xref: /netbsd/sys/arch/powerpc/ibm4xx/pmap.c (revision 6550d01e)
1 /*	$NetBSD: pmap.c,v 1.65 2011/01/14 02:06:29 rmind 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 1.65 2011/01/14 02:06:29 rmind 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 
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 <powerpc/ibm4xx/spr.h>
88 #include <machine/tlb.h>
89 
90 /*
91  * kernmap is an array of PTEs large enough to map in
92  * 4GB.  At 16KB/page it is 256K entries or 2MB.
93  */
94 #define KERNMAP_SIZE	((0xffffffffU/PAGE_SIZE)+1)
95 void *kernmap;
96 
97 #define MINCTX		2
98 #define NUMCTX		256
99 
100 volatile struct pmap *ctxbusy[NUMCTX];
101 
102 #define TLBF_USED	0x1
103 #define	TLBF_REF	0x2
104 #define	TLBF_LOCKED	0x4
105 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
106 
107 typedef struct tlb_info_s {
108 	char	ti_flags;
109 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
110 	u_int	ti_va;
111 } tlb_info_t;
112 
113 volatile tlb_info_t tlb_info[NTLB];
114 /* We'll use a modified FIFO replacement policy cause it's cheap */
115 volatile int tlbnext;
116 
117 static int tlb_nreserved = 0;
118 static int pmap_bootstrap_done = 0;
119 
120 /* Event counters */
121 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
122 	NULL, "cpu", "tlbmiss");
123 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
124 	NULL, "cpu", "tlbhit");
125 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
126 	NULL, "cpu", "tlbflush");
127 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
128 	NULL, "cpu", "tlbenter");
129 
130 struct pmap kernel_pmap_;
131 struct pmap *const kernel_pmap_ptr = &kernel_pmap_;
132 
133 static int npgs;
134 static u_int nextavail;
135 #ifndef MSGBUFADDR
136 extern paddr_t msgbuf_paddr;
137 #endif
138 
139 static struct mem_region *mem, *avail;
140 
141 /*
142  * This is a cache of referenced/modified bits.
143  * Bits herein are shifted by ATTRSHFT.
144  */
145 static char *pmap_attrib;
146 
147 #define PV_WIRED	0x1
148 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
149 #define PV_UNWIRE(pv)	((pv)->pv_va &= ~PV_WIRED)
150 #define PV_ISWIRED(pv)	((pv)->pv_va & PV_WIRED)
151 #define PV_CMPVA(va,pv)	(!(((pv)->pv_va ^ (va)) & (~PV_WIRED)))
152 
153 struct pv_entry {
154 	struct pv_entry *pv_next;	/* Linked list of mappings */
155 	vaddr_t pv_va;			/* virtual address of mapping */
156 	struct pmap *pv_pm;
157 };
158 
159 /* Each index corresponds to TLB_SIZE_* value. */
160 static size_t tlbsize[] = {
161 	1024, 		/* TLB_SIZE_1K */
162 	4096, 		/* TLB_SIZE_4K */
163 	16384, 		/* TLB_SIZE_16K */
164 	65536, 		/* TLB_SIZE_64K */
165 	262144, 	/* TLB_SIZE_256K */
166 	1048576, 	/* TLB_SIZE_1M */
167 	4194304, 	/* TLB_SIZE_4M */
168 	16777216, 	/* TLB_SIZE_16M */
169 };
170 
171 struct pv_entry *pv_table;
172 static struct pool pv_pool;
173 
174 static int pmap_initialized;
175 
176 static int ctx_flush(int);
177 
178 inline struct pv_entry *pa_to_pv(paddr_t);
179 static inline char *pa_to_attr(paddr_t);
180 
181 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
182 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
183 
184 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t, int);
185 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
186 
187 static int ppc4xx_tlb_size_mask(size_t, int *, int *);
188 
189 
190 inline struct pv_entry *
191 pa_to_pv(paddr_t pa)
192 {
193 	int bank, pg;
194 
195 	bank = vm_physseg_find(atop(pa), &pg);
196 	if (bank == -1)
197 		return NULL;
198 	return &VM_PHYSMEM_PTR(bank)->pmseg.pvent[pg];
199 }
200 
201 static inline char *
202 pa_to_attr(paddr_t pa)
203 {
204 	int bank, pg;
205 
206 	bank = vm_physseg_find(atop(pa), &pg);
207 	if (bank == -1)
208 		return NULL;
209 	return &VM_PHYSMEM_PTR(bank)->pmseg.attrs[pg];
210 }
211 
212 /*
213  * Insert PTE into page table.
214  */
215 int
216 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
217 {
218 	int seg = STIDX(va);
219 	int ptn = PTIDX(va);
220 	u_int oldpte;
221 
222 	if (!pm->pm_ptbl[seg]) {
223 		/* Don't allocate a page to clear a non-existent mapping. */
224 		if (!pte)
225 			return (0);
226 		/* Allocate a page XXXX this will sleep! */
227 		pm->pm_ptbl[seg] =
228 		    (uint *)uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
229 		    UVM_KMF_WIRED | UVM_KMF_ZERO);
230 	}
231 	oldpte = pm->pm_ptbl[seg][ptn];
232 	pm->pm_ptbl[seg][ptn] = pte;
233 
234 	/* Flush entry. */
235 	ppc4xx_tlb_flush(va, pm->pm_ctx);
236 	if (oldpte != pte) {
237 		if (pte == 0)
238 			pm->pm_stats.resident_count--;
239 		else
240 			pm->pm_stats.resident_count++;
241 	}
242 	return (1);
243 }
244 
245 /*
246  * Get a pointer to a PTE in a page table.
247  */
248 volatile u_int *
249 pte_find(struct pmap *pm, vaddr_t va)
250 {
251 	int seg = STIDX(va);
252 	int ptn = PTIDX(va);
253 
254 	if (pm->pm_ptbl[seg])
255 		return (&pm->pm_ptbl[seg][ptn]);
256 
257 	return (NULL);
258 }
259 
260 /*
261  * This is called during initppc, before the system is really initialized.
262  */
263 void
264 pmap_bootstrap(u_int kernelstart, u_int kernelend)
265 {
266 	struct mem_region *mp, *mp1;
267 	int cnt, i;
268 	u_int s, e, sz;
269 
270 	tlbnext = tlb_nreserved;
271 
272 	/*
273 	 * Allocate the kernel page table at the end of
274 	 * kernel space so it's in the locked TTE.
275 	 */
276 	kernmap = (void *)kernelend;
277 
278 	/*
279 	 * Initialize kernel page table.
280 	 */
281 	for (i = 0; i < STSZ; i++) {
282 		pmap_kernel()->pm_ptbl[i] = 0;
283 	}
284 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
285 
286 	/*
287 	 * Announce page-size to the VM-system
288 	 */
289 	uvmexp.pagesize = NBPG;
290 	uvm_setpagesize();
291 
292 	/*
293 	 * Get memory.
294 	 */
295 	mem_regions(&mem, &avail);
296 	for (mp = mem; mp->size; mp++) {
297 		physmem += btoc(mp->size);
298 		printf("+%lx,",mp->size);
299 	}
300 	printf("\n");
301 	ppc4xx_tlb_init();
302 	/*
303 	 * Count the number of available entries.
