pmap.c revision 30701
159191Skris/*
259191Skris * Copyright (c) 1991 Regents of the University of California.
359191Skris * All rights reserved.
459191Skris * Copyright (c) 1994 John S. Dyson
559191Skris * All rights reserved.
659191Skris * Copyright (c) 1994 David Greenman
759191Skris * All rights reserved.
859191Skris *
959191Skris * This code is derived from software contributed to Berkeley by
1059191Skris * the Systems Programming Group of the University of Utah Computer
1159191Skris * Science Department and William Jolitz of UUNET Technologies Inc.
1259191Skris *
1359191Skris * Redistribution and use in source and binary forms, with or without
1459191Skris * modification, are permitted provided that the following conditions
1559191Skris * are met:
1659191Skris * 1. Redistributions of source code must retain the above copyright
17109998Smarkm *    notice, this list of conditions and the following disclaimer.
18109998Smarkm * 2. Redistributions in binary form must reproduce the above copyright
1959191Skris *    notice, this list of conditions and the following disclaimer in the
2059191Skris *    documentation and/or other materials provided with the distribution.
2159191Skris * 3. All advertising materials mentioning features or use of this software
2259191Skris *    must display the following acknowledgement:
2359191Skris *	This product includes software developed by the University of
2459191Skris *	California, Berkeley and its contributors.
2559191Skris * 4. Neither the name of the University nor the names of its contributors
2659191Skris *    may be used to endorse or promote products derived from this software
2759191Skris *    without specific prior written permission.
2859191Skris *
2959191Skris * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
3059191Skris * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
3159191Skris * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
3259191Skris * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33109998Smarkm * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34109998Smarkm * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35109998Smarkm * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
3659191Skris * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
3759191Skris * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
3859191Skris * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
3959191Skris * SUCH DAMAGE.
4059191Skris *
4159191Skris *	from:	@(#)pmap.c	7.7 (Berkeley)	5/12/91
42 *	$Id: pmap.c,v 1.164 1997/10/24 23:41:04 dyson Exp $
43 */
44
45/*
46 *	Manages physical address maps.
47 *
48 *	In addition to hardware address maps, this
49 *	module is called upon to provide software-use-only
50 *	maps which may or may not be stored in the same
51 *	form as hardware maps.  These pseudo-maps are
52 *	used to store intermediate results from copy
53 *	operations to and from address spaces.
54 *
55 *	Since the information managed by this module is
56 *	also stored by the logical address mapping module,
57 *	this module may throw away valid virtual-to-physical
58 *	mappings at almost any time.  However, invalidations
59 *	of virtual-to-physical mappings must be done as
60 *	requested.
61 *
62 *	In order to cope with hardware architectures which
63 *	make virtual-to-physical map invalidates expensive,
64 *	this module may delay invalidate or reduced protection
65 *	operations until such time as they are actually
66 *	necessary.  This module is given full information as
67 *	to which processors are currently using which maps,
68 *	and to when physical maps must be made correct.
69 */
70
71#include "opt_cpu.h"
72
73#include <sys/param.h>
74#include <sys/systm.h>
75#include <sys/proc.h>
76#include <sys/malloc.h>
77#include <sys/msgbuf.h>
78#include <sys/vmmeter.h>
79#include <sys/mman.h>
80
81#include <vm/vm.h>
82#include <vm/vm_param.h>
83#include <vm/vm_prot.h>
84#include <sys/lock.h>
85#include <vm/vm_kern.h>
86#include <vm/vm_page.h>
87#include <vm/vm_map.h>
88#include <vm/vm_object.h>
89#include <vm/vm_extern.h>
90#include <vm/vm_pageout.h>
91#include <vm/vm_pager.h>
92#include <vm/vm_zone.h>
93
94#include <sys/user.h>
95
96#include <machine/cpu.h>
97#include <machine/cputypes.h>
98#include <machine/md_var.h>
99#include <machine/specialreg.h>
100#if defined(SMP) || defined(APIC_IO)
101#include <machine/smp.h>
102#include <machine/apic.h>
103#endif /* SMP || APIC_IO */
104
105#define PMAP_KEEP_PDIRS
106#ifndef PMAP_SHPGPERPROC
107#define PMAP_SHPGPERPROC 200
108#endif
109
110#if defined(DIAGNOSTIC)
111#define PMAP_DIAGNOSTIC
112#endif
113
114#if !defined(PMAP_DIAGNOSTIC)
115#define PMAP_INLINE __inline
116#else
117#define PMAP_INLINE
118#endif
119
120#define PTPHINT
121
122/*
123 * Get PDEs and PTEs for user/kernel address space
124 */
125#define	pmap_pde(m, v)	(&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
126#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
127
128#define pmap_pde_v(pte)		((*(int *)pte & PG_V) != 0)
129#define pmap_pte_w(pte)		((*(int *)pte & PG_W) != 0)
130#define pmap_pte_m(pte)		((*(int *)pte & PG_M) != 0)
131#define pmap_pte_u(pte)		((*(int *)pte & PG_A) != 0)
132#define pmap_pte_v(pte)		((*(int *)pte & PG_V) != 0)
133
134#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
135#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
136
137/*
138 * Given a map and a machine independent protection code,
139 * convert to a vax protection code.
140 */
141#define pte_prot(m, p)	(protection_codes[p])
142static int protection_codes[8];
143
144#define	pa_index(pa)		atop((pa) - vm_first_phys)
145#define	pa_to_pvh(pa)		(&pv_table[pa_index(pa)])
146
147static struct pmap kernel_pmap_store;
148pmap_t kernel_pmap;
149extern pd_entry_t my_idlePTD;
150
151vm_offset_t avail_start;	/* PA of first available physical page */
152vm_offset_t avail_end;		/* PA of last available physical page */
153vm_offset_t virtual_avail;	/* VA of first avail page (after kernel bss) */
154vm_offset_t virtual_end;	/* VA of last avail page (end of kernel AS) */
155static boolean_t pmap_initialized = FALSE;	/* Has pmap_init completed? */
156static vm_offset_t vm_first_phys;
157int pgeflag;		/* PG_G or-in */
158int pseflag;		/* PG_PS or-in */
159int pv_npg;
160
161int nkpt;
162vm_offset_t kernel_vm_end;
163
164extern vm_offset_t clean_sva, clean_eva;
165
166/*
167 * Data for the pv entry allocation mechanism
168 */
169vm_zone_t pvzone;
170struct vm_zone pvzone_store;
171struct vm_object pvzone_obj;
172int pv_entry_count=0, pv_entry_max=0,
173	pv_entry_high_water=0, pv_entry_low_water=0;
174int pmap_pagedaemon_waken = 0;
175#define NPVINIT 8192
176struct pv_entry pvinit[NPVINIT];
177
178/*
179 * All those kernel PT submaps that BSD is so fond of
180 */
181pt_entry_t *CMAP1 = 0;
182static pt_entry_t *CMAP2, *ptmmap;
183static pv_table_t *pv_table;
184caddr_t CADDR1 = 0, ptvmmap = 0;
185static caddr_t CADDR2;
186static pt_entry_t *msgbufmap;
187struct msgbuf *msgbufp=0;
188
189#ifdef SMP
190extern char prv_CPAGE1[], prv_CPAGE2[], prv_CPAGE3[];
191extern pt_entry_t *prv_CMAP1, *prv_CMAP2, *prv_CMAP3;
192extern pd_entry_t *IdlePTDS[];
193extern pt_entry_t SMP_prvpt[];
194#endif
195
196pt_entry_t *PMAP1 = 0;
197unsigned *PADDR1 = 0;
198
199static PMAP_INLINE void	free_pv_entry __P((pv_entry_t pv));
200static unsigned * get_ptbase __P((pmap_t pmap));
201static pv_entry_t get_pv_entry __P((void));
202static void	i386_protection_init __P((void));
203static void	pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem));
204
205static PMAP_INLINE int	pmap_is_managed __P((vm_offset_t pa));
206static void	pmap_remove_all __P((vm_offset_t pa));
207static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va,
208				      vm_offset_t pa, vm_page_t mpte));
209static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq,
210					vm_offset_t sva));
211static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va));
212static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv,
213					vm_offset_t va));
214static boolean_t pmap_testbit __P((vm_offset_t pa, int bit));
215static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va,
216		vm_page_t mpte, vm_offset_t pa));
217
218static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va));
219
220static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p));
221static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex));
222static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va));
223static vm_page_t pmap_page_alloc __P((vm_object_t object, vm_pindex_t pindex));
224static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex));
225static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t));
226vm_offset_t pmap_kmem_choose(vm_offset_t addr) ;
227void pmap_collect(void);
228
229#define PDSTACKMAX 6
230static vm_offset_t pdstack[PDSTACKMAX];
231static int pdstackptr;
232unsigned pdir4mb;
233
234/*
235 *	Routine:	pmap_pte
236 *	Function:
237 *		Extract the page table entry associated
238 *		with the given map/virtual_address pair.
239 */
240
241PMAP_INLINE unsigned *
242pmap_pte(pmap, va)
243	register pmap_t pmap;
244	vm_offset_t va;
245{
246	unsigned *pdeaddr;
247
248	if (pmap) {
249		pdeaddr = (unsigned *) pmap_pde(pmap, va);
250		if (*pdeaddr & PG_PS)
251			return pdeaddr;
252		if (*pdeaddr) {
253			return get_ptbase(pmap) + i386_btop(va);
254		}
255	}
256	return (0);
257}
258
259/*
260 * Move the kernel virtual free pointer to the next
261 * 4MB.  This is used to help improve performance
262 * by using a large (4MB) page for much of the kernel
263 * (.text, .data, .bss)
264 */
265vm_offset_t
266pmap_kmem_choose(vm_offset_t addr) {
267	vm_offset_t newaddr = addr;
268#ifndef DISABLE_PSE
269	if (cpu_feature & CPUID_PSE) {
270		newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
271	}
272#endif
273	return newaddr;
274}
275
276/*
277 *	Bootstrap the system enough to run with virtual memory.
278 *
279 *	On the i386 this is called after mapping has already been enabled
280 *	and just syncs the pmap module with what has already been done.
281 *	[We can't call it easily with mapping off since the kernel is not
282 *	mapped with PA == VA, hence we would have to relocate every address
283 *	from the linked base (virtual) address "KERNBASE" to the actual
284 *	(physical) address starting relative to 0]
285 */
286void
287pmap_bootstrap(firstaddr, loadaddr)
288	vm_offset_t firstaddr;
289	vm_offset_t loadaddr;
290{
291	vm_offset_t va;
292	pt_entry_t *pte;
293	int i, j;
294
295	avail_start = firstaddr;
296
297	/*
298	 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too
299	 * large. It should instead be correctly calculated in locore.s and
300	 * not based on 'first' (which is a physical address, not a virtual
301	 * address, for the start of unused physical memory). The kernel
302	 * page tables are NOT double mapped and thus should not be included
303	 * in this calculation.
304	 */
305	virtual_avail = (vm_offset_t) KERNBASE + firstaddr;
306	virtual_avail = pmap_kmem_choose(virtual_avail);
307
308	virtual_end = VM_MAX_KERNEL_ADDRESS;
309
310	/*
311	 * Initialize protection array.
312	 */
313	i386_protection_init();
314
315	/*
316	 * The kernel's pmap is statically allocated so we don't have to use
317	 * pmap_create, which is unlikely to work correctly at this part of
318	 * the boot sequence (XXX and which no longer exists).
319	 */
320	kernel_pmap = &kernel_pmap_store;
321
322	kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD);
323
324	kernel_pmap->pm_count = 1;
325	TAILQ_INIT(&kernel_pmap->pm_pvlist);
326	nkpt = NKPT;
327
328	/*
329	 * Reserve some special page table entries/VA space for temporary
330	 * mapping of pages.
331	 */
332#define	SYSMAP(c, p, v, n)	\
333	v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
334
335	va = virtual_avail;
336	pte = (pt_entry_t *) pmap_pte(kernel_pmap, va);
337
338	/*
339	 * CMAP1/CMAP2 are used for zeroing and copying pages.
340	 */
341	SYSMAP(caddr_t, CMAP1, CADDR1, 1)
342	SYSMAP(caddr_t, CMAP2, CADDR2, 1)
343
344	/*
345	 * ptvmmap is used for reading arbitrary physical pages via /dev/mem.
346	 * XXX ptmmap is not used.
347	 */
348	SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
349
350	/*
351	 * msgbufp is used to map the system message buffer.
352	 * XXX msgbufmap is not used.
