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 1.5 2003/06/18 18:29:55 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);
853 td->td_pcb = (struct pcb *)
854 ((char *)p->p_addr + UPAGES * PAGE_SIZE) - 1;
860 * Dispose of a thread, unlink from its related proc (if any)
863 pmap_dispose_thread(struct thread *td)
866 td->td_proc->p_thread = NULL;
869 zfree(thread_zone, td);
873 * Create the UPAGES for a new process.
874 * This routine directly affects the fork perf for a process.
884 unsigned *ptek, oldpte;
887 * allocate object for the upages
889 if ((upobj = p->p_upages_obj) == NULL) {
890 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES);
891 p->p_upages_obj = upobj;
894 /* get a kernel virtual address for the UPAGES for this proc */
895 if ((up = p->p_addr) == NULL) {
896 up = (struct user *) kmem_alloc_nofault(kernel_map,
899 panic("pmap_new_proc: u_map allocation failed");
903 ptek = (unsigned *) vtopte((vm_offset_t) up);
906 for(i=0;i<UPAGES;i++) {
908 * Get a kernel stack page
910 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
918 oldpte = *(ptek + i);
920 * Enter the page into the kernel address space.
922 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag;
924 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) {
925 invlpg((vm_offset_t) up + i * PAGE_SIZE);
932 vm_page_flag_clear(m, PG_ZERO);
933 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
934 m->valid = VM_PAGE_BITS_ALL;
941 * Dispose the UPAGES for a process that has exited.
942 * This routine directly impacts the exit perf of a process.
951 unsigned *ptek, oldpte;
953 upobj = p->p_upages_obj;
955 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr);
956 for(i=0;i<UPAGES;i++) {
958 if ((m = vm_page_lookup(upobj, i)) == NULL)
959 panic("pmap_dispose_proc: upage already missing???");
963 oldpte = *(ptek + i);
965 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386))
966 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE);
967 vm_page_unwire(m, 0);
970 #if defined(I386_CPU)
971 if (cpu_class <= CPUCLASS_386)
976 * If the process got swapped out some of its UPAGES might have gotten
977 * swapped. Just get rid of the object to clean up the swap use
978 * proactively. NOTE! might block waiting for paging I/O to complete.
980 if (upobj->type == OBJT_SWAP) {
981 p->p_upages_obj = NULL;
982 vm_object_deallocate(upobj);
987 * Allow the UPAGES for a process to be prejudicially paged out.
997 upobj = p->p_upages_obj;
999 * let the upages be paged
1001 for(i=0;i<UPAGES;i++) {
1002 if ((m = vm_page_lookup(upobj, i)) == NULL)
1003 panic("pmap_swapout_proc: upage already missing???");
1005 vm_page_unwire(m, 0);
1006 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
1011 * Bring the UPAGES for a specified process back in.
1021 upobj = p->p_upages_obj;
1022 for(i=0;i<UPAGES;i++) {
1024 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1026 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
1027 VM_PAGE_TO_PHYS(m));
1029 if (m->valid != VM_PAGE_BITS_ALL) {
1030 rv = vm_pager_get_pages(upobj, &m, 1, 0);
1031 if (rv != VM_PAGER_OK)
1032 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid);
1033 m = vm_page_lookup(upobj, i);
1034 m->valid = VM_PAGE_BITS_ALL;
1039 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE);
1043 /***************************************************
1044 * Page table page management routines.....
1045 ***************************************************/
1048 * This routine unholds page table pages, and if the hold count
1049 * drops to zero, then it decrements the wire count.
1052 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) {
1054 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
1057 if (m->hold_count == 0) {
1060 * unmap the page table page
1062 pmap->pm_pdir[m->pindex] = 0;
1063 --pmap->pm_stats.resident_count;
1064 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1065 (((unsigned) PTDpde) & PG_FRAME)) {
1067 * Do a invltlb to make the invalidated mapping
1068 * take effect immediately.
1070 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex);
1071 pmap_TLB_invalidate(pmap, pteva);
1074 if (pmap->pm_ptphint == m)
1075 pmap->pm_ptphint = NULL;
1078 * If the page is finally unwired, simply free it.
1081 if (m->wire_count == 0) {
1085 vm_page_free_zero(m);
1093 static PMAP_INLINE int
1094 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m)
1097 if (m->hold_count == 0)
1098 return _pmap_unwire_pte_hold(pmap, m);
1104 * After removing a page table entry, this routine is used to
1105 * conditionally free the page, and manage the hold/wire counts.
1108 pmap_unuse_pt(pmap, va, mpte)
1114 if (va >= UPT_MIN_ADDRESS)
1118 ptepindex = (va >> PDRSHIFT);
1119 if (pmap->pm_ptphint &&
1120 (pmap->pm_ptphint->pindex == ptepindex)) {
1121 mpte = pmap->pm_ptphint;
1123 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1124 pmap->pm_ptphint = mpte;
1128 return pmap_unwire_pte_hold(pmap, mpte);
1136 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1137 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD);
1139 pmap->pm_active = 0;
1140 pmap->pm_ptphint = NULL;
1141 TAILQ_INIT(&pmap->pm_pvlist);
1142 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1146 * Initialize a preallocated and zeroed pmap structure,
1147 * such as one in a vmspace structure.
1151 register struct pmap *pmap;
1156 * No need to allocate page table space yet but we do need a valid
1157 * page directory table.
