xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: pmap.c,v 1.91 2014/03/03 15:36:36 macallan 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 1.91 2014/03/03 15:36:36 macallan Exp $");
67 
68 #define	PMAP_NOOPNAMES
69 
70 #include "opt_ppcarch.h"
71 #include "opt_altivec.h"
72 #include "opt_multiprocessor.h"
73 #include "opt_pmap.h"
74 
75 #include <sys/param.h>
76 #include <sys/proc.h>
77 #include <sys/pool.h>
78 #include <sys/queue.h>
79 #include <sys/device.h>		/* for evcnt */
80 #include <sys/systm.h>
81 #include <sys/atomic.h>
82 
83 #include <uvm/uvm.h>
84 
85 #include <machine/powerpc.h>
86 #include <powerpc/bat.h>
87 #include <powerpc/pcb.h>
88 #include <powerpc/psl.h>
89 #include <powerpc/spr.h>
90 #include <powerpc/oea/spr.h>
91 #include <powerpc/oea/sr_601.h>
92 
93 #ifdef ALTIVEC
94 extern int pmap_use_altivec;
95 #endif
96 
97 #ifdef PMAP_MEMLIMIT
98 static paddr_t pmap_memlimit = PMAP_MEMLIMIT;
99 #else
100 static paddr_t pmap_memlimit = -PAGE_SIZE;		/* there is no limit */
101 #endif
102 
103 extern struct pmap kernel_pmap_;
104 static unsigned int pmap_pages_stolen;
105 static u_long pmap_pte_valid;
106 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
107 static u_long pmap_pvo_enter_depth;
108 static u_long pmap_pvo_remove_depth;
109 #endif
110 
111 #ifndef MSGBUFADDR
112 extern paddr_t msgbuf_paddr;
113 #endif
114 
115 static struct mem_region *mem, *avail;
116 static u_int mem_cnt, avail_cnt;
117 
118 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
119 # define	PMAP_OEA 1
120 #endif
121 
122 #if defined(PMAP_OEA)
123 #define	_PRIxpte	"lx"
124 #else
125 #define	_PRIxpte	PRIx64
126 #endif
127 #define	_PRIxpa		"lx"
128 #define	_PRIxva		"lx"
129 #define	_PRIsr  	"lx"
130 
131 #ifdef PMAP_NEEDS_FIXUP
132 #if defined(PMAP_OEA)
133 #define	PMAPNAME(name)	pmap32_##name
134 #elif defined(PMAP_OEA64)
135 #define	PMAPNAME(name)	pmap64_##name
136 #elif defined(PMAP_OEA64_BRIDGE)
137 #define	PMAPNAME(name)	pmap64bridge_##name
138 #else
139 #error unknown variant for pmap
140 #endif
141 #endif /* PMAP_NEEDS_FIXUP */
142 
143 #ifdef PMAPNAME
144 #define	STATIC			static
145 #define pmap_pte_spill		PMAPNAME(pte_spill)
146 #define pmap_real_memory	PMAPNAME(real_memory)
147 #define pmap_init		PMAPNAME(init)
148 #define pmap_virtual_space	PMAPNAME(virtual_space)
149 #define pmap_create		PMAPNAME(create)
150 #define pmap_reference		PMAPNAME(reference)
151 #define pmap_destroy		PMAPNAME(destroy)
152 #define pmap_copy		PMAPNAME(copy)
153 #define pmap_update		PMAPNAME(update)
154 #define pmap_enter		PMAPNAME(enter)
155 #define pmap_remove		PMAPNAME(remove)
156 #define pmap_kenter_pa		PMAPNAME(kenter_pa)
157 #define pmap_kremove		PMAPNAME(kremove)
158 #define pmap_extract		PMAPNAME(extract)
159 #define pmap_protect		PMAPNAME(protect)
160 #define pmap_unwire		PMAPNAME(unwire)
161 #define pmap_page_protect	PMAPNAME(page_protect)
162 #define pmap_query_bit		PMAPNAME(query_bit)
163 #define pmap_clear_bit		PMAPNAME(clear_bit)
164 
165 #define pmap_activate		PMAPNAME(activate)
166 #define pmap_deactivate		PMAPNAME(deactivate)
167 
168 #define pmap_pinit		PMAPNAME(pinit)
169 #define pmap_procwr		PMAPNAME(procwr)
170 
171 #define pmap_pool		PMAPNAME(pool)
172 #define pmap_upvo_pool		PMAPNAME(upvo_pool)
173 #define pmap_mpvo_pool		PMAPNAME(mpvo_pool)
174 #define pmap_pvo_table		PMAPNAME(pvo_table)
175 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
176 #define pmap_pte_print		PMAPNAME(pte_print)
177 #define pmap_pteg_check		PMAPNAME(pteg_check)
178 #define pmap_print_mmruregs	PMAPNAME(print_mmuregs)
179 #define pmap_print_pte		PMAPNAME(print_pte)
180 #define pmap_pteg_dist		PMAPNAME(pteg_dist)
181 #endif
182 #if defined(DEBUG) || defined(PMAPCHECK)
183 #define	pmap_pvo_verify		PMAPNAME(pvo_verify)
184 #define pmapcheck		PMAPNAME(check)
185 #endif
186 #if defined(DEBUG) || defined(PMAPDEBUG)
187 #define pmapdebug		PMAPNAME(debug)
188 #endif
189 #define pmap_steal_memory	PMAPNAME(steal_memory)
190 #define pmap_bootstrap		PMAPNAME(bootstrap)
191 #else
192 #define	STATIC			/* nothing */
193 #endif /* PMAPNAME */
194 
195 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
196 STATIC void pmap_real_memory(paddr_t *, psize_t *);
197 STATIC void pmap_init(void);
198 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
199 STATIC pmap_t pmap_create(void);
200 STATIC void pmap_reference(pmap_t);
201 STATIC void pmap_destroy(pmap_t);
202 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
203 STATIC void pmap_update(pmap_t);
204 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
205 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
206 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t, u_int);
207 STATIC void pmap_kremove(vaddr_t, vsize_t);
208 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
209 
210 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
211 STATIC void pmap_unwire(pmap_t, vaddr_t);
212 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
213 STATIC bool pmap_query_bit(struct vm_page *, int);
214 STATIC bool pmap_clear_bit(struct vm_page *, int);
215 
216 STATIC void pmap_activate(struct lwp *);
217 STATIC void pmap_deactivate(struct lwp *);
218 
219 STATIC void pmap_pinit(pmap_t pm);
220 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
221 
222 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
223 STATIC void pmap_pte_print(volatile struct pte *);
224 STATIC void pmap_pteg_check(void);
225 STATIC void pmap_print_mmuregs(void);
226 STATIC void pmap_print_pte(pmap_t, vaddr_t);
227 STATIC void pmap_pteg_dist(void);
228 #endif
229 #if defined(DEBUG) || defined(PMAPCHECK)
230 STATIC void pmap_pvo_verify(void);
231 #endif
232 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
233 STATIC void pmap_bootstrap(paddr_t, paddr_t);
234 
235 #ifdef PMAPNAME
236 const struct pmap_ops PMAPNAME(ops) = {
237 	.pmapop_pte_spill = pmap_pte_spill,
238 	.pmapop_real_memory = pmap_real_memory,
239 	.pmapop_init = pmap_init,
240 	.pmapop_virtual_space = pmap_virtual_space,
241 	.pmapop_create = pmap_create,
242 	.pmapop_reference = pmap_reference,
243 	.pmapop_destroy = pmap_destroy,
244 	.pmapop_copy = pmap_copy,
245 	.pmapop_update = pmap_update,
246 	.pmapop_enter = pmap_enter,
247 	.pmapop_remove = pmap_remove,
248 	.pmapop_kenter_pa = pmap_kenter_pa,
249 	.pmapop_kremove = pmap_kremove,
250 	.pmapop_extract = pmap_extract,
251 	.pmapop_protect = pmap_protect,
252 	.pmapop_unwire = pmap_unwire,
253 	.pmapop_page_protect = pmap_page_protect,
254 	.pmapop_query_bit = pmap_query_bit,
255 	.pmapop_clear_bit = pmap_clear_bit,
256 	.pmapop_activate = pmap_activate,
257 	.pmapop_deactivate = pmap_deactivate,
258 	.pmapop_pinit = pmap_pinit,
259 	.pmapop_procwr = pmap_procwr,
260 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
261 	.pmapop_pte_print = pmap_pte_print,
262 	.pmapop_pteg_check = pmap_pteg_check,
263 	.pmapop_print_mmuregs = pmap_print_mmuregs,
264 	.pmapop_print_pte = pmap_print_pte,
265 	.pmapop_pteg_dist = pmap_pteg_dist,
266 #else
267 	.pmapop_pte_print = NULL,
268 	.pmapop_pteg_check = NULL,
269 	.pmapop_print_mmuregs = NULL,
270 	.pmapop_print_pte = NULL,
271 	.pmapop_pteg_dist = NULL,
272 #endif
273 #if defined(DEBUG) || defined(PMAPCHECK)
274 	.pmapop_pvo_verify = pmap_pvo_verify,
275 #else
276 	.pmapop_pvo_verify = NULL,
277 #endif
278 	.pmapop_steal_memory = pmap_steal_memory,
279 	.pmapop_bootstrap = pmap_bootstrap,
280 };
281 #endif /* !PMAPNAME */
282 
283 /*
284  * The following structure is aligned to 32 bytes
285  */
286 struct pvo_entry {
287 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
288 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
289 	struct pte pvo_pte;			/* Prebuilt PTE */
290 	pmap_t pvo_pmap;			/* ptr to owning pmap */
291 	vaddr_t pvo_vaddr;			/* VA of entry */
292 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
293 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
294 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
295 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
296 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
297 #define	PVO_WIRED_P(pvo)	((pvo)->pvo_vaddr & PVO_WIRED)
298 #define	PVO_MANAGED_P(pvo)	((pvo)->pvo_vaddr & PVO_MANAGED)
299 #define	PVO_EXECUTABLE_P(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
300 #define	PVO_ENTER_INSERT	0		/* PVO has been removed */
301 #define	PVO_SPILL_UNSET		1		/* PVO has been evicted */
302 #define	PVO_SPILL_SET		2		/* PVO has been spilled */
303 #define	PVO_SPILL_INSERT	3		/* PVO has been inserted */
304 #define	PVO_PMAP_PAGE_PROTECT	4		/* PVO has changed */
305 #define	PVO_PMAP_PROTECT	5		/* PVO has changed */
306 #define	PVO_REMOVE		6		/* PVO has been removed */
307 #define	PVO_WHERE_MASK		15
308 #define	PVO_WHERE_SHFT		8
309 } __attribute__ ((aligned (32)));
310 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
311 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
312 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
313 #define	PVO_PTEGIDX_CLR(pvo)	\
314 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
315 #define	PVO_PTEGIDX_SET(pvo,i)	\
316 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
317 #define	PVO_WHERE(pvo,w)	\
318 	((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
319 	 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
320 
321 TAILQ_HEAD(pvo_tqhead, pvo_entry);
322 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
323 static struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
324 static struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
325 
326 struct pool pmap_pool;		/* pool for pmap structures */
327 struct pool pmap_upvo_pool;	/* pool for pvo entries for unmanaged pages */
328 struct pool pmap_mpvo_pool;	/* pool for pvo entries for managed pages */
329 
330 /*
331  * We keep a cache of unmanaged pages to be used for pvo entries for
332  * unmanaged pages.
