_entry);
 1595         }
 1596 
 1597         if (topdown) {
 1598                 KASSERT(orig_hint >= tmp->next->start - length ||
 1599                     tmp->next->start - length > tmp->next->start);
 1600                 hint = tmp->next->start - length;
 1601         } else {
 1602                 KASSERT(orig_hint <= tmp->end);
 1603                 hint = tmp->end;
 1604         }
 1605         switch (uvm_map_space_avail(&hint, length, uoffset, align,
 1606             topdown, tmp)) {
 1607         case 1:
 1608                 entry = tmp;
 1609                 goto found;
 1610         case -1:
 1611                 goto wraparound;
 1612         }
 1613 
 1614         /*
 1615          * The tree fails to find an entry because of offset or alignment
 1616          * restrictions.  Search the list instead.
 1617          */
 1618  listsearch:
 1619         /*
 1620          * Look through the rest of the map, trying to fit a new region in
 1621          * the gap between existing regions, or after the very last region.
 1622          * note: entry->end = base VA of current gap,
 1623          *       entry->next->start = VA of end of current gap
 1624          */
 1625 
 1626         for (;;) {
 1627                 /* Update hint for current gap. */
 1628                 hint = topdown ? entry->next->start - length : entry->end;
 1629 
 1630                 /* See if it fits. */
 1631                 switch (uvm_map_space_avail(&hint, length, uoffset, align,
 1632                     topdown, entry)) {
 1633                 case 1:
 1634                         goto found;
 1635                 case -1:
 1636                         goto wraparound;
 1637                 }
 1638 
 1639                 /* Advance to next/previous gap */
 1640                 if (topdown) {
 1641                         if (entry == &map->header) {
 1642                                 UVMHIST_LOG(maphist, "<- failed (off start)",
 1643                                     0,0,0,0);
 1644                                 goto notfound;
 1645                         }
 1646                         entry = entry->prev;
 1647                 } else {
 1648                         entry = entry->next;
 1649                         if (entry == &map->header) {
 1650                                 UVMHIST_LOG(maphist, "<- failed (off end)",
 1651                                     0,0,0,0);
 1652                                 goto notfound;
 1653                         }
 1654                 }
 1655         }
 1656 
 1657  found:
 1658         SAVE_HINT(map, map->hint, entry);
 1659         *result = hint;
 1660         UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
 1661         KASSERT( topdown || hint >= orig_hint);
 1662         KASSERT(!topdown || hint <= orig_hint);
 1663         KASSERT(entry->end <= hint);
 1664         KASSERT(hint + length <= entry->next->start);
 1665         return (entry);
 1666 
 1667  wraparound:
 1668         UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0);
 1669 
 1670         return (NULL);
 1671 
 1672  notfound:
 1673         UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0);
 1674 
 1675         return (NULL);
 1676 }
 1677 
 1678 /*
 1679  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
 1680  */
 1681 
 1682 /*
 1683  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
 1684  *
 1685  * => caller must check alignment and size
 1686  * => map must be locked by caller
 1687  * => we return a list of map entries that we've remove from the map
 1688  *    in "entry_list"
 1689  */
 1690 
 1691 void
 1692 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end,
 1693     struct vm_map_entry **entry_list /* OUT */)
 1694 {
 1695         struct vm_map_entry *entry, *first_entry, *next;
 1696         vaddr_t len;
 1697         UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
 1698 
 1699         UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
 1700             map, start, end, 0);
 1701         VM_MAP_RANGE_CHECK(map, start, end);
 1702 
 1703         uvm_tree_sanity(map, "unmap_remove entry");
 1704 
 1705         /*
 1706          * find first entry
 1707          */
 1708 
 1709         if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) {
 1710                 /* clip and go... */
 1711                 entry = first_entry;
 1712                 UVM_MAP_CLIP_START(map, entry, start);
 1713                 /* critical!  prevents stale hint */
 1714                 SAVE_HINT(map, entry, entry->prev);
 1715         } else {
 1716                 entry = first_entry->next;
 1717         }
 1718 
 1719         /*
 1720          * Save the free space hint
 1721          */
 1722 
 1723         if (map->first_free->start >= start)
 1724                 map->first_free = entry->prev;
 1725 
 1726         /*
 1727          * note: we now re-use first_entry for a different task.  we remove
 1728          * a number of map entries from the map and save them in a linked
 1729          * list headed by "first_entry".  once we remove them from the map
 1730          * the caller should unlock the map and drop the references to the
 1731          * backing objects [c.f. uvm_unmap_detach].  the object is to
 1732          * separate unmapping from reference dropping.  why?
 1733          *   [1] the map has to be locked for unmapping
 1734          *   [2] the map need not be locked for reference dropping
 1735          *   [3] dropping references may trigger pager I/O, and if we hit
 1736          *       a pager that does synchronous I/O we may have to wait for it.
 1737          *   [4] we would like all waiting for I/O to occur with maps unlocked
 1738          *       so that we don't block other threads.
 1739          */
 1740 
 1741         first_entry = NULL;
 1742         *entry_list = NULL;
 1743 
 1744         /*
 1745          * break up the area into map entry sized regions and unmap.  note
 1746          * that all mappings have to be removed before we can even consider
 1747          * dropping references to amaps or VM objects (otherwise we could end
 1748          * up with a mapping to a page on the free list which would be very bad)
 1749          */
 1750 
 1751         while ((entry != &map->header) && (entry->start < end)) {
 1752                 UVM_MAP_CLIP_END(map, entry, end);
 1753                 next = entry->next;
 1754                 len = entry->end - entry->start;
 1755 
 1756                 /*
 1757                  * unwire before removing addresses from the pmap; otherwise
 1758                  * unwiring will put the entries back into the pmap (XXX).
 1759                  */
 1760 
 1761                 if (VM_MAPENT_ISWIRED(entry)) {
 1762                         uvm_map_entry_unwire(map, entry);
 1763                 }
 1764                 if ((map->flags & VM_MAP_PAGEABLE) == 0) {
 1765 
 1766                         /*
 1767                          * if the map is non-pageable, any pages mapped there
 1768                          * must be wired and entered with pmap_kenter_pa(),
 1769                          * and we should free any such pages immediately.
 1770                          * this is mostly used for kmem_map and mb_map.
 1771                          */
 1772 
 1773                         uvm_km_pgremove_intrsafe(entry->start, entry->end);
 1774                         pmap_kremove(entry->start, len);
 1775                 } else if (UVM_ET_ISOBJ(entry) &&
 1776                            UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
 1777                         KASSERT(vm_map_pmap(map) == pmap_kernel());
 1778 
 1779                         /*
 1780                          * note: kernel object mappings are currently used in
 1781                          * two ways:
 1782                          *  [1] "normal" mappings of pages in the kernel object
 1783                          *  [2] uvm_km_valloc'd allocations in which we
 1784                          *      pmap_enter in some non-kernel-object page
 1785                          *      (e.g. vmapbuf).
 1786                          *
 1787                          * for case [1], we need to remove the mapping from
 1788                          * the pmap and then remove the page from the kernel
 1789                          * object (because, once pages in a kernel object are
 1790                          * unmapped they are no longer needed, unlike, say,
 1791                          * a vnode where you might want the data to persist
 1792                          * until flushed out of a queue).
 1793                          *
 1794                          * for case [2], we need to remove the mapping from
 1795                          * the pmap.  there shouldn't be any pages at the
 1796                          * specified offset in the kernel object [but it
 1797                          * doesn't hurt to call uvm_km_pgremove just to be
 1798                          * safe?]
 1799                          *
 1800                          * uvm_km_pgremove currently does the following:
 1801                          *   for pages in the kernel object in range:
 1802                          *     - drops the swap slot
 1803                          *     - uvm_pagefree the page
 1804                          */
 1805 
 1806                         /*
 1807                          * remove mappings from pmap and drop the pages
 1808                          * from the object.  offsets are always relative
 1809                          * to vm_map_min(kernel_map).
 1810                          */
 1811 
 1812                         pmap_remove(pmap_kernel(), entry->start,
 1813                             entry->start + len);
 1814                         uvm_km_pgremove(entry->object.uvm_obj,
 1815                             entry->start - vm_map_min(kernel_map),
 1816                             entry->end - vm_map_min(kernel_map));
 1817 
 1818                         /*
 1819                          * null out kernel_object reference, we've just
 1820                          * dropped it
 1821                          */
 1822 
 1823                         entry->etype &= ~UVM_ET_OBJ;
 1824                         entry->object.uvm_obj = NULL;
 1825                 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
 1826 
 1827                         /*
 1828                          * remove mappings the standard way.
 1829                          */
 1830 
 1831                         pmap_remove(map->pmap, entry->start, entry->end);
 1832                 }
 1833 
 1834                 /*
 1835                  * remove entry from map and put it on our list of entries
 1836                  * that we've nuked.  then go to next entry.
 1837                  */
 1838 
 1839                 UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
 1840 
 1841                 /* critical!  prevents stale hint */
 1842                 SAVE_HINT(map, entry, entry->prev);
 1843 
 1844                 uvm_map_entry_unlink(map, entry);
 1845                 KASSERT(map->size >= len);
 1846                 map->size -= len;
 1847                 entry->prev = NULL;
 1848                 entry->next = first_entry;
 1849                 first_entry = entry;
 1850                 entry = next;
 1851         }
 1852         if ((map->flags & VM_MAP_DYING) == 0) {
 1853                 pmap_update(vm_map_pmap(map));
 1854         }
 1855 
 1856         uvm_tree_sanity(map, "unmap_remove leave");
 1857 
 1858         /*
 1859          * now we've cleaned up the map and are ready for the caller to drop
 1860          * references to the mapped objects.
 1861          */
 1862 
 1863         *entry_list = first_entry;
 1864         UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
 1865 }
 1866 
 1867 /*
 1868  * uvm_unmap_detach: drop references in a chain of map entries
 1869  *
 1870  * => we will free the map entries as we traverse the list.
