xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: pmap.c,v 1.65 2009/04/21 21:29:59 cegger 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 1.65 2009/04/21 21:29:59 cegger 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/user.h>
79 #include <sys/pool.h>
80 #include <sys/queue.h>
81 #include <sys/device.h>		/* for evcnt */
82 #include <sys/systm.h>
83 #include <sys/atomic.h>
84 
85 #include <uvm/uvm.h>
86 
87 #include <machine/pcb.h>
88 #include <machine/powerpc.h>
89 #include <powerpc/spr.h>
90 #include <powerpc/oea/sr_601.h>
91 #include <powerpc/bat.h>
92 #include <powerpc/stdarg.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 int physmem;
116 #ifndef MSGBUFADDR
117 extern paddr_t msgbuf_paddr;
118 #endif
119 
120 static struct mem_region *mem, *avail;
121 static u_int mem_cnt, avail_cnt;
122 
123 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
124 # define	PMAP_OEA 1
125 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA64) && !defined(PPC_OEA64_BRIDGE)
126 #  define	PMAPNAME(name)	pmap_##name
127 # endif
128 #endif
129 
130 #if defined(PMAP_OEA64)
131 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA) && !defined(PPC_OEA64_BRIDGE)
132 #  define	PMAPNAME(name)	pmap_##name
133 # endif
134 #endif
135 
136 #if defined(PMAP_OEA64_BRIDGE)
137 # if defined(PMAP_EXCLUDE_DECLS) && !defined(PPC_OEA) && !defined(PPC_OEA64)
138 #  define	PMAPNAME(name)	pmap_##name
139 # endif
140 #endif
141 
142 #if defined(PMAP_OEA)
143 #define	_PRIxpte	"lx"
144 #else
145 #define	_PRIxpte	PRIx64
146 #endif
147 #define	_PRIxpa		"lx"
148 #define	_PRIxva		"lx"
149 #define	_PRIsr  	"lx"
150 
151 #if defined(PMAP_EXCLUDE_DECLS) && !defined(PMAPNAME)
152 #if defined(PMAP_OEA)
153 #define	PMAPNAME(name)	pmap32_##name
154 #elif defined(PMAP_OEA64)
155 #define	PMAPNAME(name)	pmap64_##name
156 #elif defined(PMAP_OEA64_BRIDGE)
157 #define	PMAPNAME(name)	pmap64bridge_##name
158 #else
159 #error unknown variant for pmap
160 #endif
161 #endif /* PMAP_EXLCUDE_DECLS && !PMAPNAME */
162 
163 #if defined(PMAPNAME)
164 #define	STATIC			static
165 #define pmap_pte_spill		PMAPNAME(pte_spill)
166 #define pmap_real_memory	PMAPNAME(real_memory)
167 #define pmap_init		PMAPNAME(init)
168 #define pmap_virtual_space	PMAPNAME(virtual_space)
169 #define pmap_create		PMAPNAME(create)
170 #define pmap_reference		PMAPNAME(reference)
171 #define pmap_destroy		PMAPNAME(destroy)
172 #define pmap_copy		PMAPNAME(copy)
173 #define pmap_update		PMAPNAME(update)
174 #define pmap_collect		PMAPNAME(collect)
175 #define pmap_enter		PMAPNAME(enter)
176 #define pmap_remove		PMAPNAME(remove)
177 #define pmap_kenter_pa		PMAPNAME(kenter_pa)
178 #define pmap_kremove		PMAPNAME(kremove)
179 #define pmap_extract		PMAPNAME(extract)
180 #define pmap_protect		PMAPNAME(protect)
181 #define pmap_unwire		PMAPNAME(unwire)
182 #define pmap_page_protect	PMAPNAME(page_protect)
183 #define pmap_query_bit		PMAPNAME(query_bit)
184 #define pmap_clear_bit		PMAPNAME(clear_bit)
185 
186 #define pmap_activate		PMAPNAME(activate)
187 #define pmap_deactivate		PMAPNAME(deactivate)
188 
189 #define pmap_pinit		PMAPNAME(pinit)
190 #define pmap_procwr		PMAPNAME(procwr)
191 
192 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
193 #define pmap_pte_print		PMAPNAME(pte_print)
194 #define pmap_pteg_check		PMAPNAME(pteg_check)
195 #define pmap_print_mmruregs	PMAPNAME(print_mmuregs)
196 #define pmap_print_pte		PMAPNAME(print_pte)
197 #define pmap_pteg_dist		PMAPNAME(pteg_dist)
198 #endif
199 #if defined(DEBUG) || defined(PMAPCHECK)
200 #define	pmap_pvo_verify		PMAPNAME(pvo_verify)
201 #define pmapcheck		PMAPNAME(check)
202 #endif
203 #if defined(DEBUG) || defined(PMAPDEBUG)
204 #define pmapdebug		PMAPNAME(debug)
205 #endif
206 #define pmap_steal_memory	PMAPNAME(steal_memory)
207 #define pmap_bootstrap		PMAPNAME(bootstrap)
208 #else
209 #define	STATIC			/* nothing */
210 #endif /* PMAPNAME */
211 
212 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
213 STATIC void pmap_real_memory(paddr_t *, psize_t *);
214 STATIC void pmap_init(void);
215 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
216 STATIC pmap_t pmap_create(void);
217 STATIC void pmap_reference(pmap_t);
218 STATIC void pmap_destroy(pmap_t);
219 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
220 STATIC void pmap_update(pmap_t);
221 STATIC void pmap_collect(pmap_t);
222 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
223 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
224 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t);
225 STATIC void pmap_kremove(vaddr_t, vsize_t);
226 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
227 
228 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
229 STATIC void pmap_unwire(pmap_t, vaddr_t);
230 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
231 STATIC bool pmap_query_bit(struct vm_page *, int);
232 STATIC bool pmap_clear_bit(struct vm_page *, int);
233 
234 STATIC void pmap_activate(struct lwp *);
235 STATIC void pmap_deactivate(struct lwp *);
236 
237 STATIC void pmap_pinit(pmap_t pm);
238 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
239 
240 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
241 STATIC void pmap_pte_print(volatile struct pte *);
242 STATIC void pmap_pteg_check(void);
243 STATIC void pmap_print_mmuregs(void);
244 STATIC void pmap_print_pte(pmap_t, vaddr_t);
245 STATIC void pmap_pteg_dist(void);
246 #endif
247 #if defined(DEBUG) || defined(PMAPCHECK)
248 STATIC void pmap_pvo_verify(void);
249 #endif
250 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
251 STATIC void pmap_bootstrap(paddr_t, paddr_t);
252 
253 #ifdef PMAPNAME
254 const struct pmap_ops PMAPNAME(ops) = {
255 	.pmapop_pte_spill = pmap_pte_spill,
256 	.pmapop_real_memory = pmap_real_memory,
257 	.pmapop_init = pmap_init,
258 	.pmapop_virtual_space = pmap_virtual_space,
259 	.pmapop_create = pmap_create,
260 	.pmapop_reference = pmap_reference,
261 	.pmapop_destroy = pmap_destroy,
262 	.pmapop_copy = pmap_copy,
263 	.pmapop_update = pmap_update,
264 	.pmapop_collect = pmap_collect,
265 	.pmapop_enter = pmap_enter,
266 	.pmapop_remove = pmap_remove,
267 	.pmapop_kenter_pa = pmap_kenter_pa,
268 	.pmapop_kremove = pmap_kremove,
269 	.pmapop_extract = pmap_extract,
270 	.pmapop_protect = pmap_protect,
271 	.pmapop_unwire = pmap_unwire,
272 	.pmapop_page_protect = pmap_page_protect,
273 	.pmapop_query_bit = pmap_query_bit,
274 	.pmapop_clear_bit = pmap_clear_bit,
275 	.pmapop_activate = pmap_activate,
276 	.pmapop_deactivate = pmap_deactivate,
277 	.pmapop_pinit = pmap_pinit,
278 	.pmapop_procwr = pmap_procwr,
279 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
280 	.pmapop_pte_print = pmap_pte_print,
281 	.pmapop_pteg_check = pmap_pteg_check,
282 	.pmapop_print_mmuregs = pmap_print_mmuregs,
283 	.pmapop_print_pte = pmap_print_pte,
284 	.pmapop_pteg_dist = pmap_pteg_dist,
285 #else
286 	.pmapop_pte_print = NULL,
287 	.pmapop_pteg_check = NULL,
288 	.pmapop_print_mmuregs = NULL,
289 	.pmapop_print_pte = NULL,
290 	.pmapop_pteg_dist = NULL,
291 #endif
292 #if defined(DEBUG) || defined(PMAPCHECK)
293 	.pmapop_pvo_verify = pmap_pvo_verify,
294 #else
295 	.pmapop_pvo_verify = NULL,
296 #endif
297 	.pmapop_steal_memory = pmap_steal_memory,
298 	.pmapop_bootstrap = pmap_bootstrap,
299 };
300 #endif /* !PMAPNAME */
301 
302 /*
303  * The following structure is aligned to 32 bytes
304  */
305 struct pvo_entry {
306 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
307 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
308 	struct pte pvo_pte;			/* Prebuilt PTE */
309 	pmap_t pvo_pmap;			/* ptr to owning pmap */
310 	vaddr_t pvo_vaddr;			/* VA of entry */
311 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
312 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
313 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
314 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
315 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
316 #define	PVO_WIRED_P(pvo)	((pvo)->pvo_vaddr & PVO_WIRED)
317 #define	PVO_MANAGED_P(pvo)	((pvo)->pvo_vaddr & PVO_MANAGED)
318 #define	PVO_EXECUTABLE_P(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
319 #define	PVO_ENTER_INSERT	0		/* PVO has been removed */
320 #define	PVO_SPILL_UNSET		1		/* PVO has been evicted */
321 #define	PVO_SPILL_SET		2		/* PVO has been spilled */
322 #define	PVO_SPILL_INSERT	3		/* PVO has been inserted */
323 #define	PVO_PMAP_PAGE_PROTECT	4		/* PVO has changed */
324 #define	PVO_PMAP_PROTECT	5		/* PVO has changed */
325 #define	PVO_REMOVE		6		/* PVO has been removed */
326 #define	PVO_WHERE_MASK		15
327 #define	PVO_WHERE_SHFT		8
328 } __attribute__ ((aligned (32)));
329 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
330 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
331 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
332 #define	PVO_PTEGIDX_CLR(pvo)	\
333 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
334 #define	PVO_PTEGIDX_SET(pvo,i)	\
335 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
336 #define	PVO_WHERE(pvo,w)	\
337 	((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
338 	 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
339 
340 TAILQ_HEAD(pvo_tqhead, pvo_entry);
341 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
342 static struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
343 static struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
344 
345 struct pool pmap_pool;		/* pool for pmap structures */
346 struct pool pmap_upvo_pool;	/* pool for pvo entries for unmanaged pages */
347 struct pool pmap_mpvo_pool;	/* pool for pvo entries for managed pages */
348 
349 /*
350  * We keep a cache of unmanaged pages to be used for pvo entries for
351  * unmanaged pages.
