pmap.c revision 1887
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 1.27 1994/08/03 02:45:28 davidg Exp $
43 */
44
45/*
46 * Derived from hp300 version by Mike Hibler, this version by William
47 * Jolitz uses a recursive map [a pde points to the page directory] to
48 * map the page tables using the pagetables themselves. This is done to
49 * reduce the impact on kernel virtual memory for lots of sparse address
50 * space, and to reduce the cost of memory to each process.
51 *
52 *	Derived from: hp300/@(#)pmap.c	7.1 (Berkeley) 12/5/90
53 */
54/*
55 * Major modifications by John S. Dyson primarily to support
56 * pageable page tables, eliminating pmap_attributes,
57 * discontiguous memory pages, and using more efficient string
58 * instructions. Jan 13, 1994.  Further modifications on Mar 2, 1994,
59 * general clean-up and efficiency mods.
60 */
61
62/*
63 *	Manages physical address maps.
64 *
65 *	In addition to hardware address maps, this
66 *	module is called upon to provide software-use-only
67 *	maps which may or may not be stored in the same
68 *	form as hardware maps.  These pseudo-maps are
69 *	used to store intermediate results from copy
70 *	operations to and from address spaces.
71 *
72 *	Since the information managed by this module is
73 *	also stored by the logical address mapping module,
74 *	this module may throw away valid virtual-to-physical
75 *	mappings at almost any time.  However, invalidations
76 *	of virtual-to-physical mappings must be done as
77 *	requested.
78 *
79 *	In order to cope with hardware architectures which
80 *	make virtual-to-physical map invalidates expensive,
81 *	this module may delay invalidate or reduced protection
82 *	operations until such time as they are actually
83 *	necessary.  This module is given full information as
84 *	to which processors are currently using which maps,
85 *	and to when physical maps must be made correct.
86 */
87
88#include <sys/param.h>
89#include <sys/proc.h>
90#include <sys/malloc.h>
91#include <sys/user.h>
92
93#include <vm/vm.h>
94#include <vm/vm_kern.h>
95#include <vm/vm_page.h>
96
97#include <i386/include/cpufunc.h>
98#include <i386/include/cputypes.h>
99
100#include <i386/isa/isa.h>
101
102/*
103 * Allocate various and sundry SYSMAPs used in the days of old VM
104 * and not yet converted.  XXX.
105 */
106#define BSDVM_COMPAT	1
107
108/*
109 * Get PDEs and PTEs for user/kernel address space
110 */
111#define	pmap_pde(m, v)	(&((m)->pm_pdir[((vm_offset_t)(v) >> PD_SHIFT)&1023]))
112#define pdir_pde(m, v) (m[((vm_offset_t)(v) >> PD_SHIFT)&1023])
113
114#define pmap_pte_pa(pte)	(*(int *)(pte) & PG_FRAME)
115
116#define pmap_pde_v(pte)		((*(int *)pte & PG_V) != 0)
117#define pmap_pte_w(pte)		((*(int *)pte & PG_W) != 0)
118#define pmap_pte_m(pte)		((*(int *)pte & PG_M) != 0)
119#define pmap_pte_u(pte)		((*(int *)pte & PG_U) != 0)
120#define pmap_pte_v(pte)		((*(int *)pte & PG_V) != 0)
121
122#define pmap_pte_set_w(pte, v)		((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
123#define pmap_pte_set_prot(pte, v)	((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
124
125/*
126 * Given a map and a machine independent protection code,
127 * convert to a vax protection code.
128 */
129#define pte_prot(m, p)	(protection_codes[p])
130int	protection_codes[8];
131
132struct pmap	kernel_pmap_store;
133pmap_t		kernel_pmap;
134
135vm_offset_t	phys_avail[6];	/* 2 entries + 1 null */
136vm_offset_t    	avail_start;	/* PA of first available physical page */
137vm_offset_t	avail_end;	/* PA of last available physical page */
138vm_size_t	mem_size;	/* memory size in bytes */
139vm_offset_t	virtual_avail;  /* VA of first avail page (after kernel bss)*/
140vm_offset_t	virtual_end;	/* VA of last avail page (end of kernel AS) */
141int		i386pagesperpage;	/* PAGE_SIZE / I386_PAGE_SIZE */
142boolean_t	pmap_initialized = FALSE;	/* Has pmap_init completed? */
143vm_offset_t	vm_first_phys, vm_last_phys;
144
145static inline boolean_t		pmap_testbit();
146static inline void		pmap_changebit();
147static inline int		pmap_is_managed();
148static inline void		*vm_get_pmap();
149static inline void		vm_put_pmap();
150inline void			pmap_use_pt();
151inline void			pmap_unuse_pt();
152inline pt_entry_t * pmap_pte();
153static inline pv_entry_t	get_pv_entry();
154void				pmap_alloc_pv_entry();
155void				pmap_clear_modify();
156static void			i386_protection_init();
157
158void	pmap_kenter	__P((vm_offset_t, vm_offset_t));
159void	pmap_kremove	__P((vm_offset_t));
160void	pmap_qenter	__P((vm_offset_t, vm_page_t *, int));
161void	pmap_qremove	__P((vm_offset_t, int));
162
163extern vm_offset_t clean_sva, clean_eva;
164extern int cpu_class;
165
166#if BSDVM_COMPAT
167#include "msgbuf.h"
168
169/*
170 * All those kernel PT submaps that BSD is so fond of
171 */
172pt_entry_t *CMAP1, *CMAP2, *ptmmap;
173caddr_t		CADDR1, CADDR2, ptvmmap;
174pt_entry_t *msgbufmap;
175struct msgbuf	*msgbufp;
176#endif
177
178void init_pv_entries(int) ;
179
180/*
181 *	Routine:	pmap_pte
182 *	Function:
183 *		Extract the page table entry associated
184 *		with the given map/virtual_address pair.
185 * [ what about induced faults -wfj]
186 */
187
188inline pt_entry_t *
189const pmap_pte(pmap, va)
190	register pmap_t	pmap;
191	vm_offset_t va;
192{
193
194	if (pmap && *pmap_pde(pmap, va)) {
195		vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
196		/* are we current address space or kernel? */
197		if ( (pmap == kernel_pmap) || (frame == ((int) PTDpde & PG_FRAME)))
198			return ((pt_entry_t *) vtopte(va));
199		/* otherwise, we are alternate address space */
200		else {
201			if ( frame != ((int) APTDpde & PG_FRAME) ) {
202				APTDpde = pmap->pm_pdir[PTDPTDI];
203				tlbflush();
204			}
205			return((pt_entry_t *) avtopte(va));
206		}
207	}
208	return(0);
209}
210
211/*
212 *	Routine:	pmap_extract
213 *	Function:
214 *		Extract the physical page address associated
215 *		with the given map/virtual_address pair.