1159 if (pmap->pm_pdir == NULL)
1161 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE);
1164 * allocate object for the ptes
1166 if (pmap->pm_pteobj == NULL)
1167 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1);
1170 * allocate the page directory page
1172 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI,
1173 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1175 ptdpg->wire_count = 1;
1179 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/
1180 ptdpg->valid = VM_PAGE_BITS_ALL;
1182 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1183 if ((ptdpg->flags & PG_ZERO) == 0)
1184 bzero(pmap->pm_pdir, PAGE_SIZE);
1187 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1190 /* install self-referential address mapping entry */
1191 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1192 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1195 pmap->pm_active = 0;
1196 pmap->pm_ptphint = NULL;
1197 TAILQ_INIT(&pmap->pm_pvlist);
1198 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1202 * Wire in kernel global address entries. To avoid a race condition
1203 * between pmap initialization and pmap_growkernel, this procedure
1204 * should be called after the vmspace is attached to the process
1205 * but before this pmap is activated.
1211 /* XXX copies current process, does not fill in MPPTDI */
1212 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1216 pmap_release_free_page(pmap, p)
1220 unsigned *pde = (unsigned *) pmap->pm_pdir;
1222 * This code optimizes the case of freeing non-busy
1223 * page-table pages. Those pages are zero now, and
1224 * might as well be placed directly into the zero queue.
1226 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
1232 * Remove the page table page from the processes address space.
1235 pmap->pm_stats.resident_count--;
1237 if (p->hold_count) {
1238 panic("pmap_release: freeing held page table page");
1241 * Page directory pages need to have the kernel
1242 * stuff cleared, so they can go into the zero queue also.
1244 if (p->pindex == PTDPTDI) {
1245 bzero(pde + KPTDI, nkpt * PTESIZE);
1250 pmap_kremove((vm_offset_t) pmap->pm_pdir);
1253 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1254 pmap->pm_ptphint = NULL;
1258 vm_page_free_zero(p);
1263 * this routine is called if the page table page is not
1267 _pmap_allocpte(pmap, ptepindex)
1271 vm_offset_t pteva, ptepa;
1275 * Find or fabricate a new pagetable page
1277 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1278 VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1280 KASSERT(m->queue == PQ_NONE,
1281 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1283 if (m->wire_count == 0)
1288 * Increment the hold count for the page table page
1289 * (denoting a new mapping.)
1294 * Map the pagetable page into the process address space, if
1295 * it isn't already there.
1298 pmap->pm_stats.resident_count++;
1300 ptepa = VM_PAGE_TO_PHYS(m);
1301 pmap->pm_pdir[ptepindex] =
1302 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1305 * Set the page table hint
1307 pmap->pm_ptphint = m;
1310 * Try to use the new mapping, but if we cannot, then
1311 * do it with the routine that maps the page explicitly.
1313 if ((m->flags & PG_ZERO) == 0) {
1314 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) ==
1315 (((unsigned) PTDpde) & PG_FRAME)) {
1316 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex);
1317 bzero((caddr_t) pteva, PAGE_SIZE);
1319 pmap_zero_page(ptepa);
1323 m->valid = VM_PAGE_BITS_ALL;
1324 vm_page_flag_clear(m, PG_ZERO);
1325 vm_page_flag_set(m, PG_MAPPED);
1332 pmap_allocpte(pmap, va)
1341 * Calculate pagetable page index
1343 ptepindex = va >> PDRSHIFT;
1346 * Get the page directory entry
1348 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1351 * This supports switching from a 4MB page to a
1354 if (ptepa & PG_PS) {
1355 pmap->pm_pdir[ptepindex] = 0;
1361 * If the page table page is mapped, we just increment the
1362 * hold count, and activate it.
1366 * In order to get the page table page, try the
1369 if (pmap->pm_ptphint &&
1370 (pmap->pm_ptphint->pindex == ptepindex)) {
1371 m = pmap->pm_ptphint;
1373 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1374 pmap->pm_ptphint = m;
1380 * Here if the pte page isn't mapped, or if it has been deallocated.
1382 return _pmap_allocpte(pmap, ptepindex);
1386 /***************************************************
1387 * Pmap allocation/deallocation routines.
1388 ***************************************************/
1391 * Release any resources held by the given physical map.
1392 * Called when a pmap initialized by pmap_pinit is being released.
1393 * Should only be called if the map contains no valid mappings.
1397 register struct pmap *pmap;
1399 vm_page_t p,n,ptdpg;
1400 vm_object_t object = pmap->pm_pteobj;
1403 #if defined(DIAGNOSTIC)
1404 if (object->ref_count != 1)
1405 panic("pmap_release: pteobj reference count != 1");
1410 curgeneration = object->generation;
1411 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) {
1412 n = TAILQ_NEXT(p, listq);
1413 if (p->pindex == PTDPTDI) {
1418 if (!pmap_release_free_page(pmap, p) &&
1419 (object->generation != curgeneration))
1424 if (ptdpg && !pmap_release_free_page(pmap, ptdpg))
1429 kvm_size(SYSCTL_HANDLER_ARGS)
1431 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE;
1433 return sysctl_handle_long(oidp, &ksize, 0, req);
1435 SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD,
1436 0, 0, kvm_size, "IU", "Size of KVM");
1439 kvm_free(SYSCTL_HANDLER_ARGS)
1441 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end;
1443 return sysctl_handle_long(oidp, &kfree, 0, req);
1445 SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD,
1446 0, 0, kvm_free, "IU", "Amount of KVM free");
1449 * grow the number of kernel page table entries, if needed
1452 pmap_growkernel(vm_offset_t addr)
1457 vm_offset_t ptppaddr;
1462 if (kernel_vm_end == 0) {
1463 kernel_vm_end = KERNBASE;
1465 while (pdir_pde(PTD, kernel_vm_end)) {
1466 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1470 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1471 while (kernel_vm_end < addr) {
1472 if (pdir_pde(PTD, kernel_vm_end)) {
1473 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1478 * This index is bogus, but out of the way
1480 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM);
1482 panic("pmap_growkernel: no memory to grow kernel");
1487 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1488 pmap_zero_page(ptppaddr);
1489 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1490 pdir_pde(PTD, kernel_vm_end) = newpdir;
1492 LIST_FOREACH(p, &allproc, p_list) {
1494 pmap = vmspace_pmap(p->p_vmspace);
1495 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1498 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir;
1499 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1505 * Retire the given physical map from service.