352  */
353 struct pvo_page {
354 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
355 };
356 SIMPLEQ_HEAD(pvop_head, pvo_page);
357 static struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head);
358 static struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head);
359 u_long pmap_upvop_free;
360 u_long pmap_upvop_maxfree;
361 u_long pmap_mpvop_free;
362 u_long pmap_mpvop_maxfree;
363 
364 static void *pmap_pool_ualloc(struct pool *, int);
365 static void *pmap_pool_malloc(struct pool *, int);
366 
367 static void pmap_pool_ufree(struct pool *, void *);
368 static void pmap_pool_mfree(struct pool *, void *);
369 
370 static struct pool_allocator pmap_pool_mallocator = {
371 	.pa_alloc = pmap_pool_malloc,
372 	.pa_free = pmap_pool_mfree,
373 	.pa_pagesz = 0,
374 };
375 
376 static struct pool_allocator pmap_pool_uallocator = {
377 	.pa_alloc = pmap_pool_ualloc,
378 	.pa_free = pmap_pool_ufree,
379 	.pa_pagesz = 0,
380 };
381 
382 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
383 void pmap_pte_print(volatile struct pte *);
384 void pmap_pteg_check(void);
385 void pmap_pteg_dist(void);
386 void pmap_print_pte(pmap_t, vaddr_t);
387 void pmap_print_mmuregs(void);
388 #endif
389 
390 #if defined(DEBUG) || defined(PMAPCHECK)
391 #ifdef PMAPCHECK
392 int pmapcheck = 1;
393 #else
394 int pmapcheck = 0;
395 #endif
396 void pmap_pvo_verify(void);
397 static void pmap_pvo_check(const struct pvo_entry *);
398 #define	PMAP_PVO_CHECK(pvo)	 		\
399 	do {					\
400 		if (pmapcheck)			\
401 			pmap_pvo_check(pvo);	\
402 	} while (0)
403 #else
404 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
405 #endif
406 static int pmap_pte_insert(int, struct pte *);
407 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
408 	vaddr_t, paddr_t, register_t, int);
409 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
410 static void pmap_pvo_free(struct pvo_entry *);
411 static void pmap_pvo_free_list(struct pvo_head *);
412 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
413 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
414 static struct pvo_entry *pmap_pvo_reclaim(struct pmap *);
415 static void pvo_set_exec(struct pvo_entry *);
416 static void pvo_clear_exec(struct pvo_entry *);
417 
418 static void tlbia(void);
419 
420 static void pmap_release(pmap_t);
421 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
422 
423 static uint32_t pmap_pvo_reclaim_nextidx;
424 #ifdef DEBUG
425 static int pmap_pvo_reclaim_debugctr;
426 #endif
427 
428 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
429 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
430 
431 static int pmap_initialized;
432 
433 #if defined(DEBUG) || defined(PMAPDEBUG)
434 #define	PMAPDEBUG_BOOT		0x0001
435 #define	PMAPDEBUG_PTE		0x0002
436 #define	PMAPDEBUG_EXEC		0x0008
437 #define	PMAPDEBUG_PVOENTER	0x0010
438 #define	PMAPDEBUG_PVOREMOVE	0x0020
439 #define	PMAPDEBUG_ACTIVATE	0x0100
440 #define	PMAPDEBUG_CREATE	0x0200
441 #define	PMAPDEBUG_ENTER		0x1000
442 #define	PMAPDEBUG_KENTER	0x2000
443 #define	PMAPDEBUG_KREMOVE	0x4000
444 #define	PMAPDEBUG_REMOVE	0x8000
445 
446 unsigned int pmapdebug = 0;
447 
448 # define DPRINTF(x)		printf x
449 # define DPRINTFN(n, x)		if (pmapdebug & PMAPDEBUG_ ## n) printf x
450 #else
451 # define DPRINTF(x)
452 # define DPRINTFN(n, x)
453 #endif
454 
455 
456 #ifdef PMAPCOUNTERS
457 /*
458  * From pmap_subr.c
459  */
460 extern struct evcnt pmap_evcnt_mappings;
461 extern struct evcnt pmap_evcnt_unmappings;
462 
463 extern struct evcnt pmap_evcnt_kernel_mappings;
464 extern struct evcnt pmap_evcnt_kernel_unmappings;
465 
466 extern struct evcnt pmap_evcnt_mappings_replaced;
467 
468 extern struct evcnt pmap_evcnt_exec_mappings;
469 extern struct evcnt pmap_evcnt_exec_cached;
470 
471 extern struct evcnt pmap_evcnt_exec_synced;
472 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
473 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
474 
475 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
476 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
477 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
478 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
479 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
480 
481 extern struct evcnt pmap_evcnt_updates;
482 extern struct evcnt pmap_evcnt_collects;
483 extern struct evcnt pmap_evcnt_copies;
484 
485 extern struct evcnt pmap_evcnt_ptes_spilled;
486 extern struct evcnt pmap_evcnt_ptes_unspilled;
487 extern struct evcnt pmap_evcnt_ptes_evicted;
488 
489 extern struct evcnt pmap_evcnt_ptes_primary[8];
490 extern struct evcnt pmap_evcnt_ptes_secondary[8];
491 extern struct evcnt pmap_evcnt_ptes_removed;
492 extern struct evcnt pmap_evcnt_ptes_changed;
493 extern struct evcnt pmap_evcnt_pvos_reclaimed;
494 extern struct evcnt pmap_evcnt_pvos_failed;
495 
496 extern struct evcnt pmap_evcnt_zeroed_pages;
497 extern struct evcnt pmap_evcnt_copied_pages;
498 extern struct evcnt pmap_evcnt_idlezeroed_pages;
499 
500 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
501 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
502 #else
503 #define	PMAPCOUNT(ev)	((void) 0)
504 #define	PMAPCOUNT2(ev)	((void) 0)
505 #endif
506 
507 #define	TLBIE(va)	__asm volatile("tlbie %0" :: "r"(va))
508 
509 /* XXXSL: this needs to be moved to assembler */
510 #define	TLBIEL(va)	__asm __volatile("tlbie %0" :: "r"(va))
511 
512 #define	TLBSYNC()	__asm volatile("tlbsync")
513 #define	SYNC()		__asm volatile("sync")
514 #define	EIEIO()		__asm volatile("eieio")
515 #define	DCBST(va)	__asm __volatile("dcbst 0,%0" :: "r"(va))
516 #define	MFMSR()		mfmsr()
517 #define	MTMSR(psl)	mtmsr(psl)
518 #define	MFPVR()		mfpvr()
519 #define	MFSRIN(va)	mfsrin(va)
520 #define	MFTB()		mfrtcltbl()
521 
522 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
523 static inline register_t
524 mfsrin(vaddr_t va)
525 {
526 	register_t sr;
527 	__asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
528 	return sr;
529 }
530 #endif	/* PMAP_OEA*/
531 
532 #if defined (PMAP_OEA64_BRIDGE)
533 extern void mfmsr64 (register64_t *result);
534 #endif /* PMAP_OEA64_BRIDGE */
535 
536 #define	PMAP_LOCK()		KERNEL_LOCK(1, NULL)
537 #define	PMAP_UNLOCK()		KERNEL_UNLOCK_ONE(NULL)
538 
539 static inline register_t
540 pmap_interrupts_off(void)
541 {
542 	register_t msr = MFMSR();
543 	if (msr & PSL_EE)
544 		MTMSR(msr & ~PSL_EE);
545 	return msr;
546 }
547 
548 static void
549 pmap_interrupts_restore(register_t msr)
550 {
551 	if (msr & PSL_EE)
552 		MTMSR(msr);
553 }
554 
555 static inline u_int32_t
556 mfrtcltbl(void)
557 {
558 #ifdef PPC_OEA601
559 	if ((MFPVR() >> 16) == MPC601)
560 		return (mfrtcl() >> 7);
561 	else
562 #endif
563 		return (mftbl());
564 }
565 
566 /*
567  * These small routines may have to be replaced,
568  * if/when we support processors other that the 604.
569  */
570 
571 void
572 tlbia(void)
573 {
574 	char *i;
575 
576 	SYNC();
577 #if defined(PMAP_OEA)
578 	/*
579 	 * Why not use "tlbia"?  Because not all processors implement it.
580 	 *
581 	 * This needs to be a per-CPU callback to do the appropriate thing
582 	 * for the CPU. XXX
583 	 */
584 	for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
585 		TLBIE(i);
586 		EIEIO();
587 		SYNC();
588 	}
589 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
590 	/* This is specifically for the 970, 970UM v1.6 pp. 140. */
591 	for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
592 		TLBIEL(i);
593 		EIEIO();
594 		SYNC();
595 	}
596 #endif
597 	TLBSYNC();
598 	SYNC();
599 }
600 
601 static inline register_t
602 va_to_vsid(const struct pmap *pm, vaddr_t addr)
603 {
604 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
605 	return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT;
606 #else /* PMAP_OEA64 */
607 #if 0
608 	const struct ste *ste;
609 	register_t hash;
610 	int i;
611 
612 	hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH;
613 
614 	/*
615 	 * Try the primary group first
616 	 */
617 	ste = pm->pm_stes[hash].stes;
618 	for (i = 0; i < 8; i++, ste++) {
619 		if (ste->ste_hi & STE_V) &&
620 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
621 			return ste;
622 	}
623 
624 	/*
625 	 * Then the secondary group.
626 	 */
627 	ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes;
628 	for (i = 0; i < 8; i++, ste++) {
629 		if (ste->ste_hi & STE_V) &&
630 		   (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID))
631 			return addr;
632 	}
633 
634 	return NULL;
635 #else
636 	/*
637 	 * Rather than searching the STE groups for the VSID, we know
638 	 * how we generate that from the ESID and so do that.
639 	 */
640 	return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
641 #endif
642 #endif /* PMAP_OEA */
643 }
644 
645 static inline register_t
646 va_to_pteg(const struct pmap *pm, vaddr_t addr)
647 {
648 	register_t hash;
649 
650 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
651 	return hash & pmap_pteg_mask;
652 }
653 
654 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
655 /*
656  * Given a PTE in the page table, calculate the VADDR that hashes to it.
657  * The only bit of magic is that the top 4 bits of the address doesn't
658  * technically exist in the PTE.  But we know we reserved 4 bits of the
659  * VSID for it so that's how we get it.
660  */
661 static vaddr_t
662 pmap_pte_to_va(volatile const struct pte *pt)
663 {
664 	vaddr_t va;
665 	uintptr_t ptaddr = (uintptr_t) pt;
666 
667 	if (pt->pte_hi & PTE_HID)
668 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
669 
670 	/* PPC Bits 10-19  PPC64 Bits 42-51 */
671 #if defined(PMAP_OEA)
672 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
673 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
674 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
675 #endif
676 	va <<= ADDR_PIDX_SHFT;
677 
678 	/* PPC Bits 4-9  PPC64 Bits 36-41 */
679 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
680 
681 #if defined(PMAP_OEA64)
682 	/* PPC63 Bits 0-35 */
683 	/* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
684 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
685 	/* PPC Bits 0-3 */
686 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
687 #endif
688 
689 	return va;
690 }
691 #endif
692 
693 static inline struct pvo_head *
694 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
695 {
696 	struct vm_page *pg;
697 
698 	pg = PHYS_TO_VM_PAGE(pa);
699 	if (pg_p != NULL)
700 		*pg_p = pg;
701 	if (pg == NULL)
702 		return &pmap_pvo_unmanaged;
703 	return &pg->mdpage.mdpg_pvoh;
704 }
705 
706 static inline struct pvo_head *
707 vm_page_to_pvoh(struct vm_page *pg)
708 {
709 	return &pg->mdpage.mdpg_pvoh;
710 }
711 
712 
713 static inline void
714 pmap_attr_clear(struct vm_page *pg, int ptebit)
715 {
716 	pg->mdpage.mdpg_attrs &= ~ptebit;
717 }
718 
719 static inline int
720 pmap_attr_fetch(struct vm_page *pg)
721 {
722 	return pg->mdpage.mdpg_attrs;
723 }
724 
725 static inline void
726 pmap_attr_save(struct vm_page *pg, int ptebit)
727 {
728 	pg->mdpage.mdpg_attrs |= ptebit;
729 }
730 
731 static inline int
732 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
733 {
734 	if (pt->pte_hi == pvo_pt->pte_hi
735 #if 0
736 	    && ((pt->pte_lo ^ pvo_pt->pte_lo) &
737 	        ~(PTE_REF|PTE_CHG)) == 0
738 #endif
739 	    )
740 		return 1;
741 	return 0;
742 }
743 
744 static inline void
745 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
746 {
747 	/*
748 	 * Construct the PTE.  Default to IMB initially.  Valid bit
749 	 * only gets set when the real pte is set in memory.
750 	 *
751 	 * Note: Don't set the valid bit for correct operation of tlb update.
752 	 */
753 #if defined(PMAP_OEA)
754 	pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
755 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
756 	pt->pte_lo = pte_lo;
757 #elif defined (PMAP_OEA64_BRIDGE)
758 	pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT)
759 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
760 	pt->pte_lo = (u_int64_t) pte_lo;
761 #elif defined (PMAP_OEA64)
762 #error PMAP_OEA64 not supported
763 #endif /* PMAP_OEA */
764 }
765 
766 static inline void
767 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
768 {
769 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
770 }
771 
772 static inline void
773 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
774 {
775 	/*
776 	 * As shown in Section 7.6.3.2.3
777 	 */
778 	pt->pte_lo &= ~ptebit;
779 	TLBIE(va);
780 	SYNC();
781 	EIEIO();
782 	TLBSYNC();
783 	SYNC();
784 #ifdef MULTIPROCESSOR
785 	DCBST(pt);
786 #endif
787 }
788 
789 static inline void
790 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
791 {
792 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
793 	if (pvo_pt->pte_hi & PTE_VALID)
794 		panic("pte_set: setting an already valid pte %p", pvo_pt);
795 #endif
796 	pvo_pt->pte_hi |= PTE_VALID;
797 
798 	/*
799 	 * Update the PTE as defined in section 7.6.3.1
800 	 * Note that the REF/CHG bits are from pvo_pt and thus should
801 	 * have been saved so this routine can restore them (if desired).
802 	 */
803 	pt->pte_lo = pvo_pt->pte_lo;
804 	EIEIO();
805 	pt->pte_hi = pvo_pt->pte_hi;
806 	TLBSYNC();
807 	SYNC();
808 #ifdef MULTIPROCESSOR
809 	DCBST(pt);
810 #endif
811 	pmap_pte_valid++;
812 }
813 
814 static inline void
815 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
816 {
817 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
818 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
819 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
820 	if ((pt->pte_hi & PTE_VALID) == 0)
821 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
822 #endif
823 
824 	pvo_pt->pte_hi &= ~PTE_VALID;
825 	/*
826 	 * Force the ref & chg bits back into the PTEs.
827 	 */
828 	SYNC();
829 	/*
830 	 * Invalidate the pte ... (Section 7.6.3.3)
831 	 */
832 	pt->pte_hi &= ~PTE_VALID;
833 	SYNC();
834 	TLBIE(va);
835 	SYNC();
836 	EIEIO();
837 	TLBSYNC();
838 	SYNC();
839 	/*
840 	 * Save the ref & chg bits ...
841 	 */
842 	pmap_pte_synch(pt, pvo_pt);
843 	pmap_pte_valid--;
844 }
845 
846 static inline void
847 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
848 {
849 	/*
850 	 * Invalidate the PTE
851 	 */
852 	pmap_pte_unset(pt, pvo_pt, va);
853 	pmap_pte_set(pt, pvo_pt);
854 }
855 
856 /*
857  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
858  * (either primary or secondary location).
859  *
860  * Note: both the destination and source PTEs must not have PTE_VALID set.
861  */
862 
863 static int
864 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
865 {
866 	volatile struct pte *pt;
867 	int i;
868 
869 #if defined(DEBUG)
870 	DPRINTFN(PTE, ("pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
871 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo));
872 #endif
873 	/*
874 	 * First try primary hash.
875 	 */
876 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
877 		if ((pt->pte_hi & PTE_VALID) == 0) {
878 			pvo_pt->pte_hi &= ~PTE_HID;
879 			pmap_pte_set(pt, pvo_pt);
880 			return i;
881 		}
882 	}
883 
884 	/*
885 	 * Now try secondary hash.
886 	 */
887 	ptegidx ^= pmap_pteg_mask;
888 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
889 		if ((pt->pte_hi & PTE_VALID) == 0) {
890 			pvo_pt->pte_hi |= PTE_HID;
891 			pmap_pte_set(pt, pvo_pt);
892 			return i;
893 		}
894 	}
895 	return -1;
896 }
897 
898 /*
899  * Spill handler.
900  *
901  * Tries to spill a page table entry from the overflow area.
902  * This runs in either real mode (if dealing with a exception spill)
903  * or virtual mode when dealing with manually spilling one of the
904  * kernel's pte entries.  In either case, interrupts are already
905  * disabled.
906  */
907 
908 int
909 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec)
910 {
911 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
912 	struct pvo_entry *pvo;
913 	/* XXX: gcc -- vpvoh is always set at either *1* or *2* */
914 	struct pvo_tqhead *pvoh, *vpvoh = NULL;
915 	int ptegidx, i, j;
916 	volatile struct pteg *pteg;
917 	volatile struct pte *pt;
918 
919 	PMAP_LOCK();
920 
921 	ptegidx = va_to_pteg(pm, addr);
922 
923 	/*
924 	 * Have to substitute some entry. Use the primary hash for this.
