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/platform/pc32/i386/pmap.c,v 1.18 2003/07/13 07:10:06 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, *ptmmap;
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 * ptvmmap is used for reading arbitrary physical pages via
339 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
342 * msgbufp is used to map the system message buffer.
343 * XXX msgbufmap is not used.
345 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
346 atop(round_page(MSGBUF_SIZE)))
351 for (i = 0; i < NKPT; i++)
355 * PG_G is terribly broken on SMP because we IPI invltlb's in some
356 * cases rather then invl1pg. Actually, I don't even know why it
357 * works under UP because self-referential page table mappings
362 if (cpu_feature & CPUID_PGE)
367 * Initialize the 4MB page size flag
371 * The 4MB page version of the initial
372 * kernel page mapping.
376 #if !defined(DISABLE_PSE)
377 if (cpu_feature & CPUID_PSE) {
380 * Note that we have enabled PSE mode
383 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
384 ptditmp &= ~(NBPDR - 1);
385 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
390 * Enable the PSE mode. If we are SMP we can't do this
391 * now because the APs will not be able to use it when
394 load_cr4(rcr4() | CR4_PSE);
397 * We can do the mapping here for the single processor
398 * case. We simply ignore the old page table page from
402 * For SMP, we still need 4K pages to bootstrap APs,
403 * PSE will be enabled as soon as all APs are up.
405 PTD[KPTDI] = (pd_entry_t) ptditmp;
406 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp;
412 if (cpu_apic_address == 0)
413 panic("pmap_bootstrap: no local apic!");
415 /* local apic is mapped on last page */
416 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
417 (cpu_apic_address & PG_FRAME));
420 /* BSP does this itself, AP's get it pre-set */
421 gd = &CPU_prvspace[0].mdglobaldata;
422 gd->gd_CMAP1 = &SMPpt[1];
423 gd->gd_CMAP2 = &SMPpt[2];
424 gd->gd_CMAP3 = &SMPpt[3];
425 gd->gd_PMAP1 = &SMPpt[4];
426 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
427 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
428 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
429 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
436 * Set 4mb pdir for mp startup
441 if (pseflag && (cpu_feature & CPUID_PSE)) {
442 load_cr4(rcr4() | CR4_PSE);
443 if (pdir4mb && mycpu->gd_cpuid == 0) { /* only on BSP */
444 kernel_pmap->pm_pdir[KPTDI] =
445 PTD[KPTDI] = (pd_entry_t)pdir4mb;
453 * Initialize the pmap module.
454 * Called by vm_init, to initialize any structures that the pmap
455 * system needs to map virtual memory.
456 * pmap_init has been enhanced to support in a fairly consistant
457 * way, discontiguous physical memory.
460 pmap_init(phys_start, phys_end)
461 vm_offset_t phys_start, phys_end;
467 * object for kernel page table pages
469 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
472 * Allocate memory for random pmap data structures. Includes the
476 for(i = 0; i < vm_page_array_size; i++) {
479 m = &vm_page_array[i];
480 TAILQ_INIT(&m->md.pv_list);
481 m->md.pv_list_count = 0;
485 * init the pv free list
487 initial_pvs = vm_page_array_size;
488 if (initial_pvs < MINPV)
490 pvzone = &pvzone_store;
491 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
492 initial_pvs * sizeof (struct pv_entry));
493 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
497 * Now it is safe to enable pv_table recording.
499 pmap_initialized = TRUE;
503 * Initialize the address space (zone) for the pv_entries. Set a
504 * high water mark so that the system can recover from excessive
505 * numbers of pv entries.
510 int shpgperproc = PMAP_SHPGPERPROC;
512 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
513 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
514 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
515 pv_entry_high_water = 9 * (pv_entry_max / 10);
516 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
520 /***************************************************
521 * Low level helper routines.....
522 ***************************************************/
524 #if defined(PMAP_DIAGNOSTIC)
527 * This code checks for non-writeable/modified pages.
528 * This should be an invalid condition.
531 pmap_nw_modified(pt_entry_t ptea)
537 if ((pte & (PG_M|PG_RW)) == PG_M)
546 * this routine defines the region(s) of memory that should
547 * not be tested for the modified bit.
549 static PMAP_INLINE int
550 pmap_track_modified(vm_offset_t va)
552 if ((va < clean_sva) || (va >= clean_eva))
558 static PMAP_INLINE void
559 invltlb_1pg(vm_offset_t va)
561 #if defined(I386_CPU)
562 if (cpu_class == CPUCLASS_386) {
572 pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
575 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
576 cpu_invlpg((void *)va);
577 if (pmap->pm_active & mycpu->gd_other_cpus)
586 pmap_TLB_invalidate_all(pmap_t pmap)
589 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
591 if (pmap->pm_active & mycpu->gd_other_cpus)
603 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
605 /* are we current address space or kernel? */
606 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
607 return (unsigned *) PTmap;
609 /* otherwise, we are alternate address space */
610 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
611 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
613 /* The page directory is not shared between CPUs */
619 return (unsigned *) APTmap;
623 * Super fast pmap_pte routine best used when scanning
624 * the pv lists. This eliminates many coarse-grained
625 * invltlb calls. Note that many of the pv list
626 * scans are across different pmaps. It is very wasteful
627 * to do an entire invltlb for checking a single mapping.
631 pmap_pte_quick(pmap_t pmap, vm_offset_t va)
633 struct mdglobaldata *gd = mdcpu;
636 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
637 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
638 unsigned index = i386_btop(va);
639 /* are we current address space or kernel? */
640 if ((pmap == kernel_pmap) ||
641 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
642 return (unsigned *) PTmap + index;
644 newpf = pde & PG_FRAME;
645 if ( ((* (unsigned *) gd->gd_PMAP1) & PG_FRAME) != newpf) {
646 * (unsigned *) gd->gd_PMAP1 = newpf | PG_RW | PG_V;
647 cpu_invlpg(gd->gd_PADDR1);
649 return gd->gd_PADDR1 + ((unsigned) index & (NPTEPG - 1));
655 * Routine: pmap_extract
657 * Extract the physical page address associated
658 * with the given map/virtual_address pair.
661 pmap_extract(pmap, va)
662 register pmap_t pmap;
666 vm_offset_t pdirindex;
667 pdirindex = va >> PDRSHIFT;
668 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
670 if ((rtval & PG_PS) != 0) {
671 rtval &= ~(NBPDR - 1);
672 rtval |= va & (NBPDR - 1);
675 pte = get_ptbase(pmap) + i386_btop(va);
676 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
683 /***************************************************
684 * Low level mapping routines.....
