mmu_oea.c revision 90643
131492Swollman/*
231492Swollman * Copyright (c) 2001 The NetBSD Foundation, Inc.
31553Srgrimes * All rights reserved.
431492Swollman *
51553Srgrimes * This code is derived from software contributed to The NetBSD Foundation
61553Srgrimes * by Matt Thomas <matt@3am-software.com> of Allegro Networks, Inc.
71553Srgrimes *
831492Swollman * Redistribution and use in source and binary forms, with or without
931492Swollman * modification, are permitted provided that the following conditions
101553Srgrimes * are met:
111553Srgrimes * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 *    must display the following acknowledgement:
18 *        This product includes software developed by the NetBSD
19 *        Foundation, Inc. and its contributors.
20 * 4. Neither the name of The NetBSD Foundation nor the names of its
21 *    contributors may be used to endorse or promote products derived
22 *    from this software without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
25 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
26 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
28 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
31 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
32 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
33 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
34 * POSSIBILITY OF SUCH DAMAGE.
35 */
36/*
37 * Copyright (C) 1995, 1996 Wolfgang Solfrank.
38 * Copyright (C) 1995, 1996 TooLs GmbH.
39 * All rights reserved.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 *    notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 *    notice, this list of conditions and the following disclaimer in the
48 *    documentation and/or other materials provided with the distribution.
49 * 3. All advertising materials mentioning features or use of this software
50 *    must display the following acknowledgement:
51 *	This product includes software developed by TooLs GmbH.
52 * 4. The name of TooLs GmbH may not be used to endorse or promote products
53 *    derived from this software without specific prior written permission.
54 *
55 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
56 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
57 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
58 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
59 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
61 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
62 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
63 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
64 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65 *
66 * $NetBSD: pmap.c,v 1.28 2000/03/26 20:42:36 kleink Exp $
67 */
68/*
69 * Copyright (C) 2001 Benno Rice.
70 * All rights reserved.
71 *
72 * Redistribution and use in source and binary forms, with or without
73 * modification, are permitted provided that the following conditions
74 * are met:
75 * 1. Redistributions of source code must retain the above copyright
76 *    notice, this list of conditions and the following disclaimer.
77 * 2. Redistributions in binary form must reproduce the above copyright
78 *    notice, this list of conditions and the following disclaimer in the
79 *    documentation and/or other materials provided with the distribution.
80 *
81 * THIS SOFTWARE IS PROVIDED BY Benno Rice ``AS IS'' AND ANY EXPRESS OR
82 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
83 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
84 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
85 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
86 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
87 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
88 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
89 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
90 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
91 */
92
93#ifndef lint
94static const char rcsid[] =
95  "$FreeBSD: head/sys/powerpc/aim/mmu_oea.c 90643 2002-02-14 01:39:11Z benno $";
96#endif /* not lint */
97
98/*
99 * Manages physical address maps.
100 *
101 * In addition to hardware address maps, this module is called upon to
102 * provide software-use-only maps which may or may not be stored in the
103 * same form as hardware maps.  These pseudo-maps are used to store
104 * intermediate results from copy operations to and from address spaces.
105 *
106 * Since the information managed by this module is also stored by the
107 * logical address mapping module, this module may throw away valid virtual
108 * to physical mappings at almost any time.  However, invalidations of
109 * mappings must be done as requested.
110 *
111 * In order to cope with hardware architectures which make virtual to
112 * physical map invalidates expensive, this module may delay invalidate
113 * reduced protection operations until such time as they are actually
114 * necessary.  This module is given full information as to which processors
115 * are currently using which maps, and to when physical maps must be made
116 * correct.
117 */
118
119#include <sys/param.h>
120#include <sys/kernel.h>
121#include <sys/ktr.h>
122#include <sys/lock.h>
123#include <sys/msgbuf.h>
124#include <sys/mutex.h>
125#include <sys/proc.h>
126#include <sys/sysctl.h>
127#include <sys/systm.h>
128#include <sys/vmmeter.h>
129
130#include <dev/ofw/openfirm.h>
131
132#include <vm/vm.h>
133#include <vm/vm_param.h>
134#include <vm/vm_kern.h>
135#include <vm/vm_page.h>
136#include <vm/vm_map.h>
137#include <vm/vm_object.h>
138#include <vm/vm_extern.h>
139#include <vm/vm_pageout.h>
140#include <vm/vm_pager.h>
141#include <vm/vm_zone.h>
142
143#include <machine/bat.h>
144#include <machine/frame.h>
145#include <machine/md_var.h>
146#include <machine/psl.h>
147#include <machine/pte.h>
148#include <machine/sr.h>
149
150#define	PMAP_DEBUG
151
152#define TODO	panic("%s: not implemented", __func__);
153
154#define	PMAP_LOCK(pm)
155#define	PMAP_UNLOCK(pm)
156
157#define	TLBIE(va)	__asm __volatile("tlbie %0" :: "r"(va))
158#define	TLBSYNC()	__asm __volatile("tlbsync");
159#define	SYNC()		__asm __volatile("sync");
160#define	EIEIO()		__asm __volatile("eieio");
161
162#define	VSID_MAKE(sr, hash)	((sr) | (((hash) & 0xfffff) << 4))
163#define	VSID_TO_SR(vsid)	((vsid) & 0xf)
164#define	VSID_TO_HASH(vsid)	(((vsid) >> 4) & 0xfffff)
165
166#define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
167#define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
168#define	PVO_WIRED		0x0010		/* PVO entry is wired */
169#define	PVO_MANAGED		0x0020		/* PVO entry is managed */
170#define	PVO_EXECUTABLE		0x0040		/* PVO entry is executable */
171#define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
172#define	PVO_ISEXECUTABLE(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
173#define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
174#define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
175#define	PVO_PTEGIDX_CLR(pvo)	\
176	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
177#define	PVO_PTEGIDX_SET(pvo, i)	\
178	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
179
180#define	PMAP_PVO_CHECK(pvo)
181
182struct mem_region {
183	vm_offset_t	mr_start;
184	vm_offset_t	mr_size;
185};
186
187struct ofw_map {
188	vm_offset_t	om_va;
189	vm_size_t	om_len;
190	vm_offset_t	om_pa;
191	u_int		om_mode;
192};
193
194int	pmap_bootstrapped = 0;
195
196/*
197 * Virtual and physical address of message buffer.
198 */
199struct		msgbuf *msgbufp;
200vm_offset_t	msgbuf_phys;
201
202/*
203 * Physical addresses of first and last available physical page.
204 */
205vm_offset_t avail_start;
206vm_offset_t avail_end;
207
208/*
209 * Map of physical memory regions.
210 */
211vm_offset_t	phys_avail[128];
212u_int		phys_avail_count;
213static struct	mem_region regions[128];
214static struct	ofw_map translations[128];
215static int	translations_size;
216
217/*
218 * First and last available kernel virtual addresses.
219 */
220vm_offset_t virtual_avail;
221vm_offset_t virtual_end;
222vm_offset_t kernel_vm_end;
223
224/*
225 * Kernel pmap.
226 */
227struct pmap kernel_pmap_store;
228extern struct pmap ofw_pmap;
229
230/*
231 * PTEG data.
232 */
233static struct	pteg *pmap_pteg_table;
234u_int		pmap_pteg_count;
235u_int		pmap_pteg_mask;
236
237/*
238 * PVO data.
239 */
240struct	pvo_head *pmap_pvo_table;		/* pvo entries by pteg index */
241struct	pvo_head pmap_pvo_kunmanaged =
242    LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
243struct	pvo_head pmap_pvo_unmanaged =
244    LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
245
246vm_zone_t	pmap_upvo_zone;	/* zone for pvo entries for unmanaged pages */
247vm_zone_t	pmap_mpvo_zone;	/* zone for pvo entries for managed pages */
248struct		vm_zone pmap_upvo_zone_store;
249struct		vm_zone pmap_mpvo_zone_store;
250struct		vm_object pmap_upvo_zone_obj;
251struct		vm_object pmap_mpvo_zone_obj;
252
253#define	PMAP_PVO_SIZE	1024
254struct		pvo_entry pmap_upvo_pool[PMAP_PVO_SIZE];
255
256#define	VSID_NBPW	(sizeof(u_int32_t) * 8)
257static u_int	pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
258
259static boolean_t pmap_initialized = FALSE;
260
261/*
262 * Statistics.