333  */
334 struct pvo_page {
335 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
336 };
337 SIMPLEQ_HEAD(pvop_head, pvo_page);
338 static struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head);
339 static struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head);
340 static u_long pmap_upvop_free;
341 static u_long pmap_upvop_maxfree;
342 static u_long pmap_mpvop_free;
343 static u_long pmap_mpvop_maxfree;
344 
345 static void *pmap_pool_ualloc(struct pool *, int);
346 static void *pmap_pool_malloc(struct pool *, int);
347 
348 static void pmap_pool_ufree(struct pool *, void *);
349 static void pmap_pool_mfree(struct pool *, void *);
350 
351 static struct pool_allocator pmap_pool_mallocator = {
352 	.pa_alloc = pmap_pool_malloc,
353 	.pa_free = pmap_pool_mfree,
354 	.pa_pagesz = 0,
355 };
356 
357 static struct pool_allocator pmap_pool_uallocator = {
358 	.pa_alloc = pmap_pool_ualloc,
359 	.pa_free = pmap_pool_ufree,
360 	.pa_pagesz = 0,
361 };
362 
363 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
364 void pmap_pte_print(volatile struct pte *);
365 void pmap_pteg_check(void);
366 void pmap_pteg_dist(void);
367 void pmap_print_pte(pmap_t, vaddr_t);
368 void pmap_print_mmuregs(void);
369 #endif
370 
371 #if defined(DEBUG) || defined(PMAPCHECK)
372 #ifdef PMAPCHECK
373 int pmapcheck = 1;
374 #else
375 int pmapcheck = 0;
376 #endif
377 void pmap_pvo_verify(void);
378 static void pmap_pvo_check(const struct pvo_entry *);
379 #define	PMAP_PVO_CHECK(pvo)	 		\
380 	do {					\
381 		if (pmapcheck)			\
382 			pmap_pvo_check(pvo);	\
383 	} while (0)
384 #else
385 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
386 #endif
387 static int pmap_pte_insert(int, struct pte *);
388 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
389 	vaddr_t, paddr_t, register_t, int);
390 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
391 static void pmap_pvo_free(struct pvo_entry *);
392 static void pmap_pvo_free_list(struct pvo_head *);
393 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
394 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
395 static struct pvo_entry *pmap_pvo_reclaim(struct pmap *);
396 static void pvo_set_exec(struct pvo_entry *);
397 static void pvo_clear_exec(struct pvo_entry *);
398 
399 static void tlbia(void);
400 
401 static void pmap_release(pmap_t);
402 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
403 
404 static uint32_t pmap_pvo_reclaim_nextidx;
405 #ifdef DEBUG
406 static int pmap_pvo_reclaim_debugctr;
407 #endif
408 
409 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
410 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
411 
412 static int pmap_initialized;
413 
414 #if defined(DEBUG) || defined(PMAPDEBUG)
415 #define	PMAPDEBUG_BOOT		0x0001
416 #define	PMAPDEBUG_PTE		0x0002
417 #define	PMAPDEBUG_EXEC		0x0008
418 #define	PMAPDEBUG_PVOENTER	0x0010
419 #define	PMAPDEBUG_PVOREMOVE	0x0020
420 #define	PMAPDEBUG_ACTIVATE	0x0100
421 #define	PMAPDEBUG_CREATE	0x0200
422 #define	PMAPDEBUG_ENTER		0x1000
423 #define	PMAPDEBUG_KENTER	0x2000
424 #define	PMAPDEBUG_KREMOVE	0x4000
425 #define	PMAPDEBUG_REMOVE	0x8000
426 
427 unsigned int pmapdebug = 0;
428 
429 # define DPRINTF(x, ...)	printf(x, __VA_ARGS__)
430 # define DPRINTFN(n, x, ...)	do if (pmapdebug & PMAPDEBUG_ ## n) printf(x, __VA_ARGS__); while (0)
431 #else
432 # define DPRINTF(x, ...)	do { } while (0)
433 # define DPRINTFN(n, x, ...)	do { } while (0)
434 #endif
435 
436 
437 #ifdef PMAPCOUNTERS
438 /*
439  * From pmap_subr.c
440  */
441 extern struct evcnt pmap_evcnt_mappings;
442 extern struct evcnt pmap_evcnt_unmappings;
443 
444 extern struct evcnt pmap_evcnt_kernel_mappings;
445 extern struct evcnt pmap_evcnt_kernel_unmappings;
446 
447 extern struct evcnt pmap_evcnt_mappings_replaced;
448 
449 extern struct evcnt pmap_evcnt_exec_mappings;
450 extern struct evcnt pmap_evcnt_exec_cached;
451 
452 extern struct evcnt pmap_evcnt_exec_synced;
453 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
454 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
455 
456 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
457 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
458 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
459 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
460 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
461 
462 extern struct evcnt pmap_evcnt_updates;
463 extern struct evcnt pmap_evcnt_collects;
464 extern struct evcnt pmap_evcnt_copies;
465 
466 extern struct evcnt pmap_evcnt_ptes_spilled;
467 extern struct evcnt pmap_evcnt_ptes_unspilled;
468 extern struct evcnt pmap_evcnt_ptes_evicted;
469 
470 extern struct evcnt pmap_evcnt_ptes_primary[8];
471 extern struct evcnt pmap_evcnt_ptes_secondary[8];
472 extern struct evcnt pmap_evcnt_ptes_removed;
473 extern struct evcnt pmap_evcnt_ptes_changed;
474 extern struct evcnt pmap_evcnt_pvos_reclaimed;
475 extern struct evcnt pmap_evcnt_pvos_failed;
476 
477 extern struct evcnt pmap_evcnt_zeroed_pages;
478 extern struct evcnt pmap_evcnt_copied_pages;
479 extern struct evcnt pmap_evcnt_idlezeroed_pages;
480 
481 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
482 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
483 #else
484 #define	PMAPCOUNT(ev)	((void) 0)
485 #define	PMAPCOUNT2(ev)	((void) 0)
486 #endif
487 
488 #define	TLBIE(va)	__asm volatile("tlbie %0" :: "r"(va))
489 
490 /* XXXSL: this needs to be moved to assembler */
491 #define	TLBIEL(va)	__asm __volatile("tlbie %0" :: "r"(va))
492 
493 #ifdef MD_TLBSYNC
494 #define TLBSYNC()	MD_TLBSYNC()
495 #else
496 #define	TLBSYNC()	__asm volatile("tlbsync")
497 #endif
498 #define	SYNC()		__asm volatile("sync")
499 #define	EIEIO()		__asm volatile("eieio")
500 #define	DCBST(va)	__asm __volatile("dcbst 0,%0" :: "r"(va))
501 #define	MFMSR()		mfmsr()
502 #define	MTMSR(psl)	mtmsr(psl)
503 #define	MFPVR()		mfpvr()
504 #define	MFSRIN(va)	mfsrin(va)
505 #define	MFTB()		mfrtcltbl()
506 
507 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
508 static inline register_t
509 mfsrin(vaddr_t va)
510 {
511 	register_t sr;
512 	__asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
513 	return sr;
514 }
515 #endif	/* PMAP_OEA*/
516 
517 #if defined (PMAP_OEA64_BRIDGE)
518 extern void mfmsr64 (register64_t *result);
519 #endif /* PMAP_OEA64_BRIDGE */
520 
521 #define	PMAP_LOCK()		KERNEL_LOCK(1, NULL)
522 #define	PMAP_UNLOCK()		KERNEL_UNLOCK_ONE(NULL)
523 
524 static inline register_t
525 pmap_interrupts_off(void)
526 {
527 	register_t msr = MFMSR();
528 	if (msr & PSL_EE)
529 		MTMSR(msr & ~PSL_EE);
530 	return msr;
531 }
532 
533 static void
534 pmap_interrupts_restore(register_t msr)
535 {
536 	if (msr & PSL_EE)
537 		MTMSR(msr);
538 }
539 
540 static inline u_int32_t
541 mfrtcltbl(void)
542 {
543 #ifdef PPC_OEA601
544 	if ((MFPVR() >> 16) == MPC601)
545 		return (mfrtcl() >> 7);
546 	else
547 #endif
548 		return (mftbl());
549 }
550 
551 /*
552  * These small routines may have to be replaced,
553  * if/when we support processors other that the 604.
554  */
555 
556 void
557 tlbia(void)
558 {
559 	char *i;
560 
561 	SYNC();
562 #if defined(PMAP_OEA)
563 	/*
564 	 * Why not use "tlbia"?  Because not all processors implement it.
565 	 *
566 	 * This needs to be a per-CPU callback to do the appropriate thing
567 	 * for the CPU. XXX
568 	 */
569 	for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
570 		TLBIE(i);
571 		EIEIO();
572 		SYNC();
573 	}
574 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
575 	/* This is specifically for the 970, 970UM v1.6 pp. 140. */
576 	for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
577 		TLBIEL(i);
578 		EIEIO();
579 		SYNC();
580 	}
581 #endif
582 	TLBSYNC();
583 	SYNC();
584 }
585 
586 static inline register_t
587 va_to_vsid(const struct pmap *pm, vaddr_t addr)
588 {
589 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
590 	return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT;
591 #else /* PMAP_OEA64 */
592 #if 0
593 	const struct ste *ste;
594 	register_t hash;
595 	int i;
596 
597 	hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH;
598 
599 	/*
600 	 * Try the primary group first
601 	 */
602 	ste = pm->pm_stes[hash].stes;
603 	for (i = 0; i < 8; i++, ste++) {
604 		if (ste->ste_hi & STE_V) &&
605 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
606 			return ste;
607 	}
608 
609 	/*
610 	 * Then the secondary group.
611 	 */
612 	ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes;
613 	for (i = 0; i < 8; i++, ste++) {
614 		if (ste->ste_hi & STE_V) &&
615 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
616 			return addr;
617 	}
618 
619 	return NULL;
620 #else
621 	/*
622 	 * Rather than searching the STE groups for the VSID, we know
623 	 * how we generate that from the ESID and so do that.
624 	 */
625 	return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
626 #endif
627 #endif /* PMAP_OEA */
628 }
629 
630 static inline register_t
631 va_to_pteg(const struct pmap *pm, vaddr_t addr)
632 {
633 	register_t hash;
634 
635 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
636 	return hash & pmap_pteg_mask;
637 }
638 
639 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
640 /*
641  * Given a PTE in the page table, calculate the VADDR that hashes to it.
642  * The only bit of magic is that the top 4 bits of the address doesn't
643  * technically exist in the PTE.  But we know we reserved 4 bits of the
644  * VSID for it so that's how we get it.
645  */
646 static vaddr_t
647 pmap_pte_to_va(volatile const struct pte *pt)
648 {
649 	vaddr_t va;
650 	uintptr_t ptaddr = (uintptr_t) pt;
651 
652 	if (pt->pte_hi & PTE_HID)
653 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
654 
655 	/* PPC Bits 10-19  PPC64 Bits 42-51 */
656 #if defined(PMAP_OEA)
657 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
658 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
659 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
660 #endif
661 	va <<= ADDR_PIDX_SHFT;
662 
663 	/* PPC Bits 4-9  PPC64 Bits 36-41 */
664 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
665 
666 #if defined(PMAP_OEA64)
667 	/* PPC63 Bits 0-35 */
668 	/* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
669 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
670 	/* PPC Bits 0-3 */
671 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
672 #endif
673 
674 	return va;
675 }
676 #endif
677 
678 static inline struct pvo_head *
679 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
680 {
681 	struct vm_page *pg;
682 	struct vm_page_md *md;
683 
684 	pg = PHYS_TO_VM_PAGE(pa);
685 	if (pg_p != NULL)
686 		*pg_p = pg;
687 	if (pg == NULL)
688 		return &pmap_pvo_unmanaged;
689 	md = VM_PAGE_TO_MD(pg);
690 	return &md->mdpg_pvoh;
691 }
692 
693 static inline struct pvo_head *
694 vm_page_to_pvoh(struct vm_page *pg)
695 {
696 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
697 
698 	return &md->mdpg_pvoh;
699 }
700 
701 
702 static inline void
703 pmap_attr_clear(struct vm_page *pg, int ptebit)
704 {
705 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
706 
707 	md->mdpg_attrs &= ~ptebit;
708 }
709 
710 static inline int
711 pmap_attr_fetch(struct vm_page *pg)
712 {
713 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
714 
715 	return md->mdpg_attrs;
716 }
717 
718 static inline void
719 pmap_attr_save(struct vm_page *pg, int ptebit)
720 {
721 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
722 
723 	md->mdpg_attrs |= ptebit;
724 }
725 
726 static inline int
727 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
728 {
729 	if (pt->pte_hi == pvo_pt->pte_hi
730 #if 0
731 	    && ((pt->pte_lo ^ pvo_pt->pte_lo) &
732 	        ~(PTE_REF|PTE_CHG)) == 0
733 #endif
734 	    )
735 		return 1;
736 	return 0;
737 }
738 
739 static inline void
740 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
741 {
742 	/*
743 	 * Construct the PTE.  Default to IMB initially.  Valid bit
744 	 * only gets set when the real pte is set in memory.
745 	 *
746 	 * Note: Don't set the valid bit for correct operation of tlb update.
747 	 */
748 #if defined(PMAP_OEA)
749 	pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
750 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
751 	pt->pte_lo = pte_lo;
752 #elif defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA64)
753 	pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT)
754 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
755 	pt->pte_lo = (u_int64_t) pte_lo;
756 #endif /* PMAP_OEA */
757 }
758 
759 static inline void
760 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
761 {
762 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
763 }
764 
765 static inline void
766 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
767 {
768 	/*
769 	 * As shown in Section 7.6.3.2.3
770 	 */
771 	pt->pte_lo &= ~ptebit;
772 	TLBIE(va);
773 	SYNC();
774 	EIEIO();
775 	TLBSYNC();
776 	SYNC();
777 #ifdef MULTIPROCESSOR
778 	DCBST(pt);
779 #endif
780 }
781 
782 static inline void
783 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
784 {
785 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
786 	if (pvo_pt->pte_hi & PTE_VALID)
787 		panic("pte_set: setting an already valid pte %p", pvo_pt);
788 #endif
789 	pvo_pt->pte_hi |= PTE_VALID;
790 
791 	/*
792 	 * Update the PTE as defined in section 7.6.3.1
793 	 * Note that the REF/CHG bits are from pvo_pt and thus should
794 	 * have been saved so this routine can restore them (if desired).
795 	 */
796 	pt->pte_lo = pvo_pt->pte_lo;
797 	EIEIO();
798 	pt->pte_hi = pvo_pt->pte_hi;
799 	TLBSYNC();
800 	SYNC();
801 #ifdef MULTIPROCESSOR
802 	DCBST(pt);
803 #endif
804 	pmap_pte_valid++;
805 }
806 
807 static inline void
808 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
809 {
810 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
811 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
812 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
813 	if ((pt->pte_hi & PTE_VALID) == 0)
814 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
815 #endif
816 
817 	pvo_pt->pte_hi &= ~PTE_VALID;
818 	/*
819 	 * Force the ref & chg bits back into the PTEs.
820 	 */
821 	SYNC();
822 	/*
823 	 * Invalidate the pte ... (Section 7.6.3.3)
824 	 */
825 	pt->pte_hi &= ~PTE_VALID;
826 	SYNC();
827 	TLBIE(va);
828 	SYNC();
829 	EIEIO();
830 	TLBSYNC();
831 	SYNC();
832 	/*
833 	 * Save the ref & chg bits ...
834 	 */
835 	pmap_pte_synch(pt, pvo_pt);
836 	pmap_pte_valid--;
837 }
838 
839 static inline void
840 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
841 {
842 	/*
843 	 * Invalidate the PTE
844 	 */
845 	pmap_pte_unset(pt, pvo_pt, va);
846 	pmap_pte_set(pt, pvo_pt);
847 }
848 
849 /*
850  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
851  * (either primary or secondary location).
852  *
853  * Note: both the destination and source PTEs must not have PTE_VALID set.
854  */
855 
856 static int
857 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
858 {
859 	volatile struct pte *pt;
860 	int i;
861 
862 #if defined(DEBUG)
863 	DPRINTFN(PTE, "pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
864 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo);
865 #endif
866 	/*
867 	 * First try primary hash.
868 	 */
869 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
870 		if ((pt->pte_hi & PTE_VALID) == 0) {
871 			pvo_pt->pte_hi &= ~PTE_HID;
872 			pmap_pte_set(pt, pvo_pt);
873 			return i;
874 		}
875 	}
876 
877 	/*
878 	 * Now try secondary hash.
879 	 */
880 	ptegidx ^= pmap_pteg_mask;
881 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
882 		if ((pt->pte_hi & PTE_VALID) == 0) {
883 			pvo_pt->pte_hi |= PTE_HID;
884 			pmap_pte_set(pt, pvo_pt);
885 			return i;
886 		}
887 	}
888 	return -1;
889 }
890 
891 /*
892  * Spill handler.
893  *
894  * Tries to spill a page table entry from the overflow area.
895  * This runs in either real mode (if dealing with a exception spill)
896  * or virtual mode when dealing with manually spilling one of the
897  * kernel's pte entries.  In either case, interrupts are already
898  * disabled.
