1df8bae1dSRodney W. Grimes /* 2df8bae1dSRodney W. Grimes * Copyright (c) 1991, 1993 3df8bae1dSRodney W. Grimes * The Regents of the University of California. All rights reserved. 426f9a767SRodney W. Grimes * Copyright (c) 1994 John S. Dyson 526f9a767SRodney W. Grimes * All rights reserved. 626f9a767SRodney W. Grimes * Copyright (c) 1994 David Greenman 726f9a767SRodney W. Grimes * All rights reserved. 826f9a767SRodney W. Grimes * 9df8bae1dSRodney W. Grimes * 10df8bae1dSRodney W. Grimes * This code is derived from software contributed to Berkeley by 11df8bae1dSRodney W. Grimes * The Mach Operating System project at Carnegie-Mellon University. 12df8bae1dSRodney W. Grimes * 13df8bae1dSRodney W. Grimes * Redistribution and use in source and binary forms, with or without 14df8bae1dSRodney W. Grimes * modification, are permitted provided that the following conditions 15df8bae1dSRodney W. Grimes * are met: 16df8bae1dSRodney W. Grimes * 1. Redistributions of source code must retain the above copyright 17df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer. 18df8bae1dSRodney W. Grimes * 2. Redistributions in binary form must reproduce the above copyright 19df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer in the 20df8bae1dSRodney W. Grimes * documentation and/or other materials provided with the distribution. 21df8bae1dSRodney W. Grimes * 3. All advertising materials mentioning features or use of this software 22df8bae1dSRodney W. Grimes * must display the following acknowledgement: 23df8bae1dSRodney W. Grimes * This product includes software developed by the University of 24df8bae1dSRodney W. Grimes * California, Berkeley and its contributors. 25df8bae1dSRodney W. Grimes * 4. Neither the name of the University nor the names of its contributors 26df8bae1dSRodney W. Grimes * may be used to endorse or promote products derived from this software 27df8bae1dSRodney W. Grimes * without specific prior written permission. 28df8bae1dSRodney W. Grimes * 29df8bae1dSRodney W. Grimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30df8bae1dSRodney W. Grimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31df8bae1dSRodney W. Grimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32df8bae1dSRodney W. Grimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33df8bae1dSRodney W. Grimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34df8bae1dSRodney W. Grimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35df8bae1dSRodney W. Grimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36df8bae1dSRodney W. Grimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37df8bae1dSRodney W. Grimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38df8bae1dSRodney W. Grimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39df8bae1dSRodney W. Grimes * SUCH DAMAGE. 40df8bae1dSRodney W. Grimes * 413c4dd356SDavid Greenman * from: @(#)vm_fault.c 8.4 (Berkeley) 1/12/94 42df8bae1dSRodney W. Grimes * 43df8bae1dSRodney W. Grimes * 44df8bae1dSRodney W. Grimes * Copyright (c) 1987, 1990 Carnegie-Mellon University. 45df8bae1dSRodney W. Grimes * All rights reserved. 46df8bae1dSRodney W. Grimes * 47df8bae1dSRodney W. Grimes * Authors: Avadis Tevanian, Jr., Michael Wayne Young 48df8bae1dSRodney W. Grimes * 49df8bae1dSRodney W. Grimes * Permission to use, copy, modify and distribute this software and 50df8bae1dSRodney W. Grimes * its documentation is hereby granted, provided that both the copyright 51df8bae1dSRodney W. Grimes * notice and this permission notice appear in all copies of the 52df8bae1dSRodney W. Grimes * software, derivative works or modified versions, and any portions 53df8bae1dSRodney W. Grimes * thereof, and that both notices appear in supporting documentation. 54df8bae1dSRodney W. Grimes * 55df8bae1dSRodney W. Grimes * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 56df8bae1dSRodney W. Grimes * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 57df8bae1dSRodney W. Grimes * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 58df8bae1dSRodney W. Grimes * 59df8bae1dSRodney W. Grimes * Carnegie Mellon requests users of this software to return to 60df8bae1dSRodney W. Grimes * 61df8bae1dSRodney W. Grimes * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 62df8bae1dSRodney W. Grimes * School of Computer Science 63df8bae1dSRodney W. Grimes * Carnegie Mellon University 64df8bae1dSRodney W. Grimes * Pittsburgh PA 15213-3890 65df8bae1dSRodney W. Grimes * 66df8bae1dSRodney W. Grimes * any improvements or extensions that they make and grant Carnegie the 67df8bae1dSRodney W. Grimes * rights to redistribute these changes. 683c4dd356SDavid Greenman * 69bd7e5f99SJohn Dyson * $Id: vm_fault.c,v 1.39 1995/12/11 04:58:06 dyson Exp $ 70df8bae1dSRodney W. Grimes */ 71df8bae1dSRodney W. Grimes 72df8bae1dSRodney W. Grimes /* 73df8bae1dSRodney W. Grimes * Page fault handling module. 74df8bae1dSRodney W. Grimes */ 75df8bae1dSRodney W. Grimes 76df8bae1dSRodney W. Grimes #include <sys/param.h> 77df8bae1dSRodney W. Grimes #include <sys/systm.h> 7826f9a767SRodney W. Grimes #include <sys/proc.h> 7924a1cce3SDavid Greenman #include <sys/vnode.h> 8005f0fdd2SPoul-Henning Kamp #include <sys/resource.h> 8105f0fdd2SPoul-Henning Kamp #include <sys/signalvar.h> 8226f9a767SRodney W. Grimes #include <sys/resourcevar.h> 83efeaf95aSDavid Greenman #include <sys/vmmeter.h> 84df8bae1dSRodney W. Grimes 85df8bae1dSRodney W. Grimes #include <vm/vm.h> 86efeaf95aSDavid Greenman #include <vm/vm_param.h> 87efeaf95aSDavid Greenman #include <vm/vm_prot.h> 88efeaf95aSDavid Greenman #include <vm/lock.h> 89efeaf95aSDavid Greenman #include <vm/pmap.h> 90efeaf95aSDavid Greenman #include <vm/vm_map.h> 91efeaf95aSDavid Greenman #include <vm/vm_object.h> 92df8bae1dSRodney W. Grimes #include <vm/vm_page.h> 93df8bae1dSRodney W. Grimes #include <vm/vm_pageout.h> 94a83c285cSDavid Greenman #include <vm/vm_kern.h> 9524a1cce3SDavid Greenman #include <vm/vm_pager.h> 9624a1cce3SDavid Greenman #include <vm/vnode_pager.h> 97cd41fc12SDavid Greenman #include <vm/swap_pager.h> 98efeaf95aSDavid Greenman #include <vm/vm_extern.h> 99df8bae1dSRodney W. Grimes 10022ba64e8SJohn Dyson int vm_fault_additional_pages __P((vm_page_t, int, int, vm_page_t *, int *)); 10126f9a767SRodney W. Grimes 10226f9a767SRodney W. Grimes #define VM_FAULT_READ_AHEAD 4 10326f9a767SRodney W. Grimes #define VM_FAULT_READ_BEHIND 3 10426f9a767SRodney W. Grimes #define VM_FAULT_READ (VM_FAULT_READ_AHEAD+VM_FAULT_READ_BEHIND+1) 10526f9a767SRodney W. Grimes 106df8bae1dSRodney W. Grimes /* 107df8bae1dSRodney W. Grimes * vm_fault: 108df8bae1dSRodney W. Grimes * 109df8bae1dSRodney W. Grimes * Handle a page fault occuring at the given address, 110df8bae1dSRodney W. Grimes * requiring the given permissions, in the map specified. 