pmap.c revision 16197
1/* 2 * Copyright (c) 1991 Regents of the University of California. 3 * All rights reserved. 4 * Copyright (c) 1994 John S. Dyson 5 * All rights reserved. 6 * Copyright (c) 1994 David Greenman 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * the Systems Programming Group of the University of Utah Computer 11 * Science Department and William Jolitz of UUNET Technologies Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by the University of 24 * California, Berkeley and its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91 42 * $Id: pmap.c,v 1.101 1996/06/07 02:36:08 dyson Exp $ 43 */ 44 45/* 46 * Derived from hp300 version by Mike Hibler, this version by William 47 * Jolitz uses a recursive map [a pde points to the page directory] to 48 * map the page tables using the pagetables themselves. This is done to 49 * reduce the impact on kernel virtual memory for lots of sparse address 50 * space, and to reduce the cost of memory to each process. 51 * 52 * Derived from: hp300/@(#)pmap.c 7.1 (Berkeley) 12/5/90 53 */ 54 55/* 56 * Manages physical address maps. 57 * 58 * In addition to hardware address maps, this 59 * module is called upon to provide software-use-only 60 * maps which may or may not be stored in the same 61 * form as hardware maps. These pseudo-maps are 62 * used to store intermediate results from copy 63 * operations to and from address spaces. 64 * 65 * Since the information managed by this module is 66 * also stored by the logical address mapping module, 67 * this module may throw away valid virtual-to-physical 68 * mappings at almost any time. However, invalidations 69 * of virtual-to-physical mappings must be done as 70 * requested. 71 * 72 * In order to cope with hardware architectures which 73 * make virtual-to-physical map invalidates expensive, 74 * this module may delay invalidate or reduced protection 75 * operations until such time as they are actually 76 * necessary. This module is given full information as 77 * to which processors are currently using which maps, 78 * and to when physical maps must be made correct. 79 */ 80 81#include <sys/param.h> 82#include <sys/systm.h> 83#include <sys/proc.h> 84#include <sys/malloc.h> 85#include <sys/msgbuf.h> 86#include <sys/queue.h> 87#include <sys/vmmeter.h> 88#include <sys/mman.h> 89 90#include <vm/vm.h> 91#include <vm/vm_param.h> 92#include <vm/vm_prot.h> 93#include <vm/lock.h> 94#include <vm/vm_kern.h> 95#include <vm/vm_page.h> 96#include <vm/vm_map.h> 97#include <vm/vm_object.h> 98#include <vm/vm_extern.h> 99#include <vm/vm_pageout.h> 100 101#include <machine/pcb.h> 102#include <machine/cputypes.h> 103#include <machine/md_var.h> 104 105#include <i386/isa/isa.h> 106 107#define PMAP_KEEP_PDIRS 108 109#if defined(DIAGNOSTIC) 110#define PMAP_DIAGNOSTIC 111#endif 112 113static void init_pv_entries __P((int)); 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) 126 127#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 128#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 129 130/* 131 * Given a map and a machine independent protection code, 132 * convert to a vax protection code. 133 */ 134#define pte_prot(m, p) (protection_codes[p]) 135static int protection_codes[8]; 136 137static struct pmap kernel_pmap_store; 138pmap_t kernel_pmap; 139 140vm_offset_t avail_start; /* PA of first available physical page */ 141vm_offset_t avail_end; /* PA of last available physical page */ 142vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 143vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 144static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 145static vm_offset_t vm_first_phys; 146 147static int nkpt; 148static vm_page_t nkpg; 149vm_offset_t kernel_vm_end; 150 151extern vm_offset_t clean_sva, clean_eva; 152extern int cpu_class; 153 154#define PV_FREELIST_MIN ((PAGE_SIZE / sizeof (struct pv_entry)) / 2) 155 156/* 157 * Data for the pv entry allocation mechanism 158 */ 159static int pv_freelistcnt; 160static pv_entry_t pv_freelist; 161static vm_offset_t pvva; 162static int npvvapg; 163 164/* 165 * All those kernel PT submaps that BSD is so fond of 166 */ 167pt_entry_t *CMAP1; 168static pt_entry_t *CMAP2, *ptmmap; 169static pv_entry_t *pv_table; 170caddr_t CADDR1, ptvmmap; 171static caddr_t CADDR2; 172static pt_entry_t *msgbufmap; 173struct msgbuf *msgbufp; 174 175static void free_pv_entry __P((pv_entry_t pv)); 176static __inline unsigned * get_ptbase __P((pmap_t pmap)); 177static pv_entry_t get_pv_entry __P((void)); 178static void i386_protection_init __P((void)); 179static void pmap_alloc_pv_entry __P((void)); 180static void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 181static void pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 182 vm_offset_t pa)); 183static int pmap_is_managed __P((vm_offset_t pa)); 184static void pmap_remove_all __P((vm_offset_t pa)); 185static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 186static __inline int pmap_remove_entry __P((struct pmap *pmap, pv_entry_t *pv, 187 vm_offset_t va)); 188static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 189 vm_offset_t sva)); 190static boolean_t 191 pmap_testbit __P((vm_offset_t pa, int bit)); 192static __inline void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 193 vm_page_t mpte, vm_offset_t pa)); 194 195static __inline vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 196static void pmap_remove_pte_mapping __P((vm_offset_t pa)); 197static __inline int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 198static vm_page_t _pmap_allocpte __P((pmap_t pmap, vm_offset_t va, int ptepindex)); 199 200#define PDSTACKMAX 16 201static vm_offset_t pdstack[PDSTACKMAX]; 202static int pdstackptr; 203 204#if defined(PMAP_DIAGNOSTIC) 205 206/* 207 * This code checks for non-writeable/modified pages. 208 * This should be an invalid condition. 209 */ 210static int 211pmap_nw_modified(pt_entry_t ptea) { 212 int pte; 213 214 pte = (int) ptea; 215 216 if ((pte & (PG_M|PG_RW)) == PG_M) 217 return 1; 218 else 219 return 0; 220} 221#endif 222 223/* 224 * The below are finer grained pmap_update routines. These eliminate 225 * the gratuitious tlb flushes on non-i386 architectures. 226 */ 227static __inline void 228pmap_update_1pg( vm_offset_t va) { 229#if defined(I386_CPU) 230 if (cpu_class == CPUCLASS_386) 231 pmap_update(); 232 else 233#endif 234 __asm __volatile(".byte 0xf,0x1,0x38": :"a" (va)); 235} 236 237static __inline void 238pmap_update_2pg( vm_offset_t va1, vm_offset_t va2) { 239#if defined(I386_CPU) 240 if (cpu_class == CPUCLASS_386) { 241 pmap_update(); 242 } else 243#endif 244 { 245 __asm __volatile(".byte 0xf,0x1,0x38": :"a" (va1)); 246 __asm __volatile(".byte 0xf,0x1,0x38": :"a" (va2)); 247 } 248} 249 250static __inline __pure unsigned * 251get_ptbase(pmap) 252 pmap_t pmap; 253{ 254 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 255 256 /* are we current address space or kernel? */ 257 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 258 return (unsigned *) PTmap; 259 } 260 /* otherwise, we are alternate address space */ 261 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 262 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 263 pmap_update(); 264 } 265 return (unsigned *) APTmap; 266} 267 268/* 269 * Routine: pmap_pte 270 * Function: 271 * Extract the page table entry associated 272 * with the given map/virtual_address pair. 273 */ 274 275__inline unsigned * __pure 276pmap_pte(pmap, va) 277 register pmap_t pmap; 278 vm_offset_t va; 279{ 280 if (pmap && *pmap_pde(pmap, va)) { 281 return get_ptbase(pmap) + i386_btop(va); 282 } 283 return (0); 284} 285 286/* 287 * Routine: pmap_extract 288 * Function: 289 * Extract the physical page address associated 290 * with the given map/virtual_address pair. 291 */ 292vm_offset_t __pure 293pmap_extract(pmap, va) 294 register pmap_t pmap; 295 vm_offset_t va; 296{ 297 if (pmap && *pmap_pde(pmap, va)) { 298 unsigned *pte; 299 pte = get_ptbase(pmap) + i386_btop(va); 300 return ((*pte & PG_FRAME) | (va & PAGE_MASK)); 301 } 302 return 0; 303 304} 305 306/* 307 * Add a list of wired pages to the kva 308 * this routine is only used for temporary 309 * kernel mappings that do not need to have 310 * page modification or references recorded. 