This commit was generated by cvs2svn to compensate for changes in r57419,

[FreeBSD/FreeBSD.git] / sys / alpha / alpha / 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.
 * Copyright (c) 1998 Doug Rabson
 * 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
 *      from:   i386 Id: pmap.c,v 1.193 1998/04/19 15:22:48 bde Exp
 *              with some ideas from NetBSD's alpha pmap
 * $FreeBSD$
 */

/*
 *      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.
 */

/*
 * Notes for alpha pmap.
 * 
 * On alpha, pm_pdeobj will hold lev1, lev2 and lev3 page tables.
 * Indices from 0 to NUSERLEV3MAPS-1 will map user lev3 page tables,
 * indices from NUSERLEV3MAPS to NUSERLEV3MAPS+NUSERLEV2MAPS-1 will
 * map user lev2 page tables and index NUSERLEV3MAPS+NUSERLEV2MAPS
 * will map the lev1 page table.  The lev1 table will self map at
 * address VADDR(PTLEV1I,0,0).
 * 
 * The vm_object kptobj holds the kernel page tables on i386 (62 or 63
 * of them, depending on whether the system is SMP).  On alpha, kptobj
 * will hold the lev3 and lev2 page tables for K1SEG.  Indices 0 to
 * NKLEV3MAPS-1 will map kernel lev3 page tables and indices
 * NKLEV3MAPS to NKLEV3MAPS+NKLEV2MAPS will map lev2 page tables. (XXX
 * should the kernel Lev1map be inserted into this object?).
 * 
 * pvtmmap is not needed for alpha since K0SEG maps all of physical
 * memory. CADDR1 and CADDR2 are not needed for the same reason.  The
 * only places outside pmap and machdep which use CADDR1 are xxdump
 * routines which use them for dumping physical pages.
 * 
 * 
 * alpha virtual memory map:
 * 
 * 
 *  Address                                                     Lev1 index
 * 
 *                      ---------------------------------
 *  0000000000000000    |                               |       0
 *                      |                               |
 *                      |                               |
 *                      |                               |
 *                      |                               |
 *                     ---                             ---
 *                              User space (USEG)
 *                     ---                             ---
 *                      |                               |
 *                      |                               |
 *                      |                               |
 *                      |                               |
 *  000003ffffffffff    |                               |       511=UMAXLEV1I
 *                      ---------------------------------
 *  fffffc0000000000    |                               |       512=K0SEGLEV1I
 *                      |       Kernel code/data/bss    |
 *                      |                               |
 *                      |                               |
 *                      |                               |
 *                     ---                             ---
 *                              K0SEG
 *                     ---                             ---
 *                      |                               |
 *                      |       1-1 physical/virtual    |
 *                      |                               |
 *                      |                               |
 *  fffffdffffffffff    |                               |
 *                      ---------------------------------
 *  fffffe0000000000    |                               |       768=K1SEGLEV1I
 *                      |       Kernel dynamic data     |
 *                      |                               |
 *                      |                               |
 *                      |                               |
 *                     ---                             ---
 *                              K1SEG
 *                     ---                             ---
 *                      |                               |
 *                      |       mapped by ptes          |
 *                      |                               |
 *                      |                               |
 *  fffffff7ffffffff    |                               |
 *                      ---------------------------------
 *  fffffffe00000000    |                               |       1023=PTLEV1I
 *                      |       PTmap (pte self map)    |
 *  ffffffffffffffff    |                               |
 *                      ---------------------------------
 * 
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/msgbuf.h>
#include <sys/vmmeter.h>
#include <sys/mman.h>

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <sys/lock.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/vm_zone.h>

#include <sys/user.h>

#include <machine/md_var.h>

#ifndef PMAP_SHPGPERPROC
#define PMAP_SHPGPERPROC 200
#endif

#if defined(DIAGNOSTIC)
#define PMAP_DIAGNOSTIC
#endif

#define MINPV 2048

#if 0
#define PMAP_DIAGNOSTIC
#define PMAP_DEBUG
#endif

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

#if 0

static void
pmap_break(void)
{
}

/* #define PMAP_DEBUG_VA(va) if ((va) == 0x120058000) pmap_break(); else */

#endif

#ifndef PMAP_DEBUG_VA
#define PMAP_DEBUG_VA(va) do {} while(0)
#endif

/*
 * Some macros for manipulating virtual addresses
 */
#define ALPHA_L1SIZE            (1L << ALPHA_L1SHIFT)
#define ALPHA_L2SIZE            (1L << ALPHA_L2SHIFT)

#define alpha_l1trunc(va)       ((va) & ~(ALPHA_L1SIZE-1))
#define alpha_l2trunc(va)       ((va) & ~(ALPHA_L2SIZE-1))

/*
 * Get PDEs and PTEs for user/kernel address space
 */
#define pmap_pte_w(pte)         ((*(pte) & PG_W) != 0)
#define pmap_pte_managed(pte)   ((*(pte) & PG_MANAGED) != 0)
#define pmap_pte_v(pte)         ((*(pte) & PG_V) != 0)
#define pmap_pte_pa(pte)        alpha_ptob(ALPHA_PTE_TO_PFN(*(pte)))
#define pmap_pte_prot(pte)      (*(pte) & PG_PROT)

#define pmap_pte_set_w(pte, v) ((v)?(*pte |= PG_W):(*pte &= ~PG_W))
#define pmap_pte_set_prot(pte, v) ((*pte &= ~PG_PROT), (*pte |= (v)))

/*
 * Given a map and a machine independent protection code,
 * convert to an alpha protection code.
 */
#define pte_prot(m, p)          (protection_codes[m == pmap_kernel() ? 0 : 1][p])
int     protection_codes[2][8];

#define pa_index(pa)            atop((pa) - vm_first_phys)
#define pa_to_pvh(pa)           (&pv_table[pa_index(pa)])

/*
 * Return non-zero if this pmap is currently active
 */
#define pmap_isactive(pmap)     (pmap->pm_active)

/* 
 * Extract level 1, 2 and 3 page table indices from a va
 */
#define PTMASK  ((1 << ALPHA_PTSHIFT) - 1)

#define pmap_lev1_index(va)     (((va) >> ALPHA_L1SHIFT) & PTMASK)
#define pmap_lev2_index(va)     (((va) >> ALPHA_L2SHIFT) & PTMASK)
#define pmap_lev3_index(va)     (((va) >> ALPHA_L3SHIFT) & PTMASK)

/*
 * Given a physical address, construct a pte
 */
#define pmap_phys_to_pte(pa)    ALPHA_PTE_FROM_PFN(alpha_btop(pa))

/*
 * Given a page frame number, construct a k0seg va
 */
#define pmap_k0seg_to_pfn(va)   alpha_btop(ALPHA_K0SEG_TO_PHYS(va))

/*
 * Given a pte, construct a k0seg va
 */
#define pmap_k0seg_to_pte(va)   ALPHA_PTE_FROM_PFN(pmap_k0seg_to_pfn(va))

/*
 * Lev1map:
 *
 *      Kernel level 1 page table.  This maps all kernel level 2
 *      page table pages, and is used as a template for all user
 *      pmap level 1 page tables.  When a new user level 1 page
 *      table is allocated, all Lev1map PTEs for kernel addresses
 *      are copied to the new map.
 *
 * Lev2map:
 *
 *      Initial set of kernel level 2 page table pages.  These
 *      map the kernel level 3 page table pages.  As kernel
 *      level 3 page table pages are added, more level 2 page
 *      table pages may be added to map them.  These pages are
 *      never freed.
 *
 * Lev3map:
 *
 *      Initial set of kernel level 3 page table pages.  These
 *      map pages in K1SEG.  More level 3 page table pages may
 *      be added at run-time if additional K1SEG address space
 *      is required.  These pages are never freed.
 *
 * Lev2mapsize:
 *
 *      Number of entries in the initial Lev2map.
 *
 * Lev3mapsize:
 *
 *      Number of entries in the initial Lev3map.
 *
 * NOTE: When mappings are inserted into the kernel pmap, all
 * level 2 and level 3 page table pages must already be allocated
 * and mapped into the parent page table.
 */
pt_entry_t      *Lev1map, *Lev2map, *Lev3map;
vm_size_t       Lev2mapsize, Lev3mapsize;

/*
 * Statically allocated kernel pmap
 */
static struct pmap kernel_pmap_store;
pmap_t kernel_pmap;

vm_offset_t avail_start;        /* PA of first available physical page */
vm_offset_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? */
static vm_offset_t vm_first_phys;
static int pv_npg;

static vm_object_t kptobj;

static int nklev3, nklev2;
vm_offset_t kernel_vm_end;

/*
 * Data for the ASN allocator
 */
static int pmap_maxasn;
static int pmap_nextasn = 0;
static u_int pmap_current_asngen = 1;
static pmap_t pmap_active = 0;

/*
 * Data for the pv entry allocation mechanism
 */
static vm_zone_t pvzone;
static struct vm_zone pvzone_store;
static struct vm_object pvzone_obj;
static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
static int pmap_pagedaemon_waken = 0;
static struct pv_entry *pvinit;

/*
 * All those kernel PT submaps that BSD is so fond of
 */
pt_entry_t *CMAP1 = 0;
static pt_entry_t *CMAP2;
static pv_table_t *pv_table;
caddr_t CADDR1;
static caddr_t CADDR2;

static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv));
static pv_entry_t get_pv_entry __P((void));
static void     alpha_protection_init __P((void));
static void     pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem));

static PMAP_INLINE int  pmap_is_managed __P((vm_offset_t pa));
static void     pmap_remove_all __P((vm_offset_t pa));
static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va,
                                      vm_offset_t pa, vm_page_t mpte));
static int pmap_remove_pte __P((pmap_t pmap, pt_entry_t* ptq, vm_offset_t sva));
static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va));
static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv,
                                        vm_offset_t va));
static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va,
                vm_page_t mpte, vm_offset_t pa));

static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va));

static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p));
static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex));
static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex));
static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t));


