xref: /freebsd/sys/powerpc/aim/mmu_oea.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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/platform.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_cpu_bootstrap(mmu_t, int);
327 void moea_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
328 void *moea_mapdev(mmu_t, vm_offset_t, vm_size_t);
329 void moea_unmapdev(mmu_t, vm_offset_t, vm_size_t);
330 vm_offset_t moea_kextract(mmu_t, vm_offset_t);
331 void moea_kenter(mmu_t, vm_offset_t, vm_offset_t);
332 boolean_t moea_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
333 boolean_t moea_page_executable(mmu_t, vm_page_t);
334 
335 static mmu_method_t moea_methods[] = {
336 	MMUMETHOD(mmu_change_wiring,	moea_change_wiring),
337 	MMUMETHOD(mmu_clear_modify,	moea_clear_modify),
338 	MMUMETHOD(mmu_clear_reference,	moea_clear_reference),
339 	MMUMETHOD(mmu_copy_page,	moea_copy_page),
340 	MMUMETHOD(mmu_enter,		moea_enter),
341 	MMUMETHOD(mmu_enter_object,	moea_enter_object),
342 	MMUMETHOD(mmu_enter_quick,	moea_enter_quick),
343 	MMUMETHOD(mmu_extract,		moea_extract),
344 	MMUMETHOD(mmu_extract_and_hold,	moea_extract_and_hold),
345 	MMUMETHOD(mmu_init,		moea_init),
346 	MMUMETHOD(mmu_is_modified,	moea_is_modified),
347 	MMUMETHOD(mmu_ts_referenced,	moea_ts_referenced),
348 	MMUMETHOD(mmu_map,     		moea_map),
349 	MMUMETHOD(mmu_page_exists_quick,moea_page_exists_quick),
350 	MMUMETHOD(mmu_page_wired_mappings,moea_page_wired_mappings),
351 	MMUMETHOD(mmu_pinit,		moea_pinit),
352 	MMUMETHOD(mmu_pinit0,		moea_pinit0),
353 	MMUMETHOD(mmu_protect,		moea_protect),
354 	MMUMETHOD(mmu_qenter,		moea_qenter),
355 	MMUMETHOD(mmu_qremove,		moea_qremove),
356 	MMUMETHOD(mmu_release,		moea_release),
357 	MMUMETHOD(mmu_remove,		moea_remove),
358 	MMUMETHOD(mmu_remove_all,      	moea_remove_all),
359 	MMUMETHOD(mmu_remove_write,	moea_remove_write),
360 	MMUMETHOD(mmu_zero_page,       	moea_zero_page),
361 	MMUMETHOD(mmu_zero_page_area,	moea_zero_page_area),
362 	MMUMETHOD(mmu_zero_page_idle,	moea_zero_page_idle),
363 	MMUMETHOD(mmu_activate,		moea_activate),
364 	MMUMETHOD(mmu_deactivate,      	moea_deactivate),
365 
366 	/* Internal interfaces */
367 	MMUMETHOD(mmu_bootstrap,       	moea_bootstrap),
368 	MMUMETHOD(mmu_cpu_bootstrap,   	moea_cpu_bootstrap),
369 	MMUMETHOD(mmu_mapdev,		moea_mapdev),
370 	MMUMETHOD(mmu_unmapdev,		moea_unmapdev),
371 	MMUMETHOD(mmu_kextract,		moea_kextract),
372 	MMUMETHOD(mmu_kenter,		moea_kenter),
373 	MMUMETHOD(mmu_dev_direct_mapped,moea_dev_direct_mapped),
374 	MMUMETHOD(mmu_page_executable,	moea_page_executable),
375 
376 	{ 0, 0 }
377 };
378 
379 static mmu_def_t oea_mmu = {
380 	MMU_TYPE_OEA,
381 	moea_methods,
382 	0
383 };
384 MMU_DEF(oea_mmu);
385 
386 static void
387 tlbie(vm_offset_t va)
388 {
389 
390 	mtx_lock_spin(&tlbie_mtx);
391 	__asm __volatile("tlbie %0" :: "r"(va));
392 	__asm __volatile("tlbsync");
393 	powerpc_sync();
394 	mtx_unlock_spin(&tlbie_mtx);
395 }
396 
397 static void
398 tlbia(void)
399 {
400 	vm_offset_t va;
401 
402 	for (va = 0; va < 0x00040000; va += 0x00001000) {
403 		__asm __volatile("tlbie %0" :: "r"(va));
404 		powerpc_sync();
405 	}
406 	__asm __volatile("tlbsync");
407 	powerpc_sync();
408 }
409 
410 static __inline int
411 va_to_sr(u_int *sr, vm_offset_t va)
412 {
413 	return (sr[(uintptr_t)va >> ADDR_SR_SHFT]);
414 }
415 
416 static __inline u_int
417 va_to_pteg(u_int sr, vm_offset_t addr)
418 {
419 	u_int hash;
420 
421 	hash = (sr & SR_VSID_MASK) ^ (((u_int)addr & ADDR_PIDX) >>
422 	    ADDR_PIDX_SHFT);
423 	return (hash & moea_pteg_mask);
424 }
425 
426 static __inline struct pvo_head *
427 pa_to_pvoh(vm_offset_t pa, vm_page_t *pg_p)
428 {
429 	struct	vm_page *pg;
430 
431 	pg = PHYS_TO_VM_PAGE(pa);
432 
433 	if (pg_p != NULL)
434 		*pg_p = pg;
435 
436 	if (pg == NULL)
437 		return (&moea_pvo_unmanaged);
438 
439 	return (&pg->md.mdpg_pvoh);
440 }
441 
442 static __inline struct pvo_head *
443 vm_page_to_pvoh(vm_page_t m)
444 {
445 
446 	return (&m->md.mdpg_pvoh);
447 }
448 
449 static __inline void
450 moea_attr_clear(vm_page_t m, int ptebit)
451 {
452 
453 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
454 	m->md.mdpg_attrs &= ~ptebit;
455 }
456 
457 static __inline int
458 moea_attr_fetch(vm_page_t m)
459 {
460 
461 	return (m->md.mdpg_attrs);
462 }
463 
464 static __inline void
465 moea_attr_save(vm_page_t m, int ptebit)
466 {
467 
468 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
469 	m->md.mdpg_attrs |= ptebit;
470 }
471 
472 static __inline int
473 moea_pte_compare(const struct pte *pt, const struct pte *pvo_pt)
474 {
475 	if (pt->pte_hi == pvo_pt->pte_hi)
476 		return (1);
477 
478 	return (0);
479 }
480 
481 static __inline int
482 moea_pte_match(struct pte *pt, u_int sr, vm_offset_t va, int which)
483 {
484 	return (pt->pte_hi & ~PTE_VALID) ==
485 	    (((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
486 	    ((va >> ADDR_API_SHFT) & PTE_API) | which);
487 }
488 
489 static __inline void
490 moea_pte_create(struct pte *pt, u_int sr, vm_offset_t va, u_int pte_lo)
491 {
492 
493 	mtx_assert(&moea_table_mutex, MA_OWNED);
494 
495 	/*
496 	 * Construct a PTE.  Default to IMB initially.  Valid bit only gets
497 	 * set when the real pte is set in memory.
498 	 *
499 	 * Note: Don't set the valid bit for correct operation of tlb update.
500 	 */
501 	pt->pte_hi = ((sr & SR_VSID_MASK) << PTE_VSID_SHFT) |
502 	    (((va & ADDR_PIDX) >> ADDR_API_SHFT) & PTE_API);
503 	pt->pte_lo = pte_lo;
504 }
505 
506 static __inline void
507 moea_pte_synch(struct pte *pt, struct pte *pvo_pt)
508 {
509 
510 	mtx_assert(&moea_table_mutex, MA_OWNED);
511 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF | PTE_CHG);
512 }
513 
514 static __inline void
515 moea_pte_clear(struct pte *pt, vm_offset_t va, int ptebit)
516 {
517 
518 	mtx_assert(&moea_table_mutex, MA_OWNED);
519 
520 	/*
521 	 * As shown in Section 7.6.3.2.3
522 	 */
523 	pt->pte_lo &= ~ptebit;
524 	tlbie(va);
525 }
526 
527 static __inline void
528 moea_pte_set(struct pte *pt, struct pte *pvo_pt)
529 {
530 
531 	mtx_assert(&moea_table_mutex, MA_OWNED);
532 	pvo_pt->pte_hi |= PTE_VALID;
533 
534 	/*
535 	 * Update the PTE as defined in section 7.6.3.1.
536 	 * Note that the REF/CHG bits are from pvo_pt and thus should havce
537 	 * been saved so this routine can restore them (if desired).
538 	 */
539 	pt->pte_lo = pvo_pt->pte_lo;
540 	powerpc_sync();
541 	pt->pte_hi = pvo_pt->pte_hi;
542 	powerpc_sync();
543 	moea_pte_valid++;
544 }
545 
546 static __inline void
547 moea_pte_unset(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
548 {
549 
550 	mtx_assert(&moea_table_mutex, MA_OWNED);
551 	pvo_pt->pte_hi &= ~PTE_VALID;
552 
553 	/*
554 	 * Force the reg & chg bits back into the PTEs.
