xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: pmap.c,v 1.75 2011/01/18 01:02:55 matt 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 1.75 2011/01/18 01:02:55 matt Exp $");
67 
68 #define	PMAP_NOOPNAMES
69 
70 #include "opt_ppcarch.h"
71 #include "opt_altivec.h"
72 #include "opt_multiprocessor.h"
73 #include "opt_pmap.h"
74 
75 #include <sys/param.h>
76 #include <sys/malloc.h>
77 #include <sys/proc.h>
78 #include <sys/pool.h>
79 #include <sys/queue.h>
80 #include <sys/device.h>		/* for evcnt */
81 #include <sys/systm.h>
82 #include <sys/atomic.h>
83 
84 #include <uvm/uvm.h>
85 
86 #include <machine/pcb.h>
87 #include <machine/powerpc.h>
88 #include <powerpc/spr.h>
89 #include <powerpc/bat.h>
90 #include <powerpc/stdarg.h>
91 #include <powerpc/oea/spr.h>
92 #include <powerpc/oea/sr_601.h>
93 
94 #ifdef ALTIVEC
95 int pmap_use_altivec;
96 #endif
97 
98 volatile struct pteg *pmap_pteg_table;
99 unsigned int pmap_pteg_cnt;
100 unsigned int pmap_pteg_mask;
101 #ifdef PMAP_MEMLIMIT
102 static paddr_t pmap_memlimit = PMAP_MEMLIMIT;
103 #else
104 static paddr_t pmap_memlimit = -PAGE_SIZE;		/* there is no limit */
105 #endif
106 
107 struct pmap kernel_pmap_;
108 unsigned int pmap_pages_stolen;
109 u_long pmap_pte_valid;
110 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
111 u_long pmap_pvo_enter_depth;
112 u_long pmap_pvo_remove_depth;
113 #endif
114 
115 #ifndef MSGBUFADDR
116 extern paddr_t msgbuf_paddr;
117 #endif
118 
119 static struct mem_region *mem, *avail;
120 static u_int mem_cnt, avail_cnt;
121 
122 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
123 # define	PMAP_OEA 1
124 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA64) && !defined(PPC_OEA64_BRIDGE)
125 #  define	PMAPNAME(name)	pmap_##name
126 # endif
127 #endif
128 
129 #if defined(PMAP_OEA64)
130 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA) && !defined(PPC_OEA64_BRIDGE)
131 #  define	PMAPNAME(name)	pmap_##name
132 # endif
133 #endif
134 
135 #if defined(PMAP_OEA64_BRIDGE)
136 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA) && !defined(PPC_OEA64)
137 #  define	PMAPNAME(name)	pmap_##name
138 # endif
139 #endif
140 
141 #if defined(PMAP_OEA)
142 #define	_PRIxpte	"lx"
143 #else
144 #define	_PRIxpte	PRIx64
145 #endif
146 #define	_PRIxpa		"lx"
147 #define	_PRIxva		"lx"
148 #define	_PRIsr  	"lx"
149 
150 #if defined(PMAP_EXCLUDE_DECLS) && !defined(PMAPNAME)
151 #if defined(PMAP_OEA)
152 #define	PMAPNAME(name)	pmap32_##name
153 #elif defined(PMAP_OEA64)
154 #define	PMAPNAME(name)	pmap64_##name
155 #elif defined(PMAP_OEA64_BRIDGE)
156 #define	PMAPNAME(name)	pmap64bridge_##name
157 #else
158 #error unknown variant for pmap
159 #endif
160 #endif /* PMAP_EXLCUDE_DECLS && !PMAPNAME */
161 
162 #if defined(PMAPNAME)
163 #define	STATIC			static
164 #define pmap_pte_spill		PMAPNAME(pte_spill)
165 #define pmap_real_memory	PMAPNAME(real_memory)
166 #define pmap_init		PMAPNAME(init)
167 #define pmap_virtual_space	PMAPNAME(virtual_space)
168 #define pmap_create		PMAPNAME(create)
169 #define pmap_reference		PMAPNAME(reference)
170 #define pmap_destroy		PMAPNAME(destroy)
171 #define pmap_copy		PMAPNAME(copy)
172 #define pmap_update		PMAPNAME(update)
173 #define pmap_enter		PMAPNAME(enter)
174 #define pmap_remove		PMAPNAME(remove)
175 #define pmap_kenter_pa		PMAPNAME(kenter_pa)
176 #define pmap_kremove		PMAPNAME(kremove)
177 #define pmap_extract		PMAPNAME(extract)
178 #define pmap_protect		PMAPNAME(protect)
179 #define pmap_unwire		PMAPNAME(unwire)
180 #define pmap_page_protect	PMAPNAME(page_protect)
181 #define pmap_query_bit		PMAPNAME(query_bit)
182 #define pmap_clear_bit		PMAPNAME(clear_bit)
183 
184 #define pmap_activate		PMAPNAME(activate)
185 #define pmap_deactivate		PMAPNAME(deactivate)
186 
187 #define pmap_pinit		PMAPNAME(pinit)
188 #define pmap_procwr		PMAPNAME(procwr)
189 
190 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
191 #define pmap_pte_print		PMAPNAME(pte_print)
192 #define pmap_pteg_check		PMAPNAME(pteg_check)
193 #define pmap_print_mmruregs	PMAPNAME(print_mmuregs)
194 #define pmap_print_pte		PMAPNAME(print_pte)
195 #define pmap_pteg_dist		PMAPNAME(pteg_dist)
196 #endif
197 #if defined(DEBUG) || defined(PMAPCHECK)
198 #define	pmap_pvo_verify		PMAPNAME(pvo_verify)
199 #define pmapcheck		PMAPNAME(check)
200 #endif
201 #if defined(DEBUG) || defined(PMAPDEBUG)
202 #define pmapdebug		PMAPNAME(debug)
203 #endif
204 #define pmap_steal_memory	PMAPNAME(steal_memory)
205 #define pmap_bootstrap		PMAPNAME(bootstrap)
206 #else
207 #define	STATIC			/* nothing */
208 #endif /* PMAPNAME */
209 
210 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
211 STATIC void pmap_real_memory(paddr_t *, psize_t *);
212 STATIC void pmap_init(void);
213 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
214 STATIC pmap_t pmap_create(void);
215 STATIC void pmap_reference(pmap_t);
216 STATIC void pmap_destroy(pmap_t);
217 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
218 STATIC void pmap_update(pmap_t);
219 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
220 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
221 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t, u_int);
222 STATIC void pmap_kremove(vaddr_t, vsize_t);
223 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
224 
225 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
226 STATIC void pmap_unwire(pmap_t, vaddr_t);
227 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
228 STATIC bool pmap_query_bit(struct vm_page *, int);
229 STATIC bool pmap_clear_bit(struct vm_page *, int);
230 
231 STATIC void pmap_activate(struct lwp *);
232 STATIC void pmap_deactivate(struct lwp *);
233 
234 STATIC void pmap_pinit(pmap_t pm);
235 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
236 
237 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
238 STATIC void pmap_pte_print(volatile struct pte *);
239 STATIC void pmap_pteg_check(void);
240 STATIC void pmap_print_mmuregs(void);
241 STATIC void pmap_print_pte(pmap_t, vaddr_t);
242 STATIC void pmap_pteg_dist(void);
243 #endif
244 #if defined(DEBUG) || defined(PMAPCHECK)
245 STATIC void pmap_pvo_verify(void);
246 #endif
247 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
248 STATIC void pmap_bootstrap(paddr_t, paddr_t);
249 
250 #ifdef PMAPNAME
251 const struct pmap_ops PMAPNAME(ops) = {
252 	.pmapop_pte_spill = pmap_pte_spill,
253 	.pmapop_real_memory = pmap_real_memory,
254 	.pmapop_init = pmap_init,
255 	.pmapop_virtual_space = pmap_virtual_space,
256 	.pmapop_create = pmap_create,
257 	.pmapop_reference = pmap_reference,
258 	.pmapop_destroy = pmap_destroy,
259 	.pmapop_copy = pmap_copy,
260 	.pmapop_update = pmap_update,
261 	.pmapop_enter = pmap_enter,
262 	.pmapop_remove = pmap_remove,
263 	.pmapop_kenter_pa = pmap_kenter_pa,
264 	.pmapop_kremove = pmap_kremove,
265 	.pmapop_extract = pmap_extract,
266 	.pmapop_protect = pmap_protect,
267 	.pmapop_unwire = pmap_unwire,
268 	.pmapop_page_protect = pmap_page_protect,
269 	.pmapop_query_bit = pmap_query_bit,
270 	.pmapop_clear_bit = pmap_clear_bit,
271 	.pmapop_activate = pmap_activate,
272 	.pmapop_deactivate = pmap_deactivate,
273 	.pmapop_pinit = pmap_pinit,
274 	.pmapop_procwr = pmap_procwr,
275 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
276 	.pmapop_pte_print = pmap_pte_print,
277 	.pmapop_pteg_check = pmap_pteg_check,
278 	.pmapop_print_mmuregs = pmap_print_mmuregs,
279 	.pmapop_print_pte = pmap_print_pte,
280 	.pmapop_pteg_dist = pmap_pteg_dist,
281 #else
282 	.pmapop_pte_print = NULL,
283 	.pmapop_pteg_check = NULL,
284 	.pmapop_print_mmuregs = NULL,
285 	.pmapop_print_pte = NULL,
286 	.pmapop_pteg_dist = NULL,
287 #endif
288 #if defined(DEBUG) || defined(PMAPCHECK)
289 	.pmapop_pvo_verify = pmap_pvo_verify,
290 #else
291 	.pmapop_pvo_verify = NULL,
292 #endif
293 	.pmapop_steal_memory = pmap_steal_memory,
294 	.pmapop_bootstrap = pmap_bootstrap,
295 };
296 #endif /* !PMAPNAME */
297 
298 /*
299  * The following structure is aligned to 32 bytes
300  */
301 struct pvo_entry {
302 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
303 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
304 	struct pte pvo_pte;			/* Prebuilt PTE */
305 	pmap_t pvo_pmap;			/* ptr to owning pmap */
306 	vaddr_t pvo_vaddr;			/* VA of entry */
307 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
308 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
309 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
310 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
311 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
312 #define	PVO_WIRED_P(pvo)	((pvo)->pvo_vaddr & PVO_WIRED)
313 #define	PVO_MANAGED_P(pvo)	((pvo)->pvo_vaddr & PVO_MANAGED)
314 #define	PVO_EXECUTABLE_P(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
315 #define	PVO_ENTER_INSERT	0		/* PVO has been removed */
316 #define	PVO_SPILL_UNSET		1		/* PVO has been evicted */
317 #define	PVO_SPILL_SET		2		/* PVO has been spilled */
318 #define	PVO_SPILL_INSERT	3		/* PVO has been inserted */
319 #define	PVO_PMAP_PAGE_PROTECT	4		/* PVO has changed */
320 #define	PVO_PMAP_PROTECT	5		/* PVO has changed */
321 #define	PVO_REMOVE		6		/* PVO has been removed */
322 #define	PVO_WHERE_MASK		15
323 #define	PVO_WHERE_SHFT		8
324 } __attribute__ ((aligned (32)));
325 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
326 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
327 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
328 #define	PVO_PTEGIDX_CLR(pvo)	\
329 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
330 #define	PVO_PTEGIDX_SET(pvo,i)	\
331 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
332 #define	PVO_WHERE(pvo,w)	\
333 	((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
334 	 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
335 
336 TAILQ_HEAD(pvo_tqhead, pvo_entry);
337 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
338 static struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
339 static struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
340 
341 struct pool pmap_pool;		/* pool for pmap structures */
342 struct pool pmap_upvo_pool;	/* pool for pvo entries for unmanaged pages */
343 struct pool pmap_mpvo_pool;	/* pool for pvo entries for managed pages */
344 
345 /*
346  * We keep a cache of unmanaged pages to be used for pvo entries for
347  * unmanaged pages.
