pmap.c revision 1890
1/*
2 * Copyright (c) 1991 Regents of the University of California.
3 * All rights reserved.
4 * Copyright (c) 1994 John S. Dyson
5 * All rights reserved.
6 * Copyright (c) 1994 David Greenman
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * the Systems Programming Group of the University of Utah Computer
11 * Science Department and William Jolitz of UUNET Technologies Inc.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 *    notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 *    notice, this list of conditions and the following disclaimer in the
20 *    documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 *    must display the following acknowledgement:
23 *	This product includes software developed by the University of
24 *	California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 *    may be used to endorse or promote products derived from this software
27 *    without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * SUCH DAMAGE.
40 *
41 *	from:	@(#)pmap.c	7.7 (Berkeley)	5/12/91
42 *	$Id: pmap.c,v 1.28 1994/08/06 09:15:15 davidg Exp $
43 */
44
45/*
46 * Derived from hp300 version by Mike Hibler, this version by William
47 * Jolitz uses a recursive map [a pde points to the page directory] to
48 * map the page tables using the pagetables themselves. This is done to
49 * reduce the impact on kernel virtual memory for lots of sparse address
50 * space, and to reduce the cost of memory to each process.
51 *
52 *	Derived from: hp300/@(#)pmap.c	7.1 (Berkeley) 12/5/90
53 */
54/*
55 * Major modifications by John S. Dyson primarily to support
56 * pageable page tables, eliminating pmap_attributes,
57 * discontiguous memory pages, and using more efficient string
58 * instructions. Jan 13, 1994.  Further modifications on Mar 2, 1994,
59 * general clean-up and efficiency mods.
60 */
61
62/*
63 *	Manages physical address maps.
64 *
65 *	In addition to hardware address maps, this
66 *	module is called upon to provide software-use-only
67 *	maps which may or may not be stored in the same
68 *	form as hardware maps.  These pseudo-maps are
69 *	used to store intermediate results from copy
70 *	operations to and from address spaces.
71 *
72 *	Since the information managed by this module is
73 *	also stored by the logical address mapping module,
74 *	this module may throw away valid virtual-to-physical
75 *	mappings at almost any time.  However, invalidations
76 *	of virtual-to-physical mappings must be done as
77 *	requested.
78 *
79 *	In order to cope with hardware architectures which
80 *	make virtual-to-physical map invalidates expensive,
81 *	this module may delay invalidate or reduced protection
82 *	operations until such time as they are actually
83 *	necessary.  This module is given full information as
84 *	to which processors are currently using which maps,
85 *	and to when physical maps must be made correct.
86 */
87
88#include <sys/param.h>
89#include <sys/proc.h>
90#include <sys/malloc.h>
91#include <sys/user.h>
92
93#include <vm/vm.h>
94#include <vm/vm_kern.h>
95#include <vm/vm_page.h>
96
97#include <i386/include/cpufunc.h>
98#include <i386/include/cputypes.h>
99
100#include <i386/isa/isa.h>
101
102/*
103 * Allocate various and sundry SYSMAPs used in the days of old VM
104 * and not yet converted.  XXX.
105 */
106#define BSDVM_COMPAT	1
107
108/*
109 * Get PDEs and PTEs for user/kernel address space
110 */
111#define	pmap_pde(m, v)	(&((m)->pm_pdir[((vm_offset_t)(v) >> PD_SHIFT)&1023]))
112#define pdir_pde(m, v) (m[((vm_offset_t)(v) >> PD_SHIFT)&1023])
113
114#define pmap_pte_pa(pte)	(*(int *)(pte) & PG_FRAME)
115
116#define pmap_pde_v(pte)		((*(int *)pte & PG_V) != 0)
117#define pmap_pte_w(pte)		((*(int *)pte & PG_W) != 0)
118#define pmap_pte_m(pte)		((*(int *)pte & PG_M) != 0)
119#define pmap_pte_u(pte)		((*(int *)pte & PG_U) != 0)
120#define pmap_pte_v(pte)		((*(int *)pte & PG_V) != 0)
121
122#define pmap_pte_set_w(pte, v)		((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
123#define pmap_pte_set_prot(pte, v)	((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
124
125/*
126 * Given a map and a machine independent protection code,
127 * convert to a vax protection code.
128 */
129#define pte_prot(m, p)	(protection_codes[p])
130int	protection_codes[8];
131
132struct pmap	kernel_pmap_store;
133pmap_t		kernel_pmap;
134
135vm_offset_t	phys_avail[6];		/* 2 entries + 1 null */
136vm_offset_t    	avail_start;		/* PA of first available physical page */
137vm_offset_t	avail_end;		/* PA of last available physical page */
138vm_size_t	mem_size;		/* memory size in bytes */
139vm_offset_t	virtual_avail; 		/* VA of first avail page (after kernel bss)*/
140vm_offset_t	virtual_end;		/* VA of last avail page (end of kernel AS) */
141int		i386pagesperpage;	/* PAGE_SIZE / I386_PAGE_SIZE */
142boolean_t	pmap_initialized = FALSE; /* Has pmap_init completed? */
143vm_offset_t	vm_first_phys, vm_last_phys;
144
145static inline boolean_t		pmap_testbit();
146static inline void		pmap_changebit();
147static inline int		pmap_is_managed();
148static inline void *		vm_get_pmap();
149static inline void		vm_put_pmap();
150static void			i386_protection_init();
151static void			pmap_alloc_pv_entry();
152static inline pv_entry_t	get_pv_entry();
153static inline void		pmap_use_pt();
154static inline void		pmap_unuse_pt();
155
156inline pt_entry_t * pmap_pte();
157
158extern vm_offset_t clean_sva, clean_eva;
159extern int cpu_class;
160
161#if BSDVM_COMPAT
162#include "msgbuf.h"
163
164/*
165 * All those kernel PT submaps that BSD is so fond of
166 */
167pt_entry_t *CMAP1, *CMAP2, *ptmmap;
168caddr_t		CADDR1, CADDR2, ptvmmap;
169pt_entry_t *msgbufmap;
170struct msgbuf	*msgbufp;
171#endif
172
173void init_pv_entries(int) ;
174
175/*
176 *	Routine:	pmap_pte
177 *	Function:
178 *		Extract the page table entry associated
179 *		with the given map/virtual_address pair.
180 * [ what about induced faults -wfj]
181 */
182
183inline pt_entry_t *
184const pmap_pte(pmap, va)
185	register pmap_t	pmap;
186	vm_offset_t va;
187{
188
189	if (pmap && *pmap_pde(pmap, va)) {
190		vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
191		/* are we current address space or kernel? */
192		if ( (pmap == kernel_pmap) || (frame == ((int) PTDpde & PG_FRAME)))
193			return ((pt_entry_t *) vtopte(va));
194		/* otherwise, we are alternate address space */
195		else {
196			if ( frame != ((int) APTDpde & PG_FRAME) ) {
197				APTDpde = pmap->pm_pdir[PTDPTDI];
198				tlbflush();
199			}
200			return((pt_entry_t *) avtopte(va));
201		}
202	}
203	return(0);
204}
205
206/*
207 *	Routine:	pmap_extract
208 *	Function:
209 *		Extract the physical page address associated
210 *		with the given map/virtual_address pair.
