pmap.c revision 424
1/*
2 * Copyright (c) 1991 Regents of the University of California.
3 * All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * the Systems Programming Group of the University of Utah Computer
7 * Science Department and William Jolitz of UUNET Technologies Inc.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 *    must display the following acknowledgement:
19 *	This product includes software developed by the University of
20 *	California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 *    may be used to endorse or promote products derived from this software
23 *    without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 *	@(#)pmap.c	7.7 (Berkeley)	5/12/91
38 *
39 * PATCHES MAGIC                LEVEL   PATCH THAT GOT US HERE
40 * --------------------         -----   ----------------------
41 * CURRENT PATCH LEVEL:         1       00063
42 * --------------------         -----   ----------------------
43 *
44 * 28 Nov 1991	Poul-Henning Kamp	Speedup processing.
45 */
46static char rcsid[] = "$Header: /a/cvs/386BSD/src/sys/i386/i386/pmap.c,v 1.2 1993/07/27 10:52:19 davidg Exp $";
47
48/*
49 * Derived from hp300 version by Mike Hibler, this version by William
50 * Jolitz uses a recursive map [a pde points to the page directory] to
51 * map the page tables using the pagetables themselves. This is done to
52 * reduce the impact on kernel virtual memory for lots of sparse address
53 * space, and to reduce the cost of memory to each process.
54 *
55 *	Derived from: hp300/@(#)pmap.c	7.1 (Berkeley) 12/5/90
56 */
57
58/*
59 *	Reno i386 version, from Mike Hibler's hp300 version.
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 "param.h"
89#include "proc.h"
90#include "malloc.h"
91#include "user.h"
92
93#include "vm/vm.h"
94#include "vm/vm_kern.h"
95#include "vm/vm_page.h"
96/*#include "vm/vm_pageout.h"*/
97
98#include "i386/isa/isa.h"
99
100/*
101 * Allocate various and sundry SYSMAPs used in the days of old VM
102 * and not yet converted.  XXX.
103 */
104#define BSDVM_COMPAT	1
105
106#ifdef DEBUG
107struct {
108	int kernel;	/* entering kernel mapping */
109	int user;	/* entering user mapping */
110	int ptpneeded;	/* needed to allocate a PT page */
111	int pwchange;	/* no mapping change, just wiring or protection */
112	int wchange;	/* no mapping change, just wiring */
113	int mchange;	/* was mapped but mapping to different page */
114	int managed;	/* a managed page */
115	int firstpv;	/* first mapping for this PA */
116	int secondpv;	/* second mapping for this PA */
117	int ci;		/* cache inhibited */
118	int unmanaged;	/* not a managed page */
119	int flushes;	/* cache flushes */
120} enter_stats;
121struct {
122	int calls;
123	int removes;
124	int pvfirst;
125	int pvsearch;
126	int ptinvalid;
127	int uflushes;
128	int sflushes;
129} remove_stats;
130
131int debugmap = 0;
132int pmapdebug = 0 /* 0xffff */;
133#define PDB_FOLLOW	0x0001
134#define PDB_INIT	0x0002
135#define PDB_ENTER	0x0004
136#define PDB_REMOVE	0x0008
137#define PDB_CREATE	0x0010
138#define PDB_PTPAGE	0x0020
139#define PDB_CACHE	0x0040
140#define PDB_BITS	0x0080
141#define PDB_COLLECT	0x0100
142#define PDB_PROTECT	0x0200
143#define PDB_PDRTAB	0x0400
144#define PDB_PARANOIA	0x2000
145#define PDB_WIRING	0x4000
146#define PDB_PVDUMP	0x8000
147
148int pmapvacflush = 0;
149#define	PVF_ENTER	0x01
150#define	PVF_REMOVE	0x02
151#define	PVF_PROTECT	0x04
152#define	PVF_TOTAL	0x80
153#endif
154
155/*
156 * Get PDEs and PTEs for user/kernel address space
157 */
158#define	pmap_pde(m, v)	(&((m)->pm_pdir[((vm_offset_t)(v) >> PD_SHIFT)&1023]))
159
160#define pmap_pte_pa(pte)	(*(int *)(pte) & PG_FRAME)
161
162#define pmap_pde_v(pte)		((pte)->pd_v)
163#define pmap_pte_w(pte)		((pte)->pg_w)
164/* #define pmap_pte_ci(pte)	((pte)->pg_ci) */
165#define pmap_pte_m(pte)		((pte)->pg_m)
166#define pmap_pte_u(pte)		((pte)->pg_u)
167#define pmap_pte_v(pte)		((pte)->pg_v)
168#define pmap_pte_set_w(pte, v)		((pte)->pg_w = (v))
169#define pmap_pte_set_prot(pte, v)	((pte)->pg_prot = (v))
170
171/*
172 * Given a map and a machine independent protection code,
173 * convert to a vax protection code.
174 */
175#define pte_prot(m, p)	(protection_codes[p])
176int	protection_codes[8];
177
178struct pmap	kernel_pmap_store;
179pmap_t		kernel_pmap;
180
181vm_offset_t    	avail_start;	/* PA of first available physical page */
182vm_offset_t	avail_end;	/* PA of last available physical page */
183vm_size_t	mem_size;	/* memory size in bytes */
184vm_offset_t	virtual_avail;  /* VA of first avail page (after kernel bss)*/
185vm_offset_t	virtual_end;	/* VA of last avail page (end of kernel AS) */
186vm_offset_t	vm_first_phys;	/* PA of first managed page */
187vm_offset_t	vm_last_phys;	/* PA just past last managed page */
188int		i386pagesperpage;	/* PAGE_SIZE / I386_PAGE_SIZE */
189boolean_t	pmap_initialized = FALSE;	/* Has pmap_init completed? */
190char		*pmap_attributes;	/* reference and modify bits */
191
192boolean_t	pmap_testbit();
193void		pmap_clear_modify();
194
195#if BSDVM_COMPAT
196#include "msgbuf.h"
197
198/*
199 * All those kernel PT submaps that BSD is so fond of
200 */
201struct pte	*CMAP1, *CMAP2, *mmap;
202caddr_t		CADDR1, CADDR2, vmmap;
203struct pte	*msgbufmap;
204struct msgbuf	*msgbufp;
205#endif
206
207/*
208 *	Bootstrap the system enough to run with virtual memory.
209 *	Map the kernel's code and data, and allocate the system page table.
210 *
211 *	On the I386 this is called after mapping has already been enabled
212 *	and just syncs the pmap module with what has already been done.
213 *	[We can't call it easily with mapping off since the kernel is not
214 *	mapped with PA == VA, hence we would have to relocate every address
215 *	from the linked base (virtual) address 0xFE000000 to the actual
216 *	(physical) address starting relative to 0]
217 */
218struct pte *pmap_pte();
219
220void
221pmap_bootstrap(firstaddr, loadaddr)
222	vm_offset_t firstaddr;
223	vm_offset_t loadaddr;
224{
225#if BSDVM_COMPAT
226	vm_offset_t va;
227	struct pte *pte;
228#endif
229	extern vm_offset_t maxmem, physmem;
230extern int IdlePTD;
231
232	avail_start = firstaddr;
233	avail_end = maxmem << PG_SHIFT;
234
235	/* XXX: allow for msgbuf */
236	avail_end -= i386_round_page(sizeof(struct msgbuf));
237
238	mem_size = physmem << PG_SHIFT;
239	virtual_avail = (vm_offset_t)atdevbase + 0x100000 - 0xa0000 + 10*NBPG;
240	virtual_end = VM_MAX_KERNEL_ADDRESS;
241	i386pagesperpage = PAGE_SIZE / I386_PAGE_SIZE;
242
243	/*
244	 * Initialize protection array.
