2 * Copyright (c) 2006 The DragonFly Project. All rights reserved.
3 * Copyright (c) 1991 Regents of the University of California.
5 * Copyright (c) 1994 John S. Dyson
7 * Copyright (c) 1994 David Greenman
9 * Copyright (c) 2004-2006 Matthew Dillon
10 * All rights reserved.
12 * Redistribution and use in source and binary forms, with or without
13 * 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
20 * the documentation and/or other materials provided with the
22 * 3. Neither the name of The DragonFly Project nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific, prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
28 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
29 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
30 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
31 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
32 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
33 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
34 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
35 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
36 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91
40 * $FreeBSD: src/sys/i386/i386/pmap.c,v 1.250.2.18 2002/03/06 22:48:53 silby Exp $
41 * $DragonFly: src/sys/platform/vkernel/platform/pmap.c,v 1.5 2007/01/07 08:37:37 dillon Exp $
44 #include <sys/types.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
49 #include <sys/vkernel.h>
51 #include <sys/thread.h>
53 #include <sys/vmspace.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_extern.h>
58 #include <vm/vm_kern.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_zone.h>
61 #include <vm/vm_pageout.h>
63 #include <machine/md_var.h>
64 #include <machine/pcb.h>
65 #include <machine/pmap_inval.h>
66 #include <machine/globaldata.h>
70 struct pmap kernel_pmap;
72 static struct vm_zone pvzone;
73 static struct vm_object pvzone_obj;
74 static TAILQ_HEAD(,pmap) pmap_list = TAILQ_HEAD_INITIALIZER(pmap_list);
75 static int pv_entry_count;
76 static int pv_entry_max;
77 static int pv_entry_high_water;
78 static int pmap_pagedaemon_waken;
79 static boolean_t pmap_initialized = FALSE;
80 static int protection_codes[8];
82 static void i386_protection_init(void);
83 static void pmap_remove_all(vm_page_t m);
84 static int pmap_release_free_page(struct pmap *pmap, vm_page_t p);
87 #ifndef PMAP_SHPGPERPROC
88 #define PMAP_SHPGPERPROC 200
91 #define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
93 #define pte_prot(m, p) \
94 (protection_codes[p & (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)])
100 struct pv_entry *pvinit;
102 for (i = 0; i < vm_page_array_size; i++) {
105 m = &vm_page_array[i];
106 TAILQ_INIT(&m->md.pv_list);
107 m->md.pv_list_count = 0;
110 i = vm_page_array_size;
113 pvinit = (struct pv_entry *)kmem_alloc(&kernel_map, i*sizeof(*pvinit));
114 zbootinit(&pvzone, "PV ENTRY", sizeof(*pvinit), pvinit, i);
115 pmap_initialized = TRUE;
121 int shpgperproc = PMAP_SHPGPERPROC;
123 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
124 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
125 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
126 pv_entry_high_water = 9 * (pv_entry_max / 10);
127 zinitna(&pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1);
131 * Bootstrap the kernel_pmap so it can be used with pmap_enter().
133 * NOTE! pm_pdir for the kernel pmap is offset so VA's translate
134 * directly into PTD indexes (PTA is also offset for the same reason).
135 * This is necessary because, for now, KVA is not mapped at address 0.
137 * Page table pages are not managed like they are in normal pmaps, so
138 * no pteobj is needed.
143 vm_pindex_t i = (vm_offset_t)KernelPTD >> PAGE_SHIFT;
145 kernel_pmap.pm_pdir = KernelPTD - (KvaStart >> SEG_SHIFT);
146 kernel_pmap.pm_pdirpte = KernelPTA[i];
147 kernel_pmap.pm_count = 1;
148 kernel_pmap.pm_active = (cpumask_t)-1;
149 TAILQ_INIT(&kernel_pmap.pm_pvlist);
150 i386_protection_init();
154 * Initialize pmap0/vmspace0 . Since process 0 never enters user mode we
155 * just dummy it up so it works well enough for fork().
157 * In DragonFly, process pmaps may only be used to manipulate user address
158 * space, never kernel address space.
161 pmap_pinit0(struct pmap *pmap)
166 /************************************************************************
167 * Procedures to manage whole physical maps *
168 ************************************************************************
170 * Initialize a preallocated and zeroed pmap structure,
171 * such as one in a vmspace structure.
174 pmap_pinit(struct pmap *pmap)
180 * No need to allocate page table space yet but we do need a valid
181 * page directory table.
183 if (pmap->pm_pdir == NULL) {
185 (pd_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
189 * allocate object for the pte array and page directory
191 npages = VPTE_PAGETABLE_SIZE +
192 (VM_MAX_USER_ADDRESS / PAGE_SIZE) * sizeof(vpte_t);
193 npages = (npages + PAGE_MASK) / PAGE_SIZE;
195 if (pmap->pm_pteobj == NULL)
196 pmap->pm_pteobj = vm_object_allocate(OBJT_DEFAULT, npages);
197 pmap->pm_pdindex = npages - 1;
200 * allocate the page directory page
202 ptdpg = vm_page_grab(pmap->pm_pteobj, pmap->pm_pdindex,
203 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
205 ptdpg->wire_count = 1;
206 ++vmstats.v_wire_count;
208 /* not usually mapped */
209 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY);
210 ptdpg->valid = VM_PAGE_BITS_ALL;
212 pmap_kenter((vm_offset_t)pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
213 pmap->pm_pdirpte = KernelPTA[(vm_offset_t)pmap->pm_pdir >> PAGE_SHIFT];
214 if ((ptdpg->flags & PG_ZERO) == 0)
215 bzero(pmap->pm_pdir, PAGE_SIZE);
219 pmap->pm_ptphint = NULL;
220 TAILQ_INIT(&pmap->pm_pvlist);
221 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
225 * Wire in kernel global address entries. To avoid a race condition
226 * between pmap initialization and pmap_growkernel, this procedure
227 * adds the pmap to the master list (which growkernel scans to update),
228 * then copies the template.
230 * In a virtual kernel there are no kernel global address entries.
233 pmap_pinit2(struct pmap *pmap)
236 TAILQ_INSERT_TAIL(&pmap_list, pmap, pm_pmnode);
241 * Release all resources held by the given physical map.
243 * Should only be called if the map contains no valid mappings.
245 static int pmap_release_callback(struct vm_page *p, void *data);
248 pmap_release(struct pmap *pmap)
250 vm_object_t object = pmap->pm_pteobj;
251 struct rb_vm_page_scan_info info;
253 KKASSERT(pmap != &kernel_pmap);
255 #if defined(DIAGNOSTIC)
256 if (object->ref_count != 1)
257 panic("pmap_release: pteobj reference count != 1");
261 info.object = object;
263 TAILQ_REMOVE(&pmap_list, pmap, pm_pmnode);
270 info.limit = object->generation;
272 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
273 pmap_release_callback, &info);
274 if (info.error == 0 && info.mpte) {
275 if (!pmap_release_free_page(pmap, info.mpte))
279 } while (info.error);
283 pmap_release_callback(struct vm_page *p, void *data)
285 struct rb_vm_page_scan_info *info = data;
287 if (p->pindex == info->pmap->pm_pdindex) {
291 if (!pmap_release_free_page(info->pmap, p)) {
295 if (info->object->generation != info->limit) {
303 * Retire the given physical map from service. Should only be called if
304 * the map contains no valid mappings.
307 pmap_destroy(pmap_t pmap)
314 count = --pmap->pm_count;
317 panic("destroying a pmap is not yet implemented");
322 * Add a reference to the specified pmap.
325 pmap_reference(pmap_t pmap)
332 /************************************************************************
333 * VMSPACE MANAGEMENT *
334 ************************************************************************
336 * The VMSPACE management we do in our virtual kernel must be reflected
337 * in the real kernel. This is accomplished by making vmspace system
338 * calls to the real kernel.
