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 $
43 * $DragonFly: src/sys/i386/i386/Attic/pmap.c,v 1.13 2003/06/28 04:16:02 dillon Exp $
47 * Manages physical address maps.
49 * In addition to hardware address maps, this
50 * module is called upon to provide software-use-only
51 * maps which may or may not be stored in the same
52 * form as hardware maps. These pseudo-maps are
53 * used to store intermediate results from copy
54 * operations to and from address spaces.
56 * Since the information managed by this module is
57 * also stored by the logical address mapping module,
58 * this module may throw away valid virtual-to-physical
59 * mappings at almost any time. However, invalidations
60 * of virtual-to-physical mappings must be done as
63 * In order to cope with hardware architectures which
64 * make virtual-to-physical map invalidates expensive,
65 * this module may delay invalidate or reduced protection
66 * operations until such time as they are actually
67 * necessary. This module is given full information as
68 * to which processors are currently using which maps,
69 * and to when physical maps must be made correct.
72 #include "opt_disable_pse.h"
74 #include "opt_msgbuf.h"
75 #include "opt_user_ldt.h"
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/kernel.h>
81 #include <sys/msgbuf.h>
82 #include <sys/vmmeter.h>
86 #include <vm/vm_param.h>
87 #include <sys/sysctl.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vm_zone.h>
100 #include <machine/cputypes.h>
101 #include <machine/md_var.h>
102 #include <machine/specialreg.h>
103 #if defined(SMP) || defined(APIC_IO)
104 #include <machine/smp.h>
105 #include <machine/apic.h>
106 #endif /* SMP || APIC_IO */
107 #include <machine/globaldata.h>
109 #define PMAP_KEEP_PDIRS
110 #ifndef PMAP_SHPGPERPROC
111 #define PMAP_SHPGPERPROC 200
114 #if defined(DIAGNOSTIC)
115 #define PMAP_DIAGNOSTIC
120 #if !defined(PMAP_DIAGNOSTIC)
121 #define PMAP_INLINE __inline
127 * Get PDEs and PTEs for user/kernel address space
129 #define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
130 #define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
132 #define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0)
133 #define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0)
134 #define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0)
135 #define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0)
136 #define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0)
138 #define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
139 #define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
142 * Given a map and a machine independent protection code,
143 * convert to a vax protection code.
145 #define pte_prot(m, p) (protection_codes[p])
146 static int protection_codes[8];
148 static struct pmap kernel_pmap_store;
151 vm_offset_t avail_start; /* PA of first available physical page */
152 vm_offset_t avail_end; /* PA of last available physical page */
153 vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */
154 vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */
155 static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
156 static int pgeflag; /* PG_G or-in */
157 static int pseflag; /* PG_PS or-in */
159 static vm_object_t kptobj;
162 vm_offset_t kernel_vm_end;
165 * Data for the pv entry allocation mechanism
167 static vm_zone_t pvzone;
168 static struct vm_zone pvzone_store;
169 static struct vm_object pvzone_obj;
170 static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
171 static int pmap_pagedaemon_waken = 0;
172 static struct pv_entry *pvinit;
175 * All those kernel PT submaps that BSD is so fond of
177 pt_entry_t *CMAP1 = 0;
178 caddr_t CADDR1 = 0, ptvmmap = 0;
179 static pt_entry_t *msgbufmap;
180 struct msgbuf *msgbufp=0;
185 static pt_entry_t *pt_crashdumpmap;
186 static caddr_t crashdumpmap;
188 extern pt_entry_t *SMPpt;
190 static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv));
191 static unsigned * get_ptbase __P((pmap_t pmap));
192 static pv_entry_t get_pv_entry __P((void));
193 static void i386_protection_init __P((void));
194 static __inline void pmap_changebit __P((vm_page_t m, int bit, boolean_t setem));
196 static void pmap_remove_all __P((vm_page_t m));
197 static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va,
198 vm_page_t m, vm_page_t mpte));
199 static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq,
201 static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va));
202 static int pmap_remove_entry __P((struct pmap *pmap, vm_page_t m,
204 static boolean_t pmap_testbit __P((vm_page_t m, int bit));
205 static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va,
206 vm_page_t mpte, vm_page_t m));
208 static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va));
210 static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p));
211 static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex));
212 static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va));
213 static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex));
214 static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t));
215 static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
217 static unsigned pdir4mb;
222 * Extract the page table entry associated
223 * with the given map/virtual_address pair.
226 PMAP_INLINE unsigned *
228 register pmap_t pmap;
234 pdeaddr = (unsigned *) pmap_pde(pmap, va);
235 if (*pdeaddr & PG_PS)
238 return get_ptbase(pmap) + i386_btop(va);
245 * Move the kernel virtual free pointer to the next
246 * 4MB. This is used to help improve performance
247 * by using a large (4MB) page for much of the kernel
248 * (.text, .data, .bss)
251 pmap_kmem_choose(vm_offset_t addr)
253 vm_offset_t newaddr = addr;
255 if (cpu_feature & CPUID_PSE) {
256 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
263 * Bootstrap the system enough to run with virtual memory.
265 * On the i386 this is called after mapping has already been enabled
266 * and just syncs the pmap module with what has already been done.
267 * [We can't call it easily with mapping off since the kernel is not
268 * mapped with PA == VA, hence we would have to relocate every address
269 * from the linked base (virtual) address "KERNBASE" to the actual
270 * (physical) address starting relative to 0]
273 pmap_bootstrap(firstaddr, loadaddr)
274 vm_offset_t firstaddr;
275 vm_offset_t loadaddr;
279 struct mdglobaldata *gd;
282 avail_start = firstaddr;
285 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too
286 * large. It should instead be correctly calculated in locore.s and
287 * not based on 'first' (which is a physical address, not a virtual
288 * address, for the start of unused physical memory). The kernel
289 * page tables are NOT double mapped and thus should not be included
290 * in this calculation.
292 virtual_avail = (vm_offset_t) KERNBASE + firstaddr;
293 virtual_avail = pmap_kmem_choose(virtual_avail);
295 virtual_end = VM_MAX_KERNEL_ADDRESS;
298 * Initialize protection array.
300 i386_protection_init();
303 * The kernel's pmap is statically allocated so we don't have to use
304 * pmap_create, which is unlikely to work correctly at this part of
305 * the boot sequence (XXX and which no longer exists).
307 kernel_pmap = &kernel_pmap_store;
309 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD);
310 kernel_pmap->pm_count = 1;
311 kernel_pmap->pm_active = -1; /* don't allow deactivation */
312 TAILQ_INIT(&kernel_pmap->pm_pvlist);
316 * Reserve some special page table entries/VA space for temporary
319 #define SYSMAP(c, p, v, n) \
320 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
323 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va);
326 * CMAP1/CMAP2 are used for zeroing and copying pages.
328 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
333 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
336 * msgbufp is used to map the system message buffer.
337 * XXX msgbufmap is not used.
339 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
340 atop(round_page(MSGBUF_SIZE)))
345 for (i = 0; i < NKPT; i++)
348 /* XXX - see also mp_machdep.c */
349 if (ncpus == 1 && (cpu_feature & CPUID_PGE))
355 * Initialize the 4MB page size flag
359 * The 4MB page version of the initial
360 * kernel page mapping.
