kernel - Remove debugging from i386 pmap
[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
166/*
a93980ab
MD
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
MD
175
176/*
984263bc
MD
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
224/*
840de426
MD
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
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
267/*
4107b0c0
MD
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
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
405 /*
e731d345
MD
406 * ptvmmap is used for reading arbitrary physical pages via
407 * /dev/mem.
408 */
409 SYSMAP(caddr_t, ptmmap, ptvmmap, 1)
410
411 /*
984263bc
MD
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 */
429#ifdef SMP
430 pgeflag = 0;
431#else
432 if (cpu_feature & CPUID_PGE)
984263bc 433 pgeflag = PG_G;
a2a5ad0d 434#endif
984263bc
MD
435
436/*
437 * Initialize the 4MB page size flag
438 */
439 pseflag = 0;
440/*
441 * The 4MB page version of the initial
442 * kernel page mapping.
443 */
444 pdir4mb = 0;
445
446#if !defined(DISABLE_PSE)
447 if (cpu_feature & CPUID_PSE) {
448 unsigned ptditmp;
449 /*
450 * Note that we have enabled PSE mode
451 */
452 pseflag = PG_PS;
453 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE));
454 ptditmp &= ~(NBPDR - 1);
455 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag;
456 pdir4mb = ptditmp;
457
8a8d5d85
MD
458#ifndef SMP
459 /*
460 * Enable the PSE mode. If we are SMP we can't do this
461 * now because the APs will not be able to use it when
462 * they boot up.
463 */
464 load_cr4(rcr4() | CR4_PSE);
984263bc 465
8a8d5d85
MD
466 /*
467 * We can do the mapping here for the single processor
468 * case. We simply ignore the old page table page from
469 * now on.
470 */
471 /*
472 * For SMP, we still need 4K pages to bootstrap APs,
473 * PSE will be enabled as soon as all APs are up.
474 */
b5b32410 475 PTD[KPTDI] = (pd_entry_t)ptditmp;
fbbaeba3 476 kernel_pmap.pm_pdir[KPTDI] = (pd_entry_t)ptditmp;
0f7a3396 477 cpu_invltlb();
8a8d5d85 478#endif
984263bc
MD
479 }
480#endif
984263bc 481
81c04d07
MD
482 /*
483 * We need to finish setting up the globaldata page for the BSP.
484 * locore has already populated the page table for the mdglobaldata
485 * portion.
486 */
487 pg = MDGLOBALDATA_BASEALLOC_PAGES;
85100692 488 gd = &CPU_prvspace[0].mdglobaldata;
81c04d07
MD
489 gd->gd_CMAP1 = &SMPpt[pg + 0];
490 gd->gd_CMAP2 = &SMPpt[pg + 1];
491 gd->gd_CMAP3 = &SMPpt[pg + 2];
492 gd->gd_PMAP1 = &SMPpt[pg + 3];
9388fcaa 493 gd->gd_GDMAP1 = &PTD[APTDPTDI];
85100692
MD
494 gd->gd_CADDR1 = CPU_prvspace[0].CPAGE1;
495 gd->gd_CADDR2 = CPU_prvspace[0].CPAGE2;
496 gd->gd_CADDR3 = CPU_prvspace[0].CPAGE3;
497 gd->gd_PADDR1 = (unsigned *)CPU_prvspace[0].PPAGE1;
9388fcaa 498 gd->gd_GDADDR1= (unsigned *)VADDR(APTDPTDI, 0);
984263bc 499
0f7a3396 500 cpu_invltlb();
984263bc
MD
501}
502
503#ifdef SMP
504/*
505 * Set 4mb pdir for mp startup
506 */
507void
508pmap_set_opt(void)
509{
510 if (pseflag && (cpu_feature & CPUID_PSE)) {
511 load_cr4(rcr4() | CR4_PSE);
72740893 512 if (pdir4mb && mycpu->gd_cpuid == 0) { /* only on BSP */
fbbaeba3 513 kernel_pmap.pm_pdir[KPTDI] =
984263bc
MD
514 PTD[KPTDI] = (pd_entry_t)pdir4mb;
515 cpu_invltlb();
516 }
517 }
518}
519#endif
520
521/*
4107b0c0
MD
522 * Initialize the pmap module, called by vm_init()
523 *
524 * Called from the low level boot code only.
984263bc
MD
525 */
526void
e7252eda 527pmap_init(void)
984263bc
MD
528{
529 int i;
530 int initial_pvs;
531
532 /*
533 * object for kernel page table pages
534 */
535 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE);
536
537 /*
538 * Allocate memory for random pmap data structures. Includes the
539 * pv_head_table.
540 */
541
542 for(i = 0; i < vm_page_array_size; i++) {
543 vm_page_t m;
544
545 m = &vm_page_array[i];
546 TAILQ_INIT(&m->md.pv_list);
547 m->md.pv_list_count = 0;
548 }
549
550 /*
551 * init the pv free list
552 */
553 initial_pvs = vm_page_array_size;
554 if (initial_pvs < MINPV)
555 initial_pvs = MINPV;
556 pvzone = &pvzone_store;
948209ce
MD
557 pvinit = (void *)kmem_alloc(&kernel_map,
558 initial_pvs * sizeof (struct pv_entry));
559 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry),
560 pvinit, initial_pvs);
984263bc
MD
561
562 /*
563 * Now it is safe to enable pv_table recording.
564 */
565 pmap_initialized = TRUE;
566}
567
568/*
569 * Initialize the address space (zone) for the pv_entries. Set a
570 * high water mark so that the system can recover from excessive
571 * numbers of pv entries.
4107b0c0
MD
572 *
573 * Called from the low level boot code only.
984263bc
MD
574 */
575void
f123d5a1 576pmap_init2(void)
984263bc
MD
577{
578 int shpgperproc = PMAP_SHPGPERPROC;
948209ce 579 int entry_max;
984263bc
MD
580
581 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
582 pv_entry_max = shpgperproc * maxproc + vm_page_array_size;
948209ce
MD
583
584#ifdef PMAP_PVLIMIT
585 /*
586 * Horrible hack for systems with a lot of memory running i386.
587 * the calculated pv_entry_max can wind up eating a ton of KVM
588 * so put a cap on the number of entries if the user did not
589 * change any of the values. This saves about 44MB of KVM on
590 * boxes with 3+GB of ram.
591 *
592 * On the flip side, this makes it more likely that some setups
593 * will run out of pv entries. Those sysads will have to bump
594 * the limit up with vm.pamp.pv_entries or vm.pmap.shpgperproc.
595 */
596 if (shpgperproc == PMAP_SHPGPERPROC) {
597 if (pv_entry_max > PMAP_PVLIMIT)
598 pv_entry_max = PMAP_PVLIMIT;
599 }
600#endif
984263bc
MD
601 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
602 pv_entry_high_water = 9 * (pv_entry_max / 10);
948209ce
MD
603
604 /*
605 * Subtract out pages already installed in the zone (hack)
606 */
607 entry_max = pv_entry_max - vm_page_array_size;
608 if (entry_max <= 0)
609 entry_max = 1;
610
611 zinitna(pvzone, &pvzone_obj, NULL, 0, entry_max, ZONE_INTERRUPT, 1);
984263bc
MD
612}
613
614
615/***************************************************
616 * Low level helper routines.....
617 ***************************************************/
618
5926987a
MD
619#ifdef PMAP_DEBUG
620
621static void
622test_m_maps_pv(vm_page_t m, pv_entry_t pv)
623{
624 pv_entry_t spv;
625
74b9d1ec 626 crit_enter();
5926987a
MD
627#ifdef PMAP_DEBUG
628 KKASSERT(pv->pv_m == m);
629#endif
630 TAILQ_FOREACH(spv, &m->md.pv_list, pv_list) {
74b9d1ec
MD
631 if (pv == spv) {
632 crit_exit();
5926987a 633 return;
74b9d1ec 634 }
5926987a 635 }
74b9d1ec 636 crit_exit();
ed20d0e3 637 panic("test_m_maps_pv: failed m %p pv %p", m, pv);
5926987a
MD
638}
639
640static void
641ptbase_assert(struct pmap *pmap)
642{
643 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
644
645 /* are we current address space or kernel? */
4107b0c0 646 if (pmap == &kernel_pmap || frame == (((unsigned)PTDpde) & PG_FRAME))
5926987a 647 return;
3558dcda 648 KKASSERT(frame == (*mdcpu->gd_GDMAP1 & PG_FRAME));
5926987a
MD
649}
650
651#else
652
653#define test_m_maps_pv(m, pv)
654#define ptbase_assert(pmap)
655
656#endif
657
984263bc
MD
658#if defined(PMAP_DIAGNOSTIC)
659
660/*
661 * This code checks for non-writeable/modified pages.
662 * This should be an invalid condition.
663 */
664static int
665pmap_nw_modified(pt_entry_t ptea)
666{
667 int pte;
668
669 pte = (int) ptea;
670
671 if ((pte & (PG_M|PG_RW)) == PG_M)
672 return 1;
673 else
674 return 0;
675}
676#endif
677
678
679/*
4107b0c0
MD
680 * This routine defines the region(s) of memory that should not be tested
681 * for the modified bit.
682 *
683 * No requirements.
984263bc
MD
684 */
685static PMAP_INLINE int
686pmap_track_modified(vm_offset_t va)
687{
688 if ((va < clean_sva) || (va >= clean_eva))
689 return 1;
690 else
691 return 0;
692}
693
c1692ddf
MD
694/*
695 * Retrieve the mapped page table base for a particular pmap. Use our self
696 * mapping for the kernel_pmap or our current pmap.
697 *
698 * For foreign pmaps we use the per-cpu page table map. Since this involves
699 * installing a ptd it's actually (per-process x per-cpu). However, we
700 * still cannot depend on our mapping to survive thread switches because
701 * the process might be threaded and switching to another thread for the
702 * same process on the same cpu will allow that other thread to make its
703 * own mapping.
704 *
705 * This could be a bit confusing but the jist is for something like the
706 * vkernel which uses foreign pmaps all the time this represents a pretty
707 * good cache that avoids unnecessary invltlb()s.
4107b0c0
MD
708 *
709 * The caller must hold vm_token and the returned value is only valid
710 * until the caller blocks or releases the token.
c1692ddf 711 */
984263bc 712static unsigned *
e0e69b7d 713get_ptbase(pmap_t pmap)
984263bc
MD
714{
715 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME;
c1692ddf 716 struct mdglobaldata *gd = mdcpu;
984263bc 717
4107b0c0
MD
718 ASSERT_LWKT_TOKEN_HELD(&vm_token);
719
5926987a
MD
720 /*
721 * We can use PTmap if the pmap is our current address space or
722 * the kernel address space.
723 */
fbbaeba3 724 if (pmap == &kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) {
984263bc
MD
725 return (unsigned *) PTmap;
726 }
e0e69b7d 727
5926987a 728 /*
c1692ddf
MD
729 * Otherwise we use the per-cpu alternative page table map. Each
730 * cpu gets its own map. Because of this we cannot use this map
731 * from interrupts or threads which can preempt.
be3aecf7
MD
732 *
733 * Even if we already have the map cached we may still have to
734 * invalidate the TLB if another cpu modified a PDE in the map.
5926987a 735 */
c1692ddf
MD
736 KKASSERT(gd->mi.gd_intr_nesting_level == 0 &&
737 (gd->mi.gd_curthread->td_flags & TDF_INTTHREAD) == 0);
e0e69b7d 738
c1692ddf
MD
739 if ((*gd->gd_GDMAP1 & PG_FRAME) != frame) {
740 *gd->gd_GDMAP1 = frame | PG_RW | PG_V;
be3aecf7
MD
741 pmap->pm_cached |= gd->mi.gd_cpumask;
742 cpu_invltlb();
743 } else if ((pmap->pm_cached & gd->mi.gd_cpumask) == 0) {
744 pmap->pm_cached |= gd->mi.gd_cpumask;
984263bc 745 cpu_invltlb();
a93980ab
MD
746 } else if (dreadful_invltlb) {
747 cpu_invltlb();
984263bc 748 }
c1692ddf 749 return ((unsigned *)gd->gd_GDADDR1);
984263bc
MD
750}
751
752/*
e0e69b7d
MD
753 * pmap_extract:
754 *
4107b0c0 755 * Extract the physical page address associated with the map/VA pair.
e0e69b7d 756 *
4107b0c0 757 * The caller may hold vm_token if it desires non-blocking operation.
984263bc 758 */
6ef943a3 759vm_paddr_t
840de426 760pmap_extract(pmap_t pmap, vm_offset_t va)
984263bc
MD
761{
762 vm_offset_t rtval;
763 vm_offset_t pdirindex;
840de426 764
4107b0c0 765 lwkt_gettoken(&vm_token);
984263bc
MD
766 pdirindex = va >> PDRSHIFT;
767 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) {
768 unsigned *pte;
769 if ((rtval & PG_PS) != 0) {
770 rtval &= ~(NBPDR - 1);
771 rtval |= va & (NBPDR - 1);
4107b0c0
MD
772 } else {
773 pte = get_ptbase(pmap) + i386_btop(va);
774 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK));
984263bc 775 }
4107b0c0
MD
776 } else {
777 rtval = 0;
984263bc 778 }
4107b0c0
MD
779 lwkt_reltoken(&vm_token);
780 return rtval;
f6bf3af1
MD
781}
782
984263bc
MD
783/***************************************************
784 * Low level mapping routines.....
785 ***************************************************/
786
787/*
4107b0c0
MD
788 * Map a wired VM page to a KVA, fully SMP synchronized.
789 *
790 * No requirements, non blocking.
