xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision a5847cc334d9a7029f6352b847e9e8d71a0f9e0c)
1 /*	$NetBSD: pmap.c,v 1.82 2011/07/17 20:54:46 joerg 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 1.82 2011/07/17 20:54:46 joerg 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/powerpc.h>
87 #include <powerpc/bat.h>
88 #include <powerpc/pcb.h>
89 #include <powerpc/psl.h>
90 #include <powerpc/spr.h>
91 #include <powerpc/oea/spr.h>
92 #include <powerpc/oea/sr_601.h>
93 
94 #ifdef ALTIVEC
95 int pmap_use_altivec;
96 #endif
97 
98 volatile struct pteg *pmap_pteg_table;
99 unsigned int pmap_pteg_cnt;
100 unsigned int pmap_pteg_mask;
101 #ifdef PMAP_MEMLIMIT
102 static paddr_t pmap_memlimit = PMAP_MEMLIMIT;
103 #else
104 static paddr_t pmap_memlimit = -PAGE_SIZE;		/* there is no limit */
105 #endif
106 
107 struct pmap kernel_pmap_;
108 unsigned int pmap_pages_stolen;
109 u_long pmap_pte_valid;
110 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
111 u_long pmap_pvo_enter_depth;
112 u_long pmap_pvo_remove_depth;
113 #endif
114 
115 #ifndef MSGBUFADDR
116 extern paddr_t msgbuf_paddr;
117 #endif
118 
119 static struct mem_region *mem, *avail;
120 static u_int mem_cnt, avail_cnt;
121 
122 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
123 # define	PMAP_OEA 1
124 #endif
125 
126 #if defined(PMAP_OEA)
127 #define	_PRIxpte	"lx"
128 #else
129 #define	_PRIxpte	PRIx64
130 #endif
131 #define	_PRIxpa		"lx"
132 #define	_PRIxva		"lx"
133 #define	_PRIsr  	"lx"
134 
135 #ifdef PMAP_NEEDS_FIXUP
136 #if defined(PMAP_OEA)
137 #define	PMAPNAME(name)	pmap32_##name
138 #elif defined(PMAP_OEA64)
139 #define	PMAPNAME(name)	pmap64_##name
140 #elif defined(PMAP_OEA64_BRIDGE)
141 #define	PMAPNAME(name)	pmap64bridge_##name
142 #else
143 #error unknown variant for pmap
144 #endif
145 #endif /* PMAP_NEEDS_FIXUP */
146 
147 #ifdef PMAPNAME
148 #define	STATIC			static
149 #define pmap_pte_spill		PMAPNAME(pte_spill)
150 #define pmap_real_memory	PMAPNAME(real_memory)
151 #define pmap_init		PMAPNAME(init)
152 #define pmap_virtual_space	PMAPNAME(virtual_space)
153 #define pmap_create		PMAPNAME(create)
154 #define pmap_reference		PMAPNAME(reference)
155 #define pmap_destroy		PMAPNAME(destroy)
156 #define pmap_copy		PMAPNAME(copy)
157 #define pmap_update		PMAPNAME(update)
158 #define pmap_enter		PMAPNAME(enter)
159 #define pmap_remove		PMAPNAME(remove)
160 #define pmap_kenter_pa		PMAPNAME(kenter_pa)
161 #define pmap_kremove		PMAPNAME(kremove)
162 #define pmap_extract		PMAPNAME(extract)
163 #define pmap_protect		PMAPNAME(protect)
164 #define pmap_unwire		PMAPNAME(unwire)
165 #define pmap_page_protect	PMAPNAME(page_protect)
166 #define pmap_query_bit		PMAPNAME(query_bit)
167 #define pmap_clear_bit		PMAPNAME(clear_bit)
168 
169 #define pmap_activate		PMAPNAME(activate)
170 #define pmap_deactivate		PMAPNAME(deactivate)
171 
172 #define pmap_pinit		PMAPNAME(pinit)
173 #define pmap_procwr		PMAPNAME(procwr)
174 
175 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
176 #define pmap_pte_print		PMAPNAME(pte_print)
177 #define pmap_pteg_check		PMAPNAME(pteg_check)
178 #define pmap_print_mmruregs	PMAPNAME(print_mmuregs)
179 #define pmap_print_pte		PMAPNAME(print_pte)
180 #define pmap_pteg_dist		PMAPNAME(pteg_dist)
181 #endif
182 #if defined(DEBUG) || defined(PMAPCHECK)
183 #define	pmap_pvo_verify		PMAPNAME(pvo_verify)
184 #define pmapcheck		PMAPNAME(check)
185 #endif
186 #if defined(DEBUG) || defined(PMAPDEBUG)
187 #define pmapdebug		PMAPNAME(debug)
188 #endif
189 #define pmap_steal_memory	PMAPNAME(steal_memory)
190 #define pmap_bootstrap		PMAPNAME(bootstrap)
191 #else
192 #define	STATIC			/* nothing */
193 #endif /* PMAPNAME */
194 
195 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
196 STATIC void pmap_real_memory(paddr_t *, psize_t *);
197 STATIC void pmap_init(void);
198 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
199 STATIC pmap_t pmap_create(void);
200 STATIC void pmap_reference(pmap_t);
201 STATIC void pmap_destroy(pmap_t);
202 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
203 STATIC void pmap_update(pmap_t);
204 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
205 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
206 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t, u_int);
207 STATIC void pmap_kremove(vaddr_t, vsize_t);
208 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
209 
210 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
211 STATIC void pmap_unwire(pmap_t, vaddr_t);
212 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
213 STATIC bool pmap_query_bit(struct vm_page *, int);
214 STATIC bool pmap_clear_bit(struct vm_page *, int);
215 
216 STATIC void pmap_activate(struct lwp *);
217 STATIC void pmap_deactivate(struct lwp *);
218 
219 STATIC void pmap_pinit(pmap_t pm);
220 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
221 
222 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
223 STATIC void pmap_pte_print(volatile struct pte *);
224 STATIC void pmap_pteg_check(void);
225 STATIC void pmap_print_mmuregs(void);
226 STATIC void pmap_print_pte(pmap_t, vaddr_t);
227 STATIC void pmap_pteg_dist(void);
228 #endif
229 #if defined(DEBUG) || defined(PMAPCHECK)
230 STATIC void pmap_pvo_verify(void);
231 #endif
232 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
233 STATIC void pmap_bootstrap(paddr_t, paddr_t);
234 
235 #ifdef PMAPNAME
236 const struct pmap_ops PMAPNAME(ops) = {
237 	.pmapop_pte_spill = pmap_pte_spill,
238 	.pmapop_real_memory = pmap_real_memory,
239 	.pmapop_init = pmap_init,
240 	.pmapop_virtual_space = pmap_virtual_space,
241 	.pmapop_create = pmap_create,
242 	.pmapop_reference = pmap_reference,
243 	.pmapop_destroy = pmap_destroy,
244 	.pmapop_copy = pmap_copy,
245 	.pmapop_update = pmap_update,
246 	.pmapop_enter = pmap_enter,
247 	.pmapop_remove = pmap_remove,
248 	.pmapop_kenter_pa = pmap_kenter_pa,
249 	.pmapop_kremove = pmap_kremove,
250 	.pmapop_extract = pmap_extract,
251 	.pmapop_protect = pmap_protect,
252 	.pmapop_unwire = pmap_unwire,
253 	.pmapop_page_protect = pmap_page_protect,
254 	.pmapop_query_bit = pmap_query_bit,
255 	.pmapop_clear_bit = pmap_clear_bit,
256 	.pmapop_activate = pmap_activate,
257 	.pmapop_deactivate = pmap_deactivate,
258 	.pmapop_pinit = pmap_pinit,
259 	.pmapop_procwr = pmap_procwr,
260 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
261 	.pmapop_pte_print = pmap_pte_print,
262 	.pmapop_pteg_check = pmap_pteg_check,
263 	.pmapop_print_mmuregs = pmap_print_mmuregs,
264 	.pmapop_print_pte = pmap_print_pte,
265 	.pmapop_pteg_dist = pmap_pteg_dist,
266 #else
267 	.pmapop_pte_print = NULL,
268 	.pmapop_pteg_check = NULL,
269 	.pmapop_print_mmuregs = NULL,
270 	.pmapop_print_pte = NULL,
271 	.pmapop_pteg_dist = NULL,
272 #endif
273 #if defined(DEBUG) || defined(PMAPCHECK)
274 	.pmapop_pvo_verify = pmap_pvo_verify,
275 #else
276 	.pmapop_pvo_verify = NULL,
277 #endif
278 	.pmapop_steal_memory = pmap_steal_memory,
279 	.pmapop_bootstrap = pmap_bootstrap,
280 };
281 #endif /* !PMAPNAME */
282 
283 /*
284  * The following structure is aligned to 32 bytes
285  */
286 struct pvo_entry {
287 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
288 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
289 	struct pte pvo_pte;			/* Prebuilt PTE */
290 	pmap_t pvo_pmap;			/* ptr to owning pmap */
291 	vaddr_t pvo_vaddr;			/* VA of entry */
292 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
293 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
294 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
295 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
296 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
297 #define	PVO_WIRED_P(pvo)	((pvo)->pvo_vaddr & PVO_WIRED)
298 #define	PVO_MANAGED_P(pvo)	((pvo)->pvo_vaddr & PVO_MANAGED)
299 #define	PVO_EXECUTABLE_P(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
300 #define	PVO_ENTER_INSERT	0		/* PVO has been removed */
301 #define	PVO_SPILL_UNSET		1		/* PVO has been evicted */
302 #define	PVO_SPILL_SET		2		/* PVO has been spilled */
303 #define	PVO_SPILL_INSERT	3		/* PVO has been inserted */
304 #define	PVO_PMAP_PAGE_PROTECT	4		/* PVO has changed */
305 #define	PVO_PMAP_PROTECT	5		/* PVO has changed */
306 #define	PVO_REMOVE		6		/* PVO has been removed */
307 #define	PVO_WHERE_MASK		15
308 #define	PVO_WHERE_SHFT		8
309 } __attribute__ ((aligned (32)));
310 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
311 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
312 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
313 #define	PVO_PTEGIDX_CLR(pvo)	\
314 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
315 #define	PVO_PTEGIDX_SET(pvo,i)	\
316 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
317 #define	PVO_WHERE(pvo,w)	\
318 	((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
319 	 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
320 
321 TAILQ_HEAD(pvo_tqhead, pvo_entry);
322 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
323 static struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
324 static struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
325 
326 struct pool pmap_pool;		/* pool for pmap structures */
327 struct pool pmap_upvo_pool;	/* pool for pvo entries for unmanaged pages */
328 struct pool pmap_mpvo_pool;	/* pool for pvo entries for managed pages */
329 
330 /*
331  * We keep a cache of unmanaged pages to be used for pvo entries for
332  * unmanaged pages.