211 */
212
213vm_offset_t
214pmap_extract(pmap, va)
215	register pmap_t	pmap;
216	vm_offset_t va;
217{
218	pd_entry_t save;
219	vm_offset_t pa;
220	int s;
221
222	if (pmap && *pmap_pde(pmap, va)) {
223		vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
224		/* are we current address space or kernel? */
225		if ( (pmap == kernel_pmap)
226			|| (frame == ((int) PTDpde & PG_FRAME)) ) {
227			pa = *(int *) vtopte(va);
228		/* otherwise, we are alternate address space */
229		} else {
230			if ( frame != ((int) APTDpde & PG_FRAME)) {
231				APTDpde = pmap->pm_pdir[PTDPTDI];
232				tlbflush();
233			}
234			pa = *(int *) avtopte(va);
235		}
236		pa = (pa & PG_FRAME) | (va & ~PG_FRAME);
237		return pa;
238	}
239	return 0;
240
241}
242
243/*
244 * determine if a page is managed (memory vs. device)
245 */
246static inline int
247pmap_is_managed(pa)
248	vm_offset_t pa;
249{
250	int i;
251
252	if (!pmap_initialized)
253		return 0;
254
255	for (i = 0; phys_avail[i + 1]; i += 2) {
256		if (pa >= phys_avail[i] && pa < phys_avail[i + 1])
257			return 1;
258	}
259	return 0;
260}
261
262/*
263 * find the vm_page_t of a pte (only) given va of pte and pmap
264 */
265inline vm_page_t
266pmap_pte_vm_page(pmap, pt)
267	pmap_t pmap;
268	vm_offset_t pt;
269{
270	pt = i386_trunc_page( pt);
271	pt = (pt - UPT_MIN_ADDRESS) / NBPG;
272	pt = ((vm_offset_t) pmap->pm_pdir[pt]) & PG_FRAME;
273	return PHYS_TO_VM_PAGE(pt);
274}
275
276/*
277 * Wire a page table page
278 */
279inline void
280pmap_use_pt(pmap, va)
281	pmap_t pmap;
282	vm_offset_t va;
283{
284	vm_offset_t pt;
285
286	if (va >= VM_MAX_ADDRESS || !pmap_initialized)
287		return;
288
289	pt = (vm_offset_t) vtopte(va);
290	vm_page_hold( pmap_pte_vm_page(pmap, pt));
291}
292
293/*
294 * Unwire a page table page
295 */
296inline void
297pmap_unuse_pt(pmap, va)
298	pmap_t pmap;
299	vm_offset_t va;
300{
301	vm_offset_t pt;
302
303	if (va >= VM_MAX_ADDRESS || !pmap_initialized)
304		return;
305
306	pt = (vm_offset_t) vtopte(va);
307	vm_page_unhold( pmap_pte_vm_page(pmap, pt));
308}
309
310/* [ macro again?, should I force kstack into user map here? -wfj ] */
311void
312pmap_activate(pmap, pcbp)
313	register pmap_t pmap;
314	struct pcb *pcbp;
315{
316	PMAP_ACTIVATE(pmap, pcbp);
317}
318
319/*
320 *	Bootstrap the system enough to run with virtual memory.
321 *	Map the kernel's code and data, and allocate the system page table.
322 *
323 *	On the I386 this is called after mapping has already been enabled
324 *	and just syncs the pmap module with what has already been done.
325 *	[We can't call it easily with mapping off since the kernel is not
326 *	mapped with PA == VA, hence we would have to relocate every address
327 *	from the linked base (virtual) address "KERNBASE" to the actual
328 *	(physical) address starting relative to 0]
329 */
330
331#define DMAPAGES 8
332void
333pmap_bootstrap(firstaddr, loadaddr)
334	vm_offset_t firstaddr;
335	vm_offset_t loadaddr;
336{
337#if BSDVM_COMPAT
338	vm_offset_t va;
339	pt_entry_t *pte;
340#endif
341	extern int IdlePTD;
342
343	avail_start = firstaddr + DMAPAGES*NBPG;
344
345	virtual_avail = (vm_offset_t) KERNBASE + avail_start;
346	virtual_end = VM_MAX_KERNEL_ADDRESS;
347	i386pagesperpage = PAGE_SIZE / NBPG;
348
349	/*
350	 * Initialize protection array.
351	 */
352	i386_protection_init();
353
354	/*
355	 * The kernel's pmap is statically allocated so we don't
356	 * have to use pmap_create, which is unlikely to work
357	 * correctly at this part of the boot sequence.
358	 */
359	kernel_pmap = &kernel_pmap_store;
360
361	kernel_pmap->pm_pdir = (pd_entry_t *)(KERNBASE + IdlePTD);
362
363	simple_lock_init(&kernel_pmap->pm_lock);
364	kernel_pmap->pm_count = 1;
365
366#if BSDVM_COMPAT
367	/*
368	 * Allocate all the submaps we need
369	 */
370#define	SYSMAP(c, p, v, n)	\
371	v = (c)va; va += ((n)*NBPG); p = pte; pte += (n);
372
373	va = virtual_avail;
374	pte = pmap_pte(kernel_pmap, va);
375
376	SYSMAP(caddr_t		,CMAP1		,CADDR1	   ,1		)
377	SYSMAP(caddr_t		,CMAP2		,CADDR2	   ,1		)
378	SYSMAP(caddr_t		,ptmmap		,ptvmmap	   ,1		)
379	SYSMAP(struct msgbuf *	,msgbufmap	,msgbufp   ,1		)
380	virtual_avail = va;
381#endif
382	/*
383	 * reserve special hunk of memory for use by bus dma as a bounce
384	 * buffer (contiguous virtual *and* physical memory). for now,
385	 * assume vm does not use memory beneath hole, and we know that
386	 * the bootstrap uses top 32k of base memory. -wfj
387	 */
388	{
389		extern vm_offset_t isaphysmem;
390		isaphysmem = va;
391
392		virtual_avail = pmap_map(va, firstaddr,
393				firstaddr + DMAPAGES*NBPG, VM_PROT_ALL);
394	}
395
396	*(int *)CMAP1 = *(int *)CMAP2 = *(int *)PTD = 0;
397	tlbflush();
398
399}
400
401/*
402 *	Initialize the pmap module.
403 *	Called by vm_init, to initialize any structures that the pmap
404 *	system needs to map virtual memory.
405 *	pmap_init has been enhanced to support in a fairly consistant
406 *	way, discontiguous physical memory.
407 */
408void
409pmap_init(phys_start, phys_end)
410	vm_offset_t	phys_start, phys_end;
411{
412	vm_offset_t	addr, addr2;
413	vm_size_t	npg, s;
414	int		rv;
415	int i;
416	extern int KPTphys;
417	extern int IdlePTD;
418
419	/*
420	 * Now that kernel map has been allocated, we can mark as
421	 * unavailable regions which we have mapped in locore.
422	 */
423	addr = atdevbase;
424	(void) vm_map_find(kernel_map, NULL, (vm_offset_t) 0,
425			   &addr, (0x100000-0xa0000), FALSE);
426
427	addr = (vm_offset_t) KERNBASE + IdlePTD;
428	vm_object_reference(kernel_object);
429	(void) vm_map_find(kernel_map, kernel_object, addr,
430			   &addr, (4 + NKPT) * NBPG, FALSE);
431
432
433	/*
434	 * calculate the number of pv_entries needed
435	 */
436	vm_first_phys = phys_avail[0];
437	for (i = 0; phys_avail[i + 1]; i += 2) ;
438	npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / NBPG;
439
440	/*
441	 * Allocate memory for random pmap data structures.  Includes the
442	 * pv_head_table.
443	 */
444	s = (vm_size_t) (sizeof(struct pv_entry) * npg);
445	s = i386_round_page(s);
446	addr = (vm_offset_t) kmem_alloc(kernel_map, s);
447	pv_table = (pv_entry_t) addr;
448
449	/*
450	 * init the pv free list
451	 */
452	init_pv_entries(npg);
453	/*
454	 * Now it is safe to enable pv_table recording.
