2 * Copyright (c) 1991 Regents of the University of California.
4 * Copyright (c) 1994 John S. Dyson
6 * Copyright (c) 1994 David Greenman
9 * This code is derived from software contributed to Berkeley by
10 * the Systems Programming Group of the University of Utah Computer
11 * Science Department and William Jolitz of UUNET Technologies Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91
42 * $FreeBSD: src/sys/i386/i386/pmap.c,v 1.250.2.18 2002/03/06 22:48:53 silby Exp $
46 * Manages physical address maps.
48 * In addition to hardware address maps, this
49 * module is called upon to provide software-use-only
50 * maps which may or may not be stored in the same
51 * form as hardware maps. These pseudo-maps are
52 * used to store intermediate results from copy
53 * operations to and from address spaces.
55 * Since the information managed by this module is
56 * also stored by the logical address mapping module,
57 * this module may throw away valid virtual-to-physical
58 * mappings at almost any time. However, invalidations
59 * of virtual-to-physical mappings must be done as
62 * In order to cope with hardware architectures which
63 * make virtual-to-physical map invalidates expensive,
64 * this module may delay invalidate or reduced protection
65 * operations until such time as they are actually
66 * necessary. This module is given full information as
67 * to which processors are currently using which maps,
68 * and to when physical maps must be made correct.
71 #include "opt_disable_pse.h"
73 #include "opt_msgbuf.h"
74 #include "opt_user_ldt.h"
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
80 #include <sys/msgbuf.h>
81 #include <sys/vmmeter.h>
85 #include <vm/vm_param.h>
86 #include <sys/sysctl.h>
88 #include <vm/vm_kern.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_pageout.h>
94 #include <vm/vm_pager.h>
95 #include <vm/vm_zone.h>
99 #include <machine/cputypes.h>
100 #include <machine/md_var.h>
101 #include <machine/specialreg.h>
102 #if defined(SMP) || defined(APIC_IO)
103 #include <machine/smp.h>
104 #include <machine/apic.h>
105 #include <machine/segments.h>
106 #include <machine/tss.h>
107 #include <machine/globaldata.h>
108 #endif /* SMP || APIC_IO */
110 #define PMAP_KEEP_PDIRS
111 #ifndef PMAP_SHPGPERPROC
112 #define PMAP_SHPGPERPROC 200
115 #if defined(DIAGNOSTIC)
116 #define PMAP_DIAGNOSTIC
121 #if !defined(PMAP_DIAGNOSTIC)
122 #define PMAP_INLINE __inline
128 * Get PDEs and PTEs for user/kernel address space
130 #define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
131 #define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
133 #define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0)
134 #define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0)
135 #define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0)
136 #define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0)
137 #define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0)
139 #define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
140 #define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
143 * Given a map and a machine independent protection code,
144 * convert to a vax protection code.
146 #define pte_prot(m, p) (protection_codes[p])
147 static int protection_codes[8];
149 static struct pmap kernel_pmap_store;
152 vm_offset_t avail_start; /* PA of first available physical page */
153 vm_offset_t avail_end; /* PA of last available physical page */
154 vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */
155 vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */
156 static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
157 static int pgeflag; /* PG_G or-in */
158 static int pseflag; /* PG_PS or-in */
160 static vm_object_t kptobj;
163 vm_offset_t kernel_vm_end;
166 * Data for the pv entry allocation mechanism
168 static vm_zone_t pvzone;
169 static struct vm_zone pvzone_store;
170 static struct vm_object pvzone_obj;
171 static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
172 static int pmap_pagedaemon_waken = 0;
173 static struct pv_entry *pvinit;
176 * All those kernel PT submaps that BSD is so fond of
178 pt_entry_t *CMAP1 = 0;
179 static pt_entry_t *CMAP2, *ptmmap;
180 caddr_t CADDR1 = 0, ptvmmap = 0;
181 static caddr_t CADDR2;
182 static pt_entry_t *msgbufmap;
183 struct msgbuf *msgbufp=0;
188 static pt_entry_t *pt_crashdumpmap;
189 static caddr_t crashdumpmap;
192 extern pt_entry_t *SMPpt;
194 static pt_entry_t *PMAP1 = 0;
195 static unsigned *PADDR1 = 0;
198 static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv));
199 static unsigned * get_ptbase __P((pmap_t pmap));
200 static pv_entry_t get_pv_entry __P((void));
201 static void i386_protection_init __P((void));
202 static __inline void pmap_changebit __P((vm_page_t m, int bit, boolean_t setem));
204 static void pmap_remove_all __P((vm_page_t m));
205 static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va,
206 vm_page_t m, vm_page_t mpte));
207 static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq,
209 static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va));
210 static int pmap_remove_entry __P((struct pmap *pmap, vm_page_t m,
212 static boolean_t pmap_testbit __P((vm_page_t m, int bit));
213 static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va,
214 vm_page_t mpte, vm_page_t m));
216 static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va));
218 static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p));
219 static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex));
220 static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va));
221 static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex));
222 static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t));
223 static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
225 static unsigned pdir4mb;
230 * Extract the page table entry associated
231 * with the given map/virtual_address pair.
234 PMAP_INLINE unsigned *
236 register pmap_t pmap;
242 pdeaddr = (unsigned *) pmap_pde(pmap, va);
243 if (*pdeaddr & PG_PS)
246 return get_ptbase(pmap) + i386_btop(va);
253 * Move the kernel virtual free pointer to the next
254 * 4MB. This is used to help improve performance
255 * by using a large (4MB) page for much of the kernel
256 * (.text, .data, .bss)
259 pmap_kmem_choose(vm_offset_t addr)
261 vm_offset_t newaddr = addr;
263 if (cpu_feature & CPUID_PSE) {
264 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
271 * Bootstrap the system enough to run with virtual memory.
273 * On the i386 this is called after mapping has already been enabled
274 * and just syncs the pmap module with what has already been done.
275 * [We can't call it easily with mapping off since the kernel is not
276 * mapped with PA == VA, hence we would have to relocate every address
277 * from the linked base (virtual) address "KERNBASE" to the actual
278 * (physical) address starting relative to 0]
281 pmap_bootstrap(firstaddr, loadaddr)
282 vm_offset_t firstaddr;
283 vm_offset_t loadaddr;
288 struct globaldata *gd;
292 avail_start = firstaddr;
295 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too
296 * large. It should instead be correctly calculated in locore.s and
297 * not based on 'first' (which is a physical address, not a virtual
298 * address, for the start of unused physical memory). The kernel
299 * page tables are NOT double mapped and thus should not be included
300 * in this calculation.
302 virtual_avail = (vm_offset_t) KERNBASE + firstaddr;
303 virtual_avail = pmap_kmem_choose(virtual_avail);
305 virtual_end = VM_MAX_KERNEL_ADDRESS;
308 * Initialize protection array.
310 i386_protection_init();
313 * The kernel's pmap is statically allocated so we don't have to use
314 * pmap_create, which is unlikely to work correctly at this part of
315 * the boot sequence (XXX and which no longer exists).
317 kernel_pmap = &kernel_pmap_store;
319 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD);
320 kernel_pmap->pm_count = 1;
321 kernel_pmap->pm_active = -1; /* don't allow deactivation */
322 TAILQ_INIT(&kernel_pmap->pm_pvlist);
326 * Reserve some special page table entries/VA space for temporary
329 #define SYSMAP(c, p, v, n) \
330 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
333 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va);
336 * CMAP1/CMAP2 are used for zeroing and copying pages.
