i386/i386/pmap.c

/*-
 * Copyright (c) 1991 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 Systems Programming Group of the University of Utah Computer
 * Science Department and William Jolitz of UUNET Technologies Inc.
 *
 * 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. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Berkeley and its contributors.
 * 4. 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:	@(#)pmap.c	7.7 (Berkeley)	5/12/91
 */
/*-
 * Copyright (c) 2003 Networks Associates Technology, Inc.
 * All rights reserved.
 *
 * This software was developed for the FreeBSD Project by Jake Burkholder,
 * Safeport Network Services, and Network Associates Laboratories, the
 * Security Research Division of Network Associates, Inc. under
 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA
 * CHATS research program.
 *
 * 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.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
 */

#include <sys/cdefs.h>
__FBSDID("$FreeBSD: head/sys/i386/i386/pmap.c 130932 2004-06-22 19:35:43Z alc $");

/*
 *	Manages physical address maps.
 *
 *	In addition to hardware address maps, this
 *	module is called upon to provide software-use-only
 *	maps which may or may not be stored in the same
 *	form as hardware maps.  These pseudo-maps are
 *	used to store intermediate results from copy
 *	operations to and from address spaces.
 *
 *	Since the information managed by this module is
 *	also stored by the logical address mapping module,
 *	this module may throw away valid virtual-to-physical
 *	mappings at almost any time.  However, invalidations
 *	of virtual-to-physical mappings must be done as
 *	requested.
 *
 *	In order to cope with hardware architectures which
 *	make virtual-to-physical map invalidates expensive,
 *	this module may delay invalidate or reduced protection
 *	operations until such time as they are actually
 *	necessary.  This module is given full information as
 *	to which processors are currently using which maps,
 *	and to when physical maps must be made correct.
 */

#include "opt_cpu.h"
#include "opt_pmap.h"
#include "opt_msgbuf.h"
#include "opt_kstack_pages.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mman.h>
#include <sys/msgbuf.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/sx.h>
#include <sys/user.h>
#include <sys/vmmeter.h>
#include <sys/sched.h>
#include <sys/sysctl.h>
#ifdef SMP
#include <sys/smp.h>
#endif

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

#include <machine/cpu.h>
#include <machine/cputypes.h>
#include <machine/md_var.h>
#include <machine/specialreg.h>
#ifdef SMP
#include <machine/smp.h>
#endif

#if !defined(CPU_ENABLE_SSE) && defined(I686_CPU)
#define CPU_ENABLE_SSE
#endif
#if defined(CPU_DISABLE_SSE)
#undef CPU_ENABLE_SSE
#endif

#ifndef PMAP_SHPGPERPROC
#define PMAP_SHPGPERPROC 200
#endif

#if defined(DIAGNOSTIC)
#define PMAP_DIAGNOSTIC
#endif

#define MINPV 2048

#if !defined(PMAP_DIAGNOSTIC)
#define PMAP_INLINE __inline
#else
#define PMAP_INLINE
#endif

/*
 * Get PDEs and PTEs for user/kernel address space
 */
#define	pmap_pde(m, v)	(&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])

#define pmap_pde_v(pte)		((*(int *)pte & PG_V) != 0)
#define pmap_pte_w(pte)		((*(int *)pte & PG_W) != 0)
#define pmap_pte_m(pte)		((*(int *)pte & PG_M) != 0)
#define pmap_pte_u(pte)		((*(int *)pte & PG_A) != 0)
#define pmap_pte_v(pte)		((*(int *)pte & PG_V) != 0)

#define pmap_pte_set_w(pte, v)	((v) ? atomic_set_int((u_int *)(pte), PG_W) : \
    atomic_clear_int((u_int *)(pte), PG_W))
#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))

struct pmap kernel_pmap_store;
LIST_HEAD(pmaplist, pmap);
static struct pmaplist allpmaps;
static struct mtx allpmaps_lock;
#ifdef SMP
static struct mtx lazypmap_lock;
#endif

vm_paddr_t avail_end;	/* PA of last available physical page */
vm_offset_t virtual_avail;	/* VA of first avail page (after kernel bss) */
vm_offset_t virtual_end;	/* VA of last avail page (end of kernel AS) */
static boolean_t pmap_initialized = FALSE;	/* Has pmap_init completed? */
int pgeflag = 0;		/* PG_G or-in */
int pseflag = 0;		/* PG_PS or-in */

static int nkpt;
vm_offset_t kernel_vm_end;
extern u_int32_t KERNend;

#ifdef PAE
static uma_zone_t pdptzone;
#endif

/*
 * Data for the pv entry allocation mechanism
 */
static uma_zone_t pvzone;
static struct vm_object pvzone_obj;
static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
int pmap_pagedaemon_waken;

/*
 * All those kernel PT submaps that BSD is so fond of
 */
pt_entry_t *CMAP1 = 0;
static pt_entry_t *CMAP2, *CMAP3;
caddr_t CADDR1 = 0, ptvmmap = 0;
static caddr_t CADDR2, CADDR3;
static struct mtx CMAPCADDR12_lock;
struct msgbuf *msgbufp = 0;

/*
 * Crashdump maps.
 */
static caddr_t crashdumpmap;

#ifdef SMP
extern pt_entry_t *SMPpt;
#endif
static pt_entry_t *PMAP1 = 0, *PMAP2;
static pt_entry_t *PADDR1 = 0, *PADDR2;
#ifdef SMP
static int PMAP1cpu;
static int PMAP1changedcpu;
SYSCTL_INT(_debug, OID_AUTO, PMAP1changedcpu, CTLFLAG_RD,
	   &PMAP1changedcpu, 0,
	   "Number of times pmap_pte_quick changed CPU with same PMAP1");
#endif
static int PMAP1changed;
SYSCTL_INT(_debug, OID_AUTO, PMAP1changed, CTLFLAG_RD,
	   &PMAP1changed, 0,
	   "Number of times pmap_pte_quick changed PMAP1");
static int PMAP1unchanged;
SYSCTL_INT(_debug, OID_AUTO, PMAP1unchanged, CTLFLAG_RD,
	   &PMAP1unchanged, 0,
	   "Number of times pmap_pte_quick didn't change PMAP1");

static PMAP_INLINE void	free_pv_entry(pv_entry_t pv);
static pv_entry_t get_pv_entry(void);
static void	pmap_clear_ptes(vm_page_t m, int bit);

static int pmap_remove_pte(pmap_t pmap, pt_entry_t *ptq, vm_offset_t sva);
static void pmap_remove_page(struct pmap *pmap, vm_offset_t va);
static int pmap_remove_entry(struct pmap *pmap, vm_page_t m,
					vm_offset_t va);
static void pmap_insert_entry(pmap_t pmap, vm_offset_t va,
		vm_page_t mpte, vm_page_t m);

static vm_page_t pmap_allocpte(pmap_t pmap, vm_offset_t va);

static vm_page_t _pmap_allocpte(pmap_t pmap, unsigned ptepindex);
static pt_entry_t *pmap_pte_quick(pmap_t pmap, vm_offset_t va);
static int pmap_unuse_pt(pmap_t, vm_offset_t, vm_page_t);
static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
#ifdef PAE
static void *pmap_pdpt_allocf(uma_zone_t zone, int bytes, u_int8_t *flags, int wait);
#endif

CTASSERT(1 << PDESHIFT == sizeof(pd_entry_t));
CTASSERT(1 << PTESHIFT == sizeof(pt_entry_t));

/*
 * Move the kernel virtual free pointer to the next
 * 4MB.  This is used to help improve performance
 * by using a large (4MB) page for much of the kernel
 * (.text, .data, .bss)
 */
static vm_offset_t
pmap_kmem_choose(vm_offset_t addr)
{
	vm_offset_t newaddr = addr;

#ifndef DISABLE_PSE
	if (cpu_feature & CPUID_PSE)
		newaddr = (addr + PDRMASK) & ~PDRMASK;
#endif
	return newaddr;
}

/*
 *	Bootstrap the system enough to run with virtual memory.
 *
 *	On the i386 this is called after mapping has already been enabled
 *	and just syncs the pmap module with what has already been done.
 *	[We can't call it easily with mapping off since the kernel is not
 *	mapped with PA == VA, hence we would have to relocate every address
 *	from the linked base (virtual) address "KERNBASE" to the actual
 *	(physical) address starting relative to 0]
 */
void
pmap_bootstrap(firstaddr, loadaddr)
	vm_paddr_t firstaddr;
	vm_paddr_t loadaddr;
{
	vm_offset_t va;
	pt_entry_t *pte, *unused;
	int i;

	/*
	 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too
	 * large. It should instead be correctly calculated in locore.s and
	 * not based on 'first' (which is a physical address, not a virtual
	 * address, for the start of unused physical memory). The kernel
	 * page tables are NOT double mapped and thus should not be included
	 * in this calculation.
	 */
	virtual_avail = (vm_offset_t) KERNBASE + firstaddr;
	virtual_avail = pmap_kmem_choose(virtual_avail);

	virtual_end = VM_MAX_KERNEL_ADDRESS;

	/*
	 * Initialize the kernel pmap (which is statically allocated).
	 */
	PMAP_LOCK_INIT(kernel_pmap);
	kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD);
#ifdef PAE
	kernel_pmap->pm_pdpt = (pdpt_entry_t *) (KERNBASE + (u_int)IdlePDPT);
#endif
	kernel_pmap->pm_active = -1;	/* don't allow deactivation */
	TAILQ_INIT(&kernel_pmap->pm_pvlist);
	LIST_INIT(&allpmaps);
#ifdef SMP
	mtx_init(&lazypmap_lock, "lazypmap", NULL, MTX_SPIN);
#endif
	mtx_init(&allpmaps_lock, "allpmaps", NULL, MTX_SPIN);
	mtx_lock_spin(&allpmaps_lock);
	LIST_INSERT_HEAD(&allpmaps, kernel_pmap, pm_list);
	mtx_unlock_spin(&allpmaps_lock);
	nkpt = NKPT;

	/*
	 * Reserve some special page table entries/VA space for temporary
	 * mapping of pages.
	 */
#define	SYSMAP(c, p, v, n)	\
	v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);

	va = virtual_avail;
	pte = vtopte(va);

	/*
	 * CMAP1/CMAP2 are used for zeroing and copying pages.
	 * CMAP3 is used for the idle process page zeroing.
	 */
	SYSMAP(caddr_t, CMAP1, CADDR1, 1)
	SYSMAP(caddr_t, CMAP2, CADDR2, 1)
	SYSMAP(caddr_t, CMAP3, CADDR3, 1)
	*CMAP3 = 0;

	mtx_init(&CMAPCADDR12_lock, "CMAPCADDR12", NULL, MTX_DEF);

