xref: /netbsd-src/sys/arch/powerpc/oea/pmap.c (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /*	$NetBSD: pmap.c,v 1.10 2003/05/10 21:10:37 thorpej 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  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *        This product includes software developed by the NetBSD
20  *        Foundation, Inc. and its contributors.
21  * 4. Neither the name of The NetBSD Foundation nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (C) 1995, 1996 Wolfgang Solfrank.
40  * Copyright (C) 1995, 1996 TooLs GmbH.
41  * All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by TooLs GmbH.
54  * 4. The name of TooLs GmbH may not be used to endorse or promote products
55  *    derived from this software without specific prior written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
58  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
59  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
60  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
61  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
62  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
63  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
64  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
65  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
66  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 #include "opt_altivec.h"
70 #include "opt_pmap.h"
71 #include <sys/param.h>
72 #include <sys/malloc.h>
73 #include <sys/proc.h>
74 #include <sys/user.h>
75 #include <sys/pool.h>
76 #include <sys/queue.h>
77 #include <sys/device.h>		/* for evcnt */
78 #include <sys/systm.h>
79 
80 #if __NetBSD_Version__ < 105010000
81 #include <vm/vm.h>
82 #include <vm/vm_kern.h>
83 #define	splvm()		splimp()
84 #endif
85 
86 #include <uvm/uvm.h>
87 
88 #include <machine/pcb.h>
89 #include <machine/powerpc.h>
90 #include <powerpc/spr.h>
91 #include <powerpc/oea/sr_601.h>
92 #if __NetBSD_Version__ > 105010000
93 #include <powerpc/oea/bat.h>
94 #else
95 #include <powerpc/bat.h>
96 #endif
97 
98 #if defined(DEBUG) || defined(PMAPCHECK)
99 #define	STATIC
100 #else
101 #define	STATIC	static
102 #endif
103 
104 #ifdef ALTIVEC
105 int pmap_use_altivec;
106 #endif
107 
108 volatile struct pteg *pmap_pteg_table;
109 unsigned int pmap_pteg_cnt;
110 unsigned int pmap_pteg_mask;
111 paddr_t pmap_memlimit = -PAGE_SIZE;		/* there is no limit */
112 
113 struct pmap kernel_pmap_;
114 unsigned int pmap_pages_stolen;
115 u_long pmap_pte_valid;
116 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
117 u_long pmap_pvo_enter_depth;
118 u_long pmap_pvo_remove_depth;
119 #endif
120 
121 int physmem;
122 #ifndef MSGBUFADDR
123 extern paddr_t msgbuf_paddr;
124 #endif
125 
126 static struct mem_region *mem, *avail;
127 static u_int mem_cnt, avail_cnt;
128 
129 #ifdef __HAVE_PMAP_PHYSSEG
130 /*
131  * This is a cache of referenced/modified bits.
132  * Bits herein are shifted by ATTRSHFT.
133  */
134 #define	ATTR_SHFT	4
135 struct pmap_physseg pmap_physseg;
136 #endif
137 
138 /*
139  * The following structure is exactly 32 bytes long (one cacheline).
140  */
141 struct pvo_entry {
142 	LIST_ENTRY(pvo_entry) pvo_vlink;	/* Link to common virt page */
143 	TAILQ_ENTRY(pvo_entry) pvo_olink;	/* Link to overflow entry */
144 	struct pte pvo_pte;			/* Prebuilt PTE */
145 	pmap_t pvo_pmap;			/* ptr to owning pmap */
146 	vaddr_t pvo_vaddr;			/* VA of entry */
147 #define	PVO_PTEGIDX_MASK	0x0007		/* which PTEG slot */
148 #define	PVO_PTEGIDX_VALID	0x0008		/* slot is valid */
149 #define	PVO_WIRED		0x0010		/* PVO entry is wired */
150 #define	PVO_MANAGED		0x0020		/* PVO e. for managed page */
151 #define	PVO_EXECUTABLE		0x0040		/* PVO e. for executable page */
152 };
153 #define	PVO_VADDR(pvo)		((pvo)->pvo_vaddr & ~ADDR_POFF)
154 #define	PVO_ISEXECUTABLE(pvo)	((pvo)->pvo_vaddr & PVO_EXECUTABLE)
155 #define	PVO_PTEGIDX_GET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
156 #define	PVO_PTEGIDX_ISSET(pvo)	((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
157 #define	PVO_PTEGIDX_CLR(pvo)	\
158 	((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
159 #define	PVO_PTEGIDX_SET(pvo,i)	\
160 	((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
161 
162 TAILQ_HEAD(pvo_tqhead, pvo_entry);
163 struct pvo_tqhead *pmap_pvo_table;	/* pvo entries by ptegroup index */
164 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged);	/* list of unmanaged pages */
165 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged);	/* list of unmanaged pages */
166 
167 struct pool pmap_pool;		/* pool for pmap structures */
168 struct pool pmap_upvo_pool;	/* pool for pvo entries for unmanaged pages */
169 struct pool pmap_mpvo_pool;	/* pool for pvo entries for managed pages */
170 
171 /*
172  * We keep a cache of unmanaged pages to be used for pvo entries for
173  * unmanaged pages.
174  */
175 struct pvo_page {
176 	SIMPLEQ_ENTRY(pvo_page) pvop_link;
177 };
178 SIMPLEQ_HEAD(pvop_head, pvo_page);
179 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head);
180 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head);
181 u_long pmap_upvop_free;
182 u_long pmap_upvop_maxfree;
183 u_long pmap_mpvop_free;
184 u_long pmap_mpvop_maxfree;
185 
186 STATIC void *pmap_pool_ualloc(struct pool *, int);
187 STATIC void *pmap_pool_malloc(struct pool *, int);
188 
189 STATIC void pmap_pool_ufree(struct pool *, void *);
190 STATIC void pmap_pool_mfree(struct pool *, void *);
191 
192 static struct pool_allocator pmap_pool_mallocator = {
193 	pmap_pool_malloc, pmap_pool_mfree, 0,
194 };
195 
196 static struct pool_allocator pmap_pool_uallocator = {
197 	pmap_pool_ualloc, pmap_pool_ufree, 0,
198 };
199 
200 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
201 void pmap_pte_print(volatile struct pte *);
202 #endif
203 
204 #ifdef DDB
205 void pmap_pteg_check(void);
206 void pmap_pteg_dist(void);
207 void pmap_print_pte(pmap_t, vaddr_t);
208 void pmap_print_mmuregs(void);
209 #endif
210 
211 #if defined(DEBUG) || defined(PMAPCHECK)
212 #ifdef PMAPCHECK
213 int pmapcheck = 1;
214 #else
215 int pmapcheck = 0;
216 #endif
217 void pmap_pvo_verify(void);
218 STATIC void pmap_pvo_check(const struct pvo_entry *);
219 #define	PMAP_PVO_CHECK(pvo)	 		\
220 	do {					\
221 		if (pmapcheck)			\
222 			pmap_pvo_check(pvo);	\
223 	} while (0)
224 #else
225 #define	PMAP_PVO_CHECK(pvo)	do { } while (/*CONSTCOND*/0)
226 #endif
227 STATIC int pmap_pte_insert(int, struct pte *);
228 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
229 	vaddr_t, paddr_t, register_t, int);
230 STATIC void pmap_pvo_remove(struct pvo_entry *, int);
231 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
232 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
233 
234 STATIC void tlbia(void);
235 
236 STATIC void pmap_release(pmap_t);
237 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int);
238 
239 #define	VSID_NBPW	(sizeof(uint32_t) * 8)
240 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
241 
242 static int pmap_initialized;
243 
244 #if defined(DEBUG) || defined(PMAPDEBUG)
245 #define	PMAPDEBUG_BOOT		0x0001
246 #define	PMAPDEBUG_PTE		0x0002
247 #define	PMAPDEBUG_EXEC		0x0008
248 #define	PMAPDEBUG_PVOENTER	0x0010
249 #define	PMAPDEBUG_PVOREMOVE	0x0020
250 #define	PMAPDEBUG_ACTIVATE	0x0100
251 #define	PMAPDEBUG_CREATE	0x0200
252 #define	PMAPDEBUG_ENTER		0x1000
253 #define	PMAPDEBUG_KENTER	0x2000
254 #define	PMAPDEBUG_KREMOVE	0x4000
255 #define	PMAPDEBUG_REMOVE	0x8000
256 unsigned int pmapdebug = 0;
257 # define DPRINTF(x)		printf x
258 # define DPRINTFN(n, x)		if (pmapdebug & PMAPDEBUG_ ## n) printf x
259 #else
260 # define DPRINTF(x)
261 # define DPRINTFN(n, x)
262 #endif
263 
264 
265 #ifdef PMAPCOUNTERS
266 #define	PMAPCOUNT(ev)	((pmap_evcnt_ ## ev).ev_count++)
267 #define	PMAPCOUNT2(ev)	((ev).ev_count++)
268 
269 struct evcnt pmap_evcnt_mappings =
270     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
271 	    "pmap", "pages mapped");
272 struct evcnt pmap_evcnt_unmappings =
273     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings,
274 	    "pmap", "pages unmapped");
275 
276 struct evcnt pmap_evcnt_kernel_mappings =
277     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
278 	    "pmap", "kernel pages mapped");
279 struct evcnt pmap_evcnt_kernel_unmappings =
280     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings,
281 	    "pmap", "kernel pages unmapped");
282 
283 struct evcnt pmap_evcnt_mappings_replaced =
284     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
285 	    "pmap", "page mappings replaced");
286 
287 struct evcnt pmap_evcnt_exec_mappings =
288     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings,
289 	    "pmap", "exec pages mapped");
290 struct evcnt pmap_evcnt_exec_cached =
291     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings,
292 	    "pmap", "exec pages cached");
293 
294 struct evcnt pmap_evcnt_exec_synced =
295     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings,
296 	    "pmap", "exec pages synced");
297 struct evcnt pmap_evcnt_exec_synced_clear_modify =
298     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings,
299 	    "pmap", "exec pages synced (CM)");
300 
301 struct evcnt pmap_evcnt_exec_uncached_page_protect =
302     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings,
303 	    "pmap", "exec pages uncached (PP)");
304 struct evcnt pmap_evcnt_exec_uncached_clear_modify =
305     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings,
306 	    "pmap", "exec pages uncached (CM)");
307 struct evcnt pmap_evcnt_exec_uncached_zero_page =
308     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings,
309 	    "pmap", "exec pages uncached (ZP)");
310 struct evcnt pmap_evcnt_exec_uncached_copy_page =
311     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings,
312 	    "pmap", "exec pages uncached (CP)");
313 
314 struct evcnt pmap_evcnt_updates =
315     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
316 	    "pmap", "updates");
317 struct evcnt pmap_evcnt_collects =
318     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
319 	    "pmap", "collects");
320 struct evcnt pmap_evcnt_copies =
321     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
322 	    "pmap", "copies");
323 
324 struct evcnt pmap_evcnt_ptes_spilled =
325     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
326 	    "pmap", "ptes spilled from overflow");
327 struct evcnt pmap_evcnt_ptes_unspilled =
328     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
329 	    "pmap", "ptes not spilled");
330 struct evcnt pmap_evcnt_ptes_evicted =
331     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
332 	    "pmap", "ptes evicted");
333 
334 struct evcnt pmap_evcnt_ptes_primary[8] = {
335     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
336 	    "pmap", "ptes added at primary[0]"),
337     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
338 	    "pmap", "ptes added at primary[1]"),
339     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
340 	    "pmap", "ptes added at primary[2]"),
341     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
342 	    "pmap", "ptes added at primary[3]"),
343 
344     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
345 	    "pmap", "ptes added at primary[4]"),
346     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
347 	    "pmap", "ptes added at primary[5]"),
348     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
349 	    "pmap", "ptes added at primary[6]"),
350     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
351 	    "pmap", "ptes added at primary[7]"),
352 };
353 struct evcnt pmap_evcnt_ptes_secondary[8] = {
354     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
355 	    "pmap", "ptes added at secondary[0]"),
356     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
357 	    "pmap", "ptes added at secondary[1]"),
358     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
359 	    "pmap", "ptes added at secondary[2]"),
360     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
361 	    "pmap", "ptes added at secondary[3]"),
362 
363     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
364 	    "pmap", "ptes added at secondary[4]"),
365     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
366 	    "pmap", "ptes added at secondary[5]"),
367     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
368 	    "pmap", "ptes added at secondary[6]"),
369     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
370 	    "pmap", "ptes added at secondary[7]"),
371 };
372 struct evcnt pmap_evcnt_ptes_removed =
373     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
374 	    "pmap", "ptes removed");
375 struct evcnt pmap_evcnt_ptes_changed =
376     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL,
377 	    "pmap", "ptes changed");
378 
379 /*
380  * From pmap_subr.c
381  */
382 extern struct evcnt pmap_evcnt_zeroed_pages;
383 extern struct evcnt pmap_evcnt_copied_pages;
384 extern struct evcnt pmap_evcnt_idlezeroed_pages;
385 #else
386 #define	PMAPCOUNT(ev)	((void) 0)
387 #define	PMAPCOUNT2(ev)	((void) 0)
388 #endif
389 
390 #define	TLBIE(va)	__asm __volatile("tlbie %0" :: "r"(va))
391 #define	TLBSYNC()	__asm __volatile("tlbsync")
392 #define	SYNC()		__asm __volatile("sync")
393 #define	EIEIO()		__asm __volatile("eieio")
394 #define	MFMSR()		mfmsr()
395 #define	MTMSR(psl)	mtmsr(psl)
396 #define	MFPVR()		mfpvr()
397 #define	MFSRIN(va)	mfsrin(va)
398 #define	MFTB()		mfrtcltbl()
399 
400 static __inline register_t
401 mfsrin(vaddr_t va)
402 {
403 	register_t sr;
404 	__asm __volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
405 	return sr;
406 }
407 
408 static __inline register_t
409 pmap_interrupts_off(void)
410 {
411 	register_t msr = MFMSR();
412 	if (msr & PSL_EE)
413 		MTMSR(msr & ~PSL_EE);
414 	return msr;
415 }
416 
417 static void
418 pmap_interrupts_restore(register_t msr)
419 {
420 	if (msr & PSL_EE)
421 		MTMSR(msr);
422 }
423 
424 static __inline u_int32_t
425 mfrtcltbl(void)
426 {
427 
428 	if ((MFPVR() >> 16) == MPC601)
429 		return (mfrtcl() >> 7);
430 	else
431 		return (mftbl());
432 }
433 
434 /*
435  * These small routines may have to be replaced,
436  * if/when we support processors other that the 604.