 1871  */
 1872 
 1873 void
 1874 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags)
 1875 {
 1876         struct vm_map_entry *next_entry;
 1877         UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
 1878 
 1879         while (first_entry) {
 1880                 KASSERT(!VM_MAPENT_ISWIRED(first_entry));
 1881                 UVMHIST_LOG(maphist,
 1882                     "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
 1883                     first_entry, first_entry->aref.ar_amap,
 1884                     first_entry->object.uvm_obj,
 1885                     UVM_ET_ISSUBMAP(first_entry));
 1886 
 1887                 /*
 1888                  * drop reference to amap, if we've got one
 1889                  */
 1890 
 1891                 if (first_entry->aref.ar_amap)
 1892                         uvm_map_unreference_amap(first_entry, flags);
 1893 
 1894                 /*
 1895                  * drop reference to our backing object, if we've got one
 1896                  */
 1897 
 1898                 KASSERT(!UVM_ET_ISSUBMAP(first_entry));
 1899                 if (UVM_ET_ISOBJ(first_entry) &&
 1900                     first_entry->object.uvm_obj->pgops->pgo_detach) {
 1901                         (*first_entry->object.uvm_obj->pgops->pgo_detach)
 1902                                 (first_entry->object.uvm_obj);
 1903                 }
 1904                 next_entry = first_entry->next;
 1905                 uvm_mapent_free(first_entry);
 1906                 first_entry = next_entry;
 1907         }
 1908         UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
 1909 }
 1910 
 1911 /*
 1912  *   E X T R A C T I O N   F U N C T I O N S
 1913  */
 1914 
 1915 /*
 1916  * uvm_map_reserve: reserve space in a vm_map for future use.
 1917  *
 1918  * => we reserve space in a map by putting a dummy map entry in the
 1919  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
 1920  * => map should be unlocked (we will write lock it)
 1921  * => we return true if we were able to reserve space
 1922  * => XXXCDC: should be inline?
 1923  */
 1924 
 1925 int
 1926 uvm_map_reserve(struct vm_map *map, vsize_t size,
 1927     vaddr_t offset      /* hint for pmap_prefer */,
 1928     vsize_t align       /* alignment hint */,
 1929     vaddr_t *raddr      /* IN:hint, OUT: reserved VA */)
 1930 {
 1931         UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
 1932 
 1933         UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
 1934             map,size,offset,raddr);
 1935 
 1936         size = round_page(size);
 1937         if (*raddr < vm_map_min(map))
 1938                 *raddr = vm_map_min(map);               /* hint */
 1939 
 1940         /*
 1941          * reserve some virtual space.
 1942          */
 1943 
 1944         if (uvm_map(map, raddr, size, NULL, offset, 0,
 1945             UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
 1946             UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) {
 1947             UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
 1948                 return (FALSE);
 1949         }
 1950 
 1951         UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
 1952         return (TRUE);
 1953 }
 1954 
 1955 /*
 1956  * uvm_map_replace: replace a reserved (blank) area of memory with
 1957  * real mappings.
 1958  *
 1959  * => caller must WRITE-LOCK the map
 1960  * => we return TRUE if replacement was a success
 1961  * => we expect the newents chain to have nnewents entrys on it and
 1962  *    we expect newents->prev to point to the last entry on the list
 1963  * => note newents is allowed to be NULL
 1964  */
 1965 
 1966 int
 1967 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end,
 1968     struct vm_map_entry *newents, int nnewents)
 1969 {
 1970         struct vm_map_entry *oldent, *last;
 1971 
 1972         uvm_tree_sanity(map, "map_replace entry");
 1973 
 1974         /*
 1975          * first find the blank map entry at the specified address
 1976          */
 1977 
 1978         if (!uvm_map_lookup_entry(map, start, &oldent)) {
 1979                 return (FALSE);
 1980         }
 1981 
 1982         /*
 1983          * check to make sure we have a proper blank entry
 1984          */
 1985 
 1986         if (oldent->start != start || oldent->end != end ||
 1987             oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
 1988                 return (FALSE);
 1989         }
 1990 
 1991 #ifdef DIAGNOSTIC
 1992 
 1993         /*
 1994          * sanity check the newents chain
 1995          */
 1996 
 1997         {
 1998                 struct vm_map_entry *tmpent = newents;
 1999                 int nent = 0;
 2000                 vaddr_t cur = start;
 2001 
 2002                 while (tmpent) {
 2003                         nent++;
 2004                         if (tmpent->start < cur)
 2005                                 panic("uvm_map_replace1");
 2006                         if (tmpent->start > tmpent->end || tmpent->end > end) {
 2007                 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
 2008                             tmpent->start, tmpent->end, end);
 2009                                 panic("uvm_map_replace2");
 2010                         }
 2011                         cur = tmpent->end;
 2012                         if (tmpent->next) {
 2013                                 if (tmpent->next->prev != tmpent)
 2014                                         panic("uvm_map_replace3");
 2015                         } else {
 2016                                 if (newents->prev != tmpent)
 2017                                         panic("uvm_map_replace4");
 2018                         }
 2019                         tmpent = tmpent->next;
 2020                 }
 2021                 if (nent != nnewents)
 2022                         panic("uvm_map_replace5");
 2023         }
 2024 #endif
 2025 
 2026         /*
 2027          * map entry is a valid blank!   replace it.   (this does all the
 2028          * work of map entry link/unlink...).
 2029          */
 2030 
 2031         if (newents) {
 2032                 last = newents->prev;
 2033 
 2034                 /* critical: flush stale hints out of map */
 2035                 SAVE_HINT(map, map->hint, newents);
 2036                 if (map->first_free == oldent)
 2037                         map->first_free = last;
 2038 
 2039                 last->next = oldent->next;
 2040                 last->next->prev = last;
 2041 
 2042                 /* Fix RB tree */
 2043                 uvm_rb_remove(map, oldent);
 2044 
 2045                 newents->prev = oldent->prev;
 2046                 newents->prev->next = newents;
 2047                 map->nentries = map->nentries + (nnewents - 1);
 2048 
 2049                 /* Fixup the RB tree */
 2050                 {
 2051                         int i;
 2052                         struct vm_map_entry *tmp;
 2053 
 2054                         tmp = newents;
 2055                         for (i = 0; i < nnewents && tmp; i++) {
 2056                                 uvm_rb_insert(map, tmp);
 2057                                 tmp = tmp->next;
 2058                         }
 2059                 }
 2060         } else {
 2061 
 2062                 /* critical: flush stale hints out of map */
 2063                 SAVE_HINT(map, map->hint, oldent->prev);
 2064                 if (map->first_free == oldent)
 2065                         map->first_free = oldent->prev;
 2066 
 2067                 /* NULL list of new entries: just remove the old one */
 2068                 uvm_map_entry_unlink(map, oldent);
 2069         }
 2070 
 2071         uvm_tree_sanity(map, "map_replace leave");
 2072 
 2073         /*
 2074          * now we can free the old blank entry, unlock the map and return.
 2075          */
 2076 
 2077         uvm_mapent_free(oldent);
 2078         return (TRUE);
 2079 }
 2080 
 2081 /*
 2082  * uvm_map_extract: extract a mapping from a map and put it somewhere
 2083  *      (maybe removing the old mapping)
 2084  *
 2085  * => maps should be unlocked (we will write lock them)
 2086  * => returns 0 on success, error code otherwise
 2087  * => start must be page aligned
 2088  * => len must be page sized
 2089  * => flags:
 2090  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
 2091  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
 2092  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
 2093  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
 2094  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
 2095  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
 2096  *             be used from within the kernel in a kernel level map <<<
 2097  */
 2098 
 2099 int
 2100 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len,
 2101     struct vm_map *dstmap, vaddr_t *dstaddrp, int flags)
 2102 {
 2103         vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge;
 2104         struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
 2105             *deadentry, *oldentry;
 2106         vsize_t elen;
 2107         int nchain, error, copy_ok;
 2108         UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
 2109 
 2110         UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
 2111             len,0);
 2112         UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
 2113 
 2114         uvm_tree_sanity(srcmap, "map_extract src enter");
 2115         uvm_tree_sanity(dstmap, "map_extract dst enter");
 2116 
 2117         /*
 2118          * step 0: sanity check: start must be on a page boundary, length
 2119          * must be page sized.  can't ask for CONTIG/QREF if you asked for
 2120          * REMOVE.
 2121          */
 2122 
 2123         KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
 2124         KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
 2125                 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
 2126 
 2127         /*
 2128          * step 1: reserve space in the target map for the extracted area
 2129          */
 2130 
 2131         dstaddr = vm_map_min(dstmap);
 2132         if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE)
 2133                 return (ENOMEM);
 2134         *dstaddrp = dstaddr;    /* pass address back to caller */
 2135         UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
 2136 
 2137         /*
 2138          * step 2: setup for the extraction process loop by init'ing the
 2139          * map entry chain, locking src map, and looking up the first useful
 2140          * entry in the map.
 2141          */
 2142 
 2143         end = start + len;
 2144         newend = dstaddr + len;
 2145         chain = endchain = NULL;
 2146         nchain = 0;
 2147         vm_map_lock(srcmap);
 2148 
 2149         if (uvm_map_lookup_entry(srcmap, start, &entry)) {
 2150 
 2151                 /* "start" is within an entry */
 2152                 if (flags & UVM_EXTRACT_QREF) {
 2153 
 2154                         /*
 2155                          * for quick references we don't clip the entry, so
 2156                          * the entry may map space "before" the starting
 2157                          * virtual address... this is the "fudge" factor
 2158                          * (which can be non-zero only the first time
 2159                          * through the "while" loop in step 3).
 2160                          */
 2161 
 2162                         fudge = start - entry->start;
 2163                 } else {
 2164 
 2165                         /*
 2166                          * normal reference: we clip the map to fit (thus
 2167                          * fudge is zero)
 2168                          */
 2169 
 2170                         UVM_MAP_CLIP_START(srcmap, entry, start);
 2171                         SAVE_HINT(srcmap, srcmap->hint, entry->prev);
 2172                         fudge = 0;
 2173                 }
 2174         } else {
 2175 
 2176                 /* "start" is not within an entry ... skip to next entry */
 2177                 if (flags & UVM_EXTRACT_CONTIG) {
 2178                         error = EINVAL;
 2179                         goto bad;    /* definite hole here ... */
 2180                 }
 2181 
 2182                 entry = entry->next;
 2183                 fudge = 0;
 2184         }
 2185 
 2186         /* save values from srcmap for step 6 */
 2187         orig_entry = entry;
 2188         orig_fudge = fudge;
 2189 
 2190         /*
 2191          * step 3: now start looping through the map entries, extracting
 2192          * as we go.
