4 * Copyright (c) 1991 Regents of the University of California.
6 * Copyright (c) 1994 John S. Dyson
8 * Copyright (c) 1994 David Greenman
11 * This code is derived from software contributed to Berkeley by
12 * the Systems Programming Group of the University of Utah Computer
13 * Science Department and William Jolitz of UUNET Technologies Inc.
15 * Redistribution and use in source and binary forms, with or without
16 * modification, are permitted provided that the following conditions
18 * 1. Redistributions of source code must retain the above copyright
19 * notice, this list of conditions and the following disclaimer.
20 * 2. Redistributions in binary form must reproduce the above copyright
21 * notice, this list of conditions and the following disclaimer in the
22 * documentation and/or other materials provided with the distribution.
23 * 3. All advertising materials mentioning features or use of this software
24 * must display the following acknowledgement:
25 * This product includes software developed by the University of
26 * California, Berkeley and its contributors.
27 * 4. Neither the name of the University nor the names of its contributors
28 * may be used to endorse or promote products derived from this software
29 * without specific prior written permission.
31 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91
44 * $FreeBSD: src/sys/i386/i386/pmap.c,v 1.250.2.18 2002/03/06 22:48:53 silby Exp $
48 * Manages physical address maps.
50 * In most cases we hold page table pages busy in order to manipulate them.
53 * PMAP_DEBUG - see platform/pc32/include/pmap.h
56 #include "opt_disable_pse.h"
58 #include "opt_msgbuf.h"
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
64 #include <sys/msgbuf.h>
65 #include <sys/vmmeter.h>
67 #include <sys/thread.h>
70 #include <vm/vm_param.h>
71 #include <sys/sysctl.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vm_zone.h>
83 #include <sys/thread2.h>
84 #include <sys/sysref2.h>
85 #include <sys/spinlock2.h>
86 #include <vm/vm_page2.h>
88 #include <machine/cputypes.h>
89 #include <machine/md_var.h>
90 #include <machine/specialreg.h>
91 #include <machine/smp.h>
92 #include <machine_base/apic/apicreg.h>
93 #include <machine/globaldata.h>
94 #include <machine/pmap.h>
95 #include <machine/pmap_inval.h>
97 #define PMAP_KEEP_PDIRS
98 #ifndef PMAP_SHPGPERPROC
99 #define PMAP_SHPGPERPROC 200
100 #define PMAP_PVLIMIT 1400000 /* i386 kvm problems */
103 #if defined(DIAGNOSTIC)
104 #define PMAP_DIAGNOSTIC
109 #if !defined(PMAP_DIAGNOSTIC)
110 #define PMAP_INLINE __inline
116 * Get PDEs and PTEs for user/kernel address space
118 #define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT]))
119 #define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT])
121 #define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0)
122 #define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0)
123 #define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0)
124 #define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0)
125 #define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0)
128 * Given a map and a machine independent protection code,
129 * convert to a vax protection code.
131 #define pte_prot(m, p) \
132 (protection_codes[p & (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)])
133 static int protection_codes[8];
135 struct pmap kernel_pmap;
136 static TAILQ_HEAD(,pmap) pmap_list = TAILQ_HEAD_INITIALIZER(pmap_list);
138 vm_paddr_t avail_start; /* PA of first available physical page */
139 vm_paddr_t avail_end; /* PA of last available physical page */
140 vm_offset_t virtual_start; /* VA of first avail page (after kernel bss) */
141 vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */
142 vm_offset_t virtual2_start;
143 vm_offset_t virtual2_end;
144 vm_offset_t KvaStart; /* VA start of KVA space */
145 vm_offset_t KvaEnd; /* VA end of KVA space (non-inclusive) */
146 vm_offset_t KvaSize; /* max size of kernel virtual address space */
147 static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
148 static int pgeflag; /* PG_G or-in */
149 static int pseflag; /* PG_PS or-in */
151 static vm_object_t kptobj;
154 vm_offset_t kernel_vm_end;
157 * Data for the pv entry allocation mechanism
159 static vm_zone_t pvzone;
160 static struct vm_zone pvzone_store;
161 static struct vm_object pvzone_obj;
162 static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
163 static int pmap_pagedaemon_waken = 0;
164 static struct pv_entry *pvinit;
167 * Considering all the issues I'm having with pmap caching, if breakage
168 * continues to occur, and for debugging, I've added a sysctl that will
169 * just do an unconditional invltlb.
171 static int dreadful_invltlb;
173 SYSCTL_INT(_vm, OID_AUTO, dreadful_invltlb,
174 CTLFLAG_RW, &dreadful_invltlb, 0, "Debugging sysctl to force invltlb on pmap operations");
177 * All those kernel PT submaps that BSD is so fond of
179 pt_entry_t *CMAP1 = NULL, *ptmmap;
180 caddr_t CADDR1 = NULL, ptvmmap = NULL;
181 static pt_entry_t *msgbufmap;
182 struct msgbuf *msgbufp=NULL;
187 static pt_entry_t *pt_crashdumpmap;
188 static caddr_t crashdumpmap;
190 extern pt_entry_t *SMPpt;
192 static PMAP_INLINE void free_pv_entry (pv_entry_t pv);
193 static unsigned * get_ptbase (pmap_t pmap);
194 static pv_entry_t get_pv_entry (void);
195 static void i386_protection_init (void);
196 static __inline void pmap_clearbit (vm_page_t m, int bit);
198 static void pmap_remove_all (vm_page_t m);
199 static void pmap_remove_pte (struct pmap *pmap, unsigned *ptq,
200 vm_offset_t sva, pmap_inval_info_t info);
201 static void pmap_remove_page (struct pmap *pmap,
202 vm_offset_t va, pmap_inval_info_t info);
203 static void pmap_remove_entry (struct pmap *pmap, vm_page_t m,
204 vm_offset_t va, pmap_inval_info_t info);
205 static boolean_t pmap_testbit (vm_page_t m, int bit);
206 static void pmap_insert_entry (pmap_t pmap, pv_entry_t pv,
207 vm_offset_t va, vm_page_t mpte, vm_page_t m);
209 static vm_page_t pmap_allocpte (pmap_t pmap, vm_offset_t va);
211 static int pmap_release_free_page (pmap_t pmap, vm_page_t p);
212 static vm_page_t _pmap_allocpte (pmap_t pmap, unsigned ptepindex);
213 static unsigned * pmap_pte_quick (pmap_t pmap, vm_offset_t va);
214 static vm_page_t pmap_page_lookup (vm_object_t object, vm_pindex_t pindex);
215 static void pmap_unuse_pt (pmap_t, vm_offset_t, vm_page_t, pmap_inval_info_t);
216 static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
218 static void pmap_hold(pmap_t pmap);
219 static void pmap_drop(pmap_t pmap);
220 static void pmap_wait(pmap_t pmap, int count);
222 static unsigned pdir4mb;
225 * Move the kernel virtual free pointer to the next
226 * 4MB. This is used to help improve performance
227 * by using a large (4MB) page for much of the kernel
228 * (.text, .data, .bss)
232 pmap_kmem_choose(vm_offset_t addr)
234 vm_offset_t newaddr = addr;
236 if (cpu_feature & CPUID_PSE) {
237 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
244 * This function returns a pointer to the pte entry in the pmap and has
245 * the side effect of potentially retaining a cached mapping of the pmap.
247 * The caller must hold vm_token and the returned value is only valid
248 * until the caller blocks or releases the token.
252 pmap_pte(pmap_t pmap, vm_offset_t va)
256 ASSERT_LWKT_TOKEN_HELD(&vm_token);
258 pdeaddr = (unsigned *) pmap_pde(pmap, va);
259 if (*pdeaddr & PG_PS)
262 return get_ptbase(pmap) + i386_btop(va);
268 * pmap_pte using the kernel_pmap
270 * Used for debugging, no requirements.
273 pmap_kernel_pte(vm_offset_t va)
277 pdeaddr = (unsigned *) pmap_pde(&kernel_pmap, va);
278 if (*pdeaddr & PG_PS)
281 return (unsigned *)vtopte(va);
288 * Super fast pmap_pte routine best used when scanning the pv lists.
289 * This eliminates many course-grained invltlb calls. Note that many of
290 * the pv list scans are across different pmaps and it is very wasteful
291 * to do an entire invltlb when checking a single mapping.
293 * Should only be called while in a critical section.
295 * The caller must hold vm_token and the returned value is only valid
296 * until the caller blocks or releases the token.
300 pmap_pte_quick(pmap_t pmap, vm_offset_t va)
302 struct mdglobaldata *gd = mdcpu;
305 ASSERT_LWKT_TOKEN_HELD(&vm_token);
306 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) {
307 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
308 unsigned index = i386_btop(va);
309 /* are we current address space or kernel? */
310 if ((pmap == &kernel_pmap) ||
311 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
312 return (unsigned *) PTmap + index;
314 newpf = pde & PG_FRAME;
315 if (((*(unsigned *)gd->gd_PMAP1) & PG_FRAME) != newpf) {
316 *(unsigned *)gd->gd_PMAP1 = newpf | PG_RW | PG_V;
317 cpu_invlpg(gd->gd_PADDR1);
319 return gd->gd_PADDR1 + (index & (NPTEPG - 1));
326 * Bootstrap the system enough to run with virtual memory.
328 * On the i386 this is called after mapping has already been enabled
329 * and just syncs the pmap module with what has already been done.
330 * [We can't call it easily with mapping off since the kernel is not
331 * mapped with PA == VA, hence we would have to relocate every address
332 * from the linked base (virtual) address "KERNBASE" to the actual
333 * (physical) address starting relative to 0]
336 pmap_bootstrap(vm_paddr_t firstaddr, vm_paddr_t loadaddr)
340 struct mdglobaldata *gd;
344 KvaStart = (vm_offset_t)VADDR(PTDPTDI, 0);
345 KvaSize = (vm_offset_t)VADDR(APTDPTDI, 0) - KvaStart;
346 KvaEnd = KvaStart + KvaSize;
348 avail_start = firstaddr;
351 * XXX The calculation of virtual_start is wrong. It's NKPT*PAGE_SIZE
352 * too large. It should instead be correctly calculated in locore.s and
353 * not based on 'first' (which is a physical address, not a virtual
354 * address, for the start of unused physical memory). The kernel
355 * page tables are NOT double mapped and thus should not be included
356 * in this calculation.
358 virtual_start = (vm_offset_t) KERNBASE + firstaddr;
359 virtual_start = pmap_kmem_choose(virtual_start);
360 virtual_end = VADDR(KPTDI+NKPDE-1, NPTEPG-1);
363 * Initialize protection array.
365 i386_protection_init();
368 * The kernel's pmap is statically allocated so we don't have to use
369 * pmap_create, which is unlikely to work correctly at this part of
370 * the boot sequence (XXX and which no longer exists).