437  */
438 
439 void
440 tlbia(void)
441 {
442 	caddr_t i;
443 
444 	SYNC();
445 	/*
446 	 * Why not use "tlbia"?  Because not all processors implement it.
447 	 *
448 	 * This needs to be a per-cpu callback to do the appropriate thing
449 	 * for the CPU. XXX
450 	 */
451 	for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) {
452 		TLBIE(i);
453 		EIEIO();
454 		SYNC();
455 	}
456 	TLBSYNC();
457 	SYNC();
458 }
459 
460 static __inline register_t
461 va_to_vsid(const struct pmap *pm, vaddr_t addr)
462 {
463 	return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID);
464 }
465 
466 static __inline register_t
467 va_to_pteg(const struct pmap *pm, vaddr_t addr)
468 {
469 	register_t hash;
470 
471 	hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
472 	return hash & pmap_pteg_mask;
473 }
474 
475 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
476 /*
477  * Given a PTE in the page table, calculate the VADDR that hashes to it.
478  * The only bit of magic is that the top 4 bits of the address doesn't
479  * technically exist in the PTE.  But we know we reserved 4 bits of the
480  * VSID for it so that's how we get it.
481  */
482 static vaddr_t
483 pmap_pte_to_va(volatile const struct pte *pt)
484 {
485 	vaddr_t va;
486 	uintptr_t ptaddr = (uintptr_t) pt;
487 
488 	if (pt->pte_hi & PTE_HID)
489 		ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
490 
491 	/* PPC Bits 10-19 */
492 	va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
493 	va <<= ADDR_PIDX_SHFT;
494 
495 	/* PPC Bits 4-9 */
496 	va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
497 
498 	/* PPC Bits 0-3 */
499 	va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
500 
501 	return va;
502 }
503 #endif
504 
505 static __inline struct pvo_head *
506 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
507 {
508 #ifdef __HAVE_VM_PAGE_MD
509 	struct vm_page *pg;
510 
511 	pg = PHYS_TO_VM_PAGE(pa);
512 	if (pg_p != NULL)
513 		*pg_p = pg;
514 	if (pg == NULL)
515 		return &pmap_pvo_unmanaged;
516 	return &pg->mdpage.mdpg_pvoh;
517 #endif
518 #ifdef __HAVE_PMAP_PHYSSEG
519 	int bank, pg;
520 
521 	bank = vm_physseg_find(atop(pa), &pg);
522 	if (pg_p != NULL)
523 		*pg_p = pg;
524 	if (bank == -1)
525 		return &pmap_pvo_unmanaged;
526 	return &vm_physmem[bank].pmseg.pvoh[pg];
527 #endif
528 }
529 
530 static __inline struct pvo_head *
531 vm_page_to_pvoh(struct vm_page *pg)
532 {
533 #ifdef __HAVE_VM_PAGE_MD
534 	return &pg->mdpage.mdpg_pvoh;
535 #endif
536 #ifdef __HAVE_PMAP_PHYSSEG
537 	return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL);
538 #endif
539 }
540 
541 
542 #ifdef __HAVE_PMAP_PHYSSEG
543 static __inline char *
544 pa_to_attr(paddr_t pa)
545 {
546 	int bank, pg;
547 
548 	bank = vm_physseg_find(atop(pa), &pg);
549 	if (bank == -1)
550 		return NULL;
551 	return &vm_physmem[bank].pmseg.attrs[pg];
552 }
553 #endif
554 
555 static __inline void
556 pmap_attr_clear(struct vm_page *pg, int ptebit)
557 {
558 #ifdef __HAVE_PMAP_PHYSSEG
559 	*pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT);
560 #endif
561 #ifdef __HAVE_VM_PAGE_MD
562 	pg->mdpage.mdpg_attrs &= ~ptebit;
563 #endif
564 }
565 
566 static __inline int
567 pmap_attr_fetch(struct vm_page *pg)
568 {
569 #ifdef __HAVE_PMAP_PHYSSEG
570 	return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT;
571 #endif
572 #ifdef __HAVE_VM_PAGE_MD
573 	return pg->mdpage.mdpg_attrs;
574 #endif
575 }
576 
577 static __inline void
578 pmap_attr_save(struct vm_page *pg, int ptebit)
579 {
580 #ifdef __HAVE_PMAP_PHYSSEG
581 	*pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT);
582 #endif
583 #ifdef __HAVE_VM_PAGE_MD
584 	pg->mdpage.mdpg_attrs |= ptebit;
585 #endif
586 }
587 
588 static __inline int
589 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
590 {
591 	if (pt->pte_hi == pvo_pt->pte_hi
592 #if 0
593 	    && ((pt->pte_lo ^ pvo_pt->pte_lo) &
594 	        ~(PTE_REF|PTE_CHG)) == 0
595 #endif
596 	    )
597 		return 1;
598 	return 0;
599 }
600 
601 static __inline void
602 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
603 {
604 	/*
605 	 * Construct the PTE.  Default to IMB initially.  Valid bit
606 	 * only gets set when the real pte is set in memory.
607 	 *
608 	 * Note: Don't set the valid bit for correct operation of tlb update.
609 	 */
610 	pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
611 	    | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
612 	pt->pte_lo = pte_lo;
613 }
614 
615 static __inline void
616 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
617 {
618 	pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
619 }
620 
621 static __inline void
622 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
623 {
624 	/*
625 	 * As shown in Section 7.6.3.2.3
626 	 */
627 	pt->pte_lo &= ~ptebit;
628 	TLBIE(va);
629 	SYNC();
630 	EIEIO();
631 	TLBSYNC();
632 	SYNC();
633 }
634 
635 static __inline void
636 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
637 {
638 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
639 	if (pvo_pt->pte_hi & PTE_VALID)
640 		panic("pte_set: setting an already valid pte %p", pvo_pt);
641 #endif
642 	pvo_pt->pte_hi |= PTE_VALID;
643 	/*
644 	 * Update the PTE as defined in section 7.6.3.1
645 	 * Note that the REF/CHG bits are from pvo_pt and thus should
646 	 * have been saved so this routine can restore them (if desired).
647 	 */
648 	pt->pte_lo = pvo_pt->pte_lo;
649 	EIEIO();
650 	pt->pte_hi = pvo_pt->pte_hi;
651 	SYNC();
652 	pmap_pte_valid++;
653 }
654 
655 static __inline void
656 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
657 {
658 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
659 	if ((pvo_pt->pte_hi & PTE_VALID) == 0)
660 		panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
661 	if ((pt->pte_hi & PTE_VALID) == 0)
662 		panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
663 #endif
664 
665 	pvo_pt->pte_hi &= ~PTE_VALID;
666 	/*
667 	 * Force the ref & chg bits back into the PTEs.
668 	 */
669 	SYNC();
670 	/*
671 	 * Invalidate the pte ... (Section 7.6.3.3)
672 	 */
673 	pt->pte_hi &= ~PTE_VALID;
674 	SYNC();
675 	TLBIE(va);
676 	SYNC();
677 	EIEIO();
678 	TLBSYNC();
679 	SYNC();
680 	/*
681 	 * Save the ref & chg bits ...
682 	 */
683 	pmap_pte_synch(pt, pvo_pt);
684 	pmap_pte_valid--;
685 }
686 
687 static __inline void
688 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
689 {
690 	/*
691 	 * Invalidate the PTE
692 	 */
693 	pmap_pte_unset(pt, pvo_pt, va);
694 	pmap_pte_set(pt, pvo_pt);
695 }
696 
697 /*
698  * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
699  * (either primary or secondary location).
700  *
701  * Note: both the destination and source PTEs must not have PTE_VALID set.
702  */
703 
704 STATIC int
705 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
706 {
707 	volatile struct pte *pt;
708 	int i;
709 
710 #if defined(DEBUG)
711 	DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%lx 0x%lx\n",
712 		ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo));
713 #endif
714 	/*
715 	 * First try primary hash.
716 	 */
717 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
718 		if ((pt->pte_hi & PTE_VALID) == 0) {
719 			pvo_pt->pte_hi &= ~PTE_HID;
720 			pmap_pte_set(pt, pvo_pt);
721 			return i;
722 		}
723 	}
724 
725 	/*
726 	 * Now try secondary hash.
727 	 */
728 	ptegidx ^= pmap_pteg_mask;
729 	for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
730 		if ((pt->pte_hi & PTE_VALID) == 0) {
731 			pvo_pt->pte_hi |= PTE_HID;
732 			pmap_pte_set(pt, pvo_pt);
733 			return i;
734 		}
735 	}
736 	return -1;
737 }
738 
739 /*
740  * Spill handler.
741  *
742  * Tries to spill a page table entry from the overflow area.
743  * This runs in either real mode (if dealing with a exception spill)
744  * or virtual mode when dealing with manually spilling one of the
745  * kernel's pte entries.  In either case, interrupts are already
746  * disabled.
747  */
748 int
749 pmap_pte_spill(struct pmap *pm, vaddr_t addr)
750 {
751 	struct pvo_entry *source_pvo, *victim_pvo, *next_pvo;
752 	struct pvo_entry *pvo;
753 	struct pvo_tqhead *pvoh, *vpvoh;
754 	int ptegidx, i, j;
755 	volatile struct pteg *pteg;
756 	volatile struct pte *pt;
757 
758 	ptegidx = va_to_pteg(pm, addr);
759 
760 	/*
761 	 * Have to substitute some entry. Use the primary hash for this.
762 	 *
763 	 * Use low bits of timebase as random generator
764 	 */
765 	pteg = &pmap_pteg_table[ptegidx];
766 	i = MFTB() & 7;
767 	pt = &pteg->pt[i];
768 
769 	source_pvo = NULL;
770 	victim_pvo = NULL;
771 	pvoh = &pmap_pvo_table[ptegidx];
772 	TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
773 
774 		/*
775 		 * We need to find pvo entry for this address...
776 		 */
777 		PMAP_PVO_CHECK(pvo);		/* sanity check */
778 
779 		/*
780 		 * If we haven't found the source and we come to a PVO with
781 		 * a valid PTE, then we know we can't find it because all
782 		 * evicted PVOs always are first in the list.
783 		 */
784 		if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID))
785 			break;
786 		if (source_pvo == NULL && pm == pvo->pvo_pmap &&
787 		    addr == PVO_VADDR(pvo)) {
788 
789 			/*
790 			 * Now we have found the entry to be spilled into the
791 			 * pteg.  Attempt to insert it into the page table.