685 ***************************************************/
688 * add a wired page to the kva
689 * note that in order for the mapping to take effect -- you
690 * should do a invltlb after doing the pmap_kenter...
695 register vm_offset_t pa;
697 register unsigned *pte;
700 npte = pa | PG_RW | PG_V | pgeflag;
701 pte = (unsigned *)vtopte(va);
708 * remove a page from the kernel pagetables
714 register unsigned *pte;
716 pte = (unsigned *)vtopte(va);
722 * Used to map a range of physical addresses into kernel
723 * virtual address space.
725 * For now, VM is already on, we only need to map the
729 pmap_map(virt, start, end, prot)
735 while (start < end) {
736 pmap_kenter(virt, start);
745 * Add a list of wired pages to the kva
746 * this routine is only used for temporary
747 * kernel mappings that do not need to have
748 * page modification or references recorded.
749 * Note that old mappings are simply written
750 * over. The page *must* be wired.
753 pmap_qenter(va, m, count)
760 end_va = va + count * PAGE_SIZE;
762 while (va < end_va) {
765 pte = (unsigned *)vtopte(va);
766 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
768 cpu_invlpg((void *)va);
781 * this routine jerks page mappings from the
782 * kernel -- it is meant only for temporary mappings.
785 pmap_qremove(va, count)
791 end_va = va + count*PAGE_SIZE;
793 while (va < end_va) {
796 pte = (unsigned *)vtopte(va);
799 cpu_invlpg((void *)va);
811 pmap_page_lookup(object, pindex)
817 m = vm_page_lookup(object, pindex);
818 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
824 * Create a new thread and optionally associate it with a (new) process.
827 pmap_init_thread(thread_t td)
829 td->td_pcb = (struct pcb *)(td->td_kstack + UPAGES * PAGE_SIZE) - 1;
830 td->td_sp = (char *)td->td_pcb - 16;
834 * Create the UPAGES for a new process.
835 * This routine directly affects the fork perf for a process.
838 pmap_init_proc(struct proc *p, struct thread *td)
840 p->p_addr = (void *)td->td_kstack;
843 td->td_switch = cpu_heavy_switch;
847 bzero(p->p_addr, sizeof(*p->p_addr));
851 * Dispose the UPAGES for a process that has exited.
852 * This routine directly impacts the exit perf of a process.
855 pmap_dispose_proc(struct proc *p)
859 KASSERT(p->p_lock == 0, ("attempt to dispose referenced proc! %p", p));
861 if ((td = p->p_thread) != NULL) {
870 * Allow the UPAGES for a process to be prejudicially paged out.
881 upobj = p->p_upages_obj;
883 * let the upages be paged
885 for(i=0;i<UPAGES;i++) {
886 if ((m = vm_page_lookup(upobj, i)) == NULL)
887 panic("pmap_swapout_proc: upage already missing???");
889 vm_page_unwire(m, 0);
890 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
896 * Bring the UPAGES for a specified process back in.
907 upobj = p->p_upages_obj;
908 for(i=0;i<UPAGES;i++) {
910 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
912 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
915 if (m->valid != VM_PAGE_BITS_ALL) {
916 rv = vm_pager_get_pages(upobj, &m, 1, 0);
917 if (rv != VM_PAGER_OK)
918 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
919 m = vm_page_lookup(upobj, i);
920 m->valid = VM_PAGE_BITS_ALL;
925 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
930 /***************************************************
931 * Page table page management routines.....
932 ***************************************************/
935 * This routine unholds page table pages, and if the hold count
936 * drops to zero, then it decrements the wire count.
939 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
941 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
944 if (m->hold_count == 0) {
947 * unmap the page table page
949 pmap->pm_pdir[m->pindex] = 0;
950 --pmap->pm_stats.resident_count;
951 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
952 (((unsigned) PTDpde) & PG_FRAME)) {
954 * Do a invltlb to make the invalidated mapping
955 * take effect immediately.
957 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
958 pmap_TLB_invalidate(pmap, pteva);
961 if (pmap->pm_ptphint == m)
962 pmap->pm_ptphint = NULL;
965 * If the page is finally unwired, simply free it.
968 if (m->wire_count == 0) {
972 vm_page_free_zero(m);
973 --vmstats.v_wire_count;
980 static PMAP_INLINE int
981 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
984 if (m->hold_count == 0)
985 return _pmap_unwire_pte_hold(pmap, m);
991 * After removing a page table entry, this routine is used to
992 * conditionally free the page, and manage the hold/wire counts.
995 pmap_unuse_pt(pmap, va, mpte)
1001 if (va >= UPT_MIN_ADDRESS)
1005 ptepindex = (va >> PDRSHIFT);
1006 if (pmap->pm_ptphint &&
1007 (pmap->pm_ptphint->pindex == ptepindex)) {
1008 mpte = pmap->pm_ptphint;
1010 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1011 pmap->pm_ptphint = mpte;
1015 return pmap_unwire_pte_hold(pmap, mpte);
1023 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1024 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1026 pmap->pm_active = 0;
1027 pmap->pm_ptphint = NULL;
1028 TAILQ_INIT(&pmap->pm_pvlist);
1029 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1033 * Initialize a preallocated and zeroed pmap structure,
1034 * such as one in a vmspace structure.
1038 register struct pmap *pmap;
1043 * No need to allocate page table space yet but we do need a valid
1044 * page directory table.
1046 if (pmap->pm_pdir == NULL)
1048 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1051 * allocate object for the ptes
1053 if (pmap->pm_pteobj == NULL)
1054 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1057 * allocate the page directory page
1059 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1060 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1062 ptdpg->wire_count = 1;
1063 ++vmstats.v_wire_count;
1066 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1067 ptdpg->valid = VM_PAGE_BITS_ALL;
1069 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1070 if ((ptdpg->flags & PG_ZERO) == 0)
1071 bzero(pmap->pm_pdir, PAGE_SIZE);
1073 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1075 /* install self-referential address mapping entry */
1076 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1077 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1080 pmap->pm_active = 0;
1081 pmap->pm_ptphint = NULL;
1082 TAILQ_INIT(&pmap->pm_pvlist);
1083 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1087 * Wire in kernel global address entries. To avoid a race condition
1088 * between pmap initialization and pmap_growkernel, this procedure
1089 * should be called after the vmspace is attached to the process
1090 * but before this pmap is activated.
