xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: pmap.c,v 1.95 2018/01/27 23:07:36 chs 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 1.95 2018/01/27 23:07:36 chs Exp $");
67 
68 #define	PMAP_NOOPNAMES
69 
70 #include "opt_ppcarch.h"
71 #include "opt_altivec.h"
72 #include "opt_multiprocessor.h"
73 #include "opt_pmap.h"
74 
75 #include <sys/param.h>
76 #include <sys/proc.h>
77 #include <sys/pool.h>
78 #include <sys/queue.h>
79 #include <sys/device.h>		/* for evcnt */
80 #include <sys/systm.h>
81 #include <sys/atomic.h>
82 
83 #include <uvm/uvm.h>
84 #include <uvm/uvm_physseg.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 extern int pmap_use_altivec;
96 #endif
97 
98 #ifdef PMAP_MEMLIMIT
99 static paddr_t pmap_memlimit = PMAP_MEMLIMIT;
100 #else
101 static paddr_t pmap_memlimit = -PAGE_SIZE;		/* there is no limit */
102 #endif
103 
104 extern struct pmap kernel_pmap_;
105 static unsigned int pmap_pages_stolen;
106 static u_long pmap_pte_valid;
107 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
108 static u_long pmap_pvo_enter_depth;
109 static u_long pmap_pvo_remove_depth;
110 #endif
111 
112 #ifndef MSGBUFADDR
113 extern paddr_t msgbuf_paddr;
114 #endif
115 
116 static struct mem_region *mem, *avail;
117 static u_int mem_cnt, avail_cnt;
118 
119 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
120 # define	PMAP_OEA 1
121 #endif
122 
123 #if defined(PMAP_OEA)
124 #define	_PRIxpte	"lx"
125 #else
126 #define	_PRIxpte	PRIx64
127 #endif
128 #define	_PRIxpa		"lx"
129 #define	_PRIxva		"lx"
130 #define	_PRIsr  	"lx"
131 
132 #ifdef PMAP_NEEDS_FIXUP
133 #if defined(PMAP_OEA)
134 #define	PMAPNAME(name)	pmap32_##name
135 #elif defined(PMAP_OEA64)
136 #define	PMAPNAME(name)	pmap64_##name
137 #elif defined(PMAP_OEA64_BRIDGE)
138 #define	PMAPNAME(name)	pmap64bridge_##name
139 #else
140 #error unknown variant for pmap
141 #endif
142 #endif /* PMAP_NEEDS_FIXUP */
143 
144 #ifdef PMAPNAME
145 #define	STATIC			static
146 #define pmap_pte_spill		PMAPNAME(pte_spill)
147 #define pmap_real_memory	PMAPNAME(real_memory)
148 #define pmap_init		PMAPNAME(init)
149 #define pmap_virtual_space	PMAPNAME(virtual_space)
150 #define pmap_create		PMAPNAME(create)
151 #define pmap_reference		PMAPNAME(reference)
152 #define pmap_destroy		PMAPNAME(destroy)
153 #define pmap_copy		PMAPNAME(copy)
154 #define pmap_update		PMAPNAME(update)
155 #define pmap_enter		PMAPNAME(enter)
156 #define pmap_remove		PMAPNAME(remove)
157 #define pmap_kenter_pa		PMAPNAME(kenter_pa)
158 #define pmap_kremove		PMAPNAME(kremove)
159 #define pmap_extract		PMAPNAME(extract)
160 #define pmap_protect		PMAPNAME(protect)
161 #define pmap_unwire		PMAPNAME(unwire)
162 #define pmap_page_protect	PMAPNAME(page_protect)
163 #define pmap_query_bit		PMAPNAME(query_bit)
164 #define pmap_clear_bit		PMAPNAME(clear_bit)
165 
166 #define pmap_activate		PMAPNAME(activate)
167 #define pmap_deactivate		PMAPNAME(deactivate)
168 
169 #define pmap_pinit		PMAPNAME(pinit)
170 #define pmap_procwr		PMAPNAME(procwr)
171 
172 #define pmap_pool		PMAPNAME(pool)
173 #define pmap_upvo_pool		PMAPNAME(upvo_pool)
174 #define pmap_mpvo_pool		PMAPNAME(mpvo_pool)
175 #define pmap_pvo_table		PMAPNAME(pvo_table)
176 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
177 #define pmap_pte_print		PMAPNAME(pte_print)
178 #define pmap_pteg_check		PMAPNAME(pteg_check)
179 #define pmap_print_mmruregs	PMAPNAME(print_mmuregs)
180 #define pmap_print_pte		PMAPNAME(print_pte)
181 #define pmap_pteg_dist		PMAPNAME(pteg_dist)
182 #endif
183 #if defined(DEBUG) || defined(PMAPCHECK)
184 #define	pmap_pvo_verify		PMAPNAME(pvo_verify)
185 #define pmapcheck		PMAPNAME(check)
186 #endif
187 #if defined(DEBUG) || defined(PMAPDEBUG)
188 #define pmapdebug		PMAPNAME(debug)
189 #endif
190 #define pmap_steal_memory	PMAPNAME(steal_memory)
191 #define pmap_bootstrap		PMAPNAME(bootstrap)
192 #else
193 #define	STATIC			/* nothing */
194 #endif /* PMAPNAME */
195 
196 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
197 STATIC void pmap_real_memory(paddr_t *, psize_t *);
198 STATIC void pmap_init(void);
199 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
200 STATIC pmap_t pmap_create(void);
201 STATIC void pmap_reference(pmap_t);
202 STATIC void pmap_destroy(pmap_t);
203 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
204 STATIC void pmap_update(pmap_t);
205 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
206 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
207 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t, u_int);
208 STATIC void pmap_kremove(vaddr_t, vsize_t);
209 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
210 
211 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
212 STATIC void pmap_unwire(pmap_t, vaddr_t);
213 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
214 STATIC bool pmap_query_bit(struct vm_page *, int);
215 STATIC bool pmap_clear_bit(struct vm_page *, int);
216 
217 STATIC void pmap_activate(struct lwp *);
218 STATIC void pmap_deactivate(struct lwp *);
219 
220 STATIC void pmap_pinit(pmap_t pm);
221 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
222 
223 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
224 STATIC void pmap_pte_print(volatile struct pte *);
225 STATIC void pmap_pteg_check(void);
226 STATIC void pmap_print_mmuregs(void);
227 STATIC void pmap_print_pte(pmap_t, vaddr_t);
228 STATIC void pmap_pteg_dist(void);
229 #endif
230 #if defined(DEBUG) || defined(PMAPCHECK)
231 STATIC void pmap_pvo_verify(void);
232 #endif
233 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
234 STATIC void pmap_bootstrap(paddr_t, paddr_t);
235 
236 #ifdef PMAPNAME
237 const struct pmap_ops PMAPNAME(ops) = {
238 	.pmapop_pte_spill = pmap_pte_spill,
239 	.pmapop_real_memory = pmap_real_memory,
240 	.pmapop_init = pmap_init,
241 	.pmapop_virtual_space = pmap_virtual_space,
242 	.pmapop_create = pmap_create,
243 	.pmapop_reference = pmap_reference,
244 	.pmapop_destroy = pmap_destroy,
245 	.pmapop_copy = pmap_copy,
246 	.pmapop_update = pmap_update,
247 	.pmapop_enter = pmap_enter,
248 	.pmapop_remove = pmap_remove,
249 	.pmapop_kenter_pa = pmap_kenter_pa,
250 	.pmapop_kremove = pmap_kremove,
251 	.pmapop_extract = pmap_extract,
252 	.pmapop_protect = pmap_protect,
253 	.pmapop_unwire = pmap_unwire,
254 	.pmapop_page_protect = pmap_page_protect,
255 	.pmapop_query_bit = pmap_query_bit,
256 	.pmapop_clear_bit = pmap_clear_bit,
257 	.pmapop_activate = pmap_activate,
258 	.pmapop_deactivate = pmap_deactivate,
259 	.pmapop_pinit = pmap_pinit,
260 	.pmapop_procwr = pmap_procwr,
261 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
262 	.pmapop_pte_print = pmap_pte_print,
263 	.pmapop_pteg_check = pmap_pteg_check,
264 	.pmapop_print_mmuregs = pmap_print_mmuregs,
265 	.pmapop_print_pte = pmap_print_pte,
266 	.pmapop_pteg_dist = pmap_pteg_dist,
267 #else
268 	.pmapop_pte_print = NULL,
269 	.pmapop_pteg_check = NULL,
270 	.pmapop_print_mmuregs = NULL,
271 	.pmapop_print_pte = NULL,
272 	.pmapop_pteg_dist = NULL,
273 #endif
274 #if defined(DEBUG) || defined(PMAPCHECK)
275 	.pmapop_pvo_verify = pmap_pvo_verify,
276 #else
277 	.pmapop_pvo_verify = NULL,
278 #endif
279 	.pmapop_steal_memory = pmap_steal_memory,
280 	.pmapop_bootstrap = pmap_bootstrap,
281 };
282 #endif /* !PMAPNAME */
283 
284 /*
285  * The following structure is aligned to 32 bytes
286  */
287 struct pvo_entry {
288 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
289 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
290 	struct pte pvo_pte;			/* Prebuilt PTE */
291 	pmap_t pvo_pmap;			/* ptr to owning pmap */
292 	vaddr_t pvo_vaddr;			/* VA of entry */
293 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
294 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
295 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
296 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
297 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
298 #define	PVO_WIRED_P(pvo)	((pvo)->pvo_vaddr & PVO_WIRED)
299 #define	PVO_MANAGED_P(pvo)	((pvo)->pvo_vaddr & PVO_MANAGED)
300 #define	PVO_EXECUTABLE_P(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
301 #define	PVO_ENTER_INSERT	0		/* PVO has been removed */
302 #define	PVO_SPILL_UNSET		1		/* PVO has been evicted */
303 #define	PVO_SPILL_SET		2		/* PVO has been spilled */
304 #define	PVO_SPILL_INSERT	3		/* PVO has been inserted */
305 #define	PVO_PMAP_PAGE_PROTECT	4		/* PVO has changed */
306 #define	PVO_PMAP_PROTECT	5		/* PVO has changed */
307 #define	PVO_REMOVE		6		/* PVO has been removed */
308 #define	PVO_WHERE_MASK		15
309 #define	PVO_WHERE_SHFT		8
310 } __attribute__ ((aligned (32)));
311 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
312 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
313 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
314 #define	PVO_PTEGIDX_CLR(pvo)	\
315 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
316 #define	PVO_PTEGIDX_SET(pvo,i)	\
317 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
318 #define	PVO_WHERE(pvo,w)	\
319 	((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
320 	 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
321 
322 TAILQ_HEAD(pvo_tqhead, pvo_entry);
323 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
324 static struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
325 static struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
326 
327 struct pool pmap_pool;		/* pool for pmap structures */
328 struct pool pmap_upvo_pool;	/* pool for pvo entries for unmanaged pages */
329 struct pool pmap_mpvo_pool;	/* pool for pvo entries for managed pages */
330 
331 /*
332  * We keep a cache of unmanaged pages to be used for pvo entries for
333  * unmanaged pages.