304 	 */
305 	for (cnt = 0, mp = avail; mp->size; mp++)
306 		cnt++;
307 
308 	/*
309 	 * Page align all regions.
310 	 * Non-page aligned memory isn't very interesting to us.
311 	 * Also, sort the entries for ascending addresses.
312 	 */
313 	kernelstart &= ~PGOFSET;
314 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
315 	for (mp = avail; mp->size; mp++) {
316 		s = mp->start;
317 		e = mp->start + mp->size;
318 		printf("%08x-%08x -> ",s,e);
319 		/*
320 		 * Check whether this region holds all of the kernel.
321 		 */
322 		if (s < kernelstart && e > kernelend) {
323 			avail[cnt].start = kernelend;
324 			avail[cnt++].size = e - kernelend;
325 			e = kernelstart;
326 		}
327 		/*
328 		 * Look whether this regions starts within the kernel.
329 		 */
330 		if (s >= kernelstart && s < kernelend) {
331 			if (e <= kernelend)
332 				goto empty;
333 			s = kernelend;
334 		}
335 		/*
336 		 * Now look whether this region ends within the kernel.
337 		 */
338 		if (e > kernelstart && e <= kernelend) {
339 			if (s >= kernelstart)
340 				goto empty;
341 			e = kernelstart;
342 		}
343 		/*
344 		 * Now page align the start and size of the region.
345 		 */
346 		s = round_page(s);
347 		e = trunc_page(e);
348 		if (e < s)
349 			e = s;
350 		sz = e - s;
351 		printf("%08x-%08x = %x\n",s,e,sz);
352 		/*
353 		 * Check whether some memory is left here.
354 		 */
355 		if (sz == 0) {
356 		empty:
357 			memmove(mp, mp + 1,
358 				(cnt - (mp - avail)) * sizeof *mp);
359 			cnt--;
360 			mp--;
361 			continue;
362 		}
363 		/*
364 		 * Do an insertion sort.
365 		 */
366 		npgs += btoc(sz);
367 		for (mp1 = avail; mp1 < mp; mp1++)
368 			if (s < mp1->start)
369 				break;
370 		if (mp1 < mp) {
371 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
372 			mp1->start = s;
373 			mp1->size = sz;
374 		} else {
375 			mp->start = s;
376 			mp->size = sz;
377 		}
378 	}
379 
380 	/*
381 	 * We cannot do pmap_steal_memory here,
382 	 * since we don't run with translation enabled yet.
383 	 */
384 #ifndef MSGBUFADDR
385 	/*
386 	 * allow for msgbuf
387 	 */
388 	sz = round_page(MSGBUFSIZE);
389 	mp = NULL;
390 	for (mp1 = avail; mp1->size; mp1++)
391 		if (mp1->size >= sz)
392 			mp = mp1;
393 	if (mp == NULL)
394 		panic("not enough memory?");
395 
396 	npgs -= btoc(sz);
397 	msgbuf_paddr = mp->start + mp->size - sz;
398 	mp->size -= sz;
399 	if (mp->size <= 0)
400 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
401 #endif
402 
403 	for (mp = avail; mp->size; mp++)
404 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
405 			atop(mp->start), atop(mp->start + mp->size),
406 			VM_FREELIST_DEFAULT);
407 
408 	/*
409 	 * Initialize kernel pmap and hardware.
410 	 */
411 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
412 	pmap_kernel()->pm_ctx = KERNEL_PID;
413 	nextavail = avail->start;
414 
415 	evcnt_attach_static(&tlbmiss_ev);
416 	evcnt_attach_static(&tlbhit_ev);
417 	evcnt_attach_static(&tlbflush_ev);
418 	evcnt_attach_static(&tlbenter_ev);
419 
420 	pmap_bootstrap_done = 1;
421 }
422 
423 /*
424  * Restrict given range to physical memory
425  *
426  * (Used by /dev/mem)
427  */
428 void
429 pmap_real_memory(paddr_t *start, psize_t *size)
430 {
431 	struct mem_region *mp;
432 
433 	for (mp = mem; mp->size; mp++) {
434 		if (*start + *size > mp->start &&
435 		    *start < mp->start + mp->size) {
436 			if (*start < mp->start) {
437 				*size -= mp->start - *start;
438 				*start = mp->start;
439 			}
440 			if (*start + *size > mp->start + mp->size)
441 				*size = mp->start + mp->size - *start;
442 			return;
443 		}
444 	}
445 	*size = 0;
446 }
447 
448 /*
449  * Initialize anything else for pmap handling.
450  * Called during vm_init().
451  */
452 void
453 pmap_init(void)
454 {
455 	struct pv_entry *pv;
456 	vsize_t sz;
457 	vaddr_t addr;
458 	int i, s;
459 	int bank;
460 	char *attr;
461 
462 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
463 	sz = round_page(sz);
464 	addr = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED | UVM_KMF_ZERO);
465 	s = splvm();
466 	pv = pv_table = (struct pv_entry *)addr;
467 	for (i = npgs; --i >= 0;)
468 		pv++->pv_pm = NULL;
469 	pmap_attrib = (char *)pv;
470 	memset(pv, 0, npgs);
471 
472 	pv = pv_table;
473 	attr = pmap_attrib;
474 	for (bank = 0; bank < vm_nphysseg; bank++) {
475 		sz = VM_PHYSMEM_PTR(bank)->end - VM_PHYSMEM_PTR(bank)->start;
476 		VM_PHYSMEM_PTR(bank)->pmseg.pvent = pv;
477 		VM_PHYSMEM_PTR(bank)->pmseg.attrs = attr;
478 		pv += sz;
479 		attr += sz;
480 	}
481 
482 	pmap_initialized = 1;
483 	splx(s);
484 
485 	/* Setup a pool for additional pvlist structures */
486 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL,
487 	    IPL_VM);
488 }
489 
490 /*
491  * How much virtual space is available to the kernel?
492  */
493 void
494 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
495 {
496 
497 #if 0
498 	/*
499 	 * Reserve one segment for kernel virtual memory
500 	 */
501 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
502 	*end = *start + SEGMENT_LENGTH;
503 #else
504 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
505 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
506 #endif
507 }
508 
509 #ifdef PMAP_GROWKERNEL
510 /*
511  * Preallocate kernel page tables to a specified VA.
512  * This simply loops through the first TTE for each
513  * page table from the beginning of the kernel pmap,
514  * reads the entry, and if the result is
515  * zero (either invalid entry or no page table) it stores
516  * a zero there, populating page tables in the process.
517  * This is not the most efficient technique but i don't
518  * expect it to be called that often.
519  */
520 extern struct vm_page *vm_page_alloc1(void);
521 extern void vm_page_free1(struct vm_page *);
522 
523 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
524 
525 vaddr_t
526 pmap_growkernel(vaddr_t maxkvaddr)
527 {
528 	int s;
529 	int seg;
530 	paddr_t pg;
531 	struct pmap *pm = pmap_kernel();
532 
533 	s = splvm();
534 
535 	/* Align with the start of a page table */
536 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
537 	     kbreak += PTMAP) {
538 		seg = STIDX(kbreak);
539 
540 		if (pte_find(pm, kbreak))
541 			continue;
542 
543 		if (uvm.page_init_done) {
544 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
545 		} else {
546 			if (!uvm_page_physget(&pg))
547 				panic("pmap_growkernel: no memory");
548 		}
549 		if (!pg)
550 			panic("pmap_growkernel: no pages");
551 		pmap_zero_page((paddr_t)pg);
552 
553 		/* XXX This is based on all phymem being addressable */
554 		pm->pm_ptbl[seg] = (u_int *)pg;
555 	}
556 	splx(s);
557 	return (kbreak);
558 }
559 
560 /*
561  *	vm_page_alloc1:
562  *
563  *	Allocate and return a memory cell with no associated object.