311 * Note that old mappings are simply written 312 * over. The page *must* be wired. 313 */ 314void 315pmap_qenter(va, m, count) 316 vm_offset_t va; 317 vm_page_t *m; 318 int count; 319{ 320 int i; 321 register unsigned *pte; 322 323 for (i = 0; i < count; i++) { 324 vm_offset_t tva = va + i * PAGE_SIZE; 325 unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V; 326 unsigned opte; 327 pte = (unsigned *)vtopte(tva); 328 opte = *pte; 329 *pte = npte; 330 if (opte) 331 pmap_update_1pg(tva); 332 } 333} 334/* 335 * this routine jerks page mappings from the 336 * kernel -- it is meant only for temporary mappings. 337 */ 338void 339pmap_qremove(va, count) 340 vm_offset_t va; 341 int count; 342{ 343 int i; 344 register unsigned *pte; 345 346 for (i = 0; i < count; i++) { 347 pte = (unsigned *)vtopte(va); 348 *pte = 0; 349 pmap_update_1pg(va); 350 va += PAGE_SIZE; 351 } 352} 353 354/* 355 * add a wired page to the kva 356 * note that in order for the mapping to take effect -- you 357 * should do a pmap_update after doing the pmap_kenter... 358 */ 359__inline void 360pmap_kenter(va, pa) 361 vm_offset_t va; 362 register vm_offset_t pa; 363{ 364 register unsigned *pte; 365 unsigned npte, opte; 366 367 npte = pa | PG_RW | PG_V; 368 pte = (unsigned *)vtopte(va); 369 opte = *pte; 370 *pte = npte; 371 if (opte) 372 pmap_update_1pg(va); 373} 374 375/* 376 * remove a page from the kernel pagetables 377 */ 378__inline void 379pmap_kremove(va) 380 vm_offset_t va; 381{ 382 register unsigned *pte; 383 384 pte = (unsigned *)vtopte(va); 385 *pte = 0; 386 pmap_update_1pg(va); 387} 388 389/* 390 * determine if a page is managed (memory vs. device) 391 */ 392static __inline __pure int 393pmap_is_managed(pa) 394 vm_offset_t pa; 395{ 396 int i; 397 398 if (!pmap_initialized) 399 return 0; 400 401 for (i = 0; phys_avail[i + 1]; i += 2) { 402 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 403 return 1; 404 } 405 return 0; 406} 407 408/* 409 * This routine unholds page table pages, and if the hold count 410 * drops to zero, then it decrements the wire count. 411 */ 412static __inline int 413pmap_unwire_pte_hold(vm_page_t m) { 414 vm_page_unhold(m); 415 if (m->hold_count == 0) { 416 --m->wire_count; 417 if (m->wire_count == 0) { 418 --cnt.v_wire_count; 419 m->dirty = 0; 420 vm_page_deactivate(m); 421 } 422 return 1; 423 } 424 return 0; 425} 426 427#if !defined(PMAP_DIAGNOSTIC) 428__inline 429#endif 430int 431pmap_unuse_pt(pmap, va, mpte) 432 pmap_t pmap; 433 vm_offset_t va; 434 vm_page_t mpte; 435{ 436 if (va >= UPT_MIN_ADDRESS) 437 return 0; 438 439 if (mpte == NULL) { 440 vm_offset_t ptepa; 441 ptepa = ((vm_offset_t) *pmap_pde(pmap, va)) /* & PG_FRAME */; 442#if defined(PMAP_DIAGNOSTIC) 443 if (!ptepa) 444 panic("pmap_unuse_pt: pagetable page missing, va: 0x%x", va); 445#endif 446 mpte = PHYS_TO_VM_PAGE(ptepa); 447 } 448 449#if defined(PMAP_DIAGNOSTIC) 450 if (mpte->hold_count == 0) { 451 panic("pmap_unuse_pt: hold count < 0, va: 0x%x", va); 452 } 453#endif 454 455/* 456 * We don't free page-table-pages anymore because it can have a negative 457 * impact on perf at times. Now we just deactivate, and it'll get cleaned 458 * up if needed... Also, if the page ends up getting used, it will be 459 * brought back into the process address space by pmap_allocpte and be 460 * reactivated. 461 */ 462 return pmap_unwire_pte_hold(mpte); 463} 464 465/* 466 * Bootstrap the system enough to run with virtual memory. 467 * 468 * On the i386 this is called after mapping has already been enabled 469 * and just syncs the pmap module with what has already been done. 470 * [We can't call it easily with mapping off since the kernel is not 471 * mapped with PA == VA, hence we would have to relocate every address 472 * from the linked base (virtual) address "KERNBASE" to the actual 473 * (physical) address starting relative to 0] 474 */ 475void 476pmap_bootstrap(firstaddr, loadaddr) 477 vm_offset_t firstaddr; 478 vm_offset_t loadaddr; 479{ 480 vm_offset_t va; 481 pt_entry_t *pte; 482 483 avail_start = firstaddr; 484 485 /* 486 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 487 * large. It should instead be correctly calculated in locore.s and 488 * not based on 'first' (which is a physical address, not a virtual 489 * address, for the start of unused physical memory). The kernel 490 * page tables are NOT double mapped and thus should not be included 491 * in this calculation. 492 */ 493 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 494 virtual_end = VM_MAX_KERNEL_ADDRESS; 495 496 /* 497 * Initialize protection array. 498 */ 499 i386_protection_init(); 500 501 /* 502 * The kernel's pmap is statically allocated so we don't have to use 503 * pmap_create, which is unlikely to work correctly at this part of 504 * the boot sequence (XXX and which no longer exists). 505 */ 506 kernel_pmap = &kernel_pmap_store; 507 508 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + IdlePTD); 509 510 kernel_pmap->pm_count = 1; 511 nkpt = NKPT; 512 513 /* 514 * Reserve some special page table entries/VA space for temporary 515 * mapping of pages. 516 */ 517#define SYSMAP(c, p, v, n) \ 518 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 519 520 va = virtual_avail; 521 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 522 523 /* 524 * CMAP1/CMAP2 are used for zeroing and copying pages. 525 */ 526 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 527 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 528 529 /* 530 * ptmmap is used for reading arbitrary physical pages via /dev/mem. 531 */ 532 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 533 534 /* 535 * msgbufmap is used to map the system message buffer. 536 */ 537 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 1) 538 539 virtual_avail = va; 540 541 *(int *) CMAP1 = *(int *) CMAP2 = *(int *) PTD = 0; 542 pmap_update(); 543 544} 545 546/* 547 * Initialize the pmap module. 548 * Called by vm_init, to initialize any structures that the pmap 549 * system needs to map virtual memory. 550 * pmap_init has been enhanced to support in a fairly consistant 551 * way, discontiguous physical memory. 552 */ 553void 554pmap_init(phys_start, phys_end) 555 vm_offset_t phys_start, phys_end; 556{ 557 vm_offset_t addr; 558 vm_size_t npg, s; 559 int i; 560 561 /* 562 * calculate the number of pv_entries needed 563 */ 564 vm_first_phys = phys_avail[0]; 565 for (i = 0; phys_avail[i + 1]; i += 2); 566 npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 567 568 /* 569 * Allocate memory for random pmap data structures. Includes the 570 * pv_head_table. 571 */ 572 s = (vm_size_t) (sizeof(struct pv_entry *) * npg); 573 s = round_page(s); 574 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 575 pv_table = (pv_entry_t *) addr; 576 577 /* 578 * init the pv free list 579 */ 580 init_pv_entries(npg); 581 /* 582 * Now it is safe to enable pv_table recording. 583 */ 584 pmap_initialized = TRUE; 585} 586 587/* 588 * Used to map a range of physical addresses into kernel 589 * virtual address space. 590 * 591 * For now, VM is already on, we only need to map the 592 * specified memory. 