1096 /* XXX copies current process, does not fill in MPPTDI */
1097 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1101 pmap_release_free_page(pmap, p)
1105 unsigned *pde = (unsigned *) pmap->pm_pdir;
1107 * This code optimizes the case of freeing non-busy
1108 * page-table pages. Those pages are zero now, and
1109 * might as well be placed directly into the zero queue.
1111 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1117 * Remove the page table page from the processes address space.
1120 pmap->pm_stats.resident_count--;
1122 if (p->hold_count) {
1123 panic("pmap_release: freeing held page table page");
1126 * Page directory pages need to have the kernel
1127 * stuff cleared, so they can go into the zero queue also.
1129 if (p->pindex == PTDPTDI) {
1130 bzero(pde + KPTDI, nkpt * PTESIZE);
1133 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1136 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1137 pmap->pm_ptphint = NULL;
1140 vmstats.v_wire_count--;
1141 vm_page_free_zero(p);
1146 * this routine is called if the page table page is not
1150 _pmap_allocpte(pmap, ptepindex)
1154 vm_offset_t pteva, ptepa;
1158 * Find or fabricate a new pagetable page
1160 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1161 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1163 KASSERT(m->queue == PQ_NONE,
1164 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1166 if (m->wire_count == 0)
1167 vmstats.v_wire_count++;
1171 * Increment the hold count for the page table page
1172 * (denoting a new mapping.)
1177 * Map the pagetable page into the process address space, if
1178 * it isn't already there.
1181 pmap->pm_stats.resident_count++;
1183 ptepa = VM_PAGE_TO_PHYS(m);
1184 pmap->pm_pdir[ptepindex] =
1185 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1188 * Set the page table hint
1190 pmap->pm_ptphint = m;
1193 * Try to use the new mapping, but if we cannot, then
1194 * do it with the routine that maps the page explicitly.
1196 if ((m->flags & PG_ZERO) == 0) {
1197 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1198 (((unsigned) PTDpde) & PG_FRAME)) {
1199 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1200 bzero((caddr_t) pteva, PAGE_SIZE);
1202 pmap_zero_page(ptepa);
1206 m->valid = VM_PAGE_BITS_ALL;
1207 vm_page_flag_clear(m, PG_ZERO);
1208 vm_page_flag_set(m, PG_MAPPED);
1215 pmap_allocpte(pmap, va)
1224 * Calculate pagetable page index
1226 ptepindex = va >> PDRSHIFT;
1229 * Get the page directory entry
1231 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1234 * This supports switching from a 4MB page to a
1237 if (ptepa & PG_PS) {
1238 pmap->pm_pdir[ptepindex] = 0;
1244 * If the page table page is mapped, we just increment the
1245 * hold count, and activate it.
1249 * In order to get the page table page, try the
1252 if (pmap->pm_ptphint &&
1253 (pmap->pm_ptphint->pindex == ptepindex)) {
1254 m = pmap->pm_ptphint;
1256 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1257 pmap->pm_ptphint = m;
1263 * Here if the pte page isn't mapped, or if it has been deallocated.
1265 return _pmap_allocpte(pmap, ptepindex);
1269 /***************************************************
1270 * Pmap allocation/deallocation routines.
1271 ***************************************************/
1274 * Release any resources held by the given physical map.
1275 * Called when a pmap initialized by pmap_pinit is being released.
1276 * Should only be called if the map contains no valid mappings.
1280 register struct pmap *pmap;
1282 vm_page_t p,n,ptdpg;
1283 vm_object_t object = pmap->pm_pteobj;
1286 #if defined(DIAGNOSTIC)
1287 if (object->ref_count != 1)
1288 panic("pmap_release: pteobj reference count != 1");
1293 curgeneration = object->generation;
1294 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1295 n = TAILQ_NEXT(p, listq);
1296 if (p->pindex == PTDPTDI) {
1301 if (!pmap_release_free_page(pmap, p) &&
1302 (object->generation != curgeneration))
1307 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1312 kvm_size(SYSCTL_HANDLER_ARGS)
1314 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1316 return sysctl_handle_long(oidp, &ksize, 0, req);
1318 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1319 0, 0, kvm_size, "IU", "Size of KVM");
1322 kvm_free(SYSCTL_HANDLER_ARGS)
1324 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1326 return sysctl_handle_long(oidp, &kfree, 0, req);
1328 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1329 0, 0, kvm_free, "IU", "Amount of KVM free");
1332 * grow the number of kernel page table entries, if needed
1335 pmap_growkernel(vm_offset_t addr)
1340 vm_offset_t ptppaddr;
1345 if (kernel_vm_end == 0) {
1346 kernel_vm_end = KERNBASE;
1348 while (pdir_pde(PTD, kernel_vm_end)) {
1349 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1353 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1354 while (kernel_vm_end < addr) {
1355 if (pdir_pde(PTD, kernel_vm_end)) {
1356 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1361 * This index is bogus, but out of the way
1363 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1365 panic("pmap_growkernel: no memory to grow kernel");
1370 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1371 pmap_zero_page(ptppaddr);
1372 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1373 pdir_pde(PTD, kernel_vm_end) = newpdir;
1375 LIST_FOREACH(p, &allproc, p_list) {
1377 pmap = vmspace_pmap(p->p_vmspace);
1378 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1381 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1382 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1388 * Retire the given physical map from service.
1389 * Should only be called if the map contains
1390 * no valid mappings.
1394 register pmap_t pmap;
1401 count = --pmap->pm_count;
1404 panic("destroying a pmap is not yet implemented");
1409 * Add a reference to the specified pmap.
1412 pmap_reference(pmap)
1420 /***************************************************
1421 * page management routines.
1422 ***************************************************/
1425 * free the pv_entry back to the free list. This function may be
1426 * called from an interrupt.
1428 static PMAP_INLINE void
1437 * get a new pv_entry, allocating a block from the system
1438 * when needed. This function may be called from an interrupt.
