xref: /netbsd-src/sys/arch/powerpc/ibm4xx/pmap.c (revision d25ffa98a4bfca1fe272f3c182496ec9934faac7)
1 /*	$NetBSD: pmap.c,v 1.67 2011/06/18 06:41:41 matt 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 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.67 2011/06/18 06:41:41 matt Exp $");
71 
72 #include <sys/param.h>
73 #include <sys/malloc.h>
74 #include <sys/proc.h>
75 #include <sys/queue.h>
76 #include <sys/systm.h>
77 #include <sys/pool.h>
78 #include <sys/device.h>
79 #include <sys/cpu.h>
80 
81 #include <uvm/uvm.h>
82 
83 #include <machine/pcb.h>
84 #include <machine/powerpc.h>
85 #include <machine/tlb.h>
86 
87 #include <powerpc/spr.h>
88 #include <powerpc/ibm4xx/spr.h>
89 
90 #include <powerpc/ibm4xx/cpu.h>
91 
92 /*
93  * kernmap is an array of PTEs large enough to map in
94  * 4GB.  At 16KB/page it is 256K entries or 2MB.
95  */
96 #define KERNMAP_SIZE	((0xffffffffU/PAGE_SIZE)+1)
97 void *kernmap;
98 
99 #define MINCTX		2
100 #define NUMCTX		256
101 
102 volatile struct pmap *ctxbusy[NUMCTX];
103 
104 #define TLBF_USED	0x1
105 #define	TLBF_REF	0x2
106 #define	TLBF_LOCKED	0x4
107 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
108 
109 typedef struct tlb_info_s {
110 	char	ti_flags;
111 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
112 	u_int	ti_va;
113 } tlb_info_t;
114 
115 volatile tlb_info_t tlb_info[NTLB];
116 /* We'll use a modified FIFO replacement policy cause it's cheap */
117 volatile int tlbnext;
118 
119 static int tlb_nreserved = 0;
120 static int pmap_bootstrap_done = 0;
121 
122 /* Event counters */
123 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
124 	NULL, "cpu", "tlbmiss");
125 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
126 	NULL, "cpu", "tlbhit");
127 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
128 	NULL, "cpu", "tlbflush");
129 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
130 	NULL, "cpu", "tlbenter");
131 EVCNT_ATTACH_STATIC(tlbmiss_ev);
132 EVCNT_ATTACH_STATIC(tlbhit_ev);
133 EVCNT_ATTACH_STATIC(tlbflush_ev);
134 EVCNT_ATTACH_STATIC(tlbenter_ev);
135 
136 struct pmap kernel_pmap_;
137 struct pmap *const kernel_pmap_ptr = &kernel_pmap_;
138 
139 static int npgs;
140 static u_int nextavail;
141 #ifndef MSGBUFADDR
142 extern paddr_t msgbuf_paddr;
143 #endif
144 
145 static struct mem_region *mem, *avail;
146 
147 /*
148  * This is a cache of referenced/modified bits.
149  * Bits herein are shifted by ATTRSHFT.
150  */
151 static char *pmap_attrib;
152 
153 #define PV_WIRED	0x1
154 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
155 #define PV_UNWIRE(pv)	((pv)->pv_va &= ~PV_WIRED)
156 #define PV_ISWIRED(pv)	((pv)->pv_va & PV_WIRED)
157 #define PV_CMPVA(va,pv)	(!(((pv)->pv_va ^ (va)) & (~PV_WIRED)))
158 
159 struct pv_entry {
160 	struct pv_entry *pv_next;	/* Linked list of mappings */
161 	vaddr_t pv_va;			/* virtual address of mapping */
162 	struct pmap *pv_pm;
163 };
164 
165 /* Each index corresponds to TLB_SIZE_* value. */
166 static size_t tlbsize[] = {
167 	1024, 		/* TLB_SIZE_1K */
168 	4096, 		/* TLB_SIZE_4K */
169 	16384, 		/* TLB_SIZE_16K */
170 	65536, 		/* TLB_SIZE_64K */
171 	262144, 	/* TLB_SIZE_256K */
172 	1048576, 	/* TLB_SIZE_1M */
173 	4194304, 	/* TLB_SIZE_4M */
174 	16777216, 	/* TLB_SIZE_16M */
175 };
176 
177 struct pv_entry *pv_table;
178 static struct pool pv_pool;
179 
180 static int pmap_initialized;
181 
182 static int ctx_flush(int);
183 
184 inline struct pv_entry *pa_to_pv(paddr_t);
185 static inline char *pa_to_attr(paddr_t);
186 
187 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
188 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
189 
190 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t, int);
191 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
192 
193 static int ppc4xx_tlb_size_mask(size_t, int *, int *);
194 
195 
196 inline struct pv_entry *
197 pa_to_pv(paddr_t pa)
198 {
199 	int bank, pg;
200 
201 	bank = vm_physseg_find(atop(pa), &pg);
202 	if (bank == -1)
203 		return NULL;
204 	return &VM_PHYSMEM_PTR(bank)->pmseg.pvent[pg];
205 }
206 
207 static inline char *
208 pa_to_attr(paddr_t pa)
209 {
210 	int bank, pg;
211 
212 	bank = vm_physseg_find(atop(pa), &pg);
213 	if (bank == -1)
214 		return NULL;
215 	return &VM_PHYSMEM_PTR(bank)->pmseg.attrs[pg];
216 }
217 
218 /*
219  * Insert PTE into page table.
220  */
221 int
222 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
223 {
224 	int seg = STIDX(va);
225 	int ptn = PTIDX(va);
226 	u_int oldpte;
227 
228 	if (!pm->pm_ptbl[seg]) {
229 		/* Don't allocate a page to clear a non-existent mapping. */
230 		if (!pte)
231 			return (0);
232 		/* Allocate a page XXXX this will sleep! */
233 		pm->pm_ptbl[seg] =
234 		    (uint *)uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
235 		    UVM_KMF_WIRED | UVM_KMF_ZERO);
236 	}
237 	oldpte = pm->pm_ptbl[seg][ptn];
238 	pm->pm_ptbl[seg][ptn] = pte;
239 
240 	/* Flush entry. */
241 	ppc4xx_tlb_flush(va, pm->pm_ctx);
242 	if (oldpte != pte) {
243 		if (pte == 0)
244 			pm->pm_stats.resident_count--;
245 		else
246 			pm->pm_stats.resident_count++;
247 	}
248 	return (1);
249 }
250 
251 /*
252  * Get a pointer to a PTE in a page table.
253  */
254 volatile u_int *
255 pte_find(struct pmap *pm, vaddr_t va)
256 {
257 	int seg = STIDX(va);
258 	int ptn = PTIDX(va);
259 
260 	if (pm->pm_ptbl[seg])
261 		return (&pm->pm_ptbl[seg][ptn]);
262 
263 	return (NULL);
264 }
265 
266 /*
267  * This is called during initppc, before the system is really initialized.
268  */
269 void
270 pmap_bootstrap(u_int kernelstart, u_int kernelend)
271 {
272 	struct mem_region *mp, *mp1;
273 	int cnt, i;
274 	u_int s, e, sz;
275 
276 	tlbnext = tlb_nreserved;
277 
278 	/*
279 	 * Allocate the kernel page table at the end of
280 	 * kernel space so it's in the locked TTE.