111df8bae1dSRodney W. Grimes * If successful, the page is inserted into the 112df8bae1dSRodney W. Grimes * associated physical map. 113df8bae1dSRodney W. Grimes * 114df8bae1dSRodney W. Grimes * NOTE: the given address should be truncated to the 115df8bae1dSRodney W. Grimes * proper page address. 116df8bae1dSRodney W. Grimes * 117df8bae1dSRodney W. Grimes * KERN_SUCCESS is returned if the page fault is handled; otherwise, 118df8bae1dSRodney W. Grimes * a standard error specifying why the fault is fatal is returned. 119df8bae1dSRodney W. Grimes * 120df8bae1dSRodney W. Grimes * 121df8bae1dSRodney W. Grimes * The map in question must be referenced, and remains so. 122df8bae1dSRodney W. Grimes * Caller may hold no locks. 123df8bae1dSRodney W. Grimes */ 124df8bae1dSRodney W. Grimes int 125df8bae1dSRodney W. Grimes vm_fault(map, vaddr, fault_type, change_wiring) 126df8bae1dSRodney W. Grimes vm_map_t map; 127df8bae1dSRodney W. Grimes vm_offset_t vaddr; 128df8bae1dSRodney W. Grimes vm_prot_t fault_type; 129df8bae1dSRodney W. Grimes boolean_t change_wiring; 130df8bae1dSRodney W. Grimes { 131df8bae1dSRodney W. Grimes vm_object_t first_object; 132a316d390SJohn Dyson vm_pindex_t first_pindex; 133df8bae1dSRodney W. Grimes vm_map_entry_t entry; 134df8bae1dSRodney W. Grimes register vm_object_t object; 135a316d390SJohn Dyson register vm_pindex_t pindex; 13626f9a767SRodney W. Grimes vm_page_t m; 137df8bae1dSRodney W. Grimes vm_page_t first_m; 138df8bae1dSRodney W. Grimes vm_prot_t prot; 139df8bae1dSRodney W. Grimes int result; 140df8bae1dSRodney W. Grimes boolean_t wired; 141df8bae1dSRodney W. Grimes boolean_t su; 142df8bae1dSRodney W. Grimes boolean_t lookup_still_valid; 143df8bae1dSRodney W. Grimes vm_page_t old_m; 144df8bae1dSRodney W. Grimes vm_object_t next_object; 14526f9a767SRodney W. Grimes vm_page_t marray[VM_FAULT_READ]; 14626f9a767SRodney W. Grimes int hardfault = 0; 147f6b04d2bSDavid Greenman struct vnode *vp = NULL; 148df8bae1dSRodney W. Grimes 149b8d95f16SDavid Greenman cnt.v_vm_faults++; /* needs lock XXX */ 150df8bae1dSRodney W. Grimes /* 151df8bae1dSRodney W. Grimes * Recovery actions 152df8bae1dSRodney W. Grimes */ 153df8bae1dSRodney W. Grimes #define FREE_PAGE(m) { \ 154df8bae1dSRodney W. Grimes PAGE_WAKEUP(m); \ 155df8bae1dSRodney W. Grimes vm_page_free(m); \ 156df8bae1dSRodney W. Grimes } 157df8bae1dSRodney W. Grimes 158df8bae1dSRodney W. Grimes #define RELEASE_PAGE(m) { \ 159df8bae1dSRodney W. Grimes PAGE_WAKEUP(m); \ 160bd7e5f99SJohn Dyson if (m->queue != PQ_ACTIVE) vm_page_activate(m); \ 161df8bae1dSRodney W. Grimes } 162df8bae1dSRodney W. Grimes 163df8bae1dSRodney W. Grimes #define UNLOCK_MAP { \ 164df8bae1dSRodney W. Grimes if (lookup_still_valid) { \ 165df8bae1dSRodney W. Grimes vm_map_lookup_done(map, entry); \ 166df8bae1dSRodney W. Grimes lookup_still_valid = FALSE; \ 167df8bae1dSRodney W. Grimes } \ 168df8bae1dSRodney W. Grimes } 169df8bae1dSRodney W. Grimes 170df8bae1dSRodney W. Grimes #define UNLOCK_THINGS { \ 171f919ebdeSDavid Greenman vm_object_pip_wakeup(object); \ 172df8bae1dSRodney W. Grimes if (object != first_object) { \ 173df8bae1dSRodney W. Grimes FREE_PAGE(first_m); \ 174f919ebdeSDavid Greenman vm_object_pip_wakeup(first_object); \ 175df8bae1dSRodney W. Grimes } \ 176df8bae1dSRodney W. Grimes UNLOCK_MAP; \ 17724a1cce3SDavid Greenman if (vp != NULL) VOP_UNLOCK(vp); \ 178df8bae1dSRodney W. Grimes } 179df8bae1dSRodney W. Grimes 180df8bae1dSRodney W. Grimes #define UNLOCK_AND_DEALLOCATE { \ 181df8bae1dSRodney W. Grimes UNLOCK_THINGS; \ 182df8bae1dSRodney W. Grimes vm_object_deallocate(first_object); \ 183df8bae1dSRodney W. Grimes } 184df8bae1dSRodney W. Grimes 18526f9a767SRodney W. Grimes 186df8bae1dSRodney W. Grimes RetryFault:; 187df8bae1dSRodney W. Grimes 188df8bae1dSRodney W. Grimes /* 1890d94caffSDavid Greenman * Find the backing store object and offset into it to begin the 1900d94caffSDavid Greenman * search. 191df8bae1dSRodney W. Grimes */ 192df8bae1dSRodney W. Grimes 19322ba64e8SJohn Dyson if ((result = vm_map_lookup(&map, vaddr, 19422ba64e8SJohn Dyson fault_type, &entry, &first_object, 195a316d390SJohn Dyson &first_pindex, &prot, &wired, &su)) != KERN_SUCCESS) { 196df8bae1dSRodney W. Grimes return (result); 197df8bae1dSRodney W. Grimes } 198f6b04d2bSDavid Greenman 19924a1cce3SDavid Greenman vp = vnode_pager_lock(first_object); 200f6b04d2bSDavid Greenman 201df8bae1dSRodney W. Grimes lookup_still_valid = TRUE; 202df8bae1dSRodney W. Grimes 203df8bae1dSRodney W. Grimes if (wired) 204df8bae1dSRodney W. Grimes fault_type = prot; 205df8bae1dSRodney W. Grimes 206df8bae1dSRodney W. Grimes first_m = NULL; 207df8bae1dSRodney W. Grimes 208df8bae1dSRodney W. Grimes /* 2090d94caffSDavid Greenman * Make a reference to this object to prevent its disposal while we 2100d94caffSDavid Greenman * are messing with it. Once we have the reference, the map is free 2110d94caffSDavid Greenman * to be diddled. Since objects reference their shadows (and copies), 2120d94caffSDavid Greenman * they will stay around as well. 213df8bae1dSRodney W. Grimes */ 214df8bae1dSRodney W. Grimes 215df8bae1dSRodney W. Grimes first_object->ref_count++; 216df8bae1dSRodney W. Grimes first_object->paging_in_progress++; 217df8bae1dSRodney W. Grimes 218df8bae1dSRodney W. Grimes /* 219df8bae1dSRodney W. Grimes * INVARIANTS (through entire routine): 220df8bae1dSRodney W. Grimes * 2210d94caffSDavid Greenman * 1) At all times, we must either have the object lock or a busy 22224a1cce3SDavid Greenman * page in some object to prevent some other process from trying to 2230d94caffSDavid Greenman * bring in the same page. 224df8bae1dSRodney W. Grimes * 2250d94caffSDavid Greenman * Note that we cannot hold any locks during the pager access or when 2260d94caffSDavid Greenman * waiting for memory, so we use a busy page then. 227df8bae1dSRodney W. Grimes * 2280d94caffSDavid Greenman * Note also that we aren't as concerned about more than one thead 2290d94caffSDavid Greenman * attempting to pager_data_unlock the same page at once, so we don't 2300d94caffSDavid Greenman * hold the page as busy then, but do record the highest unlock value 2310d94caffSDavid Greenman * so far. [Unlock requests may also be delivered out of order.] 