xref: /freebsd/sys/powerpc/aim/mmu_oea.c (revision 2357939bc239bd5334a169b62313806178dd8f30)
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 #include <sys/cdefs.h>
94 __FBSDID("$FreeBSD$");
95 
96 /*
97  * Manages physical address maps.
98  *
99  * In addition to hardware address maps, this module is called upon to
100  * provide software-use-only maps which may or may not be stored in the
101  * same form as hardware maps.  These pseudo-maps are used to store
102  * intermediate results from copy operations to and from address spaces.
103  *
104  * Since the information managed by this module is also stored by the
105  * logical address mapping module, this module may throw away valid virtual
106  * to physical mappings at almost any time.  However, invalidations of
107  * mappings must be done as requested.
108  *
109  * In order to cope with hardware architectures which make virtual to
110  * physical map invalidates expensive, this module may delay invalidate
111  * reduced protection operations until such time as they are actually
112  * necessary.  This module is given full information as to which processors
113  * are currently using which maps, and to when physical maps must be made
114  * correct.
115  */
116 
117 #include "opt_kstack_pages.h"
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/cpu.h>
144 #include <machine/powerpc.h>
145 #include <machine/bat.h>
146 #include <machine/frame.h>
147 #include <machine/md_var.h>
148 #include <machine/psl.h>
149 #include <machine/pte.h>
150 #include <machine/sr.h>
151 
152 #define	PMAP_DEBUG
153 
154 #define TODO	panic("%s: not implemented", __func__);
155 
156 #define	PMAP_LOCK(pm)
157 #define	PMAP_UNLOCK(pm)
158 
159 #define	TLBIE(va)	__asm __volatile("tlbie %0" :: "r"(va))
160 #define	TLBSYNC()	__asm __volatile("tlbsync");
161 #define	SYNC()		__asm __volatile("sync");
162 #define	EIEIO()		__asm __volatile("eieio");
163 
164 #define	VSID_MAKE(sr, hash)	((sr) | (((hash) & 0xfffff) << 4))
165 #define	VSID_TO_SR(vsid)	((vsid) & 0xf)
166 #define	VSID_TO_HASH(vsid)	(((vsid) >> 4) & 0xfffff)
167 
168 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
169 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
170 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
171 #define	PVO_MANAGED		0x0020		/* PVO entry is managed */
172 #define	PVO_EXECUTABLE		0x0040		/* PVO entry is executable */
173 #define	PVO_BOOTSTRAP		0x0080		/* PVO entry allocated during
174 						   bootstrap */
175 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
176 #define	PVO_ISEXECUTABLE(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
177 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
178 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
179 #define	PVO_PTEGIDX_CLR(pvo)	\
180 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
181 #define	PVO_PTEGIDX_SET(pvo, i)	\
182 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
183 
184 #define	PMAP_PVO_CHECK(pvo)
185 
186 struct ofw_map {
187 	vm_offset_t	om_va;
188 	vm_size_t	om_len;
189 	vm_offset_t	om_pa;
190 	u_int		om_mode;
191 };
192 
193 int	pmap_bootstrapped = 0;
194 
195 /*
196  * Virtual and physical address of message buffer.
197  */
198 struct		msgbuf *msgbufp;
199 vm_offset_t	msgbuf_phys;
200 
201 int pmap_pagedaemon_waken;
202 
203 /*
204  * Map of physical memory regions.
205  */
206 vm_offset_t	phys_avail[128];
207 u_int		phys_avail_count;
208 static struct	mem_region *regions;
209 static struct	mem_region *pregions;
210 int		regions_sz, pregions_sz;
211 static struct	ofw_map *translations;
212 
213 /*
214  * First and last available kernel virtual addresses.
215  */
216 vm_offset_t virtual_avail;
217 vm_offset_t virtual_end;
218 vm_offset_t kernel_vm_end;
219 
220 /*
221  * Kernel pmap.
222  */
223 struct pmap kernel_pmap_store;
224 extern struct pmap ofw_pmap;
225 
226 /*
227  * PTEG data.
228  */
229 static struct	pteg *pmap_pteg_table;
230 u_int		pmap_pteg_count;
231 u_int		pmap_pteg_mask;
232 
233 /*
234  * PVO data.
235  */
236 struct	pvo_head *pmap_pvo_table;		/* pvo entries by pteg index */
237 struct	pvo_head pmap_pvo_kunmanaged =
238     LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
239 struct	pvo_head pmap_pvo_unmanaged =
240     LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
241 
242 uma_zone_t	pmap_upvo_zone;	/* zone for pvo entries for unmanaged pages */
243 uma_zone_t	pmap_mpvo_zone;	/* zone for pvo entries for managed pages */
244 
245 #define	BPVO_POOL_SIZE	32768
246 static struct	pvo_entry *pmap_bpvo_pool;
247 static int	pmap_bpvo_pool_index = 0;
248 
249 #define	VSID_NBPW	(sizeof(u_int32_t) * 8)
250 static u_int	pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
251 
252 static boolean_t pmap_initialized = FALSE;
253 
254 /*
255  * Statistics.
256  */
257 u_int	pmap_pte_valid = 0;
258 u_int	pmap_pte_overflow = 0;
259 u_int	pmap_pte_replacements = 0;
260 u_int	pmap_pvo_entries = 0;
261 u_int	pmap_pvo_enter_calls = 0;
262 u_int	pmap_pvo_remove_calls = 0;
263 u_int	pmap_pte_spills = 0;
264 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_valid, CTLFLAG_RD, &pmap_pte_valid,
265     0, "");
266 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_overflow, CTLFLAG_RD,
267     &pmap_pte_overflow, 0, "");
268 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_replacements, CTLFLAG_RD,
269     &pmap_pte_replacements, 0, "");
270 SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_entries, CTLFLAG_RD, &pmap_pvo_entries,
271     0, "");
272 SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_enter_calls, CTLFLAG_RD,
273     &pmap_pvo_enter_calls, 0, "");
274 SYSCTL_INT(_machdep, OID_AUTO, pmap_pvo_remove_calls, CTLFLAG_RD,
275     &pmap_pvo_remove_calls, 0, "");
276 SYSCTL_INT(_machdep, OID_AUTO, pmap_pte_spills, CTLFLAG_RD,
277     &pmap_pte_spills, 0, "");
278 
279 struct	pvo_entry *pmap_pvo_zeropage;
280 
281 vm_offset_t	pmap_rkva_start = VM_MIN_KERNEL_ADDRESS;
282 u_int		pmap_rkva_count = 4;
283 
284 /*
285  * Allocate physical memory for use in pmap_bootstrap.
286  */
287 static vm_offset_t	pmap_bootstrap_alloc(vm_size_t, u_int);
288 
289 /*
290  * PTE calls.
291  */
292 static int		pmap_pte_insert(u_int, struct pte *);
293 
294 /*
295  * PVO calls.
296  */
297 static int	pmap_pvo_enter(pmap_t, uma_zone_t, struct pvo_head *,
298 		    vm_offset_t, vm_offset_t, u_int, int);
299 static void	pmap_pvo_remove(struct pvo_entry *, int);
300 static struct	pvo_entry *pmap_pvo_find_va(pmap_t, vm_offset_t, int *);
301 static struct	pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
302 
303 /*
304  * Utility routines.
305  */
306 static struct		pvo_entry *pmap_rkva_alloc(void);
307 static void		pmap_pa_map(struct pvo_entry *, vm_offset_t,
308 			    struct pte *, int *);
309 static void		pmap_pa_unmap(struct pvo_entry *, struct pte *, int *);
310 static void		pmap_syncicache(vm_offset_t, vm_size_t);
311 static boolean_t	pmap_query_bit(vm_page_t, int);
312 static u_int		pmap_clear_bit(vm_page_t, int, int *);
313 static void		tlbia(void);
314 
315 static __inline int
316 va_to_sr(u_int *sr, vm_offset_t va)
317 {
318 	return (sr[(uintptr_t)va >> ADDR_SR_SHFT]);
319 }
320 
321 static __inline u_int
322 va_to_pteg(u_int sr, vm_offset_t addr)
323 {
324 	u_int hash;
325 
326 	hash = (sr & SR_VSID_MASK) ^ (((u_int)addr & ADDR_PIDX) >>
327 	    ADDR_PIDX_SHFT);
328 	return (hash & pmap_pteg_mask);
329 }
330 
331 static __inline struct pvo_head *
332 pa_to_pvoh(vm_offset_t pa, vm_page_t *pg_p)
333 {
334 	struct	vm_page *pg;
335 
336 	pg = PHYS_TO_VM_PAGE(pa);
337 
338 	if (pg_p != NULL)
339 		*pg_p = pg;
340 
341 	if (pg == NULL)
342 		return (&pmap_pvo_unmanaged);
343 
344 	return (&pg->md.mdpg_pvoh);
345 }
346 
347 static __inline struct pvo_head *
348 vm_page_to_pvoh(vm_page_t m)
349 {
350 
351 	return (&m->md.mdpg_pvoh);
352 }
353 
354 static __inline void
355 pmap_attr_clear(vm_page_t m, int ptebit)
356 {
357 
358 	m->md.mdpg_attrs &= ~ptebit;
359 }
360 
361 static __inline int
362 pmap_attr_fetch(vm_page_t m)
363 {
364 
365 	return (m->md.mdpg_attrs);
366 }
367 
368 static __inline void
369 pmap_attr_save(vm_page_t m, int ptebit)
370 {
371 
372 	m->md.mdpg_attrs |= ptebit;
373 }
374 
375 static __inline int
376 pmap_pte_compare(const struct pte *pt, const struct pte *pvo_pt)
377 {
378 	if (pt->pte_hi == pvo_pt->pte_hi)
379 		return (1);
380 
381 	return (0);
382 }
383 
384 static __inline int
385 pmap_pte_match(struct pte *pt, u_int sr, vm_offset_t va, int which)
386 {
387 	return (pt->pte_hi & ~PTE_VALID) ==
388 	    (((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
389 	    ((va >> ADDR_API_SHFT) & PTE_API) | which);
390 }
391 
392 static __inline void
393 pmap_pte_create(struct pte *pt, u_int sr, vm_offset_t va, u_int pte_lo)
394 {
395 	/*
396 	 * Construct a PTE.  Default to IMB initially.  Valid bit only gets
397 	 * set when the real pte is set in memory.