593 */ 594vm_offset_t 595pmap_map(virt, start, end, prot) 596 vm_offset_t virt; 597 vm_offset_t start; 598 vm_offset_t end; 599 int prot; 600{ 601 while (start < end) { 602 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 603 virt += PAGE_SIZE; 604 start += PAGE_SIZE; 605 } 606 return (virt); 607} 608 609/* 610 * Initialize a preallocated and zeroed pmap structure, 611 * such as one in a vmspace structure. 612 */ 613void 614pmap_pinit(pmap) 615 register struct pmap *pmap; 616{ 617 vm_page_t ptdpg; 618 /* 619 * No need to allocate page table space yet but we do need a valid 620 * page directory table. 621 */ 622 623 if (pdstackptr > 0) { 624 --pdstackptr; 625 pmap->pm_pdir = 626 (pd_entry_t *)pdstack[pdstackptr]; 627 } else { 628 pmap->pm_pdir = 629 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 630 } 631 632 /* 633 * allocate object for the ptes 634 */ 635 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 636 637 /* 638 * allocate the page directory page 639 */ 640retry: 641 ptdpg = vm_page_alloc( pmap->pm_pteobj, PTDPTDI, VM_ALLOC_ZERO); 642 if (ptdpg == NULL) { 643 VM_WAIT; 644 goto retry; 645 } 646 vm_page_wire(ptdpg); 647 ptdpg->flags &= ~(PG_MAPPED|PG_BUSY); /* not mapped normally */ 648 ptdpg->valid = VM_PAGE_BITS_ALL; 649 650 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 651 if ((ptdpg->flags & PG_ZERO) == 0) 652 bzero(pmap->pm_pdir, PAGE_SIZE); 653 654 /* wire in kernel global address entries */ 655 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 656 657 /* install self-referential address mapping entry */ 658 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 659 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW; 660 661 pmap->pm_count = 1; 662} 663 664static int 665pmap_release_free_page(pmap, p) 666 struct pmap *pmap; 667 vm_page_t p; 668{ 669 int s; 670 /* 671 * This code optimizes the case of freeing non-busy 672 * page-table pages. Those pages are zero now, and 673 * might as well be placed directly into the zero queue. 674 */ 675 s = splvm(); 676 if (p->flags & PG_BUSY) { 677 p->flags |= PG_WANTED; 678 tsleep(p, PVM, "pmaprl", 0); 679 splx(s); 680 return 0; 681 } 682 683 pmap_remove_pte_mapping(VM_PAGE_TO_PHYS(p)); 684 685 if (p->hold_count) { 686#if defined(PMAP_DIAGNOSTIC) 687 panic("pmap_release: freeing held page table page"); 688#endif 689 /* 690 * HACK ALERT!!! 691 * If this failure happens, we must clear the page, because 692 * there is likely a mapping still valid. This condition 693 * is an error, but at least this zero operation will mitigate 694 * some Sig-11's or crashes, because this page is thought 695 * to be zero. This is a robustness fix, and not meant to 696 * be a long term work-around. 697 */ 698 pmap_zero_page(VM_PAGE_TO_PHYS(p)); 699 } 700 /* 701 * Page directory pages need to have the kernel 702 * stuff cleared, so they can go into the zero queue also. 703 */ 704 if (p->pindex == PTDPTDI) { 705 unsigned *pde = (unsigned *) pmap->pm_pdir; 706 bzero(pde + KPTDI, nkpt * PTESIZE); 707 pde[APTDPTDI] = 0; 708 pde[PTDPTDI] = 0; 709 pmap_kremove((vm_offset_t) pmap->pm_pdir); 710 } 711 712 vm_page_free_zero(p); 713 splx(s); 714 return 1; 715} 716 717/* 718 * Release any resources held by the given physical map. 719 * Called when a pmap initialized by pmap_pinit is being released. 720 * Should only be called if the map contains no valid mappings. 721 */ 722void 723pmap_release(pmap) 724 register struct pmap *pmap; 725{ 726 vm_page_t p,n,ptdpg; 727 vm_object_t object = pmap->pm_pteobj; 728 int s; 729 730 if (object->ref_count != 1) 731 panic("pmap_release: pteobj reference count != 1"); 732 733 ptdpg = NULL; 734retry: 735 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 736 n = TAILQ_NEXT(p, listq); 737 if (p->pindex == PTDPTDI) { 738 ptdpg = p; 739 continue; 740 } 741 if (!pmap_release_free_page(pmap, p)) 742 goto retry; 743 } 744 if (ptdpg == NULL) 745 panic("pmap_release: missing page table directory page"); 746 747 if (!pmap_release_free_page(pmap, ptdpg)) 748 goto retry; 749 750 vm_object_deallocate(object); 751 if (pdstackptr < PDSTACKMAX) { 752 pdstack[pdstackptr] = (vm_offset_t) pmap->pm_pdir; 753 ++pdstackptr; 754 } else { 755 kmem_free(kernel_map, (vm_offset_t) pmap->pm_pdir, PAGE_SIZE); 756 } 757} 758 759/* 760 * grow the number of kernel page table entries, if needed 761 */ 762 763void 764pmap_growkernel(vm_offset_t addr) 765{ 766 struct proc *p; 767 struct pmap *pmap; 768 int s; 769 770 s = splhigh(); 771 if (kernel_vm_end == 0) { 772 kernel_vm_end = KERNBASE; 773 nkpt = 0; 774 while (pdir_pde(PTD, kernel_vm_end)) { 775 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 776 ++nkpt; 777 } 778 } 779 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 780 while (kernel_vm_end < addr) { 781 if (pdir_pde(PTD, kernel_vm_end)) { 782 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 783 continue; 784 } 785 ++nkpt; 786 if (!nkpg) { 787 nkpg = vm_page_alloc(kernel_object, 0, VM_ALLOC_SYSTEM); 788 if (!nkpg) 789 panic("pmap_growkernel: no memory to grow kernel"); 790 vm_page_wire(nkpg); 791 vm_page_remove(nkpg); 792 pmap_zero_page(VM_PAGE_TO_PHYS(nkpg)); 793 } 794 pdir_pde(PTD, kernel_vm_end) = (pd_entry_t) (VM_PAGE_TO_PHYS(nkpg) | PG_V | PG_RW); 795 nkpg = NULL; 796 797 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 798 if (p->p_vmspace) { 799 pmap = &p->p_vmspace->vm_pmap; 800 *pmap_pde(pmap, kernel_vm_end) = pdir_pde(PTD, kernel_vm_end); 801 } 802 } 803 *pmap_pde(kernel_pmap, kernel_vm_end) = pdir_pde(PTD, kernel_vm_end); 804 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 805 } 806 splx(s); 807} 808 809/* 810 * Retire the given physical map from service. 811 * Should only be called if the map contains 812 * no valid mappings. 813 */ 814void 815pmap_destroy(pmap) 816 register pmap_t pmap; 817{ 818 int count; 819 820 if (pmap == NULL) 821 return; 822 823 count = --pmap->pm_count; 824 if (count == 0) { 825 pmap_release(pmap); 826 free((caddr_t) pmap, M_VMPMAP); 827 } 828} 829 830/* 831 * Add a reference to the specified pmap. 832 */ 833void 834pmap_reference(pmap) 835 pmap_t pmap; 836{ 837 if (pmap != NULL) { 838 pmap->pm_count++; 839 } 840} 841 842/* 843 * free the pv_entry back to the free list 844 */ 845static __inline void 846free_pv_entry(pv) 847 pv_entry_t pv; 848{ 849 ++pv_freelistcnt; 850 pv->pv_next = pv_freelist; 851 pv_freelist = pv; 852} 853 854/* 855 * get a new pv_entry, allocating a block from the system 856 * when needed. 857 * the memory allocation is performed bypassing the malloc code 858 * because of the possibility of allocations at interrupt time. 859 */ 860static __inline pv_entry_t 861get_pv_entry() 862{ 863 pv_entry_t tmp; 864 865 /* 866 * get more pv_entry pages if needed 867 */ 868 if (pv_freelistcnt < PV_FREELIST_MIN || pv_freelist == 0) { 869 pmap_alloc_pv_entry(); 870 } 871 /* 872 * get a pv_entry off of the free list 873 */ 874 --pv_freelistcnt; 875 tmp = pv_freelist; 876 pv_freelist = tmp->pv_next; 877 return tmp; 878} 879 880/* 881 * This *strange* allocation routine eliminates the possibility of a malloc 882 * failure (*FATAL*) for a pv_entry_t data structure. 883 * also -- this code is MUCH MUCH faster than the malloc equiv... 