	/*
	 * Crashdump maps.
	 */
	SYSMAP(caddr_t, unused, crashdumpmap, MAXDUMPPGS)

	/*
	 * ptvmmap is used for reading arbitrary physical pages via /dev/mem.
	 */
	SYSMAP(caddr_t, unused, ptvmmap, 1)

	/*
	 * msgbufp is used to map the system message buffer.
	 */
	SYSMAP(struct msgbuf *, unused, msgbufp, atop(round_page(MSGBUF_SIZE)))

	/*
	 * ptemap is used for pmap_pte_quick
	 */
	SYSMAP(pt_entry_t *, PMAP1, PADDR1, 1);
	SYSMAP(pt_entry_t *, PMAP2, PADDR2, 1);

	virtual_avail = va;

	*CMAP1 = *CMAP2 = 0;
	for (i = 0; i < NKPT; i++)
		PTD[i] = 0;

	/* Turn on PG_G on kernel page(s) */
	pmap_set_pg();
}

/*
 * Set PG_G on kernel pages.  Only the BSP calls this when SMP is turned on.
 */
void
pmap_set_pg(void)
{
	pd_entry_t pdir;
	pt_entry_t *pte;
	vm_offset_t va, endva;
	int i;

	if (pgeflag == 0)
		return;

	i = KERNLOAD/NBPDR;
	endva = KERNBASE + KERNend;

	if (pseflag) {
		va = KERNBASE + KERNLOAD;
		while (va  < endva) {
			pdir = kernel_pmap->pm_pdir[KPTDI+i];
			pdir |= pgeflag;
			kernel_pmap->pm_pdir[KPTDI+i] = PTD[KPTDI+i] = pdir;
			invltlb();	/* Play it safe, invltlb() every time */
			i++;
			va += NBPDR;
		}
	} else {
		va = (vm_offset_t)btext;
		while (va < endva) {
			pte = vtopte(va);
			if (*pte)
				*pte |= pgeflag;
			invltlb();	/* Play it safe, invltlb() every time */
			va += PAGE_SIZE;
		}
	}
}

#ifdef PAE
static void *
pmap_pdpt_allocf(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
{
	*flags = UMA_SLAB_PRIV;
	return (contigmalloc(PAGE_SIZE, NULL, 0, 0x0ULL, 0xffffffffULL, 1, 0));
}
#endif

/*
 *	Initialize the pmap module.
 *	Called by vm_init, to initialize any structures that the pmap
 *	system needs to map virtual memory.
 *	pmap_init has been enhanced to support in a fairly consistant
 *	way, discontiguous physical memory.
 */
void
pmap_init(void)
{
	int i;

	/*
	 * Allocate memory for random pmap data structures.  Includes the
	 * pv_head_table.
	 */

	for(i = 0; i < vm_page_array_size; i++) {
		vm_page_t m;

		m = &vm_page_array[i];
		TAILQ_INIT(&m->md.pv_list);
		m->md.pv_list_count = 0;
	}

	/*
	 * init the pv free list
	 */
	pvzone = uma_zcreate("PV ENTRY", sizeof (struct pv_entry), NULL, NULL,
	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
	uma_prealloc(pvzone, MINPV);

#ifdef PAE
	pdptzone = uma_zcreate("PDPT", NPGPTD * sizeof(pdpt_entry_t), NULL,
	    NULL, NULL, NULL, (NPGPTD * sizeof(pdpt_entry_t)) - 1,
	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
	uma_zone_set_allocf(pdptzone, pmap_pdpt_allocf);
#endif

	/*
	 * Now it is safe to enable pv_table recording.
	 */
	pmap_initialized = TRUE;
}

/*
 * Initialize the address space (zone) for the pv_entries.  Set a
 * high water mark so that the system can recover from excessive
 * numbers of pv entries.
 */
void
pmap_init2()
{
	int shpgperproc = PMAP_SHPGPERPROC;

	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
	pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
	pv_entry_high_water = 9 * (pv_entry_max / 10);
	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
}


/***************************************************
 * Low level helper routines.....
 ***************************************************/

#if defined(PMAP_DIAGNOSTIC)

/*
 * This code checks for non-writeable/modified pages.
 * This should be an invalid condition.
 */
static int
pmap_nw_modified(pt_entry_t ptea)
{
	int pte;

	pte = (int) ptea;

	if ((pte & (PG_M|PG_RW)) == PG_M)
		return 1;
	else
		return 0;
}
#endif


/*
 * this routine defines the region(s) of memory that should
 * not be tested for the modified bit.
 */
static PMAP_INLINE int
pmap_track_modified(vm_offset_t va)
{
	if ((va < kmi.clean_sva) || (va >= kmi.clean_eva))
		return 1;
	else
		return 0;
}

#ifdef I386_CPU
/*
 * i386 only has "invalidate everything" and no SMP to worry about.
 */
PMAP_INLINE void
pmap_invalidate_page(pmap_t pmap, vm_offset_t va)
{

	if (pmap == kernel_pmap || pmap->pm_active)
		invltlb();
}

PMAP_INLINE void
pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
{

	if (pmap == kernel_pmap || pmap->pm_active)
		invltlb();
}

PMAP_INLINE void
pmap_invalidate_all(pmap_t pmap)
{

	if (pmap == kernel_pmap || pmap->pm_active)
		invltlb();
}
#else /* !I386_CPU */
#ifdef SMP
/*
 * For SMP, these functions have to use the IPI mechanism for coherence.
 */
void
pmap_invalidate_page(pmap_t pmap, vm_offset_t va)
{
	u_int cpumask;
	u_int other_cpus;

	if (smp_started) {
		if (!(read_eflags() & PSL_I))
			panic("%s: interrupts disabled", __func__);
		mtx_lock_spin(&smp_tlb_mtx);
	} else
		critical_enter();
	/*
	 * We need to disable interrupt preemption but MUST NOT have
	 * interrupts disabled here.
	 * XXX we may need to hold schedlock to get a coherent pm_active
	 * XXX critical sections disable interrupts again
	 */
	if (pmap == kernel_pmap || pmap->pm_active == all_cpus) {
		invlpg(va);
		smp_invlpg(va);
	} else {
		cpumask = PCPU_GET(cpumask);
		other_cpus = PCPU_GET(other_cpus);
		if (pmap->pm_active & cpumask)
			invlpg(va);
		if (pmap->pm_active & other_cpus)
			smp_masked_invlpg(pmap->pm_active & other_cpus, va);
	}
	if (smp_started)
		mtx_unlock_spin(&smp_tlb_mtx);
	else
		critical_exit();
}

void
pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
{
	u_int cpumask;
	u_int other_cpus;
	vm_offset_t addr;

	if (smp_started) {
		if (!(read_eflags() & PSL_I))
			panic("%s: interrupts disabled", __func__);
		mtx_lock_spin(&smp_tlb_mtx);
	} else
		critical_enter();
	/*
	 * We need to disable interrupt preemption but MUST NOT have
	 * interrupts disabled here.
	 * XXX we may need to hold schedlock to get a coherent pm_active
	 * XXX critical sections disable interrupts again
	 */
	if (pmap == kernel_pmap || pmap->pm_active == all_cpus) {
		for (addr = sva; addr < eva; addr += PAGE_SIZE)
			invlpg(addr);
		smp_invlpg_range(sva, eva);
	} else {
		cpumask = PCPU_GET(cpumask);
		other_cpus = PCPU_GET(other_cpus);
		if (pmap->pm_active & cpumask)
			for (addr = sva; addr < eva; addr += PAGE_SIZE)
				invlpg(addr);
		if (pmap->pm_active & other_cpus)
			smp_masked_invlpg_range(pmap->pm_active & other_cpus,
			    sva, eva);
	}
	if (smp_started)
		mtx_unlock_spin(&smp_tlb_mtx);
	else
		critical_exit();
}

void
pmap_invalidate_all(pmap_t pmap)
{
	u_int cpumask;
	u_int other_cpus;

	if (smp_started) {
		if (!(read_eflags() & PSL_I))
			panic("%s: interrupts disabled", __func__);
		mtx_lock_spin(&smp_tlb_mtx);
	} else
		critical_enter();
	/*
	 * We need to disable interrupt preemption but MUST NOT have
	 * interrupts disabled here.
	 * XXX we may need to hold schedlock to get a coherent pm_active
	 * XXX critical sections disable interrupts again
	 */
	if (pmap == kernel_pmap || pmap->pm_active == all_cpus) {
		invltlb();
		smp_invltlb();
	} else {
		cpumask = PCPU_GET(cpumask);
		other_cpus = PCPU_GET(other_cpus);
		if (pmap->pm_active & cpumask)
			invltlb();
		if (pmap->pm_active & other_cpus)
			smp_masked_invltlb(pmap->pm_active & other_cpus);
	}
	if (smp_started)
		mtx_unlock_spin(&smp_tlb_mtx);
	else
		critical_exit();
}
#else /* !SMP */
/*
 * Normal, non-SMP, 486+ invalidation functions.
 * We inline these within pmap.c for speed.
 */
PMAP_INLINE void
pmap_invalidate_page(pmap_t pmap, vm_offset_t va)
{

	if (pmap == kernel_pmap || pmap->pm_active)
		invlpg(va);
}

PMAP_INLINE void
pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
{
	vm_offset_t addr;

	if (pmap == kernel_pmap || pmap->pm_active)
		for (addr = sva; addr < eva; addr += PAGE_SIZE)
			invlpg(addr);
}

PMAP_INLINE void
pmap_invalidate_all(pmap_t pmap)
{

	if (pmap == kernel_pmap || pmap->pm_active)
		invltlb();
}
#endif /* !SMP */
#endif /* !I386_CPU */

/*
 * Are we current address space or kernel?  N.B. We return FALSE when
 * a pmap's page table is in use because a kernel thread is borrowing
 * it.  The borrowed page table can change spontaneously, making any
 * dependence on its continued use subject to a race condition.
 */
static __inline int
pmap_is_current(pmap_t pmap)
{

	return (pmap == kernel_pmap ||
		(pmap == vmspace_pmap(curthread->td_proc->p_vmspace) &&
	    (pmap->pm_pdir[PTDPTDI] & PG_FRAME) == (PTDpde[0] & PG_FRAME)));
}

/*
 * If the given pmap is not the current pmap, Giant must be held.
 */
pt_entry_t *
pmap_pte(pmap_t pmap, vm_offset_t va)
{
	pd_entry_t newpf;
	pd_entry_t *pde;

	pde = pmap_pde(pmap, va);
	if (*pde & PG_PS)
		return (pde);
	if (*pde != 0) {
		/* are we current address space or kernel? */
		if (pmap_is_current(pmap))
			return (vtopte(va));
		GIANT_REQUIRED;
		newpf = *pde & PG_FRAME;
		if ((*PMAP2 & PG_FRAME) != newpf) {
			*PMAP2 = newpf | PG_RW | PG_V | PG_A | PG_M;
			pmap_invalidate_page(kernel_pmap, (vm_offset_t)PADDR2);
		}
		return (PADDR2 + (i386_btop(va) & (NPTEPG - 1)));
	}
	return (0);
}

static __inline void
invlcaddr(void *caddr)
{
#ifdef I386_CPU
	invltlb();
#else
	invlpg((u_int)caddr);
#endif
}