348  */
349 struct pvo_page {
350 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
351 };
352 SIMPLEQ_HEAD(pvop_head, pvo_page);
353 static struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head);
354 static struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head);
355 u_long pmap_upvop_free;
356 u_long pmap_upvop_maxfree;
357 u_long pmap_mpvop_free;
358 u_long pmap_mpvop_maxfree;
359 
360 static void *pmap_pool_ualloc(struct pool *, int);
361 static void *pmap_pool_malloc(struct pool *, int);
362 
363 static void pmap_pool_ufree(struct pool *, void *);
364 static void pmap_pool_mfree(struct pool *, void *);
365 
366 static struct pool_allocator pmap_pool_mallocator = {
367 	.pa_alloc = pmap_pool_malloc,
368 	.pa_free = pmap_pool_mfree,
369 	.pa_pagesz = 0,
370 };
371 
372 static struct pool_allocator pmap_pool_uallocator = {
373 	.pa_alloc = pmap_pool_ualloc,
374 	.pa_free = pmap_pool_ufree,
375 	.pa_pagesz = 0,
376 };
377 
378 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
379 void pmap_pte_print(volatile struct pte *);
380 void pmap_pteg_check(void);
381 void pmap_pteg_dist(void);
382 void pmap_print_pte(pmap_t, vaddr_t);
383 void pmap_print_mmuregs(void);
384 #endif
385 
386 #if defined(DEBUG) || defined(PMAPCHECK)
387 #ifdef PMAPCHECK
388 int pmapcheck = 1;
389 #else
390 int pmapcheck = 0;
391 #endif
392 void pmap_pvo_verify(void);
393 static void pmap_pvo_check(const struct pvo_entry *);
394 #define	PMAP_PVO_CHECK(pvo)	 		\
395 	do {					\
396 		if (pmapcheck)			\
397 			pmap_pvo_check(pvo);	\
398 	} while (0)
399 #else
400 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
401 #endif
402 static int pmap_pte_insert(int, struct pte *);
403 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
404 	vaddr_t, paddr_t, register_t, int);
405 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
406 static void pmap_pvo_free(struct pvo_entry *);
407 static void pmap_pvo_free_list(struct pvo_head *);
408 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
409 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
410 static struct pvo_entry *pmap_pvo_reclaim(struct pmap *);
411 static void pvo_set_exec(struct pvo_entry *);
412 static void pvo_clear_exec(struct pvo_entry *);
413 
414 static void tlbia(void);
415 
416 static void pmap_release(pmap_t);
417 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
418 
419 static uint32_t pmap_pvo_reclaim_nextidx;
420 #ifdef DEBUG
421 static int pmap_pvo_reclaim_debugctr;
422 #endif
423 
424 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
425 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
426 
427 static int pmap_initialized;
428 
429 #if defined(DEBUG) || defined(PMAPDEBUG)
430 #define	PMAPDEBUG_BOOT		0x0001
431 #define	PMAPDEBUG_PTE		0x0002
432 #define	PMAPDEBUG_EXEC		0x0008
433 #define	PMAPDEBUG_PVOENTER	0x0010
434 #define	PMAPDEBUG_PVOREMOVE	0x0020
435 #define	PMAPDEBUG_ACTIVATE	0x0100
436 #define	PMAPDEBUG_CREATE	0x0200
437 #define	PMAPDEBUG_ENTER		0x1000
438 #define	PMAPDEBUG_KENTER	0x2000
439 #define	PMAPDEBUG_KREMOVE	0x4000
440 #define	PMAPDEBUG_REMOVE	0x8000
441 
442 unsigned int pmapdebug = 0;
443 
444 # define DPRINTF(x)		printf x
445 # define DPRINTFN(n, x)		if (pmapdebug & PMAPDEBUG_ ## n) printf x
446 #else
447 # define DPRINTF(x)
448 # define DPRINTFN(n, x)
449 #endif
450 
451 
452 #ifdef PMAPCOUNTERS
453 /*
454  * From pmap_subr.c
455  */
456 extern struct evcnt pmap_evcnt_mappings;
457 extern struct evcnt pmap_evcnt_unmappings;
458 
459 extern struct evcnt pmap_evcnt_kernel_mappings;
460 extern struct evcnt pmap_evcnt_kernel_unmappings;
461 
462 extern struct evcnt pmap_evcnt_mappings_replaced;
463 
464 extern struct evcnt pmap_evcnt_exec_mappings;
465 extern struct evcnt pmap_evcnt_exec_cached;
466 
467 extern struct evcnt pmap_evcnt_exec_synced;
468 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
469 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
470 
471 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
472 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
473 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
474 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
475 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
476 
477 extern struct evcnt pmap_evcnt_updates;
478 extern struct evcnt pmap_evcnt_collects;
479 extern struct evcnt pmap_evcnt_copies;
480 
481 extern struct evcnt pmap_evcnt_ptes_spilled;
482 extern struct evcnt pmap_evcnt_ptes_unspilled;
483 extern struct evcnt pmap_evcnt_ptes_evicted;
484 
485 extern struct evcnt pmap_evcnt_ptes_primary[8];
486 extern struct evcnt pmap_evcnt_ptes_secondary[8];
487 extern struct evcnt pmap_evcnt_ptes_removed;
488 extern struct evcnt pmap_evcnt_ptes_changed;
489 extern struct evcnt pmap_evcnt_pvos_reclaimed;
490 extern struct evcnt pmap_evcnt_pvos_failed;
491 
492 extern struct evcnt pmap_evcnt_zeroed_pages;
493 extern struct evcnt pmap_evcnt_copied_pages;
494 extern struct evcnt pmap_evcnt_idlezeroed_pages;
495 
496 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
497 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
498 #else
499 #define	PMAPCOUNT(ev)	((void) 0)
500 #define	PMAPCOUNT2(ev)	((void) 0)
501 #endif
502 
503 #define	TLBIE(va)	__asm volatile("tlbie %0" :: "r"(va))
504 
505 /* XXXSL: this needs to be moved to assembler */
506 #define	TLBIEL(va)	__asm __volatile("tlbie %0" :: "r"(va))
507 
508 #define	TLBSYNC()	__asm volatile("tlbsync")
509 #define	SYNC()		__asm volatile("sync")
510 #define	EIEIO()		__asm volatile("eieio")
511 #define	DCBST(va)	__asm __volatile("dcbst 0,%0" :: "r"(va))
512 #define	MFMSR()		mfmsr()
513 #define	MTMSR(psl)	mtmsr(psl)
514 #define	MFPVR()		mfpvr()
515 #define	MFSRIN(va)	mfsrin(va)
516 #define	MFTB()		mfrtcltbl()
517 
518 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
519 static inline register_t
520 mfsrin(vaddr_t va)
521 {
522 	register_t sr;
523 	__asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
524 	return sr;
525 }
526 #endif	/* PMAP_OEA*/
527 
528 #if defined (PMAP_OEA64_BRIDGE)
529 extern void mfmsr64 (register64_t *result);
530 #endif /* PMAP_OEA64_BRIDGE */
531 
532 #define	PMAP_LOCK()		KERNEL_LOCK(1, NULL)
533 #define	PMAP_UNLOCK()		KERNEL_UNLOCK_ONE(NULL)
534 
535 static inline register_t
536 pmap_interrupts_off(void)
537 {
538 	register_t msr = MFMSR();
539 	if (msr & PSL_EE)
540 		MTMSR(msr & ~PSL_EE);
541 	return msr;
542 }
543 
544 static void
545 pmap_interrupts_restore(register_t msr)
546 {
547 	if (msr & PSL_EE)
548 		MTMSR(msr);
549 }
550 
551 static inline u_int32_t
552 mfrtcltbl(void)
553 {
554 #ifdef PPC_OEA601
555 	if ((MFPVR() >> 16) == MPC601)
556 		return (mfrtcl() >> 7);
557 	else
558 #endif
559 		return (mftbl());
560 }
561 
562 /*
563  * These small routines may have to be replaced,
564  * if/when we support processors other that the 604.
565  */
566 
567 void
568 tlbia(void)
569 {
570 	char *i;
571 
572 	SYNC();
573 #if defined(PMAP_OEA)
574 	/*
575 	 * Why not use "tlbia"?  Because not all processors implement it.
576 	 *
577 	 * This needs to be a per-CPU callback to do the appropriate thing
578 	 * for the CPU. XXX
579 	 */
580 	for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
581 		TLBIE(i);
582 		EIEIO();
583 		SYNC();
584 	}
585 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
586 	/* This is specifically for the 970, 970UM v1.6 pp. 140. */
587 	for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
588 		TLBIEL(i);
589 		EIEIO();
590 		SYNC();
591 	}
592 #endif
593 	TLBSYNC();
594 	SYNC();
595 }
596 
597 static inline register_t
598 va_to_vsid(const struct pmap *pm, vaddr_t addr)
599 {
600 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
601 	return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT;
602 #else /* PMAP_OEA64 */
603 #if 0
604 	const struct ste *ste;
605 	register_t hash;
606 	int i;
607 
608 	hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH;
609 
610 	/*
611 	 * Try the primary group first
612 	 */
613 	ste = pm->pm_stes[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 ste;
618 	}
619 
620 	/*
621 	 * Then the secondary group.
622 	 */
623 	ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes;
624 	for (i = 0; i < 8; i++, ste++) {
625 		if (ste->ste_hi & STE_V) &&
626 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
627 			return addr;
628 	}
629 
630 	return NULL;
631 #else
632 	/*
633 	 * Rather than searching the STE groups for the VSID, we know
634 	 * how we generate that from the ESID and so do that.
635 	 */
636 	return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
637 #endif
638 #endif /* PMAP_OEA */
639 }
640 
641 static inline register_t
642 va_to_pteg(const struct pmap *pm, vaddr_t addr)
643 {
644 	register_t hash;
645 
646 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
647 	return hash & pmap_pteg_mask;
648 }
649 
650 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
651 /*
652  * Given a PTE in the page table, calculate the VADDR that hashes to it.
653  * The only bit of magic is that the top 4 bits of the address doesn't
654  * technically exist in the PTE.  But we know we reserved 4 bits of the
655  * VSID for it so that's how we get it.
656  */
657 static vaddr_t
658 pmap_pte_to_va(volatile const struct pte *pt)
659 {
660 	vaddr_t va;
661 	uintptr_t ptaddr = (uintptr_t) pt;
662 
663 	if (pt->pte_hi & PTE_HID)
664 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
665 
666 	/* PPC Bits 10-19  PPC64 Bits 42-51 */
667 #if defined(PMAP_OEA)
668 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
669 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
670 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
671 #endif
672 	va <<= ADDR_PIDX_SHFT;
673 
674 	/* PPC Bits 4-9  PPC64 Bits 36-41 */
675 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
676 
677 #if defined(PMAP_OEA64)
678 	/* PPC63 Bits 0-35 */
679 	/* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
680 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
681 	/* PPC Bits 0-3 */
682 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
683 #endif
684 
685 	return va;
686 }
687 #endif
688 
689 static inline struct pvo_head *
690 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
691 {
692 	struct vm_page *pg;
693 	struct vm_page_md *md;
694 
695 	pg = PHYS_TO_VM_PAGE(pa);
696 	if (pg_p != NULL)
697 		*pg_p = pg;
698 	if (pg == NULL)
699 		return &pmap_pvo_unmanaged;
700 	md = VM_PAGE_TO_MD(pg);
701 	return &md->mdpg_pvoh;
702 }
703 
704 static inline struct pvo_head *
705 vm_page_to_pvoh(struct vm_page *pg)
706 {
707 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
708 
709 	return &md->mdpg_pvoh;
710 }
711 
712 
713 static inline void
714 pmap_attr_clear(struct vm_page *pg, int ptebit)
715 {
716 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
717 
718 	md->mdpg_attrs &= ~ptebit;
719 }
720 
721 static inline int
722 pmap_attr_fetch(struct vm_page *pg)
723 {
724 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
725 
726 	return md->mdpg_attrs;
727 }
728 
729 static inline void
730 pmap_attr_save(struct vm_page *pg, int ptebit)
731 {
732 	struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
733 
734 	md->mdpg_attrs |= ptebit;
735 }
736 
737 static inline int
738 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
739 {
740 	if (pt->pte_hi == pvo_pt->pte_hi
741 #if 0
742 	    && ((pt->pte_lo ^ pvo_pt->pte_lo) &
743 	        ~(PTE_REF|PTE_CHG)) == 0
744 #endif
745 	    )
746 		return 1;
747 	return 0;
748 }
749 
750 static inline void
751 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
752 {
753 	/*
754 	 * Construct the PTE.  Default to IMB initially.  Valid bit
755 	 * only gets set when the real pte is set in memory.
756 	 *
757 	 * Note: Don't set the valid bit for correct operation of tlb update.
758 	 */
759 #if defined(PMAP_OEA)
760 	pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
761 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
762 	pt->pte_lo = pte_lo;
763 #elif defined (PMAP_OEA64_BRIDGE)
764 	pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT)
765 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
766 	pt->pte_lo = (u_int64_t) pte_lo;
767 #elif defined (PMAP_OEA64)
768 #error PMAP_OEA64 not supported
769 #endif /* PMAP_OEA */
770 }
771 
772 static inline void
773 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
774 {
775 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
776 }
777 
778 static inline void
779 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
780 {
781 	/*
782 	 * As shown in Section 7.6.3.2.3
783 	 */
784 	pt->pte_lo &= ~ptebit;
785 	TLBIE(va);
786 	SYNC();
787 	EIEIO();
788 	TLBSYNC();
789 	SYNC();
790 #ifdef MULTIPROCESSOR
791 	DCBST(pt);
792 #endif
793 }
794 
795 static inline void
796 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
797 {
798 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
799 	if (pvo_pt->pte_hi & PTE_VALID)
800 		panic("pte_set: setting an already valid pte %p", pvo_pt);
801 #endif
802 	pvo_pt->pte_hi |= PTE_VALID;
803 
804 	/*
805 	 * Update the PTE as defined in section 7.6.3.1
806 	 * Note that the REF/CHG bits are from pvo_pt and thus should
807 	 * have been saved so this routine can restore them (if desired).
808 	 */
809 	pt->pte_lo = pvo_pt->pte_lo;
810 	EIEIO();
811 	pt->pte_hi = pvo_pt->pte_hi;
812 	TLBSYNC();
813 	SYNC();
814 #ifdef MULTIPROCESSOR
815 	DCBST(pt);
816 #endif
817 	pmap_pte_valid++;
818 }
819 
820 static inline void
821 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
822 {
823 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
824 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
825 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
826 	if ((pt->pte_hi & PTE_VALID) == 0)
827 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
828 #endif
829 
830 	pvo_pt->pte_hi &= ~PTE_VALID;
831 	/*
832 	 * Force the ref & chg bits back into the PTEs.
833 	 */
834 	SYNC();
835 	/*
836 	 * Invalidate the pte ... (Section 7.6.3.3)
837 	 */
838 	pt->pte_hi &= ~PTE_VALID;
839 	SYNC();
840 	TLBIE(va);
841 	SYNC();
842 	EIEIO();
843 	TLBSYNC();
844 	SYNC();
845 	/*
846 	 * Save the ref & chg bits ...
847 	 */
848 	pmap_pte_synch(pt, pvo_pt);
849 	pmap_pte_valid--;
850 }
851 
852 static inline void
853 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
854 {
855 	/*
856 	 * Invalidate the PTE
857 	 */
858 	pmap_pte_unset(pt, pvo_pt, va);
859 	pmap_pte_set(pt, pvo_pt);
860 }
861 
862 /*
863  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
864  * (either primary or secondary location).
865  *
866  * Note: both the destination and source PTEs must not have PTE_VALID set.
867  */
868 
869 static int
870 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
871 {
872 	volatile struct pte *pt;
873 	int i;
874 
875 #if defined(DEBUG)
876 	DPRINTFN(PTE, ("pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
877 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo));
878 #endif
879 	/*
880 	 * First try primary hash.
881 	 */
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 
890 	/*
891 	 * Now try secondary hash.
892 	 */
893 	ptegidx ^= pmap_pteg_mask;
894 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
895 		if ((pt->pte_hi & PTE_VALID) == 0) {
896 			pvo_pt->pte_hi |= PTE_HID;
897 			pmap_pte_set(pt, pvo_pt);
898 			return i;
899 		}
900 	}
901 	return -1;
902 }
903 
904 /*
905  * Spill handler.
