xref: /netbsd-src/sys/arch/powerpc/ibm4xx/pmap.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: pmap.c,v 1.60 2009/11/21 17:40:29 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 1.60 2009/11/21 17:40:29 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 <machine/tlb.h>
88 
89 /*
90  * kernmap is an array of PTEs large enough to map in
91  * 4GB.  At 16KB/page it is 256K entries or 2MB.
92  */
93 #define KERNMAP_SIZE	((0xffffffffU/PAGE_SIZE)+1)
94 void *kernmap;
95 
96 #define MINCTX		2
97 #define NUMCTX		256
98 
99 volatile struct pmap *ctxbusy[NUMCTX];
100 
101 #define TLBF_USED	0x1
102 #define	TLBF_REF	0x2
103 #define	TLBF_LOCKED	0x4
104 #define	TLB_LOCKED(i)	(tlb_info[(i)].ti_flags & TLBF_LOCKED)
105 
106 typedef struct tlb_info_s {
107 	char	ti_flags;
108 	char	ti_ctx;		/* TLB_PID assiciated with the entry */
109 	u_int	ti_va;
110 } tlb_info_t;
111 
112 volatile tlb_info_t tlb_info[NTLB];
113 /* We'll use a modified FIFO replacement policy cause it's cheap */
114 volatile int tlbnext;
115 
116 static int tlb_nreserved = 0;
117 static int pmap_bootstrap_done = 0;
118 
119 /* Event counters */
120 struct evcnt tlbmiss_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
121 	NULL, "cpu", "tlbmiss");
122 struct evcnt tlbhit_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
123 	NULL, "cpu", "tlbhit");
124 struct evcnt tlbflush_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
125 	NULL, "cpu", "tlbflush");
126 struct evcnt tlbenter_ev = EVCNT_INITIALIZER(EVCNT_TYPE_TRAP,
127 	NULL, "cpu", "tlbenter");
128 
129 struct pmap kernel_pmap_;
130 struct pmap *const kernel_pmap_ptr = &kernel_pmap_;
131 
132 static int npgs;
133 static u_int nextavail;
134 #ifndef MSGBUFADDR
135 extern paddr_t msgbuf_paddr;
136 #endif
137 
138 static struct mem_region *mem, *avail;
139 
140 /*
141  * This is a cache of referenced/modified bits.
142  * Bits herein are shifted by ATTRSHFT.
143  */
144 static char *pmap_attrib;
145 
146 #define PV_WIRED	0x1
147 #define PV_WIRE(pv)	((pv)->pv_va |= PV_WIRED)
148 #define PV_UNWIRE(pv)	((pv)->pv_va &= ~PV_WIRED)
149 #define PV_ISWIRED(pv)	((pv)->pv_va & PV_WIRED)
150 #define PV_CMPVA(va,pv)	(!(((pv)->pv_va ^ (va)) & (~PV_WIRED)))
151 
152 struct pv_entry {
153 	struct pv_entry *pv_next;	/* Linked list of mappings */
154 	vaddr_t pv_va;			/* virtual address of mapping */
155 	struct pmap *pv_pm;
156 };
157 
158 /* Each index corresponds to TLB_SIZE_* value. */
159 static size_t tlbsize[] = {
160 	1024, 		/* TLB_SIZE_1K */
161 	4096, 		/* TLB_SIZE_4K */
162 	16384, 		/* TLB_SIZE_16K */
163 	65536, 		/* TLB_SIZE_64K */
164 	262144, 	/* TLB_SIZE_256K */
165 	1048576, 	/* TLB_SIZE_1M */
166 	4194304, 	/* TLB_SIZE_4M */
167 	16777216, 	/* TLB_SIZE_16M */
168 };
169 
170 struct pv_entry *pv_table;
171 static struct pool pv_pool;
172 
173 static int pmap_initialized;
174 
175 static int ctx_flush(int);
176 
177 inline struct pv_entry *pa_to_pv(paddr_t);
178 static inline char *pa_to_attr(paddr_t);
179 
180 static inline volatile u_int *pte_find(struct pmap *, vaddr_t);
181 static inline int pte_enter(struct pmap *, vaddr_t, u_int);
182 
183 static inline int pmap_enter_pv(struct pmap *, vaddr_t, paddr_t, int);
184 static void pmap_remove_pv(struct pmap *, vaddr_t, paddr_t);
185 
186 static int ppc4xx_tlb_size_mask(size_t, int *, int *);
187 
188 
189 inline struct pv_entry *
190 pa_to_pv(paddr_t pa)
191 {
192 	int bank, pg;
193 
194 	bank = vm_physseg_find(atop(pa), &pg);
195 	if (bank == -1)
196 		return NULL;
197 	return &vm_physmem[bank].pmseg.pvent[pg];
198 }
199 
200 static inline char *
201 pa_to_attr(paddr_t pa)
202 {
203 	int bank, pg;
204 
205 	bank = vm_physseg_find(atop(pa), &pg);
206 	if (bank == -1)
207 		return NULL;
208 	return &vm_physmem[bank].pmseg.attrs[pg];
209 }
210 
211 /*
212  * Insert PTE into page table.
213  */
214 int
215 pte_enter(struct pmap *pm, vaddr_t va, u_int pte)
216 {
217 	int seg = STIDX(va);
218 	int ptn = PTIDX(va);
219 	u_int oldpte;
220 
221 	if (!pm->pm_ptbl[seg]) {
222 		/* Don't allocate a page to clear a non-existent mapping. */
223 		if (!pte)
224 			return (0);
225 		/* Allocate a page XXXX this will sleep! */
226 		pm->pm_ptbl[seg] =
227 		    (uint *)uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
228 		    UVM_KMF_WIRED | UVM_KMF_ZERO);
229 	}
230 	oldpte = pm->pm_ptbl[seg][ptn];
231 	pm->pm_ptbl[seg][ptn] = pte;
232 
233 	/* Flush entry. */
234 	ppc4xx_tlb_flush(va, pm->pm_ctx);
235 	if (oldpte != pte) {
236 		if (pte == 0)
237 			pm->pm_stats.resident_count--;
238 		else
239 			pm->pm_stats.resident_count++;
240 	}
241 	return (1);
242 }
243 
244 /*
245  * Get a pointer to a PTE in a page table.
246  */
247 volatile u_int *
248 pte_find(struct pmap *pm, vaddr_t va)
249 {
250 	int seg = STIDX(va);
251 	int ptn = PTIDX(va);
252 
253 	if (pm->pm_ptbl[seg])
254 		return (&pm->pm_ptbl[seg][ptn]);
255 
256 	return (NULL);
257 }
258 
259 /*
260  * This is called during initppc, before the system is really initialized.
261  */
262 void
263 pmap_bootstrap(u_int kernelstart, u_int kernelend)
264 {
265 	struct mem_region *mp, *mp1;
266 	int cnt, i;
267 	u_int s, e, sz;
268 
269 	tlbnext = tlb_nreserved;
270 
271 	/*
272 	 * Allocate the kernel page table at the end of
273 	 * kernel space so it's in the locked TTE.
274 	 */
275 	kernmap = (void *)kernelend;
276 
277 	/*
278 	 * Initialize kernel page table.
279 	 */
280 	for (i = 0; i < STSZ; i++) {
281 		pmap_kernel()->pm_ptbl[i] = 0;
282 	}
283 	ctxbusy[0] = ctxbusy[1] = pmap_kernel();
284 
285 	/*
286 	 * Announce page-size to the VM-system
287 	 */
288 	uvmexp.pagesize = NBPG;
289 	uvm_setpagesize();
290 
291 	/*
292 	 * Get memory.
