xref: /freebsd/sys/powerpc/aim/mmu_oea64.c (revision 1a4fcaebe30b3067a19baf8871a27942f4bb32cf)
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 <sys/kdb.h>
131 
132 #include <dev/ofw/openfirm.h>
133 
134 #include <vm/vm.h>
135 #include <vm/vm_param.h>
136 #include <vm/vm_kern.h>
137 #include <vm/vm_page.h>
138 #include <vm/vm_map.h>
139 #include <vm/vm_object.h>
140 #include <vm/vm_extern.h>
141 #include <vm/vm_pageout.h>
142 #include <vm/vm_pager.h>
143 #include <vm/uma.h>
144 
145 #include <machine/cpu.h>
146 #include <machine/platform.h>
147 #include <machine/frame.h>
148 #include <machine/md_var.h>
149 #include <machine/psl.h>
150 #include <machine/bat.h>
151 #include <machine/pte.h>
152 #include <machine/sr.h>
153 #include <machine/trap.h>
154 #include <machine/mmuvar.h>
155 
156 #include "mmu_if.h"
157 
158 #define	MOEA_DEBUG
159 
160 #define TODO	panic("%s: not implemented", __func__);
161 
162 static __inline u_int32_t
163 cntlzw(volatile u_int32_t a) {
164 	u_int32_t b;
165 	__asm ("cntlzw %0, %1" : "=r"(b) : "r"(a));
166 	return b;
167 }
168 
169 static __inline uint64_t
170 va_to_vsid(pmap_t pm, vm_offset_t va)
171 {
172 	return ((pm->pm_sr[(uintptr_t)va >> ADDR_SR_SHFT]) & SR_VSID_MASK);
173 }
174 
175 #define	TLBSYNC()	__asm __volatile("tlbsync; ptesync");
176 #define	SYNC()		__asm __volatile("sync");
177 #define	EIEIO()		__asm __volatile("eieio");
178 
179 /*
180  * The tlbie instruction must be executed in 64-bit mode
181  * so we have to twiddle MSR[SF] around every invocation.
182  * Just to add to the fun, exceptions must be off as well
183  * so that we can't trap in 64-bit mode. What a pain.
184  */
185 
186 static __inline void
187 TLBIE(pmap_t pmap, vm_offset_t va) {
188 	register_t msr;
189 	register_t scratch;
190 
191 	uint64_t vpn;
192 	register_t vpn_hi, vpn_lo;
193 
194 #if 1
195 	/*
196 	 * CPU documentation says that tlbie takes the VPN, not the
197 	 * VA. I think the code below does this correctly. We will see.
198 	 */
199 
200 	vpn = (uint64_t)(va & ADDR_PIDX);
201 	if (pmap != NULL)
202 		vpn |= (va_to_vsid(pmap,va) << 28);
203 #else
204 	vpn = va;
205 #endif
206 
207 	vpn_hi = (uint32_t)(vpn >> 32);
208 	vpn_lo = (uint32_t)vpn;
209 
210 	__asm __volatile("\
211 	    mfmsr %0; \
212 	    clrldi %1,%0,49; \
213 	    insrdi %1,1,1,0; \
214 	    mtmsrd %1; \
215 	    ptesync; \
216 	    \
217 	    sld %1,%2,%4; \
218 	    or %1,%1,%3; \
219 	    tlbie %1; \
220 	    \
221 	    mtmsrd %0; \
222 	    eieio; \
223 	    tlbsync; \
224 	    ptesync;"
225 	: "=r"(msr), "=r"(scratch) : "r"(vpn_hi), "r"(vpn_lo), "r"(32));
226 }
227 
228 #define DISABLE_TRANS(msr)	msr = mfmsr(); mtmsr(msr & ~PSL_DR); isync()
229 #define ENABLE_TRANS(msr)	mtmsr(msr); isync()
230 
231 #define	VSID_MAKE(sr, hash)	((sr) | (((hash) & 0xfffff) << 4))
232 #define	VSID_TO_SR(vsid)	((vsid) & 0xf)
233 #define	VSID_TO_HASH(vsid)	(((vsid) >> 4) & 0xfffff)
234 
235 #define	PVO_PTEGIDX_MASK	0x007		/* which PTEG slot */
236 #define	PVO_PTEGIDX_VALID	0x008		/* slot is valid */
237 #define	PVO_WIRED		0x010		/* PVO entry is wired */
238 #define	PVO_MANAGED		0x020		/* PVO entry is managed */
239 #define	PVO_BOOTSTRAP		0x080		/* PVO entry allocated during
240 						   bootstrap */
241 #define PVO_FAKE		0x100		/* fictitious phys page */
242 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
243 #define PVO_ISFAKE(pvo)		((pvo)->pvo_vaddr & PVO_FAKE)
244 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
245 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
246 #define	PVO_PTEGIDX_CLR(pvo)	\
247 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
248 #define	PVO_PTEGIDX_SET(pvo, i)	\
249 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
250 
251 #define	MOEA_PVO_CHECK(pvo)
252 
253 #define LOCK_TABLE() mtx_lock(&moea64_table_mutex)
254 #define UNLOCK_TABLE() mtx_unlock(&moea64_table_mutex);
255 #define ASSERT_TABLE_LOCK() mtx_assert(&moea64_table_mutex, MA_OWNED)
256 
257 struct ofw_map {
258 	vm_offset_t	om_va;
259 	vm_size_t	om_len;
260 	vm_offset_t	om_pa_hi;
261 	vm_offset_t	om_pa_lo;
262 	u_int		om_mode;
263 };
264 
265 /*
266  * Map of physical memory regions.
267  */
268 static struct	mem_region *regions;
269 static struct	mem_region *pregions;
270 extern u_int	phys_avail_count;
271 extern int	regions_sz, pregions_sz;
272 extern int	ofw_real_mode;
273 static struct	ofw_map translations[64];
274 
275 extern struct pmap ofw_pmap;
276 
277 extern void bs_remap_earlyboot(void);
278 
279 
280 /*
281  * Lock for the pteg and pvo tables.
282  */
283 struct mtx	moea64_table_mutex;
284 
285 /*
286  * PTEG data.
287  */
288 static struct	lpteg *moea64_pteg_table;
289 u_int		moea64_pteg_count;
290 u_int		moea64_pteg_mask;
291 
292 /*
293  * PVO data.
294  */
295 struct	pvo_head *moea64_pvo_table;		/* pvo entries by pteg index */
296 /* lists of unmanaged pages */
297 struct	pvo_head moea64_pvo_kunmanaged =
298     LIST_HEAD_INITIALIZER(moea64_pvo_kunmanaged);
299 struct	pvo_head moea64_pvo_unmanaged =
300     LIST_HEAD_INITIALIZER(moea64_pvo_unmanaged);
301 
302 uma_zone_t	moea64_upvo_zone; /* zone for pvo entries for unmanaged pages */
303 uma_zone_t	moea64_mpvo_zone; /* zone for pvo entries for managed pages */
304 
305 vm_offset_t	pvo_allocator_start;
306 vm_offset_t	pvo_allocator_end;
307 
308 #define	BPVO_POOL_SIZE	327680
309 static struct	pvo_entry *moea64_bpvo_pool;
310 static int	moea64_bpvo_pool_index = 0;
311 
312 #define	VSID_NBPW	(sizeof(u_int32_t) * 8)
313 static u_int	moea64_vsid_bitmap[NPMAPS / VSID_NBPW];
314 
315 static boolean_t moea64_initialized = FALSE;
316 
317 /*
318  * Statistics.
319  */
320 u_int	moea64_pte_valid = 0;
321 u_int	moea64_pte_overflow = 0;
322 u_int	moea64_pvo_entries = 0;
323 u_int	moea64_pvo_enter_calls = 0;
324 u_int	moea64_pvo_remove_calls = 0;
325 SYSCTL_INT(_machdep, OID_AUTO, moea64_pte_valid, CTLFLAG_RD,
326     &moea64_pte_valid, 0, "");
327 SYSCTL_INT(_machdep, OID_AUTO, moea64_pte_overflow, CTLFLAG_RD,
328     &moea64_pte_overflow, 0, "");
329 SYSCTL_INT(_machdep, OID_AUTO, moea64_pvo_entries, CTLFLAG_RD,
330     &moea64_pvo_entries, 0, "");
331 SYSCTL_INT(_machdep, OID_AUTO, moea64_pvo_enter_calls, CTLFLAG_RD,
332     &moea64_pvo_enter_calls, 0, "");
333 SYSCTL_INT(_machdep, OID_AUTO, moea64_pvo_remove_calls, CTLFLAG_RD,
334     &moea64_pvo_remove_calls, 0, "");
335 
336 vm_offset_t	moea64_scratchpage_va[2];
337 struct	pvo_entry *moea64_scratchpage_pvo[2];
338 struct	lpte 	*moea64_scratchpage_pte[2];
339 struct	mtx	moea64_scratchpage_mtx;
340 
341 /*
342  * Allocate physical memory for use in moea64_bootstrap.
343  */
344 static vm_offset_t	moea64_bootstrap_alloc(vm_size_t, u_int);
345 
346 /*
347  * PTE calls.
348  */
349 static int		moea64_pte_insert(u_int, struct lpte *);
350 
351 /*
352  * PVO calls.
353  */
354 static int	moea64_pvo_enter(pmap_t, uma_zone_t, struct pvo_head *,
355 		    vm_offset_t, vm_offset_t, uint64_t, int, int);
356 static void	moea64_pvo_remove(struct pvo_entry *, int);
357 static struct	pvo_entry *moea64_pvo_find_va(pmap_t, vm_offset_t, int *);
358 static struct	lpte *moea64_pvo_to_pte(const struct pvo_entry *, int);
359 
360 /*
361  * Utility routines.
