xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision d16b7486a53dcb8072b60ec6fcb4373a2d0c27b7)
1 /*	$NetBSD: pmap.c,v 1.114 2022/05/09 11:39:44 rin Exp $	*/
2 /*-
3  * Copyright (c) 2001 The NetBSD Foundation, Inc.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to The NetBSD Foundation
7  * by Matt Thomas <matt@3am-software.com> of Allegro Networks, Inc.
8  *
9  * Support for PPC64 Bridge mode added by Sanjay Lal <sanjayl@kymasys.com>
10  * of Kyma Systems LLC.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
23  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
24  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (C) 1995, 1996 Wolfgang Solfrank.
36  * Copyright (C) 1995, 1996 TooLs GmbH.
37  * All rights reserved.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. All advertising materials mentioning features or use of this software
48  *    must display the following acknowledgement:
49  *	This product includes software developed by TooLs GmbH.
50  * 4. The name of TooLs GmbH may not be used to endorse or promote products
51  *    derived from this software without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
54  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
55  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
56  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
57  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
58  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
59  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
60  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
61  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
62  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
63  */
64 
65 #include <sys/cdefs.h>
66 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.114 2022/05/09 11:39:44 rin Exp $");
67 
68 #define	PMAP_NOOPNAMES
69 
70 #ifdef _KERNEL_OPT
71 #include "opt_altivec.h"
72 #include "opt_multiprocessor.h"
73 #include "opt_pmap.h"
74 #include "opt_ppcarch.h"
75 #endif
76 
77 #include <sys/param.h>
78 #include <sys/proc.h>
79 #include <sys/pool.h>
80 #include <sys/queue.h>
81 #include <sys/device.h>		/* for evcnt */
82 #include <sys/systm.h>
83 #include <sys/atomic.h>
84 
85 #include <uvm/uvm.h>
86 #include <uvm/uvm_physseg.h>
87 
88 #include <machine/powerpc.h>
89 #include <powerpc/bat.h>
90 #include <powerpc/pcb.h>
91 #include <powerpc/psl.h>
92 #include <powerpc/spr.h>
93 #include <powerpc/oea/spr.h>
94 #include <powerpc/oea/sr_601.h>
95 
96 #ifdef ALTIVEC
97 extern int pmap_use_altivec;
98 #endif
99 
100 #ifdef PMAP_MEMLIMIT
101 static paddr_t pmap_memlimit = PMAP_MEMLIMIT;
102 #else
103 static paddr_t pmap_memlimit = -PAGE_SIZE;		/* there is no limit */
104 #endif
105 
106 extern struct pmap kernel_pmap_;
107 static unsigned int pmap_pages_stolen;
108 static u_long pmap_pte_valid;
109 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
110 static u_long pmap_pvo_enter_depth;
111 static u_long pmap_pvo_remove_depth;
112 #endif
113 
114 #ifndef MSGBUFADDR
115 extern paddr_t msgbuf_paddr;
116 #endif
117 
118 static struct mem_region *mem, *avail;
119 static u_int mem_cnt, avail_cnt;
120 
121 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
122 # define	PMAP_OEA 1
123 #endif
124 
125 #if defined(PMAP_OEA)
126 #define	_PRIxpte	"lx"
127 #else
128 #define	_PRIxpte	PRIx64
129 #endif
130 #define	_PRIxpa		"lx"
131 #define	_PRIxva		"lx"
132 #define	_PRIsr  	"lx"
133 
134 #ifdef PMAP_NEEDS_FIXUP
135 #if defined(PMAP_OEA)
136 #define	PMAPNAME(name)	pmap32_##name
137 #elif defined(PMAP_OEA64)
138 #define	PMAPNAME(name)	pmap64_##name
139 #elif defined(PMAP_OEA64_BRIDGE)
140 #define	PMAPNAME(name)	pmap64bridge_##name
141 #else
142 #error unknown variant for pmap
143 #endif
144 #endif /* PMAP_NEEDS_FIXUP */
145 
146 #ifdef PMAPNAME
147 #define	STATIC			static
148 #define pmap_pte_spill		PMAPNAME(pte_spill)
149 #define pmap_real_memory	PMAPNAME(real_memory)
150 #define pmap_init		PMAPNAME(init)
151 #define pmap_virtual_space	PMAPNAME(virtual_space)
152 #define pmap_create		PMAPNAME(create)
153 #define pmap_reference		PMAPNAME(reference)
154 #define pmap_destroy		PMAPNAME(destroy)
155 #define pmap_copy		PMAPNAME(copy)
156 #define pmap_update		PMAPNAME(update)
157 #define pmap_enter		PMAPNAME(enter)
158 #define pmap_remove		PMAPNAME(remove)
159 #define pmap_kenter_pa		PMAPNAME(kenter_pa)
160 #define pmap_kremove		PMAPNAME(kremove)
161 #define pmap_extract		PMAPNAME(extract)
162 #define pmap_protect		PMAPNAME(protect)
163 #define pmap_unwire		PMAPNAME(unwire)
164 #define pmap_page_protect	PMAPNAME(page_protect)
165 #define	pmap_pv_protect		PMAPNAME(pv_protect)
166 #define pmap_query_bit		PMAPNAME(query_bit)
167 #define pmap_clear_bit		PMAPNAME(clear_bit)
168 
169 #define pmap_activate		PMAPNAME(activate)
170 #define pmap_deactivate		PMAPNAME(deactivate)
171 
172 #define pmap_pinit		PMAPNAME(pinit)
173 #define pmap_procwr		PMAPNAME(procwr)
174 
175 #define pmap_pool		PMAPNAME(pool)
176 #define pmap_pvo_pool		PMAPNAME(pvo_pool)
177 #define pmap_pvo_table		PMAPNAME(pvo_table)
178 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
179 #define pmap_pte_print		PMAPNAME(pte_print)
180 #define pmap_pteg_check		PMAPNAME(pteg_check)
181 #define pmap_print_mmruregs	PMAPNAME(print_mmuregs)
182 #define pmap_print_pte		PMAPNAME(print_pte)
183 #define pmap_pteg_dist		PMAPNAME(pteg_dist)
184 #endif
185 #if defined(DEBUG) || defined(PMAPCHECK)
186 #define	pmap_pvo_verify		PMAPNAME(pvo_verify)
187 #define pmapcheck		PMAPNAME(check)
188 #endif
189 #if defined(DEBUG) || defined(PMAPDEBUG)
190 #define pmapdebug		PMAPNAME(debug)
191 #endif
192 #define pmap_steal_memory	PMAPNAME(steal_memory)
193 #define pmap_bootstrap		PMAPNAME(bootstrap)
194 #define pmap_bootstrap1		PMAPNAME(bootstrap1)
195 #define pmap_bootstrap2		PMAPNAME(bootstrap2)
196 #else
197 #define	STATIC			/* nothing */
198 #endif /* PMAPNAME */
199 
200 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
201 STATIC void pmap_real_memory(paddr_t *, psize_t *);
202 STATIC void pmap_init(void);
203 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
204 STATIC pmap_t pmap_create(void);
205 STATIC void pmap_reference(pmap_t);
206 STATIC void pmap_destroy(pmap_t);
207 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
208 STATIC void pmap_update(pmap_t);
209 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
210 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
211 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t, u_int);
212 STATIC void pmap_kremove(vaddr_t, vsize_t);
213 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
214 
215 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
216 STATIC void pmap_unwire(pmap_t, vaddr_t);
217 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
218 STATIC void pmap_pv_protect(paddr_t, vm_prot_t);
219 STATIC bool pmap_query_bit(struct vm_page *, int);
220 STATIC bool pmap_clear_bit(struct vm_page *, int);
221 
222 STATIC void pmap_activate(struct lwp *);
223 STATIC void pmap_deactivate(struct lwp *);
224 
225 STATIC void pmap_pinit(pmap_t pm);
226 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
227 
228 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
229 STATIC void pmap_pte_print(volatile struct pte *);
230 STATIC void pmap_pteg_check(void);
231 STATIC void pmap_print_mmuregs(void);
232 STATIC void pmap_print_pte(pmap_t, vaddr_t);
233 STATIC void pmap_pteg_dist(void);
234 #endif
235 #if defined(DEBUG) || defined(PMAPCHECK)
236 STATIC void pmap_pvo_verify(void);
237 #endif
238 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
239 STATIC void pmap_bootstrap(paddr_t, paddr_t);
240 STATIC void pmap_bootstrap1(paddr_t, paddr_t);
241 STATIC void pmap_bootstrap2(void);
242 
243 #ifdef PMAPNAME
244 const struct pmap_ops PMAPNAME(ops) = {
245 	.pmapop_pte_spill = pmap_pte_spill,
246 	.pmapop_real_memory = pmap_real_memory,
247 	.pmapop_init = pmap_init,
248 	.pmapop_virtual_space = pmap_virtual_space,
249 	.pmapop_create = pmap_create,
250 	.pmapop_reference = pmap_reference,
251 	.pmapop_destroy = pmap_destroy,
252 	.pmapop_copy = pmap_copy,
253 	.pmapop_update = pmap_update,
254 	.pmapop_enter = pmap_enter,
255 	.pmapop_remove = pmap_remove,
256 	.pmapop_kenter_pa = pmap_kenter_pa,
257 	.pmapop_kremove = pmap_kremove,
258 	.pmapop_extract = pmap_extract,
259 	.pmapop_protect = pmap_protect,
260 	.pmapop_unwire = pmap_unwire,
261 	.pmapop_page_protect = pmap_page_protect,
262 	.pmapop_pv_protect = pmap_pv_protect,
263 	.pmapop_query_bit = pmap_query_bit,
264 	.pmapop_clear_bit = pmap_clear_bit,
265 	.pmapop_activate = pmap_activate,
266 	.pmapop_deactivate = pmap_deactivate,
267 	.pmapop_pinit = pmap_pinit,
268 	.pmapop_procwr = pmap_procwr,
269 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
270 	.pmapop_pte_print = pmap_pte_print,
271 	.pmapop_pteg_check = pmap_pteg_check,
272 	.pmapop_print_mmuregs = pmap_print_mmuregs,
273 	.pmapop_print_pte = pmap_print_pte,
274 	.pmapop_pteg_dist = pmap_pteg_dist,
275 #else
276 	.pmapop_pte_print = NULL,
277 	.pmapop_pteg_check = NULL,
278 	.pmapop_print_mmuregs = NULL,
279 	.pmapop_print_pte = NULL,
280 	.pmapop_pteg_dist = NULL,
281 #endif
282 #if defined(DEBUG) || defined(PMAPCHECK)
283 	.pmapop_pvo_verify = pmap_pvo_verify,
284 #else
285 	.pmapop_pvo_verify = NULL,
286 #endif
287 	.pmapop_steal_memory = pmap_steal_memory,
288 	.pmapop_bootstrap = pmap_bootstrap,
289 	.pmapop_bootstrap1 = pmap_bootstrap1,
290 	.pmapop_bootstrap2 = pmap_bootstrap2,
291 };
292 #endif /* !PMAPNAME */
293 
294 /*
295  * The following structure is aligned to 32 bytes
296  */
297 struct pvo_entry {
298 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
299 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
300 	struct pte pvo_pte;			/* Prebuilt PTE */
301 	pmap_t pvo_pmap;			/* ptr to owning pmap */
302 	vaddr_t pvo_vaddr;			/* VA of entry */
303 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
304 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
305 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
306 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
307 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
308 #define	PVO_WIRED_P(pvo)	((pvo)->pvo_vaddr & PVO_WIRED)
309 #define	PVO_MANAGED_P(pvo)	((pvo)->pvo_vaddr & PVO_MANAGED)
310 #define	PVO_EXECUTABLE_P(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
311 #define	PVO_ENTER_INSERT	0		/* PVO has been removed */
312 #define	PVO_SPILL_UNSET		1		/* PVO has been evicted */
313 #define	PVO_SPILL_SET		2		/* PVO has been spilled */
314 #define	PVO_SPILL_INSERT	3		/* PVO has been inserted */
315 #define	PVO_PMAP_PAGE_PROTECT	4		/* PVO has changed */
316 #define	PVO_PMAP_PROTECT	5		/* PVO has changed */
317 #define	PVO_REMOVE		6		/* PVO has been removed */
318 #define	PVO_WHERE_MASK		15
319 #define	PVO_WHERE_SHFT		8
320 } __attribute__ ((aligned (32)));
321 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
322 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
323 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
324 #define	PVO_PTEGIDX_CLR(pvo)	\
325 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
326 #define	PVO_PTEGIDX_SET(pvo,i)	\
327 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
328 #define	PVO_WHERE(pvo,w)	\
329 	((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
330 	 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
331 
332 TAILQ_HEAD(pvo_tqhead, pvo_entry);
333 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
334 
335 struct pool pmap_pool;		/* pool for pmap structures */
336 struct pool pmap_pvo_pool;	/* pool for pvo entries */
337 
338 /*
339  * We keep a cache of unmanaged pages to be used for pvo entries for
340  * unmanaged pages.
