xref: /freebsd/sys/powerpc/aim/mmu_oea.c (revision ae83180158c4c937f170e31eff311b18c0286a93)
1 /*
2  * Copyright (c) 2001 The NetBSD Foundation, Inc.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to The NetBSD Foundation
6  * by Matt Thomas <matt@3am-software.com> of Allegro Networks, Inc.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 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
94 static const char rcsid[] =
95   "$FreeBSD$";
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/uma.h>
142 
143 #include <machine/powerpc.h>
144 #include <machine/bat.h>
145 #include <machine/frame.h>
146 #include <machine/md_var.h>
147 #include <machine/psl.h>
148 #include <machine/pte.h>
149 #include <machine/sr.h>
150 
151 #define	PMAP_DEBUG
152 
153 #define TODO	panic("%s: not implemented", __func__);
154 
155 #define	PMAP_LOCK(pm)
156 #define	PMAP_UNLOCK(pm)
157 
158 #define	TLBIE(va)	__asm __volatile("tlbie %0" :: "r"(va))
159 #define	TLBSYNC()	__asm __volatile("tlbsync");
160 #define	SYNC()		__asm __volatile("sync");
161 #define	EIEIO()		__asm __volatile("eieio");
162 
163 #define	VSID_MAKE(sr, hash)	((sr) | (((hash) & 0xfffff) << 4))
164 #define	VSID_TO_SR(vsid)	((vsid) & 0xf)
165 #define	VSID_TO_HASH(vsid)	(((vsid) >> 4) & 0xfffff)
166 
167 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
168 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
169 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
170 #define	PVO_MANAGED		0x0020		/* PVO entry is managed */
171 #define	PVO_EXECUTABLE		0x0040		/* PVO entry is executable */
172 #define	PVO_BOOTSTRAP		0x0080		/* PVO entry allocated during
173 						   bootstrap */
174 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
175 #define	PVO_ISEXECUTABLE(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
176 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
177 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
178 #define	PVO_PTEGIDX_CLR(pvo)	\
179 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
180 #define	PVO_PTEGIDX_SET(pvo, i)	\
181 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
182 
183 #define	PMAP_PVO_CHECK(pvo)
184 
185 struct ofw_map {
186 	vm_offset_t	om_va;
187 	vm_size_t	om_len;
188 	vm_offset_t	om_pa;
189 	u_int		om_mode;
190 };
191 
192 int	pmap_bootstrapped = 0;
193 
194 /*
195  * Virtual and physical address of message buffer.
196  */
197 struct		msgbuf *msgbufp;
198 vm_offset_t	msgbuf_phys;
199 
200 /*
201  * Physical addresses of first and last available physical page.
202  */
203 vm_offset_t avail_start;
204 vm_offset_t avail_end;
205 
206 /*
207  * Map of physical memory regions.
208  */
209 vm_offset_t	phys_avail[128];
210 u_int		phys_avail_count;
211 static struct	mem_region *regions;
212 static struct	mem_region *pregions;
213 int		regions_sz, pregions_sz;
214 static struct	ofw_map *translations;
215 
216 /*
217  * First and last available kernel virtual addresses.
218  */
219 vm_offset_t virtual_avail;
220 vm_offset_t virtual_end;
221 vm_offset_t kernel_vm_end;
222 
223 /*
224  * Kernel pmap.
225  */
226 struct pmap kernel_pmap_store;
227 extern struct pmap ofw_pmap;
228 
229 /*
230  * PTEG data.
231  */
232 static struct	pteg *pmap_pteg_table;
233 u_int		pmap_pteg_count;
234 u_int		pmap_pteg_mask;
235 
236 /*
237  * PVO data.
238  */
239 struct	pvo_head *pmap_pvo_table;		/* pvo entries by pteg index */
240 struct	pvo_head pmap_pvo_kunmanaged =
241     LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
242 struct	pvo_head pmap_pvo_unmanaged =
243     LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
244 
245 uma_zone_t	pmap_upvo_zone;	/* zone for pvo entries for unmanaged pages */
246 uma_zone_t	pmap_mpvo_zone;	/* zone for pvo entries for managed pages */
247 struct		vm_object pmap_upvo_zone_obj;
248 struct		vm_object pmap_mpvo_zone_obj;
249 static vm_object_t	pmap_pvo_obj;
250 static u_int		pmap_pvo_count;
251 
252 #define	BPVO_POOL_SIZE	32768
253 static struct	pvo_entry *pmap_bpvo_pool;
254 static int	pmap_bpvo_pool_index = 0;
255 
256 #define	VSID_NBPW	(sizeof(u_int32_t) * 8)
257 static u_int	pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
258 
259 static boolean_t pmap_initialized = FALSE;
260 
261 /*
262  * Statistics.
263  */
264 u_int	pmap_pte_valid = 0;
265 u_int	pmap_pte_overflow = 0;
266 u_int	pmap_pte_replacements = 0;
267 u_int	pmap_pvo_entries = 0;
268 u_int	pmap_pvo_enter_calls = 0;
269 u_int	pmap_pvo_remove_calls = 0;
270 u_int	pmap_pte_spills = 0;
271 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_valid, CTLFLAG_RD, &pmap_pte_valid,
272     0, "");
273 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_overflow, CTLFLAG_RD,
274     &pmap_pte_overflow, 0, "");
275 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_replacements, CTLFLAG_RD,
276     &pmap_pte_replacements, 0, "");
277 SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_entries, CTLFLAG_RD, &pmap_pvo_entries,
278     0, "");
279 SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_enter_calls, CTLFLAG_RD,
280     &pmap_pvo_enter_calls, 0, "");
281 SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_remove_calls, CTLFLAG_RD,
282     &pmap_pvo_remove_calls, 0, "");
283 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_spills, CTLFLAG_RD,
284     &pmap_pte_spills, 0, "");
285 
286 struct	pvo_entry *pmap_pvo_zeropage;
287 
288 vm_offset_t	pmap_rkva_start = VM_MIN_KERNEL_ADDRESS;
289 u_int		pmap_rkva_count = 4;
290 
291 /*
292  * Allocate physical memory for use in pmap_bootstrap.
293  */
294 static vm_offset_t	pmap_bootstrap_alloc(vm_size_t, u_int);
295 
296 /*
297  * PTE calls.
298  */
299 static int		pmap_pte_insert(u_int, struct pte *);
300 
301 /*
302  * PVO calls.
303  */
304 static int	pmap_pvo_enter(pmap_t, uma_zone_t, struct pvo_head *,
305 		    vm_offset_t, vm_offset_t, u_int, int);
306 static void	pmap_pvo_remove(struct pvo_entry *, int);
307 static struct	pvo_entry *pmap_pvo_find_va(pmap_t, vm_offset_t, int *);
308 static struct	pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
309 
310 /*
311  * Utility routines.
312  */
313 static void *		pmap_pvo_allocf(uma_zone_t, int, u_int8_t *, int);
314 static struct		pvo_entry *pmap_rkva_alloc(void);
315 static void		pmap_pa_map(struct pvo_entry *, vm_offset_t,
316 			    struct pte *, int *);
317 static void		pmap_pa_unmap(struct pvo_entry *, struct pte *, int *);
318 static void		pmap_syncicache(vm_offset_t, vm_size_t);
319 static boolean_t	pmap_query_bit(vm_page_t, int);
320 static boolean_t	pmap_clear_bit(vm_page_t, int);
321 static void		tlbia(void);
322 
323 static __inline int
324 va_to_sr(u_int *sr, vm_offset_t va)
325 {
326 	return (sr[(uintptr_t)va >> ADDR_SR_SHFT]);
327 }
328 
329 static __inline u_int
330 va_to_pteg(u_int sr, vm_offset_t addr)
331 {
332 	u_int hash;
333 
334 	hash = (sr & SR_VSID_MASK) ^ (((u_int)addr & ADDR_PIDX) >>
335 	    ADDR_PIDX_SHFT);
336 	return (hash & pmap_pteg_mask);
337 }
338 
339 static __inline struct pvo_head *
340 pa_to_pvoh(vm_offset_t pa, vm_page_t *pg_p)
341 {
342 	struct	vm_page *pg;
343 
344 	pg = PHYS_TO_VM_PAGE(pa);
345 
346 	if (pg_p != NULL)
347 		*pg_p = pg;
348 
349 	if (pg == NULL)
350 		return (&pmap_pvo_unmanaged);
351 
352 	return (&pg->md.mdpg_pvoh);
353 }
354 
355 static __inline struct pvo_head *
356 vm_page_to_pvoh(vm_page_t m)
357 {
358 
359 	return (&m->md.mdpg_pvoh);
360 }
361 
362 static __inline void
363 pmap_attr_clear(vm_page_t m, int ptebit)
364 {
365 
366 	m->md.mdpg_attrs &= ~ptebit;
367 }
368 
369 static __inline int
370 pmap_attr_fetch(vm_page_t m)
371 {
372 
373 	return (m->md.mdpg_attrs);
374 }
375 
376 static __inline void
377 pmap_attr_save(vm_page_t m, int ptebit)
378 {
379 
380 	m->md.mdpg_attrs |= ptebit;
381 }
382 
383 static __inline int
384 pmap_pte_compare(const struct pte *pt, const struct pte *pvo_pt)
385 {
386 	if (pt->pte_hi == pvo_pt->pte_hi)
387 		return (1);
388 
389 	return (0);
390 }
391 
392 static __inline int
393 pmap_pte_match(struct pte *pt, u_int sr, vm_offset_t va, int which)
394 {
395 	return (pt->pte_hi & ~PTE_VALID) ==
396 	    (((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
397 	    ((va >> ADDR_API_SHFT) & PTE_API) | which);
398 }
399 
400 static __inline void
401 pmap_pte_create(struct pte *pt, u_int sr, vm_offset_t va, u_int pte_lo)
402 {
403 	/*
404 	 * Construct a PTE.  Default to IMB initially.  Valid bit only gets
405 	 * set when the real pte is set in memory.
