kernel - Greatly improve concurrent fork's and concurrent exec's

[dragonfly.git] / sys / vm / vm_fault.c

/*
 * (MPSAFE)
 *
 * Copyright (c) 1991, 1993
 *      The Regents of the University of California.  All rights reserved.
 * Copyright (c) 1994 John S. Dyson
 * All rights reserved.
 * Copyright (c) 1994 David Greenman
 * All rights reserved.
 *
 *
 * This code is derived from software contributed to Berkeley by
 * The Mach Operating System project at Carnegie-Mellon University.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *      from: @(#)vm_fault.c    8.4 (Berkeley) 1/12/94
 *
 *
 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
 * All rights reserved.
 *
 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
 *
 * Permission to use, copy, modify and distribute this software and
 * its documentation is hereby granted, provided that both the copyright
 * notice and this permission notice appear in all copies of the
 * software, derivative works or modified versions, and any portions
 * thereof, and that both notices appear in supporting documentation.
 *
 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
 * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
 *
 * Carnegie Mellon requests users of this software to return to
 *
 *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
 *  School of Computer Science
 *  Carnegie Mellon University
 *  Pittsburgh PA 15213-3890
 *
 * any improvements or extensions that they make and grant Carnegie the
 * rights to redistribute these changes.
 *
 * $FreeBSD: src/sys/vm/vm_fault.c,v 1.108.2.8 2002/02/26 05:49:27 silby Exp $
 * $DragonFly: src/sys/vm/vm_fault.c,v 1.47 2008/07/01 02:02:56 dillon Exp $
 */

/*
 *      Page fault handling module.
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/proc.h>
#include <sys/vnode.h>
#include <sys/resourcevar.h>
#include <sys/vmmeter.h>
#include <sys/vkernel.h>
#include <sys/lock.h>
#include <sys/sysctl.h>

#include <cpu/lwbuf.h>

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
#include <vm/vm_kern.h>
#include <vm/vm_pager.h>
#include <vm/vnode_pager.h>
#include <vm/vm_extern.h>

#include <sys/thread2.h>
#include <vm/vm_page2.h>

struct faultstate {
        vm_page_t m;
        vm_object_t object;
        vm_pindex_t pindex;
        vm_prot_t prot;
        vm_page_t first_m;
        vm_object_t first_object;
        vm_prot_t first_prot;
        vm_map_t map;
        vm_map_entry_t entry;
        int lookup_still_valid;
        int hardfault;
        int fault_flags;
        int map_generation;
        int shared;
        int first_shared;
        boolean_t wired;
        struct vnode *vp;
};

static int debug_cluster = 0;
SYSCTL_INT(_vm, OID_AUTO, debug_cluster, CTLFLAG_RW, &debug_cluster, 0, "");
int vm_shared_fault = 1;
TUNABLE_INT("vm.shared_fault", &vm_shared_fault);
SYSCTL_INT(_vm, OID_AUTO, shared_fault, CTLFLAG_RW, &vm_shared_fault, 0,
           "Allow shared token on vm_object");
static long vm_shared_hit = 0;
SYSCTL_LONG(_vm, OID_AUTO, shared_hit, CTLFLAG_RW, &vm_shared_hit, 0,
           "Successful shared faults");
static long vm_shared_count = 0;
SYSCTL_LONG(_vm, OID_AUTO, shared_count, CTLFLAG_RW, &vm_shared_count, 0,
           "Shared fault attempts");
static long vm_shared_miss = 0;
SYSCTL_LONG(_vm, OID_AUTO, shared_miss, CTLFLAG_RW, &vm_shared_miss, 0,
           "Unsuccessful shared faults");

static int vm_fault_object(struct faultstate *, vm_pindex_t, vm_prot_t, int);
static int vm_fault_vpagetable(struct faultstate *, vm_pindex_t *,
                        vpte_t, int, int);
#if 0
static int vm_fault_additional_pages (vm_page_t, int, int, vm_page_t *, int *);
#endif
static void vm_set_nosync(vm_page_t m, vm_map_entry_t entry);
static void vm_prefault(pmap_t pmap, vm_offset_t addra,
                        vm_map_entry_t entry, int prot, int fault_flags);
static void vm_prefault_quick(pmap_t pmap, vm_offset_t addra,
                        vm_map_entry_t entry, int prot, int fault_flags);

static __inline void
release_page(struct faultstate *fs)
{
        vm_page_deactivate(fs->m);
        vm_page_wakeup(fs->m);
        fs->m = NULL;
}

/*
 * NOTE: Once unlocked any cached fs->entry becomes invalid, any reuse
 *       requires relocking and then checking the timestamp.
 *
 * NOTE: vm_map_lock_read() does not bump fs->map->timestamp so we do
 *       not have to update fs->map_generation here.
 *
 * NOTE: This function can fail due to a deadlock against the caller's
 *       holding of a vm_page BUSY.
 */
static __inline int
relock_map(struct faultstate *fs)
{
        int error;

        if (fs->lookup_still_valid == FALSE && fs->map) {
                error = vm_map_lock_read_to(fs->map);
                if (error == 0)
                        fs->lookup_still_valid = TRUE;
        } else {
                error = 0;
        }
        return error;
}

static __inline void
unlock_map(struct faultstate *fs)
{
        if (fs->lookup_still_valid && fs->map) {
                vm_map_lookup_done(fs->map, fs->entry, 0);
                fs->lookup_still_valid = FALSE;
        }
}

/*
 * Clean up after a successful call to vm_fault_object() so another call
 * to vm_fault_object() can be made.
 */
static void
_cleanup_successful_fault(struct faultstate *fs, int relock)
{
        /*
         * We allocated a junk page for a COW operation that did
         * not occur, the page must be freed.
         */
        if (fs->object != fs->first_object) {
                KKASSERT(fs->first_shared == 0);
                vm_page_free(fs->first_m);
                vm_object_pip_wakeup(fs->object);
                fs->first_m = NULL;
        }

        /*
         * Reset fs->object.
         */
        fs->object = fs->first_object;
        if (relock && fs->lookup_still_valid == FALSE) {
                if (fs->map)
                        vm_map_lock_read(fs->map);
                fs->lookup_still_valid = TRUE;
        }
}

static void
_unlock_things(struct faultstate *fs, int dealloc)
{
        _cleanup_successful_fault(fs, 0);
        if (dealloc) {
                /*vm_object_deallocate(fs->first_object);*/
                /*fs->first_object = NULL; drop used later on */
        }
        unlock_map(fs); 
        if (fs->vp != NULL) { 
                vput(fs->vp);
                fs->vp = NULL;
        }
}

#define unlock_things(fs) _unlock_things(fs, 0)
#define unlock_and_deallocate(fs) _unlock_things(fs, 1)
#define cleanup_successful_fault(fs) _cleanup_successful_fault(fs, 1)

/*
 * TRYPAGER 
 *
 * Determine if the pager for the current object *might* contain the page.
 *
 * We only need to try the pager if this is not a default object (default
 * objects are zero-fill and have no real pager), and if we are not taking
 * a wiring fault or if the FS entry is wired.
 */
#define TRYPAGER(fs)    \
                (fs->object->type != OBJT_DEFAULT && \
                (((fs->fault_flags & VM_FAULT_WIRE_MASK) == 0) || fs->wired))

/*
 * vm_fault:
 *
 * Handle a page fault occuring at the given address, requiring the given
 * permissions, in the map specified.  If successful, the page is inserted
 * into the associated physical map.
 *
 * NOTE: The given address should be truncated to the proper page address.
 *
 * KERN_SUCCESS is returned if the page fault is handled; otherwise,
 * a standard error specifying why the fault is fatal is returned.
 *
 * The map in question must be referenced, and remains so.
 * The caller may hold no locks.
 * No other requirements.
 */
int
vm_fault(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type, int fault_flags)
{
        int result;
        vm_pindex_t first_pindex;
        struct faultstate fs;
        struct lwp *lp;
        int growstack;
        int retry = 0;

        vm_page_pcpu_cache();
        fs.hardfault = 0;
        fs.fault_flags = fault_flags;
        fs.vp = NULL;
        fs.shared = vm_shared_fault;
        fs.first_shared = vm_shared_fault;
        growstack = 1;
        if (vm_shared_fault)
                ++vm_shared_count;

        /*
         * vm_map interactions
         */
        if ((lp = curthread->td_lwp) != NULL)
                lp->lwp_flags |= LWP_PAGING;
        lwkt_gettoken(&map->token);

RetryFault:
        /*
         * Find the vm_map_entry representing the backing store and resolve
         * the top level object and page index.  This may have the side
         * effect of executing a copy-on-write on the map entry and/or
         * creating a shadow object, but will not COW any actual VM pages.
         *
         * On success fs.map is left read-locked and various other fields 
         * are initialized but not otherwise referenced or locked.
         *
         * NOTE!  vm_map_lookup will try to upgrade the fault_type to
         * VM_FAULT_WRITE if the map entry is a virtual page table and also
         * writable, so we can set the 'A'accessed bit in the virtual page
         * table entry.
         */
        fs.map = map;
        result = vm_map_lookup(&fs.map, vaddr, fault_type,
                               &fs.entry, &fs.first_object,
                               &first_pindex, &fs.first_prot, &fs.wired);

        /*
         * If the lookup failed or the map protections are incompatible,
         * the fault generally fails.  However, if the caller is trying
         * to do a user wiring we have more work to do.
         */
        if (result != KERN_SUCCESS) {
                if (result != KERN_PROTECTION_FAILURE ||
                    (fs.fault_flags & VM_FAULT_WIRE_MASK) != VM_FAULT_USER_WIRE)
                {
                        if (result == KERN_INVALID_ADDRESS && growstack &&
                            map != &kernel_map && curproc != NULL) {
                                result = vm_map_growstack(curproc, vaddr);
                                if (result == KERN_SUCCESS) {
                                        growstack = 0;
                                        ++retry;
                                        goto RetryFault;
                                }
                                result = KERN_FAILURE;
                        }
                        goto done;
                }

                /*
                 * If we are user-wiring a r/w segment, and it is COW, then
                 * we need to do the COW operation.  Note that we don't
                 * currently COW RO sections now, because it is NOT desirable
                 * to COW .text.  We simply keep .text from ever being COW'ed
                 * and take the heat that one cannot debug wired .text sections.
                 */
                result = vm_map_lookup(&fs.map, vaddr,
                                       VM_PROT_READ|VM_PROT_WRITE|
                                        VM_PROT_OVERRIDE_WRITE,
                                       &fs.entry, &fs.first_object,
                                       &first_pindex, &fs.first_prot,
                                       &fs.wired);
                if (result != KERN_SUCCESS) {
                        result = KERN_FAILURE;
                        goto done;
                }