/*
 * Super fast pmap_pte routine best used when scanning
 * the pv lists.  This eliminates many coarse-grained
 * invltlb calls.  Note that many of the pv list
 * scans are across different pmaps.  It is very wasteful
 * to do an entire invltlb for checking a single mapping.
 *
 * If the given pmap is not the current pmap, vm_page_queue_mtx
 * must be held and curthread pinned to a CPU.
 */
static pt_entry_t *
pmap_pte_quick(pmap_t pmap, vm_offset_t va)
{
	pd_entry_t newpf;
	pd_entry_t *pde;

	pde = pmap_pde(pmap, va);
	if (*pde & PG_PS)
		return (pde);
	if (*pde != 0) {
		/* are we current address space or kernel? */
		if (pmap_is_current(pmap))
			return (vtopte(va));
		mtx_assert(&vm_page_queue_mtx, MA_OWNED);
		KASSERT(curthread->td_pinned > 0, ("curthread not pinned"));
		newpf = *pde & PG_FRAME;
		if ((*PMAP1 & PG_FRAME) != newpf) {
			*PMAP1 = newpf | PG_RW | PG_V | PG_A | PG_M;
#ifdef SMP
			PMAP1cpu = PCPU_GET(cpuid);
#endif
			invlcaddr(PADDR1);
			PMAP1changed++;
		} else
#ifdef SMP
		if (PMAP1cpu != PCPU_GET(cpuid)) {
			PMAP1cpu = PCPU_GET(cpuid);
			invlcaddr(PADDR1);
			PMAP1changedcpu++;
		} else
#endif
			PMAP1unchanged++;
		return (PADDR1 + (i386_btop(va) & (NPTEPG - 1)));
	}
	return (0);
}

/*
 *	Routine:	pmap_extract
 *	Function:
 *		Extract the physical page address associated
 *		with the given map/virtual_address pair.
 */
vm_paddr_t
pmap_extract(pmap, va)
	register pmap_t pmap;
	vm_offset_t va;
{
	vm_paddr_t rtval;
	pt_entry_t *pte;
	pd_entry_t pde;

	if (pmap == 0)
		return 0;
	PMAP_LOCK(pmap);
	pde = pmap->pm_pdir[va >> PDRSHIFT];
	if (pde != 0) {
		if ((pde & PG_PS) != 0) {
			rtval = (pde & ~PDRMASK) | (va & PDRMASK);
			PMAP_UNLOCK(pmap);
			return rtval;
		}
		pte = pmap_pte(pmap, va);
		rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
		PMAP_UNLOCK(pmap);
		return rtval;
	}
	PMAP_UNLOCK(pmap);
	return 0;

}

/*
 *	Routine:	pmap_extract_and_hold
 *	Function:
 *		Atomically extract and hold the physical page
 *		with the given pmap and virtual address pair
 *		if that mapping permits the given protection.
 */
vm_page_t
pmap_extract_and_hold(pmap_t pmap, vm_offset_t va, vm_prot_t prot)
{
	pd_entry_t pde;
	pt_entry_t pte;
	vm_page_t m;

	m = NULL;
	if (pmap == NULL)
		return (m);
	vm_page_lock_queues();
	PMAP_LOCK(pmap);
	pde = *pmap_pde(pmap, va);
	if (pde != 0) {
		if (pde & PG_PS) {
			if ((pde & PG_RW) || (prot & VM_PROT_WRITE) == 0) {
				m = PHYS_TO_VM_PAGE((pde & ~PDRMASK) |
				    (va & PDRMASK));
				vm_page_hold(m);
			}
		} else {
			pte = *pmap_pte_quick(pmap, va);
			if (pte != 0 &&
			    ((pte & PG_RW) || (prot & VM_PROT_WRITE) == 0)) {
				m = PHYS_TO_VM_PAGE(pte & PG_FRAME);
				vm_page_hold(m);
			}
		}
	}
	vm_page_unlock_queues();
	PMAP_UNLOCK(pmap);
	return (m);
}

/***************************************************
 * Low level mapping routines.....
 ***************************************************/

/*
 * Add a wired page to the kva.
 * Note: not SMP coherent.
 */
PMAP_INLINE void
pmap_kenter(vm_offset_t va, vm_paddr_t pa)
{
	pt_entry_t *pte;

	pte = vtopte(va);
	pte_store(pte, pa | PG_RW | PG_V | pgeflag);
}

/*
 * Remove a page from the kernel pagetables.
 * Note: not SMP coherent.
 */
PMAP_INLINE void
pmap_kremove(vm_offset_t va)
{
	pt_entry_t *pte;

	pte = vtopte(va);
	pte_clear(pte);
}

/*
 *	Used to map a range of physical addresses into kernel
 *	virtual address space.
 *
 *	The value passed in '*virt' is a suggested virtual address for
 *	the mapping. Architectures which can support a direct-mapped
 *	physical to virtual region can return the appropriate address
 *	within that region, leaving '*virt' unchanged. Other
 *	architectures should map the pages starting at '*virt' and
 *	update '*virt' with the first usable address after the mapped
 *	region.
 */
vm_offset_t
pmap_map(vm_offset_t *virt, vm_paddr_t start, vm_paddr_t end, int prot)
{
	vm_offset_t va, sva;

	va = sva = *virt;
	while (start < end) {
		pmap_kenter(va, start);
		va += PAGE_SIZE;
		start += PAGE_SIZE;
	}
	pmap_invalidate_range(kernel_pmap, sva, va);
	*virt = va;
	return (sva);
}


/*
 * Add a list of wired pages to the kva
 * this routine is only used for temporary
 * kernel mappings that do not need to have
 * page modification or references recorded.
 * Note that old mappings are simply written
 * over.  The page *must* be wired.
 * Note: SMP coherent.  Uses a ranged shootdown IPI.
 */
void
pmap_qenter(vm_offset_t sva, vm_page_t *m, int count)
{
	vm_offset_t va;

	va = sva;
	while (count-- > 0) {
		pmap_kenter(va, VM_PAGE_TO_PHYS(*m));
		va += PAGE_SIZE;
		m++;
	}
	pmap_invalidate_range(kernel_pmap, sva, va);
}

/*
 * This routine tears out page mappings from the
 * kernel -- it is meant only for temporary mappings.
 * Note: SMP coherent.  Uses a ranged shootdown IPI.
 */
void
pmap_qremove(vm_offset_t sva, int count)
{
	vm_offset_t va;

	va = sva;
	while (count-- > 0) {
		pmap_kremove(va);
		va += PAGE_SIZE;
	}
	pmap_invalidate_range(kernel_pmap, sva, va);
}

/***************************************************
 * Page table page management routines.....
 ***************************************************/

/*
 * This routine unholds page table pages, and if the hold count
 * drops to zero, then it decrements the wire count.
 */
static int
_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
{

	while (vm_page_sleep_if_busy(m, FALSE, "pmuwpt"))
		vm_page_lock_queues();

	if (m->hold_count == 0) {
		vm_offset_t pteva;
		/*
		 * unmap the page table page
		 */
		pmap->pm_pdir[m->pindex] = 0;
		--pmap->pm_stats.resident_count;
		/*
		 * We never unwire a kernel page table page, making a
		 * check for the kernel_pmap unnecessary.
		 */
		if ((pmap->pm_pdir[PTDPTDI] & PG_FRAME) == (PTDpde[0] & PG_FRAME)) {
			/*
			 * Do an invltlb to make the invalidated mapping
			 * take effect immediately.
			 */
			pteva = VM_MAXUSER_ADDRESS + i386_ptob(m->pindex);
			pmap_invalidate_page(pmap, pteva);
		}

		/*
		 * If the page is finally unwired, simply free it.
		 */
		--m->wire_count;
		if (m->wire_count == 0) {
			vm_page_busy(m);
			vm_page_free_zero(m);
			atomic_subtract_int(&cnt.v_wire_count, 1);
		}
		return 1;
	}
	return 0;
}

static PMAP_INLINE int
pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
{
	vm_page_unhold(m);
	if (m->hold_count == 0)
		return _pmap_unwire_pte_hold(pmap, m);
	else
		return 0;
}

/*
 * After removing a page table entry, this routine is used to
 * conditionally free the page, and manage the hold/wire counts.
 */
static int
pmap_unuse_pt(pmap_t pmap, vm_offset_t va, vm_page_t mpte)
{

	if (va >= VM_MAXUSER_ADDRESS)
		return 0;

	return pmap_unwire_pte_hold(pmap, mpte);
}

void
pmap_pinit0(pmap)
	struct pmap *pmap;
{

	PMAP_LOCK_INIT(pmap);
	pmap->pm_pdir = (pd_entry_t *)(KERNBASE + (vm_offset_t)IdlePTD);
#ifdef PAE
	pmap->pm_pdpt = (pdpt_entry_t *)(KERNBASE + (vm_offset_t)IdlePDPT);
#endif
	pmap->pm_active = 0;
	PCPU_SET(curpmap, pmap);
	TAILQ_INIT(&pmap->pm_pvlist);
	bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
	mtx_lock_spin(&allpmaps_lock);
	LIST_INSERT_HEAD(&allpmaps, pmap, pm_list);
	mtx_unlock_spin(&allpmaps_lock);
}

/*
 * Initialize a preallocated and zeroed pmap structure,
 * such as one in a vmspace structure.
 */
void
pmap_pinit(pmap)
	register struct pmap *pmap;
{
	vm_page_t m, ptdpg[NPGPTD];
	vm_paddr_t pa;
	static int color;
	int i;

	PMAP_LOCK_INIT(pmap);

	/*
	 * No need to allocate page table space yet but we do need a valid
	 * page directory table.
	 */
	if (pmap->pm_pdir == NULL) {
		pmap->pm_pdir = (pd_entry_t *)kmem_alloc_nofault(kernel_map,
		    NBPTD);
#ifdef PAE
		pmap->pm_pdpt = uma_zalloc(pdptzone, M_WAITOK | M_ZERO);
		KASSERT(((vm_offset_t)pmap->pm_pdpt &
		    ((NPGPTD * sizeof(pdpt_entry_t)) - 1)) == 0,
		    ("pmap_pinit: pdpt misaligned"));
		KASSERT(pmap_kextract((vm_offset_t)pmap->pm_pdpt) < (4ULL<<30),
		    ("pmap_pinit: pdpt above 4g"));
#endif
	}