1506 * Should only be called if the map contains
1507 * no valid mappings.
1511 register pmap_t pmap;
1518 count = --pmap->pm_count;
1521 panic("destroying a pmap is not yet implemented");
1526 * Add a reference to the specified pmap.
1529 pmap_reference(pmap)
1537 /***************************************************
1538 * page management routines.
1539 ***************************************************/
1542 * free the pv_entry back to the free list
1544 static PMAP_INLINE void
1553 * get a new pv_entry, allocating a block from the system
1555 * the memory allocation is performed bypassing the malloc code
1556 * because of the possibility of allocations at interrupt time.
1562 if (pv_entry_high_water &&
1563 (pv_entry_count > pv_entry_high_water) &&
1564 (pmap_pagedaemon_waken == 0)) {
1565 pmap_pagedaemon_waken = 1;
1566 wakeup (&vm_pages_needed);
1568 return zalloci(pvzone);
1572 * This routine is very drastic, but can save the system
1580 static int warningdone=0;
1582 if (pmap_pagedaemon_waken == 0)
1585 if (warningdone < 5) {
1586 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1590 for(i = 0; i < vm_page_array_size; i++) {
1591 m = &vm_page_array[i];
1592 if (m->wire_count || m->hold_count || m->busy ||
1593 (m->flags & PG_BUSY))
1597 pmap_pagedaemon_waken = 0;
1602 * If it is the first entry on the list, it is actually
1603 * in the header and we must copy the following entry up
1604 * to the header. Otherwise we must search the list for
1605 * the entry. In either case we free the now unused entry.
1609 pmap_remove_entry(pmap, m, va)
1619 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1620 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1621 if (pmap == pv->pv_pmap && va == pv->pv_va)
1625 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1626 if (va == pv->pv_va)
1634 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem);
1635 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1636 m->md.pv_list_count--;
1637 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1638 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1640 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1649 * Create a pv entry for page at pa for
1653 pmap_insert_entry(pmap, va, mpte, m)
1664 pv = get_pv_entry();
1669 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1670 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1671 m->md.pv_list_count++;
1677 * pmap_remove_pte: do the things to unmap a page in a process
1680 pmap_remove_pte(pmap, ptq, va)
1688 oldpte = loadandclear(ptq);
1690 pmap->pm_stats.wired_count -= 1;
1692 * Machines that don't support invlpg, also don't support
1697 pmap->pm_stats.resident_count -= 1;
1698 if (oldpte & PG_MANAGED) {
1699 m = PHYS_TO_VM_PAGE(oldpte);
1700 if (oldpte & PG_M) {
1701 #if defined(PMAP_DIAGNOSTIC)
1702 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1704 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1708 if (pmap_track_modified(va))
1712 vm_page_flag_set(m, PG_REFERENCED);
1713 return pmap_remove_entry(pmap, m, va);
1715 return pmap_unuse_pt(pmap, va, NULL);
1722 * Remove a single page from a process address space
1725 pmap_remove_page(pmap, va)
1727 register vm_offset_t va;
1729 register unsigned *ptq;
1732 * if there is no pte for this address, just skip it!!!
1734 if (*pmap_pde(pmap, va) == 0) {
1739 * get a local va for mappings for this pmap.
1741 ptq = get_ptbase(pmap) + i386_btop(va);
1743 (void) pmap_remove_pte(pmap, ptq, va);
1744 pmap_TLB_invalidate(pmap, va);
1750 * Remove the given range of addresses from the specified map.
1752 * It is assumed that the start and end are properly
1753 * rounded to the page size.
1756 pmap_remove(pmap, sva, eva)
1758 register vm_offset_t sva;
1759 register vm_offset_t eva;
1761 register unsigned *ptbase;
1763 vm_offset_t ptpaddr;
1764 vm_offset_t sindex, eindex;
1770 if (pmap->pm_stats.resident_count == 0)
1774 * special handling of removing one page. a very
1775 * common operation and easy to short circuit some
1778 if (((sva + PAGE_SIZE) == eva) &&
1779 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1780 pmap_remove_page(pmap, sva);
1787 * Get a local virtual address for the mappings that are being
1790 ptbase = get_ptbase(pmap);
1792 sindex = i386_btop(sva);
1793 eindex = i386_btop(eva);
1795 for (; sindex < eindex; sindex = pdnxt) {
1799 * Calculate index for next page table.
1801 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1802 if (pmap->pm_stats.resident_count == 0)
1805 pdirindex = sindex / NPDEPG;
1806 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1807 pmap->pm_pdir[pdirindex] = 0;
1808 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1814 * Weed out invalid mappings. Note: we assume that the page
1815 * directory table is always allocated, and in kernel virtual.
1821 * Limit our scan to either the end of the va represented
1822 * by the current page table page, or to the end of the
1823 * range being removed.
1825 if (pdnxt > eindex) {
1829 for ( ;sindex != pdnxt; sindex++) {
1831 if (ptbase[sindex] == 0) {
1834 va = i386_ptob(sindex);
1837 if (pmap_remove_pte(pmap,
1838 ptbase + sindex, va))
1844 pmap_TLB_invalidate_all(pmap);
1848 * Routine: pmap_remove_all
1850 * Removes this physical page from
1851 * all physical maps in which it resides.
1852 * Reflects back modify bits to the pager.
1855 * Original versions of this routine were very
1856 * inefficient because they iteratively called
1857 * pmap_remove (slow...)