334  */
335 struct pvo_page {
336 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
337 };
338 SIMPLEQ_HEAD(pvop_head, pvo_page);
339 static struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head);
340 static struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head);
341 static u_long pmap_upvop_free;
342 static u_long pmap_upvop_maxfree;
343 static u_long pmap_mpvop_free;
344 static u_long pmap_mpvop_maxfree;
345 
346 static void *pmap_pool_ualloc(struct pool *, int);
347 static void *pmap_pool_malloc(struct pool *, int);
348 
349 static void pmap_pool_ufree(struct pool *, void *);
350 static void pmap_pool_mfree(struct pool *, void *);
351 
352 static struct pool_allocator pmap_pool_mallocator = {
353 	.pa_alloc = pmap_pool_malloc,
354 	.pa_free = pmap_pool_mfree,
355 	.pa_pagesz = 0,
356 };
357 
358 static struct pool_allocator pmap_pool_uallocator = {
359 	.pa_alloc = pmap_pool_ualloc,
360 	.pa_free = pmap_pool_ufree,
361 	.pa_pagesz = 0,
362 };
363 
364 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
365 void pmap_pte_print(volatile struct pte *);
366 void pmap_pteg_check(void);
367 void pmap_pteg_dist(void);
368 void pmap_print_pte(pmap_t, vaddr_t);
369 void pmap_print_mmuregs(void);
370 #endif
371 
372 #if defined(DEBUG) || defined(PMAPCHECK)
373 #ifdef PMAPCHECK
374 int pmapcheck = 1;
375 #else
376 int pmapcheck = 0;
377 #endif
378 void pmap_pvo_verify(void);
379 static void pmap_pvo_check(const struct pvo_entry *);
380 #define	PMAP_PVO_CHECK(pvo)	 		\
381 	do {					\
382 		if (pmapcheck)			\
383 			pmap_pvo_check(pvo);	\
384 	} while (0)
385 #else
386 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
387 #endif
388 static int pmap_pte_insert(int, struct pte *);
389 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
390 	vaddr_t, paddr_t, register_t, int);
391 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
392 static void pmap_pvo_free(struct pvo_entry *);
393 static void pmap_pvo_free_list(struct pvo_head *);
394 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
395 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
396 static struct pvo_entry *pmap_pvo_reclaim(struct pmap *);
397 static void pvo_set_exec(struct pvo_entry *);
398 static void pvo_clear_exec(struct pvo_entry *);
399 
400 static void tlbia(void);
401 
402 static void pmap_release(pmap_t);
403 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
404 
405 static uint32_t pmap_pvo_reclaim_nextidx;
406 #ifdef DEBUG
407 static int pmap_pvo_reclaim_debugctr;
408 #endif
409 
410 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
411 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
412 
413 static int pmap_initialized;
414 
415 #if defined(DEBUG) || defined(PMAPDEBUG)
416 #define	PMAPDEBUG_BOOT		0x0001
417 #define	PMAPDEBUG_PTE		0x0002
418 #define	PMAPDEBUG_EXEC		0x0008
419 #define	PMAPDEBUG_PVOENTER	0x0010
420 #define	PMAPDEBUG_PVOREMOVE	0x0020
421 #define	PMAPDEBUG_ACTIVATE	0x0100
422 #define	PMAPDEBUG_CREATE	0x0200
423 #define	PMAPDEBUG_ENTER		0x1000
424 #define	PMAPDEBUG_KENTER	0x2000
425 #define	PMAPDEBUG_KREMOVE	0x4000
426 #define	PMAPDEBUG_REMOVE	0x8000
427 
428 unsigned int pmapdebug = 0;
429 
430 # define DPRINTF(x, ...)	printf(x, __VA_ARGS__)
431 # define DPRINTFN(n, x, ...)	do if (pmapdebug & PMAPDEBUG_ ## n) printf(x, __VA_ARGS__); while (0)
432 #else
433 # define DPRINTF(x, ...)	do { } while (0)
434 # define DPRINTFN(n, x, ...)	do { } while (0)
435 #endif
436 
437 
438 #ifdef PMAPCOUNTERS
439 /*
440  * From pmap_subr.c
441  */
442 extern struct evcnt pmap_evcnt_mappings;
443 extern struct evcnt pmap_evcnt_unmappings;
444 
445 extern struct evcnt pmap_evcnt_kernel_mappings;
446 extern struct evcnt pmap_evcnt_kernel_unmappings;
447 
448 extern struct evcnt pmap_evcnt_mappings_replaced;
449 
450 extern struct evcnt pmap_evcnt_exec_mappings;
451 extern struct evcnt pmap_evcnt_exec_cached;
452 
453 extern struct evcnt pmap_evcnt_exec_synced;
454 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
455 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
456 
457 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
458 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
459 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
460 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
461 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
462 
463 extern struct evcnt pmap_evcnt_updates;
464 extern struct evcnt pmap_evcnt_collects;
465 extern struct evcnt pmap_evcnt_copies;
466 
467 extern struct evcnt pmap_evcnt_ptes_spilled;
468 extern struct evcnt pmap_evcnt_ptes_unspilled;
469 extern struct evcnt pmap_evcnt_ptes_evicted;
470 
471 extern struct evcnt pmap_evcnt_ptes_primary[8];
472 extern struct evcnt pmap_evcnt_ptes_secondary[8];
473 extern struct evcnt pmap_evcnt_ptes_removed;
474 extern struct evcnt pmap_evcnt_ptes_changed;
475 extern struct evcnt pmap_evcnt_pvos_reclaimed;
476 extern struct evcnt pmap_evcnt_pvos_failed;
477 
478 extern struct evcnt pmap_evcnt_zeroed_pages;
479 extern struct evcnt pmap_evcnt_copied_pages;
480 extern struct evcnt pmap_evcnt_idlezeroed_pages;
481 
482 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
483 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
484 #else
485 #define	PMAPCOUNT(ev)	((void) 0)
486 #define	PMAPCOUNT2(ev)	((void) 0)
487 #endif
488 
489 #define	TLBIE(va)	__asm volatile("tlbie %0" :: "r"(va))
490 
491 /* XXXSL: this needs to be moved to assembler */
492 #define	TLBIEL(va)	__asm __volatile("tlbie %0" :: "r"(va))
493 
494 #ifdef MD_TLBSYNC
495 #define TLBSYNC()	MD_TLBSYNC()
496 #else
497 #define	TLBSYNC()	__asm volatile("tlbsync")
498 #endif
499 #define	SYNC()		__asm volatile("sync")
500 #define	EIEIO()		__asm volatile("eieio")
501 #define	DCBST(va)	__asm __volatile("dcbst 0,%0" :: "r"(va))
502 #define	MFMSR()		mfmsr()
503 #define	MTMSR(psl)	mtmsr(psl)
504 #define	MFPVR()		mfpvr()
505 #define	MFSRIN(va)	mfsrin(va)
506 #define	MFTB()		mfrtcltbl()
507 
508 #if defined(DDB) && !defined(PMAP_OEA64)
509 static inline register_t
510 mfsrin(vaddr_t va)
511 {
512 	register_t sr;
513 	__asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
514 	return sr;
515 }
516 #endif	/* DDB && !PMAP_OEA64 */
517 
518 #if defined (PMAP_OEA64_BRIDGE)
519 extern void mfmsr64 (register64_t *result);
520 #endif /* PMAP_OEA64_BRIDGE */
521 
522 #define	PMAP_LOCK()		KERNEL_LOCK(1, NULL)
523 #define	PMAP_UNLOCK()		KERNEL_UNLOCK_ONE(NULL)
524 
525 static inline register_t
526 pmap_interrupts_off(void)
527 {
528 	register_t msr = MFMSR();
529 	if (msr & PSL_EE)
530 		MTMSR(msr & ~PSL_EE);
531 	return msr;
532 }
533 
534 static void
535 pmap_interrupts_restore(register_t msr)
536 {
537 	if (msr & PSL_EE)
538 		MTMSR(msr);
539 }
540 
541 static inline u_int32_t
542 mfrtcltbl(void)
543 {
544 #ifdef PPC_OEA601
545 	if ((MFPVR() >> 16) == MPC601)
546 		return (mfrtcl() >> 7);
547 	else
548 #endif
549 		return (mftbl());
550 }
551 
552 /*
553  * These small routines may have to be replaced,
554  * if/when we support processors other that the 604.
555  */
556 
557 void
558 tlbia(void)
559 {
560 	char *i;
561 
562 	SYNC();
563 #if defined(PMAP_OEA)
564 	/*
565 	 * Why not use "tlbia"?  Because not all processors implement it.
566 	 *
567 	 * This needs to be a per-CPU callback to do the appropriate thing
568 	 * for the CPU. XXX
569 	 */
570 	for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
571 		TLBIE(i);
572 		EIEIO();
573 		SYNC();
574 	}
575 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
576 	/* This is specifically for the 970, 970UM v1.6 pp. 140. */
577 	for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
578 		TLBIEL(i);
579 		EIEIO();
580 		SYNC();
581 	}
582 #endif
583 	TLBSYNC();
584 	SYNC();
585 }
586 
587 static inline register_t
588 va_to_vsid(const struct pmap *pm, vaddr_t addr)
589 {
590 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
591 	return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT;
592 #else /* PMAP_OEA64 */
593 #if 0
594 	const struct ste *ste;
595 	register_t hash;
596 	int i;
597 
598 	hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH;
599 
600 	/*
601 	 * Try the primary group first
602 	 */
603 	ste = pm->pm_stes[hash].stes;
604 	for (i = 0; i < 8; i++, ste++) {
605 		if (ste->ste_hi & STE_V) &&
606 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
607 			return ste;
608 	}
609 
610 	/*
611 	 * Then the secondary group.
612 	 */
613 	ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes;
614 	for (i = 0; i < 8; i++, ste++) {
615 		if (ste->ste_hi & STE_V) &&
616 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
617 			return addr;
618 	}
619 
620 	return NULL;
621 #else
622 	/*
623 	 * Rather than searching the STE groups for the VSID, we know
624 	 * how we generate that from the ESID and so do that.
625 	 */
626 	return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
627 #endif
628 #endif /* PMAP_OEA */
629 }
630 
631 static inline register_t
632 va_to_pteg(const struct pmap *pm, vaddr_t addr)
633 {
634 	register_t hash;
635 
636 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
637 	return hash & pmap_pteg_mask;
638 }
639 
640 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
641 /*
642  * Given a PTE in the page table, calculate the VADDR that hashes to it.
643  * The only bit of magic is that the top 4 bits of the address doesn't
644  * technically exist in the PTE.  But we know we reserved 4 bits of the
645  * VSID for it so that's how we get it.
646  */
647 static vaddr_t
648 pmap_pte_to_va(volatile const struct pte *pt)
649 {
650 	vaddr_t va;
651 	uintptr_t ptaddr = (uintptr_t) pt;
652 
653 	if (pt->pte_hi & PTE_HID)
654 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
655 
656 	/* PPC Bits 10-19  PPC64 Bits 42-51 */
657 #if defined(PMAP_OEA)
658 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
659 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
660 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
661 #endif
662 	va <<= ADDR_PIDX_SHFT;
663 
664 	/* PPC Bits 4-9  PPC64 Bits 36-41 */
665 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
666 
667 #if defined(PMAP_OEA64)
668 	/* PPC63 Bits 0-35 */
669 	/* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
670 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
671 	/* PPC Bits 0-3 */
672 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
673 #endif
674 
675 	return va;
676 }
677 #endif
678 
679 static inline struct pvo_head *
680 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
681 {
682 	struct vm_page *pg;
683 	struct vm_page_md *md;
684 
685 	pg = PHYS_TO_VM_PAGE(pa);
686 	if (pg_p != NULL)
687 		*pg_p = pg;
688 	if (pg == NULL)
689 		return &pmap_pvo_unmanaged;
690 	md = VM_PAGE_TO_MD(pg);
691 	return &md->mdpg_pvoh;
692 }
693 
694 static inline struct pvo_head *
695 vm_page_to_pvoh(struct vm_page *pg)
696 {
697 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
698 
699 	return &md->mdpg_pvoh;
700 }
701 
702 
703 static inline void
704 pmap_attr_clear(struct vm_page *pg, int ptebit)
705 {
706 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
707 
708 	md->mdpg_attrs &= ~ptebit;
709 }
710 
711 static inline int
712 pmap_attr_fetch(struct vm_page *pg)
713 {
714 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
715 
716 	return md->mdpg_attrs;
717 }
718 
719 static inline void
720 pmap_attr_save(struct vm_page *pg, int ptebit)
721 {
722 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
723 
724 	md->mdpg_attrs |= ptebit;
725 }
726 
727 static inline int
728 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
729 {
730 	if (pt->pte_hi == pvo_pt->pte_hi
731 #if 0
732 	    && ((pt->pte_lo ^ pvo_pt->pte_lo) &
733 	        ~(PTE_REF|PTE_CHG)) == 0
734 #endif
735 	    )
736 		return 1;
737 	return 0;
738 }
739 
740 static inline void
741 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
742 {
743 	/*
744 	 * Construct the PTE.  Default to IMB initially.  Valid bit
745 	 * only gets set when the real pte is set in memory.
746 	 *
747 	 * Note: Don't set the valid bit for correct operation of tlb update.
748 	 */
749 #if defined(PMAP_OEA)
750 	pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
751 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
752 	pt->pte_lo = pte_lo;
753 #elif defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA64)
754 	pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT)
755 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
756 	pt->pte_lo = (u_int64_t) pte_lo;
757 #endif /* PMAP_OEA */
758 }
759 
760 static inline void
761 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
762 {
763 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
764 }
765 
766 static inline void
767 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
768 {
769 	/*
770 	 * As shown in Section 7.6.3.2.3
771 	 */
772 	pt->pte_lo &= ~ptebit;
773 	TLBIE(va);
774 	SYNC();
775 	EIEIO();
776 	TLBSYNC();
777 	SYNC();
778 #ifdef MULTIPROCESSOR
779 	DCBST(pt);
780 #endif
781 }
782 
783 static inline void
784 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
785 {
786 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
787 	if (pvo_pt->pte_hi & PTE_VALID)
788 		panic("pte_set: setting an already valid pte %p", pvo_pt);
789 #endif
790 	pvo_pt->pte_hi |= PTE_VALID;
791 
792 	/*
793 	 * Update the PTE as defined in section 7.6.3.1
794 	 * Note that the REF/CHG bits are from pvo_pt and thus should
795 	 * have been saved so this routine can restore them (if desired).
796 	 */
797 	pt->pte_lo = pvo_pt->pte_lo;
798 	EIEIO();
799 	pt->pte_hi = pvo_pt->pte_hi;
800 	TLBSYNC();
801 	SYNC();
802 #ifdef MULTIPROCESSOR
803 	DCBST(pt);
804 #endif
805 	pmap_pte_valid++;
806 }
807 
808 static inline void
809 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
810 {
811 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
812 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
813 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
814 	if ((pt->pte_hi & PTE_VALID) == 0)
815 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
816 #endif
817 
818 	pvo_pt->pte_hi &= ~PTE_VALID;
819 	/*
820 	 * Force the ref & chg bits back into the PTEs.
821 	 */
822 	SYNC();
823 	/*
824 	 * Invalidate the pte ... (Section 7.6.3.3)
825 	 */
826 	pt->pte_hi &= ~PTE_VALID;
827 	SYNC();
828 	TLBIE(va);
829 	SYNC();
830 	EIEIO();
831 	TLBSYNC();
832 	SYNC();
833 	/*
834 	 * Save the ref & chg bits ...
835 	 */
836 	pmap_pte_synch(pt, pvo_pt);
837 	pmap_pte_valid--;
838 }
839 
840 static inline void
841 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
842 {
843 	/*
844 	 * Invalidate the PTE
845 	 */
846 	pmap_pte_unset(pt, pvo_pt, va);
847 	pmap_pte_set(pt, pvo_pt);
848 }
849 
850 /*
851  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
852  * (either primary or secondary location).
853  *
854  * Note: both the destination and source PTEs must not have PTE_VALID set.
855  */
856 
857 static int
858 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
859 {
860 	volatile struct pte *pt;
861 	int i;
862 
863 #if defined(DEBUG)
864 	DPRINTFN(PTE, "pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
865 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo);
866 #endif
867 	/*
868 	 * First try primary hash.
869 	 */
870 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
871 		if ((pt->pte_hi & PTE_VALID) == 0) {
872 			pvo_pt->pte_hi &= ~PTE_HID;
873 			pmap_pte_set(pt, pvo_pt);
874 			return i;
875 		}
876 	}
877 
878 	/*
879 	 * Now try secondary hash.
880 	 */
881 	ptegidx ^= pmap_pteg_mask;
882 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
883 		if ((pt->pte_hi & PTE_VALID) == 0) {
884 			pvo_pt->pte_hi |= PTE_HID;
885 			pmap_pte_set(pt, pvo_pt);
886 			return i;
887 		}
888 	}
889 	return -1;
890 }
891 
892 /*
893  * Spill handler.
894  *
895  * Tries to spill a page table entry from the overflow area.
896  * This runs in either real mode (if dealing with a exception spill)
897  * or virtual mode when dealing with manually spilling one of the
898  * kernel's pte entries.  In either case, interrupts are already
899  * disabled.
