2 * Copyright (c) 1991 Regents of the University of California.
4 * Copyright (c) 1994 John S. Dyson
6 * Copyright (c) 1994 David Greenman
9 * This code is derived from software contributed to Berkeley by
10 * the Systems Programming Group of the University of Utah Computer
11 * Science Department and William Jolitz of UUNET Technologies Inc.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91
42 * $FreeBSD: src/sys/i386/i386/pmap.c,v 1.250.2.18 2002/03/06 22:48:53 silby Exp $
43 * $DragonFly: src/sys/platform/pc32/i386/pmap.c,v 1.4 2003/06/18 16:30:09 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 = &SMP_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 = SMP_prvspace[0].CPAGE1;
438 gd->gd_prv_CADDR2 = SMP_prvspace[0].CPAGE2;
439 gd->gd_prv_CADDR3 = SMP_prvspace[0].CPAGE3;
440 gd->gd_prv_PADDR1 = (unsigned *)SMP_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.
847 pmap_new_thread(struct proc *p)
849 struct thread *td = zalloc(thread_zone);
858 * Dispose of a thread, unlink from its related proc (if any)
861 pmap_dispose_thread(struct thread *td)
864 td->td_proc->p_thread = NULL;
867 zfree(thread_zone, td);
871 * Create the UPAGES for a new process.
872 * This routine directly affects the fork perf for a process.
882 unsigned *ptek, oldpte;
885 * allocate object for the upages
887 if ((upobj = p->p_upages_obj) == NULL) {
888 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES);
889 p->p_upages_obj = upobj;
892 /* get a kernel virtual address for the UPAGES for this proc */
893 if ((up = p->p_addr) == NULL) {
894 up = (struct user *) kmem_alloc_nofault(kernel_map,
897 panic("pmap_new_proc: u_map allocation failed");
901 ptek = (unsigned *) vtopte((vm_offset_t) up);
904 for(i=0;i<UPAGES;i++) {
906 * Get a kernel stack page
908 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
916 oldpte = *(ptek + i);
918 * Enter the page into the kernel address space.
920 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag;
922 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) {
923 invlpg((vm_offset_t) up + i * PAGE_SIZE);
930 vm_page_flag_clear(m, PG_ZERO);
931 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
932 m->valid = VM_PAGE_BITS_ALL;
939 * Dispose the UPAGES for a process that has exited.
940 * This routine directly impacts the exit perf of a process.
949 unsigned *ptek, oldpte;
951 upobj = p->p_upages_obj;
953 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr);
954 for(i=0;i<UPAGES;i++) {
956 if ((m = vm_page_lookup(upobj, i)) == NULL)
957 panic("pmap_dispose_proc: upage already missing???");
961 oldpte = *(ptek + i);
963 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386))
964 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE);
965 vm_page_unwire(m, 0);
968 #if defined(I386_CPU)
969 if (cpu_class <= CPUCLASS_386)
974 * If the process got swapped out some of its UPAGES might have gotten
975 * swapped. Just get rid of the object to clean up the swap use
976 * proactively. NOTE! might block waiting for paging I/O to complete.
978 if (upobj->type == OBJT_SWAP) {
979 p->p_upages_obj = NULL;
980 vm_object_deallocate(upobj);
985 * Allow the UPAGES for a process to be prejudicially paged out.
995 upobj = p->p_upages_obj;
997 * let the upages be paged
999 for(i=0;i<UPAGES;i++) {
1000 if ((m = vm_page_lookup(upobj, i)) == NULL)
1001 panic("pmap_swapout_proc: upage already missing???");
1003 vm_page_unwire(m, 0);
1004 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
1009 * Bring the UPAGES for a specified process back in.
1019 upobj = p->p_upages_obj;
1020 for(i=0;i<UPAGES;i++) {
1022 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1024 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
1025 VM_PAGE_TO_PHYS(m));
1027 if (m->valid != VM_PAGE_BITS_ALL) {
1028 rv = vm_pager_get_pages(upobj, &m, 1, 0);
1029 if (rv != VM_PAGER_OK)
1030 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
1031 m = vm_page_lookup(upobj, i);
1032 m->valid = VM_PAGE_BITS_ALL;
1037 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
1041 /***************************************************
1042 * Page table page management routines.....
1043 ***************************************************/
1046 * This routine unholds page table pages, and if the hold count
1047 * drops to zero, then it decrements the wire count.
1050 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
1052 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
1055 if (m->hold_count == 0) {
1058 * unmap the page table page
1060 pmap->pm_pdir[m->pindex] = 0;
1061 --pmap->pm_stats.resident_count;
1062 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1063 (((unsigned) PTDpde) & PG_FRAME)) {
1065 * Do a invltlb to make the invalidated mapping
1066 * take effect immediately.
1068 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
1069 pmap_TLB_invalidate(pmap, pteva);
1072 if (pmap->pm_ptphint == m)
1073 pmap->pm_ptphint = NULL;
1076 * If the page is finally unwired, simply free it.
1079 if (m->wire_count == 0) {
1083 vm_page_free_zero(m);
1091 static PMAP_INLINE int
1092 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
1095 if (m->hold_count == 0)
1096 return _pmap_unwire_pte_hold(pmap, m);
1102 * After removing a page table entry, this routine is used to
1103 * conditionally free the page, and manage the hold/wire counts.
1106 pmap_unuse_pt(pmap, va, mpte)
1112 if (va >= UPT_MIN_ADDRESS)
1116 ptepindex = (va >> PDRSHIFT);
1117 if (pmap->pm_ptphint &&
1118 (pmap->pm_ptphint->pindex == ptepindex)) {
1119 mpte = pmap->pm_ptphint;
1121 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1122 pmap->pm_ptphint = mpte;
1126 return pmap_unwire_pte_hold(pmap, mpte);
1134 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1135 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1137 pmap->pm_active = 0;
1138 pmap->pm_ptphint = NULL;
1139 TAILQ_INIT(&pmap->pm_pvlist);
1140 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1144 * Initialize a preallocated and zeroed pmap structure,
1145 * such as one in a vmspace structure.