372 * The kernel_pmap's pm_pteobj is used only for locking and not
375 kernel_pmap.pm_pdir = (pd_entry_t *)(KERNBASE + (u_int)IdlePTD);
376 kernel_pmap.pm_count = 1;
377 kernel_pmap.pm_active = (cpumask_t)-1 & ~CPUMASK_LOCK;
378 kernel_pmap.pm_pteobj = &kernel_object;
379 TAILQ_INIT(&kernel_pmap.pm_pvlist);
380 TAILQ_INIT(&kernel_pmap.pm_pvlist_free);
381 spin_init(&kernel_pmap.pm_spin);
382 lwkt_token_init(&kernel_pmap.pm_token, "kpmap_tok");
386 * Reserve some special page table entries/VA space for temporary
389 #define SYSMAP(c, p, v, n) \
390 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
393 pte = (pt_entry_t *) pmap_kernel_pte(va);
396 * CMAP1/CMAP2 are used for zeroing and copying pages.
398 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
403 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
406 * ptvmmap is used for reading arbitrary physical pages via
409 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
412 * msgbufp is used to map the system message buffer.
413 * XXX msgbufmap is not used.
415 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
416 atop(round_page(MSGBUF_SIZE)))
421 for (i = 0; i < NKPT; i++)
425 * PG_G is terribly broken on SMP because we IPI invltlb's in some
426 * cases rather then invl1pg. Actually, I don't even know why it
427 * works under UP because self-referential page table mappings
432 if (cpu_feature & CPUID_PGE)
437 * Initialize the 4MB page size flag
441 * The 4MB page version of the initial
442 * kernel page mapping.
446 #if !defined(DISABLE_PSE)
447 if (cpu_feature & CPUID_PSE) {
450 * Note that we have enabled PSE mode
453 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
454 ptditmp &= ~(NBPDR - 1);
455 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
460 * Enable the PSE mode. If we are SMP we can't do this
461 * now because the APs will not be able to use it when
464 load_cr4(rcr4() | CR4_PSE);
467 * We can do the mapping here for the single processor
468 * case. We simply ignore the old page table page from
472 * For SMP, we still need 4K pages to bootstrap APs,
473 * PSE will be enabled as soon as all APs are up.
475 PTD[KPTDI] = (pd_entry_t)ptditmp;
476 kernel_pmap.pm_pdir[KPTDI] = (pd_entry_t)ptditmp;
483 * We need to finish setting up the globaldata page for the BSP.
484 * locore has already populated the page table for the mdglobaldata
487 pg = MDGLOBALDATA_BASEALLOC_PAGES;
488 gd = &CPU_prvspace[0].mdglobaldata;
489 gd->gd_CMAP1 = &SMPpt[pg + 0];
490 gd->gd_CMAP2 = &SMPpt[pg + 1];
491 gd->gd_CMAP3 = &SMPpt[pg + 2];
492 gd->gd_PMAP1 = &SMPpt[pg + 3];
493 gd->gd_GDMAP1 = &PTD[APTDPTDI];
494 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
495 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
496 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
497 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
498 gd->gd_GDADDR1= (unsigned *)VADDR(APTDPTDI, 0);
505 * Set 4mb pdir for mp startup
510 if (pseflag && (cpu_feature & CPUID_PSE)) {
511 load_cr4(rcr4() | CR4_PSE);
512 if (pdir4mb && mycpu->gd_cpuid == 0) { /* only on BSP */
513 kernel_pmap.pm_pdir[KPTDI] =
514 PTD[KPTDI] = (pd_entry_t)pdir4mb;
522 * Initialize the pmap module, called by vm_init()
524 * Called from the low level boot code only.
533 * object for kernel page table pages
535 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
538 * Allocate memory for random pmap data structures. Includes the
542 for(i = 0; i < vm_page_array_size; i++) {
545 m = &vm_page_array[i];
546 TAILQ_INIT(&m->md.pv_list);
547 m->md.pv_list_count = 0;
551 * init the pv free list
553 initial_pvs = vm_page_array_size;
554 if (initial_pvs < MINPV)
556 pvzone = &pvzone_store;
557 pvinit = (void *)kmem_alloc(&kernel_map,
558 initial_pvs * sizeof (struct pv_entry));
559 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry),
560 pvinit, initial_pvs);
563 * Now it is safe to enable pv_table recording.
565 pmap_initialized = TRUE;
569 * Initialize the address space (zone) for the pv_entries. Set a
570 * high water mark so that the system can recover from excessive
571 * numbers of pv entries.
573 * Called from the low level boot code only.
578 int shpgperproc = PMAP_SHPGPERPROC;
581 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
582 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
586 * Horrible hack for systems with a lot of memory running i386.
587 * the calculated pv_entry_max can wind up eating a ton of KVM
588 * so put a cap on the number of entries if the user did not
589 * change any of the values. This saves about 44MB of KVM on
590 * boxes with 3+GB of ram.
592 * On the flip side, this makes it more likely that some setups
593 * will run out of pv entries. Those sysads will have to bump
594 * the limit up with vm.pamp.pv_entries or vm.pmap.shpgperproc.
596 if (shpgperproc == PMAP_SHPGPERPROC) {
597 if (pv_entry_max > PMAP_PVLIMIT)
598 pv_entry_max = PMAP_PVLIMIT;
601 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
602 pv_entry_high_water = 9 * (pv_entry_max / 10);
605 * Subtract out pages already installed in the zone (hack)
607 entry_max = pv_entry_max - vm_page_array_size;
611 zinitna(pvzone, &pvzone_obj, NULL, 0, entry_max, ZONE_INTERRUPT, 1);
615 /***************************************************
616 * Low level helper routines.....
617 ***************************************************/
622 test_m_maps_pv(vm_page_t m, pv_entry_t pv)
628 KKASSERT(pv->pv_m == m);
630 TAILQ_FOREACH(spv, &m->md.pv_list, pv_list) {
637 panic("test_m_maps_pv: failed m %p pv %p", m, pv);
641 ptbase_assert(struct pmap *pmap)
643 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
645 /* are we current address space or kernel? */
646 if (pmap == &kernel_pmap || frame == (((unsigned)PTDpde) & PG_FRAME))
648 KKASSERT(frame == (*mdcpu->gd_GDMAP1 & PG_FRAME));
653 #define test_m_maps_pv(m, pv)
654 #define ptbase_assert(pmap)
658 #if defined(PMAP_DIAGNOSTIC)
661 * This code checks for non-writeable/modified pages.
662 * This should be an invalid condition.
665 pmap_nw_modified(pt_entry_t ptea)
671 if ((pte & (PG_M|PG_RW)) == PG_M)
680 * This routine defines the region(s) of memory that should not be tested
681 * for the modified bit.
685 static PMAP_INLINE int
686 pmap_track_modified(vm_offset_t va)
688 if ((va < clean_sva) || (va >= clean_eva))
695 * Retrieve the mapped page table base for a particular pmap. Use our self
696 * mapping for the kernel_pmap or our current pmap.
698 * For foreign pmaps we use the per-cpu page table map. Since this involves
699 * installing a ptd it's actually (per-process x per-cpu). However, we
700 * still cannot depend on our mapping to survive thread switches because
701 * the process might be threaded and switching to another thread for the
702 * same process on the same cpu will allow that other thread to make its
705 * This could be a bit confusing but the jist is for something like the
706 * vkernel which uses foreign pmaps all the time this represents a pretty
707 * good cache that avoids unnecessary invltlb()s.
709 * The caller must hold vm_token and the returned value is only valid
710 * until the caller blocks or releases the token.
713 get_ptbase(pmap_t pmap)
715 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
716 struct mdglobaldata *gd = mdcpu;
718 ASSERT_LWKT_TOKEN_HELD(&vm_token);
721 * We can use PTmap if the pmap is our current address space or
722 * the kernel address space.
724 if (pmap == &kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
725 return (unsigned *) PTmap;
729 * Otherwise we use the per-cpu alternative page table map. Each
730 * cpu gets its own map. Because of this we cannot use this map
731 * from interrupts or threads which can preempt.
733 * Even if we already have the map cached we may still have to
734 * invalidate the TLB if another cpu modified a PDE in the map.
736 KKASSERT(gd->mi.gd_intr_nesting_level == 0 &&
737 (gd->mi.gd_curthread->td_flags & TDF_INTTHREAD) == 0);
739 if ((*gd->gd_GDMAP1 & PG_FRAME) != frame) {
740 *gd->gd_GDMAP1 = frame | PG_RW | PG_V;
741 pmap->pm_cached |= gd->mi.gd_cpumask;
743 } else if ((pmap->pm_cached & gd->mi.gd_cpumask) == 0) {
744 pmap->pm_cached |= gd->mi.gd_cpumask;
746 } else if (dreadful_invltlb) {
749 return ((unsigned *)gd->gd_GDADDR1);
755 * Extract the physical page address associated with the map/VA pair.
757 * The caller may hold vm_token if it desires non-blocking operation.
760 pmap_extract(pmap_t pmap, vm_offset_t va)
763 vm_offset_t pdirindex;
765 lwkt_gettoken(&vm_token);
766 pdirindex = va >> PDRSHIFT;
767 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
769 if ((rtval & PG_PS) != 0) {
770 rtval &= ~(NBPDR - 1);
771 rtval |= va & (NBPDR - 1);
773 pte = get_ptbase(pmap) + i386_btop(va);
774 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
779 lwkt_reltoken(&vm_token);
783 /***************************************************
784 * Low level mapping routines.....
785 ***************************************************/
788 * Map a wired VM page to a KVA, fully SMP synchronized.
790 * No requirements, non blocking.
793 pmap_kenter(vm_offset_t va, vm_paddr_t pa)
797 pmap_inval_info info;
799 pmap_inval_init(&info);
800 npte = pa | PG_RW | PG_V | pgeflag;
801 pte = (unsigned *)vtopte(va);
802 pmap_inval_interlock(&info, &kernel_pmap, va);
804 pmap_inval_deinterlock(&info, &kernel_pmap);
805 pmap_inval_done(&info);
809 * Map a wired VM page to a KVA, synchronized on current cpu only.
811 * No requirements, non blocking.
814 pmap_kenter_quick(vm_offset_t va, vm_paddr_t pa)
819 npte = pa | PG_RW | PG_V | pgeflag;
820 pte = (unsigned *)vtopte(va);
822 cpu_invlpg((void *)va);
826 * Synchronize a previously entered VA on all cpus.
828 * No requirements, non blocking.
831 pmap_kenter_sync(vm_offset_t va)
833 pmap_inval_info info;
835 pmap_inval_init(&info);
836 pmap_inval_interlock(&info, &kernel_pmap, va);
837 pmap_inval_deinterlock(&info, &kernel_pmap);
838 pmap_inval_done(&info);
842 * Synchronize a previously entered VA on the current cpu only.
844 * No requirements, non blocking.
847 pmap_kenter_sync_quick(vm_offset_t va)
849 cpu_invlpg((void *)va);
853 * Remove a page from the kernel pagetables, fully SMP synchronized.
855 * No requirements, non blocking.