/*
 *      Routine:        pmap_lev1pte
 *      Function:
 *              Extract the level 1 page table entry associated
 *              with the given map/virtual_address pair.
 */
static PMAP_INLINE pt_entry_t*
pmap_lev1pte(pmap_t pmap, vm_offset_t va)
{
        if (!pmap)
                return 0;
        return &pmap->pm_lev1[pmap_lev1_index(va)];
}

/*
 *      Routine:        pmap_lev2pte
 *      Function:
 *              Extract the level 2 page table entry associated
 *              with the given map/virtual_address pair.
 */
static PMAP_INLINE pt_entry_t*
pmap_lev2pte(pmap_t pmap, vm_offset_t va)
{
        pt_entry_t* l1pte;
        pt_entry_t* l2map;

        l1pte = pmap_lev1pte(pmap, va);
        if (!pmap_pte_v(l1pte))
                return 0;

        l2map = (pt_entry_t*) ALPHA_PHYS_TO_K0SEG(pmap_pte_pa(l1pte));
        return &l2map[pmap_lev2_index(va)];
}

/*
 *      Routine:        pmap_lev3pte
 *      Function:
 *              Extract the level 3 page table entry associated
 *              with the given map/virtual_address pair.
 */
static PMAP_INLINE pt_entry_t*
pmap_lev3pte(pmap_t pmap, vm_offset_t va)
{
        pt_entry_t* l2pte;
        pt_entry_t* l3map;

        l2pte = pmap_lev2pte(pmap, va);
        if (!l2pte || !pmap_pte_v(l2pte))
                return 0;

        l3map = (pt_entry_t*) ALPHA_PHYS_TO_K0SEG(pmap_pte_pa(l2pte));
        return &l3map[pmap_lev3_index(va)];
}

vm_offset_t
pmap_steal_memory(vm_size_t size)
{
        vm_size_t bank_size;
        vm_offset_t pa, va;

        size = round_page(size);

        bank_size = phys_avail[1] - phys_avail[0];
        while (size > bank_size) {
                int i;
                for (i = 0; phys_avail[i+2]; i+= 2) {
                        phys_avail[i] = phys_avail[i+2];
                        phys_avail[i+1] = phys_avail[i+3];
                }
                phys_avail[i] = 0;
                phys_avail[i+1] = 0;
                if (!phys_avail[0])
                        panic("pmap_steal_memory: out of memory");
                bank_size = phys_avail[1] - phys_avail[0];
        }

        pa = phys_avail[0];
        phys_avail[0] += size;

        va = ALPHA_PHYS_TO_K0SEG(pa);
        bzero((caddr_t) va, size);
        return va;
}

extern pt_entry_t rom_pte;                      /* XXX */
extern int prom_mapped;                         /* XXX */

/*
 *      Bootstrap the system enough to run with virtual memory.
 */
void
pmap_bootstrap(vm_offset_t ptaddr, u_int maxasn)
{
        pt_entry_t newpte;
        pt_entry_t* pte;
        vm_offset_t va;
        int i;

        /*
         * Setup ASNs
         */
        pmap_nextasn = 0;
        pmap_maxasn = maxasn;
        pmap_current_asngen = 1;

        /*
         * Allocate a level 1 map for the kernel.
         */
        Lev1map = (pt_entry_t*) pmap_steal_memory(PAGE_SIZE);

        /*
         * Allocate a level 2 map for the kernel
         */
        Lev2map = (pt_entry_t*) pmap_steal_memory(PAGE_SIZE);
        Lev2mapsize = PAGE_SIZE;

        /*
         * Allocate some level 3 maps for the kernel
         */
        Lev3map = (pt_entry_t*) pmap_steal_memory(PAGE_SIZE*NKPT);
        Lev3mapsize = NKPT * PAGE_SIZE;

        /* Map all of the level 2 maps */
        for (i = 0; i < howmany(Lev2mapsize, PAGE_SIZE); i++) {
                unsigned long pfn =
                        pmap_k0seg_to_pfn((vm_offset_t) Lev2map) + i;
                newpte = ALPHA_PTE_FROM_PFN(pfn);
                newpte |= PG_V | PG_ASM | PG_KRE | PG_KWE | PG_W;
                Lev1map[K1SEGLEV1I + i] = newpte;
        }


        /* Setup the mapping for the prom console */
        {

                if (pmap_uses_prom_console()) {
                        /* XXX save old pte so that we can remap prom if necessary */
                        rom_pte = *(pt_entry_t *)ptaddr & ~PG_ASM;      /* XXX */
                }
                prom_mapped = 0;

                /*
                 * Actually, this code lies.  The prom is still mapped, and will
                 * remain so until the context switch after alpha_init() returns.
                 * Printfs using the firmware before then will end up frobbing
                 * Lev1map unnecessarily, but that's OK.
                 */
        }

        /*
         * Level 1 self mapping.
         *
         * Don't set PG_ASM since the self-mapping is different for each
         * address space.
         */
        newpte = pmap_k0seg_to_pte((vm_offset_t) Lev1map);
        newpte |= PG_V | PG_KRE | PG_KWE;
        Lev1map[PTLEV1I] = newpte;

        /* Map all of the level 3 maps */
        for (i = 0; i < howmany(Lev3mapsize, PAGE_SIZE); i++) {
                unsigned long pfn =
                        pmap_k0seg_to_pfn((vm_offset_t) Lev3map) + i;
                newpte = ALPHA_PTE_FROM_PFN(pfn);
                newpte |= PG_V | PG_ASM | PG_KRE | PG_KWE | PG_W;
                Lev2map[i] = newpte;
        }

        avail_start = phys_avail[0];
        for (i = 0; phys_avail[i+2]; i+= 2) ;
        avail_end = phys_avail[i+1];

        virtual_avail = VM_MIN_KERNEL_ADDRESS;
        virtual_end = VPTBASE;

        /*
         * Initialize protection array.
         */
        alpha_protection_init();

        /*
         * The kernel's pmap is statically allocated so we don't have to use
         * pmap_create, which is unlikely to work correctly at this part of
         * the boot sequence (XXX and which no longer exists).
         */
        kernel_pmap = &kernel_pmap_store;
        kernel_pmap->pm_lev1 = Lev1map;
        kernel_pmap->pm_count = 1;
        kernel_pmap->pm_active = 1;
        kernel_pmap->pm_asn = 0;
        kernel_pmap->pm_asngen = pmap_current_asngen;
        pmap_nextasn = 1;
        TAILQ_INIT(&kernel_pmap->pm_pvlist);
        nklev3 = NKPT;
        nklev2 = 1;

        /*
         * 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 = pmap_lev3pte(kernel_pmap, va);

        /*
         * CMAP1/CMAP2 are used for zeroing and copying pages.
         */
        SYSMAP(caddr_t, CMAP1, CADDR1, 1)
        SYSMAP(caddr_t, CMAP2, CADDR2, 1)

        virtual_avail = va;

        *CMAP1 = *CMAP2 = 0;

        /*
         * Set up proc0's PCB such that the ptbr points to the right place
         * and has the kernel pmap's.
         */
        proc0.p_addr->u_pcb.pcb_hw.apcb_ptbr =
            ALPHA_K0SEG_TO_PHYS((vm_offset_t)Lev1map) >> PAGE_SHIFT;
        proc0.p_addr->u_pcb.pcb_hw.apcb_asn = 0;
}

int
pmap_uses_prom_console()
{
#if 0
        extern int cputype;

#if defined(NEW_SCC_DRIVER)
        return (cputype == ST_DEC_21000);
#else
        return (cputype == ST_DEC_21000
             || cputype == ST_DEC_3000_300
             || cputype == ST_DEC_3000_500);
#endif /* NEW_SCC_DRIVER */
#endif

        return 1;
}

/*
 *      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(phys_start, phys_end)
        vm_offset_t phys_start, phys_end;
{
        vm_offset_t addr;
        vm_size_t s;
        int i;
        int initial_pvs;

        /*
         * calculate the number of pv_entries needed
         */
        vm_first_phys = phys_avail[0];
        for (i = 0; phys_avail[i + 1]; i += 2);
        pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE;

        /*
         * Allocate memory for random pmap data structures.  Includes the
         * pv_head_table.
         */
        s = (vm_size_t) (sizeof(pv_table_t) * pv_npg);
        s = round_page(s);

        addr = (vm_offset_t) kmem_alloc(kernel_map, s);
        pv_table = (pv_table_t *) addr;
        for(i = 0; i < pv_npg; i++) {
                vm_offset_t pa;
                TAILQ_INIT(&pv_table[i].pv_list);
                pv_table[i].pv_list_count = 0;
                pa = vm_first_phys + i * PAGE_SIZE;
                pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa);
        }

        /*
         * init the pv free list
         */
        initial_pvs = pv_npg;
        if (initial_pvs < MINPV)
                initial_pvs = MINPV;
        pvzone = &pvzone_store;
        pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
                initial_pvs * sizeof (struct pv_entry));
        zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, pv_npg);
        /*
         * object for kernel page table pages
         */
        kptobj = vm_object_allocate(OBJT_DEFAULT, NKLEV3MAPS + NKLEV2MAPS);

        /*
         * 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()
{
        pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg;
        pv_entry_high_water = 9 * (pv_entry_max / 10);
        zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
}

/*
 *      Used to map a range of physical addresses into kernel
 *      virtual address space.
 *
 *      For now, VM is already on, we only need to map the
 *      specified memory.
 */
vm_offset_t
pmap_map(vm_offset_t virt, vm_offset_t start, vm_offset_t end, int prot)
{
        while (start < end) {
                pmap_enter(kernel_pmap, virt, start, prot, FALSE);
                virt += PAGE_SIZE;
                start += PAGE_SIZE;
        }
        return (virt);
}