                /*
                 * If we don't COW now, on a user wire, the user will never
                 * be able to write to the mapping.  If we don't make this
                 * restriction, the bookkeeping would be nearly impossible.
                 *
                 * XXX We have a shared lock, this will have a MP race but
                 * I don't see how it can hurt anything.
                 */
                if ((fs.entry->protection & VM_PROT_WRITE) == 0)
                        fs.entry->max_protection &= ~VM_PROT_WRITE;
        }

        /*
         * fs.map is read-locked
         *
         * Misc checks.  Save the map generation number to detect races.
         */
        fs.map_generation = fs.map->timestamp;
        fs.lookup_still_valid = TRUE;
        fs.first_m = NULL;
        fs.object = fs.first_object;    /* so unlock_and_deallocate works */

        if (fs.entry->eflags & (MAP_ENTRY_NOFAULT | MAP_ENTRY_KSTACK)) {
                if (fs.entry->eflags & MAP_ENTRY_NOFAULT) {
                        panic("vm_fault: fault on nofault entry, addr: %p",
                              (void *)vaddr);
                }
                if ((fs.entry->eflags & MAP_ENTRY_KSTACK) &&
                    vaddr >= fs.entry->start &&
                    vaddr < fs.entry->start + PAGE_SIZE) {
                        panic("vm_fault: fault on stack guard, addr: %p",
                              (void *)vaddr);
                }
        }

        /*
         * A system map entry may return a NULL object.  No object means
         * no pager means an unrecoverable kernel fault.
         */
        if (fs.first_object == NULL) {
                panic("vm_fault: unrecoverable fault at %p in entry %p",
                        (void *)vaddr, fs.entry);
        }

        /*
         * Fail here if not a trivial anonymous page fault and TDF_NOFAULT
         * is set.
         */
        if ((curthread->td_flags & TDF_NOFAULT) &&
            (retry ||
             fs.first_object->type == OBJT_VNODE ||
             fs.first_object->backing_object)) {
                result = KERN_FAILURE;
                unlock_things(&fs);
                goto done2;
        }

        /*
         * If the entry is wired we cannot change the page protection.
         */
        if (fs.wired)
                fault_type = fs.first_prot;

        /*
         * We generally want to avoid unnecessary exclusive modes on backing
         * and terminal objects because this can seriously interfere with
         * heavily fork()'d processes (particularly /bin/sh scripts).
         *
         * However, we also want to avoid unnecessary retries due to needed
         * shared->exclusive promotion for common faults.  Exclusive mode is
         * always needed if any page insertion, rename, or free occurs in an
         * object (and also indirectly if any I/O is done).
         *
         * The main issue here is going to be fs.first_shared.  If the
         * first_object has a backing object which isn't shadowed and the
         * process is single-threaded we might as well use an exclusive
         * lock/chain right off the bat.
         */
        if (fs.first_shared && fs.first_object->backing_object &&
            LIST_EMPTY(&fs.first_object->shadow_head) &&
            curthread->td_proc && curthread->td_proc->p_nthreads == 1) {
                fs.first_shared = 0;
        }


        /*
         * Obtain a top-level object lock, shared or exclusive depending
         * on fs.first_shared.  If a shared lock winds up being insufficient
         * we will retry with an exclusive lock.
         *
         * The vnode pager lock is always shared.
         */
        if (fs.first_shared)
                vm_object_hold_shared(fs.first_object);
        else
                vm_object_hold(fs.first_object);
        if (fs.vp == NULL)
                fs.vp = vnode_pager_lock(fs.first_object);

        /*
         * The page we want is at (first_object, first_pindex), but if the
         * vm_map_entry is VM_MAPTYPE_VPAGETABLE we have to traverse the
         * page table to figure out the actual pindex.
         *
         * NOTE!  DEVELOPMENT IN PROGRESS, THIS IS AN INITIAL IMPLEMENTATION
         * ONLY
         */
        if (fs.entry->maptype == VM_MAPTYPE_VPAGETABLE) {
                result = vm_fault_vpagetable(&fs, &first_pindex,
                                             fs.entry->aux.master_pde,
                                             fault_type, 1);
                if (result == KERN_TRY_AGAIN) {
                        vm_object_drop(fs.first_object);
                        ++retry;
                        goto RetryFault;
                }
                if (result != KERN_SUCCESS)
                        goto done;
        }

        /*
         * Now we have the actual (object, pindex), fault in the page.  If
         * vm_fault_object() fails it will unlock and deallocate the FS
         * data.   If it succeeds everything remains locked and fs->object
         * will have an additional PIP count if it is not equal to
         * fs->first_object
         *
         * vm_fault_object will set fs->prot for the pmap operation.  It is
         * allowed to set VM_PROT_WRITE if fault_type == VM_PROT_READ if the
         * page can be safely written.  However, it will force a read-only
         * mapping for a read fault if the memory is managed by a virtual
         * page table.
         *
         * If the fault code uses the shared object lock shortcut
         * we must not try to burst (we can't allocate VM pages).
         */
        result = vm_fault_object(&fs, first_pindex, fault_type, 1);
        if (result == KERN_TRY_AGAIN) {
                vm_object_drop(fs.first_object);
                ++retry;
                goto RetryFault;
        }
        if (result != KERN_SUCCESS)
                goto done;

        /*
         * On success vm_fault_object() does not unlock or deallocate, and fs.m
         * will contain a busied page.
         *
         * Enter the page into the pmap and do pmap-related adjustments.
         */
        KKASSERT(fs.lookup_still_valid == TRUE);
        vm_page_flag_set(fs.m, PG_REFERENCED);
        pmap_enter(fs.map->pmap, vaddr, fs.m, fs.prot, fs.wired, fs.entry);
        mycpu->gd_cnt.v_vm_faults++;
        if (curthread->td_lwp)
                ++curthread->td_lwp->lwp_ru.ru_minflt;

        /*KKASSERT(fs.m->queue == PQ_NONE); page-in op may deactivate page */
        KKASSERT(fs.m->flags & PG_BUSY);

        /*
         * If the page is not wired down, then put it where the pageout daemon
         * can find it.
         */
        if (fs.fault_flags & VM_FAULT_WIRE_MASK) {
                if (fs.wired)
                        vm_page_wire(fs.m);
                else
                        vm_page_unwire(fs.m, 1);
        } else {
                vm_page_activate(fs.m);
        }
        vm_page_wakeup(fs.m);

        /*
         * Burst in a few more pages if possible.  The fs.map should still
         * be locked.  To avoid interlocking against a vnode->getblk
         * operation we had to be sure to unbusy our primary vm_page above
         * first.
         *
         * A normal burst can continue down backing store, only execute
         * if we are holding an exclusive lock, otherwise the exclusive
         * locks the burst code gets might cause excessive SMP collisions.
         *
         * A quick burst can be utilized when there is no backing object
         * (i.e. a shared file mmap).
         */
        if ((fault_flags & VM_FAULT_BURST) &&
            (fs.fault_flags & VM_FAULT_WIRE_MASK) == 0 &&
            fs.wired == 0) {
                if (fs.first_shared == 0 && fs.shared == 0) {
                        vm_prefault(fs.map->pmap, vaddr,
                                    fs.entry, fs.prot, fault_flags);
                } else {
                        vm_prefault_quick(fs.map->pmap, vaddr,
                                          fs.entry, fs.prot, fault_flags);
                }
        }

        /*
         * Unlock everything, and return
         */
        unlock_things(&fs);

        if (curthread->td_lwp) {
                if (fs.hardfault) {
                        curthread->td_lwp->lwp_ru.ru_majflt++;
                } else {
                        curthread->td_lwp->lwp_ru.ru_minflt++;
                }
        }

        /*vm_object_deallocate(fs.first_object);*/
        /*fs.m = NULL; */
        /*fs.first_object = NULL; must still drop later */

        result = KERN_SUCCESS;
done:
        if (fs.first_object)
                vm_object_drop(fs.first_object);
done2:
        lwkt_reltoken(&map->token);
        if (lp)
                lp->lwp_flags &= ~LWP_PAGING;
        if (vm_shared_fault && fs.shared == 0)
                ++vm_shared_miss;
        return (result);
}

/*
 * Fault in the specified virtual address in the current process map, 
 * returning a held VM page or NULL.  See vm_fault_page() for more 
 * information.
 *
 * No requirements.
 */
vm_page_t
vm_fault_page_quick(vm_offset_t va, vm_prot_t fault_type, int *errorp)
{
        struct lwp *lp = curthread->td_lwp;
        vm_page_t m;

        m = vm_fault_page(&lp->lwp_vmspace->vm_map, va, 
                          fault_type, VM_FAULT_NORMAL, errorp);
        return(m);
}

/*
 * Fault in the specified virtual address in the specified map, doing all
 * necessary manipulation of the object store and all necessary I/O.  Return
 * a held VM page or NULL, and set *errorp.  The related pmap is not
 * updated.
 *
 * The returned page will be properly dirtied if VM_PROT_WRITE was specified,
 * and marked PG_REFERENCED as well.
 *
 * If the page cannot be faulted writable and VM_PROT_WRITE was specified, an
 * error will be returned.
 *
 * No requirements.
 */
vm_page_t
vm_fault_page(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
              int fault_flags, int *errorp)
{
        vm_pindex_t first_pindex;
        struct faultstate fs;
        int result;
        int retry = 0;
        vm_prot_t orig_fault_type = fault_type;

        fs.hardfault = 0;
        fs.fault_flags = fault_flags;
        KKASSERT((fault_flags & VM_FAULT_WIRE_MASK) == 0);

        /*
         * Dive the pmap (concurrency possible).  If we find the
         * appropriate page we can terminate early and quickly.
         */
        fs.m = pmap_fault_page_quick(map->pmap, vaddr, fault_type);
        if (fs.m) {
                *errorp = 0;
                return(fs.m);
        }

        /*
         * Otherwise take a concurrency hit and do a formal page
         * fault.
         */
        fs.shared = vm_shared_fault;
        fs.first_shared = vm_shared_fault;
        fs.vp = NULL;
        lwkt_gettoken(&map->token);

RetryFault:
        /*
         * Find the vm_map_entry representing the backing store and resolve
         * the top level object and page index.  This may have the side
         * effect of executing a copy-on-write on the map entry and/or
         * creating a shadow object, but will not COW any actual VM pages.
         *
         * On success fs.map is left read-locked and various other fields 
         * are initialized but not otherwise referenced or locked.
         *
         * NOTE!  vm_map_lookup will upgrade the fault_type to VM_FAULT_WRITE
         * if the map entry is a virtual page table and also writable,
         * so we can set the 'A'accessed bit in the virtual page table entry.
         */
        fs.map = map;
        result = vm_map_lookup(&fs.map, vaddr, fault_type,
                               &fs.entry, &fs.first_object,
                               &first_pindex, &fs.first_prot, &fs.wired);

        if (result != KERN_SUCCESS) {
                *errorp = result;
                fs.m = NULL;
                goto done;
        }

