pmap.c revision 1440
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 1.24 1994/04/20 07:06:14 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 "param.h"
89#include "systm.h"
90#include "proc.h"
91#include "malloc.h"
92#include "user.h"
93#include "i386/include/cpufunc.h"
94#include "i386/include/cputypes.h"
95
96#include "vm/vm.h"
97#include "vm/vm_kern.h"
98#include "vm/vm_page.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();
150inline void			pmap_use_pt();
151inline void			pmap_unuse_pt();
152inline pt_entry_t * const	pmap_pte();
153static inline pv_entry_t	get_pv_entry();
154void				pmap_alloc_pv_entry();
155void				pmap_clear_modify();
156void				i386_protection_init();
157extern vm_offset_t pager_sva, pager_eva;
158extern int cpu_class;
159
160#if BSDVM_COMPAT
161#include "msgbuf.h"
162
163/*
164 * All those kernel PT submaps that BSD is so fond of
165 */
166pt_entry_t *CMAP1, *CMAP2, *mmap;
167caddr_t		CADDR1, CADDR2, vmmap;
168pt_entry_t *msgbufmap;
169struct msgbuf	*msgbufp;
170#endif
171
172void init_pv_entries(int) ;
173
174/*
175 *	Routine:	pmap_pte
176 *	Function:
177 *		Extract the page table entry associated
178 *		with the given map/virtual_address pair.
179 * [ what about induced faults -wfj]
180 */
181
182inline pt_entry_t *
183const pmap_pte(pmap, va)
184	register pmap_t	pmap;
185	vm_offset_t va;
186{
187
188	if (pmap && *pmap_pde(pmap, va)) {
189		vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
190		/* are we current address space or kernel? */
191		if ( (pmap == kernel_pmap) || (frame == ((int) PTDpde & PG_FRAME)))
192			return ((pt_entry_t *) vtopte(va));
193		/* otherwise, we are alternate address space */
194		else {
195			if ( frame != ((int) APTDpde & PG_FRAME) ) {
196				APTDpde = pmap->pm_pdir[PTDPTDI];
197				tlbflush();
198			}
199			return((pt_entry_t *) avtopte(va));
200		}
201	}
202	return(0);
203}
204
205/*
206 *	Routine:	pmap_extract
207 *	Function:
208 *		Extract the physical page address associated
209 *		with the given map/virtual_address pair.
210 */
211
212vm_offset_t
213pmap_extract(pmap, va)
214	register pmap_t	pmap;
215	vm_offset_t va;
216{
217	pd_entry_t save;
218	vm_offset_t pa;
219	int s;
220
221	if (pmap && *pmap_pde(pmap, va)) {
222		vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
223		/* are we current address space or kernel? */
224		if ( (pmap == kernel_pmap)
225			|| (frame == ((int) PTDpde & PG_FRAME)) ) {
226			pa = *(int *) vtopte(va);
227		/* otherwise, we are alternate address space */
228		} else {
229			if ( frame != ((int) APTDpde & PG_FRAME)) {
230				APTDpde = pmap->pm_pdir[PTDPTDI];
231				tlbflush();
232			}
233			pa = *(int *) avtopte(va);
234		}
235		pa = (pa & PG_FRAME) | (va & ~PG_FRAME);
236		return pa;
237	}
238	return 0;
239
240}
241
242/*
243 * determine if a page is managed (memory vs. device)
244 */
245static inline int
246pmap_is_managed(pa)
247	vm_offset_t pa;
248{
249	int i;
250
251	if (!pmap_initialized)
252		return 0;
253
254	for (i = 0; phys_avail[i + 1]; i += 2) {
255		if (pa >= phys_avail[i] && pa < phys_avail[i + 1])
256			return 1;
257	}
258	return 0;
259}
260
261/*
262 * find the vm_page_t of a pte (only) given va of pte and pmap
263 */
264inline vm_page_t
265pmap_pte_vm_page(pmap, pt)
266	pmap_t pmap;
267	vm_offset_t pt;
268{
269	pt = i386_trunc_page( pt);
270	pt = (pt - UPT_MIN_ADDRESS) / NBPG;
271	pt = ((vm_offset_t) pmap->pm_pdir[pt]) & PG_FRAME;
272	return PHYS_TO_VM_PAGE(pt);
273}
274
275/*
276 * Wire a page table page
277 */
278inline void
279pmap_use_pt(pmap, va)
280	pmap_t pmap;
281	vm_offset_t va;
282{
283	vm_offset_t pt;
284
285	if (va >= VM_MAX_ADDRESS || !pmap_initialized)
286		return;
287
288	pt = (vm_offset_t) vtopte(va);
289	vm_page_hold( pmap_pte_vm_page(pmap, pt));
290}
291
292/*
293 * Unwire a page table page
294 */
295inline void
296pmap_unuse_pt(pmap, va)
297	pmap_t pmap;
298	vm_offset_t va;
299{
300	vm_offset_t pt;
301
302	if (va >= VM_MAX_ADDRESS || !pmap_initialized)
303		return;
304
305	pt = (vm_offset_t) vtopte(va);
306	vm_page_unhold( pmap_pte_vm_page(pmap, pt));
307}
308
309/* [ macro again?, should I force kstack into user map here? -wfj ] */
310void
311pmap_activate(pmap, pcbp)
312	register pmap_t pmap;
313	struct pcb *pcbp;
314{
315	PMAP_ACTIVATE(pmap, pcbp);
316}
317
318/*
319 *	Bootstrap the system enough to run with virtual memory.
320 *	Map the kernel's code and data, and allocate the system page table.
321 *
322 *	On the I386 this is called after mapping has already been enabled
323 *	and just syncs the pmap module with what has already been done.
324 *	[We can't call it easily with mapping off since the kernel is not
325 *	mapped with PA == VA, hence we would have to relocate every address
326 *	from the linked base (virtual) address "KERNBASE" to the actual
327 *	(physical) address starting relative to 0]
328 */
329
330#define DMAPAGES 8
331void
332pmap_bootstrap(firstaddr, loadaddr)
333	vm_offset_t firstaddr;
334	vm_offset_t loadaddr;
335{
336#if BSDVM_COMPAT
337	vm_offset_t va;
338	pt_entry_t *pte;
339#endif
340	extern int IdlePTD;
341
342	avail_start = firstaddr + DMAPAGES*NBPG;
343
344	virtual_avail = (vm_offset_t) KERNBASE + avail_start;
345	virtual_end = VM_MAX_KERNEL_ADDRESS;
346	i386pagesperpage = PAGE_SIZE / NBPG;
347
348	/*
349	 * Initialize protection array.