858 pmap_kremove(vm_offset_t va)
861 pmap_inval_info info;
863 pmap_inval_init(&info);
864 pte = (unsigned *)vtopte(va);
865 pmap_inval_interlock(&info, &kernel_pmap, va);
867 pmap_inval_deinterlock(&info, &kernel_pmap);
868 pmap_inval_done(&info);
872 * Remove a page from the kernel pagetables, synchronized on current cpu only.
874 * No requirements, non blocking.
877 pmap_kremove_quick(vm_offset_t va)
880 pte = (unsigned *)vtopte(va);
882 cpu_invlpg((void *)va);
886 * Adjust the permissions of a page in the kernel page table,
887 * synchronized on the current cpu only.
889 * No requirements, non blocking.
892 pmap_kmodify_rw(vm_offset_t va)
894 atomic_set_int(vtopte(va), PG_RW);
895 cpu_invlpg((void *)va);
899 * Adjust the permissions of a page in the kernel page table,
900 * synchronized on the current cpu only.
902 * No requirements, non blocking.
905 pmap_kmodify_nc(vm_offset_t va)
907 atomic_set_int(vtopte(va), PG_N);
908 cpu_invlpg((void *)va);
912 * Map a range of physical addresses into kernel virtual address space.
914 * No requirements, non blocking.
917 pmap_map(vm_offset_t *virtp, vm_paddr_t start, vm_paddr_t end, int prot)
919 vm_offset_t sva, virt;
922 while (start < end) {
923 pmap_kenter(virt, start);
932 * Add a list of wired pages to the kva, fully SMP synchronized.
934 * No requirements, non blocking.
937 pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
941 end_va = va + count * PAGE_SIZE;
943 while (va < end_va) {
946 pte = (unsigned *)vtopte(va);
947 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
948 cpu_invlpg((void *)va);
953 smp_invltlb(); /* XXX */
958 * Remove pages from KVA, fully SMP synchronized.
960 * No requirements, non blocking.
963 pmap_qremove(vm_offset_t va, int count)
967 end_va = va + count*PAGE_SIZE;
969 while (va < end_va) {
972 pte = (unsigned *)vtopte(va);
974 cpu_invlpg((void *)va);
983 * This routine works like vm_page_lookup() but also blocks as long as the
984 * page is busy. This routine does not busy the page it returns.
986 * The caller must hold the object.
989 pmap_page_lookup(vm_object_t object, vm_pindex_t pindex)
993 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
994 m = vm_page_lookup_busy_wait(object, pindex, FALSE, "pplookp");
1000 * Create a new thread and optionally associate it with a (new) process.
1001 * NOTE! the new thread's cpu may not equal the current cpu.
1004 pmap_init_thread(thread_t td)
1006 /* enforce pcb placement */
1007 td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_size) - 1;
1008 td->td_savefpu = &td->td_pcb->pcb_save;
1009 td->td_sp = (char *)td->td_pcb - 16;
1013 * This routine directly affects the fork perf for a process.
1016 pmap_init_proc(struct proc *p)
1020 /***************************************************
1021 * Page table page management routines.....
1022 ***************************************************/
1025 * This routine unwires page table pages, removing and freeing the page
1026 * tale page when the wire count drops to 0.
1028 * The caller must hold vm_token.
1029 * This function can block.
1032 _pmap_unwire_pte(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
1035 * Wait until we can busy the page ourselves. We cannot have
1036 * any active flushes if we block.
1038 vm_page_busy_wait(m, FALSE, "pmuwpt");
1039 KASSERT(m->queue == PQ_NONE,
1040 ("_pmap_unwire_pte: %p->queue != PQ_NONE", m));
1042 if (m->wire_count == 1) {
1044 * Unmap the page table page.
1046 * NOTE: We must clear pm_cached for all cpus, including
1047 * the current one, when clearing a page directory
1050 pmap_inval_interlock(info, pmap, -1);
1051 KKASSERT(pmap->pm_pdir[m->pindex]);
1052 pmap->pm_pdir[m->pindex] = 0;
1053 pmap->pm_cached = 0;
1054 pmap_inval_deinterlock(info, pmap);
1056 KKASSERT(pmap->pm_stats.resident_count > 0);
1057 --pmap->pm_stats.resident_count;
1059 if (pmap->pm_ptphint == m)
1060 pmap->pm_ptphint = NULL;
1063 * This was our last hold, the page had better be unwired
1064 * after we decrement wire_count.
1066 * FUTURE NOTE: shared page directory page could result in
1067 * multiple wire counts.
1069 vm_page_unwire(m, 0);
1070 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1072 vm_page_free_zero(m);
1075 KKASSERT(m->wire_count > 1);
1076 if (vm_page_unwire_quick(m))
1077 panic("pmap_unwire_pte: Insufficient wire_count");
1084 * The caller must hold vm_token.
1086 * This function can block.
1088 * This function can race the wire_count 2->1 case because the page
1089 * is not busied during the unwire_quick operation. An eventual
1090 * pmap_release() will catch the case.
1092 static PMAP_INLINE int
1093 pmap_unwire_pte(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
1095 KKASSERT(m->wire_count > 0);
1096 if (m->wire_count > 1) {
1097 if (vm_page_unwire_quick(m))
1098 panic("pmap_unwire_pte: Insufficient wire_count");
1101 return _pmap_unwire_pte(pmap, m, info);
1106 * After removing a (user) page table entry, this routine is used to
1107 * conditionally free the page, and manage the hold/wire counts.
1109 * The caller must hold vm_token.
1110 * This function can block regardless.
1113 pmap_unuse_pt(pmap_t pmap, vm_offset_t va, vm_page_t mpte,
1114 pmap_inval_info_t info)
1118 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1120 if (va >= UPT_MIN_ADDRESS)
1124 ptepindex = (va >> PDRSHIFT);
1125 if ((mpte = pmap->pm_ptphint) != NULL &&
1126 mpte->pindex == ptepindex &&
1127 (mpte->flags & PG_BUSY) == 0) {
1130 mpte = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
1131 pmap->pm_ptphint = mpte;
1132 vm_page_wakeup(mpte);
1135 pmap_unwire_pte(pmap, mpte, info);
1139 * Initialize pmap0/vmspace0. This pmap is not added to pmap_list because
1140 * it, and IdlePTD, represents the template used to update all other pmaps.
1142 * On architectures where the kernel pmap is not integrated into the user
1143 * process pmap, this pmap represents the process pmap, not the kernel pmap.
1144 * kernel_pmap should be used to directly access the kernel_pmap.
1149 pmap_pinit0(struct pmap *pmap)
1152 (pd_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
1153 pmap_kenter((vm_offset_t)pmap->pm_pdir, (vm_offset_t) IdlePTD);
1155 pmap->pm_active = 0;
1156 pmap->pm_cached = 0;
1157 pmap->pm_ptphint = NULL;
1158 TAILQ_INIT(&pmap->pm_pvlist);
1159 TAILQ_INIT(&pmap->pm_pvlist_free);
1160 spin_init(&pmap->pm_spin);
1161 lwkt_token_init(&pmap->pm_token, "pmap_tok");
1162 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1166 * Initialize a preallocated and zeroed pmap structure,
1167 * such as one in a vmspace structure.
1172 pmap_pinit(struct pmap *pmap)
1177 * No need to allocate page table space yet but we do need a valid
1178 * page directory table.
1180 if (pmap->pm_pdir == NULL) {
1182 (pd_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
1186 * Allocate an object for the ptes
1188 if (pmap->pm_pteobj == NULL)
1189 pmap->pm_pteobj = vm_object_allocate(OBJT_DEFAULT, PTDPTDI + 1);
1192 * Allocate the page directory page, unless we already have
1193 * one cached. If we used the cached page the wire_count will
1194 * already be set appropriately.
1196 if ((ptdpg = pmap->pm_pdirm) == NULL) {
1197 ptdpg = vm_page_grab(pmap->pm_pteobj, PTDPTDI,
1198 VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
1200 pmap->pm_pdirm = ptdpg;
1201 vm_page_flag_clear(ptdpg, PG_MAPPED);
1202 vm_page_wire(ptdpg);
1203 KKASSERT(ptdpg->valid == VM_PAGE_BITS_ALL);
1204 pmap_kenter((vm_offset_t)pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
1205 vm_page_wakeup(ptdpg);
1207 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
1209 /* install self-referential address mapping entry */
1210 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1211 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1214 pmap->pm_active = 0;
1215 pmap->pm_cached = 0;
1216 pmap->pm_ptphint = NULL;
1217 TAILQ_INIT(&pmap->pm_pvlist);
1218 TAILQ_INIT(&pmap->pm_pvlist_free);
1219 spin_init(&pmap->pm_spin);
1220 lwkt_token_init(&pmap->pm_token, "pmap_tok");
1221 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1222 pmap->pm_stats.resident_count = 1;
1226 * Clean up a pmap structure so it can be physically freed. This routine
1227 * is called by the vmspace dtor function. A great deal of pmap data is
1228 * left passively mapped to improve vmspace management so we have a bit
1229 * of cleanup work to do here.
1234 pmap_puninit(pmap_t pmap)
1238 pmap_wait(pmap, -1);
1239 KKASSERT(pmap->pm_active == 0);
1240 if ((p = pmap->pm_pdirm) != NULL) {
1241 KKASSERT(pmap->pm_pdir != NULL);
1242 pmap_kremove((vm_offset_t)pmap->pm_pdir);
1243 vm_page_busy_wait(p, FALSE, "pgpun");
1244 vm_page_unwire(p, 0);
1245 vm_page_free_zero(p);
1246 pmap->pm_pdirm = NULL;
1248 if (pmap->pm_pdir) {
1249 kmem_free(&kernel_map, (vm_offset_t)pmap->pm_pdir, PAGE_SIZE);
1250 pmap->pm_pdir = NULL;
1252 if (pmap->pm_pteobj) {
1253 vm_object_deallocate(pmap->pm_pteobj);
1254 pmap->pm_pteobj = NULL;
1259 * Wire in kernel global address entries. To avoid a race condition
1260 * between pmap initialization and pmap_growkernel, this procedure
1261 * adds the pmap to the master list (which growkernel scans to update),
1262 * then copies the template.
1267 pmap_pinit2(struct pmap *pmap)
1270 * XXX copies current process, does not fill in MPPTDI
1272 spin_lock(&pmap_spin);
1273 TAILQ_INSERT_TAIL(&pmap_list, pmap, pm_pmnode);
1274 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
1275 spin_unlock(&pmap_spin);
1279 * Attempt to release and free a vm_page in a pmap. Returns 1 on success,
1280 * 0 on failure (if the procedure had to sleep).
1282 * When asked to remove the page directory page itself, we actually just
1283 * leave it cached so we do not have to incur the SMP inval overhead of
1284 * removing the kernel mapping. pmap_puninit() will take care of it.
1286 * The caller must hold vm_token.
1287 * This function can block regardless.
