xref: /netbsd-src/sys/arch/powerpc/ibm4xx/pmap.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: pmap.c,v 1.3 2001/07/22 13:08:09 wiz Exp $	*/
2 
3 /*
4  * Copyright 2001 Wasabi Systems, Inc.
5  * All rights reserved.
6  *
7  * Written by Eduardo Horvath and Simon Burge for Wasabi Systems, 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 for the NetBSD Project by
20  *      Wasabi Systems, Inc.
21  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
22  *    or promote products derived from this software without specific prior
23  *    written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
26  * 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 WASABI SYSTEMS, INC
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 #undef NOCACHE
70 
71 #include <sys/param.h>
72 #include <sys/malloc.h>
73 #include <sys/proc.h>
74 #include <sys/user.h>
75 #include <sys/queue.h>
76 #include <sys/systm.h>
77 #include <sys/pool.h>
78 #include <sys/device.h>
79 
80 #include <uvm/uvm.h>
81 
82 #include <machine/pcb.h>
83 #include <machine/powerpc.h>
84 
85 #include <powerpc/spr.h>
86 #include <powerpc/ibm4xx/tlb.h>
87 
88 
89 #define	CACHE_LINE	32
90 
91 /*
92  * kernmap is an array of PTEs large enough to map in
93  * 4GB.  At 16KB/page it is 256K entries or 2MB.
94  */
95 #define KERNMAP_SIZE	((0xffffffffU/NBPG)+1)
96 caddr_t kernmap;
97 
98 #define MINCTX		2
99 #define NUMCTX		256
100 volatile struct pmap *ctxbusy[NUMCTX];
101 
102 #define TLBF_USED	0x1
103 #define	TLBF_REF	0x2
104 #define	TLBF_LOCKED	0x4
105 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
106 typedef struct tlb_info_s {
107 	char	ti_flags;
108 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
109 	u_int	ti_va;
110 } tlb_info_t;
111 
112 volatile tlb_info_t tlb_info[NTLB];
113 /* We'll use a modified FIFO replacement policy cause it's cheap */
114 volatile int tlbnext = TLB_NRESERVED;
115 
116 u_long dtlb_miss_count = 0;
117 u_long itlb_miss_count = 0;
118 u_long ktlb_miss_count = 0;
119 u_long utlb_miss_count = 0;
120 
121 /* Event counters -- XXX type `INTR' so we can see them with vmstat -i */
122 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
123 	NULL, "cpu", "tlbmiss");
124 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
125 	NULL, "cpu", "tlbhit");
126 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
127 	NULL, "cpu", "tlbflush");
128 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_INTR,
129 	NULL, "cpu", "tlbenter");
130 
131 struct pmap kernel_pmap_;
132 
133 int physmem;
134 static int npgs;
135 static u_int nextavail;
136 #ifndef MSGBUFADDR
137 extern paddr_t msgbuf_paddr;
138 #endif
139 
140 static struct mem_region *mem, *avail;
141 
142 /*
143  * This is a cache of referenced/modified bits.
144  * Bits herein are shifted by ATTRSHFT.
145  */
146 static char *pmap_attrib;
147 
148 #define PV_WIRED	0x1
149 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
150 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
151 
152 struct pv_entry {
153 	struct pv_entry *pv_next;	/* Linked list of mappings */
154 	vaddr_t pv_va;			/* virtual address of mapping */
155 	struct pmap *pv_pm;
156 };
157 
158 struct pv_entry *pv_table;
159 static struct pool pv_pool;
160 
161 static int pmap_initialized;
162 
163 static int ctx_flush(int);
164 
165 static inline void dcache_flush_page(vaddr_t);
166 static inline void icache_flush_page(vaddr_t);
167 static inline void dcache_flush(vaddr_t, vsize_t);
168 static inline void icache_flush(vaddr_t, vsize_t);
169 
170 inline struct pv_entry *pa_to_pv(paddr_t);
171 static inline char *pa_to_attr(paddr_t);
172 
173 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
174 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
175 
176 static void pmap_pinit(pmap_t);
177 static void pmap_release(pmap_t);
178 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
179 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
180 
181 /*
182  * These small routines may have to be replaced,
183  * if/when we support processors other that the 604.
184  */
185 
186 static inline void
187 dcache_flush_page(vaddr_t va)
188 {
189 	int i;
190 
191 	for (i = 0; i < NBPG; i += CACHE_LINE)
192 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
193 	asm volatile("sync;isync" : : );
194 }
195 
196 static inline void
197 icache_flush_page(vaddr_t va)
198 {
199 	int i;
200 
201 	for (i = 0; i < NBPG; i += CACHE_LINE)
202 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
203 	asm volatile("sync;isync" : : );
204 }
205 
206 static inline void
207 dcache_flush(vaddr_t va, vsize_t len)
208 {
209 	int i;
210 
211 	if (len == 0)
212 		return;
213 
214 	/* Make sure we flush all cache lines */
215 	len += va & (CACHE_LINE-1);
216 	for (i = 0; i < len; i += CACHE_LINE)
217 		asm volatile("dcbf %0,%1" : : "r" (va), "r" (i));
218 	asm volatile("sync;isync" : : );
219 }
220 
221 static inline void
222 icache_flush(vaddr_t va, vsize_t len)
223 {
224 	int i;
225 
226 	if (len == 0)
227 		return;
228 
229 	/* Make sure we flush all cache lines */
230 	len += va & (CACHE_LINE-1);
231 	for (i = 0; i < len; i += CACHE_LINE)
232 		asm volatile("icbi %0,%1" : : "r" (va), "r" (i));
233 	asm volatile("sync;isync" : : );
234 }
235 
236 inline struct pv_entry *
237 pa_to_pv(paddr_t pa)
238 {
239 	int bank, pg;
240 
241 	bank = vm_physseg_find(atop(pa), &pg);
242 	if (bank == -1)
243 		return NULL;
244 	return &vm_physmem[bank].pmseg.pvent[pg];
245 }
246 
247 static inline char *
248 pa_to_attr(paddr_t pa)
249 {
250 	int bank, pg;
251 
252 	bank = vm_physseg_find(atop(pa), &pg);
253 	if (bank == -1)
254 		return NULL;
255 	return &vm_physmem[bank].pmseg.attrs[pg];
256 }
257 
258 /*
259  * Insert PTE into page table.