293 	 */
294 	mem_regions(&mem, &avail);
295 	for (mp = mem; mp->size; mp++) {
296 		physmem += btoc(mp->size);
297 		printf("+%lx,",mp->size);
298 	}
299 	printf("\n");
300 	ppc4xx_tlb_init();
301 	/*
302 	 * Count the number of available entries.
303 	 */
304 	for (cnt = 0, mp = avail; mp->size; mp++)
305 		cnt++;
306 
307 	/*
308 	 * Page align all regions.
309 	 * Non-page aligned memory isn't very interesting to us.
310 	 * Also, sort the entries for ascending addresses.
311 	 */
312 	kernelstart &= ~PGOFSET;
313 	kernelend = (kernelend + PGOFSET) & ~PGOFSET;
314 	for (mp = avail; mp->size; mp++) {
315 		s = mp->start;
316 		e = mp->start + mp->size;
317 		printf("%08x-%08x -> ",s,e);
318 		/*
319 		 * Check whether this region holds all of the kernel.
320 		 */
321 		if (s < kernelstart && e > kernelend) {
322 			avail[cnt].start = kernelend;
323 			avail[cnt++].size = e - kernelend;
324 			e = kernelstart;
325 		}
326 		/*
327 		 * Look whether this regions starts within the kernel.
328 		 */
329 		if (s >= kernelstart && s < kernelend) {
330 			if (e <= kernelend)
331 				goto empty;
332 			s = kernelend;
333 		}
334 		/*
335 		 * Now look whether this region ends within the kernel.
336 		 */
337 		if (e > kernelstart && e <= kernelend) {
338 			if (s >= kernelstart)
339 				goto empty;
340 			e = kernelstart;
341 		}
342 		/*
343 		 * Now page align the start and size of the region.
344 		 */
345 		s = round_page(s);
346 		e = trunc_page(e);
347 		if (e < s)
348 			e = s;
349 		sz = e - s;
350 		printf("%08x-%08x = %x\n",s,e,sz);
351 		/*
352 		 * Check whether some memory is left here.
353 		 */
354 		if (sz == 0) {
355 		empty:
356 			memmove(mp, mp + 1,
357 				(cnt - (mp - avail)) * sizeof *mp);
358 			cnt--;
359 			mp--;
360 			continue;
361 		}
362 		/*
363 		 * Do an insertion sort.
364 		 */
365 		npgs += btoc(sz);
366 		for (mp1 = avail; mp1 < mp; mp1++)
367 			if (s < mp1->start)
368 				break;
369 		if (mp1 < mp) {
370 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
371 			mp1->start = s;
372 			mp1->size = sz;
373 		} else {
374 			mp->start = s;
375 			mp->size = sz;
376 		}
377 	}
378 
379 	/*
380 	 * We cannot do pmap_steal_memory here,
381 	 * since we don't run with translation enabled yet.
382 	 */
383 #ifndef MSGBUFADDR
384 	/*
385 	 * allow for msgbuf
386 	 */
387 	sz = round_page(MSGBUFSIZE);
388 	mp = NULL;
389 	for (mp1 = avail; mp1->size; mp1++)
390 		if (mp1->size >= sz)
391 			mp = mp1;
392 	if (mp == NULL)
393 		panic("not enough memory?");
394 
395 	npgs -= btoc(sz);
396 	msgbuf_paddr = mp->start + mp->size - sz;
397 	mp->size -= sz;
398 	if (mp->size <= 0)
399 		memmove(mp, mp + 1, (cnt - (mp - avail)) * sizeof *mp);
400 #endif
401 
402 	for (mp = avail; mp->size; mp++)
403 		uvm_page_physload(atop(mp->start), atop(mp->start + mp->size),
404 			atop(mp->start), atop(mp->start + mp->size),
405 			VM_FREELIST_DEFAULT);
406 
407 	/*
408 	 * Initialize kernel pmap and hardware.
409 	 */
410 	/* Setup TLB pid allocator so it knows we alreadu using PID 1 */
411 	pmap_kernel()->pm_ctx = KERNEL_PID;
412 	nextavail = avail->start;
413 
414 	evcnt_attach_static(&tlbmiss_ev);
415 	evcnt_attach_static(&tlbhit_ev);
416 	evcnt_attach_static(&tlbflush_ev);
417 	evcnt_attach_static(&tlbenter_ev);
418 
419 	pmap_bootstrap_done = 1;
420 }
421 
422 /*
423  * Restrict given range to physical memory
424  *
425  * (Used by /dev/mem)
426  */
427 void
428 pmap_real_memory(paddr_t *start, psize_t *size)
429 {
430 	struct mem_region *mp;
431 
432 	for (mp = mem; mp->size; mp++) {
433 		if (*start + *size > mp->start &&
434 		    *start < mp->start + mp->size) {
435 			if (*start < mp->start) {
436 				*size -= mp->start - *start;
437 				*start = mp->start;
438 			}
439 			if (*start + *size > mp->start + mp->size)
440 				*size = mp->start + mp->size - *start;
441 			return;
442 		}
443 	}
444 	*size = 0;
445 }
446 
447 /*
448  * Initialize anything else for pmap handling.
449  * Called during vm_init().
450  */
451 void
452 pmap_init(void)
453 {
454 	struct pv_entry *pv;
455 	vsize_t sz;
456 	vaddr_t addr;
457 	int i, s;
458 	int bank;
459 	char *attr;
460 
461 	sz = (vsize_t)((sizeof(struct pv_entry) + 1) * npgs);
462 	sz = round_page(sz);
463 	addr = uvm_km_alloc(kernel_map, sz, 0, UVM_KMF_WIRED | UVM_KMF_ZERO);
464 	s = splvm();
465 	pv = pv_table = (struct pv_entry *)addr;
466 	for (i = npgs; --i >= 0;)
467 		pv++->pv_pm = NULL;
468 	pmap_attrib = (char *)pv;
469 	memset(pv, 0, npgs);
470 
471 	pv = pv_table;
472 	attr = pmap_attrib;
473 	for (bank = 0; bank < vm_nphysseg; bank++) {
474 		sz = vm_physmem[bank].end - vm_physmem[bank].start;
475 		vm_physmem[bank].pmseg.pvent = pv;
476 		vm_physmem[bank].pmseg.attrs = attr;
477 		pv += sz;
478 		attr += sz;
479 	}
480 
481 	pmap_initialized = 1;
482 	splx(s);
483 
484 	/* Setup a pool for additional pvlist structures */
485 	pool_init(&pv_pool, sizeof(struct pv_entry), 0, 0, 0, "pv_entry", NULL,
486 	    IPL_VM);
487 }
488 
489 /*
490  * How much virtual space is available to the kernel?
491  */
492 void
493 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
494 {
495 
496 #if 0
497 	/*
498 	 * Reserve one segment for kernel virtual memory
499 	 */
500 	*start = (vaddr_t)(KERNEL_SR << ADDR_SR_SHFT);
501 	*end = *start + SEGMENT_LENGTH;
502 #else
503 	*start = (vaddr_t) VM_MIN_KERNEL_ADDRESS;
504 	*end = (vaddr_t) VM_MAX_KERNEL_ADDRESS;
505 #endif
506 }
507 
508 #ifdef PMAP_GROWKERNEL
509 /*
510  * Preallocate kernel page tables to a specified VA.
511  * This simply loops through the first TTE for each
512  * page table from the beginning of the kernel pmap,
513  * reads the entry, and if the result is
514  * zero (either invalid entry or no page table) it stores
515  * a zero there, populating page tables in the process.
516  * This is not the most efficient technique but i don't
517  * expect it to be called that often.
