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 1.7 2003/06/20 02:09:50 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 #include <machine/segments.h>
107 #include <machine/tss.h>
108 #include <machine/globaldata.h>
109 #endif /* SMP || APIC_IO */
111 #define PMAP_KEEP_PDIRS
112 #ifndef PMAP_SHPGPERPROC
113 #define PMAP_SHPGPERPROC 200
116 #if defined(DIAGNOSTIC)
117 #define PMAP_DIAGNOSTIC
122 #if !defined(PMAP_DIAGNOSTIC)
123 #define PMAP_INLINE __inline
129 * Get PDEs and PTEs for user/kernel address space
131 #define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
132 #define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
134 #define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0)
135 #define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0)
136 #define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0)
137 #define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0)
138 #define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0)
140 #define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W))
141 #define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v)))
144 * Given a map and a machine independent protection code,
145 * convert to a vax protection code.
147 #define pte_prot(m, p) (protection_codes[p])
148 static int protection_codes[8];
150 static struct pmap kernel_pmap_store;
153 vm_offset_t avail_start; /* PA of first available physical page */
154 vm_offset_t avail_end; /* PA of last available physical page */
155 vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */
156 vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */
157 static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
158 static int pgeflag; /* PG_G or-in */
159 static int pseflag; /* PG_PS or-in */
161 static vm_object_t kptobj;
164 vm_offset_t kernel_vm_end;
167 * Data for the pv entry allocation mechanism
169 static vm_zone_t pvzone;
170 static struct vm_zone pvzone_store;
171 static struct vm_object pvzone_obj;
172 static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
173 static int pmap_pagedaemon_waken = 0;
174 static struct pv_entry *pvinit;
177 * All those kernel PT submaps that BSD is so fond of
179 pt_entry_t *CMAP1 = 0;
180 static pt_entry_t *CMAP2, *ptmmap;
181 caddr_t CADDR1 = 0, ptvmmap = 0;
182 static caddr_t CADDR2;
183 static pt_entry_t *msgbufmap;
184 struct msgbuf *msgbufp=0;
189 static pt_entry_t *pt_crashdumpmap;
190 static caddr_t crashdumpmap;
193 extern pt_entry_t *SMPpt;
195 static pt_entry_t *PMAP1 = 0;
196 static unsigned *PADDR1 = 0;
199 static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv));
200 static unsigned * get_ptbase __P((pmap_t pmap));
201 static pv_entry_t get_pv_entry __P((void));
202 static void i386_protection_init __P((void));
203 static __inline void pmap_changebit __P((vm_page_t m, int bit, boolean_t setem));
205 static void pmap_remove_all __P((vm_page_t m));
206 static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va,
207 vm_page_t m, vm_page_t mpte));
208 static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq,
210 static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va));
211 static int pmap_remove_entry __P((struct pmap *pmap, vm_page_t m,
213 static boolean_t pmap_testbit __P((vm_page_t m, int bit));
214 static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va,
215 vm_page_t mpte, vm_page_t m));
217 static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va));
219 static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p));
220 static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex));
221 static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va));
222 static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex));
223 static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t));
224 static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
226 static unsigned pdir4mb;
231 * Extract the page table entry associated
232 * with the given map/virtual_address pair.
235 PMAP_INLINE unsigned *
237 register pmap_t pmap;
243 pdeaddr = (unsigned *) pmap_pde(pmap, va);
244 if (*pdeaddr & PG_PS)
247 return get_ptbase(pmap) + i386_btop(va);
254 * Move the kernel virtual free pointer to the next
255 * 4MB. This is used to help improve performance
256 * by using a large (4MB) page for much of the kernel
257 * (.text, .data, .bss)
260 pmap_kmem_choose(vm_offset_t addr)
262 vm_offset_t newaddr = addr;
264 if (cpu_feature & CPUID_PSE) {
265 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
272 * Bootstrap the system enough to run with virtual memory.
274 * On the i386 this is called after mapping has already been enabled
275 * and just syncs the pmap module with what has already been done.
276 * [We can't call it easily with mapping off since the kernel is not
277 * mapped with PA == VA, hence we would have to relocate every address
278 * from the linked base (virtual) address "KERNBASE" to the actual
279 * (physical) address starting relative to 0]
282 pmap_bootstrap(firstaddr, loadaddr)
283 vm_offset_t firstaddr;
284 vm_offset_t loadaddr;
289 struct globaldata *gd;
293 avail_start = firstaddr;
296 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too
297 * large. It should instead be correctly calculated in locore.s and
298 * not based on 'first' (which is a physical address, not a virtual
299 * address, for the start of unused physical memory). The kernel
300 * page tables are NOT double mapped and thus should not be included
301 * in this calculation.
303 virtual_avail = (vm_offset_t) KERNBASE + firstaddr;
304 virtual_avail = pmap_kmem_choose(virtual_avail);
306 virtual_end = VM_MAX_KERNEL_ADDRESS;
309 * Initialize protection array.
311 i386_protection_init();
314 * The kernel's pmap is statically allocated so we don't have to use
315 * pmap_create, which is unlikely to work correctly at this part of
316 * the boot sequence (XXX and which no longer exists).
318 kernel_pmap = &kernel_pmap_store;
320 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD);
321 kernel_pmap->pm_count = 1;
322 kernel_pmap->pm_active = -1; /* don't allow deactivation */
323 TAILQ_INIT(&kernel_pmap->pm_pvlist);
327 * Reserve some special page table entries/VA space for temporary
330 #define SYSMAP(c, p, v, n) \
331 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
334 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va);
337 * CMAP1/CMAP2 are used for zeroing and copying pages.
339 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
340 SYSMAP(caddr_t, CMAP2, CADDR2, 1)
345 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
348 * ptvmmap is used for reading arbitrary physical pages via /dev/mem.
349 * XXX ptmmap is not used.
351 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
354 * msgbufp is used to map the system message buffer.
355 * XXX msgbufmap is not used.
357 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
358 atop(round_page(MSGBUF_SIZE)))
362 * ptemap is used for pmap_pte_quick
364 SYSMAP(unsigned *, PMAP1, PADDR1, 1);
369 *(int *) CMAP1 = *(int *) CMAP2 = 0;
370 for (i = 0; i < NKPT; i++)
374 #if !defined(SMP) /* XXX - see also mp_machdep.c */
375 if (cpu_feature & CPUID_PGE) {
381 * Initialize the 4MB page size flag
385 * The 4MB page version of the initial
386 * kernel page mapping.
390 #if !defined(DISABLE_PSE)
391 if (cpu_feature & CPUID_PSE) {
394 * Note that we have enabled PSE mode
397 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
398 ptditmp &= ~(NBPDR - 1);
399 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
404 * Enable the PSE mode.
406 load_cr4(rcr4() | CR4_PSE);
409 * We can do the mapping here for the single processor
410 * case. We simply ignore the old page table page from
414 * For SMP, we still need 4K pages to bootstrap APs,
415 * PSE will be enabled as soon as all APs are up.