        /*
         * fs.map is read-locked
         *
         * Misc checks.  Save the map generation number to detect races.
         */
        fs.map_generation = fs.map->timestamp;
        fs.lookup_still_valid = TRUE;
        fs.first_m = NULL;
        fs.object = fs.first_object;    /* so unlock_and_deallocate works */

        if (fs.entry->eflags & MAP_ENTRY_NOFAULT) {
                panic("vm_fault: fault on nofault entry, addr: %lx",
                    (u_long)vaddr);
        }

        /*
         * A system map entry may return a NULL object.  No object means
         * no pager means an unrecoverable kernel fault.
         */
        if (fs.first_object == NULL) {
                panic("vm_fault: unrecoverable fault at %p in entry %p",
                        (void *)vaddr, fs.entry);
        }

        /*
         * Fail here if not a trivial anonymous page fault and TDF_NOFAULT
         * is set.
         */
        if ((curthread->td_flags & TDF_NOFAULT) &&
            (retry ||
             fs.first_object->type == OBJT_VNODE ||
             fs.first_object->backing_object)) {
                *errorp = KERN_FAILURE;
                unlock_things(&fs);
                goto done2;
        }

        /*
         * If the entry is wired we cannot change the page protection.
         */
        if (fs.wired)
                fault_type = fs.first_prot;

        /*
         * Make a reference to this object to prevent its disposal while we
         * are messing with it.  Once we have the reference, the map is free
         * to be diddled.  Since objects reference their shadows (and copies),
         * they will stay around as well.
         *
         * The reference should also prevent an unexpected collapse of the
         * parent that might move pages from the current object into the
         * parent unexpectedly, resulting in corruption.
         *
         * Bump the paging-in-progress count to prevent size changes (e.g.
         * truncation operations) during I/O.  This must be done after
         * obtaining the vnode lock in order to avoid possible deadlocks.
         */
        if (fs.first_shared)
                vm_object_hold_shared(fs.first_object);
        else
                vm_object_hold(fs.first_object);
        if (fs.vp == NULL)
                fs.vp = vnode_pager_lock(fs.first_object);      /* shared */

        /*
         * The page we want is at (first_object, first_pindex), but if the
         * vm_map_entry is VM_MAPTYPE_VPAGETABLE we have to traverse the
         * page table to figure out the actual pindex.
         *
         * NOTE!  DEVELOPMENT IN PROGRESS, THIS IS AN INITIAL IMPLEMENTATION
         * ONLY
         */
        if (fs.entry->maptype == VM_MAPTYPE_VPAGETABLE) {
                result = vm_fault_vpagetable(&fs, &first_pindex,
                                             fs.entry->aux.master_pde,
                                             fault_type, 1);
                if (result == KERN_TRY_AGAIN) {
                        vm_object_drop(fs.first_object);
                        ++retry;
                        goto RetryFault;
                }
                if (result != KERN_SUCCESS) {
                        *errorp = result;
                        fs.m = NULL;
                        goto done;
                }
        }

        /*
         * Now we have the actual (object, pindex), fault in the page.  If
         * vm_fault_object() fails it will unlock and deallocate the FS
         * data.   If it succeeds everything remains locked and fs->object
         * will have an additinal PIP count if it is not equal to
         * fs->first_object
         */
        fs.m = NULL;
        result = vm_fault_object(&fs, first_pindex, fault_type, 1);

        if (result == KERN_TRY_AGAIN) {
                vm_object_drop(fs.first_object);
                ++retry;
                goto RetryFault;
        }
        if (result != KERN_SUCCESS) {
                *errorp = result;
                fs.m = NULL;
                goto done;
        }

        if ((orig_fault_type & VM_PROT_WRITE) &&
            (fs.prot & VM_PROT_WRITE) == 0) {
                *errorp = KERN_PROTECTION_FAILURE;
                unlock_and_deallocate(&fs);
                fs.m = NULL;
                goto done;
        }

        /*
         * DO NOT UPDATE THE PMAP!!!  This function may be called for
         * a pmap unrelated to the current process pmap, in which case
         * the current cpu core will not be listed in the pmap's pm_active
         * mask.  Thus invalidation interlocks will fail to work properly.
         *
         * (for example, 'ps' uses procfs to read program arguments from
         * each process's stack).
         *
         * In addition to the above this function will be called to acquire
         * a page that might already be faulted in, re-faulting it
         * continuously is a waste of time.
         *
         * XXX could this have been the cause of our random seg-fault
         *     issues?  procfs accesses user stacks.
         */
        vm_page_flag_set(fs.m, PG_REFERENCED);
#if 0
        pmap_enter(fs.map->pmap, vaddr, fs.m, fs.prot, fs.wired, NULL);
        mycpu->gd_cnt.v_vm_faults++;
        if (curthread->td_lwp)
                ++curthread->td_lwp->lwp_ru.ru_minflt;
#endif

        /*
         * On success vm_fault_object() does not unlock or deallocate, and fs.m
         * will contain a busied page.  So we must unlock here after having
         * messed with the pmap.
         */
        unlock_things(&fs);

        /*
         * Return a held page.  We are not doing any pmap manipulation so do
         * not set PG_MAPPED.  However, adjust the page flags according to
         * the fault type because the caller may not use a managed pmapping
         * (so we don't want to lose the fact that the page will be dirtied
         * if a write fault was specified).
         */
        vm_page_hold(fs.m);
        vm_page_activate(fs.m);
        if (fault_type & VM_PROT_WRITE)
                vm_page_dirty(fs.m);

        if (curthread->td_lwp) {
                if (fs.hardfault) {
                        curthread->td_lwp->lwp_ru.ru_majflt++;
                } else {
                        curthread->td_lwp->lwp_ru.ru_minflt++;
                }
        }

        /*
         * Unlock everything, and return the held page.
         */
        vm_page_wakeup(fs.m);
        /*vm_object_deallocate(fs.first_object);*/
        /*fs.first_object = NULL; */
        *errorp = 0;

done:
        if (fs.first_object)
                vm_object_drop(fs.first_object);
done2:
        lwkt_reltoken(&map->token);
        return(fs.m);
}

/*
 * Fault in the specified (object,offset), dirty the returned page as
 * needed.  If the requested fault_type cannot be done NULL and an
 * error is returned.
 *
 * A held (but not busied) page is returned.
 *
 * The passed in object must be held as specified by the shared
 * argument.
 */
vm_page_t
vm_fault_object_page(vm_object_t object, vm_ooffset_t offset,
                     vm_prot_t fault_type, int fault_flags,
                     int *sharedp, int *errorp)
{
        int result;
        vm_pindex_t first_pindex;
        struct faultstate fs;
        struct vm_map_entry entry;

        ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
        bzero(&entry, sizeof(entry));
        entry.object.vm_object = object;
        entry.maptype = VM_MAPTYPE_NORMAL;
        entry.protection = entry.max_protection = fault_type;

        fs.hardfault = 0;
        fs.fault_flags = fault_flags;
        fs.map = NULL;
        fs.shared = vm_shared_fault;
        fs.first_shared = *sharedp;
        fs.vp = NULL;
        KKASSERT((fault_flags & VM_FAULT_WIRE_MASK) == 0);

RetryFault:
        *sharedp = fs.first_shared;
        first_pindex = OFF_TO_IDX(offset);
        fs.first_object = object;
        fs.entry = &entry;
        fs.first_prot = fault_type;
        fs.wired = 0;
        /*fs.map_generation = 0; unused */

        /*
         * Make a reference to this object to prevent its disposal while we
         * are messing with it.  Once we have the reference, the map is free
         * to be diddled.  Since objects reference their shadows (and copies),
         * they will stay around as well.
         *
         * The reference should also prevent an unexpected collapse of the
         * parent that might move pages from the current object into the
         * parent unexpectedly, resulting in corruption.
         *
         * Bump the paging-in-progress count to prevent size changes (e.g.
         * truncation operations) during I/O.  This must be done after
         * obtaining the vnode lock in order to avoid possible deadlocks.
         */
        if (fs.vp == NULL)
                fs.vp = vnode_pager_lock(fs.first_object);

        fs.lookup_still_valid = TRUE;
        fs.first_m = NULL;
        fs.object = fs.first_object;    /* so unlock_and_deallocate works */

#if 0
        /* XXX future - ability to operate on VM object using vpagetable */
        if (fs.entry->maptype == VM_MAPTYPE_VPAGETABLE) {
                result = vm_fault_vpagetable(&fs, &first_pindex,
                                             fs.entry->aux.master_pde,
                                             fault_type, 0);
                if (result == KERN_TRY_AGAIN) {
                        if (fs.first_shared == 0 && *sharedp)
                                vm_object_upgrade(object);
                        goto RetryFault;
                }
                if (result != KERN_SUCCESS) {
                        *errorp = result;
                        return (NULL);
                }
        }
#endif

        /*
         * Now we have the actual (object, pindex), fault in the page.  If
         * vm_fault_object() fails it will unlock and deallocate the FS
         * data.   If it succeeds everything remains locked and fs->object
         * will have an additinal PIP count if it is not equal to
         * fs->first_object
         *
         * On KERN_TRY_AGAIN vm_fault_object() leaves fs.first_object intact.
         * We may have to upgrade its lock to handle the requested fault.
         */
        result = vm_fault_object(&fs, first_pindex, fault_type, 0);

        if (result == KERN_TRY_AGAIN) {
                if (fs.first_shared == 0 && *sharedp)
                        vm_object_upgrade(object);
                goto RetryFault;
        }
        if (result != KERN_SUCCESS) {
                *errorp = result;
                return(NULL);
        }

        if ((fault_type & VM_PROT_WRITE) && (fs.prot & VM_PROT_WRITE) == 0) {
                *errorp = KERN_PROTECTION_FAILURE;
                unlock_and_deallocate(&fs);
                return(NULL);
        }

        /*
         * On success vm_fault_object() does not unlock or deallocate, so we
         * do it here.  Note that the returned fs.m will be busied.
         */
        unlock_things(&fs);

        /*
         * Return a held page.  We are not doing any pmap manipulation so do
         * not set PG_MAPPED.  However, adjust the page flags according to
         * the fault type because the caller may not use a managed pmapping
         * (so we don't want to lose the fact that the page will be dirtied
         * if a write fault was specified).
         */
        vm_page_hold(fs.m);
        vm_page_activate(fs.m);
        if ((fault_type & VM_PROT_WRITE) || (fault_flags & VM_FAULT_DIRTY))
                vm_page_dirty(fs.m);
        if (fault_flags & VM_FAULT_UNSWAP)
                swap_pager_unswapped(fs.m);

        /*
         * Indicate that the page was accessed.
         */
        vm_page_flag_set(fs.m, PG_REFERENCED);

        if (curthread->td_lwp) {
                if (fs.hardfault) {
                        curthread->td_lwp->lwp_ru.ru_majflt++;
                } else {
                        curthread->td_lwp->lwp_ru.ru_minflt++;
                }
        }

        /*
         * Unlock everything, and return the held page.
         */
        vm_page_wakeup(fs.m);
        /*vm_object_deallocate(fs.first_object);*/
        /*fs.first_object = NULL; */

