xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision abb0f93cd77b67f080613360c65701f85e5f5cfe)
1 /*	$NetBSD: pmap.c,v 1.69 2009/11/21 17:40:29 rmind 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 1.69 2009/11/21 17:40:29 rmind 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/malloc.h>
77 #include <sys/proc.h>
78 #include <sys/pool.h>
79 #include <sys/queue.h>
80 #include <sys/device.h>		/* for evcnt */
81 #include <sys/systm.h>
82 #include <sys/atomic.h>
83 
84 #include <uvm/uvm.h>
85 
86 #include <machine/pcb.h>
87 #include <machine/powerpc.h>
88 #include <powerpc/spr.h>
89 #include <powerpc/oea/sr_601.h>
90 #include <powerpc/bat.h>
91 #include <powerpc/stdarg.h>
92 
93 #ifdef ALTIVEC
94 int pmap_use_altivec;
95 #endif
96 
97 volatile struct pteg *pmap_pteg_table;
98 unsigned int pmap_pteg_cnt;
99 unsigned int pmap_pteg_mask;
100 #ifdef PMAP_MEMLIMIT
101 static paddr_t pmap_memlimit = PMAP_MEMLIMIT;
102 #else
103 static paddr_t pmap_memlimit = -PAGE_SIZE;		/* there is no limit */
104 #endif
105 
106 struct pmap kernel_pmap_;
107 unsigned int pmap_pages_stolen;
108 u_long pmap_pte_valid;
109 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
110 u_long pmap_pvo_enter_depth;
111 u_long pmap_pvo_remove_depth;
112 #endif
113 
114 #ifndef MSGBUFADDR
115 extern paddr_t msgbuf_paddr;
116 #endif
117 
118 static struct mem_region *mem, *avail;
119 static u_int mem_cnt, avail_cnt;
120 
121 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
122 # define	PMAP_OEA 1
123 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA64) && !defined(PPC_OEA64_BRIDGE)
124 #  define	PMAPNAME(name)	pmap_##name
125 # endif
126 #endif
127 
128 #if defined(PMAP_OEA64)
129 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA) && !defined(PPC_OEA64_BRIDGE)
130 #  define	PMAPNAME(name)	pmap_##name
131 # endif
132 #endif
133 
134 #if defined(PMAP_OEA64_BRIDGE)
135 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA) && !defined(PPC_OEA64)
136 #  define	PMAPNAME(name)	pmap_##name
137 # endif
138 #endif
139 
140 #if defined(PMAP_OEA)
141 #define	_PRIxpte	"lx"
142 #else
143 #define	_PRIxpte	PRIx64
144 #endif
145 #define	_PRIxpa		"lx"
146 #define	_PRIxva		"lx"
147 #define	_PRIsr  	"lx"
148 
149 #if defined(PMAP_EXCLUDE_DECLS) && !defined(PMAPNAME)
150 #if defined(PMAP_OEA)
151 #define	PMAPNAME(name)	pmap32_##name
152 #elif defined(PMAP_OEA64)
153 #define	PMAPNAME(name)	pmap64_##name
154 #elif defined(PMAP_OEA64_BRIDGE)
155 #define	PMAPNAME(name)	pmap64bridge_##name
156 #else
157 #error unknown variant for pmap
158 #endif
159 #endif /* PMAP_EXLCUDE_DECLS && !PMAPNAME */
160 
161 #if defined(PMAPNAME)
162 #define	STATIC			static
163 #define pmap_pte_spill		PMAPNAME(pte_spill)
164 #define pmap_real_memory	PMAPNAME(real_memory)
165 #define pmap_init		PMAPNAME(init)
166 #define pmap_virtual_space	PMAPNAME(virtual_space)
167 #define pmap_create		PMAPNAME(create)
168 #define pmap_reference		PMAPNAME(reference)
169 #define pmap_destroy		PMAPNAME(destroy)
170 #define pmap_copy		PMAPNAME(copy)
171 #define pmap_update		PMAPNAME(update)
172 #define pmap_enter		PMAPNAME(enter)
173 #define pmap_remove		PMAPNAME(remove)
174 #define pmap_kenter_pa		PMAPNAME(kenter_pa)
175 #define pmap_kremove		PMAPNAME(kremove)
176 #define pmap_extract		PMAPNAME(extract)
177 #define pmap_protect		PMAPNAME(protect)
178 #define pmap_unwire		PMAPNAME(unwire)
179 #define pmap_page_protect	PMAPNAME(page_protect)
180 #define pmap_query_bit		PMAPNAME(query_bit)
181 #define pmap_clear_bit		PMAPNAME(clear_bit)
182 
183 #define pmap_activate		PMAPNAME(activate)
184 #define pmap_deactivate		PMAPNAME(deactivate)
185 
186 #define pmap_pinit		PMAPNAME(pinit)
187 #define pmap_procwr		PMAPNAME(procwr)
188 
189 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
190 #define pmap_pte_print		PMAPNAME(pte_print)
191 #define pmap_pteg_check		PMAPNAME(pteg_check)
192 #define pmap_print_mmruregs	PMAPNAME(print_mmuregs)
193 #define pmap_print_pte		PMAPNAME(print_pte)
194 #define pmap_pteg_dist		PMAPNAME(pteg_dist)
195 #endif
196 #if defined(DEBUG) || defined(PMAPCHECK)
197 #define	pmap_pvo_verify		PMAPNAME(pvo_verify)
198 #define pmapcheck		PMAPNAME(check)
199 #endif
200 #if defined(DEBUG) || defined(PMAPDEBUG)
201 #define pmapdebug		PMAPNAME(debug)
202 #endif
203 #define pmap_steal_memory	PMAPNAME(steal_memory)
204 #define pmap_bootstrap		PMAPNAME(bootstrap)
205 #else
206 #define	STATIC			/* nothing */
207 #endif /* PMAPNAME */
208 
209 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
210 STATIC void pmap_real_memory(paddr_t *, psize_t *);
211 STATIC void pmap_init(void);
212 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
213 STATIC pmap_t pmap_create(void);
214 STATIC void pmap_reference(pmap_t);
215 STATIC void pmap_destroy(pmap_t);
216 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
217 STATIC void pmap_update(pmap_t);
218 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
219 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
220 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t, u_int);
221 STATIC void pmap_kremove(vaddr_t, vsize_t);
222 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
223 
224 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
225 STATIC void pmap_unwire(pmap_t, vaddr_t);
226 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
227 STATIC bool pmap_query_bit(struct vm_page *, int);
228 STATIC bool pmap_clear_bit(struct vm_page *, int);
229 
230 STATIC void pmap_activate(struct lwp *);
231 STATIC void pmap_deactivate(struct lwp *);
232 
233 STATIC void pmap_pinit(pmap_t pm);
234 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
235 
236 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
237 STATIC void pmap_pte_print(volatile struct pte *);
238 STATIC void pmap_pteg_check(void);
239 STATIC void pmap_print_mmuregs(void);
240 STATIC void pmap_print_pte(pmap_t, vaddr_t);
241 STATIC void pmap_pteg_dist(void);
242 #endif
243 #if defined(DEBUG) || defined(PMAPCHECK)
244 STATIC void pmap_pvo_verify(void);
245 #endif
246 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
247 STATIC void pmap_bootstrap(paddr_t, paddr_t);
248 
249 #ifdef PMAPNAME
250 const struct pmap_ops PMAPNAME(ops) = {
251 	.pmapop_pte_spill = pmap_pte_spill,
252 	.pmapop_real_memory = pmap_real_memory,
253 	.pmapop_init = pmap_init,
254 	.pmapop_virtual_space = pmap_virtual_space,
255 	.pmapop_create = pmap_create,
256 	.pmapop_reference = pmap_reference,
257 	.pmapop_destroy = pmap_destroy,
258 	.pmapop_copy = pmap_copy,
259 	.pmapop_update = pmap_update,
260 	.pmapop_enter = pmap_enter,
261 	.pmapop_remove = pmap_remove,
262 	.pmapop_kenter_pa = pmap_kenter_pa,
263 	.pmapop_kremove = pmap_kremove,
264 	.pmapop_extract = pmap_extract,
265 	.pmapop_protect = pmap_protect,
266 	.pmapop_unwire = pmap_unwire,
267 	.pmapop_page_protect = pmap_page_protect,
268 	.pmapop_query_bit = pmap_query_bit,
269 	.pmapop_clear_bit = pmap_clear_bit,
270 	.pmapop_activate = pmap_activate,
271 	.pmapop_deactivate = pmap_deactivate,
272 	.pmapop_pinit = pmap_pinit,
273 	.pmapop_procwr = pmap_procwr,
274 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
275 	.pmapop_pte_print = pmap_pte_print,
276 	.pmapop_pteg_check = pmap_pteg_check,
277 	.pmapop_print_mmuregs = pmap_print_mmuregs,
278 	.pmapop_print_pte = pmap_print_pte,
279 	.pmapop_pteg_dist = pmap_pteg_dist,
280 #else
281 	.pmapop_pte_print = NULL,
282 	.pmapop_pteg_check = NULL,
283 	.pmapop_print_mmuregs = NULL,
284 	.pmapop_print_pte = NULL,
285 	.pmapop_pteg_dist = NULL,
286 #endif
287 #if defined(DEBUG) || defined(PMAPCHECK)
288 	.pmapop_pvo_verify = pmap_pvo_verify,
289 #else
290 	.pmapop_pvo_verify = NULL,
291 #endif
292 	.pmapop_steal_memory = pmap_steal_memory,
293 	.pmapop_bootstrap = pmap_bootstrap,
294 };
295 #endif /* !PMAPNAME */
296 
297 /*
298  * The following structure is aligned to 32 bytes
299  */
300 struct pvo_entry {
301 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
302 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
303 	struct pte pvo_pte;			/* Prebuilt PTE */
304 	pmap_t pvo_pmap;			/* ptr to owning pmap */
305 	vaddr_t pvo_vaddr;			/* VA of entry */
306 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
307 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
308 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
309 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
310 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
311 #define	PVO_WIRED_P(pvo)	((pvo)->pvo_vaddr & PVO_WIRED)
312 #define	PVO_MANAGED_P(pvo)	((pvo)->pvo_vaddr & PVO_MANAGED)
313 #define	PVO_EXECUTABLE_P(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
314 #define	PVO_ENTER_INSERT	0		/* PVO has been removed */
315 #define	PVO_SPILL_UNSET		1		/* PVO has been evicted */
316 #define	PVO_SPILL_SET		2		/* PVO has been spilled */
317 #define	PVO_SPILL_INSERT	3		/* PVO has been inserted */
318 #define	PVO_PMAP_PAGE_PROTECT	4		/* PVO has changed */
319 #define	PVO_PMAP_PROTECT	5		/* PVO has changed */
320 #define	PVO_REMOVE		6		/* PVO has been removed */
321 #define	PVO_WHERE_MASK		15
322 #define	PVO_WHERE_SHFT		8
323 } __attribute__ ((aligned (32)));
324 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
325 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
326 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
327 #define	PVO_PTEGIDX_CLR(pvo)	\
328 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
329 #define	PVO_PTEGIDX_SET(pvo,i)	\
330 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
331 #define	PVO_WHERE(pvo,w)	\
332 	((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
333 	 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
334 
335 TAILQ_HEAD(pvo_tqhead, pvo_entry);
336 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
337 static struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
338 static struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
339 
340 struct pool pmap_pool;		/* pool for pmap structures */
341 struct pool pmap_upvo_pool;	/* pool for pvo entries for unmanaged pages */
342 struct pool pmap_mpvo_pool;	/* pool for pvo entries for managed pages */
343 
344 /*
345  * We keep a cache of unmanaged pages to be used for pvo entries for
346  * unmanaged pages.
