xref: /netbsd-src/sys/arch/powerpc/ibm4xx/pmap.c (revision b8c616269f5ebf18ab2e35cb8099d683130a177c)
1 /*	$NetBSD: pmap.c,v 1.17 2003/01/18 06:23:30 thorpej 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 PPC_4XX_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/cpu.h>
83 #include <machine/pcb.h>
84 #include <machine/powerpc.h>
85 
86 #include <powerpc/spr.h>
87 #include <machine/tlb.h>
88 
89 /*
90  * kernmap is an array of PTEs large enough to map in
91  * 4GB.  At 16KB/page it is 256K entries or 2MB.
92  */
93 #define KERNMAP_SIZE	((0xffffffffU/NBPG)+1)
94 caddr_t kernmap;
95 
96 #define MINCTX		2
97 #define NUMCTX		256
98 volatile struct pmap *ctxbusy[NUMCTX];
99 
100 #define TLBF_USED	0x1
101 #define	TLBF_REF	0x2
102 #define	TLBF_LOCKED	0x4
103 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
104 typedef struct tlb_info_s {
105 	char	ti_flags;
106 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
107 	u_int	ti_va;
108 } tlb_info_t;
109 
110 volatile tlb_info_t tlb_info[NTLB];
111 /* We'll use a modified FIFO replacement policy cause it's cheap */
112 volatile int tlbnext = TLB_NRESERVED;
113 
114 u_long dtlb_miss_count = 0;
115 u_long itlb_miss_count = 0;
116 u_long ktlb_miss_count = 0;
117 u_long utlb_miss_count = 0;
118 
119 /* Event counters */
120 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
121 	NULL, "cpu", "tlbmiss");
122 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
123 	NULL, "cpu", "tlbhit");
124 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
125 	NULL, "cpu", "tlbflush");
126 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
127 	NULL, "cpu", "tlbenter");
128 
129 struct pmap kernel_pmap_;
130 
131 int physmem;
132 static int npgs;
133 static u_int nextavail;
134 #ifndef MSGBUFADDR
135 extern paddr_t msgbuf_paddr;
136 #endif
137 
138 static struct mem_region *mem, *avail;
139 
140 /*
141  * This is a cache of referenced/modified bits.
142  * Bits herein are shifted by ATTRSHFT.
143  */
144 static char *pmap_attrib;
145 
146 #define PV_WIRED	0x1
147 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
148 #define	PV_CMPVA(va,pv)	(!(((pv)->pv_va^(va))&(~PV_WIRED)))
149 
150 struct pv_entry {
151 	struct pv_entry *pv_next;	/* Linked list of mappings */
152 	vaddr_t pv_va;			/* virtual address of mapping */
153 	struct pmap *pv_pm;
154 };
155 
156 struct pv_entry *pv_table;
157 static struct pool pv_pool;
158 
159 static int pmap_initialized;
160 
161 static int ctx_flush(int);
162 
163 inline struct pv_entry *pa_to_pv(paddr_t);
164 static inline char *pa_to_attr(paddr_t);
165 
166 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
167 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
168 
169 static void pmap_pinit(pmap_t);
170 static void pmap_release(pmap_t);
171 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t);
172 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
173 
174 
175 inline struct pv_entry *
176 pa_to_pv(paddr_t pa)
177 {
178 	int bank, pg;
179 
180 	bank = vm_physseg_find(atop(pa), &pg);
181 	if (bank == -1)
182 		return NULL;
183 	return &vm_physmem[bank].pmseg.pvent[pg];
184 }
185 
186 static inline char *
187 pa_to_attr(paddr_t pa)
188 {
189 	int bank, pg;
190 
191 	bank = vm_physseg_find(atop(pa), &pg);
192 	if (bank == -1)
193 		return NULL;
194 	return &vm_physmem[bank].pmseg.attrs[pg];
195 }
196 
197 /*
198  * Insert PTE into page table.
199  */
200 int
201 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
202 {
203 	int seg = STIDX(va);
204 	int ptn = PTIDX(va);
205 	paddr_t pa;
206 
207 	if (!pm->pm_ptbl[seg]) {
208 		/* Don't allocate a page to clear a non-existent mapping. */
209 		if (!pte) return (1);
210 		/* Allocate a page XXXX this will sleep! */
211 		pa = 0;
212 		pm->pm_ptbl[seg] = (uint *)uvm_km_alloc1(kernel_map, NBPG, 1);
213 	}
214 	pm->pm_ptbl[seg][ptn] = pte;
215 
216 	/* Flush entry. */
217 	ppc4xx_tlb_flush(va, pm->pm_ctx);
218 	return (1);
219 }
220 
221 /*
222  * Get a pointer to a PTE in a page table.
223  */
224 volatile u_int *
225 pte_find(struct pmap *pm, vaddr_t va)
226 {
227 	int seg = STIDX(va);
228 	int ptn = PTIDX(va);
229 
230 	if (pm->pm_ptbl[seg])
231 		return (&pm->pm_ptbl[seg][ptn]);
232 
233 	return (NULL);
234 }
235 
236 /*
237  * This is called during initppc, before the system is really initialized.
238  */
239 void
240 pmap_bootstrap(u_int kernelstart, u_int kernelend)
241 {
242 	struct mem_region *mp, *mp1;
243 	int cnt, i;
244 	u_int s, e, sz;
245 
246 	/*
247 	 * Allocate the kernel page table at the end of
248 	 * kernel space so it's in the locked TTE.
249 	 */
250 	kernmap = (caddr_t)kernelend;
251 
252 	/*
253 	 * Initialize kernel page table.
254 	 */
255 	for (i = 0; i < STSZ; i++) {
256 		pmap_kernel()->pm_ptbl[i] = 0;
257 	}
258 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
259 
260 	/*
261 	 * Announce page-size to the VM-system
262 	 */
263 	uvmexp.pagesize = NBPG;
264 	uvm_setpagesize();
265 
266 	/*
267 	 * Get memory.
268 	 */
269 	mem_regions(&mem, &avail);
270 	for (mp = mem; mp->size; mp++) {
271 		physmem += btoc(mp->size);
272 		printf("+%lx,",mp->size);
273 	}
274 	printf("\n");
275 	ppc4xx_tlb_init();
276 	/*
277 	 * Count the number of available entries.
278 	 */
279 	for (cnt = 0, mp = avail; mp->size; mp++)
280 		cnt++;
281 
282 	/*
283 	 * Page align all regions.
284 	 * Non-page aligned memory isn't very interesting to us.
285 	 * Also, sort the entries for ascending addresses.
286 	 */
287 	kernelstart &= ~PGOFSET;
288 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
289 	for (mp = avail; mp->size; mp++) {
290 		s = mp->start;
291 		e = mp->start + mp->size;
292 		printf("%08x-%08x -> ",s,e);
293 		/*
294 		 * Check whether this region holds all of the kernel.