333  */
334 struct pvo_page {
335 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
336 };
337 SIMPLEQ_HEAD(pvop_head, pvo_page);
338 static struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head);
339 static struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head);
340 u_long pmap_upvop_free;
341 u_long pmap_upvop_maxfree;
342 u_long pmap_mpvop_free;
343 u_long pmap_mpvop_maxfree;
344 
345 static void *pmap_pool_ualloc(struct pool *, int);
346 static void *pmap_pool_malloc(struct pool *, int);
347 
348 static void pmap_pool_ufree(struct pool *, void *);
349 static void pmap_pool_mfree(struct pool *, void *);
350 
351 static struct pool_allocator pmap_pool_mallocator = {
352 	.pa_alloc = pmap_pool_malloc,
353 	.pa_free = pmap_pool_mfree,
354 	.pa_pagesz = 0,
355 };
356 
357 static struct pool_allocator pmap_pool_uallocator = {
358 	.pa_alloc = pmap_pool_ualloc,
359 	.pa_free = pmap_pool_ufree,
360 	.pa_pagesz = 0,
361 };
362 
363 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
364 void pmap_pte_print(volatile struct pte *);
365 void pmap_pteg_check(void);
366 void pmap_pteg_dist(void);
367 void pmap_print_pte(pmap_t, vaddr_t);
368 void pmap_print_mmuregs(void);
369 #endif
370 
371 #if defined(DEBUG) || defined(PMAPCHECK)
372 #ifdef PMAPCHECK
373 int pmapcheck = 1;
374 #else
375 int pmapcheck = 0;
376 #endif
377 void pmap_pvo_verify(void);
378 static void pmap_pvo_check(const struct pvo_entry *);
379 #define	PMAP_PVO_CHECK(pvo)	 		\
380 	do {					\
381 		if (pmapcheck)			\
382 			pmap_pvo_check(pvo);	\
383 	} while (0)
384 #else
385 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
386 #endif
387 static int pmap_pte_insert(int, struct pte *);
388 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
389 	vaddr_t, paddr_t, register_t, int);
390 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
391 static void pmap_pvo_free(struct pvo_entry *);
392 static void pmap_pvo_free_list(struct pvo_head *);
393 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
394 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
395 static struct pvo_entry *pmap_pvo_reclaim(struct pmap *);
396 static void pvo_set_exec(struct pvo_entry *);
397 static void pvo_clear_exec(struct pvo_entry *);
398 
399 static void tlbia(void);
400 
401 static void pmap_release(pmap_t);
402 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
403 
404 static uint32_t pmap_pvo_reclaim_nextidx;
405 #ifdef DEBUG
406 static int pmap_pvo_reclaim_debugctr;
407 #endif
408 
409 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
410 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
411 
412 static int pmap_initialized;
413 
414 #if defined(DEBUG) || defined(PMAPDEBUG)
415 #define	PMAPDEBUG_BOOT		0x0001
416 #define	PMAPDEBUG_PTE		0x0002
417 #define	PMAPDEBUG_EXEC		0x0008
418 #define	PMAPDEBUG_PVOENTER	0x0010
419 #define	PMAPDEBUG_PVOREMOVE	0x0020
420 #define	PMAPDEBUG_ACTIVATE	0x0100
421 #define	PMAPDEBUG_CREATE	0x0200
422 #define	PMAPDEBUG_ENTER		0x1000
423 #define	PMAPDEBUG_KENTER	0x2000
424 #define	PMAPDEBUG_KREMOVE	0x4000
425 #define	PMAPDEBUG_REMOVE	0x8000
426 
427 unsigned int pmapdebug = 0;
428 
429 # define DPRINTF(x)		printf x
430 # define DPRINTFN(n, x)		if (pmapdebug & PMAPDEBUG_ ## n) printf x
431 #else
432 # define DPRINTF(x)
433 # define DPRINTFN(n, x)
434 #endif
435 
436 
437 #ifdef PMAPCOUNTERS
438 /*
439  * From pmap_subr.c
440  */
441 extern struct evcnt pmap_evcnt_mappings;
442 extern struct evcnt pmap_evcnt_unmappings;
443 
444 extern struct evcnt pmap_evcnt_kernel_mappings;
445 extern struct evcnt pmap_evcnt_kernel_unmappings;
446 
447 extern struct evcnt pmap_evcnt_mappings_replaced;
448 
449 extern struct evcnt pmap_evcnt_exec_mappings;
450 extern struct evcnt pmap_evcnt_exec_cached;
451 
452 extern struct evcnt pmap_evcnt_exec_synced;
453 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
454 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
455 
456 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
457 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
458 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
459 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
460 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
461 
462 extern struct evcnt pmap_evcnt_updates;
463 extern struct evcnt pmap_evcnt_collects;
464 extern struct evcnt pmap_evcnt_copies;
465 
466 extern struct evcnt pmap_evcnt_ptes_spilled;
467 extern struct evcnt pmap_evcnt_ptes_unspilled;
468 extern struct evcnt pmap_evcnt_ptes_evicted;
469 
470 extern struct evcnt pmap_evcnt_ptes_primary[8];
471 extern struct evcnt pmap_evcnt_ptes_secondary[8];
472 extern struct evcnt pmap_evcnt_ptes_removed;
473 extern struct evcnt pmap_evcnt_ptes_changed;
474 extern struct evcnt pmap_evcnt_pvos_reclaimed;
475 extern struct evcnt pmap_evcnt_pvos_failed;
476 
477 extern struct evcnt pmap_evcnt_zeroed_pages;
478 extern struct evcnt pmap_evcnt_copied_pages;
479 extern struct evcnt pmap_evcnt_idlezeroed_pages;
480 
481 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
482 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
483 #else
484 #define	PMAPCOUNT(ev)	((void) 0)
485 #define	PMAPCOUNT2(ev)	((void) 0)
486 #endif
487 
488 #define	TLBIE(va)	__asm volatile("tlbie %0" :: "r"(va))
489 
490 /* XXXSL: this needs to be moved to assembler */
491 #define	TLBIEL(va)	__asm __volatile("tlbie %0" :: "r"(va))
492 
493 #define	TLBSYNC()	__asm volatile("tlbsync")
494 #define	SYNC()		__asm volatile("sync")
495 #define	EIEIO()		__asm volatile("eieio")
496 #define	DCBST(va)	__asm __volatile("dcbst 0,%0" :: "r"(va))
497 #define	MFMSR()		mfmsr()
498 #define	MTMSR(psl)	mtmsr(psl)
499 #define	MFPVR()		mfpvr()
500 #define	MFSRIN(va)	mfsrin(va)
501 #define	MFTB()		mfrtcltbl()
502 
503 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
504 static inline register_t
505 mfsrin(vaddr_t va)
506 {
507 	register_t sr;
508 	__asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
509 	return sr;
510 }
511 #endif	/* PMAP_OEA*/
512 
513 #if defined (PMAP_OEA64_BRIDGE)
514 extern void mfmsr64 (register64_t *result);
515 #endif /* PMAP_OEA64_BRIDGE */
516 
517 #define	PMAP_LOCK()		KERNEL_LOCK(1, NULL)
518 #define	PMAP_UNLOCK()		KERNEL_UNLOCK_ONE(NULL)
519 
520 static inline register_t
521 pmap_interrupts_off(void)
522 {
523 	register_t msr = MFMSR();
524 	if (msr & PSL_EE)
525 		MTMSR(msr & ~PSL_EE);
526 	return msr;
527 }
528 
529 static void
530 pmap_interrupts_restore(register_t msr)
531 {
532 	if (msr & PSL_EE)
533 		MTMSR(msr);
534 }
535 
536 static inline u_int32_t
537 mfrtcltbl(void)
538 {
539 #ifdef PPC_OEA601
540 	if ((MFPVR() >> 16) == MPC601)
541 		return (mfrtcl() >> 7);
542 	else
543 #endif
544 		return (mftbl());
545 }
546 
547 /*
548  * These small routines may have to be replaced,
549  * if/when we support processors other that the 604.
550  */
551 
552 void
553 tlbia(void)
554 {
555 	char *i;
556 
557 	SYNC();
558 #if defined(PMAP_OEA)
559 	/*
560 	 * Why not use "tlbia"?  Because not all processors implement it.
561 	 *
562 	 * This needs to be a per-CPU callback to do the appropriate thing
563 	 * for the CPU. XXX
564 	 */
565 	for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
566 		TLBIE(i);
567 		EIEIO();
568 		SYNC();
569 	}
570 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
571 	/* This is specifically for the 970, 970UM v1.6 pp. 140. */
572 	for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
573 		TLBIEL(i);
574 		EIEIO();
575 		SYNC();
576 	}
577 #endif
578 	TLBSYNC();
579 	SYNC();
580 }
581 
582 static inline register_t
583 va_to_vsid(const struct pmap *pm, vaddr_t addr)
584 {
585 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
586 	return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT;
587 #else /* PMAP_OEA64 */
588 #if 0
589 	const struct ste *ste;
590 	register_t hash;
591 	int i;
592 
593 	hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH;
594 
595 	/*
596 	 * Try the primary group first
597 	 */
598 	ste = pm->pm_stes[hash].stes;
599 	for (i = 0; i < 8; i++, ste++) {
600 		if (ste->ste_hi & STE_V) &&
601 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
602 			return ste;
603 	}
604 
605 	/*
606 	 * Then the secondary group.