347  */
348 struct pvo_page {
349 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
350 };
351 SIMPLEQ_HEAD(pvop_head, pvo_page);
352 static struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head);
353 static struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head);
354 u_long pmap_upvop_free;
355 u_long pmap_upvop_maxfree;
356 u_long pmap_mpvop_free;
357 u_long pmap_mpvop_maxfree;
358 
359 static void *pmap_pool_ualloc(struct pool *, int);
360 static void *pmap_pool_malloc(struct pool *, int);
361 
362 static void pmap_pool_ufree(struct pool *, void *);
363 static void pmap_pool_mfree(struct pool *, void *);
364 
365 static struct pool_allocator pmap_pool_mallocator = {
366 	.pa_alloc = pmap_pool_malloc,
367 	.pa_free = pmap_pool_mfree,
368 	.pa_pagesz = 0,
369 };
370 
371 static struct pool_allocator pmap_pool_uallocator = {
372 	.pa_alloc = pmap_pool_ualloc,
373 	.pa_free = pmap_pool_ufree,
374 	.pa_pagesz = 0,
375 };
376 
377 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
378 void pmap_pte_print(volatile struct pte *);
379 void pmap_pteg_check(void);
380 void pmap_pteg_dist(void);
381 void pmap_print_pte(pmap_t, vaddr_t);
382 void pmap_print_mmuregs(void);
383 #endif
384 
385 #if defined(DEBUG) || defined(PMAPCHECK)
386 #ifdef PMAPCHECK
387 int pmapcheck = 1;
388 #else
389 int pmapcheck = 0;
390 #endif
391 void pmap_pvo_verify(void);
392 static void pmap_pvo_check(const struct pvo_entry *);
393 #define	PMAP_PVO_CHECK(pvo)	 		\
394 	do {					\
395 		if (pmapcheck)			\
396 			pmap_pvo_check(pvo);	\
397 	} while (0)
398 #else
399 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
400 #endif
401 static int pmap_pte_insert(int, struct pte *);
402 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
403 	vaddr_t, paddr_t, register_t, int);
404 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
405 static void pmap_pvo_free(struct pvo_entry *);
406 static void pmap_pvo_free_list(struct pvo_head *);
407 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
408 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
409 static struct pvo_entry *pmap_pvo_reclaim(struct pmap *);
410 static void pvo_set_exec(struct pvo_entry *);
411 static void pvo_clear_exec(struct pvo_entry *);
412 
413 static void tlbia(void);
414 
415 static void pmap_release(pmap_t);
416 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
417 
418 static uint32_t pmap_pvo_reclaim_nextidx;
419 #ifdef DEBUG
420 static int pmap_pvo_reclaim_debugctr;
421 #endif
422 
423 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
424 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
425 
426 static int pmap_initialized;
427 
428 #if defined(DEBUG) || defined(PMAPDEBUG)
429 #define	PMAPDEBUG_BOOT		0x0001
430 #define	PMAPDEBUG_PTE		0x0002
431 #define	PMAPDEBUG_EXEC		0x0008
432 #define	PMAPDEBUG_PVOENTER	0x0010
433 #define	PMAPDEBUG_PVOREMOVE	0x0020
434 #define	PMAPDEBUG_ACTIVATE	0x0100
435 #define	PMAPDEBUG_CREATE	0x0200
436 #define	PMAPDEBUG_ENTER		0x1000
437 #define	PMAPDEBUG_KENTER	0x2000
438 #define	PMAPDEBUG_KREMOVE	0x4000
439 #define	PMAPDEBUG_REMOVE	0x8000
440 
441 unsigned int pmapdebug = 0;
442 
443 # define DPRINTF(x)		printf x
444 # define DPRINTFN(n, x)		if (pmapdebug & PMAPDEBUG_ ## n) printf x
445 #else
446 # define DPRINTF(x)
447 # define DPRINTFN(n, x)
448 #endif
449 
450 
451 #ifdef PMAPCOUNTERS
452 /*
453  * From pmap_subr.c
454  */
455 extern struct evcnt pmap_evcnt_mappings;
456 extern struct evcnt pmap_evcnt_unmappings;
457 
458 extern struct evcnt pmap_evcnt_kernel_mappings;
459 extern struct evcnt pmap_evcnt_kernel_unmappings;
460 
461 extern struct evcnt pmap_evcnt_mappings_replaced;
462 
463 extern struct evcnt pmap_evcnt_exec_mappings;
464 extern struct evcnt pmap_evcnt_exec_cached;
465 
466 extern struct evcnt pmap_evcnt_exec_synced;
467 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
468 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
469 
470 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
471 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
472 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
473 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
474 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
475 
476 extern struct evcnt pmap_evcnt_updates;
477 extern struct evcnt pmap_evcnt_collects;
478 extern struct evcnt pmap_evcnt_copies;
479 
480 extern struct evcnt pmap_evcnt_ptes_spilled;
481 extern struct evcnt pmap_evcnt_ptes_unspilled;
482 extern struct evcnt pmap_evcnt_ptes_evicted;
483 
484 extern struct evcnt pmap_evcnt_ptes_primary[8];
485 extern struct evcnt pmap_evcnt_ptes_secondary[8];
486 extern struct evcnt pmap_evcnt_ptes_removed;
487 extern struct evcnt pmap_evcnt_ptes_changed;
488 extern struct evcnt pmap_evcnt_pvos_reclaimed;
489 extern struct evcnt pmap_evcnt_pvos_failed;
490 
491 extern struct evcnt pmap_evcnt_zeroed_pages;
492 extern struct evcnt pmap_evcnt_copied_pages;
493 extern struct evcnt pmap_evcnt_idlezeroed_pages;
494 
495 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
496 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
497 #else
498 #define	PMAPCOUNT(ev)	((void) 0)
499 #define	PMAPCOUNT2(ev)	((void) 0)
500 #endif
501 
502 #define	TLBIE(va)	__asm volatile("tlbie %0" :: "r"(va))
503 
504 /* XXXSL: this needs to be moved to assembler */
505 #define	TLBIEL(va)	__asm __volatile("tlbie %0" :: "r"(va))
506 
507 #define	TLBSYNC()	__asm volatile("tlbsync")
508 #define	SYNC()		__asm volatile("sync")
509 #define	EIEIO()		__asm volatile("eieio")
510 #define	DCBST(va)	__asm __volatile("dcbst 0,%0" :: "r"(va))
511 #define	MFMSR()		mfmsr()
512 #define	MTMSR(psl)	mtmsr(psl)
513 #define	MFPVR()		mfpvr()
514 #define	MFSRIN(va)	mfsrin(va)
515 #define	MFTB()		mfrtcltbl()
516 
517 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
518 static inline register_t
519 mfsrin(vaddr_t va)
520 {
521 	register_t sr;
522 	__asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
523 	return sr;
524 }
525 #endif	/* PMAP_OEA*/
526 
527 #if defined (PMAP_OEA64_BRIDGE)
528 extern void mfmsr64 (register64_t *result);
529 #endif /* PMAP_OEA64_BRIDGE */
530 
531 #define	PMAP_LOCK()		KERNEL_LOCK(1, NULL)
532 #define	PMAP_UNLOCK()		KERNEL_UNLOCK_ONE(NULL)
533 
534 static inline register_t
535 pmap_interrupts_off(void)
536 {
537 	register_t msr = MFMSR();
538 	if (msr & PSL_EE)
539 		MTMSR(msr & ~PSL_EE);
540 	return msr;
541 }
542 
543 static void
544 pmap_interrupts_restore(register_t msr)
545 {
546 	if (msr & PSL_EE)
547 		MTMSR(msr);
548 }
549 
550 static inline u_int32_t
551 mfrtcltbl(void)
552 {
553 #ifdef PPC_OEA601
554 	if ((MFPVR() >> 16) == MPC601)
555 		return (mfrtcl() >> 7);
556 	else
557 #endif
558 		return (mftbl());
559 }
560 
561 /*
562  * These small routines may have to be replaced,
563  * if/when we support processors other that the 604.
564  */
565 
566 void
567 tlbia(void)
568 {
569 	char *i;
570 
571 	SYNC();
572 #if defined(PMAP_OEA)
573 	/*
574 	 * Why not use "tlbia"?  Because not all processors implement it.
575 	 *
576 	 * This needs to be a per-CPU callback to do the appropriate thing
577 	 * for the CPU. XXX
578 	 */
579 	for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
580 		TLBIE(i);
581 		EIEIO();
582 		SYNC();
583 	}
584 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
585 	/* This is specifically for the 970, 970UM v1.6 pp. 140. */
586 	for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
587 		TLBIEL(i);
588 		EIEIO();
589 		SYNC();
590 	}
591 #endif
592 	TLBSYNC();
593 	SYNC();
594 }
595 
596 static inline register_t
597 va_to_vsid(const struct pmap *pm, vaddr_t addr)
598 {
599 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
600 	return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT;
601 #else /* PMAP_OEA64 */
602 #if 0
603 	const struct ste *ste;
604 	register_t hash;
605 	int i;
606 
607 	hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH;
608 
609 	/*
610 	 * Try the primary group first
611 	 */
612 	ste = pm->pm_stes[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 ste;
617 	}
618 
619 	/*
620 	 * Then the secondary group.
621 	 */
622 	ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes;
623 	for (i = 0; i < 8; i++, ste++) {
624 		if (ste->ste_hi & STE_V) &&
625 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
626 			return addr;
627 	}
628 
629 	return NULL;
630 #else
631 	/*
632 	 * Rather than searching the STE groups for the VSID, we know
633 	 * how we generate that from the ESID and so do that.
634 	 */
635 	return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
636 #endif
637 #endif /* PMAP_OEA */
638 }
639 
640 static inline register_t
641 va_to_pteg(const struct pmap *pm, vaddr_t addr)
642 {
643 	register_t hash;
644 
645 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
646 	return hash & pmap_pteg_mask;
647 }
648 
649 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
650 /*
651  * Given a PTE in the page table, calculate the VADDR that hashes to it.
652  * The only bit of magic is that the top 4 bits of the address doesn't
653  * technically exist in the PTE.  But we know we reserved 4 bits of the
654  * VSID for it so that's how we get it.
655  */
656 static vaddr_t
657 pmap_pte_to_va(volatile const struct pte *pt)
658 {
659 	vaddr_t va;
660 	uintptr_t ptaddr = (uintptr_t) pt;
661 
662 	if (pt->pte_hi & PTE_HID)
663 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
664 
665 	/* PPC Bits 10-19  PPC64 Bits 42-51 */
666 #if defined(PMAP_OEA)
667 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
668 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
669 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
670 #endif
671 	va <<= ADDR_PIDX_SHFT;
672 
673 	/* PPC Bits 4-9  PPC64 Bits 36-41 */
674 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
675 
676 #if defined(PMAP_OEA64)
677 	/* PPC63 Bits 0-35 */
678 	/* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
679 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
680 	/* PPC Bits 0-3 */
681 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
682 #endif
683 
684 	return va;
685 }
686 #endif
687 
688 static inline struct pvo_head *
689 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
690 {
691 	struct vm_page *pg;
692 
693 	pg = PHYS_TO_VM_PAGE(pa);
694 	if (pg_p != NULL)
695 		*pg_p = pg;
696 	if (pg == NULL)
697 		return &pmap_pvo_unmanaged;
698 	return &pg->mdpage.mdpg_pvoh;
699 }
700 
701 static inline struct pvo_head *
702 vm_page_to_pvoh(struct vm_page *pg)
703 {
704 	return &pg->mdpage.mdpg_pvoh;
705 }
706 
707 
708 static inline void
709 pmap_attr_clear(struct vm_page *pg, int ptebit)
710 {
711 	pg->mdpage.mdpg_attrs &= ~ptebit;
712 }
713 
714 static inline int
715 pmap_attr_fetch(struct vm_page *pg)
716 {
717 	return pg->mdpage.mdpg_attrs;
718 }
719 
720 static inline void
721 pmap_attr_save(struct vm_page *pg, int ptebit)
722 {
723 	pg->mdpage.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)
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 #elif defined (PMAP_OEA64)
757 #error PMAP_OEA64 not supported
758 #endif /* PMAP_OEA */
759 }
760 
761 static inline void
762 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
763 {
764 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
765 }
766 
767 static inline void
768 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
769 {
770 	/*
771 	 * As shown in Section 7.6.3.2.3
772 	 */
773 	pt->pte_lo &= ~ptebit;
774 	TLBIE(va);
775 	SYNC();
776 	EIEIO();
777 	TLBSYNC();
778 	SYNC();
779 #ifdef MULTIPROCESSOR
780 	DCBST(pt);
781 #endif
782 }
783 
784 static inline void
785 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
786 {
787 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
788 	if (pvo_pt->pte_hi & PTE_VALID)
789 		panic("pte_set: setting an already valid pte %p", pvo_pt);
790 #endif
791 	pvo_pt->pte_hi |= PTE_VALID;
792 
793 	/*
794 	 * Update the PTE as defined in section 7.6.3.1
795 	 * Note that the REF/CHG bits are from pvo_pt and thus should
796 	 * have been saved so this routine can restore them (if desired).
797 	 */
798 	pt->pte_lo = pvo_pt->pte_lo;
799 	EIEIO();
800 	pt->pte_hi = pvo_pt->pte_hi;
801 	TLBSYNC();
802 	SYNC();
803 #ifdef MULTIPROCESSOR
804 	DCBST(pt);
805 #endif
806 	pmap_pte_valid++;
807 }
808 
809 static inline void
810 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
811 {
812 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
813 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
814 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
815 	if ((pt->pte_hi & PTE_VALID) == 0)
816 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
817 #endif
818 
819 	pvo_pt->pte_hi &= ~PTE_VALID;
820 	/*
821 	 * Force the ref & chg bits back into the PTEs.