1290 pmap_release_free_page(struct pmap *pmap, vm_page_t p)
1292 unsigned *pde = (unsigned *) pmap->pm_pdir;
1295 * This code optimizes the case of freeing non-busy
1296 * page-table pages. Those pages are zero now, and
1297 * might as well be placed directly into the zero queue.
1299 if (vm_page_busy_try(p, FALSE)) {
1300 vm_page_sleep_busy(p, FALSE, "pmaprl");
1304 KKASSERT(pmap->pm_stats.resident_count > 0);
1305 KKASSERT(pde[p->pindex]);
1308 * page table page's wire_count must be 1. Caller is the pmap
1309 * termination code which holds the pm_pteobj, there is a race
1310 * if someone else is trying to hold the VM object in order to
1311 * clean up a wire_count.
1313 if (p->wire_count != 1) {
1314 if (pmap->pm_pteobj->hold_count <= 1)
1315 panic("pmap_release: freeing wired page table page");
1316 kprintf("pmap_release_free_page: unwire race detected\n");
1318 tsleep(p, 0, "pmapx", 1);
1323 * Remove the page table page from the processes address space.
1325 pmap->pm_cached = 0;
1327 --pmap->pm_stats.resident_count;
1328 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1329 pmap->pm_ptphint = NULL;
1332 * We leave the page directory page cached, wired, and mapped in
1333 * the pmap until the dtor function (pmap_puninit()) gets called.
1334 * However, still clean it up so we can set PG_ZERO.
1336 * The pmap has already been removed from the pmap_list in the
1339 if (p->pindex == PTDPTDI) {
1340 bzero(pde + KPTDI, nkpt * PTESIZE);
1341 bzero(pde + MPPTDI, (NPDEPG - MPPTDI) * PTESIZE);
1342 vm_page_flag_set(p, PG_ZERO);
1346 * This case can occur if a pmap_unwire_pte() loses a race
1347 * while the page is unbusied.
1349 /*panic("pmap_release: page should already be gone %p", p);*/
1350 vm_page_flag_clear(p, PG_MAPPED);
1351 vm_page_unwire(p, 0);
1352 vm_page_free_zero(p);
1358 * This routine is called if the page table page is not mapped correctly.
1360 * The caller must hold vm_token.
1363 _pmap_allocpte(pmap_t pmap, unsigned ptepindex)
1369 * Find or fabricate a new pagetable page. Setting VM_ALLOC_ZERO
1370 * will zero any new page and mark it valid.
1372 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
1373 VM_ALLOC_NORMAL | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
1375 KASSERT(m->queue == PQ_NONE,
1376 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1379 * Increment the wire count for the page we will be returning to
1385 * It is possible that someone else got in and mapped by the page
1386 * directory page while we were blocked, if so just unbusy and
1387 * return the wired page.
1389 if ((ptepa = pmap->pm_pdir[ptepindex]) != 0) {
1390 KKASSERT((ptepa & PG_FRAME) == VM_PAGE_TO_PHYS(m));
1396 * Map the pagetable page into the process address space, if
1397 * it isn't already there.
1399 * NOTE: For safety clear pm_cached for all cpus including the
1400 * current one when adding a PDE to the map.
1402 ++pmap->pm_stats.resident_count;
1404 ptepa = VM_PAGE_TO_PHYS(m);
1405 pmap->pm_pdir[ptepindex] =
1406 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
1407 pmap->pm_cached = 0;
1410 * Set the page table hint
1412 pmap->pm_ptphint = m;
1413 vm_page_flag_set(m, PG_MAPPED);
1420 * Allocate a page table entry for a va.
1422 * The caller must hold vm_token.
1425 pmap_allocpte(pmap_t pmap, vm_offset_t va)
1431 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1434 * Calculate pagetable page index
1436 ptepindex = va >> PDRSHIFT;
1439 * Get the page directory entry
1441 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1444 * This supports switching from a 4MB page to a
1447 if (ptepa & PG_PS) {
1448 pmap->pm_pdir[ptepindex] = 0;
1455 * If the page table page is mapped, we just increment the
1456 * wire count, and activate it.
1460 * In order to get the page table page, try the
1463 if ((mpte = pmap->pm_ptphint) != NULL &&
1464 (mpte->pindex == ptepindex) &&
1465 (mpte->flags & PG_BUSY) == 0) {
1466 vm_page_wire_quick(mpte);
1468 mpte = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
1469 pmap->pm_ptphint = mpte;
1470 vm_page_wire_quick(mpte);
1471 vm_page_wakeup(mpte);
1476 * Here if the pte page isn't mapped, or if it has been deallocated.
1478 return _pmap_allocpte(pmap, ptepindex);
1482 /***************************************************
1483 * Pmap allocation/deallocation routines.
1484 ***************************************************/
1487 * Release any resources held by the given physical map.
1488 * Called when a pmap initialized by pmap_pinit is being released.
1489 * Should only be called if the map contains no valid mappings.
1491 * Caller must hold pmap->pm_token
1493 static int pmap_release_callback(struct vm_page *p, void *data);
1496 pmap_release(struct pmap *pmap)
1498 vm_object_t object = pmap->pm_pteobj;
1499 struct rb_vm_page_scan_info info;
1501 KASSERT(pmap->pm_active == 0,
1502 ("pmap still active! %08x", pmap->pm_active));
1503 #if defined(DIAGNOSTIC)
1504 if (object->ref_count != 1)
1505 panic("pmap_release: pteobj reference count != 1");
1509 info.object = object;
1511 spin_lock(&pmap_spin);
1512 TAILQ_REMOVE(&pmap_list, pmap, pm_pmnode);
1513 spin_unlock(&pmap_spin);
1515 vm_object_hold(object);
1516 /*lwkt_gettoken(&vm_token);*/
1520 info.limit = object->generation;
1522 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
1523 pmap_release_callback, &info);
1524 if (info.error == 0 && info.mpte) {
1525 if (!pmap_release_free_page(pmap, info.mpte))
1528 } while (info.error);
1529 /*lwkt_reltoken(&vm_token);*/
1530 vm_object_drop(object);
1532 pmap->pm_cached = 0;
1536 * The caller must hold vm_token.
1539 pmap_release_callback(struct vm_page *p, void *data)
1541 struct rb_vm_page_scan_info *info = data;
1543 if (p->pindex == PTDPTDI) {
1547 if (!pmap_release_free_page(info->pmap, p)) {
1551 if (info->object->generation != info->limit) {
1559 * Grow the number of kernel page table entries, if needed.
1564 pmap_growkernel(vm_offset_t kstart, vm_offset_t kend)
1566 vm_offset_t addr = kend;
1568 vm_offset_t ptppaddr;
1572 vm_object_hold(kptobj);
1573 if (kernel_vm_end == 0) {
1574 kernel_vm_end = KERNBASE;
1576 while (pdir_pde(PTD, kernel_vm_end)) {
1577 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1578 ~(PAGE_SIZE * NPTEPG - 1);
1582 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1583 while (kernel_vm_end < addr) {
1584 if (pdir_pde(PTD, kernel_vm_end)) {
1585 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1586 ~(PAGE_SIZE * NPTEPG - 1);
1591 * This index is bogus, but out of the way
1593 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_NORMAL |
1595 VM_ALLOC_INTERRUPT);
1597 panic("pmap_growkernel: no memory to grow kernel");
1600 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1601 pmap_zero_page(ptppaddr);
1602 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1603 pdir_pde(PTD, kernel_vm_end) = newpdir;
1604 *pmap_pde(&kernel_pmap, kernel_vm_end) = newpdir;
1608 * This update must be interlocked with pmap_pinit2.
1610 spin_lock(&pmap_spin);
1611 TAILQ_FOREACH(pmap, &pmap_list, pm_pmnode) {
1612 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1614 spin_unlock(&pmap_spin);
1615 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1616 ~(PAGE_SIZE * NPTEPG - 1);
1618 vm_object_drop(kptobj);
1622 * Add a reference to the specified pmap.
1627 pmap_reference(pmap_t pmap)
1630 lwkt_gettoken(&vm_token);
1632 lwkt_reltoken(&vm_token);
1637 * vm_token must be held
1641 pmap_hold(pmap_t pmap)
1647 * vm_token must be held
1651 pmap_drop(pmap_t pmap)
1654 if (pmap->pm_count == (int)0x80000000)
1660 pmap_wait(pmap_t pmap, int count)
1662 lwkt_gettoken(&vm_token);
1663 pmap->pm_count += count;
1664 if (pmap->pm_count & 0x7FFFFFFF) {
1665 while (pmap->pm_count & 0x7FFFFFFF) {
1666 pmap->pm_count |= 0x80000000;
1667 tsleep(pmap, 0, "pmapd", 0);
1668 pmap->pm_count &= ~0x80000000;
1669 kprintf("pmap_wait: race averted\n");
1672 lwkt_reltoken(&vm_token);
1675 /***************************************************
1676 * page management routines.
1677 ***************************************************/
1680 * free the pv_entry back to the free list. This function may be
1681 * called from an interrupt.
1683 * The caller must hold vm_token.
1685 static PMAP_INLINE void
1686 free_pv_entry(pv_entry_t pv)
1688 struct mdglobaldata *gd;
1691 KKASSERT(pv->pv_m != NULL);
1696 if (gd->gd_freepv == NULL)
1703 * get a new pv_entry, allocating a block from the system
1704 * when needed. This function may be called from an interrupt thread.
1706 * THIS FUNCTION CAN BLOCK ON THE ZALLOC TOKEN, serialization of other
1707 * tokens (aka vm_token) to be temporarily lost.
1709 * The caller must hold vm_token.
1714 struct mdglobaldata *gd;
1718 if (pv_entry_high_water &&
1719 (pv_entry_count > pv_entry_high_water) &&
1720 (pmap_pagedaemon_waken == 0)) {
1721 pmap_pagedaemon_waken = 1;
1722 wakeup (&vm_pages_needed);
1725 if ((pv = gd->gd_freepv) != NULL)
1726 gd->gd_freepv = NULL;
1728 pv = zalloc(pvzone);
1733 * This routine is very drastic, but can save the system
1743 static int warningdone=0;
1745 if (pmap_pagedaemon_waken == 0)
1747 lwkt_gettoken(&vm_token);
1748 pmap_pagedaemon_waken = 0;
1750 if (warningdone < 5) {
1751 kprintf("pmap_collect: collecting pv entries -- "
1752 "suggest increasing PMAP_SHPGPERPROC\n");
1756 for (i = 0; i < vm_page_array_size; i++) {
1757 m = &vm_page_array[i];
1758 if (m->wire_count || m->hold_count)
1760 if (vm_page_busy_try(m, TRUE) == 0) {
1761 if (m->wire_count == 0 && m->hold_count == 0) {
1767 lwkt_reltoken(&vm_token);
1772 * Remove the pv entry and unwire the page table page related to the
1773 * pte the caller has cleared from the page table.
1775 * The caller must hold vm_token.
