kernel: Make SMP support default (and non-optional).
[dragonfly.git] / sys / platform / pc32 / i386 / pmap.c
CommitLineData
984263bc 1/*
4107b0c0
MD
2 * (MPSAFE)
3 *
984263bc
MD
4 * Copyright (c) 1991 Regents of the University of California.
5 * All rights reserved.
6 * Copyright (c) 1994 John S. Dyson
7 * All rights reserved.
8 * Copyright (c) 1994 David Greenman
9 * All rights reserved.
10 *
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.
14 *
15 * Redistribution and use in source and binary forms, with or without
16 * modification, are permitted provided that the following conditions
17 * are met:
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.
30 *
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
41 * SUCH DAMAGE.
42 *
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 $
45 */
46
47/*
4107b0c0 48 * Manages physical address maps.
984263bc 49 *
b12defdc 50 * In most cases we hold page table pages busy in order to manipulate them.
984263bc 51 */
5926987a
MD
52/*
53 * PMAP_DEBUG - see platform/pc32/include/pmap.h
54 */
984263bc
MD
55
56#include "opt_disable_pse.h"
57#include "opt_pmap.h"
58#include "opt_msgbuf.h"
984263bc
MD
59
60#include <sys/param.h>
61#include <sys/systm.h>
62#include <sys/kernel.h>
63#include <sys/proc.h>
64#include <sys/msgbuf.h>
65#include <sys/vmmeter.h>
66#include <sys/mman.h>
b12defdc 67#include <sys/thread.h>
984263bc
MD
68
69#include <vm/vm.h>
70#include <vm/vm_param.h>
71#include <sys/sysctl.h>
72#include <sys/lock.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>
81
82#include <sys/user.h>
e0e69b7d 83#include <sys/thread2.h>
e3161323 84#include <sys/sysref2.h>
b12defdc 85#include <sys/spinlock2.h>
90244566 86#include <vm/vm_page2.h>
984263bc
MD
87
88#include <machine/cputypes.h>
89#include <machine/md_var.h>
90#include <machine/specialreg.h>
984263bc 91#include <machine/smp.h>
a9295349 92#include <machine_base/apic/apicreg.h>
85100692 93#include <machine/globaldata.h>
0f7a3396
MD
94#include <machine/pmap.h>
95#include <machine/pmap_inval.h>
984263bc
MD
96
97#define PMAP_KEEP_PDIRS
98#ifndef PMAP_SHPGPERPROC
99#define PMAP_SHPGPERPROC 200
948209ce 100#define PMAP_PVLIMIT 1400000 /* i386 kvm problems */
984263bc
MD
101#endif
102
103#if defined(DIAGNOSTIC)
104#define PMAP_DIAGNOSTIC
105#endif
106
107#define MINPV 2048
108
109#if !defined(PMAP_DIAGNOSTIC)
110#define PMAP_INLINE __inline
111#else
112#define PMAP_INLINE
113#endif
114
115/*
116 * Get PDEs and PTEs for user/kernel address space
117 */
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])
120
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)
126
984263bc
MD
127/*
128 * Given a map and a machine independent protection code,
129 * convert to a vax protection code.
130 */
639a9b43
MD
131#define pte_prot(m, p) \
132 (protection_codes[p & (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)])
984263bc
MD
133static int protection_codes[8];
134
fbbaeba3 135struct pmap kernel_pmap;
54a764e8
MD
136static TAILQ_HEAD(,pmap) pmap_list = TAILQ_HEAD_INITIALIZER(pmap_list);
137
e880033d 138vm_paddr_t avail_start; /* PA of first available physical page */
6ef943a3 139vm_paddr_t avail_end; /* PA of last available physical page */
e880033d 140vm_offset_t virtual_start; /* VA of first avail page (after kernel bss) */
984263bc 141vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */
791c6551
MD
142vm_offset_t virtual2_start;
143vm_offset_t virtual2_end;
c439ad8f
MD
144vm_offset_t KvaStart; /* VA start of KVA space */
145vm_offset_t KvaEnd; /* VA end of KVA space (non-inclusive) */
146vm_offset_t KvaSize; /* max size of kernel virtual address space */
984263bc
MD
147static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */
148static int pgeflag; /* PG_G or-in */
149static int pseflag; /* PG_PS or-in */
150
151static vm_object_t kptobj;
152
153static int nkpt;
154vm_offset_t kernel_vm_end;
155
156/*
157 * Data for the pv entry allocation mechanism
158 */
159static vm_zone_t pvzone;
160static struct vm_zone pvzone_store;
161static struct vm_object pvzone_obj;
162static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0;
163static int pmap_pagedaemon_waken = 0;
164static struct pv_entry *pvinit;
165
a93980ab
MD
166/*
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.
170 */
171static int dreadful_invltlb;
172
173SYSCTL_INT(_vm, OID_AUTO, dreadful_invltlb,
9733f757 174 CTLFLAG_RW, &dreadful_invltlb, 0, "Debugging sysctl to force invltlb on pmap operations");
a93980ab 175
984263bc
MD
176/*
177 * All those kernel PT submaps that BSD is so fond of
178 */
4090d6ff 179pt_entry_t *CMAP1 = NULL, *ptmmap;
4c0cc8bb 180caddr_t CADDR1 = NULL, ptvmmap = NULL;
984263bc 181static pt_entry_t *msgbufmap;
4090d6ff 182struct msgbuf *msgbufp=NULL;
984263bc
MD
183
184/*
185 * Crashdump maps.
186 */
187static pt_entry_t *pt_crashdumpmap;
188static caddr_t crashdumpmap;
189
984263bc 190extern pt_entry_t *SMPpt;
984263bc 191
3ae0cd58
RG
192static PMAP_INLINE void free_pv_entry (pv_entry_t pv);
193static unsigned * get_ptbase (pmap_t pmap);
194static pv_entry_t get_pv_entry (void);
195static void i386_protection_init (void);
5e8d0349 196static __inline void pmap_clearbit (vm_page_t m, int bit);
3ae0cd58 197
554cf9ac
MD
198static void pmap_remove_all (vm_page_t m);
199static void pmap_remove_pte (struct pmap *pmap, unsigned *ptq,
0f7a3396
MD
200 vm_offset_t sva, pmap_inval_info_t info);
201static void pmap_remove_page (struct pmap *pmap,
202 vm_offset_t va, pmap_inval_info_t info);
554cf9ac 203static void pmap_remove_entry (struct pmap *pmap, vm_page_t m,
0f7a3396 204 vm_offset_t va, pmap_inval_info_t info);
3ae0cd58 205static boolean_t pmap_testbit (vm_page_t m, int bit);
2bb9cc6f
MD
206static void pmap_insert_entry (pmap_t pmap, pv_entry_t pv,
207 vm_offset_t va, vm_page_t mpte, vm_page_t m);
3ae0cd58
RG
208
209static vm_page_t pmap_allocpte (pmap_t pmap, vm_offset_t va);
210
211static int pmap_release_free_page (pmap_t pmap, vm_page_t p);
212static vm_page_t _pmap_allocpte (pmap_t pmap, unsigned ptepindex);
213static unsigned * pmap_pte_quick (pmap_t pmap, vm_offset_t va);
214static vm_page_t pmap_page_lookup (vm_object_t object, vm_pindex_t pindex);
554cf9ac 215static void pmap_unuse_pt (pmap_t, vm_offset_t, vm_page_t, pmap_inval_info_t);
984263bc
MD
216static vm_offset_t pmap_kmem_choose(vm_offset_t addr);
217
3321ee05
MD
218static void pmap_hold(pmap_t pmap);
219static void pmap_drop(pmap_t pmap);
220static void pmap_wait(pmap_t pmap, int count);
221
984263bc
MD
222static unsigned pdir4mb;
223
840de426
MD
224/*
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)
229 */
4107b0c0
MD
230static
231vm_offset_t
840de426
MD
232pmap_kmem_choose(vm_offset_t addr)
233{
234 vm_offset_t newaddr = addr;
235#ifndef DISABLE_PSE
236 if (cpu_feature & CPUID_PSE) {
237 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1);
238 }
239#endif
240 return newaddr;
241}
242
984263bc 243/*
4107b0c0
MD
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.
e0e69b7d 246 *
4107b0c0
MD
247 * The caller must hold vm_token and the returned value is only valid
248 * until the caller blocks or releases the token.
984263bc 249 */
4107b0c0
MD
250static
251unsigned *
840de426 252pmap_pte(pmap_t pmap, vm_offset_t va)
984263bc
MD
253{
254 unsigned *pdeaddr;
255
4107b0c0 256 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc
MD
257 if (pmap) {
258 pdeaddr = (unsigned *) pmap_pde(pmap, va);
259 if (*pdeaddr & PG_PS)
260 return pdeaddr;
4107b0c0 261 if (*pdeaddr)
984263bc 262 return get_ptbase(pmap) + i386_btop(va);
984263bc
MD
263 }
264 return (0);
265}
266
4107b0c0
MD
267/*
268 * pmap_pte using the kernel_pmap
269 *
270 * Used for debugging, no requirements.
271 */
272unsigned *
273pmap_kernel_pte(vm_offset_t va)
274{
275 unsigned *pdeaddr;
276
277 pdeaddr = (unsigned *) pmap_pde(&kernel_pmap, va);
278 if (*pdeaddr & PG_PS)
279 return pdeaddr;
280 if (*pdeaddr)
281 return (unsigned *)vtopte(va);
282 return(0);
283}
284
984263bc 285/*
e0e69b7d
MD
286 * pmap_pte_quick:
287 *
c1692ddf
MD
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.
e0e69b7d 292 *
c1692ddf
MD
293 * Should only be called while in a critical section.
294 *
4107b0c0
MD
295 * The caller must hold vm_token and the returned value is only valid
296 * until the caller blocks or releases the token.
984263bc 297 */
4107b0c0
MD
298static
299unsigned *
840de426 300pmap_pte_quick(pmap_t pmap, vm_offset_t va)
984263bc 301{
840de426
MD
302 struct mdglobaldata *gd = mdcpu;
303 unsigned pde, newpf;
304
4107b0c0 305 ASSERT_LWKT_TOKEN_HELD(&vm_token);
840de426
MD
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? */
fbbaeba3 310 if ((pmap == &kernel_pmap) ||
840de426
MD
311 (frame == (((unsigned) PTDpde) & PG_FRAME))) {
312 return (unsigned *) PTmap + index;
313 }
314 newpf = pde & PG_FRAME;
4107b0c0
MD
315 if (((*(unsigned *)gd->gd_PMAP1) & PG_FRAME) != newpf) {
316 *(unsigned *)gd->gd_PMAP1 = newpf | PG_RW | PG_V;
840de426
MD
317 cpu_invlpg(gd->gd_PADDR1);
318 }
06bb314f 319 return gd->gd_PADDR1 + (index & (NPTEPG - 1));
984263bc 320 }
840de426 321 return (0);
984263bc
MD
322}
323
840de426 324
984263bc 325/*
4107b0c0 326 * Bootstrap the system enough to run with virtual memory.
984263bc 327 *
4107b0c0
MD
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]
984263bc
MD
334 */
335void
f123d5a1 336pmap_bootstrap(vm_paddr_t firstaddr, vm_paddr_t loadaddr)
984263bc
MD
337{
338 vm_offset_t va;
339 pt_entry_t *pte;
85100692 340 struct mdglobaldata *gd;
984263bc 341 int i;
81c04d07 342 int pg;
984263bc 343
c439ad8f
MD
344 KvaStart = (vm_offset_t)VADDR(PTDPTDI, 0);
345 KvaSize = (vm_offset_t)VADDR(APTDPTDI, 0) - KvaStart;
346 KvaEnd = KvaStart + KvaSize;
347
984263bc
MD
348 avail_start = firstaddr;
349
350 /*
e880033d
MD
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
984263bc
MD
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.
357 */
e880033d
MD
358 virtual_start = (vm_offset_t) KERNBASE + firstaddr;
359 virtual_start = pmap_kmem_choose(virtual_start);
c439ad8f 360 virtual_end = VADDR(KPTDI+NKPDE-1, NPTEPG-1);
984263bc
MD
361
362 /*
363 * Initialize protection array.
364 */
365 i386_protection_init();
366
367 /*
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).
b12defdc
MD
371 *
372 * The kernel_pmap's pm_pteobj is used only for locking and not
373 * for mmu pages.
984263bc 374 */
fbbaeba3
MD
375 kernel_pmap.pm_pdir = (pd_entry_t *)(KERNBASE + (u_int)IdlePTD);
376 kernel_pmap.pm_count = 1;
c2fb025d 377 kernel_pmap.pm_active = (cpumask_t)-1 & ~CPUMASK_LOCK;
b12defdc 378 kernel_pmap.pm_pteobj = &kernel_object;
fbbaeba3 379 TAILQ_INIT(&kernel_pmap.pm_pvlist);
b12defdc
MD
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");
984263bc
MD
383 nkpt = NKPT;
384
385 /*
386 * Reserve some special page table entries/VA space for temporary
387 * mapping of pages.
388 */
389#define SYSMAP(c, p, v, n) \
390 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n);
391
e880033d 392 va = virtual_start;
4107b0c0 393 pte = (pt_entry_t *) pmap_kernel_pte(va);
984263bc
MD
394
395 /*
396 * CMAP1/CMAP2 are used for zeroing and copying pages.
397 */
398 SYSMAP(caddr_t, CMAP1, CADDR1, 1)
984263bc
MD
399
400 /*
401 * Crashdump maps.
402 */
403 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS);
404
e731d345
MD
405 /*
406 * ptvmmap is used for reading arbitrary physical pages via
407 * /dev/mem.
408 */
409 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
410
984263bc
MD
411 /*
412 * msgbufp is used to map the system message buffer.
413 * XXX msgbufmap is not used.
414 */
415 SYSMAP(struct msgbuf *, msgbufmap, msgbufp,
416 atop(round_page(MSGBUF_SIZE)))
417
e880033d 418 virtual_start = va;
984263bc 419
17a9f566 420 *(int *) CMAP1 = 0;
984263bc
MD
421 for (i = 0; i < NKPT; i++)
422 PTD[i] = 0;
423
a2a5ad0d
MD
424 /*
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
428 */
a2a5ad0d 429 pgeflag = 0;
984263bc
MD
430
431/*
432 * Initialize the 4MB page size flag
433 */
434 pseflag = 0;
435/*
436 * The 4MB page version of the initial
437 * kernel page mapping.
438 */
439 pdir4mb = 0;
440
441#if !defined(DISABLE_PSE)
442 if (cpu_feature & CPUID_PSE) {
443 unsigned ptditmp;
444 /*
445 * Note that we have enabled PSE mode
446 */
447 pseflag = PG_PS;
448 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
449 ptditmp &= ~(NBPDR - 1);
450 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
451 pdir4mb = ptditmp;
984263bc
MD
452 }
453#endif
984263bc 454
81c04d07
MD
455 /*
456 * We need to finish setting up the globaldata page for the BSP.
457 * locore has already populated the page table for the mdglobaldata
458 * portion.