984263bc 791 */
24712b90 792void
6ef943a3 793pmap_kenter(vm_offset_t va, vm_paddr_t pa)
984263bc 794{
840de426 795 unsigned *pte;
0f7a3396
MD
796 unsigned npte;
797 pmap_inval_info info;
984263bc 798
0f7a3396 799 pmap_inval_init(&info);
984263bc
MD
800 npte = pa | PG_RW | PG_V | pgeflag;
801 pte = (unsigned *)vtopte(va);
c2fb025d 802 pmap_inval_interlock(&info, &kernel_pmap, va);
984263bc 803 *pte = npte;
c2fb025d
MD
804 pmap_inval_deinterlock(&info, &kernel_pmap);
805 pmap_inval_done(&info);
984263bc
MD
806}
807
6d1ec6fa 808/*
4107b0c0
MD
809 * Map a wired VM page to a KVA, synchronized on current cpu only.
810 *
811 * No requirements, non blocking.
6d1ec6fa 812 */
24712b90
MD
813void
814pmap_kenter_quick(vm_offset_t va, vm_paddr_t pa)
815{
816 unsigned *pte;
817 unsigned npte;
818
819 npte = pa | PG_RW | PG_V | pgeflag;
820 pte = (unsigned *)vtopte(va);
821 *pte = npte;
822 cpu_invlpg((void *)va);
823}
824
4107b0c0
MD
825/*
826 * Synchronize a previously entered VA on all cpus.
827 *
828 * No requirements, non blocking.
829 */
24712b90
MD
830void
831pmap_kenter_sync(vm_offset_t va)
832{
833 pmap_inval_info info;
834
835 pmap_inval_init(&info);
c2fb025d
MD
836 pmap_inval_interlock(&info, &kernel_pmap, va);
837 pmap_inval_deinterlock(&info, &kernel_pmap);
838 pmap_inval_done(&info);
24712b90
MD
839}
840
4107b0c0
MD
841/*
842 * Synchronize a previously entered VA on the current cpu only.
843 *
844 * No requirements, non blocking.
845 */
24712b90
MD
846void
847pmap_kenter_sync_quick(vm_offset_t va)
848{
849 cpu_invlpg((void *)va);
850}
851
984263bc 852/*
4107b0c0
MD
853 * Remove a page from the kernel pagetables, fully SMP synchronized.
854 *
855 * No requirements, non blocking.
984263bc 856 */
24712b90 857void
840de426 858pmap_kremove(vm_offset_t va)
984263bc 859{
840de426 860 unsigned *pte;
0f7a3396 861 pmap_inval_info info;
984263bc 862
0f7a3396 863 pmap_inval_init(&info);
984263bc 864 pte = (unsigned *)vtopte(va);
c2fb025d 865 pmap_inval_interlock(&info, &kernel_pmap, va);
984263bc 866 *pte = 0;
c2fb025d
MD
867 pmap_inval_deinterlock(&info, &kernel_pmap);
868 pmap_inval_done(&info);
984263bc
MD
869}
870
4107b0c0
MD
871/*
872 * Remove a page from the kernel pagetables, synchronized on current cpu only.
873 *
874 * No requirements, non blocking.
875 */
24712b90
MD
876void
877pmap_kremove_quick(vm_offset_t va)
878{
879 unsigned *pte;
880 pte = (unsigned *)vtopte(va);
881 *pte = 0;
882 cpu_invlpg((void *)va);
883}
884
984263bc 885/*
4107b0c0
MD
886 * Adjust the permissions of a page in the kernel page table,
887 * synchronized on the current cpu only.
888 *
889 * No requirements, non blocking.
9ad680a3
MD
890 */
891void
892pmap_kmodify_rw(vm_offset_t va)
893{
4107b0c0 894 atomic_set_int(vtopte(va), PG_RW);
9ad680a3
MD
895 cpu_invlpg((void *)va);
896}
897
4107b0c0
MD
898/*
899 * Adjust the permissions of a page in the kernel page table,
900 * synchronized on the current cpu only.
901 *
902 * No requirements, non blocking.
903 */
9ad680a3
MD
904void
905pmap_kmodify_nc(vm_offset_t va)
906{
4107b0c0 907 atomic_set_int(vtopte(va), PG_N);
9ad680a3
MD
908 cpu_invlpg((void *)va);
909}
910
911/*
4107b0c0 912 * Map a range of physical addresses into kernel virtual address space.
984263bc 913 *
4107b0c0 914 * No requirements, non blocking.
984263bc
MD
915 */
916vm_offset_t
8e5e6f1b 917pmap_map(vm_offset_t *virtp, vm_paddr_t start, vm_paddr_t end, int prot)
984263bc 918{
8e5e6f1b
AH
919 vm_offset_t sva, virt;
920
921 sva = virt = *virtp;
984263bc
MD
922 while (start < end) {
923 pmap_kenter(virt, start);
924 virt += PAGE_SIZE;
925 start += PAGE_SIZE;
926 }
8e5e6f1b
AH
927 *virtp = virt;
928 return (sva);
984263bc
MD
929}
930
984263bc 931/*
4107b0c0
MD
932 * Add a list of wired pages to the kva, fully SMP synchronized.
933 *
934 * No requirements, non blocking.
984263bc
MD
935 */
936void
840de426 937pmap_qenter(vm_offset_t va, vm_page_t *m, int count)
984263bc
MD
938{
939 vm_offset_t end_va;
940
941 end_va = va + count * PAGE_SIZE;
942
943 while (va < end_va) {
944 unsigned *pte;
945
946 pte = (unsigned *)vtopte(va);
947 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag;
984263bc 948 cpu_invlpg((void *)va);
984263bc
MD
949 va += PAGE_SIZE;
950 m++;
951 }
952#ifdef SMP
0f7a3396 953 smp_invltlb(); /* XXX */
984263bc
MD
954#endif
955}
956
957/*
4107b0c0 958 * Remove pages from KVA, fully SMP synchronized.
7155fc7d 959 *
4107b0c0 960 * No requirements, non blocking.
984263bc
MD
961 */
962void
840de426 963pmap_qremove(vm_offset_t va, int count)
984263bc
MD
964{
965 vm_offset_t end_va;
966
967 end_va = va + count*PAGE_SIZE;
968
969 while (va < end_va) {
970 unsigned *pte;
971
972 pte = (unsigned *)vtopte(va);
973 *pte = 0;
984263bc 974 cpu_invlpg((void *)va);
984263bc
MD
975 va += PAGE_SIZE;
976 }
977#ifdef SMP
978 smp_invltlb();
979#endif
980}
981
06ecca5a
MD
982/*
983 * This routine works like vm_page_lookup() but also blocks as long as the
984 * page is busy. This routine does not busy the page it returns.
985 *
b12defdc 986 * The caller must hold the object.
06ecca5a 987 */
984263bc 988static vm_page_t
840de426 989pmap_page_lookup(vm_object_t object, vm_pindex_t pindex)
984263bc
MD
990{
991 vm_page_t m;
06ecca5a 992
b12defdc
MD
993 ASSERT_LWKT_TOKEN_HELD(vm_object_token(object));
994 m = vm_page_lookup_busy_wait(object, pindex, FALSE, "pplookp");
17cde63e 995
06ecca5a 996 return(m);
984263bc
MD
997}
998
999/*
263e4574 1000 * Create a new thread and optionally associate it with a (new) process.
6ef943a3 1001 * NOTE! the new thread's cpu may not equal the current cpu.
263e4574 1002 */
7d0bac62
MD
1003void
1004pmap_init_thread(thread_t td)
263e4574 1005{
f470d0c8 1006 /* enforce pcb placement */
f470d0c8 1007 td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_size) - 1;
65d6ce10 1008 td->td_savefpu = &td->td_pcb->pcb_save;
7d0bac62 1009 td->td_sp = (char *)td->td_pcb - 16;
263e4574
MD
1010}
1011
1012/*
984263bc
MD
1013 * This routine directly affects the fork perf for a process.
1014 */
1015void
13d13d89 1016pmap_init_proc(struct proc *p)
984263bc 1017{
984263bc
MD
1018}
1019
984263bc
MD
1020/***************************************************
1021 * Page table page management routines.....
1022 ***************************************************/
1023
1024/*
90244566
MD
1025 * This routine unwires page table pages, removing and freeing the page
1026 * tale page when the wire count drops to 0.
4107b0c0
MD
1027 *
1028 * The caller must hold vm_token.
1029 * This function can block.
984263bc
MD
1030 */
1031static int
90244566 1032_pmap_unwire_pte(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
840de426 1033{
17cde63e
MD
1034 /*
1035 * Wait until we can busy the page ourselves. We cannot have
1036 * any active flushes if we block.
1037 */
b12defdc 1038 vm_page_busy_wait(m, FALSE, "pmuwpt");
eec2b734 1039 KASSERT(m->queue == PQ_NONE,
90244566 1040 ("_pmap_unwire_pte: %p->queue != PQ_NONE", m));
984263bc 1041
90244566 1042 if (m->wire_count == 1) {
984263bc 1043 /*
be3aecf7
MD
1044 * Unmap the page table page.
1045 *
1046 * NOTE: We must clear pm_cached for all cpus, including
1047 * the current one, when clearing a page directory
1048 * entry.
984263bc 1049 */
c2fb025d 1050 pmap_inval_interlock(info, pmap, -1);
2247fe02 1051 KKASSERT(pmap->pm_pdir[m->pindex]);
984263bc 1052 pmap->pm_pdir[m->pindex] = 0;
be3aecf7 1053 pmap->pm_cached = 0;
c2fb025d 1054 pmap_inval_deinterlock(info, pmap);
eec2b734
MD
1055
1056 KKASSERT(pmap->pm_stats.resident_count > 0);
984263bc 1057 --pmap->pm_stats.resident_count;
984263bc
MD
1058
1059 if (pmap->pm_ptphint == m)
1060 pmap->pm_ptphint = NULL;
1061
1062 /*
eec2b734
MD
1063 * This was our last hold, the page had better be unwired
1064 * after we decrement wire_count.
1065 *
1066 * FUTURE NOTE: shared page directory page could result in
1067 * multiple wire counts.
984263bc 1068 */
90244566 1069 vm_page_unwire(m, 0);
17cde63e 1070 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
eec2b734
MD
1071 vm_page_flash(m);
1072 vm_page_free_zero(m);
984263bc 1073 return 1;
17cde63e 1074 } else {
90244566
MD
1075 KKASSERT(m->wire_count > 1);
1076 if (vm_page_unwire_quick(m))
1077 panic("pmap_unwire_pte: Insufficient wire_count");
b12defdc 1078 vm_page_wakeup(m);
17cde63e 1079 return 0;
984263bc 1080 }
984263bc
MD
1081}
1082
4107b0c0
MD
1083/*
1084 * The caller must hold vm_token.
92ba8d28 1085 *
4107b0c0 1086 * This function can block.
92ba8d28
MD
1087 *
1088 * This function can race the wire_count 2->1 case because the page
1089 * is not busied during the unwire_quick operation. An eventual
1090 * pmap_release() will catch the case.
4107b0c0 1091 */
984263bc 1092static PMAP_INLINE int
90244566 1093pmap_unwire_pte(pmap_t pmap, vm_page_t m, pmap_inval_info_t info)
984263bc 1094{
90244566
MD
1095 KKASSERT(m->wire_count > 0);
1096 if (m->wire_count > 1) {
1097 if (vm_page_unwire_quick(m))
1098 panic("pmap_unwire_pte: Insufficient wire_count");
984263bc 1099 return 0;
eec2b734 1100 } else {
90244566 1101 return _pmap_unwire_pte(pmap, m, info);
eec2b734 1102 }
984263bc
MD
1103}
1104
1105/*
4107b0c0 1106 * After removing a (user) page table entry, this routine is used to
984263bc 1107 * conditionally free the page, and manage the hold/wire counts.
5926987a 1108 *
4107b0c0
MD
1109 * The caller must hold vm_token.
1110 * This function can block regardless.
984263bc 1111 */
554cf9ac 1112static void
0f7a3396 1113pmap_unuse_pt(pmap_t pmap, vm_offset_t va, vm_page_t mpte,
4107b0c0 1114 pmap_inval_info_t info)
984263bc
MD
1115{
1116 unsigned ptepindex;
4107b0c0 1117
b12defdc
MD
1118 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1119
984263bc 1120 if (va >= UPT_MIN_ADDRESS)
554cf9ac 1121 return;
984263bc
MD
1122
1123 if (mpte == NULL) {
1124 ptepindex = (va >> PDRSHIFT);
b1482674
MD
1125 if ((mpte = pmap->pm_ptphint) != NULL &&
1126 mpte->pindex == ptepindex &&
1127 (mpte->flags & PG_BUSY) == 0) {
1128 ; /* use mpte */
984263bc 1129 } else {
b12defdc 1130 mpte = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
984263bc 1131 pmap->pm_ptphint = mpte;
b12defdc 1132 vm_page_wakeup(mpte);
984263bc
MD
1133 }
1134 }
554cf9ac 1135 pmap_unwire_pte(pmap, mpte, info);
984263bc
MD
1136}
1137
54a764e8 1138/*
fbbaeba3
MD
1139 * Initialize pmap0/vmspace0. This pmap is not added to pmap_list because
1140 * it, and IdlePTD, represents the template used to update all other pmaps.
1141 *
1142 * On architectures where the kernel pmap is not integrated into the user
1143 * process pmap, this pmap represents the process pmap, not the kernel pmap.
1144 * kernel_pmap should be used to directly access the kernel_pmap.
4107b0c0
MD
1145 *
1146 * No requirements.
54a764e8 1147 */
984263bc 1148void
840de426 1149pmap_pinit0(struct pmap *pmap)
984263bc
MD
1150{
1151 pmap->pm_pdir =
e4846942 1152 (pd_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
24712b90 1153 pmap_kenter((vm_offset_t)pmap->pm_pdir, (vm_offset_t) IdlePTD);
984263bc
MD
1154 pmap->pm_count = 1;
1155 pmap->pm_active = 0;
be3aecf7 1156 pmap->pm_cached = 0;
984263bc
MD
1157 pmap->pm_ptphint = NULL;
1158 TAILQ_INIT(&pmap->pm_pvlist);
b12defdc
MD
1159 TAILQ_INIT(&pmap->pm_pvlist_free);
1160 spin_init(&pmap->pm_spin);
1161 lwkt_token_init(&pmap->pm_token, "pmap_tok");
984263bc
MD
1162 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
1163}
1164
1165/*
1166 * Initialize a preallocated and zeroed pmap structure,
1167 * such as one in a vmspace structure.