925 	 * Use low bits of timebase as random generator.  Make sure we are
926 	 * not picking a kernel pte for replacement.
927 	 */
928 	pteg = &pmap_pteg_table[ptegidx];
929 	i = MFTB() & 7;
930 	for (j = 0; j < 8; j++) {
931 		pt = &pteg->pt[i];
932 		if ((pt->pte_hi & PTE_VALID) == 0)
933 			break;
934 		if (VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT)
935 				< PHYSMAP_VSIDBITS)
936 			break;
937 		i = (i + 1) & 7;
938 	}
939 	KASSERT(j < 8);
940 
941 	source_pvo = NULL;
942 	victim_pvo = NULL;
943 	pvoh = &pmap_pvo_table[ptegidx];
944 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
945 
946 		/*
947 		 * We need to find pvo entry for this address...
948 		 */
949 		PMAP_PVO_CHECK(pvo);		/* sanity check */
950 
951 		/*
952 		 * If we haven't found the source and we come to a PVO with
953 		 * a valid PTE, then we know we can't find it because all
954 		 * evicted PVOs always are first in the list.
955 		 */
956 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
957 			break;
958 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
959 		    addr == PVO_VADDR(pvo)) {
960 
961 			/*
962 			 * Now we have found the entry to be spilled into the
963 			 * pteg.  Attempt to insert it into the page table.
964 			 */
965 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
966 			if (j >= 0) {
967 				PVO_PTEGIDX_SET(pvo, j);
968 				PMAP_PVO_CHECK(pvo);	/* sanity check */
969 				PVO_WHERE(pvo, SPILL_INSERT);
970 				pvo->pvo_pmap->pm_evictions--;
971 				PMAPCOUNT(ptes_spilled);
972 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
973 				    ? pmap_evcnt_ptes_secondary
974 				    : pmap_evcnt_ptes_primary)[j]);
975 
976 				/*
977 				 * Since we keep the evicted entries at the
978 				 * from of the PVO list, we need move this
979 				 * (now resident) PVO after the evicted
980 				 * entries.
981 				 */
982 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
983 
984 				/*
985 				 * If we don't have to move (either we were the
986 				 * last entry or the next entry was valid),
987 				 * don't change our position.  Otherwise
988 				 * move ourselves to the tail of the queue.
989 				 */
990 				if (next_pvo != NULL &&
991 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
992 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
993 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
994 				}
995 				PMAP_UNLOCK();
996 				return 1;
997 			}
998 			source_pvo = pvo;
999 			if (exec && !PVO_EXECUTABLE_P(source_pvo)) {
1000 				return 0;
1001 			}
1002 			if (victim_pvo != NULL)
1003 				break;
1004 		}
1005 
1006 		/*
1007 		 * We also need the pvo entry of the victim we are replacing
1008 		 * so save the R & C bits of the PTE.
1009 		 */
1010 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
1011 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
1012 			vpvoh = pvoh;			/* *1* */
1013 			victim_pvo = pvo;
1014 			if (source_pvo != NULL)
1015 				break;
1016 		}
1017 	}
1018 
1019 	if (source_pvo == NULL) {
1020 		PMAPCOUNT(ptes_unspilled);
1021 		PMAP_UNLOCK();
1022 		return 0;
1023 	}
1024 
1025 	if (victim_pvo == NULL) {
1026 		if ((pt->pte_hi & PTE_HID) == 0)
1027 			panic("pmap_pte_spill: victim p-pte (%p) has "
1028 			    "no pvo entry!", pt);
1029 
1030 		/*
1031 		 * If this is a secondary PTE, we need to search
1032 		 * its primary pvo bucket for the matching PVO.
1033 		 */
1034 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */
1035 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1036 			PMAP_PVO_CHECK(pvo);		/* sanity check */
1037 
1038 			/*
1039 			 * We also need the pvo entry of the victim we are
1040 			 * replacing so save the R & C bits of the PTE.
1041 			 */
1042 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1043 				victim_pvo = pvo;
1044 				break;
1045 			}
1046 		}
1047 		if (victim_pvo == NULL)
1048 			panic("pmap_pte_spill: victim s-pte (%p) has "
1049 			    "no pvo entry!", pt);
1050 	}
1051 
1052 	/*
1053 	 * The victim should be not be a kernel PVO/PTE entry.
1054 	 */
1055 	KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1056 	KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1057 	KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1058 
1059 	/*
1060 	 * We are invalidating the TLB entry for the EA for the
1061 	 * we are replacing even though its valid; If we don't
1062 	 * we lose any ref/chg bit changes contained in the TLB
1063 	 * entry.
1064 	 */
1065 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1066 
1067 	/*
1068 	 * To enforce the PVO list ordering constraint that all
1069 	 * evicted entries should come before all valid entries,
1070 	 * move the source PVO to the tail of its list and the
1071 	 * victim PVO to the head of its list (which might not be
1072 	 * the same list, if the victim was using the secondary hash).
1073 	 */
1074 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
1075 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
1076 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1077 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1078 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1079 	pmap_pte_set(pt, &source_pvo->pvo_pte);
1080 	victim_pvo->pvo_pmap->pm_evictions++;
1081 	source_pvo->pvo_pmap->pm_evictions--;
1082 	PVO_WHERE(victim_pvo, SPILL_UNSET);
1083 	PVO_WHERE(source_pvo, SPILL_SET);
1084 
1085 	PVO_PTEGIDX_CLR(victim_pvo);
1086 	PVO_PTEGIDX_SET(source_pvo, i);
1087 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1088 	PMAPCOUNT(ptes_spilled);
1089 	PMAPCOUNT(ptes_evicted);
1090 	PMAPCOUNT(ptes_removed);
1091 
1092 	PMAP_PVO_CHECK(victim_pvo);
1093 	PMAP_PVO_CHECK(source_pvo);
1094 
1095 	PMAP_UNLOCK();
1096 	return 1;
1097 }
1098 
1099 /*
1100  * Restrict given range to physical memory
1101  */
1102 void
1103 pmap_real_memory(paddr_t *start, psize_t *size)
1104 {
1105 	struct mem_region *mp;
1106 
1107 	for (mp = mem; mp->size; mp++) {
1108 		if (*start + *size > mp->start
1109 		    && *start < mp->start + mp->size) {
1110 			if (*start < mp->start) {
1111 				*size -= mp->start - *start;
1112 				*start = mp->start;
1113 			}
1114 			if (*start + *size > mp->start + mp->size)
1115 				*size = mp->start + mp->size - *start;
1116 			return;
1117 		}
1118 	}
1119 	*size = 0;
1120 }
1121 
1122 /*
1123  * Initialize anything else for pmap handling.
1124  * Called during vm_init().
1125  */
1126 void
1127 pmap_init(void)
1128 {
1129 	pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry),
1130 	    sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl",
1131 	    &pmap_pool_mallocator, IPL_NONE);
1132 
1133 	pool_setlowat(&pmap_mpvo_pool, 1008);
1134 
1135 	pmap_initialized = 1;
1136 
1137 }
1138 
1139 /*
1140  * How much virtual space does the kernel get?
1141  */
1142 void
1143 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1144 {
1145 	/*
1146 	 * For now, reserve one segment (minus some overhead) for kernel
1147 	 * virtual memory
1148 	 */
1149 	*start = VM_MIN_KERNEL_ADDRESS;
1150 	*end = VM_MAX_KERNEL_ADDRESS;
1151 }
1152 
1153 /*
1154  * Allocate, initialize, and return a new physical map.
1155  */
1156 pmap_t
1157 pmap_create(void)
1158 {
1159 	pmap_t pm;
1160 
1161 	pm = pool_get(&pmap_pool, PR_WAITOK);
1162 	memset((void *)pm, 0, sizeof *pm);
1163 	pmap_pinit(pm);
1164 
1165 	DPRINTFN(CREATE,("pmap_create: pm %p:\n"
1166 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1167 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1168 	    "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1169 	    "    %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1170 	    pm,
1171 	    pm->pm_sr[0], pm->pm_sr[1],
1172 	    pm->pm_sr[2], pm->pm_sr[3],
1173 	    pm->pm_sr[4], pm->pm_sr[5],
1174 	    pm->pm_sr[6], pm->pm_sr[7],
1175 	    pm->pm_sr[8], pm->pm_sr[9],
1176 	    pm->pm_sr[10], pm->pm_sr[11],
1177 	    pm->pm_sr[12], pm->pm_sr[13],
1178 	    pm->pm_sr[14], pm->pm_sr[15]));
1179 	return pm;
1180 }
1181 
1182 /*
1183  * Initialize a preallocated and zeroed pmap structure.
1184  */
1185 void
1186 pmap_pinit(pmap_t pm)
1187 {
1188 	register_t entropy = MFTB();
1189 	register_t mask;
1190 	int i;
1191 
1192 	/*
1193 	 * Allocate some segment registers for this pmap.
1194 	 */
1195 	pm->pm_refs = 1;
1196 	PMAP_LOCK();
1197 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1198 		static register_t pmap_vsidcontext;
1199 		register_t hash;
1200 		unsigned int n;
1201 
1202 		/* Create a new value by multiplying by a prime adding in
1203 		 * entropy from the timebase register.  This is to make the
1204 		 * VSID more random so that the PT Hash function collides
1205 		 * less often. (note that the prime causes gcc to do shifts
1206 		 * instead of a multiply)
1207 		 */
1208 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1209 		hash = pmap_vsidcontext & (NPMAPS - 1);
1210 		if (hash == 0) {		/* 0 is special, avoid it */
1211 			entropy += 0xbadf00d;
1212 			continue;
1213 		}
1214 		n = hash >> 5;
1215 		mask = 1L << (hash & (VSID_NBPW-1));
1216 		hash = pmap_vsidcontext;
1217 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1218 			/* anything free in this bucket? */
1219 			if (~pmap_vsid_bitmap[n] == 0) {
1220 				entropy = hash ^ (hash >> 16);
1221 				continue;
1222 			}
1223 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1224 			mask = 1L << i;
1225 			hash &= ~(VSID_NBPW-1);
1226 			hash |= i;
1227 		}
1228 		hash &= PTE_VSID >> PTE_VSID_SHFT;
1229 		pmap_vsid_bitmap[n] |= mask;
1230 		pm->pm_vsid = hash;
1231 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1232 		for (i = 0; i < 16; i++)
1233 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1234 			    SR_NOEXEC;
1235 #endif
1236 		PMAP_UNLOCK();
1237 		return;
1238 	}
1239 	PMAP_UNLOCK();
1240 	panic("pmap_pinit: out of segments");
1241 }
1242 
1243 /*
1244  * Add a reference to the given pmap.
1245  */
1246 void
1247 pmap_reference(pmap_t pm)
1248 {
1249 	atomic_inc_uint(&pm->pm_refs);
1250 }
1251 
1252 /*
1253  * Retire the given pmap from service.
1254  * Should only be called if the map contains no valid mappings.
1255  */
1256 void
1257 pmap_destroy(pmap_t pm)
1258 {
1259 	if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1260 		pmap_release(pm);
1261 		pool_put(&pmap_pool, pm);
1262 	}
1263 }
1264 
1265 /*
1266  * Release any resources held by the given physical map.
1267  * Called when a pmap initialized by pmap_pinit is being released.
1268  */
1269 void
1270 pmap_release(pmap_t pm)
1271 {
1272 	int idx, mask;
1273 
1274 	KASSERT(pm->pm_stats.resident_count == 0);
1275 	KASSERT(pm->pm_stats.wired_count == 0);
1276 
1277 	PMAP_LOCK();
1278 	if (pm->pm_sr[0] == 0)
1279 		panic("pmap_release");
1280 	idx = pm->pm_vsid & (NPMAPS-1);
1281 	mask = 1 << (idx % VSID_NBPW);
1282 	idx /= VSID_NBPW;
1283 
1284 	KASSERT(pmap_vsid_bitmap[idx] & mask);
1285 	pmap_vsid_bitmap[idx] &= ~mask;
1286 	PMAP_UNLOCK();
1287 }
1288 
1289 /*
1290  * Copy the range specified by src_addr/len
1291  * from the source map to the range dst_addr/len
1292  * in the destination map.
1293  *
1294  * This routine is only advisory and need not do anything.
1295  */
1296 void
1297 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1298 	vsize_t len, vaddr_t src_addr)
1299 {
1300 	PMAPCOUNT(copies);
1301 }
1302 
1303 /*
1304  * Require that all active physical maps contain no
1305  * incorrect entries NOW.
1306  */
1307 void
1308 pmap_update(struct pmap *pmap)
1309 {
1310 	PMAPCOUNT(updates);
1311 	TLBSYNC();
1312 }
1313 
1314 /*
1315  * Garbage collects the physical map system for
1316  * pages which are no longer used.
1317  * Success need not be guaranteed -- that is, there
1318  * may well be pages which are not referenced, but
1319  * others may be collected.
1320  * Called by the pageout daemon when pages are scarce.
1321  */
1322 void
1323 pmap_collect(pmap_t pm)
1324 {
1325 	PMAPCOUNT(collects);
1326 }
1327 
1328 static inline int
1329 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1330 {
1331 	int pteidx;
1332 	/*
1333 	 * We can find the actual pte entry without searching by
1334 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1335 	 * and by noticing the HID bit.
1336 	 */
1337 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1338 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1339 		pteidx ^= pmap_pteg_mask * 8;
1340 	return pteidx;
1341 }
1342 
1343 volatile struct pte *
1344 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1345 {
1346 	volatile struct pte *pt;
1347 
1348 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1349 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1350 		return NULL;
1351 #endif
1352 
1353 	/*
1354 	 * If we haven't been supplied the ptegidx, calculate it.