406 	 *
407 	 * Note: Don't set the valid bit for correct operation of tlb update.
408 	 */
409 	pt->pte_hi = ((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
410 	    (((va & ADDR_PIDX) >> ADDR_API_SHFT) & PTE_API);
411 	pt->pte_lo = pte_lo;
412 }
413 
414 static __inline void
415 pmap_pte_synch(struct pte *pt, struct pte *pvo_pt)
416 {
417 
418 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF | PTE_CHG);
419 }
420 
421 static __inline void
422 pmap_pte_clear(struct pte *pt, vm_offset_t va, int ptebit)
423 {
424 
425 	/*
426 	 * As shown in Section 7.6.3.2.3
427 	 */
428 	pt->pte_lo &= ~ptebit;
429 	TLBIE(va);
430 	EIEIO();
431 	TLBSYNC();
432 	SYNC();
433 }
434 
435 static __inline void
436 pmap_pte_set(struct pte *pt, struct pte *pvo_pt)
437 {
438 
439 	pvo_pt->pte_hi |= PTE_VALID;
440 
441 	/*
442 	 * Update the PTE as defined in section 7.6.3.1.
443 	 * Note that the REF/CHG bits are from pvo_pt and thus should havce
444 	 * been saved so this routine can restore them (if desired).
445 	 */
446 	pt->pte_lo = pvo_pt->pte_lo;
447 	EIEIO();
448 	pt->pte_hi = pvo_pt->pte_hi;
449 	SYNC();
450 	pmap_pte_valid++;
451 }
452 
453 static __inline void
454 pmap_pte_unset(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
455 {
456 
457 	pvo_pt->pte_hi &= ~PTE_VALID;
458 
459 	/*
460 	 * Force the reg & chg bits back into the PTEs.
461 	 */
462 	SYNC();
463 
464 	/*
465 	 * Invalidate the pte.
466 	 */
467 	pt->pte_hi &= ~PTE_VALID;
468 
469 	SYNC();
470 	TLBIE(va);
471 	EIEIO();
472 	TLBSYNC();
473 	SYNC();
474 
475 	/*
476 	 * Save the reg & chg bits.
477 	 */
478 	pmap_pte_synch(pt, pvo_pt);
479 	pmap_pte_valid--;
480 }
481 
482 static __inline void
483 pmap_pte_change(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
484 {
485 
486 	/*
487 	 * Invalidate the PTE
488 	 */
489 	pmap_pte_unset(pt, pvo_pt, va);
490 	pmap_pte_set(pt, pvo_pt);
491 }
492 
493 /*
494  * Quick sort callout for comparing memory regions.
495  */
496 static int	mr_cmp(const void *a, const void *b);
497 static int	om_cmp(const void *a, const void *b);
498 
499 static int
500 mr_cmp(const void *a, const void *b)
501 {
502 	const struct	mem_region *regiona;
503 	const struct	mem_region *regionb;
504 
505 	regiona = a;
506 	regionb = b;
507 	if (regiona->mr_start < regionb->mr_start)
508 		return (-1);
509 	else if (regiona->mr_start > regionb->mr_start)
510 		return (1);
511 	else
512 		return (0);
513 }
514 
515 static int
516 om_cmp(const void *a, const void *b)
517 {
518 	const struct	ofw_map *mapa;
519 	const struct	ofw_map *mapb;
520 
521 	mapa = a;
522 	mapb = b;
523 	if (mapa->om_pa < mapb->om_pa)
524 		return (-1);
525 	else if (mapa->om_pa > mapb->om_pa)
526 		return (1);
527 	else
528 		return (0);
529 }
530 
531 void
532 pmap_bootstrap(vm_offset_t kernelstart, vm_offset_t kernelend)
533 {
534 	ihandle_t	mmui;
535 	phandle_t	chosen, mmu;
536 	int		sz;
537 	int		i, j;
538 	vm_size_t	size, physsz;
539 	vm_offset_t	pa, va, off;
540 	u_int		batl, batu;
541 
542         /*
543          * Set up BAT0 to only map the lowest 256 MB area
544          */
545         battable[0x0].batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
546         battable[0x0].batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
547 
548         /*
549          * Map PCI memory space.
550          */
551         battable[0x8].batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
552         battable[0x8].batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
553 
554         battable[0x9].batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
555         battable[0x9].batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
556 
557         battable[0xa].batl = BATL(0xa0000000, BAT_I|BAT_G, BAT_PP_RW);
558         battable[0xa].batu = BATU(0xa0000000, BAT_BL_256M, BAT_Vs);
559 
560         battable[0xb].batl = BATL(0xb0000000, BAT_I|BAT_G, BAT_PP_RW);
561         battable[0xb].batu = BATU(0xb0000000, BAT_BL_256M, BAT_Vs);
562 
563         /*
564          * Map obio devices.
565          */
566         battable[0xf].batl = BATL(0xf0000000, BAT_I|BAT_G, BAT_PP_RW);
567         battable[0xf].batu = BATU(0xf0000000, BAT_BL_256M, BAT_Vs);
568 
569 	/*
570 	 * Use an IBAT and a DBAT to map the bottom segment of memory
571 	 * where we are.
572 	 */
573 	batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
574 	batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
575 	__asm ("mtibatu 0,%0; mtibatl 0,%1; mtdbatu 0,%0; mtdbatl 0,%1"
576 	    :: "r"(batu), "r"(batl));
577 
578 #if 0
579 	/* map frame buffer */
580 	batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
581 	batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
582 	__asm ("mtdbatu 1,%0; mtdbatl 1,%1"
583 	    :: "r"(batu), "r"(batl));
584 #endif
585 
586 #if 1
587 	/* map pci space */
588 	batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
589 	batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
590 	__asm ("mtdbatu 1,%0; mtdbatl 1,%1"
591 	    :: "r"(batu), "r"(batl));
592 #endif
593 
594 	/*
595 	 * Set the start and end of kva.
596 	 */
597 	virtual_avail = VM_MIN_KERNEL_ADDRESS;
598 	virtual_end = VM_MAX_KERNEL_ADDRESS;
599 
600 	mem_regions(&pregions, &pregions_sz, &regions, &regions_sz);
601 	CTR0(KTR_PMAP, "pmap_bootstrap: physical memory");
602 
603 	qsort(pregions, pregions_sz, sizeof(*pregions), mr_cmp);
604 	for (i = 0; i < pregions_sz; i++) {
605 		CTR3(KTR_PMAP, "physregion: %#x - %#x (%#x)",
606 			pregions[i].mr_start,
607 			pregions[i].mr_start + pregions[i].mr_size,
608 			pregions[i].mr_size);
609 	}
610 
611 	if (sizeof(phys_avail)/sizeof(phys_avail[0]) < regions_sz)
612 		panic("pmap_bootstrap: phys_avail too small");
613 	qsort(regions, regions_sz, sizeof(*regions), mr_cmp);
614 	phys_avail_count = 0;
615 	physsz = 0;
616 	for (i = 0, j = 0; i < regions_sz; i++, j += 2) {
617 		CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start,
618 		    regions[i].mr_start + regions[i].mr_size,
619 		    regions[i].mr_size);
620 		phys_avail[j] = regions[i].mr_start;
621 		phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size;
622 		phys_avail_count++;
623 		physsz += regions[i].mr_size;
624 	}
625 	physmem = btoc(physsz);
626 
627 	/*
628 	 * Allocate PTEG table.
629 	 */
630 #ifdef PTEGCOUNT
631 	pmap_pteg_count = PTEGCOUNT;
632 #else
633 	pmap_pteg_count = 0x1000;
634 
635 	while (pmap_pteg_count < physmem)
636 		pmap_pteg_count <<= 1;
637 
638 	pmap_pteg_count >>= 1;
639 #endif /* PTEGCOUNT */
640 
641 	size = pmap_pteg_count * sizeof(struct pteg);
642 	CTR2(KTR_PMAP, "pmap_bootstrap: %d PTEGs, %d bytes", pmap_pteg_count,
643 	    size);
644 	pmap_pteg_table = (struct pteg *)pmap_bootstrap_alloc(size, size);
645 	CTR1(KTR_PMAP, "pmap_bootstrap: PTEG table at %p", pmap_pteg_table);
646 	bzero((void *)pmap_pteg_table, pmap_pteg_count * sizeof(struct pteg));
647 	pmap_pteg_mask = pmap_pteg_count - 1;
648 
649 	/*
650 	 * Allocate pv/overflow lists.
