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 1.25 2003/11/04 01:05:28 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>
99 #include <sys/thread2.h>
101 #include <machine/cputypes.h>
102 #include <machine/md_var.h>
103 #include <machine/specialreg.h>
104 #if defined(SMP) || defined(APIC_IO)
105 #include <machine/smp.h>
106 #include <machine/apic.h>
107 #endif /* SMP || APIC_IO */
108 #include <machine/globaldata.h>
110 #define PMAP_KEEP_PDIRS
111 #ifndef PMAP_SHPGPERPROC
112 #define PMAP_SHPGPERPROC 200
115 #if defined(DIAGNOSTIC)
116 #define PMAP_DIAGNOSTIC
121 #if !defined(PMAP_DIAGNOSTIC)
122 #define PMAP_INLINE __inline
128 * Get PDEs and PTEs for user/kernel address space
130 #define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
131 #define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
133 #define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0)
134 #define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0)
135 #define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0)
136 #define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0)
137 #define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0)
139 #define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
140 #define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
143 * Given a map and a machine independent protection code,
144 * convert to a vax protection code.
146 #define pte_prot(m, p) (protection_codes[p])
147 static int protection_codes[8];
149 static struct pmap kernel_pmap_store;
152 vm_paddr_t avail_start; /* PA of first available physical page */
153 vm_paddr_t avail_end; /* PA of last available physical page */
154 vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */
155 vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */
156 static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
157 static int pgeflag; /* PG_G or-in */
158 static int pseflag; /* PG_PS or-in */
160 static vm_object_t kptobj;
163 vm_offset_t kernel_vm_end;
166 * Data for the pv entry allocation mechanism
168 static vm_zone_t pvzone;
169 static struct vm_zone pvzone_store;
170 static struct vm_object pvzone_obj;
171 static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
172 static int pmap_pagedaemon_waken = 0;
173 static struct pv_entry *pvinit;
176 * All those kernel PT submaps that BSD is so fond of
178 pt_entry_t *CMAP1 = 0, *ptmmap;
179 caddr_t CADDR1 = 0, ptvmmap = 0;
180 static pt_entry_t *msgbufmap;
181 struct msgbuf *msgbufp=0;
186 static pt_entry_t *pt_crashdumpmap;
187 static caddr_t crashdumpmap;
189 extern pt_entry_t *SMPpt;
191 static PMAP_INLINE void free_pv_entry (pv_entry_t pv);
192 static unsigned * get_ptbase (pmap_t pmap);
193 static pv_entry_t get_pv_entry (void);
194 static void i386_protection_init (void);
195 static __inline void pmap_changebit (vm_page_t m, int bit, boolean_t setem);
197 static void pmap_remove_all (vm_page_t m);
198 static vm_page_t pmap_enter_quick (pmap_t pmap, vm_offset_t va,
199 vm_page_t m, vm_page_t mpte);
200 static int pmap_remove_pte (struct pmap *pmap, unsigned *ptq,
202 static void pmap_remove_page (struct pmap *pmap, vm_offset_t va);
203 static int pmap_remove_entry (struct pmap *pmap, vm_page_t m,
205 static boolean_t pmap_testbit (vm_page_t m, int bit);
206 static void pmap_insert_entry (pmap_t pmap, vm_offset_t va,
207 vm_page_t mpte, vm_page_t m);
209 static vm_page_t pmap_allocpte (pmap_t pmap, vm_offset_t va);
211 static int pmap_release_free_page (pmap_t pmap, vm_page_t p);
212 static vm_page_t _pmap_allocpte (pmap_t pmap, unsigned ptepindex);
213 static unsigned * pmap_pte_quick (pmap_t pmap, vm_offset_t va);
214 static vm_page_t pmap_page_lookup (vm_object_t object, vm_pindex_t pindex);
215 static int pmap_unuse_pt (pmap_t, vm_offset_t, vm_page_t);
216 static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
218 static unsigned pdir4mb;
221 * Move the kernel virtual free pointer to the next
222 * 4MB. This is used to help improve performance
223 * by using a large (4MB) page for much of the kernel
224 * (.text, .data, .bss)
227 pmap_kmem_choose(vm_offset_t addr)
229 vm_offset_t newaddr = addr;
231 if (cpu_feature & CPUID_PSE) {
232 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
241 * Extract the page table entry associated with the given map/virtual
244 * This function may NOT be called from an interrupt.
246 PMAP_INLINE unsigned *
247 pmap_pte(pmap_t pmap, vm_offset_t va)
252 pdeaddr = (unsigned *) pmap_pde(pmap, va);
253 if (*pdeaddr & PG_PS)
256 return get_ptbase(pmap) + i386_btop(va);
265 * Super fast pmap_pte routine best used when scanning the pv lists.
266 * This eliminates many course-grained invltlb calls. Note that many of
267 * the pv list scans are across different pmaps and it is very wasteful
268 * to do an entire invltlb when checking a single mapping.
270 * Should only be called while splvm() is held or from a critical
274 pmap_pte_quick(pmap_t pmap, vm_offset_t va)
276 struct mdglobaldata *gd = mdcpu;
279 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
280 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
281 unsigned index = i386_btop(va);
282 /* are we current address space or kernel? */
283 if ((pmap == kernel_pmap) ||
284 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
285 return (unsigned *) PTmap + index;
287 newpf = pde & PG_FRAME;
288 if ( ((* (unsigned *) gd->gd_PMAP1) & PG_FRAME) != newpf) {
289 * (unsigned *) gd->gd_PMAP1 = newpf | PG_RW | PG_V;
290 cpu_invlpg(gd->gd_PADDR1);
292 return gd->gd_PADDR1 + ((unsigned) index & (NPTEPG - 1));
299 * Bootstrap the system enough to run with virtual memory.
301 * On the i386 this is called after mapping has already been enabled
302 * and just syncs the pmap module with what has already been done.
303 * [We can't call it easily with mapping off since the kernel is not
304 * mapped with PA == VA, hence we would have to relocate every address
305 * from the linked base (virtual) address "KERNBASE" to the actual
306 * (physical) address starting relative to 0]
309 pmap_bootstrap(firstaddr, loadaddr)
310 vm_paddr_t firstaddr;
315 struct mdglobaldata *gd;
318 avail_start = firstaddr;
321 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too
322 * large. It should instead be correctly calculated in locore.s and
323 * not based on 'first' (which is a physical address, not a virtual
324 * address, for the start of unused physical memory). The kernel
325 * page tables are NOT double mapped and thus should not be included
326 * in this calculation.
328 virtual_avail = (vm_offset_t) KERNBASE + firstaddr;
329 virtual_avail = pmap_kmem_choose(virtual_avail);
331 virtual_end = VM_MAX_KERNEL_ADDRESS;
334 * Initialize protection array.
336 i386_protection_init();
339 * The kernel's pmap is statically allocated so we don't have to use
340 * pmap_create, which is unlikely to work correctly at this part of
341 * the boot sequence (XXX and which no longer exists).
343 kernel_pmap = &kernel_pmap_store;
345 kernel_pmap->pm_pdir = (pd_entry_t *)(KERNBASE + (u_int)IdlePTD);
346 kernel_pmap->pm_count = 1;
347 kernel_pmap->pm_active = -1; /* don't allow deactivation */
348 TAILQ_INIT(&kernel_pmap->pm_pvlist);
352 * Reserve some special page table entries/VA space for temporary
355 #define SYSMAP(c, p, v, n) \
356 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
359 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va);
362 * CMAP1/CMAP2 are used for zeroing and copying pages.