338 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
339 SYSMAP(caddr_t, CMAP2, CADDR2, 1)
344 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
347 * ptvmmap is used for reading arbitrary physical pages via /dev/mem.
348 * XXX ptmmap is not used.
350 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
353 * msgbufp is used to map the system message buffer.
354 * XXX msgbufmap is not used.
356 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
357 atop(round_page(MSGBUF_SIZE)))
361 * ptemap is used for pmap_pte_quick
363 SYSMAP(unsigned *, PMAP1, PADDR1, 1);
368 *(int *) CMAP1 = *(int *) CMAP2 = 0;
369 for (i = 0; i < NKPT; i++)
373 #if !defined(SMP) /* XXX - see also mp_machdep.c */
374 if (cpu_feature & CPUID_PGE) {
380 * Initialize the 4MB page size flag
384 * The 4MB page version of the initial
385 * kernel page mapping.
389 #if !defined(DISABLE_PSE)
390 if (cpu_feature & CPUID_PSE) {
393 * Note that we have enabled PSE mode
396 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
397 ptditmp &= ~(NBPDR - 1);
398 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
403 * Enable the PSE mode.
405 load_cr4(rcr4() | CR4_PSE);
408 * We can do the mapping here for the single processor
409 * case. We simply ignore the old page table page from
413 * For SMP, we still need 4K pages to bootstrap APs,
414 * PSE will be enabled as soon as all APs are up.
416 PTD[KPTDI] = (pd_entry_t) ptditmp;
417 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp;
424 if (cpu_apic_address == 0)
425 panic("pmap_bootstrap: no local apic!");
427 /* local apic is mapped on last page */
428 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
429 (cpu_apic_address & PG_FRAME));
431 /* BSP does this itself, AP's get it pre-set */
432 gd = &SMP_prvspace[0].globaldata;
433 gd->gd_prv_CMAP1 = &SMPpt[1];
434 gd->gd_prv_CMAP2 = &SMPpt[2];
435 gd->gd_prv_CMAP3 = &SMPpt[3];
436 gd->gd_prv_PMAP1 = &SMPpt[4];
437 gd->gd_prv_CADDR1 = SMP_prvspace[0].CPAGE1;
438 gd->gd_prv_CADDR2 = SMP_prvspace[0].CPAGE2;
439 gd->gd_prv_CADDR3 = SMP_prvspace[0].CPAGE3;
440 gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[0].PPAGE1;
448 * Set 4mb pdir for mp startup
453 if (pseflag && (cpu_feature & CPUID_PSE)) {
454 load_cr4(rcr4() | CR4_PSE);
455 if (pdir4mb && cpuid == 0) { /* only on BSP */
456 kernel_pmap->pm_pdir[KPTDI] =
457 PTD[KPTDI] = (pd_entry_t)pdir4mb;
465 * Initialize the pmap module.
466 * Called by vm_init, to initialize any structures that the pmap
467 * system needs to map virtual memory.
468 * pmap_init has been enhanced to support in a fairly consistant
469 * way, discontiguous physical memory.
472 pmap_init(phys_start, phys_end)
473 vm_offset_t phys_start, phys_end;
479 * object for kernel page table pages
481 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
484 * Allocate memory for random pmap data structures. Includes the
488 for(i = 0; i < vm_page_array_size; i++) {
491 m = &vm_page_array[i];
492 TAILQ_INIT(&m->md.pv_list);
493 m->md.pv_list_count = 0;
497 * init the pv free list
499 initial_pvs = vm_page_array_size;
500 if (initial_pvs < MINPV)
502 pvzone = &pvzone_store;
503 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
504 initial_pvs * sizeof (struct pv_entry));
505 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
509 * Now it is safe to enable pv_table recording.
511 pmap_initialized = TRUE;
515 * Initialize the address space (zone) for the pv_entries. Set a
516 * high water mark so that the system can recover from excessive
517 * numbers of pv entries.
522 int shpgperproc = PMAP_SHPGPERPROC;
524 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
525 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
526 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
527 pv_entry_high_water = 9 * (pv_entry_max / 10);
528 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
532 /***************************************************
533 * Low level helper routines.....
534 ***************************************************/
536 #if defined(PMAP_DIAGNOSTIC)
539 * This code checks for non-writeable/modified pages.
540 * This should be an invalid condition.
543 pmap_nw_modified(pt_entry_t ptea)
549 if ((pte & (PG_M|PG_RW)) == PG_M)
558 * this routine defines the region(s) of memory that should
559 * not be tested for the modified bit.
561 static PMAP_INLINE int
562 pmap_track_modified(vm_offset_t va)
564 if ((va < clean_sva) || (va >= clean_eva))
570 static PMAP_INLINE void
571 invltlb_1pg(vm_offset_t va)
573 #if defined(I386_CPU)
574 if (cpu_class == CPUCLASS_386) {
584 pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
587 if (pmap->pm_active & (1 << cpuid))
588 cpu_invlpg((void *)va);
589 if (pmap->pm_active & other_cpus)
598 pmap_TLB_invalidate_all(pmap_t pmap)
601 if (pmap->pm_active & (1 << cpuid))
603 if (pmap->pm_active & other_cpus)
615 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
617 /* are we current address space or kernel? */
618 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
619 return (unsigned *) PTmap;
621 /* otherwise, we are alternate address space */
622 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
623 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
625 /* The page directory is not shared between CPUs */
631 return (unsigned *) APTmap;
635 * Super fast pmap_pte routine best used when scanning
636 * the pv lists. This eliminates many coarse-grained
637 * invltlb calls. Note that many of the pv list
638 * scans are across different pmaps. It is very wasteful
639 * to do an entire invltlb for checking a single mapping.
643 pmap_pte_quick(pmap, va)
644 register pmap_t pmap;
648 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
649 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
650 unsigned index = i386_btop(va);
651 /* are we current address space or kernel? */
652 if ((pmap == kernel_pmap) ||
653 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
654 return (unsigned *) PTmap + index;
656 newpf = pde & PG_FRAME;
658 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) {
659 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V;
660 cpu_invlpg(prv_PADDR1);
662 return prv_PADDR1 + ((unsigned) index & (NPTEPG - 1));
664 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) {
665 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V;
666 invltlb_1pg((vm_offset_t) PADDR1);
668 return PADDR1 + ((unsigned) index & (NPTEPG - 1));
675 * Routine: pmap_extract
677 * Extract the physical page address associated
678 * with the given map/virtual_address pair.
681 pmap_extract(pmap, va)
682 register pmap_t pmap;
686 vm_offset_t pdirindex;
687 pdirindex = va >> PDRSHIFT;
688 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
690 if ((rtval & PG_PS) != 0) {
691 rtval &= ~(NBPDR - 1);
692 rtval |= va & (NBPDR - 1);
695 pte = get_ptbase(pmap) + i386_btop(va);
696 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
703 /***************************************************
704 * Low level mapping routines.....
705 ***************************************************/
708 * add a wired page to the kva
709 * note that in order for the mapping to take effect -- you
710 * should do a invltlb after doing the pmap_kenter...
715 register vm_offset_t pa;
717 register unsigned *pte;
720 npte = pa | PG_RW | PG_V | pgeflag;
721 pte = (unsigned *)vtopte(va);
728 * remove a page from the kernel pagetables
734 register unsigned *pte;
736 pte = (unsigned *)vtopte(va);
742 * Used to map a range of physical addresses into kernel
743 * virtual address space.
745 * For now, VM is already on, we only need to map the
749 pmap_map(virt, start, end, prot)
755 while (start < end) {
756 pmap_kenter(virt, start);
765 * Add a list of wired pages to the kva
766 * this routine is only used for temporary
767 * kernel mappings that do not need to have
768 * page modification or references recorded.