899  */
900 
901 int
902 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec)
903 {
904 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
905 	struct pvo_entry *pvo;
906 	/* XXX: gcc -- vpvoh is always set at either *1* or *2* */
907 	struct pvo_tqhead *pvoh, *vpvoh = NULL;
908 	int ptegidx, i, j;
909 	volatile struct pteg *pteg;
910 	volatile struct pte *pt;
911 
912 	PMAP_LOCK();
913 
914 	ptegidx = va_to_pteg(pm, addr);
915 
916 	/*
917 	 * Have to substitute some entry. Use the primary hash for this.
918 	 * Use low bits of timebase as random generator.  Make sure we are
919 	 * not picking a kernel pte for replacement.
920 	 */
921 	pteg = &pmap_pteg_table[ptegidx];
922 	i = MFTB() & 7;
923 	for (j = 0; j < 8; j++) {
924 		pt = &pteg->pt[i];
925 		if ((pt->pte_hi & PTE_VALID) == 0)
926 			break;
927 		if (VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT)
928 				< PHYSMAP_VSIDBITS)
929 			break;
930 		i = (i + 1) & 7;
931 	}
932 	KASSERT(j < 8);
933 
934 	source_pvo = NULL;
935 	victim_pvo = NULL;
936 	pvoh = &pmap_pvo_table[ptegidx];
937 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
938 
939 		/*
940 		 * We need to find pvo entry for this address...
941 		 */
942 		PMAP_PVO_CHECK(pvo);		/* sanity check */
943 
944 		/*
945 		 * If we haven't found the source and we come to a PVO with
946 		 * a valid PTE, then we know we can't find it because all
947 		 * evicted PVOs always are first in the list.
948 		 */
949 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
950 			break;
951 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
952 		    addr == PVO_VADDR(pvo)) {
953 
954 			/*
955 			 * Now we have found the entry to be spilled into the
956 			 * pteg.  Attempt to insert it into the page table.
957 			 */
958 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
959 			if (j >= 0) {
960 				PVO_PTEGIDX_SET(pvo, j);
961 				PMAP_PVO_CHECK(pvo);	/* sanity check */
962 				PVO_WHERE(pvo, SPILL_INSERT);
963 				pvo->pvo_pmap->pm_evictions--;
964 				PMAPCOUNT(ptes_spilled);
965 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
966 				    ? pmap_evcnt_ptes_secondary
967 				    : pmap_evcnt_ptes_primary)[j]);
968 
969 				/*
970 				 * Since we keep the evicted entries at the
971 				 * from of the PVO list, we need move this
972 				 * (now resident) PVO after the evicted
973 				 * entries.
974 				 */
975 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
976 
977 				/*
978 				 * If we don't have to move (either we were the
979 				 * last entry or the next entry was valid),
980 				 * don't change our position.  Otherwise
981 				 * move ourselves to the tail of the queue.
982 				 */
983 				if (next_pvo != NULL &&
984 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
985 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
986 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
987 				}
988 				PMAP_UNLOCK();
989 				return 1;
990 			}
991 			source_pvo = pvo;
992 			if (exec && !PVO_EXECUTABLE_P(source_pvo)) {
993 				return 0;
994 			}
995 			if (victim_pvo != NULL)
996 				break;
997 		}
998 
999 		/*
1000 		 * We also need the pvo entry of the victim we are replacing
1001 		 * so save the R & C bits of the PTE.
1002 		 */
1003 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
1004 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
1005 			vpvoh = pvoh;			/* *1* */
1006 			victim_pvo = pvo;
1007 			if (source_pvo != NULL)
1008 				break;
1009 		}
1010 	}
1011 
1012 	if (source_pvo == NULL) {
1013 		PMAPCOUNT(ptes_unspilled);
1014 		PMAP_UNLOCK();
1015 		return 0;
1016 	}
1017 
1018 	if (victim_pvo == NULL) {
1019 		if ((pt->pte_hi & PTE_HID) == 0)
1020 			panic("pmap_pte_spill: victim p-pte (%p) has "
1021 			    "no pvo entry!", pt);
1022 
1023 		/*
1024 		 * If this is a secondary PTE, we need to search
1025 		 * its primary pvo bucket for the matching PVO.
1026 		 */
1027 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */
1028 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1029 			PMAP_PVO_CHECK(pvo);		/* sanity check */
1030 
1031 			/*
1032 			 * We also need the pvo entry of the victim we are
1033 			 * replacing so save the R & C bits of the PTE.
1034 			 */
1035 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1036 				victim_pvo = pvo;
1037 				break;
1038 			}
1039 		}
1040 		if (victim_pvo == NULL)
1041 			panic("pmap_pte_spill: victim s-pte (%p) has "
1042 			    "no pvo entry!", pt);
1043 	}
1044 
1045 	/*
1046 	 * The victim should be not be a kernel PVO/PTE entry.
1047 	 */
1048 	KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1049 	KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1050 	KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1051 
1052 	/*
1053 	 * We are invalidating the TLB entry for the EA for the
1054 	 * we are replacing even though its valid; If we don't
1055 	 * we lose any ref/chg bit changes contained in the TLB
1056 	 * entry.
1057 	 */
1058 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1059 
1060 	/*
1061 	 * To enforce the PVO list ordering constraint that all
1062 	 * evicted entries should come before all valid entries,
1063 	 * move the source PVO to the tail of its list and the
1064 	 * victim PVO to the head of its list (which might not be
1065 	 * the same list, if the victim was using the secondary hash).
1066 	 */
1067 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
1068 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
1069 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1070 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1071 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1072 	pmap_pte_set(pt, &source_pvo->pvo_pte);
1073 	victim_pvo->pvo_pmap->pm_evictions++;
1074 	source_pvo->pvo_pmap->pm_evictions--;
1075 	PVO_WHERE(victim_pvo, SPILL_UNSET);
1076 	PVO_WHERE(source_pvo, SPILL_SET);
1077 
1078 	PVO_PTEGIDX_CLR(victim_pvo);
1079 	PVO_PTEGIDX_SET(source_pvo, i);
1080 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1081 	PMAPCOUNT(ptes_spilled);
1082 	PMAPCOUNT(ptes_evicted);
1083 	PMAPCOUNT(ptes_removed);
1084 
1085 	PMAP_PVO_CHECK(victim_pvo);
1086 	PMAP_PVO_CHECK(source_pvo);
1087 
1088 	PMAP_UNLOCK();
1089 	return 1;
1090 }
1091 
1092 /*
1093  * Restrict given range to physical memory
1094  */
1095 void
1096 pmap_real_memory(paddr_t *start, psize_t *size)
1097 {
1098 	struct mem_region *mp;
1099 
1100 	for (mp = mem; mp->size; mp++) {
1101 		if (*start + *size > mp->start
1102 		    && *start < mp->start + mp->size) {
1103 			if (*start < mp->start) {
1104 				*size -= mp->start - *start;
1105 				*start = mp->start;
1106 			}
1107 			if (*start + *size > mp->start + mp->size)
1108 				*size = mp->start + mp->size - *start;
1109 			return;
1110 		}
1111 	}
1112 	*size = 0;
1113 }
1114 
1115 /*
1116  * Initialize anything else for pmap handling.
1117  * Called during vm_init().
1118  */
1119 void
1120 pmap_init(void)
1121 {
1122 	pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry),
1123 	    sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl",
1124 	    &pmap_pool_mallocator, IPL_NONE);
1125 
1126 	pool_setlowat(&pmap_mpvo_pool, 1008);
1127 
1128 	pmap_initialized = 1;
1129 
1130 }
1131 
1132 /*
1133  * How much virtual space does the kernel get?
1134  */
1135 void
1136 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1137 {
1138 	/*
1139 	 * For now, reserve one segment (minus some overhead) for kernel
1140 	 * virtual memory
1141 	 */
1142 	*start = VM_MIN_KERNEL_ADDRESS;
1143 	*end = VM_MAX_KERNEL_ADDRESS;
1144 }
1145 
1146 /*
1147  * Allocate, initialize, and return a new physical map.
1148  */
1149 pmap_t
1150 pmap_create(void)
1151 {
1152 	pmap_t pm;
1153 
1154 	pm = pool_get(&pmap_pool, PR_WAITOK);
1155 	KASSERT((vaddr_t)pm < VM_MIN_KERNEL_ADDRESS);
1156 	memset((void *)pm, 0, sizeof *pm);
1157 	pmap_pinit(pm);
1158 
1159 	DPRINTFN(CREATE, "pmap_create: pm %p:\n"
1160 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1161 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1162 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1163 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1164 	    pm,
1165 	    pm->pm_sr[0], pm->pm_sr[1],
1166 	    pm->pm_sr[2], pm->pm_sr[3],
1167 	    pm->pm_sr[4], pm->pm_sr[5],
1168 	    pm->pm_sr[6], pm->pm_sr[7],
1169 	    pm->pm_sr[8], pm->pm_sr[9],
1170 	    pm->pm_sr[10], pm->pm_sr[11],
1171 	    pm->pm_sr[12], pm->pm_sr[13],
1172 	    pm->pm_sr[14], pm->pm_sr[15]);
1173 	return pm;
1174 }
1175 
1176 /*
1177  * Initialize a preallocated and zeroed pmap structure.
1178  */
1179 void
1180 pmap_pinit(pmap_t pm)
1181 {
1182 	register_t entropy = MFTB();
1183 	register_t mask;
1184 	int i;
1185 
1186 	/*
1187 	 * Allocate some segment registers for this pmap.
1188 	 */
1189 	pm->pm_refs = 1;
1190 	PMAP_LOCK();
1191 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1192 		static register_t pmap_vsidcontext;
1193 		register_t hash;
1194 		unsigned int n;
1195 
1196 		/* Create a new value by multiplying by a prime adding in
1197 		 * entropy from the timebase register.  This is to make the
1198 		 * VSID more random so that the PT Hash function collides
1199 		 * less often. (note that the prime causes gcc to do shifts
1200 		 * instead of a multiply)
1201 		 */
1202 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1203 		hash = pmap_vsidcontext & (NPMAPS - 1);
1204 		if (hash == 0) {		/* 0 is special, avoid it */
1205 			entropy += 0xbadf00d;
1206 			continue;
1207 		}
1208 		n = hash >> 5;
1209 		mask = 1L << (hash & (VSID_NBPW-1));
1210 		hash = pmap_vsidcontext;
1211 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1212 			/* anything free in this bucket? */
1213 			if (~pmap_vsid_bitmap[n] == 0) {
1214 				entropy = hash ^ (hash >> 16);
1215 				continue;
1216 			}
1217 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1218 			mask = 1L << i;
1219 			hash &= ~(VSID_NBPW-1);
1220 			hash |= i;
1221 		}
1222 		hash &= PTE_VSID >> PTE_VSID_SHFT;
1223 		pmap_vsid_bitmap[n] |= mask;
1224 		pm->pm_vsid = hash;
1225 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1226 		for (i = 0; i < 16; i++)
1227 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1228 			    SR_NOEXEC;
1229 #endif
1230 		PMAP_UNLOCK();
1231 		return;
1232 	}
1233 	PMAP_UNLOCK();
1234 	panic("pmap_pinit: out of segments");
1235 }
1236 
1237 /*
1238  * Add a reference to the given pmap.
1239  */
1240 void
1241 pmap_reference(pmap_t pm)
1242 {
1243 	atomic_inc_uint(&pm->pm_refs);
1244 }
1245 
1246 /*
1247  * Retire the given pmap from service.
1248  * Should only be called if the map contains no valid mappings.
1249  */
1250 void
1251 pmap_destroy(pmap_t pm)
1252 {
1253 	if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1254 		pmap_release(pm);
1255 		pool_put(&pmap_pool, pm);
1256 	}
1257 }
1258 
1259 /*
1260  * Release any resources held by the given physical map.
1261  * Called when a pmap initialized by pmap_pinit is being released.
1262  */
1263 void
1264 pmap_release(pmap_t pm)
1265 {
1266 	int idx, mask;
1267 
1268 	KASSERT(pm->pm_stats.resident_count == 0);
1269 	KASSERT(pm->pm_stats.wired_count == 0);
1270 
1271 	PMAP_LOCK();
1272 	if (pm->pm_sr[0] == 0)
1273 		panic("pmap_release");
1274 	idx = pm->pm_vsid & (NPMAPS-1);
1275 	mask = 1 << (idx % VSID_NBPW);
1276 	idx /= VSID_NBPW;
1277 
1278 	KASSERT(pmap_vsid_bitmap[idx] & mask);
1279 	pmap_vsid_bitmap[idx] &= ~mask;
1280 	PMAP_UNLOCK();
1281 }
1282 
1283 /*
1284  * Copy the range specified by src_addr/len
1285  * from the source map to the range dst_addr/len
1286  * in the destination map.
1287  *
1288  * This routine is only advisory and need not do anything.
1289  */
1290 void
1291 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1292 	vsize_t len, vaddr_t src_addr)
1293 {
1294 	PMAPCOUNT(copies);
1295 }
1296 
1297 /*
1298  * Require that all active physical maps contain no
1299  * incorrect entries NOW.
1300  */
1301 void
1302 pmap_update(struct pmap *pmap)
1303 {
1304 	PMAPCOUNT(updates);
1305 	TLBSYNC();
1306 }
1307 
1308 static inline int
1309 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1310 {
1311 	int pteidx;
1312 	/*
1313 	 * We can find the actual pte entry without searching by
1314 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1315 	 * and by noticing the HID bit.
1316 	 */
1317 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1318 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1319 		pteidx ^= pmap_pteg_mask * 8;
1320 	return pteidx;
1321 }
1322 
1323 volatile struct pte *
1324 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1325 {
1326 	volatile struct pte *pt;
1327 
1328 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1329 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1330 		return NULL;
1331 #endif
1332 
1333 	/*
1334 	 * If we haven't been supplied the ptegidx, calculate it.
