xref: /freebsd/sys/powerpc/aim/mmu_oea.c (revision 94942af266ac119ede0ca836f9aa5a5ac0582938)
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 #include <machine/mmuvar.h>
152 
153 #include "mmu_if.h"
154 
155 #define	MOEA_DEBUG
156 
157 #define TODO	panic("%s: not implemented", __func__);
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	0x007		/* which PTEG slot */
169 #define	PVO_PTEGIDX_VALID	0x008		/* slot is valid */
170 #define	PVO_WIRED		0x010		/* PVO entry is wired */
171 #define	PVO_MANAGED		0x020		/* PVO entry is managed */
172 #define	PVO_EXECUTABLE		0x040		/* PVO entry is executable */
173 #define	PVO_BOOTSTRAP		0x080		/* PVO entry allocated during
174 						   bootstrap */
175 #define PVO_FAKE		0x100		/* fictitious phys page */
176 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
177 #define	PVO_ISEXECUTABLE(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
178 #define PVO_ISFAKE(pvo)		((pvo)->pvo_vaddr & PVO_FAKE)
179 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
180 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
181 #define	PVO_PTEGIDX_CLR(pvo)	\
182 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
183 #define	PVO_PTEGIDX_SET(pvo, i)	\
184 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
185 
186 #define	MOEA_PVO_CHECK(pvo)
187 
188 struct ofw_map {
189 	vm_offset_t	om_va;
190 	vm_size_t	om_len;
191 	vm_offset_t	om_pa;
192 	u_int		om_mode;
193 };
194 
195 /*
196  * Map of physical memory regions.
197  */
198 static struct	mem_region *regions;
199 static struct	mem_region *pregions;
200 u_int           phys_avail_count;
201 int		regions_sz, pregions_sz;
202 static struct	ofw_map *translations;
203 
204 extern struct pmap ofw_pmap;
205 
206 
207 
208 /*
209  * Lock for the pteg and pvo tables.
210  */
211 struct mtx	moea_table_mutex;
212 
213 /*
214  * PTEG data.
215  */
216 static struct	pteg *moea_pteg_table;
217 u_int		moea_pteg_count;
218 u_int		moea_pteg_mask;
219 
220 /*
221  * PVO data.
222  */
223 struct	pvo_head *moea_pvo_table;		/* pvo entries by pteg index */
224 struct	pvo_head moea_pvo_kunmanaged =
225     LIST_HEAD_INITIALIZER(moea_pvo_kunmanaged);	/* list of unmanaged pages */
226 struct	pvo_head moea_pvo_unmanaged =
227     LIST_HEAD_INITIALIZER(moea_pvo_unmanaged);	/* list of unmanaged pages */
228 
229 uma_zone_t	moea_upvo_zone;	/* zone for pvo entries for unmanaged pages */
230 uma_zone_t	moea_mpvo_zone;	/* zone for pvo entries for managed pages */
231 
232 #define	BPVO_POOL_SIZE	32768
233 static struct	pvo_entry *moea_bpvo_pool;
234 static int	moea_bpvo_pool_index = 0;
235 
236 #define	VSID_NBPW	(sizeof(u_int32_t) * 8)
237 static u_int	moea_vsid_bitmap[NPMAPS / VSID_NBPW];
238 
239 static boolean_t moea_initialized = FALSE;
240 
241 /*
242  * Statistics.
243  */
244 u_int	moea_pte_valid = 0;
245 u_int	moea_pte_overflow = 0;
246 u_int	moea_pte_replacements = 0;
247 u_int	moea_pvo_entries = 0;
248 u_int	moea_pvo_enter_calls = 0;
249 u_int	moea_pvo_remove_calls = 0;
250 u_int	moea_pte_spills = 0;
251 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_valid, CTLFLAG_RD, &moea_pte_valid,
252     0, "");
253 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_overflow, CTLFLAG_RD,
254     &moea_pte_overflow, 0, "");
255 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_replacements, CTLFLAG_RD,
256     &moea_pte_replacements, 0, "");
257 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_entries, CTLFLAG_RD, &moea_pvo_entries,
258     0, "");
259 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_enter_calls, CTLFLAG_RD,
260     &moea_pvo_enter_calls, 0, "");
261 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_remove_calls, CTLFLAG_RD,
262     &moea_pvo_remove_calls, 0, "");
263 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_spills, CTLFLAG_RD,
264     &moea_pte_spills, 0, "");
265 
266 struct	pvo_entry *moea_pvo_zeropage;
267 struct	mtx	moea_pvo_zeropage_mtx;
268 
269 vm_offset_t	moea_rkva_start = VM_MIN_KERNEL_ADDRESS;
270 u_int		moea_rkva_count = 4;
271 
272 /*
273  * Allocate physical memory for use in moea_bootstrap.
274  */
275 static vm_offset_t	moea_bootstrap_alloc(vm_size_t, u_int);
276 
277 /*
278  * PTE calls.
279  */
280 static int		moea_pte_insert(u_int, struct pte *);
281 
282 /*
283  * PVO calls.
284  */
285 static int	moea_pvo_enter(pmap_t, uma_zone_t, struct pvo_head *,
286 		    vm_offset_t, vm_offset_t, u_int, int);
287 static void	moea_pvo_remove(struct pvo_entry *, int);
288 static struct	pvo_entry *moea_pvo_find_va(pmap_t, vm_offset_t, int *);
289 static struct	pte *moea_pvo_to_pte(const struct pvo_entry *, int);
290 
291 /*
292  * Utility routines.
293  */
294 static void		moea_enter_locked(pmap_t, vm_offset_t, vm_page_t,
295 			    vm_prot_t, boolean_t);
296 static struct		pvo_entry *moea_rkva_alloc(mmu_t);
297 static void		moea_pa_map(struct pvo_entry *, vm_offset_t,
298 			    struct pte *, int *);
299 static void		moea_pa_unmap(struct pvo_entry *, struct pte *, int *);
300 static void		moea_syncicache(vm_offset_t, vm_size_t);
301 static boolean_t	moea_query_bit(vm_page_t, int);
302 static u_int		moea_clear_bit(vm_page_t, int, int *);
303 static void		moea_kremove(mmu_t, vm_offset_t);
304 static void		tlbia(void);
305 int		moea_pte_spill(vm_offset_t);
306 
307 /*
308  * Kernel MMU interface
309  */
310 void moea_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
311 void moea_clear_modify(mmu_t, vm_page_t);
312 void moea_clear_reference(mmu_t, vm_page_t);
313 void moea_copy_page(mmu_t, vm_page_t, vm_page_t);
314 void moea_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t, vm_prot_t, boolean_t);
315 void moea_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t, vm_page_t,
316     vm_prot_t);
317 void moea_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t, vm_prot_t);
318 vm_paddr_t moea_extract(mmu_t, pmap_t, vm_offset_t);
319 vm_page_t moea_extract_and_hold(mmu_t, pmap_t, vm_offset_t, vm_prot_t);
320 void moea_init(mmu_t);
321 boolean_t moea_is_modified(mmu_t, vm_page_t);
322 boolean_t moea_ts_referenced(mmu_t, vm_page_t);
323 vm_offset_t moea_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t, int);
324 boolean_t moea_page_exists_quick(mmu_t, pmap_t, vm_page_t);
325 void moea_pinit(mmu_t, pmap_t);
326 void moea_pinit0(mmu_t, pmap_t);
327 void moea_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t, vm_prot_t);
328 void moea_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
329 void moea_qremove(mmu_t, vm_offset_t, int);
330 void moea_release(mmu_t, pmap_t);
331 void moea_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
332 void moea_remove_all(mmu_t, vm_page_t);
333 void moea_remove_write(mmu_t, vm_page_t);
334 void moea_zero_page(mmu_t, vm_page_t);
335 void moea_zero_page_area(mmu_t, vm_page_t, int, int);
336 void moea_zero_page_idle(mmu_t, vm_page_t);
337 void moea_activate(mmu_t, struct thread *);
338 void moea_deactivate(mmu_t, struct thread *);
339 void moea_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
340 void *moea_mapdev(mmu_t, vm_offset_t, vm_size_t);
341 void moea_unmapdev(mmu_t, vm_offset_t, vm_size_t);
342 vm_offset_t moea_kextract(mmu_t, vm_offset_t);
343 void moea_kenter(mmu_t, vm_offset_t, vm_offset_t);
344 boolean_t moea_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
345 boolean_t moea_page_executable(mmu_t, vm_page_t);
346 
347 static mmu_method_t moea_methods[] = {
348 	MMUMETHOD(mmu_change_wiring,	moea_change_wiring),
349 	MMUMETHOD(mmu_clear_modify,	moea_clear_modify),
350 	MMUMETHOD(mmu_clear_reference,	moea_clear_reference),
351 	MMUMETHOD(mmu_copy_page,	moea_copy_page),
352 	MMUMETHOD(mmu_enter,		moea_enter),
353 	MMUMETHOD(mmu_enter_object,	moea_enter_object),
354 	MMUMETHOD(mmu_enter_quick,	moea_enter_quick),
355 	MMUMETHOD(mmu_extract,		moea_extract),
356 	MMUMETHOD(mmu_extract_and_hold,	moea_extract_and_hold),
357 	MMUMETHOD(mmu_init,		moea_init),
358 	MMUMETHOD(mmu_is_modified,	moea_is_modified),
359 	MMUMETHOD(mmu_ts_referenced,	moea_ts_referenced),
360 	MMUMETHOD(mmu_map,     		moea_map),
361 	MMUMETHOD(mmu_page_exists_quick,moea_page_exists_quick),
362 	MMUMETHOD(mmu_pinit,		moea_pinit),
363 	MMUMETHOD(mmu_pinit0,		moea_pinit0),
364 	MMUMETHOD(mmu_protect,		moea_protect),
365 	MMUMETHOD(mmu_qenter,		moea_qenter),
366 	MMUMETHOD(mmu_qremove,		moea_qremove),
367 	MMUMETHOD(mmu_release,		moea_release),
368 	MMUMETHOD(mmu_remove,		moea_remove),
369 	MMUMETHOD(mmu_remove_all,      	moea_remove_all),
370 	MMUMETHOD(mmu_remove_write,	moea_remove_write),
371 	MMUMETHOD(mmu_zero_page,       	moea_zero_page),
372 	MMUMETHOD(mmu_zero_page_area,	moea_zero_page_area),
373 	MMUMETHOD(mmu_zero_page_idle,	moea_zero_page_idle),
374 	MMUMETHOD(mmu_activate,		moea_activate),
375 	MMUMETHOD(mmu_deactivate,      	moea_deactivate),
376 
377 	/* Internal interfaces */
378 	MMUMETHOD(mmu_bootstrap,       	moea_bootstrap),
379 	MMUMETHOD(mmu_mapdev,		moea_mapdev),
380 	MMUMETHOD(mmu_unmapdev,		moea_unmapdev),
381 	MMUMETHOD(mmu_kextract,		moea_kextract),
382 	MMUMETHOD(mmu_kenter,		moea_kenter),
383 	MMUMETHOD(mmu_dev_direct_mapped,moea_dev_direct_mapped),
384 	MMUMETHOD(mmu_page_executable,	moea_page_executable),
385 
386 	{ 0, 0 }
387 };
388 
389 static mmu_def_t oea_mmu = {
390 	MMU_TYPE_OEA,
391 	moea_methods,
392 	0
393 };
394 MMU_DEF(oea_mmu);
395 
396 
397 static __inline int
398 va_to_sr(u_int *sr, vm_offset_t va)
399 {
400 	return (sr[(uintptr_t)va >> ADDR_SR_SHFT]);
401 }
402 
403 static __inline u_int
404 va_to_pteg(u_int sr, vm_offset_t addr)
405 {
406 	u_int hash;
407 
408 	hash = (sr & SR_VSID_MASK) ^ (((u_int)addr & ADDR_PIDX) >>
409 	    ADDR_PIDX_SHFT);
410 	return (hash & moea_pteg_mask);
411 }
412 
413 static __inline struct pvo_head *
414 pa_to_pvoh(vm_offset_t pa, vm_page_t *pg_p)
415 {
416 	struct	vm_page *pg;
417 
418 	pg = PHYS_TO_VM_PAGE(pa);
419 
420 	if (pg_p != NULL)
421 		*pg_p = pg;
422 
423 	if (pg == NULL)
424 		return (&moea_pvo_unmanaged);
425 
426 	return (&pg->md.mdpg_pvoh);
427 }
428 
429 static __inline struct pvo_head *
430 vm_page_to_pvoh(vm_page_t m)
431 {
432 
433 	return (&m->md.mdpg_pvoh);
434 }
435 
436 static __inline void
437 moea_attr_clear(vm_page_t m, int ptebit)
438 {
439 
440 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
441 	m->md.mdpg_attrs &= ~ptebit;
442 }
443 
444 static __inline int
445 moea_attr_fetch(vm_page_t m)
446 {
447 
448 	return (m->md.mdpg_attrs);
449 }
450 
451 static __inline void
452 moea_attr_save(vm_page_t m, int ptebit)
453 {
454 
455 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
456 	m->md.mdpg_attrs |= ptebit;
457 }
458 
459 static __inline int
460 moea_pte_compare(const struct pte *pt, const struct pte *pvo_pt)
461 {
462 	if (pt->pte_hi == pvo_pt->pte_hi)
463 		return (1);
464 
465 	return (0);
466 }
467 
468 static __inline int
469 moea_pte_match(struct pte *pt, u_int sr, vm_offset_t va, int which)
470 {
471 	return (pt->pte_hi & ~PTE_VALID) ==
472 	    (((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
473 	    ((va >> ADDR_API_SHFT) & PTE_API) | which);
474 }
475 
476 static __inline void
477 moea_pte_create(struct pte *pt, u_int sr, vm_offset_t va, u_int pte_lo)
478 {
479 
480 	mtx_assert(&moea_table_mutex, MA_OWNED);
481 
482 	/*
483 	 * Construct a PTE.  Default to IMB initially.  Valid bit only gets
484 	 * set when the real pte is set in memory.