769 * Note that old mappings are simply written
770 * over. The page *must* be wired.
773 pmap_qenter(va, m, count)
780 end_va = va + count * PAGE_SIZE;
782 while (va < end_va) {
785 pte = (unsigned *)vtopte(va);
786 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
788 cpu_invlpg((void *)va);
801 * this routine jerks page mappings from the
802 * kernel -- it is meant only for temporary mappings.
805 pmap_qremove(va, count)
811 end_va = va + count*PAGE_SIZE;
813 while (va < end_va) {
816 pte = (unsigned *)vtopte(va);
819 cpu_invlpg((void *)va);
831 pmap_page_lookup(object, pindex)
837 m = vm_page_lookup(object, pindex);
838 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
844 * Create the UPAGES for a new process.
845 * This routine directly affects the fork perf for a process.
855 unsigned *ptek, oldpte;
858 * allocate object for the upages
860 if ((upobj = p->p_upages_obj) == NULL) {
861 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES);
862 p->p_upages_obj = upobj;
865 /* get a kernel virtual address for the UPAGES for this proc */
866 if ((up = p->p_addr) == NULL) {
867 up = (struct user *) kmem_alloc_nofault(kernel_map,
870 panic("pmap_new_proc: u_map allocation failed");
874 ptek = (unsigned *) vtopte((vm_offset_t) up);
877 for(i=0;i<UPAGES;i++) {
879 * Get a kernel stack page
881 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
889 oldpte = *(ptek + i);
891 * Enter the page into the kernel address space.
893 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag;
895 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) {
896 invlpg((vm_offset_t) up + i * PAGE_SIZE);
903 vm_page_flag_clear(m, PG_ZERO);
904 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
905 m->valid = VM_PAGE_BITS_ALL;
912 * Dispose the UPAGES for a process that has exited.
913 * This routine directly impacts the exit perf of a process.
922 unsigned *ptek, oldpte;
924 upobj = p->p_upages_obj;
926 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr);
927 for(i=0;i<UPAGES;i++) {
929 if ((m = vm_page_lookup(upobj, i)) == NULL)
930 panic("pmap_dispose_proc: upage already missing???");
934 oldpte = *(ptek + i);
936 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386))
937 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE);
938 vm_page_unwire(m, 0);
941 #if defined(I386_CPU)
942 if (cpu_class <= CPUCLASS_386)
947 * If the process got swapped out some of its UPAGES might have gotten
948 * swapped. Just get rid of the object to clean up the swap use
949 * proactively. NOTE! might block waiting for paging I/O to complete.
951 if (upobj->type == OBJT_SWAP) {
952 p->p_upages_obj = NULL;
953 vm_object_deallocate(upobj);
958 * Allow the UPAGES for a process to be prejudicially paged out.
968 upobj = p->p_upages_obj;
970 * let the upages be paged
972 for(i=0;i<UPAGES;i++) {
973 if ((m = vm_page_lookup(upobj, i)) == NULL)
974 panic("pmap_swapout_proc: upage already missing???");
976 vm_page_unwire(m, 0);
977 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
982 * Bring the UPAGES for a specified process back in.
992 upobj = p->p_upages_obj;
993 for(i=0;i<UPAGES;i++) {
995 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
997 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
1000 if (m->valid != VM_PAGE_BITS_ALL) {
1001 rv = vm_pager_get_pages(upobj, &m, 1, 0);
1002 if (rv != VM_PAGER_OK)
1003 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
1004 m = vm_page_lookup(upobj, i);
1005 m->valid = VM_PAGE_BITS_ALL;
1010 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
1014 /***************************************************
1015 * Page table page management routines.....
1016 ***************************************************/
1019 * This routine unholds page table pages, and if the hold count
1020 * drops to zero, then it decrements the wire count.
1023 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
1025 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
1028 if (m->hold_count == 0) {
1031 * unmap the page table page
1033 pmap->pm_pdir[m->pindex] = 0;
1034 --pmap->pm_stats.resident_count;
1035 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1036 (((unsigned) PTDpde) & PG_FRAME)) {
1038 * Do a invltlb to make the invalidated mapping
1039 * take effect immediately.
1041 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
1042 pmap_TLB_invalidate(pmap, pteva);
1045 if (pmap->pm_ptphint == m)
1046 pmap->pm_ptphint = NULL;
1049 * If the page is finally unwired, simply free it.
1052 if (m->wire_count == 0) {
1056 vm_page_free_zero(m);
1064 static PMAP_INLINE int
1065 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
1068 if (m->hold_count == 0)
1069 return _pmap_unwire_pte_hold(pmap, m);
1075 * After removing a page table entry, this routine is used to
1076 * conditionally free the page, and manage the hold/wire counts.
1079 pmap_unuse_pt(pmap, va, mpte)
1085 if (va >= UPT_MIN_ADDRESS)
1089 ptepindex = (va >> PDRSHIFT);
1090 if (pmap->pm_ptphint &&
1091 (pmap->pm_ptphint->pindex == ptepindex)) {
1092 mpte = pmap->pm_ptphint;
1094 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1095 pmap->pm_ptphint = mpte;
1099 return pmap_unwire_pte_hold(pmap, mpte);
1107 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1108 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1110 pmap->pm_active = 0;
1111 pmap->pm_ptphint = NULL;
1112 TAILQ_INIT(&pmap->pm_pvlist);
1113 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1117 * Initialize a preallocated and zeroed pmap structure,
1118 * such as one in a vmspace structure.
1122 register struct pmap *pmap;
1127 * No need to allocate page table space yet but we do need a valid
1128 * page directory table.
1130 if (pmap->pm_pdir == NULL)
1132 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1135 * allocate object for the ptes
1137 if (pmap->pm_pteobj == NULL)
1138 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1141 * allocate the page directory page
1143 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1144 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1146 ptdpg->wire_count = 1;
1150 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1151 ptdpg->valid = VM_PAGE_BITS_ALL;
1153 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1154 if ((ptdpg->flags & PG_ZERO) == 0)
1155 bzero(pmap->pm_pdir, PAGE_SIZE);
1158 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1161 /* install self-referential address mapping entry */
1162 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1163 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1166 pmap->pm_active = 0;
1167 pmap->pm_ptphint = NULL;
1168 TAILQ_INIT(&pmap->pm_pvlist);
1169 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1173 * Wire in kernel global address entries. To avoid a race condition
1174 * between pmap initialization and pmap_growkernel, this procedure
1175 * should be called after the vmspace is attached to the process
1176 * but before this pmap is activated.
1182 /* XXX copies current process, does not fill in MPPTDI */
1183 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1187 pmap_release_free_page(pmap, p)
1191 unsigned *pde = (unsigned *) pmap->pm_pdir;
1193 * This code optimizes the case of freeing non-busy
1194 * page-table pages. Those pages are zero now, and
1195 * might as well be placed directly into the zero queue.
1197 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1203 * Remove the page table page from the processes address space.
1206 pmap->pm_stats.resident_count--;
1208 if (p->hold_count) {
1209 panic("pmap_release: freeing held page table page");
1212 * Page directory pages need to have the kernel
1213 * stuff cleared, so they can go into the zero queue also.