232df8bae1dSRodney W. Grimes * 2330d94caffSDavid Greenman * 2) Once we have a busy page, we must remove it from the pageout 2340d94caffSDavid Greenman * queues, so that the pageout daemon will not grab it away. 235df8bae1dSRodney W. Grimes * 23624a1cce3SDavid Greenman * 3) To prevent another process from racing us down the shadow chain 2370d94caffSDavid Greenman * and entering a new page in the top object before we do, we must 2380d94caffSDavid Greenman * keep a busy page in the top object while following the shadow 2390d94caffSDavid Greenman * chain. 240df8bae1dSRodney W. Grimes * 2410d94caffSDavid Greenman * 4) We must increment paging_in_progress on any object for which 2420d94caffSDavid Greenman * we have a busy page, to prevent vm_object_collapse from removing 2430d94caffSDavid Greenman * the busy page without our noticing. 244df8bae1dSRodney W. Grimes */ 245df8bae1dSRodney W. Grimes 246df8bae1dSRodney W. Grimes /* 247df8bae1dSRodney W. Grimes * Search for the page at object/offset. 248df8bae1dSRodney W. Grimes */ 249df8bae1dSRodney W. Grimes 250df8bae1dSRodney W. Grimes object = first_object; 251a316d390SJohn Dyson pindex = first_pindex; 252df8bae1dSRodney W. Grimes 253df8bae1dSRodney W. Grimes /* 254df8bae1dSRodney W. Grimes * See whether this page is resident 255df8bae1dSRodney W. Grimes */ 256df8bae1dSRodney W. Grimes 257df8bae1dSRodney W. Grimes while (TRUE) { 258a316d390SJohn Dyson m = vm_page_lookup(object, pindex); 259df8bae1dSRodney W. Grimes if (m != NULL) { 260df8bae1dSRodney W. Grimes /* 2610d94caffSDavid Greenman * If the page is being brought in, wait for it and 2620d94caffSDavid Greenman * then retry. 263df8bae1dSRodney W. Grimes */ 2640d94caffSDavid Greenman if ((m->flags & PG_BUSY) || m->busy) { 26516f62314SDavid Greenman int s; 2660d94caffSDavid Greenman 267df8bae1dSRodney W. Grimes UNLOCK_THINGS; 26816f62314SDavid Greenman s = splhigh(); 2690d94caffSDavid Greenman if ((m->flags & PG_BUSY) || m->busy) { 2700d94caffSDavid Greenman m->flags |= PG_WANTED | PG_REFERENCED; 271976e77fcSDavid Greenman cnt.v_intrans++; 27224a1cce3SDavid Greenman tsleep(m, PSWP, "vmpfw", 0); 27326f9a767SRodney W. Grimes } 27416f62314SDavid Greenman splx(s); 275df8bae1dSRodney W. Grimes vm_object_deallocate(first_object); 276df8bae1dSRodney W. Grimes goto RetryFault; 277df8bae1dSRodney W. Grimes } 278f6b04d2bSDavid Greenman 279df8bae1dSRodney W. Grimes /* 28024a1cce3SDavid Greenman * Mark page busy for other processes, and the pagedaemon. 281df8bae1dSRodney W. Grimes */ 282df8bae1dSRodney W. Grimes m->flags |= PG_BUSY; 283bd7e5f99SJohn Dyson if ((m->queue == PQ_CACHE) && 28422ba64e8SJohn Dyson (cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_reserved) { 28522ba64e8SJohn Dyson UNLOCK_AND_DEALLOCATE; 28622ba64e8SJohn Dyson VM_WAIT; 28722ba64e8SJohn Dyson PAGE_WAKEUP(m); 28822ba64e8SJohn Dyson goto RetryFault; 28922ba64e8SJohn Dyson } 29022ba64e8SJohn Dyson 291bd7e5f99SJohn Dyson if (m->valid && 292bd7e5f99SJohn Dyson ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) && 293f919ebdeSDavid Greenman m->object != kernel_object && m->object != kmem_object) { 2940d94caffSDavid Greenman goto readrest; 2950d94caffSDavid Greenman } 296df8bae1dSRodney W. Grimes break; 297df8bae1dSRodney W. Grimes } 29824a1cce3SDavid Greenman if (((object->type != OBJT_DEFAULT) && (!change_wiring || wired)) 299df8bae1dSRodney W. Grimes || (object == first_object)) { 300df8bae1dSRodney W. Grimes 301a316d390SJohn Dyson if (pindex >= object->size) { 3025f55e841SDavid Greenman UNLOCK_AND_DEALLOCATE; 3035f55e841SDavid Greenman return (KERN_PROTECTION_FAILURE); 3045f55e841SDavid Greenman } 30522ba64e8SJohn Dyson 306df8bae1dSRodney W. Grimes /* 3070d94caffSDavid Greenman * Allocate a new page for this object/offset pair. 308df8bae1dSRodney W. Grimes */ 309a316d390SJohn Dyson m = vm_page_alloc(object, pindex, 310bd7e5f99SJohn Dyson vp?VM_ALLOC_NORMAL:VM_ALLOC_ZERO); 311df8bae1dSRodney W. Grimes 312df8bae1dSRodney W. Grimes if (m == NULL) { 313df8bae1dSRodney W. Grimes UNLOCK_AND_DEALLOCATE; 314df8bae1dSRodney W. Grimes VM_WAIT; 315df8bae1dSRodney W. Grimes goto RetryFault; 316df8bae1dSRodney W. Grimes } 317df8bae1dSRodney W. Grimes } 3180d94caffSDavid Greenman readrest: 31924a1cce3SDavid Greenman if (object->type != OBJT_DEFAULT && (!change_wiring || wired)) { 320df8bae1dSRodney W. Grimes int rv; 32126f9a767SRodney W. Grimes int faultcount; 32226f9a767SRodney W. Grimes int reqpage; 323df8bae1dSRodney W. Grimes 324df8bae1dSRodney W. Grimes /* 3250d94caffSDavid Greenman * now we find out if any other pages should be paged 3260d94caffSDavid Greenman * in at this time this routine checks to see if the 3270d94caffSDavid Greenman * pages surrounding this fault reside in the same 3280d94caffSDavid Greenman * object as the page for this fault. If they do, 3290d94caffSDavid Greenman * then they are faulted in also into the object. The 3300d94caffSDavid Greenman * array "marray" returned contains an array of 3310d94caffSDavid Greenman * vm_page_t structs where one of them is the 3320d94caffSDavid Greenman * vm_page_t passed to the routine. The reqpage 3330d94caffSDavid Greenman * return value is the index into the marray for the 3340d94caffSDavid Greenman * vm_page_t passed to the routine. 33526f9a767SRodney W. Grimes */ 33605f0fdd2SPoul-Henning Kamp faultcount = vm_fault_additional_pages( 33705f0fdd2SPoul-Henning Kamp m, VM_FAULT_READ_BEHIND, VM_FAULT_READ_AHEAD, 33805f0fdd2SPoul-Henning Kamp marray, &reqpage); 339df8bae1dSRodney W. Grimes 340df8bae1dSRodney W. Grimes /* 3410d94caffSDavid Greenman * Call the pager to retrieve the data, if any, after 3420d94caffSDavid Greenman * releasing the lock on the map. 343df8bae1dSRodney W. Grimes */ 344df8bae1dSRodney W. Grimes UNLOCK_MAP; 345df8bae1dSRodney W. Grimes 34626f9a767SRodney W. Grimes rv = faultcount ? 