/***************************************************
 * Manipulate TLBs for a pmap
 ***************************************************/

static void
pmap_invalidate_asn(pmap_t pmap)
{
        pmap->pm_asngen = 0;
}

static void
pmap_invalidate_page(pmap_t pmap, vm_offset_t va)
{
        if (pmap_isactive(pmap)) {
                ALPHA_TBIS(va);
                alpha_pal_imb();                /* XXX overkill? */
        } else
                pmap_invalidate_asn(pmap);
}

static void
pmap_invalidate_all(pmap_t pmap)
{
        if (pmap_isactive(pmap)) {
                ALPHA_TBIA();
                alpha_pal_imb();                /* XXX overkill? */
        } else
                pmap_invalidate_asn(pmap);
}

static void
pmap_get_asn(pmap_t pmap)
{
        if (pmap->pm_asngen != pmap_current_asngen) {
                if (pmap_nextasn > pmap_maxasn) {
                        /*
                         * Start a new ASN generation.
                         *
                         * Invalidate all per-process mappings and I-cache
                         */
                        pmap_nextasn = 0;
                        pmap_current_asngen++;

                        if (pmap_current_asngen == 0) {
                                /*
                                 * Clear the pm_asngen of all pmaps.
                                 * This is safe since it is only called from
                                 * pmap_activate after it has deactivated
                                 * the old pmap.
                                 */
                                struct proc *p;
                                pmap_t tpmap;
                               
#ifdef PMAP_DIAGNOSTIC
                                printf("pmap_get_asn: generation rollover\n");
#endif
                                pmap_current_asngen = 1;
                                LIST_FOREACH(p, &allproc, p_list) {
                                        if (p->p_vmspace) {
                                                tpmap = vmspace_pmap(p->p_vmspace);
                                                tpmap->pm_asngen = 0;
                                        }
                                }
                        }

                        /*
                         * Since we are about to start re-using ASNs, we must
                         * clear out the TLB and the I-cache since they are tagged
                         * with the ASN.
                         */
                        ALPHA_TBIAP();
                        alpha_pal_imb();        /* XXX overkill? */
                }
                pmap->pm_asn = pmap_nextasn++;
                pmap->pm_asngen = pmap_current_asngen;
        }
}

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



/*
 * 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 < clean_sva) || (va >= clean_eva)) 
                return 1;
        else
                return 0;
}

/*
 *      Routine:        pmap_extract
 *      Function:
 *              Extract the physical page address associated
 *              with the given map/virtual_address pair.
 */
vm_offset_t 
pmap_extract(pmap, va)
        register pmap_t pmap;
        vm_offset_t va;
{
        pt_entry_t* pte = pmap_lev3pte(pmap, va);
        if (pte)
                return alpha_ptob(ALPHA_PTE_TO_PFN(*pte));
        else
                return 0;
}

/*
 * determine if a page is managed (memory vs. device)
 */
static PMAP_INLINE int
pmap_is_managed(pa)
        vm_offset_t pa;
{
        int i;

        if (!pmap_initialized)
                return 0;

        for (i = 0; phys_avail[i + 1]; i += 2) {
                if (pa < phys_avail[i + 1] && pa >= phys_avail[i])
                        return 1;
        }
        return 0;
}


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

/*
 * 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.
 */
void
pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
{
        int i;
        pt_entry_t *pte;

        for (i = 0; i < count; i++) {
                vm_offset_t tva = va + i * PAGE_SIZE;
                pt_entry_t npte = pmap_phys_to_pte(VM_PAGE_TO_PHYS(m[i]))
                        | PG_ASM | PG_KRE | PG_KWE | PG_V;
                pt_entry_t opte;
                pte = vtopte(tva);
                opte = *pte;
                PMAP_DEBUG_VA(va);
                *pte = npte;
                if (opte)
                        pmap_invalidate_page(kernel_pmap, tva);
        }
}

/*
 * this routine jerks page mappings from the
 * kernel -- it is meant only for temporary mappings.
 */
void
pmap_qremove(va, count)
        vm_offset_t va;
        int count;
{
        int i;
        register pt_entry_t *pte;

        for (i = 0; i < count; i++) {
                pte = vtopte(va);
                PMAP_DEBUG_VA(va);
                *pte = 0;
                pmap_invalidate_page(kernel_pmap, va);
                va += PAGE_SIZE;
        }
}

/*
 * add a wired page to the kva
 * note that in order for the mapping to take effect -- you
 * should do a invltlb after doing the pmap_kenter...
 */
PMAP_INLINE void 
pmap_kenter(vm_offset_t va, vm_offset_t pa)
{
        pt_entry_t *pte;
        pt_entry_t npte, opte;

        npte = pmap_phys_to_pte(pa) | PG_ASM | PG_KRE | PG_KWE | PG_V;
        pte = vtopte(va);
        opte = *pte;
        PMAP_DEBUG_VA(va);
        *pte = npte;
        if (opte)
                pmap_invalidate_page(kernel_pmap, va);
}

/*
 * remove a page from the kernel pagetables
 */
PMAP_INLINE void
pmap_kremove(vm_offset_t va)
{
        register pt_entry_t *pte;

        pte = vtopte(va);
        PMAP_DEBUG_VA(va);
        *pte = 0;
        pmap_invalidate_page(kernel_pmap, va);
}

static vm_page_t
pmap_page_lookup(vm_object_t object, vm_pindex_t pindex)
{
        vm_page_t m;
retry:
        m = vm_page_lookup(object, pindex);
        if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
                goto retry;
        return m;
}

/*
 * Create the UPAGES for a new process.
 * This routine directly affects the fork perf for a process.
 */
void
pmap_new_proc(struct proc *p)
{
        int i;
        vm_object_t upobj;
        vm_page_t m;
        struct user *up;
        pt_entry_t *ptek, oldpte;

        /*
         * allocate object for the upages
         */
        if ((upobj = p->p_upages_obj) == NULL) {
                upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES);
                p->p_upages_obj = upobj;
        }

        /* get a kernel virtual address for the UPAGES for this proc */
        if ((up = p->p_addr) == NULL) {
                up = (struct user *) kmem_alloc_nofault(kernel_map,
                                UPAGES * PAGE_SIZE);
#if !defined(MAX_PERF)
                if (up == NULL)
                        panic("pmap_new_proc: u_map allocation failed");
#endif
                p->p_addr = up;
        }

        ptek = vtopte((vm_offset_t) up);

        for(i=0;i<UPAGES;i++) {
                /*
                 * Get a kernel stack page
                 */
                m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);

                /*
                 * Wire the page
                 */
                m->wire_count++;
                cnt.v_wire_count++;

                oldpte = *(ptek + i);
                /*
                 * Enter the page into the kernel address space.
                 */
                *(ptek + i) = pmap_phys_to_pte(VM_PAGE_TO_PHYS(m))
                        | PG_ASM | PG_KRE | PG_KWE | PG_V;
                if (oldpte) 
                        pmap_invalidate_page(kernel_pmap,
                                             (vm_offset_t)up + i * PAGE_SIZE);

                vm_page_wakeup(m);
                vm_page_flag_clear(m, PG_ZERO);
                vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
                m->valid = VM_PAGE_BITS_ALL;
        }
}

/*
 * Dispose the UPAGES for a process that has exited.
 * This routine directly impacts the exit perf of a process.
 */
void
pmap_dispose_proc(p)
        struct proc *p;
{
        int i;
        vm_object_t upobj;
        vm_page_t m;
        pt_entry_t *ptek, oldpte;

        upobj = p->p_upages_obj;

        ptek = vtopte((vm_offset_t) p->p_addr);
        for(i=0;i<UPAGES;i++) {

                if ((m = vm_page_lookup(upobj, i)) == NULL)
                        panic("pmap_dispose_proc: upage already missing???");

                vm_page_busy(m);

                oldpte = *(ptek + i);
                *(ptek + i) = 0;
                pmap_invalidate_page(kernel_pmap, 
                                     (vm_offset_t)p->p_addr + i * PAGE_SIZE);
                vm_page_unwire(m, 0);
                vm_page_free(m);
        }
}

/*
 * Allow the UPAGES for a process to be prejudicially paged out.
 */
void
pmap_swapout_proc(p)
        struct proc *p;
{
        int i;
        vm_object_t upobj;
        vm_page_t m;

        /*
         * Make sure we aren't fpcurproc.
         */
        alpha_fpstate_save(p, 1);

        upobj = p->p_upages_obj;
        /*
         * let the upages be paged
         */
        for(i=0;i<UPAGES;i++) {
                if ((m = vm_page_lookup(upobj, i)) == NULL)
                        panic("pmap_swapout_proc: upage already missing???");
                vm_page_dirty(m);
                vm_page_unwire(m, 0);
                pmap_kremove((vm_offset_t)p->p_addr + PAGE_SIZE * i);
        }
}

/*
 * Bring the UPAGES for a specified process back in.
 */
void
pmap_swapin_proc(p)
        struct proc *p;
{
        int i,rv;
        vm_object_t upobj;
        vm_page_t m;

        upobj = p->p_upages_obj;
        for(i=0;i<UPAGES;i++) {

                m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);

                pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
                        VM_PAGE_TO_PHYS(m));

                if (m->valid != VM_PAGE_BITS_ALL) {
                        rv = vm_pager_get_pages(upobj, &m, 1, 0);
#if !defined(MAX_PERF)
                        if (rv != VM_PAGER_OK)
                                panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
#endif
                        m = vm_page_lookup(upobj, i);
                        m->valid = VM_PAGE_BITS_ALL;
                }

                vm_page_wire(m);
                vm_page_wakeup(m);
                vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
        }

        /*
         * The pcb may be at a different physical address now so cache the
         * new address.
         */
        p->p_md.md_pcbpaddr = (void*) vtophys((vm_offset_t) &p->p_addr->u_pcb);
}