341  */
342 struct pvo_page {
343 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
344 };
345 SIMPLEQ_HEAD(pvop_head, pvo_page);
346 static struct pvop_head pmap_pvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_pvop_head);
347 static u_long pmap_pvop_free;
348 static u_long pmap_pvop_maxfree;
349 
350 static void *pmap_pool_alloc(struct pool *, int);
351 static void pmap_pool_free(struct pool *, void *);
352 
353 static struct pool_allocator pmap_pool_allocator = {
354 	.pa_alloc = pmap_pool_alloc,
355 	.pa_free = pmap_pool_free,
356 	.pa_pagesz = 0,
357 };
358 
359 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
360 void pmap_pte_print(volatile struct pte *);
361 void pmap_pteg_check(void);
362 void pmap_pteg_dist(void);
363 void pmap_print_pte(pmap_t, vaddr_t);
364 void pmap_print_mmuregs(void);
365 #endif
366 
367 #if defined(DEBUG) || defined(PMAPCHECK)
368 #ifdef PMAPCHECK
369 int pmapcheck = 1;
370 #else
371 int pmapcheck = 0;
372 #endif
373 void pmap_pvo_verify(void);
374 static void pmap_pvo_check(const struct pvo_entry *);
375 #define	PMAP_PVO_CHECK(pvo)	 		\
376 	do {					\
377 		if (pmapcheck)			\
378 			pmap_pvo_check(pvo);	\
379 	} while (0)
380 #else
381 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
382 #endif
383 static int pmap_pte_insert(int, struct pte *);
384 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
385 	vaddr_t, paddr_t, register_t, int);
386 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
387 static void pmap_pvo_free(struct pvo_entry *);
388 static void pmap_pvo_free_list(struct pvo_head *);
389 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
390 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
391 static struct pvo_entry *pmap_pvo_reclaim(struct pmap *);
392 static void pvo_set_exec(struct pvo_entry *);
393 static void pvo_clear_exec(struct pvo_entry *);
394 
395 static void tlbia(void);
396 
397 static void pmap_release(pmap_t);
398 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
399 
400 static uint32_t pmap_pvo_reclaim_nextidx;
401 #ifdef DEBUG
402 static int pmap_pvo_reclaim_debugctr;
403 #endif
404 
405 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
406 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
407 
408 static int pmap_initialized;
409 
410 #if defined(DEBUG) || defined(PMAPDEBUG)
411 #define	PMAPDEBUG_BOOT		0x0001
412 #define	PMAPDEBUG_PTE		0x0002
413 #define	PMAPDEBUG_EXEC		0x0008
414 #define	PMAPDEBUG_PVOENTER	0x0010
415 #define	PMAPDEBUG_PVOREMOVE	0x0020
416 #define	PMAPDEBUG_ACTIVATE	0x0100
417 #define	PMAPDEBUG_CREATE	0x0200
418 #define	PMAPDEBUG_ENTER		0x1000
419 #define	PMAPDEBUG_KENTER	0x2000
420 #define	PMAPDEBUG_KREMOVE	0x4000
421 #define	PMAPDEBUG_REMOVE	0x8000
422 
423 unsigned int pmapdebug = 0;
424 
425 # define DPRINTF(x, ...)	printf(x, __VA_ARGS__)
426 # define DPRINTFN(n, x, ...)	do if (pmapdebug & PMAPDEBUG_ ## n) printf(x, __VA_ARGS__); while (0)
427 #else
428 # define DPRINTF(x, ...)	do { } while (0)
429 # define DPRINTFN(n, x, ...)	do { } while (0)
430 #endif
431 
432 
433 #ifdef PMAPCOUNTERS
434 /*
435  * From pmap_subr.c
436  */
437 extern struct evcnt pmap_evcnt_mappings;
438 extern struct evcnt pmap_evcnt_unmappings;
439 
440 extern struct evcnt pmap_evcnt_kernel_mappings;
441 extern struct evcnt pmap_evcnt_kernel_unmappings;
442 
443 extern struct evcnt pmap_evcnt_mappings_replaced;
444 
445 extern struct evcnt pmap_evcnt_exec_mappings;
446 extern struct evcnt pmap_evcnt_exec_cached;
447 
448 extern struct evcnt pmap_evcnt_exec_synced;
449 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
450 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
451 
452 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
453 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
454 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
455 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
456 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
457 
458 extern struct evcnt pmap_evcnt_updates;
459 extern struct evcnt pmap_evcnt_collects;
460 extern struct evcnt pmap_evcnt_copies;
461 
462 extern struct evcnt pmap_evcnt_ptes_spilled;
463 extern struct evcnt pmap_evcnt_ptes_unspilled;
464 extern struct evcnt pmap_evcnt_ptes_evicted;
465 
466 extern struct evcnt pmap_evcnt_ptes_primary[8];
467 extern struct evcnt pmap_evcnt_ptes_secondary[8];
468 extern struct evcnt pmap_evcnt_ptes_removed;
469 extern struct evcnt pmap_evcnt_ptes_changed;
470 extern struct evcnt pmap_evcnt_pvos_reclaimed;
471 extern struct evcnt pmap_evcnt_pvos_failed;
472 
473 extern struct evcnt pmap_evcnt_zeroed_pages;
474 extern struct evcnt pmap_evcnt_copied_pages;
475 extern struct evcnt pmap_evcnt_idlezeroed_pages;
476 
477 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
478 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
479 #else
480 #define	PMAPCOUNT(ev)	((void) 0)
481 #define	PMAPCOUNT2(ev)	((void) 0)
482 #endif
483 
484 #define	TLBIE(va)	__asm volatile("tlbie %0" :: "r"(va) : "memory")
485 
486 /* XXXSL: this needs to be moved to assembler */
487 #define	TLBIEL(va)	__asm volatile("tlbie %0" :: "r"(va) : "memory")
488 
489 #ifdef MD_TLBSYNC
490 #define TLBSYNC()	MD_TLBSYNC()
491 #else
492 #define	TLBSYNC()	__asm volatile("tlbsync" ::: "memory")
493 #endif
494 #define	SYNC()		__asm volatile("sync" ::: "memory")
495 #define	EIEIO()		__asm volatile("eieio" ::: "memory")
496 #define	DCBST(va)	__asm volatile("dcbst 0,%0" :: "r"(va) : "memory")
497 #define	MFMSR()		mfmsr()
498 #define	MTMSR(psl)	mtmsr(psl)
499 #define	MFPVR()		mfpvr()
500 #define	MFSRIN(va)	mfsrin(va)
501 #define	MFTB()		mfrtcltbl()
502 
503 #if defined(DDB) && !defined(PMAP_OEA64)
504 static inline register_t
505 mfsrin(vaddr_t va)
506 {
507 	register_t sr;
508 	__asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
509 	return sr;
510 }
511 #endif	/* DDB && !PMAP_OEA64 */
512 
513 #if defined (PMAP_OEA64_BRIDGE)
514 extern void mfmsr64 (register64_t *result);
515 #endif /* PMAP_OEA64_BRIDGE */
516 
517 #define	PMAP_LOCK()		KERNEL_LOCK(1, NULL)
518 #define	PMAP_UNLOCK()		KERNEL_UNLOCK_ONE(NULL)
519 
520 static inline register_t
521 pmap_interrupts_off(void)
522 {
523 	register_t msr = MFMSR();
524 	if (msr & PSL_EE)
525 		MTMSR(msr & ~PSL_EE);
526 	return msr;
527 }
528 
529 static void
530 pmap_interrupts_restore(register_t msr)
531 {
532 	if (msr & PSL_EE)
533 		MTMSR(msr);
534 }
535 
536 static inline u_int32_t
537 mfrtcltbl(void)
538 {
539 #ifdef PPC_OEA601
540 	if ((MFPVR() >> 16) == MPC601)
541 		return (mfrtcl() >> 7);
542 	else
543 #endif
544 		return (mftbl());
545 }
546 
547 /*
548  * These small routines may have to be replaced,
549  * if/when we support processors other that the 604.
550  */
551 
552 void
553 tlbia(void)
554 {
555 	char *i;
556 
557 	SYNC();
558 #if defined(PMAP_OEA)
559 	/*
560 	 * Why not use "tlbia"?  Because not all processors implement it.
561 	 *
562 	 * This needs to be a per-CPU callback to do the appropriate thing
563 	 * for the CPU. XXX
564 	 */
565 	for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
566 		TLBIE(i);
567 		EIEIO();
568 		SYNC();
569 	}
570 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
571 	/* This is specifically for the 970, 970UM v1.6 pp. 140. */
572 	for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
573 		TLBIEL(i);
574 		EIEIO();
575 		SYNC();
576 	}
577 #endif
578 	TLBSYNC();
579 	SYNC();
580 }
581 
582 static inline register_t
583 va_to_vsid(const struct pmap *pm, vaddr_t addr)
584 {
585 	/*
586 	 * Rather than searching the STE groups for the VSID or extracting
587 	 * it from the SR, we know how we generate that from the ESID and
588 	 * so do that.
589 	 *
590 	 * This makes the code the same for OEA and OEA64, and also allows
591 	 * us to generate a correct-for-that-address-space VSID even if the
592 	 * pmap contains a different SR value at any given moment (e.g.
593 	 * kernel pmap on a 601 that is using I/O segments).
594 	 */
595 	return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
596 }
597 
598 static inline register_t
599 va_to_pteg(const struct pmap *pm, vaddr_t addr)
600 {
601 	register_t hash;
602 
603 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
604 	return hash & pmap_pteg_mask;
605 }
606 
607 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
608 /*
609  * Given a PTE in the page table, calculate the VADDR that hashes to it.
610  * The only bit of magic is that the top 4 bits of the address doesn't
611  * technically exist in the PTE.  But we know we reserved 4 bits of the
612  * VSID for it so that's how we get it.
613  */
614 static vaddr_t
615 pmap_pte_to_va(volatile const struct pte *pt)
616 {
617 	vaddr_t va;
618 	uintptr_t ptaddr = (uintptr_t) pt;
619 
620 	if (pt->pte_hi & PTE_HID)
621 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
622 
623 	/* PPC Bits 10-19  PPC64 Bits 42-51 */
624 #if defined(PMAP_OEA)
625 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
626 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
627 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
628 #endif
629 	va <<= ADDR_PIDX_SHFT;
630 
631 	/* PPC Bits 4-9  PPC64 Bits 36-41 */
632 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
633 
634 #if defined(PMAP_OEA64)
635 	/* PPC63 Bits 0-35 */
636 	/* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
637 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
638 	/* PPC Bits 0-3 */
639 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
640 #endif
641 
642 	return va;
643 }
644 #endif
645 
646 static inline struct pvo_head *
647 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
648 {
649 	struct vm_page *pg;
650 	struct vm_page_md *md;
651 	struct pmap_page *pp;
652 
653 	pg = PHYS_TO_VM_PAGE(pa);
654 	if (pg_p != NULL)
655 		*pg_p = pg;
656 	if (pg == NULL) {
657 		if ((pp = pmap_pv_tracked(pa)) != NULL)
658 			return &pp->pp_pvoh;
659 		return NULL;
660 	}
661 	md = VM_PAGE_TO_MD(pg);
662 	return &md->mdpg_pvoh;
663 }
664 
665 static inline struct pvo_head *
666 vm_page_to_pvoh(struct vm_page *pg)
667 {
668 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
669 
670 	return &md->mdpg_pvoh;
671 }
672 
673 static inline void
674 pmap_pp_attr_clear(struct pmap_page *pp, int ptebit)
675 {
676 
677 	pp->pp_attrs &= ~ptebit;
678 }
679 
680 static inline void
681 pmap_attr_clear(struct vm_page *pg, int ptebit)
682 {
683 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
684 
685 	pmap_pp_attr_clear(&md->mdpg_pp, ptebit);
686 }
687 
688 static inline int
689 pmap_pp_attr_fetch(struct pmap_page *pp)
690 {
691 
692 	return pp->pp_attrs;
693 }
694 
695 static inline int
696 pmap_attr_fetch(struct vm_page *pg)
697 {
698 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
699 
700 	return pmap_pp_attr_fetch(&md->mdpg_pp);
701 }
702 
703 static inline void
704 pmap_attr_save(struct vm_page *pg, int ptebit)
705 {
706 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
707 
708 	md->mdpg_attrs |= ptebit;
709 }
710 
711 static inline int
712 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
713 {
714 	if (pt->pte_hi == pvo_pt->pte_hi
715 #if 0
716 	    && ((pt->pte_lo ^ pvo_pt->pte_lo) &
717 	        ~(PTE_REF|PTE_CHG)) == 0
718 #endif
719 	    )
720 		return 1;
721 	return 0;
722 }
723 
724 static inline void
725 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
726 {
727 	/*
728 	 * Construct the PTE.  Default to IMB initially.  Valid bit
729 	 * only gets set when the real pte is set in memory.
730 	 *
731 	 * Note: Don't set the valid bit for correct operation of tlb update.
732 	 */
733 #if defined(PMAP_OEA)
734 	pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
735 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
736 	pt->pte_lo = pte_lo;
737 #elif defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA64)
738 	pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT)
739 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
740 	pt->pte_lo = (u_int64_t) pte_lo;
741 #endif /* PMAP_OEA */
742 }
743 
744 static inline void
745 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
746 {
747 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
748 }
749 
750 static inline void
751 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
752 {
753 	/*
754 	 * As shown in Section 7.6.3.2.3
755 	 */
756 	pt->pte_lo &= ~ptebit;
757 	TLBIE(va);
758 	SYNC();
759 	EIEIO();
760 	TLBSYNC();
761 	SYNC();
762 #ifdef MULTIPROCESSOR
763 	DCBST(pt);
764 #endif
765 }
766 
767 static inline void
768 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
769 {
770 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
771 	if (pvo_pt->pte_hi & PTE_VALID)
772 		panic("pte_set: setting an already valid pte %p", pvo_pt);
773 #endif
774 	pvo_pt->pte_hi |= PTE_VALID;
775 
776 	/*
777 	 * Update the PTE as defined in section 7.6.3.1
778 	 * Note that the REF/CHG bits are from pvo_pt and thus should
779 	 * have been saved so this routine can restore them (if desired).
780 	 */
781 	pt->pte_lo = pvo_pt->pte_lo;
782 	EIEIO();
783 	pt->pte_hi = pvo_pt->pte_hi;
784 	TLBSYNC();
785 	SYNC();
786 #ifdef MULTIPROCESSOR
787 	DCBST(pt);
788 #endif
789 	pmap_pte_valid++;
790 }
791 
792 static inline void
793 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
794 {
795 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
796 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
797 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
798 	if ((pt->pte_hi & PTE_VALID) == 0)
799 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
800 #endif
801 
802 	pvo_pt->pte_hi &= ~PTE_VALID;
803 	/*
804 	 * Force the ref & chg bits back into the PTEs.
805 	 */
806 	SYNC();
807 	/*
808 	 * Invalidate the pte ... (Section 7.6.3.3)
809 	 */
810 	pt->pte_hi &= ~PTE_VALID;
811 	SYNC();
812 	TLBIE(va);
813 	SYNC();
814 	EIEIO();
815 	TLBSYNC();
816 	SYNC();
817 	/*
818 	 * Save the ref & chg bits ...
819 	 */
820 	pmap_pte_synch(pt, pvo_pt);
821 	pmap_pte_valid--;
822 }
823 
824 static inline void
825 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
826 {
827 	/*
828 	 * Invalidate the PTE
829 	 */
830 	pmap_pte_unset(pt, pvo_pt, va);
831 	pmap_pte_set(pt, pvo_pt);
832 }
833 
834 /*
835  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
836  * (either primary or secondary location).
837  *
838  * Note: both the destination and source PTEs must not have PTE_VALID set.
839  */
840 
841 static int
842 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
843 {
844 	volatile struct pte *pt;
845 	int i;
846 
847 #if defined(DEBUG)
848 	DPRINTFN(PTE, "pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
849 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo);
850 #endif
851 	/*
852 	 * First try primary hash.
853 	 */
854 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
855 		if ((pt->pte_hi & PTE_VALID) == 0) {
856 			pvo_pt->pte_hi &= ~PTE_HID;
857 			pmap_pte_set(pt, pvo_pt);
858 			return i;
859 		}
860 	}
861 
862 	/*
863 	 * Now try secondary hash.