364 #if !defined(DISABLE_PSE)
365 if (cpu_feature & CPUID_PSE) {
368 * Note that we have enabled PSE mode
371 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
372 ptditmp &= ~(NBPDR - 1);
373 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
378 * Enable the PSE mode.
380 load_cr4(rcr4() | CR4_PSE);
383 * We can do the mapping here for the single processor
384 * case. We simply ignore the old page table page from
388 * For SMP, we still need 4K pages to bootstrap APs,
389 * PSE will be enabled as soon as all APs are up.
391 PTD[KPTDI] = (pd_entry_t) ptditmp;
392 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp;
398 if (cpu_apic_address == 0)
399 panic("pmap_bootstrap: no local apic!");
401 /* local apic is mapped on last page */
402 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
403 (cpu_apic_address & PG_FRAME));
406 /* BSP does this itself, AP's get it pre-set */
407 gd = &CPU_prvspace[0].mdglobaldata;
408 gd->gd_CMAP1 = &SMPpt[1];
409 gd->gd_CMAP2 = &SMPpt[2];
410 gd->gd_CMAP3 = &SMPpt[3];
411 gd->gd_PMAP1 = &SMPpt[4];
412 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
413 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
414 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
415 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
422 * Set 4mb pdir for mp startup
427 if (pseflag && (cpu_feature & CPUID_PSE)) {
428 load_cr4(rcr4() | CR4_PSE);
429 if (pdir4mb && cpuid == 0) { /* only on BSP */
430 kernel_pmap->pm_pdir[KPTDI] =
431 PTD[KPTDI] = (pd_entry_t)pdir4mb;
439 * Initialize the pmap module.
440 * Called by vm_init, to initialize any structures that the pmap
441 * system needs to map virtual memory.
442 * pmap_init has been enhanced to support in a fairly consistant
443 * way, discontiguous physical memory.
446 pmap_init(phys_start, phys_end)
447 vm_offset_t phys_start, phys_end;
453 * object for kernel page table pages
455 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
458 * Allocate memory for random pmap data structures. Includes the
462 for(i = 0; i < vm_page_array_size; i++) {
465 m = &vm_page_array[i];
466 TAILQ_INIT(&m->md.pv_list);
467 m->md.pv_list_count = 0;
471 * init the pv free list
473 initial_pvs = vm_page_array_size;
474 if (initial_pvs < MINPV)
476 pvzone = &pvzone_store;
477 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
478 initial_pvs * sizeof (struct pv_entry));
479 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
483 * Now it is safe to enable pv_table recording.
485 pmap_initialized = TRUE;
489 * Initialize the address space (zone) for the pv_entries. Set a
490 * high water mark so that the system can recover from excessive
491 * numbers of pv entries.
496 int shpgperproc = PMAP_SHPGPERPROC;
498 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
499 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
500 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
501 pv_entry_high_water = 9 * (pv_entry_max / 10);
502 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
506 /***************************************************
507 * Low level helper routines.....
508 ***************************************************/
510 #if defined(PMAP_DIAGNOSTIC)
513 * This code checks for non-writeable/modified pages.
514 * This should be an invalid condition.
517 pmap_nw_modified(pt_entry_t ptea)
523 if ((pte & (PG_M|PG_RW)) == PG_M)
532 * this routine defines the region(s) of memory that should
533 * not be tested for the modified bit.
535 static PMAP_INLINE int
536 pmap_track_modified(vm_offset_t va)
538 if ((va < clean_sva) || (va >= clean_eva))
544 static PMAP_INLINE void
545 invltlb_1pg(vm_offset_t va)
547 #if defined(I386_CPU)
548 if (cpu_class == CPUCLASS_386) {
558 pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
561 if (pmap->pm_active & (1 << cpuid))
562 cpu_invlpg((void *)va);
563 if (pmap->pm_active & other_cpus)
572 pmap_TLB_invalidate_all(pmap_t pmap)
575 if (pmap->pm_active & (1 << cpuid))
577 if (pmap->pm_active & other_cpus)
589 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
591 /* are we current address space or kernel? */
592 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
593 return (unsigned *) PTmap;
595 /* otherwise, we are alternate address space */
596 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
597 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
599 /* The page directory is not shared between CPUs */
605 return (unsigned *) APTmap;
609 * Super fast pmap_pte routine best used when scanning
610 * the pv lists. This eliminates many coarse-grained
611 * invltlb calls. Note that many of the pv list
612 * scans are across different pmaps. It is very wasteful
613 * to do an entire invltlb for checking a single mapping.
617 pmap_pte_quick(pmap_t pmap, vm_offset_t va)
619 struct mdglobaldata *gd = mdcpu;
622 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
623 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
624 unsigned index = i386_btop(va);
625 /* are we current address space or kernel? */
626 if ((pmap == kernel_pmap) ||
627 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
628 return (unsigned *) PTmap + index;
630 newpf = pde & PG_FRAME;
631 if ( ((* (unsigned *) gd->gd_PMAP1) & PG_FRAME) != newpf) {
632 * (unsigned *) gd->gd_PMAP1 = newpf | PG_RW | PG_V;
633 cpu_invlpg(gd->gd_PADDR1);
635 return gd->gd_PADDR1 + ((unsigned) index & (NPTEPG - 1));
641 * Routine: pmap_extract
643 * Extract the physical page address associated
644 * with the given map/virtual_address pair.
647 pmap_extract(pmap, va)
648 register pmap_t pmap;
652 vm_offset_t pdirindex;
653 pdirindex = va >> PDRSHIFT;
654 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
656 if ((rtval & PG_PS) != 0) {
657 rtval &= ~(NBPDR - 1);
658 rtval |= va & (NBPDR - 1);
661 pte = get_ptbase(pmap) + i386_btop(va);
662 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
669 /***************************************************
670 * Low level mapping routines.....
671 ***************************************************/
674 * add a wired page to the kva
675 * note that in order for the mapping to take effect -- you
676 * should do a invltlb after doing the pmap_kenter...
681 register vm_offset_t pa;
683 register unsigned *pte;
686 npte = pa | PG_RW | PG_V | pgeflag;
687 pte = (unsigned *)vtopte(va);
694 * remove a page from the kernel pagetables
700 register unsigned *pte;
702 pte = (unsigned *)vtopte(va);
708 * Used to map a range of physical addresses into kernel
709 * virtual address space.
711 * For now, VM is already on, we only need to map the
715 pmap_map(virt, start, end, prot)
721 while (start < end) {
722 pmap_kenter(virt, start);
731 * Add a list of wired pages to the kva
732 * this routine is only used for temporary
733 * kernel mappings that do not need to have
734 * page modification or references recorded.
735 * Note that old mappings are simply written
736 * over. The page *must* be wired.
739 pmap_qenter(va, m, count)
746 end_va = va + count * PAGE_SIZE;
748 while (va < end_va) {
751 pte = (unsigned *)vtopte(va);
752 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
754 cpu_invlpg((void *)va);
767 * this routine jerks page mappings from the
768 * kernel -- it is meant only for temporary mappings.