364 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
369 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
372 * ptvmmap is used for reading arbitrary physical pages via
375 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
378 * msgbufp is used to map the system message buffer.
379 * XXX msgbufmap is not used.
381 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
382 atop(round_page(MSGBUF_SIZE)))
387 for (i = 0; i < NKPT; i++)
391 * PG_G is terribly broken on SMP because we IPI invltlb's in some
392 * cases rather then invl1pg. Actually, I don't even know why it
393 * works under UP because self-referential page table mappings
398 if (cpu_feature & CPUID_PGE)
403 * Initialize the 4MB page size flag
407 * The 4MB page version of the initial
408 * kernel page mapping.
412 #if !defined(DISABLE_PSE)
413 if (cpu_feature & CPUID_PSE) {
416 * Note that we have enabled PSE mode
419 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
420 ptditmp &= ~(NBPDR - 1);
421 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
426 * Enable the PSE mode. If we are SMP we can't do this
427 * now because the APs will not be able to use it when
430 load_cr4(rcr4() | CR4_PSE);
433 * We can do the mapping here for the single processor
434 * case. We simply ignore the old page table page from
438 * For SMP, we still need 4K pages to bootstrap APs,
439 * PSE will be enabled as soon as all APs are up.
441 PTD[KPTDI] = (pd_entry_t)ptditmp;
442 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t)ptditmp;
448 if (cpu_apic_address == 0)
449 panic("pmap_bootstrap: no local apic!");
451 /* local apic is mapped on last page */
452 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
453 (cpu_apic_address & PG_FRAME));
456 /* BSP does this itself, AP's get it pre-set */
457 gd = &CPU_prvspace[0].mdglobaldata;
458 gd->gd_CMAP1 = &SMPpt[1];
459 gd->gd_CMAP2 = &SMPpt[2];
460 gd->gd_CMAP3 = &SMPpt[3];
461 gd->gd_PMAP1 = &SMPpt[4];
462 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
463 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
464 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
465 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
472 * Set 4mb pdir for mp startup
477 if (pseflag && (cpu_feature & CPUID_PSE)) {
478 load_cr4(rcr4() | CR4_PSE);
479 if (pdir4mb && mycpu->gd_cpuid == 0) { /* only on BSP */
480 kernel_pmap->pm_pdir[KPTDI] =
481 PTD[KPTDI] = (pd_entry_t)pdir4mb;
489 * Initialize the pmap module.
490 * Called by vm_init, to initialize any structures that the pmap
491 * system needs to map virtual memory.
492 * pmap_init has been enhanced to support in a fairly consistant
493 * way, discontiguous physical memory.
496 pmap_init(phys_start, phys_end)
497 vm_paddr_t phys_start, phys_end;
503 * object for kernel page table pages
505 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
508 * Allocate memory for random pmap data structures. Includes the
512 for(i = 0; i < vm_page_array_size; i++) {
515 m = &vm_page_array[i];
516 TAILQ_INIT(&m->md.pv_list);
517 m->md.pv_list_count = 0;
521 * init the pv free list
523 initial_pvs = vm_page_array_size;
524 if (initial_pvs < MINPV)
526 pvzone = &pvzone_store;
527 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
528 initial_pvs * sizeof (struct pv_entry));
529 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
533 * Now it is safe to enable pv_table recording.
535 pmap_initialized = TRUE;
539 * Initialize the address space (zone) for the pv_entries. Set a
540 * high water mark so that the system can recover from excessive
541 * numbers of pv entries.
546 int shpgperproc = PMAP_SHPGPERPROC;
548 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
549 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
550 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
551 pv_entry_high_water = 9 * (pv_entry_max / 10);
552 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
556 /***************************************************
557 * Low level helper routines.....
558 ***************************************************/
560 #if defined(PMAP_DIAGNOSTIC)
563 * This code checks for non-writeable/modified pages.
564 * This should be an invalid condition.
567 pmap_nw_modified(pt_entry_t ptea)
573 if ((pte & (PG_M|PG_RW)) == PG_M)
582 * this routine defines the region(s) of memory that should
583 * not be tested for the modified bit.
585 static PMAP_INLINE int
586 pmap_track_modified(vm_offset_t va)
588 if ((va < clean_sva) || (va >= clean_eva))
594 static PMAP_INLINE void
595 invltlb_1pg(vm_offset_t va)
597 #if defined(I386_CPU)
598 if (cpu_class == CPUCLASS_386) {
608 pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
611 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
612 cpu_invlpg((void *)va);
613 if (pmap->pm_active & mycpu->gd_other_cpus)
622 pmap_TLB_invalidate_all(pmap_t pmap)
625 if (pmap->pm_active & (1 << mycpu->gd_cpuid))
627 if (pmap->pm_active & mycpu->gd_other_cpus)
636 get_ptbase(pmap_t pmap)
638 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
639 struct globaldata *gd = mycpu;
641 /* are we current address space or kernel? */
642 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
643 return (unsigned *) PTmap;
646 /* otherwise, we are alternate address space */
647 KKASSERT(gd->gd_intr_nesting_level == 0 && (gd->gd_curthread->td_flags & TDF_INTTHREAD) == 0);
649 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
650 APTDpde = (pd_entry_t)(frame | PG_RW | PG_V);
652 /* The page directory is not shared between CPUs */
658 return (unsigned *) APTmap;
664 * Extract the physical page address associated with the map/VA pair.
666 * This function may not be called from an interrupt if the pmap is
670 pmap_extract(pmap_t pmap, vm_offset_t va)
673 vm_offset_t pdirindex;
675 pdirindex = va >> PDRSHIFT;
676 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
678 if ((rtval & PG_PS) != 0) {
679 rtval &= ~(NBPDR - 1);
680 rtval |= va & (NBPDR - 1);
683 pte = get_ptbase(pmap) + i386_btop(va);
684 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
691 /***************************************************
692 * Low level mapping routines.....
693 ***************************************************/
696 * add a wired page to the kva
697 * note that in order for the mapping to take effect -- you
698 * should do a invltlb after doing the pmap_kenter...
701 pmap_kenter(vm_offset_t va, vm_paddr_t pa)
706 npte = pa | PG_RW | PG_V | pgeflag;
707 pte = (unsigned *)vtopte(va);
714 * remove a page from the kernel pagetables
717 pmap_kremove(vm_offset_t va)
721 pte = (unsigned *)vtopte(va);
727 * Used to map a range of physical addresses into kernel
728 * virtual address space.
730 * For now, VM is already on, we only need to map the
734 pmap_map(vm_offset_t virt, vm_paddr_t start, vm_paddr_t end, int prot)
736 while (start < end) {
737 pmap_kenter(virt, start);
746 * Add a list of wired pages to the kva
747 * this routine is only used for temporary
748 * kernel mappings that do not need to have
749 * page modification or references recorded.
750 * Note that old mappings are simply written
751 * over. The page *must* be wired.
754 pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
758 end_va = va + count * PAGE_SIZE;
760 while (va < end_va) {
763 pte = (unsigned *)vtopte(va);
764 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
766 cpu_invlpg((void *)va);
779 * this routine jerks page mappings from the
780 * kernel -- it is meant only for temporary mappings.