651 	 */
652 	size = sizeof(struct pvo_head) * pmap_pteg_count;
653 	pmap_pvo_table = (struct pvo_head *)pmap_bootstrap_alloc(size,
654 	    PAGE_SIZE);
655 	CTR1(KTR_PMAP, "pmap_bootstrap: PVO table at %p", pmap_pvo_table);
656 	for (i = 0; i < pmap_pteg_count; i++)
657 		LIST_INIT(&pmap_pvo_table[i]);
658 
659 	/*
660 	 * Allocate the message buffer.
661 	 */
662 	msgbuf_phys = pmap_bootstrap_alloc(MSGBUF_SIZE, 0);
663 
664 	/*
665 	 * Initialise the unmanaged pvo pool.
666 	 */
667 	pmap_bpvo_pool = (struct pvo_entry *)pmap_bootstrap_alloc(
668 		BPVO_POOL_SIZE*sizeof(struct pvo_entry), 0);
669 	pmap_bpvo_pool_index = 0;
670 
671 	/*
672 	 * Make sure kernel vsid is allocated as well as VSID 0.
673 	 */
674 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW]
675 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
676 	pmap_vsid_bitmap[0] |= 1;
677 
678 	/*
679 	 * Set up the OpenFirmware pmap and add it's mappings.
680 	 */
681 	pmap_pinit(&ofw_pmap);
682 	ofw_pmap.pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
683 	if ((chosen = OF_finddevice("/chosen")) == -1)
684 		panic("pmap_bootstrap: can't find /chosen");
685 	OF_getprop(chosen, "mmu", &mmui, 4);
686 	if ((mmu = OF_instance_to_package(mmui)) == -1)
687 		panic("pmap_bootstrap: can't get mmu package");
688 	if ((sz = OF_getproplen(mmu, "translations")) == -1)
689 		panic("pmap_bootstrap: can't get ofw translation count");
690 	translations = NULL;
691 	for (i = 0; phys_avail[i + 2] != 0; i += 2) {
692 		if (phys_avail[i + 1] >= sz)
693 			translations = (struct ofw_map *)phys_avail[i];
694 	}
695 	if (translations == NULL)
696 		panic("pmap_bootstrap: no space to copy translations");
697 	bzero(translations, sz);
698 	if (OF_getprop(mmu, "translations", translations, sz) == -1)
699 		panic("pmap_bootstrap: can't get ofw translations");
700 	CTR0(KTR_PMAP, "pmap_bootstrap: translations");
701 	sz /= sizeof(*translations);
702 	qsort(translations, sz, sizeof (*translations), om_cmp);
703 	for (i = 0; i < sz; i++) {
704 		CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x",
705 		    translations[i].om_pa, translations[i].om_va,
706 		    translations[i].om_len);
707 
708 		/* Drop stuff below something? */
709 
710 		/* Enter the pages? */
711 		for (off = 0; off < translations[i].om_len; off += PAGE_SIZE) {
712 			struct	vm_page m;
713 
714 			m.phys_addr = translations[i].om_pa + off;
715 			pmap_enter(&ofw_pmap, translations[i].om_va + off, &m,
716 			    VM_PROT_ALL, 1);
717 		}
718 	}
719 #ifdef SMP
720 	TLBSYNC();
721 #endif
722 
723 	/*
724 	 * Initialize the kernel pmap (which is statically allocated).
725 	 */
726 	for (i = 0; i < 16; i++) {
727 		kernel_pmap->pm_sr[i] = EMPTY_SEGMENT;
728 	}
729 	kernel_pmap->pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
730 	kernel_pmap->pm_active = ~0;
731 
732 	/*
733 	 * Allocate a kernel stack with a guard page for thread0 and map it
734 	 * into the kernel page map.
735 	 */
736 	pa = pmap_bootstrap_alloc(KSTACK_PAGES * PAGE_SIZE, 0);
737 	kstack0_phys = pa;
738 	kstack0 = virtual_avail + (KSTACK_GUARD_PAGES * PAGE_SIZE);
739 	CTR2(KTR_PMAP, "pmap_bootstrap: kstack0 at %#x (%#x)", kstack0_phys,
740 	    kstack0);
741 	virtual_avail += (KSTACK_PAGES + KSTACK_GUARD_PAGES) * PAGE_SIZE;
742 	for (i = 0; i < KSTACK_PAGES; i++) {
743 		pa = kstack0_phys + i * PAGE_SIZE;
744 		va = kstack0 + i * PAGE_SIZE;
745 		pmap_kenter(va, pa);
746 		TLBIE(va);
747 	}
748 
749 	/*
750 	 * Calculate the first and last available physical addresses.
751 	 */
752 	avail_start = phys_avail[0];
753 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
754 		;
755 	avail_end = phys_avail[i + 1];
756 	Maxmem = powerpc_btop(avail_end);
757 
758 	/*
759 	 * Allocate virtual address space for the message buffer.
760 	 */
761 	msgbufp = (struct msgbuf *)virtual_avail;
762 	virtual_avail += round_page(MSGBUF_SIZE);
763 
764 	/*
765 	 * Initialize hardware.
766 	 */
767 	for (i = 0; i < 16; i++) {
768 		mtsrin(i << ADDR_SR_SHFT, EMPTY_SEGMENT);
769 	}
770 	__asm __volatile ("mtsr %0,%1"
771 	    :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
772 	__asm __volatile ("sync; mtsdr1 %0; isync"
773 	    :: "r"((u_int)pmap_pteg_table | (pmap_pteg_mask >> 10)));
774 	tlbia();
775 
776 	pmap_bootstrapped++;
777 }
778 
779 /*
780  * Activate a user pmap.  The pmap must be activated before it's address
781  * space can be accessed in any way.
782  */
783 void
784 pmap_activate(struct thread *td)
785 {
786 	pmap_t	pm, pmr;
787 
788 	/*
789 	 * Load all the data we need up front to encourasge the compiler to
790 	 * not issue any loads while we have interrupts disabled below.
791 	 */
792 	pm = &td->td_proc->p_vmspace->vm_pmap;
793 
794 	KASSERT(pm->pm_active == 0, ("pmap_activate: pmap already active?"));
795 
796 	if ((pmr = (pmap_t)pmap_kextract((vm_offset_t)pm)) == NULL)
797 		pmr = pm;
798 
799 	pm->pm_active |= PCPU_GET(cpumask);
800 	PCPU_SET(curpmap, pmr);
801 }
802 
803 void
804 pmap_deactivate(struct thread *td)
805 {
806 	pmap_t	pm;
807 
808 	pm = &td->td_proc->p_vmspace->vm_pmap;
809 	pm->pm_active &= ~(PCPU_GET(cpumask));
810 	PCPU_SET(curpmap, NULL);
811 }
812 
813 vm_offset_t
814 pmap_addr_hint(vm_object_t object, vm_offset_t va, vm_size_t size)
815 {
816 
817 	return (va);
818 }
819 
820 void
821 pmap_change_wiring(pmap_t pm, vm_offset_t va, boolean_t wired)
822 {
823 	struct	pvo_entry *pvo;
824 
825 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
826 
827 	if (pvo != NULL) {
828 		if (wired) {
829 			if ((pvo->pvo_vaddr & PVO_WIRED) == 0)
830 				pm->pm_stats.wired_count++;
831 			pvo->pvo_vaddr |= PVO_WIRED;
832 		} else {
833 			if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
834 				pm->pm_stats.wired_count--;
835 			pvo->pvo_vaddr &= ~PVO_WIRED;
836 		}
837 	}
838 }
839 
840 void
841 pmap_clear_modify(vm_page_t m)
842 {
843 
844 	if (m->flags * PG_FICTITIOUS)
845 		return;
846 	pmap_clear_bit(m, PTE_CHG);
847 }
848 
849 void
850 pmap_collect(void)
851 {
852 	TODO;
853 }
854 
855 void
856 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
857 	  vm_size_t len, vm_offset_t src_addr)
858 {
859 
860 	/*
861 	 * This is not needed as it's mainly an optimisation.
862 	 * It may want to be implemented later though.
863 	 */
864 }
865 
866 void
867 pmap_copy_page(vm_page_t msrc, vm_page_t mdst)
868 {
869 	vm_offset_t	dst;
870 	vm_offset_t	src;
871 
872 	dst = VM_PAGE_TO_PHYS(mdst);
873 	src = VM_PAGE_TO_PHYS(msrc);
874 
875 	kcopy((void *)src, (void *)dst, PAGE_SIZE);
876 }
877 
878 /*
879  * Zero a page of physical memory by temporarily mapping it into the tlb.
