MP Implmentation 3/4: MAJOR progress on SMP, full userland MP is now working!
[dragonfly.git] / sys / i386 / i386 / pmap.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1991 Regents of the University of California.
3 * All rights reserved.
4 * Copyright (c) 1994 John S. Dyson
5 * All rights reserved.
6 * Copyright (c) 1994 David Greenman
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * the Systems Programming Group of the University of Utah Computer
11 * Science Department and William Jolitz of UUNET Technologies Inc.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * SUCH DAMAGE.
40 *
41 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91
42 * $FreeBSD: src/sys/i386/i386/pmap.c,v 1.250.2.18 2002/03/06 22:48:53 silby Exp $
a2a5ad0d 43 * $DragonFly: src/sys/i386/i386/Attic/pmap.c,v 1.17 2003/07/10 04:47:53 dillon Exp $
984263bc
MD
44 */
45
46/*
47 * Manages physical address maps.
48 *
49 * In addition to hardware address maps, this
50 * module is called upon to provide software-use-only
51 * maps which may or may not be stored in the same
52 * form as hardware maps. These pseudo-maps are
53 * used to store intermediate results from copy
54 * operations to and from address spaces.
55 *
56 * Since the information managed by this module is
57 * also stored by the logical address mapping module,
58 * this module may throw away valid virtual-to-physical
59 * mappings at almost any time. However, invalidations
60 * of virtual-to-physical mappings must be done as
61 * requested.
62 *
63 * In order to cope with hardware architectures which
64 * make virtual-to-physical map invalidates expensive,
65 * this module may delay invalidate or reduced protection
66 * operations until such time as they are actually
67 * necessary. This module is given full information as
68 * to which processors are currently using which maps,
69 * and to when physical maps must be made correct.
70 */
71
72#include "opt_disable_pse.h"
73#include "opt_pmap.h"
74#include "opt_msgbuf.h"
75#include "opt_user_ldt.h"
76
77#include <sys/param.h>
78#include <sys/systm.h>
79#include <sys/kernel.h>
80#include <sys/proc.h>
81#include <sys/msgbuf.h>
82#include <sys/vmmeter.h>
83#include <sys/mman.h>
84
85#include <vm/vm.h>
86#include <vm/vm_param.h>
87#include <sys/sysctl.h>
88#include <sys/lock.h>
89#include <vm/vm_kern.h>
90#include <vm/vm_page.h>
91#include <vm/vm_map.h>
92#include <vm/vm_object.h>
93#include <vm/vm_extern.h>
94#include <vm/vm_pageout.h>
95#include <vm/vm_pager.h>
96#include <vm/vm_zone.h>
97
98#include <sys/user.h>
99
100#include <machine/cputypes.h>
101#include <machine/md_var.h>
102#include <machine/specialreg.h>
103#if defined(SMP) || defined(APIC_IO)
104#include <machine/smp.h>
105#include <machine/apic.h>
984263bc 106#endif /* SMP || APIC_IO */
85100692 107#include <machine/globaldata.h>
984263bc
MD
108
109#define PMAP_KEEP_PDIRS
110#ifndef PMAP_SHPGPERPROC
111#define PMAP_SHPGPERPROC 200
112#endif
113
114#if defined(DIAGNOSTIC)
115#define PMAP_DIAGNOSTIC
116#endif
117
118#define MINPV 2048
119
120#if !defined(PMAP_DIAGNOSTIC)
121#define PMAP_INLINE __inline
122#else
123#define PMAP_INLINE
124#endif
125
126/*
127 * Get PDEs and PTEs for user/kernel address space
128 */
129#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
130#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
131
132#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0)
133#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0)
134#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0)
135#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0)
136#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0)
137
138#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
139#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
140
141/*
142 * Given a map and a machine independent protection code,
143 * convert to a vax protection code.
144 */
145#define pte_prot(m, p) (protection_codes[p])
146static int protection_codes[8];
147
148static struct pmap kernel_pmap_store;
149pmap_t kernel_pmap;
150
151vm_offset_t avail_start; /* PA of first available physical page */
152vm_offset_t avail_end; /* PA of last available physical page */
153vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */
154vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */
155static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
156static int pgeflag; /* PG_G or-in */
157static int pseflag; /* PG_PS or-in */
158
159static vm_object_t kptobj;
160
161static int nkpt;
162vm_offset_t kernel_vm_end;
163
164/*
165 * Data for the pv entry allocation mechanism
166 */
167static vm_zone_t pvzone;
168static struct vm_zone pvzone_store;
169static struct vm_object pvzone_obj;
170static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
171static int pmap_pagedaemon_waken = 0;
172static struct pv_entry *pvinit;
173
174/*
175 * All those kernel PT submaps that BSD is so fond of
176 */
177pt_entry_t *CMAP1 = 0;
984263bc 178caddr_t CADDR1 = 0, ptvmmap = 0;
984263bc
MD
179static pt_entry_t *msgbufmap;
180struct msgbuf *msgbufp=0;
181
182/*
183 * Crashdump maps.
184 */
185static pt_entry_t *pt_crashdumpmap;
186static caddr_t crashdumpmap;
187
984263bc 188extern pt_entry_t *SMPpt;
984263bc
MD
189
190static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv));
191static unsigned * get_ptbase __P((pmap_t pmap));
192static pv_entry_t get_pv_entry __P((void));
193static void i386_protection_init __P((void));
194static __inline void pmap_changebit __P((vm_page_t m, int bit, boolean_t setem));
195
196static void pmap_remove_all __P((vm_page_t m));
197static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va,
198 vm_page_t m, vm_page_t mpte));
199static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq,
200 vm_offset_t sva));
201static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va));
202static int pmap_remove_entry __P((struct pmap *pmap, vm_page_t m,
203 vm_offset_t va));
204static boolean_t pmap_testbit __P((vm_page_t m, int bit));
205static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va,
206 vm_page_t mpte, vm_page_t m));
207
208static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va));
209
210static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p));
211static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex));
212static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va));
213static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex));
214static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t));
215static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
216
217static unsigned pdir4mb;
218
219/*
220 * Routine: pmap_pte
221 * Function:
222 * Extract the page table entry associated
223 * with the given map/virtual_address pair.
224 */
225
226PMAP_INLINE unsigned *
227pmap_pte(pmap, va)
228 register pmap_t pmap;
229 vm_offset_t va;
230{
231 unsigned *pdeaddr;
232
233 if (pmap) {
234 pdeaddr = (unsigned *) pmap_pde(pmap, va);
235 if (*pdeaddr & PG_PS)
236 return pdeaddr;
237 if (*pdeaddr) {
238 return get_ptbase(pmap) + i386_btop(va);
239 }
240 }
241 return (0);
242}
243
244/*
245 * Move the kernel virtual free pointer to the next
246 * 4MB. This is used to help improve performance
247 * by using a large (4MB) page for much of the kernel
248 * (.text, .data, .bss)
249 */
250static vm_offset_t
251pmap_kmem_choose(vm_offset_t addr)
252{
253 vm_offset_t newaddr = addr;
254#ifndef DISABLE_PSE
255 if (cpu_feature & CPUID_PSE) {
256 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
257 }
258#endif
259 return newaddr;
260}
261
262/*
263 * Bootstrap the system enough to run with virtual memory.
264 *
265 * On the i386 this is called after mapping has already been enabled
266 * and just syncs the pmap module with what has already been done.
267 * [We can't call it easily with mapping off since the kernel is not
268 * mapped with PA == VA, hence we would have to relocate every address
269 * from the linked base (virtual) address "KERNBASE" to the actual
270 * (physical) address starting relative to 0]
271 */
272void
273pmap_bootstrap(firstaddr, loadaddr)
274 vm_offset_t firstaddr;
275 vm_offset_t loadaddr;
276{
277 vm_offset_t va;
278 pt_entry_t *pte;
85100692 279 struct mdglobaldata *gd;
984263bc
MD
280 int i;
281
282 avail_start = firstaddr;
283
284 /*
285 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too
286 * large. It should instead be correctly calculated in locore.s and
287 * not based on 'first' (which is a physical address, not a virtual
288 * address, for the start of unused physical memory). The kernel
289 * page tables are NOT double mapped and thus should not be included
290 * in this calculation.
291 */
292 virtual_avail = (vm_offset_t) KERNBASE + firstaddr;
293 virtual_avail = pmap_kmem_choose(virtual_avail);
294
295 virtual_end = VM_MAX_KERNEL_ADDRESS;
296
297 /*
298 * Initialize protection array.
299 */
300 i386_protection_init();
301
302 /*
303 * The kernel's pmap is statically allocated so we don't have to use
304 * pmap_create, which is unlikely to work correctly at this part of
305 * the boot sequence (XXX and which no longer exists).
306 */
307 kernel_pmap = &kernel_pmap_store;
308
309 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD);
310 kernel_pmap->pm_count = 1;
311 kernel_pmap->pm_active = -1; /* don't allow deactivation */
312 TAILQ_INIT(&kernel_pmap->pm_pvlist);
313 nkpt = NKPT;
314
315 /*
316 * Reserve some special page table entries/VA space for temporary
317 * mapping of pages.
318 */
319#define SYSMAP(c, p, v, n) \
320 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
321
322 va = virtual_avail;
323 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va);
324
325 /*
326 * CMAP1/CMAP2 are used for zeroing and copying pages.
327 */
328 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
984263bc
MD
329
330 /*
331 * Crashdump maps.
332 */
333 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
334
984263bc
MD
335 /*
336 * msgbufp is used to map the system message buffer.
337 * XXX msgbufmap is not used.
338 */
339 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
340 atop(round_page(MSGBUF_SIZE)))
341
984263bc
MD
342 virtual_avail = va;
343
17a9f566 344 *(int *) CMAP1 = 0;
984263bc
MD
345 for (i = 0; i < NKPT; i++)
346 PTD[i] = 0;
347
a2a5ad0d
MD
348 /*
349 * PG_G is terribly broken on SMP because we IPI invltlb's in some
350 * cases rather then invl1pg. Actually, I don't even know why it
351 * works under UP because self-referential page table mappings
352 */
353#ifdef SMP
354 pgeflag = 0;
355#else
356 if (cpu_feature & CPUID_PGE)
984263bc 357 pgeflag = PG_G;
a2a5ad0d 358#endif
984263bc
MD
359
360/*
361 * Initialize the 4MB page size flag
362 */
363 pseflag = 0;
364/*
365 * The 4MB page version of the initial
366 * kernel page mapping.
367 */
368 pdir4mb = 0;
369
370#if !defined(DISABLE_PSE)
371 if (cpu_feature & CPUID_PSE) {
372 unsigned ptditmp;
373 /*
374 * Note that we have enabled PSE mode
375 */
376 pseflag = PG_PS;
377 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
378 ptditmp &= ~(NBPDR - 1);
379 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
380 pdir4mb = ptditmp;
381
8a8d5d85
MD
382#ifndef SMP
383 /*
384 * Enable the PSE mode. If we are SMP we can't do this
385 * now because the APs will not be able to use it when
386 * they boot up.
387 */
388 load_cr4(rcr4() | CR4_PSE);
984263bc 389
8a8d5d85
MD
390 /*
391 * We can do the mapping here for the single processor
392 * case. We simply ignore the old page table page from
393 * now on.
394 */
395 /*
396 * For SMP, we still need 4K pages to bootstrap APs,
397 * PSE will be enabled as soon as all APs are up.