1444 if (pv_entry_high_water &&
1445 (pv_entry_count > pv_entry_high_water) &&
1446 (pmap_pagedaemon_waken == 0)) {
1447 pmap_pagedaemon_waken = 1;
1448 wakeup (&vm_pages_needed);
1450 return zalloc(pvzone);
1454 * This routine is very drastic, but can save the system
1462 static int warningdone=0;
1464 if (pmap_pagedaemon_waken == 0)
1467 if (warningdone < 5) {
1468 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1472 for(i = 0; i < vm_page_array_size; i++) {
1473 m = &vm_page_array[i];
1474 if (m->wire_count || m->hold_count || m->busy ||
1475 (m->flags & PG_BUSY))
1479 pmap_pagedaemon_waken = 0;
1484 * If it is the first entry on the list, it is actually
1485 * in the header and we must copy the following entry up
1486 * to the header. Otherwise we must search the list for
1487 * the entry. In either case we free the now unused entry.
1491 pmap_remove_entry(pmap, m, va)
1501 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1502 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1503 if (pmap == pv->pv_pmap && va == pv->pv_va)
1507 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1508 if (va == pv->pv_va)
1516 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1517 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1518 m->md.pv_list_count--;
1519 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1520 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1522 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1531 * Create a pv entry for page at pa for
1535 pmap_insert_entry(pmap, va, mpte, m)
1546 pv = get_pv_entry();
1551 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1552 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1553 m->md.pv_list_count++;
1559 * pmap_remove_pte: do the things to unmap a page in a process
1562 pmap_remove_pte(pmap, ptq, va)
1570 oldpte = loadandclear(ptq);
1572 pmap->pm_stats.wired_count -= 1;
1574 * Machines that don't support invlpg, also don't support
1579 pmap->pm_stats.resident_count -= 1;
1580 if (oldpte & PG_MANAGED) {
1581 m = PHYS_TO_VM_PAGE(oldpte);
1582 if (oldpte & PG_M) {
1583 #if defined(PMAP_DIAGNOSTIC)
1584 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1586 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1590 if (pmap_track_modified(va))
1594 vm_page_flag_set(m, PG_REFERENCED);
1595 return pmap_remove_entry(pmap, m, va);
1597 return pmap_unuse_pt(pmap, va, NULL);
1604 * Remove a single page from a process address space
1607 pmap_remove_page(pmap, va)
1609 register vm_offset_t va;
1611 register unsigned *ptq;
1614 * if there is no pte for this address, just skip it!!!
1616 if (*pmap_pde(pmap, va) == 0) {
1621 * get a local va for mappings for this pmap.
1623 ptq = get_ptbase(pmap) + i386_btop(va);
1625 (void) pmap_remove_pte(pmap, ptq, va);
1626 pmap_TLB_invalidate(pmap, va);
1632 * Remove the given range of addresses from the specified map.
1634 * It is assumed that the start and end are properly
1635 * rounded to the page size.
1638 pmap_remove(pmap, sva, eva)
1640 register vm_offset_t sva;
1641 register vm_offset_t eva;
1643 register unsigned *ptbase;
1645 vm_offset_t ptpaddr;
1646 vm_offset_t sindex, eindex;
1652 if (pmap->pm_stats.resident_count == 0)
1656 * special handling of removing one page. a very
1657 * common operation and easy to short circuit some
1660 if (((sva + PAGE_SIZE) == eva) &&
1661 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1662 pmap_remove_page(pmap, sva);
1669 * Get a local virtual address for the mappings that are being
1672 ptbase = get_ptbase(pmap);
1674 sindex = i386_btop(sva);
1675 eindex = i386_btop(eva);
1677 for (; sindex < eindex; sindex = pdnxt) {
1681 * Calculate index for next page table.
1683 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1684 if (pmap->pm_stats.resident_count == 0)
1687 pdirindex = sindex / NPDEPG;
1688 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1689 pmap->pm_pdir[pdirindex] = 0;
1690 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1696 * Weed out invalid mappings. Note: we assume that the page
1697 * directory table is always allocated, and in kernel virtual.
1703 * Limit our scan to either the end of the va represented
1704 * by the current page table page, or to the end of the
1705 * range being removed.
1707 if (pdnxt > eindex) {
1711 for ( ;sindex != pdnxt; sindex++) {
1713 if (ptbase[sindex] == 0) {
1716 va = i386_ptob(sindex);
1719 if (pmap_remove_pte(pmap,
1720 ptbase + sindex, va))
1726 pmap_TLB_invalidate_all(pmap);
1730 * Routine: pmap_remove_all
1732 * Removes this physical page from
1733 * all physical maps in which it resides.
1734 * Reflects back modify bits to the pager.
1737 * Original versions of this routine were very
1738 * inefficient because they iteratively called
1739 * pmap_remove (slow...)
1746 register pv_entry_t pv;
1747 register unsigned *pte, tpte;
1750 #if defined(PMAP_DIAGNOSTIC)
1752 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1755 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1756 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1761 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1762 pv->pv_pmap->pm_stats.resident_count--;
1764 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1766 tpte = loadandclear(pte);
1768 pv->pv_pmap->pm_stats.wired_count--;
1771 vm_page_flag_set(m, PG_REFERENCED);
1774 * Update the vm_page_t clean and reference bits.
1777 #if defined(PMAP_DIAGNOSTIC)
1778 if (pmap_nw_modified((pt_entry_t) tpte)) {
1780 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1784 if (pmap_track_modified(pv->pv_va))
1787 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1789 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1790 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1791 m->md.pv_list_count--;
1792 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1796 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1802 * Set the physical protection on the
1803 * specified range of this map as requested.
1806 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1808 register unsigned *ptbase;
1809 vm_offset_t pdnxt, ptpaddr;
1810 vm_pindex_t sindex, eindex;
1816 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1817 pmap_remove(pmap, sva, eva);
1821 if (prot & VM_PROT_WRITE)
1826 ptbase = get_ptbase(pmap);
1828 sindex = i386_btop(sva);
1829 eindex = i386_btop(eva);
1831 for (; sindex < eindex; sindex = pdnxt) {
1835 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1837 pdirindex = sindex / NPDEPG;
1838 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1839 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1840 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1846 * Weed out invalid mappings. Note: we assume that the page
1847 * directory table is always allocated, and in kernel virtual.
1852 if (pdnxt > eindex) {
1856 for (; sindex != pdnxt; sindex++) {
1861 pbits = ptbase[sindex];
1863 if (pbits & PG_MANAGED) {
1866 m = PHYS_TO_VM_PAGE(pbits);
1867 vm_page_flag_set(m, PG_REFERENCED);
1871 if (pmap_track_modified(i386_ptob(sindex))) {
1873 m = PHYS_TO_VM_PAGE(pbits);
1882 if (pbits != ptbase[sindex]) {
1883 ptbase[sindex] = pbits;
1889 pmap_TLB_invalidate_all(pmap);
1893 * Insert the given physical page (p) at
1894 * the specified virtual address (v) in the
1895 * target physical map with the protection requested.