906  *
907  * Tries to spill a page table entry from the overflow area.
908  * This runs in either real mode (if dealing with a exception spill)
909  * or virtual mode when dealing with manually spilling one of the
910  * kernel's pte entries.  In either case, interrupts are already
911  * disabled.
912  */
913 
914 int
915 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec)
916 {
917 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
918 	struct pvo_entry *pvo;
919 	/* XXX: gcc -- vpvoh is always set at either *1* or *2* */
920 	struct pvo_tqhead *pvoh, *vpvoh = NULL;
921 	int ptegidx, i, j;
922 	volatile struct pteg *pteg;
923 	volatile struct pte *pt;
924 
925 	PMAP_LOCK();
926 
927 	ptegidx = va_to_pteg(pm, addr);
928 
929 	/*
930 	 * Have to substitute some entry. Use the primary hash for this.
931 	 * Use low bits of timebase as random generator.  Make sure we are
932 	 * not picking a kernel pte for replacement.
933 	 */
934 	pteg = &pmap_pteg_table[ptegidx];
935 	i = MFTB() & 7;
936 	for (j = 0; j < 8; j++) {
937 		pt = &pteg->pt[i];
938 		if ((pt->pte_hi & PTE_VALID) == 0)
939 			break;
940 		if (VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT)
941 				< PHYSMAP_VSIDBITS)
942 			break;
943 		i = (i + 1) & 7;
944 	}
945 	KASSERT(j < 8);
946 
947 	source_pvo = NULL;
948 	victim_pvo = NULL;
949 	pvoh = &pmap_pvo_table[ptegidx];
950 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
951 
952 		/*
953 		 * We need to find pvo entry for this address...
954 		 */
955 		PMAP_PVO_CHECK(pvo);		/* sanity check */
956 
957 		/*
958 		 * If we haven't found the source and we come to a PVO with
959 		 * a valid PTE, then we know we can't find it because all
960 		 * evicted PVOs always are first in the list.
961 		 */
962 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
963 			break;
964 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
965 		    addr == PVO_VADDR(pvo)) {
966 
967 			/*
968 			 * Now we have found the entry to be spilled into the
969 			 * pteg.  Attempt to insert it into the page table.
970 			 */
971 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
972 			if (j >= 0) {
973 				PVO_PTEGIDX_SET(pvo, j);
974 				PMAP_PVO_CHECK(pvo);	/* sanity check */
975 				PVO_WHERE(pvo, SPILL_INSERT);
976 				pvo->pvo_pmap->pm_evictions--;
977 				PMAPCOUNT(ptes_spilled);
978 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
979 				    ? pmap_evcnt_ptes_secondary
980 				    : pmap_evcnt_ptes_primary)[j]);
981 
982 				/*
983 				 * Since we keep the evicted entries at the
984 				 * from of the PVO list, we need move this
985 				 * (now resident) PVO after the evicted
986 				 * entries.
987 				 */
988 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
989 
990 				/*
991 				 * If we don't have to move (either we were the
992 				 * last entry or the next entry was valid),
993 				 * don't change our position.  Otherwise
994 				 * move ourselves to the tail of the queue.
995 				 */
996 				if (next_pvo != NULL &&
997 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
998 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
999 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1000 				}
1001 				PMAP_UNLOCK();
1002 				return 1;
1003 			}
1004 			source_pvo = pvo;
1005 			if (exec && !PVO_EXECUTABLE_P(source_pvo)) {
1006 				return 0;
1007 			}
1008 			if (victim_pvo != NULL)
1009 				break;
1010 		}
1011 
1012 		/*
1013 		 * We also need the pvo entry of the victim we are replacing
1014 		 * so save the R & C bits of the PTE.
1015 		 */
1016 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
1017 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
1018 			vpvoh = pvoh;			/* *1* */
1019 			victim_pvo = pvo;
1020 			if (source_pvo != NULL)
1021 				break;
1022 		}
1023 	}
1024 
1025 	if (source_pvo == NULL) {
1026 		PMAPCOUNT(ptes_unspilled);
1027 		PMAP_UNLOCK();
1028 		return 0;
1029 	}
1030 
1031 	if (victim_pvo == NULL) {
1032 		if ((pt->pte_hi & PTE_HID) == 0)
1033 			panic("pmap_pte_spill: victim p-pte (%p) has "
1034 			    "no pvo entry!", pt);
1035 
1036 		/*
1037 		 * If this is a secondary PTE, we need to search
1038 		 * its primary pvo bucket for the matching PVO.
1039 		 */
1040 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */
1041 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1042 			PMAP_PVO_CHECK(pvo);		/* sanity check */
1043 
1044 			/*
1045 			 * We also need the pvo entry of the victim we are
1046 			 * replacing so save the R & C bits of the PTE.
1047 			 */
1048 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1049 				victim_pvo = pvo;
1050 				break;
1051 			}
1052 		}
1053 		if (victim_pvo == NULL)
1054 			panic("pmap_pte_spill: victim s-pte (%p) has "
1055 			    "no pvo entry!", pt);
1056 	}
1057 
1058 	/*
1059 	 * The victim should be not be a kernel PVO/PTE entry.
1060 	 */
1061 	KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1062 	KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1063 	KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1064 
1065 	/*
1066 	 * We are invalidating the TLB entry for the EA for the
1067 	 * we are replacing even though its valid; If we don't
1068 	 * we lose any ref/chg bit changes contained in the TLB
1069 	 * entry.
1070 	 */
1071 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1072 
1073 	/*
1074 	 * To enforce the PVO list ordering constraint that all
1075 	 * evicted entries should come before all valid entries,
1076 	 * move the source PVO to the tail of its list and the
1077 	 * victim PVO to the head of its list (which might not be
1078 	 * the same list, if the victim was using the secondary hash).
1079 	 */
1080 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
1081 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
1082 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1083 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1084 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1085 	pmap_pte_set(pt, &source_pvo->pvo_pte);
1086 	victim_pvo->pvo_pmap->pm_evictions++;
1087 	source_pvo->pvo_pmap->pm_evictions--;
1088 	PVO_WHERE(victim_pvo, SPILL_UNSET);
1089 	PVO_WHERE(source_pvo, SPILL_SET);
1090 
1091 	PVO_PTEGIDX_CLR(victim_pvo);
1092 	PVO_PTEGIDX_SET(source_pvo, i);
1093 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1094 	PMAPCOUNT(ptes_spilled);
1095 	PMAPCOUNT(ptes_evicted);
1096 	PMAPCOUNT(ptes_removed);
1097 
1098 	PMAP_PVO_CHECK(victim_pvo);
1099 	PMAP_PVO_CHECK(source_pvo);
1100 
1101 	PMAP_UNLOCK();
1102 	return 1;
1103 }
1104 
1105 /*
1106  * Restrict given range to physical memory
1107  */
1108 void
1109 pmap_real_memory(paddr_t *start, psize_t *size)
1110 {
1111 	struct mem_region *mp;
1112 
1113 	for (mp = mem; mp->size; mp++) {
1114 		if (*start + *size > mp->start
1115 		    && *start < mp->start + mp->size) {
1116 			if (*start < mp->start) {
1117 				*size -= mp->start - *start;
1118 				*start = mp->start;
1119 			}
1120 			if (*start + *size > mp->start + mp->size)
1121 				*size = mp->start + mp->size - *start;
1122 			return;
1123 		}
1124 	}
1125 	*size = 0;
1126 }
1127 
1128 /*
1129  * Initialize anything else for pmap handling.
1130  * Called during vm_init().
1131  */
1132 void
1133 pmap_init(void)
1134 {
1135 	pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry),
1136 	    sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl",
1137 	    &pmap_pool_mallocator, IPL_NONE);
1138 
1139 	pool_setlowat(&pmap_mpvo_pool, 1008);
1140 
1141 	pmap_initialized = 1;
1142 
1143 }
1144 
1145 /*
1146  * How much virtual space does the kernel get?
1147  */
1148 void
1149 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1150 {
1151 	/*
1152 	 * For now, reserve one segment (minus some overhead) for kernel
1153 	 * virtual memory
1154 	 */
1155 	*start = VM_MIN_KERNEL_ADDRESS;
1156 	*end = VM_MAX_KERNEL_ADDRESS;
1157 }
1158 
1159 /*
1160  * Allocate, initialize, and return a new physical map.
1161  */
1162 pmap_t
1163 pmap_create(void)
1164 {
1165 	pmap_t pm;
1166 
1167 	pm = pool_get(&pmap_pool, PR_WAITOK);
1168 	memset((void *)pm, 0, sizeof *pm);
1169 	pmap_pinit(pm);
1170 
1171 	DPRINTFN(CREATE,("pmap_create: pm %p:\n"
1172 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1173 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1174 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1175 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1176 	    pm,
1177 	    pm->pm_sr[0], pm->pm_sr[1],
1178 	    pm->pm_sr[2], pm->pm_sr[3],
1179 	    pm->pm_sr[4], pm->pm_sr[5],
1180 	    pm->pm_sr[6], pm->pm_sr[7],
1181 	    pm->pm_sr[8], pm->pm_sr[9],
1182 	    pm->pm_sr[10], pm->pm_sr[11],
1183 	    pm->pm_sr[12], pm->pm_sr[13],
1184 	    pm->pm_sr[14], pm->pm_sr[15]));
1185 	return pm;
1186 }
1187 
1188 /*
1189  * Initialize a preallocated and zeroed pmap structure.
1190  */
1191 void
1192 pmap_pinit(pmap_t pm)
1193 {
1194 	register_t entropy = MFTB();
1195 	register_t mask;
1196 	int i;
1197 
1198 	/*
1199 	 * Allocate some segment registers for this pmap.
1200 	 */
1201 	pm->pm_refs = 1;
1202 	PMAP_LOCK();
1203 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1204 		static register_t pmap_vsidcontext;
1205 		register_t hash;
1206 		unsigned int n;
1207 
1208 		/* Create a new value by multiplying by a prime adding in
1209 		 * entropy from the timebase register.  This is to make the
1210 		 * VSID more random so that the PT Hash function collides
1211 		 * less often. (note that the prime causes gcc to do shifts
1212 		 * instead of a multiply)
1213 		 */
1214 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1215 		hash = pmap_vsidcontext & (NPMAPS - 1);
1216 		if (hash == 0) {		/* 0 is special, avoid it */
1217 			entropy += 0xbadf00d;
1218 			continue;
1219 		}
1220 		n = hash >> 5;
1221 		mask = 1L << (hash & (VSID_NBPW-1));
1222 		hash = pmap_vsidcontext;
1223 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1224 			/* anything free in this bucket? */
1225 			if (~pmap_vsid_bitmap[n] == 0) {
1226 				entropy = hash ^ (hash >> 16);
1227 				continue;
1228 			}
1229 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1230 			mask = 1L << i;
1231 			hash &= ~(VSID_NBPW-1);
1232 			hash |= i;
1233 		}
1234 		hash &= PTE_VSID >> PTE_VSID_SHFT;
1235 		pmap_vsid_bitmap[n] |= mask;
1236 		pm->pm_vsid = hash;
1237 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1238 		for (i = 0; i < 16; i++)
1239 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1240 			    SR_NOEXEC;
1241 #endif
1242 		PMAP_UNLOCK();
1243 		return;
1244 	}
1245 	PMAP_UNLOCK();
1246 	panic("pmap_pinit: out of segments");
1247 }
1248 
1249 /*
1250  * Add a reference to the given pmap.
1251  */
1252 void
1253 pmap_reference(pmap_t pm)
1254 {
1255 	atomic_inc_uint(&pm->pm_refs);
1256 }
1257 
1258 /*
1259  * Retire the given pmap from service.
1260  * Should only be called if the map contains no valid mappings.
1261  */
1262 void
1263 pmap_destroy(pmap_t pm)
1264 {
1265 	if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1266 		pmap_release(pm);
1267 		pool_put(&pmap_pool, pm);
1268 	}
1269 }
1270 
1271 /*
1272  * Release any resources held by the given physical map.
1273  * Called when a pmap initialized by pmap_pinit is being released.
1274  */
1275 void
1276 pmap_release(pmap_t pm)
1277 {
1278 	int idx, mask;
1279 
1280 	KASSERT(pm->pm_stats.resident_count == 0);
1281 	KASSERT(pm->pm_stats.wired_count == 0);
1282 
1283 	PMAP_LOCK();
1284 	if (pm->pm_sr[0] == 0)
1285 		panic("pmap_release");
1286 	idx = pm->pm_vsid & (NPMAPS-1);
1287 	mask = 1 << (idx % VSID_NBPW);
1288 	idx /= VSID_NBPW;
1289 
1290 	KASSERT(pmap_vsid_bitmap[idx] & mask);
1291 	pmap_vsid_bitmap[idx] &= ~mask;
1292 	PMAP_UNLOCK();
1293 }
1294 
1295 /*
1296  * Copy the range specified by src_addr/len
1297  * from the source map to the range dst_addr/len
1298  * in the destination map.
1299  *
1300  * This routine is only advisory and need not do anything.
1301  */
1302 void
1303 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1304 	vsize_t len, vaddr_t src_addr)
1305 {
1306 	PMAPCOUNT(copies);
1307 }
1308 
1309 /*
1310  * Require that all active physical maps contain no
1311  * incorrect entries NOW.
1312  */
1313 void
1314 pmap_update(struct pmap *pmap)
1315 {
1316 	PMAPCOUNT(updates);
1317 	TLBSYNC();
1318 }
1319 
1320 static inline int
1321 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1322 {
1323 	int pteidx;
1324 	/*
1325 	 * We can find the actual pte entry without searching by
1326 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1327 	 * and by noticing the HID bit.
1328 	 */
1329 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1330 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1331 		pteidx ^= pmap_pteg_mask * 8;
1332 	return pteidx;
1333 }
1334 
1335 volatile struct pte *
1336 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1337 {
1338 	volatile struct pte *pt;
1339 
1340 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1341 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1342 		return NULL;
1343 #endif
1344 
1345 	/*
1346 	 * If we haven't been supplied the ptegidx, calculate it.