1335 	 */
1336 	if (pteidx == -1) {
1337 		int ptegidx;
1338 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1339 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1340 	}
1341 
1342 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1343 
1344 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1345 	return pt;
1346 #else
1347 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1348 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1349 		    "pvo but no valid pte index", pvo);
1350 	}
1351 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1352 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1353 		    "pvo but no valid pte", pvo);
1354 	}
1355 
1356 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1357 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1358 #if defined(DEBUG) || defined(PMAPCHECK)
1359 			pmap_pte_print(pt);
1360 #endif
1361 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1362 			    "pmap_pteg_table %p but invalid in pvo",
1363 			    pvo, pt);
1364 		}
1365 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1366 #if defined(DEBUG) || defined(PMAPCHECK)
1367 			pmap_pte_print(pt);
1368 #endif
1369 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1370 			    "not match pte %p in pmap_pteg_table",
1371 			    pvo, pt);
1372 		}
1373 		return pt;
1374 	}
1375 
1376 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1377 #if defined(DEBUG) || defined(PMAPCHECK)
1378 		pmap_pte_print(pt);
1379 #endif
1380 		panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1381 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1382 	}
1383 	return NULL;
1384 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1385 }
1386 
1387 struct pvo_entry *
1388 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1389 {
1390 	struct pvo_entry *pvo;
1391 	int ptegidx;
1392 
1393 	va &= ~ADDR_POFF;
1394 	ptegidx = va_to_pteg(pm, va);
1395 
1396 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1397 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1398 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1399 			panic("pmap_pvo_find_va: invalid pvo %p on "
1400 			    "list %#x (%p)", pvo, ptegidx,
1401 			     &pmap_pvo_table[ptegidx]);
1402 #endif
1403 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1404 			if (pteidx_p)
1405 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1406 			return pvo;
1407 		}
1408 	}
1409 	if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH))
1410 		panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1411 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1412 	return NULL;
1413 }
1414 
1415 #if defined(DEBUG) || defined(PMAPCHECK)
1416 void
1417 pmap_pvo_check(const struct pvo_entry *pvo)
1418 {
1419 	struct pvo_head *pvo_head;
1420 	struct pvo_entry *pvo0;
1421 	volatile struct pte *pt;
1422 	int failed = 0;
1423 
1424 	PMAP_LOCK();
1425 
1426 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1427 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1428 
1429 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1430 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1431 		    pvo, pvo->pvo_pmap);
1432 		failed = 1;
1433 	}
1434 
1435 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1436 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1437 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1438 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1439 		failed = 1;
1440 	}
1441 
1442 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1443 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1444 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1445 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1446 		failed = 1;
1447 	}
1448 
1449 	if (PVO_MANAGED_P(pvo)) {
1450 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1451 	} else {
1452 		if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) {
1453 			printf("pmap_pvo_check: pvo %p: non kernel address "
1454 			    "on kernel unmanaged list\n", pvo);
1455 			failed = 1;
1456 		}
1457 		pvo_head = &pmap_pvo_kunmanaged;
1458 	}
1459 	LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1460 		if (pvo0 == pvo)
1461 			break;
1462 	}
1463 	if (pvo0 == NULL) {
1464 		printf("pmap_pvo_check: pvo %p: not present "
1465 		    "on its vlist head %p\n", pvo, pvo_head);
1466 		failed = 1;
1467 	}
1468 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1469 		printf("pmap_pvo_check: pvo %p: not present "
1470 		    "on its olist head\n", pvo);
1471 		failed = 1;
1472 	}
1473 	pt = pmap_pvo_to_pte(pvo, -1);
1474 	if (pt == NULL) {
1475 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1476 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1477 			    "no PTE\n", pvo);
1478 			failed = 1;
1479 		}
1480 	} else {
1481 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1482 		    (uintptr_t) pt >=
1483 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1484 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1485 			    "pteg table\n", pvo, pt);
1486 			failed = 1;
1487 		}
1488 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1489 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1490 			    "no PTE\n", pvo);
1491 			failed = 1;
1492 		}
1493 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1494 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1495 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1496 			    pvo->pvo_pte.pte_hi,
1497 			    pt->pte_hi);
1498 			failed = 1;
1499 		}
1500 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1501 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1502 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1503 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1504 			    (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1505 			    (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1506 			failed = 1;
1507 		}
1508 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1509 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1510 			    " doesn't not match PVO's VA %#" _PRIxva "\n",
1511 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1512 			failed = 1;
1513 		}
1514 		if (failed)
1515 			pmap_pte_print(pt);
1516 	}
1517 	if (failed)
1518 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1519 		    pvo->pvo_pmap);
1520 
1521 	PMAP_UNLOCK();
1522 }
1523 #endif /* DEBUG || PMAPCHECK */
1524 
1525 /*
1526  * Search the PVO table looking for a non-wired entry.
1527  * If we find one, remove it and return it.
1528  */
1529 
1530 struct pvo_entry *
1531 pmap_pvo_reclaim(struct pmap *pm)
1532 {
1533 	struct pvo_tqhead *pvoh;
1534 	struct pvo_entry *pvo;
1535 	uint32_t idx, endidx;
1536 
1537 	endidx = pmap_pvo_reclaim_nextidx;
1538 	for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1539 	     idx = (idx + 1) & pmap_pteg_mask) {
1540 		pvoh = &pmap_pvo_table[idx];
1541 		TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1542 			if (!PVO_WIRED_P(pvo)) {
1543 				pmap_pvo_remove(pvo, -1, NULL);
1544 				pmap_pvo_reclaim_nextidx = idx;
1545 				PMAPCOUNT(pvos_reclaimed);
1546 				return pvo;
1547 			}
1548 		}
1549 	}
1550 	return NULL;
1551 }
1552 
1553 /*
1554  * This returns whether this is the first mapping of a page.
1555  */
1556 int
1557 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1558 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1559 {
1560 	struct pvo_entry *pvo;
1561 	struct pvo_tqhead *pvoh;
1562 	register_t msr;
1563 	int ptegidx;
1564 	int i;
1565 	int poolflags = PR_NOWAIT;
1566 
1567 	/*
1568 	 * Compute the PTE Group index.
1569 	 */
1570 	va &= ~ADDR_POFF;
1571 	ptegidx = va_to_pteg(pm, va);
1572 
1573 	msr = pmap_interrupts_off();
1574 
1575 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1576 	if (pmap_pvo_remove_depth > 0)
1577 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1578 	if (++pmap_pvo_enter_depth > 1)
1579 		panic("pmap_pvo_enter: called recursively!");
1580 #endif
1581 
1582 	/*
1583 	 * Remove any existing mapping for this page.  Reuse the
1584 	 * pvo entry if there a mapping.
1585 	 */
1586 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1587 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1588 #ifdef DEBUG
1589 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1590 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1591 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1592 			   va < VM_MIN_KERNEL_ADDRESS) {
1593 				printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1594 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1595 				printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1596 				    pvo->pvo_pte.pte_hi,
1597 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1598 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1599 #ifdef DDBX
1600 				Debugger();
1601 #endif
1602 			}
1603 #endif
1604 			PMAPCOUNT(mappings_replaced);
1605 			pmap_pvo_remove(pvo, -1, NULL);
1606 			break;
1607 		}
1608 	}
1609 
1610 	/*
1611 	 * If we aren't overwriting an mapping, try to allocate
1612 	 */
1613 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1614 	--pmap_pvo_enter_depth;
1615 #endif
1616 	pmap_interrupts_restore(msr);
1617 	if (pvo) {
1618 		pmap_pvo_free(pvo);
1619 	}
1620 	pvo = pool_get(pl, poolflags);
1621 	KASSERT((vaddr_t)pvo < VM_MIN_KERNEL_ADDRESS);
1622 
1623 #ifdef DEBUG
1624 	/*
1625 	 * Exercise pmap_pvo_reclaim() a little.
1626 	 */
1627 	if (pvo && (flags & PMAP_CANFAIL) != 0 &&
1628 	    pmap_pvo_reclaim_debugctr++ > 0x1000 &&
1629 	    (pmap_pvo_reclaim_debugctr & 0xff) == 0) {
1630 		pool_put(pl, pvo);
1631 		pvo = NULL;
1632 	}
1633 #endif
1634 
1635 	msr = pmap_interrupts_off();
1636 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1637 	++pmap_pvo_enter_depth;
1638 #endif
1639 	if (pvo == NULL) {
1640 		pvo = pmap_pvo_reclaim(pm);
1641 		if (pvo == NULL) {
1642 			if ((flags & PMAP_CANFAIL) == 0)
1643 				panic("pmap_pvo_enter: failed");
1644 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1645 			pmap_pvo_enter_depth--;
1646 #endif
1647 			PMAPCOUNT(pvos_failed);
1648 			pmap_interrupts_restore(msr);
1649 			return ENOMEM;
1650 		}
1651 	}
1652 
1653 	pvo->pvo_vaddr = va;
1654 	pvo->pvo_pmap = pm;
1655 	pvo->pvo_vaddr &= ~ADDR_POFF;
1656 	if (flags & VM_PROT_EXECUTE) {
1657 		PMAPCOUNT(exec_mappings);
1658 		pvo_set_exec(pvo);
1659 	}
1660 	if (flags & PMAP_WIRED)
1661 		pvo->pvo_vaddr |= PVO_WIRED;
1662 	if (pvo_head != &pmap_pvo_kunmanaged) {
1663 		pvo->pvo_vaddr |= PVO_MANAGED;
1664 		PMAPCOUNT(mappings);
1665 	} else {
1666 		PMAPCOUNT(kernel_mappings);
1667 	}
1668 	pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1669 
1670 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1671 	if (PVO_WIRED_P(pvo))
1672 		pvo->pvo_pmap->pm_stats.wired_count++;
1673 	pvo->pvo_pmap->pm_stats.resident_count++;
1674 #if defined(DEBUG)
1675 /*	if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */
1676 		DPRINTFN(PVOENTER,
1677 		    "pmap_pvo_enter: pvo %p: pm %p va %#" _PRIxva " pa %#" _PRIxpa "\n",
1678 		    pvo, pm, va, pa);
1679 #endif
1680 
1681 	/*
1682 	 * We hope this succeeds but it isn't required.
1683 	 */
1684 	pvoh = &pmap_pvo_table[ptegidx];
1685 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1686 	if (i >= 0) {
1687 		PVO_PTEGIDX_SET(pvo, i);
1688 		PVO_WHERE(pvo, ENTER_INSERT);
1689 		PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1690 		    ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1691 		TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1692 
1693 	} else {
1694 		/*
1695 		 * Since we didn't have room for this entry (which makes it
1696 		 * and evicted entry), place it at the head of the list.
1697 		 */
1698 		TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1699 		PMAPCOUNT(ptes_evicted);
1700 		pm->pm_evictions++;
1701 		/*
1702 		 * If this is a kernel page, make sure it's active.
1703 		 */
1704 		if (pm == pmap_kernel()) {
1705 			i = pmap_pte_spill(pm, va, false);
1706 			KASSERT(i);
1707 		}
1708 	}
1709 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1710 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1711 	pmap_pvo_enter_depth--;
1712 #endif
1713 	pmap_interrupts_restore(msr);
1714 	return 0;
1715 }
1716 
1717 static void
1718 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol)
1719 {
1720 	volatile struct pte *pt;
1721 	int ptegidx;
1722 
1723 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1724 	if (++pmap_pvo_remove_depth > 1)
1725 		panic("pmap_pvo_remove: called recursively!");
1726 #endif
1727 
1728 	/*
1729 	 * If we haven't been supplied the ptegidx, calculate it.
1730 	 */
1731 	if (pteidx == -1) {
1732 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1733 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1734 	} else {
1735 		ptegidx = pteidx >> 3;
1736 		if (pvo->pvo_pte.pte_hi & PTE_HID)
1737 			ptegidx ^= pmap_pteg_mask;
1738 	}
1739 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1740 
1741 	/*
1742 	 * If there is an active pte entry, we need to deactivate it
1743 	 * (and save the ref & chg bits).
1744 	 */
1745 	pt = pmap_pvo_to_pte(pvo, pteidx);
1746 	if (pt != NULL) {
1747 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1748 		PVO_WHERE(pvo, REMOVE);
1749 		PVO_PTEGIDX_CLR(pvo);
1750 		PMAPCOUNT(ptes_removed);
1751 	} else {
1752 		KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1753 		pvo->pvo_pmap->pm_evictions--;
1754 	}
1755 
1756 	/*
1757 	 * Account for executable mappings.
1758 	 */
1759 	if (PVO_EXECUTABLE_P(pvo))
1760 		pvo_clear_exec(pvo);
1761 
1762 	/*
1763 	 * Update our statistics.
1764 	 */
1765 	pvo->pvo_pmap->pm_stats.resident_count--;
1766 	if (PVO_WIRED_P(pvo))
1767 		pvo->pvo_pmap->pm_stats.wired_count--;
1768 
1769 	/*
1770 	 * Save the REF/CHG bits into their cache if the page is managed.
1771 	 */
1772 	if (PVO_MANAGED_P(pvo)) {
1773 		register_t ptelo = pvo->pvo_pte.pte_lo;
1774 		struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1775 
1776 		if (pg != NULL) {
1777 			/*
1778 			 * If this page was changed and it is mapped exec,
1779 			 * invalidate it.
1780 			 */
1781 			if ((ptelo & PTE_CHG) &&
1782 			    (pmap_attr_fetch(pg) & PTE_EXEC)) {
1783 				struct pvo_head *pvoh = vm_page_to_pvoh(pg);
1784 				if (LIST_EMPTY(pvoh)) {
1785 					DPRINTFN(EXEC, "[pmap_pvo_remove: "
1786 					    "%#" _PRIxpa ": clear-exec]\n",
1787 					    VM_PAGE_TO_PHYS(pg));
1788 					pmap_attr_clear(pg, PTE_EXEC);
1789 					PMAPCOUNT(exec_uncached_pvo_remove);
1790 				} else {
1791 					DPRINTFN(EXEC, "[pmap_pvo_remove: "
1792 					    "%#" _PRIxpa ": syncicache]\n",
1793 					    VM_PAGE_TO_PHYS(pg));
1794 					pmap_syncicache(VM_PAGE_TO_PHYS(pg),
1795 					    PAGE_SIZE);
1796 					PMAPCOUNT(exec_synced_pvo_remove);
1797 				}
1798 			}
1799 
1800 			pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1801 		}
1802 		PMAPCOUNT(unmappings);
1803 	} else {
1804 		PMAPCOUNT(kernel_unmappings);
1805 	}
1806 
1807 	/*
1808 	 * Remove the PVO from its lists and return it to the pool.