822 	 */
823 	SYNC();
824 	/*
825 	 * Invalidate the pte ... (Section 7.6.3.3)
826 	 */
827 	pt->pte_hi &= ~PTE_VALID;
828 	SYNC();
829 	TLBIE(va);
830 	SYNC();
831 	EIEIO();
832 	TLBSYNC();
833 	SYNC();
834 	/*
835 	 * Save the ref & chg bits ...
836 	 */
837 	pmap_pte_synch(pt, pvo_pt);
838 	pmap_pte_valid--;
839 }
840 
841 static inline void
842 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
843 {
844 	/*
845 	 * Invalidate the PTE
846 	 */
847 	pmap_pte_unset(pt, pvo_pt, va);
848 	pmap_pte_set(pt, pvo_pt);
849 }
850 
851 /*
852  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
853  * (either primary or secondary location).
854  *
855  * Note: both the destination and source PTEs must not have PTE_VALID set.
856  */
857 
858 static int
859 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
860 {
861 	volatile struct pte *pt;
862 	int i;
863 
864 #if defined(DEBUG)
865 	DPRINTFN(PTE, ("pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
866 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo));
867 #endif
868 	/*
869 	 * First try primary hash.
870 	 */
871 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
872 		if ((pt->pte_hi & PTE_VALID) == 0) {
873 			pvo_pt->pte_hi &= ~PTE_HID;
874 			pmap_pte_set(pt, pvo_pt);
875 			return i;
876 		}
877 	}
878 
879 	/*
880 	 * Now try secondary hash.
881 	 */
882 	ptegidx ^= pmap_pteg_mask;
883 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
884 		if ((pt->pte_hi & PTE_VALID) == 0) {
885 			pvo_pt->pte_hi |= PTE_HID;
886 			pmap_pte_set(pt, pvo_pt);
887 			return i;
888 		}
889 	}
890 	return -1;
891 }
892 
893 /*
894  * Spill handler.
895  *
896  * Tries to spill a page table entry from the overflow area.
897  * This runs in either real mode (if dealing with a exception spill)
898  * or virtual mode when dealing with manually spilling one of the
899  * kernel's pte entries.  In either case, interrupts are already
900  * disabled.
901  */
902 
903 int
904 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec)
905 {
906 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
907 	struct pvo_entry *pvo;
908 	/* XXX: gcc -- vpvoh is always set at either *1* or *2* */
909 	struct pvo_tqhead *pvoh, *vpvoh = NULL;
910 	int ptegidx, i, j;
911 	volatile struct pteg *pteg;
912 	volatile struct pte *pt;
913 
914 	PMAP_LOCK();
915 
916 	ptegidx = va_to_pteg(pm, addr);
917 
918 	/*
919 	 * Have to substitute some entry. Use the primary hash for this.
920 	 * Use low bits of timebase as random generator.  Make sure we are
921 	 * not picking a kernel pte for replacement.
922 	 */
923 	pteg = &pmap_pteg_table[ptegidx];
924 	i = MFTB() & 7;
925 	for (j = 0; j < 8; j++) {
926 		pt = &pteg->pt[i];
927 		if ((pt->pte_hi & PTE_VALID) == 0)
928 			break;
929 		if (VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT)
930 				< PHYSMAP_VSIDBITS)
931 			break;
932 		i = (i + 1) & 7;
933 	}
934 	KASSERT(j < 8);
935 
936 	source_pvo = NULL;
937 	victim_pvo = NULL;
938 	pvoh = &pmap_pvo_table[ptegidx];
939 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
940 
941 		/*
942 		 * We need to find pvo entry for this address...
943 		 */
944 		PMAP_PVO_CHECK(pvo);		/* sanity check */
945 
946 		/*
947 		 * If we haven't found the source and we come to a PVO with
948 		 * a valid PTE, then we know we can't find it because all
949 		 * evicted PVOs always are first in the list.
950 		 */
951 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
952 			break;
953 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
954 		    addr == PVO_VADDR(pvo)) {
955 
956 			/*
957 			 * Now we have found the entry to be spilled into the
958 			 * pteg.  Attempt to insert it into the page table.
959 			 */
960 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
961 			if (j >= 0) {
962 				PVO_PTEGIDX_SET(pvo, j);
963 				PMAP_PVO_CHECK(pvo);	/* sanity check */
964 				PVO_WHERE(pvo, SPILL_INSERT);
965 				pvo->pvo_pmap->pm_evictions--;
966 				PMAPCOUNT(ptes_spilled);
967 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
968 				    ? pmap_evcnt_ptes_secondary
969 				    : pmap_evcnt_ptes_primary)[j]);
970 
971 				/*
972 				 * Since we keep the evicted entries at the
973 				 * from of the PVO list, we need move this
974 				 * (now resident) PVO after the evicted
975 				 * entries.
976 				 */
977 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
978 
979 				/*
980 				 * If we don't have to move (either we were the
981 				 * last entry or the next entry was valid),
982 				 * don't change our position.  Otherwise
983 				 * move ourselves to the tail of the queue.
984 				 */
985 				if (next_pvo != NULL &&
986 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
987 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
988 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
989 				}
990 				PMAP_UNLOCK();
991 				return 1;
992 			}
993 			source_pvo = pvo;
994 			if (exec && !PVO_EXECUTABLE_P(source_pvo)) {
995 				return 0;
996 			}
997 			if (victim_pvo != NULL)
998 				break;
999 		}
1000 
1001 		/*
1002 		 * We also need the pvo entry of the victim we are replacing
1003 		 * so save the R & C bits of the PTE.
1004 		 */
1005 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
1006 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
1007 			vpvoh = pvoh;			/* *1* */
1008 			victim_pvo = pvo;
1009 			if (source_pvo != NULL)
1010 				break;
1011 		}
1012 	}
1013 
1014 	if (source_pvo == NULL) {
1015 		PMAPCOUNT(ptes_unspilled);
1016 		PMAP_UNLOCK();
1017 		return 0;
1018 	}
1019 
1020 	if (victim_pvo == NULL) {
1021 		if ((pt->pte_hi & PTE_HID) == 0)
1022 			panic("pmap_pte_spill: victim p-pte (%p) has "
1023 			    "no pvo entry!", pt);
1024 
1025 		/*
1026 		 * If this is a secondary PTE, we need to search
1027 		 * its primary pvo bucket for the matching PVO.
1028 		 */
1029 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */
1030 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1031 			PMAP_PVO_CHECK(pvo);		/* sanity check */
1032 
1033 			/*
1034 			 * We also need the pvo entry of the victim we are
1035 			 * replacing so save the R & C bits of the PTE.
1036 			 */
1037 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1038 				victim_pvo = pvo;
1039 				break;
1040 			}
1041 		}
1042 		if (victim_pvo == NULL)
1043 			panic("pmap_pte_spill: victim s-pte (%p) has "
1044 			    "no pvo entry!", pt);
1045 	}
1046 
1047 	/*
1048 	 * The victim should be not be a kernel PVO/PTE entry.
1049 	 */
1050 	KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1051 	KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1052 	KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1053 
1054 	/*
1055 	 * We are invalidating the TLB entry for the EA for the
1056 	 * we are replacing even though its valid; If we don't
1057 	 * we lose any ref/chg bit changes contained in the TLB
1058 	 * entry.
1059 	 */
1060 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1061 
1062 	/*
1063 	 * To enforce the PVO list ordering constraint that all
1064 	 * evicted entries should come before all valid entries,
1065 	 * move the source PVO to the tail of its list and the
1066 	 * victim PVO to the head of its list (which might not be
1067 	 * the same list, if the victim was using the secondary hash).
1068 	 */
1069 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
1070 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
1071 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1072 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1073 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1074 	pmap_pte_set(pt, &source_pvo->pvo_pte);
1075 	victim_pvo->pvo_pmap->pm_evictions++;
1076 	source_pvo->pvo_pmap->pm_evictions--;
1077 	PVO_WHERE(victim_pvo, SPILL_UNSET);
1078 	PVO_WHERE(source_pvo, SPILL_SET);
1079 
1080 	PVO_PTEGIDX_CLR(victim_pvo);
1081 	PVO_PTEGIDX_SET(source_pvo, i);
1082 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1083 	PMAPCOUNT(ptes_spilled);
1084 	PMAPCOUNT(ptes_evicted);
1085 	PMAPCOUNT(ptes_removed);
1086 
1087 	PMAP_PVO_CHECK(victim_pvo);
1088 	PMAP_PVO_CHECK(source_pvo);
1089 
1090 	PMAP_UNLOCK();
1091 	return 1;
1092 }
1093 
1094 /*
1095  * Restrict given range to physical memory
1096  */
1097 void
1098 pmap_real_memory(paddr_t *start, psize_t *size)
1099 {
1100 	struct mem_region *mp;
1101 
1102 	for (mp = mem; mp->size; mp++) {
1103 		if (*start + *size > mp->start
1104 		    && *start < mp->start + mp->size) {
1105 			if (*start < mp->start) {
1106 				*size -= mp->start - *start;
1107 				*start = mp->start;
1108 			}
1109 			if (*start + *size > mp->start + mp->size)
1110 				*size = mp->start + mp->size - *start;
1111 			return;
1112 		}
1113 	}
1114 	*size = 0;
1115 }
1116 
1117 /*
1118  * Initialize anything else for pmap handling.
1119  * Called during vm_init().
1120  */
1121 void
1122 pmap_init(void)
1123 {
1124 	pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry),
1125 	    sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl",
1126 	    &pmap_pool_mallocator, IPL_NONE);
1127 
1128 	pool_setlowat(&pmap_mpvo_pool, 1008);
1129 
1130 	pmap_initialized = 1;
1131 
1132 }
1133 
1134 /*
1135  * How much virtual space does the kernel get?
1136  */
1137 void
1138 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1139 {
1140 	/*
1141 	 * For now, reserve one segment (minus some overhead) for kernel
1142 	 * virtual memory
1143 	 */
1144 	*start = VM_MIN_KERNEL_ADDRESS;
1145 	*end = VM_MAX_KERNEL_ADDRESS;
1146 }
1147 
1148 /*
1149  * Allocate, initialize, and return a new physical map.
1150  */
1151 pmap_t
1152 pmap_create(void)
1153 {
1154 	pmap_t pm;
1155 
1156 	pm = pool_get(&pmap_pool, PR_WAITOK);
1157 	memset((void *)pm, 0, sizeof *pm);
1158 	pmap_pinit(pm);
1159 
1160 	DPRINTFN(CREATE,("pmap_create: pm %p:\n"
1161 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1162 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1163 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1164 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1165 	    pm,
1166 	    pm->pm_sr[0], pm->pm_sr[1],
1167 	    pm->pm_sr[2], pm->pm_sr[3],
1168 	    pm->pm_sr[4], pm->pm_sr[5],
1169 	    pm->pm_sr[6], pm->pm_sr[7],
1170 	    pm->pm_sr[8], pm->pm_sr[9],
1171 	    pm->pm_sr[10], pm->pm_sr[11],
1172 	    pm->pm_sr[12], pm->pm_sr[13],
1173 	    pm->pm_sr[14], pm->pm_sr[15]));
1174 	return pm;
1175 }
1176 
1177 /*
1178  * Initialize a preallocated and zeroed pmap structure.
1179  */
1180 void
1181 pmap_pinit(pmap_t pm)
1182 {
1183 	register_t entropy = MFTB();
1184 	register_t mask;
1185 	int i;
1186 
1187 	/*
1188 	 * Allocate some segment registers for this pmap.
1189 	 */
1190 	pm->pm_refs = 1;
1191 	PMAP_LOCK();
1192 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1193 		static register_t pmap_vsidcontext;
1194 		register_t hash;
1195 		unsigned int n;
1196 
1197 		/* Create a new value by multiplying by a prime adding in
1198 		 * entropy from the timebase register.  This is to make the
1199 		 * VSID more random so that the PT Hash function collides
1200 		 * less often. (note that the prime causes gcc to do shifts
1201 		 * instead of a multiply)
1202 		 */
1203 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1204 		hash = pmap_vsidcontext & (NPMAPS - 1);
1205 		if (hash == 0) {		/* 0 is special, avoid it */
1206 			entropy += 0xbadf00d;
1207 			continue;
1208 		}
1209 		n = hash >> 5;
1210 		mask = 1L << (hash & (VSID_NBPW-1));
1211 		hash = pmap_vsidcontext;
1212 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1213 			/* anything free in this bucket? */
1214 			if (~pmap_vsid_bitmap[n] == 0) {
1215 				entropy = hash ^ (hash >> 16);
1216 				continue;
1217 			}
1218 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1219 			mask = 1L << i;
1220 			hash &= ~(VSID_NBPW-1);
1221 			hash |= i;
1222 		}
1223 		hash &= PTE_VSID >> PTE_VSID_SHFT;
1224 		pmap_vsid_bitmap[n] |= mask;
1225 		pm->pm_vsid = hash;
1226 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1227 		for (i = 0; i < 16; i++)
1228 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1229 			    SR_NOEXEC;
1230 #endif
1231 		PMAP_UNLOCK();
1232 		return;
1233 	}
1234 	PMAP_UNLOCK();
1235 	panic("pmap_pinit: out of segments");
1236 }
1237 
1238 /*
1239  * Add a reference to the given pmap.