281 	 */
282 	kernmap = (void *)kernelend;
283 
284 	/*
285 	 * Initialize kernel page table.
286 	 */
287 	for (i = 0; i < STSZ; i++) {
288 		pmap_kernel()->pm_ptbl[i] = 0;
289 	}
290 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
291 
292 	/*
293 	 * Announce page-size to the VM-system
294 	 */
295 	uvmexp.pagesize = NBPG;
296 	uvm_setpagesize();
297 
298 	/*
299 	 * Get memory.
300 	 */
301 	mem_regions(&mem, &avail);
302 	for (mp = mem; mp->size; mp++) {
303 		physmem += btoc(mp->size);
304 		printf("+%lx,",mp->size);
305 	}
306 	printf("\n");
307 	ppc4xx_tlb_init();
308 	/*
309 	 * Count the number of available entries.
310 	 */
311 	for (cnt = 0, mp = avail; mp->size; mp++)
312 		cnt++;
313 
314 	/*
315 	 * Page align all regions.
316 	 * Non-page aligned memory isn't very interesting to us.
317 	 * Also, sort the entries for ascending addresses.
318 	 */
319 	kernelstart &= ~PGOFSET;
320 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
321 	for (mp = avail; mp->size; mp++) {
322 		s = mp->start;
323 		e = mp->start + mp->size;
324 		printf("%08x-%08x -> ",s,e);
325 		/*
326 		 * Check whether this region holds all of the kernel.
327 		 */
328 		if (s < kernelstart && e > kernelend) {
329 			avail[cnt].start = kernelend;
330 			avail[cnt++].size = e - kernelend;
331 			e = kernelstart;
332 		}
333 		/*
334 		 * Look whether this regions starts within the kernel.
335 		 */
336 		if (s >= kernelstart && s < kernelend) {
337 			if (e <= kernelend)
338 				goto empty;
339 			s = kernelend;
340 		}
341 		/*
342 		 * Now look whether this region ends within the kernel.
343 		 */
344 		if (e > kernelstart && e <= kernelend) {
345 			if (s >= kernelstart)
346 				goto empty;
347 			e = kernelstart;
348 		}
349 		/*
350 		 * Now page align the start and size of the region.
351 		 */
352 		s = round_page(s);
353 		e = trunc_page(e);
354 		if (e < s)
355 			e = s;
356 		sz = e - s;
357 		printf("%08x-%08x = %x\n",s,e,sz);
358 		/*
359 		 * Check whether some memory is left here.
360 		 */
361 		if (sz == 0) {
362 		empty:
363 			memmove(mp, mp + 1,
364 				(cnt - (mp - avail)) * sizeof *mp);
365 			cnt--;
366 			mp--;
367 			continue;
368 		}
369 		/*
370 		 * Do an insertion sort.
371 		 */
372 		npgs += btoc(sz);
373 		for (mp1 = avail; mp1 < mp; mp1++)
374 			if (s < mp1->start)
375 				break;
376 		if (mp1 < mp) {
377 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
378 			mp1->start = s;
379 			mp1->size = sz;
380 		} else {
381 			mp->start = s;
382 			mp->size = sz;
383 		}
384 	}
385 
386 	/*
387 	 * We cannot do pmap_steal_memory here,
388 	 * since we don't run with translation enabled yet.
389 	 */
390 #ifndef MSGBUFADDR
391 	/*
392 	 * allow for msgbuf
393 	 */
394 	sz = round_page(MSGBUFSIZE);
395 	mp = NULL;
396 	for (mp1 = avail; mp1->size; mp1++)
397 		if (mp1->size >= sz)
398 			mp = mp1;
399 	if (mp == NULL)
400 		panic("not enough memory?");
401 
402 	npgs -= btoc(sz);
403 	msgbuf_paddr = mp->start + mp->size - sz;
404 	mp->size -= sz;
405 	if (mp->size <= 0)
406 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
407 #endif
408 
409 	for (mp = avail; mp->size; mp++)
410 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
411 			atop(mp->start), atop(mp->start + mp->size),
412 			VM_FREELIST_DEFAULT);
413 
414 	/*
415 	 * Initialize kernel pmap and hardware.
416 	 */
417 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
418 	pmap_kernel()->pm_ctx = KERNEL_PID;
419 	nextavail = avail->start;
420 
421 	pmap_bootstrap_done = 1;
422 }
423 
424 /*
425  * Restrict given range to physical memory
426  *
427  * (Used by /dev/mem)
428  */
429 void
430 pmap_real_memory(paddr_t *start, psize_t *size)
431 {
432 	struct mem_region *mp;
433 
434 	for (mp = mem; mp->size; mp++) {
435 		if (*start + *size > mp->start &&
436 		    *start < mp->start + mp->size) {
437 			if (*start < mp->start) {
438 				*size -= mp->start - *start;
439 				*start = mp->start;
440 			}
441 			if (*start + *size > mp->start + mp->size)
442 				*size = mp->start + mp->size - *start;
443 			return;
444 		}
445 	}
446 	*size = 0;
447 }
448 
449 /*
450  * Initialize anything else for pmap handling.
451  * Called during vm_init().
452  */
453 void
454 pmap_init(void)
455 {
456 	struct pv_entry *pv;
457 	vsize_t sz;
458 	vaddr_t addr;
459 	int i, s;
460 	int bank;
461 	char *attr;
462 
463 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
464 	sz = round_page(sz);
465 	addr = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED | UVM_KMF_ZERO);
466 	s = splvm();
467 	pv = pv_table = (struct pv_entry *)addr;
468 	for (i = npgs; --i >= 0;)
469 		pv++->pv_pm = NULL;
470 	pmap_attrib = (char *)pv;
471 	memset(pv, 0, npgs);
472 
473 	pv = pv_table;
474 	attr = pmap_attrib;
475 	for (bank = 0; bank < vm_nphysseg; bank++) {
476 		sz = VM_PHYSMEM_PTR(bank)->end - VM_PHYSMEM_PTR(bank)->start;
477 		VM_PHYSMEM_PTR(bank)->pmseg.pvent = pv;
478 		VM_PHYSMEM_PTR(bank)->pmseg.attrs = attr;
479 		pv += sz;
480 		attr += sz;
481 	}
482 
483 	pmap_initialized = 1;
484 	splx(s);
485 
486 	/* Setup a pool for additional pvlist structures */
487 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL,
488 	    IPL_VM);
489 }
490 
491 /*
492  * How much virtual space is available to the kernel?
493  */
494 void
495 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
496 {
497 
498 #if 0
499 	/*
500 	 * Reserve one segment for kernel virtual memory
501 	 */
502 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
503 	*end = *start + SEGMENT_LENGTH;
504 #else
505 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
506 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
507 #endif
508 }
509 
510 #ifdef PMAP_GROWKERNEL
511 /*
512  * Preallocate kernel page tables to a specified VA.
513  * This simply loops through the first TTE for each
514  * page table from the beginning of the kernel pmap,
515  * reads the entry, and if the result is
516  * zero (either invalid entry or no page table) it stores
517  * a zero there, populating page tables in the process.
518  * This is not the most efficient technique but i don't
519  * expect it to be called that often.