1897 * If specified, the page will be wired down, meaning
1898 * that the related pte can not be reclaimed.
1900 * NB: This is the only routine which MAY NOT lazy-evaluate
1901 * or lose information. That is, this routine must actually
1902 * insert this page into the given map NOW.
1905 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1909 register unsigned *pte;
1911 vm_offset_t origpte, newpte;
1918 #ifdef PMAP_DIAGNOSTIC
1919 if (va > VM_MAX_KERNEL_ADDRESS)
1920 panic("pmap_enter: toobig");
1921 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
1922 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
1927 * In the case that a page table page is not
1928 * resident, we are creating it here.
1930 if (va < UPT_MIN_ADDRESS) {
1931 mpte = pmap_allocpte(pmap, va);
1933 #if 0 && defined(PMAP_DIAGNOSTIC)
1935 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
1936 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
1937 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
1938 pmap->pm_pdir[PTDPTDI], origpte, va);
1941 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
1942 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
1943 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
1944 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
1945 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
1946 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
1947 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
1953 pte = pmap_pte(pmap, va);
1956 * Page Directory table entry not valid, we need a new PT page
1959 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
1960 (void *)pmap->pm_pdir[PTDPTDI], va);
1963 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
1964 origpte = *(vm_offset_t *)pte;
1965 opa = origpte & PG_FRAME;
1967 if (origpte & PG_PS)
1968 panic("pmap_enter: attempted pmap_enter on 4MB page");
1971 * Mapping has not changed, must be protection or wiring change.
1973 if (origpte && (opa == pa)) {
1975 * Wiring change, just update stats. We don't worry about
1976 * wiring PT pages as they remain resident as long as there
1977 * are valid mappings in them. Hence, if a user page is wired,
1978 * the PT page will be also.
1980 if (wired && ((origpte & PG_W) == 0))
1981 pmap->pm_stats.wired_count++;
1982 else if (!wired && (origpte & PG_W))
1983 pmap->pm_stats.wired_count--;
1985 #if defined(PMAP_DIAGNOSTIC)
1986 if (pmap_nw_modified((pt_entry_t) origpte)) {
1988 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
1994 * Remove extra pte reference
1999 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
2000 if ((origpte & PG_RW) == 0) {
2003 cpu_invlpg((void *)va);
2004 if (pmap->pm_active & mycpu->gd_other_cpus)
2014 * We might be turning off write access to the page,
2015 * so we go ahead and sense modify status.
2017 if (origpte & PG_MANAGED) {
2018 if ((origpte & PG_M) && pmap_track_modified(va)) {
2020 om = PHYS_TO_VM_PAGE(opa);
2028 * Mapping has changed, invalidate old range and fall through to
2029 * handle validating new mapping.
2033 err = pmap_remove_pte(pmap, pte, va);
2035 panic("pmap_enter: pte vanished, va: 0x%x", va);
2039 * Enter on the PV list if part of our managed memory. Note that we
2040 * raise IPL while manipulating pv_table since pmap_enter can be
2041 * called at interrupt time.
2043 if (pmap_initialized &&
2044 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2045 pmap_insert_entry(pmap, va, mpte, m);
2050 * Increment counters
2052 pmap->pm_stats.resident_count++;
2054 pmap->pm_stats.wired_count++;
2058 * Now validate mapping with desired protection/wiring.
2060 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2064 if (va < UPT_MIN_ADDRESS)
2066 if (pmap == kernel_pmap)
2070 * if the mapping or permission bits are different, we need
2071 * to update the pte.
2073 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2074 *pte = newpte | PG_A;
2077 cpu_invlpg((void *)va);
2078 if (pmap->pm_active & mycpu->gd_other_cpus)
2088 * this code makes some *MAJOR* assumptions:
2089 * 1. Current pmap & pmap exists.
2092 * 4. No page table pages.
2093 * 5. Tlbflush is deferred to calling procedure.
2094 * 6. Page IS managed.
2095 * but is *MUCH* faster than pmap_enter...
2099 pmap_enter_quick(pmap, va, m, mpte)
2100 register pmap_t pmap;
2109 * In the case that a page table page is not
2110 * resident, we are creating it here.
2112 if (va < UPT_MIN_ADDRESS) {
2117 * Calculate pagetable page index
2119 ptepindex = va >> PDRSHIFT;
2120 if (mpte && (mpte->pindex == ptepindex)) {
2125 * Get the page directory entry
2127 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2130 * If the page table page is mapped, we just increment
2131 * the hold count, and activate it.
2135 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2136 if (pmap->pm_ptphint &&
2137 (pmap->pm_ptphint->pindex == ptepindex)) {
2138 mpte = pmap->pm_ptphint;
2140 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2141 pmap->pm_ptphint = mpte;
2147 mpte = _pmap_allocpte(pmap, ptepindex);
2155 * This call to vtopte makes the assumption that we are
2156 * entering the page into the current pmap. In order to support
2157 * quick entry into any pmap, one would likely use pmap_pte_quick.
2158 * But that isn't as quick as vtopte.
2160 pte = (unsigned *)vtopte(va);
2163 pmap_unwire_pte_hold(pmap, mpte);
2168 * Enter on the PV list if part of our managed memory. Note that we
2169 * raise IPL while manipulating pv_table since pmap_enter can be
2170 * called at interrupt time.
2172 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2173 pmap_insert_entry(pmap, va, mpte, m);
2176 * Increment counters
2178 pmap->pm_stats.resident_count++;
2180 pa = VM_PAGE_TO_PHYS(m);
2183 * Now validate mapping with RO protection
2185 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2186 *pte = pa | PG_V | PG_U;
2188 *pte = pa | PG_V | PG_U | PG_MANAGED;
2194 * Make a temporary mapping for a physical address. This is only intended
2195 * to be used for panic dumps.
2198 pmap_kenter_temporary(vm_offset_t pa, int i)
2200 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2201 return ((void *)crashdumpmap);
2204 #define MAX_INIT_PT (96)
2206 * pmap_object_init_pt preloads the ptes for a given object
2207 * into the specified pmap. This eliminates the blast of soft
2208 * faults on process startup and immediately after an mmap.
2211 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2224 if (pmap == NULL || object == NULL)
2228 * This code maps large physical mmap regions into the
2229 * processor address space. Note that some shortcuts
2230 * are taken, but the code works.