485 	 *
486 	 * Note: Don't set the valid bit for correct operation of tlb update.
487 	 */
488 	pt->pte_hi = ((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
489 	    (((va & ADDR_PIDX) >> ADDR_API_SHFT) & PTE_API);
490 	pt->pte_lo = pte_lo;
491 }
492 
493 static __inline void
494 moea_pte_synch(struct pte *pt, struct pte *pvo_pt)
495 {
496 
497 	mtx_assert(&moea_table_mutex, MA_OWNED);
498 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF | PTE_CHG);
499 }
500 
501 static __inline void
502 moea_pte_clear(struct pte *pt, vm_offset_t va, int ptebit)
503 {
504 
505 	mtx_assert(&moea_table_mutex, MA_OWNED);
506 
507 	/*
508 	 * As shown in Section 7.6.3.2.3
509 	 */
510 	pt->pte_lo &= ~ptebit;
511 	TLBIE(va);
512 	EIEIO();
513 	TLBSYNC();
514 	SYNC();
515 }
516 
517 static __inline void
518 moea_pte_set(struct pte *pt, struct pte *pvo_pt)
519 {
520 
521 	mtx_assert(&moea_table_mutex, MA_OWNED);
522 	pvo_pt->pte_hi |= PTE_VALID;
523 
524 	/*
525 	 * Update the PTE as defined in section 7.6.3.1.
526 	 * Note that the REF/CHG bits are from pvo_pt and thus should havce
527 	 * been saved so this routine can restore them (if desired).
528 	 */
529 	pt->pte_lo = pvo_pt->pte_lo;
530 	EIEIO();
531 	pt->pte_hi = pvo_pt->pte_hi;
532 	SYNC();
533 	moea_pte_valid++;
534 }
535 
536 static __inline void
537 moea_pte_unset(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
538 {
539 
540 	mtx_assert(&moea_table_mutex, MA_OWNED);
541 	pvo_pt->pte_hi &= ~PTE_VALID;
542 
543 	/*
544 	 * Force the reg & chg bits back into the PTEs.
545 	 */
546 	SYNC();
547 
548 	/*
549 	 * Invalidate the pte.
550 	 */
551 	pt->pte_hi &= ~PTE_VALID;
552 
553 	SYNC();
554 	TLBIE(va);
555 	EIEIO();
556 	TLBSYNC();
557 	SYNC();
558 
559 	/*
560 	 * Save the reg & chg bits.
561 	 */
562 	moea_pte_synch(pt, pvo_pt);
563 	moea_pte_valid--;
564 }
565 
566 static __inline void
567 moea_pte_change(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
568 {
569 
570 	/*
571 	 * Invalidate the PTE
572 	 */
573 	moea_pte_unset(pt, pvo_pt, va);
574 	moea_pte_set(pt, pvo_pt);
575 }
576 
577 /*
578  * Quick sort callout for comparing memory regions.
579  */
580 static int	mr_cmp(const void *a, const void *b);
581 static int	om_cmp(const void *a, const void *b);
582 
583 static int
584 mr_cmp(const void *a, const void *b)
585 {
586 	const struct	mem_region *regiona;
587 	const struct	mem_region *regionb;
588 
589 	regiona = a;
590 	regionb = b;
591 	if (regiona->mr_start < regionb->mr_start)
592 		return (-1);
593 	else if (regiona->mr_start > regionb->mr_start)
594 		return (1);
595 	else
596 		return (0);
597 }
598 
599 static int
600 om_cmp(const void *a, const void *b)
601 {
602 	const struct	ofw_map *mapa;
603 	const struct	ofw_map *mapb;
604 
605 	mapa = a;
606 	mapb = b;
607 	if (mapa->om_pa < mapb->om_pa)
608 		return (-1);
609 	else if (mapa->om_pa > mapb->om_pa)
610 		return (1);
611 	else
612 		return (0);
613 }
614 
615 void
616 moea_bootstrap(mmu_t mmup, vm_offset_t kernelstart, vm_offset_t kernelend)
617 {
618 	ihandle_t	mmui;
619 	phandle_t	chosen, mmu;
620 	int		sz;
621 	int		i, j;
622 	int		ofw_mappings;
623 	vm_size_t	size, physsz, hwphyssz;
624 	vm_offset_t	pa, va, off;
625 	u_int		batl, batu;
626 
627         /*
628          * Set up BAT0 to map the lowest 256 MB area
629          */
630         battable[0x0].batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
631         battable[0x0].batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
632 
633         /*
634          * Map PCI memory space.
635          */
636         battable[0x8].batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
637         battable[0x8].batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
638 
639         battable[0x9].batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
640         battable[0x9].batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
641 
642         battable[0xa].batl = BATL(0xa0000000, BAT_I|BAT_G, BAT_PP_RW);
643         battable[0xa].batu = BATU(0xa0000000, BAT_BL_256M, BAT_Vs);
644 
645         battable[0xb].batl = BATL(0xb0000000, BAT_I|BAT_G, BAT_PP_RW);
646         battable[0xb].batu = BATU(0xb0000000, BAT_BL_256M, BAT_Vs);
647 
648         /*
649          * Map obio devices.
650          */
651         battable[0xf].batl = BATL(0xf0000000, BAT_I|BAT_G, BAT_PP_RW);
652         battable[0xf].batu = BATU(0xf0000000, BAT_BL_256M, BAT_Vs);
653 
654 	/*
655 	 * Use an IBAT and a DBAT to map the bottom segment of memory
656 	 * where we are.
657 	 */
658 	batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
659 	batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
660 	__asm (".balign 32; \n"
661 	       "mtibatu 0,%0; mtibatl 0,%1; isync; \n"
662 	       "mtdbatu 0,%0; mtdbatl 0,%1; isync"
663 	    :: "r"(batu), "r"(batl));
664 
665 #if 0
666 	/* map frame buffer */
667 	batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
668 	batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
669 	__asm ("mtdbatu 1,%0; mtdbatl 1,%1; isync"
670 	    :: "r"(batu), "r"(batl));
671 #endif
672 
673 #if 1
674 	/* map pci space */
675 	batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
676 	batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
677 	__asm ("mtdbatu 1,%0; mtdbatl 1,%1; isync"
678 	    :: "r"(batu), "r"(batl));
679 #endif
680 
681 	/*
682 	 * Set the start and end of kva.
683 	 */
684 	virtual_avail = VM_MIN_KERNEL_ADDRESS;
685 	virtual_end = VM_MAX_KERNEL_ADDRESS;
686 
687 	mem_regions(&pregions, &pregions_sz, &regions, &regions_sz);
688 	CTR0(KTR_PMAP, "moea_bootstrap: physical memory");
689 
690 	qsort(pregions, pregions_sz, sizeof(*pregions), mr_cmp);
691 	for (i = 0; i < pregions_sz; i++) {
692 		vm_offset_t pa;
693 		vm_offset_t end;
694 
695 		CTR3(KTR_PMAP, "physregion: %#x - %#x (%#x)",
696 			pregions[i].mr_start,
697 			pregions[i].mr_start + pregions[i].mr_size,
698 			pregions[i].mr_size);
699 		/*
700 		 * Install entries into the BAT table to allow all
701 		 * of physmem to be convered by on-demand BAT entries.
702 		 * The loop will sometimes set the same battable element
703 		 * twice, but that's fine since they won't be used for
704 		 * a while yet.
