kernel - Check PG_MARKER in pmap_object_init_pt_callback()
[dragonfly.git] / sys / platform / vkernel64 / platform / pmap.c
CommitLineData
da673940 1/*
f7230403
MD
2 * (MPSAFE)
3 *
da673940
JG
4 * Copyright (c) 1991 Regents of the University of California.
5 * Copyright (c) 1994 John S. Dyson
6 * Copyright (c) 1994 David Greenman
7 * Copyright (c) 2003 Peter Wemm
8 * Copyright (c) 2005-2008 Alan L. Cox <alc@cs.rice.edu>
9 * Copyright (c) 2008, 2009 The DragonFly Project.
10 * Copyright (c) 2008, 2009 Jordan Gordeev.
11 * All rights reserved.
12 *
13 * This code is derived from software contributed to Berkeley by
14 * the Systems Programming Group of the University of Utah Computer
15 * Science Department and William Jolitz of UUNET Technologies Inc.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions
19 * are met:
20 * 1. Redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer.
22 * 2. Redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution.
25 * 3. All advertising materials mentioning features or use of this software
26 * must display the following acknowledgement:
27 * This product includes software developed by the University of
28 * California, Berkeley and its contributors.
29 * 4. Neither the name of the University nor the names of its contributors
30 * may be used to endorse or promote products derived from this software
31 * without specific prior written permission.
32 *
33 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
34 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
35 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
36 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
37 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
38 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
39 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
41 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
42 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 * SUCH DAMAGE.
44 *
45 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91
46 * $FreeBSD: src/sys/i386/i386/pmap.c,v 1.250.2.18 2002/03/06 22:48:53 silby Exp $
da673940
JG
47 */
48
49/*
f7230403 50 * Manages physical address maps.
da673940
JG
51 */
52
53#if JG
54#include "opt_pmap.h"
55#endif
56#include "opt_msgbuf.h"
57
58#include <sys/param.h>
59#include <sys/systm.h>
60#include <sys/kernel.h>
61#include <sys/proc.h>
62#include <sys/msgbuf.h>
63#include <sys/vmmeter.h>
64#include <sys/mman.h>
65#include <sys/vmspace.h>
66
67#include <vm/vm.h>
68#include <vm/vm_param.h>
69#include <sys/sysctl.h>
70#include <sys/lock.h>
71#include <vm/vm_kern.h>
72#include <vm/vm_page.h>
73#include <vm/vm_map.h>
74#include <vm/vm_object.h>
75#include <vm/vm_extern.h>
76#include <vm/vm_pageout.h>
77#include <vm/vm_pager.h>
78#include <vm/vm_zone.h>
79
80#include <sys/user.h>
81#include <sys/thread2.h>
82#include <sys/sysref2.h>
b12defdc 83#include <sys/spinlock2.h>
da673940
JG
84
85#include <machine/cputypes.h>
86#include <machine/md_var.h>
87#include <machine/specialreg.h>
88#include <machine/smp.h>
89#include <machine/globaldata.h>
90#include <machine/pmap.h>
91#include <machine/pmap_inval.h>
92
93#include <ddb/ddb.h>
94
95#include <stdio.h>
96#include <assert.h>
97#include <stdlib.h>
98
99#define PMAP_KEEP_PDIRS
100#ifndef PMAP_SHPGPERPROC
f1d3f422 101#define PMAP_SHPGPERPROC 1000
da673940
JG
102#endif
103
104#if defined(DIAGNOSTIC)
105#define PMAP_DIAGNOSTIC
106#endif
107
108#define MINPV 2048
109
110#if !defined(PMAP_DIAGNOSTIC)
111#define PMAP_INLINE __inline
112#else
113#define PMAP_INLINE
114#endif
115
116/*
117 * Get PDEs and PTEs for user/kernel address space
118 */
119static pd_entry_t *pmap_pde(pmap_t pmap, vm_offset_t va);
120#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
121
122#define pmap_pde_v(pte) ((*(pd_entry_t *)pte & VPTE_V) != 0)
123#define pmap_pte_w(pte) ((*(pt_entry_t *)pte & VPTE_WIRED) != 0)
124#define pmap_pte_m(pte) ((*(pt_entry_t *)pte & VPTE_M) != 0)
125#define pmap_pte_u(pte) ((*(pt_entry_t *)pte & VPTE_A) != 0)
126#define pmap_pte_v(pte) ((*(pt_entry_t *)pte & VPTE_V) != 0)
127
128/*
129 * Given a map and a machine independent protection code,
130 * convert to a vax protection code.
131 */
132#define pte_prot(m, p) \
133 (protection_codes[p & (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)])
134static int protection_codes[8];
135
136struct pmap kernel_pmap;
137static TAILQ_HEAD(,pmap) pmap_list = TAILQ_HEAD_INITIALIZER(pmap_list);
138
139static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
140
141static vm_object_t kptobj;
142
143static int nkpt;
144
145static uint64_t KPDphys; /* phys addr of kernel level 2 */
146uint64_t KPDPphys; /* phys addr of kernel level 3 */
147uint64_t KPML4phys; /* phys addr of kernel level 4 */
148
149
150/*
151 * Data for the pv entry allocation mechanism
152 */
153static vm_zone_t pvzone;
154static struct vm_zone pvzone_store;
155static struct vm_object pvzone_obj;
156static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
157static int pmap_pagedaemon_waken = 0;
158static struct pv_entry *pvinit;
159
160/*
161 * All those kernel PT submaps that BSD is so fond of
162 */
163pt_entry_t *CMAP1 = 0, *ptmmap;
164caddr_t CADDR1 = 0;
165static pt_entry_t *msgbufmap;
166
167uint64_t KPTphys;
168
169static PMAP_INLINE void free_pv_entry (pv_entry_t pv);
170static pv_entry_t get_pv_entry (void);
171static void i386_protection_init (void);
172static __inline void pmap_clearbit (vm_page_t m, int bit);
173
174static void pmap_remove_all (vm_page_t m);
da673940
JG
175static int pmap_remove_pte (struct pmap *pmap, pt_entry_t *ptq,
176 vm_offset_t sva);
177static void pmap_remove_page (struct pmap *pmap, vm_offset_t va);
178static int pmap_remove_entry (struct pmap *pmap, vm_page_t m,
179 vm_offset_t va);
180static boolean_t pmap_testbit (vm_page_t m, int bit);
181static void pmap_insert_entry (pmap_t pmap, vm_offset_t va,
182 vm_page_t mpte, vm_page_t m);
183
184static vm_page_t pmap_allocpte (pmap_t pmap, vm_offset_t va);
185
186static int pmap_release_free_page (pmap_t pmap, vm_page_t p);
187static vm_page_t _pmap_allocpte (pmap_t pmap, vm_pindex_t ptepindex);
2c2e847c 188#if JGPMAP32
da673940 189static pt_entry_t * pmap_pte_quick (pmap_t pmap, vm_offset_t va);
2c2e847c 190#endif
da673940 191static vm_page_t pmap_page_lookup (vm_object_t object, vm_pindex_t pindex);
da673940
JG
192static int pmap_unuse_pt (pmap_t, vm_offset_t, vm_page_t);
193
194/*
195 * pmap_pte_quick:
196 *
197 * Super fast pmap_pte routine best used when scanning the pv lists.
198 * This eliminates many course-grained invltlb calls. Note that many of
199 * the pv list scans are across different pmaps and it is very wasteful
200 * to do an entire invltlb when checking a single mapping.
201 *
202 * Should only be called while in a critical section.
203 */
2c2e847c 204#if JGPMAP32
da673940
JG
205static __inline pt_entry_t *pmap_pte(pmap_t pmap, vm_offset_t va);
206
207static pt_entry_t *
208pmap_pte_quick(pmap_t pmap, vm_offset_t va)
209{
210 return pmap_pte(pmap, va);
211}
2c2e847c 212#endif
da673940
JG
213
214/* Return a non-clipped PD index for a given VA */
215static __inline vm_pindex_t
216pmap_pde_pindex(vm_offset_t va)
217{
218 return va >> PDRSHIFT;
219}
220
221/* Return various clipped indexes for a given VA */
222static __inline vm_pindex_t
223pmap_pte_index(vm_offset_t va)
224{
225
226 return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1));
227}
228
229static __inline vm_pindex_t
230pmap_pde_index(vm_offset_t va)
231{
232
233 return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
234}
235
236static __inline vm_pindex_t
237pmap_pdpe_index(vm_offset_t va)
238{
239
240 return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1));
241}
242
243static __inline vm_pindex_t
244pmap_pml4e_index(vm_offset_t va)
245{
246
247 return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1));
248}
249
250/* Return a pointer to the PML4 slot that corresponds to a VA */
251static __inline pml4_entry_t *
252pmap_pml4e(pmap_t pmap, vm_offset_t va)
253{
254
255 return (&pmap->pm_pml4[pmap_pml4e_index(va)]);
256}
257
258/* Return a pointer to the PDP slot that corresponds to a VA */
259static __inline pdp_entry_t *
260pmap_pml4e_to_pdpe(pml4_entry_t *pml4e, vm_offset_t va)
261{
262 pdp_entry_t *pdpe;
263
264 pdpe = (pdp_entry_t *)PHYS_TO_DMAP(*pml4e & VPTE_FRAME);
265 return (&pdpe[pmap_pdpe_index(va)]);
266}
267
268/* Return a pointer to the PDP slot that corresponds to a VA */
269static __inline pdp_entry_t *
270pmap_pdpe(pmap_t pmap, vm_offset_t va)
271{
272 pml4_entry_t *pml4e;
273
274 pml4e = pmap_pml4e(pmap, va);
275 if ((*pml4e & VPTE_V) == 0)
276 return NULL;
277 return (pmap_pml4e_to_pdpe(pml4e, va));
278}
279
280/* Return a pointer to the PD slot that corresponds to a VA */
281static __inline pd_entry_t *
282pmap_pdpe_to_pde(pdp_entry_t *pdpe, vm_offset_t va)
283{
284 pd_entry_t *pde;
285
286 pde = (pd_entry_t *)PHYS_TO_DMAP(*pdpe & VPTE_FRAME);
287 return (&pde[pmap_pde_index(va)]);
288}
289
290/* Return a pointer to the PD slot that corresponds to a VA */
291static __inline pd_entry_t *
292pmap_pde(pmap_t pmap, vm_offset_t va)
293{
294 pdp_entry_t *pdpe;
295
296 pdpe = pmap_pdpe(pmap, va);
297 if (pdpe == NULL || (*pdpe & VPTE_V) == 0)
298 return NULL;
299 return (pmap_pdpe_to_pde(pdpe, va));
300}
301
302/* Return a pointer to the PT slot that corresponds to a VA */
303static __inline pt_entry_t *
304pmap_pde_to_pte(pd_entry_t *pde, vm_offset_t va)
305{
306 pt_entry_t *pte;
307
308 pte = (pt_entry_t *)PHYS_TO_DMAP(*pde & VPTE_FRAME);
309 return (&pte[pmap_pte_index(va)]);
310}
311
312/* Return a pointer to the PT slot that corresponds to a VA */
313static __inline pt_entry_t *
314pmap_pte(pmap_t pmap, vm_offset_t va)
315{
316 pd_entry_t *pde;
317
318 pde = pmap_pde(pmap, va);
319 if (pde == NULL || (*pde & VPTE_V) == 0)
320 return NULL;
321 if ((*pde & VPTE_PS) != 0) /* compat with i386 pmap_pte() */
322 return ((pt_entry_t *)pde);
323 return (pmap_pde_to_pte(pde, va));
324}
325
326
327#if JGV
328PMAP_INLINE pt_entry_t *
329vtopte(vm_offset_t va)
330{
b12defdc
MD
331 uint64_t mask = ((1ul << (NPTEPGSHIFT + NPDEPGSHIFT +
332 NPDPEPGSHIFT + NPML4EPGSHIFT)) - 1);
da673940
JG
333
334 return (PTmap + ((va >> PAGE_SHIFT) & mask));
335}
336
337static __inline pd_entry_t *
338vtopde(vm_offset_t va)
339{
b12defdc
MD
340 uint64_t mask = ((1ul << (NPDEPGSHIFT + NPDPEPGSHIFT +
341 NPML4EPGSHIFT)) - 1);
da673940
JG
342
343 return (PDmap + ((va >> PDRSHIFT) & mask));
344}
345#else
2c2e847c 346static PMAP_INLINE pt_entry_t *
da673940
JG
347vtopte(vm_offset_t va)
348{
349 pt_entry_t *x;
350 x = pmap_pte(&kernel_pmap, va);
351 assert(x != NULL);
352 return x;
353}
354
355static __inline pd_entry_t *
356vtopde(vm_offset_t va)
357{
358 pd_entry_t *x;
359 x = pmap_pde(&kernel_pmap, va);
360 assert(x != NULL);
361 return x;
362}
363#endif
364
365static uint64_t
366allocpages(vm_paddr_t *firstaddr, int n)
367{
368 uint64_t ret;
369
370 ret = *firstaddr;
371#if JGV
372 bzero((void *)ret, n * PAGE_SIZE);
373#endif
374 *firstaddr += n * PAGE_SIZE;
375 return (ret);
376}
377
2c2e847c 378static void
da673940
JG
379create_pagetables(vm_paddr_t *firstaddr, int64_t ptov_offset)
380{
381 int i;
382 pml4_entry_t *KPML4virt;
383 pdp_entry_t *KPDPvirt;
384 pd_entry_t *KPDvirt;
385 pt_entry_t *KPTvirt;
386 int kpml4i = pmap_pml4e_index(ptov_offset);
387 int kpdpi = pmap_pdpe_index(ptov_offset);
388
ad54aa11
MD
389 /*
390 * Calculate NKPT - number of kernel page tables. We have to
391 * accomodoate prealloction of the vm_page_array, dump bitmap,
392 * MSGBUF_SIZE, and other stuff. Be generous.
393 *
394 * Maxmem is in pages.