353	 */
354	SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
355	       atop(round_page(sizeof(struct msgbuf))))
356
357	/*
358	 * ptemap is used for pmap_pte_quick
359	 */
360	SYSMAP(unsigned *, PMAP1, PADDR1, 1);
361
362	virtual_avail = va;
363
364	*(int *) CMAP1 = *(int *) CMAP2 = 0;
365	*(int *) PTD = 0;
366
367
368	pgeflag = 0;
369#if !defined(SMP)
370	if (cpu_feature & CPUID_PGE) {
371		pgeflag = PG_G;
372	}
373#endif
374
375/*
376 * Initialize the 4MB page size flag
377 */
378	pseflag = 0;
379/*
380 * The 4MB page version of the initial
381 * kernel page mapping.
382 */
383	pdir4mb = 0;
384
385#if !defined(DISABLE_PSE)
386	if (cpu_feature & CPUID_PSE) {
387		unsigned ptditmp;
388		/*
389		 * Enable the PSE mode
390		 */
391		load_cr4(rcr4() | CR4_PSE);
392
393		/*
394		 * Note that we have enabled PSE mode
395		 */
396		pseflag = PG_PS;
397		ptditmp = (unsigned) kernel_pmap->pm_pdir[KPTDI];
398		ptditmp &= ~(NBPDR - 1);
399		ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
400		pdir4mb = ptditmp;
401		/*
402		 * We can do the mapping here for the single processor
403		 * case.  We simply ignore the old page table page from
404		 * now on.
405		 */
406#if !defined(SMP)
407		PTD[KPTDI] = (pd_entry_t) ptditmp;
408		kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp;
409		invltlb();
410#endif
411	}
412#endif
413
414#ifdef SMP
415	if (cpu_apic_address == 0)
416		panic("pmap_bootstrap: no local apic!");
417
418	/* 0 = private page */
419	/* 1 = page table page */
420	/* 2 = local apic */
421	/* 16-31 = io apics */
422	SMP_prvpt[2] = (pt_entry_t)(PG_V | PG_RW | pgeflag | ((u_long)cpu_apic_address & PG_FRAME));
423
424	for (i = 0; i < mp_napics; i++) {
425		for (j = 0; j < 16; j++) {
426			/* same page frame as a previous IO apic? */
427			if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) ==
428			    ((u_long)io_apic_address[0] & PG_FRAME)) {
429				ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE];
430				break;
431			}
432			/* use this slot if available */
433			if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 0) {
434				SMP_prvpt[j + 16] = (pt_entry_t)(PG_V | PG_RW | pgeflag |
435				    ((u_long)io_apic_address[i] & PG_FRAME));
436				ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE];
437				break;
438			}
439		}
440		if (j == 16)
441			panic("no space to map IO apic %d!", i);
442	}
443
444	/* BSP does this itself, AP's get it pre-set */
445	prv_CMAP1 = (pt_entry_t *)&SMP_prvpt[5];
446	prv_CMAP2 = (pt_entry_t *)&SMP_prvpt[6];
447	prv_CMAP3 = (pt_entry_t *)&SMP_prvpt[7];
448#endif
449
450	invltlb();
451
452}
453
454/*
455 * Set 4mb pdir for mp startup, and global flags
456 */
457void
458pmap_set_opt(unsigned *pdir) {
459	int i;
460
461	if (pseflag && (cpu_feature & CPUID_PSE)) {
462		load_cr4(rcr4() | CR4_PSE);
463		if (pdir4mb) {
464			(unsigned) pdir[KPTDI] = pdir4mb;
465		}
466	}
467
468	if (pgeflag && (cpu_feature & CPUID_PGE)) {
469		load_cr4(rcr4() | CR4_PGE);
470		for(i = KPTDI; i < KPTDI + nkpt; i++) {
471			if (pdir[i]) {
472				pdir[i] |= PG_G;
473			}
474		}
475	}
476}
477
478/*
479 * Setup the PTD for the boot processor
480 */
481void
482pmap_set_opt_bsp(void)
483{
484	pmap_set_opt((unsigned *)kernel_pmap->pm_pdir);
485	pmap_set_opt((unsigned *)PTD);
486	invltlb();
487}
488
489/*
490 *	Initialize the pmap module.
491 *	Called by vm_init, to initialize any structures that the pmap
492 *	system needs to map virtual memory.
493 *	pmap_init has been enhanced to support in a fairly consistant
494 *	way, discontiguous physical memory.
495 */
496void
497pmap_init(phys_start, phys_end)
498	vm_offset_t phys_start, phys_end;
499{
500	vm_offset_t addr;
501	vm_size_t s;
502	int i;
503
504	/*
505	 * calculate the number of pv_entries needed
506	 */
507	vm_first_phys = phys_avail[0];
508	for (i = 0; phys_avail[i + 1]; i += 2);
509	pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE;
510
511	/*
512	 * Allocate memory for random pmap data structures.  Includes the
513	 * pv_head_table.
514	 */
515	s = (vm_size_t) (sizeof(pv_table_t) * pv_npg);
516	s = round_page(s);
517
518	addr = (vm_offset_t) kmem_alloc(kernel_map, s);
519	pv_table = (pv_table_t *) addr;
520	for(i = 0; i < pv_npg; i++) {
521		vm_offset_t pa;
522		TAILQ_INIT(&pv_table[i].pv_list);
523		pv_table[i].pv_list_count = 0;
524		pa = vm_first_phys + i * PAGE_SIZE;
525		pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa);
526	}
527
528	/*
529	 * init the pv free list
530	 */
531	pvzone = &pvzone_store;
532	zbootinit(pvzone, "PV ENTRY", sizeof(pvinit[0]), pvinit, NPVINIT);
533
534	/*
535	 * Now it is safe to enable pv_table recording.
536	 */
537	pmap_initialized = TRUE;
538}
539
540void
541pmap_init2() {
542<<<<<<< pmap.c
543	pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg;
544	pv_entry_high_water = 9 * (pv_entry_max / 10);
545	pv_entry_low_water = 4 * (pv_entry_max / 10);
546	zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
547=======
548	zinitna(pvzone, &pvzone_obj, NULL, 0,
549		PMAP_SHPGPERPROC * maxproc + pv_npg, ZONE_INTERRUPT, 1);
550>>>>>>> 1.164
551}
552
553/*
554 *	Used to map a range of physical addresses into kernel
555 *	virtual address space.
556 *
557 *	For now, VM is already on, we only need to map the
558 *	specified memory.
559 */
560vm_offset_t
561pmap_map(virt, start, end, prot)
562	vm_offset_t virt;
563	vm_offset_t start;
564	vm_offset_t end;
565	int prot;
566{
567	while (start < end) {
568		pmap_enter(kernel_pmap, virt, start, prot, FALSE);
569		virt += PAGE_SIZE;
570		start += PAGE_SIZE;
571	}
572	return (virt);
573}
574
575
576/***************************************************
577 * Low level helper routines.....
578 ***************************************************/
579
580#if defined(PMAP_DIAGNOSTIC)
581
582/*
583 * This code checks for non-writeable/modified pages.
584 * This should be an invalid condition.
585 */
586static int
587pmap_nw_modified(pt_entry_t ptea) {
588	int pte;
589
590	pte = (int) ptea;
591
592	if ((pte & (PG_M|PG_RW)) == PG_M)
593		return 1;
594	else
595		return 0;
596}
597#endif
598
599
600/*
601 * this routine defines the region(s) of memory that should
602 * not be tested for the modified bit.
603 */
604static PMAP_INLINE int
605pmap_track_modified( vm_offset_t va) {
606	if ((va < clean_sva) || (va >= clean_eva))
607		return 1;
608	else
609		return 0;
610}
611
612static PMAP_INLINE void
613invltlb_1pg( vm_offset_t va) {
614#if defined(I386_CPU)
615	if (cpu_class == CPUCLASS_386) {
616		invltlb();
617	} else
618#endif
619	{
620		invlpg(va);
621	}
622}
623
624static PMAP_INLINE void
625invltlb_2pg( vm_offset_t va1, vm_offset_t va2) {
626#if defined(I386_CPU)
627	if (cpu_class == CPUCLASS_386) {
628		invltlb();
629	} else
630#endif
631	{
632		invlpg(va1);
633		invlpg(va2);
634	}
635}
636
637static unsigned *
638get_ptbase(pmap)
639	pmap_t pmap;
640{
641	unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
642
643	/* are we current address space or kernel? */
644	if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
645		return (unsigned *) PTmap;
646	}
647	/* otherwise, we are alternate address space */
648	if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
649		APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
650		invltlb();
651	}
652	return (unsigned *) APTmap;
653}
654
655/*
656 * Super fast pmap_pte routine best used when scanning
657 * the pv lists.  This eliminates many coarse-grained
658 * invltlb calls.  Note that many of the pv list
659 * scans are across different pmaps.  It is very wasteful
660 * to do an entire invltlb for checking a single mapping.
661 */
662
663static unsigned *
664pmap_pte_quick(pmap, va)
665	register pmap_t pmap;
666	vm_offset_t va;
667{
668	unsigned pde, newpf;
669	if (pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) {
670		unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
671		unsigned index = i386_btop(va);
672		/* are we current address space or kernel? */
673		if ((pmap == kernel_pmap) ||
674			(frame == (((unsigned) PTDpde) & PG_FRAME))) {
675			return (unsigned *) PTmap + index;
676		}
677		newpf = pde & PG_FRAME;
678		if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) {
679			* (unsigned *) PMAP1 = newpf | PG_RW | PG_V;
680			invltlb_1pg((vm_offset_t) PADDR1);
681		}
682		return PADDR1 + ((unsigned) index & (NPTEPG - 1));
683	}
684	return (0);
685}
686
687/*
688 *	Routine:	pmap_extract
689 *	Function:
690 *		Extract the physical page address associated
691 *		with the given map/virtual_address pair.
692 */
693vm_offset_t
694pmap_extract(pmap, va)
695	register pmap_t pmap;
696	vm_offset_t va;
697{
698	vm_offset_t rtval;
699	vm_offset_t pdirindex;
700	pdirindex = va >> PDRSHIFT;
701	if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
702		unsigned *pte;
703		if ((rtval & PG_PS) != 0) {
704			rtval &= ~(NBPDR - 1);
705			rtval |= va & (NBPDR - 1);
706			return rtval;
707		}
708		pte = get_ptbase(pmap) + i386_btop(va);
709		rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
710		return rtval;
711	}
712	return 0;
713
714}
715
716/*
717 * determine if a page is managed (memory vs. device)
718 */
719static PMAP_INLINE int
720pmap_is_managed(pa)
721	vm_offset_t pa;
722{
723	int i;
724
725	if (!pmap_initialized)
726		return 0;
727
728	for (i = 0; phys_avail[i + 1]; i += 2) {
729		if (pa < phys_avail[i + 1] && pa >= phys_avail[i])
730			return 1;
731	}
732	return 0;
733}
734
735
736/***************************************************
737 * Low level mapping routines.....
738 ***************************************************/
739
740/*
741 * Add a list of wired pages to the kva
742 * this routine is only used for temporary
743 * kernel mappings that do not need to have
744 * page modification or references recorded.
745 * Note that old mappings are simply written
746 * over.  The page *must* be wired.
747 */
748void
749pmap_qenter(va, m, count)
750	vm_offset_t va;
751	vm_page_t *m;
752	int count;
753{
754	int i;
755	register unsigned *pte;
756
757	for (i = 0; i < count; i++) {
758		vm_offset_t tva = va + i * PAGE_SIZE;
759		unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | pgeflag;
760		unsigned opte;
761		pte = (unsigned *)vtopte(tva);
762		opte = *pte;
763		*pte = npte;
764		if (opte)
765			invltlb_1pg(tva);
766	}
767}
768
769/*
770 * this routine jerks page mappings from the
771 * kernel -- it is meant only for temporary mappings.
772 */
773void
774pmap_qremove(va, count)
775	vm_offset_t va;
776	int count;
777{
778	int i;
779	register unsigned *pte;
780
781	for (i = 0; i < count; i++) {
782		pte = (unsigned *)vtopte(va);
783		*pte = 0;
784		invltlb_1pg(va);
785		va += PAGE_SIZE;
786	}
787}
788
789/*
790 * add a wired page to the kva
791 * note that in order for the mapping to take effect -- you
792 * should do a invltlb after doing the pmap_kenter...