398 */
399 PTD[KPTDI] = (pd_entry_t) ptditmp;
400 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp;
401 invltlb();
402#endif
984263bc
MD
403 }
404#endif
17a9f566 405#ifdef APIC_IO
984263bc
MD
406 if (cpu_apic_address == 0)
407 panic("pmap_bootstrap: no local apic!");
408
409 /* local apic is mapped on last page */
410 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
411 (cpu_apic_address & PG_FRAME));
17a9f566 412#endif
984263bc
MD
413
414 /* BSP does this itself, AP's get it pre-set */
85100692
MD
415 gd = &CPU_prvspace[0].mdglobaldata;
416 gd->gd_CMAP1 = &SMPpt[1];
417 gd->gd_CMAP2 = &SMPpt[2];
418 gd->gd_CMAP3 = &SMPpt[3];
419 gd->gd_PMAP1 = &SMPpt[4];
420 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
421 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
422 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
423 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
984263bc
MD
424
425 invltlb();
426}
427
428#ifdef SMP
429/*
430 * Set 4mb pdir for mp startup
431 */
432void
433pmap_set_opt(void)
434{
435 if (pseflag && (cpu_feature & CPUID_PSE)) {
436 load_cr4(rcr4() | CR4_PSE);
72740893 437 if (pdir4mb && mycpu->gd_cpuid == 0) { /* only on BSP */
984263bc
MD
438 kernel_pmap->pm_pdir[KPTDI] =
439 PTD[KPTDI] = (pd_entry_t)pdir4mb;
440 cpu_invltlb();
441 }
442 }
443}
444#endif
445
446/*
447 * Initialize the pmap module.
448 * Called by vm_init, to initialize any structures that the pmap
449 * system needs to map virtual memory.
450 * pmap_init has been enhanced to support in a fairly consistant
451 * way, discontiguous physical memory.
452 */
453void
454pmap_init(phys_start, phys_end)
455 vm_offset_t phys_start, phys_end;
456{
457 int i;
458 int initial_pvs;
459
460 /*
461 * object for kernel page table pages
462 */
463 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
464
465 /*
466 * Allocate memory for random pmap data structures. Includes the
467 * pv_head_table.
468 */
469
470 for(i = 0; i < vm_page_array_size; i++) {
471 vm_page_t m;
472
473 m = &vm_page_array[i];
474 TAILQ_INIT(&m->md.pv_list);
475 m->md.pv_list_count = 0;
476 }
477
478 /*
479 * init the pv free list
480 */
481 initial_pvs = vm_page_array_size;
482 if (initial_pvs < MINPV)
483 initial_pvs = MINPV;
484 pvzone = &pvzone_store;
485 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
486 initial_pvs * sizeof (struct pv_entry));
487 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
488 vm_page_array_size);
489
490 /*
491 * Now it is safe to enable pv_table recording.
492 */
493 pmap_initialized = TRUE;
494}
495
496/*
497 * Initialize the address space (zone) for the pv_entries. Set a
498 * high water mark so that the system can recover from excessive
499 * numbers of pv entries.
500 */
501void
502pmap_init2()
503{
504 int shpgperproc = PMAP_SHPGPERPROC;
505
506 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
507 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
508 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
509 pv_entry_high_water = 9 * (pv_entry_max / 10);
510 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
511}
512
513
514/***************************************************
515 * Low level helper routines.....
516 ***************************************************/
517
518#if defined(PMAP_DIAGNOSTIC)
519
520/*
521 * This code checks for non-writeable/modified pages.
522 * This should be an invalid condition.
523 */
524static int
525pmap_nw_modified(pt_entry_t ptea)
526{
527 int pte;
528
529 pte = (int) ptea;
530
531 if ((pte & (PG_M|PG_RW)) == PG_M)
532 return 1;
533 else
534 return 0;
535}
536#endif
537
538
539/*
540 * this routine defines the region(s) of memory that should
541 * not be tested for the modified bit.
542 */
543static PMAP_INLINE int
544pmap_track_modified(vm_offset_t va)
545{
546 if ((va < clean_sva) || (va >= clean_eva))
547 return 1;
548 else
549 return 0;
550}
551
552static PMAP_INLINE void
553invltlb_1pg(vm_offset_t va)
554{
555#if defined(I386_CPU)
556 if (cpu_class == CPUCLASS_386) {
557 invltlb();
558 } else
559#endif
560 {
561 invlpg(va);
562 }
563}
564
565static __inline void
566pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
567{
568#if defined(SMP)
72740893 569 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
984263bc 570 cpu_invlpg((void *)va);
72740893 571 if (pmap->pm_active & mycpu->gd_other_cpus)
984263bc
MD
572 smp_invltlb();
573#else
574 if (pmap->pm_active)
575 invltlb_1pg(va);
576#endif
577}
578
579static __inline void
580pmap_TLB_invalidate_all(pmap_t pmap)
581{
582#if defined(SMP)
72740893 583 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
984263bc 584 cpu_invltlb();
72740893 585 if (pmap->pm_active & mycpu->gd_other_cpus)
984263bc
MD
586 smp_invltlb();
587#else
588 if (pmap->pm_active)
589 invltlb();
590#endif
591}
592
593static unsigned *
594get_ptbase(pmap)
595 pmap_t pmap;
596{
597 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
598
599 /* are we current address space or kernel? */
600 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
601 return (unsigned *) PTmap;
602 }
603 /* otherwise, we are alternate address space */
604 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
605 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
606#if defined(SMP)
607 /* The page directory is not shared between CPUs */
608 cpu_invltlb();
609#else
610 invltlb();
611#endif
612 }
613 return (unsigned *) APTmap;
614}
615
616/*
617 * Super fast pmap_pte routine best used when scanning
618 * the pv lists. This eliminates many coarse-grained
619 * invltlb calls. Note that many of the pv list
620 * scans are across different pmaps. It is very wasteful
621 * to do an entire invltlb for checking a single mapping.
622 */
623
624static unsigned *
17a9f566 625pmap_pte_quick(pmap_t pmap, vm_offset_t va)
984263bc 626{
85100692 627 struct mdglobaldata *gd = mdcpu;
984263bc 628 unsigned pde, newpf;
17a9f566 629
984263bc
MD
630 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
631 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
632 unsigned index = i386_btop(va);
633 /* are we current address space or kernel? */
634 if ((pmap == kernel_pmap) ||
635 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
636 return (unsigned *) PTmap + index;
637 }
638 newpf = pde & PG_FRAME;
85100692
MD
639 if ( ((* (unsigned *) gd->gd_PMAP1) & PG_FRAME) != newpf) {
640 * (unsigned *) gd->gd_PMAP1 = newpf | PG_RW | PG_V;
641 cpu_invlpg(gd->gd_PADDR1);
984263bc 642 }
85100692 643 return gd->gd_PADDR1 + ((unsigned) index & (NPTEPG - 1));
984263bc
MD
644 }
645 return (0);
646}
647
648/*
649 * Routine: pmap_extract
650 * Function:
651 * Extract the physical page address associated
652 * with the given map/virtual_address pair.
653 */
654vm_offset_t
655pmap_extract(pmap, va)
656 register pmap_t pmap;
657 vm_offset_t va;
658{
659 vm_offset_t rtval;
660 vm_offset_t pdirindex;
661 pdirindex = va >> PDRSHIFT;
662 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
663 unsigned *pte;
664 if ((rtval & PG_PS) != 0) {
665 rtval &= ~(NBPDR - 1);
666 rtval |= va & (NBPDR - 1);
667 return rtval;
668 }
669 pte = get_ptbase(pmap) + i386_btop(va);
670 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
671 return rtval;
672 }
673 return 0;
674
675}
676
677/***************************************************
678 * Low level mapping routines.....
679 ***************************************************/
680
681/*
682 * add a wired page to the kva
683 * note that in order for the mapping to take effect -- you
684 * should do a invltlb after doing the pmap_kenter...
685 */
686PMAP_INLINE void
687pmap_kenter(va, pa)
688 vm_offset_t va;
689 register vm_offset_t pa;
690{
691 register unsigned *pte;
692 unsigned npte, opte;
693
694 npte = pa | PG_RW | PG_V | pgeflag;
695 pte = (unsigned *)vtopte(va);
696 opte = *pte;
697 *pte = npte;
698 invltlb_1pg(va);
699}
700
701/*
702 * remove a page from the kernel pagetables
703 */
704PMAP_INLINE void
705pmap_kremove(va)
706 vm_offset_t va;
707{
708 register unsigned *pte;
709
710 pte = (unsigned *)vtopte(va);
711 *pte = 0;
712 invltlb_1pg(va);
713}
714
715/*
716 * Used to map a range of physical addresses into kernel
717 * virtual address space.
718 *
719 * For now, VM is already on, we only need to map the
720 * specified memory.
721 */
722vm_offset_t
723pmap_map(virt, start, end, prot)
724 vm_offset_t virt;
725 vm_offset_t start;
726 vm_offset_t end;
727 int prot;
728{
729 while (start < end) {
730 pmap_kenter(virt, start);
731 virt += PAGE_SIZE;
732 start += PAGE_SIZE;
733 }
734 return (virt);
735}
736
737
738/*
739 * Add a list of wired pages to the kva
740 * this routine is only used for temporary
741 * kernel mappings that do not need to have
742 * page modification or references recorded.
743 * Note that old mappings are simply written
744 * over. The page *must* be wired.
745 */
746void
747pmap_qenter(va, m, count)
748 vm_offset_t va;
749 vm_page_t *m;
750 int count;
751{
752 vm_offset_t end_va;
753
754 end_va = va + count * PAGE_SIZE;
755
756 while (va < end_va) {
757 unsigned *pte;
758
759 pte = (unsigned *)vtopte(va);
760 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
761#ifdef SMP
762 cpu_invlpg((void *)va);
763#else
764 invltlb_1pg(va);
765#endif
766 va += PAGE_SIZE;
767 m++;
768 }
769#ifdef SMP
770 smp_invltlb();
771#endif
772}
773
774/*
775 * this routine jerks page mappings from the
776 * kernel -- it is meant only for temporary mappings.
777 */
778void
779pmap_qremove(va, count)
780 vm_offset_t va;
781 int count;
782{
783 vm_offset_t end_va;
784
785 end_va = va + count*PAGE_SIZE;
786
787 while (va < end_va) {
788 unsigned *pte;
789
790 pte = (unsigned *)vtopte(va);
791 *pte = 0;
792#ifdef SMP
793 cpu_invlpg((void *)va);
794#else
795 invltlb_1pg(va);
796#endif
797 va += PAGE_SIZE;
798 }
799#ifdef SMP
800 smp_invltlb();
801#endif
802}
803
804static vm_page_t
805pmap_page_lookup(object, pindex)
806 vm_object_t object;
807 vm_pindex_t pindex;
808{
809 vm_page_t m;
810retry:
811 m = vm_page_lookup(object, pindex);
812 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
813 goto retry;
814 return m;
815}
816
263e4574
MD
817/*
818 * Create a new thread and optionally associate it with a (new) process.
819 */
7d0bac62
MD
820void
821pmap_init_thread(thread_t td)
263e4574 822{
7e1d4bf4 823 td->td_pcb = (struct pcb *)(td->td_kstack + UPAGES * PAGE_SIZE) - 1;
7d0bac62 824 td->td_sp = (char *)td->td_pcb - 16;
263e4574
MD
825}
826
984263bc
MD
827/*
828 * Create the UPAGES for a new process.
829 * This routine directly affects the fork perf for a process.
830 */
831void
7d0bac62 832pmap_init_proc(struct proc *p, struct thread *td)
984263bc 833{
7e1d4bf4
MD
834 p->p_addr = (void *)td->td_kstack;
835 p->p_thread = td;
836 td->td_proc = p;
8ad65e08 837 td->td_switch = cpu_heavy_switch;
8a8d5d85
MD
838#ifdef SMP
839 td->td_mpcount = 1;
840#endif
7e1d4bf4 841 bzero(p->p_addr, sizeof(*p->p_addr));
984263bc
MD
842}
843
844/*
845 * Dispose the UPAGES for a process that has exited.