216 */
217
218vm_offset_t
219pmap_extract(pmap, va)
220	register pmap_t	pmap;
221	vm_offset_t va;
222{
223	pd_entry_t save;
224	vm_offset_t pa;
225	int s;
226
227	if (pmap && *pmap_pde(pmap, va)) {
228		vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
229		/* are we current address space or kernel? */
230		if ( (pmap == kernel_pmap)
231			|| (frame == ((int) PTDpde & PG_FRAME)) ) {
232			pa = *(int *) vtopte(va);
233		/* otherwise, we are alternate address space */
234		} else {
235			if ( frame != ((int) APTDpde & PG_FRAME)) {
236				APTDpde = pmap->pm_pdir[PTDPTDI];
237				tlbflush();
238			}
239			pa = *(int *) avtopte(va);
240		}
241		pa = (pa & PG_FRAME) | (va & ~PG_FRAME);
242		return pa;
243	}
244	return 0;
245
246}
247
248/*
249 * determine if a page is managed (memory vs. device)
250 */
251static inline int
252pmap_is_managed(pa)
253	vm_offset_t pa;
254{
255	int i;
256
257	if (!pmap_initialized)
258		return 0;
259
260	for (i = 0; phys_avail[i + 1]; i += 2) {
261		if (pa >= phys_avail[i] && pa < phys_avail[i + 1])
262			return 1;
263	}
264	return 0;
265}
266
267/*
268 * find the vm_page_t of a pte (only) given va of pte and pmap
269 */
270inline vm_page_t
271pmap_pte_vm_page(pmap, pt)
272	pmap_t pmap;
273	vm_offset_t pt;
274{
275	pt = i386_trunc_page( pt);
276	pt = (pt - UPT_MIN_ADDRESS) / NBPG;
277	pt = ((vm_offset_t) pmap->pm_pdir[pt]) & PG_FRAME;
278	return PHYS_TO_VM_PAGE(pt);
279}
280
281/*
282 * Wire a page table page
283 */
284inline void
285pmap_use_pt(pmap, va)
286	pmap_t pmap;
287	vm_offset_t va;
288{
289	vm_offset_t pt;
290
291	if (va >= VM_MAX_ADDRESS || !pmap_initialized)
292		return;
293
294	pt = (vm_offset_t) vtopte(va);
295	vm_page_hold( pmap_pte_vm_page(pmap, pt));
296}
297
298/*
299 * Unwire a page table page
300 */
301inline void
302pmap_unuse_pt(pmap, va)
303	pmap_t pmap;
304	vm_offset_t va;
305{
306	vm_offset_t pt;
307
308	if (va >= VM_MAX_ADDRESS || !pmap_initialized)
309		return;
310
311	pt = (vm_offset_t) vtopte(va);
312	vm_page_unhold( pmap_pte_vm_page(pmap, pt));
313}
314
315/* [ macro again?, should I force kstack into user map here? -wfj ] */
316void
317pmap_activate(pmap, pcbp)
318	register pmap_t pmap;
319	struct pcb *pcbp;
320{
321	PMAP_ACTIVATE(pmap, pcbp);
322}
323
324/*
325 *	Bootstrap the system enough to run with virtual memory.
326 *	Map the kernel's code and data, and allocate the system page table.
327 *
328 *	On the I386 this is called after mapping has already been enabled
329 *	and just syncs the pmap module with what has already been done.
330 *	[We can't call it easily with mapping off since the kernel is not
331 *	mapped with PA == VA, hence we would have to relocate every address
332 *	from the linked base (virtual) address "KERNBASE" to the actual
333 *	(physical) address starting relative to 0]
334 */
335
336#define DMAPAGES 8
337void
338pmap_bootstrap(firstaddr, loadaddr)
339	vm_offset_t firstaddr;
340	vm_offset_t loadaddr;
341{
342#if BSDVM_COMPAT
343	vm_offset_t va;
344	pt_entry_t *pte;
345#endif
346	extern int IdlePTD;
347
348	avail_start = firstaddr + DMAPAGES*NBPG;
349
350	virtual_avail = (vm_offset_t) KERNBASE + avail_start;
351	virtual_end = VM_MAX_KERNEL_ADDRESS;
352	i386pagesperpage = PAGE_SIZE / NBPG;
353
354	/*
355	 * Initialize protection array.
356	 */
357	i386_protection_init();
358
359	/*
360	 * The kernel's pmap is statically allocated so we don't
361	 * have to use pmap_create, which is unlikely to work
362	 * correctly at this part of the boot sequence.
363	 */
364	kernel_pmap = &kernel_pmap_store;
365
366	kernel_pmap->pm_pdir = (pd_entry_t *)(KERNBASE + IdlePTD);
367
368	simple_lock_init(&kernel_pmap->pm_lock);
369	kernel_pmap->pm_count = 1;
370
371#if BSDVM_COMPAT
372	/*
373	 * Allocate all the submaps we need
374	 */
375#define	SYSMAP(c, p, v, n)	\
376	v = (c)va; va += ((n)*NBPG); p = pte; pte += (n);
377
378	va = virtual_avail;
379	pte = pmap_pte(kernel_pmap, va);
380
381	SYSMAP(caddr_t		,CMAP1		,CADDR1	   ,1		)
382	SYSMAP(caddr_t		,CMAP2		,CADDR2	   ,1		)
383	SYSMAP(caddr_t		,ptmmap		,ptvmmap	   ,1		)
384	SYSMAP(struct msgbuf *	,msgbufmap	,msgbufp   ,1		)
385	virtual_avail = va;
386#endif
387	/*
388	 * reserve special hunk of memory for use by bus dma as a bounce
389	 * buffer (contiguous virtual *and* physical memory). for now,
390	 * assume vm does not use memory beneath hole, and we know that
391	 * the bootstrap uses top 32k of base memory. -wfj
392	 */
393	{
394		extern vm_offset_t isaphysmem;
395		isaphysmem = va;
396
397		virtual_avail = pmap_map(va, firstaddr,
398				firstaddr + DMAPAGES*NBPG, VM_PROT_ALL);
399	}
400
401	*(int *)CMAP1 = *(int *)CMAP2 = *(int *)PTD = 0;
402	tlbflush();
403
404}
405
406/*
407 *	Initialize the pmap module.
408 *	Called by vm_init, to initialize any structures that the pmap
409 *	system needs to map virtual memory.
410 *	pmap_init has been enhanced to support in a fairly consistant
411 *	way, discontiguous physical memory.
412 */
413void
414pmap_init(phys_start, phys_end)
415	vm_offset_t	phys_start, phys_end;
416{
417	vm_offset_t	addr, addr2;
418	vm_size_t	npg, s;
419	int		rv;
420	int i;
421	extern int KPTphys;
422	extern int IdlePTD;
423
424	/*
425	 * Now that kernel map has been allocated, we can mark as
426	 * unavailable regions which we have mapped in locore.
427	 */
428	addr = atdevbase;
429	(void) vm_map_find(kernel_map, NULL, (vm_offset_t) 0,
430			   &addr, (0x100000-0xa0000), FALSE);
431
432	addr = (vm_offset_t) KERNBASE + IdlePTD;
433	vm_object_reference(kernel_object);
434	(void) vm_map_find(kernel_map, kernel_object, addr,
435			   &addr, (4 + NKPT) * NBPG, FALSE);
436
437
438	/*
439	 * calculate the number of pv_entries needed
440	 */
441	vm_first_phys = phys_avail[0];
442	for (i = 0; phys_avail[i + 1]; i += 2) ;
443	npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / NBPG;
444
445	/*
446	 * Allocate memory for random pmap data structures.  Includes the
447	 * pv_head_table.
448	 */
449	s = (vm_size_t) (sizeof(struct pv_entry) * npg);
450	s = i386_round_page(s);
451	addr = (vm_offset_t) kmem_alloc(kernel_map, s);
452	pv_table = (pv_entry_t) addr;
453
454	/*
455	 * init the pv free list
456	 */
457	init_pv_entries(npg);
458	/*
459	 * Now it is safe to enable pv_table recording.
460	 */
461	pmap_initialized = TRUE;
462}
463
464/*
465 *	Used to map a range of physical addresses into kernel
466 *	virtual address space.
467 *
468 *	For now, VM is already on, we only need to map the
469 *	specified memory.