34724a1cce3SDavid Greenman vm_pager_get_pages(object, marray, faultcount, 34824a1cce3SDavid Greenman reqpage) : VM_PAGER_FAIL; 34922ba64e8SJohn Dyson 35026f9a767SRodney W. Grimes if (rv == VM_PAGER_OK) { 351df8bae1dSRodney W. Grimes /* 3520d94caffSDavid Greenman * Found the page. Leave it busy while we play 3530d94caffSDavid Greenman * with it. 354df8bae1dSRodney W. Grimes */ 35526f9a767SRodney W. Grimes 356df8bae1dSRodney W. Grimes /* 3570d94caffSDavid Greenman * Relookup in case pager changed page. Pager 3580d94caffSDavid Greenman * is responsible for disposition of old page 3590d94caffSDavid Greenman * if moved. 360df8bae1dSRodney W. Grimes */ 361a316d390SJohn Dyson m = vm_page_lookup(object, pindex); 362f6b04d2bSDavid Greenman if( !m) { 363f6b04d2bSDavid Greenman UNLOCK_AND_DEALLOCATE; 364f6b04d2bSDavid Greenman goto RetryFault; 365f6b04d2bSDavid Greenman } 366f6b04d2bSDavid Greenman 36726f9a767SRodney W. Grimes hardfault++; 368df8bae1dSRodney W. Grimes break; 369df8bae1dSRodney W. Grimes } 370df8bae1dSRodney W. Grimes /* 3710d94caffSDavid Greenman * Remove the bogus page (which does not exist at this 3720d94caffSDavid Greenman * object/offset); before doing so, we must get back 3730d94caffSDavid Greenman * our object lock to preserve our invariant. 374df8bae1dSRodney W. Grimes * 37524a1cce3SDavid Greenman * Also wake up any other process that may want to bring 3760d94caffSDavid Greenman * in this page. 377df8bae1dSRodney W. Grimes * 3780d94caffSDavid Greenman * If this is the top-level object, we must leave the 37924a1cce3SDavid Greenman * busy page to prevent another process from rushing 3800d94caffSDavid Greenman * past us, and inserting the page in that object at 3810d94caffSDavid Greenman * the same time that we are. 382df8bae1dSRodney W. Grimes */ 38326f9a767SRodney W. Grimes 384a83c285cSDavid Greenman if (rv == VM_PAGER_ERROR) 385a83c285cSDavid Greenman printf("vm_fault: pager input (probably hardware) error, PID %d failure\n", 386a83c285cSDavid Greenman curproc->p_pid); 38726f9a767SRodney W. Grimes /* 388a83c285cSDavid Greenman * Data outside the range of the pager or an I/O error 38926f9a767SRodney W. Grimes */ 390a83c285cSDavid Greenman /* 3910d94caffSDavid Greenman * XXX - the check for kernel_map is a kludge to work 3920d94caffSDavid Greenman * around having the machine panic on a kernel space 3930d94caffSDavid Greenman * fault w/ I/O error. 394a83c285cSDavid Greenman */ 395a83c285cSDavid Greenman if (((map != kernel_map) && (rv == VM_PAGER_ERROR)) || (rv == VM_PAGER_BAD)) { 39626f9a767SRodney W. Grimes FREE_PAGE(m); 39726f9a767SRodney W. Grimes UNLOCK_AND_DEALLOCATE; 398a83c285cSDavid Greenman return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE); 39926f9a767SRodney W. Grimes } 400df8bae1dSRodney W. Grimes if (object != first_object) { 401df8bae1dSRodney W. Grimes FREE_PAGE(m); 40226f9a767SRodney W. Grimes /* 40326f9a767SRodney W. Grimes * XXX - we cannot just fall out at this 40426f9a767SRodney W. Grimes * point, m has been freed and is invalid! 40526f9a767SRodney W. Grimes */ 406df8bae1dSRodney W. Grimes } 407df8bae1dSRodney W. Grimes } 408df8bae1dSRodney W. Grimes /* 40924a1cce3SDavid Greenman * We get here if the object has default pager (or unwiring) or the 4100d94caffSDavid Greenman * pager doesn't have the page. 411df8bae1dSRodney W. Grimes */ 412df8bae1dSRodney W. Grimes if (object == first_object) 413df8bae1dSRodney W. Grimes first_m = m; 414df8bae1dSRodney W. Grimes 415df8bae1dSRodney W. Grimes /* 4160d94caffSDavid Greenman * Move on to the next object. Lock the next object before 4170d94caffSDavid Greenman * unlocking the current one. 418df8bae1dSRodney W. Grimes */ 419df8bae1dSRodney W. Grimes 420a316d390SJohn Dyson pindex += OFF_TO_IDX(object->backing_object_offset); 42124a1cce3SDavid Greenman next_object = object->backing_object; 422df8bae1dSRodney W. Grimes if (next_object == NULL) { 423df8bae1dSRodney W. Grimes /* 4240d94caffSDavid Greenman * If there's no object left, fill the page in the top 4250d94caffSDavid Greenman * object with zeros. 426df8bae1dSRodney W. Grimes */ 427df8bae1dSRodney W. Grimes if (object != first_object) { 428f919ebdeSDavid Greenman vm_object_pip_wakeup(object); 429df8bae1dSRodney W. Grimes 430df8bae1dSRodney W. Grimes object = first_object; 431a316d390SJohn Dyson pindex = first_pindex; 432df8bae1dSRodney W. Grimes m = first_m; 433df8bae1dSRodney W. Grimes } 434df8bae1dSRodney W. Grimes first_m = NULL; 435df8bae1dSRodney W. Grimes 436f70f05f2SJohn Dyson if ((m->flags & PG_ZERO) == 0) 437df8bae1dSRodney W. Grimes vm_page_zero_fill(m); 4380d94caffSDavid Greenman m->valid = VM_PAGE_BITS_ALL; 439df8bae1dSRodney W. Grimes cnt.v_zfod++; 440df8bae1dSRodney W. Grimes break; 4410d94caffSDavid Greenman } else { 44226f9a767SRodney W. Grimes if (object != first_object) { 443f919ebdeSDavid Greenman vm_object_pip_wakeup(object); 444c0503609SDavid Greenman } 445df8bae1dSRodney W. Grimes object = next_object; 446df8bae1dSRodney W. Grimes object->paging_in_progress++; 447df8bae1dSRodney W. Grimes } 448df8bae1dSRodney W. Grimes } 449df8bae1dSRodney W. Grimes 450f919ebdeSDavid Greenman if ((m->flags & PG_BUSY) == 0) 451f919ebdeSDavid Greenman panic("vm_fault: not busy after main loop"); 452df8bae1dSRodney W. Grimes 453df8bae1dSRodney W. Grimes /* 4540d94caffSDavid Greenman * PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock 455df8bae1dSRodney W. Grimes * is held.] 456df8bae1dSRodney W. Grimes */ 457df8bae1dSRodney W. Grimes 458df8bae1dSRodney W. Grimes old_m = m; /* save page that would be copied */ 459df8bae1dSRodney W. Grimes 460df8bae1dSRodney W. Grimes /* 4610d94caffSDavid Greenman * If the page is being written, but isn't already owned by the 4620d94caffSDavid Greenman * top-level object, we have to copy it into a new page owned by the 4630d94caffSDavid Greenman * top-level object. 