518  */
519 extern struct vm_page *vm_page_alloc1(void);
520 extern void vm_page_free1(struct vm_page *);
521 
522 vaddr_t kbreak = VM_MIN_KERNEL_ADDRESS;
523 
524 vaddr_t
525 pmap_growkernel(vaddr_t maxkvaddr)
526 {
527 	int s;
528 	int seg;
529 	paddr_t pg;
530 	struct pmap *pm = pmap_kernel();
531 
532 	s = splvm();
533 
534 	/* Align with the start of a page table */
535 	for (kbreak &= ~(PTMAP-1); kbreak < maxkvaddr;
536 	     kbreak += PTMAP) {
537 		seg = STIDX(kbreak);
538 
539 		if (pte_find(pm, kbreak))
540 			continue;
541 
542 		if (uvm.page_init_done) {
543 			pg = (paddr_t)VM_PAGE_TO_PHYS(vm_page_alloc1());
544 		} else {
545 			if (!uvm_page_physget(&pg))
546 				panic("pmap_growkernel: no memory");
547 		}
548 		if (!pg)
549 			panic("pmap_growkernel: no pages");
550 		pmap_zero_page((paddr_t)pg);
551 
552 		/* XXX This is based on all phymem being addressable */
553 		pm->pm_ptbl[seg] = (u_int *)pg;
554 	}
555 	splx(s);
556 	return (kbreak);
557 }
558 
559 /*
560  *	vm_page_alloc1:
561  *
562  *	Allocate and return a memory cell with no associated object.
563  */
564 struct vm_page *
565 vm_page_alloc1(void)
566 {
567 	struct vm_page *pg;
568 
569 	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
570 	if (pg) {
571 		pg->wire_count = 1;	/* no mappings yet */
572 		pg->flags &= ~PG_BUSY;	/* never busy */
573 	}
574 	return pg;
575 }
576 
577 /*
578  *	vm_page_free1:
579  *
580  *	Returns the given page to the free list,
581  *	disassociating it with any VM object.
582  *
583  *	Object and page must be locked prior to entry.
584  */
585 void
586 vm_page_free1(struct vm_page *pg)
587 {
588 #ifdef DIAGNOSTIC
589 	if (pg->flags != (PG_CLEAN|PG_FAKE)) {
590 		printf("Freeing invalid page %p\n", pg);
591 		printf("pa = %llx\n", (unsigned long long)VM_PAGE_TO_PHYS(pg));
592 #ifdef DDB
593 		Debugger();
594 #endif
595 		return;
596 	}
597 #endif
598 	pg->flags |= PG_BUSY;
599 	pg->wire_count = 0;
600 	uvm_pagefree(pg);
601 }
602 #endif
603 
604 /*
605  * Create and return a physical map.
606  */
607 struct pmap *
608 pmap_create(void)
609 {
610 	struct pmap *pm;
611 
612 	pm = malloc(sizeof *pm, M_VMPMAP, M_WAITOK);
613 	memset(pm, 0, sizeof *pm);
614 	pm->pm_refs = 1;
615 	return pm;
616 }
617 
618 /*
619  * Add a reference to the given pmap.
620  */
621 void
622 pmap_reference(struct pmap *pm)
623 {
624 
625 	pm->pm_refs++;
626 }
627 
628 /*
629  * Retire the given pmap from service.
630  * Should only be called if the map contains no valid mappings.
631  */
632 void
633 pmap_destroy(struct pmap *pm)
634 {
635 	int i;
636 
637 	if (--pm->pm_refs > 0) {
638 		return;
639 	}
640 	KASSERT(pm->pm_stats.resident_count == 0);
641 	KASSERT(pm->pm_stats.wired_count == 0);
642 	for (i = 0; i < STSZ; i++)
643 		if (pm->pm_ptbl[i]) {
644 			uvm_km_free(kernel_map, (vaddr_t)pm->pm_ptbl[i],
645 			    PAGE_SIZE, UVM_KMF_WIRED);
646 			pm->pm_ptbl[i] = NULL;
647 		}
648 	if (pm->pm_ctx)
649 		ctx_free(pm);
650 	free(pm, M_VMPMAP);
651 }
652 
653 /*
654  * Copy the range specified by src_addr/len
655  * from the source map to the range dst_addr/len
656  * in the destination map.
657  *
658  * This routine is only advisory and need not do anything.
659  */
660 void
661 pmap_copy(struct pmap *dst_pmap, struct pmap *src_pmap, vaddr_t dst_addr,
662 	  vsize_t len, vaddr_t src_addr)
663 {
664 }
665 
666 /*
667  * Require that all active physical maps contain no
668  * incorrect entries NOW.
669  */
670 void
671 pmap_update(struct pmap *pmap)
672 {
673 }
674 
675 /*
676  * Fill the given physical page with zeroes.
677  */
678 void
679 pmap_zero_page(paddr_t pa)
680 {
681 
682 #ifdef PPC_4XX_NOCACHE
683 	memset((void *)pa, 0, PAGE_SIZE);
684 #else
685 	int i;
686 
687 	for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
688 		__asm volatile ("dcbz 0,%0" :: "r"(pa));
689 		pa += CACHELINESIZE;
690 	}
691 #endif
692 }
693 
694 /*
695  * Copy the given physical source page to its destination.
696  */
697 void
698 pmap_copy_page(paddr_t src, paddr_t dst)
699 {
700 
701 	memcpy((void *)dst, (void *)src, PAGE_SIZE);
702 	dcache_flush_page(dst);
703 }
704 
705 /*
706  * This returns != 0 on success.
707  */
708 static inline int
709 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa, int flags)
710 {
711 	struct pv_entry *pv, *npv = NULL;
712 	int s;
713 
714 	if (!pmap_initialized)
715 		return 0;
716 
717 	s = splvm();
718 	pv = pa_to_pv(pa);
719 	if (!pv->pv_pm) {
720 		/*
721 		 * No entries yet, use header as the first entry.
722 		 */
723 		pv->pv_va = va;
724 		pv->pv_pm = pm;
725 		pv->pv_next = NULL;
726 	} else {
727 		/*
728 		 * There is at least one other VA mapping this page.
729 		 * Place this entry after the header.
730 		 */
731 		npv = pool_get(&pv_pool, PR_NOWAIT);
732 		if (npv == NULL) {
733 			if ((flags & PMAP_CANFAIL) == 0)
734 				panic("pmap_enter_pv: failed");
735 			splx(s);
736 			return 0;
737 		}
738 		npv->pv_va = va;
739 		npv->pv_pm = pm;
740 		npv->pv_next = pv->pv_next;
741 		pv->pv_next = npv;
742 		pv = npv;
743 	}
744 	if (flags & PMAP_WIRED) {
745 		PV_WIRE(pv);
746 		pm->pm_stats.wired_count++;
747 	}
748 	splx(s);
749 	return (1);
750 }
751 
752 static void
753 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
754 {
755 	struct pv_entry *pv, *npv;
756 
757 	/*
758 	 * Remove from the PV table.
759 	 */
760 	pv = pa_to_pv(pa);
761 	if (!pv)
762 		return;
763 
764 	/*
765 	 * If it is the first entry on the list, it is actually
766 	 * in the header and we must copy the following entry up
767 	 * to the header.  Otherwise we must search the list for
768 	 * the entry.  In either case we free the now unused entry.