520  */
521 extern struct vm_page *vm_page_alloc1(void);
522 extern void vm_page_free1(struct vm_page *);
523 
524 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
525 
526 vaddr_t
527 pmap_growkernel(vaddr_t maxkvaddr)
528 {
529 	int s;
530 	int seg;
531 	paddr_t pg;
532 	struct pmap *pm = pmap_kernel();
533 
534 	s = splvm();
535 
536 	/* Align with the start of a page table */
537 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
538 	     kbreak += PTMAP) {
539 		seg = STIDX(kbreak);
540 
541 		if (pte_find(pm, kbreak))
542 			continue;
543 
544 		if (uvm.page_init_done) {
545 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
546 		} else {
547 			if (!uvm_page_physget(&pg))
548 				panic("pmap_growkernel: no memory");
549 		}
550 		if (!pg)
551 			panic("pmap_growkernel: no pages");
552 		pmap_zero_page((paddr_t)pg);
553 
554 		/* XXX This is based on all phymem being addressable */
555 		pm->pm_ptbl[seg] = (u_int *)pg;
556 	}
557 	splx(s);
558 	return (kbreak);
559 }
560 
561 /*
562  *	vm_page_alloc1:
563  *
564  *	Allocate and return a memory cell with no associated object.
565  */
566 struct vm_page *
567 vm_page_alloc1(void)
568 {
569 	struct vm_page *pg;
570 
571 	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
572 	if (pg) {
573 		pg->wire_count = 1;	/* no mappings yet */
574 		pg->flags &= ~PG_BUSY;	/* never busy */
575 	}
576 	return pg;
577 }
578 
579 /*
580  *	vm_page_free1:
581  *
582  *	Returns the given page to the free list,
583  *	disassociating it with any VM object.
584  *
585  *	Object and page must be locked prior to entry.
586  */
587 void
588 vm_page_free1(struct vm_page *pg)
589 {
590 #ifdef DIAGNOSTIC
591 	if (pg->flags != (PG_CLEAN|PG_FAKE)) {
592 		printf("Freeing invalid page %p\n", pg);
593 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(pg));
594 #ifdef DDB
595 		Debugger();
596 #endif
597 		return;
598 	}
599 #endif
600 	pg->flags |= PG_BUSY;
601 	pg->wire_count = 0;
602 	uvm_pagefree(pg);
603 }
604 #endif
605 
606 /*
607  * Create and return a physical map.
608  */
609 struct pmap *
610 pmap_create(void)
611 {
612 	struct pmap *pm;
613 
614 	pm = malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
615 	memset(pm, 0, sizeof *pm);
616 	pm->pm_refs = 1;
617 	return pm;
618 }
619 
620 /*
621  * Add a reference to the given pmap.
622  */
623 void
624 pmap_reference(struct pmap *pm)
625 {
626 
627 	pm->pm_refs++;
628 }
629 
630 /*
631  * Retire the given pmap from service.
632  * Should only be called if the map contains no valid mappings.
633  */
634 void
635 pmap_destroy(struct pmap *pm)
636 {
637 	int i;
638 
639 	if (--pm->pm_refs > 0) {
640 		return;
641 	}
642 	KASSERT(pm->pm_stats.resident_count == 0);
643 	KASSERT(pm->pm_stats.wired_count == 0);
644 	for (i = 0; i < STSZ; i++)
645 		if (pm->pm_ptbl[i]) {
646 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i],
647 			    PAGE_SIZE, UVM_KMF_WIRED);
648 			pm->pm_ptbl[i] = NULL;
649 		}
650 	if (pm->pm_ctx)
651 		ctx_free(pm);
652 	free(pm, M_VMPMAP);
653 }
654 
655 /*
656  * Copy the range specified by src_addr/len
657  * from the source map to the range dst_addr/len
658  * in the destination map.
659  *
660  * This routine is only advisory and need not do anything.
661  */
662 void
663 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
664 	  vsize_t len, vaddr_t src_addr)
665 {
666 }
667 
668 /*
669  * Require that all active physical maps contain no
670  * incorrect entries NOW.
671  */
672 void
673 pmap_update(struct pmap *pmap)
674 {
675 }
676 
677 /*
678  * Fill the given physical page with zeroes.
679  */
680 void
681 pmap_zero_page(paddr_t pa)
682 {
683 
684 #ifdef PPC_4XX_NOCACHE
685 	memset((void *)pa, 0, PAGE_SIZE);
686 #else
687 	int i;
688 
689 	for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
690 		__asm volatile ("dcbz 0,%0" :: "r"(pa));
691 		pa += CACHELINESIZE;
692 	}
693 #endif
694 }
695 
696 /*
697  * Copy the given physical source page to its destination.
698  */
699 void
700 pmap_copy_page(paddr_t src, paddr_t dst)
701 {
702 
703 	memcpy((void *)dst, (void *)src, PAGE_SIZE);
704 	dcache_flush_page(dst);
705 }
706 
707 /*
708  * This returns != 0 on success.
709  */
710 static inline int
711 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa, int flags)
712 {
713 	struct pv_entry *pv, *npv = NULL;
714 	int s;
715 
716 	if (!pmap_initialized)
717 		return 0;
718 
719 	s = splvm();
720 	pv = pa_to_pv(pa);
721 	if (!pv->pv_pm) {
722 		/*
723 		 * No entries yet, use header as the first entry.
724 		 */
725 		pv->pv_va = va;
726 		pv->pv_pm = pm;
727 		pv->pv_next = NULL;
728 	} else {
729 		/*
730 		 * There is at least one other VA mapping this page.
731 		 * Place this entry after the header.
732 		 */
733 		npv = pool_get(&pv_pool, PR_NOWAIT);
734 		if (npv == NULL) {
735 			if ((flags & PMAP_CANFAIL) == 0)
736 				panic("pmap_enter_pv: failed");
737 			splx(s);
738 			return 0;
739 		}
740 		npv->pv_va = va;
741 		npv->pv_pm = pm;
742 		npv->pv_next = pv->pv_next;
743 		pv->pv_next = npv;
744 		pv = npv;
745 	}
746 	if (flags & PMAP_WIRED) {
747 		PV_WIRE(pv);
748 		pm->pm_stats.wired_count++;
749 	}
750 	splx(s);
751 	return (1);
752 }
753 
754 static void
755 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
756 {
757 	struct pv_entry *pv, *npv;
758 
759 	/*
760 	 * Remove from the PV table.
761 	 */
762 	pv = pa_to_pv(pa);
763 	if (!pv)
764 		return;
765 
766 	/*
767 	 * If it is the first entry on the list, it is actually
768 	 * in the header and we must copy the following entry up
769 	 * to the header.  Otherwise we must search the list for
770 	 * the entry.  In either case we free the now unused entry.
771 	 */
772 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
773 		if (PV_ISWIRED(pv)) {
774 			pm->pm_stats.wired_count--;
775 		}
776 		if ((npv = pv->pv_next)) {
777 			*pv = *npv;
778 			pool_put(&pv_pool, npv);
779 		} else
780 			pv->pv_pm = NULL;
781 	} else {
782 		for (; (npv = pv->pv_next) != NULL; pv = npv)
783 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
784 				break;
785 		if (npv) {
786 			pv->pv_next = npv->pv_next;
787 			if (PV_ISWIRED(npv)) {
788 				pm->pm_stats.wired_count--;
789 			}
790 			pool_put(&pv_pool, npv);
791 		}
792 	}
793 }
794 
795 /*
796  * Insert physical page at pa into the given pmap at virtual address va.