398 	 *
399 	 * Note: Don't set the valid bit for correct operation of tlb update.
400 	 */
401 	pt->pte_hi = ((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
402 	    (((va & ADDR_PIDX) >> ADDR_API_SHFT) & PTE_API);
403 	pt->pte_lo = pte_lo;
404 }
405 
406 static __inline void
407 pmap_pte_synch(struct pte *pt, struct pte *pvo_pt)
408 {
409 
410 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF | PTE_CHG);
411 }
412 
413 static __inline void
414 pmap_pte_clear(struct pte *pt, vm_offset_t va, int ptebit)
415 {
416 
417 	/*
418 	 * As shown in Section 7.6.3.2.3
419 	 */
420 	pt->pte_lo &= ~ptebit;
421 	TLBIE(va);
422 	EIEIO();
423 	TLBSYNC();
424 	SYNC();
425 }
426 
427 static __inline void
428 pmap_pte_set(struct pte *pt, struct pte *pvo_pt)
429 {
430 
431 	pvo_pt->pte_hi |= PTE_VALID;
432 
433 	/*
434 	 * Update the PTE as defined in section 7.6.3.1.
435 	 * Note that the REF/CHG bits are from pvo_pt and thus should havce
436 	 * been saved so this routine can restore them (if desired).
437 	 */
438 	pt->pte_lo = pvo_pt->pte_lo;
439 	EIEIO();
440 	pt->pte_hi = pvo_pt->pte_hi;
441 	SYNC();
442 	pmap_pte_valid++;
443 }
444 
445 static __inline void
446 pmap_pte_unset(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
447 {
448 
449 	pvo_pt->pte_hi &= ~PTE_VALID;
450 
451 	/*
452 	 * Force the reg & chg bits back into the PTEs.
453 	 */
454 	SYNC();
455 
456 	/*
457 	 * Invalidate the pte.
458 	 */
459 	pt->pte_hi &= ~PTE_VALID;
460 
461 	SYNC();
462 	TLBIE(va);
463 	EIEIO();
464 	TLBSYNC();
465 	SYNC();
466 
467 	/*
468 	 * Save the reg & chg bits.
469 	 */
470 	pmap_pte_synch(pt, pvo_pt);
471 	pmap_pte_valid--;
472 }
473 
474 static __inline void
475 pmap_pte_change(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
476 {
477 
478 	/*
479 	 * Invalidate the PTE
480 	 */
481 	pmap_pte_unset(pt, pvo_pt, va);
482 	pmap_pte_set(pt, pvo_pt);
483 }
484 
485 /*
486  * Quick sort callout for comparing memory regions.
487  */
488 static int	mr_cmp(const void *a, const void *b);
489 static int	om_cmp(const void *a, const void *b);
490 
491 static int
492 mr_cmp(const void *a, const void *b)
493 {
494 	const struct	mem_region *regiona;
495 	const struct	mem_region *regionb;
496 
497 	regiona = a;
498 	regionb = b;
499 	if (regiona->mr_start < regionb->mr_start)
500 		return (-1);
501 	else if (regiona->mr_start > regionb->mr_start)
502 		return (1);
503 	else
504 		return (0);
505 }
506 
507 static int
508 om_cmp(const void *a, const void *b)
509 {
510 	const struct	ofw_map *mapa;
511 	const struct	ofw_map *mapb;
512 
513 	mapa = a;
514 	mapb = b;
515 	if (mapa->om_pa < mapb->om_pa)
516 		return (-1);
517 	else if (mapa->om_pa > mapb->om_pa)
518 		return (1);
519 	else
520 		return (0);
521 }
522 
523 void
524 pmap_bootstrap(vm_offset_t kernelstart, vm_offset_t kernelend)
525 {
526 	ihandle_t	mmui;
527 	phandle_t	chosen, mmu;
528 	int		sz;
529 	int		i, j;
530 	int		ofw_mappings;
531 	vm_size_t	size, physsz;
532 	vm_offset_t	pa, va, off;
533 	u_int		batl, batu;
534 
535         /*
536          * Set up BAT0 to map the lowest 256 MB area
537          */
538         battable[0x0].batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
539         battable[0x0].batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
540 
541         /*
542          * Map PCI memory space.
543          */
544         battable[0x8].batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
545         battable[0x8].batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
546 
547         battable[0x9].batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
548         battable[0x9].batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
549 
550         battable[0xa].batl = BATL(0xa0000000, BAT_I|BAT_G, BAT_PP_RW);
551         battable[0xa].batu = BATU(0xa0000000, BAT_BL_256M, BAT_Vs);
552 
553         battable[0xb].batl = BATL(0xb0000000, BAT_I|BAT_G, BAT_PP_RW);
554         battable[0xb].batu = BATU(0xb0000000, BAT_BL_256M, BAT_Vs);
555 
556         /*
557          * Map obio devices.
558          */
559         battable[0xf].batl = BATL(0xf0000000, BAT_I|BAT_G, BAT_PP_RW);
560         battable[0xf].batu = BATU(0xf0000000, BAT_BL_256M, BAT_Vs);
561 
562 	/*
563 	 * Use an IBAT and a DBAT to map the bottom segment of memory
564 	 * where we are.
565 	 */
566 	batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
567 	batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
568 	__asm ("mtibatu 0,%0; mtibatl 0,%1; mtdbatu 0,%0; mtdbatl 0,%1"
569 	    :: "r"(batu), "r"(batl));
570 
571 #if 0
572 	/* map frame buffer */
573 	batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
574 	batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
575 	__asm ("mtdbatu 1,%0; mtdbatl 1,%1"
576 	    :: "r"(batu), "r"(batl));
577 #endif
578 
579 #if 1
580 	/* map pci space */
581 	batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
582 	batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
583 	__asm ("mtdbatu 1,%0; mtdbatl 1,%1"
584 	    :: "r"(batu), "r"(batl));
585 #endif
586 
587 	/*
588 	 * Set the start and end of kva.
589 	 */
590 	virtual_avail = VM_MIN_KERNEL_ADDRESS;
591 	virtual_end = VM_MAX_KERNEL_ADDRESS;
592 
593 	mem_regions(&pregions, &pregions_sz, &regions, &regions_sz);
594 	CTR0(KTR_PMAP, "pmap_bootstrap: physical memory");
595 
596 	qsort(pregions, pregions_sz, sizeof(*pregions), mr_cmp);
597 	for (i = 0; i < pregions_sz; i++) {
598 		vm_offset_t pa;
599 		vm_offset_t end;
600 
601 		CTR3(KTR_PMAP, "physregion: %#x - %#x (%#x)",
602 			pregions[i].mr_start,
603 			pregions[i].mr_start + pregions[i].mr_size,
604 			pregions[i].mr_size);
605 		/*
606 		 * Install entries into the BAT table to allow all
607 		 * of physmem to be convered by on-demand BAT entries.
608 		 * The loop will sometimes set the same battable element
609 		 * twice, but that's fine since they won't be used for
610 		 * a while yet.
611 		 */
612 		pa = pregions[i].mr_start & 0xf0000000;
613 		end = pregions[i].mr_start + pregions[i].mr_size;
614 		do {
615                         u_int n = pa >> ADDR_SR_SHFT;
616 
617 			battable[n].batl = BATL(pa, BAT_M, BAT_PP_RW);
618 			battable[n].batu = BATU(pa, BAT_BL_256M, BAT_Vs);
619 			pa += SEGMENT_LENGTH;
620 		} while (pa < end);
621 	}
622 
623 	if (sizeof(phys_avail)/sizeof(phys_avail[0]) < regions_sz)
624 		panic("pmap_bootstrap: phys_avail too small");
625 	qsort(regions, regions_sz, sizeof(*regions), mr_cmp);
626 	phys_avail_count = 0;
627 	physsz = 0;
628 	for (i = 0, j = 0; i < regions_sz; i++, j += 2) {
629 		CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start,
630 		    regions[i].mr_start + regions[i].mr_size,
631 		    regions[i].mr_size);
632 		phys_avail[j] = regions[i].mr_start;
633 		phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size;
634 		phys_avail_count++;
635 		physsz += regions[i].mr_size;
636 	}
637 	physmem = btoc(physsz);
638 
639 	/*
640 	 * Allocate PTEG table.
641 	 */
642 #ifdef PTEGCOUNT
643 	pmap_pteg_count = PTEGCOUNT;
644 #else
645 	pmap_pteg_count = 0x1000;
646 
647 	while (pmap_pteg_count < physmem)
648 		pmap_pteg_count <<= 1;
649 
650 	pmap_pteg_count >>= 1;
651 #endif /* PTEGCOUNT */
652 
653 	size = pmap_pteg_count * sizeof(struct pteg);
654 	CTR2(KTR_PMAP, "pmap_bootstrap: %d PTEGs, %d bytes", pmap_pteg_count,
655 	    size);
656 	pmap_pteg_table = (struct pteg *)pmap_bootstrap_alloc(size, size);
657 	CTR1(KTR_PMAP, "pmap_bootstrap: PTEG table at %p", pmap_pteg_table);
658 	bzero((void *)pmap_pteg_table, pmap_pteg_count * sizeof(struct pteg));
659 	pmap_pteg_mask = pmap_pteg_count - 1;
660 
661 	/*
662 	 * Allocate pv/overflow lists.