2233 (object->type == OBJT_DEVICE) &&
2234 ((addr & (NBPDR - 1)) == 0) &&
2235 ((size & (NBPDR - 1)) == 0) ) {
2238 unsigned int ptepindex;
2242 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2246 p = vm_page_lookup(object, pindex);
2247 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2251 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2256 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2261 p = vm_page_lookup(object, pindex);
2265 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2266 if (ptepa & (NBPDR - 1)) {
2270 p->valid = VM_PAGE_BITS_ALL;
2272 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2273 npdes = size >> PDRSHIFT;
2274 for(i=0;i<npdes;i++) {
2275 pmap->pm_pdir[ptepindex] =
2276 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2280 vm_page_flag_set(p, PG_MAPPED);
2285 psize = i386_btop(size);
2287 if ((object->type != OBJT_VNODE) ||
2288 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2289 (object->resident_page_count > MAX_INIT_PT))) {
2293 if (psize + pindex > object->size) {
2294 if (object->size < pindex)
2296 psize = object->size - pindex;
2301 * if we are processing a major portion of the object, then scan the
2304 if (psize > (object->resident_page_count >> 2)) {
2307 for (p = TAILQ_FIRST(&object->memq);
2308 ((objpgs > 0) && (p != NULL));
2309 p = TAILQ_NEXT(p, listq)) {
2312 if (tmpidx < pindex) {
2316 if (tmpidx >= psize) {
2320 * don't allow an madvise to blow away our really
2321 * free pages allocating pv entries.
2323 if ((limit & MAP_PREFAULT_MADVISE) &&
2324 vmstats.v_free_count < vmstats.v_free_reserved) {
2327 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2329 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2330 if ((p->queue - p->pc) == PQ_CACHE)
2331 vm_page_deactivate(p);
2333 mpte = pmap_enter_quick(pmap,
2334 addr + i386_ptob(tmpidx), p, mpte);
2335 vm_page_flag_set(p, PG_MAPPED);
2342 * else lookup the pages one-by-one.
2344 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2346 * don't allow an madvise to blow away our really
2347 * free pages allocating pv entries.
2349 if ((limit & MAP_PREFAULT_MADVISE) &&
2350 vmstats.v_free_count < vmstats.v_free_reserved) {
2353 p = vm_page_lookup(object, tmpidx + pindex);
2355 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2357 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2358 if ((p->queue - p->pc) == PQ_CACHE)
2359 vm_page_deactivate(p);
2361 mpte = pmap_enter_quick(pmap,
2362 addr + i386_ptob(tmpidx), p, mpte);
2363 vm_page_flag_set(p, PG_MAPPED);
2372 * pmap_prefault provides a quick way of clustering
2373 * pagefaults into a processes address space. It is a "cousin"
2374 * of pmap_object_init_pt, except it runs at page fault time instead
2379 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2381 static int pmap_prefault_pageorder[] = {
2382 -PAGE_SIZE, PAGE_SIZE,
2383 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2384 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2385 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2389 pmap_prefault(pmap, addra, entry)
2392 vm_map_entry_t entry;
2401 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2404 object = entry->object.vm_object;
2406 starta = addra - PFBAK * PAGE_SIZE;
2407 if (starta < entry->start) {
2408 starta = entry->start;
2409 } else if (starta > addra) {
2414 for (i = 0; i < PAGEORDER_SIZE; i++) {
2415 vm_object_t lobject;
2418 addr = addra + pmap_prefault_pageorder[i];
2419 if (addr > addra + (PFFOR * PAGE_SIZE))
2422 if (addr < starta || addr >= entry->end)
2425 if ((*pmap_pde(pmap, addr)) == NULL)
2428 pte = (unsigned *) vtopte(addr);
2432 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2434 for (m = vm_page_lookup(lobject, pindex);
2435 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2436 lobject = lobject->backing_object) {
2437 if (lobject->backing_object_offset & PAGE_MASK)
2439 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2440 m = vm_page_lookup(lobject->backing_object, pindex);
2444 * give-up when a page is not in memory
2449 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2451 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2453 if ((m->queue - m->pc) == PQ_CACHE) {
2454 vm_page_deactivate(m);
2457 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2458 vm_page_flag_set(m, PG_MAPPED);
2465 * Routine: pmap_change_wiring
2466 * Function: Change the wiring attribute for a map/virtual-address
2468 * In/out conditions:
2469 * The mapping must already exist in the pmap.
2472 pmap_change_wiring(pmap, va, wired)
2473 register pmap_t pmap;
2477 register unsigned *pte;
2482 pte = pmap_pte(pmap, va);
2484 if (wired && !pmap_pte_w(pte))
2485 pmap->pm_stats.wired_count++;
2486 else if (!wired && pmap_pte_w(pte))
2487 pmap->pm_stats.wired_count--;
2490 * Wiring is not a hardware characteristic so there is no need to
2493 pmap_pte_set_w(pte, wired);
2499 * Copy the range specified by src_addr/len
2500 * from the source map to the range dst_addr/len
2501 * in the destination map.
2503 * This routine is only advisory and need not do anything.
2507 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2508 pmap_t dst_pmap, src_pmap;
2509 vm_offset_t dst_addr;
2511 vm_offset_t src_addr;
2514 vm_offset_t end_addr = src_addr + len;
2516 unsigned src_frame, dst_frame;
2519 if (dst_addr != src_addr)
2522 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2523 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2527 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2528 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2529 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2531 /* The page directory is not shared between CPUs */
2538 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2539 unsigned *src_pte, *dst_pte;
2540 vm_page_t dstmpte, srcmpte;
2541 vm_offset_t srcptepaddr;
2544 if (addr >= UPT_MIN_ADDRESS)
2545 panic("pmap_copy: invalid to pmap_copy page tables\n");
2548 * Don't let optional prefaulting of pages make us go
2549 * way below the low water mark of free pages or way
2550 * above high water mark of used pv entries.