797  */
798 int
799 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
800 {
801 	int s;
802 	u_int tte;
803 	bool managed;
804 
805 	/*
806 	 * Have to remove any existing mapping first.
807 	 */
808 	pmap_remove(pm, va, va + PAGE_SIZE);
809 
810 	if (flags & PMAP_WIRED)
811 		flags |= prot;
812 
813 	managed = uvm_pageismanaged(pa);
814 
815 	/*
816 	 * Generate TTE.
817 	 */
818 	tte = TTE_PA(pa);
819 	/* XXXX -- need to support multiple page sizes. */
820 	tte |= TTE_SZ_16K;
821 #ifdef	DIAGNOSTIC
822 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
823 		(PME_NOCACHE | PME_WRITETHROUG))
824 		panic("pmap_enter: uncached & writethrough");
825 #endif
826 	if (flags & PME_NOCACHE)
827 		/* Must be I/O mapping */
828 		tte |= TTE_I | TTE_G;
829 #ifdef PPC_4XX_NOCACHE
830 	tte |= TTE_I;
831 #else
832 	else if (flags & PME_WRITETHROUG)
833 		/* Uncached and writethrough are not compatible */
834 		tte |= TTE_W;
835 #endif
836 	if (pm == pmap_kernel())
837 		tte |= TTE_ZONE(ZONE_PRIV);
838 	else
839 		tte |= TTE_ZONE(ZONE_USER);
840 
841 	if (flags & VM_PROT_WRITE)
842 		tte |= TTE_WR;
843 
844 	if (flags & VM_PROT_EXECUTE)
845 		tte |= TTE_EX;
846 
847 	/*
848 	 * Now record mapping for later back-translation.
849 	 */
850 	if (pmap_initialized && managed) {
851 		char *attr;
852 
853 		if (!pmap_enter_pv(pm, va, pa, flags)) {
854 			/* Could not enter pv on a managed page */
855 			return 1;
856 		}
857 
858 		/* Now set attributes. */
859 		attr = pa_to_attr(pa);
860 #ifdef DIAGNOSTIC
861 		if (!attr)
862 			panic("managed but no attr");
863 #endif
864 		if (flags & VM_PROT_ALL)
865 			*attr |= PMAP_ATTR_REF;
866 		if (flags & VM_PROT_WRITE)
867 			*attr |= PMAP_ATTR_CHG;
868 	}
869 
870 	s = splvm();
871 
872 	/* Insert page into page table. */
873 	pte_enter(pm, va, tte);
874 
875 	/* If this is a real fault, enter it in the tlb */
876 	if (tte && ((flags & PMAP_WIRED) == 0)) {
877 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
878 	}
879 	splx(s);
880 
881 	/* Flush the real memory from the instruction cache. */
882 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
883 		__syncicache((void *)pa, PAGE_SIZE);
884 
885 	return 0;
886 }
887 
888 void
889 pmap_unwire(struct pmap *pm, vaddr_t va)
890 {
891 	struct pv_entry *pv;
892 	paddr_t pa;
893 	int s;
894 
895 	if (!pmap_extract(pm, va, &pa)) {
896 		return;
897 	}
898 
899 	pv = pa_to_pv(pa);
900 	if (!pv)
901 		return;
902 
903 	s = splvm();
904 	while (pv != NULL) {
905 		if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
906 			if (PV_ISWIRED(pv)) {
907 				PV_UNWIRE(pv);
908 				pm->pm_stats.wired_count--;
909 			}
910 			break;
911 		}
912 		pv = pv->pv_next;
913 	}
914 	splx(s);
915 }
916 
917 void
918 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
919 {
920 	int s;
921 	u_int tte;
922 	struct pmap *pm = pmap_kernel();
923 
924 	/*
925 	 * Have to remove any existing mapping first.
926 	 */
927 
928 	/*
929 	 * Generate TTE.
930 	 *
931 	 * XXXX
932 	 *
933 	 * Since the kernel does not handle execution privileges properly,
934 	 * we will handle read and execute permissions together.
935 	 */
936 	tte = 0;
937 	if (prot & VM_PROT_ALL) {
938 
939 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
940 		/* XXXX -- need to support multiple page sizes. */
941 		tte |= TTE_SZ_16K;
942 #ifdef DIAGNOSTIC
943 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
944 			(PME_NOCACHE | PME_WRITETHROUG))
945 			panic("pmap_kenter_pa: uncached & writethrough");
946 #endif
947 		if (prot & PME_NOCACHE)
948 			/* Must be I/O mapping */
949 			tte |= TTE_I | TTE_G;
950 #ifdef PPC_4XX_NOCACHE
951 		tte |= TTE_I;
952 #else
953 		else if (prot & PME_WRITETHROUG)
954 			/* Uncached and writethrough are not compatible */
955 			tte |= TTE_W;
956 #endif
957 		if (prot & VM_PROT_WRITE)
958 			tte |= TTE_WR;
959 	}
960 
961 	s = splvm();
962 
963 	/* Insert page into page table. */
964 	pte_enter(pm, va, tte);
965 	splx(s);
966 }
967 
968 void
969 pmap_kremove(vaddr_t va, vsize_t len)
970 {
971 
972 	while (len > 0) {
973 		pte_enter(pmap_kernel(), va, 0);
974 		va += PAGE_SIZE;
975 		len -= PAGE_SIZE;
976 	}
977 }
978 
979 /*
980  * Remove the given range of mapping entries.
981  */
982 void
983 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
984 {
985 	int s;
986 	paddr_t pa;
987 	volatile u_int *ptp;
988 
989 	s = splvm();
990 	while (va < endva) {
991 
992 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
993 			pa = TTE_PA(pa);
994 			pmap_remove_pv(pm, va, pa);
995 			*ptp = 0;
996 			ppc4xx_tlb_flush(va, pm->pm_ctx);
997 			pm->pm_stats.resident_count--;
998 		}
999 		va += PAGE_SIZE;
1000 	}
1001 
1002 	splx(s);
1003 }
1004 
1005 /*
1006  * Get the physical page address for the given pmap/virtual address.
1007  */
1008 bool
1009 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
1010 {
1011 	int seg = STIDX(va);
1012 	int ptn = PTIDX(va);
1013 	u_int pa = 0;
1014 	int s;
1015 
1016 	s = splvm();
1017 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
1018 		*pap = TTE_PA(pa) | (va & PGOFSET);
1019 	}
1020 	splx(s);
1021 	return (pa != 0);
1022 }
1023 
1024 /*
1025  * Lower the protection on the specified range of this pmap.
1026  *
1027  * There are only two cases: either the protection is going to 0,
1028  * or it is going to read-only.