455	 */
456	pmap_initialized = TRUE;
457}
458
459/*
460 *	Used to map a range of physical addresses into kernel
461 *	virtual address space.
462 *
463 *	For now, VM is already on, we only need to map the
464 *	specified memory.
465 */
466vm_offset_t
467pmap_map(virt, start, end, prot)
468	vm_offset_t	virt;
469	vm_offset_t	start;
470	vm_offset_t	end;
471	int		prot;
472{
473	while (start < end) {
474		pmap_enter(kernel_pmap, virt, start, prot, FALSE);
475		virt += PAGE_SIZE;
476		start += PAGE_SIZE;
477	}
478	return(virt);
479}
480
481/*
482 *	Create and return a physical map.
483 *
484 *	If the size specified for the map
485 *	is zero, the map is an actual physical
486 *	map, and may be referenced by the
487 *	hardware.
488 *
489 *	If the size specified is non-zero,
490 *	the map will be used in software only, and
491 *	is bounded by that size.
492 *
493 * [ just allocate a ptd and mark it uninitialize -- should we track
494 *   with a table which process has which ptd? -wfj ]
495 */
496
497pmap_t
498pmap_create(size)
499	vm_size_t	size;
500{
501	register pmap_t pmap;
502
503	/*
504	 * Software use map does not need a pmap
505	 */
506	if (size)
507		return(NULL);
508
509	pmap = (pmap_t) malloc(sizeof *pmap, M_VMPMAP, M_WAITOK);
510	bzero(pmap, sizeof(*pmap));
511	pmap_pinit(pmap);
512	return (pmap);
513}
514
515
516struct pmaplist {
517	struct pmaplist *next;
518};
519
520static inline void *
521vm_get_pmap()
522{
523	struct pmaplist *rtval;
524
525	rtval = (struct pmaplist *)kmem_alloc(kernel_map, ctob(1));
526	bzero(rtval, ctob(1));
527	return rtval;
528}
529
530static inline void
531vm_put_pmap(up)
532	struct pmaplist *up;
533{
534	kmem_free(kernel_map, (vm_offset_t)up, ctob(1));
535}
536
537/*
538 * Initialize a preallocated and zeroed pmap structure,
539 * such as one in a vmspace structure.
540 */
541void
542pmap_pinit(pmap)
543	register struct pmap *pmap;
544{
545	/*
546	 * No need to allocate page table space yet but we do need a
547	 * valid page directory table.
548	 */
549	pmap->pm_pdir = (pd_entry_t *) vm_get_pmap();
550
551	/* wire in kernel global address entries */
552	bcopy(PTD+KPTDI, pmap->pm_pdir+KPTDI, NKPT*PTESIZE);
553
554	/* install self-referential address mapping entry */
555	*(int *)(pmap->pm_pdir+PTDPTDI) =
556		((int)pmap_kextract((vm_offset_t)pmap->pm_pdir)) | PG_V | PG_KW;
557
558	pmap->pm_count = 1;
559	simple_lock_init(&pmap->pm_lock);
560}
561
562/*
563 *	Retire the given physical map from service.
564 *	Should only be called if the map contains
565 *	no valid mappings.
566 */
567void
568pmap_destroy(pmap)
569	register pmap_t pmap;
570{
571	int count;
572
573	if (pmap == NULL)
574		return;
575
576	simple_lock(&pmap->pm_lock);
577	count = --pmap->pm_count;
578	simple_unlock(&pmap->pm_lock);
579	if (count == 0) {
580		pmap_release(pmap);
581		free((caddr_t)pmap, M_VMPMAP);
582	}
583}
584
585/*
586 * Release any resources held by the given physical map.
587 * Called when a pmap initialized by pmap_pinit is being released.
588 * Should only be called if the map contains no valid mappings.
589 */
590void
591pmap_release(pmap)
592	register struct pmap *pmap;
593{
594	vm_put_pmap((struct pmaplist *) pmap->pm_pdir);
595}
596
597/*
598 *	Add a reference to the specified pmap.
599 */
600void
601pmap_reference(pmap)
602	pmap_t	pmap;
603{
604	if (pmap != NULL) {
605		simple_lock(&pmap->pm_lock);
606		pmap->pm_count++;
607		simple_unlock(&pmap->pm_lock);
608	}
609}
610
611#define PV_FREELIST_MIN ((NBPG / sizeof (struct pv_entry)) / 2)
612
613/*
614 * Data for the pv entry allocation mechanism
615 */
616int pv_freelistcnt;
617pv_entry_t pv_freelist;
618vm_offset_t pvva;
619int npvvapg;
620
621/*
622 * free the pv_entry back to the free list
623 */
624inline static void
625free_pv_entry(pv)
626	pv_entry_t pv;
627{
628	if (!pv) return;
629	++pv_freelistcnt;
630	pv->pv_next = pv_freelist;
631	pv_freelist = pv;
632}
633
634/*
635 * get a new pv_entry, allocating a block from the system
636 * when needed.
637 * the memory allocation is performed bypassing the malloc code
638 * because of the possibility of allocations at interrupt time.
639 */
640static inline pv_entry_t
641get_pv_entry()
642{
643	pv_entry_t tmp;
644
645	/*
646	 * get more pv_entry pages if needed
647	 */
648	while (pv_freelistcnt < PV_FREELIST_MIN || pv_freelist == 0) {
649		pmap_alloc_pv_entry();
650	}
651
652	/*
653	 * get a pv_entry off of the free list
654	 */
655	--pv_freelistcnt;
656	tmp = pv_freelist;
657	pv_freelist = tmp->pv_next;
658	tmp->pv_pmap = 0;
659	tmp->pv_va = 0;
660	tmp->pv_next = 0;
661	return tmp;
662}
663
664/*
665 * this *strange* allocation routine *statistically* eliminates the
666 * *possibility* of a malloc failure (*FATAL*) for a pv_entry_t data structure.
667 * also -- this code is MUCH MUCH faster than the malloc equiv...
668 */
669static void
670pmap_alloc_pv_entry()
671{
672	/*
673	 * do we have any pre-allocated map-pages left?
674	 */
675	if (npvvapg) {
676		vm_page_t m;
677		/*
678		 * we do this to keep recursion away
679		 */
680		pv_freelistcnt += PV_FREELIST_MIN;
681		/*
682		 * allocate a physical page out of the vm system
683		 */
684		if (m = vm_page_alloc(kernel_object, pvva-vm_map_min(kernel_map))) {
685			int newentries;
686			int i;
687			pv_entry_t entry;
688			newentries = (NBPG/sizeof (struct pv_entry));
689			/*
690			 * wire the page
691			 */
692			vm_page_wire(m);
693			m->flags &= ~PG_BUSY;
694			/*
695			 * let the kernel see it
696			 */
697			pmap_kenter(pvva, VM_PAGE_TO_PHYS(m));
698
699			entry = (pv_entry_t) pvva;
700			/*
701			 * update the allocation pointers
702			 */
703			pvva += NBPG;
704			--npvvapg;
705
706			/*
707			 * free the entries into the free list
708			 */
709			for (i = 0; i < newentries; i++) {
710				free_pv_entry(entry);
711				entry++;
712			}
713		}
714		pv_freelistcnt -= PV_FREELIST_MIN;
715	}
716	if (!pv_freelist)
717		panic("get_pv_entry: cannot get a pv_entry_t");
718}
719
720
721
722/*
723 * init the pv_entry allocation system
724 */
725#define PVSPERPAGE 64
726void
727init_pv_entries(npg)
728	int npg;
729{
730	/*
731	 * allocate enough kvm space for PVSPERPAGE entries per page (lots)
732	 * kvm space is fairly cheap, be generous!!!  (the system can panic
733	 * if this is too small.)