884 * We really need to do the slab allocator thingie here. 885 */ 886static void 887pmap_alloc_pv_entry() 888{ 889 /* 890 * do we have any pre-allocated map-pages left? 891 */ 892 if (npvvapg) { 893 vm_page_t m; 894 895 /* 896 * allocate a physical page out of the vm system 897 */ 898 m = vm_page_alloc(kernel_object, 899 OFF_TO_IDX(pvva - vm_map_min(kernel_map)), 900 VM_ALLOC_INTERRUPT); 901 if (m) { 902 int newentries; 903 int i; 904 pv_entry_t entry; 905 906 newentries = (PAGE_SIZE / sizeof(struct pv_entry)); 907 /* 908 * wire the page 909 */ 910 vm_page_wire(m); 911 m->flags &= ~PG_BUSY; 912 /* 913 * let the kernel see it 914 */ 915 pmap_kenter(pvva, VM_PAGE_TO_PHYS(m)); 916 917 entry = (pv_entry_t) pvva; 918 /* 919 * update the allocation pointers 920 */ 921 pvva += PAGE_SIZE; 922 --npvvapg; 923 924 /* 925 * free the entries into the free list 926 */ 927 for (i = 0; i < newentries; i++) { 928 free_pv_entry(entry); 929 entry++; 930 } 931 } 932 } 933 if (!pv_freelist) 934 panic("get_pv_entry: cannot get a pv_entry_t"); 935} 936 937/* 938 * init the pv_entry allocation system 939 */ 940#define PVSPERPAGE 64 941void 942init_pv_entries(npg) 943 int npg; 944{ 945 /* 946 * allocate enough kvm space for PVSPERPAGE entries per page (lots) 947 * kvm space is fairly cheap, be generous!!! (the system can panic if 948 * this is too small.) 949 */ 950 npvvapg = ((npg * PVSPERPAGE) * sizeof(struct pv_entry) 951 + PAGE_SIZE - 1) / PAGE_SIZE; 952 pvva = kmem_alloc_pageable(kernel_map, npvvapg * PAGE_SIZE); 953 /* 954 * get the first batch of entries 955 */ 956 pmap_alloc_pv_entry(); 957} 958 959/* 960 * If it is the first entry on the list, it is actually 961 * in the header and we must copy the following entry up 962 * to the header. Otherwise we must search the list for 963 * the entry. In either case we free the now unused entry. 964 */ 965static __inline int 966pmap_remove_entry(pmap, ppv, va) 967 struct pmap *pmap; 968 pv_entry_t *ppv; 969 vm_offset_t va; 970{ 971 pv_entry_t npv; 972 int s; 973 974 s = splvm(); 975 for (npv = *ppv; npv; (ppv = &npv->pv_next, npv = *ppv)) { 976 if (pmap == npv->pv_pmap && va == npv->pv_va) { 977 int rtval = pmap_unuse_pt(pmap, va, npv->pv_ptem); 978 *ppv = npv->pv_next; 979 free_pv_entry(npv); 980 splx(s); 981 return rtval; 982 } 983 } 984 splx(s); 985 return 0; 986} 987 988/* 989 * pmap_remove_pte: do the things to unmap a page in a process 990 */ 991static 992#if !defined(PMAP_DIAGNOSTIC) 993__inline 994#endif 995int 996pmap_remove_pte(pmap, ptq, va) 997 struct pmap *pmap; 998 unsigned *ptq; 999 vm_offset_t va; 1000{ 1001 unsigned oldpte; 1002 pv_entry_t *ppv; 1003 int rtval; 1004 1005 oldpte = *ptq; 1006 *ptq = 0; 1007 if (oldpte & PG_W) 1008 pmap->pm_stats.wired_count -= 1; 1009 pmap->pm_stats.resident_count -= 1; 1010 if (oldpte & PG_MANAGED) { 1011 if (oldpte & PG_M) { 1012#if defined(PMAP_DIAGNOSTIC) 1013 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1014 printf("pmap_remove: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, (int) oldpte); 1015 } 1016#endif 1017 1018 if (va < clean_sva || va >= clean_eva) { 1019 if ((va < UPT_MIN_ADDRESS) || (va >= UPT_MAX_ADDRESS)) 1020 PHYS_TO_VM_PAGE(oldpte)->dirty = VM_PAGE_BITS_ALL; 1021 } 1022 } 1023 ppv = pa_to_pvh(oldpte); 1024 rtval = pmap_remove_entry(pmap, ppv, va); 1025#if defined(notyet) 1026 if (*ppv == NULL) { 1027 PHYS_TO_VM_PAGE(oldpte)->flags &= ~PG_MAPPED; 1028 } 1029#endif 1030 return rtval; 1031 } else { 1032 return pmap_unuse_pt(pmap, va, NULL); 1033 } 1034 1035 return 0; 1036} 1037 1038/* 1039 * Remove a single page from a process address space 1040 */ 1041static __inline void 1042pmap_remove_page(pmap, va) 1043 struct pmap *pmap; 1044 register vm_offset_t va; 1045{ 1046 register unsigned *ptq; 1047 1048 /* 1049 * if there is no pte for this address, just skip it!!! 1050 */ 1051 if (*pmap_pde(pmap, va) == 0) { 1052 return; 1053 } 1054 1055 /* 1056 * get a local va for mappings for this pmap. 1057 */ 1058 ptq = get_ptbase(pmap) + i386_btop(va); 1059 if (*ptq) { 1060 (void) pmap_remove_pte(pmap, ptq, va); 1061 pmap_update_1pg(va); 1062 } 1063 return; 1064} 1065 1066/* 1067 * Remove the given range of addresses from the specified map. 1068 * 1069 * It is assumed that the start and end are properly 1070 * rounded to the page size. 1071 */ 1072void 1073pmap_remove(pmap, sva, eva) 1074 struct pmap *pmap; 1075 register vm_offset_t sva; 1076 register vm_offset_t eva; 1077{ 1078 register unsigned *ptbase; 1079 vm_offset_t pdnxt; 1080 vm_offset_t ptpaddr; 1081 vm_offset_t sindex, eindex; 1082 vm_page_t mpte; 1083 int anyvalid; 1084 1085 if (pmap == NULL) 1086 return; 1087 1088 /* 1089 * special handling of removing one page. a very 1090 * common operation and easy to short circuit some 1091 * code. 1092 */ 1093 if ((sva + PAGE_SIZE) == eva) { 1094 pmap_remove_page(pmap, sva); 1095 return; 1096 } 1097 1098 anyvalid = 0; 1099 1100 /* 1101 * Get a local virtual address for the mappings that are being 1102 * worked with. 1103 */ 1104 ptbase = get_ptbase(pmap); 1105 1106 sindex = i386_btop(sva); 1107 eindex = i386_btop(eva); 1108 1109 for (; sindex < eindex; sindex = pdnxt) { 1110 1111 /* 1112 * Calculate index for next page table. 1113 */ 1114 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1115 ptpaddr = (vm_offset_t) *pmap_pde(pmap, i386_ptob(sindex)); 1116 1117 /* 1118 * Weed out invalid mappings. Note: we assume that the page 1119 * directory table is always allocated, and in kernel virtual. 1120 */ 1121 if (ptpaddr == 0) 1122 continue; 1123 1124 if (sindex < i386_btop(UPT_MIN_ADDRESS)) { 1125 /* 1126 * get the vm_page_t for the page table page 1127 */ 1128 mpte = PHYS_TO_VM_PAGE(ptpaddr); 1129 1130 /* 1131 * if the pte isn't wired, just skip it. 1132 */ 1133 if (mpte->wire_count == 0) 1134 continue; 1135 } 1136 1137 /* 1138 * Limit our scan to either the end of the va represented 1139 * by the current page table page, or to the end of the 1140 * range being removed. 1141 */ 1142 if (pdnxt > eindex) { 1143 pdnxt = eindex; 1144 } 1145 1146 for ( ;sindex != pdnxt; sindex++) { 1147 vm_offset_t va; 1148 if (ptbase[sindex] == 0) { 1149 continue; 1150 } 1151 va = i386_ptob(sindex); 1152 anyvalid = 1; 1153 if (pmap_remove_pte(pmap, 1154 ptbase + sindex, va)) 1155 break; 1156 } 1157 } 1158 1159 if (anyvalid) { 1160 pmap_update(); 1161 } 1162} 1163 1164/* 1165 * Remove pte mapping, don't do everything that we would do 1166 * for normal pages because many things aren't necessary (like 1167 * pmap_update())... 1168 */ 1169void 1170pmap_remove_pte_mapping(pa) 1171 vm_offset_t pa; 1172{ 1173 register pv_entry_t pv, *ppv, npv; 1174 register unsigned *pte; 1175 vm_offset_t va; 1176 int anyvalid = 0; 1177 1178 ppv = pa_to_pvh(pa); 1179 1180 for (pv = *ppv; pv; pv=pv->pv_next) { 1181 unsigned tpte; 1182 struct pmap *pmap; 1183 1184 pmap = pv->pv_pmap; 1185 pte = get_ptbase(pmap) + i386_btop(pv->pv_va); 1186 if (tpte = *pte) { 1187 pmap->pm_stats.resident_count--; 1188 *pte = 0; 1189 if (tpte & PG_W) 1190 pmap->pm_stats.wired_count--; 1191 } 1192 } 1193 1194 for (pv = *ppv; pv; pv = npv) { 1195 npv = pv->pv_next; 1196 free_pv_entry(pv); 1197 } 1198 *ppv = NULL; 1199} 1200 1201/* 1202 * Routine: pmap_remove_all 1203 * Function: 1204 * Removes this physical page from 1205 * all physical maps in which it resides. 1206 * Reflects back modify bits to the pager. 1207 * 1208 * Notes: 1209 * Original versions of this routine were very 1210 * inefficient because they iteratively called 1211 * pmap_remove (slow...) 1212 */ 1213static void 1214pmap_remove_all(pa) 1215 vm_offset_t pa; 1216{ 1217 register pv_entry_t pv, *ppv, npv; 1218 register unsigned *pte, *ptbase; 1219 vm_offset_t va; 1220 vm_page_t m; 1221 int s; 1222 1223#if defined(PMAP_DIAGNOSTIC) 1224 /* 1225 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1226 * pages! 