395 */
396 nkpt = (Maxmem * (sizeof(struct vm_page) * 2) + MSGBUF_SIZE) / NBPDR;
da673940 397
ad54aa11
MD
398 /*
399 * Allocate pages
400 */
da673940
JG
401 KPML4phys = allocpages(firstaddr, 1);
402 KPDPphys = allocpages(firstaddr, NKPML4E);
403 KPDphys = allocpages(firstaddr, NKPDPE);
ad54aa11 404 KPTphys = allocpages(firstaddr, nkpt);
da673940
JG
405
406 KPML4virt = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys);
407 KPDPvirt = (pdp_entry_t *)PHYS_TO_DMAP(KPDPphys);
408 KPDvirt = (pd_entry_t *)PHYS_TO_DMAP(KPDphys);
409 KPTvirt = (pt_entry_t *)PHYS_TO_DMAP(KPTphys);
410
411 bzero(KPML4virt, 1 * PAGE_SIZE);
412 bzero(KPDPvirt, NKPML4E * PAGE_SIZE);
413 bzero(KPDvirt, NKPDPE * PAGE_SIZE);
ad54aa11 414 bzero(KPTvirt, nkpt * PAGE_SIZE);
da673940
JG
415
416 /* Now map the page tables at their location within PTmap */
ad54aa11 417 for (i = 0; i < nkpt; i++) {
da673940
JG
418 KPDvirt[i] = KPTphys + (i << PAGE_SHIFT);
419 KPDvirt[i] |= VPTE_R | VPTE_W | VPTE_V;
420 }
421
422 /* And connect up the PD to the PDP */
423 for (i = 0; i < NKPDPE; i++) {
424 KPDPvirt[i + kpdpi] = KPDphys + (i << PAGE_SHIFT);
425 KPDPvirt[i + kpdpi] |= VPTE_R | VPTE_W | VPTE_V;
426 }
427
428 /* And recursively map PML4 to itself in order to get PTmap */
429 KPML4virt[PML4PML4I] = KPML4phys;
430 KPML4virt[PML4PML4I] |= VPTE_R | VPTE_W | VPTE_V;
431
432 /* Connect the KVA slot up to the PML4 */
433 KPML4virt[kpml4i] = KPDPphys;
434 KPML4virt[kpml4i] |= VPTE_R | VPTE_W | VPTE_V;
435}
436
437/*
438 * Bootstrap the system enough to run with virtual memory.
439 *
440 * On the i386 this is called after mapping has already been enabled
441 * and just syncs the pmap module with what has already been done.
442 * [We can't call it easily with mapping off since the kernel is not
443 * mapped with PA == VA, hence we would have to relocate every address
444 * from the linked base (virtual) address "KERNBASE" to the actual
445 * (physical) address starting relative to 0]
446 */
447void
448pmap_bootstrap(vm_paddr_t *firstaddr, int64_t ptov_offset)
449{
450 vm_offset_t va;
451 pt_entry_t *pte;
452
453 /*
454 * Create an initial set of page tables to run the kernel in.
455 */
456 create_pagetables(firstaddr, ptov_offset);
457
458 virtual_start = KvaStart + *firstaddr;
459 virtual_end = KvaEnd;
460
461 /*
462 * Initialize protection array.
463 */
464 i386_protection_init();
465
466 /*
467 * The kernel's pmap is statically allocated so we don't have to use
468 * pmap_create, which is unlikely to work correctly at this part of
469 * the boot sequence (XXX and which no longer exists).
b12defdc
MD
470 *
471 * The kernel_pmap's pm_pteobj is used only for locking and not
472 * for mmu pages.
da673940
JG
473 */
474 kernel_pmap.pm_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys);
475 kernel_pmap.pm_count = 1;
476 kernel_pmap.pm_active = (cpumask_t)-1; /* don't allow deactivation */
b12defdc 477 kernel_pmap.pm_pteobj = &kernel_object;
da673940 478 TAILQ_INIT(&kernel_pmap.pm_pvlist);
b12defdc
MD
479 TAILQ_INIT(&kernel_pmap.pm_pvlist_free);
480 lwkt_token_init(&kernel_pmap.pm_token, "kpmap_tok");
481 spin_init(&kernel_pmap.pm_spin);
da673940
JG
482
483 /*
484 * Reserve some special page table entries/VA space for temporary
485 * mapping of pages.
486 */
487#define SYSMAP(c, p, v, n) \
488 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
489
490 va = virtual_start;
491 pte = pmap_pte(&kernel_pmap, va);
492
493 /*
494 * CMAP1/CMAP2 are used for zeroing and copying pages.
495 */
496 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
497
498#if JGV
499 /*
500 * Crashdump maps.
501 */
502 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
503#endif
504
505 /*
506 * ptvmmap is used for reading arbitrary physical pages via
507 * /dev/mem.
508 */
509 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
510
511 /*
512 * msgbufp is used to map the system message buffer.
513 * XXX msgbufmap is not used.
514 */
515 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
516 atop(round_page(MSGBUF_SIZE)))
517
518 virtual_start = va;
519
520 *CMAP1 = 0;
521
522 cpu_invltlb();
523}
524
525/*
526 * Initialize the pmap module.
527 * Called by vm_init, to initialize any structures that the pmap
528 * system needs to map virtual memory.
529 * pmap_init has been enhanced to support in a fairly consistant
530 * way, discontiguous physical memory.
531 */
532void
533pmap_init(void)
534{
535 int i;
536 int initial_pvs;
537
538 /*
539 * object for kernel page table pages
540 */
541 /* JG I think the number can be arbitrary */
542 kptobj = vm_object_allocate(OBJT_DEFAULT, 5);
543
544 /*
545 * Allocate memory for random pmap data structures. Includes the
546 * pv_head_table.
547 */
548
549 for(i = 0; i < vm_page_array_size; i++) {
550 vm_page_t m;
551
552 m = &vm_page_array[i];
553 TAILQ_INIT(&m->md.pv_list);
554 m->md.pv_list_count = 0;
555 }
556
557 /*
558 * init the pv free list
559 */
560 initial_pvs = vm_page_array_size;
561 if (initial_pvs < MINPV)
562 initial_pvs = MINPV;
563 pvzone = &pvzone_store;
564 pvinit = (struct pv_entry *) kmem_alloc(&kernel_map,
565 initial_pvs * sizeof (struct pv_entry));
566 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
567 initial_pvs);
568
569 /*
570 * Now it is safe to enable pv_table recording.
571 */
572 pmap_initialized = TRUE;
573}
574
575/*
576 * Initialize the address space (zone) for the pv_entries. Set a
577 * high water mark so that the system can recover from excessive
578 * numbers of pv entries.
579 */
580void
581pmap_init2(void)
582{
583 int shpgperproc = PMAP_SHPGPERPROC;
584
585 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
586 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
587 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
588 pv_entry_high_water = 9 * (pv_entry_max / 10);
589 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
590}
591
592
593/***************************************************
594 * Low level helper routines.....
595 ***************************************************/
596
597/*
598 * The modification bit is not tracked for any pages in this range. XXX
599 * such pages in this maps should always use pmap_k*() functions and not
600 * be managed anyhow.
601 *
602 * XXX User and kernel address spaces are independant for virtual kernels,
603 * this function only applies to the kernel pmap.
604 */
605static int
606pmap_track_modified(pmap_t pmap, vm_offset_t va)
607{
608 if (pmap != &kernel_pmap)
609 return 1;
610 if ((va < clean_sva) || (va >= clean_eva))
611 return 1;
612 else
613 return 0;
614}
615
616/*
f7230403 617 * Extract the physical page address associated with the map/VA pair.
da673940 618 *
f7230403 619 * No requirements.
da673940
JG
620 */
621vm_paddr_t
622pmap_extract(pmap_t pmap, vm_offset_t va)
623{
624 vm_paddr_t rtval;
625 pt_entry_t *pte;
626 pd_entry_t pde, *pdep;
627
f7230403 628 lwkt_gettoken(&vm_token);
da673940
JG
629 rtval = 0;
630 pdep = pmap_pde(pmap, va);
631 if (pdep != NULL) {
632 pde = *pdep;
633 if (pde) {
634 if ((pde & VPTE_PS) != 0) {
635 /* JGV */
636 rtval = (pde & PG_PS_FRAME) | (va & PDRMASK);
637 } else {
638 pte = pmap_pde_to_pte(pdep, va);
639 rtval = (*pte & VPTE_FRAME) | (va & PAGE_MASK);
640 }
641 }
642 }
f7230403 643 lwkt_reltoken(&vm_token);
da673940
JG
644 return rtval;
645}
646
647/*
648 * Routine: pmap_kextract
649 * Function:
650 * Extract the physical page address associated
651 * kernel virtual address.
652 */
653vm_paddr_t
654pmap_kextract(vm_offset_t va)
655{
656 pd_entry_t pde;
657 vm_paddr_t pa;
658
659 KKASSERT(va >= KvaStart && va < KvaEnd);
660
661 /*
662 * The DMAP region is not included in [KvaStart, KvaEnd)
663 */
664#if 0
665 if (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS) {
666 pa = DMAP_TO_PHYS(va);
667 } else {
668#endif
669 pde = *vtopde(va);
670 if (pde & VPTE_PS) {
671 /* JGV */
672 pa = (pde & PG_PS_FRAME) | (va & PDRMASK);
673 } else {
674 /*
675 * Beware of a concurrent promotion that changes the
676 * PDE at this point! For example, vtopte() must not
677 * be used to access the PTE because it would use the
678 * new PDE. It is, however, safe to use the old PDE
679 * because the page table page is preserved by the
680 * promotion.
681 */
682 pa = *pmap_pde_to_pte(&pde, va);
683 pa = (pa & VPTE_FRAME) | (va & PAGE_MASK);
684 }
685#if 0
686 }
687#endif
688 return pa;
689}
690
691/***************************************************
692 * Low level mapping routines.....
693 ***************************************************/
694
695/*
696 * Enter a mapping into kernel_pmap. Mappings created in this fashion
697 * are not managed. Mappings must be immediately accessible on all cpus.
698 *
699 * Call pmap_inval_pte() to invalidate the virtual pte and clean out the
700 * real pmap and handle related races before storing the new vpte.
701 */
702void
703pmap_kenter(vm_offset_t va, vm_paddr_t pa)
704{
705 pt_entry_t *pte;
706 pt_entry_t npte;
707
708 KKASSERT(va >= KvaStart && va < KvaEnd);
709 npte = pa | VPTE_R | VPTE_W | VPTE_V;
710 pte = vtopte(va);
711 if (*pte & VPTE_V)
712 pmap_inval_pte(pte, &kernel_pmap, va);
713 *pte = npte;
714}
715
716/*
717 * Enter an unmanaged KVA mapping for the private use of the current
718 * cpu only. pmap_kenter_sync() may be called to make the mapping usable
719 * by other cpus.
720 *
721 * It is illegal for the mapping to be accessed by other cpus unleess
722 * pmap_kenter_sync*() is called.
723 */
724void
725pmap_kenter_quick(vm_offset_t va, vm_paddr_t pa)
726{
727 pt_entry_t *pte;
728 pt_entry_t npte;
729
730 KKASSERT(va >= KvaStart && va < KvaEnd);
731
732 npte = (vpte_t)pa | VPTE_R | VPTE_W | VPTE_V;
733 pte = vtopte(va);
734 if (*pte & VPTE_V)
735 pmap_inval_pte_quick(pte, &kernel_pmap, va);
736 *pte = npte;
737 //cpu_invlpg((void *)va);
738}
739
740/*
741 * Synchronize a kvm mapping originally made for the private use on
742 * some other cpu so it can be used on all cpus.
743 *
744 * XXX add MADV_RESYNC to improve performance.
745 */
746void
747pmap_kenter_sync(vm_offset_t va)
748{
749 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
750}
751
752/*
753 * Synchronize a kvm mapping originally made for the private use on
754 * some other cpu so it can be used on our cpu. Turns out to be the
755 * same madvise() call, because we have to sync the real pmaps anyway.
756 *
757 * XXX add MADV_RESYNC to improve performance.
758 */
759void
760pmap_kenter_sync_quick(vm_offset_t va)
761{
762 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
763}
764
765/*
766 * Remove an unmanaged mapping created with pmap_kenter*().
767 */
768void
769pmap_kremove(vm_offset_t va)
770{
771 pt_entry_t *pte;
772
773 KKASSERT(va >= KvaStart && va < KvaEnd);
774
775 pte = vtopte(va);
776 if (*pte & VPTE_V)
777 pmap_inval_pte(pte, &kernel_pmap, va);
778 *pte = 0;
779}
780
781/*
782 * Remove an unmanaged mapping created with pmap_kenter*() but synchronize
783 * only with this cpu.
784 *
785 * Unfortunately because we optimize new entries by testing VPTE_V later
786 * on, we actually still have to synchronize with all the cpus. XXX maybe
787 * store a junk value and test against 0 in the other places instead?
788 */
789void
790pmap_kremove_quick(vm_offset_t va)
791{
792 pt_entry_t *pte;
793
794 KKASSERT(va >= KvaStart && va < KvaEnd);
795
796 pte = vtopte(va);
797 if (*pte & VPTE_V)
798 pmap_inval_pte(pte, &kernel_pmap, va); /* NOT _quick */
799 *pte = 0;
800}
801
802/*
803 * Used to map a range of physical addresses into kernel
804 * virtual address space.
805 *
806 * For now, VM is already on, we only need to map the
807 * specified memory.
808 */
809vm_offset_t
0e6594a8 810pmap_map(vm_offset_t *virtp, vm_paddr_t start, vm_paddr_t end, int prot)
da673940
JG
811{
812 return PHYS_TO_DMAP(start);
813}
814
815
816/*
817 * Map a set of unmanaged VM pages into KVM.
818 */
819void
820pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
821{
822 vm_offset_t end_va;
823
824 end_va = va + count * PAGE_SIZE;
825 KKASSERT(va >= KvaStart && end_va < KvaEnd);
826
827 while (va < end_va) {
828 pt_entry_t *pte;
829
830 pte = vtopte(va);
831 if (*pte & VPTE_V)
832 pmap_inval_pte(pte, &kernel_pmap, va);
833 *pte = VM_PAGE_TO_PHYS(*m) | VPTE_R | VPTE_W | VPTE_V;
834 va += PAGE_SIZE;
835 m++;
836 }
837}
838
839/*
da673940
JG
840 * Undo the effects of pmap_qenter*().
841 */
842void
843pmap_qremove(vm_offset_t va, int count)
844{
845 vm_offset_t end_va;
846
847 end_va = va + count * PAGE_SIZE;
848 KKASSERT(va >= KvaStart && end_va < KvaEnd);
849
850 while (va < end_va) {
851 pt_entry_t *pte;
852
853 pte = vtopte(va);
854 if (*pte & VPTE_V)
855 pmap_inval_pte(pte, &kernel_pmap, va);
856 *pte = 0;
857 va += PAGE_SIZE;
858 }
859}
860
861/*
862 * This routine works like vm_page_lookup() but also blocks as long as the
863 * page is busy. This routine does not busy the page it returns.