1240  */
1241 void
1242 pmap_reference(pmap_t pm)
1243 {
1244 	atomic_inc_uint(&pm->pm_refs);
1245 }
1246 
1247 /*
1248  * Retire the given pmap from service.
1249  * Should only be called if the map contains no valid mappings.
1250  */
1251 void
1252 pmap_destroy(pmap_t pm)
1253 {
1254 	if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1255 		pmap_release(pm);
1256 		pool_put(&pmap_pool, pm);
1257 	}
1258 }
1259 
1260 /*
1261  * Release any resources held by the given physical map.
1262  * Called when a pmap initialized by pmap_pinit is being released.
1263  */
1264 void
1265 pmap_release(pmap_t pm)
1266 {
1267 	int idx, mask;
1268 
1269 	KASSERT(pm->pm_stats.resident_count == 0);
1270 	KASSERT(pm->pm_stats.wired_count == 0);
1271 
1272 	PMAP_LOCK();
1273 	if (pm->pm_sr[0] == 0)
1274 		panic("pmap_release");
1275 	idx = pm->pm_vsid & (NPMAPS-1);
1276 	mask = 1 << (idx % VSID_NBPW);
1277 	idx /= VSID_NBPW;
1278 
1279 	KASSERT(pmap_vsid_bitmap[idx] & mask);
1280 	pmap_vsid_bitmap[idx] &= ~mask;
1281 	PMAP_UNLOCK();
1282 }
1283 
1284 /*
1285  * Copy the range specified by src_addr/len
1286  * from the source map to the range dst_addr/len
1287  * in the destination map.
1288  *
1289  * This routine is only advisory and need not do anything.
1290  */
1291 void
1292 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1293 	vsize_t len, vaddr_t src_addr)
1294 {
1295 	PMAPCOUNT(copies);
1296 }
1297 
1298 /*
1299  * Require that all active physical maps contain no
1300  * incorrect entries NOW.
1301  */
1302 void
1303 pmap_update(struct pmap *pmap)
1304 {
1305 	PMAPCOUNT(updates);
1306 	TLBSYNC();
1307 }
1308 
1309 static inline int
1310 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1311 {
1312 	int pteidx;
1313 	/*
1314 	 * We can find the actual pte entry without searching by
1315 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1316 	 * and by noticing the HID bit.
1317 	 */
1318 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1319 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1320 		pteidx ^= pmap_pteg_mask * 8;
1321 	return pteidx;
1322 }
1323 
1324 volatile struct pte *
1325 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1326 {
1327 	volatile struct pte *pt;
1328 
1329 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1330 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1331 		return NULL;
1332 #endif
1333 
1334 	/*
1335 	 * If we haven't been supplied the ptegidx, calculate it.
1336 	 */
1337 	if (pteidx == -1) {
1338 		int ptegidx;
1339 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1340 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1341 	}
1342 
1343 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1344 
1345 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1346 	return pt;
1347 #else
1348 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1349 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1350 		    "pvo but no valid pte index", pvo);
1351 	}
1352 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1353 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1354 		    "pvo but no valid pte", pvo);
1355 	}
1356 
1357 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1358 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1359 #if defined(DEBUG) || defined(PMAPCHECK)
1360 			pmap_pte_print(pt);
1361 #endif
1362 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1363 			    "pmap_pteg_table %p but invalid in pvo",
1364 			    pvo, pt);
1365 		}
1366 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1367 #if defined(DEBUG) || defined(PMAPCHECK)
1368 			pmap_pte_print(pt);
1369 #endif
1370 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1371 			    "not match pte %p in pmap_pteg_table",
1372 			    pvo, pt);
1373 		}
1374 		return pt;
1375 	}
1376 
1377 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1378 #if defined(DEBUG) || defined(PMAPCHECK)
1379 		pmap_pte_print(pt);
1380 #endif
1381 		panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1382 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1383 	}
1384 	return NULL;
1385 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1386 }
1387 
1388 struct pvo_entry *
1389 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1390 {
1391 	struct pvo_entry *pvo;
1392 	int ptegidx;
1393 
1394 	va &= ~ADDR_POFF;
1395 	ptegidx = va_to_pteg(pm, va);
1396 
1397 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1398 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1399 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1400 			panic("pmap_pvo_find_va: invalid pvo %p on "
1401 			    "list %#x (%p)", pvo, ptegidx,
1402 			     &pmap_pvo_table[ptegidx]);
1403 #endif
1404 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1405 			if (pteidx_p)
1406 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1407 			return pvo;
1408 		}
1409 	}
1410 	if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH))
1411 		panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1412 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1413 	return NULL;
1414 }
1415 
1416 #if defined(DEBUG) || defined(PMAPCHECK)
1417 void
1418 pmap_pvo_check(const struct pvo_entry *pvo)
1419 {
1420 	struct pvo_head *pvo_head;
1421 	struct pvo_entry *pvo0;
1422 	volatile struct pte *pt;
1423 	int failed = 0;
1424 
1425 	PMAP_LOCK();
1426 
1427 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1428 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1429 
1430 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1431 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1432 		    pvo, pvo->pvo_pmap);
1433 		failed = 1;
1434 	}
1435 
1436 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1437 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1438 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1439 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1440 		failed = 1;
1441 	}
1442 
1443 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1444 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1445 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1446 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1447 		failed = 1;
1448 	}
1449 
1450 	if (PVO_MANAGED_P(pvo)) {
1451 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1452 	} else {
1453 		if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) {
1454 			printf("pmap_pvo_check: pvo %p: non kernel address "
1455 			    "on kernel unmanaged list\n", pvo);
1456 			failed = 1;
1457 		}
1458 		pvo_head = &pmap_pvo_kunmanaged;
1459 	}
1460 	LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1461 		if (pvo0 == pvo)
1462 			break;
1463 	}
1464 	if (pvo0 == NULL) {
1465 		printf("pmap_pvo_check: pvo %p: not present "
1466 		    "on its vlist head %p\n", pvo, pvo_head);
1467 		failed = 1;
1468 	}
1469 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1470 		printf("pmap_pvo_check: pvo %p: not present "
1471 		    "on its olist head\n", pvo);
1472 		failed = 1;
1473 	}
1474 	pt = pmap_pvo_to_pte(pvo, -1);
1475 	if (pt == NULL) {
1476 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1477 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1478 			    "no PTE\n", pvo);
1479 			failed = 1;
1480 		}
1481 	} else {
1482 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1483 		    (uintptr_t) pt >=
1484 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1485 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1486 			    "pteg table\n", pvo, pt);
1487 			failed = 1;
1488 		}
1489 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1490 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1491 			    "no PTE\n", pvo);
1492 			failed = 1;
1493 		}
1494 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1495 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1496 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1497 			    pvo->pvo_pte.pte_hi,
1498 			    pt->pte_hi);
1499 			failed = 1;
1500 		}
1501 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1502 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1503 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1504 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1505 			    (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1506 			    (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1507 			failed = 1;
1508 		}
1509 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1510 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1511 			    " doesn't not match PVO's VA %#" _PRIxva "\n",
1512 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1513 			failed = 1;
1514 		}
1515 		if (failed)
1516 			pmap_pte_print(pt);
1517 	}
1518 	if (failed)
1519 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1520 		    pvo->pvo_pmap);
1521 
1522 	PMAP_UNLOCK();
1523 }
1524 #endif /* DEBUG || PMAPCHECK */
1525 
1526 /*
1527  * Search the PVO table looking for a non-wired entry.
1528  * If we find one, remove it and return it.
1529  */
1530 
1531 struct pvo_entry *
1532 pmap_pvo_reclaim(struct pmap *pm)
1533 {
1534 	struct pvo_tqhead *pvoh;
1535 	struct pvo_entry *pvo;
1536 	uint32_t idx, endidx;
1537 
1538 	endidx = pmap_pvo_reclaim_nextidx;
1539 	for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1540 	     idx = (idx + 1) & pmap_pteg_mask) {
1541 		pvoh = &pmap_pvo_table[idx];
1542 		TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1543 			if (!PVO_WIRED_P(pvo)) {
1544 				pmap_pvo_remove(pvo, -1, NULL);
1545 				pmap_pvo_reclaim_nextidx = idx;
1546 				PMAPCOUNT(pvos_reclaimed);
1547 				return pvo;
1548 			}
1549 		}
1550 	}
1551 	return NULL;
1552 }
1553 
1554 /*
1555  * This returns whether this is the first mapping of a page.
1556  */
1557 int
1558 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1559 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1560 {
1561 	struct pvo_entry *pvo;
1562 	struct pvo_tqhead *pvoh;
1563 	register_t msr;
1564 	int ptegidx;
1565 	int i;
1566 	int poolflags = PR_NOWAIT;
1567 
1568 	/*
1569 	 * Compute the PTE Group index.
1570 	 */
1571 	va &= ~ADDR_POFF;
1572 	ptegidx = va_to_pteg(pm, va);
1573 
1574 	msr = pmap_interrupts_off();
1575 
1576 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1577 	if (pmap_pvo_remove_depth > 0)
1578 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1579 	if (++pmap_pvo_enter_depth > 1)
1580 		panic("pmap_pvo_enter: called recursively!");
1581 #endif
1582 
1583 	/*
1584 	 * Remove any existing mapping for this page.  Reuse the
1585 	 * pvo entry if there a mapping.
1586 	 */
1587 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1588 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1589 #ifdef DEBUG
1590 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1591 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1592 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1593 			   va < VM_MIN_KERNEL_ADDRESS) {
1594 				printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1595 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1596 				printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1597 				    pvo->pvo_pte.pte_hi,
1598 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1599 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1600 #ifdef DDBX
1601 				Debugger();
1602 #endif
1603 			}
1604 #endif
1605 			PMAPCOUNT(mappings_replaced);
1606 			pmap_pvo_remove(pvo, -1, NULL);
1607 			break;
1608 		}
1609 	}
1610 
1611 	/*
1612 	 * If we aren't overwriting an mapping, try to allocate
1613 	 */
1614 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1615 	--pmap_pvo_enter_depth;
1616 #endif
1617 	pmap_interrupts_restore(msr);
1618 	if (pvo) {
1619 		pmap_pvo_free(pvo);
1620 	}
1621 	pvo = pool_get(pl, poolflags);
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 pvo_flags;
1900 	u_int was_exec = 0;
1901 
1902 	PMAP_LOCK();
1903 
1904 	if (__predict_false(!pmap_initialized)) {
1905 		pvo_head = &pmap_pvo_kunmanaged;
1906 		pl = &pmap_upvo_pool;
1907 		pvo_flags = 0;
1908 		pg = NULL;
1909 		was_exec = PTE_EXEC;
1910 	} else {
1911 		pvo_head = pa_to_pvoh(pa, &pg);
1912 		pl = &pmap_mpvo_pool;
1913 		pvo_flags = PVO_MANAGED;
1914 	}
1915 
1916 	DPRINTFN(ENTER,
1917 	    ("pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1918 	    pm, va, pa, prot, flags));
1919 
1920 	/*
1921 	 * If this is a managed page, and it's the first reference to the
1922 	 * page clear the execness of the page.  Otherwise fetch the execness.
1923 	 */
1924 	if (pg != NULL)
1925 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1926 
1927 	DPRINTFN(ENTER, (" was_exec=%d", was_exec));
1928 
1929 	/*
1930 	 * Assume the page is cache inhibited and access is guarded unless
1931 	 * it's in our available memory array.  If it is in the memory array,
1932 	 * asssume it's in memory coherent memory.
1933 	 */
1934 	pte_lo = PTE_IG;
1935 	if ((flags & PMAP_NC) == 0) {
1936 		for (mp = mem; mp->size; mp++) {
1937 			if (pa >= mp->start && pa < mp->start + mp->size) {
1938 				pte_lo = PTE_M;
1939 				break;
1940 			}
1941 		}
1942 	}
1943 
1944 	if (prot & VM_PROT_WRITE)
1945 		pte_lo |= PTE_BW;
1946 	else
1947 		pte_lo |= PTE_BR;
1948 
1949 	/*
1950 	 * If this was in response to a fault, "pre-fault" the PTE's
1951 	 * changed/referenced bit appropriately.
1952 	 */
1953 	if (flags & VM_PROT_WRITE)
1954 		pte_lo |= PTE_CHG;
1955 	if (flags & VM_PROT_ALL)
1956 		pte_lo |= PTE_REF;
1957 
1958 	/*
1959 	 * We need to know if this page can be executable
1960 	 */
1961 	flags |= (prot & VM_PROT_EXECUTE);
1962 
1963 	/*
1964 	 * Record mapping for later back-translation and pte spilling.