846 * This routine directly impacts the exit perf of a process.
847 */
7e1d4bf4
MD
848struct thread *
849pmap_dispose_proc(struct proc *p)
984263bc 850{
7e1d4bf4
MD
851 struct thread *td;
852
f1d1c3fa
MD
853 KASSERT(p->p_lock == 0, ("attempt to dispose referenced proc! %p", p));
854
7e1d4bf4
MD
855 if ((td = p->p_thread) != NULL) {
856 p->p_thread = NULL;
857 td->td_proc = NULL;
858 }
859 p->p_addr = NULL;
860 return(td);
984263bc
MD
861}
862
863/*
864 * Allow the UPAGES for a process to be prejudicially paged out.
865 */
866void
867pmap_swapout_proc(p)
868 struct proc *p;
869{
7e1d4bf4 870#if 0
984263bc
MD
871 int i;
872 vm_object_t upobj;
873 vm_page_t m;
874
875 upobj = p->p_upages_obj;
876 /*
877 * let the upages be paged
878 */
879 for(i=0;i<UPAGES;i++) {
880 if ((m = vm_page_lookup(upobj, i)) == NULL)
881 panic("pmap_swapout_proc: upage already missing???");
882 vm_page_dirty(m);
883 vm_page_unwire(m, 0);
884 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
885 }
7e1d4bf4 886#endif
984263bc
MD
887}
888
889/*
890 * Bring the UPAGES for a specified process back in.
891 */
892void
893pmap_swapin_proc(p)
894 struct proc *p;
895{
7e1d4bf4 896#if 0
984263bc
MD
897 int i,rv;
898 vm_object_t upobj;
899 vm_page_t m;
900
901 upobj = p->p_upages_obj;
902 for(i=0;i<UPAGES;i++) {
903
904 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
905
906 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
907 VM_PAGE_TO_PHYS(m));
908
909 if (m->valid != VM_PAGE_BITS_ALL) {
910 rv = vm_pager_get_pages(upobj, &m, 1, 0);
911 if (rv != VM_PAGER_OK)
912 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
913 m = vm_page_lookup(upobj, i);
914 m->valid = VM_PAGE_BITS_ALL;
915 }
916
917 vm_page_wire(m);
918 vm_page_wakeup(m);
919 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
920 }
7e1d4bf4 921#endif
984263bc
MD
922}
923
924/***************************************************
925 * Page table page management routines.....
926 ***************************************************/
927
928/*
929 * This routine unholds page table pages, and if the hold count
930 * drops to zero, then it decrements the wire count.
931 */
932static int
933_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
934
935 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
936 ;
937
938 if (m->hold_count == 0) {
939 vm_offset_t pteva;
940 /*
941 * unmap the page table page
942 */
943 pmap->pm_pdir[m->pindex] = 0;
944 --pmap->pm_stats.resident_count;
945 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
946 (((unsigned) PTDpde) & PG_FRAME)) {
947 /*
948 * Do a invltlb to make the invalidated mapping
949 * take effect immediately.
950 */
951 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
952 pmap_TLB_invalidate(pmap, pteva);
953 }
954
955 if (pmap->pm_ptphint == m)
956 pmap->pm_ptphint = NULL;
957
958 /*
959 * If the page is finally unwired, simply free it.
960 */
961 --m->wire_count;
962 if (m->wire_count == 0) {
963
964 vm_page_flash(m);
965 vm_page_busy(m);
966 vm_page_free_zero(m);
12e4aaff 967 --vmstats.v_wire_count;
984263bc
MD
968 }
969 return 1;
970 }
971 return 0;
972}
973
974static PMAP_INLINE int
975pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
976{
977 vm_page_unhold(m);
978 if (m->hold_count == 0)
979 return _pmap_unwire_pte_hold(pmap, m);
980 else
981 return 0;
982}
983
984/*
985 * After removing a page table entry, this routine is used to
986 * conditionally free the page, and manage the hold/wire counts.
987 */
988static int
989pmap_unuse_pt(pmap, va, mpte)
990 pmap_t pmap;
991 vm_offset_t va;
992 vm_page_t mpte;
993{
994 unsigned ptepindex;
995 if (va >= UPT_MIN_ADDRESS)
996 return 0;
997
998 if (mpte == NULL) {
999 ptepindex = (va >> PDRSHIFT);
1000 if (pmap->pm_ptphint &&
1001 (pmap->pm_ptphint->pindex == ptepindex)) {
1002 mpte = pmap->pm_ptphint;
1003 } else {
1004 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1005 pmap->pm_ptphint = mpte;
1006 }
1007 }
1008
1009 return pmap_unwire_pte_hold(pmap, mpte);
1010}
1011
1012void
1013pmap_pinit0(pmap)
1014 struct pmap *pmap;
1015{
1016 pmap->pm_pdir =
1017 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1018 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1019 pmap->pm_count = 1;
1020 pmap->pm_active = 0;
1021 pmap->pm_ptphint = NULL;
1022 TAILQ_INIT(&pmap->pm_pvlist);
1023 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1024}
1025
1026/*
1027 * Initialize a preallocated and zeroed pmap structure,
1028 * such as one in a vmspace structure.
1029 */
1030void
1031pmap_pinit(pmap)
1032 register struct pmap *pmap;
1033{
1034 vm_page_t ptdpg;
1035
1036 /*
1037 * No need to allocate page table space yet but we do need a valid
1038 * page directory table.
1039 */
1040 if (pmap->pm_pdir == NULL)
1041 pmap->pm_pdir =
1042 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1043
1044 /*
1045 * allocate object for the ptes
1046 */
1047 if (pmap->pm_pteobj == NULL)
1048 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1049
1050 /*
1051 * allocate the page directory page
1052 */
1053 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1054 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1055
1056 ptdpg->wire_count = 1;
12e4aaff 1057 ++vmstats.v_wire_count;
984263bc
MD
1058
1059
1060 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1061 ptdpg->valid = VM_PAGE_BITS_ALL;
1062
1063 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1064 if ((ptdpg->flags & PG_ZERO) == 0)
1065 bzero(pmap->pm_pdir, PAGE_SIZE);
1066
984263bc 1067 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
984263bc
MD
1068
1069 /* install self-referential address mapping entry */
1070 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1071 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1072
1073 pmap->pm_count = 1;
1074 pmap->pm_active = 0;
1075 pmap->pm_ptphint = NULL;
1076 TAILQ_INIT(&pmap->pm_pvlist);
1077 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1078}
1079
1080/*
1081 * Wire in kernel global address entries. To avoid a race condition
1082 * between pmap initialization and pmap_growkernel, this procedure
1083 * should be called after the vmspace is attached to the process
1084 * but before this pmap is activated.
1085 */
1086void
1087pmap_pinit2(pmap)
1088 struct pmap *pmap;
1089{
1090 /* XXX copies current process, does not fill in MPPTDI */
1091 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1092}
1093
1094static int
1095pmap_release_free_page(pmap, p)
1096 struct pmap *pmap;
1097 vm_page_t p;
1098{
1099 unsigned *pde = (unsigned *) pmap->pm_pdir;
1100 /*
1101 * This code optimizes the case of freeing non-busy
1102 * page-table pages. Those pages are zero now, and
1103 * might as well be placed directly into the zero queue.
1104 */
1105 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1106 return 0;
1107
1108 vm_page_busy(p);
1109
1110 /*
1111 * Remove the page table page from the processes address space.
1112 */
1113 pde[p->pindex] = 0;
1114 pmap->pm_stats.resident_count--;
1115
1116 if (p->hold_count) {
1117 panic("pmap_release: freeing held page table page");
1118 }
1119 /*
1120 * Page directory pages need to have the kernel
1121 * stuff cleared, so they can go into the zero queue also.
1122 */
1123 if (p->pindex == PTDPTDI) {
1124 bzero(pde + KPTDI, nkpt * PTESIZE);
984263bc 1125 pde[MPPTDI] = 0;
984263bc
MD
1126 pde[APTDPTDI] = 0;
1127 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1128 }
1129
1130 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1131 pmap->pm_ptphint = NULL;
1132
1133 p->wire_count--;
12e4aaff 1134 vmstats.v_wire_count--;
984263bc
MD
1135 vm_page_free_zero(p);
1136 return 1;
1137}
1138
1139/*
1140 * this routine is called if the page table page is not
1141 * mapped correctly.
1142 */
1143static vm_page_t
1144_pmap_allocpte(pmap, ptepindex)
1145 pmap_t pmap;
1146 unsigned ptepindex;
1147{
1148 vm_offset_t pteva, ptepa;
1149 vm_page_t m;
1150
1151 /*
1152 * Find or fabricate a new pagetable page
1153 */
1154 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1155 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1156
1157 KASSERT(m->queue == PQ_NONE,
1158 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1159
1160 if (m->wire_count == 0)
12e4aaff 1161 vmstats.v_wire_count++;
984263bc
MD
1162 m->wire_count++;
1163
1164 /*
1165 * Increment the hold count for the page table page
1166 * (denoting a new mapping.)
1167 */
1168 m->hold_count++;
1169
1170 /*
1171 * Map the pagetable page into the process address space, if
1172 * it isn't already there.
1173 */
1174
1175 pmap->pm_stats.resident_count++;
1176
1177 ptepa = VM_PAGE_TO_PHYS(m);
1178 pmap->pm_pdir[ptepindex] =
1179 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1180
1181 /*
1182 * Set the page table hint
1183 */
1184 pmap->pm_ptphint = m;
1185
1186 /*
1187 * Try to use the new mapping, but if we cannot, then
1188 * do it with the routine that maps the page explicitly.
1189 */
1190 if ((m->flags & PG_ZERO) == 0) {
1191 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1192 (((unsigned) PTDpde) & PG_FRAME)) {
1193 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1194 bzero((caddr_t) pteva, PAGE_SIZE);
1195 } else {
1196 pmap_zero_page(ptepa);
1197 }
1198 }
1199
1200 m->valid = VM_PAGE_BITS_ALL;
1201 vm_page_flag_clear(m, PG_ZERO);
1202 vm_page_flag_set(m, PG_MAPPED);
1203 vm_page_wakeup(m);
1204
1205 return m;
1206}
1207
1208static vm_page_t
1209pmap_allocpte(pmap, va)
1210 pmap_t pmap;
1211 vm_offset_t va;
1212{
1213 unsigned ptepindex;
1214 vm_offset_t ptepa;
1215 vm_page_t m;
1216
1217 /*
1218 * Calculate pagetable page index
1219 */
1220 ptepindex = va >> PDRSHIFT;
1221
1222 /*
1223 * Get the page directory entry
1224 */
1225 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1226
1227 /*
1228 * This supports switching from a 4MB page to a
1229 * normal 4K page.
1230 */
1231 if (ptepa & PG_PS) {
1232 pmap->pm_pdir[ptepindex] = 0;
1233 ptepa = 0;
1234 invltlb();
1235 }
1236
1237 /*
1238 * If the page table page is mapped, we just increment the
1239 * hold count, and activate it.
1240 */
1241 if (ptepa) {
1242 /*
1243 * In order to get the page table page, try the
1244 * hint first.
1245 */
1246 if (pmap->pm_ptphint &&
1247 (pmap->pm_ptphint->pindex == ptepindex)) {
1248 m = pmap->pm_ptphint;
1249 } else {
1250 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1251 pmap->pm_ptphint = m;
1252 }
1253 m->hold_count++;
1254 return m;
1255 }
1256 /*
1257 * Here if the pte page isn't mapped, or if it has been deallocated.