1149 register struct pmap *pmap;
1154 * No need to allocate page table space yet but we do need a valid
1155 * page directory table.
1157 if (pmap->pm_pdir == NULL)
1159 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1162 * allocate object for the ptes
1164 if (pmap->pm_pteobj == NULL)
1165 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1168 * allocate the page directory page
1170 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1171 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1173 ptdpg->wire_count = 1;
1177 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1178 ptdpg->valid = VM_PAGE_BITS_ALL;
1180 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1181 if ((ptdpg->flags & PG_ZERO) == 0)
1182 bzero(pmap->pm_pdir, PAGE_SIZE);
1185 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1188 /* install self-referential address mapping entry */
1189 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1190 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1193 pmap->pm_active = 0;
1194 pmap->pm_ptphint = NULL;
1195 TAILQ_INIT(&pmap->pm_pvlist);
1196 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1200 * Wire in kernel global address entries. To avoid a race condition
1201 * between pmap initialization and pmap_growkernel, this procedure
1202 * should be called after the vmspace is attached to the process
1203 * but before this pmap is activated.
1209 /* XXX copies current process, does not fill in MPPTDI */
1210 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1214 pmap_release_free_page(pmap, p)
1218 unsigned *pde = (unsigned *) pmap->pm_pdir;
1220 * This code optimizes the case of freeing non-busy
1221 * page-table pages. Those pages are zero now, and
1222 * might as well be placed directly into the zero queue.
1224 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1230 * Remove the page table page from the processes address space.
1233 pmap->pm_stats.resident_count--;
1235 if (p->hold_count) {
1236 panic("pmap_release: freeing held page table page");
1239 * Page directory pages need to have the kernel
1240 * stuff cleared, so they can go into the zero queue also.
1242 if (p->pindex == PTDPTDI) {
1243 bzero(pde + KPTDI, nkpt * PTESIZE);
1248 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1251 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1252 pmap->pm_ptphint = NULL;
1256 vm_page_free_zero(p);
1261 * this routine is called if the page table page is not
1265 _pmap_allocpte(pmap, ptepindex)
1269 vm_offset_t pteva, ptepa;
1273 * Find or fabricate a new pagetable page
1275 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1276 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1278 KASSERT(m->queue == PQ_NONE,
1279 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1281 if (m->wire_count == 0)
1286 * Increment the hold count for the page table page
1287 * (denoting a new mapping.)
1292 * Map the pagetable page into the process address space, if
1293 * it isn't already there.
1296 pmap->pm_stats.resident_count++;
1298 ptepa = VM_PAGE_TO_PHYS(m);
1299 pmap->pm_pdir[ptepindex] =
1300 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1303 * Set the page table hint
1305 pmap->pm_ptphint = m;
1308 * Try to use the new mapping, but if we cannot, then
1309 * do it with the routine that maps the page explicitly.
1311 if ((m->flags & PG_ZERO) == 0) {
1312 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1313 (((unsigned) PTDpde) & PG_FRAME)) {
1314 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1315 bzero((caddr_t) pteva, PAGE_SIZE);
1317 pmap_zero_page(ptepa);
1321 m->valid = VM_PAGE_BITS_ALL;
1322 vm_page_flag_clear(m, PG_ZERO);
1323 vm_page_flag_set(m, PG_MAPPED);
1330 pmap_allocpte(pmap, va)
1339 * Calculate pagetable page index
1341 ptepindex = va >> PDRSHIFT;
1344 * Get the page directory entry
1346 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1349 * This supports switching from a 4MB page to a
1352 if (ptepa & PG_PS) {
1353 pmap->pm_pdir[ptepindex] = 0;
1359 * If the page table page is mapped, we just increment the
1360 * hold count, and activate it.
1364 * In order to get the page table page, try the
1367 if (pmap->pm_ptphint &&
1368 (pmap->pm_ptphint->pindex == ptepindex)) {
1369 m = pmap->pm_ptphint;
1371 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1372 pmap->pm_ptphint = m;
1378 * Here if the pte page isn't mapped, or if it has been deallocated.
1380 return _pmap_allocpte(pmap, ptepindex);
1384 /***************************************************
1385 * Pmap allocation/deallocation routines.
1386 ***************************************************/
1389 * Release any resources held by the given physical map.
1390 * Called when a pmap initialized by pmap_pinit is being released.
1391 * Should only be called if the map contains no valid mappings.
1395 register struct pmap *pmap;
1397 vm_page_t p,n,ptdpg;
1398 vm_object_t object = pmap->pm_pteobj;
1401 #if defined(DIAGNOSTIC)
1402 if (object->ref_count != 1)
1403 panic("pmap_release: pteobj reference count != 1");
1408 curgeneration = object->generation;
1409 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1410 n = TAILQ_NEXT(p, listq);
1411 if (p->pindex == PTDPTDI) {
1416 if (!pmap_release_free_page(pmap, p) &&
1417 (object->generation != curgeneration))
1422 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1427 kvm_size(SYSCTL_HANDLER_ARGS)
1429 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1431 return sysctl_handle_long(oidp, &ksize, 0, req);
1433 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1434 0, 0, kvm_size, "IU", "Size of KVM");
1437 kvm_free(SYSCTL_HANDLER_ARGS)
1439 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1441 return sysctl_handle_long(oidp, &kfree, 0, req);
1443 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1444 0, 0, kvm_free, "IU", "Amount of KVM free");
1447 * grow the number of kernel page table entries, if needed
1450 pmap_growkernel(vm_offset_t addr)
1455 vm_offset_t ptppaddr;
1460 if (kernel_vm_end == 0) {
1461 kernel_vm_end = KERNBASE;
1463 while (pdir_pde(PTD, kernel_vm_end)) {
1464 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1468 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1469 while (kernel_vm_end < addr) {
1470 if (pdir_pde(PTD, kernel_vm_end)) {
1471 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1476 * This index is bogus, but out of the way
1478 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1480 panic("pmap_growkernel: no memory to grow kernel");
1485 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1486 pmap_zero_page(ptppaddr);
1487 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1488 pdir_pde(PTD, kernel_vm_end) = newpdir;
1490 LIST_FOREACH(p, &allproc, p_list) {
1492 pmap = vmspace_pmap(p->p_vmspace);
1493 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1496 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1497 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1503 * Retire the given physical map from service.