459 */
460 pg = MDGLOBALDATA_BASEALLOC_PAGES;
85100692 461 gd = &CPU_prvspace[0].mdglobaldata;
81c04d07
MD
462 gd->gd_CMAP1 = &SMPpt[pg + 0];
463 gd->gd_CMAP2 = &SMPpt[pg + 1];
464 gd->gd_CMAP3 = &SMPpt[pg + 2];
465 gd->gd_PMAP1 = &SMPpt[pg + 3];
9388fcaa 466 gd->gd_GDMAP1 = &PTD[APTDPTDI];
85100692
MD
467 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
468 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
469 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
470 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
9388fcaa 471 gd->gd_GDADDR1= (unsigned *)VADDR(APTDPTDI, 0);
984263bc 472
0f7a3396 473 cpu_invltlb();
984263bc
MD
474}
475
984263bc
MD
476/*
477 * Set 4mb pdir for mp startup
478 */
479void
480pmap_set_opt(void)
481{
482 if (pseflag && (cpu_feature & CPUID_PSE)) {
483 load_cr4(rcr4() | CR4_PSE);
72740893 484 if (pdir4mb && mycpu->gd_cpuid == 0) { /* only on BSP */
fbbaeba3 485 kernel_pmap.pm_pdir[KPTDI] =
984263bc
MD
486 PTD[KPTDI] = (pd_entry_t)pdir4mb;
487 cpu_invltlb();
488 }
489 }
490}
984263bc
MD
491
492/*
4107b0c0
MD
493 * Initialize the pmap module, called by vm_init()
494 *
495 * Called from the low level boot code only.
984263bc
MD
496 */
497void
e7252eda 498pmap_init(void)
984263bc
MD
499{
500 int i;
501 int initial_pvs;
502
503 /*
504 * object for kernel page table pages
505 */
506 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
507
508 /*
509 * Allocate memory for random pmap data structures. Includes the
510 * pv_head_table.
511 */
512
513 for(i = 0; i < vm_page_array_size; i++) {
514 vm_page_t m;
515
516 m = &vm_page_array[i];
517 TAILQ_INIT(&m->md.pv_list);
518 m->md.pv_list_count = 0;
519 }
520
521 /*
522 * init the pv free list
523 */
524 initial_pvs = vm_page_array_size;
525 if (initial_pvs < MINPV)
526 initial_pvs = MINPV;
527 pvzone = &pvzone_store;
948209ce
MD
528 pvinit = (void *)kmem_alloc(&kernel_map,
529 initial_pvs * sizeof (struct pv_entry));
530 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry),
531 pvinit, initial_pvs);
984263bc
MD
532
533 /*
534 * Now it is safe to enable pv_table recording.
535 */
536 pmap_initialized = TRUE;
537}
538
539/*
540 * Initialize the address space (zone) for the pv_entries. Set a
541 * high water mark so that the system can recover from excessive
542 * numbers of pv entries.
4107b0c0
MD
543 *
544 * Called from the low level boot code only.
984263bc
MD
545 */
546void
f123d5a1 547pmap_init2(void)
984263bc
MD
548{
549 int shpgperproc = PMAP_SHPGPERPROC;
948209ce 550 int entry_max;
984263bc
MD
551
552 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
553 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
948209ce
MD
554
555#ifdef PMAP_PVLIMIT
556 /*
557 * Horrible hack for systems with a lot of memory running i386.
558 * the calculated pv_entry_max can wind up eating a ton of KVM
559 * so put a cap on the number of entries if the user did not
560 * change any of the values. This saves about 44MB of KVM on
561 * boxes with 3+GB of ram.
562 *
563 * On the flip side, this makes it more likely that some setups
564 * will run out of pv entries. Those sysads will have to bump
565 * the limit up with vm.pamp.pv_entries or vm.pmap.shpgperproc.
566 */
567 if (shpgperproc == PMAP_SHPGPERPROC) {
568 if (pv_entry_max > PMAP_PVLIMIT)
569 pv_entry_max = PMAP_PVLIMIT;
570 }
571#endif
984263bc
MD
572 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
573 pv_entry_high_water = 9 * (pv_entry_max / 10);
948209ce
MD
574
575 /*
576 * Subtract out pages already installed in the zone (hack)
577 */
578 entry_max = pv_entry_max - vm_page_array_size;
579 if (entry_max <= 0)
580 entry_max = 1;
581
582 zinitna(pvzone, &pvzone_obj, NULL, 0, entry_max, ZONE_INTERRUPT, 1);
984263bc
MD
583}
584
585
586/***************************************************
587 * Low level helper routines.....
588 ***************************************************/
589
5926987a
MD
590#ifdef PMAP_DEBUG
591
592static void
593test_m_maps_pv(vm_page_t m, pv_entry_t pv)
594{
595 pv_entry_t spv;
596
74b9d1ec 597 crit_enter();
5926987a
MD
598#ifdef PMAP_DEBUG
599 KKASSERT(pv->pv_m == m);
600#endif
601 TAILQ_FOREACH(spv, &m->md.pv_list, pv_list) {
74b9d1ec
MD
602 if (pv == spv) {
603 crit_exit();
5926987a 604 return;
74b9d1ec 605 }
5926987a 606 }
74b9d1ec 607 crit_exit();
ed20d0e3 608 panic("test_m_maps_pv: failed m %p pv %p", m, pv);
5926987a
MD
609}
610
611static void
612ptbase_assert(struct pmap *pmap)
613{
614 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
615
616 /* are we current address space or kernel? */
4107b0c0 617 if (pmap == &kernel_pmap || frame == (((unsigned)PTDpde) & PG_FRAME))
5926987a 618 return;
3558dcda 619 KKASSERT(frame == (*mdcpu->gd_GDMAP1 & PG_FRAME));
5926987a
MD
620}
621
622#else
623
624#define test_m_maps_pv(m, pv)
625#define ptbase_assert(pmap)
626
627#endif
628
984263bc
MD
629#if defined(PMAP_DIAGNOSTIC)
630
631/*
632 * This code checks for non-writeable/modified pages.
633 * This should be an invalid condition.
634 */
635static int
636pmap_nw_modified(pt_entry_t ptea)
637{
638 int pte;
639
640 pte = (int) ptea;
641
642 if ((pte & (PG_M|PG_RW)) == PG_M)
643 return 1;
644 else
645 return 0;
646}
647#endif
648
649
650/*
4107b0c0
MD
651 * This routine defines the region(s) of memory that should not be tested
652 * for the modified bit.
653 *
654 * No requirements.
984263bc
MD
655 */
656static PMAP_INLINE int
657pmap_track_modified(vm_offset_t va)
658{
659 if ((va < clean_sva) || (va >= clean_eva))
660 return 1;
661 else
662 return 0;
663}
664
c1692ddf
MD
665/*
666 * Retrieve the mapped page table base for a particular pmap. Use our self
667 * mapping for the kernel_pmap or our current pmap.
668 *
669 * For foreign pmaps we use the per-cpu page table map. Since this involves
670 * installing a ptd it's actually (per-process x per-cpu). However, we
671 * still cannot depend on our mapping to survive thread switches because
672 * the process might be threaded and switching to another thread for the
673 * same process on the same cpu will allow that other thread to make its
674 * own mapping.
675 *
676 * This could be a bit confusing but the jist is for something like the
677 * vkernel which uses foreign pmaps all the time this represents a pretty
678 * good cache that avoids unnecessary invltlb()s.
4107b0c0
MD
679 *
680 * The caller must hold vm_token and the returned value is only valid
681 * until the caller blocks or releases the token.
c1692ddf 682 */
984263bc 683static unsigned *
e0e69b7d 684get_ptbase(pmap_t pmap)
984263bc
MD
685{
686 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
c1692ddf 687 struct mdglobaldata *gd = mdcpu;
984263bc 688
4107b0c0
MD
689 ASSERT_LWKT_TOKEN_HELD(&vm_token);
690
5926987a
MD
691 /*
692 * We can use PTmap if the pmap is our current address space or
693 * the kernel address space.
694 */
fbbaeba3 695 if (pmap == &kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
984263bc
MD
696 return (unsigned *) PTmap;
697 }
e0e69b7d 698
5926987a 699 /*
c1692ddf
MD
700 * Otherwise we use the per-cpu alternative page table map. Each
701 * cpu gets its own map. Because of this we cannot use this map
702 * from interrupts or threads which can preempt.
be3aecf7
MD
703 *
704 * Even if we already have the map cached we may still have to
705 * invalidate the TLB if another cpu modified a PDE in the map.
5926987a 706 */
c1692ddf
MD
707 KKASSERT(gd->mi.gd_intr_nesting_level == 0 &&
708 (gd->mi.gd_curthread->td_flags & TDF_INTTHREAD) == 0);
e0e69b7d 709
c1692ddf
MD
710 if ((*gd->gd_GDMAP1 & PG_FRAME) != frame) {
711 *gd->gd_GDMAP1 = frame | PG_RW | PG_V;
be3aecf7
MD
712 pmap->pm_cached |= gd->mi.gd_cpumask;
713 cpu_invltlb();
714 } else if ((pmap->pm_cached & gd->mi.gd_cpumask) == 0) {
715 pmap->pm_cached |= gd->mi.gd_cpumask;
984263bc 716 cpu_invltlb();
a93980ab
MD
717 } else if (dreadful_invltlb) {
718 cpu_invltlb();
984263bc 719 }
c1692ddf 720 return ((unsigned *)gd->gd_GDADDR1);
984263bc
MD
721}
722
984263bc 723/*
e0e69b7d
MD
724 * pmap_extract:
725 *
4107b0c0 726 * Extract the physical page address associated with the map/VA pair.
e0e69b7d 727 *
4107b0c0 728 * The caller may hold vm_token if it desires non-blocking operation.
984263bc 729 */
6ef943a3 730vm_paddr_t
840de426 731pmap_extract(pmap_t pmap, vm_offset_t va)
984263bc
MD
732{
733 vm_offset_t rtval;
734 vm_offset_t pdirindex;
840de426 735
4107b0c0 736 lwkt_gettoken(&vm_token);
984263bc
MD
737 pdirindex = va >> PDRSHIFT;
738 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
739 unsigned *pte;
740 if ((rtval & PG_PS) != 0) {
741 rtval &= ~(NBPDR - 1);
742 rtval |= va & (NBPDR - 1);
4107b0c0
MD
743 } else {
744 pte = get_ptbase(pmap) + i386_btop(va);
745 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
984263bc 746 }
4107b0c0
MD
747 } else {
748 rtval = 0;
984263bc 749 }
4107b0c0
MD
750 lwkt_reltoken(&vm_token);
751 return rtval;
f6bf3af1
MD
752}
753
984263bc
MD
754/***************************************************
755 * Low level mapping routines.....
756 ***************************************************/
757
758/*
4107b0c0
MD
759 * Map a wired VM page to a KVA, fully SMP synchronized.
760 *
761 * No requirements, non blocking.
984263bc 762 */
24712b90 763void
6ef943a3 764pmap_kenter(vm_offset_t va, vm_paddr_t pa)
984263bc 765{
840de426 766 unsigned *pte;
0f7a3396
MD
767 unsigned npte;
768 pmap_inval_info info;
984263bc 769
0f7a3396 770 pmap_inval_init(&info);
984263bc
MD
771 npte = pa | PG_RW | PG_V | pgeflag;
772 pte = (unsigned *)vtopte(va);
c2fb025d 773 pmap_inval_interlock(&info, &kernel_pmap, va);
984263bc 774 *pte = npte;
c2fb025d
MD
775 pmap_inval_deinterlock(&info, &kernel_pmap);
776 pmap_inval_done(&info);
984263bc
MD
777}
778
6d1ec6fa 779/*
4107b0c0
MD
780 * Map a wired VM page to a KVA, synchronized on current cpu only.
781 *
782 * No requirements, non blocking.
6d1ec6fa 783 */
24712b90
MD
784void
785pmap_kenter_quick(vm_offset_t va, vm_paddr_t pa)
786{
787 unsigned *pte;
788 unsigned npte;
789
790 npte = pa | PG_RW | PG_V | pgeflag;
791 pte = (unsigned *)vtopte(va);
792 *pte = npte;
793 cpu_invlpg((void *)va);
794}
795
4107b0c0
MD
796/*
797 * Synchronize a previously entered VA on all cpus.
798 *
799 * No requirements, non blocking.
800 */
24712b90
MD
801void
802pmap_kenter_sync(vm_offset_t va)
803{
804 pmap_inval_info info;
805
806 pmap_inval_init(&info);
c2fb025d
MD
807 pmap_inval_interlock(&info, &kernel_pmap, va);
808 pmap_inval_deinterlock(&info, &kernel_pmap);
809 pmap_inval_done(&info);
24712b90
MD
810}
811
4107b0c0
MD
812/*
813 * Synchronize a previously entered VA on the current cpu only.
814 *
815 * No requirements, non blocking.
816 */
24712b90
MD
817void
818pmap_kenter_sync_quick(vm_offset_t va)
819{
820 cpu_invlpg((void *)va);
821}
822
984263bc 823/*
4107b0c0
MD
824 * Remove a page from the kernel pagetables, fully SMP synchronized.
825 *
826 * No requirements, non blocking.
984263bc 827 */
24712b90 828void
840de426 829pmap_kremove(vm_offset_t va)
984263bc 830{
840de426 831 unsigned *pte;
0f7a3396 832 pmap_inval_info info;
984263bc 833
0f7a3396 834 pmap_inval_init(&info);
984263bc 835 pte = (unsigned *)vtopte(va);
c2fb025d 836 pmap_inval_interlock(&info, &kernel_pmap, va);
984263bc 837 *pte = 0;
c2fb025d
MD
838 pmap_inval_deinterlock(&info, &kernel_pmap);
839 pmap_inval_done(&info);
984263bc
MD
840}
841
4107b0c0
MD
842/*
843 * Remove a page from the kernel pagetables, synchronized on current cpu only.
844 *
845 * No requirements, non blocking.
846 */
24712b90
MD
847void
848pmap_kremove_quick(vm_offset_t va)
849{
850 unsigned *pte;
851 pte = (unsigned *)vtopte(va);
852 *pte = 0;
853 cpu_invlpg((void *)va);
854}
855
9ad680a3 856/*
4107b0c0
MD
857 * Adjust the permissions of a page in the kernel page table,
858 * synchronized on the current cpu only.
859 *
860 * No requirements, non blocking.
9ad680a3
MD
861 */
862void
863pmap_kmodify_rw(vm_offset_t va)
864{
4107b0c0 865 atomic_set_int(vtopte(va), PG_RW);
9ad680a3
MD
866 cpu_invlpg((void *)va);
867}
868
4107b0c0
MD
869/*
870 * Adjust the permissions of a page in the kernel page table,
871 * synchronized on the current cpu only.
872 *
873 * No requirements, non blocking.
874 */
9ad680a3
MD
875void
876pmap_kmodify_nc(vm_offset_t va)
877{
4107b0c0 878 atomic_set_int(vtopte(va), PG_N);
9ad680a3
MD
879 cpu_invlpg((void *)va);
880}
881
984263bc 882/*
4107b0c0 883 * Map a range of physical addresses into kernel virtual address space.
984263bc 884 *
4107b0c0 885 * No requirements, non blocking.