	/*
	 * allocate the page directory page(s)
	 */
	for (i = 0; i < NPGPTD;) {
		m = vm_page_alloc(NULL, color++,
		    VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED |
		    VM_ALLOC_ZERO);
		if (m == NULL)
			VM_WAIT;
		else {
			ptdpg[i++] = m;
		}
	}

	pmap_qenter((vm_offset_t)pmap->pm_pdir, ptdpg, NPGPTD);

	for (i = 0; i < NPGPTD; i++) {
		if ((ptdpg[i]->flags & PG_ZERO) == 0)
			bzero(pmap->pm_pdir + (i * NPDEPG), PAGE_SIZE);
	}

	mtx_lock_spin(&allpmaps_lock);
	LIST_INSERT_HEAD(&allpmaps, pmap, pm_list);
	mtx_unlock_spin(&allpmaps_lock);
	/* Wire in kernel global address entries. */
	/* XXX copies current process, does not fill in MPPTDI */
	bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * sizeof(pd_entry_t));
#ifdef SMP
	pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
#endif

	/* install self-referential address mapping entry(s) */
	for (i = 0; i < NPGPTD; i++) {
		pa = VM_PAGE_TO_PHYS(ptdpg[i]);
		pmap->pm_pdir[PTDPTDI + i] = pa | PG_V | PG_RW | PG_A | PG_M;
#ifdef PAE
		pmap->pm_pdpt[i] = pa | PG_V;
#endif
	}

	pmap->pm_active = 0;
	TAILQ_INIT(&pmap->pm_pvlist);
	bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
}

/*
 * this routine is called if the page table page is not
 * mapped correctly.
 */
static vm_page_t
_pmap_allocpte(pmap, ptepindex)
	pmap_t	pmap;
	unsigned ptepindex;
{
	vm_paddr_t ptepa;
	vm_page_t m;

	/*
	 * Allocate a page table page.
	 */
	if ((m = vm_page_alloc(NULL, ptepindex, VM_ALLOC_NOOBJ |
	    VM_ALLOC_WIRED | VM_ALLOC_ZERO)) == NULL) {
		VM_WAIT;
		/*
		 * Indicate the need to retry.  While waiting, the page table
		 * page may have been allocated.
		 */
		return (NULL);
	}
	if ((m->flags & PG_ZERO) == 0)
		pmap_zero_page(m);

	KASSERT(m->queue == PQ_NONE,
		("_pmap_allocpte: %p->queue != PQ_NONE", m));

	/*
	 * Increment the hold count for the page table page
	 * (denoting a new mapping.)
	 */
	m->hold_count++;

	/*
	 * Map the pagetable page into the process address space, if
	 * it isn't already there.
	 */

	pmap->pm_stats.resident_count++;

	ptepa = VM_PAGE_TO_PHYS(m);
	pmap->pm_pdir[ptepindex] =
		(pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);

	return m;
}

static vm_page_t
pmap_allocpte(pmap_t pmap, vm_offset_t va)
{
	unsigned ptepindex;
	pd_entry_t ptepa;
	vm_page_t m;

	/*
	 * Calculate pagetable page index
	 */
	ptepindex = va >> PDRSHIFT;
retry:
	/*
	 * Get the page directory entry
	 */
	ptepa = pmap->pm_pdir[ptepindex];

	/*
	 * This supports switching from a 4MB page to a
	 * normal 4K page.
	 */
	if (ptepa & PG_PS) {
		pmap->pm_pdir[ptepindex] = 0;
		ptepa = 0;
		pmap_invalidate_all(kernel_pmap);
	}

	/*
	 * If the page table page is mapped, we just increment the
	 * hold count, and activate it.
	 */
	if (ptepa) {
		m = PHYS_TO_VM_PAGE(ptepa);
		m->hold_count++;
	} else {
		/*
		 * Here if the pte page isn't mapped, or if it has
		 * been deallocated.
		 */
		m = _pmap_allocpte(pmap, ptepindex);
		if (m == NULL)
			goto retry;
	}
	return (m);
}


/***************************************************
* Pmap allocation/deallocation routines.
 ***************************************************/

#ifdef SMP
/*
 * Deal with a SMP shootdown of other users of the pmap that we are
 * trying to dispose of.  This can be a bit hairy.
 */
static u_int *lazymask;
static u_int lazyptd;
static volatile u_int lazywait;

void pmap_lazyfix_action(void);

void
pmap_lazyfix_action(void)
{
	u_int mymask = PCPU_GET(cpumask);

	if (rcr3() == lazyptd)
		load_cr3(PCPU_GET(curpcb)->pcb_cr3);
	atomic_clear_int(lazymask, mymask);
	atomic_store_rel_int(&lazywait, 1);
}

static void
pmap_lazyfix_self(u_int mymask)
{

	if (rcr3() == lazyptd)
		load_cr3(PCPU_GET(curpcb)->pcb_cr3);
	atomic_clear_int(lazymask, mymask);
}


static void
pmap_lazyfix(pmap_t pmap)
{
	u_int mymask = PCPU_GET(cpumask);
	u_int mask;
	register u_int spins;

	while ((mask = pmap->pm_active) != 0) {
		spins = 50000000;
		mask = mask & -mask;	/* Find least significant set bit */
		mtx_lock_spin(&lazypmap_lock);
#ifdef PAE
		lazyptd = vtophys(pmap->pm_pdpt);
#else
		lazyptd = vtophys(pmap->pm_pdir);
#endif
		if (mask == mymask) {
			lazymask = &pmap->pm_active;
			pmap_lazyfix_self(mymask);
		} else {
			atomic_store_rel_int((u_int *)&lazymask,
			    (u_int)&pmap->pm_active);
			atomic_store_rel_int(&lazywait, 0);
			ipi_selected(mask, IPI_LAZYPMAP);
			while (lazywait == 0) {
				ia32_pause();
				if (--spins == 0)
					break;
			}
		}
		mtx_unlock_spin(&lazypmap_lock);
		if (spins == 0)
			printf("pmap_lazyfix: spun for 50000000\n");
	}
}

#else	/* SMP */

/*
 * Cleaning up on uniprocessor is easy.  For various reasons, we're
 * unlikely to have to even execute this code, including the fact
 * that the cleanup is deferred until the parent does a wait(2), which
 * means that another userland process has run.
 */
static void
pmap_lazyfix(pmap_t pmap)
{
	u_int cr3;

	cr3 = vtophys(pmap->pm_pdir);
	if (cr3 == rcr3()) {
		load_cr3(PCPU_GET(curpcb)->pcb_cr3);
		pmap->pm_active &= ~(PCPU_GET(cpumask));
	}
}
#endif	/* SMP */

/*
 * Release any resources held by the given physical map.
 * Called when a pmap initialized by pmap_pinit is being released.
 * Should only be called if the map contains no valid mappings.
 */
void
pmap_release(pmap_t pmap)
{
	vm_page_t m, ptdpg[NPGPTD];
	int i;

	KASSERT(pmap->pm_stats.resident_count == 0,
	    ("pmap_release: pmap resident count %ld != 0",
	    pmap->pm_stats.resident_count));

	pmap_lazyfix(pmap);
	mtx_lock_spin(&allpmaps_lock);
	LIST_REMOVE(pmap, pm_list);
	mtx_unlock_spin(&allpmaps_lock);

	for (i = 0; i < NPGPTD; i++)
		ptdpg[i] = PHYS_TO_VM_PAGE(pmap->pm_pdir[PTDPTDI + i]);

	bzero(pmap->pm_pdir + PTDPTDI, (nkpt + NPGPTD) *
	    sizeof(*pmap->pm_pdir));
#ifdef SMP
	pmap->pm_pdir[MPPTDI] = 0;
#endif

	pmap_qremove((vm_offset_t)pmap->pm_pdir, NPGPTD);

	vm_page_lock_queues();
	for (i = 0; i < NPGPTD; i++) {
		m = ptdpg[i];
#ifdef PAE
		KASSERT(VM_PAGE_TO_PHYS(m) == (pmap->pm_pdpt[i] & PG_FRAME),
		    ("pmap_release: got wrong ptd page"));
#endif
		m->wire_count--;
		atomic_subtract_int(&cnt.v_wire_count, 1);
		vm_page_free_zero(m);
	}
	vm_page_unlock_queues();
	PMAP_LOCK_DESTROY(pmap);
}

static int
kvm_size(SYSCTL_HANDLER_ARGS)
{
	unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;

	return sysctl_handle_long(oidp, &ksize, 0, req);
}
SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
    0, 0, kvm_size, "IU", "Size of KVM");

static int
kvm_free(SYSCTL_HANDLER_ARGS)
{
	unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;

	return sysctl_handle_long(oidp, &kfree, 0, req);
}
SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
    0, 0, kvm_free, "IU", "Amount of KVM free");

/*
 * grow the number of kernel page table entries, if needed
 */
void
pmap_growkernel(vm_offset_t addr)
{
	struct pmap *pmap;
	vm_paddr_t ptppaddr;
	vm_page_t nkpg;
	pd_entry_t newpdir;
	pt_entry_t *pde;

	mtx_assert(&kernel_map->system_mtx, MA_OWNED);
	if (kernel_vm_end == 0) {
		kernel_vm_end = KERNBASE;
		nkpt = 0;
		while (pdir_pde(PTD, kernel_vm_end)) {
			kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
			nkpt++;
		}
	}
	addr = roundup2(addr, PAGE_SIZE * NPTEPG);
	while (kernel_vm_end < addr) {
		if (pdir_pde(PTD, kernel_vm_end)) {
			kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
			continue;
		}

		/*
		 * This index is bogus, but out of the way
		 */
		nkpg = vm_page_alloc(NULL, nkpt,
		    VM_ALLOC_NOOBJ | VM_ALLOC_SYSTEM | VM_ALLOC_WIRED);
		if (!nkpg)
			panic("pmap_growkernel: no memory to grow kernel");

		nkpt++;

		pmap_zero_page(nkpg);
		ptppaddr = VM_PAGE_TO_PHYS(nkpg);
		newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
		pdir_pde(PTD, kernel_vm_end) = newpdir;

		mtx_lock_spin(&allpmaps_lock);
		LIST_FOREACH(pmap, &allpmaps, pm_list) {
			pde = pmap_pde(pmap, kernel_vm_end);
			pde_store(pde, newpdir);
		}
		mtx_unlock_spin(&allpmaps_lock);
		kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
	}
}


/***************************************************
 * page management routines.
 ***************************************************/

/*
 * free the pv_entry back to the free list
 */
static PMAP_INLINE void
free_pv_entry(pv_entry_t pv)
{
	pv_entry_count--;
	uma_zfree(pvzone, pv);
}