771 pmap_qremove(va, count)
777 end_va = va + count*PAGE_SIZE;
779 while (va < end_va) {
782 pte = (unsigned *)vtopte(va);
785 cpu_invlpg((void *)va);
797 pmap_page_lookup(object, pindex)
803 m = vm_page_lookup(object, pindex);
804 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
810 * Create a new thread and optionally associate it with a (new) process.
813 pmap_init_thread(thread_t td)
815 td->td_pcb = (struct pcb *)(td->td_kstack + UPAGES * PAGE_SIZE) - 1;
816 td->td_sp = (char *)td->td_pcb - 16;
820 * Create the UPAGES for a new process.
821 * This routine directly affects the fork perf for a process.
824 pmap_init_proc(struct proc *p, struct thread *td)
826 p->p_addr = (void *)td->td_kstack;
829 td->td_switch = cpu_heavy_switch;
830 bzero(p->p_addr, sizeof(*p->p_addr));
834 * Dispose the UPAGES for a process that has exited.
835 * This routine directly impacts the exit perf of a process.
838 pmap_dispose_proc(struct proc *p)
842 KASSERT(p->p_lock == 0, ("attempt to dispose referenced proc! %p", p));
844 if ((td = p->p_thread) != NULL) {
853 * Allow the UPAGES for a process to be prejudicially paged out.
864 upobj = p->p_upages_obj;
866 * let the upages be paged
868 for(i=0;i<UPAGES;i++) {
869 if ((m = vm_page_lookup(upobj, i)) == NULL)
870 panic("pmap_swapout_proc: upage already missing???");
872 vm_page_unwire(m, 0);
873 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
879 * Bring the UPAGES for a specified process back in.
890 upobj = p->p_upages_obj;
891 for(i=0;i<UPAGES;i++) {
893 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
895 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
898 if (m->valid != VM_PAGE_BITS_ALL) {
899 rv = vm_pager_get_pages(upobj, &m, 1, 0);
900 if (rv != VM_PAGER_OK)
901 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
902 m = vm_page_lookup(upobj, i);
903 m->valid = VM_PAGE_BITS_ALL;
908 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
913 /***************************************************
914 * Page table page management routines.....
915 ***************************************************/
918 * This routine unholds page table pages, and if the hold count
919 * drops to zero, then it decrements the wire count.
922 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
924 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
927 if (m->hold_count == 0) {
930 * unmap the page table page
932 pmap->pm_pdir[m->pindex] = 0;
933 --pmap->pm_stats.resident_count;
934 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
935 (((unsigned) PTDpde) & PG_FRAME)) {
937 * Do a invltlb to make the invalidated mapping
938 * take effect immediately.
940 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
941 pmap_TLB_invalidate(pmap, pteva);
944 if (pmap->pm_ptphint == m)
945 pmap->pm_ptphint = NULL;
948 * If the page is finally unwired, simply free it.
951 if (m->wire_count == 0) {
955 vm_page_free_zero(m);
963 static PMAP_INLINE int
964 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
967 if (m->hold_count == 0)
968 return _pmap_unwire_pte_hold(pmap, m);
974 * After removing a page table entry, this routine is used to
975 * conditionally free the page, and manage the hold/wire counts.
978 pmap_unuse_pt(pmap, va, mpte)
984 if (va >= UPT_MIN_ADDRESS)
988 ptepindex = (va >> PDRSHIFT);
989 if (pmap->pm_ptphint &&
990 (pmap->pm_ptphint->pindex == ptepindex)) {
991 mpte = pmap->pm_ptphint;
993 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
994 pmap->pm_ptphint = mpte;
998 return pmap_unwire_pte_hold(pmap, mpte);
1006 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1007 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1009 pmap->pm_active = 0;
1010 pmap->pm_ptphint = NULL;
1011 TAILQ_INIT(&pmap->pm_pvlist);
1012 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1016 * Initialize a preallocated and zeroed pmap structure,
1017 * such as one in a vmspace structure.
1021 register struct pmap *pmap;
1026 * No need to allocate page table space yet but we do need a valid
1027 * page directory table.
1029 if (pmap->pm_pdir == NULL)
1031 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1034 * allocate object for the ptes
1036 if (pmap->pm_pteobj == NULL)
1037 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1040 * allocate the page directory page
1042 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1043 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1045 ptdpg->wire_count = 1;
1049 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1050 ptdpg->valid = VM_PAGE_BITS_ALL;
1052 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1053 if ((ptdpg->flags & PG_ZERO) == 0)
1054 bzero(pmap->pm_pdir, PAGE_SIZE);
1056 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1058 /* install self-referential address mapping entry */
1059 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1060 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1063 pmap->pm_active = 0;
1064 pmap->pm_ptphint = NULL;
1065 TAILQ_INIT(&pmap->pm_pvlist);
1066 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1070 * Wire in kernel global address entries. To avoid a race condition
1071 * between pmap initialization and pmap_growkernel, this procedure
1072 * should be called after the vmspace is attached to the process
1073 * but before this pmap is activated.
1079 /* XXX copies current process, does not fill in MPPTDI */
1080 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1084 pmap_release_free_page(pmap, p)
1088 unsigned *pde = (unsigned *) pmap->pm_pdir;
1090 * This code optimizes the case of freeing non-busy
1091 * page-table pages. Those pages are zero now, and
1092 * might as well be placed directly into the zero queue.
1094 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1100 * Remove the page table page from the processes address space.
1103 pmap->pm_stats.resident_count--;
1105 if (p->hold_count) {
1106 panic("pmap_release: freeing held page table page");
1109 * Page directory pages need to have the kernel
1110 * stuff cleared, so they can go into the zero queue also.
1112 if (p->pindex == PTDPTDI) {
1113 bzero(pde + KPTDI, nkpt * PTESIZE);
1116 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1119 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1120 pmap->pm_ptphint = NULL;
1124 vm_page_free_zero(p);
1129 * this routine is called if the page table page is not
1133 _pmap_allocpte(pmap, ptepindex)
1137 vm_offset_t pteva, ptepa;
1141 * Find or fabricate a new pagetable page
1143 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1144 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1146 KASSERT(m->queue == PQ_NONE,
1147 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1149 if (m->wire_count == 0)
1154 * Increment the hold count for the page table page
1155 * (denoting a new mapping.)
1160 * Map the pagetable page into the process address space, if
1161 * it isn't already there.
1164 pmap->pm_stats.resident_count++;
1166 ptepa = VM_PAGE_TO_PHYS(m);
1167 pmap->pm_pdir[ptepindex] =
1168 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1171 * Set the page table hint
1173 pmap->pm_ptphint = m;
1176 * Try to use the new mapping, but if we cannot, then
1177 * do it with the routine that maps the page explicitly.
1179 if ((m->flags & PG_ZERO) == 0) {
1180 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1181 (((unsigned) PTDpde) & PG_FRAME)) {
1182 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1183 bzero((caddr_t) pteva, PAGE_SIZE);
1185 pmap_zero_page(ptepa);
1189 m->valid = VM_PAGE_BITS_ALL;
1190 vm_page_flag_clear(m, PG_ZERO);
1191 vm_page_flag_set(m, PG_MAPPED);
1198 pmap_allocpte(pmap, va)
1207 * Calculate pagetable page index
1209 ptepindex = va >> PDRSHIFT;
1212 * Get the page directory entry
1214 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1217 * This supports switching from a 4MB page to a
1220 if (ptepa & PG_PS) {
1221 pmap->pm_pdir[ptepindex] = 0;
1227 * If the page table page is mapped, we just increment the
1228 * hold count, and activate it.