2552 if (vmstats.v_free_count < vmstats.v_free_reserved ||
2553 pv_entry_count > pv_entry_high_water)
2556 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2557 ptepindex = addr >> PDRSHIFT;
2559 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2560 if (srcptepaddr == 0)
2563 if (srcptepaddr & PG_PS) {
2564 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2565 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2566 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2571 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2572 if ((srcmpte == NULL) ||
2573 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2576 if (pdnxt > end_addr)
2579 src_pte = (unsigned *) vtopte(addr);
2580 dst_pte = (unsigned *) avtopte(addr);
2581 while (addr < pdnxt) {
2585 * we only virtual copy managed pages
2587 if ((ptetemp & PG_MANAGED) != 0) {
2589 * We have to check after allocpte for the
2590 * pte still being around... allocpte can
2593 dstmpte = pmap_allocpte(dst_pmap, addr);
2594 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2596 * Clear the modified and
2597 * accessed (referenced) bits
2600 m = PHYS_TO_VM_PAGE(ptetemp);
2601 *dst_pte = ptetemp & ~(PG_M | PG_A);
2602 dst_pmap->pm_stats.resident_count++;
2603 pmap_insert_entry(dst_pmap, addr,
2606 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2608 if (dstmpte->hold_count >= srcmpte->hold_count)
2619 * Routine: pmap_kernel
2621 * Returns the physical map handle for the kernel.
2626 return (kernel_pmap);
2630 * pmap_zero_page zeros the specified hardware page by mapping
2631 * the page into KVM and using bzero to clear its contents.
2634 pmap_zero_page(vm_offset_t phys)
2636 struct mdglobaldata *gd = mdcpu;
2638 if (*(int *)gd->gd_CMAP3)
2639 panic("pmap_zero_page: CMAP3 busy");
2641 *(int *)gd->gd_CMAP3 =
2642 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2643 cpu_invlpg(gd->gd_CADDR3);
2645 #if defined(I686_CPU)
2646 if (cpu_class == CPUCLASS_686)
2647 i686_pagezero(gd->gd_CADDR3);
2650 bzero(gd->gd_CADDR3, PAGE_SIZE);
2652 *(int *) gd->gd_CMAP3 = 0;
2656 * pmap_zero_page_area zeros the specified hardware page by mapping
2657 * the page into KVM and using bzero to clear its contents.
2659 * off and size may not cover an area beyond a single hardware page.
2662 pmap_zero_page_area(phys, off, size)
2667 struct mdglobaldata *gd = mdcpu;
2669 if (*(int *) gd->gd_CMAP3)
2670 panic("pmap_zero_page: CMAP3 busy");
2672 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2673 cpu_invlpg(gd->gd_CADDR3);
2675 #if defined(I686_CPU)
2676 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2677 i686_pagezero(gd->gd_CADDR3);
2680 bzero((char *)gd->gd_CADDR3 + off, size);
2682 *(int *) gd->gd_CMAP3 = 0;
2686 * pmap_copy_page copies the specified (machine independent)
2687 * page by mapping the page into virtual memory and using
2688 * bcopy to copy the page, one machine dependent page at a
2692 pmap_copy_page(src, dst)
2696 struct mdglobaldata *gd = mdcpu;
2698 if (*(int *) gd->gd_CMAP1)
2699 panic("pmap_copy_page: CMAP1 busy");
2700 if (*(int *) gd->gd_CMAP2)
2701 panic("pmap_copy_page: CMAP2 busy");
2703 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2704 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2706 cpu_invlpg(gd->gd_CADDR1);
2707 cpu_invlpg(gd->gd_CADDR2);
2709 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
2711 *(int *) gd->gd_CMAP1 = 0;
2712 *(int *) gd->gd_CMAP2 = 0;
2717 * Routine: pmap_pageable
2719 * Make the specified pages (by pmap, offset)
2720 * pageable (or not) as requested.
2722 * A page which is not pageable may not take
2723 * a fault; therefore, its page table entry
2724 * must remain valid for the duration.
2726 * This routine is merely advisory; pmap_enter
2727 * will specify that these pages are to be wired
2728 * down (or not) as appropriate.
2731 pmap_pageable(pmap, sva, eva, pageable)
2733 vm_offset_t sva, eva;
2739 * Returns true if the pmap's pv is one of the first
2740 * 16 pvs linked to from this page. This count may
2741 * be changed upwards or downwards in the future; it
2742 * is only necessary that true be returned for a small
2743 * subset of pmaps for proper page aging.
2746 pmap_page_exists_quick(pmap, m)
2754 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2759 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2760 if (pv->pv_pmap == pmap) {
2772 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2774 * Remove all pages from specified address space
2775 * this aids process exit speeds. Also, this code
2776 * is special cased for current process only, but
2777 * can have the more generic (and slightly slower)
2778 * mode enabled. This is much faster than pmap_remove
2779 * in the case of running down an entire address space.
2782 pmap_remove_pages(pmap, sva, eva)
2784 vm_offset_t sva, eva;
2786 unsigned *pte, tpte;
2791 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2792 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2793 printf("warning: pmap_remove_pages called with non-current pmap\n");
2799 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2803 if (pv->pv_va >= eva || pv->pv_va < sva) {
2804 npv = TAILQ_NEXT(pv, pv_plist);
2808 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2809 pte = (unsigned *)vtopte(pv->pv_va);
2811 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2816 * We cannot remove wired pages from a process' mapping at this time
2819 npv = TAILQ_NEXT(pv, pv_plist);
2824 m = PHYS_TO_VM_PAGE(tpte);
2826 KASSERT(m < &vm_page_array[vm_page_array_size],
2827 ("pmap_remove_pages: bad tpte %x", tpte));
2829 pv->pv_pmap->pm_stats.resident_count--;
2832 * Update the vm_page_t clean and reference bits.
2839 npv = TAILQ_NEXT(pv, pv_plist);
2840 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2842 m->md.pv_list_count--;
2843 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2844 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2845 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2848 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2852 pmap_TLB_invalidate_all(pmap);
2856 * pmap_testbit tests bits in pte's
2857 * note that the testbit/changebit routines are inline,
2858 * and a lot of things compile-time evaluate.
2861 pmap_testbit(m, bit)
2869 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2872 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
2877 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2879 * if the bit being tested is the modified bit, then
2880 * mark clean_map and ptes as never
2883 if (bit & (PG_A|PG_M)) {
2884 if (!pmap_track_modified(pv->pv_va))
2888 #if defined(PMAP_DIAGNOSTIC)
2890 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
2894 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2905 * this routine is used to modify bits in ptes
2907 static __inline void
2908 pmap_changebit(m, bit, setem)
2913 register pv_entry_t pv;
2914 register unsigned *pte;
2917 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2923 * Loop over all current mappings setting/clearing as appropos If
2924 * setting RO do we need to clear the VAC?