1355 	 */
1356 	if (pteidx == -1) {
1357 		int ptegidx;
1358 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1359 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1360 	}
1361 
1362 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1363 
1364 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1365 	return pt;
1366 #else
1367 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1368 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1369 		    "pvo but no valid pte index", pvo);
1370 	}
1371 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1372 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1373 		    "pvo but no valid pte", pvo);
1374 	}
1375 
1376 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1377 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1378 #if defined(DEBUG) || defined(PMAPCHECK)
1379 			pmap_pte_print(pt);
1380 #endif
1381 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1382 			    "pmap_pteg_table %p but invalid in pvo",
1383 			    pvo, pt);
1384 		}
1385 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1386 #if defined(DEBUG) || defined(PMAPCHECK)
1387 			pmap_pte_print(pt);
1388 #endif
1389 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1390 			    "not match pte %p in pmap_pteg_table",
1391 			    pvo, pt);
1392 		}
1393 		return pt;
1394 	}
1395 
1396 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1397 #if defined(DEBUG) || defined(PMAPCHECK)
1398 		pmap_pte_print(pt);
1399 #endif
1400 		panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1401 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1402 	}
1403 	return NULL;
1404 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1405 }
1406 
1407 struct pvo_entry *
1408 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1409 {
1410 	struct pvo_entry *pvo;
1411 	int ptegidx;
1412 
1413 	va &= ~ADDR_POFF;
1414 	ptegidx = va_to_pteg(pm, va);
1415 
1416 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1417 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1418 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1419 			panic("pmap_pvo_find_va: invalid pvo %p on "
1420 			    "list %#x (%p)", pvo, ptegidx,
1421 			     &pmap_pvo_table[ptegidx]);
1422 #endif
1423 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1424 			if (pteidx_p)
1425 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1426 			return pvo;
1427 		}
1428 	}
1429 	if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH))
1430 		panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1431 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1432 	return NULL;
1433 }
1434 
1435 #if defined(DEBUG) || defined(PMAPCHECK)
1436 void
1437 pmap_pvo_check(const struct pvo_entry *pvo)
1438 {
1439 	struct pvo_head *pvo_head;
1440 	struct pvo_entry *pvo0;
1441 	volatile struct pte *pt;
1442 	int failed = 0;
1443 
1444 	PMAP_LOCK();
1445 
1446 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1447 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1448 
1449 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1450 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1451 		    pvo, pvo->pvo_pmap);
1452 		failed = 1;
1453 	}
1454 
1455 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1456 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1457 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1458 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1459 		failed = 1;
1460 	}
1461 
1462 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1463 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1464 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1465 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1466 		failed = 1;
1467 	}
1468 
1469 	if (PVO_MANAGED_P(pvo)) {
1470 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1471 	} else {
1472 		if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) {
1473 			printf("pmap_pvo_check: pvo %p: non kernel address "
1474 			    "on kernel unmanaged list\n", pvo);
1475 			failed = 1;
1476 		}
1477 		pvo_head = &pmap_pvo_kunmanaged;
1478 	}
1479 	LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1480 		if (pvo0 == pvo)
1481 			break;
1482 	}
1483 	if (pvo0 == NULL) {
1484 		printf("pmap_pvo_check: pvo %p: not present "
1485 		    "on its vlist head %p\n", pvo, pvo_head);
1486 		failed = 1;
1487 	}
1488 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1489 		printf("pmap_pvo_check: pvo %p: not present "
1490 		    "on its olist head\n", pvo);
1491 		failed = 1;
1492 	}
1493 	pt = pmap_pvo_to_pte(pvo, -1);
1494 	if (pt == NULL) {
1495 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1496 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1497 			    "no PTE\n", pvo);
1498 			failed = 1;
1499 		}
1500 	} else {
1501 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1502 		    (uintptr_t) pt >=
1503 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1504 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1505 			    "pteg table\n", pvo, pt);
1506 			failed = 1;
1507 		}
1508 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1509 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1510 			    "no PTE\n", pvo);
1511 			failed = 1;
1512 		}
1513 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1514 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1515 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1516 			    pvo->pvo_pte.pte_hi,
1517 			    pt->pte_hi);
1518 			failed = 1;
1519 		}
1520 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1521 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1522 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1523 			    "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1524 			    (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1525 			    (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1526 			failed = 1;
1527 		}
1528 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1529 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1530 			    " doesn't not match PVO's VA %#" _PRIxva "\n",
1531 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1532 			failed = 1;
1533 		}
1534 		if (failed)
1535 			pmap_pte_print(pt);
1536 	}
1537 	if (failed)
1538 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1539 		    pvo->pvo_pmap);
1540 
1541 	PMAP_UNLOCK();
1542 }
1543 #endif /* DEBUG || PMAPCHECK */
1544 
1545 /*
1546  * Search the PVO table looking for a non-wired entry.
1547  * If we find one, remove it and return it.
1548  */
1549 
1550 struct pvo_entry *
1551 pmap_pvo_reclaim(struct pmap *pm)
1552 {
1553 	struct pvo_tqhead *pvoh;
1554 	struct pvo_entry *pvo;
1555 	uint32_t idx, endidx;
1556 
1557 	endidx = pmap_pvo_reclaim_nextidx;
1558 	for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1559 	     idx = (idx + 1) & pmap_pteg_mask) {
1560 		pvoh = &pmap_pvo_table[idx];
1561 		TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1562 			if (!PVO_WIRED_P(pvo)) {
1563 				pmap_pvo_remove(pvo, -1, NULL);
1564 				pmap_pvo_reclaim_nextidx = idx;
1565 				PMAPCOUNT(pvos_reclaimed);
1566 				return pvo;
1567 			}
1568 		}
1569 	}
1570 	return NULL;
1571 }
1572 
1573 /*
1574  * This returns whether this is the first mapping of a page.
1575  */
1576 int
1577 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1578 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1579 {
1580 	struct pvo_entry *pvo;
1581 	struct pvo_tqhead *pvoh;
1582 	register_t msr;
1583 	int ptegidx;
1584 	int i;
1585 	int poolflags = PR_NOWAIT;
1586 
1587 	/*
1588 	 * Compute the PTE Group index.
1589 	 */
1590 	va &= ~ADDR_POFF;
1591 	ptegidx = va_to_pteg(pm, va);
1592 
1593 	msr = pmap_interrupts_off();
1594 
1595 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1596 	if (pmap_pvo_remove_depth > 0)
1597 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1598 	if (++pmap_pvo_enter_depth > 1)
1599 		panic("pmap_pvo_enter: called recursively!");
1600 #endif
1601 
1602 	/*
1603 	 * Remove any existing mapping for this page.  Reuse the
1604 	 * pvo entry if there a mapping.
1605 	 */
1606 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1607 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1608 #ifdef DEBUG
1609 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1610 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1611 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1612 			   va < VM_MIN_KERNEL_ADDRESS) {
1613 				printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1614 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1615 				printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1616 				    pvo->pvo_pte.pte_hi,
1617 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1618 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1619 #ifdef DDBX
1620 				Debugger();
1621 #endif
1622 			}
1623 #endif
1624 			PMAPCOUNT(mappings_replaced);
1625 			pmap_pvo_remove(pvo, -1, NULL);
1626 			break;
1627 		}
1628 	}
1629 
1630 	/*
1631 	 * If we aren't overwriting an mapping, try to allocate
1632 	 */
1633 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1634 	--pmap_pvo_enter_depth;
1635 #endif
1636 	pmap_interrupts_restore(msr);
1637 	if (pvo) {
1638 		pmap_pvo_free(pvo);
1639 	}
1640 	pvo = pool_get(pl, poolflags);
1641 
1642 #ifdef DEBUG
1643 	/*
1644 	 * Exercise pmap_pvo_reclaim() a little.
1645 	 */
1646 	if (pvo && (flags & PMAP_CANFAIL) != 0 &&
1647 	    pmap_pvo_reclaim_debugctr++ > 0x1000 &&
1648 	    (pmap_pvo_reclaim_debugctr & 0xff) == 0) {
1649 		pool_put(pl, pvo);
1650 		pvo = NULL;
1651 	}
1652 #endif
1653 
1654 	msr = pmap_interrupts_off();
1655 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1656 	++pmap_pvo_enter_depth;
1657 #endif
1658 	if (pvo == NULL) {
1659 		pvo = pmap_pvo_reclaim(pm);
1660 		if (pvo == NULL) {
1661 			if ((flags & PMAP_CANFAIL) == 0)
1662 				panic("pmap_pvo_enter: failed");
1663 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1664 			pmap_pvo_enter_depth--;
1665 #endif
1666 			PMAPCOUNT(pvos_failed);
1667 			pmap_interrupts_restore(msr);
1668 			return ENOMEM;
1669 		}
1670 	}
1671 
1672 	pvo->pvo_vaddr = va;
1673 	pvo->pvo_pmap = pm;
1674 	pvo->pvo_vaddr &= ~ADDR_POFF;
1675 	if (flags & VM_PROT_EXECUTE) {
1676 		PMAPCOUNT(exec_mappings);
1677 		pvo_set_exec(pvo);
1678 	}
1679 	if (flags & PMAP_WIRED)
1680 		pvo->pvo_vaddr |= PVO_WIRED;
1681 	if (pvo_head != &pmap_pvo_kunmanaged) {
1682 		pvo->pvo_vaddr |= PVO_MANAGED;
1683 		PMAPCOUNT(mappings);
1684 	} else {
1685 		PMAPCOUNT(kernel_mappings);
1686 	}
1687 	pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1688 
1689 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1690 	if (PVO_WIRED_P(pvo))
1691 		pvo->pvo_pmap->pm_stats.wired_count++;
1692 	pvo->pvo_pmap->pm_stats.resident_count++;
1693 #if defined(DEBUG)
1694 /*	if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */
1695 		DPRINTFN(PVOENTER,
1696 		    ("pmap_pvo_enter: pvo %p: pm %p va %#" _PRIxva " pa %#" _PRIxpa "\n",
1697 		    pvo, pm, va, pa));
1698 #endif
1699 
1700 	/*
1701 	 * We hope this succeeds but it isn't required.
1702 	 */
1703 	pvoh = &pmap_pvo_table[ptegidx];
1704 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1705 	if (i >= 0) {
1706 		PVO_PTEGIDX_SET(pvo, i);
1707 		PVO_WHERE(pvo, ENTER_INSERT);
1708 		PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1709 		    ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1710 		TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1711 
1712 	} else {
1713 		/*
1714 		 * Since we didn't have room for this entry (which makes it
1715 		 * and evicted entry), place it at the head of the list.
1716 		 */
1717 		TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1718 		PMAPCOUNT(ptes_evicted);
1719 		pm->pm_evictions++;
1720 		/*
1721 		 * If this is a kernel page, make sure it's active.
1722 		 */
1723 		if (pm == pmap_kernel()) {
1724 			i = pmap_pte_spill(pm, va, false);
1725 			KASSERT(i);
1726 		}
1727 	}
1728 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1729 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1730 	pmap_pvo_enter_depth--;
1731 #endif
1732 	pmap_interrupts_restore(msr);
1733 	return 0;
1734 }
1735 
1736 static void
1737 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol)
1738 {
1739 	volatile struct pte *pt;
1740 	int ptegidx;
1741 
1742 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1743 	if (++pmap_pvo_remove_depth > 1)
1744 		panic("pmap_pvo_remove: called recursively!");
1745 #endif
1746 
1747 	/*
1748 	 * If we haven't been supplied the ptegidx, calculate it.
1749 	 */
1750 	if (pteidx == -1) {
1751 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1752 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1753 	} else {
1754 		ptegidx = pteidx >> 3;
1755 		if (pvo->pvo_pte.pte_hi & PTE_HID)
1756 			ptegidx ^= pmap_pteg_mask;
1757 	}
1758 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1759 
1760 	/*
1761 	 * If there is an active pte entry, we need to deactivate it
1762 	 * (and save the ref & chg bits).
1763 	 */
1764 	pt = pmap_pvo_to_pte(pvo, pteidx);
1765 	if (pt != NULL) {
1766 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1767 		PVO_WHERE(pvo, REMOVE);
1768 		PVO_PTEGIDX_CLR(pvo);
1769 		PMAPCOUNT(ptes_removed);
1770 	} else {
1771 		KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1772 		pvo->pvo_pmap->pm_evictions--;
1773 	}
1774 
1775 	/*
1776 	 * Account for executable mappings.
1777 	 */
1778 	if (PVO_EXECUTABLE_P(pvo))
1779 		pvo_clear_exec(pvo);
1780 
1781 	/*
1782 	 * Update our statistics.
1783 	 */
1784 	pvo->pvo_pmap->pm_stats.resident_count--;
1785 	if (PVO_WIRED_P(pvo))
1786 		pvo->pvo_pmap->pm_stats.wired_count--;
1787 
1788 	/*
1789 	 * Save the REF/CHG bits into their cache if the page is managed.
1790 	 */
1791 	if (PVO_MANAGED_P(pvo)) {
1792 		register_t ptelo = pvo->pvo_pte.pte_lo;
1793 		struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1794 
1795 		if (pg != NULL) {
1796 			/*
1797 			 * If this page was changed and it is mapped exec,
1798 			 * invalidate it.
1799 			 */
1800 			if ((ptelo & PTE_CHG) &&
1801 			    (pmap_attr_fetch(pg) & PTE_EXEC)) {
1802 				struct pvo_head *pvoh = vm_page_to_pvoh(pg);
1803 				if (LIST_EMPTY(pvoh)) {
1804 					DPRINTFN(EXEC, ("[pmap_pvo_remove: "
1805 					    "%#" _PRIxpa ": clear-exec]\n",
1806 					    VM_PAGE_TO_PHYS(pg)));
1807 					pmap_attr_clear(pg, PTE_EXEC);
1808 					PMAPCOUNT(exec_uncached_pvo_remove);
1809 				} else {
1810 					DPRINTFN(EXEC, ("[pmap_pvo_remove: "
1811 					    "%#" _PRIxpa ": syncicache]\n",
1812 					    VM_PAGE_TO_PHYS(pg)));
1813 					pmap_syncicache(VM_PAGE_TO_PHYS(pg),
1814 					    PAGE_SIZE);
1815 					PMAPCOUNT(exec_synced_pvo_remove);
1816 				}
1817 			}
1818 
1819 			pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1820 		}
1821 		PMAPCOUNT(unmappings);
1822 	} else {
1823 		PMAPCOUNT(kernel_unmappings);
1824 	}
1825 
1826 	/*
1827 	 * Remove the PVO from its lists and return it to the pool.
