xref: /freebsd/sys/powerpc/aim/mmu_oea.c (revision b3aaa0cc21c63d388230c7ef2a80abd631ff20d5)
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/smp.h>
151 #include <machine/sr.h>
152 #include <machine/mmuvar.h>
153 
154 #include "mmu_if.h"
155 
156 #define	MOEA_DEBUG
157 
158 #define TODO	panic("%s: not implemented", __func__);
159 
160 #define	VSID_MAKE(sr, hash)	((sr) | (((hash) & 0xfffff) << 4))
161 #define	VSID_TO_SR(vsid)	((vsid) & 0xf)
162 #define	VSID_TO_HASH(vsid)	(((vsid) >> 4) & 0xfffff)
163 
164 #define	PVO_PTEGIDX_MASK	0x007		/* which PTEG slot */
165 #define	PVO_PTEGIDX_VALID	0x008		/* slot is valid */
166 #define	PVO_WIRED		0x010		/* PVO entry is wired */
167 #define	PVO_MANAGED		0x020		/* PVO entry is managed */
168 #define	PVO_EXECUTABLE		0x040		/* PVO entry is executable */
169 #define	PVO_BOOTSTRAP		0x080		/* PVO entry allocated during
170 						   bootstrap */
171 #define PVO_FAKE		0x100		/* fictitious phys page */
172 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
173 #define	PVO_ISEXECUTABLE(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
174 #define PVO_ISFAKE(pvo)		((pvo)->pvo_vaddr & PVO_FAKE)
175 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
176 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
177 #define	PVO_PTEGIDX_CLR(pvo)	\
178 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
179 #define	PVO_PTEGIDX_SET(pvo, i)	\
180 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
181 
182 #define	MOEA_PVO_CHECK(pvo)
183 
184 struct ofw_map {
185 	vm_offset_t	om_va;
186 	vm_size_t	om_len;
187 	vm_offset_t	om_pa;
188 	u_int		om_mode;
189 };
190 
191 /*
192  * Map of physical memory regions.
193  */
194 static struct	mem_region *regions;
195 static struct	mem_region *pregions;
196 u_int           phys_avail_count;
197 int		regions_sz, pregions_sz;
198 static struct	ofw_map *translations;
199 
200 extern struct pmap ofw_pmap;
201 
202 /*
203  * Lock for the pteg and pvo tables.
204  */
205 struct mtx	moea_table_mutex;
206 
207 /* tlbie instruction synchronization */
208 static struct mtx tlbie_mtx;
209 
210 /*
211  * PTEG data.
212  */
213 static struct	pteg *moea_pteg_table;
214 u_int		moea_pteg_count;
215 u_int		moea_pteg_mask;
216 
217 /*
218  * PVO data.
219  */
220 struct	pvo_head *moea_pvo_table;		/* pvo entries by pteg index */
221 struct	pvo_head moea_pvo_kunmanaged =
222     LIST_HEAD_INITIALIZER(moea_pvo_kunmanaged);	/* list of unmanaged pages */
223 struct	pvo_head moea_pvo_unmanaged =
224     LIST_HEAD_INITIALIZER(moea_pvo_unmanaged);	/* list of unmanaged pages */
225 
226 uma_zone_t	moea_upvo_zone;	/* zone for pvo entries for unmanaged pages */
227 uma_zone_t	moea_mpvo_zone;	/* zone for pvo entries for managed pages */
228 
229 #define	BPVO_POOL_SIZE	32768
230 static struct	pvo_entry *moea_bpvo_pool;
231 static int	moea_bpvo_pool_index = 0;
232 
233 #define	VSID_NBPW	(sizeof(u_int32_t) * 8)
234 static u_int	moea_vsid_bitmap[NPMAPS / VSID_NBPW];
235 
236 static boolean_t moea_initialized = FALSE;
237 
238 /*
239  * Statistics.
240  */
241 u_int	moea_pte_valid = 0;
242 u_int	moea_pte_overflow = 0;
243 u_int	moea_pte_replacements = 0;
244 u_int	moea_pvo_entries = 0;
245 u_int	moea_pvo_enter_calls = 0;
246 u_int	moea_pvo_remove_calls = 0;
247 u_int	moea_pte_spills = 0;
248 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_valid, CTLFLAG_RD, &moea_pte_valid,
249     0, "");
250 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_overflow, CTLFLAG_RD,
251     &moea_pte_overflow, 0, "");
252 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_replacements, CTLFLAG_RD,
253     &moea_pte_replacements, 0, "");
254 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_entries, CTLFLAG_RD, &moea_pvo_entries,
255     0, "");
256 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_enter_calls, CTLFLAG_RD,
257     &moea_pvo_enter_calls, 0, "");
258 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_remove_calls, CTLFLAG_RD,
259     &moea_pvo_remove_calls, 0, "");
260 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_spills, CTLFLAG_RD,
261     &moea_pte_spills, 0, "");
262 
263 /*
264  * Allocate physical memory for use in moea_bootstrap.
265  */
266 static vm_offset_t	moea_bootstrap_alloc(vm_size_t, u_int);
267 
268 /*
269  * PTE calls.
270  */
271 static int		moea_pte_insert(u_int, struct pte *);
272 
273 /*
274  * PVO calls.
275  */
276 static int	moea_pvo_enter(pmap_t, uma_zone_t, struct pvo_head *,
277 		    vm_offset_t, vm_offset_t, u_int, int);
278 static void	moea_pvo_remove(struct pvo_entry *, int);
279 static struct	pvo_entry *moea_pvo_find_va(pmap_t, vm_offset_t, int *);
280 static struct	pte *moea_pvo_to_pte(const struct pvo_entry *, int);
281 
282 /*
283  * Utility routines.
284  */
285 static void		moea_enter_locked(pmap_t, vm_offset_t, vm_page_t,
286 			    vm_prot_t, boolean_t);
287 static void		moea_syncicache(vm_offset_t, vm_size_t);
288 static boolean_t	moea_query_bit(vm_page_t, int);
289 static u_int		moea_clear_bit(vm_page_t, int, int *);
290 static void		moea_kremove(mmu_t, vm_offset_t);
291 int		moea_pte_spill(vm_offset_t);
292 
293 /*
294  * Kernel MMU interface
295  */
296 void moea_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
297 void moea_clear_modify(mmu_t, vm_page_t);
298 void moea_clear_reference(mmu_t, vm_page_t);
299 void moea_copy_page(mmu_t, vm_page_t, vm_page_t);
300 void moea_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t, vm_prot_t, boolean_t);
301 void moea_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t, vm_page_t,
302     vm_prot_t);
303 void moea_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t, vm_prot_t);
304 vm_paddr_t moea_extract(mmu_t, pmap_t, vm_offset_t);
305 vm_page_t moea_extract_and_hold(mmu_t, pmap_t, vm_offset_t, vm_prot_t);
306 void moea_init(mmu_t);
307 boolean_t moea_is_modified(mmu_t, vm_page_t);
308 boolean_t moea_ts_referenced(mmu_t, vm_page_t);
309 vm_offset_t moea_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t, int);
310 boolean_t moea_page_exists_quick(mmu_t, pmap_t, vm_page_t);
311 int moea_page_wired_mappings(mmu_t, vm_page_t);
312 void moea_pinit(mmu_t, pmap_t);
313 void moea_pinit0(mmu_t, pmap_t);
314 void moea_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t, vm_prot_t);
315 void moea_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
316 void moea_qremove(mmu_t, vm_offset_t, int);
317 void moea_release(mmu_t, pmap_t);
318 void moea_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
319 void moea_remove_all(mmu_t, vm_page_t);
320 void moea_remove_write(mmu_t, vm_page_t);
321 void moea_zero_page(mmu_t, vm_page_t);
322 void moea_zero_page_area(mmu_t, vm_page_t, int, int);
323 void moea_zero_page_idle(mmu_t, vm_page_t);
324 void moea_activate(mmu_t, struct thread *);
325 void moea_deactivate(mmu_t, struct thread *);
326 void moea_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
327 void *moea_mapdev(mmu_t, vm_offset_t, vm_size_t);
328 void moea_unmapdev(mmu_t, vm_offset_t, vm_size_t);
329 vm_offset_t moea_kextract(mmu_t, vm_offset_t);
330 void moea_kenter(mmu_t, vm_offset_t, vm_offset_t);
331 boolean_t moea_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
332 boolean_t moea_page_executable(mmu_t, vm_page_t);
333 
334 static mmu_method_t moea_methods[] = {
335 	MMUMETHOD(mmu_change_wiring,	moea_change_wiring),
336 	MMUMETHOD(mmu_clear_modify,	moea_clear_modify),
337 	MMUMETHOD(mmu_clear_reference,	moea_clear_reference),
338 	MMUMETHOD(mmu_copy_page,	moea_copy_page),
339 	MMUMETHOD(mmu_enter,		moea_enter),
340 	MMUMETHOD(mmu_enter_object,	moea_enter_object),
341 	MMUMETHOD(mmu_enter_quick,	moea_enter_quick),
342 	MMUMETHOD(mmu_extract,		moea_extract),
343 	MMUMETHOD(mmu_extract_and_hold,	moea_extract_and_hold),
344 	MMUMETHOD(mmu_init,		moea_init),
345 	MMUMETHOD(mmu_is_modified,	moea_is_modified),
346 	MMUMETHOD(mmu_ts_referenced,	moea_ts_referenced),
347 	MMUMETHOD(mmu_map,     		moea_map),
348 	MMUMETHOD(mmu_page_exists_quick,moea_page_exists_quick),
349 	MMUMETHOD(mmu_page_wired_mappings,moea_page_wired_mappings),
350 	MMUMETHOD(mmu_pinit,		moea_pinit),
351 	MMUMETHOD(mmu_pinit0,		moea_pinit0),
352 	MMUMETHOD(mmu_protect,		moea_protect),
353 	MMUMETHOD(mmu_qenter,		moea_qenter),
354 	MMUMETHOD(mmu_qremove,		moea_qremove),