900  */
901 
902 int
903 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec)
904 {
905 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
906 	struct pvo_entry *pvo;
907 	/* XXX: gcc -- vpvoh is always set at either *1* or *2* */
908 	struct pvo_tqhead *pvoh, *vpvoh = NULL;
909 	int ptegidx, i, j;
910 	volatile struct pteg *pteg;
911 	volatile struct pte *pt;
912 
913 	PMAP_LOCK();
914 
915 	ptegidx = va_to_pteg(pm, addr);
916 
917 	/*
918 	 * Have to substitute some entry. Use the primary hash for this.
919 	 * Use low bits of timebase as random generator.  Make sure we are
920 	 * not picking a kernel pte for replacement.
921 	 */
922 	pteg = &pmap_pteg_table[ptegidx];
923 	i = MFTB() & 7;
924 	for (j = 0; j < 8; j++) {
925 		pt = &pteg->pt[i];
926 		if ((pt->pte_hi & PTE_VALID) == 0)
927 			break;
928 		if (VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT)
929 				< PHYSMAP_VSIDBITS)
930 			break;
931 		i = (i + 1) & 7;
932 	}
933 	KASSERT(j < 8);
934 
935 	source_pvo = NULL;
936 	victim_pvo = NULL;
937 	pvoh = &pmap_pvo_table[ptegidx];
938 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
939 
940 		/*
941 		 * We need to find pvo entry for this address...
942 		 */
943 		PMAP_PVO_CHECK(pvo);		/* sanity check */
944 
945 		/*
946 		 * If we haven't found the source and we come to a PVO with
947 		 * a valid PTE, then we know we can't find it because all
948 		 * evicted PVOs always are first in the list.
949 		 */
950 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
951 			break;
952 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
953 		    addr == PVO_VADDR(pvo)) {
954 
955 			/*
956 			 * Now we have found the entry to be spilled into the
957 			 * pteg.  Attempt to insert it into the page table.
958 			 */
959 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
960 			if (j >= 0) {
961 				PVO_PTEGIDX_SET(pvo, j);
962 				PMAP_PVO_CHECK(pvo);	/* sanity check */
963 				PVO_WHERE(pvo, SPILL_INSERT);
964 				pvo->pvo_pmap->pm_evictions--;
965 				PMAPCOUNT(ptes_spilled);
966 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
967 				    ? pmap_evcnt_ptes_secondary
968 				    : pmap_evcnt_ptes_primary)[j]);
969 
970 				/*
971 				 * Since we keep the evicted entries at the
972 				 * from of the PVO list, we need move this
973 				 * (now resident) PVO after the evicted
974 				 * entries.
975 				 */
976 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
977 
978 				/*
979 				 * If we don't have to move (either we were the
980 				 * last entry or the next entry was valid),
981 				 * don't change our position.  Otherwise
982 				 * move ourselves to the tail of the queue.
983 				 */
984 				if (next_pvo != NULL &&
985 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
986 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
987 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
988 				}
989 				PMAP_UNLOCK();
990 				return 1;
991 			}
992 			source_pvo = pvo;
993 			if (exec && !PVO_EXECUTABLE_P(source_pvo)) {
994 				return 0;
995 			}
996 			if (victim_pvo != NULL)
997 				break;
998 		}
999 
1000 		/*
1001 		 * We also need the pvo entry of the victim we are replacing
1002 		 * so save the R & C bits of the PTE.
1003 		 */
1004 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
1005 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
1006 			vpvoh = pvoh;			/* *1* */
1007 			victim_pvo = pvo;
1008 			if (source_pvo != NULL)
1009 				break;
1010 		}
1011 	}
1012 
1013 	if (source_pvo == NULL) {
1014 		PMAPCOUNT(ptes_unspilled);
1015 		PMAP_UNLOCK();
1016 		return 0;
1017 	}
1018 
1019 	if (victim_pvo == NULL) {
1020 		if ((pt->pte_hi & PTE_HID) == 0)
1021 			panic("pmap_pte_spill: victim p-pte (%p) has "
1022 			    "no pvo entry!", pt);
1023 
1024 		/*
1025 		 * If this is a secondary PTE, we need to search
1026 		 * its primary pvo bucket for the matching PVO.
1027 		 */
1028 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */
1029 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1030 			PMAP_PVO_CHECK(pvo);		/* sanity check */
1031 
1032 			/*
1033 			 * We also need the pvo entry of the victim we are
1034 			 * replacing so save the R & C bits of the PTE.
1035 			 */
1036 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1037 				victim_pvo = pvo;
1038 				break;
1039 			}
1040 		}
1041 		if (victim_pvo == NULL)
1042 			panic("pmap_pte_spill: victim s-pte (%p) has "
1043 			    "no pvo entry!", pt);
1044 	}
1045 
1046 	/*
1047 	 * The victim should be not be a kernel PVO/PTE entry.
1048 	 */
1049 	KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1050 	KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1051 	KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1052 
1053 	/*
1054 	 * We are invalidating the TLB entry for the EA for the
1055 	 * we are replacing even though its valid; If we don't
1056 	 * we lose any ref/chg bit changes contained in the TLB
1057 	 * entry.
1058 	 */
1059 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1060 
1061 	/*
1062 	 * To enforce the PVO list ordering constraint that all
1063 	 * evicted entries should come before all valid entries,
1064 	 * move the source PVO to the tail of its list and the
1065 	 * victim PVO to the head of its list (which might not be
1066 	 * the same list, if the victim was using the secondary hash).
1067 	 */
1068 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
1069 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
1070 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1071 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1072 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1073 	pmap_pte_set(pt, &source_pvo->pvo_pte);
1074 	victim_pvo->pvo_pmap->pm_evictions++;
1075 	source_pvo->pvo_pmap->pm_evictions--;
1076 	PVO_WHERE(victim_pvo, SPILL_UNSET);
1077 	PVO_WHERE(source_pvo, SPILL_SET);
1078 
1079 	PVO_PTEGIDX_CLR(victim_pvo);
1080 	PVO_PTEGIDX_SET(source_pvo, i);
1081 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1082 	PMAPCOUNT(ptes_spilled);
1083 	PMAPCOUNT(ptes_evicted);
1084 	PMAPCOUNT(ptes_removed);
1085 
1086 	PMAP_PVO_CHECK(victim_pvo);
1087 	PMAP_PVO_CHECK(source_pvo);
1088 
1089 	PMAP_UNLOCK();
1090 	return 1;
1091 }
1092 
1093 /*
1094  * Restrict given range to physical memory
1095  */
1096 void
1097 pmap_real_memory(paddr_t *start, psize_t *size)
1098 {
1099 	struct mem_region *mp;
1100 
1101 	for (mp = mem; mp->size; mp++) {
1102 		if (*start + *size > mp->start
1103 		    && *start < mp->start + mp->size) {
1104 			if (*start < mp->start) {
1105 				*size -= mp->start - *start;
1106 				*start = mp->start;
1107 			}
1108 			if (*start + *size > mp->start + mp->size)
1109 				*size = mp->start + mp->size - *start;
1110 			return;
1111 		}
1112 	}
1113 	*size = 0;
1114 }
1115 
1116 /*
1117  * Initialize anything else for pmap handling.
1118  * Called during vm_init().
1119  */
1120 void
1121 pmap_init(void)
1122 {
1123 	pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry),
1124 	    sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl",
1125 	    &pmap_pool_mallocator, IPL_NONE);
1126 
1127 	pool_setlowat(&pmap_mpvo_pool, 1008);
1128 
1129 	pmap_initialized = 1;
1130 
1131 }
1132 
1133 /*
1134  * How much virtual space does the kernel get?
1135  */
1136 void
1137 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1138 {
1139 	/*
1140 	 * For now, reserve one segment (minus some overhead) for kernel
1141 	 * virtual memory
1142 	 */
1143 	*start = VM_MIN_KERNEL_ADDRESS;
1144 	*end = VM_MAX_KERNEL_ADDRESS;
1145 }
1146 
1147 /*
1148  * Allocate, initialize, and return a new physical map.
1149  */
1150 pmap_t
1151 pmap_create(void)
1152 {
1153 	pmap_t pm;
1154 
1155 	pm = pool_get(&pmap_pool, PR_WAITOK);
1156 	KASSERT((vaddr_t)pm < VM_MIN_KERNEL_ADDRESS);
1157 	memset((void *)pm, 0, sizeof *pm);
1158 	pmap_pinit(pm);
1159 
1160 	DPRINTFN(CREATE, "pmap_create: pm %p:\n"
1161 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1162 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1163 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1164 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1165 	    pm,
1166 	    pm->pm_sr[0], pm->pm_sr[1],
1167 	    pm->pm_sr[2], pm->pm_sr[3],
1168 	    pm->pm_sr[4], pm->pm_sr[5],
1169 	    pm->pm_sr[6], pm->pm_sr[7],
1170 	    pm->pm_sr[8], pm->pm_sr[9],
1171 	    pm->pm_sr[10], pm->pm_sr[11],
1172 	    pm->pm_sr[12], pm->pm_sr[13],
1173 	    pm->pm_sr[14], pm->pm_sr[15]);
1174 	return pm;
1175 }
1176 
1177 /*
1178  * Initialize a preallocated and zeroed pmap structure.
1179  */
1180 void
1181 pmap_pinit(pmap_t pm)
1182 {
1183 	register_t entropy = MFTB();
1184 	register_t mask;
1185 	int i;
1186 
1187 	/*
1188 	 * Allocate some segment registers for this pmap.
1189 	 */
1190 	pm->pm_refs = 1;
1191 	PMAP_LOCK();
1192 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1193 		static register_t pmap_vsidcontext;
1194 		register_t hash;
1195 		unsigned int n;
1196 
1197 		/* Create a new value by multiplying by a prime adding in
1198 		 * entropy from the timebase register.  This is to make the
1199 		 * VSID more random so that the PT Hash function collides
1200 		 * less often. (note that the prime causes gcc to do shifts
1201 		 * instead of a multiply)
1202 		 */
1203 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1204 		hash = pmap_vsidcontext & (NPMAPS - 1);
1205 		if (hash == 0) {		/* 0 is special, avoid it */
1206 			entropy += 0xbadf00d;
1207 			continue;
1208 		}
1209 		n = hash >> 5;
1210 		mask = 1L << (hash & (VSID_NBPW-1));
1211 		hash = pmap_vsidcontext;
1212 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1213 			/* anything free in this bucket? */
1214 			if (~pmap_vsid_bitmap[n] == 0) {
1215 				entropy = hash ^ (hash >> 16);
1216 				continue;
1217 			}
1218 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1219 			mask = 1L << i;
1220 			hash &= ~(VSID_NBPW-1);
1221 			hash |= i;
1222 		}
1223 		hash &= PTE_VSID >> PTE_VSID_SHFT;
1224 		pmap_vsid_bitmap[n] |= mask;
1225 		pm->pm_vsid = hash;
1226 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1227 		for (i = 0; i < 16; i++)
1228 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1229 			    SR_NOEXEC;
1230 #endif
1231 		PMAP_UNLOCK();
1232 		return;
1233 	}
1234 	PMAP_UNLOCK();
1235 	panic("pmap_pinit: out of segments");
1236 }
1237 
1238 /*
1239  * Add a reference to the given pmap.
1240  */
1241 void
1242 pmap_reference(pmap_t pm)
1243 {
1244 	atomic_inc_uint(&pm->pm_refs);
1245 }
1246 
1247 /*
1248  * Retire the given pmap from service.
1249  * Should only be called if the map contains no valid mappings.
1250  */
1251 void
1252 pmap_destroy(pmap_t pm)
1253 {
1254 	if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1255 		pmap_release(pm);
1256 		pool_put(&pmap_pool, pm);
1257 	}
1258 }
1259 
1260 /*
1261  * Release any resources held by the given physical map.
1262  * Called when a pmap initialized by pmap_pinit is being released.
1263  */
1264 void
1265 pmap_release(pmap_t pm)
1266 {
1267 	int idx, mask;
1268 
1269 	KASSERT(pm->pm_stats.resident_count == 0);
1270 	KASSERT(pm->pm_stats.wired_count == 0);
1271 
1272 	PMAP_LOCK();
1273 	if (pm->pm_sr[0] == 0)
1274 		panic("pmap_release");
1275 	idx = pm->pm_vsid & (NPMAPS-1);
1276 	mask = 1 << (idx % VSID_NBPW);
1277 	idx /= VSID_NBPW;
1278 
1279 	KASSERT(pmap_vsid_bitmap[idx] & mask);
1280 	pmap_vsid_bitmap[idx] &= ~mask;
1281 	PMAP_UNLOCK();
1282 }
1283 
1284 /*
1285  * Copy the range specified by src_addr/len
1286  * from the source map to the range dst_addr/len
1287  * in the destination map.
1288  *
1289  * This routine is only advisory and need not do anything.
1290  */
1291 void
1292 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1293 	vsize_t len, vaddr_t src_addr)
1294 {
1295 	PMAPCOUNT(copies);
1296 }
1297 
1298 /*
1299  * Require that all active physical maps contain no
1300  * incorrect entries NOW.
1301  */
1302 void
1303 pmap_update(struct pmap *pmap)
1304 {
1305 	PMAPCOUNT(updates);
1306 	TLBSYNC();
1307 }
1308 
1309 static inline int
1310 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1311 {
1312 	int pteidx;
1313 	/*
1314 	 * We can find the actual pte entry without searching by
1315 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1316 	 * and by noticing the HID bit.
1317 	 */
1318 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1319 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1320 		pteidx ^= pmap_pteg_mask * 8;
1321 	return pteidx;
1322 }
1323 
1324 volatile struct pte *
1325 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1326 {
1327 	volatile struct pte *pt;
1328 
1329 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1330 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1331 		return NULL;
1332 #endif
1333 
1334 	/*
1335 	 * If we haven't been supplied the ptegidx, calculate it.