245	 */
246	i386_protection_init();
247
248	/*
249	 * The kernel's pmap is statically allocated so we don't
250	 * have to use pmap_create, which is unlikely to work
251	 * correctly at this part of the boot sequence.
252	 */
253	kernel_pmap = &kernel_pmap_store;
254
255#ifdef notdef
256	/*
257	 * Create Kernel page directory table and page maps.
258	 * [ currently done in locore. i have wild and crazy ideas -wfj ]
259	 */
260	bzero(firstaddr, 4*NBPG);
261	kernel_pmap->pm_pdir = firstaddr + VM_MIN_KERNEL_ADDRESS;
262	kernel_pmap->pm_ptab = firstaddr + VM_MIN_KERNEL_ADDRESS + NBPG;
263
264	firstaddr += NBPG;
265	for (x = i386_btod(VM_MIN_KERNEL_ADDRESS);
266		x < i386_btod(VM_MIN_KERNEL_ADDRESS)+3; x++) {
267			struct pde *pde;
268		pde = kernel_pmap->pm_pdir + x;
269		*(int *)pde = firstaddr + x*NBPG | PG_V | PG_KW;
270	}
271#else
272	kernel_pmap->pm_pdir = (pd_entry_t *)(0xfe000000 + IdlePTD);
273#endif
274
275
276	simple_lock_init(&kernel_pmap->pm_lock);
277	kernel_pmap->pm_count = 1;
278
279#if BSDVM_COMPAT
280	/*
281	 * Allocate all the submaps we need
282	 */
283#define	SYSMAP(c, p, v, n)	\
284	v = (c)va; va += ((n)*I386_PAGE_SIZE); p = pte; pte += (n);
285
286	va = virtual_avail;
287	pte = pmap_pte(kernel_pmap, va);
288
289	SYSMAP(caddr_t		,CMAP1		,CADDR1	   ,1		)
290	SYSMAP(caddr_t		,CMAP2		,CADDR2	   ,1		)
291	SYSMAP(caddr_t		,mmap		,vmmap	   ,1		)
292	SYSMAP(struct msgbuf *	,msgbufmap	,msgbufp   ,1		)
293	virtual_avail = va;
294#endif
295	/*
296	 * reserve special hunk of memory for use by bus dma as a bounce
297	 * buffer (contiguous virtual *and* physical memory). for now,
298	 * assume vm does not use memory beneath hole, and we know that
299	 * the bootstrap uses top 32k of base memory. -wfj
300	 */
301	{
302		extern vm_offset_t isaphysmem;
303	isaphysmem = va;
304
305	virtual_avail = pmap_map(va, 0xa0000 - 32*1024, 0xa0000, VM_PROT_ALL);
306	}
307
308	*(int *)PTD = 0;
309	load_cr3(rcr3());
310
311}
312
313/*
314 *	Initialize the pmap module.
315 *	Called by vm_init, to initialize any structures that the pmap
316 *	system needs to map virtual memory.
317 */
318void
319pmap_init(phys_start, phys_end)
320	vm_offset_t	phys_start, phys_end;
321{
322	vm_offset_t	addr, addr2;
323	vm_size_t	npg, s;
324	int		rv;
325	extern int KPTphys;
326
327#ifdef DEBUG
328	if (pmapdebug & PDB_FOLLOW)
329		printf("pmap_init(%x, %x)\n", phys_start, phys_end);
330#endif
331	/*
332	 * Now that kernel map has been allocated, we can mark as
333	 * unavailable regions which we have mapped in locore.
334	 */
335	addr = atdevbase;
336	(void) vm_map_find(kernel_map, NULL, (vm_offset_t) 0,
337			   &addr, (0x100000-0xa0000), FALSE);
338
339	addr = (vm_offset_t) 0xfe000000+KPTphys/* *NBPG */;
340	vm_object_reference(kernel_object);
341	(void) vm_map_find(kernel_map, kernel_object, addr,
342			   &addr, 2*NBPG, FALSE);
343
344	/*
345	 * Allocate memory for random pmap data structures.  Includes the
346	 * pv_head_table and pmap_attributes.
347	 */
348	npg = atop(phys_end - phys_start);
349	s = (vm_size_t) (sizeof(struct pv_entry) * npg + npg);
350	s = round_page(s);
351	addr = (vm_offset_t) kmem_alloc(kernel_map, s);
352	pv_table = (pv_entry_t) addr;
353	addr += sizeof(struct pv_entry) * npg;
354	pmap_attributes = (char *) addr;
355#ifdef DEBUG
356	if (pmapdebug & PDB_INIT)
357		printf("pmap_init: %x bytes (%x pgs): tbl %x attr %x\n",
358		       s, npg, pv_table, pmap_attributes);
359#endif
360
361	/*
362	 * Now it is safe to enable pv_table recording.
363	 */
364	vm_first_phys = phys_start;
365	vm_last_phys = phys_end;
366	pmap_initialized = TRUE;
367}
368
369/*
370 *	Used to map a range of physical addresses into kernel
371 *	virtual address space.
372 *
373 *	For now, VM is already on, we only need to map the
374 *	specified memory.
375 */
376vm_offset_t
377pmap_map(virt, start, end, prot)
378	vm_offset_t	virt;
379	vm_offset_t	start;
380	vm_offset_t	end;
381	int		prot;
382{
383#ifdef DEBUG
384	if (pmapdebug & PDB_FOLLOW)
385		printf("pmap_map(%x, %x, %x, %x)\n", virt, start, end, prot);
386#endif
387	while (start < end) {
388		pmap_enter(kernel_pmap, virt, start, prot, FALSE);
389		virt += PAGE_SIZE;
390		start += PAGE_SIZE;
391	}
392	return(virt);
393}
394
395/*
396 *	Create and return a physical map.
397 *
398 *	If the size specified for the map
399 *	is zero, the map is an actual physical
400 *	map, and may be referenced by the
401 *	hardware.
402 *
403 *	If the size specified is non-zero,
404 *	the map will be used in software only, and
405 *	is bounded by that size.
406 *
407 * [ just allocate a ptd and mark it uninitialize -- should we track
408 *   with a table which process has which ptd? -wfj ]
409 */
410
411pmap_t
412pmap_create(size)
413	vm_size_t	size;
414{
415	register pmap_t pmap;
416
417#ifdef DEBUG
418	if (pmapdebug & (PDB_FOLLOW|PDB_CREATE))
419		printf("pmap_create(%x)\n", size);
420#endif
421	/*
422	 * Software use map does not need a pmap
423	 */
424	if (size)
425		return(NULL);
426
427	/* XXX: is it ok to wait here? */
428	pmap = (pmap_t) malloc(sizeof *pmap, M_VMPMAP, M_WAITOK);
429#ifdef notifwewait
430	if (pmap == NULL)
431		panic("pmap_create: cannot allocate a pmap");
432#endif
433	bzero(pmap, sizeof(*pmap));
434	pmap_pinit(pmap);
435	return (pmap);
436}
437
438/*
439 * Initialize a preallocated and zeroed pmap structure,
440 * such as one in a vmspace structure.