793 */
794PMAP_INLINE void
795pmap_kenter(va, pa)
796	vm_offset_t va;
797	register vm_offset_t pa;
798{
799	register unsigned *pte;
800	unsigned npte, opte;
801
802	npte = pa | PG_RW | PG_V | pgeflag;
803	pte = (unsigned *)vtopte(va);
804	opte = *pte;
805	*pte = npte;
806	if (opte)
807		invltlb_1pg(va);
808}
809
810/*
811 * remove a page from the kernel pagetables
812 */
813PMAP_INLINE void
814pmap_kremove(va)
815	vm_offset_t va;
816{
817	register unsigned *pte;
818
819	pte = (unsigned *)vtopte(va);
820	*pte = 0;
821	invltlb_1pg(va);
822}
823
824static vm_page_t
825pmap_page_alloc(object, pindex)
826	vm_object_t object;
827	vm_pindex_t pindex;
828{
829	vm_page_t m;
830	m = vm_page_alloc(object, pindex, VM_ALLOC_ZERO);
831	if (m == NULL) {
832		VM_WAIT;
833	}
834	return m;
835}
836
837static vm_page_t
838pmap_page_lookup(object, pindex)
839	vm_object_t object;
840	vm_pindex_t pindex;
841{
842	vm_page_t m;
843retry:
844	m = vm_page_lookup(object, pindex);
845	if (m) {
846		if (m->flags & PG_BUSY) {
847			m->flags |= PG_WANTED;
848			tsleep(m, PVM, "pplookp", 0);
849			goto retry;
850		}
851	}
852
853	return m;
854}
855
856/*
857 * Create the UPAGES for a new process.
858 * This routine directly affects the fork perf for a process.
859 */
860void
861pmap_new_proc(p)
862	struct proc *p;
863{
864	int i;
865	vm_object_t upobj;
866	vm_page_t m;
867	struct user *up;
868	unsigned *ptek;
869
870	/*
871	 * allocate object for the upages
872	 */
873	upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES);
874	p->p_upages_obj = upobj;
875
876	/* get a kernel virtual address for the UPAGES for this proc */
877	up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * PAGE_SIZE);
878	if (up == NULL)
879		panic("pmap_new_proc: u_map allocation failed");
880
881	ptek = (unsigned *) vtopte((vm_offset_t) up);
882
883	for(i=0;i<UPAGES;i++) {
884		/*
885		 * Get a kernel stack page
886		 */
887		while ((m = vm_page_alloc(upobj,
888			i, VM_ALLOC_NORMAL)) == NULL) {
889			VM_WAIT;
890		}
891
892		/*
893		 * Wire the page
894		 */
895		m->wire_count++;
896		++cnt.v_wire_count;
897
898		/*
899		 * Enter the page into the kernel address space.
900		 */
901		*(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag;
902
903		m->flags &= ~(PG_ZERO|PG_BUSY);
904		m->flags |= PG_MAPPED|PG_WRITEABLE;
905		m->valid = VM_PAGE_BITS_ALL;
906	}
907
908	p->p_addr = up;
909}
910
911/*
912 * Dispose the UPAGES for a process that has exited.
913 * This routine directly impacts the exit perf of a process.
914 */
915void
916pmap_dispose_proc(p)
917	struct proc *p;
918{
919	int i;
920	vm_object_t upobj;
921	vm_page_t m;
922	unsigned *ptek;
923
924	ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr);
925
926	upobj = p->p_upages_obj;
927
928	for(i=0;i<UPAGES;i++) {
929		unsigned oldpte;
930		if ((m = vm_page_lookup(upobj, i)) == NULL)
931			panic("pmap_dispose_proc: upage already missing???");
932		oldpte = *(ptek + i);
933		*(ptek + i) = 0;
934		if (oldpte & PG_G)
935			invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE);
936		vm_page_unwire(m);
937		vm_page_free(m);
938	}
939
940	vm_object_deallocate(upobj);
941
942	kmem_free(u_map, (vm_offset_t)p->p_addr, ctob(UPAGES));
943}
944
945/*
946 * Allow the UPAGES for a process to be prejudicially paged out.
947 */
948void
949pmap_swapout_proc(p)
950	struct proc *p;
951{
952	int i;
953	vm_object_t upobj;
954	vm_page_t m;
955
956	upobj = p->p_upages_obj;
957	/*
958	 * let the upages be paged
959	 */
960	for(i=0;i<UPAGES;i++) {
961		if ((m = vm_page_lookup(upobj, i)) == NULL)
962			panic("pmap_swapout_proc: upage already missing???");
963		m->dirty = VM_PAGE_BITS_ALL;
964		vm_page_unwire(m);
965		vm_page_deactivate(m);
966		pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
967	}
968}
969
970/*
971 * Bring the UPAGES for a specified process back in.
972 */
973void
974pmap_swapin_proc(p)
975	struct proc *p;
976{
977	int i;
978	vm_object_t upobj;
979	vm_page_t m;
980	unsigned *pte;
981
982	upobj = p->p_upages_obj;
983	for(i=0;i<UPAGES;i++) {
984		int s;
985		s = splvm();
986retry:
987		if ((m = vm_page_lookup(upobj, i)) == NULL) {
988			if ((m = vm_page_alloc(upobj, i, VM_ALLOC_NORMAL)) == NULL) {
989				VM_WAIT;
990				goto retry;
991			}
992		} else {
993			if ((m->flags & PG_BUSY) || m->busy) {
994				m->flags |= PG_WANTED;
995				tsleep(m, PVM, "swinuw",0);
996				goto retry;
997			}
998			m->flags |= PG_BUSY;
999		}
1000		vm_page_wire(m);
1001		splx(s);
1002
1003		pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
1004			VM_PAGE_TO_PHYS(m));
1005
1006		if (m->valid != VM_PAGE_BITS_ALL) {
1007			int rv;
1008			rv = vm_pager_get_pages(upobj, &m, 1, 0);
1009			if (rv != VM_PAGER_OK)
1010				panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
1011			m->valid = VM_PAGE_BITS_ALL;
1012		}
1013		PAGE_WAKEUP(m);
1014		m->flags |= PG_MAPPED|PG_WRITEABLE;
1015	}
1016}
1017
1018/***************************************************
1019 * Page table page management routines.....
1020 ***************************************************/
1021
1022/*
1023 * This routine unholds page table pages, and if the hold count
1024 * drops to zero, then it decrements the wire count.
1025 */
1026static int
1027_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
1028	int s;
1029
1030	if (m->flags & PG_BUSY) {
1031		s = splvm();
1032		while (m->flags & PG_BUSY) {
1033			m->flags |= PG_WANTED;
1034			tsleep(m, PVM, "pmuwpt", 0);
1035		}
1036		splx(s);
1037	}
1038
1039	if (m->hold_count == 0) {
1040		vm_offset_t pteva;
1041		/*
1042		 * unmap the page table page
1043		 */
1044		pmap->pm_pdir[m->pindex] = 0;
1045		--pmap->pm_stats.resident_count;
1046		if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1047			(((unsigned) PTDpde) & PG_FRAME)) {
1048			/*
1049			 * Do a invltlb to make the invalidated mapping
1050			 * take effect immediately.
1051			 */
1052			pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
1053			invltlb_1pg(pteva);
1054		}
1055
1056#if defined(PTPHINT)
1057		if (pmap->pm_ptphint == m)
1058			pmap->pm_ptphint = NULL;
1059#endif
1060
1061		/*
1062		 * If the page is finally unwired, simply free it.
1063		 */
1064		--m->wire_count;
1065		if (m->wire_count == 0) {
1066
1067			if (m->flags & PG_WANTED) {
1068				m->flags &= ~PG_WANTED;
1069				wakeup(m);
1070			}
1071
1072			vm_page_free_zero(m);
1073			--cnt.v_wire_count;
1074		}
1075		return 1;
1076	}
1077	return 0;
1078}
1079
1080__inline static int
1081pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
1082	vm_page_unhold(m);
1083	if (m->hold_count == 0)
1084		return _pmap_unwire_pte_hold(pmap, m);
1085	else
1086		return 0;
1087}
1088
1089/*
1090 * After removing a page table entry, this routine is used to
1091 * conditionally free the page, and manage the hold/wire counts.
1092 */
1093static int
1094pmap_unuse_pt(pmap, va, mpte)
1095	pmap_t pmap;
1096	vm_offset_t va;
1097	vm_page_t mpte;
1098{
1099	unsigned ptepindex;
1100	if (va >= UPT_MIN_ADDRESS)
1101		return 0;
1102
1103	if (mpte == NULL) {
1104		ptepindex = (va >> PDRSHIFT);
1105#if defined(PTPHINT)
1106		if (pmap->pm_ptphint &&
1107			(pmap->pm_ptphint->pindex == ptepindex)) {
1108			mpte = pmap->pm_ptphint;
1109		} else {
1110			mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1111			pmap->pm_ptphint = mpte;
1112		}
1113#else
1114		mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1115#endif
1116	}
1117
1118	return pmap_unwire_pte_hold(pmap, mpte);
1119}
1120
1121/*
1122 * Initialize a preallocated and zeroed pmap structure,
1123 * such as one in a vmspace structure.
1124 */
1125void
1126pmap_pinit(pmap)
1127	register struct pmap *pmap;
1128{
1129	vm_page_t ptdpg;
1130	/*
1131	 * No need to allocate page table space yet but we do need a valid
1132	 * page directory table.
1133	 */
1134
1135	if (pdstackptr > 0) {
1136		--pdstackptr;
1137		pmap->pm_pdir = (pd_entry_t *)pdstack[pdstackptr];
1138	} else {
1139		pmap->pm_pdir =
1140			(pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1141	}
1142
1143	/*
1144	 * allocate object for the ptes
1145	 */
1146	pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1147
1148	/*
1149	 * allocate the page directory page
1150	 */
1151retry:
1152	ptdpg = pmap_page_alloc( pmap->pm_pteobj, PTDPTDI);
1153	if (ptdpg == NULL)
1154		goto retry;
1155
1156	ptdpg->wire_count = 1;
1157	++cnt.v_wire_count;
1158
1159	ptdpg->flags &= ~(PG_MAPPED|PG_BUSY);	/* not mapped normally */
1160	ptdpg->valid = VM_PAGE_BITS_ALL;
1161
1162	pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1163	if ((ptdpg->flags & PG_ZERO) == 0)
1164		bzero(pmap->pm_pdir, PAGE_SIZE);
1165
1166	/* wire in kernel global address entries */
1167	/* XXX copies current process, does not fill in MPPTDI */
1168	bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1169
1170	/* install self-referential address mapping entry */
1171	*(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1172		VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW;
1173
1174	pmap->pm_flags = 0;
1175	pmap->pm_count = 1;
1176	pmap->pm_ptphint = NULL;
1177	TAILQ_INIT(&pmap->pm_pvlist);
1178}
1179
1180static int
1181pmap_release_free_page(pmap, p)
1182	struct pmap *pmap;
1183	vm_page_t p;
1184{
1185	int s;
1186	unsigned *pde = (unsigned *) pmap->pm_pdir;
1187	/*
1188	 * This code optimizes the case of freeing non-busy
1189	 * page-table pages.  Those pages are zero now, and
1190	 * might as well be placed directly into the zero queue.
1191	 */
1192	s = splvm();
1193	if (p->flags & PG_BUSY) {
1194		p->flags |= PG_WANTED;
1195		tsleep(p, PVM, "pmaprl", 0);
1196		splx(s);
1197		return 0;
1198	}
1199
1200	if (p->flags & PG_WANTED) {
1201		p->flags &= ~PG_WANTED;
1202		wakeup(p);
1203	}
1204
1205	/*
1206	 * Remove the page table page from the processes address space.
1207	 */
1208	pde[p->pindex] = 0;
1209	--pmap->pm_stats.resident_count;
1210
1211	if (p->hold_count)  {
1212		panic("pmap_release: freeing held page table page");
1213	}
1214	/*
1215	 * Page directory pages need to have the kernel
1216	 * stuff cleared, so they can go into the zero queue also.
1217	 */
1218	if (p->pindex == PTDPTDI) {
1219		bzero(pde + KPTDI, nkpt * PTESIZE);
1220#ifdef SMP
1221		pde[MPPTDI] = 0;
1222#endif
1223		pde[APTDPTDI] = 0;
1224		pmap_kremove((vm_offset_t) pmap->pm_pdir);
1225	}
1226
1227#if defined(PTPHINT)
1228	if (pmap->pm_ptphint &&
1229		(pmap->pm_ptphint->pindex == p->pindex))
1230		pmap->pm_ptphint = NULL;
1231#endif
1232
1233	vm_page_free_zero(p);
1234	splx(s);
1235	return 1;
1236}
1237
1238/*
1239 * this routine is called if the page table page is not
1240 * mapped correctly.