1227 */ 1228 if (!pmap_is_managed(pa)) { 1229 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%lx", pa); 1230 } 1231#endif 1232 1233 m = PHYS_TO_VM_PAGE(pa); 1234 ppv = pa_to_pvh(pa); 1235 1236 s = splvm(); 1237 for (pv = *ppv; pv; pv=pv->pv_next) { 1238 int tpte; 1239 struct pmap *pmap; 1240 1241 pmap = pv->pv_pmap; 1242 ptbase = get_ptbase(pmap); 1243 va = pv->pv_va; 1244 if (*pmap_pde(pmap, va) == 0) 1245 continue; 1246 pte = ptbase + i386_btop(va); 1247 if (tpte = ((int) *pte)) { 1248 pmap->pm_stats.resident_count--; 1249 *pte = 0; 1250 if (tpte & PG_W) 1251 pmap->pm_stats.wired_count--; 1252 /* 1253 * Update the vm_page_t clean and reference bits. 1254 */ 1255 if (tpte & PG_M) { 1256#if defined(PMAP_DIAGNOSTIC) 1257 if (pmap_nw_modified((pt_entry_t) tpte)) { 1258 printf("pmap_remove_all: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, tpte); 1259 } 1260#endif 1261 if ((va >= UPT_MIN_ADDRESS) && 1262 (va < UPT_MAX_ADDRESS)) 1263 continue; 1264 1265 if (va < clean_sva || va >= clean_eva) { 1266 m->dirty = VM_PAGE_BITS_ALL; 1267 } 1268 } 1269 } 1270 } 1271 1272 for (pv = *ppv; pv; pv = npv) { 1273 npv = pv->pv_next; 1274 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1275 free_pv_entry(pv); 1276 } 1277 *ppv = NULL; 1278 1279 splx(s); 1280} 1281 1282/* 1283 * Set the physical protection on the 1284 * specified range of this map as requested. 1285 */ 1286void 1287pmap_protect(pmap, sva, eva, prot) 1288 register pmap_t pmap; 1289 vm_offset_t sva, eva; 1290 vm_prot_t prot; 1291{ 1292 register unsigned *pte; 1293 register vm_offset_t va; 1294 register unsigned *ptbase; 1295 vm_offset_t pdnxt; 1296 vm_offset_t ptpaddr; 1297 vm_offset_t sindex, eindex; 1298 vm_page_t mpte; 1299 int anyvalid; 1300 1301 1302 if (pmap == NULL) 1303 return; 1304 1305 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1306 pmap_remove(pmap, sva, eva); 1307 return; 1308 } 1309 if (prot & VM_PROT_WRITE) 1310 return; 1311 1312 anyvalid = 0; 1313 1314 ptbase = get_ptbase(pmap); 1315 1316 sindex = i386_btop(sva); 1317 eindex = i386_btop(eva); 1318 1319 for (; sindex < eindex; sindex = pdnxt) { 1320 1321 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1322 ptpaddr = (vm_offset_t) *pmap_pde(pmap, i386_ptob(sindex)); 1323 1324 /* 1325 * Weed out invalid mappings. Note: we assume that the page 1326 * directory table is always allocated, and in kernel virtual. 1327 */ 1328 if (ptpaddr == 0) 1329 continue; 1330 1331 /* 1332 * Don't look at kernel page table pages 1333 */ 1334 if (sindex < i386_btop(UPT_MIN_ADDRESS)) { 1335 mpte = PHYS_TO_VM_PAGE(ptpaddr); 1336 1337 if (mpte->wire_count == 0) 1338 continue; 1339 } 1340 1341 if (pdnxt > eindex) { 1342 pdnxt = eindex; 1343 } 1344 1345 for (; sindex != pdnxt; sindex++) { 1346 1347 unsigned pbits = ptbase[sindex]; 1348 1349 if (pbits & PG_RW) { 1350 if (pbits & PG_M) { 1351 vm_page_t m = PHYS_TO_VM_PAGE(pbits); 1352 m->dirty = VM_PAGE_BITS_ALL; 1353 } 1354 ptbase[sindex] = pbits & ~(PG_M|PG_RW); 1355 anyvalid = 1; 1356 } 1357 } 1358 } 1359 if (anyvalid) 1360 pmap_update(); 1361} 1362 1363/* 1364 * Create a pv entry for page at pa for 1365 * (pmap, va). 1366 */ 1367static __inline void 1368pmap_insert_entry(pmap, va, mpte, pa) 1369 pmap_t pmap; 1370 vm_offset_t va; 1371 vm_page_t mpte; 1372 vm_offset_t pa; 1373{ 1374 1375 int s; 1376 pv_entry_t *ppv, pv; 1377 1378 s = splvm(); 1379 pv = get_pv_entry(); 1380 pv->pv_va = va; 1381 pv->pv_pmap = pmap; 1382 pv->pv_ptem = mpte; 1383 1384 ppv = pa_to_pvh(pa); 1385 if (*ppv) 1386 pv->pv_next = *ppv; 1387 else 1388 pv->pv_next = NULL; 1389 *ppv = pv; 1390 splx(s); 1391} 1392 1393/* 1394 * this routine is called if the page table page is not 1395 * mapped correctly. 1396 */ 1397static vm_page_t 1398_pmap_allocpte(pmap, va, ptepindex) 1399 pmap_t pmap; 1400 vm_offset_t va; 1401 int ptepindex; 1402{ 1403 vm_offset_t pteva, ptepa; 1404 vm_page_t m; 1405 int s; 1406 1407 /* 1408 * Find or fabricate a new pagetable page 1409 */ 1410retry: 1411 m = vm_page_lookup(pmap->pm_pteobj, ptepindex); 1412 if (m == NULL) { 1413 m = vm_page_alloc(pmap->pm_pteobj, ptepindex, VM_ALLOC_ZERO); 1414 if (m == NULL) { 1415 VM_WAIT; 1416 goto retry; 1417 } 1418 if ((m->flags & PG_ZERO) == 0) 1419 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 1420 m->flags &= ~(PG_ZERO|PG_BUSY); 1421 m->valid = VM_PAGE_BITS_ALL; 1422 } else { 1423 if ((m->flags & PG_BUSY) || m->busy) { 1424 m->flags |= PG_WANTED; 1425 tsleep(m, PVM, "ptewai", 0); 1426 goto retry; 1427 } 1428 } 1429 1430 /* 1431 * mark the object writeable 1432 */ 1433 pmap->pm_pteobj->flags |= OBJ_WRITEABLE; 1434 1435 if (m->hold_count == 0) { 1436 s = splvm(); 1437 vm_page_unqueue(m); 1438 splx(s); 1439 ++m->wire_count; 1440 ++cnt.v_wire_count; 1441 } 1442 1443 /* 1444 * Increment the hold count for the page table page 1445 * (denoting a new mapping.) 1446 */ 1447 ++m->hold_count; 1448 1449 /* 1450 * Map the pagetable page into the process address space, if 1451 * it isn't already there. 1452 */ 1453 pteva = ((vm_offset_t) vtopte(va)) & PG_FRAME; 1454 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1455 if (ptepa == 0) { 1456 pv_entry_t pv, *ppv; 1457 1458 pmap->pm_stats.resident_count++; 1459 1460 s = splvm(); 1461 pv = get_pv_entry(); 1462 1463 pv->pv_va = pteva; 1464 pv->pv_pmap = pmap; 1465 pv->pv_next = NULL; 1466 pv->pv_ptem = NULL; 1467 1468 ptepa = VM_PAGE_TO_PHYS(m); 1469 ppv = pa_to_pvh(ptepa); 1470#if defined(PMAP_DIAGNOSTIC) 1471 if (*ppv) 1472 panic("pmap_allocpte: page is already mapped"); 1473#endif 1474 *ppv = pv; 1475 splx(s); 1476 pmap_update_1pg(pteva); 1477 } else { 1478#if defined(PMAP_DIAGNOSTIC) 1479 if (VM_PAGE_TO_PHYS(m) != (ptepa & PG_FRAME)) 1480 panic("pmap_allocpte: mismatch"); 1481#endif 1482 } 1483 pmap->pm_pdir[ptepindex] = 1484 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_MANAGED); 1485 m->flags |= PG_MAPPED; 1486 return m; 1487} 1488 1489static __inline vm_page_t 1490pmap_allocpte(pmap, va) 1491 pmap_t pmap; 1492 vm_offset_t va; 1493{ 1494 int ptepindex; 1495 vm_offset_t ptepa; 1496 vm_page_t m; 1497 1498 /* 1499 * Calculate pagetable page index 1500 */ 1501 ptepindex = va >> PDRSHIFT; 1502 1503 /* 1504 * Get the page directory entry 1505 */ 1506 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1507 1508 /* 1509 * If the page table page is mapped, we just increment the 1510 * hold count, and activate it. 1511 */ 1512 if ((ptepa & (PG_RW|PG_U|PG_V)) == (PG_RW|PG_U|PG_V)) { 1513 m = PHYS_TO_VM_PAGE(ptepa); 1514 if (m->hold_count == 0) { 1515 int s = splvm(); 1516 vm_page_unqueue(m); 1517 splx(s); 1518 ++m->wire_count; 1519 ++cnt.v_wire_count; 1520 } 1521 ++m->hold_count; 1522 return m; 1523 } 1524 return _pmap_allocpte(pmap, va, ptepindex); 1525} 1526 1527/* 1528 * Insert the given physical page (p) at 1529 * the specified virtual address (v) in the 1530 * target physical map with the protection requested. 1531 * 1532 * If specified, the page will be wired down, meaning 1533 * that the related pte can not be reclaimed. 1534 * 1535 * NB: This is the only routine which MAY NOT lazy-evaluate 1536 * or lose information. That is, this routine must actually 1537 * insert this page into the given map NOW. 1538 */ 1539void 1540pmap_enter(pmap, va, pa, prot, wired) 1541 register pmap_t pmap; 1542 vm_offset_t va; 1543 register vm_offset_t pa; 1544 vm_prot_t prot; 1545 boolean_t wired; 1546{ 1547 register unsigned *pte; 1548 vm_offset_t opa; 1549 vm_offset_t origpte, newpte; 1550 vm_page_t mpte; 1551 1552 if (pmap == NULL) 1553 return; 1554 1555 va &= PG_FRAME; 1556 if (va > VM_MAX_KERNEL_ADDRESS) 1557 panic("pmap_enter: toobig"); 1558 1559 mpte = NULL; 1560 /* 1561 * In the case that a page table page is not 1562 * resident, we are creating it here. 1563 */ 1564 if (va < UPT_MIN_ADDRESS) 1565 mpte = pmap_allocpte(pmap, va); 1566 1567 pte = pmap_pte(pmap, va); 1568 /* 1569 * Page Directory table entry not valid, we need a new PT page 1570 */ 1571 if (pte == NULL) { 1572 printf("kernel page directory invalid pdir=%p, va=0x%lx\n", 1573 pmap->pm_pdir[PTDPTDI], va); 1574 panic("invalid kernel page directory"); 1575 } 1576 1577 origpte = *(vm_offset_t *)pte; 1578 pa &= PG_FRAME; 1579 opa = origpte & PG_FRAME; 1580 1581 /* 1582 * Mapping has not changed, must be protection or wiring change. 