1347 	 */
1348 	if (pteidx == -1) {
1349 		int ptegidx;
1350 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1351 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1352 	}
1353 
1354 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1355 
1356 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1357 	return pt;
1358 #else
1359 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1360 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1361 		    "pvo but no valid pte index", pvo);
1362 	}
1363 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1364 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1365 		    "pvo but no valid pte", pvo);
1366 	}
1367 
1368 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1369 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1370 #if defined(DEBUG) || defined(PMAPCHECK)
1371 			pmap_pte_print(pt);
1372 #endif
1373 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1374 			    "pmap_pteg_table %p but invalid in pvo",
1375 			    pvo, pt);
1376 		}
1377 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1378 #if defined(DEBUG) || defined(PMAPCHECK)
1379 			pmap_pte_print(pt);
1380 #endif
1381 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1382 			    "not match pte %p in pmap_pteg_table",
1383 			    pvo, pt);
1384 		}
1385 		return pt;
1386 	}
1387 
1388 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1389 #if defined(DEBUG) || defined(PMAPCHECK)
1390 		pmap_pte_print(pt);
1391 #endif
1392 		panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1393 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1394 	}
1395 	return NULL;
1396 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1397 }
1398 
1399 struct pvo_entry *
1400 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1401 {
1402 	struct pvo_entry *pvo;
1403 	int ptegidx;
1404 
1405 	va &= ~ADDR_POFF;
1406 	ptegidx = va_to_pteg(pm, va);
1407 
1408 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1409 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1410 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1411 			panic("pmap_pvo_find_va: invalid pvo %p on "
1412 			    "list %#x (%p)", pvo, ptegidx,
1413 			     &pmap_pvo_table[ptegidx]);
1414 #endif
1415 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1416 			if (pteidx_p)
1417 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1418 			return pvo;
1419 		}
1420 	}
1421 	if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH))
1422 		panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1423 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1424 	return NULL;
1425 }
1426 
1427 #if defined(DEBUG) || defined(PMAPCHECK)
1428 void
1429 pmap_pvo_check(const struct pvo_entry *pvo)
1430 {
1431 	struct pvo_head *pvo_head;
1432 	struct pvo_entry *pvo0;
1433 	volatile struct pte *pt;
1434 	int failed = 0;
1435 
1436 	PMAP_LOCK();
1437 
1438 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1439 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1440 
1441 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1442 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1443 		    pvo, pvo->pvo_pmap);
1444 		failed = 1;
1445 	}
1446 
1447 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1448 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1449 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1450 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1451 		failed = 1;
1452 	}
1453 
1454 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1455 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1456 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1457 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1458 		failed = 1;
1459 	}
1460 
1461 	if (PVO_MANAGED_P(pvo)) {
1462 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1463 	} else {
1464 		if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) {
1465 			printf("pmap_pvo_check: pvo %p: non kernel address "
1466 			    "on kernel unmanaged list\n", pvo);
1467 			failed = 1;
1468 		}
1469 		pvo_head = &pmap_pvo_kunmanaged;
1470 	}
1471 	LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1472 		if (pvo0 == pvo)
1473 			break;
1474 	}
1475 	if (pvo0 == NULL) {
1476 		printf("pmap_pvo_check: pvo %p: not present "
1477 		    "on its vlist head %p\n", pvo, pvo_head);
1478 		failed = 1;
1479 	}
1480 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1481 		printf("pmap_pvo_check: pvo %p: not present "
1482 		    "on its olist head\n", pvo);
1483 		failed = 1;
1484 	}
1485 	pt = pmap_pvo_to_pte(pvo, -1);
1486 	if (pt == NULL) {
1487 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1488 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1489 			    "no PTE\n", pvo);
1490 			failed = 1;
1491 		}
1492 	} else {
1493 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1494 		    (uintptr_t) pt >=
1495 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1496 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1497 			    "pteg table\n", pvo, pt);
1498 			failed = 1;
1499 		}
1500 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1501 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1502 			    "no PTE\n", pvo);
1503 			failed = 1;
1504 		}
1505 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1506 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1507 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1508 			    pvo->pvo_pte.pte_hi,
1509 			    pt->pte_hi);
1510 			failed = 1;
1511 		}
1512 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1513 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1514 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1515 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1516 			    (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1517 			    (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1518 			failed = 1;
1519 		}
1520 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1521 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1522 			    " doesn't not match PVO's VA %#" _PRIxva "\n",
1523 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1524 			failed = 1;
1525 		}
1526 		if (failed)
1527 			pmap_pte_print(pt);
1528 	}
1529 	if (failed)
1530 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1531 		    pvo->pvo_pmap);
1532 
1533 	PMAP_UNLOCK();
1534 }
1535 #endif /* DEBUG || PMAPCHECK */
1536 
1537 /*
1538  * Search the PVO table looking for a non-wired entry.
1539  * If we find one, remove it and return it.
1540  */
1541 
1542 struct pvo_entry *
1543 pmap_pvo_reclaim(struct pmap *pm)
1544 {
1545 	struct pvo_tqhead *pvoh;
1546 	struct pvo_entry *pvo;
1547 	uint32_t idx, endidx;
1548 
1549 	endidx = pmap_pvo_reclaim_nextidx;
1550 	for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1551 	     idx = (idx + 1) & pmap_pteg_mask) {
1552 		pvoh = &pmap_pvo_table[idx];
1553 		TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1554 			if (!PVO_WIRED_P(pvo)) {
1555 				pmap_pvo_remove(pvo, -1, NULL);
1556 				pmap_pvo_reclaim_nextidx = idx;
1557 				PMAPCOUNT(pvos_reclaimed);
1558 				return pvo;
1559 			}
1560 		}
1561 	}
1562 	return NULL;
1563 }
1564 
1565 /*
1566  * This returns whether this is the first mapping of a page.
1567  */
1568 int
1569 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1570 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1571 {
1572 	struct pvo_entry *pvo;
1573 	struct pvo_tqhead *pvoh;
1574 	register_t msr;
1575 	int ptegidx;
1576 	int i;
1577 	int poolflags = PR_NOWAIT;
1578 
1579 	/*
1580 	 * Compute the PTE Group index.
1581 	 */
1582 	va &= ~ADDR_POFF;
1583 	ptegidx = va_to_pteg(pm, va);
1584 
1585 	msr = pmap_interrupts_off();
1586 
1587 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1588 	if (pmap_pvo_remove_depth > 0)
1589 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1590 	if (++pmap_pvo_enter_depth > 1)
1591 		panic("pmap_pvo_enter: called recursively!");
1592 #endif
1593 
1594 	/*
1595 	 * Remove any existing mapping for this page.  Reuse the
1596 	 * pvo entry if there a mapping.
1597 	 */
1598 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1599 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1600 #ifdef DEBUG
1601 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1602 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1603 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1604 			   va < VM_MIN_KERNEL_ADDRESS) {
1605 				printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1606 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1607 				printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1608 				    pvo->pvo_pte.pte_hi,
1609 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1610 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1611 #ifdef DDBX
1612 				Debugger();
1613 #endif
1614 			}
1615 #endif
1616 			PMAPCOUNT(mappings_replaced);
1617 			pmap_pvo_remove(pvo, -1, NULL);
1618 			break;
1619 		}
1620 	}
1621 
1622 	/*
1623 	 * If we aren't overwriting an mapping, try to allocate
1624 	 */
1625 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1626 	--pmap_pvo_enter_depth;
1627 #endif
1628 	pmap_interrupts_restore(msr);
1629 	if (pvo) {
1630 		pmap_pvo_free(pvo);
1631 	}
1632 	pvo = pool_get(pl, poolflags);
1633 
1634 #ifdef DEBUG
1635 	/*
1636 	 * Exercise pmap_pvo_reclaim() a little.
1637 	 */
1638 	if (pvo && (flags & PMAP_CANFAIL) != 0 &&
1639 	    pmap_pvo_reclaim_debugctr++ > 0x1000 &&
1640 	    (pmap_pvo_reclaim_debugctr & 0xff) == 0) {
1641 		pool_put(pl, pvo);
1642 		pvo = NULL;
1643 	}
1644 #endif
1645 
1646 	msr = pmap_interrupts_off();
1647 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1648 	++pmap_pvo_enter_depth;
1649 #endif
1650 	if (pvo == NULL) {
1651 		pvo = pmap_pvo_reclaim(pm);
1652 		if (pvo == NULL) {
1653 			if ((flags & PMAP_CANFAIL) == 0)
1654 				panic("pmap_pvo_enter: failed");
1655 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1656 			pmap_pvo_enter_depth--;
1657 #endif
1658 			PMAPCOUNT(pvos_failed);
1659 			pmap_interrupts_restore(msr);
1660 			return ENOMEM;
1661 		}
1662 	}
1663 
1664 	pvo->pvo_vaddr = va;
1665 	pvo->pvo_pmap = pm;
1666 	pvo->pvo_vaddr &= ~ADDR_POFF;
1667 	if (flags & VM_PROT_EXECUTE) {
1668 		PMAPCOUNT(exec_mappings);
1669 		pvo_set_exec(pvo);
1670 	}
1671 	if (flags & PMAP_WIRED)
1672 		pvo->pvo_vaddr |= PVO_WIRED;
1673 	if (pvo_head != &pmap_pvo_kunmanaged) {
1674 		pvo->pvo_vaddr |= PVO_MANAGED;
1675 		PMAPCOUNT(mappings);
1676 	} else {
1677 		PMAPCOUNT(kernel_mappings);
1678 	}
1679 	pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1680 
1681 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1682 	if (PVO_WIRED_P(pvo))
1683 		pvo->pvo_pmap->pm_stats.wired_count++;
1684 	pvo->pvo_pmap->pm_stats.resident_count++;
1685 #if defined(DEBUG)
1686 /*	if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */
1687 		DPRINTFN(PVOENTER,
1688 		    ("pmap_pvo_enter: pvo %p: pm %p va %#" _PRIxva " pa %#" _PRIxpa "\n",
1689 		    pvo, pm, va, pa));
1690 #endif
1691 
1692 	/*
1693 	 * We hope this succeeds but it isn't required.
1694 	 */
1695 	pvoh = &pmap_pvo_table[ptegidx];
1696 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1697 	if (i >= 0) {
1698 		PVO_PTEGIDX_SET(pvo, i);
1699 		PVO_WHERE(pvo, ENTER_INSERT);
1700 		PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1701 		    ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1702 		TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1703 
1704 	} else {
1705 		/*
1706 		 * Since we didn't have room for this entry (which makes it
1707 		 * and evicted entry), place it at the head of the list.
1708 		 */
1709 		TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1710 		PMAPCOUNT(ptes_evicted);
1711 		pm->pm_evictions++;
1712 		/*
1713 		 * If this is a kernel page, make sure it's active.
1714 		 */
1715 		if (pm == pmap_kernel()) {
1716 			i = pmap_pte_spill(pm, va, false);
1717 			KASSERT(i);
1718 		}
1719 	}
1720 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1721 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1722 	pmap_pvo_enter_depth--;
1723 #endif
1724 	pmap_interrupts_restore(msr);
1725 	return 0;
1726 }
1727 
1728 static void
1729 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol)
1730 {
1731 	volatile struct pte *pt;
1732 	int ptegidx;
1733 
1734 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1735 	if (++pmap_pvo_remove_depth > 1)
1736 		panic("pmap_pvo_remove: called recursively!");
1737 #endif
1738 
1739 	/*
1740 	 * If we haven't been supplied the ptegidx, calculate it.
1741 	 */
1742 	if (pteidx == -1) {
1743 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1744 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1745 	} else {
1746 		ptegidx = pteidx >> 3;
1747 		if (pvo->pvo_pte.pte_hi & PTE_HID)
1748 			ptegidx ^= pmap_pteg_mask;
1749 	}
1750 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1751 
1752 	/*
1753 	 * If there is an active pte entry, we need to deactivate it
1754 	 * (and save the ref & chg bits).
1755 	 */
1756 	pt = pmap_pvo_to_pte(pvo, pteidx);
1757 	if (pt != NULL) {
1758 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1759 		PVO_WHERE(pvo, REMOVE);
1760 		PVO_PTEGIDX_CLR(pvo);
1761 		PMAPCOUNT(ptes_removed);
1762 	} else {
1763 		KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1764 		pvo->pvo_pmap->pm_evictions--;
1765 	}
1766 
1767 	/*
1768 	 * Account for executable mappings.
1769 	 */
1770 	if (PVO_EXECUTABLE_P(pvo))
1771 		pvo_clear_exec(pvo);
1772 
1773 	/*
1774 	 * Update our statistics.
1775 	 */
1776 	pvo->pvo_pmap->pm_stats.resident_count--;
1777 	if (PVO_WIRED_P(pvo))
1778 		pvo->pvo_pmap->pm_stats.wired_count--;
1779 
1780 	/*
1781 	 * Save the REF/CHG bits into their cache if the page is managed.
1782 	 */
1783 	if (PVO_MANAGED_P(pvo)) {
1784 		register_t ptelo = pvo->pvo_pte.pte_lo;
1785 		struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1786 
1787 		if (pg != NULL) {
1788 			/*
1789 			 * If this page was changed and it is mapped exec,
1790 			 * invalidate it.
1791 			 */
1792 			if ((ptelo & PTE_CHG) &&
1793 			    (pmap_attr_fetch(pg) & PTE_EXEC)) {
1794 				struct pvo_head *pvoh = vm_page_to_pvoh(pg);
1795 				if (LIST_EMPTY(pvoh)) {
1796 					DPRINTFN(EXEC, ("[pmap_pvo_remove: "
1797 					    "%#" _PRIxpa ": clear-exec]\n",
1798 					    VM_PAGE_TO_PHYS(pg)));
1799 					pmap_attr_clear(pg, PTE_EXEC);
1800 					PMAPCOUNT(exec_uncached_pvo_remove);
1801 				} else {
1802 					DPRINTFN(EXEC, ("[pmap_pvo_remove: "
1803 					    "%#" _PRIxpa ": syncicache]\n",
1804 					    VM_PAGE_TO_PHYS(pg)));
1805 					pmap_syncicache(VM_PAGE_TO_PHYS(pg),
1806 					    PAGE_SIZE);
1807 					PMAPCOUNT(exec_synced_pvo_remove);
1808 				}
1809 			}
1810 
1811 			pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1812 		}
1813 		PMAPCOUNT(unmappings);
1814 	} else {
1815 		PMAPCOUNT(kernel_unmappings);
1816 	}
1817 
1818 	/*
1819 	 * Remove the PVO from its lists and return it to the pool.
