pmap.c revision 1.48 1 /* $NetBSD: pmap.c,v 1.48 2007/03/12 18:18:26 ad 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 1.48 2007/03/12 18:18:26 ad Exp $");
71
72 #include <sys/param.h>
73 #include <sys/malloc.h>
74 #include <sys/proc.h>
75 #include <sys/user.h>
76 #include <sys/queue.h>
77 #include <sys/systm.h>
78 #include <sys/pool.h>
79 #include <sys/device.h>
80
81 #include <uvm/uvm.h>
82
83 #include <machine/cpu.h>
84 #include <machine/pcb.h>
85 #include <machine/powerpc.h>
86
87 #include <powerpc/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
132 int physmem;
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, bool);
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[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[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[bank].end - vm_physmem[bank].start;
476 vm_physmem[bank].pmseg.pvent = pv;
477 vm_physmem[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 __P((void));
521 extern void vm_page_free1 __P((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 * Garbage collects the physical map system for
678 * pages which are no longer used.
679 * Success need not be guaranteed -- that is, there
680 * may well be pages which are not referenced, but
681 * others may be collected.
682 * Called by the pageout daemon when pages are scarce.
683 */
684 void
685 pmap_collect(struct pmap *pm)
686 {
687 }
688
689 /*
690 * Fill the given physical page with zeroes.
691 */
692 void
693 pmap_zero_page(paddr_t pa)
694 {
695
696 #ifdef PPC_4XX_NOCACHE
697 memset((void *)pa, 0, PAGE_SIZE);
698 #else
699 int i;
700
701 for (i = PAGE_SIZE/CACHELINESIZE; i > 0; i--) {
702 __asm volatile ("dcbz 0,%0" :: "r"(pa));
703 pa += CACHELINESIZE;
704 }
705 #endif
706 }
707
708 /*
709 * Copy the given physical source page to its destination.
710 */
711 void
712 pmap_copy_page(paddr_t src, paddr_t dst)
713 {
714
715 memcpy((void *)dst, (void *)src, PAGE_SIZE);
716 dcache_flush_page(dst);
717 }
718
719 /*
720 * This returns whether this is the first mapping of a page.
721 */
722 static inline int
723 pmap_enter_pv(struct pmap *pm, vaddr_t va, paddr_t pa, bool wired)
724 {
725 struct pv_entry *pv, *npv = NULL;
726 int s;
727
728 if (!pmap_initialized)
729 return 0;
730
731 s = splvm();
732 pv = pa_to_pv(pa);
733 if (!pv->pv_pm) {
734 /*
735 * No entries yet, use header as the first entry.
736 */
737 pv->pv_va = va;
738 pv->pv_pm = pm;
739 pv->pv_next = NULL;
740 } else {
741 /*
742 * There is at least one other VA mapping this page.
743 * Place this entry after the header.
744 */
745 npv = pool_get(&pv_pool, PR_WAITOK);
746 npv->pv_va = va;
747 npv->pv_pm = pm;
748 npv->pv_next = pv->pv_next;
749 pv->pv_next = npv;
750 pv = npv;
751 }
752 if (wired) {
753 PV_WIRE(pv);
754 pm->pm_stats.wired_count++;
755 }
756 splx(s);
757 return (1);
758 }
759
760 static void
761 pmap_remove_pv(struct pmap *pm, vaddr_t va, paddr_t pa)
762 {
763 struct pv_entry *pv, *npv;
764
765 /*
766 * Remove from the PV table.
767 */
768 pv = pa_to_pv(pa);
769 if (!pv)
770 return;
771
772 /*
773 * If it is the first entry on the list, it is actually
774 * in the header and we must copy the following entry up
775 * to the header. Otherwise we must search the list for
776 * the entry. In either case we free the now unused entry.
777 */
778 if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
779 if (PV_ISWIRED(pv)) {
780 pm->pm_stats.wired_count--;
781 }
782 if ((npv = pv->pv_next)) {
783 *pv = *npv;
784 pool_put(&pv_pool, npv);
785 } else
786 pv->pv_pm = NULL;
787 } else {
788 for (; (npv = pv->pv_next) != NULL; pv = npv)
789 if (pm == npv->pv_pm && PV_CMPVA(va, npv))
790 break;
791 if (npv) {
792 pv->pv_next = npv->pv_next;
793 if (PV_ISWIRED(npv)) {
794 pm->pm_stats.wired_count--;
795 }
796 pool_put(&pv_pool, npv);
797 }
798 }
799 }
800
801 /*
802 * Insert physical page at pa into the given pmap at virtual address va.
803 */
804 int
805 pmap_enter(struct pmap *pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags)
806 {
807 int s;
808 u_int tte;
809 int managed;
810
811 /*
812 * Have to remove any existing mapping first.
813 */
814 pmap_remove(pm, va, va + PAGE_SIZE);
815
816 if (flags & PMAP_WIRED)
817 flags |= prot;
818
819 managed = 0;
820 if (vm_physseg_find(atop(pa), NULL) != -1)
821 managed = 1;
822
823 /*
824 * Generate TTE.