4107b0c0
MD
1168 *
1169 * No requirements.
984263bc
MD
1170 */
1171void
840de426 1172pmap_pinit(struct pmap *pmap)
984263bc
MD
1173{
1174 vm_page_t ptdpg;
1175
1176 /*
1177 * No need to allocate page table space yet but we do need a valid
1178 * page directory table.
1179 */
b5b32410 1180 if (pmap->pm_pdir == NULL) {
984263bc 1181 pmap->pm_pdir =
e4846942 1182 (pd_entry_t *)kmem_alloc_pageable(&kernel_map, PAGE_SIZE);
b5b32410 1183 }
984263bc
MD
1184
1185 /*
c3834cb2 1186 * Allocate an object for the ptes
984263bc
MD
1187 */
1188 if (pmap->pm_pteobj == NULL)
c3834cb2 1189 pmap->pm_pteobj = vm_object_allocate(OBJT_DEFAULT, PTDPTDI + 1);
984263bc
MD
1190
1191 /*
c3834cb2
MD
1192 * Allocate the page directory page, unless we already have
1193 * one cached. If we used the cached page the wire_count will
1194 * already be set appropriately.
984263bc 1195 */
c3834cb2
MD
1196 if ((ptdpg = pmap->pm_pdirm) == NULL) {
1197 ptdpg = vm_page_grab(pmap->pm_pteobj, PTDPTDI,
d2d8515b
MD
1198 VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
1199 VM_ALLOC_ZERO);
c3834cb2 1200 pmap->pm_pdirm = ptdpg;
b12defdc
MD
1201 vm_page_flag_clear(ptdpg, PG_MAPPED);
1202 vm_page_wire(ptdpg);
d2d8515b 1203 KKASSERT(ptdpg->valid == VM_PAGE_BITS_ALL);
c3834cb2 1204 pmap_kenter((vm_offset_t)pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg));
b12defdc 1205 vm_page_wakeup(ptdpg);
c3834cb2 1206 }
984263bc 1207 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI];
984263bc
MD
1208
1209 /* install self-referential address mapping entry */
1210 *(unsigned *) (pmap->pm_pdir + PTDPTDI) =
1211 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M;
1212
1213 pmap->pm_count = 1;
1214 pmap->pm_active = 0;
be3aecf7 1215 pmap->pm_cached = 0;
984263bc
MD
1216 pmap->pm_ptphint = NULL;
1217 TAILQ_INIT(&pmap->pm_pvlist);
b12defdc
MD
1218 TAILQ_INIT(&pmap->pm_pvlist_free);
1219 spin_init(&pmap->pm_spin);
1220 lwkt_token_init(&pmap->pm_token, "pmap_tok");
984263bc 1221 bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
eec2b734 1222 pmap->pm_stats.resident_count = 1;
984263bc
MD
1223}
1224
1225/*
c3834cb2
MD
1226 * Clean up a pmap structure so it can be physically freed. This routine
1227 * is called by the vmspace dtor function. A great deal of pmap data is
1228 * left passively mapped to improve vmspace management so we have a bit
1229 * of cleanup work to do here.
4107b0c0
MD
1230 *
1231 * No requirements.
e3161323
MD
1232 */
1233void
1234pmap_puninit(pmap_t pmap)
1235{
c3834cb2
MD
1236 vm_page_t p;
1237
3321ee05 1238 pmap_wait(pmap, -1);
e3161323 1239 KKASSERT(pmap->pm_active == 0);
c3834cb2
MD
1240 if ((p = pmap->pm_pdirm) != NULL) {
1241 KKASSERT(pmap->pm_pdir != NULL);
1242 pmap_kremove((vm_offset_t)pmap->pm_pdir);
b12defdc 1243 vm_page_busy_wait(p, FALSE, "pgpun");
90244566 1244 vm_page_unwire(p, 0);
c3834cb2
MD
1245 vm_page_free_zero(p);
1246 pmap->pm_pdirm = NULL;
1247 }
e3161323
MD
1248 if (pmap->pm_pdir) {
1249 kmem_free(&kernel_map, (vm_offset_t)pmap->pm_pdir, PAGE_SIZE);
1250 pmap->pm_pdir = NULL;
1251 }
1252 if (pmap->pm_pteobj) {
1253 vm_object_deallocate(pmap->pm_pteobj);
1254 pmap->pm_pteobj = NULL;
1255 }
1256}
1257
1258/*
984263bc
MD
1259 * Wire in kernel global address entries. To avoid a race condition
1260 * between pmap initialization and pmap_growkernel, this procedure
54a764e8
MD
1261 * adds the pmap to the master list (which growkernel scans to update),
1262 * then copies the template.
4107b0c0
MD
1263 *
1264 * No requirements.
984263bc
MD
1265 */
1266void
840de426 1267pmap_pinit2(struct pmap *pmap)
984263bc 1268{
b12defdc
MD
1269 /*
1270 * XXX copies current process, does not fill in MPPTDI
1271 */
1272 spin_lock(&pmap_spin);
54a764e8 1273 TAILQ_INSERT_TAIL(&pmap_list, pmap, pm_pmnode);
984263bc 1274 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE);
b12defdc 1275 spin_unlock(&pmap_spin);
984263bc
MD
1276}
1277
344ad853 1278/*
eec2b734 1279 * Attempt to release and free a vm_page in a pmap. Returns 1 on success,
344ad853 1280 * 0 on failure (if the procedure had to sleep).
c3834cb2
MD
1281 *
1282 * When asked to remove the page directory page itself, we actually just
1283 * leave it cached so we do not have to incur the SMP inval overhead of
1284 * removing the kernel mapping. pmap_puninit() will take care of it.
4107b0c0
MD
1285 *
1286 * The caller must hold vm_token.
1287 * This function can block regardless.
344ad853 1288 */
984263bc 1289static int
840de426 1290pmap_release_free_page(struct pmap *pmap, vm_page_t p)
984263bc
MD
1291{
1292 unsigned *pde = (unsigned *) pmap->pm_pdir;
4107b0c0 1293
984263bc
MD
1294 /*
1295 * This code optimizes the case of freeing non-busy
1296 * page-table pages. Those pages are zero now, and
1297 * might as well be placed directly into the zero queue.
1298 */
b12defdc
MD
1299 if (vm_page_busy_try(p, FALSE)) {
1300 vm_page_sleep_busy(p, FALSE, "pmaprl");
984263bc 1301 return 0;
b12defdc 1302 }
984263bc 1303
eec2b734 1304 KKASSERT(pmap->pm_stats.resident_count > 0);
2247fe02 1305 KKASSERT(pde[p->pindex]);
984263bc 1306
b1482674
MD
1307 /*
1308 * page table page's wire_count must be 1. Caller is the pmap
1309 * termination code which holds the pm_pteobj, there is a race
1310 * if someone else is trying to hold the VM object in order to
1311 * clean up a wire_count.
1312 */
90244566 1313 if (p->wire_count != 1) {
b1482674
MD
1314 if (pmap->pm_pteobj->hold_count <= 1)
1315 panic("pmap_release: freeing wired page table page");
1316 kprintf("pmap_release_free_page: unwire race detected\n");
1317 vm_page_wakeup(p);
1318 tsleep(p, 0, "pmapx", 1);
1319 return 0;
984263bc 1320 }
b1482674
MD
1321
1322 /*
1323 * Remove the page table page from the processes address space.
1324 */
1325 pmap->pm_cached = 0;
1326 pde[p->pindex] = 0;
1327 --pmap->pm_stats.resident_count;
c3834cb2
MD
1328 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex))
1329 pmap->pm_ptphint = NULL;
1330
984263bc 1331 /*
c3834cb2
MD
1332 * We leave the page directory page cached, wired, and mapped in
1333 * the pmap until the dtor function (pmap_puninit()) gets called.
1334 * However, still clean it up so we can set PG_ZERO.
c1692ddf
MD
1335 *
1336 * The pmap has already been removed from the pmap_list in the
1337 * PTDPTDI case.
984263bc
MD
1338 */
1339 if (p->pindex == PTDPTDI) {
1340 bzero(pde + KPTDI, nkpt * PTESIZE);
9388fcaa 1341 bzero(pde + MPPTDI, (NPDEPG - MPPTDI) * PTESIZE);
c3834cb2
MD
1342 vm_page_flag_set(p, PG_ZERO);
1343 vm_page_wakeup(p);
1344 } else {
92ba8d28
MD
1345 /*
1346 * This case can occur if a pmap_unwire_pte() loses a race
1347 * while the page is unbusied.
1348 */
1349 /*panic("pmap_release: page should already be gone %p", p);*/
1350 vm_page_flag_clear(p, PG_MAPPED);
90244566 1351 vm_page_unwire(p, 0);
c3834cb2 1352 vm_page_free_zero(p);
984263bc 1353 }
984263bc
MD
1354 return 1;
1355}
1356
1357/*
4107b0c0
MD
1358 * This routine is called if the page table page is not mapped correctly.
1359 *
1360 * The caller must hold vm_token.
984263bc
MD
1361 */
1362static vm_page_t
840de426 1363_pmap_allocpte(pmap_t pmap, unsigned ptepindex)
984263bc 1364{
480c83b6 1365 vm_offset_t ptepa;
984263bc
MD
1366 vm_page_t m;
1367
1368 /*
d2d8515b
MD
1369 * Find or fabricate a new pagetable page. Setting VM_ALLOC_ZERO
1370 * will zero any new page and mark it valid.
984263bc
MD
1371 */
1372 m = vm_page_grab(pmap->pm_pteobj, ptepindex,
d2d8515b 1373 VM_ALLOC_NORMAL | VM_ALLOC_ZERO | VM_ALLOC_RETRY);
984263bc
MD
1374
1375 KASSERT(m->queue == PQ_NONE,
1376 ("_pmap_allocpte: %p->queue != PQ_NONE", m));
1377
eec2b734 1378 /*
90244566 1379 * Increment the wire count for the page we will be returning to
eec2b734
MD
1380 * the caller.
1381 */
90244566 1382 vm_page_wire(m);
eec2b734
MD
1383
1384 /*
1385 * It is possible that someone else got in and mapped by the page
1386 * directory page while we were blocked, if so just unbusy and
90244566 1387 * return the wired page.
eec2b734
MD
1388 */
1389 if ((ptepa = pmap->pm_pdir[ptepindex]) != 0) {
1390 KKASSERT((ptepa & PG_FRAME) == VM_PAGE_TO_PHYS(m));
1391 vm_page_wakeup(m);
1392 return(m);
1393 }
1394
984263bc
MD
1395 /*
1396 * Map the pagetable page into the process address space, if
1397 * it isn't already there.
be3aecf7
MD
1398 *
1399 * NOTE: For safety clear pm_cached for all cpus including the
1400 * current one when adding a PDE to the map.
984263bc 1401 */
eec2b734 1402 ++pmap->pm_stats.resident_count;
984263bc
MD
1403
1404 ptepa = VM_PAGE_TO_PHYS(m);
1405 pmap->pm_pdir[ptepindex] =
1406 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M);
be3aecf7 1407 pmap->pm_cached = 0;
984263bc
MD
1408
1409 /*
1410 * Set the page table hint
1411 */
1412 pmap->pm_ptphint = m;
984263bc
MD
1413 vm_page_flag_set(m, PG_MAPPED);
1414 vm_page_wakeup(m);
1415
1416 return m;
1417}
1418
4107b0c0
MD
1419/*
1420 * Allocate a page table entry for a va.
1421 *
1422 * The caller must hold vm_token.
1423 */
984263bc 1424static vm_page_t
840de426 1425pmap_allocpte(pmap_t pmap, vm_offset_t va)
984263bc
MD
1426{
1427 unsigned ptepindex;
1428 vm_offset_t ptepa;
b1482674 1429 vm_page_t mpte;
984263bc 1430
b12defdc
MD
1431 ASSERT_LWKT_TOKEN_HELD(vm_object_token(pmap->pm_pteobj));
1432
984263bc
MD
1433 /*
1434 * Calculate pagetable page index
1435 */
1436 ptepindex = va >> PDRSHIFT;
1437
1438 /*
1439 * Get the page directory entry
1440 */
1441 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
1442
1443 /*
1444 * This supports switching from a 4MB page to a
1445 * normal 4K page.
1446 */
1447 if (ptepa & PG_PS) {
1448 pmap->pm_pdir[ptepindex] = 0;
1449 ptepa = 0;
0f7a3396 1450 smp_invltlb();
54341a3b 1451 cpu_invltlb();
984263bc
MD
1452 }
1453
1454 /*
1455 * If the page table page is mapped, we just increment the
90244566 1456 * wire count, and activate it.
984263bc
MD
1457 */
1458 if (ptepa) {
1459 /*
1460 * In order to get the page table page, try the
1461 * hint first.
1462 */
b1482674
MD
1463 if ((mpte = pmap->pm_ptphint) != NULL &&
1464 (mpte->pindex == ptepindex) &&
1465 (mpte->flags & PG_BUSY) == 0) {
1466 vm_page_wire_quick(mpte);
984263bc 1467 } else {
b1482674
MD
1468 mpte = pmap_page_lookup(pmap->pm_pteobj, ptepindex);
1469 pmap->pm_ptphint = mpte;
1470 vm_page_wire_quick(mpte);
1471 vm_page_wakeup(mpte);
984263bc 1472 }
b1482674 1473 return mpte;
984263bc
MD
1474 }
1475 /*
1476 * Here if the pte page isn't mapped, or if it has been deallocated.