564  */
565 struct vm_page *
566 vm_page_alloc1(void)
567 {
568 	struct vm_page *pg;
569 
570 	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
571 	if (pg) {
572 		pg->wire_count = 1;	/* no mappings yet */
573 		pg->flags &= ~PG_BUSY;	/* never busy */
574 	}
575 	return pg;
576 }
577 
578 /*
579  *	vm_page_free1:
580  *
581  *	Returns the given page to the free list,
582  *	disassociating it with any VM object.
583  *
584  *	Object and page must be locked prior to entry.
585  */
586 void
587 vm_page_free1(struct vm_page *pg)
588 {
589 #ifdef DIAGNOSTIC
590 	if (pg->flags != (PG_CLEAN|PG_FAKE)) {
591 		printf("Freeing invalid page %p\n", pg);
592 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(pg));
593 #ifdef DDB
594 		Debugger();
595 #endif
596 		return;
597 	}
598 #endif
599 	pg->flags |= PG_BUSY;
600 	pg->wire_count = 0;
601 	uvm_pagefree(pg);
602 }
603 #endif
604 
605 /*
606  * Create and return a physical map.
607  */
608 struct pmap *
609 pmap_create(void)
610 {
611 	struct pmap *pm;
612 
613 	pm = malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
614 	memset(pm, 0, sizeof *pm);
615 	pm->pm_refs = 1;
616 	return pm;
617 }
618 
619 /*
620  * Add a reference to the given pmap.
621  */
622 void
623 pmap_reference(struct pmap *pm)
624 {
625 
626 	pm->pm_refs++;
627 }
628 
629 /*
630  * Retire the given pmap from service.
631  * Should only be called if the map contains no valid mappings.
632  */
633 void
634 pmap_destroy(struct pmap *pm)
635 {
636 	int i;
637 
638 	if (--pm->pm_refs > 0) {
639 		return;
640 	}
641 	KASSERT(pm->pm_stats.resident_count == 0);
642 	KASSERT(pm->pm_stats.wired_count == 0);
643 	for (i = 0; i < STSZ; i++)
644 		if (pm->pm_ptbl[i]) {
645 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i],
646 			    PAGE_SIZE, UVM_KMF_WIRED);
647 			pm->pm_ptbl[i] = NULL;
648 		}
649 	if (pm->pm_ctx)
650 		ctx_free(pm);
651 	free(pm, M_VMPMAP);
652 }
653 
654 /*
655  * Copy the range specified by src_addr/len
656  * from the source map to the range dst_addr/len
657  * in the destination map.
658  *
659  * This routine is only advisory and need not do anything.
660  */
661 void
662 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
663 	  vsize_t len, vaddr_t src_addr)
664 {
665 }
666 
667 /*
668  * Require that all active physical maps contain no
669  * incorrect entries NOW.
670  */
671 void
672 pmap_update(struct pmap *pmap)
673 {
674 }
675 
676 /*
677  * Fill the given physical page with zeroes.
678  */
679 void
680 pmap_zero_page(paddr_t pa)
681 {
682 
683 #ifdef PPC_4XX_NOCACHE
684 	memset((void *)pa, 0, PAGE_SIZE);
685 #else
686 	int i;
687 
688 	for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
689 		__asm volatile ("dcbz 0,%0" :: "r"(pa));
690 		pa += CACHELINESIZE;
691 	}
692 #endif
693 }
694 
695 /*
696  * Copy the given physical source page to its destination.
697  */
698 void
699 pmap_copy_page(paddr_t src, paddr_t dst)
700 {
701 
702 	memcpy((void *)dst, (void *)src, PAGE_SIZE);
703 	dcache_flush_page(dst);
704 }
705 
706 /*
707  * This returns != 0 on success.
708  */
709 static inline int
710 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa, int flags)
711 {
712 	struct pv_entry *pv, *npv = NULL;
713 	int s;
714 
715 	if (!pmap_initialized)
716 		return 0;
717 
718 	s = splvm();
719 	pv = pa_to_pv(pa);
720 	if (!pv->pv_pm) {
721 		/*
722 		 * No entries yet, use header as the first entry.
723 		 */
724 		pv->pv_va = va;
725 		pv->pv_pm = pm;
726 		pv->pv_next = NULL;
727 	} else {
728 		/*
729 		 * There is at least one other VA mapping this page.
730 		 * Place this entry after the header.
731 		 */
732 		npv = pool_get(&pv_pool, PR_NOWAIT);
733 		if (npv == NULL) {
734 			if ((flags & PMAP_CANFAIL) == 0)
735 				panic("pmap_enter_pv: failed");
736 			splx(s);
737 			return 0;
738 		}
739 		npv->pv_va = va;
740 		npv->pv_pm = pm;
741 		npv->pv_next = pv->pv_next;
742 		pv->pv_next = npv;
743 		pv = npv;
744 	}
745 	if (flags & PMAP_WIRED) {
746 		PV_WIRE(pv);
747 		pm->pm_stats.wired_count++;
748 	}
749 	splx(s);
750 	return (1);
751 }
752 
753 static void
754 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
755 {
756 	struct pv_entry *pv, *npv;
757 
758 	/*
759 	 * Remove from the PV table.
760 	 */
761 	pv = pa_to_pv(pa);
762 	if (!pv)
763 		return;
764 
765 	/*
766 	 * If it is the first entry on the list, it is actually
767 	 * in the header and we must copy the following entry up
768 	 * to the header.  Otherwise we must search the list for
769 	 * the entry.  In either case we free the now unused entry.
770 	 */
771 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
772 		if (PV_ISWIRED(pv)) {
773 			pm->pm_stats.wired_count--;
774 		}
775 		if ((npv = pv->pv_next)) {
776 			*pv = *npv;
777 			pool_put(&pv_pool, npv);
778 		} else
779 			pv->pv_pm = NULL;
780 	} else {
781 		for (; (npv = pv->pv_next) != NULL; pv = npv)
782 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
783 				break;
784 		if (npv) {
785 			pv->pv_next = npv->pv_next;
786 			if (PV_ISWIRED(npv)) {
787 				pm->pm_stats.wired_count--;
788 			}
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, u_int flags)
799 {
800 	int s;
801 	u_int tte;
802 	bool managed;
803 
804 	/*
805 	 * Have to remove any existing mapping first.
806 	 */
807 	pmap_remove(pm, va, va + PAGE_SIZE);
808 
809 	if (flags & PMAP_WIRED)
810 		flags |= prot;
811 
812 	managed = uvm_pageismanaged(pa);
813 
814 	/*
815 	 * Generate TTE.