864 	 */
865 	ptegidx ^= pmap_pteg_mask;
866 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
867 		if ((pt->pte_hi & PTE_VALID) == 0) {
868 			pvo_pt->pte_hi |= PTE_HID;
869 			pmap_pte_set(pt, pvo_pt);
870 			return i;
871 		}
872 	}
873 	return -1;
874 }
875 
876 /*
877  * Spill handler.
878  *
879  * Tries to spill a page table entry from the overflow area.
880  * This runs in either real mode (if dealing with a exception spill)
881  * or virtual mode when dealing with manually spilling one of the
882  * kernel's pte entries.  In either case, interrupts are already
883  * disabled.
884  */
885 
886 int
887 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec)
888 {
889 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
890 	struct pvo_entry *pvo;
891 	/* XXX: gcc -- vpvoh is always set at either *1* or *2* */
892 	struct pvo_tqhead *pvoh, *vpvoh = NULL;
893 	int ptegidx, i, j;
894 	volatile struct pteg *pteg;
895 	volatile struct pte *pt;
896 
897 	PMAP_LOCK();
898 
899 	ptegidx = va_to_pteg(pm, addr);
900 
901 	/*
902 	 * Have to substitute some entry. Use the primary hash for this.
903 	 * Use low bits of timebase as random generator.  Make sure we are
904 	 * not picking a kernel pte for replacement.
905 	 */
906 	pteg = &pmap_pteg_table[ptegidx];
907 	i = MFTB() & 7;
908 	for (j = 0; j < 8; j++) {
909 		pt = &pteg->pt[i];
910 		if ((pt->pte_hi & PTE_VALID) == 0)
911 			break;
912 		if (VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT)
913 				< PHYSMAP_VSIDBITS)
914 			break;
915 		i = (i + 1) & 7;
916 	}
917 	KASSERT(j < 8);
918 
919 	source_pvo = NULL;
920 	victim_pvo = NULL;
921 	pvoh = &pmap_pvo_table[ptegidx];
922 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
923 
924 		/*
925 		 * We need to find pvo entry for this address...
926 		 */
927 		PMAP_PVO_CHECK(pvo);		/* sanity check */
928 
929 		/*
930 		 * If we haven't found the source and we come to a PVO with
931 		 * a valid PTE, then we know we can't find it because all
932 		 * evicted PVOs always are first in the list.
933 		 */
934 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
935 			break;
936 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
937 		    addr == PVO_VADDR(pvo)) {
938 
939 			/*
940 			 * Now we have found the entry to be spilled into the
941 			 * pteg.  Attempt to insert it into the page table.
942 			 */
943 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
944 			if (j >= 0) {
945 				PVO_PTEGIDX_SET(pvo, j);
946 				PMAP_PVO_CHECK(pvo);	/* sanity check */
947 				PVO_WHERE(pvo, SPILL_INSERT);
948 				pvo->pvo_pmap->pm_evictions--;
949 				PMAPCOUNT(ptes_spilled);
950 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
951 				    ? pmap_evcnt_ptes_secondary
952 				    : pmap_evcnt_ptes_primary)[j]);
953 
954 				/*
955 				 * Since we keep the evicted entries at the
956 				 * from of the PVO list, we need move this
957 				 * (now resident) PVO after the evicted
958 				 * entries.
959 				 */
960 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
961 
962 				/*
963 				 * If we don't have to move (either we were the
964 				 * last entry or the next entry was valid),
965 				 * don't change our position.  Otherwise
966 				 * move ourselves to the tail of the queue.
967 				 */
968 				if (next_pvo != NULL &&
969 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
970 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
971 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
972 				}
973 				PMAP_UNLOCK();
974 				return 1;
975 			}
976 			source_pvo = pvo;
977 			if (exec && !PVO_EXECUTABLE_P(source_pvo)) {
978 				PMAP_UNLOCK();
979 				return 0;
980 			}
981 			if (victim_pvo != NULL)
982 				break;
983 		}
984 
985 		/*
986 		 * We also need the pvo entry of the victim we are replacing
987 		 * so save the R & C bits of the PTE.
988 		 */
989 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
990 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
991 			vpvoh = pvoh;			/* *1* */
992 			victim_pvo = pvo;
993 			if (source_pvo != NULL)
994 				break;
995 		}
996 	}
997 
998 	if (source_pvo == NULL) {
999 		PMAPCOUNT(ptes_unspilled);
1000 		PMAP_UNLOCK();
1001 		return 0;
1002 	}
1003 
1004 	if (victim_pvo == NULL) {
1005 		if ((pt->pte_hi & PTE_HID) == 0)
1006 			panic("pmap_pte_spill: victim p-pte (%p) has "
1007 			    "no pvo entry!", pt);
1008 
1009 		/*
1010 		 * If this is a secondary PTE, we need to search
1011 		 * its primary pvo bucket for the matching PVO.
1012 		 */
1013 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */
1014 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1015 			PMAP_PVO_CHECK(pvo);		/* sanity check */
1016 
1017 			/*
1018 			 * We also need the pvo entry of the victim we are
1019 			 * replacing so save the R & C bits of the PTE.
1020 			 */
1021 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1022 				victim_pvo = pvo;
1023 				break;
1024 			}
1025 		}
1026 		if (victim_pvo == NULL)
1027 			panic("pmap_pte_spill: victim s-pte (%p) has "
1028 			    "no pvo entry!", pt);
1029 	}
1030 
1031 	/*
1032 	 * The victim should be not be a kernel PVO/PTE entry.
1033 	 */
1034 	KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1035 	KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1036 	KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1037 
1038 	/*
1039 	 * We are invalidating the TLB entry for the EA for the
1040 	 * we are replacing even though its valid; If we don't
1041 	 * we lose any ref/chg bit changes contained in the TLB
1042 	 * entry.
1043 	 */
1044 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1045 
1046 	/*
1047 	 * To enforce the PVO list ordering constraint that all
1048 	 * evicted entries should come before all valid entries,
1049 	 * move the source PVO to the tail of its list and the
1050 	 * victim PVO to the head of its list (which might not be
1051 	 * the same list, if the victim was using the secondary hash).
1052 	 */
1053 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
1054 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
1055 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1056 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1057 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1058 	pmap_pte_set(pt, &source_pvo->pvo_pte);
1059 	victim_pvo->pvo_pmap->pm_evictions++;
1060 	source_pvo->pvo_pmap->pm_evictions--;
1061 	PVO_WHERE(victim_pvo, SPILL_UNSET);
1062 	PVO_WHERE(source_pvo, SPILL_SET);
1063 
1064 	PVO_PTEGIDX_CLR(victim_pvo);
1065 	PVO_PTEGIDX_SET(source_pvo, i);
1066 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1067 	PMAPCOUNT(ptes_spilled);
1068 	PMAPCOUNT(ptes_evicted);
1069 	PMAPCOUNT(ptes_removed);
1070 
1071 	PMAP_PVO_CHECK(victim_pvo);
1072 	PMAP_PVO_CHECK(source_pvo);
1073 
1074 	PMAP_UNLOCK();
1075 	return 1;
1076 }
1077 
1078 /*
1079  * Restrict given range to physical memory
1080  */
1081 void
1082 pmap_real_memory(paddr_t *start, psize_t *size)
1083 {
1084 	struct mem_region *mp;
1085 
1086 	for (mp = mem; mp->size; mp++) {
1087 		if (*start + *size > mp->start
1088 		    && *start < mp->start + mp->size) {
1089 			if (*start < mp->start) {
1090 				*size -= mp->start - *start;
1091 				*start = mp->start;
1092 			}
1093 			if (*start + *size > mp->start + mp->size)
1094 				*size = mp->start + mp->size - *start;
1095 			return;
1096 		}
1097 	}
1098 	*size = 0;
1099 }
1100 
1101 /*
1102  * Initialize anything else for pmap handling.
1103  * Called during vm_init().
1104  */
1105 void
1106 pmap_init(void)
1107 {
1108 
1109 	pmap_initialized = 1;
1110 }
1111 
1112 /*
1113  * How much virtual space does the kernel get?
1114  */
1115 void
1116 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1117 {
1118 	/*
1119 	 * For now, reserve one segment (minus some overhead) for kernel
1120 	 * virtual memory
1121 	 */
1122 	*start = VM_MIN_KERNEL_ADDRESS;
1123 	*end = VM_MAX_KERNEL_ADDRESS;
1124 }
1125 
1126 /*
1127  * Allocate, initialize, and return a new physical map.
1128  */
1129 pmap_t
1130 pmap_create(void)
1131 {
1132 	pmap_t pm;
1133 
1134 	pm = pool_get(&pmap_pool, PR_WAITOK);
1135 	KASSERT((vaddr_t)pm < VM_MIN_KERNEL_ADDRESS);
1136 	memset((void *)pm, 0, sizeof *pm);
1137 	pmap_pinit(pm);
1138 
1139 	DPRINTFN(CREATE, "pmap_create: pm %p:\n"
1140 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1141 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1142 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1143 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1144 	    pm,
1145 	    pm->pm_sr[0], pm->pm_sr[1],
1146 	    pm->pm_sr[2], pm->pm_sr[3],
1147 	    pm->pm_sr[4], pm->pm_sr[5],
1148 	    pm->pm_sr[6], pm->pm_sr[7],
1149 	    pm->pm_sr[8], pm->pm_sr[9],
1150 	    pm->pm_sr[10], pm->pm_sr[11],
1151 	    pm->pm_sr[12], pm->pm_sr[13],
1152 	    pm->pm_sr[14], pm->pm_sr[15]);
1153 	return pm;
1154 }
1155 
1156 /*
1157  * Initialize a preallocated and zeroed pmap structure.
1158  */
1159 void
1160 pmap_pinit(pmap_t pm)
1161 {
1162 	register_t entropy = MFTB();
1163 	register_t mask;
1164 	int i;
1165 
1166 	/*
1167 	 * Allocate some segment registers for this pmap.
1168 	 */
1169 	pm->pm_refs = 1;
1170 	PMAP_LOCK();
1171 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1172 		static register_t pmap_vsidcontext;
1173 		register_t hash;
1174 		unsigned int n;
1175 
1176 		/* Create a new value by multiplying by a prime adding in
1177 		 * entropy from the timebase register.  This is to make the
1178 		 * VSID more random so that the PT Hash function collides
1179 		 * less often. (note that the prime causes gcc to do shifts
1180 		 * instead of a multiply)
1181 		 */
1182 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1183 		hash = pmap_vsidcontext & (NPMAPS - 1);
1184 		if (hash == 0) {		/* 0 is special, avoid it */
1185 			entropy += 0xbadf00d;
1186 			continue;
1187 		}
1188 		n = hash >> 5;
1189 		mask = 1L << (hash & (VSID_NBPW-1));
1190 		hash = pmap_vsidcontext;
1191 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1192 			/* anything free in this bucket? */
1193 			if (~pmap_vsid_bitmap[n] == 0) {
1194 				entropy = hash ^ (hash >> 16);
1195 				continue;
1196 			}
1197 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1198 			mask = 1L << i;
1199 			hash &= ~(VSID_NBPW-1);
1200 			hash |= i;
1201 		}
1202 		hash &= PTE_VSID >> PTE_VSID_SHFT;
1203 		pmap_vsid_bitmap[n] |= mask;
1204 		pm->pm_vsid = hash;
1205 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1206 		for (i = 0; i < 16; i++)
1207 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1208 			    SR_NOEXEC;
1209 #endif
1210 		PMAP_UNLOCK();
1211 		return;
1212 	}
1213 	PMAP_UNLOCK();
1214 	panic("pmap_pinit: out of segments");
1215 }
1216 
1217 /*
1218  * Add a reference to the given pmap.
1219  */
1220 void
1221 pmap_reference(pmap_t pm)
1222 {
1223 	atomic_inc_uint(&pm->pm_refs);
1224 }
1225 
1226 /*
1227  * Retire the given pmap from service.
1228  * Should only be called if the map contains no valid mappings.
1229  */
1230 void
1231 pmap_destroy(pmap_t pm)
1232 {
1233 	membar_release();
1234 	if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1235 		membar_acquire();
1236 		pmap_release(pm);
1237 		pool_put(&pmap_pool, pm);
1238 	}
1239 }
1240 
1241 /*
1242  * Release any resources held by the given physical map.
1243  * Called when a pmap initialized by pmap_pinit is being released.
1244  */
1245 void
1246 pmap_release(pmap_t pm)
1247 {
1248 	int idx, mask;
1249 
1250 	KASSERT(pm->pm_stats.resident_count == 0);
1251 	KASSERT(pm->pm_stats.wired_count == 0);
1252 
1253 	PMAP_LOCK();
1254 	if (pm->pm_sr[0] == 0)
1255 		panic("pmap_release");
1256 	idx = pm->pm_vsid & (NPMAPS-1);
1257 	mask = 1 << (idx % VSID_NBPW);
1258 	idx /= VSID_NBPW;
1259 
1260 	KASSERT(pmap_vsid_bitmap[idx] & mask);
1261 	pmap_vsid_bitmap[idx] &= ~mask;
1262 	PMAP_UNLOCK();
1263 }
1264 
1265 /*
1266  * Copy the range specified by src_addr/len
1267  * from the source map to the range dst_addr/len
1268  * in the destination map.
1269  *
1270  * This routine is only advisory and need not do anything.
1271  */
1272 void
1273 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1274 	vsize_t len, vaddr_t src_addr)
1275 {
1276 	PMAPCOUNT(copies);
1277 }
1278 
1279 /*
1280  * Require that all active physical maps contain no
1281  * incorrect entries NOW.
1282  */
1283 void
1284 pmap_update(struct pmap *pmap)
1285 {
1286 	PMAPCOUNT(updates);
1287 	TLBSYNC();
1288 }
1289 
1290 static inline int
1291 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1292 {
1293 	int pteidx;
1294 	/*
1295 	 * We can find the actual pte entry without searching by
1296 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1297 	 * and by noticing the HID bit.
1298 	 */
1299 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1300 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1301 		pteidx ^= pmap_pteg_mask * 8;
1302 	return pteidx;
1303 }
1304 
1305 volatile struct pte *
1306 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1307 {
1308 	volatile struct pte *pt;
1309 
1310 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1311 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1312 		return NULL;
1313 #endif
1314 
1315 	/*
1316 	 * If we haven't been supplied the ptegidx, calculate it.