350	 */
351	i386_protection_init();
352
353	/*
354	 * The kernel's pmap is statically allocated so we don't
355	 * have to use pmap_create, which is unlikely to work
356	 * correctly at this part of the boot sequence.
357	 */
358	kernel_pmap = &kernel_pmap_store;
359
360	kernel_pmap->pm_pdir = (pd_entry_t *)(KERNBASE + IdlePTD);
361
362	simple_lock_init(&kernel_pmap->pm_lock);
363	kernel_pmap->pm_count = 1;
364
365#if BSDVM_COMPAT
366	/*
367	 * Allocate all the submaps we need
368	 */
369#define	SYSMAP(c, p, v, n)	\
370	v = (c)va; va += ((n)*NBPG); p = pte; pte += (n);
371
372	va = virtual_avail;
373	pte = pmap_pte(kernel_pmap, va);
374
375	SYSMAP(caddr_t		,CMAP1		,CADDR1	   ,1		)
376	SYSMAP(caddr_t		,CMAP2		,CADDR2	   ,1		)
377	SYSMAP(caddr_t		,mmap		,vmmap	   ,1		)
378	SYSMAP(struct msgbuf *	,msgbufmap	,msgbufp   ,1		)
379	virtual_avail = va;
380#endif
381	/*
382	 * reserve special hunk of memory for use by bus dma as a bounce
383	 * buffer (contiguous virtual *and* physical memory). for now,
384	 * assume vm does not use memory beneath hole, and we know that
385	 * the bootstrap uses top 32k of base memory. -wfj
386	 */
387	{
388		extern vm_offset_t isaphysmem;
389		isaphysmem = va;
390
391		virtual_avail = pmap_map(va, firstaddr,
392				firstaddr + DMAPAGES*NBPG, VM_PROT_ALL);
393	}
394
395	*(int *)PTD = 0;
396	tlbflush();
397
398}
399
400/*
401 *	Initialize the pmap module.
402 *	Called by vm_init, to initialize any structures that the pmap
403 *	system needs to map virtual memory.
404 *	pmap_init has been enhanced to support in a fairly consistant
405 *	way, discontiguous physical memory.
406 */
407void
408pmap_init(phys_start, phys_end)
409	vm_offset_t	phys_start, phys_end;
410{
411	vm_offset_t	addr, addr2;
412	vm_size_t	npg, s;
413	int		rv;
414	int i;
415	extern int KPTphys;
416	extern int IdlePTD;
417
418	/*
419	 * Now that kernel map has been allocated, we can mark as
420	 * unavailable regions which we have mapped in locore.
421	 */
422	addr = atdevbase;
423	(void) vm_map_find(kernel_map, NULL, (vm_offset_t) 0,
424			   &addr, (0x100000-0xa0000), FALSE);
425
426	addr = (vm_offset_t) KERNBASE + IdlePTD;
427	vm_object_reference(kernel_object);
428	(void) vm_map_find(kernel_map, kernel_object, addr,
429			   &addr, (4 + NKPT) * NBPG, FALSE);
430
431
432	/*
433	 * calculate the number of pv_entries needed
434	 */
435	vm_first_phys = phys_avail[0];
436	for (i = 0; phys_avail[i + 1]; i += 2) ;
437	npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / NBPG;
438
439	/*
440	 * Allocate memory for random pmap data structures.  Includes the
441	 * pv_head_table.
442	 */
443	s = (vm_size_t) (sizeof(struct pv_entry) * npg);
444	s = i386_round_page(s);
445	addr = (vm_offset_t) kmem_alloc(kernel_map, s);
446	pv_table = (pv_entry_t) addr;
447
448	/*
449	 * init the pv free list
450	 */
451	init_pv_entries(npg);
452	/*
453	 * Now it is safe to enable pv_table recording.
454	 */
455	pmap_initialized = TRUE;
456}
457
458/*
459 *	Used to map a range of physical addresses into kernel
460 *	virtual address space.
461 *
462 *	For now, VM is already on, we only need to map the
463 *	specified memory.
464 */
465vm_offset_t
466pmap_map(virt, start, end, prot)
467	vm_offset_t	virt;
468	vm_offset_t	start;
469	vm_offset_t	end;
470	int		prot;
471{
472	while (start < end) {
473		pmap_enter(kernel_pmap, virt, start, prot, FALSE);
474		virt += PAGE_SIZE;
475		start += PAGE_SIZE;
476	}
477	return(virt);
478}
479
480/*
481 *	Create and return a physical map.
482 *
483 *	If the size specified for the map
484 *	is zero, the map is an actual physical
485 *	map, and may be referenced by the
486 *	hardware.
487 *
488 *	If the size specified is non-zero,
489 *	the map will be used in software only, and
490 *	is bounded by that size.
491 *
492 * [ just allocate a ptd and mark it uninitialize -- should we track
493 *   with a table which process has which ptd? -wfj ]
494 */
495
496pmap_t
497pmap_create(size)
498	vm_size_t	size;
499{
500	register pmap_t pmap;
501
502	/*
503	 * Software use map does not need a pmap
504	 */
505	if (size)
506		return(NULL);
507
508	pmap = (pmap_t) malloc(sizeof *pmap, M_VMPMAP, M_WAITOK);
509	bzero(pmap, sizeof(*pmap));
510	pmap_pinit(pmap);
511	return (pmap);
512}
513
514
515struct pmaplist {
516	struct pmaplist *next;
517};
518
519static inline void *
520vm_get_pmap()
521{
522	struct pmaplist *rtval;
523
524	rtval = (struct pmaplist *)kmem_alloc(kernel_map, ctob(1));
525	bzero(rtval, ctob(1));
526	return rtval;
527}
528
529static inline void
530vm_put_pmap(up)
531	struct pmaplist *up;
532{
533	kmem_free(kernel_map, up, ctob(1));
534}
535
536/*
537 * Initialize a preallocated and zeroed pmap structure,
538 * such as one in a vmspace structure.
539 */
540void
541pmap_pinit(pmap)
542	register struct pmap *pmap;
543{
544	/*
545	 * No need to allocate page table space yet but we do need a
546	 * valid page directory table.