/*
 * get a new pv_entry, allocating a block from the system
 * when needed.
 * the memory allocation is performed bypassing the malloc code
 * because of the possibility of allocations at interrupt time.
 */
static pv_entry_t
get_pv_entry(void)
{
	pv_entry_count++;
	if (pv_entry_high_water &&
		(pv_entry_count > pv_entry_high_water) &&
		(pmap_pagedaemon_waken == 0)) {
		pmap_pagedaemon_waken = 1;
		wakeup (&vm_pages_needed);
	}
	return uma_zalloc(pvzone, M_NOWAIT);
}


static int
pmap_remove_entry(pmap_t pmap, vm_page_t m, vm_offset_t va)
{
	pv_entry_t pv;
	int rtval;

	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
	if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
		TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
			if (pmap == pv->pv_pmap && va == pv->pv_va)
				break;
		}
	} else {
		TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
			if (va == pv->pv_va)
				break;
		}
	}

	rtval = 0;
	if (pv) {
		rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
		TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
		m->md.pv_list_count--;
		if (TAILQ_FIRST(&m->md.pv_list) == NULL)
			vm_page_flag_clear(m, PG_WRITEABLE);

		TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
		free_pv_entry(pv);
	}

	return rtval;
}

/*
 * Create a pv entry for page at pa for
 * (pmap, va).
 */
static void
pmap_insert_entry(pmap_t pmap, vm_offset_t va, vm_page_t mpte, vm_page_t m)
{

	pv_entry_t pv;

	pv = get_pv_entry();
	pv->pv_va = va;
	pv->pv_pmap = pmap;
	pv->pv_ptem = mpte;

	vm_page_lock_queues();
	TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
	TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
	m->md.pv_list_count++;

	vm_page_unlock_queues();
}

/*
 * pmap_remove_pte: do the things to unmap a page in a process
 */
static int
pmap_remove_pte(pmap_t pmap, pt_entry_t *ptq, vm_offset_t va)
{
	pt_entry_t oldpte;
	vm_page_t m, mpte;

	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
	oldpte = pte_load_clear(ptq);
	if (oldpte & PG_W)
		pmap->pm_stats.wired_count -= 1;
	/*
	 * Machines that don't support invlpg, also don't support
	 * PG_G.
	 */
	if (oldpte & PG_G)
		pmap_invalidate_page(kernel_pmap, va);
	pmap->pm_stats.resident_count -= 1;
	if (oldpte & PG_MANAGED) {
		m = PHYS_TO_VM_PAGE(oldpte);
		if (oldpte & PG_M) {
#if defined(PMAP_DIAGNOSTIC)
			if (pmap_nw_modified((pt_entry_t) oldpte)) {
				printf(
	"pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
				    va, oldpte);
			}
#endif
			if (pmap_track_modified(va))
				vm_page_dirty(m);
		}
		if (oldpte & PG_A)
			vm_page_flag_set(m, PG_REFERENCED);
		return pmap_remove_entry(pmap, m, va);
	} else {
		mpte = PHYS_TO_VM_PAGE(*pmap_pde(pmap, va));
		return pmap_unuse_pt(pmap, va, mpte);
	}
}

/*
 * Remove a single page from a process address space
 */
static void
pmap_remove_page(pmap_t pmap, vm_offset_t va)
{
	pt_entry_t *pte;

	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
	if ((pte = pmap_pte(pmap, va)) == NULL || *pte == 0)
		return;
	pmap_remove_pte(pmap, pte, va);
	pmap_invalidate_page(pmap, va);
}

/*
 *	Remove the given range of addresses from the specified map.
 *
 *	It is assumed that the start and end are properly
 *	rounded to the page size.
 */
void
pmap_remove(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
{
	vm_offset_t pdnxt;
	pd_entry_t ptpaddr;
	pt_entry_t *pte;
	int anyvalid;

	if (pmap == NULL)
		return;

	/*
	 * Perform an unsynchronized read.  This is, however, safe.
	 */
	if (pmap->pm_stats.resident_count == 0)
		return;
	PMAP_LOCK(pmap);

	/*
	 * special handling of removing one page.  a very
	 * common operation and easy to short circuit some
	 * code.
	 */
	if ((sva + PAGE_SIZE == eva) &&
	    ((pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
		pmap_remove_page(pmap, sva);
		PMAP_UNLOCK(pmap);
		return;
	}

	anyvalid = 0;

	for (; sva < eva; sva = pdnxt) {
		unsigned pdirindex;

		/*
		 * Calculate index for next page table.
		 */
		pdnxt = (sva + NBPDR) & ~PDRMASK;
		if (pmap->pm_stats.resident_count == 0)
			break;

		pdirindex = sva >> PDRSHIFT;
		ptpaddr = pmap->pm_pdir[pdirindex];

		/*
		 * Weed out invalid mappings. Note: we assume that the page
		 * directory table is always allocated, and in kernel virtual.
		 */
		if (ptpaddr == 0)
			continue;

		/*
		 * Check for large page.
		 */
		if ((ptpaddr & PG_PS) != 0) {
			pmap->pm_pdir[pdirindex] = 0;
			pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
			anyvalid = 1;
			continue;
		}

		/*
		 * Limit our scan to either the end of the va represented
		 * by the current page table page, or to the end of the
		 * range being removed.
		 */
		if (pdnxt > eva)
			pdnxt = eva;

		for (; sva != pdnxt; sva += PAGE_SIZE) {
			if ((pte = pmap_pte(pmap, sva)) == NULL ||
			    *pte == 0)
				continue;
			anyvalid = 1;
			if (pmap_remove_pte(pmap, pte, sva))
				break;
		}
	}

	if (anyvalid)
		pmap_invalidate_all(pmap);
	PMAP_UNLOCK(pmap);
}

/*
 *	Routine:	pmap_remove_all
 *	Function:
 *		Removes this physical page from
 *		all physical maps in which it resides.
 *		Reflects back modify bits to the pager.
 *
 *	Notes:
 *		Original versions of this routine were very
 *		inefficient because they iteratively called
 *		pmap_remove (slow...)
 */

void
pmap_remove_all(vm_page_t m)
{
	register pv_entry_t pv;
	pt_entry_t *pte, tpte;

#if defined(PMAP_DIAGNOSTIC)
	/*
	 * XXX This makes pmap_remove_all() illegal for non-managed pages!
	 */
	if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
		panic("pmap_remove_all: illegal for unmanaged page, va: 0x%x",
		    VM_PAGE_TO_PHYS(m));
	}
#endif
	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
	sched_pin();
	while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
		PMAP_LOCK(pv->pv_pmap);
		pv->pv_pmap->pm_stats.resident_count--;
		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
		tpte = pte_load_clear(pte);
		if (tpte & PG_W)
			pv->pv_pmap->pm_stats.wired_count--;
		if (tpte & PG_A)
			vm_page_flag_set(m, PG_REFERENCED);

		/*
		 * Update the vm_page_t clean and reference bits.
		 */
		if (tpte & PG_M) {
#if defined(PMAP_DIAGNOSTIC)
			if (pmap_nw_modified((pt_entry_t) tpte)) {
				printf(
	"pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
				    pv->pv_va, tpte);
			}
#endif
			if (pmap_track_modified(pv->pv_va))
				vm_page_dirty(m);
		}
		pmap_invalidate_page(pv->pv_pmap, pv->pv_va);
		TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
		TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
		m->md.pv_list_count--;
		pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
		PMAP_UNLOCK(pv->pv_pmap);
		free_pv_entry(pv);
	}
	vm_page_flag_clear(m, PG_WRITEABLE);
	sched_unpin();
}

/*
 *	Set the physical protection on the
 *	specified range of this map as requested.
 */
void
pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
{
	vm_offset_t pdnxt;
	pd_entry_t ptpaddr;
	int anychanged;

	if (pmap == NULL)
		return;

	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
		pmap_remove(pmap, sva, eva);
		return;
	}

	if (prot & VM_PROT_WRITE)
		return;

	anychanged = 0;

	PMAP_LOCK(pmap);
	for (; sva < eva; sva = pdnxt) {
		unsigned pdirindex;

		pdnxt = (sva + NBPDR) & ~PDRMASK;

		pdirindex = sva >> PDRSHIFT;
		ptpaddr = pmap->pm_pdir[pdirindex];

		/*
		 * Weed out invalid mappings. Note: we assume that the page
		 * directory table is always allocated, and in kernel virtual.
		 */
		if (ptpaddr == 0)
			continue;

		/*
		 * Check for large page.
		 */
		if ((ptpaddr & PG_PS) != 0) {
			pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
			pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
			anychanged = 1;
			continue;
		}

		if (pdnxt > eva)
			pdnxt = eva;

		for (; sva != pdnxt; sva += PAGE_SIZE) {
			pt_entry_t pbits;
			pt_entry_t *pte;
			vm_page_t m;

			if ((pte = pmap_pte(pmap, sva)) == NULL)
				continue;
			pbits = *pte;
			if (pbits & PG_MANAGED) {
				m = NULL;
				if (pbits & PG_A) {
					m = PHYS_TO_VM_PAGE(pbits);
					vm_page_flag_set(m, PG_REFERENCED);
					pbits &= ~PG_A;
				}
				if ((pbits & PG_M) != 0 &&
				    pmap_track_modified(sva)) {
					if (m == NULL)
						m = PHYS_TO_VM_PAGE(pbits);
					vm_page_dirty(m);
					pbits &= ~PG_M;
				}
			}

			pbits &= ~PG_RW;

			if (pbits != *pte) {
				pte_store(pte, pbits);
				anychanged = 1;
			}
		}
	}
	if (anychanged)
		pmap_invalidate_all(pmap);
	PMAP_UNLOCK(pmap);
}

/*
 *	Insert the given physical page (p) at
 *	the specified virtual address (v) in the
 *	target physical map with the protection requested.
 *
 *	If specified, the page will be wired down, meaning
 *	that the related pte can not be reclaimed.
 *
 *	NB:  This is the only routine which MAY NOT lazy-evaluate
 *	or lose information.  That is, this routine must actually
 *	insert this page into the given map NOW.
 */
void
pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
	   boolean_t wired)
{
	vm_paddr_t pa;
	register pt_entry_t *pte;
	vm_paddr_t opa;
	pt_entry_t origpte, newpte;
	vm_page_t mpte;

	if (pmap == NULL)
		return;

	va &= PG_FRAME;
#ifdef PMAP_DIAGNOSTIC
	if (va > VM_MAX_KERNEL_ADDRESS)
		panic("pmap_enter: toobig");
	if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
		panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
#endif

	mpte = NULL;
	/*
	 * In the case that a page table page is not
	 * resident, we are creating it here.
	 */
	if (va < VM_MAXUSER_ADDRESS) {
		mpte = pmap_allocpte(pmap, va);
	}
#if 0 && defined(PMAP_DIAGNOSTIC)
	else {
		pd_entry_t *pdeaddr = pmap_pde(pmap, va);
		origpte = *pdeaddr;
		if ((origpte & PG_V) == 0) {
			panic("pmap_enter: invalid kernel page table page, pdir=%p, pde=%p, va=%p\n",
				pmap->pm_pdir[PTDPTDI], origpte, va);
		}
	}
#endif

	pte = pmap_pte(pmap, va);