1317 	 */
1318 	if (pteidx == -1) {
1319 		int ptegidx;
1320 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1321 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1322 	}
1323 
1324 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1325 
1326 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1327 	return pt;
1328 #else
1329 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1330 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1331 		    "pvo but no valid pte index", pvo);
1332 	}
1333 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1334 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1335 		    "pvo but no valid pte", pvo);
1336 	}
1337 
1338 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1339 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1340 #if defined(DEBUG) || defined(PMAPCHECK)
1341 			pmap_pte_print(pt);
1342 #endif
1343 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1344 			    "pmap_pteg_table %p but invalid in pvo",
1345 			    pvo, pt);
1346 		}
1347 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1348 #if defined(DEBUG) || defined(PMAPCHECK)
1349 			pmap_pte_print(pt);
1350 #endif
1351 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1352 			    "not match pte %p in pmap_pteg_table",
1353 			    pvo, pt);
1354 		}
1355 		return pt;
1356 	}
1357 
1358 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1359 #if defined(DEBUG) || defined(PMAPCHECK)
1360 		pmap_pte_print(pt);
1361 #endif
1362 		panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1363 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1364 	}
1365 	return NULL;
1366 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1367 }
1368 
1369 struct pvo_entry *
1370 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1371 {
1372 	struct pvo_entry *pvo;
1373 	int ptegidx;
1374 
1375 	va &= ~ADDR_POFF;
1376 	ptegidx = va_to_pteg(pm, va);
1377 
1378 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1379 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1380 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1381 			panic("pmap_pvo_find_va: invalid pvo %p on "
1382 			    "list %#x (%p)", pvo, ptegidx,
1383 			     &pmap_pvo_table[ptegidx]);
1384 #endif
1385 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1386 			if (pteidx_p)
1387 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1388 			return pvo;
1389 		}
1390 	}
1391 	if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH))
1392 		panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1393 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1394 	return NULL;
1395 }
1396 
1397 #if defined(DEBUG) || defined(PMAPCHECK)
1398 void
1399 pmap_pvo_check(const struct pvo_entry *pvo)
1400 {
1401 	struct pvo_head *pvo_head;
1402 	struct pvo_entry *pvo0;
1403 	volatile struct pte *pt;
1404 	int failed = 0;
1405 
1406 	PMAP_LOCK();
1407 
1408 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1409 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1410 
1411 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1412 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1413 		    pvo, pvo->pvo_pmap);
1414 		failed = 1;
1415 	}
1416 
1417 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1418 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1419 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1420 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1421 		failed = 1;
1422 	}
1423 
1424 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1425 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1426 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1427 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1428 		failed = 1;
1429 	}
1430 
1431 	if (PVO_MANAGED_P(pvo)) {
1432 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1433 		LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1434 			if (pvo0 == pvo)
1435 				break;
1436 		}
1437 		if (pvo0 == NULL) {
1438 			printf("pmap_pvo_check: pvo %p: not present "
1439 			       "on its vlist head %p\n", pvo, pvo_head);
1440 			failed = 1;
1441 		}
1442 	} else {
1443 		KASSERT(pvo->pvo_vaddr >= VM_MIN_KERNEL_ADDRESS);
1444 		if (__predict_false(pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS))
1445 			failed = 1;
1446 	}
1447 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1448 		printf("pmap_pvo_check: pvo %p: not present "
1449 		    "on its olist head\n", pvo);
1450 		failed = 1;
1451 	}
1452 	pt = pmap_pvo_to_pte(pvo, -1);
1453 	if (pt == NULL) {
1454 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1455 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1456 			    "no PTE\n", pvo);
1457 			failed = 1;
1458 		}
1459 	} else {
1460 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1461 		    (uintptr_t) pt >=
1462 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1463 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1464 			    "pteg table\n", pvo, pt);
1465 			failed = 1;
1466 		}
1467 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1468 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1469 			    "no PTE\n", pvo);
1470 			failed = 1;
1471 		}
1472 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1473 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1474 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1475 			    pvo->pvo_pte.pte_hi,
1476 			    pt->pte_hi);
1477 			failed = 1;
1478 		}
1479 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1480 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1481 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1482 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1483 			    (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1484 			    (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1485 			failed = 1;
1486 		}
1487 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1488 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1489 			    " doesn't not match PVO's VA %#" _PRIxva "\n",
1490 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1491 			failed = 1;
1492 		}
1493 		if (failed)
1494 			pmap_pte_print(pt);
1495 	}
1496 	if (failed)
1497 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1498 		    pvo->pvo_pmap);
1499 
1500 	PMAP_UNLOCK();
1501 }
1502 #endif /* DEBUG || PMAPCHECK */
1503 
1504 /*
1505  * Search the PVO table looking for a non-wired entry.
1506  * If we find one, remove it and return it.
1507  */
1508 
1509 struct pvo_entry *
1510 pmap_pvo_reclaim(struct pmap *pm)
1511 {
1512 	struct pvo_tqhead *pvoh;
1513 	struct pvo_entry *pvo;
1514 	uint32_t idx, endidx;
1515 
1516 	endidx = pmap_pvo_reclaim_nextidx;
1517 	for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1518 	     idx = (idx + 1) & pmap_pteg_mask) {
1519 		pvoh = &pmap_pvo_table[idx];
1520 		TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1521 			if (!PVO_WIRED_P(pvo)) {
1522 				pmap_pvo_remove(pvo, -1, NULL);
1523 				pmap_pvo_reclaim_nextidx = idx;
1524 				PMAPCOUNT(pvos_reclaimed);
1525 				return pvo;
1526 			}
1527 		}
1528 	}
1529 	return NULL;
1530 }
1531 
1532 /*
1533  * This returns whether this is the first mapping of a page.
1534  */
1535 int
1536 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1537 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1538 {
1539 	struct pvo_entry *pvo;
1540 	struct pvo_tqhead *pvoh;
1541 	register_t msr;
1542 	int ptegidx;
1543 	int i;
1544 	int poolflags = PR_NOWAIT;
1545 
1546 	/*
1547 	 * Compute the PTE Group index.
1548 	 */
1549 	va &= ~ADDR_POFF;
1550 	ptegidx = va_to_pteg(pm, va);
1551 
1552 	msr = pmap_interrupts_off();
1553 
1554 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1555 	if (pmap_pvo_remove_depth > 0)
1556 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1557 	if (++pmap_pvo_enter_depth > 1)
1558 		panic("pmap_pvo_enter: called recursively!");
1559 #endif
1560 
1561 	/*
1562 	 * Remove any existing mapping for this page.  Reuse the
1563 	 * pvo entry if there a mapping.
1564 	 */
1565 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1566 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1567 #ifdef DEBUG
1568 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1569 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1570 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1571 			   va < VM_MIN_KERNEL_ADDRESS) {
1572 				printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1573 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1574 				printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1575 				    pvo->pvo_pte.pte_hi,
1576 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1577 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1578 #ifdef DDBX
1579 				Debugger();
1580 #endif
1581 			}
1582 #endif
1583 			PMAPCOUNT(mappings_replaced);
1584 			pmap_pvo_remove(pvo, -1, NULL);
1585 			break;
1586 		}
1587 	}
1588 
1589 	/*
1590 	 * If we aren't overwriting an mapping, try to allocate
1591 	 */
1592 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1593 	--pmap_pvo_enter_depth;
1594 #endif
1595 	pmap_interrupts_restore(msr);
1596 	if (pvo == NULL) {
1597 		pvo = pool_get(pl, poolflags);
1598 	}
1599 	KASSERT((vaddr_t)pvo < VM_MIN_KERNEL_ADDRESS);
1600 
1601 #ifdef DEBUG
1602 	/*
1603 	 * Exercise pmap_pvo_reclaim() a little.
1604 	 */
1605 	if (pvo && (flags & PMAP_CANFAIL) != 0 &&
1606 	    pmap_pvo_reclaim_debugctr++ > 0x1000 &&
1607 	    (pmap_pvo_reclaim_debugctr & 0xff) == 0) {
1608 		pool_put(pl, pvo);
1609 		pvo = NULL;
1610 	}
1611 #endif
1612 
1613 	msr = pmap_interrupts_off();
1614 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1615 	++pmap_pvo_enter_depth;
1616 #endif
1617 	if (pvo == NULL) {
1618 		pvo = pmap_pvo_reclaim(pm);
1619 		if (pvo == NULL) {
1620 			if ((flags & PMAP_CANFAIL) == 0)
1621 				panic("pmap_pvo_enter: failed");
1622 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1623 			pmap_pvo_enter_depth--;
1624 #endif
1625 			PMAPCOUNT(pvos_failed);
1626 			pmap_interrupts_restore(msr);
1627 			return ENOMEM;
1628 		}
1629 	}
1630 
1631 	pvo->pvo_vaddr = va;
1632 	pvo->pvo_pmap = pm;
1633 	pvo->pvo_vaddr &= ~ADDR_POFF;
1634 	if (flags & VM_PROT_EXECUTE) {
1635 		PMAPCOUNT(exec_mappings);
1636 		pvo_set_exec(pvo);
1637 	}
1638 	if (flags & PMAP_WIRED)
1639 		pvo->pvo_vaddr |= PVO_WIRED;
1640 	if (pvo_head != NULL) {
1641 		pvo->pvo_vaddr |= PVO_MANAGED;
1642 		PMAPCOUNT(mappings);
1643 	} else {
1644 		PMAPCOUNT(kernel_mappings);
1645 	}
1646 	pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1647 
1648 	if (pvo_head != NULL)
1649 		LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1650 	if (PVO_WIRED_P(pvo))
1651 		pvo->pvo_pmap->pm_stats.wired_count++;
1652 	pvo->pvo_pmap->pm_stats.resident_count++;
1653 #if defined(DEBUG)
1654 /*	if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */
1655 		DPRINTFN(PVOENTER,
1656 		    "pmap_pvo_enter: pvo %p: pm %p va %#" _PRIxva " pa %#" _PRIxpa "\n",
1657 		    pvo, pm, va, pa);
1658 #endif
1659 
1660 	/*
1661 	 * We hope this succeeds but it isn't required.
1662 	 */
1663 	pvoh = &pmap_pvo_table[ptegidx];
1664 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1665 	if (i >= 0) {
1666 		PVO_PTEGIDX_SET(pvo, i);
1667 		PVO_WHERE(pvo, ENTER_INSERT);
1668 		PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1669 		    ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1670 		TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1671 
1672 	} else {
1673 		/*
1674 		 * Since we didn't have room for this entry (which makes it
1675 		 * and evicted entry), place it at the head of the list.
1676 		 */
1677 		TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1678 		PMAPCOUNT(ptes_evicted);
1679 		pm->pm_evictions++;
1680 		/*
1681 		 * If this is a kernel page, make sure it's active.
1682 		 */
1683 		if (pm == pmap_kernel()) {
1684 			i = pmap_pte_spill(pm, va, false);
1685 			KASSERT(i);
1686 		}
1687 	}
1688 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1689 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1690 	pmap_pvo_enter_depth--;
1691 #endif
1692 	pmap_interrupts_restore(msr);
1693 	return 0;
1694 }
1695 
1696 static void
1697 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol)
1698 {
1699 	volatile struct pte *pt;
1700 	int ptegidx;
1701 
1702 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1703 	if (++pmap_pvo_remove_depth > 1)
1704 		panic("pmap_pvo_remove: called recursively!");
1705 #endif
1706 
1707 	/*
1708 	 * If we haven't been supplied the ptegidx, calculate it.
1709 	 */
1710 	if (pteidx == -1) {
1711 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1712 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1713 	} else {
1714 		ptegidx = pteidx >> 3;
1715 		if (pvo->pvo_pte.pte_hi & PTE_HID)
1716 			ptegidx ^= pmap_pteg_mask;
1717 	}
1718 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1719 
1720 	/*
1721 	 * If there is an active pte entry, we need to deactivate it
1722 	 * (and save the ref & chg bits).
1723 	 */
1724 	pt = pmap_pvo_to_pte(pvo, pteidx);
1725 	if (pt != NULL) {
1726 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1727 		PVO_WHERE(pvo, REMOVE);
1728 		PVO_PTEGIDX_CLR(pvo);
1729 		PMAPCOUNT(ptes_removed);
1730 	} else {
1731 		KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1732 		pvo->pvo_pmap->pm_evictions--;
1733 	}
1734 
1735 	/*
1736 	 * Account for executable mappings.
1737 	 */
1738 	if (PVO_EXECUTABLE_P(pvo))
1739 		pvo_clear_exec(pvo);
1740 
1741 	/*
1742 	 * Update our statistics.
1743 	 */
1744 	pvo->pvo_pmap->pm_stats.resident_count--;
1745 	if (PVO_WIRED_P(pvo))
1746 		pvo->pvo_pmap->pm_stats.wired_count--;
1747 
1748 	/*
1749 	 * If the page is managed:
1750 	 * Save the REF/CHG bits into their cache.
1751 	 * Remove the PVO from the P/V list.
1752 	 */
1753 	if (PVO_MANAGED_P(pvo)) {
1754 		register_t ptelo = pvo->pvo_pte.pte_lo;
1755 		struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1756 
1757 		if (pg != NULL) {
1758 			/*
1759 			 * If this page was changed and it is mapped exec,
1760 			 * invalidate it.
1761 			 */
1762 			if ((ptelo & PTE_CHG) &&
1763 			    (pmap_attr_fetch(pg) & PTE_EXEC)) {
1764 				struct pvo_head *pvoh = vm_page_to_pvoh(pg);
1765 				if (LIST_EMPTY(pvoh)) {
1766 					DPRINTFN(EXEC, "[pmap_pvo_remove: "
1767 					    "%#" _PRIxpa ": clear-exec]\n",
1768 					    VM_PAGE_TO_PHYS(pg));
1769 					pmap_attr_clear(pg, PTE_EXEC);
1770 					PMAPCOUNT(exec_uncached_pvo_remove);
1771 				} else {
1772 					DPRINTFN(EXEC, "[pmap_pvo_remove: "
1773 					    "%#" _PRIxpa ": syncicache]\n",
1774 					    VM_PAGE_TO_PHYS(pg));
1775 					pmap_syncicache(VM_PAGE_TO_PHYS(pg),
1776 					    PAGE_SIZE);
1777 					PMAPCOUNT(exec_synced_pvo_remove);
1778 				}
1779 			}
1780 
1781 			pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1782 		}
1783 		LIST_REMOVE(pvo, pvo_vlink);
1784 		PMAPCOUNT(unmappings);
1785 	} else {
1786 		PMAPCOUNT(kernel_unmappings);
1787 	}
1788 
1789 	/*
1790 	 * Remove the PVO from its list and return it to the pool.