547	 */
548	pmap->pm_pdir = (pd_entry_t *) vm_get_pmap();
549
550	/* wire in kernel global address entries */
551	bcopy(PTD+KPTDI, pmap->pm_pdir+KPTDI, NKPT*PTESIZE);
552
553	/* install self-referential address mapping entry */
554	*(int *)(pmap->pm_pdir+PTDPTDI) =
555		((int)pmap_kextract((vm_offset_t)pmap->pm_pdir)) | PG_V | PG_KW;
556
557	pmap->pm_count = 1;
558	simple_lock_init(&pmap->pm_lock);
559}
560
561/*
562 *	Retire the given physical map from service.
563 *	Should only be called if the map contains
564 *	no valid mappings.
565 */
566void
567pmap_destroy(pmap)
568	register pmap_t pmap;
569{
570	int count;
571
572	if (pmap == NULL)
573		return;
574
575	simple_lock(&pmap->pm_lock);
576	count = --pmap->pm_count;
577	simple_unlock(&pmap->pm_lock);
578	if (count == 0) {
579		pmap_release(pmap);
580		free((caddr_t)pmap, M_VMPMAP);
581	}
582}
583
584/*
585 * Release any resources held by the given physical map.
586 * Called when a pmap initialized by pmap_pinit is being released.
587 * Should only be called if the map contains no valid mappings.
588 */
589void
590pmap_release(pmap)
591	register struct pmap *pmap;
592{
593	vm_put_pmap((struct pmaplist *) pmap->pm_pdir);
594}
595
596/*
597 *	Add a reference to the specified pmap.
598 */
599void
600pmap_reference(pmap)
601	pmap_t	pmap;
602{
603	if (pmap != NULL) {
604		simple_lock(&pmap->pm_lock);
605		pmap->pm_count++;
606		simple_unlock(&pmap->pm_lock);
607	}
608}
609
610#define PV_FREELIST_MIN ((NBPG / sizeof (struct pv_entry)) / 2)
611
612/*
613 * Data for the pv entry allocation mechanism
614 */
615int pv_freelistcnt;
616pv_entry_t pv_freelist;
617vm_offset_t pvva;
618int npvvapg;
619
620/*
621 * free the pv_entry back to the free list
622 */
623inline static void
624free_pv_entry(pv)
625	pv_entry_t pv;
626{
627	if (!pv) return;
628	++pv_freelistcnt;
629	pv->pv_next = pv_freelist;
630	pv_freelist = pv;
631}
632
633/*
634 * get a new pv_entry, allocating a block from the system
635 * when needed.
636 * the memory allocation is performed bypassing the malloc code
637 * because of the possibility of allocations at interrupt time.
638 */
639static inline pv_entry_t
640get_pv_entry()
641{
642	pv_entry_t tmp;
643
644	/*
645	 * get more pv_entry pages if needed
646	 */
647	while (pv_freelistcnt < PV_FREELIST_MIN || pv_freelist == 0) {
648		pmap_alloc_pv_entry();
649	}
650
651	/*
652	 * get a pv_entry off of the free list
653	 */
654	--pv_freelistcnt;
655	tmp = pv_freelist;
656	pv_freelist = tmp->pv_next;
657	tmp->pv_pmap = 0;
658	tmp->pv_va = 0;
659	tmp->pv_next = 0;
660	return tmp;
661}
662
663/*
664 * this *strange* allocation routine *statistically* eliminates the
665 * *possibility* of a malloc failure (*FATAL*) for a pv_entry_t data structure.
666 * also -- this code is MUCH MUCH faster than the malloc equiv...
667 */
668void
669pmap_alloc_pv_entry()
670{
671	/*
672	 * do we have any pre-allocated map-pages left?
673	 */
674	if (npvvapg) {
675		vm_page_t m;
676		/*
677		 * we do this to keep recursion away
678		 */
679		pv_freelistcnt += PV_FREELIST_MIN;
680		/*
681		 * allocate a physical page out of the vm system
682		 */
683		if (m = vm_page_alloc(kernel_object, pvva-vm_map_min(kernel_map))) {
684			int newentries;
685			int i;
686			pv_entry_t entry;
687			newentries = (NBPG/sizeof (struct pv_entry));
688			/*
689			 * wire the page
690			 */
691			vm_page_wire(m);
692			m->flags &= ~PG_BUSY;
693			/*
694			 * let the kernel see it
695			 */
696			pmap_enter(vm_map_pmap(kernel_map), pvva,
697				VM_PAGE_TO_PHYS(m), VM_PROT_DEFAULT,1);
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 = splimp();
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) && (sva < USRSTACK || sva > UPT_MAX_ADDRESS))
853				|| (sva >= USRSTACK && sva < USRSTACK+(UPAGES*NBPG))) {
854				if (sva < pager_sva || sva >= pager_eva) {
855					m = PHYS_TO_VM_PAGE(pa);
856					m->flags &= ~PG_CLEAN;
857				}
858			}
859
860			pv = pa_to_pvh(pa);
861			pmap_remove_entry(pmap, pv, sva);
862			pmap_unuse_pt(pmap, sva);
863		}
864		tlbflush();
865		return;
866	}
867
868	sva = i386_btop(sva);
869	eva = i386_btop(eva);
870
871	while (sva < eva) {
872		/*
873		 * Weed out invalid mappings.
874		 * Note: we assume that the page directory table is
875	 	 * always allocated, and in kernel virtual.
876		 */
877
878		if ( *pmap_pde(pmap, i386_ptob(sva)) == 0 ) {
879			/* We can race ahead here, straight to next pde.. */
880	nextpde:
881			sva = ((sva + NPTEPG) & ~(NPTEPG - 1));
882			continue;
883		}
884
885		ptq = ptp + sva;
886
887		/*
888		 * search for page table entries, use string operations
889		 * that are much faster than
890		 * explicitly scanning when page tables are not fully
891		 * populated.