1232 * In order to get the page table page, try the
1235 if (pmap->pm_ptphint &&
1236 (pmap->pm_ptphint->pindex == ptepindex)) {
1237 m = pmap->pm_ptphint;
1239 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1240 pmap->pm_ptphint = m;
1246 * Here if the pte page isn't mapped, or if it has been deallocated.
1248 return _pmap_allocpte(pmap, ptepindex);
1252 /***************************************************
1253 * Pmap allocation/deallocation routines.
1254 ***************************************************/
1257 * Release any resources held by the given physical map.
1258 * Called when a pmap initialized by pmap_pinit is being released.
1259 * Should only be called if the map contains no valid mappings.
1263 register struct pmap *pmap;
1265 vm_page_t p,n,ptdpg;
1266 vm_object_t object = pmap->pm_pteobj;
1269 #if defined(DIAGNOSTIC)
1270 if (object->ref_count != 1)
1271 panic("pmap_release: pteobj reference count != 1");
1276 curgeneration = object->generation;
1277 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1278 n = TAILQ_NEXT(p, listq);
1279 if (p->pindex == PTDPTDI) {
1284 if (!pmap_release_free_page(pmap, p) &&
1285 (object->generation != curgeneration))
1290 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1295 kvm_size(SYSCTL_HANDLER_ARGS)
1297 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1299 return sysctl_handle_long(oidp, &ksize, 0, req);
1301 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1302 0, 0, kvm_size, "IU", "Size of KVM");
1305 kvm_free(SYSCTL_HANDLER_ARGS)
1307 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1309 return sysctl_handle_long(oidp, &kfree, 0, req);
1311 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1312 0, 0, kvm_free, "IU", "Amount of KVM free");
1315 * grow the number of kernel page table entries, if needed
1318 pmap_growkernel(vm_offset_t addr)
1323 vm_offset_t ptppaddr;
1328 if (kernel_vm_end == 0) {
1329 kernel_vm_end = KERNBASE;
1331 while (pdir_pde(PTD, kernel_vm_end)) {
1332 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1336 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1337 while (kernel_vm_end < addr) {
1338 if (pdir_pde(PTD, kernel_vm_end)) {
1339 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1344 * This index is bogus, but out of the way
1346 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1348 panic("pmap_growkernel: no memory to grow kernel");
1353 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1354 pmap_zero_page(ptppaddr);
1355 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1356 pdir_pde(PTD, kernel_vm_end) = newpdir;
1358 LIST_FOREACH(p, &allproc, p_list) {
1360 pmap = vmspace_pmap(p->p_vmspace);
1361 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1364 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1365 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1371 * Retire the given physical map from service.
1372 * Should only be called if the map contains
1373 * no valid mappings.
1377 register pmap_t pmap;
1384 count = --pmap->pm_count;
1387 panic("destroying a pmap is not yet implemented");
1392 * Add a reference to the specified pmap.
1395 pmap_reference(pmap)
1403 /***************************************************
1404 * page management routines.
1405 ***************************************************/
1408 * free the pv_entry back to the free list
1410 static PMAP_INLINE void
1419 * get a new pv_entry, allocating a block from the system
1421 * the memory allocation is performed bypassing the malloc code
1422 * because of the possibility of allocations at interrupt time.
1428 if (pv_entry_high_water &&
1429 (pv_entry_count > pv_entry_high_water) &&
1430 (pmap_pagedaemon_waken == 0)) {
1431 pmap_pagedaemon_waken = 1;
1432 wakeup (&vm_pages_needed);
1434 return zalloci(pvzone);
1438 * This routine is very drastic, but can save the system
1446 static int warningdone=0;
1448 if (pmap_pagedaemon_waken == 0)
1451 if (warningdone < 5) {
1452 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1456 for(i = 0; i < vm_page_array_size; i++) {
1457 m = &vm_page_array[i];
1458 if (m->wire_count || m->hold_count || m->busy ||
1459 (m->flags & PG_BUSY))
1463 pmap_pagedaemon_waken = 0;
1468 * If it is the first entry on the list, it is actually
1469 * in the header and we must copy the following entry up
1470 * to the header. Otherwise we must search the list for
1471 * the entry. In either case we free the now unused entry.
1475 pmap_remove_entry(pmap, m, va)
1485 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1486 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1487 if (pmap == pv->pv_pmap && va == pv->pv_va)
1491 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1492 if (va == pv->pv_va)
1500 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1501 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1502 m->md.pv_list_count--;
1503 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1504 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1506 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1515 * Create a pv entry for page at pa for
1519 pmap_insert_entry(pmap, va, mpte, m)
1530 pv = get_pv_entry();
1535 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1536 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1537 m->md.pv_list_count++;
1543 * pmap_remove_pte: do the things to unmap a page in a process
1546 pmap_remove_pte(pmap, ptq, va)
1554 oldpte = loadandclear(ptq);
1556 pmap->pm_stats.wired_count -= 1;
1558 * Machines that don't support invlpg, also don't support
1563 pmap->pm_stats.resident_count -= 1;
1564 if (oldpte & PG_MANAGED) {
1565 m = PHYS_TO_VM_PAGE(oldpte);
1566 if (oldpte & PG_M) {
1567 #if defined(PMAP_DIAGNOSTIC)
1568 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1570 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1574 if (pmap_track_modified(va))
1578 vm_page_flag_set(m, PG_REFERENCED);
1579 return pmap_remove_entry(pmap, m, va);
1581 return pmap_unuse_pt(pmap, va, NULL);
1588 * Remove a single page from a process address space
1591 pmap_remove_page(pmap, va)
1593 register vm_offset_t va;
1595 register unsigned *ptq;
1598 * if there is no pte for this address, just skip it!!!
1600 if (*pmap_pde(pmap, va) == 0) {
1605 * get a local va for mappings for this pmap.
1607 ptq = get_ptbase(pmap) + i386_btop(va);
1609 (void) pmap_remove_pte(pmap, ptq, va);
1610 pmap_TLB_invalidate(pmap, va);
1616 * Remove the given range of addresses from the specified map.
1618 * It is assumed that the start and end are properly
1619 * rounded to the page size.
1622 pmap_remove(pmap, sva, eva)
1624 register vm_offset_t sva;
1625 register vm_offset_t eva;
1627 register unsigned *ptbase;
1629 vm_offset_t ptpaddr;
1630 vm_offset_t sindex, eindex;
1636 if (pmap->pm_stats.resident_count == 0)
1640 * special handling of removing one page. a very
1641 * common operation and easy to short circuit some
1644 if (((sva + PAGE_SIZE) == eva) &&
1645 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1646 pmap_remove_page(pmap, sva);
1653 * Get a local virtual address for the mappings that are being
1656 ptbase = get_ptbase(pmap);
1658 sindex = i386_btop(sva);
1659 eindex = i386_btop(eva);
1661 for (; sindex < eindex; sindex = pdnxt) {
1665 * Calculate index for next page table.