734	 */
735	npvvapg = ((npg*PVSPERPAGE) * sizeof(struct pv_entry) + NBPG - 1)/NBPG;
736	pvva = kmem_alloc_pageable(kernel_map, npvvapg * NBPG);
737	/*
738	 * get the first batch of entries
739	 */
740	free_pv_entry(get_pv_entry());
741}
742
743static pt_entry_t *
744get_pt_entry(pmap)
745	pmap_t pmap;
746{
747	pt_entry_t *ptp;
748	vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
749	/* are we current address space or kernel? */
750	if (pmap == kernel_pmap || frame == ((int) PTDpde & PG_FRAME)) {
751		ptp=PTmap;
752	/* otherwise, we are alternate address space */
753	} else {
754		if ( frame != ((int) APTDpde & PG_FRAME)) {
755			APTDpde = pmap->pm_pdir[PTDPTDI];
756			tlbflush();
757		}
758		ptp=APTmap;
759	     }
760	return ptp;
761}
762
763/*
764 * If it is the first entry on the list, it is actually
765 * in the header and we must copy the following entry up
766 * to the header.  Otherwise we must search the list for
767 * the entry.  In either case we free the now unused entry.
768 */
769void
770pmap_remove_entry(pmap, pv, va)
771	struct pmap *pmap;
772	pv_entry_t pv;
773	vm_offset_t va;
774{
775	pv_entry_t npv;
776	int wired;
777	int s;
778	s = splhigh();
779	if (pmap == pv->pv_pmap && va == pv->pv_va) {
780		npv = pv->pv_next;
781		if (npv) {
782			*pv = *npv;
783			free_pv_entry(npv);
784		} else {
785			pv->pv_pmap = NULL;
786		}
787	} else {
788		for (npv = pv->pv_next; npv; npv = npv->pv_next) {
789			if (pmap == npv->pv_pmap && va == npv->pv_va) {
790				break;
791			}
792			pv = npv;
793		}
794		if (npv) {
795			pv->pv_next = npv->pv_next;
796			free_pv_entry(npv);
797		}
798	}
799	splx(s);
800}
801
802/*
803 *	Remove the given range of addresses from the specified map.
804 *
805 *	It is assumed that the start and end are properly
806 *	rounded to the page size.
807 */
808void
809pmap_remove(pmap, sva, eva)
810	struct pmap *pmap;
811	register vm_offset_t sva;
812	register vm_offset_t eva;
813{
814	register pt_entry_t *ptp,*ptq;
815	vm_offset_t pa;
816	register pv_entry_t pv;
817	vm_offset_t va;
818	vm_page_t m;
819	pt_entry_t oldpte;
820
821	if (pmap == NULL)
822		return;
823
824	ptp = get_pt_entry(pmap);
825
826/*
827 * special handling of removing one page.  a very
828 * common operation and easy to short circuit some
829 * code.
830 */
831	if( (sva + NBPG) == eva) {
832
833		if( *pmap_pde( pmap, sva) == 0)
834			return;
835
836		ptq = ptp + i386_btop(sva);
837
838		if( !*ptq)
839			return;
840		/*
841		 * Update statistics
842		 */
843		if (pmap_pte_w(ptq))
844			pmap->pm_stats.wired_count--;
845		pmap->pm_stats.resident_count--;
846
847		pa = pmap_pte_pa(ptq);
848		oldpte = *ptq;
849		*ptq = 0;
850
851		if (pmap_is_managed(pa)) {
852			if ((int) oldpte & (PG_M | PG_U)) {
853				if ((sva < USRSTACK || sva > UPT_MAX_ADDRESS) ||
854				    (sva >= USRSTACK && sva < USRSTACK+(UPAGES*NBPG))) {
855					if (sva < clean_sva || sva >= clean_eva) {
856						m = PHYS_TO_VM_PAGE(pa);
857						if ((int) oldpte & PG_M) {
858							m->flags &= ~PG_CLEAN;
859						}
860						if ((int) oldpte & PG_U) {
861							m->flags |= PG_REFERENCED;
862						}
863					}
864				}
865			}
866
867			pv = pa_to_pvh(pa);
868			pmap_remove_entry(pmap, pv, sva);
869			pmap_unuse_pt(pmap, sva);
870		}
871		tlbflush();
872		return;
873	}
874
875	sva = i386_btop(sva);
876	eva = i386_btop(eva);
877
878	while (sva < eva) {
879		/*
880		 * Weed out invalid mappings.
881		 * Note: we assume that the page directory table is
882	 	 * always allocated, and in kernel virtual.
883		 */
884
885		if ( *pmap_pde(pmap, i386_ptob(sva)) == 0 ) {
886			/* We can race ahead here, straight to next pde.. */
887	nextpde:
888			sva = ((sva + NPTEPG) & ~(NPTEPG - 1));
889			continue;
890		}
891
892		ptq = ptp + sva;
893
894		/*
895		 * search for page table entries, use string operations
896		 * that are much faster than
897		 * explicitly scanning when page tables are not fully
898		 * populated.
899		 */
900		if ( *ptq == 0) {
901			vm_offset_t pdnxt = ((sva + NPTEPG) & ~(NPTEPG - 1));
902			vm_offset_t nscan = pdnxt - sva;
903			int found = 0;
904
905			if ((nscan + sva) > eva)
906				nscan = eva - sva;
907
908			asm("xorl %%eax,%%eax;cld;repe;scasl;jz 1f;incl %%eax;1:;"
909				:"=D"(ptq),"=a"(found)
910				:"c"(nscan),"0"(ptq)
911				:"cx");
912
913			if( !found) {
914				sva = pdnxt;
915				continue;
916			}
917			ptq -= 1;
918
919			sva = ptq - ptp;
920		}
921
922		/*
923		 * Update statistics
924		 */
925		oldpte = *ptq;
926		if (((int)oldpte) & PG_W)
927			pmap->pm_stats.wired_count--;
928		pmap->pm_stats.resident_count--;
929
930		/*
931		 * Invalidate the PTEs.
932		 * XXX: should cluster them up and invalidate as many
933		 * as possible at once.
934		 */
935		*ptq = 0;
936
937		va = i386_ptob(sva);
938
939		/*
940		 * Remove from the PV table (raise IPL since we
941		 * may be called at interrupt time).
942		 */
943		pa = ((int)oldpte) & PG_FRAME;
944		if (!pmap_is_managed(pa)) {
945			++sva;
946			continue;
947		}
948
949		if ((((int) oldpte & PG_M) && (va < USRSTACK || va > UPT_MAX_ADDRESS))
950			|| (va >= USRSTACK && va < USRSTACK+(UPAGES*NBPG))) {
951			if (va < clean_sva || va >= clean_eva ) {
952				m = PHYS_TO_VM_PAGE(pa);
953				m->flags &= ~PG_CLEAN;
954			}
955		}
956
957		pv = pa_to_pvh(pa);
958		pmap_remove_entry(pmap, pv, va);
959		pmap_unuse_pt(pmap, va);
960		++sva;
961	}
962	tlbflush();
963}
964
965/*
966 *	Routine:	pmap_remove_all
967 *	Function:
968 *		Removes this physical page from
969 *		all physical maps in which it resides.
970 *		Reflects back modify bits to the pager.
971 *
972 *	Notes:
973 *		Original versions of this routine were very
974 *		inefficient because they iteratively called
975 *		pmap_remove (slow...)