1583 */ 1584 if (opa == pa) { 1585 /* 1586 * Wiring change, just update stats. We don't worry about 1587 * wiring PT pages as they remain resident as long as there 1588 * are valid mappings in them. Hence, if a user page is wired, 1589 * the PT page will be also. 1590 */ 1591 if (wired && ((origpte & PG_W) == 0)) 1592 pmap->pm_stats.wired_count++; 1593 else if (!wired && (origpte & PG_W)) 1594 pmap->pm_stats.wired_count--; 1595 1596#if defined(PMAP_DIAGNOSTIC) 1597 if (pmap_nw_modified((pt_entry_t) origpte)) { 1598 printf("pmap_enter: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, origpte); 1599 } 1600#endif 1601 1602 /* 1603 * We might be turning off write access to the page, 1604 * so we go ahead and sense modify status. 1605 */ 1606 if (origpte & PG_MANAGED) { 1607 vm_page_t m; 1608 if (origpte & PG_M) { 1609 m = PHYS_TO_VM_PAGE(pa); 1610 m->dirty = VM_PAGE_BITS_ALL; 1611 } 1612 pa |= PG_MANAGED; 1613 } 1614 1615 if (mpte) 1616 --mpte->hold_count; 1617 1618 goto validate; 1619 } 1620 /* 1621 * Mapping has changed, invalidate old range and fall through to 1622 * handle validating new mapping. 1623 */ 1624 if (opa) 1625 (void) pmap_remove_pte(pmap, pte, va); 1626 1627 /* 1628 * Enter on the PV list if part of our managed memory Note that we 1629 * raise IPL while manipulating pv_table since pmap_enter can be 1630 * called at interrupt time. 1631 */ 1632 if (pmap_is_managed(pa)) { 1633 pmap_insert_entry(pmap, va, mpte, pa); 1634 pa |= PG_MANAGED; 1635 } 1636 1637 /* 1638 * Increment counters 1639 */ 1640 pmap->pm_stats.resident_count++; 1641 if (wired) 1642 pmap->pm_stats.wired_count++; 1643 1644validate: 1645 /* 1646 * Now validate mapping with desired protection/wiring. 1647 */ 1648 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 1649 1650 if (wired) 1651 newpte |= PG_W; 1652 if (va < UPT_MIN_ADDRESS) 1653 newpte |= PG_U; 1654 1655 /* 1656 * if the mapping or permission bits are different, we need 1657 * to update the pte. 1658 */ 1659 if ((origpte & ~(PG_M|PG_A)) != newpte) { 1660 *pte = newpte; 1661 if (origpte) 1662 pmap_update_1pg(va); 1663 } 1664} 1665 1666/* 1667 * this code makes some *MAJOR* assumptions: 1668 * 1. Current pmap & pmap exists. 1669 * 2. Not wired. 1670 * 3. Read access. 1671 * 4. No page table pages. 1672 * 5. Tlbflush is deferred to calling procedure. 1673 * 6. Page IS managed. 1674 * but is *MUCH* faster than pmap_enter... 1675 */ 1676 1677static void 1678pmap_enter_quick(pmap, va, pa) 1679 register pmap_t pmap; 1680 vm_offset_t va; 1681 register vm_offset_t pa; 1682{ 1683 register unsigned *pte; 1684 vm_page_t mpte; 1685 1686 mpte = NULL; 1687 /* 1688 * In the case that a page table page is not 1689 * resident, we are creating it here. 1690 */ 1691 if (va < UPT_MIN_ADDRESS) 1692 mpte = pmap_allocpte(pmap, va); 1693 1694 pte = (unsigned *)vtopte(va); 1695 if (*pte) 1696 (void) pmap_remove_pte(pmap, pte, va); 1697 1698 /* 1699 * Enter on the PV list if part of our managed memory Note that we 1700 * raise IPL while manipulating pv_table since pmap_enter can be 1701 * called at interrupt time. 1702 */ 1703 pmap_insert_entry(pmap, va, mpte, pa); 1704 1705 /* 1706 * Increment counters 1707 */ 1708 pmap->pm_stats.resident_count++; 1709 1710 /* 1711 * Now validate mapping with RO protection 1712 */ 1713 *pte = pa | PG_V | PG_U | PG_MANAGED; 1714 1715 return; 1716} 1717 1718#define MAX_INIT_PT (96) 1719/* 1720 * pmap_object_init_pt preloads the ptes for a given object 1721 * into the specified pmap. This eliminates the blast of soft 1722 * faults on process startup and immediately after an mmap. 1723 */ 1724void 1725pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 1726 pmap_t pmap; 1727 vm_offset_t addr; 1728 vm_object_t object; 1729 vm_pindex_t pindex; 1730 vm_size_t size; 1731 int limit; 1732{ 1733 vm_offset_t tmpidx; 1734 int psize; 1735 vm_page_t p; 1736 int objpgs; 1737 1738 psize = (size >> PAGE_SHIFT); 1739 1740 if (!pmap || (object->type != OBJT_VNODE) || 1741 (limit && (psize > MAX_INIT_PT) && 1742 (object->resident_page_count > MAX_INIT_PT))) { 1743 return; 1744 } 1745 1746 /* 1747 * if we are processing a major portion of the object, then scan the 1748 * entire thing. 1749 */ 1750 if (psize > (object->size >> 2)) { 1751 objpgs = psize; 1752 1753 for (p = TAILQ_FIRST(&object->memq); 1754 ((objpgs > 0) && (p != NULL)); 1755 p = TAILQ_NEXT(p, listq)) { 1756 1757 tmpidx = p->pindex; 1758 if (tmpidx < pindex) { 1759 continue; 1760 } 1761 tmpidx -= pindex; 1762 if (tmpidx >= psize) { 1763 continue; 1764 } 1765 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 1766 (p->busy == 0) && 1767 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 1768 if (p->queue == PQ_CACHE) 1769 vm_page_deactivate(p); 1770 p->flags |= PG_BUSY; 1771 pmap_enter_quick(pmap, 1772 addr + (tmpidx << PAGE_SHIFT), 1773 VM_PAGE_TO_PHYS(p)); 1774 p->flags |= PG_MAPPED; 1775 PAGE_WAKEUP(p); 1776 } 1777 objpgs -= 1; 1778 } 1779 } else { 1780 /* 1781 * else lookup the pages one-by-one. 1782 */ 1783 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 1784 p = vm_page_lookup(object, tmpidx + pindex); 1785 if (p && 1786 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 1787 (p->busy == 0) && 1788 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 1789 if (p->queue == PQ_CACHE) 1790 vm_page_deactivate(p); 1791 p->flags |= PG_BUSY; 1792 pmap_enter_quick(pmap, 1793 addr + (tmpidx << PAGE_SHIFT), 1794 VM_PAGE_TO_PHYS(p)); 1795 p->flags |= PG_MAPPED; 1796 PAGE_WAKEUP(p); 1797 } 1798 } 1799 } 1800 return; 1801} 1802 1803/* 1804 * pmap_prefault provides a quick way of clustering 1805 * pagefaults into a processes address space. It is a "cousin" 1806 * of pmap_object_init_pt, except it runs at page fault time instead 1807 * of mmap time. 1808 */ 1809#define PFBAK 2 1810#define PFFOR 2 1811#define PAGEORDER_SIZE (PFBAK+PFFOR) 1812 1813static int pmap_prefault_pageorder[] = { 1814 -PAGE_SIZE, PAGE_SIZE, -2 * PAGE_SIZE, 2 * PAGE_SIZE 1815}; 1816 1817void 1818pmap_prefault(pmap, addra, entry, object) 1819 pmap_t pmap; 1820 vm_offset_t addra; 1821 vm_map_entry_t entry; 1822 vm_object_t object; 1823{ 1824 int i; 1825 vm_offset_t starta; 1826 vm_offset_t addr; 1827 vm_pindex_t pindex; 1828 vm_page_t m; 1829 1830 if (entry->object.vm_object != object) 1831 return; 1832 1833 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) 1834 return; 1835 1836 starta = addra - PFBAK * PAGE_SIZE; 1837 if (starta < entry->start) { 1838 starta = entry->start; 1839 } else if (starta > addra) { 1840 starta = 0; 1841 } 1842 1843 for (i = 0; i < PAGEORDER_SIZE; i++) { 1844 vm_object_t lobject; 1845 unsigned *pte; 1846 1847 addr = addra + pmap_prefault_pageorder[i]; 1848 if (addr < starta || addr >= entry->end) 1849 continue; 1850 1851 if ((*pmap_pde(pmap, addr)) == NULL) 1852 continue; 1853 1854 pte = (unsigned *) vtopte(addr); 1855 if (*pte) 1856 continue; 1857 1858 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 1859 lobject = object; 1860 for (m = vm_page_lookup(lobject, pindex); 1861 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 1862 lobject = lobject->backing_object) { 1863 if (lobject->backing_object_offset & PAGE_MASK) 1864 break; 1865 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 1866 m = vm_page_lookup(lobject->backing_object, pindex); 1867 } 1868 1869 /* 1870 * give-up when a page is not in memory 1871 */ 1872 if (m == NULL) 1873 break; 1874 1875 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 1876 (m->busy == 0) && 1877 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 