        *errorp = 0;
        return(fs.m);
}

/*
 * Translate the virtual page number (first_pindex) that is relative
 * to the address space into a logical page number that is relative to the
 * backing object.  Use the virtual page table pointed to by (vpte).
 *
 * This implements an N-level page table.  Any level can terminate the
 * scan by setting VPTE_PS.   A linear mapping is accomplished by setting
 * VPTE_PS in the master page directory entry set via mcontrol(MADV_SETMAP).
 */
static
int
vm_fault_vpagetable(struct faultstate *fs, vm_pindex_t *pindex,
                    vpte_t vpte, int fault_type, int allow_nofault)
{
        struct lwbuf *lwb;
        struct lwbuf lwb_cache;
        int vshift = VPTE_FRAME_END - PAGE_SHIFT; /* index bits remaining */
        int result = KERN_SUCCESS;
        vpte_t *ptep;

        ASSERT_LWKT_TOKEN_HELD(vm_object_token(fs->first_object));
        for (;;) {
                /*
                 * We cannot proceed if the vpte is not valid, not readable
                 * for a read fault, or not writable for a write fault.
                 */
                if ((vpte & VPTE_V) == 0) {
                        unlock_and_deallocate(fs);
                        return (KERN_FAILURE);
                }
                if ((fault_type & VM_PROT_WRITE) && (vpte & VPTE_RW) == 0) {
                        unlock_and_deallocate(fs);
                        return (KERN_FAILURE);
                }
                if ((vpte & VPTE_PS) || vshift == 0)
                        break;
                KKASSERT(vshift >= VPTE_PAGE_BITS);

                /*
                 * Get the page table page.  Nominally we only read the page
                 * table, but since we are actively setting VPTE_M and VPTE_A,
                 * tell vm_fault_object() that we are writing it. 
                 *
                 * There is currently no real need to optimize this.
                 */
                result = vm_fault_object(fs, (vpte & VPTE_FRAME) >> PAGE_SHIFT,
                                         VM_PROT_READ|VM_PROT_WRITE,
                                         allow_nofault);
                if (result != KERN_SUCCESS)
                        return (result);

                /*
                 * Process the returned fs.m and look up the page table
                 * entry in the page table page.
                 */
                vshift -= VPTE_PAGE_BITS;
                lwb = lwbuf_alloc(fs->m, &lwb_cache);
                ptep = ((vpte_t *)lwbuf_kva(lwb) +
                        ((*pindex >> vshift) & VPTE_PAGE_MASK));
                vpte = *ptep;

                /*
                 * Page table write-back.  If the vpte is valid for the
                 * requested operation, do a write-back to the page table.
                 *
                 * XXX VPTE_M is not set properly for page directory pages.
                 * It doesn't get set in the page directory if the page table
                 * is modified during a read access.
                 */
                vm_page_activate(fs->m);
                if ((fault_type & VM_PROT_WRITE) && (vpte & VPTE_V) &&
                    (vpte & VPTE_RW)) {
                        if ((vpte & (VPTE_M|VPTE_A)) != (VPTE_M|VPTE_A)) {
                                atomic_set_long(ptep, VPTE_M | VPTE_A);
                                vm_page_dirty(fs->m);
                        }
                }
                if ((fault_type & VM_PROT_READ) && (vpte & VPTE_V)) {
                        if ((vpte & VPTE_A) == 0) {
                                atomic_set_long(ptep, VPTE_A);
                                vm_page_dirty(fs->m);
                        }
                }
                lwbuf_free(lwb);
                vm_page_flag_set(fs->m, PG_REFERENCED);
                vm_page_wakeup(fs->m);
                fs->m = NULL;
                cleanup_successful_fault(fs);
        }
        /*
         * Combine remaining address bits with the vpte.
         */
        /* JG how many bits from each? */
        *pindex = ((vpte & VPTE_FRAME) >> PAGE_SHIFT) +
                  (*pindex & ((1L << vshift) - 1));
        return (KERN_SUCCESS);
}


/*
 * This is the core of the vm_fault code.
 *
 * Do all operations required to fault-in (fs.first_object, pindex).  Run
 * through the shadow chain as necessary and do required COW or virtual
 * copy operations.  The caller has already fully resolved the vm_map_entry
 * and, if appropriate, has created a copy-on-write layer.  All we need to
 * do is iterate the object chain.
 *
 * On failure (fs) is unlocked and deallocated and the caller may return or
 * retry depending on the failure code.  On success (fs) is NOT unlocked or
 * deallocated, fs.m will contained a resolved, busied page, and fs.object
 * will have an additional PIP count if it is not equal to fs.first_object.
 *
 * If locks based on fs->first_shared or fs->shared are insufficient,
 * clear the appropriate field(s) and return RETRY.  COWs require that
 * first_shared be 0, while page allocations (or frees) require that
 * shared be 0.  Renames require that both be 0.
 *
 * fs->first_object must be held on call.
 */
static
int
vm_fault_object(struct faultstate *fs, vm_pindex_t first_pindex,
                vm_prot_t fault_type, int allow_nofault)
{
        vm_object_t next_object;
        vm_pindex_t pindex;
        int error;

        ASSERT_LWKT_TOKEN_HELD(vm_object_token(fs->first_object));
        fs->prot = fs->first_prot;
        fs->object = fs->first_object;
        pindex = first_pindex;

        vm_object_chain_acquire(fs->first_object, fs->shared);
        vm_object_pip_add(fs->first_object, 1);

        /* 
         * If a read fault occurs we try to make the page writable if
         * possible.  There are three cases where we cannot make the
         * page mapping writable:
         *
         * (1) The mapping is read-only or the VM object is read-only,
         *     fs->prot above will simply not have VM_PROT_WRITE set.
         *
         * (2) If the mapping is a virtual page table we need to be able
         *     to detect writes so we can set VPTE_M in the virtual page
         *     table.
         *
         * (3) If the VM page is read-only or copy-on-write, upgrading would
         *     just result in an unnecessary COW fault.
         *
         * VM_PROT_VPAGED is set if faulting via a virtual page table and
         * causes adjustments to the 'M'odify bit to also turn off write
         * access to force a re-fault.
         */
        if (fs->entry->maptype == VM_MAPTYPE_VPAGETABLE) {
                if ((fault_type & VM_PROT_WRITE) == 0)
                        fs->prot &= ~VM_PROT_WRITE;
        }

        if (curthread->td_lwp && curthread->td_lwp->lwp_vmspace &&
            pmap_emulate_ad_bits(&curthread->td_lwp->lwp_vmspace->vm_pmap)) {
                if ((fault_type & VM_PROT_WRITE) == 0)
                        fs->prot &= ~VM_PROT_WRITE;
        }

        /* vm_object_hold(fs->object); implied b/c object == first_object */

        for (;;) {
                /*
                 * The entire backing chain from first_object to object
                 * inclusive is chainlocked.
                 *
                 * If the object is dead, we stop here
                 */
                if (fs->object->flags & OBJ_DEAD) {
                        vm_object_pip_wakeup(fs->first_object);
                        vm_object_chain_release_all(fs->first_object,
                                                    fs->object);
                        if (fs->object != fs->first_object)
                                vm_object_drop(fs->object);
                        unlock_and_deallocate(fs);
                        return (KERN_PROTECTION_FAILURE);
                }

                /*
                 * See if the page is resident.  Wait/Retry if the page is
                 * busy (lots of stuff may have changed so we can't continue
                 * in that case).
                 *
                 * We can theoretically allow the soft-busy case on a read
                 * fault if the page is marked valid, but since such
                 * pages are typically already pmap'd, putting that
                 * special case in might be more effort then it is
                 * worth.  We cannot under any circumstances mess
                 * around with a vm_page_t->busy page except, perhaps,
                 * to pmap it.
                 */
                fs->m = vm_page_lookup_busy_try(fs->object, pindex,
                                                TRUE, &error);
                if (error) {
                        vm_object_pip_wakeup(fs->first_object);
                        vm_object_chain_release_all(fs->first_object,
                                                    fs->object);
                        if (fs->object != fs->first_object)
                                vm_object_drop(fs->object);
                        unlock_things(fs);
                        vm_page_sleep_busy(fs->m, TRUE, "vmpfw");
                        mycpu->gd_cnt.v_intrans++;
                        /*vm_object_deallocate(fs->first_object);*/
                        /*fs->first_object = NULL;*/
                        fs->m = NULL;
                        return (KERN_TRY_AGAIN);
                }
                if (fs->m) {
                        /*
                         * The page is busied for us.
                         *
                         * If reactivating a page from PQ_CACHE we may have
                         * to rate-limit.
                         */
                        int queue = fs->m->queue;
                        vm_page_unqueue_nowakeup(fs->m);

                        if ((queue - fs->m->pc) == PQ_CACHE && 
                            vm_page_count_severe()) {
                                vm_page_activate(fs->m);
                                vm_page_wakeup(fs->m);
                                fs->m = NULL;
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                if (allow_nofault == 0 ||
                                    (curthread->td_flags & TDF_NOFAULT) == 0) {
                                        vm_wait_pfault();
                                }
                                return (KERN_TRY_AGAIN);
                        }

                        /*
                         * If it still isn't completely valid (readable),
                         * or if a read-ahead-mark is set on the VM page,
                         * jump to readrest, else we found the page and
                         * can return.
                         *
                         * We can release the spl once we have marked the
                         * page busy.
                         */
                        if (fs->m->object != &kernel_object) {
                                if ((fs->m->valid & VM_PAGE_BITS_ALL) !=
                                    VM_PAGE_BITS_ALL) {
                                        goto readrest;
                                }
                                if (fs->m->flags & PG_RAM) {
                                        if (debug_cluster)
                                                kprintf("R");
                                        vm_page_flag_clear(fs->m, PG_RAM);
                                        goto readrest;
                                }
                        }
                        break; /* break to PAGE HAS BEEN FOUND */
                }

                /*
                 * Page is not resident, If this is the search termination
                 * or the pager might contain the page, allocate a new page.
                 */
                if (TRYPAGER(fs) || fs->object == fs->first_object) {
                        /*
                         * Allocating, must be exclusive.
                         */
                        if (fs->object == fs->first_object &&
                            fs->first_shared) {
                                fs->first_shared = 0;
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                return (KERN_TRY_AGAIN);
                        }
                        if (fs->object != fs->first_object &&
                            fs->shared) {
                                fs->first_shared = 0;
                                fs->shared = 0;
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                return (KERN_TRY_AGAIN);
                        }

                        /*
                         * If the page is beyond the object size we fail
                         */
                        if (pindex >= fs->object->size) {
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                return (KERN_PROTECTION_FAILURE);
                        }