892		 */
893		if ( *ptq == 0) {
894			vm_offset_t pdnxt = ((sva + NPTEPG) & ~(NPTEPG - 1));
895			vm_offset_t nscan = pdnxt - sva;
896			int found = 0;
897
898			if ((nscan + sva) > eva)
899				nscan = eva - sva;
900
901			asm("xorl %%eax,%%eax;cld;repe;scasl;jz 1f;incl %%eax;1:;"
902				:"=D"(ptq),"=a"(found)
903				:"c"(nscan),"0"(ptq)
904				:"cx");
905
906			if( !found) {
907				sva = pdnxt;
908				continue;
909			}
910			ptq -= 1;
911
912			sva = ptq - ptp;
913		}
914
915		/*
916		 * Update statistics
917		 */
918		oldpte = *ptq;
919		if (((int)oldpte) & PG_W)
920			pmap->pm_stats.wired_count--;
921		pmap->pm_stats.resident_count--;
922
923		/*
924		 * Invalidate the PTEs.
925		 * XXX: should cluster them up and invalidate as many
926		 * as possible at once.
927		 */
928		*ptq = 0;
929
930		va = i386_ptob(sva);
931
932		/*
933		 * Remove from the PV table (raise IPL since we
934		 * may be called at interrupt time).
935		 */
936		pa = ((int)oldpte) & PG_FRAME;
937		if (!pmap_is_managed(pa)) {
938			++sva;
939			continue;
940		}
941
942		if ((((int) oldpte & PG_M) && (va < USRSTACK || va > UPT_MAX_ADDRESS))
943			|| (va >= USRSTACK && va < USRSTACK+(UPAGES*NBPG))) {
944			if (va < pager_sva || va >= pager_eva) {
945				m = PHYS_TO_VM_PAGE(pa);
946				m->flags &= ~PG_CLEAN;
947			}
948		}
949
950		pv = pa_to_pvh(pa);
951		pmap_remove_entry(pmap, pv, va);
952		pmap_unuse_pt(pmap, va);
953		++sva;
954	}
955	tlbflush();
956}
957
958/*
959 *	Routine:	pmap_remove_all
960 *	Function:
961 *		Removes this physical page from
962 *		all physical maps in which it resides.
963 *		Reflects back modify bits to the pager.
964 *
965 *	Notes:
966 *		Original versions of this routine were very
967 *		inefficient because they iteratively called
968 *		pmap_remove (slow...)
969 */
970void
971pmap_remove_all(pa)
972	vm_offset_t pa;
973{
974	register pv_entry_t pv, npv;
975	register pt_entry_t *pte, *ptp;
976	vm_offset_t va;
977	struct pmap *pmap;
978	struct map *map;
979	vm_page_t m;
980	int s;
981
982	/*
983	 * Not one of ours
984	 */
985	if (!pmap_is_managed(pa))
986		return;
987
988	pa = i386_trunc_page(pa);
989	pv = pa_to_pvh(pa);
990	m = PHYS_TO_VM_PAGE(pa);
991
992	s = splimp();
993	while (pv->pv_pmap != NULL) {
994		pmap = pv->pv_pmap;
995		ptp = get_pt_entry(pmap);
996		va = i386_btop(pv->pv_va);
997		pte = ptp + va;
998		if (pmap_pte_w(pte))
999			pmap->pm_stats.wired_count--;
1000		if ( *pte)
1001			pmap->pm_stats.resident_count--;
1002
1003		/*
1004		 * update the vm_page_t clean bit
1005		 */
1006		if ( (m->flags & PG_CLEAN) &&
1007			((((int) *pte) & PG_M) && (pv->pv_va < USRSTACK || pv->pv_va > UPT_MAX_ADDRESS))
1008			|| (pv->pv_va >= USRSTACK && pv->pv_va < USRSTACK+(UPAGES*NBPG))) {
1009			if (pv->pv_va < pager_sva || pv->pv_va >= pager_eva) {
1010				m->flags &= ~PG_CLEAN;
1011			}
1012		}
1013
1014		*pte = 0;
1015		pmap_unuse_pt(pmap, pv->pv_va);
1016
1017		npv = pv->pv_next;
1018		if (npv) {
1019			*pv = *npv;
1020			free_pv_entry(npv);
1021		} else {
1022			pv->pv_pmap = NULL;
1023		}
1024	}
1025	splx(s);
1026	tlbflush();
1027}
1028
1029
1030/*
1031 *	Set the physical protection on the
1032 *	specified range of this map as requested.
1033 */
1034void
1035pmap_protect(pmap, sva, eva, prot)
1036	register pmap_t	pmap;
1037	vm_offset_t	sva, eva;
1038	vm_prot_t	prot;
1039{
1040	register pt_entry_t *pte;
1041	register vm_offset_t va;
1042	int i386prot;
1043	register pt_entry_t *ptp;
1044	int evap = i386_btop(eva);
1045	int s;
1046
1047	if (pmap == NULL)
1048		return;
1049
1050	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1051		pmap_remove(pmap, sva, eva);
1052		return;
1053	}
1054	if (prot & VM_PROT_WRITE)
1055		return;
1056
1057	ptp = get_pt_entry(pmap);
1058
1059	va = sva;
1060	while (va < eva) {
1061		int found=0;
1062		int svap;
1063		vm_offset_t nscan;
1064		/*
1065		 * Page table page is not allocated.
1066		 * Skip it, we don't want to force allocation
1067		 * of unnecessary PTE pages just to set the protection.
1068		 */
1069		if (! *pmap_pde(pmap, va)) {
1070			/* XXX: avoid address wrap around */
1071nextpde:
1072			if (va >= i386_trunc_pdr((vm_offset_t)-1))
1073				break;
1074			va = i386_round_pdr(va + PAGE_SIZE);
1075			continue;
1076		}
1077
1078		pte = ptp + i386_btop(va);
1079
1080		if( *pte == 0) {
1081		/*
1082		 * scan for a non-empty pte
1083		 */
1084			svap = pte - ptp;
1085			nscan = ((svap + NPTEPG) & ~(NPTEPG - 1)) - svap;
1086
1087			if (nscan + svap > evap)
1088				nscan = evap - svap;
1089
1090			found = 0;
1091			if (nscan)
1092				asm("xorl %%eax,%%eax;cld;repe;scasl;jz 1f;incl %%eax;1:;"
1093					:"=D"(pte),"=a"(found)
1094					:"c"(nscan),"0"(pte):"cx");
1095
1096			if( !found)
1097				goto nextpde;
1098
1099			pte -= 1;
1100			svap = pte - ptp;
1101
1102			va = i386_ptob(svap);
1103		}
1104
1105		i386prot = pte_prot(pmap, prot);
1106		if (va < UPT_MAX_ADDRESS) {
1107			i386prot |= PG_u;
1108			if( va >= UPT_MIN_ADDRESS)
1109				i386prot |= PG_RW;
1110		}
1111		pmap_pte_set_prot(pte, i386prot);
1112		va += PAGE_SIZE;
1113	}
1114	tlbflush();
1115}
1116
1117/*
1118 *	Insert the given physical page (p) at
1119 *	the specified virtual address (v) in the
1120 *	target physical map with the protection requested.