816 	 */
817 	tte = TTE_PA(pa);
818 	/* XXXX -- need to support multiple page sizes. */
819 	tte |= TTE_SZ_16K;
820 #ifdef	DIAGNOSTIC
821 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
822 		(PME_NOCACHE | PME_WRITETHROUG))
823 		panic("pmap_enter: uncached & writethrough");
824 #endif
825 	if (flags & PME_NOCACHE)
826 		/* Must be I/O mapping */
827 		tte |= TTE_I | TTE_G;
828 #ifdef PPC_4XX_NOCACHE
829 	tte |= TTE_I;
830 #else
831 	else if (flags & PME_WRITETHROUG)
832 		/* Uncached and writethrough are not compatible */
833 		tte |= TTE_W;
834 #endif
835 	if (pm == pmap_kernel())
836 		tte |= TTE_ZONE(ZONE_PRIV);
837 	else
838 		tte |= TTE_ZONE(ZONE_USER);
839 
840 	if (flags & VM_PROT_WRITE)
841 		tte |= TTE_WR;
842 
843 	if (flags & VM_PROT_EXECUTE)
844 		tte |= TTE_EX;
845 
846 	/*
847 	 * Now record mapping for later back-translation.
848 	 */
849 	if (pmap_initialized && managed) {
850 		char *attr;
851 
852 		if (!pmap_enter_pv(pm, va, pa, flags)) {
853 			/* Could not enter pv on a managed page */
854 			return 1;
855 		}
856 
857 		/* Now set attributes. */
858 		attr = pa_to_attr(pa);
859 #ifdef DIAGNOSTIC
860 		if (!attr)
861 			panic("managed but no attr");
862 #endif
863 		if (flags & VM_PROT_ALL)
864 			*attr |= PMAP_ATTR_REF;
865 		if (flags & VM_PROT_WRITE)
866 			*attr |= PMAP_ATTR_CHG;
867 	}
868 
869 	s = splvm();
870 
871 	/* Insert page into page table. */
872 	pte_enter(pm, va, tte);
873 
874 	/* If this is a real fault, enter it in the tlb */
875 	if (tte && ((flags & PMAP_WIRED) == 0)) {
876 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
877 	}
878 	splx(s);
879 
880 	/* Flush the real memory from the instruction cache. */
881 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
882 		__syncicache((void *)pa, PAGE_SIZE);
883 
884 	return 0;
885 }
886 
887 void
888 pmap_unwire(struct pmap *pm, vaddr_t va)
889 {
890 	struct pv_entry *pv;
891 	paddr_t pa;
892 	int s;
893 
894 	if (!pmap_extract(pm, va, &pa)) {
895 		return;
896 	}
897 
898 	pv = pa_to_pv(pa);
899 	if (!pv)
900 		return;
901 
902 	s = splvm();
903 	while (pv != NULL) {
904 		if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
905 			if (PV_ISWIRED(pv)) {
906 				PV_UNWIRE(pv);
907 				pm->pm_stats.wired_count--;
908 			}
909 			break;
910 		}
911 		pv = pv->pv_next;
912 	}
913 	splx(s);
914 }
915 
916 void
917 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
918 {
919 	int s;
920 	u_int tte;
921 	struct pmap *pm = pmap_kernel();
922 
923 	/*
924 	 * Have to remove any existing mapping first.
925 	 */
926 
927 	/*
928 	 * Generate TTE.
929 	 *
930 	 * XXXX
931 	 *
932 	 * Since the kernel does not handle execution privileges properly,
933 	 * we will handle read and execute permissions together.
934 	 */
935 	tte = 0;
936 	if (prot & VM_PROT_ALL) {
937 
938 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
939 		/* XXXX -- need to support multiple page sizes. */
940 		tte |= TTE_SZ_16K;
941 #ifdef DIAGNOSTIC
942 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
943 			(PME_NOCACHE | PME_WRITETHROUG))
944 			panic("pmap_kenter_pa: uncached & writethrough");
945 #endif
946 		if (prot & PME_NOCACHE)
947 			/* Must be I/O mapping */
948 			tte |= TTE_I | TTE_G;
949 #ifdef PPC_4XX_NOCACHE
950 		tte |= TTE_I;
951 #else
952 		else if (prot & PME_WRITETHROUG)
953 			/* Uncached and writethrough are not compatible */
954 			tte |= TTE_W;
955 #endif
956 		if (prot & VM_PROT_WRITE)
957 			tte |= TTE_WR;
958 	}
959 
960 	s = splvm();
961 
962 	/* Insert page into page table. */
963 	pte_enter(pm, va, tte);
964 	splx(s);
965 }
966 
967 void
968 pmap_kremove(vaddr_t va, vsize_t len)
969 {
970 
971 	while (len > 0) {
972 		pte_enter(pmap_kernel(), va, 0);
973 		va += PAGE_SIZE;
974 		len -= PAGE_SIZE;
975 	}
976 }
977 
978 /*
979  * Remove the given range of mapping entries.
980  */
981 void
982 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
983 {
984 	int s;
985 	paddr_t pa;
986 	volatile u_int *ptp;
987 
988 	s = splvm();
989 	while (va < endva) {
990 
991 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
992 			pa = TTE_PA(pa);
993 			pmap_remove_pv(pm, va, pa);
994 			*ptp = 0;
995 			ppc4xx_tlb_flush(va, pm->pm_ctx);
996 			pm->pm_stats.resident_count--;
997 		}
998 		va += PAGE_SIZE;
999 	}
1000 
1001 	splx(s);
1002 }
1003 
1004 /*
1005  * Get the physical page address for the given pmap/virtual address.
1006  */
1007 bool
1008 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
1009 {
1010 	int seg = STIDX(va);
1011 	int ptn = PTIDX(va);
1012 	u_int pa = 0;
1013 	int s;
1014 
1015 	s = splvm();
1016 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
1017 		*pap = TTE_PA(pa) | (va & PGOFSET);
1018 	}
1019 	splx(s);
1020 	return (pa != 0);
1021 }
1022 
1023 /*
1024  * Lower the protection on the specified range of this pmap.
1025  *
1026  * There are only two cases: either the protection is going to 0,
1027  * or it is going to read-only.
1028  */
1029 void
1030 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1031 {
1032 	volatile u_int *ptp;
1033 	int s, bic;
1034 
1035 	if ((prot & VM_PROT_READ) == 0) {
1036 		pmap_remove(pm, sva, eva);
1037 		return;
1038 	}
1039 	bic = 0;
1040 	if ((prot & VM_PROT_WRITE) == 0) {
1041 		bic |= TTE_WR;
1042 	}
1043 	if ((prot & VM_PROT_EXECUTE) == 0) {
1044 		bic |= TTE_EX;
1045 	}
1046 	if (bic == 0) {
1047 		return;
1048 	}
1049 	s = splvm();
1050 	while (sva < eva) {
1051 		if ((ptp = pte_find(pm, sva)) != NULL) {
1052 			*ptp &= ~bic;
1053 			ppc4xx_tlb_flush(sva, pm->pm_ctx);
1054 		}
1055 		sva += PAGE_SIZE;
1056 	}
1057 	splx(s);
1058 }
1059 
1060 bool
1061 pmap_check_attr(struct vm_page *pg, u_int mask, int clear)
1062 {
1063 	paddr_t pa;
1064 	char *attr;
1065 	int s, rv;
1066 
1067 	/*
1068 	 * First modify bits in cache.