825 */
826 tte = TTE_PA(pa);
827 /* XXXX -- need to support multiple page sizes. */
828 tte |= TTE_SZ_16K;
829 #ifdef DIAGNOSTIC
830 if ((flags & (PME_NOCACHE | PME_WRITETHROUG)) ==
831 (PME_NOCACHE | PME_WRITETHROUG))
832 panic("pmap_enter: uncached & writethrough");
833 #endif
834 if (flags & PME_NOCACHE)
835 /* Must be I/O mapping */
836 tte |= TTE_I | TTE_G;
837 #ifdef PPC_4XX_NOCACHE
838 tte |= TTE_I;
839 #else
840 else if (flags & PME_WRITETHROUG)
841 /* Uncached and writethrough are not compatible */
842 tte |= TTE_W;
843 #endif
844 if (pm == pmap_kernel())
845 tte |= TTE_ZONE(ZONE_PRIV);
846 else
847 tte |= TTE_ZONE(ZONE_USER);
848
849 if (flags & VM_PROT_WRITE)
850 tte |= TTE_WR;
851
852 if (flags & VM_PROT_EXECUTE)
853 tte |= TTE_EX;
854
855 /*
856 * Now record mapping for later back-translation.
857 */
858 if (pmap_initialized && managed) {
859 char *attr;
860
861 if (!pmap_enter_pv(pm, va, pa, flags & PMAP_WIRED)) {
862 /* Could not enter pv on a managed page */
863 return 1;
864 }
865
866 /* Now set attributes. */
867 attr = pa_to_attr(pa);
868 #ifdef DIAGNOSTIC
869 if (!attr)
870 panic("managed but no attr");
871 #endif
872 if (flags & VM_PROT_ALL)
873 *attr |= PMAP_ATTR_REF;
874 if (flags & VM_PROT_WRITE)
875 *attr |= PMAP_ATTR_CHG;
876 }
877
878 s = splvm();
879
880 /* Insert page into page table. */
881 pte_enter(pm, va, tte);
882
883 /* If this is a real fault, enter it in the tlb */
884 if (tte && ((flags & PMAP_WIRED) == 0)) {
885 ppc4xx_tlb_enter(pm->pm_ctx, va, tte);
886 }
887 splx(s);
888
889 /* Flush the real memory from the instruction cache. */
890 if ((prot & VM_PROT_EXECUTE) && (tte & TTE_I) == 0)
891 __syncicache((void *)pa, PAGE_SIZE);
892
893 return 0;
894 }
895
896 void
897 pmap_unwire(struct pmap *pm, vaddr_t va)
898 {
899 struct pv_entry *pv;
900 paddr_t pa;
901 int s;
902
903 if (!pmap_extract(pm, va, &pa)) {
904 return;
905 }
906
907 pv = pa_to_pv(pa);
908 if (!pv)
909 return;
910
911 s = splvm();
912 while (pv != NULL) {
913 if (pm == pv->pv_pm && PV_CMPVA(va, pv)) {
914 if (PV_ISWIRED(pv)) {
915 PV_UNWIRE(pv);
916 pm->pm_stats.wired_count--;
917 }
918 break;
919 }
920 pv = pv->pv_next;
921 }
922 splx(s);
923 }
924
925 void
926 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot)
927 {
928 int s;
929 u_int tte;
930 struct pmap *pm = pmap_kernel();
931
932 /*
933 * Have to remove any existing mapping first.
934 */
935
936 /*
937 * Generate TTE.
938 *
939 * XXXX
940 *
941 * Since the kernel does not handle execution privileges properly,
942 * we will handle read and execute permissions together.
943 */
944 tte = 0;
945 if (prot & VM_PROT_ALL) {
946
947 tte = TTE_PA(pa) | TTE_EX | TTE_ZONE(ZONE_PRIV);
948 /* XXXX -- need to support multiple page sizes. */
949 tte |= TTE_SZ_16K;
950 #ifdef DIAGNOSTIC
951 if ((prot & (PME_NOCACHE | PME_WRITETHROUG)) ==
952 (PME_NOCACHE | PME_WRITETHROUG))
953 panic("pmap_kenter_pa: uncached & writethrough");
954 #endif
955 if (prot & PME_NOCACHE)
956 /* Must be I/O mapping */
957 tte |= TTE_I | TTE_G;
958 #ifdef PPC_4XX_NOCACHE
959 tte |= TTE_I;
960 #else
961 else if (prot & PME_WRITETHROUG)
962 /* Uncached and writethrough are not compatible */
963 tte |= TTE_W;
964 #endif
965 if (prot & VM_PROT_WRITE)
966 tte |= TTE_WR;
967 }
968
969 s = splvm();
970
971 /* Insert page into page table. */
972 pte_enter(pm, va, tte);
973 splx(s);
974 }
975
976 void
977 pmap_kremove(vaddr_t va, vsize_t len)
978 {
979
980 while (len > 0) {
981 pte_enter(pmap_kernel(), va, 0);
982 va += PAGE_SIZE;
983 len -= PAGE_SIZE;
984 }
985 }
986
987 /*
988 * Remove the given range of mapping entries.
989 */
990 void
991 pmap_remove(struct pmap *pm, vaddr_t va, vaddr_t endva)
992 {
993 int s;
994 paddr_t pa;
995 volatile u_int *ptp;
996
997 s = splvm();
998 while (va < endva) {
999
1000 if ((ptp = pte_find(pm, va)) && (pa = *ptp)) {
1001 pa = TTE_PA(pa);
1002 pmap_remove_pv(pm, va, pa);
1003 *ptp = 0;
1004 ppc4xx_tlb_flush(va, pm->pm_ctx);
1005 pm->pm_stats.resident_count--;
1006 }
1007 va += PAGE_SIZE;
1008 }
1009
1010 splx(s);
1011 }
1012
1013 /*
1014 * Get the physical page address for the given pmap/virtual address.
