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 1.17 2003/07/10 04:47:53 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++)
349 * PG_G is terribly broken on SMP because we IPI invltlb's in some
350 * cases rather then invl1pg. Actually, I don't even know why it
351 * works under UP because self-referential page table mappings
356 if (cpu_feature & CPUID_PGE)
361 * Initialize the 4MB page size flag
365 * The 4MB page version of the initial
366 * kernel page mapping.
370 #if !defined(DISABLE_PSE)
371 if (cpu_feature & CPUID_PSE) {
374 * Note that we have enabled PSE mode
377 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
378 ptditmp &= ~(NBPDR - 1);
379 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
384 * Enable the PSE mode. If we are SMP we can't do this
385 * now because the APs will not be able to use it when
388 load_cr4(rcr4() | CR4_PSE);
391 * We can do the mapping here for the single processor
392 * case. We simply ignore the old page table page from
396 * For SMP, we still need 4K pages to bootstrap APs,
397 * PSE will be enabled as soon as all APs are up.
399 PTD[KPTDI] = (pd_entry_t) ptditmp;
400 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp;
406 if (cpu_apic_address == 0)
407 panic("pmap_bootstrap: no local apic!");
409 /* local apic is mapped on last page */
410 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
411 (cpu_apic_address & PG_FRAME));
414 /* BSP does this itself, AP's get it pre-set */
415 gd = &CPU_prvspace[0].mdglobaldata;
416 gd->gd_CMAP1 = &SMPpt[1];
417 gd->gd_CMAP2 = &SMPpt[2];
418 gd->gd_CMAP3 = &SMPpt[3];
419 gd->gd_PMAP1 = &SMPpt[4];
420 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
421 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
422 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
423 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
430 * Set 4mb pdir for mp startup
435 if (pseflag && (cpu_feature & CPUID_PSE)) {
436 load_cr4(rcr4() | CR4_PSE);
437 if (pdir4mb && mycpu->gd_cpuid == 0) { /* only on BSP */
438 kernel_pmap->pm_pdir[KPTDI] =
439 PTD[KPTDI] = (pd_entry_t)pdir4mb;
447 * Initialize the pmap module.
448 * Called by vm_init, to initialize any structures that the pmap
449 * system needs to map virtual memory.
450 * pmap_init has been enhanced to support in a fairly consistant
451 * way, discontiguous physical memory.
454 pmap_init(phys_start, phys_end)
455 vm_offset_t phys_start, phys_end;
461 * object for kernel page table pages
463 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
466 * Allocate memory for random pmap data structures. Includes the
470 for(i = 0; i < vm_page_array_size; i++) {
473 m = &vm_page_array[i];
474 TAILQ_INIT(&m->md.pv_list);
475 m->md.pv_list_count = 0;
479 * init the pv free list
481 initial_pvs = vm_page_array_size;
482 if (initial_pvs < MINPV)
484 pvzone = &pvzone_store;
485 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
486 initial_pvs * sizeof (struct pv_entry));
487 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
491 * Now it is safe to enable pv_table recording.
493 pmap_initialized = TRUE;
497 * Initialize the address space (zone) for the pv_entries. Set a
498 * high water mark so that the system can recover from excessive
499 * numbers of pv entries.
504 int shpgperproc = PMAP_SHPGPERPROC;
506 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
507 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
508 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
509 pv_entry_high_water = 9 * (pv_entry_max / 10);
510 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
514 /***************************************************
515 * Low level helper routines.....
516 ***************************************************/
518 #if defined(PMAP_DIAGNOSTIC)
521 * This code checks for non-writeable/modified pages.
522 * This should be an invalid condition.
525 pmap_nw_modified(pt_entry_t ptea)
531 if ((pte & (PG_M|PG_RW)) == PG_M)
540 * this routine defines the region(s) of memory that should
541 * not be tested for the modified bit.
543 static PMAP_INLINE int
544 pmap_track_modified(vm_offset_t va)
546 if ((va < clean_sva) || (va >= clean_eva))
552 static PMAP_INLINE void
553 invltlb_1pg(vm_offset_t va)
555 #if defined(I386_CPU)
556 if (cpu_class == CPUCLASS_386) {
566 pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
569 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
570 cpu_invlpg((void *)va);
571 if (pmap->pm_active & mycpu->gd_other_cpus)
580 pmap_TLB_invalidate_all(pmap_t pmap)
583 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
585 if (pmap->pm_active & mycpu->gd_other_cpus)
597 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
599 /* are we current address space or kernel? */
600 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
601 return (unsigned *) PTmap;
603 /* otherwise, we are alternate address space */
604 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
605 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
607 /* The page directory is not shared between CPUs */
613 return (unsigned *) APTmap;
617 * Super fast pmap_pte routine best used when scanning
618 * the pv lists. This eliminates many coarse-grained
619 * invltlb calls. Note that many of the pv list
620 * scans are across different pmaps. It is very wasteful
621 * to do an entire invltlb for checking a single mapping.
625 pmap_pte_quick(pmap_t pmap, vm_offset_t va)
627 struct mdglobaldata *gd = mdcpu;
630 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
631 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
632 unsigned index = i386_btop(va);
633 /* are we current address space or kernel? */
634 if ((pmap == kernel_pmap) ||
635 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
636 return (unsigned *) PTmap + index;
638 newpf = pde & PG_FRAME;
639 if ( ((* (unsigned *) gd->gd_PMAP1) & PG_FRAME) != newpf) {
640 * (unsigned *) gd->gd_PMAP1 = newpf | PG_RW | PG_V;
641 cpu_invlpg(gd->gd_PADDR1);
643 return gd->gd_PADDR1 + ((unsigned) index & (NPTEPG - 1));
649 * Routine: pmap_extract
651 * Extract the physical page address associated
652 * with the given map/virtual_address pair.
655 pmap_extract(pmap, va)
656 register pmap_t pmap;
660 vm_offset_t pdirindex;
661 pdirindex = va >> PDRSHIFT;
662 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
664 if ((rtval & PG_PS) != 0) {
665 rtval &= ~(NBPDR - 1);
666 rtval |= va & (NBPDR - 1);
669 pte = get_ptbase(pmap) + i386_btop(va);
670 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
677 /***************************************************
678 * Low level mapping routines.....
679 ***************************************************/
682 * add a wired page to the kva
683 * note that in order for the mapping to take effect -- you
684 * should do a invltlb after doing the pmap_kenter...
