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 1.2 2003/06/17 04:28:35 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;
425 if (cpu_apic_address == 0)
426 panic("pmap_bootstrap: no local apic!");
428 /* local apic is mapped on last page */
429 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag |
430 (cpu_apic_address & PG_FRAME));
432 /* BSP does this itself, AP's get it pre-set */
433 gd = &SMP_prvspace[0].globaldata;
434 gd->gd_prv_CMAP1 = &SMPpt[1];
435 gd->gd_prv_CMAP2 = &SMPpt[2];
436 gd->gd_prv_CMAP3 = &SMPpt[3];
437 gd->gd_prv_PMAP1 = &SMPpt[4];
438 gd->gd_prv_CADDR1 = SMP_prvspace[0].CPAGE1;
439 gd->gd_prv_CADDR2 = SMP_prvspace[0].CPAGE2;
440 gd->gd_prv_CADDR3 = SMP_prvspace[0].CPAGE3;
441 gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[0].PPAGE1;
449 * Set 4mb pdir for mp startup
454 if (pseflag && (cpu_feature & CPUID_PSE)) {
455 load_cr4(rcr4() | CR4_PSE);
456 if (pdir4mb && cpuid == 0) { /* only on BSP */
457 kernel_pmap->pm_pdir[KPTDI] =
458 PTD[KPTDI] = (pd_entry_t)pdir4mb;
466 * Initialize the pmap module.
467 * Called by vm_init, to initialize any structures that the pmap
468 * system needs to map virtual memory.
469 * pmap_init has been enhanced to support in a fairly consistant
470 * way, discontiguous physical memory.
473 pmap_init(phys_start, phys_end)
474 vm_offset_t phys_start, phys_end;
480 * object for kernel page table pages
482 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
485 * Allocate memory for random pmap data structures. Includes the
489 for(i = 0; i < vm_page_array_size; i++) {
492 m = &vm_page_array[i];
493 TAILQ_INIT(&m->md.pv_list);
494 m->md.pv_list_count = 0;
498 * init the pv free list
500 initial_pvs = vm_page_array_size;
501 if (initial_pvs < MINPV)
503 pvzone = &pvzone_store;
504 pvinit = (struct pv_entry *) kmem_alloc(kernel_map,
505 initial_pvs * sizeof (struct pv_entry));
506 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit,
510 * Now it is safe to enable pv_table recording.
512 pmap_initialized = TRUE;
516 * Initialize the address space (zone) for the pv_entries. Set a
517 * high water mark so that the system can recover from excessive
518 * numbers of pv entries.
523 int shpgperproc = PMAP_SHPGPERPROC;
525 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
526 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
527 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
528 pv_entry_high_water = 9 * (pv_entry_max / 10);
529 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
533 /***************************************************
534 * Low level helper routines.....
535 ***************************************************/
537 #if defined(PMAP_DIAGNOSTIC)
540 * This code checks for non-writeable/modified pages.
541 * This should be an invalid condition.
544 pmap_nw_modified(pt_entry_t ptea)
550 if ((pte & (PG_M|PG_RW)) == PG_M)
559 * this routine defines the region(s) of memory that should
560 * not be tested for the modified bit.
562 static PMAP_INLINE int
563 pmap_track_modified(vm_offset_t va)
565 if ((va < clean_sva) || (va >= clean_eva))
571 static PMAP_INLINE void
572 invltlb_1pg(vm_offset_t va)
574 #if defined(I386_CPU)
575 if (cpu_class == CPUCLASS_386) {
585 pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va)
588 if (pmap->pm_active & (1 << cpuid))
589 cpu_invlpg((void *)va);
590 if (pmap->pm_active & other_cpus)
599 pmap_TLB_invalidate_all(pmap_t pmap)
602 if (pmap->pm_active & (1 << cpuid))
604 if (pmap->pm_active & other_cpus)
616 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
618 /* are we current address space or kernel? */
619 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
620 return (unsigned *) PTmap;
622 /* otherwise, we are alternate address space */
623 if (frame != (((unsigned) APTDpde) & PG_FRAME)) {
624 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V);
626 /* The page directory is not shared between CPUs */
632 return (unsigned *) APTmap;
636 * Super fast pmap_pte routine best used when scanning
637 * the pv lists. This eliminates many coarse-grained
638 * invltlb calls. Note that many of the pv list
639 * scans are across different pmaps. It is very wasteful
640 * to do an entire invltlb for checking a single mapping.
644 pmap_pte_quick(pmap, va)
645 register pmap_t pmap;
649 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
650 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
651 unsigned index = i386_btop(va);
652 /* are we current address space or kernel? */
653 if ((pmap == kernel_pmap) ||
654 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
655 return (unsigned *) PTmap + index;
657 newpf = pde & PG_FRAME;
659 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) {
660 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V;
661 cpu_invlpg(prv_PADDR1);
663 return prv_PADDR1 + ((unsigned) index & (NPTEPG - 1));
665 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) {
666 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V;
667 invltlb_1pg((vm_offset_t) PADDR1);
669 return PADDR1 + ((unsigned) index & (NPTEPG - 1));
676 * Routine: pmap_extract
678 * Extract the physical page address associated
679 * with the given map/virtual_address pair.
682 pmap_extract(pmap, va)
683 register pmap_t pmap;
687 vm_offset_t pdirindex;
688 pdirindex = va >> PDRSHIFT;
689 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
691 if ((rtval & PG_PS) != 0) {
692 rtval &= ~(NBPDR - 1);
693 rtval |= va & (NBPDR - 1);
696 pte = get_ptbase(pmap) + i386_btop(va);
697 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
704 /***************************************************
705 * Low level mapping routines.....
706 ***************************************************/
709 * add a wired page to the kva
710 * note that in order for the mapping to take effect -- you
711 * should do a invltlb after doing the pmap_kenter...
716 register vm_offset_t pa;
718 register unsigned *pte;
721 npte = pa | PG_RW | PG_V | pgeflag;
722 pte = (unsigned *)vtopte(va);
729 * remove a page from the kernel pagetables
735 register unsigned *pte;
737 pte = (unsigned *)vtopte(va);
743 * Used to map a range of physical addresses into kernel
744 * virtual address space.
746 * For now, VM is already on, we only need to map the
750 pmap_map(virt, start, end, prot)
756 while (start < end) {
757 pmap_kenter(virt, start);
766 * Add a list of wired pages to the kva
767 * this routine is only used for temporary
768 * kernel mappings that do not need to have
769 * page modification or references recorded.