1015 */
1016 bool
1017 pmap_extract(struct pmap *pm, vaddr_t va, paddr_t *pap)
1018 {
1019 int seg = STIDX(va);
1020 int ptn = PTIDX(va);
1021 u_int pa = 0;
1022 int s;
1023
1024 s = splvm();
1025 if (pm->pm_ptbl[seg] && (pa = pm->pm_ptbl[seg][ptn])) {
1026 *pap = TTE_PA(pa) | (va & PGOFSET);
1027 }
1028 splx(s);
1029 return (pa != 0);
1030 }
1031
1032 /*
1033 * Lower the protection on the specified range of this pmap.
1034 *
1035 * There are only two cases: either the protection is going to 0,
1036 * or it is going to read-only.
1037 */
1038 void
1039 pmap_protect(struct pmap *pm, vaddr_t sva, vaddr_t eva, vm_prot_t prot)
1040 {
1041 volatile u_int *ptp;
1042 int s, bic;
1043
1044 if ((prot & VM_PROT_READ) == 0) {
1045 pmap_remove(pm, sva, eva);
1046 return;
1047 }
1048 bic = 0;
1049 if ((prot & VM_PROT_WRITE) == 0) {
1050 bic |= TTE_WR;
1051 }
1052 if ((prot & VM_PROT_EXECUTE) == 0) {
1053 bic |= TTE_EX;
1054 }
1055 if (bic == 0) {
1056 return;
1057 }
1058 s = splvm();
1059 while (sva < eva) {
1060 if ((ptp = pte_find(pm, sva)) != NULL) {
1061 *ptp &= ~bic;
1062 ppc4xx_tlb_flush(sva, pm->pm_ctx);
1063 }
1064 sva += PAGE_SIZE;
1065 }
1066 splx(s);
1067 }
1068
1069 bool
1070 pmap_check_attr(struct vm_page *pg, u_int mask, int clear)
1071 {
1072 paddr_t pa;
1073 char *attr;
1074 int s, rv;
1075
1076 /*
1077 * First modify bits in cache.
1078 */
1079 pa = VM_PAGE_TO_PHYS(pg);
1080 attr = pa_to_attr(pa);
1081 if (attr == NULL)
1082 return false;
1083
1084 s = splvm();
1085 rv = ((*attr & mask) != 0);
1086 if (clear) {
1087 *attr &= ~mask;
1088 pmap_page_protect(pg, mask == PMAP_ATTR_CHG ? VM_PROT_READ : 0);
1089 }
1090 splx(s);
1091 return rv;
1092 }
1093
1094
1095 /*
1096 * Lower the protection on the specified physical page.
1097 *
1098 * There are only two cases: either the protection is going to 0,
1099 * or it is going to read-only.
1100 */
1101 void
1102 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
1103 {
1104 paddr_t pa = VM_PAGE_TO_PHYS(pg);
1105 vaddr_t va;
1106 struct pv_entry *pvh, *pv, *npv;
1107 struct pmap *pm;
1108
1109 pvh = pa_to_pv(pa);
1110 if (pvh == NULL)
1111 return;
1112
1113 /* Handle extra pvs which may be deleted in the operation */
1114 for (pv = pvh->pv_next; pv; pv = npv) {
1115 npv = pv->pv_next;
1116
1117 pm = pv->pv_pm;
1118 va = pv->pv_va;
1119 pmap_protect(pm, va, va + PAGE_SIZE, prot);
1120 }
1121 /* Now check the head pv */
1122 if (pvh->pv_pm) {
1123 pv = pvh;
1124 pm = pv->pv_pm;
1125 va = pv->pv_va;
1126 pmap_protect(pm, va, va + PAGE_SIZE, prot);
1127 }
1128 }
1129
1130 /*
1131 * Activate the address space for the specified process. If the process
1132 * is the current process, load the new MMU context.
1133 */
1134 void
1135 pmap_activate(struct lwp *l)
1136 {
1137 #if 0
1138 struct pcb *pcb = &l->l_proc->p_addr->u_pcb;
1139 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
1140
1141 /*
1142 * XXX Normally performed in cpu_fork().
1143 */
1144 printf("pmap_activate(%p), pmap=%p\n",l,pmap);
1145 pcb->pcb_pm = pmap;
1146 #endif
1147 }
1148
1149 /*
1150 * Deactivate the specified process's address space.
1151 */
1152 void
1153 pmap_deactivate(struct lwp *l)
1154 {
1155 }
1156
1157 /*
1158 * Synchronize caches corresponding to [addr, addr+len) in p.
1159 */
1160 void
1161 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
1162 {
1163 struct pmap *pm = p->p_vmspace->vm_map.pmap;
1164 int msr, ctx, opid, step;
1165
1166 step = CACHELINESIZE;
1167
1168 /*
1169 * Need to turn off IMMU and switch to user context.
1170 * (icbi uses DMMU).