976 */
977void
978pmap_remove_all(pa)
979	vm_offset_t pa;
980{
981	register pv_entry_t pv, npv;
982	register pt_entry_t *pte, *ptp;
983	vm_offset_t va;
984	struct pmap *pmap;
985	struct map *map;
986	vm_page_t m;
987	int s;
988	int anyvalid = 0;
989
990	/*
991	 * Not one of ours
992	 */
993	if (!pmap_is_managed(pa))
994		return;
995
996	pa = i386_trunc_page(pa);
997	pv = pa_to_pvh(pa);
998	m = PHYS_TO_VM_PAGE(pa);
999
1000	s = splhigh();
1001	while (pv->pv_pmap != NULL) {
1002		pmap = pv->pv_pmap;
1003		ptp = get_pt_entry(pmap);
1004		va = i386_btop(pv->pv_va);
1005		pte = ptp + va;
1006		if (pmap_pte_w(pte))
1007			pmap->pm_stats.wired_count--;
1008		if ( *pte) {
1009			pmap->pm_stats.resident_count--;
1010			anyvalid++;
1011
1012			/*
1013			 * update the vm_page_t clean bit
1014			 */
1015			if ( (m->flags & PG_CLEAN) &&
1016				((((int) *pte) & PG_M) && (pv->pv_va < USRSTACK || pv->pv_va > UPT_MAX_ADDRESS))
1017				|| (pv->pv_va >= USRSTACK && pv->pv_va < USRSTACK+(UPAGES*NBPG))) {
1018				if (pv->pv_va < clean_sva || pv->pv_va >= clean_eva) {
1019					m->flags &= ~PG_CLEAN;
1020				}
1021			}
1022
1023			*pte = 0;
1024		}
1025		pmap_unuse_pt(pmap, pv->pv_va);
1026
1027		npv = pv->pv_next;
1028		if (npv) {
1029			*pv = *npv;
1030			free_pv_entry(npv);
1031		} else {
1032			pv->pv_pmap = NULL;
1033		}
1034	}
1035	splx(s);
1036	if (anyvalid)
1037		tlbflush();
1038}
1039
1040
1041/*
1042 *	Set the physical protection on the
1043 *	specified range of this map as requested.
1044 */
1045void
1046pmap_protect(pmap, sva, eva, prot)
1047	register pmap_t	pmap;
1048	vm_offset_t	sva, eva;
1049	vm_prot_t	prot;
1050{
1051	register pt_entry_t *pte;
1052	register vm_offset_t va;
1053	int i386prot;
1054	register pt_entry_t *ptp;
1055	int evap = i386_btop(eva);
1056	int s;
1057	int anyvalid = 0;;
1058
1059	if (pmap == NULL)
1060		return;
1061
1062	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1063		pmap_remove(pmap, sva, eva);
1064		return;
1065	}
1066	if (prot & VM_PROT_WRITE)
1067		return;
1068
1069	ptp = get_pt_entry(pmap);
1070
1071	va = sva;
1072	while (va < eva) {
1073		int found=0;
1074		int svap;
1075		vm_offset_t nscan;
1076		/*
1077		 * Page table page is not allocated.
1078		 * Skip it, we don't want to force allocation
1079		 * of unnecessary PTE pages just to set the protection.
1080		 */
1081		if (! *pmap_pde(pmap, va)) {
1082			/* XXX: avoid address wrap around */
1083nextpde:
1084			if (va >= i386_trunc_pdr((vm_offset_t)-1))
1085				break;
1086			va = i386_round_pdr(va + PAGE_SIZE);
1087			continue;
1088		}
1089
1090		pte = ptp + i386_btop(va);
1091
1092		if( *pte == 0) {
1093		/*
1094		 * scan for a non-empty pte
1095		 */
1096			svap = pte - ptp;
1097			nscan = ((svap + NPTEPG) & ~(NPTEPG - 1)) - svap;
1098
1099			if (nscan + svap > evap)
1100				nscan = evap - svap;
1101
1102			found = 0;
1103			if (nscan)
1104				asm("xorl %%eax,%%eax;cld;repe;scasl;jz 1f;incl %%eax;1:;"
1105					:"=D"(pte),"=a"(found)
1106					:"c"(nscan),"0"(pte):"cx");
1107
1108			if( !found)
1109				goto nextpde;
1110
1111			pte -= 1;
1112			svap = pte - ptp;
1113
1114			va = i386_ptob(svap);
1115		}
1116
1117		anyvalid++;
1118
1119		i386prot = pte_prot(pmap, prot);
1120		if (va < UPT_MAX_ADDRESS) {
1121			i386prot |= PG_u;
1122			if( va >= UPT_MIN_ADDRESS)
1123				i386prot |= PG_RW;
1124		}
1125		pmap_pte_set_prot(pte, i386prot);
1126		va += PAGE_SIZE;
1127	}
1128	if (anyvalid)
1129		tlbflush();
1130}
1131
1132/*
1133 *	Insert the given physical page (p) at
1134 *	the specified virtual address (v) in the
1135 *	target physical map with the protection requested.
1136 *
1137 *	If specified, the page will be wired down, meaning
1138 *	that the related pte can not be reclaimed.
1139 *
1140 *	NB:  This is the only routine which MAY NOT lazy-evaluate
1141 *	or lose information.  That is, this routine must actually
1142 *	insert this page into the given map NOW.
1143 */
1144void
1145pmap_enter(pmap, va, pa, prot, wired)
1146	register pmap_t pmap;
1147	vm_offset_t va;
1148	register vm_offset_t pa;
1149	vm_prot_t prot;
1150	boolean_t wired;
1151{
1152	register pt_entry_t *pte;
1153	register pt_entry_t npte;
1154	vm_offset_t opa;
1155	int cacheable=1;
1156	int ptevalid = 0;
1157
1158	if (pmap == NULL)
1159		return;
1160
1161	va = i386_trunc_page(va);
1162	pa = i386_trunc_page(pa);
1163	if (va > VM_MAX_KERNEL_ADDRESS)panic("pmap_enter: toobig");
1164
1165	/*
1166	 * Page Directory table entry not valid, we need a new PT page
1167	 */
1168	if ( *pmap_pde(pmap, va) == 0) {
1169		pg("ptdi %x, va %x", pmap->pm_pdir[PTDPTDI], va);
1170	}
1171
1172	pte = pmap_pte(pmap, va);
1173	opa = pmap_pte_pa(pte);
1174
1175	/*
1176	 * Mapping has not changed, must be protection or wiring change.
1177	 */
1178	if (opa == pa) {
1179		/*
1180		 * Wiring change, just update stats.
1181		 * We don't worry about wiring PT pages as they remain
1182		 * resident as long as there are valid mappings in them.
1183		 * Hence, if a user page is wired, the PT page will be also.
1184		 */
1185		if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
1186			if (wired)
1187				pmap->pm_stats.wired_count++;
1188			else
1189				pmap->pm_stats.wired_count--;
1190		}
1191		goto validate;
1192	}
1193
1194	/*
1195	 * Mapping has changed, invalidate old range and fall through to
1196	 * handle validating new mapping.
1197	 */
1198	if (opa) {
1199		pmap_remove(pmap, va, va + PAGE_SIZE);
1200	}
1201
1202	/*
1203	 * Enter on the PV list if part of our managed memory
1204	 * Note that we raise IPL while manipulating pv_table
1205	 * since pmap_enter can be called at interrupt time.
1206	 */
1207	if (pmap_is_managed(pa)) {
1208		register pv_entry_t pv, npv;
1209		int s;
1210
1211		pv = pa_to_pvh(pa);
1212		s = splhigh();
1213		/*
1214		 * No entries yet, use header as the first entry
1215		 */
1216		if (pv->pv_pmap == NULL) {
1217			pv->pv_va = va;
1218			pv->pv_pmap = pmap;
1219			pv->pv_next = NULL;
1220		}
1221		/*
1222		 * There is at least one other VA mapping this page.