1258 */
1259 return _pmap_allocpte(pmap, ptepindex);
1260}
1261
1262
1263/***************************************************
1264* Pmap allocation/deallocation routines.
1265 ***************************************************/
1266
1267/*
1268 * Release any resources held by the given physical map.
1269 * Called when a pmap initialized by pmap_pinit is being released.
1270 * Should only be called if the map contains no valid mappings.
1271 */
1272void
1273pmap_release(pmap)
1274 register struct pmap *pmap;
1275{
1276 vm_page_t p,n,ptdpg;
1277 vm_object_t object = pmap->pm_pteobj;
1278 int curgeneration;
1279
1280#if defined(DIAGNOSTIC)
1281 if (object->ref_count != 1)
1282 panic("pmap_release: pteobj reference count != 1");
1283#endif
1284
1285 ptdpg = NULL;
1286retry:
1287 curgeneration = object->generation;
1288 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1289 n = TAILQ_NEXT(p, listq);
1290 if (p->pindex == PTDPTDI) {
1291 ptdpg = p;
1292 continue;
1293 }
1294 while (1) {
1295 if (!pmap_release_free_page(pmap, p) &&
1296 (object->generation != curgeneration))
1297 goto retry;
1298 }
1299 }
1300
1301 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1302 goto retry;
1303}
1304\f
1305static int
1306kvm_size(SYSCTL_HANDLER_ARGS)
1307{
1308 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1309
1310 return sysctl_handle_long(oidp, &ksize, 0, req);
1311}
1312SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1313 0, 0, kvm_size, "IU", "Size of KVM");
1314
1315static int
1316kvm_free(SYSCTL_HANDLER_ARGS)
1317{
1318 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1319
1320 return sysctl_handle_long(oidp, &kfree, 0, req);
1321}
1322SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1323 0, 0, kvm_free, "IU", "Amount of KVM free");
1324
1325/*
1326 * grow the number of kernel page table entries, if needed
1327 */
1328void
1329pmap_growkernel(vm_offset_t addr)
1330{
1331 struct proc *p;
1332 struct pmap *pmap;
1333 int s;
1334 vm_offset_t ptppaddr;
1335 vm_page_t nkpg;
1336 pd_entry_t newpdir;
1337
1338 s = splhigh();
1339 if (kernel_vm_end == 0) {
1340 kernel_vm_end = KERNBASE;
1341 nkpt = 0;
1342 while (pdir_pde(PTD, kernel_vm_end)) {
1343 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1344 nkpt++;
1345 }
1346 }
1347 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1348 while (kernel_vm_end < addr) {
1349 if (pdir_pde(PTD, kernel_vm_end)) {
1350 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1351 continue;
1352 }
1353
1354 /*
1355 * This index is bogus, but out of the way
1356 */
1357 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1358 if (!nkpg)
1359 panic("pmap_growkernel: no memory to grow kernel");
1360
1361 nkpt++;
1362
1363 vm_page_wire(nkpg);
1364 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1365 pmap_zero_page(ptppaddr);
1366 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1367 pdir_pde(PTD, kernel_vm_end) = newpdir;
1368
1369 LIST_FOREACH(p, &allproc, p_list) {
1370 if (p->p_vmspace) {
1371 pmap = vmspace_pmap(p->p_vmspace);
1372 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1373 }
1374 }
1375 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1376 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1377 }
1378 splx(s);
1379}
1380
1381/*
1382 * Retire the given physical map from service.
1383 * Should only be called if the map contains
1384 * no valid mappings.
1385 */
1386void
1387pmap_destroy(pmap)
1388 register pmap_t pmap;
1389{
1390 int count;
1391
1392 if (pmap == NULL)
1393 return;
1394
1395 count = --pmap->pm_count;
1396 if (count == 0) {
1397 pmap_release(pmap);
1398 panic("destroying a pmap is not yet implemented");
1399 }
1400}
1401
1402/*
1403 * Add a reference to the specified pmap.
1404 */
1405void
1406pmap_reference(pmap)
1407 pmap_t pmap;
1408{
1409 if (pmap != NULL) {
1410 pmap->pm_count++;
1411 }
1412}
1413
1414/***************************************************
1415* page management routines.
1416 ***************************************************/
1417
1418/*
8a8d5d85
MD
1419 * free the pv_entry back to the free list. This function may be
1420 * called from an interrupt.
984263bc
MD
1421 */
1422static PMAP_INLINE void
1423free_pv_entry(pv)
1424 pv_entry_t pv;
1425{
1426 pv_entry_count--;
8a8d5d85 1427 zfree(pvzone, pv);
984263bc
MD
1428}
1429
1430/*
1431 * get a new pv_entry, allocating a block from the system
8a8d5d85 1432 * when needed. This function may be called from an interrupt.
984263bc
MD
1433 */
1434static pv_entry_t
1435get_pv_entry(void)
1436{
1437 pv_entry_count++;
1438 if (pv_entry_high_water &&
1439 (pv_entry_count > pv_entry_high_water) &&
1440 (pmap_pagedaemon_waken == 0)) {
1441 pmap_pagedaemon_waken = 1;
1442 wakeup (&vm_pages_needed);
1443 }
8a8d5d85 1444 return zalloc(pvzone);
984263bc
MD
1445}
1446
1447/*
1448 * This routine is very drastic, but can save the system
1449 * in a pinch.
1450 */
1451void
1452pmap_collect()
1453{
1454 int i;
1455 vm_page_t m;
1456 static int warningdone=0;
1457
1458 if (pmap_pagedaemon_waken == 0)
1459 return;
1460
1461 if (warningdone < 5) {
1462 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1463 warningdone++;
1464 }
1465
1466 for(i = 0; i < vm_page_array_size; i++) {
1467 m = &vm_page_array[i];
1468 if (m->wire_count || m->hold_count || m->busy ||
1469 (m->flags & PG_BUSY))
1470 continue;
1471 pmap_remove_all(m);
1472 }
1473 pmap_pagedaemon_waken = 0;
1474}
1475
1476
1477/*
1478 * If it is the first entry on the list, it is actually
1479 * in the header and we must copy the following entry up
1480 * to the header. Otherwise we must search the list for
1481 * the entry. In either case we free the now unused entry.
1482 */
1483
1484static int
1485pmap_remove_entry(pmap, m, va)
1486 struct pmap *pmap;
1487 vm_page_t m;
1488 vm_offset_t va;
1489{
1490 pv_entry_t pv;
1491 int rtval;
1492 int s;
1493
1494 s = splvm();
1495 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1496 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1497 if (pmap == pv->pv_pmap && va == pv->pv_va)
1498 break;
1499 }
1500 } else {
1501 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1502 if (va == pv->pv_va)
1503 break;
1504 }
1505 }
1506
1507 rtval = 0;
1508 if (pv) {
1509
1510 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1511 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1512 m->md.pv_list_count--;
1513 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1514 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1515
1516 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1517 free_pv_entry(pv);
1518 }
1519
1520 splx(s);
1521 return rtval;
1522}
1523
1524/*
1525 * Create a pv entry for page at pa for
1526 * (pmap, va).
1527 */
1528static void
1529pmap_insert_entry(pmap, va, mpte, m)
1530 pmap_t pmap;
1531 vm_offset_t va;
1532 vm_page_t mpte;
1533 vm_page_t m;
1534{
1535
1536 int s;
1537 pv_entry_t pv;
1538
1539 s = splvm();
1540 pv = get_pv_entry();
1541 pv->pv_va = va;
1542 pv->pv_pmap = pmap;
1543 pv->pv_ptem = mpte;
1544
1545 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1546 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1547 m->md.pv_list_count++;
1548
1549 splx(s);
1550}
1551
1552/*
1553 * pmap_remove_pte: do the things to unmap a page in a process
1554 */
1555static int
1556pmap_remove_pte(pmap, ptq, va)
1557 struct pmap *pmap;
1558 unsigned *ptq;
1559 vm_offset_t va;
1560{
1561 unsigned oldpte;
1562 vm_page_t m;
1563
1564 oldpte = loadandclear(ptq);
1565 if (oldpte & PG_W)
1566 pmap->pm_stats.wired_count -= 1;
1567 /*
1568 * Machines that don't support invlpg, also don't support
1569 * PG_G.
1570 */
1571 if (oldpte & PG_G)
1572 invlpg(va);
1573 pmap->pm_stats.resident_count -= 1;
1574 if (oldpte & PG_MANAGED) {
1575 m = PHYS_TO_VM_PAGE(oldpte);
1576 if (oldpte & PG_M) {
1577#if defined(PMAP_DIAGNOSTIC)
1578 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1579 printf(
1580 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1581 va, oldpte);
1582 }
1583#endif
1584 if (pmap_track_modified(va))
1585 vm_page_dirty(m);
1586 }
1587 if (oldpte & PG_A)
1588 vm_page_flag_set(m, PG_REFERENCED);
1589 return pmap_remove_entry(pmap, m, va);
1590 } else {
1591 return pmap_unuse_pt(pmap, va, NULL);
1592 }
1593
1594 return 0;
1595}
1596
1597/*
1598 * Remove a single page from a process address space
1599 */
1600static void
1601pmap_remove_page(pmap, va)
1602 struct pmap *pmap;
1603 register vm_offset_t va;
1604{
1605 register unsigned *ptq;
1606
1607 /*
1608 * if there is no pte for this address, just skip it!!!
1609 */
1610 if (*pmap_pde(pmap, va) == 0) {
1611 return;
1612 }
1613
1614 /*
1615 * get a local va for mappings for this pmap.
1616 */
1617 ptq = get_ptbase(pmap) + i386_btop(va);
1618 if (*ptq) {
1619 (void) pmap_remove_pte(pmap, ptq, va);
1620 pmap_TLB_invalidate(pmap, va);
1621 }
1622 return;
1623}
1624
1625/*
1626 * Remove the given range of addresses from the specified map.
1627 *
1628 * It is assumed that the start and end are properly
1629 * rounded to the page size.
1630 */
1631void
1632pmap_remove(pmap, sva, eva)
1633 struct pmap *pmap;
1634 register vm_offset_t sva;
1635 register vm_offset_t eva;
1636{
1637 register unsigned *ptbase;
1638 vm_offset_t pdnxt;
1639 vm_offset_t ptpaddr;
1640 vm_offset_t sindex, eindex;
1641 int anyvalid;
1642
1643 if (pmap == NULL)
1644 return;
1645
1646 if (pmap->pm_stats.resident_count == 0)
1647 return;
1648
1649 /*
1650 * special handling of removing one page. a very
1651 * common operation and easy to short circuit some
1652 * code.
1653 */
1654 if (((sva + PAGE_SIZE) == eva) &&
1655 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1656 pmap_remove_page(pmap, sva);
1657 return;
1658 }
1659
1660 anyvalid = 0;
1661
1662 /*
1663 * Get a local virtual address for the mappings that are being
1664 * worked with.
1665 */
1666 ptbase = get_ptbase(pmap);
1667
1668 sindex = i386_btop(sva);
1669 eindex = i386_btop(eva);
1670
1671 for (; sindex < eindex; sindex = pdnxt) {
1672 unsigned pdirindex;
1673
1674 /*
1675 * Calculate index for next page table.
1676 */
1677 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1678 if (pmap->pm_stats.resident_count == 0)
1679 break;
1680
1681 pdirindex = sindex / NPDEPG;
1682 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1683 pmap->pm_pdir[pdirindex] = 0;
1684 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1685 anyvalid++;
1686 continue;
1687 }
1688
1689 /*
1690 * Weed out invalid mappings. Note: we assume that the page
1691 * directory table is always allocated, and in kernel virtual.