607 	 */
608 	ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes;
609 	for (i = 0; i < 8; i++, ste++) {
610 		if (ste->ste_hi & STE_V) &&
611 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
612 			return addr;
613 	}
614 
615 	return NULL;
616 #else
617 	/*
618 	 * Rather than searching the STE groups for the VSID, we know
619 	 * how we generate that from the ESID and so do that.
620 	 */
621 	return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
622 #endif
623 #endif /* PMAP_OEA */
624 }
625 
626 static inline register_t
627 va_to_pteg(const struct pmap *pm, vaddr_t addr)
628 {
629 	register_t hash;
630 
631 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
632 	return hash & pmap_pteg_mask;
633 }
634 
635 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
636 /*
637  * Given a PTE in the page table, calculate the VADDR that hashes to it.
638  * The only bit of magic is that the top 4 bits of the address doesn't
639  * technically exist in the PTE.  But we know we reserved 4 bits of the
640  * VSID for it so that's how we get it.
641  */
642 static vaddr_t
643 pmap_pte_to_va(volatile const struct pte *pt)
644 {
645 	vaddr_t va;
646 	uintptr_t ptaddr = (uintptr_t) pt;
647 
648 	if (pt->pte_hi & PTE_HID)
649 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
650 
651 	/* PPC Bits 10-19  PPC64 Bits 42-51 */
652 #if defined(PMAP_OEA)
653 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
654 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
655 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
656 #endif
657 	va <<= ADDR_PIDX_SHFT;
658 
659 	/* PPC Bits 4-9  PPC64 Bits 36-41 */
660 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
661 
662 #if defined(PMAP_OEA64)
663 	/* PPC63 Bits 0-35 */
664 	/* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
665 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
666 	/* PPC Bits 0-3 */
667 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
668 #endif
669 
670 	return va;
671 }
672 #endif
673 
674 static inline struct pvo_head *
675 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
676 {
677 	struct vm_page *pg;
678 	struct vm_page_md *md;
679 
680 	pg = PHYS_TO_VM_PAGE(pa);
681 	if (pg_p != NULL)
682 		*pg_p = pg;
683 	if (pg == NULL)
684 		return &pmap_pvo_unmanaged;
685 	md = VM_PAGE_TO_MD(pg);
686 	return &md->mdpg_pvoh;
687 }
688 
689 static inline struct pvo_head *
690 vm_page_to_pvoh(struct vm_page *pg)
691 {
692 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
693 
694 	return &md->mdpg_pvoh;
695 }
696 
697 
698 static inline void
699 pmap_attr_clear(struct vm_page *pg, int ptebit)
700 {
701 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
702 
703 	md->mdpg_attrs &= ~ptebit;
704 }
705 
706 static inline int
707 pmap_attr_fetch(struct vm_page *pg)
708 {
709 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
710 
711 	return md->mdpg_attrs;
712 }
713 
714 static inline void
715 pmap_attr_save(struct vm_page *pg, int ptebit)
716 {
717 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
718 
719 	md->mdpg_attrs |= ptebit;
720 }
721 
722 static inline int
723 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
724 {
725 	if (pt->pte_hi == pvo_pt->pte_hi
726 #if 0
727 	    && ((pt->pte_lo ^ pvo_pt->pte_lo) &
728 	        ~(PTE_REF|PTE_CHG)) == 0
729 #endif
730 	    )
731 		return 1;
732 	return 0;
733 }
734 
735 static inline void
736 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
737 {
738 	/*
739 	 * Construct the PTE.  Default to IMB initially.  Valid bit
740 	 * only gets set when the real pte is set in memory.
741 	 *
742 	 * Note: Don't set the valid bit for correct operation of tlb update.
743 	 */
744 #if defined(PMAP_OEA)
745 	pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
746 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
747 	pt->pte_lo = pte_lo;
748 #elif defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA64)
749 	pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT)
750 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
751 	pt->pte_lo = (u_int64_t) pte_lo;
752 #endif /* PMAP_OEA */
753 }
754 
755 static inline void
756 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
757 {
758 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
759 }
760 
761 static inline void
762 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
763 {
764 	/*
765 	 * As shown in Section 7.6.3.2.3
766 	 */
767 	pt->pte_lo &= ~ptebit;
768 	TLBIE(va);
769 	SYNC();
770 	EIEIO();
771 	TLBSYNC();
772 	SYNC();
773 #ifdef MULTIPROCESSOR
774 	DCBST(pt);
775 #endif
776 }
777 
778 static inline void
779 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
780 {
781 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
782 	if (pvo_pt->pte_hi & PTE_VALID)
783 		panic("pte_set: setting an already valid pte %p", pvo_pt);
784 #endif
785 	pvo_pt->pte_hi |= PTE_VALID;
786 
787 	/*
788 	 * Update the PTE as defined in section 7.6.3.1
789 	 * Note that the REF/CHG bits are from pvo_pt and thus should
790 	 * have been saved so this routine can restore them (if desired).
791 	 */
792 	pt->pte_lo = pvo_pt->pte_lo;
793 	EIEIO();
794 	pt->pte_hi = pvo_pt->pte_hi;
795 	TLBSYNC();
796 	SYNC();
797 #ifdef MULTIPROCESSOR
798 	DCBST(pt);
799 #endif
800 	pmap_pte_valid++;
801 }
802 
803 static inline void
804 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
805 {
806 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
807 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
808 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
809 	if ((pt->pte_hi & PTE_VALID) == 0)
810 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
811 #endif
812 
813 	pvo_pt->pte_hi &= ~PTE_VALID;
814 	/*
815 	 * Force the ref & chg bits back into the PTEs.
816 	 */
817 	SYNC();
818 	/*
819 	 * Invalidate the pte ... (Section 7.6.3.3)
820 	 */
821 	pt->pte_hi &= ~PTE_VALID;
822 	SYNC();
823 	TLBIE(va);
824 	SYNC();
825 	EIEIO();
826 	TLBSYNC();
827 	SYNC();
828 	/*
829 	 * Save the ref & chg bits ...
830 	 */
831 	pmap_pte_synch(pt, pvo_pt);
832 	pmap_pte_valid--;
833 }
834 
835 static inline void
836 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
837 {
838 	/*
839 	 * Invalidate the PTE
840 	 */
841 	pmap_pte_unset(pt, pvo_pt, va);
842 	pmap_pte_set(pt, pvo_pt);
843 }
844 
845 /*
846  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
847  * (either primary or secondary location).
848  *
849  * Note: both the destination and source PTEs must not have PTE_VALID set.
850  */
851 
852 static int
853 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
854 {
855 	volatile struct pte *pt;
856 	int i;
857 
858 #if defined(DEBUG)
859 	DPRINTFN(PTE, ("pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
860 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo));
861 #endif
862 	/*
863 	 * First try primary hash.
864 	 */
865 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
866 		if ((pt->pte_hi & PTE_VALID) == 0) {
867 			pvo_pt->pte_hi &= ~PTE_HID;
868 			pmap_pte_set(pt, pvo_pt);
869 			return i;
870 		}
871 	}
872 
873 	/*
874 	 * Now try secondary hash.
875 	 */
876 	ptegidx ^= pmap_pteg_mask;
877 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
878 		if ((pt->pte_hi & PTE_VALID) == 0) {
879 			pvo_pt->pte_hi |= PTE_HID;
880 			pmap_pte_set(pt, pvo_pt);
881 			return i;
882 		}
883 	}
884 	return -1;
885 }
886 
887 /*
888  * Spill handler.
889  *
890  * Tries to spill a page table entry from the overflow area.
891  * This runs in either real mode (if dealing with a exception spill)
892  * or virtual mode when dealing with manually spilling one of the
893  * kernel's pte entries.  In either case, interrupts are already
894  * disabled.
895  */
896 
897 int
898 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec)
899 {
900 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
901 	struct pvo_entry *pvo;
902 	/* XXX: gcc -- vpvoh is always set at either *1* or *2* */
903 	struct pvo_tqhead *pvoh, *vpvoh = NULL;
904 	int ptegidx, i, j;
905 	volatile struct pteg *pteg;
906 	volatile struct pte *pt;
907 
908 	PMAP_LOCK();
909 
910 	ptegidx = va_to_pteg(pm, addr);
911 
912 	/*
913 	 * Have to substitute some entry. Use the primary hash for this.
914 	 * Use low bits of timebase as random generator.  Make sure we are
915 	 * not picking a kernel pte for replacement.
916 	 */
917 	pteg = &pmap_pteg_table[ptegidx];
918 	i = MFTB() & 7;
919 	for (j = 0; j < 8; j++) {
920 		pt = &pteg->pt[i];
921 		if ((pt->pte_hi & PTE_VALID) == 0)
922 			break;
923 		if (VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT)
924 				< PHYSMAP_VSIDBITS)
925 			break;
926 		i = (i + 1) & 7;
927 	}
928 	KASSERT(j < 8);
929 
930 	source_pvo = NULL;
931 	victim_pvo = NULL;
932 	pvoh = &pmap_pvo_table[ptegidx];
933 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
934 
935 		/*
936 		 * We need to find pvo entry for this address...
937 		 */
938 		PMAP_PVO_CHECK(pvo);		/* sanity check */
939 
940 		/*
941 		 * If we haven't found the source and we come to a PVO with
942 		 * a valid PTE, then we know we can't find it because all
943 		 * evicted PVOs always are first in the list.
944 		 */
945 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
946 			break;
947 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
948 		    addr == PVO_VADDR(pvo)) {
949 
950 			/*
951 			 * Now we have found the entry to be spilled into the
952 			 * pteg.  Attempt to insert it into the page table.
953 			 */
954 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
955 			if (j >= 0) {
956 				PVO_PTEGIDX_SET(pvo, j);
957 				PMAP_PVO_CHECK(pvo);	/* sanity check */
958 				PVO_WHERE(pvo, SPILL_INSERT);
959 				pvo->pvo_pmap->pm_evictions--;
960 				PMAPCOUNT(ptes_spilled);
961 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
962 				    ? pmap_evcnt_ptes_secondary
963 				    : pmap_evcnt_ptes_primary)[j]);
964 
965 				/*
966 				 * Since we keep the evicted entries at the
967 				 * from of the PVO list, we need move this
968 				 * (now resident) PVO after the evicted
969 				 * entries.