689 register vm_offset_t pa;
691 register unsigned *pte;
694 npte = pa | PG_RW | PG_V | pgeflag;
695 pte = (unsigned *)vtopte(va);
702 * remove a page from the kernel pagetables
708 register unsigned *pte;
710 pte = (unsigned *)vtopte(va);
716 * Used to map a range of physical addresses into kernel
717 * virtual address space.
719 * For now, VM is already on, we only need to map the
723 pmap_map(virt, start, end, prot)
729 while (start < end) {
730 pmap_kenter(virt, start);
739 * Add a list of wired pages to the kva
740 * this routine is only used for temporary
741 * kernel mappings that do not need to have
742 * page modification or references recorded.
743 * Note that old mappings are simply written
744 * over. The page *must* be wired.
747 pmap_qenter(va, m, count)
754 end_va = va + count * PAGE_SIZE;
756 while (va < end_va) {
759 pte = (unsigned *)vtopte(va);
760 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
762 cpu_invlpg((void *)va);
775 * this routine jerks page mappings from the
776 * kernel -- it is meant only for temporary mappings.
779 pmap_qremove(va, count)
785 end_va = va + count*PAGE_SIZE;
787 while (va < end_va) {
790 pte = (unsigned *)vtopte(va);
793 cpu_invlpg((void *)va);
805 pmap_page_lookup(object, pindex)
811 m = vm_page_lookup(object, pindex);
812 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
818 * Create a new thread and optionally associate it with a (new) process.
821 pmap_init_thread(thread_t td)
823 td->td_pcb = (struct pcb *)(td->td_kstack + UPAGES * PAGE_SIZE) - 1;
824 td->td_sp = (char *)td->td_pcb - 16;
828 * Create the UPAGES for a new process.
829 * This routine directly affects the fork perf for a process.
832 pmap_init_proc(struct proc *p, struct thread *td)
834 p->p_addr = (void *)td->td_kstack;
837 td->td_switch = cpu_heavy_switch;
841 bzero(p->p_addr, sizeof(*p->p_addr));
845 * Dispose the UPAGES for a process that has exited.
846 * This routine directly impacts the exit perf of a process.
849 pmap_dispose_proc(struct proc *p)
853 KASSERT(p->p_lock == 0, ("attempt to dispose referenced proc! %p", p));
855 if ((td = p->p_thread) != NULL) {
864 * Allow the UPAGES for a process to be prejudicially paged out.
875 upobj = p->p_upages_obj;
877 * let the upages be paged
879 for(i=0;i<UPAGES;i++) {
880 if ((m = vm_page_lookup(upobj, i)) == NULL)
881 panic("pmap_swapout_proc: upage already missing???");
883 vm_page_unwire(m, 0);
884 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
890 * Bring the UPAGES for a specified process back in.
901 upobj = p->p_upages_obj;
902 for(i=0;i<UPAGES;i++) {
904 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
906 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
909 if (m->valid != VM_PAGE_BITS_ALL) {
910 rv = vm_pager_get_pages(upobj, &m, 1, 0);
911 if (rv != VM_PAGER_OK)
912 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
913 m = vm_page_lookup(upobj, i);
914 m->valid = VM_PAGE_BITS_ALL;
919 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
924 /***************************************************
925 * Page table page management routines.....
926 ***************************************************/
929 * This routine unholds page table pages, and if the hold count
930 * drops to zero, then it decrements the wire count.
933 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
935 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
938 if (m->hold_count == 0) {
941 * unmap the page table page
943 pmap->pm_pdir[m->pindex] = 0;
944 --pmap->pm_stats.resident_count;
945 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
946 (((unsigned) PTDpde) & PG_FRAME)) {
948 * Do a invltlb to make the invalidated mapping
949 * take effect immediately.
951 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
952 pmap_TLB_invalidate(pmap, pteva);
955 if (pmap->pm_ptphint == m)
956 pmap->pm_ptphint = NULL;
959 * If the page is finally unwired, simply free it.
962 if (m->wire_count == 0) {
966 vm_page_free_zero(m);
967 --vmstats.v_wire_count;
974 static PMAP_INLINE int
975 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
978 if (m->hold_count == 0)
979 return _pmap_unwire_pte_hold(pmap, m);
985 * After removing a page table entry, this routine is used to
986 * conditionally free the page, and manage the hold/wire counts.
989 pmap_unuse_pt(pmap, va, mpte)
995 if (va >= UPT_MIN_ADDRESS)
999 ptepindex = (va >> PDRSHIFT);
1000 if (pmap->pm_ptphint &&
1001 (pmap->pm_ptphint->pindex == ptepindex)) {
1002 mpte = pmap->pm_ptphint;
1004 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1005 pmap->pm_ptphint = mpte;
1009 return pmap_unwire_pte_hold(pmap, mpte);
1017 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1018 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1020 pmap->pm_active = 0;
1021 pmap->pm_ptphint = NULL;
1022 TAILQ_INIT(&pmap->pm_pvlist);
1023 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1027 * Initialize a preallocated and zeroed pmap structure,
1028 * such as one in a vmspace structure.
1032 register struct pmap *pmap;
1037 * No need to allocate page table space yet but we do need a valid
1038 * page directory table.
1040 if (pmap->pm_pdir == NULL)
1042 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1045 * allocate object for the ptes
1047 if (pmap->pm_pteobj == NULL)
1048 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1051 * allocate the page directory page
1053 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1054 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1056 ptdpg->wire_count = 1;
1057 ++vmstats.v_wire_count;
1060 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1061 ptdpg->valid = VM_PAGE_BITS_ALL;
1063 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1064 if ((ptdpg->flags & PG_ZERO) == 0)
1065 bzero(pmap->pm_pdir, PAGE_SIZE);
1067 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1069 /* install self-referential address mapping entry */
1070 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1071 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1074 pmap->pm_active = 0;
1075 pmap->pm_ptphint = NULL;
1076 TAILQ_INIT(&pmap->pm_pvlist);
1077 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1081 * Wire in kernel global address entries. To avoid a race condition
1082 * between pmap initialization and pmap_growkernel, this procedure
1083 * should be called after the vmspace is attached to the process
1084 * but before this pmap is activated.
1090 /* XXX copies current process, does not fill in MPPTDI */
1091 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1095 pmap_release_free_page(pmap, p)
1099 unsigned *pde = (unsigned *) pmap->pm_pdir;
1101 * This code optimizes the case of freeing non-busy
1102 * page-table pages. Those pages are zero now, and
1103 * might as well be placed directly into the zero queue.