1965 	 * This will overwrite any existing mapping.
1966 	 */
1967 	error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags);
1968 
1969 	/*
1970 	 * Flush the real page from the instruction cache if this page is
1971 	 * mapped executable and cacheable and has not been flushed since
1972 	 * the last time it was modified.
1973 	 */
1974 	if (error == 0 &&
1975             (flags & VM_PROT_EXECUTE) &&
1976             (pte_lo & PTE_I) == 0 &&
1977 	    was_exec == 0) {
1978 		DPRINTFN(ENTER, (" syncicache"));
1979 		PMAPCOUNT(exec_synced);
1980 		pmap_syncicache(pa, PAGE_SIZE);
1981 		if (pg != NULL) {
1982 			pmap_attr_save(pg, PTE_EXEC);
1983 			PMAPCOUNT(exec_cached);
1984 #if defined(DEBUG) || defined(PMAPDEBUG)
1985 			if (pmapdebug & PMAPDEBUG_ENTER)
1986 				printf(" marked-as-exec");
1987 			else if (pmapdebug & PMAPDEBUG_EXEC)
1988 				printf("[pmap_enter: %#" _PRIxpa ": marked-as-exec]\n",
1989 				    VM_PAGE_TO_PHYS(pg));
1990 
1991 #endif
1992 		}
1993 	}
1994 
1995 	DPRINTFN(ENTER, (": error=%d\n", error));
1996 
1997 	PMAP_UNLOCK();
1998 
1999 	return error;
2000 }
2001 
2002 void
2003 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
2004 {
2005 	struct mem_region *mp;
2006 	register_t pte_lo;
2007 	int error;
2008 
2009 #if defined (PMAP_OEA64_BRIDGE)
2010 	if (va < VM_MIN_KERNEL_ADDRESS)
2011 		panic("pmap_kenter_pa: attempt to enter "
2012 		    "non-kernel address %#" _PRIxva "!", va);
2013 #endif
2014 
2015 	DPRINTFN(KENTER,
2016 	    ("pmap_kenter_pa(%#" _PRIxva ",%#" _PRIxpa ",%#x)\n", va, pa, prot));
2017 
2018 	PMAP_LOCK();
2019 
2020 	/*
2021 	 * Assume the page is cache inhibited and access is guarded unless
2022 	 * it's in our available memory array.  If it is in the memory array,
2023 	 * asssume it's in memory coherent memory.
2024 	 */
2025 	pte_lo = PTE_IG;
2026 	if ((prot & PMAP_NC) == 0) {
2027 		for (mp = mem; mp->size; mp++) {
2028 			if (pa >= mp->start && pa < mp->start + mp->size) {
2029 				pte_lo = PTE_M;
2030 				break;
2031 			}
2032 		}
2033 	}
2034 
2035 	if (prot & VM_PROT_WRITE)
2036 		pte_lo |= PTE_BW;
2037 	else
2038 		pte_lo |= PTE_BR;
2039 
2040 	/*
2041 	 * We don't care about REF/CHG on PVOs on the unmanaged list.
2042 	 */
2043 	error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool,
2044 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED);
2045 
2046 	if (error != 0)
2047 		panic("pmap_kenter_pa: failed to enter va %#" _PRIxva " pa %#" _PRIxpa ": %d",
2048 		      va, pa, error);
2049 
2050 	PMAP_UNLOCK();
2051 }
2052 
2053 void
2054 pmap_kremove(vaddr_t va, vsize_t len)
2055 {
2056 	if (va < VM_MIN_KERNEL_ADDRESS)
2057 		panic("pmap_kremove: attempt to remove "
2058 		    "non-kernel address %#" _PRIxva "!", va);
2059 
2060 	DPRINTFN(KREMOVE,("pmap_kremove(%#" _PRIxva ",%#" _PRIxva ")\n", va, len));
2061 	pmap_remove(pmap_kernel(), va, va + len);
2062 }
2063 
2064 /*
2065  * Remove the given range of mapping entries.
2066  */
2067 void
2068 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
2069 {
2070 	struct pvo_head pvol;
2071 	struct pvo_entry *pvo;
2072 	register_t msr;
2073 	int pteidx;
2074 
2075 	PMAP_LOCK();
2076 	LIST_INIT(&pvol);
2077 	msr = pmap_interrupts_off();
2078 	for (; va < endva; va += PAGE_SIZE) {
2079 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2080 		if (pvo != NULL) {
2081 			pmap_pvo_remove(pvo, pteidx, &pvol);
2082 		}
2083 	}
2084 	pmap_interrupts_restore(msr);
2085 	pmap_pvo_free_list(&pvol);
2086 	PMAP_UNLOCK();
2087 }
2088 
2089 /*
2090  * Get the physical page address for the given pmap/virtual address.
2091  */
2092 bool
2093 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
2094 {
2095 	struct pvo_entry *pvo;
2096 	register_t msr;
2097 
2098 	PMAP_LOCK();
2099 
2100 	/*
2101 	 * If this is a kernel pmap lookup, also check the battable
2102 	 * and if we get a hit, translate the VA to a PA using the
2103 	 * BAT entries.  Don't check for VM_MAX_KERNEL_ADDRESS is
2104 	 * that will wrap back to 0.
2105 	 */
2106 	if (pm == pmap_kernel() &&
2107 	    (va < VM_MIN_KERNEL_ADDRESS ||
2108 	     (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
2109 		KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
2110 #if defined (PMAP_OEA)
2111 #ifdef PPC_OEA601
2112 		if ((MFPVR() >> 16) == MPC601) {
2113 			register_t batu = battable[va >> 23].batu;
2114 			register_t batl = battable[va >> 23].batl;
2115 			register_t sr = iosrtable[va >> ADDR_SR_SHFT];
2116 			if (BAT601_VALID_P(batl) &&
2117 			    BAT601_VA_MATCH_P(batu, batl, va)) {
2118 				register_t mask =
2119 				    (~(batl & BAT601_BSM) << 17) & ~0x1ffffL;
2120 				if (pap)
2121 					*pap = (batl & mask) | (va & ~mask);
2122 				PMAP_UNLOCK();
2123 				return true;
2124 			} else if (SR601_VALID_P(sr) &&
2125 				   SR601_PA_MATCH_P(sr, va)) {
2126 				if (pap)
2127 					*pap = va;
2128 				PMAP_UNLOCK();
2129 				return true;
2130 			}
2131 		} else
2132 #endif /* PPC_OEA601 */
2133 		{
2134 			register_t batu = battable[va >> ADDR_SR_SHFT].batu;
2135 			if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) {
2136 				register_t batl =
2137 				    battable[va >> ADDR_SR_SHFT].batl;
2138 				register_t mask =
2139 				    (~(batu & BAT_BL) << 15) & ~0x1ffffL;
2140 				if (pap)
2141 					*pap = (batl & mask) | (va & ~mask);
2142 				PMAP_UNLOCK();
2143 				return true;
2144 			}
2145 		}
2146 		return false;
2147 #elif defined (PMAP_OEA64_BRIDGE)
2148 	if (va >= SEGMENT_LENGTH)
2149 		panic("%s: pm: %s va >= SEGMENT_LENGTH, va: 0x%08lx\n",
2150 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
2151 	else {
2152 		if (pap)
2153 			*pap = va;
2154 			PMAP_UNLOCK();
2155 			return true;
2156 	}
2157 #elif defined (PMAP_OEA64)
2158 #error PPC_OEA64 not supported
2159 #endif /* PPC_OEA */
2160 	}
2161 
2162 	msr = pmap_interrupts_off();
2163 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2164 	if (pvo != NULL) {
2165 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2166 		if (pap)
2167 			*pap = (pvo->pvo_pte.pte_lo & PTE_RPGN)
2168 			    | (va & ADDR_POFF);
2169 	}
2170 	pmap_interrupts_restore(msr);
2171 	PMAP_UNLOCK();
2172 	return pvo != NULL;
2173 }
2174 
2175 /*
2176  * Lower the protection on the specified range of this pmap.
2177  */
2178 void
2179 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
2180 {
2181 	struct pvo_entry *pvo;
2182 	volatile struct pte *pt;
2183 	register_t msr;
2184 	int pteidx;
2185 
2186 	/*
2187 	 * Since this routine only downgrades protection, we should
2188 	 * always be called with at least one bit not set.
2189 	 */
2190 	KASSERT(prot != VM_PROT_ALL);
2191 
2192 	/*
2193 	 * If there is no protection, this is equivalent to
2194 	 * remove the pmap from the pmap.
2195 	 */
2196 	if ((prot & VM_PROT_READ) == 0) {
2197 		pmap_remove(pm, va, endva);
2198 		return;
2199 	}
2200 
2201 	PMAP_LOCK();
2202 
2203 	msr = pmap_interrupts_off();
2204 	for (; va < endva; va += PAGE_SIZE) {
2205 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2206 		if (pvo == NULL)
2207 			continue;
2208 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2209 
2210 		/*
2211 		 * Revoke executable if asked to do so.
2212 		 */
2213 		if ((prot & VM_PROT_EXECUTE) == 0)
2214 			pvo_clear_exec(pvo);
2215 
2216 #if 0
2217 		/*
2218 		 * If the page is already read-only, no change
2219 		 * needs to be made.
2220 		 */
2221 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
2222 			continue;
2223 #endif
2224 		/*
2225 		 * Grab the PTE pointer before we diddle with
2226 		 * the cached PTE copy.
2227 		 */
2228 		pt = pmap_pvo_to_pte(pvo, pteidx);
2229 		/*
2230 		 * Change the protection of the page.
2231 		 */
2232 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2233 		pvo->pvo_pte.pte_lo |= PTE_BR;
2234 
2235 		/*
2236 		 * If the PVO is in the page table, update
2237 		 * that pte at well.
2238 		 */
2239 		if (pt != NULL) {
2240 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2241 			PVO_WHERE(pvo, PMAP_PROTECT);
2242 			PMAPCOUNT(ptes_changed);
2243 		}
2244 
2245 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2246 	}
2247 	pmap_interrupts_restore(msr);
2248 	PMAP_UNLOCK();
2249 }
2250 
2251 void
2252 pmap_unwire(pmap_t pm, vaddr_t va)
2253 {
2254 	struct pvo_entry *pvo;
2255 	register_t msr;
2256 
2257 	PMAP_LOCK();
2258 	msr = pmap_interrupts_off();
2259 	pvo = pmap_pvo_find_va(pm, va, NULL);
2260 	if (pvo != NULL) {
2261 		if (PVO_WIRED_P(pvo)) {
2262 			pvo->pvo_vaddr &= ~PVO_WIRED;
2263 			pm->pm_stats.wired_count--;
2264 		}
2265 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2266 	}
2267 	pmap_interrupts_restore(msr);
2268 	PMAP_UNLOCK();
2269 }
2270 
2271 /*
2272  * Lower the protection on the specified physical page.
2273  */
2274 void
2275 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
2276 {
2277 	struct pvo_head *pvo_head, pvol;
2278 	struct pvo_entry *pvo, *next_pvo;
2279 	volatile struct pte *pt;
2280 	register_t msr;
2281 
2282 	PMAP_LOCK();
2283 
2284 	KASSERT(prot != VM_PROT_ALL);
2285 	LIST_INIT(&pvol);
2286 	msr = pmap_interrupts_off();
2287 
2288 	/*
2289 	 * When UVM reuses a page, it does a pmap_page_protect with
2290 	 * VM_PROT_NONE.  At that point, we can clear the exec flag
2291 	 * since we know the page will have different contents.
2292 	 */
2293 	if ((prot & VM_PROT_READ) == 0) {
2294 		DPRINTFN(EXEC, ("[pmap_page_protect: %#" _PRIxpa ": clear-exec]\n",
2295 		    VM_PAGE_TO_PHYS(pg)));
2296 		if (pmap_attr_fetch(pg) & PTE_EXEC) {
2297 			PMAPCOUNT(exec_uncached_page_protect);
2298 			pmap_attr_clear(pg, PTE_EXEC);
2299 		}
2300 	}
2301 
2302 	pvo_head = vm_page_to_pvoh(pg);
2303 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2304 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
2305 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2306 
2307 		/*
2308 		 * Downgrading to no mapping at all, we just remove the entry.
2309 		 */
2310 		if ((prot & VM_PROT_READ) == 0) {
2311 			pmap_pvo_remove(pvo, -1, &pvol);
2312 			continue;
2313 		}
2314 
2315 		/*
2316 		 * If EXEC permission is being revoked, just clear the
2317 		 * flag in the PVO.
2318 		 */
2319 		if ((prot & VM_PROT_EXECUTE) == 0)
2320 			pvo_clear_exec(pvo);
2321 
2322 		/*
2323 		 * If this entry is already RO, don't diddle with the
2324 		 * page table.