663 	 */
664 	size = sizeof(struct pvo_head) * pmap_pteg_count;
665 	pmap_pvo_table = (struct pvo_head *)pmap_bootstrap_alloc(size,
666 	    PAGE_SIZE);
667 	CTR1(KTR_PMAP, "pmap_bootstrap: PVO table at %p", pmap_pvo_table);
668 	for (i = 0; i < pmap_pteg_count; i++)
669 		LIST_INIT(&pmap_pvo_table[i]);
670 
671 	/*
672 	 * Allocate the message buffer.
673 	 */
674 	msgbuf_phys = pmap_bootstrap_alloc(MSGBUF_SIZE, 0);
675 
676 	/*
677 	 * Initialise the unmanaged pvo pool.
678 	 */
679 	pmap_bpvo_pool = (struct pvo_entry *)pmap_bootstrap_alloc(
680 		BPVO_POOL_SIZE*sizeof(struct pvo_entry), 0);
681 	pmap_bpvo_pool_index = 0;
682 
683 	/*
684 	 * Make sure kernel vsid is allocated as well as VSID 0.
685 	 */
686 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW]
687 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
688 	pmap_vsid_bitmap[0] |= 1;
689 
690 	/*
691 	 * Set up the OpenFirmware pmap and add it's mappings.
692 	 */
693 	pmap_pinit(&ofw_pmap);
694 	ofw_pmap.pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
695 	ofw_pmap.pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT;
696 	if ((chosen = OF_finddevice("/chosen")) == -1)
697 		panic("pmap_bootstrap: can't find /chosen");
698 	OF_getprop(chosen, "mmu", &mmui, 4);
699 	if ((mmu = OF_instance_to_package(mmui)) == -1)
700 		panic("pmap_bootstrap: can't get mmu package");
701 	if ((sz = OF_getproplen(mmu, "translations")) == -1)
702 		panic("pmap_bootstrap: can't get ofw translation count");
703 	translations = NULL;
704 	for (i = 0; phys_avail[i + 2] != 0; i += 2) {
705 		if (phys_avail[i + 1] >= sz)
706 			translations = (struct ofw_map *)phys_avail[i];
707 	}
708 	if (translations == NULL)
709 		panic("pmap_bootstrap: no space to copy translations");
710 	bzero(translations, sz);
711 	if (OF_getprop(mmu, "translations", translations, sz) == -1)
712 		panic("pmap_bootstrap: can't get ofw translations");
713 	CTR0(KTR_PMAP, "pmap_bootstrap: translations");
714 	sz /= sizeof(*translations);
715 	qsort(translations, sz, sizeof (*translations), om_cmp);
716 	for (i = 0, ofw_mappings = 0; i < sz; i++) {
717 		CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x",
718 		    translations[i].om_pa, translations[i].om_va,
719 		    translations[i].om_len);
720 
721 		/*
722 		 * If the mapping is 1:1, let the RAM and device on-demand
723 		 * BAT tables take care of the translation.
724 		 */
725 		if (translations[i].om_va == translations[i].om_pa)
726 			continue;
727 
728 		/* Enter the pages */
729 		for (off = 0; off < translations[i].om_len; off += PAGE_SIZE) {
730 			struct	vm_page m;
731 
732 			m.phys_addr = translations[i].om_pa + off;
733 			pmap_enter(&ofw_pmap, translations[i].om_va + off, &m,
734 				   VM_PROT_ALL, 1);
735 			ofw_mappings++;
736 		}
737 	}
738 #ifdef SMP
739 	TLBSYNC();
740 #endif
741 
742 	/*
743 	 * Initialize the kernel pmap (which is statically allocated).
744 	 */
745 	for (i = 0; i < 16; i++) {
746 		kernel_pmap->pm_sr[i] = EMPTY_SEGMENT;
747 	}
748 	kernel_pmap->pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
749 	kernel_pmap->pm_sr[KERNEL2_SR] = KERNEL_SEGMENT;
750 	kernel_pmap->pm_active = ~0;
751 
752 	/*
753 	 * Allocate a kernel stack with a guard page for thread0 and map it
754 	 * into the kernel page map.
755 	 */
756 	pa = pmap_bootstrap_alloc(KSTACK_PAGES * PAGE_SIZE, 0);
757 	kstack0_phys = pa;
758 	kstack0 = virtual_avail + (KSTACK_GUARD_PAGES * PAGE_SIZE);
759 	CTR2(KTR_PMAP, "pmap_bootstrap: kstack0 at %#x (%#x)", kstack0_phys,
760 	    kstack0);
761 	virtual_avail += (KSTACK_PAGES + KSTACK_GUARD_PAGES) * PAGE_SIZE;
762 	for (i = 0; i < KSTACK_PAGES; i++) {
763 		pa = kstack0_phys + i * PAGE_SIZE;
764 		va = kstack0 + i * PAGE_SIZE;
765 		pmap_kenter(va, pa);
766 		TLBIE(va);
767 	}
768 
769 	/*
770 	 * Calculate the last available physical address.
771 	 */
772 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
773 		;
774 	Maxmem = powerpc_btop(phys_avail[i + 1]);
775 
776 	/*
777 	 * Allocate virtual address space for the message buffer.
778 	 */
779 	msgbufp = (struct msgbuf *)virtual_avail;
780 	virtual_avail += round_page(MSGBUF_SIZE);
781 
782 	/*
783 	 * Initialize hardware.
784 	 */
785 	for (i = 0; i < 16; i++) {
786 		mtsrin(i << ADDR_SR_SHFT, EMPTY_SEGMENT);
787 	}
788 	__asm __volatile ("mtsr %0,%1"
789 	    :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
790 	__asm __volatile ("sync; mtsdr1 %0; isync"
791 	    :: "r"((u_int)pmap_pteg_table | (pmap_pteg_mask >> 10)));
792 	tlbia();
793 
794 	pmap_bootstrapped++;
795 }
796 
797 /*
798  * Activate a user pmap.  The pmap must be activated before it's address
799  * space can be accessed in any way.
800  */
801 void
802 pmap_activate(struct thread *td)
803 {
804 	pmap_t	pm, pmr;
805 
806 	/*
807 	 * Load all the data we need up front to encourage the compiler to
808 	 * not issue any loads while we have interrupts disabled below.
809 	 */
810 	pm = &td->td_proc->p_vmspace->vm_pmap;
811 
812 	if ((pmr = (pmap_t)pmap_kextract((vm_offset_t)pm)) == NULL)
813 		pmr = pm;
814 
815 	pm->pm_active |= PCPU_GET(cpumask);
816 	PCPU_SET(curpmap, pmr);
817 }
818 
819 void
820 pmap_deactivate(struct thread *td)
821 {
822 	pmap_t	pm;
823 
824 	pm = &td->td_proc->p_vmspace->vm_pmap;
825 	pm->pm_active &= ~(PCPU_GET(cpumask));
826 	PCPU_SET(curpmap, NULL);
827 }
828 
829 vm_offset_t
830 pmap_addr_hint(vm_object_t object, vm_offset_t va, vm_size_t size)
831 {
832 
833 	return (va);
834 }
835 
836 void
837 pmap_change_wiring(pmap_t pm, vm_offset_t va, boolean_t wired)
838 {
839 	struct	pvo_entry *pvo;
840 
841 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
842 
843 	if (pvo != NULL) {
844 		if (wired) {
845 			if ((pvo->pvo_vaddr & PVO_WIRED) == 0)
846 				pm->pm_stats.wired_count++;
847 			pvo->pvo_vaddr |= PVO_WIRED;
848 		} else {
849 			if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
850 				pm->pm_stats.wired_count--;
851 			pvo->pvo_vaddr &= ~PVO_WIRED;
852 		}
853 	}
854 }
855 
856 void
857 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
858 	  vm_size_t len, vm_offset_t src_addr)
859 {
860 
861 	/*
862 	 * This is not needed as it's mainly an optimisation.
863 	 * It may want to be implemented later though.
864 	 */
865 }
866 
867 void
868 pmap_copy_page(vm_page_t msrc, vm_page_t mdst)
869 {
870 	vm_offset_t	dst;
871 	vm_offset_t	src;
872 
873 	dst = VM_PAGE_TO_PHYS(mdst);
874 	src = VM_PAGE_TO_PHYS(msrc);
875 
876 	kcopy((void *)src, (void *)dst, PAGE_SIZE);
877 }
878 
879 /*
880  * Zero a page of physical memory by temporarily mapping it into the tlb.