769 	 */
770 	if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
771 		if (PV_ISWIRED(pv)) {
772 			pm->pm_stats.wired_count--;
773 		}
774 		if ((npv = pv->pv_next)) {
775 			*pv = *npv;
776 			pool_put(&pv_pool, npv);
777 		} else
778 			pv->pv_pm = NULL;
779 	} else {
780 		for (; (npv = pv->pv_next) != NULL; pv = npv)
781 			if (pm == npv->pv_pm && PV_CMPVA(va, npv))
782 				break;
783 		if (npv) {
784 			pv->pv_next = npv->pv_next;
785 			if (PV_ISWIRED(npv)) {
786 				pm->pm_stats.wired_count--;
787 			}
788 			pool_put(&pv_pool, npv);
789 		}
790 	}
791 }
792 
793 /*
794  * Insert physical page at pa into the given pmap at virtual address va.
795  */
796 int
797 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
798 {
799 	int s;
800 	u_int tte;
801 	bool managed;
802 
803 	/*
804 	 * Have to remove any existing mapping first.
805 	 */
806 	pmap_remove(pm, va, va + PAGE_SIZE);
807 
808 	if (flags & PMAP_WIRED)
809 		flags |= prot;
810 
811 	managed = uvm_pageismanaged(pa);
812 
813 	/*
814 	 * Generate TTE.
815 	 */
816 	tte = TTE_PA(pa);
817 	/* XXXX -- need to support multiple page sizes. */
818 	tte |= TTE_SZ_16K;
819 #ifdef	DIAGNOSTIC
820 	if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
821 		(PME_NOCACHE | PME_WRITETHROUG))
822 		panic("pmap_enter: uncached & writethrough");
823 #endif
824 	if (flags & PME_NOCACHE)
825 		/* Must be I/O mapping */
826 		tte |= TTE_I | TTE_G;
827 #ifdef PPC_4XX_NOCACHE
828 	tte |= TTE_I;
829 #else
830 	else if (flags & PME_WRITETHROUG)
831 		/* Uncached and writethrough are not compatible */
832 		tte |= TTE_W;
833 #endif
834 	if (pm == pmap_kernel())
835 		tte |= TTE_ZONE(ZONE_PRIV);
836 	else
837 		tte |= TTE_ZONE(ZONE_USER);
838 
839 	if (flags & VM_PROT_WRITE)
840 		tte |= TTE_WR;
841 
842 	if (flags & VM_PROT_EXECUTE)
843 		tte |= TTE_EX;
844 
845 	/*
846 	 * Now record mapping for later back-translation.
847 	 */
848 	if (pmap_initialized && managed) {
849 		char *attr;
850 
851 		if (!pmap_enter_pv(pm, va, pa, flags)) {
852 			/* Could not enter pv on a managed page */
853 			return 1;
854 		}
855 
856 		/* Now set attributes. */
857 		attr = pa_to_attr(pa);
858 #ifdef DIAGNOSTIC
859 		if (!attr)
860 			panic("managed but no attr");
861 #endif
862 		if (flags & VM_PROT_ALL)
863 			*attr |= PMAP_ATTR_REF;
864 		if (flags & VM_PROT_WRITE)
865 			*attr |= PMAP_ATTR_CHG;
866 	}
867 
868 	s = splvm();
869 
870 	/* Insert page into page table. */
871 	pte_enter(pm, va, tte);
872 
873 	/* If this is a real fault, enter it in the tlb */
874 	if (tte && ((flags & PMAP_WIRED) == 0)) {
875 		ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
876 	}
877 	splx(s);
878 
879 	/* Flush the real memory from the instruction cache. */
880 	if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
881 		__syncicache((void *)pa, PAGE_SIZE);
882 
883 	return 0;
884 }
885 
886 void
887 pmap_unwire(struct pmap *pm, vaddr_t va)
888 {
889 	struct pv_entry *pv;
890 	paddr_t pa;
891 	int s;
892 
893 	if (!pmap_extract(pm, va, &pa)) {
894 		return;
895 	}
896 
897 	pv = pa_to_pv(pa);
898 	if (!pv)
899 		return;
900 
901 	s = splvm();
902 	while (pv != NULL) {
903 		if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
904 			if (PV_ISWIRED(pv)) {
905 				PV_UNWIRE(pv);
906 				pm->pm_stats.wired_count--;
907 			}
908 			break;
909 		}
910 		pv = pv->pv_next;
911 	}
912 	splx(s);
913 }
914 
915 void
916 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
917 {
918 	int s;
919 	u_int tte;
920 	struct pmap *pm = pmap_kernel();
921 
922 	/*
923 	 * Have to remove any existing mapping first.
924 	 */
925 
926 	/*
927 	 * Generate TTE.
928 	 *
929 	 * XXXX
930 	 *
931 	 * Since the kernel does not handle execution privileges properly,
932 	 * we will handle read and execute permissions together.
933 	 */
934 	tte = 0;
935 	if (prot & VM_PROT_ALL) {
936 
937 		tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
938 		/* XXXX -- need to support multiple page sizes. */
939 		tte |= TTE_SZ_16K;
940 #ifdef DIAGNOSTIC
941 		if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
942 			(PME_NOCACHE | PME_WRITETHROUG))
943 			panic("pmap_kenter_pa: uncached & writethrough");
944 #endif
945 		if (prot & PME_NOCACHE)
946 			/* Must be I/O mapping */
947 			tte |= TTE_I | TTE_G;
948 #ifdef PPC_4XX_NOCACHE
949 		tte |= TTE_I;
950 #else
951 		else if (prot & PME_WRITETHROUG)
952 			/* Uncached and writethrough are not compatible */
953 			tte |= TTE_W;
954 #endif
955 		if (prot & VM_PROT_WRITE)
956 			tte |= TTE_WR;
957 	}
958 
959 	s = splvm();
960 
961 	/* Insert page into page table. */
962 	pte_enter(pm, va, tte);
963 	splx(s);
964 }
965 
966 void
967 pmap_kremove(vaddr_t va, vsize_t len)
968 {
969 
970 	while (len > 0) {
971 		pte_enter(pmap_kernel(), va, 0);
972 		va += PAGE_SIZE;
973 		len -= PAGE_SIZE;
974 	}
975 }
976 
977 /*
978  * Remove the given range of mapping entries.
979  */
980 void
981 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
982 {
983 	int s;
984 	paddr_t pa;
985 	volatile u_int *ptp;
986 
987 	s = splvm();
988 	while (va < endva) {
989 
990 		if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
991 			pa = TTE_PA(pa);
992 			pmap_remove_pv(pm, va, pa);
993 			*ptp = 0;
994 			ppc4xx_tlb_flush(va, pm->pm_ctx);
995 			pm->pm_stats.resident_count--;
996 		}
997 		va += PAGE_SIZE;
998 	}
999 
1000 	splx(s);
1001 }
1002 
1003 /*
1004  * Get the physical page address for the given pmap/virtual address.
1005  */
1006 bool
1007 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
1008 {
1009 	int seg = STIDX(va);
1010 	int ptn = PTIDX(va);
1011 	u_int pa = 0;
1012 	int s;
1013 
1014 	s = splvm();
1015 	if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
1016 		*pap = TTE_PA(pa) | (va & PGOFSET);
1017 	}
1018 	splx(s);
1019 	return (pa != 0);
1020 }
1021 
1022 /*
1023  * Lower the protection on the specified range of this pmap.
1024  *
1025  * There are only two cases: either the protection is going to 0,
1026  * or it is going to read-only.