1504 * Should only be called if the map contains
1505 * no valid mappings.
1509 register pmap_t pmap;
1516 count = --pmap->pm_count;
1519 panic("destroying a pmap is not yet implemented");
1524 * Add a reference to the specified pmap.
1527 pmap_reference(pmap)
1535 /***************************************************
1536 * page management routines.
1537 ***************************************************/
1540 * free the pv_entry back to the free list
1542 static PMAP_INLINE void
1551 * get a new pv_entry, allocating a block from the system
1553 * the memory allocation is performed bypassing the malloc code
1554 * because of the possibility of allocations at interrupt time.
1560 if (pv_entry_high_water &&
1561 (pv_entry_count > pv_entry_high_water) &&
1562 (pmap_pagedaemon_waken == 0)) {
1563 pmap_pagedaemon_waken = 1;
1564 wakeup (&vm_pages_needed);
1566 return zalloci(pvzone);
1570 * This routine is very drastic, but can save the system
1578 static int warningdone=0;
1580 if (pmap_pagedaemon_waken == 0)
1583 if (warningdone < 5) {
1584 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1588 for(i = 0; i < vm_page_array_size; i++) {
1589 m = &vm_page_array[i];
1590 if (m->wire_count || m->hold_count || m->busy ||
1591 (m->flags & PG_BUSY))
1595 pmap_pagedaemon_waken = 0;
1600 * If it is the first entry on the list, it is actually
1601 * in the header and we must copy the following entry up
1602 * to the header. Otherwise we must search the list for
1603 * the entry. In either case we free the now unused entry.
1607 pmap_remove_entry(pmap, m, va)
1617 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1618 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1619 if (pmap == pv->pv_pmap && va == pv->pv_va)
1623 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1624 if (va == pv->pv_va)
1632 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1633 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1634 m->md.pv_list_count--;
1635 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1636 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1638 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1647 * Create a pv entry for page at pa for
1651 pmap_insert_entry(pmap, va, mpte, m)
1662 pv = get_pv_entry();
1667 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1668 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1669 m->md.pv_list_count++;
1675 * pmap_remove_pte: do the things to unmap a page in a process
1678 pmap_remove_pte(pmap, ptq, va)
1686 oldpte = loadandclear(ptq);
1688 pmap->pm_stats.wired_count -= 1;
1690 * Machines that don't support invlpg, also don't support
1695 pmap->pm_stats.resident_count -= 1;
1696 if (oldpte & PG_MANAGED) {
1697 m = PHYS_TO_VM_PAGE(oldpte);
1698 if (oldpte & PG_M) {
1699 #if defined(PMAP_DIAGNOSTIC)
1700 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1702 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1706 if (pmap_track_modified(va))
1710 vm_page_flag_set(m, PG_REFERENCED);
1711 return pmap_remove_entry(pmap, m, va);
1713 return pmap_unuse_pt(pmap, va, NULL);
1720 * Remove a single page from a process address space
1723 pmap_remove_page(pmap, va)
1725 register vm_offset_t va;
1727 register unsigned *ptq;
1730 * if there is no pte for this address, just skip it!!!
1732 if (*pmap_pde(pmap, va) == 0) {
1737 * get a local va for mappings for this pmap.
1739 ptq = get_ptbase(pmap) + i386_btop(va);
1741 (void) pmap_remove_pte(pmap, ptq, va);
1742 pmap_TLB_invalidate(pmap, va);
1748 * Remove the given range of addresses from the specified map.
1750 * It is assumed that the start and end are properly
1751 * rounded to the page size.
1754 pmap_remove(pmap, sva, eva)
1756 register vm_offset_t sva;
1757 register vm_offset_t eva;
1759 register unsigned *ptbase;
1761 vm_offset_t ptpaddr;
1762 vm_offset_t sindex, eindex;
1768 if (pmap->pm_stats.resident_count == 0)
1772 * special handling of removing one page. a very
1773 * common operation and easy to short circuit some
1776 if (((sva + PAGE_SIZE) == eva) &&
1777 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1778 pmap_remove_page(pmap, sva);
1785 * Get a local virtual address for the mappings that are being
1788 ptbase = get_ptbase(pmap);
1790 sindex = i386_btop(sva);
1791 eindex = i386_btop(eva);
1793 for (; sindex < eindex; sindex = pdnxt) {
1797 * Calculate index for next page table.
1799 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1800 if (pmap->pm_stats.resident_count == 0)
1803 pdirindex = sindex / NPDEPG;
1804 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1805 pmap->pm_pdir[pdirindex] = 0;
1806 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1812 * Weed out invalid mappings. Note: we assume that the page
1813 * directory table is always allocated, and in kernel virtual.
1819 * Limit our scan to either the end of the va represented
1820 * by the current page table page, or to the end of the
1821 * range being removed.