1667 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1668 if (pmap->pm_stats.resident_count == 0)
1671 pdirindex = sindex / NPDEPG;
1672 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1673 pmap->pm_pdir[pdirindex] = 0;
1674 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1680 * Weed out invalid mappings. Note: we assume that the page
1681 * directory table is always allocated, and in kernel virtual.
1687 * Limit our scan to either the end of the va represented
1688 * by the current page table page, or to the end of the
1689 * range being removed.
1691 if (pdnxt > eindex) {
1695 for ( ;sindex != pdnxt; sindex++) {
1697 if (ptbase[sindex] == 0) {
1700 va = i386_ptob(sindex);
1703 if (pmap_remove_pte(pmap,
1704 ptbase + sindex, va))
1710 pmap_TLB_invalidate_all(pmap);
1714 * Routine: pmap_remove_all
1716 * Removes this physical page from
1717 * all physical maps in which it resides.
1718 * Reflects back modify bits to the pager.
1721 * Original versions of this routine were very
1722 * inefficient because they iteratively called
1723 * pmap_remove (slow...)
1730 register pv_entry_t pv;
1731 register unsigned *pte, tpte;
1734 #if defined(PMAP_DIAGNOSTIC)
1736 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1739 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1740 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1745 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1746 pv->pv_pmap->pm_stats.resident_count--;
1748 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1750 tpte = loadandclear(pte);
1752 pv->pv_pmap->pm_stats.wired_count--;
1755 vm_page_flag_set(m, PG_REFERENCED);
1758 * Update the vm_page_t clean and reference bits.
1761 #if defined(PMAP_DIAGNOSTIC)
1762 if (pmap_nw_modified((pt_entry_t) tpte)) {
1764 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1768 if (pmap_track_modified(pv->pv_va))
1771 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1773 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1774 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1775 m->md.pv_list_count--;
1776 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1780 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1786 * Set the physical protection on the
1787 * specified range of this map as requested.
1790 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1792 register unsigned *ptbase;
1793 vm_offset_t pdnxt, ptpaddr;
1794 vm_pindex_t sindex, eindex;
1800 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1801 pmap_remove(pmap, sva, eva);
1805 if (prot & VM_PROT_WRITE)
1810 ptbase = get_ptbase(pmap);
1812 sindex = i386_btop(sva);
1813 eindex = i386_btop(eva);
1815 for (; sindex < eindex; sindex = pdnxt) {
1819 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1821 pdirindex = sindex / NPDEPG;
1822 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1823 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1824 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1830 * Weed out invalid mappings. Note: we assume that the page
1831 * directory table is always allocated, and in kernel virtual.
1836 if (pdnxt > eindex) {
1840 for (; sindex != pdnxt; sindex++) {
1845 pbits = ptbase[sindex];
1847 if (pbits & PG_MANAGED) {
1850 m = PHYS_TO_VM_PAGE(pbits);
1851 vm_page_flag_set(m, PG_REFERENCED);
1855 if (pmap_track_modified(i386_ptob(sindex))) {
1857 m = PHYS_TO_VM_PAGE(pbits);
1866 if (pbits != ptbase[sindex]) {
1867 ptbase[sindex] = pbits;
1873 pmap_TLB_invalidate_all(pmap);
1877 * Insert the given physical page (p) at
1878 * the specified virtual address (v) in the
1879 * target physical map with the protection requested.
1881 * If specified, the page will be wired down, meaning
1882 * that the related pte can not be reclaimed.
1884 * NB: This is the only routine which MAY NOT lazy-evaluate
1885 * or lose information. That is, this routine must actually
1886 * insert this page into the given map NOW.
1889 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1893 register unsigned *pte;
1895 vm_offset_t origpte, newpte;
1902 #ifdef PMAP_DIAGNOSTIC
1903 if (va > VM_MAX_KERNEL_ADDRESS)
1904 panic("pmap_enter: toobig");
1905 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
1906 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
1911 * In the case that a page table page is not
1912 * resident, we are creating it here.
1914 if (va < UPT_MIN_ADDRESS) {
1915 mpte = pmap_allocpte(pmap, va);
1917 #if 0 && defined(PMAP_DIAGNOSTIC)
1919 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
1920 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
1921 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
1922 pmap->pm_pdir[PTDPTDI], origpte, va);
1925 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
1926 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
1927 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
1928 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
1929 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
1930 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
1931 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
1937 pte = pmap_pte(pmap, va);
1940 * Page Directory table entry not valid, we need a new PT page
1943 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
1944 (void *)pmap->pm_pdir[PTDPTDI], va);
1947 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
1948 origpte = *(vm_offset_t *)pte;
1949 opa = origpte & PG_FRAME;
1951 if (origpte & PG_PS)
1952 panic("pmap_enter: attempted pmap_enter on 4MB page");
1955 * Mapping has not changed, must be protection or wiring change.
1957 if (origpte && (opa == pa)) {
1959 * Wiring change, just update stats. We don't worry about
1960 * wiring PT pages as they remain resident as long as there
1961 * are valid mappings in them. Hence, if a user page is wired,
1962 * the PT page will be also.
1964 if (wired && ((origpte & PG_W) == 0))
1965 pmap->pm_stats.wired_count++;
1966 else if (!wired && (origpte & PG_W))
1967 pmap->pm_stats.wired_count--;
1969 #if defined(PMAP_DIAGNOSTIC)
1970 if (pmap_nw_modified((pt_entry_t) origpte)) {
1972 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
1978 * Remove extra pte reference
1983 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
1984 if ((origpte & PG_RW) == 0) {
1987 cpu_invlpg((void *)va);
1988 if (pmap->pm_active & other_cpus)
1998 * We might be turning off write access to the page,
1999 * so we go ahead and sense modify status.
2001 if (origpte & PG_MANAGED) {
2002 if ((origpte & PG_M) && pmap_track_modified(va)) {
2004 om = PHYS_TO_VM_PAGE(opa);
2012 * Mapping has changed, invalidate old range and fall through to
2013 * handle validating new mapping.
2017 err = pmap_remove_pte(pmap, pte, va);
2019 panic("pmap_enter: pte vanished, va: 0x%x", va);
2023 * Enter on the PV list if part of our managed memory. Note that we
2024 * raise IPL while manipulating pv_table since pmap_enter can be
2025 * called at interrupt time.
2027 if (pmap_initialized &&
2028 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2029 pmap_insert_entry(pmap, va, mpte, m);
2034 * Increment counters
2036 pmap->pm_stats.resident_count++;
2038 pmap->pm_stats.wired_count++;
2042 * Now validate mapping with desired protection/wiring.
2044 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2048 if (va < UPT_MIN_ADDRESS)
2050 if (pmap == kernel_pmap)
2054 * if the mapping or permission bits are different, we need
2055 * to update the pte.
2057 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2058 *pte = newpte | PG_A;
2061 cpu_invlpg((void *)va);
2062 if (pmap->pm_active & other_cpus)
2072 * this code makes some *MAJOR* assumptions:
2073 * 1. Current pmap & pmap exists.
2076 * 4. No page table pages.