355 	MMUMETHOD(mmu_release,		moea_release),
356 	MMUMETHOD(mmu_remove,		moea_remove),
357 	MMUMETHOD(mmu_remove_all,      	moea_remove_all),
358 	MMUMETHOD(mmu_remove_write,	moea_remove_write),
359 	MMUMETHOD(mmu_zero_page,       	moea_zero_page),
360 	MMUMETHOD(mmu_zero_page_area,	moea_zero_page_area),
361 	MMUMETHOD(mmu_zero_page_idle,	moea_zero_page_idle),
362 	MMUMETHOD(mmu_activate,		moea_activate),
363 	MMUMETHOD(mmu_deactivate,      	moea_deactivate),
364 
365 	/* Internal interfaces */
366 	MMUMETHOD(mmu_bootstrap,       	moea_bootstrap),
367 	MMUMETHOD(mmu_mapdev,		moea_mapdev),
368 	MMUMETHOD(mmu_unmapdev,		moea_unmapdev),
369 	MMUMETHOD(mmu_kextract,		moea_kextract),
370 	MMUMETHOD(mmu_kenter,		moea_kenter),
371 	MMUMETHOD(mmu_dev_direct_mapped,moea_dev_direct_mapped),
372 	MMUMETHOD(mmu_page_executable,	moea_page_executable),
373 
374 	{ 0, 0 }
375 };
376 
377 static mmu_def_t oea_mmu = {
378 	MMU_TYPE_OEA,
379 	moea_methods,
380 	0
381 };
382 MMU_DEF(oea_mmu);
383 
384 static void
385 tlbie(vm_offset_t va)
386 {
387 
388 	mtx_lock_spin(&tlbie_mtx);
389 	__asm __volatile("tlbie %0" :: "r"(va));
390 	__asm __volatile("tlbsync");
391 	powerpc_sync();
392 	mtx_unlock_spin(&tlbie_mtx);
393 }
394 
395 static void
396 tlbia(void)
397 {
398 	vm_offset_t va;
399 
400 	for (va = 0; va < 0x00040000; va += 0x00001000) {
401 		__asm __volatile("tlbie %0" :: "r"(va));
402 		powerpc_sync();
403 	}
404 	__asm __volatile("tlbsync");
405 	powerpc_sync();
406 }
407 
408 static __inline int
409 va_to_sr(u_int *sr, vm_offset_t va)
410 {
411 	return (sr[(uintptr_t)va >> ADDR_SR_SHFT]);
412 }
413 
414 static __inline u_int
415 va_to_pteg(u_int sr, vm_offset_t addr)
416 {
417 	u_int hash;
418 
419 	hash = (sr & SR_VSID_MASK) ^ (((u_int)addr & ADDR_PIDX) >>
420 	    ADDR_PIDX_SHFT);
421 	return (hash & moea_pteg_mask);
422 }
423 
424 static __inline struct pvo_head *
425 pa_to_pvoh(vm_offset_t pa, vm_page_t *pg_p)
426 {
427 	struct	vm_page *pg;
428 
429 	pg = PHYS_TO_VM_PAGE(pa);
430 
431 	if (pg_p != NULL)
432 		*pg_p = pg;
433 
434 	if (pg == NULL)
435 		return (&moea_pvo_unmanaged);
436 
437 	return (&pg->md.mdpg_pvoh);
438 }
439 
440 static __inline struct pvo_head *
441 vm_page_to_pvoh(vm_page_t m)
442 {
443 
444 	return (&m->md.mdpg_pvoh);
445 }
446 
447 static __inline void
448 moea_attr_clear(vm_page_t m, int ptebit)
449 {
450 
451 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
452 	m->md.mdpg_attrs &= ~ptebit;
453 }
454 
455 static __inline int
456 moea_attr_fetch(vm_page_t m)
457 {
458 
459 	return (m->md.mdpg_attrs);
460 }
461 
462 static __inline void
463 moea_attr_save(vm_page_t m, int ptebit)
464 {
465 
466 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
467 	m->md.mdpg_attrs |= ptebit;
468 }
469 
470 static __inline int
471 moea_pte_compare(const struct pte *pt, const struct pte *pvo_pt)
472 {
473 	if (pt->pte_hi == pvo_pt->pte_hi)
474 		return (1);
475 
476 	return (0);
477 }
478 
479 static __inline int
480 moea_pte_match(struct pte *pt, u_int sr, vm_offset_t va, int which)
481 {
482 	return (pt->pte_hi & ~PTE_VALID) ==
483 	    (((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
484 	    ((va >> ADDR_API_SHFT) & PTE_API) | which);
485 }
486 
487 static __inline void
488 moea_pte_create(struct pte *pt, u_int sr, vm_offset_t va, u_int pte_lo)
489 {
490 
491 	mtx_assert(&moea_table_mutex, MA_OWNED);
492 
493 	/*
494 	 * Construct a PTE.  Default to IMB initially.  Valid bit only gets
495 	 * set when the real pte is set in memory.
496 	 *
497 	 * Note: Don't set the valid bit for correct operation of tlb update.
498 	 */
499 	pt->pte_hi = ((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
500 	    (((va & ADDR_PIDX) >> ADDR_API_SHFT) & PTE_API);
501 	pt->pte_lo = pte_lo;
502 }
503 
504 static __inline void
505 moea_pte_synch(struct pte *pt, struct pte *pvo_pt)
506 {
507 
508 	mtx_assert(&moea_table_mutex, MA_OWNED);
509 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF | PTE_CHG);
510 }
511 
512 static __inline void
513 moea_pte_clear(struct pte *pt, vm_offset_t va, int ptebit)
514 {
515 
516 	mtx_assert(&moea_table_mutex, MA_OWNED);
517 
518 	/*
519 	 * As shown in Section 7.6.3.2.3
520 	 */
521 	pt->pte_lo &= ~ptebit;
522 	tlbie(va);
523 }
524 
525 static __inline void
526 moea_pte_set(struct pte *pt, struct pte *pvo_pt)
527 {
528 
529 	mtx_assert(&moea_table_mutex, MA_OWNED);
530 	pvo_pt->pte_hi |= PTE_VALID;
531 
532 	/*
533 	 * Update the PTE as defined in section 7.6.3.1.
534 	 * Note that the REF/CHG bits are from pvo_pt and thus should havce
535 	 * been saved so this routine can restore them (if desired).
536 	 */
537 	pt->pte_lo = pvo_pt->pte_lo;
538 	powerpc_sync();
539 	pt->pte_hi = pvo_pt->pte_hi;
540 	powerpc_sync();
541 	moea_pte_valid++;
542 }
543 
544 static __inline void
545 moea_pte_unset(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
546 {
547 
548 	mtx_assert(&moea_table_mutex, MA_OWNED);
549 	pvo_pt->pte_hi &= ~PTE_VALID;
550 
551 	/*
552 	 * Force the reg & chg bits back into the PTEs.
553 	 */
554 	powerpc_sync();
555 
556 	/*
557 	 * Invalidate the pte.
558 	 */
559 	pt->pte_hi &= ~PTE_VALID;
560 
561 	tlbie(va);
562 
563 	/*
564 	 * Save the reg & chg bits.
565 	 */
566 	moea_pte_synch(pt, pvo_pt);
567 	moea_pte_valid--;
568 }
569 
570 static __inline void
571 moea_pte_change(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
572 {
573 
574 	/*
575 	 * Invalidate the PTE
576 	 */
577 	moea_pte_unset(pt, pvo_pt, va);
578 	moea_pte_set(pt, pvo_pt);
579 }
580 
581 /*
582  * Quick sort callout for comparing memory regions.
583  */
584 static int	mr_cmp(const void *a, const void *b);
585 static int	om_cmp(const void *a, const void *b);
586 
587 static int
588 mr_cmp(const void *a, const void *b)
589 {
590 	const struct	mem_region *regiona;
591 	const struct	mem_region *regionb;
592 
593 	regiona = a;
594 	regionb = b;
595 	if (regiona->mr_start < regionb->mr_start)
596 		return (-1);
597 	else if (regiona->mr_start > regionb->mr_start)
598 		return (1);
599 	else
600 		return (0);
601 }
602 
603 static int
604 om_cmp(const void *a, const void *b)
605 {
606 	const struct	ofw_map *mapa;
607 	const struct	ofw_map *mapb;
608 
609 	mapa = a;
610 	mapb = b;
611 	if (mapa->om_pa < mapb->om_pa)
612 		return (-1);
613 	else if (mapa->om_pa > mapb->om_pa)
614 		return (1);
615 	else
616 		return (0);
617 }
618 
619 void
620 pmap_cpu_bootstrap(int ap)
621 {
622 	u_int sdr;
623 	int i;
624 
625 	if (ap) {
626 		powerpc_sync();
627 		__asm __volatile("mtdbatu 0,%0" :: "r"(battable[0].batu));
628 		__asm __volatile("mtdbatl 0,%0" :: "r"(battable[0].batl));
629 		isync();
630 		__asm __volatile("mtibatu 0,%0" :: "r"(battable[0].batu));
631 		__asm __volatile("mtibatl 0,%0" :: "r"(battable[0].batl));
632 		isync();
633 	}
634 
635 	__asm __volatile("mtdbatu 1,%0" :: "r"(battable[8].batu));
636 	__asm __volatile("mtdbatl 1,%0" :: "r"(battable[8].batl));
637 	isync();
638 
639 	__asm __volatile("mtibatu 1,%0" :: "r"(0));
640 	__asm __volatile("mtdbatu 2,%0" :: "r"(0));
641 	__asm __volatile("mtibatu 2,%0" :: "r"(0));
642 	__asm __volatile("mtdbatu 3,%0" :: "r"(0));
643 	__asm __volatile("mtibatu 3,%0" :: "r"(0));
644 	isync();
645 
646 	for (i = 0; i < 16; i++)
647 		mtsrin(i << ADDR_SR_SHFT, EMPTY_SEGMENT);
648 
649 	__asm __volatile("mtsr %0,%1" :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
650 	__asm __volatile("mtsr %0,%1" :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
651 	powerpc_sync();
652 
653 	sdr = (u_int)moea_pteg_table | (moea_pteg_mask >> 10);
654 	__asm __volatile("mtsdr1 %0" :: "r"(sdr));
655 	isync();
656 
657 	tlbia();
658 }
659 
660 void
661 moea_bootstrap(mmu_t mmup, vm_offset_t kernelstart, vm_offset_t kernelend)
662 {
663 	ihandle_t	mmui;
664 	phandle_t	chosen, mmu;
665 	int		sz;
666 	int		i, j;
667 	int		ofw_mappings;
668 	vm_size_t	size, physsz, hwphyssz;
669 	vm_offset_t	pa, va, off;
670 
671         /*
672          * Set up BAT0 to map the lowest 256 MB area
673          */
674         battable[0x0].batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
675         battable[0x0].batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
676 
677         /*
678          * Map PCI memory space.