783 pmap_qremove(vm_offset_t va, int count)
787 end_va = va + count*PAGE_SIZE;
789 while (va < end_va) {
792 pte = (unsigned *)vtopte(va);
795 cpu_invlpg((void *)va);
807 pmap_page_lookup(vm_object_t object, vm_pindex_t 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.
819 * NOTE! the new thread's cpu may not equal the current cpu.
822 pmap_init_thread(thread_t td)
824 td->td_pcb = (struct pcb *)(td->td_kstack + UPAGES * PAGE_SIZE) - 1;
825 td->td_sp = (char *)td->td_pcb - 16;
829 * Create the UPAGES for a new process.
830 * This routine directly affects the fork perf for a process.
833 pmap_init_proc(struct proc *p, struct thread *td)
835 p->p_addr = (void *)td->td_kstack;
838 td->td_switch = cpu_heavy_switch;
842 bzero(p->p_addr, sizeof(*p->p_addr));
846 * Dispose the UPAGES for a process that has exited.
847 * This routine directly impacts the exit perf of a process.
850 pmap_dispose_proc(struct proc *p)
854 KASSERT(p->p_lock == 0, ("attempt to dispose referenced proc! %p", p));
856 if ((td = p->p_thread) != NULL) {
865 * Allow the UPAGES for a process to be prejudicially paged out.
868 pmap_swapout_proc(struct proc *p)
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.
893 pmap_swapin_proc(struct proc *p)
900 upobj = p->p_upages_obj;
901 for(i=0;i<UPAGES;i++) {
903 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
905 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
908 if (m->valid != VM_PAGE_BITS_ALL) {
909 rv = vm_pager_get_pages(upobj, &m, 1, 0);
910 if (rv != VM_PAGER_OK)
911 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
912 m = vm_page_lookup(upobj, i);
913 m->valid = VM_PAGE_BITS_ALL;
918 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
923 /***************************************************
924 * Page table page management routines.....
925 ***************************************************/
928 * This routine unholds page table pages, and if the hold count
929 * drops to zero, then it decrements the wire count.
932 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
934 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
937 if (m->hold_count == 0) {
940 * unmap the page table page
942 pmap->pm_pdir[m->pindex] = 0;
943 --pmap->pm_stats.resident_count;
944 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
945 (((unsigned) PTDpde) & PG_FRAME)) {
947 * Do a invltlb to make the invalidated mapping
948 * take effect immediately.
950 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
951 pmap_TLB_invalidate(pmap, pteva);
954 if (pmap->pm_ptphint == m)
955 pmap->pm_ptphint = NULL;
958 * If the page is finally unwired, simply free it.
961 if (m->wire_count == 0) {
965 vm_page_free_zero(m);
966 --vmstats.v_wire_count;
973 static PMAP_INLINE int
974 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
977 if (m->hold_count == 0)
978 return _pmap_unwire_pte_hold(pmap, m);
984 * After removing a page table entry, this routine is used to
985 * conditionally free the page, and manage the hold/wire counts.
988 pmap_unuse_pt(pmap_t pmap, vm_offset_t va, vm_page_t mpte)
991 if (va >= UPT_MIN_ADDRESS)
995 ptepindex = (va >> PDRSHIFT);
996 if (pmap->pm_ptphint &&
997 (pmap->pm_ptphint->pindex == ptepindex)) {
998 mpte = pmap->pm_ptphint;
1000 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1001 pmap->pm_ptphint = mpte;
1005 return pmap_unwire_pte_hold(pmap, mpte);
1009 pmap_pinit0(struct pmap *pmap)
1012 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1013 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1015 pmap->pm_active = 0;
1016 pmap->pm_ptphint = NULL;
1017 TAILQ_INIT(&pmap->pm_pvlist);
1018 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1022 * Initialize a preallocated and zeroed pmap structure,
1023 * such as one in a vmspace structure.
1026 pmap_pinit(struct pmap *pmap)
1031 * No need to allocate page table space yet but we do need a valid
1032 * page directory table.
1034 if (pmap->pm_pdir == NULL) {
1036 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1040 * allocate object for the ptes
1042 if (pmap->pm_pteobj == NULL)
1043 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1046 * allocate the page directory page
1048 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1049 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1051 ptdpg->wire_count = 1;
1052 ++vmstats.v_wire_count;
1055 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1056 ptdpg->valid = VM_PAGE_BITS_ALL;
1058 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1059 if ((ptdpg->flags & PG_ZERO) == 0)
1060 bzero(pmap->pm_pdir, PAGE_SIZE);
1062 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1064 /* install self-referential address mapping entry */
1065 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1066 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1069 pmap->pm_active = 0;
1070 pmap->pm_ptphint = NULL;
1071 TAILQ_INIT(&pmap->pm_pvlist);
1072 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1076 * Wire in kernel global address entries. To avoid a race condition
1077 * between pmap initialization and pmap_growkernel, this procedure
1078 * should be called after the vmspace is attached to the process
1079 * but before this pmap is activated.
1082 pmap_pinit2(struct pmap *pmap)
1084 /* XXX copies current process, does not fill in MPPTDI */
1085 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1089 pmap_release_free_page(struct pmap *pmap, vm_page_t p)
1091 unsigned *pde = (unsigned *) pmap->pm_pdir;
1093 * This code optimizes the case of freeing non-busy
1094 * page-table pages. Those pages are zero now, and
1095 * might as well be placed directly into the zero queue.
1097 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1103 * Remove the page table page from the processes address space.
1106 pmap->pm_stats.resident_count--;
1108 if (p->hold_count) {
1109 panic("pmap_release: freeing held page table page");
1112 * Page directory pages need to have the kernel
1113 * stuff cleared, so they can go into the zero queue also.
1115 if (p->pindex == PTDPTDI) {
1116 bzero(pde + KPTDI, nkpt * PTESIZE);
1119 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1122 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1123 pmap->pm_ptphint = NULL;
1126 vmstats.v_wire_count--;
1127 vm_page_free_zero(p);
1132 * this routine is called if the page table page is not
1136 _pmap_allocpte(pmap_t pmap, unsigned ptepindex)
1138 vm_offset_t pteva, ptepa;
1142 * Find or fabricate a new pagetable page
1144 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1145 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1147 KASSERT(m->queue == PQ_NONE,
1148 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1150 if (m->wire_count == 0)
1151 vmstats.v_wire_count++;
1155 * Increment the hold count for the page table page
1156 * (denoting a new mapping.)
1161 * Map the pagetable page into the process address space, if
1162 * it isn't already there.
1165 pmap->pm_stats.resident_count++;
1167 ptepa = VM_PAGE_TO_PHYS(m);
1168 pmap->pm_pdir[ptepindex] =
1169 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1172 * Set the page table hint
1174 pmap->pm_ptphint = m;
1177 * Try to use the new mapping, but if we cannot, then
1178 * do it with the routine that maps the page explicitly.