792 			 */
793 			j = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
794 			if (j >= 0) {
795 				PVO_PTEGIDX_SET(pvo, j);
796 				PMAP_PVO_CHECK(pvo);	/* sanity check */
797 				pvo->pvo_pmap->pm_evictions--;
798 				PMAPCOUNT(ptes_spilled);
799 				PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
800 				    ? pmap_evcnt_ptes_secondary
801 				    : pmap_evcnt_ptes_primary)[j]);
802 
803 				/*
804 				 * Since we keep the evicted entries at the
805 				 * from of the PVO list, we need move this
806 				 * (now resident) PVO after the evicted
807 				 * entries.
808 				 */
809 				next_pvo = TAILQ_NEXT(pvo, pvo_olink);
810 
811 				/*
812 				 * If we don't have to move (either we were the
813 				 * last entry or the next entry was valid),
814 				 * don't change our position.  Otherwise
815 				 * move ourselves to the tail of the queue.
816 				 */
817 				if (next_pvo != NULL &&
818 				    !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) {
819 					TAILQ_REMOVE(pvoh, pvo, pvo_olink);
820 					TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
821 				}
822 				return 1;
823 			}
824 			source_pvo = pvo;
825 			if (victim_pvo != NULL)
826 				break;
827 		}
828 
829 		/*
830 		 * We also need the pvo entry of the victim we are replacing
831 		 * so save the R & C bits of the PTE.
832 		 */
833 		if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL &&
834 		    pmap_pte_compare(pt, &pvo->pvo_pte)) {
835 			vpvoh = pvoh;
836 			victim_pvo = pvo;
837 			if (source_pvo != NULL)
838 				break;
839 		}
840 	}
841 
842 	if (source_pvo == NULL) {
843 		PMAPCOUNT(ptes_unspilled);
844 		return 0;
845 	}
846 
847 	if (victim_pvo == NULL) {
848 		if ((pt->pte_hi & PTE_HID) == 0)
849 			panic("pmap_pte_spill: victim p-pte (%p) has "
850 			    "no pvo entry!", pt);
851 
852 		/*
853 		 * If this is a secondary PTE, we need to search
854 		 * its primary pvo bucket for the matching PVO.
855 		 */
856 		vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask];
857 		TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
858 			PMAP_PVO_CHECK(pvo);		/* sanity check */
859 
860 			/*
861 			 * We also need the pvo entry of the victim we are
862 			 * replacing so save the R & C bits of the PTE.
863 			 */
864 			if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
865 				victim_pvo = pvo;
866 				break;
867 			}
868 		}
869 		if (victim_pvo == NULL)
870 			panic("pmap_pte_spill: victim s-pte (%p) has "
871 			    "no pvo entry!", pt);
872 	}
873 
874 	/*
875 	 * We are invalidating the TLB entry for the EA for the
876 	 * we are replacing even though its valid; If we don't
877 	 * we lose any ref/chg bit changes contained in the TLB
878 	 * entry.
879 	 */
880 	source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
881 
882 	/*
883 	 * To enforce the PVO list ordering constraint that all
884 	 * evicted entries should come before all valid entries,
885 	 * move the source PVO to the tail of its list and the
886 	 * victim PVO to the head of its list (which might not be
887 	 * the same list, if the victim was using the secondary hash).
888 	 */
889 	TAILQ_REMOVE(pvoh, source_pvo, pvo_olink);
890 	TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink);
891 	TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
892 	TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
893 	pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
894 	pmap_pte_set(pt, &source_pvo->pvo_pte);
895 	victim_pvo->pvo_pmap->pm_evictions++;
896 	source_pvo->pvo_pmap->pm_evictions--;
897 
898 	PVO_PTEGIDX_CLR(victim_pvo);
899 	PVO_PTEGIDX_SET(source_pvo, i);
900 	PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
901 	PMAPCOUNT(ptes_spilled);
902 	PMAPCOUNT(ptes_evicted);
903 	PMAPCOUNT(ptes_removed);
904 
905 	PMAP_PVO_CHECK(victim_pvo);
906 	PMAP_PVO_CHECK(source_pvo);
907 	return 1;
908 }
909 
910 /*
911  * Restrict given range to physical memory
912  */
913 void
914 pmap_real_memory(paddr_t *start, psize_t *size)
915 {
916 	struct mem_region *mp;
917 
918 	for (mp = mem; mp->size; mp++) {
919 		if (*start + *size > mp->start
920 		    && *start < mp->start + mp->size) {
921 			if (*start < mp->start) {
922 				*size -= mp->start - *start;
923 				*start = mp->start;
924 			}
925 			if (*start + *size > mp->start + mp->size)
926 				*size = mp->start + mp->size - *start;
927 			return;
928 		}
929 	}
930 	*size = 0;
931 }
932 
933 /*
934  * Initialize anything else for pmap handling.
935  * Called during vm_init().
936  */
937 void
938 pmap_init(void)
939 {
940 	int s;
941 #ifdef __HAVE_PMAP_PHYSSEG
942 	struct pvo_tqhead *pvoh;
943 	int bank;
944 	long sz;
945 	char *attr;
946 
947 	s = splvm();
948 	pvoh = pmap_physseg.pvoh;
949 	attr = pmap_physseg.attrs;
950 	for (bank = 0; bank < vm_nphysseg; bank++) {
951 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
952 		vm_physmem[bank].pmseg.pvoh = pvoh;
953 		vm_physmem[bank].pmseg.attrs = attr;
954 		for (; sz > 0; sz--, pvoh++, attr++) {
955 			TAILQ_INIT(pvoh);
956 			*attr = 0;
957 		}
958 	}
959 	splx(s);
960 #endif
961 
962 	s = splvm();
963 	pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry),
964 	    sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl",
965 	    &pmap_pool_mallocator);
966 
967 	pool_setlowat(&pmap_mpvo_pool, 1008);
968 
969 	pmap_initialized = 1;
970 	splx(s);
971 
972 #ifdef PMAPCOUNTERS
973 	evcnt_attach_static(&pmap_evcnt_mappings);
974 	evcnt_attach_static(&pmap_evcnt_mappings_replaced);
975 	evcnt_attach_static(&pmap_evcnt_unmappings);
976 
977 	evcnt_attach_static(&pmap_evcnt_kernel_mappings);
978 	evcnt_attach_static(&pmap_evcnt_kernel_unmappings);
979 
980 	evcnt_attach_static(&pmap_evcnt_exec_mappings);
981 	evcnt_attach_static(&pmap_evcnt_exec_cached);
982 	evcnt_attach_static(&pmap_evcnt_exec_synced);
983 	evcnt_attach_static(&pmap_evcnt_exec_synced_clear_modify);
984 
985 	evcnt_attach_static(&pmap_evcnt_exec_uncached_page_protect);
986 	evcnt_attach_static(&pmap_evcnt_exec_uncached_clear_modify);
987 	evcnt_attach_static(&pmap_evcnt_exec_uncached_zero_page);
988 	evcnt_attach_static(&pmap_evcnt_exec_uncached_copy_page);
989 
990 	evcnt_attach_static(&pmap_evcnt_zeroed_pages);
991 	evcnt_attach_static(&pmap_evcnt_copied_pages);
992 	evcnt_attach_static(&pmap_evcnt_idlezeroed_pages);
993 
994 	evcnt_attach_static(&pmap_evcnt_updates);
995 	evcnt_attach_static(&pmap_evcnt_collects);
996 	evcnt_attach_static(&pmap_evcnt_copies);
997 
998 	evcnt_attach_static(&pmap_evcnt_ptes_spilled);
999 	evcnt_attach_static(&pmap_evcnt_ptes_unspilled);
1000 	evcnt_attach_static(&pmap_evcnt_ptes_evicted);
1001 	evcnt_attach_static(&pmap_evcnt_ptes_removed);
1002 	evcnt_attach_static(&pmap_evcnt_ptes_changed);
1003 	evcnt_attach_static(&pmap_evcnt_ptes_primary[0]);
1004 	evcnt_attach_static(&pmap_evcnt_ptes_primary[1]);
1005 	evcnt_attach_static(&pmap_evcnt_ptes_primary[2]);
1006 	evcnt_attach_static(&pmap_evcnt_ptes_primary[3]);
1007 	evcnt_attach_static(&pmap_evcnt_ptes_primary[4]);
1008 	evcnt_attach_static(&pmap_evcnt_ptes_primary[5]);
1009 	evcnt_attach_static(&pmap_evcnt_ptes_primary[6]);
1010 	evcnt_attach_static(&pmap_evcnt_ptes_primary[7]);
1011 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[0]);
1012 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[1]);
1013 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[2]);
1014 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[3]);
1015 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[4]);
1016 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[5]);
1017 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[6]);
1018 	evcnt_attach_static(&pmap_evcnt_ptes_secondary[7]);
1019 #endif
1020 }
1021 
1022 /*
1023  * How much virtual space does the kernel get?
1024  */
1025 void
1026 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1027 {
1028 	/*
1029 	 * For now, reserve one segment (minus some overhead) for kernel
1030 	 * virtual memory
1031 	 */
1032 	*start = VM_MIN_KERNEL_ADDRESS;
1033 	*end = VM_MAX_KERNEL_ADDRESS;
1034 }
1035 
1036 /*
1037  * Allocate, initialize, and return a new physical map.
1038  */
1039 pmap_t
1040 pmap_create(void)
1041 {
1042 	pmap_t pm;
1043 
1044 	pm = pool_get(&pmap_pool, PR_WAITOK);
1045 	memset((caddr_t)pm, 0, sizeof *pm);
1046 	pmap_pinit(pm);
1047 
1048 	DPRINTFN(CREATE,("pmap_create: pm %p:\n"
1049 	    "\t%06lx %06lx %06lx %06lx    %06lx %06lx %06lx %06lx\n"
1050 	    "\t%06lx %06lx %06lx %06lx    %06lx %06lx %06lx %06lx\n", pm,
1051 	    pm->pm_sr[0], pm->pm_sr[1], pm->pm_sr[2], pm->pm_sr[3],
1052 	    pm->pm_sr[4], pm->pm_sr[5], pm->pm_sr[6], pm->pm_sr[7],
1053 	    pm->pm_sr[8], pm->pm_sr[9], pm->pm_sr[10], pm->pm_sr[11],
1054 	    pm->pm_sr[12], pm->pm_sr[13], pm->pm_sr[14], pm->pm_sr[15]));
1055 	return pm;
1056 }
1057 
1058 /*
1059  * Initialize a preallocated and zeroed pmap structure.
1060  */
1061 void
1062 pmap_pinit(pmap_t pm)
1063 {
1064 	register_t entropy = MFTB();
1065 	register_t mask;
1066 	int i;
1067 
1068 	/*
1069 	 * Allocate some segment registers for this pmap.
1070 	 */
1071 	pm->pm_refs = 1;
1072 	for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1073 		static register_t pmap_vsidcontext;
1074 		register_t hash;
1075 		unsigned int n;
1076 
1077 		/* Create a new value by multiplying by a prime adding in
1078 		 * entropy from the timebase register.  This is to make the
1079 		 * VSID more random so that the PT Hash function collides
1080 		 * less often. (note that the prime causes gcc to do shifts
1081 		 * instead of a multiply)
1082 		 */
1083 		pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1084 		hash = pmap_vsidcontext & (NPMAPS - 1);
1085 		if (hash == 0)			/* 0 is special, avoid it */
1086 			continue;
1087 		n = hash >> 5;
1088 		mask = 1L << (hash & (VSID_NBPW-1));
1089 		hash = pmap_vsidcontext;
1090 		if (pmap_vsid_bitmap[n] & mask) {	/* collision? */
1091 			/* anything free in this bucket? */
1092 			if (~pmap_vsid_bitmap[n] == 0) {
1093 				entropy = hash >> PTE_VSID_SHFT;
1094 				continue;
1095 			}
1096 			i = ffs(~pmap_vsid_bitmap[n]) - 1;
1097 			mask = 1L << i;
1098 			hash &= ~(VSID_NBPW-1);
1099 			hash |= i;
1100 		}
1101 		/*
1102 		 * Make sure clear out SR_KEY_LEN bits because we put our
1103 		 * our data in those bits (to identify the segment).
1104 		 */
1105 		hash &= PTE_VSID >> (PTE_VSID_SHFT + SR_KEY_LEN);
1106 		pmap_vsid_bitmap[n] |= mask;
1107 		for (i = 0; i < 16; i++)
1108 			pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY;
1109 		return;
1110 	}
1111 	panic("pmap_pinit: out of segments");
1112 }
1113 
1114 /*
1115  * Add a reference to the given pmap.
1116  */
1117 void
1118 pmap_reference(pmap_t pm)
1119 {
1120 	pm->pm_refs++;
1121 }
1122 
1123 /*
1124  * Retire the given pmap from service.
1125  * Should only be called if the map contains no valid mappings.