1791 	 */
1792 	TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1793 	if (pvol) {
1794 		LIST_INSERT_HEAD(pvol, pvo, pvo_vlink);
1795 	}
1796 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1797 	pmap_pvo_remove_depth--;
1798 #endif
1799 }
1800 
1801 void
1802 pmap_pvo_free(struct pvo_entry *pvo)
1803 {
1804 
1805 	pool_put(&pmap_pvo_pool, pvo);
1806 }
1807 
1808 void
1809 pmap_pvo_free_list(struct pvo_head *pvol)
1810 {
1811 	struct pvo_entry *pvo, *npvo;
1812 
1813 	for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) {
1814 		npvo = LIST_NEXT(pvo, pvo_vlink);
1815 		LIST_REMOVE(pvo, pvo_vlink);
1816 		pmap_pvo_free(pvo);
1817 	}
1818 }
1819 
1820 /*
1821  * Mark a mapping as executable.
1822  * If this is the first executable mapping in the segment,
1823  * clear the noexec flag.
1824  */
1825 static void
1826 pvo_set_exec(struct pvo_entry *pvo)
1827 {
1828 	struct pmap *pm = pvo->pvo_pmap;
1829 
1830 	if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) {
1831 		return;
1832 	}
1833 	pvo->pvo_vaddr |= PVO_EXECUTABLE;
1834 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1835 	{
1836 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1837 		if (pm->pm_exec[sr]++ == 0) {
1838 			pm->pm_sr[sr] &= ~SR_NOEXEC;
1839 		}
1840 	}
1841 #endif
1842 }
1843 
1844 /*
1845  * Mark a mapping as non-executable.
1846  * If this was the last executable mapping in the segment,
1847  * set the noexec flag.
1848  */
1849 static void
1850 pvo_clear_exec(struct pvo_entry *pvo)
1851 {
1852 	struct pmap *pm = pvo->pvo_pmap;
1853 
1854 	if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) {
1855 		return;
1856 	}
1857 	pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1858 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1859 	{
1860 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1861 		if (--pm->pm_exec[sr] == 0) {
1862 			pm->pm_sr[sr] |= SR_NOEXEC;
1863 		}
1864 	}
1865 #endif
1866 }
1867 
1868 /*
1869  * Insert physical page at pa into the given pmap at virtual address va.
1870  */
1871 int
1872 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1873 {
1874 	struct mem_region *mp;
1875 	struct pvo_head *pvo_head;
1876 	struct vm_page *pg;
1877 	register_t pte_lo;
1878 	int error;
1879 	u_int was_exec = 0;
1880 
1881 	PMAP_LOCK();
1882 
1883 	if (__predict_false(!pmap_initialized)) {
1884 		pvo_head = NULL;
1885 		pg = NULL;
1886 		was_exec = PTE_EXEC;
1887 
1888 	} else {
1889 		pvo_head = pa_to_pvoh(pa, &pg);
1890 	}
1891 
1892 	DPRINTFN(ENTER,
1893 	    "pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1894 	    pm, va, pa, prot, flags);
1895 
1896 	/*
1897 	 * If this is a managed page, and it's the first reference to the
1898 	 * page clear the execness of the page.  Otherwise fetch the execness.
1899 	 */
1900 	if (pg != NULL)
1901 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1902 
1903 	DPRINTFN(ENTER, " was_exec=%d", was_exec);
1904 
1905 	/*
1906 	 * Assume the page is cache inhibited and access is guarded unless
1907 	 * it's in our available memory array.  If it is in the memory array,
1908 	 * asssume it's in memory coherent memory.
1909 	 */
1910 	if (flags & PMAP_MD_PREFETCHABLE) {
1911 		pte_lo = 0;
1912 	} else
1913 		pte_lo = PTE_G;
1914 
1915 	if ((flags & PMAP_NOCACHE) == 0) {
1916 		for (mp = mem; mp->size; mp++) {
1917 			if (pa >= mp->start && pa < mp->start + mp->size) {
1918 				pte_lo = PTE_M;
1919 				break;
1920 			}
1921 		}
1922 #ifdef MULTIPROCESSOR
1923 		if (((mfpvr() >> 16) & 0xffff) == MPC603e)
1924 			pte_lo = PTE_M;
1925 #endif
1926 	} else {
1927 		pte_lo |= PTE_I;
1928 	}
1929 
1930 	if (prot & VM_PROT_WRITE)
1931 		pte_lo |= PTE_BW;
1932 	else
1933 		pte_lo |= PTE_BR;
1934 
1935 	/*
1936 	 * If this was in response to a fault, "pre-fault" the PTE's
1937 	 * changed/referenced bit appropriately.
1938 	 */
1939 	if (flags & VM_PROT_WRITE)
1940 		pte_lo |= PTE_CHG;
1941 	if (flags & VM_PROT_ALL)
1942 		pte_lo |= PTE_REF;
1943 
1944 	/*
1945 	 * We need to know if this page can be executable
1946 	 */
1947 	flags |= (prot & VM_PROT_EXECUTE);
1948 
1949 	/*
1950 	 * Record mapping for later back-translation and pte spilling.
1951 	 * This will overwrite any existing mapping.
1952 	 */
1953 	error = pmap_pvo_enter(pm, &pmap_pvo_pool, pvo_head, va, pa, pte_lo, flags);
1954 
1955 	/*
1956 	 * Flush the real page from the instruction cache if this page is
1957 	 * mapped executable and cacheable and has not been flushed since
1958 	 * the last time it was modified.
1959 	 */
1960 	if (error == 0 &&
1961             (flags & VM_PROT_EXECUTE) &&
1962             (pte_lo & PTE_I) == 0 &&
1963 	    was_exec == 0) {
1964 		DPRINTFN(ENTER, " %s", "syncicache");
1965 		PMAPCOUNT(exec_synced);
1966 		pmap_syncicache(pa, PAGE_SIZE);
1967 		if (pg != NULL) {
1968 			pmap_attr_save(pg, PTE_EXEC);
1969 			PMAPCOUNT(exec_cached);
1970 #if defined(DEBUG) || defined(PMAPDEBUG)
1971 			if (pmapdebug & PMAPDEBUG_ENTER)
1972 				printf(" marked-as-exec");
1973 			else if (pmapdebug & PMAPDEBUG_EXEC)
1974 				printf("[pmap_enter: %#" _PRIxpa ": marked-as-exec]\n",
1975 				    VM_PAGE_TO_PHYS(pg));
1976 #endif
1977 		}
1978 	}
1979 
1980 	DPRINTFN(ENTER, ": error=%d\n", error);
1981 
1982 	PMAP_UNLOCK();
1983 
1984 	return error;
1985 }
1986 
1987 void
1988 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1989 {
1990 	struct mem_region *mp;
1991 	register_t pte_lo;
1992 	int error;
1993 
1994 #if defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA)
1995 	if (va < VM_MIN_KERNEL_ADDRESS)
1996 		panic("pmap_kenter_pa: attempt to enter "
1997 		    "non-kernel address %#" _PRIxva "!", va);
1998 #endif
1999 
2000 	DPRINTFN(KENTER,
2001 	    "pmap_kenter_pa(%#" _PRIxva ",%#" _PRIxpa ",%#x)\n", va, pa, prot);
2002 
2003 	PMAP_LOCK();
2004 
2005 	/*
2006 	 * Assume the page is cache inhibited and access is guarded unless
2007 	 * it's in our available memory array.  If it is in the memory array,
2008 	 * asssume it's in memory coherent memory.
2009 	 */
2010 	pte_lo = PTE_IG;
2011 	if ((flags & PMAP_NOCACHE) == 0) {
2012 		for (mp = mem; mp->size; mp++) {
2013 			if (pa >= mp->start && pa < mp->start + mp->size) {
2014 				pte_lo = PTE_M;
2015 				break;
2016 			}
2017 		}
2018 #ifdef MULTIPROCESSOR
2019 		if (((mfpvr() >> 16) & 0xffff) == MPC603e)
2020 			pte_lo = PTE_M;
2021 #endif
2022 	}
2023 
2024 	if (prot & VM_PROT_WRITE)
2025 		pte_lo |= PTE_BW;
2026 	else
2027 		pte_lo |= PTE_BR;
2028 
2029 	/*
2030 	 * We don't care about REF/CHG on PVOs on the unmanaged list.
2031 	 */
2032 	error = pmap_pvo_enter(pmap_kernel(), &pmap_pvo_pool,
2033 	    NULL, va, pa, pte_lo, prot|PMAP_WIRED);
2034 
2035 	if (error != 0)
2036 		panic("pmap_kenter_pa: failed to enter va %#" _PRIxva " pa %#" _PRIxpa ": %d",
2037 		      va, pa, error);
2038 
2039 	PMAP_UNLOCK();
2040 }
2041 
2042 void
2043 pmap_kremove(vaddr_t va, vsize_t len)
2044 {
2045 	if (va < VM_MIN_KERNEL_ADDRESS)
2046 		panic("pmap_kremove: attempt to remove "
2047 		    "non-kernel address %#" _PRIxva "!", va);
2048 
2049 	DPRINTFN(KREMOVE, "pmap_kremove(%#" _PRIxva ",%#" _PRIxva ")\n", va, len);
2050 	pmap_remove(pmap_kernel(), va, va + len);
2051 }
2052 
2053 /*
2054  * Remove the given range of mapping entries.
2055  */
2056 void
2057 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
2058 {
2059 	struct pvo_head pvol;
2060 	struct pvo_entry *pvo;
2061 	register_t msr;
2062 	int pteidx;
2063 
2064 	PMAP_LOCK();
2065 	LIST_INIT(&pvol);
2066 	msr = pmap_interrupts_off();
2067 	for (; va < endva; va += PAGE_SIZE) {
2068 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2069 		if (pvo != NULL) {
2070 			pmap_pvo_remove(pvo, pteidx, &pvol);
2071 		}
2072 	}
2073 	pmap_interrupts_restore(msr);
2074 	pmap_pvo_free_list(&pvol);
2075 	PMAP_UNLOCK();
2076 }
2077 
2078 #if defined(PMAP_OEA)
2079 #ifdef PPC_OEA601
2080 bool
2081 pmap_extract_ioseg601(vaddr_t va, paddr_t *pap)
2082 {
2083 	if ((MFPVR() >> 16) != MPC601)
2084 		return false;
2085 
2086 	const register_t sr = iosrtable[va >> ADDR_SR_SHFT];
2087 
2088 	if (SR601_VALID_P(sr) && SR601_PA_MATCH_P(sr, va)) {
2089 		if (pap)
2090 			*pap = va;
2091 		return true;
2092 	}
2093 	return false;
2094 }
2095 
2096 static bool
2097 pmap_extract_battable601(vaddr_t va, paddr_t *pap)
2098 {
2099 	const register_t batu = battable[va >> 23].batu;
2100 	const register_t batl = battable[va >> 23].batl;
2101 
2102 	if (BAT601_VALID_P(batl) && BAT601_VA_MATCH_P(batu, batl, va)) {
2103 		const register_t mask =
2104 		    (~(batl & BAT601_BSM) << 17) & ~0x1ffffL;
2105 		if (pap)
2106 			*pap = (batl & mask) | (va & ~mask);
2107 		return true;
2108 	}
2109 	return false;
2110 }
2111 #endif /* PPC_OEA601 */
2112 
2113 bool
2114 pmap_extract_battable(vaddr_t va, paddr_t *pap)
2115 {
2116 #ifdef PPC_OEA601
2117 	if ((MFPVR() >> 16) == MPC601)
2118 		return pmap_extract_battable601(va, pap);
2119 #endif /* PPC_OEA601 */
2120 
2121 	if (oeacpufeat & OEACPU_NOBAT)
2122 		return false;
2123 
2124 	const register_t batu = battable[BAT_VA2IDX(va)].batu;
2125 
2126 	if (BAT_VALID_P(batu, 0) && BAT_VA_MATCH_P(batu, va)) {
2127 		const register_t batl = battable[BAT_VA2IDX(va)].batl;
2128 		const register_t mask =
2129 		    (~(batu & (BAT_XBL|BAT_BL)) << 15) & ~0x1ffffL;
2130 		if (pap)
2131 			*pap = (batl & mask) | (va & ~mask);
2132 		return true;
2133 	}
2134 	return false;
2135 }
2136 #endif /* PMAP_OEA */
2137 
2138 /*
2139  * Get the physical page address for the given pmap/virtual address.
2140  */
2141 bool
2142 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
2143 {
2144 	struct pvo_entry *pvo;
2145 	register_t msr;
2146 
2147 	PMAP_LOCK();
2148 
2149 	/*
2150 	 * If this is the kernel pmap, check the battable and I/O
2151 	 * segments for a hit.  This is done only for regions outside
2152 	 * VM_MIN_KERNEL_ADDRESS-VM_MAX_KERNEL_ADDRESS.
2153 	 *
2154 	 * Be careful when checking VM_MAX_KERNEL_ADDRESS; you don't
2155 	 * want to wrap around to 0.
2156 	 */
2157 	if (pm == pmap_kernel() &&
2158 	    (va < VM_MIN_KERNEL_ADDRESS ||
2159 	     (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
2160 		KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
2161 #if defined(PMAP_OEA)
2162 #ifdef PPC_OEA601
2163 		if (pmap_extract_ioseg601(va, pap)) {
2164 			PMAP_UNLOCK();
2165 			return true;
2166 		}
2167 #endif /* PPC_OEA601 */
2168 		if (pmap_extract_battable(va, pap)) {
2169 			PMAP_UNLOCK();
2170 			return true;
2171 		}
2172 		/*
2173 		 * We still check the HTAB...
2174 		 */
2175 #elif defined(PMAP_OEA64_BRIDGE)
2176 		if (va < SEGMENT_LENGTH) {
2177 			if (pap)
2178 				*pap = va;
2179 			PMAP_UNLOCK();
2180 			return true;
2181 		}
2182 		/*
2183 		 * We still check the HTAB...
2184 		 */
2185 #elif defined(PMAP_OEA64)
2186 #error PPC_OEA64 not supported
2187 #endif /* PPC_OEA */
2188 	}
2189 
2190 	msr = pmap_interrupts_off();
2191 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2192 	if (pvo != NULL) {
2193 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2194 		if (pap)
2195 			*pap = (pvo->pvo_pte.pte_lo & PTE_RPGN)
2196 			    | (va & ADDR_POFF);
2197 	}
2198 	pmap_interrupts_restore(msr);
2199 	PMAP_UNLOCK();
2200 	return pvo != NULL;
2201 }
2202 
2203 /*
2204  * Lower the protection on the specified range of this pmap.
2205  */
2206 void
2207 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
2208 {
2209 	struct pvo_entry *pvo;
2210 	volatile struct pte *pt;
2211 	register_t msr;
2212 	int pteidx;
2213 
2214 	/*
2215 	 * Since this routine only downgrades protection, we should
2216 	 * always be called with at least one bit not set.