1477 */
1478 return _pmap_allocpte(pmap, ptepindex);
1479}
1480
1481
1482/***************************************************
1f804340 1483 * Pmap allocation/deallocation routines.
984263bc
MD
1484 ***************************************************/
1485
1486/*
1487 * Release any resources held by the given physical map.
1488 * Called when a pmap initialized by pmap_pinit is being released.
1489 * Should only be called if the map contains no valid mappings.
4107b0c0 1490 *
b12defdc 1491 * Caller must hold pmap->pm_token
984263bc 1492 */
1f804340
MD
1493static int pmap_release_callback(struct vm_page *p, void *data);
1494
984263bc 1495void
840de426 1496pmap_release(struct pmap *pmap)
984263bc 1497{
984263bc 1498 vm_object_t object = pmap->pm_pteobj;
1f804340 1499 struct rb_vm_page_scan_info info;
984263bc 1500
4107b0c0
MD
1501 KASSERT(pmap->pm_active == 0,
1502 ("pmap still active! %08x", pmap->pm_active));
984263bc
MD
1503#if defined(DIAGNOSTIC)
1504 if (object->ref_count != 1)
1505 panic("pmap_release: pteobj reference count != 1");
1506#endif
1507
1f804340
MD
1508 info.pmap = pmap;
1509 info.object = object;
b12defdc
MD
1510
1511 spin_lock(&pmap_spin);
54a764e8 1512 TAILQ_REMOVE(&pmap_list, pmap, pm_pmnode);
b12defdc 1513 spin_unlock(&pmap_spin);
1f804340 1514
b12defdc 1515 vm_object_hold(object);
2160c64a 1516 /*lwkt_gettoken(&vm_token);*/
1f804340 1517 do {
1f804340
MD
1518 info.error = 0;
1519 info.mpte = NULL;
1520 info.limit = object->generation;
1521
1522 vm_page_rb_tree_RB_SCAN(&object->rb_memq, NULL,
1523 pmap_release_callback, &info);
1524 if (info.error == 0 && info.mpte) {
1525 if (!pmap_release_free_page(pmap, info.mpte))
1526 info.error = 1;
984263bc 1527 }
1f804340 1528 } while (info.error);
2160c64a 1529 /*lwkt_reltoken(&vm_token);*/
2f2d9e58 1530 vm_object_drop(object);
b12defdc
MD
1531
1532 pmap->pm_cached = 0;
1f804340
MD
1533}
1534
4107b0c0
MD
1535/*
1536 * The caller must hold vm_token.
1537 */
1f804340
MD
1538static int
1539pmap_release_callback(struct vm_page *p, void *data)
1540{
1541 struct rb_vm_page_scan_info *info = data;
1542
1543 if (p->pindex == PTDPTDI) {
1544 info->mpte = p;
1545 return(0);
344ad853 1546 }
1f804340
MD
1547 if (!pmap_release_free_page(info->pmap, p)) {
1548 info->error = 1;
1549 return(-1);
1550 }
1551 if (info->object->generation != info->limit) {
1552 info->error = 1;
1553 return(-1);
1554 }
1555 return(0);
984263bc 1556}
984263bc
MD
1557
1558/*
0e5797fe 1559 * Grow the number of kernel page table entries, if needed.
4107b0c0
MD
1560 *
1561 * No requirements.
984263bc
MD
1562 */
1563void
a8cf2878 1564pmap_growkernel(vm_offset_t kstart, vm_offset_t kend)
984263bc 1565{
a8cf2878 1566 vm_offset_t addr = kend;
54a764e8 1567 struct pmap *pmap;
984263bc
MD
1568 vm_offset_t ptppaddr;
1569 vm_page_t nkpg;
1570 pd_entry_t newpdir;
1571
b12defdc 1572 vm_object_hold(kptobj);
984263bc
MD
1573 if (kernel_vm_end == 0) {
1574 kernel_vm_end = KERNBASE;
1575 nkpt = 0;
1576 while (pdir_pde(PTD, kernel_vm_end)) {
4107b0c0
MD
1577 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1578 ~(PAGE_SIZE * NPTEPG - 1);
984263bc
MD
1579 nkpt++;
1580 }
1581 }
1582 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1);
1583 while (kernel_vm_end < addr) {
1584 if (pdir_pde(PTD, kernel_vm_end)) {
4107b0c0
MD
1585 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1586 ~(PAGE_SIZE * NPTEPG - 1);
984263bc
MD
1587 continue;
1588 }
1589
1590 /*
1591 * This index is bogus, but out of the way
1592 */
4107b0c0
MD
1593 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_NORMAL |
1594 VM_ALLOC_SYSTEM |
1595 VM_ALLOC_INTERRUPT);
dc1fd4b3 1596 if (nkpg == NULL)
984263bc
MD
1597 panic("pmap_growkernel: no memory to grow kernel");
1598
984263bc
MD
1599 vm_page_wire(nkpg);
1600 ptppaddr = VM_PAGE_TO_PHYS(nkpg);
1601 pmap_zero_page(ptppaddr);
1602 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M);
1603 pdir_pde(PTD, kernel_vm_end) = newpdir;
fbbaeba3 1604 *pmap_pde(&kernel_pmap, kernel_vm_end) = newpdir;
0e5797fe
MD
1605 nkpt++;
1606
1607 /*
54a764e8 1608 * This update must be interlocked with pmap_pinit2.
0e5797fe 1609 */
b12defdc 1610 spin_lock(&pmap_spin);
54a764e8
MD
1611 TAILQ_FOREACH(pmap, &pmap_list, pm_pmnode) {
1612 *pmap_pde(pmap, kernel_vm_end) = newpdir;
1613 }
b12defdc 1614 spin_unlock(&pmap_spin);
54a764e8
MD
1615 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) &
1616 ~(PAGE_SIZE * NPTEPG - 1);
984263bc 1617 }
b12defdc 1618 vm_object_drop(kptobj);
984263bc
MD
1619}
1620
1621/*
4107b0c0
MD
1622 * Add a reference to the specified pmap.
1623 *
1624 * No requirements.
984263bc
MD
1625 */
1626void
840de426 1627pmap_reference(pmap_t pmap)
984263bc 1628{
4107b0c0
MD
1629 if (pmap) {
1630 lwkt_gettoken(&vm_token);
1631 ++pmap->pm_count;
1632 lwkt_reltoken(&vm_token);
984263bc
MD
1633 }
1634}
1635
3321ee05
MD
1636/*
1637 * vm_token must be held
1638 */
1639static
1640void
1641pmap_hold(pmap_t pmap)
1642{
1643 ++pmap->pm_count;
1644}
1645
1646/*
1647 * vm_token must be held
1648 */
1649static
1650void
1651pmap_drop(pmap_t pmap)
1652{
1653 --pmap->pm_count;
22c4e116 1654 if (pmap->pm_count == (int)0x80000000)
3321ee05
MD
1655 wakeup(pmap);
1656}
1657
1658static
1659void
1660pmap_wait(pmap_t pmap, int count)
1661{
1662 lwkt_gettoken(&vm_token);
1663 pmap->pm_count += count;
1664 if (pmap->pm_count & 0x7FFFFFFF) {
1665 while (pmap->pm_count & 0x7FFFFFFF) {
1666 pmap->pm_count |= 0x80000000;
1667 tsleep(pmap, 0, "pmapd", 0);
1668 pmap->pm_count &= ~0x80000000;
1669 kprintf("pmap_wait: race averted\n");
1670 }
1671 }
1672 lwkt_reltoken(&vm_token);
1673}
1674
984263bc 1675/***************************************************
4107b0c0 1676 * page management routines.
984263bc
MD
1677 ***************************************************/
1678
1679/*
8a8d5d85
MD
1680 * free the pv_entry back to the free list. This function may be
1681 * called from an interrupt.
4107b0c0
MD
1682 *
1683 * The caller must hold vm_token.
984263bc
MD
1684 */
1685static PMAP_INLINE void
840de426 1686free_pv_entry(pv_entry_t pv)
984263bc 1687{
2bb9cc6f
MD
1688 struct mdglobaldata *gd;
1689
5926987a
MD
1690#ifdef PMAP_DEBUG
1691 KKASSERT(pv->pv_m != NULL);
1692 pv->pv_m = NULL;
1693#endif
2bb9cc6f 1694 gd = mdcpu;
984263bc 1695 pv_entry_count--;
2bb9cc6f
MD
1696 if (gd->gd_freepv == NULL)
1697 gd->gd_freepv = pv;
1698 else
1699 zfree(pvzone, pv);
984263bc
MD
1700}
1701
1702/*
1703 * get a new pv_entry, allocating a block from the system
2bb9cc6f
MD
1704 * when needed. This function may be called from an interrupt thread.
1705 *
1706 * THIS FUNCTION CAN BLOCK ON THE ZALLOC TOKEN, serialization of other
1707 * tokens (aka vm_token) to be temporarily lost.
4107b0c0
MD
1708 *
1709 * The caller must hold vm_token.
984263bc
MD
1710 */
1711static pv_entry_t
1712get_pv_entry(void)
1713{
2bb9cc6f
MD
1714 struct mdglobaldata *gd;
1715 pv_entry_t pv;
1716
984263bc
MD
1717 pv_entry_count++;
1718 if (pv_entry_high_water &&
20479584
MD
1719 (pv_entry_count > pv_entry_high_water) &&
1720 (pmap_pagedaemon_waken == 0)) {
984263bc
MD
1721 pmap_pagedaemon_waken = 1;
1722 wakeup (&vm_pages_needed);
1723 }
2bb9cc6f
MD
1724 gd = mdcpu;
1725 if ((pv = gd->gd_freepv) != NULL)
1726 gd->gd_freepv = NULL;
1727 else
1728 pv = zalloc(pvzone);
1729 return pv;
984263bc
MD
1730}
1731
1732/*
1733 * This routine is very drastic, but can save the system
1734 * in a pinch.
4107b0c0
MD
1735 *
1736 * No requirements.
984263bc
MD
1737 */
1738void
840de426 1739pmap_collect(void)
984263bc
MD
1740{
1741 int i;
1742 vm_page_t m;
1743 static int warningdone=0;
1744
1745 if (pmap_pagedaemon_waken == 0)
1746 return;
4107b0c0 1747 lwkt_gettoken(&vm_token);
20479584 1748 pmap_pagedaemon_waken = 0;
984263bc
MD
1749
1750 if (warningdone < 5) {
948209ce
MD
1751 kprintf("pmap_collect: collecting pv entries -- "
1752 "suggest increasing PMAP_SHPGPERPROC\n");
984263bc
MD
1753 warningdone++;
1754 }
1755
b12defdc 1756 for (i = 0; i < vm_page_array_size; i++) {
984263bc 1757 m = &vm_page_array[i];
b12defdc 1758 if (m->wire_count || m->hold_count)
984263bc 1759 continue;
b12defdc
MD
1760 if (vm_page_busy_try(m, TRUE) == 0) {
1761 if (m->wire_count == 0 && m->hold_count == 0) {
1762 pmap_remove_all(m);
1763 }
1764 vm_page_wakeup(m);
4107b0c0 1765 }
984263bc 1766 }
4107b0c0 1767 lwkt_reltoken(&vm_token);
984263bc
MD
1768}
1769
1770
1771/*
b1482674
MD
1772 * Remove the pv entry and unwire the page table page related to the
1773 * pte the caller has cleared from the page table.
4107b0c0
MD
1774 *
1775 * The caller must hold vm_token.
984263bc 1776 */
554cf9ac 1777static void
0f7a3396 1778pmap_remove_entry(struct pmap *pmap, vm_page_t m,
4107b0c0 1779 vm_offset_t va, pmap_inval_info_t info)
984263bc
MD
1780{
1781 pv_entry_t pv;
984263bc 1782
b1482674
MD
1783 /*
1784 * Cannot block
1785 */
4107b0c0 1786 ASSERT_LWKT_TOKEN_HELD(&vm_token);
984263bc
MD
1787 if (m->md.pv_list_count < pmap->pm_stats.resident_count) {
1788 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
1789 if (pmap == pv->pv_pmap && va == pv->pv_va)
1790 break;
1791 }
1792 } else {
1793 TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) {
5926987a
MD
1794#ifdef PMAP_DEBUG
1795 KKASSERT(pv->pv_pmap == pmap);
1796#endif
1797 if (va == pv->pv_va)
984263bc
MD
1798 break;
1799 }
1800 }
5926987a 1801 KKASSERT(pv);
984263bc 1802
b1482674
MD
1803 /*
1804 * Cannot block
1805 */
5926987a
MD
1806 test_m_maps_pv(m, pv);
1807 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
1808 m->md.pv_list_count--;
cef01e15
MD
1809 if (m->object)
1810 atomic_add_int(&m->object->agg_pv_list_count, -1);
5926987a
MD
1811 if (TAILQ_EMPTY(&m->md.pv_list))
1812 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
1813 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
1814 ++pmap->pm_generation;
b1482674
MD
1815
1816 /*
1817 * This can block.
1818 */
b12defdc 1819 vm_object_hold(pmap->pm_pteobj);
554cf9ac 1820 pmap_unuse_pt(pmap, va, pv->pv_ptem, info);
b12defdc 1821 vm_object_drop(pmap->pm_pteobj);
5926987a 1822 free_pv_entry(pv);
984263bc
MD
1823}
1824
1825/*
4107b0c0
MD
1826 * Create a pv entry for page at pa for (pmap, va).
1827 *
1828 * The caller must hold vm_token.