1809 	 */
1810 	LIST_REMOVE(pvo, pvo_vlink);
1811 	TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1812 	if (pvol) {
1813 		LIST_INSERT_HEAD(pvol, pvo, pvo_vlink);
1814 	}
1815 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1816 	pmap_pvo_remove_depth--;
1817 #endif
1818 }
1819 
1820 void
1821 pmap_pvo_free(struct pvo_entry *pvo)
1822 {
1823 
1824 	pool_put(PVO_MANAGED_P(pvo) ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo);
1825 }
1826 
1827 void
1828 pmap_pvo_free_list(struct pvo_head *pvol)
1829 {
1830 	struct pvo_entry *pvo, *npvo;
1831 
1832 	for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) {
1833 		npvo = LIST_NEXT(pvo, pvo_vlink);
1834 		LIST_REMOVE(pvo, pvo_vlink);
1835 		pmap_pvo_free(pvo);
1836 	}
1837 }
1838 
1839 /*
1840  * Mark a mapping as executable.
1841  * If this is the first executable mapping in the segment,
1842  * clear the noexec flag.
1843  */
1844 static void
1845 pvo_set_exec(struct pvo_entry *pvo)
1846 {
1847 	struct pmap *pm = pvo->pvo_pmap;
1848 
1849 	if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) {
1850 		return;
1851 	}
1852 	pvo->pvo_vaddr |= PVO_EXECUTABLE;
1853 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1854 	{
1855 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1856 		if (pm->pm_exec[sr]++ == 0) {
1857 			pm->pm_sr[sr] &= ~SR_NOEXEC;
1858 		}
1859 	}
1860 #endif
1861 }
1862 
1863 /*
1864  * Mark a mapping as non-executable.
1865  * If this was the last executable mapping in the segment,
1866  * set the noexec flag.
1867  */
1868 static void
1869 pvo_clear_exec(struct pvo_entry *pvo)
1870 {
1871 	struct pmap *pm = pvo->pvo_pmap;
1872 
1873 	if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) {
1874 		return;
1875 	}
1876 	pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1877 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1878 	{
1879 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1880 		if (--pm->pm_exec[sr] == 0) {
1881 			pm->pm_sr[sr] |= SR_NOEXEC;
1882 		}
1883 	}
1884 #endif
1885 }
1886 
1887 /*
1888  * Insert physical page at pa into the given pmap at virtual address va.
1889  */
1890 int
1891 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1892 {
1893 	struct mem_region *mp;
1894 	struct pvo_head *pvo_head;
1895 	struct vm_page *pg;
1896 	struct pool *pl;
1897 	register_t pte_lo;
1898 	int error;
1899 	u_int was_exec = 0;
1900 
1901 	PMAP_LOCK();
1902 
1903 	if (__predict_false(!pmap_initialized)) {
1904 		pvo_head = &pmap_pvo_kunmanaged;
1905 		pl = &pmap_upvo_pool;
1906 		pg = NULL;
1907 		was_exec = PTE_EXEC;
1908 	} else {
1909 		pvo_head = pa_to_pvoh(pa, &pg);
1910 		pl = &pmap_mpvo_pool;
1911 	}
1912 
1913 	DPRINTFN(ENTER,
1914 	    "pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1915 	    pm, va, pa, prot, flags);
1916 
1917 	/*
1918 	 * If this is a managed page, and it's the first reference to the
1919 	 * page clear the execness of the page.  Otherwise fetch the execness.
1920 	 */
1921 	if (pg != NULL)
1922 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1923 
1924 	DPRINTFN(ENTER, " was_exec=%d", was_exec);
1925 
1926 	/*
1927 	 * Assume the page is cache inhibited and access is guarded unless
1928 	 * it's in our available memory array.  If it is in the memory array,
1929 	 * asssume it's in memory coherent memory.
1930 	 */
1931 	if (flags & PMAP_MD_PREFETCHABLE) {
1932 		pte_lo = 0;
1933 	} else
1934 		pte_lo = PTE_G;
1935 
1936 	if ((flags & PMAP_NOCACHE) == 0) {
1937 		for (mp = mem; mp->size; mp++) {
1938 			if (pa >= mp->start && pa < mp->start + mp->size) {
1939 				pte_lo = PTE_M;
1940 				break;
1941 			}
1942 		}
1943 #ifdef MULTIPROCESSOR
1944 		if (((mfpvr() >> 16) & 0xffff) == MPC603e)
1945 			pte_lo = PTE_M;
1946 #endif
1947 	} else {
1948 		pte_lo |= PTE_I;
1949 	}
1950 
1951 	if (prot & VM_PROT_WRITE)
1952 		pte_lo |= PTE_BW;
1953 	else
1954 		pte_lo |= PTE_BR;
1955 
1956 	/*
1957 	 * If this was in response to a fault, "pre-fault" the PTE's
1958 	 * changed/referenced bit appropriately.
1959 	 */
1960 	if (flags & VM_PROT_WRITE)
1961 		pte_lo |= PTE_CHG;
1962 	if (flags & VM_PROT_ALL)
1963 		pte_lo |= PTE_REF;
1964 
1965 	/*
1966 	 * We need to know if this page can be executable
1967 	 */
1968 	flags |= (prot & VM_PROT_EXECUTE);
1969 
1970 	/*
1971 	 * Record mapping for later back-translation and pte spilling.
1972 	 * This will overwrite any existing mapping.
1973 	 */
1974 	error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags);
1975 
1976 	/*
1977 	 * Flush the real page from the instruction cache if this page is
1978 	 * mapped executable and cacheable and has not been flushed since
1979 	 * the last time it was modified.
1980 	 */
1981 	if (error == 0 &&
1982             (flags & VM_PROT_EXECUTE) &&
1983             (pte_lo & PTE_I) == 0 &&
1984 	    was_exec == 0) {
1985 		DPRINTFN(ENTER, " %s", "syncicache");
1986 		PMAPCOUNT(exec_synced);
1987 		pmap_syncicache(pa, PAGE_SIZE);
1988 		if (pg != NULL) {
1989 			pmap_attr_save(pg, PTE_EXEC);
1990 			PMAPCOUNT(exec_cached);
1991 #if defined(DEBUG) || defined(PMAPDEBUG)
1992 			if (pmapdebug & PMAPDEBUG_ENTER)
1993 				printf(" marked-as-exec");
1994 			else if (pmapdebug & PMAPDEBUG_EXEC)
1995 				printf("[pmap_enter: %#" _PRIxpa ": marked-as-exec]\n",
1996 				    VM_PAGE_TO_PHYS(pg));
1997 
1998 #endif
1999 		}
2000 	}
2001 
2002 	DPRINTFN(ENTER, ": error=%d\n", error);
2003 
2004 	PMAP_UNLOCK();
2005 
2006 	return error;
2007 }
2008 
2009 void
2010 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
2011 {
2012 	struct mem_region *mp;
2013 	register_t pte_lo;
2014 	int error;
2015 
2016 #if defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA)
2017 	if (va < VM_MIN_KERNEL_ADDRESS)
2018 		panic("pmap_kenter_pa: attempt to enter "
2019 		    "non-kernel address %#" _PRIxva "!", va);
2020 #endif
2021 
2022 	DPRINTFN(KENTER,
2023 	    "pmap_kenter_pa(%#" _PRIxva ",%#" _PRIxpa ",%#x)\n", va, pa, prot);
2024 
2025 	PMAP_LOCK();
2026 
2027 	/*
2028 	 * Assume the page is cache inhibited and access is guarded unless
2029 	 * it's in our available memory array.  If it is in the memory array,
2030 	 * asssume it's in memory coherent memory.
2031 	 */
2032 	pte_lo = PTE_IG;
2033 	if ((flags & PMAP_NOCACHE) == 0) {
2034 		for (mp = mem; mp->size; mp++) {
2035 			if (pa >= mp->start && pa < mp->start + mp->size) {
2036 				pte_lo = PTE_M;
2037 				break;
2038 			}
2039 		}
2040 #ifdef MULTIPROCESSOR
2041 		if (((mfpvr() >> 16) & 0xffff) == MPC603e)
2042 			pte_lo = PTE_M;
2043 #endif
2044 	}
2045 
2046 	if (prot & VM_PROT_WRITE)
2047 		pte_lo |= PTE_BW;
2048 	else
2049 		pte_lo |= PTE_BR;
2050 
2051 	/*
2052 	 * We don't care about REF/CHG on PVOs on the unmanaged list.
2053 	 */
2054 	error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool,
2055 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED);
2056 
2057 	if (error != 0)
2058 		panic("pmap_kenter_pa: failed to enter va %#" _PRIxva " pa %#" _PRIxpa ": %d",
2059 		      va, pa, error);
2060 
2061 	PMAP_UNLOCK();
2062 }
2063 
2064 void
2065 pmap_kremove(vaddr_t va, vsize_t len)
2066 {
2067 	if (va < VM_MIN_KERNEL_ADDRESS)
2068 		panic("pmap_kremove: attempt to remove "
2069 		    "non-kernel address %#" _PRIxva "!", va);
2070 
2071 	DPRINTFN(KREMOVE, "pmap_kremove(%#" _PRIxva ",%#" _PRIxva ")\n", va, len);
2072 	pmap_remove(pmap_kernel(), va, va + len);
2073 }
2074 
2075 /*
2076  * Remove the given range of mapping entries.
2077  */
2078 void
2079 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
2080 {
2081 	struct pvo_head pvol;
2082 	struct pvo_entry *pvo;
2083 	register_t msr;
2084 	int pteidx;
2085 
2086 	PMAP_LOCK();
2087 	LIST_INIT(&pvol);
2088 	msr = pmap_interrupts_off();
2089 	for (; va < endva; va += PAGE_SIZE) {
2090 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2091 		if (pvo != NULL) {
2092 			pmap_pvo_remove(pvo, pteidx, &pvol);
2093 		}
2094 	}
2095 	pmap_interrupts_restore(msr);
2096 	pmap_pvo_free_list(&pvol);
2097 	PMAP_UNLOCK();
2098 }
2099 
2100 /*
2101  * Get the physical page address for the given pmap/virtual address.
2102  */
2103 bool
2104 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
2105 {
2106 	struct pvo_entry *pvo;
2107 	register_t msr;
2108 
2109 	PMAP_LOCK();
2110 
2111 	/*
2112 	 * If this is a kernel pmap lookup, also check the battable
2113 	 * and if we get a hit, translate the VA to a PA using the
2114 	 * BAT entries.  Don't check for VM_MAX_KERNEL_ADDRESS is
2115 	 * that will wrap back to 0.
2116 	 */
2117 	if (pm == pmap_kernel() &&
2118 	    (va < VM_MIN_KERNEL_ADDRESS ||
2119 	     (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
2120 		KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
2121 #if defined (PMAP_OEA)
2122 #ifdef PPC_OEA601
2123 		if ((MFPVR() >> 16) == MPC601) {
2124 			register_t batu = battable[va >> 23].batu;
2125 			register_t batl = battable[va >> 23].batl;
2126 			register_t sr = iosrtable[va >> ADDR_SR_SHFT];
2127 			if (BAT601_VALID_P(batl) &&
2128 			    BAT601_VA_MATCH_P(batu, batl, va)) {
2129 				register_t mask =
2130 				    (~(batl & BAT601_BSM) << 17) & ~0x1ffffL;
2131 				if (pap)
2132 					*pap = (batl & mask) | (va & ~mask);
2133 				PMAP_UNLOCK();
2134 				return true;
2135 			} else if (SR601_VALID_P(sr) &&
2136 				   SR601_PA_MATCH_P(sr, va)) {
2137 				if (pap)
2138 					*pap = va;
2139 				PMAP_UNLOCK();
2140 				return true;
2141 			}
2142 		} else
2143 #endif /* PPC_OEA601 */
2144 		{
2145 			register_t batu = battable[BAT_VA2IDX(va)].batu;
2146 			if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) {
2147 				register_t batl = battable[BAT_VA2IDX(va)].batl;
2148 				register_t mask =
2149 				    (~(batu & (BAT_XBL|BAT_BL)) << 15) & ~0x1ffffL;
2150 				if (pap)
2151 					*pap = (batl & mask) | (va & ~mask);
2152 				PMAP_UNLOCK();
2153 				return true;
2154 			}
2155 		}
2156 		return false;
2157 #elif defined (PMAP_OEA64_BRIDGE)
2158 	if (va >= SEGMENT_LENGTH)
2159 		panic("%s: pm: %s va >= SEGMENT_LENGTH, va: 0x%08lx\n",
2160 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
2161 	else {
2162 		if (pap)
2163 			*pap = va;
2164 			PMAP_UNLOCK();
2165 			return true;
2166 	}
2167 #elif defined (PMAP_OEA64)
2168 #error PPC_OEA64 not supported
2169 #endif /* PPC_OEA */
2170 	}
2171 
2172 	msr = pmap_interrupts_off();
2173 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2174 	if (pvo != NULL) {
2175 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2176 		if (pap)
2177 			*pap = (pvo->pvo_pte.pte_lo & PTE_RPGN)
2178 			    | (va & ADDR_POFF);
2179 	}
2180 	pmap_interrupts_restore(msr);
2181 	PMAP_UNLOCK();
2182 	return pvo != NULL;
2183 }
2184 
2185 /*
2186  * Lower the protection on the specified range of this pmap.
2187  */
2188 void
2189 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
2190 {
2191 	struct pvo_entry *pvo;
2192 	volatile struct pte *pt;
2193 	register_t msr;
2194 	int pteidx;
2195 
2196 	/*
2197 	 * Since this routine only downgrades protection, we should
2198 	 * always be called with at least one bit not set.
2199 	 */
2200 	KASSERT(prot != VM_PROT_ALL);
2201 
2202 	/*
2203 	 * If there is no protection, this is equivalent to
2204 	 * remove the pmap from the pmap.
2205 	 */
2206 	if ((prot & VM_PROT_READ) == 0) {
2207 		pmap_remove(pm, va, endva);
2208 		return;
2209 	}
2210 
2211 	PMAP_LOCK();
2212 
2213 	msr = pmap_interrupts_off();
2214 	for (; va < endva; va += PAGE_SIZE) {
2215 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2216 		if (pvo == NULL)
2217 			continue;
2218 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2219 
2220 		/*
2221 		 * Revoke executable if asked to do so.