1336 	 */
1337 	if (pteidx == -1) {
1338 		int ptegidx;
1339 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1340 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1341 	}
1342 
1343 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1344 
1345 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1346 	return pt;
1347 #else
1348 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1349 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1350 		    "pvo but no valid pte index", pvo);
1351 	}
1352 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1353 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1354 		    "pvo but no valid pte", pvo);
1355 	}
1356 
1357 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1358 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1359 #if defined(DEBUG) || defined(PMAPCHECK)
1360 			pmap_pte_print(pt);
1361 #endif
1362 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1363 			    "pmap_pteg_table %p but invalid in pvo",
1364 			    pvo, pt);
1365 		}
1366 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1367 #if defined(DEBUG) || defined(PMAPCHECK)
1368 			pmap_pte_print(pt);
1369 #endif
1370 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1371 			    "not match pte %p in pmap_pteg_table",
1372 			    pvo, pt);
1373 		}
1374 		return pt;
1375 	}
1376 
1377 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1378 #if defined(DEBUG) || defined(PMAPCHECK)
1379 		pmap_pte_print(pt);
1380 #endif
1381 		panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1382 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1383 	}
1384 	return NULL;
1385 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1386 }
1387 
1388 struct pvo_entry *
1389 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1390 {
1391 	struct pvo_entry *pvo;
1392 	int ptegidx;
1393 
1394 	va &= ~ADDR_POFF;
1395 	ptegidx = va_to_pteg(pm, va);
1396 
1397 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1398 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1399 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1400 			panic("pmap_pvo_find_va: invalid pvo %p on "
1401 			    "list %#x (%p)", pvo, ptegidx,
1402 			     &pmap_pvo_table[ptegidx]);
1403 #endif
1404 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1405 			if (pteidx_p)
1406 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1407 			return pvo;
1408 		}
1409 	}
1410 	if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH))
1411 		panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1412 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1413 	return NULL;
1414 }
1415 
1416 #if defined(DEBUG) || defined(PMAPCHECK)
1417 void
1418 pmap_pvo_check(const struct pvo_entry *pvo)
1419 {
1420 	struct pvo_head *pvo_head;
1421 	struct pvo_entry *pvo0;
1422 	volatile struct pte *pt;
1423 	int failed = 0;
1424 
1425 	PMAP_LOCK();
1426 
1427 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1428 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1429 
1430 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1431 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1432 		    pvo, pvo->pvo_pmap);
1433 		failed = 1;
1434 	}
1435 
1436 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1437 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1438 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1439 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1440 		failed = 1;
1441 	}
1442 
1443 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1444 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1445 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1446 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1447 		failed = 1;
1448 	}
1449 
1450 	if (PVO_MANAGED_P(pvo)) {
1451 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1452 	} else {
1453 		if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) {
1454 			printf("pmap_pvo_check: pvo %p: non kernel address "
1455 			    "on kernel unmanaged list\n", pvo);
1456 			failed = 1;
1457 		}
1458 		pvo_head = &pmap_pvo_kunmanaged;
1459 	}
1460 	LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1461 		if (pvo0 == pvo)
1462 			break;
1463 	}
1464 	if (pvo0 == NULL) {
1465 		printf("pmap_pvo_check: pvo %p: not present "
1466 		    "on its vlist head %p\n", pvo, pvo_head);
1467 		failed = 1;
1468 	}
1469 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1470 		printf("pmap_pvo_check: pvo %p: not present "
1471 		    "on its olist head\n", pvo);
1472 		failed = 1;
1473 	}
1474 	pt = pmap_pvo_to_pte(pvo, -1);
1475 	if (pt == NULL) {
1476 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1477 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1478 			    "no PTE\n", pvo);
1479 			failed = 1;
1480 		}
1481 	} else {
1482 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1483 		    (uintptr_t) pt >=
1484 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1485 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1486 			    "pteg table\n", pvo, pt);
1487 			failed = 1;
1488 		}
1489 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1490 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1491 			    "no PTE\n", pvo);
1492 			failed = 1;
1493 		}
1494 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1495 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1496 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1497 			    pvo->pvo_pte.pte_hi,
1498 			    pt->pte_hi);
1499 			failed = 1;
1500 		}
1501 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1502 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1503 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1504 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1505 			    (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1506 			    (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1507 			failed = 1;
1508 		}
1509 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1510 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1511 			    " doesn't not match PVO's VA %#" _PRIxva "\n",
1512 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1513 			failed = 1;
1514 		}
1515 		if (failed)
1516 			pmap_pte_print(pt);
1517 	}
1518 	if (failed)
1519 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1520 		    pvo->pvo_pmap);
1521 
1522 	PMAP_UNLOCK();
1523 }
1524 #endif /* DEBUG || PMAPCHECK */
1525 
1526 /*
1527  * Search the PVO table looking for a non-wired entry.
1528  * If we find one, remove it and return it.
1529  */
1530 
1531 struct pvo_entry *
1532 pmap_pvo_reclaim(struct pmap *pm)
1533 {
1534 	struct pvo_tqhead *pvoh;
1535 	struct pvo_entry *pvo;
1536 	uint32_t idx, endidx;
1537 
1538 	endidx = pmap_pvo_reclaim_nextidx;
1539 	for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1540 	     idx = (idx + 1) & pmap_pteg_mask) {
1541 		pvoh = &pmap_pvo_table[idx];
1542 		TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1543 			if (!PVO_WIRED_P(pvo)) {
1544 				pmap_pvo_remove(pvo, -1, NULL);
1545 				pmap_pvo_reclaim_nextidx = idx;
1546 				PMAPCOUNT(pvos_reclaimed);
1547 				return pvo;
1548 			}
1549 		}
1550 	}
1551 	return NULL;
1552 }
1553 
1554 static struct pool *
1555 pmap_pvo_pl(struct pvo_entry *pvo)
1556 {
1557 
1558 	return PVO_MANAGED_P(pvo) ? &pmap_mpvo_pool : &pmap_upvo_pool;
1559 }
1560 
1561 /*
1562  * This returns whether this is the first mapping of a page.
1563  */
1564 int
1565 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1566 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1567 {
1568 	struct pvo_entry *pvo;
1569 	struct pvo_tqhead *pvoh;
1570 	register_t msr;
1571 	int ptegidx;
1572 	int i;
1573 	int poolflags = PR_NOWAIT;
1574 
1575 	/*
1576 	 * Compute the PTE Group index.
1577 	 */
1578 	va &= ~ADDR_POFF;
1579 	ptegidx = va_to_pteg(pm, va);
1580 
1581 	msr = pmap_interrupts_off();
1582 
1583 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1584 	if (pmap_pvo_remove_depth > 0)
1585 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1586 	if (++pmap_pvo_enter_depth > 1)
1587 		panic("pmap_pvo_enter: called recursively!");
1588 #endif
1589 
1590 	/*
1591 	 * Remove any existing mapping for this page.  Reuse the
1592 	 * pvo entry if there a mapping.
1593 	 */
1594 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1595 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1596 #ifdef DEBUG
1597 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1598 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1599 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1600 			   va < VM_MIN_KERNEL_ADDRESS) {
1601 				printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1602 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1603 				printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1604 				    pvo->pvo_pte.pte_hi,
1605 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1606 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1607 #ifdef DDBX
1608 				Debugger();
1609 #endif
1610 			}
1611 #endif
1612 			PMAPCOUNT(mappings_replaced);
1613 			pmap_pvo_remove(pvo, -1, NULL);
1614 			break;
1615 		}
1616 	}
1617 
1618 	/*
1619 	 * If we aren't overwriting an mapping, try to allocate
1620 	 */
1621 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1622 	--pmap_pvo_enter_depth;
1623 #endif
1624 	pmap_interrupts_restore(msr);
1625 	if (pvo) {
1626 		KASSERT(pmap_pvo_pl(pvo) == pl);
1627 	} else {
1628 		pvo = pool_get(pl, poolflags);
1629 	}
1630 	KASSERT((vaddr_t)pvo < VM_MIN_KERNEL_ADDRESS);
1631 
1632 #ifdef DEBUG
1633 	/*
1634 	 * Exercise pmap_pvo_reclaim() a little.
1635 	 */
1636 	if (pvo && (flags & PMAP_CANFAIL) != 0 &&
1637 	    pmap_pvo_reclaim_debugctr++ > 0x1000 &&
1638 	    (pmap_pvo_reclaim_debugctr & 0xff) == 0) {
1639 		pool_put(pl, pvo);
1640 		pvo = NULL;
1641 	}
1642 #endif
1643 
1644 	msr = pmap_interrupts_off();
1645 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1646 	++pmap_pvo_enter_depth;
1647 #endif
1648 	if (pvo == NULL) {
1649 		pvo = pmap_pvo_reclaim(pm);
1650 		if (pvo == NULL) {
1651 			if ((flags & PMAP_CANFAIL) == 0)
1652 				panic("pmap_pvo_enter: failed");
1653 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1654 			pmap_pvo_enter_depth--;
1655 #endif
1656 			PMAPCOUNT(pvos_failed);
1657 			pmap_interrupts_restore(msr);
1658 			return ENOMEM;
1659 		}
1660 	}
1661 
1662 	pvo->pvo_vaddr = va;
1663 	pvo->pvo_pmap = pm;
1664 	pvo->pvo_vaddr &= ~ADDR_POFF;
1665 	if (flags & VM_PROT_EXECUTE) {
1666 		PMAPCOUNT(exec_mappings);
1667 		pvo_set_exec(pvo);
1668 	}
1669 	if (flags & PMAP_WIRED)
1670 		pvo->pvo_vaddr |= PVO_WIRED;
1671 	if (pvo_head != &pmap_pvo_kunmanaged) {
1672 		pvo->pvo_vaddr |= PVO_MANAGED;
1673 		PMAPCOUNT(mappings);
1674 	} else {
1675 		PMAPCOUNT(kernel_mappings);
1676 	}
1677 	pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1678 
1679 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1680 	if (PVO_WIRED_P(pvo))
1681 		pvo->pvo_pmap->pm_stats.wired_count++;
1682 	pvo->pvo_pmap->pm_stats.resident_count++;
1683 #if defined(DEBUG)
1684 /*	if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */
1685 		DPRINTFN(PVOENTER,
1686 		    "pmap_pvo_enter: pvo %p: pm %p va %#" _PRIxva " pa %#" _PRIxpa "\n",
1687 		    pvo, pm, va, pa);
1688 #endif
1689 
1690 	/*
1691 	 * We hope this succeeds but it isn't required.
1692 	 */
1693 	pvoh = &pmap_pvo_table[ptegidx];
1694 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1695 	if (i >= 0) {
1696 		PVO_PTEGIDX_SET(pvo, i);
1697 		PVO_WHERE(pvo, ENTER_INSERT);
1698 		PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1699 		    ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1700 		TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1701 
1702 	} else {
1703 		/*
1704 		 * Since we didn't have room for this entry (which makes it
1705 		 * and evicted entry), place it at the head of the list.
1706 		 */
1707 		TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1708 		PMAPCOUNT(ptes_evicted);
1709 		pm->pm_evictions++;
1710 		/*
1711 		 * If this is a kernel page, make sure it's active.
1712 		 */
1713 		if (pm == pmap_kernel()) {
1714 			i = pmap_pte_spill(pm, va, false);
1715 			KASSERT(i);
1716 		}
1717 	}
1718 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1719 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1720 	pmap_pvo_enter_depth--;
1721 #endif
1722 	pmap_interrupts_restore(msr);
1723 	return 0;
1724 }
1725 
1726 static void
1727 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol)
1728 {
1729 	volatile struct pte *pt;
1730 	int ptegidx;
1731 
1732 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1733 	if (++pmap_pvo_remove_depth > 1)
1734 		panic("pmap_pvo_remove: called recursively!");
1735 #endif
1736 
1737 	/*
1738 	 * If we haven't been supplied the ptegidx, calculate it.
1739 	 */
1740 	if (pteidx == -1) {
1741 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1742 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1743 	} else {
1744 		ptegidx = pteidx >> 3;
1745 		if (pvo->pvo_pte.pte_hi & PTE_HID)
1746 			ptegidx ^= pmap_pteg_mask;
1747 	}
1748 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1749 
1750 	/*
1751 	 * If there is an active pte entry, we need to deactivate it
1752 	 * (and save the ref & chg bits).
1753 	 */
1754 	pt = pmap_pvo_to_pte(pvo, pteidx);
1755 	if (pt != NULL) {
1756 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1757 		PVO_WHERE(pvo, REMOVE);
1758 		PVO_PTEGIDX_CLR(pvo);
1759 		PMAPCOUNT(ptes_removed);
1760 	} else {
1761 		KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1762 		pvo->pvo_pmap->pm_evictions--;
1763 	}
1764 
1765 	/*
1766 	 * Account for executable mappings.
1767 	 */
1768 	if (PVO_EXECUTABLE_P(pvo))
1769 		pvo_clear_exec(pvo);
1770 
1771 	/*
1772 	 * Update our statistics.
1773 	 */
1774 	pvo->pvo_pmap->pm_stats.resident_count--;
1775 	if (PVO_WIRED_P(pvo))
1776 		pvo->pvo_pmap->pm_stats.wired_count--;
1777 
1778 	/*
1779 	 * Save the REF/CHG bits into their cache if the page is managed.
1780 	 */
1781 	if (PVO_MANAGED_P(pvo)) {
1782 		register_t ptelo = pvo->pvo_pte.pte_lo;
1783 		struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1784 
1785 		if (pg != NULL) {
1786 			/*
1787 			 * If this page was changed and it is mapped exec,
1788 			 * invalidate it.
1789 			 */
1790 			if ((ptelo & PTE_CHG) &&
1791 			    (pmap_attr_fetch(pg) & PTE_EXEC)) {
1792 				struct pvo_head *pvoh = vm_page_to_pvoh(pg);
1793 				if (LIST_EMPTY(pvoh)) {
1794 					DPRINTFN(EXEC, "[pmap_pvo_remove: "
1795 					    "%#" _PRIxpa ": clear-exec]\n",
1796 					    VM_PAGE_TO_PHYS(pg));
1797 					pmap_attr_clear(pg, PTE_EXEC);
1798 					PMAPCOUNT(exec_uncached_pvo_remove);
1799 				} else {
1800 					DPRINTFN(EXEC, "[pmap_pvo_remove: "
1801 					    "%#" _PRIxpa ": syncicache]\n",
1802 					    VM_PAGE_TO_PHYS(pg));
1803 					pmap_syncicache(VM_PAGE_TO_PHYS(pg),
1804 					    PAGE_SIZE);
1805 					PMAPCOUNT(exec_synced_pvo_remove);
1806 				}
1807 			}
1808 
1809 			pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1810 		}
1811 		PMAPCOUNT(unmappings);
1812 	} else {
1813 		PMAPCOUNT(kernel_unmappings);
1814 	}
1815 
1816 	/*
1817 	 * Remove the PVO from its lists and return it to the pool.