1069 	 */
1070 	pa = VM_PAGE_TO_PHYS(pg);
1071 	attr = pa_to_attr(pa);
1072 	if (attr == NULL)
1073 		return false;
1074 
1075 	s = splvm();
1076 	rv = ((*attr & mask) != 0);
1077 	if (clear) {
1078 		*attr &= ~mask;
1079 		pmap_page_protect(pg, mask == PMAP_ATTR_CHG ? VM_PROT_READ : 0);
1080 	}
1081 	splx(s);
1082 	return rv;
1083 }
1084 
1085 
1086 /*
1087  * Lower the protection on the specified physical page.
1088  *
1089  * There are only two cases: either the protection is going to 0,
1090  * or it is going to read-only.
1091  */
1092 void
1093 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1094 {
1095 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
1096 	vaddr_t va;
1097 	struct pv_entry *pvh, *pv, *npv;
1098 	struct pmap *pm;
1099 
1100 	pvh = pa_to_pv(pa);
1101 	if (pvh == NULL)
1102 		return;
1103 
1104 	/* Handle extra pvs which may be deleted in the operation */
1105 	for (pv = pvh->pv_next; pv; pv = npv) {
1106 		npv = pv->pv_next;
1107 
1108 		pm = pv->pv_pm;
1109 		va = pv->pv_va;
1110 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
1111 	}
1112 	/* Now check the head pv */
1113 	if (pvh->pv_pm) {
1114 		pv = pvh;
1115 		pm = pv->pv_pm;
1116 		va = pv->pv_va;
1117 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
1118 	}
1119 }
1120 
1121 /*
1122  * Activate the address space for the specified process.  If the process
1123  * is the current process, load the new MMU context.
1124  */
1125 void
1126 pmap_activate(struct lwp *l)
1127 {
1128 #if 0
1129 	struct pcb *pcb = lwp_getpcb(l);
1130 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
1131 
1132 	/*
1133 	 * XXX Normally performed in cpu_lwp_fork().
1134 	 */
1135 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
1136 	pcb->pcb_pm = pmap;
1137 #endif
1138 }
1139 
1140 /*
1141  * Deactivate the specified process's address space.
1142  */
1143 void
1144 pmap_deactivate(struct lwp *l)
1145 {
1146 }
1147 
1148 /*
1149  * Synchronize caches corresponding to [addr, addr+len) in p.
1150  */
1151 void
1152 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
1153 {
1154 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
1155 	int msr, ctx, opid, step;
1156 
1157 	step = CACHELINESIZE;
1158 
1159 	/*
1160 	 * Need to turn off IMMU and switch to user context.
1161 	 * (icbi uses DMMU).
1162 	 */
1163 	if (!(ctx = pm->pm_ctx)) {
1164 		/* No context -- assign it one */
1165 		ctx_alloc(pm);
1166 		ctx = pm->pm_ctx;
1167 	}
1168 	__asm volatile("mfmsr %0;"
1169 		"li %1, %7;"
1170 		"andc %1,%0,%1;"
1171 		"mtmsr %1;"
1172 		"sync;isync;"
1173 		"mfpid %1;"
1174 		"mtpid %2;"
1175 		"sync; isync;"
1176 		"1:"
1177 		"dcbf 0,%3;"
1178 		"icbi 0,%3;"
1179 		"add %3,%3,%5;"
1180 		"addc. %4,%4,%6;"
1181 		"bge 1b;"
1182 		"mtpid %1;"
1183 		"mtmsr %0;"
1184 		"sync; isync"
1185 		: "=&r" (msr), "=&r" (opid)
1186 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step),
1187 		  "K" (PSL_IR | PSL_DR));
1188 }
1189 
1190 
1191 /* This has to be done in real mode !!! */
1192 void
1193 ppc4xx_tlb_flush(vaddr_t va, int pid)
1194 {
1195 	u_long i, found;
1196 	u_long msr;
1197 
1198 	/* If there's no context then it can't be mapped. */
1199 	if (!pid)
1200 		return;
1201 
1202 	__asm( 	"mfpid %1;"		/* Save PID */
1203 		"mfmsr %2;"		/* Save MSR */
1204 		"li %0,0;"		/* Now clear MSR */
1205 		"mtmsr %0;"
1206 		"mtpid %4;"		/* Set PID */
1207 		"sync;"
1208 		"tlbsx. %0,0,%3;"	/* Search TLB */
1209 		"sync;"
1210 		"mtpid %1;"		/* Restore PID */
1211 		"mtmsr %2;"		/* Restore MSR */
1212 		"sync;isync;"
1213 		"li %1,1;"
1214 		"beq 1f;"
1215 		"li %1,0;"
1216 		"1:"
1217 		: "=&r" (i), "=&r" (found), "=&r" (msr)
1218 		: "r" (va), "r" (pid));
1219 	if (found && !TLB_LOCKED(i)) {
1220 
1221 		/* Now flush translation */
1222 		__asm volatile(
1223 			"tlbwe %0,%1,0;"
1224 			"sync;isync;"
1225 			: : "r" (0), "r" (i));
1226 
1227 		tlb_info[i].ti_ctx = 0;
1228 		tlb_info[i].ti_flags = 0;
1229 		tlbnext = i;
1230 		/* Successful flushes */
1231 		tlbflush_ev.ev_count++;
1232 	}
1233 }
1234 
1235 void
1236 ppc4xx_tlb_flush_all(void)
1237 {
1238 	u_long i;
1239 
1240 	for (i = 0; i < NTLB; i++)
1241 		if (!TLB_LOCKED(i)) {
1242 			__asm volatile(
1243 				"tlbwe %0,%1,0;"
1244 				"sync;isync;"
1245 				: : "r" (0), "r" (i));
1246 			tlb_info[i].ti_ctx = 0;
1247 			tlb_info[i].ti_flags = 0;
1248 		}
1249 
1250 	__asm volatile("sync;isync");
1251 }
1252 
1253 /* Find a TLB entry to evict. */
1254 static int
1255 ppc4xx_tlb_find_victim(void)
1256 {
1257 	int flags;
1258 
1259 	for (;;) {
1260 		if (++tlbnext >= NTLB)
1261 			tlbnext = tlb_nreserved;
1262 		flags = tlb_info[tlbnext].ti_flags;
1263 		if (!(flags & TLBF_USED) ||
1264 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
1265 			u_long va, stack = (u_long)&va;
1266 
1267 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
1268 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
1269 			     (flags & TLBF_USED)) {
1270 				/* Kernel stack page */
1271 				flags |= TLBF_USED;
1272 				tlb_info[tlbnext].ti_flags = flags;
1273 			} else {
1274 				/* Found it! */
1275 				return (tlbnext);
1276 			}
1277 		} else {
1278 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
1279 		}
1280 	}
1281 }
1282 
1283 void
1284 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
1285 {
1286 	u_long th, tl, idx;
1287 	tlbpid_t pid;
1288 	u_short msr;
1289 	paddr_t pa;