1105 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1111 * Remove the page table page from the processes address space.
1114 pmap->pm_stats.resident_count--;
1116 if (p->hold_count) {
1117 panic("pmap_release: freeing held page table page");
1120 * Page directory pages need to have the kernel
1121 * stuff cleared, so they can go into the zero queue also.
1123 if (p->pindex == PTDPTDI) {
1124 bzero(pde + KPTDI, nkpt * PTESIZE);
1127 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1130 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1131 pmap->pm_ptphint = NULL;
1134 vmstats.v_wire_count--;
1135 vm_page_free_zero(p);
1140 * this routine is called if the page table page is not
1144 _pmap_allocpte(pmap, ptepindex)
1148 vm_offset_t pteva, ptepa;
1152 * Find or fabricate a new pagetable page
1154 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1155 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1157 KASSERT(m->queue == PQ_NONE,
1158 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1160 if (m->wire_count == 0)
1161 vmstats.v_wire_count++;
1165 * Increment the hold count for the page table page
1166 * (denoting a new mapping.)
1171 * Map the pagetable page into the process address space, if
1172 * it isn't already there.
1175 pmap->pm_stats.resident_count++;
1177 ptepa = VM_PAGE_TO_PHYS(m);
1178 pmap->pm_pdir[ptepindex] =
1179 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1182 * Set the page table hint
1184 pmap->pm_ptphint = m;
1187 * Try to use the new mapping, but if we cannot, then
1188 * do it with the routine that maps the page explicitly.
1190 if ((m->flags & PG_ZERO) == 0) {
1191 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1192 (((unsigned) PTDpde) & PG_FRAME)) {
1193 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1194 bzero((caddr_t) pteva, PAGE_SIZE);
1196 pmap_zero_page(ptepa);
1200 m->valid = VM_PAGE_BITS_ALL;
1201 vm_page_flag_clear(m, PG_ZERO);
1202 vm_page_flag_set(m, PG_MAPPED);
1209 pmap_allocpte(pmap, va)
1218 * Calculate pagetable page index
1220 ptepindex = va >> PDRSHIFT;
1223 * Get the page directory entry
1225 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1228 * This supports switching from a 4MB page to a
1231 if (ptepa & PG_PS) {
1232 pmap->pm_pdir[ptepindex] = 0;
1238 * If the page table page is mapped, we just increment the
1239 * hold count, and activate it.
1243 * In order to get the page table page, try the
1246 if (pmap->pm_ptphint &&
1247 (pmap->pm_ptphint->pindex == ptepindex)) {
1248 m = pmap->pm_ptphint;
1250 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1251 pmap->pm_ptphint = m;
1257 * Here if the pte page isn't mapped, or if it has been deallocated.
1259 return _pmap_allocpte(pmap, ptepindex);
1263 /***************************************************
1264 * Pmap allocation/deallocation routines.
1265 ***************************************************/
1268 * Release any resources held by the given physical map.
1269 * Called when a pmap initialized by pmap_pinit is being released.
1270 * Should only be called if the map contains no valid mappings.
1274 register struct pmap *pmap;
1276 vm_page_t p,n,ptdpg;
1277 vm_object_t object = pmap->pm_pteobj;
1280 #if defined(DIAGNOSTIC)
1281 if (object->ref_count != 1)
1282 panic("pmap_release: pteobj reference count != 1");
1287 curgeneration = object->generation;
1288 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1289 n = TAILQ_NEXT(p, listq);
1290 if (p->pindex == PTDPTDI) {
1295 if (!pmap_release_free_page(pmap, p) &&
1296 (object->generation != curgeneration))
1301 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1306 kvm_size(SYSCTL_HANDLER_ARGS)
1308 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1310 return sysctl_handle_long(oidp, &ksize, 0, req);
1312 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1313 0, 0, kvm_size, "IU", "Size of KVM");
1316 kvm_free(SYSCTL_HANDLER_ARGS)
1318 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1320 return sysctl_handle_long(oidp, &kfree, 0, req);
1322 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1323 0, 0, kvm_free, "IU", "Amount of KVM free");
1326 * grow the number of kernel page table entries, if needed
1329 pmap_growkernel(vm_offset_t addr)
1334 vm_offset_t ptppaddr;
1339 if (kernel_vm_end == 0) {
1340 kernel_vm_end = KERNBASE;
1342 while (pdir_pde(PTD, kernel_vm_end)) {
1343 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1347 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1348 while (kernel_vm_end < addr) {
1349 if (pdir_pde(PTD, kernel_vm_end)) {
1350 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1355 * This index is bogus, but out of the way
1357 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1359 panic("pmap_growkernel: no memory to grow kernel");
1364 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1365 pmap_zero_page(ptppaddr);
1366 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1367 pdir_pde(PTD, kernel_vm_end) = newpdir;
1369 LIST_FOREACH(p, &allproc, p_list) {
1371 pmap = vmspace_pmap(p->p_vmspace);
1372 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1375 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1376 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1382 * Retire the given physical map from service.
1383 * Should only be called if the map contains
1384 * no valid mappings.
1388 register pmap_t pmap;
1395 count = --pmap->pm_count;
1398 panic("destroying a pmap is not yet implemented");
1403 * Add a reference to the specified pmap.
1406 pmap_reference(pmap)
1414 /***************************************************
1415 * page management routines.
1416 ***************************************************/
1419 * free the pv_entry back to the free list. This function may be
1420 * called from an interrupt.
1422 static PMAP_INLINE void
1431 * get a new pv_entry, allocating a block from the system
1432 * when needed. This function may be called from an interrupt.
1438 if (pv_entry_high_water &&
1439 (pv_entry_count > pv_entry_high_water) &&
1440 (pmap_pagedaemon_waken == 0)) {
1441 pmap_pagedaemon_waken = 1;
1442 wakeup (&vm_pages_needed);
1444 return zalloc(pvzone);
1448 * This routine is very drastic, but can save the system
1456 static int warningdone=0;
1458 if (pmap_pagedaemon_waken == 0)
1461 if (warningdone < 5) {
1462 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1466 for(i = 0; i < vm_page_array_size; i++) {
1467 m = &vm_page_array[i];
1468 if (m->wire_count || m->hold_count || m->busy ||
1469 (m->flags & PG_BUSY))
1473 pmap_pagedaemon_waken = 0;
1478 * If it is the first entry on the list, it is actually
1479 * in the header and we must copy the following entry up
1480 * to the header. Otherwise we must search the list for
1481 * the entry. In either case we free the now unused entry.