362  */
363 static void		moea64_bridge_bootstrap(mmu_t mmup,
364 			    vm_offset_t kernelstart, vm_offset_t kernelend);
365 static void		moea64_bridge_cpu_bootstrap(mmu_t, int ap);
366 static void		moea64_enter_locked(pmap_t, vm_offset_t, vm_page_t,
367 			    vm_prot_t, boolean_t);
368 static boolean_t	moea64_query_bit(vm_page_t, u_int64_t);
369 static u_int		moea64_clear_bit(vm_page_t, u_int64_t, u_int64_t *);
370 static void		moea64_kremove(mmu_t, vm_offset_t);
371 static void		moea64_syncicache(pmap_t pmap, vm_offset_t va,
372 			    vm_offset_t pa, vm_size_t sz);
373 static void		tlbia(void);
374 
375 /*
376  * Kernel MMU interface
377  */
378 void moea64_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
379 void moea64_clear_modify(mmu_t, vm_page_t);
380 void moea64_clear_reference(mmu_t, vm_page_t);
381 void moea64_copy_page(mmu_t, vm_page_t, vm_page_t);
382 void moea64_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t, vm_prot_t, boolean_t);
383 void moea64_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t, vm_page_t,
384     vm_prot_t);
385 void moea64_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t, vm_prot_t);
386 vm_paddr_t moea64_extract(mmu_t, pmap_t, vm_offset_t);
387 vm_page_t moea64_extract_and_hold(mmu_t, pmap_t, vm_offset_t, vm_prot_t);
388 void moea64_init(mmu_t);
389 boolean_t moea64_is_modified(mmu_t, vm_page_t);
390 boolean_t moea64_ts_referenced(mmu_t, vm_page_t);
391 vm_offset_t moea64_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t, int);
392 boolean_t moea64_page_exists_quick(mmu_t, pmap_t, vm_page_t);
393 int moea64_page_wired_mappings(mmu_t, vm_page_t);
394 void moea64_pinit(mmu_t, pmap_t);
395 void moea64_pinit0(mmu_t, pmap_t);
396 void moea64_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t, vm_prot_t);
397 void moea64_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
398 void moea64_qremove(mmu_t, vm_offset_t, int);
399 void moea64_release(mmu_t, pmap_t);
400 void moea64_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
401 void moea64_remove_all(mmu_t, vm_page_t);
402 void moea64_remove_write(mmu_t, vm_page_t);
403 void moea64_zero_page(mmu_t, vm_page_t);
404 void moea64_zero_page_area(mmu_t, vm_page_t, int, int);
405 void moea64_zero_page_idle(mmu_t, vm_page_t);
406 void moea64_activate(mmu_t, struct thread *);
407 void moea64_deactivate(mmu_t, struct thread *);
408 void *moea64_mapdev(mmu_t, vm_offset_t, vm_size_t);
409 void moea64_unmapdev(mmu_t, vm_offset_t, vm_size_t);
410 vm_offset_t moea64_kextract(mmu_t, vm_offset_t);
411 void moea64_kenter(mmu_t, vm_offset_t, vm_offset_t);
412 boolean_t moea64_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
413 static void moea64_sync_icache(mmu_t, pmap_t, vm_offset_t, vm_size_t);
414 
415 static mmu_method_t moea64_bridge_methods[] = {
416 	MMUMETHOD(mmu_change_wiring,	moea64_change_wiring),
417 	MMUMETHOD(mmu_clear_modify,	moea64_clear_modify),
418 	MMUMETHOD(mmu_clear_reference,	moea64_clear_reference),
419 	MMUMETHOD(mmu_copy_page,	moea64_copy_page),
420 	MMUMETHOD(mmu_enter,		moea64_enter),
421 	MMUMETHOD(mmu_enter_object,	moea64_enter_object),
422 	MMUMETHOD(mmu_enter_quick,	moea64_enter_quick),
423 	MMUMETHOD(mmu_extract,		moea64_extract),
424 	MMUMETHOD(mmu_extract_and_hold,	moea64_extract_and_hold),
425 	MMUMETHOD(mmu_init,		moea64_init),
426 	MMUMETHOD(mmu_is_modified,	moea64_is_modified),
427 	MMUMETHOD(mmu_ts_referenced,	moea64_ts_referenced),
428 	MMUMETHOD(mmu_map,     		moea64_map),
429 	MMUMETHOD(mmu_page_exists_quick,moea64_page_exists_quick),
430 	MMUMETHOD(mmu_page_wired_mappings,moea64_page_wired_mappings),
431 	MMUMETHOD(mmu_pinit,		moea64_pinit),
432 	MMUMETHOD(mmu_pinit0,		moea64_pinit0),
433 	MMUMETHOD(mmu_protect,		moea64_protect),
434 	MMUMETHOD(mmu_qenter,		moea64_qenter),
435 	MMUMETHOD(mmu_qremove,		moea64_qremove),
436 	MMUMETHOD(mmu_release,		moea64_release),
437 	MMUMETHOD(mmu_remove,		moea64_remove),
438 	MMUMETHOD(mmu_remove_all,      	moea64_remove_all),
439 	MMUMETHOD(mmu_remove_write,	moea64_remove_write),
440 	MMUMETHOD(mmu_sync_icache,	moea64_sync_icache),
441 	MMUMETHOD(mmu_zero_page,       	moea64_zero_page),
442 	MMUMETHOD(mmu_zero_page_area,	moea64_zero_page_area),
443 	MMUMETHOD(mmu_zero_page_idle,	moea64_zero_page_idle),
444 	MMUMETHOD(mmu_activate,		moea64_activate),
445 	MMUMETHOD(mmu_deactivate,      	moea64_deactivate),
446 
447 	/* Internal interfaces */
448 	MMUMETHOD(mmu_bootstrap,       	moea64_bridge_bootstrap),
449 	MMUMETHOD(mmu_cpu_bootstrap,   	moea64_bridge_cpu_bootstrap),
450 	MMUMETHOD(mmu_mapdev,		moea64_mapdev),
451 	MMUMETHOD(mmu_unmapdev,		moea64_unmapdev),
452 	MMUMETHOD(mmu_kextract,		moea64_kextract),
453 	MMUMETHOD(mmu_kenter,		moea64_kenter),
454 	MMUMETHOD(mmu_dev_direct_mapped,moea64_dev_direct_mapped),
455 
456 	{ 0, 0 }
457 };
458 
459 static mmu_def_t oea64_bridge_mmu = {
460 	MMU_TYPE_G5,
461 	moea64_bridge_methods,
462 	0
463 };
464 MMU_DEF(oea64_bridge_mmu);
465 
466 static __inline u_int
467 va_to_pteg(uint64_t vsid, vm_offset_t addr)
468 {
469 	u_int hash;
470 
471 	hash = vsid ^ (((uint64_t)addr & ADDR_PIDX) >>
472 	    ADDR_PIDX_SHFT);
473 	return (hash & moea64_pteg_mask);
474 }
475 
476 static __inline struct pvo_head *
477 pa_to_pvoh(vm_offset_t pa, vm_page_t *pg_p)
478 {
479 	struct	vm_page *pg;
480 
481 	pg = PHYS_TO_VM_PAGE(pa);
482 
483 	if (pg_p != NULL)
484 		*pg_p = pg;
485 
486 	if (pg == NULL)
487 		return (&moea64_pvo_unmanaged);
488 
489 	return (&pg->md.mdpg_pvoh);
490 }
491 
492 static __inline struct pvo_head *
493 vm_page_to_pvoh(vm_page_t m)
494 {
495 
496 	return (&m->md.mdpg_pvoh);
497 }
498 
499 static __inline void
500 moea64_attr_clear(vm_page_t m, u_int64_t ptebit)
501 {
502 
503 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
504 	m->md.mdpg_attrs &= ~ptebit;
505 }
506 
507 static __inline u_int64_t
508 moea64_attr_fetch(vm_page_t m)
509 {
510 
511 	return (m->md.mdpg_attrs);
512 }
513 
514 static __inline void
515 moea64_attr_save(vm_page_t m, u_int64_t ptebit)
516 {
517 
518 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
519 	m->md.mdpg_attrs |= ptebit;
520 }
521 
522 static __inline int
523 moea64_pte_compare(const struct lpte *pt, const struct lpte *pvo_pt)
524 {
525 	if (pt->pte_hi == pvo_pt->pte_hi)
526 		return (1);
527 
528 	return (0);
529 }
530 
531 static __inline int
532 moea64_pte_match(struct lpte *pt, uint64_t vsid, vm_offset_t va, int which)
533 {
534 	return (pt->pte_hi & ~LPTE_VALID) ==
535 	    ((vsid << LPTE_VSID_SHIFT) |
536 	    ((uint64_t)(va >> ADDR_API_SHFT64) & LPTE_API) | which);
537 }
538 
539 static __inline void
540 moea64_pte_create(struct lpte *pt, uint64_t vsid, vm_offset_t va,
541     uint64_t pte_lo)
542 {
543 	ASSERT_TABLE_LOCK();
544 
545 	/*
546 	 * Construct a PTE.  Default to IMB initially.  Valid bit only gets
547 	 * set when the real pte is set in memory.
548 	 *
549 	 * Note: Don't set the valid bit for correct operation of tlb update.
550 	 */
551 	pt->pte_hi = (vsid << LPTE_VSID_SHIFT) |
552 	    (((uint64_t)(va & ADDR_PIDX) >> ADDR_API_SHFT64) & LPTE_API);
553 
554 	pt->pte_lo = pte_lo;
555 }
556 
557 static __inline void
558 moea64_pte_synch(struct lpte *pt, struct lpte *pvo_pt)
559 {
560 
561 	ASSERT_TABLE_LOCK();
562 
563 	pvo_pt->pte_lo |= pt->pte_lo & (LPTE_REF | LPTE_CHG);
564 }
565 
566 static __inline void
567 moea64_pte_clear(struct lpte *pt, pmap_t pmap, vm_offset_t va, u_int64_t ptebit)
568 {
569 	ASSERT_TABLE_LOCK();
570 
571 	/*
572 	 * As shown in Section 7.6.3.2.3
573 	 */
574 	pt->pte_lo &= ~ptebit;
575 	TLBIE(pmap,va);
576 }
577 
578 static __inline void
579 moea64_pte_set(struct lpte *pt, struct lpte *pvo_pt)
580 {
581 
582 	ASSERT_TABLE_LOCK();
583 	pvo_pt->pte_hi |= LPTE_VALID;
584 
585 	/*
586 	 * Update the PTE as defined in section 7.6.3.1.
587 	 * Note that the REF/CHG bits are from pvo_pt and thus should have
588 	 * been saved so this routine can restore them (if desired).
589 	 */
590 	pt->pte_lo = pvo_pt->pte_lo;
591 	EIEIO();
592 	pt->pte_hi = pvo_pt->pte_hi;
593 	SYNC();
594 	moea64_pte_valid++;
595 }
596 
597 static __inline void
598 moea64_pte_unset(struct lpte *pt, struct lpte *pvo_pt, pmap_t pmap, vm_offset_t va)
599 {
600 	ASSERT_TABLE_LOCK();
601 	pvo_pt->pte_hi &= ~LPTE_VALID;
602 
603 	/*
604 	 * Force the reg & chg bits back into the PTEs.
605 	 */
606 	SYNC();
607 
608 	/*
609 	 * Invalidate the pte.
610 	 */
611 	pt->pte_hi &= ~LPTE_VALID;
612 
613 	TLBIE(pmap,va);
614 
615 	/*
616 	 * Save the reg & chg bits.
617 	 */
618 	moea64_pte_synch(pt, pvo_pt);
619 	moea64_pte_valid--;
620 }
621 
622 static __inline void
623 moea64_pte_change(struct lpte *pt, struct lpte *pvo_pt, pmap_t pmap, vm_offset_t va)
624 {
625 
626 	/*
627 	 * Invalidate the PTE
628 	 */
629 	moea64_pte_unset(pt, pvo_pt, pmap, va);
630 	moea64_pte_set(pt, pvo_pt);
631 }
632 
633 static __inline uint64_t
634 moea64_calc_wimg(vm_offset_t pa)
635 {
636 	uint64_t pte_lo;
637 	int i;
638 
639 	/*
640 	 * Assume the page is cache inhibited and access is guarded unless
641 	 * it's in our available memory array.
642 	 */
643 	pte_lo = LPTE_I | LPTE_G;
644 	for (i = 0; i < pregions_sz; i++) {
645 		if ((pa >= pregions[i].mr_start) &&
646 		    (pa < (pregions[i].mr_start + pregions[i].mr_size))) {
647 			pte_lo &= ~(LPTE_I | LPTE_G);
648 			pte_lo |= LPTE_M;
649 			break;
650 		}
651 	}
652 
653 	return pte_lo;
654 }
655 
656 /*
657  * Quick sort callout for comparing memory regions.