864 *
865 * Unless the caller is managing objects whos pages are in a known state,
866 * the call should be made with a critical section held so the page's object
867 * association remains valid on return.
868 */
869static vm_page_t
870pmap_page_lookup(vm_object_t object, vm_pindex_t pindex)
871{
872 vm_page_t m;
873
b12defdc
MD
874 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
875 m = vm_page_lookup_busy_wait(object, pindex, FALSE, "pplookp");
da673940
JG
876
877 return(m);
878}
879
880/*
881 * Create a new thread and optionally associate it with a (new) process.
882 * NOTE! the new thread's cpu may not equal the current cpu.
883 */
884void
885pmap_init_thread(thread_t td)
886{
887 /* enforce pcb placement */
888 td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_size) - 1;
889 td->td_savefpu = &td->td_pcb->pcb_save;
a76ca9b9 890 td->td_sp = (char *)td->td_pcb - 16; /* JG is -16 needed on x86_64? */
da673940
JG
891}
892
893/*
894 * This routine directly affects the fork perf for a process.
895 */
896void
897pmap_init_proc(struct proc *p)
898{
899}
900
da673940
JG
901/***************************************************
902 * Page table page management routines.....
903 ***************************************************/
904
2c2e847c
SW
905static __inline int pmap_unwire_pte_hold(pmap_t pmap, vm_offset_t va,
906 vm_page_t m);
907
da673940
JG
908/*
909 * This routine unholds page table pages, and if the hold count
910 * drops to zero, then it decrements the wire count.
911 *
912 * We must recheck that this is the last hold reference after busy-sleeping
913 * on the page.
914 */
915static int
916_pmap_unwire_pte_hold(pmap_t pmap, vm_offset_t va, vm_page_t m)
917{
b12defdc 918 vm_page_busy_wait(m, FALSE, "pmuwpt");
da673940
JG
919 KASSERT(m->queue == PQ_NONE,
920 ("_pmap_unwire_pte_hold: %p->queue != PQ_NONE", m));
921
922 if (m->hold_count == 1) {
923 /*
924 * Unmap the page table page.
925 */
926 //abort(); /* JG */
da673940
JG
927 /* pmap_inval_add(info, pmap, -1); */
928
929 if (m->pindex >= (NUPDE + NUPDPE)) {
930 /* PDP page */
931 pml4_entry_t *pml4;
932 pml4 = pmap_pml4e(pmap, va);
933 *pml4 = 0;
934 } else if (m->pindex >= NUPDE) {
935 /* PD page */
936 pdp_entry_t *pdp;
937 pdp = pmap_pdpe(pmap, va);
938 *pdp = 0;
939 } else {
940 /* PT page */
941 pd_entry_t *pd;
942 pd = pmap_pde(pmap, va);
943 *pd = 0;
944 }
945
946 KKASSERT(pmap->pm_stats.resident_count > 0);
947 --pmap->pm_stats.resident_count;
948
949 if (pmap->pm_ptphint == m)
950 pmap->pm_ptphint = NULL;
951
952 if (m->pindex < NUPDE) {
953 /* We just released a PT, unhold the matching PD */
954 vm_page_t pdpg;
955
956 pdpg = PHYS_TO_VM_PAGE(*pmap_pdpe(pmap, va) & VPTE_FRAME);
957 pmap_unwire_pte_hold(pmap, va, pdpg);
958 }
959 if (m->pindex >= NUPDE && m->pindex < (NUPDE + NUPDPE)) {
960 /* We just released a PD, unhold the matching PDP */
961 vm_page_t pdppg;
962
963 pdppg = PHYS_TO_VM_PAGE(*pmap_pml4e(pmap, va) & VPTE_FRAME);
964 pmap_unwire_pte_hold(pmap, va, pdppg);
965 }
966
967 /*
968 * This was our last hold, the page had better be unwired
969 * after we decrement wire_count.
970 *
971 * FUTURE NOTE: shared page directory page could result in
972 * multiple wire counts.
973 */
974 vm_page_unhold(m);
975 --m->wire_count;
976 KKASSERT(m->wire_count == 0);
b12defdc 977 atomic_add_int(&vmstats.v_wire_count, -1);
da673940
JG
978 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
979 vm_page_flash(m);
980 vm_page_free_zero(m);
981 return 1;
982 } else {
983 KKASSERT(m->hold_count > 1);
984 vm_page_unhold(m);
b12defdc 985 vm_page_wakeup(m);
da673940
JG
986 return 0;
987 }
988}
989
990static __inline int
991pmap_unwire_pte_hold(pmap_t pmap, vm_offset_t va, vm_page_t m)
992{
993 KKASSERT(m->hold_count > 0);
994 if (m->hold_count > 1) {
995 vm_page_unhold(m);
996 return 0;
997 } else {
998 return _pmap_unwire_pte_hold(pmap, va, m);
999 }
1000}
1001
1002/*
1003 * After removing a page table entry, this routine is used to
1004 * conditionally free the page, and manage the hold/wire counts.
1005 */
1006static int
1007pmap_unuse_pt(pmap_t pmap, vm_offset_t va, vm_page_t mpte)
1008{
1009 /* JG Use FreeBSD/amd64 or FreeBSD/i386 ptepde approaches? */
1010 vm_pindex_t ptepindex;
1011
b12defdc
MD
1012 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1013
da673940
JG
1014 if (mpte == NULL) {
1015 /*
1016 * page table pages in the kernel_pmap are not managed.
1017 */
1018 if (pmap == &kernel_pmap)
1019 return(0);
1020 ptepindex = pmap_pde_pindex(va);
1021 if (pmap->pm_ptphint &&
1022 (pmap->pm_ptphint->pindex == ptepindex)) {
1023 mpte = pmap->pm_ptphint;
1024 } else {
1025 mpte = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
1026 pmap->pm_ptphint = mpte;
b12defdc 1027 vm_page_wakeup(mpte);
da673940
JG
1028 }
1029 }
1030
1031 return pmap_unwire_pte_hold(pmap, va, mpte);
1032}
1033
1034/*
1035 * Initialize pmap0/vmspace0 . Since process 0 never enters user mode we
1036 * just dummy it up so it works well enough for fork().
1037 *
1038 * In DragonFly, process pmaps may only be used to manipulate user address
1039 * space, never kernel address space.
1040 */
1041void
1042pmap_pinit0(struct pmap *pmap)
1043{
1044 pmap_pinit(pmap);
1045}
1046
1047/*
1048 * Initialize a preallocated and zeroed pmap structure,
1049 * such as one in a vmspace structure.
1050 */
1051void
1052pmap_pinit(struct pmap *pmap)
1053{
1054 vm_page_t ptdpg;
1055
1056 /*
1057 * No need to allocate page table space yet but we do need a valid
1058 * page directory table.
1059 */
1060 if (pmap->pm_pml4 == NULL) {
1061 pmap->pm_pml4 =
1062 (pml4_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
1063 }
1064
1065 /*
1066 * Allocate an object for the ptes
1067 */
1068 if (pmap->pm_pteobj == NULL)
1069 pmap->pm_pteobj = vm_object_allocate(OBJT_DEFAULT, NUPDE + NUPDPE + PML4PML4I + 1);
1070
1071 /*
1072 * Allocate the page directory page, unless we already have
1073 * one cached. If we used the cached page the wire_count will
1074 * already be set appropriately.
1075 */
1076 if ((ptdpg = pmap->pm_pdirm) == NULL) {
d2d8515b
MD
1077 ptdpg = vm_page_grab(pmap->pm_pteobj,
1078 NUPDE + NUPDPE + PML4PML4I,
1079 VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
1080 VM_ALLOC_ZERO);
da673940 1081 pmap->pm_pdirm = ptdpg;
b12defdc 1082 vm_page_flag_clear(ptdpg, PG_MAPPED);
54341a3b 1083 vm_page_wire(ptdpg);
b12defdc 1084 vm_page_wakeup(ptdpg);
da673940
JG
1085 pmap_kenter((vm_offset_t)pmap->pm_pml4, VM_PAGE_TO_PHYS(ptdpg));
1086 }
da673940
JG
1087 pmap->pm_count = 1;
1088 pmap->pm_active = 0;
1089 pmap->pm_ptphint = NULL;
1090 TAILQ_INIT(&pmap->pm_pvlist);
b12defdc
MD
1091 TAILQ_INIT(&pmap->pm_pvlist_free);
1092 spin_init(&pmap->pm_spin);
1093 lwkt_token_init(&pmap->pm_token, "pmap_tok");
da673940
JG
1094 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1095 pmap->pm_stats.resident_count = 1;
1096}
1097
1098/*
1099 * Clean up a pmap structure so it can be physically freed. This routine
1100 * is called by the vmspace dtor function. A great deal of pmap data is
1101 * left passively mapped to improve vmspace management so we have a bit
1102 * of cleanup work to do here.
f7230403
MD
1103 *
1104 * No requirements.
da673940
JG
1105 */
1106void
1107pmap_puninit(pmap_t pmap)
1108{
1109 vm_page_t p;
1110
1111 KKASSERT(pmap->pm_active == 0);
1112 if ((p = pmap->pm_pdirm) != NULL) {
1113 KKASSERT(pmap->pm_pml4 != NULL);
1114 pmap_kremove((vm_offset_t)pmap->pm_pml4);
b12defdc 1115 vm_page_busy_wait(p, FALSE, "pgpun");
da673940 1116 p->wire_count--;
b12defdc 1117 atomic_add_int(&vmstats.v_wire_count, -1);
da673940
JG
1118 vm_page_free_zero(p);
1119 pmap->pm_pdirm = NULL;
1120 }
1121 if (pmap->pm_pml4) {
1122 kmem_free(&kernel_map, (vm_offset_t)pmap->pm_pml4, PAGE_SIZE);
1123 pmap->pm_pml4 = NULL;
1124 }
1125 if (pmap->pm_pteobj) {
1126 vm_object_deallocate(pmap->pm_pteobj);
1127 pmap->pm_pteobj = NULL;
1128 }
1129}
1130
1131/*
1132 * Wire in kernel global address entries. To avoid a race condition
1133 * between pmap initialization and pmap_growkernel, this procedure
1134 * adds the pmap to the master list (which growkernel scans to update),
1135 * then copies the template.
1136 *
1137 * In a virtual kernel there are no kernel global address entries.
f7230403
MD
1138 *
1139 * No requirements.
da673940
JG
1140 */
1141void
1142pmap_pinit2(struct pmap *pmap)
1143{
b12defdc 1144 spin_lock(&pmap_spin);
da673940 1145 TAILQ_INSERT_TAIL(&pmap_list, pmap, pm_pmnode);
b12defdc 1146 spin_unlock(&pmap_spin);
da673940
JG
1147}
1148
1149/*
1150 * Attempt to release and free a vm_page in a pmap. Returns 1 on success,
1151 * 0 on failure (if the procedure had to sleep).
1152 *
1153 * When asked to remove the page directory page itself, we actually just
1154 * leave it cached so we do not have to incur the SMP inval overhead of
1155 * removing the kernel mapping. pmap_puninit() will take care of it.
1156 */
1157static int
1158pmap_release_free_page(struct pmap *pmap, vm_page_t p)
1159{
da673940
JG
1160 /*
1161 * This code optimizes the case of freeing non-busy
1162 * page-table pages. Those pages are zero now, and
1163 * might as well be placed directly into the zero queue.
1164 */
b12defdc
MD
1165 if (vm_page_busy_try(p, FALSE)) {
1166 vm_page_sleep_busy(p, FALSE, "pmaprl");
da673940 1167 return 0;
b12defdc 1168 }
da673940
JG
1169
1170 /*
1171 * Remove the page table page from the processes address space.
1172 */
1173 if (p->pindex == NUPDE + NUPDPE + PML4PML4I) {
1174 /*
1175 * We are the pml4 table itself.
1176 */
1177 /* XXX anything to do here? */
1178 } else if (p->pindex >= (NUPDE + NUPDPE)) {
1179 /*
1180 * We are a PDP page.
1181 * We look for the PML4 entry that points to us.
1182 */
1183 vm_page_t m4 = vm_page_lookup(pmap->pm_pteobj, NUPDE + NUPDPE + PML4PML4I);
1184 KKASSERT(m4 != NULL);
2c2e847c 1185 pml4_entry_t *pml4 = (pml4_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m4));
da673940
JG
1186 int idx = (p->pindex - (NUPDE + NUPDPE)) % NPML4EPG;
1187 KKASSERT(pml4[idx] != 0);
1188 pml4[idx] = 0;
1189 m4->hold_count--;
1190 /* JG What about wire_count? */
1191 } else if (p->pindex >= NUPDE) {
1192 /*
1193 * We are a PD page.
1194 * We look for the PDP entry that points to us.
1195 */
1196 vm_page_t m3 = vm_page_lookup(pmap->pm_pteobj, NUPDE + NUPDPE + (p->pindex - NUPDE) / NPDPEPG);
1197 KKASSERT(m3 != NULL);
2c2e847c 1198 pdp_entry_t *pdp = (pdp_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m3));
da673940
JG
1199 int idx = (p->pindex - NUPDE) % NPDPEPG;
1200 KKASSERT(pdp[idx] != 0);
1201 pdp[idx] = 0;
1202 m3->hold_count--;
1203 /* JG What about wire_count? */
1204 } else {
1205 /* We are a PT page.
1206 * We look for the PD entry that points to us.
1207 */
1208 vm_page_t m2 = vm_page_lookup(pmap->pm_pteobj, NUPDE + p->pindex / NPDEPG);
1209 KKASSERT(m2 != NULL);
2c2e847c 1210 pd_entry_t *pd = (pd_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m2));
da673940
JG
1211 int idx = p->pindex % NPDEPG;
1212 pd[idx] = 0;
1213 m2->hold_count--;
1214 /* JG What about wire_count? */
1215 }
1216 KKASSERT(pmap->pm_stats.resident_count > 0);
1217 --pmap->pm_stats.resident_count;
1218
1219 if (p->hold_count) {
1220 panic("pmap_release: freeing held page table page");
1221 }
1222 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1223 pmap->pm_ptphint = NULL;
1224
1225 /*
1226 * We leave the top-level page table page cached, wired, and mapped in
1227 * the pmap until the dtor function (pmap_puninit()) gets called.
1228 * However, still clean it up so we can set PG_ZERO.