464df8bae1dSRodney W. Grimes */ 465df8bae1dSRodney W. Grimes 466df8bae1dSRodney W. Grimes if (object != first_object) { 467df8bae1dSRodney W. Grimes /* 4680d94caffSDavid Greenman * We only really need to copy if we want to write it. 469df8bae1dSRodney W. Grimes */ 470df8bae1dSRodney W. Grimes 471df8bae1dSRodney W. Grimes if (fault_type & VM_PROT_WRITE) { 472df8bae1dSRodney W. Grimes 473df8bae1dSRodney W. Grimes /* 4740d94caffSDavid Greenman * If we try to collapse first_object at this point, 4750d94caffSDavid Greenman * we may deadlock when we try to get the lock on an 4760d94caffSDavid Greenman * intermediate object (since we have the bottom 4770d94caffSDavid Greenman * object locked). We can't unlock the bottom object, 4780d94caffSDavid Greenman * because the page we found may move (by collapse) if 4790d94caffSDavid Greenman * we do. 480df8bae1dSRodney W. Grimes * 4810d94caffSDavid Greenman * Instead, we first copy the page. Then, when we have 4820d94caffSDavid Greenman * no more use for the bottom object, we unlock it and 4830d94caffSDavid Greenman * try to collapse. 484df8bae1dSRodney W. Grimes * 4850d94caffSDavid Greenman * Note that we copy the page even if we didn't need 4860d94caffSDavid Greenman * to... that's the breaks. 487df8bae1dSRodney W. Grimes */ 488df8bae1dSRodney W. Grimes 489df8bae1dSRodney W. Grimes /* 4900d94caffSDavid Greenman * We already have an empty page in first_object - use 4910d94caffSDavid Greenman * it. 492df8bae1dSRodney W. Grimes */ 493df8bae1dSRodney W. Grimes 494df8bae1dSRodney W. Grimes vm_page_copy(m, first_m); 4950d94caffSDavid Greenman first_m->valid = VM_PAGE_BITS_ALL; 496df8bae1dSRodney W. Grimes 497df8bae1dSRodney W. Grimes /* 4980d94caffSDavid Greenman * If another map is truly sharing this page with us, 4990d94caffSDavid Greenman * we have to flush all uses of the original page, 5000d94caffSDavid Greenman * since we can't distinguish those which want the 5010d94caffSDavid Greenman * original from those which need the new copy. 502df8bae1dSRodney W. Grimes * 5030d94caffSDavid Greenman * XXX If we know that only one map has access to this 5040d94caffSDavid Greenman * page, then we could avoid the pmap_page_protect() 5050d94caffSDavid Greenman * call. 506df8bae1dSRodney W. Grimes */ 507df8bae1dSRodney W. Grimes 508bd7e5f99SJohn Dyson if (m->queue != PQ_ACTIVE) 509df8bae1dSRodney W. Grimes vm_page_activate(m); 510df8bae1dSRodney W. Grimes 511df8bae1dSRodney W. Grimes /* 512df8bae1dSRodney W. Grimes * We no longer need the old page or object. 513df8bae1dSRodney W. Grimes */ 514df8bae1dSRodney W. Grimes PAGE_WAKEUP(m); 515f919ebdeSDavid Greenman vm_object_pip_wakeup(object); 516df8bae1dSRodney W. Grimes 517df8bae1dSRodney W. Grimes /* 518df8bae1dSRodney W. Grimes * Only use the new page below... 519df8bae1dSRodney W. Grimes */ 520df8bae1dSRodney W. Grimes 521df8bae1dSRodney W. Grimes cnt.v_cow_faults++; 522df8bae1dSRodney W. Grimes m = first_m; 523df8bae1dSRodney W. Grimes object = first_object; 524a316d390SJohn Dyson pindex = first_pindex; 525df8bae1dSRodney W. Grimes 526df8bae1dSRodney W. Grimes /* 5270d94caffSDavid Greenman * Now that we've gotten the copy out of the way, 5280d94caffSDavid Greenman * let's try to collapse the top object. 52924a1cce3SDavid Greenman * 530df8bae1dSRodney W. Grimes * But we have to play ugly games with 531df8bae1dSRodney W. Grimes * paging_in_progress to do that... 532df8bae1dSRodney W. Grimes */ 533f919ebdeSDavid Greenman vm_object_pip_wakeup(object); 534df8bae1dSRodney W. Grimes vm_object_collapse(object); 535df8bae1dSRodney W. Grimes object->paging_in_progress++; 5360d94caffSDavid Greenman } else { 537df8bae1dSRodney W. Grimes prot &= ~VM_PROT_WRITE; 538df8bae1dSRodney W. Grimes } 539df8bae1dSRodney W. Grimes } 540df8bae1dSRodney W. Grimes 541df8bae1dSRodney W. Grimes /* 5420d94caffSDavid Greenman * We must verify that the maps have not changed since our last 5430d94caffSDavid Greenman * lookup. 544df8bae1dSRodney W. Grimes */ 545df8bae1dSRodney W. Grimes 546df8bae1dSRodney W. Grimes if (!lookup_still_valid) { 547df8bae1dSRodney W. Grimes vm_object_t retry_object; 548a316d390SJohn Dyson vm_pindex_t retry_pindex; 549df8bae1dSRodney W. Grimes vm_prot_t retry_prot; 550df8bae1dSRodney W. Grimes 551df8bae1dSRodney W. Grimes /* 5520d94caffSDavid Greenman * Since map entries may be pageable, make sure we can take a 5530d94caffSDavid Greenman * page fault on them. 554df8bae1dSRodney W. Grimes */ 555df8bae1dSRodney W. Grimes 556df8bae1dSRodney W. Grimes /* 55724a1cce3SDavid Greenman * To avoid trying to write_lock the map while another process 5580d94caffSDavid Greenman * has it read_locked (in vm_map_pageable), we do not try for 5590d94caffSDavid Greenman * write permission. If the page is still writable, we will 5600d94caffSDavid Greenman * get write permission. If it is not, or has been marked 5610d94caffSDavid Greenman * needs_copy, we enter the mapping without write permission, 5620d94caffSDavid Greenman * and will merely take another fault. 563df8bae1dSRodney W. Grimes */ 5640d94caffSDavid Greenman result = vm_map_lookup(&map, vaddr, fault_type & ~VM_PROT_WRITE, 565a316d390SJohn Dyson &entry, &retry_object, &retry_pindex, &retry_prot, &wired, &su); 566df8bae1dSRodney W. Grimes 567df8bae1dSRodney W. Grimes /* 5680d94caffSDavid Greenman * If we don't need the page any longer, put it on the active 5690d94caffSDavid Greenman * list (the easiest thing to do here). If no one needs it, 5700d94caffSDavid Greenman * pageout will grab it eventually. 571df8bae1dSRodney W. Grimes */ 572df8bae1dSRodney W. Grimes 573df8bae1dSRodney W. Grimes if (result != KERN_SUCCESS) { 574df8bae1dSRodney W. Grimes RELEASE_PAGE(m); 575df8bae1dSRodney W. Grimes UNLOCK_AND_DEALLOCATE; 576df8bae1dSRodney W. Grimes return (result); 577df8bae1dSRodney W. Grimes } 578df8bae1dSRodney W. Grimes lookup_still_valid = TRUE; 579df8bae1dSRodney W. Grimes 580df8bae1dSRodney W. Grimes if ((retry_object != first_object) || 581a316d390SJohn Dyson (retry_pindex != first_pindex)) { 582df8bae1dSRodney W. Grimes RELEASE_PAGE(m); 583df8bae1dSRodney W. Grimes UNLOCK_AND_DEALLOCATE; 584df8bae1dSRodney W. Grimes goto RetryFault; 585df8bae1dSRodney W. Grimes } 586df8bae1dSRodney W. Grimes /* 5870d94caffSDavid Greenman * Check whether the protection has changed or the object has 5880d94caffSDavid Greenman * been copied while we left the map unlocked. Changing from 5890d94caffSDavid Greenman * read to write permission is OK - we leave the page 5900d94caffSDavid Greenman * write-protected, and catch the write fault. Changing from 5910d94caffSDavid Greenman * write to read permission means that we can't mark the page 5920d94caffSDavid Greenman * write-enabled after all. 593df8bae1dSRodney W. Grimes */ 594df8bae1dSRodney W. Grimes prot &= retry_prot; 595df8bae1dSRodney W. Grimes } 596df8bae1dSRodney W. Grimes /* 5970d94caffSDavid Greenman * (the various bits we're fiddling with here are locked by the 5980d94caffSDavid Greenman * object's lock) 599df8bae1dSRodney W. Grimes */ 600df8bae1dSRodney W. Grimes 601df8bae1dSRodney W. Grimes /* 6020d94caffSDavid Greenman * It's critically important that a wired-down page be faulted only 6030d94caffSDavid Greenman * once in each map for which it is wired. 