658  */
659 static int	mr_cmp(const void *a, const void *b);
660 static int	om_cmp(const void *a, const void *b);
661 
662 static int
663 mr_cmp(const void *a, const void *b)
664 {
665 	const struct	mem_region *regiona;
666 	const struct	mem_region *regionb;
667 
668 	regiona = a;
669 	regionb = b;
670 	if (regiona->mr_start < regionb->mr_start)
671 		return (-1);
672 	else if (regiona->mr_start > regionb->mr_start)
673 		return (1);
674 	else
675 		return (0);
676 }
677 
678 static int
679 om_cmp(const void *a, const void *b)
680 {
681 	const struct	ofw_map *mapa;
682 	const struct	ofw_map *mapb;
683 
684 	mapa = a;
685 	mapb = b;
686 	if (mapa->om_pa_hi < mapb->om_pa_hi)
687 		return (-1);
688 	else if (mapa->om_pa_hi > mapb->om_pa_hi)
689 		return (1);
690 	else if (mapa->om_pa_lo < mapb->om_pa_lo)
691 		return (-1);
692 	else if (mapa->om_pa_lo > mapb->om_pa_lo)
693 		return (1);
694 	else
695 		return (0);
696 }
697 
698 static void
699 moea64_bridge_cpu_bootstrap(mmu_t mmup, int ap)
700 {
701 	int i = 0;
702 
703 	/*
704 	 * Initialize segment registers and MMU
705 	 */
706 
707 	mtmsr(mfmsr() & ~PSL_DR & ~PSL_IR); isync();
708 	for (i = 0; i < 16; i++) {
709 		mtsrin(i << ADDR_SR_SHFT, kernel_pmap->pm_sr[i]);
710 	}
711 	__asm __volatile ("sync; mtsdr1 %0; isync"
712 	    :: "r"((u_int)moea64_pteg_table
713 		     | (32 - cntlzw(moea64_pteg_mask >> 11))));
714 	tlbia();
715 }
716 
717 static void
718 moea64_bridge_bootstrap(mmu_t mmup, vm_offset_t kernelstart, vm_offset_t kernelend)
719 {
720 	ihandle_t	mmui;
721 	phandle_t	chosen;
722 	phandle_t	mmu;
723 	int		sz;
724 	int		i, j;
725 	int		ofw_mappings;
726 	vm_size_t	size, physsz, hwphyssz;
727 	vm_offset_t	pa, va, off;
728 	uint32_t	msr;
729 	void		*dpcpu;
730 
731 	/* We don't have a direct map since there is no BAT */
732 	hw_direct_map = 0;
733 
734 	/* Make sure battable is zero, since we have no BAT */
735 	for (i = 0; i < 16; i++) {
736 		battable[i].batu = 0;
737 		battable[i].batl = 0;
738 	}
739 
740 	/* Get physical memory regions from firmware */
741 	mem_regions(&pregions, &pregions_sz, &regions, &regions_sz);
742 	CTR0(KTR_PMAP, "moea64_bootstrap: physical memory");
743 
744 	qsort(pregions, pregions_sz, sizeof(*pregions), mr_cmp);
745 	if (sizeof(phys_avail)/sizeof(phys_avail[0]) < regions_sz)
746 		panic("moea64_bootstrap: phys_avail too small");
747 	qsort(regions, regions_sz, sizeof(*regions), mr_cmp);
748 	phys_avail_count = 0;
749 	physsz = 0;
750 	hwphyssz = 0;
751 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
752 	for (i = 0, j = 0; i < regions_sz; i++, j += 2) {
753 		CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start,
754 		    regions[i].mr_start + regions[i].mr_size,
755 		    regions[i].mr_size);
756 		if (hwphyssz != 0 &&
757 		    (physsz + regions[i].mr_size) >= hwphyssz) {
758 			if (physsz < hwphyssz) {
759 				phys_avail[j] = regions[i].mr_start;
760 				phys_avail[j + 1] = regions[i].mr_start +
761 				    hwphyssz - physsz;
762 				physsz = hwphyssz;
763 				phys_avail_count++;
764 			}
765 			break;
766 		}
767 		phys_avail[j] = regions[i].mr_start;
768 		phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size;
769 		phys_avail_count++;
770 		physsz += regions[i].mr_size;
771 	}
772 	physmem = btoc(physsz);
773 
774 	/*
775 	 * Allocate PTEG table.
776 	 */
777 #ifdef PTEGCOUNT
778 	moea64_pteg_count = PTEGCOUNT;
779 #else
780 	moea64_pteg_count = 0x1000;
781 
782 	while (moea64_pteg_count < physmem)
783 		moea64_pteg_count <<= 1;
784 #endif /* PTEGCOUNT */
785 
786 	size = moea64_pteg_count * sizeof(struct lpteg);
787 	CTR2(KTR_PMAP, "moea64_bootstrap: %d PTEGs, %d bytes",
788 	    moea64_pteg_count, size);
789 
790 	/*
791 	 * We now need to allocate memory. This memory, to be allocated,
792 	 * has to reside in a page table. The page table we are about to
793 	 * allocate. We don't have BAT. So drop to data real mode for a minute
794 	 * as a measure of last resort. We do this a couple times.
795 	 */
796 
797 	moea64_pteg_table = (struct lpteg *)moea64_bootstrap_alloc(size, size);
798 	DISABLE_TRANS(msr);
799 	bzero((void *)moea64_pteg_table, moea64_pteg_count * sizeof(struct lpteg));
800 	ENABLE_TRANS(msr);
801 
802 	moea64_pteg_mask = moea64_pteg_count - 1;
803 
804 	CTR1(KTR_PMAP, "moea64_bootstrap: PTEG table at %p", moea64_pteg_table);
805 
806 	/*
807 	 * Allocate pv/overflow lists.
808 	 */
809 	size = sizeof(struct pvo_head) * moea64_pteg_count;
810 
811 	moea64_pvo_table = (struct pvo_head *)moea64_bootstrap_alloc(size,
812 	    PAGE_SIZE);
813 	CTR1(KTR_PMAP, "moea64_bootstrap: PVO table at %p", moea64_pvo_table);
814 
815 	DISABLE_TRANS(msr);
816 	for (i = 0; i < moea64_pteg_count; i++)
817 		LIST_INIT(&moea64_pvo_table[i]);
818 	ENABLE_TRANS(msr);
819 
820 	/*
821 	 * Initialize the lock that synchronizes access to the pteg and pvo
822 	 * tables.
823 	 */
824 	mtx_init(&moea64_table_mutex, "pmap table", NULL, MTX_DEF |
825 	    MTX_RECURSE);
826 
827 	/*
828 	 * Initialise the unmanaged pvo pool.
829 	 */
830 	moea64_bpvo_pool = (struct pvo_entry *)moea64_bootstrap_alloc(
831 		BPVO_POOL_SIZE*sizeof(struct pvo_entry), 0);
832 	moea64_bpvo_pool_index = 0;
833 
834 	/*
835 	 * Make sure kernel vsid is allocated as well as VSID 0.
836 	 */
837 	moea64_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW]
838 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
839 	moea64_vsid_bitmap[0] |= 1;
840 
841 	/*
842 	 * Initialize the kernel pmap (which is statically allocated).
843 	 */
844 	for (i = 0; i < 16; i++)
845 		kernel_pmap->pm_sr[i] = EMPTY_SEGMENT + i;
846 
847 	kernel_pmap->pmap_phys = kernel_pmap;
848 	kernel_pmap->pm_active = ~0;
849 
850 	PMAP_LOCK_INIT(kernel_pmap);
851 
852 	/*
853 	 * Now map in all the other buffers we allocated earlier
854 	 */
855 
856 	DISABLE_TRANS(msr);
857 	size = moea64_pteg_count * sizeof(struct lpteg);
858 	off = (vm_offset_t)(moea64_pteg_table);
859 	for (pa = off; pa < off + size; pa += PAGE_SIZE)
860 		moea64_kenter(mmup, pa, pa);
861 	size = sizeof(struct pvo_head) * moea64_pteg_count;
862 	off = (vm_offset_t)(moea64_pvo_table);
863 	for (pa = off; pa < off + size; pa += PAGE_SIZE)
864 		moea64_kenter(mmup, pa, pa);
865 	size = BPVO_POOL_SIZE*sizeof(struct pvo_entry);
866 	off = (vm_offset_t)(moea64_bpvo_pool);
867 	for (pa = off; pa < off + size; pa += PAGE_SIZE)
868 		moea64_kenter(mmup, pa, pa);
869 	ENABLE_TRANS(msr);
870 
871 	/*
872 	 * Map certain important things, like ourselves and the exception
873 	 * vectors
874 	 */
875 
876 	DISABLE_TRANS(msr);
877 	for (pa = kernelstart & ~PAGE_MASK; pa < kernelend; pa += PAGE_SIZE)
878 		moea64_kenter(mmup, pa, pa);
879 	for (pa = EXC_RSVD; pa < EXC_LAST; pa += PAGE_SIZE)
880 		moea64_kenter(mmup, pa, pa);
881 	ENABLE_TRANS(msr);
882 
883 	if (!ofw_real_mode) {
884 	    /*
885 	     * Set up the Open Firmware pmap and add its mappings.
886 	     */
887 
888 	    moea64_pinit(mmup, &ofw_pmap);
889 	    ofw_pmap.pm_sr[KERNEL_SR] = kernel_pmap->pm_sr[KERNEL_SR];
890 	    ofw_pmap.pm_sr[KERNEL2_SR] = kernel_pmap->pm_sr[KERNEL2_SR];
891 
892 	    if ((chosen = OF_finddevice("/chosen")) == -1)
893 		panic("moea64_bootstrap: can't find /chosen");
894 	    OF_getprop(chosen, "mmu", &mmui, 4);
895 	    if ((mmu = OF_instance_to_package(mmui)) == -1)
896 		panic("moea64_bootstrap: can't get mmu package");
897 	    if ((sz = OF_getproplen(mmu, "translations")) == -1)
898 		panic("moea64_bootstrap: can't get ofw translation count");
899 
900 	    bzero(translations, sz);
901 	    if (OF_getprop(mmu, "translations", translations, sz) == -1)
902 		panic("moea64_bootstrap: can't get ofw translations");
903 
904 	    CTR0(KTR_PMAP, "moea64_bootstrap: translations");
905 	    sz /= sizeof(*translations);
906 	    qsort(translations, sz, sizeof (*translations), om_cmp);
907 
908 	    for (i = 0, ofw_mappings = 0; i < sz; i++) {
909 		CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x",
910 		    (uint32_t)(translations[i].om_pa_lo), translations[i].om_va,
911 		    translations[i].om_len);
912 
913 		if (translations[i].om_pa_lo % PAGE_SIZE)
914 			panic("OFW translation not page-aligned!");
915 
916 		if (translations[i].om_pa_hi)
917 			panic("OFW translations above 32-bit boundary!");
918 
919 		/* Now enter the pages for this mapping */
920 
921 		/*
922 		 * Lock the ofw pmap. pmap_kenter(), which we use for the
923 		 * pages the kernel also needs, does its own locking.
924 		 */
925 		PMAP_LOCK(&ofw_pmap);
926 		DISABLE_TRANS(msr);
927 		for (off = 0; off < translations[i].om_len; off += PAGE_SIZE) {
928 			struct vm_page m;
929 
930 			/* Map low memory mappings into the kernel pmap, too.
931 			 * These are typically mappings made by the loader,
932 			 * so we need them if we want to keep executing. */
933 
934 			if (translations[i].om_va + off < SEGMENT_LENGTH)
935 				moea64_kenter(mmup, translations[i].om_va + off,
936 				    translations[i].om_va + off);
937 
938 			m.phys_addr = translations[i].om_pa_lo + off;
939 			moea64_enter_locked(&ofw_pmap,
940 			    translations[i].om_va + off, &m, VM_PROT_ALL, 1);
941 
942 			ofw_mappings++;
943 		}
944 		ENABLE_TRANS(msr);
945 		PMAP_UNLOCK(&ofw_pmap);
946 	    }
947 	}
948 
949 #ifdef SMP
950 	TLBSYNC();
951 #endif
952 
953 	/*
954 	 * Calculate the last available physical address.