984263bc
MD
886 */
887vm_offset_t
8e5e6f1b 888pmap_map(vm_offset_t *virtp, vm_paddr_t start, vm_paddr_t end, int prot)
984263bc 889{
8e5e6f1b
AH
890 vm_offset_t sva, virt;
891
892 sva = virt = *virtp;
984263bc
MD
893 while (start < end) {
894 pmap_kenter(virt, start);
895 virt += PAGE_SIZE;
896 start += PAGE_SIZE;
897 }
8e5e6f1b
AH
898 *virtp = virt;
899 return (sva);
984263bc
MD
900}
901
984263bc 902/*
4107b0c0
MD
903 * Add a list of wired pages to the kva, fully SMP synchronized.
904 *
905 * No requirements, non blocking.
984263bc
MD
906 */
907void
840de426 908pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
984263bc
MD
909{
910 vm_offset_t end_va;
911
912 end_va = va + count * PAGE_SIZE;
913
914 while (va < end_va) {
915 unsigned *pte;
916
917 pte = (unsigned *)vtopte(va);
918 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
984263bc 919 cpu_invlpg((void *)va);
984263bc
MD
920 va += PAGE_SIZE;
921 m++;
922 }
0f7a3396 923 smp_invltlb(); /* XXX */
984263bc
MD
924}
925
926/*
4107b0c0 927 * Remove pages from KVA, fully SMP synchronized.
7155fc7d 928 *
4107b0c0 929 * No requirements, non blocking.
984263bc
MD
930 */
931void
840de426 932pmap_qremove(vm_offset_t va, int count)
984263bc
MD
933{
934 vm_offset_t end_va;
935
936 end_va = va + count*PAGE_SIZE;
937
938 while (va < end_va) {
939 unsigned *pte;
940
941 pte = (unsigned *)vtopte(va);
942 *pte = 0;
984263bc 943 cpu_invlpg((void *)va);
984263bc
MD
944 va += PAGE_SIZE;
945 }
984263bc 946 smp_invltlb();
984263bc
MD
947}
948
06ecca5a
MD
949/*
950 * This routine works like vm_page_lookup() but also blocks as long as the
951 * page is busy. This routine does not busy the page it returns.
952 *
b12defdc 953 * The caller must hold the object.
06ecca5a 954 */
984263bc 955static vm_page_t
840de426 956pmap_page_lookup(vm_object_t object, vm_pindex_t pindex)
984263bc
MD
957{
958 vm_page_t m;
06ecca5a 959
b12defdc
MD
960 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
961 m = vm_page_lookup_busy_wait(object, pindex, FALSE, "pplookp");
17cde63e 962
06ecca5a 963 return(m);
984263bc
MD
964}
965
263e4574
MD
966/*
967 * Create a new thread and optionally associate it with a (new) process.
6ef943a3 968 * NOTE! the new thread's cpu may not equal the current cpu.
263e4574 969 */
7d0bac62
MD
970void
971pmap_init_thread(thread_t td)
263e4574 972{
f470d0c8 973 /* enforce pcb placement */
f470d0c8 974 td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_size) - 1;
65d6ce10 975 td->td_savefpu = &td->td_pcb->pcb_save;
7d0bac62 976 td->td_sp = (char *)td->td_pcb - 16;
263e4574
MD
977}
978
984263bc 979/*
984263bc
MD
980 * This routine directly affects the fork perf for a process.
981 */
982void
13d13d89 983pmap_init_proc(struct proc *p)
984263bc 984{
984263bc
MD
985}
986
984263bc
MD
987/***************************************************
988 * Page table page management routines.....
989 ***************************************************/
990
991/*
90244566
MD
992 * This routine unwires page table pages, removing and freeing the page
993 * tale page when the wire count drops to 0.
4107b0c0
MD
994 *
995 * The caller must hold vm_token.
996 * This function can block.
984263bc
MD
997 */
998static int
90244566 999_pmap_unwire_pte(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
840de426 1000{
17cde63e
MD
1001 /*
1002 * Wait until we can busy the page ourselves. We cannot have
1003 * any active flushes if we block.
1004 */
b12defdc 1005 vm_page_busy_wait(m, FALSE, "pmuwpt");
eec2b734 1006 KASSERT(m->queue == PQ_NONE,
90244566 1007 ("_pmap_unwire_pte: %p->queue != PQ_NONE", m));
984263bc 1008
90244566 1009 if (m->wire_count == 1) {
984263bc 1010 /*
be3aecf7
MD
1011 * Unmap the page table page.
1012 *
1013 * NOTE: We must clear pm_cached for all cpus, including
1014 * the current one, when clearing a page directory
1015 * entry.
984263bc 1016 */
c2fb025d 1017 pmap_inval_interlock(info, pmap, -1);
2247fe02 1018 KKASSERT(pmap->pm_pdir[m->pindex]);
984263bc 1019 pmap->pm_pdir[m->pindex] = 0;
be3aecf7 1020 pmap->pm_cached = 0;
c2fb025d 1021 pmap_inval_deinterlock(info, pmap);
eec2b734
MD
1022
1023 KKASSERT(pmap->pm_stats.resident_count > 0);
984263bc 1024 --pmap->pm_stats.resident_count;
984263bc
MD
1025
1026 if (pmap->pm_ptphint == m)
1027 pmap->pm_ptphint = NULL;
1028
1029 /*
eec2b734
MD
1030 * This was our last hold, the page had better be unwired
1031 * after we decrement wire_count.
1032 *
1033 * FUTURE NOTE: shared page directory page could result in
1034 * multiple wire counts.
984263bc 1035 */
90244566 1036 vm_page_unwire(m, 0);
17cde63e 1037 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
eec2b734
MD
1038 vm_page_flash(m);
1039 vm_page_free_zero(m);
984263bc 1040 return 1;
17cde63e 1041 } else {
90244566
MD
1042 KKASSERT(m->wire_count > 1);
1043 if (vm_page_unwire_quick(m))
1044 panic("pmap_unwire_pte: Insufficient wire_count");
b12defdc 1045 vm_page_wakeup(m);
17cde63e 1046 return 0;
984263bc 1047 }
984263bc
MD
1048}
1049
4107b0c0
MD
1050/*
1051 * The caller must hold vm_token.
92ba8d28 1052 *
4107b0c0 1053 * This function can block.
92ba8d28
MD
1054 *
1055 * This function can race the wire_count 2->1 case because the page
1056 * is not busied during the unwire_quick operation. An eventual
1057 * pmap_release() will catch the case.
4107b0c0 1058 */
984263bc 1059static PMAP_INLINE int
90244566 1060pmap_unwire_pte(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
984263bc 1061{
90244566
MD
1062 KKASSERT(m->wire_count > 0);
1063 if (m->wire_count > 1) {
1064 if (vm_page_unwire_quick(m))
1065 panic("pmap_unwire_pte: Insufficient wire_count");
984263bc 1066 return 0;
eec2b734 1067 } else {
90244566 1068 return _pmap_unwire_pte(pmap, m, info);
eec2b734 1069 }
984263bc
MD
1070}
1071
1072/*
4107b0c0 1073 * After removing a (user) page table entry, this routine is used to
984263bc 1074 * conditionally free the page, and manage the hold/wire counts.
5926987a 1075 *
4107b0c0
MD
1076 * The caller must hold vm_token.
1077 * This function can block regardless.
984263bc 1078 */
554cf9ac 1079static void
0f7a3396 1080pmap_unuse_pt(pmap_t pmap, vm_offset_t va, vm_page_t mpte,
4107b0c0 1081 pmap_inval_info_t info)
984263bc
MD
1082{
1083 unsigned ptepindex;
4107b0c0 1084
b12defdc
MD
1085 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1086
984263bc 1087 if (va >= UPT_MIN_ADDRESS)
554cf9ac 1088 return;
984263bc
MD
1089
1090 if (mpte == NULL) {
1091 ptepindex = (va >> PDRSHIFT);
b1482674
MD
1092 if ((mpte = pmap->pm_ptphint) != NULL &&
1093 mpte->pindex == ptepindex &&
1094 (mpte->flags & PG_BUSY) == 0) {
1095 ; /* use mpte */
984263bc 1096 } else {
b12defdc 1097 mpte = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
984263bc 1098 pmap->pm_ptphint = mpte;
b12defdc 1099 vm_page_wakeup(mpte);
984263bc
MD
1100 }
1101 }
554cf9ac 1102 pmap_unwire_pte(pmap, mpte, info);
984263bc
MD
1103}
1104
54a764e8 1105/*
fbbaeba3
MD
1106 * Initialize pmap0/vmspace0. This pmap is not added to pmap_list because
1107 * it, and IdlePTD, represents the template used to update all other pmaps.
1108 *
1109 * On architectures where the kernel pmap is not integrated into the user
1110 * process pmap, this pmap represents the process pmap, not the kernel pmap.
1111 * kernel_pmap should be used to directly access the kernel_pmap.
4107b0c0
MD
1112 *
1113 * No requirements.
54a764e8 1114 */
984263bc 1115void
840de426 1116pmap_pinit0(struct pmap *pmap)
984263bc
MD
1117{
1118 pmap->pm_pdir =
e4846942 1119 (pd_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
24712b90 1120 pmap_kenter((vm_offset_t)pmap->pm_pdir, (vm_offset_t) IdlePTD);
984263bc
MD
1121 pmap->pm_count = 1;
1122 pmap->pm_active = 0;
be3aecf7 1123 pmap->pm_cached = 0;
984263bc
MD
1124 pmap->pm_ptphint = NULL;
1125 TAILQ_INIT(&pmap->pm_pvlist);
b12defdc
MD
1126 TAILQ_INIT(&pmap->pm_pvlist_free);
1127 spin_init(&pmap->pm_spin);
1128 lwkt_token_init(&pmap->pm_token, "pmap_tok");
984263bc
MD
1129 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1130}
1131
1132/*
1133 * Initialize a preallocated and zeroed pmap structure,
1134 * such as one in a vmspace structure.
4107b0c0
MD
1135 *
1136 * No requirements.
984263bc
MD
1137 */
1138void
840de426 1139pmap_pinit(struct pmap *pmap)
984263bc
MD
1140{
1141 vm_page_t ptdpg;
1142
1143 /*
1144 * No need to allocate page table space yet but we do need a valid
1145 * page directory table.
1146 */
b5b32410 1147 if (pmap->pm_pdir == NULL) {
984263bc 1148 pmap->pm_pdir =
e4846942 1149 (pd_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
b5b32410 1150 }
984263bc
MD
1151
1152 /*
c3834cb2 1153 * Allocate an object for the ptes
984263bc
MD
1154 */
1155 if (pmap->pm_pteobj == NULL)
c3834cb2 1156 pmap->pm_pteobj = vm_object_allocate(OBJT_DEFAULT, PTDPTDI + 1);
984263bc
MD
1157
1158 /*
c3834cb2
MD
1159 * Allocate the page directory page, unless we already have
1160 * one cached. If we used the cached page the wire_count will
1161 * already be set appropriately.
984263bc 1162 */
c3834cb2
MD
1163 if ((ptdpg = pmap->pm_pdirm) == NULL) {
1164 ptdpg = vm_page_grab(pmap->pm_pteobj, PTDPTDI,
d2d8515b
MD
1165 VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
1166 VM_ALLOC_ZERO);
c3834cb2 1167 pmap->pm_pdirm = ptdpg;
b12defdc
MD
1168 vm_page_flag_clear(ptdpg, PG_MAPPED);
1169 vm_page_wire(ptdpg);
d2d8515b 1170 KKASSERT(ptdpg->valid == VM_PAGE_BITS_ALL);
c3834cb2 1171 pmap_kenter((vm_offset_t)pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
b12defdc 1172 vm_page_wakeup(ptdpg);
c3834cb2 1173 }
984263bc 1174 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
984263bc
MD
1175
1176 /* install self-referential address mapping entry */
1177 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1178 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1179
1180 pmap->pm_count = 1;
1181 pmap->pm_active = 0;
be3aecf7 1182 pmap->pm_cached = 0;
984263bc
MD
1183 pmap->pm_ptphint = NULL;
1184 TAILQ_INIT(&pmap->pm_pvlist);
b12defdc
MD
1185 TAILQ_INIT(&pmap->pm_pvlist_free);
1186 spin_init(&pmap->pm_spin);
1187 lwkt_token_init(&pmap->pm_token, "pmap_tok");
984263bc 1188 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
eec2b734 1189 pmap->pm_stats.resident_count = 1;
984263bc
MD
1190}
1191
e3161323 1192/*
c3834cb2
MD
1193 * Clean up a pmap structure so it can be physically freed. This routine
1194 * is called by the vmspace dtor function. A great deal of pmap data is
1195 * left passively mapped to improve vmspace management so we have a bit
1196 * of cleanup work to do here.
4107b0c0
MD
1197 *
1198 * No requirements.
e3161323
MD
1199 */
1200void
1201pmap_puninit(pmap_t pmap)
1202{
c3834cb2
MD
1203 vm_page_t p;
1204
3321ee05 1205 pmap_wait(pmap, -1);
e3161323 1206 KKASSERT(pmap->pm_active == 0);
c3834cb2
MD
1207 if ((p = pmap->pm_pdirm) != NULL) {
1208 KKASSERT(pmap->pm_pdir != NULL);
1209 pmap_kremove((vm_offset_t)pmap->pm_pdir);
b12defdc 1210 vm_page_busy_wait(p, FALSE, "pgpun");
90244566 1211 vm_page_unwire(p, 0);
c3834cb2
MD
1212 vm_page_free_zero(p);
1213 pmap->pm_pdirm = NULL;
1214 }
e3161323
MD
1215 if (pmap->pm_pdir) {
1216 kmem_free(&kernel_map, (vm_offset_t)pmap->pm_pdir, PAGE_SIZE);
1217 pmap->pm_pdir = NULL;
1218 }
1219 if (pmap->pm_pteobj) {
1220 vm_object_deallocate(pmap->pm_pteobj);
1221 pmap->pm_pteobj = NULL;
1222 }
1223}
1224
984263bc
MD
1225/*
1226 * Wire in kernel global address entries. To avoid a race condition
1227 * between pmap initialization and pmap_growkernel, this procedure
54a764e8
MD
1228 * adds the pmap to the master list (which growkernel scans to update),
1229 * then copies the template.
4107b0c0
MD
1230 *
1231 * No requirements.
984263bc
MD
1232 */
1233void
840de426 1234pmap_pinit2(struct pmap *pmap)
984263bc 1235{
b12defdc
MD
1236 /*
1237 * XXX copies current process, does not fill in MPPTDI
1238 */
1239 spin_lock(&pmap_spin);
54a764e8 1240 TAILQ_INSERT_TAIL(&pmap_list, pmap, pm_pmnode);
984263bc 1241 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
b12defdc 1242 spin_unlock(&pmap_spin);
984263bc
MD
1243}
1244
344ad853 1245/*
eec2b734 1246 * Attempt to release and free a vm_page in a pmap. Returns 1 on success,
344ad853 1247 * 0 on failure (if the procedure had to sleep).
c3834cb2
MD
1248 *
1249 * When asked to remove the page directory page itself, we actually just
1250 * leave it cached so we do not have to incur the SMP inval overhead of
1251 * removing the kernel mapping. pmap_puninit() will take care of it.
4107b0c0
MD
1252 *
1253 * The caller must hold vm_token.
1254 * This function can block regardless.