341 cpu_vmspace_alloc(struct vmspace *vm)
346 #define LAST_EXTENT (VM_MAX_USER_ADDRESS - 0x80000000)
348 if (vmspace_create(vm, 0, NULL) < 0)
349 panic("vmspace_create() failed");
351 rp = vmspace_mmap(vm, (void *)0x00000000, 0x40000000,
352 PROT_READ|PROT_WRITE,
353 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED,
355 if (rp == MAP_FAILED)
356 panic("vmspace_mmap: failed1");
357 rp = vmspace_mmap(vm, (void *)0x40000000, 0x40000000,
358 PROT_READ|PROT_WRITE,
359 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED,
360 MemImageFd, 0x40000000);
361 if (rp == MAP_FAILED)
362 panic("vmspace_mmap: failed2");
363 rp = vmspace_mmap(vm, (void *)0x80000000, LAST_EXTENT,
364 PROT_READ|PROT_WRITE,
365 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED,
366 MemImageFd, 0x80000000);
367 if (rp == MAP_FAILED)
368 panic("vmspace_mmap: failed3");
370 r = vmspace_mcontrol(vm, (void *)0x00000000, 0x40000000, MADV_SETMAP,
371 vmspace_pmap(vm)->pm_pdirpte);
373 panic("vmspace_mcontrol: failed1");
374 r = vmspace_mcontrol(vm, (void *)0x40000000, 0x40000000, MADV_SETMAP,
375 vmspace_pmap(vm)->pm_pdirpte);
377 panic("vmspace_mcontrol: failed2");
378 r = vmspace_mcontrol(vm, (void *)0x80000000, LAST_EXTENT, MADV_SETMAP,
379 vmspace_pmap(vm)->pm_pdirpte);
381 panic("vmspace_mcontrol: failed3");
385 cpu_vmspace_free(struct vmspace *vm)
387 if (vmspace_destroy(vm) < 0)
388 panic("vmspace_destroy() failed");
391 /************************************************************************
392 * Procedures which operate directly on the kernel PMAP *
393 ************************************************************************/
396 * This maps the requested page table and gives us access to it.
399 get_ptbase(struct pmap *pmap, vm_offset_t va)
401 struct mdglobaldata *gd = mdcpu;
403 if (pmap == &kernel_pmap) {
404 KKASSERT(va >= KvaStart && va < KvaEnd);
405 return(KernelPTA + (va >> PAGE_SHIFT));
406 } else if (pmap->pm_pdir == gd->gd_PT1pdir) {
407 return(gd->gd_PT1map + (va >> PAGE_SHIFT));
408 } else if (pmap->pm_pdir == gd->gd_PT2pdir) {
409 return(gd->gd_PT2map + (va >> PAGE_SHIFT));
413 * Otherwise choose one or the other and map the page table
414 * in the KVA space reserved for it.
416 KKASSERT(gd->mi.gd_intr_nesting_level == 0 &&
417 (gd->mi.gd_curthread->td_flags & TDF_INTTHREAD) == 0);
419 if ((gd->gd_PTflip = 1 - gd->gd_PTflip) == 0) {
420 gd->gd_PT1pdir = pmap->pm_pdir;
421 *gd->gd_PT1pde = pmap->pm_pdirpte;
422 madvise(gd->gd_PT1map, SEG_SIZE, MADV_INVAL);
423 return(gd->gd_PT1map + (va >> PAGE_SHIFT));
425 gd->gd_PT2pdir = pmap->pm_pdir;
426 *gd->gd_PT2pde = pmap->pm_pdirpte;
427 madvise(gd->gd_PT2map, SEG_SIZE, MADV_INVAL);
428 return(gd->gd_PT2map + (va >> PAGE_SHIFT));
433 get_ptbase1(struct pmap *pmap, vm_offset_t va)
435 struct mdglobaldata *gd = mdcpu;
437 if (pmap == &kernel_pmap) {
438 KKASSERT(va >= KvaStart && va < KvaEnd);
439 return(KernelPTA + (va >> PAGE_SHIFT));
440 } else if (pmap->pm_pdir == gd->gd_PT1pdir) {
441 return(gd->gd_PT1map + (va >> PAGE_SHIFT));
443 KKASSERT(gd->mi.gd_intr_nesting_level == 0 &&
444 (gd->mi.gd_curthread->td_flags & TDF_INTTHREAD) == 0);
445 gd->gd_PT1pdir = pmap->pm_pdir;
446 *gd->gd_PT1pde = pmap->pm_pdirpte;
447 madvise(gd->gd_PT1map, SEG_SIZE, MADV_INVAL);
448 return(gd->gd_PT1map + (va >> PAGE_SHIFT));
452 get_ptbase2(struct pmap *pmap, vm_offset_t va)
454 struct mdglobaldata *gd = mdcpu;
456 if (pmap == &kernel_pmap) {
457 KKASSERT(va >= KvaStart && va < KvaEnd);
458 return(KernelPTA + (va >> PAGE_SHIFT));
459 } else if (pmap->pm_pdir == gd->gd_PT2pdir) {
460 return(gd->gd_PT2map + (va >> PAGE_SHIFT));
462 KKASSERT(gd->mi.gd_intr_nesting_level == 0 &&
463 (gd->mi.gd_curthread->td_flags & TDF_INTTHREAD) == 0);
464 gd->gd_PT2pdir = pmap->pm_pdir;
465 *gd->gd_PT2pde = pmap->pm_pdirpte;
466 madvise(gd->gd_PT2map, SEG_SIZE, MADV_INVAL);
467 return(gd->gd_PT2map + (va >> PAGE_SHIFT));
471 * Return a pointer to the page table entry for the specified va in the
472 * specified pmap. NULL is returned if there is no valid page table page
475 static __inline vpte_t *
476 pmap_pte(struct pmap *pmap, vm_offset_t va)
480 ptep = &pmap->pm_pdir[va >> SEG_SHIFT];
484 return (get_ptbase(pmap, va));
490 * Enter a mapping into kernel_pmap. Mappings created in this fashion
494 pmap_kenter(vm_offset_t va, vm_paddr_t pa)
499 pmap_inval_info info;
502 KKASSERT(va >= KvaStart && va < KvaEnd);
503 npte = (vpte_t)pa | VPTE_R | VPTE_W | VPTE_V;
504 ptep = KernelPTA + (va >> PAGE_SHIFT);
505 if (*ptep & VPTE_V) {
507 pmap_inval_init(&info);
508 pmap_inval_add(&info, &kernel_pmap, va);
512 pmap_inval_flush(&info);
514 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
522 pmap_kenter_sync(vm_offset_t va)
524 pmap_inval_info info;
526 pmap_inval_init(&info);
527 pmap_inval_add(&info, &kernel_pmap, va);
528 pmap_inval_flush(&info);
532 pmap_kenter_sync_quick(vm_offset_t va)
534 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
538 * XXX these need to be recoded. They are not used in any critical path.
541 pmap_kmodify_rw(vm_offset_t va)
543 *pmap_kpte(va) |= VPTE_R | VPTE_W;
544 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
548 pmap_kmodify_nc(vm_offset_t va)
551 *pmap_kpte(va) |= VPTE_N;
552 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
557 * Map a contiguous range of physical memory to a KVM
560 pmap_map(vm_offset_t virt, vm_paddr_t start, vm_paddr_t end, int prot)
562 while (start < end) {
563 pmap_kenter(virt, start);
571 pmap_kpte(vm_offset_t va)
575 KKASSERT(va >= KvaStart && va < KvaEnd);
576 ptep = KernelPTA + (va >> PAGE_SHIFT);
581 * Enter a mapping into kernel_pmap without any SMP interactions.
583 * Mappings created in this fashion are not managed.
586 pmap_kenter_quick(vm_offset_t va, vm_paddr_t pa)
591 KKASSERT(va >= KvaStart && va < KvaEnd);
593 npte = (vpte_t)pa | VPTE_R | VPTE_W | VPTE_V;
594 ptep = KernelPTA + (va >> PAGE_SHIFT);
595 if (*ptep & VPTE_V) {
597 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
604 * Make a temporary mapping for a physical address. This is only intended
605 * to be used for panic dumps.
608 pmap_kenter_temporary(vm_paddr_t pa, int i)
610 pmap_kenter(crashdumpmap + (i * PAGE_SIZE), pa);
611 return ((void *)crashdumpmap);
615 * Remove an unmanaged mapping created with pmap_kenter*().
618 pmap_kremove(vm_offset_t va)
622 pmap_inval_info info;
625 KKASSERT(va >= KvaStart && va < KvaEnd);
627 ptep = KernelPTA + (va >> PAGE_SHIFT);
628 if (*ptep & VPTE_V) {
630 pmap_inval_init(&info);
631 pmap_inval_add(&info, &kernel_pmap, va);
635 pmap_inval_flush(&info);
637 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
646 * Remove an unmanaged mapping created with pmap_kenter*() without
647 * going through any SMP interactions.