955 	 */
956 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
957 		;
958 	Maxmem = powerpc_btop(phys_avail[i + 1]);
959 
960 	/*
961 	 * Initialize MMU and remap early physical mappings
962 	 */
963 	moea64_bridge_cpu_bootstrap(mmup,0);
964 	mtmsr(mfmsr() | PSL_DR | PSL_IR); isync();
965 	pmap_bootstrapped++;
966 	bs_remap_earlyboot();
967 
968 	/*
969 	 * Set the start and end of kva.
970 	 */
971 	virtual_avail = VM_MIN_KERNEL_ADDRESS;
972 	virtual_end = VM_MAX_KERNEL_ADDRESS;
973 
974 	/*
975 	 * Allocate some stupid buffer regions.
976 	 */
977 
978 	pvo_allocator_start = virtual_avail;
979 	virtual_avail += SEGMENT_LENGTH/4;
980 	pvo_allocator_end = virtual_avail;
981 
982 	/*
983 	 * Allocate some things for page zeroing
984 	 */
985 
986 	mtx_init(&moea64_scratchpage_mtx, "pvo zero page", NULL, MTX_DEF);
987 	for (i = 0; i < 2; i++) {
988 		moea64_scratchpage_va[i] = virtual_avail;
989 		virtual_avail += PAGE_SIZE;
990 
991 		moea64_kenter(mmup,moea64_scratchpage_va[i],kernelstart);
992 
993 		LOCK_TABLE();
994 		moea64_scratchpage_pvo[i] = moea64_pvo_find_va(kernel_pmap,
995 		    moea64_scratchpage_va[i],&j);
996 		moea64_scratchpage_pte[i] = moea64_pvo_to_pte(
997 		    moea64_scratchpage_pvo[i],j);
998 		UNLOCK_TABLE();
999 	}
1000 
1001 	/*
1002 	 * Allocate a kernel stack with a guard page for thread0 and map it
1003 	 * into the kernel page map.
1004 	 */
1005 	pa = moea64_bootstrap_alloc(KSTACK_PAGES * PAGE_SIZE, PAGE_SIZE);
1006 	va = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1007 	virtual_avail = va + KSTACK_PAGES * PAGE_SIZE;
1008 	CTR2(KTR_PMAP, "moea_bootstrap: kstack0 at %#x (%#x)", pa, va);
1009 	thread0.td_kstack = va;
1010 	thread0.td_kstack_pages = KSTACK_PAGES;
1011 	for (i = 0; i < KSTACK_PAGES; i++) {
1012 		moea64_kenter(mmup, va, pa);;
1013 		pa += PAGE_SIZE;
1014 		va += PAGE_SIZE;
1015 	}
1016 
1017 	/*
1018 	 * Allocate virtual address space for the message buffer.
1019 	 */
1020 	pa = msgbuf_phys = moea64_bootstrap_alloc(MSGBUF_SIZE, PAGE_SIZE);
1021 	msgbufp = (struct msgbuf *)virtual_avail;
1022 	va = virtual_avail;
1023 	virtual_avail += round_page(MSGBUF_SIZE);
1024 	while (va < virtual_avail) {
1025 		moea64_kenter(mmup, va, pa);;
1026 		pa += PAGE_SIZE;
1027 		va += PAGE_SIZE;
1028 	}
1029 
1030 	/*
1031 	 * Allocate virtual address space for the dynamic percpu area.
1032 	 */
1033 	pa = moea64_bootstrap_alloc(DPCPU_SIZE, PAGE_SIZE);
1034 	dpcpu = (void *)virtual_avail;
1035 	va = virtual_avail;
1036 	virtual_avail += DPCPU_SIZE;
1037 	while (va < virtual_avail) {
1038 		moea64_kenter(mmup, va, pa);;
1039 		pa += PAGE_SIZE;
1040 		va += PAGE_SIZE;
1041 	}
1042 	dpcpu_init(dpcpu, 0);
1043 }
1044 
1045 /*
1046  * Activate a user pmap.  The pmap must be activated before it's address
1047  * space can be accessed in any way.
1048  */
1049 void
1050 moea64_activate(mmu_t mmu, struct thread *td)
1051 {
1052 	pmap_t	pm, pmr;
1053 
1054 	/*
1055 	 * Load all the data we need up front to encourage the compiler to
1056 	 * not issue any loads while we have interrupts disabled below.
1057 	 */
1058 	pm = &td->td_proc->p_vmspace->vm_pmap;
1059 	pmr = pm->pmap_phys;
1060 
1061 	pm->pm_active |= PCPU_GET(cpumask);
1062 	PCPU_SET(curpmap, pmr);
1063 }
1064 
1065 void
1066 moea64_deactivate(mmu_t mmu, struct thread *td)
1067 {
1068 	pmap_t	pm;
1069 
1070 	pm = &td->td_proc->p_vmspace->vm_pmap;
1071 	pm->pm_active &= ~(PCPU_GET(cpumask));
1072 	PCPU_SET(curpmap, NULL);
1073 }
1074 
1075 void
1076 moea64_change_wiring(mmu_t mmu, pmap_t pm, vm_offset_t va, boolean_t wired)
1077 {
1078 	struct	pvo_entry *pvo;
1079 
1080 	PMAP_LOCK(pm);
1081 	pvo = moea64_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1082 
1083 	if (pvo != NULL) {
1084 		if (wired) {
1085 			if ((pvo->pvo_vaddr & PVO_WIRED) == 0)
1086 				pm->pm_stats.wired_count++;
1087 			pvo->pvo_vaddr |= PVO_WIRED;
1088 		} else {
1089 			if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
1090 				pm->pm_stats.wired_count--;
1091 			pvo->pvo_vaddr &= ~PVO_WIRED;
1092 		}
1093 	}
1094 	PMAP_UNLOCK(pm);
1095 }
1096 
1097 /*
1098  * Zero a page of physical memory by temporarily mapping it into the tlb.
1099  */
1100 void
1101 moea64_zero_page(mmu_t mmu, vm_page_t m)
1102 {
1103 	moea64_zero_page_area(mmu,m,0,PAGE_SIZE);
1104 }
1105 
1106 /*
1107  * This goes through and sets the physical address of our
1108  * special scratch PTE to the PA we want to zero or copy. Because
1109  * of locking issues (this can get called in pvo_enter() by
1110  * the UMA allocator), we can't use most other utility functions here
1111  */
1112 
1113 static __inline
1114 void moea64_set_scratchpage_pa(int which, vm_offset_t pa) {
1115 	moea64_scratchpage_pvo[which]->pvo_pte.lpte.pte_lo &=
1116 	    (~LPTE_WIMG & ~LPTE_RPGN);
1117 	moea64_scratchpage_pvo[which]->pvo_pte.lpte.pte_lo |=
1118 	    moea64_calc_wimg(pa) | (uint64_t)pa;
1119 
1120 	moea64_scratchpage_pte[which]->pte_hi &= ~LPTE_VALID;
1121 	TLBIE(kernel_pmap, moea64_scratchpage_va[which]);
1122 
1123 	moea64_scratchpage_pte[which]->pte_lo =
1124 	    moea64_scratchpage_pvo[which]->pvo_pte.lpte.pte_lo;
1125 	EIEIO();
1126 
1127 	moea64_scratchpage_pte[which]->pte_hi |= LPTE_VALID;
1128 	TLBIE(kernel_pmap, moea64_scratchpage_va[which]);
1129 }
1130 
1131 void
1132 moea64_copy_page(mmu_t mmu, vm_page_t msrc, vm_page_t mdst)
1133 {
1134 	vm_offset_t	dst;
1135 	vm_offset_t	src;
1136 
1137 	dst = VM_PAGE_TO_PHYS(mdst);
1138 	src = VM_PAGE_TO_PHYS(msrc);
1139 
1140 	mtx_lock(&moea64_scratchpage_mtx);
1141 
1142 	moea64_set_scratchpage_pa(0,src);
1143 	moea64_set_scratchpage_pa(1,dst);
1144 
1145 	kcopy((void *)moea64_scratchpage_va[0],
1146 	    (void *)moea64_scratchpage_va[1], PAGE_SIZE);
1147 
1148 	__syncicache((void *)moea64_scratchpage_va[1],PAGE_SIZE);
1149 
1150 	mtx_unlock(&moea64_scratchpage_mtx);
1151 }
1152 
1153 void
1154 moea64_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
1155 {
1156 	vm_offset_t pa = VM_PAGE_TO_PHYS(m);
1157 
1158 	if (!moea64_initialized)
1159 		panic("moea64_zero_page: can't zero pa %#x", pa);
1160 	if (size + off > PAGE_SIZE)
1161 		panic("moea64_zero_page: size + off > PAGE_SIZE");
1162 
1163 	mtx_lock(&moea64_scratchpage_mtx);
1164 
1165 	moea64_set_scratchpage_pa(0,pa);
1166 	bzero((caddr_t)moea64_scratchpage_va[0] + off, size);
1167 	__syncicache((void *)moea64_scratchpage_va[0],PAGE_SIZE);
1168 
1169 	mtx_unlock(&moea64_scratchpage_mtx);
1170 }
1171 
1172 void
1173 moea64_zero_page_idle(mmu_t mmu, vm_page_t m)
1174 {
1175 
1176 	moea64_zero_page(mmu, m);
1177 }
1178 
1179 /*
1180  * Map the given physical page at the specified virtual address in the
1181  * target pmap with the protection requested.  If specified the page
1182  * will be wired down.
1183  */
1184 void
1185 moea64_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1186     vm_prot_t prot, boolean_t wired)
1187 {
1188 
1189 	vm_page_lock_queues();
1190 	PMAP_LOCK(pmap);
1191 	moea64_enter_locked(pmap, va, m, prot, wired);
1192 	vm_page_unlock_queues();
1193 	PMAP_UNLOCK(pmap);
1194 }
1195 
1196 /*
1197  * Map the given physical page at the specified virtual address in the
1198  * target pmap with the protection requested.  If specified the page
1199  * will be wired down.
1200  *
1201  * The page queues and pmap must be locked.
1202  */
1203 
1204 static void
1205 moea64_enter_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1206     boolean_t wired)
1207 {
1208 	struct		pvo_head *pvo_head;
1209 	uma_zone_t	zone;
1210 	vm_page_t	pg;
1211 	uint64_t	pte_lo;
1212 	u_int		pvo_flags;
1213 	int		error;
1214 
1215 	if (!moea64_initialized) {
1216 		pvo_head = &moea64_pvo_kunmanaged;
1217 		pg = NULL;
1218 		zone = moea64_upvo_zone;
1219 		pvo_flags = 0;
1220 	} else {
1221 		pvo_head = vm_page_to_pvoh(m);
1222 		pg = m;
1223 		zone = moea64_mpvo_zone;
1224 		pvo_flags = PVO_MANAGED;
1225 	}
1226 
1227 	if (pmap_bootstrapped)
1228 		mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1229 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1230 
1231 	/* XXX change the pvo head for fake pages */
1232 	if ((m->flags & PG_FICTITIOUS) == PG_FICTITIOUS) {
1233 		pvo_flags &= ~PVO_MANAGED;
1234 		pvo_head = &moea64_pvo_kunmanaged;
1235 		zone = moea64_upvo_zone;
1236 	}
1237 
1238 	pte_lo = moea64_calc_wimg(VM_PAGE_TO_PHYS(m));
1239 
1240 	if (prot & VM_PROT_WRITE) {
1241 		pte_lo |= LPTE_BW;
1242 		if (pmap_bootstrapped)
1243 			vm_page_flag_set(m, PG_WRITEABLE);
1244 	} else
1245 		pte_lo |= LPTE_BR;
1246 
1247 	if (prot & VM_PROT_EXECUTE)
1248 		pvo_flags |= VM_PROT_EXECUTE;
1249 
1250 	if (wired)
1251 		pvo_flags |= PVO_WIRED;
1252 
1253 	if ((m->flags & PG_FICTITIOUS) != 0)
1254 		pvo_flags |= PVO_FAKE;
1255 
1256 	error = moea64_pvo_enter(pmap, zone, pvo_head, va, VM_PAGE_TO_PHYS(m),
1257 	    pte_lo, pvo_flags, 0);
1258 
1259 	if (pmap == kernel_pmap)
1260 		TLBIE(pmap, va);
1261 
1262 	/*
1263 	 * Flush the page from the instruction cache if this page is
1264 	 * mapped executable and cacheable.