1223		 * Place this entry after the header.
1224		 */
1225		else {
1226			npv = get_pv_entry();
1227			npv->pv_va = va;
1228			npv->pv_pmap = pmap;
1229			npv->pv_next = pv->pv_next;
1230			pv->pv_next = npv;
1231		}
1232		splx(s);
1233		cacheable = 1;
1234	} else {
1235		cacheable = 0;
1236	}
1237
1238	pmap_use_pt(pmap, va);
1239
1240	/*
1241	 * Increment counters
1242	 */
1243	pmap->pm_stats.resident_count++;
1244	if (wired)
1245		pmap->pm_stats.wired_count++;
1246
1247validate:
1248	/*
1249	 * Now validate mapping with desired protection/wiring.
1250	 */
1251	npte = (pt_entry_t) ( (int) (pa | pte_prot(pmap, prot) | PG_V));
1252	/*
1253	 * for correctness:
1254	 */
1255	if( !cacheable)
1256		(int) npte |= PG_N;
1257
1258	/*
1259	 * When forking (copy-on-write, etc):
1260	 * A process will turn off write permissions for any of its writable
1261	 * pages.  If the data (object) is only referred to by one process, the
1262	 * processes map is modified directly as opposed to using the
1263	 * object manipulation routine.  When using pmap_protect, the
1264	 * modified bits are not kept in the vm_page_t data structure.
1265	 * Therefore, when using pmap_enter in vm_fault to bring back
1266	 * writability of a page, there has been no memory of the
1267	 * modified or referenced bits except at the pte level.
1268	 * this clause supports the carryover of the modified and
1269	 * used (referenced) bits.
1270	 */
1271	if (pa == opa)
1272		(int) npte |= (int) *pte & (PG_M|PG_U);
1273
1274
1275	if (wired)
1276		(int) npte |= PG_W;
1277	if (va < UPT_MIN_ADDRESS)
1278		(int) npte |= PG_u;
1279	else if (va < UPT_MAX_ADDRESS)
1280		(int) npte |= PG_u | PG_RW;
1281
1282	if(*pte != npte) {
1283		if (*pte)
1284			ptevalid++;
1285		*pte = npte;
1286	}
1287	if (ptevalid)
1288		tlbflush();
1289}
1290
1291/*
1292 * Add a list of wired pages to the kva
1293 * this routine is only used for temporary
1294 * kernel mappings that do not need to have
1295 * page modification or references recorded.
1296 * Note that old mappings are simply written
1297 * over.  The page *must* be wired.
1298 */
1299void
1300pmap_qenter(va, m, count)
1301	vm_offset_t va;
1302	vm_page_t *m;
1303	int count;
1304{
1305	int i;
1306	int anyvalid = 0;
1307	register pt_entry_t *pte;
1308
1309	for(i=0;i<count;i++) {
1310		pte = vtopte(va + i * NBPG);
1311		if (*pte)
1312			anyvalid++;
1313		*pte = (pt_entry_t) ( (int) (VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | PG_W));
1314	}
1315	if (anyvalid)
1316		tlbflush();
1317}
1318/*
1319 * this routine jerks page mappings from the
1320 * kernel -- it is meant only for temporary mappings.
1321 */
1322void
1323pmap_qremove(va, count)
1324	vm_offset_t va;
1325	int count;
1326{
1327	int i;
1328	register pt_entry_t *pte;
1329	for(i=0;i<count;i++) {
1330		pte = vtopte(va + i * NBPG);
1331		*pte = 0;
1332	}
1333	tlbflush();
1334}
1335
1336/*
1337 * add a wired page to the kva
1338 * note that in order for the mapping to take effect -- you
1339 * should do a tlbflush after doing the pmap_kenter...
1340 */
1341void
1342pmap_kenter(va, pa)
1343	vm_offset_t va;
1344	register vm_offset_t pa;
1345{
1346	register pt_entry_t *pte;
1347	pte = vtopte(va);
1348
1349	*pte = (pt_entry_t) ( (int) (pa | PG_RW | PG_V | PG_W));
1350}
1351
1352/*
1353 * remove a page from the kernel pagetables
1354 */
1355void
1356pmap_kremove( va)
1357	vm_offset_t va;
1358{
1359	register pt_entry_t *pte;
1360	pte = vtopte(va);
1361
1362	*pte = (pt_entry_t) 0;
1363	tlbflush();
1364}
1365
1366/*
1367 * this code makes some *MAJOR* assumptions:
1368 * 1. Current pmap & pmap exists.
1369 * 2. Not wired.
1370 * 3. Read access.
1371 * 4. No page table pages.
1372 * 5. Tlbflush is deferred to calling procedure.
1373 * 6. Page IS managed.
1374 * but is *MUCH* faster than pmap_enter...
1375 */
1376
1377static inline int
1378pmap_enter_quick(pmap, va, pa)
1379	register pmap_t pmap;
1380	vm_offset_t va;
1381	register vm_offset_t pa;
1382{
1383	register pt_entry_t *pte;
1384	register pv_entry_t pv, npv;
1385	int s;
1386	int anyvalid = 0;
1387
1388	/*
1389	 * Enter on the PV list if part of our managed memory
1390	 * Note that we raise IPL while manipulating pv_table
1391	 * since pmap_enter can be called at interrupt time.
1392	 */
1393
1394	pte = vtopte(va);
1395	if (pmap_pte_pa(pte)) {
1396		pmap_remove(pmap, va, va + PAGE_SIZE);
1397	}
1398
1399	pv = pa_to_pvh(pa);
1400	s = splhigh();
1401	/*
1402	 * No entries yet, use header as the first entry
1403	 */
1404	if (pv->pv_pmap == NULL) {
1405		pv->pv_va = va;
1406		pv->pv_pmap = pmap;
1407		pv->pv_next = NULL;
1408	}
1409	/*
1410	 * There is at least one other VA mapping this page.
1411	 * Place this entry after the header.
1412	 */
1413	else {
1414		npv = get_pv_entry();
1415		npv->pv_va = va;
1416		npv->pv_pmap = pmap;
1417		npv->pv_next = pv->pv_next;
1418		pv->pv_next = npv;
1419	}
1420	splx(s);
1421
1422	pmap_use_pt(pmap, va);
1423
1424	/*
1425	 * Increment counters
1426	 */
1427	pmap->pm_stats.resident_count++;
1428
1429validate:
1430
1431	if (*pte)
1432		anyvalid++;
1433	/*
1434	 * Now validate mapping with desired protection/wiring.
1435	 */
1436	*pte = (pt_entry_t) ( (int) (pa | PG_V | PG_u));
1437
1438	return (anyvalid);
1439}
1440
1441/*
1442 * pmap_object_init_pt preloads the ptes for a given object
1443 * into the specified pmap.  This eliminates the blast of soft
1444 * faults on process startup and immediately after an mmap.
1445 */
1446void
1447pmap_object_init_pt(pmap, addr, object, offset, size)
1448	pmap_t pmap;
1449	vm_offset_t addr;
1450	vm_object_t object;
1451	vm_offset_t offset;
1452	vm_offset_t size;
1453{
1454
1455	vm_offset_t tmpoff;
1456	vm_page_t p;
1457	int s;
1458	vm_offset_t v, lastv=0;
1459	pt_entry_t pte;
1460	extern vm_map_t kernel_map;
1461	vm_offset_t objbytes;
1462	int anyvalid = 0;
1463
1464	if (!pmap)
1465		return;
1466
1467	/*
1468	 * if we are processing a major portion of the object, then
1469	 * scan the entire thing.