679          */
680         battable[0x8].batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
681         battable[0x8].batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
682 
683         battable[0x9].batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
684         battable[0x9].batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
685 
686         battable[0xa].batl = BATL(0xa0000000, BAT_I|BAT_G, BAT_PP_RW);
687         battable[0xa].batu = BATU(0xa0000000, BAT_BL_256M, BAT_Vs);
688 
689         battable[0xb].batl = BATL(0xb0000000, BAT_I|BAT_G, BAT_PP_RW);
690         battable[0xb].batu = BATU(0xb0000000, BAT_BL_256M, BAT_Vs);
691 
692         /*
693          * Map obio devices.
694          */
695         battable[0xf].batl = BATL(0xf0000000, BAT_I|BAT_G, BAT_PP_RW);
696         battable[0xf].batu = BATU(0xf0000000, BAT_BL_256M, BAT_Vs);
697 
698 	/*
699 	 * Use an IBAT and a DBAT to map the bottom segment of memory
700 	 * where we are.
701 	 */
702 	__asm (".balign 32; \n"
703 	       "mtibatu 0,%0; mtibatl 0,%1; isync; \n"
704 	       "mtdbatu 0,%0; mtdbatl 0,%1; isync"
705 	    :: "r"(battable[0].batu), "r"(battable[0].batl));
706 
707 	/* map pci space */
708 	__asm __volatile("mtdbatu 1,%0" :: "r"(battable[8].batu));
709 	__asm __volatile("mtdbatl 1,%0" :: "r"(battable[8].batl));
710 	isync();
711 
712 	mem_regions(&pregions, &pregions_sz, &regions, &regions_sz);
713 	CTR0(KTR_PMAP, "moea_bootstrap: physical memory");
714 
715 	qsort(pregions, pregions_sz, sizeof(*pregions), mr_cmp);
716 	for (i = 0; i < pregions_sz; i++) {
717 		vm_offset_t pa;
718 		vm_offset_t end;
719 
720 		CTR3(KTR_PMAP, "physregion: %#x - %#x (%#x)",
721 			pregions[i].mr_start,
722 			pregions[i].mr_start + pregions[i].mr_size,
723 			pregions[i].mr_size);
724 		/*
725 		 * Install entries into the BAT table to allow all
726 		 * of physmem to be convered by on-demand BAT entries.
727 		 * The loop will sometimes set the same battable element
728 		 * twice, but that's fine since they won't be used for
729 		 * a while yet.
730 		 */
731 		pa = pregions[i].mr_start & 0xf0000000;
732 		end = pregions[i].mr_start + pregions[i].mr_size;
733 		do {
734                         u_int n = pa >> ADDR_SR_SHFT;
735 
736 			battable[n].batl = BATL(pa, BAT_M, BAT_PP_RW);
737 			battable[n].batu = BATU(pa, BAT_BL_256M, BAT_Vs);
738 			pa += SEGMENT_LENGTH;
739 		} while (pa < end);
740 	}
741 
742 	if (sizeof(phys_avail)/sizeof(phys_avail[0]) < regions_sz)
743 		panic("moea_bootstrap: phys_avail too small");
744 	qsort(regions, regions_sz, sizeof(*regions), mr_cmp);
745 	phys_avail_count = 0;
746 	physsz = 0;
747 	hwphyssz = 0;
748 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
749 	for (i = 0, j = 0; i < regions_sz; i++, j += 2) {
750 		CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start,
751 		    regions[i].mr_start + regions[i].mr_size,
752 		    regions[i].mr_size);
753 		if (hwphyssz != 0 &&
754 		    (physsz + regions[i].mr_size) >= hwphyssz) {
755 			if (physsz < hwphyssz) {
756 				phys_avail[j] = regions[i].mr_start;
757 				phys_avail[j + 1] = regions[i].mr_start +
758 				    hwphyssz - physsz;
759 				physsz = hwphyssz;
760 				phys_avail_count++;
761 			}
762 			break;
763 		}
764 		phys_avail[j] = regions[i].mr_start;
765 		phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size;
766 		phys_avail_count++;
767 		physsz += regions[i].mr_size;
768 	}
769 	physmem = btoc(physsz);
770 
771 	/*
772 	 * Allocate PTEG table.
773 	 */
774 #ifdef PTEGCOUNT
775 	moea_pteg_count = PTEGCOUNT;
776 #else
777 	moea_pteg_count = 0x1000;
778 
779 	while (moea_pteg_count < physmem)
780 		moea_pteg_count <<= 1;
781 
782 	moea_pteg_count >>= 1;
783 #endif /* PTEGCOUNT */
784 
785 	size = moea_pteg_count * sizeof(struct pteg);
786 	CTR2(KTR_PMAP, "moea_bootstrap: %d PTEGs, %d bytes", moea_pteg_count,
787 	    size);
788 	moea_pteg_table = (struct pteg *)moea_bootstrap_alloc(size, size);
789 	CTR1(KTR_PMAP, "moea_bootstrap: PTEG table at %p", moea_pteg_table);
790 	bzero((void *)moea_pteg_table, moea_pteg_count * sizeof(struct pteg));
791 	moea_pteg_mask = moea_pteg_count - 1;
792 
793 	/*
794 	 * Allocate pv/overflow lists.
795 	 */
796 	size = sizeof(struct pvo_head) * moea_pteg_count;
797 	moea_pvo_table = (struct pvo_head *)moea_bootstrap_alloc(size,
798 	    PAGE_SIZE);
799 	CTR1(KTR_PMAP, "moea_bootstrap: PVO table at %p", moea_pvo_table);
800 	for (i = 0; i < moea_pteg_count; i++)
801 		LIST_INIT(&moea_pvo_table[i]);
802 
803 	/*
804 	 * Initialize the lock that synchronizes access to the pteg and pvo
805 	 * tables.
806 	 */
807 	mtx_init(&moea_table_mutex, "pmap table", NULL, MTX_DEF |
808 	    MTX_RECURSE);
809 
810 	mtx_init(&tlbie_mtx, "tlbie", NULL, MTX_SPIN);
811 
812 	/*
813 	 * Initialise the unmanaged pvo pool.
814 	 */
815 	moea_bpvo_pool = (struct pvo_entry *)moea_bootstrap_alloc(
816 		BPVO_POOL_SIZE*sizeof(struct pvo_entry), 0);
817 	moea_bpvo_pool_index = 0;
818 
819 	/*
820 	 * Make sure kernel vsid is allocated as well as VSID 0.
821 	 */
822 	moea_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW]
823 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
824 	moea_vsid_bitmap[0] |= 1;
825 
826 	/*
827 	 * Set up the Open Firmware pmap and add it's mappings.
828 	 */
829 	moea_pinit(mmup, &ofw_pmap);
830 	ofw_pmap.pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
831 	ofw_pmap.pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT;
832 	if ((chosen = OF_finddevice("/chosen")) == -1)
833 		panic("moea_bootstrap: can't find /chosen");
834 	OF_getprop(chosen, "mmu", &mmui, 4);
835 	if ((mmu = OF_instance_to_package(mmui)) == -1)
836 		panic("moea_bootstrap: can't get mmu package");
837 	if ((sz = OF_getproplen(mmu, "translations")) == -1)
838 		panic("moea_bootstrap: can't get ofw translation count");
839 	translations = NULL;
840 	for (i = 0; phys_avail[i] != 0; i += 2) {
841 		if (phys_avail[i + 1] >= sz) {
842 			translations = (struct ofw_map *)phys_avail[i];
843 			break;
844 		}
845 	}
846 	if (translations == NULL)
847 		panic("moea_bootstrap: no space to copy translations");
848 	bzero(translations, sz);
849 	if (OF_getprop(mmu, "translations", translations, sz) == -1)
850 		panic("moea_bootstrap: can't get ofw translations");
851 	CTR0(KTR_PMAP, "moea_bootstrap: translations");
852 	sz /= sizeof(*translations);
853 	qsort(translations, sz, sizeof (*translations), om_cmp);
854 	for (i = 0, ofw_mappings = 0; i < sz; i++) {
855 		CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x",
856 		    translations[i].om_pa, translations[i].om_va,
857 		    translations[i].om_len);
858 
859 		/*
860 		 * If the mapping is 1:1, let the RAM and device on-demand
861 		 * BAT tables take care of the translation.
862 		 */
863 		if (translations[i].om_va == translations[i].om_pa)
864 			continue;
865 
866 		/* Enter the pages */
867 		for (off = 0; off < translations[i].om_len; off += PAGE_SIZE) {
868 			struct	vm_page m;
869 
870 			m.phys_addr = translations[i].om_pa + off;
871 			PMAP_LOCK(&ofw_pmap);
872 			moea_enter_locked(&ofw_pmap,
873 				   translations[i].om_va + off, &m,
874 				   VM_PROT_ALL, 1);
875 			PMAP_UNLOCK(&ofw_pmap);
876 			ofw_mappings++;
877 		}
878 	}
879 
880 	/*
881 	 * Calculate the last available physical address.
882 	 */
883 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
884 		;
885 	Maxmem = powerpc_btop(phys_avail[i + 1]);
886 
887 	/*
888 	 * Initialize the kernel pmap (which is statically allocated).
889 	 */
890 	PMAP_LOCK_INIT(kernel_pmap);
891 	for (i = 0; i < 16; i++) {
892 		kernel_pmap->pm_sr[i] = EMPTY_SEGMENT;
893 	}
894 	kernel_pmap->pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
895 	kernel_pmap->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT;
896 	kernel_pmap->pm_active = ~0;
897 
898 	pmap_cpu_bootstrap(0);
899 
900 	pmap_bootstrapped++;
901 
902 	/*
903 	 * Set the start and end of kva.
904 	 */
905 	virtual_avail = VM_MIN_KERNEL_ADDRESS;
906 	virtual_end = VM_MAX_KERNEL_ADDRESS;
907 
908 	/*
909 	 * Allocate a kernel stack with a guard page for thread0 and map it
910 	 * into the kernel page map.
911 	 */
912 	pa = moea_bootstrap_alloc(KSTACK_PAGES * PAGE_SIZE, PAGE_SIZE);
913 	va = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
914 	virtual_avail = va + KSTACK_PAGES * PAGE_SIZE;
915 	CTR2(KTR_PMAP, "moea_bootstrap: kstack0 at %#x (%#x)", pa, va);
916 	thread0.td_kstack = va;
917 	thread0.td_kstack_pages = KSTACK_PAGES;
918 	for (i = 0; i < KSTACK_PAGES; i++) {
919 		moea_kenter(mmup, va, pa);;
920 		pa += PAGE_SIZE;
921 		va += PAGE_SIZE;
922 	}
923 
924 	/*
925 	 * Allocate virtual address space for the message buffer.