970 				 */
971 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
972 
973 				/*
974 				 * If we don't have to move (either we were the
975 				 * last entry or the next entry was valid),
976 				 * don't change our position.  Otherwise
977 				 * move ourselves to the tail of the queue.
978 				 */
979 				if (next_pvo != NULL &&
980 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
981 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
982 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
983 				}
984 				PMAP_UNLOCK();
985 				return 1;
986 			}
987 			source_pvo = pvo;
988 			if (exec && !PVO_EXECUTABLE_P(source_pvo)) {
989 				return 0;
990 			}
991 			if (victim_pvo != NULL)
992 				break;
993 		}
994 
995 		/*
996 		 * We also need the pvo entry of the victim we are replacing
997 		 * so save the R & C bits of the PTE.
998 		 */
999 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
1000 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
1001 			vpvoh = pvoh;			/* *1* */
1002 			victim_pvo = pvo;
1003 			if (source_pvo != NULL)
1004 				break;
1005 		}
1006 	}
1007 
1008 	if (source_pvo == NULL) {
1009 		PMAPCOUNT(ptes_unspilled);
1010 		PMAP_UNLOCK();
1011 		return 0;
1012 	}
1013 
1014 	if (victim_pvo == NULL) {
1015 		if ((pt->pte_hi & PTE_HID) == 0)
1016 			panic("pmap_pte_spill: victim p-pte (%p) has "
1017 			    "no pvo entry!", pt);
1018 
1019 		/*
1020 		 * If this is a secondary PTE, we need to search
1021 		 * its primary pvo bucket for the matching PVO.
1022 		 */
1023 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */
1024 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1025 			PMAP_PVO_CHECK(pvo);		/* sanity check */
1026 
1027 			/*
1028 			 * We also need the pvo entry of the victim we are
1029 			 * replacing so save the R & C bits of the PTE.
1030 			 */
1031 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1032 				victim_pvo = pvo;
1033 				break;
1034 			}
1035 		}
1036 		if (victim_pvo == NULL)
1037 			panic("pmap_pte_spill: victim s-pte (%p) has "
1038 			    "no pvo entry!", pt);
1039 	}
1040 
1041 	/*
1042 	 * The victim should be not be a kernel PVO/PTE entry.
1043 	 */
1044 	KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1045 	KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1046 	KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1047 
1048 	/*
1049 	 * We are invalidating the TLB entry for the EA for the
1050 	 * we are replacing even though its valid; If we don't
1051 	 * we lose any ref/chg bit changes contained in the TLB
1052 	 * entry.
1053 	 */
1054 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1055 
1056 	/*
1057 	 * To enforce the PVO list ordering constraint that all
1058 	 * evicted entries should come before all valid entries,
1059 	 * move the source PVO to the tail of its list and the
1060 	 * victim PVO to the head of its list (which might not be
1061 	 * the same list, if the victim was using the secondary hash).
1062 	 */
1063 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
1064 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
1065 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1066 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1067 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1068 	pmap_pte_set(pt, &source_pvo->pvo_pte);
1069 	victim_pvo->pvo_pmap->pm_evictions++;
1070 	source_pvo->pvo_pmap->pm_evictions--;
1071 	PVO_WHERE(victim_pvo, SPILL_UNSET);
1072 	PVO_WHERE(source_pvo, SPILL_SET);
1073 
1074 	PVO_PTEGIDX_CLR(victim_pvo);
1075 	PVO_PTEGIDX_SET(source_pvo, i);
1076 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1077 	PMAPCOUNT(ptes_spilled);
1078 	PMAPCOUNT(ptes_evicted);
1079 	PMAPCOUNT(ptes_removed);
1080 
1081 	PMAP_PVO_CHECK(victim_pvo);
1082 	PMAP_PVO_CHECK(source_pvo);
1083 
1084 	PMAP_UNLOCK();
1085 	return 1;
1086 }
1087 
1088 /*
1089  * Restrict given range to physical memory
1090  */
1091 void
1092 pmap_real_memory(paddr_t *start, psize_t *size)
1093 {
1094 	struct mem_region *mp;
1095 
1096 	for (mp = mem; mp->size; mp++) {
1097 		if (*start + *size > mp->start
1098 		    && *start < mp->start + mp->size) {
1099 			if (*start < mp->start) {
1100 				*size -= mp->start - *start;
1101 				*start = mp->start;
1102 			}
1103 			if (*start + *size > mp->start + mp->size)
1104 				*size = mp->start + mp->size - *start;
1105 			return;
1106 		}
1107 	}
1108 	*size = 0;
1109 }
1110 
1111 /*
1112  * Initialize anything else for pmap handling.
1113  * Called during vm_init().
1114  */
1115 void
1116 pmap_init(void)
1117 {
1118 	pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry),
1119 	    sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl",
1120 	    &pmap_pool_mallocator, IPL_NONE);
1121 
1122 	pool_setlowat(&pmap_mpvo_pool, 1008);
1123 
1124 	pmap_initialized = 1;
1125 
1126 }
1127 
1128 /*
1129  * How much virtual space does the kernel get?
1130  */
1131 void
1132 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1133 {
1134 	/*
1135 	 * For now, reserve one segment (minus some overhead) for kernel
1136 	 * virtual memory
1137 	 */
1138 	*start = VM_MIN_KERNEL_ADDRESS;
1139 	*end = VM_MAX_KERNEL_ADDRESS;
1140 }
1141 
1142 /*
1143  * Allocate, initialize, and return a new physical map.
1144  */
1145 pmap_t
1146 pmap_create(void)
1147 {
1148 	pmap_t pm;
1149 
1150 	pm = pool_get(&pmap_pool, PR_WAITOK);
1151 	memset((void *)pm, 0, sizeof *pm);
1152 	pmap_pinit(pm);
1153 
1154 	DPRINTFN(CREATE,("pmap_create: pm %p:\n"
1155 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1156 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1157 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1158 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1159 	    pm,
1160 	    pm->pm_sr[0], pm->pm_sr[1],
1161 	    pm->pm_sr[2], pm->pm_sr[3],
1162 	    pm->pm_sr[4], pm->pm_sr[5],
1163 	    pm->pm_sr[6], pm->pm_sr[7],
1164 	    pm->pm_sr[8], pm->pm_sr[9],
1165 	    pm->pm_sr[10], pm->pm_sr[11],
1166 	    pm->pm_sr[12], pm->pm_sr[13],
1167 	    pm->pm_sr[14], pm->pm_sr[15]));
1168 	return pm;
1169 }
1170 
1171 /*
1172  * Initialize a preallocated and zeroed pmap structure.
1173  */
1174 void
1175 pmap_pinit(pmap_t pm)
1176 {
1177 	register_t entropy = MFTB();
1178 	register_t mask;
1179 	int i;
1180 
1181 	/*
1182 	 * Allocate some segment registers for this pmap.
1183 	 */
1184 	pm->pm_refs = 1;
1185 	PMAP_LOCK();
1186 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1187 		static register_t pmap_vsidcontext;
1188 		register_t hash;
1189 		unsigned int n;
1190 
1191 		/* Create a new value by multiplying by a prime adding in
1192 		 * entropy from the timebase register.  This is to make the
1193 		 * VSID more random so that the PT Hash function collides
1194 		 * less often. (note that the prime causes gcc to do shifts
1195 		 * instead of a multiply)
1196 		 */
1197 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1198 		hash = pmap_vsidcontext & (NPMAPS - 1);
1199 		if (hash == 0) {		/* 0 is special, avoid it */
1200 			entropy += 0xbadf00d;
1201 			continue;
1202 		}
1203 		n = hash >> 5;
1204 		mask = 1L << (hash & (VSID_NBPW-1));
1205 		hash = pmap_vsidcontext;
1206 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1207 			/* anything free in this bucket? */
1208 			if (~pmap_vsid_bitmap[n] == 0) {
1209 				entropy = hash ^ (hash >> 16);
1210 				continue;
1211 			}
1212 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1213 			mask = 1L << i;
1214 			hash &= ~(VSID_NBPW-1);
1215 			hash |= i;
1216 		}
1217 		hash &= PTE_VSID >> PTE_VSID_SHFT;
1218 		pmap_vsid_bitmap[n] |= mask;
1219 		pm->pm_vsid = hash;
1220 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1221 		for (i = 0; i < 16; i++)
1222 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1223 			    SR_NOEXEC;
1224 #endif
1225 		PMAP_UNLOCK();
1226 		return;
1227 	}
1228 	PMAP_UNLOCK();
1229 	panic("pmap_pinit: out of segments");
1230 }
1231 
1232 /*
1233  * Add a reference to the given pmap.
1234  */
1235 void
1236 pmap_reference(pmap_t pm)
1237 {
1238 	atomic_inc_uint(&pm->pm_refs);
1239 }
1240 
1241 /*
1242  * Retire the given pmap from service.
1243  * Should only be called if the map contains no valid mappings.
1244  */
1245 void
1246 pmap_destroy(pmap_t pm)
1247 {
1248 	if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1249 		pmap_release(pm);
1250 		pool_put(&pmap_pool, pm);
1251 	}
1252 }
1253 
1254 /*
1255  * Release any resources held by the given physical map.
1256  * Called when a pmap initialized by pmap_pinit is being released.
1257  */
1258 void
1259 pmap_release(pmap_t pm)
1260 {
1261 	int idx, mask;
1262 
1263 	KASSERT(pm->pm_stats.resident_count == 0);
1264 	KASSERT(pm->pm_stats.wired_count == 0);
1265 
1266 	PMAP_LOCK();
1267 	if (pm->pm_sr[0] == 0)
1268 		panic("pmap_release");
1269 	idx = pm->pm_vsid & (NPMAPS-1);
1270 	mask = 1 << (idx % VSID_NBPW);
1271 	idx /= VSID_NBPW;
1272 
1273 	KASSERT(pmap_vsid_bitmap[idx] & mask);
1274 	pmap_vsid_bitmap[idx] &= ~mask;
1275 	PMAP_UNLOCK();
1276 }
1277 
1278 /*
1279  * Copy the range specified by src_addr/len
1280  * from the source map to the range dst_addr/len
1281  * in the destination map.