417 PTD[KPTDI] = (pd_entry_t) ptditmp;
418 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp;
424 if (cpu_apic_address == 0)
425 panic("pmap_bootstrap: no local apic!");
427 /* local apic is mapped on last page */
428 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
429 (cpu_apic_address & PG_FRAME));
431 /* BSP does this itself, AP's get it pre-set */
432 gd = &CPU_prvspace[0].globaldata;
433 gd->gd_prv_CMAP1 = &SMPpt[1];
434 gd->gd_prv_CMAP2 = &SMPpt[2];
435 gd->gd_prv_CMAP3 = &SMPpt[3];
436 gd->gd_prv_PMAP1 = &SMPpt[4];
437 gd->gd_prv_CADDR1 = CPU_prvspace[0].CPAGE1;
438 gd->gd_prv_CADDR2 = CPU_prvspace[0].CPAGE2;
439 gd->gd_prv_CADDR3 = CPU_prvspace[0].CPAGE3;
440 gd->gd_prv_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
448 * Set 4mb pdir for mp startup
453 if (pseflag && (cpu_feature & CPUID_PSE)) {
454 load_cr4(rcr4() | CR4_PSE);
455 if (pdir4mb && cpuid == 0) { /* only on BSP */
456 kernel_pmap->pm_pdir[KPTDI] =
457 PTD[KPTDI] = (pd_entry_t)pdir4mb;
465 * Initialize the pmap module.
466 * Called by vm_init, to initialize any structures that the pmap
467 * system needs to map virtual memory.
468 * pmap_init has been enhanced to support in a fairly consistant
469 * way, discontiguous physical memory.
472 pmap_init(phys_start, phys_end)
473 vm_offset_t phys_start, phys_end;
479 * object for kernel page table pages
481 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
484 * Allocate memory for random pmap data structures. Includes the
488 for(i = 0; i < vm_page_array_size; i++) {
491 m = &vm_page_array[i];
492 TAILQ_INIT(&m->md.pv_list);
493 m->md.pv_list_count = 0;
497 * init the pv free list
499 initial_pvs = vm_page_array_size;
500 if (initial_pvs < MINPV)
502 pvzone = &pvzone_store;
503 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
504 initial_pvs * sizeof (struct pv_entry));
505 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
509 * Now it is safe to enable pv_table recording.
511 pmap_initialized = TRUE;
515 * Initialize the address space (zone) for the pv_entries. Set a
516 * high water mark so that the system can recover from excessive
517 * numbers of pv entries.
522 int shpgperproc = PMAP_SHPGPERPROC;
524 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
525 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
526 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
527 pv_entry_high_water = 9 * (pv_entry_max / 10);
528 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
532 /***************************************************
533 * Low level helper routines.....
534 ***************************************************/
536 #if defined(PMAP_DIAGNOSTIC)
539 * This code checks for non-writeable/modified pages.
540 * This should be an invalid condition.
543 pmap_nw_modified(pt_entry_t ptea)
549 if ((pte & (PG_M|PG_RW)) == PG_M)
558 * this routine defines the region(s) of memory that should
559 * not be tested for the modified bit.
561 static PMAP_INLINE int
562 pmap_track_modified(vm_offset_t va)
564 if ((va < clean_sva) || (va >= clean_eva))
570 static PMAP_INLINE void
571 invltlb_1pg(vm_offset_t va)
573 #if defined(I386_CPU)
574 if (cpu_class == CPUCLASS_386) {
584 pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
587 if (pmap->pm_active & (1 << cpuid))
588 cpu_invlpg((void *)va);
589 if (pmap->pm_active & other_cpus)
598 pmap_TLB_invalidate_all(pmap_t pmap)
601 if (pmap->pm_active & (1 << cpuid))
603 if (pmap->pm_active & other_cpus)
615 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
617 /* are we current address space or kernel? */
618 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
619 return (unsigned *) PTmap;
621 /* otherwise, we are alternate address space */
622 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
623 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
625 /* The page directory is not shared between CPUs */
631 return (unsigned *) APTmap;
635 * Super fast pmap_pte routine best used when scanning
636 * the pv lists. This eliminates many coarse-grained
637 * invltlb calls. Note that many of the pv list
638 * scans are across different pmaps. It is very wasteful
639 * to do an entire invltlb for checking a single mapping.
643 pmap_pte_quick(pmap, va)
644 register pmap_t pmap;
648 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
649 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
650 unsigned index = i386_btop(va);
651 /* are we current address space or kernel? */
652 if ((pmap == kernel_pmap) ||
653 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
654 return (unsigned *) PTmap + index;
656 newpf = pde & PG_FRAME;
658 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) {
659 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V;
660 cpu_invlpg(prv_PADDR1);
662 return prv_PADDR1 + ((unsigned) index & (NPTEPG - 1));
664 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) {
665 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V;
666 invltlb_1pg((vm_offset_t) PADDR1);
668 return PADDR1 + ((unsigned) index & (NPTEPG - 1));
675 * Routine: pmap_extract
677 * Extract the physical page address associated
678 * with the given map/virtual_address pair.
681 pmap_extract(pmap, va)
682 register pmap_t pmap;
686 vm_offset_t pdirindex;
687 pdirindex = va >> PDRSHIFT;
688 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
690 if ((rtval & PG_PS) != 0) {
691 rtval &= ~(NBPDR - 1);
692 rtval |= va & (NBPDR - 1);
695 pte = get_ptbase(pmap) + i386_btop(va);
696 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
703 /***************************************************
704 * Low level mapping routines.....
705 ***************************************************/
708 * add a wired page to the kva
709 * note that in order for the mapping to take effect -- you
710 * should do a invltlb after doing the pmap_kenter...
715 register vm_offset_t pa;
717 register unsigned *pte;
720 npte = pa | PG_RW | PG_V | pgeflag;
721 pte = (unsigned *)vtopte(va);
728 * remove a page from the kernel pagetables
734 register unsigned *pte;
736 pte = (unsigned *)vtopte(va);
742 * Used to map a range of physical addresses into kernel
743 * virtual address space.
745 * For now, VM is already on, we only need to map the
749 pmap_map(virt, start, end, prot)
755 while (start < end) {
756 pmap_kenter(virt, start);
765 * Add a list of wired pages to the kva
766 * this routine is only used for temporary
767 * kernel mappings that do not need to have
768 * page modification or references recorded.
769 * Note that old mappings are simply written
770 * over. The page *must* be wired.
773 pmap_qenter(va, m, count)
780 end_va = va + count * PAGE_SIZE;
782 while (va < end_va) {
785 pte = (unsigned *)vtopte(va);
786 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
788 cpu_invlpg((void *)va);
801 * this routine jerks page mappings from the
802 * kernel -- it is meant only for temporary mappings.
805 pmap_qremove(va, count)
811 end_va = va + count*PAGE_SIZE;
813 while (va < end_va) {
816 pte = (unsigned *)vtopte(va);
819 cpu_invlpg((void *)va);
831 pmap_page_lookup(object, pindex)
837 m = vm_page_lookup(object, pindex);
838 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
844 * Create a new thread and optionally associate it with a (new) process.
852 if (mycpu->gd_tdfreecount > 0) {
853 --mycpu->gd_tdfreecount;
854 td = TAILQ_FIRST(&mycpu->gd_tdfreeq);
855 KASSERT(td != NULL, ("unexpected null cache td"));
856 TAILQ_REMOVE(&mycpu->gd_tdfreeq, td, td_threadq);
858 td = zalloc(thread_zone);
860 (void *)kmem_alloc(kernel_map, UPAGES * PAGE_SIZE);
862 td->td_pcb = (struct pcb *)(td->td_kstack + UPAGES * PAGE_SIZE) - 1;
867 * Dispose of a thread, unlink it from its related proc (if any). Keep
868 * CACHE_NTHREAD threads around for fast-startup.
871 pmap_dispose_thread(struct thread *td)
874 if (mycpu->gd_tdfreecount < CACHE_NTHREADS) {
875 ++mycpu->gd_tdfreecount;
876 TAILQ_INSERT_HEAD(&mycpu->gd_tdfreeq, td, td_threadq);
879 kmem_free(kernel_map,
880 (vm_offset_t)td->td_kstack, UPAGES * PAGE_SIZE);
881 td->td_kstack = NULL;
883 zfree(thread_zone, td);
888 * Create the UPAGES for a new process.