926 	 */
927 	pa = msgbuf_phys = moea_bootstrap_alloc(MSGBUF_SIZE, PAGE_SIZE);
928 	msgbufp = (struct msgbuf *)virtual_avail;
929 	va = virtual_avail;
930 	virtual_avail += round_page(MSGBUF_SIZE);
931 	while (va < virtual_avail) {
932 		moea_kenter(mmup, va, pa);;
933 		pa += PAGE_SIZE;
934 		va += PAGE_SIZE;
935 	}
936 }
937 
938 /*
939  * Activate a user pmap.  The pmap must be activated before it's address
940  * space can be accessed in any way.
941  */
942 void
943 moea_activate(mmu_t mmu, struct thread *td)
944 {
945 	pmap_t	pm, pmr;
946 
947 	/*
948 	 * Load all the data we need up front to encourage the compiler to
949 	 * not issue any loads while we have interrupts disabled below.
950 	 */
951 	pm = &td->td_proc->p_vmspace->vm_pmap;
952 	pmr = pm->pmap_phys;
953 
954 	pm->pm_active |= PCPU_GET(cpumask);
955 	PCPU_SET(curpmap, pmr);
956 }
957 
958 void
959 moea_deactivate(mmu_t mmu, struct thread *td)
960 {
961 	pmap_t	pm;
962 
963 	pm = &td->td_proc->p_vmspace->vm_pmap;
964 	pm->pm_active &= ~PCPU_GET(cpumask);
965 	PCPU_SET(curpmap, NULL);
966 }
967 
968 void
969 moea_change_wiring(mmu_t mmu, pmap_t pm, vm_offset_t va, boolean_t wired)
970 {
971 	struct	pvo_entry *pvo;
972 
973 	PMAP_LOCK(pm);
974 	pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
975 
976 	if (pvo != NULL) {
977 		if (wired) {
978 			if ((pvo->pvo_vaddr & PVO_WIRED) == 0)
979 				pm->pm_stats.wired_count++;
980 			pvo->pvo_vaddr |= PVO_WIRED;
981 		} else {
982 			if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
983 				pm->pm_stats.wired_count--;
984 			pvo->pvo_vaddr &= ~PVO_WIRED;
985 		}
986 	}
987 	PMAP_UNLOCK(pm);
988 }
989 
990 void
991 moea_copy_page(mmu_t mmu, vm_page_t msrc, vm_page_t mdst)
992 {
993 	vm_offset_t	dst;
994 	vm_offset_t	src;
995 
996 	dst = VM_PAGE_TO_PHYS(mdst);
997 	src = VM_PAGE_TO_PHYS(msrc);
998 
999 	kcopy((void *)src, (void *)dst, PAGE_SIZE);
1000 }
1001 
1002 /*
1003  * Zero a page of physical memory by temporarily mapping it into the tlb.
1004  */
1005 void
1006 moea_zero_page(mmu_t mmu, vm_page_t m)
1007 {
1008 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1009 	void *va = (void *)pa;
1010 
1011 	bzero(va, PAGE_SIZE);
1012 }
1013 
1014 void
1015 moea_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
1016 {
1017 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1018 	void *va = (void *)(pa + off);
1019 
1020 	bzero(va, size);
1021 }
1022 
1023 void
1024 moea_zero_page_idle(mmu_t mmu, vm_page_t m)
1025 {
1026 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1027 	void *va = (void *)pa;
1028 
1029 	bzero(va, PAGE_SIZE);
1030 }
1031 
1032 /*
1033  * Map the given physical page at the specified virtual address in the
1034  * target pmap with the protection requested.  If specified the page
1035  * will be wired down.
1036  */
1037 void
1038 moea_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1039 	   boolean_t wired)
1040 {
1041 
1042 	vm_page_lock_queues();
1043 	PMAP_LOCK(pmap);
1044 	moea_enter_locked(pmap, va, m, prot, wired);
1045 	vm_page_unlock_queues();
1046 	PMAP_UNLOCK(pmap);
1047 }
1048 
1049 /*
1050  * Map the given physical page at the specified virtual address in the
1051  * target pmap with the protection requested.  If specified the page
1052  * will be wired down.
1053  *
1054  * The page queues and pmap must be locked.
1055  */
1056 static void
1057 moea_enter_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1058     boolean_t wired)
1059 {
1060 	struct		pvo_head *pvo_head;
1061 	uma_zone_t	zone;
1062 	vm_page_t	pg;
1063 	u_int		pte_lo, pvo_flags, was_exec, i;
1064 	int		error;
1065 
1066 	if (!moea_initialized) {
1067 		pvo_head = &moea_pvo_kunmanaged;
1068 		zone = moea_upvo_zone;
1069 		pvo_flags = 0;
1070 		pg = NULL;
1071 		was_exec = PTE_EXEC;
1072 	} else {
1073 		pvo_head = vm_page_to_pvoh(m);
1074 		pg = m;
1075 		zone = moea_mpvo_zone;
1076 		pvo_flags = PVO_MANAGED;
1077 		was_exec = 0;
1078 	}
1079 	if (pmap_bootstrapped)
1080 		mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1081 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1082 
1083 	/* XXX change the pvo head for fake pages */
1084 	if ((m->flags & PG_FICTITIOUS) == PG_FICTITIOUS)
1085 		pvo_head = &moea_pvo_kunmanaged;
1086 
1087 	/*
1088 	 * If this is a managed page, and it's the first reference to the page,
1089 	 * clear the execness of the page.  Otherwise fetch the execness.
1090 	 */
1091 	if ((pg != NULL) && ((m->flags & PG_FICTITIOUS) == 0)) {
1092 		if (LIST_EMPTY(pvo_head)) {
1093 			moea_attr_clear(pg, PTE_EXEC);
1094 		} else {
1095 			was_exec = moea_attr_fetch(pg) & PTE_EXEC;
1096 		}
1097 	}
1098 
1099 	/*
1100 	 * Assume the page is cache inhibited and access is guarded unless
1101 	 * it's in our available memory array.
1102 	 */
1103 	pte_lo = PTE_I | PTE_G;
1104 	for (i = 0; i < pregions_sz; i++) {
1105 		if ((VM_PAGE_TO_PHYS(m) >= pregions[i].mr_start) &&
1106 		    (VM_PAGE_TO_PHYS(m) <
1107 			(pregions[i].mr_start + pregions[i].mr_size))) {
1108 			pte_lo = PTE_M;
1109 			break;
1110 		}
1111 	}
1112 
1113 	if (prot & VM_PROT_WRITE) {
1114 		pte_lo |= PTE_BW;
1115 		if (pmap_bootstrapped)
1116 			vm_page_flag_set(m, PG_WRITEABLE);
1117 	} else
1118 		pte_lo |= PTE_BR;
1119 
1120 	if (prot & VM_PROT_EXECUTE)
1121 		pvo_flags |= PVO_EXECUTABLE;
1122 
1123 	if (wired)
1124 		pvo_flags |= PVO_WIRED;
1125 
1126 	if ((m->flags & PG_FICTITIOUS) != 0)
1127 		pvo_flags |= PVO_FAKE;
1128 
1129 	error = moea_pvo_enter(pmap, zone, pvo_head, va, VM_PAGE_TO_PHYS(m),
1130 	    pte_lo, pvo_flags);
1131 
1132 	/*
1133 	 * Flush the real page from the instruction cache if this page is
1134 	 * mapped executable and cacheable and was not previously mapped (or
1135 	 * was not mapped executable).
1136 	 */
1137 	if (error == 0 && (pvo_flags & PVO_EXECUTABLE) &&
1138 	    (pte_lo & PTE_I) == 0 && was_exec == 0) {
1139 		/*
1140 		 * Flush the real memory from the cache.
1141 		 */
1142 		moea_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1143 		if (pg != NULL)
1144 			moea_attr_save(pg, PTE_EXEC);
1145 	}
1146 
1147 	/* XXX syncicache always until problems are sorted */
1148 	moea_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1149 }
1150 
1151 /*
1152  * Maps a sequence of resident pages belonging to the same object.
1153  * The sequence begins with the given page m_start.  This page is
1154  * mapped at the given virtual address start.  Each subsequent page is
1155  * mapped at a virtual address that is offset from start by the same
1156  * amount as the page is offset from m_start within the object.  The
1157  * last page in the sequence is the page with the largest offset from
1158  * m_start that can be mapped at a virtual address less than the given
1159  * virtual address end.  Not every virtual page between start and end
1160  * is mapped; only those for which a resident page exists with the
1161  * corresponding offset from m_start are mapped.
1162  */
1163 void
1164 moea_enter_object(mmu_t mmu, pmap_t pm, vm_offset_t start, vm_offset_t end,
1165     vm_page_t m_start, vm_prot_t prot)
1166 {
1167 	vm_page_t m;
1168 	vm_pindex_t diff, psize;
1169 
1170 	psize = atop(end - start);
1171 	m = m_start;
1172 	PMAP_LOCK(pm);
1173 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1174 		moea_enter_locked(pm, start + ptoa(diff), m, prot &
1175 		    (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1176 		m = TAILQ_NEXT(m, listq);
1177 	}
1178 	PMAP_UNLOCK(pm);
1179 }
1180 
1181 void
1182 moea_enter_quick(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_page_t m,
1183     vm_prot_t prot)
1184 {
1185 
1186 	PMAP_LOCK(pm);
1187 	moea_enter_locked(pm, va, m, prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1188 	    FALSE);
1189 	PMAP_UNLOCK(pm);
1190 
1191 }
1192 
1193 vm_paddr_t
1194 moea_extract(mmu_t mmu, pmap_t pm, vm_offset_t va)
1195 {
1196 	struct	pvo_entry *pvo;
1197 	vm_paddr_t pa;
1198 
1199 	PMAP_LOCK(pm);
1200 	pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1201 	if (pvo == NULL)
1202 		pa = 0;
1203 	else
1204 		pa = (pvo->pvo_pte.pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF);
1205 	PMAP_UNLOCK(pm);
1206 	return (pa);
1207 }
1208 
1209 /*
1210  * Atomically extract and hold the physical page with the given
1211  * pmap and virtual address pair if that mapping permits the given
1212  * protection.