555 	 */
556 	powerpc_sync();
557 
558 	/*
559 	 * Invalidate the pte.
560 	 */
561 	pt->pte_hi &= ~PTE_VALID;
562 
563 	tlbie(va);
564 
565 	/*
566 	 * Save the reg & chg bits.
567 	 */
568 	moea_pte_synch(pt, pvo_pt);
569 	moea_pte_valid--;
570 }
571 
572 static __inline void
573 moea_pte_change(struct pte *pt, struct pte *pvo_pt, vm_offset_t va)
574 {
575 
576 	/*
577 	 * Invalidate the PTE
578 	 */
579 	moea_pte_unset(pt, pvo_pt, va);
580 	moea_pte_set(pt, pvo_pt);
581 }
582 
583 /*
584  * Quick sort callout for comparing memory regions.
585  */
586 static int	mr_cmp(const void *a, const void *b);
587 static int	om_cmp(const void *a, const void *b);
588 
589 static int
590 mr_cmp(const void *a, const void *b)
591 {
592 	const struct	mem_region *regiona;
593 	const struct	mem_region *regionb;
594 
595 	regiona = a;
596 	regionb = b;
597 	if (regiona->mr_start < regionb->mr_start)
598 		return (-1);
599 	else if (regiona->mr_start > regionb->mr_start)
600 		return (1);
601 	else
602 		return (0);
603 }
604 
605 static int
606 om_cmp(const void *a, const void *b)
607 {
608 	const struct	ofw_map *mapa;
609 	const struct	ofw_map *mapb;
610 
611 	mapa = a;
612 	mapb = b;
613 	if (mapa->om_pa < mapb->om_pa)
614 		return (-1);
615 	else if (mapa->om_pa > mapb->om_pa)
616 		return (1);
617 	else
618 		return (0);
619 }
620 
621 void
622 moea_cpu_bootstrap(mmu_t mmup, int ap)
623 {
624 	u_int sdr;
625 	int i;
626 
627 	if (ap) {
628 		powerpc_sync();
629 		__asm __volatile("mtdbatu 0,%0" :: "r"(battable[0].batu));
630 		__asm __volatile("mtdbatl 0,%0" :: "r"(battable[0].batl));
631 		isync();
632 		__asm __volatile("mtibatu 0,%0" :: "r"(battable[0].batu));
633 		__asm __volatile("mtibatl 0,%0" :: "r"(battable[0].batl));
634 		isync();
635 	}
636 
637 	__asm __volatile("mtdbatu 1,%0" :: "r"(battable[8].batu));
638 	__asm __volatile("mtdbatl 1,%0" :: "r"(battable[8].batl));
639 	isync();
640 
641 	__asm __volatile("mtibatu 1,%0" :: "r"(0));
642 	__asm __volatile("mtdbatu 2,%0" :: "r"(0));
643 	__asm __volatile("mtibatu 2,%0" :: "r"(0));
644 	__asm __volatile("mtdbatu 3,%0" :: "r"(0));
645 	__asm __volatile("mtibatu 3,%0" :: "r"(0));
646 	isync();
647 
648 	for (i = 0; i < 16; i++)
649 		mtsrin(i << ADDR_SR_SHFT, EMPTY_SEGMENT);
650 
651 	__asm __volatile("mtsr %0,%1" :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
652 	__asm __volatile("mtsr %0,%1" :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
653 	powerpc_sync();
654 
655 	sdr = (u_int)moea_pteg_table | (moea_pteg_mask >> 10);
656 	__asm __volatile("mtsdr1 %0" :: "r"(sdr));
657 	isync();
658 
659 	tlbia();
660 }
661 
662 void
663 moea_bootstrap(mmu_t mmup, vm_offset_t kernelstart, vm_offset_t kernelend)
664 {
665 	ihandle_t	mmui;
666 	phandle_t	chosen, mmu;
667 	int		sz;
668 	int		i, j;
669 	int		ofw_mappings;
670 	vm_size_t	size, physsz, hwphyssz;
671 	vm_offset_t	pa, va, off;
672 	void		*dpcpu;
673 
674         /*
675          * Set up BAT0 to map the lowest 256 MB area
676          */
677         battable[0x0].batl = BATL(0x00000000, BAT_M, BAT_PP_RW);
678         battable[0x0].batu = BATU(0x00000000, BAT_BL_256M, BAT_Vs);
679 
680         /*
681          * Map PCI memory space.
682          */
683         battable[0x8].batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW);
684         battable[0x8].batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs);
685 
686         battable[0x9].batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW);
687         battable[0x9].batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs);
688 
689         battable[0xa].batl = BATL(0xa0000000, BAT_I|BAT_G, BAT_PP_RW);
690         battable[0xa].batu = BATU(0xa0000000, BAT_BL_256M, BAT_Vs);
691 
692         battable[0xb].batl = BATL(0xb0000000, BAT_I|BAT_G, BAT_PP_RW);
693         battable[0xb].batu = BATU(0xb0000000, BAT_BL_256M, BAT_Vs);
694 
695         /*
696          * Map obio devices.
697          */
698         battable[0xf].batl = BATL(0xf0000000, BAT_I|BAT_G, BAT_PP_RW);
699         battable[0xf].batu = BATU(0xf0000000, BAT_BL_256M, BAT_Vs);
700 
701 	/*
702 	 * Use an IBAT and a DBAT to map the bottom segment of memory
703 	 * where we are.
704 	 */
705 	__asm (".balign 32; \n"
706 	       "mtibatu 0,%0; mtibatl 0,%1; isync; \n"
707 	       "mtdbatu 0,%0; mtdbatl 0,%1; isync"
708 	    :: "r"(battable[0].batu), "r"(battable[0].batl));
709 
710 	/* map pci space */
711 	__asm __volatile("mtdbatu 1,%0" :: "r"(battable[8].batu));
712 	__asm __volatile("mtdbatl 1,%0" :: "r"(battable[8].batl));
713 	isync();
714 
715 	/* set global direct map flag */
716 	hw_direct_map = 1;
717 
718 	mem_regions(&pregions, &pregions_sz, &regions, &regions_sz);
719 	CTR0(KTR_PMAP, "moea_bootstrap: physical memory");
720 
721 	qsort(pregions, pregions_sz, sizeof(*pregions), mr_cmp);
722 	for (i = 0; i < pregions_sz; i++) {
723 		vm_offset_t pa;
724 		vm_offset_t end;
725 
726 		CTR3(KTR_PMAP, "physregion: %#x - %#x (%#x)",
727 			pregions[i].mr_start,
728 			pregions[i].mr_start + pregions[i].mr_size,
729 			pregions[i].mr_size);
730 		/*
731 		 * Install entries into the BAT table to allow all
732 		 * of physmem to be convered by on-demand BAT entries.
733 		 * The loop will sometimes set the same battable element
734 		 * twice, but that's fine since they won't be used for
735 		 * a while yet.
736 		 */
737 		pa = pregions[i].mr_start & 0xf0000000;
738 		end = pregions[i].mr_start + pregions[i].mr_size;
739 		do {
740                         u_int n = pa >> ADDR_SR_SHFT;
741 
742 			battable[n].batl = BATL(pa, BAT_M, BAT_PP_RW);
743 			battable[n].batu = BATU(pa, BAT_BL_256M, BAT_Vs);
744 			pa += SEGMENT_LENGTH;
745 		} while (pa < end);
746 	}
747 
748 	if (sizeof(phys_avail)/sizeof(phys_avail[0]) < regions_sz)
749 		panic("moea_bootstrap: phys_avail too small");
750 	qsort(regions, regions_sz, sizeof(*regions), mr_cmp);
751 	phys_avail_count = 0;
752 	physsz = 0;
753 	hwphyssz = 0;
754 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
755 	for (i = 0, j = 0; i < regions_sz; i++, j += 2) {
756 		CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start,
757 		    regions[i].mr_start + regions[i].mr_size,
758 		    regions[i].mr_size);
759 		if (hwphyssz != 0 &&
760 		    (physsz + regions[i].mr_size) >= hwphyssz) {
761 			if (physsz < hwphyssz) {
762 				phys_avail[j] = regions[i].mr_start;
763 				phys_avail[j + 1] = regions[i].mr_start +
764 				    hwphyssz - physsz;
765 				physsz = hwphyssz;
766 				phys_avail_count++;
767 			}
768 			break;
769 		}
770 		phys_avail[j] = regions[i].mr_start;
771 		phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size;
772 		phys_avail_count++;
773 		physsz += regions[i].mr_size;
774 	}
775 	physmem = btoc(physsz);
776 
777 	/*
778 	 * Allocate PTEG table.
779 	 */
780 #ifdef PTEGCOUNT
781 	moea_pteg_count = PTEGCOUNT;
782 #else
783 	moea_pteg_count = 0x1000;
784 
785 	while (moea_pteg_count < physmem)
786 		moea_pteg_count <<= 1;
787 
788 	moea_pteg_count >>= 1;
789 #endif /* PTEGCOUNT */
790 
791 	size = moea_pteg_count * sizeof(struct pteg);
792 	CTR2(KTR_PMAP, "moea_bootstrap: %d PTEGs, %d bytes", moea_pteg_count,
793 	    size);
794 	moea_pteg_table = (struct pteg *)moea_bootstrap_alloc(size, size);
795 	CTR1(KTR_PMAP, "moea_bootstrap: PTEG table at %p", moea_pteg_table);
796 	bzero((void *)moea_pteg_table, moea_pteg_count * sizeof(struct pteg));
797 	moea_pteg_mask = moea_pteg_count - 1;
798 
799 	/*
800 	 * Allocate pv/overflow lists.