	/*
	 * Page Directory table entry not valid, we need a new PT page
	 */
	if (pte == NULL) {
		panic("pmap_enter: invalid page directory pdir=%#jx, va=%#x\n",
			(uintmax_t)pmap->pm_pdir[PTDPTDI], va);
	}

	pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
	origpte = *pte;
	opa = origpte & PG_FRAME;

	if (origpte & PG_PS) {
		/*
		 * Yes, I know this will truncate upper address bits for PAE,
		 * but I'm actually more interested in the lower bits
		 */
		printf("pmap_enter: va %p, pte %p, origpte %p\n",
		    (void *)va, (void *)pte, (void *)(uintptr_t)origpte);
		panic("pmap_enter: attempted pmap_enter on 4MB page");
	}

	/*
	 * Mapping has not changed, must be protection or wiring change.
	 */
	if (origpte && (opa == pa)) {
		/*
		 * Wiring change, just update stats. We don't worry about
		 * wiring PT pages as they remain resident as long as there
		 * are valid mappings in them. Hence, if a user page is wired,
		 * the PT page will be also.
		 */
		if (wired && ((origpte & PG_W) == 0))
			pmap->pm_stats.wired_count++;
		else if (!wired && (origpte & PG_W))
			pmap->pm_stats.wired_count--;

#if defined(PMAP_DIAGNOSTIC)
		if (pmap_nw_modified((pt_entry_t) origpte)) {
			printf(
	"pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
			    va, origpte);
		}
#endif

		/*
		 * Remove extra pte reference
		 */
		if (mpte)
			mpte->hold_count--;

		/*
		 * We might be turning off write access to the page,
		 * so we go ahead and sense modify status.
		 */
		if (origpte & PG_MANAGED) {
			if ((origpte & PG_M) && pmap_track_modified(va)) {
				vm_page_t om;
				om = PHYS_TO_VM_PAGE(opa);
				vm_page_dirty(om);
			}
			pa |= PG_MANAGED;
		}
		goto validate;
	}
	/*
	 * Mapping has changed, invalidate old range and fall through to
	 * handle validating new mapping.
	 */
	if (opa) {
		int err;
		vm_page_lock_queues();
		PMAP_LOCK(pmap);
		err = pmap_remove_pte(pmap, pte, va);
		PMAP_UNLOCK(pmap);
		vm_page_unlock_queues();
		if (err)
			panic("pmap_enter: pte vanished, va: 0x%x", va);
	}

	/*
	 * Enter on the PV list if part of our managed memory. Note that we
	 * raise IPL while manipulating pv_table since pmap_enter can be
	 * called at interrupt time.
	 */
	if (pmap_initialized &&
	    (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
		pmap_insert_entry(pmap, va, mpte, m);
		pa |= PG_MANAGED;
	}

	/*
	 * Increment counters
	 */
	pmap->pm_stats.resident_count++;
	if (wired)
		pmap->pm_stats.wired_count++;

validate:
	/*
	 * Now validate mapping with desired protection/wiring.
	 */
	newpte = (pt_entry_t)(pa | PG_V);
	if ((prot & VM_PROT_WRITE) != 0)
		newpte |= PG_RW;
	if (wired)
		newpte |= PG_W;
	if (va < VM_MAXUSER_ADDRESS)
		newpte |= PG_U;
	if (pmap == kernel_pmap)
		newpte |= pgeflag;

	/*
	 * if the mapping or permission bits are different, we need
	 * to update the pte.
	 */
	if ((origpte & ~(PG_M|PG_A)) != newpte) {
		pte_store(pte, newpte | PG_A);
		/*if (origpte)*/ {
			pmap_invalidate_page(pmap, va);
		}
	}
}

/*
 * this code makes some *MAJOR* assumptions:
 * 1. Current pmap & pmap exists.
 * 2. Not wired.
 * 3. Read access.
 * 4. No page table pages.
 * 5. Tlbflush is deferred to calling procedure.
 * 6. Page IS managed.
 * but is *MUCH* faster than pmap_enter...
 */

vm_page_t
pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_page_t mpte)
{
	pt_entry_t *pte;
	vm_paddr_t pa;

	/*
	 * In the case that a page table page is not
	 * resident, we are creating it here.
	 */
	if (va < VM_MAXUSER_ADDRESS) {
		unsigned ptepindex;
		pd_entry_t ptepa;

		/*
		 * Calculate pagetable page index
		 */
		ptepindex = va >> PDRSHIFT;
		if (mpte && (mpte->pindex == ptepindex)) {
			mpte->hold_count++;
		} else {
retry:
			/*
			 * Get the page directory entry
			 */
			ptepa = pmap->pm_pdir[ptepindex];

			/*
			 * If the page table page is mapped, we just increment
			 * the hold count, and activate it.
			 */
			if (ptepa) {
				if (ptepa & PG_PS)
					panic("pmap_enter_quick: unexpected mapping into 4MB page");
				mpte = PHYS_TO_VM_PAGE(ptepa);
				mpte->hold_count++;
			} else {
				mpte = _pmap_allocpte(pmap, ptepindex);
				if (mpte == NULL)
					goto retry;
			}
		}
	} else {
		mpte = NULL;
	}

	/*
	 * This call to vtopte makes the assumption that we are
	 * entering the page into the current pmap.  In order to support
	 * quick entry into any pmap, one would likely use pmap_pte_quick.
	 * But that isn't as quick as vtopte.
	 */
	pte = vtopte(va);
	if (*pte) {
		if (mpte != NULL) {
			vm_page_lock_queues();
			pmap_unwire_pte_hold(pmap, mpte);
			vm_page_unlock_queues();
		}
		return 0;
	}

	/*
	 * Enter on the PV list if part of our managed memory. Note that we
	 * raise IPL while manipulating pv_table since pmap_enter can be
	 * called at interrupt time.
	 */
	if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
		pmap_insert_entry(pmap, va, mpte, m);

	/*
	 * Increment counters
	 */
	pmap->pm_stats.resident_count++;

	pa = VM_PAGE_TO_PHYS(m);

	/*
	 * Now validate mapping with RO protection
	 */
	if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
		pte_store(pte, pa | PG_V | PG_U);
	else
		pte_store(pte, pa | PG_V | PG_U | PG_MANAGED);

	return mpte;
}

/*
 * Make a temporary mapping for a physical address.  This is only intended
 * to be used for panic dumps.
 */
void *
pmap_kenter_temporary(vm_paddr_t pa, int i)
{
	vm_offset_t va;

	va = (vm_offset_t)crashdumpmap + (i * PAGE_SIZE);
	pmap_kenter(va, pa);
#ifndef I386_CPU
	invlpg(va);
#else
	invltlb();
#endif
	return ((void *)crashdumpmap);
}

/*
 * This code maps large physical mmap regions into the
 * processor address space.  Note that some shortcuts
 * are taken, but the code works.
 */
void
pmap_object_init_pt(pmap_t pmap, vm_offset_t addr,
		    vm_object_t object, vm_pindex_t pindex,
		    vm_size_t size)
{
	vm_page_t p;

	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
	KASSERT(object->type == OBJT_DEVICE,
	    ("pmap_object_init_pt: non-device object"));
	if (pseflag &&
	    ((addr & (NBPDR - 1)) == 0) && ((size & (NBPDR - 1)) == 0)) {
		int i;
		vm_page_t m[1];
		unsigned int ptepindex;
		int npdes;
		pd_entry_t ptepa;

		if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
			return;
retry:
		p = vm_page_lookup(object, pindex);
		if (p != NULL) {
			vm_page_lock_queues();
			if (vm_page_sleep_if_busy(p, FALSE, "init4p"))
				goto retry;
		} else {
			p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
			if (p == NULL)
				return;
			m[0] = p;

			if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
				vm_page_lock_queues();
				vm_page_free(p);
				vm_page_unlock_queues();
				return;
			}

			p = vm_page_lookup(object, pindex);
			vm_page_lock_queues();
			vm_page_wakeup(p);
		}
		vm_page_unlock_queues();

		ptepa = VM_PAGE_TO_PHYS(p);
		if (ptepa & (NBPDR - 1))
			return;

		p->valid = VM_PAGE_BITS_ALL;

		pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
		npdes = size >> PDRSHIFT;
		for(i = 0; i < npdes; i++) {
			pde_store(&pmap->pm_pdir[ptepindex],
			    ptepa | PG_U | PG_RW | PG_V | PG_PS);
			ptepa += NBPDR;
			ptepindex += 1;
		}
		pmap_invalidate_all(pmap);
	}
}

/*
 *	Routine:	pmap_change_wiring
 *	Function:	Change the wiring attribute for a map/virtual-address
 *			pair.
 *	In/out conditions:
 *			The mapping must already exist in the pmap.
 */
void
pmap_change_wiring(pmap, va, wired)
	register pmap_t pmap;
	vm_offset_t va;
	boolean_t wired;
{
	register pt_entry_t *pte;

	if (pmap == NULL)
		return;

	PMAP_LOCK(pmap);
	pte = pmap_pte(pmap, va);

	if (wired && !pmap_pte_w(pte))
		pmap->pm_stats.wired_count++;
	else if (!wired && pmap_pte_w(pte))
		pmap->pm_stats.wired_count--;

	/*
	 * Wiring is not a hardware characteristic so there is no need to
	 * invalidate TLB.
	 */
	pmap_pte_set_w(pte, wired);
	PMAP_UNLOCK(pmap);
}


/*
 *	Copy the range specified by src_addr/len
 *	from the source map to the range dst_addr/len
 *	in the destination map.
 *
 *	This routine is only advisory and need not do anything.
 */

void
pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr, vm_size_t len,
	  vm_offset_t src_addr)
{
	vm_offset_t addr;
	vm_offset_t end_addr = src_addr + len;
	vm_offset_t pdnxt;
	vm_page_t m;

	if (dst_addr != src_addr)
		return;

	if (!pmap_is_current(src_pmap))
		return;

	for (addr = src_addr; addr < end_addr; addr = pdnxt) {
		pt_entry_t *src_pte, *dst_pte;
		vm_page_t dstmpte, srcmpte;
		pd_entry_t srcptepaddr;
		unsigned ptepindex;

		if (addr >= UPT_MIN_ADDRESS)
			panic("pmap_copy: invalid to pmap_copy page tables\n");