1029  */
1030 void
1031 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1032 {
1033 	volatile u_int *ptp;
1034 	int s, bic;
1035 
1036 	if ((prot & VM_PROT_READ) == 0) {
1037 		pmap_remove(pm, sva, eva);
1038 		return;
1039 	}
1040 	bic = 0;
1041 	if ((prot & VM_PROT_WRITE) == 0) {
1042 		bic |= TTE_WR;
1043 	}
1044 	if ((prot & VM_PROT_EXECUTE) == 0) {
1045 		bic |= TTE_EX;
1046 	}
1047 	if (bic == 0) {
1048 		return;
1049 	}
1050 	s = splvm();
1051 	while (sva < eva) {
1052 		if ((ptp = pte_find(pm, sva)) != NULL) {
1053 			*ptp &= ~bic;
1054 			ppc4xx_tlb_flush(sva, pm->pm_ctx);
1055 		}
1056 		sva += PAGE_SIZE;
1057 	}
1058 	splx(s);
1059 }
1060 
1061 bool
1062 pmap_check_attr(struct vm_page *pg, u_int mask, int clear)
1063 {
1064 	paddr_t pa;
1065 	char *attr;
1066 	int s, rv;
1067 
1068 	/*
1069 	 * First modify bits in cache.
1070 	 */
1071 	pa = VM_PAGE_TO_PHYS(pg);
1072 	attr = pa_to_attr(pa);
1073 	if (attr == NULL)
1074 		return false;
1075 
1076 	s = splvm();
1077 	rv = ((*attr & mask) != 0);
1078 	if (clear) {
1079 		*attr &= ~mask;
1080 		pmap_page_protect(pg, mask == PMAP_ATTR_CHG ? VM_PROT_READ : 0);
1081 	}
1082 	splx(s);
1083 	return rv;
1084 }
1085 
1086 
1087 /*
1088  * Lower the protection on the specified physical page.
1089  *
1090  * There are only two cases: either the protection is going to 0,
1091  * or it is going to read-only.
1092  */
1093 void
1094 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1095 {
1096 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
1097 	vaddr_t va;
1098 	struct pv_entry *pvh, *pv, *npv;
1099 	struct pmap *pm;
1100 
1101 	pvh = pa_to_pv(pa);
1102 	if (pvh == NULL)
1103 		return;
1104 
1105 	/* Handle extra pvs which may be deleted in the operation */
1106 	for (pv = pvh->pv_next; pv; pv = npv) {
1107 		npv = pv->pv_next;
1108 
1109 		pm = pv->pv_pm;
1110 		va = pv->pv_va;
1111 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
1112 	}
1113 	/* Now check the head pv */
1114 	if (pvh->pv_pm) {
1115 		pv = pvh;
1116 		pm = pv->pv_pm;
1117 		va = pv->pv_va;
1118 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
1119 	}
1120 }
1121 
1122 /*
1123  * Activate the address space for the specified process.  If the process
1124  * is the current process, load the new MMU context.
1125  */
1126 void
1127 pmap_activate(struct lwp *l)
1128 {
1129 #if 0
1130 	struct pcb *pcb = lwp_getpcb(l);
1131 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
1132 
1133 	/*
1134 	 * XXX Normally performed in cpu_lwp_fork().
1135 	 */
1136 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
1137 	pcb->pcb_pm = pmap;
1138 #endif
1139 }
1140 
1141 /*
1142  * Deactivate the specified process's address space.
1143  */
1144 void
1145 pmap_deactivate(struct lwp *l)
1146 {
1147 }
1148 
1149 /*
1150  * Synchronize caches corresponding to [addr, addr+len) in p.
1151  */
1152 void
1153 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
1154 {
1155 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
1156 	int msr, ctx, opid, step;
1157 
1158 	step = CACHELINESIZE;
1159 
1160 	/*
1161 	 * Need to turn off IMMU and switch to user context.
1162 	 * (icbi uses DMMU).
1163 	 */
1164 	if (!(ctx = pm->pm_ctx)) {
1165 		/* No context -- assign it one */
1166 		ctx_alloc(pm);
1167 		ctx = pm->pm_ctx;
1168 	}
1169 	__asm volatile("mfmsr %0;"
1170 		"li %1, %7;"
1171 		"andc %1,%0,%1;"
1172 		"mtmsr %1;"
1173 		"sync;isync;"
1174 		"mfpid %1;"
1175 		"mtpid %2;"
1176 		"sync; isync;"
1177 		"1:"
1178 		"dcbf 0,%3;"
1179 		"icbi 0,%3;"
1180 		"add %3,%3,%5;"
1181 		"addc. %4,%4,%6;"
1182 		"bge 1b;"
1183 		"mtpid %1;"
1184 		"mtmsr %0;"
1185 		"sync; isync"
1186 		: "=&r" (msr), "=&r" (opid)
1187 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step),
1188 		  "K" (PSL_IR | PSL_DR));
1189 }
1190 
1191 
1192 /* This has to be done in real mode !!! */
1193 void
1194 ppc4xx_tlb_flush(vaddr_t va, int pid)
1195 {
1196 	u_long i, found;
1197 	u_long msr;
1198 
1199 	/* If there's no context then it can't be mapped. */
1200 	if (!pid)
1201 		return;
1202 
1203 	__asm( 	"mfpid %1;"		/* Save PID */
1204 		"mfmsr %2;"		/* Save MSR */
1205 		"li %0,0;"		/* Now clear MSR */
1206 		"mtmsr %0;"
1207 		"mtpid %4;"		/* Set PID */
1208 		"sync;"
1209 		"tlbsx. %0,0,%3;"	/* Search TLB */
1210 		"sync;"
1211 		"mtpid %1;"		/* Restore PID */
1212 		"mtmsr %2;"		/* Restore MSR */
1213 		"sync;isync;"
1214 		"li %1,1;"
1215 		"beq 1f;"
1216 		"li %1,0;"
1217 		"1:"
1218 		: "=&r" (i), "=&r" (found), "=&r" (msr)
1219 		: "r" (va), "r" (pid));
1220 	if (found && !TLB_LOCKED(i)) {
1221 
1222 		/* Now flush translation */
1223 		__asm volatile(
1224 			"tlbwe %0,%1,0;"
1225 			"sync;isync;"
1226 			: : "r" (0), "r" (i));
1227 
1228 		tlb_info[i].ti_ctx = 0;
1229 		tlb_info[i].ti_flags = 0;
1230 		tlbnext = i;
1231 		/* Successful flushes */
1232 		tlbflush_ev.ev_count++;
1233 	}
1234 }
1235 
1236 void
1237 ppc4xx_tlb_flush_all(void)
1238 {
1239 	u_long i;
1240 
1241 	for (i = 0; i < NTLB; i++)
1242 		if (!TLB_LOCKED(i)) {
1243 			__asm volatile(
1244 				"tlbwe %0,%1,0;"
1245 				"sync;isync;"
1246 				: : "r" (0), "r" (i));
1247 			tlb_info[i].ti_ctx = 0;
1248 			tlb_info[i].ti_flags = 0;
1249 		}
1250 
1251 	__asm volatile("sync;isync");
1252 }
1253 
1254 /* Find a TLB entry to evict. */
1255 static int
1256 ppc4xx_tlb_find_victim(void)
1257 {
1258 	int flags;
1259 
1260 	for (;;) {
1261 		if (++tlbnext >= NTLB)
1262 			tlbnext = tlb_nreserved;
1263 		flags = tlb_info[tlbnext].ti_flags;
1264 		if (!(flags & TLBF_USED) ||
1265 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
1266 			u_long va, stack = (u_long)&va;
1267 
1268 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