260  */
261 int
262 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
263 {
264 	int seg = STIDX(va);
265 	int ptn = PTIDX(va);
266 	paddr_t pa;
267 
268 	if (!pm->pm_ptbl[seg]) {
269 		/* Don't allocate a page to clear a non-existent mapping. */
270 		if (!pte) return (1);
271 		/* Allocate a page XXXX this will sleep! */
272 		pa = 0;
273 		pm->pm_ptbl[seg] = (uint *)uvm_km_alloc1(kernel_map, NBPG, 1);
274 	}
275 	pm->pm_ptbl[seg][ptn] = pte;
276 
277 	/* Flush entry. */
278 	ppc4xx_tlb_flush(va, pm->pm_ctx);
279 	return (1);
280 }
281 
282 /*
283  * Get a pointer to a PTE in a page table.
284  */
285 volatile u_int *
286 pte_find(struct pmap *pm, vaddr_t va)
287 {
288 	int seg = STIDX(va);
289 	int ptn = PTIDX(va);
290 
291 	if (pm->pm_ptbl[seg])
292 		return (&pm->pm_ptbl[seg][ptn]);
293 
294 	return (NULL);
295 }
296 
297 /*
298  * This is called during initppc, before the system is really initialized.
299  */
300 void
301 pmap_bootstrap(u_int kernelstart, u_int kernelend)
302 {
303 	struct mem_region *mp, *mp1;
304 	int cnt, i;
305 	u_int s, e, sz;
306 
307 	/*
308 	 * Allocate the kernel page table at the end of
309 	 * kernel space so it's in the locked TTE.
310 	 */
311 	kernmap = (caddr_t)kernelend;
312 	kernelend += KERNMAP_SIZE*sizeof(struct pte);
313 
314 	/*
315 	 * Initialize kernel page table.
316 	 */
317 	memset(kernmap, 0, KERNMAP_SIZE*sizeof(struct pte));
318 	for (i = 0; i < STSZ; i++) {
319 		pmap_kernel()->pm_ptbl[i] = (u_int *)(kernmap + i*NBPG);
320 	}
321 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
322 
323 	/*
324 	 * Announce page-size to the VM-system
325 	 */
326 	uvmexp.pagesize = NBPG;
327 	uvm_setpagesize();
328 
329 	/*
330 	 * Get memory.
331 	 */
332 	mem_regions(&mem, &avail);
333 	for (mp = mem; mp->size; mp++) {
334 		physmem += btoc(mp->size);
335 		printf("+%lx,",mp->size);
336 	}
337 	printf("\n");
338 	ppc4xx_tlb_init();
339 	/*
340 	 * Count the number of available entries.
341 	 */
342 	for (cnt = 0, mp = avail; mp->size; mp++)
343 		cnt++;
344 
345 	/*
346 	 * Page align all regions.
347 	 * Non-page aligned memory isn't very interesting to us.
348 	 * Also, sort the entries for ascending addresses.
349 	 */
350 	kernelstart &= ~PGOFSET;
351 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
352 	for (mp = avail; mp->size; mp++) {
353 		s = mp->start;
354 		e = mp->start + mp->size;
355 		printf("%08x-%08x -> ",s,e);
356 		/*
357 		 * Check whether this region holds all of the kernel.
358 		 */
359 		if (s < kernelstart && e > kernelend) {
360 			avail[cnt].start = kernelend;
361 			avail[cnt++].size = e - kernelend;
362 			e = kernelstart;
363 		}
364 		/*
365 		 * Look whether this regions starts within the kernel.
366 		 */
367 		if (s >= kernelstart && s < kernelend) {
368 			if (e <= kernelend)
369 				goto empty;
370 			s = kernelend;
371 		}
372 		/*
373 		 * Now look whether this region ends within the kernel.
374 		 */
375 		if (e > kernelstart && e <= kernelend) {
376 			if (s >= kernelstart)
377 				goto empty;
378 			e = kernelstart;
379 		}
380 		/*
381 		 * Now page align the start and size of the region.
382 		 */
383 		s = round_page(s);
384 		e = trunc_page(e);
385 		if (e < s)
386 			e = s;
387 		sz = e - s;
388 		printf("%08x-%08x = %x\n",s,e,sz);
389 		/*
390 		 * Check whether some memory is left here.
391 		 */
392 		if (sz == 0) {
393 		empty:
394 			memmove(mp, mp + 1,
395 				(cnt - (mp - avail)) * sizeof *mp);
396 			cnt--;
397 			mp--;
398 			continue;
399 		}
400 		/*
401 		 * Do an insertion sort.
402 		 */
403 		npgs += btoc(sz);
404 		for (mp1 = avail; mp1 < mp; mp1++)
405 			if (s < mp1->start)
406 				break;
407 		if (mp1 < mp) {
408 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
409 			mp1->start = s;
410 			mp1->size = sz;
411 		} else {
412 			mp->start = s;
413 			mp->size = sz;
414 		}
415 	}
416 
417 	/*
418 	 * We cannot do pmap_steal_memory here,
419 	 * since we don't run with translation enabled yet.
420 	 */
421 #ifndef MSGBUFADDR
422 	/*
423 	 * allow for msgbuf
424 	 */
425 	sz = round_page(MSGBUFSIZE);
426 	mp = NULL;
427 	for (mp1 = avail; mp1->size; mp1++)
428 		if (mp1->size >= sz)
429 			mp = mp1;
430 	if (mp == NULL)
431 		panic("not enough memory?");
432 
433 	npgs -= btoc(sz);
434 	msgbuf_paddr = mp->start + mp->size - sz;
435 	mp->size -= sz;
436 	if (mp->size <= 0)
437 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
438 #endif
439 
440 	printf("Loading pages\n");
441 	for (mp = avail; mp->size; mp++)
442 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
443 			atop(mp->start), atop(mp->start + mp->size),
444 			VM_FREELIST_DEFAULT);
445 
446 	/*
447 	 * Initialize kernel pmap and hardware.
448 	 */
449 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
450 	pmap_kernel()->pm_ctx = KERNEL_PID;
451 	nextavail = avail->start;
452 
453 
454 	evcnt_attach_static(&tlbhit_ev);
455 	evcnt_attach_static(&tlbmiss_ev);
456 	evcnt_attach_static(&tlbflush_ev);
457 	evcnt_attach_static(&tlbenter_ev);
458 	printf("Done\n");
459 }
460 
461 /*
462  * Restrict given range to physical memory
463  *
464  * (Used by /dev/mem)
465  */
466 void
467 pmap_real_memory(paddr_t *start, psize_t *size)
468 {
469 	struct mem_region *mp;
470 
471 	for (mp = mem; mp->size; mp++) {
472 		if (*start + *size > mp->start &&
473 		    *start < mp->start + mp->size) {
474 			if (*start < mp->start) {
475 				*size -= mp->start - *start;
476 				*start = mp->start;
477 			}
478 			if (*start + *size > mp->start + mp->size)
479 				*size = mp->start + mp->size - *start;
480 			return;
481 		}
482 	}
483 	*size = 0;
484 }
485 
486 /*
487  * Initialize anything else for pmap handling.