604df8bae1dSRodney W. Grimes */ 605df8bae1dSRodney W. Grimes 606df8bae1dSRodney W. Grimes /* 6070d94caffSDavid Greenman * Put this page into the physical map. We had to do the unlock above 6080d94caffSDavid Greenman * because pmap_enter may cause other faults. We don't put the page 6090d94caffSDavid Greenman * back on the active queue until later so that the page-out daemon 6100d94caffSDavid Greenman * won't find us (yet). 611df8bae1dSRodney W. Grimes */ 612df8bae1dSRodney W. Grimes 6132ddba215SDavid Greenman if (prot & VM_PROT_WRITE) { 614f919ebdeSDavid Greenman m->flags |= PG_WRITEABLE; 615aef922f5SJohn Dyson m->object->flags |= OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY; 6162ddba215SDavid Greenman /* 6172ddba215SDavid Greenman * If the fault is a write, we know that this page is being 6182ddba215SDavid Greenman * written NOW. This will save on the pmap_is_modified() calls 6192ddba215SDavid Greenman * later. 6202ddba215SDavid Greenman */ 6212ddba215SDavid Greenman if (fault_type & VM_PROT_WRITE) { 6222ddba215SDavid Greenman m->dirty = VM_PAGE_BITS_ALL; 6232ddba215SDavid Greenman } 6242ddba215SDavid Greenman } 625f6b04d2bSDavid Greenman 626f70f05f2SJohn Dyson m->flags |= PG_MAPPED|PG_REFERENCED; 627ced399eeSJohn Dyson m->flags &= ~PG_ZERO; 628f919ebdeSDavid Greenman 629df8bae1dSRodney W. Grimes pmap_enter(map->pmap, vaddr, VM_PAGE_TO_PHYS(m), prot, wired); 6301e9122e5SDavid Greenman #if 0 631f6b04d2bSDavid Greenman if (change_wiring == 0 && wired == 0) 6326d40c3d3SDavid Greenman pmap_prefault(map->pmap, vaddr, entry, first_object); 6331e9122e5SDavid Greenman #endif 634df8bae1dSRodney W. Grimes 635df8bae1dSRodney W. Grimes /* 6360d94caffSDavid Greenman * If the page is not wired down, then put it where the pageout daemon 6370d94caffSDavid Greenman * can find it. 638df8bae1dSRodney W. Grimes */ 639df8bae1dSRodney W. Grimes if (change_wiring) { 640df8bae1dSRodney W. Grimes if (wired) 641df8bae1dSRodney W. Grimes vm_page_wire(m); 642df8bae1dSRodney W. Grimes else 643df8bae1dSRodney W. Grimes vm_page_unwire(m); 6440d94caffSDavid Greenman } else { 645bd7e5f99SJohn Dyson if (m->queue != PQ_ACTIVE) 646df8bae1dSRodney W. Grimes vm_page_activate(m); 64726f9a767SRodney W. Grimes } 64826f9a767SRodney W. Grimes 649a1f6d91cSDavid Greenman if (curproc && (curproc->p_flag & P_INMEM) && curproc->p_stats) { 65026f9a767SRodney W. Grimes if (hardfault) { 65126f9a767SRodney W. Grimes curproc->p_stats->p_ru.ru_majflt++; 65226f9a767SRodney W. Grimes } else { 65326f9a767SRodney W. Grimes curproc->p_stats->p_ru.ru_minflt++; 65426f9a767SRodney W. Grimes } 65526f9a767SRodney W. Grimes } 656df8bae1dSRodney W. Grimes 657df8bae1dSRodney W. Grimes /* 658df8bae1dSRodney W. Grimes * Unlock everything, and return 659df8bae1dSRodney W. Grimes */ 660df8bae1dSRodney W. Grimes 661df8bae1dSRodney W. Grimes PAGE_WAKEUP(m); 662df8bae1dSRodney W. Grimes UNLOCK_AND_DEALLOCATE; 663df8bae1dSRodney W. Grimes 664df8bae1dSRodney W. Grimes return (KERN_SUCCESS); 665df8bae1dSRodney W. Grimes 666df8bae1dSRodney W. Grimes } 667df8bae1dSRodney W. Grimes 668df8bae1dSRodney W. Grimes /* 669df8bae1dSRodney W. Grimes * vm_fault_wire: 670df8bae1dSRodney W. Grimes * 671df8bae1dSRodney W. Grimes * Wire down a range of virtual addresses in a map. 672df8bae1dSRodney W. Grimes */ 673df8bae1dSRodney W. Grimes int 674df8bae1dSRodney W. Grimes vm_fault_wire(map, start, end) 675df8bae1dSRodney W. Grimes vm_map_t map; 676df8bae1dSRodney W. Grimes vm_offset_t start, end; 677df8bae1dSRodney W. Grimes { 67826f9a767SRodney W. Grimes 679df8bae1dSRodney W. Grimes register vm_offset_t va; 680df8bae1dSRodney W. Grimes register pmap_t pmap; 681df8bae1dSRodney W. Grimes int rv; 682df8bae1dSRodney W. Grimes 683df8bae1dSRodney W. Grimes pmap = vm_map_pmap(map); 684df8bae1dSRodney W. Grimes 685df8bae1dSRodney W. Grimes /* 6860d94caffSDavid Greenman * Inform the physical mapping system that the range of addresses may 6870d94caffSDavid Greenman * not fault, so that page tables and such can be locked down as well. 688df8bae1dSRodney W. Grimes */ 689df8bae1dSRodney W. Grimes 690df8bae1dSRodney W. Grimes pmap_pageable(pmap, start, end, FALSE); 691df8bae1dSRodney W. Grimes 692df8bae1dSRodney W. Grimes /* 6930d94caffSDavid Greenman * We simulate a fault to get the page and enter it in the physical 6940d94caffSDavid Greenman * map. 695df8bae1dSRodney W. Grimes */ 696df8bae1dSRodney W. Grimes 697df8bae1dSRodney W. Grimes for (va = start; va < end; va += PAGE_SIZE) { 6986d40c3d3SDavid Greenman 699a1f6d91cSDavid Greenman while( curproc != pageproc && 7006d40c3d3SDavid Greenman (cnt.v_free_count <= cnt.v_pageout_free_min)) 7016d40c3d3SDavid Greenman VM_WAIT; 7026d40c3d3SDavid Greenman 703a1f6d91cSDavid Greenman rv = vm_fault(map, va, VM_PROT_READ|VM_PROT_WRITE, TRUE); 704df8bae1dSRodney W. Grimes if (rv) { 705df8bae1dSRodney W. Grimes if (va != start) 706df8bae1dSRodney W. Grimes vm_fault_unwire(map, start, va); 707df8bae1dSRodney W. Grimes return (rv); 708df8bae1dSRodney W. Grimes } 709df8bae1dSRodney W. Grimes } 710df8bae1dSRodney W. Grimes return (KERN_SUCCESS); 711df8bae1dSRodney W. Grimes } 712df8bae1dSRodney W. Grimes 713df8bae1dSRodney W. Grimes 714df8bae1dSRodney W. Grimes /* 715df8bae1dSRodney W. Grimes * vm_fault_unwire: 716df8bae1dSRodney W. Grimes * 717df8bae1dSRodney W. Grimes * Unwire a range of virtual addresses in a map. 718df8bae1dSRodney W. Grimes */ 71926f9a767SRodney W. Grimes void 72026f9a767SRodney W. Grimes vm_fault_unwire(map, start, end) 721df8bae1dSRodney W. Grimes vm_map_t map; 722df8bae1dSRodney W. Grimes vm_offset_t start, end; 723df8bae1dSRodney W. Grimes { 724df8bae1dSRodney W. Grimes 725df8bae1dSRodney W. Grimes register vm_offset_t va, pa; 726df8bae1dSRodney W. Grimes register pmap_t pmap; 727df8bae1dSRodney W. Grimes 728df8bae1dSRodney W. Grimes pmap = vm_map_pmap(map); 729df8bae1dSRodney W. Grimes 730df8bae1dSRodney W. Grimes /* 7310d94caffSDavid Greenman * Since the pages are wired down, we must be able to get their 7320d94caffSDavid Greenman * mappings from the physical map system. 