1282  *
1283  * This routine is only advisory and need not do anything.
1284  */
1285 void
1286 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1287 	vsize_t len, vaddr_t src_addr)
1288 {
1289 	PMAPCOUNT(copies);
1290 }
1291 
1292 /*
1293  * Require that all active physical maps contain no
1294  * incorrect entries NOW.
1295  */
1296 void
1297 pmap_update(struct pmap *pmap)
1298 {
1299 	PMAPCOUNT(updates);
1300 	TLBSYNC();
1301 }
1302 
1303 static inline int
1304 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1305 {
1306 	int pteidx;
1307 	/*
1308 	 * We can find the actual pte entry without searching by
1309 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1310 	 * and by noticing the HID bit.
1311 	 */
1312 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1313 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1314 		pteidx ^= pmap_pteg_mask * 8;
1315 	return pteidx;
1316 }
1317 
1318 volatile struct pte *
1319 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1320 {
1321 	volatile struct pte *pt;
1322 
1323 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1324 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1325 		return NULL;
1326 #endif
1327 
1328 	/*
1329 	 * If we haven't been supplied the ptegidx, calculate it.
1330 	 */
1331 	if (pteidx == -1) {
1332 		int ptegidx;
1333 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1334 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1335 	}
1336 
1337 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1338 
1339 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1340 	return pt;
1341 #else
1342 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1343 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1344 		    "pvo but no valid pte index", pvo);
1345 	}
1346 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1347 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1348 		    "pvo but no valid pte", pvo);
1349 	}
1350 
1351 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1352 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1353 #if defined(DEBUG) || defined(PMAPCHECK)
1354 			pmap_pte_print(pt);
1355 #endif
1356 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1357 			    "pmap_pteg_table %p but invalid in pvo",
1358 			    pvo, pt);
1359 		}
1360 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1361 #if defined(DEBUG) || defined(PMAPCHECK)
1362 			pmap_pte_print(pt);
1363 #endif
1364 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1365 			    "not match pte %p in pmap_pteg_table",
1366 			    pvo, pt);
1367 		}
1368 		return pt;
1369 	}
1370 
1371 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1372 #if defined(DEBUG) || defined(PMAPCHECK)
1373 		pmap_pte_print(pt);
1374 #endif
1375 		panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1376 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1377 	}
1378 	return NULL;
1379 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1380 }
1381 
1382 struct pvo_entry *
1383 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1384 {
1385 	struct pvo_entry *pvo;
1386 	int ptegidx;
1387 
1388 	va &= ~ADDR_POFF;
1389 	ptegidx = va_to_pteg(pm, va);
1390 
1391 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1392 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1393 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1394 			panic("pmap_pvo_find_va: invalid pvo %p on "
1395 			    "list %#x (%p)", pvo, ptegidx,
1396 			     &pmap_pvo_table[ptegidx]);
1397 #endif
1398 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1399 			if (pteidx_p)
1400 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1401 			return pvo;
1402 		}
1403 	}
1404 	if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH))
1405 		panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1406 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1407 	return NULL;
1408 }
1409 
1410 #if defined(DEBUG) || defined(PMAPCHECK)
1411 void
1412 pmap_pvo_check(const struct pvo_entry *pvo)
1413 {
1414 	struct pvo_head *pvo_head;
1415 	struct pvo_entry *pvo0;
1416 	volatile struct pte *pt;
1417 	int failed = 0;
1418 
1419 	PMAP_LOCK();
1420 
1421 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1422 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1423 
1424 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1425 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1426 		    pvo, pvo->pvo_pmap);
1427 		failed = 1;
1428 	}
1429 
1430 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1431 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1432 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1433 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1434 		failed = 1;
1435 	}
1436 
1437 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1438 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1439 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1440 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1441 		failed = 1;
1442 	}
1443 
1444 	if (PVO_MANAGED_P(pvo)) {
1445 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1446 	} else {
1447 		if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) {
1448 			printf("pmap_pvo_check: pvo %p: non kernel address "
1449 			    "on kernel unmanaged list\n", pvo);
1450 			failed = 1;
1451 		}
1452 		pvo_head = &pmap_pvo_kunmanaged;
1453 	}
1454 	LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1455 		if (pvo0 == pvo)
1456 			break;
1457 	}
1458 	if (pvo0 == NULL) {
1459 		printf("pmap_pvo_check: pvo %p: not present "
1460 		    "on its vlist head %p\n", pvo, pvo_head);
1461 		failed = 1;
1462 	}
1463 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1464 		printf("pmap_pvo_check: pvo %p: not present "
1465 		    "on its olist head\n", pvo);
1466 		failed = 1;
1467 	}
1468 	pt = pmap_pvo_to_pte(pvo, -1);
1469 	if (pt == NULL) {
1470 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1471 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1472 			    "no PTE\n", pvo);
1473 			failed = 1;
1474 		}
1475 	} else {
1476 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1477 		    (uintptr_t) pt >=
1478 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1479 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1480 			    "pteg table\n", pvo, pt);
1481 			failed = 1;
1482 		}
1483 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1484 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1485 			    "no PTE\n", pvo);
1486 			failed = 1;
1487 		}
1488 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1489 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1490 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1491 			    pvo->pvo_pte.pte_hi,
1492 			    pt->pte_hi);
1493 			failed = 1;
1494 		}
1495 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1496 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1497 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1498 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1499 			    (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1500 			    (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1501 			failed = 1;
1502 		}
1503 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1504 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1505 			    " doesn't not match PVO's VA %#" _PRIxva "\n",
1506 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1507 			failed = 1;
1508 		}
1509 		if (failed)
1510 			pmap_pte_print(pt);
1511 	}
1512 	if (failed)
1513 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1514 		    pvo->pvo_pmap);
1515 
1516 	PMAP_UNLOCK();
1517 }
1518 #endif /* DEBUG || PMAPCHECK */
1519 
1520 /*
1521  * Search the PVO table looking for a non-wired entry.
1522  * If we find one, remove it and return it.
1523  */
1524 
1525 struct pvo_entry *
1526 pmap_pvo_reclaim(struct pmap *pm)
1527 {
1528 	struct pvo_tqhead *pvoh;
1529 	struct pvo_entry *pvo;
1530 	uint32_t idx, endidx;
1531 
1532 	endidx = pmap_pvo_reclaim_nextidx;
1533 	for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1534 	     idx = (idx + 1) & pmap_pteg_mask) {
1535 		pvoh = &pmap_pvo_table[idx];
1536 		TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1537 			if (!PVO_WIRED_P(pvo)) {
1538 				pmap_pvo_remove(pvo, -1, NULL);
1539 				pmap_pvo_reclaim_nextidx = idx;
1540 				PMAPCOUNT(pvos_reclaimed);
1541 				return pvo;
1542 			}
1543 		}
1544 	}
1545 	return NULL;
1546 }
1547 
1548 /*
1549  * This returns whether this is the first mapping of a page.
1550  */
1551 int
1552 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1553 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1554 {
1555 	struct pvo_entry *pvo;
1556 	struct pvo_tqhead *pvoh;
1557 	register_t msr;
1558 	int ptegidx;
1559 	int i;
1560 	int poolflags = PR_NOWAIT;
1561 
1562 	/*
1563 	 * Compute the PTE Group index.
1564 	 */
1565 	va &= ~ADDR_POFF;
1566 	ptegidx = va_to_pteg(pm, va);
1567 
1568 	msr = pmap_interrupts_off();
1569 
1570 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1571 	if (pmap_pvo_remove_depth > 0)
1572 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1573 	if (++pmap_pvo_enter_depth > 1)
1574 		panic("pmap_pvo_enter: called recursively!");
1575 #endif
1576 
1577 	/*
1578 	 * Remove any existing mapping for this page.  Reuse the
1579 	 * pvo entry if there a mapping.
1580 	 */
1581 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1582 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1583 #ifdef DEBUG
1584 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1585 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1586 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1587 			   va < VM_MIN_KERNEL_ADDRESS) {
1588 				printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1589 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1590 				printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1591 				    pvo->pvo_pte.pte_hi,
1592 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1593 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1594 #ifdef DDBX
1595 				Debugger();
1596 #endif
1597 			}
1598 #endif
1599 			PMAPCOUNT(mappings_replaced);
1600 			pmap_pvo_remove(pvo, -1, NULL);
1601 			break;
1602 		}
1603 	}
1604 
1605 	/*
1606 	 * If we aren't overwriting an mapping, try to allocate
1607 	 */
1608 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1609 	--pmap_pvo_enter_depth;
1610 #endif
1611 	pmap_interrupts_restore(msr);
1612 	if (pvo) {
1613 		pmap_pvo_free(pvo);
1614 	}
1615 	pvo = pool_get(pl, poolflags);
1616 
1617 #ifdef DEBUG
1618 	/*
1619 	 * Exercise pmap_pvo_reclaim() a little.