344ad853 1255 */
984263bc 1256static int
840de426 1257pmap_release_free_page(struct pmap *pmap, vm_page_t p)
984263bc
MD
1258{
1259 unsigned *pde = (unsigned *) pmap->pm_pdir;
4107b0c0 1260
984263bc
MD
1261 /*
1262 * This code optimizes the case of freeing non-busy
1263 * page-table pages. Those pages are zero now, and
1264 * might as well be placed directly into the zero queue.
1265 */
b12defdc
MD
1266 if (vm_page_busy_try(p, FALSE)) {
1267 vm_page_sleep_busy(p, FALSE, "pmaprl");
984263bc 1268 return 0;
b12defdc 1269 }
984263bc 1270
eec2b734 1271 KKASSERT(pmap->pm_stats.resident_count > 0);
2247fe02 1272 KKASSERT(pde[p->pindex]);
984263bc 1273
b1482674
MD
1274 /*
1275 * page table page's wire_count must be 1. Caller is the pmap
1276 * termination code which holds the pm_pteobj, there is a race
1277 * if someone else is trying to hold the VM object in order to
1278 * clean up a wire_count.
1279 */
90244566 1280 if (p->wire_count != 1) {
b1482674
MD
1281 if (pmap->pm_pteobj->hold_count <= 1)
1282 panic("pmap_release: freeing wired page table page");
1283 kprintf("pmap_release_free_page: unwire race detected\n");
1284 vm_page_wakeup(p);
1285 tsleep(p, 0, "pmapx", 1);
1286 return 0;
984263bc 1287 }
b1482674
MD
1288
1289 /*
1290 * Remove the page table page from the processes address space.
1291 */
1292 pmap->pm_cached = 0;
1293 pde[p->pindex] = 0;
1294 --pmap->pm_stats.resident_count;
c3834cb2
MD
1295 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1296 pmap->pm_ptphint = NULL;
1297
984263bc 1298 /*
c3834cb2
MD
1299 * We leave the page directory page cached, wired, and mapped in
1300 * the pmap until the dtor function (pmap_puninit()) gets called.
1301 * However, still clean it up so we can set PG_ZERO.
c1692ddf
MD
1302 *
1303 * The pmap has already been removed from the pmap_list in the
1304 * PTDPTDI case.
984263bc
MD
1305 */
1306 if (p->pindex == PTDPTDI) {
1307 bzero(pde + KPTDI, nkpt * PTESIZE);
9388fcaa 1308 bzero(pde + MPPTDI, (NPDEPG - MPPTDI) * PTESIZE);
c3834cb2
MD
1309 vm_page_flag_set(p, PG_ZERO);
1310 vm_page_wakeup(p);
1311 } else {
92ba8d28
MD
1312 /*
1313 * This case can occur if a pmap_unwire_pte() loses a race
1314 * while the page is unbusied.
1315 */
1316 /*panic("pmap_release: page should already be gone %p", p);*/
1317 vm_page_flag_clear(p, PG_MAPPED);
90244566 1318 vm_page_unwire(p, 0);
c3834cb2 1319 vm_page_free_zero(p);
984263bc 1320 }
984263bc
MD
1321 return 1;
1322}
1323
1324/*
4107b0c0
MD
1325 * This routine is called if the page table page is not mapped correctly.
1326 *
1327 * The caller must hold vm_token.
984263bc
MD
1328 */
1329static vm_page_t
840de426 1330_pmap_allocpte(pmap_t pmap, unsigned ptepindex)
984263bc 1331{
480c83b6 1332 vm_offset_t ptepa;
984263bc
MD
1333 vm_page_t m;
1334
1335 /*
d2d8515b
MD
1336 * Find or fabricate a new pagetable page. Setting VM_ALLOC_ZERO
1337 * will zero any new page and mark it valid.
984263bc
MD
1338 */
1339 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
d2d8515b 1340 VM_ALLOC_NORMAL | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
984263bc
MD
1341
1342 KASSERT(m->queue == PQ_NONE,
1343 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1344
eec2b734 1345 /*
90244566 1346 * Increment the wire count for the page we will be returning to
eec2b734
MD
1347 * the caller.
1348 */
90244566 1349 vm_page_wire(m);
eec2b734
MD
1350
1351 /*
1352 * It is possible that someone else got in and mapped by the page
1353 * directory page while we were blocked, if so just unbusy and
90244566 1354 * return the wired page.
eec2b734
MD
1355 */
1356 if ((ptepa = pmap->pm_pdir[ptepindex]) != 0) {
1357 KKASSERT((ptepa & PG_FRAME) == VM_PAGE_TO_PHYS(m));
1358 vm_page_wakeup(m);
1359 return(m);
1360 }
1361
984263bc
MD
1362 /*
1363 * Map the pagetable page into the process address space, if
1364 * it isn't already there.
be3aecf7
MD
1365 *
1366 * NOTE: For safety clear pm_cached for all cpus including the
1367 * current one when adding a PDE to the map.
984263bc 1368 */
eec2b734 1369 ++pmap->pm_stats.resident_count;
984263bc
MD
1370
1371 ptepa = VM_PAGE_TO_PHYS(m);
1372 pmap->pm_pdir[ptepindex] =
1373 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
be3aecf7 1374 pmap->pm_cached = 0;
984263bc
MD
1375
1376 /*
1377 * Set the page table hint
1378 */
1379 pmap->pm_ptphint = m;
984263bc
MD
1380 vm_page_flag_set(m, PG_MAPPED);
1381 vm_page_wakeup(m);
1382
1383 return m;
1384}
1385
4107b0c0
MD
1386/*
1387 * Allocate a page table entry for a va.
1388 *
1389 * The caller must hold vm_token.
1390 */
984263bc 1391static vm_page_t
840de426 1392pmap_allocpte(pmap_t pmap, vm_offset_t va)
984263bc
MD
1393{
1394 unsigned ptepindex;
1395 vm_offset_t ptepa;
b1482674 1396 vm_page_t mpte;
984263bc 1397
b12defdc
MD
1398 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1399
984263bc
MD
1400 /*
1401 * Calculate pagetable page index
1402 */
1403 ptepindex = va >> PDRSHIFT;
1404
1405 /*
1406 * Get the page directory entry
1407 */
1408 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1409
1410 /*
1411 * This supports switching from a 4MB page to a
1412 * normal 4K page.
1413 */
1414 if (ptepa & PG_PS) {
1415 pmap->pm_pdir[ptepindex] = 0;
1416 ptepa = 0;
0f7a3396 1417 smp_invltlb();
54341a3b 1418 cpu_invltlb();
984263bc
MD
1419 }
1420
1421 /*
1422 * If the page table page is mapped, we just increment the
90244566 1423 * wire count, and activate it.
984263bc
MD
1424 */
1425 if (ptepa) {
1426 /*
1427 * In order to get the page table page, try the
1428 * hint first.
1429 */
b1482674
MD
1430 if ((mpte = pmap->pm_ptphint) != NULL &&
1431 (mpte->pindex == ptepindex) &&
1432 (mpte->flags & PG_BUSY) == 0) {
1433 vm_page_wire_quick(mpte);
984263bc 1434 } else {
b1482674
MD
1435 mpte = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
1436 pmap->pm_ptphint = mpte;
1437 vm_page_wire_quick(mpte);
1438 vm_page_wakeup(mpte);
984263bc 1439 }
b1482674 1440 return mpte;
984263bc
MD
1441 }
1442 /*
1443 * Here if the pte page isn't mapped, or if it has been deallocated.
1444 */
1445 return _pmap_allocpte(pmap, ptepindex);
1446}
1447
1448
1449/***************************************************
1f804340 1450 * Pmap allocation/deallocation routines.
984263bc
MD
1451 ***************************************************/
1452
1453/*
1454 * Release any resources held by the given physical map.
1455 * Called when a pmap initialized by pmap_pinit is being released.
1456 * Should only be called if the map contains no valid mappings.
4107b0c0 1457 *
b12defdc 1458 * Caller must hold pmap->pm_token
984263bc 1459 */
1f804340
MD
1460static int pmap_release_callback(struct vm_page *p, void *data);
1461
984263bc 1462void
840de426 1463pmap_release(struct pmap *pmap)
984263bc 1464{
984263bc 1465 vm_object_t object = pmap->pm_pteobj;
1f804340 1466 struct rb_vm_page_scan_info info;
984263bc 1467
4107b0c0
MD
1468 KASSERT(pmap->pm_active == 0,
1469 ("pmap still active! %08x", pmap->pm_active));
984263bc
MD
1470#if defined(DIAGNOSTIC)
1471 if (object->ref_count != 1)
1472 panic("pmap_release: pteobj reference count != 1");
1473#endif
1474
1f804340
MD
1475 info.pmap = pmap;
1476 info.object = object;
b12defdc
MD
1477
1478 spin_lock(&pmap_spin);
54a764e8 1479 TAILQ_REMOVE(&pmap_list, pmap, pm_pmnode);
b12defdc 1480 spin_unlock(&pmap_spin);
1f804340 1481
b12defdc 1482 vm_object_hold(object);
2160c64a 1483 /*lwkt_gettoken(&vm_token);*/
1f804340 1484 do {
1f804340
MD
1485 info.error = 0;
1486 info.mpte = NULL;
1487 info.limit = object->generation;
1488
1489 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
1490 pmap_release_callback, &info);
1491 if (info.error == 0 && info.mpte) {
1492 if (!pmap_release_free_page(pmap, info.mpte))
1493 info.error = 1;
984263bc 1494 }
1f804340 1495 } while (info.error);
2160c64a 1496 /*lwkt_reltoken(&vm_token);*/
2f2d9e58 1497 vm_object_drop(object);
b12defdc
MD
1498
1499 pmap->pm_cached = 0;
1f804340
MD
1500}
1501
4107b0c0
MD
1502/*
1503 * The caller must hold vm_token.
1504 */
1f804340
MD
1505static int
1506pmap_release_callback(struct vm_page *p, void *data)
1507{
1508 struct rb_vm_page_scan_info *info = data;
1509
1510 if (p->pindex == PTDPTDI) {
1511 info->mpte = p;
1512 return(0);
344ad853 1513 }
1f804340
MD
1514 if (!pmap_release_free_page(info->pmap, p)) {
1515 info->error = 1;
1516 return(-1);
1517 }
1518 if (info->object->generation != info->limit) {
1519 info->error = 1;
1520 return(-1);
1521 }
1522 return(0);
984263bc 1523}
984263bc
MD
1524
1525/*
0e5797fe 1526 * Grow the number of kernel page table entries, if needed.
4107b0c0
MD
1527 *
1528 * No requirements.
984263bc
MD
1529 */
1530void
a8cf2878 1531pmap_growkernel(vm_offset_t kstart, vm_offset_t kend)
984263bc 1532{
a8cf2878 1533 vm_offset_t addr = kend;
54a764e8 1534 struct pmap *pmap;
984263bc
MD
1535 vm_offset_t ptppaddr;
1536 vm_page_t nkpg;
1537 pd_entry_t newpdir;
1538
b12defdc 1539 vm_object_hold(kptobj);
984263bc
MD
1540 if (kernel_vm_end == 0) {
1541 kernel_vm_end = KERNBASE;
1542 nkpt = 0;
1543 while (pdir_pde(PTD, kernel_vm_end)) {
4107b0c0
MD
1544 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1545 ~(PAGE_SIZE * NPTEPG - 1);
984263bc
MD
1546 nkpt++;
1547 }
1548 }
1549 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1550 while (kernel_vm_end < addr) {
1551 if (pdir_pde(PTD, kernel_vm_end)) {
4107b0c0
MD
1552 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1553 ~(PAGE_SIZE * NPTEPG - 1);
984263bc
MD
1554 continue;
1555 }
1556
1557 /*
1558 * This index is bogus, but out of the way
1559 */
4107b0c0
MD
1560 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_NORMAL |
1561 VM_ALLOC_SYSTEM |
1562 VM_ALLOC_INTERRUPT);
dc1fd4b3 1563 if (nkpg == NULL)
984263bc
MD
1564 panic("pmap_growkernel: no memory to grow kernel");
1565
984263bc
MD
1566 vm_page_wire(nkpg);
1567 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1568 pmap_zero_page(ptppaddr);
1569 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1570 pdir_pde(PTD, kernel_vm_end) = newpdir;
fbbaeba3 1571 *pmap_pde(&kernel_pmap, kernel_vm_end) = newpdir;
0e5797fe
MD
1572 nkpt++;
1573
1574 /*
54a764e8 1575 * This update must be interlocked with pmap_pinit2.
0e5797fe 1576 */
b12defdc 1577 spin_lock(&pmap_spin);
54a764e8
MD
1578 TAILQ_FOREACH(pmap, &pmap_list, pm_pmnode) {
1579 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1580 }
b12defdc 1581 spin_unlock(&pmap_spin);
54a764e8
MD
1582 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1583 ~(PAGE_SIZE * NPTEPG - 1);
984263bc 1584 }
b12defdc 1585 vm_object_drop(kptobj);
984263bc
MD
1586}
1587
984263bc 1588/*
4107b0c0
MD
1589 * Add a reference to the specified pmap.
1590 *
1591 * No requirements.
984263bc
MD
1592 */
1593void
840de426 1594pmap_reference(pmap_t pmap)
984263bc 1595{
4107b0c0
MD
1596 if (pmap) {
1597 lwkt_gettoken(&vm_token);
1598 ++pmap->pm_count;
1599 lwkt_reltoken(&vm_token);
984263bc
MD
1600 }
1601}
1602
3321ee05
MD
1603/*
1604 * vm_token must be held
1605 */
1606static
1607void
1608pmap_hold(pmap_t pmap)
1609{
1610 ++pmap->pm_count;
1611}
1612
1613/*
1614 * vm_token must be held
1615 */
1616static
1617void
1618pmap_drop(pmap_t pmap)
1619{
1620 --pmap->pm_count;
22c4e116 1621 if (pmap->pm_count == (int)0x80000000)
3321ee05
MD
1622 wakeup(pmap);
1623}
1624
1625static
1626void
1627pmap_wait(pmap_t pmap, int count)
1628{
1629 lwkt_gettoken(&vm_token);
1630 pmap->pm_count += count;
1631 if (pmap->pm_count & 0x7FFFFFFF) {
1632 while (pmap->pm_count & 0x7FFFFFFF) {
1633 pmap->pm_count |= 0x80000000;
1634 tsleep(pmap, 0, "pmapd", 0);
1635 pmap->pm_count &= ~0x80000000;
1636 kprintf("pmap_wait: race averted\n");
1637 }
1638 }
1639 lwkt_reltoken(&vm_token);
1640}
1641
984263bc 1642/***************************************************
4107b0c0 1643 * page management routines.
984263bc
MD
1644 ***************************************************/
1645
1646/*
8a8d5d85
MD
1647 * free the pv_entry back to the free list. This function may be
1648 * called from an interrupt.
4107b0c0
MD
1649 *
1650 * The caller must hold vm_token.