1864 register pv_entry_t pv;
1865 register unsigned *pte, tpte;
1868 #if defined(PMAP_DIAGNOSTIC)
1870 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1873 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1874 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m));
1879 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1880 pv->pv_pmap->pm_stats.resident_count--;
1882 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
1884 tpte = loadandclear(pte);
1886 pv->pv_pmap->pm_stats.wired_count--;
1889 vm_page_flag_set(m, PG_REFERENCED);
1892 * Update the vm_page_t clean and reference bits.
1895 #if defined(PMAP_DIAGNOSTIC)
1896 if (pmap_nw_modified((pt_entry_t) tpte)) {
1898 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1902 if (pmap_track_modified(pv->pv_va))
1905 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
1907 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1908 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1909 m->md.pv_list_count--;
1910 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
1914 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1920 * Set the physical protection on the
1921 * specified range of this map as requested.
1924 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1926 register unsigned *ptbase;
1927 vm_offset_t pdnxt, ptpaddr;
1928 vm_pindex_t sindex, eindex;
1934 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1935 pmap_remove(pmap, sva, eva);
1939 if (prot & VM_PROT_WRITE)
1944 ptbase = get_ptbase(pmap);
1946 sindex = i386_btop(sva);
1947 eindex = i386_btop(eva);
1949 for (; sindex < eindex; sindex = pdnxt) {
1953 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1955 pdirindex = sindex / NPDEPG;
1956 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
1957 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
1958 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1964 * Weed out invalid mappings. Note: we assume that the page
1965 * directory table is always allocated, and in kernel virtual.
1970 if (pdnxt > eindex) {
1974 for (; sindex != pdnxt; sindex++) {
1979 pbits = ptbase[sindex];
1981 if (pbits & PG_MANAGED) {
1984 m = PHYS_TO_VM_PAGE(pbits);
1985 vm_page_flag_set(m, PG_REFERENCED);
1989 if (pmap_track_modified(i386_ptob(sindex))) {
1991 m = PHYS_TO_VM_PAGE(pbits);
2000 if (pbits != ptbase[sindex]) {
2001 ptbase[sindex] = pbits;
2007 pmap_TLB_invalidate_all(pmap);
2011 * Insert the given physical page (p) at
2012 * the specified virtual address (v) in the
2013 * target physical map with the protection requested.
2015 * If specified, the page will be wired down, meaning
2016 * that the related pte can not be reclaimed.
2018 * NB: This is the only routine which MAY NOT lazy-evaluate
2019 * or lose information. That is, this routine must actually
2020 * insert this page into the given map NOW.
2023 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
2027 register unsigned *pte;
2029 vm_offset_t origpte, newpte;
2036 #ifdef PMAP_DIAGNOSTIC
2037 if (va > VM_MAX_KERNEL_ADDRESS)
2038 panic("pmap_enter: toobig");
2039 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS))
2040 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va);
2045 * In the case that a page table page is not
2046 * resident, we are creating it here.
2048 if (va < UPT_MIN_ADDRESS) {
2049 mpte = pmap_allocpte(pmap, va);
2051 #if 0 && defined(PMAP_DIAGNOSTIC)
2053 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va);
2054 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) {
2055 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n",
2056 pmap->pm_pdir[PTDPTDI], origpte, va);
2059 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid];
2060 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) {
2061 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr))
2062 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr);
2063 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr);
2064 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n",
2065 pmap->pm_pdir[PTDPTDI], newpte, origpte, va);
2071 pte = pmap_pte(pmap, va);
2074 * Page Directory table entry not valid, we need a new PT page
2077 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n",
2078 (void *)pmap->pm_pdir[PTDPTDI], va);
2081 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2082 origpte = *(vm_offset_t *)pte;
2083 opa = origpte & PG_FRAME;
2085 if (origpte & PG_PS)
2086 panic("pmap_enter: attempted pmap_enter on 4MB page");
2089 * Mapping has not changed, must be protection or wiring change.
2091 if (origpte && (opa == pa)) {
2093 * Wiring change, just update stats. We don't worry about
2094 * wiring PT pages as they remain resident as long as there
2095 * are valid mappings in them. Hence, if a user page is wired,
2096 * the PT page will be also.
2098 if (wired && ((origpte & PG_W) == 0))
2099 pmap->pm_stats.wired_count++;
2100 else if (!wired && (origpte & PG_W))
2101 pmap->pm_stats.wired_count--;
2103 #if defined(PMAP_DIAGNOSTIC)
2104 if (pmap_nw_modified((pt_entry_t) origpte)) {
2106 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
2112 * Remove extra pte reference
2117 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) {
2118 if ((origpte & PG_RW) == 0) {
2121 cpu_invlpg((void *)va);
2122 if (pmap->pm_active & other_cpus)
2132 * We might be turning off write access to the page,
2133 * so we go ahead and sense modify status.
2135 if (origpte & PG_MANAGED) {
2136 if ((origpte & PG_M) && pmap_track_modified(va)) {
2138 om = PHYS_TO_VM_PAGE(opa);
2146 * Mapping has changed, invalidate old range and fall through to
2147 * handle validating new mapping.
2151 err = pmap_remove_pte(pmap, pte, va);
2153 panic("pmap_enter: pte vanished, va: 0x%x", va);
2157 * Enter on the PV list if part of our managed memory. Note that we
2158 * raise IPL while manipulating pv_table since pmap_enter can be
2159 * called at interrupt time.
2161 if (pmap_initialized &&
2162 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2163 pmap_insert_entry(pmap, va, mpte, m);
2168 * Increment counters
2170 pmap->pm_stats.resident_count++;
2172 pmap->pm_stats.wired_count++;
2176 * Now validate mapping with desired protection/wiring.
2178 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2182 if (va < UPT_MIN_ADDRESS)
2184 if (pmap == kernel_pmap)
2188 * if the mapping or permission bits are different, we need
2189 * to update the pte.