470 */
471vm_offset_t
472pmap_map(virt, start, end, prot)
473	vm_offset_t	virt;
474	vm_offset_t	start;
475	vm_offset_t	end;
476	int		prot;
477{
478	while (start < end) {
479		pmap_enter(kernel_pmap, virt, start, prot, FALSE);
480		virt += PAGE_SIZE;
481		start += PAGE_SIZE;
482	}
483	return(virt);
484}
485
486/*
487 *	Create and return a physical map.
488 *
489 *	If the size specified for the map
490 *	is zero, the map is an actual physical
491 *	map, and may be referenced by the
492 *	hardware.
493 *
494 *	If the size specified is non-zero,
495 *	the map will be used in software only, and
496 *	is bounded by that size.
497 *
498 * [ just allocate a ptd and mark it uninitialize -- should we track
499 *   with a table which process has which ptd? -wfj ]
500 */
501
502pmap_t
503pmap_create(size)
504	vm_size_t	size;
505{
506	register pmap_t pmap;
507
508	/*
509	 * Software use map does not need a pmap
510	 */
511	if (size)
512		return(NULL);
513
514	pmap = (pmap_t) malloc(sizeof *pmap, M_VMPMAP, M_WAITOK);
515	bzero(pmap, sizeof(*pmap));
516	pmap_pinit(pmap);
517	return (pmap);
518}
519
520
521struct pmaplist {
522	struct pmaplist *next;
523};
524
525static inline void *
526vm_get_pmap()
527{
528	struct pmaplist *rtval;
529
530	rtval = (struct pmaplist *)kmem_alloc(kernel_map, ctob(1));
531	bzero(rtval, ctob(1));
532	return rtval;
533}
534
535static inline void
536vm_put_pmap(up)
537	struct pmaplist *up;
538{
539	kmem_free(kernel_map, (vm_offset_t)up, ctob(1));
540}
541
542/*
543 * Initialize a preallocated and zeroed pmap structure,
544 * such as one in a vmspace structure.
545 */
546void
547pmap_pinit(pmap)
548	register struct pmap *pmap;
549{
550	/*
551	 * No need to allocate page table space yet but we do need a
552	 * valid page directory table.
553	 */
554	pmap->pm_pdir = (pd_entry_t *) vm_get_pmap();
555
556	/* wire in kernel global address entries */
557	bcopy(PTD+KPTDI, pmap->pm_pdir+KPTDI, NKPT*PTESIZE);
558
559	/* install self-referential address mapping entry */
560	*(int *)(pmap->pm_pdir+PTDPTDI) =
561		((int)pmap_kextract((vm_offset_t)pmap->pm_pdir)) | PG_V | PG_KW;
562
563	pmap->pm_count = 1;
564	simple_lock_init(&pmap->pm_lock);
565}
566
567/*
568 *	Retire the given physical map from service.
569 *	Should only be called if the map contains
570 *	no valid mappings.
571 */
572void
573pmap_destroy(pmap)
574	register pmap_t pmap;
575{
576	int count;
577
578	if (pmap == NULL)
579		return;
580
581	simple_lock(&pmap->pm_lock);
582	count = --pmap->pm_count;
583	simple_unlock(&pmap->pm_lock);
584	if (count == 0) {
585		pmap_release(pmap);
586		free((caddr_t)pmap, M_VMPMAP);
587	}
588}
589
590/*
591 * Release any resources held by the given physical map.
592 * Called when a pmap initialized by pmap_pinit is being released.
593 * Should only be called if the map contains no valid mappings.
594 */
595void
596pmap_release(pmap)
597	register struct pmap *pmap;
598{
599	vm_put_pmap((struct pmaplist *) pmap->pm_pdir);
600}
601
602/*
603 *	Add a reference to the specified pmap.
604 */
605void
606pmap_reference(pmap)
607	pmap_t	pmap;
608{
609	if (pmap != NULL) {
610		simple_lock(&pmap->pm_lock);
611		pmap->pm_count++;
612		simple_unlock(&pmap->pm_lock);
613	}
614}
615
616#define PV_FREELIST_MIN ((NBPG / sizeof (struct pv_entry)) / 2)
617
618/*
619 * Data for the pv entry allocation mechanism
620 */
621int pv_freelistcnt;
622pv_entry_t pv_freelist;
623vm_offset_t pvva;
624int npvvapg;
625
626/*
627 * free the pv_entry back to the free list
628 */
629inline static void
630free_pv_entry(pv)
631	pv_entry_t pv;
632{
633	if (!pv) return;
634	++pv_freelistcnt;
635	pv->pv_next = pv_freelist;
636	pv_freelist = pv;
637}
638
639/*
640 * get a new pv_entry, allocating a block from the system
641 * when needed.
642 * the memory allocation is performed bypassing the malloc code
643 * because of the possibility of allocations at interrupt time.
644 */
645static inline pv_entry_t
646get_pv_entry()
647{
648	pv_entry_t tmp;
649
650	/*
651	 * get more pv_entry pages if needed
652	 */
653	while (pv_freelistcnt < PV_FREELIST_MIN || pv_freelist == 0) {
654		pmap_alloc_pv_entry();
655	}
656
657	/*
658	 * get a pv_entry off of the free list
659	 */
660	--pv_freelistcnt;
661	tmp = pv_freelist;
662	pv_freelist = tmp->pv_next;
663	tmp->pv_pmap = 0;
664	tmp->pv_va = 0;
665	tmp->pv_next = 0;
666	return tmp;
667}
668
669/*
670 * this *strange* allocation routine *statistically* eliminates the
671 * *possibility* of a malloc failure (*FATAL*) for a pv_entry_t data structure.
672 * also -- this code is MUCH MUCH faster than the malloc equiv...
673 */
674void
675pmap_alloc_pv_entry()
676{
677	/*
678	 * do we have any pre-allocated map-pages left?
679	 */
680	if (npvvapg) {
681		vm_page_t m;
682		/*
683		 * we do this to keep recursion away
684		 */
685		pv_freelistcnt += PV_FREELIST_MIN;
686		/*
687		 * allocate a physical page out of the vm system
688		 */
689		if (m = vm_page_alloc(kernel_object, pvva-vm_map_min(kernel_map))) {
690			int newentries;
691			int i;
692			pv_entry_t entry;
693			newentries = (NBPG/sizeof (struct pv_entry));
694			/*
695			 * wire the page
696			 */
697			vm_page_wire(m);
698			m->flags &= ~PG_BUSY;
699			/*
700			 * let the kernel see it
701			 */
702			pmap_kenter(pvva, VM_PAGE_TO_PHYS(m));
703
704			entry = (pv_entry_t) pvva;
705			/*
706			 * update the allocation pointers
707			 */
708			pvva += NBPG;
709			--npvvapg;
710
711			/*
712			 * free the entries into the free list
713			 */
714			for (i = 0; i < newentries; i++) {
715				free_pv_entry(entry);
716				entry++;
717			}
718		}
719		pv_freelistcnt -= PV_FREELIST_MIN;
720	}
721	if (!pv_freelist)
722		panic("get_pv_entry: cannot get a pv_entry_t");
723}
724
725
726
727/*
728 * init the pv_entry allocation system
729 */
730#define PVSPERPAGE 64
731void
732init_pv_entries(npg)
733	int npg;
734{
735	/*
736	 * allocate enough kvm space for PVSPERPAGE entries per page (lots)
737	 * kvm space is fairly cheap, be generous!!!  (the system can panic
738	 * if this is too small.)