441 */
442void
443pmap_pinit(pmap)
444	register struct pmap *pmap;
445{
446
447#ifdef DEBUG
448	if (pmapdebug & (PDB_FOLLOW|PDB_CREATE))
449		pg("pmap_pinit(%x)\n", pmap);
450#endif
451
452	/*
453	 * No need to allocate page table space yet but we do need a
454	 * valid page directory table.
455	 */
456	pmap->pm_pdir = (pd_entry_t *) kmem_alloc(kernel_map, NBPG);
457
458	/* wire in kernel global address entries */
459	bcopy(PTD+KPTDI_FIRST, pmap->pm_pdir+KPTDI_FIRST,
460		(KPTDI_LAST-KPTDI_FIRST+1)*4);
461
462	/* install self-referential address mapping entry */
463	*(int *)(pmap->pm_pdir+PTDPTDI) =
464		(int)pmap_extract(kernel_pmap, pmap->pm_pdir) | PG_V | PG_KW;
465
466	pmap->pm_count = 1;
467	simple_lock_init(&pmap->pm_lock);
468}
469
470/*
471 *	Retire the given physical map from service.
472 *	Should only be called if the map contains
473 *	no valid mappings.
474 */
475void
476pmap_destroy(pmap)
477	register pmap_t pmap;
478{
479	int count;
480
481#ifdef DEBUG
482	if (pmapdebug & PDB_FOLLOW)
483		printf("pmap_destroy(%x)\n", pmap);
484#endif
485	if (pmap == NULL)
486		return;
487
488	simple_lock(&pmap->pm_lock);
489	count = --pmap->pm_count;
490	simple_unlock(&pmap->pm_lock);
491	if (count == 0) {
492		pmap_release(pmap);
493		free((caddr_t)pmap, M_VMPMAP);
494	}
495}
496
497/*
498 * Release any resources held by the given physical map.
499 * Called when a pmap initialized by pmap_pinit is being released.
500 * Should only be called if the map contains no valid mappings.
501 */
502void
503pmap_release(pmap)
504	register struct pmap *pmap;
505{
506
507#ifdef DEBUG
508	if (pmapdebug & PDB_FOLLOW)
509		pg("pmap_release(%x)\n", pmap);
510#endif
511#ifdef notdef /* DIAGNOSTIC */
512	/* count would be 0 from pmap_destroy... */
513	simple_lock(&pmap->pm_lock);
514	if (pmap->pm_count != 1)
515		panic("pmap_release count");
516#endif
517	kmem_free(kernel_map, (vm_offset_t)pmap->pm_pdir, NBPG);
518}
519
520/*
521 *	Add a reference to the specified pmap.
522 */
523void
524pmap_reference(pmap)
525	pmap_t	pmap;
526{
527#ifdef DEBUG
528	if (pmapdebug & PDB_FOLLOW)
529		printf("pmap_reference(%x)", pmap);
530#endif
531	if (pmap != NULL) {
532		simple_lock(&pmap->pm_lock);
533		pmap->pm_count++;
534		simple_unlock(&pmap->pm_lock);
535	}
536}
537
538/*
539 *	Remove the given range of addresses from the specified map.
540 *
541 *	It is assumed that the start and end are properly
542 *	rounded to the page size.
543 */
544void
545pmap_remove(pmap, sva, eva)
546	struct pmap *pmap;
547	register vm_offset_t sva;
548	register vm_offset_t eva;
549{
550	register pt_entry_t *ptp,*ptq;
551	vm_offset_t va;
552	vm_offset_t pa;
553	pt_entry_t *pte;
554	pv_entry_t pv, npv;
555	int ix;
556	int s, bits;
557#ifdef DEBUG
558	pt_entry_t opte;
559
560	if (pmapdebug & (PDB_FOLLOW|PDB_REMOVE|PDB_PROTECT))
561		pg("pmap_remove(%x, %x, %x)", pmap, sva, eva);
562#endif
563
564	if (pmap == NULL)
565		return;
566
567	/* are we current address space or kernel? */
568	if (pmap->pm_pdir[PTDPTDI].pd_pfnum == PTDpde.pd_pfnum
569		|| pmap == kernel_pmap)
570		ptp=PTmap;
571
572	/* otherwise, we are alternate address space */
573	else {
574		if (pmap->pm_pdir[PTDPTDI].pd_pfnum
575			!= APTDpde.pd_pfnum) {
576			APTDpde = pmap->pm_pdir[PTDPTDI];
577			tlbflush();
578		}
579		ptp=APTmap;
580	     }
581#ifdef DEBUG
582	remove_stats.calls++;
583#endif
584
585	/* this is essential since we must check the PDE(sva) for precense */
586	while (sva <= eva && !pmap_pde_v(pmap_pde(pmap, sva)))
587		sva = (sva & PD_MASK) + (1<<PD_SHIFT);
588	sva = i386_btop(sva);
589	eva = i386_btop(eva);
590
591	for (; sva < eva; sva++) {
592		/*
593		 * Weed out invalid mappings.
594		 * Note: we assume that the page directory table is
595	 	 * always allocated, and in kernel virtual.
596		 */
597		ptq=ptp+sva;
598		while((sva & 0x3ff) && !pmap_pte_pa(ptq))
599		    {
600		    if(++sva >= eva)
601		        return;
602		    ptq++;
603		    }
604
605
606		if(!(sva & 0x3ff)) /* Only check once in a while */
607 		    {
608		    if (!pmap_pde_v(pmap_pde(pmap, i386_ptob(sva))))
609			{
610			/* We can race ahead here, straight to next pde.. */
611			sva = (sva & 0xffc00) + (1<<10) -1 ;
612			continue;
613			}
614		    }
615	        if(!pmap_pte_pa(ptp+sva))
616		    continue;
617
618		pte = ptp + sva;
619		pa = pmap_pte_pa(pte);
620		va = i386_ptob(sva);
621#ifdef DEBUG
622		opte = *pte;
623		remove_stats.removes++;
624#endif
625		/*
626		 * Update statistics
627		 */
628		if (pmap_pte_w(pte))
629			pmap->pm_stats.wired_count--;
630		pmap->pm_stats.resident_count--;
631
632		/*
633		 * Invalidate the PTEs.
634		 * XXX: should cluster them up and invalidate as many
635		 * as possible at once.