1126  */
1127 void
1128 pmap_destroy(pmap_t pm)
1129 {
1130 	if (--pm->pm_refs == 0) {
1131 		pmap_release(pm);
1132 		pool_put(&pmap_pool, pm);
1133 	}
1134 }
1135 
1136 /*
1137  * Release any resources held by the given physical map.
1138  * Called when a pmap initialized by pmap_pinit is being released.
1139  */
1140 void
1141 pmap_release(pmap_t pm)
1142 {
1143 	int idx, mask;
1144 
1145 	if (pm->pm_sr[0] == 0)
1146 		panic("pmap_release");
1147 	idx = VSID_TO_HASH(pm->pm_sr[0]) & (NPMAPS-1);
1148 	mask = 1 << (idx % VSID_NBPW);
1149 	idx /= VSID_NBPW;
1150 	pmap_vsid_bitmap[idx] &= ~mask;
1151 }
1152 
1153 /*
1154  * Copy the range specified by src_addr/len
1155  * from the source map to the range dst_addr/len
1156  * in the destination map.
1157  *
1158  * This routine is only advisory and need not do anything.
1159  */
1160 void
1161 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1162 	vsize_t len, vaddr_t src_addr)
1163 {
1164 	PMAPCOUNT(copies);
1165 }
1166 
1167 /*
1168  * Require that all active physical maps contain no
1169  * incorrect entries NOW.
1170  */
1171 void
1172 pmap_update(struct pmap *pmap)
1173 {
1174 	PMAPCOUNT(updates);
1175 	TLBSYNC();
1176 }
1177 
1178 /*
1179  * Garbage collects the physical map system for
1180  * pages which are no longer used.
1181  * Success need not be guaranteed -- that is, there
1182  * may well be pages which are not referenced, but
1183  * others may be collected.
1184  * Called by the pageout daemon when pages are scarce.
1185  */
1186 void
1187 pmap_collect(pmap_t pm)
1188 {
1189 	PMAPCOUNT(collects);
1190 }
1191 
1192 static __inline int
1193 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1194 {
1195 	int pteidx;
1196 	/*
1197 	 * We can find the actual pte entry without searching by
1198 	 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1199 	 * and by noticing the HID bit.
1200 	 */
1201 	pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1202 	if (pvo->pvo_pte.pte_hi & PTE_HID)
1203 		pteidx ^= pmap_pteg_mask * 8;
1204 	return pteidx;
1205 }
1206 
1207 volatile struct pte *
1208 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1209 {
1210 	volatile struct pte *pt;
1211 
1212 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1213 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1214 		return NULL;
1215 #endif
1216 
1217 	/*
1218 	 * If we haven't been supplied the ptegidx, calculate it.
1219 	 */
1220 	if (pteidx == -1) {
1221 		int ptegidx;
1222 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1223 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1224 	}
1225 
1226 	pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1227 
1228 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1229 	return pt;
1230 #else
1231 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1232 		panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1233 		    "pvo but no valid pte index", pvo);
1234 	}
1235 	if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1236 		panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1237 		    "pvo but no valid pte", pvo);
1238 	}
1239 
1240 	if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1241 		if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1242 #if defined(DEBUG) || defined(PMAPCHECK)
1243 			pmap_pte_print(pt);
1244 #endif
1245 			panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1246 			    "pmap_pteg_table %p but invalid in pvo",
1247 			    pvo, pt);
1248 		}
1249 		if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1250 #if defined(DEBUG) || defined(PMAPCHECK)
1251 			pmap_pte_print(pt);
1252 #endif
1253 			panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1254 			    "not match pte %p in pmap_pteg_table",
1255 			    pvo, pt);
1256 		}
1257 		return pt;
1258 	}
1259 
1260 	if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1261 #if defined(DEBUG) || defined(PMAPCHECK)
1262 		pmap_pte_print(pt);
1263 #endif
1264 		panic("pmap_pvo_to_pte: pvo %p: has invalid pte %p in "
1265 		    "pmap_pteg_table but valid in pvo", pvo, pt);
1266 	}
1267 	return NULL;
1268 #endif	/* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1269 }
1270 
1271 struct pvo_entry *
1272 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1273 {
1274 	struct pvo_entry *pvo;
1275 	int ptegidx;
1276 
1277 	va &= ~ADDR_POFF;
1278 	ptegidx = va_to_pteg(pm, va);
1279 
1280 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1281 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1282 		if ((uintptr_t) pvo >= SEGMENT_LENGTH)
1283 			panic("pmap_pvo_find_va: invalid pvo %p on "
1284 			    "list %#x (%p)", pvo, ptegidx,
1285 			     &pmap_pvo_table[ptegidx]);
1286 #endif
1287 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1288 			if (pteidx_p)
1289 				*pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1290 			return pvo;
1291 		}
1292 	}
1293 	return NULL;
1294 }
1295 
1296 #if defined(DEBUG) || defined(PMAPCHECK)
1297 void
1298 pmap_pvo_check(const struct pvo_entry *pvo)
1299 {
1300 	struct pvo_head *pvo_head;
1301 	struct pvo_entry *pvo0;
1302 	volatile struct pte *pt;
1303 	int failed = 0;
1304 
1305 	if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH)
1306 		panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1307 
1308 	if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) {
1309 		printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1310 		    pvo, pvo->pvo_pmap);
1311 		failed = 1;
1312 	}
1313 
1314 	if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH ||
1315 	    (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1316 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1317 		    pvo, TAILQ_NEXT(pvo, pvo_olink));
1318 		failed = 1;
1319 	}
1320 
1321 	if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH ||
1322 	    (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1323 		printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1324 		    pvo, LIST_NEXT(pvo, pvo_vlink));
1325 		failed = 1;
1326 	}
1327 
1328 	if (pvo->pvo_vaddr & PVO_MANAGED) {
1329 		pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1330 	} else {
1331 		if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) {
1332 			printf("pmap_pvo_check: pvo %p: non kernel address "
1333 			    "on kernel unmanaged list\n", pvo);
1334 			failed = 1;
1335 		}
1336 		pvo_head = &pmap_pvo_kunmanaged;
1337 	}
1338 	LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1339 		if (pvo0 == pvo)
1340 			break;
1341 	}
1342 	if (pvo0 == NULL) {
1343 		printf("pmap_pvo_check: pvo %p: not present "
1344 		    "on its vlist head %p\n", pvo, pvo_head);
1345 		failed = 1;
1346 	}
1347 	if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1348 		printf("pmap_pvo_check: pvo %p: not present "
1349 		    "on its olist head\n", pvo);
1350 		failed = 1;
1351 	}
1352 	pt = pmap_pvo_to_pte(pvo, -1);
1353 	if (pt == NULL) {
1354 		if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1355 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1356 			    "no PTE\n", pvo);
1357 			failed = 1;
1358 		}
1359 	} else {
1360 		if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1361 		    (uintptr_t) pt >=
1362 		    (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1363 			printf("pmap_pvo_check: pvo %p: pte %p not in "
1364 			    "pteg table\n", pvo, pt);
1365 			failed = 1;
1366 		}
1367 		if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1368 			printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1369 			    "no PTE\n", pvo);
1370 			failed = 1;
1371 		}
1372 		if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1373 			printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1374 			    "%#lx/%#lx\n", pvo, pvo->pvo_pte.pte_hi, pt->pte_hi);
1375 			failed = 1;
1376 		}
1377 		if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1378 		    (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1379 			printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1380 			    "%#lx/%#lx\n", pvo,
1381 			    pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN),
1382 			    pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN));
1383 			failed = 1;
1384 		}
1385 		if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1386 			printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx"
1387 			    " doesn't not match PVO's VA %#lx\n",
1388 			    pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1389 			failed = 1;
1390 		}
1391 		if (failed)
1392 			pmap_pte_print(pt);
1393 	}
1394 	if (failed)
1395 		panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1396 		    pvo->pvo_pmap);
1397 }
1398 #endif /* DEBUG || PMAPCHECK */
1399 
1400 /*
1401  * This returns whether this is the first mapping of a page.
1402  */
1403 int
1404 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1405 	vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1406 {
1407 	struct pvo_entry *pvo;
1408 	struct pvo_tqhead *pvoh;
1409 	register_t msr;
1410 	int ptegidx;
1411 	int i;
1412 	int poolflags = PR_NOWAIT;
1413 
1414 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1415 	if (pmap_pvo_remove_depth > 0)
1416 		panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1417 	if (++pmap_pvo_enter_depth > 1)
1418 		panic("pmap_pvo_enter: called recursively!");
1419 #endif
1420 
1421 	/*
1422 	 * Compute the PTE Group index.
1423 	 */
1424 	va &= ~ADDR_POFF;
1425 	ptegidx = va_to_pteg(pm, va);
1426 
1427 	msr = pmap_interrupts_off();
1428 	/*
1429 	 * Remove any existing mapping for this page.  Reuse the
1430 	 * pvo entry if there a mapping.
1431 	 */
1432 	TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1433 		if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1434 #ifdef DEBUG
1435 			if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1436 			    ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1437 			    ~(PTE_REF|PTE_CHG)) == 0 &&
1438 			   va < VM_MIN_KERNEL_ADDRESS) {
1439 				printf("pmap_pvo_enter: pvo %p: dup %#lx/%#lx\n",
1440 				    pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1441 				printf("pmap_pvo_enter: pte_hi=%#lx sr=%#lx\n",
1442 				    pvo->pvo_pte.pte_hi,
1443 				    pm->pm_sr[va >> ADDR_SR_SHFT]);
1444 				pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1445 #ifdef DDBX
1446 				Debugger();
1447 #endif
1448 			}
1449 #endif
1450 			PMAPCOUNT(mappings_replaced);
1451 			pmap_pvo_remove(pvo, -1);
1452 			break;
1453 		}
1454 	}
1455 
1456 	/*
1457 	 * If we aren't overwriting an mapping, try to allocate
1458 	 */
1459 	pmap_interrupts_restore(msr);
1460 	pvo = pool_get(pl, poolflags);
1461 	msr = pmap_interrupts_off();
1462 	if (pvo == NULL) {
1463 #if 0
1464 		pvo = pmap_pvo_reclaim(pm);
1465 		if (pvo == NULL) {
1466 #endif
1467 			if ((flags & PMAP_CANFAIL) == 0)
1468 				panic("pmap_pvo_enter: failed");
1469 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1470 			pmap_pvo_enter_depth--;
1471 #endif
1472 			pmap_interrupts_restore(msr);
1473 			return ENOMEM;
1474 #if 0
1475 		}
1476 #endif
1477 	}
1478 	pvo->pvo_vaddr = va;
1479 	pvo->pvo_pmap = pm;
1480 	pvo->pvo_vaddr &= ~ADDR_POFF;
1481 	if (flags & VM_PROT_EXECUTE) {
1482 		PMAPCOUNT(exec_mappings);
1483 		pvo->pvo_vaddr |= PVO_EXECUTABLE;
1484 	}
1485 	if (flags & PMAP_WIRED)
1486 		pvo->pvo_vaddr |= PVO_WIRED;
1487 	if (pvo_head != &pmap_pvo_kunmanaged) {
1488 		pvo->pvo_vaddr |= PVO_MANAGED;
1489 		PMAPCOUNT(mappings);
1490 	} else {
1491 		PMAPCOUNT(kernel_mappings);
1492 	}
1493 	pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1494 
1495 	LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1496 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1497 		pvo->pvo_pmap->pm_stats.wired_count++;
1498 	pvo->pvo_pmap->pm_stats.resident_count++;
1499 #if defined(DEBUG)
1500 	if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS)
1501 		DPRINTFN(PVOENTER,
1502 		    ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n",
1503 		    pvo, pm, va, pa));
1504 #endif
1505 
1506 	/*
1507 	 * We hope this succeeds but it isn't required.
1508 	 */
1509 	pvoh = &pmap_pvo_table[ptegidx];
1510 	i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1511 	if (i >= 0) {
1512 		PVO_PTEGIDX_SET(pvo, i);
1513 		PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1514 		    ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1515 		TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1516 	} else {
1517 
1518 		/*
1519 		 * Since we didn't have room for this entry (which makes it
1520 		 * and evicted entry), place it at the head of the list.
1521 		 */
1522 		TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1523 		PMAPCOUNT(ptes_evicted);
1524 		pm->pm_evictions++;
1525 	}
1526 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1527 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1528 	pmap_pvo_enter_depth--;
1529 #endif
1530 	pmap_interrupts_restore(msr);
1531 	return 0;
1532 }
1533 
1534 void
1535 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx)
1536 {
1537 	volatile struct pte *pt;
1538 	int ptegidx;
1539 
1540 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1541 	if (++pmap_pvo_remove_depth > 1)
1542 		panic("pmap_pvo_remove: called recursively!");
1543 #endif
1544 
1545 	/*
1546 	 * If we haven't been supplied the ptegidx, calculate it.