1180 if ((m->flags & PG_ZERO) == 0) {
1181 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1182 (((unsigned) PTDpde) & PG_FRAME)) {
1183 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1184 bzero((caddr_t) pteva, PAGE_SIZE);
1186 pmap_zero_page(ptepa);
1190 m->valid = VM_PAGE_BITS_ALL;
1191 vm_page_flag_clear(m, PG_ZERO);
1192 vm_page_flag_set(m, PG_MAPPED);
1199 pmap_allocpte(pmap_t pmap, vm_offset_t va)
1206 * Calculate pagetable page index
1208 ptepindex = va >> PDRSHIFT;
1211 * Get the page directory entry
1213 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1216 * This supports switching from a 4MB page to a
1219 if (ptepa & PG_PS) {
1220 pmap->pm_pdir[ptepindex] = 0;
1226 * If the page table page is mapped, we just increment the
1227 * hold count, and activate it.
1231 * In order to get the page table page, try the
1234 if (pmap->pm_ptphint &&
1235 (pmap->pm_ptphint->pindex == ptepindex)) {
1236 m = pmap->pm_ptphint;
1238 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1239 pmap->pm_ptphint = m;
1245 * Here if the pte page isn't mapped, or if it has been deallocated.
1247 return _pmap_allocpte(pmap, ptepindex);
1251 /***************************************************
1252 * Pmap allocation/deallocation routines.
1253 ***************************************************/
1256 * Release any resources held by the given physical map.
1257 * Called when a pmap initialized by pmap_pinit is being released.
1258 * Should only be called if the map contains no valid mappings.
1261 pmap_release(struct pmap *pmap)
1263 vm_page_t p,n,ptdpg;
1264 vm_object_t object = pmap->pm_pteobj;
1267 #if defined(DIAGNOSTIC)
1268 if (object->ref_count != 1)
1269 panic("pmap_release: pteobj reference count != 1");
1274 curgeneration = object->generation;
1275 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1276 n = TAILQ_NEXT(p, listq);
1277 if (p->pindex == PTDPTDI) {
1282 if (!pmap_release_free_page(pmap, p) &&
1283 (object->generation != curgeneration))
1288 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1293 kvm_size(SYSCTL_HANDLER_ARGS)
1295 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1297 return sysctl_handle_long(oidp, &ksize, 0, req);
1299 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1300 0, 0, kvm_size, "IU", "Size of KVM");
1303 kvm_free(SYSCTL_HANDLER_ARGS)
1305 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1307 return sysctl_handle_long(oidp, &kfree, 0, req);
1309 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1310 0, 0, kvm_free, "IU", "Amount of KVM free");
1313 * grow the number of kernel page table entries, if needed
1316 pmap_growkernel(vm_offset_t addr)
1321 vm_offset_t ptppaddr;
1326 if (kernel_vm_end == 0) {
1327 kernel_vm_end = KERNBASE;
1329 while (pdir_pde(PTD, kernel_vm_end)) {
1330 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1334 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1335 while (kernel_vm_end < addr) {
1336 if (pdir_pde(PTD, kernel_vm_end)) {
1337 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1342 * This index is bogus, but out of the way
1344 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1346 panic("pmap_growkernel: no memory to grow kernel");
1351 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1352 pmap_zero_page(ptppaddr);
1353 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1354 pdir_pde(PTD, kernel_vm_end) = newpdir;
1356 FOREACH_PROC_IN_SYSTEM(p) {
1358 pmap = vmspace_pmap(p->p_vmspace);
1359 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1362 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1363 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1369 * Retire the given physical map from service.
1370 * Should only be called if the map contains
1371 * no valid mappings.
1374 pmap_destroy(pmap_t pmap)
1381 count = --pmap->pm_count;
1384 panic("destroying a pmap is not yet implemented");
1389 * Add a reference to the specified pmap.
1392 pmap_reference(pmap_t pmap)
1399 /***************************************************
1400 * page management routines.
1401 ***************************************************/
1404 * free the pv_entry back to the free list. This function may be
1405 * called from an interrupt.
1407 static PMAP_INLINE void
1408 free_pv_entry(pv_entry_t pv)
1415 * get a new pv_entry, allocating a block from the system
1416 * when needed. This function may be called from an interrupt.
1422 if (pv_entry_high_water &&
1423 (pv_entry_count > pv_entry_high_water) &&
1424 (pmap_pagedaemon_waken == 0)) {
1425 pmap_pagedaemon_waken = 1;
1426 wakeup (&vm_pages_needed);
1428 return zalloc(pvzone);
1432 * This routine is very drastic, but can save the system
1440 static int warningdone=0;
1442 if (pmap_pagedaemon_waken == 0)
1445 if (warningdone < 5) {
1446 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1450 for(i = 0; i < vm_page_array_size; i++) {
1451 m = &vm_page_array[i];
1452 if (m->wire_count || m->hold_count || m->busy ||
1453 (m->flags & PG_BUSY))
1457 pmap_pagedaemon_waken = 0;
1462 * If it is the first entry on the list, it is actually
1463 * in the header and we must copy the following entry up
1464 * to the header. Otherwise we must search the list for
1465 * the entry. In either case we free the now unused entry.
1469 pmap_remove_entry(struct pmap *pmap, vm_page_t m, vm_offset_t va)
1476 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1477 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1478 if (pmap == pv->pv_pmap && va == pv->pv_va)
1482 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1483 if (va == pv->pv_va)
1491 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1492 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1493 m->md.pv_list_count--;
1494 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1495 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1497 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1506 * Create a pv entry for page at pa for
1510 pmap_insert_entry(pmap_t pmap, vm_offset_t va, vm_page_t mpte, vm_page_t m)
1516 pv = get_pv_entry();
1521 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1522 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1523 m->md.pv_list_count++;
1529 * pmap_remove_pte: do the things to unmap a page in a process
1532 pmap_remove_pte(struct pmap *pmap, unsigned *ptq, vm_offset_t va)
1537 oldpte = loadandclear(ptq);
1539 pmap->pm_stats.wired_count -= 1;
1541 * Machines that don't support invlpg, also don't support
1546 pmap->pm_stats.resident_count -= 1;
1547 if (oldpte & PG_MANAGED) {
1548 m = PHYS_TO_VM_PAGE(oldpte);
1549 if (oldpte & PG_M) {
1550 #if defined(PMAP_DIAGNOSTIC)
1551 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1553 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1557 if (pmap_track_modified(va))
1561 vm_page_flag_set(m, PG_REFERENCED);
1562 return pmap_remove_entry(pmap, m, va);
1564 return pmap_unuse_pt(pmap, va, NULL);
1573 * Remove a single page from a process address space.
1575 * This function may not be called from an interrupt if the pmap is
1579 pmap_remove_page(struct pmap *pmap, vm_offset_t va)
1584 * if there is no pte for this address, just skip it!!! Otherwise
1585 * get a local va for mappings for this pmap and remove the entry.
1587 if (*pmap_pde(pmap, va) != 0) {
1588 ptq = get_ptbase(pmap) + i386_btop(va);
1590 (void) pmap_remove_pte(pmap, ptq, va);
1591 pmap_TLB_invalidate(pmap, va);
1599 * Remove the given range of addresses from the specified map.
1601 * It is assumed that the start and end are properly
1602 * rounded to the page size.