984263bc
MD
1829 */
1830static void
2bb9cc6f
MD
1831pmap_insert_entry(pmap_t pmap, pv_entry_t pv, vm_offset_t va,
1832 vm_page_t mpte, vm_page_t m)
984263bc 1833{
5926987a
MD
1834#ifdef PMAP_DEBUG
1835 KKASSERT(pv->pv_m == NULL);
1836 pv->pv_m = m;
1837#endif
984263bc
MD
1838 pv->pv_va = va;
1839 pv->pv_pmap = pmap;
1840 pv->pv_ptem = mpte;
1841
1842 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist);
1843 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list);
5926987a 1844 ++pmap->pm_generation;
984263bc 1845 m->md.pv_list_count++;
cef01e15
MD
1846 if (m->object)
1847 atomic_add_int(&m->object->agg_pv_list_count, 1);
984263bc
MD
1848}
1849
1850/*
5926987a
MD
1851 * pmap_remove_pte: do the things to unmap a page in a process.
1852 *
4107b0c0
MD
1853 * The caller must hold vm_token.
1854 *
1855 * WARNING! As with most other pmap functions this one can block, so
1856 * callers using temporary page table mappings must reload
1857 * them.
984263bc 1858 */
554cf9ac 1859static void
0f7a3396 1860pmap_remove_pte(struct pmap *pmap, unsigned *ptq, vm_offset_t va,
5926987a 1861 pmap_inval_info_t info)
984263bc
MD
1862{
1863 unsigned oldpte;
1864 vm_page_t m;
1865
5926987a 1866 ptbase_assert(pmap);
c2fb025d 1867 pmap_inval_interlock(info, pmap, va);
5926987a 1868 ptbase_assert(pmap);
984263bc
MD
1869 oldpte = loadandclear(ptq);
1870 if (oldpte & PG_W)
1871 pmap->pm_stats.wired_count -= 1;
c2fb025d 1872 pmap_inval_deinterlock(info, pmap);
90244566 1873 KKASSERT(oldpte & PG_V);
984263bc
MD
1874 /*
1875 * Machines that don't support invlpg, also don't support
0f7a3396
MD
1876 * PG_G. XXX PG_G is disabled for SMP so don't worry about
1877 * the SMP case.
984263bc
MD
1878 */
1879 if (oldpte & PG_G)
41a01a4d 1880 cpu_invlpg((void *)va);
eec2b734
MD
1881 KKASSERT(pmap->pm_stats.resident_count > 0);
1882 --pmap->pm_stats.resident_count;
984263bc
MD
1883 if (oldpte & PG_MANAGED) {
1884 m = PHYS_TO_VM_PAGE(oldpte);
1885 if (oldpte & PG_M) {
1886#if defined(PMAP_DIAGNOSTIC)
1887 if (pmap_nw_modified((pt_entry_t) oldpte)) {
d557216f
MD
1888 kprintf("pmap_remove: modified page not "
1889 "writable: va: %p, pte: 0x%lx\n",
1890 (void *)va, (long)oldpte);
984263bc
MD
1891 }
1892#endif
1893 if (pmap_track_modified(va))
1894 vm_page_dirty(m);
1895 }
1896 if (oldpte & PG_A)
1897 vm_page_flag_set(m, PG_REFERENCED);
554cf9ac 1898 pmap_remove_entry(pmap, m, va, info);
984263bc 1899 } else {
554cf9ac 1900 pmap_unuse_pt(pmap, va, NULL, info);
984263bc 1901 }
984263bc
MD
1902}
1903
1904/*
5926987a 1905 * Remove a single page from a process address space.
e0e69b7d 1906 *
4107b0c0 1907 * The caller must hold vm_token.
984263bc
MD
1908 */
1909static void
0f7a3396 1910pmap_remove_page(struct pmap *pmap, vm_offset_t va, pmap_inval_info_t info)
984263bc 1911{
840de426 1912 unsigned *ptq;
984263bc
MD
1913
1914 /*
90244566 1915 * If there is no pte for this address, just skip it!!! Otherwise
e0e69b7d 1916 * get a local va for mappings for this pmap and remove the entry.
984263bc 1917 */
e0e69b7d
MD
1918 if (*pmap_pde(pmap, va) != 0) {
1919 ptq = get_ptbase(pmap) + i386_btop(va);
1920 if (*ptq) {
0f7a3396 1921 pmap_remove_pte(pmap, ptq, va, info);
5926987a 1922 /* ptq invalid */
e0e69b7d 1923 }
984263bc 1924 }
984263bc
MD
1925}
1926
1927/*
4107b0c0 1928 * Remove the given range of addresses from the specified map.
984263bc 1929 *
4107b0c0
MD
1930 * It is assumed that the start and end are properly rounded to the page
1931 * size.
e0e69b7d 1932 *
4107b0c0 1933 * No requirements.
984263bc
MD
1934 */
1935void
840de426 1936pmap_remove(struct pmap *pmap, vm_offset_t sva, vm_offset_t eva)
984263bc 1937{
840de426 1938 unsigned *ptbase;
984263bc
MD
1939 vm_offset_t pdnxt;
1940 vm_offset_t ptpaddr;
1941 vm_offset_t sindex, eindex;
0f7a3396 1942 struct pmap_inval_info info;
984263bc
MD
1943
1944 if (pmap == NULL)
1945 return;
1946
b12defdc 1947 vm_object_hold(pmap->pm_pteobj);
4107b0c0
MD
1948 lwkt_gettoken(&vm_token);
1949 if (pmap->pm_stats.resident_count == 0) {
1950 lwkt_reltoken(&vm_token);
b12defdc 1951 vm_object_drop(pmap->pm_pteobj);
984263bc 1952 return;
4107b0c0 1953 }
984263bc 1954
0f7a3396
MD
1955 pmap_inval_init(&info);
1956
984263bc
MD
1957 /*
1958 * special handling of removing one page. a very
1959 * common operation and easy to short circuit some
1960 * code.
1961 */
1962 if (((sva + PAGE_SIZE) == eva) &&
1963 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) {
0f7a3396 1964 pmap_remove_page(pmap, sva, &info);
c2fb025d 1965 pmap_inval_done(&info);
4107b0c0 1966 lwkt_reltoken(&vm_token);
b12defdc 1967 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
1968 return;
1969 }
1970
984263bc
MD
1971 /*
1972 * Get a local virtual address for the mappings that are being
1973 * worked with.
1974 */
984263bc
MD
1975 sindex = i386_btop(sva);
1976 eindex = i386_btop(eva);
1977
554cf9ac 1978 while (sindex < eindex) {
984263bc
MD
1979 unsigned pdirindex;
1980
1981 /*
554cf9ac 1982 * Stop scanning if no pages are left
984263bc 1983 */
984263bc
MD
1984 if (pmap->pm_stats.resident_count == 0)
1985 break;
1986
554cf9ac
MD
1987 /*
1988 * Calculate index for next page table, limited by eindex.
1989 */
1990 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
1991 if (pdnxt > eindex)
1992 pdnxt = eindex;
1993
984263bc 1994 pdirindex = sindex / NPDEPG;
554cf9ac
MD
1995 ptpaddr = (unsigned)pmap->pm_pdir[pdirindex];
1996 if (ptpaddr & PG_PS) {
c2fb025d 1997 pmap_inval_interlock(&info, pmap, -1);
984263bc
MD
1998 pmap->pm_pdir[pdirindex] = 0;
1999 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
be3aecf7 2000 pmap->pm_cached = 0;
c2fb025d 2001 pmap_inval_deinterlock(&info, pmap);
554cf9ac 2002 sindex = pdnxt;
984263bc
MD
2003 continue;
2004 }
2005
2006 /*
2007 * Weed out invalid mappings. Note: we assume that the page
2008 * directory table is always allocated, and in kernel virtual.
2009 */
554cf9ac
MD
2010 if (ptpaddr == 0) {
2011 sindex = pdnxt;
984263bc 2012 continue;
984263bc
MD
2013 }
2014
8790d7d8 2015 /*
554cf9ac
MD
2016 * Sub-scan the page table page. pmap_remove_pte() can
2017 * block on us, invalidating ptbase, so we must reload
2018 * ptbase and we must also check whether the page directory
2019 * page is still present.
8790d7d8 2020 */
554cf9ac 2021 while (sindex < pdnxt) {
984263bc 2022 vm_offset_t va;
8790d7d8
MD
2023
2024 ptbase = get_ptbase(pmap);
554cf9ac
MD
2025 if (ptbase[sindex]) {
2026 va = i386_ptob(sindex);
2027 pmap_remove_pte(pmap, ptbase + sindex,
2028 va, &info);
2029 }
2030 if (pmap->pm_pdir[pdirindex] == 0 ||
2031 (pmap->pm_pdir[pdirindex] & PG_PS)) {
984263bc 2032 break;
554cf9ac
MD
2033 }
2034 ++sindex;
984263bc
MD
2035 }
2036 }
c2fb025d 2037 pmap_inval_done(&info);
4107b0c0 2038 lwkt_reltoken(&vm_token);
b12defdc 2039 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
2040}
2041
2042/*
4107b0c0
MD
2043 * Removes this physical page from all physical maps in which it resides.
2044 * Reflects back modify bits to the pager.
984263bc 2045 *
3321ee05 2046 * vm_token must be held by caller.
984263bc 2047 */
984263bc 2048static void
840de426 2049pmap_remove_all(vm_page_t m)
984263bc 2050{
0f7a3396 2051 struct pmap_inval_info info;
840de426 2052 unsigned *pte, tpte;
0f7a3396 2053 pv_entry_t pv;
3321ee05 2054 pmap_t pmap;
984263bc 2055
bee81bdd
SS
2056 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2057 return;
bcc6a8ac
MD
2058 if (TAILQ_EMPTY(&m->md.pv_list))
2059 return;
984263bc 2060
0f7a3396 2061 pmap_inval_init(&info);
984263bc 2062 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) {
3321ee05
MD
2063 pmap = pv->pv_pmap;
2064 KKASSERT(pmap->pm_stats.resident_count > 0);
3321ee05 2065 pmap_hold(pmap);
2160c64a
MD
2066 vm_object_hold(pmap->pm_pteobj);
2067
2068 if (pv != TAILQ_FIRST(&m->md.pv_list)) {
2160c64a
MD
2069 vm_object_drop(pmap->pm_pteobj);
2070 pmap_drop(pmap);
2071 continue;
2072 }
984263bc 2073
1c10d640 2074 --pmap->pm_stats.resident_count;
3321ee05
MD
2075 pte = pmap_pte_quick(pmap, pv->pv_va);
2076 pmap_inval_interlock(&info, pmap, pv->pv_va);
984263bc
MD
2077 tpte = loadandclear(pte);
2078 if (tpte & PG_W)
3321ee05
MD
2079 pmap->pm_stats.wired_count--;
2080 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2081 if (tpte & PG_A)
2082 vm_page_flag_set(m, PG_REFERENCED);
c2fb025d 2083 KKASSERT(PHYS_TO_VM_PAGE(tpte) == m);
984263bc
MD
2084
2085 /*
2086 * Update the vm_page_t clean and reference bits.
2087 */
2088 if (tpte & PG_M) {
2089#if defined(PMAP_DIAGNOSTIC)
2090 if (pmap_nw_modified((pt_entry_t) tpte)) {
d557216f
MD
2091 kprintf("pmap_remove_all: modified page "
2092 "not writable: va: %p, pte: 0x%lx\n",
2093 (void *)pv->pv_va, (long)tpte);
984263bc
MD
2094 }
2095#endif
2096 if (pmap_track_modified(pv->pv_va))
2097 vm_page_dirty(m);
2098 }
5926987a
MD
2099#ifdef PMAP_DEBUG
2100 KKASSERT(pv->pv_m == m);
2101#endif
2bb9cc6f 2102 KKASSERT(pv == TAILQ_FIRST(&m->md.pv_list));
984263bc 2103 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
3321ee05
MD
2104 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
2105 ++pmap->pm_generation;
984263bc 2106 m->md.pv_list_count--;
cef01e15
MD
2107 if (m->object)
2108 atomic_add_int(&m->object->agg_pv_list_count, -1);
17cde63e
MD
2109 if (TAILQ_EMPTY(&m->md.pv_list))
2110 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
3321ee05
MD
2111 pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem, &info);
2112 vm_object_drop(pmap->pm_pteobj);
984263bc 2113 free_pv_entry(pv);
3321ee05 2114 pmap_drop(pmap);
984263bc 2115 }
17cde63e 2116 KKASSERT((m->flags & (PG_MAPPED|PG_WRITEABLE)) == 0);
c2fb025d 2117 pmap_inval_done(&info);
984263bc
MD
2118}
2119
2120/*
4107b0c0
MD
2121 * Set the physical protection on the specified range of this map
2122 * as requested.
e0e69b7d 2123 *
4107b0c0 2124 * No requirements.
984263bc
MD
2125 */
2126void
2127pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot)
2128{
840de426 2129 unsigned *ptbase;
984263bc
MD
2130 vm_offset_t pdnxt, ptpaddr;
2131 vm_pindex_t sindex, eindex;
0f7a3396 2132 pmap_inval_info info;
984263bc
MD
2133
2134 if (pmap == NULL)
2135 return;
2136
2137 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2138 pmap_remove(pmap, sva, eva);
2139 return;
2140 }
2141
2142 if (prot & VM_PROT_WRITE)
2143 return;
2144
4107b0c0 2145 lwkt_gettoken(&vm_token);
0f7a3396 2146 pmap_inval_init(&info);
984263bc
MD
2147
2148 ptbase = get_ptbase(pmap);
2149
2150 sindex = i386_btop(sva);
2151 eindex = i386_btop(eva);
2152
2153 for (; sindex < eindex; sindex = pdnxt) {
984263bc
MD
2154 unsigned pdirindex;
2155
2156 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1));
2157
2158 pdirindex = sindex / NPDEPG;
2159 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) {
c2fb025d 2160 pmap_inval_interlock(&info, pmap, -1);
55f2596a 2161 pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW);
984263bc 2162 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE;
c2fb025d 2163 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2164 continue;
2165 }
2166
2167 /*
2168 * Weed out invalid mappings. Note: we assume that the page
2169 * directory table is always allocated, and in kernel virtual.