295 		 */
296 		if (s < kernelstart && e > kernelend) {
297 			avail[cnt].start = kernelend;
298 			avail[cnt++].size = e - kernelend;
299 			e = kernelstart;
300 		}
301 		/*
302 		 * Look whether this regions starts within the kernel.
303 		 */
304 		if (s >= kernelstart && s < kernelend) {
305 			if (e <= kernelend)
306 				goto empty;
307 			s = kernelend;
308 		}
309 		/*
310 		 * Now look whether this region ends within the kernel.
311 		 */
312 		if (e > kernelstart && e <= kernelend) {
313 			if (s >= kernelstart)
314 				goto empty;
315 			e = kernelstart;
316 		}
317 		/*
318 		 * Now page align the start and size of the region.
319 		 */
320 		s = round_page(s);
321 		e = trunc_page(e);
322 		if (e < s)
323 			e = s;
324 		sz = e - s;
325 		printf("%08x-%08x = %x\n",s,e,sz);
326 		/*
327 		 * Check whether some memory is left here.
328 		 */
329 		if (sz == 0) {
330 		empty:
331 			memmove(mp, mp + 1,
332 				(cnt - (mp - avail)) * sizeof *mp);
333 			cnt--;
334 			mp--;
335 			continue;
336 		}
337 		/*
338 		 * Do an insertion sort.
339 		 */
340 		npgs += btoc(sz);
341 		for (mp1 = avail; mp1 < mp; mp1++)
342 			if (s < mp1->start)
343 				break;
344 		if (mp1 < mp) {
345 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
346 			mp1->start = s;
347 			mp1->size = sz;
348 		} else {
349 			mp->start = s;
350 			mp->size = sz;
351 		}
352 	}
353 
354 	/*
355 	 * We cannot do pmap_steal_memory here,
356 	 * since we don't run with translation enabled yet.
357 	 */
358 #ifndef MSGBUFADDR
359 	/*
360 	 * allow for msgbuf
361 	 */
362 	sz = round_page(MSGBUFSIZE);
363 	mp = NULL;
364 	for (mp1 = avail; mp1->size; mp1++)
365 		if (mp1->size >= sz)
366 			mp = mp1;
367 	if (mp == NULL)
368 		panic("not enough memory?");
369 
370 	npgs -= btoc(sz);
371 	msgbuf_paddr = mp->start + mp->size - sz;
372 	mp->size -= sz;
373 	if (mp->size <= 0)
374 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
375 #endif
376 
377 	printf("Loading pages\n");
378 	for (mp = avail; mp->size; mp++)
379 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
380 			atop(mp->start), atop(mp->start + mp->size),
381 			VM_FREELIST_DEFAULT);
382 
383 	/*
384 	 * Initialize kernel pmap and hardware.
385 	 */
386 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
387 	pmap_kernel()->pm_ctx = KERNEL_PID;
388 	nextavail = avail->start;
389 
390 
391 	evcnt_attach_static(&tlbhit_ev);
392 	evcnt_attach_static(&tlbmiss_ev);
393 	evcnt_attach_static(&tlbflush_ev);
394 	evcnt_attach_static(&tlbenter_ev);
395 	printf("Done\n");
396 }
397 
398 /*
399  * Restrict given range to physical memory
400  *
401  * (Used by /dev/mem)
402  */
403 void
404 pmap_real_memory(paddr_t *start, psize_t *size)
405 {
406 	struct mem_region *mp;
407 
408 	for (mp = mem; mp->size; mp++) {
409 		if (*start + *size > mp->start &&
410 		    *start < mp->start + mp->size) {
411 			if (*start < mp->start) {
412 				*size -= mp->start - *start;
413 				*start = mp->start;
414 			}
415 			if (*start + *size > mp->start + mp->size)
416 				*size = mp->start + mp->size - *start;
417 			return;
418 		}
419 	}
420 	*size = 0;
421 }
422 
423 /*
424  * Initialize anything else for pmap handling.
425  * Called during vm_init().
426  */
427 void
428 pmap_init(void)
429 {
430 	struct pv_entry *pv;
431 	vsize_t sz;
432 	vaddr_t addr;
433 	int i, s;
434 	int bank;
435 	char *attr;
436 
437 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
438 	sz = round_page(sz);
439 	addr = uvm_km_zalloc(kernel_map, sz);
440 	s = splvm();
441 	pv = pv_table = (struct pv_entry *)addr;
442 	for (i = npgs; --i >= 0;)
443 		pv++->pv_pm = NULL;
444 	pmap_attrib = (char *)pv;
445 	memset(pv, 0, npgs);
446 
447 	pv = pv_table;
448 	attr = pmap_attrib;
449 	for (bank = 0; bank < vm_nphysseg; bank++) {
450 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
451 		vm_physmem[bank].pmseg.pvent = pv;
452 		vm_physmem[bank].pmseg.attrs = attr;
453 		pv += sz;
454 		attr += sz;
455 	}
456 
457 	pmap_initialized = 1;
458 	splx(s);
459 
460 	/* Setup a pool for additional pvlist structures */
461 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL);
462 }
463 
464 /*
465  * How much virtual space is available to the kernel?
466  */
467 void
468 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
469 {
470 
471 #if 0
472 	/*
473 	 * Reserve one segment for kernel virtual memory
474 	 */
475 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
476 	*end = *start + SEGMENT_LENGTH;
477 #else
478 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
479 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
480 #endif
481 }
482 
483 #ifdef PMAP_GROWKERNEL
484 /*
485  * Preallocate kernel page tables to a specified VA.
486  * This simply loops through the first TTE for each
487  * page table from the beginning of the kernel pmap,
488  * reads the entry, and if the result is
489  * zero (either invalid entry or no page table) it stores
490  * a zero there, populating page tables in the process.
491  * This is not the most efficient technique but i don't
492  * expect it to be called that often.
493  */
494 extern struct vm_page *vm_page_alloc1 __P((void));
495 extern void vm_page_free1 __P((struct vm_page *));
496 
497 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
498 
499 vaddr_t
500 pmap_growkernel(maxkvaddr)
501 	vaddr_t maxkvaddr;
502 {
503 	int s;
504 	int seg;
505 	paddr_t pg;
506 	struct pmap *pm = pmap_kernel();
507 
508 	s = splvm();
509 
510 	/* Align with the start of a page table */
511 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
512 	     kbreak += PTMAP) {
513 		seg = STIDX(kbreak);
514 
515 		if (pte_find(pm, kbreak)) continue;
516 
517 		if (uvm.page_init_done) {
518 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
519 		} else {
520 			if (!uvm_page_physget(&pg))
521 				panic("pmap_growkernel: no memory");
522 		}
523 		if (!pg) panic("pmap_growkernel: no pages");
524 		pmap_zero_page((paddr_t)pg);
525 
526 		/* XXX This is based on all phymem being addressable */
527 		pm->pm_ptbl[seg] = (u_int *)pg;
528 	}
529 	splx(s);
530 	return (kbreak);
531 }
532 
533 /*
534  *	vm_page_alloc1:
535  *
536  *	Allocate and return a memory cell with no associated object.