1878 1879 if (m->queue == PQ_CACHE) { 1880 vm_page_deactivate(m); 1881 } 1882 m->flags |= PG_BUSY; 1883 pmap_enter_quick(pmap, addr, VM_PAGE_TO_PHYS(m)); 1884 m->flags |= PG_MAPPED; 1885 PAGE_WAKEUP(m); 1886 } 1887 } 1888} 1889 1890/* 1891 * Routine: pmap_change_wiring 1892 * Function: Change the wiring attribute for a map/virtual-address 1893 * pair. 1894 * In/out conditions: 1895 * The mapping must already exist in the pmap. 1896 */ 1897void 1898pmap_change_wiring(pmap, va, wired) 1899 register pmap_t pmap; 1900 vm_offset_t va; 1901 boolean_t wired; 1902{ 1903 register unsigned *pte; 1904 1905 if (pmap == NULL) 1906 return; 1907 1908 pte = pmap_pte(pmap, va); 1909 1910 if (wired && !pmap_pte_w(pte)) 1911 pmap->pm_stats.wired_count++; 1912 else if (!wired && pmap_pte_w(pte)) 1913 pmap->pm_stats.wired_count--; 1914 1915 /* 1916 * Wiring is not a hardware characteristic so there is no need to 1917 * invalidate TLB. 1918 */ 1919 pmap_pte_set_w(pte, wired); 1920} 1921 1922 1923 1924/* 1925 * Copy the range specified by src_addr/len 1926 * from the source map to the range dst_addr/len 1927 * in the destination map. 1928 * 1929 * This routine is only advisory and need not do anything. 1930 */ 1931void 1932pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 1933 pmap_t dst_pmap, src_pmap; 1934 vm_offset_t dst_addr; 1935 vm_size_t len; 1936 vm_offset_t src_addr; 1937{ 1938 vm_offset_t addr; 1939 vm_offset_t end_addr = src_addr + len; 1940 vm_offset_t pdnxt; 1941 unsigned src_frame, dst_frame; 1942 1943 if (dst_addr != src_addr) 1944 return; 1945 1946 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 1947 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) 1948 return; 1949 1950 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 1951 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 1952 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 1953 pmap_update(); 1954 } 1955 1956 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 1957 unsigned *src_pte, *dst_pte; 1958 vm_page_t dstmpte, srcmpte; 1959 vm_offset_t srcptepaddr; 1960 1961 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 1962 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[addr >> PDRSHIFT]; 1963 if (srcptepaddr == 0) { 1964 continue; 1965 } 1966 1967 srcmpte = PHYS_TO_VM_PAGE(srcptepaddr); 1968 if (srcmpte->hold_count == 0) 1969 continue; 1970 1971 if (pdnxt > end_addr) 1972 pdnxt = end_addr; 1973 1974 src_pte = (unsigned *) vtopte(addr); 1975 dst_pte = (unsigned *) avtopte(addr); 1976 while (addr < pdnxt) { 1977 unsigned ptetemp; 1978 ptetemp = *src_pte; 1979 /* 1980 * we only virtual copy managed pages 1981 */ 1982 if ((ptetemp & PG_MANAGED) != 0) { 1983 /* 1984 * We have to check after allocpte for the 1985 * pte still being around... allocpte can 1986 * block. 1987 */ 1988 dstmpte = pmap_allocpte(dst_pmap, addr); 1989 if (ptetemp = *src_pte) { 1990 /* 1991 * Simply clear the modified and accessed (referenced) 1992 * bits. 1993 */ 1994 *dst_pte = ptetemp & ~(PG_M|PG_A); 1995 dst_pmap->pm_stats.resident_count++; 1996 pmap_insert_entry(dst_pmap, addr, dstmpte, 1997 (ptetemp & PG_FRAME)); 1998 } else { 1999 pmap_unwire_pte_hold(dstmpte); 2000 } 2001 if (dstmpte->hold_count >= srcmpte->hold_count) 2002 break; 2003 } 2004 addr += PAGE_SIZE; 2005 ++src_pte; 2006 ++dst_pte; 2007 } 2008 } 2009} 2010 2011/* 2012 * Routine: pmap_kernel 2013 * Function: 2014 * Returns the physical map handle for the kernel. 2015 */ 2016pmap_t 2017pmap_kernel() 2018{ 2019 return (kernel_pmap); 2020} 2021 2022/* 2023 * pmap_zero_page zeros the specified (machine independent) 2024 * page by mapping the page into virtual memory and using 2025 * bzero to clear its contents, one machine dependent page 2026 * at a time. 2027 */ 2028void 2029pmap_zero_page(phys) 2030 vm_offset_t phys; 2031{ 2032 if (*(int *) CMAP2) 2033 panic("pmap_zero_page: CMAP busy"); 2034 2035 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME); 2036 bzero(CADDR2, PAGE_SIZE); 2037 *(int *) CMAP2 = 0; 2038 pmap_update_1pg((vm_offset_t) CADDR2); 2039} 2040 2041/* 2042 * pmap_copy_page copies the specified (machine independent) 2043 * page by mapping the page into virtual memory and using 2044 * bcopy to copy the page, one machine dependent page at a 2045 * time. 2046 */ 2047void 2048pmap_copy_page(src, dst) 2049 vm_offset_t src; 2050 vm_offset_t dst; 2051{ 2052 if (*(int *) CMAP1 || *(int *) CMAP2) 2053 panic("pmap_copy_page: CMAP busy"); 2054 2055 *(int *) CMAP1 = PG_V | PG_RW | (src & PG_FRAME); 2056 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME); 2057 2058#if __GNUC__ > 1 2059 memcpy(CADDR2, CADDR1, PAGE_SIZE); 2060#else 2061 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2062#endif 2063 *(int *) CMAP1 = 0; 2064 *(int *) CMAP2 = 0; 2065 pmap_update_2pg( (vm_offset_t) CADDR1, (vm_offset_t) CADDR2); 2066} 2067 2068 2069/* 2070 * Routine: pmap_pageable 2071 * Function: 2072 * Make the specified pages (by pmap, offset) 2073 * pageable (or not) as requested. 2074 * 2075 * A page which is not pageable may not take 2076 * a fault; therefore, its page table entry 2077 * must remain valid for the duration. 2078 * 2079 * This routine is merely advisory; pmap_enter 2080 * will specify that these pages are to be wired 2081 * down (or not) as appropriate. 2082 */ 2083void 2084pmap_pageable(pmap, sva, eva, pageable) 2085 pmap_t pmap; 2086 vm_offset_t sva, eva; 2087 boolean_t pageable; 2088{ 2089} 2090 2091/* 2092 * this routine returns true if a physical page resides 2093 * in the given pmap. 2094 */ 2095boolean_t 2096pmap_page_exists(pmap, pa) 2097 pmap_t pmap; 2098 vm_offset_t pa; 2099{ 2100 register pv_entry_t *ppv, pv; 2101 int s; 2102 2103 if (!pmap_is_managed(pa)) 2104 return FALSE; 2105 2106 s = splvm(); 2107 2108 ppv = pa_to_pvh(pa); 2109 /* 2110 * Not found, check current mappings returning immediately if found. 2111 */ 2112 for (pv = *ppv; pv; pv = pv->pv_next) { 2113 if (pv->pv_pmap == pmap) { 2114 splx(s); 2115 return TRUE; 2116 } 2117 } 2118 splx(s); 2119 return (FALSE); 2120} 2121 2122/* 2123 * pmap_testbit tests bits in pte's 2124 * note that the testbit/changebit routines are inline, 2125 * and a lot of things compile-time evaluate. 2126 */ 2127static __inline boolean_t 2128pmap_testbit(pa, bit) 2129 register vm_offset_t pa; 2130 int bit; 2131{ 2132 register pv_entry_t *ppv, pv; 2133 unsigned *pte; 2134 int s; 2135 2136 if (!pmap_is_managed(pa)) 2137 return FALSE; 2138 2139 s = splvm(); 2140 2141 ppv = pa_to_pvh(pa); 2142 /* 2143 * Not found, check current mappings returning immediately if found. 2144 */ 2145 for (pv = *ppv ;pv; pv = pv->pv_next) { 2146 /* 2147 * if the bit being tested is the modified bit, then 2148 * mark UPAGES as always modified, and ptes as never 2149 * modified. 