2077 * 5. Tlbflush is deferred to calling procedure.
2078 * 6. Page IS managed.
2079 * but is *MUCH* faster than pmap_enter...
2083 pmap_enter_quick(pmap, va, m, mpte)
2084 register pmap_t pmap;
2093 * In the case that a page table page is not
2094 * resident, we are creating it here.
2096 if (va < UPT_MIN_ADDRESS) {
2101 * Calculate pagetable page index
2103 ptepindex = va >> PDRSHIFT;
2104 if (mpte && (mpte->pindex == ptepindex)) {
2109 * Get the page directory entry
2111 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2114 * If the page table page is mapped, we just increment
2115 * the hold count, and activate it.
2119 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2120 if (pmap->pm_ptphint &&
2121 (pmap->pm_ptphint->pindex == ptepindex)) {
2122 mpte = pmap->pm_ptphint;
2124 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2125 pmap->pm_ptphint = mpte;
2131 mpte = _pmap_allocpte(pmap, ptepindex);
2139 * This call to vtopte makes the assumption that we are
2140 * entering the page into the current pmap. In order to support
2141 * quick entry into any pmap, one would likely use pmap_pte_quick.
2142 * But that isn't as quick as vtopte.
2144 pte = (unsigned *)vtopte(va);
2147 pmap_unwire_pte_hold(pmap, mpte);
2152 * Enter on the PV list if part of our managed memory. Note that we
2153 * raise IPL while manipulating pv_table since pmap_enter can be
2154 * called at interrupt time.
2156 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2157 pmap_insert_entry(pmap, va, mpte, m);
2160 * Increment counters
2162 pmap->pm_stats.resident_count++;
2164 pa = VM_PAGE_TO_PHYS(m);
2167 * Now validate mapping with RO protection
2169 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2170 *pte = pa | PG_V | PG_U;
2172 *pte = pa | PG_V | PG_U | PG_MANAGED;
2178 * Make a temporary mapping for a physical address. This is only intended
2179 * to be used for panic dumps.
2182 pmap_kenter_temporary(vm_offset_t pa, int i)
2184 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2185 return ((void *)crashdumpmap);
2188 #define MAX_INIT_PT (96)
2190 * pmap_object_init_pt preloads the ptes for a given object
2191 * into the specified pmap. This eliminates the blast of soft
2192 * faults on process startup and immediately after an mmap.
2195 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2208 if (pmap == NULL || object == NULL)
2212 * This code maps large physical mmap regions into the
2213 * processor address space. Note that some shortcuts
2214 * are taken, but the code works.
2217 (object->type == OBJT_DEVICE) &&
2218 ((addr & (NBPDR - 1)) == 0) &&
2219 ((size & (NBPDR - 1)) == 0) ) {
2222 unsigned int ptepindex;
2226 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2230 p = vm_page_lookup(object, pindex);
2231 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2235 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2240 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2245 p = vm_page_lookup(object, pindex);
2249 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2250 if (ptepa & (NBPDR - 1)) {
2254 p->valid = VM_PAGE_BITS_ALL;
2256 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2257 npdes = size >> PDRSHIFT;
2258 for(i=0;i<npdes;i++) {
2259 pmap->pm_pdir[ptepindex] =
2260 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2264 vm_page_flag_set(p, PG_MAPPED);
2269 psize = i386_btop(size);
2271 if ((object->type != OBJT_VNODE) ||
2272 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2273 (object->resident_page_count > MAX_INIT_PT))) {
2277 if (psize + pindex > object->size) {
2278 if (object->size < pindex)
2280 psize = object->size - pindex;
2285 * if we are processing a major portion of the object, then scan the
2288 if (psize > (object->resident_page_count >> 2)) {
2291 for (p = TAILQ_FIRST(&object->memq);
2292 ((objpgs > 0) && (p != NULL));
2293 p = TAILQ_NEXT(p, listq)) {
2296 if (tmpidx < pindex) {
2300 if (tmpidx >= psize) {
2304 * don't allow an madvise to blow away our really
2305 * free pages allocating pv entries.
2307 if ((limit & MAP_PREFAULT_MADVISE) &&
2308 cnt.v_free_count < cnt.v_free_reserved) {
2311 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2313 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2314 if ((p->queue - p->pc) == PQ_CACHE)
2315 vm_page_deactivate(p);
2317 mpte = pmap_enter_quick(pmap,
2318 addr + i386_ptob(tmpidx), p, mpte);
2319 vm_page_flag_set(p, PG_MAPPED);
2326 * else lookup the pages one-by-one.
2328 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2330 * don't allow an madvise to blow away our really
2331 * free pages allocating pv entries.
2333 if ((limit & MAP_PREFAULT_MADVISE) &&
2334 cnt.v_free_count < cnt.v_free_reserved) {
2337 p = vm_page_lookup(object, tmpidx + pindex);
2339 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2341 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2342 if ((p->queue - p->pc) == PQ_CACHE)
2343 vm_page_deactivate(p);
2345 mpte = pmap_enter_quick(pmap,
2346 addr + i386_ptob(tmpidx), p, mpte);
2347 vm_page_flag_set(p, PG_MAPPED);
2356 * pmap_prefault provides a quick way of clustering
2357 * pagefaults into a processes address space. It is a "cousin"
2358 * of pmap_object_init_pt, except it runs at page fault time instead
2363 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2365 static int pmap_prefault_pageorder[] = {
2366 -PAGE_SIZE, PAGE_SIZE,
2367 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2368 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2369 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2373 pmap_prefault(pmap, addra, entry)
2376 vm_map_entry_t entry;
2385 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2388 object = entry->object.vm_object;
2390 starta = addra - PFBAK * PAGE_SIZE;
2391 if (starta < entry->start) {
2392 starta = entry->start;
2393 } else if (starta > addra) {
2398 for (i = 0; i < PAGEORDER_SIZE; i++) {
2399 vm_object_t lobject;
2402 addr = addra + pmap_prefault_pageorder[i];
2403 if (addr > addra + (PFFOR * PAGE_SIZE))
2406 if (addr < starta || addr >= entry->end)
2409 if ((*pmap_pde(pmap, addr)) == NULL)
2412 pte = (unsigned *) vtopte(addr);
2416 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2418 for (m = vm_page_lookup(lobject, pindex);
2419 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2420 lobject = lobject->backing_object) {
2421 if (lobject->backing_object_offset & PAGE_MASK)
2423 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2424 m = vm_page_lookup(lobject->backing_object, pindex);
2428 * give-up when a page is not in memory
2433 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2435 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2437 if ((m->queue - m->pc) == PQ_CACHE) {
2438 vm_page_deactivate(m);
2441 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2442 vm_page_flag_set(m, PG_MAPPED);
2449 * Routine: pmap_change_wiring
2450 * Function: Change the wiring attribute for a map/virtual-address
2452 * In/out conditions:
2453 * The mapping must already exist in the pmap.