1027  */
1028 void
1029 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1030 {
1031 	volatile u_int *ptp;
1032 	int s, bic;
1033 
1034 	if ((prot & VM_PROT_READ) == 0) {
1035 		pmap_remove(pm, sva, eva);
1036 		return;
1037 	}
1038 	bic = 0;
1039 	if ((prot & VM_PROT_WRITE) == 0) {
1040 		bic |= TTE_WR;
1041 	}
1042 	if ((prot & VM_PROT_EXECUTE) == 0) {
1043 		bic |= TTE_EX;
1044 	}
1045 	if (bic == 0) {
1046 		return;
1047 	}
1048 	s = splvm();
1049 	while (sva < eva) {
1050 		if ((ptp = pte_find(pm, sva)) != NULL) {
1051 			*ptp &= ~bic;
1052 			ppc4xx_tlb_flush(sva, pm->pm_ctx);
1053 		}
1054 		sva += PAGE_SIZE;
1055 	}
1056 	splx(s);
1057 }
1058 
1059 bool
1060 pmap_check_attr(struct vm_page *pg, u_int mask, int clear)
1061 {
1062 	paddr_t pa;
1063 	char *attr;
1064 	int s, rv;
1065 
1066 	/*
1067 	 * First modify bits in cache.
1068 	 */
1069 	pa = VM_PAGE_TO_PHYS(pg);
1070 	attr = pa_to_attr(pa);
1071 	if (attr == NULL)
1072 		return false;
1073 
1074 	s = splvm();
1075 	rv = ((*attr & mask) != 0);
1076 	if (clear) {
1077 		*attr &= ~mask;
1078 		pmap_page_protect(pg, mask == PMAP_ATTR_CHG ? VM_PROT_READ : 0);
1079 	}
1080 	splx(s);
1081 	return rv;
1082 }
1083 
1084 
1085 /*
1086  * Lower the protection on the specified physical page.
1087  *
1088  * There are only two cases: either the protection is going to 0,
1089  * or it is going to read-only.
1090  */
1091 void
1092 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1093 {
1094 	paddr_t pa = VM_PAGE_TO_PHYS(pg);
1095 	vaddr_t va;
1096 	struct pv_entry *pvh, *pv, *npv;
1097 	struct pmap *pm;
1098 
1099 	pvh = pa_to_pv(pa);
1100 	if (pvh == NULL)
1101 		return;
1102 
1103 	/* Handle extra pvs which may be deleted in the operation */
1104 	for (pv = pvh->pv_next; pv; pv = npv) {
1105 		npv = pv->pv_next;
1106 
1107 		pm = pv->pv_pm;
1108 		va = pv->pv_va;
1109 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
1110 	}
1111 	/* Now check the head pv */
1112 	if (pvh->pv_pm) {
1113 		pv = pvh;
1114 		pm = pv->pv_pm;
1115 		va = pv->pv_va;
1116 		pmap_protect(pm, va, va + PAGE_SIZE, prot);
1117 	}
1118 }
1119 
1120 /*
1121  * Activate the address space for the specified process.  If the process
1122  * is the current process, load the new MMU context.
1123  */
1124 void
1125 pmap_activate(struct lwp *l)
1126 {
1127 #if 0
1128 	struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
1129 	pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
1130 
1131 	/*
1132 	 * XXX Normally performed in cpu_fork().
1133 	 */
1134 	printf("pmap_activate(%p), pmap=%p\n",l,pmap);
1135 	pcb->pcb_pm = pmap;
1136 #endif
1137 }
1138 
1139 /*
1140  * Deactivate the specified process's address space.
1141  */
1142 void
1143 pmap_deactivate(struct lwp *l)
1144 {
1145 }
1146 
1147 /*
1148  * Synchronize caches corresponding to [addr, addr+len) in p.
1149  */
1150 void
1151 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
1152 {
1153 	struct pmap *pm = p->p_vmspace->vm_map.pmap;
1154 	int msr, ctx, opid, step;
1155 
1156 	step = CACHELINESIZE;
1157 
1158 	/*
1159 	 * Need to turn off IMMU and switch to user context.
1160 	 * (icbi uses DMMU).
1161 	 */
1162 	if (!(ctx = pm->pm_ctx)) {
1163 		/* No context -- assign it one */
1164 		ctx_alloc(pm);
1165 		ctx = pm->pm_ctx;
1166 	}
1167 	__asm volatile("mfmsr %0;"
1168 		"li %1, %7;"
1169 		"andc %1,%0,%1;"
1170 		"mtmsr %1;"
1171 		"sync;isync;"
1172 		"mfpid %1;"
1173 		"mtpid %2;"
1174 		"sync; isync;"
1175 		"1:"
1176 		"dcbf 0,%3;"
1177 		"icbi 0,%3;"
1178 		"add %3,%3,%5;"
1179 		"addc. %4,%4,%6;"
1180 		"bge 1b;"
1181 		"mtpid %1;"
1182 		"mtmsr %0;"
1183 		"sync; isync"
1184 		: "=&r" (msr), "=&r" (opid)
1185 		: "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step),
1186 		  "K" (PSL_IR | PSL_DR));
1187 }
1188 
1189 
1190 /* This has to be done in real mode !!! */
1191 void
1192 ppc4xx_tlb_flush(vaddr_t va, int pid)
1193 {
1194 	u_long i, found;
1195 	u_long msr;
1196 
1197 	/* If there's no context then it can't be mapped. */
1198 	if (!pid)
1199 		return;
1200 
1201 	__asm( 	"mfpid %1;"		/* Save PID */
1202 		"mfmsr %2;"		/* Save MSR */
1203 		"li %0,0;"		/* Now clear MSR */
1204 		"mtmsr %0;"
1205 		"mtpid %4;"		/* Set PID */
1206 		"sync;"
1207 		"tlbsx. %0,0,%3;"	/* Search TLB */
1208 		"sync;"
1209 		"mtpid %1;"		/* Restore PID */
1210 		"mtmsr %2;"		/* Restore MSR */
1211 		"sync;isync;"
1212 		"li %1,1;"
1213 		"beq 1f;"
1214 		"li %1,0;"
1215 		"1:"
1216 		: "=&r" (i), "=&r" (found), "=&r" (msr)
1217 		: "r" (va), "r" (pid));
1218 	if (found && !TLB_LOCKED(i)) {
1219 
1220 		/* Now flush translation */
1221 		__asm volatile(
1222 			"tlbwe %0,%1,0;"
1223 			"sync;isync;"
1224 			: : "r" (0), "r" (i));
1225 
1226 		tlb_info[i].ti_ctx = 0;
1227 		tlb_info[i].ti_flags = 0;
1228 		tlbnext = i;
1229 		/* Successful flushes */
1230 		tlbflush_ev.ev_count++;
1231 	}
1232 }
1233 
1234 void
1235 ppc4xx_tlb_flush_all(void)
1236 {
1237 	u_long i;
1238 
1239 	for (i = 0; i < NTLB; i++)
1240 		if (!TLB_LOCKED(i)) {
1241 			__asm volatile(
1242 				"tlbwe %0,%1,0;"
1243 				"sync;isync;"
1244 				: : "r" (0), "r" (i));
1245 			tlb_info[i].ti_ctx = 0;
1246 			tlb_info[i].ti_flags = 0;
1247 		}
1248 
1249 	__asm volatile("sync;isync");
1250 }
1251 
1252 /* Find a TLB entry to evict. */
1253 static int
1254 ppc4xx_tlb_find_victim(void)
1255 {
1256 	int flags;
1257 
1258 	for (;;) {
1259 		if (++tlbnext >= NTLB)
1260 			tlbnext = tlb_nreserved;
1261 		flags = tlb_info[tlbnext].ti_flags;
1262 		if (!(flags & TLBF_USED) ||
1263 			(flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
1264 			u_long va, stack = (u_long)&va;