488  * Called during vm_init().
489  */
490 void
491 pmap_init(void)
492 {
493 	struct pv_entry *pv;
494 	vsize_t sz;
495 	vaddr_t addr;
496 	int i, s;
497 	int bank;
498 	char *attr;
499 
500 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
501 	sz = round_page(sz);
502 	addr = uvm_km_zalloc(kernel_map, sz);
503 	s = splvm();
504 	pv = pv_table = (struct pv_entry *)addr;
505 	for (i = npgs; --i >= 0;)
506 		pv++->pv_pm = NULL;
507 	pmap_attrib = (char *)pv;
508 	memset(pv, 0, npgs);
509 
510 	pv = pv_table;
511 	attr = pmap_attrib;
512 	for (bank = 0; bank < vm_nphysseg; bank++) {
513 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
514 		vm_physmem[bank].pmseg.pvent = pv;
515 		vm_physmem[bank].pmseg.attrs = attr;
516 		pv += sz;
517 		attr += sz;
518 	}
519 
520 	pmap_initialized = 1;
521 	splx(s);
522 
523 	/* Setup a pool for additional pvlist structures */
524 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", 0,
525 		  NULL, NULL, 0);
526 }
527 
528 /*
529  * How much virtual space is available to the kernel?
530  */
531 void
532 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
533 {
534 
535 #if 0
536 	/*
537 	 * Reserve one segment for kernel virtual memory
538 	 */
539 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
540 	*end = *start + SEGMENT_LENGTH;
541 #else
542 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
543 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
544 #endif
545 }
546 
547 /*
548  * Create and return a physical map.
549  */
550 struct pmap *
551 pmap_create(void)
552 {
553 	struct pmap *pm;
554 
555 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
556 	memset((caddr_t)pm, 0, sizeof *pm);
557 	pmap_pinit(pm);
558 	return pm;
559 }
560 
561 /*
562  * Initialize a preallocated and zeroed pmap structure.
563  */
564 void
565 pmap_pinit(struct pmap *pm)
566 {
567 	int i;
568 
569 	/*
570 	 * Allocate some segment registers for this pmap.
571 	 */
572 	pm->pm_refs = 1;
573 	for (i = 0; i < STSZ; i++)
574 		pm->pm_ptbl[i] = NULL;
575 }
576 
577 /*
578  * Add a reference to the given pmap.
579  */
580 void
581 pmap_reference(struct pmap *pm)
582 {
583 
584 	pm->pm_refs++;
585 }
586 
587 /*
588  * Retire the given pmap from service.
589  * Should only be called if the map contains no valid mappings.
590  */
591 void
592 pmap_destroy(struct pmap *pm)
593 {
594 
595 	if (--pm->pm_refs == 0) {
596 		pmap_release(pm);
597 		free((caddr_t)pm, M_VMPMAP);
598 	}
599 }
600 
601 /*
602  * Release any resources held by the given physical map.
603  * Called when a pmap initialized by pmap_pinit is being released.
604  */
605 static void
606 pmap_release(struct pmap *pm)
607 {
608 	int i;
609 
610 	for (i = 0; i < STSZ; i++)
611 		if (pm->pm_ptbl[i]) {
612 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i], NBPG);
613 			pm->pm_ptbl[i] = NULL;
614 		}
615 	if (pm->pm_ctx) ctx_free(pm);
616 }
617 
618 /*
619  * Copy the range specified by src_addr/len
620  * from the source map to the range dst_addr/len
621  * in the destination map.
622  *
623  * This routine is only advisory and need not do anything.
624  */
625 void
626 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
627 	  vsize_t len, vaddr_t src_addr)
628 {
629 }
630 
631 /*
632  * Require that all active physical maps contain no
633  * incorrect entries NOW.
634  */
635 void
636 pmap_update(void)
637 {
638 }
639 
640 /*
641  * Garbage collects the physical map system for
642  * pages which are no longer used.
643  * Success need not be guaranteed -- that is, there
644  * may well be pages which are not referenced, but
645  * others may be collected.
646  * Called by the pageout daemon when pages are scarce.
647  */
648 void
649 pmap_collect(struct pmap *pm)
650 {
651 }
652 
653 /*
654  * Fill the given physical page with zeroes.
655  */
656 void
657 pmap_zero_page(paddr_t pa)
658 {
659 
660 #ifdef NOCACHE
661 	memset((caddr_t)pa, 0, NBPG);
662 #else
663 	int i;
664 
665 	for (i = NBPG/CACHELINESIZE; i > 0; i--) {
666 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
667 		pa += CACHELINESIZE;
668 	}
669 #endif
670 }
671 
672 /*
673  * Copy the given physical source page to its destination.
674  */
675 void
676 pmap_copy_page(paddr_t src, paddr_t dst)
677 {
678 
679 	memcpy((caddr_t)dst, (caddr_t)src, NBPG);
680 	dcache_flush_page(dst);
681 }
682 
683 /*
684  * This returns whether this is the first mapping of a page.
685  */
686 static inline int
687 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
688 {
689 	struct pv_entry *pv, *npv = NULL;
690 	int s;
691 
692 	if (!pmap_initialized)
693 		return 0;
694 
695 	s = splvm();
696 
697 	pv = pa_to_pv(pa);
698 	if (!pv->pv_pm) {
699 		/*
700 		 * No entries yet, use header as the first entry.
701 		 */
702 		pv->pv_va = va;
703 		pv->pv_pm = pm;
704 		pv->pv_next = NULL;
705 	} else {
706 		/*
707 		 * There is at least one other VA mapping this page.
708 		 * Place this entry after the header.
709 		 */
710 		npv = pool_get(&pv_pool, PR_WAITOK);
711 		if (!npv) return (0);
712 		npv->pv_va = va;
713 		npv->pv_pm = pm;
714 		npv->pv_next = pv->pv_next;
715 		pv->pv_next = npv;
716 	}
717 	splx(s);
718 	return (1);
719 }
720 
721 static void
722 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
723 {
724 	struct pv_entry *pv, *npv;
725 
726 	/*
727 	 * Remove from the PV table.