 2193          */
 2194 
 2195         while (entry->start < end && entry != &srcmap->header) {
 2196 
 2197                 /* if we are not doing a quick reference, clip it */
 2198                 if ((flags & UVM_EXTRACT_QREF) == 0)
 2199                         UVM_MAP_CLIP_END(srcmap, entry, end);
 2200 
 2201                 /* clear needs_copy (allow chunking) */
 2202                 if (UVM_ET_ISNEEDSCOPY(entry)) {
 2203                         amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end);
 2204                         if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
 2205                                 error = ENOMEM;
 2206                                 goto bad;
 2207                         }
 2208 
 2209                         /* amap_copy could clip (during chunk)!  update fudge */
 2210                         if (fudge) {
 2211                                 fudge = start - entry->start;
 2212                                 orig_fudge = fudge;
 2213                         }
 2214                 }
 2215 
 2216                 /* calculate the offset of this from "start" */
 2217                 oldoffset = (entry->start + fudge) - start;
 2218 
 2219                 /* allocate a new map entry */
 2220                 newentry = uvm_mapent_alloc(dstmap, 0);
 2221                 if (newentry == NULL) {
 2222                         error = ENOMEM;
 2223                         goto bad;
 2224                 }
 2225 
 2226                 /* set up new map entry */
 2227                 newentry->next = NULL;
 2228                 newentry->prev = endchain;
 2229                 newentry->start = dstaddr + oldoffset;
 2230                 newentry->end =
 2231                     newentry->start + (entry->end - (entry->start + fudge));
 2232                 if (newentry->end > newend || newentry->end < newentry->start)
 2233                         newentry->end = newend;
 2234                 newentry->object.uvm_obj = entry->object.uvm_obj;
 2235                 if (newentry->object.uvm_obj) {
 2236                         if (newentry->object.uvm_obj->pgops->pgo_reference)
 2237                                 newentry->object.uvm_obj->pgops->
 2238                                     pgo_reference(newentry->object.uvm_obj);
 2239                                 newentry->offset = entry->offset + fudge;
 2240                 } else {
 2241                         newentry->offset = 0;
 2242                 }
 2243                 newentry->etype = entry->etype;
 2244                 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
 2245                         entry->max_protection : entry->protection;
 2246                 newentry->max_protection = entry->max_protection;
 2247                 newentry->inheritance = entry->inheritance;
 2248                 newentry->wired_count = 0;
 2249                 newentry->aref.ar_amap = entry->aref.ar_amap;
 2250                 if (newentry->aref.ar_amap) {
 2251                         newentry->aref.ar_pageoff =
 2252                             entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
 2253                         uvm_map_reference_amap(newentry, AMAP_SHARED |
 2254                             ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
 2255                 } else {
 2256                         newentry->aref.ar_pageoff = 0;
 2257                 }
 2258                 newentry->advice = entry->advice;
 2259 
 2260                 /* now link it on the chain */
 2261                 nchain++;
 2262                 if (endchain == NULL) {
 2263                         chain = endchain = newentry;
 2264                 } else {
 2265                         endchain->next = newentry;
 2266                         endchain = newentry;
 2267                 }
 2268 
 2269                 /* end of 'while' loop! */
 2270                 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
 2271                     (entry->next == &srcmap->header ||
 2272                     entry->next->start != entry->end)) {
 2273                         error = EINVAL;
 2274                         goto bad;
 2275                 }
 2276                 entry = entry->next;
 2277                 fudge = 0;
 2278         }
 2279 
 2280         /*
 2281          * step 4: close off chain (in format expected by uvm_map_replace)
 2282          */
 2283 
 2284         if (chain)
 2285                 chain->prev = endchain;
 2286 
 2287         /*
 2288          * step 5: attempt to lock the dest map so we can pmap_copy.
 2289          * note usage of copy_ok:
 2290          *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
 2291          *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
 2292          */
 2293 
 2294         if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) {
 2295                 copy_ok = 1;
 2296                 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
 2297                     nchain)) {
 2298                         if (srcmap != dstmap)
 2299                                 vm_map_unlock(dstmap);
 2300                         error = EIO;
 2301                         goto bad;
 2302                 }
 2303         } else {
 2304                 copy_ok = 0;
 2305                 /* replace defered until step 7 */
 2306         }
 2307 
 2308         /*
 2309          * step 6: traverse the srcmap a second time to do the following:
 2310          *  - if we got a lock on the dstmap do pmap_copy
 2311          *  - if UVM_EXTRACT_REMOVE remove the entries
 2312          * we make use of orig_entry and orig_fudge (saved in step 2)
 2313          */
 2314 
 2315         if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
 2316 
 2317                 /* purge possible stale hints from srcmap */
 2318                 if (flags & UVM_EXTRACT_REMOVE) {
 2319                         SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
 2320                         if (srcmap->first_free->start >= start)
 2321                                 srcmap->first_free = orig_entry->prev;
 2322                 }
 2323 
 2324                 entry = orig_entry;
 2325                 fudge = orig_fudge;
 2326                 deadentry = NULL;       /* for UVM_EXTRACT_REMOVE */
 2327 
 2328                 while (entry->start < end && entry != &srcmap->header) {
 2329                         if (copy_ok) {
 2330                                 oldoffset = (entry->start + fudge) - start;
 2331                                 elen = MIN(end, entry->end) -
 2332                                     (entry->start + fudge);
 2333                                 pmap_copy(dstmap->pmap, srcmap->pmap,
 2334                                     dstaddr + oldoffset, elen,
 2335                                     entry->start + fudge);
 2336                         }
 2337 
 2338                         /* we advance "entry" in the following if statement */
 2339                         if (flags & UVM_EXTRACT_REMOVE) {
 2340                                 pmap_remove(srcmap->pmap, entry->start,
 2341                                                 entry->end);
 2342                                 oldentry = entry;       /* save entry */
 2343                                 entry = entry->next;    /* advance */
 2344                                 uvm_map_entry_unlink(srcmap, oldentry);
 2345                                                         /* add to dead list */
 2346                                 oldentry->next = deadentry;
 2347                                 deadentry = oldentry;
 2348                         } else {
 2349                                 entry = entry->next;            /* advance */
 2350                         }
 2351 
 2352                         /* end of 'while' loop */
 2353                         fudge = 0;
 2354                 }
 2355                 pmap_update(srcmap->pmap);
 2356 
 2357                 /*
 2358                  * unlock dstmap.  we will dispose of deadentry in
 2359                  * step 7 if needed
 2360                  */
 2361 
 2362                 if (copy_ok && srcmap != dstmap)
 2363                         vm_map_unlock(dstmap);
 2364 
 2365         } else {
 2366                 deadentry = NULL;
 2367         }
 2368 
 2369         /*
 2370          * step 7: we are done with the source map, unlock.   if copy_ok
 2371          * is 0 then we have not replaced the dummy mapping in dstmap yet
 2372          * and we need to do so now.
 2373          */
 2374 
 2375         vm_map_unlock(srcmap);
 2376         if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
 2377                 uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
 2378 
 2379         /* now do the replacement if we didn't do it in step 5 */
 2380         if (copy_ok == 0) {
 2381                 vm_map_lock(dstmap);
 2382                 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
 2383                     nchain);
 2384                 vm_map_unlock(dstmap);
 2385 
 2386                 if (error == FALSE) {
 2387                         error = EIO;
 2388                         goto bad2;
 2389                 }
 2390         }
 2391 
 2392         uvm_tree_sanity(srcmap, "map_extract src leave");
 2393         uvm_tree_sanity(dstmap, "map_extract dst leave");
 2394 
 2395         return (0);
 2396 
 2397         /*
 2398          * bad: failure recovery
 2399          */
 2400 bad:
 2401         vm_map_unlock(srcmap);
 2402 bad2:                   /* src already unlocked */
 2403         if (chain)
 2404                 uvm_unmap_detach(chain,
 2405                     (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
 2406 
 2407         uvm_tree_sanity(srcmap, "map_extract src err leave");
 2408         uvm_tree_sanity(dstmap, "map_extract dst err leave");
 2409 
 2410         uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
 2411         return (error);
 2412 }
 2413 
 2414 /* end of extraction functions */
 2415 
 2416 /*
 2417  * uvm_map_submap: punch down part of a map into a submap
 2418  *
 2419  * => only the kernel_map is allowed to be submapped
 2420  * => the purpose of submapping is to break up the locking granularity
 2421  *      of a larger map
 2422  * => the range specified must have been mapped previously with a uvm_map()
 2423  *      call [with uobj==NULL] to create a blank map entry in the main map.
 2424  *      [And it had better still be blank!]
 2425  * => maps which contain submaps should never be copied or forked.
 2426  * => to remove a submap, use uvm_unmap() on the main map
 2427  *      and then uvm_map_deallocate() the submap.
 2428  * => main map must be unlocked.
 2429  * => submap must have been init'd and have a zero reference count.
 2430  *      [need not be locked as we don't actually reference it]
 2431  */
 2432 
 2433 int
 2434 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end,
 2435     struct vm_map *submap)
 2436 {
 2437         struct vm_map_entry *entry;
 2438         int error;
 2439 
 2440         vm_map_lock(map);
 2441         VM_MAP_RANGE_CHECK(map, start, end);
 2442 
 2443         if (uvm_map_lookup_entry(map, start, &entry)) {
 2444                 UVM_MAP_CLIP_START(map, entry, start);
 2445                 UVM_MAP_CLIP_END(map, entry, end);              /* to be safe */
 2446         } else {
 2447                 entry = NULL;
 2448         }
 2449 
 2450         if (entry != NULL &&
 2451             entry->start == start && entry->end == end &&
 2452             entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
 2453             !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
 2454                 entry->etype |= UVM_ET_SUBMAP;
 2455                 entry->object.sub_map = submap;
 2456                 entry->offset = 0;
 2457                 uvm_map_reference(submap);
 2458                 error = 0;
 2459         } else {
 2460                 error = EINVAL;
 2461         }
 2462         vm_map_unlock(map);
 2463         return error;
 2464 }
 2465 
 2466 
 2467 /*
 2468  * uvm_map_protect: change map protection
 2469  *
 2470  * => set_max means set max_protection.
 2471  * => map must be unlocked.
 2472  */
 2473 
 2474 #define MASK(entry)     (UVM_ET_ISCOPYONWRITE(entry) ? \
 2475                          ~VM_PROT_WRITE : VM_PROT_ALL)
 2476 
 2477 int
 2478 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end,
 2479     vm_prot_t new_prot, boolean_t set_max)
 2480 {
 2481         struct vm_map_entry *current, *entry;
 2482         int error = 0;
 2483         UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
 2484         UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
 2485                     map, start, end, new_prot);
 2486 
 2487         vm_map_lock(map);
 2488         VM_MAP_RANGE_CHECK(map, start, end);
 2489         if (uvm_map_lookup_entry(map, start, &entry)) {
 2490                 UVM_MAP_CLIP_START(map, entry, start);
 2491         } else {
 2492                 entry = entry->next;
 2493         }
 2494 
 2495         /*
 2496          * make a first pass to check for protection violations.