537  */
538 struct vm_page *
539 vm_page_alloc1()
540 {
541 	struct vm_page *pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
542 	if (pg) {
543 		pg->wire_count = 1;	/* no mappings yet */
544 		pg->flags &= ~PG_BUSY;	/* never busy */
545 	}
546 	return pg;
547 }
548 
549 /*
550  *	vm_page_free1:
551  *
552  *	Returns the given page to the free list,
553  *	disassociating it with any VM object.
554  *
555  *	Object and page must be locked prior to entry.
556  */
557 void
558 vm_page_free1(mem)
559 	struct vm_page *mem;
560 {
561 #ifdef DIAGNOSTIC
562 	if (mem->flags != (PG_CLEAN|PG_FAKE)) {
563 		printf("Freeing invalid page %p\n", mem);
564 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(mem));
565 #ifdef DDB
566 		Debugger();
567 #endif
568 		return;
569 	}
570 #endif
571 	mem->flags |= PG_BUSY;
572 	mem->wire_count = 0;
573 	uvm_pagefree(mem);
574 }
575 #endif
576 
577 /*
578  * Create and return a physical map.
579  */
580 struct pmap *
581 pmap_create(void)
582 {
583 	struct pmap *pm;
584 
585 	pm = (struct pmap *)malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
586 	memset((caddr_t)pm, 0, sizeof *pm);
587 	pmap_pinit(pm);
588 	return pm;
589 }
590 
591 /*
592  * Initialize a preallocated and zeroed pmap structure.
593  */
594 void
595 pmap_pinit(struct pmap *pm)
596 {
597 	int i;
598 
599 	/*
600 	 * Allocate some segment registers for this pmap.
601 	 */
602 	pm->pm_refs = 1;
603 	for (i = 0; i < STSZ; i++)
604 		pm->pm_ptbl[i] = NULL;
605 }
606 
607 /*
608  * Add a reference to the given pmap.
609  */
610 void
611 pmap_reference(struct pmap *pm)
612 {
613 
614 	pm->pm_refs++;
615 }
616 
617 /*
618  * Retire the given pmap from service.
619  * Should only be called if the map contains no valid mappings.
620  */
621 void
622 pmap_destroy(struct pmap *pm)
623 {
624 
625 	if (--pm->pm_refs == 0) {
626 		pmap_release(pm);
627 		free((caddr_t)pm, M_VMPMAP);
628 	}
629 }
630 
631 /*
632  * Release any resources held by the given physical map.
633  * Called when a pmap initialized by pmap_pinit is being released.
634  */
635 static void
636 pmap_release(struct pmap *pm)
637 {
638 	int i;
639 
640 	for (i = 0; i < STSZ; i++)
641 		if (pm->pm_ptbl[i]) {
642 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i], NBPG);
643 			pm->pm_ptbl[i] = NULL;
644 		}
645 	if (pm->pm_ctx) ctx_free(pm);
646 }
647 
648 /*
649  * Copy the range specified by src_addr/len
650  * from the source map to the range dst_addr/len
651  * in the destination map.
652  *
653  * This routine is only advisory and need not do anything.
654  */
655 void
656 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
657 	  vsize_t len, vaddr_t src_addr)
658 {
659 }
660 
661 /*
662  * Require that all active physical maps contain no
663  * incorrect entries NOW.
664  */
665 void
666 pmap_update(struct pmap *pmap)
667 {
668 }
669 
670 /*
671  * Garbage collects the physical map system for
672  * pages which are no longer used.
673  * Success need not be guaranteed -- that is, there
674  * may well be pages which are not referenced, but
675  * others may be collected.
676  * Called by the pageout daemon when pages are scarce.
677  */
678 void
679 pmap_collect(struct pmap *pm)
680 {
681 }
682 
683 /*
684  * Fill the given physical page with zeroes.
685  */
686 void
687 pmap_zero_page(paddr_t pa)
688 {
689 
690 #ifdef PPC_4XX_NOCACHE
691 	memset((caddr_t)pa, 0, NBPG);
692 #else
693 	int i;
694 
695 	for (i = NBPG/CACHELINESIZE; i > 0; i--) {
696 		__asm __volatile ("dcbz 0,%0" :: "r"(pa));
697 		pa += CACHELINESIZE;
698 	}
699 #endif
700 }
701 
702 /*
703  * Copy the given physical source page to its destination.
704  */
705 void
706 pmap_copy_page(paddr_t src, paddr_t dst)
707 {
708 
709 	memcpy((caddr_t)dst, (caddr_t)src, NBPG);
710 	dcache_flush_page(dst);
711 }
712 
713 /*
714  * This returns whether this is the first mapping of a page.
715  */
716 static inline int
717 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
718 {
719 	struct pv_entry *pv, *npv = NULL;
720 	int s;
721 
722 	if (!pmap_initialized)
723 		return 0;
724 
725 	s = splvm();
726 
727 	pv = pa_to_pv(pa);
728 for (npv = pv; npv; npv = npv->pv_next)
729 if (npv->pv_va == va && npv->pv_pm == pm) {
730 printf("Duplicate pv: va %lx pm %p\n", va, pm);
731 #ifdef DDB
732 Debugger();
733 #endif
734 return (1);
735 }
736 
737 	if (!pv->pv_pm) {
738 		/*
739 		 * No entries yet, use header as the first entry.
740 		 */
741 		pv->pv_va = va;
742 		pv->pv_pm = pm;
743 		pv->pv_next = NULL;
744 	} else {
745 		/*
746 		 * There is at least one other VA mapping this page.
747 		 * Place this entry after the header.
748 		 */
749 		npv = pool_get(&pv_pool, PR_WAITOK);
750 		if (!npv) return (0);
751 		npv->pv_va = va;
752 		npv->pv_pm = pm;
753 		npv->pv_next = pv->pv_next;
754 		pv->pv_next = npv;
755 	}
756 	splx(s);
757 	return (1);
758 }
759 
760 static void
761 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
762 {
763 	struct pv_entry *pv, *npv;
764 
765 	/*
766 	 * Remove from the PV table.
767 	 */
768 	pv = pa_to_pv(pa);
769 	if (!pv) return;
770 
771 	/*
772 	 * If it is the first entry on the list, it is actually
773 	 * in the header and we must copy the following entry up
774 	 * to the header.  Otherwise we must search the list for
775 	 * the entry.  In either case we free the now unused entry.