1604 * This function may not be called from an interrupt if the pmap is
1608 pmap_remove(struct pmap *pmap, vm_offset_t sva, vm_offset_t eva)
1612 vm_offset_t ptpaddr;
1613 vm_offset_t sindex, eindex;
1619 if (pmap->pm_stats.resident_count == 0)
1623 * special handling of removing one page. a very
1624 * common operation and easy to short circuit some
1627 if (((sva + PAGE_SIZE) == eva) &&
1628 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1629 pmap_remove_page(pmap, sva);
1636 * Get a local virtual address for the mappings that are being
1639 ptbase = get_ptbase(pmap);
1641 sindex = i386_btop(sva);
1642 eindex = i386_btop(eva);
1644 for (; sindex < eindex; sindex = pdnxt) {
1648 * Calculate index for next page table.
1650 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1651 if (pmap->pm_stats.resident_count == 0)
1654 pdirindex = sindex / NPDEPG;
1655 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1656 pmap->pm_pdir[pdirindex] = 0;
1657 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1663 * Weed out invalid mappings. Note: we assume that the page
1664 * directory table is always allocated, and in kernel virtual.
1670 * Limit our scan to either the end of the va represented
1671 * by the current page table page, or to the end of the
1672 * range being removed.
1674 if (pdnxt > eindex) {
1678 for ( ;sindex != pdnxt; sindex++) {
1680 if (ptbase[sindex] == 0) {
1683 va = i386_ptob(sindex);
1686 if (pmap_remove_pte(pmap,
1687 ptbase + sindex, va))
1693 pmap_TLB_invalidate_all(pmap);
1699 * Removes this physical page from all physical maps in which it resides.
1700 * Reflects back modify bits to the pager.
1702 * This routine may not be called from an interrupt.
1706 pmap_remove_all(vm_page_t m)
1709 unsigned *pte, tpte;
1712 #if defined(PMAP_DIAGNOSTIC)
1714 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1717 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1718 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%08llx", (long long)VM_PAGE_TO_PHYS(m));
1723 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1724 pv->pv_pmap->pm_stats.resident_count--;
1726 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1728 tpte = loadandclear(pte);
1730 pv->pv_pmap->pm_stats.wired_count--;
1733 vm_page_flag_set(m, PG_REFERENCED);
1736 * Update the vm_page_t clean and reference bits.
1739 #if defined(PMAP_DIAGNOSTIC)
1740 if (pmap_nw_modified((pt_entry_t) tpte)) {
1742 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1746 if (pmap_track_modified(pv->pv_va))
1749 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1751 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1752 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1753 m->md.pv_list_count--;
1754 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1758 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1766 * Set the physical protection on the specified range of this map
1769 * This function may not be called from an interrupt if the map is
1770 * not the kernel_pmap.
1773 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1776 vm_offset_t pdnxt, ptpaddr;
1777 vm_pindex_t sindex, eindex;
1783 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1784 pmap_remove(pmap, sva, eva);
1788 if (prot & VM_PROT_WRITE)
1793 ptbase = get_ptbase(pmap);
1795 sindex = i386_btop(sva);
1796 eindex = i386_btop(eva);
1798 for (; sindex < eindex; sindex = pdnxt) {
1802 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1804 pdirindex = sindex / NPDEPG;
1805 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1806 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1807 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1813 * Weed out invalid mappings. Note: we assume that the page
1814 * directory table is always allocated, and in kernel virtual.
1819 if (pdnxt > eindex) {
1823 for (; sindex != pdnxt; sindex++) {
1828 pbits = ptbase[sindex];
1830 if (pbits & PG_MANAGED) {
1833 m = PHYS_TO_VM_PAGE(pbits);
1834 vm_page_flag_set(m, PG_REFERENCED);
1838 if (pmap_track_modified(i386_ptob(sindex))) {
1840 m = PHYS_TO_VM_PAGE(pbits);
1849 if (pbits != ptbase[sindex]) {
1850 ptbase[sindex] = pbits;
1856 pmap_TLB_invalidate_all(pmap);
1860 * Insert the given physical page (p) at
1861 * the specified virtual address (v) in the
1862 * target physical map with the protection requested.
1864 * If specified, the page will be wired down, meaning
1865 * that the related pte can not be reclaimed.
1867 * NB: This is the only routine which MAY NOT lazy-evaluate
1868 * or lose information. That is, this routine must actually
1869 * insert this page into the given map NOW.
1872 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1878 vm_offset_t origpte, newpte;
1885 #ifdef PMAP_DIAGNOSTIC
1886 if (va > VM_MAX_KERNEL_ADDRESS)
1887 panic("pmap_enter: toobig");
1888 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
1889 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
1894 * In the case that a page table page is not
1895 * resident, we are creating it here.
1897 if (va < UPT_MIN_ADDRESS) {
1898 mpte = pmap_allocpte(pmap, va);
1900 #if 0 && defined(PMAP_DIAGNOSTIC)
1902 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
1903 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
1904 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
1905 pmap->pm_pdir[PTDPTDI], origpte, va);
1908 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
1909 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
1910 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
1911 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
1912 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
1913 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
1914 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
1920 pte = pmap_pte(pmap, va);
1923 * Page Directory table entry not valid, we need a new PT page
1926 panic("pmap_enter: invalid page directory pdir=%x, va=0x%x\n",
1927 (unsigned) pmap->pm_pdir[PTDPTDI], va);
1930 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
1931 origpte = *(vm_offset_t *)pte;
1932 opa = origpte & PG_FRAME;
1934 if (origpte & PG_PS)
1935 panic("pmap_enter: attempted pmap_enter on 4MB page");
1938 * Mapping has not changed, must be protection or wiring change.
1940 if (origpte && (opa == pa)) {
1942 * Wiring change, just update stats. We don't worry about
1943 * wiring PT pages as they remain resident as long as there
1944 * are valid mappings in them. Hence, if a user page is wired,
1945 * the PT page will be also.
1947 if (wired && ((origpte & PG_W) == 0))
1948 pmap->pm_stats.wired_count++;
1949 else if (!wired && (origpte & PG_W))
1950 pmap->pm_stats.wired_count--;
1952 #if defined(PMAP_DIAGNOSTIC)
1953 if (pmap_nw_modified((pt_entry_t) origpte)) {
1955 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
1961 * Remove extra pte reference
1966 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
1967 if ((origpte & PG_RW) == 0) {
1970 cpu_invlpg((void *)va);
1971 if (pmap->pm_active & mycpu->gd_other_cpus)
1981 * We might be turning off write access to the page,
1982 * so we go ahead and sense modify status.
1984 if (origpte & PG_MANAGED) {
1985 if ((origpte & PG_M) && pmap_track_modified(va)) {
1987 om = PHYS_TO_VM_PAGE(opa);
1995 * Mapping has changed, invalidate old range and fall through to
1996 * handle validating new mapping.
2000 err = pmap_remove_pte(pmap, pte, va);
2002 panic("pmap_enter: pte vanished, va: 0x%x", va);
2006 * Enter on the PV list if part of our managed memory. Note that we
2007 * raise IPL while manipulating pv_table since pmap_enter can be
2008 * called at interrupt time.
2010 if (pmap_initialized &&
2011 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2012 pmap_insert_entry(pmap, va, mpte, m);
2017 * Increment counters
2019 pmap->pm_stats.resident_count++;
2021 pmap->pm_stats.wired_count++;
2025 * Now validate mapping with desired protection/wiring.
2027 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2031 if (va < UPT_MIN_ADDRESS)
2033 if (pmap == kernel_pmap)
2037 * if the mapping or permission bits are different, we need
2038 * to update the pte.