739	 */
740	npvvapg = ((npg*PVSPERPAGE) * sizeof(struct pv_entry) + NBPG - 1)/NBPG;
741	pvva = kmem_alloc_pageable(kernel_map, npvvapg * NBPG);
742	/*
743	 * get the first batch of entries
744	 */
745	free_pv_entry(get_pv_entry());
746}
747
748static pt_entry_t *
749get_pt_entry(pmap)
750	pmap_t pmap;
751{
752	pt_entry_t *ptp;
753	vm_offset_t frame = (int) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
754	/* are we current address space or kernel? */
755	if (pmap == kernel_pmap || frame == ((int) PTDpde & PG_FRAME)) {
756		ptp=PTmap;
757	/* otherwise, we are alternate address space */
758	} else {
759		if ( frame != ((int) APTDpde & PG_FRAME)) {
760			APTDpde = pmap->pm_pdir[PTDPTDI];
761			tlbflush();
762		}
763		ptp=APTmap;
764	     }
765	return ptp;
766}
767
768/*
769 * If it is the first entry on the list, it is actually
770 * in the header and we must copy the following entry up
771 * to the header.  Otherwise we must search the list for
772 * the entry.  In either case we free the now unused entry.
773 */
774void
775pmap_remove_entry(pmap, pv, va)
776	struct pmap *pmap;
777	pv_entry_t pv;
778	vm_offset_t va;
779{
780	pv_entry_t npv;
781	int wired;
782	int s;
783	s = splhigh();
784	if (pmap == pv->pv_pmap && va == pv->pv_va) {
785		npv = pv->pv_next;
786		if (npv) {
787			*pv = *npv;
788			free_pv_entry(npv);
789		} else {
790			pv->pv_pmap = NULL;
791		}
792	} else {
793		for (npv = pv->pv_next; npv; npv = npv->pv_next) {
794			if (pmap == npv->pv_pmap && va == npv->pv_va) {
795				break;
796			}
797			pv = npv;
798		}
799		if (npv) {
800			pv->pv_next = npv->pv_next;
801			free_pv_entry(npv);
802		}
803	}
804	splx(s);
805}
806
807/*
808 *	Remove the given range of addresses from the specified map.
809 *
810 *	It is assumed that the start and end are properly
811 *	rounded to the page size.
812 */
813void
814pmap_remove(pmap, sva, eva)
815	struct pmap *pmap;
816	register vm_offset_t sva;
817	register vm_offset_t eva;
818{
819	register pt_entry_t *ptp,*ptq;
820	vm_offset_t pa;
821	register pv_entry_t pv;
822	vm_offset_t va;
823	vm_page_t m;
824	pt_entry_t oldpte;
825
826	if (pmap == NULL)
827		return;
828
829	ptp = get_pt_entry(pmap);
830
831/*
832 * special handling of removing one page.  a very
833 * common operation and easy to short circuit some
834 * code.
835 */
836	if( (sva + NBPG) == eva) {
837
838		if( *pmap_pde( pmap, sva) == 0)
839			return;
840
841		ptq = ptp + i386_btop(sva);
842
843		if( !*ptq)
844			return;
845		/*
846		 * Update statistics
847		 */
848		if (pmap_pte_w(ptq))
849			pmap->pm_stats.wired_count--;
850		pmap->pm_stats.resident_count--;
851
852		pa = pmap_pte_pa(ptq);
853		oldpte = *ptq;
854		*ptq = 0;
855
856		if (pmap_is_managed(pa)) {
857			if ((int) oldpte & (PG_M | PG_U)) {
858				if ((sva < USRSTACK || sva > UPT_MAX_ADDRESS) ||
859				    (sva >= USRSTACK && sva < USRSTACK+(UPAGES*NBPG))) {
860					if (sva < clean_sva || sva >= clean_eva) {
861						m = PHYS_TO_VM_PAGE(pa);
862						if ((int) oldpte & PG_M) {
863							m->flags &= ~PG_CLEAN;
864						}
865						if ((int) oldpte & PG_U) {
866							m->flags |= PG_REFERENCED;
867						}
868					}
869				}
870			}
871
872			pv = pa_to_pvh(pa);
873			pmap_remove_entry(pmap, pv, sva);
874			pmap_unuse_pt(pmap, sva);
875		}
876		tlbflush();
877		return;
878	}
879
880	sva = i386_btop(sva);
881	eva = i386_btop(eva);
882
883	while (sva < eva) {
884		/*
885		 * Weed out invalid mappings.
886		 * Note: we assume that the page directory table is
887	 	 * always allocated, and in kernel virtual.
888		 */
889
890		if ( *pmap_pde(pmap, i386_ptob(sva)) == 0 ) {
891			/* We can race ahead here, straight to next pde.. */
892	nextpde:
893			sva = ((sva + NPTEPG) & ~(NPTEPG - 1));
894			continue;
895		}
896
897		ptq = ptp + sva;
898
899		/*
900		 * search for page table entries, use string operations
901		 * that are much faster than
902		 * explicitly scanning when page tables are not fully
903		 * populated.
904		 */
905		if ( *ptq == 0) {
906			vm_offset_t pdnxt = ((sva + NPTEPG) & ~(NPTEPG - 1));
907			vm_offset_t nscan = pdnxt - sva;
908			int found = 0;
909
910			if ((nscan + sva) > eva)
911				nscan = eva - sva;
912
913			asm("xorl %%eax,%%eax;cld;repe;scasl;jz 1f;incl %%eax;1:;"
914				:"=D"(ptq),"=a"(found)
915				:"c"(nscan),"0"(ptq)
916				:"cx");
917
918			if( !found) {
919				sva = pdnxt;
920				continue;
921			}
922			ptq -= 1;
923
924			sva = ptq - ptp;
925		}
926
927		/*
928		 * Update statistics
929		 */
930		oldpte = *ptq;
931		if (((int)oldpte) & PG_W)
932			pmap->pm_stats.wired_count--;
933		pmap->pm_stats.resident_count--;
934
935		/*
936		 * Invalidate the PTEs.
937		 * XXX: should cluster them up and invalidate as many
938		 * as possible at once.
939		 */
940		*ptq = 0;
941
942		va = i386_ptob(sva);
943
944		/*
945		 * Remove from the PV table (raise IPL since we
946		 * may be called at interrupt time).
947		 */
948		pa = ((int)oldpte) & PG_FRAME;
949		if (!pmap_is_managed(pa)) {
950			++sva;
951			continue;
952		}
953
954		if ((((int) oldpte & PG_M) && (va < USRSTACK || va > UPT_MAX_ADDRESS))
955			|| (va >= USRSTACK && va < USRSTACK+(UPAGES*NBPG))) {
956			if (va < clean_sva || va >= clean_eva ) {
957				m = PHYS_TO_VM_PAGE(pa);
958				m->flags &= ~PG_CLEAN;
959			}
960		}
961
962		pv = pa_to_pvh(pa);
963		pmap_remove_entry(pmap, pv, va);
964		pmap_unuse_pt(pmap, va);
965		++sva;
966	}
967	tlbflush();
968}
969
970/*
971 *	Routine:	pmap_remove_all
972 *	Function:
973 *		Removes this physical page from
974 *		all physical maps in which it resides.
975 *		Reflects back modify bits to the pager.
976 *
977 *	Notes:
978 *		Original versions of this routine were very
979 *		inefficient because they iteratively called
980 *		pmap_remove (slow...)