1171 */
1172 if (!(ctx = pm->pm_ctx)) {
1173 /* No context -- assign it one */
1174 ctx_alloc(pm);
1175 ctx = pm->pm_ctx;
1176 }
1177 __asm volatile("mfmsr %0;"
1178 "li %1, %7;"
1179 "andc %1,%0,%1;"
1180 "mtmsr %1;"
1181 "sync;isync;"
1182 "mfpid %1;"
1183 "mtpid %2;"
1184 "sync; isync;"
1185 "1:"
1186 "dcbf 0,%3;"
1187 "icbi 0,%3;"
1188 "add %3,%3,%5;"
1189 "addc. %4,%4,%6;"
1190 "bge 1b;"
1191 "mtpid %1;"
1192 "mtmsr %0;"
1193 "sync; isync"
1194 : "=&r" (msr), "=&r" (opid)
1195 : "r" (ctx), "r" (va), "r" (len), "r" (step), "r" (-step),
1196 "K" (PSL_IR | PSL_DR));
1197 }
1198
1199
1200 /* This has to be done in real mode !!! */
1201 void
1202 ppc4xx_tlb_flush(vaddr_t va, int pid)
1203 {
1204 u_long i, found;
1205 u_long msr;
1206
1207 /* If there's no context then it can't be mapped. */
1208 if (!pid)
1209 return;
1210
1211 __asm( "mfpid %1;" /* Save PID */
1212 "mfmsr %2;" /* Save MSR */
1213 "li %0,0;" /* Now clear MSR */
1214 "mtmsr %0;"
1215 "mtpid %4;" /* Set PID */
1216 "sync;"
1217 "tlbsx. %0,0,%3;" /* Search TLB */
1218 "sync;"
1219 "mtpid %1;" /* Restore PID */
1220 "mtmsr %2;" /* Restore MSR */
1221 "sync;isync;"
1222 "li %1,1;"
1223 "beq 1f;"
1224 "li %1,0;"
1225 "1:"
1226 : "=&r" (i), "=&r" (found), "=&r" (msr)
1227 : "r" (va), "r" (pid));
1228 if (found && !TLB_LOCKED(i)) {
1229
1230 /* Now flush translation */
1231 __asm volatile(
1232 "tlbwe %0,%1,0;"
1233 "sync;isync;"
1234 : : "r" (0), "r" (i));
1235
1236 tlb_info[i].ti_ctx = 0;
1237 tlb_info[i].ti_flags = 0;
1238 tlbnext = i;
1239 /* Successful flushes */
1240 tlbflush_ev.ev_count++;
1241 }
1242 }
1243
1244 void
1245 ppc4xx_tlb_flush_all(void)
1246 {
1247 u_long i;
1248
1249 for (i = 0; i < NTLB; i++)
1250 if (!TLB_LOCKED(i)) {
1251 __asm volatile(
1252 "tlbwe %0,%1,0;"
1253 "sync;isync;"
1254 : : "r" (0), "r" (i));
1255 tlb_info[i].ti_ctx = 0;
1256 tlb_info[i].ti_flags = 0;
1257 }
1258
1259 __asm volatile("sync;isync");
1260 }
1261
1262 /* Find a TLB entry to evict. */
1263 static int
1264 ppc4xx_tlb_find_victim(void)
1265 {
1266 int flags;
1267
1268 for (;;) {
1269 if (++tlbnext >= NTLB)
1270 tlbnext = tlb_nreserved;
1271 flags = tlb_info[tlbnext].ti_flags;
1272 if (!(flags & TLBF_USED) ||
1273 (flags & (TLBF_LOCKED | TLBF_REF)) == 0) {
1274 u_long va, stack = (u_long)&va;
1275
1276 if (!((tlb_info[tlbnext].ti_va ^ stack) & (~PGOFSET)) &&
1277 (tlb_info[tlbnext].ti_ctx == KERNEL_PID) &&
1278 (flags & TLBF_USED)) {
1279 /* Kernel stack page */
1280 flags |= TLBF_USED;
1281 tlb_info[tlbnext].ti_flags = flags;
1282 } else {
1283 /* Found it! */
1284 return (tlbnext);
1285 }
1286 } else {
1287 tlb_info[tlbnext].ti_flags = (flags & ~TLBF_REF);
1288 }
1289 }
1290 }
1291
1292 void
1293 ppc4xx_tlb_enter(int ctx, vaddr_t va, u_int pte)
1294 {
1295 u_long th, tl, idx;
1296 tlbpid_t pid;
1297 u_short msr;
1298 paddr_t pa;
1299 int s, sz;
1300
1301 tlbenter_ev.ev_count++;
1302
1303 sz = (pte & TTE_SZ_MASK) >> TTE_SZ_SHIFT;
1304 pa = (pte & TTE_RPN_MASK(sz));
1305 th = (va & TLB_EPN_MASK) | (sz << TLB_SIZE_SHFT) | TLB_VALID;
1306 tl = (pte & ~TLB_RPN_MASK) | pa;
1307 tl |= ppc4xx_tlbflags(va, pa);
1308
1309 s = splhigh();
1310 idx = ppc4xx_tlb_find_victim();
1311
1312 #ifdef DIAGNOSTIC
1313 if ((idx < tlb_nreserved) || (idx >= NTLB)) {
1314 panic("ppc4xx_tlb_enter: replacing entry %ld", idx);
1315 }
1316 #endif
1317
1318 tlb_info[idx].ti_va = (va & TLB_EPN_MASK);
1319 tlb_info[idx].ti_ctx = ctx;
1320 tlb_info[idx].ti_flags = TLBF_USED | TLBF_REF;
1321
1322 __asm volatile(
1323 "mfmsr %0;" /* Save MSR */
1324 "li %1,0;"
1325 "tlbwe %1,%3,0;" /* Invalidate old entry. */
1326 "mtmsr %1;" /* Clear MSR */
1327 "mfpid %1;" /* Save old PID */
1328 "mtpid %2;" /* Load translation ctx */
1329 "sync; isync;"
1330 #ifdef DEBUG