1229 */
1230 if (p->pindex == NUPDE + NUPDPE + PML4PML4I) {
1231 bzero(pmap->pm_pml4, PAGE_SIZE);
1232 vm_page_flag_set(p, PG_ZERO);
1233 vm_page_wakeup(p);
1234 } else {
1235 abort();
1236 p->wire_count--;
b12defdc 1237 atomic_add_int(&vmstats.v_wire_count, -1);
da673940
JG
1238 /* JG eventually revert to using vm_page_free_zero() */
1239 vm_page_free(p);
1240 }
1241 return 1;
1242}
1243
1244/*
1245 * this routine is called if the page table page is not
1246 * mapped correctly.
1247 */
1248static vm_page_t
1249_pmap_allocpte(pmap_t pmap, vm_pindex_t ptepindex)
1250{
1251 vm_page_t m, pdppg, pdpg;
1252
1253 /*
b12defdc
MD
1254 * Find or fabricate a new pagetable page. Handle allocation
1255 * races by checking m->valid.
da673940
JG
1256 */
1257 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
d2d8515b 1258 VM_ALLOC_NORMAL | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
da673940
JG
1259
1260 KASSERT(m->queue == PQ_NONE,
1261 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1262
1263 /*
1264 * Increment the hold count for the page we will be returning to
1265 * the caller.
1266 */
1267 m->hold_count++;
54341a3b 1268 vm_page_wire(m);
da673940
JG
1269
1270 /*
1271 * Map the pagetable page into the process address space, if
1272 * it isn't already there.
1273 */
da673940
JG
1274 ++pmap->pm_stats.resident_count;
1275
1276 if (ptepindex >= (NUPDE + NUPDPE)) {
1277 pml4_entry_t *pml4;
1278 vm_pindex_t pml4index;
1279
1280 /* Wire up a new PDP page */
1281 pml4index = ptepindex - (NUPDE + NUPDPE);
1282 pml4 = &pmap->pm_pml4[pml4index];
1283 *pml4 = VM_PAGE_TO_PHYS(m) | VPTE_R | VPTE_W | VPTE_V |
1284 VPTE_A | VPTE_M;
1285 } else if (ptepindex >= NUPDE) {
1286 vm_pindex_t pml4index;
1287 vm_pindex_t pdpindex;
1288 pml4_entry_t *pml4;
1289 pdp_entry_t *pdp;
1290
1291 /* Wire up a new PD page */
1292 pdpindex = ptepindex - NUPDE;
1293 pml4index = pdpindex >> NPML4EPGSHIFT;
1294
1295 pml4 = &pmap->pm_pml4[pml4index];
1296 if ((*pml4 & VPTE_V) == 0) {
1297 /* Have to allocate a new PDP page, recurse */
1298 if (_pmap_allocpte(pmap, NUPDE + NUPDPE + pml4index)
1299 == NULL) {
1300 --m->wire_count;
1301 vm_page_free(m);
1302 return (NULL);
1303 }
1304 } else {
1305 /* Add reference to the PDP page */
1306 pdppg = PHYS_TO_VM_PAGE(*pml4 & VPTE_FRAME);
1307 pdppg->hold_count++;
1308 }
1309 pdp = (pdp_entry_t *)PHYS_TO_DMAP(*pml4 & VPTE_FRAME);
1310
1311 /* Now find the pdp page */
1312 pdp = &pdp[pdpindex & ((1ul << NPDPEPGSHIFT) - 1)];
1313 KKASSERT(*pdp == 0); /* JG DEBUG64 */
1314 *pdp = VM_PAGE_TO_PHYS(m) | VPTE_R | VPTE_W | VPTE_V |
1315 VPTE_A | VPTE_M;
1316 } else {
1317 vm_pindex_t pml4index;
1318 vm_pindex_t pdpindex;
1319 pml4_entry_t *pml4;
1320 pdp_entry_t *pdp;
1321 pd_entry_t *pd;
1322
1323 /* Wire up a new PT page */
1324 pdpindex = ptepindex >> NPDPEPGSHIFT;
1325 pml4index = pdpindex >> NPML4EPGSHIFT;
1326
1327 /* First, find the pdp and check that its valid. */
1328 pml4 = &pmap->pm_pml4[pml4index];
1329 if ((*pml4 & VPTE_V) == 0) {
1330 /* We miss a PDP page. We ultimately need a PD page.
1331 * Recursively allocating a PD page will allocate
1332 * the missing PDP page and will also allocate
1333 * the PD page we need.
1334 */
1335 /* Have to allocate a new PD page, recurse */
1336 if (_pmap_allocpte(pmap, NUPDE + pdpindex)
1337 == NULL) {
1338 --m->wire_count;
1339 vm_page_free(m);
1340 return (NULL);
1341 }
1342 pdp = (pdp_entry_t *)PHYS_TO_DMAP(*pml4 & VPTE_FRAME);
1343 pdp = &pdp[pdpindex & ((1ul << NPDPEPGSHIFT) - 1)];
1344 } else {
1345 pdp = (pdp_entry_t *)PHYS_TO_DMAP(*pml4 & VPTE_FRAME);
1346 pdp = &pdp[pdpindex & ((1ul << NPDPEPGSHIFT) - 1)];
1347 if ((*pdp & VPTE_V) == 0) {
1348 /* Have to allocate a new PD page, recurse */
1349 if (_pmap_allocpte(pmap, NUPDE + pdpindex)
1350 == NULL) {
1351 --m->wire_count;
1352 vm_page_free(m);
1353 return (NULL);
1354 }
1355 } else {
1356 /* Add reference to the PD page */
1357 pdpg = PHYS_TO_VM_PAGE(*pdp & VPTE_FRAME);
1358 pdpg->hold_count++;
1359 }
1360 }
1361 pd = (pd_entry_t *)PHYS_TO_DMAP(*pdp & VPTE_FRAME);
1362
1363 /* Now we know where the page directory page is */
1364 pd = &pd[ptepindex & ((1ul << NPDEPGSHIFT) - 1)];
1365 KKASSERT(*pd == 0); /* JG DEBUG64 */
1366 *pd = VM_PAGE_TO_PHYS(m) | VPTE_R | VPTE_W | VPTE_V |
1367 VPTE_A | VPTE_M;
1368 }
1369
1370 /*
1371 * Set the page table hint
1372 */
1373 pmap->pm_ptphint = m;
da673940
JG
1374 vm_page_flag_set(m, PG_MAPPED);
1375 vm_page_wakeup(m);
1376
1377 return m;
1378}
1379
1380/*
1381 * Determine the page table page required to access the VA in the pmap
1382 * and allocate it if necessary. Return a held vm_page_t for the page.
1383 *
1384 * Only used with user pmaps.
1385 */
1386static vm_page_t
1387pmap_allocpte(pmap_t pmap, vm_offset_t va)
1388{
1389 vm_pindex_t ptepindex;
1390 pd_entry_t *pd;
1391 vm_page_t m;
1392
b12defdc
MD
1393 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1394
da673940
JG
1395 /*
1396 * Calculate pagetable page index
1397 */
1398 ptepindex = pmap_pde_pindex(va);
1399
1400 /*
1401 * Get the page directory entry
1402 */
1403 pd = pmap_pde(pmap, va);
1404
1405 /*
1406 * This supports switching from a 2MB page to a
1407 * normal 4K page.
1408 */
1409 if (pd != NULL && (*pd & (VPTE_PS | VPTE_V)) == (VPTE_PS | VPTE_V)) {
1410 panic("no promotion/demotion yet");
1411 *pd = 0;
1412 pd = NULL;
1413 /*cpu_invltlb();*/
1414 /*smp_invltlb();*/
1415 }
1416
1417 /*
1418 * If the page table page is mapped, we just increment the
1419 * hold count, and activate it.
1420 */
1421 if (pd != NULL && (*pd & VPTE_V) != 0) {
1422 /* YYY hint is used here on i386 */
b12defdc 1423 m = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
da673940 1424 pmap->pm_ptphint = m;
b12defdc
MD
1425 vm_page_hold(m);
1426 vm_page_wakeup(m);
da673940
JG
1427 return m;
1428 }
1429 /*
1430 * Here if the pte page isn't mapped, or if it has been deallocated.
1431 */
1432 return _pmap_allocpte(pmap, ptepindex);
1433}
1434
1435
1436/***************************************************
1437 * Pmap allocation/deallocation routines.
1438 ***************************************************/
1439
1440/*
1441 * Release any resources held by the given physical map.
1442 * Called when a pmap initialized by pmap_pinit is being released.
1443 * Should only be called if the map contains no valid mappings.
f7230403 1444 *
b12defdc 1445 * Caller must hold pmap->pm_token
da673940
JG
1446 */
1447static int pmap_release_callback(struct vm_page *p, void *data);
1448
1449void
1450pmap_release(struct pmap *pmap)
1451{
1452 vm_object_t object = pmap->pm_pteobj;
1453 struct rb_vm_page_scan_info info;
1454
1455 KKASSERT(pmap != &kernel_pmap);
1456
1457#if defined(DIAGNOSTIC)
1458 if (object->ref_count != 1)
1459 panic("pmap_release: pteobj reference count != 1");
1460#endif
1461
1462 info.pmap = pmap;
1463 info.object = object;
b12defdc
MD
1464
1465 spin_lock(&pmap_spin);
da673940 1466 TAILQ_REMOVE(&pmap_list, pmap, pm_pmnode);
b12defdc 1467 spin_unlock(&pmap_spin);
da673940 1468
b12defdc 1469 vm_object_hold(object);
da673940 1470 do {
da673940
JG
1471 info.error = 0;
1472 info.mpte = NULL;
1473 info.limit = object->generation;
1474
1475 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
1476 pmap_release_callback, &info);
1477 if (info.error == 0 && info.mpte) {
1478 if (!pmap_release_free_page(pmap, info.mpte))
1479 info.error = 1;
1480 }
da673940 1481 } while (info.error);
b12defdc 1482 vm_object_drop(object);
da673940
JG
1483}
1484
1485static int
1486pmap_release_callback(struct vm_page *p, void *data)
1487{
1488 struct rb_vm_page_scan_info *info = data;
1489
1490 if (p->pindex == NUPDE + NUPDPE + PML4PML4I) {
1491 info->mpte = p;
1492 return(0);
1493 }
1494 if (!pmap_release_free_page(info->pmap, p)) {
1495 info->error = 1;
1496 return(-1);
1497 }
1498 if (info->object->generation != info->limit) {
1499 info->error = 1;
1500 return(-1);
1501 }
1502 return(0);
1503}
1504
1505/*
1506 * Grow the number of kernel page table entries, if needed.
f7230403
MD
1507 *
1508 * No requirements.
da673940 1509 */
da673940 1510void
a8cf2878 1511pmap_growkernel(vm_offset_t kstart, vm_offset_t kend)
da673940 1512{
a8cf2878 1513 vm_offset_t addr;
da673940
JG
1514 vm_paddr_t paddr;
1515 vm_offset_t ptppaddr;
1516 vm_page_t nkpg;
1517 pd_entry_t *pde, newpdir;
1518 pdp_entry_t newpdp;
1519
a8cf2878
MD
1520 addr = kend;
1521
b12defdc 1522 vm_object_hold(kptobj);
da673940
JG
1523 if (kernel_vm_end == 0) {
1524 kernel_vm_end = KvaStart;
1525 nkpt = 0;
1526 while ((*pmap_pde(&kernel_pmap, kernel_vm_end) & VPTE_V) != 0) {
1527 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1528 nkpt++;
1529 if (kernel_vm_end - 1 >= kernel_map.max_offset) {
1530 kernel_vm_end = kernel_map.max_offset;
1531 break;
1532 }
1533 }
1534 }
1535 addr = roundup2(addr, PAGE_SIZE * NPTEPG);
1536 if (addr - 1 >= kernel_map.max_offset)
1537 addr = kernel_map.max_offset;
1538 while (kernel_vm_end < addr) {
1539 pde = pmap_pde(&kernel_pmap, kernel_vm_end);
1540 if (pde == NULL) {
1541 /* We need a new PDP entry */
1542 nkpg = vm_page_alloc(kptobj, nkpt,
1543 VM_ALLOC_NORMAL | VM_ALLOC_SYSTEM
1544 | VM_ALLOC_INTERRUPT);
a8cf2878
MD
1545 if (nkpg == NULL) {
1546 panic("pmap_growkernel: no memory to "
1547 "grow kernel");
1548 }
da673940
JG
1549 paddr = VM_PAGE_TO_PHYS(nkpg);
1550 if ((nkpg->flags & PG_ZERO) == 0)
1551 pmap_zero_page(paddr);
1552 vm_page_flag_clear(nkpg, PG_ZERO);
a8cf2878
MD
1553 newpdp = (pdp_entry_t)(paddr | VPTE_V | VPTE_R |
1554 VPTE_W | VPTE_A | VPTE_M);
da673940
JG
1555 *pmap_pdpe(&kernel_pmap, kernel_vm_end) = newpdp;
1556 nkpt++;
1557 continue; /* try again */
1558 }
1559 if ((*pde & VPTE_V) != 0) {
a8cf2878
MD
1560 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1561 ~(PAGE_SIZE * NPTEPG - 1);
da673940
JG
1562 if (kernel_vm_end - 1 >= kernel_map.max_offset) {
1563 kernel_vm_end = kernel_map.max_offset;
1564 break;
1565 }
1566 continue;
1567 }
1568
1569 /*
1570 * This index is bogus, but out of the way
1571 */
1572 nkpg = vm_page_alloc(kptobj, nkpt,
a8cf2878
MD
1573 VM_ALLOC_NORMAL |
1574 VM_ALLOC_SYSTEM |
1575 VM_ALLOC_INTERRUPT);
da673940
JG
1576 if (nkpg == NULL)
1577 panic("pmap_growkernel: no memory to grow kernel");
1578
1579 vm_page_wire(nkpg);
1580 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1581 pmap_zero_page(ptppaddr);
1582 vm_page_flag_clear(nkpg, PG_ZERO);
a8cf2878
MD
1583 newpdir = (pd_entry_t)(ptppaddr | VPTE_V | VPTE_R |
1584 VPTE_W | VPTE_A | VPTE_M);
da673940
JG
1585 *pmap_pde(&kernel_pmap, kernel_vm_end) = newpdir;
1586 nkpt++;
1587
a8cf2878
MD
1588 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1589 ~(PAGE_SIZE * NPTEPG - 1);
da673940
JG
1590 if (kernel_vm_end - 1 >= kernel_map.max_offset) {
1591 kernel_vm_end = kernel_map.max_offset;
1592 break;
1593 }
1594 }
b12defdc 1595 vm_object_drop(kptobj);
da673940
JG
1596}
1597
1598/*
f7230403
MD
1599 * Retire the given physical map from service. Should only be called
1600 * if the map contains no valid mappings.