881  */
882 void
883 pmap_zero_page(vm_page_t m)
884 {
885 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
886 	caddr_t va;
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 	if (pa >= SEGMENT_LENGTH)
902 		pmap_pa_unmap(pmap_pvo_zeropage, NULL, NULL);
903 }
904 
905 void
906 pmap_zero_page_area(vm_page_t m, int off, int size)
907 {
908 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
909 	caddr_t va;
910 
911 	if (pa < SEGMENT_LENGTH) {
912 		va = (caddr_t) pa;
913 	} else if (pmap_initialized) {
914 		if (pmap_pvo_zeropage == NULL)
915 			pmap_pvo_zeropage = pmap_rkva_alloc();
916 		pmap_pa_map(pmap_pvo_zeropage, pa, NULL, NULL);
917 		va = (caddr_t)PVO_VADDR(pmap_pvo_zeropage);
918 	} else {
919 		panic("pmap_zero_page: can't zero pa %#x", pa);
920 	}
921 
922 	bzero(va + off, size);
923 
924 	if (pa >= SEGMENT_LENGTH)
925 		pmap_pa_unmap(pmap_pvo_zeropage, NULL, NULL);
926 }
927 
928 void
929 pmap_zero_page_idle(vm_page_t m)
930 {
931 
932 	/* XXX this is called outside of Giant, is pmap_zero_page safe? */
933 	/* XXX maybe have a dedicated mapping for this to avoid the problem? */
934 	mtx_lock(&Giant);
935 	pmap_zero_page(m);
936 	mtx_unlock(&Giant);
937 }
938 
939 /*
940  * Map the given physical page at the specified virtual address in the
941  * target pmap with the protection requested.  If specified the page
942  * will be wired down.
943  */
944 void
945 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
946 	   boolean_t wired)
947 {
948 	struct		pvo_head *pvo_head;
949 	uma_zone_t	zone;
950 	vm_page_t	pg;
951 	u_int		pte_lo, pvo_flags, was_exec, i;
952 	int		error;
953 
954 	if (!pmap_initialized) {
955 		pvo_head = &pmap_pvo_kunmanaged;
956 		zone = pmap_upvo_zone;
957 		pvo_flags = 0;
958 		pg = NULL;
959 		was_exec = PTE_EXEC;
960 	} else {
961 		pvo_head = vm_page_to_pvoh(m);
962 		pg = m;
963 		zone = pmap_mpvo_zone;
964 		pvo_flags = PVO_MANAGED;
965 		was_exec = 0;
966 	}
967 
968 	/*
969 	 * If this is a managed page, and it's the first reference to the page,
970 	 * clear the execness of the page.  Otherwise fetch the execness.
971 	 */
972 	if (pg != NULL) {
973 		if (LIST_EMPTY(pvo_head)) {
974 			pmap_attr_clear(pg, PTE_EXEC);
975 		} else {
976 			was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
977 		}
978 	}
979 
980 
981 	/*
982 	 * Assume the page is cache inhibited and access is guarded unless
983 	 * it's in our available memory array.
984 	 */
985 	pte_lo = PTE_I | PTE_G;
986 	for (i = 0; i < pregions_sz; i++) {
987 		if ((VM_PAGE_TO_PHYS(m) >= pregions[i].mr_start) &&
988 		    (VM_PAGE_TO_PHYS(m) <
989 			(pregions[i].mr_start + pregions[i].mr_size))) {
990 			pte_lo &= ~(PTE_I | PTE_G);
991 			break;
992 		}
993 	}
994 
995 	if (prot & VM_PROT_WRITE)
996 		pte_lo |= PTE_BW;
997 	else
998 		pte_lo |= PTE_BR;
999 
1000 	pvo_flags |= (prot & VM_PROT_EXECUTE);
1001 
1002 	if (wired)
1003 		pvo_flags |= PVO_WIRED;
1004 
1005 	error = pmap_pvo_enter(pmap, zone, pvo_head, va, VM_PAGE_TO_PHYS(m),
1006 	    pte_lo, pvo_flags);
1007 
1008 	/*
1009 	 * Flush the real page from the instruction cache if this page is
1010 	 * mapped executable and cacheable and was not previously mapped (or
1011 	 * was not mapped executable).
1012 	 */
1013 	if (error == 0 && (pvo_flags & PVO_EXECUTABLE) &&
1014 	    (pte_lo & PTE_I) == 0 && was_exec == 0) {
1015 		/*
1016 		 * Flush the real memory from the cache.
1017 		 */
1018 		pmap_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1019 		if (pg != NULL)
1020 			pmap_attr_save(pg, PTE_EXEC);
1021 	}
1022 
1023 	/* XXX syncicache always until problems are sorted */
1024 	pmap_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1025 }
1026 
1027 vm_page_t
1028 pmap_enter_quick(pmap_t pm, vm_offset_t va, vm_page_t m, vm_page_t mpte)
1029 {
1030 
1031 	pmap_enter(pm, va, m, VM_PROT_READ | VM_PROT_EXECUTE, FALSE);
1032 	return (NULL);
1033 }
1034 
1035 vm_offset_t
1036 pmap_extract(pmap_t pm, vm_offset_t va)
1037 {
1038 	struct	pvo_entry *pvo;
1039 
1040 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1041 
1042 	if (pvo != NULL) {
1043 		return ((pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF));
1044 	}
1045 
1046 	return (0);
1047 }
1048 
1049 /*
1050  * Atomically extract and hold the physical page with the given
1051  * pmap and virtual address pair if that mapping permits the given
1052  * protection.
1053  */
1054 vm_page_t
1055 pmap_extract_and_hold(pmap_t pmap, vm_offset_t va, vm_prot_t prot)
1056 {
1057 	vm_paddr_t pa;
1058 	vm_page_t m;
1059 
1060 	m = NULL;
1061 	mtx_lock(&Giant);
1062 	if ((pa = pmap_extract(pmap, va)) != 0) {
1063 		m = PHYS_TO_VM_PAGE(pa);
1064 		vm_page_lock_queues();
1065 		vm_page_hold(m);
1066 		vm_page_unlock_queues();
1067 	}
1068 	mtx_unlock(&Giant);
1069 	return (m);
1070 }
1071 
1072 /*
1073  * Grow the number of kernel page table entries.  Unneeded.
1074  */
1075 void
1076 pmap_growkernel(vm_offset_t addr)
1077 {
1078 }
1079 
1080 void
1081 pmap_init(void)
1082 {
1083 
1084 	CTR0(KTR_PMAP, "pmap_init");
1085 
1086 	pmap_upvo_zone = uma_zcreate("UPVO entry", sizeof (struct pvo_entry),
1087 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1088 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1089 	pmap_mpvo_zone = uma_zcreate("MPVO entry", sizeof(struct pvo_entry),
1090 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1091 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1092 	pmap_initialized = TRUE;
1093 }
1094 
1095 void
1096 pmap_init2(void)
1097 {
1098 
1099 	CTR0(KTR_PMAP, "pmap_init2");
1100 }
1101 
1102 boolean_t
1103 pmap_is_modified(vm_page_t m)
1104 {
1105 
1106 	if ((m->flags & (PG_FICTITIOUS |PG_UNMANAGED)) != 0)
1107 		return (FALSE);
1108 
1109 	return (pmap_query_bit(m, PTE_CHG));
1110 }
1111 
1112 /*
1113  *	pmap_is_prefaultable:
1114  *
1115  *	Return whether or not the specified virtual address is elgible
1116  *	for prefault.
1117  */
1118 boolean_t
1119 pmap_is_prefaultable(pmap_t pmap, vm_offset_t addr)
1120 {
1121 
1122 	return (FALSE);
1123 }
1124 
1125 void
1126 pmap_clear_reference(vm_page_t m)
1127 {
1128 
1129 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1130 		return;
1131 	pmap_clear_bit(m, PTE_REF, NULL);
1132 }
1133 
1134 void
1135 pmap_clear_modify(vm_page_t m)
1136 {
1137 
1138 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1139 		return;
1140 	pmap_clear_bit(m, PTE_CHG, NULL);
1141 }
1142 
1143 /*
1144  *	pmap_ts_referenced:
1145  *
1146  *	Return a count of reference bits for a page, clearing those bits.
1147  *	It is not necessary for every reference bit to be cleared, but it
1148  *	is necessary that 0 only be returned when there are truly no
1149  *	reference bits set.
1150  *
1151  *	XXX: The exact number of bits to check and clear is a matter that
1152  *	should be tested and standardized at some point in the future for
1153  *	optimal aging of shared pages.
1154  */
1155 int
1156 pmap_ts_referenced(vm_page_t m)
1157 {
1158 	int count;
1159 
1160 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1161 		return (0);
1162 
1163 	count = pmap_clear_bit(m, PTE_REF, NULL);
1164 
1165 	return (count);
1166 }
1167 
1168 /*
1169  * Map a wired page into kernel virtual address space.
1170  */
1171 void
1172 pmap_kenter(vm_offset_t va, vm_offset_t pa)
1173 {
1174 	u_int		pte_lo;
1175 	int		error;
1176 	int		i;
1177 
1178 #if 0
1179 	if (va < VM_MIN_KERNEL_ADDRESS)
1180 		panic("pmap_kenter: attempt to enter non-kernel address %#x",
1181 		    va);
1182 #endif
1183 
1184 	pte_lo = PTE_I | PTE_G;
1185 	for (i = 0; i < pregions_sz; i++) {
1186 		if ((pa >= pregions[i].mr_start) &&
1187 		    (pa < (pregions[i].mr_start + pregions[i].mr_size))) {
1188 			pte_lo &= ~(PTE_I | PTE_G);
1189 			break;
1190 		}
1191 	}
1192 
1193 	error = pmap_pvo_enter(kernel_pmap, pmap_upvo_zone,
1194 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, PVO_WIRED);
1195 
1196 	if (error != 0 && error != ENOENT)
1197 		panic("pmap_kenter: failed to enter va %#x pa %#x: %d", va,
1198 		    pa, error);
1199 
1200 	/*
1201 	 * Flush the real memory from the instruction cache.
1202 	 */
1203 	if ((pte_lo & (PTE_I | PTE_G)) == 0) {
1204 		pmap_syncicache(pa, PAGE_SIZE);
1205 	}
1206 }
1207 
1208 /*
1209  * Extract the physical page address associated with the given kernel virtual
1210  * address.