636		 */
637#ifdef DEBUG
638		if (pmapdebug & PDB_REMOVE)
639			printf("remove: inv %x ptes at %x(%x) ",
640			       i386pagesperpage, pte, *(int *)pte);
641#endif
642		bits = ix = 0;
643		do {
644			bits |= *(int *)pte & (PG_U|PG_M);
645			*(int *)pte++ = 0;
646			/*TBIS(va + ix * I386_PAGE_SIZE);*/
647		} while (++ix != i386pagesperpage);
648		if (curproc && pmap == &curproc->p_vmspace->vm_pmap)
649			pmap_activate(pmap, (struct pcb *)curproc->p_addr);
650		/* are we current address space or kernel? */
651		/*if (pmap->pm_pdir[PTDPTDI].pd_pfnum == PTDpde.pd_pfnum
652			|| pmap == kernel_pmap)
653		load_cr3(curpcb->pcb_ptd);*/
654		tlbflush();
655
656#ifdef needednotdone
657reduce wiring count on page table pages as references drop
658#endif
659
660		/*
661		 * Remove from the PV table (raise IPL since we
662		 * may be called at interrupt time).
663		 */
664		if (pa < vm_first_phys || pa >= vm_last_phys)
665			continue;
666		pv = pa_to_pvh(pa);
667		s = splimp();
668		/*
669		 * If it is the first entry on the list, it is actually
670		 * in the header and we must copy the following entry up
671		 * to the header.  Otherwise we must search the list for
672		 * the entry.  In either case we free the now unused entry.
673		 */
674		if (pmap == pv->pv_pmap && va == pv->pv_va) {
675			npv = pv->pv_next;
676			if (npv) {
677				*pv = *npv;
678				free((caddr_t)npv, M_VMPVENT);
679			} else
680				pv->pv_pmap = NULL;
681#ifdef DEBUG
682			remove_stats.pvfirst++;
683#endif
684		} else {
685			for (npv = pv->pv_next; npv; npv = npv->pv_next) {
686#ifdef DEBUG
687				remove_stats.pvsearch++;
688#endif
689				if (pmap == npv->pv_pmap && va == npv->pv_va)
690					break;
691				pv = npv;
692			}
693#ifdef DEBUG
694			if (npv == NULL)
695				panic("pmap_remove: PA not in pv_tab");
696#endif
697			pv->pv_next = npv->pv_next;
698			free((caddr_t)npv, M_VMPVENT);
699			pv = pa_to_pvh(pa);
700		}
701
702#ifdef notdef
703[tally number of pagetable pages, if sharing of ptpages adjust here]
704#endif
705		/*
706		 * Update saved attributes for managed page
707		 */
708		pmap_attributes[pa_index(pa)] |= bits;
709		splx(s);
710	}
711#ifdef notdef
712[cache and tlb flushing, if needed]
713#endif
714}
715
716/*
717 *	Routine:	pmap_remove_all
718 *	Function:
719 *		Removes this physical page from
720 *		all physical maps in which it resides.
721 *		Reflects back modify bits to the pager.
722 */
723void
724pmap_remove_all(pa)
725	vm_offset_t pa;
726{
727	register pv_entry_t pv;
728	int s;
729
730#ifdef DEBUG
731	if (pmapdebug & (PDB_FOLLOW|PDB_REMOVE|PDB_PROTECT))
732		printf("pmap_remove_all(%x)", pa);
733	/*pmap_pvdump(pa);*/
734#endif
735	/*
736	 * Not one of ours
737	 */
738	if (pa < vm_first_phys || pa >= vm_last_phys)
739		return;
740
741	pv = pa_to_pvh(pa);
742	s = splimp();
743	/*
744	 * Do it the easy way for now
745	 */
746	while (pv->pv_pmap != NULL) {
747#ifdef DEBUG
748		if (!pmap_pde_v(pmap_pde(pv->pv_pmap, pv->pv_va)) ||
749		    pmap_pte_pa(pmap_pte(pv->pv_pmap, pv->pv_va)) != pa)
750			panic("pmap_remove_all: bad mapping");
751#endif
752		pmap_remove(pv->pv_pmap, pv->pv_va, pv->pv_va + PAGE_SIZE);
753	}
754	splx(s);
755}
756
757/*
758 *	Routine:	pmap_copy_on_write
759 *	Function:
760 *		Remove write privileges from all
761 *		physical maps for this physical page.
762 */
763void
764pmap_copy_on_write(pa)
765	vm_offset_t pa;
766{
767#ifdef DEBUG
768	if (pmapdebug & (PDB_FOLLOW|PDB_PROTECT))
769		printf("pmap_copy_on_write(%x)", pa);
770#endif
771	pmap_changebit(pa, PG_RO, TRUE);
772}
773
774/*
775 *	Set the physical protection on the
776 *	specified range of this map as requested.
777 */
778void
779pmap_protect(pmap, sva, eva, prot)
780	register pmap_t	pmap;
781	vm_offset_t	sva, eva;
782	vm_prot_t	prot;
783{
784	register pt_entry_t *pte;
785	register vm_offset_t va;
786	register int ix;
787	int i386prot;
788	boolean_t firstpage = TRUE;
789	register pt_entry_t *ptp;
790
791#ifdef DEBUG
792	if (pmapdebug & (PDB_FOLLOW|PDB_PROTECT))
793		printf("pmap_protect(%x, %x, %x, %x)", pmap, sva, eva, prot);
794#endif
795	if (pmap == NULL)
796		return;
797
798	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
799		pmap_remove(pmap, sva, eva);
800		return;
801	}
802	if (prot & VM_PROT_WRITE)
803		return;
804
805	/* are we current address space or kernel? */
806	if (pmap->pm_pdir[PTDPTDI].pd_pfnum == PTDpde.pd_pfnum
807		|| pmap == kernel_pmap)
808		ptp=PTmap;
809
810	/* otherwise, we are alternate address space */
811	else {
812		if (pmap->pm_pdir[PTDPTDI].pd_pfnum
813			!= APTDpde.pd_pfnum) {
814			APTDpde = pmap->pm_pdir[PTDPTDI];
815			tlbflush();
816		}
817		ptp=APTmap;
818	     }
819	for (va = sva; va < eva; va += PAGE_SIZE) {
820		/*
821		 * Page table page is not allocated.
822		 * Skip it, we don't want to force allocation
823		 * of unnecessary PTE pages just to set the protection.
824		 */
825		if (!pmap_pde_v(pmap_pde(pmap, va))) {
826			/* XXX: avoid address wrap around */
827			if (va >= i386_trunc_pdr((vm_offset_t)-1))
828				break;
829			va = i386_round_pdr(va + PAGE_SIZE) - PAGE_SIZE;
830			continue;
831		}
832
833		pte = ptp + i386_btop(va);
834
835		/*
836		 * Page not valid.  Again, skip it.
837		 * Should we do this?  Or set protection anyway?
838		 */
839		if (!pmap_pte_v(pte))
840			continue;
841
842		ix = 0;
843		i386prot = pte_prot(pmap, prot);
844		if(va < UPT_MAX_ADDRESS)
845			i386prot |= 2 /*PG_u*/;
846		do {
847			/* clear VAC here if PG_RO? */
848			pmap_pte_set_prot(pte++, i386prot);
849			/*TBIS(va + ix * I386_PAGE_SIZE);*/
850		} while (++ix != i386pagesperpage);
851	}
852	if (curproc && pmap == &curproc->p_vmspace->vm_pmap)
853		pmap_activate(pmap, (struct pcb *)curproc->p_addr);
854}
855
856/*
857 *	Insert the given physical page (p) at
858 *	the specified virtual address (v) in the
859 *	target physical map with the protection requested.