770 * Note that old mappings are simply written
771 * over. The page *must* be wired.
774 pmap_qenter(va, m, count)
781 end_va = va + count * PAGE_SIZE;
783 while (va < end_va) {
786 pte = (unsigned *)vtopte(va);
787 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
789 cpu_invlpg((void *)va);
802 * this routine jerks page mappings from the
803 * kernel -- it is meant only for temporary mappings.
806 pmap_qremove(va, count)
812 end_va = va + count*PAGE_SIZE;
814 while (va < end_va) {
817 pte = (unsigned *)vtopte(va);
820 cpu_invlpg((void *)va);
832 pmap_page_lookup(object, pindex)
838 m = vm_page_lookup(object, pindex);
839 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
845 * Create the UPAGES for a new process.
846 * This routine directly affects the fork perf for a process.
856 unsigned *ptek, oldpte;
859 * allocate object for the upages
861 if ((upobj = p->p_upages_obj) == NULL) {
862 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES);
863 p->p_upages_obj = upobj;
866 /* get a kernel virtual address for the UPAGES for this proc */
867 if ((up = p->p_addr) == NULL) {
868 up = (struct user *) kmem_alloc_nofault(kernel_map,
871 panic("pmap_new_proc: u_map allocation failed");
875 ptek = (unsigned *) vtopte((vm_offset_t) up);
878 for(i=0;i<UPAGES;i++) {
880 * Get a kernel stack page
882 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
890 oldpte = *(ptek + i);
892 * Enter the page into the kernel address space.
894 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag;
896 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) {
897 invlpg((vm_offset_t) up + i * PAGE_SIZE);
904 vm_page_flag_clear(m, PG_ZERO);
905 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
906 m->valid = VM_PAGE_BITS_ALL;
913 * Dispose the UPAGES for a process that has exited.
914 * This routine directly impacts the exit perf of a process.
923 unsigned *ptek, oldpte;
925 upobj = p->p_upages_obj;
927 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr);
928 for(i=0;i<UPAGES;i++) {
930 if ((m = vm_page_lookup(upobj, i)) == NULL)
931 panic("pmap_dispose_proc: upage already missing???");
935 oldpte = *(ptek + i);
937 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386))
938 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE);
939 vm_page_unwire(m, 0);
942 #if defined(I386_CPU)
943 if (cpu_class <= CPUCLASS_386)
948 * If the process got swapped out some of its UPAGES might have gotten
949 * swapped. Just get rid of the object to clean up the swap use
950 * proactively. NOTE! might block waiting for paging I/O to complete.
952 if (upobj->type == OBJT_SWAP) {
953 p->p_upages_obj = NULL;
954 vm_object_deallocate(upobj);
959 * Allow the UPAGES for a process to be prejudicially paged out.
969 upobj = p->p_upages_obj;
971 * let the upages be paged
973 for(i=0;i<UPAGES;i++) {
974 if ((m = vm_page_lookup(upobj, i)) == NULL)
975 panic("pmap_swapout_proc: upage already missing???");
977 vm_page_unwire(m, 0);
978 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
983 * Bring the UPAGES for a specified process back in.
993 upobj = p->p_upages_obj;
994 for(i=0;i<UPAGES;i++) {
996 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
998 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
1001 if (m->valid != VM_PAGE_BITS_ALL) {
1002 rv = vm_pager_get_pages(upobj, &m, 1, 0);
1003 if (rv != VM_PAGER_OK)
1004 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
1005 m = vm_page_lookup(upobj, i);
1006 m->valid = VM_PAGE_BITS_ALL;
1011 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
1015 /***************************************************
1016 * Page table page management routines.....
1017 ***************************************************/
1020 * This routine unholds page table pages, and if the hold count
1021 * drops to zero, then it decrements the wire count.
1024 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
1026 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
1029 if (m->hold_count == 0) {
1032 * unmap the page table page
1034 pmap->pm_pdir[m->pindex] = 0;
1035 --pmap->pm_stats.resident_count;
1036 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1037 (((unsigned) PTDpde) & PG_FRAME)) {
1039 * Do a invltlb to make the invalidated mapping
1040 * take effect immediately.
1042 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
1043 pmap_TLB_invalidate(pmap, pteva);
1046 if (pmap->pm_ptphint == m)
1047 pmap->pm_ptphint = NULL;
1050 * If the page is finally unwired, simply free it.
1053 if (m->wire_count == 0) {
1057 vm_page_free_zero(m);
1065 static PMAP_INLINE int
1066 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
1069 if (m->hold_count == 0)
1070 return _pmap_unwire_pte_hold(pmap, m);
1076 * After removing a page table entry, this routine is used to
1077 * conditionally free the page, and manage the hold/wire counts.
1080 pmap_unuse_pt(pmap, va, mpte)
1086 if (va >= UPT_MIN_ADDRESS)
1090 ptepindex = (va >> PDRSHIFT);
1091 if (pmap->pm_ptphint &&
1092 (pmap->pm_ptphint->pindex == ptepindex)) {
1093 mpte = pmap->pm_ptphint;
1095 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1096 pmap->pm_ptphint = mpte;
1100 return pmap_unwire_pte_hold(pmap, mpte);
1108 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1109 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1111 pmap->pm_active = 0;
1112 pmap->pm_ptphint = NULL;
1113 TAILQ_INIT(&pmap->pm_pvlist);
1114 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1118 * Initialize a preallocated and zeroed pmap structure,
1119 * such as one in a vmspace structure.
1123 register struct pmap *pmap;
1128 * No need to allocate page table space yet but we do need a valid
1129 * page directory table.
1131 if (pmap->pm_pdir == NULL)
1133 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1136 * allocate object for the ptes
1138 if (pmap->pm_pteobj == NULL)
1139 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1142 * allocate the page directory page
1144 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1145 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1147 ptdpg->wire_count = 1;
1151 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1152 ptdpg->valid = VM_PAGE_BITS_ALL;
1154 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1155 if ((ptdpg->flags & PG_ZERO) == 0)
1156 bzero(pmap->pm_pdir, PAGE_SIZE);
1159 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1162 /* install self-referential address mapping entry */
1163 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1164 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1167 pmap->pm_active = 0;
1168 pmap->pm_ptphint = NULL;
1169 TAILQ_INIT(&pmap->pm_pvlist);
1170 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1174 * Wire in kernel global address entries. To avoid a race condition
1175 * between pmap initialization and pmap_growkernel, this procedure
1176 * should be called after the vmspace is attached to the process
1177 * but before this pmap is activated.