2040 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2041 *pte = newpte | PG_A;
2044 cpu_invlpg((void *)va);
2045 if (pmap->pm_active & mycpu->gd_other_cpus)
2055 * this code makes some *MAJOR* assumptions:
2056 * 1. Current pmap & pmap exists.
2059 * 4. No page table pages.
2060 * 5. Tlbflush is deferred to calling procedure.
2061 * 6. Page IS managed.
2062 * but is *MUCH* faster than pmap_enter...
2066 pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_page_t mpte)
2072 * In the case that a page table page is not
2073 * resident, we are creating it here.
2075 if (va < UPT_MIN_ADDRESS) {
2080 * Calculate pagetable page index
2082 ptepindex = va >> PDRSHIFT;
2083 if (mpte && (mpte->pindex == ptepindex)) {
2088 * Get the page directory entry
2090 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2093 * If the page table page is mapped, we just increment
2094 * the hold count, and activate it.
2098 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2099 if (pmap->pm_ptphint &&
2100 (pmap->pm_ptphint->pindex == ptepindex)) {
2101 mpte = pmap->pm_ptphint;
2103 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2104 pmap->pm_ptphint = mpte;
2110 mpte = _pmap_allocpte(pmap, ptepindex);
2118 * This call to vtopte makes the assumption that we are
2119 * entering the page into the current pmap. In order to support
2120 * quick entry into any pmap, one would likely use pmap_pte_quick.
2121 * But that isn't as quick as vtopte.
2123 pte = (unsigned *)vtopte(va);
2126 pmap_unwire_pte_hold(pmap, mpte);
2131 * Enter on the PV list if part of our managed memory. Note that we
2132 * raise IPL while manipulating pv_table since pmap_enter can be
2133 * called at interrupt time.
2135 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2136 pmap_insert_entry(pmap, va, mpte, m);
2139 * Increment counters
2141 pmap->pm_stats.resident_count++;
2143 pa = VM_PAGE_TO_PHYS(m);
2146 * Now validate mapping with RO protection
2148 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2149 *pte = pa | PG_V | PG_U;
2151 *pte = pa | PG_V | PG_U | PG_MANAGED;
2157 * Make a temporary mapping for a physical address. This is only intended
2158 * to be used for panic dumps.
2161 pmap_kenter_temporary(vm_paddr_t pa, int i)
2163 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2164 return ((void *)crashdumpmap);
2167 #define MAX_INIT_PT (96)
2169 * pmap_object_init_pt preloads the ptes for a given object
2170 * into the specified pmap. This eliminates the blast of soft
2171 * faults on process startup and immediately after an mmap.
2174 pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_object_t object,
2175 vm_pindex_t pindex, vm_size_t size, int limit)
2182 if (pmap == NULL || object == NULL)
2186 * This code maps large physical mmap regions into the
2187 * processor address space. Note that some shortcuts
2188 * are taken, but the code works.
2191 (object->type == OBJT_DEVICE) &&
2192 ((addr & (NBPDR - 1)) == 0) &&
2193 ((size & (NBPDR - 1)) == 0) ) {
2196 unsigned int ptepindex;
2200 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2204 p = vm_page_lookup(object, pindex);
2205 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2209 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2214 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2219 p = vm_page_lookup(object, pindex);
2223 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2224 if (ptepa & (NBPDR - 1)) {
2228 p->valid = VM_PAGE_BITS_ALL;
2230 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2231 npdes = size >> PDRSHIFT;
2232 for(i=0;i<npdes;i++) {
2233 pmap->pm_pdir[ptepindex] =
2234 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2238 vm_page_flag_set(p, PG_MAPPED);
2243 psize = i386_btop(size);
2245 if ((object->type != OBJT_VNODE) ||
2246 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2247 (object->resident_page_count > MAX_INIT_PT))) {
2251 if (psize + pindex > object->size) {
2252 if (object->size < pindex)
2254 psize = object->size - pindex;
2259 * if we are processing a major portion of the object, then scan the
2262 if (psize > (object->resident_page_count >> 2)) {
2265 for (p = TAILQ_FIRST(&object->memq);
2266 ((objpgs > 0) && (p != NULL));
2267 p = TAILQ_NEXT(p, listq)) {
2270 if (tmpidx < pindex) {
2274 if (tmpidx >= psize) {
2278 * don't allow an madvise to blow away our really
2279 * free pages allocating pv entries.
2281 if ((limit & MAP_PREFAULT_MADVISE) &&
2282 vmstats.v_free_count < vmstats.v_free_reserved) {
2285 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2287 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2288 if ((p->queue - p->pc) == PQ_CACHE)
2289 vm_page_deactivate(p);
2291 mpte = pmap_enter_quick(pmap,
2292 addr + i386_ptob(tmpidx), p, mpte);
2293 vm_page_flag_set(p, PG_MAPPED);
2300 * else lookup the pages one-by-one.
2302 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2304 * don't allow an madvise to blow away our really
2305 * free pages allocating pv entries.
2307 if ((limit & MAP_PREFAULT_MADVISE) &&
2308 vmstats.v_free_count < vmstats.v_free_reserved) {
2311 p = vm_page_lookup(object, tmpidx + pindex);
2313 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2315 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2316 if ((p->queue - p->pc) == PQ_CACHE)
2317 vm_page_deactivate(p);
2319 mpte = pmap_enter_quick(pmap,
2320 addr + i386_ptob(tmpidx), p, mpte);
2321 vm_page_flag_set(p, PG_MAPPED);
2330 * pmap_prefault provides a quick way of clustering
2331 * pagefaults into a processes address space. It is a "cousin"
2332 * of pmap_object_init_pt, except it runs at page fault time instead
2337 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2339 static int pmap_prefault_pageorder[] = {
2340 -PAGE_SIZE, PAGE_SIZE,
2341 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2342 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2343 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2347 pmap_prefault(pmap_t pmap, vm_offset_t addra, vm_map_entry_t entry)
2356 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2359 object = entry->object.vm_object;
2361 starta = addra - PFBAK * PAGE_SIZE;
2362 if (starta < entry->start) {
2363 starta = entry->start;
2364 } else if (starta > addra) {
2369 for (i = 0; i < PAGEORDER_SIZE; i++) {
2370 vm_object_t lobject;
2373 addr = addra + pmap_prefault_pageorder[i];
2374 if (addr > addra + (PFFOR * PAGE_SIZE))
2377 if (addr < starta || addr >= entry->end)
2380 if ((*pmap_pde(pmap, addr)) == NULL)
2383 pte = (unsigned *) vtopte(addr);
2387 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2389 for (m = vm_page_lookup(lobject, pindex);
2390 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2391 lobject = lobject->backing_object) {
2392 if (lobject->backing_object_offset & PAGE_MASK)
2394 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2395 m = vm_page_lookup(lobject->backing_object, pindex);
2399 * give-up when a page is not in memory
2404 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2406 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2408 if ((m->queue - m->pc) == PQ_CACHE) {
2409 vm_page_deactivate(m);
2412 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2413 vm_page_flag_set(m, PG_MAPPED);
2420 * Routine: pmap_change_wiring
2421 * Function: Change the wiring attribute for a map/virtual-address
2423 * In/out conditions:
2424 * The mapping must already exist in the pmap.