801 	 */
802 	size = sizeof(struct pvo_head) * moea_pteg_count;
803 	moea_pvo_table = (struct pvo_head *)moea_bootstrap_alloc(size,
804 	    PAGE_SIZE);
805 	CTR1(KTR_PMAP, "moea_bootstrap: PVO table at %p", moea_pvo_table);
806 	for (i = 0; i < moea_pteg_count; i++)
807 		LIST_INIT(&moea_pvo_table[i]);
808 
809 	/*
810 	 * Initialize the lock that synchronizes access to the pteg and pvo
811 	 * tables.
812 	 */
813 	mtx_init(&moea_table_mutex, "pmap table", NULL, MTX_DEF |
814 	    MTX_RECURSE);
815 
816 	mtx_init(&tlbie_mtx, "tlbie", NULL, MTX_SPIN);
817 
818 	/*
819 	 * Initialise the unmanaged pvo pool.
820 	 */
821 	moea_bpvo_pool = (struct pvo_entry *)moea_bootstrap_alloc(
822 		BPVO_POOL_SIZE*sizeof(struct pvo_entry), 0);
823 	moea_bpvo_pool_index = 0;
824 
825 	/*
826 	 * Make sure kernel vsid is allocated as well as VSID 0.
827 	 */
828 	moea_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW]
829 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
830 	moea_vsid_bitmap[0] |= 1;
831 
832 	/*
833 	 * Set up the Open Firmware pmap and add it's mappings.
834 	 */
835 	moea_pinit(mmup, &ofw_pmap);
836 	ofw_pmap.pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
837 	ofw_pmap.pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT;
838 	if ((chosen = OF_finddevice("/chosen")) == -1)
839 		panic("moea_bootstrap: can't find /chosen");
840 	OF_getprop(chosen, "mmu", &mmui, 4);
841 	if ((mmu = OF_instance_to_package(mmui)) == -1)
842 		panic("moea_bootstrap: can't get mmu package");
843 	if ((sz = OF_getproplen(mmu, "translations")) == -1)
844 		panic("moea_bootstrap: can't get ofw translation count");
845 	translations = NULL;
846 	for (i = 0; phys_avail[i] != 0; i += 2) {
847 		if (phys_avail[i + 1] >= sz) {
848 			translations = (struct ofw_map *)phys_avail[i];
849 			break;
850 		}
851 	}
852 	if (translations == NULL)
853 		panic("moea_bootstrap: no space to copy translations");
854 	bzero(translations, sz);
855 	if (OF_getprop(mmu, "translations", translations, sz) == -1)
856 		panic("moea_bootstrap: can't get ofw translations");
857 	CTR0(KTR_PMAP, "moea_bootstrap: translations");
858 	sz /= sizeof(*translations);
859 	qsort(translations, sz, sizeof (*translations), om_cmp);
860 	for (i = 0, ofw_mappings = 0; i < sz; i++) {
861 		CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x",
862 		    translations[i].om_pa, translations[i].om_va,
863 		    translations[i].om_len);
864 
865 		/*
866 		 * If the mapping is 1:1, let the RAM and device on-demand
867 		 * BAT tables take care of the translation.
868 		 */
869 		if (translations[i].om_va == translations[i].om_pa)
870 			continue;
871 
872 		/* Enter the pages */
873 		for (off = 0; off < translations[i].om_len; off += PAGE_SIZE) {
874 			struct	vm_page m;
875 
876 			m.phys_addr = translations[i].om_pa + off;
877 			PMAP_LOCK(&ofw_pmap);
878 			moea_enter_locked(&ofw_pmap,
879 				   translations[i].om_va + off, &m,
880 				   VM_PROT_ALL, 1);
881 			PMAP_UNLOCK(&ofw_pmap);
882 			ofw_mappings++;
883 		}
884 	}
885 
886 	/*
887 	 * Calculate the last available physical address.
888 	 */
889 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
890 		;
891 	Maxmem = powerpc_btop(phys_avail[i + 1]);
892 
893 	/*
894 	 * Initialize the kernel pmap (which is statically allocated).
895 	 */
896 	PMAP_LOCK_INIT(kernel_pmap);
897 	for (i = 0; i < 16; i++) {
898 		kernel_pmap->pm_sr[i] = EMPTY_SEGMENT;
899 	}
900 	kernel_pmap->pm_sr[KERNEL_SR] = KERNEL_SEGMENT;
901 	kernel_pmap->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT;
902 	kernel_pmap->pm_active = ~0;
903 
904 	moea_cpu_bootstrap(mmup,0);
905 
906 	pmap_bootstrapped++;
907 
908 	/*
909 	 * Set the start and end of kva.
910 	 */
911 	virtual_avail = VM_MIN_KERNEL_ADDRESS;
912 	virtual_end = VM_MAX_KERNEL_ADDRESS;
913 
914 	/*
915 	 * Allocate a kernel stack with a guard page for thread0 and map it
916 	 * into the kernel page map.
917 	 */
918 	pa = moea_bootstrap_alloc(KSTACK_PAGES * PAGE_SIZE, PAGE_SIZE);
919 	va = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
920 	virtual_avail = va + KSTACK_PAGES * PAGE_SIZE;
921 	CTR2(KTR_PMAP, "moea_bootstrap: kstack0 at %#x (%#x)", pa, va);
922 	thread0.td_kstack = va;
923 	thread0.td_kstack_pages = KSTACK_PAGES;
924 	for (i = 0; i < KSTACK_PAGES; i++) {
925 		moea_kenter(mmup, va, pa);;
926 		pa += PAGE_SIZE;
927 		va += PAGE_SIZE;
928 	}
929 
930 	/*
931 	 * Allocate virtual address space for the message buffer.
932 	 */
933 	pa = msgbuf_phys = moea_bootstrap_alloc(MSGBUF_SIZE, PAGE_SIZE);
934 	msgbufp = (struct msgbuf *)virtual_avail;
935 	va = virtual_avail;
936 	virtual_avail += round_page(MSGBUF_SIZE);
937 	while (va < virtual_avail) {
938 		moea_kenter(mmup, va, pa);;
939 		pa += PAGE_SIZE;
940 		va += PAGE_SIZE;
941 	}
942 
943 	/*
944 	 * Allocate virtual address space for the dynamic percpu area.
945 	 */
946 	pa = moea_bootstrap_alloc(DPCPU_SIZE, PAGE_SIZE);
947 	dpcpu = (void *)virtual_avail;
948 	va = virtual_avail;
949 	virtual_avail += DPCPU_SIZE;
950 	while (va < virtual_avail) {
951 		moea_kenter(mmup, va, pa);;
952 		pa += PAGE_SIZE;
953 		va += PAGE_SIZE;
954 	}
955 	dpcpu_init(dpcpu, 0);
956 }
957 
958 /*
959  * Activate a user pmap.  The pmap must be activated before it's address
960  * space can be accessed in any way.
961  */
962 void
963 moea_activate(mmu_t mmu, struct thread *td)
964 {
965 	pmap_t	pm, pmr;
966 
967 	/*
968 	 * Load all the data we need up front to encourage the compiler to
969 	 * not issue any loads while we have interrupts disabled below.
970 	 */
971 	pm = &td->td_proc->p_vmspace->vm_pmap;
972 	pmr = pm->pmap_phys;
973 
974 	pm->pm_active |= PCPU_GET(cpumask);
975 	PCPU_SET(curpmap, pmr);
976 }
977 
978 void
979 moea_deactivate(mmu_t mmu, struct thread *td)
980 {
981 	pmap_t	pm;
982 
983 	pm = &td->td_proc->p_vmspace->vm_pmap;
984 	pm->pm_active &= ~PCPU_GET(cpumask);
985 	PCPU_SET(curpmap, NULL);
986 }
987 
988 void
989 moea_change_wiring(mmu_t mmu, pmap_t pm, vm_offset_t va, boolean_t wired)
990 {
991 	struct	pvo_entry *pvo;
992 
993 	PMAP_LOCK(pm);
994 	pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
995 
996 	if (pvo != NULL) {
997 		if (wired) {
998 			if ((pvo->pvo_vaddr & PVO_WIRED) == 0)
999 				pm->pm_stats.wired_count++;
1000 			pvo->pvo_vaddr |= PVO_WIRED;
1001 		} else {
1002 			if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
1003 				pm->pm_stats.wired_count--;
1004 			pvo->pvo_vaddr &= ~PVO_WIRED;
1005 		}
1006 	}
1007 	PMAP_UNLOCK(pm);
1008 }
1009 
1010 void
1011 moea_copy_page(mmu_t mmu, vm_page_t msrc, vm_page_t mdst)
1012 {
1013 	vm_offset_t	dst;
1014 	vm_offset_t	src;
1015 
1016 	dst = VM_PAGE_TO_PHYS(mdst);
1017 	src = VM_PAGE_TO_PHYS(msrc);
1018 
1019 	kcopy((void *)src, (void *)dst, PAGE_SIZE);
1020 }
1021 
1022 /*
1023  * Zero a page of physical memory by temporarily mapping it into the tlb.