1215 if (p->pindex == PTDPTDI) {
1216 bzero(pde + KPTDI, nkpt * PTESIZE);
1221 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1224 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1225 pmap->pm_ptphint = NULL;
1229 vm_page_free_zero(p);
1234 * this routine is called if the page table page is not
1238 _pmap_allocpte(pmap, ptepindex)
1242 vm_offset_t pteva, ptepa;
1246 * Find or fabricate a new pagetable page
1248 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1249 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1251 KASSERT(m->queue == PQ_NONE,
1252 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1254 if (m->wire_count == 0)
1259 * Increment the hold count for the page table page
1260 * (denoting a new mapping.)
1265 * Map the pagetable page into the process address space, if
1266 * it isn't already there.
1269 pmap->pm_stats.resident_count++;
1271 ptepa = VM_PAGE_TO_PHYS(m);
1272 pmap->pm_pdir[ptepindex] =
1273 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1276 * Set the page table hint
1278 pmap->pm_ptphint = m;
1281 * Try to use the new mapping, but if we cannot, then
1282 * do it with the routine that maps the page explicitly.
1284 if ((m->flags & PG_ZERO) == 0) {
1285 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1286 (((unsigned) PTDpde) & PG_FRAME)) {
1287 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1288 bzero((caddr_t) pteva, PAGE_SIZE);
1290 pmap_zero_page(ptepa);
1294 m->valid = VM_PAGE_BITS_ALL;
1295 vm_page_flag_clear(m, PG_ZERO);
1296 vm_page_flag_set(m, PG_MAPPED);
1303 pmap_allocpte(pmap, va)
1312 * Calculate pagetable page index
1314 ptepindex = va >> PDRSHIFT;
1317 * Get the page directory entry
1319 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1322 * This supports switching from a 4MB page to a
1325 if (ptepa & PG_PS) {
1326 pmap->pm_pdir[ptepindex] = 0;
1332 * If the page table page is mapped, we just increment the
1333 * hold count, and activate it.
1337 * In order to get the page table page, try the
1340 if (pmap->pm_ptphint &&
1341 (pmap->pm_ptphint->pindex == ptepindex)) {
1342 m = pmap->pm_ptphint;
1344 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1345 pmap->pm_ptphint = m;
1351 * Here if the pte page isn't mapped, or if it has been deallocated.
1353 return _pmap_allocpte(pmap, ptepindex);
1357 /***************************************************
1358 * Pmap allocation/deallocation routines.
1359 ***************************************************/
1362 * Release any resources held by the given physical map.
1363 * Called when a pmap initialized by pmap_pinit is being released.
1364 * Should only be called if the map contains no valid mappings.
1368 register struct pmap *pmap;
1370 vm_page_t p,n,ptdpg;
1371 vm_object_t object = pmap->pm_pteobj;
1374 #if defined(DIAGNOSTIC)
1375 if (object->ref_count != 1)
1376 panic("pmap_release: pteobj reference count != 1");
1381 curgeneration = object->generation;
1382 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1383 n = TAILQ_NEXT(p, listq);
1384 if (p->pindex == PTDPTDI) {
1389 if (!pmap_release_free_page(pmap, p) &&
1390 (object->generation != curgeneration))
1395 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1400 kvm_size(SYSCTL_HANDLER_ARGS)
1402 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1404 return sysctl_handle_long(oidp, &ksize, 0, req);
1406 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1407 0, 0, kvm_size, "IU", "Size of KVM");
1410 kvm_free(SYSCTL_HANDLER_ARGS)
1412 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1414 return sysctl_handle_long(oidp, &kfree, 0, req);
1416 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1417 0, 0, kvm_free, "IU", "Amount of KVM free");
1420 * grow the number of kernel page table entries, if needed
1423 pmap_growkernel(vm_offset_t addr)
1428 vm_offset_t ptppaddr;
1433 if (kernel_vm_end == 0) {
1434 kernel_vm_end = KERNBASE;
1436 while (pdir_pde(PTD, kernel_vm_end)) {
1437 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1441 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1442 while (kernel_vm_end < addr) {
1443 if (pdir_pde(PTD, kernel_vm_end)) {
1444 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1449 * This index is bogus, but out of the way
1451 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1453 panic("pmap_growkernel: no memory to grow kernel");
1458 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1459 pmap_zero_page(ptppaddr);
1460 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1461 pdir_pde(PTD, kernel_vm_end) = newpdir;
1463 LIST_FOREACH(p, &allproc, p_list) {
1465 pmap = vmspace_pmap(p->p_vmspace);
1466 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1469 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1470 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1476 * Retire the given physical map from service.
1477 * Should only be called if the map contains
1478 * no valid mappings.
1482 register pmap_t pmap;
1489 count = --pmap->pm_count;
1492 panic("destroying a pmap is not yet implemented");
1497 * Add a reference to the specified pmap.
1500 pmap_reference(pmap)
1508 /***************************************************
1509 * page management routines.
1510 ***************************************************/
1513 * free the pv_entry back to the free list
1515 static PMAP_INLINE void
1524 * get a new pv_entry, allocating a block from the system
1526 * the memory allocation is performed bypassing the malloc code
1527 * because of the possibility of allocations at interrupt time.
1533 if (pv_entry_high_water &&
1534 (pv_entry_count > pv_entry_high_water) &&
1535 (pmap_pagedaemon_waken == 0)) {
1536 pmap_pagedaemon_waken = 1;
1537 wakeup (&vm_pages_needed);
1539 return zalloci(pvzone);
1543 * This routine is very drastic, but can save the system
1551 static int warningdone=0;
1553 if (pmap_pagedaemon_waken == 0)
1556 if (warningdone < 5) {
1557 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1561 for(i = 0; i < vm_page_array_size; i++) {
1562 m = &vm_page_array[i];
1563 if (m->wire_count || m->hold_count || m->busy ||
1564 (m->flags & PG_BUSY))
1568 pmap_pagedaemon_waken = 0;
1573 * If it is the first entry on the list, it is actually
1574 * in the header and we must copy the following entry up
1575 * to the header. Otherwise we must search the list for
1576 * the entry. In either case we free the now unused entry.
1580 pmap_remove_entry(pmap, m, va)
1590 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1591 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1592 if (pmap == pv->pv_pmap && va == pv->pv_va)
1596 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1597 if (va == pv->pv_va)
1605 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1606 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1607 m->md.pv_list_count--;
1608 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1609 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1611 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1620 * Create a pv entry for page at pa for
1624 pmap_insert_entry(pmap, va, mpte, m)
1635 pv = get_pv_entry();
1640 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1641 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1642 m->md.pv_list_count++;
1648 * pmap_remove_pte: do the things to unmap a page in a process
1651 pmap_remove_pte(pmap, ptq, va)
1659 oldpte = loadandclear(ptq);
1661 pmap->pm_stats.wired_count -= 1;
1663 * Machines that don't support invlpg, also don't support
1668 pmap->pm_stats.resident_count -= 1;
1669 if (oldpte & PG_MANAGED) {
1670 m = PHYS_TO_VM_PAGE(oldpte);
1671 if (oldpte & PG_M) {
1672 #if defined(PMAP_DIAGNOSTIC)
1673 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1675 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1679 if (pmap_track_modified(va))
1683 vm_page_flag_set(m, PG_REFERENCED);
1684 return pmap_remove_entry(pmap, m, va);
1686 return pmap_unuse_pt(pmap, va, NULL);
1693 * Remove a single page from a process address space
1696 pmap_remove_page(pmap, va)
1698 register vm_offset_t va;
1700 register unsigned *ptq;
1703 * if there is no pte for this address, just skip it!!!
1705 if (*pmap_pde(pmap, va) == 0) {
1710 * get a local va for mappings for this pmap.