		/*
		 * Don't let optional prefaulting of pages make us go
		 * way below the low water mark of free pages or way
		 * above high water mark of used pv entries.
		 */
		if (cnt.v_free_count < cnt.v_free_reserved ||
		    pv_entry_count > pv_entry_high_water)
			break;

		pdnxt = (addr + NBPDR) & ~PDRMASK;
		ptepindex = addr >> PDRSHIFT;

		srcptepaddr = src_pmap->pm_pdir[ptepindex];
		if (srcptepaddr == 0)
			continue;

		if (srcptepaddr & PG_PS) {
			if (dst_pmap->pm_pdir[ptepindex] == 0) {
				dst_pmap->pm_pdir[ptepindex] = srcptepaddr;
				dst_pmap->pm_stats.resident_count +=
				    NBPDR / PAGE_SIZE;
			}
			continue;
		}

		srcmpte = PHYS_TO_VM_PAGE(srcptepaddr);
		if (srcmpte->hold_count == 0 || (srcmpte->flags & PG_BUSY))
			continue;

		if (pdnxt > end_addr)
			pdnxt = end_addr;

		src_pte = vtopte(addr);
		while (addr < pdnxt) {
			pt_entry_t ptetemp;
			ptetemp = *src_pte;
			/*
			 * we only virtual copy managed pages
			 */
			if ((ptetemp & PG_MANAGED) != 0) {
				/*
				 * We have to check after allocpte for the
				 * pte still being around...  allocpte can
				 * block.
				 */
				dstmpte = pmap_allocpte(dst_pmap, addr);
				dst_pte = pmap_pte(dst_pmap, addr);
				if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
					/*
					 * Clear the modified and
					 * accessed (referenced) bits
					 * during the copy.
					 */
					m = PHYS_TO_VM_PAGE(ptetemp);
					*dst_pte = ptetemp & ~(PG_M | PG_A);
					dst_pmap->pm_stats.resident_count++;
					pmap_insert_entry(dst_pmap, addr,
						dstmpte, m);
	 			} else {
					vm_page_lock_queues();
					pmap_unwire_pte_hold(dst_pmap, dstmpte);
					vm_page_unlock_queues();
				}
				if (dstmpte->hold_count >= srcmpte->hold_count)
					break;
			}
			addr += PAGE_SIZE;
			src_pte++;
		}
	}
}

static __inline void
pagezero(void *page)
{
#if defined(I686_CPU)
	if (cpu_class == CPUCLASS_686) {
#if defined(CPU_ENABLE_SSE)
		if (cpu_feature & CPUID_SSE2)
			sse2_pagezero(page);
		else
#endif
			i686_pagezero(page);
	} else
#endif
		bzero(page, PAGE_SIZE);
}

/*
 *	pmap_zero_page zeros the specified hardware page by mapping
 *	the page into KVM and using bzero to clear its contents.
 */
void
pmap_zero_page(vm_page_t m)
{

	mtx_lock(&CMAPCADDR12_lock);
	if (*CMAP2)
		panic("pmap_zero_page: CMAP2 busy");
	sched_pin();
	*CMAP2 = PG_V | PG_RW | VM_PAGE_TO_PHYS(m) | PG_A | PG_M;
	invlcaddr(CADDR2);
	pagezero(CADDR2);
	*CMAP2 = 0;
	sched_unpin();
	mtx_unlock(&CMAPCADDR12_lock);
}

/*
 *	pmap_zero_page_area zeros the specified hardware page by mapping
 *	the page into KVM and using bzero to clear its contents.
 *
 *	off and size may not cover an area beyond a single hardware page.
 */
void
pmap_zero_page_area(vm_page_t m, int off, int size)
{

	mtx_lock(&CMAPCADDR12_lock);
	if (*CMAP2)
		panic("pmap_zero_page: CMAP2 busy");
	sched_pin();
	*CMAP2 = PG_V | PG_RW | VM_PAGE_TO_PHYS(m) | PG_A | PG_M;
	invlcaddr(CADDR2);
	if (off == 0 && size == PAGE_SIZE)
		pagezero(CADDR2);
	else
		bzero((char *)CADDR2 + off, size);
	*CMAP2 = 0;
	sched_unpin();
	mtx_unlock(&CMAPCADDR12_lock);
}

/*
 *	pmap_zero_page_idle zeros the specified hardware page by mapping
 *	the page into KVM and using bzero to clear its contents.  This
 *	is intended to be called from the vm_pagezero process only and
 *	outside of Giant.
 */
void
pmap_zero_page_idle(vm_page_t m)
{

	if (*CMAP3)
		panic("pmap_zero_page: CMAP3 busy");
	sched_pin();
	*CMAP3 = PG_V | PG_RW | VM_PAGE_TO_PHYS(m) | PG_A | PG_M;
	invlcaddr(CADDR3);
	pagezero(CADDR3);
	*CMAP3 = 0;
	sched_unpin();
}

/*
 *	pmap_copy_page copies the specified (machine independent)
 *	page by mapping the page into virtual memory and using
 *	bcopy to copy the page, one machine dependent page at a
 *	time.
 */
void
pmap_copy_page(vm_page_t src, vm_page_t dst)
{

	mtx_lock(&CMAPCADDR12_lock);
	if (*CMAP1)
		panic("pmap_copy_page: CMAP1 busy");
	if (*CMAP2)
		panic("pmap_copy_page: CMAP2 busy");
	sched_pin();
#ifdef I386_CPU
	invltlb();
#else
	invlpg((u_int)CADDR1);
	invlpg((u_int)CADDR2);
#endif
	*CMAP1 = PG_V | VM_PAGE_TO_PHYS(src) | PG_A;
	*CMAP2 = PG_V | PG_RW | VM_PAGE_TO_PHYS(dst) | PG_A | PG_M;
	bcopy(CADDR1, CADDR2, PAGE_SIZE);
	*CMAP1 = 0;
	*CMAP2 = 0;
	sched_unpin();
	mtx_unlock(&CMAPCADDR12_lock);
}

/*
 * Returns true if the pmap's pv is one of the first
 * 16 pvs linked to from this page.  This count may
 * be changed upwards or downwards in the future; it
 * is only necessary that true be returned for a small
 * subset of pmaps for proper page aging.
 */
boolean_t
pmap_page_exists_quick(pmap, m)
	pmap_t pmap;
	vm_page_t m;
{
	pv_entry_t pv;
	int loops = 0;

	if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
		return FALSE;

	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
	TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
		if (pv->pv_pmap == pmap) {
			return TRUE;
		}
		loops++;
		if (loops >= 16)
			break;
	}
	return (FALSE);
}

#define PMAP_REMOVE_PAGES_CURPROC_ONLY
/*
 * Remove all pages from specified address space
 * this aids process exit speeds.  Also, this code
 * is special cased for current process only, but
 * can have the more generic (and slightly slower)
 * mode enabled.  This is much faster than pmap_remove
 * in the case of running down an entire address space.
 */
void
pmap_remove_pages(pmap, sva, eva)
	pmap_t pmap;
	vm_offset_t sva, eva;
{
	pt_entry_t *pte, tpte;
	vm_page_t m;
	pv_entry_t pv, npv;

#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
	if (!curthread || (pmap != vmspace_pmap(curthread->td_proc->p_vmspace))) {
		printf("warning: pmap_remove_pages called with non-current pmap\n");
		return;
	}
#endif
	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
	PMAP_LOCK(pmap);
	sched_pin();
	for (pv = TAILQ_FIRST(&pmap->pm_pvlist); pv; pv = npv) {

		if (pv->pv_va >= eva || pv->pv_va < sva) {
			npv = TAILQ_NEXT(pv, pv_plist);
			continue;
		}

#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
		pte = vtopte(pv->pv_va);
#else
		pte = pmap_pte_quick(pmap, pv->pv_va);
#endif
		tpte = *pte;

		if (tpte == 0) {
			printf("TPTE at %p  IS ZERO @ VA %08x\n",
							pte, pv->pv_va);
			panic("bad pte");
		}

/*
 * We cannot remove wired pages from a process' mapping at this time
 */
		if (tpte & PG_W) {
			npv = TAILQ_NEXT(pv, pv_plist);
			continue;
		}

		m = PHYS_TO_VM_PAGE(tpte);
		KASSERT(m->phys_addr == (tpte & PG_FRAME),
		    ("vm_page_t %p phys_addr mismatch %016jx %016jx",
		    m, (uintmax_t)m->phys_addr, (uintmax_t)tpte));

		KASSERT(m < &vm_page_array[vm_page_array_size],
			("pmap_remove_pages: bad tpte %#jx", (uintmax_t)tpte));

		pmap->pm_stats.resident_count--;

		pte_clear(pte);

		/*
		 * Update the vm_page_t clean and reference bits.
		 */
		if (tpte & PG_M) {
			vm_page_dirty(m);
		}

		npv = TAILQ_NEXT(pv, pv_plist);
		TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);

		m->md.pv_list_count--;
		TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
		if (TAILQ_EMPTY(&m->md.pv_list))
			vm_page_flag_clear(m, PG_WRITEABLE);

		pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem);
		free_pv_entry(pv);
	}
	sched_unpin();
	pmap_invalidate_all(pmap);
	PMAP_UNLOCK(pmap);
}

/*
 *	pmap_is_modified:
 *
 *	Return whether or not the specified physical page was modified
 *	in any physical maps.
 */
boolean_t
pmap_is_modified(vm_page_t m)
{
	pv_entry_t pv;
	pt_entry_t *pte;
	boolean_t rv;

	rv = FALSE;
	if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
		return (rv);

	sched_pin();
	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
	TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
		/*
		 * if the bit being tested is the modified bit, then
		 * mark clean_map and ptes as never
		 * modified.
		 */
		if (!pmap_track_modified(pv->pv_va))
			continue;
#if defined(PMAP_DIAGNOSTIC)
		if (!pv->pv_pmap) {
			printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
			continue;
		}
#endif
		PMAP_LOCK(pv->pv_pmap);
		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
		rv = (*pte & PG_M) != 0;
		PMAP_UNLOCK(pv->pv_pmap);
		if (rv)
			break;
	}
	sched_unpin();
	return (rv);
}

/*
 *	pmap_is_prefaultable:
 *
 *	Return whether or not the specified virtual address is elgible
 *	for prefault.
 */
boolean_t
pmap_is_prefaultable(pmap_t pmap, vm_offset_t addr)
{
	pt_entry_t *pte;
	boolean_t rv;

	rv = FALSE;
	PMAP_LOCK(pmap);
	if (*pmap_pde(pmap, addr)) {
		pte = vtopte(addr);
		rv = *pte == 0;
	}
	PMAP_UNLOCK(pmap);
	return (rv);
}