2170 */
2171 if (ptpaddr == 0)
2172 continue;
2173
2174 if (pdnxt > eindex) {
2175 pdnxt = eindex;
2176 }
2177
2178 for (; sindex != pdnxt; sindex++) {
984263bc 2179 unsigned pbits;
c2fb025d 2180 unsigned cbits;
984263bc
MD
2181 vm_page_t m;
2182
17cde63e 2183 /*
d5b2d319 2184 * XXX non-optimal.
17cde63e 2185 */
c2fb025d
MD
2186 pmap_inval_interlock(&info, pmap, i386_ptob(sindex));
2187again:
984263bc 2188 pbits = ptbase[sindex];
c2fb025d 2189 cbits = pbits;
984263bc
MD
2190
2191 if (pbits & PG_MANAGED) {
2192 m = NULL;
2193 if (pbits & PG_A) {
2194 m = PHYS_TO_VM_PAGE(pbits);
2195 vm_page_flag_set(m, PG_REFERENCED);
c2fb025d 2196 cbits &= ~PG_A;
984263bc
MD
2197 }
2198 if (pbits & PG_M) {
2199 if (pmap_track_modified(i386_ptob(sindex))) {
2200 if (m == NULL)
2201 m = PHYS_TO_VM_PAGE(pbits);
2202 vm_page_dirty(m);
c2fb025d 2203 cbits &= ~PG_M;
984263bc
MD
2204 }
2205 }
2206 }
c2fb025d
MD
2207 cbits &= ~PG_RW;
2208 if (pbits != cbits &&
2209 !atomic_cmpset_int(ptbase + sindex, pbits, cbits)) {
2210 goto again;
984263bc 2211 }
c2fb025d 2212 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
2213 }
2214 }
c2fb025d 2215 pmap_inval_done(&info);
4107b0c0 2216 lwkt_reltoken(&vm_token);
984263bc
MD
2217}
2218
2219/*
4107b0c0
MD
2220 * Insert the given physical page (p) at the specified virtual address (v)
2221 * in the target physical map with the protection requested.
984263bc 2222 *
4107b0c0
MD
2223 * If specified, the page will be wired down, meaning that the related pte
2224 * cannot be reclaimed.
984263bc 2225 *
4107b0c0 2226 * No requirements.
984263bc
MD
2227 */
2228void
2229pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
921c891e 2230 boolean_t wired, vm_map_entry_t entry __unused)
984263bc 2231{
6ef943a3 2232 vm_paddr_t pa;
840de426 2233 unsigned *pte;
6ef943a3 2234 vm_paddr_t opa;
984263bc
MD
2235 vm_offset_t origpte, newpte;
2236 vm_page_t mpte;
0f7a3396 2237 pmap_inval_info info;
2bb9cc6f 2238 pv_entry_t pv;
984263bc
MD
2239
2240 if (pmap == NULL)
2241 return;
2242
2243 va &= PG_FRAME;
2244#ifdef PMAP_DIAGNOSTIC
c439ad8f 2245 if (va >= KvaEnd)
984263bc 2246 panic("pmap_enter: toobig");
d557216f
MD
2247 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) {
2248 panic("pmap_enter: invalid to pmap_enter page "
2249 "table pages (va: %p)", (void *)va);
2250 }
984263bc 2251#endif
fbbaeba3
MD
2252 if (va < UPT_MAX_ADDRESS && pmap == &kernel_pmap) {
2253 kprintf("Warning: pmap_enter called on UVA with kernel_pmap\n");
7ce2998e 2254 print_backtrace(-1);
fbbaeba3
MD
2255 }
2256 if (va >= UPT_MAX_ADDRESS && pmap != &kernel_pmap) {
2257 kprintf("Warning: pmap_enter called on KVA without kernel_pmap\n");
7ce2998e 2258 print_backtrace(-1);
fbbaeba3 2259 }
984263bc 2260
b12defdc 2261 vm_object_hold(pmap->pm_pteobj);
4107b0c0
MD
2262 lwkt_gettoken(&vm_token);
2263
984263bc 2264 /*
2bb9cc6f
MD
2265 * This can block, get it before we do anything important.
2266 */
2267 if (pmap_initialized &&
2268 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2269 pv = get_pv_entry();
2270 } else {
2271 pv = NULL;
2272 }
2273
2274 /*
984263bc
MD
2275 * In the case that a page table page is not
2276 * resident, we are creating it here.
2277 */
17cde63e 2278 if (va < UPT_MIN_ADDRESS)
984263bc 2279 mpte = pmap_allocpte(pmap, va);
17cde63e
MD
2280 else
2281 mpte = NULL;
984263bc 2282
b12defdc
MD
2283 if ((prot & VM_PROT_NOSYNC) == 0)
2284 pmap_inval_init(&info);
984263bc
MD
2285 pte = pmap_pte(pmap, va);
2286
2287 /*
2288 * Page Directory table entry not valid, we need a new PT page
2289 */
2290 if (pte == NULL) {
ed20d0e3 2291 panic("pmap_enter: invalid page directory pdir=0x%lx, va=%p",
d557216f 2292 (long)pmap->pm_pdir[PTDPTDI], (void *)va);
984263bc
MD
2293 }
2294
2295 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME;
2296 origpte = *(vm_offset_t *)pte;
2297 opa = origpte & PG_FRAME;
2298
2299 if (origpte & PG_PS)
2300 panic("pmap_enter: attempted pmap_enter on 4MB page");
2301
2302 /*
2303 * Mapping has not changed, must be protection or wiring change.
2304 */
2305 if (origpte && (opa == pa)) {
2306 /*
2307 * Wiring change, just update stats. We don't worry about
2308 * wiring PT pages as they remain resident as long as there
2309 * are valid mappings in them. Hence, if a user page is wired,
2310 * the PT page will be also.
2311 */
2312 if (wired && ((origpte & PG_W) == 0))
2313 pmap->pm_stats.wired_count++;
2314 else if (!wired && (origpte & PG_W))
2315 pmap->pm_stats.wired_count--;
2316
2317#if defined(PMAP_DIAGNOSTIC)
2318 if (pmap_nw_modified((pt_entry_t) origpte)) {
d557216f
MD
2319 kprintf("pmap_enter: modified page not "
2320 "writable: va: %p, pte: 0x%lx\n",
2321 (void *)va, (long )origpte);
984263bc
MD
2322 }
2323#endif
2324
2325 /*
984263bc
MD
2326 * We might be turning off write access to the page,
2327 * so we go ahead and sense modify status.
2328 */
2329 if (origpte & PG_MANAGED) {
2330 if ((origpte & PG_M) && pmap_track_modified(va)) {
2331 vm_page_t om;
2332 om = PHYS_TO_VM_PAGE(opa);
2333 vm_page_dirty(om);
2334 }
2335 pa |= PG_MANAGED;
17cde63e 2336 KKASSERT(m->flags & PG_MAPPED);
984263bc
MD
2337 }
2338 goto validate;
2339 }
2340 /*
2341 * Mapping has changed, invalidate old range and fall through to
2342 * handle validating new mapping.
5926987a
MD
2343 *
2344 * Since we have a ref on the page directory page pmap_pte()
2345 * will always return non-NULL.
2346 *
2347 * NOTE: pmap_remove_pte() can block and cause the temporary ptbase
2348 * to get wiped. reload the ptbase. I'm not sure if it is
2349 * also possible to race another pmap_enter() but check for
2350 * that case too.
984263bc 2351 */
5926987a 2352 while (opa) {
5926987a
MD
2353 KKASSERT((origpte & PG_FRAME) ==
2354 (*(vm_offset_t *)pte & PG_FRAME));
39efc543
MD
2355 if (prot & VM_PROT_NOSYNC) {
2356 prot &= ~VM_PROT_NOSYNC;
2357 pmap_inval_init(&info);
2358 }
554cf9ac 2359 pmap_remove_pte(pmap, pte, va, &info);
5926987a
MD
2360 pte = pmap_pte(pmap, va);
2361 origpte = *(vm_offset_t *)pte;
2362 opa = origpte & PG_FRAME;
2363 if (opa) {
2364 kprintf("pmap_enter: Warning, raced pmap %p va %p\n",
2365 pmap, (void *)va);
2366 }
984263bc
MD
2367 }
2368
2369 /*
2370 * Enter on the PV list if part of our managed memory. Note that we
2371 * raise IPL while manipulating pv_table since pmap_enter can be
2372 * called at interrupt time.
2373 */
2374 if (pmap_initialized &&
2375 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2bb9cc6f
MD
2376 pmap_insert_entry(pmap, pv, va, mpte, m);
2377 pv = NULL;
5926987a 2378 ptbase_assert(pmap);
984263bc 2379 pa |= PG_MANAGED;
17cde63e 2380 vm_page_flag_set(m, PG_MAPPED);
984263bc
MD
2381 }
2382
2383 /*
2384 * Increment counters
2385 */
eec2b734 2386 ++pmap->pm_stats.resident_count;
984263bc
MD
2387 if (wired)
2388 pmap->pm_stats.wired_count++;
5926987a 2389 KKASSERT(*pte == 0);
984263bc
MD
2390
2391validate:
2392 /*
2393 * Now validate mapping with desired protection/wiring.
2394 */
5926987a 2395 ptbase_assert(pmap);
984263bc
MD
2396 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V);
2397
2398 if (wired)
2399 newpte |= PG_W;
2400 if (va < UPT_MIN_ADDRESS)
2401 newpte |= PG_U;
fbbaeba3 2402 if (pmap == &kernel_pmap)
984263bc
MD
2403 newpte |= pgeflag;
2404
2405 /*
2bb9cc6f
MD
2406 * If the mapping or permission bits are different, we need
2407 * to update the pte. If the pte is already present we have
2408 * to get rid of the extra wire-count on mpte we had obtained
2409 * above.
b1482674
MD
2410 *
2411 * mpte has a new wire_count, which also serves to prevent the
2412 * page table page from getting ripped out while we work. If we
2413 * are modifying an existing pte instead of installing a new one
2414 * we have to drop it.
984263bc
MD
2415 */
2416 if ((origpte & ~(PG_M|PG_A)) != newpte) {
b12defdc
MD
2417 if (prot & VM_PROT_NOSYNC)
2418 cpu_invlpg((void *)va);
2419 else
2420 pmap_inval_interlock(&info, pmap, va);
5926987a 2421 ptbase_assert(pmap);
2bb9cc6f
MD
2422
2423 if (*pte) {
2424 KKASSERT((*pte & PG_FRAME) == (newpte & PG_FRAME));
2425 if (vm_page_unwire_quick(mpte))
2426 panic("pmap_enter: Insufficient wire_count");
2427 }
2428
984263bc 2429 *pte = newpte | PG_A;
b12defdc
MD
2430 if ((prot & VM_PROT_NOSYNC) == 0)
2431 pmap_inval_deinterlock(&info, pmap);
17cde63e
MD
2432 if (newpte & PG_RW)
2433 vm_page_flag_set(m, PG_WRITEABLE);
b1482674
MD
2434 } else {
2435 if (*pte) {
2436 KKASSERT((*pte & PG_FRAME) == (newpte & PG_FRAME));
2437 if (vm_page_unwire_quick(mpte))
2438 panic("pmap_enter: Insufficient wire_count");
2439 }
984263bc 2440 }
b1482674
MD
2441
2442 /*
2443 * NOTE: mpte invalid after this point if we block.
2444 */
c695044a 2445 KKASSERT((newpte & PG_MANAGED) == 0 || (m->flags & PG_MAPPED));
b12defdc
MD
2446 if ((prot & VM_PROT_NOSYNC) == 0)
2447 pmap_inval_done(&info);
2bb9cc6f
MD
2448 if (pv)
2449 free_pv_entry(pv);
4107b0c0 2450 lwkt_reltoken(&vm_token);
b12defdc 2451 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
2452}
2453
2454/*
17cde63e
MD
2455 * This code works like pmap_enter() but assumes VM_PROT_READ and not-wired.
2456 * This code also assumes that the pmap has no pre-existing entry for this
2457 * VA.
2458 *
2459 * This code currently may only be used on user pmaps, not kernel_pmap.
4107b0c0
MD
2460 *
2461 * No requirements.
984263bc 2462 */
1b9d3514 2463void
17cde63e 2464pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m)
984263bc
MD
2465{
2466 unsigned *pte;
6ef943a3 2467 vm_paddr_t pa;
17cde63e
MD
2468 vm_page_t mpte;
2469 unsigned ptepindex;
2470 vm_offset_t ptepa;
0f7a3396 2471 pmap_inval_info info;
2bb9cc6f 2472 pv_entry_t pv;
0f7a3396 2473
b12defdc 2474 vm_object_hold(pmap->pm_pteobj);
4107b0c0 2475 lwkt_gettoken(&vm_token);
2bb9cc6f
MD
2476
2477 /*
2478 * This can block, get it before we do anything important.
2479 */
2480 if (pmap_initialized &&
2481 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2482 pv = get_pv_entry();
2483 } else {
2484 pv = NULL;
2485 }
2486
0f7a3396 2487 pmap_inval_init(&info);
984263bc 2488
fbbaeba3
MD
2489 if (va < UPT_MAX_ADDRESS && pmap == &kernel_pmap) {
2490 kprintf("Warning: pmap_enter_quick called on UVA with kernel_pmap\n");
7ce2998e 2491 print_backtrace(-1);
fbbaeba3
MD
2492 }
2493 if (va >= UPT_MAX_ADDRESS && pmap != &kernel_pmap) {
2494 kprintf("Warning: pmap_enter_quick called on KVA without kernel_pmap\n");
7ce2998e 2495 print_backtrace(-1);
fbbaeba3
MD
2496 }
2497
17cde63e
MD
2498 KKASSERT(va < UPT_MIN_ADDRESS); /* assert used on user pmaps only */
2499
984263bc 2500 /*
17cde63e
MD
2501 * Calculate the page table page (mpte), allocating it if necessary.