1269 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
1270 			     (flags & TLBF_USED)) {
1271 				/* Kernel stack page */
1272 				flags |= TLBF_USED;
1273 				tlb_info[tlbnext].ti_flags = flags;
1274 			} else {
1275 				/* Found it! */
1276 				return (tlbnext);
1277 			}
1278 		} else {
1279 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
1280 		}
1281 	}
1282 }
1283 
1284 void
1285 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
1286 {
1287 	u_long th, tl, idx;
1288 	tlbpid_t pid;
1289 	u_short msr;
1290 	paddr_t pa;
1291 	int s, sz;
1292 
1293 	tlbenter_ev.ev_count++;
1294 
1295 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
1296 	pa = (pte & TTE_RPN_MASK(sz));
1297 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
1298 	tl = (pte & ~TLB_RPN_MASK) | pa;
1299 	tl |= ppc4xx_tlbflags(va, pa);
1300 
1301 	s = splhigh();
1302 	idx = ppc4xx_tlb_find_victim();
1303 
1304 #ifdef DIAGNOSTIC
1305 	if ((idx < tlb_nreserved) || (idx >= NTLB)) {
1306 		panic("ppc4xx_tlb_enter: replacing entry %ld", idx);
1307 	}
1308 #endif
1309 
1310 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
1311 	tlb_info[idx].ti_ctx = ctx;
1312 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
1313 
1314 	__asm volatile(
1315 		"mfmsr %0;"			/* Save MSR */
1316 		"li %1,0;"
1317 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
1318 		"mtmsr %1;"			/* Clear MSR */
1319 		"mfpid %1;"			/* Save old PID */
1320 		"mtpid %2;"			/* Load translation ctx */
1321 		"sync; isync;"
1322 #ifdef DEBUG
1323 		"andi. %3,%3,63;"
1324 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
1325 #endif
1326 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
1327 		"sync; isync;"
1328 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
1329 		"sync; isync;"
1330 	: "=&r" (msr), "=&r" (pid)
1331 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
1332 	splx(s);
1333 }
1334 
1335 void
1336 ppc4xx_tlb_init(void)
1337 {
1338 	int i;
1339 
1340 	/* Mark reserved TLB entries */
1341 	for (i = 0; i < tlb_nreserved; i++) {
1342 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
1343 		tlb_info[i].ti_ctx = KERNEL_PID;
1344 	}
1345 
1346 	/* Setup security zones */
1347 	/* Z0 - accessible by kernel only if TLB entry permissions allow
1348 	 * Z1,Z2 - access is controlled by TLB entry permissions
1349 	 * Z3 - full access regardless of TLB entry permissions
1350 	 */
1351 
1352 	__asm volatile(
1353 		"mtspr %0,%1;"
1354 		"sync;"
1355 		::  "K"(SPR_ZPR), "r" (0x1b000000));
1356 }
1357 
1358 /*
1359  * ppc4xx_tlb_size_mask:
1360  *
1361  * 	Roundup size to supported page size, return TLBHI mask and real size.
1362  */
1363 static int
1364 ppc4xx_tlb_size_mask(size_t size, int *mask, int *rsiz)
1365 {
1366 	int 			i;
1367 
1368 	for (i = 0; i < __arraycount(tlbsize); i++)
1369 		if (size <= tlbsize[i]) {
1370 			*mask = (i << TLB_SIZE_SHFT);
1371 			*rsiz = tlbsize[i];
1372 			return (0);
1373 		}
1374 	return (EINVAL);
1375 }
1376 
1377 /*
1378  * ppc4xx_tlb_mapiodev:
1379  *
1380  * 	Lookup virtual address of mapping previously entered via
1381  * 	ppc4xx_tlb_reserve. Search TLB directly so that we don't
1382  * 	need to waste extra storage for reserved mappings. Note
1383  * 	that reading TLBHI also sets PID, but all reserved mappings
1384  * 	use KERNEL_PID, so the side effect is nil.
1385  */
1386 void *
1387 ppc4xx_tlb_mapiodev(paddr_t base, psize_t len)
1388 {
1389 	paddr_t 		pa;
1390 	vaddr_t 		va;
1391 	u_int 			lo, hi, sz;
1392 	int 			i;
1393 
1394 	/* tlb_nreserved is only allowed to grow, so this is safe. */
1395 	for (i = 0; i < tlb_nreserved; i++) {
1396 		__asm volatile (
1397 		    "	tlbre %0,%2,1 	\n" 	/* TLBLO */
1398 		    "	tlbre %1,%2,0 	\n" 	/* TLBHI */
1399 		    : "=&r" (lo), "=&r" (hi)
1400 		    : "r" (i));
1401 
1402 		KASSERT(hi & TLB_VALID);
1403 		KASSERT(mfspr(SPR_PID) == KERNEL_PID);
1404 
1405 		pa = (lo & TLB_RPN_MASK);
1406 		if (base < pa)
1407 			continue;
1408 
1409 		sz = tlbsize[(hi & TLB_SIZE_MASK) >> TLB_SIZE_SHFT];
1410 		if ((base + len) > (pa + sz))
1411 			continue;
1412 
1413 		va = (hi & TLB_EPN_MASK) + (base & (sz - 1)); 	/* sz = 2^n */
1414 		return (void *)(va);
1415 	}
1416 
1417 	return (NULL);
1418 }
1419 
1420 /*
1421  * ppc4xx_tlb_reserve:
1422  *
1423  * 	Map physical range to kernel virtual chunk via reserved TLB entry.
1424  */
1425 void
1426 ppc4xx_tlb_reserve(paddr_t pa, vaddr_t va, size_t size, int flags)
1427 {
1428 	u_int 			lo, hi;
1429 	int 			szmask, rsize;
1430 
1431 	/* Called before pmap_bootstrap(), va outside kernel space. */
1432 	KASSERT(va < VM_MIN_KERNEL_ADDRESS || va >= VM_MAX_KERNEL_ADDRESS);
1433 	KASSERT(! pmap_bootstrap_done);
1434 	KASSERT(tlb_nreserved < NTLB);
1435 
1436 	/* Resolve size. */
1437 	if (ppc4xx_tlb_size_mask(size, &szmask, &rsize) != 0)
1438 		panic("ppc4xx_tlb_reserve: entry %d, %zuB too large",
1439 		    size, tlb_nreserved);
1440 
1441 	/* Real size will be power of two >= 1024, so this is OK. */
1442 	pa &= ~(rsize - 1); 	/* RPN */
1443 	va &= ~(rsize - 1); 	/* EPN */
1444 
1445 	lo = pa | TLB_WR | flags;
1446 	hi = va | TLB_VALID | szmask;
1447 
1448 #ifdef PPC_4XX_NOCACHE
1449 	lo |= TLB_I;
1450 #endif
1451 
1452 	__asm volatile(
1453 	    "	tlbwe %1,%0,1 	\n" 	/* write TLBLO */
1454 	    "	tlbwe %2,%0,0 	\n" 	/* write TLBHI */
1455 	    "   sync 		\n"
1456 	    "	isync 		\n"
1457 	    : : "r" (tlb_nreserved), "r" (lo), "r" (hi));
1458 
1459 	tlb_nreserved++;
1460 }
1461 
1462 /*
1463  * We should pass the ctx in from trap code.
1464  */
1465 int
1466 pmap_tlbmiss(vaddr_t va, int ctx)
1467 {
1468 	volatile u_int *pte;
1469 	u_long tte;
1470 
1471 	tlbmiss_ev.ev_count++;
1472 
1473 	/*
1474 	 * We will reserve 0 upto VM_MIN_KERNEL_ADDRESS for va == pa mappings.