/***************************************************
 * 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_offset_t va, vm_page_t m)
{

        while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
                ;

        if (m->hold_count == 0) {
                vm_offset_t pteva;
                pt_entry_t* pte;

                /*
                 * unmap the page table page
                 */
                if (m->pindex >= NUSERLEV3MAPS) {
                        /* Level 2 page table */
                        pte = pmap_lev1pte(pmap, va);
                        pteva = (vm_offset_t) PTlev2 + alpha_ptob(m->pindex - NUSERLEV3MAPS);
                } else {
                        /* Level 3 page table */
                        pte = pmap_lev2pte(pmap, va);
                        pteva = (vm_offset_t) PTmap + alpha_ptob(m->pindex);
                }

                *pte = 0;

                if (m->pindex < NUSERLEV3MAPS) {
                        /* unhold the level 2 page table */
                        vm_page_t lev2pg;
                        lev2pg = pmap_page_lookup(pmap->pm_pteobj,
                                                  NUSERLEV3MAPS + pmap_lev1_index(va));
                        vm_page_unhold(lev2pg);
                        if (lev2pg->hold_count == 0)
                                _pmap_unwire_pte_hold(pmap, va, lev2pg);
                }

                --pmap->pm_stats.resident_count;
                /*
                 * Do a invltlb to make the invalidated mapping
                 * take effect immediately.
                 */
                pmap_invalidate_page(pmap, pteva);

                if (pmap->pm_ptphint == m)
                        pmap->pm_ptphint = NULL;

                /*
                 * If the page is finally unwired, simply free it.
                 */
                --m->wire_count;
                if (m->wire_count == 0) {

                        if (m->flags & PG_WANTED) {
                                vm_page_flag_clear(m, PG_WANTED);
                                wakeup(m);
                        }

                        vm_page_busy(m);
                        vm_page_free_zero(m);
                        --cnt.v_wire_count;
                }
                return 1;
        }
        return 0;
}

static PMAP_INLINE int
pmap_unwire_pte_hold(pmap_t pmap, vm_offset_t va, vm_page_t m)
{
        vm_page_unhold(m);
        if (m->hold_count == 0)
                return _pmap_unwire_pte_hold(pmap, va, 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)
{
        unsigned ptepindex;
        if (va >= VM_MAXUSER_ADDRESS)
                return 0;

        if (mpte == NULL) {
                ptepindex = (va >> ALPHA_L2SHIFT);
                if (pmap->pm_ptphint &&
                        (pmap->pm_ptphint->pindex == ptepindex)) {
                        mpte = pmap->pm_ptphint;
                } else {
                        mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
                        pmap->pm_ptphint = mpte;
                }
        }

        return pmap_unwire_pte_hold(pmap, va, mpte);
}

void
pmap_pinit0(pmap)
        struct pmap *pmap;
{
        pmap->pm_lev1 = Lev1map;
        pmap->pm_flags = 0;
        pmap->pm_count = 1;
        pmap->pm_ptphint = NULL;
        pmap->pm_active = 0;
        pmap->pm_asn = 0;
        pmap->pm_asngen = 0;
        TAILQ_INIT(&pmap->pm_pvlist);
        bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
}

/*
 * 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 lev1pg;

        /*
         * allocate object for the ptes
         */
        if (pmap->pm_pteobj == NULL)
                pmap->pm_pteobj = vm_object_allocate(OBJT_DEFAULT, NUSERLEV3MAPS + NUSERLEV2MAPS + 1);

        /*
         * allocate the page directory page
         */
        lev1pg = vm_page_grab(pmap->pm_pteobj, NUSERLEV3MAPS + NUSERLEV2MAPS,
                              VM_ALLOC_NORMAL | VM_ALLOC_RETRY);

        lev1pg->wire_count = 1;
        ++cnt.v_wire_count;

        vm_page_flag_clear(lev1pg, PG_MAPPED | PG_BUSY);        /* not mapped normally */
        lev1pg->valid = VM_PAGE_BITS_ALL;

        pmap->pm_lev1 = (pt_entry_t*) ALPHA_PHYS_TO_K0SEG(VM_PAGE_TO_PHYS(lev1pg));
        if ((lev1pg->flags & PG_ZERO) == 0)
                bzero(pmap->pm_lev1, PAGE_SIZE);


        /* install self-referential address mapping entry (not PG_ASM) */
        pmap->pm_lev1[PTLEV1I] = pmap_phys_to_pte(VM_PAGE_TO_PHYS(lev1pg))
                | PG_V | PG_KRE | PG_KWE;

        pmap->pm_flags = 0;
        pmap->pm_count = 1;
        pmap->pm_ptphint = NULL;
        pmap->pm_active = 0;
        pmap->pm_asn = 0;
        pmap->pm_asngen = 0;
        TAILQ_INIT(&pmap->pm_pvlist);
        bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
}

/*
 * Wire in kernel global address entries.  To avoid a race condition
 * between pmap initialization and pmap_growkernel, this procedure
 * should be called after the vmspace is attached to the process
 * but before this pmap is activated.
 */
void
pmap_pinit2(pmap)
        struct pmap *pmap;
{
        bcopy(PTlev1 + K1SEGLEV1I, pmap->pm_lev1 + K1SEGLEV1I, nklev2 * PTESIZE);
}

static int
pmap_release_free_page(pmap_t pmap, vm_page_t p)
{
        pt_entry_t* pte;
        pt_entry_t* l2map;

        if (p->pindex >= NUSERLEV3MAPS + NUSERLEV2MAPS)
                /* level 1 page table */
                pte = &pmap->pm_lev1[PTLEV1I];
        else if (p->pindex >= NUSERLEV3MAPS)
                /* level 2 page table */
                pte = &pmap->pm_lev1[p->pindex - NUSERLEV3MAPS];
        else {
                /* level 3 page table */
                pte = &pmap->pm_lev1[p->pindex >> ALPHA_PTSHIFT];
                l2map = (pt_entry_t*) ALPHA_PHYS_TO_K0SEG(pmap_pte_pa(pte));
                pte = &l2map[p->pindex & ((1 << ALPHA_PTSHIFT) - 1)];
        }

        /*
         * This code optimizes the case of freeing non-busy
         * page-table pages.  Those pages are zero now, and
         * might as well be placed directly into the zero queue.
         */
        if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
                return 0;

        vm_page_busy(p);

        /*
         * Remove the page table page from the processes address space.
         */
        *pte = 0;
        pmap->pm_stats.resident_count--;

#ifdef PMAP_DEBUG
        if (p->hold_count)  {
                panic("pmap_release: freeing held page table page");
        }
#endif
        /*
         * Level1  pages need to have the kernel
         * stuff cleared, so they can go into the zero queue also.
         */
        if (p->pindex == NUSERLEV3MAPS + NUSERLEV2MAPS)
                bzero(pmap->pm_lev1 + K1SEGLEV1I, nklev2 * PTESIZE);

        if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
                pmap->pm_ptphint = NULL;

#ifdef PMAP_DEBUG
        {
            u_long *lp = (u_long*) ALPHA_PHYS_TO_K0SEG(VM_PAGE_TO_PHYS(p));
            u_long *ep = (u_long*) ((char*) lp + PAGE_SIZE);
            for (; lp < ep; lp++)
                if (*lp != 0)
                    panic("pmap_release_free_page: page not zero");
        }
#endif

        p->wire_count--;
        cnt.v_wire_count--;
        vm_page_free_zero(p);
        return 1;
}

/*
 * 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;
{
        pt_entry_t* pte;
        vm_offset_t ptepa;
        vm_page_t m;

        /*
         * Find or fabricate a new pagetable page
         */
        m = vm_page_grab(pmap->pm_pteobj, ptepindex,
                        VM_ALLOC_ZERO | VM_ALLOC_RETRY);

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

        if (m->wire_count == 0)
                cnt.v_wire_count++;
        m->wire_count++;

        /*
         * 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);

        if (ptepindex >= NUSERLEV3MAPS) {
                pte = &pmap->pm_lev1[ptepindex - NUSERLEV3MAPS];
        } else {
                int l1index = ptepindex >> ALPHA_PTSHIFT;
                pt_entry_t* l1pte = &pmap->pm_lev1[l1index];
                pt_entry_t* l2map;
                if (!pmap_pte_v(l1pte))
                        _pmap_allocpte(pmap, NUSERLEV3MAPS + l1index);
                else {
                        vm_page_t l2page =
                                pmap_page_lookup(pmap->pm_pteobj,
                                                 NUSERLEV3MAPS + l1index);
                        l2page->hold_count++;
                }
                l2map = (pt_entry_t*) ALPHA_PHYS_TO_K0SEG(pmap_pte_pa(l1pte));
                pte = &l2map[ptepindex & ((1 << ALPHA_PTSHIFT) - 1)];
        }

        *pte = pmap_phys_to_pte(ptepa) | PG_KRE | PG_KWE | PG_V;

        /*
         * Set the page table hint
         */
        pmap->pm_ptphint = m;

        if ((m->flags & PG_ZERO) == 0)
                bzero((caddr_t) ALPHA_PHYS_TO_K0SEG(ptepa), PAGE_SIZE);

        m->valid = VM_PAGE_BITS_ALL;
        vm_page_flag_clear(m, PG_ZERO | PG_BUSY);
        vm_page_flag_set(m, PG_MAPPED);

        return m;
}

static vm_page_t
pmap_allocpte(pmap_t pmap, vm_offset_t va)
{
        unsigned ptepindex;
        pt_entry_t* lev2pte;
        vm_page_t m;

        /*
         * Calculate pagetable page index
         */
        ptepindex = va >> (PAGE_SHIFT + ALPHA_PTSHIFT);

        /*
         * Get the level2 entry
         */
        lev2pte = pmap_lev2pte(pmap, va);