1778 pmap_remove_entry(struct pmap *pmap, vm_page_t m,
1779 vm_offset_t va, pmap_inval_info_t info)
1786 ASSERT_LWKT_TOKEN_HELD(&vm_token);
1787 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1788 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1789 if (pmap == pv->pv_pmap && va == pv->pv_va)
1793 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
1795 KKASSERT(pv->pv_pmap == pmap);
1797 if (va == pv->pv_va)
1806 test_m_maps_pv(m, pv);
1807 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1808 m->md.pv_list_count--;
1810 atomic_add_int(&m->object->agg_pv_list_count, -1);
1811 if (TAILQ_EMPTY(&m->md.pv_list))
1812 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1813 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1814 ++pmap->pm_generation;
1819 vm_object_hold(pmap->pm_pteobj);
1820 pmap_unuse_pt(pmap, va, pv->pv_ptem, info);
1821 vm_object_drop(pmap->pm_pteobj);
1826 * Create a pv entry for page at pa for (pmap, va).
1828 * The caller must hold vm_token.
1831 pmap_insert_entry(pmap_t pmap, pv_entry_t pv, vm_offset_t va,
1832 vm_page_t mpte, vm_page_t m)
1835 KKASSERT(pv->pv_m == NULL);
1842 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1843 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
1844 ++pmap->pm_generation;
1845 m->md.pv_list_count++;
1847 atomic_add_int(&m->object->agg_pv_list_count, 1);
1851 * pmap_remove_pte: do the things to unmap a page in a process.
1853 * The caller must hold vm_token.
1855 * WARNING! As with most other pmap functions this one can block, so
1856 * callers using temporary page table mappings must reload
1860 pmap_remove_pte(struct pmap *pmap, unsigned *ptq, vm_offset_t va,
1861 pmap_inval_info_t info)
1866 ptbase_assert(pmap);
1867 pmap_inval_interlock(info, pmap, va);
1868 ptbase_assert(pmap);
1869 oldpte = loadandclear(ptq);
1871 pmap->pm_stats.wired_count -= 1;
1872 pmap_inval_deinterlock(info, pmap);
1873 KKASSERT(oldpte & PG_V);
1875 * Machines that don't support invlpg, also don't support
1876 * PG_G. XXX PG_G is disabled for SMP so don't worry about
1880 cpu_invlpg((void *)va);
1881 KKASSERT(pmap->pm_stats.resident_count > 0);
1882 --pmap->pm_stats.resident_count;
1883 if (oldpte & PG_MANAGED) {
1884 m = PHYS_TO_VM_PAGE(oldpte);
1885 if (oldpte & PG_M) {
1886 #if defined(PMAP_DIAGNOSTIC)
1887 if (pmap_nw_modified((pt_entry_t) oldpte)) {
1888 kprintf("pmap_remove: modified page not "
1889 "writable: va: %p, pte: 0x%lx\n",
1890 (void *)va, (long)oldpte);
1893 if (pmap_track_modified(va))
1897 vm_page_flag_set(m, PG_REFERENCED);
1898 pmap_remove_entry(pmap, m, va, info);
1900 pmap_unuse_pt(pmap, va, NULL, info);
1905 * Remove a single page from a process address space.
1907 * The caller must hold vm_token.
1910 pmap_remove_page(struct pmap *pmap, vm_offset_t va, pmap_inval_info_t info)
1915 * If there is no pte for this address, just skip it!!! Otherwise
1916 * get a local va for mappings for this pmap and remove the entry.
1918 if (*pmap_pde(pmap, va) != 0) {
1919 ptq = get_ptbase(pmap) + i386_btop(va);
1921 pmap_remove_pte(pmap, ptq, va, info);
1928 * Remove the given range of addresses from the specified map.
1930 * It is assumed that the start and end are properly rounded to the page
1936 pmap_remove(struct pmap *pmap, vm_offset_t sva, vm_offset_t eva)
1940 vm_offset_t ptpaddr;
1941 vm_offset_t sindex, eindex;
1942 struct pmap_inval_info info;
1947 vm_object_hold(pmap->pm_pteobj);
1948 lwkt_gettoken(&vm_token);
1949 if (pmap->pm_stats.resident_count == 0) {
1950 lwkt_reltoken(&vm_token);
1951 vm_object_drop(pmap->pm_pteobj);
1955 pmap_inval_init(&info);
1958 * special handling of removing one page. a very
1959 * common operation and easy to short circuit some
1962 if (((sva + PAGE_SIZE) == eva) &&
1963 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
1964 pmap_remove_page(pmap, sva, &info);
1965 pmap_inval_done(&info);
1966 lwkt_reltoken(&vm_token);
1967 vm_object_drop(pmap->pm_pteobj);
1972 * Get a local virtual address for the mappings that are being
1975 sindex = i386_btop(sva);
1976 eindex = i386_btop(eva);
1978 while (sindex < eindex) {
1982 * Stop scanning if no pages are left
1984 if (pmap->pm_stats.resident_count == 0)
1988 * Calculate index for next page table, limited by eindex.
1990 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1994 pdirindex = sindex / NPDEPG;
1995 ptpaddr = (unsigned)pmap->pm_pdir[pdirindex];
1996 if (ptpaddr & PG_PS) {
1997 pmap_inval_interlock(&info, pmap, -1);
1998 pmap->pm_pdir[pdirindex] = 0;
1999 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
2000 pmap->pm_cached = 0;
2001 pmap_inval_deinterlock(&info, pmap);
2007 * Weed out invalid mappings. Note: we assume that the page
2008 * directory table is always allocated, and in kernel virtual.
2016 * Sub-scan the page table page. pmap_remove_pte() can
2017 * block on us, invalidating ptbase, so we must reload
2018 * ptbase and we must also check whether the page directory
2019 * page is still present.
2021 while (sindex < pdnxt) {
2024 ptbase = get_ptbase(pmap);
2025 if (ptbase[sindex]) {
2026 va = i386_ptob(sindex);
2027 pmap_remove_pte(pmap, ptbase + sindex,
2030 if (pmap->pm_pdir[pdirindex] == 0 ||
2031 (pmap->pm_pdir[pdirindex] & PG_PS)) {
2037 pmap_inval_done(&info);
2038 lwkt_reltoken(&vm_token);
2039 vm_object_drop(pmap->pm_pteobj);
2043 * Removes this physical page from all physical maps in which it resides.
2044 * Reflects back modify bits to the pager.
2046 * vm_token must be held by caller.
2049 pmap_remove_all(vm_page_t m)
2051 struct pmap_inval_info info;
2052 unsigned *pte, tpte;
2056 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2058 if (TAILQ_EMPTY(&m->md.pv_list))
2061 pmap_inval_init(&info);
2062 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
2064 KKASSERT(pmap->pm_stats.resident_count > 0);
2066 vm_object_hold(pmap->pm_pteobj);
2068 if (pv != TAILQ_FIRST(&m->md.pv_list)) {
2069 vm_object_drop(pmap->pm_pteobj);
2074 --pmap->pm_stats.resident_count;
2075 pte = pmap_pte_quick(pmap, pv->pv_va);
2076 pmap_inval_interlock(&info, pmap, pv->pv_va);
2077 tpte = loadandclear(pte);
2079 pmap->pm_stats.wired_count--;
2080 pmap_inval_deinterlock(&info, pmap);
2082 vm_page_flag_set(m, PG_REFERENCED);
2083 KKASSERT(PHYS_TO_VM_PAGE(tpte) == m);
2086 * Update the vm_page_t clean and reference bits.
2089 #if defined(PMAP_DIAGNOSTIC)
2090 if (pmap_nw_modified((pt_entry_t) tpte)) {
2091 kprintf("pmap_remove_all: modified page "
2092 "not writable: va: %p, pte: 0x%lx\n",
2093 (void *)pv->pv_va, (long)tpte);
2096 if (pmap_track_modified(pv->pv_va))
2100 KKASSERT(pv->pv_m == m);
2102 KKASSERT(pv == TAILQ_FIRST(&m->md.pv_list));
2103 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
2104 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
2105 ++pmap->pm_generation;
2106 m->md.pv_list_count--;
2108 atomic_add_int(&m->object->agg_pv_list_count, -1);
2109 if (TAILQ_EMPTY(&m->md.pv_list))
2110 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
2111 pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem, &info);
2112 vm_object_drop(pmap->pm_pteobj);
2116 KKASSERT((m->flags & (PG_MAPPED|PG_WRITEABLE)) == 0);
2117 pmap_inval_done(&info);
2121 * Set the physical protection on the specified range of this map
2127 pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
2130 vm_offset_t pdnxt, ptpaddr;
2131 vm_pindex_t sindex, eindex;
2132 pmap_inval_info info;
2137 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2138 pmap_remove(pmap, sva, eva);
2142 if (prot & VM_PROT_WRITE)
2145 lwkt_gettoken(&vm_token);
2146 pmap_inval_init(&info);
2148 ptbase = get_ptbase(pmap);
2150 sindex = i386_btop(sva);
2151 eindex = i386_btop(eva);
2153 for (; sindex < eindex; sindex = pdnxt) {
2156 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
2158 pdirindex = sindex / NPDEPG;
2159 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
2160 pmap_inval_interlock(&info, pmap, -1);
2161 pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
2162 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
2163 pmap_inval_deinterlock(&info, pmap);
2168 * Weed out invalid mappings. Note: we assume that the page
2169 * directory table is always allocated, and in kernel virtual.
2174 if (pdnxt > eindex) {
2178 for (; sindex != pdnxt; sindex++) {
2186 pmap_inval_interlock(&info, pmap, i386_ptob(sindex));
2188 pbits = ptbase[sindex];
2191 if (pbits & PG_MANAGED) {
2194 m = PHYS_TO_VM_PAGE(pbits);
2195 vm_page_flag_set(m, PG_REFERENCED);
2199 if (pmap_track_modified(i386_ptob(sindex))) {
2201 m = PHYS_TO_VM_PAGE(pbits);
2208 if (pbits != cbits &&
2209 !atomic_cmpset_int(ptbase + sindex, pbits, cbits)) {
2212 pmap_inval_deinterlock(&info, pmap);
2215 pmap_inval_done(&info);
2216 lwkt_reltoken(&vm_token);
2220 * Insert the given physical page (p) at the specified virtual address (v)
2221 * in the target physical map with the protection requested.
2223 * If specified, the page will be wired down, meaning that the related pte
2224 * cannot be reclaimed.
2229 pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
2230 boolean_t wired, vm_map_entry_t entry __unused)
2235 vm_offset_t origpte, newpte;
2237 pmap_inval_info info;
2244 #ifdef PMAP_DIAGNOSTIC
2246 panic("pmap_enter: toobig");
2247 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) {
2248 panic("pmap_enter: invalid to pmap_enter page "
2249 "table pages (va: %p)", (void *)va);
2252 if (va < UPT_MAX_ADDRESS && pmap == &kernel_pmap) {
2253 kprintf("Warning: pmap_enter called on UVA with kernel_pmap\n");
2254 print_backtrace(-1);
2256 if (va >= UPT_MAX_ADDRESS && pmap != &kernel_pmap) {
2257 kprintf("Warning: pmap_enter called on KVA without kernel_pmap\n");
2258 print_backtrace(-1);
2261 vm_object_hold(pmap->pm_pteobj);
2262 lwkt_gettoken(&vm_token);
2265 * This can block, get it before we do anything important.