 2497          */
 2498 
 2499         current = entry;
 2500         while ((current != &map->header) && (current->start < end)) {
 2501                 if (UVM_ET_ISSUBMAP(current)) {
 2502                         error = EINVAL;
 2503                         goto out;
 2504                 }
 2505                 if ((new_prot & current->max_protection) != new_prot) {
 2506                         error = EACCES;
 2507                         goto out;
 2508                 }
 2509                 /*
 2510                  * Don't allow VM_PROT_EXECUTE to be set on entries that
 2511                  * point to vnodes that are associated with a NOEXEC file
 2512                  * system.
 2513                  */
 2514                 if (UVM_ET_ISOBJ(current) &&
 2515                     UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
 2516                         struct vnode *vp =
 2517                             (struct vnode *) current->object.uvm_obj;
 2518 
 2519                         if ((new_prot & VM_PROT_EXECUTE) != 0 &&
 2520                             (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
 2521                                 error = EACCES;
 2522                                 goto out;
 2523                         }
 2524                 }
 2525                 current = current->next;
 2526         }
 2527 
 2528         /* go back and fix up protections (no need to clip this time). */
 2529 
 2530         current = entry;
 2531         while ((current != &map->header) && (current->start < end)) {
 2532                 vm_prot_t old_prot;
 2533 
 2534                 UVM_MAP_CLIP_END(map, current, end);
 2535                 old_prot = current->protection;
 2536                 if (set_max)
 2537                         current->protection =
 2538                             (current->max_protection = new_prot) & old_prot;
 2539                 else
 2540                         current->protection = new_prot;
 2541 
 2542                 /*
 2543                  * update physical map if necessary.  worry about copy-on-write
 2544                  * here -- CHECK THIS XXX
 2545                  */
 2546 
 2547                 if (current->protection != old_prot) {
 2548                         /* update pmap! */
 2549                         pmap_protect(map->pmap, current->start, current->end,
 2550                             current->protection & MASK(entry));
 2551 
 2552                         /*
 2553                          * If this entry points at a vnode, and the
 2554                          * protection includes VM_PROT_EXECUTE, mark
 2555                          * the vnode as VEXECMAP.
 2556                          */
 2557                         if (UVM_ET_ISOBJ(current)) {
 2558                                 struct uvm_object *uobj =
 2559                                     current->object.uvm_obj;
 2560 
 2561                                 if (UVM_OBJ_IS_VNODE(uobj) &&
 2562                                     (current->protection & VM_PROT_EXECUTE))
 2563                                         vn_markexec((struct vnode *) uobj);
 2564                         }
 2565                 }
 2566 
 2567                 /*
 2568                  * If the map is configured to lock any future mappings,
 2569                  * wire this entry now if the old protection was VM_PROT_NONE
 2570                  * and the new protection is not VM_PROT_NONE.
 2571                  */
 2572 
 2573                 if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
 2574                     VM_MAPENT_ISWIRED(entry) == 0 &&
 2575                     old_prot == VM_PROT_NONE &&
 2576                     new_prot != VM_PROT_NONE) {
 2577                         if (uvm_map_pageable(map, entry->start,
 2578                             entry->end, FALSE,
 2579                             UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
 2580 
 2581                                 /*
 2582                                  * If locking the entry fails, remember the
 2583                                  * error if it's the first one.  Note we
 2584                                  * still continue setting the protection in
 2585                                  * the map, but will return the error
 2586                                  * condition regardless.
 2587                                  *
 2588                                  * XXX Ignore what the actual error is,
 2589                                  * XXX just call it a resource shortage
 2590                                  * XXX so that it doesn't get confused
 2591                                  * XXX what uvm_map_protect() itself would
 2592                                  * XXX normally return.
 2593                                  */
 2594 
 2595                                 error = ENOMEM;
 2596                         }
 2597                 }
 2598                 current = current->next;
 2599         }
 2600         pmap_update(map->pmap);
 2601 
 2602  out:
 2603         vm_map_unlock(map);
 2604         UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
 2605         return error;
 2606 }
 2607 
 2608 #undef  MASK
 2609 
 2610 /*
 2611  * uvm_map_inherit: set inheritance code for range of addrs in map.
 2612  *
 2613  * => map must be unlocked
 2614  * => note that the inherit code is used during a "fork".  see fork
 2615  *      code for details.
 2616  */
 2617 
 2618 int
 2619 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end,
 2620     vm_inherit_t new_inheritance)
 2621 {
 2622         struct vm_map_entry *entry, *temp_entry;
 2623         UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
 2624         UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
 2625             map, start, end, new_inheritance);
 2626 
 2627         switch (new_inheritance) {
 2628         case MAP_INHERIT_NONE:
 2629         case MAP_INHERIT_COPY:
 2630         case MAP_INHERIT_SHARE:
 2631                 break;
 2632         default:
 2633                 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
 2634                 return EINVAL;
 2635         }
 2636 
 2637         vm_map_lock(map);
 2638         VM_MAP_RANGE_CHECK(map, start, end);
 2639         if (uvm_map_lookup_entry(map, start, &temp_entry)) {
 2640                 entry = temp_entry;
 2641                 UVM_MAP_CLIP_START(map, entry, start);
 2642         }  else {
 2643                 entry = temp_entry->next;
 2644         }
 2645         while ((entry != &map->header) && (entry->start < end)) {
 2646                 UVM_MAP_CLIP_END(map, entry, end);
 2647                 entry->inheritance = new_inheritance;
 2648                 entry = entry->next;
 2649         }
 2650         vm_map_unlock(map);
 2651         UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
 2652         return 0;
 2653 }
 2654 
 2655 /*
 2656  * uvm_map_advice: set advice code for range of addrs in map.
 2657  *
 2658  * => map must be unlocked
 2659  */
 2660 
 2661 int
 2662 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice)
 2663 {
 2664         struct vm_map_entry *entry, *temp_entry;
 2665         UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
 2666         UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
 2667             map, start, end, new_advice);
 2668 
 2669         vm_map_lock(map);
 2670         VM_MAP_RANGE_CHECK(map, start, end);
 2671         if (uvm_map_lookup_entry(map, start, &temp_entry)) {
 2672                 entry = temp_entry;
 2673                 UVM_MAP_CLIP_START(map, entry, start);
 2674         } else {
 2675                 entry = temp_entry->next;
 2676         }
 2677 
 2678         /*
 2679          * XXXJRT: disallow holes?
 2680          */
 2681 
 2682         while ((entry != &map->header) && (entry->start < end)) {
 2683                 UVM_MAP_CLIP_END(map, entry, end);
 2684 
 2685                 switch (new_advice) {
 2686                 case MADV_NORMAL:
 2687                 case MADV_RANDOM:
 2688                 case MADV_SEQUENTIAL:
 2689                         /* nothing special here */
 2690                         break;
 2691 
 2692                 default:
 2693                         vm_map_unlock(map);
 2694                         UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
 2695                         return EINVAL;
 2696                 }
 2697                 entry->advice = new_advice;
 2698                 entry = entry->next;
 2699         }
 2700 
 2701         vm_map_unlock(map);
 2702         UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
 2703         return 0;
 2704 }
 2705 
 2706 /*
 2707  * uvm_map_pageable: sets the pageability of a range in a map.
 2708  *
 2709  * => wires map entries.  should not be used for transient page locking.
 2710  *      for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
 2711  * => regions sepcified as not pageable require lock-down (wired) memory
 2712  *      and page tables.
 2713  * => map must never be read-locked
 2714  * => if islocked is TRUE, map is already write-locked
 2715  * => we always unlock the map, since we must downgrade to a read-lock
 2716  *      to call uvm_fault_wire()
 2717  * => XXXCDC: check this and try and clean it up.
 2718  */
 2719 
 2720 int
 2721 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end,
 2722     boolean_t new_pageable, int lockflags)
 2723 {
 2724         struct vm_map_entry *entry, *start_entry, *failed_entry;
 2725         int rv;
 2726 #ifdef DIAGNOSTIC
 2727         u_int timestamp_save;
 2728 #endif
 2729         UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
 2730         UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
 2731                     map, start, end, new_pageable);
 2732         KASSERT(map->flags & VM_MAP_PAGEABLE);
 2733 
 2734         if ((lockflags & UVM_LK_ENTER) == 0)
 2735                 vm_map_lock(map);
 2736         VM_MAP_RANGE_CHECK(map, start, end);
 2737 
 2738         /*
 2739          * only one pageability change may take place at one time, since
 2740          * uvm_fault_wire assumes it will be called only once for each
 2741          * wiring/unwiring.  therefore, we have to make sure we're actually
 2742          * changing the pageability for the entire region.  we do so before
 2743          * making any changes.
 2744          */
 2745 
 2746         if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) {
 2747                 if ((lockflags & UVM_LK_EXIT) == 0)
 2748                         vm_map_unlock(map);
 2749 
 2750                 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
 2751                 return EFAULT;
 2752         }
 2753         entry = start_entry;
 2754 
 2755         /*
 2756          * handle wiring and unwiring separately.
 2757          */
 2758 
 2759         if (new_pageable) {             /* unwire */
 2760                 UVM_MAP_CLIP_START(map, entry, start);
 2761 
 2762                 /*
 2763                  * unwiring.  first ensure that the range to be unwired is
 2764                  * really wired down and that there are no holes.
 2765                  */
 2766 
 2767                 while ((entry != &map->header) && (entry->start < end)) {
 2768                         if (entry->wired_count == 0 ||
 2769                             (entry->end < end &&
 2770                              (entry->next == &map->header ||
 2771                               entry->next->start > entry->end))) {
 2772                                 if ((lockflags & UVM_LK_EXIT) == 0)
 2773                                         vm_map_unlock(map);
 2774                                 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
 2775                                 return EINVAL;
 2776                         }
 2777                         entry = entry->next;
 2778                 }
 2779 
 2780                 /*
 2781                  * POSIX 1003.1b - a single munlock call unlocks a region,
 2782                  * regardless of the number of mlock calls made on that
 2783                  * region.
 2784                  */
 2785 
 2786                 entry = start_entry;
 2787                 while ((entry != &map->header) && (entry->start < end)) {
 2788                         UVM_MAP_CLIP_END(map, entry, end);
 2789                         if (VM_MAPENT_ISWIRED(entry))
 2790                                 uvm_map_entry_unwire(map, entry);
 2791                         entry = entry->next;
 2792                 }
 2793                 if ((lockflags & UVM_LK_EXIT) == 0)
 2794                         vm_map_unlock(map);
 2795                 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
 2796                 return 0;
 2797         }
 2798 
 2799         /*
 2800          * wire case: in two passes [XXXCDC: ugly block of code here]
 2801          *
 2802          * 1: holding the write lock, we create any anonymous maps that need
 2803          *    to be created.  then we clip each map entry to the region to
 2804          *    be wired and increment its wiring count.