        /*
         * If the page table page is mapped, we just increment the
         * hold count, and activate it.
         */
        if (lev2pte && pmap_pte_v(lev2pte)) {
                /*
                 * In order to get the page table page, try the
                 * hint first.
                 */
                if (pmap->pm_ptphint &&
                        (pmap->pm_ptphint->pindex == ptepindex)) {
                        m = pmap->pm_ptphint;
                } else {
                        m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
                        pmap->pm_ptphint = m;
                }
                m->hold_count++;
                return m;
        }
        /*
         * Here if the pte page isn't mapped, or if it has been deallocated.
         */
        return _pmap_allocpte(pmap, ptepindex);
}


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

/*
 * 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 p,n,lev1pg;
        vm_object_t object = pmap->pm_pteobj;
        int curgeneration;

#if defined(DIAGNOSTIC)
        if (object->ref_count != 1)
                panic("pmap_release: pteobj reference count != 1");
#endif
        
        lev1pg = NULL;
retry:
        curgeneration = object->generation;
        for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
                n = TAILQ_NEXT(p, listq);
                if (p->pindex >= NUSERLEV3MAPS) {
                        continue;
                }
                while (1) {
                        if (!pmap_release_free_page(pmap, p) &&
                                (object->generation != curgeneration))
                                goto retry;
                }
        }
        for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
                n = TAILQ_NEXT(p, listq);
                if (p->pindex < NUSERLEV3MAPS) {
                        /* can this happen?  maybe panic */
                        goto retry;
                }
                if (p->pindex >= NUSERLEV3MAPS + NUSERLEV2MAPS) {
                        lev1pg = p;
                        continue;
                }
                while (1) {
                        if (!pmap_release_free_page(pmap, p) &&
                                (object->generation != curgeneration))
                                goto retry;
                }
        }

        if (lev1pg && !pmap_release_free_page(pmap, lev1pg))
                goto retry;
}

/*
 * grow the number of kernel page table entries, if needed
 */
void
pmap_growkernel(vm_offset_t addr)
{
        /* XXX come back to this */
        struct proc *p;
        struct pmap *pmap;
        int s;
        pt_entry_t* pte;
        pt_entry_t newlev1, newlev2;
        vm_offset_t pa;
        vm_page_t nkpg;

        s = splhigh();

        if (kernel_vm_end == 0) {
                kernel_vm_end = VM_MIN_KERNEL_ADDRESS;;

                /* Count the level 2 page tables */
                nklev2 = 0;
                nklev3 = 0;
                while (pmap_pte_v(pmap_lev1pte(kernel_pmap, kernel_vm_end))) {
                        nklev2++;
                        nklev3 += (1L << ALPHA_PTSHIFT);
                        kernel_vm_end += ALPHA_L1SIZE;
                }
                        
                /* Count the level 3 page tables in the last level 2 page table */
                kernel_vm_end -= ALPHA_L1SIZE;
                nklev3 -= (1 << ALPHA_PTSHIFT);
                while (pmap_pte_v(pmap_lev2pte(kernel_pmap, kernel_vm_end))) {
                        nklev3++;
                        kernel_vm_end += ALPHA_L2SIZE;
                }
        }

        addr = (addr + ALPHA_L2SIZE) & ~(ALPHA_L2SIZE - 1);
        while (kernel_vm_end < addr) {
                /*
                 * If the level 1 pte is invalid, allocate a new level 2 page table
                 */
                pte = pmap_lev1pte(kernel_pmap, kernel_vm_end);
                if (!pmap_pte_v(pte)) {
                        int pindex = NKLEV3MAPS + pmap_lev1_index(kernel_vm_end) - K1SEGLEV1I;

                        nkpg = vm_page_alloc(kptobj, pindex, VM_ALLOC_SYSTEM);
#if !defined(MAX_PERF)
                        if (!nkpg)
                                panic("pmap_growkernel: no memory to grow kernel");
#endif
                        printf("pmap_growkernel: growing to %lx\n", addr);
                        printf("pmap_growkernel: adding new level2 page table\n");

                        nklev2++;
                        vm_page_wire(nkpg);
                        pa = VM_PAGE_TO_PHYS(nkpg);
                        pmap_zero_page(pa);

                        newlev1 = pmap_phys_to_pte(pa)
                                | PG_V | PG_ASM | PG_KRE | PG_KWE;

                        LIST_FOREACH(p, &allproc, p_list) {
                                if (p->p_vmspace) {
                                        pmap = vmspace_pmap(p->p_vmspace);
                                        *pmap_lev1pte(pmap, kernel_vm_end) = newlev1;
                                }
                        }
                        *pte = newlev1;
                        pmap_invalidate_all(kernel_pmap);
                }

                /*
                 * If the level 2 pte is invalid, allocate a new level 3 page table
                 */
                pte = pmap_lev2pte(kernel_pmap, kernel_vm_end);
                if (pmap_pte_v(pte)) {
                        kernel_vm_end = (kernel_vm_end + ALPHA_L2SIZE) & ~(ALPHA_L2SIZE - 1);
                        continue;
                }

                /*
                 * This index is bogus, but out of the way
                 */
                nkpg = vm_page_alloc(kptobj, nklev3, VM_ALLOC_SYSTEM);
#if !defined(MAX_PERF)
                if (!nkpg)
                        panic("pmap_growkernel: no memory to grow kernel");
#endif

                nklev3++;

                vm_page_wire(nkpg);
                pa = VM_PAGE_TO_PHYS(nkpg);
                pmap_zero_page(pa);
                newlev2 = pmap_phys_to_pte(pa) | PG_V | PG_ASM | PG_KRE | PG_KWE;
                *pte = newlev2;

                kernel_vm_end = (kernel_vm_end + ALPHA_L2SIZE) & ~(ALPHA_L2SIZE - 1);
        }
        splx(s);
}

/*
 *      Retire the given physical map from service.
 *      Should only be called if the map contains
 *      no valid mappings.
 */
void
pmap_destroy(pmap_t pmap)
{
        int count;

        if (pmap == NULL)
                return;

        count = --pmap->pm_count;
        if (count == 0) {
                pmap_release(pmap);
#if !defined(MAX_PERF)
                panic("destroying a pmap is not yet implemented");
#endif
        }
}

/*
 *      Add a reference to the specified pmap.
 */
void
pmap_reference(pmap_t pmap)
{
        if (pmap != NULL) {
                pmap->pm_count++;
        }
}

/***************************************************
* 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--;
        zfreei(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 zalloci(pvzone);
}

/*
 * This routine is very drastic, but can save the system
 * in a pinch.
 */
void
pmap_collect()
{
        pv_table_t *ppv;
        int i;
        vm_offset_t pa;
        vm_page_t m;
        static int warningdone=0;

        if (pmap_pagedaemon_waken == 0)
                return;

        if (warningdone < 5) {
                printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
                warningdone++;
        }

        for(i = 0; i < pv_npg; i++) {
                if ((ppv = &pv_table[i]) == 0)
                        continue;
                m = ppv->pv_vm_page;
                if ((pa = VM_PAGE_TO_PHYS(m)) == 0)
                        continue;
                if (m->wire_count || m->hold_count || m->busy ||
                        (m->flags & PG_BUSY))
                        continue;
                pmap_remove_all(pa);
        }
        pmap_pagedaemon_waken = 0;
}
        

/*
 * If it is the first entry on the list, it is actually
 * in the header and we must copy the following entry up
 * to the header.  Otherwise we must search the list for
 * the entry.  In either case we free the now unused entry.
 */

static int
pmap_remove_entry(pmap_t pmap, pv_table_t* ppv, vm_offset_t va)
{
        pv_entry_t pv;
        int rtval;
        int s;

        s = splvm();
        if (ppv->pv_list_count < pmap->pm_stats.resident_count) {
                for (pv = TAILQ_FIRST(&ppv->pv_list);
                        pv;
                        pv = TAILQ_NEXT(pv, pv_list)) {
                        if (pmap == pv->pv_pmap && va == pv->pv_va) 
                                break;
                }
        } else {
                for (pv = TAILQ_FIRST(&pmap->pm_pvlist);
                        pv;
                        pv = TAILQ_NEXT(pv, pv_plist)) {
                        if (va == pv->pv_va) 
                                break;
                }
        }

        rtval = 0;
        if (pv) {
                rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
                TAILQ_REMOVE(&ppv->pv_list, pv, pv_list);
                ppv->pv_list_count--;
                if (TAILQ_FIRST(&ppv->pv_list) == NULL)
                        vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE);

                TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
                free_pv_entry(pv);
        }
                        
        splx(s);
        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_offset_t pa)
{

        int s;
        pv_entry_t pv;
        pv_table_t *ppv;

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

        TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);

        ppv = pa_to_pvh(pa);
        TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list);
        ppv->pv_list_count++;

        splx(s);
}

/*
 * 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;
        pv_table_t *ppv;

        oldpte = *ptq;
        PMAP_DEBUG_VA(va);
        *ptq = 0;
        if (oldpte & PG_W)
                pmap->pm_stats.wired_count -= 1;

        pmap->pm_stats.resident_count -= 1;
        if (oldpte & PG_MANAGED) {
                ppv = pa_to_pvh(pmap_pte_pa(&oldpte));
                return pmap_remove_entry(pmap, ppv, va);
        } else {
                return pmap_unuse_pt(pmap, va, NULL);
        }

        return 0;
}

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

        ptq = pmap_lev3pte(pmap, va);
        
        /*
         * if there is no pte for this address, just skip it!!!
         */
        if (!ptq || !pmap_pte_v(ptq))
                return;

        /*
         * get a local va for mappings for this pmap.
         */
        (void) pmap_remove_pte(pmap, ptq, va);
        pmap_invalidate_page(pmap, va);

        return;
}

/*
 *      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 va, nva;

        if (pmap == NULL)
                return;

        if (pmap->pm_stats.resident_count == 0)
                return;