1265 	 */
1266 	if ((pte_lo & (LPTE_I | LPTE_G | LPTE_NOEXEC)) == 0) {
1267 		moea64_syncicache(pmap, va, VM_PAGE_TO_PHYS(m), PAGE_SIZE);
1268 	}
1269 }
1270 
1271 static void
1272 moea64_syncicache(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_size_t sz)
1273 {
1274 	/*
1275 	 * This is much trickier than on older systems because
1276 	 * we can't sync the icache on physical addresses directly
1277 	 * without a direct map. Instead we check a couple of cases
1278 	 * where the memory is already mapped in and, failing that,
1279 	 * use the same trick we use for page zeroing to create
1280 	 * a temporary mapping for this physical address.
1281 	 */
1282 
1283 	if (!pmap_bootstrapped) {
1284 		/*
1285 		 * If PMAP is not bootstrapped, we are likely to be
1286 		 * in real mode.
1287 		 */
1288 		__syncicache((void *)pa, sz);
1289 	} else if (pmap == kernel_pmap) {
1290 		__syncicache((void *)va, sz);
1291 	} else {
1292 		/* Use the scratch page to set up a temp mapping */
1293 
1294 		mtx_lock(&moea64_scratchpage_mtx);
1295 
1296 		moea64_set_scratchpage_pa(1,pa);
1297 		__syncicache((void *)moea64_scratchpage_va[1], sz);
1298 
1299 		mtx_unlock(&moea64_scratchpage_mtx);
1300 	}
1301 }
1302 
1303 /*
1304  * Maps a sequence of resident pages belonging to the same object.
1305  * The sequence begins with the given page m_start.  This page is
1306  * mapped at the given virtual address start.  Each subsequent page is
1307  * mapped at a virtual address that is offset from start by the same
1308  * amount as the page is offset from m_start within the object.  The
1309  * last page in the sequence is the page with the largest offset from
1310  * m_start that can be mapped at a virtual address less than the given
1311  * virtual address end.  Not every virtual page between start and end
1312  * is mapped; only those for which a resident page exists with the
1313  * corresponding offset from m_start are mapped.
1314  */
1315 void
1316 moea64_enter_object(mmu_t mmu, pmap_t pm, vm_offset_t start, vm_offset_t end,
1317     vm_page_t m_start, vm_prot_t prot)
1318 {
1319 	vm_page_t m;
1320 	vm_pindex_t diff, psize;
1321 
1322 	psize = atop(end - start);
1323 	m = m_start;
1324 	PMAP_LOCK(pm);
1325 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1326 		moea64_enter_locked(pm, start + ptoa(diff), m, prot &
1327 		    (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1328 		m = TAILQ_NEXT(m, listq);
1329 	}
1330 	PMAP_UNLOCK(pm);
1331 }
1332 
1333 void
1334 moea64_enter_quick(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_page_t m,
1335     vm_prot_t prot)
1336 {
1337 	PMAP_LOCK(pm);
1338 	moea64_enter_locked(pm, va, m, prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1339 	    FALSE);
1340 	PMAP_UNLOCK(pm);
1341 
1342 }
1343 
1344 vm_paddr_t
1345 moea64_extract(mmu_t mmu, pmap_t pm, vm_offset_t va)
1346 {
1347 	struct	pvo_entry *pvo;
1348 	vm_paddr_t pa;
1349 
1350 	PMAP_LOCK(pm);
1351 	pvo = moea64_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1352 	if (pvo == NULL)
1353 		pa = 0;
1354 	else
1355 		pa = (pvo->pvo_pte.lpte.pte_lo & LPTE_RPGN) | (va & ADDR_POFF);
1356 	PMAP_UNLOCK(pm);
1357 	return (pa);
1358 }
1359 
1360 /*
1361  * Atomically extract and hold the physical page with the given
1362  * pmap and virtual address pair if that mapping permits the given
1363  * protection.
1364  */
1365 vm_page_t
1366 moea64_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_prot_t prot)
1367 {
1368 	struct	pvo_entry *pvo;
1369 	vm_page_t m;
1370 
1371 	m = NULL;
1372 	vm_page_lock_queues();
1373 	PMAP_LOCK(pmap);
1374 	pvo = moea64_pvo_find_va(pmap, va & ~ADDR_POFF, NULL);
1375 	if (pvo != NULL && (pvo->pvo_pte.lpte.pte_hi & LPTE_VALID) &&
1376 	    ((pvo->pvo_pte.lpte.pte_lo & LPTE_PP) == LPTE_RW ||
1377 	     (prot & VM_PROT_WRITE) == 0)) {
1378 		m = PHYS_TO_VM_PAGE(pvo->pvo_pte.lpte.pte_lo & LPTE_RPGN);
1379 		vm_page_hold(m);
1380 	}
1381 	vm_page_unlock_queues();
1382 	PMAP_UNLOCK(pmap);
1383 	return (m);
1384 }
1385 
1386 static void *
1387 moea64_uma_page_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
1388 {
1389 	/*
1390 	 * This entire routine is a horrible hack to avoid bothering kmem
1391 	 * for new KVA addresses. Because this can get called from inside
1392 	 * kmem allocation routines, calling kmem for a new address here
1393 	 * can lead to multiply locking non-recursive mutexes.
1394 	 */
1395 	static vm_pindex_t color;
1396         vm_offset_t va;
1397 
1398         vm_page_t m;
1399         int pflags, needed_lock;
1400 
1401 	*flags = UMA_SLAB_PRIV;
1402 	needed_lock = !PMAP_LOCKED(kernel_pmap);
1403 
1404 	if (needed_lock)
1405 		PMAP_LOCK(kernel_pmap);
1406 
1407         if ((wait & (M_NOWAIT|M_USE_RESERVE)) == M_NOWAIT)
1408                 pflags = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
1409         else
1410                 pflags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED;
1411         if (wait & M_ZERO)
1412                 pflags |= VM_ALLOC_ZERO;
1413 
1414         for (;;) {
1415                 m = vm_page_alloc(NULL, color++, pflags | VM_ALLOC_NOOBJ);
1416                 if (m == NULL) {
1417                         if (wait & M_NOWAIT)
1418                                 return (NULL);
1419                         VM_WAIT;
1420                 } else
1421                         break;
1422         }
1423 
1424 	va = pvo_allocator_start;
1425 	pvo_allocator_start += PAGE_SIZE;
1426 
1427 	if (pvo_allocator_start >= pvo_allocator_end)
1428 		panic("Ran out of PVO allocator buffer space!");
1429 
1430 	/* Now call pvo_enter in recursive mode */
1431 	moea64_pvo_enter(kernel_pmap, moea64_upvo_zone,
1432 	    &moea64_pvo_kunmanaged, va,  VM_PAGE_TO_PHYS(m), LPTE_M,
1433 	    PVO_WIRED | PVO_BOOTSTRAP, 1);
1434 
1435 	TLBIE(kernel_pmap, va);
1436 
1437 	if (needed_lock)
1438 		PMAP_UNLOCK(kernel_pmap);
1439 
1440 	if ((wait & M_ZERO) && (m->flags & PG_ZERO) == 0)
1441                 bzero((void *)va, PAGE_SIZE);
1442 
1443 	return (void *)va;
1444 }
1445 
1446 void
1447 moea64_init(mmu_t mmu)
1448 {
1449 
1450 	CTR0(KTR_PMAP, "moea64_init");
1451 
1452 	moea64_upvo_zone = uma_zcreate("UPVO entry", sizeof (struct pvo_entry),
1453 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1454 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1455 	moea64_mpvo_zone = uma_zcreate("MPVO entry", sizeof(struct pvo_entry),
1456 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1457 	    UMA_ZONE_VM | UMA_ZONE_NOFREE);
1458 
1459 	if (!hw_direct_map) {
1460 		uma_zone_set_allocf(moea64_upvo_zone,moea64_uma_page_alloc);
1461 		uma_zone_set_allocf(moea64_mpvo_zone,moea64_uma_page_alloc);
1462 	}
1463 
1464 	moea64_initialized = TRUE;
1465 }
1466 
1467 boolean_t
1468 moea64_is_modified(mmu_t mmu, vm_page_t m)
1469 {
1470 
1471 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1472 		return (FALSE);
1473 
1474 	return (moea64_query_bit(m, LPTE_CHG));
1475 }
1476 
1477 void
1478 moea64_clear_reference(mmu_t mmu, vm_page_t m)
1479 {
1480 
1481 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1482 		return;
1483 	moea64_clear_bit(m, LPTE_REF, NULL);
1484 }
1485 
1486 void
1487 moea64_clear_modify(mmu_t mmu, vm_page_t m)
1488 {
1489 
1490 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1491 		return;
1492 	moea64_clear_bit(m, LPTE_CHG, NULL);
1493 }
1494 
1495 /*
1496  * Clear the write and modified bits in each of the given page's mappings.
1497  */
1498 void
1499 moea64_remove_write(mmu_t mmu, vm_page_t m)
1500 {
1501 	struct	pvo_entry *pvo;
1502 	struct	lpte *pt;
1503 	pmap_t	pmap;
1504 	uint64_t lo;
1505 
1506 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1507 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1508 	    (m->flags & PG_WRITEABLE) == 0)
1509 		return;
1510 	lo = moea64_attr_fetch(m);
1511 	SYNC();
1512 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1513 		pmap = pvo->pvo_pmap;
1514 		PMAP_LOCK(pmap);
1515 		if ((pvo->pvo_pte.lpte.pte_lo & LPTE_PP) != LPTE_BR) {
1516 			LOCK_TABLE();
1517 			pt = moea64_pvo_to_pte(pvo, -1);
1518 			pvo->pvo_pte.lpte.pte_lo &= ~LPTE_PP;
1519 			pvo->pvo_pte.lpte.pte_lo |= LPTE_BR;
1520 			if (pt != NULL) {
1521 				moea64_pte_synch(pt, &pvo->pvo_pte.lpte);
1522 				lo |= pvo->pvo_pte.lpte.pte_lo;
1523 				pvo->pvo_pte.lpte.pte_lo &= ~LPTE_CHG;
1524 				moea64_pte_change(pt, &pvo->pvo_pte.lpte,
1525 				    pvo->pvo_pmap, pvo->pvo_vaddr);
1526 			}
1527 			UNLOCK_TABLE();
1528 		}
1529 		PMAP_UNLOCK(pmap);
1530 	}
1531 	if ((lo & LPTE_CHG) != 0) {
1532 		moea64_attr_clear(m, LPTE_CHG);
1533 		vm_page_dirty(m);
1534 	}
1535 	vm_page_flag_clear(m, PG_WRITEABLE);
1536 }
1537 
1538 /*
1539  *	moea64_ts_referenced:
1540  *
1541  *	Return a count of reference bits for a page, clearing those bits.
1542  *	It is not necessary for every reference bit to be cleared, but it
1543  *	is necessary that 0 only be returned when there are truly no
1544  *	reference bits set.