1211  */
1212 vm_offset_t
1213 pmap_kextract(vm_offset_t va)
1214 {
1215 	struct		pvo_entry *pvo;
1216 
1217 #ifdef UMA_MD_SMALL_ALLOC
1218 	/*
1219 	 * Allow direct mappings
1220 	 */
1221 	if (va < VM_MIN_KERNEL_ADDRESS) {
1222 		return (va);
1223 	}
1224 #endif
1225 
1226 	pvo = pmap_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL);
1227 	KASSERT(pvo != NULL, ("pmap_kextract: no addr found"));
1228 	if (pvo == NULL) {
1229 		return (0);
1230 	}
1231 
1232 	return ((pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF));
1233 }
1234 
1235 /*
1236  * Remove a wired page from kernel virtual address space.
1237  */
1238 void
1239 pmap_kremove(vm_offset_t va)
1240 {
1241 
1242 	pmap_remove(kernel_pmap, va, va + PAGE_SIZE);
1243 }
1244 
1245 /*
1246  * Map a range of physical addresses into kernel virtual address space.
1247  *
1248  * The value passed in *virt is a suggested virtual address for the mapping.
1249  * Architectures which can support a direct-mapped physical to virtual region
1250  * can return the appropriate address within that region, leaving '*virt'
1251  * unchanged.  We cannot and therefore do not; *virt is updated with the
1252  * first usable address after the mapped region.
1253  */
1254 vm_offset_t
1255 pmap_map(vm_offset_t *virt, vm_offset_t pa_start, vm_offset_t pa_end, int prot)
1256 {
1257 	vm_offset_t	sva, va;
1258 
1259 	sva = *virt;
1260 	va = sva;
1261 	for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE)
1262 		pmap_kenter(va, pa_start);
1263 	*virt = va;
1264 	return (sva);
1265 }
1266 
1267 int
1268 pmap_mincore(pmap_t pmap, vm_offset_t addr)
1269 {
1270 	TODO;
1271 	return (0);
1272 }
1273 
1274 void
1275 pmap_object_init_pt(pmap_t pm, vm_offset_t addr, vm_object_t object,
1276 		    vm_pindex_t pindex, vm_size_t size)
1277 {
1278 
1279 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
1280 	KASSERT(object->type == OBJT_DEVICE,
1281 	    ("pmap_object_init_pt: non-device object"));
1282 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1283 	    ("pmap_object_init_pt: non current pmap"));
1284 }
1285 
1286 /*
1287  * Lower the permission for all mappings to a given page.
1288  */
1289 void
1290 pmap_page_protect(vm_page_t m, vm_prot_t prot)
1291 {
1292 	struct	pvo_head *pvo_head;
1293 	struct	pvo_entry *pvo, *next_pvo;
1294 	struct	pte *pt;
1295 
1296 	/*
1297 	 * Since the routine only downgrades protection, if the
1298 	 * maximal protection is desired, there isn't any change
1299 	 * to be made.
1300 	 */
1301 	if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) ==
1302 	    (VM_PROT_READ|VM_PROT_WRITE))
1303 		return;
1304 
1305 	pvo_head = vm_page_to_pvoh(m);
1306 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1307 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1308 		PMAP_PVO_CHECK(pvo);	/* sanity check */
1309 
1310 		/*
1311 		 * Downgrading to no mapping at all, we just remove the entry.
1312 		 */
1313 		if ((prot & VM_PROT_READ) == 0) {
1314 			pmap_pvo_remove(pvo, -1);
1315 			continue;
1316 		}
1317 
1318 		/*
1319 		 * If EXEC permission is being revoked, just clear the flag
1320 		 * in the PVO.
1321 		 */
1322 		if ((prot & VM_PROT_EXECUTE) == 0)
1323 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1324 
1325 		/*
1326 		 * If this entry is already RO, don't diddle with the page
1327 		 * table.
1328 		 */
1329 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
1330 			PMAP_PVO_CHECK(pvo);
1331 			continue;
1332 		}
1333 
1334 		/*
1335 		 * Grab the PTE before we diddle the bits so pvo_to_pte can
1336 		 * verify the pte contents are as expected.
1337 		 */
1338 		pt = pmap_pvo_to_pte(pvo, -1);
1339 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
1340 		pvo->pvo_pte.pte_lo |= PTE_BR;
1341 		if (pt != NULL)
1342 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1343 		PMAP_PVO_CHECK(pvo);	/* sanity check */
1344 	}
1345 }
1346 
1347 /*
1348  * Returns true if the pmap's pv is one of the first
1349  * 16 pvs linked to from this page.  This count may
1350  * be changed upwards or downwards in the future; it
1351  * is only necessary that true be returned for a small
1352  * subset of pmaps for proper page aging.
1353  */
1354 boolean_t
1355 pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
1356 {
1357         int loops;
1358 	struct pvo_entry *pvo;
1359 
1360         if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
1361                 return FALSE;
1362 
1363 	loops = 0;
1364 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1365 		if (pvo->pvo_pmap == pmap)
1366 			return (TRUE);
1367 		if (++loops >= 16)
1368 			break;
1369 	}
1370 
1371 	return (FALSE);
1372 }
1373 
1374 static u_int	pmap_vsidcontext;
1375 
1376 void
1377 pmap_pinit(pmap_t pmap)
1378 {
1379 	int	i, mask;
1380 	u_int	entropy;
1381 
1382 	KASSERT((int)pmap < VM_MIN_KERNEL_ADDRESS, ("pmap_pinit: virt pmap"));
1383 
1384 	entropy = 0;
1385 	__asm __volatile("mftb %0" : "=r"(entropy));
1386 
1387 	/*
1388 	 * Allocate some segment registers for this pmap.
1389 	 */
1390 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1391 		u_int	hash, n;
1392 
1393 		/*
1394 		 * Create a new value by mutiplying by a prime and adding in
1395 		 * entropy from the timebase register.  This is to make the
1396 		 * VSID more random so that the PT hash function collides
1397 		 * less often.  (Note that the prime casues gcc to do shifts
1398 		 * instead of a multiply.)
1399 		 */
1400 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1401 		hash = pmap_vsidcontext & (NPMAPS - 1);
1402 		if (hash == 0)		/* 0 is special, avoid it */
1403 			continue;
1404 		n = hash >> 5;
1405 		mask = 1 << (hash & (VSID_NBPW - 1));
1406 		hash = (pmap_vsidcontext & 0xfffff);
1407 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1408 			/* anything free in this bucket? */
1409 			if (pmap_vsid_bitmap[n] == 0xffffffff) {
1410 				entropy = (pmap_vsidcontext >> 20);
1411 				continue;
1412 			}
1413 			i = ffs(~pmap_vsid_bitmap[i]) - 1;
1414 			mask = 1 << i;
1415 			hash &= 0xfffff & ~(VSID_NBPW - 1);
1416 			hash |= i;
1417 		}
1418 		pmap_vsid_bitmap[n] |= mask;
1419 		for (i = 0; i < 16; i++)
1420 			pmap->pm_sr[i] = VSID_MAKE(i, hash);
1421 		return;
1422 	}
1423 
1424 	panic("pmap_pinit: out of segments");
1425 }
1426 
1427 /*
1428  * Initialize the pmap associated with process 0.
1429  */
1430 void
1431 pmap_pinit0(pmap_t pm)
1432 {
1433 
1434 	pmap_pinit(pm);
1435 	bzero(&pm->pm_stats, sizeof(pm->pm_stats));
1436 }
1437 
1438 /*
1439  * Set the physical protection on the specified range of this map as requested.
1440  */
1441 void
1442 pmap_protect(pmap_t pm, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1443 {
1444 	struct	pvo_entry *pvo;
1445 	struct	pte *pt;
1446 	int	pteidx;
1447 
1448 	CTR4(KTR_PMAP, "pmap_protect: pm=%p sva=%#x eva=%#x prot=%#x", pm, sva,
1449 	    eva, prot);
1450 
1451 
1452 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1453 	    ("pmap_protect: non current pmap"));
1454 
1455 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1456 		pmap_remove(pm, sva, eva);
1457 		return;
1458 	}
1459 
1460 	for (; sva < eva; sva += PAGE_SIZE) {
1461 		pvo = pmap_pvo_find_va(pm, sva, &pteidx);
1462 		if (pvo == NULL)
1463 			continue;
1464 
1465 		if ((prot & VM_PROT_EXECUTE) == 0)
1466 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1467 
1468 		/*
1469 		 * Grab the PTE pointer before we diddle with the cached PTE
1470 		 * copy.
1471 		 */
1472 		pt = pmap_pvo_to_pte(pvo, pteidx);
1473 		/*
1474 		 * Change the protection of the page.
1475 		 */
1476 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
1477 		pvo->pvo_pte.pte_lo |= PTE_BR;
1478 
1479 		/*
1480 		 * If the PVO is in the page table, update that pte as well.
1481 		 */
1482 		if (pt != NULL)
1483 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1484 	}
1485 }
1486 
1487 /*
1488  * Map a list of wired pages into kernel virtual address space.  This is
1489  * intended for temporary mappings which do not need page modification or
1490  * references recorded.  Existing mappings in the region are overwritten.
1491  */
1492 void
1493 pmap_qenter(vm_offset_t sva, vm_page_t *m, int count)
1494 {
1495 	vm_offset_t va;
1496 
1497 	va = sva;
1498 	while (count-- > 0) {
1499 		pmap_kenter(va, VM_PAGE_TO_PHYS(*m));
1500 		va += PAGE_SIZE;
1501 		m++;
1502 	}
1503 }
1504 
1505 /*
1506  * Remove page mappings from kernel virtual address space.  Intended for
1507  * temporary mappings entered by pmap_qenter.