1818 	 */
1819 	LIST_REMOVE(pvo, pvo_vlink);
1820 	TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1821 	if (pvol) {
1822 		LIST_INSERT_HEAD(pvol, pvo, pvo_vlink);
1823 	}
1824 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1825 	pmap_pvo_remove_depth--;
1826 #endif
1827 }
1828 
1829 void
1830 pmap_pvo_free(struct pvo_entry *pvo)
1831 {
1832 
1833 	pool_put(pmap_pvo_pl(pvo), pvo);
1834 }
1835 
1836 void
1837 pmap_pvo_free_list(struct pvo_head *pvol)
1838 {
1839 	struct pvo_entry *pvo, *npvo;
1840 
1841 	for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) {
1842 		npvo = LIST_NEXT(pvo, pvo_vlink);
1843 		LIST_REMOVE(pvo, pvo_vlink);
1844 		pmap_pvo_free(pvo);
1845 	}
1846 }
1847 
1848 /*
1849  * Mark a mapping as executable.
1850  * If this is the first executable mapping in the segment,
1851  * clear the noexec flag.
1852  */
1853 static void
1854 pvo_set_exec(struct pvo_entry *pvo)
1855 {
1856 	struct pmap *pm = pvo->pvo_pmap;
1857 
1858 	if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) {
1859 		return;
1860 	}
1861 	pvo->pvo_vaddr |= PVO_EXECUTABLE;
1862 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1863 	{
1864 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1865 		if (pm->pm_exec[sr]++ == 0) {
1866 			pm->pm_sr[sr] &= ~SR_NOEXEC;
1867 		}
1868 	}
1869 #endif
1870 }
1871 
1872 /*
1873  * Mark a mapping as non-executable.
1874  * If this was the last executable mapping in the segment,
1875  * set the noexec flag.
1876  */
1877 static void
1878 pvo_clear_exec(struct pvo_entry *pvo)
1879 {
1880 	struct pmap *pm = pvo->pvo_pmap;
1881 
1882 	if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) {
1883 		return;
1884 	}
1885 	pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1886 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1887 	{
1888 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1889 		if (--pm->pm_exec[sr] == 0) {
1890 			pm->pm_sr[sr] |= SR_NOEXEC;
1891 		}
1892 	}
1893 #endif
1894 }
1895 
1896 /*
1897  * Insert physical page at pa into the given pmap at virtual address va.
1898  */
1899 int
1900 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1901 {
1902 	struct mem_region *mp;
1903 	struct pvo_head *pvo_head;
1904 	struct vm_page *pg;
1905 	struct pool *pl;
1906 	register_t pte_lo;
1907 	int error;
1908 	u_int was_exec = 0;
1909 
1910 	PMAP_LOCK();
1911 
1912 	if (__predict_false(!pmap_initialized)) {
1913 		pvo_head = &pmap_pvo_kunmanaged;
1914 		pl = &pmap_upvo_pool;
1915 		pg = NULL;
1916 		was_exec = PTE_EXEC;
1917 	} else {
1918 		pvo_head = pa_to_pvoh(pa, &pg);
1919 		pl = &pmap_mpvo_pool;
1920 	}
1921 
1922 	DPRINTFN(ENTER,
1923 	    "pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1924 	    pm, va, pa, prot, flags);
1925 
1926 	/*
1927 	 * If this is a managed page, and it's the first reference to the
1928 	 * page clear the execness of the page.  Otherwise fetch the execness.
1929 	 */
1930 	if (pg != NULL)
1931 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1932 
1933 	DPRINTFN(ENTER, " was_exec=%d", was_exec);
1934 
1935 	/*
1936 	 * Assume the page is cache inhibited and access is guarded unless
1937 	 * it's in our available memory array.  If it is in the memory array,
1938 	 * asssume it's in memory coherent memory.
1939 	 */
1940 	if (flags & PMAP_MD_PREFETCHABLE) {
1941 		pte_lo = 0;
1942 	} else
1943 		pte_lo = PTE_G;
1944 
1945 	if ((flags & PMAP_NOCACHE) == 0) {
1946 		for (mp = mem; mp->size; mp++) {
1947 			if (pa >= mp->start && pa < mp->start + mp->size) {
1948 				pte_lo = PTE_M;
1949 				break;
1950 			}
1951 		}
1952 #ifdef MULTIPROCESSOR
1953 		if (((mfpvr() >> 16) & 0xffff) == MPC603e)
1954 			pte_lo = PTE_M;
1955 #endif
1956 	} else {
1957 		pte_lo |= PTE_I;
1958 	}
1959 
1960 	if (prot & VM_PROT_WRITE)
1961 		pte_lo |= PTE_BW;
1962 	else
1963 		pte_lo |= PTE_BR;
1964 
1965 	/*
1966 	 * If this was in response to a fault, "pre-fault" the PTE's
1967 	 * changed/referenced bit appropriately.
1968 	 */
1969 	if (flags & VM_PROT_WRITE)
1970 		pte_lo |= PTE_CHG;
1971 	if (flags & VM_PROT_ALL)
1972 		pte_lo |= PTE_REF;
1973 
1974 	/*
1975 	 * We need to know if this page can be executable
1976 	 */
1977 	flags |= (prot & VM_PROT_EXECUTE);
1978 
1979 	/*
1980 	 * Record mapping for later back-translation and pte spilling.
1981 	 * This will overwrite any existing mapping.
1982 	 */
1983 	error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags);
1984 
1985 	/*
1986 	 * Flush the real page from the instruction cache if this page is
1987 	 * mapped executable and cacheable and has not been flushed since
1988 	 * the last time it was modified.
1989 	 */
1990 	if (error == 0 &&
1991             (flags & VM_PROT_EXECUTE) &&
1992             (pte_lo & PTE_I) == 0 &&
1993 	    was_exec == 0) {
1994 		DPRINTFN(ENTER, " %s", "syncicache");
1995 		PMAPCOUNT(exec_synced);
1996 		pmap_syncicache(pa, PAGE_SIZE);
1997 		if (pg != NULL) {
1998 			pmap_attr_save(pg, PTE_EXEC);
1999 			PMAPCOUNT(exec_cached);
2000 #if defined(DEBUG) || defined(PMAPDEBUG)
2001 			if (pmapdebug & PMAPDEBUG_ENTER)
2002 				printf(" marked-as-exec");
2003 			else if (pmapdebug & PMAPDEBUG_EXEC)
2004 				printf("[pmap_enter: %#" _PRIxpa ": marked-as-exec]\n",
2005 				    VM_PAGE_TO_PHYS(pg));
2006 
2007 #endif
2008 		}
2009 	}
2010 
2011 	DPRINTFN(ENTER, ": error=%d\n", error);
2012 
2013 	PMAP_UNLOCK();
2014 
2015 	return error;
2016 }
2017 
2018 void
2019 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
2020 {
2021 	struct mem_region *mp;
2022 	register_t pte_lo;
2023 	int error;
2024 
2025 #if defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA)
2026 	if (va < VM_MIN_KERNEL_ADDRESS)
2027 		panic("pmap_kenter_pa: attempt to enter "
2028 		    "non-kernel address %#" _PRIxva "!", va);
2029 #endif
2030 
2031 	DPRINTFN(KENTER,
2032 	    "pmap_kenter_pa(%#" _PRIxva ",%#" _PRIxpa ",%#x)\n", va, pa, prot);
2033 
2034 	PMAP_LOCK();
2035 
2036 	/*
2037 	 * Assume the page is cache inhibited and access is guarded unless
2038 	 * it's in our available memory array.  If it is in the memory array,
2039 	 * asssume it's in memory coherent memory.
2040 	 */
2041 	pte_lo = PTE_IG;
2042 	if ((flags & PMAP_NOCACHE) == 0) {
2043 		for (mp = mem; mp->size; mp++) {
2044 			if (pa >= mp->start && pa < mp->start + mp->size) {
2045 				pte_lo = PTE_M;
2046 				break;
2047 			}
2048 		}
2049 #ifdef MULTIPROCESSOR
2050 		if (((mfpvr() >> 16) & 0xffff) == MPC603e)
2051 			pte_lo = PTE_M;
2052 #endif
2053 	}
2054 
2055 	if (prot & VM_PROT_WRITE)
2056 		pte_lo |= PTE_BW;
2057 	else
2058 		pte_lo |= PTE_BR;
2059 
2060 	/*
2061 	 * We don't care about REF/CHG on PVOs on the unmanaged list.
2062 	 */
2063 	error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool,
2064 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED);
2065 
2066 	if (error != 0)
2067 		panic("pmap_kenter_pa: failed to enter va %#" _PRIxva " pa %#" _PRIxpa ": %d",
2068 		      va, pa, error);
2069 
2070 	PMAP_UNLOCK();
2071 }
2072 
2073 void
2074 pmap_kremove(vaddr_t va, vsize_t len)
2075 {
2076 	if (va < VM_MIN_KERNEL_ADDRESS)
2077 		panic("pmap_kremove: attempt to remove "
2078 		    "non-kernel address %#" _PRIxva "!", va);
2079 
2080 	DPRINTFN(KREMOVE, "pmap_kremove(%#" _PRIxva ",%#" _PRIxva ")\n", va, len);
2081 	pmap_remove(pmap_kernel(), va, va + len);
2082 }
2083 
2084 /*
2085  * Remove the given range of mapping entries.
2086  */
2087 void
2088 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
2089 {
2090 	struct pvo_head pvol;
2091 	struct pvo_entry *pvo;
2092 	register_t msr;
2093 	int pteidx;
2094 
2095 	PMAP_LOCK();
2096 	LIST_INIT(&pvol);
2097 	msr = pmap_interrupts_off();
2098 	for (; va < endva; va += PAGE_SIZE) {
2099 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2100 		if (pvo != NULL) {
2101 			pmap_pvo_remove(pvo, pteidx, &pvol);
2102 		}
2103 	}
2104 	pmap_interrupts_restore(msr);
2105 	pmap_pvo_free_list(&pvol);
2106 	PMAP_UNLOCK();
2107 }
2108 
2109 /*
2110  * Get the physical page address for the given pmap/virtual address.
2111  */
2112 bool
2113 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
2114 {
2115 	struct pvo_entry *pvo;
2116 	register_t msr;
2117 
2118 	PMAP_LOCK();
2119 
2120 	/*
2121 	 * If this is a kernel pmap lookup, also check the battable
2122 	 * and if we get a hit, translate the VA to a PA using the
2123 	 * BAT entries.  Don't check for VM_MAX_KERNEL_ADDRESS is
2124 	 * that will wrap back to 0.
2125 	 */
2126 	if (pm == pmap_kernel() &&
2127 	    (va < VM_MIN_KERNEL_ADDRESS ||
2128 	     (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
2129 		KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
2130 #if defined (PMAP_OEA)
2131 #ifdef PPC_OEA601
2132 		if ((MFPVR() >> 16) == MPC601) {
2133 			register_t batu = battable[va >> 23].batu;
2134 			register_t batl = battable[va >> 23].batl;
2135 			register_t sr = iosrtable[va >> ADDR_SR_SHFT];
2136 			if (BAT601_VALID_P(batl) &&
2137 			    BAT601_VA_MATCH_P(batu, batl, va)) {
2138 				register_t mask =
2139 				    (~(batl & BAT601_BSM) << 17) & ~0x1ffffL;
2140 				if (pap)
2141 					*pap = (batl & mask) | (va & ~mask);
2142 				PMAP_UNLOCK();
2143 				return true;
2144 			} else if (SR601_VALID_P(sr) &&
2145 				   SR601_PA_MATCH_P(sr, va)) {
2146 				if (pap)
2147 					*pap = va;
2148 				PMAP_UNLOCK();
2149 				return true;
2150 			}
2151 		} else
2152 #endif /* PPC_OEA601 */
2153 		{
2154 			register_t batu = battable[BAT_VA2IDX(va)].batu;
2155 			if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) {
2156 				register_t batl = battable[BAT_VA2IDX(va)].batl;
2157 				register_t mask =
2158 				    (~(batu & (BAT_XBL|BAT_BL)) << 15) & ~0x1ffffL;
2159 				if (pap)
2160 					*pap = (batl & mask) | (va & ~mask);
2161 				PMAP_UNLOCK();
2162 				return true;
2163 			}
2164 		}
2165 		return false;
2166 #elif defined (PMAP_OEA64_BRIDGE)
2167 	if (va >= SEGMENT_LENGTH)
2168 		panic("%s: pm: %s va >= SEGMENT_LENGTH, va: 0x%08lx\n",
2169 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
2170 	else {
2171 		if (pap)
2172 			*pap = va;
2173 			PMAP_UNLOCK();
2174 			return true;
2175 	}
2176 #elif defined (PMAP_OEA64)
2177 #error PPC_OEA64 not supported
2178 #endif /* PPC_OEA */
2179 	}
2180 
2181 	msr = pmap_interrupts_off();
2182 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2183 	if (pvo != NULL) {
2184 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2185 		if (pap)
2186 			*pap = (pvo->pvo_pte.pte_lo & PTE_RPGN)
2187 			    | (va & ADDR_POFF);
2188 	}
2189 	pmap_interrupts_restore(msr);
2190 	PMAP_UNLOCK();
2191 	return pvo != NULL;
2192 }
2193 
2194 /*
2195  * Lower the protection on the specified range of this pmap.
2196  */
2197 void
2198 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
2199 {
2200 	struct pvo_entry *pvo;
2201 	volatile struct pte *pt;
2202 	register_t msr;
2203 	int pteidx;
2204 
2205 	/*
2206 	 * Since this routine only downgrades protection, we should
2207 	 * always be called with at least one bit not set.
2208 	 */
2209 	KASSERT(prot != VM_PROT_ALL);
2210 
2211 	/*
2212 	 * If there is no protection, this is equivalent to
2213 	 * remove the pmap from the pmap.
2214 	 */
2215 	if ((prot & VM_PROT_READ) == 0) {
2216 		pmap_remove(pm, va, endva);
2217 		return;
2218 	}
2219 
2220 	PMAP_LOCK();
2221 
2222 	msr = pmap_interrupts_off();
2223 	for (; va < endva; va += PAGE_SIZE) {
2224 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2225 		if (pvo == NULL)
2226 			continue;
2227 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2228 
2229 		/*
2230 		 * Revoke executable if asked to do so.