860 *
861 *	If specified, the page will be wired down, meaning
862 *	that the related pte can not be reclaimed.
863 *
864 *	NB:  This is the only routine which MAY NOT lazy-evaluate
865 *	or lose information.  That is, this routine must actually
866 *	insert this page into the given map NOW.
867 */
868void
869pmap_enter(pmap, va, pa, prot, wired)
870	register pmap_t pmap;
871	vm_offset_t va;
872	register vm_offset_t pa;
873	vm_prot_t prot;
874	boolean_t wired;
875{
876	register pt_entry_t *pte;
877	register int npte, ix;
878	vm_offset_t opa;
879	boolean_t cacheable = TRUE;
880	boolean_t checkpv = TRUE;
881
882#ifdef DEBUG
883	if (pmapdebug & (PDB_FOLLOW|PDB_ENTER))
884		printf("pmap_enter(%x, %x, %x, %x, %x)",
885		       pmap, va, pa, prot, wired);
886#endif
887	if (pmap == NULL)
888		return;
889
890	if(va > VM_MAX_KERNEL_ADDRESS)panic("pmap_enter: toobig");
891	/* also, should not muck with PTD va! */
892
893#ifdef DEBUG
894	if (pmap == kernel_pmap)
895		enter_stats.kernel++;
896	else
897		enter_stats.user++;
898#endif
899
900	/*
901	 * Page Directory table entry not valid, we need a new PT page
902	 */
903	if (!pmap_pde_v(pmap_pde(pmap, va))) {
904		printf("ptdi %x\n", pmap->pm_pdir[PTDPTDI]);
905		panic("Page Table Directory Invalid (ptdi)");
906	}
907
908	pte = pmap_pte(pmap, va);
909	opa = pmap_pte_pa(pte);
910#ifdef DEBUG
911	if (pmapdebug & PDB_ENTER)
912		printf("enter: pte %x, *pte %x ", pte, *(int *)pte);
913#endif
914
915	/*
916	 * Mapping has not changed, must be protection or wiring change.
917	 */
918	if (opa == pa) {
919#ifdef DEBUG
920		enter_stats.pwchange++;
921#endif
922		/*
923		 * Wiring change, just update stats.
924		 * We don't worry about wiring PT pages as they remain
925		 * resident as long as there are valid mappings in them.
926		 * Hence, if a user page is wired, the PT page will be also.
927		 */
928		if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
929#ifdef DEBUG
930			if (pmapdebug & PDB_ENTER)
931				pg("enter: wiring change -> %x ", wired);
932#endif
933			if (wired)
934				pmap->pm_stats.wired_count++;
935			else
936				pmap->pm_stats.wired_count--;
937#ifdef DEBUG
938			enter_stats.wchange++;
939#endif
940		}
941		goto validate;
942	}
943
944	/*
945	 * Mapping has changed, invalidate old range and fall through to
946	 * handle validating new mapping.
947	 */
948	if (opa) {
949#ifdef DEBUG
950		if (pmapdebug & PDB_ENTER)
951			printf("enter: removing old mapping %x pa %x ", va, opa);
952#endif
953		pmap_remove(pmap, va, va + PAGE_SIZE);
954#ifdef DEBUG
955		enter_stats.mchange++;
956#endif
957	}
958
959	/*
960	 * Enter on the PV list if part of our managed memory
961	 * Note that we raise IPL while manipulating pv_table
962	 * since pmap_enter can be called at interrupt time.
963	 */
964	if (pa >= vm_first_phys && pa < vm_last_phys) {
965		register pv_entry_t pv, npv;
966		int s;
967
968#ifdef DEBUG
969		enter_stats.managed++;
970#endif
971		pv = pa_to_pvh(pa);
972		s = splimp();
973#ifdef DEBUG
974		if (pmapdebug & PDB_ENTER)
975			printf("enter: pv at %x: %x/%x/%x ",
976			       pv, pv->pv_va, pv->pv_pmap, pv->pv_next);
977#endif
978		/*
979		 * No entries yet, use header as the first entry
980		 */
981		if (pv->pv_pmap == NULL) {
982#ifdef DEBUG
983			enter_stats.firstpv++;
984#endif
985			pv->pv_va = va;
986			pv->pv_pmap = pmap;
987			pv->pv_next = NULL;
988			pv->pv_flags = 0;
989		}
990		/*
991		 * There is at least one other VA mapping this page.
992		 * Place this entry after the header.
993		 */
994		else {
995			/*printf("second time: ");*/
996#ifdef DEBUG
997			for (npv = pv; npv; npv = npv->pv_next)
998				if (pmap == npv->pv_pmap && va == npv->pv_va)
999					panic("pmap_enter: already in pv_tab");
1000#endif
1001			npv = (pv_entry_t)
1002				malloc(sizeof *npv, M_VMPVENT, M_NOWAIT);
1003			npv->pv_va = va;
1004			npv->pv_pmap = pmap;
1005			npv->pv_next = pv->pv_next;
1006			pv->pv_next = npv;
1007#ifdef DEBUG
1008			if (!npv->pv_next)
1009				enter_stats.secondpv++;
1010#endif
1011		}
1012		splx(s);
1013	}
1014	/*
1015	 * Assumption: if it is not part of our managed memory
1016	 * then it must be device memory which may be volitile.
1017	 */
1018	if (pmap_initialized) {
1019		checkpv = cacheable = FALSE;
1020#ifdef DEBUG
1021		enter_stats.unmanaged++;
1022#endif
1023	}
1024
1025	/*
1026	 * Increment counters
1027	 */
1028	pmap->pm_stats.resident_count++;
1029	if (wired)
1030		pmap->pm_stats.wired_count++;
1031
1032validate:
1033	/*
1034	 * Now validate mapping with desired protection/wiring.
1035	 * Assume uniform modified and referenced status for all
1036	 * I386 pages in a MACH page.
1037	 */
1038	npte = (pa & PG_FRAME) | pte_prot(pmap, prot) | PG_V;
1039	npte |= (*(int *)pte & (PG_M|PG_U));
1040	if (wired)
1041		npte |= PG_W;
1042	if(va < UPT_MIN_ADDRESS)
1043		npte |= PG_u;
1044	else if(va < UPT_MAX_ADDRESS)
1045		npte |= PG_u | PG_RW;
1046#ifdef DEBUG
1047	if (pmapdebug & PDB_ENTER)
1048		printf("enter: new pte value %x ", npte);
1049#endif
1050	ix = 0;
1051	do {
1052		*(int *)pte++ = npte;
1053		/*TBIS(va);*/
1054		npte += I386_PAGE_SIZE;
1055		va += I386_PAGE_SIZE;
1056	} while (++ix != i386pagesperpage);
1057	pte--;
1058#ifdef DEBUGx
1059cache, tlb flushes
1060#endif
1061/*pads(pmap);*/
1062	/*load_cr3(((struct pcb *)curproc->p_addr)->pcb_ptd);*/
1063	tlbflush();
1064}
1065
1066/*
1067 *      pmap_page_protect:
1068 *
1069 *      Lower the permission for all mappings to a given page.