705 		 */
706 		pa = pregions[i].mr_start & 0xf0000000;
707 		end = pregions[i].mr_start + pregions[i].mr_size;
708 		do {
709                         u_int n = pa >> ADDR_SR_SHFT;
710 
711 			battable[n].batl = BATL(pa, BAT_M, BAT_PP_RW);
712 			battable[n].batu = BATU(pa, BAT_BL_256M, BAT_Vs);
713 			pa += SEGMENT_LENGTH;
714 		} while (pa < end);
715 	}
716 
717 	if (sizeof(phys_avail)/sizeof(phys_avail[0]) < regions_sz)
718 		panic("moea_bootstrap: phys_avail too small");
719 	qsort(regions, regions_sz, sizeof(*regions), mr_cmp);
720 	phys_avail_count = 0;
721 	physsz = 0;
722 	hwphyssz = 0;
723 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
724 	for (i = 0, j = 0; i < regions_sz; i++, j += 2) {
725 		CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start,
726 		    regions[i].mr_start + regions[i].mr_size,
727 		    regions[i].mr_size);
728 		if (hwphyssz != 0 &&
729 		    (physsz + regions[i].mr_size) >= hwphyssz) {
730 			if (physsz < hwphyssz) {
731 				phys_avail[j] = regions[i].mr_start;
732 				phys_avail[j + 1] = regions[i].mr_start +
733 				    hwphyssz - physsz;
734 				physsz = hwphyssz;
735 				phys_avail_count++;
736 			}
737 			break;
738 		}
739 		phys_avail[j] = regions[i].mr_start;
740 		phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size;
741 		phys_avail_count++;
742 		physsz += regions[i].mr_size;
743 	}
744 	physmem = btoc(physsz);
745 
746 	/*
747 	 * Allocate PTEG table.
748 	 */
749 #ifdef PTEGCOUNT
750 	moea_pteg_count = PTEGCOUNT;
751 #else
752 	moea_pteg_count = 0x1000;
753 
754 	while (moea_pteg_count < physmem)
755 		moea_pteg_count <<= 1;
756 
757 	moea_pteg_count >>= 1;
758 #endif /* PTEGCOUNT */
759 
760 	size = moea_pteg_count * sizeof(struct pteg);
761 	CTR2(KTR_PMAP, "moea_bootstrap: %d PTEGs, %d bytes", moea_pteg_count,
762 	    size);
763 	moea_pteg_table = (struct pteg *)moea_bootstrap_alloc(size, size);
764 	CTR1(KTR_PMAP, "moea_bootstrap: PTEG table at %p", moea_pteg_table);
765 	bzero((void *)moea_pteg_table, moea_pteg_count * sizeof(struct pteg));
766 	moea_pteg_mask = moea_pteg_count - 1;
767 
768 	/*
769 	 * Allocate pv/overflow lists.
770 	 */
771 	size = sizeof(struct pvo_head) * moea_pteg_count;
772 	moea_pvo_table = (struct pvo_head *)moea_bootstrap_alloc(size,
773 	    PAGE_SIZE);
774 	CTR1(KTR_PMAP, "moea_bootstrap: PVO table at %p", moea_pvo_table);
775 	for (i = 0; i < moea_pteg_count; i++)
776 		LIST_INIT(&moea_pvo_table[i]);
777 
778 	/*
779 	 * Initialize the lock that synchronizes access to the pteg and pvo
780 	 * tables.
781 	 */
782 	mtx_init(&moea_table_mutex, "pmap table", NULL, MTX_DEF |
783 	    MTX_RECURSE);
784 
785 	/*
786 	 * Allocate the message buffer.
787 	 */
788 	msgbuf_phys = moea_bootstrap_alloc(MSGBUF_SIZE, 0);
789 
790 	/*
791 	 * Initialise the unmanaged pvo pool.
792 	 */
793 	moea_bpvo_pool = (struct pvo_entry *)moea_bootstrap_alloc(
794 		BPVO_POOL_SIZE*sizeof(struct pvo_entry), 0);
795 	moea_bpvo_pool_index = 0;
796 
797 	/*
798 	 * Make sure kernel vsid is allocated as well as VSID 0.
799 	 */
800 	moea_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW]
801 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
802 	moea_vsid_bitmap[0] |= 1;
803 
804 	/*
805 	 * Set up the Open Firmware pmap and add it's mappings.
806 	 */
807 	moea_pinit(mmup, &ofw_pmap);
808 	ofw_pmap.pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
809 	ofw_pmap.pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT;
810 	if ((chosen = OF_finddevice("/chosen")) == -1)
811 		panic("moea_bootstrap: can't find /chosen");
812 	OF_getprop(chosen, "mmu", &mmui, 4);
813 	if ((mmu = OF_instance_to_package(mmui)) == -1)
814 		panic("moea_bootstrap: can't get mmu package");
815 	if ((sz = OF_getproplen(mmu, "translations")) == -1)
816 		panic("moea_bootstrap: can't get ofw translation count");
817 	translations = NULL;
818 	for (i = 0; phys_avail[i] != 0; i += 2) {
819 		if (phys_avail[i + 1] >= sz) {
820 			translations = (struct ofw_map *)phys_avail[i];
821 			break;
822 		}
823 	}
824 	if (translations == NULL)
825 		panic("moea_bootstrap: no space to copy translations");
826 	bzero(translations, sz);
827 	if (OF_getprop(mmu, "translations", translations, sz) == -1)
828 		panic("moea_bootstrap: can't get ofw translations");
829 	CTR0(KTR_PMAP, "moea_bootstrap: translations");
830 	sz /= sizeof(*translations);
831 	qsort(translations, sz, sizeof (*translations), om_cmp);
832 	for (i = 0, ofw_mappings = 0; i < sz; i++) {
833 		CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x",
834 		    translations[i].om_pa, translations[i].om_va,
835 		    translations[i].om_len);
836 
837 		/*
838 		 * If the mapping is 1:1, let the RAM and device on-demand
839 		 * BAT tables take care of the translation.
840 		 */
841 		if (translations[i].om_va == translations[i].om_pa)
842 			continue;
843 
844 		/* Enter the pages */
845 		for (off = 0; off < translations[i].om_len; off += PAGE_SIZE) {
846 			struct	vm_page m;
847 
848 			m.phys_addr = translations[i].om_pa + off;
849 			PMAP_LOCK(&ofw_pmap);
850 			moea_enter_locked(&ofw_pmap,
851 				   translations[i].om_va + off, &m,
852 				   VM_PROT_ALL, 1);
853 			PMAP_UNLOCK(&ofw_pmap);
854 			ofw_mappings++;
855 		}
856 	}
857 #ifdef SMP
858 	TLBSYNC();
859 #endif
860 
861 	/*
862 	 * Initialize the kernel pmap (which is statically allocated).
863 	 */
864 	PMAP_LOCK_INIT(kernel_pmap);
865 	for (i = 0; i < 16; i++) {
866 		kernel_pmap->pm_sr[i] = EMPTY_SEGMENT;
867 	}
868 	kernel_pmap->pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
869 	kernel_pmap->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT;
870 	kernel_pmap->pm_active = ~0;
871 
872 	/*
873 	 * Allocate a kernel stack with a guard page for thread0 and map it
874 	 * into the kernel page map.
875 	 */
876 	pa = moea_bootstrap_alloc(KSTACK_PAGES * PAGE_SIZE, 0);
877 	kstack0_phys = pa;
878 	kstack0 = virtual_avail + (KSTACK_GUARD_PAGES * PAGE_SIZE);
879 	CTR2(KTR_PMAP, "moea_bootstrap: kstack0 at %#x (%#x)", kstack0_phys,
880 	    kstack0);
881 	virtual_avail += (KSTACK_PAGES + KSTACK_GUARD_PAGES) * PAGE_SIZE;
882 	for (i = 0; i < KSTACK_PAGES; i++) {
883 		pa = kstack0_phys + i * PAGE_SIZE;
884 		va = kstack0 + i * PAGE_SIZE;
885 		moea_kenter(mmup, va, pa);
886 		TLBIE(va);
887 	}
888 
889 	/*
890 	 * Calculate the last available physical address.
891 	 */
892 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
893 		;
894 	Maxmem = powerpc_btop(phys_avail[i + 1]);
895 
896 	/*
897 	 * Allocate virtual address space for the message buffer.
898 	 */
899 	msgbufp = (struct msgbuf *)virtual_avail;
900 	virtual_avail += round_page(MSGBUF_SIZE);
901 
902 	/*
903 	 * Initialize hardware.
904 	 */
905 	for (i = 0; i < 16; i++) {
906 		mtsrin(i << ADDR_SR_SHFT, EMPTY_SEGMENT);
907 	}
908 	__asm __volatile ("mtsr %0,%1"
909 	    :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
910 	__asm __volatile ("mtsr %0,%1"
911 	    :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
912 	__asm __volatile ("sync; mtsdr1 %0; isync"
913 	    :: "r"((u_int)moea_pteg_table | (moea_pteg_mask >> 10)));
914 	tlbia();
915 
916 	pmap_bootstrapped++;
917 }
918 
919 /*
920  * Activate a user pmap.  The pmap must be activated before it's address
921  * space can be accessed in any way.
922  */
923 void
924 moea_activate(mmu_t mmu, struct thread *td)
925 {
926 	pmap_t	pm, pmr;
927 
928 	/*
929 	 * Load all the data we need up front to encourage the compiler to
930 	 * not issue any loads while we have interrupts disabled below.
931 	 */
932 	pm = &td->td_proc->p_vmspace->vm_pmap;
933 
934 	if ((pmr = (pmap_t)moea_kextract(mmu, (vm_offset_t)pm)) == NULL)
935 		pmr = pm;
936 
937 	pm->pm_active |= PCPU_GET(cpumask);
938 	PCPU_SET(curpmap, pmr);
939 }
940 
941 void
942 moea_deactivate(mmu_t mmu, struct thread *td)
943 {
944 	pmap_t	pm;
945 
946 	pm = &td->td_proc->p_vmspace->vm_pmap;
947 	pm->pm_active &= ~(PCPU_GET(cpumask));
948 	PCPU_SET(curpmap, NULL);
949 }
950 
951 void
952 moea_change_wiring(mmu_t mmu, pmap_t pm, vm_offset_t va, boolean_t wired)
953 {
954 	struct	pvo_entry *pvo;
955 
956 	PMAP_LOCK(pm);
957 	pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
958 
959 	if (pvo != NULL) {
960 		if (wired) {
961 			if ((pvo->pvo_vaddr & PVO_WIRED) == 0)
962 				pm->pm_stats.wired_count++;
963 			pvo->pvo_vaddr |= PVO_WIRED;
964 		} else {
965 			if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
966 				pm->pm_stats.wired_count--;
967 			pvo->pvo_vaddr &= ~PVO_WIRED;
968 		}
969 	}
970 	PMAP_UNLOCK(pm);
971 }
972 
973 void
974 moea_copy_page(mmu_t mmu, vm_page_t msrc, vm_page_t mdst)
975 {
976 	vm_offset_t	dst;
977 	vm_offset_t	src;
978 
979 	dst = VM_PAGE_TO_PHYS(mdst);
980 	src = VM_PAGE_TO_PHYS(msrc);
981 
982 	kcopy((void *)src, (void *)dst, PAGE_SIZE);
983 }
984 
985 /*
986  * Zero a page of physical memory by temporarily mapping it into the tlb.