1241 */
1242static vm_page_t
1243_pmap_allocpte(pmap, ptepindex)
1244	pmap_t	pmap;
1245	unsigned ptepindex;
1246{
1247	vm_offset_t pteva, ptepa;
1248	vm_page_t m;
1249	int needszero = 0;
1250
1251	/*
1252	 * Find or fabricate a new pagetable page
1253	 */
1254retry:
1255	m = vm_page_lookup(pmap->pm_pteobj, ptepindex);
1256	if (m == NULL) {
1257		m = pmap_page_alloc(pmap->pm_pteobj, ptepindex);
1258		if (m == NULL)
1259			goto retry;
1260		if ((m->flags & PG_ZERO) == 0)
1261			needszero = 1;
1262		m->flags &= ~(PG_ZERO|PG_BUSY);
1263		m->valid = VM_PAGE_BITS_ALL;
1264	} else {
1265		if ((m->flags & PG_BUSY) || m->busy) {
1266			m->flags |= PG_WANTED;
1267			tsleep(m, PVM, "ptewai", 0);
1268			goto retry;
1269		}
1270	}
1271
1272	if (m->queue != PQ_NONE) {
1273		int s = splvm();
1274		vm_page_unqueue(m);
1275		splx(s);
1276	}
1277
1278	if (m->wire_count == 0)
1279		++cnt.v_wire_count;
1280	++m->wire_count;
1281
1282	/*
1283	 * Increment the hold count for the page table page
1284	 * (denoting a new mapping.)
1285	 */
1286	++m->hold_count;
1287
1288	/*
1289	 * Map the pagetable page into the process address space, if
1290	 * it isn't already there.
1291	 */
1292
1293	pmap->pm_stats.resident_count++;
1294
1295	ptepa = VM_PAGE_TO_PHYS(m);
1296	pmap->pm_pdir[ptepindex] = (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V);
1297
1298#if defined(PTPHINT)
1299	/*
1300	 * Set the page table hint
1301	 */
1302	pmap->pm_ptphint = m;
1303#endif
1304
1305	/*
1306	 * Try to use the new mapping, but if we cannot, then
1307	 * do it with the routine that maps the page explicitly.
1308	 */
1309	if (needszero) {
1310		if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1311			(((unsigned) PTDpde) & PG_FRAME)) {
1312			pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1313			bzero((caddr_t) pteva, PAGE_SIZE);
1314		} else {
1315			pmap_zero_page(ptepa);
1316		}
1317	}
1318
1319	m->valid = VM_PAGE_BITS_ALL;
1320	m->flags |= PG_MAPPED;
1321
1322	return m;
1323}
1324
1325static vm_page_t
1326pmap_allocpte(pmap, va)
1327	pmap_t	pmap;
1328	vm_offset_t va;
1329{
1330	unsigned ptepindex;
1331	vm_offset_t ptepa;
1332	vm_page_t m;
1333
1334	/*
1335	 * Calculate pagetable page index
1336	 */
1337	ptepindex = va >> PDRSHIFT;
1338
1339	/*
1340	 * Get the page directory entry
1341	 */
1342	ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1343
1344	/*
1345	 * This supports switching from a 4MB page to a
1346	 * normal 4K page.
1347	 */
1348	if (ptepa & PG_PS) {
1349		pmap->pm_pdir[ptepindex] = 0;
1350		ptepa = 0;
1351		invltlb();
1352	}
1353
1354	/*
1355	 * If the page table page is mapped, we just increment the
1356	 * hold count, and activate it.
1357	 */
1358	if (ptepa) {
1359#if defined(PTPHINT)
1360		/*
1361		 * In order to get the page table page, try the
1362		 * hint first.
1363		 */
1364		if (pmap->pm_ptphint &&
1365			(pmap->pm_ptphint->pindex == ptepindex)) {
1366			m = pmap->pm_ptphint;
1367		} else {
1368			m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1369			pmap->pm_ptphint = m;
1370		}
1371#else
1372		m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1373#endif
1374		++m->hold_count;
1375		return m;
1376	}
1377	/*
1378	 * Here if the pte page isn't mapped, or if it has been deallocated.
1379	 */
1380	return _pmap_allocpte(pmap, ptepindex);
1381}
1382
1383
1384/***************************************************
1385* Pmap allocation/deallocation routines.
1386 ***************************************************/
1387
1388/*
1389 * Release any resources held by the given physical map.
1390 * Called when a pmap initialized by pmap_pinit is being released.
1391 * Should only be called if the map contains no valid mappings.
1392 */
1393void
1394pmap_release(pmap)
1395	register struct pmap *pmap;
1396{
1397	vm_page_t p,n,ptdpg;
1398	vm_object_t object = pmap->pm_pteobj;
1399
1400#if defined(DIAGNOSTIC)
1401	if (object->ref_count != 1)
1402		panic("pmap_release: pteobj reference count != 1");
1403#endif
1404
1405	ptdpg = NULL;
1406retry:
1407	for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1408		n = TAILQ_NEXT(p, listq);
1409		if (p->pindex == PTDPTDI) {
1410			ptdpg = p;
1411			continue;
1412		}
1413		if (!pmap_release_free_page(pmap, p))
1414			goto retry;
1415	}
1416
1417	if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1418		goto retry;
1419
1420	vm_object_deallocate(object);
1421	if (pdstackptr < PDSTACKMAX) {
1422		pdstack[pdstackptr] = (vm_offset_t) pmap->pm_pdir;
1423		++pdstackptr;
1424	} else {
1425		int pdstmp = pdstackptr - 1;
1426		kmem_free(kernel_map, pdstack[pdstmp], PAGE_SIZE);
1427		pdstack[pdstmp] = (vm_offset_t) pmap->pm_pdir;
1428	}
1429	pmap->pm_pdir = 0;
1430}
1431
1432/*
1433 * grow the number of kernel page table entries, if needed
1434 */
1435void
1436pmap_growkernel(vm_offset_t addr)
1437{
1438	struct proc *p;
1439	struct pmap *pmap;
1440	int s;
1441	vm_offset_t ptpkva, ptppaddr;
1442	vm_page_t nkpg;
1443#ifdef SMP
1444	int i;
1445#endif
1446	pd_entry_t newpdir;
1447	vm_pindex_t ptpidx;
1448
1449	s = splhigh();
1450	if (kernel_vm_end == 0) {
1451		kernel_vm_end = KERNBASE;
1452		nkpt = 0;
1453		while (pdir_pde(PTD, kernel_vm_end)) {
1454			kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1455			++nkpt;
1456		}
1457	}
1458	addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1459	while (kernel_vm_end < addr) {
1460		if (pdir_pde(PTD, kernel_vm_end)) {
1461			kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1462			continue;
1463		}
1464		++nkpt;
1465		ptpkva = (vm_offset_t) vtopte(addr);
1466		ptpidx = (ptpkva >> PAGE_SHIFT);
1467		/*
1468		 * This index is bogus, but out of the way
1469		 */
1470		nkpg = vm_page_alloc(kernel_object,
1471			ptpidx, VM_ALLOC_SYSTEM);
1472		if (!nkpg)
1473			panic("pmap_growkernel: no memory to grow kernel");
1474
1475		vm_page_wire(nkpg);
1476		vm_page_remove(nkpg);
1477		ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1478		pmap_zero_page(ptppaddr);
1479		newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW);
1480		pdir_pde(PTD, kernel_vm_end) = newpdir;
1481
1482#ifdef SMP
1483		for (i = 0; i < mp_ncpus; i++) {
1484			if (IdlePTDS[i])
1485				pdir_pde(IdlePTDS[i], kernel_vm_end) = newpdir;
1486		}
1487#endif
1488
1489		for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
1490			if (p->p_vmspace) {
1491				pmap = &p->p_vmspace->vm_pmap;
1492				*pmap_pde(pmap, kernel_vm_end) = newpdir;
1493			}
1494		}
1495		*pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1496		kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1497	}
1498	splx(s);
1499}
1500
1501/*
1502 *	Retire the given physical map from service.
1503 *	Should only be called if the map contains
1504 *	no valid mappings.
1505 */
1506void
1507pmap_destroy(pmap)
1508	register pmap_t pmap;
1509{
1510	int count;
1511
1512	if (pmap == NULL)
1513		return;
1514
1515	count = --pmap->pm_count;
1516	if (count == 0) {
1517		pmap_release(pmap);
1518		panic("destroying a pmap is not yet implemented");
1519		/* free((caddr_t) pmap, M_VMPMAP); */
1520	}
1521}
1522
1523/*
1524 *	Add a reference to the specified pmap.
1525 */
1526void
1527pmap_reference(pmap)
1528	pmap_t pmap;
1529{
1530	if (pmap != NULL) {
1531		pmap->pm_count++;
1532	}
1533}
1534
1535/***************************************************
1536* page management routines.
1537 ***************************************************/
1538
1539/*
1540 * free the pv_entry back to the free list
1541 */
1542static inline void
1543free_pv_entry(pv)
1544	pv_entry_t pv;
1545{
1546	pv_entry_count--;
1547	zfreei(pvzone, pv);
1548}
1549
1550/*
1551 * get a new pv_entry, allocating a block from the system
1552 * when needed.
1553 * the memory allocation is performed bypassing the malloc code
1554 * because of the possibility of allocations at interrupt time.
1555 */
1556static pv_entry_t
1557get_pv_entry(void)
1558{
1559	pv_entry_count++;
1560	if ((pv_entry_count > pv_entry_high_water) &&
1561		(pmap_pagedaemon_waken == 0)) {
1562		pmap_pagedaemon_waken = 1;
1563		wakeup (&vm_pages_needed);
1564	}
1565	return zalloci(pvzone);
1566}
1567
1568/*
1569 * This routine is very drastic, but can save the system
1570 * in a pinch.
1571 */
1572void
1573pmap_collect() {
1574	pv_table_t *ppv;
1575	pv_entry_t pv;
1576	int i;
1577	vm_offset_t pa;
1578	vm_page_t m;
1579	static int warningdone=0;
1580
1581	if (pmap_pagedaemon_waken == 0)
1582		return;
1583
1584	if (warningdone < 5) {
1585		printf("pmap_collect: collecting pv entries -- increase PMAP_SHPGPERPROC");
1586		warningdone++;
1587	}
1588
1589	for(i = 0; i < pv_npg; i++) {
1590		if ((ppv = &pv_table[i]) == 0)
1591			continue;
1592		m = ppv->pv_vm_page;
1593		if ((pa = VM_PAGE_TO_PHYS(m)) == 0)
1594			continue;
1595		if (m->wire_count || m->hold_count || m->busy || (m->flags & PG_BUSY))
1596			continue;
1597		pmap_remove_all(pa);
1598	}
1599	pmap_pagedaemon_waken = 0;
1600}
1601
1602
1603/*
1604 * If it is the first entry on the list, it is actually
1605 * in the header and we must copy the following entry up
1606 * to the header.  Otherwise we must search the list for
1607 * the entry.  In either case we free the now unused entry.
1608 */
1609
1610static int
1611pmap_remove_entry(pmap, ppv, va)
1612	struct pmap *pmap;
1613	pv_table_t *ppv;
1614	vm_offset_t va;
1615{
1616	pv_entry_t pv;
1617	int rtval;
1618	int s;
1619
1620	s = splvm();
1621	if (ppv->pv_list_count < pmap->pm_stats.resident_count) {
1622		for (pv = TAILQ_FIRST(&ppv->pv_list);
1623			pv;
1624			pv = TAILQ_NEXT(pv, pv_list)) {
1625			if (pmap == pv->pv_pmap && va == pv->pv_va)
1626				break;
1627		}
1628	} else {
1629		for (pv = TAILQ_FIRST(&pmap->pm_pvlist);
1630			pv;
1631			pv = TAILQ_NEXT(pv, pv_plist)) {
1632			if (va == pv->pv_va)
1633				break;
1634		}
1635	}
1636
1637	rtval = 0;
1638	if (pv) {
1639		rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1640		TAILQ_REMOVE(&ppv->pv_list, pv, pv_list);
1641		--ppv->pv_list_count;
1642		if (TAILQ_FIRST(&ppv->pv_list) == NULL) {
1643			ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE);
1644		}
1645
1646		TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1647		free_pv_entry(pv);
1648	}
1649
1650	splx(s);
1651	return rtval;
1652}
1653
1654/*
1655 * Create a pv entry for page at pa for
1656 * (pmap, va).