2191 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2192 *pte = newpte | PG_A;
2195 cpu_invlpg((void *)va);
2196 if (pmap->pm_active & other_cpus)
2206 * this code makes some *MAJOR* assumptions:
2207 * 1. Current pmap & pmap exists.
2210 * 4. No page table pages.
2211 * 5. Tlbflush is deferred to calling procedure.
2212 * 6. Page IS managed.
2213 * but is *MUCH* faster than pmap_enter...
2217 pmap_enter_quick(pmap, va, m, mpte)
2218 register pmap_t pmap;
2227 * In the case that a page table page is not
2228 * resident, we are creating it here.
2230 if (va < UPT_MIN_ADDRESS) {
2235 * Calculate pagetable page index
2237 ptepindex = va >> PDRSHIFT;
2238 if (mpte && (mpte->pindex == ptepindex)) {
2243 * Get the page directory entry
2245 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2248 * If the page table page is mapped, we just increment
2249 * the hold count, and activate it.
2253 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2254 if (pmap->pm_ptphint &&
2255 (pmap->pm_ptphint->pindex == ptepindex)) {
2256 mpte = pmap->pm_ptphint;
2258 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
2259 pmap->pm_ptphint = mpte;
2265 mpte = _pmap_allocpte(pmap, ptepindex);
2273 * This call to vtopte makes the assumption that we are
2274 * entering the page into the current pmap. In order to support
2275 * quick entry into any pmap, one would likely use pmap_pte_quick.
2276 * But that isn't as quick as vtopte.
2278 pte = (unsigned *)vtopte(va);
2281 pmap_unwire_pte_hold(pmap, mpte);
2286 * Enter on the PV list if part of our managed memory. Note that we
2287 * raise IPL while manipulating pv_table since pmap_enter can be
2288 * called at interrupt time.
2290 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
2291 pmap_insert_entry(pmap, va, mpte, m);
2294 * Increment counters
2296 pmap->pm_stats.resident_count++;
2298 pa = VM_PAGE_TO_PHYS(m);
2301 * Now validate mapping with RO protection
2303 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2304 *pte = pa | PG_V | PG_U;
2306 *pte = pa | PG_V | PG_U | PG_MANAGED;
2312 * Make a temporary mapping for a physical address. This is only intended
2313 * to be used for panic dumps.
2316 pmap_kenter_temporary(vm_offset_t pa, int i)
2318 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2319 return ((void *)crashdumpmap);
2322 #define MAX_INIT_PT (96)
2324 * pmap_object_init_pt preloads the ptes for a given object
2325 * into the specified pmap. This eliminates the blast of soft
2326 * faults on process startup and immediately after an mmap.
2329 pmap_object_init_pt(pmap, addr, object, pindex, size, limit)
2342 if (pmap == NULL || object == NULL)
2346 * This code maps large physical mmap regions into the
2347 * processor address space. Note that some shortcuts
2348 * are taken, but the code works.
2351 (object->type == OBJT_DEVICE) &&
2352 ((addr & (NBPDR - 1)) == 0) &&
2353 ((size & (NBPDR - 1)) == 0) ) {
2356 unsigned int ptepindex;
2360 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)])
2364 p = vm_page_lookup(object, pindex);
2365 if (p && vm_page_sleep_busy(p, FALSE, "init4p"))
2369 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL);
2374 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) {
2379 p = vm_page_lookup(object, pindex);
2383 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p);
2384 if (ptepa & (NBPDR - 1)) {
2388 p->valid = VM_PAGE_BITS_ALL;
2390 pmap->pm_stats.resident_count += size >> PAGE_SHIFT;
2391 npdes = size >> PDRSHIFT;
2392 for(i=0;i<npdes;i++) {
2393 pmap->pm_pdir[ptepindex] =
2394 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS);
2398 vm_page_flag_set(p, PG_MAPPED);
2403 psize = i386_btop(size);
2405 if ((object->type != OBJT_VNODE) ||
2406 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2407 (object->resident_page_count > MAX_INIT_PT))) {
2411 if (psize + pindex > object->size) {
2412 if (object->size < pindex)
2414 psize = object->size - pindex;
2419 * if we are processing a major portion of the object, then scan the
2422 if (psize > (object->resident_page_count >> 2)) {
2425 for (p = TAILQ_FIRST(&object->memq);
2426 ((objpgs > 0) && (p != NULL));
2427 p = TAILQ_NEXT(p, listq)) {
2430 if (tmpidx < pindex) {
2434 if (tmpidx >= psize) {
2438 * don't allow an madvise to blow away our really
2439 * free pages allocating pv entries.
2441 if ((limit & MAP_PREFAULT_MADVISE) &&
2442 cnt.v_free_count < cnt.v_free_reserved) {
2445 if (((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);
2460 * else lookup the pages one-by-one.
2462 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) {
2464 * don't allow an madvise to blow away our really
2465 * free pages allocating pv entries.