728 	 */
729 	pv = pa_to_pv(pa);
730 	if (!pv) return;
731 
732 	/*
733 	 * If it is the first entry on the list, it is actually
734 	 * in the header and we must copy the following entry up
735 	 * to the header.  Otherwise we must search the list for
736 	 * the entry.  In either case we free the now unused entry.
737 	 */
738 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
739 		if ((npv = pv->pv_next)) {
740 			*pv = *npv;
741 			pool_put(&pv_pool, npv);
742 		} else
743 			pv->pv_pm = NULL;
744 	} else {
745 		for (; (npv = pv->pv_next) != NULL; pv = npv)
746 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
747 				break;
748 		if (npv) {
749 			pv->pv_next = npv->pv_next;
750 			pool_put(&pv_pool, npv);
751 		}
752 	}
753 }
754 
755 /*
756  * Insert physical page at pa into the given pmap at virtual address va.
757  */
758 int
759 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
760 {
761 	int s;
762 	u_int tte;
763 	int managed;
764 
765 	/*
766 	 * Have to remove any existing mapping first.
767 	 */
768 	pmap_remove(pm, va, va + NBPG);
769 
770 	if (flags & PMAP_WIRED) flags |= prot;
771 
772 	/* If it has no protections don't bother w/the rest */
773 	if (!(flags & VM_PROT_ALL))
774 		return (0);
775 
776 	managed = 0;
777 	if (vm_physseg_find(atop(pa), NULL) != -1)
778 		managed = 1;
779 
780 	/*
781 	 * Generate TTE.
782 	 *
783 	 * XXXX
784 	 *
785 	 * Since the kernel does not handle execution privileges properly,
786 	 * we will handle read and execute permissions together.
787 	 */
788 	tte = TTE_PA(pa) | TTE_EX;
789 	/* XXXX -- need to support multiple page sizes. */
790 	tte |= TTE_SZ_16K;
791 #ifdef	DIAGNOSTIC
792 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
793 		(PME_NOCACHE | PME_WRITETHROUG))
794 		panic("pmap_enter: uncached & writethrough\n");
795 #endif
796 	if (flags & PME_NOCACHE)
797 		/* Must be I/O mapping */
798 		tte |= TTE_I | TTE_G;
799 #ifdef NOCACHE
800 	tte |= TTE_I;
801 #else
802 	else if (flags & PME_WRITETHROUG)
803 		/* Uncached and writethrough are not compatible */
804 		tte |= TTE_W;
805 #endif
806 	if (pm == pmap_kernel())
807 		tte |= TTE_ZONE(ZONE_PRIV);
808 	else
809 		tte |= TTE_ZONE(ZONE_USER);
810 
811 	if (flags & VM_PROT_WRITE)
812 		tte |= TTE_WR;
813 
814 	/*
815 	 * Now record mapping for later back-translation.
816 	 */
817 	if (pmap_initialized && managed) {
818 		char *attr;
819 
820 		if (!pmap_enter_pv(pm, va, pa)) {
821 			/* Could not enter pv on a managed page */
822 			return 1;
823 		}
824 
825 		/* Now set attributes. */
826 		attr = pa_to_attr(pa);
827 #ifdef DIAGNOSTIC
828 		if (!attr)
829 			panic("managed but no attr\n");
830 #endif
831 		if (flags & VM_PROT_ALL)
832 			*attr |= PTE_HI_REF;
833 		if (flags & VM_PROT_WRITE)
834 			*attr |= PTE_HI_CHG;
835 	}
836 
837 	s = splvm();
838 	pm->pm_stats.resident_count++;
839 
840 	/* Insert page into page table. */
841 	pte_enter(pm, va, tte);
842 
843 	/* If this is a real fault, enter it in the tlb */
844 	if (tte && ((flags & PMAP_WIRED) == 0)) {
845 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
846 	}
847 	splx(s);
848 	return 0;
849 }
850 
851 void
852 pmap_unwire(struct pmap *pm, vaddr_t va)
853 {
854 	struct pv_entry *pv, *npv;
855 	paddr_t pa;
856 	int s = splvm();
857 
858         if (pm == NULL) {
859                 return;
860         }
861 
862 	if (!pmap_extract(pm, va, &pa)) {
863 		return;
864 	}
865 
866 	va |= PV_WIRED;
867 
868 	pv = pa_to_pv(pa);
869 	if (!pv) return;
870 
871 	/*
872 	 * If it is the first entry on the list, it is actually
873 	 * in the header and we must copy the following entry up
874 	 * to the header.  Otherwise we must search the list for
875 	 * the entry.  In either case we free the now unused entry.
876 	 */
877 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
878 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
879 			npv->pv_va &= ~PV_WIRED;
880 			break;
881 		}
882 	}
883 	splx(s);
884 }
885 
886 void
887 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
888 {
889 	int s;
890 	u_int tte;
891 	struct pmap *pm = pmap_kernel();
892 
893 	/*
894 	 * Have to remove any existing mapping first.
895 	 */
896 
897 	/*
898 	 * Generate TTE.
899 	 *
900 	 * XXXX
901 	 *
902 	 * Since the kernel does not handle execution privileges properly,
903 	 * we will handle read and execute permissions together.
904 	 */
905 	tte = 0;
906 	if (prot & VM_PROT_ALL) {
907 
908 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
909 		/* XXXX -- need to support multiple page sizes. */
910 		tte |= TTE_SZ_16K;
911 #ifdef DIAGNOSTIC
912 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
913 			(PME_NOCACHE | PME_WRITETHROUG))
914 			panic("pmap_kenter_pa: uncached & writethrough\n");
915 #endif
916 		if (prot & PME_NOCACHE)
917 			/* Must be I/O mapping */
918 			tte |= TTE_I | TTE_G;
919 #ifdef NOCACHE
920 		tte |= TTE_I;
921 #else
922 		else if (prot & PME_WRITETHROUG)
923 			/* Uncached and writethrough are not compatible */
924 			tte |= TTE_W;
925 #endif
926 		if (prot & VM_PROT_WRITE)
927 			tte |= TTE_WR;
928 	}
929 
930 	s = splvm();
931 	pm->pm_stats.resident_count++;
932 
933 	/* Insert page into page table. */
934 	pte_enter(pm, va, tte);
935 	splx(s);
936 }
937 
938 void
939 pmap_kremove(vaddr_t va, vsize_t len)
940 {
941 
942 	while (len > 0) {
943 		pte_enter(pmap_kernel(), va, 0);
944 		va += PAGE_SIZE;
945 		len -= PAGE_SIZE;
946 	}
947 }
948 
949 /*
950  * Remove the given range of mapping entries.