2325 		 */
2326 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2327 			PMAP_PVO_CHECK(pvo);
2328 			continue;
2329 		}
2330 
2331 		/*
2332 		 * Grab the PTE before the we diddle the bits so
2333 		 * pvo_to_pte can verify the pte contents are as
2334 		 * expected.
2335 		 */
2336 		pt = pmap_pvo_to_pte(pvo, -1);
2337 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2338 		pvo->pvo_pte.pte_lo |= PTE_BR;
2339 		if (pt != NULL) {
2340 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2341 			PVO_WHERE(pvo, PMAP_PAGE_PROTECT);
2342 			PMAPCOUNT(ptes_changed);
2343 		}
2344 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2345 	}
2346 	pmap_interrupts_restore(msr);
2347 	pmap_pvo_free_list(&pvol);
2348 
2349 	PMAP_UNLOCK();
2350 }
2351 
2352 /*
2353  * Activate the address space for the specified process.  If the process
2354  * is the current process, load the new MMU context.
2355  */
2356 void
2357 pmap_activate(struct lwp *l)
2358 {
2359 	struct pcb *pcb = lwp_getpcb(l);
2360 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2361 
2362 	DPRINTFN(ACTIVATE,
2363 	    ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp));
2364 
2365 	/*
2366 	 * XXX Normally performed in cpu_fork().
2367 	 */
2368 	pcb->pcb_pm = pmap;
2369 
2370 	/*
2371 	* In theory, the SR registers need only be valid on return
2372 	* to user space wait to do them there.
2373 	*/
2374 	if (l == curlwp) {
2375 		/* Store pointer to new current pmap. */
2376 		curpm = pmap;
2377 	}
2378 }
2379 
2380 /*
2381  * Deactivate the specified process's address space.
2382  */
2383 void
2384 pmap_deactivate(struct lwp *l)
2385 {
2386 }
2387 
2388 bool
2389 pmap_query_bit(struct vm_page *pg, int ptebit)
2390 {
2391 	struct pvo_entry *pvo;
2392 	volatile struct pte *pt;
2393 	register_t msr;
2394 
2395 	PMAP_LOCK();
2396 
2397 	if (pmap_attr_fetch(pg) & ptebit) {
2398 		PMAP_UNLOCK();
2399 		return true;
2400 	}
2401 
2402 	msr = pmap_interrupts_off();
2403 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2404 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2405 		/*
2406 		 * See if we saved the bit off.  If so cache, it and return
2407 		 * success.
2408 		 */
2409 		if (pvo->pvo_pte.pte_lo & ptebit) {
2410 			pmap_attr_save(pg, ptebit);
2411 			PMAP_PVO_CHECK(pvo);		/* sanity check */
2412 			pmap_interrupts_restore(msr);
2413 			PMAP_UNLOCK();
2414 			return true;
2415 		}
2416 	}
2417 	/*
2418 	 * No luck, now go thru the hard part of looking at the ptes
2419 	 * themselves.  Sync so any pending REF/CHG bits are flushed
2420 	 * to the PTEs.
2421 	 */
2422 	SYNC();
2423 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2424 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2425 		/*
2426 		 * See if this pvo have a valid PTE.  If so, fetch the
2427 		 * REF/CHG bits from the valid PTE.  If the appropriate
2428 		 * ptebit is set, cache, it and return success.
2429 		 */
2430 		pt = pmap_pvo_to_pte(pvo, -1);
2431 		if (pt != NULL) {
2432 			pmap_pte_synch(pt, &pvo->pvo_pte);
2433 			if (pvo->pvo_pte.pte_lo & ptebit) {
2434 				pmap_attr_save(pg, ptebit);
2435 				PMAP_PVO_CHECK(pvo);		/* sanity check */
2436 				pmap_interrupts_restore(msr);
2437 				PMAP_UNLOCK();
2438 				return true;
2439 			}
2440 		}
2441 	}
2442 	pmap_interrupts_restore(msr);
2443 	PMAP_UNLOCK();
2444 	return false;
2445 }
2446 
2447 bool
2448 pmap_clear_bit(struct vm_page *pg, int ptebit)
2449 {
2450 	struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2451 	struct pvo_entry *pvo;
2452 	volatile struct pte *pt;
2453 	register_t msr;
2454 	int rv = 0;
2455 
2456 	PMAP_LOCK();
2457 	msr = pmap_interrupts_off();
2458 
2459 	/*
2460 	 * Fetch the cache value
2461 	 */
2462 	rv |= pmap_attr_fetch(pg);
2463 
2464 	/*
2465 	 * Clear the cached value.
2466 	 */
2467 	pmap_attr_clear(pg, ptebit);
2468 
2469 	/*
2470 	 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2471 	 * can reset the right ones).  Note that since the pvo entries and
2472 	 * list heads are accessed via BAT0 and are never placed in the
2473 	 * page table, we don't have to worry about further accesses setting
2474 	 * the REF/CHG bits.
2475 	 */
2476 	SYNC();
2477 
2478 	/*
2479 	 * For each pvo entry, clear pvo's ptebit.  If this pvo have a
2480 	 * valid PTE.  If so, clear the ptebit from the valid PTE.
2481 	 */
2482 	LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2483 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2484 		pt = pmap_pvo_to_pte(pvo, -1);
2485 		if (pt != NULL) {
2486 			/*
2487 			 * Only sync the PTE if the bit we are looking
2488 			 * for is not already set.
2489 			 */
2490 			if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2491 				pmap_pte_synch(pt, &pvo->pvo_pte);
2492 			/*
2493 			 * If the bit we are looking for was already set,
2494 			 * clear that bit in the pte.
2495 			 */
2496 			if (pvo->pvo_pte.pte_lo & ptebit)
2497 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2498 		}
2499 		rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2500 		pvo->pvo_pte.pte_lo &= ~ptebit;
2501 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2502 	}
2503 	pmap_interrupts_restore(msr);
2504 
2505 	/*
2506 	 * If we are clearing the modify bit and this page was marked EXEC
2507 	 * and the user of the page thinks the page was modified, then we
2508 	 * need to clean it from the icache if it's mapped or clear the EXEC
2509 	 * bit if it's not mapped.  The page itself might not have the CHG
2510 	 * bit set if the modification was done via DMA to the page.
2511 	 */
2512 	if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2513 		if (LIST_EMPTY(pvoh)) {
2514 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#" _PRIxpa ": clear-exec]\n",
2515 			    VM_PAGE_TO_PHYS(pg)));
2516 			pmap_attr_clear(pg, PTE_EXEC);
2517 			PMAPCOUNT(exec_uncached_clear_modify);
2518 		} else {
2519 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#" _PRIxpa ": syncicache]\n",
2520 			    VM_PAGE_TO_PHYS(pg)));
2521 			pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE);
2522 			PMAPCOUNT(exec_synced_clear_modify);
2523 		}
2524 	}
2525 	PMAP_UNLOCK();
2526 	return (rv & ptebit) != 0;
2527 }
2528 
2529 void
2530 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2531 {
2532 	struct pvo_entry *pvo;
2533 	size_t offset = va & ADDR_POFF;
2534 	int s;
2535 
2536 	PMAP_LOCK();
2537 	s = splvm();
2538 	while (len > 0) {
2539 		size_t seglen = PAGE_SIZE - offset;
2540 		if (seglen > len)
2541 			seglen = len;
2542 		pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2543 		if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) {
2544 			pmap_syncicache(
2545 			    (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2546 			PMAP_PVO_CHECK(pvo);
2547 		}
2548 		va += seglen;
2549 		len -= seglen;
2550 		offset = 0;
2551 	}
2552 	splx(s);
2553 	PMAP_UNLOCK();
2554 }
2555 
2556 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2557 void
2558 pmap_pte_print(volatile struct pte *pt)
2559 {
2560 	printf("PTE %p: ", pt);
2561 
2562 #if defined(PMAP_OEA)
2563 	/* High word: */
2564 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2565 #else
2566 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2567 #endif /* PMAP_OEA */
2568 
2569 	printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2570 	printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2571 
2572 	printf("%#" _PRIxpte " %#" _PRIxpte "",
2573 	    (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2574 	    pt->pte_hi & PTE_API);
2575 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2576 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2577 #else
2578 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2579 #endif /* PMAP_OEA */
2580 
2581 	/* Low word: */
2582 #if defined (PMAP_OEA)
2583 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2584 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2585 #else
2586 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2587 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2588 #endif
2589 	printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2590 	printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2591 	printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2592 	printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2593 	printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2594 	printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2595 	switch (pt->pte_lo & PTE_PP) {
2596 	case PTE_BR: printf("br]\n"); break;
2597 	case PTE_BW: printf("bw]\n"); break;
2598 	case PTE_SO: printf("so]\n"); break;
2599 	case PTE_SW: printf("sw]\n"); break;
2600 	}
2601 }
2602 #endif
2603 
2604 #if defined(DDB)
2605 void
2606 pmap_pteg_check(void)
2607 {
2608 	volatile struct pte *pt;
2609 	int i;
2610 	int ptegidx;
2611 	u_int p_valid = 0;
2612 	u_int s_valid = 0;
2613 	u_int invalid = 0;
2614 
2615 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2616 		for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2617 			if (pt->pte_hi & PTE_VALID) {
2618 				if (pt->pte_hi & PTE_HID)
2619 					s_valid++;
2620 				else
2621 				{
2622 					p_valid++;
2623 				}
2624 			} else
2625 				invalid++;
2626 		}
2627 	}
2628 	printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2629 		p_valid, p_valid, s_valid, s_valid,
2630 		invalid, invalid);
2631 }
2632 
2633 void
2634 pmap_print_mmuregs(void)
2635 {
2636 	int i;
2637 	u_int cpuvers;
2638 #ifndef PMAP_OEA64
2639 	vaddr_t addr;
2640 	register_t soft_sr[16];
2641 #endif
2642 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2643 	struct bat soft_ibat[4];
2644 	struct bat soft_dbat[4];
2645 #endif
2646 	paddr_t sdr1;
2647 
2648 	cpuvers = MFPVR() >> 16;
2649 	__asm volatile ("mfsdr1 %0" : "=r"(sdr1));
2650 #ifndef PMAP_OEA64
2651 	addr = 0;
2652 	for (i = 0; i < 16; i++) {
2653 		soft_sr[i] = MFSRIN(addr);
2654 		addr += (1 << ADDR_SR_SHFT);
2655 	}
2656 #endif
2657 
2658 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2659 	/* read iBAT (601: uBAT) registers */
2660 	__asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2661 	__asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2662 	__asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2663 	__asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2664 	__asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2665 	__asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2666 	__asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2667 	__asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2668 
2669 
2670 	if (cpuvers != MPC601) {
2671 		/* read dBAT registers */
2672 		__asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2673 		__asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2674 		__asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2675 		__asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2676 		__asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2677 		__asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2678 		__asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2679 		__asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2680 	}
2681 #endif
2682 
2683 	printf("SDR1:\t%#" _PRIxpa "\n", sdr1);
2684 #ifndef PMAP_OEA64
2685 	printf("SR[]:\t");
2686 	for (i = 0; i < 4; i++)
2687 		printf("0x%08lx,   ", soft_sr[i]);
2688 	printf("\n\t");
2689 	for ( ; i < 8; i++)
2690 		printf("0x%08lx,   ", soft_sr[i]);
2691 	printf("\n\t");
2692 	for ( ; i < 12; i++)
2693 		printf("0x%08lx,   ", soft_sr[i]);
2694 	printf("\n\t");
2695 	for ( ; i < 16; i++)
2696 		printf("0x%08lx,   ", soft_sr[i]);
2697 	printf("\n");
2698 #endif
2699 
2700 #if defined(PMAP_OEA) || defined(PMAP_OEA_BRIDGE)
2701 	printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2702 	for (i = 0; i < 4; i++) {
2703 		printf("0x%08lx 0x%08lx, ",
2704 			soft_ibat[i].batu, soft_ibat[i].batl);
2705 		if (i == 1)
2706 			printf("\n\t");
2707 	}
2708 	if (cpuvers != MPC601) {
2709 		printf("\ndBAT[]:\t");
2710 		for (i = 0; i < 4; i++) {
2711 			printf("0x%08lx 0x%08lx, ",
2712 				soft_dbat[i].batu, soft_dbat[i].batl);