2467 if ((limit & MAP_PREFAULT_MADVISE) &&
2468 cnt.v_free_count < cnt.v_free_reserved) {
2471 p = vm_page_lookup(object, tmpidx + pindex);
2473 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2475 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2476 if ((p->queue - p->pc) == PQ_CACHE)
2477 vm_page_deactivate(p);
2479 mpte = pmap_enter_quick(pmap,
2480 addr + i386_ptob(tmpidx), p, mpte);
2481 vm_page_flag_set(p, PG_MAPPED);
2490 * pmap_prefault provides a quick way of clustering
2491 * pagefaults into a processes address space. It is a "cousin"
2492 * of pmap_object_init_pt, except it runs at page fault time instead
2497 #define PAGEORDER_SIZE (PFBAK+PFFOR)
2499 static int pmap_prefault_pageorder[] = {
2500 -PAGE_SIZE, PAGE_SIZE,
2501 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
2502 -3 * PAGE_SIZE, 3 * PAGE_SIZE
2503 -4 * PAGE_SIZE, 4 * PAGE_SIZE
2507 pmap_prefault(pmap, addra, entry)
2510 vm_map_entry_t entry;
2519 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace)))
2522 object = entry->object.vm_object;
2524 starta = addra - PFBAK * PAGE_SIZE;
2525 if (starta < entry->start) {
2526 starta = entry->start;
2527 } else if (starta > addra) {
2532 for (i = 0; i < PAGEORDER_SIZE; i++) {
2533 vm_object_t lobject;
2536 addr = addra + pmap_prefault_pageorder[i];
2537 if (addr > addra + (PFFOR * PAGE_SIZE))
2540 if (addr < starta || addr >= entry->end)
2543 if ((*pmap_pde(pmap, addr)) == NULL)
2546 pte = (unsigned *) vtopte(addr);
2550 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2552 for (m = vm_page_lookup(lobject, pindex);
2553 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object));
2554 lobject = lobject->backing_object) {
2555 if (lobject->backing_object_offset & PAGE_MASK)
2557 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2558 m = vm_page_lookup(lobject->backing_object, pindex);
2562 * give-up when a page is not in memory
2567 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2569 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2571 if ((m->queue - m->pc) == PQ_CACHE) {
2572 vm_page_deactivate(m);
2575 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2576 vm_page_flag_set(m, PG_MAPPED);
2583 * Routine: pmap_change_wiring
2584 * Function: Change the wiring attribute for a map/virtual-address
2586 * In/out conditions:
2587 * The mapping must already exist in the pmap.
2590 pmap_change_wiring(pmap, va, wired)
2591 register pmap_t pmap;
2595 register unsigned *pte;
2600 pte = pmap_pte(pmap, va);
2602 if (wired && !pmap_pte_w(pte))
2603 pmap->pm_stats.wired_count++;
2604 else if (!wired && pmap_pte_w(pte))
2605 pmap->pm_stats.wired_count--;
2608 * Wiring is not a hardware characteristic so there is no need to
2611 pmap_pte_set_w(pte, wired);
2617 * Copy the range specified by src_addr/len
2618 * from the source map to the range dst_addr/len
2619 * in the destination map.
2621 * This routine is only advisory and need not do anything.
2625 pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr)
2626 pmap_t dst_pmap, src_pmap;
2627 vm_offset_t dst_addr;
2629 vm_offset_t src_addr;
2632 vm_offset_t end_addr = src_addr + len;
2634 unsigned src_frame, dst_frame;
2637 if (dst_addr != src_addr)
2640 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2641 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) {
2645 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME;
2646 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) {
2647 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V);
2649 /* The page directory is not shared between CPUs */
2656 for(addr = src_addr; addr < end_addr; addr = pdnxt) {
2657 unsigned *src_pte, *dst_pte;
2658 vm_page_t dstmpte, srcmpte;
2659 vm_offset_t srcptepaddr;
2662 if (addr >= UPT_MIN_ADDRESS)
2663 panic("pmap_copy: invalid to pmap_copy page tables\n");
2666 * Don't let optional prefaulting of pages make us go
2667 * way below the low water mark of free pages or way
2668 * above high water mark of used pv entries.
2670 if (cnt.v_free_count < cnt.v_free_reserved ||
2671 pv_entry_count > pv_entry_high_water)
2674 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2675 ptepindex = addr >> PDRSHIFT;
2677 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2678 if (srcptepaddr == 0)
2681 if (srcptepaddr & PG_PS) {
2682 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2683 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2684 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2689 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2690 if ((srcmpte == NULL) ||
2691 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2694 if (pdnxt > end_addr)
2697 src_pte = (unsigned *) vtopte(addr);
2698 dst_pte = (unsigned *) avtopte(addr);
2699 while (addr < pdnxt) {
2703 * we only virtual copy managed pages
2705 if ((ptetemp & PG_MANAGED) != 0) {
2707 * We have to check after allocpte for the
2708 * pte still being around... allocpte can
2711 dstmpte = pmap_allocpte(dst_pmap, addr);
2712 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2714 * Clear the modified and
2715 * accessed (referenced) bits
2718 m = PHYS_TO_VM_PAGE(ptetemp);
2719 *dst_pte = ptetemp & ~(PG_M | PG_A);
2720 dst_pmap->pm_stats.resident_count++;
2721 pmap_insert_entry(dst_pmap, addr,
2724 pmap_unwire_pte_hold(dst_pmap, dstmpte);
2726 if (dstmpte->hold_count >= srcmpte->hold_count)
2737 * Routine: pmap_kernel
2739 * Returns the physical map handle for the kernel.
2744 return (kernel_pmap);
2748 * pmap_zero_page zeros the specified hardware page by mapping
2749 * the page into KVM and using bzero to clear its contents.
2752 pmap_zero_page(phys)
2756 if (*(int *) prv_CMAP3)
2757 panic("pmap_zero_page: prv_CMAP3 busy");
2759 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2760 cpu_invlpg(prv_CADDR3);
2762 #if defined(I686_CPU)
2763 if (cpu_class == CPUCLASS_686)
2764 i686_pagezero(prv_CADDR3);
2767 bzero(prv_CADDR3, PAGE_SIZE);
2769 *(int *) prv_CMAP3 = 0;
2772 panic("pmap_zero_page: CMAP2 busy");
2774 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2775 invltlb_1pg((vm_offset_t)CADDR2);
2777 #if defined(I686_CPU)
2778 if (cpu_class == CPUCLASS_686)
2779 i686_pagezero(CADDR2);
2782 bzero(CADDR2, PAGE_SIZE);
2788 * pmap_zero_page_area zeros the specified hardware page by mapping
2789 * the page into KVM and using bzero to clear its contents.
2791 * off and size may not cover an area beyond a single hardware page.