1331 "andi. %3,%3,63;"
1332 "tweqi %3,0;" /* XXXXX DEBUG trap on index 0 */
1333 #endif
1334 "tlbwe %4,%3,1; tlbwe %5,%3,0;" /* Set TLB */
1335 "sync; isync;"
1336 "mtpid %1; mtmsr %0;" /* Restore PID and MSR */
1337 "sync; isync;"
1338 : "=&r" (msr), "=&r" (pid)
1339 : "r" (ctx), "r" (idx), "r" (tl), "r" (th));
1340 splx(s);
1341 }
1342
1343 void
1344 ppc4xx_tlb_init(void)
1345 {
1346 int i;
1347
1348 /* Mark reserved TLB entries */
1349 for (i = 0; i < tlb_nreserved; i++) {
1350 tlb_info[i].ti_flags = TLBF_LOCKED | TLBF_USED;
1351 tlb_info[i].ti_ctx = KERNEL_PID;
1352 }
1353
1354 /* Setup security zones */
1355 /* Z0 - accessible by kernel only if TLB entry permissions allow
1356 * Z1,Z2 - access is controlled by TLB entry permissions
1357 * Z3 - full access regardless of TLB entry permissions
1358 */
1359
1360 __asm volatile(
1361 "mtspr %0,%1;"
1362 "sync;"
1363 :: "K"(SPR_ZPR), "r" (0x1b000000));
1364 }
1365
1366 /*
1367 * ppc4xx_tlb_size_mask:
1368 *
1369 * Roundup size to supported page size, return TLBHI mask and real size.
1370 */
1371 static int
1372 ppc4xx_tlb_size_mask(size_t size, int *mask, int *rsiz)
1373 {
1374 int i;
1375
1376 for (i = 0; i < __arraycount(tlbsize); i++)
1377 if (size <= tlbsize[i]) {
1378 *mask = (i << TLB_SIZE_SHFT);
1379 *rsiz = tlbsize[i];
1380 return (0);
1381 }
1382 return (EINVAL);
1383 }
1384
1385 /*
1386 * ppc4xx_tlb_mapiodev:
1387 *
1388 * Lookup virtual address of mapping previously entered via
1389 * ppc4xx_tlb_reserve. Search TLB directly so that we don't
1390 * need to waste extra storage for reserved mappings. Note
1391 * that reading TLBHI also sets PID, but all reserved mappings
1392 * use KERNEL_PID, so the side effect is nil.
1393 */
1394 void *
1395 ppc4xx_tlb_mapiodev(paddr_t base, psize_t len)
1396 {
1397 paddr_t pa;
1398 vaddr_t va;
1399 u_int lo, hi, sz;
1400 int i;
1401
1402 /* tlb_nreserved is only allowed to grow, so this is safe. */
1403 for (i = 0; i < tlb_nreserved; i++) {
1404 __asm volatile (
1405 " tlbre %0,%2,1 \n" /* TLBLO */
1406 " tlbre %1,%2,0 \n" /* TLBHI */
1407 : "=&r" (lo), "=&r" (hi)
1408 : "r" (i));
1409
1410 KASSERT(hi & TLB_VALID);
1411 KASSERT(mfspr(SPR_PID) == KERNEL_PID);
1412
1413 pa = (lo & TLB_RPN_MASK);
1414 if (base < pa)
1415 continue;
1416
1417 sz = tlbsize[(hi & TLB_SIZE_MASK) >> TLB_SIZE_SHFT];
1418 if ((base + len) > (pa + sz))
1419 continue;
1420
1421 va = (hi & TLB_EPN_MASK) + (base & (sz - 1)); /* sz = 2^n */
1422 return (void *)(va);
1423 }
1424
1425 return (NULL);
1426 }
1427
1428 /*
1429 * ppc4xx_tlb_reserve:
1430 *
1431 * Map physical range to kernel virtual chunk via reserved TLB entry.
1432 */
1433 void
1434 ppc4xx_tlb_reserve(paddr_t pa, vaddr_t va, size_t size, int flags)
1435 {
1436 u_int lo, hi;
1437 int szmask, rsize;
1438
1439 /* Called before pmap_bootstrap(), va outside kernel space. */
1440 KASSERT(va < VM_MIN_KERNEL_ADDRESS || va >= VM_MAX_KERNEL_ADDRESS);
1441 KASSERT(! pmap_bootstrap_done);
1442 KASSERT(tlb_nreserved < NTLB);
1443
1444 /* Resolve size. */
1445 if (ppc4xx_tlb_size_mask(size, &szmask, &rsize) != 0)
1446 panic("ppc4xx_tlb_reserve: entry %d, %zuB too large",
1447 size, tlb_nreserved);
1448
1449 /* Real size will be power of two >= 1024, so this is OK. */
1450 pa &= ~(rsize - 1); /* RPN */
1451 va &= ~(rsize - 1); /* EPN */
1452
1453 lo = pa | TLB_WR | flags;
1454 hi = va | TLB_VALID | szmask;
1455
1456 #ifdef PPC_4XX_NOCACHE
1457 lo |= TLB_I;
1458 #endif
1459
1460 __asm volatile(
1461 " tlbwe %1,%0,1 \n" /* write TLBLO */
1462 " tlbwe %2,%0,0 \n" /* write TLBHI */
1463 " sync \n"
1464 " isync \n"
1465 : : "r" (tlb_nreserved), "r" (lo), "r" (hi));
1466
1467 tlb_nreserved++;
1468 }
1469
1470 /*
1471 * We should pass the ctx in from trap code.