                        /*
                         * Allocate a new page for this object/offset pair.
                         *
                         * It is possible for the allocation to race, so
                         * handle the case.
                         */
                        fs->m = NULL;
                        if (!vm_page_count_severe()) {
                                fs->m = vm_page_alloc(fs->object, pindex,
                                    ((fs->vp || fs->object->backing_object) ?
                                        VM_ALLOC_NULL_OK | VM_ALLOC_NORMAL :
                                        VM_ALLOC_NULL_OK | VM_ALLOC_NORMAL |
                                        VM_ALLOC_USE_GD | VM_ALLOC_ZERO));
                        }
                        if (fs->m == NULL) {
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                if (allow_nofault == 0 ||
                                    (curthread->td_flags & TDF_NOFAULT) == 0) {
                                        vm_wait_pfault();
                                }
                                return (KERN_TRY_AGAIN);
                        }

                        /*
                         * Fall through to readrest.  We have a new page which
                         * will have to be paged (since m->valid will be 0).
                         */
                }

readrest:
                /*
                 * We have found an invalid or partially valid page, a
                 * page with a read-ahead mark which might be partially or
                 * fully valid (and maybe dirty too), or we have allocated
                 * a new page.
                 *
                 * Attempt to fault-in the page if there is a chance that the
                 * pager has it, and potentially fault in additional pages
                 * at the same time.
                 *
                 * If TRYPAGER is true then fs.m will be non-NULL and busied
                 * for us.
                 */
                if (TRYPAGER(fs)) {
                        int rv;
                        int seqaccess;
                        u_char behavior = vm_map_entry_behavior(fs->entry);

                        if (behavior == MAP_ENTRY_BEHAV_RANDOM)
                                seqaccess = 0;
                        else
                                seqaccess = -1;

                        /*
                         * Doing I/O may synchronously insert additional
                         * pages so we can't be shared at this point either.
                         *
                         * NOTE: We can't free fs->m here in the allocated
                         *       case (fs->object != fs->first_object) as
                         *       this would require an exclusively locked
                         *       VM object.
                         */
                        if (fs->object == fs->first_object &&
                            fs->first_shared) {
                                vm_page_deactivate(fs->m);
                                vm_page_wakeup(fs->m);
                                fs->m = NULL;
                                fs->first_shared = 0;
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                return (KERN_TRY_AGAIN);
                        }
                        if (fs->object != fs->first_object &&
                            fs->shared) {
                                vm_page_deactivate(fs->m);
                                vm_page_wakeup(fs->m);
                                fs->m = NULL;
                                fs->first_shared = 0;
                                fs->shared = 0;
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                return (KERN_TRY_AGAIN);
                        }

                        /*
                         * Avoid deadlocking against the map when doing I/O.
                         * fs.object and the page is PG_BUSY'd.
                         *
                         * NOTE: Once unlocked, fs->entry can become stale
                         *       so this will NULL it out.
                         *
                         * NOTE: fs->entry is invalid until we relock the
                         *       map and verify that the timestamp has not
                         *       changed.
                         */
                        unlock_map(fs);

                        /*
                         * Acquire the page data.  We still hold a ref on
                         * fs.object and the page has been PG_BUSY's.
                         *
                         * The pager may replace the page (for example, in
                         * order to enter a fictitious page into the
                         * object).  If it does so it is responsible for
                         * cleaning up the passed page and properly setting
                         * the new page PG_BUSY.
                         *
                         * If we got here through a PG_RAM read-ahead
                         * mark the page may be partially dirty and thus
                         * not freeable.  Don't bother checking to see
                         * if the pager has the page because we can't free
                         * it anyway.  We have to depend on the get_page
                         * operation filling in any gaps whether there is
                         * backing store or not.
                         */
                        rv = vm_pager_get_page(fs->object, &fs->m, seqaccess);

                        if (rv == VM_PAGER_OK) {
                                /*
                                 * Relookup in case pager changed page. Pager
                                 * is responsible for disposition of old page
                                 * if moved.
                                 *
                                 * XXX other code segments do relookups too.
                                 * It's a bad abstraction that needs to be
                                 * fixed/removed.
                                 */
                                fs->m = vm_page_lookup(fs->object, pindex);
                                if (fs->m == NULL) {
                                        vm_object_pip_wakeup(fs->first_object);
                                        vm_object_chain_release_all(
                                                fs->first_object, fs->object);
                                        if (fs->object != fs->first_object)
                                                vm_object_drop(fs->object);
                                        unlock_and_deallocate(fs);
                                        return (KERN_TRY_AGAIN);
                                }
                                ++fs->hardfault;
                                break; /* break to PAGE HAS BEEN FOUND */
                        }

                        /*
                         * Remove the bogus page (which does not exist at this
                         * object/offset); before doing so, we must get back
                         * our object lock to preserve our invariant.
                         *
                         * Also wake up any other process that may want to bring
                         * in this page.
                         *
                         * If this is the top-level object, we must leave the
                         * busy page to prevent another process from rushing
                         * past us, and inserting the page in that object at
                         * the same time that we are.
                         */
                        if (rv == VM_PAGER_ERROR) {
                                if (curproc) {
                                        kprintf("vm_fault: pager read error, "
                                                "pid %d (%s)\n",
                                                curproc->p_pid,
                                                curproc->p_comm);
                                } else {
                                        kprintf("vm_fault: pager read error, "
                                                "thread %p (%s)\n",
                                                curthread,
                                                curproc->p_comm);
                                }
                        }

                        /*
                         * Data outside the range of the pager or an I/O error
                         *
                         * The page may have been wired during the pagein,
                         * e.g. by the buffer cache, and cannot simply be
                         * freed.  Call vnode_pager_freepage() to deal with it.
                         *
                         * Also note that we cannot free the page if we are
                         * holding the related object shared. XXX not sure
                         * what to do in that case.
                         */
                        if (fs->object != fs->first_object) {
                                vnode_pager_freepage(fs->m);
                                fs->m = NULL;
                                /*
                                 * XXX - we cannot just fall out at this
                                 * point, m has been freed and is invalid!
                                 */
                        }
                        /*
                         * XXX - the check for kernel_map is a kludge to work
                         * around having the machine panic on a kernel space
                         * fault w/ I/O error.
                         */
                        if (((fs->map != &kernel_map) &&
                            (rv == VM_PAGER_ERROR)) || (rv == VM_PAGER_BAD)) {
                                if (fs->m) {
                                        if (fs->first_shared) {
                                                vm_page_deactivate(fs->m);
                                                vm_page_wakeup(fs->m);
                                        } else {
                                                vnode_pager_freepage(fs->m);
                                        }
                                        fs->m = NULL;
                                }
                                vm_object_pip_wakeup(fs->first_object);
                                vm_object_chain_release_all(fs->first_object,
                                                            fs->object);
                                if (fs->object != fs->first_object)
                                        vm_object_drop(fs->object);
                                unlock_and_deallocate(fs);
                                if (rv == VM_PAGER_ERROR)
                                        return (KERN_FAILURE);
                                else
                                        return (KERN_PROTECTION_FAILURE);
                                /* NOT REACHED */
                        }
                }

                /*
                 * We get here if the object has a default pager (or unwiring) 
                 * or the pager doesn't have the page.
                 *
                 * fs->first_m will be used for the COW unless we find a
                 * deeper page to be mapped read-only, in which case the
                 * unlock*(fs) will free first_m.
                 */
                if (fs->object == fs->first_object)
                        fs->first_m = fs->m;

                /*
                 * Move on to the next object.  The chain lock should prevent
                 * the backing_object from getting ripped out from under us.
                 *
                 * The object lock for the next object is governed by
                 * fs->shared.
                 */
                if ((next_object = fs->object->backing_object) != NULL) {
                        if (fs->shared)
                                vm_object_hold_shared(next_object);
                        else
                                vm_object_hold(next_object);
                        vm_object_chain_acquire(next_object, fs->shared);
                        KKASSERT(next_object == fs->object->backing_object);
                        pindex += OFF_TO_IDX(fs->object->backing_object_offset);
                }

                if (next_object == NULL) {
                        /*
                         * If there's no object left, fill the page in the top
                         * object with zeros.
                         */
                        if (fs->object != fs->first_object) {
#if 0
                                if (fs->first_object->backing_object !=
                                    fs->object) {
                                        vm_object_hold(fs->first_object->backing_object);
                                }
#endif
                                vm_object_chain_release_all(
                                        fs->first_object->backing_object,
                                        fs->object);
#if 0
                                if (fs->first_object->backing_object !=
                                    fs->object) {
                                        vm_object_drop(fs->first_object->backing_object);
                                }
#endif
                                vm_object_pip_wakeup(fs->object);
                                vm_object_drop(fs->object);
                                fs->object = fs->first_object;
                                pindex = first_pindex;
                                fs->m = fs->first_m;
                        }
                        fs->first_m = NULL;

                        /*
                         * Zero the page if necessary and mark it valid.
                         */
                        if ((fs->m->flags & PG_ZERO) == 0) {
                                vm_page_zero_fill(fs->m);
                        } else {
#ifdef PMAP_DEBUG
                                pmap_page_assertzero(VM_PAGE_TO_PHYS(fs->m));
#endif
                                vm_page_flag_clear(fs->m, PG_ZERO);
                                mycpu->gd_cnt.v_ozfod++;
                        }
                        mycpu->gd_cnt.v_zfod++;
                        fs->m->valid = VM_PAGE_BITS_ALL;
                        break;  /* break to PAGE HAS BEEN FOUND */
                }
                if (fs->object != fs->first_object) {
                        vm_object_pip_wakeup(fs->object);
                        vm_object_lock_swap();
                        vm_object_drop(fs->object);
                }
                KASSERT(fs->object != next_object,
                        ("object loop %p", next_object));
                fs->object = next_object;
                vm_object_pip_add(fs->object, 1);
        }

        /*
         * PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock
         * is held.]
         *
         * object still held.
         *
         * local shared variable may be different from fs->shared.
         *
         * If the page is being written, but isn't already owned by the
         * top-level object, we have to copy it into a new page owned by the
         * top-level object.
         */
        KASSERT((fs->m->flags & PG_BUSY) != 0,
                ("vm_fault: not busy after main loop"));

        if (fs->object != fs->first_object) {
                /*
                 * We only really need to copy if we want to write it.
                 */
                if (fault_type & VM_PROT_WRITE) {
                        /*
                         * This allows pages to be virtually copied from a 
                         * backing_object into the first_object, where the 
                         * backing object has no other refs to it, and cannot
                         * gain any more refs.  Instead of a bcopy, we just 
                         * move the page from the backing object to the 
                         * first object.  Note that we must mark the page 
                         * dirty in the first object so that it will go out 
                         * to swap when needed.
                         */
                        if (
                                /*
                                 * Must be holding exclusive locks
                                 */
                                fs->first_shared == 0 &&
                                fs->shared == 0 &&
                                /*
                                 * Map, if present, has not changed
                                 */
                                (fs->map == NULL ||
                                fs->map_generation == fs->map->timestamp) &&
                                /*
                                 * Only one shadow object
                                 */
                                (fs->object->shadow_count == 1) &&
                                /*
                                 * No COW refs, except us
                                 */
                                (fs->object->ref_count == 1) &&
                                /*
                                 * No one else can look this object up
                                 */
                                (fs->object->handle == NULL) &&
                                /*
                                 * No other ways to look the object up
                                 */
                                ((fs->object->type == OBJT_DEFAULT) ||
                                 (fs->object->type == OBJT_SWAP)) &&
                                /*
                                 * We don't chase down the shadow chain
                                 */
                                (fs->object == fs->first_object->backing_object) &&