1657 */
1658static void
1659pmap_insert_entry(pmap, va, mpte, pa)
1660	pmap_t pmap;
1661	vm_offset_t va;
1662	vm_page_t mpte;
1663	vm_offset_t pa;
1664{
1665
1666	int s;
1667	pv_entry_t pv;
1668	pv_table_t *ppv;
1669
1670	s = splvm();
1671	pv = get_pv_entry();
1672	pv->pv_va = va;
1673	pv->pv_pmap = pmap;
1674	pv->pv_ptem = mpte;
1675
1676	TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1677
1678	ppv = pa_to_pvh(pa);
1679	TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list);
1680	++ppv->pv_list_count;
1681
1682	splx(s);
1683}
1684
1685/*
1686 * pmap_remove_pte: do the things to unmap a page in a process
1687 */
1688static int
1689pmap_remove_pte(pmap, ptq, va)
1690	struct pmap *pmap;
1691	unsigned *ptq;
1692	vm_offset_t va;
1693{
1694	unsigned oldpte;
1695	pv_table_t *ppv;
1696
1697	oldpte = *ptq;
1698	*ptq = 0;
1699	if (oldpte & PG_W)
1700		pmap->pm_stats.wired_count -= 1;
1701	/*
1702	 * Machines that don't support invlpg, also don't support
1703	 * PG_G.
1704	 */
1705	if (oldpte & PG_G)
1706		invlpg(va);
1707	pmap->pm_stats.resident_count -= 1;
1708	if (oldpte & PG_MANAGED) {
1709		ppv = pa_to_pvh(oldpte);
1710		if (oldpte & PG_M) {
1711#if defined(PMAP_DIAGNOSTIC)
1712			if (pmap_nw_modified((pt_entry_t) oldpte)) {
1713				printf("pmap_remove: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, (int) oldpte);
1714			}
1715#endif
1716			if (pmap_track_modified(va))
1717				ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL;
1718		}
1719		return pmap_remove_entry(pmap, ppv, va);
1720	} else {
1721		return pmap_unuse_pt(pmap, va, NULL);
1722	}
1723
1724	return 0;
1725}
1726
1727/*
1728 * Remove a single page from a process address space
1729 */
1730static void
1731pmap_remove_page(pmap, va)
1732	struct pmap *pmap;
1733	register vm_offset_t va;
1734{
1735	register unsigned *ptq;
1736
1737	/*
1738	 * if there is no pte for this address, just skip it!!!
1739	 */
1740	if (*pmap_pde(pmap, va) == 0) {
1741		return;
1742	}
1743
1744	/*
1745	 * get a local va for mappings for this pmap.
1746	 */
1747	ptq = get_ptbase(pmap) + i386_btop(va);
1748	if (*ptq) {
1749		(void) pmap_remove_pte(pmap, ptq, va);
1750		invltlb_1pg(va);
1751	}
1752	return;
1753}
1754
1755/*
1756 *	Remove the given range of addresses from the specified map.
1757 *
1758 *	It is assumed that the start and end are properly
1759 *	rounded to the page size.
1760 */
1761void
1762pmap_remove(pmap, sva, eva)
1763	struct pmap *pmap;
1764	register vm_offset_t sva;
1765	register vm_offset_t eva;
1766{
1767	register unsigned *ptbase;
1768	vm_offset_t pdnxt;
1769	vm_offset_t ptpaddr;
1770	vm_offset_t sindex, eindex;
1771	int anyvalid;
1772
1773	if (pmap == NULL)
1774		return;
1775
1776	if (pmap->pm_stats.resident_count == 0)
1777		return;
1778
1779	/*
1780	 * special handling of removing one page.  a very
1781	 * common operation and easy to short circuit some
1782	 * code.
1783	 */
1784	if (((sva + PAGE_SIZE) == eva) &&
1785		(((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1786		pmap_remove_page(pmap, sva);
1787		return;
1788	}
1789
1790	anyvalid = 0;
1791
1792	/*
1793	 * Get a local virtual address for the mappings that are being
1794	 * worked with.
1795	 */
1796	ptbase = get_ptbase(pmap);
1797
1798	sindex = i386_btop(sva);
1799	eindex = i386_btop(eva);
1800
1801	for (; sindex < eindex; sindex = pdnxt) {
1802		unsigned pdirindex;
1803
1804		/*
1805		 * Calculate index for next page table.
1806		 */
1807		pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1808		if (pmap->pm_stats.resident_count == 0)
1809			break;
1810
1811		pdirindex = sindex / NPDEPG;
1812		if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1813			pmap->pm_pdir[pdirindex] = 0;
1814			pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1815			anyvalid++;
1816			continue;
1817		}
1818
1819		/*
1820		 * Weed out invalid mappings. Note: we assume that the page
1821		 * directory table is always allocated, and in kernel virtual.
1822		 */
1823		if (ptpaddr == 0)
1824			continue;
1825
1826		/*
1827		 * Limit our scan to either the end of the va represented
1828		 * by the current page table page, or to the end of the
1829		 * range being removed.
1830		 */
1831		if (pdnxt > eindex) {
1832			pdnxt = eindex;
1833		}
1834
1835		for ( ;sindex != pdnxt; sindex++) {
1836			vm_offset_t va;
1837			if (ptbase[sindex] == 0) {
1838				continue;
1839			}
1840			va = i386_ptob(sindex);
1841
1842			anyvalid++;
1843			if (pmap_remove_pte(pmap,
1844				ptbase + sindex, va))
1845				break;
1846		}
1847	}
1848
1849	if (anyvalid) {
1850		invltlb();
1851	}
1852}
1853
1854/*
1855 *	Routine:	pmap_remove_all
1856 *	Function:
1857 *		Removes this physical page from
1858 *		all physical maps in which it resides.
1859 *		Reflects back modify bits to the pager.
1860 *
1861 *	Notes:
1862 *		Original versions of this routine were very
1863 *		inefficient because they iteratively called
1864 *		pmap_remove (slow...)
1865 */
1866
1867static void
1868pmap_remove_all(pa)
1869	vm_offset_t pa;
1870{
1871	register pv_entry_t pv;
1872	pv_table_t *ppv;
1873	register unsigned *pte, tpte;
1874	int nmodify;
1875	int update_needed;
1876	int s;
1877
1878	nmodify = 0;
1879	update_needed = 0;
1880#if defined(PMAP_DIAGNOSTIC)
1881	/*
1882	 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1883	 * pages!
1884	 */
1885	if (!pmap_is_managed(pa)) {
1886		panic("pmap_page_protect: illegal for unmanaged page, va: 0x%lx", pa);
1887	}
1888#endif
1889
1890	s = splvm();
1891	ppv = pa_to_pvh(pa);
1892	while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) {
1893		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1894
1895		pv->pv_pmap->pm_stats.resident_count--;
1896
1897		tpte = *pte;
1898		*pte = 0;
1899		if (tpte & PG_W)
1900			pv->pv_pmap->pm_stats.wired_count--;
1901		/*
1902		 * Update the vm_page_t clean and reference bits.
1903		 */
1904		if (tpte & PG_M) {
1905#if defined(PMAP_DIAGNOSTIC)
1906			if (pmap_nw_modified((pt_entry_t) tpte)) {
1907				printf("pmap_remove_all: modified page not writable: va: 0x%lx, pte: 0x%lx\n", pv->pv_va, tpte);
1908			}
1909#endif
1910			if (pmap_track_modified(pv->pv_va))
1911				ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL;
1912		}
1913		if (!update_needed &&
1914			((!curproc || (&curproc->p_vmspace->vm_pmap == pv->pv_pmap)) ||
1915			(pv->pv_pmap == kernel_pmap))) {
1916			update_needed = 1;
1917		}
1918
1919		TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1920		TAILQ_REMOVE(&ppv->pv_list, pv, pv_list);
1921		--ppv->pv_list_count;
1922		pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1923		free_pv_entry(pv);
1924	}
1925	ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE);
1926
1927
1928	if (update_needed)
1929		invltlb();
1930	splx(s);
1931	return;
1932}
1933
1934/*
1935 *	Set the physical protection on the
1936 *	specified range of this map as requested.
1937 */
1938void
1939pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1940{
1941	register unsigned *ptbase;
1942	vm_offset_t pdnxt;
1943	vm_offset_t ptpaddr;
1944	vm_offset_t sindex, eindex;
1945	int anychanged;
1946
1947
1948	if (pmap == NULL)
1949		return;
1950
1951	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1952		pmap_remove(pmap, sva, eva);
1953		return;
1954	}
1955
1956	anychanged = 0;
1957
1958	ptbase = get_ptbase(pmap);
1959
1960	sindex = i386_btop(sva);
1961	eindex = i386_btop(eva);
1962
1963	for (; sindex < eindex; sindex = pdnxt) {
1964
1965		unsigned pdirindex;
1966
1967		pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1968
1969		pdirindex = sindex / NPDEPG;
1970		if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1971			(unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1972			pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1973			anychanged++;
1974			continue;
1975		}
1976
1977		/*
1978		 * Weed out invalid mappings. Note: we assume that the page
1979		 * directory table is always allocated, and in kernel virtual.
1980		 */
1981		if (ptpaddr == 0)
1982			continue;
1983
1984		if (pdnxt > eindex) {
1985			pdnxt = eindex;
1986		}
1987
1988		for (; sindex != pdnxt; sindex++) {
1989
1990			unsigned pbits = ptbase[sindex];
1991
1992			if (prot & VM_PROT_WRITE) {
1993				if ((pbits & (PG_RW|PG_V)) == PG_V) {
1994					if (pbits & PG_MANAGED) {
1995						vm_page_t m = PHYS_TO_VM_PAGE(pbits);
1996						m->flags |= PG_WRITEABLE;
1997						m->object->flags |= OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY;
1998					}
1999					ptbase[sindex] = pbits | PG_RW;
2000					anychanged = 1;
2001				}
2002			} else if (pbits & PG_RW) {
2003				if (pbits & PG_M) {
2004					vm_offset_t sva = i386_ptob(sindex);
2005					if ((pbits & PG_MANAGED) && pmap_track_modified(sva)) {
2006						vm_page_t m = PHYS_TO_VM_PAGE(pbits);
2007						m->dirty = VM_PAGE_BITS_ALL;
2008					}
2009				}
2010				ptbase[sindex] = pbits & ~(PG_M|PG_RW);
2011				anychanged = 1;
2012			}
2013		}
2014	}
2015	if (anychanged)
2016		invltlb();
2017}
2018
2019/*
2020 *	Insert the given physical page (p) at
2021 *	the specified virtual address (v) in the
2022 *	target physical map with the protection requested.
2023 *
2024 *	If specified, the page will be wired down, meaning
2025 *	that the related pte can not be reclaimed.
2026 *
2027 *	NB:  This is the only routine which MAY NOT lazy-evaluate
2028 *	or lose information.  That is, this routine must actually
2029 *	insert this page into the given map NOW.
2030 */
2031void
2032pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot,
2033	   boolean_t wired)
2034{
2035	register unsigned *pte;
2036	vm_offset_t opa;
2037	vm_offset_t origpte, newpte;
2038	vm_page_t mpte;
2039
2040	if (pmap == NULL)
2041		return;
2042
2043	va &= PG_FRAME;
2044#ifdef PMAP_DIAGNOSTIC
2045	if (va > VM_MAX_KERNEL_ADDRESS)
2046		panic("pmap_enter: toobig");
2047	if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
2048		panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
2049#endif
2050
2051	mpte = NULL;
2052	/*
2053	 * In the case that a page table page is not
2054	 * resident, we are creating it here.
2055	 */
2056	if (va < UPT_MIN_ADDRESS) {
2057		mpte = pmap_allocpte(pmap, va);
2058	}
2059#if 0 && defined(PMAP_DIAGNOSTIC)
2060	else {
2061		vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
2062		if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
2063			panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
2064				pmap->pm_pdir[PTDPTDI], origpte, va);
2065		}
2066		if (smp_active) {
2067			pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
2068			if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
2069				if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
2070					printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
2071				printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
2072				panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
2073					pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
2074			}
2075		}
2076	}
2077#endif
2078
2079	pte = pmap_pte(pmap, va);
2080	/*
2081	 * Page Directory table entry not valid, we need a new PT page
2082	 */
2083	if (pte == NULL) {
2084		panic("pmap_enter: invalid page directory, pdir=%p, va=0x%lx\n",
2085			pmap->pm_pdir[PTDPTDI], va);
2086	}
2087
2088	origpte = *(vm_offset_t *)pte;
2089	pa &= PG_FRAME;
2090	opa = origpte & PG_FRAME;
2091	if (origpte & PG_PS)
2092		panic("pmap_enter: attempted pmap_enter on 4MB page");
2093
2094	/*
2095	 * Mapping has not changed, must be protection or wiring change.