2456 pmap_change_wiring(pmap, va, wired)
2457 register pmap_t pmap;
2461 register unsigned *pte;
2466 pte = pmap_pte(pmap, va);
2468 if (wired && !pmap_pte_w(pte))
2469 pmap->pm_stats.wired_count++;
2470 else if (!wired && pmap_pte_w(pte))
2471 pmap->pm_stats.wired_count--;
2474 * Wiring is not a hardware characteristic so there is no need to
2477 pmap_pte_set_w(pte, wired);
2483 * Copy the range specified by src_addr/len
2484 * from the source map to the range dst_addr/len
2485 * in the destination map.
2487 * This routine is only advisory and need not do anything.
2491 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2492 pmap_t dst_pmap, src_pmap;
2493 vm_offset_t dst_addr;
2495 vm_offset_t src_addr;
2498 vm_offset_t end_addr = src_addr + len;
2500 unsigned src_frame, dst_frame;
2503 if (dst_addr != src_addr)
2506 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2507 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2511 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2512 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2513 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2515 /* The page directory is not shared between CPUs */
2522 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2523 unsigned *src_pte, *dst_pte;
2524 vm_page_t dstmpte, srcmpte;
2525 vm_offset_t srcptepaddr;
2528 if (addr >= UPT_MIN_ADDRESS)
2529 panic("pmap_copy: invalid to pmap_copy page tables\n");
2532 * Don't let optional prefaulting of pages make us go
2533 * way below the low water mark of free pages or way
2534 * above high water mark of used pv entries.
2536 if (cnt.v_free_count < cnt.v_free_reserved ||
2537 pv_entry_count > pv_entry_high_water)
2540 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2541 ptepindex = addr >> PDRSHIFT;
2543 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2544 if (srcptepaddr == 0)
2547 if (srcptepaddr & PG_PS) {
2548 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2549 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2550 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2555 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2556 if ((srcmpte == NULL) ||
2557 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2560 if (pdnxt > end_addr)
2563 src_pte = (unsigned *) vtopte(addr);
2564 dst_pte = (unsigned *) avtopte(addr);
2565 while (addr < pdnxt) {
2569 * we only virtual copy managed pages
2571 if ((ptetemp & PG_MANAGED) != 0) {
2573 * We have to check after allocpte for the
2574 * pte still being around... allocpte can
2577 dstmpte = pmap_allocpte(dst_pmap, addr);
2578 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2580 * Clear the modified and
2581 * accessed (referenced) bits
2584 m = PHYS_TO_VM_PAGE(ptetemp);
2585 *dst_pte = ptetemp & ~(PG_M | PG_A);
2586 dst_pmap->pm_stats.resident_count++;
2587 pmap_insert_entry(dst_pmap, addr,
2590 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2592 if (dstmpte->hold_count >= srcmpte->hold_count)
2603 * Routine: pmap_kernel
2605 * Returns the physical map handle for the kernel.
2610 return (kernel_pmap);
2614 * pmap_zero_page zeros the specified hardware page by mapping
2615 * the page into KVM and using bzero to clear its contents.
2618 pmap_zero_page(vm_offset_t phys)
2620 struct mdglobaldata *gd = mdcpu;
2622 if (*(int *)gd->gd_CMAP3)
2623 panic("pmap_zero_page: CMAP3 busy");
2625 *(int *)gd->gd_CMAP3 =
2626 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2627 cpu_invlpg(gd->gd_CADDR3);
2629 #if defined(I686_CPU)
2630 if (cpu_class == CPUCLASS_686)
2631 i686_pagezero(gd->gd_CADDR3);
2634 bzero(gd->gd_CADDR3, PAGE_SIZE);
2636 *(int *) gd->gd_CMAP3 = 0;
2640 * pmap_zero_page_area zeros the specified hardware page by mapping
2641 * the page into KVM and using bzero to clear its contents.
2643 * off and size may not cover an area beyond a single hardware page.
2646 pmap_zero_page_area(phys, off, size)
2651 struct mdglobaldata *gd = mdcpu;
2653 if (*(int *) gd->gd_CMAP3)
2654 panic("pmap_zero_page: CMAP3 busy");
2656 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2657 cpu_invlpg(gd->gd_CADDR3);
2659 #if defined(I686_CPU)
2660 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2661 i686_pagezero(gd->gd_CADDR3);
2664 bzero((char *)gd->gd_CADDR3 + off, size);
2666 *(int *) gd->gd_CMAP3 = 0;
2670 * pmap_copy_page copies the specified (machine independent)
2671 * page by mapping the page into virtual memory and using
2672 * bcopy to copy the page, one machine dependent page at a
2676 pmap_copy_page(src, dst)
2680 struct mdglobaldata *gd = mdcpu;
2682 if (*(int *) gd->gd_CMAP1)
2683 panic("pmap_copy_page: CMAP1 busy");
2684 if (*(int *) gd->gd_CMAP2)
2685 panic("pmap_copy_page: CMAP2 busy");
2687 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2688 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2690 cpu_invlpg(gd->gd_CADDR1);
2691 cpu_invlpg(gd->gd_CADDR2);
2693 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
2695 *(int *) gd->gd_CMAP1 = 0;
2696 *(int *) gd->gd_CMAP2 = 0;
2701 * Routine: pmap_pageable
2703 * Make the specified pages (by pmap, offset)
2704 * pageable (or not) as requested.
2706 * A page which is not pageable may not take
2707 * a fault; therefore, its page table entry
2708 * must remain valid for the duration.
2710 * This routine is merely advisory; pmap_enter
2711 * will specify that these pages are to be wired
2712 * down (or not) as appropriate.
2715 pmap_pageable(pmap, sva, eva, pageable)
2717 vm_offset_t sva, eva;
2723 * Returns true if the pmap's pv is one of the first
2724 * 16 pvs linked to from this page. This count may
2725 * be changed upwards or downwards in the future; it
2726 * is only necessary that true be returned for a small
2727 * subset of pmaps for proper page aging.
2730 pmap_page_exists_quick(pmap, m)
2738 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2743 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2744 if (pv->pv_pmap == pmap) {
2756 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2758 * Remove all pages from specified address space
2759 * this aids process exit speeds. Also, this code
2760 * is special cased for current process only, but
2761 * can have the more generic (and slightly slower)
2762 * mode enabled. This is much faster than pmap_remove
2763 * in the case of running down an entire address space.
2766 pmap_remove_pages(pmap, sva, eva)
2768 vm_offset_t sva, eva;
2770 unsigned *pte, tpte;
2775 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2776 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2777 printf("warning: pmap_remove_pages called with non-current pmap\n");
2783 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2787 if (pv->pv_va >= eva || pv->pv_va < sva) {
2788 npv = TAILQ_NEXT(pv, pv_plist);
2792 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2793 pte = (unsigned *)vtopte(pv->pv_va);
2795 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2800 * We cannot remove wired pages from a process' mapping at this time
2803 npv = TAILQ_NEXT(pv, pv_plist);
2808 m = PHYS_TO_VM_PAGE(tpte);
2810 KASSERT(m < &vm_page_array[vm_page_array_size],
2811 ("pmap_remove_pages: bad tpte %x", tpte));
2813 pv->pv_pmap->pm_stats.resident_count--;
2816 * Update the vm_page_t clean and reference bits.
2823 npv = TAILQ_NEXT(pv, pv_plist);
2824 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2826 m->md.pv_list_count--;
2827 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2828 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2829 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2832 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2836 pmap_TLB_invalidate_all(pmap);
2840 * pmap_testbit tests bits in pte's
2841 * note that the testbit/changebit routines are inline,
2842 * and a lot of things compile-time evaluate.