1121 *
1122 *	If specified, the page will be wired down, meaning
1123 *	that the related pte can not be reclaimed.
1124 *
1125 *	NB:  This is the only routine which MAY NOT lazy-evaluate
1126 *	or lose information.  That is, this routine must actually
1127 *	insert this page into the given map NOW.
1128 */
1129void
1130pmap_enter(pmap, va, pa, prot, wired)
1131	register pmap_t pmap;
1132	vm_offset_t va;
1133	register vm_offset_t pa;
1134	vm_prot_t prot;
1135	boolean_t wired;
1136{
1137	register pt_entry_t *pte;
1138	register pt_entry_t npte;
1139	vm_offset_t opa;
1140	int cacheable=1;
1141
1142	if (pmap == NULL)
1143		return;
1144
1145	va = i386_trunc_page(va);
1146	pa = i386_trunc_page(pa);
1147	if (va > VM_MAX_KERNEL_ADDRESS)panic("pmap_enter: toobig");
1148
1149	/*
1150	 * Page Directory table entry not valid, we need a new PT page
1151	 */
1152	if ( *pmap_pde(pmap, va) == 0) {
1153		pg("ptdi %x, va %x", pmap->pm_pdir[PTDPTDI], va);
1154	}
1155
1156	pte = pmap_pte(pmap, va);
1157	opa = pmap_pte_pa(pte);
1158
1159	/*
1160	 * Mapping has not changed, must be protection or wiring change.
1161	 */
1162	if (opa == pa) {
1163		/*
1164		 * Wiring change, just update stats.
1165		 * We don't worry about wiring PT pages as they remain
1166		 * resident as long as there are valid mappings in them.
1167		 * Hence, if a user page is wired, the PT page will be also.
1168		 */
1169		if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
1170			if (wired)
1171				pmap->pm_stats.wired_count++;
1172			else
1173				pmap->pm_stats.wired_count--;
1174		}
1175		goto validate;
1176	}
1177
1178	/*
1179	 * Mapping has changed, invalidate old range and fall through to
1180	 * handle validating new mapping.
1181	 */
1182	if (opa) {
1183		pmap_remove(pmap, va, va + PAGE_SIZE);
1184	}
1185
1186	/*
1187	 * Enter on the PV list if part of our managed memory
1188	 * Note that we raise IPL while manipulating pv_table
1189	 * since pmap_enter can be called at interrupt time.
1190	 */
1191	if (pmap_is_managed(pa)) {
1192		register pv_entry_t pv, npv;
1193		int s;
1194
1195		pv = pa_to_pvh(pa);
1196		s = splimp();
1197		/*
1198		 * No entries yet, use header as the first entry
1199		 */
1200		if (pv->pv_pmap == NULL) {
1201			pv->pv_va = va;
1202			pv->pv_pmap = pmap;
1203			pv->pv_next = NULL;
1204		}
1205		/*
1206		 * There is at least one other VA mapping this page.
1207		 * Place this entry after the header.
1208		 */
1209		else {
1210			npv = get_pv_entry();
1211			npv->pv_va = va;
1212			npv->pv_pmap = pmap;
1213			npv->pv_next = pv->pv_next;
1214			pv->pv_next = npv;
1215		}
1216		splx(s);
1217		cacheable = 1;
1218	} else {
1219		cacheable = 0;
1220	}
1221
1222	pmap_use_pt(pmap, va);
1223
1224	/*
1225	 * Increment counters
1226	 */
1227	pmap->pm_stats.resident_count++;
1228	if (wired)
1229		pmap->pm_stats.wired_count++;
1230
1231validate:
1232	/*
1233	 * Now validate mapping with desired protection/wiring.
1234	 */
1235	npte = (pt_entry_t) ( (int) (pa | pte_prot(pmap, prot) | PG_V));
1236	/*
1237	 * for correctness:
1238	 */
1239	if( !cacheable)
1240		(int) npte |= PG_N;
1241
1242	/*
1243	 * When forking (copy-on-write, etc):
1244	 * A process will turn off write permissions for any of its writable
1245	 * pages.  If the data (object) is only referred to by one process, the
1246	 * processes map is modified directly as opposed to using the
1247	 * object manipulation routine.  When using pmap_protect, the
1248	 * modified bits are not kept in the vm_page_t data structure.
1249	 * Therefore, when using pmap_enter in vm_fault to bring back
1250	 * writability of a page, there has been no memory of the
1251	 * modified or referenced bits except at the pte level.
1252	 * this clause supports the carryover of the modified and
1253	 * used (referenced) bits.
1254	 */
1255	if (pa == opa)
1256		(int) npte |= (int) *pte & (PG_M|PG_U);
1257
1258
1259	if (wired)
1260		(int) npte |= PG_W;
1261	if (va < UPT_MIN_ADDRESS)
1262		(int) npte |= PG_u;
1263	else if (va < UPT_MAX_ADDRESS)
1264		(int) npte |= PG_u | PG_RW;
1265
1266	if( *pte != npte) {
1267		*pte = npte;
1268		tlbflush();
1269	}
1270}
1271
1272/*
1273 * add a wired page to the kva
1274 */
1275void
1276pmap_kenter(va, pa)
1277	vm_offset_t va;
1278	register vm_offset_t pa;
1279{
1280	register pt_entry_t *pte;
1281	register pv_entry_t pv, npv;
1282	vm_offset_t opa;
1283	int s;
1284
1285	/*
1286	 * Enter on the PV list if part of our managed memory
1287	 * Note that we raise IPL while manipulating pv_table
1288	 * since pmap_enter can be called at interrupt time.
1289	 */
1290
1291	pte = vtopte(va);
1292
1293	opa = pmap_pte_pa(pte);
1294	/*
1295	 * Mapping has not changed, must be protection or wiring change.
1296	 */
1297	if (opa == pa) {
1298		/*
1299		 * Wiring change, just update stats.
1300		 * We don't worry about wiring PT pages as they remain
1301		 * resident as long as there are valid mappings in them.
1302		 * Hence, if a user page is wired, the PT page will be also.