1712 ptq = get_ptbase(pmap) + i386_btop(va);
1714 (void) pmap_remove_pte(pmap, ptq, va);
1715 pmap_TLB_invalidate(pmap, va);
1721 * Remove the given range of addresses from the specified map.
1723 * It is assumed that the start and end are properly
1724 * rounded to the page size.
1727 pmap_remove(pmap, sva, eva)
1729 register vm_offset_t sva;
1730 register vm_offset_t eva;
1732 register unsigned *ptbase;
1734 vm_offset_t ptpaddr;
1735 vm_offset_t sindex, eindex;
1741 if (pmap->pm_stats.resident_count == 0)
1745 * special handling of removing one page. a very
1746 * common operation and easy to short circuit some
1749 if (((sva + PAGE_SIZE) == eva) &&
1750 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1751 pmap_remove_page(pmap, sva);
1758 * Get a local virtual address for the mappings that are being
1761 ptbase = get_ptbase(pmap);
1763 sindex = i386_btop(sva);
1764 eindex = i386_btop(eva);
1766 for (; sindex < eindex; sindex = pdnxt) {
1770 * Calculate index for next page table.
1772 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1773 if (pmap->pm_stats.resident_count == 0)
1776 pdirindex = sindex / NPDEPG;
1777 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1778 pmap->pm_pdir[pdirindex] = 0;
1779 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1785 * Weed out invalid mappings. Note: we assume that the page
1786 * directory table is always allocated, and in kernel virtual.
1792 * Limit our scan to either the end of the va represented
1793 * by the current page table page, or to the end of the
1794 * range being removed.
1796 if (pdnxt > eindex) {
1800 for ( ;sindex != pdnxt; sindex++) {
1802 if (ptbase[sindex] == 0) {
1805 va = i386_ptob(sindex);
1808 if (pmap_remove_pte(pmap,
1809 ptbase + sindex, va))
1815 pmap_TLB_invalidate_all(pmap);
1819 * Routine: pmap_remove_all
1821 * Removes this physical page from
1822 * all physical maps in which it resides.
1823 * Reflects back modify bits to the pager.
1826 * Original versions of this routine were very
1827 * inefficient because they iteratively called
1828 * pmap_remove (slow...)
1835 register pv_entry_t pv;
1836 register unsigned *pte, tpte;
1839 #if defined(PMAP_DIAGNOSTIC)
1841 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1844 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1845 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1850 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1851 pv->pv_pmap->pm_stats.resident_count--;
1853 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1855 tpte = loadandclear(pte);
1857 pv->pv_pmap->pm_stats.wired_count--;
1860 vm_page_flag_set(m, PG_REFERENCED);
1863 * Update the vm_page_t clean and reference bits.
1866 #if defined(PMAP_DIAGNOSTIC)
1867 if (pmap_nw_modified((pt_entry_t) tpte)) {
1869 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1873 if (pmap_track_modified(pv->pv_va))
1876 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1878 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1879 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1880 m->md.pv_list_count--;
1881 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1885 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1891 * Set the physical protection on the
1892 * specified range of this map as requested.
1895 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1897 register unsigned *ptbase;
1898 vm_offset_t pdnxt, ptpaddr;
1899 vm_pindex_t sindex, eindex;
1905 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1906 pmap_remove(pmap, sva, eva);
1910 if (prot & VM_PROT_WRITE)
1915 ptbase = get_ptbase(pmap);
1917 sindex = i386_btop(sva);
1918 eindex = i386_btop(eva);
1920 for (; sindex < eindex; sindex = pdnxt) {
1924 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1926 pdirindex = sindex / NPDEPG;
1927 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1928 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1929 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1935 * Weed out invalid mappings. Note: we assume that the page
1936 * directory table is always allocated, and in kernel virtual.
1941 if (pdnxt > eindex) {
1945 for (; sindex != pdnxt; sindex++) {
1950 pbits = ptbase[sindex];
1952 if (pbits & PG_MANAGED) {
1955 m = PHYS_TO_VM_PAGE(pbits);
1956 vm_page_flag_set(m, PG_REFERENCED);
1960 if (pmap_track_modified(i386_ptob(sindex))) {
1962 m = PHYS_TO_VM_PAGE(pbits);
1971 if (pbits != ptbase[sindex]) {
1972 ptbase[sindex] = pbits;
1978 pmap_TLB_invalidate_all(pmap);
1982 * Insert the given physical page (p) at
1983 * the specified virtual address (v) in the
1984 * target physical map with the protection requested.
1986 * If specified, the page will be wired down, meaning
1987 * that the related pte can not be reclaimed.
1989 * NB: This is the only routine which MAY NOT lazy-evaluate
1990 * or lose information. That is, this routine must actually
1991 * insert this page into the given map NOW.
1994 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1998 register unsigned *pte;
2000 vm_offset_t origpte, newpte;
2007 #ifdef PMAP_DIAGNOSTIC
2008 if (va > VM_MAX_KERNEL_ADDRESS)
2009 panic("pmap_enter: toobig");
2010 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
2011 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
2016 * In the case that a page table page is not
2017 * resident, we are creating it here.
2019 if (va < UPT_MIN_ADDRESS) {
2020 mpte = pmap_allocpte(pmap, va);
2022 #if 0 && defined(PMAP_DIAGNOSTIC)
2024 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
2025 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
2026 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
2027 pmap->pm_pdir[PTDPTDI], origpte, va);
2030 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
2031 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
2032 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
2033 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
2034 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
2035 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
2036 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
2042 pte = pmap_pte(pmap, va);
2045 * Page Directory table entry not valid, we need a new PT page
2048 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
2049 (void *)pmap->pm_pdir[PTDPTDI], va);
2052 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2053 origpte = *(vm_offset_t *)pte;
2054 opa = origpte & PG_FRAME;
2056 if (origpte & PG_PS)
2057 panic("pmap_enter: attempted pmap_enter on 4MB page");
2060 * Mapping has not changed, must be protection or wiring change.
2062 if (origpte && (opa == pa)) {
2064 * Wiring change, just update stats. We don't worry about
2065 * wiring PT pages as they remain resident as long as there
2066 * are valid mappings in them. Hence, if a user page is wired,
2067 * the PT page will be also.
2069 if (wired && ((origpte & PG_W) == 0))
2070 pmap->pm_stats.wired_count++;
2071 else if (!wired && (origpte & PG_W))
2072 pmap->pm_stats.wired_count--;
2074 #if defined(PMAP_DIAGNOSTIC)
2075 if (pmap_nw_modified((pt_entry_t) origpte)) {
2077 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
2083 * Remove extra pte reference
2088 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
2089 if ((origpte & PG_RW) == 0) {
2092 cpu_invlpg((void *)va);
2093 if (pmap->pm_active & other_cpus)
2103 * We might be turning off write access to the page,
2104 * so we go ahead and sense modify status.
2106 if (origpte & PG_MANAGED) {
2107 if ((origpte & PG_M) && pmap_track_modified(va)) {
2109 om = PHYS_TO_VM_PAGE(opa);
2117 * Mapping has changed, invalidate old range and fall through to
2118 * handle validating new mapping.
2122 err = pmap_remove_pte(pmap, pte, va);
2124 panic("pmap_enter: pte vanished, va: 0x%x", va);
2128 * Enter on the PV list if part of our managed memory. Note that we
2129 * raise IPL while manipulating pv_table since pmap_enter can be
2130 * called at interrupt time.
2132 if (pmap_initialized &&
2133 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2134 pmap_insert_entry(pmap, va, mpte, m);
2139 * Increment counters
2141 pmap->pm_stats.resident_count++;
2143 pmap->pm_stats.wired_count++;
2147 * Now validate mapping with desired protection/wiring.