889 * This routine directly affects the fork perf for a process.
892 pmap_new_proc(struct proc *p, struct thread *td)
894 p->p_addr = (void *)td->td_kstack;
897 td->td_switch = cpu_heavy_switch;
898 bzero(p->p_addr, sizeof(*p->p_addr));
905 unsigned *ptek, oldpte;
908 * allocate object for the upages
910 if ((upobj = p->p_upages_obj) == NULL) {
911 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES);
912 p->p_upages_obj = upobj;
915 /* get a kernel virtual address for the UPAGES for this proc */
916 if ((up = p->p_addr) == NULL) {
917 up = (struct user *) kmem_alloc_nofault(kernel_map,
920 panic("pmap_new_proc: u_map allocation failed");
924 ptek = (unsigned *) vtopte((vm_offset_t) up);
927 for(i=0;i<UPAGES;i++) {
929 * Get a kernel stack page
931 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
939 oldpte = *(ptek + i);
941 * Enter the page into the kernel address space.
943 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag;
945 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) {
946 invlpg((vm_offset_t) up + i * PAGE_SIZE);
953 vm_page_flag_clear(m, PG_ZERO);
954 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
955 m->valid = VM_PAGE_BITS_ALL;
963 * Dispose the UPAGES for a process that has exited.
964 * This routine directly impacts the exit perf of a process.
967 pmap_dispose_proc(struct proc *p)
971 if ((td = p->p_thread) != NULL) {
981 unsigned *ptek, oldpte;
983 upobj = p->p_upages_obj;
985 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr);
986 for(i=0;i<UPAGES;i++) {
988 if ((m = vm_page_lookup(upobj, i)) == NULL)
989 panic("pmap_dispose_proc: upage already missing???");
993 oldpte = *(ptek + i);
995 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386))
996 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE);
997 vm_page_unwire(m, 0);
1000 #if defined(I386_CPU)
1001 if (cpu_class <= CPUCLASS_386)
1006 * If the process got swapped out some of its UPAGES might have gotten
1007 * swapped. Just get rid of the object to clean up the swap use
1008 * proactively. NOTE! might block waiting for paging I/O to complete.
1010 if (upobj->type == OBJT_SWAP) {
1011 p->p_upages_obj = NULL;
1012 vm_object_deallocate(upobj);
1018 * Allow the UPAGES for a process to be prejudicially paged out.
1021 pmap_swapout_proc(p)
1029 upobj = p->p_upages_obj;
1031 * let the upages be paged
1033 for(i=0;i<UPAGES;i++) {
1034 if ((m = vm_page_lookup(upobj, i)) == NULL)
1035 panic("pmap_swapout_proc: upage already missing???");
1037 vm_page_unwire(m, 0);
1038 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
1044 * Bring the UPAGES for a specified process back in.
1055 upobj = p->p_upages_obj;
1056 for(i=0;i<UPAGES;i++) {
1058 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1060 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
1061 VM_PAGE_TO_PHYS(m));
1063 if (m->valid != VM_PAGE_BITS_ALL) {
1064 rv = vm_pager_get_pages(upobj, &m, 1, 0);
1065 if (rv != VM_PAGER_OK)
1066 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
1067 m = vm_page_lookup(upobj, i);
1068 m->valid = VM_PAGE_BITS_ALL;
1073 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
1078 /***************************************************
1079 * Page table page management routines.....
1080 ***************************************************/
1083 * This routine unholds page table pages, and if the hold count
1084 * drops to zero, then it decrements the wire count.
1087 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
1089 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
1092 if (m->hold_count == 0) {
1095 * unmap the page table page
1097 pmap->pm_pdir[m->pindex] = 0;
1098 --pmap->pm_stats.resident_count;
1099 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1100 (((unsigned) PTDpde) & PG_FRAME)) {
1102 * Do a invltlb to make the invalidated mapping
1103 * take effect immediately.
1105 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
1106 pmap_TLB_invalidate(pmap, pteva);
1109 if (pmap->pm_ptphint == m)
1110 pmap->pm_ptphint = NULL;
1113 * If the page is finally unwired, simply free it.
1116 if (m->wire_count == 0) {
1120 vm_page_free_zero(m);
1128 static PMAP_INLINE int
1129 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
1132 if (m->hold_count == 0)
1133 return _pmap_unwire_pte_hold(pmap, m);
1139 * After removing a page table entry, this routine is used to
1140 * conditionally free the page, and manage the hold/wire counts.
1143 pmap_unuse_pt(pmap, va, mpte)
1149 if (va >= UPT_MIN_ADDRESS)
1153 ptepindex = (va >> PDRSHIFT);
1154 if (pmap->pm_ptphint &&
1155 (pmap->pm_ptphint->pindex == ptepindex)) {
1156 mpte = pmap->pm_ptphint;
1158 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1159 pmap->pm_ptphint = mpte;
1163 return pmap_unwire_pte_hold(pmap, mpte);
1171 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1172 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1174 pmap->pm_active = 0;
1175 pmap->pm_ptphint = NULL;
1176 TAILQ_INIT(&pmap->pm_pvlist);
1177 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1181 * Initialize a preallocated and zeroed pmap structure,
1182 * such as one in a vmspace structure.
1186 register struct pmap *pmap;
1191 * No need to allocate page table space yet but we do need a valid
1192 * page directory table.
1194 if (pmap->pm_pdir == NULL)
1196 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1199 * allocate object for the ptes
1201 if (pmap->pm_pteobj == NULL)
1202 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1205 * allocate the page directory page
1207 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1208 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1210 ptdpg->wire_count = 1;
1214 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1215 ptdpg->valid = VM_PAGE_BITS_ALL;
1217 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1218 if ((ptdpg->flags & PG_ZERO) == 0)
1219 bzero(pmap->pm_pdir, PAGE_SIZE);
1222 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1225 /* install self-referential address mapping entry */
1226 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1227 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1230 pmap->pm_active = 0;
1231 pmap->pm_ptphint = NULL;
1232 TAILQ_INIT(&pmap->pm_pvlist);
1233 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1237 * Wire in kernel global address entries. To avoid a race condition
1238 * between pmap initialization and pmap_growkernel, this procedure
1239 * should be called after the vmspace is attached to the process
1240 * but before this pmap is activated.
1246 /* XXX copies current process, does not fill in MPPTDI */
1247 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1251 pmap_release_free_page(pmap, p)
1255 unsigned *pde = (unsigned *) pmap->pm_pdir;
1257 * This code optimizes the case of freeing non-busy
1258 * page-table pages. Those pages are zero now, and
1259 * might as well be placed directly into the zero queue.
1261 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1267 * Remove the page table page from the processes address space.
1270 pmap->pm_stats.resident_count--;
1272 if (p->hold_count) {
1273 panic("pmap_release: freeing held page table page");
1276 * Page directory pages need to have the kernel
1277 * stuff cleared, so they can go into the zero queue also.
1279 if (p->pindex == PTDPTDI) {
1280 bzero(pde + KPTDI, nkpt * PTESIZE);
1285 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1288 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1289 pmap->pm_ptphint = NULL;
1293 vm_page_free_zero(p);
1298 * this routine is called if the page table page is not
1302 _pmap_allocpte(pmap, ptepindex)
1306 vm_offset_t pteva, ptepa;
1310 * Find or fabricate a new pagetable page
1312 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1313 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1315 KASSERT(m->queue == PQ_NONE,
1316 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1318 if (m->wire_count == 0)
1323 * Increment the hold count for the page table page
1324 * (denoting a new mapping.)