1024  */
1025 void
1026 moea_zero_page(mmu_t mmu, vm_page_t m)
1027 {
1028 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1029 	void *va = (void *)pa;
1030 
1031 	bzero(va, PAGE_SIZE);
1032 }
1033 
1034 void
1035 moea_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
1036 {
1037 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1038 	void *va = (void *)(pa + off);
1039 
1040 	bzero(va, size);
1041 }
1042 
1043 void
1044 moea_zero_page_idle(mmu_t mmu, vm_page_t m)
1045 {
1046 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1047 	void *va = (void *)pa;
1048 
1049 	bzero(va, PAGE_SIZE);
1050 }
1051 
1052 /*
1053  * Map the given physical page at the specified virtual address in the
1054  * target pmap with the protection requested.  If specified the page
1055  * will be wired down.
1056  */
1057 void
1058 moea_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1059 	   boolean_t wired)
1060 {
1061 
1062 	vm_page_lock_queues();
1063 	PMAP_LOCK(pmap);
1064 	moea_enter_locked(pmap, va, m, prot, wired);
1065 	vm_page_unlock_queues();
1066 	PMAP_UNLOCK(pmap);
1067 }
1068 
1069 /*
1070  * Map the given physical page at the specified virtual address in the
1071  * target pmap with the protection requested.  If specified the page
1072  * will be wired down.
1073  *
1074  * The page queues and pmap must be locked.
1075  */
1076 static void
1077 moea_enter_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1078     boolean_t wired)
1079 {
1080 	struct		pvo_head *pvo_head;
1081 	uma_zone_t	zone;
1082 	vm_page_t	pg;
1083 	u_int		pte_lo, pvo_flags, was_exec, i;
1084 	int		error;
1085 
1086 	if (!moea_initialized) {
1087 		pvo_head = &moea_pvo_kunmanaged;
1088 		zone = moea_upvo_zone;
1089 		pvo_flags = 0;
1090 		pg = NULL;
1091 		was_exec = PTE_EXEC;
1092 	} else {
1093 		pvo_head = vm_page_to_pvoh(m);
1094 		pg = m;
1095 		zone = moea_mpvo_zone;
1096 		pvo_flags = PVO_MANAGED;
1097 		was_exec = 0;
1098 	}
1099 	if (pmap_bootstrapped)
1100 		mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1101 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1102 
1103 	/* XXX change the pvo head for fake pages */
1104 	if ((m->flags & PG_FICTITIOUS) == PG_FICTITIOUS) {
1105 		pvo_flags &= ~PVO_MANAGED;
1106 		pvo_head = &moea_pvo_kunmanaged;
1107 		zone = moea_upvo_zone;
1108 	}
1109 
1110 	/*
1111 	 * If this is a managed page, and it's the first reference to the page,
1112 	 * clear the execness of the page.  Otherwise fetch the execness.
1113 	 */
1114 	if ((pg != NULL) && ((m->flags & PG_FICTITIOUS) == 0)) {
1115 		if (LIST_EMPTY(pvo_head)) {
1116 			moea_attr_clear(pg, PTE_EXEC);
1117 		} else {
1118 			was_exec = moea_attr_fetch(pg) & PTE_EXEC;
1119 		}
1120 	}
1121 
1122 	/*
1123 	 * Assume the page is cache inhibited and access is guarded unless
1124 	 * it's in our available memory array.
1125 	 */
1126 	pte_lo = PTE_I | PTE_G;
1127 	for (i = 0; i < pregions_sz; i++) {
1128 		if ((VM_PAGE_TO_PHYS(m) >= pregions[i].mr_start) &&
1129 		    (VM_PAGE_TO_PHYS(m) <
1130 			(pregions[i].mr_start + pregions[i].mr_size))) {
1131 			pte_lo = PTE_M;
1132 			break;
1133 		}
1134 	}
1135 
1136 	if (prot & VM_PROT_WRITE) {
1137 		pte_lo |= PTE_BW;
1138 		if (pmap_bootstrapped)
1139 			vm_page_flag_set(m, PG_WRITEABLE);
1140 	} else
1141 		pte_lo |= PTE_BR;
1142 
1143 	if (prot & VM_PROT_EXECUTE)
1144 		pvo_flags |= PVO_EXECUTABLE;
1145 
1146 	if (wired)
1147 		pvo_flags |= PVO_WIRED;
1148 
1149 	if ((m->flags & PG_FICTITIOUS) != 0)
1150 		pvo_flags |= PVO_FAKE;
1151 
1152 	error = moea_pvo_enter(pmap, zone, pvo_head, va, VM_PAGE_TO_PHYS(m),
1153 	    pte_lo, pvo_flags);
1154 
1155 	/*
1156 	 * Flush the real page from the instruction cache if this page is
1157 	 * mapped executable and cacheable and was not previously mapped (or
1158 	 * was not mapped executable).
1159 	 */
1160 	if (error == 0 && (pvo_flags & PVO_EXECUTABLE) &&
1161 	    (pte_lo & PTE_I) == 0 && was_exec == 0) {
1162 		/*
1163 		 * Flush the real memory from the cache.
1164 		 */
1165 		moea_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1166 		if (pg != NULL)
1167 			moea_attr_save(pg, PTE_EXEC);
1168 	}
1169 
1170 	/* XXX syncicache always until problems are sorted */
1171 	moea_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1172 }
1173 
1174 /*
1175  * Maps a sequence of resident pages belonging to the same object.
1176  * The sequence begins with the given page m_start.  This page is
1177  * mapped at the given virtual address start.  Each subsequent page is
1178  * mapped at a virtual address that is offset from start by the same
1179  * amount as the page is offset from m_start within the object.  The
1180  * last page in the sequence is the page with the largest offset from
1181  * m_start that can be mapped at a virtual address less than the given
1182  * virtual address end.  Not every virtual page between start and end
1183  * is mapped; only those for which a resident page exists with the
1184  * corresponding offset from m_start are mapped.
1185  */
1186 void
1187 moea_enter_object(mmu_t mmu, pmap_t pm, vm_offset_t start, vm_offset_t end,
1188     vm_page_t m_start, vm_prot_t prot)
1189 {
1190 	vm_page_t m;
1191 	vm_pindex_t diff, psize;
1192 
1193 	psize = atop(end - start);
1194 	m = m_start;
1195 	PMAP_LOCK(pm);
1196 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1197 		moea_enter_locked(pm, start + ptoa(diff), m, prot &
1198 		    (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1199 		m = TAILQ_NEXT(m, listq);
1200 	}
1201 	PMAP_UNLOCK(pm);
1202 }
1203 
1204 void
1205 moea_enter_quick(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_page_t m,
1206     vm_prot_t prot)
1207 {
1208 
1209 	PMAP_LOCK(pm);
1210 	moea_enter_locked(pm, va, m, prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1211 	    FALSE);
1212 	PMAP_UNLOCK(pm);
1213 
1214 }
1215 
1216 vm_paddr_t
1217 moea_extract(mmu_t mmu, pmap_t pm, vm_offset_t va)
1218 {
1219 	struct	pvo_entry *pvo;
1220 	vm_paddr_t pa;
1221 
1222 	PMAP_LOCK(pm);
1223 	pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1224 	if (pvo == NULL)
1225 		pa = 0;
1226 	else
1227 		pa = (pvo->pvo_pte.pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF);
1228 	PMAP_UNLOCK(pm);
1229 	return (pa);
1230 }
1231 
1232 /*
1233  * Atomically extract and hold the physical page with the given
1234  * pmap and virtual address pair if that mapping permits the given
1235  * protection.
1236  */
1237 vm_page_t
1238 moea_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_prot_t prot)
1239 {
1240 	struct	pvo_entry *pvo;
1241 	vm_page_t m;
1242 
1243 	m = NULL;
1244 	vm_page_lock_queues();
1245 	PMAP_LOCK(pmap);
1246 	pvo = moea_pvo_find_va(pmap, va & ~ADDR_POFF, NULL);
1247 	if (pvo != NULL && (pvo->pvo_pte.pte.pte_hi & PTE_VALID) &&
1248 	    ((pvo->pvo_pte.pte.pte_lo & PTE_PP) == PTE_RW ||
1249 	     (prot & VM_PROT_WRITE) == 0)) {
1250 		m = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte.pte_lo & PTE_RPGN);
1251 		vm_page_hold(m);
1252 	}
1253 	vm_page_unlock_queues();
1254 	PMAP_UNLOCK(pmap);
1255 	return (m);
1256 }
1257 
1258 void
1259 moea_init(mmu_t mmu)
1260 {
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 	powerpc_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.pte_lo & PTE_PP) != PTE_BR) {
1320 			pt = moea_pvo_to_pte(pvo, -1);
1321 			pvo->pvo_pte.pte.pte_lo &= ~PTE_PP;
1322 			pvo->pvo_pte.pte.pte_lo |= PTE_BR;
1323 			if (pt != NULL) {
1324 				moea_pte_synch(pt, &pvo->pvo_pte.pte);
1325 				lo |= pvo->pvo_pte.pte.pte_lo;
1326 				pvo->pvo_pte.pte.pte_lo &= ~PTE_CHG;
1327 				moea_pte_change(pt, &pvo->pvo_pte.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_M;
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 	/*
1419 	 * Allow direct mappings on 32-bit OEA
1420 	 */
1421 	if (va < VM_MIN_KERNEL_ADDRESS) {
1422 		return (va);
1423 	}
1424 
1425 	PMAP_LOCK(kernel_pmap);
1426 	pvo = moea_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL);
1427 	KASSERT(pvo != NULL, ("moea_kextract: no addr found"));
1428 	pa = (pvo->pvo_pte.pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF);
1429 	PMAP_UNLOCK(kernel_pmap);
1430 	return (pa);
1431 }
1432 
1433 /*
1434  * Remove a wired page from kernel virtual address space.