1290 	int s, sz;
1291 
1292 	tlbenter_ev.ev_count++;
1293 
1294 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
1295 	pa = (pte & TTE_RPN_MASK(sz));
1296 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
1297 	tl = (pte & ~TLB_RPN_MASK) | pa;
1298 	tl |= ppc4xx_tlbflags(va, pa);
1299 
1300 	s = splhigh();
1301 	idx = ppc4xx_tlb_find_victim();
1302 
1303 #ifdef DIAGNOSTIC
1304 	if ((idx < tlb_nreserved) || (idx >= NTLB)) {
1305 		panic("ppc4xx_tlb_enter: replacing entry %ld", idx);
1306 	}
1307 #endif
1308 
1309 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
1310 	tlb_info[idx].ti_ctx = ctx;
1311 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
1312 
1313 	__asm volatile(
1314 		"mfmsr %0;"			/* Save MSR */
1315 		"li %1,0;"
1316 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
1317 		"mtmsr %1;"			/* Clear MSR */
1318 		"mfpid %1;"			/* Save old PID */
1319 		"mtpid %2;"			/* Load translation ctx */
1320 		"sync; isync;"
1321 #ifdef DEBUG
1322 		"andi. %3,%3,63;"
1323 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
1324 #endif
1325 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
1326 		"sync; isync;"
1327 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
1328 		"sync; isync;"
1329 	: "=&r" (msr), "=&r" (pid)
1330 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
1331 	splx(s);
1332 }
1333 
1334 void
1335 ppc4xx_tlb_init(void)
1336 {
1337 	int i;
1338 
1339 	/* Mark reserved TLB entries */
1340 	for (i = 0; i < tlb_nreserved; i++) {
1341 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
1342 		tlb_info[i].ti_ctx = KERNEL_PID;
1343 	}
1344 
1345 	/* Setup security zones */
1346 	/* Z0 - accessible by kernel only if TLB entry permissions allow
1347 	 * Z1,Z2 - access is controlled by TLB entry permissions
1348 	 * Z3 - full access regardless of TLB entry permissions
1349 	 */
1350 
1351 	__asm volatile(
1352 		"mtspr %0,%1;"
1353 		"sync;"
1354 		::  "K"(SPR_ZPR), "r" (0x1b000000));
1355 }
1356 
1357 /*
1358  * ppc4xx_tlb_size_mask:
1359  *
1360  * 	Roundup size to supported page size, return TLBHI mask and real size.
1361  */
1362 static int
1363 ppc4xx_tlb_size_mask(size_t size, int *mask, int *rsiz)
1364 {
1365 	int 			i;
1366 
1367 	for (i = 0; i < __arraycount(tlbsize); i++)
1368 		if (size <= tlbsize[i]) {
1369 			*mask = (i << TLB_SIZE_SHFT);
1370 			*rsiz = tlbsize[i];
1371 			return (0);
1372 		}
1373 	return (EINVAL);
1374 }
1375 
1376 /*
1377  * ppc4xx_tlb_mapiodev:
1378  *
1379  * 	Lookup virtual address of mapping previously entered via
1380  * 	ppc4xx_tlb_reserve. Search TLB directly so that we don't
1381  * 	need to waste extra storage for reserved mappings. Note
1382  * 	that reading TLBHI also sets PID, but all reserved mappings
1383  * 	use KERNEL_PID, so the side effect is nil.
1384  */
1385 void *
1386 ppc4xx_tlb_mapiodev(paddr_t base, psize_t len)
1387 {
1388 	paddr_t 		pa;
1389 	vaddr_t 		va;
1390 	u_int 			lo, hi, sz;
1391 	int 			i;
1392 
1393 	/* tlb_nreserved is only allowed to grow, so this is safe. */
1394 	for (i = 0; i < tlb_nreserved; i++) {
1395 		__asm volatile (
1396 		    "	tlbre %0,%2,1 	\n" 	/* TLBLO */
1397 		    "	tlbre %1,%2,0 	\n" 	/* TLBHI */
1398 		    : "=&r" (lo), "=&r" (hi)
1399 		    : "r" (i));
1400 
1401 		KASSERT(hi & TLB_VALID);
1402 		KASSERT(mfspr(SPR_PID) == KERNEL_PID);
1403 
1404 		pa = (lo & TLB_RPN_MASK);
1405 		if (base < pa)
1406 			continue;
1407 
1408 		sz = tlbsize[(hi & TLB_SIZE_MASK) >> TLB_SIZE_SHFT];
1409 		if ((base + len) > (pa + sz))
1410 			continue;
1411 
1412 		va = (hi & TLB_EPN_MASK) + (base & (sz - 1)); 	/* sz = 2^n */
1413 		return (void *)(va);
1414 	}
1415 
1416 	return (NULL);
1417 }
1418 
1419 /*
1420  * ppc4xx_tlb_reserve:
1421  *
1422  * 	Map physical range to kernel virtual chunk via reserved TLB entry.
1423  */
1424 void
1425 ppc4xx_tlb_reserve(paddr_t pa, vaddr_t va, size_t size, int flags)
1426 {
1427 	u_int 			lo, hi;
1428 	int 			szmask, rsize;
1429 
1430 	/* Called before pmap_bootstrap(), va outside kernel space. */
1431 	KASSERT(va < VM_MIN_KERNEL_ADDRESS || va >= VM_MAX_KERNEL_ADDRESS);
1432 	KASSERT(! pmap_bootstrap_done);
1433 	KASSERT(tlb_nreserved < NTLB);
1434 
1435 	/* Resolve size. */
1436 	if (ppc4xx_tlb_size_mask(size, &szmask, &rsize) != 0)
1437 		panic("ppc4xx_tlb_reserve: entry %d, %zuB too large",
1438 		    size, tlb_nreserved);
1439 
1440 	/* Real size will be power of two >= 1024, so this is OK. */
1441 	pa &= ~(rsize - 1); 	/* RPN */
1442 	va &= ~(rsize - 1); 	/* EPN */
1443 
1444 	lo = pa | TLB_WR | flags;
1445 	hi = va | TLB_VALID | szmask;
1446 
1447 #ifdef PPC_4XX_NOCACHE
1448 	lo |= TLB_I;
1449 #endif
1450 
1451 	__asm volatile(
1452 	    "	tlbwe %1,%0,1 	\n" 	/* write TLBLO */
1453 	    "	tlbwe %2,%0,0 	\n" 	/* write TLBHI */
1454 	    "   sync 		\n"
1455 	    "	isync 		\n"
1456 	    : : "r" (tlb_nreserved), "r" (lo), "r" (hi));
1457 
1458 	tlb_nreserved++;
1459 }
1460 
1461 /*
1462  * We should pass the ctx in from trap code.
1463  */
1464 int
1465 pmap_tlbmiss(vaddr_t va, int ctx)
1466 {
1467 	volatile u_int *pte;
1468 	u_long tte;
1469 
1470 	tlbmiss_ev.ev_count++;
1471 
1472 	/*
1473 	 * We will reserve 0 upto VM_MIN_KERNEL_ADDRESS for va == pa mappings.
1474 	 * Physical RAM is expected to live in this range, care must be taken
1475 	 * to not clobber 0 upto ${physmem} with device mappings in machdep
1476 	 * code.