650 pmap_kremove_quick(vm_offset_t va)
654 KKASSERT(va >= KvaStart && va < KvaEnd);
656 ptep = KernelPTA + (va >> PAGE_SHIFT);
657 if (*ptep & VPTE_V) {
659 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
666 * Extract the physical address from the kernel_pmap that is associated
667 * with the specified virtual address.
670 pmap_kextract(vm_offset_t va)
675 KKASSERT(va >= KvaStart && va < KvaEnd);
677 ptep = KernelPTA + (va >> PAGE_SHIFT);
678 pa = (vm_paddr_t)(*ptep & VPTE_FRAME) | (va & PAGE_MASK);
683 * Map a set of unmanaged VM pages into KVM.
686 pmap_qenter(vm_offset_t va, struct vm_page **m, int count)
688 KKASSERT(va >= KvaStart && va + count * PAGE_SIZE < KvaEnd);
692 ptep = KernelPTA + (va >> PAGE_SHIFT);
694 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
695 *ptep = (vpte_t)(*m)->phys_addr | VPTE_R | VPTE_W | VPTE_V;
707 * Map a set of VM pages to kernel virtual memory. If a mapping changes
708 * clear the supplied mask. The caller handles any SMP interactions.
709 * The mask is used to provide the caller with hints on what SMP interactions
713 pmap_qenter2(vm_offset_t va, struct vm_page **m, int count, cpumask_t *mask)
715 cpumask_t cmask = mycpu->gd_cpumask;
717 KKASSERT(va >= KvaStart && va + count * PAGE_SIZE < KvaEnd);
722 ptep = KernelPTA + (va >> PAGE_SHIFT);
723 npte = (vpte_t)(*m)->phys_addr | VPTE_R | VPTE_W | VPTE_V;
727 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
728 } else if ((*mask & cmask) == 0) {
729 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
739 * Undo the effects of pmap_qenter*().
742 pmap_qremove(vm_offset_t va, int count)
744 KKASSERT(va >= KvaStart && va + count * PAGE_SIZE < KvaEnd);
748 ptep = KernelPTA + (va >> PAGE_SHIFT);
750 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
761 /************************************************************************
762 * Misc support glue called by machine independant code *
763 ************************************************************************
765 * These routines are called by machine independant code to operate on
766 * certain machine-dependant aspects of processes, threads, and pmaps.
770 * Initialize MD portions of the thread structure.
773 pmap_init_thread(thread_t td)
775 /* enforce pcb placement */
776 td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_size) - 1;
777 td->td_savefpu = &td->td_pcb->pcb_save;
778 td->td_sp = (char *)td->td_pcb - 16;
782 * Initialize MD portions of a process structure. XXX this aint MD
785 pmap_init_proc(struct proc *p, struct thread *td)
787 p->p_addr = (void *)td->td_kstack;
790 td->td_lwp = &p->p_lwp;
791 td->td_switch = cpu_heavy_switch;
793 KKASSERT(td->td_mpcount == 1);
795 bzero(p->p_addr, sizeof(*p->p_addr));
799 * Destroy the UPAGES for a process that has exited and disassociate
800 * the process from its thread.
803 pmap_dispose_proc(struct proc *p)
807 KASSERT(p->p_lock == 0, ("attempt to dispose referenced proc! %p", p));
809 if ((td = p->p_thread) != NULL) {
818 * We pre-allocate all page table pages for kernel virtual memory so
819 * this routine will only be called if KVM has been exhausted.
822 pmap_growkernel(vm_offset_t size)
824 panic("KVM exhausted");
828 * The modification bit is not tracked for any pages in this range. XXX
829 * such pages in this maps should always use pmap_k*() functions and not
833 pmap_track_modified(vm_offset_t va)
835 if ((va < clean_sva) || (va >= clean_eva))
841 /************************************************************************
842 * Procedures supporting managed page table pages *
843 ************************************************************************
845 * These procedures are used to track managed page table pages. These pages
846 * use the page table page's vm_page_t to track PTEs in the page. The
847 * page table pages themselves are arranged in a VM object, pmap->pm_pteobj.
849 * This allows the system to throw away page table pages for user processes
850 * at will and reinstantiate them on demand.
854 * This routine works like vm_page_lookup() but also blocks as long as the
855 * page is busy. This routine does not busy the page it returns.
857 * Unless the caller is managing objects whos pages are in a known state,
858 * the call should be made with a critical section held so the page's object
859 * association remains valid on return.
862 pmap_page_lookup(vm_object_t object, vm_pindex_t pindex)
867 m = vm_page_lookup(object, pindex);
868 if (m && vm_page_sleep_busy(m, FALSE, "pplookp"))
874 * This routine unholds page table pages, and if the hold count
875 * drops to zero, then it decrements the wire count.
878 _pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
880 pmap_inval_flush(info);
881 while (vm_page_sleep_busy(m, FALSE, "pmuwpt"))
884 if (m->hold_count == 0) {
886 * unmap the page table page
888 pmap_inval_add(info, pmap, -1);
889 pmap->pm_pdir[m->pindex] = 0;
890 --pmap->pm_stats.resident_count;
892 if (pmap->pm_ptphint == m)
893 pmap->pm_ptphint = NULL;
896 * If the page is finally unwired, simply free it.
899 if (m->wire_count == 0) {
902 vm_page_free_zero(m);
903 --vmstats.v_wire_count;
911 pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
914 if (m->hold_count == 0)
915 return _pmap_unwire_pte_hold(pmap, m, info);
921 * After removing a page table entry, this routine is used to
922 * conditionally free the page, and manage the hold/wire counts.
925 pmap_unuse_pt(pmap_t pmap, vm_offset_t va, vm_page_t mpte,
926 pmap_inval_info_t info)
932 * page table pages in the kernel_pmap are not managed.
934 if (pmap == &kernel_pmap)
936 ptepindex = (va >> PDRSHIFT);
937 if (pmap->pm_ptphint &&
938 (pmap->pm_ptphint->pindex == ptepindex)) {
939 mpte = pmap->pm_ptphint;
941 pmap_inval_flush(info);
942 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
943 pmap->pm_ptphint = mpte;
946 return pmap_unwire_pte_hold(pmap, mpte, info);
950 * Attempt to release and free an vm_page in a pmap. Returns 1 on success,
951 * 0 on failure (if the procedure had to sleep).
954 pmap_release_free_page(struct pmap *pmap, vm_page_t p)
956 vpte_t *pde = pmap->pm_pdir;
958 * This code optimizes the case of freeing non-busy
959 * page-table pages. Those pages are zero now, and
960 * might as well be placed directly into the zero queue.
962 if (vm_page_sleep_busy(p, FALSE, "pmaprl"))
968 * Remove the page table page from the processes address space.
971 pmap->pm_stats.resident_count--;
974 panic("pmap_release: freeing held page table page");
977 * Page directory pages need to have the kernel stuff cleared, so
978 * they can go into the zero queue also.
980 * In virtual kernels there is no 'kernel stuff'. For the moment
981 * I just make sure the whole thing has been zero'd even though
982 * it should already be completely zero'd.
984 if (p->pindex == pmap->pm_pdindex) {
985 bzero(pde, VPTE_PAGETABLE_SIZE);
986 pmap_kremove((vm_offset_t)pmap->pm_pdir);
990 * Clear the matching hint
992 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
993 pmap->pm_ptphint = NULL;
996 * And throw the page away. The page is completely zero'd out so
997 * optimize the free call.
1000 vmstats.v_wire_count--;
1001 vm_page_free_zero(p);
1006 * This routine is called if the page table page is not mapped in the page
1009 * The routine is broken up into two parts for readability.
1012 _pmap_allocpte(pmap_t pmap, unsigned ptepindex)
1018 * Find or fabricate a new pagetable page
1020 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1021 VM_ALLOC_NORMAL | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1023 KASSERT(m->queue == PQ_NONE,
1024 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1026 if (m->wire_count == 0)
1027 vmstats.v_wire_count++;
1031 * Increment the hold count for the page table page
1032 * (denoting a new mapping.)
1037 * Map the pagetable page into the process address space, if
1038 * it isn't already there.