1601 *
1602 * No requirements.
da673940
JG
1603 */
1604void
1605pmap_destroy(pmap_t pmap)
1606{
da673940
JG
1607 if (pmap == NULL)
1608 return;
1609
f7230403
MD
1610 lwkt_gettoken(&vm_token);
1611 if (--pmap->pm_count == 0) {
da673940
JG
1612 pmap_release(pmap);
1613 panic("destroying a pmap is not yet implemented");
1614 }
f7230403 1615 lwkt_reltoken(&vm_token);
da673940
JG
1616}
1617
1618/*
f7230403
MD
1619 * Add a reference to the specified pmap.
1620 *
1621 * No requirements.
da673940
JG
1622 */
1623void
1624pmap_reference(pmap_t pmap)
1625{
f7230403
MD
1626 if (pmap) {
1627 lwkt_gettoken(&vm_token);
1628 ++pmap->pm_count;
1629 lwkt_reltoken(&vm_token);
da673940
JG
1630 }
1631}
1632
1633/************************************************************************
1634 * VMSPACE MANAGEMENT *
1635 ************************************************************************
1636 *
1637 * The VMSPACE management we do in our virtual kernel must be reflected
1638 * in the real kernel. This is accomplished by making vmspace system
1639 * calls to the real kernel.
1640 */
1641void
1642cpu_vmspace_alloc(struct vmspace *vm)
1643{
1644 int r;
1645 void *rp;
1646 vpte_t vpte;
1647
1648#define USER_SIZE (VM_MAX_USER_ADDRESS - VM_MIN_USER_ADDRESS)
1649
1650 if (vmspace_create(&vm->vm_pmap, 0, NULL) < 0)
1651 panic("vmspace_create() failed");
1652
1653 rp = vmspace_mmap(&vm->vm_pmap, VM_MIN_USER_ADDRESS, USER_SIZE,
1654 PROT_READ|PROT_WRITE,
1655 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED,
1656 MemImageFd, 0);
1657 if (rp == MAP_FAILED)
1658 panic("vmspace_mmap: failed");
1659 vmspace_mcontrol(&vm->vm_pmap, VM_MIN_USER_ADDRESS, USER_SIZE,
1660 MADV_NOSYNC, 0);
1661 vpte = VM_PAGE_TO_PHYS(vmspace_pmap(vm)->pm_pdirm) | VPTE_R | VPTE_W | VPTE_V;
1662 r = vmspace_mcontrol(&vm->vm_pmap, VM_MIN_USER_ADDRESS, USER_SIZE,
1663 MADV_SETMAP, vpte);
1664 if (r < 0)
1665 panic("vmspace_mcontrol: failed");
1666}
1667
1668void
1669cpu_vmspace_free(struct vmspace *vm)
1670{
1671 if (vmspace_destroy(&vm->vm_pmap) < 0)
1672 panic("vmspace_destroy() failed");
1673}
1674
1675/***************************************************
1676* page management routines.
1677 ***************************************************/
1678
1679/*
1680 * free the pv_entry back to the free list. This function may be
1681 * called from an interrupt.
1682 */
1683static __inline void
1684free_pv_entry(pv_entry_t pv)
1685{
1686 pv_entry_count--;
1687 KKASSERT(pv_entry_count >= 0);
1688 zfree(pvzone, pv);
1689}
1690
1691/*
1692 * get a new pv_entry, allocating a block from the system
1693 * when needed. This function may be called from an interrupt.
1694 */
1695static pv_entry_t
1696get_pv_entry(void)
1697{
1698 pv_entry_count++;
1699 if (pv_entry_high_water &&
1700 (pv_entry_count > pv_entry_high_water) &&
1701 (pmap_pagedaemon_waken == 0)) {
1702 pmap_pagedaemon_waken = 1;
1703 wakeup(&vm_pages_needed);
1704 }
1705 return zalloc(pvzone);
1706}
1707
1708/*
1709 * This routine is very drastic, but can save the system
1710 * in a pinch.
f7230403
MD
1711 *
1712 * No requirements.
da673940
JG
1713 */
1714void
1715pmap_collect(void)
1716{
1717 int i;
1718 vm_page_t m;
1719 static int warningdone=0;
1720
1721 if (pmap_pagedaemon_waken == 0)
1722 return;
f7230403 1723 lwkt_gettoken(&vm_token);
da673940
JG
1724 pmap_pagedaemon_waken = 0;
1725
1726 if (warningdone < 5) {
1727 kprintf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1728 warningdone++;
1729 }
1730
b12defdc 1731 for (i = 0; i < vm_page_array_size; i++) {
da673940 1732 m = &vm_page_array[i];
b12defdc 1733 if (m->wire_count || m->hold_count)
da673940 1734 continue;
b12defdc
MD
1735 if (vm_page_busy_try(m, TRUE) == 0) {
1736 if (m->wire_count == 0 && m->hold_count == 0) {
1737 pmap_remove_all(m);
1738 }
1739 vm_page_wakeup(m);
1740 }
da673940 1741 }
f7230403 1742 lwkt_reltoken(&vm_token);
da673940
JG
1743}
1744
1745
1746/*
1747 * If it is the first entry on the list, it is actually
1748 * in the header and we must copy the following entry up
1749 * to the header. Otherwise we must search the list for
1750 * the entry. In either case we free the now unused entry.
b12defdc
MD
1751 *
1752 * caller must hold vm_token.
da673940
JG
1753 */
1754static int
1755pmap_remove_entry(struct pmap *pmap, vm_page_t m, vm_offset_t va)
1756{
1757 pv_entry_t pv;
1758 int rtval;
1759
da673940
JG
1760 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1761 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1762 if (pmap == pv->pv_pmap && va == pv->pv_va)
1763 break;
1764 }
1765 } else {
1766 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1767 if (va == pv->pv_va)
1768 break;
1769 }
1770 }
1771
1772 /*
1773 * Note that pv_ptem is NULL if the page table page itself is not
1774 * managed, even if the page being removed IS managed.
1775 */
1776 rtval = 0;
1777 /* JGXXX When can 'pv' be NULL? */
1778 if (pv) {
1779 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1780 m->md.pv_list_count--;
b12defdc 1781 atomic_add_int(&m->object->agg_pv_list_count, -1);
da673940
JG
1782 KKASSERT(m->md.pv_list_count >= 0);
1783 if (TAILQ_EMPTY(&m->md.pv_list))
1784 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1785 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1786 ++pmap->pm_generation;
b12defdc
MD
1787 KKASSERT(pmap->pm_pteobj != NULL);
1788 vm_object_hold(pmap->pm_pteobj);
da673940 1789 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
b12defdc 1790 vm_object_drop(pmap->pm_pteobj);
da673940
JG
1791 free_pv_entry(pv);
1792 }
da673940
JG
1793 return rtval;
1794}
1795
1796/*
1797 * Create a pv entry for page at pa for (pmap, va). If the page table page
1798 * holding the VA is managed, mpte will be non-NULL.
1799 */
1800static void
1801pmap_insert_entry(pmap_t pmap, vm_offset_t va, vm_page_t mpte, vm_page_t m)
1802{
1803 pv_entry_t pv;
1804
1805 crit_enter();
1806 pv = get_pv_entry();
1807 pv->pv_va = va;
1808 pv->pv_pmap = pmap;
1809 pv->pv_ptem = mpte;
1810
1811 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1812 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1813 m->md.pv_list_count++;
b12defdc 1814 atomic_add_int(&m->object->agg_pv_list_count, 1);
da673940
JG
1815
1816 crit_exit();
1817}
1818
1819/*
1820 * pmap_remove_pte: do the things to unmap a page in a process
1821 */
1822static int
1823pmap_remove_pte(struct pmap *pmap, pt_entry_t *ptq, vm_offset_t va)
1824{
1825 pt_entry_t oldpte;
1826 vm_page_t m;
1827
1828 oldpte = pmap_inval_loadandclear(ptq, pmap, va);
1829 if (oldpte & VPTE_WIRED)
1830 --pmap->pm_stats.wired_count;
1831 KKASSERT(pmap->pm_stats.wired_count >= 0);
1832
1833#if 0
1834 /*
1835 * Machines that don't support invlpg, also don't support
1836 * PG_G. XXX PG_G is disabled for SMP so don't worry about
1837 * the SMP case.
1838 */
1839 if (oldpte & PG_G)
1840 cpu_invlpg((void *)va);
1841#endif
1842 KKASSERT(pmap->pm_stats.resident_count > 0);
1843 --pmap->pm_stats.resident_count;
1844 if (oldpte & VPTE_MANAGED) {
1845 m = PHYS_TO_VM_PAGE(oldpte);
1846 if (oldpte & VPTE_M) {
1847#if defined(PMAP_DIAGNOSTIC)
1848 if (pmap_nw_modified((pt_entry_t) oldpte)) {
b12defdc
MD
1849 kprintf("pmap_remove: modified page not "
1850 "writable: va: 0x%lx, pte: 0x%lx\n",
1851 va, oldpte);
da673940
JG
1852 }
1853#endif
1854 if (pmap_track_modified(pmap, va))
1855 vm_page_dirty(m);
1856 }
1857 if (oldpte & VPTE_A)
1858 vm_page_flag_set(m, PG_REFERENCED);
1859 return pmap_remove_entry(pmap, m, va);
1860 } else {
1861 return pmap_unuse_pt(pmap, va, NULL);
1862 }
1863
1864 return 0;
1865}
1866
1867/*
1868 * pmap_remove_page:
1869 *
1870 * Remove a single page from a process address space.
1871 *
1872 * This function may not be called from an interrupt if the pmap is
1873 * not kernel_pmap.
1874 */
1875static void
1876pmap_remove_page(struct pmap *pmap, vm_offset_t va)
1877{
1878 pt_entry_t *pte;
1879
1880 pte = pmap_pte(pmap, va);
1881 if (pte == NULL)
1882 return;
1883 if ((*pte & VPTE_V) == 0)
1884 return;
1885 pmap_remove_pte(pmap, pte, va);
1886}
1887
1888/*
f7230403 1889 * Remove the given range of addresses from the specified map.
da673940 1890 *
f7230403
MD
1891 * It is assumed that the start and end are properly rounded to
1892 * the page size.
da673940 1893 *
f7230403
MD
1894 * This function may not be called from an interrupt if the pmap is
1895 * not kernel_pmap.
da673940 1896 *
f7230403 1897 * No requirements.
da673940
JG
1898 */
1899void
1900pmap_remove(struct pmap *pmap, vm_offset_t sva, vm_offset_t eva)
1901{
1902 vm_offset_t va_next;
1903 pml4_entry_t *pml4e;
1904 pdp_entry_t *pdpe;
1905 pd_entry_t ptpaddr, *pde;
1906 pt_entry_t *pte;
1907
1908 if (pmap == NULL)
1909 return;
1910
b12defdc 1911 vm_object_hold(pmap->pm_pteobj);
f7230403 1912 lwkt_gettoken(&vm_token);
da673940 1913 KKASSERT(pmap->pm_stats.resident_count >= 0);
f7230403
MD
1914 if (pmap->pm_stats.resident_count == 0) {
1915 lwkt_reltoken(&vm_token);
b12defdc 1916 vm_object_drop(pmap->pm_pteobj);
da673940 1917 return;
f7230403 1918 }
da673940
JG
1919
1920 /*
1921 * special handling of removing one page. a very
1922 * common operation and easy to short circuit some
1923 * code.
1924 */
1925 if (sva + PAGE_SIZE == eva) {
1926 pde = pmap_pde(pmap, sva);
1927 if (pde && (*pde & VPTE_PS) == 0) {
1928 pmap_remove_page(pmap, sva);
f7230403 1929 lwkt_reltoken(&vm_token);
b12defdc 1930 vm_object_drop(pmap->pm_pteobj);
da673940
JG
1931 return;
1932 }
1933 }
1934
1935 for (; sva < eva; sva = va_next) {
1936 pml4e = pmap_pml4e(pmap, sva);
1937 if ((*pml4e & VPTE_V) == 0) {
1938 va_next = (sva + NBPML4) & ~PML4MASK;
1939 if (va_next < sva)
1940 va_next = eva;
1941 continue;
1942 }
1943
1944 pdpe = pmap_pml4e_to_pdpe(pml4e, sva);
1945 if ((*pdpe & VPTE_V) == 0) {
1946 va_next = (sva + NBPDP) & ~PDPMASK;
1947 if (va_next < sva)
1948 va_next = eva;
1949 continue;
1950 }
1951
1952 /*
1953 * Calculate index for next page table.
1954 */
1955 va_next = (sva + NBPDR) & ~PDRMASK;
1956 if (va_next < sva)
1957 va_next = eva;
1958
1959 pde = pmap_pdpe_to_pde(pdpe, sva);
1960 ptpaddr = *pde;
1961
1962 /*
1963 * Weed out invalid mappings.
1964 */
1965 if (ptpaddr == 0)
1966 continue;
1967
1968 /*
1969 * Check for large page.
1970 */
1971 if ((ptpaddr & VPTE_PS) != 0) {
1972 /* JG FreeBSD has more complex treatment here */
1973 KKASSERT(*pde != 0);
1974 pmap_inval_pde(pde, pmap, sva);
1975 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1976 continue;
1977 }
1978
1979 /*
1980 * Limit our scan to either the end of the va represented
1981 * by the current page table page, or to the end of the
1982 * range being removed.
1983 */
1984 if (va_next > eva)
1985 va_next = eva;
1986
1987 /*
1988 * NOTE: pmap_remove_pte() can block.
1989 */
1990 for (pte = pmap_pde_to_pte(pde, sva); sva != va_next; pte++,
1991 sva += PAGE_SIZE) {
1992 if (*pte == 0)
1993 continue;
1994 if (pmap_remove_pte(pmap, pte, sva))
1995 break;
1996 }
1997 }
f7230403 1998 lwkt_reltoken(&vm_token);
b12defdc 1999 vm_object_drop(pmap->pm_pteobj);
da673940
JG
2000}
2001
2002/*
f7230403
MD
2003 * Removes this physical page from all physical maps in which it resides.
2004 * Reflects back modify bits to the pager.
da673940 2005 *
f7230403 2006 * This routine may not be called from an interrupt.
da673940 2007 *
f7230403 2008 * No requirements.
da673940 2009 */
da673940
JG
2010static void
2011pmap_remove_all(vm_page_t m)
2012{
2013 pt_entry_t *pte, tpte;
2014 pv_entry_t pv;
2015
2016#if defined(PMAP_DIAGNOSTIC)
2017 /*
2018 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
2019 * pages!