1213  */
1214 vm_page_t
1215 moea_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_prot_t prot)
1216 {
1217 	struct	pvo_entry *pvo;
1218 	vm_page_t m;
1219 
1220 	m = NULL;
1221 	vm_page_lock_queues();
1222 	PMAP_LOCK(pmap);
1223 	pvo = moea_pvo_find_va(pmap, va & ~ADDR_POFF, NULL);
1224 	if (pvo != NULL && (pvo->pvo_pte.pte.pte_hi & PTE_VALID) &&
1225 	    ((pvo->pvo_pte.pte.pte_lo & PTE_PP) == PTE_RW ||
1226 	     (prot & VM_PROT_WRITE) == 0)) {
1227 		m = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte.pte_lo & PTE_RPGN);
1228 		vm_page_hold(m);
1229 	}
1230 	vm_page_unlock_queues();
1231 	PMAP_UNLOCK(pmap);
1232 	return (m);
1233 }
1234 
1235 void
1236 moea_init(mmu_t mmu)
1237 {
1238 
1239 	moea_upvo_zone = uma_zcreate("UPVO entry", sizeof (struct pvo_entry),
1240 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1241 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1242 	moea_mpvo_zone = uma_zcreate("MPVO entry", sizeof(struct pvo_entry),
1243 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1244 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1245 	moea_initialized = TRUE;
1246 }
1247 
1248 boolean_t
1249 moea_is_modified(mmu_t mmu, vm_page_t m)
1250 {
1251 
1252 	if ((m->flags & (PG_FICTITIOUS |PG_UNMANAGED)) != 0)
1253 		return (FALSE);
1254 
1255 	return (moea_query_bit(m, PTE_CHG));
1256 }
1257 
1258 void
1259 moea_clear_reference(mmu_t mmu, vm_page_t m)
1260 {
1261 
1262 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1263 		return;
1264 	moea_clear_bit(m, PTE_REF, NULL);
1265 }
1266 
1267 void
1268 moea_clear_modify(mmu_t mmu, vm_page_t m)
1269 {
1270 
1271 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1272 		return;
1273 	moea_clear_bit(m, PTE_CHG, NULL);
1274 }
1275 
1276 /*
1277  * Clear the write and modified bits in each of the given page's mappings.
1278  */
1279 void
1280 moea_remove_write(mmu_t mmu, vm_page_t m)
1281 {
1282 	struct	pvo_entry *pvo;
1283 	struct	pte *pt;
1284 	pmap_t	pmap;
1285 	u_int	lo;
1286 
1287 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1288 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1289 	    (m->flags & PG_WRITEABLE) == 0)
1290 		return;
1291 	lo = moea_attr_fetch(m);
1292 	powerpc_sync();
1293 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1294 		pmap = pvo->pvo_pmap;
1295 		PMAP_LOCK(pmap);
1296 		if ((pvo->pvo_pte.pte.pte_lo & PTE_PP) != PTE_BR) {
1297 			pt = moea_pvo_to_pte(pvo, -1);
1298 			pvo->pvo_pte.pte.pte_lo &= ~PTE_PP;
1299 			pvo->pvo_pte.pte.pte_lo |= PTE_BR;
1300 			if (pt != NULL) {
1301 				moea_pte_synch(pt, &pvo->pvo_pte.pte);
1302 				lo |= pvo->pvo_pte.pte.pte_lo;
1303 				pvo->pvo_pte.pte.pte_lo &= ~PTE_CHG;
1304 				moea_pte_change(pt, &pvo->pvo_pte.pte,
1305 				    pvo->pvo_vaddr);
1306 				mtx_unlock(&moea_table_mutex);
1307 			}
1308 		}
1309 		PMAP_UNLOCK(pmap);
1310 	}
1311 	if ((lo & PTE_CHG) != 0) {
1312 		moea_attr_clear(m, PTE_CHG);
1313 		vm_page_dirty(m);
1314 	}
1315 	vm_page_flag_clear(m, PG_WRITEABLE);
1316 }
1317 
1318 /*
1319  *	moea_ts_referenced:
1320  *
1321  *	Return a count of reference bits for a page, clearing those bits.
1322  *	It is not necessary for every reference bit to be cleared, but it
1323  *	is necessary that 0 only be returned when there are truly no
1324  *	reference bits set.
1325  *
1326  *	XXX: The exact number of bits to check and clear is a matter that
1327  *	should be tested and standardized at some point in the future for
1328  *	optimal aging of shared pages.
1329  */
1330 boolean_t
1331 moea_ts_referenced(mmu_t mmu, vm_page_t m)
1332 {
1333 	int count;
1334 
1335 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1336 		return (0);
1337 
1338 	count = moea_clear_bit(m, PTE_REF, NULL);
1339 
1340 	return (count);
1341 }
1342 
1343 /*
1344  * Map a wired page into kernel virtual address space.
1345  */
1346 void
1347 moea_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1348 {
1349 	u_int		pte_lo;
1350 	int		error;
1351 	int		i;
1352 
1353 #if 0
1354 	if (va < VM_MIN_KERNEL_ADDRESS)
1355 		panic("moea_kenter: attempt to enter non-kernel address %#x",
1356 		    va);
1357 #endif
1358 
1359 	pte_lo = PTE_I | PTE_G;
1360 	for (i = 0; i < pregions_sz; i++) {
1361 		if ((pa >= pregions[i].mr_start) &&
1362 		    (pa < (pregions[i].mr_start + pregions[i].mr_size))) {
1363 			pte_lo = PTE_M;
1364 			break;
1365 		}
1366 	}
1367 
1368 	PMAP_LOCK(kernel_pmap);
1369 	error = moea_pvo_enter(kernel_pmap, moea_upvo_zone,
1370 	    &moea_pvo_kunmanaged, va, pa, pte_lo, PVO_WIRED);
1371 
1372 	if (error != 0 && error != ENOENT)
1373 		panic("moea_kenter: failed to enter va %#x pa %#x: %d", va,
1374 		    pa, error);
1375 
1376 	/*
1377 	 * Flush the real memory from the instruction cache.
1378 	 */
1379 	if ((pte_lo & (PTE_I | PTE_G)) == 0) {
1380 		moea_syncicache(pa, PAGE_SIZE);
1381 	}
1382 	PMAP_UNLOCK(kernel_pmap);
1383 }
1384 
1385 /*
1386  * Extract the physical page address associated with the given kernel virtual
1387  * address.
1388  */
1389 vm_offset_t
1390 moea_kextract(mmu_t mmu, vm_offset_t va)
1391 {
1392 	struct		pvo_entry *pvo;
1393 	vm_paddr_t pa;
1394 
1395 	/*
1396 	 * Allow direct mappings on 32-bit OEA
1397 	 */
1398 	if (va < VM_MIN_KERNEL_ADDRESS) {
1399 		return (va);
1400 	}
1401 
1402 	PMAP_LOCK(kernel_pmap);
1403 	pvo = moea_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL);
1404 	KASSERT(pvo != NULL, ("moea_kextract: no addr found"));
1405 	pa = (pvo->pvo_pte.pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF);
1406 	PMAP_UNLOCK(kernel_pmap);
1407 	return (pa);
1408 }
1409 
1410 /*
1411  * Remove a wired page from kernel virtual address space.
1412  */
1413 void
1414 moea_kremove(mmu_t mmu, vm_offset_t va)
1415 {
1416 
1417 	moea_remove(mmu, kernel_pmap, va, va + PAGE_SIZE);
1418 }
1419 
1420 /*
1421  * Map a range of physical addresses into kernel virtual address space.
1422  *
1423  * The value passed in *virt is a suggested virtual address for the mapping.
1424  * Architectures which can support a direct-mapped physical to virtual region
1425  * can return the appropriate address within that region, leaving '*virt'
1426  * unchanged.  We cannot and therefore do not; *virt is updated with the
1427  * first usable address after the mapped region.
1428  */
1429 vm_offset_t
1430 moea_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1431     vm_offset_t pa_end, int prot)
1432 {
1433 	vm_offset_t	sva, va;
1434 
1435 	sva = *virt;
1436 	va = sva;
1437 	for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE)
1438 		moea_kenter(mmu, va, pa_start);
1439 	*virt = va;
1440 	return (sva);
1441 }
1442 
1443 /*
1444  * Returns true if the pmap's pv is one of the first
1445  * 16 pvs linked to from this page.  This count may
1446  * be changed upwards or downwards in the future; it
1447  * is only necessary that true be returned for a small
1448  * subset of pmaps for proper page aging.
1449  */
1450 boolean_t
1451 moea_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
1452 {
1453         int loops;
1454 	struct pvo_entry *pvo;
1455 
1456         if (!moea_initialized || (m->flags & PG_FICTITIOUS))
1457                 return FALSE;
1458 
1459 	loops = 0;
1460 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1461 		if (pvo->pvo_pmap == pmap)
1462 			return (TRUE);
1463 		if (++loops >= 16)
1464 			break;
1465 	}
1466 
1467 	return (FALSE);
1468 }
1469 
1470 /*
1471  * Return the number of managed mappings to the given physical page
1472  * that are wired.
1473  */
1474 int
1475 moea_page_wired_mappings(mmu_t mmu, vm_page_t m)
1476 {
1477 	struct pvo_entry *pvo;
1478 	int count;
1479 
1480 	count = 0;
1481 	if (!moea_initialized || (m->flags & PG_FICTITIOUS) != 0)
1482 		return (count);
1483 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1484 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink)
1485 		if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
1486 			count++;
1487 	return (count);
1488 }
1489 
1490 static u_int	moea_vsidcontext;
1491 
1492 void
1493 moea_pinit(mmu_t mmu, pmap_t pmap)
1494 {
1495 	int	i, mask;
1496 	u_int	entropy;
1497 
1498 	KASSERT((int)pmap < VM_MIN_KERNEL_ADDRESS, ("moea_pinit: virt pmap"));
1499 	PMAP_LOCK_INIT(pmap);
1500 
1501 	entropy = 0;
1502 	__asm __volatile("mftb %0" : "=r"(entropy));
1503 
1504 	if ((pmap->pmap_phys = (pmap_t)moea_kextract(mmu, (vm_offset_t)pmap))
1505 	    == NULL) {
1506 		pmap->pmap_phys = pmap;
1507 	}
1508 
1509 
1510 	/*
1511 	 * Allocate some segment registers for this pmap.
1512 	 */
1513 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1514 		u_int	hash, n;
1515 
1516 		/*
1517 		 * Create a new value by mutiplying by a prime and adding in
1518 		 * entropy from the timebase register.  This is to make the
1519 		 * VSID more random so that the PT hash function collides
1520 		 * less often.  (Note that the prime casues gcc to do shifts
1521 		 * instead of a multiply.)