2267 if (pmap_initialized &&
2268 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2269 pv = get_pv_entry();
2275 * In the case that a page table page is not
2276 * resident, we are creating it here.
2278 if (va < UPT_MIN_ADDRESS)
2279 mpte = pmap_allocpte(pmap, va);
2283 if ((prot & VM_PROT_NOSYNC) == 0)
2284 pmap_inval_init(&info);
2285 pte = pmap_pte(pmap, va);
2288 * Page Directory table entry not valid, we need a new PT page
2291 panic("pmap_enter: invalid page directory pdir=0x%lx, va=%p",
2292 (long)pmap->pm_pdir[PTDPTDI], (void *)va);
2295 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2296 origpte = *(vm_offset_t *)pte;
2297 opa = origpte & PG_FRAME;
2299 if (origpte & PG_PS)
2300 panic("pmap_enter: attempted pmap_enter on 4MB page");
2303 * Mapping has not changed, must be protection or wiring change.
2305 if (origpte && (opa == pa)) {
2307 * Wiring change, just update stats. We don't worry about
2308 * wiring PT pages as they remain resident as long as there
2309 * are valid mappings in them. Hence, if a user page is wired,
2310 * the PT page will be also.
2312 if (wired && ((origpte & PG_W) == 0))
2313 pmap->pm_stats.wired_count++;
2314 else if (!wired && (origpte & PG_W))
2315 pmap->pm_stats.wired_count--;
2317 #if defined(PMAP_DIAGNOSTIC)
2318 if (pmap_nw_modified((pt_entry_t) origpte)) {
2319 kprintf("pmap_enter: modified page not "
2320 "writable: va: %p, pte: 0x%lx\n",
2321 (void *)va, (long )origpte);
2326 * We might be turning off write access to the page,
2327 * so we go ahead and sense modify status.
2329 if (origpte & PG_MANAGED) {
2330 if ((origpte & PG_M) && pmap_track_modified(va)) {
2332 om = PHYS_TO_VM_PAGE(opa);
2336 KKASSERT(m->flags & PG_MAPPED);
2341 * Mapping has changed, invalidate old range and fall through to
2342 * handle validating new mapping.
2344 * Since we have a ref on the page directory page pmap_pte()
2345 * will always return non-NULL.
2347 * NOTE: pmap_remove_pte() can block and cause the temporary ptbase
2348 * to get wiped. reload the ptbase. I'm not sure if it is
2349 * also possible to race another pmap_enter() but check for
2353 KKASSERT((origpte & PG_FRAME) ==
2354 (*(vm_offset_t *)pte & PG_FRAME));
2355 if (prot & VM_PROT_NOSYNC) {
2356 prot &= ~VM_PROT_NOSYNC;
2357 pmap_inval_init(&info);
2359 pmap_remove_pte(pmap, pte, va, &info);
2360 pte = pmap_pte(pmap, va);
2361 origpte = *(vm_offset_t *)pte;
2362 opa = origpte & PG_FRAME;
2364 kprintf("pmap_enter: Warning, raced pmap %p va %p\n",
2370 * Enter on the PV list if part of our managed memory. Note that we
2371 * raise IPL while manipulating pv_table since pmap_enter can be
2372 * called at interrupt time.
2374 if (pmap_initialized &&
2375 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2376 pmap_insert_entry(pmap, pv, va, mpte, m);
2378 ptbase_assert(pmap);
2380 vm_page_flag_set(m, PG_MAPPED);
2384 * Increment counters
2386 ++pmap->pm_stats.resident_count;
2388 pmap->pm_stats.wired_count++;
2389 KKASSERT(*pte == 0);
2393 * Now validate mapping with desired protection/wiring.
2395 ptbase_assert(pmap);
2396 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2400 if (va < UPT_MIN_ADDRESS)
2402 if (pmap == &kernel_pmap)
2406 * If the mapping or permission bits are different, we need
2407 * to update the pte. If the pte is already present we have
2408 * to get rid of the extra wire-count on mpte we had obtained
2411 * mpte has a new wire_count, which also serves to prevent the
2412 * page table page from getting ripped out while we work. If we
2413 * are modifying an existing pte instead of installing a new one
2414 * we have to drop it.
2416 if ((origpte & ~(PG_M|PG_A)) != newpte) {
2417 if (prot & VM_PROT_NOSYNC)
2418 cpu_invlpg((void *)va);
2420 pmap_inval_interlock(&info, pmap, va);
2421 ptbase_assert(pmap);
2424 KKASSERT((*pte & PG_FRAME) == (newpte & PG_FRAME));
2425 if (mpte && vm_page_unwire_quick(mpte))
2426 panic("pmap_enter: Insufficient wire_count");
2429 *pte = newpte | PG_A;
2430 if ((prot & VM_PROT_NOSYNC) == 0)
2431 pmap_inval_deinterlock(&info, pmap);
2433 vm_page_flag_set(m, PG_WRITEABLE);
2436 KKASSERT((*pte & PG_FRAME) == (newpte & PG_FRAME));
2437 if (mpte && vm_page_unwire_quick(mpte))
2438 panic("pmap_enter: Insufficient wire_count");
2443 * NOTE: mpte invalid after this point if we block.
2445 KKASSERT((newpte & PG_MANAGED) == 0 || (m->flags & PG_MAPPED));
2446 if ((prot & VM_PROT_NOSYNC) == 0)
2447 pmap_inval_done(&info);
2450 lwkt_reltoken(&vm_token);
2451 vm_object_drop(pmap->pm_pteobj);
2455 * This code works like pmap_enter() but assumes VM_PROT_READ and not-wired.
2456 * This code also assumes that the pmap has no pre-existing entry for this
2459 * This code currently may only be used on user pmaps, not kernel_pmap.
2464 pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m)
2471 pmap_inval_info info;
2474 vm_object_hold(pmap->pm_pteobj);
2475 lwkt_gettoken(&vm_token);
2478 * This can block, get it before we do anything important.
2480 if (pmap_initialized &&
2481 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2482 pv = get_pv_entry();
2487 pmap_inval_init(&info);
2489 if (va < UPT_MAX_ADDRESS && pmap == &kernel_pmap) {
2490 kprintf("Warning: pmap_enter_quick called on UVA with kernel_pmap\n");
2491 print_backtrace(-1);
2493 if (va >= UPT_MAX_ADDRESS && pmap != &kernel_pmap) {
2494 kprintf("Warning: pmap_enter_quick called on KVA without kernel_pmap\n");
2495 print_backtrace(-1);
2498 KKASSERT(va < UPT_MIN_ADDRESS); /* assert used on user pmaps only */
2501 * Calculate the page table page (mpte), allocating it if necessary.
2503 * A held page table page (mpte), or NULL, is passed onto the
2504 * section following.
2506 if (va < UPT_MIN_ADDRESS) {
2508 * Calculate pagetable page index
2510 ptepindex = va >> PDRSHIFT;
2514 * Get the page directory entry
2516 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2519 * If the page table page is mapped, we just increment
2520 * the wire count, and activate it.
2524 panic("pmap_enter_quick: unexpected mapping into 4MB page");
2525 if ((mpte = pmap->pm_ptphint) != NULL &&
2526 (mpte->pindex == ptepindex) &&
2527 (mpte->flags & PG_BUSY) == 0) {
2528 vm_page_wire_quick(mpte);
2530 mpte = pmap_page_lookup(pmap->pm_pteobj,
2532 pmap->pm_ptphint = mpte;
2533 vm_page_wire_quick(mpte);
2534 vm_page_wakeup(mpte);
2537 mpte = _pmap_allocpte(pmap, ptepindex);
2539 } while (mpte == NULL);
2542 /* this code path is not yet used */
2546 * With a valid (and held) page directory page, we can just use
2547 * vtopte() to get to the pte. If the pte is already present
2548 * we do not disturb it.
2550 pte = (unsigned *)vtopte(va);
2552 KKASSERT(*pte & PG_V);
2553 pa = VM_PAGE_TO_PHYS(m);
2554 KKASSERT(((*pte ^ pa) & PG_FRAME) == 0);
2555 pmap_inval_done(&info);
2557 pmap_unwire_pte(pmap, mpte, &info);
2562 lwkt_reltoken(&vm_token);
2563 vm_object_drop(pmap->pm_pteobj);
2568 * Enter on the PV list if part of our managed memory
2570 if (pmap_initialized &&
2571 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2572 pmap_insert_entry(pmap, pv, va, mpte, m);
2574 vm_page_flag_set(m, PG_MAPPED);
2578 * Increment counters
2580 ++pmap->pm_stats.resident_count;
2582 pa = VM_PAGE_TO_PHYS(m);
2585 * Now validate mapping with RO protection
2587 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2588 *pte = pa | PG_V | PG_U;
2590 *pte = pa | PG_V | PG_U | PG_MANAGED;
2591 /* pmap_inval_add(&info, pmap, va); shouldn't be needed inval->valid */
2592 pmap_inval_done(&info);
2597 lwkt_reltoken(&vm_token);
2598 vm_object_drop(pmap->pm_pteobj);
2602 * Make a temporary mapping for a physical address. This is only intended
2603 * to be used for panic dumps.
2605 * The caller is responsible for calling smp_invltlb().
2610 pmap_kenter_temporary(vm_paddr_t pa, long i)
2612 pmap_kenter_quick((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
2613 return ((void *)crashdumpmap);
2616 #define MAX_INIT_PT (96)
2619 * This routine preloads the ptes for a given object into the specified pmap.
2620 * This eliminates the blast of soft faults on process startup and
2621 * immediately after an mmap.
2625 static int pmap_object_init_pt_callback(vm_page_t p, void *data);
2628 pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_prot_t prot,
2629 vm_object_t object, vm_pindex_t pindex,
2630 vm_size_t size, int limit)
2632 struct rb_vm_page_scan_info info;
2637 * We can't preinit if read access isn't set or there is no pmap
2640 if ((prot & VM_PROT_READ) == 0 || pmap == NULL || object == NULL)
2644 * We can't preinit if the pmap is not the current pmap
2646 lp = curthread->td_lwp;
2647 if (lp == NULL || pmap != vmspace_pmap(lp->lwp_vmspace))
2650 psize = i386_btop(size);
2652 if ((object->type != OBJT_VNODE) ||
2653 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2654 (object->resident_page_count > MAX_INIT_PT))) {
2658 if (psize + pindex > object->size) {
2659 if (object->size < pindex)
2661 psize = object->size - pindex;
2668 * Use a red-black scan to traverse the requested range and load
2669 * any valid pages found into the pmap.