1265 
1266 			if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
1267 			    (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
1268 			     (flags & TLBF_USED)) {
1269 				/* Kernel stack page */
1270 				flags |= TLBF_USED;
1271 				tlb_info[tlbnext].ti_flags = flags;
1272 			} else {
1273 				/* Found it! */
1274 				return (tlbnext);
1275 			}
1276 		} else {
1277 			tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
1278 		}
1279 	}
1280 }
1281 
1282 void
1283 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
1284 {
1285 	u_long th, tl, idx;
1286 	tlbpid_t pid;
1287 	u_short msr;
1288 	paddr_t pa;
1289 	int s, sz;
1290 
1291 	tlbenter_ev.ev_count++;
1292 
1293 	sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
1294 	pa = (pte & TTE_RPN_MASK(sz));
1295 	th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
1296 	tl = (pte & ~TLB_RPN_MASK) | pa;
1297 	tl |= ppc4xx_tlbflags(va, pa);
1298 
1299 	s = splhigh();
1300 	idx = ppc4xx_tlb_find_victim();
1301 
1302 #ifdef DIAGNOSTIC
1303 	if ((idx < tlb_nreserved) || (idx >= NTLB)) {
1304 		panic("ppc4xx_tlb_enter: replacing entry %ld", idx);
1305 	}
1306 #endif
1307 
1308 	tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
1309 	tlb_info[idx].ti_ctx = ctx;
1310 	tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
1311 
1312 	__asm volatile(
1313 		"mfmsr %0;"			/* Save MSR */
1314 		"li %1,0;"
1315 		"tlbwe %1,%3,0;"		/* Invalidate old entry. */
1316 		"mtmsr %1;"			/* Clear MSR */
1317 		"mfpid %1;"			/* Save old PID */
1318 		"mtpid %2;"			/* Load translation ctx */
1319 		"sync; isync;"
1320 #ifdef DEBUG
1321 		"andi. %3,%3,63;"
1322 		"tweqi %3,0;" 			/* XXXXX DEBUG trap on index 0 */
1323 #endif
1324 		"tlbwe %4,%3,1; tlbwe %5,%3,0;"	/* Set TLB */
1325 		"sync; isync;"
1326 		"mtpid %1; mtmsr %0;"		/* Restore PID and MSR */
1327 		"sync; isync;"
1328 	: "=&r" (msr), "=&r" (pid)
1329 	: "r" (ctx), "r" (idx), "r" (tl), "r" (th));
1330 	splx(s);
1331 }
1332 
1333 void
1334 ppc4xx_tlb_init(void)
1335 {
1336 	int i;
1337 
1338 	/* Mark reserved TLB entries */
1339 	for (i = 0; i < tlb_nreserved; i++) {
1340 		tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
1341 		tlb_info[i].ti_ctx = KERNEL_PID;
1342 	}
1343 
1344 	/* Setup security zones */
1345 	/* Z0 - accessible by kernel only if TLB entry permissions allow
1346 	 * Z1,Z2 - access is controlled by TLB entry permissions
1347 	 * Z3 - full access regardless of TLB entry permissions
1348 	 */
1349 
1350 	__asm volatile(
1351 		"mtspr %0,%1;"
1352 		"sync;"
1353 		::  "K"(SPR_ZPR), "r" (0x1b000000));
1354 }
1355 
1356 /*
1357  * ppc4xx_tlb_size_mask:
1358  *
1359  * 	Roundup size to supported page size, return TLBHI mask and real size.
1360  */
1361 static int
1362 ppc4xx_tlb_size_mask(size_t size, int *mask, int *rsiz)
1363 {
1364 	int 			i;
1365 
1366 	for (i = 0; i < __arraycount(tlbsize); i++)
1367 		if (size <= tlbsize[i]) {
1368 			*mask = (i << TLB_SIZE_SHFT);
1369 			*rsiz = tlbsize[i];
1370 			return (0);
1371 		}
1372 	return (EINVAL);
1373 }
1374 
1375 /*
1376  * ppc4xx_tlb_mapiodev:
1377  *
1378  * 	Lookup virtual address of mapping previously entered via
1379  * 	ppc4xx_tlb_reserve. Search TLB directly so that we don't
1380  * 	need to waste extra storage for reserved mappings. Note
1381  * 	that reading TLBHI also sets PID, but all reserved mappings
1382  * 	use KERNEL_PID, so the side effect is nil.
1383  */
1384 void *
1385 ppc4xx_tlb_mapiodev(paddr_t base, psize_t len)
1386 {
1387 	paddr_t 		pa;
1388 	vaddr_t 		va;
1389 	u_int 			lo, hi, sz;
1390 	int 			i;
1391 
1392 	/* tlb_nreserved is only allowed to grow, so this is safe. */
1393 	for (i = 0; i < tlb_nreserved; i++) {
1394 		__asm volatile (
1395 		    "	tlbre %0,%2,1 	\n" 	/* TLBLO */
1396 		    "	tlbre %1,%2,0 	\n" 	/* TLBHI */
1397 		    : "=&r" (lo), "=&r" (hi)
1398 		    : "r" (i));
1399 
1400 		KASSERT(hi & TLB_VALID);
1401 		KASSERT(mfspr(SPR_PID) == KERNEL_PID);
1402 
1403 		pa = (lo & TLB_RPN_MASK);
1404 		if (base < pa)
1405 			continue;
1406 
1407 		sz = tlbsize[(hi & TLB_SIZE_MASK) >> TLB_SIZE_SHFT];
1408 		if ((base + len) > (pa + sz))
1409 			continue;
1410 
1411 		va = (hi & TLB_EPN_MASK) + (base & (sz - 1)); 	/* sz = 2^n */
1412 		return (void *)(va);
1413 	}
1414 
1415 	return (NULL);
1416 }
1417 
1418 /*
1419  * ppc4xx_tlb_reserve:
1420  *
1421  * 	Map physical range to kernel virtual chunk via reserved TLB entry.
1422  */
1423 void
1424 ppc4xx_tlb_reserve(paddr_t pa, vaddr_t va, size_t size, int flags)
1425 {
1426 	u_int 			lo, hi;
1427 	int 			szmask, rsize;
1428 
1429 	/* Called before pmap_bootstrap(), va outside kernel space. */
1430 	KASSERT(va < VM_MIN_KERNEL_ADDRESS || va >= VM_MAX_KERNEL_ADDRESS);
1431 	KASSERT(! pmap_bootstrap_done);
1432 	KASSERT(tlb_nreserved < NTLB);
1433 
1434 	/* Resolve size. */
1435 	if (ppc4xx_tlb_size_mask(size, &szmask, &rsize) != 0)
1436 		panic("ppc4xx_tlb_reserve: entry %d, %zuB too large",
1437 		    size, tlb_nreserved);
1438 
1439 	/* Real size will be power of two >= 1024, so this is OK. */
1440 	pa &= ~(rsize - 1); 	/* RPN */
1441 	va &= ~(rsize - 1); 	/* EPN */
1442 
1443 	lo = pa | TLB_WR | flags;
1444 	hi = va | TLB_VALID | szmask;
1445 
1446 #ifdef PPC_4XX_NOCACHE
1447 	lo |= TLB_I;
1448 #endif
1449 
1450 	__asm volatile(
1451 	    "	tlbwe %1,%0,1 	\n" 	/* write TLBLO */
1452 	    "	tlbwe %2,%0,0 	\n" 	/* write TLBHI */
1453 	    "   sync 		\n"
1454 	    "	isync 		\n"
1455 	    : : "r" (tlb_nreserved), "r" (lo), "r" (hi));
1456 
1457 	tlb_nreserved++;
1458 }
1459 
1460 /*
1461  * We should pass the ctx in from trap code.