2231 		 */
2232 		if ((prot & VM_PROT_EXECUTE) == 0)
2233 			pvo_clear_exec(pvo);
2234 
2235 #if 0
2236 		/*
2237 		 * If the page is already read-only, no change
2238 		 * needs to be made.
2239 		 */
2240 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
2241 			continue;
2242 #endif
2243 		/*
2244 		 * Grab the PTE pointer before we diddle with
2245 		 * the cached PTE copy.
2246 		 */
2247 		pt = pmap_pvo_to_pte(pvo, pteidx);
2248 		/*
2249 		 * Change the protection of the page.
2250 		 */
2251 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2252 		pvo->pvo_pte.pte_lo |= PTE_BR;
2253 
2254 		/*
2255 		 * If the PVO is in the page table, update
2256 		 * that pte at well.
2257 		 */
2258 		if (pt != NULL) {
2259 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2260 			PVO_WHERE(pvo, PMAP_PROTECT);
2261 			PMAPCOUNT(ptes_changed);
2262 		}
2263 
2264 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2265 	}
2266 	pmap_interrupts_restore(msr);
2267 	PMAP_UNLOCK();
2268 }
2269 
2270 void
2271 pmap_unwire(pmap_t pm, vaddr_t va)
2272 {
2273 	struct pvo_entry *pvo;
2274 	register_t msr;
2275 
2276 	PMAP_LOCK();
2277 	msr = pmap_interrupts_off();
2278 	pvo = pmap_pvo_find_va(pm, va, NULL);
2279 	if (pvo != NULL) {
2280 		if (PVO_WIRED_P(pvo)) {
2281 			pvo->pvo_vaddr &= ~PVO_WIRED;
2282 			pm->pm_stats.wired_count--;
2283 		}
2284 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2285 	}
2286 	pmap_interrupts_restore(msr);
2287 	PMAP_UNLOCK();
2288 }
2289 
2290 /*
2291  * Lower the protection on the specified physical page.
2292  */
2293 void
2294 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
2295 {
2296 	struct pvo_head *pvo_head, pvol;
2297 	struct pvo_entry *pvo, *next_pvo;
2298 	volatile struct pte *pt;
2299 	register_t msr;
2300 
2301 	PMAP_LOCK();
2302 
2303 	KASSERT(prot != VM_PROT_ALL);
2304 	LIST_INIT(&pvol);
2305 	msr = pmap_interrupts_off();
2306 
2307 	/*
2308 	 * When UVM reuses a page, it does a pmap_page_protect with
2309 	 * VM_PROT_NONE.  At that point, we can clear the exec flag
2310 	 * since we know the page will have different contents.
2311 	 */
2312 	if ((prot & VM_PROT_READ) == 0) {
2313 		DPRINTFN(EXEC, "[pmap_page_protect: %#" _PRIxpa ": clear-exec]\n",
2314 		    VM_PAGE_TO_PHYS(pg));
2315 		if (pmap_attr_fetch(pg) & PTE_EXEC) {
2316 			PMAPCOUNT(exec_uncached_page_protect);
2317 			pmap_attr_clear(pg, PTE_EXEC);
2318 		}
2319 	}
2320 
2321 	pvo_head = vm_page_to_pvoh(pg);
2322 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2323 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
2324 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2325 
2326 		/*
2327 		 * Downgrading to no mapping at all, we just remove the entry.
2328 		 */
2329 		if ((prot & VM_PROT_READ) == 0) {
2330 			pmap_pvo_remove(pvo, -1, &pvol);
2331 			continue;
2332 		}
2333 
2334 		/*
2335 		 * If EXEC permission is being revoked, just clear the
2336 		 * flag in the PVO.
2337 		 */
2338 		if ((prot & VM_PROT_EXECUTE) == 0)
2339 			pvo_clear_exec(pvo);
2340 
2341 		/*
2342 		 * If this entry is already RO, don't diddle with the
2343 		 * page table.
2344 		 */
2345 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2346 			PMAP_PVO_CHECK(pvo);
2347 			continue;
2348 		}
2349 
2350 		/*
2351 		 * Grab the PTE before the we diddle the bits so
2352 		 * pvo_to_pte can verify the pte contents are as
2353 		 * expected.
2354 		 */
2355 		pt = pmap_pvo_to_pte(pvo, -1);
2356 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2357 		pvo->pvo_pte.pte_lo |= PTE_BR;
2358 		if (pt != NULL) {
2359 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2360 			PVO_WHERE(pvo, PMAP_PAGE_PROTECT);
2361 			PMAPCOUNT(ptes_changed);
2362 		}
2363 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2364 	}
2365 	pmap_interrupts_restore(msr);
2366 	pmap_pvo_free_list(&pvol);
2367 
2368 	PMAP_UNLOCK();
2369 }
2370 
2371 /*
2372  * Activate the address space for the specified process.  If the process
2373  * is the current process, load the new MMU context.
2374  */
2375 void
2376 pmap_activate(struct lwp *l)
2377 {
2378 	struct pcb *pcb = lwp_getpcb(l);
2379 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2380 
2381 	DPRINTFN(ACTIVATE,
2382 	    "pmap_activate: lwp %p (curlwp %p)\n", l, curlwp);
2383 
2384 	/*
2385 	 * XXX Normally performed in cpu_lwp_fork().
2386 	 */
2387 	pcb->pcb_pm = pmap;
2388 
2389 	/*
2390 	* In theory, the SR registers need only be valid on return
2391 	* to user space wait to do them there.
2392 	*/
2393 	if (l == curlwp) {
2394 		/* Store pointer to new current pmap. */
2395 		curpm = pmap;
2396 	}
2397 }
2398 
2399 /*
2400  * Deactivate the specified process's address space.
2401  */
2402 void
2403 pmap_deactivate(struct lwp *l)
2404 {
2405 }
2406 
2407 bool
2408 pmap_query_bit(struct vm_page *pg, int ptebit)
2409 {
2410 	struct pvo_entry *pvo;
2411 	volatile struct pte *pt;
2412 	register_t msr;
2413 
2414 	PMAP_LOCK();
2415 
2416 	if (pmap_attr_fetch(pg) & ptebit) {
2417 		PMAP_UNLOCK();
2418 		return true;
2419 	}
2420 
2421 	msr = pmap_interrupts_off();
2422 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2423 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2424 		/*
2425 		 * See if we saved the bit off.  If so cache, it and return
2426 		 * success.
2427 		 */
2428 		if (pvo->pvo_pte.pte_lo & ptebit) {
2429 			pmap_attr_save(pg, ptebit);
2430 			PMAP_PVO_CHECK(pvo);		/* sanity check */
2431 			pmap_interrupts_restore(msr);
2432 			PMAP_UNLOCK();
2433 			return true;
2434 		}
2435 	}
2436 	/*
2437 	 * No luck, now go thru the hard part of looking at the ptes
2438 	 * themselves.  Sync so any pending REF/CHG bits are flushed
2439 	 * to the PTEs.
2440 	 */
2441 	SYNC();
2442 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2443 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2444 		/*
2445 		 * See if this pvo have a valid PTE.  If so, fetch the
2446 		 * REF/CHG bits from the valid PTE.  If the appropriate
2447 		 * ptebit is set, cache, it and return success.
2448 		 */
2449 		pt = pmap_pvo_to_pte(pvo, -1);
2450 		if (pt != NULL) {
2451 			pmap_pte_synch(pt, &pvo->pvo_pte);
2452 			if (pvo->pvo_pte.pte_lo & ptebit) {
2453 				pmap_attr_save(pg, ptebit);
2454 				PMAP_PVO_CHECK(pvo);		/* sanity check */
2455 				pmap_interrupts_restore(msr);
2456 				PMAP_UNLOCK();
2457 				return true;
2458 			}
2459 		}
2460 	}
2461 	pmap_interrupts_restore(msr);
2462 	PMAP_UNLOCK();
2463 	return false;
2464 }
2465 
2466 bool
2467 pmap_clear_bit(struct vm_page *pg, int ptebit)
2468 {
2469 	struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2470 	struct pvo_entry *pvo;
2471 	volatile struct pte *pt;
2472 	register_t msr;
2473 	int rv = 0;
2474 
2475 	PMAP_LOCK();
2476 	msr = pmap_interrupts_off();
2477 
2478 	/*
2479 	 * Fetch the cache value
2480 	 */
2481 	rv |= pmap_attr_fetch(pg);
2482 
2483 	/*
2484 	 * Clear the cached value.
2485 	 */
2486 	pmap_attr_clear(pg, ptebit);
2487 
2488 	/*
2489 	 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2490 	 * can reset the right ones).  Note that since the pvo entries and
2491 	 * list heads are accessed via BAT0 and are never placed in the
2492 	 * page table, we don't have to worry about further accesses setting
2493 	 * the REF/CHG bits.
2494 	 */
2495 	SYNC();
2496 
2497 	/*
2498 	 * For each pvo entry, clear pvo's ptebit.  If this pvo have a
2499 	 * valid PTE.  If so, clear the ptebit from the valid PTE.
2500 	 */
2501 	LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2502 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2503 		pt = pmap_pvo_to_pte(pvo, -1);
2504 		if (pt != NULL) {
2505 			/*
2506 			 * Only sync the PTE if the bit we are looking
2507 			 * for is not already set.
2508 			 */
2509 			if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2510 				pmap_pte_synch(pt, &pvo->pvo_pte);
2511 			/*
2512 			 * If the bit we are looking for was already set,
2513 			 * clear that bit in the pte.
2514 			 */
2515 			if (pvo->pvo_pte.pte_lo & ptebit)
2516 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2517 		}
2518 		rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2519 		pvo->pvo_pte.pte_lo &= ~ptebit;
2520 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2521 	}
2522 	pmap_interrupts_restore(msr);
2523 
2524 	/*
2525 	 * If we are clearing the modify bit and this page was marked EXEC
2526 	 * and the user of the page thinks the page was modified, then we
2527 	 * need to clean it from the icache if it's mapped or clear the EXEC
2528 	 * bit if it's not mapped.  The page itself might not have the CHG
2529 	 * bit set if the modification was done via DMA to the page.