776 	 */
777 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
778 		if ((npv = pv->pv_next)) {
779 			*pv = *npv;
780 			pool_put(&pv_pool, npv);
781 		} else
782 			pv->pv_pm = NULL;
783 	} else {
784 		for (; (npv = pv->pv_next) != NULL; pv = npv)
785 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
786 				break;
787 		if (npv) {
788 			pv->pv_next = npv->pv_next;
789 			pool_put(&pv_pool, npv);
790 		}
791 	}
792 }
793 
794 /*
795  * Insert physical page at pa into the given pmap at virtual address va.
796  */
797 int
798 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
799 {
800 	int s;
801 	u_int tte;
802 	int managed;
803 
804 	/*
805 	 * Have to remove any existing mapping first.
806 	 */
807 	pmap_remove(pm, va, va + NBPG);
808 
809 	if (flags & PMAP_WIRED) flags |= prot;
810 
811 	/* If it has no protections don't bother w/the rest */
812 	if (!(flags & VM_PROT_ALL))
813 		return (0);
814 
815 	managed = 0;
816 	if (vm_physseg_find(atop(pa), NULL) != -1)
817 		managed = 1;
818 
819 	/*
820 	 * Generate TTE.
821 	 *
822 	 * XXXX
823 	 *
824 	 * Since the kernel does not handle execution privileges properly,
825 	 * we will handle read and execute permissions together.
826 	 */
827 	tte = TTE_PA(pa) | TTE_EX;
828 	/* XXXX -- need to support multiple page sizes. */
829 	tte |= TTE_SZ_16K;
830 #ifdef	DIAGNOSTIC
831 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
832 		(PME_NOCACHE | PME_WRITETHROUG))
833 		panic("pmap_enter: uncached & writethrough");
834 #endif
835 	if (flags & PME_NOCACHE)
836 		/* Must be I/O mapping */
837 		tte |= TTE_I | TTE_G;
838 #ifdef PPC_4XX_NOCACHE
839 	tte |= TTE_I;
840 #else
841 	else if (flags & PME_WRITETHROUG)
842 		/* Uncached and writethrough are not compatible */
843 		tte |= TTE_W;
844 #endif
845 	if (pm == pmap_kernel())
846 		tte |= TTE_ZONE(ZONE_PRIV);
847 	else
848 		tte |= TTE_ZONE(ZONE_USER);
849 
850 	if (flags & VM_PROT_WRITE)
851 		tte |= TTE_WR;
852 
853 	/*
854 	 * Now record mapping for later back-translation.
855 	 */
856 	if (pmap_initialized && managed) {
857 		char *attr;
858 
859 		if (!pmap_enter_pv(pm, va, pa)) {
860 			/* Could not enter pv on a managed page */
861 			return 1;
862 		}
863 
864 		/* Now set attributes. */
865 		attr = pa_to_attr(pa);
866 #ifdef DIAGNOSTIC
867 		if (!attr)
868 			panic("managed but no attr");
869 #endif
870 		if (flags & VM_PROT_ALL)
871 			*attr |= PTE_HI_REF;
872 		if (flags & VM_PROT_WRITE)
873 			*attr |= PTE_HI_CHG;
874 	}
875 
876 	s = splvm();
877 	pm->pm_stats.resident_count++;
878 
879 	/* Insert page into page table. */
880 	pte_enter(pm, va, tte);
881 
882 	/* If this is a real fault, enter it in the tlb */
883 	if (tte && ((flags & PMAP_WIRED) == 0)) {
884 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
885 	}
886 	splx(s);
887 
888 	/* Flush the real memory from the instruction cache. */
889 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
890 		__syncicache((void *)pa, PAGE_SIZE);
891 
892 	return 0;
893 }
894 
895 void
896 pmap_unwire(struct pmap *pm, vaddr_t va)
897 {
898 	struct pv_entry *pv, *npv;
899 	paddr_t pa;
900 	int s = splvm();
901 
902 	if (pm == NULL) {
903 		return;
904 	}
905 
906 	if (!pmap_extract(pm, va, &pa)) {
907 		return;
908 	}
909 
910 	va |= PV_WIRED;
911 
912 	pv = pa_to_pv(pa);
913 	if (!pv) return;
914 
915 	/*
916 	 * If it is the first entry on the list, it is actually
917 	 * in the header and we must copy the following entry up
918 	 * to the header.  Otherwise we must search the list for
919 	 * the entry.  In either case we free the now unused entry.
920 	 */
921 	for (npv = pv; (npv = pv->pv_next) != NULL; pv = npv) {
922 		if (pm == npv->pv_pm && PV_CMPVA(va, npv)) {
923 			npv->pv_va &= ~PV_WIRED;
924 			break;
925 		}
926 	}
927 	splx(s);
928 }
929 
930 void
931 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
932 {
933 	int s;
934 	u_int tte;
935 	struct pmap *pm = pmap_kernel();
936 
937 	/*
938 	 * Have to remove any existing mapping first.
939 	 */
940 
941 	/*
942 	 * Generate TTE.
943 	 *
944 	 * XXXX
945 	 *
946 	 * Since the kernel does not handle execution privileges properly,
947 	 * we will handle read and execute permissions together.
948 	 */
949 	tte = 0;
950 	if (prot & VM_PROT_ALL) {
951 
952 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
953 		/* XXXX -- need to support multiple page sizes. */
954 		tte |= TTE_SZ_16K;
955 #ifdef DIAGNOSTIC
956 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
957 			(PME_NOCACHE | PME_WRITETHROUG))
958 			panic("pmap_kenter_pa: uncached & writethrough");
959 #endif
960 		if (prot & PME_NOCACHE)
961 			/* Must be I/O mapping */
962 			tte |= TTE_I | TTE_G;
963 #ifdef PPC_4XX_NOCACHE
964 		tte |= TTE_I;
965 #else
966 		else if (prot & PME_WRITETHROUG)
967 			/* Uncached and writethrough are not compatible */
968 			tte |= TTE_W;
969 #endif
970 		if (prot & VM_PROT_WRITE)
971 			tte |= TTE_WR;
972 	}
973 
974 	s = splvm();
975 	pm->pm_stats.resident_count++;
976 
977 	/* Insert page into page table. */
978 	pte_enter(pm, va, tte);
979 	splx(s);
980 }
981 
982 void
983 pmap_kremove(vaddr_t va, vsize_t len)
984 {
985 
986 	while (len > 0) {
987 		pte_enter(pmap_kernel(), va, 0);
988 		va += PAGE_SIZE;
989 		len -= PAGE_SIZE;
990 	}
991 }
992 
993 /*
994  * Remove the given range of mapping entries.