981 */
982void
983pmap_remove_all(pa)
984	vm_offset_t pa;
985{
986	register pv_entry_t pv, npv;
987	register pt_entry_t *pte, *ptp;
988	vm_offset_t va;
989	struct pmap *pmap;
990	struct map *map;
991	vm_page_t m;
992	int s;
993	int anyvalid = 0;
994
995	/*
996	 * Not one of ours
997	 */
998	if (!pmap_is_managed(pa))
999		return;
1000
1001	pa = i386_trunc_page(pa);
1002	pv = pa_to_pvh(pa);
1003	m = PHYS_TO_VM_PAGE(pa);
1004
1005	s = splhigh();
1006	while (pv->pv_pmap != NULL) {
1007		pmap = pv->pv_pmap;
1008		ptp = get_pt_entry(pmap);
1009		va = i386_btop(pv->pv_va);
1010		pte = ptp + va;
1011		if (pmap_pte_w(pte))
1012			pmap->pm_stats.wired_count--;
1013		if ( *pte) {
1014			pmap->pm_stats.resident_count--;
1015			anyvalid++;
1016
1017			/*
1018			 * update the vm_page_t clean bit
1019			 */
1020			if ( (m->flags & PG_CLEAN) &&
1021				((((int) *pte) & PG_M) && (pv->pv_va < USRSTACK || pv->pv_va > UPT_MAX_ADDRESS))
1022				|| (pv->pv_va >= USRSTACK && pv->pv_va < USRSTACK+(UPAGES*NBPG))) {
1023				if (pv->pv_va < clean_sva || pv->pv_va >= clean_eva) {
1024					m->flags &= ~PG_CLEAN;
1025				}
1026			}
1027
1028			*pte = 0;
1029		}
1030		pmap_unuse_pt(pmap, pv->pv_va);
1031
1032		npv = pv->pv_next;
1033		if (npv) {
1034			*pv = *npv;
1035			free_pv_entry(npv);
1036		} else {
1037			pv->pv_pmap = NULL;
1038		}
1039	}
1040	splx(s);
1041	if (anyvalid)
1042		tlbflush();
1043}
1044
1045
1046/*
1047 *	Set the physical protection on the
1048 *	specified range of this map as requested.
1049 */
1050void
1051pmap_protect(pmap, sva, eva, prot)
1052	register pmap_t	pmap;
1053	vm_offset_t	sva, eva;
1054	vm_prot_t	prot;
1055{
1056	register pt_entry_t *pte;
1057	register vm_offset_t va;
1058	int i386prot;
1059	register pt_entry_t *ptp;
1060	int evap = i386_btop(eva);
1061	int s;
1062	int anyvalid = 0;;
1063
1064	if (pmap == NULL)
1065		return;
1066
1067	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1068		pmap_remove(pmap, sva, eva);
1069		return;
1070	}
1071	if (prot & VM_PROT_WRITE)
1072		return;
1073
1074	ptp = get_pt_entry(pmap);
1075
1076	va = sva;
1077	while (va < eva) {
1078		int found=0;
1079		int svap;
1080		vm_offset_t nscan;
1081		/*
1082		 * Page table page is not allocated.
1083		 * Skip it, we don't want to force allocation
1084		 * of unnecessary PTE pages just to set the protection.
1085		 */
1086		if (! *pmap_pde(pmap, va)) {
1087			/* XXX: avoid address wrap around */
1088nextpde:
1089			if (va >= i386_trunc_pdr((vm_offset_t)-1))
1090				break;
1091			va = i386_round_pdr(va + PAGE_SIZE);
1092			continue;
1093		}
1094
1095		pte = ptp + i386_btop(va);
1096
1097		if( *pte == 0) {
1098		/*
1099		 * scan for a non-empty pte
1100		 */
1101			svap = pte - ptp;
1102			nscan = ((svap + NPTEPG) & ~(NPTEPG - 1)) - svap;
1103
1104			if (nscan + svap > evap)
1105				nscan = evap - svap;
1106
1107			found = 0;
1108			if (nscan)
1109				asm("xorl %%eax,%%eax;cld;repe;scasl;jz 1f;incl %%eax;1:;"
1110					:"=D"(pte),"=a"(found)
1111					:"c"(nscan),"0"(pte):"cx");
1112
1113			if( !found)
1114				goto nextpde;
1115
1116			pte -= 1;
1117			svap = pte - ptp;
1118
1119			va = i386_ptob(svap);
1120		}
1121
1122		anyvalid++;
1123
1124		i386prot = pte_prot(pmap, prot);
1125		if (va < UPT_MAX_ADDRESS) {
1126			i386prot |= PG_u;
1127			if( va >= UPT_MIN_ADDRESS)
1128				i386prot |= PG_RW;
1129		}
1130		pmap_pte_set_prot(pte, i386prot);
1131		va += PAGE_SIZE;
1132	}
1133	if (anyvalid)
1134		tlbflush();
1135}
1136
1137/*
1138 *	Insert the given physical page (p) at
1139 *	the specified virtual address (v) in the
1140 *	target physical map with the protection requested.
1141 *
1142 *	If specified, the page will be wired down, meaning
1143 *	that the related pte can not be reclaimed.
1144 *
1145 *	NB:  This is the only routine which MAY NOT lazy-evaluate
1146 *	or lose information.  That is, this routine must actually
1147 *	insert this page into the given map NOW.
1148 */
1149void
1150pmap_enter(pmap, va, pa, prot, wired)
1151	register pmap_t pmap;
1152	vm_offset_t va;
1153	register vm_offset_t pa;
1154	vm_prot_t prot;
1155	boolean_t wired;
1156{
1157	register pt_entry_t *pte;
1158	register pt_entry_t npte;
1159	vm_offset_t opa;
1160	int cacheable=1;
1161	int ptevalid = 0;
1162
1163	if (pmap == NULL)
1164		return;
1165
1166	va = i386_trunc_page(va);
1167	pa = i386_trunc_page(pa);
1168	if (va > VM_MAX_KERNEL_ADDRESS)panic("pmap_enter: toobig");
1169
1170	/*
1171	 * Page Directory table entry not valid, we need a new PT page
1172	 */
1173	if ( *pmap_pde(pmap, va) == 0) {
1174		pg("ptdi %x, va %x", pmap->pm_pdir[PTDPTDI], va);
1175	}
1176
1177	pte = pmap_pte(pmap, va);
1178	opa = pmap_pte_pa(pte);
1179
1180	/*
1181	 * Mapping has not changed, must be protection or wiring change.
1182	 */
1183	if (opa == pa) {
1184		/*
1185		 * Wiring change, just update stats.
1186		 * We don't worry about wiring PT pages as they remain
1187		 * resident as long as there are valid mappings in them.
1188		 * Hence, if a user page is wired, the PT page will be also.
1189		 */
1190		if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
1191			if (wired)
1192				pmap->pm_stats.wired_count++;
1193			else
1194				pmap->pm_stats.wired_count--;
1195		}
1196		goto validate;
1197	}
1198
1199	/*
1200	 * Mapping has changed, invalidate old range and fall through to
1201	 * handle validating new mapping.
1202	 */
1203	if (opa) {
1204		pmap_remove(pmap, va, va + PAGE_SIZE);
1205	}
1206
1207	/*
1208	 * Enter on the PV list if part of our managed memory
1209	 * Note that we raise IPL while manipulating pv_table
1210	 * since pmap_enter can be called at interrupt time.
1211	 */
1212	if (pmap_is_managed(pa)) {
1213		register pv_entry_t pv, npv;
1214		int s;
1215
1216		pv = pa_to_pvh(pa);
1217		s = splhigh();
1218		/*
1219		 * No entries yet, use header as the first entry
1220		 */
1221		if (pv->pv_pmap == NULL) {
1222			pv->pv_va = va;
1223			pv->pv_pmap = pmap;
1224			pv->pv_next = NULL;
1225		}
1226		/*
1227		 * There is at least one other VA mapping this page.
1228		 * Place this entry after the header.