2794 pmap_zero_page_area(phys, off, size)
2800 if (*(int *) prv_CMAP3)
2801 panic("pmap_zero_page: prv_CMAP3 busy");
2803 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2804 cpu_invlpg(prv_CADDR3);
2806 #if defined(I686_CPU)
2807 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2808 i686_pagezero(prv_CADDR3);
2811 bzero((char *)prv_CADDR3 + off, size);
2813 *(int *) prv_CMAP3 = 0;
2816 panic("pmap_zero_page: CMAP2 busy");
2818 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2819 invltlb_1pg((vm_offset_t)CADDR2);
2821 #if defined(I686_CPU)
2822 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE)
2823 i686_pagezero(CADDR2);
2826 bzero((char *)CADDR2 + off, size);
2832 * pmap_copy_page copies the specified (machine independent)
2833 * page by mapping the page into virtual memory and using
2834 * bcopy to copy the page, one machine dependent page at a
2838 pmap_copy_page(src, dst)
2843 if (*(int *) prv_CMAP1)
2844 panic("pmap_copy_page: prv_CMAP1 busy");
2845 if (*(int *) prv_CMAP2)
2846 panic("pmap_copy_page: prv_CMAP2 busy");
2848 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2849 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2851 cpu_invlpg(prv_CADDR1);
2852 cpu_invlpg(prv_CADDR2);
2854 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE);
2856 *(int *) prv_CMAP1 = 0;
2857 *(int *) prv_CMAP2 = 0;
2859 if (*(int *) CMAP1 || *(int *) CMAP2)
2860 panic("pmap_copy_page: CMAP busy");
2862 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2863 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2864 #if defined(I386_CPU)
2865 if (cpu_class == CPUCLASS_386) {
2870 invlpg((u_int)CADDR1);
2871 invlpg((u_int)CADDR2);
2874 bcopy(CADDR1, CADDR2, PAGE_SIZE);
2883 * Routine: pmap_pageable
2885 * Make the specified pages (by pmap, offset)
2886 * pageable (or not) as requested.
2888 * A page which is not pageable may not take
2889 * a fault; therefore, its page table entry
2890 * must remain valid for the duration.
2892 * This routine is merely advisory; pmap_enter
2893 * will specify that these pages are to be wired
2894 * down (or not) as appropriate.
2897 pmap_pageable(pmap, sva, eva, pageable)
2899 vm_offset_t sva, eva;
2905 * Returns true if the pmap's pv is one of the first
2906 * 16 pvs linked to from this page. This count may
2907 * be changed upwards or downwards in the future; it
2908 * is only necessary that true be returned for a small
2909 * subset of pmaps for proper page aging.
2912 pmap_page_exists_quick(pmap, m)
2920 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2925 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2926 if (pv->pv_pmap == pmap) {
2938 #define PMAP_REMOVE_PAGES_CURPROC_ONLY
2940 * Remove all pages from specified address space
2941 * this aids process exit speeds. Also, this code
2942 * is special cased for current process only, but
2943 * can have the more generic (and slightly slower)
2944 * mode enabled. This is much faster than pmap_remove
2945 * in the case of running down an entire address space.
2948 pmap_remove_pages(pmap, sva, eva)
2950 vm_offset_t sva, eva;
2952 unsigned *pte, tpte;
2957 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2958 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) {
2959 printf("warning: pmap_remove_pages called with non-current pmap\n");
2965 for(pv = TAILQ_FIRST(&pmap->pm_pvlist);
2969 if (pv->pv_va >= eva || pv->pv_va < sva) {
2970 npv = TAILQ_NEXT(pv, pv_plist);
2974 #ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY
2975 pte = (unsigned *)vtopte(pv->pv_va);
2977 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
2982 * We cannot remove wired pages from a process' mapping at this time
2985 npv = TAILQ_NEXT(pv, pv_plist);
2990 m = PHYS_TO_VM_PAGE(tpte);
2992 KASSERT(m < &vm_page_array[vm_page_array_size],
2993 ("pmap_remove_pages: bad tpte %x", tpte));
2995 pv->pv_pmap->pm_stats.resident_count--;
2998 * Update the vm_page_t clean and reference bits.
3005 npv = TAILQ_NEXT(pv, pv_plist);
3006 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
3008 m->md.pv_list_count--;
3009 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3010 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
3011 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
3014 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem);
3018 pmap_TLB_invalidate_all(pmap);
3022 * pmap_testbit tests bits in pte's
3023 * note that the testbit/changebit routines are inline,
3024 * and a lot of things compile-time evaluate.
3027 pmap_testbit(m, bit)
3035 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3038 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3043 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3045 * if the bit being tested is the modified bit, then
3046 * mark clean_map and ptes as never
3049 if (bit & (PG_A|PG_M)) {
3050 if (!pmap_track_modified(pv->pv_va))
3054 #if defined(PMAP_DIAGNOSTIC)
3056 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
3060 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3071 * this routine is used to modify bits in ptes
3073 static __inline void
3074 pmap_changebit(m, bit, setem)
3079 register pv_entry_t pv;
3080 register unsigned *pte;
3083 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3089 * Loop over all current mappings setting/clearing as appropos If
3090 * setting RO do we need to clear the VAC?
3092 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3094 * don't write protect pager mappings
3096 if (!setem && (bit == PG_RW)) {
3097 if (!pmap_track_modified(pv->pv_va))
3101 #if defined(PMAP_DIAGNOSTIC)
3103 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
3108 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3112 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3114 vm_offset_t pbits = *(vm_offset_t *)pte;
3120 *(int *)pte = pbits & ~(PG_M|PG_RW);
3122 *(int *)pte = pbits & ~bit;
3124 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3132 * pmap_page_protect:
3134 * Lower the permission for all mappings to a given page.