1692 */
1693 if (ptpaddr == 0)
1694 continue;
1695
1696 /*
1697 * Limit our scan to either the end of the va represented
1698 * by the current page table page, or to the end of the
1699 * range being removed.
1700 */
1701 if (pdnxt > eindex) {
1702 pdnxt = eindex;
1703 }
1704
1705 for ( ;sindex != pdnxt; sindex++) {
1706 vm_offset_t va;
1707 if (ptbase[sindex] == 0) {
1708 continue;
1709 }
1710 va = i386_ptob(sindex);
1711
1712 anyvalid++;
1713 if (pmap_remove_pte(pmap,
1714 ptbase + sindex, va))
1715 break;
1716 }
1717 }
1718
1719 if (anyvalid)
1720 pmap_TLB_invalidate_all(pmap);
1721}
1722
1723/*
1724 * Routine: pmap_remove_all
1725 * Function:
1726 * Removes this physical page from
1727 * all physical maps in which it resides.
1728 * Reflects back modify bits to the pager.
1729 *
1730 * Notes:
1731 * Original versions of this routine were very
1732 * inefficient because they iteratively called
1733 * pmap_remove (slow...)
1734 */
1735
1736static void
1737pmap_remove_all(m)
1738 vm_page_t m;
1739{
1740 register pv_entry_t pv;
1741 register unsigned *pte, tpte;
1742 int s;
1743
1744#if defined(PMAP_DIAGNOSTIC)
1745 /*
1746 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1747 * pages!
1748 */
1749 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1750 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1751 }
1752#endif
1753
1754 s = splvm();
1755 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1756 pv->pv_pmap->pm_stats.resident_count--;
1757
1758 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1759
1760 tpte = loadandclear(pte);
1761 if (tpte & PG_W)
1762 pv->pv_pmap->pm_stats.wired_count--;
1763
1764 if (tpte & PG_A)
1765 vm_page_flag_set(m, PG_REFERENCED);
1766
1767 /*
1768 * Update the vm_page_t clean and reference bits.
1769 */
1770 if (tpte & PG_M) {
1771#if defined(PMAP_DIAGNOSTIC)
1772 if (pmap_nw_modified((pt_entry_t) tpte)) {
1773 printf(
1774 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1775 pv->pv_va, tpte);
1776 }
1777#endif
1778 if (pmap_track_modified(pv->pv_va))
1779 vm_page_dirty(m);
1780 }
1781 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1782
1783 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1784 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1785 m->md.pv_list_count--;
1786 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1787 free_pv_entry(pv);
1788 }
1789
1790 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1791
1792 splx(s);
1793}
1794
1795/*
1796 * Set the physical protection on the
1797 * specified range of this map as requested.
1798 */
1799void
1800pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1801{
1802 register unsigned *ptbase;
1803 vm_offset_t pdnxt, ptpaddr;
1804 vm_pindex_t sindex, eindex;
1805 int anychanged;
1806
1807 if (pmap == NULL)
1808 return;
1809
1810 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1811 pmap_remove(pmap, sva, eva);
1812 return;
1813 }
1814
1815 if (prot & VM_PROT_WRITE)
1816 return;
1817
1818 anychanged = 0;
1819
1820 ptbase = get_ptbase(pmap);
1821
1822 sindex = i386_btop(sva);
1823 eindex = i386_btop(eva);
1824
1825 for (; sindex < eindex; sindex = pdnxt) {
1826
1827 unsigned pdirindex;
1828
1829 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1830
1831 pdirindex = sindex / NPDEPG;
1832 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1833 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1834 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1835 anychanged++;
1836 continue;
1837 }
1838
1839 /*
1840 * Weed out invalid mappings. Note: we assume that the page
1841 * directory table is always allocated, and in kernel virtual.
1842 */
1843 if (ptpaddr == 0)
1844 continue;
1845
1846 if (pdnxt > eindex) {
1847 pdnxt = eindex;
1848 }
1849
1850 for (; sindex != pdnxt; sindex++) {
1851
1852 unsigned pbits;
1853 vm_page_t m;
1854
1855 pbits = ptbase[sindex];
1856
1857 if (pbits & PG_MANAGED) {
1858 m = NULL;
1859 if (pbits & PG_A) {
1860 m = PHYS_TO_VM_PAGE(pbits);
1861 vm_page_flag_set(m, PG_REFERENCED);
1862 pbits &= ~PG_A;
1863 }
1864 if (pbits & PG_M) {
1865 if (pmap_track_modified(i386_ptob(sindex))) {
1866 if (m == NULL)
1867 m = PHYS_TO_VM_PAGE(pbits);
1868 vm_page_dirty(m);
1869 pbits &= ~PG_M;
1870 }
1871 }
1872 }
1873
1874 pbits &= ~PG_RW;
1875
1876 if (pbits != ptbase[sindex]) {
1877 ptbase[sindex] = pbits;
1878 anychanged = 1;
1879 }
1880 }
1881 }
1882 if (anychanged)
1883 pmap_TLB_invalidate_all(pmap);
1884}
1885
1886/*
1887 * Insert the given physical page (p) at
1888 * the specified virtual address (v) in the
1889 * target physical map with the protection requested.
1890 *
1891 * If specified, the page will be wired down, meaning
1892 * that the related pte can not be reclaimed.
1893 *
1894 * NB: This is the only routine which MAY NOT lazy-evaluate
1895 * or lose information. That is, this routine must actually
1896 * insert this page into the given map NOW.
1897 */
1898void
1899pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1900 boolean_t wired)
1901{
1902 vm_offset_t pa;
1903 register unsigned *pte;
1904 vm_offset_t opa;
1905 vm_offset_t origpte, newpte;
1906 vm_page_t mpte;
1907
1908 if (pmap == NULL)
1909 return;
1910
1911 va &= PG_FRAME;
1912#ifdef PMAP_DIAGNOSTIC
1913 if (va > VM_MAX_KERNEL_ADDRESS)
1914 panic("pmap_enter: toobig");
1915 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
1916 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
1917#endif
1918
1919 mpte = NULL;
1920 /*
1921 * In the case that a page table page is not
1922 * resident, we are creating it here.
1923 */
1924 if (va < UPT_MIN_ADDRESS) {
1925 mpte = pmap_allocpte(pmap, va);
1926 }
1927#if 0 && defined(PMAP_DIAGNOSTIC)
1928 else {
1929 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
1930 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
1931 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
1932 pmap->pm_pdir[PTDPTDI], origpte, va);
1933 }
1934 if (smp_active) {
1935 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
1936 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
1937 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
1938 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
1939 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
1940 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
1941 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
1942 }
1943 }
1944 }
1945#endif
1946
1947 pte = pmap_pte(pmap, va);
1948
1949 /*
1950 * Page Directory table entry not valid, we need a new PT page
1951 */
1952 if (pte == NULL) {
1953 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
1954 (void *)pmap->pm_pdir[PTDPTDI], va);
1955 }
1956
1957 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
1958 origpte = *(vm_offset_t *)pte;
1959 opa = origpte & PG_FRAME;
1960
1961 if (origpte & PG_PS)
1962 panic("pmap_enter: attempted pmap_enter on 4MB page");
1963
1964 /*
1965 * Mapping has not changed, must be protection or wiring change.
1966 */
1967 if (origpte && (opa == pa)) {
1968 /*
1969 * Wiring change, just update stats. We don't worry about
1970 * wiring PT pages as they remain resident as long as there
1971 * are valid mappings in them. Hence, if a user page is wired,
1972 * the PT page will be also.
1973 */
1974 if (wired && ((origpte & PG_W) == 0))
1975 pmap->pm_stats.wired_count++;
1976 else if (!wired && (origpte & PG_W))
1977 pmap->pm_stats.wired_count--;
1978
1979#if defined(PMAP_DIAGNOSTIC)
1980 if (pmap_nw_modified((pt_entry_t) origpte)) {
1981 printf(
1982 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
1983 va, origpte);
1984 }
1985#endif
1986
1987 /*
1988 * Remove extra pte reference
1989 */
1990 if (mpte)
1991 mpte->hold_count--;
1992
1993 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
1994 if ((origpte & PG_RW) == 0) {
1995 *pte |= PG_RW;
1996#ifdef SMP
1997 cpu_invlpg((void *)va);
72740893 1998 if (pmap->pm_active & mycpu->gd_other_cpus)
984263bc
MD
1999 smp_invltlb();
2000#else
2001 invltlb_1pg(va);
2002#endif
2003 }
2004 return;
2005 }
2006
2007 /*
2008 * We might be turning off write access to the page,
2009 * so we go ahead and sense modify status.
2010 */
2011 if (origpte & PG_MANAGED) {
2012 if ((origpte & PG_M) && pmap_track_modified(va)) {
2013 vm_page_t om;
2014 om = PHYS_TO_VM_PAGE(opa);
2015 vm_page_dirty(om);
2016 }
2017 pa |= PG_MANAGED;
2018 }
2019 goto validate;
2020 }
2021 /*
2022 * Mapping has changed, invalidate old range and fall through to
2023 * handle validating new mapping.
2024 */
2025 if (opa) {
2026 int err;
2027 err = pmap_remove_pte(pmap, pte, va);
2028 if (err)
2029 panic("pmap_enter: pte vanished, va: 0x%x", va);
2030 }
2031
2032 /*
2033 * Enter on the PV list if part of our managed memory. Note that we
2034 * raise IPL while manipulating pv_table since pmap_enter can be
2035 * called at interrupt time.
2036 */
2037 if (pmap_initialized &&
2038 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2039 pmap_insert_entry(pmap, va, mpte, m);
2040 pa |= PG_MANAGED;
2041 }
2042
2043 /*
2044 * Increment counters
2045 */
2046 pmap->pm_stats.resident_count++;
2047 if (wired)
2048 pmap->pm_stats.wired_count++;
2049
2050validate:
2051 /*
2052 * Now validate mapping with desired protection/wiring.
2053 */
2054 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2055
2056 if (wired)
2057 newpte |= PG_W;
2058 if (va < UPT_MIN_ADDRESS)
2059 newpte |= PG_U;
2060 if (pmap == kernel_pmap)
2061 newpte |= pgeflag;
2062
2063 /*
2064 * if the mapping or permission bits are different, we need
2065 * to update the pte.
2066 */
2067 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2068 *pte = newpte | PG_A;
2069 /*if (origpte)*/ {
2070#ifdef SMP
2071 cpu_invlpg((void *)va);
72740893 2072 if (pmap->pm_active & mycpu->gd_other_cpus)
984263bc
MD
2073 smp_invltlb();
2074#else
2075 invltlb_1pg(va);
2076#endif
2077 }
2078 }
2079}
2080
2081/*
2082 * this code makes some *MAJOR* assumptions:
2083 * 1. Current pmap & pmap exists.
2084 * 2. Not wired.
2085 * 3. Read access.
2086 * 4. No page table pages.
2087 * 5. Tlbflush is deferred to calling procedure.
2088 * 6. Page IS managed.
2089 * but is *MUCH* faster than pmap_enter...
2090 */
2091
2092static vm_page_t
2093pmap_enter_quick(pmap, va, m, mpte)
2094 register pmap_t pmap;
2095 vm_offset_t va;
2096 vm_page_t m;
2097 vm_page_t mpte;
2098{
2099 unsigned *pte;
2100 vm_offset_t pa;
2101
2102 /*
2103 * In the case that a page table page is not
2104 * resident, we are creating it here.