1820 	 */
1821 	LIST_REMOVE(pvo, pvo_vlink);
1822 	TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1823 	if (pvol) {
1824 		LIST_INSERT_HEAD(pvol, pvo, pvo_vlink);
1825 	}
1826 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1827 	pmap_pvo_remove_depth--;
1828 #endif
1829 }
1830 
1831 void
1832 pmap_pvo_free(struct pvo_entry *pvo)
1833 {
1834 
1835 	pool_put(PVO_MANAGED_P(pvo) ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo);
1836 }
1837 
1838 void
1839 pmap_pvo_free_list(struct pvo_head *pvol)
1840 {
1841 	struct pvo_entry *pvo, *npvo;
1842 
1843 	for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) {
1844 		npvo = LIST_NEXT(pvo, pvo_vlink);
1845 		LIST_REMOVE(pvo, pvo_vlink);
1846 		pmap_pvo_free(pvo);
1847 	}
1848 }
1849 
1850 /*
1851  * Mark a mapping as executable.
1852  * If this is the first executable mapping in the segment,
1853  * clear the noexec flag.
1854  */
1855 static void
1856 pvo_set_exec(struct pvo_entry *pvo)
1857 {
1858 	struct pmap *pm = pvo->pvo_pmap;
1859 
1860 	if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) {
1861 		return;
1862 	}
1863 	pvo->pvo_vaddr |= PVO_EXECUTABLE;
1864 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1865 	{
1866 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1867 		if (pm->pm_exec[sr]++ == 0) {
1868 			pm->pm_sr[sr] &= ~SR_NOEXEC;
1869 		}
1870 	}
1871 #endif
1872 }
1873 
1874 /*
1875  * Mark a mapping as non-executable.
1876  * If this was the last executable mapping in the segment,
1877  * set the noexec flag.
1878  */
1879 static void
1880 pvo_clear_exec(struct pvo_entry *pvo)
1881 {
1882 	struct pmap *pm = pvo->pvo_pmap;
1883 
1884 	if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) {
1885 		return;
1886 	}
1887 	pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1888 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1889 	{
1890 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1891 		if (--pm->pm_exec[sr] == 0) {
1892 			pm->pm_sr[sr] |= SR_NOEXEC;
1893 		}
1894 	}
1895 #endif
1896 }
1897 
1898 /*
1899  * Insert physical page at pa into the given pmap at virtual address va.
1900  */
1901 int
1902 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1903 {
1904 	struct mem_region *mp;
1905 	struct pvo_head *pvo_head;
1906 	struct vm_page *pg;
1907 	struct pool *pl;
1908 	register_t pte_lo;
1909 	int error;
1910 	u_int pvo_flags;
1911 	u_int was_exec = 0;
1912 
1913 	PMAP_LOCK();
1914 
1915 	if (__predict_false(!pmap_initialized)) {
1916 		pvo_head = &pmap_pvo_kunmanaged;
1917 		pl = &pmap_upvo_pool;
1918 		pvo_flags = 0;
1919 		pg = NULL;
1920 		was_exec = PTE_EXEC;
1921 	} else {
1922 		pvo_head = pa_to_pvoh(pa, &pg);
1923 		pl = &pmap_mpvo_pool;
1924 		pvo_flags = PVO_MANAGED;
1925 	}
1926 
1927 	DPRINTFN(ENTER,
1928 	    ("pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1929 	    pm, va, pa, prot, flags));
1930 
1931 	/*
1932 	 * If this is a managed page, and it's the first reference to the
1933 	 * page clear the execness of the page.  Otherwise fetch the execness.
1934 	 */
1935 	if (pg != NULL)
1936 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1937 
1938 	DPRINTFN(ENTER, (" was_exec=%d", was_exec));
1939 
1940 	/*
1941 	 * Assume the page is cache inhibited and access is guarded unless
1942 	 * it's in our available memory array.  If it is in the memory array,
1943 	 * asssume it's in memory coherent memory.
1944 	 */
1945 	pte_lo = PTE_IG;
1946 	if ((flags & PMAP_MD_NOCACHE) == 0) {
1947 		for (mp = mem; mp->size; mp++) {
1948 			if (pa >= mp->start && pa < mp->start + mp->size) {
1949 				pte_lo = PTE_M;
1950 				break;
1951 			}
1952 		}
1953 	}
1954 
1955 	if (prot & VM_PROT_WRITE)
1956 		pte_lo |= PTE_BW;
1957 	else
1958 		pte_lo |= PTE_BR;
1959 
1960 	/*
1961 	 * If this was in response to a fault, "pre-fault" the PTE's
1962 	 * changed/referenced bit appropriately.
1963 	 */
1964 	if (flags & VM_PROT_WRITE)
1965 		pte_lo |= PTE_CHG;
1966 	if (flags & VM_PROT_ALL)
1967 		pte_lo |= PTE_REF;
1968 
1969 	/*
1970 	 * We need to know if this page can be executable
1971 	 */
1972 	flags |= (prot & VM_PROT_EXECUTE);
1973 
1974 	/*
1975 	 * Record mapping for later back-translation and pte spilling.
1976 	 * This will overwrite any existing mapping.
1977 	 */
1978 	error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags);
1979 
1980 	/*
1981 	 * Flush the real page from the instruction cache if this page is
1982 	 * mapped executable and cacheable and has not been flushed since
1983 	 * the last time it was modified.
1984 	 */
1985 	if (error == 0 &&
1986             (flags & VM_PROT_EXECUTE) &&
1987             (pte_lo & PTE_I) == 0 &&
1988 	    was_exec == 0) {
1989 		DPRINTFN(ENTER, (" syncicache"));
1990 		PMAPCOUNT(exec_synced);
1991 		pmap_syncicache(pa, PAGE_SIZE);
1992 		if (pg != NULL) {
1993 			pmap_attr_save(pg, PTE_EXEC);
1994 			PMAPCOUNT(exec_cached);
1995 #if defined(DEBUG) || defined(PMAPDEBUG)
1996 			if (pmapdebug & PMAPDEBUG_ENTER)
1997 				printf(" marked-as-exec");
1998 			else if (pmapdebug & PMAPDEBUG_EXEC)
1999 				printf("[pmap_enter: %#" _PRIxpa ": marked-as-exec]\n",
2000 				    VM_PAGE_TO_PHYS(pg));
2001 
2002 #endif
2003 		}
2004 	}
2005 
2006 	DPRINTFN(ENTER, (": error=%d\n", error));
2007 
2008 	PMAP_UNLOCK();
2009 
2010 	return error;
2011 }
2012 
2013 void
2014 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
2015 {
2016 	struct mem_region *mp;
2017 	register_t pte_lo;
2018 	int error;
2019 
2020 #if defined (PMAP_OEA64_BRIDGE)
2021 	if (va < VM_MIN_KERNEL_ADDRESS)
2022 		panic("pmap_kenter_pa: attempt to enter "
2023 		    "non-kernel address %#" _PRIxva "!", va);
2024 #endif
2025 
2026 	DPRINTFN(KENTER,
2027 	    ("pmap_kenter_pa(%#" _PRIxva ",%#" _PRIxpa ",%#x)\n", va, pa, prot));
2028 
2029 	PMAP_LOCK();
2030 
2031 	/*
2032 	 * Assume the page is cache inhibited and access is guarded unless
2033 	 * it's in our available memory array.  If it is in the memory array,
2034 	 * asssume it's in memory coherent memory.
2035 	 */
2036 	pte_lo = PTE_IG;
2037 	if ((flags & PMAP_MD_NOCACHE) == 0) {
2038 		for (mp = mem; mp->size; mp++) {
2039 			if (pa >= mp->start && pa < mp->start + mp->size) {
2040 				pte_lo = PTE_M;
2041 				break;
2042 			}
2043 		}
2044 	}
2045 
2046 	if (prot & VM_PROT_WRITE)
2047 		pte_lo |= PTE_BW;
2048 	else
2049 		pte_lo |= PTE_BR;
2050 
2051 	/*
2052 	 * We don't care about REF/CHG on PVOs on the unmanaged list.
2053 	 */
2054 	error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool,
2055 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED);
2056 
2057 	if (error != 0)
2058 		panic("pmap_kenter_pa: failed to enter va %#" _PRIxva " pa %#" _PRIxpa ": %d",
2059 		      va, pa, error);
2060 
2061 	PMAP_UNLOCK();
2062 }
2063 
2064 void
2065 pmap_kremove(vaddr_t va, vsize_t len)
2066 {
2067 	if (va < VM_MIN_KERNEL_ADDRESS)
2068 		panic("pmap_kremove: attempt to remove "
2069 		    "non-kernel address %#" _PRIxva "!", va);
2070 
2071 	DPRINTFN(KREMOVE,("pmap_kremove(%#" _PRIxva ",%#" _PRIxva ")\n", va, len));
2072 	pmap_remove(pmap_kernel(), va, va + len);
2073 }
2074 
2075 /*
2076  * Remove the given range of mapping entries.
2077  */
2078 void
2079 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
2080 {
2081 	struct pvo_head pvol;
2082 	struct pvo_entry *pvo;
2083 	register_t msr;
2084 	int pteidx;
2085 
2086 	PMAP_LOCK();
2087 	LIST_INIT(&pvol);
2088 	msr = pmap_interrupts_off();
2089 	for (; va < endva; va += PAGE_SIZE) {
2090 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2091 		if (pvo != NULL) {
2092 			pmap_pvo_remove(pvo, pteidx, &pvol);
2093 		}
2094 	}
2095 	pmap_interrupts_restore(msr);
2096 	pmap_pvo_free_list(&pvol);
2097 	PMAP_UNLOCK();
2098 }
2099 
2100 /*
2101  * Get the physical page address for the given pmap/virtual address.
2102  */
2103 bool
2104 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
2105 {
2106 	struct pvo_entry *pvo;
2107 	register_t msr;
2108 
2109 	PMAP_LOCK();
2110 
2111 	/*
2112 	 * If this is a kernel pmap lookup, also check the battable
2113 	 * and if we get a hit, translate the VA to a PA using the
2114 	 * BAT entries.  Don't check for VM_MAX_KERNEL_ADDRESS is
2115 	 * that will wrap back to 0.
2116 	 */
2117 	if (pm == pmap_kernel() &&
2118 	    (va < VM_MIN_KERNEL_ADDRESS ||
2119 	     (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
2120 		KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
2121 #if defined (PMAP_OEA)
2122 #ifdef PPC_OEA601
2123 		if ((MFPVR() >> 16) == MPC601) {
2124 			register_t batu = battable[va >> 23].batu;
2125 			register_t batl = battable[va >> 23].batl;
2126 			register_t sr = iosrtable[va >> ADDR_SR_SHFT];
2127 			if (BAT601_VALID_P(batl) &&
2128 			    BAT601_VA_MATCH_P(batu, batl, va)) {
2129 				register_t mask =
2130 				    (~(batl & BAT601_BSM) << 17) & ~0x1ffffL;
2131 				if (pap)
2132 					*pap = (batl & mask) | (va & ~mask);
2133 				PMAP_UNLOCK();
2134 				return true;
2135 			} else if (SR601_VALID_P(sr) &&
2136 				   SR601_PA_MATCH_P(sr, va)) {
2137 				if (pap)
2138 					*pap = va;
2139 				PMAP_UNLOCK();
2140 				return true;
2141 			}
2142 		} else
2143 #endif /* PPC_OEA601 */
2144 		{
2145 			register_t batu = battable[va >> ADDR_SR_SHFT].batu;
2146 			if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) {
2147 				register_t batl =
2148 				    battable[va >> ADDR_SR_SHFT].batl;
2149 				register_t mask =
2150 				    (~(batu & BAT_BL) << 15) & ~0x1ffffL;
2151 				if (pap)
2152 					*pap = (batl & mask) | (va & ~mask);
2153 				PMAP_UNLOCK();
2154 				return true;
2155 			}
2156 		}
2157 		return false;
2158 #elif defined (PMAP_OEA64_BRIDGE)
2159 	if (va >= SEGMENT_LENGTH)
2160 		panic("%s: pm: %s va >= SEGMENT_LENGTH, va: 0x%08lx\n",
2161 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
2162 	else {
2163 		if (pap)
2164 			*pap = va;
2165 			PMAP_UNLOCK();
2166 			return true;
2167 	}
2168 #elif defined (PMAP_OEA64)
2169 #error PPC_OEA64 not supported
2170 #endif /* PPC_OEA */
2171 	}
2172 
2173 	msr = pmap_interrupts_off();
2174 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2175 	if (pvo != NULL) {
2176 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2177 		if (pap)
2178 			*pap = (pvo->pvo_pte.pte_lo & PTE_RPGN)
2179 			    | (va & ADDR_POFF);
2180 	}
2181 	pmap_interrupts_restore(msr);
2182 	PMAP_UNLOCK();
2183 	return pvo != NULL;
2184 }
2185 
2186 /*
2187  * Lower the protection on the specified range of this pmap.
2188  */
2189 void
2190 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
2191 {
2192 	struct pvo_entry *pvo;
2193 	volatile struct pte *pt;
2194 	register_t msr;
2195 	int pteidx;
2196 
2197 	/*
2198 	 * Since this routine only downgrades protection, we should
2199 	 * always be called with at least one bit not set.
2200 	 */
2201 	KASSERT(prot != VM_PROT_ALL);
2202 
2203 	/*
2204 	 * If there is no protection, this is equivalent to
2205 	 * remove the pmap from the pmap.
2206 	 */
2207 	if ((prot & VM_PROT_READ) == 0) {
2208 		pmap_remove(pm, va, endva);
2209 		return;
2210 	}
2211 
2212 	PMAP_LOCK();
2213 
2214 	msr = pmap_interrupts_off();
2215 	for (; va < endva; va += PAGE_SIZE) {
2216 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2217 		if (pvo == NULL)
2218 			continue;
2219 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2220 
2221 		/*
2222 		 * Revoke executable if asked to do so.