2713 			if (i == 1)
2714 				printf("\n\t");
2715 		}
2716 	}
2717 	printf("\n");
2718 #endif /* PMAP_OEA... */
2719 }
2720 
2721 void
2722 pmap_print_pte(pmap_t pm, vaddr_t va)
2723 {
2724 	struct pvo_entry *pvo;
2725 	volatile struct pte *pt;
2726 	int pteidx;
2727 
2728 	pvo = pmap_pvo_find_va(pm, va, &pteidx);
2729 	if (pvo != NULL) {
2730 		pt = pmap_pvo_to_pte(pvo, pteidx);
2731 		if (pt != NULL) {
2732 			printf("VA %#" _PRIxva " -> %p -> %s %#" _PRIxpte ", %#" _PRIxpte "\n",
2733 				va, pt,
2734 				pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2735 				pt->pte_hi, pt->pte_lo);
2736 		} else {
2737 			printf("No valid PTE found\n");
2738 		}
2739 	} else {
2740 		printf("Address not in pmap\n");
2741 	}
2742 }
2743 
2744 void
2745 pmap_pteg_dist(void)
2746 {
2747 	struct pvo_entry *pvo;
2748 	int ptegidx;
2749 	int depth;
2750 	int max_depth = 0;
2751 	unsigned int depths[64];
2752 
2753 	memset(depths, 0, sizeof(depths));
2754 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2755 		depth = 0;
2756 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2757 			depth++;
2758 		}
2759 		if (depth > max_depth)
2760 			max_depth = depth;
2761 		if (depth > 63)
2762 			depth = 63;
2763 		depths[depth]++;
2764 	}
2765 
2766 	for (depth = 0; depth < 64; depth++) {
2767 		printf("  [%2d]: %8u", depth, depths[depth]);
2768 		if ((depth & 3) == 3)
2769 			printf("\n");
2770 		if (depth == max_depth)
2771 			break;
2772 	}
2773 	if ((depth & 3) != 3)
2774 		printf("\n");
2775 	printf("Max depth found was %d\n", max_depth);
2776 }
2777 #endif /* DEBUG */
2778 
2779 #if defined(PMAPCHECK) || defined(DEBUG)
2780 void
2781 pmap_pvo_verify(void)
2782 {
2783 	int ptegidx;
2784 	int s;
2785 
2786 	s = splvm();
2787 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2788 		struct pvo_entry *pvo;
2789 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2790 			if ((uintptr_t) pvo >= SEGMENT_LENGTH)
2791 				panic("pmap_pvo_verify: invalid pvo %p "
2792 				    "on list %#x", pvo, ptegidx);
2793 			pmap_pvo_check(pvo);
2794 		}
2795 	}
2796 	splx(s);
2797 }
2798 #endif /* PMAPCHECK */
2799 
2800 
2801 void *
2802 pmap_pool_ualloc(struct pool *pp, int flags)
2803 {
2804 	struct pvo_page *pvop;
2805 
2806 	if (uvm.page_init_done != true) {
2807 		return (void *) uvm_pageboot_alloc(PAGE_SIZE);
2808 	}
2809 
2810 	PMAP_LOCK();
2811 	pvop = SIMPLEQ_FIRST(&pmap_upvop_head);
2812 	if (pvop != NULL) {
2813 		pmap_upvop_free--;
2814 		SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link);
2815 		PMAP_UNLOCK();
2816 		return pvop;
2817 	}
2818 	PMAP_UNLOCK();
2819 	return pmap_pool_malloc(pp, flags);
2820 }
2821 
2822 void *
2823 pmap_pool_malloc(struct pool *pp, int flags)
2824 {
2825 	struct pvo_page *pvop;
2826 	struct vm_page *pg;
2827 
2828 	PMAP_LOCK();
2829 	pvop = SIMPLEQ_FIRST(&pmap_mpvop_head);
2830 	if (pvop != NULL) {
2831 		pmap_mpvop_free--;
2832 		SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link);
2833 		PMAP_UNLOCK();
2834 		return pvop;
2835 	}
2836 	PMAP_UNLOCK();
2837  again:
2838 	pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE,
2839 	    UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256);
2840 	if (__predict_false(pg == NULL)) {
2841 		if (flags & PR_WAITOK) {
2842 			uvm_wait("plpg");
2843 			goto again;
2844 		} else {
2845 			return (0);
2846 		}
2847 	}
2848 	KDASSERT(VM_PAGE_TO_PHYS(pg) == (uintptr_t)VM_PAGE_TO_PHYS(pg));
2849 	return (void *)(uintptr_t) VM_PAGE_TO_PHYS(pg);
2850 }
2851 
2852 void
2853 pmap_pool_ufree(struct pool *pp, void *va)
2854 {
2855 	struct pvo_page *pvop;
2856 #if 0
2857 	if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) {
2858 		pmap_pool_mfree(va, size, tag);
2859 		return;
2860 	}
2861 #endif
2862 	PMAP_LOCK();
2863 	pvop = va;
2864 	SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link);
2865 	pmap_upvop_free++;
2866 	if (pmap_upvop_free > pmap_upvop_maxfree)
2867 		pmap_upvop_maxfree = pmap_upvop_free;
2868 	PMAP_UNLOCK();
2869 }
2870 
2871 void
2872 pmap_pool_mfree(struct pool *pp, void *va)
2873 {
2874 	struct pvo_page *pvop;
2875 
2876 	PMAP_LOCK();
2877 	pvop = va;
2878 	SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link);
2879 	pmap_mpvop_free++;
2880 	if (pmap_mpvop_free > pmap_mpvop_maxfree)
2881 		pmap_mpvop_maxfree = pmap_mpvop_free;
2882 	PMAP_UNLOCK();
2883 #if 0
2884 	uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va));
2885 #endif
2886 }
2887 
2888 /*
2889  * This routine in bootstraping to steal to-be-managed memory (which will
2890  * then be unmanaged).  We use it to grab from the first 256MB for our
2891  * pmap needs and above 256MB for other stuff.
2892  */
2893 vaddr_t
2894 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2895 {
2896 	vsize_t size;
2897 	vaddr_t va;
2898 	paddr_t pa = 0;
2899 	int npgs, bank;
2900 	struct vm_physseg *ps;
2901 
2902 	if (uvm.page_init_done == true)
2903 		panic("pmap_steal_memory: called _after_ bootstrap");
2904 
2905 	*vstartp = VM_MIN_KERNEL_ADDRESS;
2906 	*vendp = VM_MAX_KERNEL_ADDRESS;
2907 
2908 	size = round_page(vsize);
2909 	npgs = atop(size);
2910 
2911 	/*
2912 	 * PA 0 will never be among those given to UVM so we can use it
2913 	 * to indicate we couldn't steal any memory.
2914 	 */
2915 	for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) {
2916 		if (ps->free_list == VM_FREELIST_FIRST256 &&
2917 		    ps->avail_end - ps->avail_start >= npgs) {
2918 			pa = ptoa(ps->avail_start);
2919 			break;
2920 		}
2921 	}
2922 
2923 	if (pa == 0)
2924 		panic("pmap_steal_memory: no approriate memory to steal!");
2925 
2926 	ps->avail_start += npgs;
2927 	ps->start += npgs;
2928 
2929 	/*
2930 	 * If we've used up all the pages in the segment, remove it and
2931 	 * compact the list.
2932 	 */
2933 	if (ps->avail_start == ps->end) {
2934 		/*
2935 		 * If this was the last one, then a very bad thing has occurred
2936 		 */
2937 		if (--vm_nphysseg == 0)
2938 			panic("pmap_steal_memory: out of memory!");
2939 
2940 		printf("pmap_steal_memory: consumed bank %d\n", bank);
2941 		for (; bank < vm_nphysseg; bank++, ps++) {
2942 			ps[0] = ps[1];
2943 		}
2944 	}
2945 
2946 	va = (vaddr_t) pa;
2947 	memset((void *) va, 0, size);
2948 	pmap_pages_stolen += npgs;
2949 #ifdef DEBUG
2950 	if (pmapdebug && npgs > 1) {
2951 		u_int cnt = 0;
2952 		for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++)
2953 			cnt += ps->avail_end - ps->avail_start;
2954 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2955 		    npgs, pmap_pages_stolen, cnt);
2956 	}
2957 #endif
2958 
2959 	return va;
2960 }
2961 
2962 /*
2963  * Find a chuck of memory with right size and alignment.
2964  */
2965 paddr_t
2966 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2967 {
2968 	struct mem_region *mp;
2969 	paddr_t s, e;
2970 	int i, j;
2971 
2972 	size = round_page(size);
2973 
2974 	DPRINTFN(BOOT,
2975 	    ("pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
2976 	    size, alignment, at_end));
2977 
2978 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2979 		panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
2980 		    alignment);
2981 
2982 	if (at_end) {
2983 		if (alignment != PAGE_SIZE)
2984 			panic("pmap_boot_find_memory: invalid ending "
2985 			    "alignment %#" _PRIxpa, alignment);
2986 
2987 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
2988 			s = mp->start + mp->size - size;
2989 			if (s >= mp->start && mp->size >= size) {
2990 				DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
2991 				DPRINTFN(BOOT,
2992 				    ("pmap_boot_find_memory: b-avail[%d] start "
2993 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
2994 				     mp->start, mp->size));
2995 				mp->size -= size;
2996 				DPRINTFN(BOOT,
2997 				    ("pmap_boot_find_memory: a-avail[%d] start "
2998 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
2999 				     mp->start, mp->size));
3000 				return s;
3001 			}
3002 		}
3003 		panic("pmap_boot_find_memory: no available memory");
3004 	}
3005 
3006 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3007 		s = (mp->start + alignment - 1) & ~(alignment-1);
3008 		e = s + size;
3009 
3010 		/*
3011 		 * Is the calculated region entirely within the region?
3012 		 */
3013 		if (s < mp->start || e > mp->start + mp->size)
3014 			continue;
3015 
3016 		DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
3017 		if (s == mp->start) {
3018 			/*
3019 			 * If the block starts at the beginning of region,
3020 			 * adjust the size & start. (the region may now be
3021 			 * zero in length)
3022 			 */
3023 			DPRINTFN(BOOT,
3024 			    ("pmap_boot_find_memory: b-avail[%d] start "
3025 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3026 			mp->start += size;
3027 			mp->size -= size;
3028 			DPRINTFN(BOOT,
3029 			    ("pmap_boot_find_memory: a-avail[%d] start "
3030 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3031 		} else if (e == mp->start + mp->size) {
3032 			/*
3033 			 * If the block starts at the beginning of region,
3034 			 * adjust only the size.
3035 			 */
3036 			DPRINTFN(BOOT,
3037 			    ("pmap_boot_find_memory: b-avail[%d] start "
3038 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3039 			mp->size -= size;
3040 			DPRINTFN(BOOT,
3041 			    ("pmap_boot_find_memory: a-avail[%d] start "
3042 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3043 		} else {
3044 			/*
3045 			 * Block is in the middle of the region, so we
3046 			 * have to split it in two.
3047 			 */
3048 			for (j = avail_cnt; j > i + 1; j--) {
3049 				avail[j] = avail[j-1];
3050 			}
3051 			DPRINTFN(BOOT,
3052 			    ("pmap_boot_find_memory: b-avail[%d] start "
3053 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3054 			mp[1].start = e;
3055 			mp[1].size = mp[0].start + mp[0].size - e;
3056 			mp[0].size = s - mp[0].start;
3057 			avail_cnt++;
3058 			for (; i < avail_cnt; i++) {
3059 				DPRINTFN(BOOT,
3060 				    ("pmap_boot_find_memory: a-avail[%d] "
3061 				     "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3062 				     avail[i].start, avail[i].size));
3063 			}
3064 		}
3065 		KASSERT(s == (uintptr_t) s);
3066 		return s;
3067 	}
3068 	panic("pmap_boot_find_memory: not enough memory for "
3069 	    "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3070 }
3071 
3072 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3073 #if defined (PMAP_OEA64_BRIDGE)
3074 int
3075 pmap_setup_segment0_map(int use_large_pages, ...)
3076 {
3077     vaddr_t va;
3078 
3079     register_t pte_lo = 0x0;
3080     int ptegidx = 0, i = 0;
3081     struct pte pte;
3082     va_list ap;
3083 
3084     /* Coherent + Supervisor RW, no user access */
3085     pte_lo = PTE_M;
3086 
3087     /* XXXSL
3088      * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3089      * these have to take priority.
3090      */
3091     for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3092         ptegidx = va_to_pteg(pmap_kernel(), va);
3093         pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3094         i = pmap_pte_insert(ptegidx, &pte);
3095     }
3096 
3097     va_start(ap, use_large_pages);
3098     while (1) {
3099         paddr_t pa;
3100         size_t size;
3101 
3102         va = va_arg(ap, vaddr_t);
3103 
3104         if (va == 0)
3105             break;
3106 
3107         pa = va_arg(ap, paddr_t);
3108         size = va_arg(ap, size_t);
3109 
3110         for (; va < (va + size); va += 0x1000, pa += 0x1000) {
3111 #if 0
3112 	    printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__,  va, pa);
3113 #endif
3114             ptegidx = va_to_pteg(pmap_kernel(), va);
3115             pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3116             i = pmap_pte_insert(ptegidx, &pte);
3117         }
3118     }
3119 
3120     TLBSYNC();
3121     SYNC();
3122     return (0);
3123 }
3124 #endif /* PMAP_OEA64_BRIDGE */
3125 
3126 /*
3127  * This is not part of the defined PMAP interface and is specific to the
3128  * PowerPC architecture.  This is called during initppc, before the system
3129  * is really initialized.