2020 */
2021 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
2022 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%08llx", (long long)VM_PAGE_TO_PHYS(m));
2023 }
2024#endif
2025
f7230403 2026 lwkt_gettoken(&vm_token);
da673940
JG
2027 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
2028 KKASSERT(pv->pv_pmap->pm_stats.resident_count > 0);
2029 --pv->pv_pmap->pm_stats.resident_count;
2030
2031 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
2032 KKASSERT(pte != NULL);
2033
2034 tpte = pmap_inval_loadandclear(pte, pv->pv_pmap, pv->pv_va);
2035 if (tpte & VPTE_WIRED)
2036 pv->pv_pmap->pm_stats.wired_count--;
2037 KKASSERT(pv->pv_pmap->pm_stats.wired_count >= 0);
2038
2039 if (tpte & VPTE_A)
2040 vm_page_flag_set(m, PG_REFERENCED);
2041
2042 /*
2043 * Update the vm_page_t clean and reference bits.
2044 */
2045 if (tpte & VPTE_M) {
2046#if defined(PMAP_DIAGNOSTIC)
2047 if (pmap_nw_modified(tpte)) {
2048 kprintf(
2049 "pmap_remove_all: modified page not writable: va: 0x%lx, pte: 0x%lx\n",
2050 pv->pv_va, tpte);
2051 }
2052#endif
2053 if (pmap_track_modified(pv->pv_pmap, pv->pv_va))
2054 vm_page_dirty(m);
2055 }
2056 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2057 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2058 ++pv->pv_pmap->pm_generation;
2059 m->md.pv_list_count--;
b12defdc 2060 atomic_add_int(&m->object->agg_pv_list_count, -1);
da673940
JG
2061 KKASSERT(m->md.pv_list_count >= 0);
2062 if (TAILQ_EMPTY(&m->md.pv_list))
2063 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
b12defdc 2064 vm_object_hold(pv->pv_pmap->pm_pteobj);
da673940 2065 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
b12defdc 2066 vm_object_drop(pv->pv_pmap->pm_pteobj);
da673940
JG
2067 free_pv_entry(pv);
2068 }
2069 KKASSERT((m->flags & (PG_MAPPED|PG_WRITEABLE)) == 0);
f7230403 2070 lwkt_reltoken(&vm_token);
da673940
JG
2071}
2072
2073/*
f7230403
MD
2074 * Set the physical protection on the specified range of this map
2075 * as requested.
da673940 2076 *
f7230403
MD
2077 * This function may not be called from an interrupt if the map is
2078 * not the kernel_pmap.
da673940 2079 *
f7230403 2080 * No requirements.
da673940
JG
2081 */
2082void
2083pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
2084{
2085 vm_offset_t va_next;
2086 pml4_entry_t *pml4e;
2087 pdp_entry_t *pdpe;
2088 pd_entry_t ptpaddr, *pde;
2089 pt_entry_t *pte;
2090
2091 /* JG review for NX */
2092
2093 if (pmap == NULL)
2094 return;
2095
2096 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2097 pmap_remove(pmap, sva, eva);
2098 return;
2099 }
2100
2101 if (prot & VM_PROT_WRITE)
2102 return;
2103
f7230403
MD
2104 lwkt_gettoken(&vm_token);
2105
da673940
JG
2106 for (; sva < eva; sva = va_next) {
2107
2108 pml4e = pmap_pml4e(pmap, sva);
2109 if ((*pml4e & VPTE_V) == 0) {
2110 va_next = (sva + NBPML4) & ~PML4MASK;
2111 if (va_next < sva)
2112 va_next = eva;
2113 continue;
2114 }
2115
2116 pdpe = pmap_pml4e_to_pdpe(pml4e, sva);
2117 if ((*pdpe & VPTE_V) == 0) {
2118 va_next = (sva + NBPDP) & ~PDPMASK;
2119 if (va_next < sva)
2120 va_next = eva;
2121 continue;
2122 }
2123
2124 va_next = (sva + NBPDR) & ~PDRMASK;
2125 if (va_next < sva)
2126 va_next = eva;
2127
2128 pde = pmap_pdpe_to_pde(pdpe, sva);
2129 ptpaddr = *pde;
2130
2131 /*
2132 * Check for large page.
2133 */
2134 if ((ptpaddr & VPTE_PS) != 0) {
2135 /* JG correct? */
2136 pmap_clean_pde(pde, pmap, sva);
2137 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
2138 continue;
2139 }
2140
2141 /*
2142 * Weed out invalid mappings. Note: we assume that the page
2143 * directory table is always allocated, and in kernel virtual.
2144 */
2145 if (ptpaddr == 0)
2146 continue;
2147
2148 if (va_next > eva)
2149 va_next = eva;
2150
2151 for (pte = pmap_pde_to_pte(pde, sva); sva != va_next; pte++,
2152 sva += PAGE_SIZE) {
2c2e847c 2153 pt_entry_t pbits;
da673940
JG
2154 vm_page_t m;
2155
2156 /*
2157 * Clean managed pages and also check the accessed
2158 * bit. Just remove write perms for unmanaged
2159 * pages. Be careful of races, turning off write
2160 * access will force a fault rather then setting
2161 * the modified bit at an unexpected time.
2162 */
2163 if (*pte & VPTE_MANAGED) {
2164 pbits = pmap_clean_pte(pte, pmap, sva);
2165 m = NULL;
2166 if (pbits & VPTE_A) {
2167 m = PHYS_TO_VM_PAGE(pbits & VPTE_FRAME);
2168 vm_page_flag_set(m, PG_REFERENCED);
2169 atomic_clear_long(pte, VPTE_A);
2170 }
2171 if (pbits & VPTE_M) {
2172 if (pmap_track_modified(pmap, sva)) {
2173 if (m == NULL)
2174 m = PHYS_TO_VM_PAGE(pbits & VPTE_FRAME);
2175 vm_page_dirty(m);
2176 }
2177 }
2178 } else {
2179 pbits = pmap_setro_pte(pte, pmap, sva);
2180 }
2181 }
2182 }
f7230403 2183 lwkt_reltoken(&vm_token);
da673940
JG
2184}
2185
2186/*
2187 * Enter a managed page into a pmap. If the page is not wired related pmap
2188 * data can be destroyed at any time for later demand-operation.
2189 *
2190 * Insert the vm_page (m) at virtual address (v) in (pmap), with the
2191 * specified protection, and wire the mapping if requested.
2192 *
2193 * NOTE: This routine may not lazy-evaluate or lose information. The
2194 * page must actually be inserted into the given map NOW.
2195 *
2196 * NOTE: When entering a page at a KVA address, the pmap must be the
2197 * kernel_pmap.
f7230403
MD
2198 *
2199 * No requirements.
da673940
JG
2200 */
2201void
2202pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
2203 boolean_t wired)
2204{
2205 vm_paddr_t pa;
2206 pd_entry_t *pde;
2207 pt_entry_t *pte;
2208 vm_paddr_t opa;
2209 pt_entry_t origpte, newpte;
2210 vm_page_t mpte;
2211
2212 if (pmap == NULL)
2213 return;
2214
2215 va = trunc_page(va);
2216
b12defdc 2217 vm_object_hold(pmap->pm_pteobj);
f7230403
MD
2218 lwkt_gettoken(&vm_token);
2219
da673940
JG
2220 /*
2221 * Get the page table page. The kernel_pmap's page table pages
2222 * are preallocated and have no associated vm_page_t.
2223 */
2224 if (pmap == &kernel_pmap)
2225 mpte = NULL;
2226 else
2227 mpte = pmap_allocpte(pmap, va);
2228
2229 pde = pmap_pde(pmap, va);
2230 if (pde != NULL && (*pde & VPTE_V) != 0) {
2231 if ((*pde & VPTE_PS) != 0)
2232 panic("pmap_enter: attempted pmap_enter on 2MB page");
2233 pte = pmap_pde_to_pte(pde, va);
f7230403 2234 } else {
da673940 2235 panic("pmap_enter: invalid page directory va=%#lx", va);
f7230403 2236 }
da673940
JG
2237
2238 KKASSERT(pte != NULL);
2239 /*
2240 * Deal with races on the original mapping (though don't worry
2241 * about VPTE_A races) by cleaning it. This will force a fault
2242 * if an attempt is made to write to the page.
2243 */
2244 pa = VM_PAGE_TO_PHYS(m);
2245 origpte = pmap_clean_pte(pte, pmap, va);
2246 opa = origpte & VPTE_FRAME;
2247
2248 if (origpte & VPTE_PS)
2249 panic("pmap_enter: attempted pmap_enter on 2MB page");
2250
2251 /*
2252 * Mapping has not changed, must be protection or wiring change.
2253 */
2254 if (origpte && (opa == pa)) {
2255 /*
2256 * Wiring change, just update stats. We don't worry about
2257 * wiring PT pages as they remain resident as long as there
2258 * are valid mappings in them. Hence, if a user page is wired,
2259 * the PT page will be also.
2260 */
2261 if (wired && ((origpte & VPTE_WIRED) == 0))
2262 ++pmap->pm_stats.wired_count;
2263 else if (!wired && (origpte & VPTE_WIRED))
2264 --pmap->pm_stats.wired_count;
2265
2266 /*
2267 * Remove the extra pte reference. Note that we cannot
2268 * optimize the RO->RW case because we have adjusted the
2269 * wiring count above and may need to adjust the wiring
2270 * bits below.
2271 */
2272 if (mpte)
2273 mpte->hold_count--;
2274
2275 /*
2276 * We might be turning off write access to the page,
2277 * so we go ahead and sense modify status.
2278 */
2279 if (origpte & VPTE_MANAGED) {
2280 if ((origpte & VPTE_M) &&
2281 pmap_track_modified(pmap, va)) {
2282 vm_page_t om;
2283 om = PHYS_TO_VM_PAGE(opa);
2284 vm_page_dirty(om);
2285 }
2286 pa |= VPTE_MANAGED;
2287 KKASSERT(m->flags & PG_MAPPED);
2288 }
2289 goto validate;
2290 }
2291 /*
2292 * Mapping has changed, invalidate old range and fall through to
2293 * handle validating new mapping.
2294 */
2295 if (opa) {
2296 int err;
2297 err = pmap_remove_pte(pmap, pte, va);
2298 if (err)
2299 panic("pmap_enter: pte vanished, va: 0x%lx", va);
2300 }
2301
2302 /*
2303 * Enter on the PV list if part of our managed memory. Note that we
2304 * raise IPL while manipulating pv_table since pmap_enter can be
2305 * called at interrupt time.
2306 */
2307 if (pmap_initialized &&
2308 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2309 pmap_insert_entry(pmap, va, mpte, m);
2310 pa |= VPTE_MANAGED;
2311 vm_page_flag_set(m, PG_MAPPED);
2312 }
2313
2314 /*
2315 * Increment counters
2316 */
2317 ++pmap->pm_stats.resident_count;
2318 if (wired)
2319 pmap->pm_stats.wired_count++;
2320
2321validate:
2322 /*
2323 * Now validate mapping with desired protection/wiring.
2324 */
2325 newpte = (pt_entry_t) (pa | pte_prot(pmap, prot) | VPTE_V);
2326
2327 if (wired)
2328 newpte |= VPTE_WIRED;
2329 if (pmap != &kernel_pmap)
2330 newpte |= VPTE_U;
2331
2332 /*
2333 * If the mapping or permission bits are different from the
2334 * (now cleaned) original pte, an update is needed. We've
2335 * already downgraded or invalidated the page so all we have
2336 * to do now is update the bits.
2337 *
2338 * XXX should we synchronize RO->RW changes to avoid another
2339 * fault?
2340 */
2341 if ((origpte & ~(VPTE_W|VPTE_M|VPTE_A)) != newpte) {
2342 *pte = newpte | VPTE_A;
2343 if (newpte & VPTE_W)
2344 vm_page_flag_set(m, PG_WRITEABLE);
2345 }
2346 KKASSERT((newpte & VPTE_MANAGED) == 0 || (m->flags & PG_MAPPED));
f7230403 2347 lwkt_reltoken(&vm_token);
b12defdc 2348 vm_object_drop(pmap->pm_pteobj);
da673940
JG
2349}
2350
2351/*
2352 * This code works like pmap_enter() but assumes VM_PROT_READ and not-wired.
2353 *
2354 * Currently this routine may only be used on user pmaps, not kernel_pmap.
f7230403
MD
2355 *
2356 * No requirements.
da673940 2357 */
0e6594a8 2358void
da673940
JG
2359pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m)
2360{
2361 pt_entry_t *pte;
2362 vm_paddr_t pa;
2363 vm_page_t mpte;
2364 vm_pindex_t ptepindex;
2365 pd_entry_t *ptepa;
2366
2367 KKASSERT(pmap != &kernel_pmap);
2368
2369 KKASSERT(va >= VM_MIN_USER_ADDRESS && va < VM_MAX_USER_ADDRESS);
2370
2371 /*
2372 * Calculate pagetable page index
2373 */
2374 ptepindex = pmap_pde_pindex(va);
2375
b12defdc 2376 vm_object_hold(pmap->pm_pteobj);
f7230403
MD
2377 lwkt_gettoken(&vm_token);
2378
da673940
JG
2379 do {
2380 /*
2381 * Get the page directory entry
2382 */
2383 ptepa = pmap_pde(pmap, va);
2384
2385 /*
2386 * If the page table page is mapped, we just increment
2387 * the hold count, and activate it.
2388 */
2389 if (ptepa && (*ptepa & VPTE_V) != 0) {
2390 if (*ptepa & VPTE_PS)
2391 panic("pmap_enter_quick: unexpected mapping into 2MB page");
2392 if (pmap->pm_ptphint &&
2393 (pmap->pm_ptphint->pindex == ptepindex)) {
2394 mpte = pmap->pm_ptphint;
2395 } else {
2396 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2397 pmap->pm_ptphint = mpte;
b12defdc 2398 vm_page_wakeup(mpte);
da673940
JG
2399 }
2400 if (mpte)
2401 mpte->hold_count++;
2402 } else {
2403 mpte = _pmap_allocpte(pmap, ptepindex);
2404 }
2405 } while (mpte == NULL);
2406
2407 /*
2408 * Ok, now that the page table page has been validated, get the pte.
2409 * If the pte is already mapped undo mpte's hold_count and
2410 * just return.