1545  *
1546  *	XXX: The exact number of bits to check and clear is a matter that
1547  *	should be tested and standardized at some point in the future for
1548  *	optimal aging of shared pages.
1549  */
1550 boolean_t
1551 moea64_ts_referenced(mmu_t mmu, vm_page_t m)
1552 {
1553 	int count;
1554 
1555 	if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0)
1556 		return (0);
1557 
1558 	count = moea64_clear_bit(m, LPTE_REF, NULL);
1559 
1560 	return (count);
1561 }
1562 
1563 /*
1564  * Map a wired page into kernel virtual address space.
1565  */
1566 void
1567 moea64_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1568 {
1569 	uint64_t	pte_lo;
1570 	int		error;
1571 
1572 	if (!pmap_bootstrapped) {
1573 		if (va >= VM_MIN_KERNEL_ADDRESS && va < VM_MAX_KERNEL_ADDRESS)
1574 			panic("Trying to enter an address in KVA -- %#x!\n",pa);
1575 	}
1576 
1577 	pte_lo = moea64_calc_wimg(pa);
1578 
1579 	PMAP_LOCK(kernel_pmap);
1580 	error = moea64_pvo_enter(kernel_pmap, moea64_upvo_zone,
1581 	    &moea64_pvo_kunmanaged, va, pa, pte_lo,
1582 	    PVO_WIRED | VM_PROT_EXECUTE, 0);
1583 
1584 	TLBIE(kernel_pmap, va);
1585 
1586 	if (error != 0 && error != ENOENT)
1587 		panic("moea64_kenter: failed to enter va %#x pa %#x: %d", va,
1588 		    pa, error);
1589 
1590 	/*
1591 	 * Flush the memory from the instruction cache.
1592 	 */
1593 	if ((pte_lo & (LPTE_I | LPTE_G)) == 0) {
1594 		__syncicache((void *)va, PAGE_SIZE);
1595 	}
1596 	PMAP_UNLOCK(kernel_pmap);
1597 }
1598 
1599 /*
1600  * Extract the physical page address associated with the given kernel virtual
1601  * address.
1602  */
1603 vm_offset_t
1604 moea64_kextract(mmu_t mmu, vm_offset_t va)
1605 {
1606 	struct		pvo_entry *pvo;
1607 	vm_paddr_t pa;
1608 
1609 	PMAP_LOCK(kernel_pmap);
1610 	pvo = moea64_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL);
1611 	KASSERT(pvo != NULL, ("moea64_kextract: no addr found"));
1612 	pa = (pvo->pvo_pte.lpte.pte_lo & LPTE_RPGN) | (va & ADDR_POFF);
1613 	PMAP_UNLOCK(kernel_pmap);
1614 	return (pa);
1615 }
1616 
1617 /*
1618  * Remove a wired page from kernel virtual address space.
1619  */
1620 void
1621 moea64_kremove(mmu_t mmu, vm_offset_t va)
1622 {
1623 	moea64_remove(mmu, kernel_pmap, va, va + PAGE_SIZE);
1624 }
1625 
1626 /*
1627  * Map a range of physical addresses into kernel virtual address space.
1628  *
1629  * The value passed in *virt is a suggested virtual address for the mapping.
1630  * Architectures which can support a direct-mapped physical to virtual region
1631  * can return the appropriate address within that region, leaving '*virt'
1632  * unchanged.  We cannot and therefore do not; *virt is updated with the
1633  * first usable address after the mapped region.
1634  */
1635 vm_offset_t
1636 moea64_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1637     vm_offset_t pa_end, int prot)
1638 {
1639 	vm_offset_t	sva, va;
1640 
1641 	sva = *virt;
1642 	va = sva;
1643 	for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE)
1644 		moea64_kenter(mmu, va, pa_start);
1645 	*virt = va;
1646 
1647 	return (sva);
1648 }
1649 
1650 /*
1651  * Returns true if the pmap's pv is one of the first
1652  * 16 pvs linked to from this page.  This count may
1653  * be changed upwards or downwards in the future; it
1654  * is only necessary that true be returned for a small
1655  * subset of pmaps for proper page aging.
1656  */
1657 boolean_t
1658 moea64_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
1659 {
1660         int loops;
1661 	struct pvo_entry *pvo;
1662 
1663         if (!moea64_initialized || (m->flags & PG_FICTITIOUS))
1664                 return FALSE;
1665 
1666 	loops = 0;
1667 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
1668 		if (pvo->pvo_pmap == pmap)
1669 			return (TRUE);
1670 		if (++loops >= 16)
1671 			break;
1672 	}
1673 
1674 	return (FALSE);
1675 }
1676 
1677 /*
1678  * Return the number of managed mappings to the given physical page
1679  * that are wired.
1680  */
1681 int
1682 moea64_page_wired_mappings(mmu_t mmu, vm_page_t m)
1683 {
1684 	struct pvo_entry *pvo;
1685 	int count;
1686 
1687 	count = 0;
1688 	if (!moea64_initialized || (m->flags & PG_FICTITIOUS) != 0)
1689 		return (count);
1690 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1691 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink)
1692 		if ((pvo->pvo_vaddr & PVO_WIRED) != 0)
1693 			count++;
1694 	return (count);
1695 }
1696 
1697 static u_int	moea64_vsidcontext;
1698 
1699 void
1700 moea64_pinit(mmu_t mmu, pmap_t pmap)
1701 {
1702 	int	i, mask;
1703 	u_int	entropy;
1704 
1705 	PMAP_LOCK_INIT(pmap);
1706 
1707 	entropy = 0;
1708 	__asm __volatile("mftb %0" : "=r"(entropy));
1709 
1710 	if (pmap_bootstrapped)
1711 		pmap->pmap_phys = (pmap_t)moea64_kextract(mmu, (vm_offset_t)pmap);
1712 	else
1713 		pmap->pmap_phys = pmap;
1714 
1715 	/*
1716 	 * Allocate some segment registers for this pmap.
1717 	 */
1718 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1719 		u_int	hash, n;
1720 
1721 		/*
1722 		 * Create a new value by mutiplying by a prime and adding in
1723 		 * entropy from the timebase register.  This is to make the
1724 		 * VSID more random so that the PT hash function collides
1725 		 * less often.  (Note that the prime casues gcc to do shifts
1726 		 * instead of a multiply.)
1727 		 */
1728 		moea64_vsidcontext = (moea64_vsidcontext * 0x1105) + entropy;
1729 		hash = moea64_vsidcontext & (NPMAPS - 1);
1730 		if (hash == 0)		/* 0 is special, avoid it */
1731 			continue;
1732 		n = hash >> 5;
1733 		mask = 1 << (hash & (VSID_NBPW - 1));
1734 		hash = (moea64_vsidcontext & 0xfffff);
1735 		if (moea64_vsid_bitmap[n] & mask) {	/* collision? */
1736 			/* anything free in this bucket? */
1737 			if (moea64_vsid_bitmap[n] == 0xffffffff) {
1738 				entropy = (moea64_vsidcontext >> 20);
1739 				continue;
1740 			}
1741 			i = ffs(~moea64_vsid_bitmap[i]) - 1;
1742 			mask = 1 << i;
1743 			hash &= 0xfffff & ~(VSID_NBPW - 1);
1744 			hash |= i;
1745 		}
1746 		moea64_vsid_bitmap[n] |= mask;
1747 		for (i = 0; i < 16; i++) {
1748 			pmap->pm_sr[i] = VSID_MAKE(i, hash);
1749 		}
1750 		return;
1751 	}
1752 
1753 	panic("moea64_pinit: out of segments");
1754 }
1755 
1756 /*
1757  * Initialize the pmap associated with process 0.
1758  */
1759 void
1760 moea64_pinit0(mmu_t mmu, pmap_t pm)
1761 {
1762 	moea64_pinit(mmu, pm);
1763 	bzero(&pm->pm_stats, sizeof(pm->pm_stats));
1764 }
1765 
1766 /*
1767  * Set the physical protection on the specified range of this map as requested.
1768  */
1769 void
1770 moea64_protect(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva,
1771     vm_prot_t prot)
1772 {
1773 	struct	pvo_entry *pvo;
1774 	struct	lpte *pt;
1775 	int	pteidx;
1776 
1777 	CTR4(KTR_PMAP, "moea64_protect: pm=%p sva=%#x eva=%#x prot=%#x", pm, sva,
1778 	    eva, prot);
1779 
1780 
1781 	KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap,
1782 	    ("moea64_protect: non current pmap"));
1783 
1784 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1785 		moea64_remove(mmu, pm, sva, eva);
1786 		return;
1787 	}
1788 
1789 	vm_page_lock_queues();
1790 	PMAP_LOCK(pm);
1791 	for (; sva < eva; sva += PAGE_SIZE) {
1792 		pvo = moea64_pvo_find_va(pm, sva, &pteidx);
1793 		if (pvo == NULL)
1794 			continue;
1795 
1796 		/*
1797 		 * Grab the PTE pointer before we diddle with the cached PTE
1798 		 * copy.
1799 		 */
1800 		LOCK_TABLE();
1801 		pt = moea64_pvo_to_pte(pvo, pteidx);
1802 
1803 		/*
1804 		 * Change the protection of the page.
1805 		 */
1806 		pvo->pvo_pte.lpte.pte_lo &= ~LPTE_PP;
1807 		pvo->pvo_pte.lpte.pte_lo |= LPTE_BR;
1808 		pvo->pvo_pte.lpte.pte_lo &= ~LPTE_NOEXEC;
1809 		if ((prot & VM_PROT_EXECUTE) == 0)
1810 			pvo->pvo_pte.lpte.pte_lo |= LPTE_NOEXEC;
1811 
1812 		/*
1813 		 * If the PVO is in the page table, update that pte as well.
1814 		 */
1815 		if (pt != NULL) {
1816 			moea64_pte_change(pt, &pvo->pvo_pte.lpte,
1817 			    pvo->pvo_pmap, pvo->pvo_vaddr);
1818 			if ((pvo->pvo_pte.lpte.pte_lo &
1819 			    (LPTE_I | LPTE_G | LPTE_NOEXEC)) == 0) {
1820 				moea64_syncicache(pm, sva,
1821 				    pvo->pvo_pte.lpte.pte_lo & LPTE_RPGN,
1822 				    PAGE_SIZE);
1823 			}
1824 		}
1825 		UNLOCK_TABLE();
1826 	}
1827 	vm_page_unlock_queues();
1828 	PMAP_UNLOCK(pm);
1829 }
1830 
1831 /*
1832  * Map a list of wired pages into kernel virtual address space.  This is
1833  * intended for temporary mappings which do not need page modification or
1834  * references recorded.  Existing mappings in the region are overwritten.
1835  */
1836 void
1837 moea64_qenter(mmu_t mmu, vm_offset_t va, vm_page_t *m, int count)
1838 {
1839 	while (count-- > 0) {
1840 		moea64_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1841 		va += PAGE_SIZE;
1842 		m++;
1843 	}
1844 }
1845 
1846 /*
1847  * Remove page mappings from kernel virtual address space.  Intended for
1848  * temporary mappings entered by moea64_qenter.