1485 pmap_remove_entry(pmap, m, va)
1495 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1496 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1497 if (pmap == pv->pv_pmap && va == pv->pv_va)
1501 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1502 if (va == pv->pv_va)
1510 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1511 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1512 m->md.pv_list_count--;
1513 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1514 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1516 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1525 * Create a pv entry for page at pa for
1529 pmap_insert_entry(pmap, va, mpte, m)
1540 pv = get_pv_entry();
1545 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1546 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1547 m->md.pv_list_count++;
1553 * pmap_remove_pte: do the things to unmap a page in a process
1556 pmap_remove_pte(pmap, ptq, va)
1564 oldpte = loadandclear(ptq);
1566 pmap->pm_stats.wired_count -= 1;
1568 * Machines that don't support invlpg, also don't support
1573 pmap->pm_stats.resident_count -= 1;
1574 if (oldpte & PG_MANAGED) {
1575 m = PHYS_TO_VM_PAGE(oldpte);
1576 if (oldpte & PG_M) {
1577 #if defined(PMAP_DIAGNOSTIC)
1578 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1580 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1584 if (pmap_track_modified(va))
1588 vm_page_flag_set(m, PG_REFERENCED);
1589 return pmap_remove_entry(pmap, m, va);
1591 return pmap_unuse_pt(pmap, va, NULL);
1598 * Remove a single page from a process address space
1601 pmap_remove_page(pmap, va)
1603 register vm_offset_t va;
1605 register unsigned *ptq;
1608 * if there is no pte for this address, just skip it!!!
1610 if (*pmap_pde(pmap, va) == 0) {
1615 * get a local va for mappings for this pmap.
1617 ptq = get_ptbase(pmap) + i386_btop(va);
1619 (void) pmap_remove_pte(pmap, ptq, va);
1620 pmap_TLB_invalidate(pmap, va);
1626 * Remove the given range of addresses from the specified map.
1628 * It is assumed that the start and end are properly
1629 * rounded to the page size.
1632 pmap_remove(pmap, sva, eva)
1634 register vm_offset_t sva;
1635 register vm_offset_t eva;
1637 register unsigned *ptbase;
1639 vm_offset_t ptpaddr;
1640 vm_offset_t sindex, eindex;
1646 if (pmap->pm_stats.resident_count == 0)
1650 * special handling of removing one page. a very
1651 * common operation and easy to short circuit some
1654 if (((sva + PAGE_SIZE) == eva) &&
1655 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1656 pmap_remove_page(pmap, sva);
1663 * Get a local virtual address for the mappings that are being
1666 ptbase = get_ptbase(pmap);
1668 sindex = i386_btop(sva);
1669 eindex = i386_btop(eva);
1671 for (; sindex < eindex; sindex = pdnxt) {
1675 * Calculate index for next page table.
1677 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1678 if (pmap->pm_stats.resident_count == 0)
1681 pdirindex = sindex / NPDEPG;
1682 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1683 pmap->pm_pdir[pdirindex] = 0;
1684 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1690 * Weed out invalid mappings. Note: we assume that the page
1691 * directory table is always allocated, and in kernel virtual.
1697 * Limit our scan to either the end of the va represented
1698 * by the current page table page, or to the end of the
1699 * range being removed.
1701 if (pdnxt > eindex) {
1705 for ( ;sindex != pdnxt; sindex++) {
1707 if (ptbase[sindex] == 0) {
1710 va = i386_ptob(sindex);
1713 if (pmap_remove_pte(pmap,
1714 ptbase + sindex, va))
1720 pmap_TLB_invalidate_all(pmap);
1724 * Routine: pmap_remove_all
1726 * Removes this physical page from
1727 * all physical maps in which it resides.
1728 * Reflects back modify bits to the pager.
1731 * Original versions of this routine were very
1732 * inefficient because they iteratively called
1733 * pmap_remove (slow...)
1740 register pv_entry_t pv;
1741 register unsigned *pte, tpte;
1744 #if defined(PMAP_DIAGNOSTIC)
1746 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1749 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1750 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1755 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1756 pv->pv_pmap->pm_stats.resident_count--;
1758 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1760 tpte = loadandclear(pte);
1762 pv->pv_pmap->pm_stats.wired_count--;
1765 vm_page_flag_set(m, PG_REFERENCED);
1768 * Update the vm_page_t clean and reference bits.
1771 #if defined(PMAP_DIAGNOSTIC)
1772 if (pmap_nw_modified((pt_entry_t) tpte)) {
1774 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1778 if (pmap_track_modified(pv->pv_va))
1781 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1783 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1784 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1785 m->md.pv_list_count--;
1786 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1790 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1796 * Set the physical protection on the
1797 * specified range of this map as requested.
1800 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1802 register unsigned *ptbase;
1803 vm_offset_t pdnxt, ptpaddr;
1804 vm_pindex_t sindex, eindex;
1810 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1811 pmap_remove(pmap, sva, eva);
1815 if (prot & VM_PROT_WRITE)
1820 ptbase = get_ptbase(pmap);
1822 sindex = i386_btop(sva);
1823 eindex = i386_btop(eva);
1825 for (; sindex < eindex; sindex = pdnxt) {
1829 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1831 pdirindex = sindex / NPDEPG;
1832 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1833 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1834 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1840 * Weed out invalid mappings. Note: we assume that the page
1841 * directory table is always allocated, and in kernel virtual.
1846 if (pdnxt > eindex) {
1850 for (; sindex != pdnxt; sindex++) {
1855 pbits = ptbase[sindex];
1857 if (pbits & PG_MANAGED) {
1860 m = PHYS_TO_VM_PAGE(pbits);
1861 vm_page_flag_set(m, PG_REFERENCED);
1865 if (pmap_track_modified(i386_ptob(sindex))) {
1867 m = PHYS_TO_VM_PAGE(pbits);
1876 if (pbits != ptbase[sindex]) {
1877 ptbase[sindex] = pbits;
1883 pmap_TLB_invalidate_all(pmap);
1887 * Insert the given physical page (p) at
1888 * the specified virtual address (v) in the
1889 * target physical map with the protection requested.