1070 */
1071void
1072pmap_page_protect(phys, prot)
1073        vm_offset_t     phys;
1074        vm_prot_t       prot;
1075{
1076        switch (prot) {
1077        case VM_PROT_READ:
1078        case VM_PROT_READ|VM_PROT_EXECUTE:
1079                pmap_copy_on_write(phys);
1080                break;
1081        case VM_PROT_ALL:
1082                break;
1083        default:
1084                pmap_remove_all(phys);
1085                break;
1086        }
1087}
1088
1089/*
1090 *	Routine:	pmap_change_wiring
1091 *	Function:	Change the wiring attribute for a map/virtual-address
1092 *			pair.
1093 *	In/out conditions:
1094 *			The mapping must already exist in the pmap.
1095 */
1096void
1097pmap_change_wiring(pmap, va, wired)
1098	register pmap_t	pmap;
1099	vm_offset_t	va;
1100	boolean_t	wired;
1101{
1102	register pt_entry_t *pte;
1103	register int ix;
1104
1105#ifdef DEBUG
1106	if (pmapdebug & PDB_FOLLOW)
1107		printf("pmap_change_wiring(%x, %x, %x)", pmap, va, wired);
1108#endif
1109	if (pmap == NULL)
1110		return;
1111
1112	pte = pmap_pte(pmap, va);
1113#ifdef DEBUG
1114	/*
1115	 * Page table page is not allocated.
1116	 * Should this ever happen?  Ignore it for now,
1117	 * we don't want to force allocation of unnecessary PTE pages.
1118	 */
1119	if (!pmap_pde_v(pmap_pde(pmap, va))) {
1120		if (pmapdebug & PDB_PARANOIA)
1121			pg("pmap_change_wiring: invalid PDE for %x ", va);
1122		return;
1123	}
1124	/*
1125	 * Page not valid.  Should this ever happen?
1126	 * Just continue and change wiring anyway.
1127	 */
1128	if (!pmap_pte_v(pte)) {
1129		if (pmapdebug & PDB_PARANOIA)
1130			pg("pmap_change_wiring: invalid PTE for %x ", va);
1131	}
1132#endif
1133	if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
1134		if (wired)
1135			pmap->pm_stats.wired_count++;
1136		else
1137			pmap->pm_stats.wired_count--;
1138	}
1139	/*
1140	 * Wiring is not a hardware characteristic so there is no need
1141	 * to invalidate TLB.
1142	 */
1143	ix = 0;
1144	do {
1145		pmap_pte_set_w(pte++, wired);
1146	} while (++ix != i386pagesperpage);
1147}
1148
1149/*
1150 *	Routine:	pmap_pte
1151 *	Function:
1152 *		Extract the page table entry associated
1153 *		with the given map/virtual_address pair.
1154 * [ what about induced faults -wfj]
1155 */
1156
1157struct pte *pmap_pte(pmap, va)
1158	register pmap_t	pmap;
1159	vm_offset_t va;
1160{
1161
1162#ifdef DEBUGx
1163	if (pmapdebug & PDB_FOLLOW)
1164		printf("pmap_pte(%x, %x) ->\n", pmap, va);
1165#endif
1166	if (pmap && pmap_pde_v(pmap_pde(pmap, va))) {
1167
1168		/* are we current address space or kernel? */
1169		if (pmap->pm_pdir[PTDPTDI].pd_pfnum == PTDpde.pd_pfnum
1170			|| pmap == kernel_pmap)
1171			return ((struct pte *) vtopte(va));
1172
1173		/* otherwise, we are alternate address space */
1174		else {
1175			if (pmap->pm_pdir[PTDPTDI].pd_pfnum
1176				!= APTDpde.pd_pfnum) {
1177				APTDpde = pmap->pm_pdir[PTDPTDI];
1178				tlbflush();
1179			}
1180			return((struct pte *) avtopte(va));
1181		}
1182	}
1183	return(0);
1184}
1185
1186/*
1187 *	Routine:	pmap_extract
1188 *	Function:
1189 *		Extract the physical page address associated
1190 *		with the given map/virtual_address pair.
1191 */
1192
1193vm_offset_t
1194pmap_extract(pmap, va)
1195	register pmap_t	pmap;
1196	vm_offset_t va;
1197{
1198	register vm_offset_t pa;
1199
1200#ifdef DEBUGx
1201	if (pmapdebug & PDB_FOLLOW)
1202		pg("pmap_extract(%x, %x) -> ", pmap, va);
1203#endif
1204	pa = 0;
1205	if (pmap && pmap_pde_v(pmap_pde(pmap, va))) {
1206		pa = *(int *) pmap_pte(pmap, va);
1207	}
1208	if (pa)
1209		pa = (pa & PG_FRAME) | (va & ~PG_FRAME);
1210#ifdef DEBUGx
1211	if (pmapdebug & PDB_FOLLOW)
1212		printf("%x\n", pa);
1213#endif
1214	return(pa);
1215}
1216
1217/*
1218 *	Copy the range specified by src_addr/len
1219 *	from the source map to the range dst_addr/len
1220 *	in the destination map.
1221 *
1222 *	This routine is only advisory and need not do anything.
1223 */
1224void pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
1225	pmap_t		dst_pmap;
1226	pmap_t		src_pmap;
1227	vm_offset_t	dst_addr;
1228	vm_size_t	len;
1229	vm_offset_t	src_addr;
1230{
1231#ifdef DEBUG
1232	if (pmapdebug & PDB_FOLLOW)
1233		printf("pmap_copy(%x, %x, %x, %x, %x)",
1234		       dst_pmap, src_pmap, dst_addr, len, src_addr);
1235#endif
1236}
1237
1238/*
1239 *	Require that all active physical maps contain no
1240 *	incorrect entries NOW.  [This update includes
1241 *	forcing updates of any address map caching.]
1242 *
1243 *	Generally used to insure that a thread about
1244 *	to run will see a semantically correct world.
1245 */
1246void pmap_update()
1247{
1248#ifdef DEBUG
1249	if (pmapdebug & PDB_FOLLOW)
1250		printf("pmap_update()");
1251#endif
1252	tlbflush();
1253}
1254
1255/*
1256 *	Routine:	pmap_collect
1257 *	Function:
1258 *		Garbage collects the physical map system for
1259 *		pages which are no longer used.
1260 *		Success need not be guaranteed -- that is, there
1261 *		may well be pages which are not referenced, but
1262 *		others may be collected.
1263 *	Usage:
1264 *		Called by the pageout daemon when pages are scarce.