1040 pmap->pm_stats.resident_count++;
1042 ptepa = VM_PAGE_TO_PHYS(m);
1043 pmap->pm_pdir[ptepindex] = (vpte_t)ptepa | VPTE_R | VPTE_W | VPTE_V |
1047 * We are likely about to access this page table page, so set the
1048 * page table hint to reduce overhead.
1050 pmap->pm_ptphint = m;
1053 * Try to use the new mapping, but if we cannot, then
1054 * do it with the routine that maps the page explicitly.
1056 if ((m->flags & PG_ZERO) == 0)
1057 pmap_zero_page(ptepa);
1059 m->valid = VM_PAGE_BITS_ALL;
1060 vm_page_flag_clear(m, PG_ZERO);
1061 vm_page_flag_set(m, PG_MAPPED);
1068 * Determine the page table page required to access the VA in the pmap
1069 * and allocate it if necessary. Return a held vm_page_t for the page.
1071 * Only used with user pmaps.
1074 pmap_allocpte(pmap_t pmap, vm_offset_t va)
1081 * Calculate pagetable page index
1083 ptepindex = va >> PDRSHIFT;
1086 * Get the page directory entry
1088 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1091 * This supports switching from a 4MB page to a
1094 if (ptepa & VPTE_PS) {
1095 pmap->pm_pdir[ptepindex] = 0;
1102 * If the page table page is mapped, we just increment the
1103 * hold count, and activate it.
1107 * In order to get the page table page, try the
1110 if (pmap->pm_ptphint &&
1111 (pmap->pm_ptphint->pindex == ptepindex)) {
1112 m = pmap->pm_ptphint;
1114 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1115 pmap->pm_ptphint = m;
1121 * Here if the pte page isn't mapped, or if it has been deallocated.
1123 return _pmap_allocpte(pmap, ptepindex);
1126 /************************************************************************
1127 * Managed pages in pmaps *
1128 ************************************************************************
1130 * All pages entered into user pmaps and some pages entered into the kernel
1131 * pmap are managed, meaning that pmap_protect() and other related management
1132 * functions work on these pages.
1136 * free the pv_entry back to the free list. This function may be
1137 * called from an interrupt.
1139 static __inline void
1140 free_pv_entry(pv_entry_t pv)
1147 * get a new pv_entry, allocating a block from the system
1148 * when needed. This function may be called from an interrupt.
1154 if (pv_entry_high_water &&
1155 (pv_entry_count > pv_entry_high_water) &&
1156 (pmap_pagedaemon_waken == 0)) {
1157 pmap_pagedaemon_waken = 1;
1158 wakeup (&vm_pages_needed);
1160 return zalloc(&pvzone);
1164 * This routine is very drastic, but can save the system
1172 static int warningdone=0;
1174 if (pmap_pagedaemon_waken == 0)
1177 if (warningdone < 5) {
1178 kprintf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n");
1182 for(i = 0; i < vm_page_array_size; i++) {
1183 m = &vm_page_array[i];
1184 if (m->wire_count || m->hold_count || m->busy ||
1185 (m->flags & PG_BUSY))
1189 pmap_pagedaemon_waken = 0;
1193 * If it is the first entry on the list, it is actually
1194 * in the header and we must copy the following entry up
1195 * to the header. Otherwise we must search the list for
1196 * the entry. In either case we free the now unused entry.
1199 pmap_remove_entry(struct pmap *pmap, vm_page_t m,
1200 vm_offset_t va, pmap_inval_info_t info)
1206 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1207 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1208 if (pmap == pv->pv_pmap && va == pv->pv_va)
1212 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1213 if (va == pv->pv_va)
1219 * Note that pv_ptem is NULL if the page table page itself is not
1220 * managed, even if the page being removed IS managed.
1224 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem, info);
1225 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1226 m->md.pv_list_count--;
1227 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
1228 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1229 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1237 * Create a pv entry for page at pa for (pmap, va). If the page table page
1238 * holding the VA is managed, mpte will be non-NULL.
1241 pmap_insert_entry(pmap_t pmap, vm_offset_t va, vm_page_t mpte, vm_page_t m)
1246 pv = get_pv_entry();
1251 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1252 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1253 m->md.pv_list_count++;
1259 * pmap_remove_pte: do the things to unmap a page in a process
1262 pmap_remove_pte(struct pmap *pmap, vpte_t *ptq, vm_offset_t va,
1263 pmap_inval_info_t info)
1268 pmap_inval_add(info, pmap, va);
1269 oldpte = loadandclear(ptq);
1270 if (oldpte & VPTE_W)
1271 pmap->pm_stats.wired_count -= 1;
1273 * Machines that don't support invlpg, also don't support
1274 * VPTE_G. XXX VPTE_G is disabled for SMP so don't worry about
1277 if (oldpte & VPTE_G)
1278 madvise((void *)va, PAGE_SIZE, MADV_INVAL);
1279 pmap->pm_stats.resident_count -= 1;
1280 if (oldpte & PG_MANAGED) {
1281 m = PHYS_TO_VM_PAGE(oldpte);
1282 if (oldpte & VPTE_M) {
1283 #if defined(PMAP_DIAGNOSTIC)
1284 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1286 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n",
1290 if (pmap_track_modified(va))
1293 if (oldpte & VPTE_A)
1294 vm_page_flag_set(m, PG_REFERENCED);
1295 return pmap_remove_entry(pmap, m, va, info);
1297 return pmap_unuse_pt(pmap, va, NULL, info);
1306 * Remove a single page from a process address space.
1308 * This function may not be called from an interrupt if the pmap is
1312 pmap_remove_page(struct pmap *pmap, vm_offset_t va, pmap_inval_info_t info)
1317 * if there is no pte for this address, just skip it!!! Otherwise
1318 * get a local va for mappings for this pmap and remove the entry.
1320 if (*pmap_pde(pmap, va) != 0) {
1321 ptq = get_ptbase(pmap, va);
1323 pmap_remove_pte(pmap, ptq, va, info);
1331 * Remove the given range of addresses from the specified map.
1333 * It is assumed that the start and end are properly
1334 * rounded to the page size.
1336 * This function may not be called from an interrupt if the pmap is
1340 pmap_remove(struct pmap *pmap, vm_offset_t sva, vm_offset_t eva)
1344 vm_offset_t ptpaddr;
1345 vm_pindex_t sindex, eindex;
1347 struct pmap_inval_info info;
1352 if (pmap->pm_stats.resident_count == 0)
1355 pmap_inval_init(&info);
1358 * special handling of removing one page. a very
1359 * common operation and easy to short circuit some
1362 if (((sva + PAGE_SIZE) == eva) &&
1363 ((pmap->pm_pdir[(sva >> PDRSHIFT)] & VPTE_PS) == 0)) {
1364 pmap_remove_page(pmap, sva, &info);
1365 pmap_inval_flush(&info);
1370 * Get a local virtual address for the mappings that are being
1373 * XXX this is really messy because the kernel pmap is not relative
1376 ptbase = get_ptbase(pmap, sva);
1378 sindex = (sva >> PAGE_SHIFT);
1379 eindex = (eva >> PAGE_SHIFT);
1382 for (; sindex < eindex; sindex = pdnxt) {
1386 * Calculate index for next page table.
1388 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1389 if (pmap->pm_stats.resident_count == 0)
1392 pdirindex = sindex / NPDEPG;
1393 if (((ptpaddr = pmap->pm_pdir[pdirindex]) & VPTE_PS) != 0) {
1394 pmap_inval_add(&info, pmap, -1);
1395 pmap->pm_pdir[pdirindex] = 0;
1396 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1401 * Weed out invalid mappings. Note: we assume that the page
1402 * directory table is always allocated, and in kernel virtual.
1408 * Limit our scan to either the end of the va represented
1409 * by the current page table page, or to the end of the
1410 * range being removed.
1412 if (pdnxt > eindex) {
1416 for (; sindex != pdnxt; sindex++) {
1418 if (ptbase[sindex - sbase] == 0)
1420 va = i386_ptob(sindex);
1421 if (pmap_remove_pte(pmap, ptbase + sindex - sbase, va, &info))
1425 pmap_inval_flush(&info);
1431 * Removes this physical page from all physical maps in which it resides.
1432 * Reflects back modify bits to the pager.
1434 * This routine may not be called from an interrupt.