2223 		 */
2224 		if ((prot & VM_PROT_EXECUTE) == 0)
2225 			pvo_clear_exec(pvo);
2226 
2227 #if 0
2228 		/*
2229 		 * If the page is already read-only, no change
2230 		 * needs to be made.
2231 		 */
2232 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
2233 			continue;
2234 #endif
2235 		/*
2236 		 * Grab the PTE pointer before we diddle with
2237 		 * the cached PTE copy.
2238 		 */
2239 		pt = pmap_pvo_to_pte(pvo, pteidx);
2240 		/*
2241 		 * Change the protection of the page.
2242 		 */
2243 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2244 		pvo->pvo_pte.pte_lo |= PTE_BR;
2245 
2246 		/*
2247 		 * If the PVO is in the page table, update
2248 		 * that pte at well.
2249 		 */
2250 		if (pt != NULL) {
2251 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2252 			PVO_WHERE(pvo, PMAP_PROTECT);
2253 			PMAPCOUNT(ptes_changed);
2254 		}
2255 
2256 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2257 	}
2258 	pmap_interrupts_restore(msr);
2259 	PMAP_UNLOCK();
2260 }
2261 
2262 void
2263 pmap_unwire(pmap_t pm, vaddr_t va)
2264 {
2265 	struct pvo_entry *pvo;
2266 	register_t msr;
2267 
2268 	PMAP_LOCK();
2269 	msr = pmap_interrupts_off();
2270 	pvo = pmap_pvo_find_va(pm, va, NULL);
2271 	if (pvo != NULL) {
2272 		if (PVO_WIRED_P(pvo)) {
2273 			pvo->pvo_vaddr &= ~PVO_WIRED;
2274 			pm->pm_stats.wired_count--;
2275 		}
2276 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2277 	}
2278 	pmap_interrupts_restore(msr);
2279 	PMAP_UNLOCK();
2280 }
2281 
2282 /*
2283  * Lower the protection on the specified physical page.
2284  */
2285 void
2286 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
2287 {
2288 	struct pvo_head *pvo_head, pvol;
2289 	struct pvo_entry *pvo, *next_pvo;
2290 	volatile struct pte *pt;
2291 	register_t msr;
2292 
2293 	PMAP_LOCK();
2294 
2295 	KASSERT(prot != VM_PROT_ALL);
2296 	LIST_INIT(&pvol);
2297 	msr = pmap_interrupts_off();
2298 
2299 	/*
2300 	 * When UVM reuses a page, it does a pmap_page_protect with
2301 	 * VM_PROT_NONE.  At that point, we can clear the exec flag
2302 	 * since we know the page will have different contents.
2303 	 */
2304 	if ((prot & VM_PROT_READ) == 0) {
2305 		DPRINTFN(EXEC, ("[pmap_page_protect: %#" _PRIxpa ": clear-exec]\n",
2306 		    VM_PAGE_TO_PHYS(pg)));
2307 		if (pmap_attr_fetch(pg) & PTE_EXEC) {
2308 			PMAPCOUNT(exec_uncached_page_protect);
2309 			pmap_attr_clear(pg, PTE_EXEC);
2310 		}
2311 	}
2312 
2313 	pvo_head = vm_page_to_pvoh(pg);
2314 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2315 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
2316 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2317 
2318 		/*
2319 		 * Downgrading to no mapping at all, we just remove the entry.
2320 		 */
2321 		if ((prot & VM_PROT_READ) == 0) {
2322 			pmap_pvo_remove(pvo, -1, &pvol);
2323 			continue;
2324 		}
2325 
2326 		/*
2327 		 * If EXEC permission is being revoked, just clear the
2328 		 * flag in the PVO.
2329 		 */
2330 		if ((prot & VM_PROT_EXECUTE) == 0)
2331 			pvo_clear_exec(pvo);
2332 
2333 		/*
2334 		 * If this entry is already RO, don't diddle with the
2335 		 * page table.
2336 		 */
2337 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2338 			PMAP_PVO_CHECK(pvo);
2339 			continue;
2340 		}
2341 
2342 		/*
2343 		 * Grab the PTE before the we diddle the bits so
2344 		 * pvo_to_pte can verify the pte contents are as
2345 		 * expected.
2346 		 */
2347 		pt = pmap_pvo_to_pte(pvo, -1);
2348 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2349 		pvo->pvo_pte.pte_lo |= PTE_BR;
2350 		if (pt != NULL) {
2351 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2352 			PVO_WHERE(pvo, PMAP_PAGE_PROTECT);
2353 			PMAPCOUNT(ptes_changed);
2354 		}
2355 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2356 	}
2357 	pmap_interrupts_restore(msr);
2358 	pmap_pvo_free_list(&pvol);
2359 
2360 	PMAP_UNLOCK();
2361 }
2362 
2363 /*
2364  * Activate the address space for the specified process.  If the process
2365  * is the current process, load the new MMU context.
2366  */
2367 void
2368 pmap_activate(struct lwp *l)
2369 {
2370 	struct pcb *pcb = lwp_getpcb(l);
2371 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2372 
2373 	DPRINTFN(ACTIVATE,
2374 	    ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp));
2375 
2376 	/*
2377 	 * XXX Normally performed in cpu_lwp_fork().
2378 	 */
2379 	pcb->pcb_pm = pmap;
2380 
2381 	/*
2382 	* In theory, the SR registers need only be valid on return
2383 	* to user space wait to do them there.
2384 	*/
2385 	if (l == curlwp) {
2386 		/* Store pointer to new current pmap. */
2387 		curpm = pmap;
2388 	}
2389 }
2390 
2391 /*
2392  * Deactivate the specified process's address space.
2393  */
2394 void
2395 pmap_deactivate(struct lwp *l)
2396 {
2397 }
2398 
2399 bool
2400 pmap_query_bit(struct vm_page *pg, int ptebit)
2401 {
2402 	struct pvo_entry *pvo;
2403 	volatile struct pte *pt;
2404 	register_t msr;
2405 
2406 	PMAP_LOCK();
2407 
2408 	if (pmap_attr_fetch(pg) & ptebit) {
2409 		PMAP_UNLOCK();
2410 		return true;
2411 	}
2412 
2413 	msr = pmap_interrupts_off();
2414 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2415 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2416 		/*
2417 		 * See if we saved the bit off.  If so cache, it and return
2418 		 * success.
2419 		 */
2420 		if (pvo->pvo_pte.pte_lo & ptebit) {
2421 			pmap_attr_save(pg, ptebit);
2422 			PMAP_PVO_CHECK(pvo);		/* sanity check */
2423 			pmap_interrupts_restore(msr);
2424 			PMAP_UNLOCK();
2425 			return true;
2426 		}
2427 	}
2428 	/*
2429 	 * No luck, now go thru the hard part of looking at the ptes
2430 	 * themselves.  Sync so any pending REF/CHG bits are flushed
2431 	 * to the PTEs.
2432 	 */
2433 	SYNC();
2434 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2435 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2436 		/*
2437 		 * See if this pvo have a valid PTE.  If so, fetch the
2438 		 * REF/CHG bits from the valid PTE.  If the appropriate
2439 		 * ptebit is set, cache, it and return success.
2440 		 */
2441 		pt = pmap_pvo_to_pte(pvo, -1);
2442 		if (pt != NULL) {
2443 			pmap_pte_synch(pt, &pvo->pvo_pte);
2444 			if (pvo->pvo_pte.pte_lo & ptebit) {
2445 				pmap_attr_save(pg, ptebit);
2446 				PMAP_PVO_CHECK(pvo);		/* sanity check */
2447 				pmap_interrupts_restore(msr);
2448 				PMAP_UNLOCK();
2449 				return true;
2450 			}
2451 		}
2452 	}
2453 	pmap_interrupts_restore(msr);
2454 	PMAP_UNLOCK();
2455 	return false;
2456 }
2457 
2458 bool
2459 pmap_clear_bit(struct vm_page *pg, int ptebit)
2460 {
2461 	struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2462 	struct pvo_entry *pvo;
2463 	volatile struct pte *pt;
2464 	register_t msr;
2465 	int rv = 0;
2466 
2467 	PMAP_LOCK();
2468 	msr = pmap_interrupts_off();
2469 
2470 	/*
2471 	 * Fetch the cache value
2472 	 */
2473 	rv |= pmap_attr_fetch(pg);
2474 
2475 	/*
2476 	 * Clear the cached value.
2477 	 */
2478 	pmap_attr_clear(pg, ptebit);
2479 
2480 	/*
2481 	 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2482 	 * can reset the right ones).  Note that since the pvo entries and
2483 	 * list heads are accessed via BAT0 and are never placed in the
2484 	 * page table, we don't have to worry about further accesses setting
2485 	 * the REF/CHG bits.
2486 	 */
2487 	SYNC();
2488 
2489 	/*
2490 	 * For each pvo entry, clear pvo's ptebit.  If this pvo have a
2491 	 * valid PTE.  If so, clear the ptebit from the valid PTE.
2492 	 */
2493 	LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2494 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2495 		pt = pmap_pvo_to_pte(pvo, -1);
2496 		if (pt != NULL) {
2497 			/*
2498 			 * Only sync the PTE if the bit we are looking
2499 			 * for is not already set.
2500 			 */
2501 			if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2502 				pmap_pte_synch(pt, &pvo->pvo_pte);
2503 			/*
2504 			 * If the bit we are looking for was already set,
2505 			 * clear that bit in the pte.
2506 			 */
2507 			if (pvo->pvo_pte.pte_lo & ptebit)
2508 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2509 		}
2510 		rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2511 		pvo->pvo_pte.pte_lo &= ~ptebit;
2512 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2513 	}
2514 	pmap_interrupts_restore(msr);
2515 
2516 	/*
2517 	 * If we are clearing the modify bit and this page was marked EXEC
2518 	 * and the user of the page thinks the page was modified, then we
2519 	 * need to clean it from the icache if it's mapped or clear the EXEC
2520 	 * bit if it's not mapped.  The page itself might not have the CHG
2521 	 * bit set if the modification was done via DMA to the page.
2522 	 */
2523 	if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2524 		if (LIST_EMPTY(pvoh)) {
2525 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#" _PRIxpa ": clear-exec]\n",
2526 			    VM_PAGE_TO_PHYS(pg)));
2527 			pmap_attr_clear(pg, PTE_EXEC);
2528 			PMAPCOUNT(exec_uncached_clear_modify);
2529 		} else {
2530 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#" _PRIxpa ": syncicache]\n",
2531 			    VM_PAGE_TO_PHYS(pg)));
2532 			pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE);
2533 			PMAPCOUNT(exec_synced_clear_modify);
2534 		}
2535 	}
2536 	PMAP_UNLOCK();
2537 	return (rv & ptebit) != 0;
2538 }
2539 
2540 void
2541 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2542 {
2543 	struct pvo_entry *pvo;
2544 	size_t offset = va & ADDR_POFF;
2545 	int s;
2546 
2547 	PMAP_LOCK();
2548 	s = splvm();
2549 	while (len > 0) {
2550 		size_t seglen = PAGE_SIZE - offset;
2551 		if (seglen > len)
2552 			seglen = len;
2553 		pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2554 		if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) {
2555 			pmap_syncicache(
2556 			    (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2557 			PMAP_PVO_CHECK(pvo);
2558 		}
2559 		va += seglen;
2560 		len -= seglen;
2561 		offset = 0;
2562 	}
2563 	splx(s);
2564 	PMAP_UNLOCK();
2565 }
2566 
2567 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2568 void
2569 pmap_pte_print(volatile struct pte *pt)
2570 {
2571 	printf("PTE %p: ", pt);
2572 
2573 #if defined(PMAP_OEA)
2574 	/* High word: */
2575 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2576 #else
2577 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2578 #endif /* PMAP_OEA */
2579 
2580 	printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2581 	printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2582 
2583 	printf("%#" _PRIxpte " %#" _PRIxpte "",
2584 	    (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2585 	    pt->pte_hi & PTE_API);
2586 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2587 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2588 #else
2589 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2590 #endif /* PMAP_OEA */
2591 
2592 	/* Low word: */
2593 #if defined (PMAP_OEA)
2594 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2595 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2596 #else
2597 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2598 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2599 #endif
2600 	printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2601 	printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2602 	printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2603 	printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2604 	printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2605 	printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2606 	switch (pt->pte_lo & PTE_PP) {
2607 	case PTE_BR: printf("br]\n"); break;
2608 	case PTE_BW: printf("bw]\n"); break;
2609 	case PTE_SO: printf("so]\n"); break;
2610 	case PTE_SW: printf("sw]\n"); break;
2611 	}
2612 }
2613 #endif
2614 
2615 #if defined(DDB)
2616 void
2617 pmap_pteg_check(void)
2618 {
2619 	volatile struct pte *pt;
2620 	int i;
2621 	int ptegidx;
2622 	u_int p_valid = 0;
2623 	u_int s_valid = 0;
2624 	u_int invalid = 0;
2625 
2626 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2627 		for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2628 			if (pt->pte_hi & PTE_VALID) {
2629 				if (pt->pte_hi & PTE_HID)
2630 					s_valid++;
2631 				else
2632 				{
2633 					p_valid++;
2634 				}
2635 			} else
2636 				invalid++;
2637 		}
2638 	}
2639 	printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2640 		p_valid, p_valid, s_valid, s_valid,
2641 		invalid, invalid);
2642 }
2643 
2644 void
2645 pmap_print_mmuregs(void)
2646 {
2647 	int i;
2648 	u_int cpuvers;
2649 #ifndef PMAP_OEA64
2650 	vaddr_t addr;
2651 	register_t soft_sr[16];
2652 #endif
2653 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2654 	struct bat soft_ibat[4];
2655 	struct bat soft_dbat[4];
2656 #endif
2657 	paddr_t sdr1;
2658 
2659 	cpuvers = MFPVR() >> 16;