2222 		 */
2223 		if ((prot & VM_PROT_EXECUTE) == 0)
2224 			pvo_clear_exec(pvo);
2225 
2226 #if 0
2227 		/*
2228 		 * If the page is already read-only, no change
2229 		 * needs to be made.
2230 		 */
2231 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
2232 			continue;
2233 #endif
2234 		/*
2235 		 * Grab the PTE pointer before we diddle with
2236 		 * the cached PTE copy.
2237 		 */
2238 		pt = pmap_pvo_to_pte(pvo, pteidx);
2239 		/*
2240 		 * Change the protection of the page.
2241 		 */
2242 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2243 		pvo->pvo_pte.pte_lo |= PTE_BR;
2244 
2245 		/*
2246 		 * If the PVO is in the page table, update
2247 		 * that pte at well.
2248 		 */
2249 		if (pt != NULL) {
2250 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2251 			PVO_WHERE(pvo, PMAP_PROTECT);
2252 			PMAPCOUNT(ptes_changed);
2253 		}
2254 
2255 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2256 	}
2257 	pmap_interrupts_restore(msr);
2258 	PMAP_UNLOCK();
2259 }
2260 
2261 void
2262 pmap_unwire(pmap_t pm, vaddr_t va)
2263 {
2264 	struct pvo_entry *pvo;
2265 	register_t msr;
2266 
2267 	PMAP_LOCK();
2268 	msr = pmap_interrupts_off();
2269 	pvo = pmap_pvo_find_va(pm, va, NULL);
2270 	if (pvo != NULL) {
2271 		if (PVO_WIRED_P(pvo)) {
2272 			pvo->pvo_vaddr &= ~PVO_WIRED;
2273 			pm->pm_stats.wired_count--;
2274 		}
2275 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2276 	}
2277 	pmap_interrupts_restore(msr);
2278 	PMAP_UNLOCK();
2279 }
2280 
2281 /*
2282  * Lower the protection on the specified physical page.
2283  */
2284 void
2285 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
2286 {
2287 	struct pvo_head *pvo_head, pvol;
2288 	struct pvo_entry *pvo, *next_pvo;
2289 	volatile struct pte *pt;
2290 	register_t msr;
2291 
2292 	PMAP_LOCK();
2293 
2294 	KASSERT(prot != VM_PROT_ALL);
2295 	LIST_INIT(&pvol);
2296 	msr = pmap_interrupts_off();
2297 
2298 	/*
2299 	 * When UVM reuses a page, it does a pmap_page_protect with
2300 	 * VM_PROT_NONE.  At that point, we can clear the exec flag
2301 	 * since we know the page will have different contents.
2302 	 */
2303 	if ((prot & VM_PROT_READ) == 0) {
2304 		DPRINTFN(EXEC, "[pmap_page_protect: %#" _PRIxpa ": clear-exec]\n",
2305 		    VM_PAGE_TO_PHYS(pg));
2306 		if (pmap_attr_fetch(pg) & PTE_EXEC) {
2307 			PMAPCOUNT(exec_uncached_page_protect);
2308 			pmap_attr_clear(pg, PTE_EXEC);
2309 		}
2310 	}
2311 
2312 	pvo_head = vm_page_to_pvoh(pg);
2313 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2314 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
2315 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2316 
2317 		/*
2318 		 * Downgrading to no mapping at all, we just remove the entry.
2319 		 */
2320 		if ((prot & VM_PROT_READ) == 0) {
2321 			pmap_pvo_remove(pvo, -1, &pvol);
2322 			continue;
2323 		}
2324 
2325 		/*
2326 		 * If EXEC permission is being revoked, just clear the
2327 		 * flag in the PVO.
2328 		 */
2329 		if ((prot & VM_PROT_EXECUTE) == 0)
2330 			pvo_clear_exec(pvo);
2331 
2332 		/*
2333 		 * If this entry is already RO, don't diddle with the
2334 		 * page table.
2335 		 */
2336 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2337 			PMAP_PVO_CHECK(pvo);
2338 			continue;
2339 		}
2340 
2341 		/*
2342 		 * Grab the PTE before the we diddle the bits so
2343 		 * pvo_to_pte can verify the pte contents are as
2344 		 * expected.
2345 		 */
2346 		pt = pmap_pvo_to_pte(pvo, -1);
2347 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2348 		pvo->pvo_pte.pte_lo |= PTE_BR;
2349 		if (pt != NULL) {
2350 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2351 			PVO_WHERE(pvo, PMAP_PAGE_PROTECT);
2352 			PMAPCOUNT(ptes_changed);
2353 		}
2354 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2355 	}
2356 	pmap_interrupts_restore(msr);
2357 	pmap_pvo_free_list(&pvol);
2358 
2359 	PMAP_UNLOCK();
2360 }
2361 
2362 /*
2363  * Activate the address space for the specified process.  If the process
2364  * is the current process, load the new MMU context.
2365  */
2366 void
2367 pmap_activate(struct lwp *l)
2368 {
2369 	struct pcb *pcb = lwp_getpcb(l);
2370 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2371 
2372 	DPRINTFN(ACTIVATE,
2373 	    "pmap_activate: lwp %p (curlwp %p)\n", l, curlwp);
2374 
2375 	/*
2376 	 * XXX Normally performed in cpu_lwp_fork().
2377 	 */
2378 	pcb->pcb_pm = pmap;
2379 
2380 	/*
2381 	* In theory, the SR registers need only be valid on return
2382 	* to user space wait to do them there.
2383 	*/
2384 	if (l == curlwp) {
2385 		/* Store pointer to new current pmap. */
2386 		curpm = pmap;
2387 	}
2388 }
2389 
2390 /*
2391  * Deactivate the specified process's address space.
2392  */
2393 void
2394 pmap_deactivate(struct lwp *l)
2395 {
2396 }
2397 
2398 bool
2399 pmap_query_bit(struct vm_page *pg, int ptebit)
2400 {
2401 	struct pvo_entry *pvo;
2402 	volatile struct pte *pt;
2403 	register_t msr;
2404 
2405 	PMAP_LOCK();
2406 
2407 	if (pmap_attr_fetch(pg) & ptebit) {
2408 		PMAP_UNLOCK();
2409 		return true;
2410 	}
2411 
2412 	msr = pmap_interrupts_off();
2413 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2414 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2415 		/*
2416 		 * See if we saved the bit off.  If so cache, it and return
2417 		 * success.
2418 		 */
2419 		if (pvo->pvo_pte.pte_lo & ptebit) {
2420 			pmap_attr_save(pg, ptebit);
2421 			PMAP_PVO_CHECK(pvo);		/* sanity check */
2422 			pmap_interrupts_restore(msr);
2423 			PMAP_UNLOCK();
2424 			return true;
2425 		}
2426 	}
2427 	/*
2428 	 * No luck, now go thru the hard part of looking at the ptes
2429 	 * themselves.  Sync so any pending REF/CHG bits are flushed
2430 	 * to the PTEs.
2431 	 */
2432 	SYNC();
2433 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2434 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2435 		/*
2436 		 * See if this pvo have a valid PTE.  If so, fetch the
2437 		 * REF/CHG bits from the valid PTE.  If the appropriate
2438 		 * ptebit is set, cache, it and return success.
2439 		 */
2440 		pt = pmap_pvo_to_pte(pvo, -1);
2441 		if (pt != NULL) {
2442 			pmap_pte_synch(pt, &pvo->pvo_pte);
2443 			if (pvo->pvo_pte.pte_lo & ptebit) {
2444 				pmap_attr_save(pg, ptebit);
2445 				PMAP_PVO_CHECK(pvo);		/* sanity check */
2446 				pmap_interrupts_restore(msr);
2447 				PMAP_UNLOCK();
2448 				return true;
2449 			}
2450 		}
2451 	}
2452 	pmap_interrupts_restore(msr);
2453 	PMAP_UNLOCK();
2454 	return false;
2455 }
2456 
2457 bool
2458 pmap_clear_bit(struct vm_page *pg, int ptebit)
2459 {
2460 	struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2461 	struct pvo_entry *pvo;
2462 	volatile struct pte *pt;
2463 	register_t msr;
2464 	int rv = 0;
2465 
2466 	PMAP_LOCK();
2467 	msr = pmap_interrupts_off();
2468 
2469 	/*
2470 	 * Fetch the cache value
2471 	 */
2472 	rv |= pmap_attr_fetch(pg);
2473 
2474 	/*
2475 	 * Clear the cached value.
2476 	 */
2477 	pmap_attr_clear(pg, ptebit);
2478 
2479 	/*
2480 	 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2481 	 * can reset the right ones).  Note that since the pvo entries and
2482 	 * list heads are accessed via BAT0 and are never placed in the
2483 	 * page table, we don't have to worry about further accesses setting
2484 	 * the REF/CHG bits.
2485 	 */
2486 	SYNC();
2487 
2488 	/*
2489 	 * For each pvo entry, clear pvo's ptebit.  If this pvo have a
2490 	 * valid PTE.  If so, clear the ptebit from the valid PTE.
2491 	 */
2492 	LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2493 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2494 		pt = pmap_pvo_to_pte(pvo, -1);
2495 		if (pt != NULL) {
2496 			/*
2497 			 * Only sync the PTE if the bit we are looking
2498 			 * for is not already set.
2499 			 */
2500 			if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2501 				pmap_pte_synch(pt, &pvo->pvo_pte);
2502 			/*
2503 			 * If the bit we are looking for was already set,
2504 			 * clear that bit in the pte.
2505 			 */
2506 			if (pvo->pvo_pte.pte_lo & ptebit)
2507 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2508 		}
2509 		rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2510 		pvo->pvo_pte.pte_lo &= ~ptebit;
2511 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2512 	}
2513 	pmap_interrupts_restore(msr);
2514 
2515 	/*
2516 	 * If we are clearing the modify bit and this page was marked EXEC
2517 	 * and the user of the page thinks the page was modified, then we
2518 	 * need to clean it from the icache if it's mapped or clear the EXEC
2519 	 * bit if it's not mapped.  The page itself might not have the CHG
2520 	 * bit set if the modification was done via DMA to the page.
2521 	 */
2522 	if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2523 		if (LIST_EMPTY(pvoh)) {
2524 			DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": clear-exec]\n",
2525 			    VM_PAGE_TO_PHYS(pg));
2526 			pmap_attr_clear(pg, PTE_EXEC);
2527 			PMAPCOUNT(exec_uncached_clear_modify);
2528 		} else {
2529 			DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": syncicache]\n",
2530 			    VM_PAGE_TO_PHYS(pg));
2531 			pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE);
2532 			PMAPCOUNT(exec_synced_clear_modify);
2533 		}
2534 	}
2535 	PMAP_UNLOCK();
2536 	return (rv & ptebit) != 0;
2537 }
2538 
2539 void
2540 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2541 {
2542 	struct pvo_entry *pvo;
2543 	size_t offset = va & ADDR_POFF;
2544 	int s;
2545 
2546 	PMAP_LOCK();
2547 	s = splvm();
2548 	while (len > 0) {
2549 		size_t seglen = PAGE_SIZE - offset;
2550 		if (seglen > len)
2551 			seglen = len;
2552 		pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2553 		if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) {
2554 			pmap_syncicache(
2555 			    (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2556 			PMAP_PVO_CHECK(pvo);
2557 		}
2558 		va += seglen;
2559 		len -= seglen;
2560 		offset = 0;
2561 	}
2562 	splx(s);
2563 	PMAP_UNLOCK();
2564 }
2565 
2566 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2567 void
2568 pmap_pte_print(volatile struct pte *pt)
2569 {
2570 	printf("PTE %p: ", pt);
2571 
2572 #if defined(PMAP_OEA)
2573 	/* High word: */
2574 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2575 #else
2576 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2577 #endif /* PMAP_OEA */
2578 
2579 	printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2580 	printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2581 
2582 	printf("%#" _PRIxpte " %#" _PRIxpte "",
2583 	    (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2584 	    pt->pte_hi & PTE_API);
2585 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2586 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2587 #else
2588 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2589 #endif /* PMAP_OEA */
2590 
2591 	/* Low word: */
2592 #if defined (PMAP_OEA)
2593 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2594 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2595 #else
2596 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2597 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2598 #endif
2599 	printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2600 	printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2601 	printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2602 	printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2603 	printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2604 	printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2605 	switch (pt->pte_lo & PTE_PP) {
2606 	case PTE_BR: printf("br]\n"); break;
2607 	case PTE_BW: printf("bw]\n"); break;
2608 	case PTE_SO: printf("so]\n"); break;
2609 	case PTE_SW: printf("sw]\n"); break;
2610 	}
2611 }
2612 #endif
2613 
2614 #if defined(DDB)
2615 void
2616 pmap_pteg_check(void)
2617 {
2618 	volatile struct pte *pt;
2619 	int i;
2620 	int ptegidx;
2621 	u_int p_valid = 0;
2622 	u_int s_valid = 0;
2623 	u_int invalid = 0;
2624 
2625 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2626 		for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2627 			if (pt->pte_hi & PTE_VALID) {
2628 				if (pt->pte_hi & PTE_HID)
2629 					s_valid++;
2630 				else
2631 				{
2632 					p_valid++;
2633 				}
2634 			} else
2635 				invalid++;
2636 		}
2637 	}
2638 	printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2639 		p_valid, p_valid, s_valid, s_valid,
2640 		invalid, invalid);
2641 }
2642 
2643 void
2644 pmap_print_mmuregs(void)
2645 {
2646 	int i;
2647 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2648 	u_int cpuvers;
2649 #endif
2650 #ifndef PMAP_OEA64
2651 	vaddr_t addr;
2652 	register_t soft_sr[16];
2653 #endif
2654 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2655 	struct bat soft_ibat[4];
2656 	struct bat soft_dbat[4];
2657 #endif
2658 	paddr_t sdr1;
2659 
2660 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2661 	cpuvers = MFPVR() >> 16;
2662 #endif
2663 	__asm volatile ("mfsdr1 %0" : "=r"(sdr1));
2664 #ifndef PMAP_OEA64
2665 	addr = 0;
2666 	for (i = 0; i < 16; i++) {
2667 		soft_sr[i] = MFSRIN(addr);