1472 */
1473 int
1474 pmap_tlbmiss(vaddr_t va, int ctx)
1475 {
1476 volatile u_int *pte;
1477 u_long tte;
1478
1479 tlbmiss_ev.ev_count++;
1480
1481 /*
1482 * We will reserve 0 upto VM_MIN_KERNEL_ADDRESS for va == pa mappings.
1483 * Physical RAM is expected to live in this range, care must be taken
1484 * to not clobber 0 upto ${physmem} with device mappings in machdep
1485 * code.
1486 */
1487 if (ctx != KERNEL_PID || va >= VM_MIN_KERNEL_ADDRESS) {
1488 pte = pte_find((struct pmap *)__UNVOLATILE(ctxbusy[ctx]), va);
1489 if (pte == NULL) {
1490 /* Map unmanaged addresses directly for kernel access */
1491 return 1;
1492 }
1493 tte = *pte;
1494 if (tte == 0) {
1495 return 1;
1496 }
1497 } else {
1498 /* Create a 16MB writable mapping. */
1499 #ifdef PPC_4XX_NOCACHE
1500 tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_I |TTE_WR;
1501 #else
1502 tte = TTE_PA(va) | TTE_ZONE(ZONE_PRIV) | TTE_SZ_16M | TTE_WR;
1503 #endif
1504 }
1505 tlbhit_ev.ev_count++;
1506 ppc4xx_tlb_enter(ctx, va, tte);
1507
1508 return 0;
1509 }
1510
1511 /*
1512 * Flush all the entries matching a context from the TLB.
1513 */
1514 static int
1515 ctx_flush(int cnum)
1516 {
1517 int i;
1518
1519 /* We gotta steal this context */
1520 for (i = tlb_nreserved; i < NTLB; i++) {
1521 if (tlb_info[i].ti_ctx == cnum) {
1522 /* Can't steal ctx if it has a locked entry. */
1523 if (TLB_LOCKED(i)) {
1524 #ifdef DIAGNOSTIC
1525 printf("ctx_flush: can't invalidate "
1526 "locked mapping %d "
1527 "for context %d\n", i, cnum);
1528 #ifdef DDB
1529 Debugger();
1530 #endif
1531 #endif
1532 return (1);
1533 }
1534 #ifdef DIAGNOSTIC
1535 if (i < tlb_nreserved)
1536 panic("TLB entry %d not locked", i);
1537 #endif
1538 /* Invalidate particular TLB entry regardless of locked status */
1539 __asm volatile("tlbwe %0,%1,0" : :"r"(0),"r"(i));
1540 tlb_info[i].ti_flags = 0;
1541 }
1542 }
1543 return (0);
1544 }
1545
1546 /*
1547 * Allocate a context. If necessary, steal one from someone else.
1548 *
1549 * The new context is flushed from the TLB before returning.
1550 */
1551 int
1552 ctx_alloc(struct pmap *pm)
1553 {
1554 int s, cnum;
1555 static int next = MINCTX;
1556
1557 if (pm == pmap_kernel()) {
1558 #ifdef DIAGNOSTIC
1559 printf("ctx_alloc: kernel pmap!\n");
1560 #endif
1561 return (0);
1562 }
1563 s = splvm();
1564
1565 /* Find a likely context. */
1566 cnum = next;
1567 do {
1568 if ((++cnum) > NUMCTX)
1569 cnum = MINCTX;
1570 } while (ctxbusy[cnum] != NULL && cnum != next);
1571
1572 /* Now clean it out */
1573 oops:
1574 if (cnum < MINCTX)
1575 cnum = MINCTX; /* Never steal ctx 0 or 1 */
1576 if (ctx_flush(cnum)) {
1577 /* oops -- something's wired. */
1578 if ((++cnum) > NUMCTX)
1579 cnum = MINCTX;
1580 goto oops;
1581 }
1582
1583 if (ctxbusy[cnum]) {
1584 #ifdef DEBUG
1585 /* We should identify this pmap and clear it */
1586 printf("Warning: stealing context %d\n", cnum);
1587 #endif
1588 ctxbusy[cnum]->pm_ctx = 0;
1589 }
1590 ctxbusy[cnum] = pm;
1591 next = cnum;
1592 splx(s);
1593 pm->pm_ctx = cnum;
1594
1595 return cnum;
1596 }
1597
1598 /*
1599 * Give away a context.