1462  */
1463 int
1464 pmap_tlbmiss(vaddr_t va, int ctx)
1465 {
1466 	volatile u_int *pte;
1467 	u_long tte;
1468 
1469 	tlbmiss_ev.ev_count++;
1470 
1471 	/*
1472 	 * We will reserve 0 upto VM_MIN_KERNEL_ADDRESS for va == pa mappings.
1473 	 * Physical RAM is expected to live in this range, care must be taken
1474 	 * to not clobber 0 upto ${physmem} with device mappings in machdep
1475 	 * code.
1476 	 */
1477 	if (ctx != KERNEL_PID || va >= VM_MIN_KERNEL_ADDRESS) {
1478 		pte = pte_find((struct pmap *)__UNVOLATILE(ctxbusy[ctx]), va);
1479 		if (pte == NULL) {
1480 			/* Map unmanaged addresses directly for kernel access */
1481 			return 1;
1482 		}
1483 		tte = *pte;
1484 		if (tte == 0) {
1485 			return 1;
1486 		}
1487 	} else {
1488 		/* Create a 16MB writable mapping. */
1489 #ifdef PPC_4XX_NOCACHE
1490 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I |TTE_WR;
1491 #else
1492 		tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
1493 #endif
1494 	}
1495 	tlbhit_ev.ev_count++;
1496 	ppc4xx_tlb_enter(ctx, va, tte);
1497 
1498 	return 0;
1499 }
1500 
1501 /*
1502  * Flush all the entries matching a context from the TLB.
1503  */
1504 static int
1505 ctx_flush(int cnum)
1506 {
1507 	int i;
1508 
1509 	/* We gotta steal this context */
1510 	for (i = tlb_nreserved; i < NTLB; i++) {
1511 		if (tlb_info[i].ti_ctx == cnum) {
1512 			/* Can't steal ctx if it has a locked entry. */
1513 			if (TLB_LOCKED(i)) {
1514 #ifdef DIAGNOSTIC
1515 				printf("ctx_flush: can't invalidate "
1516 					"locked mapping %d "
1517 					"for context %d\n", i, cnum);
1518 #ifdef DDB
1519 				Debugger();
1520 #endif
1521 #endif
1522 				return (1);
1523 			}
1524 #ifdef DIAGNOSTIC
1525 			if (i < tlb_nreserved)
1526 				panic("TLB entry %d not locked", i);
1527 #endif
1528 			/* Invalidate particular TLB entry regardless of locked status */
1529 			__asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
1530 			tlb_info[i].ti_flags = 0;
1531 		}
1532 	}
1533 	return (0);
1534 }
1535 
1536 /*
1537  * Allocate a context.  If necessary, steal one from someone else.
1538  *
1539  * The new context is flushed from the TLB before returning.
1540  */
1541 int
1542 ctx_alloc(struct pmap *pm)
1543 {
1544 	int s, cnum;
1545 	static int next = MINCTX;
1546 
1547 	if (pm == pmap_kernel()) {
1548 #ifdef DIAGNOSTIC
1549 		printf("ctx_alloc: kernel pmap!\n");
1550 #endif
1551 		return (0);
1552 	}
1553 	s = splvm();
1554 
1555 	/* Find a likely context. */
1556 	cnum = next;
1557 	do {
1558 		if ((++cnum) > NUMCTX)
1559 			cnum = MINCTX;
1560 	} while (ctxbusy[cnum] != NULL && cnum != next);
1561 
1562 	/* Now clean it out */
1563 oops:
1564 	if (cnum < MINCTX)
1565 		cnum = MINCTX; /* Never steal ctx 0 or 1 */
1566 	if (ctx_flush(cnum)) {
1567 		/* oops -- something's wired. */
1568 		if ((++cnum) > NUMCTX)
1569 			cnum = MINCTX;
1570 		goto oops;
1571 	}
1572 
1573 	if (ctxbusy[cnum]) {
1574 #ifdef DEBUG
1575 		/* We should identify this pmap and clear it */
1576 		printf("Warning: stealing context %d\n", cnum);
1577 #endif
1578 		ctxbusy[cnum]->pm_ctx = 0;
1579 	}
1580 	ctxbusy[cnum] = pm;
1581 	next = cnum;
1582 	splx(s);
1583 	pm->pm_ctx = cnum;
1584 
1585 	return cnum;
1586 }
1587 
1588 /*
1589  * Give away a context.
1590  */
1591 void
1592 ctx_free(struct pmap *pm)
1593 {
1594 	int oldctx;
1595 
1596 	oldctx = pm->pm_ctx;
1597 
1598 	if (oldctx == 0)
1599 		panic("ctx_free: freeing kernel context");
1600 #ifdef DIAGNOSTIC
1601 	if (ctxbusy[oldctx] == 0)
1602 		printf("ctx_free: freeing free context %d\n", oldctx);
1603 	if (ctxbusy[oldctx] != pm) {
1604 		printf("ctx_free: freeing someone esle's context\n "
1605 		       "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
1606 		       oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
1607 #ifdef DDB
1608 		Debugger();
1609 #endif
1610 	}
1611 #endif
1612 	/* We should verify it has not been stolen and reallocated... */
1613 	ctxbusy[oldctx] = NULL;
1614 	ctx_flush(oldctx);
1615 }
1616 
1617 
1618 #ifdef DEBUG
1619 /*
1620  * Test ref/modify handling.