2530 	 */
2531 	if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2532 		if (LIST_EMPTY(pvoh)) {
2533 			DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": clear-exec]\n",
2534 			    VM_PAGE_TO_PHYS(pg));
2535 			pmap_attr_clear(pg, PTE_EXEC);
2536 			PMAPCOUNT(exec_uncached_clear_modify);
2537 		} else {
2538 			DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": syncicache]\n",
2539 			    VM_PAGE_TO_PHYS(pg));
2540 			pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE);
2541 			PMAPCOUNT(exec_synced_clear_modify);
2542 		}
2543 	}
2544 	PMAP_UNLOCK();
2545 	return (rv & ptebit) != 0;
2546 }
2547 
2548 void
2549 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2550 {
2551 	struct pvo_entry *pvo;
2552 	size_t offset = va & ADDR_POFF;
2553 	int s;
2554 
2555 	PMAP_LOCK();
2556 	s = splvm();
2557 	while (len > 0) {
2558 		size_t seglen = PAGE_SIZE - offset;
2559 		if (seglen > len)
2560 			seglen = len;
2561 		pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2562 		if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) {
2563 			pmap_syncicache(
2564 			    (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2565 			PMAP_PVO_CHECK(pvo);
2566 		}
2567 		va += seglen;
2568 		len -= seglen;
2569 		offset = 0;
2570 	}
2571 	splx(s);
2572 	PMAP_UNLOCK();
2573 }
2574 
2575 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2576 void
2577 pmap_pte_print(volatile struct pte *pt)
2578 {
2579 	printf("PTE %p: ", pt);
2580 
2581 #if defined(PMAP_OEA)
2582 	/* High word: */
2583 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2584 #else
2585 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2586 #endif /* PMAP_OEA */
2587 
2588 	printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2589 	printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2590 
2591 	printf("%#" _PRIxpte " %#" _PRIxpte "",
2592 	    (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2593 	    pt->pte_hi & PTE_API);
2594 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2595 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2596 #else
2597 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2598 #endif /* PMAP_OEA */
2599 
2600 	/* Low word: */
2601 #if defined (PMAP_OEA)
2602 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2603 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2604 #else
2605 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2606 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2607 #endif
2608 	printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2609 	printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2610 	printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2611 	printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2612 	printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2613 	printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2614 	switch (pt->pte_lo & PTE_PP) {
2615 	case PTE_BR: printf("br]\n"); break;
2616 	case PTE_BW: printf("bw]\n"); break;
2617 	case PTE_SO: printf("so]\n"); break;
2618 	case PTE_SW: printf("sw]\n"); break;
2619 	}
2620 }
2621 #endif
2622 
2623 #if defined(DDB)
2624 void
2625 pmap_pteg_check(void)
2626 {
2627 	volatile struct pte *pt;
2628 	int i;
2629 	int ptegidx;
2630 	u_int p_valid = 0;
2631 	u_int s_valid = 0;
2632 	u_int invalid = 0;
2633 
2634 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2635 		for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2636 			if (pt->pte_hi & PTE_VALID) {
2637 				if (pt->pte_hi & PTE_HID)
2638 					s_valid++;
2639 				else
2640 				{
2641 					p_valid++;
2642 				}
2643 			} else
2644 				invalid++;
2645 		}
2646 	}
2647 	printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2648 		p_valid, p_valid, s_valid, s_valid,
2649 		invalid, invalid);
2650 }
2651 
2652 void
2653 pmap_print_mmuregs(void)
2654 {
2655 	int i;
2656 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2657 	u_int cpuvers;
2658 #endif
2659 #ifndef PMAP_OEA64
2660 	vaddr_t addr;
2661 	register_t soft_sr[16];
2662 #endif
2663 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2664 	struct bat soft_ibat[4];
2665 	struct bat soft_dbat[4];
2666 #endif
2667 	paddr_t sdr1;
2668 
2669 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2670 	cpuvers = MFPVR() >> 16;
2671 #endif
2672 	__asm volatile ("mfsdr1 %0" : "=r"(sdr1));
2673 #ifndef PMAP_OEA64
2674 	addr = 0;
2675 	for (i = 0; i < 16; i++) {
2676 		soft_sr[i] = MFSRIN(addr);
2677 		addr += (1 << ADDR_SR_SHFT);
2678 	}
2679 #endif
2680 
2681 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2682 	/* read iBAT (601: uBAT) registers */
2683 	__asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2684 	__asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2685 	__asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2686 	__asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2687 	__asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2688 	__asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2689 	__asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2690 	__asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2691 
2692 
2693 	if (cpuvers != MPC601) {
2694 		/* read dBAT registers */
2695 		__asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2696 		__asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2697 		__asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2698 		__asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2699 		__asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2700 		__asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2701 		__asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2702 		__asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2703 	}
2704 #endif
2705 
2706 	printf("SDR1:\t%#" _PRIxpa "\n", sdr1);
2707 #ifndef PMAP_OEA64
2708 	printf("SR[]:\t");
2709 	for (i = 0; i < 4; i++)
2710 		printf("0x%08lx,   ", soft_sr[i]);
2711 	printf("\n\t");
2712 	for ( ; i < 8; i++)
2713 		printf("0x%08lx,   ", soft_sr[i]);
2714 	printf("\n\t");
2715 	for ( ; i < 12; i++)
2716 		printf("0x%08lx,   ", soft_sr[i]);
2717 	printf("\n\t");
2718 	for ( ; i < 16; i++)
2719 		printf("0x%08lx,   ", soft_sr[i]);
2720 	printf("\n");
2721 #endif
2722 
2723 #if defined(PMAP_OEA) || defined(PMAP_OEA_BRIDGE)
2724 	printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2725 	for (i = 0; i < 4; i++) {
2726 		printf("0x%08lx 0x%08lx, ",
2727 			soft_ibat[i].batu, soft_ibat[i].batl);
2728 		if (i == 1)
2729 			printf("\n\t");
2730 	}
2731 	if (cpuvers != MPC601) {
2732 		printf("\ndBAT[]:\t");
2733 		for (i = 0; i < 4; i++) {
2734 			printf("0x%08lx 0x%08lx, ",
2735 				soft_dbat[i].batu, soft_dbat[i].batl);
2736 			if (i == 1)
2737 				printf("\n\t");
2738 		}
2739 	}
2740 	printf("\n");
2741 #endif /* PMAP_OEA... */
2742 }
2743 
2744 void
2745 pmap_print_pte(pmap_t pm, vaddr_t va)
2746 {
2747 	struct pvo_entry *pvo;
2748 	volatile struct pte *pt;
2749 	int pteidx;
2750 
2751 	pvo = pmap_pvo_find_va(pm, va, &pteidx);
2752 	if (pvo != NULL) {
2753 		pt = pmap_pvo_to_pte(pvo, pteidx);
2754 		if (pt != NULL) {
2755 			printf("VA %#" _PRIxva " -> %p -> %s %#" _PRIxpte ", %#" _PRIxpte "\n",
2756 				va, pt,
2757 				pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2758 				pt->pte_hi, pt->pte_lo);
2759 		} else {
2760 			printf("No valid PTE found\n");
2761 		}
2762 	} else {
2763 		printf("Address not in pmap\n");
2764 	}
2765 }
2766 
2767 void
2768 pmap_pteg_dist(void)
2769 {
2770 	struct pvo_entry *pvo;
2771 	int ptegidx;
2772 	int depth;
2773 	int max_depth = 0;
2774 	unsigned int depths[64];
2775 
2776 	memset(depths, 0, sizeof(depths));
2777 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2778 		depth = 0;
2779 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2780 			depth++;
2781 		}
2782 		if (depth > max_depth)
2783 			max_depth = depth;
2784 		if (depth > 63)
2785 			depth = 63;
2786 		depths[depth]++;
2787 	}
2788 
2789 	for (depth = 0; depth < 64; depth++) {
2790 		printf("  [%2d]: %8u", depth, depths[depth]);
2791 		if ((depth & 3) == 3)
2792 			printf("\n");
2793 		if (depth == max_depth)
2794 			break;
2795 	}
2796 	if ((depth & 3) != 3)
2797 		printf("\n");
2798 	printf("Max depth found was %d\n", max_depth);
2799 }
2800 #endif /* DEBUG */
2801 
2802 #if defined(PMAPCHECK) || defined(DEBUG)
2803 void
2804 pmap_pvo_verify(void)
2805 {
2806 	int ptegidx;
2807 	int s;
2808 
2809 	s = splvm();
2810 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2811 		struct pvo_entry *pvo;
2812 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2813 			if ((uintptr_t) pvo >= SEGMENT_LENGTH)
2814 				panic("pmap_pvo_verify: invalid pvo %p "
2815 				    "on list %#x", pvo, ptegidx);
2816 			pmap_pvo_check(pvo);
2817 		}
2818 	}
2819 	splx(s);
2820 }
2821 #endif /* PMAPCHECK */
2822 
2823 
2824 void *
2825 pmap_pool_ualloc(struct pool *pp, int flags)
2826 {
2827 	struct pvo_page *pvop;
2828 
2829 	if (uvm.page_init_done != true) {
2830 		return (void *) uvm_pageboot_alloc(PAGE_SIZE);
2831 	}
2832 
2833 	PMAP_LOCK();
2834 	pvop = SIMPLEQ_FIRST(&pmap_upvop_head);
2835 	if (pvop != NULL) {
2836 		pmap_upvop_free--;
2837 		SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link);
2838 		PMAP_UNLOCK();
2839 		return pvop;
2840 	}
2841 	PMAP_UNLOCK();
2842 	return pmap_pool_malloc(pp, flags);
2843 }
2844 
2845 void *
2846 pmap_pool_malloc(struct pool *pp, int flags)
2847 {
2848 	struct pvo_page *pvop;
2849 	struct vm_page *pg;
2850 
2851 	PMAP_LOCK();
2852 	pvop = SIMPLEQ_FIRST(&pmap_mpvop_head);
2853 	if (pvop != NULL) {
2854 		pmap_mpvop_free--;
2855 		SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link);
2856 		PMAP_UNLOCK();
2857 		return pvop;
2858 	}
2859 	PMAP_UNLOCK();
2860  again:
2861 	pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE,
2862 	    UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256);
2863 	if (__predict_false(pg == NULL)) {
2864 		if (flags & PR_WAITOK) {
2865 			uvm_wait("plpg");
2866 			goto again;
2867 		} else {
2868 			return (0);
2869 		}
2870 	}
2871 	KDASSERT(VM_PAGE_TO_PHYS(pg) == (uintptr_t)VM_PAGE_TO_PHYS(pg));
2872 	return (void *)(uintptr_t) VM_PAGE_TO_PHYS(pg);
2873 }
2874 
2875 void
2876 pmap_pool_ufree(struct pool *pp, void *va)
2877 {
2878 	struct pvo_page *pvop;
2879 #if 0
2880 	if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) {
2881 		pmap_pool_mfree(va, size, tag);
2882 		return;
2883 	}
2884 #endif
2885 	PMAP_LOCK();
2886 	pvop = va;
2887 	SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link);
2888 	pmap_upvop_free++;
2889 	if (pmap_upvop_free > pmap_upvop_maxfree)
2890 		pmap_upvop_maxfree = pmap_upvop_free;
2891 	PMAP_UNLOCK();
2892 }
2893 
2894 void
2895 pmap_pool_mfree(struct pool *pp, void *va)
2896 {
2897 	struct pvo_page *pvop;
2898 
2899 	PMAP_LOCK();
2900 	pvop = va;
2901 	SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link);
2902 	pmap_mpvop_free++;
2903 	if (pmap_mpvop_free > pmap_mpvop_maxfree)
2904 		pmap_mpvop_maxfree = pmap_mpvop_free;
2905 	PMAP_UNLOCK();
2906 #if 0
2907 	uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va));
2908 #endif
2909 }
2910 
2911 /*
2912  * This routine in bootstraping to steal to-be-managed memory (which will
2913  * then be unmanaged).  We use it to grab from the first 256MB for our
2914  * pmap needs and above 256MB for other stuff.
2915  */
2916 vaddr_t
2917 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2918 {
2919 	vsize_t size;
2920 	vaddr_t va;
2921 	paddr_t start, end, pa = 0;
2922 	int npgs, freelist;
2923 	uvm_physseg_t bank;
2924 
2925 	if (uvm.page_init_done == true)
2926 		panic("pmap_steal_memory: called _after_ bootstrap");
2927 
2928 	*vstartp = VM_MIN_KERNEL_ADDRESS;
2929 	*vendp = VM_MAX_KERNEL_ADDRESS;
2930 
2931 	size = round_page(vsize);
2932 	npgs = atop(size);
2933 
2934 	/*
2935 	 * PA 0 will never be among those given to UVM so we can use it
2936 	 * to indicate we couldn't steal any memory.
2937 	 */
2938 
2939 	for (bank = uvm_physseg_get_first();
2940 	     uvm_physseg_valid_p(bank);
2941 	     bank = uvm_physseg_get_next(bank)) {
2942 
2943 		freelist = uvm_physseg_get_free_list(bank);
2944 		start = uvm_physseg_get_start(bank);
2945 		end = uvm_physseg_get_end(bank);
2946 
2947 		if (freelist == VM_FREELIST_FIRST256 &&
2948 		    (end - start) >= npgs) {
2949 			pa = ptoa(start);
2950 			break;
2951 		}
2952 	}
2953 
2954 	if (pa == 0)
2955 		panic("pmap_steal_memory: no approriate memory to steal!");
2956 
2957 	uvm_physseg_unplug(start, npgs);
2958 
2959 	va = (vaddr_t) pa;
2960 	memset((void *) va, 0, size);
2961 	pmap_pages_stolen += npgs;
2962 #ifdef DEBUG
2963 	if (pmapdebug && npgs > 1) {
2964 		u_int cnt = 0;
2965 	for (bank = uvm_physseg_get_first();
2966 	     uvm_physseg_valid_p(bank);
2967 	     bank = uvm_physseg_get_next(bank)) {
2968 		cnt += uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank);
2969 		}
2970 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2971 		    npgs, pmap_pages_stolen, cnt);
2972 	}
2973 #endif
2974 
2975 	return va;
2976 }
2977 
2978 /*
2979  * Find a chuck of memory with right size and alignment.
2980  */
2981 paddr_t
2982 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2983 {
2984 	struct mem_region *mp;
2985 	paddr_t s, e;
2986 	int i, j;
2987 
2988 	size = round_page(size);
2989 
2990 	DPRINTFN(BOOT,
2991 	    "pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
2992 	    size, alignment, at_end);
2993 
2994 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2995 		panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
2996 		    alignment);
2997 
2998 	if (at_end) {
2999 		if (alignment != PAGE_SIZE)
3000 			panic("pmap_boot_find_memory: invalid ending "
3001 			    "alignment %#" _PRIxpa, alignment);
3002 
3003 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
3004 			s = mp->start + mp->size - size;
3005 			if (s >= mp->start && mp->size >= size) {
3006 				DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3007 				DPRINTFN(BOOT,
3008 				    "pmap_boot_find_memory: b-avail[%d] start "
3009 				    "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3010 				     mp->start, mp->size);
3011 				mp->size -= size;
3012 				DPRINTFN(BOOT,
3013 				    "pmap_boot_find_memory: a-avail[%d] start "
3014 				    "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3015 				     mp->start, mp->size);
3016 				return s;
3017 			}
3018 		}
3019 		panic("pmap_boot_find_memory: no available memory");
3020 	}
3021 
3022 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3023 		s = (mp->start + alignment - 1) & ~(alignment-1);
3024 		e = s + size;
3025 
3026 		/*
3027 		 * Is the calculated region entirely within the region?
3028 		 */
3029 		if (s < mp->start || e > mp->start + mp->size)
3030 			continue;
3031 
3032 		DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3033 		if (s == mp->start) {
3034 			/*
3035 			 * If the block starts at the beginning of region,
3036 			 * adjust the size & start. (the region may now be
3037 			 * zero in length)
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->start += size;
3043 			mp->size -= size;
3044 			DPRINTFN(BOOT,
3045 			    "pmap_boot_find_memory: a-avail[%d] start "
3046 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3047 		} else if (e == mp->start + mp->size) {
3048 			/*
3049 			 * If the block starts at the beginning of region,
3050 			 * adjust only the size.
3051 			 */
3052 			DPRINTFN(BOOT,
3053 			    "pmap_boot_find_memory: b-avail[%d] start "
3054 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3055 			mp->size -= size;
3056 			DPRINTFN(BOOT,
3057 			    "pmap_boot_find_memory: a-avail[%d] start "
3058 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3059 		} else {
3060 			/*
3061 			 * Block is in the middle of the region, so we
3062 			 * have to split it in two.
3063 			 */
3064 			for (j = avail_cnt; j > i + 1; j--) {
3065 				avail[j] = avail[j-1];
3066 			}
3067 			DPRINTFN(BOOT,
3068 			    "pmap_boot_find_memory: b-avail[%d] start "
3069 			    "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3070 			mp[1].start = e;
3071 			mp[1].size = mp[0].start + mp[0].size - e;
3072 			mp[0].size = s - mp[0].start;
3073 			avail_cnt++;
3074 			for (; i < avail_cnt; i++) {
3075 				DPRINTFN(BOOT,
3076 				    "pmap_boot_find_memory: a-avail[%d] "
3077 				    "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3078 				     avail[i].start, avail[i].size);
3079 			}
3080 		}
3081 		KASSERT(s == (uintptr_t) s);
3082 		return s;
3083 	}
3084 	panic("pmap_boot_find_memory: not enough memory for "
3085 	    "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3086 }
3087 
3088 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3089 #if defined (PMAP_OEA64_BRIDGE)
3090 int
3091 pmap_setup_segment0_map(int use_large_pages, ...)
3092 {
3093     vaddr_t va, va_end;
3094 
3095     register_t pte_lo = 0x0;
3096     int ptegidx = 0;
3097     struct pte pte;
3098     va_list ap;
3099 
3100     /* Coherent + Supervisor RW, no user access */
3101     pte_lo = PTE_M;
3102 
3103     /* XXXSL
3104      * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3105      * these have to take priority.