880  */
881 void
882 pmap_zero_page(vm_page_t m)
883 {
884 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
885 	caddr_t	va;
886 	int	i;
887 
888 	if (pa < SEGMENT_LENGTH) {
889 		va = (caddr_t) pa;
890 	} else if (pmap_initialized) {
891 		if (pmap_pvo_zeropage == NULL)
892 			pmap_pvo_zeropage = pmap_rkva_alloc();
893 		pmap_pa_map(pmap_pvo_zeropage, pa, NULL, NULL);
894 		va = (caddr_t)PVO_VADDR(pmap_pvo_zeropage);
895 	} else {
896 		panic("pmap_zero_page: can't zero pa %#x", pa);
897 	}
898 
899 	bzero(va, PAGE_SIZE);
900 
901 	for (i = PAGE_SIZE / CACHELINESIZE; i > 0; i--) {
902 		__asm __volatile("dcbz 0,%0" :: "r"(va));
903 		va += CACHELINESIZE;
904 	}
905 
906 	if (pa >= SEGMENT_LENGTH)
907 		pmap_pa_unmap(pmap_pvo_zeropage, NULL, NULL);
908 }
909 
910 void
911 pmap_zero_page_area(vm_page_t m, int off, int size)
912 {
913 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
914 	caddr_t	va;
915 	int	i;
916 
917 	if (pa < SEGMENT_LENGTH) {
918 		va = (caddr_t) pa;
919 	} else if (pmap_initialized) {
920 		if (pmap_pvo_zeropage == NULL)
921 			pmap_pvo_zeropage = pmap_rkva_alloc();
922 		pmap_pa_map(pmap_pvo_zeropage, pa, NULL, NULL);
923 		va = (caddr_t)PVO_VADDR(pmap_pvo_zeropage);
924 	} else {
925 		panic("pmap_zero_page: can't zero pa %#x", pa);
926 	}
927 
928 	bzero(va, size);
929 
930 	for (i = size / CACHELINESIZE; i > 0; i--) {
931 		__asm __volatile("dcbz 0,%0" :: "r"(va));
932 		va += CACHELINESIZE;
933 	}
934 
935 	if (pa >= SEGMENT_LENGTH)
936 		pmap_pa_unmap(pmap_pvo_zeropage, NULL, NULL);
937 }
938 
939 void
940 pmap_zero_page_idle(vm_page_t m)
941 {
942 
943 	/* XXX this is called outside of Giant, is pmap_zero_page safe? */
944 	/* XXX maybe have a dedicated mapping for this to avoid the problem? */
945 	mtx_lock(&Giant);
946 	pmap_zero_page(m);
947 	mtx_unlock(&Giant);
948 }
949 
950 /*
951  * Map the given physical page at the specified virtual address in the
952  * target pmap with the protection requested.  If specified the page
953  * will be wired down.
954  */
955 void
956 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
957 	   boolean_t wired)
958 {
959 	struct		pvo_head *pvo_head;
960 	uma_zone_t	zone;
961 	vm_page_t	pg;
962 	u_int		pte_lo, pvo_flags, was_exec, i;
963 	int		error;
964 
965 	if (!pmap_initialized) {
966 		pvo_head = &pmap_pvo_kunmanaged;
967 		zone = pmap_upvo_zone;
968 		pvo_flags = 0;
969 		pg = NULL;
970 		was_exec = PTE_EXEC;
971 	} else {
972 		pvo_head = pa_to_pvoh(VM_PAGE_TO_PHYS(m), &pg);
973 		zone = pmap_mpvo_zone;
974 		pvo_flags = PVO_MANAGED;
975 		was_exec = 0;
976 	}
977 
978 	/*
979 	 * If this is a managed page, and it's the first reference to the page,
980 	 * clear the execness of the page.  Otherwise fetch the execness.
981 	 */
982 	if (pg != NULL) {
983 		if (LIST_EMPTY(pvo_head)) {
984 			pmap_attr_clear(pg, PTE_EXEC);
985 		} else {
986 			was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
987 		}
988 	}
989 
990 
991 	/*
992 	 * Assume the page is cache inhibited and access is guarded unless
993 	 * it's in our available memory array.
994 	 */
995 	pte_lo = PTE_I | PTE_G;
996 	for (i = 0; i < pregions_sz; i++) {
997 		if ((VM_PAGE_TO_PHYS(m) >= pregions[i].mr_start) &&
998 		    (VM_PAGE_TO_PHYS(m) <
999 			(pregions[i].mr_start + pregions[i].mr_size))) {
1000 			pte_lo &= ~(PTE_I | PTE_G);
1001 			break;
1002 		}
1003 	}
1004 
1005 	if (prot & VM_PROT_WRITE)
1006 		pte_lo |= PTE_BW;
1007 	else
1008 		pte_lo |= PTE_BR;
1009 
1010 	pvo_flags |= (prot & VM_PROT_EXECUTE);
1011 
1012 	if (wired)
1013 		pvo_flags |= PVO_WIRED;
1014 
1015 	error = pmap_pvo_enter(pmap, zone, pvo_head, va, VM_PAGE_TO_PHYS(m),
1016 	    pte_lo, pvo_flags);
1017 
1018 	/*
1019 	 * Flush the real page from the instruction cache if this page is
1020 	 * mapped executable and cacheable and was not previously mapped (or
1021 	 * was not mapped executable).
1022 	 */
1023 	if (error == 0 && (pvo_flags & PVO_EXECUTABLE) &&
1024 	    (pte_lo & PTE_I) == 0 && was_exec == 0) {
1025 		/*
1026 		 * Flush the real memory from the cache.
1027 		 */
1028 		pmap_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1029 		if (pg != NULL)
1030 			pmap_attr_save(pg, PTE_EXEC);
1031 	}
1032 }
1033 
1034 vm_offset_t
1035 pmap_extract(pmap_t pm, vm_offset_t va)
1036 {
1037 	struct	pvo_entry *pvo;
1038 
1039 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1040 
1041 	if (pvo != NULL) {
1042 		return ((pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF));
1043 	}
1044 
1045 	return (0);
1046 }
1047 
1048 /*
1049  * Grow the number of kernel page table entries.  Unneeded.
1050  */
1051 void
1052 pmap_growkernel(vm_offset_t addr)
1053 {
1054 }
1055 
1056 void
1057 pmap_init(vm_offset_t phys_start, vm_offset_t phys_end)
1058 {
1059 
1060 	CTR0(KTR_PMAP, "pmap_init");
1061 
1062 	pmap_pvo_obj = vm_object_allocate(OBJT_PHYS, 16);
1063 	pmap_pvo_count = 0;
1064 	pmap_upvo_zone = uma_zcreate("UPVO entry", sizeof (struct pvo_entry),
1065 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM);
1066 	uma_zone_set_allocf(pmap_upvo_zone, pmap_pvo_allocf);
1067 	pmap_mpvo_zone = uma_zcreate("MPVO entry", sizeof(struct pvo_entry),
1068 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM);
1069 	uma_zone_set_allocf(pmap_mpvo_zone, pmap_pvo_allocf);
1070 	pmap_initialized = TRUE;
1071 }
1072 
1073 void
1074 pmap_init2(void)
1075 {
1076 
1077 	CTR0(KTR_PMAP, "pmap_init2");
1078 }
1079 
1080 boolean_t
1081 pmap_is_modified(vm_page_t m)
1082 {
1083 
1084 	if (m->flags & PG_FICTITIOUS)
1085 		return (FALSE);
1086 
1087 	return (pmap_query_bit(m, PTE_CHG));
1088 }
1089 
1090 void
1091 pmap_clear_reference(vm_page_t m)
1092 {
1093 	TODO;
1094 }
1095 
1096 /*
1097  *	pmap_ts_referenced:
1098  *
1099  *	Return a count of reference bits for a page, clearing those bits.
1100  *	It is not necessary for every reference bit to be cleared, but it
1101  *	is necessary that 0 only be returned when there are truly no
1102  *	reference bits set.
1103  *
1104  *	XXX: The exact number of bits to check and clear is a matter that
1105  *	should be tested and standardized at some point in the future for
1106  *	optimal aging of shared pages.
1107  */
1108 
1109 int
1110 pmap_ts_referenced(vm_page_t m)
1111 {
1112 	TODO;
1113 	return (0);
1114 }
1115 
1116 /*
1117  * Map a wired page into kernel virtual address space.
1118  */
1119 void
1120 pmap_kenter(vm_offset_t va, vm_offset_t pa)
1121 {
1122 	u_int		pte_lo;
1123 	int		error;
1124 	int		i;
1125 
1126 #if 0
1127 	if (va < VM_MIN_KERNEL_ADDRESS)
1128 		panic("pmap_kenter: attempt to enter non-kernel address %#x",
1129 		    va);
1130 #endif
1131 
1132 	pte_lo = PTE_I | PTE_G | PTE_BW;
1133 	for (i = 0; phys_avail[i + 2] != 0; i += 2) {
1134 		if (pa >= phys_avail[i] && pa < phys_avail[i + 1]) {
1135 			pte_lo &= ~(PTE_I | PTE_G);
1136 			break;
1137 		}
1138 	}
1139 
1140 	error = pmap_pvo_enter(kernel_pmap, pmap_upvo_zone,
1141 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, PVO_WIRED);
1142 
1143 	if (error != 0 && error != ENOENT)
1144 		panic("pmap_kenter: failed to enter va %#x pa %#x: %d", va,
1145 		    pa, error);
1146 
1147 	/*
1148 	 * Flush the real memory from the instruction cache.
1149 	 */
1150 	if ((pte_lo & (PTE_I | PTE_G)) == 0) {
1151 		pmap_syncicache(pa, PAGE_SIZE);
1152 	}
1153 }
1154 
1155 /*
1156  * Extract the physical page address associated with the given kernel virtual
1157  * address.
1158  */
1159 vm_offset_t
1160 pmap_kextract(vm_offset_t va)
1161 {
1162 	struct		pvo_entry *pvo;
1163 
1164 	pvo = pmap_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL);
1165 	if (pvo == NULL) {
1166 		return (0);
1167 	}
1168 
1169 	return ((pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF));
1170 }
1171 
1172 /*
1173  * Remove a wired page from kernel virtual address space.