1183 /* XXX copies current process, does not fill in MPPTDI */
1184 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1188 pmap_release_free_page(pmap, p)
1192 unsigned *pde = (unsigned *) pmap->pm_pdir;
1194 * This code optimizes the case of freeing non-busy
1195 * page-table pages. Those pages are zero now, and
1196 * might as well be placed directly into the zero queue.
1198 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1204 * Remove the page table page from the processes address space.
1207 pmap->pm_stats.resident_count--;
1209 if (p->hold_count) {
1210 panic("pmap_release: freeing held page table page");
1213 * Page directory pages need to have the kernel
1214 * stuff cleared, so they can go into the zero queue also.
1216 if (p->pindex == PTDPTDI) {
1217 bzero(pde + KPTDI, nkpt * PTESIZE);
1222 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1225 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1226 pmap->pm_ptphint = NULL;
1230 vm_page_free_zero(p);
1235 * this routine is called if the page table page is not
1239 _pmap_allocpte(pmap, ptepindex)
1243 vm_offset_t pteva, ptepa;
1247 * Find or fabricate a new pagetable page
1249 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1250 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1252 KASSERT(m->queue == PQ_NONE,
1253 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1255 if (m->wire_count == 0)
1260 * Increment the hold count for the page table page
1261 * (denoting a new mapping.)
1266 * Map the pagetable page into the process address space, if
1267 * it isn't already there.
1270 pmap->pm_stats.resident_count++;
1272 ptepa = VM_PAGE_TO_PHYS(m);
1273 pmap->pm_pdir[ptepindex] =
1274 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1277 * Set the page table hint
1279 pmap->pm_ptphint = m;
1282 * Try to use the new mapping, but if we cannot, then
1283 * do it with the routine that maps the page explicitly.
1285 if ((m->flags & PG_ZERO) == 0) {
1286 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1287 (((unsigned) PTDpde) & PG_FRAME)) {
1288 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1289 bzero((caddr_t) pteva, PAGE_SIZE);
1291 pmap_zero_page(ptepa);
1295 m->valid = VM_PAGE_BITS_ALL;
1296 vm_page_flag_clear(m, PG_ZERO);
1297 vm_page_flag_set(m, PG_MAPPED);
1304 pmap_allocpte(pmap, va)
1313 * Calculate pagetable page index
1315 ptepindex = va >> PDRSHIFT;
1318 * Get the page directory entry
1320 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1323 * This supports switching from a 4MB page to a
1326 if (ptepa & PG_PS) {
1327 pmap->pm_pdir[ptepindex] = 0;
1333 * If the page table page is mapped, we just increment the
1334 * hold count, and activate it.
1338 * In order to get the page table page, try the
1341 if (pmap->pm_ptphint &&
1342 (pmap->pm_ptphint->pindex == ptepindex)) {
1343 m = pmap->pm_ptphint;
1345 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1346 pmap->pm_ptphint = m;
1352 * Here if the pte page isn't mapped, or if it has been deallocated.
1354 return _pmap_allocpte(pmap, ptepindex);
1358 /***************************************************
1359 * Pmap allocation/deallocation routines.
1360 ***************************************************/
1363 * Release any resources held by the given physical map.
1364 * Called when a pmap initialized by pmap_pinit is being released.
1365 * Should only be called if the map contains no valid mappings.
1369 register struct pmap *pmap;
1371 vm_page_t p,n,ptdpg;
1372 vm_object_t object = pmap->pm_pteobj;
1375 #if defined(DIAGNOSTIC)
1376 if (object->ref_count != 1)
1377 panic("pmap_release: pteobj reference count != 1");
1382 curgeneration = object->generation;
1383 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1384 n = TAILQ_NEXT(p, listq);
1385 if (p->pindex == PTDPTDI) {
1390 if (!pmap_release_free_page(pmap, p) &&
1391 (object->generation != curgeneration))
1396 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1401 kvm_size(SYSCTL_HANDLER_ARGS)
1403 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1405 return sysctl_handle_long(oidp, &ksize, 0, req);
1407 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1408 0, 0, kvm_size, "IU", "Size of KVM");
1411 kvm_free(SYSCTL_HANDLER_ARGS)
1413 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1415 return sysctl_handle_long(oidp, &kfree, 0, req);
1417 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1418 0, 0, kvm_free, "IU", "Amount of KVM free");
1421 * grow the number of kernel page table entries, if needed
1424 pmap_growkernel(vm_offset_t addr)
1429 vm_offset_t ptppaddr;
1434 if (kernel_vm_end == 0) {
1435 kernel_vm_end = KERNBASE;
1437 while (pdir_pde(PTD, kernel_vm_end)) {
1438 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1442 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1443 while (kernel_vm_end < addr) {
1444 if (pdir_pde(PTD, kernel_vm_end)) {
1445 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1450 * This index is bogus, but out of the way
1452 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1454 panic("pmap_growkernel: no memory to grow kernel");
1459 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1460 pmap_zero_page(ptppaddr);
1461 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1462 pdir_pde(PTD, kernel_vm_end) = newpdir;
1464 LIST_FOREACH(p, &allproc, p_list) {
1466 pmap = vmspace_pmap(p->p_vmspace);
1467 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1470 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1471 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1477 * Retire the given physical map from service.
1478 * Should only be called if the map contains
1479 * no valid mappings.
1483 register pmap_t pmap;
1490 count = --pmap->pm_count;
1493 panic("destroying a pmap is not yet implemented");
1498 * Add a reference to the specified pmap.
1501 pmap_reference(pmap)
1509 /***************************************************
1510 * page management routines.
1511 ***************************************************/
1514 * free the pv_entry back to the free list
1516 static PMAP_INLINE void
1525 * get a new pv_entry, allocating a block from the system
1527 * the memory allocation is performed bypassing the malloc code
1528 * because of the possibility of allocations at interrupt time.
1534 if (pv_entry_high_water &&
1535 (pv_entry_count > pv_entry_high_water) &&
1536 (pmap_pagedaemon_waken == 0)) {
1537 pmap_pagedaemon_waken = 1;
1538 wakeup (&vm_pages_needed);
1540 return zalloci(pvzone);
1544 * This routine is very drastic, but can save the system
1552 static int warningdone=0;
1554 if (pmap_pagedaemon_waken == 0)
1557 if (warningdone < 5) {
1558 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1562 for(i = 0; i < vm_page_array_size; i++) {
1563 m = &vm_page_array[i];
1564 if (m->wire_count || m->hold_count || m->busy ||
1565 (m->flags & PG_BUSY))
1569 pmap_pagedaemon_waken = 0;
1574 * If it is the first entry on the list, it is actually
1575 * in the header and we must copy the following entry up
1576 * to the header. Otherwise we must search the list for
1577 * the entry. In either case we free the now unused entry.