2096	 */
2097	if (origpte && (opa == pa)) {
2098		/*
2099		 * Wiring change, just update stats. We don't worry about
2100		 * wiring PT pages as they remain resident as long as there
2101		 * are valid mappings in them. Hence, if a user page is wired,
2102		 * the PT page will be also.
2103		 */
2104		if (wired && ((origpte & PG_W) == 0))
2105			pmap->pm_stats.wired_count++;
2106		else if (!wired && (origpte & PG_W))
2107			pmap->pm_stats.wired_count--;
2108
2109#if defined(PMAP_DIAGNOSTIC)
2110		if (pmap_nw_modified((pt_entry_t) origpte)) {
2111			printf("pmap_enter: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, origpte);
2112		}
2113#endif
2114
2115		/*
2116		 * We might be turning off write access to the page,
2117		 * so we go ahead and sense modify status.
2118		 */
2119		if (origpte & PG_MANAGED) {
2120			vm_page_t m;
2121			if (origpte & PG_M) {
2122				if (pmap_track_modified(va)) {
2123					m = PHYS_TO_VM_PAGE(pa);
2124					m->dirty = VM_PAGE_BITS_ALL;
2125				}
2126			}
2127			pa |= PG_MANAGED;
2128		}
2129
2130		if (mpte)
2131			--mpte->hold_count;
2132
2133		goto validate;
2134	}
2135	/*
2136	 * Mapping has changed, invalidate old range and fall through to
2137	 * handle validating new mapping.
2138	 */
2139	if (opa) {
2140		int err;
2141		err = pmap_remove_pte(pmap, pte, va);
2142		if (err)
2143			panic("pmap_enter: pte vanished, va: 0x%x", va);
2144	}
2145
2146	/*
2147	 * Enter on the PV list if part of our managed memory Note that we
2148	 * raise IPL while manipulating pv_table since pmap_enter can be
2149	 * called at interrupt time.
2150	 */
2151	if (pmap_is_managed(pa)) {
2152		pmap_insert_entry(pmap, va, mpte, pa);
2153		pa |= PG_MANAGED;
2154	}
2155
2156	/*
2157	 * Increment counters
2158	 */
2159	pmap->pm_stats.resident_count++;
2160	if (wired)
2161		pmap->pm_stats.wired_count++;
2162
2163validate:
2164	/*
2165	 * Now validate mapping with desired protection/wiring.
2166	 */
2167	newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2168
2169	if (wired)
2170		newpte |= PG_W;
2171	if (va < UPT_MIN_ADDRESS)
2172		newpte |= PG_U;
2173	if (pmap == kernel_pmap)
2174		newpte |= pgeflag;
2175
2176	/*
2177	 * if the mapping or permission bits are different, we need
2178	 * to update the pte.
2179	 */
2180	if ((origpte & ~(PG_M|PG_A)) != newpte) {
2181		*pte = newpte;
2182		if (origpte)
2183			invltlb_1pg(va);
2184	}
2185}
2186
2187/*
2188 * this code makes some *MAJOR* assumptions:
2189 * 1. Current pmap & pmap exists.
2190 * 2. Not wired.
2191 * 3. Read access.
2192 * 4. No page table pages.
2193 * 5. Tlbflush is deferred to calling procedure.
2194 * 6. Page IS managed.
2195 * but is *MUCH* faster than pmap_enter...
2196 */
2197
2198static vm_page_t
2199pmap_enter_quick(pmap, va, pa, mpte)
2200	register pmap_t pmap;
2201	vm_offset_t va;
2202	register vm_offset_t pa;
2203	vm_page_t mpte;
2204{
2205	register unsigned *pte;
2206
2207	/*
2208	 * In the case that a page table page is not
2209	 * resident, we are creating it here.
2210	 */
2211	if (va < UPT_MIN_ADDRESS) {
2212		unsigned ptepindex;
2213		vm_offset_t ptepa;
2214
2215		/*
2216		 * Calculate pagetable page index
2217		 */
2218		ptepindex = va >> PDRSHIFT;
2219		if (mpte && (mpte->pindex == ptepindex)) {
2220			++mpte->hold_count;
2221		} else {
2222retry:
2223			/*
2224			 * Get the page directory entry
2225			 */
2226			ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2227
2228			/*
2229			 * If the page table page is mapped, we just increment
2230			 * the hold count, and activate it.
2231			 */
2232			if (ptepa) {
2233				if (ptepa & PG_PS)
2234					panic("pmap_enter_quick: unexpected mapping into 4MB page");
2235#if defined(PTPHINT)
2236				if (pmap->pm_ptphint &&
2237					(pmap->pm_ptphint->pindex == ptepindex)) {
2238					mpte = pmap->pm_ptphint;
2239				} else {
2240					mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2241					pmap->pm_ptphint = mpte;
2242				}
2243#else
2244				mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2245#endif
2246				if (mpte == NULL)
2247					goto retry;
2248				++mpte->hold_count;
2249			} else {
2250				mpte = _pmap_allocpte(pmap, ptepindex);
2251			}
2252		}
2253	} else {
2254		mpte = NULL;
2255	}
2256
2257	/*
2258	 * This call to vtopte makes the assumption that we are
2259	 * entering the page into the current pmap.  In order to support
2260	 * quick entry into any pmap, one would likely use pmap_pte_quick.
2261	 * But that isn't as quick as vtopte.
2262	 */
2263	pte = (unsigned *)vtopte(va);
2264	if (*pte) {
2265		if (mpte)
2266			pmap_unwire_pte_hold(pmap, mpte);
2267		return 0;
2268	}
2269
2270	/*
2271	 * Enter on the PV list if part of our managed memory Note that we
2272	 * raise IPL while manipulating pv_table since pmap_enter can be
2273	 * called at interrupt time.
2274	 */
2275	pmap_insert_entry(pmap, va, mpte, pa);
2276
2277	/*
2278	 * Increment counters
2279	 */
2280	pmap->pm_stats.resident_count++;
2281
2282	/*
2283	 * Now validate mapping with RO protection
2284	 */
2285	*pte = pa | PG_V | PG_U | PG_MANAGED;
2286
2287	return mpte;
2288}
2289
2290#define MAX_INIT_PT (96)
2291/*
2292 * pmap_object_init_pt preloads the ptes for a given object
2293 * into the specified pmap.  This eliminates the blast of soft
2294 * faults on process startup and immediately after an mmap.
2295 */
2296void
2297pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2298	pmap_t pmap;
2299	vm_offset_t addr;
2300	vm_object_t object;
2301	vm_pindex_t pindex;
2302	vm_size_t size;
2303	int limit;
2304{
2305	vm_offset_t tmpidx;
2306	int psize;
2307	vm_page_t p, mpte;
2308	int objpgs;
2309
2310	if (!pmap)
2311		return;
2312
2313	/*
2314	 * This code maps large physical mmap regions into the
2315	 * processor address space.  Note that some shortcuts
2316	 * are taken, but the code works.
2317	 */
2318	if (pseflag &&
2319		(object->type == OBJT_DEVICE) &&
2320		((addr & (NBPDR - 1)) == 0) &&
2321		((size & (NBPDR - 1)) == 0) ) {
2322		int i;
2323		int s;
2324		vm_page_t m[1];
2325		unsigned int ptepindex;
2326		int npdes;
2327		vm_offset_t ptepa;
2328
2329		if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2330			return;
2331
2332		s = splhigh();
2333retry:
2334		p = vm_page_lookup(object, pindex);
2335		if (p && (p->flags & PG_BUSY)) {
2336			tsleep(p, PVM, "init4p", 0);
2337			goto retry;
2338		}
2339		splx(s);
2340
2341		if (p == NULL) {
2342			p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2343			if (p == NULL)
2344				return;
2345			m[0] = p;
2346
2347			if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2348				PAGE_WAKEUP(p);
2349				vm_page_free(p);
2350				return;
2351			}
2352
2353			p = vm_page_lookup(object, pindex);
2354			PAGE_WAKEUP(p);
2355		}
2356
2357		ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2358		if (ptepa & (NBPDR - 1)) {
2359			return;
2360		}
2361
2362		p->valid = VM_PAGE_BITS_ALL;
2363
2364		pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2365		npdes = size >> PDRSHIFT;
2366		for(i=0;i<npdes;i++) {
2367			pmap->pm_pdir[ptepindex] =
2368				(pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2369			ptepa += NBPDR;
2370			ptepindex += 1;
2371		}
2372		p->flags |= PG_MAPPED;
2373		invltlb();
2374		return;
2375	}
2376
2377	psize = i386_btop(size);
2378
2379	if ((object->type != OBJT_VNODE) ||
2380		(limit && (psize > MAX_INIT_PT) &&
2381			(object->resident_page_count > MAX_INIT_PT))) {
2382		return;
2383	}
2384
2385	if (psize + pindex > object->size)
2386		psize = object->size - pindex;
2387
2388	mpte = NULL;
2389	/*
2390	 * if we are processing a major portion of the object, then scan the
2391	 * entire thing.
2392	 */
2393	if (psize > (object->size >> 2)) {
2394		objpgs = psize;
2395
2396		for (p = TAILQ_FIRST(&object->memq);
2397		    ((objpgs > 0) && (p != NULL));
2398		    p = TAILQ_NEXT(p, listq)) {
2399
2400			tmpidx = p->pindex;
2401			if (tmpidx < pindex) {
2402				continue;
2403			}
2404			tmpidx -= pindex;
2405			if (tmpidx >= psize) {
2406				continue;
2407			}
2408			if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2409			    (p->busy == 0) &&
2410			    (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2411				if ((p->queue - p->pc) == PQ_CACHE)
2412					vm_page_deactivate(p);
2413				p->flags |= PG_BUSY;
2414				mpte = pmap_enter_quick(pmap,
2415					addr + i386_ptob(tmpidx),
2416					VM_PAGE_TO_PHYS(p), mpte);
2417				p->flags |= PG_MAPPED;
2418				PAGE_WAKEUP(p);
2419			}
2420			objpgs -= 1;
2421		}
2422	} else {
2423		/*
2424		 * else lookup the pages one-by-one.
2425		 */
2426		for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2427			p = vm_page_lookup(object, tmpidx + pindex);
2428			if (p &&
2429			    ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2430			    (p->busy == 0) &&
2431			    (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2432				if ((p->queue - p->pc) == PQ_CACHE)
2433					vm_page_deactivate(p);
2434				p->flags |= PG_BUSY;
2435				mpte = pmap_enter_quick(pmap,
2436					addr + i386_ptob(tmpidx),
2437					VM_PAGE_TO_PHYS(p), mpte);
2438				p->flags |= PG_MAPPED;
2439				PAGE_WAKEUP(p);
2440			}
2441		}
2442	}
2443	return;
2444}
2445
2446/*
2447 * pmap_prefault provides a quick way of clustering
2448 * pagefaults into a processes address space.  It is a "cousin"
2449 * of pmap_object_init_pt, except it runs at page fault time instead
2450 * of mmap time.
2451 */
2452#define PFBAK 2
2453#define PFFOR 2
2454#define PAGEORDER_SIZE (PFBAK+PFFOR)
2455
2456static int pmap_prefault_pageorder[] = {
2457	-PAGE_SIZE, PAGE_SIZE, -2 * PAGE_SIZE, 2 * PAGE_SIZE
2458};
2459
2460void
2461pmap_prefault(pmap, addra, entry, object)
2462	pmap_t pmap;
2463	vm_offset_t addra;
2464	vm_map_entry_t entry;
2465	vm_object_t object;
2466{
2467	int i;
2468	vm_offset_t starta;
2469	vm_offset_t addr;
2470	vm_pindex_t pindex;
2471	vm_page_t m, mpte;
2472
2473	if (entry->object.vm_object != object)
2474		return;
2475
2476	if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap))
2477		return;
2478
2479	starta = addra - PFBAK * PAGE_SIZE;
2480	if (starta < entry->start) {
2481		starta = entry->start;
2482	} else if (starta > addra) {
2483		starta = 0;
2484	}
2485
2486	mpte = NULL;
2487	for (i = 0; i < PAGEORDER_SIZE; i++) {
2488		vm_object_t lobject;
2489		unsigned *pte;
2490
2491		addr = addra + pmap_prefault_pageorder[i];
2492		if (addr < starta || addr >= entry->end)
2493			continue;
2494
2495		if ((*pmap_pde(pmap, addr)) == NULL)
2496			continue;
2497
2498		pte = (unsigned *) vtopte(addr);
2499		if (*pte)
2500			continue;
2501
2502		pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2503		lobject = object;
2504		for (m = vm_page_lookup(lobject, pindex);
2505		    (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2506		    lobject = lobject->backing_object) {
2507			if (lobject->backing_object_offset & PAGE_MASK)
2508				break;
2509			pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2510			m = vm_page_lookup(lobject->backing_object, pindex);
2511		}
2512
2513		/*
2514		 * give-up when a page is not in memory
2515		 */
2516		if (m == NULL)
2517			break;
2518
2519		if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2520		    (m->busy == 0) &&
2521		    (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2522
2523			if ((m->queue - m->pc) == PQ_CACHE) {
2524				vm_page_deactivate(m);
2525			}
2526			m->flags |= PG_BUSY;
2527			mpte = pmap_enter_quick(pmap, addr,
2528				VM_PAGE_TO_PHYS(m), mpte);
2529			m->flags |= PG_MAPPED;
2530			PAGE_WAKEUP(m);
2531		}
2532	}
2533}
2534
2535/*
2536 *	Routine:	pmap_change_wiring
2537 *	Function:	Change the wiring attribute for a map/virtual-address
2538 *			pair.