1229		 */
1230		else {
1231			npv = get_pv_entry();
1232			npv->pv_va = va;
1233			npv->pv_pmap = pmap;
1234			npv->pv_next = pv->pv_next;
1235			pv->pv_next = npv;
1236		}
1237		splx(s);
1238		cacheable = 1;
1239	} else {
1240		cacheable = 0;
1241	}
1242
1243	pmap_use_pt(pmap, va);
1244
1245	/*
1246	 * Increment counters
1247	 */
1248	pmap->pm_stats.resident_count++;
1249	if (wired)
1250		pmap->pm_stats.wired_count++;
1251
1252validate:
1253	/*
1254	 * Now validate mapping with desired protection/wiring.
1255	 */
1256	npte = (pt_entry_t) ( (int) (pa | pte_prot(pmap, prot) | PG_V));
1257	/*
1258	 * for correctness:
1259	 */
1260	if( !cacheable)
1261		(int) npte |= PG_N;
1262
1263	/*
1264	 * When forking (copy-on-write, etc):
1265	 * A process will turn off write permissions for any of its writable
1266	 * pages.  If the data (object) is only referred to by one process, the
1267	 * processes map is modified directly as opposed to using the
1268	 * object manipulation routine.  When using pmap_protect, the
1269	 * modified bits are not kept in the vm_page_t data structure.
1270	 * Therefore, when using pmap_enter in vm_fault to bring back
1271	 * writability of a page, there has been no memory of the
1272	 * modified or referenced bits except at the pte level.
1273	 * this clause supports the carryover of the modified and
1274	 * used (referenced) bits.
1275	 */
1276	if (pa == opa)
1277		(int) npte |= (int) *pte & (PG_M|PG_U);
1278
1279
1280	if (wired)
1281		(int) npte |= PG_W;
1282	if (va < UPT_MIN_ADDRESS)
1283		(int) npte |= PG_u;
1284	else if (va < UPT_MAX_ADDRESS)
1285		(int) npte |= PG_u | PG_RW;
1286
1287	if(*pte != npte) {
1288		if (*pte)
1289			ptevalid++;
1290		*pte = npte;
1291	}
1292	if (ptevalid)
1293		tlbflush();
1294}
1295
1296/*
1297 * Add a list of wired pages to the kva
1298 * this routine is only used for temporary
1299 * kernel mappings that do not need to have
1300 * page modification or references recorded.
1301 * Note that old mappings are simply written
1302 * over.  The page *must* be wired.
1303 */
1304void
1305pmap_qenter(va, m, count)
1306	vm_offset_t va;
1307	vm_page_t *m;
1308	int count;
1309{
1310	int i;
1311	int anyvalid = 0;
1312	register pt_entry_t *pte;
1313
1314	for(i=0;i<count;i++) {
1315		pte = vtopte(va + i * NBPG);
1316		if (*pte)
1317			anyvalid++;
1318		*pte = (pt_entry_t) ( (int) (VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | PG_W));
1319	}
1320	if (anyvalid)
1321		tlbflush();
1322}
1323/*
1324 * this routine jerks page mappings from the
1325 * kernel -- it is meant only for temporary mappings.
1326 */
1327void
1328pmap_qremove(va, count)
1329	vm_offset_t va;
1330	int count;
1331{
1332	int i;
1333	register pt_entry_t *pte;
1334	for(i=0;i<count;i++) {
1335		pte = vtopte(va + i * NBPG);
1336		*pte = 0;
1337	}
1338	tlbflush();
1339}
1340
1341/*
1342 * add a wired page to the kva
1343 * note that in order for the mapping to take effect -- you
1344 * should do a tlbflush after doing the pmap_kenter...
1345 */
1346void
1347pmap_kenter(va, pa)
1348	vm_offset_t va;
1349	register vm_offset_t pa;
1350{
1351	register pt_entry_t *pte;
1352	pte = vtopte(va);
1353
1354	*pte = (pt_entry_t) ( (int) (pa | PG_RW | PG_V | PG_W));
1355}
1356
1357/*
1358 * remove a page from the kernel pagetables
1359 */
1360void
1361pmap_kremove( va)
1362	vm_offset_t va;
1363{
1364	register pt_entry_t *pte;
1365	pte = vtopte(va);
1366
1367	*pte = (pt_entry_t) 0;
1368	tlbflush();
1369}
1370
1371/*
1372 * this code makes some *MAJOR* assumptions:
1373 * 1. Current pmap & pmap exists.
1374 * 2. Not wired.
1375 * 3. Read access.
1376 * 4. No page table pages.
1377 * 5. Tlbflush is deferred to calling procedure.
1378 * 6. Page IS managed.
1379 * but is *MUCH* faster than pmap_enter...
1380 */
1381
1382static inline int
1383pmap_enter_quick(pmap, va, pa)
1384	register pmap_t pmap;
1385	vm_offset_t va;
1386	register vm_offset_t pa;
1387{
1388	register pt_entry_t *pte;
1389	register pv_entry_t pv, npv;
1390	int s;
1391	int anyvalid = 0;
1392
1393	/*
1394	 * Enter on the PV list if part of our managed memory
1395	 * Note that we raise IPL while manipulating pv_table
1396	 * since pmap_enter can be called at interrupt time.
1397	 */
1398
1399	pte = vtopte(va);
1400	if (pmap_pte_pa(pte)) {
1401		pmap_remove(pmap, va, va + PAGE_SIZE);
1402	}
1403
1404	pv = pa_to_pvh(pa);
1405	s = splhigh();
1406	/*
1407	 * No entries yet, use header as the first entry
1408	 */
1409	if (pv->pv_pmap == NULL) {
1410		pv->pv_va = va;
1411		pv->pv_pmap = pmap;
1412		pv->pv_next = NULL;
1413	}
1414	/*
1415	 * There is at least one other VA mapping this page.
1416	 * Place this entry after the header.
1417	 */
1418	else {
1419		npv = get_pv_entry();
1420		npv->pv_va = va;
1421		npv->pv_pmap = pmap;
1422		npv->pv_next = pv->pv_next;
1423		pv->pv_next = npv;
1424	}
1425	splx(s);
1426
1427	pmap_use_pt(pmap, va);
1428
1429	/*
1430	 * Increment counters
1431	 */
1432	pmap->pm_stats.resident_count++;
1433
1434validate:
1435
1436	if (*pte)
1437		anyvalid++;
1438	/*
1439	 * Now validate mapping with desired protection/wiring.
1440	 */
1441	*pte = (pt_entry_t) ( (int) (pa | PG_V | PG_u));
1442
1443	return (anyvalid);
1444}
1445
1446/*
1447 * pmap_object_init_pt preloads the ptes for a given object
1448 * into the specified pmap.  This eliminates the blast of soft
1449 * faults on process startup and immediately after an mmap.
1450 */
1451void
1452pmap_object_init_pt(pmap, addr, object, offset, size)
1453	pmap_t pmap;
1454	vm_offset_t addr;
1455	vm_object_t object;
1456	vm_offset_t offset;
1457	vm_offset_t size;
1458{
1459
1460	vm_offset_t tmpoff;
1461	vm_page_t p;
1462	int s;
1463	vm_offset_t v, lastv=0;
1464	pt_entry_t pte;
1465	extern vm_map_t kernel_map;
1466	vm_offset_t objbytes;
1467	int anyvalid = 0;
1468
1469	if (!pmap)
1470		return;
1471
1472	/*
1473	 * if we are processing a major portion of the object, then
1474	 * scan the entire thing.