263 */
264u_int	pmap_pte_valid = 0;
265u_int	pmap_pte_overflow = 0;
266u_int	pmap_pte_replacements = 0;
267u_int	pmap_pvo_entries = 0;
268u_int	pmap_pvo_enter_calls = 0;
269u_int	pmap_pvo_remove_calls = 0;
270u_int	pmap_pte_spills = 0;
271SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_valid, CTLFLAG_RD, &pmap_pte_valid,
272    0, "");
273SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_overflow, CTLFLAG_RD,
274    &pmap_pte_overflow, 0, "");
275SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_replacements, CTLFLAG_RD,
276    &pmap_pte_replacements, 0, "");
277SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_entries, CTLFLAG_RD, &pmap_pvo_entries,
278    0, "");
279SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_enter_calls, CTLFLAG_RD,
280    &pmap_pvo_enter_calls, 0, "");
281SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_remove_calls, CTLFLAG_RD,
282    &pmap_pvo_remove_calls, 0, "");
283SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_spills, CTLFLAG_RD,
284    &pmap_pte_spills, 0, "");
285
286struct	pvo_entry *pmap_pvo_zeropage;
287
288vm_offset_t	pmap_rkva_start = VM_MIN_KERNEL_ADDRESS;
289u_int		pmap_rkva_count = 4;
290
291/*
292 * Allocate physical memory for use in pmap_bootstrap.
293 */
294static vm_offset_t	pmap_bootstrap_alloc(vm_size_t, u_int);
295
296/*
297 * PTE calls.
298 */
299static int		pmap_pte_insert(u_int, struct pte *);
300
301/*
302 * PVO calls.
303 */
304static int	pmap_pvo_enter(pmap_t, vm_zone_t, struct pvo_head *,
305		    vm_offset_t, vm_offset_t, u_int, int);
306static void	pmap_pvo_remove(struct pvo_entry *, int);
307static struct	pvo_entry *pmap_pvo_find_va(pmap_t, vm_offset_t, int *);
308static struct	pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
309
310/*
311 * Utility routines.
312 */
313static struct		pvo_entry *pmap_rkva_alloc(void);
314static void		pmap_pa_map(struct pvo_entry *, vm_offset_t,
315			    struct pte *, int *);
316static void		pmap_pa_unmap(struct pvo_entry *, struct pte *, int *);
317static void		pmap_syncicache(vm_offset_t, vm_size_t);
318static boolean_t	pmap_query_bit(vm_page_t, int);
319static boolean_t	pmap_clear_bit(vm_page_t, int);
320static void		tlbia(void);
321
322static __inline int
323va_to_sr(u_int *sr, vm_offset_t va)
324{
325	return (sr[(uintptr_t)va >> ADDR_SR_SHFT]);
326}
327
328static __inline u_int
329va_to_pteg(u_int sr, vm_offset_t addr)
330{
331	u_int hash;
332
333	hash = (sr & SR_VSID_MASK) ^ (((u_int)addr & ADDR_PIDX) >>
334	    ADDR_PIDX_SHFT);
335	return (hash & pmap_pteg_mask);
336}
337
338static __inline struct pvo_head *
339pa_to_pvoh(vm_offset_t pa)
340{
341	struct	vm_page *pg;
342
343	pg = PHYS_TO_VM_PAGE(pa);
344
345	if (pg == NULL)
346		return (&pmap_pvo_unmanaged);
347
348	return (&pg->md.mdpg_pvoh);
349}
350
351static __inline struct pvo_head *
352vm_page_to_pvoh(vm_page_t m)
353{
354
355	return (&m->md.mdpg_pvoh);
356}
357
358static __inline void
359pmap_attr_clear(vm_page_t m, int ptebit)
360{
361
362	m->md.mdpg_attrs &= ~ptebit;
363}
364
365static __inline int
366pmap_attr_fetch(vm_page_t m)
367{
368
369	return (m->md.mdpg_attrs);
370}
371
372static __inline void
373pmap_attr_save(vm_page_t m, int ptebit)
374{
375
376	m->md.mdpg_attrs |= ptebit;
377}
378
379static __inline int
380pmap_pte_compare(const struct pte *pt, const struct pte *pvo_pt)
381{
382	if (pt->pte_hi == pvo_pt->pte_hi)
383		return (1);
384
385	return (0);
386}
387
388static __inline int
389pmap_pte_match(struct pte *pt, u_int sr, vm_offset_t va, int which)
390{
391	return (pt->pte_hi & ~PTE_VALID) ==
392	    (((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
393	    ((va >> ADDR_API_SHFT) & PTE_API) | which);
394}
395
396static __inline void
397pmap_pte_create(struct pte *pt, u_int sr, vm_offset_t va, u_int pte_lo)
398{
399	/*
400	 * Construct a PTE.  Default to IMB initially.  Valid bit only gets
401	 * set when the real pte is set in memory.
402	 *
403	 * Note: Don't set the valid bit for correct operation of tlb update.
404	 */
405	pt->pte_hi = ((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
406	    (((va & ADDR_PIDX) >> ADDR_API_SHFT) & PTE_API);
407	pt->pte_lo = pte_lo;
408}
409
410static __inline void
411pmap_pte_synch(struct pte *pt, struct pte *pvo_pt)
412{
413
414	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF | PTE_CHG);
415}
416
417static __inline void
418pmap_pte_clear(struct pte *pt, vm_offset_t va, int ptebit)
419{
420
421	/*
422	 * As shown in Section 7.6.3.2.3
423	 */
424	pt->pte_lo &= ~ptebit;
425	TLBIE(va);
426	EIEIO();
427	TLBSYNC();
428	SYNC();
429}
430
431static __inline void
432pmap_pte_set(struct pte *pt, struct pte *pvo_pt)
433{
434
435	pvo_pt->pte_hi |= PTE_VALID;
436
437	/*
438	 * Update the PTE as defined in section 7.6.3.1.
439	 * Note that the REF/CHG bits are from pvo_pt and thus should havce
440	 * been saved so this routine can restore them (if desired).
441	 */
442	pt->pte_lo = pvo_pt->pte_lo;
443	EIEIO();
444	pt->pte_hi = pvo_pt->pte_hi;
445	SYNC();
446	pmap_pte_valid++;
447}
448
449static __inline void
450pmap_pte_unset(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
451{
452
453	pvo_pt->pte_hi &= ~PTE_VALID;
454
455	/*
456	 * Force the reg & chg bits back into the PTEs.
457	 */
458	SYNC();
459
460	/*
461	 * Invalidate the pte.
462	 */
463	pt->pte_hi &= ~PTE_VALID;
464
465	SYNC();
466	TLBIE(va);
467	EIEIO();
468	TLBSYNC();
469	SYNC();
470
471	/*
472	 * Save the reg & chg bits.
473	 */
474	pmap_pte_synch(pt, pvo_pt);
475	pmap_pte_valid--;
476}
477
478static __inline void
479pmap_pte_change(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
480{
481
482	/*
483	 * Invalidate the PTE
484	 */
485	pmap_pte_unset(pt, pvo_pt, va);
486	pmap_pte_set(pt, pvo_pt);
487}
488
489/*
490 * Quick sort callout for comparing memory regions.