1891 * If specified, the page will be wired down, meaning
1892 * that the related pte can not be reclaimed.
1894 * NB: This is the only routine which MAY NOT lazy-evaluate
1895 * or lose information. That is, this routine must actually
1896 * insert this page into the given map NOW.
1899 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1903 register unsigned *pte;
1905 vm_offset_t origpte, newpte;
1912 #ifdef PMAP_DIAGNOSTIC
1913 if (va > VM_MAX_KERNEL_ADDRESS)
1914 panic("pmap_enter: toobig");
1915 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
1916 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
1921 * In the case that a page table page is not
1922 * resident, we are creating it here.
1924 if (va < UPT_MIN_ADDRESS) {
1925 mpte = pmap_allocpte(pmap, va);
1927 #if 0 && defined(PMAP_DIAGNOSTIC)
1929 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
1930 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
1931 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
1932 pmap->pm_pdir[PTDPTDI], origpte, va);
1935 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
1936 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
1937 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
1938 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
1939 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
1940 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
1941 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
1947 pte = pmap_pte(pmap, va);
1950 * Page Directory table entry not valid, we need a new PT page
1953 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
1954 (void *)pmap->pm_pdir[PTDPTDI], va);
1957 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
1958 origpte = *(vm_offset_t *)pte;
1959 opa = origpte & PG_FRAME;
1961 if (origpte & PG_PS)
1962 panic("pmap_enter: attempted pmap_enter on 4MB page");
1965 * Mapping has not changed, must be protection or wiring change.
1967 if (origpte && (opa == pa)) {
1969 * Wiring change, just update stats. We don't worry about
1970 * wiring PT pages as they remain resident as long as there
1971 * are valid mappings in them. Hence, if a user page is wired,
1972 * the PT page will be also.
1974 if (wired && ((origpte & PG_W) == 0))
1975 pmap->pm_stats.wired_count++;
1976 else if (!wired && (origpte & PG_W))
1977 pmap->pm_stats.wired_count--;
1979 #if defined(PMAP_DIAGNOSTIC)
1980 if (pmap_nw_modified((pt_entry_t) origpte)) {
1982 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
1988 * Remove extra pte reference
1993 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
1994 if ((origpte & PG_RW) == 0) {
1997 cpu_invlpg((void *)va);
1998 if (pmap->pm_active & mycpu->gd_other_cpus)
2008 * We might be turning off write access to the page,
2009 * so we go ahead and sense modify status.
2011 if (origpte & PG_MANAGED) {
2012 if ((origpte & PG_M) && pmap_track_modified(va)) {
2014 om = PHYS_TO_VM_PAGE(opa);
2022 * Mapping has changed, invalidate old range and fall through to
2023 * handle validating new mapping.
2027 err = pmap_remove_pte(pmap, pte, va);
2029 panic("pmap_enter: pte vanished, va: 0x%x", va);
2033 * Enter on the PV list if part of our managed memory. Note that we
2034 * raise IPL while manipulating pv_table since pmap_enter can be
2035 * called at interrupt time.
2037 if (pmap_initialized &&
2038 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2039 pmap_insert_entry(pmap, va, mpte, m);
2044 * Increment counters
2046 pmap->pm_stats.resident_count++;
2048 pmap->pm_stats.wired_count++;
2052 * Now validate mapping with desired protection/wiring.
2054 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2058 if (va < UPT_MIN_ADDRESS)
2060 if (pmap == kernel_pmap)
2064 * if the mapping or permission bits are different, we need
2065 * to update the pte.
2067 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2068 *pte = newpte | PG_A;
2071 cpu_invlpg((void *)va);
2072 if (pmap->pm_active & mycpu->gd_other_cpus)
2082 * this code makes some *MAJOR* assumptions:
2083 * 1. Current pmap & pmap exists.
2086 * 4. No page table pages.
2087 * 5. Tlbflush is deferred to calling procedure.
2088 * 6. Page IS managed.
2089 * but is *MUCH* faster than pmap_enter...
2093 pmap_enter_quick(pmap, va, m, mpte)
2094 register pmap_t pmap;
2103 * In the case that a page table page is not
2104 * resident, we are creating it here.
2106 if (va < UPT_MIN_ADDRESS) {
2111 * Calculate pagetable page index
2113 ptepindex = va >> PDRSHIFT;
2114 if (mpte && (mpte->pindex == ptepindex)) {
2119 * Get the page directory entry
2121 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2124 * If the page table page is mapped, we just increment
2125 * the hold count, and activate it.
2129 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2130 if (pmap->pm_ptphint &&
2131 (pmap->pm_ptphint->pindex == ptepindex)) {
2132 mpte = pmap->pm_ptphint;
2134 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2135 pmap->pm_ptphint = mpte;
2141 mpte = _pmap_allocpte(pmap, ptepindex);
2149 * This call to vtopte makes the assumption that we are
2150 * entering the page into the current pmap. In order to support
2151 * quick entry into any pmap, one would likely use pmap_pte_quick.
2152 * But that isn't as quick as vtopte.
2154 pte = (unsigned *)vtopte(va);
2157 pmap_unwire_pte_hold(pmap, mpte);
2162 * Enter on the PV list if part of our managed memory. Note that we
2163 * raise IPL while manipulating pv_table since pmap_enter can be
2164 * called at interrupt time.
2166 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2167 pmap_insert_entry(pmap, va, mpte, m);
2170 * Increment counters
2172 pmap->pm_stats.resident_count++;
2174 pa = VM_PAGE_TO_PHYS(m);
2177 * Now validate mapping with RO protection
2179 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2180 *pte = pa | PG_V | PG_U;
2182 *pte = pa | PG_V | PG_U | PG_MANAGED;
2188 * Make a temporary mapping for a physical address. This is only intended
2189 * to be used for panic dumps.
2192 pmap_kenter_temporary(vm_offset_t pa, int i)
2194 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2195 return ((void *)crashdumpmap);
2198 #define MAX_INIT_PT (96)
2200 * pmap_object_init_pt preloads the ptes for a given object
2201 * into the specified pmap. This eliminates the blast of soft
2202 * faults on process startup and immediately after an mmap.
2205 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2218 if (pmap == NULL || object == NULL)
2222 * This code maps large physical mmap regions into the
2223 * processor address space. Note that some shortcuts
2224 * are taken, but the code works.