987  */
988 void
989 moea_zero_page(mmu_t mmu, vm_page_t m)
990 {
991 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
992 	caddr_t va;
993 
994 	if (pa < SEGMENT_LENGTH) {
995 		va = (caddr_t) pa;
996 	} else if (moea_initialized) {
997 		if (moea_pvo_zeropage == NULL) {
998 			moea_pvo_zeropage = moea_rkva_alloc(mmu);
999 			mtx_init(&moea_pvo_zeropage_mtx, "pvo zero page",
1000 			    NULL, MTX_DEF);
1001 		}
1002 		mtx_lock(&moea_pvo_zeropage_mtx);
1003 		moea_pa_map(moea_pvo_zeropage, pa, NULL, NULL);
1004 		va = (caddr_t)PVO_VADDR(moea_pvo_zeropage);
1005 	} else {
1006 		panic("moea_zero_page: can't zero pa %#x", pa);
1007 	}
1008 
1009 	bzero(va, PAGE_SIZE);
1010 
1011 	if (pa >= SEGMENT_LENGTH) {
1012 		moea_pa_unmap(moea_pvo_zeropage, NULL, NULL);
1013 		mtx_unlock(&moea_pvo_zeropage_mtx);
1014 	}
1015 }
1016 
1017 void
1018 moea_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
1019 {
1020 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1021 	caddr_t va;
1022 
1023 	if (pa < SEGMENT_LENGTH) {
1024 		va = (caddr_t) pa;
1025 	} else if (moea_initialized) {
1026 		if (moea_pvo_zeropage == NULL) {
1027 			moea_pvo_zeropage = moea_rkva_alloc(mmu);
1028 			mtx_init(&moea_pvo_zeropage_mtx, "pvo zero page",
1029 			    NULL, MTX_DEF);
1030 		}
1031 		mtx_lock(&moea_pvo_zeropage_mtx);
1032 		moea_pa_map(moea_pvo_zeropage, pa, NULL, NULL);
1033 		va = (caddr_t)PVO_VADDR(moea_pvo_zeropage);
1034 	} else {
1035 		panic("moea_zero_page: can't zero pa %#x", pa);
1036 	}
1037 
1038 	bzero(va + off, size);
1039 
1040 	if (pa >= SEGMENT_LENGTH) {
1041 		moea_pa_unmap(moea_pvo_zeropage, NULL, NULL);
1042 		mtx_unlock(&moea_pvo_zeropage_mtx);
1043 	}
1044 }
1045 
1046 void
1047 moea_zero_page_idle(mmu_t mmu, vm_page_t m)
1048 {
1049 
1050 	moea_zero_page(mmu, m);
1051 }
1052 
1053 /*
1054  * Map the given physical page at the specified virtual address in the
1055  * target pmap with the protection requested.  If specified the page
1056  * will be wired down.
1057  */
1058 void
1059 moea_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1060 	   boolean_t wired)
1061 {
1062 
1063 	vm_page_lock_queues();
1064 	PMAP_LOCK(pmap);
1065 	moea_enter_locked(pmap, va, m, prot, wired);
1066 	vm_page_unlock_queues();
1067 	PMAP_UNLOCK(pmap);
1068 }
1069 
1070 /*
1071  * Map the given physical page at the specified virtual address in the
1072  * target pmap with the protection requested.  If specified the page
1073  * will be wired down.
1074  *
1075  * The page queues and pmap must be locked.
1076  */
1077 static void
1078 moea_enter_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1079     boolean_t wired)
1080 {
1081 	struct		pvo_head *pvo_head;
1082 	uma_zone_t	zone;
1083 	vm_page_t	pg;
1084 	u_int		pte_lo, pvo_flags, was_exec, i;
1085 	int		error;
1086 
1087 	if (!moea_initialized) {
1088 		pvo_head = &moea_pvo_kunmanaged;
1089 		zone = moea_upvo_zone;
1090 		pvo_flags = 0;
1091 		pg = NULL;
1092 		was_exec = PTE_EXEC;
1093 	} else {
1094 		pvo_head = vm_page_to_pvoh(m);
1095 		pg = m;
1096 		zone = moea_mpvo_zone;
1097 		pvo_flags = PVO_MANAGED;
1098 		was_exec = 0;
1099 	}
1100 	if (pmap_bootstrapped)
1101 		mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1102 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1103 
1104 	/* XXX change the pvo head for fake pages */
1105 	if ((m->flags & PG_FICTITIOUS) == PG_FICTITIOUS)
1106 		pvo_head = &moea_pvo_kunmanaged;
1107 
1108 	/*
1109 	 * If this is a managed page, and it's the first reference to the page,
1110 	 * clear the execness of the page.  Otherwise fetch the execness.
1111 	 */
1112 	if ((pg != NULL) && ((m->flags & PG_FICTITIOUS) == 0)) {
1113 		if (LIST_EMPTY(pvo_head)) {
1114 			moea_attr_clear(pg, PTE_EXEC);
1115 		} else {
1116 			was_exec = moea_attr_fetch(pg) & PTE_EXEC;
1117 		}
1118 	}
1119 
1120 	/*
1121 	 * Assume the page is cache inhibited and access is guarded unless
1122 	 * it's in our available memory array.
1123 	 */
1124 	pte_lo = PTE_I | PTE_G;
1125 	for (i = 0; i < pregions_sz; i++) {
1126 		if ((VM_PAGE_TO_PHYS(m) >= pregions[i].mr_start) &&
1127 		    (VM_PAGE_TO_PHYS(m) <
1128 			(pregions[i].mr_start + pregions[i].mr_size))) {
1129 			pte_lo &= ~(PTE_I | PTE_G);
1130 			break;
1131 		}
1132 	}
1133 
1134 	if (prot & VM_PROT_WRITE) {
1135 		pte_lo |= PTE_BW;
1136 		if (pmap_bootstrapped)
1137 			vm_page_flag_set(m, PG_WRITEABLE);
1138 	} else
1139 		pte_lo |= PTE_BR;
1140 
1141 	if (prot & VM_PROT_EXECUTE)
1142 		pvo_flags |= PVO_EXECUTABLE;
1143 
1144 	if (wired)
1145 		pvo_flags |= PVO_WIRED;
1146 
1147 	if ((m->flags & PG_FICTITIOUS) != 0)
1148 		pvo_flags |= PVO_FAKE;
1149 
1150 	error = moea_pvo_enter(pmap, zone, pvo_head, va, VM_PAGE_TO_PHYS(m),
1151 	    pte_lo, pvo_flags);
1152 
1153 	/*
1154 	 * Flush the real page from the instruction cache if this page is
1155 	 * mapped executable and cacheable and was not previously mapped (or
1156 	 * was not mapped executable).
1157 	 */
1158 	if (error == 0 && (pvo_flags & PVO_EXECUTABLE) &&
1159 	    (pte_lo & PTE_I) == 0 && was_exec == 0) {
1160 		/*
1161 		 * Flush the real memory from the cache.
1162 		 */
1163 		moea_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1164 		if (pg != NULL)
1165 			moea_attr_save(pg, PTE_EXEC);
1166 	}
1167 
1168 	/* XXX syncicache always until problems are sorted */
1169 	moea_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1170 }
1171 
1172 /*
1173  * Maps a sequence of resident pages belonging to the same object.
1174  * The sequence begins with the given page m_start.  This page is
1175  * mapped at the given virtual address start.  Each subsequent page is
1176  * mapped at a virtual address that is offset from start by the same
1177  * amount as the page is offset from m_start within the object.  The
1178  * last page in the sequence is the page with the largest offset from
1179  * m_start that can be mapped at a virtual address less than the given
1180  * virtual address end.  Not every virtual page between start and end
1181  * is mapped; only those for which a resident page exists with the
1182  * corresponding offset from m_start are mapped.
1183  */
1184 void
1185 moea_enter_object(mmu_t mmu, pmap_t pm, vm_offset_t start, vm_offset_t end,
1186     vm_page_t m_start, vm_prot_t prot)
1187 {
1188 	vm_page_t m;
1189 	vm_pindex_t diff, psize;
1190 
1191 	psize = atop(end - start);
1192 	m = m_start;
1193 	PMAP_LOCK(pm);
1194 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1195 		moea_enter_locked(pm, start + ptoa(diff), m, prot &
1196 		    (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1197 		m = TAILQ_NEXT(m, listq);
1198 	}
1199 	PMAP_UNLOCK(pm);
1200 }
1201 
1202 void
1203 moea_enter_quick(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_page_t m,
1204     vm_prot_t prot)
1205 {
1206 
1207 	PMAP_LOCK(pm);
1208 	moea_enter_locked(pm, va, m, prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1209 	    FALSE);
1210 	PMAP_UNLOCK(pm);
1211 
1212 }
1213 
1214 vm_paddr_t
1215 moea_extract(mmu_t mmu, pmap_t pm, vm_offset_t va)
1216 {
1217 	struct	pvo_entry *pvo;
1218 	vm_paddr_t pa;
1219 
1220 	PMAP_LOCK(pm);
1221 	pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1222 	if (pvo == NULL)
1223 		pa = 0;
1224 	else
1225 		pa = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF);
1226 	PMAP_UNLOCK(pm);
1227 	return (pa);
1228 }
1229 
1230 /*
1231  * Atomically extract and hold the physical page with the given
1232  * pmap and virtual address pair if that mapping permits the given
1233  * protection.
1234  */
1235 vm_page_t
1236 moea_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_prot_t prot)
1237 {
1238 	struct	pvo_entry *pvo;
1239 	vm_page_t m;
1240 
1241 	m = NULL;
1242 	vm_page_lock_queues();
1243 	PMAP_LOCK(pmap);
1244 	pvo = moea_pvo_find_va(pmap, va & ~ADDR_POFF, NULL);
1245 	if (pvo != NULL && (pvo->pvo_pte.pte_hi & PTE_VALID) &&
1246 	    ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_RW ||
1247 	     (prot & VM_PROT_WRITE) == 0)) {
1248 		m = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte_lo & PTE_RPGN);
1249 		vm_page_hold(m);
1250 	}
1251 	vm_page_unlock_queues();
1252 	PMAP_UNLOCK(pmap);
1253 	return (m);
1254 }
1255 
1256 void
1257 moea_init(mmu_t mmu)
1258 {
1259 
1260 	CTR0(KTR_PMAP, "moea_init");
1261 
1262 	moea_upvo_zone = uma_zcreate("UPVO entry", sizeof (struct pvo_entry),
1263 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1264 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1265 	moea_mpvo_zone = uma_zcreate("MPVO entry", sizeof(struct pvo_entry),
1266 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1267 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1268 	moea_initialized = TRUE;
1269 }
1270 
1271 boolean_t
1272 moea_is_modified(mmu_t mmu, vm_page_t m)
1273 {
1274 
1275 	if ((m->flags & (PG_FICTITIOUS |PG_UNMANAGED)) != 0)
1276 		return (FALSE);
1277 
1278 	return (moea_query_bit(m, PTE_CHG));
1279 }
1280 
1281 void
1282 moea_clear_reference(mmu_t mmu, vm_page_t m)
1283 {
1284 
1285 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1286 		return;
1287 	moea_clear_bit(m, PTE_REF, NULL);
1288 }
1289 
1290 void
1291 moea_clear_modify(mmu_t mmu, vm_page_t m)
1292 {
1293 
1294 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1295 		return;
1296 	moea_clear_bit(m, PTE_CHG, NULL);
1297 }
1298 
1299 /*
1300  * Clear the write and modified bits in each of the given page's mappings.