951  */
952 void
953 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
954 {
955 	int s;
956 	paddr_t pa;
957 	volatile u_int *ptp;
958 
959 	s = splvm();
960 	while (va < endva) {
961 
962 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
963 			pa = TTE_PA(pa);
964 			pmap_remove_pv(pm, va, pa);
965 			*ptp = 0;
966 			ppc4xx_tlb_flush(va, pm->pm_ctx);
967 			pm->pm_stats.resident_count--;
968 		}
969 		va += NBPG;
970 	}
971 
972 	splx(s);
973 }
974 
975 /*
976  * Get the physical page address for the given pmap/virtual address.
977  */
978 boolean_t
979 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
980 {
981 	int seg = STIDX(va);
982 	int ptn = PTIDX(va);
983 	u_int pa = 0;
984 	int s = splvm();
985 
986 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
987 		*pap = TTE_PA(pa) | (va & PGOFSET);
988 	}
989 	splx(s);
990 	return (pa != 0);
991 }
992 
993 /*
994  * Lower the protection on the specified range of this pmap.
995  *
996  * There are only two cases: either the protection is going to 0,
997  * or it is going to read-only.
998  */
999 void
1000 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1001 {
1002 	volatile u_int *ptp;
1003 	int s;
1004 
1005 	if (prot & VM_PROT_READ) {
1006 		s = splvm();
1007 		while (sva < eva) {
1008 			if ((ptp = pte_find(pm, sva)) != NULL) {
1009 				*ptp &= ~TTE_WR;
1010 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
1011 			}
1012 			sva += NBPG;
1013 		}
1014 		splx(s);
1015 		return;
1016 	}
1017 	pmap_remove(pm, sva, eva);
1018 }
1019 
1020 boolean_t
1021 check_attr(struct vm_page *pg, u_int mask, int clear)
1022 {
1023 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
1024 	int s;
1025 	char *attr;
1026 	int rv;
1027 
1028 	/*
1029 	 * First modify bits in cache.
1030 	 */
1031 	s = splvm();
1032 	attr = pa_to_attr(pa);
1033 	if (attr == NULL)
1034 		return FALSE;
1035 
1036 	rv = ((*attr & mask) != 0);
1037 	if (clear)
1038 		*attr &= ~mask;
1039 
1040 	splx(s);
1041 	return rv;
1042 }
1043 
1044 
1045 /*
1046  * Lower the protection on the specified physical page.
1047  *
1048  * There are only two cases: either the protection is going to 0,
1049  * or it is going to read-only.
1050  */
1051 void
1052 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1053 {
1054 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
1055 	vaddr_t va;
1056 	struct pv_entry *pvh, *pv, *npv;
1057 	struct pmap *pm;
1058 
1059 	pvh = pa_to_pv(pa);
1060 	if (pvh == NULL)
1061 		return;
1062 
1063 	/* Handle extra pvs which may be deleted in the operation */
1064 	for (pv = pvh->pv_next; pv; pv = npv) {
1065 		npv = pv->pv_next;
1066 
1067 		pm = pv->pv_pm;
1068 		va = pv->pv_va;
1069 		pmap_protect(pm, va, va+NBPG, prot);
1070 	}
1071 	/* Now check the head pv */
1072 	if (pvh->pv_pm) {
1073 		pv = pvh;
1074 		pm = pv->pv_pm;
1075 		va = pv->pv_va;
1076 		pmap_protect(pm, va, va+NBPG, prot);
1077 	}
1078 }
1079 
1080 /*
1081  * Activate the address space for the specified process.  If the process
1082  * is the current process, load the new MMU context.
1083  */
1084 void
1085 pmap_activate(struct proc *p)
1086 {
1087 #if 0
1088 	struct pcb *pcb = &p->p_addr->u_pcb;
1089 	pmap_t pmap = p->p_vmspace->vm_map.pmap;
1090 
1091 	/*
1092 	 * XXX Normally performed in cpu_fork().
1093 	 */
1094 	printf("pmap_activate(%p), pmap=%p\n",p,pmap);
1095 	if (pcb->pcb_pm != pmap) {
1096 		pcb->pcb_pm = pmap;
1097 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
1098 		    (paddr_t *)&pcb->pcb_pmreal);
1099 	}
1100 
1101 	if (p == curproc) {
1102 		/* Store pointer to new current pmap. */
1103 		curpm = pcb->pcb_pmreal;
1104 	}
1105 #endif
1106 }
1107 
1108 /*
1109  * Deactivate the specified process's address space.
1110  */
1111 void
1112 pmap_deactivate(struct proc *p)
1113 {
1114 }
1115 
1116 /*
1117  * Synchronize caches corresponding to [addr, addr+len) in p.
1118  */
1119 void
1120 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
1121 {
1122 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
1123 	int msr, ctx, opid;
1124 
1125 
1126 	/*
1127 	 * Need to turn off IMMU and switch to user context.
1128 	 * (icbi uses DMMU).