2502 *
2503 * A held page table page (mpte), or NULL, is passed onto the
2504 * section following.
984263bc
MD
2505 */
2506 if (va < UPT_MIN_ADDRESS) {
984263bc
MD
2507 /*
2508 * Calculate pagetable page index
2509 */
2510 ptepindex = va >> PDRSHIFT;
17cde63e
MD
2511
2512 do {
984263bc
MD
2513 /*
2514 * Get the page directory entry
2515 */
2516 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex];
2517
2518 /*
2519 * If the page table page is mapped, we just increment
90244566 2520 * the wire count, and activate it.
984263bc
MD
2521 */
2522 if (ptepa) {
2523 if (ptepa & PG_PS)
2524 panic("pmap_enter_quick: unexpected mapping into 4MB page");
b1482674
MD
2525 if ((mpte = pmap->pm_ptphint) != NULL &&
2526 (mpte->pindex == ptepindex) &&
2527 (mpte->flags & PG_BUSY) == 0) {
2bb9cc6f 2528 vm_page_wire_quick(mpte);
984263bc 2529 } else {
2bb9cc6f
MD
2530 mpte = pmap_page_lookup(pmap->pm_pteobj,
2531 ptepindex);
984263bc 2532 pmap->pm_ptphint = mpte;
2bb9cc6f 2533 vm_page_wire_quick(mpte);
b12defdc 2534 vm_page_wakeup(mpte);
984263bc 2535 }
984263bc
MD
2536 } else {
2537 mpte = _pmap_allocpte(pmap, ptepindex);
2538 }
17cde63e 2539 } while (mpte == NULL);
984263bc
MD
2540 } else {
2541 mpte = NULL;
17cde63e 2542 /* this code path is not yet used */
984263bc
MD
2543 }
2544
2545 /*
17cde63e
MD
2546 * With a valid (and held) page directory page, we can just use
2547 * vtopte() to get to the pte. If the pte is already present
2548 * we do not disturb it.
984263bc
MD
2549 */
2550 pte = (unsigned *)vtopte(va);
554cf9ac
MD
2551 if (*pte) {
2552 KKASSERT(*pte & PG_V);
17cde63e
MD
2553 pa = VM_PAGE_TO_PHYS(m);
2554 KKASSERT(((*pte ^ pa) & PG_FRAME) == 0);
c2fb025d 2555 pmap_inval_done(&info);
b1482674
MD
2556 if (mpte)
2557 pmap_unwire_pte(pmap, mpte, &info);
b1482674 2558 if (pv) {
2bb9cc6f 2559 free_pv_entry(pv);
b1482674
MD
2560 /* pv = NULL; */
2561 }
554cf9ac
MD
2562 lwkt_reltoken(&vm_token);
2563 vm_object_drop(pmap->pm_pteobj);
17cde63e 2564 return;
984263bc
MD
2565 }
2566
2567 /*
17cde63e 2568 * Enter on the PV list if part of our managed memory
984263bc 2569 */
2bb9cc6f
MD
2570 if (pmap_initialized &&
2571 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
2572 pmap_insert_entry(pmap, pv, va, mpte, m);
2573 pv = NULL;
17cde63e
MD
2574 vm_page_flag_set(m, PG_MAPPED);
2575 }
984263bc
MD
2576
2577 /*
2578 * Increment counters
2579 */
eec2b734 2580 ++pmap->pm_stats.resident_count;
984263bc
MD
2581
2582 pa = VM_PAGE_TO_PHYS(m);
2583
2584 /*
2585 * Now validate mapping with RO protection
2586 */
2587 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
2588 *pte = pa | PG_V | PG_U;
2589 else
2590 *pte = pa | PG_V | PG_U | PG_MANAGED;
17cde63e 2591/* pmap_inval_add(&info, pmap, va); shouldn't be needed inval->valid */
c2fb025d 2592 pmap_inval_done(&info);
b1482674 2593 if (pv) {
2bb9cc6f 2594 free_pv_entry(pv);
b1482674
MD
2595 /* pv = NULL; */
2596 }
4107b0c0 2597 lwkt_reltoken(&vm_token);
b12defdc 2598 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
2599}
2600
2601/*
2602 * Make a temporary mapping for a physical address. This is only intended
2603 * to be used for panic dumps.
4107b0c0 2604 *
fb8345e6
MD
2605 * The caller is responsible for calling smp_invltlb().
2606 *
4107b0c0 2607 * No requirements.
984263bc
MD
2608 */
2609void *
8e5ea5f7 2610pmap_kenter_temporary(vm_paddr_t pa, long i)
984263bc 2611{
fb8345e6 2612 pmap_kenter_quick((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa);
984263bc
MD
2613 return ((void *)crashdumpmap);
2614}
2615
2616#define MAX_INIT_PT (96)
06ecca5a 2617
984263bc 2618/*
06ecca5a
MD
2619 * This routine preloads the ptes for a given object into the specified pmap.
2620 * This eliminates the blast of soft faults on process startup and
2621 * immediately after an mmap.
4107b0c0
MD
2622 *
2623 * No requirements.
984263bc 2624 */
1f804340
MD
2625static int pmap_object_init_pt_callback(vm_page_t p, void *data);
2626
984263bc 2627void
083a7402
MD
2628pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_prot_t prot,
2629 vm_object_t object, vm_pindex_t pindex,
2630 vm_size_t size, int limit)
984263bc 2631{
1f804340 2632 struct rb_vm_page_scan_info info;
287ebb09 2633 struct lwp *lp;
984263bc 2634 int psize;
984263bc 2635
54a764e8
MD
2636 /*
2637 * We can't preinit if read access isn't set or there is no pmap
2638 * or object.
2639 */
083a7402 2640 if ((prot & VM_PROT_READ) == 0 || pmap == NULL || object == NULL)
984263bc
MD
2641 return;
2642
54a764e8
MD
2643 /*
2644 * We can't preinit if the pmap is not the current pmap
2645 */
287ebb09
MD
2646 lp = curthread->td_lwp;
2647 if (lp == NULL || pmap != vmspace_pmap(lp->lwp_vmspace))
54a764e8
MD
2648 return;
2649
984263bc
MD
2650 psize = i386_btop(size);
2651
2652 if ((object->type != OBJT_VNODE) ||
2653 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
2654 (object->resident_page_count > MAX_INIT_PT))) {
2655 return;
2656 }
2657
2658 if (psize + pindex > object->size) {
2659 if (object->size < pindex)
2660 return;
2661 psize = object->size - pindex;
2662 }
2663
1f804340
MD
2664 if (psize == 0)
2665 return;
06ecca5a 2666
984263bc 2667 /*
1f804340
MD
2668 * Use a red-black scan to traverse the requested range and load
2669 * any valid pages found into the pmap.
06ecca5a 2670 *
9acd5bbb
MD
2671 * We cannot safely scan the object's memq unless we are in a
2672 * critical section since interrupts can remove pages from objects.
984263bc 2673 */
1f804340
MD
2674 info.start_pindex = pindex;
2675 info.end_pindex = pindex + psize - 1;
2676 info.limit = limit;
2677 info.mpte = NULL;
2678 info.addr = addr;
2679 info.pmap = pmap;
2680
2f2d9e58 2681 vm_object_hold(object);
1f804340
MD
2682 vm_page_rb_tree_RB_SCAN(&object->rb_memq, rb_vm_page_scancmp,
2683 pmap_object_init_pt_callback, &info);
2f2d9e58 2684 vm_object_drop(object);
1f804340 2685}
06ecca5a 2686
4107b0c0
MD
2687/*
2688 * The caller must hold vm_token.
2689 */
1f804340
MD
2690static
2691int
2692pmap_object_init_pt_callback(vm_page_t p, void *data)
2693{
2694 struct rb_vm_page_scan_info *info = data;
2695 vm_pindex_t rel_index;
2696 /*
2697 * don't allow an madvise to blow away our really
2698 * free pages allocating pv entries.
2699 */
2700 if ((info->limit & MAP_PREFAULT_MADVISE) &&
2701 vmstats.v_free_count < vmstats.v_free_reserved) {
2702 return(-1);
984263bc 2703 }
0d987a03
MD
2704
2705 /*
2706 * Ignore list markers and ignore pages we cannot instantly
2707 * busy (while holding the object token).
2708 */
2709 if (p->flags & PG_MARKER)
2710 return 0;
b12defdc
MD
2711 if (vm_page_busy_try(p, TRUE))
2712 return 0;
1f804340 2713 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
b12defdc 2714 (p->flags & PG_FICTITIOUS) == 0) {
1f804340
MD
2715 if ((p->queue - p->pc) == PQ_CACHE)
2716 vm_page_deactivate(p);
1f804340 2717 rel_index = p->pindex - info->start_pindex;
17cde63e
MD
2718 pmap_enter_quick(info->pmap,
2719 info->addr + i386_ptob(rel_index), p);
1f804340 2720 }
b12defdc 2721 vm_page_wakeup(p);
1f804340 2722 return(0);
984263bc
MD
2723}
2724
2725/*
1b9d3514
MD
2726 * Return TRUE if the pmap is in shape to trivially
2727 * pre-fault the specified address.
2728 *
2729 * Returns FALSE if it would be non-trivial or if a
2730 * pte is already loaded into the slot.
4107b0c0
MD
2731 *
2732 * No requirements.
984263bc 2733 */
1b9d3514
MD
2734int
2735pmap_prefault_ok(pmap_t pmap, vm_offset_t addr)
984263bc 2736{
1b9d3514 2737 unsigned *pte;
4107b0c0 2738 int ret;
984263bc 2739
4107b0c0
MD
2740 lwkt_gettoken(&vm_token);
2741 if ((*pmap_pde(pmap, addr)) == 0) {
2742 ret = 0;
2743 } else {
2744 pte = (unsigned *) vtopte(addr);
2745 ret = (*pte) ? 0 : 1;
2746 }
2747 lwkt_reltoken(&vm_token);
2748 return(ret);
984263bc
MD
2749}
2750
2751/*
4107b0c0
MD
2752 * Change the wiring attribute for a map/virtual-adderss pair. The mapping
2753 * must already exist.
2754 *
2755 * No requirements.
984263bc
MD
2756 */
2757void
921c891e
MD
2758pmap_change_wiring(pmap_t pmap, vm_offset_t va, boolean_t wired,
2759 vm_map_entry_t entry __unused)
984263bc 2760{
840de426 2761 unsigned *pte;
984263bc
MD
2762
2763 if (pmap == NULL)
2764 return;
2765
4107b0c0 2766 lwkt_gettoken(&vm_token);
984263bc
MD
2767 pte = pmap_pte(pmap, va);
2768
2769 if (wired && !pmap_pte_w(pte))
2770 pmap->pm_stats.wired_count++;
2771 else if (!wired && pmap_pte_w(pte))
2772 pmap->pm_stats.wired_count--;
2773
2774 /*
2775 * Wiring is not a hardware characteristic so there is no need to
0f7a3396
MD
2776 * invalidate TLB. However, in an SMP environment we must use
2777 * a locked bus cycle to update the pte (if we are not using
2778 * the pmap_inval_*() API that is)... it's ok to do this for simple
2779 * wiring changes.
984263bc 2780 */
0f7a3396
MD
2781#ifdef SMP
2782 if (wired)
2783 atomic_set_int(pte, PG_W);
2784 else
2785 atomic_clear_int(pte, PG_W);
2786#else
2787 if (wired)
2788 atomic_set_int_nonlocked(pte, PG_W);
2789 else
2790 atomic_clear_int_nonlocked(pte, PG_W);
2791#endif
4107b0c0 2792 lwkt_reltoken(&vm_token);
984263bc
MD
2793}
2794
984263bc 2795/*
4107b0c0
MD
2796 * Copy the range specified by src_addr/len from the source map to the
2797 * range dst_addr/len in the destination map.
2798 *
2799 * This routine is only advisory and need not do anything.
984263bc 2800 *
4107b0c0 2801 * No requirements.
984263bc 2802 */
984263bc 2803void
840de426 2804pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr,
4107b0c0 2805 vm_size_t len, vm_offset_t src_addr)
984263bc 2806{
4107b0c0 2807 /* does nothing */
984263bc
MD
2808}
2809
2810/*
4107b0c0
MD
2811 * Zero the specified PA by mapping the page into KVM and clearing its
2812 * contents.
e0e69b7d 2813 *
4107b0c0 2814 * No requirements.
984263bc
MD
2815 */
2816void
6ef943a3 2817pmap_zero_page(vm_paddr_t phys)
984263bc 2818{
85100692 2819 struct mdglobaldata *gd = mdcpu;
17a9f566 2820
e0e69b7d 2821 crit_enter();
85100692
MD
2822 if (*(int *)gd->gd_CMAP3)
2823 panic("pmap_zero_page: CMAP3 busy");
85100692 2824 *(int *)gd->gd_CMAP3 =
17a9f566 2825 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
85100692 2826 cpu_invlpg(gd->gd_CADDR3);
1fa15583 2827 bzero(gd->gd_CADDR3, PAGE_SIZE);
85100692 2828 *(int *) gd->gd_CMAP3 = 0;
e0e69b7d 2829 crit_exit();
8100156a
MD
2830}
2831
2832/*
4107b0c0 2833 * Assert that a page is empty, panic if it isn't.
8100156a 2834 *
4107b0c0 2835 * No requirements.