1437 pmap_remove_all(vm_page_t m)
1439 struct pmap_inval_info info;
1443 #if defined(PMAP_DIAGNOSTIC)
1445 * XXX this makes pmap_page_protect(NONE) illegal for non-managed
1448 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) {
1449 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%08llx", (long long)VM_PAGE_TO_PHYS(m));
1453 pmap_inval_init(&info);
1455 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
1456 pv->pv_pmap->pm_stats.resident_count--;
1458 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
1459 pmap_inval_add(&info, pv->pv_pmap, pv->pv_va);
1461 tpte = loadandclear(pte);
1463 pv->pv_pmap->pm_stats.wired_count--;
1466 vm_page_flag_set(m, PG_REFERENCED);
1469 * Update the vm_page_t clean and reference bits.
1471 if (tpte & VPTE_M) {
1472 #if defined(PMAP_DIAGNOSTIC)
1473 if (pmap_nw_modified((pt_entry_t) tpte)) {
1475 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n",
1479 if (pmap_track_modified(pv->pv_va))
1482 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
1483 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1484 m->md.pv_list_count--;
1485 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem, &info);
1489 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1491 pmap_inval_flush(&info);
1497 * Set the physical protection on the specified range of this map
1500 * This function may not be called from an interrupt if the map is
1501 * not the kernel_pmap.
1504 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
1507 vm_offset_t pdnxt, ptpaddr;
1508 vm_pindex_t sindex, eindex;
1510 pmap_inval_info info;
1515 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1516 pmap_remove(pmap, sva, eva);
1520 if (prot & VM_PROT_WRITE)
1523 pmap_inval_init(&info);
1525 ptbase = get_ptbase(pmap, sva);
1527 sindex = (sva >> PAGE_SHIFT);
1528 eindex = (eva >> PAGE_SHIFT);
1531 for (; sindex < eindex; sindex = pdnxt) {
1535 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1537 pdirindex = sindex / NPDEPG;
1538 if (((ptpaddr = pmap->pm_pdir[pdirindex]) & VPTE_PS) != 0) {
1539 pmap_inval_add(&info, pmap, -1);
1540 pmap->pm_pdir[pdirindex] &= ~(VPTE_M|VPTE_W);
1541 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
1546 * Weed out invalid mappings. Note: we assume that the page
1547 * directory table is always allocated, and in kernel virtual.
1552 if (pdnxt > eindex) {
1556 for (; sindex != pdnxt; sindex++) {
1561 /* XXX this isn't optimal */
1562 pmap_inval_add(&info, pmap, i386_ptob(sindex));
1563 pbits = ptbase[sindex - sbase];
1565 if (pbits & PG_MANAGED) {
1567 if (pbits & VPTE_A) {
1568 m = PHYS_TO_VM_PAGE(pbits);
1569 vm_page_flag_set(m, PG_REFERENCED);
1572 if (pbits & VPTE_M) {
1573 if (pmap_track_modified(i386_ptob(sindex))) {
1575 m = PHYS_TO_VM_PAGE(pbits);
1584 if (pbits != ptbase[sindex - sbase]) {
1585 ptbase[sindex - sbase] = pbits;
1589 pmap_inval_flush(&info);
1593 * Enter a managed page into a pmap. If the page is not wired related pmap
1594 * data can be destroyed at any time for later demand-operation.
1596 * Insert the vm_page (m) at virtual address (v) in (pmap), with the
1597 * specified protection, and wire the mapping if requested.
1599 * NOTE: This routine may not lazy-evaluate or lose information. The
1600 * page must actually be inserted into the given map NOW.
1602 * NOTE: When entering a page at a KVA address, the pmap must be the
1606 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
1612 vm_offset_t origpte, newpte;
1614 pmap_inval_info info;
1622 * Get the page table page. The kernel_pmap's page table pages
1623 * are preallocated and have no associated vm_page_t.
1625 if (pmap == &kernel_pmap)
1628 mpte = pmap_allocpte(pmap, va);
1630 pmap_inval_init(&info);
1631 pte = pmap_pte(pmap, va);
1634 * Page Directory table entry not valid, we need a new PT page
1635 * and pmap_allocpte() didn't give us one. Oops!
1638 panic("pmap_enter: invalid page directory pmap=%p, va=0x%p\n",
1642 pa = VM_PAGE_TO_PHYS(m) & VPTE_FRAME;
1643 pmap_inval_add(&info, pmap, va); /* XXX non-optimal */
1645 opa = origpte & VPTE_FRAME;
1647 if (origpte & VPTE_PS)
1648 panic("pmap_enter: attempted pmap_enter on 4MB page");
1651 * Mapping has not changed, must be protection or wiring change.
1653 if (origpte && (opa == pa)) {
1655 * Wiring change, just update stats. We don't worry about
1656 * wiring PT pages as they remain resident as long as there
1657 * are valid mappings in them. Hence, if a user page is wired,
1658 * the PT page will be also.
1660 if (wired && ((origpte & VPTE_W) == 0))
1661 pmap->pm_stats.wired_count++;
1662 else if (!wired && (origpte & VPTE_W))
1663 pmap->pm_stats.wired_count--;
1665 #if defined(PMAP_DIAGNOSTIC)
1666 if (pmap_nw_modified((pt_entry_t) origpte)) {
1668 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n",
1674 * Remove the extra pte reference. Note that we cannot
1675 * optimize the RO->RW case because we have adjusted the
1676 * wiring count above and may need to adjust the wiring
1683 * We might be turning off write access to the page,
1684 * so we go ahead and sense modify status.
1686 if (origpte & PG_MANAGED) {
1687 if ((origpte & VPTE_M) && pmap_track_modified(va)) {
1689 om = PHYS_TO_VM_PAGE(opa);
1697 * Mapping has changed, invalidate old range and fall through to
1698 * handle validating new mapping.
1702 err = pmap_remove_pte(pmap, pte, va, &info);
1704 panic("pmap_enter: pte vanished, va: 0x%x", va);
1708 * Enter on the PV list if part of our managed memory. Note that we
1709 * raise IPL while manipulating pv_table since pmap_enter can be
1710 * called at interrupt time.
1712 if (pmap_initialized &&
1713 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
1714 pmap_insert_entry(pmap, va, mpte, m);
1719 * Increment counters
1721 pmap->pm_stats.resident_count++;
1723 pmap->pm_stats.wired_count++;
1727 * Now validate mapping with desired protection/wiring.
1729 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | VPTE_V);
1736 * if the mapping or permission bits are different, we need
1737 * to update the pte.
1739 if ((origpte & ~(VPTE_M|VPTE_A)) != newpte) {
1740 *pte = newpte | VPTE_A;
1742 pmap_inval_flush(&info);
1746 * This is a quick version of pmap_enter(). It is used only under the
1747 * following conditions:
1749 * (1) The pmap is not the kernel_pmap
1750 * (2) The page is not to be wired into the map
1751 * (3) The page is to mapped read-only in the pmap (initially that is)
1752 * (4) The calling procedure is responsible for flushing the TLB
1753 * (5) The page is always managed
1754 * (6) There is no prior mapping at the VA
1758 pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_page_t mpte)
1762 pmap_inval_info info;
1766 KKASSERT(pmap != &kernel_pmap);
1767 pmap_inval_init(&info);
1769 KKASSERT(va >= VM_MIN_USER_ADDRESS && va < VM_MAX_USER_ADDRESS);
1772 * Instantiate the page table page if required
1776 * Calculate pagetable page index
1778 ptepindex = va >> PDRSHIFT;
1779 if (mpte && (mpte->pindex == ptepindex)) {
1784 * Get the page directory entry
1786 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1789 * If the page table page is mapped, we just increment
1790 * the hold count, and activate it.
1793 if (ptepa & VPTE_PS)
1794 panic("pmap_enter_quick: unexpected mapping into 4MB page");
1795 if (pmap->pm_ptphint &&
1796 (pmap->pm_ptphint->pindex == ptepindex)) {
1797 mpte = pmap->pm_ptphint;
1799 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex);
1800 pmap->pm_ptphint = mpte;
1806 mpte = _pmap_allocpte(pmap, ptepindex);
1811 * Ok, now that the page table page has been validated, get the pte.
1812 * If the pte is already mapped undo mpte's hold_count and
1815 pte = pmap_pte(pmap, va);
1818 pmap_unwire_pte_hold(pmap, mpte, &info);
1823 * Enter on the PV list if part of our managed memory. Note that we
1824 * raise IPL while manipulating pv_table since pmap_enter can be
1825 * called at interrupt time.