733df8bae1dSRodney W. Grimes */ 734df8bae1dSRodney W. Grimes 735df8bae1dSRodney W. Grimes for (va = start; va < end; va += PAGE_SIZE) { 736df8bae1dSRodney W. Grimes pa = pmap_extract(pmap, va); 737df8bae1dSRodney W. Grimes if (pa == (vm_offset_t) 0) { 738df8bae1dSRodney W. Grimes panic("unwire: page not in pmap"); 739df8bae1dSRodney W. Grimes } 740df8bae1dSRodney W. Grimes pmap_change_wiring(pmap, va, FALSE); 741df8bae1dSRodney W. Grimes vm_page_unwire(PHYS_TO_VM_PAGE(pa)); 742df8bae1dSRodney W. Grimes } 743df8bae1dSRodney W. Grimes 744df8bae1dSRodney W. Grimes /* 7450d94caffSDavid Greenman * Inform the physical mapping system that the range of addresses may 7460d94caffSDavid Greenman * fault, so that page tables and such may be unwired themselves. 747df8bae1dSRodney W. Grimes */ 748df8bae1dSRodney W. Grimes 749df8bae1dSRodney W. Grimes pmap_pageable(pmap, start, end, TRUE); 750df8bae1dSRodney W. Grimes 751df8bae1dSRodney W. Grimes } 752df8bae1dSRodney W. Grimes 753df8bae1dSRodney W. Grimes /* 754df8bae1dSRodney W. Grimes * Routine: 755df8bae1dSRodney W. Grimes * vm_fault_copy_entry 756df8bae1dSRodney W. Grimes * Function: 757df8bae1dSRodney W. Grimes * Copy all of the pages from a wired-down map entry to another. 758df8bae1dSRodney W. Grimes * 759df8bae1dSRodney W. Grimes * In/out conditions: 760df8bae1dSRodney W. Grimes * The source and destination maps must be locked for write. 761df8bae1dSRodney W. Grimes * The source map entry must be wired down (or be a sharing map 762df8bae1dSRodney W. Grimes * entry corresponding to a main map entry that is wired down). 763df8bae1dSRodney W. Grimes */ 764df8bae1dSRodney W. Grimes 76526f9a767SRodney W. Grimes void 76626f9a767SRodney W. Grimes vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry) 767df8bae1dSRodney W. Grimes vm_map_t dst_map; 768df8bae1dSRodney W. Grimes vm_map_t src_map; 769df8bae1dSRodney W. Grimes vm_map_entry_t dst_entry; 770df8bae1dSRodney W. Grimes vm_map_entry_t src_entry; 771df8bae1dSRodney W. Grimes { 772df8bae1dSRodney W. Grimes vm_object_t dst_object; 773df8bae1dSRodney W. Grimes vm_object_t src_object; 774a316d390SJohn Dyson vm_ooffset_t dst_offset; 775a316d390SJohn Dyson vm_ooffset_t src_offset; 776df8bae1dSRodney W. Grimes vm_prot_t prot; 777df8bae1dSRodney W. Grimes vm_offset_t vaddr; 778df8bae1dSRodney W. Grimes vm_page_t dst_m; 779df8bae1dSRodney W. Grimes vm_page_t src_m; 780df8bae1dSRodney W. Grimes 781df8bae1dSRodney W. Grimes #ifdef lint 782df8bae1dSRodney W. Grimes src_map++; 7830d94caffSDavid Greenman #endif /* lint */ 784df8bae1dSRodney W. Grimes 785df8bae1dSRodney W. Grimes src_object = src_entry->object.vm_object; 786df8bae1dSRodney W. Grimes src_offset = src_entry->offset; 787df8bae1dSRodney W. Grimes 788df8bae1dSRodney W. Grimes /* 7890d94caffSDavid Greenman * Create the top-level object for the destination entry. (Doesn't 7900d94caffSDavid Greenman * actually shadow anything - we copy the pages directly.) 791df8bae1dSRodney W. Grimes */ 79224a1cce3SDavid Greenman dst_object = vm_object_allocate(OBJT_DEFAULT, 793a316d390SJohn Dyson (vm_size_t) OFF_TO_IDX(dst_entry->end - dst_entry->start)); 794df8bae1dSRodney W. Grimes 795df8bae1dSRodney W. Grimes dst_entry->object.vm_object = dst_object; 796df8bae1dSRodney W. Grimes dst_entry->offset = 0; 797df8bae1dSRodney W. Grimes 798df8bae1dSRodney W. Grimes prot = dst_entry->max_protection; 799df8bae1dSRodney W. Grimes 800df8bae1dSRodney W. Grimes /* 8010d94caffSDavid Greenman * Loop through all of the pages in the entry's range, copying each 8020d94caffSDavid Greenman * one from the source object (it should be there) to the destination 8030d94caffSDavid Greenman * object. 804df8bae1dSRodney W. Grimes */ 805df8bae1dSRodney W. Grimes for (vaddr = dst_entry->start, dst_offset = 0; 806df8bae1dSRodney W. Grimes vaddr < dst_entry->end; 807df8bae1dSRodney W. Grimes vaddr += PAGE_SIZE, dst_offset += PAGE_SIZE) { 808df8bae1dSRodney W. Grimes 809df8bae1dSRodney W. Grimes /* 810df8bae1dSRodney W. Grimes * Allocate a page in the destination object 811df8bae1dSRodney W. Grimes */ 812df8bae1dSRodney W. Grimes do { 813a316d390SJohn Dyson dst_m = vm_page_alloc(dst_object, 814a316d390SJohn Dyson OFF_TO_IDX(dst_offset), VM_ALLOC_NORMAL); 815df8bae1dSRodney W. Grimes if (dst_m == NULL) { 816df8bae1dSRodney W. Grimes VM_WAIT; 817df8bae1dSRodney W. Grimes } 818df8bae1dSRodney W. Grimes } while (dst_m == NULL); 819df8bae1dSRodney W. Grimes 820df8bae1dSRodney W. Grimes /* 821df8bae1dSRodney W. Grimes * Find the page in the source object, and copy it in. 8220d94caffSDavid Greenman * (Because the source is wired down, the page will be in 8230d94caffSDavid Greenman * memory.) 824df8bae1dSRodney W. Grimes */ 825a316d390SJohn Dyson src_m = vm_page_lookup(src_object, 826a316d390SJohn Dyson OFF_TO_IDX(dst_offset + src_offset)); 827df8bae1dSRodney W. Grimes if (src_m == NULL) 828df8bae1dSRodney W. Grimes panic("vm_fault_copy_wired: page missing"); 829df8bae1dSRodney W. Grimes 830df8bae1dSRodney W. Grimes vm_page_copy(src_m, dst_m); 831df8bae1dSRodney W. Grimes 832df8bae1dSRodney W. Grimes /* 833df8bae1dSRodney W. Grimes * Enter it in the pmap... 834df8bae1dSRodney W. Grimes */ 835df8bae1dSRodney W. Grimes 836ced399eeSJohn Dyson dst_m->flags |= PG_WRITEABLE|PG_MAPPED; 837ccbb2f72SJohn Dyson dst_m->flags &= ~PG_ZERO; 838df8bae1dSRodney W. Grimes pmap_enter(dst_map->pmap, vaddr, VM_PAGE_TO_PHYS(dst_m), 839df8bae1dSRodney W. Grimes prot, FALSE); 840df8bae1dSRodney W. Grimes 841df8bae1dSRodney W. Grimes /* 842df8bae1dSRodney W. Grimes * Mark it no longer busy, and put it on the active list. 843df8bae1dSRodney W. Grimes */ 844df8bae1dSRodney W. Grimes vm_page_activate(dst_m); 845df8bae1dSRodney W. Grimes PAGE_WAKEUP(dst_m); 846df8bae1dSRodney W. Grimes } 847df8bae1dSRodney W. Grimes } 84826f9a767SRodney W. Grimes 84926f9a767SRodney W. Grimes 85026f9a767SRodney W. Grimes /* 85126f9a767SRodney W. Grimes * This routine checks around the requested page for other pages that 85222ba64e8SJohn Dyson * might be able to be faulted in. This routine brackets the viable 85322ba64e8SJohn Dyson * pages for the pages to be paged