1581 pmap_remove_entry(pmap, m, va)
1591 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1592 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1593 if (pmap == pv->pv_pmap && va == pv->pv_va)
1597 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1598 if (va == pv->pv_va)
1606 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1607 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1608 m->md.pv_list_count--;
1609 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1610 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1612 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1621 * Create a pv entry for page at pa for
1625 pmap_insert_entry(pmap, va, mpte, m)
1636 pv = get_pv_entry();
1641 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1642 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1643 m->md.pv_list_count++;
1649 * pmap_remove_pte: do the things to unmap a page in a process
1652 pmap_remove_pte(pmap, ptq, va)
1660 oldpte = loadandclear(ptq);
1662 pmap->pm_stats.wired_count -= 1;
1664 * Machines that don't support invlpg, also don't support
1669 pmap->pm_stats.resident_count -= 1;
1670 if (oldpte & PG_MANAGED) {
1671 m = PHYS_TO_VM_PAGE(oldpte);
1672 if (oldpte & PG_M) {
1673 #if defined(PMAP_DIAGNOSTIC)
1674 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1676 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1680 if (pmap_track_modified(va))
1684 vm_page_flag_set(m, PG_REFERENCED);
1685 return pmap_remove_entry(pmap, m, va);
1687 return pmap_unuse_pt(pmap, va, NULL);
1694 * Remove a single page from a process address space
1697 pmap_remove_page(pmap, va)
1699 register vm_offset_t va;
1701 register unsigned *ptq;
1704 * if there is no pte for this address, just skip it!!!
1706 if (*pmap_pde(pmap, va) == 0) {
1711 * get a local va for mappings for this pmap.
1713 ptq = get_ptbase(pmap) + i386_btop(va);
1715 (void) pmap_remove_pte(pmap, ptq, va);
1716 pmap_TLB_invalidate(pmap, va);
1722 * Remove the given range of addresses from the specified map.
1724 * It is assumed that the start and end are properly
1725 * rounded to the page size.
1728 pmap_remove(pmap, sva, eva)
1730 register vm_offset_t sva;
1731 register vm_offset_t eva;
1733 register unsigned *ptbase;
1735 vm_offset_t ptpaddr;
1736 vm_offset_t sindex, eindex;
1742 if (pmap->pm_stats.resident_count == 0)
1746 * special handling of removing one page. a very
1747 * common operation and easy to short circuit some
1750 if (((sva + PAGE_SIZE) == eva) &&
1751 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1752 pmap_remove_page(pmap, sva);
1759 * Get a local virtual address for the mappings that are being
1762 ptbase = get_ptbase(pmap);
1764 sindex = i386_btop(sva);
1765 eindex = i386_btop(eva);
1767 for (; sindex < eindex; sindex = pdnxt) {
1771 * Calculate index for next page table.
1773 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1774 if (pmap->pm_stats.resident_count == 0)
1777 pdirindex = sindex / NPDEPG;
1778 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1779 pmap->pm_pdir[pdirindex] = 0;
1780 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1786 * Weed out invalid mappings. Note: we assume that the page
1787 * directory table is always allocated, and in kernel virtual.
1793 * Limit our scan to either the end of the va represented
1794 * by the current page table page, or to the end of the
1795 * range being removed.
1797 if (pdnxt > eindex) {
1801 for ( ;sindex != pdnxt; sindex++) {
1803 if (ptbase[sindex] == 0) {
1806 va = i386_ptob(sindex);
1809 if (pmap_remove_pte(pmap,
1810 ptbase + sindex, va))
1816 pmap_TLB_invalidate_all(pmap);
1820 * Routine: pmap_remove_all
1822 * Removes this physical page from
1823 * all physical maps in which it resides.
1824 * Reflects back modify bits to the pager.
1827 * Original versions of this routine were very
1828 * inefficient because they iteratively called
1829 * pmap_remove (slow...)
1836 register pv_entry_t pv;
1837 register unsigned *pte, tpte;
1840 #if defined(PMAP_DIAGNOSTIC)
1842 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1845 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1846 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1851 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1852 pv->pv_pmap->pm_stats.resident_count--;
1854 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1856 tpte = loadandclear(pte);
1858 pv->pv_pmap->pm_stats.wired_count--;
1861 vm_page_flag_set(m, PG_REFERENCED);
1864 * Update the vm_page_t clean and reference bits.
1867 #if defined(PMAP_DIAGNOSTIC)
1868 if (pmap_nw_modified((pt_entry_t) tpte)) {
1870 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1874 if (pmap_track_modified(pv->pv_va))
1877 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1879 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1880 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1881 m->md.pv_list_count--;
1882 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1886 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1892 * Set the physical protection on the
1893 * specified range of this map as requested.
1896 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1898 register unsigned *ptbase;
1899 vm_offset_t pdnxt, ptpaddr;
1900 vm_pindex_t sindex, eindex;
1906 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1907 pmap_remove(pmap, sva, eva);
1911 if (prot & VM_PROT_WRITE)
1916 ptbase = get_ptbase(pmap);
1918 sindex = i386_btop(sva);
1919 eindex = i386_btop(eva);
1921 for (; sindex < eindex; sindex = pdnxt) {
1925 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1927 pdirindex = sindex / NPDEPG;
1928 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1929 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1930 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1936 * Weed out invalid mappings. Note: we assume that the page
1937 * directory table is always allocated, and in kernel virtual.
1942 if (pdnxt > eindex) {
1946 for (; sindex != pdnxt; sindex++) {
1951 pbits = ptbase[sindex];
1953 if (pbits & PG_MANAGED) {
1956 m = PHYS_TO_VM_PAGE(pbits);
1957 vm_page_flag_set(m, PG_REFERENCED);
1961 if (pmap_track_modified(i386_ptob(sindex))) {
1963 m = PHYS_TO_VM_PAGE(pbits);
1972 if (pbits != ptbase[sindex]) {
1973 ptbase[sindex] = pbits;
1979 pmap_TLB_invalidate_all(pmap);
1983 * Insert the given physical page (p) at
1984 * the specified virtual address (v) in the
1985 * target physical map with the protection requested.