2539 *	In/out conditions:
2540 *			The mapping must already exist in the pmap.
2541 */
2542void
2543pmap_change_wiring(pmap, va, wired)
2544	register pmap_t pmap;
2545	vm_offset_t va;
2546	boolean_t wired;
2547{
2548	register unsigned *pte;
2549
2550	if (pmap == NULL)
2551		return;
2552
2553	pte = pmap_pte(pmap, va);
2554
2555	if (wired && !pmap_pte_w(pte))
2556		pmap->pm_stats.wired_count++;
2557	else if (!wired && pmap_pte_w(pte))
2558		pmap->pm_stats.wired_count--;
2559
2560	/*
2561	 * Wiring is not a hardware characteristic so there is no need to
2562	 * invalidate TLB.
2563	 */
2564	pmap_pte_set_w(pte, wired);
2565}
2566
2567
2568
2569/*
2570 *	Copy the range specified by src_addr/len
2571 *	from the source map to the range dst_addr/len
2572 *	in the destination map.
2573 *
2574 *	This routine is only advisory and need not do anything.
2575 */
2576
2577void
2578pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2579	pmap_t dst_pmap, src_pmap;
2580	vm_offset_t dst_addr;
2581	vm_size_t len;
2582	vm_offset_t src_addr;
2583{
2584	vm_offset_t addr;
2585	vm_offset_t end_addr = src_addr + len;
2586	vm_offset_t pdnxt;
2587	unsigned src_frame, dst_frame;
2588
2589	if (dst_addr != src_addr)
2590		return;
2591
2592	src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2593	if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2594		return;
2595	}
2596
2597	dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2598	if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2599		APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2600		invltlb();
2601	}
2602
2603	for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2604		unsigned *src_pte, *dst_pte;
2605		vm_page_t dstmpte, srcmpte;
2606		vm_offset_t srcptepaddr;
2607		unsigned ptepindex;
2608
2609		if (addr >= UPT_MIN_ADDRESS)
2610			panic("pmap_copy: invalid to pmap_copy page tables\n");
2611
2612		pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2613		ptepindex = addr >> PDRSHIFT;
2614
2615		srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2616		if (srcptepaddr == 0)
2617			continue;
2618
2619		if (srcptepaddr & PG_PS) {
2620			if (dst_pmap->pm_pdir[ptepindex] == 0) {
2621				dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2622				dst_pmap->pm_stats.resident_count += NBPDR;
2623			}
2624			continue;
2625		}
2626
2627		srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2628		if ((srcmpte == NULL) ||
2629			(srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2630			continue;
2631
2632		if (pdnxt > end_addr)
2633			pdnxt = end_addr;
2634
2635		src_pte = (unsigned *) vtopte(addr);
2636		dst_pte = (unsigned *) avtopte(addr);
2637		while (addr < pdnxt) {
2638			unsigned ptetemp;
2639			ptetemp = *src_pte;
2640			/*
2641			 * we only virtual copy managed pages
2642			 */
2643			if ((ptetemp & PG_MANAGED) != 0) {
2644				/*
2645				 * We have to check after allocpte for the
2646				 * pte still being around...  allocpte can
2647				 * block.
2648				 */
2649				dstmpte = pmap_allocpte(dst_pmap, addr);
2650				if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2651					/*
2652					 * Clear the modified and
2653					 * accessed (referenced) bits
2654					 * during the copy.
2655					 */
2656					*dst_pte = ptetemp & ~(PG_M|PG_A);
2657					dst_pmap->pm_stats.resident_count++;
2658					pmap_insert_entry(dst_pmap, addr,
2659						dstmpte,
2660						(ptetemp & PG_FRAME));
2661	 			} else {
2662					pmap_unwire_pte_hold(dst_pmap, dstmpte);
2663				}
2664				if (dstmpte->hold_count >= srcmpte->hold_count)
2665					break;
2666			}
2667			addr += PAGE_SIZE;
2668			++src_pte;
2669			++dst_pte;
2670		}
2671	}
2672}
2673
2674/*
2675 *	Routine:	pmap_kernel
2676 *	Function:
2677 *		Returns the physical map handle for the kernel.
2678 */
2679pmap_t
2680pmap_kernel()
2681{
2682	return (kernel_pmap);
2683}
2684
2685/*
2686 *	pmap_zero_page zeros the specified (machine independent)
2687 *	page by mapping the page into virtual memory and using
2688 *	bzero to clear its contents, one machine dependent page
2689 *	at a time.
2690 */
2691void
2692pmap_zero_page(phys)
2693	vm_offset_t phys;
2694{
2695#ifdef SMP
2696	if (*(int *) prv_CMAP3)
2697		panic("pmap_zero_page: prv_CMAP3 busy");
2698
2699	*(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME);
2700	invltlb_1pg((vm_offset_t) &prv_CPAGE3);
2701
2702	bzero(&prv_CPAGE3, PAGE_SIZE);
2703
2704	*(int *) prv_CMAP3 = 0;
2705	invltlb_1pg((vm_offset_t) &prv_CPAGE3);
2706#else
2707	if (*(int *) CMAP2)
2708		panic("pmap_zero_page: CMAP busy");
2709
2710	*(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME);
2711	bzero(CADDR2, PAGE_SIZE);
2712	*(int *) CMAP2 = 0;
2713	invltlb_1pg((vm_offset_t) CADDR2);
2714#endif
2715}
2716
2717/*
2718 *	pmap_copy_page copies the specified (machine independent)
2719 *	page by mapping the page into virtual memory and using
2720 *	bcopy to copy the page, one machine dependent page at a
2721 *	time.
2722 */
2723void
2724pmap_copy_page(src, dst)
2725	vm_offset_t src;
2726	vm_offset_t dst;
2727{
2728#ifdef SMP
2729	if (*(int *) prv_CMAP1)
2730		panic("pmap_copy_page: prv_CMAP1 busy");
2731	if (*(int *) prv_CMAP2)
2732		panic("pmap_copy_page: prv_CMAP2 busy");
2733
2734	*(int *) prv_CMAP1 = PG_V | PG_RW | (src & PG_FRAME);
2735	*(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME);
2736
2737	invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2);
2738
2739	bcopy(&prv_CPAGE1, &prv_CPAGE2, PAGE_SIZE);
2740
2741	*(int *) prv_CMAP1 = 0;
2742	*(int *) prv_CMAP2 = 0;
2743	invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2);
2744#else
2745	if (*(int *) CMAP1 || *(int *) CMAP2)
2746		panic("pmap_copy_page: CMAP busy");
2747
2748	*(int *) CMAP1 = PG_V | PG_RW | (src & PG_FRAME);
2749	*(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME);
2750
2751	bcopy(CADDR1, CADDR2, PAGE_SIZE);
2752
2753	*(int *) CMAP1 = 0;
2754	*(int *) CMAP2 = 0;
2755	invltlb_2pg( (vm_offset_t) CADDR1, (vm_offset_t) CADDR2);
2756#endif
2757}
2758
2759
2760/*
2761 *	Routine:	pmap_pageable
2762 *	Function:
2763 *		Make the specified pages (by pmap, offset)
2764 *		pageable (or not) as requested.
2765 *
2766 *		A page which is not pageable may not take
2767 *		a fault; therefore, its page table entry
2768 *		must remain valid for the duration.
2769 *
2770 *		This routine is merely advisory; pmap_enter
2771 *		will specify that these pages are to be wired
2772 *		down (or not) as appropriate.
2773 */
2774void
2775pmap_pageable(pmap, sva, eva, pageable)
2776	pmap_t pmap;
2777	vm_offset_t sva, eva;
2778	boolean_t pageable;
2779{
2780}
2781
2782/*
2783 * this routine returns true if a physical page resides
2784 * in the given pmap.
2785 */
2786boolean_t
2787pmap_page_exists(pmap, pa)
2788	pmap_t pmap;
2789	vm_offset_t pa;
2790{
2791	register pv_entry_t pv;
2792	pv_table_t *ppv;
2793	int s;
2794
2795	if (!pmap_is_managed(pa))
2796		return FALSE;
2797
2798	s = splvm();
2799
2800	ppv = pa_to_pvh(pa);
2801	/*
2802	 * Not found, check current mappings returning immediately if found.
2803	 */
2804	for (pv = TAILQ_FIRST(&ppv->pv_list);
2805		pv;
2806		pv = TAILQ_NEXT(pv, pv_list)) {
2807		if (pv->pv_pmap == pmap) {
2808			splx(s);
2809			return TRUE;
2810		}
2811	}
2812	splx(s);
2813	return (FALSE);
2814}
2815
2816#define PMAP_REMOVE_PAGES_CURPROC_ONLY
2817/*
2818 * Remove all pages from specified address space
2819 * this aids process exit speeds.  Also, this code
2820 * is special cased for current process only, but
2821 * can have the more generic (and slightly slower)
2822 * mode enabled.  This is much faster than pmap_remove
2823 * in the case of running down an entire address space.
2824 */
2825void
2826pmap_remove_pages(pmap, sva, eva)
2827	pmap_t pmap;
2828	vm_offset_t sva, eva;
2829{
2830	unsigned *pte, tpte;
2831	pv_table_t *ppv;
2832	pv_entry_t pv, npv;
2833	int s;
2834
2835#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2836	if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) {
2837		printf("warning: pmap_remove_pages called with non-current pmap\n");
2838		return;
2839	}
2840#endif
2841
2842	s = splvm();
2843	for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2844		pv;
2845		pv = npv) {
2846
2847		if (pv->pv_va >= eva || pv->pv_va < sva) {
2848			npv = TAILQ_NEXT(pv, pv_plist);
2849			continue;
2850		}
2851
2852#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2853		pte = (unsigned *)vtopte(pv->pv_va);
2854#else
2855		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2856#endif
2857		tpte = *pte;
2858
2859/*
2860 * We cannot remove wired pages from a process' mapping at this time
2861 */
2862		if (tpte & PG_W) {
2863			npv = TAILQ_NEXT(pv, pv_plist);
2864			continue;
2865		}
2866		*pte = 0;
2867
2868		ppv = pa_to_pvh(tpte);
2869
2870		pv->pv_pmap->pm_stats.resident_count--;
2871
2872		/*
2873		 * Update the vm_page_t clean and reference bits.
2874		 */
2875		if (tpte & PG_M) {
2876			ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL;
2877		}
2878
2879
2880		npv = TAILQ_NEXT(pv, pv_plist);
2881		TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2882
2883		--ppv->pv_list_count;
2884		TAILQ_REMOVE(&ppv->pv_list, pv, pv_list);
2885		if (TAILQ_FIRST(&ppv->pv_list) == NULL) {
2886			ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE);
2887		}
2888
2889		pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2890		free_pv_entry(pv);
2891	}
2892	splx(s);
2893	invltlb();
2894}
2895
2896/*
2897 * pmap_testbit tests bits in pte's
2898 * note that the testbit/changebit routines are inline,
2899 * and a lot of things compile-time evaluate.
2900 */
2901static boolean_t
2902pmap_testbit(pa, bit)
2903	register vm_offset_t pa;
2904	int bit;
2905{
2906	register pv_entry_t pv;
2907	pv_table_t *ppv;
2908	unsigned *pte;
2909	int s;
2910
2911	if (!pmap_is_managed(pa))
2912		return FALSE;
2913
2914	ppv = pa_to_pvh(pa);
2915	if (TAILQ_FIRST(&ppv->pv_list) == NULL)
2916		return FALSE;
2917
2918	s = splvm();
2919
2920	for (pv = TAILQ_FIRST(&ppv->pv_list);
2921		pv;
2922		pv = TAILQ_NEXT(pv, pv_list)) {
2923
2924		/*
2925		 * if the bit being tested is the modified bit, then
2926		 * mark clean_map and ptes as never
2927		 * modified.