1301  */
1302 void
1303 moea_remove_write(mmu_t mmu, vm_page_t m)
1304 {
1305 	struct	pvo_entry *pvo;
1306 	struct	pte *pt;
1307 	pmap_t	pmap;
1308 	u_int	lo;
1309 
1310 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1311 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1312 	    (m->flags & PG_WRITEABLE) == 0)
1313 		return;
1314 	lo = moea_attr_fetch(m);
1315 	SYNC();
1316 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1317 		pmap = pvo->pvo_pmap;
1318 		PMAP_LOCK(pmap);
1319 		if ((pvo->pvo_pte.pte_lo & PTE_PP) != PTE_BR) {
1320 			pt = moea_pvo_to_pte(pvo, -1);
1321 			pvo->pvo_pte.pte_lo &= ~PTE_PP;
1322 			pvo->pvo_pte.pte_lo |= PTE_BR;
1323 			if (pt != NULL) {
1324 				moea_pte_synch(pt, &pvo->pvo_pte);
1325 				lo |= pvo->pvo_pte.pte_lo;
1326 				pvo->pvo_pte.pte_lo &= ~PTE_CHG;
1327 				moea_pte_change(pt, &pvo->pvo_pte,
1328 				    pvo->pvo_vaddr);
1329 				mtx_unlock(&moea_table_mutex);
1330 			}
1331 		}
1332 		PMAP_UNLOCK(pmap);
1333 	}
1334 	if ((lo & PTE_CHG) != 0) {
1335 		moea_attr_clear(m, PTE_CHG);
1336 		vm_page_dirty(m);
1337 	}
1338 	vm_page_flag_clear(m, PG_WRITEABLE);
1339 }
1340 
1341 /*
1342  *	moea_ts_referenced:
1343  *
1344  *	Return a count of reference bits for a page, clearing those bits.
1345  *	It is not necessary for every reference bit to be cleared, but it
1346  *	is necessary that 0 only be returned when there are truly no
1347  *	reference bits set.
1348  *
1349  *	XXX: The exact number of bits to check and clear is a matter that
1350  *	should be tested and standardized at some point in the future for
1351  *	optimal aging of shared pages.
1352  */
1353 boolean_t
1354 moea_ts_referenced(mmu_t mmu, vm_page_t m)
1355 {
1356 	int count;
1357 
1358 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1359 		return (0);
1360 
1361 	count = moea_clear_bit(m, PTE_REF, NULL);
1362 
1363 	return (count);
1364 }
1365 
1366 /*
1367  * Map a wired page into kernel virtual address space.
1368  */
1369 void
1370 moea_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1371 {
1372 	u_int		pte_lo;
1373 	int		error;
1374 	int		i;
1375 
1376 #if 0
1377 	if (va < VM_MIN_KERNEL_ADDRESS)
1378 		panic("moea_kenter: attempt to enter non-kernel address %#x",
1379 		    va);
1380 #endif
1381 
1382 	pte_lo = PTE_I | PTE_G;
1383 	for (i = 0; i < pregions_sz; i++) {
1384 		if ((pa >= pregions[i].mr_start) &&
1385 		    (pa < (pregions[i].mr_start + pregions[i].mr_size))) {
1386 			pte_lo &= ~(PTE_I | PTE_G);
1387 			break;
1388 		}
1389 	}
1390 
1391 	PMAP_LOCK(kernel_pmap);
1392 	error = moea_pvo_enter(kernel_pmap, moea_upvo_zone,
1393 	    &moea_pvo_kunmanaged, va, pa, pte_lo, PVO_WIRED);
1394 
1395 	if (error != 0 && error != ENOENT)
1396 		panic("moea_kenter: failed to enter va %#x pa %#x: %d", va,
1397 		    pa, error);
1398 
1399 	/*
1400 	 * Flush the real memory from the instruction cache.
1401 	 */
1402 	if ((pte_lo & (PTE_I | PTE_G)) == 0) {
1403 		moea_syncicache(pa, PAGE_SIZE);
1404 	}
1405 	PMAP_UNLOCK(kernel_pmap);
1406 }
1407 
1408 /*
1409  * Extract the physical page address associated with the given kernel virtual
1410  * address.
1411  */
1412 vm_offset_t
1413 moea_kextract(mmu_t mmu, vm_offset_t va)
1414 {
1415 	struct		pvo_entry *pvo;
1416 	vm_paddr_t pa;
1417 
1418 #ifdef UMA_MD_SMALL_ALLOC
1419 	/*
1420 	 * Allow direct mappings
1421 	 */
1422 	if (va < VM_MIN_KERNEL_ADDRESS) {
1423 		return (va);
1424 	}
1425 #endif
1426 
1427 	PMAP_LOCK(kernel_pmap);
1428 	pvo = moea_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL);
1429 	KASSERT(pvo != NULL, ("moea_kextract: no addr found"));
1430 	pa = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF);
1431 	PMAP_UNLOCK(kernel_pmap);
1432 	return (pa);
1433 }
1434 
1435 /*
1436  * Remove a wired page from kernel virtual address space.
1437  */
1438 void
1439 moea_kremove(mmu_t mmu, vm_offset_t va)
1440 {
1441 
1442 	moea_remove(mmu, kernel_pmap, va, va + PAGE_SIZE);
1443 }
1444 
1445 /*
1446  * Map a range of physical addresses into kernel virtual address space.
1447  *
1448  * The value passed in *virt is a suggested virtual address for the mapping.
1449  * Architectures which can support a direct-mapped physical to virtual region
1450  * can return the appropriate address within that region, leaving '*virt'
1451  * unchanged.  We cannot and therefore do not; *virt is updated with the
1452  * first usable address after the mapped region.
1453  */
1454 vm_offset_t
1455 moea_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1456     vm_offset_t pa_end, int prot)
1457 {
1458 	vm_offset_t	sva, va;
1459 
1460 	sva = *virt;
1461 	va = sva;
1462 	for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE)
1463 		moea_kenter(mmu, va, pa_start);
1464 	*virt = va;
1465 	return (sva);
1466 }
1467 
1468 /*
1469  * Returns true if the pmap's pv is one of the first
1470  * 16 pvs linked to from this page.  This count may
1471  * be changed upwards or downwards in the future; it
1472  * is only necessary that true be returned for a small
1473  * subset of pmaps for proper page aging.
1474  */
1475 boolean_t
1476 moea_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
1477 {
1478         int loops;
1479 	struct pvo_entry *pvo;
1480 
1481         if (!moea_initialized || (m->flags & PG_FICTITIOUS))
1482                 return FALSE;
1483 
1484 	loops = 0;
1485 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1486 		if (pvo->pvo_pmap == pmap)
1487 			return (TRUE);
1488 		if (++loops >= 16)
1489 			break;
1490 	}
1491 
1492 	return (FALSE);
1493 }
1494 
1495 static u_int	moea_vsidcontext;
1496 
1497 void
1498 moea_pinit(mmu_t mmu, pmap_t pmap)
1499 {
1500 	int	i, mask;
1501 	u_int	entropy;
1502 
1503 	KASSERT((int)pmap < VM_MIN_KERNEL_ADDRESS, ("moea_pinit: virt pmap"));
1504 	PMAP_LOCK_INIT(pmap);
1505 
1506 	entropy = 0;
1507 	__asm __volatile("mftb %0" : "=r"(entropy));
1508 
1509 	/*
1510 	 * Allocate some segment registers for this pmap.
1511 	 */
1512 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1513 		u_int	hash, n;
1514 
1515 		/*
1516 		 * Create a new value by mutiplying by a prime and adding in
1517 		 * entropy from the timebase register.  This is to make the
1518 		 * VSID more random so that the PT hash function collides
1519 		 * less often.  (Note that the prime casues gcc to do shifts
1520 		 * instead of a multiply.)
1521 		 */
1522 		moea_vsidcontext = (moea_vsidcontext * 0x1105) + entropy;
1523 		hash = moea_vsidcontext & (NPMAPS - 1);
1524 		if (hash == 0)		/* 0 is special, avoid it */
1525 			continue;
1526 		n = hash >> 5;
1527 		mask = 1 << (hash & (VSID_NBPW - 1));
1528 		hash = (moea_vsidcontext & 0xfffff);
1529 		if (moea_vsid_bitmap[n] & mask) {	/* collision? */
1530 			/* anything free in this bucket? */
1531 			if (moea_vsid_bitmap[n] == 0xffffffff) {
1532 				entropy = (moea_vsidcontext >> 20);
1533 				continue;
1534 			}
1535 			i = ffs(~moea_vsid_bitmap[i]) - 1;
1536 			mask = 1 << i;
1537 			hash &= 0xfffff & ~(VSID_NBPW - 1);
1538 			hash |= i;
1539 		}
1540 		moea_vsid_bitmap[n] |= mask;
1541 		for (i = 0; i < 16; i++)
1542 			pmap->pm_sr[i] = VSID_MAKE(i, hash);
1543 		return;
1544 	}
1545 
1546 	panic("moea_pinit: out of segments");
1547 }
1548 
1549 /*
1550  * Initialize the pmap associated with process 0.
1551  */
1552 void
1553 moea_pinit0(mmu_t mmu, pmap_t pm)
1554 {
1555 
1556 	moea_pinit(mmu, pm);
1557 	bzero(&pm->pm_stats, sizeof(pm->pm_stats));
1558 }
1559 
1560 /*
1561  * Set the physical protection on the specified range of this map as requested.
1562  */
1563 void
1564 moea_protect(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva,
1565     vm_prot_t prot)
1566 {
1567 	struct	pvo_entry *pvo;
1568 	struct	pte *pt;
1569 	int	pteidx;
1570 
1571 	CTR4(KTR_PMAP, "moea_protect: pm=%p sva=%#x eva=%#x prot=%#x", pm, sva,
1572 	    eva, prot);
1573 
1574 
1575 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1576 	    ("moea_protect: non current pmap"));
1577 
1578 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1579 		moea_remove(mmu, pm, sva, eva);
1580 		return;
1581 	}
1582 
1583 	vm_page_lock_queues();
1584 	PMAP_LOCK(pm);
1585 	for (; sva < eva; sva += PAGE_SIZE) {
1586 		pvo = moea_pvo_find_va(pm, sva, &pteidx);
1587 		if (pvo == NULL)
1588 			continue;
1589 
1590 		if ((prot & VM_PROT_EXECUTE) == 0)
1591 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1592 
1593 		/*
1594 		 * Grab the PTE pointer before we diddle with the cached PTE
1595 		 * copy.