1547 	 */
1548 	if (pteidx == -1) {
1549 		ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1550 		pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1551 	} else {
1552 		ptegidx = pteidx >> 3;
1553 		if (pvo->pvo_pte.pte_hi & PTE_HID)
1554 			ptegidx ^= pmap_pteg_mask;
1555 	}
1556 	PMAP_PVO_CHECK(pvo);		/* sanity check */
1557 
1558 	/*
1559 	 * If there is an active pte entry, we need to deactivate it
1560 	 * (and save the ref & chg bits).
1561 	 */
1562 	pt = pmap_pvo_to_pte(pvo, pteidx);
1563 	if (pt != NULL) {
1564 		pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1565 		PVO_PTEGIDX_CLR(pvo);
1566 		PMAPCOUNT(ptes_removed);
1567 	} else {
1568 		KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1569 		pvo->pvo_pmap->pm_evictions--;
1570 	}
1571 
1572 	/*
1573 	 * Update our statistics
1574 	 */
1575 	pvo->pvo_pmap->pm_stats.resident_count--;
1576 	if (pvo->pvo_pte.pte_lo & PVO_WIRED)
1577 		pvo->pvo_pmap->pm_stats.wired_count--;
1578 
1579 	/*
1580 	 * Save the REF/CHG bits into their cache if the page is managed.
1581 	 */
1582 	if (pvo->pvo_vaddr & PVO_MANAGED) {
1583 		register_t ptelo = pvo->pvo_pte.pte_lo;
1584 		struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1585 
1586 		if (pg != NULL) {
1587 			pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1588 		}
1589 		PMAPCOUNT(unmappings);
1590 	} else {
1591 		PMAPCOUNT(kernel_unmappings);
1592 	}
1593 
1594 	/*
1595 	 * Remove the PVO from its lists and return it to the pool.
1596 	 */
1597 	LIST_REMOVE(pvo, pvo_vlink);
1598 	TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1599 	pool_put(pvo->pvo_vaddr & PVO_MANAGED
1600 	    ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo);
1601 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1602 	pmap_pvo_remove_depth--;
1603 #endif
1604 }
1605 
1606 /*
1607  * Insert physical page at pa into the given pmap at virtual address va.
1608  */
1609 int
1610 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
1611 {
1612 	struct mem_region *mp;
1613 	struct pvo_head *pvo_head;
1614 	struct vm_page *pg;
1615 	struct pool *pl;
1616 	register_t pte_lo;
1617 	int s;
1618 	int error;
1619 	u_int pvo_flags;
1620 	u_int was_exec = 0;
1621 
1622 	if (__predict_false(!pmap_initialized)) {
1623 		pvo_head = &pmap_pvo_kunmanaged;
1624 		pl = &pmap_upvo_pool;
1625 		pvo_flags = 0;
1626 		pg = NULL;
1627 		was_exec = PTE_EXEC;
1628 	} else {
1629 		pvo_head = pa_to_pvoh(pa, &pg);
1630 		pl = &pmap_mpvo_pool;
1631 		pvo_flags = PVO_MANAGED;
1632 	}
1633 
1634 	DPRINTFN(ENTER,
1635 	    ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):",
1636 	    pm, va, pa, prot, flags));
1637 
1638 	/*
1639 	 * If this is a managed page, and it's the first reference to the
1640 	 * page clear the execness of the page.  Otherwise fetch the execness.
1641 	 */
1642 	if (pg != NULL)
1643 		was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1644 
1645 	DPRINTFN(ENTER, (" was_exec=%d", was_exec));
1646 
1647 	/*
1648 	 * Assume the page is cache inhibited and access is guarded unless
1649 	 * it's in our available memory array.  If it is in the memory array,
1650 	 * asssume it's in memory coherent memory.
1651 	 */
1652 	pte_lo = PTE_IG;
1653 	if ((flags & PMAP_NC) == 0) {
1654 		for (mp = mem; mp->size; mp++) {
1655 			if (pa >= mp->start && pa < mp->start + mp->size) {
1656 				pte_lo = PTE_M;
1657 				break;
1658 			}
1659 		}
1660 	}
1661 
1662 	if (prot & VM_PROT_WRITE)
1663 		pte_lo |= PTE_BW;
1664 	else
1665 		pte_lo |= PTE_BR;
1666 
1667 	/*
1668 	 * If this was in response to a fault, "pre-fault" the PTE's
1669 	 * changed/referenced bit appropriately.
1670 	 */
1671 	if (flags & VM_PROT_WRITE)
1672 		pte_lo |= PTE_CHG;
1673 	if (flags & (VM_PROT_READ|VM_PROT_WRITE))
1674 		pte_lo |= PTE_REF;
1675 
1676 #if 0
1677 	if (pm == pmap_kernel()) {
1678 		if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) == VM_PROT_READ)
1679 			printf("pmap_pvo_enter: Kernel RO va %#lx pa %#lx\n",
1680 				va, pa);
1681 		if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) == VM_PROT_NONE)
1682 			printf("pmap_pvo_enter: Kernel N/A va %#lx pa %#lx\n",
1683 				va, pa);
1684 	}
1685 #endif
1686 
1687 	/*
1688 	 * We need to know if this page can be executable
1689 	 */
1690 	flags |= (prot & VM_PROT_EXECUTE);
1691 
1692 	/*
1693 	 * Record mapping for later back-translation and pte spilling.
1694 	 * This will overwrite any existing mapping.
1695 	 */
1696 	s = splvm();
1697 	error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags);
1698 	splx(s);
1699 
1700 	/*
1701 	 * Flush the real page from the instruction cache if this page is
1702 	 * mapped executable and cacheable and has not been flushed since
1703 	 * the last time it was modified.
1704 	 */
1705 	if (error == 0 &&
1706             (flags & VM_PROT_EXECUTE) &&
1707             (pte_lo & PTE_I) == 0 &&
1708 	    was_exec == 0) {
1709 		DPRINTFN(ENTER, (" syncicache"));
1710 		PMAPCOUNT(exec_synced);
1711 		pmap_syncicache(pa, PAGE_SIZE);
1712 		if (pg != NULL) {
1713 			pmap_attr_save(pg, PTE_EXEC);
1714 			PMAPCOUNT(exec_cached);
1715 #if defined(DEBUG) || defined(PMAPDEBUG)
1716 			if (pmapdebug & PMAPDEBUG_ENTER)
1717 				printf(" marked-as-exec");
1718 			else if (pmapdebug & PMAPDEBUG_EXEC)
1719 				printf("[pmap_enter: %#lx: marked-as-exec]\n",
1720 				    pg->phys_addr);
1721 
1722 #endif
1723 		}
1724 	}
1725 
1726 	DPRINTFN(ENTER, (": error=%d\n", error));
1727 
1728 	return error;
1729 }
1730 
1731 void
1732 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
1733 {
1734 	struct mem_region *mp;
1735 	register_t pte_lo;
1736 	register_t msr;
1737 	int error;
1738 	int s;
1739 
1740 	if (va < VM_MIN_KERNEL_ADDRESS)
1741 		panic("pmap_kenter_pa: attempt to enter "
1742 		    "non-kernel address %#lx!", va);
1743 
1744 	DPRINTFN(KENTER,
1745 	    ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot));
1746 
1747 	/*
1748 	 * Assume the page is cache inhibited and access is guarded unless
1749 	 * it's in our available memory array.  If it is in the memory array,
1750 	 * asssume it's in memory coherent memory.
1751 	 */
1752 	pte_lo = PTE_IG;
1753 	if ((prot & PMAP_NC) == 0) {
1754 		for (mp = mem; mp->size; mp++) {
1755 			if (pa >= mp->start && pa < mp->start + mp->size) {
1756 				pte_lo = PTE_M;
1757 				break;
1758 			}
1759 		}
1760 	}
1761 
1762 	if (prot & VM_PROT_WRITE)
1763 		pte_lo |= PTE_BW;
1764 	else
1765 		pte_lo |= PTE_BR;
1766 
1767 	/*
1768 	 * We don't care about REF/CHG on PVOs on the unmanaged list.
1769 	 */
1770 	s = splvm();
1771 	msr = pmap_interrupts_off();
1772 	error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool,
1773 	    &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED);
1774 	pmap_interrupts_restore(msr);
1775 	splx(s);
1776 
1777 	if (error != 0)
1778 		panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d",
1779 		      va, pa, error);
1780 }
1781 
1782 void
1783 pmap_kremove(vaddr_t va, vsize_t len)
1784 {
1785 	if (va < VM_MIN_KERNEL_ADDRESS)
1786 		panic("pmap_kremove: attempt to remove "
1787 		    "non-kernel address %#lx!", va);
1788 
1789 	DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len));
1790 	pmap_remove(pmap_kernel(), va, va + len);
1791 }
1792 
1793 /*
1794  * Remove the given range of mapping entries.
1795  */
1796 void
1797 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
1798 {
1799 	struct pvo_entry *pvo;
1800 	register_t msr;
1801 	int pteidx;
1802 	int s;
1803 
1804 	for (; va < endva; va += PAGE_SIZE) {
1805 		s = splvm();
1806 		msr = pmap_interrupts_off();
1807 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
1808 		if (pvo != NULL) {
1809 			pmap_pvo_remove(pvo, pteidx);
1810 		}
1811 		pmap_interrupts_restore(msr);
1812 		splx(s);
1813 	}
1814 }
1815 
1816 /*
1817  * Get the physical page address for the given pmap/virtual address.
1818  */
1819 boolean_t
1820 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
1821 {
1822 	struct pvo_entry *pvo;
1823 	register_t msr;
1824 	int s;
1825 
1826 	/*
1827 	 * If this is a kernel pmap lookup, also check the battable
1828 	 * and if we get a hit, translate the VA to a PA using the
1829 	 * BAT entries.  Don't check for VM_MAX_KENREL_ADDRESS is
1830 	 * that will wrap back to 0.
1831 	 */
1832 	if (pm == pmap_kernel() &&
1833 	    (va < VM_MIN_KERNEL_ADDRESS ||
1834 	     (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
1835 		register_t batu = battable[va >> ADDR_SR_SHFT].batu;
1836 		KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
1837 		if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) {
1838 			register_t batl = battable[va >> ADDR_SR_SHFT].batl;
1839 			register_t mask = (~(batu & BAT_BL) << 15) & ~0x1ffffL;
1840 			*pap = (batl & mask) | (va & ~mask);
1841 			return TRUE;
1842 		}
1843 		return FALSE;
1844 	}
1845 
1846 	s = splvm();
1847 	msr = pmap_interrupts_off();
1848 	pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
1849 	if (pvo != NULL) {
1850 		PMAP_PVO_CHECK(pvo);		/* sanity check */
1851 		*pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF);
1852 	}
1853 	pmap_interrupts_restore(msr);
1854 	splx(s);
1855 	return pvo != NULL;
1856 }
1857 
1858 /*
1859  * Lower the protection on the specified range of this pmap.
1860  *
1861  * There are only two cases: either the protection is going to 0,
1862  * or it is going to read-only.
1863  */
1864 void
1865 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
1866 {
1867 	struct pvo_entry *pvo;
1868 	volatile struct pte *pt;
1869 	register_t msr;
1870 	int s;
1871 	int pteidx;
1872 
1873 	/*
1874 	 * Since this routine only downgrades protection, we should
1875 	 * always be called without WRITE permisison.
1876 	 */
1877 	KASSERT((prot & VM_PROT_WRITE) == 0);
1878 
1879 	/*
1880 	 * If there is no protection, this is equivalent to
1881 	 * remove the pmap from the pmap.
1882 	 */
1883 	if ((prot & VM_PROT_READ) == 0) {
1884 		pmap_remove(pm, va, endva);
1885 		return;
1886 	}
1887 
1888 	s = splvm();
1889 	msr = pmap_interrupts_off();
1890 
1891 	for (; va < endva; va += PAGE_SIZE) {
1892 		pvo = pmap_pvo_find_va(pm, va, &pteidx);
1893 		if (pvo == NULL)
1894 			continue;
1895 		PMAP_PVO_CHECK(pvo);		/* sanity check */
1896 
1897 		/*
1898 		 * Revoke executable if asked to do so.
1899 		 */
1900 		if ((prot & VM_PROT_EXECUTE) == 0)
1901 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1902 
1903 #if 0
1904 		/*
1905 		 * If the page is already read-only, no change
1906 		 * needs to be made.
1907 		 */
1908 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
1909 			continue;
1910 #endif
1911 		/*
1912 		 * Grab the PTE pointer before we diddle with
1913 		 * the cached PTE copy.
1914 		 */
1915 		pt = pmap_pvo_to_pte(pvo, pteidx);
1916 		/*
1917 		 * Change the protection of the page.