995  */
996 void
997 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
998 {
999 	int s;
1000 	paddr_t pa;
1001 	volatile u_int *ptp;
1002 
1003 	s = splvm();
1004 	while (va < endva) {
1005 
1006 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
1007 			pa = TTE_PA(pa);
1008 			pmap_remove_pv(pm, va, pa);
1009 			*ptp = 0;
1010 			ppc4xx_tlb_flush(va, pm->pm_ctx);
1011 			pm->pm_stats.resident_count--;
1012 		}
1013 		va += NBPG;
1014 	}
1015 
1016 	splx(s);
1017 }
1018 
1019 /*
1020  * Get the physical page address for the given pmap/virtual address.
1021  */
1022 boolean_t
1023 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
1024 {
1025 	int seg = STIDX(va);
1026 	int ptn = PTIDX(va);
1027 	u_int pa = 0;
1028 	int s = splvm();
1029 
1030 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
1031 		*pap = TTE_PA(pa) | (va & PGOFSET);
1032 	}
1033 	splx(s);
1034 	return (pa != 0);
1035 }
1036 
1037 /*
1038  * Lower the protection on the specified range of this pmap.
1039  *
1040  * There are only two cases: either the protection is going to 0,
1041  * or it is going to read-only.
1042  */
1043 void
1044 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1045 {
1046 	volatile u_int *ptp;
1047 	int s;
1048 
1049 	if (prot & VM_PROT_READ) {
1050 		s = splvm();
1051 		while (sva < eva) {
1052 			if ((ptp = pte_find(pm, sva)) != NULL) {
1053 				*ptp &= ~TTE_WR;
1054 				ppc4xx_tlb_flush(sva, pm->pm_ctx);
1055 			}
1056 			sva += NBPG;
1057 		}
1058 		splx(s);
1059 		return;
1060 	}
1061 	pmap_remove(pm, sva, eva);
1062 }
1063 
1064 boolean_t
1065 check_attr(struct vm_page *pg, u_int mask, int clear)
1066 {
1067 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
1068 	int s;
1069 	char *attr;
1070 	int rv;
1071 
1072 	/*
1073 	 * First modify bits in cache.
1074 	 */
1075 	s = splvm();
1076 	attr = pa_to_attr(pa);
1077 	if (attr == NULL)
1078 		return FALSE;
1079 
1080 	rv = ((*attr & mask) != 0);
1081 	if (clear) {
1082 		*attr &= ~mask;
1083 		pmap_page_protect(pg, (mask == PTE_HI_CHG) ? VM_PROT_READ : 0);
1084 	}
1085 	splx(s);
1086 	return rv;
1087 }
1088 
1089 
1090 /*
1091  * Lower the protection on the specified physical page.
1092  *
1093  * There are only two cases: either the protection is going to 0,
1094  * or it is going to read-only.
1095  */
1096 void
1097 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1098 {
1099 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
1100 	vaddr_t va;
1101 	struct pv_entry *pvh, *pv, *npv;
1102 	struct pmap *pm;
1103 
1104 	pvh = pa_to_pv(pa);
1105 	if (pvh == NULL)
1106 		return;
1107 
1108 	/* Handle extra pvs which may be deleted in the operation */
1109 	for (pv = pvh->pv_next; pv; pv = npv) {
1110 		npv = pv->pv_next;
1111 
1112 		pm = pv->pv_pm;
1113 		va = pv->pv_va;
1114 		pmap_protect(pm, va, va+NBPG, prot);
1115 	}
1116 	/* Now check the head pv */
1117 	if (pvh->pv_pm) {
1118 		pv = pvh;
1119 		pm = pv->pv_pm;
1120 		va = pv->pv_va;
1121 		pmap_protect(pm, va, va+NBPG, prot);
1122 	}
1123 }
1124 
1125 /*
1126  * Activate the address space for the specified process.  If the process
1127  * is the current process, load the new MMU context.
1128  */
1129 void
1130 pmap_activate(struct lwp *l)
1131 {
1132 #if 0
1133 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
1134 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
1135 
1136 	/*
1137 	 * XXX Normally performed in cpu_fork().
1138 	 */
1139 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
1140 	if (pcb->pcb_pm != pmap) {
1141 		pcb->pcb_pm = pmap;
1142 		(void) pmap_extract(pmap_kernel(), (vaddr_t)pcb->pcb_pm,
1143 		    (paddr_t *)&pcb->pcb_pmreal);
1144 	}
1145 
1146 	if (l == curlwp) {
1147 		/* Store pointer to new current pmap. */
1148 		curpm = pcb->pcb_pmreal;
1149 	}
1150 #endif
1151 }
1152 
1153 /*
1154  * Deactivate the specified process's address space.
1155  */
1156 void
1157 pmap_deactivate(struct lwp *l)
1158 {
1159 }
1160 
1161 /*
1162  * Synchronize caches corresponding to [addr, addr+len) in p.
1163  */
1164 void
1165 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
1166 {
1167 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
1168 	int msr, ctx, opid;
1169 
1170 
1171 	/*
1172 	 * Need to turn off IMMU and switch to user context.
1173 	 * (icbi uses DMMU).