2660 	__asm volatile ("mfsdr1 %0" : "=r"(sdr1));
2661 #ifndef PMAP_OEA64
2662 	addr = 0;
2663 	for (i = 0; i < 16; i++) {
2664 		soft_sr[i] = MFSRIN(addr);
2665 		addr += (1 << ADDR_SR_SHFT);
2666 	}
2667 #endif
2668 
2669 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2670 	/* read iBAT (601: uBAT) registers */
2671 	__asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2672 	__asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2673 	__asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2674 	__asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2675 	__asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2676 	__asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2677 	__asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2678 	__asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2679 
2680 
2681 	if (cpuvers != MPC601) {
2682 		/* read dBAT registers */
2683 		__asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2684 		__asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2685 		__asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2686 		__asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2687 		__asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2688 		__asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2689 		__asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2690 		__asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2691 	}
2692 #endif
2693 
2694 	printf("SDR1:\t%#" _PRIxpa "\n", sdr1);
2695 #ifndef PMAP_OEA64
2696 	printf("SR[]:\t");
2697 	for (i = 0; i < 4; i++)
2698 		printf("0x%08lx,   ", soft_sr[i]);
2699 	printf("\n\t");
2700 	for ( ; i < 8; i++)
2701 		printf("0x%08lx,   ", soft_sr[i]);
2702 	printf("\n\t");
2703 	for ( ; i < 12; i++)
2704 		printf("0x%08lx,   ", soft_sr[i]);
2705 	printf("\n\t");
2706 	for ( ; i < 16; i++)
2707 		printf("0x%08lx,   ", soft_sr[i]);
2708 	printf("\n");
2709 #endif
2710 
2711 #if defined(PMAP_OEA) || defined(PMAP_OEA_BRIDGE)
2712 	printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2713 	for (i = 0; i < 4; i++) {
2714 		printf("0x%08lx 0x%08lx, ",
2715 			soft_ibat[i].batu, soft_ibat[i].batl);
2716 		if (i == 1)
2717 			printf("\n\t");
2718 	}
2719 	if (cpuvers != MPC601) {
2720 		printf("\ndBAT[]:\t");
2721 		for (i = 0; i < 4; i++) {
2722 			printf("0x%08lx 0x%08lx, ",
2723 				soft_dbat[i].batu, soft_dbat[i].batl);
2724 			if (i == 1)
2725 				printf("\n\t");
2726 		}
2727 	}
2728 	printf("\n");
2729 #endif /* PMAP_OEA... */
2730 }
2731 
2732 void
2733 pmap_print_pte(pmap_t pm, vaddr_t va)
2734 {
2735 	struct pvo_entry *pvo;
2736 	volatile struct pte *pt;
2737 	int pteidx;
2738 
2739 	pvo = pmap_pvo_find_va(pm, va, &pteidx);
2740 	if (pvo != NULL) {
2741 		pt = pmap_pvo_to_pte(pvo, pteidx);
2742 		if (pt != NULL) {
2743 			printf("VA %#" _PRIxva " -> %p -> %s %#" _PRIxpte ", %#" _PRIxpte "\n",
2744 				va, pt,
2745 				pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2746 				pt->pte_hi, pt->pte_lo);
2747 		} else {
2748 			printf("No valid PTE found\n");
2749 		}
2750 	} else {
2751 		printf("Address not in pmap\n");
2752 	}
2753 }
2754 
2755 void
2756 pmap_pteg_dist(void)
2757 {
2758 	struct pvo_entry *pvo;
2759 	int ptegidx;
2760 	int depth;
2761 	int max_depth = 0;
2762 	unsigned int depths[64];
2763 
2764 	memset(depths, 0, sizeof(depths));
2765 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2766 		depth = 0;
2767 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2768 			depth++;
2769 		}
2770 		if (depth > max_depth)
2771 			max_depth = depth;
2772 		if (depth > 63)
2773 			depth = 63;
2774 		depths[depth]++;
2775 	}
2776 
2777 	for (depth = 0; depth < 64; depth++) {
2778 		printf("  [%2d]: %8u", depth, depths[depth]);
2779 		if ((depth & 3) == 3)
2780 			printf("\n");
2781 		if (depth == max_depth)
2782 			break;
2783 	}
2784 	if ((depth & 3) != 3)
2785 		printf("\n");
2786 	printf("Max depth found was %d\n", max_depth);
2787 }
2788 #endif /* DEBUG */
2789 
2790 #if defined(PMAPCHECK) || defined(DEBUG)
2791 void
2792 pmap_pvo_verify(void)
2793 {
2794 	int ptegidx;
2795 	int s;
2796 
2797 	s = splvm();
2798 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2799 		struct pvo_entry *pvo;
2800 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2801 			if ((uintptr_t) pvo >= SEGMENT_LENGTH)
2802 				panic("pmap_pvo_verify: invalid pvo %p "
2803 				    "on list %#x", pvo, ptegidx);
2804 			pmap_pvo_check(pvo);
2805 		}
2806 	}
2807 	splx(s);
2808 }
2809 #endif /* PMAPCHECK */
2810 
2811 
2812 void *
2813 pmap_pool_ualloc(struct pool *pp, int flags)
2814 {
2815 	struct pvo_page *pvop;
2816 
2817 	if (uvm.page_init_done != true) {
2818 		return (void *) uvm_pageboot_alloc(PAGE_SIZE);
2819 	}
2820 
2821 	PMAP_LOCK();
2822 	pvop = SIMPLEQ_FIRST(&pmap_upvop_head);
2823 	if (pvop != NULL) {
2824 		pmap_upvop_free--;
2825 		SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link);
2826 		PMAP_UNLOCK();
2827 		return pvop;
2828 	}
2829 	PMAP_UNLOCK();
2830 	return pmap_pool_malloc(pp, flags);
2831 }
2832 
2833 void *
2834 pmap_pool_malloc(struct pool *pp, int flags)
2835 {
2836 	struct pvo_page *pvop;
2837 	struct vm_page *pg;
2838 
2839 	PMAP_LOCK();
2840 	pvop = SIMPLEQ_FIRST(&pmap_mpvop_head);
2841 	if (pvop != NULL) {
2842 		pmap_mpvop_free--;
2843 		SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link);
2844 		PMAP_UNLOCK();
2845 		return pvop;
2846 	}
2847 	PMAP_UNLOCK();
2848  again:
2849 	pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE,
2850 	    UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256);
2851 	if (__predict_false(pg == NULL)) {
2852 		if (flags & PR_WAITOK) {
2853 			uvm_wait("plpg");
2854 			goto again;
2855 		} else {
2856 			return (0);
2857 		}
2858 	}
2859 	KDASSERT(VM_PAGE_TO_PHYS(pg) == (uintptr_t)VM_PAGE_TO_PHYS(pg));
2860 	return (void *)(uintptr_t) VM_PAGE_TO_PHYS(pg);
2861 }
2862 
2863 void
2864 pmap_pool_ufree(struct pool *pp, void *va)
2865 {
2866 	struct pvo_page *pvop;
2867 #if 0
2868 	if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) {
2869 		pmap_pool_mfree(va, size, tag);
2870 		return;
2871 	}
2872 #endif
2873 	PMAP_LOCK();
2874 	pvop = va;
2875 	SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link);
2876 	pmap_upvop_free++;
2877 	if (pmap_upvop_free > pmap_upvop_maxfree)
2878 		pmap_upvop_maxfree = pmap_upvop_free;
2879 	PMAP_UNLOCK();
2880 }
2881 
2882 void
2883 pmap_pool_mfree(struct pool *pp, void *va)
2884 {
2885 	struct pvo_page *pvop;
2886 
2887 	PMAP_LOCK();
2888 	pvop = va;
2889 	SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link);
2890 	pmap_mpvop_free++;
2891 	if (pmap_mpvop_free > pmap_mpvop_maxfree)
2892 		pmap_mpvop_maxfree = pmap_mpvop_free;
2893 	PMAP_UNLOCK();
2894 #if 0
2895 	uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va));
2896 #endif
2897 }
2898 
2899 /*
2900  * This routine in bootstraping to steal to-be-managed memory (which will
2901  * then be unmanaged).  We use it to grab from the first 256MB for our
2902  * pmap needs and above 256MB for other stuff.
2903  */
2904 vaddr_t
2905 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2906 {
2907 	vsize_t size;
2908 	vaddr_t va;
2909 	paddr_t pa = 0;
2910 	int npgs, bank;
2911 	struct vm_physseg *ps;
2912 
2913 	if (uvm.page_init_done == true)
2914 		panic("pmap_steal_memory: called _after_ bootstrap");
2915 
2916 	*vstartp = VM_MIN_KERNEL_ADDRESS;
2917 	*vendp = VM_MAX_KERNEL_ADDRESS;
2918 
2919 	size = round_page(vsize);
2920 	npgs = atop(size);
2921 
2922 	/*
2923 	 * PA 0 will never be among those given to UVM so we can use it
2924 	 * to indicate we couldn't steal any memory.
2925 	 */
2926 	for (bank = 0; bank < vm_nphysseg; bank++) {
2927 		ps = VM_PHYSMEM_PTR(bank);
2928 		if (ps->free_list == VM_FREELIST_FIRST256 &&
2929 		    ps->avail_end - ps->avail_start >= npgs) {
2930 			pa = ptoa(ps->avail_start);
2931 			break;
2932 		}
2933 	}
2934 
2935 	if (pa == 0)
2936 		panic("pmap_steal_memory: no approriate memory to steal!");
2937 
2938 	ps->avail_start += npgs;
2939 	ps->start += npgs;
2940 
2941 	/*
2942 	 * If we've used up all the pages in the segment, remove it and
2943 	 * compact the list.
2944 	 */
2945 	if (ps->avail_start == ps->end) {
2946 		/*
2947 		 * If this was the last one, then a very bad thing has occurred
2948 		 */
2949 		if (--vm_nphysseg == 0)
2950 			panic("pmap_steal_memory: out of memory!");
2951 
2952 		printf("pmap_steal_memory: consumed bank %d\n", bank);
2953 		for (; bank < vm_nphysseg; bank++, ps++) {
2954 			ps[0] = ps[1];
2955 		}
2956 	}
2957 
2958 	va = (vaddr_t) pa;
2959 	memset((void *) va, 0, size);
2960 	pmap_pages_stolen += npgs;
2961 #ifdef DEBUG
2962 	if (pmapdebug && npgs > 1) {
2963 		u_int cnt = 0;
2964 		for (bank = 0; bank < vm_nphysseg; bank++) {
2965 			ps = VM_PHYSMEM_PTR(bank);
2966 			cnt += ps->avail_end - ps->avail_start;
2967 		}
2968 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2969 		    npgs, pmap_pages_stolen, cnt);
2970 	}
2971 #endif
2972 
2973 	return va;
2974 }
2975 
2976 /*
2977  * Find a chuck of memory with right size and alignment.
2978  */
2979 paddr_t
2980 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2981 {
2982 	struct mem_region *mp;
2983 	paddr_t s, e;
2984 	int i, j;
2985 
2986 	size = round_page(size);
2987 
2988 	DPRINTFN(BOOT,
2989 	    ("pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
2990 	    size, alignment, at_end));
2991 
2992 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2993 		panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
2994 		    alignment);
2995 
2996 	if (at_end) {
2997 		if (alignment != PAGE_SIZE)
2998 			panic("pmap_boot_find_memory: invalid ending "
2999 			    "alignment %#" _PRIxpa, alignment);
3000 
3001 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
3002 			s = mp->start + mp->size - size;
3003 			if (s >= mp->start && mp->size >= size) {
3004 				DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
3005 				DPRINTFN(BOOT,
3006 				    ("pmap_boot_find_memory: b-avail[%d] start "
3007 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3008 				     mp->start, mp->size));
3009 				mp->size -= size;
3010 				DPRINTFN(BOOT,
3011 				    ("pmap_boot_find_memory: a-avail[%d] start "
3012 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3013 				     mp->start, mp->size));
3014 				return s;
3015 			}
3016 		}
3017 		panic("pmap_boot_find_memory: no available memory");
3018 	}
3019 
3020 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3021 		s = (mp->start + alignment - 1) & ~(alignment-1);
3022 		e = s + size;
3023 
3024 		/*
3025 		 * Is the calculated region entirely within the region?
3026 		 */
3027 		if (s < mp->start || e > mp->start + mp->size)
3028 			continue;
3029 
3030 		DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
3031 		if (s == mp->start) {
3032 			/*
3033 			 * If the block starts at the beginning of region,
3034 			 * adjust the size & start. (the region may now be
3035 			 * zero in length)
3036 			 */
3037 			DPRINTFN(BOOT,
3038 			    ("pmap_boot_find_memory: b-avail[%d] start "
3039 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3040 			mp->start += size;
3041 			mp->size -= size;
3042 			DPRINTFN(BOOT,
3043 			    ("pmap_boot_find_memory: a-avail[%d] start "
3044 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3045 		} else if (e == mp->start + mp->size) {
3046 			/*
3047 			 * If the block starts at the beginning of region,
3048 			 * adjust only the size.
3049 			 */
3050 			DPRINTFN(BOOT,
3051 			    ("pmap_boot_find_memory: b-avail[%d] start "
3052 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3053 			mp->size -= size;
3054 			DPRINTFN(BOOT,
3055 			    ("pmap_boot_find_memory: a-avail[%d] start "
3056 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3057 		} else {
3058 			/*
3059 			 * Block is in the middle of the region, so we
3060 			 * have to split it in two.
3061 			 */
3062 			for (j = avail_cnt; j > i + 1; j--) {
3063 				avail[j] = avail[j-1];
3064 			}
3065 			DPRINTFN(BOOT,
3066 			    ("pmap_boot_find_memory: b-avail[%d] start "
3067 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3068 			mp[1].start = e;
3069 			mp[1].size = mp[0].start + mp[0].size - e;
3070 			mp[0].size = s - mp[0].start;
3071 			avail_cnt++;
3072 			for (; i < avail_cnt; i++) {
3073 				DPRINTFN(BOOT,
3074 				    ("pmap_boot_find_memory: a-avail[%d] "
3075 				     "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3076 				     avail[i].start, avail[i].size));
3077 			}
3078 		}
3079 		KASSERT(s == (uintptr_t) s);
3080 		return s;
3081 	}
3082 	panic("pmap_boot_find_memory: not enough memory for "
3083 	    "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3084 }
3085 
3086 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3087 #if defined (PMAP_OEA64_BRIDGE)
3088 int
3089 pmap_setup_segment0_map(int use_large_pages, ...)