1918 		 */
1919 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
1920 		pvo->pvo_pte.pte_lo |= PTE_BR;
1921 
1922 		/*
1923 		 * If the PVO is in the page table, update
1924 		 * that pte at well.
1925 		 */
1926 		if (pt != NULL) {
1927 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1928 			PMAPCOUNT(ptes_changed);
1929 		}
1930 
1931 		PMAP_PVO_CHECK(pvo);		/* sanity check */
1932 	}
1933 
1934 	pmap_interrupts_restore(msr);
1935 	splx(s);
1936 }
1937 
1938 void
1939 pmap_unwire(pmap_t pm, vaddr_t va)
1940 {
1941 	struct pvo_entry *pvo;
1942 	register_t msr;
1943 	int s;
1944 
1945 	s = splvm();
1946 	msr = pmap_interrupts_off();
1947 
1948 	pvo = pmap_pvo_find_va(pm, va, NULL);
1949 	if (pvo != NULL) {
1950 		if (pvo->pvo_vaddr & PVO_WIRED) {
1951 			pvo->pvo_vaddr &= ~PVO_WIRED;
1952 			pm->pm_stats.wired_count--;
1953 		}
1954 		PMAP_PVO_CHECK(pvo);		/* sanity check */
1955 	}
1956 
1957 	pmap_interrupts_restore(msr);
1958 	splx(s);
1959 }
1960 
1961 /*
1962  * Lower the protection on the specified physical page.
1963  *
1964  * There are only two cases: either the protection is going to 0,
1965  * or it is going to read-only.
1966  */
1967 void
1968 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1969 {
1970 	struct pvo_head *pvo_head;
1971 	struct pvo_entry *pvo, *next_pvo;
1972 	volatile struct pte *pt;
1973 	register_t msr;
1974 	int s;
1975 
1976 	/*
1977 	 * Since this routine only downgrades protection, if the
1978 	 * maximal protection is desired, there isn't any change
1979 	 * to be made.
1980 	 */
1981 	KASSERT((prot & VM_PROT_WRITE) == 0);
1982 	if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) == (VM_PROT_READ|VM_PROT_WRITE))
1983 		return;
1984 
1985 	s = splvm();
1986 	msr = pmap_interrupts_off();
1987 
1988 	/*
1989 	 * When UVM reuses a page, it does a pmap_page_protect with
1990 	 * VM_PROT_NONE.  At that point, we can clear the exec flag
1991 	 * since we know the page will have different contents.
1992 	 */
1993 	if ((prot & VM_PROT_READ) == 0) {
1994 		DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n",
1995 		    pg->phys_addr));
1996 		if (pmap_attr_fetch(pg) & PTE_EXEC) {
1997 			PMAPCOUNT(exec_uncached_page_protect);
1998 			pmap_attr_clear(pg, PTE_EXEC);
1999 		}
2000 	}
2001 
2002 	pvo_head = vm_page_to_pvoh(pg);
2003 	for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2004 		next_pvo = LIST_NEXT(pvo, pvo_vlink);
2005 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2006 
2007 		/*
2008 		 * Downgrading to no mapping at all, we just remove the entry.
2009 		 */
2010 		if ((prot & VM_PROT_READ) == 0) {
2011 			pmap_pvo_remove(pvo, -1);
2012 			continue;
2013 		}
2014 
2015 		/*
2016 		 * If EXEC permission is being revoked, just clear the
2017 		 * flag in the PVO.
2018 		 */
2019 		if ((prot & VM_PROT_EXECUTE) == 0)
2020 			pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
2021 
2022 		/*
2023 		 * If this entry is already RO, don't diddle with the
2024 		 * page table.
2025 		 */
2026 		if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2027 			PMAP_PVO_CHECK(pvo);
2028 			continue;
2029 		}
2030 
2031 		/*
2032 		 * Grab the PTE before the we diddle the bits so
2033 		 * pvo_to_pte can verify the pte contents are as
2034 		 * expected.
2035 		 */
2036 		pt = pmap_pvo_to_pte(pvo, -1);
2037 		pvo->pvo_pte.pte_lo &= ~PTE_PP;
2038 		pvo->pvo_pte.pte_lo |= PTE_BR;
2039 		if (pt != NULL) {
2040 			pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2041 			PMAPCOUNT(ptes_changed);
2042 		}
2043 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2044 	}
2045 
2046 	pmap_interrupts_restore(msr);
2047 	splx(s);
2048 }
2049 
2050 /*
2051  * Activate the address space for the specified process.  If the process
2052  * is the current process, load the new MMU context.
2053  */
2054 void
2055 pmap_activate(struct lwp *l)
2056 {
2057 	struct pcb *pcb = &l->l_addr->u_pcb;
2058 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2059 
2060 	DPRINTFN(ACTIVATE,
2061 	    ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp));
2062 
2063 	/*
2064 	 * XXX Normally performed in cpu_fork().
2065 	 */
2066 	if (pcb->pcb_pm != pmap) {
2067 		pcb->pcb_pm = pmap;
2068 		pcb->pcb_pmreal = pmap;
2069 	}
2070 
2071 	/*
2072 	 * In theory, the SR registers need only be valid on return
2073 	 * to user space wait to do them there.
2074 	 */
2075 	if (l == curlwp) {
2076 		/* Store pointer to new current pmap. */
2077 		curpm = pmap;
2078 	}
2079 }
2080 
2081 /*
2082  * Deactivate the specified process's address space.
2083  */
2084 void
2085 pmap_deactivate(struct lwp *l)
2086 {
2087 }
2088 
2089 boolean_t
2090 pmap_query_bit(struct vm_page *pg, int ptebit)
2091 {
2092 	struct pvo_entry *pvo;
2093 	volatile struct pte *pt;
2094 	register_t msr;
2095 	int s;
2096 
2097 	if (pmap_attr_fetch(pg) & ptebit)
2098 		return TRUE;
2099 	s = splvm();
2100 	msr = pmap_interrupts_off();
2101 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2102 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2103 		/*
2104 		 * See if we saved the bit off.  If so cache, it and return
2105 		 * success.
2106 		 */
2107 		if (pvo->pvo_pte.pte_lo & ptebit) {
2108 			pmap_attr_save(pg, ptebit);
2109 			PMAP_PVO_CHECK(pvo);		/* sanity check */
2110 			pmap_interrupts_restore(msr);
2111 			splx(s);
2112 			return TRUE;
2113 		}
2114 	}
2115 	/*
2116 	 * No luck, now go thru the hard part of looking at the ptes
2117 	 * themselves.  Sync so any pending REF/CHG bits are flushed
2118 	 * to the PTEs.
2119 	 */
2120 	SYNC();
2121 	LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2122 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2123 		/*
2124 		 * See if this pvo have a valid PTE.  If so, fetch the
2125 		 * REF/CHG bits from the valid PTE.  If the appropriate
2126 		 * ptebit is set, cache, it and return success.
2127 		 */
2128 		pt = pmap_pvo_to_pte(pvo, -1);
2129 		if (pt != NULL) {
2130 			pmap_pte_synch(pt, &pvo->pvo_pte);
2131 			if (pvo->pvo_pte.pte_lo & ptebit) {
2132 				pmap_attr_save(pg, ptebit);
2133 				PMAP_PVO_CHECK(pvo);		/* sanity check */
2134 				pmap_interrupts_restore(msr);
2135 				splx(s);
2136 				return TRUE;
2137 			}
2138 		}
2139 	}
2140 	pmap_interrupts_restore(msr);
2141 	splx(s);
2142 	return FALSE;
2143 }
2144 
2145 boolean_t
2146 pmap_clear_bit(struct vm_page *pg, int ptebit)
2147 {
2148 	struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2149 	struct pvo_entry *pvo;
2150 	volatile struct pte *pt;
2151 	register_t msr;
2152 	int rv = 0;
2153 	int s;
2154 
2155 	s = splvm();
2156 	msr = pmap_interrupts_off();
2157 
2158 	/*
2159 	 * Fetch the cache value
2160 	 */
2161 	rv |= pmap_attr_fetch(pg);
2162 
2163 	/*
2164 	 * Clear the cached value.
2165 	 */
2166 	pmap_attr_clear(pg, ptebit);
2167 
2168 	/*
2169 	 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2170 	 * can reset the right ones).  Note that since the pvo entries and
2171 	 * list heads are accessed via BAT0 and are never placed in the
2172 	 * page table, we don't have to worry about further accesses setting
2173 	 * the REF/CHG bits.
2174 	 */
2175 	SYNC();
2176 
2177 	/*
2178 	 * For each pvo entry, clear pvo's ptebit.  If this pvo have a
2179 	 * valid PTE.  If so, clear the ptebit from the valid PTE.
2180 	 */
2181 	LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2182 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2183 		pt = pmap_pvo_to_pte(pvo, -1);
2184 		if (pt != NULL) {
2185 			/*
2186 			 * Only sync the PTE if the bit we are looking
2187 			 * for is not already set.
2188 			 */
2189 			if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2190 				pmap_pte_synch(pt, &pvo->pvo_pte);
2191 			/*
2192 			 * If the bit we are looking for was already set,
2193 			 * clear that bit in the pte.
2194 			 */
2195 			if (pvo->pvo_pte.pte_lo & ptebit)
2196 				pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2197 		}
2198 		rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2199 		pvo->pvo_pte.pte_lo &= ~ptebit;
2200 		PMAP_PVO_CHECK(pvo);		/* sanity check */
2201 	}
2202 	pmap_interrupts_restore(msr);
2203 	splx(s);
2204 	/*
2205 	 * If we are clearing the modify bit and this page was marked EXEC
2206 	 * and the user of the page thinks the page was modified, then we
2207 	 * need to clean it from the icache if it's mapped or clear the EXEC
2208 	 * bit if it's not mapped.  The page itself might not have the CHG
2209 	 * bit set if the modification was done via DMA to the page.