2427 pmap_change_wiring(pmap_t pmap, vm_offset_t va, boolean_t wired)
2434 pte = pmap_pte(pmap, va);
2436 if (wired && !pmap_pte_w(pte))
2437 pmap->pm_stats.wired_count++;
2438 else if (!wired && pmap_pte_w(pte))
2439 pmap->pm_stats.wired_count--;
2442 * Wiring is not a hardware characteristic so there is no need to
2445 pmap_pte_set_w(pte, wired);
2451 * Copy the range specified by src_addr/len
2452 * from the source map to the range dst_addr/len
2453 * in the destination map.
2455 * This routine is only advisory and need not do anything.
2458 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
2459 vm_size_t len, vm_offset_t src_addr)
2462 vm_offset_t end_addr = src_addr + len;
2464 unsigned src_frame, dst_frame;
2467 if (dst_addr != src_addr)
2470 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2471 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2475 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2476 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2477 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2479 /* The page directory is not shared between CPUs */
2486 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2487 unsigned *src_pte, *dst_pte;
2488 vm_page_t dstmpte, srcmpte;
2489 vm_offset_t srcptepaddr;
2492 if (addr >= UPT_MIN_ADDRESS)
2493 panic("pmap_copy: invalid to pmap_copy page tables\n");
2496 * Don't let optional prefaulting of pages make us go
2497 * way below the low water mark of free pages or way
2498 * above high water mark of used pv entries.
2500 if (vmstats.v_free_count < vmstats.v_free_reserved ||
2501 pv_entry_count > pv_entry_high_water)
2504 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2505 ptepindex = addr >> PDRSHIFT;
2507 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2508 if (srcptepaddr == 0)
2511 if (srcptepaddr & PG_PS) {
2512 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2513 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2514 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2519 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2520 if ((srcmpte == NULL) ||
2521 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2524 if (pdnxt > end_addr)
2527 src_pte = (unsigned *) vtopte(addr);
2528 dst_pte = (unsigned *) avtopte(addr);
2529 while (addr < pdnxt) {
2533 * we only virtual copy managed pages
2535 if ((ptetemp & PG_MANAGED) != 0) {
2537 * We have to check after allocpte for the
2538 * pte still being around... allocpte can
2541 dstmpte = pmap_allocpte(dst_pmap, addr);
2542 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2544 * Clear the modified and
2545 * accessed (referenced) bits
2548 m = PHYS_TO_VM_PAGE(ptetemp);
2549 *dst_pte = ptetemp & ~(PG_M | PG_A);
2550 dst_pmap->pm_stats.resident_count++;
2551 pmap_insert_entry(dst_pmap, addr,
2554 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2556 if (dstmpte->hold_count >= srcmpte->hold_count)
2567 * Routine: pmap_kernel
2569 * Returns the physical map handle for the kernel.
2574 return (kernel_pmap);
2580 * Zero the specified PA by mapping the page into KVM and clearing its
2583 * This function may be called from an interrupt and no locking is
2587 pmap_zero_page(vm_paddr_t phys)
2589 struct mdglobaldata *gd = mdcpu;
2592 if (*(int *)gd->gd_CMAP3)
2593 panic("pmap_zero_page: CMAP3 busy");
2594 *(int *)gd->gd_CMAP3 =
2595 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2596 cpu_invlpg(gd->gd_CADDR3);
2598 #if defined(I686_CPU)
2599 if (cpu_class == CPUCLASS_686)
2600 i686_pagezero(gd->gd_CADDR3);
2603 bzero(gd->gd_CADDR3, PAGE_SIZE);
2604 *(int *) gd->gd_CMAP3 = 0;
2611 * Zero part of a physical page by mapping it into memory and clearing
2612 * its contents with bzero.
2614 * off and size may not cover an area beyond a single hardware page.
2617 pmap_zero_page_area(vm_paddr_t phys, int off, int size)
2619 struct mdglobaldata *gd = mdcpu;
2622 if (*(int *) gd->gd_CMAP3)
2623 panic("pmap_zero_page: CMAP3 busy");
2624 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2625 cpu_invlpg(gd->gd_CADDR3);
2627 #if defined(I686_CPU)
2628 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2629 i686_pagezero(gd->gd_CADDR3);
2632 bzero((char *)gd->gd_CADDR3 + off, size);
2633 *(int *) gd->gd_CMAP3 = 0;
2640 * Copy the physical page from the source PA to the target PA.
2641 * This function may be called from an interrupt. No locking
2645 pmap_copy_page(vm_paddr_t src, vm_paddr_t dst)
2647 struct mdglobaldata *gd = mdcpu;
2650 if (*(int *) gd->gd_CMAP1)
2651 panic("pmap_copy_page: CMAP1 busy");
2652 if (*(int *) gd->gd_CMAP2)
2653 panic("pmap_copy_page: CMAP2 busy");
2655 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2656 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2658 cpu_invlpg(gd->gd_CADDR1);
2659 cpu_invlpg(gd->gd_CADDR2);
2661 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
2663 *(int *) gd->gd_CMAP1 = 0;
2664 *(int *) gd->gd_CMAP2 = 0;
2670 * Routine: pmap_pageable
2672 * Make the specified pages (by pmap, offset)
2673 * pageable (or not) as requested.
2675 * A page which is not pageable may not take
2676 * a fault; therefore, its page table entry
2677 * must remain valid for the duration.
2679 * This routine is merely advisory; pmap_enter
2680 * will specify that these pages are to be wired
2681 * down (or not) as appropriate.
2684 pmap_pageable(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, boolean_t pageable)
2689 * Returns true if the pmap's pv is one of the first
2690 * 16 pvs linked to from this page. This count may
2691 * be changed upwards or downwards in the future; it
2692 * is only necessary that true be returned for a small
2693 * subset of pmaps for proper page aging.
2696 pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
2702 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2707 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2708 if (pv->pv_pmap == pmap) {
2720 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2722 * Remove all pages from specified address space
2723 * this aids process exit speeds. Also, this code
2724 * is special cased for current process only, but
2725 * can have the more generic (and slightly slower)
2726 * mode enabled. This is much faster than pmap_remove
2727 * in the case of running down an entire address space.
2730 pmap_remove_pages(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2732 unsigned *pte, tpte;
2737 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2738 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2739 printf("warning: pmap_remove_pages called with non-current pmap\n");
2745 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2749 if (pv->pv_va >= eva || pv->pv_va < sva) {
2750 npv = TAILQ_NEXT(pv, pv_plist);
2754 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2755 pte = (unsigned *)vtopte(pv->pv_va);
2757 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2762 * We cannot remove wired pages from a process' mapping at this time
2765 npv = TAILQ_NEXT(pv, pv_plist);
2770 m = PHYS_TO_VM_PAGE(tpte);
2772 KASSERT(m < &vm_page_array[vm_page_array_size],
2773 ("pmap_remove_pages: bad tpte %x", tpte));
2775 pv->pv_pmap->pm_stats.resident_count--;
2778 * Update the vm_page_t clean and reference bits.
2785 npv = TAILQ_NEXT(pv, pv_plist);
2786 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2788 m->md.pv_list_count--;
2789 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2790 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2791 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2794 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2798 pmap_TLB_invalidate_all(pmap);
2802 * pmap_testbit tests bits in pte's
2803 * note that the testbit/changebit routines are inline,
2804 * and a lot of things compile-time evaluate.