1475 	 * Physical RAM is expected to live in this range, care must be taken
1476 	 * to not clobber 0 upto ${physmem} with device mappings in machdep
1477 	 * code.
1478 	 */
1479 	if (ctx != KERNEL_PID ||
1480 	    (va >= VM_MIN_KERNEL_ADDRESS && va < VM_MAX_KERNEL_ADDRESS)) {
1481 		pte = pte_find((struct pmap *)__UNVOLATILE(ctxbusy[ctx]), va);
1482 		if (pte == NULL) {
1483 			/* Map unmanaged addresses directly for kernel access */
1484 			return 1;
1485 		}
1486 		tte = *pte;
1487 		if (tte == 0) {
1488 			return 1;
1489 		}
1490 	} else {
1491 		/* Create a 16MB writable mapping. */
1492 #ifdef PPC_4XX_NOCACHE
1493 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I |TTE_WR;
1494 #else
1495 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
1496 #endif
1497 	}
1498 	tlbhit_ev.ev_count++;
1499 	ppc4xx_tlb_enter(ctx, va, tte);
1500 
1501 	return 0;
1502 }
1503 
1504 /*
1505  * Flush all the entries matching a context from the TLB.
1506  */
1507 static int
1508 ctx_flush(int cnum)
1509 {
1510 	int i;
1511 
1512 	/* We gotta steal this context */
1513 	for (i = tlb_nreserved; i < NTLB; i++) {
1514 		if (tlb_info[i].ti_ctx == cnum) {
1515 			/* Can't steal ctx if it has a locked entry. */
1516 			if (TLB_LOCKED(i)) {
1517 #ifdef DIAGNOSTIC
1518 				printf("ctx_flush: can't invalidate "
1519 					"locked mapping %d "
1520 					"for context %d\n", i, cnum);
1521 #ifdef DDB
1522 				Debugger();
1523 #endif
1524 #endif
1525 				return (1);
1526 			}
1527 #ifdef DIAGNOSTIC
1528 			if (i < tlb_nreserved)
1529 				panic("TLB entry %d not locked", i);
1530 #endif
1531 			/* Invalidate particular TLB entry regardless of locked status */
1532 			__asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
1533 			tlb_info[i].ti_flags = 0;
1534 		}
1535 	}
1536 	return (0);
1537 }
1538 
1539 /*
1540  * Allocate a context.  If necessary, steal one from someone else.
1541  *
1542  * The new context is flushed from the TLB before returning.
1543  */
1544 int
1545 ctx_alloc(struct pmap *pm)
1546 {
1547 	int s, cnum;
1548 	static int next = MINCTX;
1549 
1550 	if (pm == pmap_kernel()) {
1551 #ifdef DIAGNOSTIC
1552 		printf("ctx_alloc: kernel pmap!\n");
1553 #endif
1554 		return (0);
1555 	}
1556 	s = splvm();
1557 
1558 	/* Find a likely context. */
1559 	cnum = next;
1560 	do {
1561 		if ((++cnum) > NUMCTX)
1562 			cnum = MINCTX;
1563 	} while (ctxbusy[cnum] != NULL && cnum != next);
1564 
1565 	/* Now clean it out */
1566 oops:
1567 	if (cnum < MINCTX)
1568 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
1569 	if (ctx_flush(cnum)) {
1570 		/* oops -- something's wired. */
1571 		if ((++cnum) > NUMCTX)
1572 			cnum = MINCTX;
1573 		goto oops;
1574 	}
1575 
1576 	if (ctxbusy[cnum]) {
1577 #ifdef DEBUG
1578 		/* We should identify this pmap and clear it */
1579 		printf("Warning: stealing context %d\n", cnum);
1580 #endif
1581 		ctxbusy[cnum]->pm_ctx = 0;
1582 	}
1583 	ctxbusy[cnum] = pm;
1584 	next = cnum;
1585 	splx(s);
1586 	pm->pm_ctx = cnum;
1587 
1588 	return cnum;
1589 }
1590 
1591 /*
1592  * Give away a context.
1593  */
1594 void
1595 ctx_free(struct pmap *pm)
1596 {
1597 	int oldctx;
1598 
1599 	oldctx = pm->pm_ctx;
1600 
1601 	if (oldctx == 0)
1602 		panic("ctx_free: freeing kernel context");
1603 #ifdef DIAGNOSTIC
1604 	if (ctxbusy[oldctx] == 0)
1605 		printf("ctx_free: freeing free context %d\n", oldctx);
1606 	if (ctxbusy[oldctx] != pm) {
1607 		printf("ctx_free: freeing someone esle's context\n "
1608 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
1609 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
1610 #ifdef DDB
1611 		Debugger();
1612 #endif
1613 	}
1614 #endif
1615 	/* We should verify it has not been stolen and reallocated... */
1616 	ctxbusy[oldctx] = NULL;
1617 	ctx_flush(oldctx);
1618 }
1619 
1620 
1621 #ifdef DEBUG
1622 /*
1623  * Test ref/modify handling.
1624  */
1625 void pmap_testout(void);
1626 void
1627 pmap_testout(void)
1628 {
1629 	vaddr_t va;
1630 	volatile int *loc;
1631 	int val = 0;
1632 	paddr_t pa;
1633 	struct vm_page *pg;
1634 	int ref, mod;
1635 
1636 	/* Allocate a page */
1637 	va = (vaddr_t)uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
1638 	    UVM_KMF_WIRED | UVM_KMF_ZERO);
1639 	loc = (int*)va;
1640 
1641 	pmap_extract(pmap_kernel(), va, &pa);
1642 	pg = PHYS_TO_VM_PAGE(pa);
1643 	pmap_unwire(pmap_kernel(), va);
1644 
1645 	pmap_kremove(va, PAGE_SIZE);
1646 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1647 	pmap_update(pmap_kernel());
1648 
1649 	/* Now clear reference and modify */
1650 	ref = pmap_clear_reference(pg);
1651 	mod = pmap_clear_modify(pg);
1652 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1653 	       (void *)(u_long)va, (long)pa,
1654 	       ref, mod);
1655 
1656 	/* Check it's properly cleared */
1657 	ref = pmap_is_referenced(pg);
1658 	mod = pmap_is_modified(pg);
1659 	printf("Checking cleared page: ref %d, mod %d\n",
1660 	       ref, mod);
1661 
1662 	/* Reference page */
1663 	val = *loc;
1664 
1665 	ref = pmap_is_referenced(pg);
1666 	mod = pmap_is_modified(pg);
1667 	printf("Referenced page: ref %d, mod %d val %x\n",
1668 	       ref, mod, val);
1669 
1670 	/* Now clear reference and modify */
1671 	ref = pmap_clear_reference(pg);
1672 	mod = pmap_clear_modify(pg);
1673 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1674 	       (void *)(u_long)va, (long)pa,
1675 	       ref, mod);
1676 
1677 	/* Modify page */
1678 	*loc = 1;
1679 
1680 	ref = pmap_is_referenced(pg);
1681 	mod = pmap_is_modified(pg);
1682 	printf("Modified page: ref %d, mod %d\n",
1683 	       ref, mod);
1684 
1685 	/* Now clear reference and modify */
1686 	ref = pmap_clear_reference(pg);
1687 	mod = pmap_clear_modify(pg);
1688 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1689 	       (void *)(u_long)va, (long)pa,
1690 	       ref, mod);
1691 
1692 	/* Check it's properly cleared */
1693 	ref = pmap_is_referenced(pg);
1694 	mod = pmap_is_modified(pg);
1695 	printf("Checking cleared page: ref %d, mod %d\n",