2149 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2153 if (va < UPT_MIN_ADDRESS)
2155 if (pmap == kernel_pmap)
2159 * if the mapping or permission bits are different, we need
2160 * to update the pte.
2162 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2163 *pte = newpte | PG_A;
2166 cpu_invlpg((void *)va);
2167 if (pmap->pm_active & other_cpus)
2177 * this code makes some *MAJOR* assumptions:
2178 * 1. Current pmap & pmap exists.
2181 * 4. No page table pages.
2182 * 5. Tlbflush is deferred to calling procedure.
2183 * 6. Page IS managed.
2184 * but is *MUCH* faster than pmap_enter...
2188 pmap_enter_quick(pmap, va, m, mpte)
2189 register pmap_t pmap;
2198 * In the case that a page table page is not
2199 * resident, we are creating it here.
2201 if (va < UPT_MIN_ADDRESS) {
2206 * Calculate pagetable page index
2208 ptepindex = va >> PDRSHIFT;
2209 if (mpte && (mpte->pindex == ptepindex)) {
2214 * Get the page directory entry
2216 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2219 * If the page table page is mapped, we just increment
2220 * the hold count, and activate it.
2224 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2225 if (pmap->pm_ptphint &&
2226 (pmap->pm_ptphint->pindex == ptepindex)) {
2227 mpte = pmap->pm_ptphint;
2229 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2230 pmap->pm_ptphint = mpte;
2236 mpte = _pmap_allocpte(pmap, ptepindex);
2244 * This call to vtopte makes the assumption that we are
2245 * entering the page into the current pmap. In order to support
2246 * quick entry into any pmap, one would likely use pmap_pte_quick.
2247 * But that isn't as quick as vtopte.
2249 pte = (unsigned *)vtopte(va);
2252 pmap_unwire_pte_hold(pmap, mpte);
2257 * Enter on the PV list if part of our managed memory. Note that we
2258 * raise IPL while manipulating pv_table since pmap_enter can be
2259 * called at interrupt time.
2261 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2262 pmap_insert_entry(pmap, va, mpte, m);
2265 * Increment counters
2267 pmap->pm_stats.resident_count++;
2269 pa = VM_PAGE_TO_PHYS(m);
2272 * Now validate mapping with RO protection
2274 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2275 *pte = pa | PG_V | PG_U;
2277 *pte = pa | PG_V | PG_U | PG_MANAGED;
2283 * Make a temporary mapping for a physical address. This is only intended
2284 * to be used for panic dumps.
2287 pmap_kenter_temporary(vm_offset_t pa, int i)
2289 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2290 return ((void *)crashdumpmap);
2293 #define MAX_INIT_PT (96)
2295 * pmap_object_init_pt preloads the ptes for a given object
2296 * into the specified pmap. This eliminates the blast of soft
2297 * faults on process startup and immediately after an mmap.
2300 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2313 if (pmap == NULL || object == NULL)
2317 * This code maps large physical mmap regions into the
2318 * processor address space. Note that some shortcuts
2319 * are taken, but the code works.
2322 (object->type == OBJT_DEVICE) &&
2323 ((addr & (NBPDR - 1)) == 0) &&
2324 ((size & (NBPDR - 1)) == 0) ) {
2327 unsigned int ptepindex;
2331 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2335 p = vm_page_lookup(object, pindex);
2336 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2340 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2345 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2350 p = vm_page_lookup(object, pindex);
2354 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2355 if (ptepa & (NBPDR - 1)) {
2359 p->valid = VM_PAGE_BITS_ALL;
2361 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2362 npdes = size >> PDRSHIFT;
2363 for(i=0;i<npdes;i++) {
2364 pmap->pm_pdir[ptepindex] =
2365 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2369 vm_page_flag_set(p, PG_MAPPED);
2374 psize = i386_btop(size);
2376 if ((object->type != OBJT_VNODE) ||
2377 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2378 (object->resident_page_count > MAX_INIT_PT))) {
2382 if (psize + pindex > object->size) {
2383 if (object->size < pindex)
2385 psize = object->size - pindex;
2390 * if we are processing a major portion of the object, then scan the
2393 if (psize > (object->resident_page_count >> 2)) {
2396 for (p = TAILQ_FIRST(&object->memq);
2397 ((objpgs > 0) && (p != NULL));
2398 p = TAILQ_NEXT(p, listq)) {
2401 if (tmpidx < pindex) {
2405 if (tmpidx >= psize) {
2409 * don't allow an madvise to blow away our really
2410 * free pages allocating pv entries.
2412 if ((limit & MAP_PREFAULT_MADVISE) &&
2413 cnt.v_free_count < cnt.v_free_reserved) {
2416 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2418 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2419 if ((p->queue - p->pc) == PQ_CACHE)
2420 vm_page_deactivate(p);
2422 mpte = pmap_enter_quick(pmap,
2423 addr + i386_ptob(tmpidx), p, mpte);
2424 vm_page_flag_set(p, PG_MAPPED);
2431 * else lookup the pages one-by-one.
2433 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2435 * don't allow an madvise to blow away our really
2436 * free pages allocating pv entries.
2438 if ((limit & MAP_PREFAULT_MADVISE) &&
2439 cnt.v_free_count < cnt.v_free_reserved) {
2442 p = vm_page_lookup(object, tmpidx + pindex);
2444 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2446 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2447 if ((p->queue - p->pc) == PQ_CACHE)
2448 vm_page_deactivate(p);
2450 mpte = pmap_enter_quick(pmap,
2451 addr + i386_ptob(tmpidx), p, mpte);
2452 vm_page_flag_set(p, PG_MAPPED);
2461 * pmap_prefault provides a quick way of clustering
2462 * pagefaults into a processes address space. It is a "cousin"
2463 * of pmap_object_init_pt, except it runs at page fault time instead
2468 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2470 static int pmap_prefault_pageorder[] = {
2471 -PAGE_SIZE, PAGE_SIZE,
2472 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2473 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2474 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2478 pmap_prefault(pmap, addra, entry)
2481 vm_map_entry_t entry;
2490 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2493 object = entry->object.vm_object;
2495 starta = addra - PFBAK * PAGE_SIZE;
2496 if (starta < entry->start) {
2497 starta = entry->start;
2498 } else if (starta > addra) {
2503 for (i = 0; i < PAGEORDER_SIZE; i++) {
2504 vm_object_t lobject;
2507 addr = addra + pmap_prefault_pageorder[i];
2508 if (addr > addra + (PFFOR * PAGE_SIZE))
2511 if (addr < starta || addr >= entry->end)
2514 if ((*pmap_pde(pmap, addr)) == NULL)
2517 pte = (unsigned *) vtopte(addr);
2521 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2523 for (m = vm_page_lookup(lobject, pindex);
2524 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2525 lobject = lobject->backing_object) {
2526 if (lobject->backing_object_offset & PAGE_MASK)
2528 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2529 m = vm_page_lookup(lobject->backing_object, pindex);
2533 * give-up when a page is not in memory
2538 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2540 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2542 if ((m->queue - m->pc) == PQ_CACHE) {
2543 vm_page_deactivate(m);
2546 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2547 vm_page_flag_set(m, PG_MAPPED);
2554 * Routine: pmap_change_wiring
2555 * Function: Change the wiring attribute for a map/virtual-address
2557 * In/out conditions:
2558 * The mapping must already exist in the pmap.