1329 * Map the pagetable page into the process address space, if
1330 * it isn't already there.
1333 pmap->pm_stats.resident_count++;
1335 ptepa = VM_PAGE_TO_PHYS(m);
1336 pmap->pm_pdir[ptepindex] =
1337 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1340 * Set the page table hint
1342 pmap->pm_ptphint = m;
1345 * Try to use the new mapping, but if we cannot, then
1346 * do it with the routine that maps the page explicitly.
1348 if ((m->flags & PG_ZERO) == 0) {
1349 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1350 (((unsigned) PTDpde) & PG_FRAME)) {
1351 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1352 bzero((caddr_t) pteva, PAGE_SIZE);
1354 pmap_zero_page(ptepa);
1358 m->valid = VM_PAGE_BITS_ALL;
1359 vm_page_flag_clear(m, PG_ZERO);
1360 vm_page_flag_set(m, PG_MAPPED);
1367 pmap_allocpte(pmap, va)
1376 * Calculate pagetable page index
1378 ptepindex = va >> PDRSHIFT;
1381 * Get the page directory entry
1383 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1386 * This supports switching from a 4MB page to a
1389 if (ptepa & PG_PS) {
1390 pmap->pm_pdir[ptepindex] = 0;
1396 * If the page table page is mapped, we just increment the
1397 * hold count, and activate it.
1401 * In order to get the page table page, try the
1404 if (pmap->pm_ptphint &&
1405 (pmap->pm_ptphint->pindex == ptepindex)) {
1406 m = pmap->pm_ptphint;
1408 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1409 pmap->pm_ptphint = m;
1415 * Here if the pte page isn't mapped, or if it has been deallocated.
1417 return _pmap_allocpte(pmap, ptepindex);
1421 /***************************************************
1422 * Pmap allocation/deallocation routines.
1423 ***************************************************/
1426 * Release any resources held by the given physical map.
1427 * Called when a pmap initialized by pmap_pinit is being released.
1428 * Should only be called if the map contains no valid mappings.
1432 register struct pmap *pmap;
1434 vm_page_t p,n,ptdpg;
1435 vm_object_t object = pmap->pm_pteobj;
1438 #if defined(DIAGNOSTIC)
1439 if (object->ref_count != 1)
1440 panic("pmap_release: pteobj reference count != 1");
1445 curgeneration = object->generation;
1446 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1447 n = TAILQ_NEXT(p, listq);
1448 if (p->pindex == PTDPTDI) {
1453 if (!pmap_release_free_page(pmap, p) &&
1454 (object->generation != curgeneration))
1459 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1464 kvm_size(SYSCTL_HANDLER_ARGS)
1466 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1468 return sysctl_handle_long(oidp, &ksize, 0, req);
1470 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1471 0, 0, kvm_size, "IU", "Size of KVM");
1474 kvm_free(SYSCTL_HANDLER_ARGS)
1476 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1478 return sysctl_handle_long(oidp, &kfree, 0, req);
1480 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1481 0, 0, kvm_free, "IU", "Amount of KVM free");
1484 * grow the number of kernel page table entries, if needed
1487 pmap_growkernel(vm_offset_t addr)
1492 vm_offset_t ptppaddr;
1497 if (kernel_vm_end == 0) {
1498 kernel_vm_end = KERNBASE;
1500 while (pdir_pde(PTD, kernel_vm_end)) {
1501 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1505 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1506 while (kernel_vm_end < addr) {
1507 if (pdir_pde(PTD, kernel_vm_end)) {
1508 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1513 * This index is bogus, but out of the way
1515 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1517 panic("pmap_growkernel: no memory to grow kernel");
1522 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1523 pmap_zero_page(ptppaddr);
1524 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1525 pdir_pde(PTD, kernel_vm_end) = newpdir;
1527 LIST_FOREACH(p, &allproc, p_list) {
1529 pmap = vmspace_pmap(p->p_vmspace);
1530 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1533 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1534 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1540 * Retire the given physical map from service.
1541 * Should only be called if the map contains
1542 * no valid mappings.
1546 register pmap_t pmap;
1553 count = --pmap->pm_count;
1556 panic("destroying a pmap is not yet implemented");
1561 * Add a reference to the specified pmap.
1564 pmap_reference(pmap)
1572 /***************************************************
1573 * page management routines.
1574 ***************************************************/
1577 * free the pv_entry back to the free list
1579 static PMAP_INLINE void
1588 * get a new pv_entry, allocating a block from the system
1590 * the memory allocation is performed bypassing the malloc code
1591 * because of the possibility of allocations at interrupt time.
1597 if (pv_entry_high_water &&
1598 (pv_entry_count > pv_entry_high_water) &&
1599 (pmap_pagedaemon_waken == 0)) {
1600 pmap_pagedaemon_waken = 1;
1601 wakeup (&vm_pages_needed);
1603 return zalloci(pvzone);
1607 * This routine is very drastic, but can save the system
1615 static int warningdone=0;
1617 if (pmap_pagedaemon_waken == 0)
1620 if (warningdone < 5) {
1621 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1625 for(i = 0; i < vm_page_array_size; i++) {
1626 m = &vm_page_array[i];
1627 if (m->wire_count || m->hold_count || m->busy ||
1628 (m->flags & PG_BUSY))
1632 pmap_pagedaemon_waken = 0;
1637 * If it is the first entry on the list, it is actually
1638 * in the header and we must copy the following entry up
1639 * to the header. Otherwise we must search the list for
1640 * the entry. In either case we free the now unused entry.
1644 pmap_remove_entry(pmap, m, va)
1654 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1655 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1656 if (pmap == pv->pv_pmap && va == pv->pv_va)
1660 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1661 if (va == pv->pv_va)
1669 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1670 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1671 m->md.pv_list_count--;
1672 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1673 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1675 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1684 * Create a pv entry for page at pa for
1688 pmap_insert_entry(pmap, va, mpte, m)
1699 pv = get_pv_entry();
1704 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1705 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1706 m->md.pv_list_count++;
1712 * pmap_remove_pte: do the things to unmap a page in a process
1715 pmap_remove_pte(pmap, ptq, va)
1723 oldpte = loadandclear(ptq);
1725 pmap->pm_stats.wired_count -= 1;
1727 * Machines that don't support invlpg, also don't support
1732 pmap->pm_stats.resident_count -= 1;
1733 if (oldpte & PG_MANAGED) {
1734 m = PHYS_TO_VM_PAGE(oldpte);
1735 if (oldpte & PG_M) {
1736 #if defined(PMAP_DIAGNOSTIC)
1737 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1739 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1743 if (pmap_track_modified(va))
1747 vm_page_flag_set(m, PG_REFERENCED);
1748 return pmap_remove_entry(pmap, m, va);
1750 return pmap_unuse_pt(pmap, va, NULL);
1757 * Remove a single page from a process address space
1760 pmap_remove_page(pmap, va)
1762 register vm_offset_t va;
1764 register unsigned *ptq;
1767 * if there is no pte for this address, just skip it!!!
1769 if (*pmap_pde(pmap, va) == 0) {
1774 * get a local va for mappings for this pmap.
1776 ptq = get_ptbase(pmap) + i386_btop(va);
1778 (void) pmap_remove_pte(pmap, ptq, va);
1779 pmap_TLB_invalidate(pmap, va);
1785 * Remove the given range of addresses from the specified map.