491 */
492static int	mr_cmp(const void *a, const void *b);
493static int	om_cmp(const void *a, const void *b);
494
495static int
496mr_cmp(const void *a, const void *b)
497{
498	const struct	mem_region *regiona;
499	const struct	mem_region *regionb;
500
501	regiona = a;
502	regionb = b;
503	if (regiona->mr_start < regionb->mr_start)
504		return (-1);
505	else if (regiona->mr_start > regionb->mr_start)
506		return (1);
507	else
508		return (0);
509}
510
511static int
512om_cmp(const void *a, const void *b)
513{
514	const struct	ofw_map *mapa;
515	const struct	ofw_map *mapb;
516
517	mapa = a;
518	mapb = b;
519	if (mapa->om_pa < mapb->om_pa)
520		return (-1);
521	else if (mapa->om_pa > mapb->om_pa)
522		return (1);
523	else
524		return (0);
525}
526
527void
528pmap_bootstrap(vm_offset_t kernelstart, vm_offset_t kernelend)
529{
530	ihandle_t	pmem, mmui;
531	phandle_t	chosen, mmu;
532	int		sz;
533	int		i, j;
534	vm_size_t	size;
535	vm_offset_t	pa, va, off;
536	u_int		batl, batu;
537
538	/*
539	 * Use an IBAT and a DBAT to map the bottom segment of memory
540	 * where we are.
541	 */
542	batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
543	batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
544	__asm ("mtibatu 0,%0; mtibatl 0,%1; mtdbatu 0,%0; mtdbatl 0,%1"
545	    :: "r"(batu), "r"(batl));
546#if 0
547	batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
548	batl = BATL(0x80000000, BAT_M, BAT_PP_RW);
549	__asm ("mtibatu 1,%0; mtibatl 1,%1; mtdbatu 1,%0; mtdbatl 1,%1"
550	    :: "r"(batu), "r"(batl));
551#endif
552
553	/*
554	 * Set the start and end of kva.
555	 */
556	virtual_avail = VM_MIN_KERNEL_ADDRESS;
557	virtual_end = VM_MAX_KERNEL_ADDRESS;
558
559	if ((pmem = OF_finddevice("/memory")) == -1)
560		panic("pmap_bootstrap: can't locate memory device");
561	if ((sz = OF_getproplen(pmem, "available")) == -1)
562		panic("pmap_bootstrap: can't get length of available memory");
563	if (sizeof(phys_avail) < sz)
564		panic("pmap_bootstrap: phys_avail too small");
565	if (sizeof(regions) < sz)
566		panic("pmap_bootstrap: regions too small");
567	bzero(regions, sz);
568	if (OF_getprop(pmem, "available", regions, sz) == -1)
569		panic("pmap_bootstrap: can't get available memory");
570	sz /= sizeof(*regions);
571	CTR0(KTR_PMAP, "pmap_bootstrap: physical memory");
572	qsort(regions, sz, sizeof(*regions), mr_cmp);
573	phys_avail_count = 0;
574	for (i = 0, j = 0; i < sz; i++, j += 2) {
575		CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start,
576		    regions[i].mr_start + regions[i].mr_size,
577		    regions[i].mr_size);
578		phys_avail[j] = regions[i].mr_start;
579		phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size;
580		phys_avail_count++;
581	}
582
583	/*
584	 * Allocate PTEG table.
585	 */
586#ifdef PTEGCOUNT
587	pmap_pteg_count = PTEGCOUNT;
588#else
589	pmap_pteg_count = 0x1000;
590
591	while (pmap_pteg_count < physmem)
592		pmap_pteg_count <<= 1;
593
594	pmap_pteg_count >>= 1;
595#endif /* PTEGCOUNT */
596
597	size = pmap_pteg_count * sizeof(struct pteg);
598	CTR2(KTR_PMAP, "pmap_bootstrap: %d PTEGs, %d bytes", pmap_pteg_count,
599	    size);
600	pmap_pteg_table = (struct pteg *)pmap_bootstrap_alloc(size, size);
601	CTR1(KTR_PMAP, "pmap_bootstrap: PTEG table at %p", pmap_pteg_table);
602	bzero((void *)pmap_pteg_table, pmap_pteg_count * sizeof(struct pteg));
603	pmap_pteg_mask = pmap_pteg_count - 1;
604
605	/*
606	 * Allocate PTE overflow lists.
607	 */
608	size = sizeof(struct pvo_head) * pmap_pteg_count;
609	pmap_pvo_table = (struct pvo_head *)pmap_bootstrap_alloc(size,
610	    PAGE_SIZE);
611	CTR1(KTR_PMAP, "pmap_bootstrap: PVO table at %p", pmap_pvo_table);
612	for (i = 0; i < pmap_pteg_count; i++)
613		LIST_INIT(&pmap_pvo_table[i]);
614
615	/*
616	 * Allocate the message buffer.
617	 */
618	msgbuf_phys = pmap_bootstrap_alloc(MSGBUF_SIZE, 0);
619
620	/*
621	 * Initialise the unmanaged pvo pool.
622	 */
623	pmap_upvo_zone = &pmap_upvo_zone_store;
624	zbootinit(pmap_upvo_zone, "unmanaged pvo", sizeof (struct pvo_entry),
625	    pmap_upvo_pool, PMAP_PVO_SIZE);
626
627	/*
628	 * Make sure kernel vsid is allocated as well as VSID 0.
629	 */
630	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW]
631		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
632	pmap_vsid_bitmap[0] |= 1;
633
634	/*
635	 * Set up the OpenFirmware pmap and add it's mappings.
636	 */
637	pmap_pinit(&ofw_pmap);
638	ofw_pmap.pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
639	if ((chosen = OF_finddevice("/chosen")) == -1)
640		panic("pmap_bootstrap: can't find /chosen");
641	OF_getprop(chosen, "mmu", &mmui, 4);
642	if ((mmu = OF_instance_to_package(mmui)) == -1)
643		panic("pmap_bootstrap: can't get mmu package");
644	if ((sz = OF_getproplen(mmu, "translations")) == -1)
645		panic("pmap_bootstrap: can't get ofw translation count");
646	if (sizeof(translations) < sz)
647		panic("pmap_bootstrap: translations too small");
648	bzero(translations, sz);
649	if (OF_getprop(mmu, "translations", translations, sz) == -1)
650		panic("pmap_bootstrap: can't get ofw translations");
651	CTR0(KTR_PMAP, "pmap_bootstrap: translations");
652	qsort(translations, sz, sizeof (*translations), om_cmp);
653	for (i = 0; i < sz; i++) {
654		CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x",
655		    translations[i].om_pa, translations[i].om_va,
656		    translations[i].om_len);
657
658		/* Drop stuff below something? */
659
660		/* Enter the pages? */
661		for (off = 0; off < translations[i].om_len; off += PAGE_SIZE) {
662			struct	vm_page m;
663
664			m.phys_addr = translations[i].om_pa + off;
665			pmap_enter(&ofw_pmap, translations[i].om_va + off, &m,
666			    VM_PROT_ALL, 1);
667		}
668	}
669#ifdef SMP
670	TLBSYNC();
671#endif
672
673	/*
674	 * Initialize the kernel pmap (which is statically allocated).
675	 */
676	for (i = 0; i < 16; i++) {
677		kernel_pmap->pm_sr[i] = EMPTY_SEGMENT;
678	}
679	kernel_pmap->pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
680	kernel_pmap->pm_active = ~0;
681	kernel_pmap->pm_count = 1;
682
683	/*
684	 * Allocate a kernel stack with a guard page for thread0 and map it
685	 * into the kernel page map.
686	 */
687	pa = pmap_bootstrap_alloc(KSTACK_PAGES * PAGE_SIZE, 0);
688	kstack0_phys = pa;
689	kstack0 = virtual_avail + (KSTACK_GUARD_PAGES * PAGE_SIZE);
690	CTR2(KTR_PMAP, "pmap_bootstrap: kstack0 at %#x (%#x)", kstack0_phys,
691	    kstack0);
692	virtual_avail += (KSTACK_PAGES + KSTACK_GUARD_PAGES) * PAGE_SIZE;
693	for (i = 0; i < KSTACK_PAGES; i++) {
694		pa = kstack0_phys + i * PAGE_SIZE;
695		va = kstack0 + i * PAGE_SIZE;
696		pmap_kenter(va, pa);
697		TLBIE(va);
698	}
699
700	/*
701	 * Calculate the first and last available physical addresses.
702	 */
703	avail_start = phys_avail[0];
704	for (i = 0; phys_avail[i + 2] != 0; i += 2)
705		;
706	avail_end = phys_avail[i + 1];
707	Maxmem = powerpc_btop(avail_end);
708
709	/*
710	 * Allocate virtual address space for the message buffer.