1174  */
1175 void
1176 pmap_kremove(vm_offset_t va)
1177 {
1178 
1179 	pmap_remove(kernel_pmap, va, roundup(va, PAGE_SIZE));
1180 }
1181 
1182 /*
1183  * Map a range of physical addresses into kernel virtual address space.
1184  *
1185  * The value passed in *virt is a suggested virtual address for the mapping.
1186  * Architectures which can support a direct-mapped physical to virtual region
1187  * can return the appropriate address within that region, leaving '*virt'
1188  * unchanged.  We cannot and therefore do not; *virt is updated with the
1189  * first usable address after the mapped region.
1190  */
1191 vm_offset_t
1192 pmap_map(vm_offset_t *virt, vm_offset_t pa_start, vm_offset_t pa_end, int prot)
1193 {
1194 	vm_offset_t	sva, va;
1195 
1196 	sva = *virt;
1197 	va = sva;
1198 	for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE)
1199 		pmap_kenter(va, pa_start);
1200 	*virt = va;
1201 	return (sva);
1202 }
1203 
1204 int
1205 pmap_mincore(pmap_t pmap, vm_offset_t addr)
1206 {
1207 	TODO;
1208 	return (0);
1209 }
1210 
1211 void
1212 pmap_object_init_pt(pmap_t pm, vm_offset_t addr, vm_object_t object,
1213 		    vm_pindex_t pindex, vm_size_t size, int limit)
1214 {
1215 
1216 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1217 	    ("pmap_remove_pages: non current pmap"));
1218 	/* XXX */
1219 }
1220 
1221 /*
1222  * Lower the permission for all mappings to a given page.
1223  */
1224 void
1225 pmap_page_protect(vm_page_t m, vm_prot_t prot)
1226 {
1227 	struct	pvo_head *pvo_head;
1228 	struct	pvo_entry *pvo, *next_pvo;
1229 	struct	pte *pt;
1230 
1231 	/*
1232 	 * Since the routine only downgrades protection, if the
1233 	 * maximal protection is desired, there isn't any change
1234 	 * to be made.
1235 	 */
1236 	if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) ==
1237 	    (VM_PROT_READ|VM_PROT_WRITE))
1238 		return;
1239 
1240 	pvo_head = vm_page_to_pvoh(m);
1241 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1242 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1243 		PMAP_PVO_CHECK(pvo);	/* sanity check */
1244 
1245 		/*
1246 		 * Downgrading to no mapping at all, we just remove the entry.
1247 		 */
1248 		if ((prot & VM_PROT_READ) == 0) {
1249 			pmap_pvo_remove(pvo, -1);
1250 			continue;
1251 		}
1252 
1253 		/*
1254 		 * If EXEC permission is being revoked, just clear the flag
1255 		 * in the PVO.
1256 		 */
1257 		if ((prot & VM_PROT_EXECUTE) == 0)
1258 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1259 
1260 		/*
1261 		 * If this entry is already RO, don't diddle with the page
1262 		 * table.
1263 		 */
1264 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
1265 			PMAP_PVO_CHECK(pvo);
1266 			continue;
1267 		}
1268 
1269 		/*
1270 		 * Grab the PTE before we diddle the bits so pvo_to_pte can
1271 		 * verify the pte contents are as expected.
1272 		 */
1273 		pt = pmap_pvo_to_pte(pvo, -1);
1274 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
1275 		pvo->pvo_pte.pte_lo |= PTE_BR;
1276 		if (pt != NULL)
1277 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1278 		PMAP_PVO_CHECK(pvo);	/* sanity check */
1279 	}
1280 }
1281 
1282 /*
1283  * Make the specified page pageable (or not).  Unneeded.
1284  */
1285 void
1286 pmap_pageable(pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1287 	      boolean_t pageable)
1288 {
1289 }
1290 
1291 /*
1292  * Returns true if the pmap's pv is one of the first
1293  * 16 pvs linked to from this page.  This count may
1294  * be changed upwards or downwards in the future; it
1295  * is only necessary that true be returned for a small
1296  * subset of pmaps for proper page aging.
1297  */
1298 boolean_t
1299 pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
1300 {
1301 	TODO;
1302 	return (0);
1303 }
1304 
1305 static u_int	pmap_vsidcontext;
1306 
1307 void
1308 pmap_pinit(pmap_t pmap)
1309 {
1310 	int	i, mask;
1311 	u_int	entropy;
1312 
1313 	entropy = 0;
1314 	__asm __volatile("mftb %0" : "=r"(entropy));
1315 
1316 	/*
1317 	 * Allocate some segment registers for this pmap.
1318 	 */
1319 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1320 		u_int	hash, n;
1321 
1322 		/*
1323 		 * Create a new value by mutiplying by a prime and adding in
1324 		 * entropy from the timebase register.  This is to make the
1325 		 * VSID more random so that the PT hash function collides
1326 		 * less often.  (Note that the prime casues gcc to do shifts
1327 		 * instead of a multiply.)
1328 		 */
1329 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1330 		hash = pmap_vsidcontext & (NPMAPS - 1);
1331 		if (hash == 0)		/* 0 is special, avoid it */
1332 			continue;
1333 		n = hash >> 5;
1334 		mask = 1 << (hash & (VSID_NBPW - 1));
1335 		hash = (pmap_vsidcontext & 0xfffff);
1336 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1337 			/* anything free in this bucket? */
1338 			if (pmap_vsid_bitmap[n] == 0xffffffff) {
1339 				entropy = (pmap_vsidcontext >> 20);
1340 				continue;
1341 			}
1342 			i = ffs(~pmap_vsid_bitmap[i]) - 1;
1343 			mask = 1 << i;
1344 			hash &= 0xfffff & ~(VSID_NBPW - 1);
1345 			hash |= i;
1346 		}
1347 		pmap_vsid_bitmap[n] |= mask;
1348 		for (i = 0; i < 16; i++)
1349 			pmap->pm_sr[i] = VSID_MAKE(i, hash);
1350 		return;
1351 	}
1352 
1353 	panic("pmap_pinit: out of segments");
1354 }
1355 
1356 /*
1357  * Initialize the pmap associated with process 0.
1358  */
1359 void
1360 pmap_pinit0(pmap_t pm)
1361 {
1362 
1363 	pmap_pinit(pm);
1364 	bzero(&pm->pm_stats, sizeof(pm->pm_stats));
1365 }
1366 
1367 void
1368 pmap_pinit2(pmap_t pmap)
1369 {
1370 	/* XXX: Remove this stub when no longer called */
1371 }
1372 
1373 void
1374 pmap_prefault(pmap_t pm, vm_offset_t va, vm_map_entry_t entry)
1375 {
1376 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1377 	    ("pmap_prefault: non current pmap"));
1378 	/* XXX */
1379 }
1380 
1381 /*
1382  * Set the physical protection on the specified range of this map as requested.
1383  */
1384 void
1385 pmap_protect(pmap_t pm, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1386 {
1387 	struct	pvo_entry *pvo;
1388 	struct	pte *pt;
1389 	int	pteidx;
1390 
1391 	CTR4(KTR_PMAP, "pmap_protect: pm=%p sva=%#x eva=%#x prot=%#x", pm, sva,
1392 	    eva, prot);
1393 
1394 
1395 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1396 	    ("pmap_protect: non current pmap"));
1397 
1398 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1399 		pmap_remove(pm, sva, eva);
1400 		return;
1401 	}
1402 
1403 	for (; sva < eva; sva += PAGE_SIZE) {
1404 		pvo = pmap_pvo_find_va(pm, sva, &pteidx);
1405 		if (pvo == NULL)
1406 			continue;
1407 
1408 		if ((prot & VM_PROT_EXECUTE) == 0)
1409 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1410 
1411 		/*
1412 		 * Grab the PTE pointer before we diddle with the cached PTE
1413 		 * copy.
1414 		 */
1415 		pt = pmap_pvo_to_pte(pvo, pteidx);
1416 		/*
1417 		 * Change the protection of the page.
1418 		 */
1419 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
1420 		pvo->pvo_pte.pte_lo |= PTE_BR;
1421 
1422 		/*
1423 		 * If the PVO is in the page table, update that pte as well.
1424 		 */
1425 		if (pt != NULL)
1426 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1427 	}
1428 }
1429 
1430 vm_offset_t
1431 pmap_phys_address(int ppn)
1432 {
1433 	TODO;
1434 	return (0);
1435 }
1436 
1437 /*
1438  * Map a list of wired pages into kernel virtual address space.  This is
1439  * intended for temporary mappings which do not need page modification or
1440  * references recorded.  Existing mappings in the region are overwritten.
1441  */
1442 void
1443 pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
1444 {
1445 	int	i;
1446 
1447 	for (i = 0; i < count; i++, va += PAGE_SIZE)
1448 		pmap_kenter(va, VM_PAGE_TO_PHYS(m[i]));
1449 }
1450 
1451 /*
1452  * Remove page mappings from kernel virtual address space.  Intended for
1453  * temporary mappings entered by pmap_qenter.