1435  */
1436 void
1437 moea_kremove(mmu_t mmu, vm_offset_t va)
1438 {
1439 
1440 	moea_remove(mmu, kernel_pmap, va, va + PAGE_SIZE);
1441 }
1442 
1443 /*
1444  * Map a range of physical addresses into kernel virtual address space.
1445  *
1446  * The value passed in *virt is a suggested virtual address for the mapping.
1447  * Architectures which can support a direct-mapped physical to virtual region
1448  * can return the appropriate address within that region, leaving '*virt'
1449  * unchanged.  We cannot and therefore do not; *virt is updated with the
1450  * first usable address after the mapped region.
1451  */
1452 vm_offset_t
1453 moea_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1454     vm_offset_t pa_end, int prot)
1455 {
1456 	vm_offset_t	sva, va;
1457 
1458 	sva = *virt;
1459 	va = sva;
1460 	for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE)
1461 		moea_kenter(mmu, va, pa_start);
1462 	*virt = va;
1463 	return (sva);
1464 }
1465 
1466 /*
1467  * Returns true if the pmap's pv is one of the first
1468  * 16 pvs linked to from this page.  This count may
1469  * be changed upwards or downwards in the future; it
1470  * is only necessary that true be returned for a small
1471  * subset of pmaps for proper page aging.
1472  */
1473 boolean_t
1474 moea_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
1475 {
1476         int loops;
1477 	struct pvo_entry *pvo;
1478 
1479         if (!moea_initialized || (m->flags & PG_FICTITIOUS))
1480                 return FALSE;
1481 
1482 	loops = 0;
1483 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1484 		if (pvo->pvo_pmap == pmap)
1485 			return (TRUE);
1486 		if (++loops >= 16)
1487 			break;
1488 	}
1489 
1490 	return (FALSE);
1491 }
1492 
1493 /*
1494  * Return the number of managed mappings to the given physical page
1495  * that are wired.
1496  */
1497 int
1498 moea_page_wired_mappings(mmu_t mmu, vm_page_t m)
1499 {
1500 	struct pvo_entry *pvo;
1501 	int count;
1502 
1503 	count = 0;
1504 	if (!moea_initialized || (m->flags & PG_FICTITIOUS) != 0)
1505 		return (count);
1506 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1507 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink)
1508 		if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
1509 			count++;
1510 	return (count);
1511 }
1512 
1513 static u_int	moea_vsidcontext;
1514 
1515 void
1516 moea_pinit(mmu_t mmu, pmap_t pmap)
1517 {
1518 	int	i, mask;
1519 	u_int	entropy;
1520 
1521 	KASSERT((int)pmap < VM_MIN_KERNEL_ADDRESS, ("moea_pinit: virt pmap"));
1522 	PMAP_LOCK_INIT(pmap);
1523 
1524 	entropy = 0;
1525 	__asm __volatile("mftb %0" : "=r"(entropy));
1526 
1527 	if ((pmap->pmap_phys = (pmap_t)moea_kextract(mmu, (vm_offset_t)pmap))
1528 	    == NULL) {
1529 		pmap->pmap_phys = pmap;
1530 	}
1531 
1532 
1533 	/*
1534 	 * Allocate some segment registers for this pmap.
1535 	 */
1536 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1537 		u_int	hash, n;
1538 
1539 		/*
1540 		 * Create a new value by mutiplying by a prime and adding in
1541 		 * entropy from the timebase register.  This is to make the
1542 		 * VSID more random so that the PT hash function collides
1543 		 * less often.  (Note that the prime casues gcc to do shifts
1544 		 * instead of a multiply.)
1545 		 */
1546 		moea_vsidcontext = (moea_vsidcontext * 0x1105) + entropy;
1547 		hash = moea_vsidcontext & (NPMAPS - 1);
1548 		if (hash == 0)		/* 0 is special, avoid it */
1549 			continue;
1550 		n = hash >> 5;
1551 		mask = 1 << (hash & (VSID_NBPW - 1));
1552 		hash = (moea_vsidcontext & 0xfffff);
1553 		if (moea_vsid_bitmap[n] & mask) {	/* collision? */
1554 			/* anything free in this bucket? */
1555 			if (moea_vsid_bitmap[n] == 0xffffffff) {
1556 				entropy = (moea_vsidcontext >> 20);
1557 				continue;
1558 			}
1559 			i = ffs(~moea_vsid_bitmap[i]) - 1;
1560 			mask = 1 << i;
1561 			hash &= 0xfffff & ~(VSID_NBPW - 1);
1562 			hash |= i;
1563 		}
1564 		moea_vsid_bitmap[n] |= mask;
1565 		for (i = 0; i < 16; i++)
1566 			pmap->pm_sr[i] = VSID_MAKE(i, hash);
1567 		return;
1568 	}
1569 
1570 	panic("moea_pinit: out of segments");
1571 }
1572 
1573 /*
1574  * Initialize the pmap associated with process 0.
1575  */
1576 void
1577 moea_pinit0(mmu_t mmu, pmap_t pm)
1578 {
1579 
1580 	moea_pinit(mmu, pm);
1581 	bzero(&pm->pm_stats, sizeof(pm->pm_stats));
1582 }
1583 
1584 /*
1585  * Set the physical protection on the specified range of this map as requested.
1586  */
1587 void
1588 moea_protect(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva,
1589     vm_prot_t prot)
1590 {
1591 	struct	pvo_entry *pvo;
1592 	struct	pte *pt;
1593 	int	pteidx;
1594 
1595 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1596 	    ("moea_protect: non current pmap"));
1597 
1598 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1599 		moea_remove(mmu, pm, sva, eva);
1600 		return;
1601 	}
1602 
1603 	vm_page_lock_queues();
1604 	PMAP_LOCK(pm);
1605 	for (; sva < eva; sva += PAGE_SIZE) {
1606 		pvo = moea_pvo_find_va(pm, sva, &pteidx);
1607 		if (pvo == NULL)
1608 			continue;
1609 
1610 		if ((prot & VM_PROT_EXECUTE) == 0)
1611 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1612 
1613 		/*
1614 		 * Grab the PTE pointer before we diddle with the cached PTE
1615 		 * copy.
1616 		 */
1617 		pt = moea_pvo_to_pte(pvo, pteidx);
1618 		/*
1619 		 * Change the protection of the page.
1620 		 */
1621 		pvo->pvo_pte.pte.pte_lo &= ~PTE_PP;
1622 		pvo->pvo_pte.pte.pte_lo |= PTE_BR;
1623 
1624 		/*
1625 		 * If the PVO is in the page table, update that pte as well.
1626 		 */
1627 		if (pt != NULL) {
1628 			moea_pte_change(pt, &pvo->pvo_pte.pte, pvo->pvo_vaddr);
1629 			mtx_unlock(&moea_table_mutex);
1630 		}
1631 	}
1632 	vm_page_unlock_queues();
1633 	PMAP_UNLOCK(pm);
1634 }
1635 
1636 /*
1637  * Map a list of wired pages into kernel virtual address space.  This is
1638  * intended for temporary mappings which do not need page modification or
1639  * references recorded.  Existing mappings in the region are overwritten.
1640  */
1641 void
1642 moea_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1643 {
1644 	vm_offset_t va;
1645 
1646 	va = sva;
1647 	while (count-- > 0) {
1648 		moea_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1649 		va += PAGE_SIZE;
1650 		m++;
1651 	}
1652 }
1653 
1654 /*
1655  * Remove page mappings from kernel virtual address space.  Intended for
1656  * temporary mappings entered by moea_qenter.
1657  */
1658 void
1659 moea_qremove(mmu_t mmu, vm_offset_t sva, int count)
1660 {
1661 	vm_offset_t va;
1662 
1663 	va = sva;
1664 	while (count-- > 0) {
1665 		moea_kremove(mmu, va);
1666 		va += PAGE_SIZE;
1667 	}
1668 }
1669 
1670 void
1671 moea_release(mmu_t mmu, pmap_t pmap)
1672 {
1673         int idx, mask;
1674 
1675 	/*
1676 	 * Free segment register's VSID
1677 	 */
1678         if (pmap->pm_sr[0] == 0)
1679                 panic("moea_release");
1680 
1681         idx = VSID_TO_HASH(pmap->pm_sr[0]) & (NPMAPS-1);
1682         mask = 1 << (idx % VSID_NBPW);
1683         idx /= VSID_NBPW;
1684         moea_vsid_bitmap[idx] &= ~mask;
1685 	PMAP_LOCK_DESTROY(pmap);
1686 }
1687 
1688 /*
1689  * Remove the given range of addresses from the specified map.