1174 	 */
1175 	if (!(ctx = pm->pm_ctx)) {
1176 		/* No context -- assign it one */
1177 		ctx_alloc(pm);
1178 		ctx = pm->pm_ctx;
1179 	}
1180 	__asm __volatile("mfmsr %0;"
1181 		"li %1, 0x20;"
1182 		"andc %1,%0,%1;"
1183 		"mtmsr %1;"
1184 		"sync;isync;"
1185 		"mfpid %1;"
1186 		"mtpid %2;"
1187 		"sync; isync;"
1188 		"1:"
1189 		"dcbf 0,%3;"
1190 		"icbi 0,%3;"
1191 		"addi %3,%3,32;"
1192 		"addic. %4,%4,-32;"
1193 		"bge 1b;"
1194 		"mtpid %1;"
1195 		"mtmsr %0;"
1196 		"sync; isync"
1197 		: "=&r" (msr), "=&r" (opid)
1198 		: "r" (ctx), "r" (va), "r" (len));
1199 }
1200 
1201 
1202 /* This has to be done in real mode !!! */
1203 void
1204 ppc4xx_tlb_flush(vaddr_t va, int pid)
1205 {
1206 	u_long i, found;
1207 	u_long msr;
1208 
1209 	/* If there's no context then it can't be mapped. */
1210 	if (!pid) return;
1211 
1212 	asm("mfpid %1;"			/* Save PID */
1213 		"mfmsr %2;"		/* Save MSR */
1214 		"li %0,0;"		/* Now clear MSR */
1215 		"mtmsr %0;"
1216 		"mtpid %4;"		/* Set PID */
1217 		"sync;"
1218 		"tlbsx. %0,0,%3;"	/* Search TLB */
1219 		"sync;"
1220 		"mtpid %1;"		/* Restore PID */
1221 		"mtmsr %2;"		/* Restore MSR */
1222 		"sync;isync;"
1223 		"li %1,1;"
1224 		"beq 1f;"
1225 		"li %1,0;"
1226 		"1:"
1227 		: "=&r" (i), "=&r" (found), "=&r" (msr)
1228 		: "r" (va), "r" (pid));
1229 	if (found && !TLB_LOCKED(i)) {
1230 
1231 		/* Now flush translation */
1232 		asm volatile(
1233 			"tlbwe %0,%1,0;"
1234 			"sync;isync;"
1235 			: : "r" (0), "r" (i));
1236 
1237 		tlb_info[i].ti_ctx = 0;
1238 		tlb_info[i].ti_flags = 0;
1239 		tlbnext = i;
1240 		/* Successful flushes */
1241 		tlbflush_ev.ev_count++;
1242 	}
1243 }
1244 
1245 void
1246 ppc4xx_tlb_flush_all(void)
1247 {
1248 	u_long i;
1249 
1250 	for (i = 0; i < NTLB; i++)
1251 		if (!TLB_LOCKED(i)) {
1252 			asm volatile(
1253 				"tlbwe %0,%1,0;"
1254 				"sync;isync;"
1255 				: : "r" (0), "r" (i));
1256 			tlb_info[i].ti_ctx = 0;
1257 			tlb_info[i].ti_flags = 0;
1258 		}
1259 
1260 	asm volatile("sync;isync");
1261 }
1262 
1263 /* Find a TLB entry to evict. */
1264 static int
1265 ppc4xx_tlb_find_victim(void)
1266 {
1267 	int flags;
1268 
1269 	for (;;) {
1270 		if (++tlbnext >= NTLB)
1271 			tlbnext = TLB_NRESERVED;
1272 		flags = tlb_info[tlbnext].ti_flags;
1273 		if (!(flags & TLBF_USED) ||
1274 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
1275 			u_long va, stack = (u_long)&va;
1276 
1277 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
1278 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
1279 			     (flags & TLBF_USED)) {
1280 				/* Kernel stack page */
1281 				flags |= TLBF_USED;
1282 				tlb_info[tlbnext].ti_flags = flags;
1283 			} else {
1284 				/* Found it! */
1285 				return (tlbnext);
1286 			}
1287 		} else {
1288 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
1289 		}
1290 	}
1291 }
1292 
1293 void
1294 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
1295 {
1296 	u_long th, tl, idx;
1297 	tlbpid_t pid;
1298 	u_short msr;
1299 	paddr_t pa;
1300 	int s, sz;
1301 
1302 	tlbenter_ev.ev_count++;
1303 
1304 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
1305 	pa = (pte & TTE_RPN_MASK(sz));
1306 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
1307 	tl = (pte & ~TLB_RPN_MASK) | pa;
1308 	tl |= ppc4xx_tlbflags(va, pa);
1309 
1310 	s = splhigh();
1311 	idx = ppc4xx_tlb_find_victim();
1312 
1313 #ifdef DIAGNOSTIC
1314 	if ((idx < TLB_NRESERVED) || (idx >= NTLB)) {
1315 		panic("ppc4xx_tlb_enter: repacing entry %ld", idx);
1316 	}
1317 #endif
1318 
1319 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
1320 	tlb_info[idx].ti_ctx = ctx;
1321 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
1322 
1323 	asm volatile(
1324 		"mfmsr %0;"			/* Save MSR */
1325 		"li %1,0;"
1326 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
1327 		"mtmsr %1;"			/* Clear MSR */
1328 		"mfpid %1;"			/* Save old PID */
1329 		"mtpid %2;"			/* Load translation ctx */
1330 		"sync; isync;"
1331 #ifdef DEBUG
1332 		"andi. %3,%3,63;"
1333 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
1334 #endif
1335 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
1336 		"sync; isync;"
1337 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
1338 		"sync; isync;"
1339 	: "=&r" (msr), "=&r" (pid)
1340 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
1341 	splx(s);
1342 }
1343 
1344 void
1345 ppc4xx_tlb_unpin(int i)
1346 {
1347 
1348 	if (i == -1)
1349 		for (i = 0; i < TLB_NRESERVED; i++)
1350 			tlb_info[i].ti_flags &= ~TLBF_LOCKED;
1351 	else
1352 		tlb_info[i].ti_flags &= ~TLBF_LOCKED;
1353 }
1354 
1355 void
1356 ppc4xx_tlb_init(void)
1357 {
1358 	int i;
1359 
1360 	/* Mark reserved TLB entries */
1361 	for (i = 0; i < TLB_NRESERVED; i++) {
1362 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
1363 		tlb_info[i].ti_ctx = KERNEL_PID;
1364 	}
1365 
1366 	/* Setup security zones */
1367 	/* Z0 - accessible by kernel only if TLB entry permissions allow
1368 	 * Z1,Z2 - access is controlled by TLB entry permissions
1369 	 * Z3 - full access regardless of TLB entry permissions
1370 	 */
1371 
1372 	asm volatile(
1373 		"mtspr %0,%1;"
1374 		"sync;"
1375 		::  "K"(SPR_ZPR), "r" (0x1b000000));
1376 }
1377 
1378 
1379 /*
1380  * We should pass the ctx in from trap code.
1381  */
1382 int
1383 pmap_tlbmiss(vaddr_t va, int ctx)
1384 {
1385 	volatile u_int *pte;
1386 	u_long tte;
1387 
1388 	tlbmiss_ev.ev_count++;
1389 
1390 	/*
1391 	 * XXXX We will reserve 0-0x80000000 for va==pa mappings.
1392 	 */
1393 	if (ctx != KERNEL_PID || (va & 0x80000000)) {
1394 		pte = pte_find((struct pmap *)ctxbusy[ctx], va);
1395 		if (pte == NULL) {
1396 			/* Map unmanaged addresses directly for kernel access */
1397 			return 1;
1398 		}
1399 		tte = *pte;
1400 		if (tte == 0) {
1401 			return 1;
1402 		}
1403 	} else {
1404 		/* Create a 16MB writable mapping. */
1405 #ifdef PPC_4XX_NOCACHE
1406 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I | TTE_WR;
1407 #else
1408 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
1409 #endif
1410 	}
1411 	tlbhit_ev.ev_count++;
1412 	ppc4xx_tlb_enter(ctx, va, tte);
1413 
1414 	return 0;
1415 }
1416 
1417 /*
1418  * Flush all the entries matching a context from the TLB.