1129 	 */
1130 	if (!(ctx = pm->pm_ctx)) {
1131 		/* No context -- assign it one */
1132 		ctx_alloc(pm);
1133 		ctx = pm->pm_ctx;
1134 	}
1135 	__asm __volatile("mfmsr %0;"
1136 		"li %1, 0x20;"
1137 		"andc %1,%0,%1;"
1138 		"mtmsr %1;"
1139 		"sync;isync;"
1140 		"mfpid %1;"
1141 		"mtpid %2;"
1142 		"sync; isync;"
1143 		"1:"
1144 		"dcbf 0,%3;"
1145 		"icbi 0,%3;"
1146 		"addi %3,%3,32;"
1147 		"addic. %4,%4,-32;"
1148 		"bge 1b;"
1149 		"mtpid %1;"
1150 		"mtmsr %0;"
1151 		"sync; isync"
1152 		: "=&r" (msr), "=&r" (opid)
1153 		: "r" (ctx), "r" (va), "r" (len));
1154 }
1155 
1156 
1157 /* This has to be done in real mode !!! */
1158 void
1159 ppc4xx_tlb_flush(vaddr_t va, int pid)
1160 {
1161 	u_long i, found;
1162 	u_long msr;
1163 
1164 	/* If there's no context then it can't be mapped. */
1165 	if (!pid) return;
1166 
1167 	asm("mfpid %1;"			/* Save PID */
1168 		"mfmsr %2;"		/* Save MSR */
1169 		"li %0,0;"		/* Now clear MSR */
1170 		"mtmsr %0;"
1171 		"mtpid %4;"		/* Set PID */
1172 		"sync;"
1173 		"tlbsx. %0,0,%3;"	/* Search TLB */
1174 		"sync;"
1175 		"mtpid %1;"		/* Restore PID */
1176 		"mtmsr %2;"		/* Restore MSR */
1177 		"sync;isync;"
1178 		"li %1,1;"
1179 		"beq 1f;"
1180 		"li %1,0;"
1181 		"1:"
1182 		: "=&r" (i), "=&r" (found), "=&r" (msr)
1183 		: "r" (va), "r" (pid));
1184 	if (found && !TLB_LOCKED(i)) {
1185 
1186 		/* Now flush translation */
1187 		asm volatile(
1188 			"tlbwe %0,%1,0;"
1189 			"sync;isync;"
1190 			: : "r" (0), "r" (i));
1191 
1192 		tlb_info[i].ti_ctx = 0;
1193 		tlb_info[i].ti_flags = 0;
1194 		tlbnext = i;
1195 		/* Successful flushes */
1196 		tlbflush_ev.ev_count++;
1197 	}
1198 }
1199 
1200 void
1201 ppc4xx_tlb_flush_all(void)
1202 {
1203 	u_long i;
1204 
1205 	for (i = 0; i < NTLB; i++)
1206 		if (!TLB_LOCKED(i)) {
1207 			asm volatile(
1208 				"tlbwe %0,%1,0;"
1209 				"sync;isync;"
1210 				: : "r" (0), "r" (i));
1211 			tlb_info[i].ti_ctx = 0;
1212 			tlb_info[i].ti_flags = 0;
1213 		}
1214 
1215 	asm volatile("sync;isync");
1216 }
1217 
1218 /* Find a TLB entry to evict. */
1219 static int
1220 ppc4xx_tlb_find_victim(void)
1221 {
1222 	int flags;
1223 
1224 	for (;;) {
1225 		if (++tlbnext >= NTLB)
1226 			tlbnext = TLB_NRESERVED;
1227 		flags = tlb_info[tlbnext].ti_flags;
1228 		if (!(flags & TLBF_USED) ||
1229 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
1230 			u_long va, stack = (u_long)&va;
1231 
1232 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
1233 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
1234 			     (flags & TLBF_USED)) {
1235 				/* Kernel stack page */
1236 				flags |= TLBF_USED;
1237 				tlb_info[tlbnext].ti_flags = flags;
1238 			} else {
1239 				/* Found it! */
1240 				return (tlbnext);
1241 			}
1242 		} else {
1243 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
1244 		}
1245 	}
1246 }
1247 
1248 void
1249 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
1250 {
1251 	u_long th, tl, idx;
1252 	tlbpid_t pid;
1253 	u_short msr;
1254 	int s;
1255 
1256 	tlbenter_ev.ev_count++;
1257 
1258 	th = (va & TLB_EPN_MASK) |
1259 		(((pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT) << TLB_SIZE_SHFT) |
1260 		TLB_VALID;
1261 	tl = pte;
1262 
1263 	s = splhigh();
1264 	idx = ppc4xx_tlb_find_victim();
1265 
1266 #ifdef DIAGNOSTIC
1267 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
1268 		panic("ppc4xx_tlb_enter: repacing entry %ld\n", idx);
1269 	}
1270 #endif
1271 
1272 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
1273 	tlb_info[idx].ti_ctx = ctx;
1274 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
1275 
1276 	asm volatile(
1277 		"mfmsr %0;"			/* Save MSR */
1278 		"li %1,0;"
1279 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
1280 		"mtmsr %1;"			/* Clear MSR */
1281 		"mfpid %1;"			/* Save old PID */
1282 		"mtpid %2;"			/* Load translation ctx */
1283 		"sync; isync;"
1284 #ifdef DEBUG
1285 		"andi. %3,%3,63;"
1286 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
1287 #endif
1288 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
1289 		"sync; isync;"
1290 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
1291 		"sync; isync;"
1292 	: "=&r" (msr), "=&r" (pid)
1293 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
1294 	splx(s);
1295 }
1296 
1297 void
1298 ppc4xx_tlb_unpin(int i)
1299 {
1300 
1301 	if (i == -1)
1302 		for (i = 0; i < TLB_NRESERVED; i++)
1303 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
1304 	else
1305 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
1306 }
1307 
1308 void
1309 ppc4xx_tlb_init(void)
1310 {
1311 	int i;
1312 
1313 	/* Mark reserved TLB entries */
1314 	for (i = 0; i < TLB_NRESERVED; i++) {
1315 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
1316 		tlb_info[i].ti_ctx = KERNEL_PID;
1317 	}
1318 
1319 	/* Setup security zones */
1320 	/* Z0 - accessible by kernel only if TLB entry permissions allow
1321 	 * Z1,Z2 - access is controlled by TLB entry permissions
1322 	 * Z3 - full access regardless of TLB entry permissions
1323 	 */
1324 
1325 	asm volatile(
1326 		"mtspr %0,%1;"
1327 		"sync;"
1328 		::  "K"(SPR_ZPR), "r" (0x1b000000));
1329 }
1330 
1331 
1332 /*
1333  * We should pass the ctx in from trap code.
1334  */
1335 int
1336 pmap_tlbmiss(vaddr_t va, int ctx)
1337 {
1338 	volatile u_int *pte;
1339 	u_long tte;
1340 
1341 	tlbmiss_ev.ev_count++;
1342 
1343 	/*
1344 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
1345 	 */
1346 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
1347 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
1348 		if (pte == NULL) {
1349 			/* Map unmanaged addresses directly for kernel access */
1350 			return 1;
1351 		}
1352 		tte = *pte;
1353 		if (tte == 0) {
1354 			return 1;
1355 		}
1356 	} else {
1357 		/* Create a 16MB writeable mapping. */
1358 #ifdef NOCACHE
1359 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
1360 #else
1361 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
1362 #endif
1363 	}
1364 	tlbhit_ev.ev_count++;
1365 	ppc4xx_tlb_enter(ctx, va, tte);
1366 
1367 	return 0;
1368 }
1369 
1370 /*
1371  * Flush all the entries matching a context from the TLB.