1987 * If specified, the page will be wired down, meaning
1988 * that the related pte can not be reclaimed.
1990 * NB: This is the only routine which MAY NOT lazy-evaluate
1991 * or lose information. That is, this routine must actually
1992 * insert this page into the given map NOW.
1995 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1999 register unsigned *pte;
2001 vm_offset_t origpte, newpte;
2008 #ifdef PMAP_DIAGNOSTIC
2009 if (va > VM_MAX_KERNEL_ADDRESS)
2010 panic("pmap_enter: toobig");
2011 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
2012 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
2017 * In the case that a page table page is not
2018 * resident, we are creating it here.
2020 if (va < UPT_MIN_ADDRESS) {
2021 mpte = pmap_allocpte(pmap, va);
2023 #if 0 && defined(PMAP_DIAGNOSTIC)
2025 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
2026 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
2027 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
2028 pmap->pm_pdir[PTDPTDI], origpte, va);
2031 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
2032 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
2033 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
2034 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
2035 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
2036 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
2037 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
2043 pte = pmap_pte(pmap, va);
2046 * Page Directory table entry not valid, we need a new PT page
2049 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
2050 (void *)pmap->pm_pdir[PTDPTDI], va);
2053 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2054 origpte = *(vm_offset_t *)pte;
2055 opa = origpte & PG_FRAME;
2057 if (origpte & PG_PS)
2058 panic("pmap_enter: attempted pmap_enter on 4MB page");
2061 * Mapping has not changed, must be protection or wiring change.
2063 if (origpte && (opa == pa)) {
2065 * Wiring change, just update stats. We don't worry about
2066 * wiring PT pages as they remain resident as long as there
2067 * are valid mappings in them. Hence, if a user page is wired,
2068 * the PT page will be also.
2070 if (wired && ((origpte & PG_W) == 0))
2071 pmap->pm_stats.wired_count++;
2072 else if (!wired && (origpte & PG_W))
2073 pmap->pm_stats.wired_count--;
2075 #if defined(PMAP_DIAGNOSTIC)
2076 if (pmap_nw_modified((pt_entry_t) origpte)) {
2078 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
2084 * Remove extra pte reference
2089 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
2090 if ((origpte & PG_RW) == 0) {
2093 cpu_invlpg((void *)va);
2094 if (pmap->pm_active & other_cpus)
2104 * We might be turning off write access to the page,
2105 * so we go ahead and sense modify status.
2107 if (origpte & PG_MANAGED) {
2108 if ((origpte & PG_M) && pmap_track_modified(va)) {
2110 om = PHYS_TO_VM_PAGE(opa);
2118 * Mapping has changed, invalidate old range and fall through to
2119 * handle validating new mapping.
2123 err = pmap_remove_pte(pmap, pte, va);
2125 panic("pmap_enter: pte vanished, va: 0x%x", va);
2129 * Enter on the PV list if part of our managed memory. Note that we
2130 * raise IPL while manipulating pv_table since pmap_enter can be
2131 * called at interrupt time.
2133 if (pmap_initialized &&
2134 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2135 pmap_insert_entry(pmap, va, mpte, m);
2140 * Increment counters
2142 pmap->pm_stats.resident_count++;
2144 pmap->pm_stats.wired_count++;
2148 * Now validate mapping with desired protection/wiring.
2150 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2154 if (va < UPT_MIN_ADDRESS)
2156 if (pmap == kernel_pmap)
2160 * if the mapping or permission bits are different, we need
2161 * to update the pte.
2163 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2164 *pte = newpte | PG_A;
2167 cpu_invlpg((void *)va);
2168 if (pmap->pm_active & other_cpus)
2178 * this code makes some *MAJOR* assumptions:
2179 * 1. Current pmap & pmap exists.
2182 * 4. No page table pages.
2183 * 5. Tlbflush is deferred to calling procedure.
2184 * 6. Page IS managed.
2185 * but is *MUCH* faster than pmap_enter...
2189 pmap_enter_quick(pmap, va, m, mpte)
2190 register pmap_t pmap;
2199 * In the case that a page table page is not
2200 * resident, we are creating it here.
2202 if (va < UPT_MIN_ADDRESS) {
2207 * Calculate pagetable page index
2209 ptepindex = va >> PDRSHIFT;
2210 if (mpte && (mpte->pindex == ptepindex)) {
2215 * Get the page directory entry
2217 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2220 * If the page table page is mapped, we just increment
2221 * the hold count, and activate it.
2225 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2226 if (pmap->pm_ptphint &&
2227 (pmap->pm_ptphint->pindex == ptepindex)) {
2228 mpte = pmap->pm_ptphint;
2230 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2231 pmap->pm_ptphint = mpte;
2237 mpte = _pmap_allocpte(pmap, ptepindex);
2245 * This call to vtopte makes the assumption that we are
2246 * entering the page into the current pmap. In order to support
2247 * quick entry into any pmap, one would likely use pmap_pte_quick.
2248 * But that isn't as quick as vtopte.
2250 pte = (unsigned *)vtopte(va);
2253 pmap_unwire_pte_hold(pmap, mpte);
2258 * Enter on the PV list if part of our managed memory. Note that we
2259 * raise IPL while manipulating pv_table since pmap_enter can be
2260 * called at interrupt time.
2262 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2263 pmap_insert_entry(pmap, va, mpte, m);
2266 * Increment counters
2268 pmap->pm_stats.resident_count++;
2270 pa = VM_PAGE_TO_PHYS(m);
2273 * Now validate mapping with RO protection
2275 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2276 *pte = pa | PG_V | PG_U;
2278 *pte = pa | PG_V | PG_U | PG_MANAGED;
2284 * Make a temporary mapping for a physical address. This is only intended
2285 * to be used for panic dumps.
2288 pmap_kenter_temporary(vm_offset_t pa, int i)
2290 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2291 return ((void *)crashdumpmap);
2294 #define MAX_INIT_PT (96)
2296 * pmap_object_init_pt preloads the ptes for a given object
2297 * into the specified pmap. This eliminates the blast of soft
2298 * faults on process startup and immediately after an mmap.
2301 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2314 if (pmap == NULL || object == NULL)
2318 * This code maps large physical mmap regions into the
2319 * processor address space. Note that some shortcuts
2320 * are taken, but the code works.