2105 */
2106 if (va < UPT_MIN_ADDRESS) {
2107 unsigned ptepindex;
2108 vm_offset_t ptepa;
2109
2110 /*
2111 * Calculate pagetable page index
2112 */
2113 ptepindex = va >> PDRSHIFT;
2114 if (mpte && (mpte->pindex == ptepindex)) {
2115 mpte->hold_count++;
2116 } else {
2117retry:
2118 /*
2119 * Get the page directory entry
2120 */
2121 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2122
2123 /*
2124 * If the page table page is mapped, we just increment
2125 * the hold count, and activate it.
2126 */
2127 if (ptepa) {
2128 if (ptepa & PG_PS)
2129 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2130 if (pmap->pm_ptphint &&
2131 (pmap->pm_ptphint->pindex == ptepindex)) {
2132 mpte = pmap->pm_ptphint;
2133 } else {
2134 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2135 pmap->pm_ptphint = mpte;
2136 }
2137 if (mpte == NULL)
2138 goto retry;
2139 mpte->hold_count++;
2140 } else {
2141 mpte = _pmap_allocpte(pmap, ptepindex);
2142 }
2143 }
2144 } else {
2145 mpte = NULL;
2146 }
2147
2148 /*
2149 * This call to vtopte makes the assumption that we are
2150 * entering the page into the current pmap. In order to support
2151 * quick entry into any pmap, one would likely use pmap_pte_quick.
2152 * But that isn't as quick as vtopte.
2153 */
2154 pte = (unsigned *)vtopte(va);
2155 if (*pte) {
2156 if (mpte)
2157 pmap_unwire_pte_hold(pmap, mpte);
2158 return 0;
2159 }
2160
2161 /*
2162 * Enter on the PV list if part of our managed memory. Note that we
2163 * raise IPL while manipulating pv_table since pmap_enter can be
2164 * called at interrupt time.
2165 */
2166 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2167 pmap_insert_entry(pmap, va, mpte, m);
2168
2169 /*
2170 * Increment counters
2171 */
2172 pmap->pm_stats.resident_count++;
2173
2174 pa = VM_PAGE_TO_PHYS(m);
2175
2176 /*
2177 * Now validate mapping with RO protection
2178 */
2179 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2180 *pte = pa | PG_V | PG_U;
2181 else
2182 *pte = pa | PG_V | PG_U | PG_MANAGED;
2183
2184 return mpte;
2185}
2186
2187/*
2188 * Make a temporary mapping for a physical address. This is only intended
2189 * to be used for panic dumps.
2190 */
2191void *
2192pmap_kenter_temporary(vm_offset_t pa, int i)
2193{
2194 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2195 return ((void *)crashdumpmap);
2196}
2197
2198#define MAX_INIT_PT (96)
2199/*
2200 * pmap_object_init_pt preloads the ptes for a given object
2201 * into the specified pmap. This eliminates the blast of soft
2202 * faults on process startup and immediately after an mmap.
2203 */
2204void
2205pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2206 pmap_t pmap;
2207 vm_offset_t addr;
2208 vm_object_t object;
2209 vm_pindex_t pindex;
2210 vm_size_t size;
2211 int limit;
2212{
2213 vm_offset_t tmpidx;
2214 int psize;
2215 vm_page_t p, mpte;
2216 int objpgs;
2217
2218 if (pmap == NULL || object == NULL)
2219 return;
2220
2221 /*
2222 * This code maps large physical mmap regions into the
2223 * processor address space. Note that some shortcuts
2224 * are taken, but the code works.
2225 */
2226 if (pseflag &&
2227 (object->type == OBJT_DEVICE) &&
2228 ((addr & (NBPDR - 1)) == 0) &&
2229 ((size & (NBPDR - 1)) == 0) ) {
2230 int i;
2231 vm_page_t m[1];
2232 unsigned int ptepindex;
2233 int npdes;
2234 vm_offset_t ptepa;
2235
2236 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2237 return;
2238
2239retry:
2240 p = vm_page_lookup(object, pindex);
2241 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2242 goto retry;
2243
2244 if (p == NULL) {
2245 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2246 if (p == NULL)
2247 return;
2248 m[0] = p;
2249
2250 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2251 vm_page_free(p);
2252 return;
2253 }
2254
2255 p = vm_page_lookup(object, pindex);
2256 vm_page_wakeup(p);
2257 }
2258
2259 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2260 if (ptepa & (NBPDR - 1)) {
2261 return;
2262 }
2263
2264 p->valid = VM_PAGE_BITS_ALL;
2265
2266 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2267 npdes = size >> PDRSHIFT;
2268 for(i=0;i<npdes;i++) {
2269 pmap->pm_pdir[ptepindex] =
2270 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2271 ptepa += NBPDR;
2272 ptepindex += 1;
2273 }
2274 vm_page_flag_set(p, PG_MAPPED);
2275 invltlb();
2276 return;
2277 }
2278
2279 psize = i386_btop(size);
2280
2281 if ((object->type != OBJT_VNODE) ||
2282 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2283 (object->resident_page_count > MAX_INIT_PT))) {
2284 return;
2285 }
2286
2287 if (psize + pindex > object->size) {
2288 if (object->size < pindex)
2289 return;
2290 psize = object->size - pindex;
2291 }
2292
2293 mpte = NULL;
2294 /*
2295 * if we are processing a major portion of the object, then scan the
2296 * entire thing.
2297 */
2298 if (psize > (object->resident_page_count >> 2)) {
2299 objpgs = psize;
2300
2301 for (p = TAILQ_FIRST(&object->memq);
2302 ((objpgs > 0) && (p != NULL));
2303 p = TAILQ_NEXT(p, listq)) {
2304
2305 tmpidx = p->pindex;
2306 if (tmpidx < pindex) {
2307 continue;
2308 }
2309 tmpidx -= pindex;
2310 if (tmpidx >= psize) {
2311 continue;
2312 }
2313 /*
2314 * don't allow an madvise to blow away our really
2315 * free pages allocating pv entries.
2316 */
2317 if ((limit & MAP_PREFAULT_MADVISE) &&
12e4aaff 2318 vmstats.v_free_count < vmstats.v_free_reserved) {
984263bc
MD
2319 break;
2320 }
2321 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2322 (p->busy == 0) &&
2323 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2324 if ((p->queue - p->pc) == PQ_CACHE)
2325 vm_page_deactivate(p);
2326 vm_page_busy(p);
2327 mpte = pmap_enter_quick(pmap,
2328 addr + i386_ptob(tmpidx), p, mpte);
2329 vm_page_flag_set(p, PG_MAPPED);
2330 vm_page_wakeup(p);
2331 }
2332 objpgs -= 1;
2333 }
2334 } else {
2335 /*
2336 * else lookup the pages one-by-one.
2337 */
2338 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2339 /*
2340 * don't allow an madvise to blow away our really
2341 * free pages allocating pv entries.
2342 */
2343 if ((limit & MAP_PREFAULT_MADVISE) &&
12e4aaff 2344 vmstats.v_free_count < vmstats.v_free_reserved) {
984263bc
MD
2345 break;
2346 }
2347 p = vm_page_lookup(object, tmpidx + pindex);
2348 if (p &&
2349 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2350 (p->busy == 0) &&
2351 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2352 if ((p->queue - p->pc) == PQ_CACHE)
2353 vm_page_deactivate(p);
2354 vm_page_busy(p);
2355 mpte = pmap_enter_quick(pmap,
2356 addr + i386_ptob(tmpidx), p, mpte);
2357 vm_page_flag_set(p, PG_MAPPED);
2358 vm_page_wakeup(p);
2359 }
2360 }
2361 }
2362 return;
2363}
2364
2365/*
2366 * pmap_prefault provides a quick way of clustering
2367 * pagefaults into a processes address space. It is a "cousin"
2368 * of pmap_object_init_pt, except it runs at page fault time instead
2369 * of mmap time.
2370 */
2371#define PFBAK 4
2372#define PFFOR 4
2373#define PAGEORDER_SIZE (PFBAK+PFFOR)
2374
2375static int pmap_prefault_pageorder[] = {
2376 -PAGE_SIZE, PAGE_SIZE,
2377 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2378 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2379 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2380};
2381
2382void
2383pmap_prefault(pmap, addra, entry)
2384 pmap_t pmap;
2385 vm_offset_t addra;
2386 vm_map_entry_t entry;
2387{
2388 int i;
2389 vm_offset_t starta;
2390 vm_offset_t addr;
2391 vm_pindex_t pindex;
2392 vm_page_t m, mpte;
2393 vm_object_t object;
2394
2395 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2396 return;
2397
2398 object = entry->object.vm_object;
2399
2400 starta = addra - PFBAK * PAGE_SIZE;
2401 if (starta < entry->start) {
2402 starta = entry->start;
2403 } else if (starta > addra) {
2404 starta = 0;
2405 }
2406
2407 mpte = NULL;
2408 for (i = 0; i < PAGEORDER_SIZE; i++) {
2409 vm_object_t lobject;
2410 unsigned *pte;
2411
2412 addr = addra + pmap_prefault_pageorder[i];
2413 if (addr > addra + (PFFOR * PAGE_SIZE))
2414 addr = 0;
2415
2416 if (addr < starta || addr >= entry->end)
2417 continue;
2418
2419 if ((*pmap_pde(pmap, addr)) == NULL)
2420 continue;
2421
2422 pte = (unsigned *) vtopte(addr);
2423 if (*pte)
2424 continue;
2425
2426 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2427 lobject = object;
2428 for (m = vm_page_lookup(lobject, pindex);
2429 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2430 lobject = lobject->backing_object) {
2431 if (lobject->backing_object_offset & PAGE_MASK)
2432 break;
2433 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2434 m = vm_page_lookup(lobject->backing_object, pindex);
2435 }
2436
2437 /*
2438 * give-up when a page is not in memory
2439 */
2440 if (m == NULL)
2441 break;
2442
2443 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2444 (m->busy == 0) &&
2445 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2446
2447 if ((m->queue - m->pc) == PQ_CACHE) {
2448 vm_page_deactivate(m);
2449 }
2450 vm_page_busy(m);
2451 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2452 vm_page_flag_set(m, PG_MAPPED);
2453 vm_page_wakeup(m);
2454 }
2455 }
2456}
2457
2458/*
2459 * Routine: pmap_change_wiring
2460 * Function: Change the wiring attribute for a map/virtual-address
2461 * pair.
2462 * In/out conditions:
2463 * The mapping must already exist in the pmap.
2464 */
2465void
2466pmap_change_wiring(pmap, va, wired)
2467 register pmap_t pmap;
2468 vm_offset_t va;
2469 boolean_t wired;
2470{
2471 register unsigned *pte;
2472
2473 if (pmap == NULL)
2474 return;
2475
2476 pte = pmap_pte(pmap, va);
2477
2478 if (wired && !pmap_pte_w(pte))
2479 pmap->pm_stats.wired_count++;
2480 else if (!wired && pmap_pte_w(pte))
2481 pmap->pm_stats.wired_count--;
2482
2483 /*
2484 * Wiring is not a hardware characteristic so there is no need to
2485 * invalidate TLB.
2486 */
2487 pmap_pte_set_w(pte, wired);
2488}
2489
2490
2491
2492/*
2493 * Copy the range specified by src_addr/len
2494 * from the source map to the range dst_addr/len
2495 * in the destination map.
2496 *
2497 * This routine is only advisory and need not do anything.