1596 		 */
1597 		pt = moea_pvo_to_pte(pvo, pteidx);
1598 		/*
1599 		 * Change the protection of the page.
1600 		 */
1601 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
1602 		pvo->pvo_pte.pte_lo |= PTE_BR;
1603 
1604 		/*
1605 		 * If the PVO is in the page table, update that pte as well.
1606 		 */
1607 		if (pt != NULL) {
1608 			moea_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1609 			mtx_unlock(&moea_table_mutex);
1610 		}
1611 	}
1612 	vm_page_unlock_queues();
1613 	PMAP_UNLOCK(pm);
1614 }
1615 
1616 /*
1617  * Map a list of wired pages into kernel virtual address space.  This is
1618  * intended for temporary mappings which do not need page modification or
1619  * references recorded.  Existing mappings in the region are overwritten.
1620  */
1621 void
1622 moea_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1623 {
1624 	vm_offset_t va;
1625 
1626 	va = sva;
1627 	while (count-- > 0) {
1628 		moea_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1629 		va += PAGE_SIZE;
1630 		m++;
1631 	}
1632 }
1633 
1634 /*
1635  * Remove page mappings from kernel virtual address space.  Intended for
1636  * temporary mappings entered by moea_qenter.
1637  */
1638 void
1639 moea_qremove(mmu_t mmu, vm_offset_t sva, int count)
1640 {
1641 	vm_offset_t va;
1642 
1643 	va = sva;
1644 	while (count-- > 0) {
1645 		moea_kremove(mmu, va);
1646 		va += PAGE_SIZE;
1647 	}
1648 }
1649 
1650 void
1651 moea_release(mmu_t mmu, pmap_t pmap)
1652 {
1653         int idx, mask;
1654 
1655 	/*
1656 	 * Free segment register's VSID
1657 	 */
1658         if (pmap->pm_sr[0] == 0)
1659                 panic("moea_release");
1660 
1661         idx = VSID_TO_HASH(pmap->pm_sr[0]) & (NPMAPS-1);
1662         mask = 1 << (idx % VSID_NBPW);
1663         idx /= VSID_NBPW;
1664         moea_vsid_bitmap[idx] &= ~mask;
1665 	PMAP_LOCK_DESTROY(pmap);
1666 }
1667 
1668 /*
1669  * Remove the given range of addresses from the specified map.
1670  */
1671 void
1672 moea_remove(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1673 {
1674 	struct	pvo_entry *pvo;
1675 	int	pteidx;
1676 
1677 	vm_page_lock_queues();
1678 	PMAP_LOCK(pm);
1679 	for (; sva < eva; sva += PAGE_SIZE) {
1680 		pvo = moea_pvo_find_va(pm, sva, &pteidx);
1681 		if (pvo != NULL) {
1682 			moea_pvo_remove(pvo, pteidx);
1683 		}
1684 	}
1685 	PMAP_UNLOCK(pm);
1686 	vm_page_unlock_queues();
1687 }
1688 
1689 /*
1690  * Remove physical page from all pmaps in which it resides. moea_pvo_remove()
1691  * will reflect changes in pte's back to the vm_page.
1692  */
1693 void
1694 moea_remove_all(mmu_t mmu, vm_page_t m)
1695 {
1696 	struct  pvo_head *pvo_head;
1697 	struct	pvo_entry *pvo, *next_pvo;
1698 	pmap_t	pmap;
1699 
1700 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1701 
1702 	pvo_head = vm_page_to_pvoh(m);
1703 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1704 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1705 
1706 		MOEA_PVO_CHECK(pvo);	/* sanity check */
1707 		pmap = pvo->pvo_pmap;
1708 		PMAP_LOCK(pmap);
1709 		moea_pvo_remove(pvo, -1);
1710 		PMAP_UNLOCK(pmap);
1711 	}
1712 	vm_page_flag_clear(m, PG_WRITEABLE);
1713 }
1714 
1715 /*
1716  * Allocate a physical page of memory directly from the phys_avail map.
1717  * Can only be called from moea_bootstrap before avail start and end are
1718  * calculated.
1719  */
1720 static vm_offset_t
1721 moea_bootstrap_alloc(vm_size_t size, u_int align)
1722 {
1723 	vm_offset_t	s, e;
1724 	int		i, j;
1725 
1726 	size = round_page(size);
1727 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1728 		if (align != 0)
1729 			s = (phys_avail[i] + align - 1) & ~(align - 1);
1730 		else
1731 			s = phys_avail[i];
1732 		e = s + size;
1733 
1734 		if (s < phys_avail[i] || e > phys_avail[i + 1])
1735 			continue;
1736 
1737 		if (s == phys_avail[i]) {
1738 			phys_avail[i] += size;
1739 		} else if (e == phys_avail[i + 1]) {
1740 			phys_avail[i + 1] -= size;
1741 		} else {
1742 			for (j = phys_avail_count * 2; j > i; j -= 2) {
1743 				phys_avail[j] = phys_avail[j - 2];
1744 				phys_avail[j + 1] = phys_avail[j - 1];
1745 			}
1746 
1747 			phys_avail[i + 3] = phys_avail[i + 1];
1748 			phys_avail[i + 1] = s;
1749 			phys_avail[i + 2] = e;
1750 			phys_avail_count++;
1751 		}
1752 
1753 		return (s);
1754 	}
1755 	panic("moea_bootstrap_alloc: could not allocate memory");
1756 }
1757 
1758 /*
1759  * Return an unmapped pvo for a kernel virtual address.
1760  * Used by pmap functions that operate on physical pages.
1761  */
1762 static struct pvo_entry *
1763 moea_rkva_alloc(mmu_t mmu)
1764 {
1765 	struct		pvo_entry *pvo;
1766 	struct		pte *pt;
1767 	vm_offset_t	kva;
1768 	int		pteidx;
1769 
1770 	if (moea_rkva_count == 0)
1771 		panic("moea_rkva_alloc: no more reserved KVAs");
1772 
1773 	kva = moea_rkva_start + (PAGE_SIZE * --moea_rkva_count);
1774 	moea_kenter(mmu, kva, 0);
1775 
1776 	pvo = moea_pvo_find_va(kernel_pmap, kva, &pteidx);
1777 
1778 	if (pvo == NULL)
1779 		panic("moea_kva_alloc: moea_pvo_find_va failed");
1780 
1781 	pt = moea_pvo_to_pte(pvo, pteidx);
1782 
1783 	if (pt == NULL)
1784 		panic("moea_kva_alloc: moea_pvo_to_pte failed");
1785 
1786 	moea_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1787 	mtx_unlock(&moea_table_mutex);
1788 	PVO_PTEGIDX_CLR(pvo);
1789 
1790 	moea_pte_overflow++;
1791 
1792 	return (pvo);
1793 }
1794 
1795 static void
1796 moea_pa_map(struct pvo_entry *pvo, vm_offset_t pa, struct pte *saved_pt,
1797     int *depth_p)
1798 {
1799 	struct	pte *pt;
1800 
1801 	/*
1802 	 * If this pvo already has a valid pte, we need to save it so it can
1803 	 * be restored later.  We then just reload the new PTE over the old
1804 	 * slot.
1805 	 */
1806 	if (saved_pt != NULL) {
1807 		pt = moea_pvo_to_pte(pvo, -1);
1808 
1809 		if (pt != NULL) {
1810 			moea_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1811 			mtx_unlock(&moea_table_mutex);
1812 			PVO_PTEGIDX_CLR(pvo);
1813 			moea_pte_overflow++;
1814 		}
1815 
1816 		*saved_pt = pvo->pvo_pte;
1817 
1818 		pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1819 	}
1820 
1821 	pvo->pvo_pte.pte_lo |= pa;
1822 
1823 	if (!moea_pte_spill(pvo->pvo_vaddr))
1824 		panic("moea_pa_map: could not spill pvo %p", pvo);
1825 
1826 	if (depth_p != NULL)
1827 		(*depth_p)++;
1828 }
1829 
1830 static void
1831 moea_pa_unmap(struct pvo_entry *pvo, struct pte *saved_pt, int *depth_p)
1832 {
1833 	struct	pte *pt;
1834 
1835 	pt = moea_pvo_to_pte(pvo, -1);
1836 
1837 	if (pt != NULL) {
1838 		moea_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1839 		mtx_unlock(&moea_table_mutex);
1840 		PVO_PTEGIDX_CLR(pvo);
1841 		moea_pte_overflow++;
1842 	}
1843 
1844 	pvo->pvo_pte.pte_lo &= ~PTE_RPGN;
1845 
1846 	/*
1847 	 * If there is a saved PTE and it's valid, restore it and return.
1848 	 */
1849 	if (saved_pt != NULL && (saved_pt->pte_lo & PTE_RPGN) != 0) {
1850 		if (depth_p != NULL && --(*depth_p) == 0)
1851 			panic("moea_pa_unmap: restoring but depth == 0");
1852 
1853 		pvo->pvo_pte = *saved_pt;
1854 
1855 		if (!moea_pte_spill(pvo->pvo_vaddr))
1856 			panic("moea_pa_unmap: could not spill pvo %p", pvo);
1857 	}
1858 }
1859 
1860 static void
1861 moea_syncicache(vm_offset_t pa, vm_size_t len)
1862 {
1863 	__syncicache((void *)pa, len);
1864 }
1865 
1866 static void
1867 tlbia(void)
1868 {
1869 	caddr_t	i;
1870 
1871 	SYNC();
1872 	for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) {
1873 		TLBIE(i);
1874 		EIEIO();
1875 	}
1876 	TLBSYNC();
1877 	SYNC();
1878 }
1879 
1880 static int
1881 moea_pvo_enter(pmap_t pm, uma_zone_t zone, struct pvo_head *pvo_head,
1882     vm_offset_t va, vm_offset_t pa, u_int pte_lo, int flags)
1883 {
1884 	struct	pvo_entry *pvo;
1885 	u_int	sr;
1886 	int	first;
1887 	u_int	ptegidx;
1888 	int	i;
1889 	int     bootstrap;
1890 
1891 	moea_pvo_enter_calls++;
1892 	first = 0;
1893 	bootstrap = 0;
1894 
1895 	/*
1896 	 * Compute the PTE Group index.
1897 	 */
1898 	va &= ~ADDR_POFF;
1899 	sr = va_to_sr(pm->pm_sr, va);
1900 	ptegidx = va_to_pteg(sr, va);
1901 
1902 	/*
1903 	 * Remove any existing mapping for this page.  Reuse the pvo entry if
1904 	 * there is a mapping.