1690  */
1691 void
1692 moea_remove(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1693 {
1694 	struct	pvo_entry *pvo;
1695 	int	pteidx;
1696 
1697 	vm_page_lock_queues();
1698 	PMAP_LOCK(pm);
1699 	for (; sva < eva; sva += PAGE_SIZE) {
1700 		pvo = moea_pvo_find_va(pm, sva, &pteidx);
1701 		if (pvo != NULL) {
1702 			moea_pvo_remove(pvo, pteidx);
1703 		}
1704 	}
1705 	PMAP_UNLOCK(pm);
1706 	vm_page_unlock_queues();
1707 }
1708 
1709 /*
1710  * Remove physical page from all pmaps in which it resides. moea_pvo_remove()
1711  * will reflect changes in pte's back to the vm_page.
1712  */
1713 void
1714 moea_remove_all(mmu_t mmu, vm_page_t m)
1715 {
1716 	struct  pvo_head *pvo_head;
1717 	struct	pvo_entry *pvo, *next_pvo;
1718 	pmap_t	pmap;
1719 
1720 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1721 
1722 	pvo_head = vm_page_to_pvoh(m);
1723 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1724 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1725 
1726 		MOEA_PVO_CHECK(pvo);	/* sanity check */
1727 		pmap = pvo->pvo_pmap;
1728 		PMAP_LOCK(pmap);
1729 		moea_pvo_remove(pvo, -1);
1730 		PMAP_UNLOCK(pmap);
1731 	}
1732 	vm_page_flag_clear(m, PG_WRITEABLE);
1733 }
1734 
1735 /*
1736  * Allocate a physical page of memory directly from the phys_avail map.
1737  * Can only be called from moea_bootstrap before avail start and end are
1738  * calculated.
1739  */
1740 static vm_offset_t
1741 moea_bootstrap_alloc(vm_size_t size, u_int align)
1742 {
1743 	vm_offset_t	s, e;
1744 	int		i, j;
1745 
1746 	size = round_page(size);
1747 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1748 		if (align != 0)
1749 			s = (phys_avail[i] + align - 1) & ~(align - 1);
1750 		else
1751 			s = phys_avail[i];
1752 		e = s + size;
1753 
1754 		if (s < phys_avail[i] || e > phys_avail[i + 1])
1755 			continue;
1756 
1757 		if (s == phys_avail[i]) {
1758 			phys_avail[i] += size;
1759 		} else if (e == phys_avail[i + 1]) {
1760 			phys_avail[i + 1] -= size;
1761 		} else {
1762 			for (j = phys_avail_count * 2; j > i; j -= 2) {
1763 				phys_avail[j] = phys_avail[j - 2];
1764 				phys_avail[j + 1] = phys_avail[j - 1];
1765 			}
1766 
1767 			phys_avail[i + 3] = phys_avail[i + 1];
1768 			phys_avail[i + 1] = s;
1769 			phys_avail[i + 2] = e;
1770 			phys_avail_count++;
1771 		}
1772 
1773 		return (s);
1774 	}
1775 	panic("moea_bootstrap_alloc: could not allocate memory");
1776 }
1777 
1778 static void
1779 moea_syncicache(vm_offset_t pa, vm_size_t len)
1780 {
1781 	__syncicache((void *)pa, len);
1782 }
1783 
1784 static int
1785 moea_pvo_enter(pmap_t pm, uma_zone_t zone, struct pvo_head *pvo_head,
1786     vm_offset_t va, vm_offset_t pa, u_int pte_lo, int flags)
1787 {
1788 	struct	pvo_entry *pvo;
1789 	u_int	sr;
1790 	int	first;
1791 	u_int	ptegidx;
1792 	int	i;
1793 	int     bootstrap;
1794 
1795 	moea_pvo_enter_calls++;
1796 	first = 0;
1797 	bootstrap = 0;
1798 
1799 	/*
1800 	 * Compute the PTE Group index.
1801 	 */
1802 	va &= ~ADDR_POFF;
1803 	sr = va_to_sr(pm->pm_sr, va);
1804 	ptegidx = va_to_pteg(sr, va);
1805 
1806 	/*
1807 	 * Remove any existing mapping for this page.  Reuse the pvo entry if
1808 	 * there is a mapping.
1809 	 */
1810 	mtx_lock(&moea_table_mutex);
1811 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
1812 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1813 			if ((pvo->pvo_pte.pte.pte_lo & PTE_RPGN) == pa &&
1814 			    (pvo->pvo_pte.pte.pte_lo & PTE_PP) ==
1815 			    (pte_lo & PTE_PP)) {
1816 				mtx_unlock(&moea_table_mutex);
1817 				return (0);
1818 			}
1819 			moea_pvo_remove(pvo, -1);
1820 			break;
1821 		}
1822 	}
1823 
1824 	/*
1825 	 * If we aren't overwriting a mapping, try to allocate.
1826 	 */
1827 	if (moea_initialized) {
1828 		pvo = uma_zalloc(zone, M_NOWAIT);
1829 	} else {
1830 		if (moea_bpvo_pool_index >= BPVO_POOL_SIZE) {
1831 			panic("moea_enter: bpvo pool exhausted, %d, %d, %d",
1832 			      moea_bpvo_pool_index, BPVO_POOL_SIZE,
1833 			      BPVO_POOL_SIZE * sizeof(struct pvo_entry));
1834 		}
1835 		pvo = &moea_bpvo_pool[moea_bpvo_pool_index];
1836 		moea_bpvo_pool_index++;
1837 		bootstrap = 1;
1838 	}
1839 
1840 	if (pvo == NULL) {
1841 		mtx_unlock(&moea_table_mutex);
1842 		return (ENOMEM);
1843 	}
1844 
1845 	moea_pvo_entries++;
1846 	pvo->pvo_vaddr = va;
1847 	pvo->pvo_pmap = pm;
1848 	LIST_INSERT_HEAD(&moea_pvo_table[ptegidx], pvo, pvo_olink);
1849 	pvo->pvo_vaddr &= ~ADDR_POFF;
1850 	if (flags & VM_PROT_EXECUTE)
1851 		pvo->pvo_vaddr |= PVO_EXECUTABLE;
1852 	if (flags & PVO_WIRED)
1853 		pvo->pvo_vaddr |= PVO_WIRED;
1854 	if (pvo_head != &moea_pvo_kunmanaged)
1855 		pvo->pvo_vaddr |= PVO_MANAGED;
1856 	if (bootstrap)
1857 		pvo->pvo_vaddr |= PVO_BOOTSTRAP;
1858 	if (flags & PVO_FAKE)
1859 		pvo->pvo_vaddr |= PVO_FAKE;
1860 
1861 	moea_pte_create(&pvo->pvo_pte.pte, sr, va, pa | pte_lo);
1862 
1863 	/*
1864 	 * Remember if the list was empty and therefore will be the first
1865 	 * item.
1866 	 */
1867 	if (LIST_FIRST(pvo_head) == NULL)
1868 		first = 1;
1869 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1870 
1871 	if (pvo->pvo_pte.pte.pte_lo & PVO_WIRED)
1872 		pm->pm_stats.wired_count++;
1873 	pm->pm_stats.resident_count++;
1874 
1875 	/*
1876 	 * We hope this succeeds but it isn't required.
1877 	 */
1878 	i = moea_pte_insert(ptegidx, &pvo->pvo_pte.pte);
1879 	if (i >= 0) {
1880 		PVO_PTEGIDX_SET(pvo, i);
1881 	} else {
1882 		panic("moea_pvo_enter: overflow");
1883 		moea_pte_overflow++;
1884 	}
1885 	mtx_unlock(&moea_table_mutex);
1886 
1887 	return (first ? ENOENT : 0);
1888 }
1889 
1890 static void
1891 moea_pvo_remove(struct pvo_entry *pvo, int pteidx)
1892 {
1893 	struct	pte *pt;
1894 
1895 	/*
1896 	 * If there is an active pte entry, we need to deactivate it (and
1897 	 * save the ref & cfg bits).
1898 	 */
1899 	pt = moea_pvo_to_pte(pvo, pteidx);
1900 	if (pt != NULL) {
1901 		moea_pte_unset(pt, &pvo->pvo_pte.pte, pvo->pvo_vaddr);
1902 		mtx_unlock(&moea_table_mutex);
1903 		PVO_PTEGIDX_CLR(pvo);
1904 	} else {
1905 		moea_pte_overflow--;
1906 	}
1907 
1908 	/*
1909 	 * Update our statistics.
1910 	 */
1911 	pvo->pvo_pmap->pm_stats.resident_count--;
1912 	if (pvo->pvo_pte.pte.pte_lo & PVO_WIRED)
1913 		pvo->pvo_pmap->pm_stats.wired_count--;
1914 
1915 	/*
1916 	 * Save the REF/CHG bits into their cache if the page is managed.