1600 */
1601 void
1602 ctx_free(struct pmap *pm)
1603 {
1604 int oldctx;
1605
1606 oldctx = pm->pm_ctx;
1607
1608 if (oldctx == 0)
1609 panic("ctx_free: freeing kernel context");
1610 #ifdef DIAGNOSTIC
1611 if (ctxbusy[oldctx] == 0)
1612 printf("ctx_free: freeing free context %d\n", oldctx);
1613 if (ctxbusy[oldctx] != pm) {
1614 printf("ctx_free: freeing someone esle's context\n "
1615 "ctxbusy[%d] = %p, pm->pm_ctx = %p\n",
1616 oldctx, (void *)(u_long)ctxbusy[oldctx], pm);
1617 #ifdef DDB
1618 Debugger();
1619 #endif
1620 }
1621 #endif
1622 /* We should verify it has not been stolen and reallocated... */
1623 ctxbusy[oldctx] = NULL;
1624 ctx_flush(oldctx);
1625 }
1626
1627
1628 #ifdef DEBUG
1629 /*
1630 * Test ref/modify handling.
1631 */
1632 void pmap_testout __P((void));
1633 void
1634 pmap_testout()
1635 {
1636 vaddr_t va;
1637 volatile int *loc;
1638 int val = 0;
1639 paddr_t pa;
1640 struct vm_page *pg;
1641 int ref, mod;
1642
1643 /* Allocate a page */
1644 va = (vaddr_t)uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
1645 UVM_KMF_WIRED | UVM_KMF_ZERO);
1646 loc = (int*)va;
1647
1648 pmap_extract(pmap_kernel(), va, &pa);
1649 pg = PHYS_TO_VM_PAGE(pa);
1650 pmap_unwire(pmap_kernel(), va);
1651
1652 pmap_kremove(va, PAGE_SIZE);
1653 pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1654 pmap_update(pmap_kernel());
1655
1656 /* Now clear reference and modify */
1657 ref = pmap_clear_reference(pg);
1658 mod = pmap_clear_modify(pg);
1659 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1660 (void *)(u_long)va, (long)pa,
1661 ref, mod);
1662
1663 /* Check it's properly cleared */
1664 ref = pmap_is_referenced(pg);
1665 mod = pmap_is_modified(pg);
1666 printf("Checking cleared page: ref %d, mod %d\n",
1667 ref, mod);
1668
1669 /* Reference page */
1670 val = *loc;
1671
1672 ref = pmap_is_referenced(pg);
1673 mod = pmap_is_modified(pg);
1674 printf("Referenced page: ref %d, mod %d val %x\n",
1675 ref, mod, val);
1676
1677 /* Now clear reference and modify */
1678 ref = pmap_clear_reference(pg);
1679 mod = pmap_clear_modify(pg);
1680 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1681 (void *)(u_long)va, (long)pa,
1682 ref, mod);
1683
1684 /* Modify page */
1685 *loc = 1;
1686
1687 ref = pmap_is_referenced(pg);
1688 mod = pmap_is_modified(pg);
1689 printf("Modified page: ref %d, mod %d\n",
1690 ref, mod);
1691
1692 /* Now clear reference and modify */
1693 ref = pmap_clear_reference(pg);
1694 mod = pmap_clear_modify(pg);
1695 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1696 (void *)(u_long)va, (long)pa,
1697 ref, mod);
1698
1699 /* Check it's properly cleared */
1700 ref = pmap_is_referenced(pg);
1701 mod = pmap_is_modified(pg);
1702 printf("Checking cleared page: ref %d, mod %d\n",
1703 ref, mod);
1704
1705 /* Modify page */
1706 *loc = 1;
1707
1708 ref = pmap_is_referenced(pg);
1709 mod = pmap_is_modified(pg);
1710 printf("Modified page: ref %d, mod %d\n",
1711 ref, mod);
1712
1713 /* Check pmap_protect() */
1714 pmap_protect(pmap_kernel(), va, va+1, VM_PROT_READ);
1715 pmap_update(pmap_kernel());
1716 ref = pmap_is_referenced(pg);
1717 mod = pmap_is_modified(pg);
1718 printf("pmap_protect(VM_PROT_READ): ref %d, mod %d\n",
1719 ref, mod);
1720
1721 /* Now clear reference and modify */
1722 ref = pmap_clear_reference(pg);
1723 mod = pmap_clear_modify(pg);
1724 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1725 (void *)(u_long)va, (long)pa,
1726 ref, mod);
1727
1728 /* Reference page */
1729 val = *loc;
1730
1731 ref = pmap_is_referenced(pg);
1732 mod = pmap_is_modified(pg);
1733 printf("Referenced page: ref %d, mod %d val %x\n",
1734 ref, mod, val);
1735
1736 /* Now clear reference and modify */
1737 ref = pmap_clear_reference(pg);
1738 mod = pmap_clear_modify(pg);
1739 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1740 (void *)(u_long)va, (long)pa,
1741 ref, mod);
1742
1743 /* Modify page */
1744 #if 0
1745 pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1746 pmap_update(pmap_kernel());
1747 #endif
1748 *loc = 1;
1749
1750 ref = pmap_is_referenced(pg);
1751 mod = pmap_is_modified(pg);
1752 printf("Modified page: ref %d, mod %d\n",
1753 ref, mod);
1754
1755 /* Check pmap_protect() */
1756 pmap_protect(pmap_kernel(), va, va+1, VM_PROT_NONE);