1823 if (pdnxt > eindex) {
1827 for ( ;sindex != pdnxt; sindex++) {
1829 if (ptbase[sindex] == 0) {
1832 va = i386_ptob(sindex);
1835 if (pmap_remove_pte(pmap,
1836 ptbase + sindex, va))
1842 pmap_TLB_invalidate_all(pmap);
1846 * Routine: pmap_remove_all
1848 * Removes this physical page from
1849 * all physical maps in which it resides.
1850 * Reflects back modify bits to the pager.
1853 * Original versions of this routine were very
1854 * inefficient because they iteratively called
1855 * pmap_remove (slow...)
1862 register pv_entry_t pv;
1863 register unsigned *pte, tpte;
1866 #if defined(PMAP_DIAGNOSTIC)
1868 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1871 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1872 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1877 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1878 pv->pv_pmap->pm_stats.resident_count--;
1880 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1882 tpte = loadandclear(pte);
1884 pv->pv_pmap->pm_stats.wired_count--;
1887 vm_page_flag_set(m, PG_REFERENCED);
1890 * Update the vm_page_t clean and reference bits.
1893 #if defined(PMAP_DIAGNOSTIC)
1894 if (pmap_nw_modified((pt_entry_t) tpte)) {
1896 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1900 if (pmap_track_modified(pv->pv_va))
1903 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1905 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1906 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1907 m->md.pv_list_count--;
1908 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1912 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1918 * Set the physical protection on the
1919 * specified range of this map as requested.
1922 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1924 register unsigned *ptbase;
1925 vm_offset_t pdnxt, ptpaddr;
1926 vm_pindex_t sindex, eindex;
1932 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1933 pmap_remove(pmap, sva, eva);
1937 if (prot & VM_PROT_WRITE)
1942 ptbase = get_ptbase(pmap);
1944 sindex = i386_btop(sva);
1945 eindex = i386_btop(eva);
1947 for (; sindex < eindex; sindex = pdnxt) {
1951 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1953 pdirindex = sindex / NPDEPG;
1954 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1955 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1956 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1962 * Weed out invalid mappings. Note: we assume that the page
1963 * directory table is always allocated, and in kernel virtual.
1968 if (pdnxt > eindex) {
1972 for (; sindex != pdnxt; sindex++) {
1977 pbits = ptbase[sindex];
1979 if (pbits & PG_MANAGED) {
1982 m = PHYS_TO_VM_PAGE(pbits);
1983 vm_page_flag_set(m, PG_REFERENCED);
1987 if (pmap_track_modified(i386_ptob(sindex))) {
1989 m = PHYS_TO_VM_PAGE(pbits);
1998 if (pbits != ptbase[sindex]) {
1999 ptbase[sindex] = pbits;
2005 pmap_TLB_invalidate_all(pmap);
2009 * Insert the given physical page (p) at
2010 * the specified virtual address (v) in the
2011 * target physical map with the protection requested.
2013 * If specified, the page will be wired down, meaning
2014 * that the related pte can not be reclaimed.
2016 * NB: This is the only routine which MAY NOT lazy-evaluate
2017 * or lose information. That is, this routine must actually
2018 * insert this page into the given map NOW.
2021 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
2025 register unsigned *pte;
2027 vm_offset_t origpte, newpte;
2034 #ifdef PMAP_DIAGNOSTIC
2035 if (va > VM_MAX_KERNEL_ADDRESS)
2036 panic("pmap_enter: toobig");
2037 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
2038 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
2043 * In the case that a page table page is not
2044 * resident, we are creating it here.
2046 if (va < UPT_MIN_ADDRESS) {
2047 mpte = pmap_allocpte(pmap, va);
2049 #if 0 && defined(PMAP_DIAGNOSTIC)
2051 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
2052 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
2053 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
2054 pmap->pm_pdir[PTDPTDI], origpte, va);
2057 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
2058 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
2059 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
2060 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
2061 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
2062 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
2063 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
2069 pte = pmap_pte(pmap, va);
2072 * Page Directory table entry not valid, we need a new PT page
2075 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
2076 (void *)pmap->pm_pdir[PTDPTDI], va);
2079 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2080 origpte = *(vm_offset_t *)pte;
2081 opa = origpte & PG_FRAME;
2083 if (origpte & PG_PS)
2084 panic("pmap_enter: attempted pmap_enter on 4MB page");
2087 * Mapping has not changed, must be protection or wiring change.
2089 if (origpte && (opa == pa)) {
2091 * Wiring change, just update stats. We don't worry about
2092 * wiring PT pages as they remain resident as long as there
2093 * are valid mappings in them. Hence, if a user page is wired,
2094 * the PT page will be also.
2096 if (wired && ((origpte & PG_W) == 0))
2097 pmap->pm_stats.wired_count++;
2098 else if (!wired && (origpte & PG_W))
2099 pmap->pm_stats.wired_count--;
2101 #if defined(PMAP_DIAGNOSTIC)
2102 if (pmap_nw_modified((pt_entry_t) origpte)) {
2104 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
2110 * Remove extra pte reference
2115 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
2116 if ((origpte & PG_RW) == 0) {
2119 cpu_invlpg((void *)va);
2120 if (pmap->pm_active & other_cpus)
2130 * We might be turning off write access to the page,
2131 * so we go ahead and sense modify status.
2133 if (origpte & PG_MANAGED) {
2134 if ((origpte & PG_M) && pmap_track_modified(va)) {
2136 om = PHYS_TO_VM_PAGE(opa);
2144 * Mapping has changed, invalidate old range and fall through to
2145 * handle validating new mapping.
2149 err = pmap_remove_pte(pmap, pte, va);
2151 panic("pmap_enter: pte vanished, va: 0x%x", va);
2155 * Enter on the PV list if part of our managed memory. Note that we
2156 * raise IPL while manipulating pv_table since pmap_enter can be
2157 * called at interrupt time.