1372  */
1373 static int
1374 ctx_flush(int cnum)
1375 {
1376 	int i;
1377 
1378 	/* We gotta steal this context */
1379 	for (i = TLB_NRESERVED; i < NTLB; i++) {
1380 		if (tlb_info[i].ti_ctx == cnum) {
1381 			/* Can't steal ctx if it has a locked entry. */
1382 			if (TLB_LOCKED(i)) {
1383 #ifdef DIAGNOSTIC
1384 				printf("ctx_flush: can't invalidate "
1385 					"locked mapping %d "
1386 					"for context %d\n", i, cnum);
1387 				Debugger();
1388 #endif
1389 				return (1);
1390 			}
1391 #ifdef DIAGNOSTIC
1392 			if (i < TLB_NRESERVED)
1393 				panic("TLB entry %d not locked\n", i);
1394 #endif
1395 			/* Invalidate particular TLB entry regardless of locked status */
1396 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
1397 			tlb_info[i].ti_flags = 0;
1398 		}
1399 	}
1400 	return (0);
1401 }
1402 
1403 /*
1404  * Allocate a context.  If necessary, steal one from someone else.
1405  *
1406  * The new context is flushed from the TLB before returning.
1407  */
1408 int
1409 ctx_alloc(struct pmap *pm)
1410 {
1411 	int s, cnum;
1412 	static int next = MINCTX;
1413 
1414 	if (pm == pmap_kernel()) {
1415 #ifdef DIAGNOSTIC
1416 		printf("ctx_alloc: kernel pmap!\n");
1417 #endif
1418 		return (0);
1419 	}
1420 	s = splvm();
1421 
1422 	/* Find a likely context. */
1423 	cnum = next;
1424 	do {
1425 		if ((++cnum) > NUMCTX)
1426 			cnum = MINCTX;
1427 	} while (ctxbusy[cnum] != NULL && cnum != next);
1428 
1429 	/* Now clean it out */
1430 oops:
1431 	if (cnum < MINCTX)
1432 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
1433 	if (ctx_flush(cnum)) {
1434 		/* oops -- something's wired. */
1435 		if ((++cnum) > NUMCTX)
1436 			cnum = MINCTX;
1437 		goto oops;
1438 	}
1439 
1440 	if (ctxbusy[cnum]) {
1441 #ifdef DEBUG
1442 		/* We should identify this pmap and clear it */
1443 		printf("Warning: stealing context %d\n", cnum);
1444 #endif
1445 		ctxbusy[cnum]->pm_ctx = 0;
1446 	}
1447 	ctxbusy[cnum] = pm;
1448 	next = cnum;
1449 	splx(s);
1450 	pm->pm_ctx = cnum;
1451 
1452 	return cnum;
1453 }
1454 
1455 /*
1456  * Give away a context.
1457  */
1458 void
1459 ctx_free(struct pmap *pm)
1460 {
1461 	int oldctx;
1462 
1463 	oldctx = pm->pm_ctx;
1464 
1465 	if (oldctx == 0)
1466 		panic("ctx_free: freeing kernel context");
1467 #ifdef DIAGNOSTIC
1468 	if (ctxbusy[oldctx] == 0)
1469 		printf("ctx_free: freeing free context %d\n", oldctx);
1470 	if (ctxbusy[oldctx] != pm) {
1471 		printf("ctx_free: freeing someone esle's context\n "
1472 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
1473 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
1474 		Debugger();
1475 	}
1476 #endif
1477 	/* We should verify it has not been stolen and reallocated... */
1478 	ctxbusy[oldctx] = NULL;
1479 	ctx_flush(oldctx);
1480 }
1481 
1482 #ifdef DEBUG
1483 /*
1484  * Test ref/modify handling.
1485  */
1486 void pmap_testout __P((void));
1487 void
1488 pmap_testout()
1489 {
1490 	vaddr_t va;
1491 	volatile int *loc;
1492 	int val = 0;
1493 	paddr_t pa;
1494 	struct vm_page *pg;
1495 	int ref, mod;
1496 
1497 	/* Allocate a page */
1498 	va = (vaddr_t)uvm_km_alloc1(kernel_map, NBPG, 1);
1499 	loc = (int*)va;
1500 
1501 	pmap_extract(pmap_kernel(), va, &pa);
1502 	pg = PHYS_TO_VM_PAGE(pa);
1503 	pmap_unwire(pmap_kernel(), va);
1504 
1505 	pmap_remove(pmap_kernel(), va, va+1);
1506 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1507 	pmap_update();
1508 
1509 	/* Now clear reference and modify */
1510 	ref = pmap_clear_reference(pg);
1511 	mod = pmap_clear_modify(pg);
1512 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1513 	       (void *)(u_long)va, (long)pa,
1514 	       ref, mod);
1515 
1516 	/* Check it's properly cleared */
1517 	ref = pmap_is_referenced(pg);
1518 	mod = pmap_is_modified(pg);
1519 	printf("Checking cleared page: ref %d, mod %d\n",
1520 	       ref, mod);
1521 
1522 	/* Reference page */
1523 	val = *loc;
1524 
1525 	ref = pmap_is_referenced(pg);
1526 	mod = pmap_is_modified(pg);
1527 	printf("Referenced page: ref %d, mod %d val %x\n",
1528 	       ref, mod, val);
1529 
1530 	/* Now clear reference and modify */
1531 	ref = pmap_clear_reference(pg);
1532 	mod = pmap_clear_modify(pg);
1533 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1534 	       (void *)(u_long)va, (long)pa,
1535 	       ref, mod);
1536 
1537 	/* Modify page */
1538 	*loc = 1;
1539 
1540 	ref = pmap_is_referenced(pg);
1541 	mod = pmap_is_modified(pg);
1542 	printf("Modified page: ref %d, mod %d\n",
1543 	       ref, mod);
1544 
1545 	/* Now clear reference and modify */
1546 	ref = pmap_clear_reference(pg);
1547 	mod = pmap_clear_modify(pg);
1548 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1549 	       (void *)(u_long)va, (long)pa,
1550 	       ref, mod);
1551 
1552 	/* Check it's properly cleared */
1553 	ref = pmap_is_referenced(pg);
1554 	mod = pmap_is_modified(pg);
1555 	printf("Checking cleared page: ref %d, mod %d\n",
1556 	       ref, mod);
1557 
1558 	/* Modify page */
1559 	*loc = 1;
1560 
1561 	ref = pmap_is_referenced(pg);
1562 	mod = pmap_is_modified(pg);
1563 	printf("Modified page: ref %d, mod %d\n",
1564 	       ref, mod);
1565 
1566 	/* Check pmap_protect() */
1567 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
1568 	pmap_update();
1569 	ref = pmap_is_referenced(pg);
1570 	mod = pmap_is_modified(pg);
1571 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
1572 	       ref, mod);
1573 
1574 	/* Now clear reference and modify */
1575 	ref = pmap_clear_reference(pg);