1905 	 */
1906 	mtx_lock(&moea_table_mutex);
1907 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
1908 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1909 			if ((pvo->pvo_pte.pte_lo & PTE_RPGN) == pa &&
1910 			    (pvo->pvo_pte.pte_lo & PTE_PP) ==
1911 			    (pte_lo & PTE_PP)) {
1912 				mtx_unlock(&moea_table_mutex);
1913 				return (0);
1914 			}
1915 			moea_pvo_remove(pvo, -1);
1916 			break;
1917 		}
1918 	}
1919 
1920 	/*
1921 	 * If we aren't overwriting a mapping, try to allocate.
1922 	 */
1923 	if (moea_initialized) {
1924 		pvo = uma_zalloc(zone, M_NOWAIT);
1925 	} else {
1926 		if (moea_bpvo_pool_index >= BPVO_POOL_SIZE) {
1927 			panic("moea_enter: bpvo pool exhausted, %d, %d, %d",
1928 			      moea_bpvo_pool_index, BPVO_POOL_SIZE,
1929 			      BPVO_POOL_SIZE * sizeof(struct pvo_entry));
1930 		}
1931 		pvo = &moea_bpvo_pool[moea_bpvo_pool_index];
1932 		moea_bpvo_pool_index++;
1933 		bootstrap = 1;
1934 	}
1935 
1936 	if (pvo == NULL) {
1937 		mtx_unlock(&moea_table_mutex);
1938 		return (ENOMEM);
1939 	}
1940 
1941 	moea_pvo_entries++;
1942 	pvo->pvo_vaddr = va;
1943 	pvo->pvo_pmap = pm;
1944 	LIST_INSERT_HEAD(&moea_pvo_table[ptegidx], pvo, pvo_olink);
1945 	pvo->pvo_vaddr &= ~ADDR_POFF;
1946 	if (flags & VM_PROT_EXECUTE)
1947 		pvo->pvo_vaddr |= PVO_EXECUTABLE;
1948 	if (flags & PVO_WIRED)
1949 		pvo->pvo_vaddr |= PVO_WIRED;
1950 	if (pvo_head != &moea_pvo_kunmanaged)
1951 		pvo->pvo_vaddr |= PVO_MANAGED;
1952 	if (bootstrap)
1953 		pvo->pvo_vaddr |= PVO_BOOTSTRAP;
1954 	if (flags & PVO_FAKE)
1955 		pvo->pvo_vaddr |= PVO_FAKE;
1956 
1957 	moea_pte_create(&pvo->pvo_pte, sr, va, pa | pte_lo);
1958 
1959 	/*
1960 	 * Remember if the list was empty and therefore will be the first
1961 	 * item.
1962 	 */
1963 	if (LIST_FIRST(pvo_head) == NULL)
1964 		first = 1;
1965 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1966 
1967 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1968 		pm->pm_stats.wired_count++;
1969 	pm->pm_stats.resident_count++;
1970 
1971 	/*
1972 	 * We hope this succeeds but it isn't required.
1973 	 */
1974 	i = moea_pte_insert(ptegidx, &pvo->pvo_pte);
1975 	if (i >= 0) {
1976 		PVO_PTEGIDX_SET(pvo, i);
1977 	} else {
1978 		panic("moea_pvo_enter: overflow");
1979 		moea_pte_overflow++;
1980 	}
1981 	mtx_unlock(&moea_table_mutex);
1982 
1983 	return (first ? ENOENT : 0);
1984 }
1985 
1986 static void
1987 moea_pvo_remove(struct pvo_entry *pvo, int pteidx)
1988 {
1989 	struct	pte *pt;
1990 
1991 	/*
1992 	 * If there is an active pte entry, we need to deactivate it (and
1993 	 * save the ref & cfg bits).
1994 	 */
1995 	pt = moea_pvo_to_pte(pvo, pteidx);
1996 	if (pt != NULL) {
1997 		moea_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1998 		mtx_unlock(&moea_table_mutex);
1999 		PVO_PTEGIDX_CLR(pvo);
2000 	} else {
2001 		moea_pte_overflow--;
2002 	}
2003 
2004 	/*
2005 	 * Update our statistics.
2006 	 */
2007 	pvo->pvo_pmap->pm_stats.resident_count--;
2008 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
2009 		pvo->pvo_pmap->pm_stats.wired_count--;
2010 
2011 	/*
2012 	 * Save the REF/CHG bits into their cache if the page is managed.
2013 	 */
2014 	if ((pvo->pvo_vaddr & (PVO_MANAGED|PVO_FAKE)) == PVO_MANAGED) {
2015 		struct	vm_page *pg;
2016 
2017 		pg = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte_lo & PTE_RPGN);
2018 		if (pg != NULL) {
2019 			moea_attr_save(pg, pvo->pvo_pte.pte_lo &
2020 			    (PTE_REF | PTE_CHG));
2021 		}
2022 	}
2023 
2024 	/*
2025 	 * Remove this PVO from the PV list.
2026 	 */
2027 	LIST_REMOVE(pvo, pvo_vlink);
2028 
2029 	/*
2030 	 * Remove this from the overflow list and return it to the pool
2031 	 * if we aren't going to reuse it.
2032 	 */
2033 	LIST_REMOVE(pvo, pvo_olink);
2034 	if (!(pvo->pvo_vaddr & PVO_BOOTSTRAP))
2035 		uma_zfree(pvo->pvo_vaddr & PVO_MANAGED ? moea_mpvo_zone :
2036 		    moea_upvo_zone, pvo);
2037 	moea_pvo_entries--;
2038 	moea_pvo_remove_calls++;
2039 }
2040 
2041 static __inline int
2042 moea_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
2043 {
2044 	int	pteidx;
2045 
2046 	/*
2047 	 * We can find the actual pte entry without searching by grabbing
2048 	 * the PTEG index from 3 unused bits in pte_lo[11:9] and by
2049 	 * noticing the HID bit.
2050 	 */
2051 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
2052 	if (pvo->pvo_pte.pte_hi & PTE_HID)
2053 		pteidx ^= moea_pteg_mask * 8;
2054 
2055 	return (pteidx);
2056 }
2057 
2058 static struct pvo_entry *
2059 moea_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p)
2060 {
2061 	struct	pvo_entry *pvo;
2062 	int	ptegidx;
2063 	u_int	sr;
2064 
2065 	va &= ~ADDR_POFF;
2066 	sr = va_to_sr(pm->pm_sr, va);
2067 	ptegidx = va_to_pteg(sr, va);
2068 
2069 	mtx_lock(&moea_table_mutex);
2070 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
2071 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
2072 			if (pteidx_p)
2073 				*pteidx_p = moea_pvo_pte_index(pvo, ptegidx);
2074 			break;
2075 		}
2076 	}
2077 	mtx_unlock(&moea_table_mutex);
2078 
2079 	return (pvo);
2080 }
2081 
2082 static struct pte *
2083 moea_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
2084 {
2085 	struct	pte *pt;
2086 
2087 	/*
2088 	 * If we haven't been supplied the ptegidx, calculate it.
2089 	 */
2090 	if (pteidx == -1) {
2091 		int	ptegidx;
2092 		u_int	sr;
2093 
2094 		sr = va_to_sr(pvo->pvo_pmap->pm_sr, pvo->pvo_vaddr);
2095 		ptegidx = va_to_pteg(sr, pvo->pvo_vaddr);
2096 		pteidx = moea_pvo_pte_index(pvo, ptegidx);
2097 	}
2098 
2099 	pt = &moea_pteg_table[pteidx >> 3].pt[pteidx & 7];
2100 	mtx_lock(&moea_table_mutex);
2101 
2102 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
2103 		panic("moea_pvo_to_pte: pvo %p has valid pte in pvo but no "
2104 		    "valid pte index", pvo);
2105 	}
2106 
2107 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
2108 		panic("moea_pvo_to_pte: pvo %p has valid pte index in pvo "
2109 		    "pvo but no valid pte", pvo);
2110 	}
2111 
2112 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
2113 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
2114 			panic("moea_pvo_to_pte: pvo %p has valid pte in "
2115 			    "moea_pteg_table %p but invalid in pvo", pvo, pt);
2116 		}
2117 
2118 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF))
2119 		    != 0) {
2120 			panic("moea_pvo_to_pte: pvo %p pte does not match "
2121 			    "pte %p in moea_pteg_table", pvo, pt);
2122 		}
2123 
2124 		mtx_assert(&moea_table_mutex, MA_OWNED);
2125 		return (pt);
2126 	}
2127 
2128 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
2129 		panic("moea_pvo_to_pte: pvo %p has invalid pte %p in "
2130 		    "moea_pteg_table but valid in pvo", pvo, pt);
2131 	}
2132 
2133 	mtx_unlock(&moea_table_mutex);
2134 	return (NULL);
2135 }
2136 
2137 /*
2138  * XXX: THIS STUFF SHOULD BE IN pte.c?
2139  */
2140 int
2141 moea_pte_spill(vm_offset_t addr)
2142 {
2143 	struct	pvo_entry *source_pvo, *victim_pvo;
2144 	struct	pvo_entry *pvo;
2145 	int	ptegidx, i, j;
2146 	u_int	sr;
2147 	struct	pteg *pteg;
2148 	struct	pte *pt;
2149 
2150 	moea_pte_spills++;
2151 
2152 	sr = mfsrin(addr);
2153 	ptegidx = va_to_pteg(sr, addr);
2154 
2155 	/*
2156 	 * Have to substitute some entry.  Use the primary hash for this.
2157 	 * Use low bits of timebase as random generator.
2158 	 */
2159 	pteg = &moea_pteg_table[ptegidx];
2160 	mtx_lock(&moea_table_mutex);
2161 	__asm __volatile("mftb %0" : "=r"(i));
2162 	i &= 7;
2163 	pt = &pteg->pt[i];
2164 
2165 	source_pvo = NULL;
2166 	victim_pvo = NULL;
2167 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
2168 		/*
2169 		 * We need to find a pvo entry for this address.
2170 		 */
2171 		MOEA_PVO_CHECK(pvo);
2172 		if (source_pvo == NULL &&
2173 		    moea_pte_match(&pvo->pvo_pte, sr, addr,
2174 		    pvo->pvo_pte.pte_hi & PTE_HID)) {
2175 			/*
2176 			 * Now found an entry to be spilled into the pteg.
2177 			 * The PTE is now valid, so we know it's active.
2178 			 */
2179 			j = moea_pte_insert(ptegidx, &pvo->pvo_pte);
2180 
2181 			if (j >= 0) {
2182 				PVO_PTEGIDX_SET(pvo, j);
2183 				moea_pte_overflow--;
2184 				MOEA_PVO_CHECK(pvo);
2185 				mtx_unlock(&moea_table_mutex);
2186 				return (1);
2187 			}
2188 
2189 			source_pvo = pvo;
2190 
2191 			if (victim_pvo != NULL)
2192 				break;
2193 		}
2194 
2195 		/*
2196 		 * We also need the pvo entry of the victim we are replacing
2197 		 * so save the R & C bits of the PTE.