in. 85426f9a767SRodney W. Grimes * 85526f9a767SRodney W. Grimes * Inputs: 85622ba64e8SJohn Dyson * m, rbehind, rahead 85726f9a767SRodney W. Grimes * 85826f9a767SRodney W. Grimes * Outputs: 85926f9a767SRodney W. Grimes * marray (array of vm_page_t), reqpage (index of requested page) 86026f9a767SRodney W. Grimes * 86126f9a767SRodney W. Grimes * Return value: 86226f9a767SRodney W. Grimes * number of pages in marray 86326f9a767SRodney W. Grimes */ 86426f9a767SRodney W. Grimes int 86522ba64e8SJohn Dyson vm_fault_additional_pages(m, rbehind, rahead, marray, reqpage) 86626f9a767SRodney W. Grimes vm_page_t m; 86726f9a767SRodney W. Grimes int rbehind; 86822ba64e8SJohn Dyson int rahead; 86926f9a767SRodney W. Grimes vm_page_t *marray; 87026f9a767SRodney W. Grimes int *reqpage; 87126f9a767SRodney W. Grimes { 87226f9a767SRodney W. Grimes int i; 87326f9a767SRodney W. Grimes vm_object_t object; 874a316d390SJohn Dyson vm_pindex_t pindex, startpindex, endpindex, tpindex; 875a316d390SJohn Dyson vm_offset_t size; 87626f9a767SRodney W. Grimes vm_page_t rtm; 87726f9a767SRodney W. Grimes int treqpage; 878170db9c6SJohn Dyson int cbehind, cahead; 87926f9a767SRodney W. Grimes 88026f9a767SRodney W. Grimes object = m->object; 881a316d390SJohn Dyson pindex = m->pindex; 88226f9a767SRodney W. Grimes 88326f9a767SRodney W. Grimes /* 88426f9a767SRodney W. Grimes * if the requested page is not available, then give up now 88526f9a767SRodney W. Grimes */ 88626f9a767SRodney W. Grimes 887170db9c6SJohn Dyson if (!vm_pager_has_page(object, 888a316d390SJohn Dyson OFF_TO_IDX(object->paging_offset) + pindex, &cbehind, &cahead)) 88926f9a767SRodney W. Grimes return 0; 89026f9a767SRodney W. Grimes 89122ba64e8SJohn Dyson if ((cbehind == 0) && (cahead == 0)) { 89222ba64e8SJohn Dyson *reqpage = 0; 89322ba64e8SJohn Dyson marray[0] = m; 89422ba64e8SJohn Dyson return 1; 895170db9c6SJohn Dyson } 89622ba64e8SJohn Dyson 89722ba64e8SJohn Dyson if (rahead > cahead) { 89822ba64e8SJohn Dyson rahead = cahead; 89922ba64e8SJohn Dyson } 90022ba64e8SJohn Dyson 901170db9c6SJohn Dyson if (rbehind > cbehind) { 902170db9c6SJohn Dyson rbehind = cbehind; 903170db9c6SJohn Dyson } 904170db9c6SJohn Dyson 90526f9a767SRodney W. Grimes /* 90626f9a767SRodney W. Grimes * try to do any readahead that we might have free pages for. 90726f9a767SRodney W. Grimes */ 908ccbb2f72SJohn Dyson if ((rahead + rbehind) > 90922ba64e8SJohn Dyson ((cnt.v_free_count + cnt.v_cache_count) - cnt.v_free_reserved)) { 910f919ebdeSDavid Greenman pagedaemon_wakeup(); 91126f9a767SRodney W. Grimes *reqpage = 0; 91226f9a767SRodney W. Grimes marray[0] = m; 91326f9a767SRodney W. Grimes return 1; 91426f9a767SRodney W. Grimes } 91522ba64e8SJohn Dyson 91626f9a767SRodney W. Grimes /* 9170d94caffSDavid Greenman * scan backward for the read behind pages -- in memory or on disk not 9180d94caffSDavid Greenman * in same object 91926f9a767SRodney W. Grimes */ 920a316d390SJohn Dyson tpindex = pindex - 1; 921a316d390SJohn Dyson if (tpindex < pindex) { 922a316d390SJohn Dyson if (rbehind > pindex) 923a316d390SJohn Dyson rbehind = pindex; 924a316d390SJohn Dyson startpindex = pindex - rbehind; 925a316d390SJohn Dyson while (tpindex >= startpindex) { 926a316d390SJohn Dyson if (vm_page_lookup( object, tpindex)) { 927a316d390SJohn Dyson startpindex = tpindex + 1; 92826f9a767SRodney W. Grimes break; 92926f9a767SRodney W. Grimes } 930a316d390SJohn Dyson if (tpindex == 0) 93126f9a767SRodney W. Grimes break; 932a316d390SJohn Dyson tpindex -= 1; 93326f9a767SRodney W. Grimes } 934317205caSDavid Greenman } else { 935a316d390SJohn Dyson startpindex = pindex; 936317205caSDavid Greenman } 93726f9a767SRodney W. Grimes 93826f9a767SRodney W. Grimes /* 9390d94caffSDavid Greenman * scan forward for the read ahead pages -- in memory or on disk not 9400d94caffSDavid Greenman * in same object 94126f9a767SRodney W. Grimes */ 942a316d390SJohn Dyson tpindex = pindex + 1; 943a316d390SJohn Dyson endpindex = pindex + (rahead + 1); 944a316d390SJohn Dyson if (endpindex > object->size) 945a316d390SJohn Dyson endpindex = object->size; 946a316d390SJohn Dyson while (tpindex < endpindex) { 947a316d390SJohn Dyson if ( vm_page_lookup(object, tpindex)) { 94826f9a767SRodney W. Grimes break; 94926f9a767SRodney W. Grimes } 950a316d390SJohn Dyson tpindex += 1; 95126f9a767SRodney W. Grimes } 952a316d390SJohn Dyson endpindex = tpindex; 95326f9a767SRodney W. Grimes 95426f9a767SRodney W. Grimes /* calculate number of bytes of pages */ 955a316d390SJohn Dyson size = endpindex - startpindex; 95626f9a767SRodney W. Grimes 95726f9a767SRodney W. Grimes /* calculate the page offset of the required page */ 958a316d390SJohn Dyson treqpage = pindex - startpindex; 95926f9a767SRodney W. Grimes 96026f9a767SRodney W. Grimes /* see if we have space (again) */ 96122ba64e8SJohn Dyson if ((cnt.v_free_count + cnt.v_cache_count) > 96222ba64e8SJohn Dyson (cnt.v_free_reserved + size)) { 96326f9a767SRodney W. Grimes /* 96426f9a767SRodney W. Grimes * get our pages and don't block for them 96526f9a767SRodney W. Grimes */ 96626f9a767SRodney W. Grimes for (i = 0; i < size; i++) { 967170db9c6SJohn Dyson if (i != treqpage) { 968ccbb2f72SJohn Dyson rtm = vm_page_alloc(object, 969a316d390SJohn Dyson startpindex + i, 97022ba64e8SJohn Dyson VM_ALLOC_NORMAL); 971ccbb2f72SJohn Dyson if (rtm == NULL) { 972170db9c6SJohn Dyson if (i < treqpage) { 973ccbb2f72SJohn Dyson int j; 974ccbb2f72SJohn Dyson for (j = 0; j < i; j++) { 97521bf3904SJohn Dyson FREE_PAGE(marray[j]); 97626f9a767SRodney W. Grimes } 97726f9a767SRodney W. Grimes *reqpage = 0; 97826f9a767SRodney W. Grimes marray[0] = m; 97926f9a767SRodney W. Grimes return 1; 980ccbb2f72SJohn Dyson } else { 981ccbb2f72SJohn Dyson size = i; 982ccbb2f72SJohn Dyson *reqpage = treqpage; 983ccbb2f72SJohn Dyson return size; 984ccbb2f72SJohn Dyson } 985ccbb2f72SJohn Dyson } 986ccbb2f72SJohn Dyson marray[i] = rtm; 987ccbb2f72SJohn Dyson } else { 988ccbb2f72SJohn Dyson marray[i] = m; 989ccbb2f72SJohn Dyson } 99026f9a767SRodney W. Grimes } 991170db9c6SJohn Dyson 99226f9a767SRodney W. Grimes *reqpage = treqpage; 99326f9a767SRodney W. Grimes return size; 99426f9a767SRodney W. Grimes } 99526f9a767SRodney W. Grimes *reqpage = 0; 99626f9a767SRodney W. Grimes marray[0] = m; 99726f9a767SRodney W. Grimes return 1; 99826f9a767SRodney W. Grimes } 999