1827 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
1828 pmap_insert_entry(pmap, va, mpte, m);
1831 * Increment counters
1833 pmap->pm_stats.resident_count++;
1835 pa = VM_PAGE_TO_PHYS(m);
1838 * Now validate mapping with RO protection
1840 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
1841 *pte = pa | VPTE_V | VPTE_U;
1843 *pte = pa | VPTE_V | VPTE_U | VPTE_MANAGED;
1849 pmap_extract(pmap_t pmap, vm_offset_t va)
1854 if (pmap && (pte = pmap->pm_pdir[va >> SEG_SHIFT]) != 0) {
1855 if (pte & VPTE_PS) {
1856 rtval = pte & ~((vpte_t)(1 << SEG_SHIFT) - 1);
1857 rtval |= va & SEG_MASK;
1859 pte = *get_ptbase(pmap, va);
1860 rtval = (pte & VPTE_FRAME) | (va & PAGE_MASK);
1867 #define MAX_INIT_PT (96)
1870 * This routine preloads the ptes for a given object into the specified pmap.
1871 * This eliminates the blast of soft faults on process startup and
1872 * immediately after an mmap.
1874 static int pmap_object_init_pt_callback(vm_page_t p, void *data);
1877 pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_prot_t prot,
1878 vm_object_t object, vm_pindex_t pindex,
1879 vm_size_t size, int limit)
1881 struct rb_vm_page_scan_info info;
1885 * We can't preinit if read access isn't set or there is no pmap
1888 if ((prot & VM_PROT_READ) == 0 || pmap == NULL || object == NULL)
1892 * We can't preinit if the pmap is not the current pmap
1894 if (curproc == NULL || pmap != vmspace_pmap(curproc->p_vmspace))
1897 psize = size >> PAGE_SHIFT;
1899 if ((object->type != OBJT_VNODE) ||
1900 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
1901 (object->resident_page_count > MAX_INIT_PT))) {
1905 if (psize + pindex > object->size) {
1906 if (object->size < pindex)
1908 psize = object->size - pindex;
1915 * Use a red-black scan to traverse the requested range and load
1916 * any valid pages found into the pmap.
1918 * We cannot safely scan the object's memq unless we are in a
1919 * critical section since interrupts can remove pages from objects.
1921 info.start_pindex = pindex;
1922 info.end_pindex = pindex + psize - 1;
1929 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
1930 pmap_object_init_pt_callback, &info);
1936 pmap_object_init_pt_callback(vm_page_t p, void *data)
1938 struct rb_vm_page_scan_info *info = data;
1939 vm_pindex_t rel_index;
1941 * don't allow an madvise to blow away our really
1942 * free pages allocating pv entries.
1944 if ((info->limit & MAP_PREFAULT_MADVISE) &&
1945 vmstats.v_free_count < vmstats.v_free_reserved) {
1948 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
1949 (p->busy == 0) && (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
1950 if ((p->queue - p->pc) == PQ_CACHE)
1951 vm_page_deactivate(p);
1953 rel_index = p->pindex - info->start_pindex;
1954 info->mpte = pmap_enter_quick(info->pmap,
1955 info->addr + i386_ptob(rel_index),
1957 vm_page_flag_set(p, PG_MAPPED);
1964 * pmap_prefault provides a quick way of clustering pagefaults into a
1965 * processes address space. It is a "cousin" of pmap_object_init_pt,
1966 * except it runs at page fault time instead of mmap time.
1970 #define PAGEORDER_SIZE (PFBAK+PFFOR)
1972 static int pmap_prefault_pageorder[] = {
1973 -PAGE_SIZE, PAGE_SIZE,
1974 -2 * PAGE_SIZE, 2 * PAGE_SIZE,
1975 -3 * PAGE_SIZE, 3 * PAGE_SIZE,
1976 -4 * PAGE_SIZE, 4 * PAGE_SIZE
1980 pmap_prefault(pmap_t pmap, vm_offset_t addra, vm_map_entry_t entry)
1990 * We do not currently prefault mappings that use virtual page
1991 * tables. We do not prefault foreign pmaps.
1993 if (entry->maptype == VM_MAPTYPE_VPAGETABLE)
1995 if (curproc == NULL || (pmap != vmspace_pmap(curproc->p_vmspace)))
1998 object = entry->object.vm_object;
2000 starta = addra - PFBAK * PAGE_SIZE;
2001 if (starta < entry->start)
2002 starta = entry->start;
2003 else if (starta > addra)
2007 * critical section protection is required to maintain the
2008 * page/object association, interrupts can free pages and remove
2009 * them from their objects.
2013 for (i = 0; i < PAGEORDER_SIZE; i++) {
2014 vm_object_t lobject;
2017 addr = addra + pmap_prefault_pageorder[i];
2018 if (addr > addra + (PFFOR * PAGE_SIZE))
2021 if (addr < starta || addr >= entry->end)
2025 * Make sure the page table page already exists
2027 if ((*pmap_pde(pmap, addr)) == NULL)
2031 * Get a pointer to the pte and make sure that no valid page
2034 pte = get_ptbase(pmap, addr);
2039 * Get the page to be mapped
2041 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
2044 for (m = vm_page_lookup(lobject, pindex);
2045 (!m && (lobject->type == OBJT_DEFAULT) &&
2046 (lobject->backing_object));
2047 lobject = lobject->backing_object
2049 if (lobject->backing_object_offset & PAGE_MASK)
2051 pindex += (lobject->backing_object_offset >> PAGE_SHIFT);
2052 m = vm_page_lookup(lobject->backing_object, pindex);
2056 * give-up when a page is not in memory
2062 * If everything meets the requirements for pmap_enter_quick(),
2063 * then enter the page.
2066 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2068 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
2070 if ((m->queue - m->pc) == PQ_CACHE) {
2071 vm_page_deactivate(m);
2074 mpte = pmap_enter_quick(pmap, addr, m, mpte);
2075 vm_page_flag_set(m, PG_MAPPED);
2083 * Routine: pmap_change_wiring
2084 * Function: Change the wiring attribute for a map/virtual-address
2086 * In/out conditions:
2087 * The mapping must already exist in the pmap.
2090 pmap_change_wiring(pmap_t pmap, vm_offset_t va, boolean_t wired)
2097 pte = get_ptbase(pmap, va);
2099 if (wired && (*pte & VPTE_W) == 0)
2100 pmap->pm_stats.wired_count++;
2101 else if (!wired && (*pte & VPTE_W))
2102 pmap->pm_stats.wired_count--;
2105 * Wiring is not a hardware characteristic so there is no need to
2106 * invalidate TLB. However, in an SMP environment we must use
2107 * a locked bus cycle to update the pte (if we are not using
2108 * the pmap_inval_*() API that is)... it's ok to do this for simple
2113 atomic_set_int(pte, VPTE_W);
2115 atomic_clear_int(pte, VPTE_W);
2118 atomic_set_int_nonlocked(pte, VPTE_W);
2120 atomic_clear_int_nonlocked(pte, VPTE_W);
2125 * Copy the range specified by src_addr/len
2126 * from the source map to the range dst_addr/len
2127 * in the destination map.
2129 * This routine is only advisory and need not do anything.
2132 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
2133 vm_size_t len, vm_offset_t src_addr)
2135 pmap_inval_info info;
2137 vm_offset_t end_addr = src_addr + len;
2143 if (dst_addr != src_addr)
2145 if (dst_pmap->pm_pdir == NULL)
2147 if (src_pmap->pm_pdir == NULL)
2150 src_frame = get_ptbase1(src_pmap, src_addr);
2151 dst_frame = get_ptbase2(dst_pmap, src_addr);
2153 pmap_inval_init(&info);
2154 pmap_inval_add(&info, dst_pmap, -1);
2155 pmap_inval_add(&info, src_pmap, -1);
2158 * critical section protection is required to maintain the page/object
2159 * association, interrupts can free pages and remove them from
2163 for (addr = src_addr; addr < end_addr; addr = pdnxt) {
2164 vpte_t *src_pte, *dst_pte;
2165 vm_page_t dstmpte, srcmpte;
2166 vm_offset_t srcptepaddr;
2169 if (addr >= VM_MAX_USER_ADDRESS)
2170 panic("pmap_copy: invalid to pmap_copy page tables\n");
2173 * Don't let optional prefaulting of pages make us go
2174 * way below the low water mark of free pages or way
2175 * above high water mark of used pv entries.