2227 (object->type == OBJT_DEVICE) &&
2228 ((addr & (NBPDR - 1)) == 0) &&
2229 ((size & (NBPDR - 1)) == 0) ) {
2232 unsigned int ptepindex;
2236 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2240 p = vm_page_lookup(object, pindex);
2241 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2245 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2250 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2255 p = vm_page_lookup(object, pindex);
2259 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2260 if (ptepa & (NBPDR - 1)) {
2264 p->valid = VM_PAGE_BITS_ALL;
2266 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2267 npdes = size >> PDRSHIFT;
2268 for(i=0;i<npdes;i++) {
2269 pmap->pm_pdir[ptepindex] =
2270 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2274 vm_page_flag_set(p, PG_MAPPED);
2279 psize = i386_btop(size);
2281 if ((object->type != OBJT_VNODE) ||
2282 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2283 (object->resident_page_count > MAX_INIT_PT))) {
2287 if (psize + pindex > object->size) {
2288 if (object->size < pindex)
2290 psize = object->size - pindex;
2295 * if we are processing a major portion of the object, then scan the
2298 if (psize > (object->resident_page_count >> 2)) {
2301 for (p = TAILQ_FIRST(&object->memq);
2302 ((objpgs > 0) && (p != NULL));
2303 p = TAILQ_NEXT(p, listq)) {
2306 if (tmpidx < pindex) {
2310 if (tmpidx >= psize) {
2314 * don't allow an madvise to blow away our really
2315 * free pages allocating pv entries.
2317 if ((limit & MAP_PREFAULT_MADVISE) &&
2318 vmstats.v_free_count < vmstats.v_free_reserved) {
2321 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2323 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2324 if ((p->queue - p->pc) == PQ_CACHE)
2325 vm_page_deactivate(p);
2327 mpte = pmap_enter_quick(pmap,
2328 addr + i386_ptob(tmpidx), p, mpte);
2329 vm_page_flag_set(p, PG_MAPPED);
2336 * else lookup the pages one-by-one.
2338 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2340 * don't allow an madvise to blow away our really
2341 * free pages allocating pv entries.
2343 if ((limit & MAP_PREFAULT_MADVISE) &&
2344 vmstats.v_free_count < vmstats.v_free_reserved) {
2347 p = vm_page_lookup(object, tmpidx + pindex);
2349 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2351 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2352 if ((p->queue - p->pc) == PQ_CACHE)
2353 vm_page_deactivate(p);
2355 mpte = pmap_enter_quick(pmap,
2356 addr + i386_ptob(tmpidx), p, mpte);
2357 vm_page_flag_set(p, PG_MAPPED);
2366 * pmap_prefault provides a quick way of clustering
2367 * pagefaults into a processes address space. It is a "cousin"
2368 * of pmap_object_init_pt, except it runs at page fault time instead
2373 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2375 static int pmap_prefault_pageorder[] = {
2376 -PAGE_SIZE, PAGE_SIZE,
2377 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2378 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2379 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2383 pmap_prefault(pmap, addra, entry)
2386 vm_map_entry_t entry;
2395 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2398 object = entry->object.vm_object;
2400 starta = addra - PFBAK * PAGE_SIZE;
2401 if (starta < entry->start) {
2402 starta = entry->start;
2403 } else if (starta > addra) {
2408 for (i = 0; i < PAGEORDER_SIZE; i++) {
2409 vm_object_t lobject;
2412 addr = addra + pmap_prefault_pageorder[i];
2413 if (addr > addra + (PFFOR * PAGE_SIZE))
2416 if (addr < starta || addr >= entry->end)
2419 if ((*pmap_pde(pmap, addr)) == NULL)
2422 pte = (unsigned *) vtopte(addr);
2426 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2428 for (m = vm_page_lookup(lobject, pindex);
2429 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2430 lobject = lobject->backing_object) {
2431 if (lobject->backing_object_offset & PAGE_MASK)
2433 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2434 m = vm_page_lookup(lobject->backing_object, pindex);
2438 * give-up when a page is not in memory
2443 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2445 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2447 if ((m->queue - m->pc) == PQ_CACHE) {
2448 vm_page_deactivate(m);
2451 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2452 vm_page_flag_set(m, PG_MAPPED);
2459 * Routine: pmap_change_wiring
2460 * Function: Change the wiring attribute for a map/virtual-address
2462 * In/out conditions:
2463 * The mapping must already exist in the pmap.
2466 pmap_change_wiring(pmap, va, wired)
2467 register pmap_t pmap;
2471 register unsigned *pte;
2476 pte = pmap_pte(pmap, va);
2478 if (wired && !pmap_pte_w(pte))
2479 pmap->pm_stats.wired_count++;
2480 else if (!wired && pmap_pte_w(pte))
2481 pmap->pm_stats.wired_count--;
2484 * Wiring is not a hardware characteristic so there is no need to
2487 pmap_pte_set_w(pte, wired);
2493 * Copy the range specified by src_addr/len
2494 * from the source map to the range dst_addr/len
2495 * in the destination map.
2497 * This routine is only advisory and need not do anything.
2501 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2502 pmap_t dst_pmap, src_pmap;
2503 vm_offset_t dst_addr;
2505 vm_offset_t src_addr;
2508 vm_offset_t end_addr = src_addr + len;
2510 unsigned src_frame, dst_frame;
2513 if (dst_addr != src_addr)
2516 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2517 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2521 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2522 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2523 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2525 /* The page directory is not shared between CPUs */
2532 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2533 unsigned *src_pte, *dst_pte;
2534 vm_page_t dstmpte, srcmpte;
2535 vm_offset_t srcptepaddr;
2538 if (addr >= UPT_MIN_ADDRESS)
2539 panic("pmap_copy: invalid to pmap_copy page tables\n");
2542 * Don't let optional prefaulting of pages make us go
2543 * way below the low water mark of free pages or way
2544 * above high water mark of used pv entries.