1757 pmap_update(pmap_kernel());
1758 ref = pmap_is_referenced(pg);
1759 mod = pmap_is_modified(pg);
1760 printf("pmap_protect(): ref %d, mod %d\n",
1761 ref, mod);
1762
1763 /* Now clear reference and modify */
1764 ref = pmap_clear_reference(pg);
1765 mod = pmap_clear_modify(pg);
1766 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1767 (void *)(u_long)va, (long)pa,
1768 ref, mod);
1769
1770 /* Reference page */
1771 val = *loc;
1772
1773 ref = pmap_is_referenced(pg);
1774 mod = pmap_is_modified(pg);
1775 printf("Referenced page: ref %d, mod %d val %x\n",
1776 ref, mod, val);
1777
1778 /* Now clear reference and modify */
1779 ref = pmap_clear_reference(pg);
1780 mod = pmap_clear_modify(pg);
1781 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1782 (void *)(u_long)va, (long)pa,
1783 ref, mod);
1784
1785 /* Modify page */
1786 #if 0
1787 pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1788 pmap_update(pmap_kernel());
1789 #endif
1790 *loc = 1;
1791
1792 ref = pmap_is_referenced(pg);
1793 mod = pmap_is_modified(pg);
1794 printf("Modified page: ref %d, mod %d\n",
1795 ref, mod);
1796
1797 /* Check pmap_pag_protect() */
1798 pmap_page_protect(pg, VM_PROT_READ);
1799 ref = pmap_is_referenced(pg);
1800 mod = pmap_is_modified(pg);
1801 printf("pmap_page_protect(VM_PROT_READ): ref %d, mod %d\n",
1802 ref, mod);
1803
1804 /* Now clear reference and modify */
1805 ref = pmap_clear_reference(pg);
1806 mod = pmap_clear_modify(pg);
1807 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1808 (void *)(u_long)va, (long)pa,
1809 ref, mod);
1810
1811 /* Reference page */
1812 val = *loc;
1813
1814 ref = pmap_is_referenced(pg);
1815 mod = pmap_is_modified(pg);
1816 printf("Referenced page: ref %d, mod %d val %x\n",
1817 ref, mod, val);
1818
1819 /* Now clear reference and modify */
1820 ref = pmap_clear_reference(pg);
1821 mod = pmap_clear_modify(pg);
1822 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1823 (void *)(u_long)va, (long)pa,
1824 ref, mod);
1825
1826 /* Modify page */
1827 #if 0
1828 pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1829 pmap_update(pmap_kernel());
1830 #endif
1831 *loc = 1;
1832
1833 ref = pmap_is_referenced(pg);
1834 mod = pmap_is_modified(pg);
1835 printf("Modified page: ref %d, mod %d\n",
1836 ref, mod);
1837
1838 /* Check pmap_pag_protect() */
1839 pmap_page_protect(pg, VM_PROT_NONE);
1840 ref = pmap_is_referenced(pg);
1841 mod = pmap_is_modified(pg);
1842 printf("pmap_page_protect(): ref %d, mod %d\n",
1843 ref, mod);
1844
1845 /* Now clear reference and modify */
1846 ref = pmap_clear_reference(pg);
1847 mod = pmap_clear_modify(pg);
1848 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1849 (void *)(u_long)va, (long)pa,
1850 ref, mod);
1851
1852
1853 /* Reference page */
1854 val = *loc;
1855
1856 ref = pmap_is_referenced(pg);
1857 mod = pmap_is_modified(pg);
1858 printf("Referenced page: ref %d, mod %d val %x\n",
1859 ref, mod, val);
1860
1861 /* Now clear reference and modify */
1862 ref = pmap_clear_reference(pg);
1863 mod = pmap_clear_modify(pg);
1864 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1865 (void *)(u_long)va, (long)pa,
1866 ref, mod);
1867
1868 /* Modify page */
1869 #if 0
1870 pmap_enter(pmap_kernel(), va, pa, VM_PROT_ALL, 0);
1871 pmap_update(pmap_kernel());
1872 #endif
1873 *loc = 1;
1874
1875 ref = pmap_is_referenced(pg);
1876 mod = pmap_is_modified(pg);
1877 printf("Modified page: ref %d, mod %d\n",
1878 ref, mod);
1879
1880 /* Unmap page */
1881 pmap_remove(pmap_kernel(), va, va+1);
1882 pmap_update(pmap_kernel());
1883 ref = pmap_is_referenced(pg);
1884 mod = pmap_is_modified(pg);
1885 printf("Unmapped page: ref %d, mod %d\n", ref, mod);
1886
1887 /* Now clear reference and modify */
1888 ref = pmap_clear_reference(pg);
1889 mod = pmap_clear_modify(pg);
1890 printf("Clearing page va %p pa %lx: ref %d, mod %d\n",
1891 (void *)(u_long)va, (long)pa, ref, mod);
1892
1893 /* Check it's properly cleared */
1894 ref = pmap_is_referenced(pg);
1895 mod = pmap_is_modified(pg);
1896 printf("Checking cleared page: ref %d, mod %d\n",
1897 ref, mod);
1898
1899 pmap_remove(pmap_kernel(), va, va + PAGE_SIZE);
1900 pmap_kenter_pa(va, pa, VM_PROT_ALL);
1901 uvm_km_free(kernel_map, (vaddr_t)va, PAGE_SIZE, UVM_KMF_WIRED);
1902 }
1903 #endif
1904