1576 	mod = pmap_clear_modify(pg);
1577 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1578 	       (void *)(u_long)va, (long)pa,
1579 	       ref, mod);
1580 
1581 	/* Reference page */
1582 	val = *loc;
1583 
1584 	ref = pmap_is_referenced(pg);
1585 	mod = pmap_is_modified(pg);
1586 	printf("Referenced page: ref %d, mod %d val %x\n",
1587 	       ref, mod, val);
1588 
1589 	/* Now clear reference and modify */
1590 	ref = pmap_clear_reference(pg);
1591 	mod = pmap_clear_modify(pg);
1592 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1593 	       (void *)(u_long)va, (long)pa,
1594 	       ref, mod);
1595 
1596 	/* Modify page */
1597 #if 0
1598 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1599 	pmap_update();
1600 #endif
1601 	*loc = 1;
1602 
1603 	ref = pmap_is_referenced(pg);
1604 	mod = pmap_is_modified(pg);
1605 	printf("Modified page: ref %d, mod %d\n",
1606 	       ref, mod);
1607 
1608 	/* Check pmap_protect() */
1609 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
1610 	pmap_update();
1611 	ref = pmap_is_referenced(pg);
1612 	mod = pmap_is_modified(pg);
1613 	printf("pmap_protect(): ref %d, mod %d\n",
1614 	       ref, mod);
1615 
1616 	/* Now clear reference and modify */
1617 	ref = pmap_clear_reference(pg);
1618 	mod = pmap_clear_modify(pg);
1619 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1620 	       (void *)(u_long)va, (long)pa,
1621 	       ref, mod);
1622 
1623 	/* Reference page */
1624 	val = *loc;
1625 
1626 	ref = pmap_is_referenced(pg);
1627 	mod = pmap_is_modified(pg);
1628 	printf("Referenced page: ref %d, mod %d val %x\n",
1629 	       ref, mod, val);
1630 
1631 	/* Now clear reference and modify */
1632 	ref = pmap_clear_reference(pg);
1633 	mod = pmap_clear_modify(pg);
1634 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1635 	       (void *)(u_long)va, (long)pa,
1636 	       ref, mod);
1637 
1638 	/* Modify page */
1639 #if 0
1640 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1641 	pmap_update();
1642 #endif
1643 	*loc = 1;
1644 
1645 	ref = pmap_is_referenced(pg);
1646 	mod = pmap_is_modified(pg);
1647 	printf("Modified page: ref %d, mod %d\n",
1648 	       ref, mod);
1649 
1650 	/* Check pmap_pag_protect() */
1651 	pmap_page_protect(pg, VM_PROT_READ);
1652 	ref = pmap_is_referenced(pg);
1653 	mod = pmap_is_modified(pg);
1654 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
1655 	       ref, mod);
1656 
1657 	/* Now clear reference and modify */
1658 	ref = pmap_clear_reference(pg);
1659 	mod = pmap_clear_modify(pg);
1660 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1661 	       (void *)(u_long)va, (long)pa,
1662 	       ref, mod);
1663 
1664 	/* Reference page */
1665 	val = *loc;
1666 
1667 	ref = pmap_is_referenced(pg);
1668 	mod = pmap_is_modified(pg);
1669 	printf("Referenced page: ref %d, mod %d val %x\n",
1670 	       ref, mod, val);
1671 
1672 	/* Now clear reference and modify */
1673 	ref = pmap_clear_reference(pg);
1674 	mod = pmap_clear_modify(pg);
1675 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1676 	       (void *)(u_long)va, (long)pa,
1677 	       ref, mod);
1678 
1679 	/* Modify page */
1680 #if 0
1681 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1682 	pmap_update();
1683 #endif
1684 	*loc = 1;
1685 
1686 	ref = pmap_is_referenced(pg);
1687 	mod = pmap_is_modified(pg);
1688 	printf("Modified page: ref %d, mod %d\n",
1689 	       ref, mod);
1690 
1691 	/* Check pmap_pag_protect() */
1692 	pmap_page_protect(pg, VM_PROT_NONE);
1693 	ref = pmap_is_referenced(pg);
1694 	mod = pmap_is_modified(pg);
1695 	printf("pmap_page_protect(): ref %d, mod %d\n",
1696 	       ref, mod);
1697 
1698 	/* Now clear reference and modify */
1699 	ref = pmap_clear_reference(pg);
1700 	mod = pmap_clear_modify(pg);
1701 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1702 	       (void *)(u_long)va, (long)pa,
1703 	       ref, mod);
1704 
1705 
1706 	/* Reference page */
1707 	val = *loc;
1708 
1709 	ref = pmap_is_referenced(pg);
1710 	mod = pmap_is_modified(pg);
1711 	printf("Referenced page: ref %d, mod %d val %x\n",
1712 	       ref, mod, val);
1713 
1714 	/* Now clear reference and modify */
1715 	ref = pmap_clear_reference(pg);
1716 	mod = pmap_clear_modify(pg);
1717 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1718 	       (void *)(u_long)va, (long)pa,
1719 	       ref, mod);
1720 
1721 	/* Modify page */
1722 #if 0
1723 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1724 	pmap_update();
1725 #endif
1726 	*loc = 1;
1727 
1728 	ref = pmap_is_referenced(pg);
1729 	mod = pmap_is_modified(pg);
1730 	printf("Modified page: ref %d, mod %d\n",
1731 	       ref, mod);
1732 
1733 	/* Unmap page */
1734 	pmap_remove(pmap_kernel(), va, va+1);
1735 	pmap_update();
1736 	ref = pmap_is_referenced(pg);
1737 	mod = pmap_is_modified(pg);
1738 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
1739 
1740 	/* Now clear reference and modify */
1741 	ref = pmap_clear_reference(pg);
1742 	mod = pmap_clear_modify(pg);
1743 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1744 	       (void *)(u_long)va, (long)pa, ref, mod);
1745 
1746 	/* Check it's properly cleared */
1747 	ref = pmap_is_referenced(pg);
1748 	mod = pmap_is_modified(pg);
1749 	printf("Checking cleared page: ref %d, mod %d\n",
1750 	       ref, mod);
1751 
1752 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
1753 		VM_PROT_ALL|PMAP_WIRED);
1754 	uvm_km_free(kernel_map, (vaddr_t)va, NBPG);
1755 }
1756 #endif
1757