1419  */
1420 static int
1421 ctx_flush(int cnum)
1422 {
1423 	int i;
1424 
1425 	/* We gotta steal this context */
1426 	for (i = TLB_NRESERVED; i < NTLB; i++) {
1427 		if (tlb_info[i].ti_ctx == cnum) {
1428 			/* Can't steal ctx if it has a locked entry. */
1429 			if (TLB_LOCKED(i)) {
1430 #ifdef DIAGNOSTIC
1431 				printf("ctx_flush: can't invalidate "
1432 					"locked mapping %d "
1433 					"for context %d\n", i, cnum);
1434 #ifdef DDB
1435 				Debugger();
1436 #endif
1437 #endif
1438 				return (1);
1439 			}
1440 #ifdef DIAGNOSTIC
1441 			if (i < TLB_NRESERVED)
1442 				panic("TLB entry %d not locked", i);
1443 #endif
1444 			/* Invalidate particular TLB entry regardless of locked status */
1445 			asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
1446 			tlb_info[i].ti_flags = 0;
1447 		}
1448 	}
1449 	return (0);
1450 }
1451 
1452 /*
1453  * Allocate a context.  If necessary, steal one from someone else.
1454  *
1455  * The new context is flushed from the TLB before returning.
1456  */
1457 int
1458 ctx_alloc(struct pmap *pm)
1459 {
1460 	int s, cnum;
1461 	static int next = MINCTX;
1462 
1463 	if (pm == pmap_kernel()) {
1464 #ifdef DIAGNOSTIC
1465 		printf("ctx_alloc: kernel pmap!\n");
1466 #endif
1467 		return (0);
1468 	}
1469 	s = splvm();
1470 
1471 	/* Find a likely context. */
1472 	cnum = next;
1473 	do {
1474 		if ((++cnum) > NUMCTX)
1475 			cnum = MINCTX;
1476 	} while (ctxbusy[cnum] != NULL && cnum != next);
1477 
1478 	/* Now clean it out */
1479 oops:
1480 	if (cnum < MINCTX)
1481 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
1482 	if (ctx_flush(cnum)) {
1483 		/* oops -- something's wired. */
1484 		if ((++cnum) > NUMCTX)
1485 			cnum = MINCTX;
1486 		goto oops;
1487 	}
1488 
1489 	if (ctxbusy[cnum]) {
1490 #ifdef DEBUG
1491 		/* We should identify this pmap and clear it */
1492 		printf("Warning: stealing context %d\n", cnum);
1493 #endif
1494 		ctxbusy[cnum]->pm_ctx = 0;
1495 	}
1496 	ctxbusy[cnum] = pm;
1497 	next = cnum;
1498 	splx(s);
1499 	pm->pm_ctx = cnum;
1500 
1501 	return cnum;
1502 }
1503 
1504 /*
1505  * Give away a context.
1506  */
1507 void
1508 ctx_free(struct pmap *pm)
1509 {
1510 	int oldctx;
1511 
1512 	oldctx = pm->pm_ctx;
1513 
1514 	if (oldctx == 0)
1515 		panic("ctx_free: freeing kernel context");
1516 #ifdef DIAGNOSTIC
1517 	if (ctxbusy[oldctx] == 0)
1518 		printf("ctx_free: freeing free context %d\n", oldctx);
1519 	if (ctxbusy[oldctx] != pm) {
1520 		printf("ctx_free: freeing someone esle's context\n "
1521 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
1522 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
1523 #ifdef DDB
1524 		Debugger();
1525 #endif
1526 	}
1527 #endif
1528 	/* We should verify it has not been stolen and reallocated... */
1529 	ctxbusy[oldctx] = NULL;
1530 	ctx_flush(oldctx);
1531 }
1532 
1533 
1534 #ifdef DEBUG
1535 /*
1536  * Test ref/modify handling.
1537  */
1538 void pmap_testout __P((void));
1539 void
1540 pmap_testout()
1541 {
1542 	vaddr_t va;
1543 	volatile int *loc;
1544 	int val = 0;
1545 	paddr_t pa;
1546 	struct vm_page *pg;
1547 	int ref, mod;
1548 
1549 	/* Allocate a page */
1550 	va = (vaddr_t)uvm_km_alloc1(kernel_map, NBPG, 1);
1551 	loc = (int*)va;
1552 
1553 	pmap_extract(pmap_kernel(), va, &pa);
1554 	pg = PHYS_TO_VM_PAGE(pa);
1555 	pmap_unwire(pmap_kernel(), va);
1556 
1557 	pmap_remove(pmap_kernel(), va, va+1);
1558 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1559 	pmap_update(pmap_kernel());
1560 
1561 	/* Now clear reference and modify */
1562 	ref = pmap_clear_reference(pg);
1563 	mod = pmap_clear_modify(pg);
1564 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1565 	       (void *)(u_long)va, (long)pa,
1566 	       ref, mod);
1567 
1568 	/* Check it's properly cleared */
1569 	ref = pmap_is_referenced(pg);
1570 	mod = pmap_is_modified(pg);
1571 	printf("Checking cleared page: ref %d, mod %d\n",
1572 	       ref, mod);
1573 
1574 	/* Reference page */
1575 	val = *loc;
1576 
1577 	ref = pmap_is_referenced(pg);
1578 	mod = pmap_is_modified(pg);
1579 	printf("Referenced page: ref %d, mod %d val %x\n",
1580 	       ref, mod, val);
1581 
1582 	/* Now clear reference and modify */
1583 	ref = pmap_clear_reference(pg);
1584 	mod = pmap_clear_modify(pg);
1585 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1586 	       (void *)(u_long)va, (long)pa,
1587 	       ref, mod);
1588 
1589 	/* Modify page */
1590 	*loc = 1;
1591 
1592 	ref = pmap_is_referenced(pg);
1593 	mod = pmap_is_modified(pg);
1594 	printf("Modified page: ref %d, mod %d\n",
1595 	       ref, mod);
1596 
1597 	/* Now clear reference and modify */
1598 	ref = pmap_clear_reference(pg);
1599 	mod = pmap_clear_modify(pg);
1600 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1601 	       (void *)(u_long)va, (long)pa,
1602 	       ref, mod);
1603 
1604 	/* Check it's properly cleared */
1605 	ref = pmap_is_referenced(pg);
1606 	mod = pmap_is_modified(pg);
1607 	printf("Checking cleared page: ref %d, mod %d\n",
1608 	       ref, mod);
1609 
1610 	/* Modify page */
1611 	*loc = 1;
1612 
1613 	ref = pmap_is_referenced(pg);
1614 	mod = pmap_is_modified(pg);
1615 	printf("Modified page: ref %d, mod %d\n",
1616 	       ref, mod);
1617 
1618 	/* Check pmap_protect() */
1619 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
1620 	pmap_update(pmap_kernel());
1621 	ref = pmap_is_referenced(pg);
1622 	mod = pmap_is_modified(pg);
1623 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
1624 	       ref, mod);
1625 
1626 	/* Now clear reference and modify */
1627 	ref = pmap_clear_reference(pg);