2217 	 */
2218 	KASSERT(prot != VM_PROT_ALL);
2219 
2220 	/*
2221 	 * If there is no protection, this is equivalent to
2222 	 * remove the pmap from the pmap.
2223 	 */
2224 	if ((prot & VM_PROT_READ) == 0) {
2225 		pmap_remove(pm, va, endva);
2226 		return;
2227 	}
2228 
2229 	PMAP_LOCK();
2230 
2231 	msr = pmap_interrupts_off();
2232 	for (; va < endva; va += PAGE_SIZE) {
2233 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2234 		if (pvo == NULL)
2235 			continue;
2236 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2237 
2238 		/*
2239 		 * Revoke executable if asked to do so.
2240 		 */
2241 		if ((prot & VM_PROT_EXECUTE) == 0)
2242 			pvo_clear_exec(pvo);
2243 
2244 #if 0
2245 		/*
2246 		 * If the page is already read-only, no change
2247 		 * needs to be made.
2248 		 */
2249 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
2250 			continue;
2251 #endif
2252 		/*
2253 		 * Grab the PTE pointer before we diddle with
2254 		 * the cached PTE copy.
2255 		 */
2256 		pt = pmap_pvo_to_pte(pvo, pteidx);
2257 		/*
2258 		 * Change the protection of the page.
2259 		 */
2260 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2261 		pvo->pvo_pte.pte_lo |= PTE_BR;
2262 
2263 		/*
2264 		 * If the PVO is in the page table, update
2265 		 * that pte at well.
2266 		 */
2267 		if (pt != NULL) {
2268 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2269 			PVO_WHERE(pvo, PMAP_PROTECT);
2270 			PMAPCOUNT(ptes_changed);
2271 		}
2272 
2273 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2274 	}
2275 	pmap_interrupts_restore(msr);
2276 	PMAP_UNLOCK();
2277 }
2278 
2279 void
2280 pmap_unwire(pmap_t pm, vaddr_t va)
2281 {
2282 	struct pvo_entry *pvo;
2283 	register_t msr;
2284 
2285 	PMAP_LOCK();
2286 	msr = pmap_interrupts_off();
2287 	pvo = pmap_pvo_find_va(pm, va, NULL);
2288 	if (pvo != NULL) {
2289 		if (PVO_WIRED_P(pvo)) {
2290 			pvo->pvo_vaddr &= ~PVO_WIRED;
2291 			pm->pm_stats.wired_count--;
2292 		}
2293 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2294 	}
2295 	pmap_interrupts_restore(msr);
2296 	PMAP_UNLOCK();
2297 }
2298 
2299 static void
2300 pmap_pp_protect(struct pmap_page *pp, paddr_t pa, vm_prot_t prot)
2301 {
2302 	struct pvo_head *pvo_head, pvol;
2303 	struct pvo_entry *pvo, *next_pvo;
2304 	volatile struct pte *pt;
2305 	register_t msr;
2306 
2307 	PMAP_LOCK();
2308 
2309 	KASSERT(prot != VM_PROT_ALL);
2310 	LIST_INIT(&pvol);
2311 	msr = pmap_interrupts_off();
2312 
2313 	/*
2314 	 * When UVM reuses a page, it does a pmap_page_protect with
2315 	 * VM_PROT_NONE.  At that point, we can clear the exec flag
2316 	 * since we know the page will have different contents.
2317 	 */
2318 	if ((prot & VM_PROT_READ) == 0) {
2319 		DPRINTFN(EXEC, "[pmap_page_protect: %#" _PRIxpa ": clear-exec]\n",
2320 		    pa);
2321 		if (pmap_pp_attr_fetch(pp) & PTE_EXEC) {
2322 			PMAPCOUNT(exec_uncached_page_protect);
2323 			pmap_pp_attr_clear(pp, PTE_EXEC);
2324 		}
2325 	}
2326 
2327 	pvo_head = &pp->pp_pvoh;
2328 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2329 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
2330 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2331 
2332 		/*
2333 		 * Downgrading to no mapping at all, we just remove the entry.
2334 		 */
2335 		if ((prot & VM_PROT_READ) == 0) {
2336 			pmap_pvo_remove(pvo, -1, &pvol);
2337 			continue;
2338 		}
2339 
2340 		/*
2341 		 * If EXEC permission is being revoked, just clear the
2342 		 * flag in the PVO.
2343 		 */
2344 		if ((prot & VM_PROT_EXECUTE) == 0)
2345 			pvo_clear_exec(pvo);
2346 
2347 		/*
2348 		 * If this entry is already RO, don't diddle with the
2349 		 * page table.
2350 		 */
2351 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2352 			PMAP_PVO_CHECK(pvo);
2353 			continue;
2354 		}
2355 
2356 		/*
2357 		 * Grab the PTE before the we diddle the bits so
2358 		 * pvo_to_pte can verify the pte contents are as
2359 		 * expected.
2360 		 */
2361 		pt = pmap_pvo_to_pte(pvo, -1);
2362 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2363 		pvo->pvo_pte.pte_lo |= PTE_BR;
2364 		if (pt != NULL) {
2365 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2366 			PVO_WHERE(pvo, PMAP_PAGE_PROTECT);
2367 			PMAPCOUNT(ptes_changed);
2368 		}
2369 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2370 	}
2371 	pmap_interrupts_restore(msr);
2372 	pmap_pvo_free_list(&pvol);
2373 
2374 	PMAP_UNLOCK();
2375 }
2376 
2377 /*
2378  * Lower the protection on the specified physical page.
2379  */
2380 void
2381 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
2382 {
2383 	struct vm_page_md *md = VM_PAGE_TO_MD(pg);
2384 
2385 	pmap_pp_protect(&md->mdpg_pp, VM_PAGE_TO_PHYS(pg), prot);
2386 }
2387 
2388 /*
2389  * Lower the protection on the physical page at the specified physical
2390  * address, which may not be managed and so may not have a struct
2391  * vm_page.
2392  */
2393 void
2394 pmap_pv_protect(paddr_t pa, vm_prot_t prot)
2395 {
2396 	struct pmap_page *pp;
2397 
2398 	if ((pp = pmap_pv_tracked(pa)) == NULL)
2399 		return;
2400 	pmap_pp_protect(pp, pa, prot);
2401 }
2402 
2403 /*
2404  * Activate the address space for the specified process.  If the process
2405  * is the current process, load the new MMU context.
2406  */
2407 void
2408 pmap_activate(struct lwp *l)
2409 {
2410 	struct pcb *pcb = lwp_getpcb(l);
2411 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2412 
2413 	DPRINTFN(ACTIVATE,
2414 	    "pmap_activate: lwp %p (curlwp %p)\n", l, curlwp);
2415 
2416 	/*
2417 	 * XXX Normally performed in cpu_lwp_fork().
2418 	 */
2419 	pcb->pcb_pm = pmap;
2420 
2421 	/*
2422 	* In theory, the SR registers need only be valid on return
2423 	* to user space wait to do them there.
2424 	*/
2425 	if (l == curlwp) {
2426 		/* Store pointer to new current pmap. */
2427 		curpm = pmap;
2428 	}
2429 }
2430 
2431 /*
2432  * Deactivate the specified process's address space.
2433  */
2434 void
2435 pmap_deactivate(struct lwp *l)
2436 {
2437 }
2438 
2439 bool
2440 pmap_query_bit(struct vm_page *pg, int ptebit)
2441 {
2442 	struct pvo_entry *pvo;
2443 	volatile struct pte *pt;
2444 	register_t msr;
2445 
2446 	PMAP_LOCK();
2447 
2448 	if (pmap_attr_fetch(pg) & ptebit) {
2449 		PMAP_UNLOCK();
2450 		return true;
2451 	}
2452 
2453 	msr = pmap_interrupts_off();
2454 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2455 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2456 		/*
2457 		 * See if we saved the bit off.  If so cache, it and return
2458 		 * success.
2459 		 */
2460 		if (pvo->pvo_pte.pte_lo & ptebit) {
2461 			pmap_attr_save(pg, ptebit);
2462 			PMAP_PVO_CHECK(pvo);		/* sanity check */
2463 			pmap_interrupts_restore(msr);
2464 			PMAP_UNLOCK();
2465 			return true;
2466 		}
2467 	}
2468 	/*
2469 	 * No luck, now go thru the hard part of looking at the ptes
2470 	 * themselves.  Sync so any pending REF/CHG bits are flushed
2471 	 * to the PTEs.
2472 	 */
2473 	SYNC();
2474 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2475 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2476 		/*
2477 		 * See if this pvo have a valid PTE.  If so, fetch the
2478 		 * REF/CHG bits from the valid PTE.  If the appropriate
2479 		 * ptebit is set, cache, it and return success.
2480 		 */
2481 		pt = pmap_pvo_to_pte(pvo, -1);
2482 		if (pt != NULL) {
2483 			pmap_pte_synch(pt, &pvo->pvo_pte);
2484 			if (pvo->pvo_pte.pte_lo & ptebit) {
2485 				pmap_attr_save(pg, ptebit);
2486 				PMAP_PVO_CHECK(pvo);		/* sanity check */
2487 				pmap_interrupts_restore(msr);
2488 				PMAP_UNLOCK();
2489 				return true;
2490 			}
2491 		}
2492 	}
2493 	pmap_interrupts_restore(msr);
2494 	PMAP_UNLOCK();
2495 	return false;
2496 }
2497 
2498 bool
2499 pmap_clear_bit(struct vm_page *pg, int ptebit)
2500 {
2501 	struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2502 	struct pvo_entry *pvo;
2503 	volatile struct pte *pt;
2504 	register_t msr;
2505 	int rv = 0;
2506 
2507 	PMAP_LOCK();
2508 	msr = pmap_interrupts_off();
2509 
2510 	/*
2511 	 * Fetch the cache value
2512 	 */
2513 	rv |= pmap_attr_fetch(pg);
2514 
2515 	/*
2516 	 * Clear the cached value.
2517 	 */
2518 	pmap_attr_clear(pg, ptebit);
2519 
2520 	/*
2521 	 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2522 	 * can reset the right ones).  Note that since the pvo entries and
2523 	 * list heads are accessed via BAT0 and are never placed in the
2524 	 * page table, we don't have to worry about further accesses setting
2525 	 * the REF/CHG bits.
2526 	 */
2527 	SYNC();
2528 
2529 	/*
2530 	 * For each pvo entry, clear pvo's ptebit.  If this pvo have a
2531 	 * valid PTE.  If so, clear the ptebit from the valid PTE.
2532 	 */
2533 	LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2534 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2535 		pt = pmap_pvo_to_pte(pvo, -1);
2536 		if (pt != NULL) {
2537 			/*
2538 			 * Only sync the PTE if the bit we are looking
2539 			 * for is not already set.
2540 			 */
2541 			if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2542 				pmap_pte_synch(pt, &pvo->pvo_pte);
2543 			/*
2544 			 * If the bit we are looking for was already set,
2545 			 * clear that bit in the pte.
2546 			 */
2547 			if (pvo->pvo_pte.pte_lo & ptebit)
2548 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2549 		}
2550 		rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2551 		pvo->pvo_pte.pte_lo &= ~ptebit;
2552 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2553 	}
2554 	pmap_interrupts_restore(msr);
2555 
2556 	/*
2557 	 * If we are clearing the modify bit and this page was marked EXEC
2558 	 * and the user of the page thinks the page was modified, then we
2559 	 * need to clean it from the icache if it's mapped or clear the EXEC
2560 	 * bit if it's not mapped.  The page itself might not have the CHG
2561 	 * bit set if the modification was done via DMA to the page.