                                /*
                                 * grab the lock if we need to
                                 */
                                (fs->lookup_still_valid ||
                                 fs->map == NULL ||
                                 lockmgr(&fs->map->lock, LK_EXCLUSIVE|LK_NOWAIT) == 0)
                            ) {
                                /*
                                 * (first_m) and (m) are both busied.  We have
                                 * move (m) into (first_m)'s object/pindex
                                 * in an atomic fashion, then free (first_m).
                                 *
                                 * first_object is held so second remove
                                 * followed by the rename should wind
                                 * up being atomic.  vm_page_free() might
                                 * block so we don't do it until after the
                                 * rename.
                                 */
                                fs->lookup_still_valid = 1;
                                vm_page_protect(fs->first_m, VM_PROT_NONE);
                                vm_page_remove(fs->first_m);
                                vm_page_rename(fs->m, fs->first_object,
                                               first_pindex);
                                vm_page_free(fs->first_m);
                                fs->first_m = fs->m;
                                fs->m = NULL;
                                mycpu->gd_cnt.v_cow_optim++;
                        } else {
                                /*
                                 * Oh, well, lets copy it.
                                 *
                                 * Why are we unmapping the original page
                                 * here?  Well, in short, not all accessors
                                 * of user memory go through the pmap.  The
                                 * procfs code doesn't have access user memory
                                 * via a local pmap, so vm_fault_page*()
                                 * can't call pmap_enter().  And the umtx*()
                                 * code may modify the COW'd page via a DMAP
                                 * or kernel mapping and not via the pmap,
                                 * leaving the original page still mapped
                                 * read-only into the pmap.
                                 *
                                 * So we have to remove the page from at
                                 * least the current pmap if it is in it.
                                 * Just remove it from all pmaps.
                                 */
                                KKASSERT(fs->first_shared == 0);
                                vm_page_copy(fs->m, fs->first_m);
                                vm_page_protect(fs->m, VM_PROT_NONE);
                                vm_page_event(fs->m, VMEVENT_COW);
                        }

                        /*
                         * We no longer need the old page or object.
                         */
                        if (fs->m)
                                release_page(fs);

                        /*
                         * We intend to revert to first_object, undo the
                         * chain lock through to that.
                         */
#if 0
                        if (fs->first_object->backing_object != fs->object)
                                vm_object_hold(fs->first_object->backing_object);
#endif
                        vm_object_chain_release_all(
                                        fs->first_object->backing_object,
                                        fs->object);
#if 0
                        if (fs->first_object->backing_object != fs->object)
                                vm_object_drop(fs->first_object->backing_object);
#endif

                        /*
                         * fs->object != fs->first_object due to above 
                         * conditional
                         */
                        vm_object_pip_wakeup(fs->object);
                        vm_object_drop(fs->object);

                        /*
                         * Only use the new page below...
                         */
                        mycpu->gd_cnt.v_cow_faults++;
                        fs->m = fs->first_m;
                        fs->object = fs->first_object;
                        pindex = first_pindex;
                } else {
                        /*
                         * If it wasn't a write fault avoid having to copy
                         * the page by mapping it read-only.
                         */
                        fs->prot &= ~VM_PROT_WRITE;
                }
        }

        /*
         * Relock the map if necessary, then check the generation count.
         * relock_map() will update fs->timestamp to account for the
         * relocking if necessary.
         *
         * If the count has changed after relocking then all sorts of
         * crap may have happened and we have to retry.
         *
         * NOTE: The relock_map() can fail due to a deadlock against
         *       the vm_page we are holding BUSY.
         */
        if (fs->lookup_still_valid == FALSE && fs->map) {
                if (relock_map(fs) ||
                    fs->map->timestamp != fs->map_generation) {
                        release_page(fs);
                        vm_object_pip_wakeup(fs->first_object);
                        vm_object_chain_release_all(fs->first_object,
                                                    fs->object);
                        if (fs->object != fs->first_object)
                                vm_object_drop(fs->object);
                        unlock_and_deallocate(fs);
                        return (KERN_TRY_AGAIN);
                }
        }

        /*
         * If the fault is a write, we know that this page is being
         * written NOW so dirty it explicitly to save on pmap_is_modified()
         * calls later.
         *
         * If this is a NOSYNC mmap we do not want to set PG_NOSYNC
         * if the page is already dirty to prevent data written with
         * the expectation of being synced from not being synced.
         * Likewise if this entry does not request NOSYNC then make
         * sure the page isn't marked NOSYNC.  Applications sharing
         * data should use the same flags to avoid ping ponging.
         *
         * Also tell the backing pager, if any, that it should remove
         * any swap backing since the page is now dirty.
         */
        vm_page_activate(fs->m);
        if (fs->prot & VM_PROT_WRITE) {
                vm_object_set_writeable_dirty(fs->m->object);
                vm_set_nosync(fs->m, fs->entry);
                if (fs->fault_flags & VM_FAULT_DIRTY) {
                        vm_page_dirty(fs->m);
                        swap_pager_unswapped(fs->m);
                }
        }

        vm_object_pip_wakeup(fs->first_object);
        vm_object_chain_release_all(fs->first_object, fs->object);
        if (fs->object != fs->first_object)
                vm_object_drop(fs->object);

        /*
         * Page had better still be busy.  We are still locked up and 
         * fs->object will have another PIP reference if it is not equal
         * to fs->first_object.
         */
        KASSERT(fs->m->flags & PG_BUSY,
                ("vm_fault: page %p not busy!", fs->m));

        /*
         * Sanity check: page must be completely valid or it is not fit to
         * map into user space.  vm_pager_get_pages() ensures this.
         */
        if (fs->m->valid != VM_PAGE_BITS_ALL) {
                vm_page_zero_invalid(fs->m, TRUE);
                kprintf("Warning: page %p partially invalid on fault\n", fs->m);
        }
        vm_page_flag_clear(fs->m, PG_ZERO);

        return (KERN_SUCCESS);
}

/*
 * Hold each of the physical pages that are mapped by the specified range of
 * virtual addresses, ["addr", "addr" + "len"), if those mappings are valid
 * and allow the specified types of access, "prot".  If all of the implied
 * pages are successfully held, then the number of held pages is returned
 * together with pointers to those pages in the array "ma".  However, if any
 * of the pages cannot be held, -1 is returned.
 */
int
vm_fault_quick_hold_pages(vm_map_t map, vm_offset_t addr, vm_size_t len,
    vm_prot_t prot, vm_page_t *ma, int max_count)
{
        vm_offset_t start, end;
        int i, npages, error;

        start = trunc_page(addr);
        end = round_page(addr + len);

        npages = howmany(end - start, PAGE_SIZE);

        if (npages > max_count)
                return -1;

        for (i = 0; i < npages; i++) {
                // XXX error handling
                ma[i] = vm_fault_page_quick(start + (i * PAGE_SIZE),
                        prot,
                        &error);
        }

        return npages;
}

/*
 * Wire down a range of virtual addresses in a map.  The entry in question
 * should be marked in-transition and the map must be locked.  We must
 * release the map temporarily while faulting-in the page to avoid a
 * deadlock.  Note that the entry may be clipped while we are blocked but
 * will never be freed.
 *
 * No requirements.
 */
int
vm_fault_wire(vm_map_t map, vm_map_entry_t entry, boolean_t user_wire)
{
        boolean_t fictitious;
        vm_offset_t start;
        vm_offset_t end;
        vm_offset_t va;
        vm_paddr_t pa;
        vm_page_t m;
        pmap_t pmap;
        int rv;

        lwkt_gettoken(&map->token);

        pmap = vm_map_pmap(map);
        start = entry->start;
        end = entry->end;
        fictitious = entry->object.vm_object &&
                        ((entry->object.vm_object->type == OBJT_DEVICE) ||
                         (entry->object.vm_object->type == OBJT_MGTDEVICE));
        if (entry->eflags & MAP_ENTRY_KSTACK)
                start += PAGE_SIZE;
        map->timestamp++;
        vm_map_unlock(map);

        /*
         * We simulate a fault to get the page and enter it in the physical
         * map.
         */
        for (va = start; va < end; va += PAGE_SIZE) {
                if (user_wire) {
                        rv = vm_fault(map, va, VM_PROT_READ, 
                                        VM_FAULT_USER_WIRE);
                } else {
                        rv = vm_fault(map, va, VM_PROT_READ|VM_PROT_WRITE,
                                        VM_FAULT_CHANGE_WIRING);
                }
                if (rv) {
                        while (va > start) {
                                va -= PAGE_SIZE;
                                if ((pa = pmap_extract(pmap, va)) == 0)
                                        continue;
                                pmap_change_wiring(pmap, va, FALSE, entry);
                                if (!fictitious) {
                                        m = PHYS_TO_VM_PAGE(pa);
                                        vm_page_busy_wait(m, FALSE, "vmwrpg");
                                        vm_page_unwire(m, 1);
                                        vm_page_wakeup(m);
                                }
                        }
                        goto done;
                }
        }
        rv = KERN_SUCCESS;
done:
        vm_map_lock(map);
        lwkt_reltoken(&map->token);
        return (rv);
}

/*
 * Unwire a range of virtual addresses in a map.  The map should be
 * locked.
 */
void
vm_fault_unwire(vm_map_t map, vm_map_entry_t entry)
{
        boolean_t fictitious;
        vm_offset_t start;
        vm_offset_t end;
        vm_offset_t va;
        vm_paddr_t pa;
        vm_page_t m;
        pmap_t pmap;

        lwkt_gettoken(&map->token);

        pmap = vm_map_pmap(map);
        start = entry->start;
        end = entry->end;
        fictitious = entry->object.vm_object &&
                        ((entry->object.vm_object->type == OBJT_DEVICE) ||
                         (entry->object.vm_object->type == OBJT_MGTDEVICE));
        if (entry->eflags & MAP_ENTRY_KSTACK)
                start += PAGE_SIZE;

        /*
         * Since the pages are wired down, we must be able to get their
         * mappings from the physical map system.
         */
        for (va = start; va < end; va += PAGE_SIZE) {
                pa = pmap_extract(pmap, va);
                if (pa != 0) {
                        pmap_change_wiring(pmap, va, FALSE, entry);
                        if (!fictitious) {
                                m = PHYS_TO_VM_PAGE(pa);
                                vm_page_busy_wait(m, FALSE, "vmwupg");
                                vm_page_unwire(m, 1);
                                vm_page_wakeup(m);
                        }
                }
        }
        lwkt_reltoken(&map->token);
}