2323 (object->type == OBJT_DEVICE) &&
2324 ((addr & (NBPDR - 1)) == 0) &&
2325 ((size & (NBPDR - 1)) == 0) ) {
2328 unsigned int ptepindex;
2332 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2336 p = vm_page_lookup(object, pindex);
2337 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2341 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2346 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2351 p = vm_page_lookup(object, pindex);
2355 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2356 if (ptepa & (NBPDR - 1)) {
2360 p->valid = VM_PAGE_BITS_ALL;
2362 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2363 npdes = size >> PDRSHIFT;
2364 for(i=0;i<npdes;i++) {
2365 pmap->pm_pdir[ptepindex] =
2366 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2370 vm_page_flag_set(p, PG_MAPPED);
2375 psize = i386_btop(size);
2377 if ((object->type != OBJT_VNODE) ||
2378 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2379 (object->resident_page_count > MAX_INIT_PT))) {
2383 if (psize + pindex > object->size) {
2384 if (object->size < pindex)
2386 psize = object->size - pindex;
2391 * if we are processing a major portion of the object, then scan the
2394 if (psize > (object->resident_page_count >> 2)) {
2397 for (p = TAILQ_FIRST(&object->memq);
2398 ((objpgs > 0) && (p != NULL));
2399 p = TAILQ_NEXT(p, listq)) {
2402 if (tmpidx < pindex) {
2406 if (tmpidx >= psize) {
2410 * don't allow an madvise to blow away our really
2411 * free pages allocating pv entries.
2413 if ((limit & MAP_PREFAULT_MADVISE) &&
2414 cnt.v_free_count < cnt.v_free_reserved) {
2417 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2419 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2420 if ((p->queue - p->pc) == PQ_CACHE)
2421 vm_page_deactivate(p);
2423 mpte = pmap_enter_quick(pmap,
2424 addr + i386_ptob(tmpidx), p, mpte);
2425 vm_page_flag_set(p, PG_MAPPED);
2432 * else lookup the pages one-by-one.
2434 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
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 p = vm_page_lookup(object, tmpidx + pindex);
2445 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2447 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2448 if ((p->queue - p->pc) == PQ_CACHE)
2449 vm_page_deactivate(p);
2451 mpte = pmap_enter_quick(pmap,
2452 addr + i386_ptob(tmpidx), p, mpte);
2453 vm_page_flag_set(p, PG_MAPPED);
2462 * pmap_prefault provides a quick way of clustering
2463 * pagefaults into a processes address space. It is a "cousin"
2464 * of pmap_object_init_pt, except it runs at page fault time instead
2469 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2471 static int pmap_prefault_pageorder[] = {
2472 -PAGE_SIZE, PAGE_SIZE,
2473 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2474 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2475 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2479 pmap_prefault(pmap, addra, entry)
2482 vm_map_entry_t entry;
2491 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2494 object = entry->object.vm_object;
2496 starta = addra - PFBAK * PAGE_SIZE;
2497 if (starta < entry->start) {
2498 starta = entry->start;
2499 } else if (starta > addra) {
2504 for (i = 0; i < PAGEORDER_SIZE; i++) {
2505 vm_object_t lobject;
2508 addr = addra + pmap_prefault_pageorder[i];
2509 if (addr > addra + (PFFOR * PAGE_SIZE))
2512 if (addr < starta || addr >= entry->end)
2515 if ((*pmap_pde(pmap, addr)) == NULL)
2518 pte = (unsigned *) vtopte(addr);
2522 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2524 for (m = vm_page_lookup(lobject, pindex);
2525 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2526 lobject = lobject->backing_object) {
2527 if (lobject->backing_object_offset & PAGE_MASK)
2529 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2530 m = vm_page_lookup(lobject->backing_object, pindex);
2534 * give-up when a page is not in memory
2539 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2541 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2543 if ((m->queue - m->pc) == PQ_CACHE) {
2544 vm_page_deactivate(m);
2547 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2548 vm_page_flag_set(m, PG_MAPPED);
2555 * Routine: pmap_change_wiring
2556 * Function: Change the wiring attribute for a map/virtual-address
2558 * In/out conditions:
2559 * The mapping must already exist in the pmap.
2562 pmap_change_wiring(pmap, va, wired)
2563 register pmap_t pmap;
2567 register unsigned *pte;
2572 pte = pmap_pte(pmap, va);
2574 if (wired && !pmap_pte_w(pte))
2575 pmap->pm_stats.wired_count++;
2576 else if (!wired && pmap_pte_w(pte))
2577 pmap->pm_stats.wired_count--;
2580 * Wiring is not a hardware characteristic so there is no need to
2583 pmap_pte_set_w(pte, wired);
2589 * Copy the range specified by src_addr/len
2590 * from the source map to the range dst_addr/len
2591 * in the destination map.
2593 * This routine is only advisory and need not do anything.
2597 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2598 pmap_t dst_pmap, src_pmap;
2599 vm_offset_t dst_addr;
2601 vm_offset_t src_addr;
2604 vm_offset_t end_addr = src_addr + len;
2606 unsigned src_frame, dst_frame;
2609 if (dst_addr != src_addr)
2612 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2613 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2617 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2618 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2619 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2621 /* The page directory is not shared between CPUs */
2628 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2629 unsigned *src_pte, *dst_pte;
2630 vm_page_t dstmpte, srcmpte;
2631 vm_offset_t srcptepaddr;
2634 if (addr >= UPT_MIN_ADDRESS)
2635 panic("pmap_copy: invalid to pmap_copy page tables\n");
2638 * Don't let optional prefaulting of pages make us go
2639 * way below the low water mark of free pages or way
2640 * above high water mark of used pv entries.
2642 if (cnt.v_free_count < cnt.v_free_reserved ||
2643 pv_entry_count > pv_entry_high_water)
2646 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2647 ptepindex = addr >> PDRSHIFT;
2649 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2650 if (srcptepaddr == 0)
2653 if (srcptepaddr & PG_PS) {
2654 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2655 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2656 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2661 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2662 if ((srcmpte == NULL) ||
2663 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2666 if (pdnxt > end_addr)
2669 src_pte = (unsigned *) vtopte(addr);
2670 dst_pte = (unsigned *) avtopte(addr);
2671 while (addr < pdnxt) {
2675 * we only virtual copy managed pages
2677 if ((ptetemp & PG_MANAGED) != 0) {
2679 * We have to check after allocpte for the
2680 * pte still being around... allocpte can
2683 dstmpte = pmap_allocpte(dst_pmap, addr);
2684 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2686 * Clear the modified and
2687 * accessed (referenced) bits
2690 m = PHYS_TO_VM_PAGE(ptetemp);
2691 *dst_pte = ptetemp & ~(PG_M | PG_A);
2692 dst_pmap->pm_stats.resident_count++;
2693 pmap_insert_entry(dst_pmap, addr,
2696 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2698 if (dstmpte->hold_count >= srcmpte->hold_count)
2709 * Routine: pmap_kernel
2711 * Returns the physical map handle for the kernel.