3106      */
3107     for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3108         ptegidx = va_to_pteg(pmap_kernel(), va);
3109         pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3110         (void)pmap_pte_insert(ptegidx, &pte);
3111     }
3112 
3113     va_start(ap, use_large_pages);
3114     while (1) {
3115         paddr_t pa;
3116         size_t size;
3117 
3118         va = va_arg(ap, vaddr_t);
3119 
3120         if (va == 0)
3121             break;
3122 
3123         pa = va_arg(ap, paddr_t);
3124         size = va_arg(ap, size_t);
3125 
3126         for (va_end = va + size; va < va_end; va += 0x1000, pa += 0x1000) {
3127 #if 0
3128 	    printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__,  va, pa);
3129 #endif
3130             ptegidx = va_to_pteg(pmap_kernel(), va);
3131             pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3132             (void)pmap_pte_insert(ptegidx, &pte);
3133         }
3134     }
3135     va_end(ap);
3136 
3137     TLBSYNC();
3138     SYNC();
3139     return (0);
3140 }
3141 #endif /* PMAP_OEA64_BRIDGE */
3142 
3143 /*
3144  * This is not part of the defined PMAP interface and is specific to the
3145  * PowerPC architecture.  This is called during initppc, before the system
3146  * is really initialized.
3147  */
3148 void
3149 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3150 {
3151 	struct mem_region *mp, tmp;
3152 	paddr_t s, e;
3153 	psize_t size;
3154 	int i, j;
3155 
3156 	/*
3157 	 * Get memory.
3158 	 */
3159 	mem_regions(&mem, &avail);
3160 #if defined(DEBUG)
3161 	if (pmapdebug & PMAPDEBUG_BOOT) {
3162 		printf("pmap_bootstrap: memory configuration:\n");
3163 		for (mp = mem; mp->size; mp++) {
3164 			printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3165 				mp->start, mp->size);
3166 		}
3167 		for (mp = avail; mp->size; mp++) {
3168 			printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3169 				mp->start, mp->size);
3170 		}
3171 	}
3172 #endif
3173 
3174 	/*
3175 	 * Find out how much physical memory we have and in how many chunks.
3176 	 */
3177 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3178 		if (mp->start >= pmap_memlimit)
3179 			continue;
3180 		if (mp->start + mp->size > pmap_memlimit) {
3181 			size = pmap_memlimit - mp->start;
3182 			physmem += btoc(size);
3183 		} else {
3184 			physmem += btoc(mp->size);
3185 		}
3186 		mem_cnt++;
3187 	}
3188 
3189 	/*
3190 	 * Count the number of available entries.
3191 	 */
3192 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
3193 		avail_cnt++;
3194 
3195 	/*
3196 	 * Page align all regions.
3197 	 */
3198 	kernelstart = trunc_page(kernelstart);
3199 	kernelend = round_page(kernelend);
3200 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3201 		s = round_page(mp->start);
3202 		mp->size -= (s - mp->start);
3203 		mp->size = trunc_page(mp->size);
3204 		mp->start = s;
3205 		e = mp->start + mp->size;
3206 
3207 		DPRINTFN(BOOT,
3208 		    "pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3209 		    i, mp->start, mp->size);
3210 
3211 		/*
3212 		 * Don't allow the end to run beyond our artificial limit
3213 		 */
3214 		if (e > pmap_memlimit)
3215 			e = pmap_memlimit;
3216 
3217 		/*
3218 		 * Is this region empty or strange?  skip it.
3219 		 */
3220 		if (e <= s) {
3221 			mp->start = 0;
3222 			mp->size = 0;
3223 			continue;
3224 		}
3225 
3226 		/*
3227 		 * Does this overlap the beginning of kernel?
3228 		 *   Does extend past the end of the kernel?
3229 		 */
3230 		else if (s < kernelstart && e > kernelstart) {
3231 			if (e > kernelend) {
3232 				avail[avail_cnt].start = kernelend;
3233 				avail[avail_cnt].size = e - kernelend;
3234 				avail_cnt++;
3235 			}
3236 			mp->size = kernelstart - s;
3237 		}
3238 		/*
3239 		 * Check whether this region overlaps the end of the kernel.
3240 		 */
3241 		else if (s < kernelend && e > kernelend) {
3242 			mp->start = kernelend;
3243 			mp->size = e - kernelend;
3244 		}
3245 		/*
3246 		 * Look whether this regions is completely inside the kernel.
3247 		 * Nuke it if it does.
3248 		 */
3249 		else if (s >= kernelstart && e <= kernelend) {
3250 			mp->start = 0;
3251 			mp->size = 0;
3252 		}
3253 		/*
3254 		 * If the user imposed a memory limit, enforce it.
3255 		 */
3256 		else if (s >= pmap_memlimit) {
3257 			mp->start = -PAGE_SIZE;	/* let's know why */
3258 			mp->size = 0;
3259 		}
3260 		else {
3261 			mp->start = s;
3262 			mp->size = e - s;
3263 		}
3264 		DPRINTFN(BOOT,
3265 		    "pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3266 		    i, mp->start, mp->size);
3267 	}
3268 
3269 	/*
3270 	 * Move (and uncount) all the null return to the end.
3271 	 */
3272 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3273 		if (mp->size == 0) {
3274 			tmp = avail[i];
3275 			avail[i] = avail[--avail_cnt];
3276 			avail[avail_cnt] = avail[i];
3277 		}
3278 	}
3279 
3280 	/*
3281 	 * (Bubble)sort them into ascending order.
3282 	 */
3283 	for (i = 0; i < avail_cnt; i++) {
3284 		for (j = i + 1; j < avail_cnt; j++) {
3285 			if (avail[i].start > avail[j].start) {
3286 				tmp = avail[i];
3287 				avail[i] = avail[j];
3288 				avail[j] = tmp;
3289 			}
3290 		}
3291 	}
3292 
3293 	/*
3294 	 * Make sure they don't overlap.
3295 	 */
3296 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3297 		if (mp[0].start + mp[0].size > mp[1].start) {
3298 			mp[0].size = mp[1].start - mp[0].start;
3299 		}
3300 		DPRINTFN(BOOT,
3301 		    "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3302 		    i, mp->start, mp->size);
3303 	}
3304 	DPRINTFN(BOOT,
3305 	    "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3306 	    i, mp->start, mp->size);
3307 
3308 #ifdef	PTEGCOUNT
3309 	pmap_pteg_cnt = PTEGCOUNT;
3310 #else /* PTEGCOUNT */
3311 
3312 	pmap_pteg_cnt = 0x1000;
3313 
3314 	while (pmap_pteg_cnt < physmem)
3315 		pmap_pteg_cnt <<= 1;
3316 
3317 	pmap_pteg_cnt >>= 1;
3318 #endif /* PTEGCOUNT */
3319 
3320 #ifdef DEBUG
3321 	DPRINTFN(BOOT, "pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt);
3322 #endif
3323 
3324 	/*
3325 	 * Find suitably aligned memory for PTEG hash table.
3326 	 */
3327 	size = pmap_pteg_cnt * sizeof(struct pteg);
3328 	pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3329 
3330 #ifdef DEBUG
3331 	DPRINTFN(BOOT,
3332 		"PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table);
3333 #endif
3334 
3335 
3336 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3337 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
3338 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > 256MB",
3339 		    pmap_pteg_table, size);
3340 #endif
3341 
3342 	memset(__UNVOLATILE(pmap_pteg_table), 0,
3343 		pmap_pteg_cnt * sizeof(struct pteg));
3344 	pmap_pteg_mask = pmap_pteg_cnt - 1;
3345 
3346 	/*
3347 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3348 	 * with pages.  So we just steal them before giving them to UVM.
3349 	 */
3350 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3351 	pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3352 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3353 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
3354 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > 256MB",
3355 		    pmap_pvo_table, size);
3356 #endif
3357 
3358 	for (i = 0; i < pmap_pteg_cnt; i++)
3359 		TAILQ_INIT(&pmap_pvo_table[i]);
3360 
3361 #ifndef MSGBUFADDR
3362 	/*
3363 	 * Allocate msgbuf in high memory.
3364 	 */
3365 	msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3366 #endif
3367 
3368 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3369 		paddr_t pfstart = atop(mp->start);
3370 		paddr_t pfend = atop(mp->start + mp->size);
3371 		if (mp->size == 0)
3372 			continue;
3373 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
3374 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3375 				VM_FREELIST_FIRST256);
3376 		} else if (mp->start >= SEGMENT_LENGTH) {
3377 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3378 				VM_FREELIST_DEFAULT);
3379 		} else {
3380 			pfend = atop(SEGMENT_LENGTH);
3381 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3382 				VM_FREELIST_FIRST256);
3383 			pfstart = atop(SEGMENT_LENGTH);
3384 			pfend = atop(mp->start + mp->size);
3385 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3386 				VM_FREELIST_DEFAULT);
3387 		}
3388 	}
3389 
3390 	/*
3391 	 * Make sure kernel vsid is allocated as well as VSID 0.
3392 	 */
3393 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3394 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3395 	pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3396 		|= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3397 	pmap_vsid_bitmap[0] |= 1;
3398 
3399 	/*
3400 	 * Initialize kernel pmap and hardware.
3401 	 */
3402 
3403 /* PMAP_OEA64_BRIDGE does support these instructions */
3404 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
3405 	for (i = 0; i < 16; i++) {
3406 #if defined(PPC_OEA601)
3407 	    /* XXX wedges for segment register 0xf , so set later */
3408 	    if ((iosrtable[i] & SR601_T) && ((MFPVR() >> 16) == MPC601))
3409 		    continue;
3410 #endif
3411  		pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3412 		__asm volatile ("mtsrin %0,%1"
3413  			      :: "r"(KERNELN_SEGMENT(i)|SR_PRKEY), "r"(i << ADDR_SR_SHFT));
3414 	}
3415 
3416 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3417 	__asm volatile ("mtsr %0,%1"
3418 		      :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
3419 #ifdef KERNEL2_SR
3420 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3421 	__asm volatile ("mtsr %0,%1"
3422 		      :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
3423 #endif
3424 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3425 #if defined (PMAP_OEA)
3426 	for (i = 0; i < 16; i++) {
3427 		if (iosrtable[i] & SR601_T) {
3428 			pmap_kernel()->pm_sr[i] = iosrtable[i];
3429 			__asm volatile ("mtsrin %0,%1"
3430 			    :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT));
3431 		}
3432 	}
3433 	__asm volatile ("sync; mtsdr1 %0; isync"
3434 		      :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10)));
3435 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
3436  	__asm __volatile ("sync; mtsdr1 %0; isync"
3437  		      :: "r"((uintptr_t)pmap_pteg_table | (32 - __builtin_clz(pmap_pteg_mask >> 11))));
3438 #endif
3439 	tlbia();
3440 
3441 #ifdef ALTIVEC
3442 	pmap_use_altivec = cpu_altivec;
3443 #endif
3444 
3445 #ifdef DEBUG
3446 	if (pmapdebug & PMAPDEBUG_BOOT) {
3447 		u_int cnt;
3448 		uvm_physseg_t bank;
3449 		char pbuf[9];
3450 		for (cnt = 0, bank = uvm_physseg_get_first();
3451 		     uvm_physseg_valid_p(bank);
3452 		     bank = uvm_physseg_get_next(bank)) {
3453 			cnt += uvm_physseg_get_avail_end(bank) -
3454 			    uvm_physseg_get_avail_start(bank);
3455 			printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3456 			    bank,
3457 			    ptoa(uvm_physseg_get_avail_start(bank)),
3458 			    ptoa(uvm_physseg_get_avail_end(bank)),
3459 			    ptoa(uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank)));
3460 		}
3461 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3462 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3463 		    pbuf, cnt);
3464 	}
3465 #endif
3466 
3467 	pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry),
3468 	    sizeof(struct pvo_entry), 0, 0, "pmap_upvopl",
3469 	    &pmap_pool_uallocator, IPL_VM);
3470 
3471 	pool_setlowat(&pmap_upvo_pool, 252);
3472 
3473 	pool_init(&pmap_pool, sizeof(struct pmap),
3474 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator,
3475 	    IPL_NONE);
3476 
3477 #if defined(PMAP_NEED_MAPKERNEL)
3478 	{
3479 		struct pmap *pm = pmap_kernel();
3480 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3481 		extern int etext[], kernel_text[];
3482 		vaddr_t va, va_etext = (paddr_t) etext;
3483 #endif
3484 		paddr_t pa, pa_end;
3485 		register_t sr;
3486 		struct pte pt;
3487 		unsigned int ptegidx;
3488 		int bank;
3489 
3490 		sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3491 		pm->pm_sr[0] = sr;
3492 
3493 		for (bank = 0; bank < vm_nphysseg; bank++) {
3494 			pa_end = ptoa(VM_PHYSMEM_PTR(bank)->avail_end);
3495 			pa = ptoa(VM_PHYSMEM_PTR(bank)->avail_start);
3496 			for (; pa < pa_end; pa += PAGE_SIZE) {
3497 				ptegidx = va_to_pteg(pm, pa);
3498 				pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3499 				pmap_pte_insert(ptegidx, &pt);
3500 			}
3501 		}
3502 
3503 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3504 		va = (vaddr_t) kernel_text;
3505 
3506 		for (pa = kernelstart; va < va_etext;
3507 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3508 			ptegidx = va_to_pteg(pm, va);
3509 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3510 			pmap_pte_insert(ptegidx, &pt);
3511 		}
3512 
3513 		for (; pa < kernelend;
3514 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3515 			ptegidx = va_to_pteg(pm, va);
3516 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3517 			pmap_pte_insert(ptegidx, &pt);
3518 		}
3519 
3520 		for (va = 0, pa = 0; va < kernelstart;
3521 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3522 			ptegidx = va_to_pteg(pm, va);
3523 			if (va < 0x3000)
3524 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3525 			else
3526 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3527 			pmap_pte_insert(ptegidx, &pt);
3528 		}
3529 		for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3530 		    pa += PAGE_SIZE, va += PAGE_SIZE) {
3531 			ptegidx = va_to_pteg(pm, va);
3532 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3533 			pmap_pte_insert(ptegidx, &pt);
3534 		}
3535 #endif
3536 
3537 		__asm volatile ("mtsrin %0,%1"
3538  			      :: "r"(sr), "r"(kernelstart));
3539 	}
3540 #endif
3541 
3542 #if defined(PMAPDEBUG)
3543 	if ( pmapdebug )
3544 	    pmap_print_mmuregs();
3545 #endif
3546 }
3547