1787 * It is assumed that the start and end are properly
1788 * rounded to the page size.
1791 pmap_remove(pmap, sva, eva)
1793 register vm_offset_t sva;
1794 register vm_offset_t eva;
1796 register unsigned *ptbase;
1798 vm_offset_t ptpaddr;
1799 vm_offset_t sindex, eindex;
1805 if (pmap->pm_stats.resident_count == 0)
1809 * special handling of removing one page. a very
1810 * common operation and easy to short circuit some
1813 if (((sva + PAGE_SIZE) == eva) &&
1814 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1815 pmap_remove_page(pmap, sva);
1822 * Get a local virtual address for the mappings that are being
1825 ptbase = get_ptbase(pmap);
1827 sindex = i386_btop(sva);
1828 eindex = i386_btop(eva);
1830 for (; sindex < eindex; sindex = pdnxt) {
1834 * Calculate index for next page table.
1836 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1837 if (pmap->pm_stats.resident_count == 0)
1840 pdirindex = sindex / NPDEPG;
1841 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1842 pmap->pm_pdir[pdirindex] = 0;
1843 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1849 * Weed out invalid mappings. Note: we assume that the page
1850 * directory table is always allocated, and in kernel virtual.
1856 * Limit our scan to either the end of the va represented
1857 * by the current page table page, or to the end of the
1858 * range being removed.
1860 if (pdnxt > eindex) {
1864 for ( ;sindex != pdnxt; sindex++) {
1866 if (ptbase[sindex] == 0) {
1869 va = i386_ptob(sindex);
1872 if (pmap_remove_pte(pmap,
1873 ptbase + sindex, va))
1879 pmap_TLB_invalidate_all(pmap);
1883 * Routine: pmap_remove_all
1885 * Removes this physical page from
1886 * all physical maps in which it resides.
1887 * Reflects back modify bits to the pager.
1890 * Original versions of this routine were very
1891 * inefficient because they iteratively called
1892 * pmap_remove (slow...)
1899 register pv_entry_t pv;
1900 register unsigned *pte, tpte;
1903 #if defined(PMAP_DIAGNOSTIC)
1905 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1908 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1909 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1914 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1915 pv->pv_pmap->pm_stats.resident_count--;
1917 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1919 tpte = loadandclear(pte);
1921 pv->pv_pmap->pm_stats.wired_count--;
1924 vm_page_flag_set(m, PG_REFERENCED);
1927 * Update the vm_page_t clean and reference bits.
1930 #if defined(PMAP_DIAGNOSTIC)
1931 if (pmap_nw_modified((pt_entry_t) tpte)) {
1933 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1937 if (pmap_track_modified(pv->pv_va))
1940 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1942 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1943 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1944 m->md.pv_list_count--;
1945 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1949 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1955 * Set the physical protection on the
1956 * specified range of this map as requested.
1959 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1961 register unsigned *ptbase;
1962 vm_offset_t pdnxt, ptpaddr;
1963 vm_pindex_t sindex, eindex;
1969 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1970 pmap_remove(pmap, sva, eva);
1974 if (prot & VM_PROT_WRITE)
1979 ptbase = get_ptbase(pmap);
1981 sindex = i386_btop(sva);
1982 eindex = i386_btop(eva);
1984 for (; sindex < eindex; sindex = pdnxt) {
1988 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1990 pdirindex = sindex / NPDEPG;
1991 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1992 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1993 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1999 * Weed out invalid mappings. Note: we assume that the page
2000 * directory table is always allocated, and in kernel virtual.
2005 if (pdnxt > eindex) {
2009 for (; sindex != pdnxt; sindex++) {
2014 pbits = ptbase[sindex];
2016 if (pbits & PG_MANAGED) {
2019 m = PHYS_TO_VM_PAGE(pbits);
2020 vm_page_flag_set(m, PG_REFERENCED);
2024 if (pmap_track_modified(i386_ptob(sindex))) {
2026 m = PHYS_TO_VM_PAGE(pbits);
2035 if (pbits != ptbase[sindex]) {
2036 ptbase[sindex] = pbits;
2042 pmap_TLB_invalidate_all(pmap);
2046 * Insert the given physical page (p) at
2047 * the specified virtual address (v) in the
2048 * target physical map with the protection requested.
2050 * If specified, the page will be wired down, meaning
2051 * that the related pte can not be reclaimed.
2053 * NB: This is the only routine which MAY NOT lazy-evaluate
2054 * or lose information. That is, this routine must actually
2055 * insert this page into the given map NOW.
2058 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
2062 register unsigned *pte;
2064 vm_offset_t origpte, newpte;
2071 #ifdef PMAP_DIAGNOSTIC
2072 if (va > VM_MAX_KERNEL_ADDRESS)
2073 panic("pmap_enter: toobig");
2074 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
2075 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
2080 * In the case that a page table page is not
2081 * resident, we are creating it here.
2083 if (va < UPT_MIN_ADDRESS) {
2084 mpte = pmap_allocpte(pmap, va);
2086 #if 0 && defined(PMAP_DIAGNOSTIC)
2088 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
2089 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
2090 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
2091 pmap->pm_pdir[PTDPTDI], origpte, va);
2094 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
2095 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
2096 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
2097 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
2098 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
2099 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
2100 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
2106 pte = pmap_pte(pmap, va);
2109 * Page Directory table entry not valid, we need a new PT page
2112 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
2113 (void *)pmap->pm_pdir[PTDPTDI], va);
2116 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2117 origpte = *(vm_offset_t *)pte;
2118 opa = origpte & PG_FRAME;
2120 if (origpte & PG_PS)
2121 panic("pmap_enter: attempted pmap_enter on 4MB page");
2124 * Mapping has not changed, must be protection or wiring change.
2126 if (origpte && (opa == pa)) {
2128 * Wiring change, just update stats. We don't worry about
2129 * wiring PT pages as they remain resident as long as there
2130 * are valid mappings in them. Hence, if a user page is wired,
2131 * the PT page will be also.
2133 if (wired && ((origpte & PG_W) == 0))
2134 pmap->pm_stats.wired_count++;
2135 else if (!wired && (origpte & PG_W))
2136 pmap->pm_stats.wired_count--;
2138 #if defined(PMAP_DIAGNOSTIC)
2139 if (pmap_nw_modified((pt_entry_t) origpte)) {
2141 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
2147 * Remove extra pte reference
2152 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
2153 if ((origpte & PG_RW) == 0) {
2156 cpu_invlpg((void *)va);
2157 if (pmap->pm_active & other_cpus)
2167 * We might be turning off write access to the page,
2168 * so we go ahead and sense modify status.
2170 if (origpte & PG_MANAGED) {
2171 if ((origpte & PG_M) && pmap_track_modified(va)) {
2173 om = PHYS_TO_VM_PAGE(opa);
2181 * Mapping has changed, invalidate old range and fall through to
2182 * handle validating new mapping.
2186 err = pmap_remove_pte(pmap, pte, va);
2188 panic("pmap_enter: pte vanished, va: 0x%x", va);
2192 * Enter on the PV list if part of our managed memory. Note that we
2193 * raise IPL while manipulating pv_table since pmap_enter can be
2194 * called at interrupt time.
2196 if (pmap_initialized &&
2197 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2198 pmap_insert_entry(pmap, va, mpte, m);
2203 * Increment counters
2205 pmap->pm_stats.resident_count++;
2207 pmap->pm_stats.wired_count++;
2211 * Now validate mapping with desired protection/wiring.