1522 		 */
1523 		moea_vsidcontext = (moea_vsidcontext * 0x1105) + entropy;
1524 		hash = moea_vsidcontext & (NPMAPS - 1);
1525 		if (hash == 0)		/* 0 is special, avoid it */
1526 			continue;
1527 		n = hash >> 5;
1528 		mask = 1 << (hash & (VSID_NBPW - 1));
1529 		hash = (moea_vsidcontext & 0xfffff);
1530 		if (moea_vsid_bitmap[n] & mask) {	/* collision? */
1531 			/* anything free in this bucket? */
1532 			if (moea_vsid_bitmap[n] == 0xffffffff) {
1533 				entropy = (moea_vsidcontext >> 20);
1534 				continue;
1535 			}
1536 			i = ffs(~moea_vsid_bitmap[i]) - 1;
1537 			mask = 1 << i;
1538 			hash &= 0xfffff & ~(VSID_NBPW - 1);
1539 			hash |= i;
1540 		}
1541 		moea_vsid_bitmap[n] |= mask;
1542 		for (i = 0; i < 16; i++)
1543 			pmap->pm_sr[i] = VSID_MAKE(i, hash);
1544 		return;
1545 	}
1546 
1547 	panic("moea_pinit: out of segments");
1548 }
1549 
1550 /*
1551  * Initialize the pmap associated with process 0.
1552  */
1553 void
1554 moea_pinit0(mmu_t mmu, pmap_t pm)
1555 {
1556 
1557 	moea_pinit(mmu, pm);
1558 	bzero(&pm->pm_stats, sizeof(pm->pm_stats));
1559 }
1560 
1561 /*
1562  * Set the physical protection on the specified range of this map as requested.
1563  */
1564 void
1565 moea_protect(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva,
1566     vm_prot_t prot)
1567 {
1568 	struct	pvo_entry *pvo;
1569 	struct	pte *pt;
1570 	int	pteidx;
1571 
1572 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1573 	    ("moea_protect: non current pmap"));
1574 
1575 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1576 		moea_remove(mmu, pm, sva, eva);
1577 		return;
1578 	}
1579 
1580 	vm_page_lock_queues();
1581 	PMAP_LOCK(pm);
1582 	for (; sva < eva; sva += PAGE_SIZE) {
1583 		pvo = moea_pvo_find_va(pm, sva, &pteidx);
1584 		if (pvo == NULL)
1585 			continue;
1586 
1587 		if ((prot & VM_PROT_EXECUTE) == 0)
1588 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1589 
1590 		/*
1591 		 * Grab the PTE pointer before we diddle with the cached PTE
1592 		 * copy.
1593 		 */
1594 		pt = moea_pvo_to_pte(pvo, pteidx);
1595 		/*
1596 		 * Change the protection of the page.
1597 		 */
1598 		pvo->pvo_pte.pte.pte_lo &= ~PTE_PP;
1599 		pvo->pvo_pte.pte.pte_lo |= PTE_BR;
1600 
1601 		/*
1602 		 * If the PVO is in the page table, update that pte as well.
1603 		 */
1604 		if (pt != NULL) {
1605 			moea_pte_change(pt, &pvo->pvo_pte.pte, pvo->pvo_vaddr);
1606 			mtx_unlock(&moea_table_mutex);
1607 		}
1608 	}
1609 	vm_page_unlock_queues();
1610 	PMAP_UNLOCK(pm);
1611 }
1612 
1613 /*
1614  * Map a list of wired pages into kernel virtual address space.  This is
1615  * intended for temporary mappings which do not need page modification or
1616  * references recorded.  Existing mappings in the region are overwritten.
1617  */
1618 void
1619 moea_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1620 {
1621 	vm_offset_t va;
1622 
1623 	va = sva;
1624 	while (count-- > 0) {
1625 		moea_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1626 		va += PAGE_SIZE;
1627 		m++;
1628 	}
1629 }
1630 
1631 /*
1632  * Remove page mappings from kernel virtual address space.  Intended for
1633  * temporary mappings entered by moea_qenter.
1634  */
1635 void
1636 moea_qremove(mmu_t mmu, vm_offset_t sva, int count)
1637 {
1638 	vm_offset_t va;
1639 
1640 	va = sva;
1641 	while (count-- > 0) {
1642 		moea_kremove(mmu, va);
1643 		va += PAGE_SIZE;
1644 	}
1645 }
1646 
1647 void
1648 moea_release(mmu_t mmu, pmap_t pmap)
1649 {
1650         int idx, mask;
1651 
1652 	/*
1653 	 * Free segment register's VSID
1654 	 */
1655         if (pmap->pm_sr[0] == 0)
1656                 panic("moea_release");
1657 
1658         idx = VSID_TO_HASH(pmap->pm_sr[0]) & (NPMAPS-1);
1659         mask = 1 << (idx % VSID_NBPW);
1660         idx /= VSID_NBPW;
1661         moea_vsid_bitmap[idx] &= ~mask;
1662 	PMAP_LOCK_DESTROY(pmap);
1663 }
1664 
1665 /*
1666  * Remove the given range of addresses from the specified map.
1667  */
1668 void
1669 moea_remove(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1670 {
1671 	struct	pvo_entry *pvo;
1672 	int	pteidx;
1673 
1674 	vm_page_lock_queues();
1675 	PMAP_LOCK(pm);
1676 	for (; sva < eva; sva += PAGE_SIZE) {
1677 		pvo = moea_pvo_find_va(pm, sva, &pteidx);
1678 		if (pvo != NULL) {
1679 			moea_pvo_remove(pvo, pteidx);
1680 		}
1681 	}
1682 	PMAP_UNLOCK(pm);
1683 	vm_page_unlock_queues();
1684 }
1685 
1686 /*
1687  * Remove physical page from all pmaps in which it resides. moea_pvo_remove()
1688  * will reflect changes in pte's back to the vm_page.
1689  */
1690 void
1691 moea_remove_all(mmu_t mmu, vm_page_t m)
1692 {
1693 	struct  pvo_head *pvo_head;
1694 	struct	pvo_entry *pvo, *next_pvo;
1695 	pmap_t	pmap;
1696 
1697 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1698 
1699 	pvo_head = vm_page_to_pvoh(m);
1700 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1701 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1702 
1703 		MOEA_PVO_CHECK(pvo);	/* sanity check */
1704 		pmap = pvo->pvo_pmap;
1705 		PMAP_LOCK(pmap);
1706 		moea_pvo_remove(pvo, -1);
1707 		PMAP_UNLOCK(pmap);
1708 	}
1709 	vm_page_flag_clear(m, PG_WRITEABLE);
1710 }
1711 
1712 /*
1713  * Allocate a physical page of memory directly from the phys_avail map.
1714  * Can only be called from moea_bootstrap before avail start and end are
1715  * calculated.
1716  */
1717 static vm_offset_t
1718 moea_bootstrap_alloc(vm_size_t size, u_int align)
1719 {
1720 	vm_offset_t	s, e;
1721 	int		i, j;
1722 
1723 	size = round_page(size);
1724 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1725 		if (align != 0)
1726 			s = (phys_avail[i] + align - 1) & ~(align - 1);
1727 		else
1728 			s = phys_avail[i];
1729 		e = s + size;
1730 
1731 		if (s < phys_avail[i] || e > phys_avail[i + 1])
1732 			continue;
1733 
1734 		if (s == phys_avail[i]) {
1735 			phys_avail[i] += size;
1736 		} else if (e == phys_avail[i + 1]) {
1737 			phys_avail[i + 1] -= size;
1738 		} else {
1739 			for (j = phys_avail_count * 2; j > i; j -= 2) {
1740 				phys_avail[j] = phys_avail[j - 2];
1741 				phys_avail[j + 1] = phys_avail[j - 1];
1742 			}
1743 
1744 			phys_avail[i + 3] = phys_avail[i + 1];
1745 			phys_avail[i + 1] = s;
1746 			phys_avail[i + 2] = e;
1747 			phys_avail_count++;
1748 		}
1749 
1750 		return (s);
1751 	}
1752 	panic("moea_bootstrap_alloc: could not allocate memory");
1753 }
1754 
1755 static void
1756 moea_syncicache(vm_offset_t pa, vm_size_t len)
1757 {
1758 	__syncicache((void *)pa, len);
1759 }
1760 
1761 static int
1762 moea_pvo_enter(pmap_t pm, uma_zone_t zone, struct pvo_head *pvo_head,
1763     vm_offset_t va, vm_offset_t pa, u_int pte_lo, int flags)
1764 {
1765 	struct	pvo_entry *pvo;
1766 	u_int	sr;
1767 	int	first;
1768 	u_int	ptegidx;
1769 	int	i;
1770 	int     bootstrap;
1771 
1772 	moea_pvo_enter_calls++;
1773 	first = 0;
1774 	bootstrap = 0;
1775 
1776 	/*
1777 	 * Compute the PTE Group index.
1778 	 */
1779 	va &= ~ADDR_POFF;
1780 	sr = va_to_sr(pm->pm_sr, va);
1781 	ptegidx = va_to_pteg(sr, va);
1782 
1783 	/*
1784 	 * Remove any existing mapping for this page.  Reuse the pvo entry if
1785 	 * there is a mapping.
1786 	 */
1787 	mtx_lock(&moea_table_mutex);
1788 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
1789 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1790 			if ((pvo->pvo_pte.pte.pte_lo & PTE_RPGN) == pa &&
1791 			    (pvo->pvo_pte.pte.pte_lo & PTE_PP) ==
1792 			    (pte_lo & PTE_PP)) {
1793 				mtx_unlock(&moea_table_mutex);
1794 				return (0);
1795 			}
1796 			moea_pvo_remove(pvo, -1);
1797 			break;
1798 		}
1799 	}
1800 
1801 	/*
1802 	 * If we aren't overwriting a mapping, try to allocate.
1803 	 */
1804 	if (moea_initialized) {
1805 		pvo = uma_zalloc(zone, M_NOWAIT);
1806 	} else {
1807 		if (moea_bpvo_pool_index >= BPVO_POOL_SIZE) {
1808 			panic("moea_enter: bpvo pool exhausted, %d, %d, %d",
1809 			      moea_bpvo_pool_index, BPVO_POOL_SIZE,
1810 			      BPVO_POOL_SIZE * sizeof(struct pvo_entry));
1811 		}
1812 		pvo = &moea_bpvo_pool[moea_bpvo_pool_index];
1813 		moea_bpvo_pool_index++;
1814 		bootstrap = 1;
1815 	}
1816 
1817 	if (pvo == NULL) {
1818 		mtx_unlock(&moea_table_mutex);
1819 		return (ENOMEM);
1820 	}
1821 
1822 	moea_pvo_entries++;
1823 	pvo->pvo_vaddr = va;
1824 	pvo->pvo_pmap = pm;
1825 	LIST_INSERT_HEAD(&moea_pvo_table[ptegidx], pvo, pvo_olink);
1826 	pvo->pvo_vaddr &= ~ADDR_POFF;
1827 	if (flags & VM_PROT_EXECUTE)
1828 		pvo->pvo_vaddr |= PVO_EXECUTABLE;
1829 	if (flags & PVO_WIRED)
1830 		pvo->pvo_vaddr |= PVO_WIRED;
1831 	if (pvo_head != &moea_pvo_kunmanaged)
1832 		pvo->pvo_vaddr |= PVO_MANAGED;
1833 	if (bootstrap)
1834 		pvo->pvo_vaddr |= PVO_BOOTSTRAP;
1835 	if (flags & PVO_FAKE)
1836 		pvo->pvo_vaddr |= PVO_FAKE;
1837 
1838 	moea_pte_create(&pvo->pvo_pte.pte, sr, va, pa | pte_lo);
1839 
1840 	/*
1841 	 * Remember if the list was empty and therefore will be the first
1842 	 * item.