2562 	 */
2563 	if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2564 		if (LIST_EMPTY(pvoh)) {
2565 			DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": clear-exec]\n",
2566 			    VM_PAGE_TO_PHYS(pg));
2567 			pmap_attr_clear(pg, PTE_EXEC);
2568 			PMAPCOUNT(exec_uncached_clear_modify);
2569 		} else {
2570 			DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": syncicache]\n",
2571 			    VM_PAGE_TO_PHYS(pg));
2572 			pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE);
2573 			PMAPCOUNT(exec_synced_clear_modify);
2574 		}
2575 	}
2576 	PMAP_UNLOCK();
2577 	return (rv & ptebit) != 0;
2578 }
2579 
2580 void
2581 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2582 {
2583 	struct pvo_entry *pvo;
2584 	size_t offset = va & ADDR_POFF;
2585 	int s;
2586 
2587 	PMAP_LOCK();
2588 	s = splvm();
2589 	while (len > 0) {
2590 		size_t seglen = PAGE_SIZE - offset;
2591 		if (seglen > len)
2592 			seglen = len;
2593 		pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2594 		if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) {
2595 			pmap_syncicache(
2596 			    (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2597 			PMAP_PVO_CHECK(pvo);
2598 		}
2599 		va += seglen;
2600 		len -= seglen;
2601 		offset = 0;
2602 	}
2603 	splx(s);
2604 	PMAP_UNLOCK();
2605 }
2606 
2607 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2608 void
2609 pmap_pte_print(volatile struct pte *pt)
2610 {
2611 	printf("PTE %p: ", pt);
2612 
2613 #if defined(PMAP_OEA)
2614 	/* High word: */
2615 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2616 #else
2617 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2618 #endif /* PMAP_OEA */
2619 
2620 	printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2621 	printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2622 
2623 	printf("%#" _PRIxpte " %#" _PRIxpte "",
2624 	    (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2625 	    pt->pte_hi & PTE_API);
2626 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2627 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2628 #else
2629 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2630 #endif /* PMAP_OEA */
2631 
2632 	/* Low word: */
2633 #if defined (PMAP_OEA)
2634 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2635 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2636 #else
2637 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2638 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2639 #endif
2640 	printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2641 	printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2642 	printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2643 	printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2644 	printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2645 	printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2646 	switch (pt->pte_lo & PTE_PP) {
2647 	case PTE_BR: printf("br]\n"); break;
2648 	case PTE_BW: printf("bw]\n"); break;
2649 	case PTE_SO: printf("so]\n"); break;
2650 	case PTE_SW: printf("sw]\n"); break;
2651 	}
2652 }
2653 #endif
2654 
2655 #if defined(DDB)
2656 void
2657 pmap_pteg_check(void)
2658 {
2659 	volatile struct pte *pt;
2660 	int i;
2661 	int ptegidx;
2662 	u_int p_valid = 0;
2663 	u_int s_valid = 0;
2664 	u_int invalid = 0;
2665 
2666 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2667 		for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2668 			if (pt->pte_hi & PTE_VALID) {
2669 				if (pt->pte_hi & PTE_HID)
2670 					s_valid++;
2671 				else
2672 				{
2673 					p_valid++;
2674 				}
2675 			} else
2676 				invalid++;
2677 		}
2678 	}
2679 	printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2680 		p_valid, p_valid, s_valid, s_valid,
2681 		invalid, invalid);
2682 }
2683 
2684 void
2685 pmap_print_mmuregs(void)
2686 {
2687 	int i;
2688 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2689 	u_int cpuvers;
2690 #endif
2691 #ifndef PMAP_OEA64
2692 	vaddr_t addr;
2693 	register_t soft_sr[16];
2694 #endif
2695 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2696 	struct bat soft_ibat[4];
2697 	struct bat soft_dbat[4];
2698 #endif
2699 	paddr_t sdr1;
2700 
2701 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2702 	cpuvers = MFPVR() >> 16;
2703 #endif
2704 	__asm volatile ("mfsdr1 %0" : "=r"(sdr1));
2705 #ifndef PMAP_OEA64
2706 	addr = 0;
2707 	for (i = 0; i < 16; i++) {
2708 		soft_sr[i] = MFSRIN(addr);
2709 		addr += (1 << ADDR_SR_SHFT);
2710 	}
2711 #endif
2712 
2713 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2714 	/* read iBAT (601: uBAT) registers */
2715 	__asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2716 	__asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2717 	__asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2718 	__asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2719 	__asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2720 	__asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2721 	__asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2722 	__asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2723 
2724 
2725 	if (cpuvers != MPC601) {
2726 		/* read dBAT registers */
2727 		__asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2728 		__asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2729 		__asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2730 		__asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2731 		__asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2732 		__asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2733 		__asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2734 		__asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2735 	}
2736 #endif
2737 
2738 	printf("SDR1:\t%#" _PRIxpa "\n", sdr1);
2739 #ifndef PMAP_OEA64
2740 	printf("SR[]:\t");
2741 	for (i = 0; i < 4; i++)
2742 		printf("0x%08lx,   ", soft_sr[i]);
2743 	printf("\n\t");
2744 	for ( ; i < 8; i++)
2745 		printf("0x%08lx,   ", soft_sr[i]);
2746 	printf("\n\t");
2747 	for ( ; i < 12; i++)
2748 		printf("0x%08lx,   ", soft_sr[i]);
2749 	printf("\n\t");
2750 	for ( ; i < 16; i++)
2751 		printf("0x%08lx,   ", soft_sr[i]);
2752 	printf("\n");
2753 #endif
2754 
2755 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2756 	printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2757 	for (i = 0; i < 4; i++) {
2758 		printf("0x%08lx 0x%08lx, ",
2759 			soft_ibat[i].batu, soft_ibat[i].batl);
2760 		if (i == 1)
2761 			printf("\n\t");
2762 	}
2763 	if (cpuvers != MPC601) {
2764 		printf("\ndBAT[]:\t");
2765 		for (i = 0; i < 4; i++) {
2766 			printf("0x%08lx 0x%08lx, ",
2767 				soft_dbat[i].batu, soft_dbat[i].batl);
2768 			if (i == 1)
2769 				printf("\n\t");
2770 		}
2771 	}
2772 	printf("\n");
2773 #endif /* PMAP_OEA... */
2774 }
2775 
2776 void
2777 pmap_print_pte(pmap_t pm, vaddr_t va)
2778 {
2779 	struct pvo_entry *pvo;
2780 	volatile struct pte *pt;
2781 	int pteidx;
2782 
2783 	pvo = pmap_pvo_find_va(pm, va, &pteidx);
2784 	if (pvo != NULL) {
2785 		pt = pmap_pvo_to_pte(pvo, pteidx);
2786 		if (pt != NULL) {
2787 			printf("VA %#" _PRIxva " -> %p -> %s %#" _PRIxpte ", %#" _PRIxpte "\n",
2788 				va, pt,
2789 				pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2790 				pt->pte_hi, pt->pte_lo);
2791 		} else {
2792 			printf("No valid PTE found\n");
2793 		}
2794 	} else {
2795 		printf("Address not in pmap\n");
2796 	}
2797 }
2798 
2799 void
2800 pmap_pteg_dist(void)
2801 {
2802 	struct pvo_entry *pvo;
2803 	int ptegidx;
2804 	int depth;
2805 	int max_depth = 0;
2806 	unsigned int depths[64];
2807 
2808 	memset(depths, 0, sizeof(depths));
2809 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2810 		depth = 0;
2811 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2812 			depth++;
2813 		}
2814 		if (depth > max_depth)
2815 			max_depth = depth;
2816 		if (depth > 63)
2817 			depth = 63;
2818 		depths[depth]++;
2819 	}
2820 
2821 	for (depth = 0; depth < 64; depth++) {
2822 		printf("  [%2d]: %8u", depth, depths[depth]);
2823 		if ((depth & 3) == 3)
2824 			printf("\n");
2825 		if (depth == max_depth)
2826 			break;
2827 	}
2828 	if ((depth & 3) != 3)
2829 		printf("\n");
2830 	printf("Max depth found was %d\n", max_depth);
2831 }
2832 #endif /* DEBUG */
2833 
2834 #if defined(PMAPCHECK) || defined(DEBUG)
2835 void
2836 pmap_pvo_verify(void)
2837 {
2838 	int ptegidx;
2839 	int s;
2840 
2841 	s = splvm();
2842 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2843 		struct pvo_entry *pvo;
2844 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2845 			if ((uintptr_t) pvo >= SEGMENT_LENGTH)
2846 				panic("pmap_pvo_verify: invalid pvo %p "
2847 				    "on list %#x", pvo, ptegidx);
2848 			pmap_pvo_check(pvo);
2849 		}
2850 	}
2851 	splx(s);
2852 }
2853 #endif /* PMAPCHECK */
2854 
2855 void *
2856 pmap_pool_alloc(struct pool *pp, int flags)
2857 {
2858 	struct pvo_page *pvop;
2859 	struct vm_page *pg;
2860 
2861 	if (uvm.page_init_done != true) {
2862 		return (void *) uvm_pageboot_alloc(PAGE_SIZE);
2863 	}
2864 
2865 	PMAP_LOCK();
2866 	pvop = SIMPLEQ_FIRST(&pmap_pvop_head);
2867 	if (pvop != NULL) {
2868 		pmap_pvop_free--;
2869 		SIMPLEQ_REMOVE_HEAD(&pmap_pvop_head, pvop_link);
2870 		PMAP_UNLOCK();
2871 		return pvop;
2872 	}
2873 	PMAP_UNLOCK();
2874  again:
2875 	pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE,
2876 	    UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256);
2877 	if (__predict_false(pg == NULL)) {
2878 		if (flags & PR_WAITOK) {
2879 			uvm_wait("plpg");
2880 			goto again;
2881 		} else {
2882 			return (0);
2883 		}
2884 	}
2885 	KDASSERT(VM_PAGE_TO_PHYS(pg) == (uintptr_t)VM_PAGE_TO_PHYS(pg));
2886 	return (void *)(uintptr_t) VM_PAGE_TO_PHYS(pg);
2887 }
2888 
2889 void
2890 pmap_pool_free(struct pool *pp, void *va)
2891 {
2892 	struct pvo_page *pvop;
2893 
2894 	PMAP_LOCK();
2895 	pvop = va;
2896 	SIMPLEQ_INSERT_HEAD(&pmap_pvop_head, pvop, pvop_link);
2897 	pmap_pvop_free++;
2898 	if (pmap_pvop_free > pmap_pvop_maxfree)
2899 		pmap_pvop_maxfree = pmap_pvop_free;
2900 	PMAP_UNLOCK();
2901 #if 0
2902 	uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va));
2903 #endif
2904 }
2905 
2906 /*
2907  * This routine in bootstraping to steal to-be-managed memory (which will
2908  * then be unmanaged).  We use it to grab from the first 256MB for our
2909  * pmap needs and above 256MB for other stuff.
2910  */
2911 vaddr_t
2912 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2913 {
2914 	vsize_t size;
2915 	vaddr_t va;
2916 	paddr_t start, end, pa = 0;
2917 	int npgs, freelist;
2918 	uvm_physseg_t bank;
2919 
2920 	if (uvm.page_init_done == true)
2921 		panic("pmap_steal_memory: called _after_ bootstrap");
2922 
2923 	*vstartp = VM_MIN_KERNEL_ADDRESS;
2924 	*vendp = VM_MAX_KERNEL_ADDRESS;
2925 
2926 	size = round_page(vsize);
2927 	npgs = atop(size);
2928 
2929 	/*
2930 	 * PA 0 will never be among those given to UVM so we can use it
2931 	 * to indicate we couldn't steal any memory.
2932 	 */
2933 
2934 	for (bank = uvm_physseg_get_first();
2935 	     uvm_physseg_valid_p(bank);
2936 	     bank = uvm_physseg_get_next(bank)) {
2937 
2938 		freelist = uvm_physseg_get_free_list(bank);
2939 		start = uvm_physseg_get_start(bank);
2940 		end = uvm_physseg_get_end(bank);
2941 
2942 		if (freelist == VM_FREELIST_FIRST256 &&
2943 		    (end - start) >= npgs) {
2944 			pa = ptoa(start);
2945 			break;
2946 		}
2947 	}
2948 
2949 	if (pa == 0)
2950 		panic("pmap_steal_memory: no approriate memory to steal!");
2951 
2952 	uvm_physseg_unplug(start, npgs);
2953 
2954 	va = (vaddr_t) pa;
2955 	memset((void *) va, 0, size);
2956 	pmap_pages_stolen += npgs;
2957 #ifdef DEBUG
2958 	if (pmapdebug && npgs > 1) {
2959 		u_int cnt = 0;
2960 	for (bank = uvm_physseg_get_first();
2961 	     uvm_physseg_valid_p(bank);
2962 	     bank = uvm_physseg_get_next(bank)) {
2963 		cnt += uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank);
2964 		}
2965 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2966 		    npgs, pmap_pages_stolen, cnt);
2967 	}
2968 #endif
2969 
2970 	return va;
2971 }
2972 
2973 /*
2974  * Find a chuck of memory with right size and alignment.
2975  */
2976 paddr_t
2977 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2978 {
2979 	struct mem_region *mp;
2980 	paddr_t s, e;
2981 	int i, j;
2982 
2983 	size = round_page(size);
2984 
2985 	DPRINTFN(BOOT,
2986 	    "pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
2987 	    size, alignment, at_end);
2988 
2989 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2990 		panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
2991 		    alignment);
2992 
2993 	if (at_end) {
2994 		if (alignment != PAGE_SIZE)
2995 			panic("pmap_boot_find_memory: invalid ending "
2996 			    "alignment %#" _PRIxpa, alignment);
2997 
2998 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
2999 			s = mp->start + mp->size - size;
3000 			if (s >= mp->start && mp->size >= size) {
3001 				DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3002 				DPRINTFN(BOOT,
3003 				    "pmap_boot_find_memory: b-avail[%d] start "
3004 				    "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3005 				     mp->start, mp->size);
3006 				mp->size -= size;
3007 				DPRINTFN(BOOT,
3008 				    "pmap_boot_find_memory: a-avail[%d] start "
3009 				    "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3010 				     mp->start, mp->size);
3011 				return s;
3012 			}
3013 		}
3014 		panic("pmap_boot_find_memory: no available memory");
3015 	}
3016 
3017 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3018 		s = (mp->start + alignment - 1) & ~(alignment-1);
3019 		e = s + size;
3020 
3021 		/*
3022 		 * Is the calculated region entirely within the region?
3023 		 */
3024 		if (s < mp->start || e > mp->start + mp->size)
3025 			continue;
3026 
3027 		DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3028 		if (s == mp->start) {
3029 			/*
3030 			 * If the block starts at the beginning of region,
3031 			 * adjust the size & start. (the region may now be
3032 			 * zero in length)
3033 			 */
3034 			DPRINTFN(BOOT,
3035 			    "pmap_boot_find_memory: b-avail[%d] start "
3036 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3037 			mp->start += size;
3038 			mp->size -= size;
3039 			DPRINTFN(BOOT,
3040 			    "pmap_boot_find_memory: a-avail[%d] start "
3041 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3042 		} else if (e == mp->start + mp->size) {
3043 			/*
3044 			 * If the block starts at the beginning of region,
3045 			 * adjust only the size.
3046 			 */
3047 			DPRINTFN(BOOT,
3048 			    "pmap_boot_find_memory: b-avail[%d] start "
3049 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3050 			mp->size -= size;
3051 			DPRINTFN(BOOT,
3052 			    "pmap_boot_find_memory: a-avail[%d] start "
3053 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3054 		} else {
3055 			/*
3056 			 * Block is in the middle of the region, so we
3057 			 * have to split it in two.
3058 			 */
3059 			for (j = avail_cnt; j > i + 1; j--) {
3060 				avail[j] = avail[j-1];
3061 			}
3062 			DPRINTFN(BOOT,
3063 			    "pmap_boot_find_memory: b-avail[%d] start "
3064 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3065 			mp[1].start = e;
3066 			mp[1].size = mp[0].start + mp[0].size - e;
3067 			mp[0].size = s - mp[0].start;
3068 			avail_cnt++;
3069 			for (; i < avail_cnt; i++) {
3070 				DPRINTFN(BOOT,
3071 				    "pmap_boot_find_memory: a-avail[%d] "
3072 				    "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3073 				     avail[i].start, avail[i].size);
3074 			}
3075 		}
3076 		KASSERT(s == (uintptr_t) s);
3077 		return s;
3078 	}
3079 	panic("pmap_boot_find_memory: not enough memory for "
3080 	    "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3081 }
3082 
3083 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3084 #if defined (PMAP_OEA64_BRIDGE)
3085 int
3086 pmap_setup_segment0_map(int use_large_pages, ...)
3087 {
3088     vaddr_t va, va_end;
3089 
3090     register_t pte_lo = 0x0;
3091     int ptegidx = 0;
3092     struct pte pte;
3093     va_list ap;
3094 
3095     /* Coherent + Supervisor RW, no user access */
3096     pte_lo = PTE_M;
3097 
3098     /* XXXSL
3099      * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3100      * these have to take priority.