2159 if (pmap_initialized &&
2160 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2161 pmap_insert_entry(pmap, va, mpte, m);
2166 * Increment counters
2168 pmap->pm_stats.resident_count++;
2170 pmap->pm_stats.wired_count++;
2174 * Now validate mapping with desired protection/wiring.
2176 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2180 if (va < UPT_MIN_ADDRESS)
2182 if (pmap == kernel_pmap)
2186 * if the mapping or permission bits are different, we need
2187 * to update the pte.
2189 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2190 *pte = newpte | PG_A;
2193 cpu_invlpg((void *)va);
2194 if (pmap->pm_active & other_cpus)
2204 * this code makes some *MAJOR* assumptions:
2205 * 1. Current pmap & pmap exists.
2208 * 4. No page table pages.
2209 * 5. Tlbflush is deferred to calling procedure.
2210 * 6. Page IS managed.
2211 * but is *MUCH* faster than pmap_enter...
2215 pmap_enter_quick(pmap, va, m, mpte)
2216 register pmap_t pmap;
2225 * In the case that a page table page is not
2226 * resident, we are creating it here.
2228 if (va < UPT_MIN_ADDRESS) {
2233 * Calculate pagetable page index
2235 ptepindex = va >> PDRSHIFT;
2236 if (mpte && (mpte->pindex == ptepindex)) {
2241 * Get the page directory entry
2243 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2246 * If the page table page is mapped, we just increment
2247 * the hold count, and activate it.
2251 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2252 if (pmap->pm_ptphint &&
2253 (pmap->pm_ptphint->pindex == ptepindex)) {
2254 mpte = pmap->pm_ptphint;
2256 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2257 pmap->pm_ptphint = mpte;
2263 mpte = _pmap_allocpte(pmap, ptepindex);
2271 * This call to vtopte makes the assumption that we are
2272 * entering the page into the current pmap. In order to support
2273 * quick entry into any pmap, one would likely use pmap_pte_quick.
2274 * But that isn't as quick as vtopte.
2276 pte = (unsigned *)vtopte(va);
2279 pmap_unwire_pte_hold(pmap, mpte);
2284 * Enter on the PV list if part of our managed memory. Note that we
2285 * raise IPL while manipulating pv_table since pmap_enter can be
2286 * called at interrupt time.
2288 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2289 pmap_insert_entry(pmap, va, mpte, m);
2292 * Increment counters
2294 pmap->pm_stats.resident_count++;
2296 pa = VM_PAGE_TO_PHYS(m);
2299 * Now validate mapping with RO protection
2301 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2302 *pte = pa | PG_V | PG_U;
2304 *pte = pa | PG_V | PG_U | PG_MANAGED;
2310 * Make a temporary mapping for a physical address. This is only intended
2311 * to be used for panic dumps.
2314 pmap_kenter_temporary(vm_offset_t pa, int i)
2316 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2317 return ((void *)crashdumpmap);
2320 #define MAX_INIT_PT (96)
2322 * pmap_object_init_pt preloads the ptes for a given object
2323 * into the specified pmap. This eliminates the blast of soft
2324 * faults on process startup and immediately after an mmap.
2327 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2340 if (pmap == NULL || object == NULL)
2344 * This code maps large physical mmap regions into the
2345 * processor address space. Note that some shortcuts
2346 * are taken, but the code works.
2349 (object->type == OBJT_DEVICE) &&
2350 ((addr & (NBPDR - 1)) == 0) &&
2351 ((size & (NBPDR - 1)) == 0) ) {
2354 unsigned int ptepindex;
2358 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2362 p = vm_page_lookup(object, pindex);
2363 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2367 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2372 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2377 p = vm_page_lookup(object, pindex);
2381 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2382 if (ptepa & (NBPDR - 1)) {
2386 p->valid = VM_PAGE_BITS_ALL;
2388 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2389 npdes = size >> PDRSHIFT;
2390 for(i=0;i<npdes;i++) {
2391 pmap->pm_pdir[ptepindex] =
2392 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2396 vm_page_flag_set(p, PG_MAPPED);
2401 psize = i386_btop(size);
2403 if ((object->type != OBJT_VNODE) ||
2404 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2405 (object->resident_page_count > MAX_INIT_PT))) {
2409 if (psize + pindex > object->size) {
2410 if (object->size < pindex)
2412 psize = object->size - pindex;
2417 * if we are processing a major portion of the object, then scan the
2420 if (psize > (object->resident_page_count >> 2)) {
2423 for (p = TAILQ_FIRST(&object->memq);
2424 ((objpgs > 0) && (p != NULL));
2425 p = TAILQ_NEXT(p, listq)) {
2428 if (tmpidx < pindex) {
2432 if (tmpidx >= psize) {
2436 * don't allow an madvise to blow away our really
2437 * free pages allocating pv entries.
2439 if ((limit & MAP_PREFAULT_MADVISE) &&
2440 cnt.v_free_count < cnt.v_free_reserved) {
2443 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2445 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2446 if ((p->queue - p->pc) == PQ_CACHE)
2447 vm_page_deactivate(p);
2449 mpte = pmap_enter_quick(pmap,
2450 addr + i386_ptob(tmpidx), p, mpte);
2451 vm_page_flag_set(p, PG_MAPPED);
2458 * else lookup the pages one-by-one.
2460 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2462 * don't allow an madvise to blow away our really
2463 * free pages allocating pv entries.