2668 		addr += (1 << ADDR_SR_SHFT);
2669 	}
2670 #endif
2671 
2672 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2673 	/* read iBAT (601: uBAT) registers */
2674 	__asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2675 	__asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2676 	__asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2677 	__asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2678 	__asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2679 	__asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2680 	__asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2681 	__asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2682 
2683 
2684 	if (cpuvers != MPC601) {
2685 		/* read dBAT registers */
2686 		__asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2687 		__asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2688 		__asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2689 		__asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2690 		__asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2691 		__asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2692 		__asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2693 		__asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2694 	}
2695 #endif
2696 
2697 	printf("SDR1:\t%#" _PRIxpa "\n", sdr1);
2698 #ifndef PMAP_OEA64
2699 	printf("SR[]:\t");
2700 	for (i = 0; i < 4; i++)
2701 		printf("0x%08lx,   ", soft_sr[i]);
2702 	printf("\n\t");
2703 	for ( ; i < 8; i++)
2704 		printf("0x%08lx,   ", soft_sr[i]);
2705 	printf("\n\t");
2706 	for ( ; i < 12; i++)
2707 		printf("0x%08lx,   ", soft_sr[i]);
2708 	printf("\n\t");
2709 	for ( ; i < 16; i++)
2710 		printf("0x%08lx,   ", soft_sr[i]);
2711 	printf("\n");
2712 #endif
2713 
2714 #if defined(PMAP_OEA) || defined(PMAP_OEA_BRIDGE)
2715 	printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2716 	for (i = 0; i < 4; i++) {
2717 		printf("0x%08lx 0x%08lx, ",
2718 			soft_ibat[i].batu, soft_ibat[i].batl);
2719 		if (i == 1)
2720 			printf("\n\t");
2721 	}
2722 	if (cpuvers != MPC601) {
2723 		printf("\ndBAT[]:\t");
2724 		for (i = 0; i < 4; i++) {
2725 			printf("0x%08lx 0x%08lx, ",
2726 				soft_dbat[i].batu, soft_dbat[i].batl);
2727 			if (i == 1)
2728 				printf("\n\t");
2729 		}
2730 	}
2731 	printf("\n");
2732 #endif /* PMAP_OEA... */
2733 }
2734 
2735 void
2736 pmap_print_pte(pmap_t pm, vaddr_t va)
2737 {
2738 	struct pvo_entry *pvo;
2739 	volatile struct pte *pt;
2740 	int pteidx;
2741 
2742 	pvo = pmap_pvo_find_va(pm, va, &pteidx);
2743 	if (pvo != NULL) {
2744 		pt = pmap_pvo_to_pte(pvo, pteidx);
2745 		if (pt != NULL) {
2746 			printf("VA %#" _PRIxva " -> %p -> %s %#" _PRIxpte ", %#" _PRIxpte "\n",
2747 				va, pt,
2748 				pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2749 				pt->pte_hi, pt->pte_lo);
2750 		} else {
2751 			printf("No valid PTE found\n");
2752 		}
2753 	} else {
2754 		printf("Address not in pmap\n");
2755 	}
2756 }
2757 
2758 void
2759 pmap_pteg_dist(void)
2760 {
2761 	struct pvo_entry *pvo;
2762 	int ptegidx;
2763 	int depth;
2764 	int max_depth = 0;
2765 	unsigned int depths[64];
2766 
2767 	memset(depths, 0, sizeof(depths));
2768 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2769 		depth = 0;
2770 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2771 			depth++;
2772 		}
2773 		if (depth > max_depth)
2774 			max_depth = depth;
2775 		if (depth > 63)
2776 			depth = 63;
2777 		depths[depth]++;
2778 	}
2779 
2780 	for (depth = 0; depth < 64; depth++) {
2781 		printf("  [%2d]: %8u", depth, depths[depth]);
2782 		if ((depth & 3) == 3)
2783 			printf("\n");
2784 		if (depth == max_depth)
2785 			break;
2786 	}
2787 	if ((depth & 3) != 3)
2788 		printf("\n");
2789 	printf("Max depth found was %d\n", max_depth);
2790 }
2791 #endif /* DEBUG */
2792 
2793 #if defined(PMAPCHECK) || defined(DEBUG)
2794 void
2795 pmap_pvo_verify(void)
2796 {
2797 	int ptegidx;
2798 	int s;
2799 
2800 	s = splvm();
2801 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2802 		struct pvo_entry *pvo;
2803 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2804 			if ((uintptr_t) pvo >= SEGMENT_LENGTH)
2805 				panic("pmap_pvo_verify: invalid pvo %p "
2806 				    "on list %#x", pvo, ptegidx);
2807 			pmap_pvo_check(pvo);
2808 		}
2809 	}
2810 	splx(s);
2811 }
2812 #endif /* PMAPCHECK */
2813 
2814 
2815 void *
2816 pmap_pool_ualloc(struct pool *pp, int flags)
2817 {
2818 	struct pvo_page *pvop;
2819 
2820 	if (uvm.page_init_done != true) {
2821 		return (void *) uvm_pageboot_alloc(PAGE_SIZE);
2822 	}
2823 
2824 	PMAP_LOCK();
2825 	pvop = SIMPLEQ_FIRST(&pmap_upvop_head);
2826 	if (pvop != NULL) {
2827 		pmap_upvop_free--;
2828 		SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link);
2829 		PMAP_UNLOCK();
2830 		return pvop;
2831 	}
2832 	PMAP_UNLOCK();
2833 	return pmap_pool_malloc(pp, flags);
2834 }
2835 
2836 void *
2837 pmap_pool_malloc(struct pool *pp, int flags)
2838 {
2839 	struct pvo_page *pvop;
2840 	struct vm_page *pg;
2841 
2842 	PMAP_LOCK();
2843 	pvop = SIMPLEQ_FIRST(&pmap_mpvop_head);
2844 	if (pvop != NULL) {
2845 		pmap_mpvop_free--;
2846 		SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link);
2847 		PMAP_UNLOCK();
2848 		return pvop;
2849 	}
2850 	PMAP_UNLOCK();
2851  again:
2852 	pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE,
2853 	    UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256);
2854 	if (__predict_false(pg == NULL)) {
2855 		if (flags & PR_WAITOK) {
2856 			uvm_wait("plpg");
2857 			goto again;
2858 		} else {
2859 			return (0);
2860 		}
2861 	}
2862 	KDASSERT(VM_PAGE_TO_PHYS(pg) == (uintptr_t)VM_PAGE_TO_PHYS(pg));
2863 	return (void *)(uintptr_t) VM_PAGE_TO_PHYS(pg);
2864 }
2865 
2866 void
2867 pmap_pool_ufree(struct pool *pp, void *va)
2868 {
2869 	struct pvo_page *pvop;
2870 #if 0
2871 	if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) {
2872 		pmap_pool_mfree(va, size, tag);
2873 		return;
2874 	}
2875 #endif
2876 	PMAP_LOCK();
2877 	pvop = va;
2878 	SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link);
2879 	pmap_upvop_free++;
2880 	if (pmap_upvop_free > pmap_upvop_maxfree)
2881 		pmap_upvop_maxfree = pmap_upvop_free;
2882 	PMAP_UNLOCK();
2883 }
2884 
2885 void
2886 pmap_pool_mfree(struct pool *pp, void *va)
2887 {
2888 	struct pvo_page *pvop;
2889 
2890 	PMAP_LOCK();
2891 	pvop = va;
2892 	SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link);
2893 	pmap_mpvop_free++;
2894 	if (pmap_mpvop_free > pmap_mpvop_maxfree)
2895 		pmap_mpvop_maxfree = pmap_mpvop_free;
2896 	PMAP_UNLOCK();
2897 #if 0
2898 	uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va));
2899 #endif
2900 }
2901 
2902 /*
2903  * This routine in bootstraping to steal to-be-managed memory (which will
2904  * then be unmanaged).  We use it to grab from the first 256MB for our
2905  * pmap needs and above 256MB for other stuff.
2906  */
2907 vaddr_t
2908 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2909 {
2910 	vsize_t size;
2911 	vaddr_t va;
2912 	paddr_t pa = 0;
2913 	int npgs, bank;
2914 	struct vm_physseg *ps;
2915 
2916 	if (uvm.page_init_done == true)
2917 		panic("pmap_steal_memory: called _after_ bootstrap");
2918 
2919 	*vstartp = VM_MIN_KERNEL_ADDRESS;
2920 	*vendp = VM_MAX_KERNEL_ADDRESS;
2921 
2922 	size = round_page(vsize);
2923 	npgs = atop(size);
2924 
2925 	/*
2926 	 * PA 0 will never be among those given to UVM so we can use it
2927 	 * to indicate we couldn't steal any memory.
2928 	 */
2929 	for (bank = 0; bank < vm_nphysseg; bank++) {
2930 		ps = VM_PHYSMEM_PTR(bank);
2931 		if (ps->free_list == VM_FREELIST_FIRST256 &&
2932 		    ps->avail_end - ps->avail_start >= npgs) {
2933 			pa = ptoa(ps->avail_start);
2934 			break;
2935 		}
2936 	}
2937 
2938 	if (pa == 0)
2939 		panic("pmap_steal_memory: no approriate memory to steal!");
2940 
2941 	ps->avail_start += npgs;
2942 	ps->start += npgs;
2943 
2944 	/*
2945 	 * If we've used up all the pages in the segment, remove it and
2946 	 * compact the list.
2947 	 */
2948 	if (ps->avail_start == ps->end) {
2949 		/*
2950 		 * If this was the last one, then a very bad thing has occurred
2951 		 */
2952 		if (--vm_nphysseg == 0)
2953 			panic("pmap_steal_memory: out of memory!");
2954 
2955 		printf("pmap_steal_memory: consumed bank %d\n", bank);
2956 		for (; bank < vm_nphysseg; bank++, ps++) {
2957 			ps[0] = ps[1];
2958 		}
2959 	}
2960 
2961 	va = (vaddr_t) pa;
2962 	memset((void *) va, 0, size);
2963 	pmap_pages_stolen += npgs;
2964 #ifdef DEBUG
2965 	if (pmapdebug && npgs > 1) {
2966 		u_int cnt = 0;
2967 		for (bank = 0; bank < vm_nphysseg; bank++) {
2968 			ps = VM_PHYSMEM_PTR(bank);
2969 			cnt += ps->avail_end - ps->avail_start;
2970 		}
2971 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2972 		    npgs, pmap_pages_stolen, cnt);
2973 	}
2974 #endif
2975 
2976 	return va;
2977 }
2978 
2979 /*
2980  * Find a chuck of memory with right size and alignment.
2981  */
2982 paddr_t
2983 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2984 {
2985 	struct mem_region *mp;
2986 	paddr_t s, e;
2987 	int i, j;
2988 
2989 	size = round_page(size);
2990 
2991 	DPRINTFN(BOOT,
2992 	    "pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
2993 	    size, alignment, at_end);
2994 
2995 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2996 		panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
2997 		    alignment);
2998 
2999 	if (at_end) {
3000 		if (alignment != PAGE_SIZE)
3001 			panic("pmap_boot_find_memory: invalid ending "
3002 			    "alignment %#" _PRIxpa, alignment);
3003 
3004 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
3005 			s = mp->start + mp->size - size;
3006 			if (s >= mp->start && mp->size >= size) {
3007 				DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3008 				DPRINTFN(BOOT,
3009 				    "pmap_boot_find_memory: b-avail[%d] start "
3010 				    "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3011 				     mp->start, mp->size);
3012 				mp->size -= size;
3013 				DPRINTFN(BOOT,
3014 				    "pmap_boot_find_memory: a-avail[%d] start "
3015 				    "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3016 				     mp->start, mp->size);
3017 				return s;
3018 			}
3019 		}
3020 		panic("pmap_boot_find_memory: no available memory");
3021 	}
3022 
3023 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3024 		s = (mp->start + alignment - 1) & ~(alignment-1);
3025 		e = s + size;
3026 
3027 		/*
3028 		 * Is the calculated region entirely within the region?
3029 		 */
3030 		if (s < mp->start || e > mp->start + mp->size)
3031 			continue;
3032 
3033 		DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3034 		if (s == mp->start) {
3035 			/*
3036 			 * If the block starts at the beginning of region,
3037 			 * adjust the size & start. (the region may now be
3038 			 * zero in length)
3039 			 */
3040 			DPRINTFN(BOOT,
3041 			    "pmap_boot_find_memory: b-avail[%d] start "
3042 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3043 			mp->start += size;
3044 			mp->size -= size;
3045 			DPRINTFN(BOOT,
3046 			    "pmap_boot_find_memory: a-avail[%d] start "
3047 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3048 		} else if (e == mp->start + mp->size) {
3049 			/*
3050 			 * If the block starts at the beginning of region,
3051 			 * adjust only the size.
3052 			 */
3053 			DPRINTFN(BOOT,
3054 			    "pmap_boot_find_memory: b-avail[%d] start "
3055 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3056 			mp->size -= size;
3057 			DPRINTFN(BOOT,
3058 			    "pmap_boot_find_memory: a-avail[%d] start "
3059 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3060 		} else {
3061 			/*
3062 			 * Block is in the middle of the region, so we
3063 			 * have to split it in two.
3064 			 */
3065 			for (j = avail_cnt; j > i + 1; j--) {
3066 				avail[j] = avail[j-1];
3067 			}
3068 			DPRINTFN(BOOT,
3069 			    "pmap_boot_find_memory: b-avail[%d] start "
3070 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3071 			mp[1].start = e;
3072 			mp[1].size = mp[0].start + mp[0].size - e;
3073 			mp[0].size = s - mp[0].start;
3074 			avail_cnt++;
3075 			for (; i < avail_cnt; i++) {
3076 				DPRINTFN(BOOT,
3077 				    "pmap_boot_find_memory: a-avail[%d] "
3078 				    "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3079 				     avail[i].start, avail[i].size);
3080 			}
3081 		}
3082 		KASSERT(s == (uintptr_t) s);
3083 		return s;
3084 	}
3085 	panic("pmap_boot_find_memory: not enough memory for "
3086 	    "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3087 }
3088 
3089 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3090 #if defined (PMAP_OEA64_BRIDGE)
3091 int
3092 pmap_setup_segment0_map(int use_large_pages, ...)
3093 {
3094     vaddr_t va, va_end;
3095 
3096     register_t pte_lo = 0x0;
3097     int ptegidx = 0;
3098     struct pte pte;
3099     va_list ap;
3100 
3101     /* Coherent + Supervisor RW, no user access */
3102     pte_lo = PTE_M;
3103 
3104     /* XXXSL
3105      * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3106      * these have to take priority.