2198 		 */
2199 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
2200 		    moea_pte_compare(pt, &pvo->pvo_pte)) {
2201 			victim_pvo = pvo;
2202 			if (source_pvo != NULL)
2203 				break;
2204 		}
2205 	}
2206 
2207 	if (source_pvo == NULL) {
2208 		mtx_unlock(&moea_table_mutex);
2209 		return (0);
2210 	}
2211 
2212 	if (victim_pvo == NULL) {
2213 		if ((pt->pte_hi & PTE_HID) == 0)
2214 			panic("moea_pte_spill: victim p-pte (%p) has no pvo"
2215 			    "entry", pt);
2216 
2217 		/*
2218 		 * If this is a secondary PTE, we need to search it's primary
2219 		 * pvo bucket for the matching PVO.
2220 		 */
2221 		LIST_FOREACH(pvo, &moea_pvo_table[ptegidx ^ moea_pteg_mask],
2222 		    pvo_olink) {
2223 			MOEA_PVO_CHECK(pvo);
2224 			/*
2225 			 * We also need the pvo entry of the victim we are
2226 			 * replacing so save the R & C bits of the PTE.
2227 			 */
2228 			if (moea_pte_compare(pt, &pvo->pvo_pte)) {
2229 				victim_pvo = pvo;
2230 				break;
2231 			}
2232 		}
2233 
2234 		if (victim_pvo == NULL)
2235 			panic("moea_pte_spill: victim s-pte (%p) has no pvo"
2236 			    "entry", pt);
2237 	}
2238 
2239 	/*
2240 	 * We are invalidating the TLB entry for the EA we are replacing even
2241 	 * though it's valid.  If we don't, we lose any ref/chg bit changes
2242 	 * contained in the TLB entry.
2243 	 */
2244 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
2245 
2246 	moea_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
2247 	moea_pte_set(pt, &source_pvo->pvo_pte);
2248 
2249 	PVO_PTEGIDX_CLR(victim_pvo);
2250 	PVO_PTEGIDX_SET(source_pvo, i);
2251 	moea_pte_replacements++;
2252 
2253 	MOEA_PVO_CHECK(victim_pvo);
2254 	MOEA_PVO_CHECK(source_pvo);
2255 
2256 	mtx_unlock(&moea_table_mutex);
2257 	return (1);
2258 }
2259 
2260 static int
2261 moea_pte_insert(u_int ptegidx, struct pte *pvo_pt)
2262 {
2263 	struct	pte *pt;
2264 	int	i;
2265 
2266 	mtx_assert(&moea_table_mutex, MA_OWNED);
2267 
2268 	/*
2269 	 * First try primary hash.
2270 	 */
2271 	for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2272 		if ((pt->pte_hi & PTE_VALID) == 0) {
2273 			pvo_pt->pte_hi &= ~PTE_HID;
2274 			moea_pte_set(pt, pvo_pt);
2275 			return (i);
2276 		}
2277 	}
2278 
2279 	/*
2280 	 * Now try secondary hash.
2281 	 */
2282 	ptegidx ^= moea_pteg_mask;
2283 
2284 	for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2285 		if ((pt->pte_hi & PTE_VALID) == 0) {
2286 			pvo_pt->pte_hi |= PTE_HID;
2287 			moea_pte_set(pt, pvo_pt);
2288 			return (i);
2289 		}
2290 	}
2291 
2292 	panic("moea_pte_insert: overflow");
2293 	return (-1);
2294 }
2295 
2296 static boolean_t
2297 moea_query_bit(vm_page_t m, int ptebit)
2298 {
2299 	struct	pvo_entry *pvo;
2300 	struct	pte *pt;
2301 
2302 #if 0
2303 	if (moea_attr_fetch(m) & ptebit)
2304 		return (TRUE);
2305 #endif
2306 
2307 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2308 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2309 
2310 		/*
2311 		 * See if we saved the bit off.  If so, cache it and return
2312 		 * success.
2313 		 */
2314 		if (pvo->pvo_pte.pte_lo & ptebit) {
2315 			moea_attr_save(m, ptebit);
2316 			MOEA_PVO_CHECK(pvo);	/* sanity check */
2317 			return (TRUE);
2318 		}
2319 	}
2320 
2321 	/*
2322 	 * No luck, now go through the hard part of looking at the PTEs
2323 	 * themselves.  Sync so that any pending REF/CHG bits are flushed to
2324 	 * the PTEs.
2325 	 */
2326 	SYNC();
2327 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2328 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2329 
2330 		/*
2331 		 * See if this pvo has a valid PTE.  if so, fetch the
2332 		 * REF/CHG bits from the valid PTE.  If the appropriate
2333 		 * ptebit is set, cache it and return success.
2334 		 */
2335 		pt = moea_pvo_to_pte(pvo, -1);
2336 		if (pt != NULL) {
2337 			moea_pte_synch(pt, &pvo->pvo_pte);
2338 			mtx_unlock(&moea_table_mutex);
2339 			if (pvo->pvo_pte.pte_lo & ptebit) {
2340 				moea_attr_save(m, ptebit);
2341 				MOEA_PVO_CHECK(pvo);	/* sanity check */
2342 				return (TRUE);
2343 			}
2344 		}
2345 	}
2346 
2347 	return (FALSE);
2348 }
2349 
2350 static u_int
2351 moea_clear_bit(vm_page_t m, int ptebit, int *origbit)
2352 {
2353 	u_int	count;
2354 	struct	pvo_entry *pvo;
2355 	struct	pte *pt;
2356 	int	rv;
2357 
2358 	/*
2359 	 * Clear the cached value.
2360 	 */
2361 	rv = moea_attr_fetch(m);
2362 	moea_attr_clear(m, ptebit);
2363 
2364 	/*
2365 	 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so
2366 	 * we can reset the right ones).  note that since the pvo entries and
2367 	 * list heads are accessed via BAT0 and are never placed in the page
2368 	 * table, we don't have to worry about further accesses setting the
2369 	 * REF/CHG bits.
2370 	 */
2371 	SYNC();
2372 
2373 	/*
2374 	 * For each pvo entry, clear the pvo's ptebit.  If this pvo has a
2375 	 * valid pte clear the ptebit from the valid pte.
2376 	 */
2377 	count = 0;
2378 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2379 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2380 		pt = moea_pvo_to_pte(pvo, -1);
2381 		if (pt != NULL) {
2382 			moea_pte_synch(pt, &pvo->pvo_pte);
2383 			if (pvo->pvo_pte.pte_lo & ptebit) {
2384 				count++;
2385 				moea_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2386 			}
2387 			mtx_unlock(&moea_table_mutex);
2388 		}
2389 		rv |= pvo->pvo_pte.pte_lo;
2390 		pvo->pvo_pte.pte_lo &= ~ptebit;
2391 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2392 	}
2393 
2394 	if (origbit != NULL) {
2395 		*origbit = rv;
2396 	}
2397 
2398 	return (count);
2399 }
2400 
2401 /*
2402  * Return true if the physical range is encompassed by the battable[idx]
2403  */
2404 static int
2405 moea_bat_mapped(int idx, vm_offset_t pa, vm_size_t size)
2406 {
2407 	u_int prot;
2408 	u_int32_t start;
2409 	u_int32_t end;
2410 	u_int32_t bat_ble;
2411 
2412 	/*
2413 	 * Return immediately if not a valid mapping
2414 	 */
2415 	if (!battable[idx].batu & BAT_Vs)
2416 		return (EINVAL);
2417 
2418 	/*
2419 	 * The BAT entry must be cache-inhibited, guarded, and r/w
2420 	 * so it can function as an i/o page
2421 	 */
2422 	prot = battable[idx].batl & (BAT_I|BAT_G|BAT_PP_RW);
2423 	if (prot != (BAT_I|BAT_G|BAT_PP_RW))
2424 		return (EPERM);
2425 
2426 	/*
2427 	 * The address should be within the BAT range. Assume that the
2428 	 * start address in the BAT has the correct alignment (thus
2429 	 * not requiring masking)
2430 	 */
2431 	start = battable[idx].batl & BAT_PBS;
2432 	bat_ble = (battable[idx].batu & ~(BAT_EBS)) | 0x03;
2433 	end = start | (bat_ble << 15) | 0x7fff;
2434 
2435 	if ((pa < start) || ((pa + size) > end))
2436 		return (ERANGE);
2437 
2438 	return (0);
2439 }
2440 
2441 boolean_t
2442 moea_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2443 {
2444 	int i;
2445 
2446 	/*
2447 	 * This currently does not work for entries that
2448 	 * overlap 256M BAT segments.
2449 	 */
2450 
2451 	for(i = 0; i < 16; i++)
2452 		if (moea_bat_mapped(i, pa, size) == 0)
2453 			return (0);
2454 
2455 	return (EFAULT);
2456 }
2457 
2458 boolean_t
2459 moea_page_executable(mmu_t mmu, vm_page_t pg)
2460 {
2461 	return ((moea_attr_fetch(pg) & PTE_EXEC) == PTE_EXEC);
2462 }
2463 
2464 /*
2465  * Map a set of physical memory pages into the kernel virtual
2466  * address space. Return a pointer to where it is mapped. This
2467  * routine is intended to be used for mapping device memory,
2468  * NOT real memory.
2469  */
2470 void *
2471 moea_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2472 {
2473 	vm_offset_t va, tmpva, ppa, offset;
2474 	int i;
2475 
2476 	ppa = trunc_page(pa);
2477 	offset = pa & PAGE_MASK;
2478 	size = roundup(offset + size, PAGE_SIZE);
2479 
2480 	GIANT_REQUIRED;
2481 
2482 	/*
2483 	 * If the physical address lies within a valid BAT table entry,
2484 	 * return the 1:1 mapping. This currently doesn't work
2485 	 * for regions that overlap 256M BAT segments.
2486 	 */
2487 	for (i = 0; i < 16; i++) {
2488 		if (moea_bat_mapped(i, pa, size) == 0)
2489 			return ((void *) pa);
2490 	}
2491 
2492 	va = kmem_alloc_nofault(kernel_map, size);
2493 	if (!va)
2494 		panic("moea_mapdev: Couldn't alloc kernel virtual memory");
2495 
2496 	for (tmpva = va; size > 0;) {
2497 		moea_kenter(mmu, tmpva, ppa);
2498 		TLBIE(tmpva); /* XXX or should it be invalidate-all ? */
2499 		size -= PAGE_SIZE;
2500 		tmpva += PAGE_SIZE;
2501 		ppa += PAGE_SIZE;
2502 	}
2503 
2504 	return ((void *)(va + offset));
2505 }
2506 
2507 void
2508 moea_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2509 {
2510 	vm_offset_t base, offset;
2511 
2512 	/*
2513 	 * If this is outside kernel virtual space, then it's a
2514 	 * battable entry and doesn't require unmapping
2515 	 */
2516 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2517 		base = trunc_page(va);
2518 		offset = va & PAGE_MASK;
2519 		size = roundup(offset + size, PAGE_SIZE);
2520 		kmem_free(kernel_map, base, size);
2521 	}
2522 }
2523