1303		 */
1304		if (!pmap_pte_w(pte)) {
1305			kernel_pmap->pm_stats.wired_count++;
1306		}
1307		goto validate;
1308	}
1309
1310	if (opa) {
1311		pmap_remove(kernel_pmap, va, va + PAGE_SIZE);
1312	}
1313
1314	pv = pa_to_pvh(pa);
1315	s = splimp();
1316	/*
1317	 * No entries yet, use header as the first entry
1318	 */
1319	if (pv->pv_pmap == NULL) {
1320		pv->pv_va = va;
1321		pv->pv_pmap = kernel_pmap;
1322		pv->pv_next = NULL;
1323	}
1324	/*
1325	 * There is at least one other VA mapping this page.
1326	 * Place this entry after the header.
1327	 */
1328	else {
1329		npv = get_pv_entry();
1330		npv->pv_va = va;
1331		npv->pv_pmap = kernel_pmap;
1332		npv->pv_next = pv->pv_next;
1333		pv->pv_next = npv;
1334	}
1335	splx(s);
1336
1337	/*
1338	 * Increment counters
1339	 */
1340	kernel_pmap->pm_stats.resident_count++;
1341
1342validate:
1343
1344	/*
1345	 * Now validate mapping with desired protection/wiring.
1346	 */
1347	*pte = (pt_entry_t) ( (int) (pa | PG_RW | PG_V | PG_W));
1348}
1349
1350/*
1351 * this code makes some *MAJOR* assumptions:
1352 * 1. Current pmap & pmap exists.
1353 * 2. Not wired.
1354 * 3. Read access.
1355 * 4. No page table pages.
1356 * 5. Tlbflush is deferred to calling procedure.
1357 * 6. Page IS managed.
1358 * but is *MUCH* faster than pmap_enter...
1359 */
1360
1361static inline void
1362pmap_enter_quick(pmap, va, pa)
1363	register pmap_t pmap;
1364	vm_offset_t va;
1365	register vm_offset_t pa;
1366{
1367	register pt_entry_t *pte;
1368	register pv_entry_t pv, npv;
1369	int s;
1370
1371	/*
1372	 * Enter on the PV list if part of our managed memory
1373	 * Note that we raise IPL while manipulating pv_table
1374	 * since pmap_enter can be called at interrupt time.
1375	 */
1376
1377	pte = vtopte(va);
1378	if (pmap_pte_pa(pte)) {
1379		pmap_remove(pmap, va, va + PAGE_SIZE);
1380	}
1381
1382	pv = pa_to_pvh(pa);
1383	s = splimp();
1384	/*
1385	 * No entries yet, use header as the first entry
1386	 */
1387	if (pv->pv_pmap == NULL) {
1388		pv->pv_va = va;
1389		pv->pv_pmap = pmap;
1390		pv->pv_next = NULL;
1391	}
1392	/*
1393	 * There is at least one other VA mapping this page.
1394	 * Place this entry after the header.
1395	 */
1396	else {
1397		npv = get_pv_entry();
1398		npv->pv_va = va;
1399		npv->pv_pmap = pmap;
1400		npv->pv_next = pv->pv_next;
1401		pv->pv_next = npv;
1402	}
1403	splx(s);
1404
1405	pmap_use_pt(pmap, va);
1406
1407	/*
1408	 * Increment counters
1409	 */
1410	pmap->pm_stats.resident_count++;
1411
1412validate:
1413
1414	/*
1415	 * Now validate mapping with desired protection/wiring.
1416	 */
1417	*pte = (pt_entry_t) ( (int) (pa | PG_RO | PG_V | PG_u));
1418}
1419
1420/*
1421 * pmap_object_init_pt preloads the ptes for a given object
1422 * into the specified pmap.  This eliminates the blast of soft
1423 * faults on process startup and immediately after an mmap.
1424 */
1425void
1426pmap_object_init_pt(pmap, addr, object, offset, size)
1427	pmap_t pmap;
1428	vm_offset_t addr;
1429	vm_object_t object;
1430	vm_offset_t offset;
1431	vm_offset_t size;
1432{
1433
1434	vm_offset_t tmpoff;
1435	vm_page_t p;
1436	int s;
1437	vm_offset_t v, lastv=0;
1438	pt_entry_t pte;
1439	extern vm_map_t kernel_map;
1440	vm_offset_t objbytes;
1441
1442	if (!pmap)
1443		return;
1444
1445	/*
1446	 * if we are processing a major portion of the object, then
1447	 * scan the entire thing.
1448	 */
1449	if( size > object->size / 2) {
1450		objbytes = size;
1451		p = (vm_page_t) queue_first(&object->memq);
1452		while (!queue_end(&object->memq, (queue_entry_t) p) && objbytes != 0) {
1453			tmpoff = p->offset;
1454			if( tmpoff < offset) {
1455				p = (vm_page_t) queue_next(&p->listq);
1456				continue;
1457			}
1458			tmpoff -= offset;
1459			if( tmpoff >= size) {
1460				p = (vm_page_t) queue_next(&p->listq);
1461				continue;
1462			}
1463
1464			if ((p->flags & (PG_BUSY|PG_FICTITIOUS)) == 0 ) {
1465				vm_page_hold(p);
1466				v = i386_trunc_page(((vm_offset_t)vtopte( addr+tmpoff)));
1467				/* a fault might occur here */
1468				*(volatile char *)v += 0;
1469				vm_page_unhold(p);
1470				pmap_enter_quick(pmap, addr+tmpoff, VM_PAGE_TO_PHYS(p));
1471			}
1472			p = (vm_page_t) queue_next(&p->listq);
1473			objbytes -= NBPG;
1474		}
1475	} else {
1476	/*
1477	 * else lookup the pages one-by-one.
1478	 */
1479		for(tmpoff = 0; tmpoff < size; tmpoff += NBPG) {
1480			if( p = vm_page_lookup(object, tmpoff + offset)) {
1481				if( (p->flags & (PG_BUSY|PG_FICTITIOUS)) == 0) {
1482					vm_page_hold(p);
1483					v = i386_trunc_page(((vm_offset_t)vtopte( addr+tmpoff)));
1484					/* a fault might occur here */
1485					*(volatile char *)v += 0;
1486					vm_page_unhold(p);
1487					pmap_enter_quick(pmap, addr+tmpoff, VM_PAGE_TO_PHYS(p));
1488				}
1489			}
1490		}
1491	}
1492
1493	tlbflush();
1494}
1495
1496/*
1497 *	Routine:	pmap_change_wiring
1498 *	Function:	Change the wiring attribute for a map/virtual-address
1499 *			pair.