984263bc
MD
1651 */
1652static PMAP_INLINE void
840de426 1653free_pv_entry(pv_entry_t pv)
984263bc 1654{
2bb9cc6f
MD
1655 struct mdglobaldata *gd;
1656
5926987a
MD
1657#ifdef PMAP_DEBUG
1658 KKASSERT(pv->pv_m != NULL);
1659 pv->pv_m = NULL;
1660#endif
2bb9cc6f 1661 gd = mdcpu;
984263bc 1662 pv_entry_count--;
2bb9cc6f
MD
1663 if (gd->gd_freepv == NULL)
1664 gd->gd_freepv = pv;
1665 else
1666 zfree(pvzone, pv);
984263bc
MD
1667}
1668
1669/*
1670 * get a new pv_entry, allocating a block from the system
2bb9cc6f
MD
1671 * when needed. This function may be called from an interrupt thread.
1672 *
1673 * THIS FUNCTION CAN BLOCK ON THE ZALLOC TOKEN, serialization of other
1674 * tokens (aka vm_token) to be temporarily lost.
4107b0c0
MD
1675 *
1676 * The caller must hold vm_token.
984263bc
MD
1677 */
1678static pv_entry_t
1679get_pv_entry(void)
1680{
2bb9cc6f
MD
1681 struct mdglobaldata *gd;
1682 pv_entry_t pv;
1683
984263bc
MD
1684 pv_entry_count++;
1685 if (pv_entry_high_water &&
20479584
MD
1686 (pv_entry_count > pv_entry_high_water) &&
1687 (pmap_pagedaemon_waken == 0)) {
984263bc
MD
1688 pmap_pagedaemon_waken = 1;
1689 wakeup (&vm_pages_needed);
1690 }
2bb9cc6f
MD
1691 gd = mdcpu;
1692 if ((pv = gd->gd_freepv) != NULL)
1693 gd->gd_freepv = NULL;
1694 else
1695 pv = zalloc(pvzone);
1696 return pv;
984263bc
MD
1697}
1698
1699/*
1700 * This routine is very drastic, but can save the system
1701 * in a pinch.
4107b0c0
MD
1702 *
1703 * No requirements.
984263bc
MD
1704 */
1705void
840de426 1706pmap_collect(void)
984263bc
MD
1707{
1708 int i;
1709 vm_page_t m;
1710 static int warningdone=0;
1711
1712 if (pmap_pagedaemon_waken == 0)
1713 return;
4107b0c0 1714 lwkt_gettoken(&vm_token);
20479584 1715 pmap_pagedaemon_waken = 0;
984263bc
MD
1716
1717 if (warningdone < 5) {
948209ce
MD
1718 kprintf("pmap_collect: collecting pv entries -- "
1719 "suggest increasing PMAP_SHPGPERPROC\n");
984263bc
MD
1720 warningdone++;
1721 }
1722
b12defdc 1723 for (i = 0; i < vm_page_array_size; i++) {
984263bc 1724 m = &vm_page_array[i];
b12defdc 1725 if (m->wire_count || m->hold_count)
984263bc 1726 continue;
b12defdc
MD
1727 if (vm_page_busy_try(m, TRUE) == 0) {
1728 if (m->wire_count == 0 && m->hold_count == 0) {
1729 pmap_remove_all(m);
1730 }
1731 vm_page_wakeup(m);
4107b0c0 1732 }
984263bc 1733 }
4107b0c0 1734 lwkt_reltoken(&vm_token);
984263bc
MD
1735}
1736
1737
1738/*
b1482674
MD
1739 * Remove the pv entry and unwire the page table page related to the
1740 * pte the caller has cleared from the page table.
4107b0c0
MD
1741 *
1742 * The caller must hold vm_token.
984263bc 1743 */
554cf9ac 1744static void
0f7a3396 1745pmap_remove_entry(struct pmap *pmap, vm_page_t m,
4107b0c0 1746 vm_offset_t va, pmap_inval_info_t info)
984263bc
MD
1747{
1748 pv_entry_t pv;
984263bc 1749
b1482674
MD
1750 /*
1751 * Cannot block
1752 */
4107b0c0 1753 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc
MD
1754 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1755 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1756 if (pmap == pv->pv_pmap && va == pv->pv_va)
1757 break;
1758 }
1759 } else {
1760 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
5926987a
MD
1761#ifdef PMAP_DEBUG
1762 KKASSERT(pv->pv_pmap == pmap);
1763#endif
1764 if (va == pv->pv_va)
984263bc
MD
1765 break;
1766 }
1767 }
5926987a 1768 KKASSERT(pv);
984263bc 1769
b1482674
MD
1770 /*
1771 * Cannot block
1772 */
5926987a
MD
1773 test_m_maps_pv(m, pv);
1774 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1775 m->md.pv_list_count--;
cef01e15
MD
1776 if (m->object)
1777 atomic_add_int(&m->object->agg_pv_list_count, -1);
5926987a
MD
1778 if (TAILQ_EMPTY(&m->md.pv_list))
1779 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1780 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1781 ++pmap->pm_generation;
b1482674
MD
1782
1783 /*
1784 * This can block.
1785 */
b12defdc 1786 vm_object_hold(pmap->pm_pteobj);
554cf9ac 1787 pmap_unuse_pt(pmap, va, pv->pv_ptem, info);
b12defdc 1788 vm_object_drop(pmap->pm_pteobj);
5926987a 1789 free_pv_entry(pv);
984263bc
MD
1790}
1791
1792/*
4107b0c0
MD
1793 * Create a pv entry for page at pa for (pmap, va).
1794 *
1795 * The caller must hold vm_token.
984263bc
MD
1796 */
1797static void
2bb9cc6f
MD
1798pmap_insert_entry(pmap_t pmap, pv_entry_t pv, vm_offset_t va,
1799 vm_page_t mpte, vm_page_t m)
984263bc 1800{
5926987a
MD
1801#ifdef PMAP_DEBUG
1802 KKASSERT(pv->pv_m == NULL);
1803 pv->pv_m = m;
1804#endif
984263bc
MD
1805 pv->pv_va = va;
1806 pv->pv_pmap = pmap;
1807 pv->pv_ptem = mpte;
1808
1809 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1810 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
5926987a 1811 ++pmap->pm_generation;
984263bc 1812 m->md.pv_list_count++;
cef01e15
MD
1813 if (m->object)
1814 atomic_add_int(&m->object->agg_pv_list_count, 1);
984263bc
MD
1815}
1816
1817/*
5926987a
MD
1818 * pmap_remove_pte: do the things to unmap a page in a process.
1819 *
4107b0c0
MD
1820 * The caller must hold vm_token.
1821 *
1822 * WARNING! As with most other pmap functions this one can block, so
1823 * callers using temporary page table mappings must reload
1824 * them.
984263bc 1825 */
554cf9ac 1826static void
0f7a3396 1827pmap_remove_pte(struct pmap *pmap, unsigned *ptq, vm_offset_t va,
5926987a 1828 pmap_inval_info_t info)
984263bc
MD
1829{
1830 unsigned oldpte;
1831 vm_page_t m;
1832
5926987a 1833 ptbase_assert(pmap);
c2fb025d 1834 pmap_inval_interlock(info, pmap, va);
5926987a 1835 ptbase_assert(pmap);
984263bc
MD
1836 oldpte = loadandclear(ptq);
1837 if (oldpte & PG_W)
1838 pmap->pm_stats.wired_count -= 1;
c2fb025d 1839 pmap_inval_deinterlock(info, pmap);
90244566 1840 KKASSERT(oldpte & PG_V);
984263bc
MD
1841 /*
1842 * Machines that don't support invlpg, also don't support
0f7a3396
MD
1843 * PG_G. XXX PG_G is disabled for SMP so don't worry about
1844 * the SMP case.
984263bc
MD
1845 */
1846 if (oldpte & PG_G)
41a01a4d 1847 cpu_invlpg((void *)va);
eec2b734
MD
1848 KKASSERT(pmap->pm_stats.resident_count > 0);
1849 --pmap->pm_stats.resident_count;
984263bc
MD
1850 if (oldpte & PG_MANAGED) {
1851 m = PHYS_TO_VM_PAGE(oldpte);
1852 if (oldpte & PG_M) {
1853#if defined(PMAP_DIAGNOSTIC)
1854 if (pmap_nw_modified((pt_entry_t) oldpte)) {
d557216f
MD
1855 kprintf("pmap_remove: modified page not "
1856 "writable: va: %p, pte: 0x%lx\n",
1857 (void *)va, (long)oldpte);
984263bc
MD
1858 }
1859#endif
1860 if (pmap_track_modified(va))
1861 vm_page_dirty(m);
1862 }
1863 if (oldpte & PG_A)
1864 vm_page_flag_set(m, PG_REFERENCED);
554cf9ac 1865 pmap_remove_entry(pmap, m, va, info);
984263bc 1866 } else {
554cf9ac 1867 pmap_unuse_pt(pmap, va, NULL, info);
984263bc 1868 }
984263bc
MD
1869}
1870
1871/*
5926987a 1872 * Remove a single page from a process address space.
e0e69b7d 1873 *
4107b0c0 1874 * The caller must hold vm_token.
984263bc
MD
1875 */
1876static void
0f7a3396 1877pmap_remove_page(struct pmap *pmap, vm_offset_t va, pmap_inval_info_t info)
984263bc 1878{
840de426 1879 unsigned *ptq;
984263bc
MD
1880
1881 /*
90244566 1882 * If there is no pte for this address, just skip it!!! Otherwise
e0e69b7d 1883 * get a local va for mappings for this pmap and remove the entry.
984263bc 1884 */
e0e69b7d
MD
1885 if (*pmap_pde(pmap, va) != 0) {
1886 ptq = get_ptbase(pmap) + i386_btop(va);
1887 if (*ptq) {
0f7a3396 1888 pmap_remove_pte(pmap, ptq, va, info);
5926987a 1889 /* ptq invalid */
e0e69b7d 1890 }
984263bc 1891 }
984263bc
MD
1892}
1893
1894/*
4107b0c0 1895 * Remove the given range of addresses from the specified map.
984263bc 1896 *
4107b0c0
MD
1897 * It is assumed that the start and end are properly rounded to the page
1898 * size.
e0e69b7d 1899 *
4107b0c0 1900 * No requirements.
984263bc
MD
1901 */
1902void
840de426 1903pmap_remove(struct pmap *pmap, vm_offset_t sva, vm_offset_t eva)
984263bc 1904{
840de426 1905 unsigned *ptbase;
984263bc
MD
1906 vm_offset_t pdnxt;
1907 vm_offset_t ptpaddr;
1908 vm_offset_t sindex, eindex;
0f7a3396 1909 struct pmap_inval_info info;
984263bc
MD
1910
1911 if (pmap == NULL)
1912 return;
1913
b12defdc 1914 vm_object_hold(pmap->pm_pteobj);
4107b0c0
MD
1915 lwkt_gettoken(&vm_token);
1916 if (pmap->pm_stats.resident_count == 0) {
1917 lwkt_reltoken(&vm_token);
b12defdc 1918 vm_object_drop(pmap->pm_pteobj);
984263bc 1919 return;
4107b0c0 1920 }
984263bc 1921
0f7a3396
MD
1922 pmap_inval_init(&info);
1923
984263bc
MD
1924 /*
1925 * special handling of removing one page. a very
1926 * common operation and easy to short circuit some
1927 * code.
1928 */
1929 if (((sva + PAGE_SIZE) == eva) &&
1930 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
0f7a3396 1931 pmap_remove_page(pmap, sva, &info);
c2fb025d 1932 pmap_inval_done(&info);
4107b0c0 1933 lwkt_reltoken(&vm_token);
b12defdc 1934 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
1935 return;
1936 }
1937
984263bc
MD
1938 /*
1939 * Get a local virtual address for the mappings that are being
1940 * worked with.
1941 */
984263bc
MD
1942 sindex = i386_btop(sva);
1943 eindex = i386_btop(eva);
1944
554cf9ac 1945 while (sindex < eindex) {
984263bc
MD
1946 unsigned pdirindex;
1947
1948 /*
554cf9ac 1949 * Stop scanning if no pages are left
984263bc 1950 */
984263bc
MD
1951 if (pmap->pm_stats.resident_count == 0)
1952 break;
1953
554cf9ac
MD
1954 /*
1955 * Calculate index for next page table, limited by eindex.
1956 */
1957 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1958 if (pdnxt > eindex)
1959 pdnxt = eindex;
1960
984263bc 1961 pdirindex = sindex / NPDEPG;
554cf9ac
MD
1962 ptpaddr = (unsigned)pmap->pm_pdir[pdirindex];
1963 if (ptpaddr & PG_PS) {
c2fb025d 1964 pmap_inval_interlock(&info, pmap, -1);
984263bc
MD
1965 pmap->pm_pdir[pdirindex] = 0;
1966 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
be3aecf7 1967 pmap->pm_cached = 0;
c2fb025d 1968 pmap_inval_deinterlock(&info, pmap);
554cf9ac 1969 sindex = pdnxt;
984263bc
MD
1970 continue;
1971 }
1972
1973 /*
1974 * Weed out invalid mappings. Note: we assume that the page
1975 * directory table is always allocated, and in kernel virtual.
1976 */
554cf9ac
MD
1977 if (ptpaddr == 0) {
1978 sindex = pdnxt;
984263bc 1979 continue;
984263bc
MD
1980 }
1981
8790d7d8 1982 /*
554cf9ac
MD
1983 * Sub-scan the page table page. pmap_remove_pte() can
1984 * block on us, invalidating ptbase, so we must reload
1985 * ptbase and we must also check whether the page directory
1986 * page is still present.
8790d7d8 1987 */
554cf9ac 1988 while (sindex < pdnxt) {
984263bc 1989 vm_offset_t va;
8790d7d8
MD
1990
1991 ptbase = get_ptbase(pmap);
554cf9ac
MD
1992 if (ptbase[sindex]) {
1993 va = i386_ptob(sindex);
1994 pmap_remove_pte(pmap, ptbase + sindex,
1995 va, &info);
1996 }
1997 if (pmap->pm_pdir[pdirindex] == 0 ||
1998 (pmap->pm_pdir[pdirindex] & PG_PS)) {
984263bc 1999 break;
554cf9ac
MD
2000 }
2001 ++sindex;
984263bc
MD
2002 }
2003 }
c2fb025d 2004 pmap_inval_done(&info);
4107b0c0 2005 lwkt_reltoken(&vm_token);
b12defdc 2006 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
2007}
2008
2009/*
4107b0c0
MD
2010 * Removes this physical page from all physical maps in which it resides.
2011 * Reflects back modify bits to the pager.
984263bc 2012 *
3321ee05 2013 * vm_token must be held by caller.