2465 if ((limit & MAP_PREFAULT_MADVISE) &&
2466 cnt.v_free_count < cnt.v_free_reserved) {
2469 p = vm_page_lookup(object, tmpidx + pindex);
2471 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2473 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2474 if ((p->queue - p->pc) == PQ_CACHE)
2475 vm_page_deactivate(p);
2477 mpte = pmap_enter_quick(pmap,
2478 addr + i386_ptob(tmpidx), p, mpte);
2479 vm_page_flag_set(p, PG_MAPPED);
2488 * pmap_prefault provides a quick way of clustering
2489 * pagefaults into a processes address space. It is a "cousin"
2490 * of pmap_object_init_pt, except it runs at page fault time instead
2495 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2497 static int pmap_prefault_pageorder[] = {
2498 -PAGE_SIZE, PAGE_SIZE,
2499 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2500 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2501 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2505 pmap_prefault(pmap, addra, entry)
2508 vm_map_entry_t entry;
2517 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2520 object = entry->object.vm_object;
2522 starta = addra - PFBAK * PAGE_SIZE;
2523 if (starta < entry->start) {
2524 starta = entry->start;
2525 } else if (starta > addra) {
2530 for (i = 0; i < PAGEORDER_SIZE; i++) {
2531 vm_object_t lobject;
2534 addr = addra + pmap_prefault_pageorder[i];
2535 if (addr > addra + (PFFOR * PAGE_SIZE))
2538 if (addr < starta || addr >= entry->end)
2541 if ((*pmap_pde(pmap, addr)) == NULL)
2544 pte = (unsigned *) vtopte(addr);
2548 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2550 for (m = vm_page_lookup(lobject, pindex);
2551 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2552 lobject = lobject->backing_object) {
2553 if (lobject->backing_object_offset & PAGE_MASK)
2555 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2556 m = vm_page_lookup(lobject->backing_object, pindex);
2560 * give-up when a page is not in memory
2565 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2567 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2569 if ((m->queue - m->pc) == PQ_CACHE) {
2570 vm_page_deactivate(m);
2573 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2574 vm_page_flag_set(m, PG_MAPPED);
2581 * Routine: pmap_change_wiring
2582 * Function: Change the wiring attribute for a map/virtual-address
2584 * In/out conditions:
2585 * The mapping must already exist in the pmap.
2588 pmap_change_wiring(pmap, va, wired)
2589 register pmap_t pmap;
2593 register unsigned *pte;
2598 pte = pmap_pte(pmap, va);
2600 if (wired && !pmap_pte_w(pte))
2601 pmap->pm_stats.wired_count++;
2602 else if (!wired && pmap_pte_w(pte))
2603 pmap->pm_stats.wired_count--;
2606 * Wiring is not a hardware characteristic so there is no need to
2609 pmap_pte_set_w(pte, wired);
2615 * Copy the range specified by src_addr/len
2616 * from the source map to the range dst_addr/len
2617 * in the destination map.
2619 * This routine is only advisory and need not do anything.
2623 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2624 pmap_t dst_pmap, src_pmap;
2625 vm_offset_t dst_addr;
2627 vm_offset_t src_addr;
2630 vm_offset_t end_addr = src_addr + len;
2632 unsigned src_frame, dst_frame;
2635 if (dst_addr != src_addr)
2638 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2639 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2643 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2644 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2645 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2647 /* The page directory is not shared between CPUs */
2654 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2655 unsigned *src_pte, *dst_pte;
2656 vm_page_t dstmpte, srcmpte;
2657 vm_offset_t srcptepaddr;
2660 if (addr >= UPT_MIN_ADDRESS)
2661 panic("pmap_copy: invalid to pmap_copy page tables\n");
2664 * Don't let optional prefaulting of pages make us go
2665 * way below the low water mark of free pages or way
2666 * above high water mark of used pv entries.
2668 if (cnt.v_free_count < cnt.v_free_reserved ||
2669 pv_entry_count > pv_entry_high_water)
2672 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2673 ptepindex = addr >> PDRSHIFT;
2675 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2676 if (srcptepaddr == 0)
2679 if (srcptepaddr & PG_PS) {
2680 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2681 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2682 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2687 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2688 if ((srcmpte == NULL) ||
2689 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2692 if (pdnxt > end_addr)
2695 src_pte = (unsigned *) vtopte(addr);
2696 dst_pte = (unsigned *) avtopte(addr);
2697 while (addr < pdnxt) {
2701 * we only virtual copy managed pages
2703 if ((ptetemp & PG_MANAGED) != 0) {
2705 * We have to check after allocpte for the
2706 * pte still being around... allocpte can
2709 dstmpte = pmap_allocpte(dst_pmap, addr);
2710 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2712 * Clear the modified and
2713 * accessed (referenced) bits
2716 m = PHYS_TO_VM_PAGE(ptetemp);
2717 *dst_pte = ptetemp & ~(PG_M | PG_A);
2718 dst_pmap->pm_stats.resident_count++;
2719 pmap_insert_entry(dst_pmap, addr,
2722 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2724 if (dstmpte->hold_count >= srcmpte->hold_count)
2735 * Routine: pmap_kernel
2737 * Returns the physical map handle for the kernel.
2742 return (kernel_pmap);
2746 * pmap_zero_page zeros the specified hardware page by mapping
2747 * the page into KVM and using bzero to clear its contents.
2750 pmap_zero_page(phys)
2754 if (*(int *) prv_CMAP3)
2755 panic("pmap_zero_page: prv_CMAP3 busy");
2757 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2758 cpu_invlpg(prv_CADDR3);
2760 #if defined(I686_CPU)
2761 if (cpu_class == CPUCLASS_686)
2762 i686_pagezero(prv_CADDR3);
2765 bzero(prv_CADDR3, PAGE_SIZE);
2767 *(int *) prv_CMAP3 = 0;
2770 panic("pmap_zero_page: CMAP2 busy");
2772 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2773 invltlb_1pg((vm_offset_t)CADDR2);
2775 #if defined(I686_CPU)
2776 if (cpu_class == CPUCLASS_686)
2777 i686_pagezero(CADDR2);
2780 bzero(CADDR2, PAGE_SIZE);
2786 * pmap_zero_page_area zeros the specified hardware page by mapping
2787 * the page into KVM and using bzero to clear its contents.