2213 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2217 if (va < UPT_MIN_ADDRESS)
2219 if (pmap == kernel_pmap)
2223 * if the mapping or permission bits are different, we need
2224 * to update the pte.
2226 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2227 *pte = newpte | PG_A;
2230 cpu_invlpg((void *)va);
2231 if (pmap->pm_active & other_cpus)
2241 * this code makes some *MAJOR* assumptions:
2242 * 1. Current pmap & pmap exists.
2245 * 4. No page table pages.
2246 * 5. Tlbflush is deferred to calling procedure.
2247 * 6. Page IS managed.
2248 * but is *MUCH* faster than pmap_enter...
2252 pmap_enter_quick(pmap, va, m, mpte)
2253 register pmap_t pmap;
2262 * In the case that a page table page is not
2263 * resident, we are creating it here.
2265 if (va < UPT_MIN_ADDRESS) {
2270 * Calculate pagetable page index
2272 ptepindex = va >> PDRSHIFT;
2273 if (mpte && (mpte->pindex == ptepindex)) {
2278 * Get the page directory entry
2280 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2283 * If the page table page is mapped, we just increment
2284 * the hold count, and activate it.
2288 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2289 if (pmap->pm_ptphint &&
2290 (pmap->pm_ptphint->pindex == ptepindex)) {
2291 mpte = pmap->pm_ptphint;
2293 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2294 pmap->pm_ptphint = mpte;
2300 mpte = _pmap_allocpte(pmap, ptepindex);
2308 * This call to vtopte makes the assumption that we are
2309 * entering the page into the current pmap. In order to support
2310 * quick entry into any pmap, one would likely use pmap_pte_quick.
2311 * But that isn't as quick as vtopte.
2313 pte = (unsigned *)vtopte(va);
2316 pmap_unwire_pte_hold(pmap, mpte);
2321 * Enter on the PV list if part of our managed memory. Note that we
2322 * raise IPL while manipulating pv_table since pmap_enter can be
2323 * called at interrupt time.
2325 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2326 pmap_insert_entry(pmap, va, mpte, m);
2329 * Increment counters
2331 pmap->pm_stats.resident_count++;
2333 pa = VM_PAGE_TO_PHYS(m);
2336 * Now validate mapping with RO protection
2338 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2339 *pte = pa | PG_V | PG_U;
2341 *pte = pa | PG_V | PG_U | PG_MANAGED;
2347 * Make a temporary mapping for a physical address. This is only intended
2348 * to be used for panic dumps.
2351 pmap_kenter_temporary(vm_offset_t pa, int i)
2353 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2354 return ((void *)crashdumpmap);
2357 #define MAX_INIT_PT (96)
2359 * pmap_object_init_pt preloads the ptes for a given object
2360 * into the specified pmap. This eliminates the blast of soft
2361 * faults on process startup and immediately after an mmap.
2364 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2377 if (pmap == NULL || object == NULL)
2381 * This code maps large physical mmap regions into the
2382 * processor address space. Note that some shortcuts
2383 * are taken, but the code works.
2386 (object->type == OBJT_DEVICE) &&
2387 ((addr & (NBPDR - 1)) == 0) &&
2388 ((size & (NBPDR - 1)) == 0) ) {
2391 unsigned int ptepindex;
2395 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2399 p = vm_page_lookup(object, pindex);
2400 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2404 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2409 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2414 p = vm_page_lookup(object, pindex);
2418 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2419 if (ptepa & (NBPDR - 1)) {
2423 p->valid = VM_PAGE_BITS_ALL;
2425 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2426 npdes = size >> PDRSHIFT;
2427 for(i=0;i<npdes;i++) {
2428 pmap->pm_pdir[ptepindex] =
2429 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2433 vm_page_flag_set(p, PG_MAPPED);
2438 psize = i386_btop(size);
2440 if ((object->type != OBJT_VNODE) ||
2441 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2442 (object->resident_page_count > MAX_INIT_PT))) {
2446 if (psize + pindex > object->size) {
2447 if (object->size < pindex)
2449 psize = object->size - pindex;
2454 * if we are processing a major portion of the object, then scan the
2457 if (psize > (object->resident_page_count >> 2)) {
2460 for (p = TAILQ_FIRST(&object->memq);
2461 ((objpgs > 0) && (p != NULL));
2462 p = TAILQ_NEXT(p, listq)) {
2465 if (tmpidx < pindex) {
2469 if (tmpidx >= psize) {
2473 * don't allow an madvise to blow away our really
2474 * free pages allocating pv entries.
2476 if ((limit & MAP_PREFAULT_MADVISE) &&
2477 cnt.v_free_count < cnt.v_free_reserved) {
2480 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2482 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2483 if ((p->queue - p->pc) == PQ_CACHE)
2484 vm_page_deactivate(p);
2486 mpte = pmap_enter_quick(pmap,
2487 addr + i386_ptob(tmpidx), p, mpte);
2488 vm_page_flag_set(p, PG_MAPPED);
2495 * else lookup the pages one-by-one.
2497 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2499 * don't allow an madvise to blow away our really
2500 * free pages allocating pv entries.
2502 if ((limit & MAP_PREFAULT_MADVISE) &&
2503 cnt.v_free_count < cnt.v_free_reserved) {
2506 p = vm_page_lookup(object, tmpidx + pindex);
2508 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2510 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2511 if ((p->queue - p->pc) == PQ_CACHE)
2512 vm_page_deactivate(p);
2514 mpte = pmap_enter_quick(pmap,
2515 addr + i386_ptob(tmpidx), p, mpte);
2516 vm_page_flag_set(p, PG_MAPPED);
2525 * pmap_prefault provides a quick way of clustering
2526 * pagefaults into a processes address space. It is a "cousin"
2527 * of pmap_object_init_pt, except it runs at page fault time instead
2532 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2534 static int pmap_prefault_pageorder[] = {
2535 -PAGE_SIZE, PAGE_SIZE,
2536 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2537 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2538 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2542 pmap_prefault(pmap, addra, entry)
2545 vm_map_entry_t entry;
2554 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2557 object = entry->object.vm_object;
2559 starta = addra - PFBAK * PAGE_SIZE;
2560 if (starta < entry->start) {
2561 starta = entry->start;
2562 } else if (starta > addra) {
2567 for (i = 0; i < PAGEORDER_SIZE; i++) {
2568 vm_object_t lobject;
2571 addr = addra + pmap_prefault_pageorder[i];
2572 if (addr > addra + (PFFOR * PAGE_SIZE))
2575 if (addr < starta || addr >= entry->end)
2578 if ((*pmap_pde(pmap, addr)) == NULL)
2581 pte = (unsigned *) vtopte(addr);
2585 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2587 for (m = vm_page_lookup(lobject, pindex);
2588 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2589 lobject = lobject->backing_object) {
2590 if (lobject->backing_object_offset & PAGE_MASK)
2592 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2593 m = vm_page_lookup(lobject->backing_object, pindex);
2597 * give-up when a page is not in memory
2602 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2604 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2606 if ((m->queue - m->pc) == PQ_CACHE) {
2607 vm_page_deactivate(m);
2610 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2611 vm_page_flag_set(m, PG_MAPPED);
2618 * Routine: pmap_change_wiring
2619 * Function: Change the wiring attribute for a map/virtual-address
2621 * In/out conditions:
2622 * The mapping must already exist in the pmap.