2671 * We cannot safely scan the object's memq unless we are in a
2672 * critical section since interrupts can remove pages from objects.
2674 info.start_pindex = pindex;
2675 info.end_pindex = pindex + psize - 1;
2681 vm_object_hold(object);
2682 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
2683 pmap_object_init_pt_callback, &info);
2684 vm_object_drop(object);
2688 * The caller must hold vm_token.
2692 pmap_object_init_pt_callback(vm_page_t p, void *data)
2694 struct rb_vm_page_scan_info *info = data;
2695 vm_pindex_t rel_index;
2697 * don't allow an madvise to blow away our really
2698 * free pages allocating pv entries.
2700 if ((info->limit & MAP_PREFAULT_MADVISE) &&
2701 vmstats.v_free_count < vmstats.v_free_reserved) {
2706 * Ignore list markers and ignore pages we cannot instantly
2707 * busy (while holding the object token).
2709 if (p->flags & PG_MARKER)
2711 if (vm_page_busy_try(p, TRUE))
2713 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
2714 (p->flags & PG_FICTITIOUS) == 0) {
2715 if ((p->queue - p->pc) == PQ_CACHE)
2716 vm_page_deactivate(p);
2717 rel_index = p->pindex - info->start_pindex;
2718 pmap_enter_quick(info->pmap,
2719 info->addr + i386_ptob(rel_index), p);
2726 * Return TRUE if the pmap is in shape to trivially
2727 * pre-fault the specified address.
2729 * Returns FALSE if it would be non-trivial or if a
2730 * pte is already loaded into the slot.
2735 pmap_prefault_ok(pmap_t pmap, vm_offset_t addr)
2740 lwkt_gettoken(&vm_token);
2741 if ((*pmap_pde(pmap, addr)) == 0) {
2744 pte = (unsigned *) vtopte(addr);
2745 ret = (*pte) ? 0 : 1;
2747 lwkt_reltoken(&vm_token);
2752 * Change the wiring attribute for a map/virtual-adderss pair. The mapping
2753 * must already exist.
2758 pmap_change_wiring(pmap_t pmap, vm_offset_t va, boolean_t wired,
2759 vm_map_entry_t entry __unused)
2766 lwkt_gettoken(&vm_token);
2767 pte = pmap_pte(pmap, va);
2769 if (wired && !pmap_pte_w(pte))
2770 pmap->pm_stats.wired_count++;
2771 else if (!wired && pmap_pte_w(pte))
2772 pmap->pm_stats.wired_count--;
2775 * Wiring is not a hardware characteristic so there is no need to
2776 * invalidate TLB. However, in an SMP environment we must use
2777 * a locked bus cycle to update the pte (if we are not using
2778 * the pmap_inval_*() API that is)... it's ok to do this for simple
2783 atomic_set_int(pte, PG_W);
2785 atomic_clear_int(pte, PG_W);
2788 atomic_set_int_nonlocked(pte, PG_W);
2790 atomic_clear_int_nonlocked(pte, PG_W);
2792 lwkt_reltoken(&vm_token);
2796 * Copy the range specified by src_addr/len from the source map to the
2797 * range dst_addr/len in the destination map.
2799 * This routine is only advisory and need not do anything.
2804 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
2805 vm_size_t len, vm_offset_t src_addr)
2811 * Zero the specified PA by mapping the page into KVM and clearing its
2817 pmap_zero_page(vm_paddr_t phys)
2819 struct mdglobaldata *gd = mdcpu;
2822 if (*(int *)gd->gd_CMAP3)
2823 panic("pmap_zero_page: CMAP3 busy");
2824 *(int *)gd->gd_CMAP3 =
2825 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2826 cpu_invlpg(gd->gd_CADDR3);
2827 bzero(gd->gd_CADDR3, PAGE_SIZE);
2828 *(int *) gd->gd_CMAP3 = 0;
2833 * Assert that a page is empty, panic if it isn't.
2838 pmap_page_assertzero(vm_paddr_t phys)
2840 struct mdglobaldata *gd = mdcpu;
2844 if (*(int *)gd->gd_CMAP3)
2845 panic("pmap_zero_page: CMAP3 busy");
2846 *(int *)gd->gd_CMAP3 =
2847 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2848 cpu_invlpg(gd->gd_CADDR3);
2849 for (i = 0; i < PAGE_SIZE; i += 4) {
2850 if (*(int *)((char *)gd->gd_CADDR3 + i) != 0) {
2851 panic("pmap_page_assertzero() @ %p not zero!",
2852 (void *)gd->gd_CADDR3);
2855 *(int *) gd->gd_CMAP3 = 0;
2860 * Zero part of a physical page by mapping it into memory and clearing
2861 * its contents with bzero.
2863 * off and size may not cover an area beyond a single hardware page.
2868 pmap_zero_page_area(vm_paddr_t phys, int off, int size)
2870 struct mdglobaldata *gd = mdcpu;
2873 if (*(int *) gd->gd_CMAP3)
2874 panic("pmap_zero_page: CMAP3 busy");
2875 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2876 cpu_invlpg(gd->gd_CADDR3);
2877 bzero((char *)gd->gd_CADDR3 + off, size);
2878 *(int *) gd->gd_CMAP3 = 0;
2883 * Copy the physical page from the source PA to the target PA.
2884 * This function may be called from an interrupt. No locking
2890 pmap_copy_page(vm_paddr_t src, vm_paddr_t dst)
2892 struct mdglobaldata *gd = mdcpu;
2895 if (*(int *) gd->gd_CMAP1)
2896 panic("pmap_copy_page: CMAP1 busy");
2897 if (*(int *) gd->gd_CMAP2)
2898 panic("pmap_copy_page: CMAP2 busy");
2900 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2901 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2903 cpu_invlpg(gd->gd_CADDR1);
2904 cpu_invlpg(gd->gd_CADDR2);
2906 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
2908 *(int *) gd->gd_CMAP1 = 0;
2909 *(int *) gd->gd_CMAP2 = 0;
2914 * Copy the physical page from the source PA to the target PA.
2915 * This function may be called from an interrupt. No locking
2921 pmap_copy_page_frag(vm_paddr_t src, vm_paddr_t dst, size_t bytes)
2923 struct mdglobaldata *gd = mdcpu;
2926 if (*(int *) gd->gd_CMAP1)
2927 panic("pmap_copy_page: CMAP1 busy");
2928 if (*(int *) gd->gd_CMAP2)
2929 panic("pmap_copy_page: CMAP2 busy");
2931 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2932 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2934 cpu_invlpg(gd->gd_CADDR1);
2935 cpu_invlpg(gd->gd_CADDR2);
2937 bcopy((char *)gd->gd_CADDR1 + (src & PAGE_MASK),
2938 (char *)gd->gd_CADDR2 + (dst & PAGE_MASK),
2941 *(int *) gd->gd_CMAP1 = 0;
2942 *(int *) gd->gd_CMAP2 = 0;
2947 * Returns true if the pmap's pv is one of the first
2948 * 16 pvs linked to from this page. This count may
2949 * be changed upwards or downwards in the future; it
2950 * is only necessary that true be returned for a small
2951 * subset of pmaps for proper page aging.
2956 pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
2961 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2964 lwkt_gettoken(&vm_token);
2965 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2966 if (pv->pv_pmap == pmap) {
2967 lwkt_reltoken(&vm_token);
2974 lwkt_reltoken(&vm_token);
2979 * Remove all pages from specified address space
2980 * this aids process exit speeds. Also, this code
2981 * is special cased for current process only, but
2982 * can have the more generic (and slightly slower)
2983 * mode enabled. This is much faster than pmap_remove
2984 * in the case of running down an entire address space.
2989 pmap_remove_pages(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2992 unsigned *pte, tpte;
2995 pmap_inval_info info;
2997 int32_t save_generation;
2999 lp = curthread->td_lwp;
3000 if (lp && pmap == vmspace_pmap(lp->lwp_vmspace))
3005 if (pmap->pm_pteobj)
3006 vm_object_hold(pmap->pm_pteobj);
3007 lwkt_gettoken(&vm_token);
3008 pmap_inval_init(&info);
3010 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); pv; pv = npv) {
3011 if (pv->pv_va >= eva || pv->pv_va < sva) {
3012 npv = TAILQ_NEXT(pv, pv_plist);
3016 KKASSERT(pmap == pv->pv_pmap);
3019 pte = (unsigned *)vtopte(pv->pv_va);
3021 pte = pmap_pte_quick(pmap, pv->pv_va);
3023 pmap_inval_interlock(&info, pmap, pv->pv_va);
3026 * We cannot remove wired pages from a process' mapping
3030 pmap_inval_deinterlock(&info, pmap);
3031 npv = TAILQ_NEXT(pv, pv_plist);
3035 tpte = loadandclear(pte);
3036 pmap_inval_deinterlock(&info, pmap);
3038 m = PHYS_TO_VM_PAGE(tpte);
3039 test_m_maps_pv(m, pv);
3041 KASSERT(m < &vm_page_array[vm_page_array_size],
3042 ("pmap_remove_pages: bad tpte %x", tpte));
3044 KKASSERT(pmap->pm_stats.resident_count > 0);
3045 --pmap->pm_stats.resident_count;
3048 * Update the vm_page_t clean and reference bits.
3054 npv = TAILQ_NEXT(pv, pv_plist);
3056 KKASSERT(pv->pv_m == m);
3057 KKASSERT(pv->pv_pmap == pmap);
3059 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
3060 save_generation = ++pmap->pm_generation;
3062 m->md.pv_list_count--;
3064 atomic_add_int(&m->object->agg_pv_list_count, -1);
3065 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3066 if (TAILQ_EMPTY(&m->md.pv_list))
3067 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
3069 pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem, &info);
3073 * Restart the scan if we blocked during the unuse or free
3074 * calls and other removals were made.
3076 if (save_generation != pmap->pm_generation) {
3077 kprintf("Warning: pmap_remove_pages race-A avoided\n");
3078 npv = TAILQ_FIRST(&pmap->pm_pvlist);
3081 pmap_inval_done(&info);
3082 lwkt_reltoken(&vm_token);
3083 if (pmap->pm_pteobj)
3084 vm_object_drop(pmap->pm_pteobj);
3088 * pmap_testbit tests bits in pte's
3089 * note that the testbit/clearbit routines are inline,
3090 * and a lot of things compile-time evaluate.
3092 * The caller must hold vm_token.
3095 pmap_testbit(vm_page_t m, int bit)
3100 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3103 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3106 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3108 * if the bit being tested is the modified bit, then
3109 * mark clean_map and ptes as never
3112 if (bit & (PG_A|PG_M)) {
3113 if (!pmap_track_modified(pv->pv_va))
3117 #if defined(PMAP_DIAGNOSTIC)
3119 kprintf("Null pmap (tb) at va: %p\n",
3124 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3133 * This routine is used to modify bits in ptes
3135 * The caller must hold vm_token.