711	 */
712	msgbufp = (struct msgbuf *)virtual_avail;
713	virtual_avail += round_page(MSGBUF_SIZE);
714
715	/*
716	 * Initialize hardware.
717	 */
718	for (i = 0; i < 16; i++) {
719		__asm __volatile("mtsrin %0,%1"
720		    :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT));
721	}
722	__asm __volatile ("mtsr %0,%1"
723	    :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
724	__asm __volatile ("sync; mtsdr1 %0; isync"
725	    :: "r"((u_int)pmap_pteg_table | (pmap_pteg_mask >> 10)));
726	tlbia();
727
728	pmap_bootstrapped++;
729}
730
731/*
732 * Activate a user pmap.  The pmap must be activated before it's address
733 * space can be accessed in any way.
734 */
735void
736pmap_activate(struct thread *td)
737{
738	pmap_t	pm;
739	int	i;
740
741	/*
742	 * Load all the data we need up front to encourasge the compiler to
743	 * not issue any loads while we have interrupts disabled below.
744	 */
745	pm = &td->td_proc->p_vmspace->vm_pmap;
746
747	KASSERT(pm->pm_active == 0, ("pmap_activate: pmap already active?"));
748
749	pm->pm_active |= PCPU_GET(cpumask);
750
751	/*
752	 * XXX: Address this again later?
753	 */
754	critical_enter();
755
756	for (i = 0; i < 16; i++) {
757		__asm __volatile("mtsr %0,%1" :: "r"(i), "r"(pm->pm_sr[i]));
758	}
759	__asm __volatile("sync; isync");
760
761	critical_exit();
762}
763
764vm_offset_t
765pmap_addr_hint(vm_object_t object, vm_offset_t va, vm_size_t size)
766{
767	TODO;
768	return (0);
769}
770
771void
772pmap_change_wiring(pmap_t pmap, vm_offset_t va, boolean_t wired)
773{
774	TODO;
775}
776
777void
778pmap_clear_modify(vm_page_t m)
779{
780
781	if (m->flags * PG_FICTITIOUS)
782		return;
783	pmap_clear_bit(m, PTE_CHG);
784}
785
786void
787pmap_collect(void)
788{
789	TODO;
790}
791
792void
793pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
794	  vm_size_t len, vm_offset_t src_addr)
795{
796	TODO;
797}
798
799void
800pmap_copy_page(vm_offset_t src, vm_offset_t dst)
801{
802	TODO;
803}
804
805/*
806 * Zero a page of physical memory by temporarily mapping it into the tlb.
807 */
808void
809pmap_zero_page(vm_offset_t pa)
810{
811	caddr_t	va;
812	int	i;
813
814	if (pa < SEGMENT_LENGTH) {
815		va = (caddr_t) pa;
816	} else if (pmap_initialized) {
817		if (pmap_pvo_zeropage == NULL)
818			pmap_pvo_zeropage = pmap_rkva_alloc();
819		pmap_pa_map(pmap_pvo_zeropage, pa, NULL, NULL);
820		va = (caddr_t)PVO_VADDR(pmap_pvo_zeropage);
821	} else {
822		panic("pmap_zero_page: can't zero pa %#x", pa);
823	}
824
825	bzero(va, PAGE_SIZE);
826
827	for (i = PAGE_SIZE / CACHELINESIZE; i > 0; i--) {
828		__asm __volatile("dcbz 0,%0" :: "r"(va));
829		va += CACHELINESIZE;
830	}
831
832	if (pa >= SEGMENT_LENGTH)
833		pmap_pa_unmap(pmap_pvo_zeropage, NULL, NULL);
834}
835
836void
837pmap_zero_page_area(vm_offset_t pa, int off, int size)
838{
839	TODO;
840}
841
842/*
843 * Map the given physical page at the specified virtual address in the
844 * target pmap with the protection requested.  If specified the page
845 * will be wired down.
846 */
847void
848pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
849	   boolean_t wired)
850{
851	struct		pvo_head *pvo_head;
852	vm_zone_t	zone;
853	u_int		pte_lo, pvo_flags;
854	int		error;
855
856	if (!pmap_initialized) {
857		pvo_head = &pmap_pvo_kunmanaged;
858		zone = pmap_upvo_zone;
859		pvo_flags = 0;
860	} else {
861		pvo_head = pa_to_pvoh(m->phys_addr);
862		zone = pmap_mpvo_zone;
863		pvo_flags = PVO_MANAGED;
864	}
865
866	pte_lo = PTE_I | PTE_G;
867
868	if (prot & VM_PROT_WRITE)
869		pte_lo |= PTE_BW;
870	else
871		pte_lo |= PTE_BR;
872
873	if (prot & VM_PROT_EXECUTE)
874		pvo_flags |= PVO_EXECUTABLE;
875
876	if (wired)
877		pvo_flags |= PVO_WIRED;
878
879	critical_enter();
880
881	error = pmap_pvo_enter(pmap, zone, pvo_head, va, m->phys_addr, pte_lo,
882	    pvo_flags);
883
884	critical_exit();
885
886	if (error == ENOENT) {
887		/*
888		 * Flush the real memory from the cache.
889		 */
890		if ((pvo_flags & PVO_EXECUTABLE) && (pte_lo & PTE_I) == 0) {
891			pmap_syncicache(m->phys_addr, PAGE_SIZE);
892		}
893	}
894}
895
896vm_offset_t
897pmap_extract(pmap_t pmap, vm_offset_t va)
898{
899	TODO;
900	return (0);
901}
902
903/*
904 * Grow the number of kernel page table entries.  Unneeded.
905 */
906void
907pmap_growkernel(vm_offset_t addr)
908{
909}
910
911void
912pmap_init(vm_offset_t phys_start, vm_offset_t phys_end)
913{
914
915	CTR(KTR_PMAP, "pmap_init");
916}
917
918void
919pmap_init2(void)
920{
921
922	CTR(KTR_PMAP, "pmap_init2");
923	zinitna(pmap_upvo_zone, &pmap_upvo_zone_obj, NULL, 0, PMAP_PVO_SIZE,
924	    ZONE_INTERRUPT, 1);
925	pmap_mpvo_zone = zinit("managed pvo", sizeof(struct pvo_entry),
926	    PMAP_PVO_SIZE, ZONE_INTERRUPT, 1);
927	pmap_initialized = TRUE;
928}
929
930boolean_t
931pmap_is_modified(vm_page_t m)
932{
933	TODO;
934	return (0);
935}
936
937void
938pmap_clear_reference(vm_page_t m)
939{
940	TODO;
941}
942
943int
944pmap_ts_referenced(vm_page_t m)
945{
946	TODO;
947	return (0);
948}
949
950/*
951 * Map a wired page into kernel virtual address space.
952 */
953void
954pmap_kenter(vm_offset_t va, vm_offset_t pa)
955{
956	u_int		pte_lo;
957	int		error;
958	int		i;
959
960#if 0
961	if (va < VM_MIN_KERNEL_ADDRESS)
962		panic("pmap_kenter: attempt to enter non-kernel address %#x",
963		    va);
964#endif
965
966	pte_lo = PTE_I | PTE_G | PTE_BW;
967	for (i = 0; phys_avail[i + 2] != 0; i += 2) {
968		if (pa >= phys_avail[i] && pa < phys_avail[i + 1]) {
969			pte_lo &= ~(PTE_I | PTE_G);
970			break;
971		}
972	}
973
974	critical_enter();
975
976	error = pmap_pvo_enter(kernel_pmap, pmap_upvo_zone,
977	    &pmap_pvo_kunmanaged, va, pa, pte_lo, PVO_WIRED);
978
979	critical_exit();
980
981	if (error != 0 && error != ENOENT)
982		panic("pmap_kenter: failed to enter va %#x pa %#x: %d", va,
983		    pa, error);
984
985	/*
986	 * Flush the real memory from the instruction cache.