2561 pmap_change_wiring(pmap, va, wired)
2562 register pmap_t pmap;
2566 register unsigned *pte;
2571 pte = pmap_pte(pmap, va);
2573 if (wired && !pmap_pte_w(pte))
2574 pmap->pm_stats.wired_count++;
2575 else if (!wired && pmap_pte_w(pte))
2576 pmap->pm_stats.wired_count--;
2579 * Wiring is not a hardware characteristic so there is no need to
2582 pmap_pte_set_w(pte, wired);
2588 * Copy the range specified by src_addr/len
2589 * from the source map to the range dst_addr/len
2590 * in the destination map.
2592 * This routine is only advisory and need not do anything.
2596 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2597 pmap_t dst_pmap, src_pmap;
2598 vm_offset_t dst_addr;
2600 vm_offset_t src_addr;
2603 vm_offset_t end_addr = src_addr + len;
2605 unsigned src_frame, dst_frame;
2608 if (dst_addr != src_addr)
2611 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2612 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2616 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2617 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2618 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2620 /* The page directory is not shared between CPUs */
2627 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2628 unsigned *src_pte, *dst_pte;
2629 vm_page_t dstmpte, srcmpte;
2630 vm_offset_t srcptepaddr;
2633 if (addr >= UPT_MIN_ADDRESS)
2634 panic("pmap_copy: invalid to pmap_copy page tables\n");
2637 * Don't let optional prefaulting of pages make us go
2638 * way below the low water mark of free pages or way
2639 * above high water mark of used pv entries.
2641 if (cnt.v_free_count < cnt.v_free_reserved ||
2642 pv_entry_count > pv_entry_high_water)
2645 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2646 ptepindex = addr >> PDRSHIFT;
2648 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2649 if (srcptepaddr == 0)
2652 if (srcptepaddr & PG_PS) {
2653 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2654 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2655 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2660 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2661 if ((srcmpte == NULL) ||
2662 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2665 if (pdnxt > end_addr)
2668 src_pte = (unsigned *) vtopte(addr);
2669 dst_pte = (unsigned *) avtopte(addr);
2670 while (addr < pdnxt) {
2674 * we only virtual copy managed pages
2676 if ((ptetemp & PG_MANAGED) != 0) {
2678 * We have to check after allocpte for the
2679 * pte still being around... allocpte can
2682 dstmpte = pmap_allocpte(dst_pmap, addr);
2683 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2685 * Clear the modified and
2686 * accessed (referenced) bits
2689 m = PHYS_TO_VM_PAGE(ptetemp);
2690 *dst_pte = ptetemp & ~(PG_M | PG_A);
2691 dst_pmap->pm_stats.resident_count++;
2692 pmap_insert_entry(dst_pmap, addr,
2695 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2697 if (dstmpte->hold_count >= srcmpte->hold_count)
2708 * Routine: pmap_kernel
2710 * Returns the physical map handle for the kernel.
2715 return (kernel_pmap);
2719 * pmap_zero_page zeros the specified hardware page by mapping
2720 * the page into KVM and using bzero to clear its contents.
2723 pmap_zero_page(phys)
2727 if (*(int *) prv_CMAP3)
2728 panic("pmap_zero_page: prv_CMAP3 busy");
2730 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2731 cpu_invlpg(prv_CADDR3);
2733 #if defined(I686_CPU)
2734 if (cpu_class == CPUCLASS_686)
2735 i686_pagezero(prv_CADDR3);
2738 bzero(prv_CADDR3, PAGE_SIZE);
2740 *(int *) prv_CMAP3 = 0;
2743 panic("pmap_zero_page: CMAP2 busy");
2745 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2746 invltlb_1pg((vm_offset_t)CADDR2);
2748 #if defined(I686_CPU)
2749 if (cpu_class == CPUCLASS_686)
2750 i686_pagezero(CADDR2);
2753 bzero(CADDR2, PAGE_SIZE);
2759 * pmap_zero_page_area zeros the specified hardware page by mapping
2760 * the page into KVM and using bzero to clear its contents.
2762 * off and size may not cover an area beyond a single hardware page.
2765 pmap_zero_page_area(phys, off, size)
2771 if (*(int *) prv_CMAP3)
2772 panic("pmap_zero_page: prv_CMAP3 busy");
2774 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2775 cpu_invlpg(prv_CADDR3);
2777 #if defined(I686_CPU)
2778 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2779 i686_pagezero(prv_CADDR3);
2782 bzero((char *)prv_CADDR3 + off, size);
2784 *(int *) prv_CMAP3 = 0;
2787 panic("pmap_zero_page: CMAP2 busy");
2789 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2790 invltlb_1pg((vm_offset_t)CADDR2);
2792 #if defined(I686_CPU)
2793 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2794 i686_pagezero(CADDR2);
2797 bzero((char *)CADDR2 + off, size);
2803 * pmap_copy_page copies the specified (machine independent)
2804 * page by mapping the page into virtual memory and using
2805 * bcopy to copy the page, one machine dependent page at a
2809 pmap_copy_page(src, dst)
2814 if (*(int *) prv_CMAP1)
2815 panic("pmap_copy_page: prv_CMAP1 busy");
2816 if (*(int *) prv_CMAP2)
2817 panic("pmap_copy_page: prv_CMAP2 busy");
2819 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2820 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2822 cpu_invlpg(prv_CADDR1);
2823 cpu_invlpg(prv_CADDR2);
2825 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE);
2827 *(int *) prv_CMAP1 = 0;
2828 *(int *) prv_CMAP2 = 0;
2830 if (*(int *) CMAP1 || *(int *) CMAP2)
2831 panic("pmap_copy_page: CMAP busy");
2833 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2834 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2835 #if defined(I386_CPU)
2836 if (cpu_class == CPUCLASS_386) {
2841 invlpg((u_int)CADDR1);
2842 invlpg((u_int)CADDR2);
2845 bcopy(CADDR1, CADDR2, PAGE_SIZE);
2854 * Routine: pmap_pageable
2856 * Make the specified pages (by pmap, offset)
2857 * pageable (or not) as requested.
2859 * A page which is not pageable may not take
2860 * a fault; therefore, its page table entry
2861 * must remain valid for the duration.
2863 * This routine is merely advisory; pmap_enter
2864 * will specify that these pages are to be wired
2865 * down (or not) as appropriate.
2868 pmap_pageable(pmap, sva, eva, pageable)
2870 vm_offset_t sva, eva;
2876 * Returns true if the pmap's pv is one of the first
2877 * 16 pvs linked to from this page. This count may
2878 * be changed upwards or downwards in the future; it
2879 * is only necessary that true be returned for a small
2880 * subset of pmaps for proper page aging.
2883 pmap_page_exists_quick(pmap, m)
2891 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2896 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2897 if (pv->pv_pmap == pmap) {
2909 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2911 * Remove all pages from specified address space
2912 * this aids process exit speeds. Also, this code
2913 * is special cased for current process only, but
2914 * can have the more generic (and slightly slower)
2915 * mode enabled. This is much faster than pmap_remove
2916 * in the case of running down an entire address space.
2919 pmap_remove_pages(pmap, sva, eva)
2921 vm_offset_t sva, eva;
2923 unsigned *pte, tpte;
2928 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2929 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2930 printf("warning: pmap_remove_pages called with non-current pmap\n");
2936 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2940 if (pv->pv_va >= eva || pv->pv_va < sva) {
2941 npv = TAILQ_NEXT(pv, pv_plist);
2945 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2946 pte = (unsigned *)vtopte(pv->pv_va);
2948 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2953 * We cannot remove wired pages from a process' mapping at this time
2956 npv = TAILQ_NEXT(pv, pv_plist);
2961 m = PHYS_TO_VM_PAGE(tpte);
2963 KASSERT(m < &vm_page_array[vm_page_array_size],
2964 ("pmap_remove_pages: bad tpte %x", tpte));
2966 pv->pv_pmap->pm_stats.resident_count--;
2969 * Update the vm_page_t clean and reference bits.