984263bc 2014 */
984263bc 2015static void
840de426 2016pmap_remove_all(vm_page_t m)
984263bc 2017{
0f7a3396 2018 struct pmap_inval_info info;
840de426 2019 unsigned *pte, tpte;
0f7a3396 2020 pv_entry_t pv;
3321ee05 2021 pmap_t pmap;
984263bc 2022
bee81bdd
SS
2023 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2024 return;
bcc6a8ac
MD
2025 if (TAILQ_EMPTY(&m->md.pv_list))
2026 return;
984263bc 2027
0f7a3396 2028 pmap_inval_init(&info);
984263bc 2029 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3321ee05
MD
2030 pmap = pv->pv_pmap;
2031 KKASSERT(pmap->pm_stats.resident_count > 0);
3321ee05 2032 pmap_hold(pmap);
2160c64a
MD
2033 vm_object_hold(pmap->pm_pteobj);
2034
2035 if (pv != TAILQ_FIRST(&m->md.pv_list)) {
2160c64a
MD
2036 vm_object_drop(pmap->pm_pteobj);
2037 pmap_drop(pmap);
2038 continue;
2039 }
984263bc 2040
1c10d640 2041 --pmap->pm_stats.resident_count;
3321ee05
MD
2042 pte = pmap_pte_quick(pmap, pv->pv_va);
2043 pmap_inval_interlock(&info, pmap, pv->pv_va);
984263bc
MD
2044 tpte = loadandclear(pte);
2045 if (tpte & PG_W)
3321ee05
MD
2046 pmap->pm_stats.wired_count--;
2047 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2048 if (tpte & PG_A)
2049 vm_page_flag_set(m, PG_REFERENCED);
c2fb025d 2050 KKASSERT(PHYS_TO_VM_PAGE(tpte) == m);
984263bc
MD
2051
2052 /*
2053 * Update the vm_page_t clean and reference bits.
2054 */
2055 if (tpte & PG_M) {
2056#if defined(PMAP_DIAGNOSTIC)
2057 if (pmap_nw_modified((pt_entry_t) tpte)) {
d557216f
MD
2058 kprintf("pmap_remove_all: modified page "
2059 "not writable: va: %p, pte: 0x%lx\n",
2060 (void *)pv->pv_va, (long)tpte);
984263bc
MD
2061 }
2062#endif
2063 if (pmap_track_modified(pv->pv_va))
2064 vm_page_dirty(m);
2065 }
5926987a
MD
2066#ifdef PMAP_DEBUG
2067 KKASSERT(pv->pv_m == m);
2068#endif
2bb9cc6f 2069 KKASSERT(pv == TAILQ_FIRST(&m->md.pv_list));
984263bc 2070 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3321ee05
MD
2071 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
2072 ++pmap->pm_generation;
984263bc 2073 m->md.pv_list_count--;
cef01e15
MD
2074 if (m->object)
2075 atomic_add_int(&m->object->agg_pv_list_count, -1);
17cde63e
MD
2076 if (TAILQ_EMPTY(&m->md.pv_list))
2077 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
3321ee05
MD
2078 pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem, &info);
2079 vm_object_drop(pmap->pm_pteobj);
984263bc 2080 free_pv_entry(pv);
3321ee05 2081 pmap_drop(pmap);
984263bc 2082 }
17cde63e 2083 KKASSERT((m->flags & (PG_MAPPED|PG_WRITEABLE)) == 0);
c2fb025d 2084 pmap_inval_done(&info);
984263bc
MD
2085}
2086
2087/*
4107b0c0
MD
2088 * Set the physical protection on the specified range of this map
2089 * as requested.
e0e69b7d 2090 *
4107b0c0 2091 * No requirements.
984263bc
MD
2092 */
2093void
2094pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
2095{
840de426 2096 unsigned *ptbase;
984263bc
MD
2097 vm_offset_t pdnxt, ptpaddr;
2098 vm_pindex_t sindex, eindex;
0f7a3396 2099 pmap_inval_info info;
984263bc
MD
2100
2101 if (pmap == NULL)
2102 return;
2103
2104 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2105 pmap_remove(pmap, sva, eva);
2106 return;
2107 }
2108
2109 if (prot & VM_PROT_WRITE)
2110 return;
2111
4107b0c0 2112 lwkt_gettoken(&vm_token);
0f7a3396 2113 pmap_inval_init(&info);
984263bc
MD
2114
2115 ptbase = get_ptbase(pmap);
2116
2117 sindex = i386_btop(sva);
2118 eindex = i386_btop(eva);
2119
2120 for (; sindex < eindex; sindex = pdnxt) {
984263bc
MD
2121 unsigned pdirindex;
2122
2123 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
2124
2125 pdirindex = sindex / NPDEPG;
2126 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
c2fb025d 2127 pmap_inval_interlock(&info, pmap, -1);
55f2596a 2128 pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
984263bc 2129 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
c2fb025d 2130 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2131 continue;
2132 }
2133
2134 /*
2135 * Weed out invalid mappings. Note: we assume that the page
2136 * directory table is always allocated, and in kernel virtual.
2137 */
2138 if (ptpaddr == 0)
2139 continue;
2140
2141 if (pdnxt > eindex) {
2142 pdnxt = eindex;
2143 }
2144
2145 for (; sindex != pdnxt; sindex++) {
984263bc 2146 unsigned pbits;
c2fb025d 2147 unsigned cbits;
984263bc
MD
2148 vm_page_t m;
2149
17cde63e 2150 /*
d5b2d319 2151 * XXX non-optimal.
17cde63e 2152 */
c2fb025d
MD
2153 pmap_inval_interlock(&info, pmap, i386_ptob(sindex));
2154again:
984263bc 2155 pbits = ptbase[sindex];
c2fb025d 2156 cbits = pbits;
984263bc
MD
2157
2158 if (pbits & PG_MANAGED) {
2159 m = NULL;
2160 if (pbits & PG_A) {
2161 m = PHYS_TO_VM_PAGE(pbits);
2162 vm_page_flag_set(m, PG_REFERENCED);
c2fb025d 2163 cbits &= ~PG_A;
984263bc
MD
2164 }
2165 if (pbits & PG_M) {
2166 if (pmap_track_modified(i386_ptob(sindex))) {
2167 if (m == NULL)
2168 m = PHYS_TO_VM_PAGE(pbits);
2169 vm_page_dirty(m);
c2fb025d 2170 cbits &= ~PG_M;
984263bc
MD
2171 }
2172 }
2173 }
c2fb025d
MD
2174 cbits &= ~PG_RW;
2175 if (pbits != cbits &&
2176 !atomic_cmpset_int(ptbase + sindex, pbits, cbits)) {
2177 goto again;
984263bc 2178 }
c2fb025d 2179 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2180 }
2181 }
c2fb025d 2182 pmap_inval_done(&info);
4107b0c0 2183 lwkt_reltoken(&vm_token);
984263bc
MD
2184}
2185
2186/*
4107b0c0
MD
2187 * Insert the given physical page (p) at the specified virtual address (v)
2188 * in the target physical map with the protection requested.
984263bc 2189 *
4107b0c0
MD
2190 * If specified, the page will be wired down, meaning that the related pte
2191 * cannot be reclaimed.
984263bc 2192 *
4107b0c0 2193 * No requirements.
984263bc
MD
2194 */
2195void
2196pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
921c891e 2197 boolean_t wired, vm_map_entry_t entry __unused)
984263bc 2198{
6ef943a3 2199 vm_paddr_t pa;
840de426 2200 unsigned *pte;
6ef943a3 2201 vm_paddr_t opa;
984263bc
MD
2202 vm_offset_t origpte, newpte;
2203 vm_page_t mpte;
0f7a3396 2204 pmap_inval_info info;
2bb9cc6f 2205 pv_entry_t pv;
984263bc
MD
2206
2207 if (pmap == NULL)
2208 return;
2209
2210 va &= PG_FRAME;
2211#ifdef PMAP_DIAGNOSTIC
c439ad8f 2212 if (va >= KvaEnd)
984263bc 2213 panic("pmap_enter: toobig");
d557216f
MD
2214 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) {
2215 panic("pmap_enter: invalid to pmap_enter page "
2216 "table pages (va: %p)", (void *)va);
2217 }
984263bc 2218#endif
fbbaeba3
MD
2219 if (va < UPT_MAX_ADDRESS && pmap == &kernel_pmap) {
2220 kprintf("Warning: pmap_enter called on UVA with kernel_pmap\n");
7ce2998e 2221 print_backtrace(-1);
fbbaeba3
MD
2222 }
2223 if (va >= UPT_MAX_ADDRESS && pmap != &kernel_pmap) {
2224 kprintf("Warning: pmap_enter called on KVA without kernel_pmap\n");
7ce2998e 2225 print_backtrace(-1);
fbbaeba3 2226 }
984263bc 2227
b12defdc 2228 vm_object_hold(pmap->pm_pteobj);
4107b0c0
MD
2229 lwkt_gettoken(&vm_token);
2230
2bb9cc6f
MD
2231 /*
2232 * This can block, get it before we do anything important.
2233 */
2234 if (pmap_initialized &&
2235 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2236 pv = get_pv_entry();
2237 } else {
2238 pv = NULL;
2239 }
2240
984263bc
MD
2241 /*
2242 * In the case that a page table page is not
2243 * resident, we are creating it here.
2244 */
17cde63e 2245 if (va < UPT_MIN_ADDRESS)
984263bc 2246 mpte = pmap_allocpte(pmap, va);
17cde63e
MD
2247 else
2248 mpte = NULL;
984263bc 2249
b12defdc
MD
2250 if ((prot & VM_PROT_NOSYNC) == 0)
2251 pmap_inval_init(&info);
984263bc
MD
2252 pte = pmap_pte(pmap, va);
2253
2254 /*
2255 * Page Directory table entry not valid, we need a new PT page
2256 */
2257 if (pte == NULL) {
ed20d0e3 2258 panic("pmap_enter: invalid page directory pdir=0x%lx, va=%p",
d557216f 2259 (long)pmap->pm_pdir[PTDPTDI], (void *)va);
984263bc
MD
2260 }
2261
2262 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2263 origpte = *(vm_offset_t *)pte;
2264 opa = origpte & PG_FRAME;
2265
2266 if (origpte & PG_PS)
2267 panic("pmap_enter: attempted pmap_enter on 4MB page");
2268
2269 /*
2270 * Mapping has not changed, must be protection or wiring change.
2271 */
2272 if (origpte && (opa == pa)) {
2273 /*
2274 * Wiring change, just update stats. We don't worry about
2275 * wiring PT pages as they remain resident as long as there
2276 * are valid mappings in them. Hence, if a user page is wired,
2277 * the PT page will be also.
2278 */
2279 if (wired && ((origpte & PG_W) == 0))
2280 pmap->pm_stats.wired_count++;
2281 else if (!wired && (origpte & PG_W))
2282 pmap->pm_stats.wired_count--;
2283
2284#if defined(PMAP_DIAGNOSTIC)
2285 if (pmap_nw_modified((pt_entry_t) origpte)) {
d557216f
MD
2286 kprintf("pmap_enter: modified page not "
2287 "writable: va: %p, pte: 0x%lx\n",
2288 (void *)va, (long )origpte);
984263bc
MD
2289 }
2290#endif
2291
984263bc
MD
2292 /*
2293 * We might be turning off write access to the page,
2294 * so we go ahead and sense modify status.
2295 */
2296 if (origpte & PG_MANAGED) {
2297 if ((origpte & PG_M) && pmap_track_modified(va)) {
2298 vm_page_t om;
2299 om = PHYS_TO_VM_PAGE(opa);
2300 vm_page_dirty(om);
2301 }
2302 pa |= PG_MANAGED;
17cde63e 2303 KKASSERT(m->flags & PG_MAPPED);
984263bc
MD
2304 }
2305 goto validate;
2306 }
2307 /*
2308 * Mapping has changed, invalidate old range and fall through to
2309 * handle validating new mapping.
5926987a
MD
2310 *
2311 * Since we have a ref on the page directory page pmap_pte()
2312 * will always return non-NULL.
2313 *
2314 * NOTE: pmap_remove_pte() can block and cause the temporary ptbase
2315 * to get wiped. reload the ptbase. I'm not sure if it is
2316 * also possible to race another pmap_enter() but check for
2317 * that case too.
984263bc 2318 */
5926987a 2319 while (opa) {
5926987a
MD
2320 KKASSERT((origpte & PG_FRAME) ==
2321 (*(vm_offset_t *)pte & PG_FRAME));
39efc543
MD
2322 if (prot & VM_PROT_NOSYNC) {
2323 prot &= ~VM_PROT_NOSYNC;
2324 pmap_inval_init(&info);
2325 }
554cf9ac 2326 pmap_remove_pte(pmap, pte, va, &info);
5926987a
MD
2327 pte = pmap_pte(pmap, va);
2328 origpte = *(vm_offset_t *)pte;
2329 opa = origpte & PG_FRAME;
2330 if (opa) {
2331 kprintf("pmap_enter: Warning, raced pmap %p va %p\n",
2332 pmap, (void *)va);
2333 }
984263bc
MD
2334 }
2335
2336 /*
2337 * Enter on the PV list if part of our managed memory. Note that we
2338 * raise IPL while manipulating pv_table since pmap_enter can be
2339 * called at interrupt time.
2340 */
2341 if (pmap_initialized &&
2342 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2bb9cc6f
MD
2343 pmap_insert_entry(pmap, pv, va, mpte, m);
2344 pv = NULL;
5926987a 2345 ptbase_assert(pmap);
984263bc 2346 pa |= PG_MANAGED;
17cde63e 2347 vm_page_flag_set(m, PG_MAPPED);
984263bc
MD
2348 }
2349
2350 /*
2351 * Increment counters
2352 */
eec2b734 2353 ++pmap->pm_stats.resident_count;
984263bc
MD
2354 if (wired)
2355 pmap->pm_stats.wired_count++;
5926987a 2356 KKASSERT(*pte == 0);
984263bc
MD
2357
2358validate:
2359 /*
2360 * Now validate mapping with desired protection/wiring.
2361 */
5926987a 2362 ptbase_assert(pmap);
984263bc
MD
2363 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2364
2365 if (wired)
2366 newpte |= PG_W;
2367 if (va < UPT_MIN_ADDRESS)
2368 newpte |= PG_U;
fbbaeba3 2369 if (pmap == &kernel_pmap)
984263bc
MD
2370 newpte |= pgeflag;
2371
2372 /*
2bb9cc6f
MD
2373 * If the mapping or permission bits are different, we need
2374 * to update the pte. If the pte is already present we have
2375 * to get rid of the extra wire-count on mpte we had obtained
2376 * above.
b1482674
MD
2377 *
2378 * mpte has a new wire_count, which also serves to prevent the
2379 * page table page from getting ripped out while we work. If we
2380 * are modifying an existing pte instead of installing a new one
2381 * we have to drop it.
984263bc
MD
2382 */
2383 if ((origpte & ~(PG_M|PG_A)) != newpte) {
b12defdc
MD
2384 if (prot & VM_PROT_NOSYNC)
2385 cpu_invlpg((void *)va);
2386 else
2387 pmap_inval_interlock(&info, pmap, va);
5926987a 2388 ptbase_assert(pmap);
2bb9cc6f
MD
2389
2390 if (*pte) {
2391 KKASSERT((*pte & PG_FRAME) == (newpte & PG_FRAME));
4bb41982 2392 if (mpte && vm_page_unwire_quick(mpte))
2bb9cc6f
MD
2393 panic("pmap_enter: Insufficient wire_count");
2394 }
2395
984263bc 2396 *pte = newpte | PG_A;
b12defdc
MD
2397 if ((prot & VM_PROT_NOSYNC) == 0)
2398 pmap_inval_deinterlock(&info, pmap);
17cde63e
MD
2399 if (newpte & PG_RW)
2400 vm_page_flag_set(m, PG_WRITEABLE);
b1482674
MD
2401 } else {
2402 if (*pte) {
2403 KKASSERT((*pte & PG_FRAME) == (newpte & PG_FRAME));
4bb41982 2404 if (mpte && vm_page_unwire_quick(mpte))
b1482674
MD
2405 panic("pmap_enter: Insufficient wire_count");
2406 }
984263bc 2407 }
b1482674
MD
2408
2409 /*
2410 * NOTE: mpte invalid after this point if we block.