2411 */
2412 pte = pmap_pte(pmap, va);
2413 if (*pte & VPTE_V) {
2414 KKASSERT(mpte != NULL);
2415 pmap_unwire_pte_hold(pmap, va, mpte);
2416 pa = VM_PAGE_TO_PHYS(m);
2417 KKASSERT(((*pte ^ pa) & VPTE_FRAME) == 0);
f7230403 2418 lwkt_reltoken(&vm_token);
b12defdc 2419 vm_object_drop(pmap->pm_pteobj);
da673940
JG
2420 return;
2421 }
2422
2423 /*
2424 * Enter on the PV list if part of our managed memory
2425 */
2426 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2427 pmap_insert_entry(pmap, va, mpte, m);
2428 vm_page_flag_set(m, PG_MAPPED);
2429 }
2430
2431 /*
2432 * Increment counters
2433 */
2434 ++pmap->pm_stats.resident_count;
2435
2436 pa = VM_PAGE_TO_PHYS(m);
2437
2438 /*
2439 * Now validate mapping with RO protection
2440 */
2441 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2442 *pte = (vpte_t)pa | VPTE_V | VPTE_U;
2443 else
2444 *pte = (vpte_t)pa | VPTE_V | VPTE_U | VPTE_MANAGED;
2445 /*pmap_inval_add(&info, pmap, va); shouldn't be needed 0->valid */
2446 /*pmap_inval_flush(&info); don't need for vkernel */
f7230403 2447 lwkt_reltoken(&vm_token);
b12defdc 2448 vm_object_drop(pmap->pm_pteobj);
da673940
JG
2449}
2450
2451/*
2452 * Make a temporary mapping for a physical address. This is only intended
2453 * to be used for panic dumps.
fb8345e6
MD
2454 *
2455 * The caller is responsible for calling smp_invltlb().
da673940
JG
2456 */
2457void *
8e5ea5f7 2458pmap_kenter_temporary(vm_paddr_t pa, long i)
da673940 2459{
fb8345e6 2460 pmap_kenter_quick(crashdumpmap + (i * PAGE_SIZE), pa);
da673940
JG
2461 return ((void *)crashdumpmap);
2462}
2463
2464#define MAX_INIT_PT (96)
2465
2466/*
2467 * This routine preloads the ptes for a given object into the specified pmap.
2468 * This eliminates the blast of soft faults on process startup and
2469 * immediately after an mmap.
f7230403
MD
2470 *
2471 * No requirements.
da673940
JG
2472 */
2473static int pmap_object_init_pt_callback(vm_page_t p, void *data);
2474
2475void
2476pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_prot_t prot,
2477 vm_object_t object, vm_pindex_t pindex,
2478 vm_size_t size, int limit)
2479{
2480 struct rb_vm_page_scan_info info;
2481 struct lwp *lp;
2482 vm_size_t psize;
2483
2484 /*
2485 * We can't preinit if read access isn't set or there is no pmap
2486 * or object.
2487 */
2488 if ((prot & VM_PROT_READ) == 0 || pmap == NULL || object == NULL)
2489 return;
2490
2491 /*
2492 * We can't preinit if the pmap is not the current pmap
2493 */
2494 lp = curthread->td_lwp;
2495 if (lp == NULL || pmap != vmspace_pmap(lp->lwp_vmspace))
2496 return;
2497
0e6594a8 2498 psize = x86_64_btop(size);
da673940
JG
2499
2500 if ((object->type != OBJT_VNODE) ||
2501 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2502 (object->resident_page_count > MAX_INIT_PT))) {
2503 return;
2504 }
2505
2506 if (psize + pindex > object->size) {
2507 if (object->size < pindex)
2508 return;
2509 psize = object->size - pindex;
2510 }
2511
2512 if (psize == 0)
2513 return;
2514
2515 /*
2516 * Use a red-black scan to traverse the requested range and load
2517 * any valid pages found into the pmap.
2518 *
2519 * We cannot safely scan the object's memq unless we are in a
2520 * critical section since interrupts can remove pages from objects.
2521 */
2522 info.start_pindex = pindex;
2523 info.end_pindex = pindex + psize - 1;
2524 info.limit = limit;
2525 info.mpte = NULL;
2526 info.addr = addr;
2527 info.pmap = pmap;
2528
b12defdc 2529 vm_object_hold(object);
da673940
JG
2530 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
2531 pmap_object_init_pt_callback, &info);
b12defdc 2532 vm_object_drop(object);
da673940
JG
2533}
2534
2535static
2536int
2537pmap_object_init_pt_callback(vm_page_t p, void *data)
2538{
2539 struct rb_vm_page_scan_info *info = data;
2540 vm_pindex_t rel_index;
2541 /*
2542 * don't allow an madvise to blow away our really
2543 * free pages allocating pv entries.
2544 */
2545 if ((info->limit & MAP_PREFAULT_MADVISE) &&
2546 vmstats.v_free_count < vmstats.v_free_reserved) {
2547 return(-1);
2548 }
0d987a03
MD
2549
2550 /*
2551 * Ignore list markers and ignore pages we cannot instantly
2552 * busy (while holding the object token).
2553 */
2554 if (p->flags & PG_MARKER)
2555 return 0;
b12defdc
MD
2556 if (vm_page_busy_try(p, TRUE))
2557 return 0;
da673940 2558 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
b12defdc 2559 (p->flags & PG_FICTITIOUS) == 0) {
da673940
JG
2560 if ((p->queue - p->pc) == PQ_CACHE)
2561 vm_page_deactivate(p);
da673940
JG
2562 rel_index = p->pindex - info->start_pindex;
2563 pmap_enter_quick(info->pmap,
0e6594a8 2564 info->addr + x86_64_ptob(rel_index), p);
da673940 2565 }
b12defdc 2566 vm_page_wakeup(p);
da673940
JG
2567 return(0);
2568}
2569
2570/*
0e6594a8
SW
2571 * Return TRUE if the pmap is in shape to trivially
2572 * pre-fault the specified address.
2573 *
2574 * Returns FALSE if it would be non-trivial or if a
2575 * pte is already loaded into the slot.
f7230403
MD
2576 *
2577 * No requirements.
da673940 2578 */
0e6594a8
SW
2579int
2580pmap_prefault_ok(pmap_t pmap, vm_offset_t addr)
da673940 2581{
0e6594a8
SW
2582 pt_entry_t *pte;
2583 pd_entry_t *pde;
f7230403 2584 int ret;
da673940 2585
f7230403 2586 lwkt_gettoken(&vm_token);
0e6594a8 2587 pde = pmap_pde(pmap, addr);
f7230403
MD
2588 if (pde == NULL || *pde == 0) {
2589 ret = 0;
2590 } else {
2591 pte = pmap_pde_to_pte(pde, addr);
2592 ret = (*pte) ? 0 : 1;
2593 }
2594 lwkt_reltoken(&vm_token);
2595 return (ret);
da673940
JG
2596}
2597
2598/*
f7230403
MD
2599 * Change the wiring attribute for a map/virtual-address pair.
2600 *
2601 * The mapping must already exist in the pmap.
2602 * No other requirements.
da673940
JG
2603 */
2604void
2605pmap_change_wiring(pmap_t pmap, vm_offset_t va, boolean_t wired)
2606{
2607 pt_entry_t *pte;
2608
2609 if (pmap == NULL)
2610 return;
2611
f7230403 2612 lwkt_gettoken(&vm_token);
da673940
JG
2613 pte = pmap_pte(pmap, va);
2614
2615 if (wired && !pmap_pte_w(pte))
2616 pmap->pm_stats.wired_count++;
2617 else if (!wired && pmap_pte_w(pte))
2618 pmap->pm_stats.wired_count--;
2619
2620 /*
2621 * Wiring is not a hardware characteristic so there is no need to
2622 * invalidate TLB. However, in an SMP environment we must use
2623 * a locked bus cycle to update the pte (if we are not using
2624 * the pmap_inval_*() API that is)... it's ok to do this for simple
2625 * wiring changes.
2626 */
2627 if (wired)
2628 atomic_set_long(pte, VPTE_WIRED);
2629 else
2630 atomic_clear_long(pte, VPTE_WIRED);
f7230403 2631 lwkt_reltoken(&vm_token);
da673940
JG
2632}
2633
2634/*
2635 * Copy the range specified by src_addr/len
2636 * from the source map to the range dst_addr/len
2637 * in the destination map.
2638 *
2639 * This routine is only advisory and need not do anything.
2640 */
2641void
2642pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
2643 vm_size_t len, vm_offset_t src_addr)
2644{
2645 /*
2646 * XXX BUGGY. Amoung other things srcmpte is assumed to remain
2647 * valid through blocking calls, and that's just not going to
2648 * be the case.
2649 *
2650 * FIXME!
2651 */
2652 return;
2653}
2654
2655/*
2656 * pmap_zero_page:
2657 *
2658 * Zero the specified physical page.
2659 *
2660 * This function may be called from an interrupt and no locking is
2661 * required.
2662 */
2663void
2664pmap_zero_page(vm_paddr_t phys)
2665{
2666 vm_offset_t va = PHYS_TO_DMAP(phys);
2667
2668 bzero((void *)va, PAGE_SIZE);
2669}
2670
2671/*
2672 * pmap_page_assertzero:
2673 *
2674 * Assert that a page is empty, panic if it isn't.
2675 */
2676void
2677pmap_page_assertzero(vm_paddr_t phys)
2678{
2679 int i;
2680
2681 crit_enter();
2682 vm_offset_t virt = PHYS_TO_DMAP(phys);
2683
2684 for (i = 0; i < PAGE_SIZE; i += sizeof(int)) {
2685 if (*(int *)((char *)virt + i) != 0) {
2686 panic("pmap_page_assertzero() @ %p not zero!\n",
2687 (void *)virt);
2688 }
2689 }
2690 crit_exit();
2691}
2692
2693/*
2694 * pmap_zero_page:
2695 *
2696 * Zero part of a physical page by mapping it into memory and clearing
2697 * its contents with bzero.
2698 *
2699 * off and size may not cover an area beyond a single hardware page.
2700 */
2701void
2702pmap_zero_page_area(vm_paddr_t phys, int off, int size)
2703{
2704 crit_enter();
2705 vm_offset_t virt = PHYS_TO_DMAP(phys);
2706 bzero((char *)virt + off, size);
2707 crit_exit();
2708}
2709
2710/*
2711 * pmap_copy_page:
2712 *
2713 * Copy the physical page from the source PA to the target PA.
2714 * This function may be called from an interrupt. No locking
2715 * is required.
2716 */
2717void
2718pmap_copy_page(vm_paddr_t src, vm_paddr_t dst)
2719{
2720 vm_offset_t src_virt, dst_virt;
2721
2722 crit_enter();
2723 src_virt = PHYS_TO_DMAP(src);
2724 dst_virt = PHYS_TO_DMAP(dst);
2c2e847c 2725 bcopy((void *)src_virt, (void *)dst_virt, PAGE_SIZE);
da673940
JG
2726 crit_exit();
2727}
2728
2729/*
2730 * pmap_copy_page_frag:
2731 *
2732 * Copy the physical page from the source PA to the target PA.
2733 * This function may be called from an interrupt. No locking
2734 * is required.
2735 */
2736void
2737pmap_copy_page_frag(vm_paddr_t src, vm_paddr_t dst, size_t bytes)
2738{
2739 vm_offset_t src_virt, dst_virt;
2740
2741 crit_enter();
2742 src_virt = PHYS_TO_DMAP(src);
2743 dst_virt = PHYS_TO_DMAP(dst);
2744 bcopy((char *)src_virt + (src & PAGE_MASK),
2745 (char *)dst_virt + (dst & PAGE_MASK),
2746 bytes);
2747 crit_exit();
2748}
2749
2750/*
f7230403
MD
2751 * Returns true if the pmap's pv is one of the first 16 pvs linked to
2752 * from this page. This count may be changed upwards or downwards
2753 * in the future; it is only necessary that true be returned for a small
da673940 2754 * subset of pmaps for proper page aging.
f7230403
MD
2755 *
2756 * No other requirements.
da673940
JG
2757 */
2758boolean_t
2759pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
2760{
2761 pv_entry_t pv;
2762 int loops = 0;
2763
2764 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2765 return FALSE;
2766
2767 crit_enter();
f7230403 2768 lwkt_gettoken(&vm_token);
da673940
JG
2769
2770 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2771 if (pv->pv_pmap == pmap) {
f7230403 2772 lwkt_reltoken(&vm_token);
da673940
JG
2773 crit_exit();
2774 return TRUE;
2775 }
2776 loops++;
2777 if (loops >= 16)
2778 break;
2779 }
f7230403 2780 lwkt_reltoken(&vm_token);
da673940
JG
2781 crit_exit();
2782 return (FALSE);
2783}
2784
2785/*
f7230403
MD
2786 * Remove all pages from specified address space this aids process
2787 * exit speeds. Also, this code is special cased for current
2788 * process only, but can have the more generic (and slightly slower)
2789 * mode enabled. This is much faster than pmap_remove in the case
2790 * of running down an entire address space.
2791 *
2792 * No other requirements.
da673940
JG
2793 */
2794void
2795pmap_remove_pages(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2796{
da673940
JG
2797 pt_entry_t *pte, tpte;
2798 pv_entry_t pv, npv;
2799 vm_page_t m;
da673940
JG
2800 int save_generation;
2801
b12defdc
MD
2802 if (pmap->pm_pteobj)
2803 vm_object_hold(pmap->pm_pteobj);
f7230403 2804 lwkt_gettoken(&vm_token);
b12defdc 2805
da673940
JG
2806 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); pv; pv = npv) {
2807 if (pv->pv_va >= eva || pv->pv_va < sva) {
2808 npv = TAILQ_NEXT(pv, pv_plist);
2809 continue;
2810 }
2811
2812 KKASSERT(pmap == pv->pv_pmap);
2813
2814 pte = pmap_pte(pmap, pv->pv_va);
2815
2816 /*
2817 * We cannot remove wired pages from a process' mapping
2818 * at this time
2819 */
2820 if (*pte & VPTE_WIRED) {
2821 npv = TAILQ_NEXT(pv, pv_plist);
2822 continue;
2823 }
2824 tpte = pmap_inval_loadandclear(pte, pmap, pv->pv_va);
2825
2826 m = PHYS_TO_VM_PAGE(tpte & VPTE_FRAME);
2827
2828 KASSERT(m < &vm_page_array[vm_page_array_size],
2829 ("pmap_remove_pages: bad tpte %lx", tpte));
2830
2831 KKASSERT(pmap->pm_stats.resident_count > 0);
2832 --pmap->pm_stats.resident_count;
2833
2834 /*
2835 * Update the vm_page_t clean and reference bits.