1843 	 */
1844 	if (LIST_FIRST(pvo_head) == NULL)
1845 		first = 1;
1846 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1847 
1848 	if (pvo->pvo_pte.pte.pte_lo & PVO_WIRED)
1849 		pm->pm_stats.wired_count++;
1850 	pm->pm_stats.resident_count++;
1851 
1852 	/*
1853 	 * We hope this succeeds but it isn't required.
1854 	 */
1855 	i = moea_pte_insert(ptegidx, &pvo->pvo_pte.pte);
1856 	if (i >= 0) {
1857 		PVO_PTEGIDX_SET(pvo, i);
1858 	} else {
1859 		panic("moea_pvo_enter: overflow");
1860 		moea_pte_overflow++;
1861 	}
1862 	mtx_unlock(&moea_table_mutex);
1863 
1864 	return (first ? ENOENT : 0);
1865 }
1866 
1867 static void
1868 moea_pvo_remove(struct pvo_entry *pvo, int pteidx)
1869 {
1870 	struct	pte *pt;
1871 
1872 	/*
1873 	 * If there is an active pte entry, we need to deactivate it (and
1874 	 * save the ref & cfg bits).
1875 	 */
1876 	pt = moea_pvo_to_pte(pvo, pteidx);
1877 	if (pt != NULL) {
1878 		moea_pte_unset(pt, &pvo->pvo_pte.pte, pvo->pvo_vaddr);
1879 		mtx_unlock(&moea_table_mutex);
1880 		PVO_PTEGIDX_CLR(pvo);
1881 	} else {
1882 		moea_pte_overflow--;
1883 	}
1884 
1885 	/*
1886 	 * Update our statistics.
1887 	 */
1888 	pvo->pvo_pmap->pm_stats.resident_count--;
1889 	if (pvo->pvo_pte.pte.pte_lo & PVO_WIRED)
1890 		pvo->pvo_pmap->pm_stats.wired_count--;
1891 
1892 	/*
1893 	 * Save the REF/CHG bits into their cache if the page is managed.
1894 	 */
1895 	if ((pvo->pvo_vaddr & (PVO_MANAGED|PVO_FAKE)) == PVO_MANAGED) {
1896 		struct	vm_page *pg;
1897 
1898 		pg = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte.pte_lo & PTE_RPGN);
1899 		if (pg != NULL) {
1900 			moea_attr_save(pg, pvo->pvo_pte.pte.pte_lo &
1901 			    (PTE_REF | PTE_CHG));
1902 		}
1903 	}
1904 
1905 	/*
1906 	 * Remove this PVO from the PV list.
1907 	 */
1908 	LIST_REMOVE(pvo, pvo_vlink);
1909 
1910 	/*
1911 	 * Remove this from the overflow list and return it to the pool
1912 	 * if we aren't going to reuse it.
1913 	 */
1914 	LIST_REMOVE(pvo, pvo_olink);
1915 	if (!(pvo->pvo_vaddr & PVO_BOOTSTRAP))
1916 		uma_zfree(pvo->pvo_vaddr & PVO_MANAGED ? moea_mpvo_zone :
1917 		    moea_upvo_zone, pvo);
1918 	moea_pvo_entries--;
1919 	moea_pvo_remove_calls++;
1920 }
1921 
1922 static __inline int
1923 moea_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1924 {
1925 	int	pteidx;
1926 
1927 	/*
1928 	 * We can find the actual pte entry without searching by grabbing
1929 	 * the PTEG index from 3 unused bits in pte_lo[11:9] and by
1930 	 * noticing the HID bit.
1931 	 */
1932 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1933 	if (pvo->pvo_pte.pte.pte_hi & PTE_HID)
1934 		pteidx ^= moea_pteg_mask * 8;
1935 
1936 	return (pteidx);
1937 }
1938 
1939 static struct pvo_entry *
1940 moea_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p)
1941 {
1942 	struct	pvo_entry *pvo;
1943 	int	ptegidx;
1944 	u_int	sr;
1945 
1946 	va &= ~ADDR_POFF;
1947 	sr = va_to_sr(pm->pm_sr, va);
1948 	ptegidx = va_to_pteg(sr, va);
1949 
1950 	mtx_lock(&moea_table_mutex);
1951 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
1952 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1953 			if (pteidx_p)
1954 				*pteidx_p = moea_pvo_pte_index(pvo, ptegidx);
1955 			break;
1956 		}
1957 	}
1958 	mtx_unlock(&moea_table_mutex);
1959 
1960 	return (pvo);
1961 }
1962 
1963 static struct pte *
1964 moea_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1965 {
1966 	struct	pte *pt;
1967 
1968 	/*
1969 	 * If we haven't been supplied the ptegidx, calculate it.
1970 	 */
1971 	if (pteidx == -1) {
1972 		int	ptegidx;
1973 		u_int	sr;
1974 
1975 		sr = va_to_sr(pvo->pvo_pmap->pm_sr, pvo->pvo_vaddr);
1976 		ptegidx = va_to_pteg(sr, pvo->pvo_vaddr);
1977 		pteidx = moea_pvo_pte_index(pvo, ptegidx);
1978 	}
1979 
1980 	pt = &moea_pteg_table[pteidx >> 3].pt[pteidx & 7];
1981 	mtx_lock(&moea_table_mutex);
1982 
1983 	if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1984 		panic("moea_pvo_to_pte: pvo %p has valid pte in pvo but no "
1985 		    "valid pte index", pvo);
1986 	}
1987 
1988 	if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1989 		panic("moea_pvo_to_pte: pvo %p has valid pte index in pvo "
1990 		    "pvo but no valid pte", pvo);
1991 	}
1992 
1993 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1994 		if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) == 0) {
1995 			panic("moea_pvo_to_pte: pvo %p has valid pte in "
1996 			    "moea_pteg_table %p but invalid in pvo", pvo, pt);
1997 		}
1998 
1999 		if (((pt->pte_lo ^ pvo->pvo_pte.pte.pte_lo) & ~(PTE_CHG|PTE_REF))
2000 		    != 0) {
2001 			panic("moea_pvo_to_pte: pvo %p pte does not match "
2002 			    "pte %p in moea_pteg_table", pvo, pt);
2003 		}
2004 
2005 		mtx_assert(&moea_table_mutex, MA_OWNED);
2006 		return (pt);
2007 	}
2008 
2009 	if (pvo->pvo_pte.pte.pte_hi & PTE_VALID) {
2010 		panic("moea_pvo_to_pte: pvo %p has invalid pte %p in "
2011 		    "moea_pteg_table but valid in pvo", pvo, pt);
2012 	}
2013 
2014 	mtx_unlock(&moea_table_mutex);
2015 	return (NULL);
2016 }
2017 
2018 /*
2019  * XXX: THIS STUFF SHOULD BE IN pte.c?
2020  */
2021 int
2022 moea_pte_spill(vm_offset_t addr)
2023 {
2024 	struct	pvo_entry *source_pvo, *victim_pvo;
2025 	struct	pvo_entry *pvo;
2026 	int	ptegidx, i, j;
2027 	u_int	sr;
2028 	struct	pteg *pteg;
2029 	struct	pte *pt;
2030 
2031 	moea_pte_spills++;
2032 
2033 	sr = mfsrin(addr);
2034 	ptegidx = va_to_pteg(sr, addr);
2035 
2036 	/*
2037 	 * Have to substitute some entry.  Use the primary hash for this.
2038 	 * Use low bits of timebase as random generator.
2039 	 */
2040 	pteg = &moea_pteg_table[ptegidx];
2041 	mtx_lock(&moea_table_mutex);
2042 	__asm __volatile("mftb %0" : "=r"(i));
2043 	i &= 7;
2044 	pt = &pteg->pt[i];
2045 
2046 	source_pvo = NULL;
2047 	victim_pvo = NULL;
2048 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
2049 		/*
2050 		 * We need to find a pvo entry for this address.
2051 		 */
2052 		MOEA_PVO_CHECK(pvo);
2053 		if (source_pvo == NULL &&
2054 		    moea_pte_match(&pvo->pvo_pte.pte, sr, addr,
2055 		    pvo->pvo_pte.pte.pte_hi & PTE_HID)) {
2056 			/*
2057 			 * Now found an entry to be spilled into the pteg.
2058 			 * The PTE is now valid, so we know it's active.
2059 			 */
2060 			j = moea_pte_insert(ptegidx, &pvo->pvo_pte.pte);
2061 
2062 			if (j >= 0) {
2063 				PVO_PTEGIDX_SET(pvo, j);
2064 				moea_pte_overflow--;
2065 				MOEA_PVO_CHECK(pvo);
2066 				mtx_unlock(&moea_table_mutex);
2067 				return (1);
2068 			}
2069 
2070 			source_pvo = pvo;
2071 
2072 			if (victim_pvo != NULL)
2073 				break;
2074 		}
2075 
2076 		/*
2077 		 * We also need the pvo entry of the victim we are replacing
2078 		 * so save the R & C bits of the PTE.
2079 		 */
2080 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
2081 		    moea_pte_compare(pt, &pvo->pvo_pte.pte)) {
2082 			victim_pvo = pvo;
2083 			if (source_pvo != NULL)
2084 				break;
2085 		}
2086 	}
2087 
2088 	if (source_pvo == NULL) {
2089 		mtx_unlock(&moea_table_mutex);
2090 		return (0);
2091 	}
2092 
2093 	if (victim_pvo == NULL) {
2094 		if ((pt->pte_hi & PTE_HID) == 0)
2095 			panic("moea_pte_spill: victim p-pte (%p) has no pvo"
2096 			    "entry", pt);
2097 
2098 		/*
2099 		 * If this is a secondary PTE, we need to search it's primary
2100 		 * pvo bucket for the matching PVO.