987	 */
988	if ((pte_lo & (PTE_I | PTE_G)) == 0) {
989		pmap_syncicache(pa, PAGE_SIZE);
990	}
991}
992
993vm_offset_t
994pmap_kextract(vm_offset_t va)
995{
996	TODO;
997	return (0);
998}
999
1000void
1001pmap_kremove(vm_offset_t va)
1002{
1003	TODO;
1004}
1005
1006/*
1007 * Map a range of physical addresses into kernel virtual address space.
1008 *
1009 * The value passed in *virt is a suggested virtual address for the mapping.
1010 * Architectures which can support a direct-mapped physical to virtual region
1011 * can return the appropriate address within that region, leaving '*virt'
1012 * unchanged.  We cannot and therefore do not; *virt is updated with the
1013 * first usable address after the mapped region.
1014 */
1015vm_offset_t
1016pmap_map(vm_offset_t *virt, vm_offset_t pa_start, vm_offset_t pa_end, int prot)
1017{
1018	vm_offset_t	sva, va;
1019
1020	sva = *virt;
1021	va = sva;
1022	for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE)
1023		pmap_kenter(va, pa_start);
1024	*virt = va;
1025	return (sva);
1026}
1027
1028int
1029pmap_mincore(pmap_t pmap, vm_offset_t addr)
1030{
1031	TODO;
1032	return (0);
1033}
1034
1035/*
1036 * Create the uarea for a new process.
1037 * This routine directly affects the fork perf for a process.
1038 */
1039void
1040pmap_new_proc(struct proc *p)
1041{
1042	vm_object_t	upobj;
1043	vm_offset_t	up;
1044	vm_page_t	m;
1045	u_int		i;
1046
1047	/*
1048	 * Allocate the object for the upages.
1049	 */
1050	upobj = p->p_upages_obj;
1051	if (upobj == NULL) {
1052		upobj = vm_object_allocate(OBJT_DEFAULT, UAREA_PAGES);
1053		p->p_upages_obj = upobj;
1054	}
1055
1056	/*
1057	 * Get a kernel virtual address for the uarea for this process.
1058	 */
1059	up = (vm_offset_t)p->p_uarea;
1060	if (up == 0) {
1061		up = kmem_alloc_nofault(kernel_map, UAREA_PAGES * PAGE_SIZE);
1062		if (up == 0)
1063			panic("pmap_new_proc: upage allocation failed");
1064		p->p_uarea = (struct user *)up;
1065	}
1066
1067	for (i = 0; i < UAREA_PAGES; i++) {
1068		/*
1069		 * Get a uarea page.
1070		 */
1071		m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1072
1073		/*
1074		 * Wire the page.
1075		 */
1076		m->wire_count++;
1077
1078		/*
1079		 * Enter the page into the kernel address space.
1080		 */
1081		pmap_kenter(up + i * PAGE_SIZE, VM_PAGE_TO_PHYS(m));
1082
1083		vm_page_wakeup(m);
1084		vm_page_flag_clear(m, PG_ZERO);
1085		vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
1086		m->valid = VM_PAGE_BITS_ALL;
1087	}
1088}
1089
1090void
1091pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_object_t object,
1092		    vm_pindex_t pindex, vm_size_t size, int limit)
1093{
1094	TODO;
1095}
1096
1097/*
1098 * Lower the permission for all mappings to a given page.
1099 */
1100void
1101pmap_page_protect(vm_page_t m, vm_prot_t prot)
1102{
1103	struct	pvo_head *pvo_head;
1104	struct	pvo_entry *pvo, *next_pvo;
1105	struct	pte *pt;
1106
1107	/*
1108	 * Since the routine only downgrades protection, if the
1109	 * maximal protection is desired, there isn't any change
1110	 * to be made.
1111	 */
1112	if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) ==
1113	    (VM_PROT_READ|VM_PROT_WRITE))
1114		return;
1115
1116	critical_enter();
1117
1118	pvo_head = vm_page_to_pvoh(m);
1119	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1120		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1121		PMAP_PVO_CHECK(pvo);	/* sanity check */
1122
1123		/*
1124		 * Downgrading to no mapping at all, we just remove the entry.
1125		 */
1126		if ((prot & VM_PROT_READ) == 0) {
1127			pmap_pvo_remove(pvo, -1);
1128			continue;
1129		}
1130
1131		/*
1132		 * If EXEC permission is being revoked, just clear the flag
1133		 * in the PVO.
1134		 */
1135		if ((prot & VM_PROT_EXECUTE) == 0)
1136			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1137
1138		/*
1139		 * If this entry is already RO, don't diddle with the page
1140		 * table.
1141		 */
1142		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
1143			PMAP_PVO_CHECK(pvo);
1144			continue;
1145		}
1146
1147		/*
1148		 * Grab the PTE before we diddle the bits so pvo_to_pte can
1149		 * verify the pte contents are as expected.
1150		 */
1151		pt = pmap_pvo_to_pte(pvo, -1);
1152		pvo->pvo_pte.pte_lo &= ~PTE_PP;
1153		pvo->pvo_pte.pte_lo |= PTE_BR;
1154		if (pt != NULL)
1155			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1156		PMAP_PVO_CHECK(pvo);	/* sanity check */
1157	}
1158
1159	critical_exit();
1160}
1161
1162/*
1163 * Make the specified page pageable (or not).  Unneeded.
1164 */
1165void
1166pmap_pageable(pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1167	      boolean_t pageable)
1168{
1169}
1170
1171boolean_t
1172pmap_page_exists(pmap_t pmap, vm_page_t m)
1173{
1174	TODO;
1175	return (0);
1176}
1177
1178static u_int	pmap_vsidcontext;
1179
1180void
1181pmap_pinit(pmap_t pmap)
1182{
1183	int	i, mask;
1184	u_int	entropy;
1185
1186	entropy = 0;
1187	__asm __volatile("mftb %0" : "=r"(entropy));
1188
1189	/*
1190	 * Allocate some segment registers for this pmap.
1191	 */
1192	pmap->pm_count = 1;
1193	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1194		u_int	hash, n;
1195
1196		/*
1197		 * Create a new value by mutiplying by a prime and adding in
1198		 * entropy from the timebase register.  This is to make the
1199		 * VSID more random so that the PT hash function collides
1200		 * less often.  (Note that the prime casues gcc to do shifts
1201		 * instead of a multiply.)
1202		 */
1203		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1204		hash = pmap_vsidcontext & (NPMAPS - 1);
1205		if (hash == 0)		/* 0 is special, avoid it */
1206			continue;
1207		n = hash >> 5;
1208		mask = 1 << (hash & (VSID_NBPW - 1));
1209		hash = (pmap_vsidcontext & 0xfffff);
1210		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1211			/* anything free in this bucket? */
1212			if (pmap_vsid_bitmap[n] == 0xffffffff) {
1213				entropy = (pmap_vsidcontext >> 20);
1214				continue;
1215			}
1216			i = ffs(~pmap_vsid_bitmap[i]) - 1;
1217			mask = 1 << i;
1218			hash &= 0xfffff & ~(VSID_NBPW - 1);
1219			hash |= i;
1220		}
1221		pmap_vsid_bitmap[n] |= mask;
1222		for (i = 0; i < 16; i++)
1223			pmap->pm_sr[i] = VSID_MAKE(i, hash);
1224		return;
1225	}
1226
1227	panic("pmap_pinit: out of segments");
1228}
1229
1230/*
1231 * Initialize the pmap associated with process 0.
1232 */
1233void
1234pmap_pinit0(pmap_t pm)
1235{
1236
1237	pmap_pinit(pm);
1238	bzero(&pm->pm_stats, sizeof(pm->pm_stats));
1239}
1240
1241void
1242pmap_pinit2(pmap_t pmap)
1243{
1244	/* XXX: Remove this stub when no longer called */
1245}
1246
1247void
1248pmap_prefault(pmap_t pmap, vm_offset_t va, vm_map_entry_t entry)
1249{
1250	TODO;
1251}
1252
1253void
1254pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1255{
1256	TODO;
1257}
1258
1259vm_offset_t
1260pmap_phys_address(int ppn)
1261{
1262	TODO;
1263	return (0);
1264}
1265
1266void
1267pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
1268{
1269	int	i;
1270
1271	for (i = 0; i < count; i++, va += PAGE_SIZE)
1272		pmap_kenter(va, VM_PAGE_TO_PHYS(m[i]));
1273}
1274
1275void
1276pmap_qremove(vm_offset_t va, int count)
1277{
1278	TODO;
1279}
1280
1281/*
1282 * Add a reference to the specified pmap.