1620 	 */
1621 	if (pvo && (flags & PMAP_CANFAIL) != 0 &&
1622 	    pmap_pvo_reclaim_debugctr++ > 0x1000 &&
1623 	    (pmap_pvo_reclaim_debugctr & 0xff) == 0) {
1624 		pool_put(pl, pvo);
1625 		pvo = NULL;
1626 	}
1627 #endif
1628 
1629 	msr = pmap_interrupts_off();
1630 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1631 	++pmap_pvo_enter_depth;
1632 #endif
1633 	if (pvo == NULL) {
1634 		pvo = pmap_pvo_reclaim(pm);
1635 		if (pvo == NULL) {
1636 			if ((flags & PMAP_CANFAIL) == 0)
1637 				panic("pmap_pvo_enter: failed");
1638 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1639 			pmap_pvo_enter_depth--;
1640 #endif
1641 			PMAPCOUNT(pvos_failed);
1642 			pmap_interrupts_restore(msr);
1643 			return ENOMEM;
1644 		}
1645 	}
1646 
1647 	pvo->pvo_vaddr = va;
1648 	pvo->pvo_pmap = pm;
1649 	pvo->pvo_vaddr &= ~ADDR_POFF;
1650 	if (flags & VM_PROT_EXECUTE) {
1651 		PMAPCOUNT(exec_mappings);
1652 		pvo_set_exec(pvo);
1653 	}
1654 	if (flags & PMAP_WIRED)
1655 		pvo->pvo_vaddr |= PVO_WIRED;
1656 	if (pvo_head != &pmap_pvo_kunmanaged) {
1657 		pvo->pvo_vaddr |= PVO_MANAGED;
1658 		PMAPCOUNT(mappings);
1659 	} else {
1660 		PMAPCOUNT(kernel_mappings);
1661 	}
1662 	pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1663 
1664 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1665 	if (PVO_WIRED_P(pvo))
1666 		pvo->pvo_pmap->pm_stats.wired_count++;
1667 	pvo->pvo_pmap->pm_stats.resident_count++;
1668 #if defined(DEBUG)
1669 /*	if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */
1670 		DPRINTFN(PVOENTER,
1671 		    ("pmap_pvo_enter: pvo %p: pm %p va %#" _PRIxva " pa %#" _PRIxpa "\n",
1672 		    pvo, pm, va, pa));
1673 #endif
1674 
1675 	/*
1676 	 * We hope this succeeds but it isn't required.
1677 	 */
1678 	pvoh = &pmap_pvo_table[ptegidx];
1679 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1680 	if (i >= 0) {
1681 		PVO_PTEGIDX_SET(pvo, i);
1682 		PVO_WHERE(pvo, ENTER_INSERT);
1683 		PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1684 		    ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1685 		TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1686 
1687 	} else {
1688 		/*
1689 		 * Since we didn't have room for this entry (which makes it
1690 		 * and evicted entry), place it at the head of the list.
1691 		 */
1692 		TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1693 		PMAPCOUNT(ptes_evicted);
1694 		pm->pm_evictions++;
1695 		/*
1696 		 * If this is a kernel page, make sure it's active.
1697 		 */
1698 		if (pm == pmap_kernel()) {
1699 			i = pmap_pte_spill(pm, va, false);
1700 			KASSERT(i);
1701 		}
1702 	}
1703 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1704 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1705 	pmap_pvo_enter_depth--;
1706 #endif
1707 	pmap_interrupts_restore(msr);
1708 	return 0;
1709 }
1710 
1711 static void
1712 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol)
1713 {
1714 	volatile struct pte *pt;
1715 	int ptegidx;
1716 
1717 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1718 	if (++pmap_pvo_remove_depth > 1)
1719 		panic("pmap_pvo_remove: called recursively!");
1720 #endif
1721 
1722 	/*
1723 	 * If we haven't been supplied the ptegidx, calculate it.
1724 	 */
1725 	if (pteidx == -1) {
1726 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1727 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1728 	} else {
1729 		ptegidx = pteidx >> 3;
1730 		if (pvo->pvo_pte.pte_hi & PTE_HID)
1731 			ptegidx ^= pmap_pteg_mask;
1732 	}
1733 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1734 
1735 	/*
1736 	 * If there is an active pte entry, we need to deactivate it
1737 	 * (and save the ref & chg bits).
1738 	 */
1739 	pt = pmap_pvo_to_pte(pvo, pteidx);
1740 	if (pt != NULL) {
1741 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1742 		PVO_WHERE(pvo, REMOVE);
1743 		PVO_PTEGIDX_CLR(pvo);
1744 		PMAPCOUNT(ptes_removed);
1745 	} else {
1746 		KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1747 		pvo->pvo_pmap->pm_evictions--;
1748 	}
1749 
1750 	/*
1751 	 * Account for executable mappings.
1752 	 */
1753 	if (PVO_EXECUTABLE_P(pvo))
1754 		pvo_clear_exec(pvo);
1755 
1756 	/*
1757 	 * Update our statistics.
1758 	 */
1759 	pvo->pvo_pmap->pm_stats.resident_count--;
1760 	if (PVO_WIRED_P(pvo))
1761 		pvo->pvo_pmap->pm_stats.wired_count--;
1762 
1763 	/*
1764 	 * Save the REF/CHG bits into their cache if the page is managed.
1765 	 */
1766 	if (PVO_MANAGED_P(pvo)) {
1767 		register_t ptelo = pvo->pvo_pte.pte_lo;
1768 		struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1769 
1770 		if (pg != NULL) {
1771 			/*
1772 			 * If this page was changed and it is mapped exec,
1773 			 * invalidate it.
1774 			 */
1775 			if ((ptelo & PTE_CHG) &&
1776 			    (pmap_attr_fetch(pg) & PTE_EXEC)) {
1777 				struct pvo_head *pvoh = vm_page_to_pvoh(pg);
1778 				if (LIST_EMPTY(pvoh)) {
1779 					DPRINTFN(EXEC, ("[pmap_pvo_remove: "
1780 					    "%#" _PRIxpa ": clear-exec]\n",
1781 					    VM_PAGE_TO_PHYS(pg)));
1782 					pmap_attr_clear(pg, PTE_EXEC);
1783 					PMAPCOUNT(exec_uncached_pvo_remove);
1784 				} else {
1785 					DPRINTFN(EXEC, ("[pmap_pvo_remove: "
1786 					    "%#" _PRIxpa ": syncicache]\n",
1787 					    VM_PAGE_TO_PHYS(pg)));
1788 					pmap_syncicache(VM_PAGE_TO_PHYS(pg),
1789 					    PAGE_SIZE);
1790 					PMAPCOUNT(exec_synced_pvo_remove);
1791 				}
1792 			}
1793 
1794 			pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1795 		}
1796 		PMAPCOUNT(unmappings);
1797 	} else {
1798 		PMAPCOUNT(kernel_unmappings);
1799 	}
1800 
1801 	/*
1802 	 * Remove the PVO from its lists and return it to the pool.
1803 	 */
1804 	LIST_REMOVE(pvo, pvo_vlink);
1805 	TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1806 	if (pvol) {
1807 		LIST_INSERT_HEAD(pvol, pvo, pvo_vlink);
1808 	}
1809 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1810 	pmap_pvo_remove_depth--;
1811 #endif
1812 }
1813 
1814 void
1815 pmap_pvo_free(struct pvo_entry *pvo)
1816 {
1817 
1818 	pool_put(PVO_MANAGED_P(pvo) ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo);
1819 }
1820 
1821 void
1822 pmap_pvo_free_list(struct pvo_head *pvol)
1823 {
1824 	struct pvo_entry *pvo, *npvo;
1825 
1826 	for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) {
1827 		npvo = LIST_NEXT(pvo, pvo_vlink);
1828 		LIST_REMOVE(pvo, pvo_vlink);
1829 		pmap_pvo_free(pvo);
1830 	}
1831 }
1832 
1833 /*
1834  * Mark a mapping as executable.
1835  * If this is the first executable mapping in the segment,
1836  * clear the noexec flag.
1837  */
1838 static void
1839 pvo_set_exec(struct pvo_entry *pvo)
1840 {
1841 	struct pmap *pm = pvo->pvo_pmap;
1842 
1843 	if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) {
1844 		return;
1845 	}
1846 	pvo->pvo_vaddr |= PVO_EXECUTABLE;
1847 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1848 	{
1849 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1850 		if (pm->pm_exec[sr]++ == 0) {
1851 			pm->pm_sr[sr] &= ~SR_NOEXEC;
1852 		}
1853 	}
1854 #endif
1855 }
1856 
1857 /*
1858  * Mark a mapping as non-executable.
1859  * If this was the last executable mapping in the segment,
1860  * set the noexec flag.
1861  */
1862 static void
1863 pvo_clear_exec(struct pvo_entry *pvo)
1864 {
1865 	struct pmap *pm = pvo->pvo_pmap;
1866 
1867 	if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) {
1868 		return;
1869 	}
1870 	pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1871 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1872 	{
1873 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1874 		if (--pm->pm_exec[sr] == 0) {
1875 			pm->pm_sr[sr] |= SR_NOEXEC;
1876 		}
1877 	}
1878 #endif
1879 }
1880 
1881 /*
1882  * Insert physical page at pa into the given pmap at virtual address va.
1883  */
1884 int
1885 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1886 {
1887 	struct mem_region *mp;
1888 	struct pvo_head *pvo_head;
1889 	struct vm_page *pg;
1890 	struct pool *pl;
1891 	register_t pte_lo;
1892 	int error;
1893 	u_int pvo_flags;
1894 	u_int was_exec = 0;
1895 
1896 	PMAP_LOCK();
1897 
1898 	if (__predict_false(!pmap_initialized)) {
1899 		pvo_head = &pmap_pvo_kunmanaged;
1900 		pl = &pmap_upvo_pool;
1901 		pvo_flags = 0;
1902 		pg = NULL;
1903 		was_exec = PTE_EXEC;
1904 	} else {
1905 		pvo_head = pa_to_pvoh(pa, &pg);
1906 		pl = &pmap_mpvo_pool;
1907 		pvo_flags = PVO_MANAGED;
1908 	}
1909 
1910 	DPRINTFN(ENTER,
1911 	    ("pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1912 	    pm, va, pa, prot, flags));
1913 
1914 	/*
1915 	 * If this is a managed page, and it's the first reference to the
1916 	 * page clear the execness of the page.  Otherwise fetch the execness.
1917 	 */
1918 	if (pg != NULL)
1919 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1920 
1921 	DPRINTFN(ENTER, (" was_exec=%d", was_exec));
1922 
1923 	/*
1924 	 * Assume the page is cache inhibited and access is guarded unless
1925 	 * it's in our available memory array.  If it is in the memory array,
1926 	 * asssume it's in memory coherent memory.