1477 	 */
1478 	if (ctx != KERNEL_PID ||
1479 	    (va >= VM_MIN_KERNEL_ADDRESS && va < VM_MAX_KERNEL_ADDRESS)) {
1480 		pte = pte_find((struct pmap *)__UNVOLATILE(ctxbusy[ctx]), va);
1481 		if (pte == NULL) {
1482 			/* Map unmanaged addresses directly for kernel access */
1483 			return 1;
1484 		}
1485 		tte = *pte;
1486 		if (tte == 0) {
1487 			return 1;
1488 		}
1489 	} else {
1490 		/* Create a 16MB writable mapping. */
1491 #ifdef PPC_4XX_NOCACHE
1492 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I |TTE_WR;
1493 #else
1494 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
1495 #endif
1496 	}
1497 	tlbhit_ev.ev_count++;
1498 	ppc4xx_tlb_enter(ctx, va, tte);
1499 
1500 	return 0;
1501 }
1502 
1503 /*
1504  * Flush all the entries matching a context from the TLB.
1505  */
1506 static int
1507 ctx_flush(int cnum)
1508 {
1509 	int i;
1510 
1511 	/* We gotta steal this context */
1512 	for (i = tlb_nreserved; i < NTLB; i++) {
1513 		if (tlb_info[i].ti_ctx == cnum) {
1514 			/* Can't steal ctx if it has a locked entry. */
1515 			if (TLB_LOCKED(i)) {
1516 #ifdef DIAGNOSTIC
1517 				printf("ctx_flush: can't invalidate "
1518 					"locked mapping %d "
1519 					"for context %d\n", i, cnum);
1520 #ifdef DDB
1521 				Debugger();
1522 #endif
1523 #endif
1524 				return (1);
1525 			}
1526 #ifdef DIAGNOSTIC
1527 			if (i < tlb_nreserved)
1528 				panic("TLB entry %d not locked", i);
1529 #endif
1530 			/* Invalidate particular TLB entry regardless of locked status */
1531 			__asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
1532 			tlb_info[i].ti_flags = 0;
1533 		}
1534 	}
1535 	return (0);
1536 }
1537 
1538 /*
1539  * Allocate a context.  If necessary, steal one from someone else.
1540  *
1541  * The new context is flushed from the TLB before returning.
1542  */
1543 int
1544 ctx_alloc(struct pmap *pm)
1545 {
1546 	int s, cnum;
1547 	static int next = MINCTX;
1548 
1549 	if (pm == pmap_kernel()) {
1550 #ifdef DIAGNOSTIC
1551 		printf("ctx_alloc: kernel pmap!\n");
1552 #endif
1553 		return (0);
1554 	}
1555 	s = splvm();
1556 
1557 	/* Find a likely context. */
1558 	cnum = next;
1559 	do {
1560 		if ((++cnum) > NUMCTX)
1561 			cnum = MINCTX;
1562 	} while (ctxbusy[cnum] != NULL && cnum != next);
1563 
1564 	/* Now clean it out */
1565 oops:
1566 	if (cnum < MINCTX)
1567 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
1568 	if (ctx_flush(cnum)) {
1569 		/* oops -- something's wired. */
1570 		if ((++cnum) > NUMCTX)
1571 			cnum = MINCTX;
1572 		goto oops;
1573 	}
1574 
1575 	if (ctxbusy[cnum]) {
1576 #ifdef DEBUG
1577 		/* We should identify this pmap and clear it */
1578 		printf("Warning: stealing context %d\n", cnum);
1579 #endif
1580 		ctxbusy[cnum]->pm_ctx = 0;
1581 	}
1582 	ctxbusy[cnum] = pm;
1583 	next = cnum;
1584 	splx(s);
1585 	pm->pm_ctx = cnum;
1586 
1587 	return cnum;
1588 }
1589 
1590 /*
1591  * Give away a context.
1592  */
1593 void
1594 ctx_free(struct pmap *pm)
1595 {
1596 	int oldctx;
1597 
1598 	oldctx = pm->pm_ctx;
1599 
1600 	if (oldctx == 0)
1601 		panic("ctx_free: freeing kernel context");
1602 #ifdef DIAGNOSTIC
1603 	if (ctxbusy[oldctx] == 0)
1604 		printf("ctx_free: freeing free context %d\n", oldctx);
1605 	if (ctxbusy[oldctx] != pm) {
1606 		printf("ctx_free: freeing someone esle's context\n "
1607 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
1608 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
1609 #ifdef DDB
1610 		Debugger();
1611 #endif
1612 	}
1613 #endif
1614 	/* We should verify it has not been stolen and reallocated... */
1615 	ctxbusy[oldctx] = NULL;
1616 	ctx_flush(oldctx);
1617 }
1618 
1619 
1620 #ifdef DEBUG
1621 /*
1622  * Test ref/modify handling.
1623  */
1624 void pmap_testout(void);
1625 void
1626 pmap_testout(void)
1627 {
1628 	vaddr_t va;
1629 	volatile int *loc;
1630 	int val = 0;
1631 	paddr_t pa;
1632 	struct vm_page *pg;
1633 	int ref, mod;
1634 
1635 	/* Allocate a page */
1636 	va = (vaddr_t)uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
1637 	    UVM_KMF_WIRED | UVM_KMF_ZERO);
1638 	loc = (int*)va;
1639 
1640 	pmap_extract(pmap_kernel(), va, &pa);
1641 	pg = PHYS_TO_VM_PAGE(pa);
1642 	pmap_unwire(pmap_kernel(), va);
1643 
1644 	pmap_kremove(va, PAGE_SIZE);
1645 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1646 	pmap_update(pmap_kernel());
1647 
1648 	/* Now clear reference and modify */
1649 	ref = pmap_clear_reference(pg);
1650 	mod = pmap_clear_modify(pg);
1651 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1652 	       (void *)(u_long)va, (long)pa,
1653 	       ref, mod);
1654 
1655 	/* Check it's properly cleared */
1656 	ref = pmap_is_referenced(pg);
1657 	mod = pmap_is_modified(pg);
1658 	printf("Checking cleared page: ref %d, mod %d\n",
1659 	       ref, mod);
1660 
1661 	/* Reference page */
1662 	val = *loc;
1663 
1664 	ref = pmap_is_referenced(pg);
1665 	mod = pmap_is_modified(pg);
1666 	printf("Referenced page: ref %d, mod %d val %x\n",
1667 	       ref, mod, val);
1668 
1669 	/* Now clear reference and modify */
1670 	ref = pmap_clear_reference(pg);
1671 	mod = pmap_clear_modify(pg);
1672 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1673 	       (void *)(u_long)va, (long)pa,
1674 	       ref, mod);
1675 
1676 	/* Modify page */
1677 	*loc = 1;
1678 
1679 	ref = pmap_is_referenced(pg);
1680 	mod = pmap_is_modified(pg);
1681 	printf("Modified page: ref %d, mod %d\n",
1682 	       ref, mod);
1683 
1684 	/* Now clear reference and modify */
1685 	ref = pmap_clear_reference(pg);
1686 	mod = pmap_clear_modify(pg);
1687 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1688 	       (void *)(u_long)va, (long)pa,
1689 	       ref, mod);
1690 
1691 	/* Check it's properly cleared */
1692 	ref = pmap_is_referenced(pg);
1693 	mod = pmap_is_modified(pg);
1694 	printf("Checking cleared page: ref %d, mod %d\n",
1695 	       ref, mod);
1696 
1697 	/* Modify page */
1698 	*loc = 1;
1699 
1700 	ref = pmap_is_referenced(pg);
1701 	mod = pmap_is_modified(pg);
1702 	printf("Modified page: ref %d, mod %d\n",