        /*
         * special handling of removing one page.  a very
         * common operation and easy to short circuit some
         * code.
         */
        if (sva + PAGE_SIZE == eva) {
                pmap_remove_page(pmap, sva);
                return;
        }

        for (va = sva; va < eva; va = nva) {
                if (!pmap_pte_v(pmap_lev1pte(pmap, va))) {
                        nva = alpha_l1trunc(va + ALPHA_L1SIZE);
                        continue;
                }

                if (!pmap_pte_v(pmap_lev2pte(pmap, va))) {
                        nva = alpha_l2trunc(va + ALPHA_L2SIZE);
                        continue;
                }

                pmap_remove_page(pmap, va);
                nva = va + PAGE_SIZE;
        }
}

/*
 *      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...)
 */

static void
pmap_remove_all(vm_offset_t pa)
{
        register pv_entry_t pv;
        pv_table_t *ppv;
        pt_entry_t *pte, tpte;
        int nmodify;
        int s;

        nmodify = 0;
#if defined(PMAP_DIAGNOSTIC)
        /*
         * XXX this makes pmap_page_protect(NONE) illegal for non-managed
         * pages!
         */
        if (!pmap_is_managed(pa)) {
                panic("pmap_page_protect: illegal for unmanaged page, va: 0x%lx", pa);
        }
#endif

        s = splvm();
        ppv = pa_to_pvh(pa);
        while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) {
                pte = pmap_lev3pte(pv->pv_pmap, pv->pv_va);

                pv->pv_pmap->pm_stats.resident_count--;

                if (pmap_pte_pa(pte) != pa)
                        panic("pmap_remove_all: pv_table for %lx is inconsistent", pa);

                tpte = *pte;

                PMAP_DEBUG_VA(pv->pv_va);
                *pte = 0;
                if (tpte & PG_W)
                        pv->pv_pmap->pm_stats.wired_count--;

                pmap_invalidate_page(pv->pv_pmap, pv->pv_va);

                TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
                TAILQ_REMOVE(&ppv->pv_list, pv, pv_list);
                ppv->pv_list_count--;
                pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
                free_pv_entry(pv);
        }

        vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE);

        splx(s);
        return;
}

/*
 *      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)
{
        pt_entry_t* pte;
        int newprot;

        if (pmap == NULL)
                return;

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

        if (prot & VM_PROT_WRITE)
                return;

        newprot = pte_prot(pmap, prot);

        if ((sva & PAGE_MASK) || (eva & PAGE_MASK))
                panic("pmap_protect: unaligned addresses");

        while (sva < eva) {

                /*
                 * If level 1 pte is invalid, skip this segment
                 */
                pte = pmap_lev1pte(pmap, sva);
                if (!pmap_pte_v(pte)) {
                        sva = alpha_l1trunc(sva) + ALPHA_L1SIZE;
                        continue;
                }

                /*
                 * If level 2 pte is invalid, skip this segment
                 */
                pte = pmap_lev2pte(pmap, sva);
                if (!pmap_pte_v(pte)) {
                        sva = alpha_l2trunc(sva) + ALPHA_L2SIZE;
                        continue;
                }

                /* 
                 * If level 3 pte is invalid, skip this page
                 */
                pte = pmap_lev3pte(pmap, sva);
                if (!pmap_pte_v(pte)) {
                        sva += PAGE_SIZE;
                        continue;
                }

                if (pmap_pte_prot(pte) != newprot) {
                        pmap_pte_set_prot(pte, newprot);
                        pmap_invalidate_page(pmap, sva);
                }

                sva += PAGE_SIZE;
        }
}

/*
 *      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_offset_t pa, vm_prot_t prot,
           boolean_t wired)
{
        pt_entry_t *pte;
        vm_offset_t opa;
        pt_entry_t origpte, newpte;
        vm_page_t mpte;
        int managed;

        if (pmap == NULL)
                return;

        va &= ~PAGE_MASK;
#ifdef PMAP_DIAGNOSTIC
        if (va > VM_MAX_KERNEL_ADDRESS)
                panic("pmap_enter: toobig");
#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);
        }

        pte = pmap_lev3pte(pmap, va);

#if !defined(MAX_PERF)
        /*
         * Page Directory table entry not valid, we need a new PT page
         */
        if (pte == NULL) {
                panic("pmap_enter: invalid kernel page tables pmap=%p, va=0x%lx\n", pmap, va);
        }
#endif

        origpte = *pte;
        pa &= ~PAGE_MASK;
        managed = 0;
        opa = pmap_pte_pa(pte);

        /*
         * 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--;

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

                managed = origpte & PG_MANAGED;
                goto validate;
        } 
        /*
         * Mapping has changed, invalidate old range and fall through to
         * handle validating new mapping.
         */
        if (opa) {
                int err;
                err = pmap_remove_pte(pmap, pte, va);
#if !defined(MAX_PERF)
                if (err)
                        panic("pmap_enter: pte vanished, va: 0x%lx", va);
#endif
        }

        /*
         * 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_is_managed(pa)) {
                pmap_insert_entry(pmap, va, mpte, pa);
                managed |= 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 = pmap_phys_to_pte(pa) | pte_prot(pmap, prot) | PG_V | managed;

        if (managed) {
                pv_table_t* ppv;

                /*
                 * Set up referenced/modified emulation for the new mapping
                 */
                ppv = pa_to_pvh(pa);
                if ((ppv->pv_flags & PV_TABLE_REF) == 0)
                        newpte |= PG_FOR | PG_FOW | PG_FOE;
                else if ((ppv->pv_flags & PV_TABLE_MOD) == 0)
                        newpte |= PG_FOW;
        }

        if (wired)
                newpte |= PG_W;

        /*
         * if the mapping or permission bits are different, we need
         * to update the pte.
         */
        if (origpte != newpte) {
                PMAP_DEBUG_VA(va);
                *pte = newpte;
                if (origpte)
                        pmap_invalidate_page(pmap, va);
                if (prot & VM_PROT_EXECUTE)
                        alpha_pal_imb();
        }
}

/*
 * 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...
 */

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


        /*
         * In the case that a page table page is not
         * resident, we are creating it here.
         */
        if (va < VM_MAXUSER_ADDRESS) {
                unsigned ptepindex;
                pt_entry_t* l2pte;

                /*
                 * Calculate lev2 page index
                 */
                ptepindex = va >> ALPHA_L2SHIFT;
                if (mpte && (mpte->pindex == ptepindex)) {
                        mpte->hold_count++;
                } else {
retry:
                        /*
                         * Get the level 2 entry
                         */
                        l2pte = pmap_lev2pte(pmap, va);

                        /*
                         * If the level 2 page table is mapped, we just increment
                         * the hold count, and activate it.
                         */
                        if (l2pte && pmap_pte_v(l2pte)) {
                                if (pmap->pm_ptphint &&
                                        (pmap->pm_ptphint->pindex == ptepindex)) {
                                        mpte = pmap->pm_ptphint;
                                } else {
                                        mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
                                        pmap->pm_ptphint = mpte;
                                }
                                if (mpte == NULL)
                                        goto retry;
                                mpte->hold_count++;
                        } else {
                                mpte = _pmap_allocpte(pmap, ptepindex);
                        }
                }
        } 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)
                        pmap_unwire_pte_hold(pmap, va, mpte);
                alpha_pal_imb();                /* XXX overkill? */
                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.
         */
        PMAP_DEBUG_VA(va);
        pmap_insert_entry(pmap, va, mpte, pa);

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

        /*
         * Now validate mapping with RO protection
         */
        *pte = pmap_phys_to_pte(pa) | PG_V | PG_KRE | PG_URE | PG_MANAGED;

        alpha_pal_imb();                        /* XXX overkill? */
        return mpte;
}

#define MAX_INIT_PT (96)
/*
 * pmap_object_init_pt preloads the ptes for a given object
 * into the specified pmap.  This eliminates the blast of soft
 * faults on process startup and immediately after an mmap.
 */
void
pmap_object_init_pt(pmap_t pmap, vm_offset_t addr,
                    vm_object_t object, vm_pindex_t pindex,
                    vm_size_t size, int limit)
{
        vm_offset_t tmpidx;
        int psize;
        vm_page_t p, mpte;
        int objpgs;

        if (pmap == NULL || object == NULL)
                return;

        psize = alpha_btop(size);

        if ((object->type != OBJT_VNODE) ||
                (limit && (psize > MAX_INIT_PT) &&
                        (object->resident_page_count > MAX_INIT_PT))) {
                return;
        }

        if (psize + pindex > object->size)
                psize = object->size - pindex;

        mpte = NULL;
        /*
         * if we are processing a major portion of the object, then scan the
         * entire thing.
         */
        if (psize > (object->resident_page_count >> 2)) {
                objpgs = psize;

                for (p = TAILQ_FIRST(&object->memq);
                    ((objpgs > 0) && (p != NULL));
                    p = TAILQ_NEXT(p, listq)) {

                        tmpidx = p->pindex;
                        if (tmpidx < pindex) {
                                continue;
                        }
                        tmpidx -= pindex;
                        if (tmpidx >= psize) {
                                continue;
                        }
                        if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
                            (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
                                if ((p->queue - p->pc) == PQ_CACHE)
                                        vm_page_deactivate(p);
                                vm_page_busy(p);
                                mpte = pmap_enter_quick(pmap, 
                                        addr + alpha_ptob(tmpidx),
                                        VM_PAGE_TO_PHYS(p), mpte);
                                vm_page_flag_set(p, PG_MAPPED);
                                vm_page_wakeup(p);
                        }
                        objpgs -= 1;
                }
        } else {
                /*
                 * else lookup the pages one-by-one.
                 */
                for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
                        p = vm_page_lookup(object, tmpidx + pindex);
                        if (p &&
                            ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
                            (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
                                if ((p->queue - p->pc) == PQ_CACHE)
                                        vm_page_deactivate(p);
                                vm_page_busy(p);
                                mpte = pmap_enter_quick(pmap, 
                                        addr + alpha_ptob(tmpidx),
                                        VM_PAGE_TO_PHYS(p), mpte);
                                vm_page_flag_set(p, PG_MAPPED);
                                vm_page_wakeup(p);
                        }
                }
        }
        return;
}