1828 	 */
1829 	LIST_REMOVE(pvo, pvo_vlink);
1830 	TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1831 	if (pvol) {
1832 		LIST_INSERT_HEAD(pvol, pvo, pvo_vlink);
1833 	}
1834 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1835 	pmap_pvo_remove_depth--;
1836 #endif
1837 }
1838 
1839 void
1840 pmap_pvo_free(struct pvo_entry *pvo)
1841 {
1842 
1843 	pool_put(PVO_MANAGED_P(pvo) ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo);
1844 }
1845 
1846 void
1847 pmap_pvo_free_list(struct pvo_head *pvol)
1848 {
1849 	struct pvo_entry *pvo, *npvo;
1850 
1851 	for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) {
1852 		npvo = LIST_NEXT(pvo, pvo_vlink);
1853 		LIST_REMOVE(pvo, pvo_vlink);
1854 		pmap_pvo_free(pvo);
1855 	}
1856 }
1857 
1858 /*
1859  * Mark a mapping as executable.
1860  * If this is the first executable mapping in the segment,
1861  * clear the noexec flag.
1862  */
1863 static void
1864 pvo_set_exec(struct pvo_entry *pvo)
1865 {
1866 	struct pmap *pm = pvo->pvo_pmap;
1867 
1868 	if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) {
1869 		return;
1870 	}
1871 	pvo->pvo_vaddr |= PVO_EXECUTABLE;
1872 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1873 	{
1874 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1875 		if (pm->pm_exec[sr]++ == 0) {
1876 			pm->pm_sr[sr] &= ~SR_NOEXEC;
1877 		}
1878 	}
1879 #endif
1880 }
1881 
1882 /*
1883  * Mark a mapping as non-executable.
1884  * If this was the last executable mapping in the segment,
1885  * set the noexec flag.
1886  */
1887 static void
1888 pvo_clear_exec(struct pvo_entry *pvo)
1889 {
1890 	struct pmap *pm = pvo->pvo_pmap;
1891 
1892 	if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) {
1893 		return;
1894 	}
1895 	pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1896 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1897 	{
1898 		int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1899 		if (--pm->pm_exec[sr] == 0) {
1900 			pm->pm_sr[sr] |= SR_NOEXEC;
1901 		}
1902 	}
1903 #endif
1904 }
1905 
1906 /*
1907  * Insert physical page at pa into the given pmap at virtual address va.
1908  */
1909 int
1910 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1911 {
1912 	struct mem_region *mp;
1913 	struct pvo_head *pvo_head;
1914 	struct vm_page *pg;
1915 	struct pool *pl;
1916 	register_t pte_lo;
1917 	int error;
1918 	u_int pvo_flags;
1919 	u_int was_exec = 0;
1920 
1921 	PMAP_LOCK();
1922 
1923 	if (__predict_false(!pmap_initialized)) {
1924 		pvo_head = &pmap_pvo_kunmanaged;
1925 		pl = &pmap_upvo_pool;
1926 		pvo_flags = 0;
1927 		pg = NULL;
1928 		was_exec = PTE_EXEC;
1929 	} else {
1930 		pvo_head = pa_to_pvoh(pa, &pg);
1931 		pl = &pmap_mpvo_pool;
1932 		pvo_flags = PVO_MANAGED;
1933 	}
1934 
1935 	DPRINTFN(ENTER,
1936 	    ("pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1937 	    pm, va, pa, prot, flags));
1938 
1939 	/*
1940 	 * If this is a managed page, and it's the first reference to the
1941 	 * page clear the execness of the page.  Otherwise fetch the execness.
1942 	 */
1943 	if (pg != NULL)
1944 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1945 
1946 	DPRINTFN(ENTER, (" was_exec=%d", was_exec));
1947 
1948 	/*
1949 	 * Assume the page is cache inhibited and access is guarded unless
1950 	 * it's in our available memory array.  If it is in the memory array,
1951 	 * asssume it's in memory coherent memory.
1952 	 */
1953 	pte_lo = PTE_IG;
1954 	if ((flags & PMAP_NC) == 0) {
1955 		for (mp = mem; mp->size; mp++) {
1956 			if (pa >= mp->start && pa < mp->start + mp->size) {
1957 				pte_lo = PTE_M;
1958 				break;
1959 			}
1960 		}
1961 	}
1962 
1963 	if (prot & VM_PROT_WRITE)
1964 		pte_lo |= PTE_BW;
1965 	else
1966 		pte_lo |= PTE_BR;
1967 
1968 	/*
1969 	 * If this was in response to a fault, "pre-fault" the PTE's
1970 	 * changed/referenced bit appropriately.
1971 	 */
1972 	if (flags & VM_PROT_WRITE)
1973 		pte_lo |= PTE_CHG;
1974 	if (flags & VM_PROT_ALL)
1975 		pte_lo |= PTE_REF;
1976 
1977 	/*
1978 	 * We need to know if this page can be executable
1979 	 */
1980 	flags |= (prot & VM_PROT_EXECUTE);
1981 
1982 	/*
1983 	 * Record mapping for later back-translation and pte spilling.
1984 	 * This will overwrite any existing mapping.
1985 	 */
1986 	error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags);
1987 
1988 	/*
1989 	 * Flush the real page from the instruction cache if this page is
1990 	 * mapped executable and cacheable and has not been flushed since
1991 	 * the last time it was modified.
1992 	 */
1993 	if (error == 0 &&
1994             (flags & VM_PROT_EXECUTE) &&
1995             (pte_lo & PTE_I) == 0 &&
1996 	    was_exec == 0) {
1997 		DPRINTFN(ENTER, (" syncicache"));
1998 		PMAPCOUNT(exec_synced);
1999 		pmap_syncicache(pa, PAGE_SIZE);
2000 		if (pg != NULL) {
2001 			pmap_attr_save(pg, PTE_EXEC);
2002 			PMAPCOUNT(exec_cached);
2003 #if defined(DEBUG) || defined(PMAPDEBUG)
2004 			if (pmapdebug & PMAPDEBUG_ENTER)
2005 				printf(" marked-as-exec");
2006 			else if (pmapdebug & PMAPDEBUG_EXEC)
2007 				printf("[pmap_enter: %#" _PRIxpa ": marked-as-exec]\n",
2008 				    VM_PAGE_TO_PHYS(pg));
2009 
2010 #endif
2011 		}
2012 	}
2013 
2014 	DPRINTFN(ENTER, (": error=%d\n", error));
2015 
2016 	PMAP_UNLOCK();
2017 
2018 	return error;
2019 }
2020 
2021 void
2022 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
2023 {
2024 	struct mem_region *mp;
2025 	register_t pte_lo;
2026 	int error;
2027 
2028 #if defined (PMAP_OEA64_BRIDGE)
2029 	if (va < VM_MIN_KERNEL_ADDRESS)
2030 		panic("pmap_kenter_pa: attempt to enter "
2031 		    "non-kernel address %#" _PRIxva "!", va);
2032 #endif
2033 
2034 	DPRINTFN(KENTER,
2035 	    ("pmap_kenter_pa(%#" _PRIxva ",%#" _PRIxpa ",%#x)\n", va, pa, prot));
2036 
2037 	PMAP_LOCK();
2038 
2039 	/*
2040 	 * Assume the page is cache inhibited and access is guarded unless
2041 	 * it's in our available memory array.  If it is in the memory array,
2042 	 * asssume it's in memory coherent memory.
2043 	 */
2044 	pte_lo = PTE_IG;
2045 	if ((prot & PMAP_NC) == 0) {
2046 		for (mp = mem; mp->size; mp++) {
2047 			if (pa >= mp->start && pa < mp->start + mp->size) {
2048 				pte_lo = PTE_M;
2049 				break;
2050 			}
2051 		}
2052 	}
2053 
2054 	if (prot & VM_PROT_WRITE)
2055 		pte_lo |= PTE_BW;
2056 	else
2057 		pte_lo |= PTE_BR;
2058 
2059 	/*
2060 	 * We don't care about REF/CHG on PVOs on the unmanaged list.
2061 	 */
2062 	error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool,
2063 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED);
2064 
2065 	if (error != 0)
2066 		panic("pmap_kenter_pa: failed to enter va %#" _PRIxva " pa %#" _PRIxpa ": %d",
2067 		      va, pa, error);
2068 
2069 	PMAP_UNLOCK();
2070 }
2071 
2072 void
2073 pmap_kremove(vaddr_t va, vsize_t len)
2074 {
2075 	if (va < VM_MIN_KERNEL_ADDRESS)
2076 		panic("pmap_kremove: attempt to remove "
2077 		    "non-kernel address %#" _PRIxva "!", va);
2078 
2079 	DPRINTFN(KREMOVE,("pmap_kremove(%#" _PRIxva ",%#" _PRIxva ")\n", va, len));
2080 	pmap_remove(pmap_kernel(), va, va + len);
2081 }
2082 
2083 /*
2084  * Remove the given range of mapping entries.
2085  */
2086 void
2087 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
2088 {
2089 	struct pvo_head pvol;
2090 	struct pvo_entry *pvo;
2091 	register_t msr;
2092 	int pteidx;
2093 
2094 	PMAP_LOCK();
2095 	LIST_INIT(&pvol);
2096 	msr = pmap_interrupts_off();
2097 	for (; va < endva; va += PAGE_SIZE) {
2098 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2099 		if (pvo != NULL) {
2100 			pmap_pvo_remove(pvo, pteidx, &pvol);
2101 		}
2102 	}
2103 	pmap_interrupts_restore(msr);
2104 	pmap_pvo_free_list(&pvol);
2105 	PMAP_UNLOCK();
2106 }
2107 
2108 /*
2109  * Get the physical page address for the given pmap/virtual address.
2110  */
2111 bool
2112 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
2113 {
2114 	struct pvo_entry *pvo;
2115 	register_t msr;
2116 
2117 	PMAP_LOCK();
2118 
2119 	/*
2120 	 * If this is a kernel pmap lookup, also check the battable
2121 	 * and if we get a hit, translate the VA to a PA using the
2122 	 * BAT entries.  Don't check for VM_MAX_KERNEL_ADDRESS is
2123 	 * that will wrap back to 0.
2124 	 */
2125 	if (pm == pmap_kernel() &&
2126 	    (va < VM_MIN_KERNEL_ADDRESS ||
2127 	     (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
2128 		KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
2129 #if defined (PMAP_OEA)
2130 #ifdef PPC_OEA601
2131 		if ((MFPVR() >> 16) == MPC601) {
2132 			register_t batu = battable[va >> 23].batu;
2133 			register_t batl = battable[va >> 23].batl;
2134 			register_t sr = iosrtable[va >> ADDR_SR_SHFT];
2135 			if (BAT601_VALID_P(batl) &&
2136 			    BAT601_VA_MATCH_P(batu, batl, va)) {
2137 				register_t mask =
2138 				    (~(batl & BAT601_BSM) << 17) & ~0x1ffffL;
2139 				if (pap)
2140 					*pap = (batl & mask) | (va & ~mask);
2141 				PMAP_UNLOCK();
2142 				return true;
2143 			} else if (SR601_VALID_P(sr) &&
2144 				   SR601_PA_MATCH_P(sr, va)) {
2145 				if (pap)
2146 					*pap = va;
2147 				PMAP_UNLOCK();
2148 				return true;
2149 			}
2150 		} else
2151 #endif /* PPC_OEA601 */
2152 		{
2153 			register_t batu = battable[va >> ADDR_SR_SHFT].batu;
2154 			if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) {
2155 				register_t batl =
2156 				    battable[va >> ADDR_SR_SHFT].batl;
2157 				register_t mask =
2158 				    (~(batu & BAT_BL) << 15) & ~0x1ffffL;
2159 				if (pap)
2160 					*pap = (batl & mask) | (va & ~mask);
2161 				PMAP_UNLOCK();
2162 				return true;
2163 			}
2164 		}
2165 		return false;
2166 #elif defined (PMAP_OEA64_BRIDGE)
2167 	if (va >= SEGMENT_LENGTH)
2168 		panic("%s: pm: %s va >= SEGMENT_LENGTH, va: 0x%08lx\n",
2169 		    __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
2170 	else {
2171 		if (pap)
2172 			*pap = va;
2173 			PMAP_UNLOCK();
2174 			return true;
2175 	}
2176 #elif defined (PMAP_OEA64)
2177 #error PPC_OEA64 not supported
2178 #endif /* PPC_OEA */
2179 	}
2180 
2181 	msr = pmap_interrupts_off();
2182 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2183 	if (pvo != NULL) {
2184 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2185 		if (pap)
2186 			*pap = (pvo->pvo_pte.pte_lo & PTE_RPGN)
2187 			    | (va & ADDR_POFF);
2188 	}
2189 	pmap_interrupts_restore(msr);
2190 	PMAP_UNLOCK();
2191 	return pvo != NULL;
2192 }
2193 
2194 /*
2195  * Lower the protection on the specified range of this pmap.
2196  */
2197 void
2198 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
2199 {
2200 	struct pvo_entry *pvo;
2201 	volatile struct pte *pt;
2202 	register_t msr;
2203 	int pteidx;
2204 
2205 	/*
2206 	 * Since this routine only downgrades protection, we should
2207 	 * always be called with at least one bit not set.
2208 	 */
2209 	KASSERT(prot != VM_PROT_ALL);
2210 
2211 	/*
2212 	 * If there is no protection, this is equivalent to
2213 	 * remove the pmap from the pmap.
2214 	 */
2215 	if ((prot & VM_PROT_READ) == 0) {
2216 		pmap_remove(pm, va, endva);
2217 		return;
2218 	}
2219 
2220 	PMAP_LOCK();
2221 
2222 	msr = pmap_interrupts_off();
2223 	for (; va < endva; va += PAGE_SIZE) {
2224 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
2225 		if (pvo == NULL)
2226 			continue;
2227 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2228 
2229 		/*
2230 		 * Revoke executable if asked to do so.