1283 */
1284void
1285pmap_reference(pmap_t pm)
1286{
1287
1288	if (pm != NULL)
1289		pm->pm_count++;
1290}
1291
1292void
1293pmap_release(pmap_t pmap)
1294{
1295	TODO;
1296}
1297
1298void
1299pmap_remove(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
1300{
1301	TODO;
1302}
1303
1304void
1305pmap_remove_pages(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
1306{
1307	TODO;
1308}
1309
1310void
1311pmap_swapin_proc(struct proc *p)
1312{
1313	TODO;
1314}
1315
1316void
1317pmap_swapout_proc(struct proc *p)
1318{
1319	TODO;
1320}
1321
1322/*
1323 * Create the kernel stack and pcb for a new thread.
1324 * This routine directly affects the fork perf for a process and
1325 * create performance for a thread.
1326 */
1327void
1328pmap_new_thread(struct thread *td)
1329{
1330	vm_object_t	ksobj;
1331	vm_offset_t	ks;
1332	vm_page_t	m;
1333	u_int		i;
1334
1335	/*
1336	 * Allocate object for the kstack.
1337	 */
1338	ksobj = td->td_kstack_obj;
1339	if (ksobj == NULL) {
1340		ksobj = vm_object_allocate(OBJT_DEFAULT, KSTACK_PAGES);
1341		td->td_kstack_obj = ksobj;
1342	}
1343
1344	/*
1345	 * Get a kernel virtual address for the kstack for this thread.
1346	 */
1347	ks = td->td_kstack;
1348	if (ks == 0) {
1349		ks = kmem_alloc_nofault(kernel_map,
1350		    (KSTACK_PAGES + KSTACK_GUARD_PAGES) * PAGE_SIZE);
1351		if (ks == 0)
1352			panic("pmap_new_thread: kstack allocation failed");
1353		TLBIE(ks);
1354		ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
1355		td->td_kstack = ks;
1356	}
1357
1358	for (i = 0; i < KSTACK_PAGES; i++) {
1359		/*
1360		 * Get a kernel stack page.
1361		 */
1362		m = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1363
1364		/*
1365		 * Wire the page.
1366		 */
1367		m->wire_count++;
1368
1369		/*
1370		 * Enter the page into the kernel address space.
1371		 */
1372		pmap_kenter(ks + i * PAGE_SIZE, VM_PAGE_TO_PHYS(m));
1373
1374		vm_page_wakeup(m);
1375		vm_page_flag_clear(m, PG_ZERO);
1376		vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
1377		m->valid = VM_PAGE_BITS_ALL;
1378	}
1379}
1380
1381void
1382pmap_dispose_proc(struct proc *p)
1383{
1384	TODO;
1385}
1386
1387void
1388pmap_dispose_thread(struct thread *td)
1389{
1390	TODO;
1391}
1392
1393void
1394pmap_swapin_thread(struct thread *td)
1395{
1396	TODO;
1397}
1398
1399void
1400pmap_swapout_thread(struct thread *td)
1401{
1402	TODO;
1403}
1404
1405/*
1406 * Allocate a physical page of memory directly from the phys_avail map.
1407 * Can only be called from pmap_bootstrap before avail start and end are
1408 * calculated.
1409 */
1410static vm_offset_t
1411pmap_bootstrap_alloc(vm_size_t size, u_int align)
1412{
1413	vm_offset_t	s, e;
1414	int		i, j;
1415
1416	size = round_page(size);
1417	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1418		if (align != 0)
1419			s = (phys_avail[i] + align - 1) & ~(align - 1);
1420		else
1421			s = phys_avail[i];
1422		e = s + size;
1423
1424		if (s < phys_avail[i] || e > phys_avail[i + 1])
1425			continue;
1426
1427		if (s == phys_avail[i]) {
1428			phys_avail[i] += size;
1429		} else if (e == phys_avail[i + 1]) {
1430			phys_avail[i + 1] -= size;
1431		} else {
1432			for (j = phys_avail_count * 2; j > i; j -= 2) {
1433				phys_avail[j] = phys_avail[j - 2];
1434				phys_avail[j + 1] = phys_avail[j - 1];
1435			}
1436
1437			phys_avail[i + 3] = phys_avail[i + 1];
1438			phys_avail[i + 1] = s;
1439			phys_avail[i + 2] = e;
1440			phys_avail_count++;
1441		}
1442
1443		return (s);
1444	}
1445	panic("pmap_bootstrap_alloc: could not allocate memory");
1446}
1447
1448/*
1449 * Return an unmapped pvo for a kernel virtual address.
1450 * Used by pmap functions that operate on physical pages.
1451 */
1452static struct pvo_entry *
1453pmap_rkva_alloc(void)
1454{
1455	struct		pvo_entry *pvo;
1456	struct		pte *pt;
1457	vm_offset_t	kva;
1458	int		pteidx;
1459
1460	if (pmap_rkva_count == 0)
1461		panic("pmap_rkva_alloc: no more reserved KVAs");
1462
1463	kva = pmap_rkva_start + (PAGE_SIZE * --pmap_rkva_count);
1464	pmap_kenter(kva, 0);
1465
1466	pvo = pmap_pvo_find_va(kernel_pmap, kva, &pteidx);
1467
1468	if (pvo == NULL)
1469		panic("pmap_kva_alloc: pmap_pvo_find_va failed");
1470
1471	pt = pmap_pvo_to_pte(pvo, pteidx);
1472
1473	if (pt == NULL)
1474		panic("pmap_kva_alloc: pmap_pvo_to_pte failed");
1475
1476	pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1477	PVO_PTEGIDX_CLR(pvo);
1478
1479	pmap_pte_overflow++;
1480
1481	return (pvo);
1482}
1483
1484static void
1485pmap_pa_map(struct pvo_entry *pvo, vm_offset_t pa, struct pte *saved_pt,
1486    int *depth_p)
1487{
1488	struct	pte *pt;
1489
1490	critical_enter();
1491
1492	/*
1493	 * If this pvo already has a valid pte, we need to save it so it can
1494	 * be restored later.  We then just reload the new PTE over the old
1495	 * slot.
1496	 */
1497	if (saved_pt != NULL) {
1498		pt = pmap_pvo_to_pte(pvo, -1);
1499
1500		if (pt != NULL) {
1501			pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1502			PVO_PTEGIDX_CLR(pvo);
1503			pmap_pte_overflow++;
1504		}
1505
1506		*saved_pt = pvo->pvo_pte;
1507
1508		pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1509	}
1510
1511	pvo->pvo_pte.pte_lo |= pa;
1512
1513	if (!pmap_pte_spill(pvo->pvo_vaddr))
1514		panic("pmap_pa_map: could not spill pvo %p", pvo);
1515
1516	if (depth_p != NULL)
1517		(*depth_p)++;
1518
1519	critical_exit();
1520}
1521
1522static void
1523pmap_pa_unmap(struct pvo_entry *pvo, struct pte *saved_pt, int *depth_p)
1524{
1525	struct	pte *pt;
1526
1527	critical_enter();
1528
1529	pt = pmap_pvo_to_pte(pvo, -1);
1530
1531	if (pt != NULL) {
1532		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1533		PVO_PTEGIDX_CLR(pvo);
1534		pmap_pte_overflow++;
1535	}
1536
1537	pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1538
1539	/*
1540	 * If there is a saved PTE and it's valid, restore it and return.