1500 *	In/out conditions:
1501 *			The mapping must already exist in the pmap.
1502 */
1503void
1504pmap_change_wiring(pmap, va, wired)
1505	register pmap_t	pmap;
1506	vm_offset_t	va;
1507	boolean_t	wired;
1508{
1509	register pt_entry_t *pte;
1510
1511	if (pmap == NULL)
1512		return;
1513
1514	pte = pmap_pte(pmap, va);
1515	if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
1516		if (wired)
1517			pmap->pm_stats.wired_count++;
1518		else
1519			pmap->pm_stats.wired_count--;
1520	}
1521	/*
1522	 * Wiring is not a hardware characteristic so there is no need
1523	 * to invalidate TLB.
1524	 */
1525	pmap_pte_set_w(pte, wired);
1526	/*
1527 	 * When unwiring, set the modified bit in the pte -- could have
1528	 * been changed by the kernel
1529 	 */
1530	if (!wired)
1531		(int) *pte |= PG_M;
1532}
1533
1534
1535
1536/*
1537 *	Copy the range specified by src_addr/len
1538 *	from the source map to the range dst_addr/len
1539 *	in the destination map.
1540 *
1541 *	This routine is only advisory and need not do anything.
1542 */
1543void
1544pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
1545	pmap_t dst_pmap, src_pmap;
1546	vm_offset_t	dst_addr;
1547	vm_size_t	len;
1548	vm_offset_t	src_addr;
1549{
1550}
1551/*
1552 *	Require that all active physical maps contain no
1553 *	incorrect entries NOW.  [This update includes
1554 *	forcing updates of any address map caching.]
1555 *
1556 *	Generally used to insure that a thread about
1557 *	to run will see a semantically correct world.
1558 */
1559void
1560pmap_update()
1561{
1562	tlbflush();
1563}
1564
1565/*
1566 *	Routine:	pmap_kernel
1567 *	Function:
1568 *		Returns the physical map handle for the kernel.
1569 */
1570pmap_t
1571pmap_kernel()
1572{
1573    	return (kernel_pmap);
1574}
1575
1576/*
1577 *	pmap_zero_page zeros the specified (machine independent)
1578 *	page by mapping the page into virtual memory and using
1579 *	bzero to clear its contents, one machine dependent page
1580 *	at a time.
1581 */
1582void
1583pmap_zero_page(phys)
1584	vm_offset_t phys;
1585{
1586	*(int *)CMAP2 = PG_V | PG_KW | i386_trunc_page(phys);
1587	tlbflush();
1588	bzero(CADDR2,NBPG);
1589}
1590
1591/*
1592 *	pmap_copy_page copies the specified (machine independent)
1593 *	page by mapping the page into virtual memory and using
1594 *	bcopy to copy the page, one machine dependent page at a
1595 *	time.
1596 */
1597void
1598pmap_copy_page(src, dst)
1599	vm_offset_t src;
1600	vm_offset_t dst;
1601{
1602	*(int *)CMAP1 = PG_V | PG_KW | i386_trunc_page(src);
1603	*(int *)CMAP2 = PG_V | PG_KW | i386_trunc_page(dst);
1604	tlbflush();
1605
1606#if __GNUC__ > 1
1607	memcpy(CADDR2, CADDR1, NBPG);
1608#else
1609	bcopy(CADDR1, CADDR2, NBPG);
1610#endif
1611}
1612
1613
1614/*
1615 *	Routine:	pmap_pageable
1616 *	Function:
1617 *		Make the specified pages (by pmap, offset)
1618 *		pageable (or not) as requested.
1619 *
1620 *		A page which is not pageable may not take
1621 *		a fault; therefore, its page table entry
1622 *		must remain valid for the duration.
1623 *
1624 *		This routine is merely advisory; pmap_enter
1625 *		will specify that these pages are to be wired
1626 *		down (or not) as appropriate.
1627 */
1628void
1629pmap_pageable(pmap, sva, eva, pageable)
1630	pmap_t		pmap;
1631	vm_offset_t	sva, eva;
1632	boolean_t	pageable;
1633{
1634}
1635
1636/*
1637 * this routine returns true if a physical page resides
1638 * in the given pmap.
1639 */
1640boolean_t
1641pmap_page_exists(pmap, pa)
1642	pmap_t pmap;
1643	vm_offset_t pa;
1644{
1645	register pv_entry_t pv;
1646	int s;
1647
1648	if (!pmap_is_managed(pa))
1649		return FALSE;
1650
1651	pv = pa_to_pvh(pa);
1652	s = splimp();
1653
1654	/*
1655	 * Not found, check current mappings returning
1656	 * immediately if found.
1657	 */
1658	if (pv->pv_pmap != NULL) {
1659		for (; pv; pv = pv->pv_next) {
1660			if (pv->pv_pmap == pmap) {
1661				splx(s);
1662				return TRUE;
1663			}
1664		}
1665	}
1666	splx(s);
1667	return(FALSE);
1668}
1669
1670/*
1671 * pmap_testbit tests bits in pte's
1672 * note that the testbit/changebit routines are inline,
1673 * and a lot of things compile-time evaluate.
1674 */
1675static inline boolean_t
1676pmap_testbit(pa, bit)
1677	register vm_offset_t pa;
1678	int bit;
1679{
1680	register pv_entry_t pv;
1681	pt_entry_t *pte;
1682	int s;
1683
1684	if (!pmap_is_managed(pa))
1685		return FALSE;
1686
1687	pv = pa_to_pvh(pa);
1688	s = splimp();
1689
1690	/*
1691	 * Not found, check current mappings returning
1692	 * immediately if found.
1693	 */
1694	if (pv->pv_pmap != NULL) {
1695		for (; pv; pv = pv->pv_next) {
1696			/*
1697			 * if the bit being tested is the modified bit,
1698			 * then mark UPAGES as always modified, and
1699			 * ptes as never modified.