1475	 */
1476	if( size > object->size / 2) {
1477		objbytes = size;
1478		p = object->memq.tqh_first;
1479		while ((p != NULL) && (objbytes != 0)) {
1480			tmpoff = p->offset;
1481			if( tmpoff < offset) {
1482				p = p->listq.tqe_next;
1483				continue;
1484			}
1485			tmpoff -= offset;
1486			if( tmpoff >= size) {
1487				p = p->listq.tqe_next;
1488				continue;
1489			}
1490
1491			if ((p->flags & (PG_BUSY|PG_FICTITIOUS)) == 0 ) {
1492				vm_page_hold(p);
1493				v = i386_trunc_page(((vm_offset_t)vtopte( addr+tmpoff)));
1494				/* a fault might occur here */
1495				*(volatile char *)v += 0;
1496				vm_page_unhold(p);
1497				anyvalid += pmap_enter_quick(pmap, addr+tmpoff, VM_PAGE_TO_PHYS(p));
1498			}
1499			p = p->listq.tqe_next;
1500			objbytes -= NBPG;
1501		}
1502	} else {
1503	/*
1504	 * else lookup the pages one-by-one.
1505	 */
1506		for(tmpoff = 0; tmpoff < size; tmpoff += NBPG) {
1507			if( p = vm_page_lookup(object, tmpoff + offset)) {
1508				if( (p->flags & (PG_BUSY|PG_FICTITIOUS)) == 0) {
1509					vm_page_hold(p);
1510					v = i386_trunc_page(((vm_offset_t)vtopte( addr+tmpoff)));
1511					/* a fault might occur here */
1512					*(volatile char *)v += 0;
1513					vm_page_unhold(p);
1514					anyvalid += pmap_enter_quick(pmap, addr+tmpoff, VM_PAGE_TO_PHYS(p));
1515				}
1516			}
1517		}
1518	}
1519
1520	if (anyvalid)
1521		tlbflush();
1522}
1523
1524/*
1525 *	Routine:	pmap_change_wiring
1526 *	Function:	Change the wiring attribute for a map/virtual-address
1527 *			pair.
1528 *	In/out conditions:
1529 *			The mapping must already exist in the pmap.
1530 */
1531void
1532pmap_change_wiring(pmap, va, wired)
1533	register pmap_t	pmap;
1534	vm_offset_t	va;
1535	boolean_t	wired;
1536{
1537	register pt_entry_t *pte;
1538
1539	if (pmap == NULL)
1540		return;
1541
1542	pte = pmap_pte(pmap, va);
1543	if (wired && !pmap_pte_w(pte) || !wired && pmap_pte_w(pte)) {
1544		if (wired)
1545			pmap->pm_stats.wired_count++;
1546		else
1547			pmap->pm_stats.wired_count--;
1548	}
1549	/*
1550	 * Wiring is not a hardware characteristic so there is no need
1551	 * to invalidate TLB.
1552	 */
1553	pmap_pte_set_w(pte, wired);
1554	/*
1555 	 * When unwiring, set the modified bit in the pte -- could have
1556	 * been changed by the kernel
1557 	 */
1558	if (!wired)
1559		(int) *pte |= PG_M;
1560}
1561
1562
1563
1564/*
1565 *	Copy the range specified by src_addr/len
1566 *	from the source map to the range dst_addr/len
1567 *	in the destination map.
1568 *
1569 *	This routine is only advisory and need not do anything.
1570 */
1571void
1572pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
1573	pmap_t dst_pmap, src_pmap;
1574	vm_offset_t	dst_addr;
1575	vm_size_t	len;
1576	vm_offset_t	src_addr;
1577{
1578}
1579/*
1580 *	Require that all active physical maps contain no
1581 *	incorrect entries NOW.  [This update includes
1582 *	forcing updates of any address map caching.]
1583 *
1584 *	Generally used to insure that a thread about
1585 *	to run will see a semantically correct world.
1586 */
1587void
1588pmap_update()
1589{
1590	tlbflush();
1591}
1592
1593/*
1594 *	Routine:	pmap_kernel
1595 *	Function:
1596 *		Returns the physical map handle for the kernel.
1597 */
1598pmap_t
1599pmap_kernel()
1600{
1601    	return (kernel_pmap);
1602}
1603
1604/*
1605 *	pmap_zero_page zeros the specified (machine independent)
1606 *	page by mapping the page into virtual memory and using
1607 *	bzero to clear its contents, one machine dependent page
1608 *	at a time.
1609 */
1610void
1611pmap_zero_page(phys)
1612	vm_offset_t phys;
1613{
1614	if (*(int *)CMAP2)
1615		panic("pmap_zero_page: CMAP busy");
1616
1617	*(int *)CMAP2 = PG_V | PG_KW | i386_trunc_page(phys);
1618	bzero(CADDR2,NBPG);
1619
1620	*(int *)CMAP2 = 0;
1621	tlbflush();
1622}
1623
1624/*
1625 *	pmap_copy_page copies the specified (machine independent)
1626 *	page by mapping the page into virtual memory and using
1627 *	bcopy to copy the page, one machine dependent page at a
1628 *	time.
1629 */
1630void
1631pmap_copy_page(src, dst)
1632	vm_offset_t src;
1633	vm_offset_t dst;
1634{
1635	if (*(int *)CMAP1 || *(int *)CMAP2)
1636		panic("pmap_copy_page: CMAP busy");
1637
1638	*(int *)CMAP1 = PG_V | PG_KW | i386_trunc_page(src);
1639	*(int *)CMAP2 = PG_V | PG_KW | i386_trunc_page(dst);
1640
1641#if __GNUC__ > 1
1642	memcpy(CADDR2, CADDR1, NBPG);
1643#else
1644	bcopy(CADDR1, CADDR2, NBPG);
1645#endif
1646	*(int *)CMAP1 = 0;
1647	*(int *)CMAP2 = 0;
1648	tlbflush();
1649}
1650
1651
1652/*
1653 *	Routine:	pmap_pageable
1654 *	Function:
1655 *		Make the specified pages (by pmap, offset)
1656 *		pageable (or not) as requested.
1657 *
1658 *		A page which is not pageable may not take
1659 *		a fault; therefore, its page table entry
1660 *		must remain valid for the duration.
1661 *
1662 *		This routine is merely advisory; pmap_enter
1663 *		will specify that these pages are to be wired
1664 *		down (or not) as appropriate.
1665 */
1666void
1667pmap_pageable(pmap, sva, eva, pageable)
1668	pmap_t		pmap;
1669	vm_offset_t	sva, eva;
1670	boolean_t	pageable;
1671{
1672}
1673
1674/*
1675 * this routine returns true if a physical page resides
1676 * in the given pmap.
1677 */
1678boolean_t
1679pmap_page_exists(pmap, pa)
1680	pmap_t pmap;
1681	vm_offset_t pa;
1682{
1683	register pv_entry_t pv;
1684	int s;
1685
1686	if (!pmap_is_managed(pa))
1687		return FALSE;
1688
1689	pv = pa_to_pvh(pa);
1690	s = splhigh();
1691
1692	/*
1693	 * Not found, check current mappings returning
1694	 * immediately if found.
1695	 */
1696	if (pv->pv_pmap != NULL) {
1697		for (; pv; pv = pv->pv_next) {
1698			if (pv->pv_pmap == pmap) {
1699				splx(s);
1700				return TRUE;
1701			}
1702		}
1703	}
1704	splx(s);
1705	return(FALSE);
1706}
1707
1708/*
1709 * pmap_testbit tests bits in pte's
1710 * note that the testbit/changebit routines are inline,
1711 * and a lot of things compile-time evaluate.