2836 */
2837 if (tpte & VPTE_M) {
2838 vm_page_dirty(m);
2839 }
2840
2841 npv = TAILQ_NEXT(pv, pv_plist);
2842 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
2843 save_generation = ++pmap->pm_generation;
2844
2845 m->md.pv_list_count--;
b12defdc 2846 atomic_add_int(&m->object->agg_pv_list_count, -1);
da673940
JG
2847 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2848 if (TAILQ_EMPTY(&m->md.pv_list))
2849 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2850
2851 pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem);
2852 free_pv_entry(pv);
2853
2854 /*
2855 * Restart the scan if we blocked during the unuse or free
2856 * calls and other removals were made.
2857 */
2858 if (save_generation != pmap->pm_generation) {
2859 kprintf("Warning: pmap_remove_pages race-A avoided\n");
cd2a0876 2860 npv = TAILQ_FIRST(&pmap->pm_pvlist);
da673940
JG
2861 }
2862 }
f7230403 2863 lwkt_reltoken(&vm_token);
b12defdc
MD
2864 if (pmap->pm_pteobj)
2865 vm_object_drop(pmap->pm_pteobj);
da673940
JG
2866}
2867
2868/*
2869 * pmap_testbit tests bits in active mappings of a VM page.
2870 */
2871static boolean_t
2872pmap_testbit(vm_page_t m, int bit)
2873{
2874 pv_entry_t pv;
2875 pt_entry_t *pte;
2876
2877 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2878 return FALSE;
2879
2880 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
2881 return FALSE;
2882
2883 crit_enter();
2884
2885 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2886 /*
2887 * if the bit being tested is the modified bit, then
2888 * mark clean_map and ptes as never
2889 * modified.
2890 */
2891 if (bit & (VPTE_A|VPTE_M)) {
2892 if (!pmap_track_modified(pv->pv_pmap, pv->pv_va))
2893 continue;
2894 }
2895
2896#if defined(PMAP_DIAGNOSTIC)
2897 if (pv->pv_pmap == NULL) {
2898 kprintf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va);
2899 continue;
2900 }
2901#endif
2902 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
2903 if (*pte & bit) {
2904 crit_exit();
2905 return TRUE;
2906 }
2907 }
2908 crit_exit();
2909 return (FALSE);
2910}
2911
2912/*
2913 * This routine is used to clear bits in ptes. Certain bits require special
2914 * handling, in particular (on virtual kernels) the VPTE_M (modify) bit.
2915 *
2916 * This routine is only called with certain VPTE_* bit combinations.
2917 */
2918static __inline void
2919pmap_clearbit(vm_page_t m, int bit)
2920{
2921 pv_entry_t pv;
2922 pt_entry_t *pte;
2923 pt_entry_t pbits;
2924
2925 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2926 return;
2927
2928 crit_enter();
2929
2930 /*
2931 * Loop over all current mappings setting/clearing as appropos If
2932 * setting RO do we need to clear the VAC?
2933 */
2934 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2935 /*
2936 * don't write protect pager mappings
2937 */
2938 if (bit == VPTE_W) {
2939 if (!pmap_track_modified(pv->pv_pmap, pv->pv_va))
2940 continue;
2941 }
2942
2943#if defined(PMAP_DIAGNOSTIC)
2944 if (pv->pv_pmap == NULL) {
2945 kprintf("Null pmap (cb) at va: 0x%lx\n", pv->pv_va);
2946 continue;
2947 }
2948#endif
2949
2950 /*
2951 * Careful here. We can use a locked bus instruction to
2952 * clear VPTE_A or VPTE_M safely but we need to synchronize
2953 * with the target cpus when we mess with VPTE_W.
2954 *
2955 * On virtual kernels we must force a new fault-on-write
2956 * in the real kernel if we clear the Modify bit ourselves,
2957 * otherwise the real kernel will not get a new fault and
2958 * will never set our Modify bit again.
2959 */
2960 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
2961 if (*pte & bit) {
2962 if (bit == VPTE_W) {
2963 /*
2964 * We must also clear VPTE_M when clearing
2965 * VPTE_W
2966 */
2967 pbits = pmap_clean_pte(pte, pv->pv_pmap,
2968 pv->pv_va);
2969 if (pbits & VPTE_M)
2970 vm_page_dirty(m);
2971 } else if (bit == VPTE_M) {
2972 /*
2973 * We do not have to make the page read-only
2974 * when clearing the Modify bit. The real
2975 * kernel will make the real PTE read-only
2976 * or otherwise detect the write and set
2977 * our VPTE_M again simply by us invalidating
2978 * the real kernel VA for the pmap (as we did
2979 * above). This allows the real kernel to
2980 * handle the write fault without forwarding
2981 * the fault to us.
2982 */
2983 atomic_clear_long(pte, VPTE_M);
2984 } else if ((bit & (VPTE_W|VPTE_M)) == (VPTE_W|VPTE_M)) {
2985 /*
2986 * We've been asked to clear W & M, I guess
2987 * the caller doesn't want us to update
2988 * the dirty status of the VM page.
2989 */
2990 pmap_clean_pte(pte, pv->pv_pmap, pv->pv_va);
2991 } else {
2992 /*
2993 * We've been asked to clear bits that do
2994 * not interact with hardware.
2995 */
2996 atomic_clear_long(pte, bit);
2997 }
2998 }
2999 }
3000 crit_exit();
3001}
3002
3003/*
f7230403 3004 * Lower the permission for all mappings to a given page.
da673940 3005 *
f7230403 3006 * No other requirements.
da673940
JG
3007 */
3008void
3009pmap_page_protect(vm_page_t m, vm_prot_t prot)
3010{
3011 /* JG NX support? */
3012 if ((prot & VM_PROT_WRITE) == 0) {
f7230403 3013 lwkt_gettoken(&vm_token);
da673940
JG
3014 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3015 pmap_clearbit(m, VPTE_W);
3016 vm_page_flag_clear(m, PG_WRITEABLE);
3017 } else {
3018 pmap_remove_all(m);
3019 }
f7230403 3020 lwkt_reltoken(&vm_token);
da673940
JG
3021 }
3022}
3023
3024vm_paddr_t
3025pmap_phys_address(vm_pindex_t ppn)
3026{
0e6594a8 3027 return (x86_64_ptob(ppn));
da673940
JG
3028}
3029
3030/*
f7230403
MD
3031 * Return a count of reference bits for a page, clearing those bits.
3032 * It is not necessary for every reference bit to be cleared, but it
3033 * is necessary that 0 only be returned when there are truly no
3034 * reference bits set.
da673940 3035 *
f7230403
MD
3036 * XXX: The exact number of bits to check and clear is a matter that
3037 * should be tested and standardized at some point in the future for
3038 * optimal aging of shared pages.
da673940 3039 *
f7230403 3040 * No other requirements.
da673940
JG
3041 */
3042int
3043pmap_ts_referenced(vm_page_t m)
3044{
3045 pv_entry_t pv, pvf, pvn;
3046 pt_entry_t *pte;
3047 int rtval = 0;
3048
3049 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3050 return (rtval);
3051
3052 crit_enter();
f7230403 3053 lwkt_gettoken(&vm_token);
da673940
JG
3054
3055 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3056
3057 pvf = pv;
3058
3059 do {
3060 pvn = TAILQ_NEXT(pv, pv_list);
3061
3062 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3063
3064 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3065
3066 if (!pmap_track_modified(pv->pv_pmap, pv->pv_va))
3067 continue;
3068
3069 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
3070
3071 if (pte && (*pte & VPTE_A)) {
3072#ifdef SMP
3073 atomic_clear_long(pte, VPTE_A);
3074#else
3075 atomic_clear_long_nonlocked(pte, VPTE_A);
3076#endif
3077 rtval++;
3078 if (rtval > 4) {
3079 break;
3080 }
3081 }
3082 } while ((pv = pvn) != NULL && pv != pvf);
3083 }
f7230403 3084 lwkt_reltoken(&vm_token);
da673940
JG
3085 crit_exit();
3086
3087 return (rtval);
3088}
3089
3090/*
f7230403
MD
3091 * Return whether or not the specified physical page was modified
3092 * in any physical maps.
da673940 3093 *
f7230403 3094 * No other requirements.
da673940
JG
3095 */
3096boolean_t
3097pmap_is_modified(vm_page_t m)
3098{
f7230403
MD
3099 boolean_t res;
3100
3101 lwkt_gettoken(&vm_token);
3102 res = pmap_testbit(m, VPTE_M);
3103 lwkt_reltoken(&vm_token);
3104 return (res);
da673940
JG
3105}
3106
3107/*
f7230403
MD
3108 * Clear the modify bits on the specified physical page.
3109 *
3110 * No other requirements.
da673940
JG
3111 */
3112void
3113pmap_clear_modify(vm_page_t m)
3114{
f7230403 3115 lwkt_gettoken(&vm_token);
da673940 3116 pmap_clearbit(m, VPTE_M);
f7230403 3117 lwkt_reltoken(&vm_token);
da673940
JG
3118}
3119
3120/*
f7230403 3121 * Clear the reference bit on the specified physical page.
da673940 3122 *
f7230403 3123 * No other requirements.
da673940
JG
3124 */
3125void
3126pmap_clear_reference(vm_page_t m)
3127{
f7230403 3128 lwkt_gettoken(&vm_token);
da673940 3129 pmap_clearbit(m, VPTE_A);
f7230403 3130 lwkt_reltoken(&vm_token);
da673940
JG
3131}
3132
3133/*
3134 * Miscellaneous support routines follow
3135 */
3136
3137static void
3138i386_protection_init(void)
3139{
3140 int *kp, prot;
3141
3142 kp = protection_codes;
3143 for (prot = 0; prot < 8; prot++) {
3144 if (prot & VM_PROT_READ)
3145 *kp |= VPTE_R;
3146 if (prot & VM_PROT_WRITE)
3147 *kp |= VPTE_W;
3148 if (prot & VM_PROT_EXECUTE)
3149 *kp |= VPTE_X;
3150 ++kp;
3151 }
3152}
3153
3154/*
f7230403
MD
3155 * Perform the pmap work for mincore
3156 *
3157 * No other requirements.
da673940
JG
3158 */
3159int
3160pmap_mincore(pmap_t pmap, vm_offset_t addr)
3161{
3162 pt_entry_t *ptep, pte;
3163 vm_page_t m;
3164 int val = 0;
3165
f7230403 3166 lwkt_gettoken(&vm_token);
da673940 3167 ptep = pmap_pte(pmap, addr);
da673940 3168
f7230403 3169 if (ptep && (pte = *ptep) != 0) {
8608b858 3170 vm_paddr_t pa;
da673940
JG
3171
3172 val = MINCORE_INCORE;
3173 if ((pte & VPTE_MANAGED) == 0)
f7230403 3174 goto done;
da673940
JG
3175
3176 pa = pte & VPTE_FRAME;
3177
3178 m = PHYS_TO_VM_PAGE(pa);
3179
3180 /*
3181 * Modified by us
3182 */
3183 if (pte & VPTE_M)
3184 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3185 /*
3186 * Modified by someone
3187 */
3188 else if (m->dirty || pmap_is_modified(m))
3189 val |= MINCORE_MODIFIED_OTHER;
3190 /*
3191 * Referenced by us
3192 */
3193 if (pte & VPTE_A)
3194 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3195
3196 /*
3197 * Referenced by someone
3198 */
3199 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3200 val |= MINCORE_REFERENCED_OTHER;
3201 vm_page_flag_set(m, PG_REFERENCED);
3202 }
3203 }
f7230403
MD
3204done:
3205 lwkt_reltoken(&vm_token);
da673940
JG
3206 return val;
3207}
3208
3209/*
3210 * Replace p->p_vmspace with a new one. If adjrefs is non-zero the new
3211 * vmspace will be ref'd and the old one will be deref'd.
b12defdc
MD
3212 *
3213 * Caller must hold vmspace->vm_map.token for oldvm and newvm
da673940
JG
3214 */
3215void
3216pmap_replacevm(struct proc *p, struct vmspace *newvm, int adjrefs)
3217{
3218 struct vmspace *oldvm;
3219 struct lwp *lp;
3220
3221 crit_enter();
3222 oldvm = p->p_vmspace;
3223 if (oldvm != newvm) {
3224 p->p_vmspace = newvm;
3225 KKASSERT(p->p_nthreads == 1);
3226 lp = RB_ROOT(&p->p_lwp_tree);
3227 pmap_setlwpvm(lp, newvm);
3228 if (adjrefs) {
3229 sysref_get(&newvm->vm_sysref);
3230 sysref_put(&oldvm->vm_sysref);
3231 }
3232 }
3233 crit_exit();
3234}
3235
3236/*
3237 * Set the vmspace for a LWP. The vmspace is almost universally set the
3238 * same as the process vmspace, but virtual kernels need to swap out contexts
3239 * on a per-lwp basis.
3240 */
3241void
3242pmap_setlwpvm(struct lwp *lp, struct vmspace *newvm)
3243{
3244 struct vmspace *oldvm;
3245 struct pmap *pmap;
3246
3247 crit_enter();
3248 oldvm = lp->lwp_vmspace;
3249
3250 if (oldvm != newvm) {
3251 lp->lwp_vmspace = newvm;
3252 if (curthread->td_lwp == lp) {
3253 pmap = vmspace_pmap(newvm);
3254#if defined(SMP)
da23a592 3255 atomic_set_cpumask(&pmap->pm_active, CPUMASK(mycpu->gd_cpuid));
da673940
JG
3256#else
3257 pmap->pm_active |= 1;
3258#endif
3259#if defined(SWTCH_OPTIM_STATS)
3260 tlb_flush_count++;
3261#endif
3262 pmap = vmspace_pmap(oldvm);
3263#if defined(SMP)
da23a592
MD
3264 atomic_clear_cpumask(&pmap->pm_active,
3265 CPUMASK(mycpu->gd_cpuid));
da673940 3266#else
da23a592 3267 pmap->pm_active &= ~(cpumask_t)1;
da673940
JG
3268#endif
3269 }
3270 }
3271 crit_exit();
3272}
3273
3274vm_offset_t
3275pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3276{
3277
3278 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3279 return addr;
3280 }
3281
3282 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3283 return addr;
3284}
722871d3
MD
3285
3286/*
3287 * Used by kmalloc/kfree, page already exists at va
3288 */
3289vm_page_t
3290pmap_kvtom(vm_offset_t va)
3291{
3292 vpte_t *ptep;
3293
3294 KKASSERT(va >= KvaStart && va < KvaEnd);
5e6356ab
SW
3295 ptep = vtopte(va);
3296 return(PHYS_TO_VM_PAGE(*ptep & PG_FRAME));
722871d3 3297}