1508  */
1509 void
1510 pmap_qremove(vm_offset_t sva, int count)
1511 {
1512 	vm_offset_t va;
1513 
1514 	va = sva;
1515 	while (count-- > 0) {
1516 		pmap_kremove(va);
1517 		va += PAGE_SIZE;
1518 	}
1519 }
1520 
1521 void
1522 pmap_release(pmap_t pmap)
1523 {
1524         int idx, mask;
1525 
1526 	/*
1527 	 * Free segment register's VSID
1528 	 */
1529         if (pmap->pm_sr[0] == 0)
1530                 panic("pmap_release");
1531 
1532         idx = VSID_TO_HASH(pmap->pm_sr[0]) & (NPMAPS-1);
1533         mask = 1 << (idx % VSID_NBPW);
1534         idx /= VSID_NBPW;
1535         pmap_vsid_bitmap[idx] &= ~mask;
1536 }
1537 
1538 /*
1539  * Remove the given range of addresses from the specified map.
1540  */
1541 void
1542 pmap_remove(pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1543 {
1544 	struct	pvo_entry *pvo;
1545 	int	pteidx;
1546 
1547 	for (; sva < eva; sva += PAGE_SIZE) {
1548 		pvo = pmap_pvo_find_va(pm, sva, &pteidx);
1549 		if (pvo != NULL) {
1550 			pmap_pvo_remove(pvo, pteidx);
1551 		}
1552 	}
1553 }
1554 
1555 /*
1556  * Remove physical page from all pmaps in which it resides. pmap_pvo_remove()
1557  * will reflect changes in pte's back to the vm_page.
1558  */
1559 void
1560 pmap_remove_all(vm_page_t m)
1561 {
1562 	struct  pvo_head *pvo_head;
1563 	struct	pvo_entry *pvo, *next_pvo;
1564 
1565 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1566 
1567 	pvo_head = vm_page_to_pvoh(m);
1568 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1569 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1570 
1571 		PMAP_PVO_CHECK(pvo);	/* sanity check */
1572 		pmap_pvo_remove(pvo, -1);
1573 	}
1574 	vm_page_flag_clear(m, PG_WRITEABLE);
1575 }
1576 
1577 /*
1578  * Remove all pages from specified address space, this aids process exit
1579  * speeds.  This is much faster than pmap_remove in the case of running down
1580  * an entire address space.  Only works for the current pmap.
1581  */
1582 void
1583 pmap_remove_pages(pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1584 {
1585 
1586 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1587 	    ("pmap_remove_pages: non current pmap"));
1588 	pmap_remove(pm, sva, eva);
1589 }
1590 
1591 /*
1592  * Allocate a physical page of memory directly from the phys_avail map.
1593  * Can only be called from pmap_bootstrap before avail start and end are
1594  * calculated.
1595  */
1596 static vm_offset_t
1597 pmap_bootstrap_alloc(vm_size_t size, u_int align)
1598 {
1599 	vm_offset_t	s, e;
1600 	int		i, j;
1601 
1602 	size = round_page(size);
1603 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1604 		if (align != 0)
1605 			s = (phys_avail[i] + align - 1) & ~(align - 1);
1606 		else
1607 			s = phys_avail[i];
1608 		e = s + size;
1609 
1610 		if (s < phys_avail[i] || e > phys_avail[i + 1])
1611 			continue;
1612 
1613 		if (s == phys_avail[i]) {
1614 			phys_avail[i] += size;
1615 		} else if (e == phys_avail[i + 1]) {
1616 			phys_avail[i + 1] -= size;
1617 		} else {
1618 			for (j = phys_avail_count * 2; j > i; j -= 2) {
1619 				phys_avail[j] = phys_avail[j - 2];
1620 				phys_avail[j + 1] = phys_avail[j - 1];
1621 			}
1622 
1623 			phys_avail[i + 3] = phys_avail[i + 1];
1624 			phys_avail[i + 1] = s;
1625 			phys_avail[i + 2] = e;
1626 			phys_avail_count++;
1627 		}
1628 
1629 		return (s);
1630 	}
1631 	panic("pmap_bootstrap_alloc: could not allocate memory");
1632 }
1633 
1634 /*
1635  * Return an unmapped pvo for a kernel virtual address.
1636  * Used by pmap functions that operate on physical pages.
1637  */
1638 static struct pvo_entry *
1639 pmap_rkva_alloc(void)
1640 {
1641 	struct		pvo_entry *pvo;
1642 	struct		pte *pt;
1643 	vm_offset_t	kva;
1644 	int		pteidx;
1645 
1646 	if (pmap_rkva_count == 0)
1647 		panic("pmap_rkva_alloc: no more reserved KVAs");
1648 
1649 	kva = pmap_rkva_start + (PAGE_SIZE * --pmap_rkva_count);
1650 	pmap_kenter(kva, 0);
1651 
1652 	pvo = pmap_pvo_find_va(kernel_pmap, kva, &pteidx);
1653 
1654 	if (pvo == NULL)
1655 		panic("pmap_kva_alloc: pmap_pvo_find_va failed");
1656 
1657 	pt = pmap_pvo_to_pte(pvo, pteidx);
1658 
1659 	if (pt == NULL)
1660 		panic("pmap_kva_alloc: pmap_pvo_to_pte failed");
1661 
1662 	pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1663 	PVO_PTEGIDX_CLR(pvo);
1664 
1665 	pmap_pte_overflow++;
1666 
1667 	return (pvo);
1668 }
1669 
1670 static void
1671 pmap_pa_map(struct pvo_entry *pvo, vm_offset_t pa, struct pte *saved_pt,
1672     int *depth_p)
1673 {
1674 	struct	pte *pt;
1675 
1676 	/*
1677 	 * If this pvo already has a valid pte, we need to save it so it can
1678 	 * be restored later.  We then just reload the new PTE over the old
1679 	 * slot.
1680 	 */
1681 	if (saved_pt != NULL) {
1682 		pt = pmap_pvo_to_pte(pvo, -1);
1683 
1684 		if (pt != NULL) {
1685 			pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1686 			PVO_PTEGIDX_CLR(pvo);
1687 			pmap_pte_overflow++;
1688 		}
1689 
1690 		*saved_pt = pvo->pvo_pte;
1691 
1692 		pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1693 	}
1694 
1695 	pvo->pvo_pte.pte_lo |= pa;
1696 
1697 	if (!pmap_pte_spill(pvo->pvo_vaddr))
1698 		panic("pmap_pa_map: could not spill pvo %p", pvo);
1699 
1700 	if (depth_p != NULL)
1701 		(*depth_p)++;
1702 }
1703 
1704 static void
1705 pmap_pa_unmap(struct pvo_entry *pvo, struct pte *saved_pt, int *depth_p)
1706 {
1707 	struct	pte *pt;
1708 
1709 	pt = pmap_pvo_to_pte(pvo, -1);
1710 
1711 	if (pt != NULL) {
1712 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1713 		PVO_PTEGIDX_CLR(pvo);
1714 		pmap_pte_overflow++;
1715 	}
1716 
1717 	pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1718 
1719 	/*
1720 	 * If there is a saved PTE and it's valid, restore it and return.
1721 	 */
1722 	if (saved_pt != NULL && (saved_pt->pte_lo & PTE_RPGN) != 0) {
1723 		if (depth_p != NULL && --(*depth_p) == 0)
1724 			panic("pmap_pa_unmap: restoring but depth == 0");
1725 
1726 		pvo->pvo_pte = *saved_pt;
1727 
1728 		if (!pmap_pte_spill(pvo->pvo_vaddr))
1729 			panic("pmap_pa_unmap: could not spill pvo %p", pvo);
1730 	}
1731 }
1732 
1733 static void
1734 pmap_syncicache(vm_offset_t pa, vm_size_t len)
1735 {
1736 	__syncicache((void *)pa, len);
1737 }
1738 
1739 static void
1740 tlbia(void)
1741 {
1742 	caddr_t	i;
1743 
1744 	SYNC();
1745 	for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) {
1746 		TLBIE(i);
1747 		EIEIO();
1748 	}
1749 	TLBSYNC();
1750 	SYNC();
1751 }
1752 
1753 static int
1754 pmap_pvo_enter(pmap_t pm, uma_zone_t zone, struct pvo_head *pvo_head,
1755     vm_offset_t va, vm_offset_t pa, u_int pte_lo, int flags)
1756 {
1757 	struct	pvo_entry *pvo;
1758 	u_int	sr;
1759 	int	first;
1760 	u_int	ptegidx;
1761 	int	i;
1762 	int     bootstrap;
1763 
1764 	pmap_pvo_enter_calls++;
1765 	first = 0;
1766 
1767 	bootstrap = 0;
1768 
1769 	/*
1770 	 * Compute the PTE Group index.
1771 	 */
1772 	va &= ~ADDR_POFF;
1773 	sr = va_to_sr(pm->pm_sr, va);
1774 	ptegidx = va_to_pteg(sr, va);
1775 
1776 	/*
1777 	 * Remove any existing mapping for this page.  Reuse the pvo entry if
1778 	 * there is a mapping.
1779 	 */
1780 	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1781 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1782 			if ((pvo->pvo_pte.pte_lo & PTE_RPGN) == pa &&
1783 			    (pvo->pvo_pte.pte_lo & PTE_PP) ==
1784 			    (pte_lo & PTE_PP)) {
1785 				return (0);
1786 			}
1787 			pmap_pvo_remove(pvo, -1);
1788 			break;
1789 		}
1790 	}
1791 
1792 	/*
1793 	 * If we aren't overwriting a mapping, try to allocate.