2716 return (kernel_pmap);
2720 * pmap_zero_page zeros the specified hardware page by mapping
2721 * the page into KVM and using bzero to clear its contents.
2724 pmap_zero_page(phys)
2728 if (*(int *) prv_CMAP3)
2729 panic("pmap_zero_page: prv_CMAP3 busy");
2731 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2732 cpu_invlpg(prv_CADDR3);
2734 #if defined(I686_CPU)
2735 if (cpu_class == CPUCLASS_686)
2736 i686_pagezero(prv_CADDR3);
2739 bzero(prv_CADDR3, PAGE_SIZE);
2741 *(int *) prv_CMAP3 = 0;
2744 panic("pmap_zero_page: CMAP2 busy");
2746 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2747 invltlb_1pg((vm_offset_t)CADDR2);
2749 #if defined(I686_CPU)
2750 if (cpu_class == CPUCLASS_686)
2751 i686_pagezero(CADDR2);
2754 bzero(CADDR2, PAGE_SIZE);
2760 * pmap_zero_page_area zeros the specified hardware page by mapping
2761 * the page into KVM and using bzero to clear its contents.
2763 * off and size may not cover an area beyond a single hardware page.
2766 pmap_zero_page_area(phys, off, size)
2772 if (*(int *) prv_CMAP3)
2773 panic("pmap_zero_page: prv_CMAP3 busy");
2775 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2776 cpu_invlpg(prv_CADDR3);
2778 #if defined(I686_CPU)
2779 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2780 i686_pagezero(prv_CADDR3);
2783 bzero((char *)prv_CADDR3 + off, size);
2785 *(int *) prv_CMAP3 = 0;
2788 panic("pmap_zero_page: CMAP2 busy");
2790 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2791 invltlb_1pg((vm_offset_t)CADDR2);
2793 #if defined(I686_CPU)
2794 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2795 i686_pagezero(CADDR2);
2798 bzero((char *)CADDR2 + off, size);
2804 * pmap_copy_page copies the specified (machine independent)
2805 * page by mapping the page into virtual memory and using
2806 * bcopy to copy the page, one machine dependent page at a
2810 pmap_copy_page(src, dst)
2815 if (*(int *) prv_CMAP1)
2816 panic("pmap_copy_page: prv_CMAP1 busy");
2817 if (*(int *) prv_CMAP2)
2818 panic("pmap_copy_page: prv_CMAP2 busy");
2820 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2821 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2823 cpu_invlpg(prv_CADDR1);
2824 cpu_invlpg(prv_CADDR2);
2826 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE);
2828 *(int *) prv_CMAP1 = 0;
2829 *(int *) prv_CMAP2 = 0;
2831 if (*(int *) CMAP1 || *(int *) CMAP2)
2832 panic("pmap_copy_page: CMAP busy");
2834 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2835 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2836 #if defined(I386_CPU)
2837 if (cpu_class == CPUCLASS_386) {
2842 invlpg((u_int)CADDR1);
2843 invlpg((u_int)CADDR2);
2846 bcopy(CADDR1, CADDR2, PAGE_SIZE);
2855 * Routine: pmap_pageable
2857 * Make the specified pages (by pmap, offset)
2858 * pageable (or not) as requested.
2860 * A page which is not pageable may not take
2861 * a fault; therefore, its page table entry
2862 * must remain valid for the duration.
2864 * This routine is merely advisory; pmap_enter
2865 * will specify that these pages are to be wired
2866 * down (or not) as appropriate.
2869 pmap_pageable(pmap, sva, eva, pageable)
2871 vm_offset_t sva, eva;
2877 * Returns true if the pmap's pv is one of the first
2878 * 16 pvs linked to from this page. This count may
2879 * be changed upwards or downwards in the future; it
2880 * is only necessary that true be returned for a small
2881 * subset of pmaps for proper page aging.
2884 pmap_page_exists_quick(pmap, m)
2892 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2897 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2898 if (pv->pv_pmap == pmap) {
2910 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2912 * Remove all pages from specified address space
2913 * this aids process exit speeds. Also, this code
2914 * is special cased for current process only, but
2915 * can have the more generic (and slightly slower)
2916 * mode enabled. This is much faster than pmap_remove
2917 * in the case of running down an entire address space.
2920 pmap_remove_pages(pmap, sva, eva)
2922 vm_offset_t sva, eva;
2924 unsigned *pte, tpte;
2929 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2930 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2931 printf("warning: pmap_remove_pages called with non-current pmap\n");
2937 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2941 if (pv->pv_va >= eva || pv->pv_va < sva) {
2942 npv = TAILQ_NEXT(pv, pv_plist);
2946 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2947 pte = (unsigned *)vtopte(pv->pv_va);
2949 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2954 * We cannot remove wired pages from a process' mapping at this time
2957 npv = TAILQ_NEXT(pv, pv_plist);
2962 m = PHYS_TO_VM_PAGE(tpte);
2964 KASSERT(m < &vm_page_array[vm_page_array_size],
2965 ("pmap_remove_pages: bad tpte %x", tpte));
2967 pv->pv_pmap->pm_stats.resident_count--;
2970 * Update the vm_page_t clean and reference bits.
2977 npv = TAILQ_NEXT(pv, pv_plist);
2978 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2980 m->md.pv_list_count--;
2981 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2982 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2983 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2986 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
2990 pmap_TLB_invalidate_all(pmap);
2994 * pmap_testbit tests bits in pte's
2995 * note that the testbit/changebit routines are inline,
2996 * and a lot of things compile-time evaluate.
2999 pmap_testbit(m, bit)
3007 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3010 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3015 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3017 * if the bit being tested is the modified bit, then
3018 * mark clean_map and ptes as never
3021 if (bit & (PG_A|PG_M)) {
3022 if (!pmap_track_modified(pv->pv_va))
3026 #if defined(PMAP_DIAGNOSTIC)
3028 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
3032 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3043 * this routine is used to modify bits in ptes
3045 static __inline void
3046 pmap_changebit(m, bit, setem)
3051 register pv_entry_t pv;
3052 register unsigned *pte;
3055 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3061 * Loop over all current mappings setting/clearing as appropos If
3062 * setting RO do we need to clear the VAC?