2231 		 */
2232 		if ((prot & VM_PROT_EXECUTE) == 0)
2233 			pvo_clear_exec(pvo);
2234 
2235 #if 0
2236 		/*
2237 		 * If the page is already read-only, no change
2238 		 * needs to be made.
2239 		 */
2240 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
2241 			continue;
2242 #endif
2243 		/*
2244 		 * Grab the PTE pointer before we diddle with
2245 		 * the cached PTE copy.
2246 		 */
2247 		pt = pmap_pvo_to_pte(pvo, pteidx);
2248 		/*
2249 		 * Change the protection of the page.
2250 		 */
2251 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2252 		pvo->pvo_pte.pte_lo |= PTE_BR;
2253 
2254 		/*
2255 		 * If the PVO is in the page table, update
2256 		 * that pte at well.
2257 		 */
2258 		if (pt != NULL) {
2259 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2260 			PVO_WHERE(pvo, PMAP_PROTECT);
2261 			PMAPCOUNT(ptes_changed);
2262 		}
2263 
2264 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2265 	}
2266 	pmap_interrupts_restore(msr);
2267 	PMAP_UNLOCK();
2268 }
2269 
2270 void
2271 pmap_unwire(pmap_t pm, vaddr_t va)
2272 {
2273 	struct pvo_entry *pvo;
2274 	register_t msr;
2275 
2276 	PMAP_LOCK();
2277 	msr = pmap_interrupts_off();
2278 	pvo = pmap_pvo_find_va(pm, va, NULL);
2279 	if (pvo != NULL) {
2280 		if (PVO_WIRED_P(pvo)) {
2281 			pvo->pvo_vaddr &= ~PVO_WIRED;
2282 			pm->pm_stats.wired_count--;
2283 		}
2284 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2285 	}
2286 	pmap_interrupts_restore(msr);
2287 	PMAP_UNLOCK();
2288 }
2289 
2290 /*
2291  * Lower the protection on the specified physical page.
2292  */
2293 void
2294 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
2295 {
2296 	struct pvo_head *pvo_head, pvol;
2297 	struct pvo_entry *pvo, *next_pvo;
2298 	volatile struct pte *pt;
2299 	register_t msr;
2300 
2301 	PMAP_LOCK();
2302 
2303 	KASSERT(prot != VM_PROT_ALL);
2304 	LIST_INIT(&pvol);
2305 	msr = pmap_interrupts_off();
2306 
2307 	/*
2308 	 * When UVM reuses a page, it does a pmap_page_protect with
2309 	 * VM_PROT_NONE.  At that point, we can clear the exec flag
2310 	 * since we know the page will have different contents.
2311 	 */
2312 	if ((prot & VM_PROT_READ) == 0) {
2313 		DPRINTFN(EXEC, ("[pmap_page_protect: %#" _PRIxpa ": clear-exec]\n",
2314 		    VM_PAGE_TO_PHYS(pg)));
2315 		if (pmap_attr_fetch(pg) & PTE_EXEC) {
2316 			PMAPCOUNT(exec_uncached_page_protect);
2317 			pmap_attr_clear(pg, PTE_EXEC);
2318 		}
2319 	}
2320 
2321 	pvo_head = vm_page_to_pvoh(pg);
2322 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2323 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
2324 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2325 
2326 		/*
2327 		 * Downgrading to no mapping at all, we just remove the entry.
2328 		 */
2329 		if ((prot & VM_PROT_READ) == 0) {
2330 			pmap_pvo_remove(pvo, -1, &pvol);
2331 			continue;
2332 		}
2333 
2334 		/*
2335 		 * If EXEC permission is being revoked, just clear the
2336 		 * flag in the PVO.
2337 		 */
2338 		if ((prot & VM_PROT_EXECUTE) == 0)
2339 			pvo_clear_exec(pvo);
2340 
2341 		/*
2342 		 * If this entry is already RO, don't diddle with the
2343 		 * page table.
2344 		 */
2345 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2346 			PMAP_PVO_CHECK(pvo);
2347 			continue;
2348 		}
2349 
2350 		/*
2351 		 * Grab the PTE before the we diddle the bits so
2352 		 * pvo_to_pte can verify the pte contents are as
2353 		 * expected.
2354 		 */
2355 		pt = pmap_pvo_to_pte(pvo, -1);
2356 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2357 		pvo->pvo_pte.pte_lo |= PTE_BR;
2358 		if (pt != NULL) {
2359 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2360 			PVO_WHERE(pvo, PMAP_PAGE_PROTECT);
2361 			PMAPCOUNT(ptes_changed);
2362 		}
2363 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2364 	}
2365 	pmap_interrupts_restore(msr);
2366 	pmap_pvo_free_list(&pvol);
2367 
2368 	PMAP_UNLOCK();
2369 }
2370 
2371 /*
2372  * Activate the address space for the specified process.  If the process
2373  * is the current process, load the new MMU context.
2374  */
2375 void
2376 pmap_activate(struct lwp *l)
2377 {
2378 	struct pcb *pcb = &l->l_addr->u_pcb;
2379 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2380 
2381 	DPRINTFN(ACTIVATE,
2382 	    ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp));
2383 
2384 	/*
2385 	 * XXX Normally performed in cpu_fork().
2386 	 */
2387 	pcb->pcb_pm = pmap;
2388 
2389 	/*
2390 	* In theory, the SR registers need only be valid on return
2391 	* to user space wait to do them there.
2392 	*/
2393 	if (l == curlwp) {
2394 		/* Store pointer to new current pmap. */
2395 		curpm = pmap;
2396 	}
2397 }
2398 
2399 /*
2400  * Deactivate the specified process's address space.
2401  */
2402 void
2403 pmap_deactivate(struct lwp *l)
2404 {
2405 }
2406 
2407 bool
2408 pmap_query_bit(struct vm_page *pg, int ptebit)
2409 {
2410 	struct pvo_entry *pvo;
2411 	volatile struct pte *pt;
2412 	register_t msr;
2413 
2414 	PMAP_LOCK();
2415 
2416 	if (pmap_attr_fetch(pg) & ptebit) {
2417 		PMAP_UNLOCK();
2418 		return true;
2419 	}
2420 
2421 	msr = pmap_interrupts_off();
2422 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2423 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2424 		/*
2425 		 * See if we saved the bit off.  If so cache, it and return
2426 		 * success.
2427 		 */
2428 		if (pvo->pvo_pte.pte_lo & ptebit) {
2429 			pmap_attr_save(pg, ptebit);
2430 			PMAP_PVO_CHECK(pvo);		/* sanity check */
2431 			pmap_interrupts_restore(msr);
2432 			PMAP_UNLOCK();
2433 			return true;
2434 		}
2435 	}
2436 	/*
2437 	 * No luck, now go thru the hard part of looking at the ptes
2438 	 * themselves.  Sync so any pending REF/CHG bits are flushed
2439 	 * to the PTEs.
2440 	 */
2441 	SYNC();
2442 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2443 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2444 		/*
2445 		 * See if this pvo have a valid PTE.  If so, fetch the
2446 		 * REF/CHG bits from the valid PTE.  If the appropriate
2447 		 * ptebit is set, cache, it and return success.
2448 		 */
2449 		pt = pmap_pvo_to_pte(pvo, -1);
2450 		if (pt != NULL) {
2451 			pmap_pte_synch(pt, &pvo->pvo_pte);
2452 			if (pvo->pvo_pte.pte_lo & ptebit) {
2453 				pmap_attr_save(pg, ptebit);
2454 				PMAP_PVO_CHECK(pvo);		/* sanity check */
2455 				pmap_interrupts_restore(msr);
2456 				PMAP_UNLOCK();
2457 				return true;
2458 			}
2459 		}
2460 	}
2461 	pmap_interrupts_restore(msr);
2462 	PMAP_UNLOCK();
2463 	return false;
2464 }
2465 
2466 bool
2467 pmap_clear_bit(struct vm_page *pg, int ptebit)
2468 {
2469 	struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2470 	struct pvo_entry *pvo;
2471 	volatile struct pte *pt;
2472 	register_t msr;
2473 	int rv = 0;
2474 
2475 	PMAP_LOCK();
2476 	msr = pmap_interrupts_off();
2477 
2478 	/*
2479 	 * Fetch the cache value
2480 	 */
2481 	rv |= pmap_attr_fetch(pg);
2482 
2483 	/*
2484 	 * Clear the cached value.
2485 	 */
2486 	pmap_attr_clear(pg, ptebit);
2487 
2488 	/*
2489 	 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2490 	 * can reset the right ones).  Note that since the pvo entries and
2491 	 * list heads are accessed via BAT0 and are never placed in the
2492 	 * page table, we don't have to worry about further accesses setting
2493 	 * the REF/CHG bits.
2494 	 */
2495 	SYNC();
2496 
2497 	/*
2498 	 * For each pvo entry, clear pvo's ptebit.  If this pvo have a
2499 	 * valid PTE.  If so, clear the ptebit from the valid PTE.
2500 	 */
2501 	LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2502 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2503 		pt = pmap_pvo_to_pte(pvo, -1);
2504 		if (pt != NULL) {
2505 			/*
2506 			 * Only sync the PTE if the bit we are looking
2507 			 * for is not already set.
2508 			 */
2509 			if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2510 				pmap_pte_synch(pt, &pvo->pvo_pte);
2511 			/*
2512 			 * If the bit we are looking for was already set,
2513 			 * clear that bit in the pte.
2514 			 */
2515 			if (pvo->pvo_pte.pte_lo & ptebit)
2516 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2517 		}
2518 		rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2519 		pvo->pvo_pte.pte_lo &= ~ptebit;
2520 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2521 	}
2522 	pmap_interrupts_restore(msr);
2523 
2524 	/*
2525 	 * If we are clearing the modify bit and this page was marked EXEC
2526 	 * and the user of the page thinks the page was modified, then we
2527 	 * need to clean it from the icache if it's mapped or clear the EXEC
2528 	 * bit if it's not mapped.  The page itself might not have the CHG
2529 	 * bit set if the modification was done via DMA to the page.
2530 	 */
2531 	if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2532 		if (LIST_EMPTY(pvoh)) {
2533 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#" _PRIxpa ": clear-exec]\n",
2534 			    VM_PAGE_TO_PHYS(pg)));
2535 			pmap_attr_clear(pg, PTE_EXEC);
2536 			PMAPCOUNT(exec_uncached_clear_modify);
2537 		} else {
2538 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#" _PRIxpa ": syncicache]\n",
2539 			    VM_PAGE_TO_PHYS(pg)));
2540 			pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE);
2541 			PMAPCOUNT(exec_synced_clear_modify);
2542 		}
2543 	}
2544 	PMAP_UNLOCK();
2545 	return (rv & ptebit) != 0;
2546 }
2547 
2548 void
2549 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2550 {
2551 	struct pvo_entry *pvo;
2552 	size_t offset = va & ADDR_POFF;
2553 	int s;
2554 
2555 	PMAP_LOCK();
2556 	s = splvm();
2557 	while (len > 0) {
2558 		size_t seglen = PAGE_SIZE - offset;
2559 		if (seglen > len)
2560 			seglen = len;
2561 		pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2562 		if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) {
2563 			pmap_syncicache(
2564 			    (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2565 			PMAP_PVO_CHECK(pvo);
2566 		}
2567 		va += seglen;
2568 		len -= seglen;
2569 		offset = 0;
2570 	}
2571 	splx(s);
2572 	PMAP_UNLOCK();
2573 }
2574 
2575 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2576 void
2577 pmap_pte_print(volatile struct pte *pt)
2578 {
2579 	printf("PTE %p: ", pt);
2580 
2581 #if defined(PMAP_OEA)
2582 	/* High word: */
2583 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2584 #else
2585 	printf("%#" _PRIxpte ": [", pt->pte_hi);
2586 #endif /* PMAP_OEA */
2587 
2588 	printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2589 	printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2590 
2591 	printf("%#" _PRIxpte " %#" _PRIxpte "",
2592 	    (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2593 	    pt->pte_hi & PTE_API);
2594 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2595 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2596 #else
2597 	printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2598 #endif /* PMAP_OEA */
2599 
2600 	/* Low word: */
2601 #if defined (PMAP_OEA)
2602 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2603 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2604 #else
2605 	printf(" %#" _PRIxpte ": [", pt->pte_lo);
2606 	printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2607 #endif
2608 	printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2609 	printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2610 	printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2611 	printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2612 	printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2613 	printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2614 	switch (pt->pte_lo & PTE_PP) {
2615 	case PTE_BR: printf("br]\n"); break;
2616 	case PTE_BW: printf("bw]\n"); break;
2617 	case PTE_SO: printf("so]\n"); break;
2618 	case PTE_SW: printf("sw]\n"); break;
2619 	}
2620 }
2621 #endif
2622 
2623 #if defined(DDB)
2624 void
2625 pmap_pteg_check(void)
2626 {
2627 	volatile struct pte *pt;
2628 	int i;
2629 	int ptegidx;
2630 	u_int p_valid = 0;
2631 	u_int s_valid = 0;
2632 	u_int invalid = 0;
2633 
2634 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2635 		for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2636 			if (pt->pte_hi & PTE_VALID) {
2637 				if (pt->pte_hi & PTE_HID)
2638 					s_valid++;
2639 				else
2640 				{
2641 					p_valid++;
2642 				}
2643 			} else
2644 				invalid++;
2645 		}
2646 	}
2647 	printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2648 		p_valid, p_valid, s_valid, s_valid,
2649 		invalid, invalid);
2650 }
2651 
2652 void
2653 pmap_print_mmuregs(void)
2654 {
2655 	int i;
2656 	u_int cpuvers;
2657 #ifndef PMAP_OEA64
2658 	vaddr_t addr;
2659 	register_t soft_sr[16];
2660 #endif
2661 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2662 	struct bat soft_ibat[4];
2663 	struct bat soft_dbat[4];
2664 #endif
2665 	paddr_t sdr1;
2666 
2667 	cpuvers = MFPVR() >> 16;