3107      */
3108     for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3109         ptegidx = va_to_pteg(pmap_kernel(), va);
3110         pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3111         (void)pmap_pte_insert(ptegidx, &pte);
3112     }
3113 
3114     va_start(ap, use_large_pages);
3115     while (1) {
3116         paddr_t pa;
3117         size_t size;
3118 
3119         va = va_arg(ap, vaddr_t);
3120 
3121         if (va == 0)
3122             break;
3123 
3124         pa = va_arg(ap, paddr_t);
3125         size = va_arg(ap, size_t);
3126 
3127         for (va_end = va + size; va < va_end; va += 0x1000, pa += 0x1000) {
3128 #if 0
3129 	    printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__,  va, pa);
3130 #endif
3131             ptegidx = va_to_pteg(pmap_kernel(), va);
3132             pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3133             (void)pmap_pte_insert(ptegidx, &pte);
3134         }
3135     }
3136 
3137     TLBSYNC();
3138     SYNC();
3139     return (0);
3140 }
3141 #endif /* PMAP_OEA64_BRIDGE */
3142 
3143 /*
3144  * This is not part of the defined PMAP interface and is specific to the
3145  * PowerPC architecture.  This is called during initppc, before the system
3146  * is really initialized.
3147  */
3148 void
3149 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3150 {
3151 	struct mem_region *mp, tmp;
3152 	paddr_t s, e;
3153 	psize_t size;
3154 	int i, j;
3155 
3156 	/*
3157 	 * Get memory.
3158 	 */
3159 	mem_regions(&mem, &avail);
3160 #if defined(DEBUG)
3161 	if (pmapdebug & PMAPDEBUG_BOOT) {
3162 		printf("pmap_bootstrap: memory configuration:\n");
3163 		for (mp = mem; mp->size; mp++) {
3164 			printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3165 				mp->start, mp->size);
3166 		}
3167 		for (mp = avail; mp->size; mp++) {
3168 			printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3169 				mp->start, mp->size);
3170 		}
3171 	}
3172 #endif
3173 
3174 	/*
3175 	 * Find out how much physical memory we have and in how many chunks.
3176 	 */
3177 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3178 		if (mp->start >= pmap_memlimit)
3179 			continue;
3180 		if (mp->start + mp->size > pmap_memlimit) {
3181 			size = pmap_memlimit - mp->start;
3182 			physmem += btoc(size);
3183 		} else {
3184 			physmem += btoc(mp->size);
3185 		}
3186 		mem_cnt++;
3187 	}
3188 
3189 	/*
3190 	 * Count the number of available entries.
3191 	 */
3192 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
3193 		avail_cnt++;
3194 
3195 	/*
3196 	 * Page align all regions.
3197 	 */
3198 	kernelstart = trunc_page(kernelstart);
3199 	kernelend = round_page(kernelend);
3200 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3201 		s = round_page(mp->start);
3202 		mp->size -= (s - mp->start);
3203 		mp->size = trunc_page(mp->size);
3204 		mp->start = s;
3205 		e = mp->start + mp->size;
3206 
3207 		DPRINTFN(BOOT,
3208 		    "pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3209 		    i, mp->start, mp->size);
3210 
3211 		/*
3212 		 * Don't allow the end to run beyond our artificial limit
3213 		 */
3214 		if (e > pmap_memlimit)
3215 			e = pmap_memlimit;
3216 
3217 		/*
3218 		 * Is this region empty or strange?  skip it.
3219 		 */
3220 		if (e <= s) {
3221 			mp->start = 0;
3222 			mp->size = 0;
3223 			continue;
3224 		}
3225 
3226 		/*
3227 		 * Does this overlap the beginning of kernel?
3228 		 *   Does extend past the end of the kernel?
3229 		 */
3230 		else if (s < kernelstart && e > kernelstart) {
3231 			if (e > kernelend) {
3232 				avail[avail_cnt].start = kernelend;
3233 				avail[avail_cnt].size = e - kernelend;
3234 				avail_cnt++;
3235 			}
3236 			mp->size = kernelstart - s;
3237 		}
3238 		/*
3239 		 * Check whether this region overlaps the end of the kernel.
3240 		 */
3241 		else if (s < kernelend && e > kernelend) {
3242 			mp->start = kernelend;
3243 			mp->size = e - kernelend;
3244 		}
3245 		/*
3246 		 * Look whether this regions is completely inside the kernel.
3247 		 * Nuke it if it does.
3248 		 */
3249 		else if (s >= kernelstart && e <= kernelend) {
3250 			mp->start = 0;
3251 			mp->size = 0;
3252 		}
3253 		/*
3254 		 * If the user imposed a memory limit, enforce it.
3255 		 */
3256 		else if (s >= pmap_memlimit) {
3257 			mp->start = -PAGE_SIZE;	/* let's know why */
3258 			mp->size = 0;
3259 		}
3260 		else {
3261 			mp->start = s;
3262 			mp->size = e - s;
3263 		}
3264 		DPRINTFN(BOOT,
3265 		    "pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3266 		    i, mp->start, mp->size);
3267 	}
3268 
3269 	/*
3270 	 * Move (and uncount) all the null return to the end.
3271 	 */
3272 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3273 		if (mp->size == 0) {
3274 			tmp = avail[i];
3275 			avail[i] = avail[--avail_cnt];
3276 			avail[avail_cnt] = avail[i];
3277 		}
3278 	}
3279 
3280 	/*
3281 	 * (Bubble)sort them into ascending order.
3282 	 */
3283 	for (i = 0; i < avail_cnt; i++) {
3284 		for (j = i + 1; j < avail_cnt; j++) {
3285 			if (avail[i].start > avail[j].start) {
3286 				tmp = avail[i];
3287 				avail[i] = avail[j];
3288 				avail[j] = tmp;
3289 			}
3290 		}
3291 	}
3292 
3293 	/*
3294 	 * Make sure they don't overlap.
3295 	 */
3296 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3297 		if (mp[0].start + mp[0].size > mp[1].start) {
3298 			mp[0].size = mp[1].start - mp[0].start;
3299 		}
3300 		DPRINTFN(BOOT,
3301 		    "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3302 		    i, mp->start, mp->size);
3303 	}
3304 	DPRINTFN(BOOT,
3305 	    "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3306 	    i, mp->start, mp->size);
3307 
3308 #ifdef	PTEGCOUNT
3309 	pmap_pteg_cnt = PTEGCOUNT;
3310 #else /* PTEGCOUNT */
3311 
3312 	pmap_pteg_cnt = 0x1000;
3313 
3314 	while (pmap_pteg_cnt < physmem)
3315 		pmap_pteg_cnt <<= 1;
3316 
3317 	pmap_pteg_cnt >>= 1;
3318 #endif /* PTEGCOUNT */
3319 
3320 #ifdef DEBUG
3321 	DPRINTFN(BOOT, "pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt);
3322 #endif
3323 
3324 	/*
3325 	 * Find suitably aligned memory for PTEG hash table.
3326 	 */
3327 	size = pmap_pteg_cnt * sizeof(struct pteg);
3328 	pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3329 
3330 #ifdef DEBUG
3331 	DPRINTFN(BOOT,
3332 		"PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table);
3333 #endif
3334 
3335 
3336 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3337 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
3338 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > 256MB",
3339 		    pmap_pteg_table, size);
3340 #endif
3341 
3342 	memset(__UNVOLATILE(pmap_pteg_table), 0,
3343 		pmap_pteg_cnt * sizeof(struct pteg));
3344 	pmap_pteg_mask = pmap_pteg_cnt - 1;
3345 
3346 	/*
3347 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3348 	 * with pages.  So we just steal them before giving them to UVM.
3349 	 */
3350 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3351 	pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3352 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3353 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
3354 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > 256MB",
3355 		    pmap_pvo_table, size);
3356 #endif
3357 
3358 	for (i = 0; i < pmap_pteg_cnt; i++)
3359 		TAILQ_INIT(&pmap_pvo_table[i]);
3360 
3361 #ifndef MSGBUFADDR
3362 	/*
3363 	 * Allocate msgbuf in high memory.
3364 	 */
3365 	msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3366 #endif
3367 
3368 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3369 		paddr_t pfstart = atop(mp->start);
3370 		paddr_t pfend = atop(mp->start + mp->size);
3371 		if (mp->size == 0)
3372 			continue;
3373 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
3374 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3375 				VM_FREELIST_FIRST256);
3376 		} else if (mp->start >= SEGMENT_LENGTH) {
3377 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3378 				VM_FREELIST_DEFAULT);
3379 		} else {
3380 			pfend = atop(SEGMENT_LENGTH);
3381 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3382 				VM_FREELIST_FIRST256);
3383 			pfstart = atop(SEGMENT_LENGTH);
3384 			pfend = atop(mp->start + mp->size);
3385 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3386 				VM_FREELIST_DEFAULT);
3387 		}
3388 	}
3389 
3390 	/*
3391 	 * Make sure kernel vsid is allocated as well as VSID 0.
3392 	 */
3393 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3394 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3395 	pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3396 		|= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3397 	pmap_vsid_bitmap[0] |= 1;
3398 
3399 	/*
3400 	 * Initialize kernel pmap and hardware.
3401 	 */
3402 
3403 /* PMAP_OEA64_BRIDGE does support these instructions */
3404 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
3405 	for (i = 0; i < 16; i++) {
3406 #if defined(PPC_OEA601)
3407 	    /* XXX wedges for segment register 0xf , so set later */
3408 	    if ((iosrtable[i] & SR601_T) && ((MFPVR() >> 16) == MPC601))
3409 		    continue;
3410 #endif
3411  		pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3412 		__asm volatile ("mtsrin %0,%1"
3413  			      :: "r"(KERNELN_SEGMENT(i)|SR_PRKEY), "r"(i << ADDR_SR_SHFT));
3414 	}
3415 
3416 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3417 	__asm volatile ("mtsr %0,%1"
3418 		      :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
3419 #ifdef KERNEL2_SR
3420 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3421 	__asm volatile ("mtsr %0,%1"
3422 		      :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
3423 #endif
3424 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3425 #if defined (PMAP_OEA)
3426 	for (i = 0; i < 16; i++) {
3427 		if (iosrtable[i] & SR601_T) {
3428 			pmap_kernel()->pm_sr[i] = iosrtable[i];
3429 			__asm volatile ("mtsrin %0,%1"
3430 			    :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT));
3431 		}
3432 	}
3433 	__asm volatile ("sync; mtsdr1 %0; isync"
3434 		      :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10)));
3435 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
3436  	__asm __volatile ("sync; mtsdr1 %0; isync"
3437  		      :: "r"((uintptr_t)pmap_pteg_table | (32 - __builtin_clz(pmap_pteg_mask >> 11))));
3438 #endif
3439 	tlbia();
3440 
3441 #ifdef ALTIVEC
3442 	pmap_use_altivec = cpu_altivec;
3443 #endif
3444 
3445 #ifdef DEBUG
3446 	if (pmapdebug & PMAPDEBUG_BOOT) {
3447 		u_int cnt;
3448 		int bank;
3449 		char pbuf[9];
3450 		for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) {
3451 			cnt += VM_PHYSMEM_PTR(bank)->avail_end - VM_PHYSMEM_PTR(bank)->avail_start;
3452 			printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3453 			    bank,
3454 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_start),
3455 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_end),
3456 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_end - VM_PHYSMEM_PTR(bank)->avail_start));
3457 		}
3458 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3459 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3460 		    pbuf, cnt);
3461 	}
3462 #endif
3463 
3464 	pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry),
3465 	    sizeof(struct pvo_entry), 0, 0, "pmap_upvopl",
3466 	    &pmap_pool_uallocator, IPL_VM);
3467 
3468 	pool_setlowat(&pmap_upvo_pool, 252);
3469 
3470 	pool_init(&pmap_pool, sizeof(struct pmap),
3471 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator,
3472 	    IPL_NONE);
3473 
3474 #if defined(PMAP_NEED_MAPKERNEL)
3475 	{
3476 		struct pmap *pm = pmap_kernel();
3477 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3478 		extern int etext[], kernel_text[];
3479 		vaddr_t va, va_etext = (paddr_t) etext;
3480 #endif
3481 		paddr_t pa, pa_end;
3482 		register_t sr;
3483 		struct pte pt;
3484 		unsigned int ptegidx;
3485 		int bank;
3486 
3487 		sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3488 		pm->pm_sr[0] = sr;
3489 
3490 		for (bank = 0; bank < vm_nphysseg; bank++) {
3491 			pa_end = ptoa(VM_PHYSMEM_PTR(bank)->avail_end);
3492 			pa = ptoa(VM_PHYSMEM_PTR(bank)->avail_start);
3493 			for (; pa < pa_end; pa += PAGE_SIZE) {
3494 				ptegidx = va_to_pteg(pm, pa);
3495 				pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3496 				pmap_pte_insert(ptegidx, &pt);
3497 			}
3498 		}
3499 
3500 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3501 		va = (vaddr_t) kernel_text;
3502 
3503 		for (pa = kernelstart; va < va_etext;
3504 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3505 			ptegidx = va_to_pteg(pm, va);
3506 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3507 			pmap_pte_insert(ptegidx, &pt);
3508 		}
3509 
3510 		for (; pa < kernelend;
3511 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3512 			ptegidx = va_to_pteg(pm, va);
3513 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3514 			pmap_pte_insert(ptegidx, &pt);
3515 		}
3516 
3517 		for (va = 0, pa = 0; va < kernelstart;
3518 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3519 			ptegidx = va_to_pteg(pm, va);
3520 			if (va < 0x3000)
3521 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3522 			else
3523 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3524 			pmap_pte_insert(ptegidx, &pt);
3525 		}
3526 		for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3527 		    pa += PAGE_SIZE, va += PAGE_SIZE) {
3528 			ptegidx = va_to_pteg(pm, va);
3529 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3530 			pmap_pte_insert(ptegidx, &pt);
3531 		}
3532 #endif
3533 
3534 		__asm volatile ("mtsrin %0,%1"
3535  			      :: "r"(sr), "r"(kernelstart));
3536 	}
3537 #endif
3538 
3539 #if defined(PMAPDEBUG)
3540 	if ( pmapdebug )
3541 	    pmap_print_mmuregs();
3542 #endif
3543 }
3544