1541	 */
1542	if (saved_pt != NULL && (saved_pt->pte_lo & PTE_RPGN) != 0) {
1543		if (depth_p != NULL && --(*depth_p) == 0)
1544			panic("pmap_pa_unmap: restoring but depth == 0");
1545
1546		pvo->pvo_pte = *saved_pt;
1547
1548		if (!pmap_pte_spill(pvo->pvo_vaddr))
1549			panic("pmap_pa_unmap: could not spill pvo %p", pvo);
1550	}
1551
1552	critical_exit();
1553}
1554
1555static void
1556pmap_syncicache(vm_offset_t pa, vm_size_t len)
1557{
1558	__syncicache((void *)pa, len);
1559}
1560
1561static void
1562tlbia(void)
1563{
1564	caddr_t	i;
1565
1566	SYNC();
1567	for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) {
1568		TLBIE(i);
1569		EIEIO();
1570	}
1571	TLBSYNC();
1572	SYNC();
1573}
1574
1575static int
1576pmap_pvo_enter(pmap_t pm, vm_zone_t zone, struct pvo_head *pvo_head,
1577    vm_offset_t va, vm_offset_t pa, u_int pte_lo, int flags)
1578{
1579	struct	pvo_entry *pvo;
1580	u_int	sr;
1581	int	first;
1582	u_int	ptegidx;
1583	int	i;
1584
1585	pmap_pvo_enter_calls++;
1586
1587	/*
1588	 * Compute the PTE Group index.
1589	 */
1590	va &= ~ADDR_POFF;
1591	sr = va_to_sr(pm->pm_sr, va);
1592	ptegidx = va_to_pteg(sr, va);
1593
1594	critical_enter();
1595
1596	/*
1597	 * Remove any existing mapping for this page.  Reuse the pvo entry if
1598	 * there is a mapping.
1599	 */
1600	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1601		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1602			pmap_pvo_remove(pvo, -1);
1603			break;
1604		}
1605	}
1606
1607	/*
1608	 * If we aren't overwriting a mapping, try to allocate.
1609	 */
1610	critical_exit();
1611
1612	pvo = zalloc(zone);
1613
1614	critical_enter();
1615
1616	if (pvo == NULL) {
1617		critical_exit();
1618		return (ENOMEM);
1619	}
1620
1621	pmap_pvo_entries++;
1622	pvo->pvo_vaddr = va;
1623	pvo->pvo_pmap = pm;
1624	LIST_INSERT_HEAD(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1625	pvo->pvo_vaddr &= ~ADDR_POFF;
1626	if (flags & VM_PROT_EXECUTE)
1627		pvo->pvo_vaddr |= PVO_EXECUTABLE;
1628	if (flags & PVO_WIRED)
1629		pvo->pvo_vaddr |= PVO_WIRED;
1630	if (pvo_head != &pmap_pvo_kunmanaged)
1631		pvo->pvo_vaddr |= PVO_MANAGED;
1632	pmap_pte_create(&pvo->pvo_pte, sr, va, pa | pte_lo);
1633
1634	/*
1635	 * Remember if the list was empty and therefore will be the first
1636	 * item.
1637	 */
1638	first = LIST_FIRST(pvo_head) == NULL;
1639
1640	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1641	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1642		pvo->pvo_pmap->pm_stats.wired_count++;
1643	pvo->pvo_pmap->pm_stats.resident_count++;
1644
1645	/*
1646	 * We hope this succeeds but it isn't required.
1647	 */
1648	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1649	if (i >= 0) {
1650		PVO_PTEGIDX_SET(pvo, i);
1651	} else {
1652		panic("pmap_pvo_enter: overflow");
1653		pmap_pte_overflow++;
1654	}
1655
1656	critical_exit();
1657
1658	return (first ? ENOENT : 0);
1659}
1660
1661static void
1662pmap_pvo_remove(struct pvo_entry *pvo, int pteidx)
1663{
1664	struct	pte *pt;
1665
1666	/*
1667	 * If there is an active pte entry, we need to deactivate it (and
1668	 * save the ref & cfg bits).
1669	 */
1670	pt = pmap_pvo_to_pte(pvo, pteidx);
1671	if (pt != NULL) {
1672		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1673		PVO_PTEGIDX_CLR(pvo);
1674	} else {
1675		pmap_pte_overflow--;
1676	}
1677
1678	/*
1679	 * Update our statistics.
1680	 */
1681	pvo->pvo_pmap->pm_stats.resident_count--;
1682	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1683		pvo->pvo_pmap->pm_stats.wired_count--;
1684
1685	/*
1686	 * Save the REF/CHG bits into their cache if the page is managed.
1687	 */
1688	if (pvo->pvo_vaddr & PVO_MANAGED) {
1689		struct	vm_page *pg;
1690
1691		pg = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte_lo * PTE_RPGN);
1692		if (pg != NULL) {
1693			pmap_attr_save(pg, pvo->pvo_pte.pte_lo &
1694			    (PTE_REF | PTE_CHG));
1695		}
1696	}
1697
1698	/*
1699	 * Remove this PVO from the PV list.
1700	 */
1701	LIST_REMOVE(pvo, pvo_vlink);
1702
1703	/*
1704	 * Remove this from the overflow list and return it to the pool
1705	 * if we aren't going to reuse it.
1706	 */
1707	LIST_REMOVE(pvo, pvo_olink);
1708	zfree(pvo->pvo_vaddr & PVO_MANAGED ? pmap_mpvo_zone : pmap_upvo_zone,
1709	    pvo);
1710	pmap_pvo_entries--;
1711	pmap_pvo_remove_calls++;
1712}
1713
1714static __inline int
1715pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1716{
1717	int	pteidx;
1718
1719	/*
1720	 * We can find the actual pte entry without searching by grabbing
1721	 * the PTEG index from 3 unused bits in pte_lo[11:9] and by
1722	 * noticing the HID bit.
1723	 */
1724	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1725	if (pvo->pvo_pte.pte_hi & PTE_HID)
1726		pteidx ^= pmap_pteg_mask * 8;
1727
1728	return (pteidx);
1729}
1730
1731static struct pvo_entry *
1732pmap_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p)
1733{
1734	struct	pvo_entry *pvo;
1735	int	ptegidx;
1736	u_int	sr;
1737
1738	va &= ~ADDR_POFF;
1739	sr = va_to_sr(pm->pm_sr, va);
1740	ptegidx = va_to_pteg(sr, va);
1741
1742	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1743		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1744			if (pteidx_p)
1745				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1746			return (pvo);
1747		}
1748	}
1749
1750	return (NULL);
1751}
1752
1753static struct pte *
1754pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1755{
1756	struct	pte *pt;
1757
1758	/*
1759	 * If we haven't been supplied the ptegidx, calculate it.
1760	 */
1761	if (pteidx == -1) {
1762		int	ptegidx;
1763		u_int	sr;
1764
1765		sr = va_to_sr(pvo->pvo_pmap->pm_sr, pvo->pvo_vaddr);
1766		ptegidx = va_to_pteg(sr, pvo->pvo_vaddr);
1767		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1768	}
1769
1770	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1771
1772	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1773		panic("pmap_pvo_to_pte: pvo %p has valid pte in pvo but no "
1774		    "valid pte index", pvo);
1775	}
1776
1777	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1778		panic("pmap_pvo_to_pte: pvo %p has valid pte index in pvo "
1779		    "pvo but no valid pte", pvo);
1780	}
1781
1782	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1783		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1784			panic("pmap_pvo_to_pte: pvo %p has valid pte in "
1785			    "pmap_pteg_table %p but invalid in pvo", pvo, pt);
1786		}
1787
1788		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF))
1789		    != 0) {
1790			panic("pmap_pvo_to_pte: pvo %p pte does not match "
1791			    "pte %p in pmap_pteg_table", pvo, pt);
1792		}
1793
1794		return (pt);
1795	}
1796
1797	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1798		panic("pmap_pvo_to_pte: pvo %p has invalid pte %p in "
1799		    "pmap_pteg_table but valid in pvo", pvo, pt);
1800	}
1801
1802	return (NULL);
1803}
1804
1805/*
1806 * XXX: THIS STUFF SHOULD BE IN pte.c?