2150 */ 2151 if (bit & (PG_A|PG_M)) { 2152 if ((pv->pv_va >= UPT_MIN_ADDRESS) && 2153 (pv->pv_va < UPT_MAX_ADDRESS)) { 2154 continue; 2155 } 2156 if ((pv->pv_va >= clean_sva) && 2157 (pv->pv_va < clean_eva)) { 2158 continue; 2159 } 2160 } 2161 if (!pv->pv_pmap) { 2162#if defined(PMAP_DIAGNOSTIC) 2163 printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va); 2164#endif 2165 continue; 2166 } 2167 pte = pmap_pte(pv->pv_pmap, pv->pv_va); 2168 if (pte == NULL) 2169 continue; 2170 if ((int) *pte & bit) { 2171 splx(s); 2172 return TRUE; 2173 } 2174 } 2175 splx(s); 2176 return (FALSE); 2177} 2178 2179/* 2180 * this routine is used to modify bits in ptes 2181 */ 2182static __inline void 2183pmap_changebit(pa, bit, setem) 2184 vm_offset_t pa; 2185 int bit; 2186 boolean_t setem; 2187{ 2188 register pv_entry_t pv, *ppv; 2189 register unsigned *pte; 2190 vm_offset_t va; 2191 int changed; 2192 int s; 2193 2194 if (!pmap_is_managed(pa)) 2195 return; 2196 2197 s = splvm(); 2198 2199 changed = 0; 2200 ppv = pa_to_pvh(pa); 2201 /* 2202 * Loop over all current mappings setting/clearing as appropos If 2203 * setting RO do we need to clear the VAC? 2204 */ 2205 for ( pv = *ppv; pv; pv = pv->pv_next) { 2206 va = pv->pv_va; 2207 2208 /* 2209 * don't write protect pager mappings 2210 */ 2211 if (!setem && (bit == PG_RW)) { 2212 if (va >= clean_sva && va < clean_eva) 2213 continue; 2214 } 2215 if (!pv->pv_pmap) { 2216#if defined(PMAP_DIAGNOSTIC) 2217 printf("Null pmap (cb) at va: 0x%lx\n", va); 2218#endif 2219 continue; 2220 } 2221 2222 pte = pmap_pte(pv->pv_pmap, va); 2223 if (pte == NULL) 2224 continue; 2225 if (setem) { 2226 *(int *)pte |= bit; 2227 changed = 1; 2228 } else { 2229 vm_offset_t pbits = *(vm_offset_t *)pte; 2230 if (pbits & bit) 2231 changed = 1; 2232 if (bit == PG_RW) { 2233 if (pbits & PG_M) { 2234 vm_page_t m; 2235 vm_offset_t pa = pbits & PG_FRAME; 2236 m = PHYS_TO_VM_PAGE(pa); 2237 m->dirty = VM_PAGE_BITS_ALL; 2238 } 2239 *(int *)pte = pbits & ~(PG_M|PG_RW); 2240 } else { 2241 *(int *)pte = pbits & ~bit; 2242 } 2243 } 2244 } 2245 splx(s); 2246 if (changed) 2247 pmap_update(); 2248} 2249 2250/* 2251 * pmap_page_protect: 2252 * 2253 * Lower the permission for all mappings to a given page. 2254 */ 2255void 2256pmap_page_protect(phys, prot) 2257 vm_offset_t phys; 2258 vm_prot_t prot; 2259{ 2260 if ((prot & VM_PROT_WRITE) == 0) { 2261 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) 2262 pmap_changebit(phys, PG_RW, FALSE); 2263 else { 2264 pmap_remove_all(phys); 2265 pmap_update(); 2266 } 2267 } 2268} 2269 2270vm_offset_t 2271pmap_phys_address(ppn) 2272 int ppn; 2273{ 2274 return (i386_ptob(ppn)); 2275} 2276 2277/* 2278 * pmap_is_referenced: 2279 * 2280 * Return whether or not the specified physical page was referenced 2281 * by any physical maps. 2282 */ 2283boolean_t 2284pmap_is_referenced(vm_offset_t pa) 2285{ 2286 return pmap_testbit((pa), PG_A); 2287} 2288 2289/* 2290 * pmap_is_modified: 2291 * 2292 * Return whether or not the specified physical page was modified 2293 * in any physical maps. 2294 */ 2295boolean_t 2296pmap_is_modified(vm_offset_t pa) 2297{ 2298 return pmap_testbit((pa), PG_M); 2299} 2300 2301/* 2302 * Clear the modify bits on the specified physical page. 2303 */ 2304void 2305pmap_clear_modify(vm_offset_t pa) 2306{ 2307 pmap_changebit((pa), PG_M, FALSE); 2308} 2309 2310/* 2311 * pmap_clear_reference: 2312 * 2313 * Clear the reference bit on the specified physical page. 2314 */ 2315void 2316pmap_clear_reference(vm_offset_t pa) 2317{ 2318 pmap_changebit((pa), PG_A, FALSE); 2319} 2320 2321/* 2322 * Miscellaneous support routines follow 2323 */ 2324 2325static void 2326i386_protection_init() 2327{ 2328 register int *kp, prot; 2329 2330 kp = protection_codes; 2331 for (prot = 0; prot < 8; prot++) { 2332 switch (prot) { 2333 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 2334 /* 2335 * Read access is also 0. There isn't any execute bit, 2336 * so just make it readable. 2337 */ 2338 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 2339 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 2340 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 2341 *kp++ = 0; 2342 break; 2343 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 2344 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 2345 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 2346 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 2347 *kp++ = PG_RW; 2348 break; 2349 } 2350 } 2351} 2352 2353/* 2354 * Map a set of physical memory pages into the kernel virtual 2355 * address space. Return a pointer to where it is mapped. This 2356 * routine is intended to be used for mapping device memory, 2357 * NOT real memory. The non-cacheable bits are set on each 2358 * mapped page. 2359 */ 2360void * 2361pmap_mapdev(pa, size) 2362 vm_offset_t pa; 2363 vm_size_t size; 2364{ 2365 vm_offset_t va, tmpva; 2366 unsigned *pte; 2367 2368 size = roundup(size, PAGE_SIZE); 2369 2370 va = kmem_alloc_pageable(kernel_map, size); 2371 if (!va) 2372 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 2373 2374 pa = pa & PG_FRAME; 2375 for (tmpva = va; size > 0;) { 2376 pte = (unsigned *)vtopte(tmpva); 2377 *pte = pa | PG_RW | PG_V | PG_N; 2378 size -= PAGE_SIZE; 2379 tmpva += PAGE_SIZE; 2380 pa += PAGE_SIZE; 2381 } 2382 pmap_update(); 2383 2384 return ((void *) va); 2385} 2386 2387int 2388pmap_mincore(pmap, addr) 2389 pmap_t pmap; 2390 vm_offset_t addr; 2391{ 2392 2393 unsigned *ptep, pte; 2394 int val = 0; 2395 2396 ptep = pmap_pte(pmap, addr); 2397 if (ptep == 0) { 2398 return 0; 2399 } 2400 2401 if ((pte = *ptep)) { 2402 vm_offset_t pa; 2403 val = MINCORE_INCORE; 2404 pa = pte & PG_FRAME; 2405 2406 /* 2407 * Modified by us 2408 */ 2409 if (pte & PG_M) 2410 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 2411 /* 2412 * Modified by someone 2413 */ 2414 else if (PHYS_TO_VM_PAGE(pa)->dirty || 2415 pmap_is_modified(pa)) 2416 val |= MINCORE_MODIFIED_OTHER; 2417 /* 2418 * Referenced by us 2419 */ 2420 if (pte & PG_U) 2421 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 2422 2423 /* 2424 * Referenced by someone 2425 */ 2426 else if ((PHYS_TO_VM_PAGE(pa)->flags & PG_REFERENCED) || 2427 pmap_is_referenced(pa)) 2428 val |= MINCORE_REFERENCED_OTHER; 2429 } 2430 return val; 2431} 2432 2433#if defined(PMAP_DEBUG) 2434pmap_pid_dump(int pid) { 2435 pmap_t pmap; 2436 struct proc *p; 2437 int npte = 0; 2438 int index; 2439 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 2440 if (p->p_pid != pid) 2441 continue; 2442 2443 if (p->p_vmspace) { 2444 int i,j; 2445 index = 0; 2446 pmap = &p->p_vmspace->vm_pmap; 2447 for(i=0;i<1024;i++) { 2448 pd_entry_t *pde; 2449 unsigned *pte; 2450 unsigned base = i << PDRSHIFT; 2451 2452 pde = &pmap->pm_pdir[i]; 2453 if (pde && pmap_pde_v(pde)) { 2454 for(j=0;j<1024;j++) { 2455 unsigned va = base + (j << PAGE_SHIFT); 2456 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 2457 if (index) { 2458 index = 0; 2459 printf("\n"); 2460 } 2461 return npte; 2462 } 2463 pte = pmap_pte( pmap, va); 2464 if (pte && pmap_pte_v(pte)) { 2465 vm_offset_t pa; 2466 vm_page_t m; 2467 pa = *(int *)pte; 2468 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 2469 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 2470 va, pa, m->hold_count, m->wire_count, m->flags); 2471 npte++; 2472 index++; 2473 if (index >= 2) { 2474 index = 0; 2475 printf("\n"); 2476 } else { 2477 printf(" "); 2478 } 2479 } 2480 } 2481 } 2482 } 2483 } 2484 } 2485 return npte; 2486} 2487#endif 2488 2489#if defined(DEBUG) 2490 2491static void pads __P((pmap_t pm)); 2492static void pmap_pvdump __P((vm_offset_t pa)); 2493 2494/* print address space of pmap*/ 2495static void 2496pads(pm) 2497 pmap_t pm; 2498{ 2499 unsigned va, i, j; 2500 unsigned *ptep; 2501 2502 if (pm == kernel_pmap) 2503 return; 2504 for (i = 0; i < 1024; i++) 2505 if (pm->pm_pdir[i]) 2506 for (j = 0; j < 1024; j++) { 2507 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 2508 if (pm == kernel_pmap && va < KERNBASE) 2509 continue; 2510 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 2511 continue; 2512 ptep = pmap_pte(pm, va); 2513 if (pmap_pte_v(ptep)) 2514 printf("%x:%x ", va, *(int *) ptep); 2515 }; 2516 2517} 2518 2519static void 2520pmap_pvdump(pa) 2521 vm_offset_t pa; 2522{ 2523 register pv_entry_t pv; 2524 2525 printf("pa %x", pa); 2526 for (pv = pa_to_pvh(pa); pv; pv = pv->pv_next) { 2527#ifdef used_to_be 2528 printf(" -> pmap %x, va %x, flags %x", 2529 pv->pv_pmap, pv->pv_va, pv->pv_flags); 2530#endif 2531 printf(" -> pmap %x, va %x", 2532 pv->pv_pmap, pv->pv_va); 2533 pads(pv->pv_pmap); 2534 } 2535 printf(" "); 2536} 2537#endif 2538