1794 	 */
1795 	if (pmap_initialized) {
1796 		pvo = uma_zalloc(zone, M_NOWAIT);
1797 	} else {
1798 		if (pmap_bpvo_pool_index >= BPVO_POOL_SIZE) {
1799 			panic("pmap_enter: bpvo pool exhausted, %d, %d, %d",
1800 			      pmap_bpvo_pool_index, BPVO_POOL_SIZE,
1801 			      BPVO_POOL_SIZE * sizeof(struct pvo_entry));
1802 		}
1803 		pvo = &pmap_bpvo_pool[pmap_bpvo_pool_index];
1804 		pmap_bpvo_pool_index++;
1805 		bootstrap = 1;
1806 	}
1807 
1808 	if (pvo == NULL) {
1809 		return (ENOMEM);
1810 	}
1811 
1812 	pmap_pvo_entries++;
1813 	pvo->pvo_vaddr = va;
1814 	pvo->pvo_pmap = pm;
1815 	LIST_INSERT_HEAD(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1816 	pvo->pvo_vaddr &= ~ADDR_POFF;
1817 	if (flags & VM_PROT_EXECUTE)
1818 		pvo->pvo_vaddr |= PVO_EXECUTABLE;
1819 	if (flags & PVO_WIRED)
1820 		pvo->pvo_vaddr |= PVO_WIRED;
1821 	if (pvo_head != &pmap_pvo_kunmanaged)
1822 		pvo->pvo_vaddr |= PVO_MANAGED;
1823 	if (bootstrap)
1824 		pvo->pvo_vaddr |= PVO_BOOTSTRAP;
1825 	pmap_pte_create(&pvo->pvo_pte, sr, va, pa | pte_lo);
1826 
1827 	/*
1828 	 * Remember if the list was empty and therefore will be the first
1829 	 * item.
1830 	 */
1831 	if (LIST_FIRST(pvo_head) == NULL)
1832 		first = 1;
1833 
1834 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1835 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1836 		pvo->pvo_pmap->pm_stats.wired_count++;
1837 	pvo->pvo_pmap->pm_stats.resident_count++;
1838 
1839 	/*
1840 	 * We hope this succeeds but it isn't required.
1841 	 */
1842 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1843 	if (i >= 0) {
1844 		PVO_PTEGIDX_SET(pvo, i);
1845 	} else {
1846 		panic("pmap_pvo_enter: overflow");
1847 		pmap_pte_overflow++;
1848 	}
1849 
1850 	return (first ? ENOENT : 0);
1851 }
1852 
1853 static void
1854 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx)
1855 {
1856 	struct	pte *pt;
1857 
1858 	/*
1859 	 * If there is an active pte entry, we need to deactivate it (and
1860 	 * save the ref & cfg bits).
1861 	 */
1862 	pt = pmap_pvo_to_pte(pvo, pteidx);
1863 	if (pt != NULL) {
1864 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1865 		PVO_PTEGIDX_CLR(pvo);
1866 	} else {
1867 		pmap_pte_overflow--;
1868 	}
1869 
1870 	/*
1871 	 * Update our statistics.
1872 	 */
1873 	pvo->pvo_pmap->pm_stats.resident_count--;
1874 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1875 		pvo->pvo_pmap->pm_stats.wired_count--;
1876 
1877 	/*
1878 	 * Save the REF/CHG bits into their cache if the page is managed.
1879 	 */
1880 	if (pvo->pvo_vaddr & PVO_MANAGED) {
1881 		struct	vm_page *pg;
1882 
1883 		pg = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte_lo & PTE_RPGN);
1884 		if (pg != NULL) {
1885 			pmap_attr_save(pg, pvo->pvo_pte.pte_lo &
1886 			    (PTE_REF | PTE_CHG));
1887 		}
1888 	}
1889 
1890 	/*
1891 	 * Remove this PVO from the PV list.
1892 	 */
1893 	LIST_REMOVE(pvo, pvo_vlink);
1894 
1895 	/*
1896 	 * Remove this from the overflow list and return it to the pool
1897 	 * if we aren't going to reuse it.
1898 	 */
1899 	LIST_REMOVE(pvo, pvo_olink);
1900 	if (!(pvo->pvo_vaddr & PVO_BOOTSTRAP))
1901 		uma_zfree(pvo->pvo_vaddr & PVO_MANAGED ? pmap_mpvo_zone :
1902 		    pmap_upvo_zone, pvo);
1903 	pmap_pvo_entries--;
1904 	pmap_pvo_remove_calls++;
1905 }
1906 
1907 static __inline int
1908 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1909 {
1910 	int	pteidx;
1911 
1912 	/*
1913 	 * We can find the actual pte entry without searching by grabbing
1914 	 * the PTEG index from 3 unused bits in pte_lo[11:9] and by
1915 	 * noticing the HID bit.
1916 	 */
1917 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1918 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1919 		pteidx ^= pmap_pteg_mask * 8;
1920 
1921 	return (pteidx);
1922 }
1923 
1924 static struct pvo_entry *
1925 pmap_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p)
1926 {
1927 	struct	pvo_entry *pvo;
1928 	int	ptegidx;
1929 	u_int	sr;
1930 
1931 	va &= ~ADDR_POFF;
1932 	sr = va_to_sr(pm->pm_sr, va);
1933 	ptegidx = va_to_pteg(sr, va);
1934 
1935 	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1936 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1937 			if (pteidx_p)
1938 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1939 			return (pvo);
1940 		}
1941 	}
1942 
1943 	return (NULL);
1944 }
1945 
1946 static struct pte *
1947 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1948 {
1949 	struct	pte *pt;
1950 
1951 	/*
1952 	 * If we haven't been supplied the ptegidx, calculate it.
1953 	 */
1954 	if (pteidx == -1) {
1955 		int	ptegidx;
1956 		u_int	sr;
1957 
1958 		sr = va_to_sr(pvo->pvo_pmap->pm_sr, pvo->pvo_vaddr);
1959 		ptegidx = va_to_pteg(sr, pvo->pvo_vaddr);
1960 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1961 	}
1962 
1963 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1964 
1965 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1966 		panic("pmap_pvo_to_pte: pvo %p has valid pte in pvo but no "
1967 		    "valid pte index", pvo);
1968 	}
1969 
1970 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1971 		panic("pmap_pvo_to_pte: pvo %p has valid pte index in pvo "
1972 		    "pvo but no valid pte", pvo);
1973 	}
1974 
1975 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1976 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1977 			panic("pmap_pvo_to_pte: pvo %p has valid pte in "
1978 			    "pmap_pteg_table %p but invalid in pvo", pvo, pt);
1979 		}
1980 
1981 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF))
1982 		    != 0) {
1983 			panic("pmap_pvo_to_pte: pvo %p pte does not match "
1984 			    "pte %p in pmap_pteg_table", pvo, pt);
1985 		}
1986 
1987 		return (pt);
1988 	}
1989 
1990 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1991 		panic("pmap_pvo_to_pte: pvo %p has invalid pte %p in "
1992 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1993 	}
1994 
1995 	return (NULL);
1996 }
1997 
1998 /*
1999  * XXX: THIS STUFF SHOULD BE IN pte.c?
2000  */
2001 int
2002 pmap_pte_spill(vm_offset_t addr)
2003 {
2004 	struct	pvo_entry *source_pvo, *victim_pvo;
2005 	struct	pvo_entry *pvo;
2006 	int	ptegidx, i, j;
2007 	u_int	sr;
2008 	struct	pteg *pteg;
2009 	struct	pte *pt;
2010 
2011 	pmap_pte_spills++;
2012 
2013 	sr = mfsrin(addr);
2014 	ptegidx = va_to_pteg(sr, addr);
2015 
2016 	/*
2017 	 * Have to substitute some entry.  Use the primary hash for this.
2018 	 * Use low bits of timebase as random generator.
2019 	 */
2020 	pteg = &pmap_pteg_table[ptegidx];
2021 	__asm __volatile("mftb %0" : "=r"(i));
2022 	i &= 7;
2023 	pt = &pteg->pt[i];
2024 
2025 	source_pvo = NULL;
2026 	victim_pvo = NULL;
2027 	LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2028 		/*
2029 		 * We need to find a pvo entry for this address.
2030 		 */
2031 		PMAP_PVO_CHECK(pvo);
2032 		if (source_pvo == NULL &&
2033 		    pmap_pte_match(&pvo->pvo_pte, sr, addr,
2034 		    pvo->pvo_pte.pte_hi & PTE_HID)) {
2035 			/*
2036 			 * Now found an entry to be spilled into the pteg.
2037 			 * The PTE is now valid, so we know it's active.
2038 			 */
2039 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
2040 
2041 			if (j >= 0) {
2042 				PVO_PTEGIDX_SET(pvo, j);
2043 				pmap_pte_overflow--;
2044 				PMAP_PVO_CHECK(pvo);
2045 				return (1);
2046 			}
2047 
2048 			source_pvo = pvo;
2049 
2050 			if (victim_pvo != NULL)
2051 				break;
2052 		}
2053 
2054 		/*
2055 		 * We also need the pvo entry of the victim we are replacing
2056 		 * so save the R & C bits of the PTE.