3137 pmap_page_protect(vm_page_t m, vm_prot_t prot)
3139 if ((prot & VM_PROT_WRITE) == 0) {
3140 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3141 pmap_changebit(m, PG_RW, FALSE);
3149 pmap_phys_address(ppn)
3152 return (i386_ptob(ppn));
3156 * pmap_ts_referenced:
3158 * Return a count of reference bits for a page, clearing those bits.
3159 * It is not necessary for every reference bit to be cleared, but it
3160 * is necessary that 0 only be returned when there are truly no
3161 * reference bits set.
3163 * XXX: The exact number of bits to check and clear is a matter that
3164 * should be tested and standardized at some point in the future for
3165 * optimal aging of shared pages.
3168 pmap_ts_referenced(vm_page_t m)
3170 register pv_entry_t pv, pvf, pvn;
3175 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3180 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3185 pvn = TAILQ_NEXT(pv, pv_list);
3187 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3189 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3191 if (!pmap_track_modified(pv->pv_va))
3194 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3196 if (pte && (*pte & PG_A)) {
3199 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va);
3206 } while ((pv = pvn) != NULL && pv != pvf);
3216 * Return whether or not the specified physical page was modified
3217 * in any physical maps.
3220 pmap_is_modified(vm_page_t m)
3222 return pmap_testbit(m, PG_M);
3226 * Clear the modify bits on the specified physical page.
3229 pmap_clear_modify(vm_page_t m)
3231 pmap_changebit(m, PG_M, FALSE);
3235 * pmap_clear_reference:
3237 * Clear the reference bit on the specified physical page.
3240 pmap_clear_reference(vm_page_t m)
3242 pmap_changebit(m, PG_A, FALSE);
3246 * Miscellaneous support routines follow
3250 i386_protection_init()
3252 register int *kp, prot;
3254 kp = protection_codes;
3255 for (prot = 0; prot < 8; prot++) {
3257 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3259 * Read access is also 0. There isn't any execute bit,
3260 * so just make it readable.
3262 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3263 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3264 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3267 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3268 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3269 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3270 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3278 * Map a set of physical memory pages into the kernel virtual
3279 * address space. Return a pointer to where it is mapped. This
3280 * routine is intended to be used for mapping device memory,
3284 pmap_mapdev(pa, size)
3288 vm_offset_t va, tmpva, offset;
3291 offset = pa & PAGE_MASK;
3292 size = roundup(offset + size, PAGE_SIZE);
3294 va = kmem_alloc_pageable(kernel_map, size);
3296 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3299 for (tmpva = va; size > 0;) {
3300 pte = (unsigned *)vtopte(tmpva);
3301 *pte = pa | PG_RW | PG_V | pgeflag;
3308 return ((void *)(va + offset));
3312 pmap_unmapdev(va, size)
3316 vm_offset_t base, offset;
3318 base = va & PG_FRAME;
3319 offset = va & PAGE_MASK;
3320 size = roundup(offset + size, PAGE_SIZE);
3321 kmem_free(kernel_map, base, size);
3325 * perform the pmap work for mincore
3328 pmap_mincore(pmap, addr)
3333 unsigned *ptep, pte;
3337 ptep = pmap_pte(pmap, addr);
3342 if ((pte = *ptep) != 0) {
3345 val = MINCORE_INCORE;
3346 if ((pte & PG_MANAGED) == 0)
3349 pa = pte & PG_FRAME;
3351 m = PHYS_TO_VM_PAGE(pa);
3357 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3359 * Modified by someone
3361 else if (m->dirty || pmap_is_modified(m))
3362 val |= MINCORE_MODIFIED_OTHER;
3367 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3370 * Referenced by someone
3372 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
3373 val |= MINCORE_REFERENCED_OTHER;
3374 vm_page_flag_set(m, PG_REFERENCED);
3381 pmap_activate(struct proc *p)
3385 pmap = vmspace_pmap(p->p_vmspace);
3387 pmap->pm_active |= 1 << cpuid;
3389 pmap->pm_active |= 1;
3391 #if defined(SWTCH_OPTIM_STATS)
3394 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3395 load_cr3(p->p_thread->td_pcb->pcb_cr3);
3399 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3402 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3406 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3411 #if defined(PMAP_DEBUG)
3412 pmap_pid_dump(int pid)
3418 LIST_FOREACH(p, &allproc, p_list) {
3419 if (p->p_pid != pid)
3425 pmap = vmspace_pmap(p->p_vmspace);
3426 for(i=0;i<1024;i++) {
3429 unsigned base = i << PDRSHIFT;
3431 pde = &pmap->pm_pdir[i];
3432 if (pde && pmap_pde_v(pde)) {
3433 for(j=0;j<1024;j++) {
3434 unsigned va = base + (j << PAGE_SHIFT);
3435 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) {
3442 pte = pmap_pte_quick( pmap, va);
3443 if (pte && pmap_pte_v(pte)) {
3447 m = PHYS_TO_VM_PAGE(pa);
3448 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x",
3449 va, pa, m->hold_count, m->wire_count, m->flags);
3470 static void pads __P((pmap_t pm));
3471 void pmap_pvdump __P((vm_offset_t pa));
3473 /* print address space of pmap*/
3481 if (pm == kernel_pmap)
3483 for (i = 0; i < 1024; i++)
3485 for (j = 0; j < 1024; j++) {
3486 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
3487 if (pm == kernel_pmap && va < KERNBASE)
3489 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS)
3491 ptep = pmap_pte_quick(pm, va);
3492 if (pmap_pte_v(ptep))
3493 printf("%x:%x ", va, *(int *) ptep);
3502 register pv_entry_t pv;
3505 printf("pa %x", pa);
3506 m = PHYS_TO_VM_PAGE(pa);
3507 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3509 printf(" -> pmap %p, va %x, flags %x",
3510 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
3512 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);