1703 	       ref, mod);
1704 
1705 	/* Check pmap_protect() */
1706 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
1707 	pmap_update(pmap_kernel());
1708 	ref = pmap_is_referenced(pg);
1709 	mod = pmap_is_modified(pg);
1710 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
1711 	       ref, mod);
1712 
1713 	/* Now clear reference and modify */
1714 	ref = pmap_clear_reference(pg);
1715 	mod = pmap_clear_modify(pg);
1716 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1717 	       (void *)(u_long)va, (long)pa,
1718 	       ref, mod);
1719 
1720 	/* Reference page */
1721 	val = *loc;
1722 
1723 	ref = pmap_is_referenced(pg);
1724 	mod = pmap_is_modified(pg);
1725 	printf("Referenced page: ref %d, mod %d val %x\n",
1726 	       ref, mod, val);
1727 
1728 	/* Now clear reference and modify */
1729 	ref = pmap_clear_reference(pg);
1730 	mod = pmap_clear_modify(pg);
1731 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1732 	       (void *)(u_long)va, (long)pa,
1733 	       ref, mod);
1734 
1735 	/* Modify page */
1736 #if 0
1737 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1738 	pmap_update(pmap_kernel());
1739 #endif
1740 	*loc = 1;
1741 
1742 	ref = pmap_is_referenced(pg);
1743 	mod = pmap_is_modified(pg);
1744 	printf("Modified page: ref %d, mod %d\n",
1745 	       ref, mod);
1746 
1747 	/* Check pmap_protect() */
1748 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
1749 	pmap_update(pmap_kernel());
1750 	ref = pmap_is_referenced(pg);
1751 	mod = pmap_is_modified(pg);
1752 	printf("pmap_protect(): ref %d, mod %d\n",
1753 	       ref, mod);
1754 
1755 	/* Now clear reference and modify */
1756 	ref = pmap_clear_reference(pg);
1757 	mod = pmap_clear_modify(pg);
1758 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1759 	       (void *)(u_long)va, (long)pa,
1760 	       ref, mod);
1761 
1762 	/* Reference page */
1763 	val = *loc;
1764 
1765 	ref = pmap_is_referenced(pg);
1766 	mod = pmap_is_modified(pg);
1767 	printf("Referenced page: ref %d, mod %d val %x\n",
1768 	       ref, mod, val);
1769 
1770 	/* Now clear reference and modify */
1771 	ref = pmap_clear_reference(pg);
1772 	mod = pmap_clear_modify(pg);
1773 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1774 	       (void *)(u_long)va, (long)pa,
1775 	       ref, mod);
1776 
1777 	/* Modify page */
1778 #if 0
1779 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1780 	pmap_update(pmap_kernel());
1781 #endif
1782 	*loc = 1;
1783 
1784 	ref = pmap_is_referenced(pg);
1785 	mod = pmap_is_modified(pg);
1786 	printf("Modified page: ref %d, mod %d\n",
1787 	       ref, mod);
1788 
1789 	/* Check pmap_pag_protect() */
1790 	pmap_page_protect(pg, VM_PROT_READ);
1791 	ref = pmap_is_referenced(pg);
1792 	mod = pmap_is_modified(pg);
1793 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
1794 	       ref, mod);
1795 
1796 	/* Now clear reference and modify */
1797 	ref = pmap_clear_reference(pg);
1798 	mod = pmap_clear_modify(pg);
1799 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1800 	       (void *)(u_long)va, (long)pa,
1801 	       ref, mod);
1802 
1803 	/* Reference page */
1804 	val = *loc;
1805 
1806 	ref = pmap_is_referenced(pg);
1807 	mod = pmap_is_modified(pg);
1808 	printf("Referenced page: ref %d, mod %d val %x\n",
1809 	       ref, mod, val);
1810 
1811 	/* Now clear reference and modify */
1812 	ref = pmap_clear_reference(pg);
1813 	mod = pmap_clear_modify(pg);
1814 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1815 	       (void *)(u_long)va, (long)pa,
1816 	       ref, mod);
1817 
1818 	/* Modify page */
1819 #if 0
1820 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1821 	pmap_update(pmap_kernel());
1822 #endif
1823 	*loc = 1;
1824 
1825 	ref = pmap_is_referenced(pg);
1826 	mod = pmap_is_modified(pg);
1827 	printf("Modified page: ref %d, mod %d\n",
1828 	       ref, mod);
1829 
1830 	/* Check pmap_pag_protect() */
1831 	pmap_page_protect(pg, VM_PROT_NONE);
1832 	ref = pmap_is_referenced(pg);
1833 	mod = pmap_is_modified(pg);
1834 	printf("pmap_page_protect(): ref %d, mod %d\n",
1835 	       ref, mod);
1836 
1837 	/* Now clear reference and modify */
1838 	ref = pmap_clear_reference(pg);
1839 	mod = pmap_clear_modify(pg);
1840 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1841 	       (void *)(u_long)va, (long)pa,
1842 	       ref, mod);
1843 
1844 
1845 	/* Reference page */
1846 	val = *loc;
1847 
1848 	ref = pmap_is_referenced(pg);
1849 	mod = pmap_is_modified(pg);
1850 	printf("Referenced page: ref %d, mod %d val %x\n",
1851 	       ref, mod, val);
1852 
1853 	/* Now clear reference and modify */
1854 	ref = pmap_clear_reference(pg);
1855 	mod = pmap_clear_modify(pg);
1856 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1857 	       (void *)(u_long)va, (long)pa,
1858 	       ref, mod);
1859 
1860 	/* Modify page */
1861 #if 0
1862 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1863 	pmap_update(pmap_kernel());
1864 #endif
1865 	*loc = 1;
1866 
1867 	ref = pmap_is_referenced(pg);
1868 	mod = pmap_is_modified(pg);
1869 	printf("Modified page: ref %d, mod %d\n",
1870 	       ref, mod);
1871 
1872 	/* Unmap page */
1873 	pmap_remove(pmap_kernel(), va, va+1);
1874 	pmap_update(pmap_kernel());
1875 	ref = pmap_is_referenced(pg);
1876 	mod = pmap_is_modified(pg);
1877 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
1878 
1879 	/* Now clear reference and modify */
1880 	ref = pmap_clear_reference(pg);
1881 	mod = pmap_clear_modify(pg);
1882 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1883 	       (void *)(u_long)va, (long)pa, ref, mod);
1884 
1885 	/* Check it's properly cleared */
1886 	ref = pmap_is_referenced(pg);
1887 	mod = pmap_is_modified(pg);
1888 	printf("Checking cleared page: ref %d, mod %d\n",
1889 	       ref, mod);
1890 
1891 	pmap_remove(pmap_kernel(), va, va + PAGE_SIZE);
1892 	pmap_kenter_pa(va, pa, VM_PROT_ALL, 0);
1893 	uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE, UVM_KMF_WIRED);
1894 }
1895 #endif
1896