3101      */
3102     for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3103         ptegidx = va_to_pteg(pmap_kernel(), va);
3104         pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3105         (void)pmap_pte_insert(ptegidx, &pte);
3106     }
3107 
3108     va_start(ap, use_large_pages);
3109     while (1) {
3110         paddr_t pa;
3111         size_t size;
3112 
3113         va = va_arg(ap, vaddr_t);
3114 
3115         if (va == 0)
3116             break;
3117 
3118         pa = va_arg(ap, paddr_t);
3119         size = va_arg(ap, size_t);
3120 
3121         for (va_end = va + size; va < va_end; va += 0x1000, pa += 0x1000) {
3122 #if 0
3123 	    printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__,  va, pa);
3124 #endif
3125             ptegidx = va_to_pteg(pmap_kernel(), va);
3126             pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3127             (void)pmap_pte_insert(ptegidx, &pte);
3128         }
3129     }
3130     va_end(ap);
3131 
3132     TLBSYNC();
3133     SYNC();
3134     return (0);
3135 }
3136 #endif /* PMAP_OEA64_BRIDGE */
3137 
3138 /*
3139  * Set up the bottom level of the data structures necessary for the kernel
3140  * to manage memory.  MMU hardware is programmed in pmap_bootstrap2().
3141  */
3142 void
3143 pmap_bootstrap1(paddr_t kernelstart, paddr_t kernelend)
3144 {
3145 	struct mem_region *mp, tmp;
3146 	paddr_t s, e;
3147 	psize_t size;
3148 	int i, j;
3149 
3150 	/*
3151 	 * Get memory.
3152 	 */
3153 	mem_regions(&mem, &avail);
3154 #if defined(DEBUG)
3155 	if (pmapdebug & PMAPDEBUG_BOOT) {
3156 		printf("pmap_bootstrap: memory configuration:\n");
3157 		for (mp = mem; mp->size; mp++) {
3158 			printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3159 				mp->start, mp->size);
3160 		}
3161 		for (mp = avail; mp->size; mp++) {
3162 			printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3163 				mp->start, mp->size);
3164 		}
3165 	}
3166 #endif
3167 
3168 	/*
3169 	 * Find out how much physical memory we have and in how many chunks.
3170 	 */
3171 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3172 		if (mp->start >= pmap_memlimit)
3173 			continue;
3174 		if (mp->start + mp->size > pmap_memlimit) {
3175 			size = pmap_memlimit - mp->start;
3176 			physmem += btoc(size);
3177 		} else {
3178 			physmem += btoc(mp->size);
3179 		}
3180 		mem_cnt++;
3181 	}
3182 
3183 	/*
3184 	 * Count the number of available entries.
3185 	 */
3186 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
3187 		avail_cnt++;
3188 
3189 	/*
3190 	 * Page align all regions.
3191 	 */
3192 	kernelstart = trunc_page(kernelstart);
3193 	kernelend = round_page(kernelend);
3194 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3195 		s = round_page(mp->start);
3196 		mp->size -= (s - mp->start);
3197 		mp->size = trunc_page(mp->size);
3198 		mp->start = s;
3199 		e = mp->start + mp->size;
3200 
3201 		DPRINTFN(BOOT,
3202 		    "pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3203 		    i, mp->start, mp->size);
3204 
3205 		/*
3206 		 * Don't allow the end to run beyond our artificial limit
3207 		 */
3208 		if (e > pmap_memlimit)
3209 			e = pmap_memlimit;
3210 
3211 		/*
3212 		 * Is this region empty or strange?  skip it.
3213 		 */
3214 		if (e <= s) {
3215 			mp->start = 0;
3216 			mp->size = 0;
3217 			continue;
3218 		}
3219 
3220 		/*
3221 		 * Does this overlap the beginning of kernel?
3222 		 *   Does extend past the end of the kernel?
3223 		 */
3224 		else if (s < kernelstart && e > kernelstart) {
3225 			if (e > kernelend) {
3226 				avail[avail_cnt].start = kernelend;
3227 				avail[avail_cnt].size = e - kernelend;
3228 				avail_cnt++;
3229 			}
3230 			mp->size = kernelstart - s;
3231 		}
3232 		/*
3233 		 * Check whether this region overlaps the end of the kernel.
3234 		 */
3235 		else if (s < kernelend && e > kernelend) {
3236 			mp->start = kernelend;
3237 			mp->size = e - kernelend;
3238 		}
3239 		/*
3240 		 * Look whether this regions is completely inside the kernel.
3241 		 * Nuke it if it does.
3242 		 */
3243 		else if (s >= kernelstart && e <= kernelend) {
3244 			mp->start = 0;
3245 			mp->size = 0;
3246 		}
3247 		/*
3248 		 * If the user imposed a memory limit, enforce it.
3249 		 */
3250 		else if (s >= pmap_memlimit) {
3251 			mp->start = -PAGE_SIZE;	/* let's know why */
3252 			mp->size = 0;
3253 		}
3254 		else {
3255 			mp->start = s;
3256 			mp->size = e - s;
3257 		}
3258 		DPRINTFN(BOOT,
3259 		    "pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3260 		    i, mp->start, mp->size);
3261 	}
3262 
3263 	/*
3264 	 * Move (and uncount) all the null return to the end.
3265 	 */
3266 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3267 		if (mp->size == 0) {
3268 			tmp = avail[i];
3269 			avail[i] = avail[--avail_cnt];
3270 			avail[avail_cnt] = avail[i];
3271 		}
3272 	}
3273 
3274 	/*
3275 	 * (Bubble)sort them into ascending order.
3276 	 */
3277 	for (i = 0; i < avail_cnt; i++) {
3278 		for (j = i + 1; j < avail_cnt; j++) {
3279 			if (avail[i].start > avail[j].start) {
3280 				tmp = avail[i];
3281 				avail[i] = avail[j];
3282 				avail[j] = tmp;
3283 			}
3284 		}
3285 	}
3286 
3287 	/*
3288 	 * Make sure they don't overlap.
3289 	 */
3290 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3291 		if (mp[0].start + mp[0].size > mp[1].start) {
3292 			mp[0].size = mp[1].start - mp[0].start;
3293 		}
3294 		DPRINTFN(BOOT,
3295 		    "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3296 		    i, mp->start, mp->size);
3297 	}
3298 	DPRINTFN(BOOT,
3299 	    "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3300 	    i, mp->start, mp->size);
3301 
3302 #ifdef	PTEGCOUNT
3303 	pmap_pteg_cnt = PTEGCOUNT;
3304 #else /* PTEGCOUNT */
3305 
3306 	pmap_pteg_cnt = 0x1000;
3307 
3308 	while (pmap_pteg_cnt < physmem)
3309 		pmap_pteg_cnt <<= 1;
3310 
3311 	pmap_pteg_cnt >>= 1;
3312 #endif /* PTEGCOUNT */
3313 
3314 #ifdef DEBUG
3315 	DPRINTFN(BOOT, "pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt);
3316 #endif
3317 
3318 	/*
3319 	 * Find suitably aligned memory for PTEG hash table.
3320 	 */
3321 	size = pmap_pteg_cnt * sizeof(struct pteg);
3322 	pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3323 
3324 #ifdef DEBUG
3325 	DPRINTFN(BOOT,
3326 		"PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table);
3327 #endif
3328 
3329 
3330 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3331 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
3332 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > 256MB",
3333 		    pmap_pteg_table, size);
3334 #endif
3335 
3336 	memset(__UNVOLATILE(pmap_pteg_table), 0,
3337 		pmap_pteg_cnt * sizeof(struct pteg));
3338 	pmap_pteg_mask = pmap_pteg_cnt - 1;
3339 
3340 	/*
3341 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3342 	 * with pages.  So we just steal them before giving them to UVM.
3343 	 */
3344 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3345 	pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3346 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3347 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
3348 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > 256MB",
3349 		    pmap_pvo_table, size);
3350 #endif
3351 
3352 	for (i = 0; i < pmap_pteg_cnt; i++)
3353 		TAILQ_INIT(&pmap_pvo_table[i]);
3354 
3355 #ifndef MSGBUFADDR
3356 	/*
3357 	 * Allocate msgbuf in high memory.
3358 	 */
3359 	msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3360 #endif
3361 
3362 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3363 		paddr_t pfstart = atop(mp->start);
3364 		paddr_t pfend = atop(mp->start + mp->size);
3365 		if (mp->size == 0)
3366 			continue;
3367 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
3368 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3369 				VM_FREELIST_FIRST256);
3370 		} else if (mp->start >= SEGMENT_LENGTH) {
3371 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3372 				VM_FREELIST_DEFAULT);
3373 		} else {
3374 			pfend = atop(SEGMENT_LENGTH);
3375 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3376 				VM_FREELIST_FIRST256);
3377 			pfstart = atop(SEGMENT_LENGTH);
3378 			pfend = atop(mp->start + mp->size);
3379 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3380 				VM_FREELIST_DEFAULT);
3381 		}
3382 	}
3383 
3384 	/*
3385 	 * Make sure kernel vsid is allocated as well as VSID 0.
3386 	 */
3387 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3388 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3389 	pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3390 		|= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3391 	pmap_vsid_bitmap[0] |= 1;
3392 
3393 	/*
3394 	 * Initialize kernel pmap.
3395 	 */
3396 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
3397 	for (i = 0; i < 16; i++) {
3398  		pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3399 	}
3400 	pmap_kernel()->pm_vsid = KERNEL_VSIDBITS;
3401 
3402 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3403 #ifdef KERNEL2_SR
3404 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3405 #endif
3406 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3407 
3408 #if defined(PMAP_OEA) && defined(PPC_OEA601)
3409 	if ((MFPVR() >> 16) == MPC601) {
3410 		for (i = 0; i < 16; i++) {
3411 			if (iosrtable[i] & SR601_T) {
3412 				pmap_kernel()->pm_sr[i] = iosrtable[i];
3413 			}
3414 		}
3415 	}
3416 #endif /* PMAP_OEA && PPC_OEA601 */
3417 
3418 #ifdef ALTIVEC
3419 	pmap_use_altivec = cpu_altivec;
3420 #endif
3421 
3422 #ifdef DEBUG
3423 	if (pmapdebug & PMAPDEBUG_BOOT) {
3424 		u_int cnt;
3425 		uvm_physseg_t bank;
3426 		char pbuf[9];
3427 		for (cnt = 0, bank = uvm_physseg_get_first();
3428 		     uvm_physseg_valid_p(bank);
3429 		     bank = uvm_physseg_get_next(bank)) {
3430 			cnt += uvm_physseg_get_avail_end(bank) -
3431 			    uvm_physseg_get_avail_start(bank);
3432 			printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3433 			    bank,
3434 			    ptoa(uvm_physseg_get_avail_start(bank)),
3435 			    ptoa(uvm_physseg_get_avail_end(bank)),
3436 			    ptoa(uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank)));
3437 		}
3438 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3439 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3440 		    pbuf, cnt);
3441 	}
3442 #endif
3443 
3444 	pool_init(&pmap_pvo_pool, sizeof(struct pvo_entry),
3445 	    sizeof(struct pvo_entry), 0, 0, "pmap_pvopl",
3446 	    &pmap_pool_allocator, IPL_VM);
3447 
3448 	pool_setlowat(&pmap_pvo_pool, 1008);
3449 
3450 	pool_init(&pmap_pool, sizeof(struct pmap),
3451 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_allocator,
3452 	    IPL_NONE);
3453 
3454 #if defined(PMAP_NEED_MAPKERNEL)
3455 	{
3456 		struct pmap *pm = pmap_kernel();
3457 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3458 		extern int etext[], kernel_text[];
3459 		vaddr_t va, va_etext = (paddr_t) etext;
3460 #endif
3461 		paddr_t pa, pa_end;
3462 		register_t sr;
3463 		struct pte pt;
3464 		unsigned int ptegidx;
3465 		int bank;
3466 
3467 		sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3468 		pm->pm_sr[0] = sr;
3469 
3470 		for (bank = 0; bank < vm_nphysseg; bank++) {
3471 			pa_end = ptoa(VM_PHYSMEM_PTR(bank)->avail_end);
3472 			pa = ptoa(VM_PHYSMEM_PTR(bank)->avail_start);
3473 			for (; pa < pa_end; pa += PAGE_SIZE) {
3474 				ptegidx = va_to_pteg(pm, pa);
3475 				pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3476 				pmap_pte_insert(ptegidx, &pt);
3477 			}
3478 		}
3479 
3480 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3481 		va = (vaddr_t) kernel_text;
3482 
3483 		for (pa = kernelstart; va < va_etext;
3484 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3485 			ptegidx = va_to_pteg(pm, va);
3486 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3487 			pmap_pte_insert(ptegidx, &pt);
3488 		}
3489 
3490 		for (; pa < kernelend;
3491 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3492 			ptegidx = va_to_pteg(pm, va);
3493 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3494 			pmap_pte_insert(ptegidx, &pt);
3495 		}
3496 
3497 		for (va = 0, pa = 0; va < kernelstart;
3498 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3499 			ptegidx = va_to_pteg(pm, va);
3500 			if (va < 0x3000)
3501 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3502 			else
3503 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3504 			pmap_pte_insert(ptegidx, &pt);
3505 		}
3506 		for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3507 		    pa += PAGE_SIZE, va += PAGE_SIZE) {
3508 			ptegidx = va_to_pteg(pm, va);
3509 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3510 			pmap_pte_insert(ptegidx, &pt);
3511 		}
3512 #endif /* PMAP_NEED_FULL_MAPKERNEL */
3513 	}
3514 #endif /* PMAP_NEED_MAPKERNEL */
3515 }
3516 
3517 /*
3518  * Using the data structures prepared in pmap_bootstrap1(), program
3519  * the MMU hardware.
3520  */
3521 void
3522 pmap_bootstrap2(void)
3523 {
3524 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
3525 	for (int i = 0; i < 16; i++) {
3526 		__asm volatile("mtsrin %0,%1"
3527 			:: "r"(pmap_kernel()->pm_sr[i]),
3528 			   "r"(i << ADDR_SR_SHFT));
3529 	}
3530 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3531 
3532 #if defined(PMAP_OEA)
3533 	__asm volatile("sync; mtsdr1 %0; isync"
3534 	    :
3535 	    : "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))
3536 	    : "memory");
3537 #elif defined(PMAP_OEA64) || defined(PMAP_OEA64_BRIDGE)
3538 	__asm volatile("sync; mtsdr1 %0; isync"
3539 	    :
3540 	    : "r"((uintptr_t)pmap_pteg_table |
3541 		(32 - __builtin_clz(pmap_pteg_mask >> 11)))
3542 	    : "memory");
3543 #endif
3544 	tlbia();
3545 
3546 #if defined(PMAPDEBUG)
3547 	if (pmapdebug)
3548 	    pmap_print_mmuregs();
3549 #endif
3550 }
3551 
3552 /*
3553  * This is not part of the defined PMAP interface and is specific to the
3554  * PowerPC architecture.  This is called during initppc, before the system
3555  * is really initialized.
3556  */
3557 void
3558 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3559 {
3560 	pmap_bootstrap1(kernelstart, kernelend);
3561 	pmap_bootstrap2();
3562 }
3563