2411 */
c695044a 2412 KKASSERT((newpte & PG_MANAGED) == 0 || (m->flags & PG_MAPPED));
b12defdc
MD
2413 if ((prot & VM_PROT_NOSYNC) == 0)
2414 pmap_inval_done(&info);
2bb9cc6f
MD
2415 if (pv)
2416 free_pv_entry(pv);
4107b0c0 2417 lwkt_reltoken(&vm_token);
b12defdc 2418 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
2419}
2420
2421/*
17cde63e
MD
2422 * This code works like pmap_enter() but assumes VM_PROT_READ and not-wired.
2423 * This code also assumes that the pmap has no pre-existing entry for this
2424 * VA.
2425 *
2426 * This code currently may only be used on user pmaps, not kernel_pmap.
4107b0c0
MD
2427 *
2428 * No requirements.
984263bc 2429 */
1b9d3514 2430void
17cde63e 2431pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m)
984263bc
MD
2432{
2433 unsigned *pte;
6ef943a3 2434 vm_paddr_t pa;
17cde63e
MD
2435 vm_page_t mpte;
2436 unsigned ptepindex;
2437 vm_offset_t ptepa;
0f7a3396 2438 pmap_inval_info info;
2bb9cc6f 2439 pv_entry_t pv;
0f7a3396 2440
b12defdc 2441 vm_object_hold(pmap->pm_pteobj);
4107b0c0 2442 lwkt_gettoken(&vm_token);
2bb9cc6f
MD
2443
2444 /*
2445 * This can block, get it before we do anything important.
2446 */
2447 if (pmap_initialized &&
2448 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2449 pv = get_pv_entry();
2450 } else {
2451 pv = NULL;
2452 }
2453
0f7a3396 2454 pmap_inval_init(&info);
984263bc 2455
fbbaeba3
MD
2456 if (va < UPT_MAX_ADDRESS && pmap == &kernel_pmap) {
2457 kprintf("Warning: pmap_enter_quick called on UVA with kernel_pmap\n");
7ce2998e 2458 print_backtrace(-1);
fbbaeba3
MD
2459 }
2460 if (va >= UPT_MAX_ADDRESS && pmap != &kernel_pmap) {
2461 kprintf("Warning: pmap_enter_quick called on KVA without kernel_pmap\n");
7ce2998e 2462 print_backtrace(-1);
fbbaeba3
MD
2463 }
2464
17cde63e
MD
2465 KKASSERT(va < UPT_MIN_ADDRESS); /* assert used on user pmaps only */
2466
984263bc 2467 /*
17cde63e
MD
2468 * Calculate the page table page (mpte), allocating it if necessary.
2469 *
2470 * A held page table page (mpte), or NULL, is passed onto the
2471 * section following.
984263bc
MD
2472 */
2473 if (va < UPT_MIN_ADDRESS) {
984263bc
MD
2474 /*
2475 * Calculate pagetable page index
2476 */
2477 ptepindex = va >> PDRSHIFT;
17cde63e
MD
2478
2479 do {
984263bc
MD
2480 /*
2481 * Get the page directory entry
2482 */
2483 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2484
2485 /*
2486 * If the page table page is mapped, we just increment
90244566 2487 * the wire count, and activate it.
984263bc
MD
2488 */
2489 if (ptepa) {
2490 if (ptepa & PG_PS)
2491 panic("pmap_enter_quick: unexpected mapping into 4MB page");
b1482674
MD
2492 if ((mpte = pmap->pm_ptphint) != NULL &&
2493 (mpte->pindex == ptepindex) &&
2494 (mpte->flags & PG_BUSY) == 0) {
2bb9cc6f 2495 vm_page_wire_quick(mpte);
984263bc 2496 } else {
2bb9cc6f
MD
2497 mpte = pmap_page_lookup(pmap->pm_pteobj,
2498 ptepindex);
984263bc 2499 pmap->pm_ptphint = mpte;
2bb9cc6f 2500 vm_page_wire_quick(mpte);
b12defdc 2501 vm_page_wakeup(mpte);
984263bc 2502 }
984263bc
MD
2503 } else {
2504 mpte = _pmap_allocpte(pmap, ptepindex);
2505 }
17cde63e 2506 } while (mpte == NULL);
984263bc
MD
2507 } else {
2508 mpte = NULL;
17cde63e 2509 /* this code path is not yet used */
984263bc
MD
2510 }
2511
2512 /*
17cde63e
MD
2513 * With a valid (and held) page directory page, we can just use
2514 * vtopte() to get to the pte. If the pte is already present
2515 * we do not disturb it.
984263bc
MD
2516 */
2517 pte = (unsigned *)vtopte(va);
554cf9ac
MD
2518 if (*pte) {
2519 KKASSERT(*pte & PG_V);
17cde63e
MD
2520 pa = VM_PAGE_TO_PHYS(m);
2521 KKASSERT(((*pte ^ pa) & PG_FRAME) == 0);
c2fb025d 2522 pmap_inval_done(&info);
b1482674
MD
2523 if (mpte)
2524 pmap_unwire_pte(pmap, mpte, &info);
b1482674 2525 if (pv) {
2bb9cc6f 2526 free_pv_entry(pv);
b1482674
MD
2527 /* pv = NULL; */
2528 }
554cf9ac
MD
2529 lwkt_reltoken(&vm_token);
2530 vm_object_drop(pmap->pm_pteobj);
17cde63e 2531 return;
984263bc
MD
2532 }
2533
2534 /*
17cde63e 2535 * Enter on the PV list if part of our managed memory
984263bc 2536 */
2bb9cc6f
MD
2537 if (pmap_initialized &&
2538 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2539 pmap_insert_entry(pmap, pv, va, mpte, m);
2540 pv = NULL;
17cde63e
MD
2541 vm_page_flag_set(m, PG_MAPPED);
2542 }
984263bc
MD
2543
2544 /*
2545 * Increment counters
2546 */
eec2b734 2547 ++pmap->pm_stats.resident_count;
984263bc
MD
2548
2549 pa = VM_PAGE_TO_PHYS(m);
2550
2551 /*
2552 * Now validate mapping with RO protection
2553 */
2554 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2555 *pte = pa | PG_V | PG_U;
2556 else
2557 *pte = pa | PG_V | PG_U | PG_MANAGED;
17cde63e 2558/* pmap_inval_add(&info, pmap, va); shouldn't be needed inval->valid */
c2fb025d 2559 pmap_inval_done(&info);
b1482674 2560 if (pv) {
2bb9cc6f 2561 free_pv_entry(pv);
b1482674
MD
2562 /* pv = NULL; */
2563 }
4107b0c0 2564 lwkt_reltoken(&vm_token);
b12defdc 2565 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
2566}
2567
2568/*
2569 * Make a temporary mapping for a physical address. This is only intended
2570 * to be used for panic dumps.
4107b0c0 2571 *
fb8345e6
MD
2572 * The caller is responsible for calling smp_invltlb().
2573 *
4107b0c0 2574 * No requirements.
984263bc
MD
2575 */
2576void *
8e5ea5f7 2577pmap_kenter_temporary(vm_paddr_t pa, long i)
984263bc 2578{
fb8345e6 2579 pmap_kenter_quick((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
984263bc
MD
2580 return ((void *)crashdumpmap);
2581}
2582
2583#define MAX_INIT_PT (96)
06ecca5a 2584
984263bc 2585/*
06ecca5a
MD
2586 * This routine preloads the ptes for a given object into the specified pmap.
2587 * This eliminates the blast of soft faults on process startup and
2588 * immediately after an mmap.
4107b0c0
MD
2589 *
2590 * No requirements.
984263bc 2591 */
1f804340
MD
2592static int pmap_object_init_pt_callback(vm_page_t p, void *data);
2593
984263bc 2594void
083a7402
MD
2595pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_prot_t prot,
2596 vm_object_t object, vm_pindex_t pindex,
2597 vm_size_t size, int limit)
984263bc 2598{
1f804340 2599 struct rb_vm_page_scan_info info;
287ebb09 2600 struct lwp *lp;
984263bc 2601 int psize;
984263bc 2602
54a764e8
MD
2603 /*
2604 * We can't preinit if read access isn't set or there is no pmap
2605 * or object.
2606 */
083a7402 2607 if ((prot & VM_PROT_READ) == 0 || pmap == NULL || object == NULL)
984263bc
MD
2608 return;
2609
54a764e8
MD
2610 /*
2611 * We can't preinit if the pmap is not the current pmap
2612 */
287ebb09
MD
2613 lp = curthread->td_lwp;
2614 if (lp == NULL || pmap != vmspace_pmap(lp->lwp_vmspace))
54a764e8
MD
2615 return;
2616
984263bc
MD
2617 psize = i386_btop(size);
2618
2619 if ((object->type != OBJT_VNODE) ||
2620 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2621 (object->resident_page_count > MAX_INIT_PT))) {
2622 return;
2623 }
2624
2625 if (psize + pindex > object->size) {
2626 if (object->size < pindex)
2627 return;
2628 psize = object->size - pindex;
2629 }
2630
1f804340
MD
2631 if (psize == 0)
2632 return;
06ecca5a 2633
984263bc 2634 /*
1f804340
MD
2635 * Use a red-black scan to traverse the requested range and load
2636 * any valid pages found into the pmap.
06ecca5a 2637 *
9acd5bbb
MD
2638 * We cannot safely scan the object's memq unless we are in a
2639 * critical section since interrupts can remove pages from objects.
984263bc 2640 */
1f804340
MD
2641 info.start_pindex = pindex;
2642 info.end_pindex = pindex + psize - 1;
2643 info.limit = limit;
2644 info.mpte = NULL;
2645 info.addr = addr;
2646 info.pmap = pmap;
2647
2f2d9e58 2648 vm_object_hold(object);
1f804340
MD
2649 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
2650 pmap_object_init_pt_callback, &info);
2f2d9e58 2651 vm_object_drop(object);
1f804340 2652}
06ecca5a 2653
4107b0c0
MD
2654/*
2655 * The caller must hold vm_token.
2656 */
1f804340
MD
2657static
2658int
2659pmap_object_init_pt_callback(vm_page_t p, void *data)
2660{
2661 struct rb_vm_page_scan_info *info = data;
2662 vm_pindex_t rel_index;
2663 /*
2664 * don't allow an madvise to blow away our really
2665 * free pages allocating pv entries.
2666 */
2667 if ((info->limit & MAP_PREFAULT_MADVISE) &&
2668 vmstats.v_free_count < vmstats.v_free_reserved) {
2669 return(-1);
984263bc 2670 }
0d987a03
MD
2671
2672 /*
2673 * Ignore list markers and ignore pages we cannot instantly
2674 * busy (while holding the object token).
2675 */
2676 if (p->flags & PG_MARKER)
2677 return 0;
b12defdc
MD
2678 if (vm_page_busy_try(p, TRUE))
2679 return 0;
1f804340 2680 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
b12defdc 2681 (p->flags & PG_FICTITIOUS) == 0) {
1f804340
MD
2682 if ((p->queue - p->pc) == PQ_CACHE)
2683 vm_page_deactivate(p);
1f804340 2684 rel_index = p->pindex - info->start_pindex;
17cde63e
MD
2685 pmap_enter_quick(info->pmap,
2686 info->addr + i386_ptob(rel_index), p);
1f804340 2687 }
b12defdc 2688 vm_page_wakeup(p);
1f804340 2689 return(0);
984263bc
MD
2690}
2691
2692/*
1b9d3514
MD
2693 * Return TRUE if the pmap is in shape to trivially
2694 * pre-fault the specified address.
2695 *
2696 * Returns FALSE if it would be non-trivial or if a
2697 * pte is already loaded into the slot.
4107b0c0
MD
2698 *
2699 * No requirements.
984263bc 2700 */
1b9d3514
MD
2701int
2702pmap_prefault_ok(pmap_t pmap, vm_offset_t addr)
984263bc 2703{
1b9d3514 2704 unsigned *pte;
4107b0c0 2705 int ret;
984263bc 2706
4107b0c0
MD
2707 lwkt_gettoken(&vm_token);
2708 if ((*pmap_pde(pmap, addr)) == 0) {
2709 ret = 0;
2710 } else {
2711 pte = (unsigned *) vtopte(addr);
2712 ret = (*pte) ? 0 : 1;
2713 }
2714 lwkt_reltoken(&vm_token);
2715 return(ret);
984263bc
MD
2716}
2717
2718/*
4107b0c0
MD
2719 * Change the wiring attribute for a map/virtual-adderss pair. The mapping
2720 * must already exist.
2721 *
2722 * No requirements.
984263bc
MD
2723 */
2724void
921c891e
MD
2725pmap_change_wiring(pmap_t pmap, vm_offset_t va, boolean_t wired,
2726 vm_map_entry_t entry __unused)
984263bc 2727{
840de426 2728 unsigned *pte;
984263bc
MD
2729
2730 if (pmap == NULL)
2731 return;
2732
4107b0c0 2733 lwkt_gettoken(&vm_token);
984263bc
MD
2734 pte = pmap_pte(pmap, va);
2735
2736 if (wired && !pmap_pte_w(pte))
2737 pmap->pm_stats.wired_count++;
2738 else if (!wired && pmap_pte_w(pte))
2739 pmap->pm_stats.wired_count--;
2740
2741 /*
2742 * Wiring is not a hardware characteristic so there is no need to
0f7a3396
MD
2743 * invalidate TLB. However, in an SMP environment we must use
2744 * a locked bus cycle to update the pte (if we are not using
2745 * the pmap_inval_*() API that is)... it's ok to do this for simple
2746 * wiring changes.
984263bc 2747 */
0f7a3396
MD
2748 if (wired)
2749 atomic_set_int(pte, PG_W);
2750 else
2751 atomic_clear_int(pte, PG_W);
4107b0c0 2752 lwkt_reltoken(&vm_token);
984263bc
MD
2753}
2754
984263bc 2755/*
4107b0c0
MD
2756 * Copy the range specified by src_addr/len from the source map to the
2757 * range dst_addr/len in the destination map.
2758 *
2759 * This routine is only advisory and need not do anything.
984263bc 2760 *
4107b0c0 2761 * No requirements.
984263bc 2762 */
984263bc 2763void
840de426 2764pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
4107b0c0 2765 vm_size_t len, vm_offset_t src_addr)
984263bc 2766{
4107b0c0 2767 /* does nothing */
984263bc
MD
2768}
2769
984263bc 2770/*
4107b0c0
MD
2771 * Zero the specified PA by mapping the page into KVM and clearing its
2772 * contents.
e0e69b7d 2773 *
4107b0c0 2774 * No requirements.
984263bc
MD
2775 */
2776void
6ef943a3 2777pmap_zero_page(vm_paddr_t phys)
984263bc 2778{
85100692 2779 struct mdglobaldata *gd = mdcpu;
17a9f566 2780
e0e69b7d 2781 crit_enter();
85100692
MD
2782 if (*(int *)gd->gd_CMAP3)
2783 panic("pmap_zero_page: CMAP3 busy");
85100692 2784 *(int *)gd->gd_CMAP3 =
17a9f566 2785 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
85100692 2786 cpu_invlpg(gd->gd_CADDR3);
1fa15583 2787 bzero(gd->gd_CADDR3, PAGE_SIZE);
85100692 2788 *(int *) gd->gd_CMAP3 = 0;
e0e69b7d 2789 crit_exit();
8100156a
MD
2790}
2791
2792/*
4107b0c0 2793 * Assert that a page is empty, panic if it isn't.