1927 	 */
1928 	if (flags & PMAP_MD_PREFETCHABLE) {
1929 		pte_lo = 0;
1930 	} else
1931 		pte_lo = PTE_G;
1932 
1933 	if ((flags & PMAP_NOCACHE) == 0) {
1934 		for (mp = mem; mp->size; mp++) {
1935 			if (pa >= mp->start && pa < mp->start + mp->size) {
1936 				pte_lo = PTE_M;
1937 				break;
1938 			}
1939 		}
1940 	} else {
1941 		pte_lo |= PTE_I;
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 ((flags & PMAP_NOCACHE) == 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_lwp_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 (bank = 0; bank < vm_nphysseg; bank++) {
2916 		ps = VM_PHYSMEM_PTR(bank);
2917 		if (ps->free_list == VM_FREELIST_FIRST256 &&
2918 		    ps->avail_end - ps->avail_start >= npgs) {
2919 			pa = ptoa(ps->avail_start);
2920 			break;
2921 		}
2922 	}
2923 
2924 	if (pa == 0)
2925 		panic("pmap_steal_memory: no approriate memory to steal!");
2926 
2927 	ps->avail_start += npgs;
2928 	ps->start += npgs;
2929 
2930 	/*
2931 	 * If we've used up all the pages in the segment, remove it and
2932 	 * compact the list.
2933 	 */
2934 	if (ps->avail_start == ps->end) {
2935 		/*
2936 		 * If this was the last one, then a very bad thing has occurred
2937 		 */
2938 		if (--vm_nphysseg == 0)
2939 			panic("pmap_steal_memory: out of memory!");
2940 
2941 		printf("pmap_steal_memory: consumed bank %d\n", bank);
2942 		for (; bank < vm_nphysseg; bank++, ps++) {
2943 			ps[0] = ps[1];
2944 		}
2945 	}
2946 
2947 	va = (vaddr_t) pa;
2948 	memset((void *) va, 0, size);
2949 	pmap_pages_stolen += npgs;
2950 #ifdef DEBUG
2951 	if (pmapdebug && npgs > 1) {
2952 		u_int cnt = 0;
2953 		for (bank = 0; bank < vm_nphysseg; bank++) {
2954 			ps = VM_PHYSMEM_PTR(bank);
2955 			cnt += ps->avail_end - ps->avail_start;
2956 		}
2957 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2958 		    npgs, pmap_pages_stolen, cnt);
2959 	}
2960 #endif
2961 
2962 	return va;
2963 }
2964 
2965 /*
2966  * Find a chuck of memory with right size and alignment.
2967  */
2968 paddr_t
2969 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2970 {
2971 	struct mem_region *mp;
2972 	paddr_t s, e;
2973 	int i, j;
2974 
2975 	size = round_page(size);
2976 
2977 	DPRINTFN(BOOT,
2978 	    ("pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
2979 	    size, alignment, at_end));
2980 
2981 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2982 		panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
2983 		    alignment);
2984 
2985 	if (at_end) {
2986 		if (alignment != PAGE_SIZE)
2987 			panic("pmap_boot_find_memory: invalid ending "
2988 			    "alignment %#" _PRIxpa, alignment);
2989 
2990 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
2991 			s = mp->start + mp->size - size;
2992 			if (s >= mp->start && mp->size >= size) {
2993 				DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
2994 				DPRINTFN(BOOT,
2995 				    ("pmap_boot_find_memory: b-avail[%d] start "
2996 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
2997 				     mp->start, mp->size));
2998 				mp->size -= size;
2999 				DPRINTFN(BOOT,
3000 				    ("pmap_boot_find_memory: a-avail[%d] start "
3001 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3002 				     mp->start, mp->size));
3003 				return s;
3004 			}
3005 		}
3006 		panic("pmap_boot_find_memory: no available memory");
3007 	}
3008 
3009 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3010 		s = (mp->start + alignment - 1) & ~(alignment-1);
3011 		e = s + size;
3012 
3013 		/*
3014 		 * Is the calculated region entirely within the region?
3015 		 */
3016 		if (s < mp->start || e > mp->start + mp->size)
3017 			continue;
3018 
3019 		DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
3020 		if (s == mp->start) {
3021 			/*
3022 			 * If the block starts at the beginning of region,
3023 			 * adjust the size & start. (the region may now be
3024 			 * zero in length)
3025 			 */
3026 			DPRINTFN(BOOT,
3027 			    ("pmap_boot_find_memory: b-avail[%d] start "
3028 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3029 			mp->start += size;
3030 			mp->size -= size;
3031 			DPRINTFN(BOOT,
3032 			    ("pmap_boot_find_memory: a-avail[%d] start "
3033 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3034 		} else if (e == mp->start + mp->size) {
3035 			/*
3036 			 * If the block starts at the beginning of region,
3037 			 * adjust only the size.
3038 			 */
3039 			DPRINTFN(BOOT,
3040 			    ("pmap_boot_find_memory: b-avail[%d] start "
3041 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3042 			mp->size -= size;
3043 			DPRINTFN(BOOT,
3044 			    ("pmap_boot_find_memory: a-avail[%d] start "
3045 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3046 		} else {
3047 			/*
3048 			 * Block is in the middle of the region, so we
3049 			 * have to split it in two.
3050 			 */
3051 			for (j = avail_cnt; j > i + 1; j--) {
3052 				avail[j] = avail[j-1];
3053 			}
3054 			DPRINTFN(BOOT,
3055 			    ("pmap_boot_find_memory: b-avail[%d] start "
3056 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3057 			mp[1].start = e;
3058 			mp[1].size = mp[0].start + mp[0].size - e;
3059 			mp[0].size = s - mp[0].start;
3060 			avail_cnt++;
3061 			for (; i < avail_cnt; i++) {
3062 				DPRINTFN(BOOT,
3063 				    ("pmap_boot_find_memory: a-avail[%d] "
3064 				     "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3065 				     avail[i].start, avail[i].size));
3066 			}
3067 		}
3068 		KASSERT(s == (uintptr_t) s);
3069 		return s;
3070 	}
3071 	panic("pmap_boot_find_memory: not enough memory for "
3072 	    "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3073 }
3074 
3075 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3076 #if defined (PMAP_OEA64_BRIDGE)
3077 int
3078 pmap_setup_segment0_map(int use_large_pages, ...)
3079 {
3080     vaddr_t va;
3081 
3082     register_t pte_lo = 0x0;
3083     int ptegidx = 0, i = 0;
3084     struct pte pte;
3085     va_list ap;
3086 
3087     /* Coherent + Supervisor RW, no user access */
3088     pte_lo = PTE_M;
3089 
3090     /* XXXSL
3091      * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3092      * these have to take priority.
3093      */
3094     for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3095         ptegidx = va_to_pteg(pmap_kernel(), va);
3096         pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3097         i = pmap_pte_insert(ptegidx, &pte);
3098     }
3099 
3100     va_start(ap, use_large_pages);
3101     while (1) {
3102         paddr_t pa;
3103         size_t size;
3104 
3105         va = va_arg(ap, vaddr_t);
3106 
3107         if (va == 0)
3108             break;
3109 
3110         pa = va_arg(ap, paddr_t);
3111         size = va_arg(ap, size_t);
3112 
3113         for (; va < (va + size); va += 0x1000, pa += 0x1000) {
3114 #if 0
3115 	    printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__,  va, pa);
3116 #endif
3117             ptegidx = va_to_pteg(pmap_kernel(), va);
3118             pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3119             i = pmap_pte_insert(ptegidx, &pte);
3120         }
3121     }
3122 
3123     TLBSYNC();
3124     SYNC();
3125     return (0);
3126 }
3127 #endif /* PMAP_OEA64_BRIDGE */
3128 
3129 /*
3130  * This is not part of the defined PMAP interface and is specific to the
3131  * PowerPC architecture.  This is called during initppc, before the system
3132  * is really initialized.
3133  */
3134 void
3135 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3136 {
3137 	struct mem_region *mp, tmp;
3138 	paddr_t s, e;
3139 	psize_t size;
3140 	int i, j;
3141 
3142 	/*
3143 	 * Get memory.
3144 	 */
3145 	mem_regions(&mem, &avail);
3146 #if defined(DEBUG)
3147 	if (pmapdebug & PMAPDEBUG_BOOT) {
3148 		printf("pmap_bootstrap: memory configuration:\n");
3149 		for (mp = mem; mp->size; mp++) {
3150 			printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3151 				mp->start, mp->size);
3152 		}
3153 		for (mp = avail; mp->size; mp++) {
3154 			printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3155 				mp->start, mp->size);
3156 		}
3157 	}
3158 #endif
3159 
3160 	/*
3161 	 * Find out how much physical memory we have and in how many chunks.
3162 	 */
3163 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3164 		if (mp->start >= pmap_memlimit)
3165 			continue;
3166 		if (mp->start + mp->size > pmap_memlimit) {
3167 			size = pmap_memlimit - mp->start;
3168 			physmem += btoc(size);
3169 		} else {
3170 			physmem += btoc(mp->size);
3171 		}
3172 		mem_cnt++;
3173 	}
3174 
3175 	/*
3176 	 * Count the number of available entries.
3177 	 */
3178 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
3179 		avail_cnt++;
3180 
3181 	/*
3182 	 * Page align all regions.
3183 	 */
3184 	kernelstart = trunc_page(kernelstart);
3185 	kernelend = round_page(kernelend);
3186 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3187 		s = round_page(mp->start);
3188 		mp->size -= (s - mp->start);
3189 		mp->size = trunc_page(mp->size);
3190 		mp->start = s;
3191 		e = mp->start + mp->size;
3192 
3193 		DPRINTFN(BOOT,
3194 		    ("pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3195 		    i, mp->start, mp->size));
3196 
3197 		/*
3198 		 * Don't allow the end to run beyond our artificial limit
3199 		 */
3200 		if (e > pmap_memlimit)
3201 			e = pmap_memlimit;
3202 
3203 		/*
3204 		 * Is this region empty or strange?  skip it.
3205 		 */
3206 		if (e <= s) {
3207 			mp->start = 0;
3208 			mp->size = 0;
3209 			continue;
3210 		}
3211 
3212 		/*
3213 		 * Does this overlap the beginning of kernel?
3214 		 *   Does extend past the end of the kernel?