2498 */
2499
2500void
2501pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2502 pmap_t dst_pmap, src_pmap;
2503 vm_offset_t dst_addr;
2504 vm_size_t len;
2505 vm_offset_t src_addr;
2506{
2507 vm_offset_t addr;
2508 vm_offset_t end_addr = src_addr + len;
2509 vm_offset_t pdnxt;
2510 unsigned src_frame, dst_frame;
2511 vm_page_t m;
2512
2513 if (dst_addr != src_addr)
2514 return;
2515
2516 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2517 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2518 return;
2519 }
2520
2521 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2522 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2523 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2524#if defined(SMP)
2525 /* The page directory is not shared between CPUs */
2526 cpu_invltlb();
2527#else
2528 invltlb();
2529#endif
2530 }
2531
2532 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2533 unsigned *src_pte, *dst_pte;
2534 vm_page_t dstmpte, srcmpte;
2535 vm_offset_t srcptepaddr;
2536 unsigned ptepindex;
2537
2538 if (addr >= UPT_MIN_ADDRESS)
2539 panic("pmap_copy: invalid to pmap_copy page tables\n");
2540
2541 /*
2542 * Don't let optional prefaulting of pages make us go
2543 * way below the low water mark of free pages or way
2544 * above high water mark of used pv entries.
2545 */
12e4aaff 2546 if (vmstats.v_free_count < vmstats.v_free_reserved ||
984263bc
MD
2547 pv_entry_count > pv_entry_high_water)
2548 break;
2549
2550 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2551 ptepindex = addr >> PDRSHIFT;
2552
2553 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2554 if (srcptepaddr == 0)
2555 continue;
2556
2557 if (srcptepaddr & PG_PS) {
2558 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2559 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2560 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2561 }
2562 continue;
2563 }
2564
2565 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2566 if ((srcmpte == NULL) ||
2567 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2568 continue;
2569
2570 if (pdnxt > end_addr)
2571 pdnxt = end_addr;
2572
2573 src_pte = (unsigned *) vtopte(addr);
2574 dst_pte = (unsigned *) avtopte(addr);
2575 while (addr < pdnxt) {
2576 unsigned ptetemp;
2577 ptetemp = *src_pte;
2578 /*
2579 * we only virtual copy managed pages
2580 */
2581 if ((ptetemp & PG_MANAGED) != 0) {
2582 /*
2583 * We have to check after allocpte for the
2584 * pte still being around... allocpte can
2585 * block.
2586 */
2587 dstmpte = pmap_allocpte(dst_pmap, addr);
2588 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2589 /*
2590 * Clear the modified and
2591 * accessed (referenced) bits
2592 * during the copy.
2593 */
2594 m = PHYS_TO_VM_PAGE(ptetemp);
2595 *dst_pte = ptetemp & ~(PG_M | PG_A);
2596 dst_pmap->pm_stats.resident_count++;
2597 pmap_insert_entry(dst_pmap, addr,
2598 dstmpte, m);
2599 } else {
2600 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2601 }
2602 if (dstmpte->hold_count >= srcmpte->hold_count)
2603 break;
2604 }
2605 addr += PAGE_SIZE;
2606 src_pte++;
2607 dst_pte++;
2608 }
2609 }
2610}
2611
2612/*
2613 * Routine: pmap_kernel
2614 * Function:
2615 * Returns the physical map handle for the kernel.
2616 */
2617pmap_t
2618pmap_kernel()
2619{
2620 return (kernel_pmap);
2621}
2622
2623/*
2624 * pmap_zero_page zeros the specified hardware page by mapping
2625 * the page into KVM and using bzero to clear its contents.
2626 */
2627void
17a9f566 2628pmap_zero_page(vm_offset_t phys)
984263bc 2629{
85100692 2630 struct mdglobaldata *gd = mdcpu;
17a9f566 2631
85100692
MD
2632 if (*(int *)gd->gd_CMAP3)
2633 panic("pmap_zero_page: CMAP3 busy");
984263bc 2634
85100692 2635 *(int *)gd->gd_CMAP3 =
17a9f566 2636 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
85100692 2637 cpu_invlpg(gd->gd_CADDR3);
984263bc
MD
2638
2639#if defined(I686_CPU)
2640 if (cpu_class == CPUCLASS_686)
85100692 2641 i686_pagezero(gd->gd_CADDR3);
984263bc
MD
2642 else
2643#endif
85100692 2644 bzero(gd->gd_CADDR3, PAGE_SIZE);
984263bc 2645
85100692 2646 *(int *) gd->gd_CMAP3 = 0;
984263bc
MD
2647}
2648
2649/*
2650 * pmap_zero_page_area zeros the specified hardware page by mapping
2651 * the page into KVM and using bzero to clear its contents.
2652 *
2653 * off and size may not cover an area beyond a single hardware page.
2654 */
2655void
2656pmap_zero_page_area(phys, off, size)
2657 vm_offset_t phys;
2658 int off;
2659 int size;
2660{
85100692 2661 struct mdglobaldata *gd = mdcpu;
17a9f566 2662
85100692
MD
2663 if (*(int *) gd->gd_CMAP3)
2664 panic("pmap_zero_page: CMAP3 busy");
984263bc 2665
85100692
MD
2666 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2667 cpu_invlpg(gd->gd_CADDR3);
984263bc
MD
2668
2669#if defined(I686_CPU)
2670 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
85100692 2671 i686_pagezero(gd->gd_CADDR3);
984263bc
MD
2672 else
2673#endif
85100692 2674 bzero((char *)gd->gd_CADDR3 + off, size);
984263bc 2675
85100692 2676 *(int *) gd->gd_CMAP3 = 0;
984263bc
MD
2677}
2678
2679/*
2680 * pmap_copy_page copies the specified (machine independent)
2681 * page by mapping the page into virtual memory and using
2682 * bcopy to copy the page, one machine dependent page at a
2683 * time.
2684 */
2685void
2686pmap_copy_page(src, dst)
2687 vm_offset_t src;
2688 vm_offset_t dst;
2689{
85100692 2690 struct mdglobaldata *gd = mdcpu;
17a9f566 2691
85100692
MD
2692 if (*(int *) gd->gd_CMAP1)
2693 panic("pmap_copy_page: CMAP1 busy");
2694 if (*(int *) gd->gd_CMAP2)
2695 panic("pmap_copy_page: CMAP2 busy");
984263bc 2696
85100692
MD
2697 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2698 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
984263bc 2699
85100692
MD
2700 cpu_invlpg(gd->gd_CADDR1);
2701 cpu_invlpg(gd->gd_CADDR2);
984263bc 2702
85100692 2703 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
984263bc 2704
85100692
MD
2705 *(int *) gd->gd_CMAP1 = 0;
2706 *(int *) gd->gd_CMAP2 = 0;
984263bc
MD
2707}
2708
2709
2710/*
2711 * Routine: pmap_pageable
2712 * Function:
2713 * Make the specified pages (by pmap, offset)
2714 * pageable (or not) as requested.
2715 *
2716 * A page which is not pageable may not take
2717 * a fault; therefore, its page table entry
2718 * must remain valid for the duration.
2719 *
2720 * This routine is merely advisory; pmap_enter
2721 * will specify that these pages are to be wired
2722 * down (or not) as appropriate.
2723 */
2724void
2725pmap_pageable(pmap, sva, eva, pageable)
2726 pmap_t pmap;
2727 vm_offset_t sva, eva;
2728 boolean_t pageable;
2729{
2730}
2731
2732/*
2733 * Returns true if the pmap's pv is one of the first
2734 * 16 pvs linked to from this page. This count may
2735 * be changed upwards or downwards in the future; it
2736 * is only necessary that true be returned for a small
2737 * subset of pmaps for proper page aging.
2738 */
2739boolean_t
2740pmap_page_exists_quick(pmap, m)
2741 pmap_t pmap;
2742 vm_page_t m;
2743{
2744 pv_entry_t pv;
2745 int loops = 0;
2746 int s;
2747
2748 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2749 return FALSE;
2750
2751 s = splvm();
2752
2753 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2754 if (pv->pv_pmap == pmap) {
2755 splx(s);
2756 return TRUE;
2757 }
2758 loops++;
2759 if (loops >= 16)
2760 break;
2761 }
2762 splx(s);
2763 return (FALSE);
2764}
2765
2766#define PMAP_REMOVE_PAGES_CURPROC_ONLY
2767/*
2768 * Remove all pages from specified address space
2769 * this aids process exit speeds. Also, this code
2770 * is special cased for current process only, but
2771 * can have the more generic (and slightly slower)
2772 * mode enabled. This is much faster than pmap_remove
2773 * in the case of running down an entire address space.
2774 */
2775void
2776pmap_remove_pages(pmap, sva, eva)
2777 pmap_t pmap;
2778 vm_offset_t sva, eva;
2779{
2780 unsigned *pte, tpte;
2781 pv_entry_t pv, npv;
2782 int s;
2783 vm_page_t m;
2784
2785#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2786 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2787 printf("warning: pmap_remove_pages called with non-current pmap\n");
2788 return;
2789 }
2790#endif
2791
2792 s = splvm();
2793 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2794 pv;
2795 pv = npv) {
2796
2797 if (pv->pv_va >= eva || pv->pv_va < sva) {
2798 npv = TAILQ_NEXT(pv, pv_plist);
2799 continue;
2800 }
2801
2802#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2803 pte = (unsigned *)vtopte(pv->pv_va);
2804#else
2805 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2806#endif
2807 tpte = *pte;
2808
2809/*
2810 * We cannot remove wired pages from a process' mapping at this time
2811 */
2812 if (tpte & PG_W) {
2813 npv = TAILQ_NEXT(pv, pv_plist);
2814 continue;
2815 }
2816 *pte = 0;
2817
2818 m = PHYS_TO_VM_PAGE(tpte);
2819
2820 KASSERT(m < &vm_page_array[vm_page_array_size],
2821 ("pmap_remove_pages: bad tpte %x", tpte));
2822
2823 pv->pv_pmap->pm_stats.resident_count--;
2824
2825 /*
2826 * Update the vm_page_t clean and reference bits.
2827 */
2828 if (tpte & PG_M) {
2829 vm_page_dirty(m);
2830 }
2831
2832
2833 npv = TAILQ_NEXT(pv, pv_plist);
2834 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2835
2836 m->md.pv_list_count--;
2837 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2838 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2839 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2840 }
2841
2842 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2843 free_pv_entry(pv);
2844 }
2845 splx(s);
2846 pmap_TLB_invalidate_all(pmap);
2847}
2848
2849/*
2850 * pmap_testbit tests bits in pte's
2851 * note that the testbit/changebit routines are inline,
2852 * and a lot of things compile-time evaluate.
2853 */
2854static boolean_t
2855pmap_testbit(m, bit)
2856 vm_page_t m;
2857 int bit;
2858{
2859 pv_entry_t pv;
2860 unsigned *pte;
2861 int s;
2862
2863 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2864 return FALSE;
2865
2866 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
2867 return FALSE;
2868
2869 s = splvm();
2870
2871 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2872 /*
2873 * if the bit being tested is the modified bit, then
2874 * mark clean_map and ptes as never
2875 * modified.
2876 */
2877 if (bit & (PG_A|PG_M)) {
2878 if (!pmap_track_modified(pv->pv_va))
2879 continue;
2880 }
2881
2882#if defined(PMAP_DIAGNOSTIC)
2883 if (!pv->pv_pmap) {
2884 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
2885 continue;
2886 }
2887#endif
2888 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2889 if (*pte & bit) {
2890 splx(s);
2891 return TRUE;
2892 }
2893 }
2894 splx(s);
2895 return (FALSE);
2896}
2897
2898/*
2899 * this routine is used to modify bits in ptes
2900 */
2901static __inline void
2902pmap_changebit(m, bit, setem)
2903 vm_page_t m;
2904 int bit;
2905 boolean_t setem;
2906{
2907 register pv_entry_t pv;
2908 register unsigned *pte;
2909 int s;
2910
2911 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2912 return;
2913
2914 s = splvm();
2915
2916 /*
2917 * Loop over all current mappings setting/clearing as appropos If
2918 * setting RO do we need to clear the VAC?