1265 * [ needs to be written -wfj ]
1266 */
1267void
1268pmap_collect(pmap)
1269	pmap_t		pmap;
1270{
1271	register vm_offset_t pa;
1272	register pv_entry_t pv;
1273	register int *pte;
1274	vm_offset_t kpa;
1275	int s;
1276
1277#ifdef DEBUG
1278	int *pde;
1279	int opmapdebug;
1280	printf("pmap_collect(%x) ", pmap);
1281#endif
1282	if (pmap != kernel_pmap)
1283		return;
1284
1285}
1286
1287/* [ macro again?, should I force kstack into user map here? -wfj ] */
1288void
1289pmap_activate(pmap, pcbp)
1290	register pmap_t pmap;
1291	struct pcb *pcbp;
1292{
1293int x;
1294#ifdef DEBUG
1295	if (pmapdebug & (PDB_FOLLOW|PDB_PDRTAB))
1296		pg("pmap_activate(%x, %x) ", pmap, pcbp);
1297#endif
1298	PMAP_ACTIVATE(pmap, pcbp);
1299/*printf("pde ");
1300for(x=0x3f6; x < 0x3fA; x++)
1301	printf("%x ", pmap->pm_pdir[x]);*/
1302/*pads(pmap);*/
1303/*pg(" pcb_cr3 %x", pcbp->pcb_cr3);*/
1304}
1305
1306/*
1307 *	Routine:	pmap_kernel
1308 *	Function:
1309 *		Returns the physical map handle for the kernel.
1310 */
1311pmap_t
1312pmap_kernel()
1313{
1314    	return (kernel_pmap);
1315}
1316
1317/*
1318 *	pmap_zero_page zeros the specified (machine independent)
1319 *	page by mapping the page into virtual memory and using
1320 *	bzero to clear its contents, one machine dependent page
1321 *	at a time.
1322 */
1323pmap_zero_page(phys)
1324	register vm_offset_t	phys;
1325{
1326	register int ix;
1327
1328#ifdef DEBUG
1329	if (pmapdebug & PDB_FOLLOW)
1330		printf("pmap_zero_page(%x)", phys);
1331#endif
1332	phys >>= PG_SHIFT;
1333	ix = 0;
1334	do {
1335		clearseg(phys++);
1336	} while (++ix != i386pagesperpage);
1337}
1338
1339/*
1340 *	pmap_copy_page copies the specified (machine independent)
1341 *	page by mapping the page into virtual memory and using
1342 *	bcopy to copy the page, one machine dependent page at a
1343 *	time.
1344 */
1345pmap_copy_page(src, dst)
1346	register vm_offset_t	src, dst;
1347{
1348	register int ix;
1349
1350#ifdef DEBUG
1351	if (pmapdebug & PDB_FOLLOW)
1352		printf("pmap_copy_page(%x, %x)", src, dst);
1353#endif
1354	src >>= PG_SHIFT;
1355	dst >>= PG_SHIFT;
1356	ix = 0;
1357	do {
1358		physcopyseg(src++, dst++);
1359	} while (++ix != i386pagesperpage);
1360}
1361
1362
1363/*
1364 *	Routine:	pmap_pageable
1365 *	Function:
1366 *		Make the specified pages (by pmap, offset)
1367 *		pageable (or not) as requested.
1368 *
1369 *		A page which is not pageable may not take
1370 *		a fault; therefore, its page table entry
1371 *		must remain valid for the duration.
1372 *
1373 *		This routine is merely advisory; pmap_enter
1374 *		will specify that these pages are to be wired
1375 *		down (or not) as appropriate.
1376 */
1377pmap_pageable(pmap, sva, eva, pageable)
1378	pmap_t		pmap;
1379	vm_offset_t	sva, eva;
1380	boolean_t	pageable;
1381{
1382#ifdef DEBUG
1383	if (pmapdebug & PDB_FOLLOW)
1384		printf("pmap_pageable(%x, %x, %x, %x)",
1385		       pmap, sva, eva, pageable);
1386#endif
1387	/*
1388	 * If we are making a PT page pageable then all valid
1389	 * mappings must be gone from that page.  Hence it should
1390	 * be all zeros and there is no need to clean it.
1391	 * Assumptions:
1392	 *	- we are called with only one page at a time
1393	 *	- PT pages have only one pv_table entry
1394	 */
1395	if (pmap == kernel_pmap && pageable && sva + PAGE_SIZE == eva) {
1396		register pv_entry_t pv;
1397		register vm_offset_t pa;
1398
1399#ifdef DEBUG
1400		if ((pmapdebug & (PDB_FOLLOW|PDB_PTPAGE)) == PDB_PTPAGE)
1401			printf("pmap_pageable(%x, %x, %x, %x)",
1402			       pmap, sva, eva, pageable);
1403#endif
1404		/*if (!pmap_pde_v(pmap_pde(pmap, sva)))
1405			return;*/
1406		if(pmap_pte(pmap, sva) == 0)
1407			return;
1408		pa = pmap_pte_pa(pmap_pte(pmap, sva));
1409		if (pa < vm_first_phys || pa >= vm_last_phys)
1410			return;
1411		pv = pa_to_pvh(pa);
1412		/*if (!ispt(pv->pv_va))
1413			return;*/
1414#ifdef DEBUG
1415		if (pv->pv_va != sva || pv->pv_next) {
1416			pg("pmap_pageable: bad PT page va %x next %x\n",
1417			       pv->pv_va, pv->pv_next);
1418			return;
1419		}
1420#endif
1421		/*
1422		 * Mark it unmodified to avoid pageout
1423		 */
1424		pmap_clear_modify(pa);
1425#ifdef needsomethinglikethis
1426		if (pmapdebug & PDB_PTPAGE)
1427			pg("pmap_pageable: PT page %x(%x) unmodified\n",
1428			       sva, *(int *)pmap_pte(pmap, sva));
1429		if (pmapdebug & PDB_WIRING)
1430			pmap_check_wiring("pageable", sva);
1431#endif
1432	}
1433}
1434
1435/*
1436 *	Clear the modify bits on the specified physical page.
1437 */
1438
1439void
1440pmap_clear_modify(pa)
1441	vm_offset_t	pa;
1442{
1443#ifdef DEBUG
1444	if (pmapdebug & PDB_FOLLOW)
1445		printf("pmap_clear_modify(%x)", pa);
1446#endif
1447	pmap_changebit(pa, PG_M, FALSE);
1448}
1449
1450/*
1451 *	pmap_clear_reference:
1452 *
1453 *	Clear the reference bit on the specified physical page.
1454 */
1455
1456void pmap_clear_reference(pa)
1457	vm_offset_t	pa;
1458{
1459#ifdef DEBUG
1460	if (pmapdebug & PDB_FOLLOW)
1461		printf("pmap_clear_reference(%x)", pa);
1462#endif
1463	pmap_changebit(pa, PG_U, FALSE);
1464}
1465
1466/*
1467 *	pmap_is_referenced:
1468 *
1469 *	Return whether or not the specified physical page is referenced
1470 *	by any physical maps.
1471 */
1472
1473boolean_t
1474pmap_is_referenced(pa)
1475	vm_offset_t	pa;
1476{
1477#ifdef DEBUG
1478	if (pmapdebug & PDB_FOLLOW) {
1479		boolean_t rv = pmap_testbit(pa, PG_U);
1480		printf("pmap_is_referenced(%x) -> %c", pa, "FT"[rv]);
1481		return(rv);
1482	}
1483#endif
1484	return(pmap_testbit(pa, PG_U));
1485}
1486
1487/*
1488 *	pmap_is_modified:
1489 *
1490 *	Return whether or not the specified physical page is modified
1491 *	by any physical maps.