3130  */
3131 void
3132 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3133 {
3134 	struct mem_region *mp, tmp;
3135 	paddr_t s, e;
3136 	psize_t size;
3137 	int i, j;
3138 
3139 	/*
3140 	 * Get memory.
3141 	 */
3142 	mem_regions(&mem, &avail);
3143 #if defined(DEBUG)
3144 	if (pmapdebug & PMAPDEBUG_BOOT) {
3145 		printf("pmap_bootstrap: memory configuration:\n");
3146 		for (mp = mem; mp->size; mp++) {
3147 			printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3148 				mp->start, mp->size);
3149 		}
3150 		for (mp = avail; mp->size; mp++) {
3151 			printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3152 				mp->start, mp->size);
3153 		}
3154 	}
3155 #endif
3156 
3157 	/*
3158 	 * Find out how much physical memory we have and in how many chunks.
3159 	 */
3160 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3161 		if (mp->start >= pmap_memlimit)
3162 			continue;
3163 		if (mp->start + mp->size > pmap_memlimit) {
3164 			size = pmap_memlimit - mp->start;
3165 			physmem += btoc(size);
3166 		} else {
3167 			physmem += btoc(mp->size);
3168 		}
3169 		mem_cnt++;
3170 	}
3171 
3172 	/*
3173 	 * Count the number of available entries.
3174 	 */
3175 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
3176 		avail_cnt++;
3177 
3178 	/*
3179 	 * Page align all regions.
3180 	 */
3181 	kernelstart = trunc_page(kernelstart);
3182 	kernelend = round_page(kernelend);
3183 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3184 		s = round_page(mp->start);
3185 		mp->size -= (s - mp->start);
3186 		mp->size = trunc_page(mp->size);
3187 		mp->start = s;
3188 		e = mp->start + mp->size;
3189 
3190 		DPRINTFN(BOOT,
3191 		    ("pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3192 		    i, mp->start, mp->size));
3193 
3194 		/*
3195 		 * Don't allow the end to run beyond our artificial limit
3196 		 */
3197 		if (e > pmap_memlimit)
3198 			e = pmap_memlimit;
3199 
3200 		/*
3201 		 * Is this region empty or strange?  skip it.
3202 		 */
3203 		if (e <= s) {
3204 			mp->start = 0;
3205 			mp->size = 0;
3206 			continue;
3207 		}
3208 
3209 		/*
3210 		 * Does this overlap the beginning of kernel?
3211 		 *   Does extend past the end of the kernel?
3212 		 */
3213 		else if (s < kernelstart && e > kernelstart) {
3214 			if (e > kernelend) {
3215 				avail[avail_cnt].start = kernelend;
3216 				avail[avail_cnt].size = e - kernelend;
3217 				avail_cnt++;
3218 			}
3219 			mp->size = kernelstart - s;
3220 		}
3221 		/*
3222 		 * Check whether this region overlaps the end of the kernel.
3223 		 */
3224 		else if (s < kernelend && e > kernelend) {
3225 			mp->start = kernelend;
3226 			mp->size = e - kernelend;
3227 		}
3228 		/*
3229 		 * Look whether this regions is completely inside the kernel.
3230 		 * Nuke it if it does.
3231 		 */
3232 		else if (s >= kernelstart && e <= kernelend) {
3233 			mp->start = 0;
3234 			mp->size = 0;
3235 		}
3236 		/*
3237 		 * If the user imposed a memory limit, enforce it.
3238 		 */
3239 		else if (s >= pmap_memlimit) {
3240 			mp->start = -PAGE_SIZE;	/* let's know why */
3241 			mp->size = 0;
3242 		}
3243 		else {
3244 			mp->start = s;
3245 			mp->size = e - s;
3246 		}
3247 		DPRINTFN(BOOT,
3248 		    ("pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3249 		    i, mp->start, mp->size));
3250 	}
3251 
3252 	/*
3253 	 * Move (and uncount) all the null return to the end.
3254 	 */
3255 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3256 		if (mp->size == 0) {
3257 			tmp = avail[i];
3258 			avail[i] = avail[--avail_cnt];
3259 			avail[avail_cnt] = avail[i];
3260 		}
3261 	}
3262 
3263 	/*
3264 	 * (Bubble)sort them into ascending order.
3265 	 */
3266 	for (i = 0; i < avail_cnt; i++) {
3267 		for (j = i + 1; j < avail_cnt; j++) {
3268 			if (avail[i].start > avail[j].start) {
3269 				tmp = avail[i];
3270 				avail[i] = avail[j];
3271 				avail[j] = tmp;
3272 			}
3273 		}
3274 	}
3275 
3276 	/*
3277 	 * Make sure they don't overlap.
3278 	 */
3279 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3280 		if (mp[0].start + mp[0].size > mp[1].start) {
3281 			mp[0].size = mp[1].start - mp[0].start;
3282 		}
3283 		DPRINTFN(BOOT,
3284 		    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3285 		    i, mp->start, mp->size));
3286 	}
3287 	DPRINTFN(BOOT,
3288 	    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3289 	    i, mp->start, mp->size));
3290 
3291 #ifdef	PTEGCOUNT
3292 	pmap_pteg_cnt = PTEGCOUNT;
3293 #else /* PTEGCOUNT */
3294 
3295 	pmap_pteg_cnt = 0x1000;
3296 
3297 	while (pmap_pteg_cnt < physmem)
3298 		pmap_pteg_cnt <<= 1;
3299 
3300 	pmap_pteg_cnt >>= 1;
3301 #endif /* PTEGCOUNT */
3302 
3303 #ifdef DEBUG
3304 	DPRINTFN(BOOT,
3305 		("pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt));
3306 #endif
3307 
3308 	/*
3309 	 * Find suitably aligned memory for PTEG hash table.
3310 	 */
3311 	size = pmap_pteg_cnt * sizeof(struct pteg);
3312 	pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3313 
3314 #ifdef DEBUG
3315 	DPRINTFN(BOOT,
3316 		("PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table));
3317 #endif
3318 
3319 
3320 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3321 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
3322 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > 256MB",
3323 		    pmap_pteg_table, size);
3324 #endif
3325 
3326 	memset(__UNVOLATILE(pmap_pteg_table), 0,
3327 		pmap_pteg_cnt * sizeof(struct pteg));
3328 	pmap_pteg_mask = pmap_pteg_cnt - 1;
3329 
3330 	/*
3331 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3332 	 * with pages.  So we just steal them before giving them to UVM.
3333 	 */
3334 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3335 	pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3336 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3337 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
3338 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > 256MB",
3339 		    pmap_pvo_table, size);
3340 #endif
3341 
3342 	for (i = 0; i < pmap_pteg_cnt; i++)
3343 		TAILQ_INIT(&pmap_pvo_table[i]);
3344 
3345 #ifndef MSGBUFADDR
3346 	/*
3347 	 * Allocate msgbuf in high memory.
3348 	 */
3349 	msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3350 #endif
3351 
3352 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3353 		paddr_t pfstart = atop(mp->start);
3354 		paddr_t pfend = atop(mp->start + mp->size);
3355 		if (mp->size == 0)
3356 			continue;
3357 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
3358 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3359 				VM_FREELIST_FIRST256);
3360 		} else if (mp->start >= SEGMENT_LENGTH) {
3361 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3362 				VM_FREELIST_DEFAULT);
3363 		} else {
3364 			pfend = atop(SEGMENT_LENGTH);
3365 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3366 				VM_FREELIST_FIRST256);
3367 			pfstart = atop(SEGMENT_LENGTH);
3368 			pfend = atop(mp->start + mp->size);
3369 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3370 				VM_FREELIST_DEFAULT);
3371 		}
3372 	}
3373 
3374 	/*
3375 	 * Make sure kernel vsid is allocated as well as VSID 0.
3376 	 */
3377 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3378 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3379 	pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3380 		|= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3381 	pmap_vsid_bitmap[0] |= 1;
3382 
3383 	/*
3384 	 * Initialize kernel pmap and hardware.
3385 	 */
3386 
3387 /* PMAP_OEA64_BRIDGE does support these instructions */
3388 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
3389 	for (i = 0; i < 16; i++) {
3390  		pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3391 		__asm volatile ("mtsrin %0,%1"
3392  			      :: "r"(KERNELN_SEGMENT(i)|SR_PRKEY), "r"(i << ADDR_SR_SHFT));
3393 	}
3394 
3395 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3396 	__asm volatile ("mtsr %0,%1"
3397 		      :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
3398 #ifdef KERNEL2_SR
3399 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3400 	__asm volatile ("mtsr %0,%1"
3401 		      :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
3402 #endif
3403 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3404 #if defined (PMAP_OEA)
3405 	for (i = 0; i < 16; i++) {
3406 		if (iosrtable[i] & SR601_T) {
3407 			pmap_kernel()->pm_sr[i] = iosrtable[i];
3408 			__asm volatile ("mtsrin %0,%1"
3409 			    :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT));
3410 		}
3411 	}
3412 	__asm volatile ("sync; mtsdr1 %0; isync"
3413 		      :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10)));
3414 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
3415  	__asm __volatile ("sync; mtsdr1 %0; isync"
3416  		      :: "r"((uintptr_t)pmap_pteg_table | (32 - cntlzw(pmap_pteg_mask >> 11))));
3417 #endif
3418 	tlbia();
3419 
3420 #ifdef ALTIVEC
3421 	pmap_use_altivec = cpu_altivec;
3422 #endif
3423 
3424 #ifdef DEBUG
3425 	if (pmapdebug & PMAPDEBUG_BOOT) {
3426 		u_int cnt;
3427 		int bank;
3428 		char pbuf[9];
3429 		for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) {
3430 			cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start;
3431 			printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3432 			    bank,
3433 			    ptoa(vm_physmem[bank].avail_start),
3434 			    ptoa(vm_physmem[bank].avail_end),
3435 			    ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start));
3436 		}
3437 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3438 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3439 		    pbuf, cnt);
3440 	}
3441 #endif
3442 
3443 	pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry),
3444 	    sizeof(struct pvo_entry), 0, 0, "pmap_upvopl",
3445 	    &pmap_pool_uallocator, IPL_VM);
3446 
3447 	pool_setlowat(&pmap_upvo_pool, 252);
3448 
3449 	pool_init(&pmap_pool, sizeof(struct pmap),
3450 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator,
3451 	    IPL_NONE);
3452 
3453 #if defined(PMAP_NEED_MAPKERNEL) || 1
3454 	{
3455 		struct pmap *pm = pmap_kernel();
3456 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3457 		extern int etext[], kernel_text[];
3458 		vaddr_t va, va_etext = (paddr_t) etext;
3459 #endif
3460 		paddr_t pa, pa_end;
3461 		register_t sr;
3462 		struct pte pt;
3463 		unsigned int ptegidx;
3464 		int bank;
3465 
3466 		sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3467 		pm->pm_sr[0] = sr;
3468 
3469 		for (bank = 0; bank < vm_nphysseg; bank++) {
3470 			pa_end = ptoa(vm_physmem[bank].avail_end);
3471 			pa = ptoa(vm_physmem[bank].avail_start);
3472 			for (; pa < pa_end; pa += PAGE_SIZE) {
3473 				ptegidx = va_to_pteg(pm, pa);
3474 				pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3475 				pmap_pte_insert(ptegidx, &pt);
3476 			}
3477 		}
3478 
3479 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3480 		va = (vaddr_t) kernel_text;
3481 
3482 		for (pa = kernelstart; va < va_etext;
3483 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3484 			ptegidx = va_to_pteg(pm, va);
3485 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3486 			pmap_pte_insert(ptegidx, &pt);
3487 		}
3488 
3489 		for (; pa < kernelend;
3490 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3491 			ptegidx = va_to_pteg(pm, va);
3492 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3493 			pmap_pte_insert(ptegidx, &pt);
3494 		}
3495 
3496 		for (va = 0, pa = 0; va < kernelstart;
3497 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3498 			ptegidx = va_to_pteg(pm, va);
3499 			if (va < 0x3000)
3500 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3501 			else
3502 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3503 			pmap_pte_insert(ptegidx, &pt);
3504 		}
3505 		for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3506 		    pa += PAGE_SIZE, va += PAGE_SIZE) {
3507 			ptegidx = va_to_pteg(pm, va);
3508 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3509 			pmap_pte_insert(ptegidx, &pt);
3510 		}
3511 #endif
3512 
3513 		__asm volatile ("mtsrin %0,%1"
3514  			      :: "r"(sr), "r"(kernelstart));
3515 	}
3516 #endif
3517 }
3518