/*
 * Copy all of the pages from a wired-down map entry to another.
 *
 * The source and destination maps must be locked for write.
 * The source and destination maps token must be held
 * The source map entry must be wired down (or be a sharing map
 * entry corresponding to a main map entry that is wired down).
 *
 * No other requirements.
 *
 * XXX do segment optimization
 */
void
vm_fault_copy_entry(vm_map_t dst_map, vm_map_t src_map,
                    vm_map_entry_t dst_entry, vm_map_entry_t src_entry)
{
        vm_object_t dst_object;
        vm_object_t src_object;
        vm_ooffset_t dst_offset;
        vm_ooffset_t src_offset;
        vm_prot_t prot;
        vm_offset_t vaddr;
        vm_page_t dst_m;
        vm_page_t src_m;

        src_object = src_entry->object.vm_object;
        src_offset = src_entry->offset;

        /*
         * Create the top-level object for the destination entry. (Doesn't
         * actually shadow anything - we copy the pages directly.)
         */
        vm_map_entry_allocate_object(dst_entry);
        dst_object = dst_entry->object.vm_object;

        prot = dst_entry->max_protection;

        /*
         * Loop through all of the pages in the entry's range, copying each
         * one from the source object (it should be there) to the destination
         * object.
         */
        vm_object_hold(src_object);
        vm_object_hold(dst_object);
        for (vaddr = dst_entry->start, dst_offset = 0;
            vaddr < dst_entry->end;
            vaddr += PAGE_SIZE, dst_offset += PAGE_SIZE) {

                /*
                 * Allocate a page in the destination object
                 */
                do {
                        dst_m = vm_page_alloc(dst_object,
                                              OFF_TO_IDX(dst_offset),
                                              VM_ALLOC_NORMAL);
                        if (dst_m == NULL) {
                                vm_wait(0);
                        }
                } while (dst_m == NULL);

                /*
                 * Find the page in the source object, and copy it in.
                 * (Because the source is wired down, the page will be in
                 * memory.)
                 */
                src_m = vm_page_lookup(src_object,
                                       OFF_TO_IDX(dst_offset + src_offset));
                if (src_m == NULL)
                        panic("vm_fault_copy_wired: page missing");

                vm_page_copy(src_m, dst_m);
                vm_page_event(src_m, VMEVENT_COW);

                /*
                 * Enter it in the pmap...
                 */

                vm_page_flag_clear(dst_m, PG_ZERO);
                pmap_enter(dst_map->pmap, vaddr, dst_m, prot, FALSE, dst_entry);

                /*
                 * Mark it no longer busy, and put it on the active list.
                 */
                vm_page_activate(dst_m);
                vm_page_wakeup(dst_m);
        }
        vm_object_drop(dst_object);
        vm_object_drop(src_object);
}

#if 0

/*
 * This routine checks around the requested page for other pages that
 * might be able to be faulted in.  This routine brackets the viable
 * pages for the pages to be paged in.
 *
 * Inputs:
 *      m, rbehind, rahead
 *
 * Outputs:
 *  marray (array of vm_page_t), reqpage (index of requested page)
 *
 * Return value:
 *  number of pages in marray
 */
static int
vm_fault_additional_pages(vm_page_t m, int rbehind, int rahead,
                          vm_page_t *marray, int *reqpage)
{
        int i,j;
        vm_object_t object;
        vm_pindex_t pindex, startpindex, endpindex, tpindex;
        vm_page_t rtm;
        int cbehind, cahead;

        object = m->object;
        pindex = m->pindex;

        /*
         * we don't fault-ahead for device pager
         */
        if ((object->type == OBJT_DEVICE) ||
            (object->type == OBJT_MGTDEVICE)) {
                *reqpage = 0;
                marray[0] = m;
                return 1;
        }

        /*
         * if the requested page is not available, then give up now
         */
        if (!vm_pager_has_page(object, pindex, &cbehind, &cahead)) {
                *reqpage = 0;   /* not used by caller, fix compiler warn */
                return 0;
        }

        if ((cbehind == 0) && (cahead == 0)) {
                *reqpage = 0;
                marray[0] = m;
                return 1;
        }

        if (rahead > cahead) {
                rahead = cahead;
        }

        if (rbehind > cbehind) {
                rbehind = cbehind;
        }

        /*
         * Do not do any readahead if we have insufficient free memory.
         *
         * XXX code was broken disabled before and has instability
         * with this conditonal fixed, so shortcut for now.
         */
        if (burst_fault == 0 || vm_page_count_severe()) {
                marray[0] = m;
                *reqpage = 0;
                return 1;
        }

        /*
         * scan backward for the read behind pages -- in memory 
         *
         * Assume that if the page is not found an interrupt will not
         * create it.  Theoretically interrupts can only remove (busy)
         * pages, not create new associations.
         */
        if (pindex > 0) {
                if (rbehind > pindex) {
                        rbehind = pindex;
                        startpindex = 0;
                } else {
                        startpindex = pindex - rbehind;
                }

                vm_object_hold(object);
                for (tpindex = pindex; tpindex > startpindex; --tpindex) {
                        if (vm_page_lookup(object, tpindex - 1))
                                break;
                }

                i = 0;
                while (tpindex < pindex) {
                        rtm = vm_page_alloc(object, tpindex, VM_ALLOC_SYSTEM |
                                                             VM_ALLOC_NULL_OK);
                        if (rtm == NULL) {
                                for (j = 0; j < i; j++) {
                                        vm_page_free(marray[j]);
                                }
                                vm_object_drop(object);
                                marray[0] = m;
                                *reqpage = 0;
                                return 1;
                        }
                        marray[i] = rtm;
                        ++i;
                        ++tpindex;
                }
                vm_object_drop(object);
        } else {
                i = 0;
        }

        /*
         * Assign requested page
         */
        marray[i] = m;
        *reqpage = i;
        ++i;

        /*
         * Scan forwards for read-ahead pages
         */
        tpindex = pindex + 1;
        endpindex = tpindex + rahead;
        if (endpindex > object->size)
                endpindex = object->size;

        vm_object_hold(object);
        while (tpindex < endpindex) {
                if (vm_page_lookup(object, tpindex))
                        break;
                rtm = vm_page_alloc(object, tpindex, VM_ALLOC_SYSTEM |
                                                     VM_ALLOC_NULL_OK);
                if (rtm == NULL)
                        break;
                marray[i] = rtm;
                ++i;
                ++tpindex;
        }
        vm_object_drop(object);

        return (i);
}

#endif

/*
 * vm_prefault() provides a quick way of clustering pagefaults into a
 * processes address space.  It is a "cousin" of pmap_object_init_pt,
 * except it runs at page fault time instead of mmap time.
 *
 * vm.fast_fault        Enables pre-faulting zero-fill pages
 *
 * vm.prefault_pages    Number of pages (1/2 negative, 1/2 positive) to
 *                      prefault.  Scan stops in either direction when
 *                      a page is found to already exist.
 *
 * This code used to be per-platform pmap_prefault().  It is now
 * machine-independent and enhanced to also pre-fault zero-fill pages
 * (see vm.fast_fault) as well as make them writable, which greatly
 * reduces the number of page faults programs incur.
 *
 * Application performance when pre-faulting zero-fill pages is heavily
 * dependent on the application.  Very tiny applications like /bin/echo
 * lose a little performance while applications of any appreciable size
 * gain performance.  Prefaulting multiple pages also reduces SMP
 * congestion and can improve SMP performance significantly.
 *
 * NOTE!  prot may allow writing but this only applies to the top level
 *        object.  If we wind up mapping a page extracted from a backing
 *        object we have to make sure it is read-only.
 *
 * NOTE!  The caller has already handled any COW operations on the
 *        vm_map_entry via the normal fault code.  Do NOT call this
 *        shortcut unless the normal fault code has run on this entry.
 *
 * The related map must be locked.
 * No other requirements.
 */
static int vm_prefault_pages = 8;
SYSCTL_INT(_vm, OID_AUTO, prefault_pages, CTLFLAG_RW, &vm_prefault_pages, 0,
           "Maximum number of pages to pre-fault");
static int vm_fast_fault = 1;
SYSCTL_INT(_vm, OID_AUTO, fast_fault, CTLFLAG_RW, &vm_fast_fault, 0,
           "Burst fault zero-fill regions");

/*
 * Set PG_NOSYNC if the map entry indicates so, but only if the page
 * is not already dirty by other means.  This will prevent passive
 * filesystem syncing as well as 'sync' from writing out the page.
 */
static void
vm_set_nosync(vm_page_t m, vm_map_entry_t entry)
{
        if (entry->eflags & MAP_ENTRY_NOSYNC) {
                if (m->dirty == 0)
                        vm_page_flag_set(m, PG_NOSYNC);
        } else {
                vm_page_flag_clear(m, PG_NOSYNC);
        }
}

static void
vm_prefault(pmap_t pmap, vm_offset_t addra, vm_map_entry_t entry, int prot,
            int fault_flags)
{
        struct lwp *lp;
        vm_page_t m;
        vm_offset_t addr;
        vm_pindex_t index;
        vm_pindex_t pindex;
        vm_object_t object;
        int pprot;
        int i;
        int noneg;
        int nopos;
        int maxpages;

        /*
         * Get stable max count value, disabled if set to 0
         */
        maxpages = vm_prefault_pages;
        cpu_ccfence();
        if (maxpages <= 0)
                return;

        /*
         * We do not currently prefault mappings that use virtual page
         * tables.  We do not prefault foreign pmaps.
         */
        if (entry->maptype == VM_MAPTYPE_VPAGETABLE)
                return;
        lp = curthread->td_lwp;
        if (lp == NULL || (pmap != vmspace_pmap(lp->lwp_vmspace)))
                return;

        /*
         * Limit pre-fault count to 1024 pages.
         */
        if (maxpages > 1024)
                maxpages = 1024;

        object = entry->object.vm_object;
        KKASSERT(object != NULL);
        KKASSERT(object == entry->object.vm_object);
        vm_object_hold(object);
        vm_object_chain_acquire(object, 0);

        noneg = 0;
        nopos = 0;
        for (i = 0; i < maxpages; ++i) {
                vm_object_t lobject;
                vm_object_t nobject;
                int allocated = 0;
                int error;

                /*
                 * This can eat a lot of time on a heavily contended
                 * machine so yield on the tick if needed.
                 */
                if ((i & 7) == 7)
                        lwkt_yield();

                /*
                 * Calculate the page to pre-fault, stopping the scan in
                 * each direction separately if the limit is reached.
                 */
                if (i & 1) {
                        if (noneg)
                                continue;
                        addr = addra - ((i + 1) >> 1) * PAGE_SIZE;
                } else {
                        if (nopos)
                                continue;
                        addr = addra + ((i + 2) >> 1) * PAGE_SIZE;
                }
                if (addr < entry->start) {
                        noneg = 1;
                        if (noneg && nopos)
                                break;
                        continue;
                }
                if (addr >= entry->end) {
                        nopos = 1;
                        if (noneg && nopos)
                                break;
                        continue;
                }