1454  */
1455 void
1456 pmap_qremove(vm_offset_t va, int count)
1457 {
1458 	int	i;
1459 
1460 	for (i = 0; i < count; i++, va += PAGE_SIZE)
1461 		pmap_kremove(va);
1462 }
1463 
1464 void
1465 pmap_release(pmap_t pmap)
1466 {
1467 	TODO;
1468 }
1469 
1470 /*
1471  * Remove the given range of addresses from the specified map.
1472  */
1473 void
1474 pmap_remove(pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1475 {
1476 	struct	pvo_entry *pvo;
1477 	int	pteidx;
1478 
1479 	for (; sva < eva; sva += PAGE_SIZE) {
1480 		pvo = pmap_pvo_find_va(pm, sva, &pteidx);
1481 		if (pvo != NULL) {
1482 			pmap_pvo_remove(pvo, pteidx);
1483 		}
1484 	}
1485 }
1486 
1487 /*
1488  * Remove all pages from specified address space, this aids process exit
1489  * speeds.  This is much faster than pmap_remove in the case of running down
1490  * an entire address space.  Only works for the current pmap.
1491  */
1492 void
1493 pmap_remove_pages(pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1494 {
1495 
1496 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1497 	    ("pmap_remove_pages: non current pmap"));
1498 	pmap_remove(pm, sva, eva);
1499 }
1500 
1501 /*
1502  * Create the kernel stack and pcb for a new thread.
1503  * This routine directly affects the fork perf for a process and
1504  * create performance for a thread.
1505  */
1506 void
1507 pmap_new_thread(struct thread *td)
1508 {
1509 	vm_object_t	ksobj;
1510 	vm_offset_t	ks;
1511 	vm_page_t	m;
1512 	u_int		i;
1513 
1514 	/*
1515 	 * Allocate object for the kstack.
1516 	 */
1517 	ksobj = vm_object_allocate(OBJT_DEFAULT, KSTACK_PAGES);
1518 	td->td_kstack_obj = ksobj;
1519 
1520 	/*
1521 	 * Get a kernel virtual address for the kstack for this thread.
1522 	 */
1523 	ks = kmem_alloc_nofault(kernel_map,
1524 	    (KSTACK_PAGES + KSTACK_GUARD_PAGES) * PAGE_SIZE);
1525 	if (ks == 0)
1526 		panic("pmap_new_thread: kstack allocation failed");
1527 	TLBIE(ks);
1528 	ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
1529 	td->td_kstack = ks;
1530 
1531 	for (i = 0; i < KSTACK_PAGES; i++) {
1532 		/*
1533 		 * Get a kernel stack page.
1534 		 */
1535 		m = vm_page_grab(ksobj, i,
1536 		    VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED);
1537 
1538 		/*
1539 		 * Enter the page into the kernel address space.
1540 		 */
1541 		pmap_kenter(ks + i * PAGE_SIZE, VM_PAGE_TO_PHYS(m));
1542 
1543 		vm_page_wakeup(m);
1544 		vm_page_flag_clear(m, PG_ZERO);
1545 		m->valid = VM_PAGE_BITS_ALL;
1546 	}
1547 }
1548 
1549 void
1550 pmap_dispose_thread(struct thread *td)
1551 {
1552 	TODO;
1553 }
1554 
1555 void
1556 pmap_swapin_thread(struct thread *td)
1557 {
1558 	TODO;
1559 }
1560 
1561 void
1562 pmap_swapout_thread(struct thread *td)
1563 {
1564 	TODO;
1565 }
1566 
1567 /*
1568  * Allocate a physical page of memory directly from the phys_avail map.
1569  * Can only be called from pmap_bootstrap before avail start and end are
1570  * calculated.
1571  */
1572 static vm_offset_t
1573 pmap_bootstrap_alloc(vm_size_t size, u_int align)
1574 {
1575 	vm_offset_t	s, e;
1576 	int		i, j;
1577 
1578 	size = round_page(size);
1579 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1580 		if (align != 0)
1581 			s = (phys_avail[i] + align - 1) & ~(align - 1);
1582 		else
1583 			s = phys_avail[i];
1584 		e = s + size;
1585 
1586 		if (s < phys_avail[i] || e > phys_avail[i + 1])
1587 			continue;
1588 
1589 		if (s == phys_avail[i]) {
1590 			phys_avail[i] += size;
1591 		} else if (e == phys_avail[i + 1]) {
1592 			phys_avail[i + 1] -= size;
1593 		} else {
1594 			for (j = phys_avail_count * 2; j > i; j -= 2) {
1595 				phys_avail[j] = phys_avail[j - 2];
1596 				phys_avail[j + 1] = phys_avail[j - 1];
1597 			}
1598 
1599 			phys_avail[i + 3] = phys_avail[i + 1];
1600 			phys_avail[i + 1] = s;
1601 			phys_avail[i + 2] = e;
1602 			phys_avail_count++;
1603 		}
1604 
1605 		return (s);
1606 	}
1607 	panic("pmap_bootstrap_alloc: could not allocate memory");
1608 }
1609 
1610 /*
1611  * Return an unmapped pvo for a kernel virtual address.
1612  * Used by pmap functions that operate on physical pages.
1613  */
1614 static struct pvo_entry *
1615 pmap_rkva_alloc(void)
1616 {
1617 	struct		pvo_entry *pvo;
1618 	struct		pte *pt;
1619 	vm_offset_t	kva;
1620 	int		pteidx;
1621 
1622 	if (pmap_rkva_count == 0)
1623 		panic("pmap_rkva_alloc: no more reserved KVAs");
1624 
1625 	kva = pmap_rkva_start + (PAGE_SIZE * --pmap_rkva_count);
1626 	pmap_kenter(kva, 0);
1627 
1628 	pvo = pmap_pvo_find_va(kernel_pmap, kva, &pteidx);
1629 
1630 	if (pvo == NULL)
1631 		panic("pmap_kva_alloc: pmap_pvo_find_va failed");
1632 
1633 	pt = pmap_pvo_to_pte(pvo, pteidx);
1634 
1635 	if (pt == NULL)
1636 		panic("pmap_kva_alloc: pmap_pvo_to_pte failed");
1637 
1638 	pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1639 	PVO_PTEGIDX_CLR(pvo);
1640 
1641 	pmap_pte_overflow++;
1642 
1643 	return (pvo);
1644 }
1645 
1646 static void
1647 pmap_pa_map(struct pvo_entry *pvo, vm_offset_t pa, struct pte *saved_pt,
1648     int *depth_p)
1649 {
1650 	struct	pte *pt;
1651 
1652 	/*
1653 	 * If this pvo already has a valid pte, we need to save it so it can
1654 	 * be restored later.  We then just reload the new PTE over the old
1655 	 * slot.
1656 	 */
1657 	if (saved_pt != NULL) {
1658 		pt = pmap_pvo_to_pte(pvo, -1);
1659 
1660 		if (pt != NULL) {
1661 			pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1662 			PVO_PTEGIDX_CLR(pvo);
1663 			pmap_pte_overflow++;
1664 		}
1665 
1666 		*saved_pt = pvo->pvo_pte;
1667 
1668 		pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1669 	}
1670 
1671 	pvo->pvo_pte.pte_lo |= pa;
1672 
1673 	if (!pmap_pte_spill(pvo->pvo_vaddr))
1674 		panic("pmap_pa_map: could not spill pvo %p", pvo);
1675 
1676 	if (depth_p != NULL)
1677 		(*depth_p)++;
1678 }
1679 
1680 static void
1681 pmap_pa_unmap(struct pvo_entry *pvo, struct pte *saved_pt, int *depth_p)
1682 {
1683 	struct	pte *pt;
1684 
1685 	pt = pmap_pvo_to_pte(pvo, -1);
1686 
1687 	if (pt != NULL) {
1688 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1689 		PVO_PTEGIDX_CLR(pvo);
1690 		pmap_pte_overflow++;
1691 	}
1692 
1693 	pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1694 
1695 	/*
1696 	 * If there is a saved PTE and it's valid, restore it and return.
1697 	 */
1698 	if (saved_pt != NULL && (saved_pt->pte_lo & PTE_RPGN) != 0) {
1699 		if (depth_p != NULL && --(*depth_p) == 0)
1700 			panic("pmap_pa_unmap: restoring but depth == 0");
1701 
1702 		pvo->pvo_pte = *saved_pt;
1703 
1704 		if (!pmap_pte_spill(pvo->pvo_vaddr))
1705 			panic("pmap_pa_unmap: could not spill pvo %p", pvo);
1706 	}
1707 }
1708 
1709 static void
1710 pmap_syncicache(vm_offset_t pa, vm_size_t len)
1711 {
1712 	__syncicache((void *)pa, len);
1713 }
1714 
1715 static void
1716 tlbia(void)
1717 {
1718 	caddr_t	i;
1719 
1720 	SYNC();
1721 	for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) {
1722 		TLBIE(i);
1723 		EIEIO();
1724 	}
1725 	TLBSYNC();
1726 	SYNC();
1727 }
1728 
1729 static int
1730 pmap_pvo_enter(pmap_t pm, uma_zone_t zone, struct pvo_head *pvo_head,
1731     vm_offset_t va, vm_offset_t pa, u_int pte_lo, int flags)
1732 {
1733 	struct	pvo_entry *pvo;
1734 	u_int	sr;
1735 	int	first;
1736 	u_int	ptegidx;
1737 	int	i;
1738 
1739 	pmap_pvo_enter_calls++;
1740 	first = 0;
1741 
1742 	/*
1743 	 * Compute the PTE Group index.