1849  */
1850 void
1851 moea64_qremove(mmu_t mmu, vm_offset_t va, int count)
1852 {
1853 	while (count-- > 0) {
1854 		moea64_kremove(mmu, va);
1855 		va += PAGE_SIZE;
1856 	}
1857 }
1858 
1859 void
1860 moea64_release(mmu_t mmu, pmap_t pmap)
1861 {
1862         int idx, mask;
1863 
1864 	/*
1865 	 * Free segment register's VSID
1866 	 */
1867         if (pmap->pm_sr[0] == 0)
1868                 panic("moea64_release");
1869 
1870         idx = VSID_TO_HASH(pmap->pm_sr[0]) & (NPMAPS-1);
1871         mask = 1 << (idx % VSID_NBPW);
1872         idx /= VSID_NBPW;
1873         moea64_vsid_bitmap[idx] &= ~mask;
1874 	PMAP_LOCK_DESTROY(pmap);
1875 }
1876 
1877 /*
1878  * Remove the given range of addresses from the specified map.
1879  */
1880 void
1881 moea64_remove(mmu_t mmu, pmap_t pm, vm_offset_t sva, vm_offset_t eva)
1882 {
1883 	struct	pvo_entry *pvo;
1884 	int	pteidx;
1885 
1886 	vm_page_lock_queues();
1887 	PMAP_LOCK(pm);
1888 	for (; sva < eva; sva += PAGE_SIZE) {
1889 		pvo = moea64_pvo_find_va(pm, sva, &pteidx);
1890 		if (pvo != NULL) {
1891 			moea64_pvo_remove(pvo, pteidx);
1892 		}
1893 	}
1894 	vm_page_unlock_queues();
1895 	PMAP_UNLOCK(pm);
1896 }
1897 
1898 /*
1899  * Remove physical page from all pmaps in which it resides. moea64_pvo_remove()
1900  * will reflect changes in pte's back to the vm_page.
1901  */
1902 void
1903 moea64_remove_all(mmu_t mmu, vm_page_t m)
1904 {
1905 	struct  pvo_head *pvo_head;
1906 	struct	pvo_entry *pvo, *next_pvo;
1907 	pmap_t	pmap;
1908 
1909 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
1910 
1911 	pvo_head = vm_page_to_pvoh(m);
1912 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
1913 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
1914 
1915 		MOEA_PVO_CHECK(pvo);	/* sanity check */
1916 		pmap = pvo->pvo_pmap;
1917 		PMAP_LOCK(pmap);
1918 		moea64_pvo_remove(pvo, -1);
1919 		PMAP_UNLOCK(pmap);
1920 	}
1921 	vm_page_flag_clear(m, PG_WRITEABLE);
1922 }
1923 
1924 /*
1925  * Allocate a physical page of memory directly from the phys_avail map.
1926  * Can only be called from moea64_bootstrap before avail start and end are
1927  * calculated.
1928  */
1929 static vm_offset_t
1930 moea64_bootstrap_alloc(vm_size_t size, u_int align)
1931 {
1932 	vm_offset_t	s, e;
1933 	int		i, j;
1934 
1935 	size = round_page(size);
1936 	for (i = 0; phys_avail[i + 1] != 0; i += 2) {
1937 		if (align != 0)
1938 			s = (phys_avail[i] + align - 1) & ~(align - 1);
1939 		else
1940 			s = phys_avail[i];
1941 		e = s + size;
1942 
1943 		if (s < phys_avail[i] || e > phys_avail[i + 1])
1944 			continue;
1945 
1946 		if (s == phys_avail[i]) {
1947 			phys_avail[i] += size;
1948 		} else if (e == phys_avail[i + 1]) {
1949 			phys_avail[i + 1] -= size;
1950 		} else {
1951 			for (j = phys_avail_count * 2; j > i; j -= 2) {
1952 				phys_avail[j] = phys_avail[j - 2];
1953 				phys_avail[j + 1] = phys_avail[j - 1];
1954 			}
1955 
1956 			phys_avail[i + 3] = phys_avail[i + 1];
1957 			phys_avail[i + 1] = s;
1958 			phys_avail[i + 2] = e;
1959 			phys_avail_count++;
1960 		}
1961 
1962 		return (s);
1963 	}
1964 	panic("moea64_bootstrap_alloc: could not allocate memory");
1965 }
1966 
1967 static void
1968 tlbia(void)
1969 {
1970 	vm_offset_t i;
1971 
1972 	for (i = 0; i < 0xFF000; i += 0x00001000)
1973 		TLBIE(NULL,i);
1974 }
1975 
1976 static int
1977 moea64_pvo_enter(pmap_t pm, uma_zone_t zone, struct pvo_head *pvo_head,
1978     vm_offset_t va, vm_offset_t pa, uint64_t pte_lo, int flags, int recurse)
1979 {
1980 	struct	 pvo_entry *pvo;
1981 	uint64_t vsid;
1982 	int	 first;
1983 	u_int	 ptegidx;
1984 	int	 i;
1985 	int      bootstrap;
1986 
1987 	/*
1988 	 * One nasty thing that can happen here is that the UMA calls to
1989 	 * allocate new PVOs need to map more memory, which calls pvo_enter(),
1990 	 * which calls UMA...
1991 	 *
1992 	 * We break the loop by detecting recursion and allocating out of
1993 	 * the bootstrap pool.
1994 	 */
1995 
1996 	moea64_pvo_enter_calls++;
1997 	first = 0;
1998 	bootstrap = (flags & PVO_BOOTSTRAP);
1999 
2000 	if (!moea64_initialized)
2001 		bootstrap = 1;
2002 
2003 	/*
2004 	 * Compute the PTE Group index.
2005 	 */
2006 	va &= ~ADDR_POFF;
2007 	vsid = va_to_vsid(pm, va);
2008 	ptegidx = va_to_pteg(vsid, va);
2009 
2010 	/*
2011 	 * Remove any existing mapping for this page.  Reuse the pvo entry if
2012 	 * there is a mapping.
2013 	 */
2014 	if (!recurse)
2015 		LOCK_TABLE();
2016 
2017 	LIST_FOREACH(pvo, &moea64_pvo_table[ptegidx], pvo_olink) {
2018 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
2019 			if ((pvo->pvo_pte.lpte.pte_lo & LPTE_RPGN) == pa &&
2020 			    (pvo->pvo_pte.lpte.pte_lo & LPTE_PP) ==
2021 			    (pte_lo & LPTE_PP)) {
2022 				if (!recurse)
2023 					UNLOCK_TABLE();
2024 				return (0);
2025 			}
2026 			moea64_pvo_remove(pvo, -1);
2027 			break;
2028 		}
2029 	}
2030 
2031 	/*
2032 	 * If we aren't overwriting a mapping, try to allocate.
2033 	 */
2034 	if (bootstrap) {
2035 		if (moea64_bpvo_pool_index >= BPVO_POOL_SIZE) {
2036 			panic("moea64_enter: bpvo pool exhausted, %d, %d, %d",
2037 			      moea64_bpvo_pool_index, BPVO_POOL_SIZE,
2038 			      BPVO_POOL_SIZE * sizeof(struct pvo_entry));
2039 		}
2040 		pvo = &moea64_bpvo_pool[moea64_bpvo_pool_index];
2041 		moea64_bpvo_pool_index++;
2042 		bootstrap = 1;
2043 	} else {
2044 		pvo = uma_zalloc(zone, M_NOWAIT);
2045 	}
2046 
2047 	if (pvo == NULL) {
2048 		if (!recurse)
2049 			UNLOCK_TABLE();
2050 		return (ENOMEM);
2051 	}
2052 
2053 	moea64_pvo_entries++;
2054 	pvo->pvo_vaddr = va;
2055 	pvo->pvo_pmap = pm;
2056 	LIST_INSERT_HEAD(&moea64_pvo_table[ptegidx], pvo, pvo_olink);
2057 	pvo->pvo_vaddr &= ~ADDR_POFF;
2058 
2059 	if (!(flags & VM_PROT_EXECUTE))
2060 		pte_lo |= LPTE_NOEXEC;
2061 	if (flags & PVO_WIRED)
2062 		pvo->pvo_vaddr |= PVO_WIRED;
2063 	if (pvo_head != &moea64_pvo_kunmanaged)
2064 		pvo->pvo_vaddr |= PVO_MANAGED;
2065 	if (bootstrap)
2066 		pvo->pvo_vaddr |= PVO_BOOTSTRAP;
2067 	if (flags & PVO_FAKE)
2068 		pvo->pvo_vaddr |= PVO_FAKE;
2069 
2070 	moea64_pte_create(&pvo->pvo_pte.lpte, vsid, va,
2071 	    (uint64_t)(pa) | pte_lo);
2072 
2073 	/*
2074 	 * Remember if the list was empty and therefore will be the first
2075 	 * item.
2076 	 */
2077 	if (LIST_FIRST(pvo_head) == NULL)
2078 		first = 1;
2079 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
2080 
2081 	if (pvo->pvo_pte.lpte.pte_lo & PVO_WIRED)
2082 		pm->pm_stats.wired_count++;
2083 	pm->pm_stats.resident_count++;
2084 
2085 	/*
2086 	 * We hope this succeeds but it isn't required.
2087 	 */
2088 	i = moea64_pte_insert(ptegidx, &pvo->pvo_pte.lpte);
2089 	if (i >= 0) {
2090 		PVO_PTEGIDX_SET(pvo, i);
2091 	} else {
2092 		panic("moea64_pvo_enter: overflow");
2093 		moea64_pte_overflow++;
2094 	}
2095 
2096 	if (!recurse)
2097 		UNLOCK_TABLE();
2098 
2099 	return (first ? ENOENT : 0);
2100 }
2101 
2102 static void
2103 moea64_pvo_remove(struct pvo_entry *pvo, int pteidx)
2104 {
2105 	struct	lpte *pt;
2106 
2107 	/*
2108 	 * If there is an active pte entry, we need to deactivate it (and
2109 	 * save the ref & cfg bits).
2110 	 */
2111 	LOCK_TABLE();
2112 	pt = moea64_pvo_to_pte(pvo, pteidx);
2113 	if (pt != NULL) {
2114 		moea64_pte_unset(pt, &pvo->pvo_pte.lpte, pvo->pvo_pmap,
2115 		    pvo->pvo_vaddr);
2116 		PVO_PTEGIDX_CLR(pvo);
2117 	} else {
2118 		moea64_pte_overflow--;
2119 	}
2120 	UNLOCK_TABLE();
2121 
2122 	/*
2123 	 * Update our statistics.
2124 	 */
2125 	pvo->pvo_pmap->pm_stats.resident_count--;
2126 	if (pvo->pvo_pte.lpte.pte_lo & PVO_WIRED)
2127 		pvo->pvo_pmap->pm_stats.wired_count--;
2128 
2129 	/*
2130 	 * Save the REF/CHG bits into their cache if the page is managed.
2131 	 */
2132 	if ((pvo->pvo_vaddr & (PVO_MANAGED|PVO_FAKE)) == PVO_MANAGED) {
2133 		struct	vm_page *pg;
2134 
2135 		pg = PHYS_TO_VM_PAGE(pvo->pvo_pte.lpte.pte_lo & LPTE_RPGN);
2136 		if (pg != NULL) {
2137 			moea64_attr_save(pg, pvo->pvo_pte.lpte.pte_lo &
2138 			    (LPTE_REF | LPTE_CHG));
2139 		}
2140 	}
2141 
2142 	/*
2143 	 * Remove this PVO from the PV list.
2144 	 */
2145 	LIST_REMOVE(pvo, pvo_vlink);
2146 
2147 	/*
2148 	 * Remove this from the overflow list and return it to the pool
2149 	 * if we aren't going to reuse it.