2845 pmap_testbit(m, bit)
2853 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2856 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
2861 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2863 * if the bit being tested is the modified bit, then
2864 * mark clean_map and ptes as never
2867 if (bit & (PG_A|PG_M)) {
2868 if (!pmap_track_modified(pv->pv_va))
2872 #if defined(PMAP_DIAGNOSTIC)
2874 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
2878 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2889 * this routine is used to modify bits in ptes
2891 static __inline void
2892 pmap_changebit(m, bit, setem)
2897 register pv_entry_t pv;
2898 register unsigned *pte;
2901 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2907 * Loop over all current mappings setting/clearing as appropos If
2908 * setting RO do we need to clear the VAC?
2910 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2912 * don't write protect pager mappings
2914 if (!setem && (bit == PG_RW)) {
2915 if (!pmap_track_modified(pv->pv_va))
2919 #if defined(PMAP_DIAGNOSTIC)
2921 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
2926 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2930 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2932 vm_offset_t pbits = *(vm_offset_t *)pte;
2938 *(int *)pte = pbits & ~(PG_M|PG_RW);
2940 *(int *)pte = pbits & ~bit;
2942 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2950 * pmap_page_protect:
2952 * Lower the permission for all mappings to a given page.
2955 pmap_page_protect(vm_page_t m, vm_prot_t prot)
2957 if ((prot & VM_PROT_WRITE) == 0) {
2958 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
2959 pmap_changebit(m, PG_RW, FALSE);
2967 pmap_phys_address(ppn)
2970 return (i386_ptob(ppn));
2974 * pmap_ts_referenced:
2976 * Return a count of reference bits for a page, clearing those bits.
2977 * It is not necessary for every reference bit to be cleared, but it
2978 * is necessary that 0 only be returned when there are truly no
2979 * reference bits set.
2981 * XXX: The exact number of bits to check and clear is a matter that
2982 * should be tested and standardized at some point in the future for
2983 * optimal aging of shared pages.
2986 pmap_ts_referenced(vm_page_t m)
2988 register pv_entry_t pv, pvf, pvn;
2993 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2998 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3003 pvn = TAILQ_NEXT(pv, pv_list);
3005 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3007 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3009 if (!pmap_track_modified(pv->pv_va))
3012 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3014 if (pte && (*pte & PG_A)) {
3017 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3024 } while ((pv = pvn) != NULL && pv != pvf);
3034 * Return whether or not the specified physical page was modified
3035 * in any physical maps.
3038 pmap_is_modified(vm_page_t m)
3040 return pmap_testbit(m, PG_M);
3044 * Clear the modify bits on the specified physical page.
3047 pmap_clear_modify(vm_page_t m)
3049 pmap_changebit(m, PG_M, FALSE);
3053 * pmap_clear_reference:
3055 * Clear the reference bit on the specified physical page.
3058 pmap_clear_reference(vm_page_t m)
3060 pmap_changebit(m, PG_A, FALSE);
3064 * Miscellaneous support routines follow
3068 i386_protection_init()
3070 register int *kp, prot;
3072 kp = protection_codes;
3073 for (prot = 0; prot < 8; prot++) {
3075 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3077 * Read access is also 0. There isn't any execute bit,
3078 * so just make it readable.
3080 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3081 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3082 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3085 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3086 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3087 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3088 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3096 * Map a set of physical memory pages into the kernel virtual
3097 * address space. Return a pointer to where it is mapped. This
3098 * routine is intended to be used for mapping device memory,
3102 pmap_mapdev(pa, size)
3106 vm_offset_t va, tmpva, offset;
3109 offset = pa & PAGE_MASK;
3110 size = roundup(offset + size, PAGE_SIZE);
3112 va = kmem_alloc_pageable(kernel_map, size);
3114 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3117 for (tmpva = va; size > 0;) {
3118 pte = (unsigned *)vtopte(tmpva);
3119 *pte = pa | PG_RW | PG_V | pgeflag;
3126 return ((void *)(va + offset));
3130 pmap_unmapdev(va, size)
3134 vm_offset_t base, offset;
3136 base = va & PG_FRAME;
3137 offset = va & PAGE_MASK;
3138 size = roundup(offset + size, PAGE_SIZE);
3139 kmem_free(kernel_map, base, size);
3143 * perform the pmap work for mincore
3146 pmap_mincore(pmap, addr)
3151 unsigned *ptep, pte;
3155 ptep = pmap_pte(pmap, addr);
3160 if ((pte = *ptep) != 0) {
3163 val = MINCORE_INCORE;
3164 if ((pte & PG_MANAGED) == 0)
3167 pa = pte & PG_FRAME;
3169 m = PHYS_TO_VM_PAGE(pa);
3175 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3177 * Modified by someone
3179 else if (m->dirty || pmap_is_modified(m))
3180 val |= MINCORE_MODIFIED_OTHER;
3185 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3188 * Referenced by someone
3190 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3191 val |= MINCORE_REFERENCED_OTHER;
3192 vm_page_flag_set(m, PG_REFERENCED);
3199 pmap_activate(struct proc *p)
3203 pmap = vmspace_pmap(p->p_vmspace);
3205 pmap->pm_active |= 1 << cpuid;
3207 pmap->pm_active |= 1;
3209 #if defined(SWTCH_OPTIM_STATS)
3212 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3213 load_cr3(p->p_thread->td_pcb->pcb_cr3);
3217 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3220 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3224 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3229 #if defined(PMAP_DEBUG)
3230 pmap_pid_dump(int pid)
3236 LIST_FOREACH(p, &allproc, p_list) {
3237 if (p->p_pid != pid)
3243 pmap = vmspace_pmap(p->p_vmspace);
3244 for(i=0;i<1024;i++) {
3247 unsigned base = i << PDRSHIFT;
3249 pde = &pmap->pm_pdir[i];
3250 if (pde && pmap_pde_v(pde)) {
3251 for(j=0;j<1024;j++) {
3252 unsigned va = base + (j << PAGE_SHIFT);
3253 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3260 pte = pmap_pte_quick( pmap, va);
3261 if (pte && pmap_pte_v(pte)) {
3265 m = PHYS_TO_VM_PAGE(pa);
3266 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3267 va, pa, m->hold_count, m->wire_count, m->flags);
3288 static void pads __P((pmap_t pm));
3289 void pmap_pvdump __P((vm_offset_t pa));
3291 /* print address space of pmap*/
3299 if (pm == kernel_pmap)
3301 for (i = 0; i < 1024; i++)
3303 for (j = 0; j < 1024; j++) {
3304 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3305 if (pm == kernel_pmap && va < KERNBASE)
3307 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3309 ptep = pmap_pte_quick(pm, va);
3310 if (pmap_pte_v(ptep))
3311 printf("%x:%x ", va, *(int *) ptep);
3320 register pv_entry_t pv;
3323 printf("pa %x", pa);
3324 m = PHYS_TO_VM_PAGE(pa);
3325 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3327 printf(" -> pmap %p, va %x, flags %x",
3328 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3330 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);