2546 if (vmstats.v_free_count < vmstats.v_free_reserved ||
2547 pv_entry_count > pv_entry_high_water)
2550 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2551 ptepindex = addr >> PDRSHIFT;
2553 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2554 if (srcptepaddr == 0)
2557 if (srcptepaddr & PG_PS) {
2558 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2559 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2560 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2565 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2566 if ((srcmpte == NULL) ||
2567 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2570 if (pdnxt > end_addr)
2573 src_pte = (unsigned *) vtopte(addr);
2574 dst_pte = (unsigned *) avtopte(addr);
2575 while (addr < pdnxt) {
2579 * we only virtual copy managed pages
2581 if ((ptetemp & PG_MANAGED) != 0) {
2583 * We have to check after allocpte for the
2584 * pte still being around... allocpte can
2587 dstmpte = pmap_allocpte(dst_pmap, addr);
2588 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2590 * Clear the modified and
2591 * accessed (referenced) bits
2594 m = PHYS_TO_VM_PAGE(ptetemp);
2595 *dst_pte = ptetemp & ~(PG_M | PG_A);
2596 dst_pmap->pm_stats.resident_count++;
2597 pmap_insert_entry(dst_pmap, addr,
2600 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2602 if (dstmpte->hold_count >= srcmpte->hold_count)
2613 * Routine: pmap_kernel
2615 * Returns the physical map handle for the kernel.
2620 return (kernel_pmap);
2624 * pmap_zero_page zeros the specified hardware page by mapping
2625 * the page into KVM and using bzero to clear its contents.
2628 pmap_zero_page(vm_offset_t phys)
2630 struct mdglobaldata *gd = mdcpu;
2632 if (*(int *)gd->gd_CMAP3)
2633 panic("pmap_zero_page: CMAP3 busy");
2635 *(int *)gd->gd_CMAP3 =
2636 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2637 cpu_invlpg(gd->gd_CADDR3);
2639 #if defined(I686_CPU)
2640 if (cpu_class == CPUCLASS_686)
2641 i686_pagezero(gd->gd_CADDR3);
2644 bzero(gd->gd_CADDR3, PAGE_SIZE);
2646 *(int *) gd->gd_CMAP3 = 0;
2650 * pmap_zero_page_area zeros the specified hardware page by mapping
2651 * the page into KVM and using bzero to clear its contents.
2653 * off and size may not cover an area beyond a single hardware page.
2656 pmap_zero_page_area(phys, off, size)
2661 struct mdglobaldata *gd = mdcpu;
2663 if (*(int *) gd->gd_CMAP3)
2664 panic("pmap_zero_page: CMAP3 busy");
2666 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2667 cpu_invlpg(gd->gd_CADDR3);
2669 #if defined(I686_CPU)
2670 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2671 i686_pagezero(gd->gd_CADDR3);
2674 bzero((char *)gd->gd_CADDR3 + off, size);
2676 *(int *) gd->gd_CMAP3 = 0;
2680 * pmap_copy_page copies the specified (machine independent)
2681 * page by mapping the page into virtual memory and using
2682 * bcopy to copy the page, one machine dependent page at a
2686 pmap_copy_page(src, dst)
2690 struct mdglobaldata *gd = mdcpu;
2692 if (*(int *) gd->gd_CMAP1)
2693 panic("pmap_copy_page: CMAP1 busy");
2694 if (*(int *) gd->gd_CMAP2)
2695 panic("pmap_copy_page: CMAP2 busy");
2697 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2698 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2700 cpu_invlpg(gd->gd_CADDR1);
2701 cpu_invlpg(gd->gd_CADDR2);
2703 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
2705 *(int *) gd->gd_CMAP1 = 0;
2706 *(int *) gd->gd_CMAP2 = 0;
2711 * Routine: pmap_pageable
2713 * Make the specified pages (by pmap, offset)
2714 * pageable (or not) as requested.
2716 * A page which is not pageable may not take
2717 * a fault; therefore, its page table entry
2718 * must remain valid for the duration.
2720 * This routine is merely advisory; pmap_enter
2721 * will specify that these pages are to be wired
2722 * down (or not) as appropriate.
2725 pmap_pageable(pmap, sva, eva, pageable)
2727 vm_offset_t sva, eva;
2733 * Returns true if the pmap's pv is one of the first
2734 * 16 pvs linked to from this page. This count may
2735 * be changed upwards or downwards in the future; it
2736 * is only necessary that true be returned for a small
2737 * subset of pmaps for proper page aging.
2740 pmap_page_exists_quick(pmap, m)
2748 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2753 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2754 if (pv->pv_pmap == pmap) {
2766 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2768 * Remove all pages from specified address space
2769 * this aids process exit speeds. Also, this code
2770 * is special cased for current process only, but
2771 * can have the more generic (and slightly slower)
2772 * mode enabled. This is much faster than pmap_remove
2773 * in the case of running down an entire address space.
2776 pmap_remove_pages(pmap, sva, eva)
2778 vm_offset_t sva, eva;
2780 unsigned *pte, tpte;
2785 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2786 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2787 printf("warning: pmap_remove_pages called with non-current pmap\n");
2793 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2797 if (pv->pv_va >= eva || pv->pv_va < sva) {
2798 npv = TAILQ_NEXT(pv, pv_plist);
2802 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2803 pte = (unsigned *)vtopte(pv->pv_va);
2805 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2810 * We cannot remove wired pages from a process' mapping at this time
2813 npv = TAILQ_NEXT(pv, pv_plist);
2818 m = PHYS_TO_VM_PAGE(tpte);
2820 KASSERT(m < &vm_page_array[vm_page_array_size],
2821 ("pmap_remove_pages: bad tpte %x", tpte));
2823 pv->pv_pmap->pm_stats.resident_count--;
2826 * Update the vm_page_t clean and reference bits.
2833 npv = TAILQ_NEXT(pv, pv_plist);
2834 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2836 m->md.pv_list_count--;
2837 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2838 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2839 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2842 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2846 pmap_TLB_invalidate_all(pmap);
2850 * pmap_testbit tests bits in pte's
2851 * note that the testbit/changebit routines are inline,
2852 * and a lot of things compile-time evaluate.
2855 pmap_testbit(m, bit)
2863 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2866 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
2871 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2873 * if the bit being tested is the modified bit, then
2874 * mark clean_map and ptes as never
2877 if (bit & (PG_A|PG_M)) {
2878 if (!pmap_track_modified(pv->pv_va))
2882 #if defined(PMAP_DIAGNOSTIC)
2884 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
2888 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2899 * this routine is used to modify bits in ptes
2901 static __inline void
2902 pmap_changebit(m, bit, setem)
2907 register pv_entry_t pv;
2908 register unsigned *pte;
2911 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2917 * Loop over all current mappings setting/clearing as appropos If
2918 * setting RO do we need to clear the VAC?