/*
 * pmap_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.
 */
#define PFBAK 4
#define PFFOR 4
#define PAGEORDER_SIZE (PFBAK+PFFOR)

static int pmap_prefault_pageorder[] = {
        -PAGE_SIZE, PAGE_SIZE,
        -2 * PAGE_SIZE, 2 * PAGE_SIZE,
        -3 * PAGE_SIZE, 3 * PAGE_SIZE
        -4 * PAGE_SIZE, 4 * PAGE_SIZE
};

void
pmap_prefault(pmap, addra, entry)
        pmap_t pmap;
        vm_offset_t addra;
        vm_map_entry_t entry;
{
        int i;
        vm_offset_t starta;
        vm_offset_t addr;
        vm_pindex_t pindex;
        vm_page_t m, mpte;
        vm_object_t object;

        if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
                return;

        object = entry->object.vm_object;

        starta = addra - PFBAK * PAGE_SIZE;
        if (starta < entry->start) {
                starta = entry->start;
        } else if (starta > addra) {
                starta = 0;
        }

        mpte = NULL;
        for (i = 0; i < PAGEORDER_SIZE; i++) {
                vm_object_t lobject;
                pt_entry_t *pte;

                addr = addra + pmap_prefault_pageorder[i];
                if (addr > addra + (PFFOR * PAGE_SIZE))
                        addr = 0;

                if (addr < starta || addr >= entry->end)
                        continue;

                if (!pmap_pte_v(pmap_lev1pte(pmap, addr))
                    || !pmap_pte_v(pmap_lev2pte(pmap, addr)))
                        continue;

                pte = vtopte(addr);
                if (*pte)
                        continue;

                pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
                lobject = object;
                for (m = vm_page_lookup(lobject, pindex);
                    (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
                    lobject = lobject->backing_object) {
                        if (lobject->backing_object_offset & PAGE_MASK)
                                break;
                        pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
                        m = vm_page_lookup(lobject->backing_object, pindex);
                }

                /*
                 * give-up when a page is not in memory
                 */
                if (m == NULL)
                        break;

                if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
                    (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {

                        if ((m->queue - m->pc) == PQ_CACHE) {
                                vm_page_deactivate(m);
                        }
                        vm_page_busy(m);
                        mpte = pmap_enter_quick(pmap, addr,
                                VM_PAGE_TO_PHYS(m), mpte);
                        vm_page_flag_set(m, PG_MAPPED);
                        vm_page_wakeup(m);
                }
        }
}

/*
 *      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;
{
        pt_entry_t *pte;

        if (pmap == NULL)
                return;

        pte = pmap_lev3pte(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);
}



/*
 *      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)
{
}       

/*
 *      Routine:        pmap_kernel
 *      Function:
 *              Returns the physical map handle for the kernel.
 */
pmap_t
pmap_kernel()
{
        return (kernel_pmap);
}

/*
 *      pmap_zero_page zeros the specified hardware page by
 *      mapping it into virtual memory and using bzero to clear
 *      its contents.
 */

void
pmap_zero_page(vm_offset_t pa)
{
        vm_offset_t va = ALPHA_PHYS_TO_K0SEG(pa);
        bzero((caddr_t) va, PAGE_SIZE);
}


/*
 *      pmap_zero_page_area zeros the specified hardware page by
 *      mapping it into virtual memory and using bzero to clear
 *      its contents.
 *
 *      off and size must reside within a single page.
 */

void
pmap_zero_page_area(vm_offset_t pa, int off, int size)
{
        vm_offset_t va = ALPHA_PHYS_TO_K0SEG(pa);
        bzero((char *)(caddr_t)va + off, size);
}

/*
 *      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_offset_t src, vm_offset_t dst)
{
        src = ALPHA_PHYS_TO_K0SEG(src);
        dst = ALPHA_PHYS_TO_K0SEG(dst);
        bcopy((caddr_t) src, (caddr_t) dst, PAGE_SIZE);
}


/*
 *      Routine:        pmap_pageable
 *      Function:
 *              Make the specified pages (by pmap, offset)
 *              pageable (or not) as requested.
 *
 *              A page which is not pageable may not take
 *              a fault; therefore, its page table entry
 *              must remain valid for the duration.
 *
 *              This routine is merely advisory; pmap_enter
 *              will specify that these pages are to be wired
 *              down (or not) as appropriate.
 */
void
pmap_pageable(pmap, sva, eva, pageable)
        pmap_t pmap;
        vm_offset_t sva, eva;
        boolean_t pageable;
{
}

/*
 * this routine returns true if a physical page resides
 * in the given pmap.
 */
boolean_t
pmap_page_exists(pmap, pa)
        pmap_t pmap;
        vm_offset_t pa;
{
        register pv_entry_t pv;
        pv_table_t *ppv;
        int s;

        if (!pmap_is_managed(pa))
                return FALSE;

        s = splvm();

        ppv = pa_to_pvh(pa);
        /*
         * Not found, check current mappings returning immediately if found.
         */
        for (pv = TAILQ_FIRST(&ppv->pv_list);
                pv;
                pv = TAILQ_NEXT(pv, pv_list)) {
                if (pv->pv_pmap == pmap) {
                        splx(s);
                        return TRUE;
                }
        }
        splx(s);
        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;
        pv_table_t *ppv;
        pv_entry_t pv, npv;
        int s;

#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
        if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
                printf("warning: pmap_remove_pages called with non-current pmap\n");
                return;
        }
#endif

        s = splvm();
        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(pv->pv_pmap, pv->pv_va);
#endif
                if (!pmap_pte_v(pte))
                        panic("pmap_remove_pages: page on pm_pvlist has no pte\n");
                tpte = *pte;


/*
 * We cannot remove wired pages from a process' mapping at this time
 */
                if (tpte & PG_W) {
                        npv = TAILQ_NEXT(pv, pv_plist);
                        continue;
                }
                PMAP_DEBUG_VA(pv->pv_va);
                *pte = 0;

                ppv = pa_to_pvh(pmap_pte_pa(&tpte));

                pv->pv_pmap->pm_stats.resident_count--;

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

                ppv->pv_list_count--;
                TAILQ_REMOVE(&ppv->pv_list, pv, pv_list);
                if (TAILQ_FIRST(&ppv->pv_list) == NULL) {
                        vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE);
                }

                pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
                free_pv_entry(pv);
        }
        splx(s);
        pmap_invalidate_all(pmap);
}

/*
 * this routine is used to modify bits in ptes
 */
static void
pmap_changebit(vm_offset_t pa, int bit, boolean_t setem)
{
        pv_entry_t pv;
        pv_table_t *ppv;
        pt_entry_t *pte;
        int changed;
        int s;

        if (!pmap_is_managed(pa))
                return;

        s = splvm();
        changed = 0;
        ppv = pa_to_pvh(pa);

        /*
         * Loop over all current mappings setting/clearing as appropos If
         * setting RO do we need to clear the VAC?
         */
        for (pv = TAILQ_FIRST(&ppv->pv_list);
                pv;
                pv = TAILQ_NEXT(pv, pv_list)) {

                /*
                 * don't write protect pager mappings
                 */
                if (!setem && bit == (PG_UWE|PG_KWE)) {
                        if (!pmap_track_modified(pv->pv_va))
                                continue;
                }

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

                pte = pmap_lev3pte(pv->pv_pmap, pv->pv_va);

                changed = 0;
                if (setem) {
                        *pte |= bit;
                        changed = 1;
                } else {
                        pt_entry_t pbits = *pte;
                        if (pbits & bit) {
                                changed = 1;
                                *pte = pbits & ~bit;
                        }
                }
                if (changed)
                        pmap_invalidate_page(pv->pv_pmap, pv->pv_va);
        }
        splx(s);
}

/*
 *      pmap_page_protect:
 *
 *      Lower the permission for all mappings to a given page.
 */
void
pmap_page_protect(vm_offset_t phys, vm_prot_t prot)
{
        if ((prot & VM_PROT_WRITE) == 0) {
                if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
                        pmap_changebit(phys, PG_KWE|PG_UWE, FALSE);
                } else {
                        pmap_remove_all(phys);
                }
        }
}

vm_offset_t
pmap_phys_address(ppn)
        int ppn;
{
        return (alpha_ptob(ppn));
}

/*
 *      pmap_ts_referenced:
 *
 *      Return the count of reference bits for a page, clearing all of them.
 *      
 */
int
pmap_ts_referenced(vm_offset_t pa)
{
        pv_table_t *ppv;

        if (!pmap_is_managed(pa))
                return 0;

        ppv = pa_to_pvh(pa);

        if (ppv->pv_flags & PV_TABLE_REF) {
                pmap_changebit(pa, PG_FOR|PG_FOE|PG_FOW, TRUE);
                ppv->pv_flags &= ~PV_TABLE_REF;
                return 1;
        }

        return 0;
}

/*
 *      pmap_is_modified:
 *
 *      Return whether or not the specified physical page was modified
 *      in any physical maps.
 */
boolean_t
pmap_is_modified(vm_offset_t pa)
{
        pv_table_t *ppv;

        if (!pmap_is_managed(pa))
                return FALSE;

        ppv = pa_to_pvh(pa);

        return (ppv->pv_flags & PV_TABLE_MOD) != 0;
}

/*
 *      Clear the modify bits on the specified physical page.
 */
void
pmap_clear_modify(vm_offset_t pa)
{
        pv_table_t *ppv;

        if (!pmap_is_managed(pa))
                return;

        ppv = pa_to_pvh(pa);

        if (ppv->pv_flags & PV_TABLE_MOD) {
                pmap_changebit(pa, PG_FOW, TRUE);
                ppv->pv_flags &= ~PV_TABLE_MOD;
        }
}