1492 */
1493
1494boolean_t
1495pmap_is_modified(pa)
1496	vm_offset_t	pa;
1497{
1498#ifdef DEBUG
1499	if (pmapdebug & PDB_FOLLOW) {
1500		boolean_t rv = pmap_testbit(pa, PG_M);
1501		printf("pmap_is_modified(%x) -> %c", pa, "FT"[rv]);
1502		return(rv);
1503	}
1504#endif
1505	return(pmap_testbit(pa, PG_M));
1506}
1507
1508vm_offset_t
1509pmap_phys_address(ppn)
1510	int ppn;
1511{
1512	return(i386_ptob(ppn));
1513}
1514
1515/*
1516 * Miscellaneous support routines follow
1517 */
1518
1519i386_protection_init()
1520{
1521	register int *kp, prot;
1522
1523	kp = protection_codes;
1524	for (prot = 0; prot < 8; prot++) {
1525		switch (prot) {
1526		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
1527			*kp++ = 0;
1528			break;
1529		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
1530		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
1531		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
1532			*kp++ = PG_RO;
1533			break;
1534		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
1535		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
1536		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
1537		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
1538			*kp++ = PG_RW;
1539			break;
1540		}
1541	}
1542}
1543
1544boolean_t
1545pmap_testbit(pa, bit)
1546	register vm_offset_t pa;
1547	int bit;
1548{
1549	register pv_entry_t pv;
1550	register int *pte, ix;
1551	int s;
1552
1553	if (pa < vm_first_phys || pa >= vm_last_phys)
1554		return(FALSE);
1555
1556	pv = pa_to_pvh(pa);
1557	s = splimp();
1558	/*
1559	 * Check saved info first
1560	 */
1561	if (pmap_attributes[pa_index(pa)] & bit) {
1562		splx(s);
1563		return(TRUE);
1564	}
1565	/*
1566	 * Not found, check current mappings returning
1567	 * immediately if found.
1568	 */
1569	if (pv->pv_pmap != NULL) {
1570		for (; pv; pv = pv->pv_next) {
1571			pte = (int *) pmap_pte(pv->pv_pmap, pv->pv_va);
1572			ix = 0;
1573			do {
1574				if (*pte++ & bit) {
1575					splx(s);
1576					return(TRUE);
1577				}
1578			} while (++ix != i386pagesperpage);
1579		}
1580	}
1581	splx(s);
1582	return(FALSE);
1583}
1584
1585pmap_changebit(pa, bit, setem)
1586	register vm_offset_t pa;
1587	int bit;
1588	boolean_t setem;
1589{
1590	register pv_entry_t pv;
1591	register int *pte, npte, ix;
1592	vm_offset_t va;
1593	int s;
1594	boolean_t firstpage = TRUE;
1595
1596#ifdef DEBUG
1597	if (pmapdebug & PDB_BITS)
1598		printf("pmap_changebit(%x, %x, %s)",
1599		       pa, bit, setem ? "set" : "clear");
1600#endif
1601	if (pa < vm_first_phys || pa >= vm_last_phys)
1602		return;
1603
1604	pv = pa_to_pvh(pa);
1605	s = splimp();
1606	/*
1607	 * Clear saved attributes (modify, reference)
1608	 */
1609	if (!setem)
1610		pmap_attributes[pa_index(pa)] &= ~bit;
1611	/*
1612	 * Loop over all current mappings setting/clearing as appropos
1613	 * If setting RO do we need to clear the VAC?
1614	 */
1615	if (pv->pv_pmap != NULL) {
1616#ifdef DEBUG
1617		int toflush = 0;
1618#endif
1619		for (; pv; pv = pv->pv_next) {
1620#ifdef DEBUG
1621			toflush |= (pv->pv_pmap == kernel_pmap) ? 2 : 1;
1622#endif
1623			va = pv->pv_va;
1624
1625                        /*
1626                         * XXX don't write protect pager mappings
1627                         */
1628                        if (bit == PG_RO) {
1629                                extern vm_offset_t pager_sva, pager_eva;
1630
1631                                if (va >= pager_sva && va < pager_eva)
1632                                        continue;
1633                        }
1634
1635			pte = (int *) pmap_pte(pv->pv_pmap, va);
1636			ix = 0;
1637			do {
1638				if (setem)
1639					npte = *pte | bit;
1640				else
1641					npte = *pte & ~bit;
1642				if (*pte != npte) {
1643					*pte = npte;
1644					/*TBIS(va);*/
1645				}
1646				va += I386_PAGE_SIZE;
1647				pte++;
1648			} while (++ix != i386pagesperpage);
1649
1650			if (curproc && pv->pv_pmap == &curproc->p_vmspace->vm_pmap)
1651				pmap_activate(pv->pv_pmap, (struct pcb *)curproc->p_addr);
1652		}
1653#ifdef somethinglikethis
1654		if (setem && bit == PG_RO && (pmapvacflush & PVF_PROTECT)) {
1655			if ((pmapvacflush & PVF_TOTAL) || toflush == 3)
1656				DCIA();
1657			else if (toflush == 2)
1658				DCIS();
1659			else
1660				DCIU();
1661		}
1662#endif
1663	}
1664	splx(s);
1665}
1666
1667#ifdef DEBUG
1668pmap_pvdump(pa)
1669	vm_offset_t pa;
1670{
1671	register pv_entry_t pv;
1672
1673	printf("pa %x", pa);
1674	for (pv = pa_to_pvh(pa); pv; pv = pv->pv_next) {
1675		printf(" -> pmap %x, va %x, flags %x",
1676		       pv->pv_pmap, pv->pv_va, pv->pv_flags);
1677		pads(pv->pv_pmap);
1678	}
1679	printf(" ");
1680}
1681
1682#ifdef notyet
1683pmap_check_wiring(str, va)
1684	char *str;
1685	vm_offset_t va;
1686{
1687	vm_map_entry_t entry;
1688	register int count, *pte;
1689
1690	va = trunc_page(va);
1691	if (!pmap_pde_v(pmap_pde(kernel_pmap, va)) ||
1692	    !pmap_pte_v(pmap_pte(kernel_pmap, va)))
1693		return;
1694
1695	if (!vm_map_lookup_entry(pt_map, va, &entry)) {
1696		pg("wired_check: entry for %x not found\n", va);
1697		return;
1698	}
1699	count = 0;
1700	for (pte = (int *)va; pte < (int *)(va+PAGE_SIZE); pte++)
1701		if (*pte)
1702			count++;
1703	if (entry->wired_count != count)
1704		pg("*%s*: %x: w%d/a%d\n",
1705		       str, va, entry->wired_count, count);
1706}
1707#endif
1708
1709/* print address space of pmap*/
1710pads(pm) pmap_t pm; {
1711	unsigned va, i, j;
1712	struct pte *ptep;
1713
1714	if(pm == kernel_pmap) return;
1715	for (i = 0; i < 1024; i++)
1716		if(pm->pm_pdir[i].pd_v)
1717			for (j = 0; j < 1024 ; j++) {
1718				va = (i<<22)+(j<<12);
1719				if (pm == kernel_pmap && va < 0xfe000000)
1720						continue;
1721				if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
1722						continue;
1723				ptep = pmap_pte(pm, va);
1724				if(pmap_pte_v(ptep))
1725					printf("%x:%x ", va, *(int *)ptep);
1726			} ;
1727
1728}
1729#endif
1730