8100156a
MD
2836 */
2837void
2838pmap_page_assertzero(vm_paddr_t phys)
2839{
2840 struct mdglobaldata *gd = mdcpu;
2841 int i;
2842
2843 crit_enter();
2844 if (*(int *)gd->gd_CMAP3)
2845 panic("pmap_zero_page: CMAP3 busy");
2846 *(int *)gd->gd_CMAP3 =
2847 PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2848 cpu_invlpg(gd->gd_CADDR3);
2849 for (i = 0; i < PAGE_SIZE; i += 4) {
2850 if (*(int *)((char *)gd->gd_CADDR3 + i) != 0) {
ed20d0e3 2851 panic("pmap_page_assertzero() @ %p not zero!",
8100156a
MD
2852 (void *)gd->gd_CADDR3);
2853 }
2854 }
2855 *(int *) gd->gd_CMAP3 = 0;
2856 crit_exit();
984263bc
MD
2857}
2858
2859/*
4107b0c0
MD
2860 * Zero part of a physical page by mapping it into memory and clearing
2861 * its contents with bzero.
e0e69b7d 2862 *
4107b0c0 2863 * off and size may not cover an area beyond a single hardware page.
984263bc 2864 *
4107b0c0 2865 * No requirements.
984263bc
MD
2866 */
2867void
6ef943a3 2868pmap_zero_page_area(vm_paddr_t phys, int off, int size)
984263bc 2869{
85100692 2870 struct mdglobaldata *gd = mdcpu;
17a9f566 2871
e0e69b7d 2872 crit_enter();
85100692
MD
2873 if (*(int *) gd->gd_CMAP3)
2874 panic("pmap_zero_page: CMAP3 busy");
85100692
MD
2875 *(int *) gd->gd_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M;
2876 cpu_invlpg(gd->gd_CADDR3);
1fa15583 2877 bzero((char *)gd->gd_CADDR3 + off, size);
85100692 2878 *(int *) gd->gd_CMAP3 = 0;
e0e69b7d 2879 crit_exit();
984263bc
MD
2880}
2881
2882/*
4107b0c0
MD
2883 * Copy the physical page from the source PA to the target PA.
2884 * This function may be called from an interrupt. No locking
2885 * is required.
e0e69b7d 2886 *
4107b0c0 2887 * No requirements.
984263bc
MD
2888 */
2889void
6ef943a3 2890pmap_copy_page(vm_paddr_t src, vm_paddr_t dst)
984263bc 2891{
85100692 2892 struct mdglobaldata *gd = mdcpu;
17a9f566 2893
e0e69b7d 2894 crit_enter();
85100692
MD
2895 if (*(int *) gd->gd_CMAP1)
2896 panic("pmap_copy_page: CMAP1 busy");
2897 if (*(int *) gd->gd_CMAP2)
2898 panic("pmap_copy_page: CMAP2 busy");
984263bc 2899
85100692
MD
2900 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2901 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
984263bc 2902
85100692
MD
2903 cpu_invlpg(gd->gd_CADDR1);
2904 cpu_invlpg(gd->gd_CADDR2);
984263bc 2905
85100692 2906 bcopy(gd->gd_CADDR1, gd->gd_CADDR2, PAGE_SIZE);
984263bc 2907
85100692
MD
2908 *(int *) gd->gd_CMAP1 = 0;
2909 *(int *) gd->gd_CMAP2 = 0;
e0e69b7d 2910 crit_exit();
984263bc
MD
2911}
2912
f6bf3af1 2913/*
4107b0c0
MD
2914 * Copy the physical page from the source PA to the target PA.
2915 * This function may be called from an interrupt. No locking
2916 * is required.
f6bf3af1 2917 *
4107b0c0 2918 * No requirements.
f6bf3af1
MD
2919 */
2920void
2921pmap_copy_page_frag(vm_paddr_t src, vm_paddr_t dst, size_t bytes)
2922{
2923 struct mdglobaldata *gd = mdcpu;
2924
2925 crit_enter();
2926 if (*(int *) gd->gd_CMAP1)
2927 panic("pmap_copy_page: CMAP1 busy");
2928 if (*(int *) gd->gd_CMAP2)
2929 panic("pmap_copy_page: CMAP2 busy");
2930
2931 *(int *) gd->gd_CMAP1 = PG_V | (src & PG_FRAME) | PG_A;
2932 *(int *) gd->gd_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M;
2933
2934 cpu_invlpg(gd->gd_CADDR1);
2935 cpu_invlpg(gd->gd_CADDR2);
2936
2937 bcopy((char *)gd->gd_CADDR1 + (src & PAGE_MASK),
2938 (char *)gd->gd_CADDR2 + (dst & PAGE_MASK),
2939 bytes);
2940
2941 *(int *) gd->gd_CMAP1 = 0;
2942 *(int *) gd->gd_CMAP2 = 0;
2943 crit_exit();
2944}
2945
984263bc
MD
2946/*
2947 * Returns true if the pmap's pv is one of the first
2948 * 16 pvs linked to from this page. This count may
2949 * be changed upwards or downwards in the future; it
2950 * is only necessary that true be returned for a small
2951 * subset of pmaps for proper page aging.
4107b0c0
MD
2952 *
2953 * No requirements.
984263bc
MD
2954 */
2955boolean_t
840de426 2956pmap_page_exists_quick(pmap_t pmap, vm_page_t m)
984263bc
MD
2957{
2958 pv_entry_t pv;
2959 int loops = 0;
984263bc
MD
2960
2961 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
2962 return FALSE;
2963
4107b0c0 2964 lwkt_gettoken(&vm_token);
984263bc
MD
2965 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
2966 if (pv->pv_pmap == pmap) {
11502947 2967 lwkt_reltoken(&vm_token);
984263bc
MD
2968 return TRUE;
2969 }
2970 loops++;
2971 if (loops >= 16)
2972 break;
2973 }
4107b0c0 2974 lwkt_reltoken(&vm_token);
984263bc
MD
2975 return (FALSE);
2976}
2977
984263bc
MD
2978/*
2979 * Remove all pages from specified address space
2980 * this aids process exit speeds. Also, this code
2981 * is special cased for current process only, but
2982 * can have the more generic (and slightly slower)
2983 * mode enabled. This is much faster than pmap_remove
2984 * in the case of running down an entire address space.
4107b0c0
MD
2985 *
2986 * No requirements.
984263bc
MD
2987 */
2988void
840de426 2989pmap_remove_pages(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
984263bc 2990{
287ebb09 2991 struct lwp *lp;
984263bc
MD
2992 unsigned *pte, tpte;
2993 pv_entry_t pv, npv;
984263bc 2994 vm_page_t m;
0f7a3396 2995 pmap_inval_info info;
4a22e893 2996 int iscurrentpmap;
8790d7d8 2997 int32_t save_generation;
984263bc 2998
287ebb09
MD
2999 lp = curthread->td_lwp;
3000 if (lp && pmap == vmspace_pmap(lp->lwp_vmspace))
4a22e893
MD
3001 iscurrentpmap = 1;
3002 else
3003 iscurrentpmap = 0;
984263bc 3004
b12defdc
MD
3005 if (pmap->pm_pteobj)
3006 vm_object_hold(pmap->pm_pteobj);
4107b0c0 3007 lwkt_gettoken(&vm_token);
0f7a3396 3008 pmap_inval_init(&info);
b12defdc 3009
4a22e893 3010 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); pv; pv = npv) {
984263bc
MD
3011 if (pv->pv_va >= eva || pv->pv_va < sva) {
3012 npv = TAILQ_NEXT(pv, pv_plist);
3013 continue;
3014 }
3015
8790d7d8
MD
3016 KKASSERT(pmap == pv->pv_pmap);
3017
4a22e893
MD
3018 if (iscurrentpmap)
3019 pte = (unsigned *)vtopte(pv->pv_va);
3020 else
8790d7d8 3021 pte = pmap_pte_quick(pmap, pv->pv_va);
5926987a 3022 KKASSERT(*pte);
c2fb025d 3023 pmap_inval_interlock(&info, pmap, pv->pv_va);
984263bc 3024
4a22e893
MD
3025 /*
3026 * We cannot remove wired pages from a process' mapping
3027 * at this time
3028 */
17cde63e 3029 if (*pte & PG_W) {
c2fb025d 3030 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
3031 npv = TAILQ_NEXT(pv, pv_plist);
3032 continue;
3033 }
2247fe02 3034 KKASSERT(*pte);
17cde63e 3035 tpte = loadandclear(pte);
c2fb025d 3036 pmap_inval_deinterlock(&info, pmap);
984263bc
MD
3037
3038 m = PHYS_TO_VM_PAGE(tpte);
5926987a 3039 test_m_maps_pv(m, pv);
984263bc
MD
3040
3041 KASSERT(m < &vm_page_array[vm_page_array_size],
3042 ("pmap_remove_pages: bad tpte %x", tpte));
3043
eec2b734
MD
3044 KKASSERT(pmap->pm_stats.resident_count > 0);
3045 --pmap->pm_stats.resident_count;
984263bc
MD
3046
3047 /*
3048 * Update the vm_page_t clean and reference bits.
3049 */
3050 if (tpte & PG_M) {
3051 vm_page_dirty(m);
3052 }
3053
984263bc 3054 npv = TAILQ_NEXT(pv, pv_plist);
5926987a
MD
3055#ifdef PMAP_DEBUG
3056 KKASSERT(pv->pv_m == m);
3057 KKASSERT(pv->pv_pmap == pmap);
3058#endif
8790d7d8
MD
3059 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist);
3060 save_generation = ++pmap->pm_generation;
984263bc
MD
3061
3062 m->md.pv_list_count--;
cef01e15
MD
3063 if (m->object)
3064 atomic_add_int(&m->object->agg_pv_list_count, -1);
984263bc 3065 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list);
17cde63e 3066 if (TAILQ_EMPTY(&m->md.pv_list))
984263bc 3067 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE);
984263bc 3068
8790d7d8 3069 pmap_unuse_pt(pmap, pv->pv_va, pv->pv_ptem, &info);
984263bc 3070 free_pv_entry(pv);
8790d7d8
MD
3071
3072 /*
3073 * Restart the scan if we blocked during the unuse or free
3074 * calls and other removals were made.
3075 */
3076 if (save_generation != pmap->pm_generation) {
3077 kprintf("Warning: pmap_remove_pages race-A avoided\n");
5926987a 3078 npv = TAILQ_FIRST(&pmap->pm_pvlist);
8790d7d8 3079 }
984263bc 3080 }
c2fb025d 3081 pmap_inval_done(&info);
4107b0c0 3082 lwkt_reltoken(&vm_token);
b12defdc
MD
3083 if (pmap->pm_pteobj)
3084 vm_object_drop(pmap->pm_pteobj);
984263bc
MD
3085}
3086
3087/*
3088 * pmap_testbit tests bits in pte's
5e8d0349 3089 * note that the testbit/clearbit routines are inline,
984263bc 3090 * and a lot of things compile-time evaluate.
4107b0c0
MD
3091 *
3092 * The caller must hold vm_token.
984263bc
MD
3093 */
3094static boolean_t
840de426 3095pmap_testbit(vm_page_t m, int bit)
984263bc
MD
3096{
3097 pv_entry_t pv;
3098 unsigned *pte;
984263bc
MD
3099
3100 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3101 return FALSE;
3102
3103 if (TAILQ_FIRST(&m->md.pv_list) == NULL)
3104 return FALSE;
3105
984263bc
MD
3106 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3107 /*
3108 * if the bit being tested is the modified bit, then
3109 * mark clean_map and ptes as never
3110 * modified.
3111 */
3112 if (bit & (PG_A|PG_M)) {
3113 if (!pmap_track_modified(pv->pv_va))
3114 continue;
3115 }
3116
3117#if defined(PMAP_DIAGNOSTIC)
3118 if (!pv->pv_pmap) {
d557216f
MD
3119 kprintf("Null pmap (tb) at va: %p\n",
3120 (void *)pv->pv_va);
984263bc
MD
3121 continue;
3122 }
3123#endif
3124 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
74b9d1ec 3125 if (*pte & bit) {
984263bc 3126 return TRUE;
74b9d1ec 3127 }
984263bc 3128 }
984263bc
MD
3129 return (FALSE);
3130}
3131
3132/*
4107b0c0
MD
3133 * This routine is used to modify bits in ptes
3134 *
3135 * The caller must hold vm_token.
984263bc
MD
3136 */
3137static __inline void
5e8d0349 3138pmap_clearbit(vm_page_t m, int bit)
984263bc 3139{
0f7a3396 3140 struct pmap_inval_info info;
840de426
MD
3141 pv_entry_t pv;
3142 unsigned *pte;
5e8d0349 3143 unsigned pbits;
984263bc
MD
3144
3145 if (!pmap_initialized || (m->flags & PG_FICTITIOUS))
3146 return;
3147
0f7a3396 3148 pmap_inval_init(&info);
984263bc
MD
3149
3150 /*
3151 * Loop over all current mappings setting/clearing as appropos If
3152 * setting RO do we need to clear the VAC?
3153 */
3154 TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) {
3155 /*
3156 * don't write protect pager mappings
3157 */
5e8d0349 3158 if (bit == PG_RW) {
984263bc
MD
3159 if (!pmap_track_modified(pv->pv_va))
3160 continue;
3161 }
3162
3163#if defined(PMAP_DIAGNOSTIC)
3164 if (!pv->pv_pmap) {
d557216f
MD
3165 kprintf("Null pmap (cb) at va: %p\n",
3166 (void *)pv->pv_va);
984263bc
MD
3167 continue;
3168 }
3169#endif
3170
0f7a3396
MD
3171 /*
3172 * Careful here. We can use a locked bus instruction to
3173 * clear PG_A or PG_M safely but we need to synchronize
3174 * with the target cpus when we mess with PG_RW.
70fc5283
MD
3175 *
3176 * We do not have to force synchronization when clearing
3177 * PG_M even for PTEs generated via virtual memory maps,
3178 * because the virtual kernel will invalidate the pmap
3179 * entry when/if it needs to resynchronize the Modify bit.
0f7a3396 3180 */
70fc5283 3181 if (bit & PG_RW)
c2fb025d 3182 pmap_inval_interlock(&info, pv->pv_pmap, pv->pv_va);
17cde63e
MD
3183 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va);
3184again:
5e8d0349
MD
3185 pbits = *pte;
3186 if (pbits & bit) {
3187