3090 {
3091     vaddr_t va;
3092 
3093     register_t pte_lo = 0x0;
3094     int ptegidx = 0, i = 0;
3095     struct pte pte;
3096     va_list ap;
3097 
3098     /* Coherent + Supervisor RW, no user access */
3099     pte_lo = PTE_M;
3100 
3101     /* XXXSL
3102      * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3103      * these have to take priority.
3104      */
3105     for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3106         ptegidx = va_to_pteg(pmap_kernel(), va);
3107         pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3108         i = pmap_pte_insert(ptegidx, &pte);
3109     }
3110 
3111     va_start(ap, use_large_pages);
3112     while (1) {
3113         paddr_t pa;
3114         size_t size;
3115 
3116         va = va_arg(ap, vaddr_t);
3117 
3118         if (va == 0)
3119             break;
3120 
3121         pa = va_arg(ap, paddr_t);
3122         size = va_arg(ap, size_t);
3123 
3124         for (; va < (va + size); va += 0x1000, pa += 0x1000) {
3125 #if 0
3126 	    printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__,  va, pa);
3127 #endif
3128             ptegidx = va_to_pteg(pmap_kernel(), va);
3129             pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3130             i = pmap_pte_insert(ptegidx, &pte);
3131         }
3132     }
3133 
3134     TLBSYNC();
3135     SYNC();
3136     return (0);
3137 }
3138 #endif /* PMAP_OEA64_BRIDGE */
3139 
3140 /*
3141  * This is not part of the defined PMAP interface and is specific to the
3142  * PowerPC architecture.  This is called during initppc, before the system
3143  * is really initialized.
3144  */
3145 void
3146 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3147 {
3148 	struct mem_region *mp, tmp;
3149 	paddr_t s, e;
3150 	psize_t size;
3151 	int i, j;
3152 
3153 	/*
3154 	 * Get memory.
3155 	 */
3156 	mem_regions(&mem, &avail);
3157 #if defined(DEBUG)
3158 	if (pmapdebug & PMAPDEBUG_BOOT) {
3159 		printf("pmap_bootstrap: memory configuration:\n");
3160 		for (mp = mem; mp->size; mp++) {
3161 			printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3162 				mp->start, mp->size);
3163 		}
3164 		for (mp = avail; mp->size; mp++) {
3165 			printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3166 				mp->start, mp->size);
3167 		}
3168 	}
3169 #endif
3170 
3171 	/*
3172 	 * Find out how much physical memory we have and in how many chunks.
3173 	 */
3174 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3175 		if (mp->start >= pmap_memlimit)
3176 			continue;
3177 		if (mp->start + mp->size > pmap_memlimit) {
3178 			size = pmap_memlimit - mp->start;
3179 			physmem += btoc(size);
3180 		} else {
3181 			physmem += btoc(mp->size);
3182 		}
3183 		mem_cnt++;
3184 	}
3185 
3186 	/*
3187 	 * Count the number of available entries.
3188 	 */
3189 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
3190 		avail_cnt++;
3191 
3192 	/*
3193 	 * Page align all regions.
3194 	 */
3195 	kernelstart = trunc_page(kernelstart);
3196 	kernelend = round_page(kernelend);
3197 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3198 		s = round_page(mp->start);
3199 		mp->size -= (s - mp->start);
3200 		mp->size = trunc_page(mp->size);
3201 		mp->start = s;
3202 		e = mp->start + mp->size;
3203 
3204 		DPRINTFN(BOOT,
3205 		    ("pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3206 		    i, mp->start, mp->size));
3207 
3208 		/*
3209 		 * Don't allow the end to run beyond our artificial limit
3210 		 */
3211 		if (e > pmap_memlimit)
3212 			e = pmap_memlimit;
3213 
3214 		/*
3215 		 * Is this region empty or strange?  skip it.
3216 		 */
3217 		if (e <= s) {
3218 			mp->start = 0;
3219 			mp->size = 0;
3220 			continue;
3221 		}
3222 
3223 		/*
3224 		 * Does this overlap the beginning of kernel?
3225 		 *   Does extend past the end of the kernel?
3226 		 */
3227 		else if (s < kernelstart && e > kernelstart) {
3228 			if (e > kernelend) {
3229 				avail[avail_cnt].start = kernelend;
3230 				avail[avail_cnt].size = e - kernelend;
3231 				avail_cnt++;
3232 			}
3233 			mp->size = kernelstart - s;
3234 		}
3235 		/*
3236 		 * Check whether this region overlaps the end of the kernel.
3237 		 */
3238 		else if (s < kernelend && e > kernelend) {
3239 			mp->start = kernelend;
3240 			mp->size = e - kernelend;
3241 		}
3242 		/*
3243 		 * Look whether this regions is completely inside the kernel.
3244 		 * Nuke it if it does.
3245 		 */
3246 		else if (s >= kernelstart && e <= kernelend) {
3247 			mp->start = 0;
3248 			mp->size = 0;
3249 		}
3250 		/*
3251 		 * If the user imposed a memory limit, enforce it.
3252 		 */
3253 		else if (s >= pmap_memlimit) {
3254 			mp->start = -PAGE_SIZE;	/* let's know why */
3255 			mp->size = 0;
3256 		}
3257 		else {
3258 			mp->start = s;
3259 			mp->size = e - s;
3260 		}
3261 		DPRINTFN(BOOT,
3262 		    ("pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3263 		    i, mp->start, mp->size));
3264 	}
3265 
3266 	/*
3267 	 * Move (and uncount) all the null return to the end.
3268 	 */
3269 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3270 		if (mp->size == 0) {
3271 			tmp = avail[i];
3272 			avail[i] = avail[--avail_cnt];
3273 			avail[avail_cnt] = avail[i];
3274 		}
3275 	}
3276 
3277 	/*
3278 	 * (Bubble)sort them into ascending order.
3279 	 */
3280 	for (i = 0; i < avail_cnt; i++) {
3281 		for (j = i + 1; j < avail_cnt; j++) {
3282 			if (avail[i].start > avail[j].start) {
3283 				tmp = avail[i];
3284 				avail[i] = avail[j];
3285 				avail[j] = tmp;
3286 			}
3287 		}
3288 	}
3289 
3290 	/*
3291 	 * Make sure they don't overlap.
3292 	 */
3293 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3294 		if (mp[0].start + mp[0].size > mp[1].start) {
3295 			mp[0].size = mp[1].start - mp[0].start;
3296 		}
3297 		DPRINTFN(BOOT,
3298 		    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3299 		    i, mp->start, mp->size));
3300 	}
3301 	DPRINTFN(BOOT,
3302 	    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3303 	    i, mp->start, mp->size));
3304 
3305 #ifdef	PTEGCOUNT
3306 	pmap_pteg_cnt = PTEGCOUNT;
3307 #else /* PTEGCOUNT */
3308 
3309 	pmap_pteg_cnt = 0x1000;
3310 
3311 	while (pmap_pteg_cnt < physmem)
3312 		pmap_pteg_cnt <<= 1;
3313 
3314 	pmap_pteg_cnt >>= 1;
3315 #endif /* PTEGCOUNT */
3316 
3317 #ifdef DEBUG
3318 	DPRINTFN(BOOT,
3319 		("pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt));
3320 #endif
3321 
3322 	/*
3323 	 * Find suitably aligned memory for PTEG hash table.
3324 	 */
3325 	size = pmap_pteg_cnt * sizeof(struct pteg);
3326 	pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3327 
3328 #ifdef DEBUG
3329 	DPRINTFN(BOOT,
3330 		("PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table));
3331 #endif
3332 
3333 
3334 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3335 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
3336 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > 256MB",
3337 		    pmap_pteg_table, size);
3338 #endif
3339 
3340 	memset(__UNVOLATILE(pmap_pteg_table), 0,
3341 		pmap_pteg_cnt * sizeof(struct pteg));
3342 	pmap_pteg_mask = pmap_pteg_cnt - 1;
3343 
3344 	/*
3345 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3346 	 * with pages.  So we just steal them before giving them to UVM.
3347 	 */
3348 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3349 	pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3350 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3351 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
3352 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > 256MB",
3353 		    pmap_pvo_table, size);
3354 #endif
3355 
3356 	for (i = 0; i < pmap_pteg_cnt; i++)
3357 		TAILQ_INIT(&pmap_pvo_table[i]);
3358 
3359 #ifndef MSGBUFADDR
3360 	/*
3361 	 * Allocate msgbuf in high memory.
3362 	 */
3363 	msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3364 #endif
3365 
3366 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3367 		paddr_t pfstart = atop(mp->start);
3368 		paddr_t pfend = atop(mp->start + mp->size);
3369 		if (mp->size == 0)
3370 			continue;
3371 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
3372 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3373 				VM_FREELIST_FIRST256);
3374 		} else if (mp->start >= SEGMENT_LENGTH) {
3375 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3376 				VM_FREELIST_DEFAULT);
3377 		} else {
3378 			pfend = atop(SEGMENT_LENGTH);
3379 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3380 				VM_FREELIST_FIRST256);
3381 			pfstart = atop(SEGMENT_LENGTH);
3382 			pfend = atop(mp->start + mp->size);
3383 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3384 				VM_FREELIST_DEFAULT);
3385 		}
3386 	}
3387 
3388 	/*
3389 	 * Make sure kernel vsid is allocated as well as VSID 0.
3390 	 */
3391 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3392 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3393 	pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3394 		|= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3395 	pmap_vsid_bitmap[0] |= 1;
3396 
3397 	/*
3398 	 * Initialize kernel pmap and hardware.
3399 	 */
3400 
3401 /* PMAP_OEA64_BRIDGE does support these instructions */
3402 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
3403 	for (i = 0; i < 16; i++) {
3404  		pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3405 		__asm volatile ("mtsrin %0,%1"
3406  			      :: "r"(KERNELN_SEGMENT(i)|SR_PRKEY), "r"(i << ADDR_SR_SHFT));
3407 	}
3408 
3409 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3410 	__asm volatile ("mtsr %0,%1"
3411 		      :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
3412 #ifdef KERNEL2_SR
3413 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3414 	__asm volatile ("mtsr %0,%1"
3415 		      :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
3416 #endif
3417 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3418 #if defined (PMAP_OEA)
3419 	for (i = 0; i < 16; i++) {
3420 		if (iosrtable[i] & SR601_T) {
3421 			pmap_kernel()->pm_sr[i] = iosrtable[i];
3422 			__asm volatile ("mtsrin %0,%1"
3423 			    :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT));
3424 		}
3425 	}
3426 	__asm volatile ("sync; mtsdr1 %0; isync"
3427 		      :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10)));
3428 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
3429  	__asm __volatile ("sync; mtsdr1 %0; isync"
3430  		      :: "r"((uintptr_t)pmap_pteg_table | (32 - cntlzw(pmap_pteg_mask >> 11))));
3431 #endif
3432 	tlbia();
3433 
3434 #ifdef ALTIVEC
3435 	pmap_use_altivec = cpu_altivec;
3436 #endif
3437 
3438 #ifdef DEBUG
3439 	if (pmapdebug & PMAPDEBUG_BOOT) {
3440 		u_int cnt;
3441 		int bank;
3442 		char pbuf[9];
3443 		for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) {
3444 			cnt += VM_PHYSMEM_PTR(bank)->avail_end - VM_PHYSMEM_PTR(bank)->avail_start;
3445 			printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3446 			    bank,
3447 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_start),
3448 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_end),
3449 			    ptoa(VM_PHYSMEM_PTR(bank)->avail_end - VM_PHYSMEM_PTR(bank)->avail_start));
3450 		}
3451 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3452 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3453 		    pbuf, cnt);
3454 	}
3455 #endif
3456 
3457 	pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry),
3458 	    sizeof(struct pvo_entry), 0, 0, "pmap_upvopl",
3459 	    &pmap_pool_uallocator, IPL_VM);
3460 
3461 	pool_setlowat(&pmap_upvo_pool, 252);
3462 
3463 	pool_init(&pmap_pool, sizeof(struct pmap),
3464 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator,
3465 	    IPL_NONE);
3466 
3467 #if defined(PMAP_NEED_MAPKERNEL) || 1
3468 	{
3469 		struct pmap *pm = pmap_kernel();
3470 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3471 		extern int etext[], kernel_text[];
3472 		vaddr_t va, va_etext = (paddr_t) etext;
3473 #endif
3474 		paddr_t pa, pa_end;
3475 		register_t sr;
3476 		struct pte pt;
3477 		unsigned int ptegidx;
3478 		int bank;
3479 
3480 		sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3481 		pm->pm_sr[0] = sr;
3482 
3483 		for (bank = 0; bank < vm_nphysseg; bank++) {
3484 			pa_end = ptoa(VM_PHYSMEM_PTR(bank)->avail_end);
3485 			pa = ptoa(VM_PHYSMEM_PTR(bank)->avail_start);
3486 			for (; pa < pa_end; pa += PAGE_SIZE) {
3487 				ptegidx = va_to_pteg(pm, pa);
3488 				pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3489 				pmap_pte_insert(ptegidx, &pt);
3490 			}
3491 		}
3492 
3493 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3494 		va = (vaddr_t) kernel_text;
3495 
3496 		for (pa = kernelstart; va < va_etext;
3497 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3498 			ptegidx = va_to_pteg(pm, va);
3499 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3500 			pmap_pte_insert(ptegidx, &pt);
3501 		}
3502 
3503 		for (; pa < kernelend;
3504 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3505 			ptegidx = va_to_pteg(pm, va);
3506 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3507 			pmap_pte_insert(ptegidx, &pt);
3508 		}
3509 
3510 		for (va = 0, pa = 0; va < kernelstart;
3511 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3512 			ptegidx = va_to_pteg(pm, va);
3513 			if (va < 0x3000)
3514 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3515 			else
3516 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3517 			pmap_pte_insert(ptegidx, &pt);
3518 		}
3519 		for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3520 		    pa += PAGE_SIZE, va += PAGE_SIZE) {
3521 			ptegidx = va_to_pteg(pm, va);
3522 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3523 			pmap_pte_insert(ptegidx, &pt);
3524 		}
3525 #endif
3526 
3527 		__asm volatile ("mtsrin %0,%1"
3528  			      :: "r"(sr), "r"(kernelstart));
3529 	}
3530 #endif
3531 }
3532