2625 pmap_change_wiring(pmap, va, wired)
2626 register pmap_t pmap;
2630 register unsigned *pte;
2635 pte = pmap_pte(pmap, va);
2637 if (wired && !pmap_pte_w(pte))
2638 pmap->pm_stats.wired_count++;
2639 else if (!wired && pmap_pte_w(pte))
2640 pmap->pm_stats.wired_count--;
2643 * Wiring is not a hardware characteristic so there is no need to
2646 pmap_pte_set_w(pte, wired);
2652 * Copy the range specified by src_addr/len
2653 * from the source map to the range dst_addr/len
2654 * in the destination map.
2656 * This routine is only advisory and need not do anything.
2660 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2661 pmap_t dst_pmap, src_pmap;
2662 vm_offset_t dst_addr;
2664 vm_offset_t src_addr;
2667 vm_offset_t end_addr = src_addr + len;
2669 unsigned src_frame, dst_frame;
2672 if (dst_addr != src_addr)
2675 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2676 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2680 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2681 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2682 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2684 /* The page directory is not shared between CPUs */
2691 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2692 unsigned *src_pte, *dst_pte;
2693 vm_page_t dstmpte, srcmpte;
2694 vm_offset_t srcptepaddr;
2697 if (addr >= UPT_MIN_ADDRESS)
2698 panic("pmap_copy: invalid to pmap_copy page tables\n");
2701 * Don't let optional prefaulting of pages make us go
2702 * way below the low water mark of free pages or way
2703 * above high water mark of used pv entries.
2705 if (cnt.v_free_count < cnt.v_free_reserved ||
2706 pv_entry_count > pv_entry_high_water)
2709 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2710 ptepindex = addr >> PDRSHIFT;
2712 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2713 if (srcptepaddr == 0)
2716 if (srcptepaddr & PG_PS) {
2717 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2718 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2719 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2724 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2725 if ((srcmpte == NULL) ||
2726 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2729 if (pdnxt > end_addr)
2732 src_pte = (unsigned *) vtopte(addr);
2733 dst_pte = (unsigned *) avtopte(addr);
2734 while (addr < pdnxt) {
2738 * we only virtual copy managed pages
2740 if ((ptetemp & PG_MANAGED) != 0) {
2742 * We have to check after allocpte for the
2743 * pte still being around... allocpte can
2746 dstmpte = pmap_allocpte(dst_pmap, addr);
2747 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2749 * Clear the modified and
2750 * accessed (referenced) bits
2753 m = PHYS_TO_VM_PAGE(ptetemp);
2754 *dst_pte = ptetemp & ~(PG_M | PG_A);
2755 dst_pmap->pm_stats.resident_count++;
2756 pmap_insert_entry(dst_pmap, addr,
2759 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2761 if (dstmpte->hold_count >= srcmpte->hold_count)
2772 * Routine: pmap_kernel
2774 * Returns the physical map handle for the kernel.
2779 return (kernel_pmap);
2783 * pmap_zero_page zeros the specified hardware page by mapping
2784 * the page into KVM and using bzero to clear its contents.
2787 pmap_zero_page(phys)
2791 if (*(int *) prv_CMAP3)
2792 panic("pmap_zero_page: prv_CMAP3 busy");
2794 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2795 cpu_invlpg(prv_CADDR3);
2797 #if defined(I686_CPU)
2798 if (cpu_class == CPUCLASS_686)
2799 i686_pagezero(prv_CADDR3);
2802 bzero(prv_CADDR3, PAGE_SIZE);
2804 *(int *) prv_CMAP3 = 0;
2807 panic("pmap_zero_page: CMAP2 busy");
2809 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2810 invltlb_1pg((vm_offset_t)CADDR2);
2812 #if defined(I686_CPU)
2813 if (cpu_class == CPUCLASS_686)
2814 i686_pagezero(CADDR2);
2817 bzero(CADDR2, PAGE_SIZE);
2823 * pmap_zero_page_area zeros the specified hardware page by mapping
2824 * the page into KVM and using bzero to clear its contents.
2826 * off and size may not cover an area beyond a single hardware page.
2829 pmap_zero_page_area(phys, off, size)
2835 if (*(int *) prv_CMAP3)
2836 panic("pmap_zero_page: prv_CMAP3 busy");
2838 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2839 cpu_invlpg(prv_CADDR3);
2841 #if defined(I686_CPU)
2842 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2843 i686_pagezero(prv_CADDR3);
2846 bzero((char *)prv_CADDR3 + off, size);
2848 *(int *) prv_CMAP3 = 0;
2851 panic("pmap_zero_page: CMAP2 busy");
2853 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2854 invltlb_1pg((vm_offset_t)CADDR2);
2856 #if defined(I686_CPU)
2857 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2858 i686_pagezero(CADDR2);
2861 bzero((char *)CADDR2 + off, size);
2867 * pmap_copy_page copies the specified (machine independent)
2868 * page by mapping the page into virtual memory and using
2869 * bcopy to copy the page, one machine dependent page at a
2873 pmap_copy_page(src, dst)
2878 if (*(int *) prv_CMAP1)
2879 panic("pmap_copy_page: prv_CMAP1 busy");
2880 if (*(int *) prv_CMAP2)
2881 panic("pmap_copy_page: prv_CMAP2 busy");
2883 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2884 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2886 cpu_invlpg(prv_CADDR1);
2887 cpu_invlpg(prv_CADDR2);
2889 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE);
2891 *(int *) prv_CMAP1 = 0;
2892 *(int *) prv_CMAP2 = 0;
2894 if (*(int *) CMAP1 || *(int *) CMAP2)
2895 panic("pmap_copy_page: CMAP busy");
2897 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2898 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2899 #if defined(I386_CPU)
2900 if (cpu_class == CPUCLASS_386) {
2905 invlpg((u_int)CADDR1);
2906 invlpg((u_int)CADDR2);
2909 bcopy(CADDR1, CADDR2, PAGE_SIZE);
2918 * Routine: pmap_pageable
2920 * Make the specified pages (by pmap, offset)
2921 * pageable (or not) as requested.
2923 * A page which is not pageable may not take
2924 * a fault; therefore, its page table entry
2925 * must remain valid for the duration.
2927 * This routine is merely advisory; pmap_enter
2928 * will specify that these pages are to be wired
2929 * down (or not) as appropriate.
2932 pmap_pageable(pmap, sva, eva, pageable)
2934 vm_offset_t sva, eva;
2940 * Returns true if the pmap's pv is one of the first
2941 * 16 pvs linked to from this page. This count may
2942 * be changed upwards or downwards in the future; it
2943 * is only necessary that true be returned for a small
2944 * subset of pmaps for proper page aging.
2947 pmap_page_exists_quick(pmap, m)
2955 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2960 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2961 if (pv->pv_pmap == pmap) {
2973 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2975 * Remove all pages from specified address space
2976 * this aids process exit speeds. Also, this code
2977 * is special cased for current process only, but
2978 * can have the more generic (and slightly slower)
2979 * mode enabled. This is much faster than pmap_remove
2980 * in the case of running down an entire address space.
2983 pmap_remove_pages(pmap, sva, eva)
2985 vm_offset_t sva, eva;
2987 unsigned *pte, tpte;
2992 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2993 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2994 printf("warning: pmap_remove_pages called with non-current pmap\n");
3000 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
3004 if (pv->pv_va >= eva || pv->pv_va < sva) {
3005 npv = TAILQ_NEXT(pv, pv_plist);
3009 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
3010 pte = (unsigned *)vtopte(pv->pv_va);
3012 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3017 * We cannot remove wired pages from a process' mapping at this time
3020 npv = TAILQ_NEXT(pv, pv_plist);
3025 m = PHYS_TO_VM_PAGE(tpte);
3027 KASSERT(m < &vm_page_array[vm_page_array_size],
3028 ("pmap_remove_pages: bad tpte %x", tpte));
3030 pv->pv_pmap->pm_stats.resident_count--;
3033 * Update the vm_page_t clean and reference bits.