2210 	 */
2211 	if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2212 		if (LIST_EMPTY(pvoh)) {
2213 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n",
2214 			    pg->phys_addr));
2215 			pmap_attr_clear(pg, PTE_EXEC);
2216 			PMAPCOUNT(exec_uncached_clear_modify);
2217 		} else {
2218 			DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n",
2219 			    pg->phys_addr));
2220 			pmap_syncicache(pg->phys_addr, PAGE_SIZE);
2221 			PMAPCOUNT(exec_synced_clear_modify);
2222 		}
2223 	}
2224 	return (rv & ptebit) != 0;
2225 }
2226 
2227 void
2228 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2229 {
2230 	struct pvo_entry *pvo;
2231 	size_t offset = va & ADDR_POFF;
2232 	int s;
2233 
2234 	s = splvm();
2235 	while (len > 0) {
2236 		size_t seglen = PAGE_SIZE - offset;
2237 		if (seglen > len)
2238 			seglen = len;
2239 		pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2240 		if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) {
2241 			pmap_syncicache(
2242 			    (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2243 			PMAP_PVO_CHECK(pvo);
2244 		}
2245 		va += seglen;
2246 		len -= seglen;
2247 		offset = 0;
2248 	}
2249 	splx(s);
2250 }
2251 
2252 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2253 void
2254 pmap_pte_print(volatile struct pte *pt)
2255 {
2256 	printf("PTE %p: ", pt);
2257 	/* High word: */
2258 	printf("0x%08lx: [", pt->pte_hi);
2259 	printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2260 	printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2261 	printf("0x%06lx 0x%02lx",
2262 	    (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2263 	    pt->pte_hi & PTE_API);
2264 	printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt));
2265 	/* Low word: */
2266 	printf(" 0x%08lx: [", pt->pte_lo);
2267 	printf("0x%05lx... ", pt->pte_lo >> 12);
2268 	printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2269 	printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2270 	printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2271 	printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2272 	printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2273 	printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2274 	switch (pt->pte_lo & PTE_PP) {
2275 	case PTE_BR: printf("br]\n"); break;
2276 	case PTE_BW: printf("bw]\n"); break;
2277 	case PTE_SO: printf("so]\n"); break;
2278 	case PTE_SW: printf("sw]\n"); break;
2279 	}
2280 }
2281 #endif
2282 
2283 #if defined(DDB)
2284 void
2285 pmap_pteg_check(void)
2286 {
2287 	volatile struct pte *pt;
2288 	int i;
2289 	int ptegidx;
2290 	u_int p_valid = 0;
2291 	u_int s_valid = 0;
2292 	u_int invalid = 0;
2293 
2294 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2295 		for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2296 			if (pt->pte_hi & PTE_VALID) {
2297 				if (pt->pte_hi & PTE_HID)
2298 					s_valid++;
2299 				else
2300 					p_valid++;
2301 			} else
2302 				invalid++;
2303 		}
2304 	}
2305 	printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2306 		p_valid, p_valid, s_valid, s_valid,
2307 		invalid, invalid);
2308 }
2309 
2310 void
2311 pmap_print_mmuregs(void)
2312 {
2313 	int i;
2314 	u_int cpuvers;
2315 	vaddr_t addr;
2316 	register_t soft_sr[16];
2317 	struct bat soft_ibat[4];
2318 	struct bat soft_dbat[4];
2319 	register_t sdr1;
2320 
2321 	cpuvers = MFPVR() >> 16;
2322 
2323 	__asm __volatile ("mfsdr1 %0" : "=r"(sdr1));
2324 	for (i=0; i<16; i++) {
2325 		soft_sr[i] = MFSRIN(addr);
2326 		addr += (1 << ADDR_SR_SHFT);
2327 	}
2328 
2329 	/* read iBAT (601: uBAT) registers */
2330 	__asm __volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2331 	__asm __volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2332 	__asm __volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2333 	__asm __volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2334 	__asm __volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2335 	__asm __volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2336 	__asm __volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2337 	__asm __volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2338 
2339 
2340 	if (cpuvers != MPC601) {
2341 		/* read dBAT registers */
2342 		__asm __volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2343 		__asm __volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2344 		__asm __volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2345 		__asm __volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2346 		__asm __volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2347 		__asm __volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2348 		__asm __volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2349 		__asm __volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2350 	}
2351 
2352 	printf("SDR1:\t%#lx\n", sdr1);
2353 	printf("SR[]:\t");
2354 	addr = 0;
2355 	for (i=0; i<4; i++)
2356 		printf("0x%08lx,   ", soft_sr[i]);
2357 	printf("\n\t");
2358 	for ( ; i<8; i++)
2359 		printf("0x%08lx,   ", soft_sr[i]);
2360 	printf("\n\t");
2361 	for ( ; i<12; i++)
2362 		printf("0x%08lx,   ", soft_sr[i]);
2363 	printf("\n\t");
2364 	for ( ; i<16; i++)
2365 		printf("0x%08lx,   ", soft_sr[i]);
2366 	printf("\n");
2367 
2368 	printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2369 	for (i=0; i<4; i++) {
2370 		printf("0x%08lx 0x%08lx, ",
2371 			soft_ibat[i].batu, soft_ibat[i].batl);
2372 		if (i == 1)
2373 			printf("\n\t");
2374 	}
2375 	if (cpuvers != MPC601) {
2376 		printf("\ndBAT[]:\t");
2377 		for (i=0; i<4; i++) {
2378 			printf("0x%08lx 0x%08lx, ",
2379 				soft_dbat[i].batu, soft_dbat[i].batl);
2380 			if (i == 1)
2381 				printf("\n\t");
2382 		}
2383 	}
2384 	printf("\n");
2385 }
2386 
2387 void
2388 pmap_print_pte(pmap_t pm, vaddr_t va)
2389 {
2390 	struct pvo_entry *pvo;
2391 	volatile struct pte *pt;
2392 	int pteidx;
2393 
2394 	pvo = pmap_pvo_find_va(pm, va, &pteidx);
2395 	if (pvo != NULL) {
2396 		pt = pmap_pvo_to_pte(pvo, pteidx);
2397 		if (pt != NULL) {
2398 			printf("VA %#lx -> %p -> %s %#lx, %#lx\n",
2399 				va, pt,
2400 				pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2401 				pt->pte_hi, pt->pte_lo);
2402 		} else {
2403 			printf("No valid PTE found\n");
2404 		}
2405 	} else {
2406 		printf("Address not in pmap\n");
2407 	}
2408 }
2409 
2410 void
2411 pmap_pteg_dist(void)
2412 {
2413 	struct pvo_entry *pvo;
2414 	int ptegidx;
2415 	int depth;
2416 	int max_depth = 0;
2417 	unsigned int depths[64];
2418 
2419 	memset(depths, 0, sizeof(depths));
2420 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2421 		depth = 0;
2422 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2423 			depth++;
2424 		}
2425 		if (depth > max_depth)
2426 			max_depth = depth;
2427 		if (depth > 63)
2428 			depth = 63;
2429 		depths[depth]++;
2430 	}
2431 
2432 	for (depth = 0; depth < 64; depth++) {
2433 		printf("  [%2d]: %8u", depth, depths[depth]);
2434 		if ((depth & 3) == 3)
2435 			printf("\n");
2436 		if (depth == max_depth)
2437 			break;
2438 	}
2439 	if ((depth & 3) != 3)
2440 		printf("\n");
2441 	printf("Max depth found was %d\n", max_depth);
2442 }
2443 #endif /* DEBUG */
2444 
2445 #if defined(PMAPCHECK) || defined(DEBUG)
2446 void
2447 pmap_pvo_verify(void)
2448 {
2449 	int ptegidx;
2450 	int s;
2451 
2452 	s = splvm();
2453 	for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2454 		struct pvo_entry *pvo;
2455 		TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2456 			if ((uintptr_t) pvo >= SEGMENT_LENGTH)
2457 				panic("pmap_pvo_verify: invalid pvo %p "
2458 				    "on list %#x", pvo, ptegidx);
2459 			pmap_pvo_check(pvo);
2460 		}
2461 	}
2462 	splx(s);
2463 }
2464 #endif /* PMAPCHECK */
2465 
2466 
2467 void *
2468 pmap_pool_ualloc(struct pool *pp, int flags)
2469 {
2470 	struct pvo_page *pvop;
2471 
2472 	pvop = SIMPLEQ_FIRST(&pmap_upvop_head);
2473 	if (pvop != NULL) {
2474 		pmap_upvop_free--;
2475 		SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link);
2476 		return pvop;
2477 	}
2478 	if (uvm.page_init_done != TRUE) {
2479 		return (void *) uvm_pageboot_alloc(PAGE_SIZE);
2480 	}
2481 	return pmap_pool_malloc(pp, flags);
2482 }
2483 
2484 void *
2485 pmap_pool_malloc(struct pool *pp, int flags)
2486 {
2487 	struct pvo_page *pvop;
2488 	struct vm_page *pg;
2489 
2490 	pvop = SIMPLEQ_FIRST(&pmap_mpvop_head);
2491 	if (pvop != NULL) {
2492 		pmap_mpvop_free--;
2493 		SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link);
2494 		return pvop;
2495 	}
2496  again:
2497 	pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE,
2498 	    UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256);
2499 	if (__predict_false(pg == NULL)) {
2500 		if (flags & PR_WAITOK) {
2501 			uvm_wait("plpg");
2502 			goto again;
2503 		} else {
2504 			return (0);
2505 		}
2506 	}
2507 	return (void *) VM_PAGE_TO_PHYS(pg);
2508 }
2509 
2510 void
2511 pmap_pool_ufree(struct pool *pp, void *va)
2512 {
2513 	struct pvo_page *pvop;
2514 #if 0
2515 	if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) {
2516 		pmap_pool_mfree(va, size, tag);
2517 		return;
2518 	}
2519 #endif
2520 	pvop = va;
2521 	SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link);
2522 	pmap_upvop_free++;
2523 	if (pmap_upvop_free > pmap_upvop_maxfree)
2524 		pmap_upvop_maxfree = pmap_upvop_free;
2525 }
2526 
2527 void
2528 pmap_pool_mfree(struct pool *pp, void *va)
2529 {
2530 	struct pvo_page *pvop;
2531 
2532 	pvop = va;
2533 	SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link);
2534 	pmap_mpvop_free++;
2535 	if (pmap_mpvop_free > pmap_mpvop_maxfree)
2536 		pmap_mpvop_maxfree = pmap_mpvop_free;
2537 #if 0
2538 	uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va));
2539 #endif
2540 }
2541 
2542 /*
2543  * This routine in bootstraping to steal to-be-managed memory (which will
2544  * then be unmanaged).  We use it to grab from the first 256MB for our
2545  * pmap needs and above 256MB for other stuff.
2546  */
2547 vaddr_t
2548 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2549 {
2550 	vsize_t size;
2551 	vaddr_t va;
2552 	paddr_t pa = 0;
2553 	int npgs, bank;
2554 	struct vm_physseg *ps;
2555 
2556 	if (uvm.page_init_done == TRUE)
2557 		panic("pmap_steal_memory: called _after_ bootstrap");
2558 
2559 	*vstartp = VM_MIN_KERNEL_ADDRESS;
2560 	*vendp = VM_MAX_KERNEL_ADDRESS;
2561 
2562 	size = round_page(vsize);
2563 	npgs = atop(size);
2564 
2565 	/*
2566 	 * PA 0 will never be among those given to UVM so we can use it
2567 	 * to indicate we couldn't steal any memory.
2568 	 */
2569 	for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) {
2570 		if (ps->free_list == VM_FREELIST_FIRST256 &&
2571 		    ps->avail_end - ps->avail_start >= npgs) {
2572 			pa = ptoa(ps->avail_start);
2573 			break;
2574 		}
2575 	}
2576 
2577 	if (pa == 0)
2578 		panic("pmap_steal_memory: no approriate memory to steal!");
2579 
2580 	ps->avail_start += npgs;
2581 	ps->start += npgs;
2582 
2583 	/*
2584 	 * If we've used up all the pages in the segment, remove it and
2585 	 * compact the list.
2586 	 */
2587 	if (ps->avail_start == ps->end) {
2588 		/*
2589 		 * If this was the last one, then a very bad thing has occurred
2590 		 */
2591 		if (--vm_nphysseg == 0)
2592 			panic("pmap_steal_memory: out of memory!");
2593 
2594 		printf("pmap_steal_memory: consumed bank %d\n", bank);
2595 		for (; bank < vm_nphysseg; bank++, ps++) {
2596 			ps[0] = ps[1];
2597 		}
2598 	}
2599 
2600 	va = (vaddr_t) pa;
2601 	memset((caddr_t) va, 0, size);
2602 	pmap_pages_stolen += npgs;
2603 #ifdef DEBUG
2604 	if (pmapdebug && npgs > 1) {
2605 		u_int cnt = 0;
2606 		for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++)
2607 			cnt += ps->avail_end - ps->avail_start;
2608 		printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2609 		    npgs, pmap_pages_stolen, cnt);
2610 	}
2611 #endif
2612 
2613 	return va;
2614 }
2615 
2616 /*
2617  * Find a chuck of memory with right size and alignment.
2618  */
2619 void *
2620 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2621 {
2622 	struct mem_region *mp;
2623 	paddr_t s, e;
2624 	int i, j;
2625 
2626 	size = round_page(size);
2627 
2628 	DPRINTFN(BOOT,
2629 	    ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d",
2630 	    size, alignment, at_end));
2631 
2632 	if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
2633 		panic("pmap_boot_find_memory: invalid alignment %lx",
2634 		    alignment);
2635 
2636 	if (at_end) {
2637 		if (alignment != PAGE_SIZE)
2638 			panic("pmap_boot_find_memory: invalid ending "
2639 			    "alignment %lx", alignment);
2640 
2641 		for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
2642 			s = mp->start + mp->size - size;
2643 			if (s >= mp->start && mp->size >= size) {
2644 				DPRINTFN(BOOT,(": %lx\n", s));
2645 				DPRINTFN(BOOT,
2646 				    ("pmap_boot_find_memory: b-avail[%d] start "
2647 				     "0x%lx size 0x%lx\n", mp - avail,
2648 				     mp->start, mp->size));
2649 				mp->size -= size;
2650 				DPRINTFN(BOOT,
2651 				    ("pmap_boot_find_memory: a-avail[%d] start "
2652 				     "0x%lx size 0x%lx\n", mp - avail,
2653 				     mp->start, mp->size));
2654 				return (void *) s;
2655 			}
2656 		}
2657 		panic("pmap_boot_find_memory: no available memory");
2658 	}
2659 
2660 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
2661 		s = (mp->start + alignment - 1) & ~(alignment-1);
2662 		e = s + size;
2663 
2664 		/*
2665 		 * Is the calculated region entirely within the region?
2666 		 */
2667 		if (s < mp->start || e > mp->start + mp->size)
2668 			continue;
2669 
2670 		DPRINTFN(BOOT,(": %lx\n", s));
2671 		if (s == mp->start) {
2672 			/*
2673 			 * If the block starts at the beginning of region,
2674 			 * adjust the size & start. (the region may now be
2675 			 * zero in length)
2676 			 */
2677 			DPRINTFN(BOOT,
2678 			    ("pmap_boot_find_memory: b-avail[%d] start "
2679 			     "0x%lx size 0x%lx\n", i, mp->start, mp->size));
2680 			mp->start += size;
2681 			mp->size -= size;
2682 			DPRINTFN(BOOT,
2683 			    ("pmap_boot_find_memory: a-avail[%d] start "
2684 			     "0x%lx size 0x%lx\n", i, mp->start, mp->size));
2685 		} else if (e == mp->start + mp->size) {
2686 			/*
2687 			 * If the block starts at the beginning of region,
2688 			 * adjust only the size.