3215 		 */
3216 		else if (s < kernelstart && e > kernelstart) {
3217 			if (e > kernelend) {
3218 				avail[avail_cnt].start = kernelend;
3219 				avail[avail_cnt].size = e - kernelend;
3220 				avail_cnt++;
3221 			}
3222 			mp->size = kernelstart - s;
3223 		}
3224 		/*
3225 		 * Check whether this region overlaps the end of the kernel.
3226 		 */
3227 		else if (s < kernelend && e > kernelend) {
3228 			mp->start = kernelend;
3229 			mp->size = e - kernelend;
3230 		}
3231 		/*
3232 		 * Look whether this regions is completely inside the kernel.
3233 		 * Nuke it if it does.
3234 		 */
3235 		else if (s >= kernelstart && e <= kernelend) {
3236 			mp->start = 0;
3237 			mp->size = 0;
3238 		}
3239 		/*
3240 		 * If the user imposed a memory limit, enforce it.
3241 		 */
3242 		else if (s >= pmap_memlimit) {
3243 			mp->start = -PAGE_SIZE;	/* let's know why */
3244 			mp->size = 0;
3245 		}
3246 		else {
3247 			mp->start = s;
3248 			mp->size = e - s;
3249 		}
3250 		DPRINTFN(BOOT,
3251 		    ("pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3252 		    i, mp->start, mp->size));
3253 	}
3254 
3255 	/*
3256 	 * Move (and uncount) all the null return to the end.
3257 	 */
3258 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3259 		if (mp->size == 0) {
3260 			tmp = avail[i];
3261 			avail[i] = avail[--avail_cnt];
3262 			avail[avail_cnt] = avail[i];
3263 		}
3264 	}
3265 
3266 	/*
3267 	 * (Bubble)sort them into ascending order.
3268 	 */
3269 	for (i = 0; i < avail_cnt; i++) {
3270 		for (j = i + 1; j < avail_cnt; j++) {
3271 			if (avail[i].start > avail[j].start) {
3272 				tmp = avail[i];
3273 				avail[i] = avail[j];
3274 				avail[j] = tmp;
3275 			}
3276 		}
3277 	}
3278 
3279 	/*
3280 	 * Make sure they don't overlap.
3281 	 */
3282 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3283 		if (mp[0].start + mp[0].size > mp[1].start) {
3284 			mp[0].size = mp[1].start - mp[0].start;
3285 		}
3286 		DPRINTFN(BOOT,
3287 		    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3288 		    i, mp->start, mp->size));
3289 	}
3290 	DPRINTFN(BOOT,
3291 	    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3292 	    i, mp->start, mp->size));
3293 
3294 #ifdef	PTEGCOUNT
3295 	pmap_pteg_cnt = PTEGCOUNT;
3296 #else /* PTEGCOUNT */
3297 
3298 	pmap_pteg_cnt = 0x1000;
3299 
3300 	while (pmap_pteg_cnt < physmem)
3301 		pmap_pteg_cnt <<= 1;
3302 
3303 	pmap_pteg_cnt >>= 1;
3304 #endif /* PTEGCOUNT */
3305 
3306 #ifdef DEBUG
3307 	DPRINTFN(BOOT,
3308 		("pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt));
3309 #endif
3310 
3311 	/*
3312 	 * Find suitably aligned memory for PTEG hash table.
3313 	 */
3314 	size = pmap_pteg_cnt * sizeof(struct pteg);
3315 	pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3316 
3317 #ifdef DEBUG
3318 	DPRINTFN(BOOT,
3319 		("PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table));
3320 #endif
3321 
3322 
3323 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3324 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
3325 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > 256MB",
3326 		    pmap_pteg_table, size);
3327 #endif
3328 
3329 	memset(__UNVOLATILE(pmap_pteg_table), 0,
3330 		pmap_pteg_cnt * sizeof(struct pteg));
3331 	pmap_pteg_mask = pmap_pteg_cnt - 1;
3332 
3333 	/*
3334 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3335 	 * with pages.  So we just steal them before giving them to UVM.
3336 	 */
3337 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3338 	pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3339 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3340 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
3341 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > 256MB",
3342 		    pmap_pvo_table, size);
3343 #endif
3344 
3345 	for (i = 0; i < pmap_pteg_cnt; i++)
3346 		TAILQ_INIT(&pmap_pvo_table[i]);
3347 
3348 #ifndef MSGBUFADDR
3349 	/*
3350 	 * Allocate msgbuf in high memory.
3351 	 */
3352 	msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3353 #endif
3354 
3355 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3356 		paddr_t pfstart = atop(mp->start);
3357 		paddr_t pfend = atop(mp->start + mp->size);
3358 		if (mp->size == 0)
3359 			continue;
3360 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
3361 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3362 				VM_FREELIST_FIRST256);
3363 		} else if (mp->start >= SEGMENT_LENGTH) {
3364 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3365 				VM_FREELIST_DEFAULT);
3366 		} else {
3367 			pfend = atop(SEGMENT_LENGTH);
3368 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3369 				VM_FREELIST_FIRST256);
3370 			pfstart = atop(SEGMENT_LENGTH);
3371 			pfend = atop(mp->start + mp->size);
3372 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3373 				VM_FREELIST_DEFAULT);
3374 		}
3375 	}
3376 
3377 	/*
3378 	 * Make sure kernel vsid is allocated as well as VSID 0.
3379 	 */
3380 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3381 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3382 	pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3383 		|= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3384 	pmap_vsid_bitmap[0] |= 1;
3385 
3386 	/*
3387 	 * Initialize kernel pmap and hardware.
3388 	 */
3389 
3390 /* PMAP_OEA64_BRIDGE does support these instructions */
3391 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
3392 	for (i = 0; i < 16; i++) {
3393  		pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3394 		__asm volatile ("mtsrin %0,%1"
3395  			      :: "r"(KERNELN_SEGMENT(i)|SR_PRKEY), "r"(i << ADDR_SR_SHFT));
3396 	}
3397 
3398 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3399 	__asm volatile ("mtsr %0,%1"
3400 		      :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
3401 #ifdef KERNEL2_SR
3402 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3403 	__asm volatile ("mtsr %0,%1"
3404 		      :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
3405 #endif
3406 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3407 #if defined (PMAP_OEA)
3408 	for (i = 0; i < 16; i++) {
3409 		if (iosrtable[i] & SR601_T) {
3410 			pmap_kernel()->pm_sr[i] = iosrtable[i];
3411 			__asm volatile ("mtsrin %0,%1"
3412 			    :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT));
3413 		}
3414 	}
3415 	__asm volatile ("sync; mtsdr1 %0; isync"
3416 		      :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10)));
3417 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
3418  	__asm __volatile ("sync; mtsdr1 %0; isync"
3419  		      :: "r"((uintptr_t)pmap_pteg_table | (32 - __builtin_clz(pmap_pteg_mask >> 11))));
3420 #endif
3421 	tlbia();
3422 
3423 #ifdef ALTIVEC
3424 	pmap_use_altivec = cpu_altivec;
3425 #endif
3426 
3427 #ifdef DEBUG
3428 	if (pmapdebug & PMAPDEBUG_BOOT) {
3429 		u_int cnt;
3430 		int bank;
3431 		char pbuf[9];
3432 		for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) {
3433 			cnt += VM_PHYSMEM_PTR(bank)->avail_end - VM_PHYSMEM_PTR(bank)->avail_start;
3434 			printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3435 			    bank,
3436 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_start),
3437 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_end),
3438 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_end - VM_PHYSMEM_PTR(bank)->avail_start));
3439 		}
3440 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3441 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3442 		    pbuf, cnt);
3443 	}
3444 #endif
3445 
3446 	pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry),
3447 	    sizeof(struct pvo_entry), 0, 0, "pmap_upvopl",
3448 	    &pmap_pool_uallocator, IPL_VM);
3449 
3450 	pool_setlowat(&pmap_upvo_pool, 252);
3451 
3452 	pool_init(&pmap_pool, sizeof(struct pmap),
3453 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator,
3454 	    IPL_NONE);
3455 
3456 #if defined(PMAP_NEED_MAPKERNEL) || 1
3457 	{
3458 		struct pmap *pm = pmap_kernel();
3459 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3460 		extern int etext[], kernel_text[];
3461 		vaddr_t va, va_etext = (paddr_t) etext;
3462 #endif
3463 		paddr_t pa, pa_end;
3464 		register_t sr;
3465 		struct pte pt;
3466 		unsigned int ptegidx;
3467 		int bank;
3468 
3469 		sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3470 		pm->pm_sr[0] = sr;
3471 
3472 		for (bank = 0; bank < vm_nphysseg; bank++) {
3473 			pa_end = ptoa(VM_PHYSMEM_PTR(bank)->avail_end);
3474 			pa = ptoa(VM_PHYSMEM_PTR(bank)->avail_start);
3475 			for (; pa < pa_end; pa += PAGE_SIZE) {
3476 				ptegidx = va_to_pteg(pm, pa);
3477 				pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3478 				pmap_pte_insert(ptegidx, &pt);
3479 			}
3480 		}
3481 
3482 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3483 		va = (vaddr_t) kernel_text;
3484 
3485 		for (pa = kernelstart; va < va_etext;
3486 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3487 			ptegidx = va_to_pteg(pm, va);
3488 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3489 			pmap_pte_insert(ptegidx, &pt);
3490 		}
3491 
3492 		for (; pa < kernelend;
3493 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3494 			ptegidx = va_to_pteg(pm, va);
3495 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3496 			pmap_pte_insert(ptegidx, &pt);
3497 		}
3498 
3499 		for (va = 0, pa = 0; va < kernelstart;
3500 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3501 			ptegidx = va_to_pteg(pm, va);
3502 			if (va < 0x3000)
3503 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3504 			else
3505 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3506 			pmap_pte_insert(ptegidx, &pt);
3507 		}
3508 		for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3509 		    pa += PAGE_SIZE, va += PAGE_SIZE) {
3510 			ptegidx = va_to_pteg(pm, va);
3511 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3512 			pmap_pte_insert(ptegidx, &pt);
3513 		}
3514 #endif
3515 
3516 		__asm volatile ("mtsrin %0,%1"
3517  			      :: "r"(sr), "r"(kernelstart));
3518 	}
3519 #endif
3520 }
3521