2928		 */
2929		if (bit & (PG_A|PG_M)) {
2930			if (!pmap_track_modified(pv->pv_va))
2931				continue;
2932		}
2933
2934#if defined(PMAP_DIAGNOSTIC)
2935		if (!pv->pv_pmap) {
2936			printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va);
2937			continue;
2938		}
2939#endif
2940		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2941		if (*pte & bit) {
2942			splx(s);
2943			return TRUE;
2944		}
2945	}
2946	splx(s);
2947	return (FALSE);
2948}
2949
2950/*
2951 * this routine is used to modify bits in ptes
2952 */
2953static void
2954pmap_changebit(pa, bit, setem)
2955	vm_offset_t pa;
2956	int bit;
2957	boolean_t setem;
2958{
2959	register pv_entry_t pv;
2960	pv_table_t *ppv;
2961	register unsigned *pte;
2962	int changed;
2963	int s;
2964
2965	if (!pmap_is_managed(pa))
2966		return;
2967
2968	s = splvm();
2969	changed = 0;
2970	ppv = pa_to_pvh(pa);
2971
2972	/*
2973	 * Loop over all current mappings setting/clearing as appropos If
2974	 * setting RO do we need to clear the VAC?
2975	 */
2976	for (pv = TAILQ_FIRST(&ppv->pv_list);
2977		pv;
2978		pv = TAILQ_NEXT(pv, pv_list)) {
2979
2980		/*
2981		 * don't write protect pager mappings
2982		 */
2983		if (!setem && (bit == PG_RW)) {
2984			if (!pmap_track_modified(pv->pv_va))
2985				continue;
2986		}
2987
2988#if defined(PMAP_DIAGNOSTIC)
2989		if (!pv->pv_pmap) {
2990			printf("Null pmap (cb) at va: 0x%lx\n", pv->pv_va);
2991			continue;
2992		}
2993#endif
2994
2995		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2996
2997		if (setem) {
2998			*(int *)pte |= bit;
2999			changed = 1;
3000		} else {
3001			vm_offset_t pbits = *(vm_offset_t *)pte;
3002			if (pbits & bit) {
3003				changed = 1;
3004				if (bit == PG_RW) {
3005					if (pbits & PG_M) {
3006						ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL;
3007					}
3008					*(int *)pte = pbits & ~(PG_M|PG_RW);
3009				} else {
3010					*(int *)pte = pbits & ~bit;
3011				}
3012			}
3013		}
3014	}
3015	splx(s);
3016	if (changed)
3017		invltlb();
3018}
3019
3020/*
3021 *      pmap_page_protect:
3022 *
3023 *      Lower the permission for all mappings to a given page.
3024 */
3025void
3026pmap_page_protect(vm_offset_t phys, vm_prot_t prot)
3027{
3028	if ((prot & VM_PROT_WRITE) == 0) {
3029		if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3030			pmap_changebit(phys, PG_RW, FALSE);
3031		} else {
3032			pmap_remove_all(phys);
3033		}
3034	}
3035}
3036
3037vm_offset_t
3038pmap_phys_address(ppn)
3039	int ppn;
3040{
3041	return (i386_ptob(ppn));
3042}
3043
3044/*
3045 *	pmap_ts_referenced:
3046 *
3047 *	Return the count of reference bits for a page, clearing all of them.
3048 *
3049 */
3050int
3051pmap_ts_referenced(vm_offset_t pa)
3052{
3053	register pv_entry_t pv;
3054	pv_table_t *ppv;
3055	unsigned *pte;
3056	int s;
3057	int rtval = 0;
3058
3059	if (!pmap_is_managed(pa))
3060		return FALSE;
3061
3062	s = splvm();
3063
3064	ppv = pa_to_pvh(pa);
3065
3066	if (TAILQ_FIRST(&ppv->pv_list) == NULL) {
3067		splx(s);
3068		return 0;
3069	}
3070
3071	/*
3072	 * Not found, check current mappings returning immediately if found.
3073	 */
3074	for (pv = TAILQ_FIRST(&ppv->pv_list);
3075		pv;
3076		pv = TAILQ_NEXT(pv, pv_list)) {
3077		/*
3078		 * if the bit being tested is the modified bit, then
3079		 * mark clean_map and ptes as never
3080		 * modified.
3081		 */
3082		if (!pmap_track_modified(pv->pv_va))
3083			continue;
3084
3085		pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3086		if (pte == NULL) {
3087			continue;
3088		}
3089		if (*pte & PG_A) {
3090			rtval++;
3091			*pte &= ~PG_A;
3092		}
3093	}
3094	splx(s);
3095	if (rtval) {
3096		invltlb();
3097	}
3098	return (rtval);
3099}
3100
3101/*
3102 *	pmap_is_modified:
3103 *
3104 *	Return whether or not the specified physical page was modified
3105 *	in any physical maps.
3106 */
3107boolean_t
3108pmap_is_modified(vm_offset_t pa)
3109{
3110	return pmap_testbit((pa), PG_M);
3111}
3112
3113/*
3114 *	Clear the modify bits on the specified physical page.
3115 */
3116void
3117pmap_clear_modify(vm_offset_t pa)
3118{
3119	pmap_changebit((pa), PG_M, FALSE);
3120}
3121
3122/*
3123 *	pmap_clear_reference:
3124 *
3125 *	Clear the reference bit on the specified physical page.
3126 */
3127void
3128pmap_clear_reference(vm_offset_t pa)
3129{
3130	pmap_changebit((pa), PG_A, FALSE);
3131}
3132
3133/*
3134 * Miscellaneous support routines follow
3135 */
3136
3137static void
3138i386_protection_init()
3139{
3140	register int *kp, prot;
3141
3142	kp = protection_codes;
3143	for (prot = 0; prot < 8; prot++) {
3144		switch (prot) {
3145		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3146			/*
3147			 * Read access is also 0. There isn't any execute bit,
3148			 * so just make it readable.
3149			 */
3150		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3151		case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3152		case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3153			*kp++ = 0;
3154			break;
3155		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3156		case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3157		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3158		case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3159			*kp++ = PG_RW;
3160			break;
3161		}
3162	}
3163}
3164
3165/*
3166 * Map a set of physical memory pages into the kernel virtual
3167 * address space. Return a pointer to where it is mapped. This
3168 * routine is intended to be used for mapping device memory,
3169 * NOT real memory.
3170 */
3171void *
3172pmap_mapdev(pa, size)
3173	vm_offset_t pa;
3174	vm_size_t size;
3175{
3176	vm_offset_t va, tmpva;
3177	unsigned *pte;
3178
3179	size = roundup(size, PAGE_SIZE);
3180
3181	va = kmem_alloc_pageable(kernel_map, size);
3182	if (!va)
3183		panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3184
3185	pa = pa & PG_FRAME;
3186	for (tmpva = va; size > 0;) {
3187		pte = (unsigned *)vtopte(tmpva);
3188		*pte = pa | PG_RW | PG_V | pgeflag;
3189		size -= PAGE_SIZE;
3190		tmpva += PAGE_SIZE;
3191		pa += PAGE_SIZE;
3192	}
3193	invltlb();
3194
3195	return ((void *) va);
3196}
3197
3198/*
3199 * perform the pmap work for mincore
3200 */
3201int
3202pmap_mincore(pmap, addr)
3203	pmap_t pmap;
3204	vm_offset_t addr;
3205{
3206
3207	unsigned *ptep, pte;
3208	int val = 0;
3209
3210	ptep = pmap_pte(pmap, addr);
3211	if (ptep == 0) {
3212		return 0;
3213	}
3214
3215	if (pte = *ptep) {
3216		vm_offset_t pa;
3217		val = MINCORE_INCORE;
3218		pa = pte & PG_FRAME;
3219
3220		/*
3221		 * Modified by us
3222		 */
3223		if (pte & PG_M)
3224			val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3225		/*
3226		 * Modified by someone
3227		 */
3228		else if (PHYS_TO_VM_PAGE(pa)->dirty ||
3229			pmap_is_modified(pa))
3230			val |= MINCORE_MODIFIED_OTHER;
3231		/*
3232		 * Referenced by us
3233		 */
3234		if (pte & PG_U)
3235			val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3236
3237		/*
3238		 * Referenced by someone
3239		 */
3240		else if ((PHYS_TO_VM_PAGE(pa)->flags & PG_REFERENCED) ||
3241			pmap_ts_referenced(pa)) {
3242			val |= MINCORE_REFERENCED_OTHER;
3243			PHYS_TO_VM_PAGE(pa)->flags |= PG_REFERENCED;
3244		}
3245	}
3246	return val;
3247}
3248
3249void
3250pmap_activate(struct proc *p)
3251{
3252	load_cr3(p->p_addr->u_pcb.pcb_cr3 =
3253		vtophys(p->p_vmspace->vm_pmap.pm_pdir));
3254}
3255
3256vm_offset_t
3257pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) {
3258
3259	if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3260		return addr;
3261	}
3262
3263	addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3264	return addr;
3265}
3266
3267
3268#if defined(PMAP_DEBUG)
3269pmap_pid_dump(int pid) {
3270	pmap_t pmap;
3271	struct proc *p;
3272	int npte = 0;
3273	int index;
3274	for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) {
3275		if (p->p_pid != pid)
3276			continue;
3277
3278		if (p->p_vmspace) {
3279			int i,j;
3280			index = 0;
3281			pmap = &p->p_vmspace->vm_pmap;
3282			for(i=0;i<1024;i++) {
3283				pd_entry_t *pde;
3284				unsigned *pte;
3285				unsigned base = i << PDRSHIFT;
3286
3287				pde = &pmap->pm_pdir[i];
3288				if (pde && pmap_pde_v(pde)) {
3289					for(j=0;j<1024;j++) {
3290						unsigned va = base + (j << PAGE_SHIFT);
3291						if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3292							if (index) {
3293								index = 0;
3294								printf("\n");
3295							}
3296							return npte;
3297						}
3298						pte = pmap_pte_quick( pmap, va);
3299						if (pte && pmap_pte_v(pte)) {
3300							vm_offset_t pa;
3301							vm_page_t m;
3302							pa = *(int *)pte;
3303							m = PHYS_TO_VM_PAGE((pa & PG_FRAME));
3304							printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3305								va, pa, m->hold_count, m->wire_count, m->flags);
3306							npte++;
3307							index++;
3308							if (index >= 2) {
3309								index = 0;
3310								printf("\n");
3311							} else {
3312								printf(" ");
3313							}
3314						}
3315					}
3316				}
3317			}
3318		}
3319	}
3320	return npte;
3321}
3322#endif
3323
3324#if defined(DEBUG)
3325
3326static void	pads __P((pmap_t pm));
3327static void	pmap_pvdump __P((vm_offset_t pa));
3328
3329/* print address space of pmap*/
3330static void
3331pads(pm)
3332	pmap_t pm;
3333{
3334	unsigned va, i, j;
3335	unsigned *ptep;
3336
3337	if (pm == kernel_pmap)
3338		return;
3339	for (i = 0; i < 1024; i++)
3340		if (pm->pm_pdir[i])
3341			for (j = 0; j < 1024; j++) {
3342				va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3343				if (pm == kernel_pmap && va < KERNBASE)
3344					continue;
3345				if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3346					continue;
3347				ptep = pmap_pte_quick(pm, va);
3348				if (pmap_pte_v(ptep))
3349					printf("%x:%x ", va, *(int *) ptep);
3350			};
3351
3352}
3353
3354static void
3355pmap_pvdump(pa)
3356	vm_offset_t pa;
3357{
3358	pv_table_t *ppv;
3359	register pv_entry_t pv;
3360
3361	printf("pa %x", pa);
3362	ppv = pa_to_pvh(pa);
3363	for (pv = TAILQ_FIRST(&ppv->pv_list);
3364		pv;
3365		pv = TAILQ_NEXT(pv, pv_list)) {
3366#ifdef used_to_be
3367		printf(" -> pmap %x, va %x, flags %x",
3368		    pv->pv_pmap, pv->pv_va, pv->pv_flags);
3369#endif
3370		printf(" -> pmap %x, va %x",
3371		    pv->pv_pmap, pv->pv_va);
3372		pads(pv->pv_pmap);
3373	}
3374	printf(" ");
3375}
3376#endif
3377