1744 	 */
1745 	va &= ~ADDR_POFF;
1746 	sr = va_to_sr(pm->pm_sr, va);
1747 	ptegidx = va_to_pteg(sr, va);
1748 
1749 	/*
1750 	 * Remove any existing mapping for this page.  Reuse the pvo entry if
1751 	 * there is a mapping.
1752 	 */
1753 	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1754 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1755 			if ((pvo->pvo_pte.pte_lo & PTE_RPGN) == pa &&
1756 			    (pvo->pvo_pte.pte_lo & PTE_PP) ==
1757 			    (pte_lo & PTE_PP)) {
1758 				return (0);
1759 			}
1760 			pmap_pvo_remove(pvo, -1);
1761 			break;
1762 		}
1763 	}
1764 
1765 	/*
1766 	 * If we aren't overwriting a mapping, try to allocate.
1767 	 */
1768 	if (pmap_initialized) {
1769 		pvo = uma_zalloc(zone, M_NOWAIT);
1770 	} else {
1771 		if (pmap_bpvo_pool_index >= BPVO_POOL_SIZE) {
1772 			panic("pmap_enter: bpvo pool exhausted, %d, %d, %d",
1773 			      pmap_bpvo_pool_index, BPVO_POOL_SIZE,
1774 			      BPVO_POOL_SIZE * sizeof(struct pvo_entry));
1775 		}
1776 		pvo = &pmap_bpvo_pool[pmap_bpvo_pool_index];
1777 		pmap_bpvo_pool_index++;
1778 		pvo->pvo_vaddr |= PVO_BOOTSTRAP;
1779 	}
1780 
1781 	if (pvo == NULL) {
1782 		return (ENOMEM);
1783 	}
1784 
1785 	pmap_pvo_entries++;
1786 	pvo->pvo_vaddr = va;
1787 	pvo->pvo_pmap = pm;
1788 	LIST_INSERT_HEAD(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1789 	pvo->pvo_vaddr &= ~ADDR_POFF;
1790 	if (flags & VM_PROT_EXECUTE)
1791 		pvo->pvo_vaddr |= PVO_EXECUTABLE;
1792 	if (flags & PVO_WIRED)
1793 		pvo->pvo_vaddr |= PVO_WIRED;
1794 	if (pvo_head != &pmap_pvo_kunmanaged)
1795 		pvo->pvo_vaddr |= PVO_MANAGED;
1796 	pmap_pte_create(&pvo->pvo_pte, sr, va, pa | pte_lo);
1797 
1798 	/*
1799 	 * Remember if the list was empty and therefore will be the first
1800 	 * item.
1801 	 */
1802 	if (LIST_FIRST(pvo_head) == NULL)
1803 		first = 1;
1804 
1805 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1806 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1807 		pvo->pvo_pmap->pm_stats.wired_count++;
1808 	pvo->pvo_pmap->pm_stats.resident_count++;
1809 
1810 	/*
1811 	 * We hope this succeeds but it isn't required.
1812 	 */
1813 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1814 	if (i >= 0) {
1815 		PVO_PTEGIDX_SET(pvo, i);
1816 	} else {
1817 		panic("pmap_pvo_enter: overflow");
1818 		pmap_pte_overflow++;
1819 	}
1820 
1821 	return (first ? ENOENT : 0);
1822 }
1823 
1824 static void
1825 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx)
1826 {
1827 	struct	pte *pt;
1828 
1829 	/*
1830 	 * If there is an active pte entry, we need to deactivate it (and
1831 	 * save the ref & cfg bits).
1832 	 */
1833 	pt = pmap_pvo_to_pte(pvo, pteidx);
1834 	if (pt != NULL) {
1835 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1836 		PVO_PTEGIDX_CLR(pvo);
1837 	} else {
1838 		pmap_pte_overflow--;
1839 	}
1840 
1841 	/*
1842 	 * Update our statistics.
1843 	 */
1844 	pvo->pvo_pmap->pm_stats.resident_count--;
1845 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1846 		pvo->pvo_pmap->pm_stats.wired_count--;
1847 
1848 	/*
1849 	 * Save the REF/CHG bits into their cache if the page is managed.
1850 	 */
1851 	if (pvo->pvo_vaddr & PVO_MANAGED) {
1852 		struct	vm_page *pg;
1853 
1854 		pg = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte_lo & PTE_RPGN);
1855 		if (pg != NULL) {
1856 			pmap_attr_save(pg, pvo->pvo_pte.pte_lo &
1857 			    (PTE_REF | PTE_CHG));
1858 		}
1859 	}
1860 
1861 	/*
1862 	 * Remove this PVO from the PV list.
1863 	 */
1864 	LIST_REMOVE(pvo, pvo_vlink);
1865 
1866 	/*
1867 	 * Remove this from the overflow list and return it to the pool
1868 	 * if we aren't going to reuse it.
1869 	 */
1870 	LIST_REMOVE(pvo, pvo_olink);
1871 	if (!(pvo->pvo_vaddr & PVO_BOOTSTRAP))
1872 		uma_zfree(pvo->pvo_vaddr & PVO_MANAGED ? pmap_mpvo_zone :
1873 		    pmap_upvo_zone, pvo);
1874 	pmap_pvo_entries--;
1875 	pmap_pvo_remove_calls++;
1876 }
1877 
1878 static __inline int
1879 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1880 {
1881 	int	pteidx;
1882 
1883 	/*
1884 	 * We can find the actual pte entry without searching by grabbing
1885 	 * the PTEG index from 3 unused bits in pte_lo[11:9] and by
1886 	 * noticing the HID bit.
1887 	 */
1888 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1889 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1890 		pteidx ^= pmap_pteg_mask * 8;
1891 
1892 	return (pteidx);
1893 }
1894 
1895 static struct pvo_entry *
1896 pmap_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p)
1897 {
1898 	struct	pvo_entry *pvo;
1899 	int	ptegidx;
1900 	u_int	sr;
1901 
1902 	va &= ~ADDR_POFF;
1903 	sr = va_to_sr(pm->pm_sr, va);
1904 	ptegidx = va_to_pteg(sr, va);
1905 
1906 	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1907 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1908 			if (pteidx_p)
1909 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1910 			return (pvo);
1911 		}
1912 	}
1913 
1914 	return (NULL);
1915 }
1916 
1917 static struct pte *
1918 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1919 {
1920 	struct	pte *pt;
1921 
1922 	/*
1923 	 * If we haven't been supplied the ptegidx, calculate it.
1924 	 */
1925 	if (pteidx == -1) {
1926 		int	ptegidx;
1927 		u_int	sr;
1928 
1929 		sr = va_to_sr(pvo->pvo_pmap->pm_sr, pvo->pvo_vaddr);
1930 		ptegidx = va_to_pteg(sr, pvo->pvo_vaddr);
1931 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1932 	}
1933 
1934 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1935 
1936 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1937 		panic("pmap_pvo_to_pte: pvo %p has valid pte in pvo but no "
1938 		    "valid pte index", pvo);
1939 	}
1940 
1941 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1942 		panic("pmap_pvo_to_pte: pvo %p has valid pte index in pvo "
1943 		    "pvo but no valid pte", pvo);
1944 	}
1945 
1946 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1947 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1948 			panic("pmap_pvo_to_pte: pvo %p has valid pte in "
1949 			    "pmap_pteg_table %p but invalid in pvo", pvo, pt);
1950 		}
1951 
1952 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF))
1953 		    != 0) {
1954 			panic("pmap_pvo_to_pte: pvo %p pte does not match "
1955 			    "pte %p in pmap_pteg_table", pvo, pt);
1956 		}
1957 
1958 		return (pt);
1959 	}
1960 
1961 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1962 		panic("pmap_pvo_to_pte: pvo %p has invalid pte %p in "
1963 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1964 	}
1965 
1966 	return (NULL);
1967 }
1968 
1969 static void *
1970 pmap_pvo_allocf(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
1971 {
1972 	vm_page_t	m;
1973 
1974 	if (bytes != PAGE_SIZE)
1975 		panic("pmap_pvo_allocf: benno was shortsighted.  hit him.");
1976 
1977 	*flags = UMA_SLAB_PRIV;
1978 	m = vm_page_alloc(pmap_pvo_obj, pmap_pvo_count, VM_ALLOC_SYSTEM);
1979 	if (m == NULL)
1980 		return (NULL);
1981 	pmap_pvo_count++;
1982 	return ((void *)VM_PAGE_TO_PHYS(m));
1983 }
1984 
1985 /*
1986  * XXX: THIS STUFF SHOULD BE IN pte.c?
1987  */
1988 int
1989 pmap_pte_spill(vm_offset_t addr)
1990 {
1991 	struct	pvo_entry *source_pvo, *victim_pvo;
1992 	struct	pvo_entry *pvo;
1993 	int	ptegidx, i, j;
1994 	u_int	sr;
1995 	struct	pteg *pteg;
1996 	struct	pte *pt;
1997 
1998 	pmap_pte_spills++;
1999 
2000 	sr = mfsrin(addr);
2001 	ptegidx = va_to_pteg(sr, addr);
2002 
2003 	/*
2004 	 * Have to substitute some entry.  Use the primary hash for this.