3040 npv = TAILQ_NEXT(pv, pv_plist);
3041 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
3043 m->md.pv_list_count--;
3044 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3045 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
3046 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
3049 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
3053 pmap_TLB_invalidate_all(pmap);
3057 * pmap_testbit tests bits in pte's
3058 * note that the testbit/changebit routines are inline,
3059 * and a lot of things compile-time evaluate.
3062 pmap_testbit(m, bit)
3070 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3073 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3078 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3080 * if the bit being tested is the modified bit, then
3081 * mark clean_map and ptes as never
3084 if (bit & (PG_A|PG_M)) {
3085 if (!pmap_track_modified(pv->pv_va))
3089 #if defined(PMAP_DIAGNOSTIC)
3091 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
3095 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3106 * this routine is used to modify bits in ptes
3108 static __inline void
3109 pmap_changebit(m, bit, setem)
3114 register pv_entry_t pv;
3115 register unsigned *pte;
3118 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3124 * Loop over all current mappings setting/clearing as appropos If
3125 * setting RO do we need to clear the VAC?
3127 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3129 * don't write protect pager mappings
3131 if (!setem && (bit == PG_RW)) {
3132 if (!pmap_track_modified(pv->pv_va))
3136 #if defined(PMAP_DIAGNOSTIC)
3138 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
3143 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3147 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3149 vm_offset_t pbits = *(vm_offset_t *)pte;
3155 *(int *)pte = pbits & ~(PG_M|PG_RW);
3157 *(int *)pte = pbits & ~bit;
3159 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3167 * pmap_page_protect:
3169 * Lower the permission for all mappings to a given page.
3172 pmap_page_protect(vm_page_t m, vm_prot_t prot)
3174 if ((prot & VM_PROT_WRITE) == 0) {
3175 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3176 pmap_changebit(m, PG_RW, FALSE);
3184 pmap_phys_address(ppn)
3187 return (i386_ptob(ppn));
3191 * pmap_ts_referenced:
3193 * Return a count of reference bits for a page, clearing those bits.
3194 * It is not necessary for every reference bit to be cleared, but it
3195 * is necessary that 0 only be returned when there are truly no
3196 * reference bits set.
3198 * XXX: The exact number of bits to check and clear is a matter that
3199 * should be tested and standardized at some point in the future for
3200 * optimal aging of shared pages.
3203 pmap_ts_referenced(vm_page_t m)
3205 register pv_entry_t pv, pvf, pvn;
3210 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3215 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3220 pvn = TAILQ_NEXT(pv, pv_list);
3222 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3224 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3226 if (!pmap_track_modified(pv->pv_va))
3229 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3231 if (pte && (*pte & PG_A)) {
3234 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3241 } while ((pv = pvn) != NULL && pv != pvf);
3251 * Return whether or not the specified physical page was modified
3252 * in any physical maps.
3255 pmap_is_modified(vm_page_t m)
3257 return pmap_testbit(m, PG_M);
3261 * Clear the modify bits on the specified physical page.
3264 pmap_clear_modify(vm_page_t m)
3266 pmap_changebit(m, PG_M, FALSE);
3270 * pmap_clear_reference:
3272 * Clear the reference bit on the specified physical page.
3275 pmap_clear_reference(vm_page_t m)
3277 pmap_changebit(m, PG_A, FALSE);
3281 * Miscellaneous support routines follow
3285 i386_protection_init()
3287 register int *kp, prot;
3289 kp = protection_codes;
3290 for (prot = 0; prot < 8; prot++) {
3292 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3294 * Read access is also 0. There isn't any execute bit,
3295 * so just make it readable.
3297 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3298 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3299 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3302 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3303 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3304 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3305 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3313 * Map a set of physical memory pages into the kernel virtual
3314 * address space. Return a pointer to where it is mapped. This
3315 * routine is intended to be used for mapping device memory,
3319 pmap_mapdev(pa, size)
3323 vm_offset_t va, tmpva, offset;
3326 offset = pa & PAGE_MASK;
3327 size = roundup(offset + size, PAGE_SIZE);
3329 va = kmem_alloc_pageable(kernel_map, size);
3331 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3334 for (tmpva = va; size > 0;) {
3335 pte = (unsigned *)vtopte(tmpva);
3336 *pte = pa | PG_RW | PG_V | pgeflag;
3343 return ((void *)(va + offset));
3347 pmap_unmapdev(va, size)
3351 vm_offset_t base, offset;
3353 base = va & PG_FRAME;
3354 offset = va & PAGE_MASK;
3355 size = roundup(offset + size, PAGE_SIZE);
3356 kmem_free(kernel_map, base, size);
3360 * perform the pmap work for mincore
3363 pmap_mincore(pmap, addr)
3368 unsigned *ptep, pte;
3372 ptep = pmap_pte(pmap, addr);
3377 if ((pte = *ptep) != 0) {
3380 val = MINCORE_INCORE;
3381 if ((pte & PG_MANAGED) == 0)
3384 pa = pte & PG_FRAME;
3386 m = PHYS_TO_VM_PAGE(pa);
3392 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3394 * Modified by someone
3396 else if (m->dirty || pmap_is_modified(m))
3397 val |= MINCORE_MODIFIED_OTHER;
3402 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3405 * Referenced by someone
3407 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3408 val |= MINCORE_REFERENCED_OTHER;
3409 vm_page_flag_set(m, PG_REFERENCED);
3416 pmap_activate(struct proc *p)
3420 pmap = vmspace_pmap(p->p_vmspace);
3422 pmap->pm_active |= 1 << cpuid;
3424 pmap->pm_active |= 1;
3426 #if defined(SWTCH_OPTIM_STATS)
3429 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3430 load_cr3(p->p_thread->td_pcb->pcb_cr3);
3434 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3437 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3441 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3446 #if defined(PMAP_DEBUG)
3447 pmap_pid_dump(int pid)
3453 LIST_FOREACH(p, &allproc, p_list) {
3454 if (p->p_pid != pid)
3460 pmap = vmspace_pmap(p->p_vmspace);
3461 for(i=0;i<1024;i++) {
3464 unsigned base = i << PDRSHIFT;
3466 pde = &pmap->pm_pdir[i];
3467 if (pde && pmap_pde_v(pde)) {
3468 for(j=0;j<1024;j++) {
3469 unsigned va = base + (j << PAGE_SHIFT);
3470 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3477 pte = pmap_pte_quick( pmap, va);
3478 if (pte && pmap_pte_v(pte)) {
3482 m = PHYS_TO_VM_PAGE(pa);
3483 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3484 va, pa, m->hold_count, m->wire_count, m->flags);
3505 static void pads __P((pmap_t pm));
3506 void pmap_pvdump __P((vm_offset_t pa));
3508 /* print address space of pmap*/
3516 if (pm == kernel_pmap)
3518 for (i = 0; i < 1024; i++)
3520 for (j = 0; j < 1024; j++) {
3521 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3522 if (pm == kernel_pmap && va < KERNBASE)
3524 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3526 ptep = pmap_pte_quick(pm, va);
3527 if (pmap_pte_v(ptep))
3528 printf("%x:%x ", va, *(int *) ptep);
3537 register pv_entry_t pv;
3540 printf("pa %x", pa);
3541 m = PHYS_TO_VM_PAGE(pa);
3542 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3544 printf(" -> pmap %p, va %x, flags %x",
3545 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3547 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);