/*
 *      pmap_page_is_free:
 *
 *      Called when a page is freed to allow pmap to clean up
 *      any extra state associated with the page.  In this case
 *      clear modified/referenced bits.
 */
void
pmap_page_is_free(vm_page_t m)
{
        pv_table_t *ppv;

        ppv = pa_to_pvh(VM_PAGE_TO_PHYS(m));
        ppv->pv_flags = 0;
}

/*
 *      pmap_clear_reference:
 *
 *      Clear the reference bit on the specified physical page.
 */
void
pmap_clear_reference(vm_offset_t pa)
{
        pv_table_t *ppv;

        if (!pmap_is_managed(pa))
                return;

        ppv = pa_to_pvh(pa);

        if (ppv->pv_flags & PV_TABLE_REF) {
                pmap_changebit(pa, PG_FOR|PG_FOE|PG_FOW, TRUE);
                ppv->pv_flags &= ~PV_TABLE_REF;
        }
}

/*
 * pmap_emulate_reference:
 *
 *      Emulate reference and/or modified bit hits.
 *      From NetBSD
 */
void
pmap_emulate_reference(struct proc *p, vm_offset_t v, int user, int write)
{
        pt_entry_t faultoff, *pte;
        vm_offset_t pa;
        pv_table_t *ppv;

        /*
         * Convert process and virtual address to physical address.
         */
        if (v >= VM_MIN_KERNEL_ADDRESS) {
                if (user)
                        panic("pmap_emulate_reference: user ref to kernel");
                pte = vtopte(v);
        } else {
#ifdef DIAGNOSTIC
                if (p == NULL)
                        panic("pmap_emulate_reference: bad proc");
                if (p->p_vmspace == NULL)
                        panic("pmap_emulate_reference: bad p_vmspace");
#endif
                pte = pmap_lev3pte(p->p_vmspace->vm_map.pmap, v);
        }
#ifdef DEBUG                            /* These checks are more expensive */
        if (!pmap_pte_v(pte))
                panic("pmap_emulate_reference: invalid pte");
#if 0
        /*
         * Can't do these, because cpu_fork and cpu_swapin call
         * pmap_emulate_reference(), and the bits aren't guaranteed,
         * for them...
         */
        if (write) {
                if (!(*pte & (user ? PG_UWE : PG_UWE | PG_KWE)))
                        panic("pmap_emulate_reference: write but unwritable");
                if (!(*pte & PG_FOW))
                        panic("pmap_emulate_reference: write but not FOW");
        } else {
                if (!(*pte & (user ? PG_URE : PG_URE | PG_KRE)))
                        panic("pmap_emulate_reference: !write but unreadable");
                if (!(*pte & (PG_FOR | PG_FOE)))
                        panic("pmap_emulate_reference: !write but not FOR|FOE");
        }
#endif
        /* Other diagnostics? */
#endif
        pa = pmap_pte_pa(pte);

#ifdef DIAGNOSTIC
        if ((*pte & PG_MANAGED) == 0)
                panic("pmap_emulate_reference(%p, 0x%lx, %d, %d): pa 0x%lx not managed", p, v, user, write, pa);
#endif

        /*
         * Twiddle the appropriate bits to reflect the reference
         * and/or modification..
         *
         * The rules:
         *      (1) always mark page as used, and
         *      (2) if it was a write fault, mark page as modified.
         */
        ppv = pa_to_pvh(pa);
        ppv->pv_flags |= PV_TABLE_REF;
        faultoff = PG_FOR | PG_FOE;
        vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED);
        if (write) {
                ppv->pv_flags |= PV_TABLE_MOD;
                vm_page_dirty(ppv->pv_vm_page);
                faultoff |= PG_FOW;
        }
        pmap_changebit(pa, faultoff, FALSE);
        if ((*pte & faultoff) != 0) {
#if 1
                /*
                 * XXX dfr - don't think its possible in our pmap
                 */
                /*
                 * This is apparently normal.  Why? -- cgd
                 * XXX because was being called on unmanaged pages?
                 */
                panic("warning: pmap_changebit didn't.");
#endif
                *pte &= ~faultoff;
                ALPHA_TBIS(v);
        }
}

/*
 * Miscellaneous support routines follow
 */

static void
alpha_protection_init()
{
        int prot, *kp, *up;

        kp = protection_codes[0];
        up = protection_codes[1];

        for (prot = 0; prot < 8; prot++) {
                switch (prot) {
                case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
                        *kp++ = PG_ASM;
                        *up++ = 0;
                        break;
                case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
                case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
                case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
                        *kp++ = PG_ASM | PG_KRE;
                        *up++ = PG_URE | PG_KRE;
                        break;
                case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
                        *kp++ = PG_ASM | PG_KWE;
                        *up++ = PG_UWE | PG_KWE;
                        break;
                case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
                case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
                case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
                        *kp++ = PG_ASM | PG_KWE | PG_KRE;
                        *up++ = PG_UWE | PG_URE | PG_KWE | PG_KRE;
                        break;
                }
        }
}

/*
 * 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_offset_t pa;
        vm_size_t size;
{
        return (void*) ALPHA_PHYS_TO_K0SEG(pa);
}

/*
 * perform the pmap work for mincore
 */
int
pmap_mincore(pmap, addr)
        pmap_t pmap;
        vm_offset_t addr;
{
        
        pt_entry_t *pte;
        vm_page_t m;
        int val = 0;
        
        pte = pmap_lev3pte(pmap, addr);
        if (pte == 0) {
                return 0;
        }

        if (pmap_pte_v(pte)) {
                pv_table_t *ppv;
                vm_offset_t pa;

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

                pa = pmap_pte_pa(pte);

                ppv = pa_to_pvh(pa);
                m = ppv->pv_vm_page;

                /*
                 * Modified by us
                 */
                if (ppv->pv_flags & PV_TABLE_MOD)
                        val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
                /*
                 * Modified by someone
                 */
                else if (m->dirty || pmap_is_modified(pa))
                        val |= MINCORE_MODIFIED_OTHER;
                /*
                 * Referenced by us
                 */
                if (ppv->pv_flags & PV_TABLE_REF)
                        val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;

                /*
                 * Referenced by someone
                 */
                else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(pa)) {
                        val |= MINCORE_REFERENCED_OTHER;
                        vm_page_flag_set(m, PG_REFERENCED);
                }
        } 
        return val;
}

void
pmap_activate(struct proc *p)
{
        pmap_t pmap;

        pmap = vmspace_pmap(p->p_vmspace);

        if (pmap_active && pmap != pmap_active) {
                pmap_active->pm_active = 0;
                pmap_active = 0;
        }

        p->p_addr->u_pcb.pcb_hw.apcb_ptbr =
                ALPHA_K0SEG_TO_PHYS((vm_offset_t) pmap->pm_lev1) >> PAGE_SHIFT;

        if (pmap->pm_asngen != pmap_current_asngen)
                pmap_get_asn(pmap);

        pmap_active = pmap;
        pmap->pm_active = 1;    /* XXX use bitmap for SMP */

        p->p_addr->u_pcb.pcb_hw.apcb_asn = pmap->pm_asn;

        if (p == curproc) {
                alpha_pal_swpctx((u_long)p->p_md.md_pcbpaddr);
        }
}

void
pmap_deactivate(struct proc *p)
{
        pmap_t pmap;
        pmap = vmspace_pmap(p->p_vmspace);
        pmap->pm_active = 0;
        pmap_active = 0;
}

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

        return addr;
}

#if 0
#if defined(PMAP_DEBUG)
pmap_pid_dump(int pid) {
        pmap_t pmap;
        struct proc *p;
        int npte = 0;
        int index;
        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<1024;i++) {
                                pd_entry_t *pde;
                                pt_entry_t *pte;
                                unsigned base = i << PDRSHIFT;
                                
                                pde = &pmap->pm_pdir[i];
                                if (pde && pmap_pde_v(pde)) {
                                        for(j=0;j<1024;j++) {
                                                unsigned va = base + (j << PAGE_SHIFT);
                                                if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
                                                        if (index) {
                                                                index = 0;
                                                                printf("\n");
                                                        }
                                                        return npte;
                                                }
                                                pte = pmap_pte_quick( pmap, va);
                                                if (pte && pmap_pte_v(pte)) {
                                                        vm_offset_t pa;
                                                        vm_page_t m;
                                                        pa = *(int *)pte;
                                                        m = PHYS_TO_VM_PAGE((pa & PG_FRAME));
                                                        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(" ");
                                                        }
                                                }
                                        }
                                }
                        }
                }
        }
        return npte;
}
#endif

#if defined(DEBUG)

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

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

        if (pm == kernel_pmap)
                return;
        for (i = 0; i < 1024; i++)
                if (pm->pm_pdir[i])
                        for (j = 0; j < 1024; 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_quick(pm, va);
                                if (pmap_pte_v(ptep))
                                        printf("%x:%x ", va, *(int *) ptep);
                        };

}

static void
pmap_pvdump(pa)
        vm_offset_t pa;
{
        pv_table_t *ppv;
        register pv_entry_t pv;

        printf("pa %x", pa);
        ppv = pa_to_pvh(pa);
        for (pv = TAILQ_FIRST(&ppv->pv_list);
                pv;
                pv = TAILQ_NEXT(pv, pv_list)) {
#ifdef used_to_be
                printf(" -> pmap %x, va %x, flags %x",
                    pv->pv_pmap, pv->pv_va, pv->pv_flags);
#endif
                printf(" -> pmap %x, va %x",
                    pv->pv_pmap, pv->pv_va);
                pads(pv->pv_pmap);
        }
        printf(" ");
}
#endif
#endif