8100156a 2794 *
4107b0c0 2795 * No requirements.
8100156a
MD
2796 */
2797void
2798pmap_page_assertzero(vm_paddr_t phys)
2799{
2800 struct mdglobaldata *gd = mdcpu;
2801 int i;
2802
2803 crit_enter();
2804 if (*(int *)gd->gd_CMAP3)
2805 panic("pmap_zero_page: CMAP3 busy");
2806 *(int *)gd->gd_CMAP3 =
2807 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2808 cpu_invlpg(gd->gd_CADDR3);
2809 for (i = 0; i < PAGE_SIZE; i += 4) {
2810 if (*(int *)((char *)gd->gd_CADDR3 + i) != 0) {
ed20d0e3 2811 panic("pmap_page_assertzero() @ %p not zero!",
8100156a
MD
2812 (void *)gd->gd_CADDR3);
2813 }
2814 }
2815 *(int *) gd->gd_CMAP3 = 0;
2816 crit_exit();
984263bc
MD
2817}
2818
2819/*
4107b0c0
MD
2820 * Zero part of a physical page by mapping it into memory and clearing
2821 * its contents with bzero.
e0e69b7d 2822 *
4107b0c0 2823 * off and size may not cover an area beyond a single hardware page.
984263bc 2824 *
4107b0c0 2825 * No requirements.
984263bc
MD
2826 */
2827void
6ef943a3 2828pmap_zero_page_area(vm_paddr_t phys, int off, int size)
984263bc 2829{
85100692 2830 struct mdglobaldata *gd = mdcpu;
17a9f566 2831
e0e69b7d 2832 crit_enter();
85100692
MD
2833 if (*(int *) gd->gd_CMAP3)
2834 panic("pmap_zero_page: CMAP3 busy");
85100692
MD
2835 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2836 cpu_invlpg(gd->gd_CADDR3);
1fa15583 2837 bzero((char *)gd->gd_CADDR3 + off, size);
85100692 2838 *(int *) gd->gd_CMAP3 = 0;
e0e69b7d 2839 crit_exit();
984263bc
MD
2840}
2841
2842/*
4107b0c0
MD
2843 * Copy the physical page from the source PA to the target PA.
2844 * This function may be called from an interrupt. No locking
2845 * is required.
e0e69b7d 2846 *
4107b0c0 2847 * No requirements.
984263bc
MD
2848 */
2849void
6ef943a3 2850pmap_copy_page(vm_paddr_t src, vm_paddr_t dst)
984263bc 2851{
85100692 2852 struct mdglobaldata *gd = mdcpu;
17a9f566 2853
e0e69b7d 2854 crit_enter();
85100692
MD
2855 if (*(int *) gd->gd_CMAP1)
2856 panic("pmap_copy_page: CMAP1 busy");
2857 if (*(int *) gd->gd_CMAP2)
2858 panic("pmap_copy_page: CMAP2 busy");
984263bc 2859
85100692
MD
2860 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2861 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
984263bc 2862
85100692
MD
2863 cpu_invlpg(gd->gd_CADDR1);
2864 cpu_invlpg(gd->gd_CADDR2);
984263bc 2865
85100692 2866 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
984263bc 2867
85100692
MD
2868 *(int *) gd->gd_CMAP1 = 0;
2869 *(int *) gd->gd_CMAP2 = 0;
e0e69b7d 2870 crit_exit();
984263bc
MD
2871}
2872
f6bf3af1 2873/*
4107b0c0
MD
2874 * Copy the physical page from the source PA to the target PA.
2875 * This function may be called from an interrupt. No locking
2876 * is required.
f6bf3af1 2877 *
4107b0c0 2878 * No requirements.
f6bf3af1
MD
2879 */
2880void
2881pmap_copy_page_frag(vm_paddr_t src, vm_paddr_t dst, size_t bytes)
2882{
2883 struct mdglobaldata *gd = mdcpu;
2884
2885 crit_enter();
2886 if (*(int *) gd->gd_CMAP1)
2887 panic("pmap_copy_page: CMAP1 busy");
2888 if (*(int *) gd->gd_CMAP2)
2889 panic("pmap_copy_page: CMAP2 busy");
2890
2891 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2892 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2893
2894 cpu_invlpg(gd->gd_CADDR1);
2895 cpu_invlpg(gd->gd_CADDR2);
2896
2897 bcopy((char *)gd->gd_CADDR1 + (src & PAGE_MASK),
2898 (char *)gd->gd_CADDR2 + (dst & PAGE_MASK),
2899 bytes);
2900
2901 *(int *) gd->gd_CMAP1 = 0;
2902 *(int *) gd->gd_CMAP2 = 0;
2903 crit_exit();
2904}
2905
984263bc
MD
2906/*
2907 * Returns true if the pmap's pv is one of the first
2908 * 16 pvs linked to from this page. This count may
2909 * be changed upwards or downwards in the future; it
2910 * is only necessary that true be returned for a small
2911 * subset of pmaps for proper page aging.
4107b0c0
MD
2912 *
2913 * No requirements.
984263bc
MD
2914 */
2915boolean_t
840de426 2916pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
984263bc
MD
2917{
2918 pv_entry_t pv;
2919 int loops = 0;
984263bc
MD
2920
2921 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2922 return FALSE;
2923
4107b0c0 2924 lwkt_gettoken(&vm_token);
984263bc
MD
2925 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2926 if (pv->pv_pmap == pmap) {
11502947 2927 lwkt_reltoken(&vm_token);
984263bc
MD
2928 return TRUE;
2929 }
2930 loops++;
2931 if (loops >= 16)
2932 break;
2933 }
4107b0c0 2934 lwkt_reltoken(&vm_token);
984263bc
MD
2935 return (FALSE);
2936}
2937
984263bc
MD
2938/*
2939 * Remove all pages from specified address space
2940 * this aids process exit speeds. Also, this code
2941 * is special cased for current process only, but
2942 * can have the more generic (and slightly slower)
2943 * mode enabled. This is much faster than pmap_remove
2944 * in the case of running down an entire address space.
4107b0c0
MD
2945 *
2946 * No requirements.
984263bc
MD
2947 */
2948void
840de426 2949pmap_remove_pages(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
984263bc 2950{
287ebb09 2951 struct lwp *lp;
984263bc
MD
2952 unsigned *pte, tpte;
2953 pv_entry_t pv, npv;
984263bc 2954 vm_page_t m;
0f7a3396 2955 pmap_inval_info info;
4a22e893 2956 int iscurrentpmap;
8790d7d8 2957 int32_t save_generation;
984263bc 2958
287ebb09
MD
2959 lp = curthread->td_lwp;
2960 if (lp && pmap == vmspace_pmap(lp->lwp_vmspace))
4a22e893
MD
2961 iscurrentpmap = 1;
2962 else
2963 iscurrentpmap = 0;
984263bc 2964
b12defdc
MD
2965 if (pmap->pm_pteobj)
2966 vm_object_hold(pmap->pm_pteobj);
4107b0c0 2967 lwkt_gettoken(&vm_token);
0f7a3396 2968 pmap_inval_init(&info);
b12defdc 2969
4a22e893 2970 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); pv; pv = npv) {
984263bc
MD
2971 if (pv->pv_va >= eva || pv->pv_va < sva) {
2972 npv = TAILQ_NEXT(pv, pv_plist);
2973 continue;
2974 }
2975
8790d7d8
MD
2976 KKASSERT(pmap == pv->pv_pmap);
2977
4a22e893
MD
2978 if (iscurrentpmap)
2979 pte = (unsigned *)vtopte(pv->pv_va);
2980 else
8790d7d8 2981 pte = pmap_pte_quick(pmap, pv->pv_va);
5926987a 2982 KKASSERT(*pte);
c2fb025d 2983 pmap_inval_interlock(&info, pmap, pv->pv_va);
984263bc 2984
4a22e893
MD
2985 /*
2986 * We cannot remove wired pages from a process' mapping
2987 * at this time
2988 */
17cde63e 2989 if (*pte & PG_W) {
c2fb025d 2990 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2991 npv = TAILQ_NEXT(pv, pv_plist);
2992 continue;
2993 }
2247fe02 2994 KKASSERT(*pte);
17cde63e 2995 tpte = loadandclear(pte);
c2fb025d 2996 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2997
2998 m = PHYS_TO_VM_PAGE(tpte);
5926987a 2999 test_m_maps_pv(m, pv);
984263bc
MD
3000
3001 KASSERT(m < &vm_page_array[vm_page_array_size],
3002 ("pmap_remove_pages: bad tpte %x", tpte));
3003
eec2b734
MD
3004 KKASSERT(pmap->pm_stats.resident_count > 0);
3005 --pmap->pm_stats.resident_count;
984263bc
MD
3006
3007 /*
3008 * Update the vm_page_t clean and reference bits.
3009 */
3010 if (tpte & PG_M) {
3011 vm_page_dirty(m);
3012 }
3013
984263bc 3014 npv = TAILQ_NEXT(pv, pv_plist);
5926987a
MD
3015#ifdef PMAP_DEBUG
3016 KKASSERT(pv->pv_m == m);
3017 KKASSERT(pv->pv_pmap == pmap);
3018#endif
8790d7d8
MD
3019 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
3020 save_generation = ++pmap->pm_generation;
984263bc
MD
3021
3022 m->md.pv_list_count--;
cef01e15
MD
3023 if (m->object)
3024 atomic_add_int(&m->object->agg_pv_list_count, -1);
984263bc 3025 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
17cde63e 3026 if (TAILQ_EMPTY(&m->md.pv_list))
984263bc 3027 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
984263bc 3028
8790d7d8 3029 pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem, &info);
984263bc 3030 free_pv_entry(pv);
8790d7d8
MD
3031
3032 /*
3033 * Restart the scan if we blocked during the unuse or free
3034 * calls and other removals were made.
3035 */
3036 if (save_generation != pmap->pm_generation) {
3037 kprintf("Warning: pmap_remove_pages race-A avoided\n");
5926987a 3038 npv = TAILQ_FIRST(&pmap->pm_pvlist);
8790d7d8 3039 }
984263bc 3040 }
c2fb025d 3041 pmap_inval_done(&info);
4107b0c0 3042 lwkt_reltoken(&vm_token);
b12defdc
MD
3043 if (pmap->pm_pteobj)
3044 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
3045}
3046
3047/*
3048 * pmap_testbit tests bits in pte's
5e8d0349 3049 * note that the testbit/clearbit routines are inline,
984263bc 3050 * and a lot of things compile-time evaluate.
4107b0c0
MD
3051 *
3052 * The caller must hold vm_token.
984263bc
MD
3053 */
3054static boolean_t
840de426 3055pmap_testbit(vm_page_t m, int bit)
984263bc
MD
3056{
3057 pv_entry_t pv;
3058 unsigned *pte;
984263bc
MD
3059
3060 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3061 return FALSE;
3062
3063 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3064 return FALSE;
3065
984263bc
MD
3066 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3067 /*
3068 * if the bit being tested is the modified bit, then
3069 * mark clean_map and ptes as never
3070 * modified.
3071 */
3072 if (bit & (PG_A|PG_M)) {
3073 if (!pmap_track_modified(pv->pv_va))
3074 continue;
3075 }
3076
3077#if defined(PMAP_DIAGNOSTIC)
3078 if (!pv->pv_pmap) {
d557216f
MD
3079 kprintf("Null pmap (tb) at va: %p\n",
3080 (void *)pv->pv_va);
984263bc
MD
3081 continue;
3082 }
3083#endif
3084 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
74b9d1ec 3085 if (*pte & bit) {
984263bc 3086 return TRUE;
74b9d1ec 3087 }
984263bc 3088 }
984263bc
MD
3089 return (FALSE);
3090}
3091
3092/*
4107b0c0
MD
3093 * This routine is used to modify bits in ptes
3094 *
3095 * The caller must hold vm_token.
984263bc
MD
3096 */
3097static __inline void
5e8d0349 3098pmap_clearbit(vm_page_t m, int bit)
984263bc 3099{
0f7a3396 3100 struct pmap_inval_info info;
840de426
MD
3101 pv_entry_t pv;
3102 unsigned *pte;
5e8d0349 3103 unsigned pbits;
984263bc
MD
3104
3105 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3106 return;
3107
0f7a3396 3108 pmap_inval_init(&info);
984263bc
MD
3109
3110 /*
3111 * Loop over all current mappings setting/clearing as appropos If
3112 * setting RO do we need to clear the VAC?
3113 */
3114 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3115 /*
3116 * don't write protect pager mappings
3117 */
5e8d0349 3118 if (bit == PG_RW) {
984263bc
MD
3119 if (!pmap_track_modified(pv->pv_va))
3120 continue;
3121 }
3122
3123#if defined(PMAP_DIAGNOSTIC)
3124 if (!pv->pv_pmap) {
d557216f
MD
3125 kprintf("Null pmap (cb) at va: %p\n",
3126 (void *)pv->pv_va);
984263bc
MD
3127 continue;
3128 }
3129#endif
3130
0f7a3396
MD
3131 /*
3132 * Careful here. We can use a locked bus instruction to
3133 * clear PG_A or PG_M safely but we need to synchronize
3134 * with the target cpus when we mess with PG_RW.
70fc5283
MD
3135 *
3136 * We do not have to force synchronization when clearing
3137 * PG_M even for PTEs generated via virtual memory maps,
3138 * because the virtual kernel will invalidate the pmap
3139 * entry when/if it needs to resynchronize the Modify bit.
0f7a3396 3140 */
70fc5283 3141 if (bit & PG_RW)
c2fb025d 3142 pmap_inval_interlock(&info, pv->pv_pmap, pv->pv_va);
17cde63e
MD
3143 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3144again:
5e8d0349
MD
3145 pbits = *pte;
3146 if (pbits & bit) {
3147 if (bit == PG_RW) {
17cde63e 3148 if (pbits & PG_M) {
5e8d0349 3149 vm_page_dirty(m);
17cde63e
MD
3150 atomic_clear_int(pte, PG_M|PG_RW);
3151 } else {
3152 /*
3153 * The cpu may be trying to set PG_M
3154 * simultaniously with our clearing
3155 * of PG_RW.
3156 */
3157 if (!atomic_cmpset_int(pte, pbits,
3158 pbits & ~PG_RW))
3159 goto again;
3160 }
5e8d0349
MD
3161 } else if (bit == PG_M) {
3162 /*
70fc5283
MD
3163 * We could also clear PG_RW here to force
3164 * a fault on write to redetect PG_M for
3165 * virtual kernels, but it isn't necessary
3166 * since virtual kernels invalidate the pte
3167 * when they clear the VPTE_M bit in their
3168 * virtual page tables.
5e8d0349 3169 */
70fc5283 3170 atomic_clear_int(pte, PG_M);
5e8d0349
MD
3171 } else {
3172 atomic_clear_int(pte, bit);