1917 	 */
1918 	if ((pvo->pvo_vaddr & (PVO_MANAGED|PVO_FAKE)) == PVO_MANAGED) {
1919 		struct	vm_page *pg;
1920 
1921 		pg = PHYS_TO_VM_PAGE(pvo->pvo_pte.pte.pte_lo & PTE_RPGN);
1922 		if (pg != NULL) {
1923 			moea_attr_save(pg, pvo->pvo_pte.pte.pte_lo &
1924 			    (PTE_REF | PTE_CHG));
1925 		}
1926 	}
1927 
1928 	/*
1929 	 * Remove this PVO from the PV list.
1930 	 */
1931 	LIST_REMOVE(pvo, pvo_vlink);
1932 
1933 	/*
1934 	 * Remove this from the overflow list and return it to the pool
1935 	 * if we aren't going to reuse it.
1936 	 */
1937 	LIST_REMOVE(pvo, pvo_olink);
1938 	if (!(pvo->pvo_vaddr & PVO_BOOTSTRAP))
1939 		uma_zfree(pvo->pvo_vaddr & PVO_MANAGED ? moea_mpvo_zone :
1940 		    moea_upvo_zone, pvo);
1941 	moea_pvo_entries--;
1942 	moea_pvo_remove_calls++;
1943 }
1944 
1945 static __inline int
1946 moea_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1947 {
1948 	int	pteidx;
1949 
1950 	/*
1951 	 * We can find the actual pte entry without searching by grabbing
1952 	 * the PTEG index from 3 unused bits in pte_lo[11:9] and by
1953 	 * noticing the HID bit.
1954 	 */
1955 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1956 	if (pvo->pvo_pte.pte.pte_hi & PTE_HID)
1957 		pteidx ^= moea_pteg_mask * 8;
1958 
1959 	return (pteidx);
1960 }
1961 
1962 static struct pvo_entry *
1963 moea_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p)
1964 {
1965 	struct	pvo_entry *pvo;
1966 	int	ptegidx;
1967 	u_int	sr;
1968 
1969 	va &= ~ADDR_POFF;
1970 	sr = va_to_sr(pm->pm_sr, va);
1971 	ptegidx = va_to_pteg(sr, va);
1972 
1973 	mtx_lock(&moea_table_mutex);
1974 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
1975 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1976 			if (pteidx_p)
1977 				*pteidx_p = moea_pvo_pte_index(pvo, ptegidx);
1978 			break;
1979 		}
1980 	}
1981 	mtx_unlock(&moea_table_mutex);
1982 
1983 	return (pvo);
1984 }
1985 
1986 static struct pte *
1987 moea_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1988 {
1989 	struct	pte *pt;
1990 
1991 	/*
1992 	 * If we haven't been supplied the ptegidx, calculate it.
1993 	 */
1994 	if (pteidx == -1) {
1995 		int	ptegidx;
1996 		u_int	sr;
1997 
1998 		sr = va_to_sr(pvo->pvo_pmap->pm_sr, pvo->pvo_vaddr);
1999 		ptegidx = va_to_pteg(sr, pvo->pvo_vaddr);
2000 		pteidx = moea_pvo_pte_index(pvo, ptegidx);
2001 	}
2002 
2003 	pt = &moea_pteg_table[pteidx >> 3].pt[pteidx & 7];
2004 	mtx_lock(&moea_table_mutex);
2005 
2006 	if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
2007 		panic("moea_pvo_to_pte: pvo %p has valid pte in pvo but no "
2008 		    "valid pte index", pvo);
2009 	}
2010 
2011 	if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
2012 		panic("moea_pvo_to_pte: pvo %p has valid pte index in pvo "
2013 		    "pvo but no valid pte", pvo);
2014 	}
2015 
2016 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
2017 		if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) == 0) {
2018 			panic("moea_pvo_to_pte: pvo %p has valid pte in "
2019 			    "moea_pteg_table %p but invalid in pvo", pvo, pt);
2020 		}
2021 
2022 		if (((pt->pte_lo ^ pvo->pvo_pte.pte.pte_lo) & ~(PTE_CHG|PTE_REF))
2023 		    != 0) {
2024 			panic("moea_pvo_to_pte: pvo %p pte does not match "
2025 			    "pte %p in moea_pteg_table", pvo, pt);
2026 		}
2027 
2028 		mtx_assert(&moea_table_mutex, MA_OWNED);
2029 		return (pt);
2030 	}
2031 
2032 	if (pvo->pvo_pte.pte.pte_hi & PTE_VALID) {
2033 		panic("moea_pvo_to_pte: pvo %p has invalid pte %p in "
2034 		    "moea_pteg_table but valid in pvo", pvo, pt);
2035 	}
2036 
2037 	mtx_unlock(&moea_table_mutex);
2038 	return (NULL);
2039 }
2040 
2041 /*
2042  * XXX: THIS STUFF SHOULD BE IN pte.c?
2043  */
2044 int
2045 moea_pte_spill(vm_offset_t addr)
2046 {
2047 	struct	pvo_entry *source_pvo, *victim_pvo;
2048 	struct	pvo_entry *pvo;
2049 	int	ptegidx, i, j;
2050 	u_int	sr;
2051 	struct	pteg *pteg;
2052 	struct	pte *pt;
2053 
2054 	moea_pte_spills++;
2055 
2056 	sr = mfsrin(addr);
2057 	ptegidx = va_to_pteg(sr, addr);
2058 
2059 	/*
2060 	 * Have to substitute some entry.  Use the primary hash for this.
2061 	 * Use low bits of timebase as random generator.
2062 	 */
2063 	pteg = &moea_pteg_table[ptegidx];
2064 	mtx_lock(&moea_table_mutex);
2065 	__asm __volatile("mftb %0" : "=r"(i));
2066 	i &= 7;
2067 	pt = &pteg->pt[i];
2068 
2069 	source_pvo = NULL;
2070 	victim_pvo = NULL;
2071 	LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) {
2072 		/*
2073 		 * We need to find a pvo entry for this address.
2074 		 */
2075 		MOEA_PVO_CHECK(pvo);
2076 		if (source_pvo == NULL &&
2077 		    moea_pte_match(&pvo->pvo_pte.pte, sr, addr,
2078 		    pvo->pvo_pte.pte.pte_hi & PTE_HID)) {
2079 			/*
2080 			 * Now found an entry to be spilled into the pteg.
2081 			 * The PTE is now valid, so we know it's active.
2082 			 */
2083 			j = moea_pte_insert(ptegidx, &pvo->pvo_pte.pte);
2084 
2085 			if (j >= 0) {
2086 				PVO_PTEGIDX_SET(pvo, j);
2087 				moea_pte_overflow--;
2088 				MOEA_PVO_CHECK(pvo);
2089 				mtx_unlock(&moea_table_mutex);
2090 				return (1);
2091 			}
2092 
2093 			source_pvo = pvo;
2094 
2095 			if (victim_pvo != NULL)
2096 				break;
2097 		}
2098 
2099 		/*
2100 		 * We also need the pvo entry of the victim we are replacing
2101 		 * so save the R & C bits of the PTE.
2102 		 */
2103 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
2104 		    moea_pte_compare(pt, &pvo->pvo_pte.pte)) {
2105 			victim_pvo = pvo;
2106 			if (source_pvo != NULL)
2107 				break;
2108 		}
2109 	}
2110 
2111 	if (source_pvo == NULL) {
2112 		mtx_unlock(&moea_table_mutex);
2113 		return (0);
2114 	}
2115 
2116 	if (victim_pvo == NULL) {
2117 		if ((pt->pte_hi & PTE_HID) == 0)
2118 			panic("moea_pte_spill: victim p-pte (%p) has no pvo"
2119 			    "entry", pt);
2120 
2121 		/*
2122 		 * If this is a secondary PTE, we need to search it's primary
2123 		 * pvo bucket for the matching PVO.
2124 		 */
2125 		LIST_FOREACH(pvo, &moea_pvo_table[ptegidx ^ moea_pteg_mask],
2126 		    pvo_olink) {
2127 			MOEA_PVO_CHECK(pvo);
2128 			/*
2129 			 * We also need the pvo entry of the victim we are
2130 			 * replacing so save the R & C bits of the PTE.
2131 			 */
2132 			if (moea_pte_compare(pt, &pvo->pvo_pte.pte)) {
2133 				victim_pvo = pvo;
2134 				break;
2135 			}
2136 		}
2137 
2138 		if (victim_pvo == NULL)
2139 			panic("moea_pte_spill: victim s-pte (%p) has no pvo"
2140 			    "entry", pt);
2141 	}
2142 
2143 	/*
2144 	 * We are invalidating the TLB entry for the EA we are replacing even
2145 	 * though it's valid.  If we don't, we lose any ref/chg bit changes
2146 	 * contained in the TLB entry.