/*
 *	Clear the given bit in each of the given page's ptes.
 */
static __inline void
pmap_clear_ptes(vm_page_t m, int bit)
{
	register pv_entry_t pv;
	pt_entry_t pbits, *pte;

	if (!pmap_initialized || (m->flags & PG_FICTITIOUS) ||
	    (bit == PG_RW && (m->flags & PG_WRITEABLE) == 0))
		return;

	sched_pin();
	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
	/*
	 * Loop over all current mappings setting/clearing as appropos If
	 * setting RO do we need to clear the VAC?
	 */
	TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
		/*
		 * don't write protect pager mappings
		 */
		if (bit == PG_RW) {
			if (!pmap_track_modified(pv->pv_va))
				continue;
		}

#if defined(PMAP_DIAGNOSTIC)
		if (!pv->pv_pmap) {
			printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
			continue;
		}
#endif

		PMAP_LOCK(pv->pv_pmap);
		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
		pbits = *pte;
		if (pbits & bit) {
			if (bit == PG_RW) {
				if (pbits & PG_M) {
					vm_page_dirty(m);
				}
				pte_store(pte, pbits & ~(PG_M|PG_RW));
			} else {
				pte_store(pte, pbits & ~bit);
			}
			pmap_invalidate_page(pv->pv_pmap, pv->pv_va);
		}
		PMAP_UNLOCK(pv->pv_pmap);
	}
	if (bit == PG_RW)
		vm_page_flag_clear(m, PG_WRITEABLE);
	sched_unpin();
}

/*
 *      pmap_page_protect:
 *
 *      Lower the permission for all mappings to a given page.
 */
void
pmap_page_protect(vm_page_t m, vm_prot_t prot)
{
	if ((prot & VM_PROT_WRITE) == 0) {
		if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
			pmap_clear_ptes(m, PG_RW);
		} else {
			pmap_remove_all(m);
		}
	}
}

/*
 *	pmap_ts_referenced:
 *
 *	Return a count of reference bits for a page, clearing those bits.
 *	It is not necessary for every reference bit to be cleared, but it
 *	is necessary that 0 only be returned when there are truly no
 *	reference bits set.
 *
 *	XXX: The exact number of bits to check and clear is a matter that
 *	should be tested and standardized at some point in the future for
 *	optimal aging of shared pages.
 */
int
pmap_ts_referenced(vm_page_t m)
{
	register pv_entry_t pv, pvf, pvn;
	pt_entry_t *pte;
	pt_entry_t v;
	int rtval = 0;

	if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
		return (rtval);

	sched_pin();
	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
	if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {

		pvf = pv;

		do {
			pvn = TAILQ_NEXT(pv, pv_list);

			TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);

			TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);

			if (!pmap_track_modified(pv->pv_va))
				continue;

			PMAP_LOCK(pv->pv_pmap);
			pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);

			if (pte && ((v = pte_load(pte)) & PG_A) != 0) {
				atomic_clear_int((u_int *)pte, PG_A);
				pmap_invalidate_page(pv->pv_pmap, pv->pv_va);

				rtval++;
				if (rtval > 4) {
					PMAP_UNLOCK(pv->pv_pmap);
					break;
				}
			}
			PMAP_UNLOCK(pv->pv_pmap);
		} while ((pv = pvn) != NULL && pv != pvf);
	}
	sched_unpin();

	return (rtval);
}

/*
 *	Clear the modify bits on the specified physical page.
 */
void
pmap_clear_modify(vm_page_t m)
{
	pmap_clear_ptes(m, PG_M);
}

/*
 *	pmap_clear_reference:
 *
 *	Clear the reference bit on the specified physical page.
 */
void
pmap_clear_reference(vm_page_t m)
{
	pmap_clear_ptes(m, PG_A);
}

/*
 * Miscellaneous support routines follow
 */

/*
 * Map a set of physical memory pages into the kernel virtual
 * address space. Return a pointer to where it is mapped. This
 * routine is intended to be used for mapping device memory,
 * NOT real memory.
 */
void *
pmap_mapdev(pa, size)
	vm_paddr_t pa;
	vm_size_t size;
{
	vm_offset_t va, tmpva, offset;

	offset = pa & PAGE_MASK;
	size = roundup(offset + size, PAGE_SIZE);
	pa = pa & PG_FRAME;

	if (pa < KERNLOAD && pa + size <= KERNLOAD)
		va = KERNBASE + pa;
	else
		va = kmem_alloc_nofault(kernel_map, size);
	if (!va)
		panic("pmap_mapdev: Couldn't alloc kernel virtual memory");

	for (tmpva = va; size > 0; ) {
		pmap_kenter(tmpva, pa);
		size -= PAGE_SIZE;
		tmpva += PAGE_SIZE;
		pa += PAGE_SIZE;
	}
	pmap_invalidate_range(kernel_pmap, va, tmpva);
	return ((void *)(va + offset));
}

void
pmap_unmapdev(va, size)
	vm_offset_t va;
	vm_size_t size;
{
	vm_offset_t base, offset, tmpva;

	if (va >= KERNBASE && va + size <= KERNBASE + KERNLOAD)
		return;
	base = va & PG_FRAME;
	offset = va & PAGE_MASK;
	size = roundup(offset + size, PAGE_SIZE);
	for (tmpva = base; tmpva < (base + size); tmpva += PAGE_SIZE)
		pmap_kremove(tmpva);
	pmap_invalidate_range(kernel_pmap, va, tmpva);
	kmem_free(kernel_map, base, size);
}

/*
 * perform the pmap work for mincore
 */
int
pmap_mincore(pmap, addr)
	pmap_t pmap;
	vm_offset_t addr;
{
	pt_entry_t *ptep, pte;
	vm_page_t m;
	int val = 0;

	PMAP_LOCK(pmap);
	ptep = pmap_pte(pmap, addr);
	pte = (ptep != NULL) ? *ptep : 0;
	PMAP_UNLOCK(pmap);

	if (pte != 0) {
		vm_paddr_t pa;

		val = MINCORE_INCORE;
		if ((pte & PG_MANAGED) == 0)
			return val;

		pa = pte & PG_FRAME;

		m = PHYS_TO_VM_PAGE(pa);

		/*
		 * Modified by us
		 */
		if (pte & PG_M)
			val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
		else {
			/*
			 * Modified by someone else
			 */
			vm_page_lock_queues();
			if (m->dirty || pmap_is_modified(m))
				val |= MINCORE_MODIFIED_OTHER;
			vm_page_unlock_queues();
		}
		/*
		 * Referenced by us
		 */
		if (pte & PG_A)
			val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
		else {
			/*
			 * Referenced by someone else
			 */
			vm_page_lock_queues();
			if ((m->flags & PG_REFERENCED) ||
			    pmap_ts_referenced(m)) {
				val |= MINCORE_REFERENCED_OTHER;
				vm_page_flag_set(m, PG_REFERENCED);
			}
			vm_page_unlock_queues();
		}
	}
	return val;
}

void
pmap_activate(struct thread *td)
{
	struct proc *p = td->td_proc;
	pmap_t	pmap, oldpmap;
	u_int32_t  cr3;

	critical_enter();
	pmap = vmspace_pmap(td->td_proc->p_vmspace);
	oldpmap = PCPU_GET(curpmap);
#if defined(SMP)
	atomic_clear_int(&oldpmap->pm_active, PCPU_GET(cpumask));
	atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask));
#else
	oldpmap->pm_active &= ~1;
	pmap->pm_active |= 1;
#endif
#ifdef PAE
	cr3 = vtophys(pmap->pm_pdpt);
#else
	cr3 = vtophys(pmap->pm_pdir);
#endif
	/* XXXKSE this is wrong.
	 * pmap_activate is for the current thread on the current cpu
	 */
	if (p->p_flag & P_SA) {
		/* Make sure all other cr3 entries are updated. */
		/* what if they are running?  XXXKSE (maybe abort them) */
		FOREACH_THREAD_IN_PROC(p, td) {
			td->td_pcb->pcb_cr3 = cr3;
		}
	} else {
		td->td_pcb->pcb_cr3 = cr3;
	}
	load_cr3(cr3);
	PCPU_SET(curpmap, pmap);
	critical_exit();
}

vm_offset_t
pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
{

	if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
		return addr;
	}

	addr = (addr + PDRMASK) & ~PDRMASK;
	return addr;
}


#if defined(PMAP_DEBUG)
pmap_pid_dump(int pid)
{
	pmap_t pmap;
	struct proc *p;
	int npte = 0;
	int index;

	sx_slock(&allproc_lock);
	LIST_FOREACH(p, &allproc, p_list) {
		if (p->p_pid != pid)
			continue;

		if (p->p_vmspace) {
			int i,j;
			index = 0;
			pmap = vmspace_pmap(p->p_vmspace);
			for (i = 0; i < NPDEPTD; i++) {
				pd_entry_t *pde;
				pt_entry_t *pte;
				vm_offset_t base = i << PDRSHIFT;

				pde = &pmap->pm_pdir[i];
				if (pde && pmap_pde_v(pde)) {
					for (j = 0; j < NPTEPG; j++) {
						vm_offset_t va = base + (j << PAGE_SHIFT);
						if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
							if (index) {
								index = 0;
								printf("\n");
							}
							sx_sunlock(&allproc_lock);
							return npte;
						}
						pte = pmap_pte(pmap, va);
						if (pte && pmap_pte_v(pte)) {
							pt_entry_t pa;
							vm_page_t m;
							pa = *pte;
							m = PHYS_TO_VM_PAGE(pa);
							printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
								va, pa, m->hold_count, m->wire_count, m->flags);
							npte++;
							index++;
							if (index >= 2) {
								index = 0;
								printf("\n");
							} else {
								printf(" ");
							}
						}
					}
				}
			}
		}
	}
	sx_sunlock(&allproc_lock);
	return npte;
}
#endif

#if defined(DEBUG)

static void	pads(pmap_t pm);
void		pmap_pvdump(vm_offset_t pa);

/* print address space of pmap*/
static void
pads(pm)
	pmap_t pm;
{
	int i, j;
	vm_paddr_t va;
	pt_entry_t *ptep;

	if (pm == kernel_pmap)
		return;
	for (i = 0; i < NPDEPTD; i++)
		if (pm->pm_pdir[i])
			for (j = 0; j < NPTEPG; j++) {
				va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
				if (pm == kernel_pmap && va < KERNBASE)
					continue;
				if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
					continue;
				ptep = pmap_pte(pm, va);
				if (pmap_pte_v(ptep))
					printf("%x:%x ", va, *ptep);
			};

}

void
pmap_pvdump(pa)
	vm_paddr_t pa;
{
	pv_entry_t pv;
	vm_page_t m;

	printf("pa %x", pa);
	m = PHYS_TO_VM_PAGE(pa);
	TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
		printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);
		pads(pv->pv_pmap);
	}
	printf(" ");
}
#endif