2668 	__asm volatile ("mfsdr1 %0" : "=r"(sdr1));
2669 #ifndef PMAP_OEA64
2670 	addr = 0;
2671 	for (i = 0; i < 16; i++) {
2672 		soft_sr[i] = MFSRIN(addr);
2673 		addr += (1 << ADDR_SR_SHFT);
2674 	}
2675 #endif
2676 
2677 #if defined (PMAP_OEA) || defined (PMAP_OEA_BRIDGE)
2678 	/* read iBAT (601: uBAT) registers */
2679 	__asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2680 	__asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2681 	__asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2682 	__asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2683 	__asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2684 	__asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2685 	__asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2686 	__asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2687 
2688 
2689 	if (cpuvers != MPC601) {
2690 		/* read dBAT registers */
2691 		__asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2692 		__asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2693 		__asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2694 		__asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2695 		__asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2696 		__asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2697 		__asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2698 		__asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2699 	}
2700 #endif
2701 
2702 	printf("SDR1:\t%#" _PRIxpa "\n", sdr1);
2703 #ifndef PMAP_OEA64
2704 	printf("SR[]:\t");
2705 	for (i = 0; i < 4; i++)
2706 		printf("0x%08lx,   ", soft_sr[i]);
2707 	printf("\n\t");
2708 	for ( ; i < 8; i++)
2709 		printf("0x%08lx,   ", soft_sr[i]);
2710 	printf("\n\t");
2711 	for ( ; i < 12; i++)
2712 		printf("0x%08lx,   ", soft_sr[i]);
2713 	printf("\n\t");
2714 	for ( ; i < 16; i++)
2715 		printf("0x%08lx,   ", soft_sr[i]);
2716 	printf("\n");
2717 #endif
2718 
2719 #if defined(PMAP_OEA) || defined(PMAP_OEA_BRIDGE)
2720 	printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2721 	for (i = 0; i < 4; i++) {
2722 		printf("0x%08lx 0x%08lx, ",
2723 			soft_ibat[i].batu, soft_ibat[i].batl);
2724 		if (i == 1)
2725 			printf("\n\t");
2726 	}
2727 	if (cpuvers != MPC601) {
2728 		printf("\ndBAT[]:\t");
2729 		for (i = 0; i < 4; i++) {
2730 			printf("0x%08lx 0x%08lx, ",
2731 				soft_dbat[i].batu, soft_dbat[i].batl);
2732 			if (i == 1)
2733 				printf("\n\t");
2734 		}
2735 	}
2736 	printf("\n");
2737 #endif /* PMAP_OEA... */
2738 }
2739 
2740 void
2741 pmap_print_pte(pmap_t pm, vaddr_t va)
2742 {
2743 	struct pvo_entry *pvo;
2744 	volatile struct pte *pt;
2745 	int pteidx;
2746 
2747 	pvo = pmap_pvo_find_va(pm, va, &pteidx);
2748 	if (pvo != NULL) {
2749 		pt = pmap_pvo_to_pte(pvo, pteidx);
2750 		if (pt != NULL) {
2751 			printf("VA %#" _PRIxva " -> %p -> %s %#" _PRIxpte ", %#" _PRIxpte "\n",
2752 				va, pt,
2753 				pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2754 				pt->pte_hi, pt->pte_lo);
2755 		} else {
2756 			printf("No valid PTE found\n");
2757 		}
2758 	} else {
2759 		printf("Address not in pmap\n");
2760 	}
2761 }
2762 
2763 void
2764 pmap_pteg_dist(void)
2765 {
2766 	struct pvo_entry *pvo;
2767 	int ptegidx;
2768 	int depth;
2769 	int max_depth = 0;
2770 	unsigned int depths[64];
2771 
2772 	memset(depths, 0, sizeof(depths));
2773 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2774 		depth = 0;
2775 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2776 			depth++;
2777 		}
2778 		if (depth > max_depth)
2779 			max_depth = depth;
2780 		if (depth > 63)
2781 			depth = 63;
2782 		depths[depth]++;
2783 	}
2784 
2785 	for (depth = 0; depth < 64; depth++) {
2786 		printf("  [%2d]: %8u", depth, depths[depth]);
2787 		if ((depth & 3) == 3)
2788 			printf("\n");
2789 		if (depth == max_depth)
2790 			break;
2791 	}
2792 	if ((depth & 3) != 3)
2793 		printf("\n");
2794 	printf("Max depth found was %d\n", max_depth);
2795 }
2796 #endif /* DEBUG */
2797 
2798 #if defined(PMAPCHECK) || defined(DEBUG)
2799 void
2800 pmap_pvo_verify(void)
2801 {
2802 	int ptegidx;
2803 	int s;
2804 
2805 	s = splvm();
2806 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2807 		struct pvo_entry *pvo;
2808 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2809 			if ((uintptr_t) pvo >= SEGMENT_LENGTH)
2810 				panic("pmap_pvo_verify: invalid pvo %p "
2811 				    "on list %#x", pvo, ptegidx);
2812 			pmap_pvo_check(pvo);
2813 		}
2814 	}
2815 	splx(s);
2816 }
2817 #endif /* PMAPCHECK */
2818 
2819 
2820 void *
2821 pmap_pool_ualloc(struct pool *pp, int flags)
2822 {
2823 	struct pvo_page *pvop;
2824 
2825 	if (uvm.page_init_done != true) {
2826 		return (void *) uvm_pageboot_alloc(PAGE_SIZE);
2827 	}
2828 
2829 	PMAP_LOCK();
2830 	pvop = SIMPLEQ_FIRST(&pmap_upvop_head);
2831 	if (pvop != NULL) {
2832 		pmap_upvop_free--;
2833 		SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link);
2834 		PMAP_UNLOCK();
2835 		return pvop;
2836 	}
2837 	PMAP_UNLOCK();
2838 	return pmap_pool_malloc(pp, flags);
2839 }
2840 
2841 void *
2842 pmap_pool_malloc(struct pool *pp, int flags)
2843 {
2844 	struct pvo_page *pvop;
2845 	struct vm_page *pg;
2846 
2847 	PMAP_LOCK();
2848 	pvop = SIMPLEQ_FIRST(&pmap_mpvop_head);
2849 	if (pvop != NULL) {
2850 		pmap_mpvop_free--;
2851 		SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link);
2852 		PMAP_UNLOCK();
2853 		return pvop;
2854 	}
2855 	PMAP_UNLOCK();
2856  again:
2857 	pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE,
2858 	    UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256);
2859 	if (__predict_false(pg == NULL)) {
2860 		if (flags & PR_WAITOK) {
2861 			uvm_wait("plpg");
2862 			goto again;
2863 		} else {
2864 			return (0);
2865 		}
2866 	}
2867 	KDASSERT(VM_PAGE_TO_PHYS(pg) == (uintptr_t)VM_PAGE_TO_PHYS(pg));
2868 	return (void *)(uintptr_t) VM_PAGE_TO_PHYS(pg);
2869 }
2870 
2871 void
2872 pmap_pool_ufree(struct pool *pp, void *va)
2873 {
2874 	struct pvo_page *pvop;
2875 #if 0
2876 	if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) {
2877 		pmap_pool_mfree(va, size, tag);
2878 		return;
2879 	}
2880 #endif
2881 	PMAP_LOCK();
2882 	pvop = va;
2883 	SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link);
2884 	pmap_upvop_free++;
2885 	if (pmap_upvop_free > pmap_upvop_maxfree)
2886 		pmap_upvop_maxfree = pmap_upvop_free;
2887 	PMAP_UNLOCK();
2888 }
2889 
2890 void
2891 pmap_pool_mfree(struct pool *pp, void *va)
2892 {
2893 	struct pvo_page *pvop;
2894 
2895 	PMAP_LOCK();
2896 	pvop = va;
2897 	SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link);
2898 	pmap_mpvop_free++;
2899 	if (pmap_mpvop_free > pmap_mpvop_maxfree)
2900 		pmap_mpvop_maxfree = pmap_mpvop_free;
2901 	PMAP_UNLOCK();
2902 #if 0
2903 	uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va));
2904 #endif
2905 }
2906 
2907 /*
2908  * This routine in bootstraping to steal to-be-managed memory (which will
2909  * then be unmanaged).  We use it to grab from the first 256MB for our
2910  * pmap needs and above 256MB for other stuff.
2911  */
2912 vaddr_t
2913 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2914 {
2915 	vsize_t size;
2916 	vaddr_t va;
2917 	paddr_t pa = 0;
2918 	int npgs, bank;
2919 	struct vm_physseg *ps;
2920 
2921 	if (uvm.page_init_done == true)
2922 		panic("pmap_steal_memory: called _after_ bootstrap");
2923 
2924 	*vstartp = VM_MIN_KERNEL_ADDRESS;
2925 	*vendp = VM_MAX_KERNEL_ADDRESS;
2926 
2927 	size = round_page(vsize);
2928 	npgs = atop(size);
2929 
2930 	/*
2931 	 * PA 0 will never be among those given to UVM so we can use it
2932 	 * to indicate we couldn't steal any memory.
2933 	 */
2934 	for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) {
2935 		if (ps->free_list == VM_FREELIST_FIRST256 &&
2936 		    ps->avail_end - ps->avail_start >= npgs) {
2937 			pa = ptoa(ps->avail_start);
2938 			break;
2939 		}
2940 	}
2941 
2942 	if (pa == 0)
2943 		panic("pmap_steal_memory: no approriate memory to steal!");
2944 
2945 	ps->avail_start += npgs;
2946 	ps->start += npgs;
2947 
2948 	/*
2949 	 * If we've used up all the pages in the segment, remove it and
2950 	 * compact the list.
2951 	 */
2952 	if (ps->avail_start == ps->end) {
2953 		/*
2954 		 * If this was the last one, then a very bad thing has occurred
2955 		 */
2956 		if (--vm_nphysseg == 0)
2957 			panic("pmap_steal_memory: out of memory!");
2958 
2959 		printf("pmap_steal_memory: consumed bank %d\n", bank);
2960 		for (; bank < vm_nphysseg; bank++, ps++) {
2961 			ps[0] = ps[1];
2962 		}
2963 	}
2964 
2965 	va = (vaddr_t) pa;
2966 	memset((void *) va, 0, size);
2967 	pmap_pages_stolen += npgs;
2968 #ifdef DEBUG
2969 	if (pmapdebug && npgs > 1) {
2970 		u_int cnt = 0;
2971 		for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++)
2972 			cnt += ps->avail_end - ps->avail_start;
2973 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2974 		    npgs, pmap_pages_stolen, cnt);
2975 	}
2976 #endif
2977 
2978 	return va;
2979 }
2980 
2981 /*
2982  * Find a chuck of memory with right size and alignment.
2983  */
2984 paddr_t
2985 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2986 {
2987 	struct mem_region *mp;
2988 	paddr_t s, e;
2989 	int i, j;
2990 
2991 	size = round_page(size);
2992 
2993 	DPRINTFN(BOOT,
2994 	    ("pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
2995 	    size, alignment, at_end));
2996 
2997 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2998 		panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
2999 		    alignment);
3000 
3001 	if (at_end) {
3002 		if (alignment != PAGE_SIZE)
3003 			panic("pmap_boot_find_memory: invalid ending "
3004 			    "alignment %#" _PRIxpa, alignment);
3005 
3006 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
3007 			s = mp->start + mp->size - size;
3008 			if (s >= mp->start && mp->size >= size) {
3009 				DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
3010 				DPRINTFN(BOOT,
3011 				    ("pmap_boot_find_memory: b-avail[%d] start "
3012 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3013 				     mp->start, mp->size));
3014 				mp->size -= size;
3015 				DPRINTFN(BOOT,
3016 				    ("pmap_boot_find_memory: a-avail[%d] start "
3017 				     "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3018 				     mp->start, mp->size));
3019 				return s;
3020 			}
3021 		}
3022 		panic("pmap_boot_find_memory: no available memory");
3023 	}
3024 
3025 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3026 		s = (mp->start + alignment - 1) & ~(alignment-1);
3027 		e = s + size;
3028 
3029 		/*
3030 		 * Is the calculated region entirely within the region?
3031 		 */
3032 		if (s < mp->start || e > mp->start + mp->size)
3033 			continue;
3034 
3035 		DPRINTFN(BOOT,(": %#" _PRIxpa "\n", s));
3036 		if (s == mp->start) {
3037 			/*
3038 			 * If the block starts at the beginning of region,
3039 			 * adjust the size & start. (the region may now be
3040 			 * zero in length)
3041 			 */
3042 			DPRINTFN(BOOT,
3043 			    ("pmap_boot_find_memory: b-avail[%d] start "
3044 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3045 			mp->start += size;
3046 			mp->size -= size;
3047 			DPRINTFN(BOOT,
3048 			    ("pmap_boot_find_memory: a-avail[%d] start "
3049 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3050 		} else if (e == mp->start + mp->size) {
3051 			/*
3052 			 * If the block starts at the beginning of region,
3053 			 * adjust only the size.
3054 			 */
3055 			DPRINTFN(BOOT,
3056 			    ("pmap_boot_find_memory: b-avail[%d] start "
3057 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3058 			mp->size -= size;
3059 			DPRINTFN(BOOT,
3060 			    ("pmap_boot_find_memory: a-avail[%d] start "
3061 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3062 		} else {
3063 			/*
3064 			 * Block is in the middle of the region, so we
3065 			 * have to split it in two.
3066 			 */
3067 			for (j = avail_cnt; j > i + 1; j--) {
3068 				avail[j] = avail[j-1];
3069 			}
3070 			DPRINTFN(BOOT,
3071 			    ("pmap_boot_find_memory: b-avail[%d] start "
3072 			     "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size));
3073 			mp[1].start = e;
3074 			mp[1].size = mp[0].start + mp[0].size - e;
3075 			mp[0].size = s - mp[0].start;
3076 			avail_cnt++;
3077 			for (; i < avail_cnt; i++) {
3078 				DPRINTFN(BOOT,
3079 				    ("pmap_boot_find_memory: a-avail[%d] "
3080 				     "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3081 				     avail[i].start, avail[i].size));
3082 			}
3083 		}
3084 		KASSERT(s == (uintptr_t) s);
3085 		return s;
3086 	}
3087 	panic("pmap_boot_find_memory: not enough memory for "
3088 	    "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3089 }
3090 
3091 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3092 #if defined (PMAP_OEA64_BRIDGE)
3093 int
3094 pmap_setup_segment0_map(int use_large_pages, ...)