1696 	       ref, mod);
1697 
1698 	/* Modify page */
1699 	*loc = 1;
1700 
1701 	ref = pmap_is_referenced(pg);
1702 	mod = pmap_is_modified(pg);
1703 	printf("Modified page: ref %d, mod %d\n",
1704 	       ref, mod);
1705 
1706 	/* Check pmap_protect() */
1707 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
1708 	pmap_update(pmap_kernel());
1709 	ref = pmap_is_referenced(pg);
1710 	mod = pmap_is_modified(pg);
1711 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
1712 	       ref, mod);
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 	/* Reference page */
1722 	val = *loc;
1723 
1724 	ref = pmap_is_referenced(pg);
1725 	mod = pmap_is_modified(pg);
1726 	printf("Referenced page: ref %d, mod %d val %x\n",
1727 	       ref, mod, val);
1728 
1729 	/* Now clear reference and modify */
1730 	ref = pmap_clear_reference(pg);
1731 	mod = pmap_clear_modify(pg);
1732 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1733 	       (void *)(u_long)va, (long)pa,
1734 	       ref, mod);
1735 
1736 	/* Modify page */
1737 #if 0
1738 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1739 	pmap_update(pmap_kernel());
1740 #endif
1741 	*loc = 1;
1742 
1743 	ref = pmap_is_referenced(pg);
1744 	mod = pmap_is_modified(pg);
1745 	printf("Modified page: ref %d, mod %d\n",
1746 	       ref, mod);
1747 
1748 	/* Check pmap_protect() */
1749 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
1750 	pmap_update(pmap_kernel());
1751 	ref = pmap_is_referenced(pg);
1752 	mod = pmap_is_modified(pg);
1753 	printf("pmap_protect(): ref %d, mod %d\n",
1754 	       ref, mod);
1755 
1756 	/* Now clear reference and modify */
1757 	ref = pmap_clear_reference(pg);
1758 	mod = pmap_clear_modify(pg);
1759 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1760 	       (void *)(u_long)va, (long)pa,
1761 	       ref, mod);
1762 
1763 	/* Reference page */
1764 	val = *loc;
1765 
1766 	ref = pmap_is_referenced(pg);
1767 	mod = pmap_is_modified(pg);
1768 	printf("Referenced page: ref %d, mod %d val %x\n",
1769 	       ref, mod, val);
1770 
1771 	/* Now clear reference and modify */
1772 	ref = pmap_clear_reference(pg);
1773 	mod = pmap_clear_modify(pg);
1774 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1775 	       (void *)(u_long)va, (long)pa,
1776 	       ref, mod);
1777 
1778 	/* Modify page */
1779 #if 0
1780 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1781 	pmap_update(pmap_kernel());
1782 #endif
1783 	*loc = 1;
1784 
1785 	ref = pmap_is_referenced(pg);
1786 	mod = pmap_is_modified(pg);
1787 	printf("Modified page: ref %d, mod %d\n",
1788 	       ref, mod);
1789 
1790 	/* Check pmap_pag_protect() */
1791 	pmap_page_protect(pg, VM_PROT_READ);
1792 	ref = pmap_is_referenced(pg);
1793 	mod = pmap_is_modified(pg);
1794 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
1795 	       ref, mod);
1796 
1797 	/* Now clear reference and modify */
1798 	ref = pmap_clear_reference(pg);
1799 	mod = pmap_clear_modify(pg);
1800 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1801 	       (void *)(u_long)va, (long)pa,
1802 	       ref, mod);
1803 
1804 	/* Reference page */
1805 	val = *loc;
1806 
1807 	ref = pmap_is_referenced(pg);
1808 	mod = pmap_is_modified(pg);
1809 	printf("Referenced page: ref %d, mod %d val %x\n",
1810 	       ref, mod, val);
1811 
1812 	/* Now clear reference and modify */
1813 	ref = pmap_clear_reference(pg);
1814 	mod = pmap_clear_modify(pg);
1815 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1816 	       (void *)(u_long)va, (long)pa,
1817 	       ref, mod);
1818 
1819 	/* Modify page */
1820 #if 0
1821 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1822 	pmap_update(pmap_kernel());
1823 #endif
1824 	*loc = 1;
1825 
1826 	ref = pmap_is_referenced(pg);
1827 	mod = pmap_is_modified(pg);
1828 	printf("Modified page: ref %d, mod %d\n",
1829 	       ref, mod);
1830 
1831 	/* Check pmap_pag_protect() */
1832 	pmap_page_protect(pg, VM_PROT_NONE);
1833 	ref = pmap_is_referenced(pg);
1834 	mod = pmap_is_modified(pg);
1835 	printf("pmap_page_protect(): ref %d, mod %d\n",
1836 	       ref, mod);
1837 
1838 	/* Now clear reference and modify */
1839 	ref = pmap_clear_reference(pg);
1840 	mod = pmap_clear_modify(pg);
1841 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1842 	       (void *)(u_long)va, (long)pa,
1843 	       ref, mod);
1844 
1845 
1846 	/* Reference page */
1847 	val = *loc;
1848 
1849 	ref = pmap_is_referenced(pg);
1850 	mod = pmap_is_modified(pg);
1851 	printf("Referenced page: ref %d, mod %d val %x\n",
1852 	       ref, mod, val);
1853 
1854 	/* Now clear reference and modify */
1855 	ref = pmap_clear_reference(pg);
1856 	mod = pmap_clear_modify(pg);
1857 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1858 	       (void *)(u_long)va, (long)pa,
1859 	       ref, mod);
1860 
1861 	/* Modify page */
1862 #if 0
1863 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1864 	pmap_update(pmap_kernel());
1865 #endif
1866 	*loc = 1;
1867 
1868 	ref = pmap_is_referenced(pg);
1869 	mod = pmap_is_modified(pg);
1870 	printf("Modified page: ref %d, mod %d\n",
1871 	       ref, mod);
1872 
1873 	/* Unmap page */
1874 	pmap_remove(pmap_kernel(), va, va+1);
1875 	pmap_update(pmap_kernel());
1876 	ref = pmap_is_referenced(pg);
1877 	mod = pmap_is_modified(pg);
1878 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
1879 
1880 	/* Now clear reference and modify */
1881 	ref = pmap_clear_reference(pg);
1882 	mod = pmap_clear_modify(pg);
1883 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1884 	       (void *)(u_long)va, (long)pa, ref, mod);
1885 
1886 	/* Check it's properly cleared */
1887 	ref = pmap_is_referenced(pg);
1888 	mod = pmap_is_modified(pg);
1889 	printf("Checking cleared page: ref %d, mod %d\n",
1890 	       ref, mod);
1891 
1892 	pmap_remove(pmap_kernel(), va, va + PAGE_SIZE);
1893 	pmap_kenter_pa(va, pa, VM_PROT_ALL, 0);
1894 	uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE, UVM_KMF_WIRED);
1895 }
1896 #endif
1897