1807 */
1808int
1809pmap_pte_spill(vm_offset_t addr)
1810{
1811	struct	pvo_entry *source_pvo, *victim_pvo;
1812	struct	pvo_entry *pvo;
1813	int	ptegidx, i, j;
1814	u_int	sr;
1815	struct	pteg *pteg;
1816	struct	pte *pt;
1817
1818	pmap_pte_spills++;
1819
1820	__asm __volatile("mfsrin %0,%1" : "=r"(sr) : "r"(addr));
1821	ptegidx = va_to_pteg(sr, addr);
1822
1823	/*
1824	 * Have to substitute some entry.  Use the primary hash for this.
1825	 * Use low bits of timebase as random generator.
1826	 */
1827	pteg = &pmap_pteg_table[ptegidx];
1828	__asm __volatile("mftb %0" : "=r"(i));
1829	i &= 7;
1830	pt = &pteg->pt[i];
1831
1832	source_pvo = NULL;
1833	victim_pvo = NULL;
1834	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1835		/*
1836		 * We need to find a pvo entry for this address.
1837		 */
1838		PMAP_PVO_CHECK(pvo);
1839		if (source_pvo == NULL &&
1840		    pmap_pte_match(&pvo->pvo_pte, sr, addr,
1841		    pvo->pvo_pte.pte_hi & PTE_HID)) {
1842			/*
1843			 * Now found an entry to be spilled into the pteg.
1844			 * The PTE is now valid, so we know it's active.
1845			 */
1846			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1847
1848			if (j >= 0) {
1849				PVO_PTEGIDX_SET(pvo, j);
1850				pmap_pte_overflow--;
1851				PMAP_PVO_CHECK(pvo);
1852				return (1);
1853			}
1854
1855			source_pvo = pvo;
1856
1857			if (victim_pvo != NULL)
1858				break;
1859		}
1860
1861		/*
1862		 * We also need the pvo entry of the victim we are replacing
1863		 * so save the R & C bits of the PTE.
1864		 */
1865		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
1866		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
1867			victim_pvo = pvo;
1868			if (source_pvo != NULL)
1869				break;
1870		}
1871	}
1872
1873	if (source_pvo == NULL)
1874		return (0);
1875
1876	if (victim_pvo == NULL) {
1877		if ((pt->pte_hi & PTE_HID) == 0)
1878			panic("pmap_pte_spill: victim p-pte (%p) has no pvo"
1879			    "entry", pt);
1880
1881		/*
1882		 * If this is a secondary PTE, we need to search it's primary
1883		 * pvo bucket for the matching PVO.
1884		 */
1885		LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx ^ pmap_pteg_mask],
1886		    pvo_olink) {
1887			PMAP_PVO_CHECK(pvo);
1888			/*
1889			 * We also need the pvo entry of the victim we are
1890			 * replacing so save the R & C bits of the PTE.
1891			 */
1892			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1893				victim_pvo = pvo;
1894				break;
1895			}
1896		}
1897
1898		if (victim_pvo == NULL)
1899			panic("pmap_pte_spill: victim s-pte (%p) has no pvo"
1900			    "entry", pt);
1901	}
1902
1903	/*
1904	 * We are invalidating the TLB entry for the EA we are replacing even
1905	 * though it's valid.  If we don't, we lose any ref/chg bit changes
1906	 * contained in the TLB entry.
1907	 */
1908	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1909
1910	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1911	pmap_pte_set(pt, &source_pvo->pvo_pte);
1912
1913	PVO_PTEGIDX_CLR(victim_pvo);
1914	PVO_PTEGIDX_SET(source_pvo, i);
1915	pmap_pte_replacements++;
1916
1917	PMAP_PVO_CHECK(victim_pvo);
1918	PMAP_PVO_CHECK(source_pvo);
1919
1920	return (1);
1921}
1922
1923static int
1924pmap_pte_insert(u_int ptegidx, struct pte *pvo_pt)
1925{
1926	struct	pte *pt;
1927	int	i;
1928
1929	/*
1930	 * First try primary hash.
1931	 */
1932	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
1933		if ((pt->pte_hi & PTE_VALID) == 0) {
1934			pvo_pt->pte_hi &= ~PTE_HID;
1935			pmap_pte_set(pt, pvo_pt);
1936			return (i);
1937		}
1938	}
1939
1940	/*
1941	 * Now try secondary hash.
1942	 */
1943	ptegidx ^= pmap_pteg_mask;
1944	ptegidx++;
1945	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
1946		if ((pt->pte_hi & PTE_VALID) == 0) {
1947			pvo_pt->pte_hi |= PTE_HID;
1948			pmap_pte_set(pt, pvo_pt);
1949			return (i);
1950		}
1951	}
1952
1953	panic("pmap_pte_insert: overflow");
1954	return (-1);
1955}
1956
1957static boolean_t
1958pmap_query_bit(vm_page_t m, int ptebit)
1959{
1960	struct	pvo_entry *pvo;
1961	struct	pte *pt;
1962
1963	if (pmap_attr_fetch(m) & ptebit)
1964		return (TRUE);
1965
1966	critical_enter();
1967
1968	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1969		PMAP_PVO_CHECK(pvo);	/* sanity check */
1970
1971		/*
1972		 * See if we saved the bit off.  If so, cache it and return
1973		 * success.
1974		 */
1975		if (pvo->pvo_pte.pte_lo & ptebit) {
1976			pmap_attr_save(m, ptebit);
1977			PMAP_PVO_CHECK(pvo);	/* sanity check */
1978			critical_exit();
1979			return (TRUE);
1980		}
1981	}
1982
1983	/*
1984	 * No luck, now go through the hard part of looking at the PTEs
1985	 * themselves.  Sync so that any pending REF/CHG bits are flushed to
1986	 * the PTEs.
1987	 */
1988	SYNC();
1989	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1990		PMAP_PVO_CHECK(pvo);	/* sanity check */
1991
1992		/*
1993		 * See if this pvo has a valid PTE.  if so, fetch the
1994		 * REF/CHG bits from the valid PTE.  If the appropriate
1995		 * ptebit is set, cache it and return success.
1996		 */
1997		pt = pmap_pvo_to_pte(pvo, -1);
1998		if (pt != NULL) {
1999			pmap_pte_synch(pt, &pvo->pvo_pte);
2000			if (pvo->pvo_pte.pte_lo & ptebit) {
2001				pmap_attr_save(m, ptebit);
2002				PMAP_PVO_CHECK(pvo);	/* sanity check */
2003				critical_exit();
2004				return (TRUE);
2005			}
2006		}
2007	}
2008
2009	critical_exit();
2010	return (TRUE);
2011}
2012
2013static boolean_t
2014pmap_clear_bit(vm_page_t m, int ptebit)
2015{
2016	struct	pvo_entry *pvo;
2017	struct	pte *pt;
2018	int	rv;
2019
2020	critical_enter();
2021
2022	/*
2023	 * Clear the cached value.
2024	 */
2025	rv = pmap_attr_fetch(m);
2026	pmap_attr_clear(m, ptebit);
2027
2028	/*
2029	 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so
2030	 * we can reset the right ones).  note that since the pvo entries and
2031	 * list heads are accessed via BAT0 and are never placed in the page
2032	 * table, we don't have to worry about further accesses setting the
2033	 * REF/CHG bits.
2034	 */
2035	SYNC();
2036
2037	/*
2038	 * For each pvo entry, clear the pvo's ptebit.  If this pvo has a
2039	 * valid pte clear the ptebit from the valid pte.
2040	 */
2041	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2042		PMAP_PVO_CHECK(pvo);	/* sanity check */
2043		pt = pmap_pvo_to_pte(pvo, -1);
2044		if (pt != NULL) {
2045			pmap_pte_synch(pt, &pvo->pvo_pte);
2046			if (pvo->pvo_pte.pte_lo & ptebit)
2047				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2048		}
2049		rv |= pvo->pvo_pte.pte_lo;
2050		pvo->pvo_pte.pte_lo &= ~ptebit;
2051		PMAP_PVO_CHECK(pvo);	/* sanity check */
2052	}
2053
2054	critical_exit();
2055	return ((rv & ptebit) != 0);
2056}
2057