1628 	mod = pmap_clear_modify(pg);
1629 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1630 	       (void *)(u_long)va, (long)pa,
1631 	       ref, mod);
1632 
1633 	/* Reference page */
1634 	val = *loc;
1635 
1636 	ref = pmap_is_referenced(pg);
1637 	mod = pmap_is_modified(pg);
1638 	printf("Referenced page: ref %d, mod %d val %x\n",
1639 	       ref, mod, val);
1640 
1641 	/* Now clear reference and modify */
1642 	ref = pmap_clear_reference(pg);
1643 	mod = pmap_clear_modify(pg);
1644 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1645 	       (void *)(u_long)va, (long)pa,
1646 	       ref, mod);
1647 
1648 	/* Modify page */
1649 #if 0
1650 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1651 	pmap_update(pmap_kernel());
1652 #endif
1653 	*loc = 1;
1654 
1655 	ref = pmap_is_referenced(pg);
1656 	mod = pmap_is_modified(pg);
1657 	printf("Modified page: ref %d, mod %d\n",
1658 	       ref, mod);
1659 
1660 	/* Check pmap_protect() */
1661 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
1662 	pmap_update(pmap_kernel());
1663 	ref = pmap_is_referenced(pg);
1664 	mod = pmap_is_modified(pg);
1665 	printf("pmap_protect(): ref %d, mod %d\n",
1666 	       ref, mod);
1667 
1668 	/* Now clear reference and modify */
1669 	ref = pmap_clear_reference(pg);
1670 	mod = pmap_clear_modify(pg);
1671 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1672 	       (void *)(u_long)va, (long)pa,
1673 	       ref, mod);
1674 
1675 	/* Reference page */
1676 	val = *loc;
1677 
1678 	ref = pmap_is_referenced(pg);
1679 	mod = pmap_is_modified(pg);
1680 	printf("Referenced page: ref %d, mod %d val %x\n",
1681 	       ref, mod, val);
1682 
1683 	/* Now clear reference and modify */
1684 	ref = pmap_clear_reference(pg);
1685 	mod = pmap_clear_modify(pg);
1686 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1687 	       (void *)(u_long)va, (long)pa,
1688 	       ref, mod);
1689 
1690 	/* Modify page */
1691 #if 0
1692 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1693 	pmap_update(pmap_kernel());
1694 #endif
1695 	*loc = 1;
1696 
1697 	ref = pmap_is_referenced(pg);
1698 	mod = pmap_is_modified(pg);
1699 	printf("Modified page: ref %d, mod %d\n",
1700 	       ref, mod);
1701 
1702 	/* Check pmap_pag_protect() */
1703 	pmap_page_protect(pg, VM_PROT_READ);
1704 	ref = pmap_is_referenced(pg);
1705 	mod = pmap_is_modified(pg);
1706 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
1707 	       ref, mod);
1708 
1709 	/* Now clear reference and modify */
1710 	ref = pmap_clear_reference(pg);
1711 	mod = pmap_clear_modify(pg);
1712 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1713 	       (void *)(u_long)va, (long)pa,
1714 	       ref, mod);
1715 
1716 	/* Reference page */
1717 	val = *loc;
1718 
1719 	ref = pmap_is_referenced(pg);
1720 	mod = pmap_is_modified(pg);
1721 	printf("Referenced page: ref %d, mod %d val %x\n",
1722 	       ref, mod, val);
1723 
1724 	/* Now clear reference and modify */
1725 	ref = pmap_clear_reference(pg);
1726 	mod = pmap_clear_modify(pg);
1727 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1728 	       (void *)(u_long)va, (long)pa,
1729 	       ref, mod);
1730 
1731 	/* Modify page */
1732 #if 0
1733 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1734 	pmap_update(pmap_kernel());
1735 #endif
1736 	*loc = 1;
1737 
1738 	ref = pmap_is_referenced(pg);
1739 	mod = pmap_is_modified(pg);
1740 	printf("Modified page: ref %d, mod %d\n",
1741 	       ref, mod);
1742 
1743 	/* Check pmap_pag_protect() */
1744 	pmap_page_protect(pg, VM_PROT_NONE);
1745 	ref = pmap_is_referenced(pg);
1746 	mod = pmap_is_modified(pg);
1747 	printf("pmap_page_protect(): ref %d, mod %d\n",
1748 	       ref, mod);
1749 
1750 	/* Now clear reference and modify */
1751 	ref = pmap_clear_reference(pg);
1752 	mod = pmap_clear_modify(pg);
1753 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1754 	       (void *)(u_long)va, (long)pa,
1755 	       ref, mod);
1756 
1757 
1758 	/* Reference page */
1759 	val = *loc;
1760 
1761 	ref = pmap_is_referenced(pg);
1762 	mod = pmap_is_modified(pg);
1763 	printf("Referenced page: ref %d, mod %d val %x\n",
1764 	       ref, mod, val);
1765 
1766 	/* Now clear reference and modify */
1767 	ref = pmap_clear_reference(pg);
1768 	mod = pmap_clear_modify(pg);
1769 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1770 	       (void *)(u_long)va, (long)pa,
1771 	       ref, mod);
1772 
1773 	/* Modify page */
1774 #if 0
1775 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1776 	pmap_update(pmap_kernel());
1777 #endif
1778 	*loc = 1;
1779 
1780 	ref = pmap_is_referenced(pg);
1781 	mod = pmap_is_modified(pg);
1782 	printf("Modified page: ref %d, mod %d\n",
1783 	       ref, mod);
1784 
1785 	/* Unmap page */
1786 	pmap_remove(pmap_kernel(), va, va+1);
1787 	pmap_update(pmap_kernel());
1788 	ref = pmap_is_referenced(pg);
1789 	mod = pmap_is_modified(pg);
1790 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
1791 
1792 	/* Now clear reference and modify */
1793 	ref = pmap_clear_reference(pg);
1794 	mod = pmap_clear_modify(pg);
1795 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1796 	       (void *)(u_long)va, (long)pa, ref, mod);
1797 
1798 	/* Check it's properly cleared */
1799 	ref = pmap_is_referenced(pg);
1800 	mod = pmap_is_modified(pg);
1801 	printf("Checking cleared page: ref %d, mod %d\n",
1802 	       ref, mod);
1803 
1804 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL,
1805 		VM_PROT_ALL|PMAP_WIRED);
1806 	uvm_km_free(kernel_map, (vaddr_t)va, NBPG);
1807 }
1808 #endif
1809