2005 	 * Use low bits of timebase as random generator.
2006 	 */
2007 	pteg = &pmap_pteg_table[ptegidx];
2008 	__asm __volatile("mftb %0" : "=r"(i));
2009 	i &= 7;
2010 	pt = &pteg->pt[i];
2011 
2012 	source_pvo = NULL;
2013 	victim_pvo = NULL;
2014 	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2015 		/*
2016 		 * We need to find a pvo entry for this address.
2017 		 */
2018 		PMAP_PVO_CHECK(pvo);
2019 		if (source_pvo == NULL &&
2020 		    pmap_pte_match(&pvo->pvo_pte, sr, addr,
2021 		    pvo->pvo_pte.pte_hi & PTE_HID)) {
2022 			/*
2023 			 * Now found an entry to be spilled into the pteg.
2024 			 * The PTE is now valid, so we know it's active.
2025 			 */
2026 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
2027 
2028 			if (j >= 0) {
2029 				PVO_PTEGIDX_SET(pvo, j);
2030 				pmap_pte_overflow--;
2031 				PMAP_PVO_CHECK(pvo);
2032 				return (1);
2033 			}
2034 
2035 			source_pvo = pvo;
2036 
2037 			if (victim_pvo != NULL)
2038 				break;
2039 		}
2040 
2041 		/*
2042 		 * We also need the pvo entry of the victim we are replacing
2043 		 * so save the R & C bits of the PTE.
2044 		 */
2045 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
2046 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
2047 			victim_pvo = pvo;
2048 			if (source_pvo != NULL)
2049 				break;
2050 		}
2051 	}
2052 
2053 	if (source_pvo == NULL)
2054 		return (0);
2055 
2056 	if (victim_pvo == NULL) {
2057 		if ((pt->pte_hi & PTE_HID) == 0)
2058 			panic("pmap_pte_spill: victim p-pte (%p) has no pvo"
2059 			    "entry", pt);
2060 
2061 		/*
2062 		 * If this is a secondary PTE, we need to search it's primary
2063 		 * pvo bucket for the matching PVO.
2064 		 */
2065 		LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx ^ pmap_pteg_mask],
2066 		    pvo_olink) {
2067 			PMAP_PVO_CHECK(pvo);
2068 			/*
2069 			 * We also need the pvo entry of the victim we are
2070 			 * replacing so save the R & C bits of the PTE.
2071 			 */
2072 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
2073 				victim_pvo = pvo;
2074 				break;
2075 			}
2076 		}
2077 
2078 		if (victim_pvo == NULL)
2079 			panic("pmap_pte_spill: victim s-pte (%p) has no pvo"
2080 			    "entry", pt);
2081 	}
2082 
2083 	/*
2084 	 * We are invalidating the TLB entry for the EA we are replacing even
2085 	 * though it's valid.  If we don't, we lose any ref/chg bit changes
2086 	 * contained in the TLB entry.
2087 	 */
2088 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
2089 
2090 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
2091 	pmap_pte_set(pt, &source_pvo->pvo_pte);
2092 
2093 	PVO_PTEGIDX_CLR(victim_pvo);
2094 	PVO_PTEGIDX_SET(source_pvo, i);
2095 	pmap_pte_replacements++;
2096 
2097 	PMAP_PVO_CHECK(victim_pvo);
2098 	PMAP_PVO_CHECK(source_pvo);
2099 
2100 	return (1);
2101 }
2102 
2103 static int
2104 pmap_pte_insert(u_int ptegidx, struct pte *pvo_pt)
2105 {
2106 	struct	pte *pt;
2107 	int	i;
2108 
2109 	/*
2110 	 * First try primary hash.
2111 	 */
2112 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2113 		if ((pt->pte_hi & PTE_VALID) == 0) {
2114 			pvo_pt->pte_hi &= ~PTE_HID;
2115 			pmap_pte_set(pt, pvo_pt);
2116 			return (i);
2117 		}
2118 	}
2119 
2120 	/*
2121 	 * Now try secondary hash.
2122 	 */
2123 	ptegidx ^= pmap_pteg_mask;
2124 	ptegidx++;
2125 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2126 		if ((pt->pte_hi & PTE_VALID) == 0) {
2127 			pvo_pt->pte_hi |= PTE_HID;
2128 			pmap_pte_set(pt, pvo_pt);
2129 			return (i);
2130 		}
2131 	}
2132 
2133 	panic("pmap_pte_insert: overflow");
2134 	return (-1);
2135 }
2136 
2137 static boolean_t
2138 pmap_query_bit(vm_page_t m, int ptebit)
2139 {
2140 	struct	pvo_entry *pvo;
2141 	struct	pte *pt;
2142 
2143 	if (pmap_attr_fetch(m) & ptebit)
2144 		return (TRUE);
2145 
2146 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2147 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2148 
2149 		/*
2150 		 * See if we saved the bit off.  If so, cache it and return
2151 		 * success.
2152 		 */
2153 		if (pvo->pvo_pte.pte_lo & ptebit) {
2154 			pmap_attr_save(m, ptebit);
2155 			PMAP_PVO_CHECK(pvo);	/* sanity check */
2156 			return (TRUE);
2157 		}
2158 	}
2159 
2160 	/*
2161 	 * No luck, now go through the hard part of looking at the PTEs
2162 	 * themselves.  Sync so that any pending REF/CHG bits are flushed to
2163 	 * the PTEs.
2164 	 */
2165 	SYNC();
2166 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2167 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2168 
2169 		/*
2170 		 * See if this pvo has a valid PTE.  if so, fetch the
2171 		 * REF/CHG bits from the valid PTE.  If the appropriate
2172 		 * ptebit is set, cache it and return success.
2173 		 */
2174 		pt = pmap_pvo_to_pte(pvo, -1);
2175 		if (pt != NULL) {
2176 			pmap_pte_synch(pt, &pvo->pvo_pte);
2177 			if (pvo->pvo_pte.pte_lo & ptebit) {
2178 				pmap_attr_save(m, ptebit);
2179 				PMAP_PVO_CHECK(pvo);	/* sanity check */
2180 				return (TRUE);
2181 			}
2182 		}
2183 	}
2184 
2185 	return (TRUE);
2186 }
2187 
2188 static boolean_t
2189 pmap_clear_bit(vm_page_t m, int ptebit)
2190 {
2191 	struct	pvo_entry *pvo;
2192 	struct	pte *pt;
2193 	int	rv;
2194 
2195 	/*
2196 	 * Clear the cached value.
2197 	 */
2198 	rv = pmap_attr_fetch(m);
2199 	pmap_attr_clear(m, ptebit);
2200 
2201 	/*
2202 	 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so
2203 	 * we can reset the right ones).  note that since the pvo entries and
2204 	 * list heads are accessed via BAT0 and are never placed in the page
2205 	 * table, we don't have to worry about further accesses setting the
2206 	 * REF/CHG bits.
2207 	 */
2208 	SYNC();
2209 
2210 	/*
2211 	 * For each pvo entry, clear the pvo's ptebit.  If this pvo has a
2212 	 * valid pte clear the ptebit from the valid pte.
2213 	 */
2214 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2215 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2216 		pt = pmap_pvo_to_pte(pvo, -1);
2217 		if (pt != NULL) {
2218 			pmap_pte_synch(pt, &pvo->pvo_pte);
2219 			if (pvo->pvo_pte.pte_lo & ptebit)
2220 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2221 		}
2222 		rv |= pvo->pvo_pte.pte_lo;
2223 		pvo->pvo_pte.pte_lo &= ~ptebit;
2224 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2225 	}
2226 
2227 	return ((rv & ptebit) != 0);
2228 }
2229 
2230 /*
2231  * Map a set of physical memory pages into the kernel virtual
2232  * address space. Return a pointer to where it is mapped. This
2233  * routine is intended to be used for mapping device memory,
2234  * NOT real memory.
2235  */
2236 void *
2237 pmap_mapdev(vm_offset_t pa, vm_size_t size)
2238 {
2239 	vm_offset_t va, tmpva, offset;
2240 
2241 	pa = trunc_page(pa);
2242 	offset = pa & PAGE_MASK;
2243 	size = roundup(offset + size, PAGE_SIZE);
2244 
2245 	GIANT_REQUIRED;
2246 
2247 	va = kmem_alloc_pageable(kernel_map, size);
2248 	if (!va)
2249 		panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
2250 
2251 	for (tmpva = va; size > 0;) {
2252 		pmap_kenter(tmpva, pa);
2253 		TLBIE(tmpva); /* XXX or should it be invalidate-all ? */
2254 		size -= PAGE_SIZE;
2255 		tmpva += PAGE_SIZE;
2256 		pa += PAGE_SIZE;
2257 	}
2258 
2259 	return ((void *)(va + offset));
2260 }
2261 
2262 void
2263 pmap_unmapdev(vm_offset_t va, vm_size_t size)
2264 {
2265 	vm_offset_t base, offset;
2266 
2267 	base = trunc_page(va);
2268 	offset = va & PAGE_MASK;
2269 	size = roundup(offset + size, PAGE_SIZE);
2270 	kmem_free(kernel_map, base, size);
2271 }
2272