 2805          *
 2806          * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
 2807          *    in the pages for any newly wired area (wired_count == 1).
 2808          *
 2809          *    downgrading to a read lock for uvm_fault_wire avoids a possible
 2810          *    deadlock with another thread that may have faulted on one of
 2811          *    the pages to be wired (it would mark the page busy, blocking
 2812          *    us, then in turn block on the map lock that we hold).  because
 2813          *    of problems in the recursive lock package, we cannot upgrade
 2814          *    to a write lock in vm_map_lookup.  thus, any actions that
 2815          *    require the write lock must be done beforehand.  because we
 2816          *    keep the read lock on the map, the copy-on-write status of the
 2817          *    entries we modify here cannot change.
 2818          */
 2819 
 2820         while ((entry != &map->header) && (entry->start < end)) {
 2821                 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
 2822 
 2823                         /*
 2824                          * perform actions of vm_map_lookup that need the
 2825                          * write lock on the map: create an anonymous map
 2826                          * for a copy-on-write region, or an anonymous map
 2827                          * for a zero-fill region.  (XXXCDC: submap case
 2828                          * ok?)
 2829                          */
 2830 
 2831                         if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
 2832                                 if (UVM_ET_ISNEEDSCOPY(entry) &&
 2833                                     ((entry->max_protection & VM_PROT_WRITE) ||
 2834                                      (entry->object.uvm_obj == NULL))) {
 2835                                         amap_copy(map, entry, M_WAITOK, TRUE,
 2836                                             start, end);
 2837                                         /* XXXCDC: wait OK? */
 2838                                 }
 2839                         }
 2840                 }
 2841                 UVM_MAP_CLIP_START(map, entry, start);
 2842                 UVM_MAP_CLIP_END(map, entry, end);
 2843                 entry->wired_count++;
 2844 
 2845                 /*
 2846                  * Check for holes
 2847                  */
 2848 
 2849                 if (entry->protection == VM_PROT_NONE ||
 2850                     (entry->end < end &&
 2851                      (entry->next == &map->header ||
 2852                       entry->next->start > entry->end))) {
 2853 
 2854                         /*
 2855                          * found one.  amap creation actions do not need to
 2856                          * be undone, but the wired counts need to be restored.
 2857                          */
 2858 
 2859                         while (entry != &map->header && entry->end > start) {
 2860                                 entry->wired_count--;
 2861                                 entry = entry->prev;
 2862                         }
 2863                         if ((lockflags & UVM_LK_EXIT) == 0)
 2864                                 vm_map_unlock(map);
 2865                         UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
 2866                         return EINVAL;
 2867                 }
 2868                 entry = entry->next;
 2869         }
 2870 
 2871         /*
 2872          * Pass 2.
 2873          */
 2874 
 2875 #ifdef DIAGNOSTIC
 2876         timestamp_save = map->timestamp;
 2877 #endif
 2878         vm_map_busy(map);
 2879         vm_map_downgrade(map);
 2880 
 2881         rv = 0;
 2882         entry = start_entry;
 2883         while (entry != &map->header && entry->start < end) {
 2884                 if (entry->wired_count == 1) {
 2885                         rv = uvm_fault_wire(map, entry->start, entry->end,
 2886                             VM_FAULT_WIREMAX, entry->max_protection);
 2887                         if (rv) {
 2888 
 2889                                 /*
 2890                                  * wiring failed.  break out of the loop.
 2891                                  * we'll clean up the map below, once we
 2892                                  * have a write lock again.
 2893                                  */
 2894 
 2895                                 break;
 2896                         }
 2897                 }
 2898                 entry = entry->next;
 2899         }
 2900 
 2901         if (rv) {       /* failed? */
 2902 
 2903                 /*
 2904                  * Get back to an exclusive (write) lock.
 2905                  */
 2906 
 2907                 vm_map_upgrade(map);
 2908                 vm_map_unbusy(map);
 2909 
 2910 #ifdef DIAGNOSTIC
 2911                 if (timestamp_save != map->timestamp)
 2912                         panic("uvm_map_pageable: stale map");
 2913 #endif
 2914 
 2915                 /*
 2916                  * first drop the wiring count on all the entries
 2917                  * which haven't actually been wired yet.
 2918                  */
 2919 
 2920                 failed_entry = entry;
 2921                 while (entry != &map->header && entry->start < end) {
 2922                         entry->wired_count--;
 2923                         entry = entry->next;
 2924                 }
 2925 
 2926                 /*
 2927                  * now, unwire all the entries that were successfully
 2928                  * wired above.
 2929                  */
 2930 
 2931                 entry = start_entry;
 2932                 while (entry != failed_entry) {
 2933                         entry->wired_count--;
 2934                         if (VM_MAPENT_ISWIRED(entry) == 0)
 2935                                 uvm_map_entry_unwire(map, entry);
 2936                         entry = entry->next;
 2937                 }
 2938                 if ((lockflags & UVM_LK_EXIT) == 0)
 2939                         vm_map_unlock(map);
 2940                 UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
 2941                 return (rv);
 2942         }
 2943 
 2944         /* We are holding a read lock here. */
 2945         if ((lockflags & UVM_LK_EXIT) == 0) {
 2946                 vm_map_unbusy(map);
 2947                 vm_map_unlock_read(map);
 2948         } else {
 2949 
 2950                 /*
 2951                  * Get back to an exclusive (write) lock.
 2952                  */
 2953 
 2954                 vm_map_upgrade(map);
 2955                 vm_map_unbusy(map);
 2956         }
 2957 
 2958         UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
 2959         return 0;
 2960 }
 2961 
 2962 /*
 2963  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
 2964  * all mapped regions.
 2965  *
 2966  * => map must not be locked.
 2967  * => if no flags are specified, all regions are unwired.
 2968  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
 2969  */
 2970 
 2971 int
 2972 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit)
 2973 {
 2974         struct vm_map_entry *entry, *failed_entry;
 2975         vsize_t size;
 2976         int rv;
 2977 #ifdef DIAGNOSTIC
 2978         u_int timestamp_save;
 2979 #endif
 2980         UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
 2981         UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
 2982 
 2983         KASSERT(map->flags & VM_MAP_PAGEABLE);
 2984 
 2985         vm_map_lock(map);
 2986 
 2987         /*
 2988          * handle wiring and unwiring separately.
 2989          */
 2990 
 2991         if (flags == 0) {                       /* unwire */
 2992 
 2993                 /*
 2994                  * POSIX 1003.1b -- munlockall unlocks all regions,
 2995                  * regardless of how many times mlockall has been called.
 2996                  */
 2997 
 2998                 for (entry = map->header.next; entry != &map->header;
 2999                      entry = entry->next) {
 3000                         if (VM_MAPENT_ISWIRED(entry))
 3001                                 uvm_map_entry_unwire(map, entry);
 3002                 }
 3003                 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
 3004                 vm_map_unlock(map);
 3005                 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
 3006                 return 0;
 3007         }
 3008 
 3009         if (flags & MCL_FUTURE) {
 3010 
 3011                 /*
 3012                  * must wire all future mappings; remember this.
 3013                  */
 3014 
 3015                 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0);
 3016         }
 3017 
 3018         if ((flags & MCL_CURRENT) == 0) {
 3019 
 3020                 /*
 3021                  * no more work to do!
 3022                  */
 3023 
 3024                 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
 3025                 vm_map_unlock(map);
 3026                 return 0;
 3027         }
 3028 
 3029         /*
 3030          * wire case: in three passes [XXXCDC: ugly block of code here]
 3031          *
 3032          * 1: holding the write lock, count all pages mapped by non-wired
 3033          *    entries.  if this would cause us to go over our limit, we fail.
 3034          *
 3035          * 2: still holding the write lock, we create any anonymous maps that
 3036          *    need to be created.  then we increment its wiring count.
 3037          *
 3038          * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
 3039          *    in the pages for any newly wired area (wired_count == 1).
 3040          *
 3041          *    downgrading to a read lock for uvm_fault_wire avoids a possible
 3042          *    deadlock with another thread that may have faulted on one of
 3043          *    the pages to be wired (it would mark the page busy, blocking
 3044          *    us, then in turn block on the map lock that we hold).  because
 3045          *    of problems in the recursive lock package, we cannot upgrade
 3046          *    to a write lock in vm_map_lookup.  thus, any actions that
 3047          *    require the write lock must be done beforehand.  because we
 3048          *    keep the read lock on the map, the copy-on-write status of the
 3049          *    entries we modify here cannot change.
 3050          */
 3051 
 3052         for (size = 0, entry = map->header.next; entry != &map->header;
 3053              entry = entry->next) {
 3054                 if (entry->protection != VM_PROT_NONE &&
 3055                     VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
 3056                         size += entry->end - entry->start;
 3057                 }
 3058         }
 3059 
 3060         if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
 3061                 vm_map_unlock(map);
 3062                 return ENOMEM;
 3063         }
 3064 
 3065         /* XXX non-pmap_wired_count case must be handled by caller */
 3066 #ifdef pmap_wired_count
 3067         if (limit != 0 &&
 3068             (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
 3069                 vm_map_unlock(map);
 3070                 return ENOMEM;
 3071         }
 3072 #endif
 3073 
 3074         /*
 3075          * Pass 2.
 3076          */
 3077 
 3078         for (entry = map->header.next; entry != &map->header;
 3079              entry = entry->next) {
 3080                 if (entry->protection == VM_PROT_NONE)
 3081                         continue;
 3082                 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
 3083 
 3084                         /*
 3085                          * perform actions of vm_map_lookup that need the
 3086                          * write lock on the map: create an anonymous map
 3087                          * for a copy-on-write region, or an anonymous map
 3088                          * for a zero-fill region.  (XXXCDC: submap case
 3089                          * ok?)
 3090                          */
 3091 
 3092                         if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
 3093                                 if (UVM_ET_ISNEEDSCOPY(entry) &&
 3094                                     ((entry->max_protection & VM_PROT_WRITE) ||
 3095                                      (entry->object.uvm_obj == NULL))) {
 3096                                         amap_copy(map, entry, M_WAITOK, TRUE,
 3097                                             entry->start, entry->end);
 3098                                         /* XXXCDC: wait OK? */
 3099                                 }
 3100                         }
 3101                 }
 3102                 entry->wired_count++;
 3103         }
 3104 
 3105         /*
 3106          * Pass 3.