3064 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3066 * don't write protect pager mappings
3068 if (!setem && (bit == PG_RW)) {
3069 if (!pmap_track_modified(pv->pv_va))
3073 #if defined(PMAP_DIAGNOSTIC)
3075 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
3080 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3084 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3086 vm_offset_t pbits = *(vm_offset_t *)pte;
3092 *(int *)pte = pbits & ~(PG_M|PG_RW);
3094 *(int *)pte = pbits & ~bit;
3096 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3104 * pmap_page_protect:
3106 * Lower the permission for all mappings to a given page.
3109 pmap_page_protect(vm_page_t m, vm_prot_t prot)
3111 if ((prot & VM_PROT_WRITE) == 0) {
3112 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3113 pmap_changebit(m, PG_RW, FALSE);
3121 pmap_phys_address(ppn)
3124 return (i386_ptob(ppn));
3128 * pmap_ts_referenced:
3130 * Return a count of reference bits for a page, clearing those bits.
3131 * It is not necessary for every reference bit to be cleared, but it
3132 * is necessary that 0 only be returned when there are truly no
3133 * reference bits set.
3135 * XXX: The exact number of bits to check and clear is a matter that
3136 * should be tested and standardized at some point in the future for
3137 * optimal aging of shared pages.
3140 pmap_ts_referenced(vm_page_t m)
3142 register pv_entry_t pv, pvf, pvn;
3147 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3152 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3157 pvn = TAILQ_NEXT(pv, pv_list);
3159 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3161 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3163 if (!pmap_track_modified(pv->pv_va))
3166 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3168 if (pte && (*pte & PG_A)) {
3171 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3178 } while ((pv = pvn) != NULL && pv != pvf);
3188 * Return whether or not the specified physical page was modified
3189 * in any physical maps.
3192 pmap_is_modified(vm_page_t m)
3194 return pmap_testbit(m, PG_M);
3198 * Clear the modify bits on the specified physical page.
3201 pmap_clear_modify(vm_page_t m)
3203 pmap_changebit(m, PG_M, FALSE);
3207 * pmap_clear_reference:
3209 * Clear the reference bit on the specified physical page.
3212 pmap_clear_reference(vm_page_t m)
3214 pmap_changebit(m, PG_A, FALSE);
3218 * Miscellaneous support routines follow
3222 i386_protection_init()
3224 register int *kp, prot;
3226 kp = protection_codes;
3227 for (prot = 0; prot < 8; prot++) {
3229 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3231 * Read access is also 0. There isn't any execute bit,
3232 * so just make it readable.
3234 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3235 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3236 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3239 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3240 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3241 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3242 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3250 * Map a set of physical memory pages into the kernel virtual
3251 * address space. Return a pointer to where it is mapped. This
3252 * routine is intended to be used for mapping device memory,
3256 pmap_mapdev(pa, size)
3260 vm_offset_t va, tmpva, offset;
3263 offset = pa & PAGE_MASK;
3264 size = roundup(offset + size, PAGE_SIZE);
3266 va = kmem_alloc_pageable(kernel_map, size);
3268 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3271 for (tmpva = va; size > 0;) {
3272 pte = (unsigned *)vtopte(tmpva);
3273 *pte = pa | PG_RW | PG_V | pgeflag;
3280 return ((void *)(va + offset));
3284 pmap_unmapdev(va, size)
3288 vm_offset_t base, offset;
3290 base = va & PG_FRAME;
3291 offset = va & PAGE_MASK;
3292 size = roundup(offset + size, PAGE_SIZE);
3293 kmem_free(kernel_map, base, size);
3297 * perform the pmap work for mincore
3300 pmap_mincore(pmap, addr)
3305 unsigned *ptep, pte;
3309 ptep = pmap_pte(pmap, addr);
3314 if ((pte = *ptep) != 0) {
3317 val = MINCORE_INCORE;
3318 if ((pte & PG_MANAGED) == 0)
3321 pa = pte & PG_FRAME;
3323 m = PHYS_TO_VM_PAGE(pa);
3329 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3331 * Modified by someone
3333 else if (m->dirty || pmap_is_modified(m))
3334 val |= MINCORE_MODIFIED_OTHER;
3339 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3342 * Referenced by someone
3344 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3345 val |= MINCORE_REFERENCED_OTHER;
3346 vm_page_flag_set(m, PG_REFERENCED);
3353 pmap_activate(struct proc *p)
3357 pmap = vmspace_pmap(p->p_vmspace);
3359 pmap->pm_active |= 1 << cpuid;
3361 pmap->pm_active |= 1;
3363 #if defined(SWTCH_OPTIM_STATS)
3366 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir));
3370 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3373 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3377 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3382 #if defined(PMAP_DEBUG)
3383 pmap_pid_dump(int pid)
3389 LIST_FOREACH(p, &allproc, p_list) {
3390 if (p->p_pid != pid)
3396 pmap = vmspace_pmap(p->p_vmspace);
3397 for(i=0;i<1024;i++) {
3400 unsigned base = i << PDRSHIFT;
3402 pde = &pmap->pm_pdir[i];
3403 if (pde && pmap_pde_v(pde)) {
3404 for(j=0;j<1024;j++) {
3405 unsigned va = base + (j << PAGE_SHIFT);
3406 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3413 pte = pmap_pte_quick( pmap, va);
3414 if (pte && pmap_pte_v(pte)) {
3418 m = PHYS_TO_VM_PAGE(pa);
3419 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3420 va, pa, m->hold_count, m->wire_count, m->flags);
3441 static void pads __P((pmap_t pm));
3442 void pmap_pvdump __P((vm_offset_t pa));
3444 /* print address space of pmap*/
3452 if (pm == kernel_pmap)
3454 for (i = 0; i < 1024; i++)
3456 for (j = 0; j < 1024; j++) {
3457 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3458 if (pm == kernel_pmap && va < KERNBASE)
3460 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3462 ptep = pmap_pte_quick(pm, va);
3463 if (pmap_pte_v(ptep))
3464 printf("%x:%x ", va, *(int *) ptep);
3473 register pv_entry_t pv;
3476 printf("pa %x", pa);
3477 m = PHYS_TO_VM_PAGE(pa);
3478 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3480 printf(" -> pmap %p, va %x, flags %x",
3481 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3483 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);