2689 			 */
2690 			DPRINTFN(BOOT,
2691 			    ("pmap_boot_find_memory: b-avail[%d] start "
2692 			     "0x%lx size 0x%lx\n", i, mp->start, mp->size));
2693 			mp->size -= size;
2694 			DPRINTFN(BOOT,
2695 			    ("pmap_boot_find_memory: a-avail[%d] start "
2696 			     "0x%lx size 0x%lx\n", i, mp->start, mp->size));
2697 		} else {
2698 			/*
2699 			 * Block is in the middle of the region, so we
2700 			 * have to split it in two.
2701 			 */
2702 			for (j = avail_cnt; j > i + 1; j--) {
2703 				avail[j] = avail[j-1];
2704 			}
2705 			DPRINTFN(BOOT,
2706 			    ("pmap_boot_find_memory: b-avail[%d] start "
2707 			     "0x%lx size 0x%lx\n", i, mp->start, mp->size));
2708 			mp[1].start = e;
2709 			mp[1].size = mp[0].start + mp[0].size - e;
2710 			mp[0].size = s - mp[0].start;
2711 			avail_cnt++;
2712 			for (; i < avail_cnt; i++) {
2713 				DPRINTFN(BOOT,
2714 				    ("pmap_boot_find_memory: a-avail[%d] "
2715 				     "start 0x%lx size 0x%lx\n", i,
2716 				     avail[i].start, avail[i].size));
2717 			}
2718 		}
2719 		return (void *) s;
2720 	}
2721 	panic("pmap_boot_find_memory: not enough memory for "
2722 	    "%lx/%lx allocation?", size, alignment);
2723 }
2724 
2725 /*
2726  * This is not part of the defined PMAP interface and is specific to the
2727  * PowerPC architecture.  This is called during initppc, before the system
2728  * is really initialized.
2729  */
2730 void
2731 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
2732 {
2733 	struct mem_region *mp, tmp;
2734 	paddr_t s, e;
2735 	psize_t size;
2736 	int i, j;
2737 
2738 	/*
2739 	 * Get memory.
2740 	 */
2741 	mem_regions(&mem, &avail);
2742 #if defined(DEBUG)
2743 	if (pmapdebug & PMAPDEBUG_BOOT) {
2744 		printf("pmap_bootstrap: memory configuration:\n");
2745 		for (mp = mem; mp->size; mp++) {
2746 			printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n",
2747 				mp->start, mp->size);
2748 		}
2749 		for (mp = avail; mp->size; mp++) {
2750 			printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n",
2751 				mp->start, mp->size);
2752 		}
2753 	}
2754 #endif
2755 
2756 	/*
2757 	 * Find out how much physical memory we have and in how many chunks.
2758 	 */
2759 	for (mem_cnt = 0, mp = mem; mp->size; mp++) {
2760 		if (mp->start >= pmap_memlimit)
2761 			continue;
2762 		if (mp->start + mp->size > pmap_memlimit) {
2763 			size = pmap_memlimit - mp->start;
2764 			physmem += btoc(size);
2765 		} else {
2766 			physmem += btoc(mp->size);
2767 		}
2768 		mem_cnt++;
2769 	}
2770 
2771 	/*
2772 	 * Count the number of available entries.
2773 	 */
2774 	for (avail_cnt = 0, mp = avail; mp->size; mp++)
2775 		avail_cnt++;
2776 
2777 	/*
2778 	 * Page align all regions.
2779 	 */
2780 	kernelstart = trunc_page(kernelstart);
2781 	kernelend = round_page(kernelend);
2782 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
2783 		s = round_page(mp->start);
2784 		mp->size -= (s - mp->start);
2785 		mp->size = trunc_page(mp->size);
2786 		mp->start = s;
2787 		e = mp->start + mp->size;
2788 
2789 		DPRINTFN(BOOT,
2790 		    ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n",
2791 		    i, mp->start, mp->size));
2792 
2793 		/*
2794 		 * Don't allow the end to run beyond our artificial limit
2795 		 */
2796 		if (e > pmap_memlimit)
2797 			e = pmap_memlimit;
2798 
2799 		/*
2800 		 * Is this region empty or strange?  skip it.
2801 		 */
2802 		if (e <= s) {
2803 			mp->start = 0;
2804 			mp->size = 0;
2805 			continue;
2806 		}
2807 
2808 		/*
2809 		 * Does this overlap the beginning of kernel?
2810 		 *   Does extend past the end of the kernel?
2811 		 */
2812 		else if (s < kernelstart && e > kernelstart) {
2813 			if (e > kernelend) {
2814 				avail[avail_cnt].start = kernelend;
2815 				avail[avail_cnt].size = e - kernelend;
2816 				avail_cnt++;
2817 			}
2818 			mp->size = kernelstart - s;
2819 		}
2820 		/*
2821 		 * Check whether this region overlaps the end of the kernel.
2822 		 */
2823 		else if (s < kernelend && e > kernelend) {
2824 			mp->start = kernelend;
2825 			mp->size = e - kernelend;
2826 		}
2827 		/*
2828 		 * Look whether this regions is completely inside the kernel.
2829 		 * Nuke it if it does.
2830 		 */
2831 		else if (s >= kernelstart && e <= kernelend) {
2832 			mp->start = 0;
2833 			mp->size = 0;
2834 		}
2835 		/*
2836 		 * If the user imposed a memory limit, enforce it.
2837 		 */
2838 		else if (s >= pmap_memlimit) {
2839 			mp->start = -PAGE_SIZE;	/* let's know why */
2840 			mp->size = 0;
2841 		}
2842 		else {
2843 			mp->start = s;
2844 			mp->size = e - s;
2845 		}
2846 		DPRINTFN(BOOT,
2847 		    ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n",
2848 		    i, mp->start, mp->size));
2849 	}
2850 
2851 	/*
2852 	 * Move (and uncount) all the null return to the end.
2853 	 */
2854 	for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
2855 		if (mp->size == 0) {
2856 			tmp = avail[i];
2857 			avail[i] = avail[--avail_cnt];
2858 			avail[avail_cnt] = avail[i];
2859 		}
2860 	}
2861 
2862 	/*
2863 	 * (Bubble)sort them into asecnding order.
2864 	 */
2865 	for (i = 0; i < avail_cnt; i++) {
2866 		for (j = i + 1; j < avail_cnt; j++) {
2867 			if (avail[i].start > avail[j].start) {
2868 				tmp = avail[i];
2869 				avail[i] = avail[j];
2870 				avail[j] = tmp;
2871 			}
2872 		}
2873 	}
2874 
2875 	/*
2876 	 * Make sure they don't overlap.
2877 	 */
2878 	for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
2879 		if (mp[0].start + mp[0].size > mp[1].start) {
2880 			mp[0].size = mp[1].start - mp[0].start;
2881 		}
2882 		DPRINTFN(BOOT,
2883 		    ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n",
2884 		    i, mp->start, mp->size));
2885 	}
2886 	DPRINTFN(BOOT,
2887 	    ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n",
2888 	    i, mp->start, mp->size));
2889 
2890 #ifdef	PTEGCOUNT
2891 	pmap_pteg_cnt = PTEGCOUNT;
2892 #else /* PTEGCOUNT */
2893 	pmap_pteg_cnt = 0x1000;
2894 
2895 	while (pmap_pteg_cnt < physmem)
2896 		pmap_pteg_cnt <<= 1;
2897 
2898 	pmap_pteg_cnt >>= 1;
2899 #endif /* PTEGCOUNT */
2900 
2901 	/*
2902 	 * Find suitably aligned memory for PTEG hash table.
2903 	 */
2904 	size = pmap_pteg_cnt * sizeof(struct pteg);
2905 	pmap_pteg_table = pmap_boot_find_memory(size, size, 0);
2906 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
2907 	if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH)
2908 		panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB",
2909 		    pmap_pteg_table, size);
2910 #endif
2911 
2912 	memset((void *)pmap_pteg_table, 0, pmap_pteg_cnt * sizeof(struct pteg));
2913 	pmap_pteg_mask = pmap_pteg_cnt - 1;
2914 
2915 	/*
2916 	 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
2917 	 * with pages.  So we just steal them before giving them to UVM.
2918 	 */
2919 	size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
2920 	pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0);
2921 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
2922 	if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH)
2923 		panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB",
2924 		    pmap_pvo_table, size);
2925 #endif
2926 
2927 	for (i = 0; i < pmap_pteg_cnt; i++)
2928 		TAILQ_INIT(&pmap_pvo_table[i]);
2929 
2930 #ifndef MSGBUFADDR
2931 	/*
2932 	 * Allocate msgbuf in high memory.
2933 	 */
2934 	msgbuf_paddr =
2935 	    (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
2936 #endif
2937 
2938 #ifdef __HAVE_PMAP_PHYSSEG
2939 	{
2940 		u_int npgs = 0;
2941 		for (i = 0, mp = avail; i < avail_cnt; i++, mp++)
2942 			npgs += btoc(mp->size);
2943 		size = (sizeof(struct pvo_head) + 1) * npgs;
2944 		pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0);
2945 		pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs];
2946 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
2947 		if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH)
2948 			panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB",
2949 			    pmap_physseg.pvoh, size);
2950 #endif
2951 	}
2952 #endif
2953 
2954 	for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
2955 		paddr_t pfstart = atop(mp->start);
2956 		paddr_t pfend = atop(mp->start + mp->size);
2957 		if (mp->size == 0)
2958 			continue;
2959 		if (mp->start + mp->size <= SEGMENT_LENGTH) {
2960 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
2961 				VM_FREELIST_FIRST256);
2962 		} else if (mp->start >= SEGMENT_LENGTH) {
2963 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
2964 				VM_FREELIST_DEFAULT);
2965 		} else {
2966 			pfend = atop(SEGMENT_LENGTH);
2967 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
2968 				VM_FREELIST_FIRST256);
2969 			pfstart = atop(SEGMENT_LENGTH);
2970 			pfend = atop(mp->start + mp->size);
2971 			uvm_page_physload(pfstart, pfend, pfstart, pfend,
2972 				VM_FREELIST_DEFAULT);
2973 		}
2974 	}
2975 
2976 	/*
2977 	 * Make sure kernel vsid is allocated as well as VSID 0.
2978 	 */
2979 	pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
2980 		|= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
2981 	pmap_vsid_bitmap[0] |= 1;
2982 
2983 	/*
2984 	 * Initialize kernel pmap and hardware.
2985 	 */
2986 	for (i = 0; i < 16; i++) {
2987 		pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT;
2988 		__asm __volatile ("mtsrin %0,%1"
2989 			      :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT));
2990 	}
2991 
2992 	pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
2993 	__asm __volatile ("mtsr %0,%1"
2994 		      :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT));
2995 #ifdef KERNEL2_SR
2996 	pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
2997 	__asm __volatile ("mtsr %0,%1"
2998 		      :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT));
2999 #endif
3000 	for (i = 0; i < 16; i++) {
3001 		if (iosrtable[i] & SR601_T) {
3002 			pmap_kernel()->pm_sr[i] = iosrtable[i];
3003 			__asm __volatile ("mtsrin %0,%1"
3004 			    :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT));
3005 		}
3006 	}
3007 
3008 	__asm __volatile ("sync; mtsdr1 %0; isync"
3009 		      :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10)));
3010 	tlbia();
3011 
3012 #ifdef ALTIVEC
3013 	pmap_use_altivec = cpu_altivec;
3014 #endif
3015 
3016 #ifdef DEBUG
3017 	if (pmapdebug & PMAPDEBUG_BOOT) {
3018 		u_int cnt;
3019 		int bank;
3020 		char pbuf[9];
3021 		for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) {
3022 			cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start;
3023 			printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n",
3024 			    bank,
3025 			    ptoa(vm_physmem[bank].avail_start),
3026 			    ptoa(vm_physmem[bank].avail_end),
3027 			    ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start));
3028 		}
3029 		format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3030 		printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3031 		    pbuf, cnt);
3032 	}
3033 #endif
3034 
3035 	pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry),
3036 	    sizeof(struct pvo_entry), 0, 0, "pmap_upvopl",
3037 	    &pmap_pool_uallocator);
3038 
3039 	pool_setlowat(&pmap_upvo_pool, 252);
3040 
3041 	pool_init(&pmap_pool, sizeof(struct pmap),
3042 	    sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator);
3043 }
3044