1700			 */
1701			if (bit & PG_U ) {
1702				if ((pv->pv_va >= pager_sva) && (pv->pv_va < pager_eva)) {
1703					continue;
1704				}
1705			}
1706			if (bit & PG_M ) {
1707				if (pv->pv_va >= USRSTACK) {
1708					if (pv->pv_va >= pager_sva && pv->pv_va < pager_eva) {
1709						continue;
1710					}
1711					if (pv->pv_va < USRSTACK+(UPAGES*NBPG)) {
1712						splx(s);
1713						return TRUE;
1714					}
1715					else if (pv->pv_va < UPT_MAX_ADDRESS) {
1716						splx(s);
1717						return FALSE;
1718					}
1719				}
1720			}
1721			pte = pmap_pte(pv->pv_pmap, pv->pv_va);
1722			if ((int) *pte & bit) {
1723				splx(s);
1724				return TRUE;
1725			}
1726		}
1727	}
1728	splx(s);
1729	return(FALSE);
1730}
1731
1732/*
1733 * this routine is used to modify bits in ptes
1734 */
1735static inline void
1736pmap_changebit(pa, bit, setem)
1737	vm_offset_t pa;
1738	int bit;
1739	boolean_t setem;
1740{
1741	register pv_entry_t pv;
1742	register pt_entry_t *pte, npte;
1743	vm_offset_t va;
1744	int s;
1745
1746	if (!pmap_is_managed(pa))
1747		return;
1748
1749	pv = pa_to_pvh(pa);
1750	s = splimp();
1751
1752	/*
1753	 * Loop over all current mappings setting/clearing as appropos
1754	 * If setting RO do we need to clear the VAC?
1755	 */
1756	if (pv->pv_pmap != NULL) {
1757		for (; pv; pv = pv->pv_next) {
1758			va = pv->pv_va;
1759
1760			/*
1761			 * don't write protect pager mappings
1762			 */
1763			if (!setem && (bit == PG_RW)) {
1764				if (va >= pager_sva && va < pager_eva)
1765					continue;
1766			}
1767
1768			pte = pmap_pte(pv->pv_pmap, va);
1769			if (setem)
1770				(int) npte = (int) *pte | bit;
1771			else
1772				(int) npte = (int) *pte & ~bit;
1773			*pte = npte;
1774		}
1775	}
1776	splx(s);
1777	tlbflush();
1778}
1779
1780/*
1781 *      pmap_page_protect:
1782 *
1783 *      Lower the permission for all mappings to a given page.
1784 */
1785void
1786pmap_page_protect(phys, prot)
1787        vm_offset_t     phys;
1788        vm_prot_t       prot;
1789{
1790	if ((prot & VM_PROT_WRITE) == 0) {
1791		if (prot & (VM_PROT_READ | VM_PROT_EXECUTE))
1792			pmap_changebit(phys, PG_RW, FALSE);
1793		else
1794			pmap_remove_all(phys);
1795	}
1796}
1797
1798/*
1799 *	Clear the modify bits on the specified physical page.
1800 */
1801void
1802pmap_clear_modify(pa)
1803	vm_offset_t	pa;
1804{
1805	pmap_changebit(pa, PG_M, FALSE);
1806}
1807
1808/*
1809 *	pmap_clear_reference:
1810 *
1811 *	Clear the reference bit on the specified physical page.
1812 */
1813void
1814pmap_clear_reference(pa)
1815	vm_offset_t	pa;
1816{
1817	pmap_changebit(pa, PG_U, FALSE);
1818}
1819
1820/*
1821 *	pmap_is_referenced:
1822 *
1823 *	Return whether or not the specified physical page is referenced
1824 *	by any physical maps.
1825 */
1826
1827boolean_t
1828pmap_is_referenced(pa)
1829	vm_offset_t	pa;
1830{
1831	return(pmap_testbit(pa, PG_U));
1832}
1833
1834/*
1835 *	pmap_is_modified:
1836 *
1837 *	Return whether or not the specified physical page is modified
1838 *	by any physical maps.
1839 */
1840
1841boolean_t
1842pmap_is_modified(pa)
1843	vm_offset_t	pa;
1844{
1845	return(pmap_testbit(pa, PG_M));
1846}
1847
1848/*
1849 *	Routine:	pmap_copy_on_write
1850 *	Function:
1851 *		Remove write privileges from all
1852 *		physical maps for this physical page.
1853 */
1854void
1855pmap_copy_on_write(pa)
1856	vm_offset_t pa;
1857{
1858	pmap_changebit(pa, PG_RW, FALSE);
1859}
1860
1861
1862vm_offset_t
1863pmap_phys_address(ppn)
1864	int ppn;
1865{
1866	return(i386_ptob(ppn));
1867}
1868
1869/*
1870 * Miscellaneous support routines follow
1871 */
1872
1873void
1874i386_protection_init()
1875{
1876	register int *kp, prot;
1877
1878	kp = protection_codes;
1879	for (prot = 0; prot < 8; prot++) {
1880		switch (prot) {
1881		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
1882			*kp++ = 0;
1883			break;
1884		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
1885		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
1886		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
1887			*kp++ = PG_RO;
1888			break;
1889		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
1890		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
1891		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
1892		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
1893			*kp++ = PG_RW;
1894			break;
1895		}
1896	}
1897}
1898
1899#ifdef DEBUG
1900void
1901pmap_pvdump(pa)
1902	vm_offset_t pa;
1903{
1904	register pv_entry_t pv;
1905
1906	printf("pa %x", pa);
1907	for (pv = pa_to_pvh(pa); pv; pv = pv->pv_next) {
1908		printf(" -> pmap %x, va %x, flags %x",
1909		       pv->pv_pmap, pv->pv_va, pv->pv_flags);
1910		pads(pv->pv_pmap);
1911	}
1912	printf(" ");
1913}
1914
1915/* print address space of pmap*/
1916void
1917pads(pm)
1918	pmap_t pm;
1919{
1920	unsigned va, i, j;
1921	pt_entry_t *ptep;
1922
1923	if (pm == kernel_pmap) return;
1924	for (i = 0; i < 1024; i++)
1925		if (pm->pm_pdir[i])
1926			for (j = 0; j < 1024 ; j++) {
1927				va = (i<<PD_SHIFT)+(j<<PG_SHIFT);
1928				if (pm == kernel_pmap && va < KERNBASE)
1929						continue;
1930				if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
1931						continue;
1932				ptep = pmap_pte(pm, va);
1933				if (pmap_pte_v(ptep))
1934					printf("%x:%x ", va, *(int *)ptep);
1935			} ;
1936
1937}
1938#endif
1939