2807 pmap_testbit(vm_page_t m, int bit)
2813 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2816 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
2821 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2823 * if the bit being tested is the modified bit, then
2824 * mark clean_map and ptes as never
2827 if (bit & (PG_A|PG_M)) {
2828 if (!pmap_track_modified(pv->pv_va))
2832 #if defined(PMAP_DIAGNOSTIC)
2834 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
2838 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2849 * this routine is used to modify bits in ptes
2851 static __inline void
2852 pmap_changebit(vm_page_t m, int bit, boolean_t setem)
2858 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2864 * Loop over all current mappings setting/clearing as appropos If
2865 * setting RO do we need to clear the VAC?
2867 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2869 * don't write protect pager mappings
2871 if (!setem && (bit == PG_RW)) {
2872 if (!pmap_track_modified(pv->pv_va))
2876 #if defined(PMAP_DIAGNOSTIC)
2878 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
2883 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2887 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2889 vm_offset_t pbits = *(vm_offset_t *)pte;
2895 *(int *)pte = pbits & ~(PG_M|PG_RW);
2897 *(int *)pte = pbits & ~bit;
2899 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2907 * pmap_page_protect:
2909 * Lower the permission for all mappings to a given page.
2912 pmap_page_protect(vm_page_t m, vm_prot_t prot)
2914 if ((prot & VM_PROT_WRITE) == 0) {
2915 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
2916 pmap_changebit(m, PG_RW, FALSE);
2924 pmap_phys_address(int ppn)
2926 return (i386_ptob(ppn));
2930 * pmap_ts_referenced:
2932 * Return a count of reference bits for a page, clearing those bits.
2933 * It is not necessary for every reference bit to be cleared, but it
2934 * is necessary that 0 only be returned when there are truly no
2935 * reference bits set.
2937 * XXX: The exact number of bits to check and clear is a matter that
2938 * should be tested and standardized at some point in the future for
2939 * optimal aging of shared pages.
2942 pmap_ts_referenced(vm_page_t m)
2944 pv_entry_t pv, pvf, pvn;
2949 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2954 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
2959 pvn = TAILQ_NEXT(pv, pv_list);
2961 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2963 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
2965 if (!pmap_track_modified(pv->pv_va))
2968 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2970 if (pte && (*pte & PG_A)) {
2973 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
2980 } while ((pv = pvn) != NULL && pv != pvf);
2990 * Return whether or not the specified physical page was modified
2991 * in any physical maps.
2994 pmap_is_modified(vm_page_t m)
2996 return pmap_testbit(m, PG_M);
3000 * Clear the modify bits on the specified physical page.
3003 pmap_clear_modify(vm_page_t m)
3005 pmap_changebit(m, PG_M, FALSE);
3009 * pmap_clear_reference:
3011 * Clear the reference bit on the specified physical page.
3014 pmap_clear_reference(vm_page_t m)
3016 pmap_changebit(m, PG_A, FALSE);
3020 * Miscellaneous support routines follow
3024 i386_protection_init(void)
3028 kp = protection_codes;
3029 for (prot = 0; prot < 8; prot++) {
3031 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3033 * Read access is also 0. There isn't any execute bit,
3034 * so just make it readable.
3036 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3037 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3038 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3041 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3042 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3043 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3044 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3052 * Map a set of physical memory pages into the kernel virtual
3053 * address space. Return a pointer to where it is mapped. This
3054 * routine is intended to be used for mapping device memory,
3057 * NOTE: we can't use pgeflag unless we invalidate the pages one at
3061 pmap_mapdev(vm_paddr_t pa, vm_size_t size)
3063 vm_offset_t va, tmpva, offset;
3066 offset = pa & PAGE_MASK;
3067 size = roundup(offset + size, PAGE_SIZE);
3069 va = kmem_alloc_pageable(kernel_map, size);
3071 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3074 for (tmpva = va; size > 0;) {
3075 pte = (unsigned *)vtopte(tmpva);
3076 *pte = pa | PG_RW | PG_V; /* | pgeflag; */
3083 return ((void *)(va + offset));
3087 pmap_unmapdev(vm_offset_t va, vm_size_t size)
3089 vm_offset_t base, offset;
3091 base = va & PG_FRAME;
3092 offset = va & PAGE_MASK;
3093 size = roundup(offset + size, PAGE_SIZE);
3094 kmem_free(kernel_map, base, size);
3098 * perform the pmap work for mincore
3101 pmap_mincore(pmap_t pmap, vm_offset_t addr)
3103 unsigned *ptep, pte;
3107 ptep = pmap_pte(pmap, addr);
3112 if ((pte = *ptep) != 0) {
3115 val = MINCORE_INCORE;
3116 if ((pte & PG_MANAGED) == 0)
3119 pa = pte & PG_FRAME;
3121 m = PHYS_TO_VM_PAGE(pa);
3127 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3129 * Modified by someone
3131 else if (m->dirty || pmap_is_modified(m))
3132 val |= MINCORE_MODIFIED_OTHER;
3137 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3140 * Referenced by someone
3142 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3143 val |= MINCORE_REFERENCED_OTHER;
3144 vm_page_flag_set(m, PG_REFERENCED);
3151 pmap_activate(struct proc *p)
3155 pmap = vmspace_pmap(p->p_vmspace);
3157 atomic_set_int(&pmap->pm_active, 1 << mycpu->gd_cpuid);
3159 pmap->pm_active |= 1;
3161 #if defined(SWTCH_OPTIM_STATS)
3164 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3165 load_cr3(p->p_thread->td_pcb->pcb_cr3);
3169 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3172 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3176 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3181 #if defined(PMAP_DEBUG)
3183 pmap_pid_dump(int pid)
3189 FOREACH_PROC_IN_SYSTEM(p) {
3190 if (p->p_pid != pid)
3196 pmap = vmspace_pmap(p->p_vmspace);
3197 for(i=0;i<1024;i++) {
3200 unsigned base = i << PDRSHIFT;
3202 pde = &pmap->pm_pdir[i];
3203 if (pde && pmap_pde_v(pde)) {
3204 for(j=0;j<1024;j++) {
3205 unsigned va = base + (j << PAGE_SHIFT);
3206 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3213 pte = pmap_pte_quick( pmap, va);
3214 if (pte && pmap_pte_v(pte)) {
3218 m = PHYS_TO_VM_PAGE(pa);
3219 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3220 va, pa, m->hold_count, m->wire_count, m->flags);
3241 static void pads (pmap_t pm);
3242 void pmap_pvdump (vm_paddr_t pa);
3244 /* print address space of pmap*/
3251 if (pm == kernel_pmap)
3253 for (i = 0; i < 1024; i++)
3255 for (j = 0; j < 1024; j++) {
3256 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3257 if (pm == kernel_pmap && va < KERNBASE)
3259 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3261 ptep = pmap_pte_quick(pm, va);
3262 if (pmap_pte_v(ptep))
3263 printf("%x:%x ", va, *(int *) ptep);
3269 pmap_pvdump(vm_paddr_t pa)
3274 printf("pa %08llx", (long long)pa);
3275 m = PHYS_TO_VM_PAGE(pa);
3276 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3278 printf(" -> pmap %p, va %x, flags %x",
3279 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3281 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);