3137 static __inline void
3138 pmap_clearbit(vm_page_t m, int bit)
3140 struct pmap_inval_info info;
3145 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3148 pmap_inval_init(&info);
3151 * Loop over all current mappings setting/clearing as appropos If
3152 * setting RO do we need to clear the VAC?
3154 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3156 * don't write protect pager mappings
3159 if (!pmap_track_modified(pv->pv_va))
3163 #if defined(PMAP_DIAGNOSTIC)
3165 kprintf("Null pmap (cb) at va: %p\n",
3172 * Careful here. We can use a locked bus instruction to
3173 * clear PG_A or PG_M safely but we need to synchronize
3174 * with the target cpus when we mess with PG_RW.
3176 * We do not have to force synchronization when clearing
3177 * PG_M even for PTEs generated via virtual memory maps,
3178 * because the virtual kernel will invalidate the pmap
3179 * entry when/if it needs to resynchronize the Modify bit.
3182 pmap_inval_interlock(&info, pv->pv_pmap, pv->pv_va);
3183 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3190 atomic_clear_int(pte, PG_M|PG_RW);
3193 * The cpu may be trying to set PG_M
3194 * simultaniously with our clearing
3197 if (!atomic_cmpset_int(pte, pbits,
3201 } else if (bit == PG_M) {
3203 * We could also clear PG_RW here to force
3204 * a fault on write to redetect PG_M for
3205 * virtual kernels, but it isn't necessary
3206 * since virtual kernels invalidate the pte
3207 * when they clear the VPTE_M bit in their
3208 * virtual page tables.
3210 atomic_clear_int(pte, PG_M);
3212 atomic_clear_int(pte, bit);
3216 pmap_inval_deinterlock(&info, pv->pv_pmap);
3218 pmap_inval_done(&info);
3222 * Lower the permission for all mappings to a given page.
3227 pmap_page_protect(vm_page_t m, vm_prot_t prot)
3229 if ((prot & VM_PROT_WRITE) == 0) {
3230 lwkt_gettoken(&vm_token);
3231 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) {
3232 pmap_clearbit(m, PG_RW);
3233 vm_page_flag_clear(m, PG_WRITEABLE);
3237 lwkt_reltoken(&vm_token);
3242 * Return the physical address given a physical page index.
3247 pmap_phys_address(vm_pindex_t ppn)
3249 return (i386_ptob(ppn));
3253 * Return a count of reference bits for a page, clearing those bits.
3254 * It is not necessary for every reference bit to be cleared, but it
3255 * is necessary that 0 only be returned when there are truly no
3256 * reference bits set.
3261 pmap_ts_referenced(vm_page_t m)
3263 pv_entry_t pv, pvf, pvn;
3267 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3270 lwkt_gettoken(&vm_token);
3272 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3277 pvn = TAILQ_NEXT(pv, pv_list);
3279 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3280 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
3282 if (!pmap_track_modified(pv->pv_va))
3285 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3287 if (pte && (*pte & PG_A)) {
3289 atomic_clear_int(pte, PG_A);
3291 atomic_clear_int_nonlocked(pte, PG_A);
3298 } while ((pv = pvn) != NULL && pv != pvf);
3301 lwkt_reltoken(&vm_token);
3307 * Return whether or not the specified physical page was modified
3308 * in any physical maps.
3313 pmap_is_modified(vm_page_t m)
3317 lwkt_gettoken(&vm_token);
3318 res = pmap_testbit(m, PG_M);
3319 lwkt_reltoken(&vm_token);
3324 * Clear the modify bits on the specified physical page.
3329 pmap_clear_modify(vm_page_t m)
3331 lwkt_gettoken(&vm_token);
3332 pmap_clearbit(m, PG_M);
3333 lwkt_reltoken(&vm_token);
3337 * Clear the reference bit on the specified physical page.
3342 pmap_clear_reference(vm_page_t m)
3344 lwkt_gettoken(&vm_token);
3345 pmap_clearbit(m, PG_A);
3346 lwkt_reltoken(&vm_token);
3350 * Miscellaneous support routines follow
3352 * Called from the low level boot code only.
3355 i386_protection_init(void)
3359 kp = protection_codes;
3360 for (prot = 0; prot < 8; prot++) {
3362 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE:
3364 * Read access is also 0. There isn't any execute bit,
3365 * so just make it readable.
3367 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE:
3368 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE:
3369 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE:
3372 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE:
3373 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE:
3374 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE:
3375 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE:
3383 * Map a set of physical memory pages into the kernel virtual
3384 * address space. Return a pointer to where it is mapped. This
3385 * routine is intended to be used for mapping device memory,
3388 * NOTE: we can't use pgeflag unless we invalidate the pages one at
3394 pmap_mapdev(vm_paddr_t pa, vm_size_t size)
3396 vm_offset_t va, tmpva, offset;
3399 offset = pa & PAGE_MASK;
3400 size = roundup(offset + size, PAGE_SIZE);
3402 va = kmem_alloc_nofault(&kernel_map, size, PAGE_SIZE);
3404 panic("pmap_mapdev: Couldn't alloc kernel virtual memory");
3407 for (tmpva = va; size > 0;) {
3408 pte = (unsigned *)vtopte(tmpva);
3409 *pte = pa | PG_RW | PG_V; /* | pgeflag; */
3417 return ((void *)(va + offset));
3421 pmap_mapdev_uncacheable(vm_paddr_t pa, vm_size_t size)
3423 vm_offset_t va, tmpva, offset;
3426 offset = pa & PAGE_MASK;
3427 size = roundup(offset + size, PAGE_SIZE);
3429 va = kmem_alloc_nofault(&kernel_map, size, PAGE_SIZE);
3431 panic("pmap_mapdev_uncacheable: "
3432 "Couldn't alloc kernel virtual memory");
3436 for (tmpva = va; size > 0;) {
3437 pte = (unsigned *)vtopte(tmpva);
3438 *pte = pa | PG_RW | PG_V | PG_N; /* | pgeflag; */
3446 return ((void *)(va + offset));
3453 pmap_unmapdev(vm_offset_t va, vm_size_t size)
3455 vm_offset_t base, offset;
3457 base = va & PG_FRAME;
3458 offset = va & PAGE_MASK;
3459 size = roundup(offset + size, PAGE_SIZE);
3460 pmap_qremove(va, size >> PAGE_SHIFT);
3461 kmem_free(&kernel_map, base, size);
3465 * Perform the pmap work for mincore
3467 * The caller must hold vm_token if the caller wishes a stable result,
3468 * and even in that case some bits can change due to third party accesses
3474 pmap_mincore(pmap_t pmap, vm_offset_t addr)
3476 unsigned *ptep, pte;
3480 lwkt_gettoken(&vm_token);
3481 ptep = pmap_pte(pmap, addr);
3483 if (ptep && (pte = *ptep) != 0) {
3486 val = MINCORE_INCORE;
3487 if ((pte & PG_MANAGED) == 0)
3490 pa = pte & PG_FRAME;
3492 m = PHYS_TO_VM_PAGE(pa);
3498 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER;
3499 } else if (m->dirty || pmap_is_modified(m)) {
3501 * Modified by someone else
3503 val |= MINCORE_MODIFIED_OTHER;
3510 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER;
3511 } else if ((m->flags & PG_REFERENCED) ||
3512 pmap_ts_referenced(m)) {
3514 * Referenced by someone else
3516 val |= MINCORE_REFERENCED_OTHER;
3517 vm_page_flag_set(m, PG_REFERENCED);
3521 lwkt_reltoken(&vm_token);
3526 * Replace p->p_vmspace with a new one. If adjrefs is non-zero the new
3527 * vmspace will be ref'd and the old one will be deref'd.
3529 * cr3 will be reloaded if any lwp is the current lwp.
3531 * Only called with new VM spaces.
3532 * The process must have only a single thread.
3533 * The process must hold the vmspace->vm_map.token for oldvm and newvm
3534 * No other requirements.
3537 pmap_replacevm(struct proc *p, struct vmspace *newvm, int adjrefs)
3539 struct vmspace *oldvm;
3542 oldvm = p->p_vmspace;
3543 if (oldvm != newvm) {
3545 sysref_get(&newvm->vm_sysref);
3546 p->p_vmspace = newvm;
3547 KKASSERT(p->p_nthreads == 1);
3548 lp = RB_ROOT(&p->p_lwp_tree);
3549 pmap_setlwpvm(lp, newvm);
3551 sysref_put(&oldvm->vm_sysref);
3556 * Set the vmspace for a LWP. The vmspace is almost universally set the
3557 * same as the process vmspace, but virtual kernels need to swap out contexts
3558 * on a per-lwp basis.
3560 * Always called with a lp under the caller's direct control, either
3561 * unscheduled or the current lwp.
3566 pmap_setlwpvm(struct lwp *lp, struct vmspace *newvm)
3568 struct vmspace *oldvm;
3571 oldvm = lp->lwp_vmspace;
3573 if (oldvm != newvm) {
3574 lp->lwp_vmspace = newvm;
3575 if (curthread->td_lwp == lp) {
3576 pmap = vmspace_pmap(newvm);
3578 atomic_set_cpumask(&pmap->pm_active, mycpu->gd_cpumask);
3579 if (pmap->pm_active & CPUMASK_LOCK)
3580 pmap_interlock_wait(newvm);
3582 pmap->pm_active |= 1;
3584 #if defined(SWTCH_OPTIM_STATS)
3587 curthread->td_pcb->pcb_cr3 = vtophys(pmap->pm_pdir);
3588 load_cr3(curthread->td_pcb->pcb_cr3);
3589 pmap = vmspace_pmap(oldvm);
3591 atomic_clear_cpumask(&pmap->pm_active,
3594 pmap->pm_active &= ~(cpumask_t)1;
3602 * Called when switching to a locked pmap, used to interlock against pmaps
3603 * undergoing modifications to prevent us from activating the MMU for the
3604 * target pmap until all such modifications have completed. We have to do
3605 * this because the thread making the modifications has already set up its
3606 * SMP synchronization mask.
3611 pmap_interlock_wait(struct vmspace *vm)
3613 struct pmap *pmap = &vm->vm_pmap;
3615 if (pmap->pm_active & CPUMASK_LOCK) {
3617 DEBUG_PUSH_INFO("pmap_interlock_wait");
3618 while (pmap->pm_active & CPUMASK_LOCK) {
3620 lwkt_process_ipiq();
3630 * Return a page-directory alignment hint for device mappings which will
3631 * allow the use of super-pages for the mapping.
3636 pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size)
3639 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) {
3643 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
3648 * Return whether the PGE flag is supported globally.
3653 pmap_get_pgeflag(void)
3659 * Used by kmalloc/kfree, page already exists at va
3662 pmap_kvtom(vm_offset_t va)
3664 return(PHYS_TO_VM_PAGE(*vtopte(va) & PG_FRAME));
3668 pmap_object_init(vm_object_t object)
3674 pmap_object_free(vm_object_t object)