2789 * off and size may not cover an area beyond a single hardware page.
2792 pmap_zero_page_area(phys, off, size)
2798 if (*(int *) prv_CMAP3)
2799 panic("pmap_zero_page: prv_CMAP3 busy");
2801 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2802 cpu_invlpg(prv_CADDR3);
2804 #if defined(I686_CPU)
2805 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2806 i686_pagezero(prv_CADDR3);
2809 bzero((char *)prv_CADDR3 + off, size);
2811 *(int *) prv_CMAP3 = 0;
2814 panic("pmap_zero_page: CMAP2 busy");
2816 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2817 invltlb_1pg((vm_offset_t)CADDR2);
2819 #if defined(I686_CPU)
2820 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2821 i686_pagezero(CADDR2);
2824 bzero((char *)CADDR2 + off, size);
2830 * pmap_copy_page copies the specified (machine independent)
2831 * page by mapping the page into virtual memory and using
2832 * bcopy to copy the page, one machine dependent page at a
2836 pmap_copy_page(src, dst)
2841 if (*(int *) prv_CMAP1)
2842 panic("pmap_copy_page: prv_CMAP1 busy");
2843 if (*(int *) prv_CMAP2)
2844 panic("pmap_copy_page: prv_CMAP2 busy");
2846 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2847 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2849 cpu_invlpg(prv_CADDR1);
2850 cpu_invlpg(prv_CADDR2);
2852 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE);
2854 *(int *) prv_CMAP1 = 0;
2855 *(int *) prv_CMAP2 = 0;
2857 if (*(int *) CMAP1 || *(int *) CMAP2)
2858 panic("pmap_copy_page: CMAP busy");
2860 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2861 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2862 #if defined(I386_CPU)
2863 if (cpu_class == CPUCLASS_386) {
2868 invlpg((u_int)CADDR1);
2869 invlpg((u_int)CADDR2);
2872 bcopy(CADDR1, CADDR2, PAGE_SIZE);
2881 * Routine: pmap_pageable
2883 * Make the specified pages (by pmap, offset)
2884 * pageable (or not) as requested.
2886 * A page which is not pageable may not take
2887 * a fault; therefore, its page table entry
2888 * must remain valid for the duration.
2890 * This routine is merely advisory; pmap_enter
2891 * will specify that these pages are to be wired
2892 * down (or not) as appropriate.
2895 pmap_pageable(pmap, sva, eva, pageable)
2897 vm_offset_t sva, eva;
2903 * Returns true if the pmap's pv is one of the first
2904 * 16 pvs linked to from this page. This count may
2905 * be changed upwards or downwards in the future; it
2906 * is only necessary that true be returned for a small
2907 * subset of pmaps for proper page aging.
2910 pmap_page_exists_quick(pmap, m)
2918 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2923 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2924 if (pv->pv_pmap == pmap) {
2936 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2938 * Remove all pages from specified address space
2939 * this aids process exit speeds. Also, this code
2940 * is special cased for current process only, but
2941 * can have the more generic (and slightly slower)
2942 * mode enabled. This is much faster than pmap_remove
2943 * in the case of running down an entire address space.
2946 pmap_remove_pages(pmap, sva, eva)
2948 vm_offset_t sva, eva;
2950 unsigned *pte, tpte;
2955 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2956 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2957 printf("warning: pmap_remove_pages called with non-current pmap\n");
2963 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2967 if (pv->pv_va >= eva || pv->pv_va < sva) {
2968 npv = TAILQ_NEXT(pv, pv_plist);
2972 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2973 pte = (unsigned *)vtopte(pv->pv_va);
2975 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2980 * We cannot remove wired pages from a process' mapping at this time
2983 npv = TAILQ_NEXT(pv, pv_plist);
2988 m = PHYS_TO_VM_PAGE(tpte);
2990 KASSERT(m < &vm_page_array[vm_page_array_size],
2991 ("pmap_remove_pages: bad tpte %x", tpte));
2993 pv->pv_pmap->pm_stats.resident_count--;
2996 * Update the vm_page_t clean and reference bits.
3003 npv = TAILQ_NEXT(pv, pv_plist);
3004 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
3006 m->md.pv_list_count--;
3007 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3008 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
3009 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
3012 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
3016 pmap_TLB_invalidate_all(pmap);
3020 * pmap_testbit tests bits in pte's
3021 * note that the testbit/changebit routines are inline,
3022 * and a lot of things compile-time evaluate.
3025 pmap_testbit(m, bit)
3033 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3036 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3041 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3043 * if the bit being tested is the modified bit, then
3044 * mark clean_map and ptes as never
3047 if (bit & (PG_A|PG_M)) {
3048 if (!pmap_track_modified(pv->pv_va))
3052 #if defined(PMAP_DIAGNOSTIC)
3054 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
3058 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3069 * this routine is used to modify bits in ptes
3071 static __inline void
3072 pmap_changebit(m, bit, setem)
3077 register pv_entry_t pv;
3078 register unsigned *pte;
3081 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3087 * Loop over all current mappings setting/clearing as appropos If
3088 * setting RO do we need to clear the VAC?
3090 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3092 * don't write protect pager mappings
3094 if (!setem && (bit == PG_RW)) {
3095 if (!pmap_track_modified(pv->pv_va))
3099 #if defined(PMAP_DIAGNOSTIC)
3101 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
3106 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3110 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3112 vm_offset_t pbits = *(vm_offset_t *)pte;
3118 *(int *)pte = pbits & ~(PG_M|PG_RW);
3120 *(int *)pte = pbits & ~bit;
3122 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3130 * pmap_page_protect:
3132 * Lower the permission for all mappings to a given page.