2177 if (vmstats.v_free_count < vmstats.v_free_reserved ||
2178 pv_entry_count > pv_entry_high_water)
2181 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1));
2182 ptepindex = addr >> PDRSHIFT;
2184 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex];
2185 if (srcptepaddr == 0)
2188 if (srcptepaddr & VPTE_PS) {
2189 if (dst_pmap->pm_pdir[ptepindex] == 0) {
2190 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr;
2191 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE;
2196 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex);
2197 if ((srcmpte == NULL) ||
2198 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY))
2201 if (pdnxt > end_addr)
2204 src_pte = src_frame + ((addr - src_addr) >> PAGE_SHIFT);
2205 dst_pte = dst_frame + ((addr - src_addr) >> PAGE_SHIFT);
2206 while (addr < pdnxt) {
2210 * we only virtual copy managed pages
2212 if ((ptetemp & PG_MANAGED) != 0) {
2214 * We have to check after allocpte for the
2215 * pte still being around... allocpte can
2218 dstmpte = pmap_allocpte(dst_pmap, addr);
2219 if ((*dst_pte == 0) && (ptetemp = *src_pte)) {
2221 * Clear the modified and
2222 * accessed (referenced) bits
2225 m = PHYS_TO_VM_PAGE(ptetemp);
2226 *dst_pte = ptetemp & ~(VPTE_M | VPTE_A);
2227 dst_pmap->pm_stats.resident_count++;
2228 pmap_insert_entry(dst_pmap, addr,
2231 pmap_unwire_pte_hold(dst_pmap, dstmpte, &info);
2233 if (dstmpte->hold_count >= srcmpte->hold_count)
2242 pmap_inval_flush(&info);
2248 * Zero the specified PA by mapping the page into KVM and clearing its
2251 * This function may be called from an interrupt and no locking is
2255 pmap_zero_page(vm_paddr_t phys)
2257 struct mdglobaldata *gd = mdcpu;
2261 panic("pmap_zero_page: CMAP3 busy");
2262 *gd->gd_CMAP3 = VPTE_V | VPTE_W | (phys & VPTE_FRAME) | VPTE_A | VPTE_M;
2263 madvise(gd->gd_CADDR3, PAGE_SIZE, MADV_INVAL);
2265 bzero(gd->gd_CADDR3, PAGE_SIZE);
2271 * pmap_page_assertzero:
2273 * Assert that a page is empty, panic if it isn't.
2276 pmap_page_assertzero(vm_paddr_t phys)
2278 struct mdglobaldata *gd = mdcpu;
2283 panic("pmap_zero_page: CMAP3 busy");
2284 *gd->gd_CMAP3 = VPTE_V | VPTE_R | VPTE_W |
2285 (phys & VPTE_FRAME) | VPTE_A | VPTE_M;
2286 madvise(gd->gd_CADDR3, PAGE_SIZE, MADV_INVAL);
2287 for (i = 0; i < PAGE_SIZE; i += 4) {
2288 if (*(int *)((char *)gd->gd_CADDR3 + i) != 0) {
2289 panic("pmap_page_assertzero() @ %p not zero!\n",
2290 (void *)gd->gd_CADDR3);
2300 * Zero part of a physical page by mapping it into memory and clearing
2301 * its contents with bzero.
2303 * off and size may not cover an area beyond a single hardware page.
2306 pmap_zero_page_area(vm_paddr_t phys, int off, int size)
2308 struct mdglobaldata *gd = mdcpu;
2312 panic("pmap_zero_page: CMAP3 busy");
2313 *gd->gd_CMAP3 = VPTE_V | VPTE_R | VPTE_W |
2314 (phys & VPTE_FRAME) | VPTE_A | VPTE_M;
2315 madvise(gd->gd_CADDR3, PAGE_SIZE, MADV_INVAL);
2317 bzero((char *)gd->gd_CADDR3 + off, size);
2325 * Copy the physical page from the source PA to the target PA.
2326 * This function may be called from an interrupt. No locking
2330 pmap_copy_page(vm_paddr_t src, vm_paddr_t dst)
2332 struct mdglobaldata *gd = mdcpu;
2335 if (*(int *) gd->gd_CMAP1)
2336 panic("pmap_copy_page: CMAP1 busy");
2337 if (*(int *) gd->gd_CMAP2)
2338 panic("pmap_copy_page: CMAP2 busy");
2340 *(int *) gd->gd_CMAP1 = VPTE_V | (src & PG_FRAME) | PG_A;
2341 *(int *) gd->gd_CMAP2 = VPTE_V | VPTE_R | VPTE_W | (dst & VPTE_FRAME) | VPTE_A | VPTE_M;
2343 madvise(gd->gd_CADDR1, PAGE_SIZE, MADV_INVAL);
2344 madvise(gd->gd_CADDR2, PAGE_SIZE, MADV_INVAL);
2346 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
2348 *(int *) gd->gd_CMAP1 = 0;
2349 *(int *) gd->gd_CMAP2 = 0;
2354 * pmap_copy_page_frag:
2356 * Copy the physical page from the source PA to the target PA.
2357 * This function may be called from an interrupt. No locking
2361 pmap_copy_page_frag(vm_paddr_t src, vm_paddr_t dst, size_t bytes)
2363 struct mdglobaldata *gd = mdcpu;
2366 if (*(int *) gd->gd_CMAP1)
2367 panic("pmap_copy_page: CMAP1 busy");
2368 if (*(int *) gd->gd_CMAP2)
2369 panic("pmap_copy_page: CMAP2 busy");
2371 *(int *) gd->gd_CMAP1 = VPTE_V | (src & VPTE_FRAME) | VPTE_A;
2372 *(int *) gd->gd_CMAP2 = VPTE_V | VPTE_R | VPTE_W | (dst & VPTE_FRAME) | VPTE_A | VPTE_M;
2374 madvise(gd->gd_CADDR1, PAGE_SIZE, MADV_INVAL);
2375 madvise(gd->gd_CADDR2, PAGE_SIZE, MADV_INVAL);
2377 bcopy((char *)gd->gd_CADDR1 + (src & PAGE_MASK),
2378 (char *)gd->gd_CADDR2 + (dst & PAGE_MASK),
2381 *(int *) gd->gd_CMAP1 = 0;
2382 *(int *) gd->gd_CMAP2 = 0;
2387 * Returns true if the pmap's pv is one of the first
2388 * 16 pvs linked to from this page. This count may
2389 * be changed upwards or downwards in the future; it
2390 * is only necessary that true be returned for a small
2391 * subset of pmaps for proper page aging.
2394 pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
2399 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2404 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2405 if (pv->pv_pmap == pmap) {
2418 * Remove all pages from specified address space
2419 * this aids process exit speeds. Also, this code
2420 * is special cased for current process only, but
2421 * can have the more generic (and slightly slower)
2422 * mode enabled. This is much faster than pmap_remove
2423 * in the case of running down an entire address space.
2426 pmap_remove_pages(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2431 pmap_inval_info info;
2434 if (curproc && pmap == vmspace_pmap(curproc->p_vmspace))
2439 pmap_inval_init(&info);
2441 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); pv; pv = npv) {
2442 if (pv->pv_va >= eva || pv->pv_va < sva) {
2443 npv = TAILQ_NEXT(pv, pv_plist);
2447 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
2448 if (pmap->pm_active)
2449 pmap_inval_add(&info, pv->pv_pmap, pv->pv_va);
2453 * We cannot remove wired pages from a process' mapping
2456 if (tpte & VPTE_W) {
2457 npv = TAILQ_NEXT(pv, pv_plist);
2462 m = PHYS_TO_VM_PAGE(tpte);
2464 KASSERT(m < &vm_page_array[vm_page_array_size],
2465 ("pmap_remove_pages: bad tpte %x", tpte));
2467 pv->pv_pmap->pm_stats.resident_count--;
2470 * Update the vm_page_t clean and reference bits.
2472 if (tpte & VPTE_M) {
2477 npv = TAILQ_NEXT(pv, pv_plist);
2478 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist);
2480 m->md.pv_list_count--;
2481 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2482 if (TAILQ_FIRST(&m->md.pv_list) == NULL) {
2483 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2486 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem, &info);
2489 pmap_inval_flush(&info);
2494 * pmap_testbit tests bits in pte's
2495 * note that the testbit/changebit routines are inline,
2496 * and a lot of things compile-time evaluate.