                /*
                 * Skip pages already mapped, and stop scanning in that
                 * direction.  When the scan terminates in both directions
                 * we are done.
                 */
                if (pmap_prefault_ok(pmap, addr) == 0) {
                        if (i & 1)
                                noneg = 1;
                        else
                                nopos = 1;
                        if (noneg && nopos)
                                break;
                        continue;
                }

                /*
                 * Follow the VM object chain to obtain the page to be mapped
                 * into the pmap.
                 *
                 * If we reach the terminal object without finding a page
                 * and we determine it would be advantageous, then allocate
                 * a zero-fill page for the base object.  The base object
                 * is guaranteed to be OBJT_DEFAULT for this case.
                 *
                 * In order to not have to check the pager via *haspage*()
                 * we stop if any non-default object is encountered.  e.g.
                 * a vnode or swap object would stop the loop.
                 */
                index = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
                lobject = object;
                pindex = index;
                pprot = prot;

                KKASSERT(lobject == entry->object.vm_object);
                /*vm_object_hold(lobject); implied */

                while ((m = vm_page_lookup_busy_try(lobject, pindex,
                                                    TRUE, &error)) == NULL) {
                        if (lobject->type != OBJT_DEFAULT)
                                break;
                        if (lobject->backing_object == NULL) {
                                if (vm_fast_fault == 0)
                                        break;
                                if ((prot & VM_PROT_WRITE) == 0 ||
                                    vm_page_count_min(0)) {
                                        break;
                                }

                                /*
                                 * NOTE: Allocated from base object
                                 */
                                m = vm_page_alloc(object, index,
                                                  VM_ALLOC_NORMAL |
                                                  VM_ALLOC_ZERO |
                                                  VM_ALLOC_USE_GD |
                                                  VM_ALLOC_NULL_OK);
                                if (m == NULL)
                                        break;
                                allocated = 1;
                                pprot = prot;
                                /* lobject = object .. not needed */
                                break;
                        }
                        if (lobject->backing_object_offset & PAGE_MASK)
                                break;
                        nobject = lobject->backing_object;
                        vm_object_hold(nobject);
                        KKASSERT(nobject == lobject->backing_object);
                        pindex += lobject->backing_object_offset >> PAGE_SHIFT;
                        if (lobject != object) {
                                vm_object_lock_swap();
                                vm_object_drop(lobject);
                        }
                        lobject = nobject;
                        pprot &= ~VM_PROT_WRITE;
                        vm_object_chain_acquire(lobject, 0);
                }

                /*
                 * NOTE: A non-NULL (m) will be associated with lobject if
                 *       it was found there, otherwise it is probably a
                 *       zero-fill page associated with the base object.
                 *
                 * Give-up if no page is available.
                 */
                if (m == NULL) {
                        if (lobject != object) {
#if 0
                                if (object->backing_object != lobject)
                                        vm_object_hold(object->backing_object);
#endif
                                vm_object_chain_release_all(
                                        object->backing_object, lobject);
#if 0
                                if (object->backing_object != lobject)
                                        vm_object_drop(object->backing_object);
#endif
                                vm_object_drop(lobject);
                        }
                        break;
                }

                /*
                 * The object must be marked dirty if we are mapping a
                 * writable page.  m->object is either lobject or object,
                 * both of which are still held.  Do this before we
                 * potentially drop the object.
                 */
                if (pprot & VM_PROT_WRITE)
                        vm_object_set_writeable_dirty(m->object);

                /*
                 * Do not conditionalize on PG_RAM.  If pages are present in
                 * the VM system we assume optimal caching.  If caching is
                 * not optimal the I/O gravy train will be restarted when we
                 * hit an unavailable page.  We do not want to try to restart
                 * the gravy train now because we really don't know how much
                 * of the object has been cached.  The cost for restarting
                 * the gravy train should be low (since accesses will likely
                 * be I/O bound anyway).
                 */
                if (lobject != object) {
#if 0
                        if (object->backing_object != lobject)
                                vm_object_hold(object->backing_object);
#endif
                        vm_object_chain_release_all(object->backing_object,
                                                    lobject);
#if 0
                        if (object->backing_object != lobject)
                                vm_object_drop(object->backing_object);
#endif
                        vm_object_drop(lobject);
                }

                /*
                 * Enter the page into the pmap if appropriate.  If we had
                 * allocated the page we have to place it on a queue.  If not
                 * we just have to make sure it isn't on the cache queue
                 * (pages on the cache queue are not allowed to be mapped).
                 */
                if (allocated) {
                        /*
                         * Page must be zerod.
                         */
                        if ((m->flags & PG_ZERO) == 0) {
                                vm_page_zero_fill(m);
                        } else {
#ifdef PMAP_DEBUG
                                pmap_page_assertzero(
                                                VM_PAGE_TO_PHYS(m));
#endif
                                vm_page_flag_clear(m, PG_ZERO);
                                mycpu->gd_cnt.v_ozfod++;
                        }
                        mycpu->gd_cnt.v_zfod++;
                        m->valid = VM_PAGE_BITS_ALL;

                        /*
                         * Handle dirty page case
                         */
                        if (pprot & VM_PROT_WRITE)
                                vm_set_nosync(m, entry);
                        pmap_enter(pmap, addr, m, pprot, 0, entry);
                        mycpu->gd_cnt.v_vm_faults++;
                        if (curthread->td_lwp)
                                ++curthread->td_lwp->lwp_ru.ru_minflt;
                        vm_page_deactivate(m);
                        if (pprot & VM_PROT_WRITE) {
                                /*vm_object_set_writeable_dirty(m->object);*/
                                vm_set_nosync(m, entry);
                                if (fault_flags & VM_FAULT_DIRTY) {
                                        vm_page_dirty(m);
                                        /*XXX*/
                                        swap_pager_unswapped(m);
                                }
                        }
                        vm_page_wakeup(m);
                } else if (error) {
                        /* couldn't busy page, no wakeup */
                } else if (
                    ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
                    (m->flags & PG_FICTITIOUS) == 0) {
                        /*
                         * A fully valid page not undergoing soft I/O can
                         * be immediately entered into the pmap.
                         */
                        if ((m->queue - m->pc) == PQ_CACHE)
                                vm_page_deactivate(m);
                        if (pprot & VM_PROT_WRITE) {
                                /*vm_object_set_writeable_dirty(m->object);*/
                                vm_set_nosync(m, entry);
                                if (fault_flags & VM_FAULT_DIRTY) {
                                        vm_page_dirty(m);
                                        /*XXX*/
                                        swap_pager_unswapped(m);
                                }
                        }
                        if (pprot & VM_PROT_WRITE)
                                vm_set_nosync(m, entry);
                        pmap_enter(pmap, addr, m, pprot, 0, entry);
                        mycpu->gd_cnt.v_vm_faults++;
                        if (curthread->td_lwp)
                                ++curthread->td_lwp->lwp_ru.ru_minflt;
                        vm_page_wakeup(m);
                } else {
                        vm_page_wakeup(m);
                }
        }
        vm_object_chain_release(object);
        vm_object_drop(object);
}

/*
 * Object can be held shared
 */
static void
vm_prefault_quick(pmap_t pmap, vm_offset_t addra,
                  vm_map_entry_t entry, int prot, int fault_flags)
{
        struct lwp *lp;
        vm_page_t m;
        vm_offset_t addr;
        vm_pindex_t pindex;
        vm_object_t object;
        int i;
        int noneg;
        int nopos;
        int maxpages;

        /*
         * Get stable max count value, disabled if set to 0
         */
        maxpages = vm_prefault_pages;
        cpu_ccfence();
        if (maxpages <= 0)
                return;

        /*
         * We do not currently prefault mappings that use virtual page
         * tables.  We do not prefault foreign pmaps.
         */
        if (entry->maptype == VM_MAPTYPE_VPAGETABLE)
                return;
        lp = curthread->td_lwp;
        if (lp == NULL || (pmap != vmspace_pmap(lp->lwp_vmspace)))
                return;
        object = entry->object.vm_object;
        if (object->backing_object != NULL)
                return;
        ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));

        /*
         * Limit pre-fault count to 1024 pages.
         */
        if (maxpages > 1024)
                maxpages = 1024;

        noneg = 0;
        nopos = 0;
        for (i = 0; i < maxpages; ++i) {
                int error;

                /*
                 * Calculate the page to pre-fault, stopping the scan in
                 * each direction separately if the limit is reached.
                 */
                if (i & 1) {
                        if (noneg)
                                continue;
                        addr = addra - ((i + 1) >> 1) * PAGE_SIZE;
                } else {
                        if (nopos)
                                continue;
                        addr = addra + ((i + 2) >> 1) * PAGE_SIZE;
                }
                if (addr < entry->start) {
                        noneg = 1;
                        if (noneg && nopos)
                                break;
                        continue;
                }
                if (addr >= entry->end) {
                        nopos = 1;
                        if (noneg && nopos)
                                break;
                        continue;
                }

                /*
                 * Skip pages already mapped, and stop scanning in that
                 * direction.  When the scan terminates in both directions
                 * we are done.
                 */
                if (pmap_prefault_ok(pmap, addr) == 0) {
                        if (i & 1)
                                noneg = 1;
                        else
                                nopos = 1;
                        if (noneg && nopos)
                                break;
                        continue;
                }

                /*
                 * Follow the VM object chain to obtain the page to be mapped
                 * into the pmap.  This version of the prefault code only
                 * works with terminal objects.
                 *
                 * WARNING!  We cannot call swap_pager_unswapped() with a
                 *           shared token.
                 */
                pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;

                m = vm_page_lookup_busy_try(object, pindex, TRUE, &error);
                if (m == NULL || error)
                        continue;

                if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
                    (m->flags & PG_FICTITIOUS) == 0 &&
                    ((m->flags & PG_SWAPPED) == 0 ||
                     (prot & VM_PROT_WRITE) == 0 ||
                     (fault_flags & VM_FAULT_DIRTY) == 0)) {
                        /*
                         * A fully valid page not undergoing soft I/O can
                         * be immediately entered into the pmap.
                         */
                        if ((m->queue - m->pc) == PQ_CACHE)
                                vm_page_deactivate(m);
                        if (prot & VM_PROT_WRITE) {
                                vm_object_set_writeable_dirty(m->object);
                                vm_set_nosync(m, entry);
                                if (fault_flags & VM_FAULT_DIRTY) {
                                        vm_page_dirty(m);
                                        /*XXX*/
                                        swap_pager_unswapped(m);
                                }
                        }
                        pmap_enter(pmap, addr, m, prot, 0, entry);
                        mycpu->gd_cnt.v_vm_faults++;
                        if (curthread->td_lwp)
                                ++curthread->td_lwp->lwp_ru.ru_minflt;
                }
                vm_page_wakeup(m);
        }
}