 3107          */
 3108 
 3109 #ifdef DIAGNOSTIC
 3110         timestamp_save = map->timestamp;
 3111 #endif
 3112         vm_map_busy(map);
 3113         vm_map_downgrade(map);
 3114 
 3115         rv = 0;
 3116         for (entry = map->header.next; entry != &map->header;
 3117              entry = entry->next) {
 3118                 if (entry->wired_count == 1) {
 3119                         rv = uvm_fault_wire(map, entry->start, entry->end,
 3120                             VM_FAULT_WIREMAX, entry->max_protection);
 3121                         if (rv) {
 3122 
 3123                                 /*
 3124                                  * wiring failed.  break out of the loop.
 3125                                  * we'll clean up the map below, once we
 3126                                  * have a write lock again.
 3127                                  */
 3128 
 3129                                 break;
 3130                         }
 3131                 }
 3132         }
 3133 
 3134         if (rv) {
 3135 
 3136                 /*
 3137                  * Get back an exclusive (write) lock.
 3138                  */
 3139 
 3140                 vm_map_upgrade(map);
 3141                 vm_map_unbusy(map);
 3142 
 3143 #ifdef DIAGNOSTIC
 3144                 if (timestamp_save != map->timestamp)
 3145                         panic("uvm_map_pageable_all: stale map");
 3146 #endif
 3147 
 3148                 /*
 3149                  * first drop the wiring count on all the entries
 3150                  * which haven't actually been wired yet.
 3151                  *
 3152                  * Skip VM_PROT_NONE entries like we did above.
 3153                  */
 3154 
 3155                 failed_entry = entry;
 3156                 for (/* nothing */; entry != &map->header;
 3157                      entry = entry->next) {
 3158                         if (entry->protection == VM_PROT_NONE)
 3159                                 continue;
 3160                         entry->wired_count--;
 3161                 }
 3162 
 3163                 /*
 3164                  * now, unwire all the entries that were successfully
 3165                  * wired above.
 3166                  *
 3167                  * Skip VM_PROT_NONE entries like we did above.
 3168                  */
 3169 
 3170                 for (entry = map->header.next; entry != failed_entry;
 3171                      entry = entry->next) {
 3172                         if (entry->protection == VM_PROT_NONE)
 3173                                 continue;
 3174                         entry->wired_count--;
 3175                         if (VM_MAPENT_ISWIRED(entry))
 3176                                 uvm_map_entry_unwire(map, entry);
 3177                 }
 3178                 vm_map_unlock(map);
 3179                 UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
 3180                 return (rv);
 3181         }
 3182 
 3183         /* We are holding a read lock here. */
 3184         vm_map_unbusy(map);
 3185         vm_map_unlock_read(map);
 3186 
 3187         UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
 3188         return 0;
 3189 }
 3190 
 3191 /*
 3192  * uvm_map_clean: clean out a map range
 3193  *
 3194  * => valid flags:
 3195  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
 3196  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
 3197  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
 3198  *   if (flags & PGO_FREE): any cached pages are freed after clean
 3199  * => returns an error if any part of the specified range isn't mapped
 3200  * => never a need to flush amap layer since the anonymous memory has
 3201  *      no permanent home, but may deactivate pages there
 3202  * => called from sys_msync() and sys_madvise()
 3203  * => caller must not write-lock map (read OK).
 3204  * => we may sleep while cleaning if SYNCIO [with map read-locked]
 3205  */
 3206 
 3207 int
 3208 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
 3209 {
 3210         struct vm_map_entry *current, *entry;
 3211         struct uvm_object *uobj;
 3212         struct vm_amap *amap;
 3213         struct vm_anon *anon;
 3214         struct vm_page *pg;
 3215         vaddr_t offset;
 3216         vsize_t size;
 3217         int error, refs;
 3218         UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
 3219 
 3220         UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
 3221                     map, start, end, flags);
 3222         KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
 3223                 (PGO_FREE|PGO_DEACTIVATE));
 3224 
 3225         vm_map_lock_read(map);
 3226         VM_MAP_RANGE_CHECK(map, start, end);
 3227         if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
 3228                 vm_map_unlock_read(map);
 3229                 return EFAULT;
 3230         }
 3231 
 3232         /*
 3233          * Make a first pass to check for holes.
 3234          */
 3235 
 3236         for (current = entry; current->start < end; current = current->next) {
 3237                 if (UVM_ET_ISSUBMAP(current)) {
 3238                         vm_map_unlock_read(map);
 3239                         return EINVAL;
 3240                 }
 3241                 if (end <= current->end) {
 3242                         break;
 3243                 }
 3244                 if (current->end != current->next->start) {
 3245                         vm_map_unlock_read(map);
 3246                         return EFAULT;
 3247                 }
 3248         }
 3249 
 3250         error = 0;
 3251         for (current = entry; start < end; current = current->next) {
 3252                 amap = current->aref.ar_amap;   /* top layer */
 3253                 uobj = current->object.uvm_obj; /* bottom layer */
 3254                 KASSERT(start >= current->start);
 3255 
 3256                 /*
 3257                  * No amap cleaning necessary if:
 3258                  *
 3259                  *      (1) There's no amap.
 3260                  *
 3261                  *      (2) We're not deactivating or freeing pages.
 3262                  */
 3263 
 3264                 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
 3265                         goto flush_object;
 3266 
 3267                 amap_lock(amap);
 3268                 offset = start - current->start;
 3269                 size = MIN(end, current->end) - start;
 3270                 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
 3271                         anon = amap_lookup(&current->aref, offset);
 3272                         if (anon == NULL)
 3273                                 continue;
 3274 
 3275                         simple_lock(&anon->an_lock);
 3276                         pg = anon->u.an_page;
 3277                         if (pg == NULL) {
 3278                                 simple_unlock(&anon->an_lock);
 3279                                 continue;
 3280                         }
 3281 
 3282                         switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
 3283 
 3284                         /*
 3285                          * In these first 3 cases, we just deactivate the page.
 3286                          */
 3287 
 3288                         case PGO_CLEANIT|PGO_FREE:
 3289                         case PGO_CLEANIT|PGO_DEACTIVATE:
 3290                         case PGO_DEACTIVATE:
 3291  deactivate_it:
 3292                                 /*
 3293                                  * skip the page if it's loaned or wired,
 3294                                  * since it shouldn't be on a paging queue
 3295                                  * at all in these cases.
 3296                                  */
 3297 
 3298                                 uvm_lock_pageq();
 3299                                 if (pg->loan_count != 0 ||
 3300                                     pg->wire_count != 0) {
 3301                                         uvm_unlock_pageq();
 3302                                         simple_unlock(&anon->an_lock);
 3303                                         continue;
 3304                                 }
 3305                                 KASSERT(pg->uanon == anon);
 3306                                 pmap_clear_reference(pg);
 3307                                 uvm_pagedeactivate(pg);
 3308                                 uvm_unlock_pageq();
 3309                                 simple_unlock(&anon->an_lock);
 3310                                 continue;
 3311 
 3312                         case PGO_FREE:
 3313 
 3314                                 /*
 3315                                  * If there are multiple references to
 3316                                  * the amap, just deactivate the page.
 3317                                  */
 3318 
 3319                                 if (amap_refs(amap) > 1)
 3320                                         goto deactivate_it;
 3321 
 3322                                 /* skip the page if it's wired */
 3323                                 if (pg->wire_count != 0) {
 3324                                         simple_unlock(&anon->an_lock);
 3325                                         continue;
 3326                                 }
 3327                                 amap_unadd(&current->aref, offset);
 3328                                 refs = --anon->an_ref;
 3329                                 simple_unlock(&anon->an_lock);
 3330                                 if (refs == 0)
 3331                                         uvm_anfree(anon);
 3332                                 continue;
 3333                         }
 3334                 }
 3335                 amap_unlock(amap);
 3336 
 3337  flush_object:
 3338                 /*
 3339                  * flush pages if we've got a valid backing object.
 3340                  * note that we must always clean object pages before
 3341                  * freeing them since otherwise we could reveal stale
 3342                  * data from files.
 3343                  */
 3344 
 3345                 offset = current->offset + (start - current->start);
 3346                 size = MIN(end, current->end) - start;
 3347                 if (uobj != NULL) {
 3348                         simple_lock(&uobj->vmobjlock);
 3349                         if (uobj->pgops->pgo_put != NULL)
 3350                                 error = (uobj->pgops->pgo_put)(uobj, offset,
 3351                                     offset + size, flags | PGO_CLEANIT);
 3352                         else
 3353                                 error = 0;
 3354                 }
 3355                 start += size;
 3356         }
 3357         vm_map_unlock_read(map);
 3358         return (error);
 3359 }
 3360 
 3361 
 3362 /*
 3363  * uvm_map_checkprot: check protection in map
 3364  *
 3365  * => must allow specified protection in a fully allocated region.
 3366  * => map must be read or write locked by caller.
 3367  */
 3368 
 3369 boolean_t
 3370 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end,
 3371     vm_prot_t protection)
 3372 {
 3373         struct vm_map_entry *entry;
 3374         struct vm_map_entry *tmp_entry;
 3375 
 3376         if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
 3377                 return (FALSE);
 3378         }
 3379         entry = tmp_entry;
 3380         while (start < end) {
 3381                 if (entry == &map->header) {
 3382                         return (FALSE);
 3383                 }
 3384 
 3385                 /*
 3386                  * no holes allowed
 3387                  */
 3388 
 3389                 if (start < entry->start) {
 3390                         return (FALSE);
 3391                 }
 3392 
 3393                 /*
 3394                  * check protection associated with entry
 3395                  */
 3396 
 3397                 if ((entry->protection & protection) != protection) {
 3398                         return (FALSE);
 3399                 }
 3400                 start = entry->end;
 3401                 entry = entry->next;
 3402         }
 3403         return (TRUE);
 3404 }
 3405 
 3406 /*
 3407  * uvmspace_alloc: allocate a vmspace structure.
 3408  *
 3409  * - structure includes vm_map and pmap
 3410  * - XXX: no locking on this structure
 3411  * - refcnt set to 1, rest must be init'd by caller
 3412  */
 3413 struct vmspace *
 3414 uvmspace_alloc(vaddr_t min, vaddr_t max)
 3415 {
 3416         struct vmspace *vm;
 3417         UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
 3418 
 3419         vm = pool_get(&uvm_vmspace_pool, PR_WAITOK);
 3420         uvmspace_init(vm, NULL, min, max);
 3421         UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
 3422         return (vm);
 3423 }
 3424 
 3425 /*
 3426  * uvmspace_init: initialize a vmspace structure.