2147 	 */
2148 	source_pvo->pvo_pte.pte.pte_hi &= ~PTE_HID;
2149 
2150 	moea_pte_unset(pt, &victim_pvo->pvo_pte.pte, victim_pvo->pvo_vaddr);
2151 	moea_pte_set(pt, &source_pvo->pvo_pte.pte);
2152 
2153 	PVO_PTEGIDX_CLR(victim_pvo);
2154 	PVO_PTEGIDX_SET(source_pvo, i);
2155 	moea_pte_replacements++;
2156 
2157 	MOEA_PVO_CHECK(victim_pvo);
2158 	MOEA_PVO_CHECK(source_pvo);
2159 
2160 	mtx_unlock(&moea_table_mutex);
2161 	return (1);
2162 }
2163 
2164 static int
2165 moea_pte_insert(u_int ptegidx, struct pte *pvo_pt)
2166 {
2167 	struct	pte *pt;
2168 	int	i;
2169 
2170 	mtx_assert(&moea_table_mutex, MA_OWNED);
2171 
2172 	/*
2173 	 * First try primary hash.
2174 	 */
2175 	for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2176 		if ((pt->pte_hi & PTE_VALID) == 0) {
2177 			pvo_pt->pte_hi &= ~PTE_HID;
2178 			moea_pte_set(pt, pvo_pt);
2179 			return (i);
2180 		}
2181 	}
2182 
2183 	/*
2184 	 * Now try secondary hash.
2185 	 */
2186 	ptegidx ^= moea_pteg_mask;
2187 
2188 	for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2189 		if ((pt->pte_hi & PTE_VALID) == 0) {
2190 			pvo_pt->pte_hi |= PTE_HID;
2191 			moea_pte_set(pt, pvo_pt);
2192 			return (i);
2193 		}
2194 	}
2195 
2196 	panic("moea_pte_insert: overflow");
2197 	return (-1);
2198 }
2199 
2200 static boolean_t
2201 moea_query_bit(vm_page_t m, int ptebit)
2202 {
2203 	struct	pvo_entry *pvo;
2204 	struct	pte *pt;
2205 
2206 #if 0
2207 	if (moea_attr_fetch(m) & ptebit)
2208 		return (TRUE);
2209 #endif
2210 
2211 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2212 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2213 
2214 		/*
2215 		 * See if we saved the bit off.  If so, cache it and return
2216 		 * success.
2217 		 */
2218 		if (pvo->pvo_pte.pte.pte_lo & ptebit) {
2219 			moea_attr_save(m, ptebit);
2220 			MOEA_PVO_CHECK(pvo);	/* sanity check */
2221 			return (TRUE);
2222 		}
2223 	}
2224 
2225 	/*
2226 	 * No luck, now go through the hard part of looking at the PTEs
2227 	 * themselves.  Sync so that any pending REF/CHG bits are flushed to
2228 	 * the PTEs.
2229 	 */
2230 	powerpc_sync();
2231 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2232 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2233 
2234 		/*
2235 		 * See if this pvo has a valid PTE.  if so, fetch the
2236 		 * REF/CHG bits from the valid PTE.  If the appropriate
2237 		 * ptebit is set, cache it and return success.
2238 		 */
2239 		pt = moea_pvo_to_pte(pvo, -1);
2240 		if (pt != NULL) {
2241 			moea_pte_synch(pt, &pvo->pvo_pte.pte);
2242 			mtx_unlock(&moea_table_mutex);
2243 			if (pvo->pvo_pte.pte.pte_lo & ptebit) {
2244 				moea_attr_save(m, ptebit);
2245 				MOEA_PVO_CHECK(pvo);	/* sanity check */
2246 				return (TRUE);
2247 			}
2248 		}
2249 	}
2250 
2251 	return (FALSE);
2252 }
2253 
2254 static u_int
2255 moea_clear_bit(vm_page_t m, int ptebit, int *origbit)
2256 {
2257 	u_int	count;
2258 	struct	pvo_entry *pvo;
2259 	struct	pte *pt;
2260 	int	rv;
2261 
2262 	/*
2263 	 * Clear the cached value.
2264 	 */
2265 	rv = moea_attr_fetch(m);
2266 	moea_attr_clear(m, ptebit);
2267 
2268 	/*
2269 	 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so
2270 	 * we can reset the right ones).  note that since the pvo entries and
2271 	 * list heads are accessed via BAT0 and are never placed in the page
2272 	 * table, we don't have to worry about further accesses setting the
2273 	 * REF/CHG bits.
2274 	 */
2275 	powerpc_sync();
2276 
2277 	/*
2278 	 * For each pvo entry, clear the pvo's ptebit.  If this pvo has a
2279 	 * valid pte clear the ptebit from the valid pte.
2280 	 */
2281 	count = 0;
2282 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2283 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2284 		pt = moea_pvo_to_pte(pvo, -1);
2285 		if (pt != NULL) {
2286 			moea_pte_synch(pt, &pvo->pvo_pte.pte);
2287 			if (pvo->pvo_pte.pte.pte_lo & ptebit) {
2288 				count++;
2289 				moea_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2290 			}
2291 			mtx_unlock(&moea_table_mutex);
2292 		}
2293 		rv |= pvo->pvo_pte.pte.pte_lo;
2294 		pvo->pvo_pte.pte.pte_lo &= ~ptebit;
2295 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2296 	}
2297 
2298 	if (origbit != NULL) {
2299 		*origbit = rv;
2300 	}
2301 
2302 	return (count);
2303 }
2304 
2305 /*
2306  * Return true if the physical range is encompassed by the battable[idx]
2307  */
2308 static int
2309 moea_bat_mapped(int idx, vm_offset_t pa, vm_size_t size)
2310 {
2311 	u_int prot;
2312 	u_int32_t start;
2313 	u_int32_t end;
2314 	u_int32_t bat_ble;
2315 
2316 	/*
2317 	 * Return immediately if not a valid mapping
2318 	 */
2319 	if (!battable[idx].batu & BAT_Vs)
2320 		return (EINVAL);
2321 
2322 	/*
2323 	 * The BAT entry must be cache-inhibited, guarded, and r/w
2324 	 * so it can function as an i/o page
2325 	 */
2326 	prot = battable[idx].batl & (BAT_I|BAT_G|BAT_PP_RW);
2327 	if (prot != (BAT_I|BAT_G|BAT_PP_RW))
2328 		return (EPERM);
2329 
2330 	/*
2331 	 * The address should be within the BAT range. Assume that the
2332 	 * start address in the BAT has the correct alignment (thus
2333 	 * not requiring masking)
2334 	 */
2335 	start = battable[idx].batl & BAT_PBS;
2336 	bat_ble = (battable[idx].batu & ~(BAT_EBS)) | 0x03;
2337 	end = start | (bat_ble << 15) | 0x7fff;
2338 
2339 	if ((pa < start) || ((pa + size) > end))
2340 		return (ERANGE);
2341 
2342 	return (0);
2343 }
2344 
2345 boolean_t
2346 moea_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2347 {
2348 	int i;
2349 
2350 	/*
2351 	 * This currently does not work for entries that
2352 	 * overlap 256M BAT segments.
2353 	 */
2354 
2355 	for(i = 0; i < 16; i++)
2356 		if (moea_bat_mapped(i, pa, size) == 0)
2357 			return (0);
2358 
2359 	return (EFAULT);
2360 }
2361 
2362 boolean_t
2363 moea_page_executable(mmu_t mmu, vm_page_t pg)
2364 {
2365 	return ((moea_attr_fetch(pg) & PTE_EXEC) == PTE_EXEC);
2366 }
2367 
2368 /*
2369  * Map a set of physical memory pages into the kernel virtual
2370  * address space. Return a pointer to where it is mapped. This
2371  * routine is intended to be used for mapping device memory,
2372  * NOT real memory.
2373  */
2374 void *
2375 moea_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2376 {
2377 	vm_offset_t va, tmpva, ppa, offset;
2378 	int i;
2379 
2380 	ppa = trunc_page(pa);
2381 	offset = pa & PAGE_MASK;
2382 	size = roundup(offset + size, PAGE_SIZE);
2383 
2384 	GIANT_REQUIRED;
2385 
2386 	/*
2387 	 * If the physical address lies within a valid BAT table entry,
2388 	 * return the 1:1 mapping. This currently doesn't work
2389 	 * for regions that overlap 256M BAT segments.
2390 	 */
2391 	for (i = 0; i < 16; i++) {
2392 		if (moea_bat_mapped(i, pa, size) == 0)
2393 			return ((void *) pa);
2394 	}
2395 
2396 	va = kmem_alloc_nofault(kernel_map, size);
2397 	if (!va)
2398 		panic("moea_mapdev: Couldn't alloc kernel virtual memory");
2399 
2400 	for (tmpva = va; size > 0;) {
2401 		moea_kenter(mmu, tmpva, ppa);
2402 		tlbie(tmpva);
2403 		size -= PAGE_SIZE;
2404 		tmpva += PAGE_SIZE;
2405 		ppa += PAGE_SIZE;
2406 	}
2407 
2408 	return ((void *)(va + offset));
2409 }
2410 
2411 void
2412 moea_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2413 {
2414 	vm_offset_t base, offset;
2415 
2416 	/*
2417 	 * If this is outside kernel virtual space, then it's a
2418 	 * battable entry and doesn't require unmapping
2419 	 */
2420 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2421 		base = trunc_page(va);
2422 		offset = va & PAGE_MASK;
2423 		size = roundup(offset + size, PAGE_SIZE);
2424 		kmem_free(kernel_map, base, size);
2425 	}
2426 }
2427