2920 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2922 * don't write protect pager mappings
2924 if (!setem && (bit == PG_RW)) {
2925 if (!pmap_track_modified(pv->pv_va))
2929 #if defined(PMAP_DIAGNOSTIC)
2931 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
2936 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2940 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2942 vm_offset_t pbits = *(vm_offset_t *)pte;
2948 *(int *)pte = pbits & ~(PG_M|PG_RW);
2950 *(int *)pte = pbits & ~bit;
2952 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2960 * pmap_page_protect:
2962 * Lower the permission for all mappings to a given page.
2965 pmap_page_protect(vm_page_t m, vm_prot_t prot)
2967 if ((prot & VM_PROT_WRITE) == 0) {
2968 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
2969 pmap_changebit(m, PG_RW, FALSE);
2977 pmap_phys_address(ppn)
2980 return (i386_ptob(ppn));
2984 * pmap_ts_referenced:
2986 * Return a count of reference bits for a page, clearing those bits.
2987 * It is not necessary for every reference bit to be cleared, but it
2988 * is necessary that 0 only be returned when there are truly no
2989 * reference bits set.
2991 * XXX: The exact number of bits to check and clear is a matter that
2992 * should be tested and standardized at some point in the future for
2993 * optimal aging of shared pages.
2996 pmap_ts_referenced(vm_page_t m)
2998 register pv_entry_t pv, pvf, pvn;
3003 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3008 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3013 pvn = TAILQ_NEXT(pv, pv_list);
3015 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3017 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3019 if (!pmap_track_modified(pv->pv_va))
3022 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3024 if (pte && (*pte & PG_A)) {
3027 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3034 } while ((pv = pvn) != NULL && pv != pvf);
3044 * Return whether or not the specified physical page was modified
3045 * in any physical maps.
3048 pmap_is_modified(vm_page_t m)
3050 return pmap_testbit(m, PG_M);
3054 * Clear the modify bits on the specified physical page.
3057 pmap_clear_modify(vm_page_t m)
3059 pmap_changebit(m, PG_M, FALSE);
3063 * pmap_clear_reference:
3065 * Clear the reference bit on the specified physical page.
3068 pmap_clear_reference(vm_page_t m)
3070 pmap_changebit(m, PG_A, FALSE);
3074 * Miscellaneous support routines follow
3078 i386_protection_init()
3080 register int *kp, prot;
3082 kp = protection_codes;
3083 for (prot = 0; prot < 8; prot++) {
3085 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3087 * Read access is also 0. There isn't any execute bit,
3088 * so just make it readable.
3090 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3091 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3092 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3095 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3096 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3097 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3098 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3106 * Map a set of physical memory pages into the kernel virtual
3107 * address space. Return a pointer to where it is mapped. This
3108 * routine is intended to be used for mapping device memory,
3111 * NOTE: we can't use pgeflag unless we invalidate the pages one at
3115 pmap_mapdev(pa, size)
3119 vm_offset_t va, tmpva, offset;
3122 offset = pa & PAGE_MASK;
3123 size = roundup(offset + size, PAGE_SIZE);
3125 va = kmem_alloc_pageable(kernel_map, size);
3127 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3130 for (tmpva = va; size > 0;) {
3131 pte = (unsigned *)vtopte(tmpva);
3132 *pte = pa | PG_RW | PG_V; /* | pgeflag; */
3139 return ((void *)(va + offset));
3143 pmap_unmapdev(va, size)
3147 vm_offset_t base, offset;
3149 base = va & PG_FRAME;
3150 offset = va & PAGE_MASK;
3151 size = roundup(offset + size, PAGE_SIZE);
3152 kmem_free(kernel_map, base, size);
3156 * perform the pmap work for mincore
3159 pmap_mincore(pmap, addr)
3164 unsigned *ptep, pte;
3168 ptep = pmap_pte(pmap, addr);
3173 if ((pte = *ptep) != 0) {
3176 val = MINCORE_INCORE;
3177 if ((pte & PG_MANAGED) == 0)
3180 pa = pte & PG_FRAME;
3182 m = PHYS_TO_VM_PAGE(pa);
3188 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3190 * Modified by someone
3192 else if (m->dirty || pmap_is_modified(m))
3193 val |= MINCORE_MODIFIED_OTHER;
3198 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3201 * Referenced by someone
3203 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3204 val |= MINCORE_REFERENCED_OTHER;
3205 vm_page_flag_set(m, PG_REFERENCED);
3212 pmap_activate(struct proc *p)
3216 pmap = vmspace_pmap(p->p_vmspace);
3218 atomic_set_int(&pmap->pm_active, 1 << mycpu->gd_cpuid);
3220 pmap->pm_active |= 1;
3222 #if defined(SWTCH_OPTIM_STATS)
3225 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3226 load_cr3(p->p_thread->td_pcb->pcb_cr3);
3230 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3233 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3237 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3242 #if defined(PMAP_DEBUG)
3243 pmap_pid_dump(int pid)
3249 LIST_FOREACH(p, &allproc, p_list) {
3250 if (p->p_pid != pid)
3256 pmap = vmspace_pmap(p->p_vmspace);
3257 for(i=0;i<1024;i++) {
3260 unsigned base = i << PDRSHIFT;
3262 pde = &pmap->pm_pdir[i];
3263 if (pde && pmap_pde_v(pde)) {
3264 for(j=0;j<1024;j++) {
3265 unsigned va = base + (j << PAGE_SHIFT);
3266 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3273 pte = pmap_pte_quick( pmap, va);
3274 if (pte && pmap_pte_v(pte)) {
3278 m = PHYS_TO_VM_PAGE(pa);
3279 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3280 va, pa, m->hold_count, m->wire_count, m->flags);
3301 static void pads __P((pmap_t pm));
3302 void pmap_pvdump __P((vm_offset_t pa));
3304 /* print address space of pmap*/
3312 if (pm == kernel_pmap)
3314 for (i = 0; i < 1024; i++)
3316 for (j = 0; j < 1024; j++) {
3317 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3318 if (pm == kernel_pmap && va < KERNBASE)
3320 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3322 ptep = pmap_pte_quick(pm, va);
3323 if (pmap_pte_v(ptep))
3324 printf("%x:%x ", va, *(int *) ptep);
3333 register pv_entry_t pv;
3336 printf("pa %x", pa);
3337 m = PHYS_TO_VM_PAGE(pa);
3338 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3340 printf(" -> pmap %p, va %x, flags %x",
3341 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3343 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);