1470	 */
1471	if( size > object->size / 2) {
1472		objbytes = size;
1473		p = object->memq.tqh_first;
1474		while ((p != NULL) && (objbytes != 0)) {
1475			tmpoff = p->offset;
1476			if( tmpoff < offset) {
1477				p = p->listq.tqe_next;
1478				continue;
1479			}
1480			tmpoff -= offset;
1481			if( tmpoff >= size) {
1482				p = p->listq.tqe_next;
1483				continue;
1484			}
1485
1486			if ((p->flags & (PG_BUSY|PG_FICTITIOUS)) == 0 ) {
1487				vm_page_hold(p);
1488				v = i386_trunc_page(((vm_offset_t)vtopte( addr+tmpoff)));
1489				/* a fault might occur here */
1490				*(volatile char *)v += 0;
1491				vm_page_unhold(p);
1492				anyvalid += pmap_enter_quick(pmap, addr+tmpoff, VM_PAGE_TO_PHYS(p));
1493			}
1494			p = p->listq.tqe_next;
1495			objbytes -= NBPG;
1496		}
1497	} else {
1498	/*
1499	 * else lookup the pages one-by-one.
1500	 */
1501		for(tmpoff = 0; tmpoff < size; tmpoff += NBPG) {
1502			if( p = vm_page_lookup(object, tmpoff + offset)) {
1503				if( (p->flags & (PG_BUSY|PG_FICTITIOUS)) == 0) {
1504					vm_page_hold(p);
1505					v = i386_trunc_page(((vm_offset_t)vtopte( addr+tmpoff)));
1506					/* a fault might occur here */
1507					*(volatile char *)v += 0;
1508					vm_page_unhold(p);
1509					anyvalid += pmap_enter_quick(pmap, addr+tmpoff, VM_PAGE_TO_PHYS(p));
1510				}
1511			}
1512		}
1513	}
1514
1515	if (anyvalid)
1516		tlbflush();
1517}
1518
1519/*
1520 *	Routine:	pmap_change_wiring
1521 *	Function:	Change the wiring attribute for a map/virtual-address
1522 *			pair.
1523 *	In/out conditions:
1524 *			The mapping must already exist in the pmap.
1525 */
1526void
1527pmap_change_wiring(pmap, va, wired)
1528	register pmap_t	pmap;
1529	vm_offset_t	va;
1530	boolean_t	wired;
1531{
1532	register pt_entry_t *pte;
1533
1534	if (pmap == NULL)
1535		return;
1536
1537	pte = pmap_pte(pmap, va);
1538	if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
1539		if (wired)
1540			pmap->pm_stats.wired_count++;
1541		else
1542			pmap->pm_stats.wired_count--;
1543	}
1544	/*
1545	 * Wiring is not a hardware characteristic so there is no need
1546	 * to invalidate TLB.
1547	 */
1548	pmap_pte_set_w(pte, wired);
1549	/*
1550 	 * When unwiring, set the modified bit in the pte -- could have
1551	 * been changed by the kernel
1552 	 */
1553	if (!wired)
1554		(int) *pte |= PG_M;
1555}
1556
1557
1558
1559/*
1560 *	Copy the range specified by src_addr/len
1561 *	from the source map to the range dst_addr/len
1562 *	in the destination map.
1563 *
1564 *	This routine is only advisory and need not do anything.
1565 */
1566void
1567pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
1568	pmap_t dst_pmap, src_pmap;
1569	vm_offset_t	dst_addr;
1570	vm_size_t	len;
1571	vm_offset_t	src_addr;
1572{
1573}
1574/*
1575 *	Require that all active physical maps contain no
1576 *	incorrect entries NOW.  [This update includes
1577 *	forcing updates of any address map caching.]
1578 *
1579 *	Generally used to insure that a thread about
1580 *	to run will see a semantically correct world.
1581 */
1582void
1583pmap_update()
1584{
1585	tlbflush();
1586}
1587
1588/*
1589 *	Routine:	pmap_kernel
1590 *	Function:
1591 *		Returns the physical map handle for the kernel.
1592 */
1593pmap_t
1594pmap_kernel()
1595{
1596    	return (kernel_pmap);
1597}
1598
1599/*
1600 *	pmap_zero_page zeros the specified (machine independent)
1601 *	page by mapping the page into virtual memory and using
1602 *	bzero to clear its contents, one machine dependent page
1603 *	at a time.
1604 */
1605void
1606pmap_zero_page(phys)
1607	vm_offset_t phys;
1608{
1609	if (*(int *)CMAP2)
1610		panic("pmap_zero_page: CMAP busy");
1611
1612	*(int *)CMAP2 = PG_V | PG_KW | i386_trunc_page(phys);
1613	bzero(CADDR2,NBPG);
1614
1615	*(int *)CMAP2 = 0;
1616	tlbflush();
1617}
1618
1619/*
1620 *	pmap_copy_page copies the specified (machine independent)
1621 *	page by mapping the page into virtual memory and using
1622 *	bcopy to copy the page, one machine dependent page at a
1623 *	time.
1624 */
1625void
1626pmap_copy_page(src, dst)
1627	vm_offset_t src;
1628	vm_offset_t dst;
1629{
1630	if (*(int *)CMAP1 || *(int *)CMAP2)
1631		panic("pmap_copy_page: CMAP busy");
1632
1633	*(int *)CMAP1 = PG_V | PG_KW | i386_trunc_page(src);
1634	*(int *)CMAP2 = PG_V | PG_KW | i386_trunc_page(dst);
1635
1636#if __GNUC__ > 1
1637	memcpy(CADDR2, CADDR1, NBPG);
1638#else
1639	bcopy(CADDR1, CADDR2, NBPG);
1640#endif
1641	*(int *)CMAP1 = 0;
1642	*(int *)CMAP2 = 0;
1643	tlbflush();
1644}
1645
1646
1647/*
1648 *	Routine:	pmap_pageable
1649 *	Function:
1650 *		Make the specified pages (by pmap, offset)
1651 *		pageable (or not) as requested.
1652 *
1653 *		A page which is not pageable may not take
1654 *		a fault; therefore, its page table entry
1655 *		must remain valid for the duration.
1656 *
1657 *		This routine is merely advisory; pmap_enter
1658 *		will specify that these pages are to be wired
1659 *		down (or not) as appropriate.
1660 */
1661void
1662pmap_pageable(pmap, sva, eva, pageable)
1663	pmap_t		pmap;
1664	vm_offset_t	sva, eva;
1665	boolean_t	pageable;
1666{
1667}
1668
1669/*
1670 * this routine returns true if a physical page resides
1671 * in the given pmap.
1672 */
1673boolean_t
1674pmap_page_exists(pmap, pa)
1675	pmap_t pmap;
1676	vm_offset_t pa;
1677{
1678	register pv_entry_t pv;
1679	int s;
1680
1681	if (!pmap_is_managed(pa))
1682		return FALSE;
1683
1684	pv = pa_to_pvh(pa);
1685	s = splhigh();
1686
1687	/*
1688	 * Not found, check current mappings returning
1689	 * immediately if found.
1690	 */
1691	if (pv->pv_pmap != NULL) {
1692		for (; pv; pv = pv->pv_next) {
1693			if (pv->pv_pmap == pmap) {
1694				splx(s);
1695				return TRUE;
1696			}
1697		}
1698	}
1699	splx(s);
1700	return(FALSE);
1701}
1702
1703/*
1704 * pmap_testbit tests bits in pte's
1705 * note that the testbit/changebit routines are inline,
1706 * and a lot of things compile-time evaluate.