 3427  *
 3428  * - XXX: no locking on this structure
 3429  * - refcnt set to 1, rest must be init'd by caller
 3430  */
 3431 void
 3432 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t min, vaddr_t max)
 3433 {
 3434         UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
 3435 
 3436         memset(vm, 0, sizeof(*vm));
 3437         uvm_map_setup(&vm->vm_map, min, max, VM_MAP_PAGEABLE
 3438 #ifdef __USING_TOPDOWN_VM
 3439             | VM_MAP_TOPDOWN
 3440 #endif
 3441             );
 3442         if (pmap)
 3443                 pmap_reference(pmap);
 3444         else
 3445                 pmap = pmap_create();
 3446         vm->vm_map.pmap = pmap;
 3447         vm->vm_refcnt = 1;
 3448         UVMHIST_LOG(maphist,"<- done",0,0,0,0);
 3449 }
 3450 
 3451 /*
 3452  * uvmspace_share: share a vmspace between two proceses
 3453  *
 3454  * - XXX: no locking on vmspace
 3455  * - used for vfork, threads(?)
 3456  */
 3457 
 3458 void
 3459 uvmspace_share(struct proc *p1, struct proc *p2)
 3460 {
 3461 
 3462         p2->p_vmspace = p1->p_vmspace;
 3463         p1->p_vmspace->vm_refcnt++;
 3464 }
 3465 
 3466 /*
 3467  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
 3468  *
 3469  * - XXX: no locking on vmspace
 3470  */
 3471 
 3472 void
 3473 uvmspace_unshare(struct lwp *l)
 3474 {
 3475         struct proc *p = l->l_proc;
 3476         struct vmspace *nvm, *ovm = p->p_vmspace;
 3477 
 3478         if (ovm->vm_refcnt == 1)
 3479                 /* nothing to do: vmspace isn't shared in the first place */
 3480                 return;
 3481 
 3482         /* make a new vmspace, still holding old one */
 3483         nvm = uvmspace_fork(ovm);
 3484 
 3485         pmap_deactivate(l);             /* unbind old vmspace */
 3486         p->p_vmspace = nvm;
 3487         pmap_activate(l);               /* switch to new vmspace */
 3488 
 3489         uvmspace_free(ovm);             /* drop reference to old vmspace */
 3490 }
 3491 
 3492 /*
 3493  * uvmspace_exec: the process wants to exec a new program
 3494  *
 3495  * - XXX: no locking on vmspace
 3496  */
 3497 
 3498 void
 3499 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end)
 3500 {
 3501         struct proc *p = l->l_proc;
 3502         struct vmspace *nvm, *ovm = p->p_vmspace;
 3503         struct vm_map *map = &ovm->vm_map;
 3504 
 3505 #ifdef __sparc__
 3506         /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
 3507         kill_user_windows(l);   /* before stack addresses go away */
 3508 #endif
 3509 
 3510         /*
 3511          * see if more than one process is using this vmspace...
 3512          */
 3513 
 3514         if (ovm->vm_refcnt == 1) {
 3515 
 3516                 /*
 3517                  * if p is the only process using its vmspace then we can safely
 3518                  * recycle that vmspace for the program that is being exec'd.
 3519                  */
 3520 
 3521 #ifdef SYSVSHM
 3522                 /*
 3523                  * SYSV SHM semantics require us to kill all segments on an exec
 3524                  */
 3525 
 3526                 if (ovm->vm_shm)
 3527                         shmexit(ovm);
 3528 #endif
 3529 
 3530                 /*
 3531                  * POSIX 1003.1b -- "lock future mappings" is revoked
 3532                  * when a process execs another program image.
 3533                  */
 3534 
 3535                 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE);
 3536 
 3537                 /*
 3538                  * now unmap the old program
 3539                  */
 3540 
 3541                 pmap_remove_all(map->pmap);
 3542                 uvm_unmap(map, map->min_offset, map->max_offset);
 3543                 KASSERT(map->header.prev == &map->header);
 3544                 KASSERT(map->nentries == 0);
 3545 
 3546                 /*
 3547                  * resize the map
 3548                  */
 3549 
 3550                 map->min_offset = start;
 3551                 map->max_offset = end;
 3552         } else {
 3553 
 3554                 /*
 3555                  * p's vmspace is being shared, so we can't reuse it for p since
 3556                  * it is still being used for others.   allocate a new vmspace
 3557                  * for p
 3558                  */
 3559 
 3560                 nvm = uvmspace_alloc(start, end);
 3561 
 3562                 /*
 3563                  * install new vmspace and drop our ref to the old one.
 3564                  */
 3565 
 3566                 pmap_deactivate(l);
 3567                 p->p_vmspace = nvm;
 3568                 pmap_activate(l);
 3569 
 3570                 uvmspace_free(ovm);
 3571         }
 3572 }
 3573 
 3574 /*
 3575  * uvmspace_free: free a vmspace data structure
 3576  *
 3577  * - XXX: no locking on vmspace
 3578  */
 3579 
 3580 void
 3581 uvmspace_free(struct vmspace *vm)
 3582 {
 3583         struct vm_map_entry *dead_entries;
 3584         struct vm_map *map;
 3585         UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
 3586 
 3587         UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
 3588         if (--vm->vm_refcnt > 0) {
 3589                 return;
 3590         }
 3591 
 3592         /*
 3593          * at this point, there should be no other references to the map.
 3594          * delete all of the mappings, then destroy the pmap.
 3595          */
 3596 
 3597         map = &vm->vm_map;
 3598         map->flags |= VM_MAP_DYING;
 3599         pmap_remove_all(map->pmap);
 3600 #ifdef SYSVSHM
 3601         /* Get rid of any SYSV shared memory segments. */
 3602         if (vm->vm_shm != NULL)
 3603                 shmexit(vm);
 3604 #endif
 3605         if (map->nentries) {
 3606                 uvm_unmap_remove(map, map->min_offset, map->max_offset,
 3607                     &dead_entries);
 3608                 if (dead_entries != NULL)
 3609                         uvm_unmap_detach(dead_entries, 0);
 3610         }
 3611         KASSERT(map->nentries == 0);
 3612         KASSERT(map->size == 0);
 3613         pmap_destroy(map->pmap);
 3614         pool_put(&uvm_vmspace_pool, vm);
 3615 }
 3616 
 3617 /*
 3618  *   F O R K   -   m a i n   e n t r y   p o i n t
 3619  */
 3620 /*
 3621  * uvmspace_fork: fork a process' main map
 3622  *
 3623  * => create a new vmspace for child process from parent.
 3624  * => parent's map must not be locked.
 3625  */
 3626 
 3627 struct vmspace *
 3628 uvmspace_fork(struct vmspace *vm1)
 3629 {
 3630         struct vmspace *vm2;
 3631         struct vm_map *old_map = &vm1->vm_map;
 3632         struct vm_map *new_map;
 3633         struct vm_map_entry *old_entry;
 3634         struct vm_map_entry *new_entry;
 3635         UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
 3636 
 3637         vm_map_lock(old_map);
 3638 
 3639         vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset);
 3640         memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
 3641             (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
 3642         new_map = &vm2->vm_map;           /* XXX */
 3643 
 3644         old_entry = old_map->header.next;
 3645         new_map->size = old_map->size;
 3646 
 3647         /*
 3648          * go entry-by-entry
 3649          */
 3650 
 3651         while (old_entry != &old_map->header) {
 3652 
 3653                 /*
 3654                  * first, some sanity checks on the old entry
 3655                  */
 3656 
 3657                 KASSERT(!UVM_ET_ISSUBMAP(old_entry));
 3658                 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
 3659                         !UVM_ET_ISNEEDSCOPY(old_entry));
 3660 
 3661                 switch (old_entry->inheritance) {
 3662                 case MAP_INHERIT_NONE:
 3663 
 3664                         /*
 3665                          * drop the mapping, modify size
 3666                          */
 3667                         new_map->size -= old_entry->end - old_entry->start;
 3668                         break;
 3669 
 3670                 case MAP_INHERIT_SHARE:
 3671 
 3672                         /*
 3673                          * share the mapping: this means we want the old and
 3674                          * new entries to share amaps and backing objects.
 3675                          */
 3676                         /*
 3677                          * if the old_entry needs a new amap (due to prev fork)
 3678                          * then we need to allocate it now so that we have
 3679                          * something we own to share with the new_entry.   [in
 3680                          * other words, we need to clear needs_copy]
 3681                          */
 3682 
 3683                         if (UVM_ET_ISNEEDSCOPY(old_entry)) {
 3684                                 /* get our own amap, clears needs_copy */
 3685                                 amap_copy(old_map, old_entry, M_WAITOK, FALSE,
 3686                                     0, 0);
 3687                                 /* XXXCDC: WAITOK??? */
 3688                         }
 3689 
 3690                         new_entry = uvm_mapent_alloc(new_map, 0);
 3691                         /* old_entry -> new_entry */
 3692                         uvm_mapent_copy(old_entry, new_entry);
 3693 
 3694                         /* new pmap has nothing wired in it */
 3695                         new_entry->wired_count = 0;
 3696 
 3697                         /*
 3698                          * gain reference to object backing the map (can't
 3699                          * be a submap, already checked this case).
 3700                          */
 3701 
 3702                         if (new_entry->aref.ar_amap)
 3703                                 uvm_map_reference_amap(new_entry, AMAP_SHARED);
 3704 
 3705                         if (new_entry->object.uvm_obj &&
 3706                             new_entry->object.uvm_obj->pgops->pgo_reference)
 3707                                 new_entry->object.uvm_obj->
 3708                                     pgops->pgo_reference(
 3709                                         new_entry->object.uvm_obj);
 3710 
 3711                         /* insert entry at end of new_map's entry list */
 3712                         uvm_map_entry_link(new_map, new_map->header.prev,
 3713                             new_entry);
 3714 
 3715                         break;
 3716 
 3717                 case MAP_INHERIT_COPY:
 3718 
 3719                         /*
 3720                          * copy-on-write the mapping (using mmap's
 3721                          * MAP_PRIVATE semantics)
 3722                          *
 3723                          * allocate new_entry, adjust reference counts.
 3724                          * (note that new references are read-only).
 3725                          */
 3726 
 3727                         new_entry = uvm_mapent_alloc(new_map, 0);
 3728                         /* old_entry -> new_entry */
 3729                         uvm_mapent_copy(old_entry, new_entry);
 3730 
 3731                         if (new_entry->aref.ar_amap)
 3732                                 uvm_map_reference_amap(new_entry, 0);
 3733 
 3734                         if (new_entry->object.uvm_obj &&
 3735                             new_entry->object.uvm_obj->pgops->pgo_reference)
 3736                                 new_entry->object.uvm_obj->pgops->pgo_reference
 3737                                     (new_entry->object.uvm_obj);
 3738 
 3739                         /* new pmap has nothing wired in it */
 3740                         new_entry->wired_count = 0;
 3741 
 3742                         new_entry->etype |=
 3743                             (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
 3744                         uvm_map_entry_link(new_map, new_map->header.prev,
 3745                             new_entry);
 3746 
 3747                         /*
 3748                          * the new entry will need an amap.  it will either
 3749                          * need to be copied from the old entry or created
 3750                          * from scratch (if the old entry does not have an
 3751                          * amap).  can we defer this process until later
 3752                          * (by setting "needs_copy") or do we need to copy
 3753                          * the amap now?