2499 pmap_testbit(vm_page_t m, int bit)
2504 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2507 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
2512 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2514 * if the bit being tested is the modified bit, then
2515 * mark clean_map and ptes as never
2518 if (bit & (VPTE_A|VPTE_M)) {
2519 if (!pmap_track_modified(pv->pv_va))
2523 #if defined(PMAP_DIAGNOSTIC)
2525 kprintf("Null pmap (tb) at va: 0x%x\n", pv->pv_va);
2529 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
2540 * this routine is used to modify bits in ptes
2542 static __inline void
2543 pmap_changebit(vm_page_t m, int bit, boolean_t setem)
2545 struct pmap_inval_info info;
2549 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2552 pmap_inval_init(&info);
2556 * Loop over all current mappings setting/clearing as appropos If
2557 * setting RO do we need to clear the VAC?
2559 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2561 * don't write protect pager mappings
2563 if (!setem && (bit == VPTE_W)) {
2564 if (!pmap_track_modified(pv->pv_va))
2568 #if defined(PMAP_DIAGNOSTIC)
2570 kprintf("Null pmap (cb) at va: 0x%x\n", pv->pv_va);
2576 * Careful here. We can use a locked bus instruction to
2577 * clear VPTE_A or VPTE_M safely but we need to synchronize
2578 * with the target cpus when we mess with VPTE_W.
2580 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
2582 pmap_inval_add(&info, pv->pv_pmap, pv->pv_va);
2586 atomic_set_int(pte, bit);
2588 atomic_set_int_nonlocked(pte, bit);
2591 vm_offset_t pbits = *(vm_offset_t *)pte;
2593 if (bit == VPTE_W) {
2594 if (pbits & VPTE_M) {
2598 atomic_clear_int(pte, VPTE_M|VPTE_W);
2600 atomic_clear_int_nonlocked(pte, VPTE_M|VPTE_W);
2604 atomic_clear_int(pte, bit);
2606 atomic_clear_int_nonlocked(pte, bit);
2612 pmap_inval_flush(&info);
2617 * pmap_page_protect:
2619 * Lower the permission for all mappings to a given page.
2622 pmap_page_protect(vm_page_t m, vm_prot_t prot)
2624 if ((prot & VM_PROT_WRITE) == 0) {
2625 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
2626 pmap_changebit(m, VPTE_W, FALSE);
2634 pmap_phys_address(int ppn)
2636 return (i386_ptob(ppn));
2640 * pmap_ts_referenced:
2642 * Return a count of reference bits for a page, clearing those bits.
2643 * It is not necessary for every reference bit to be cleared, but it
2644 * is necessary that 0 only be returned when there are truly no
2645 * reference bits set.
2647 * XXX: The exact number of bits to check and clear is a matter that
2648 * should be tested and standardized at some point in the future for
2649 * optimal aging of shared pages.
2652 pmap_ts_referenced(vm_page_t m)
2654 pv_entry_t pv, pvf, pvn;
2658 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2663 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
2668 pvn = TAILQ_NEXT(pv, pv_list);
2670 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2672 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
2674 if (!pmap_track_modified(pv->pv_va))
2677 pte = pmap_pte(pv->pv_pmap, pv->pv_va);
2679 if (pte && (*pte & VPTE_A)) {
2681 atomic_clear_int(pte, VPTE_A);
2683 atomic_clear_int_nonlocked(pte, VPTE_A);
2690 } while ((pv = pvn) != NULL && pv != pvf);
2700 * Return whether or not the specified physical page was modified
2701 * in any physical maps.
2704 pmap_is_modified(vm_page_t m)
2706 return pmap_testbit(m, VPTE_M);
2710 * Clear the modify bits on the specified physical page.
2713 pmap_clear_modify(vm_page_t m)
2715 pmap_changebit(m, VPTE_M, FALSE);
2719 * pmap_clear_reference:
2721 * Clear the reference bit on the specified physical page.
2724 pmap_clear_reference(vm_page_t m)
2726 pmap_changebit(m, VPTE_A, FALSE);
2730 * Miscellaneous support routines follow
2734 i386_protection_init(void)
2738 kp = protection_codes;
2739 for (prot = 0; prot < 8; prot++) {
2740 if (prot & VM_PROT_READ)
2742 if (prot & VM_PROT_WRITE)
2744 if (prot & VM_PROT_EXECUTE)
2751 * Map a set of physical memory pages into the kernel virtual
2752 * address space. Return a pointer to where it is mapped. This
2753 * routine is intended to be used for mapping device memory,
2756 * NOTE: we can't use pgeflag unless we invalidate the pages one at
2760 pmap_mapdev(vm_paddr_t pa, vm_size_t size)
2762 vm_offset_t va, tmpva, offset;
2765 offset = pa & PAGE_MASK;
2766 size = roundup(offset + size, PAGE_SIZE);
2768 va = kmem_alloc_nofault(&kernel_map, size);
2770 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
2772 pa = pa & VPTE_FRAME;
2773 for (tmpva = va; size > 0;) {
2774 pte = KernelPTA + (tmpva >> PAGE_SHIFT);
2775 *pte = pa | VPTE_R | VPTE_W | VPTE_V; /* | pgeflag; */
2783 return ((void *)(va + offset));
2787 pmap_unmapdev(vm_offset_t va, vm_size_t size)
2789 vm_offset_t base, offset;
2791 base = va & VPTE_FRAME;
2792 offset = va & PAGE_MASK;
2793 size = roundup(offset + size, PAGE_SIZE);
2794 pmap_qremove(va, size >> PAGE_SHIFT);
2795 kmem_free(&kernel_map, base, size);
2799 * perform the pmap work for mincore
2802 pmap_mincore(pmap_t pmap, vm_offset_t addr)
2808 ptep = pmap_pte(pmap, addr);
2813 if ((pte = *ptep) != 0) {
2816 val = MINCORE_INCORE;
2817 if ((pte & VPTE_MANAGED) == 0)
2820 pa = pte & VPTE_FRAME;
2822 m = PHYS_TO_VM_PAGE(pa);
2828 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
2830 * Modified by someone
2832 else if (m->dirty || pmap_is_modified(m))
2833 val |= MINCORE_MODIFIED_OTHER;
2838 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
2841 * Referenced by someone
2843 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) {
2844 val |= MINCORE_REFERENCED_OTHER;
2845 vm_page_flag_set(m, PG_REFERENCED);
2852 pmap_activate(struct proc *p)
2856 pmap = vmspace_pmap(p->p_vmspace);
2858 atomic_set_int(&pmap->pm_active, 1 << mycpu->gd_cpuid);
2860 pmap->pm_active |= 1;
2862 #if defined(SWTCH_OPTIM_STATS)
2865 panic("pmap_activate"); /* XXX store vmspace id in context */
2867 p->p_thread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
2868 load_cr3(p->p_thread->td_pcb->pcb_cr3);
2873 pmap_deactivate(struct proc *p)
2877 pmap = vmspace_pmap(p->p_vmspace);
2879 atomic_clear_int(&pmap->pm_active, 1 << mycpu->gd_cpuid);
2881 pmap->pm_active &= ~1;
2884 * XXX - note we do not adjust %cr3. The caller is expected to
2885 * activate a new pmap or do a thread-exit.
2890 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
2893 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
2897 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
2904 static void pads (pmap_t pm);
2905 void pmap_pvdump (vm_paddr_t pa);
2907 /* print address space of pmap*/
2915 if (pm == &kernel_pmap)
2917 for (i = 0; i < 1024; i++)
2919 for (j = 0; j < 1024; j++) {
2920 va = (i << PDRSHIFT) + (j << PAGE_SHIFT);
2921 if (pm == &kernel_pmap && va < KERNBASE)
2923 if (pm != &kernel_pmap && va > UPT_MAX_ADDRESS)
2925 ptep = pmap_pte(pm, va);
2926 if (ptep && (*ptep & VPTE_V)) {
2928 (void *)va, (unsigned)*ptep);
2935 pmap_pvdump(vm_paddr_t pa)
2940 kprintf("pa %08llx", (long long)pa);
2941 m = PHYS_TO_VM_PAGE(pa);
2942 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2944 kprintf(" -> pmap %p, va %x, flags %x",
2945 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags);
2947 kprintf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va);