2926 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2928 * don't write protect pager mappings
2930 if (!setem && (bit == PG_RW)) {
2931 if (!pmap_track_modified(pv->pv_va))
2935 #if defined(PMAP_DIAGNOSTIC)
2937 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
2942 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2946 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2948 vm_offset_t pbits = *(vm_offset_t *)pte;
2954 *(int *)pte = pbits & ~(PG_M|PG_RW);
2956 *(int *)pte = pbits & ~bit;
2958 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2966 * pmap_page_protect:
2968 * Lower the permission for all mappings to a given page.
2971 pmap_page_protect(vm_page_t m, vm_prot_t prot)
2973 if ((prot & VM_PROT_WRITE) == 0) {
2974 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
2975 pmap_changebit(m, PG_RW, FALSE);
2983 pmap_phys_address(ppn)
2986 return (i386_ptob(ppn));
2990 * pmap_ts_referenced:
2992 * Return a count of reference bits for a page, clearing those bits.
2993 * It is not necessary for every reference bit to be cleared, but it
2994 * is necessary that 0 only be returned when there are truly no
2995 * reference bits set.
2997 * XXX: The exact number of bits to check and clear is a matter that
2998 * should be tested and standardized at some point in the future for
2999 * optimal aging of shared pages.
3002 pmap_ts_referenced(vm_page_t m)
3004 register pv_entry_t pv, pvf, pvn;
3009 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3014 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3019 pvn = TAILQ_NEXT(pv, pv_list);
3021 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3023 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3025 if (!pmap_track_modified(pv->pv_va))
3028 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3030 if (pte && (*pte & PG_A)) {
3033 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3040 } while ((pv = pvn) != NULL && pv != pvf);
3050 * Return whether or not the specified physical page was modified
3051 * in any physical maps.
3054 pmap_is_modified(vm_page_t m)
3056 return pmap_testbit(m, PG_M);
3060 * Clear the modify bits on the specified physical page.
3063 pmap_clear_modify(vm_page_t m)
3065 pmap_changebit(m, PG_M, FALSE);
3069 * pmap_clear_reference:
3071 * Clear the reference bit on the specified physical page.
3074 pmap_clear_reference(vm_page_t m)
3076 pmap_changebit(m, PG_A, FALSE);
3080 * Miscellaneous support routines follow
3084 i386_protection_init()
3086 register int *kp, prot;
3088 kp = protection_codes;
3089 for (prot = 0; prot < 8; prot++) {
3091 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3093 * Read access is also 0. There isn't any execute bit,
3094 * so just make it readable.
3096 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3097 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3098 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3101 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3102 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3103 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3104 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3112 * Map a set of physical memory pages into the kernel virtual
3113 * address space. Return a pointer to where it is mapped. This
3114 * routine is intended to be used for mapping device memory,
3117 * NOTE: we can't use pgeflag unless we invalidate the pages one at
3121 pmap_mapdev(pa, size)
3125 vm_offset_t va, tmpva, offset;
3128 offset = pa & PAGE_MASK;
3129 size = roundup(offset + size, PAGE_SIZE);
3131 va = kmem_alloc_pageable(kernel_map, size);
3133 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3136 for (tmpva = va; size > 0;) {
3137 pte = (unsigned *)vtopte(tmpva);
3138 *pte = pa | PG_RW | PG_V; /* | pgeflag; */
3145 return ((void *)(va + offset));
3149 pmap_unmapdev(va, size)
3153 vm_offset_t base, offset;
3155 base = va & PG_FRAME;
3156 offset = va & PAGE_MASK;
3157 size = roundup(offset + size, PAGE_SIZE);
3158 kmem_free(kernel_map, base, size);
3162 * perform the pmap work for mincore
3165 pmap_mincore(pmap, addr)
3170 unsigned *ptep, pte;
3174 ptep = pmap_pte(pmap, addr);
3179 if ((pte = *ptep) != 0) {
3182 val = MINCORE_INCORE;
3183 if ((pte & PG_MANAGED) == 0)
3186 pa = pte & PG_FRAME;
3188 m = PHYS_TO_VM_PAGE(pa);
3194 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3196 * Modified by someone
3198 else if (m->dirty || pmap_is_modified(m))
3199 val |= MINCORE_MODIFIED_OTHER;
3204 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3207 * Referenced by someone
3209 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3210 val |= MINCORE_REFERENCED_OTHER;
3211 vm_page_flag_set(m, PG_REFERENCED);
3218 pmap_activate(struct proc *p)
3222 pmap = vmspace_pmap(p->p_vmspace);
3224 atomic_set_int(&pmap->pm_active, 1 << mycpu->gd_cpuid);
3226 pmap->pm_active |= 1;
3228 #if defined(SWTCH_OPTIM_STATS)
3231 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3232 load_cr3(p->p_thread->td_pcb->pcb_cr3);
3236 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3239 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3243 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3248 #if defined(PMAP_DEBUG)
3249 pmap_pid_dump(int pid)
3255 LIST_FOREACH(p, &allproc, p_list) {
3256 if (p->p_pid != pid)
3262 pmap = vmspace_pmap(p->p_vmspace);
3263 for(i=0;i<1024;i++) {
3266 unsigned base = i << PDRSHIFT;
3268 pde = &pmap->pm_pdir[i];
3269 if (pde && pmap_pde_v(pde)) {
3270 for(j=0;j<1024;j++) {
3271 unsigned va = base + (j << PAGE_SHIFT);
3272 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3279 pte = pmap_pte_quick( pmap, va);
3280 if (pte && pmap_pte_v(pte)) {
3284 m = PHYS_TO_VM_PAGE(pa);
3285 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3286 va, pa, m->hold_count, m->wire_count, m->flags);
3307 static void pads __P((pmap_t pm));
3308 void pmap_pvdump __P((vm_offset_t pa));
3310 /* print address space of pmap*/
3318 if (pm == kernel_pmap)
3320 for (i = 0; i < 1024; i++)
3322 for (j = 0; j < 1024; j++) {
3323 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3324 if (pm == kernel_pmap && va < KERNBASE)
3326 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3328 ptep = pmap_pte_quick(pm, va);
3329 if (pmap_pte_v(ptep))
3330 printf("%x:%x ", va, *(int *) ptep);
3339 register pv_entry_t pv;
3342 printf("pa %x", pa);
3343 m = PHYS_TO_VM_PAGE(pa);
3344 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3346 printf(" -> pmap %p, va %x, flags %x",
3347 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3349 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);