2976 npv = TAILQ_NEXT(pv, pv_plist);
2977 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2979 m->md.pv_list_count--;
2980 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2981 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2982 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2985 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2989 pmap_TLB_invalidate_all(pmap);
2993 * pmap_testbit tests bits in pte's
2994 * note that the testbit/changebit routines are inline,
2995 * and a lot of things compile-time evaluate.
2998 pmap_testbit(m, bit)
3006 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3009 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3014 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3016 * if the bit being tested is the modified bit, then
3017 * mark clean_map and ptes as never
3020 if (bit & (PG_A|PG_M)) {
3021 if (!pmap_track_modified(pv->pv_va))
3025 #if defined(PMAP_DIAGNOSTIC)
3027 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
3031 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3042 * this routine is used to modify bits in ptes
3044 static __inline void
3045 pmap_changebit(m, bit, setem)
3050 register pv_entry_t pv;
3051 register unsigned *pte;
3054 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3060 * Loop over all current mappings setting/clearing as appropos If
3061 * setting RO do we need to clear the VAC?
3063 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3065 * don't write protect pager mappings
3067 if (!setem && (bit == PG_RW)) {
3068 if (!pmap_track_modified(pv->pv_va))
3072 #if defined(PMAP_DIAGNOSTIC)
3074 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
3079 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3083 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3085 vm_offset_t pbits = *(vm_offset_t *)pte;
3091 *(int *)pte = pbits & ~(PG_M|PG_RW);
3093 *(int *)pte = pbits & ~bit;
3095 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3103 * pmap_page_protect:
3105 * Lower the permission for all mappings to a given page.
3108 pmap_page_protect(vm_page_t m, vm_prot_t prot)
3110 if ((prot & VM_PROT_WRITE) == 0) {
3111 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3112 pmap_changebit(m, PG_RW, FALSE);
3120 pmap_phys_address(ppn)
3123 return (i386_ptob(ppn));
3127 * pmap_ts_referenced:
3129 * Return a count of reference bits for a page, clearing those bits.
3130 * It is not necessary for every reference bit to be cleared, but it
3131 * is necessary that 0 only be returned when there are truly no
3132 * reference bits set.
3134 * XXX: The exact number of bits to check and clear is a matter that
3135 * should be tested and standardized at some point in the future for
3136 * optimal aging of shared pages.
3139 pmap_ts_referenced(vm_page_t m)
3141 register pv_entry_t pv, pvf, pvn;
3146 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3151 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3156 pvn = TAILQ_NEXT(pv, pv_list);
3158 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3160 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3162 if (!pmap_track_modified(pv->pv_va))
3165 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3167 if (pte && (*pte & PG_A)) {
3170 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3177 } while ((pv = pvn) != NULL && pv != pvf);
3187 * Return whether or not the specified physical page was modified
3188 * in any physical maps.
3191 pmap_is_modified(vm_page_t m)
3193 return pmap_testbit(m, PG_M);
3197 * Clear the modify bits on the specified physical page.
3200 pmap_clear_modify(vm_page_t m)
3202 pmap_changebit(m, PG_M, FALSE);
3206 * pmap_clear_reference:
3208 * Clear the reference bit on the specified physical page.
3211 pmap_clear_reference(vm_page_t m)
3213 pmap_changebit(m, PG_A, FALSE);
3217 * Miscellaneous support routines follow
3221 i386_protection_init()
3223 register int *kp, prot;
3225 kp = protection_codes;
3226 for (prot = 0; prot < 8; prot++) {
3228 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3230 * Read access is also 0. There isn't any execute bit,
3231 * so just make it readable.
3233 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3234 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3235 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3238 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3239 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3240 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3241 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3249 * Map a set of physical memory pages into the kernel virtual
3250 * address space. Return a pointer to where it is mapped. This
3251 * routine is intended to be used for mapping device memory,
3255 pmap_mapdev(pa, size)
3259 vm_offset_t va, tmpva, offset;
3262 offset = pa & PAGE_MASK;
3263 size = roundup(offset + size, PAGE_SIZE);
3265 va = kmem_alloc_pageable(kernel_map, size);
3267 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3270 for (tmpva = va; size > 0;) {
3271 pte = (unsigned *)vtopte(tmpva);
3272 *pte = pa | PG_RW | PG_V | pgeflag;
3279 return ((void *)(va + offset));
3283 pmap_unmapdev(va, size)
3287 vm_offset_t base, offset;
3289 base = va & PG_FRAME;
3290 offset = va & PAGE_MASK;
3291 size = roundup(offset + size, PAGE_SIZE);
3292 kmem_free(kernel_map, base, size);
3296 * perform the pmap work for mincore
3299 pmap_mincore(pmap, addr)
3304 unsigned *ptep, pte;
3308 ptep = pmap_pte(pmap, addr);
3313 if ((pte = *ptep) != 0) {
3316 val = MINCORE_INCORE;
3317 if ((pte & PG_MANAGED) == 0)
3320 pa = pte & PG_FRAME;
3322 m = PHYS_TO_VM_PAGE(pa);
3328 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3330 * Modified by someone
3332 else if (m->dirty || pmap_is_modified(m))
3333 val |= MINCORE_MODIFIED_OTHER;
3338 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3341 * Referenced by someone
3343 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3344 val |= MINCORE_REFERENCED_OTHER;
3345 vm_page_flag_set(m, PG_REFERENCED);
3352 pmap_activate(struct proc *p)
3356 pmap = vmspace_pmap(p->p_vmspace);
3358 pmap->pm_active |= 1 << cpuid;
3360 pmap->pm_active |= 1;
3362 #if defined(SWTCH_OPTIM_STATS)
3365 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir));
3369 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3372 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3376 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3381 #if defined(PMAP_DEBUG)
3382 pmap_pid_dump(int pid)
3388 LIST_FOREACH(p, &allproc, p_list) {
3389 if (p->p_pid != pid)
3395 pmap = vmspace_pmap(p->p_vmspace);
3396 for(i=0;i<1024;i++) {
3399 unsigned base = i << PDRSHIFT;
3401 pde = &pmap->pm_pdir[i];
3402 if (pde && pmap_pde_v(pde)) {
3403 for(j=0;j<1024;j++) {
3404 unsigned va = base + (j << PAGE_SHIFT);
3405 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3412 pte = pmap_pte_quick( pmap, va);
3413 if (pte && pmap_pte_v(pte)) {
3417 m = PHYS_TO_VM_PAGE(pa);
3418 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3419 va, pa, m->hold_count, m->wire_count, m->flags);
3440 static void pads __P((pmap_t pm));
3441 void pmap_pvdump __P((vm_offset_t pa));
3443 /* print address space of pmap*/
3451 if (pm == kernel_pmap)
3453 for (i = 0; i < 1024; i++)
3455 for (j = 0; j < 1024; j++) {
3456 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3457 if (pm == kernel_pmap && va < KERNBASE)
3459 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3461 ptep = pmap_pte_quick(pm, va);
3462 if (pmap_pte_v(ptep))
3463 printf("%x:%x ", va, *(int *) ptep);
3472 register pv_entry_t pv;
3475 printf("pa %x", pa);
3476 m = PHYS_TO_VM_PAGE(pa);
3477 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3479 printf(" -> pmap %p, va %x, flags %x",
3480 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3482 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);