3095 {
3096     vaddr_t va;
3097 
3098     register_t pte_lo = 0x0;
3099     int ptegidx = 0, i = 0;
3100     struct pte pte;
3101     va_list ap;
3102 
3103     /* Coherent + Supervisor RW, no user access */
3104     pte_lo = PTE_M;
3105 
3106     /* XXXSL
3107      * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3108      * these have to take priority.
3109      */
3110     for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3111         ptegidx = va_to_pteg(pmap_kernel(), va);
3112         pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3113         i = pmap_pte_insert(ptegidx, &pte);
3114     }
3115 
3116     va_start(ap, use_large_pages);
3117     while (1) {
3118         paddr_t pa;
3119         size_t size;
3120 
3121         va = va_arg(ap, vaddr_t);
3122 
3123         if (va == 0)
3124             break;
3125 
3126         pa = va_arg(ap, paddr_t);
3127         size = va_arg(ap, size_t);
3128 
3129         for (; va < (va + size); va += 0x1000, pa += 0x1000) {
3130 #if 0
3131 	    printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__,  va, pa);
3132 #endif
3133             ptegidx = va_to_pteg(pmap_kernel(), va);
3134             pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3135             i = pmap_pte_insert(ptegidx, &pte);
3136         }
3137     }
3138 
3139     TLBSYNC();
3140     SYNC();
3141     return (0);
3142 }
3143 #endif /* PMAP_OEA64_BRIDGE */
3144 
3145 /*
3146  * This is not part of the defined PMAP interface and is specific to the
3147  * PowerPC architecture.  This is called during initppc, before the system
3148  * is really initialized.
3149  */
3150 void
3151 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3152 {
3153 	struct mem_region *mp, tmp;
3154 	paddr_t s, e;
3155 	psize_t size;
3156 	int i, j;
3157 
3158 	/*
3159 	 * Get memory.
3160 	 */
3161 	mem_regions(&mem, &avail);
3162 #if defined(DEBUG)
3163 	if (pmapdebug & PMAPDEBUG_BOOT) {
3164 		printf("pmap_bootstrap: memory configuration:\n");
3165 		for (mp = mem; mp->size; mp++) {
3166 			printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3167 				mp->start, mp->size);
3168 		}
3169 		for (mp = avail; mp->size; mp++) {
3170 			printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3171 				mp->start, mp->size);
3172 		}
3173 	}
3174 #endif
3175 
3176 	/*
3177 	 * Find out how much physical memory we have and in how many chunks.
3178 	 */
3179 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3180 		if (mp->start >= pmap_memlimit)
3181 			continue;
3182 		if (mp->start + mp->size > pmap_memlimit) {
3183 			size = pmap_memlimit - mp->start;
3184 			physmem += btoc(size);
3185 		} else {
3186 			physmem += btoc(mp->size);
3187 		}
3188 		mem_cnt++;
3189 	}
3190 
3191 	/*
3192 	 * Count the number of available entries.
3193 	 */
3194 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
3195 		avail_cnt++;
3196 
3197 	/*
3198 	 * Page align all regions.
3199 	 */
3200 	kernelstart = trunc_page(kernelstart);
3201 	kernelend = round_page(kernelend);
3202 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3203 		s = round_page(mp->start);
3204 		mp->size -= (s - mp->start);
3205 		mp->size = trunc_page(mp->size);
3206 		mp->start = s;
3207 		e = mp->start + mp->size;
3208 
3209 		DPRINTFN(BOOT,
3210 		    ("pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3211 		    i, mp->start, mp->size));
3212 
3213 		/*
3214 		 * Don't allow the end to run beyond our artificial limit
3215 		 */
3216 		if (e > pmap_memlimit)
3217 			e = pmap_memlimit;
3218 
3219 		/*
3220 		 * Is this region empty or strange?  skip it.
3221 		 */
3222 		if (e <= s) {
3223 			mp->start = 0;
3224 			mp->size = 0;
3225 			continue;
3226 		}
3227 
3228 		/*
3229 		 * Does this overlap the beginning of kernel?
3230 		 *   Does extend past the end of the kernel?
3231 		 */
3232 		else if (s < kernelstart && e > kernelstart) {
3233 			if (e > kernelend) {
3234 				avail[avail_cnt].start = kernelend;
3235 				avail[avail_cnt].size = e - kernelend;
3236 				avail_cnt++;
3237 			}
3238 			mp->size = kernelstart - s;
3239 		}
3240 		/*
3241 		 * Check whether this region overlaps the end of the kernel.
3242 		 */
3243 		else if (s < kernelend && e > kernelend) {
3244 			mp->start = kernelend;
3245 			mp->size = e - kernelend;
3246 		}
3247 		/*
3248 		 * Look whether this regions is completely inside the kernel.
3249 		 * Nuke it if it does.
3250 		 */
3251 		else if (s >= kernelstart && e <= kernelend) {
3252 			mp->start = 0;
3253 			mp->size = 0;
3254 		}
3255 		/*
3256 		 * If the user imposed a memory limit, enforce it.
3257 		 */
3258 		else if (s >= pmap_memlimit) {
3259 			mp->start = -PAGE_SIZE;	/* let's know why */
3260 			mp->size = 0;
3261 		}
3262 		else {
3263 			mp->start = s;
3264 			mp->size = e - s;
3265 		}
3266 		DPRINTFN(BOOT,
3267 		    ("pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3268 		    i, mp->start, mp->size));
3269 	}
3270 
3271 	/*
3272 	 * Move (and uncount) all the null return to the end.
3273 	 */
3274 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3275 		if (mp->size == 0) {
3276 			tmp = avail[i];
3277 			avail[i] = avail[--avail_cnt];
3278 			avail[avail_cnt] = avail[i];
3279 		}
3280 	}
3281 
3282 	/*
3283 	 * (Bubble)sort them into ascending order.
3284 	 */
3285 	for (i = 0; i < avail_cnt; i++) {
3286 		for (j = i + 1; j < avail_cnt; j++) {
3287 			if (avail[i].start > avail[j].start) {
3288 				tmp = avail[i];
3289 				avail[i] = avail[j];
3290 				avail[j] = tmp;
3291 			}
3292 		}
3293 	}
3294 
3295 	/*
3296 	 * Make sure they don't overlap.
3297 	 */
3298 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3299 		if (mp[0].start + mp[0].size > mp[1].start) {
3300 			mp[0].size = mp[1].start - mp[0].start;
3301 		}
3302 		DPRINTFN(BOOT,
3303 		    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3304 		    i, mp->start, mp->size));
3305 	}
3306 	DPRINTFN(BOOT,
3307 	    ("pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3308 	    i, mp->start, mp->size));
3309 
3310 #ifdef	PTEGCOUNT
3311 	pmap_pteg_cnt = PTEGCOUNT;
3312 #else /* PTEGCOUNT */
3313 
3314 	pmap_pteg_cnt = 0x1000;
3315 
3316 	while (pmap_pteg_cnt < physmem)
3317 		pmap_pteg_cnt <<= 1;
3318 
3319 	pmap_pteg_cnt >>= 1;
3320 #endif /* PTEGCOUNT */
3321 
3322 #ifdef DEBUG
3323 	DPRINTFN(BOOT,
3324 		("pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt));
3325 #endif
3326 
3327 	/*
3328 	 * Find suitably aligned memory for PTEG hash table.
3329 	 */
3330 	size = pmap_pteg_cnt * sizeof(struct pteg);
3331 	pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3332 
3333 #ifdef DEBUG
3334 	DPRINTFN(BOOT,
3335 		("PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table));
3336 #endif
3337 
3338 
3339 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3340 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
3341 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > 256MB",
3342 		    pmap_pteg_table, size);
3343 #endif
3344 
3345 	memset(__UNVOLATILE(pmap_pteg_table), 0,
3346 		pmap_pteg_cnt * sizeof(struct pteg));
3347 	pmap_pteg_mask = pmap_pteg_cnt - 1;
3348 
3349 	/*
3350 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3351 	 * with pages.  So we just steal them before giving them to UVM.
3352 	 */
3353 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3354 	pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3355 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3356 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
3357 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > 256MB",
3358 		    pmap_pvo_table, size);
3359 #endif
3360 
3361 	for (i = 0; i < pmap_pteg_cnt; i++)
3362 		TAILQ_INIT(&pmap_pvo_table[i]);
3363 
3364 #ifndef MSGBUFADDR
3365 	/*
3366 	 * Allocate msgbuf in high memory.
3367 	 */
3368 	msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3369 #endif
3370 
3371 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3372 		paddr_t pfstart = atop(mp->start);
3373 		paddr_t pfend = atop(mp->start + mp->size);
3374 		if (mp->size == 0)
3375 			continue;
3376 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
3377 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3378 				VM_FREELIST_FIRST256);
3379 		} else if (mp->start >= SEGMENT_LENGTH) {
3380 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3381 				VM_FREELIST_DEFAULT);
3382 		} else {
3383 			pfend = atop(SEGMENT_LENGTH);
3384 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3385 				VM_FREELIST_FIRST256);
3386 			pfstart = atop(SEGMENT_LENGTH);
3387 			pfend = atop(mp->start + mp->size);
3388 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
3389 				VM_FREELIST_DEFAULT);
3390 		}
3391 	}
3392 
3393 	/*
3394 	 * Make sure kernel vsid is allocated as well as VSID 0.
3395 	 */
3396 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3397 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3398 	pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3399 		|= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3400 	pmap_vsid_bitmap[0] |= 1;
3401 
3402 	/*
3403 	 * Initialize kernel pmap and hardware.
3404 	 */
3405 
3406 /* PMAP_OEA64_BRIDGE does support these instructions */
3407 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
3408 	for (i = 0; i < 16; i++) {
3409  		pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3410 		__asm volatile ("mtsrin %0,%1"
3411  			      :: "r"(KERNELN_SEGMENT(i)|SR_PRKEY), "r"(i << ADDR_SR_SHFT));
3412 	}
3413 
3414 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3415 	__asm volatile ("mtsr %0,%1"
3416 		      :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
3417 #ifdef KERNEL2_SR
3418 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3419 	__asm volatile ("mtsr %0,%1"
3420 		      :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
3421 #endif
3422 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3423 #if defined (PMAP_OEA)
3424 	for (i = 0; i < 16; i++) {
3425 		if (iosrtable[i] & SR601_T) {
3426 			pmap_kernel()->pm_sr[i] = iosrtable[i];
3427 			__asm volatile ("mtsrin %0,%1"
3428 			    :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT));
3429 		}
3430 	}
3431 	__asm volatile ("sync; mtsdr1 %0; isync"
3432 		      :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10)));
3433 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
3434  	__asm __volatile ("sync; mtsdr1 %0; isync"
3435  		      :: "r"((uintptr_t)pmap_pteg_table | (32 - cntlzw(pmap_pteg_mask >> 11))));
3436 #endif
3437 	tlbia();
3438 
3439 #ifdef ALTIVEC
3440 	pmap_use_altivec = cpu_altivec;
3441 #endif
3442 
3443 #ifdef DEBUG
3444 	if (pmapdebug & PMAPDEBUG_BOOT) {
3445 		u_int cnt;
3446 		int bank;
3447 		char pbuf[9];
3448 		for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) {
3449 			cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start;
3450 			printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3451 			    bank,
3452 			    ptoa(vm_physmem[bank].avail_start),
3453 			    ptoa(vm_physmem[bank].avail_end),
3454 			    ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start));
3455 		}
3456 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3457 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3458 		    pbuf, cnt);
3459 	}
3460 #endif
3461 
3462 	pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry),
3463 	    sizeof(struct pvo_entry), 0, 0, "pmap_upvopl",
3464 	    &pmap_pool_uallocator, IPL_VM);
3465 
3466 	pool_setlowat(&pmap_upvo_pool, 252);
3467 
3468 	pool_init(&pmap_pool, sizeof(struct pmap),
3469 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator,
3470 	    IPL_NONE);
3471 
3472 #if defined(PMAP_NEED_MAPKERNEL) || 1
3473 	{
3474 		struct pmap *pm = pmap_kernel();
3475 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3476 		extern int etext[], kernel_text[];
3477 		vaddr_t va, va_etext = (paddr_t) etext;
3478 #endif
3479 		paddr_t pa, pa_end;
3480 		register_t sr;
3481 		struct pte pt;
3482 		unsigned int ptegidx;
3483 		int bank;
3484 
3485 		sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3486 		pm->pm_sr[0] = sr;
3487 
3488 		for (bank = 0; bank < vm_nphysseg; bank++) {
3489 			pa_end = ptoa(vm_physmem[bank].avail_end);
3490 			pa = ptoa(vm_physmem[bank].avail_start);
3491 			for (; pa < pa_end; pa += PAGE_SIZE) {
3492 				ptegidx = va_to_pteg(pm, pa);
3493 				pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3494 				pmap_pte_insert(ptegidx, &pt);
3495 			}
3496 		}
3497 
3498 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3499 		va = (vaddr_t) kernel_text;
3500 
3501 		for (pa = kernelstart; va < va_etext;
3502 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3503 			ptegidx = va_to_pteg(pm, va);
3504 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3505 			pmap_pte_insert(ptegidx, &pt);
3506 		}
3507 
3508 		for (; pa < kernelend;
3509 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3510 			ptegidx = va_to_pteg(pm, va);
3511 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3512 			pmap_pte_insert(ptegidx, &pt);
3513 		}
3514 
3515 		for (va = 0, pa = 0; va < kernelstart;
3516 		     pa += PAGE_SIZE, va += PAGE_SIZE) {
3517 			ptegidx = va_to_pteg(pm, va);
3518 			if (va < 0x3000)
3519 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3520 			else
3521 				pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3522 			pmap_pte_insert(ptegidx, &pt);
3523 		}
3524 		for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3525 		    pa += PAGE_SIZE, va += PAGE_SIZE) {
3526 			ptegidx = va_to_pteg(pm, va);
3527 			pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3528 			pmap_pte_insert(ptegidx, &pt);
3529 		}
3530 #endif
3531 
3532 		__asm volatile ("mtsrin %0,%1"
3533  			      :: "r"(sr), "r"(kernelstart));
3534 	}
3535 #endif
3536 }
3537