 3754                          *
 3755                          * we must copy the amap now if any of the following
 3756                          * conditions hold:
 3757                          * 1. the old entry has an amap and that amap is
 3758                          *    being shared.  this means that the old (parent)
 3759                          *    process is sharing the amap with another
 3760                          *    process.  if we do not clear needs_copy here
 3761                          *    we will end up in a situation where both the
 3762                          *    parent and child process are refering to the
 3763                          *    same amap with "needs_copy" set.  if the
 3764                          *    parent write-faults, the fault routine will
 3765                          *    clear "needs_copy" in the parent by allocating
 3766                          *    a new amap.   this is wrong because the
 3767                          *    parent is supposed to be sharing the old amap
 3768                          *    and the new amap will break that.
 3769                          *
 3770                          * 2. if the old entry has an amap and a non-zero
 3771                          *    wire count then we are going to have to call
 3772                          *    amap_cow_now to avoid page faults in the
 3773                          *    parent process.   since amap_cow_now requires
 3774                          *    "needs_copy" to be clear we might as well
 3775                          *    clear it here as well.
 3776                          *
 3777                          */
 3778 
 3779                         if (old_entry->aref.ar_amap != NULL) {
 3780                                 if ((amap_flags(old_entry->aref.ar_amap) &
 3781                                      AMAP_SHARED) != 0 ||
 3782                                     VM_MAPENT_ISWIRED(old_entry)) {
 3783 
 3784                                         amap_copy(new_map, new_entry, M_WAITOK,
 3785                                             FALSE, 0, 0);
 3786                                         /* XXXCDC: M_WAITOK ... ok? */
 3787                                 }
 3788                         }
 3789 
 3790                         /*
 3791                          * if the parent's entry is wired down, then the
 3792                          * parent process does not want page faults on
 3793                          * access to that memory.  this means that we
 3794                          * cannot do copy-on-write because we can't write
 3795                          * protect the old entry.   in this case we
 3796                          * resolve all copy-on-write faults now, using
 3797                          * amap_cow_now.   note that we have already
 3798                          * allocated any needed amap (above).
 3799                          */
 3800 
 3801                         if (VM_MAPENT_ISWIRED(old_entry)) {
 3802 
 3803                           /*
 3804                            * resolve all copy-on-write faults now
 3805                            * (note that there is nothing to do if
 3806                            * the old mapping does not have an amap).
 3807                            */
 3808                           if (old_entry->aref.ar_amap)
 3809                             amap_cow_now(new_map, new_entry);
 3810 
 3811                         } else {
 3812 
 3813                           /*
 3814                            * setup mappings to trigger copy-on-write faults
 3815                            * we must write-protect the parent if it has
 3816                            * an amap and it is not already "needs_copy"...
 3817                            * if it is already "needs_copy" then the parent
 3818                            * has already been write-protected by a previous
 3819                            * fork operation.
 3820                            */
 3821 
 3822                           if (old_entry->aref.ar_amap &&
 3823                               !UVM_ET_ISNEEDSCOPY(old_entry)) {
 3824                               if (old_entry->max_protection & VM_PROT_WRITE) {
 3825                                 pmap_protect(old_map->pmap,
 3826                                              old_entry->start,
 3827                                              old_entry->end,
 3828                                              old_entry->protection &
 3829                                              ~VM_PROT_WRITE);
 3830                                 pmap_update(old_map->pmap);
 3831                               }
 3832                               old_entry->etype |= UVM_ET_NEEDSCOPY;
 3833                           }
 3834                         }
 3835                         break;
 3836                 }  /* end of switch statement */
 3837                 old_entry = old_entry->next;
 3838         }
 3839 
 3840         vm_map_unlock(old_map);
 3841 
 3842 #ifdef SYSVSHM
 3843         if (vm1->vm_shm)
 3844                 shmfork(vm1, vm2);
 3845 #endif
 3846 
 3847 #ifdef PMAP_FORK
 3848         pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
 3849 #endif
 3850 
 3851         UVMHIST_LOG(maphist,"<- done",0,0,0,0);
 3852         return (vm2);
 3853 }
 3854 
 3855 
 3856 #if defined(DDB)
 3857 
 3858 /*
 3859  * DDB hooks
 3860  */
 3861 
 3862 /*
 3863  * uvm_map_printit: actually prints the map
 3864  */
 3865 
 3866 void
 3867 uvm_map_printit(struct vm_map *map, boolean_t full,
 3868     void (*pr)(const char *, ...))
 3869 {
 3870         struct vm_map_entry *entry;
 3871 
 3872         (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset);
 3873         (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
 3874             map->nentries, map->size, map->ref_count, map->timestamp,
 3875             map->flags);
 3876         (*pr)("\tpmap=%p(resident=%d)\n", map->pmap,
 3877             pmap_resident_count(map->pmap));
 3878         if (!full)
 3879                 return;
 3880         for (entry = map->header.next; entry != &map->header;
 3881             entry = entry->next) {
 3882                 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
 3883                     entry, entry->start, entry->end, entry->object.uvm_obj,
 3884                     (long long)entry->offset, entry->aref.ar_amap,
 3885                     entry->aref.ar_pageoff);
 3886                 (*pr)(
 3887                     "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
 3888                     "wc=%d, adv=%d\n",
 3889                     (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
 3890                     (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
 3891                     (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
 3892                     entry->protection, entry->max_protection,
 3893                     entry->inheritance, entry->wired_count, entry->advice);
 3894         }
 3895 }
 3896 
 3897 /*
 3898  * uvm_object_printit: actually prints the object
 3899  */
 3900 
 3901 void
 3902 uvm_object_printit(struct uvm_object *uobj, boolean_t full,
 3903     void (*pr)(const char *, ...))
 3904 {
 3905         struct vm_page *pg;
 3906         int cnt = 0;
 3907 
 3908         (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
 3909             uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages);
 3910         if (UVM_OBJ_IS_KERN_OBJECT(uobj))
 3911                 (*pr)("refs=<SYSTEM>\n");
 3912         else
 3913                 (*pr)("refs=%d\n", uobj->uo_refs);
 3914 
 3915         if (!full) {
 3916                 return;
 3917         }
 3918         (*pr)("  PAGES <pg,offset>:\n  ");
 3919         TAILQ_FOREACH(pg, &uobj->memq, listq) {
 3920                 cnt++;
 3921                 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
 3922                 if ((cnt % 3) == 0) {
 3923                         (*pr)("\n  ");
 3924                 }
 3925         }
 3926         if ((cnt % 3) != 0) {
 3927                 (*pr)("\n");
 3928         }
 3929 }
 3930 
 3931 /*
 3932  * uvm_page_printit: actually print the page
 3933  */
 3934 
 3935 static const char page_flagbits[] =
 3936         "\2\1BUSY\2WANTED\3TABLED\4CLEAN\5PAGEOUT\6RELEASED\7FAKE\10RDONLY"
 3937         "\11ZERO\15PAGER1";
 3938 static const char page_pqflagbits[] =
 3939         "\2\1FREE\2INACTIVE\3ACTIVE\5ANON\6AOBJ";
 3940 
 3941 void
 3942 uvm_page_printit(struct vm_page *pg, boolean_t full,
 3943     void (*pr)(const char *, ...))
 3944 {
 3945         struct vm_page *tpg;
 3946         struct uvm_object *uobj;
 3947         struct pglist *pgl;
 3948         char pgbuf[128];
 3949         char pqbuf[128];
 3950 
 3951         (*pr)("PAGE %p:\n", pg);
 3952         bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
 3953         bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
 3954         (*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
 3955             pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg));
 3956         (*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
 3957             pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
 3958 #if defined(UVM_PAGE_TRKOWN)
 3959         if (pg->flags & PG_BUSY)
 3960                 (*pr)("  owning process = %d, tag=%s\n",
 3961                     pg->owner, pg->owner_tag);
 3962         else
 3963                 (*pr)("  page not busy, no owner\n");
 3964 #else
 3965         (*pr)("  [page ownership tracking disabled]\n");
 3966 #endif
 3967 
 3968         if (!full)
 3969                 return;
 3970 
 3971         /* cross-verify object/anon */
 3972         if ((pg->pqflags & PQ_FREE) == 0) {
 3973                 if (pg->pqflags & PQ_ANON) {
 3974                         if (pg->uanon == NULL || pg->uanon->u.an_page != pg)
 3975                             (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
 3976                                 (pg->uanon) ? pg->uanon->u.an_page : NULL);
 3977                         else
 3978                                 (*pr)("  anon backpointer is OK\n");
 3979                 } else {
 3980                         uobj = pg->uobject;
 3981                         if (uobj) {
 3982                                 (*pr)("  checking object list\n");
 3983                                 TAILQ_FOREACH(tpg, &uobj->memq, listq) {
 3984                                         if (tpg == pg) {
 3985                                                 break;
 3986                                         }
 3987                                 }
 3988                                 if (tpg)
 3989                                         (*pr)("  page found on object list\n");
 3990                                 else
 3991                         (*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
 3992                         }
 3993                 }
 3994         }
 3995 
 3996         /* cross-verify page queue */
 3997         if (pg->pqflags & PQ_FREE) {
 3998                 int fl = uvm_page_lookup_freelist(pg);
 3999                 int color = VM_PGCOLOR_BUCKET(pg);
 4000                 pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
 4001                     ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
 4002         } else if (pg->pqflags & PQ_INACTIVE) {
 4003                 pgl = &uvm.page_inactive;
 4004         } else if (pg->pqflags & PQ_ACTIVE) {
 4005                 pgl = &uvm.page_active;
 4006         } else {
 4007                 pgl = NULL;
 4008         }
 4009 
 4010         if (pgl) {
 4011                 (*pr)("  checking pageq list\n");
 4012                 TAILQ_FOREACH(tpg, pgl, pageq) {
 4013                         if (tpg == pg) {
 4014                                 break;
 4015                         }
 4016                 }
 4017                 if (tpg)
 4018                         (*pr)("  page found on pageq list\n");
 4019                 else
 4020                         (*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
 4021         }
 4022 }
 4023 #endif

Cache object: 23ecec3fc1842cddae43416acdc05cfa


[ 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.