1707 */
1708static inline boolean_t
1709pmap_testbit(pa, bit)
1710	register vm_offset_t pa;
1711	int bit;
1712{
1713	register pv_entry_t pv;
1714	pt_entry_t *pte;
1715	int s;
1716
1717	if (!pmap_is_managed(pa))
1718		return FALSE;
1719
1720	pv = pa_to_pvh(pa);
1721	s = splhigh();
1722
1723	/*
1724	 * Not found, check current mappings returning
1725	 * immediately if found.
1726	 */
1727	if (pv->pv_pmap != NULL) {
1728		for (; pv; pv = pv->pv_next) {
1729			/*
1730			 * if the bit being tested is the modified bit,
1731			 * then mark UPAGES as always modified, and
1732			 * ptes as never modified.
1733			 */
1734			if (bit & PG_U ) {
1735				if ((pv->pv_va >= clean_sva) && (pv->pv_va < clean_eva)) {
1736					continue;
1737				}
1738			}
1739			if (bit & PG_M ) {
1740				if (pv->pv_va >= USRSTACK) {
1741					if (pv->pv_va >= clean_sva && pv->pv_va < clean_eva) {
1742						continue;
1743					}
1744					if (pv->pv_va < USRSTACK+(UPAGES*NBPG)) {
1745						splx(s);
1746						return TRUE;
1747					}
1748					else if (pv->pv_va < UPT_MAX_ADDRESS) {
1749						splx(s);
1750						return FALSE;
1751					}
1752				}
1753			}
1754			if( !pv->pv_pmap) {
1755				printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va);
1756				continue;
1757			}
1758			pte = pmap_pte(pv->pv_pmap, pv->pv_va);
1759			if ((int) *pte & bit) {
1760				splx(s);
1761				return TRUE;
1762			}
1763		}
1764	}
1765	splx(s);
1766	return(FALSE);
1767}
1768
1769/*
1770 * this routine is used to modify bits in ptes
1771 */
1772static inline void
1773pmap_changebit(pa, bit, setem)
1774	vm_offset_t pa;
1775	int bit;
1776	boolean_t setem;
1777{
1778	register pv_entry_t pv;
1779	register pt_entry_t *pte, npte;
1780	vm_offset_t va;
1781	int s;
1782
1783	if (!pmap_is_managed(pa))
1784		return;
1785
1786	pv = pa_to_pvh(pa);
1787	s = splhigh();
1788
1789	/*
1790	 * Loop over all current mappings setting/clearing as appropos
1791	 * If setting RO do we need to clear the VAC?
1792	 */
1793	if (pv->pv_pmap != NULL) {
1794		for (; pv; pv = pv->pv_next) {
1795			va = pv->pv_va;
1796
1797			/*
1798			 * don't write protect pager mappings
1799			 */
1800			if (!setem && (bit == PG_RW)) {
1801				if (va >= clean_sva && va < clean_eva)
1802					continue;
1803			}
1804
1805			if( !pv->pv_pmap) {
1806				printf("Null pmap (cb) at va: 0x%lx\n", va);
1807				continue;
1808			}
1809			pte = pmap_pte(pv->pv_pmap, va);
1810			if (setem)
1811				(int) npte = (int) *pte | bit;
1812			else
1813				(int) npte = (int) *pte & ~bit;
1814			*pte = npte;
1815		}
1816	}
1817	splx(s);
1818	tlbflush();
1819}
1820
1821/*
1822 *      pmap_page_protect:
1823 *
1824 *      Lower the permission for all mappings to a given page.
1825 */
1826void
1827pmap_page_protect(phys, prot)
1828        vm_offset_t     phys;
1829        vm_prot_t       prot;
1830{
1831	if ((prot & VM_PROT_WRITE) == 0) {
1832		if (prot & (VM_PROT_READ | VM_PROT_EXECUTE))
1833			pmap_changebit(phys, PG_RW, FALSE);
1834		else
1835			pmap_remove_all(phys);
1836	}
1837}
1838
1839/*
1840 *	Clear the modify bits on the specified physical page.
1841 */
1842void
1843pmap_clear_modify(pa)
1844	vm_offset_t	pa;
1845{
1846	pmap_changebit(pa, PG_M, FALSE);
1847}
1848
1849/*
1850 *	pmap_clear_reference:
1851 *
1852 *	Clear the reference bit on the specified physical page.
1853 */
1854void
1855pmap_clear_reference(pa)
1856	vm_offset_t	pa;
1857{
1858	pmap_changebit(pa, PG_U, FALSE);
1859}
1860
1861/*
1862 *	pmap_is_referenced:
1863 *
1864 *	Return whether or not the specified physical page is referenced
1865 *	by any physical maps.
1866 */
1867
1868boolean_t
1869pmap_is_referenced(pa)
1870	vm_offset_t	pa;
1871{
1872	return(pmap_testbit(pa, PG_U));
1873}
1874
1875/*
1876 *	pmap_is_modified:
1877 *
1878 *	Return whether or not the specified physical page is modified
1879 *	by any physical maps.
1880 */
1881
1882boolean_t
1883pmap_is_modified(pa)
1884	vm_offset_t	pa;
1885{
1886	return(pmap_testbit(pa, PG_M));
1887}
1888
1889/*
1890 *	Routine:	pmap_copy_on_write
1891 *	Function:
1892 *		Remove write privileges from all
1893 *		physical maps for this physical page.
1894 */
1895void
1896pmap_copy_on_write(pa)
1897	vm_offset_t pa;
1898{
1899	pmap_changebit(pa, PG_RW, FALSE);
1900}
1901
1902
1903vm_offset_t
1904pmap_phys_address(ppn)
1905	int ppn;
1906{
1907	return(i386_ptob(ppn));
1908}
1909
1910/*
1911 * Miscellaneous support routines follow
1912 */
1913
1914void
1915i386_protection_init()
1916{
1917	register int *kp, prot;
1918
1919	kp = protection_codes;
1920	for (prot = 0; prot < 8; prot++) {
1921		switch (prot) {
1922		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
1923		/*
1924		 * Read access is also 0. There isn't any execute
1925		 * bit, so just make it readable.
1926		 */
1927		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
1928		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
1929		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
1930			*kp++ = 0;
1931			break;
1932		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
1933		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
1934		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
1935		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
1936			*kp++ = PG_RW;
1937			break;
1938		}
1939	}
1940}
1941
1942#ifdef DEBUG
1943/* print address space of pmap*/
1944void
1945pads(pm)
1946	pmap_t pm;
1947{
1948	unsigned va, i, j;
1949	pt_entry_t *ptep;
1950
1951	if (pm == kernel_pmap) return;
1952	for (i = 0; i < 1024; i++)
1953		if (pm->pm_pdir[i])
1954			for (j = 0; j < 1024 ; j++) {
1955				va = (i<<PD_SHIFT)+(j<<PG_SHIFT);
1956				if (pm == kernel_pmap && va < KERNBASE)
1957						continue;
1958				if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
1959						continue;
1960				ptep = pmap_pte(pm, va);
1961				if (pmap_pte_v(ptep))
1962					printf("%x:%x ", va, *(int *)ptep);
1963			} ;
1964
1965}
1966
1967void
1968pmap_pvdump(pa)
1969	vm_offset_t pa;
1970{
1971	register pv_entry_t pv;
1972
1973	printf("pa %x", pa);
1974	for (pv = pa_to_pvh(pa); pv; pv = pv->pv_next) {
1975#ifdef used_to_be
1976		printf(" -> pmap %x, va %x, flags %x",
1977		       pv->pv_pmap, pv->pv_va, pv->pv_flags);
1978#endif
1979		printf(" -> pmap %x, va %x",
1980		       pv->pv_pmap, pv->pv_va);
1981		pads(pv->pv_pmap);
1982	}
1983	printf(" ");
1984}
1985#endif
1986
1987