1712 */
1713static inline boolean_t
1714pmap_testbit(pa, bit)
1715	register vm_offset_t pa;
1716	int bit;
1717{
1718	register pv_entry_t pv;
1719	pt_entry_t *pte;
1720	int s;
1721
1722	if (!pmap_is_managed(pa))
1723		return FALSE;
1724
1725	pv = pa_to_pvh(pa);
1726	s = splhigh();
1727
1728	/*
1729	 * Not found, check current mappings returning
1730	 * immediately if found.
1731	 */
1732	if (pv->pv_pmap != NULL) {
1733		for (; pv; pv = pv->pv_next) {
1734			/*
1735			 * if the bit being tested is the modified bit,
1736			 * then mark UPAGES as always modified, and
1737			 * ptes as never modified.
1738			 */
1739			if (bit & PG_U ) {
1740				if ((pv->pv_va >= clean_sva) && (pv->pv_va < clean_eva)) {
1741					continue;
1742				}
1743			}
1744			if (bit & PG_M ) {
1745				if (pv->pv_va >= USRSTACK) {
1746					if (pv->pv_va >= clean_sva && pv->pv_va < clean_eva) {
1747						continue;
1748					}
1749					if (pv->pv_va < USRSTACK+(UPAGES*NBPG)) {
1750						splx(s);
1751						return TRUE;
1752					}
1753					else if (pv->pv_va < UPT_MAX_ADDRESS) {
1754						splx(s);
1755						return FALSE;
1756					}
1757				}
1758			}
1759			if( !pv->pv_pmap) {
1760				printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va);
1761				continue;
1762			}
1763			pte = pmap_pte(pv->pv_pmap, pv->pv_va);
1764			if ((int) *pte & bit) {
1765				splx(s);
1766				return TRUE;
1767			}
1768		}
1769	}
1770	splx(s);
1771	return(FALSE);
1772}
1773
1774/*
1775 * this routine is used to modify bits in ptes
1776 */
1777static inline void
1778pmap_changebit(pa, bit, setem)
1779	vm_offset_t pa;
1780	int bit;
1781	boolean_t setem;
1782{
1783	register pv_entry_t pv;
1784	register pt_entry_t *pte, npte;
1785	vm_offset_t va;
1786	int s;
1787
1788	if (!pmap_is_managed(pa))
1789		return;
1790
1791	pv = pa_to_pvh(pa);
1792	s = splhigh();
1793
1794	/*
1795	 * Loop over all current mappings setting/clearing as appropos
1796	 * If setting RO do we need to clear the VAC?
1797	 */
1798	if (pv->pv_pmap != NULL) {
1799		for (; pv; pv = pv->pv_next) {
1800			va = pv->pv_va;
1801
1802			/*
1803			 * don't write protect pager mappings
1804			 */
1805			if (!setem && (bit == PG_RW)) {
1806				if (va >= clean_sva && va < clean_eva)
1807					continue;
1808			}
1809
1810			if( !pv->pv_pmap) {
1811				printf("Null pmap (cb) at va: 0x%lx\n", va);
1812				continue;
1813			}
1814			pte = pmap_pte(pv->pv_pmap, va);
1815			if (setem)
1816				(int) npte = (int) *pte | bit;
1817			else
1818				(int) npte = (int) *pte & ~bit;
1819			*pte = npte;
1820		}
1821	}
1822	splx(s);
1823	tlbflush();
1824}
1825
1826/*
1827 *      pmap_page_protect:
1828 *
1829 *      Lower the permission for all mappings to a given page.
1830 */
1831void
1832pmap_page_protect(phys, prot)
1833        vm_offset_t     phys;
1834        vm_prot_t       prot;
1835{
1836	if ((prot & VM_PROT_WRITE) == 0) {
1837		if (prot & (VM_PROT_READ | VM_PROT_EXECUTE))
1838			pmap_changebit(phys, PG_RW, FALSE);
1839		else
1840			pmap_remove_all(phys);
1841	}
1842}
1843
1844/*
1845 *	Clear the modify bits on the specified physical page.
1846 */
1847void
1848pmap_clear_modify(pa)
1849	vm_offset_t	pa;
1850{
1851	pmap_changebit(pa, PG_M, FALSE);
1852}
1853
1854/*
1855 *	pmap_clear_reference:
1856 *
1857 *	Clear the reference bit on the specified physical page.
1858 */
1859void
1860pmap_clear_reference(pa)
1861	vm_offset_t	pa;
1862{
1863	pmap_changebit(pa, PG_U, FALSE);
1864}
1865
1866/*
1867 *	pmap_is_referenced:
1868 *
1869 *	Return whether or not the specified physical page is referenced
1870 *	by any physical maps.
1871 */
1872
1873boolean_t
1874pmap_is_referenced(pa)
1875	vm_offset_t	pa;
1876{
1877	return(pmap_testbit(pa, PG_U));
1878}
1879
1880/*
1881 *	pmap_is_modified:
1882 *
1883 *	Return whether or not the specified physical page is modified
1884 *	by any physical maps.
1885 */
1886
1887boolean_t
1888pmap_is_modified(pa)
1889	vm_offset_t	pa;
1890{
1891	return(pmap_testbit(pa, PG_M));
1892}
1893
1894/*
1895 *	Routine:	pmap_copy_on_write
1896 *	Function:
1897 *		Remove write privileges from all
1898 *		physical maps for this physical page.
1899 */
1900void
1901pmap_copy_on_write(pa)
1902	vm_offset_t pa;
1903{
1904	pmap_changebit(pa, PG_RW, FALSE);
1905}
1906
1907
1908vm_offset_t
1909pmap_phys_address(ppn)
1910	int ppn;
1911{
1912	return(i386_ptob(ppn));
1913}
1914
1915/*
1916 * Miscellaneous support routines follow
1917 */
1918
1919void
1920i386_protection_init()
1921{
1922	register int *kp, prot;
1923
1924	kp = protection_codes;
1925	for (prot = 0; prot < 8; prot++) {
1926		switch (prot) {
1927		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
1928		/*
1929		 * Read access is also 0. There isn't any execute
1930		 * bit, so just make it readable.
1931		 */
1932		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
1933		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
1934		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
1935			*kp++ = 0;
1936			break;
1937		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
1938		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
1939		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
1940		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
1941			*kp++ = PG_RW;
1942			break;
1943		}
1944	}
1945}
1946
1947#ifdef DEBUG
1948/* print address space of pmap*/
1949void
1950pads(pm)
1951	pmap_t pm;
1952{
1953	unsigned va, i, j;
1954	pt_entry_t *ptep;
1955
1956	if (pm == kernel_pmap) return;
1957	for (i = 0; i < 1024; i++)
1958		if (pm->pm_pdir[i])
1959			for (j = 0; j < 1024 ; j++) {
1960				va = (i<<PD_SHIFT)+(j<<PG_SHIFT);
1961				if (pm == kernel_pmap && va < KERNBASE)
1962						continue;
1963				if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
1964						continue;
1965				ptep = pmap_pte(pm, va);
1966				if (pmap_pte_v(ptep))
1967					printf("%x:%x ", va, *(int *)ptep);
1968			} ;
1969
1970}
1971
1972void
1973pmap_pvdump(pa)
1974	vm_offset_t pa;
1975{
1976	register pv_entry_t pv;
1977
1978	printf("pa %x", pa);
1979	for (pv = pa_to_pvh(pa); pv; pv = pv->pv_next) {
1980#ifdef used_to_be
1981		printf(" -> pmap %x, va %x, flags %x",
1982		       pv->pv_pmap, pv->pv_va, pv->pv_flags);
1983#endif
1984		printf(" -> pmap %x, va %x",
1985		       pv->pv_pmap, pv->pv_va);
1986		pads(pv->pv_pmap);
1987	}
1988	printf(" ");
1989}
1990#endif
1991
1992