2919 */
2920 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2921 /*
2922 * don't write protect pager mappings
2923 */
2924 if (!setem && (bit == PG_RW)) {
2925 if (!pmap_track_modified(pv->pv_va))
2926 continue;
2927 }
2928
2929#if defined(PMAP_DIAGNOSTIC)
2930 if (!pv->pv_pmap) {
2931 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
2932 continue;
2933 }
2934#endif
2935
2936 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2937
2938 if (setem) {
2939 *(int *)pte |= bit;
2940 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2941 } else {
2942 vm_offset_t pbits = *(vm_offset_t *)pte;
2943 if (pbits & bit) {
2944 if (bit == PG_RW) {
2945 if (pbits & PG_M) {
2946 vm_page_dirty(m);
2947 }
2948 *(int *)pte = pbits & ~(PG_M|PG_RW);
2949 } else {
2950 *(int *)pte = pbits & ~bit;
2951 }
2952 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2953 }
2954 }
2955 }
2956 splx(s);
2957}
2958
2959/*
2960 * pmap_page_protect:
2961 *
2962 * Lower the permission for all mappings to a given page.
2963 */
2964void
2965pmap_page_protect(vm_page_t m, vm_prot_t prot)
2966{
2967 if ((prot & VM_PROT_WRITE) == 0) {
2968 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
2969 pmap_changebit(m, PG_RW, FALSE);
2970 } else {
2971 pmap_remove_all(m);
2972 }
2973 }
2974}
2975
2976vm_offset_t
2977pmap_phys_address(ppn)
2978 int ppn;
2979{
2980 return (i386_ptob(ppn));
2981}
2982
2983/*
2984 * pmap_ts_referenced:
2985 *
2986 * Return a count of reference bits for a page, clearing those bits.
2987 * It is not necessary for every reference bit to be cleared, but it
2988 * is necessary that 0 only be returned when there are truly no
2989 * reference bits set.
2990 *
2991 * XXX: The exact number of bits to check and clear is a matter that
2992 * should be tested and standardized at some point in the future for
2993 * optimal aging of shared pages.
2994 */
2995int
2996pmap_ts_referenced(vm_page_t m)
2997{
2998 register pv_entry_t pv, pvf, pvn;
2999 unsigned *pte;
3000 int s;
3001 int rtval = 0;
3002
3003 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3004 return (rtval);
3005
3006 s = splvm();
3007
3008 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3009
3010 pvf = pv;
3011
3012 do {
3013 pvn = TAILQ_NEXT(pv, pv_list);
3014
3015 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3016
3017 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3018
3019 if (!pmap_track_modified(pv->pv_va))
3020 continue;
3021
3022 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3023
3024 if (pte && (*pte & PG_A)) {
3025 *pte &= ~PG_A;
3026
3027 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3028
3029 rtval++;
3030 if (rtval > 4) {
3031 break;
3032 }
3033 }
3034 } while ((pv = pvn) != NULL && pv != pvf);
3035 }
3036 splx(s);
3037
3038 return (rtval);
3039}
3040
3041/*
3042 * pmap_is_modified:
3043 *
3044 * Return whether or not the specified physical page was modified
3045 * in any physical maps.
3046 */
3047boolean_t
3048pmap_is_modified(vm_page_t m)
3049{
3050 return pmap_testbit(m, PG_M);
3051}
3052
3053/*
3054 * Clear the modify bits on the specified physical page.
3055 */
3056void
3057pmap_clear_modify(vm_page_t m)
3058{
3059 pmap_changebit(m, PG_M, FALSE);
3060}
3061
3062/*
3063 * pmap_clear_reference:
3064 *
3065 * Clear the reference bit on the specified physical page.
3066 */
3067void
3068pmap_clear_reference(vm_page_t m)
3069{
3070 pmap_changebit(m, PG_A, FALSE);
3071}
3072
3073/*
3074 * Miscellaneous support routines follow
3075 */
3076
3077static void
3078i386_protection_init()
3079{
3080 register int *kp, prot;
3081
3082 kp = protection_codes;
3083 for (prot = 0; prot < 8; prot++) {
3084 switch (prot) {
3085 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3086 /*
3087 * Read access is also 0. There isn't any execute bit,
3088 * so just make it readable.
3089 */
3090 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3091 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3092 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3093 *kp++ = 0;
3094 break;
3095 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3096 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3097 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3098 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3099 *kp++ = PG_RW;
3100 break;
3101 }
3102 }
3103}
3104
3105/*
3106 * Map a set of physical memory pages into the kernel virtual
3107 * address space. Return a pointer to where it is mapped. This
3108 * routine is intended to be used for mapping device memory,
3109 * NOT real memory.
a2a5ad0d
MD
3110 *
3111 * NOTE: we can't use pgeflag unless we invalidate the pages one at
3112 * a time.
984263bc
MD
3113 */
3114void *
3115pmap_mapdev(pa, size)
3116 vm_offset_t pa;
3117 vm_size_t size;
3118{
3119 vm_offset_t va, tmpva, offset;
3120 unsigned *pte;
3121
3122 offset = pa & PAGE_MASK;
3123 size = roundup(offset + size, PAGE_SIZE);
3124
3125 va = kmem_alloc_pageable(kernel_map, size);
3126 if (!va)
3127 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3128
3129 pa = pa & PG_FRAME;
3130 for (tmpva = va; size > 0;) {
3131 pte = (unsigned *)vtopte(tmpva);
a2a5ad0d 3132 *pte = pa | PG_RW | PG_V; /* | pgeflag; */
984263bc
MD
3133 size -= PAGE_SIZE;
3134 tmpva += PAGE_SIZE;
3135 pa += PAGE_SIZE;
3136 }
3137 invltlb();
3138
3139 return ((void *)(va + offset));
3140}
3141
3142void
3143pmap_unmapdev(va, size)
3144 vm_offset_t va;
3145 vm_size_t size;
3146{
3147 vm_offset_t base, offset;
3148
3149 base = va & PG_FRAME;
3150 offset = va & PAGE_MASK;
3151 size = roundup(offset + size, PAGE_SIZE);
3152 kmem_free(kernel_map, base, size);
3153}
3154
3155/*
3156 * perform the pmap work for mincore
3157 */
3158int
3159pmap_mincore(pmap, addr)
3160 pmap_t pmap;
3161 vm_offset_t addr;
3162{
3163
3164 unsigned *ptep, pte;
3165 vm_page_t m;
3166 int val = 0;
3167
3168 ptep = pmap_pte(pmap, addr);
3169 if (ptep == 0) {
3170 return 0;
3171 }
3172
3173 if ((pte = *ptep) != 0) {
3174 vm_offset_t pa;
3175
3176 val = MINCORE_INCORE;
3177 if ((pte & PG_MANAGED) == 0)
3178 return val;
3179
3180 pa = pte & PG_FRAME;
3181
3182 m = PHYS_TO_VM_PAGE(pa);
3183
3184 /*
3185 * Modified by us
3186 */
3187 if (pte & PG_M)
3188 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3189 /*
3190 * Modified by someone
3191 */
3192 else if (m->dirty || pmap_is_modified(m))
3193 val |= MINCORE_MODIFIED_OTHER;
3194 /*
3195 * Referenced by us
3196 */
3197 if (pte & PG_A)
3198 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3199
3200 /*
3201 * Referenced by someone
3202 */
3203 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3204 val |= MINCORE_REFERENCED_OTHER;
3205 vm_page_flag_set(m, PG_REFERENCED);
3206 }
3207 }
3208 return val;
3209}
3210
3211void
3212pmap_activate(struct proc *p)
3213{
3214 pmap_t pmap;
3215
3216 pmap = vmspace_pmap(p->p_vmspace);
3217#if defined(SMP)
a2a5ad0d 3218 atomic_set_int(&pmap->pm_active, 1 << mycpu->gd_cpuid);
984263bc
MD
3219#else
3220 pmap->pm_active |= 1;
3221#endif
3222#if defined(SWTCH_OPTIM_STATS)
3223 tlb_flush_count++;
3224#endif
b7c628e4
MD
3225 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3226 load_cr3(p->p_thread->td_pcb->pcb_cr3);
984263bc
MD
3227}
3228
3229vm_offset_t
3230pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3231{
3232
3233 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3234 return addr;
3235 }
3236
3237 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3238 return addr;
3239}
3240
3241
3242#if defined(PMAP_DEBUG)
3243pmap_pid_dump(int pid)
3244{
3245 pmap_t pmap;
3246 struct proc *p;
3247 int npte = 0;
3248 int index;
3249 LIST_FOREACH(p, &allproc, p_list) {
3250 if (p->p_pid != pid)
3251 continue;
3252
3253 if (p->p_vmspace) {
3254 int i,j;
3255 index = 0;
3256 pmap = vmspace_pmap(p->p_vmspace);
3257 for(i=0;i<1024;i++) {
3258 pd_entry_t *pde;
3259 unsigned *pte;
3260 unsigned base = i << PDRSHIFT;
3261
3262 pde = &pmap->pm_pdir[i];
3263 if (pde && pmap_pde_v(pde)) {
3264 for(j=0;j<1024;j++) {
3265 unsigned va = base + (j << PAGE_SHIFT);
3266 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3267 if (index) {
3268 index = 0;
3269 printf("\n");
3270 }
3271 return npte;
3272 }
3273 pte = pmap_pte_quick( pmap, va);
3274 if (pte && pmap_pte_v(pte)) {
3275 vm_offset_t pa;
3276 vm_page_t m;
3277 pa = *(int *)pte;
3278 m = PHYS_TO_VM_PAGE(pa);
3279 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3280 va, pa, m->hold_count, m->wire_count, m->flags);
3281 npte++;
3282 index++;
3283 if (index >= 2) {
3284 index = 0;
3285 printf("\n");
3286 } else {
3287 printf(" ");
3288 }
3289 }
3290 }
3291 }
3292 }
3293 }
3294 }
3295 return npte;
3296}
3297#endif
3298
3299#if defined(DEBUG)
3300
3301static void pads __P((pmap_t pm));
3302void pmap_pvdump __P((vm_offset_t pa));
3303
3304/* print address space of pmap*/
3305static void
3306pads(pm)
3307 pmap_t pm;
3308{
3309 unsigned va, i, j;
3310 unsigned *ptep;
3311
3312 if (pm == kernel_pmap)
3313 return;
3314 for (i = 0; i < 1024; i++)
3315 if (pm->pm_pdir[i])
3316 for (j = 0; j < 1024; j++) {
3317 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3318 if (pm == kernel_pmap && va < KERNBASE)
3319 continue;
3320 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3321 continue;
3322 ptep = pmap_pte_quick(pm, va);
3323 if (pmap_pte_v(ptep))
3324 printf("%x:%x ", va, *(int *) ptep);
3325 };
3326
3327}
3328
3329void
3330pmap_pvdump(pa)
3331 vm_offset_t pa;
3332{
3333 register pv_entry_t pv;
3334 vm_page_t m;
3335
3336 printf("pa %x", pa);
3337 m = PHYS_TO_VM_PAGE(pa);
3338 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3339#ifdef used_to_be
3340 printf(" -> pmap %p, va %x, flags %x",
3341 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3342#endif
3343 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);
3344 pads(pv->pv_pmap);
3345 }
3346 printf(" ");
3347}
3348#endif