3135 pmap_page_protect(vm_page_t m, vm_prot_t prot)
3137 if ((prot & VM_PROT_WRITE) == 0) {
3138 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3139 pmap_changebit(m, PG_RW, FALSE);
3147 pmap_phys_address(ppn)
3150 return (i386_ptob(ppn));
3154 * pmap_ts_referenced:
3156 * Return a count of reference bits for a page, clearing those bits.
3157 * It is not necessary for every reference bit to be cleared, but it
3158 * is necessary that 0 only be returned when there are truly no
3159 * reference bits set.
3161 * XXX: The exact number of bits to check and clear is a matter that
3162 * should be tested and standardized at some point in the future for
3163 * optimal aging of shared pages.
3166 pmap_ts_referenced(vm_page_t m)
3168 register pv_entry_t pv, pvf, pvn;
3173 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3178 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3183 pvn = TAILQ_NEXT(pv, pv_list);
3185 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3187 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3189 if (!pmap_track_modified(pv->pv_va))
3192 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3194 if (pte && (*pte & PG_A)) {
3197 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3204 } while ((pv = pvn) != NULL && pv != pvf);
3214 * Return whether or not the specified physical page was modified
3215 * in any physical maps.
3218 pmap_is_modified(vm_page_t m)
3220 return pmap_testbit(m, PG_M);
3224 * Clear the modify bits on the specified physical page.
3227 pmap_clear_modify(vm_page_t m)
3229 pmap_changebit(m, PG_M, FALSE);
3233 * pmap_clear_reference:
3235 * Clear the reference bit on the specified physical page.
3238 pmap_clear_reference(vm_page_t m)
3240 pmap_changebit(m, PG_A, FALSE);
3244 * Miscellaneous support routines follow
3248 i386_protection_init()
3250 register int *kp, prot;
3252 kp = protection_codes;
3253 for (prot = 0; prot < 8; prot++) {
3255 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3257 * Read access is also 0. There isn't any execute bit,
3258 * so just make it readable.
3260 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3261 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3262 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3265 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3266 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3267 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3268 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3276 * Map a set of physical memory pages into the kernel virtual
3277 * address space. Return a pointer to where it is mapped. This
3278 * routine is intended to be used for mapping device memory,
3282 pmap_mapdev(pa, size)
3286 vm_offset_t va, tmpva, offset;
3289 offset = pa & PAGE_MASK;
3290 size = roundup(offset + size, PAGE_SIZE);
3292 va = kmem_alloc_pageable(kernel_map, size);
3294 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3297 for (tmpva = va; size > 0;) {
3298 pte = (unsigned *)vtopte(tmpva);
3299 *pte = pa | PG_RW | PG_V | pgeflag;
3306 return ((void *)(va + offset));
3310 pmap_unmapdev(va, size)
3314 vm_offset_t base, offset;
3316 base = va & PG_FRAME;
3317 offset = va & PAGE_MASK;
3318 size = roundup(offset + size, PAGE_SIZE);
3319 kmem_free(kernel_map, base, size);
3323 * perform the pmap work for mincore
3326 pmap_mincore(pmap, addr)
3331 unsigned *ptep, pte;
3335 ptep = pmap_pte(pmap, addr);
3340 if ((pte = *ptep) != 0) {
3343 val = MINCORE_INCORE;
3344 if ((pte & PG_MANAGED) == 0)
3347 pa = pte & PG_FRAME;
3349 m = PHYS_TO_VM_PAGE(pa);
3355 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3357 * Modified by someone
3359 else if (m->dirty || pmap_is_modified(m))
3360 val |= MINCORE_MODIFIED_OTHER;
3365 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3368 * Referenced by someone
3370 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3371 val |= MINCORE_REFERENCED_OTHER;
3372 vm_page_flag_set(m, PG_REFERENCED);
3379 pmap_activate(struct proc *p)
3383 pmap = vmspace_pmap(p->p_vmspace);
3385 pmap->pm_active |= 1 << cpuid;
3387 pmap->pm_active |= 1;
3389 #if defined(SWTCH_OPTIM_STATS)
3392 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir));
3396 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3399 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3403 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3408 #if defined(PMAP_DEBUG)
3409 pmap_pid_dump(int pid)
3415 LIST_FOREACH(p, &allproc, p_list) {
3416 if (p->p_pid != pid)
3422 pmap = vmspace_pmap(p->p_vmspace);
3423 for(i=0;i<1024;i++) {
3426 unsigned base = i << PDRSHIFT;
3428 pde = &pmap->pm_pdir[i];
3429 if (pde && pmap_pde_v(pde)) {
3430 for(j=0;j<1024;j++) {
3431 unsigned va = base + (j << PAGE_SHIFT);
3432 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3439 pte = pmap_pte_quick( pmap, va);
3440 if (pte && pmap_pte_v(pte)) {
3444 m = PHYS_TO_VM_PAGE(pa);
3445 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3446 va, pa, m->hold_count, m->wire_count, m->flags);
3467 static void pads __P((pmap_t pm));
3468 void pmap_pvdump __P((vm_offset_t pa));
3470 /* print address space of pmap*/
3478 if (pm == kernel_pmap)
3480 for (i = 0; i < 1024; i++)
3482 for (j = 0; j < 1024; j++) {
3483 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3484 if (pm == kernel_pmap && va < KERNBASE)
3486 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3488 ptep = pmap_pte_quick(pm, va);
3489 if (pmap_pte_v(ptep))
3490 printf("%x:%x ", va, *(int *) ptep);
3499 register pv_entry_t pv;
3502 printf("pa %x", pa);
3503 m = PHYS_TO_VM_PAGE(pa);
3504 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3506 printf(" -> pmap %p, va %x, flags %x",
3507 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3509 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);