2101 		 */
2102 		LIST_FOREACH(pvo, &moea_pvo_table[ptegidx ^ moea_pteg_mask],
2103 		    pvo_olink) {
2104 			MOEA_PVO_CHECK(pvo);
2105 			/*
2106 			 * We also need the pvo entry of the victim we are
2107 			 * replacing so save the R & C bits of the PTE.
2108 			 */
2109 			if (moea_pte_compare(pt, &pvo->pvo_pte.pte)) {
2110 				victim_pvo = pvo;
2111 				break;
2112 			}
2113 		}
2114 
2115 		if (victim_pvo == NULL)
2116 			panic("moea_pte_spill: victim s-pte (%p) has no pvo"
2117 			    "entry", pt);
2118 	}
2119 
2120 	/*
2121 	 * We are invalidating the TLB entry for the EA we are replacing even
2122 	 * though it's valid.  If we don't, we lose any ref/chg bit changes
2123 	 * contained in the TLB entry.
2124 	 */
2125 	source_pvo->pvo_pte.pte.pte_hi &= ~PTE_HID;
2126 
2127 	moea_pte_unset(pt, &victim_pvo->pvo_pte.pte, victim_pvo->pvo_vaddr);
2128 	moea_pte_set(pt, &source_pvo->pvo_pte.pte);
2129 
2130 	PVO_PTEGIDX_CLR(victim_pvo);
2131 	PVO_PTEGIDX_SET(source_pvo, i);
2132 	moea_pte_replacements++;
2133 
2134 	MOEA_PVO_CHECK(victim_pvo);
2135 	MOEA_PVO_CHECK(source_pvo);
2136 
2137 	mtx_unlock(&moea_table_mutex);
2138 	return (1);
2139 }
2140 
2141 static int
2142 moea_pte_insert(u_int ptegidx, struct pte *pvo_pt)
2143 {
2144 	struct	pte *pt;
2145 	int	i;
2146 
2147 	mtx_assert(&moea_table_mutex, MA_OWNED);
2148 
2149 	/*
2150 	 * First try primary hash.
2151 	 */
2152 	for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2153 		if ((pt->pte_hi & PTE_VALID) == 0) {
2154 			pvo_pt->pte_hi &= ~PTE_HID;
2155 			moea_pte_set(pt, pvo_pt);
2156 			return (i);
2157 		}
2158 	}
2159 
2160 	/*
2161 	 * Now try secondary hash.
2162 	 */
2163 	ptegidx ^= moea_pteg_mask;
2164 
2165 	for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2166 		if ((pt->pte_hi & PTE_VALID) == 0) {
2167 			pvo_pt->pte_hi |= PTE_HID;
2168 			moea_pte_set(pt, pvo_pt);
2169 			return (i);
2170 		}
2171 	}
2172 
2173 	panic("moea_pte_insert: overflow");
2174 	return (-1);
2175 }
2176 
2177 static boolean_t
2178 moea_query_bit(vm_page_t m, int ptebit)
2179 {
2180 	struct	pvo_entry *pvo;
2181 	struct	pte *pt;
2182 
2183 #if 0
2184 	if (moea_attr_fetch(m) & ptebit)
2185 		return (TRUE);
2186 #endif
2187 
2188 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2189 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2190 
2191 		/*
2192 		 * See if we saved the bit off.  If so, cache it and return
2193 		 * success.
2194 		 */
2195 		if (pvo->pvo_pte.pte.pte_lo & ptebit) {
2196 			moea_attr_save(m, ptebit);
2197 			MOEA_PVO_CHECK(pvo);	/* sanity check */
2198 			return (TRUE);
2199 		}
2200 	}
2201 
2202 	/*
2203 	 * No luck, now go through the hard part of looking at the PTEs
2204 	 * themselves.  Sync so that any pending REF/CHG bits are flushed to
2205 	 * the PTEs.
2206 	 */
2207 	powerpc_sync();
2208 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2209 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2210 
2211 		/*
2212 		 * See if this pvo has a valid PTE.  if so, fetch the
2213 		 * REF/CHG bits from the valid PTE.  If the appropriate
2214 		 * ptebit is set, cache it and return success.
2215 		 */
2216 		pt = moea_pvo_to_pte(pvo, -1);
2217 		if (pt != NULL) {
2218 			moea_pte_synch(pt, &pvo->pvo_pte.pte);
2219 			mtx_unlock(&moea_table_mutex);
2220 			if (pvo->pvo_pte.pte.pte_lo & ptebit) {
2221 				moea_attr_save(m, ptebit);
2222 				MOEA_PVO_CHECK(pvo);	/* sanity check */
2223 				return (TRUE);
2224 			}
2225 		}
2226 	}
2227 
2228 	return (FALSE);
2229 }
2230 
2231 static u_int
2232 moea_clear_bit(vm_page_t m, int ptebit, int *origbit)
2233 {
2234 	u_int	count;
2235 	struct	pvo_entry *pvo;
2236 	struct	pte *pt;
2237 	int	rv;
2238 
2239 	/*
2240 	 * Clear the cached value.
2241 	 */
2242 	rv = moea_attr_fetch(m);
2243 	moea_attr_clear(m, ptebit);
2244 
2245 	/*
2246 	 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so
2247 	 * we can reset the right ones).  note that since the pvo entries and
2248 	 * list heads are accessed via BAT0 and are never placed in the page
2249 	 * table, we don't have to worry about further accesses setting the
2250 	 * REF/CHG bits.
2251 	 */
2252 	powerpc_sync();
2253 
2254 	/*
2255 	 * For each pvo entry, clear the pvo's ptebit.  If this pvo has a
2256 	 * valid pte clear the ptebit from the valid pte.
2257 	 */
2258 	count = 0;
2259 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2260 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2261 		pt = moea_pvo_to_pte(pvo, -1);
2262 		if (pt != NULL) {
2263 			moea_pte_synch(pt, &pvo->pvo_pte.pte);
2264 			if (pvo->pvo_pte.pte.pte_lo & ptebit) {
2265 				count++;
2266 				moea_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2267 			}
2268 			mtx_unlock(&moea_table_mutex);
2269 		}
2270 		rv |= pvo->pvo_pte.pte.pte_lo;
2271 		pvo->pvo_pte.pte.pte_lo &= ~ptebit;
2272 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2273 	}
2274 
2275 	if (origbit != NULL) {
2276 		*origbit = rv;
2277 	}
2278 
2279 	return (count);
2280 }
2281 
2282 /*
2283  * Return true if the physical range is encompassed by the battable[idx]
2284  */
2285 static int
2286 moea_bat_mapped(int idx, vm_offset_t pa, vm_size_t size)
2287 {
2288 	u_int prot;
2289 	u_int32_t start;
2290 	u_int32_t end;
2291 	u_int32_t bat_ble;
2292 
2293 	/*
2294 	 * Return immediately if not a valid mapping
2295 	 */
2296 	if (!battable[idx].batu & BAT_Vs)
2297 		return (EINVAL);
2298 
2299 	/*
2300 	 * The BAT entry must be cache-inhibited, guarded, and r/w
2301 	 * so it can function as an i/o page
2302 	 */
2303 	prot = battable[idx].batl & (BAT_I|BAT_G|BAT_PP_RW);
2304 	if (prot != (BAT_I|BAT_G|BAT_PP_RW))
2305 		return (EPERM);
2306 
2307 	/*
2308 	 * The address should be within the BAT range. Assume that the
2309 	 * start address in the BAT has the correct alignment (thus
2310 	 * not requiring masking)
2311 	 */
2312 	start = battable[idx].batl & BAT_PBS;
2313 	bat_ble = (battable[idx].batu & ~(BAT_EBS)) | 0x03;
2314 	end = start | (bat_ble << 15) | 0x7fff;
2315 
2316 	if ((pa < start) || ((pa + size) > end))
2317 		return (ERANGE);
2318 
2319 	return (0);
2320 }
2321 
2322 boolean_t
2323 moea_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2324 {
2325 	int i;
2326 
2327 	/*
2328 	 * This currently does not work for entries that
2329 	 * overlap 256M BAT segments.
2330 	 */
2331 
2332 	for(i = 0; i < 16; i++)
2333 		if (moea_bat_mapped(i, pa, size) == 0)
2334 			return (0);
2335 
2336 	return (EFAULT);
2337 }
2338 
2339 boolean_t
2340 moea_page_executable(mmu_t mmu, vm_page_t pg)
2341 {
2342 	return ((moea_attr_fetch(pg) & PTE_EXEC) == PTE_EXEC);
2343 }
2344 
2345 /*
2346  * Map a set of physical memory pages into the kernel virtual
2347  * address space. Return a pointer to where it is mapped. This
2348  * routine is intended to be used for mapping device memory,
2349  * NOT real memory.
2350  */
2351 void *
2352 moea_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2353 {
2354 	vm_offset_t va, tmpva, ppa, offset;
2355 	int i;
2356 
2357 	ppa = trunc_page(pa);
2358 	offset = pa & PAGE_MASK;
2359 	size = roundup(offset + size, PAGE_SIZE);
2360 
2361 	GIANT_REQUIRED;
2362 
2363 	/*
2364 	 * If the physical address lies within a valid BAT table entry,
2365 	 * return the 1:1 mapping. This currently doesn't work
2366 	 * for regions that overlap 256M BAT segments.
2367 	 */
2368 	for (i = 0; i < 16; i++) {
2369 		if (moea_bat_mapped(i, pa, size) == 0)
2370 			return ((void *) pa);
2371 	}
2372 
2373 	va = kmem_alloc_nofault(kernel_map, size);
2374 	if (!va)
2375 		panic("moea_mapdev: Couldn't alloc kernel virtual memory");
2376 
2377 	for (tmpva = va; size > 0;) {
2378 		moea_kenter(mmu, tmpva, ppa);
2379 		tlbie(tmpva);
2380 		size -= PAGE_SIZE;
2381 		tmpva += PAGE_SIZE;
2382 		ppa += PAGE_SIZE;
2383 	}
2384 
2385 	return ((void *)(va + offset));
2386 }
2387 
2388 void
2389 moea_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2390 {
2391 	vm_offset_t base, offset;
2392 
2393 	/*
2394 	 * If this is outside kernel virtual space, then it's a
2395 	 * battable entry and doesn't require unmapping
2396 	 */
2397 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2398 		base = trunc_page(va);
2399 		offset = va & PAGE_MASK;
2400 		size = roundup(offset + size, PAGE_SIZE);
2401 		kmem_free(kernel_map, base, size);
2402 	}
2403 }
2404