2057 		 */
2058 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
2059 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
2060 			victim_pvo = pvo;
2061 			if (source_pvo != NULL)
2062 				break;
2063 		}
2064 	}
2065 
2066 	if (source_pvo == NULL)
2067 		return (0);
2068 
2069 	if (victim_pvo == NULL) {
2070 		if ((pt->pte_hi & PTE_HID) == 0)
2071 			panic("pmap_pte_spill: victim p-pte (%p) has no pvo"
2072 			    "entry", pt);
2073 
2074 		/*
2075 		 * If this is a secondary PTE, we need to search it's primary
2076 		 * pvo bucket for the matching PVO.
2077 		 */
2078 		LIST_FOREACH(pvo, &pmap_pvo_table[ptegidx ^ pmap_pteg_mask],
2079 		    pvo_olink) {
2080 			PMAP_PVO_CHECK(pvo);
2081 			/*
2082 			 * We also need the pvo entry of the victim we are
2083 			 * replacing so save the R & C bits of the PTE.
2084 			 */
2085 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
2086 				victim_pvo = pvo;
2087 				break;
2088 			}
2089 		}
2090 
2091 		if (victim_pvo == NULL)
2092 			panic("pmap_pte_spill: victim s-pte (%p) has no pvo"
2093 			    "entry", pt);
2094 	}
2095 
2096 	/*
2097 	 * We are invalidating the TLB entry for the EA we are replacing even
2098 	 * though it's valid.  If we don't, we lose any ref/chg bit changes
2099 	 * contained in the TLB entry.
2100 	 */
2101 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
2102 
2103 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
2104 	pmap_pte_set(pt, &source_pvo->pvo_pte);
2105 
2106 	PVO_PTEGIDX_CLR(victim_pvo);
2107 	PVO_PTEGIDX_SET(source_pvo, i);
2108 	pmap_pte_replacements++;
2109 
2110 	PMAP_PVO_CHECK(victim_pvo);
2111 	PMAP_PVO_CHECK(source_pvo);
2112 
2113 	return (1);
2114 }
2115 
2116 static int
2117 pmap_pte_insert(u_int ptegidx, struct pte *pvo_pt)
2118 {
2119 	struct	pte *pt;
2120 	int	i;
2121 
2122 	/*
2123 	 * First try primary hash.
2124 	 */
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 	/*
2134 	 * Now try secondary hash.
2135 	 */
2136 	ptegidx ^= pmap_pteg_mask;
2137 	ptegidx++;
2138 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2139 		if ((pt->pte_hi & PTE_VALID) == 0) {
2140 			pvo_pt->pte_hi |= PTE_HID;
2141 			pmap_pte_set(pt, pvo_pt);
2142 			return (i);
2143 		}
2144 	}
2145 
2146 	panic("pmap_pte_insert: overflow");
2147 	return (-1);
2148 }
2149 
2150 static boolean_t
2151 pmap_query_bit(vm_page_t m, int ptebit)
2152 {
2153 	struct	pvo_entry *pvo;
2154 	struct	pte *pt;
2155 
2156 #if 0
2157 	if (pmap_attr_fetch(m) & ptebit)
2158 		return (TRUE);
2159 #endif
2160 
2161 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2162 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2163 
2164 		/*
2165 		 * See if we saved the bit off.  If so, cache it and return
2166 		 * success.
2167 		 */
2168 		if (pvo->pvo_pte.pte_lo & ptebit) {
2169 			pmap_attr_save(m, ptebit);
2170 			PMAP_PVO_CHECK(pvo);	/* sanity check */
2171 			return (TRUE);
2172 		}
2173 	}
2174 
2175 	/*
2176 	 * No luck, now go through the hard part of looking at the PTEs
2177 	 * themselves.  Sync so that any pending REF/CHG bits are flushed to
2178 	 * the PTEs.
2179 	 */
2180 	SYNC();
2181 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2182 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2183 
2184 		/*
2185 		 * See if this pvo has a valid PTE.  if so, fetch the
2186 		 * REF/CHG bits from the valid PTE.  If the appropriate
2187 		 * ptebit is set, cache it and return success.
2188 		 */
2189 		pt = pmap_pvo_to_pte(pvo, -1);
2190 		if (pt != NULL) {
2191 			pmap_pte_synch(pt, &pvo->pvo_pte);
2192 			if (pvo->pvo_pte.pte_lo & ptebit) {
2193 				pmap_attr_save(m, ptebit);
2194 				PMAP_PVO_CHECK(pvo);	/* sanity check */
2195 				return (TRUE);
2196 			}
2197 		}
2198 	}
2199 
2200 	return (FALSE);
2201 }
2202 
2203 static u_int
2204 pmap_clear_bit(vm_page_t m, int ptebit, int *origbit)
2205 {
2206 	u_int	count;
2207 	struct	pvo_entry *pvo;
2208 	struct	pte *pt;
2209 	int	rv;
2210 
2211 	/*
2212 	 * Clear the cached value.
2213 	 */
2214 	rv = pmap_attr_fetch(m);
2215 	pmap_attr_clear(m, ptebit);
2216 
2217 	/*
2218 	 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so
2219 	 * we can reset the right ones).  note that since the pvo entries and
2220 	 * list heads are accessed via BAT0 and are never placed in the page
2221 	 * table, we don't have to worry about further accesses setting the
2222 	 * REF/CHG bits.
2223 	 */
2224 	SYNC();
2225 
2226 	/*
2227 	 * For each pvo entry, clear the pvo's ptebit.  If this pvo has a
2228 	 * valid pte clear the ptebit from the valid pte.
2229 	 */
2230 	count = 0;
2231 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2232 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2233 		pt = pmap_pvo_to_pte(pvo, -1);
2234 		if (pt != NULL) {
2235 			pmap_pte_synch(pt, &pvo->pvo_pte);
2236 			if (pvo->pvo_pte.pte_lo & ptebit) {
2237 				count++;
2238 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2239 			}
2240 		}
2241 		rv |= pvo->pvo_pte.pte_lo;
2242 		pvo->pvo_pte.pte_lo &= ~ptebit;
2243 		PMAP_PVO_CHECK(pvo);	/* sanity check */
2244 	}
2245 
2246 	if (origbit != NULL) {
2247 		*origbit = rv;
2248 	}
2249 
2250 	return (count);
2251 }
2252 
2253 /*
2254  * Return true if the physical range is encompassed by the battable[idx]
2255  */
2256 static int
2257 pmap_bat_mapped(int idx, vm_offset_t pa, vm_size_t size)
2258 {
2259 	u_int prot;
2260 	u_int32_t start;
2261 	u_int32_t end;
2262 	u_int32_t bat_ble;
2263 
2264 	/*
2265 	 * Return immediately if not a valid mapping
2266 	 */
2267 	if (!battable[idx].batu & BAT_Vs)
2268 		return (EINVAL);
2269 
2270 	/*
2271 	 * The BAT entry must be cache-inhibited, guarded, and r/w
2272 	 * so it can function as an i/o page
2273 	 */
2274 	prot = battable[idx].batl & (BAT_I|BAT_G|BAT_PP_RW);
2275 	if (prot != (BAT_I|BAT_G|BAT_PP_RW))
2276 		return (EPERM);
2277 
2278 	/*
2279 	 * The address should be within the BAT range. Assume that the
2280 	 * start address in the BAT has the correct alignment (thus
2281 	 * not requiring masking)
2282 	 */
2283 	start = battable[idx].batl & BAT_PBS;
2284 	bat_ble = (battable[idx].batu & ~(BAT_EBS)) | 0x03;
2285 	end = start | (bat_ble << 15) | 0x7fff;
2286 
2287 	if ((pa < start) || ((pa + size) > end))
2288 		return (ERANGE);
2289 
2290 	return (0);
2291 }
2292 
2293 
2294 /*
2295  * Map a set of physical memory pages into the kernel virtual
2296  * address space. Return a pointer to where it is mapped. This
2297  * routine is intended to be used for mapping device memory,
2298  * NOT real memory.
2299  */
2300 void *
2301 pmap_mapdev(vm_offset_t pa, vm_size_t size)
2302 {
2303 	vm_offset_t va, tmpva, ppa, offset;
2304 	int i;
2305 
2306 	ppa = trunc_page(pa);
2307 	offset = pa & PAGE_MASK;
2308 	size = roundup(offset + size, PAGE_SIZE);
2309 
2310 	GIANT_REQUIRED;
2311 
2312 	/*
2313 	 * If the physical address lies within a valid BAT table entry,
2314 	 * return the 1:1 mapping. This currently doesn't work
2315 	 * for regions that overlap 256M BAT segments.
2316 	 */
2317 	for (i = 0; i < 16; i++) {
2318 		if (pmap_bat_mapped(i, pa, size) == 0)
2319 			return ((void *) pa);
2320 	}
2321 
2322 	va = kmem_alloc_nofault(kernel_map, size);
2323 	if (!va)
2324 		panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
2325 
2326 	for (tmpva = va; size > 0;) {
2327 		pmap_kenter(tmpva, ppa);
2328 		TLBIE(tmpva); /* XXX or should it be invalidate-all ? */
2329 		size -= PAGE_SIZE;
2330 		tmpva += PAGE_SIZE;
2331 		ppa += PAGE_SIZE;
2332 	}
2333 
2334 	return ((void *)(va + offset));
2335 }
2336 
2337 void
2338 pmap_unmapdev(vm_offset_t va, vm_size_t size)
2339 {
2340 	vm_offset_t base, offset;
2341 
2342 	/*
2343 	 * If this is outside kernel virtual space, then it's a
2344 	 * battable entry and doesn't require unmapping
2345 	 */
2346 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2347 		base = trunc_page(va);
2348 		offset = va & PAGE_MASK;
2349 		size = roundup(offset + size, PAGE_SIZE);
2350 		kmem_free(kernel_map, base, size);
2351 	}
2352 }
2353