1621  */
1622 void pmap_testout(void);
1623 void
1624 pmap_testout(void)
1625 {
1626 	vaddr_t va;
1627 	volatile int *loc;
1628 	int val = 0;
1629 	paddr_t pa;
1630 	struct vm_page *pg;
1631 	int ref, mod;
1632 
1633 	/* Allocate a page */
1634 	va = (vaddr_t)uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
1635 	    UVM_KMF_WIRED | UVM_KMF_ZERO);
1636 	loc = (int*)va;
1637 
1638 	pmap_extract(pmap_kernel(), va, &pa);
1639 	pg = PHYS_TO_VM_PAGE(pa);
1640 	pmap_unwire(pmap_kernel(), va);
1641 
1642 	pmap_kremove(va, PAGE_SIZE);
1643 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1644 	pmap_update(pmap_kernel());
1645 
1646 	/* Now clear reference and modify */
1647 	ref = pmap_clear_reference(pg);
1648 	mod = pmap_clear_modify(pg);
1649 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1650 	       (void *)(u_long)va, (long)pa,
1651 	       ref, mod);
1652 
1653 	/* Check it's properly cleared */
1654 	ref = pmap_is_referenced(pg);
1655 	mod = pmap_is_modified(pg);
1656 	printf("Checking cleared page: ref %d, mod %d\n",
1657 	       ref, mod);
1658 
1659 	/* Reference page */
1660 	val = *loc;
1661 
1662 	ref = pmap_is_referenced(pg);
1663 	mod = pmap_is_modified(pg);
1664 	printf("Referenced page: ref %d, mod %d val %x\n",
1665 	       ref, mod, val);
1666 
1667 	/* Now clear reference and modify */
1668 	ref = pmap_clear_reference(pg);
1669 	mod = pmap_clear_modify(pg);
1670 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1671 	       (void *)(u_long)va, (long)pa,
1672 	       ref, mod);
1673 
1674 	/* Modify page */
1675 	*loc = 1;
1676 
1677 	ref = pmap_is_referenced(pg);
1678 	mod = pmap_is_modified(pg);
1679 	printf("Modified page: ref %d, mod %d\n",
1680 	       ref, mod);
1681 
1682 	/* Now clear reference and modify */
1683 	ref = pmap_clear_reference(pg);
1684 	mod = pmap_clear_modify(pg);
1685 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1686 	       (void *)(u_long)va, (long)pa,
1687 	       ref, mod);
1688 
1689 	/* Check it's properly cleared */
1690 	ref = pmap_is_referenced(pg);
1691 	mod = pmap_is_modified(pg);
1692 	printf("Checking cleared page: ref %d, mod %d\n",
1693 	       ref, mod);
1694 
1695 	/* Modify page */
1696 	*loc = 1;
1697 
1698 	ref = pmap_is_referenced(pg);
1699 	mod = pmap_is_modified(pg);
1700 	printf("Modified page: ref %d, mod %d\n",
1701 	       ref, mod);
1702 
1703 	/* Check pmap_protect() */
1704 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
1705 	pmap_update(pmap_kernel());
1706 	ref = pmap_is_referenced(pg);
1707 	mod = pmap_is_modified(pg);
1708 	printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
1709 	       ref, mod);
1710 
1711 	/* Now clear reference and modify */
1712 	ref = pmap_clear_reference(pg);
1713 	mod = pmap_clear_modify(pg);
1714 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1715 	       (void *)(u_long)va, (long)pa,
1716 	       ref, mod);
1717 
1718 	/* Reference page */
1719 	val = *loc;
1720 
1721 	ref = pmap_is_referenced(pg);
1722 	mod = pmap_is_modified(pg);
1723 	printf("Referenced page: ref %d, mod %d val %x\n",
1724 	       ref, mod, val);
1725 
1726 	/* Now clear reference and modify */
1727 	ref = pmap_clear_reference(pg);
1728 	mod = pmap_clear_modify(pg);
1729 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1730 	       (void *)(u_long)va, (long)pa,
1731 	       ref, mod);
1732 
1733 	/* Modify page */
1734 #if 0
1735 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1736 	pmap_update(pmap_kernel());
1737 #endif
1738 	*loc = 1;
1739 
1740 	ref = pmap_is_referenced(pg);
1741 	mod = pmap_is_modified(pg);
1742 	printf("Modified page: ref %d, mod %d\n",
1743 	       ref, mod);
1744 
1745 	/* Check pmap_protect() */
1746 	pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
1747 	pmap_update(pmap_kernel());
1748 	ref = pmap_is_referenced(pg);
1749 	mod = pmap_is_modified(pg);
1750 	printf("pmap_protect(): ref %d, mod %d\n",
1751 	       ref, mod);
1752 
1753 	/* Now clear reference and modify */
1754 	ref = pmap_clear_reference(pg);
1755 	mod = pmap_clear_modify(pg);
1756 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1757 	       (void *)(u_long)va, (long)pa,
1758 	       ref, mod);
1759 
1760 	/* Reference page */
1761 	val = *loc;
1762 
1763 	ref = pmap_is_referenced(pg);
1764 	mod = pmap_is_modified(pg);
1765 	printf("Referenced page: ref %d, mod %d val %x\n",
1766 	       ref, mod, val);
1767 
1768 	/* Now clear reference and modify */
1769 	ref = pmap_clear_reference(pg);
1770 	mod = pmap_clear_modify(pg);
1771 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1772 	       (void *)(u_long)va, (long)pa,
1773 	       ref, mod);
1774 
1775 	/* Modify page */
1776 #if 0
1777 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1778 	pmap_update(pmap_kernel());
1779 #endif
1780 	*loc = 1;
1781 
1782 	ref = pmap_is_referenced(pg);
1783 	mod = pmap_is_modified(pg);
1784 	printf("Modified page: ref %d, mod %d\n",
1785 	       ref, mod);
1786 
1787 	/* Check pmap_pag_protect() */
1788 	pmap_page_protect(pg, VM_PROT_READ);
1789 	ref = pmap_is_referenced(pg);
1790 	mod = pmap_is_modified(pg);
1791 	printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
1792 	       ref, mod);
1793 
1794 	/* Now clear reference and modify */
1795 	ref = pmap_clear_reference(pg);
1796 	mod = pmap_clear_modify(pg);
1797 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1798 	       (void *)(u_long)va, (long)pa,
1799 	       ref, mod);
1800 
1801 	/* Reference page */
1802 	val = *loc;
1803 
1804 	ref = pmap_is_referenced(pg);
1805 	mod = pmap_is_modified(pg);
1806 	printf("Referenced page: ref %d, mod %d val %x\n",
1807 	       ref, mod, val);
1808 
1809 	/* Now clear reference and modify */
1810 	ref = pmap_clear_reference(pg);
1811 	mod = pmap_clear_modify(pg);
1812 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1813 	       (void *)(u_long)va, (long)pa,
1814 	       ref, mod);
1815 
1816 	/* Modify page */
1817 #if 0
1818 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1819 	pmap_update(pmap_kernel());
1820 #endif
1821 	*loc = 1;
1822 
1823 	ref = pmap_is_referenced(pg);
1824 	mod = pmap_is_modified(pg);
1825 	printf("Modified page: ref %d, mod %d\n",
1826 	       ref, mod);
1827 
1828 	/* Check pmap_pag_protect() */
1829 	pmap_page_protect(pg, VM_PROT_NONE);
1830 	ref = pmap_is_referenced(pg);
1831 	mod = pmap_is_modified(pg);
1832 	printf("pmap_page_protect(): ref %d, mod %d\n",
1833 	       ref, mod);
1834 
1835 	/* Now clear reference and modify */
1836 	ref = pmap_clear_reference(pg);
1837 	mod = pmap_clear_modify(pg);
1838 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1839 	       (void *)(u_long)va, (long)pa,
1840 	       ref, mod);
1841 
1842 
1843 	/* Reference page */
1844 	val = *loc;
1845 
1846 	ref = pmap_is_referenced(pg);
1847 	mod = pmap_is_modified(pg);
1848 	printf("Referenced page: ref %d, mod %d val %x\n",
1849 	       ref, mod, val);
1850 
1851 	/* Now clear reference and modify */
1852 	ref = pmap_clear_reference(pg);
1853 	mod = pmap_clear_modify(pg);
1854 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1855 	       (void *)(u_long)va, (long)pa,
1856 	       ref, mod);
1857 
1858 	/* Modify page */
1859 #if 0
1860 	pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1861 	pmap_update(pmap_kernel());
1862 #endif
1863 	*loc = 1;
1864 
1865 	ref = pmap_is_referenced(pg);
1866 	mod = pmap_is_modified(pg);
1867 	printf("Modified page: ref %d, mod %d\n",
1868 	       ref, mod);
1869 
1870 	/* Unmap page */
1871 	pmap_remove(pmap_kernel(), va, va+1);
1872 	pmap_update(pmap_kernel());
1873 	ref = pmap_is_referenced(pg);
1874 	mod = pmap_is_modified(pg);
1875 	printf("Unmapped page: ref %d, mod %d\n", ref, mod);
1876 
1877 	/* Now clear reference and modify */
1878 	ref = pmap_clear_reference(pg);
1879 	mod = pmap_clear_modify(pg);
1880 	printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1881 	       (void *)(u_long)va, (long)pa, ref, mod);
1882 
1883 	/* Check it's properly cleared */
1884 	ref = pmap_is_referenced(pg);
1885 	mod = pmap_is_modified(pg);
1886 	printf("Checking cleared page: ref %d, mod %d\n",
1887 	       ref, mod);
1888 
1889 	pmap_remove(pmap_kernel(), va, va + PAGE_SIZE);
1890 	pmap_kenter_pa(va, pa, VM_PROT_ALL, 0);
1891 	uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE, UVM_KMF_WIRED);
1892 }
1893 #endif
1894