2150 	 */
2151 	LIST_REMOVE(pvo, pvo_olink);
2152 	if (!(pvo->pvo_vaddr & PVO_BOOTSTRAP))
2153 		uma_zfree(pvo->pvo_vaddr & PVO_MANAGED ? moea64_mpvo_zone :
2154 		    moea64_upvo_zone, pvo);
2155 	moea64_pvo_entries--;
2156 	moea64_pvo_remove_calls++;
2157 }
2158 
2159 static __inline int
2160 moea64_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
2161 {
2162 	int	pteidx;
2163 
2164 	/*
2165 	 * We can find the actual pte entry without searching by grabbing
2166 	 * the PTEG index from 3 unused bits in pte_lo[11:9] and by
2167 	 * noticing the HID bit.
2168 	 */
2169 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
2170 	if (pvo->pvo_pte.lpte.pte_hi & LPTE_HID)
2171 		pteidx ^= moea64_pteg_mask * 8;
2172 
2173 	return (pteidx);
2174 }
2175 
2176 static struct pvo_entry *
2177 moea64_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p)
2178 {
2179 	struct		pvo_entry *pvo;
2180 	int		ptegidx;
2181 	uint64_t	vsid;
2182 
2183 	va &= ~ADDR_POFF;
2184 	vsid = va_to_vsid(pm, va);
2185 	ptegidx = va_to_pteg(vsid, va);
2186 
2187 	LOCK_TABLE();
2188 	LIST_FOREACH(pvo, &moea64_pvo_table[ptegidx], pvo_olink) {
2189 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
2190 			if (pteidx_p)
2191 				*pteidx_p = moea64_pvo_pte_index(pvo, ptegidx);
2192 			break;
2193 		}
2194 	}
2195 	UNLOCK_TABLE();
2196 
2197 	return (pvo);
2198 }
2199 
2200 static struct lpte *
2201 moea64_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
2202 {
2203 	struct lpte *pt;
2204 
2205 	/*
2206 	 * If we haven't been supplied the ptegidx, calculate it.
2207 	 */
2208 	if (pteidx == -1) {
2209 		int		ptegidx;
2210 		uint64_t	vsid;
2211 
2212 		vsid = va_to_vsid(pvo->pvo_pmap, pvo->pvo_vaddr);
2213 		ptegidx = va_to_pteg(vsid, pvo->pvo_vaddr);
2214 		pteidx = moea64_pvo_pte_index(pvo, ptegidx);
2215 	}
2216 
2217 	pt = &moea64_pteg_table[pteidx >> 3].pt[pteidx & 7];
2218 
2219 	if ((pvo->pvo_pte.lpte.pte_hi & LPTE_VALID) &&
2220 	    !PVO_PTEGIDX_ISSET(pvo)) {
2221 		panic("moea64_pvo_to_pte: pvo %p has valid pte in pvo but no "
2222 		    "valid pte index", pvo);
2223 	}
2224 
2225 	if ((pvo->pvo_pte.lpte.pte_hi & LPTE_VALID) == 0 &&
2226 	    PVO_PTEGIDX_ISSET(pvo)) {
2227 		panic("moea64_pvo_to_pte: pvo %p has valid pte index in pvo "
2228 		    "pvo but no valid pte", pvo);
2229 	}
2230 
2231 	if ((pt->pte_hi ^ (pvo->pvo_pte.lpte.pte_hi & ~LPTE_VALID)) ==
2232 	    LPTE_VALID) {
2233 		if ((pvo->pvo_pte.lpte.pte_hi & LPTE_VALID) == 0) {
2234 			panic("moea64_pvo_to_pte: pvo %p has valid pte in "
2235 			    "moea64_pteg_table %p but invalid in pvo", pvo, pt);
2236 		}
2237 
2238 		if (((pt->pte_lo ^ pvo->pvo_pte.lpte.pte_lo) &
2239 		    ~(LPTE_CHG|LPTE_REF)) != 0) {
2240 			panic("moea64_pvo_to_pte: pvo %p pte does not match "
2241 			    "pte %p in moea64_pteg_table difference is %#x",
2242 			    pvo, pt,
2243 			    (uint32_t)(pt->pte_lo ^ pvo->pvo_pte.lpte.pte_lo));
2244 		}
2245 
2246 		ASSERT_TABLE_LOCK();
2247 		return (pt);
2248 	}
2249 
2250 	if (pvo->pvo_pte.lpte.pte_hi & LPTE_VALID) {
2251 		panic("moea64_pvo_to_pte: pvo %p has invalid pte %p in "
2252 		    "moea64_pteg_table but valid in pvo", pvo, pt);
2253 	}
2254 
2255 	return (NULL);
2256 }
2257 
2258 static int
2259 moea64_pte_insert(u_int ptegidx, struct lpte *pvo_pt)
2260 {
2261 	struct	lpte *pt;
2262 	int	i;
2263 
2264 	ASSERT_TABLE_LOCK();
2265 
2266 	/*
2267 	 * First try primary hash.
2268 	 */
2269 	for (pt = moea64_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2270 		if ((pt->pte_hi & LPTE_VALID) == 0) {
2271 			pvo_pt->pte_hi &= ~LPTE_HID;
2272 			moea64_pte_set(pt, pvo_pt);
2273 			return (i);
2274 		}
2275 	}
2276 
2277 	/*
2278 	 * Now try secondary hash.
2279 	 */
2280 	ptegidx ^= moea64_pteg_mask;
2281 
2282 	for (pt = moea64_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
2283 		if ((pt->pte_hi & LPTE_VALID) == 0) {
2284 			pvo_pt->pte_hi |= LPTE_HID;
2285 			moea64_pte_set(pt, pvo_pt);
2286 			return (i);
2287 		}
2288 	}
2289 
2290 	panic("moea64_pte_insert: overflow");
2291 	return (-1);
2292 }
2293 
2294 static boolean_t
2295 moea64_query_bit(vm_page_t m, u_int64_t ptebit)
2296 {
2297 	struct	pvo_entry *pvo;
2298 	struct	lpte *pt;
2299 
2300 #if 0
2301 	if (moea64_attr_fetch(m) & ptebit)
2302 		return (TRUE);
2303 #endif
2304 
2305 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2306 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2307 
2308 		/*
2309 		 * See if we saved the bit off.  If so, cache it and return
2310 		 * success.
2311 		 */
2312 		if (pvo->pvo_pte.lpte.pte_lo & ptebit) {
2313 			moea64_attr_save(m, ptebit);
2314 			MOEA_PVO_CHECK(pvo);	/* sanity check */
2315 			return (TRUE);
2316 		}
2317 	}
2318 
2319 	/*
2320 	 * No luck, now go through the hard part of looking at the PTEs
2321 	 * themselves.  Sync so that any pending REF/CHG bits are flushed to
2322 	 * the PTEs.
2323 	 */
2324 	SYNC();
2325 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2326 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2327 
2328 		/*
2329 		 * See if this pvo has a valid PTE.  if so, fetch the
2330 		 * REF/CHG bits from the valid PTE.  If the appropriate
2331 		 * ptebit is set, cache it and return success.
2332 		 */
2333 		LOCK_TABLE();
2334 		pt = moea64_pvo_to_pte(pvo, -1);
2335 		if (pt != NULL) {
2336 			moea64_pte_synch(pt, &pvo->pvo_pte.lpte);
2337 			if (pvo->pvo_pte.lpte.pte_lo & ptebit) {
2338 				UNLOCK_TABLE();
2339 
2340 				moea64_attr_save(m, ptebit);
2341 				MOEA_PVO_CHECK(pvo);	/* sanity check */
2342 				return (TRUE);
2343 			}
2344 		}
2345 		UNLOCK_TABLE();
2346 	}
2347 
2348 	return (FALSE);
2349 }
2350 
2351 static u_int
2352 moea64_clear_bit(vm_page_t m, u_int64_t ptebit, u_int64_t *origbit)
2353 {
2354 	u_int	count;
2355 	struct	pvo_entry *pvo;
2356 	struct	lpte *pt;
2357 	uint64_t rv;
2358 
2359 	/*
2360 	 * Clear the cached value.
2361 	 */
2362 	rv = moea64_attr_fetch(m);
2363 	moea64_attr_clear(m, ptebit);
2364 
2365 	/*
2366 	 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so
2367 	 * we can reset the right ones).  note that since the pvo entries and
2368 	 * list heads are accessed via BAT0 and are never placed in the page
2369 	 * table, we don't have to worry about further accesses setting the
2370 	 * REF/CHG bits.
2371 	 */
2372 	SYNC();
2373 
2374 	/*
2375 	 * For each pvo entry, clear the pvo's ptebit.  If this pvo has a
2376 	 * valid pte clear the ptebit from the valid pte.
2377 	 */
2378 	count = 0;
2379 	LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) {
2380 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2381 
2382 		LOCK_TABLE();
2383 		pt = moea64_pvo_to_pte(pvo, -1);
2384 		if (pt != NULL) {
2385 			moea64_pte_synch(pt, &pvo->pvo_pte.lpte);
2386 			if (pvo->pvo_pte.lpte.pte_lo & ptebit) {
2387 				count++;
2388 				moea64_pte_clear(pt, pvo->pvo_pmap, PVO_VADDR(pvo), ptebit);
2389 			}
2390 		}
2391 		UNLOCK_TABLE();
2392 		rv |= pvo->pvo_pte.lpte.pte_lo;
2393 		pvo->pvo_pte.lpte.pte_lo &= ~ptebit;
2394 		MOEA_PVO_CHECK(pvo);	/* sanity check */
2395 	}
2396 
2397 	if (origbit != NULL) {
2398 		*origbit = rv;
2399 	}
2400 
2401 	return (count);
2402 }
2403 
2404 boolean_t
2405 moea64_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2406 {
2407 	return (EFAULT);
2408 }
2409 
2410 /*
2411  * Map a set of physical memory pages into the kernel virtual
2412  * address space. Return a pointer to where it is mapped. This
2413  * routine is intended to be used for mapping device memory,
2414  * NOT real memory.
2415  */
2416 void *
2417 moea64_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2418 {
2419 	vm_offset_t va, tmpva, ppa, offset;
2420 
2421 	ppa = trunc_page(pa);
2422 	offset = pa & PAGE_MASK;
2423 	size = roundup(offset + size, PAGE_SIZE);
2424 
2425 	va = kmem_alloc_nofault(kernel_map, size);
2426 
2427 	if (!va)
2428 		panic("moea64_mapdev: Couldn't alloc kernel virtual memory");
2429 
2430 	for (tmpva = va; size > 0;) {
2431 		moea64_kenter(mmu, tmpva, ppa);
2432 		size -= PAGE_SIZE;
2433 		tmpva += PAGE_SIZE;
2434 		ppa += PAGE_SIZE;
2435 	}
2436 
2437 	return ((void *)(va + offset));
2438 }
2439 
2440 void
2441 moea64_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2442 {
2443 	vm_offset_t base, offset;
2444 
2445 	base = trunc_page(va);
2446 	offset = va & PAGE_MASK;
2447 	size = roundup(offset + size, PAGE_SIZE);
2448 
2449 	kmem_free(kernel_map, base, size);
2450 }
2451 
2452 static void
2453 moea64_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2454 {
2455 	struct pvo_entry *pvo;
2456 	vm_offset_t lim;
2457 	vm_paddr_t pa;
2458 	vm_size_t len;
2459 
2460 	PMAP_LOCK(pm);
2461 	while (sz > 0) {
2462 		lim = round_page(va);
2463 		len = MIN(lim - va, sz);
2464 		pvo = moea64_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2465 		if (pvo != NULL) {
2466 			pa = (pvo->pvo_pte.pte.pte_lo & PTE_RPGN) |
2467 			    (va & ADDR_POFF);
2468 			moea64_syncicache(pm, va, pa, len);
2469 		}
2470 		va += len;
2471 		sz -= len;
2472 	}
2473 	PMAP_UNLOCK(pm);
2474 }
2475