pmap.c revision 18904
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 1.125 1996/10/12 21:35:03 dyson Exp $ 43 */ 44 45/* 46 * Manages physical address maps. 47 * 48 * In addition to hardware address maps, this 49 * module is called upon to provide software-use-only 50 * maps which may or may not be stored in the same 51 * form as hardware maps. These pseudo-maps are 52 * used to store intermediate results from copy 53 * operations to and from address spaces. 54 * 55 * Since the information managed by this module is 56 * also stored by the logical address mapping module, 57 * this module may throw away valid virtual-to-physical 58 * mappings at almost any time. However, invalidations 59 * of virtual-to-physical mappings must be done as 60 * requested. 61 * 62 * In order to cope with hardware architectures which 63 * make virtual-to-physical map invalidates expensive, 64 * this module may delay invalidate or reduced protection 65 * operations until such time as they are actually 66 * necessary. This module is given full information as 67 * to which processors are currently using which maps, 68 * and to when physical maps must be made correct. 69 */ 70 71#include "opt_cpu.h" 72 73#define PMAP_LOCK 1 74#define PMAP_PVLIST 1 75 76#include <sys/param.h> 77#include <sys/systm.h> 78#include <sys/proc.h> 79#include <sys/malloc.h> 80#include <sys/msgbuf.h> 81#include <sys/queue.h> 82#include <sys/vmmeter.h> 83#include <sys/mman.h> 84 85#include <vm/vm.h> 86#include <vm/vm_param.h> 87#include <vm/vm_prot.h> 88#include <vm/lock.h> 89#include <vm/vm_kern.h> 90#include <vm/vm_page.h> 91#include <vm/vm_map.h> 92#include <vm/vm_object.h> 93#include <vm/vm_extern.h> 94#include <vm/vm_pageout.h> 95#include <vm/vm_pager.h> 96 97#include <machine/pcb.h> 98#include <machine/cputypes.h> 99#include <machine/md_var.h> 100 101#define PMAP_KEEP_PDIRS 102 103#if defined(DIAGNOSTIC) 104#define PMAP_DIAGNOSTIC 105#endif 106 107#if !defined(PMAP_DIAGNOSTIC) 108#define PMAP_INLINE __inline 109#else 110#define PMAP_INLINE 111#endif 112 113#define PTPHINT 114 115static void init_pv_entries __P((int)); 116 117/* 118 * Get PDEs and PTEs for user/kernel address space 119 */ 120#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 121#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 122 123#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 124#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 125#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 126#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 127#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 128 129#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 130#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 131 132/* 133 * Given a map and a machine independent protection code, 134 * convert to a vax protection code. 135 */ 136#define pte_prot(m, p) (protection_codes[p]) 137static int protection_codes[8]; 138 139#define pa_index(pa) atop((pa) - vm_first_phys) 140#define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) 141 142static struct pmap kernel_pmap_store; 143pmap_t kernel_pmap; 144 145vm_offset_t avail_start; /* PA of first available physical page */ 146vm_offset_t avail_end; /* PA of last available physical page */ 147vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 148vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 149static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 150static vm_offset_t vm_first_phys; 151 152static int nkpt; 153static vm_page_t nkpg; 154vm_offset_t kernel_vm_end; 155 156extern vm_offset_t clean_sva, clean_eva; 157extern int cpu_class; 158 159#define PV_FREELIST_MIN ((PAGE_SIZE / sizeof (struct pv_entry)) / 2) 160 161/* 162 * Data for the pv entry allocation mechanism 163 */ 164static int pv_freelistcnt; 165TAILQ_HEAD (,pv_entry) pv_freelist; 166static vm_offset_t pvva; 167static int npvvapg; 168 169/* 170 * All those kernel PT submaps that BSD is so fond of 171 */ 172pt_entry_t *CMAP1; 173static pt_entry_t *CMAP2, *ptmmap; 174static pv_table_t *pv_table; 175caddr_t CADDR1, ptvmmap; 176static caddr_t CADDR2; 177static pt_entry_t *msgbufmap; 178struct msgbuf *msgbufp; 179 180pt_entry_t *PMAP1; 181unsigned *PADDR1; 182 183static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 184static unsigned * get_ptbase __P((pmap_t pmap)); 185static pv_entry_t get_pv_entry __P((void)); 186static void i386_protection_init __P((void)); 187static void pmap_alloc_pv_entry __P((void)); 188static void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 189 190static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 191static void pmap_remove_all __P((vm_offset_t pa)); 192static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 193 vm_offset_t pa, vm_page_t mpte)); 194static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 195 vm_offset_t sva)); 196static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 197static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 198 vm_offset_t va)); 199static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 200static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 201 vm_page_t mpte, vm_offset_t pa)); 202 203static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 204 205static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 206static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 207static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 208static vm_page_t pmap_page_alloc __P((vm_object_t object, vm_pindex_t pindex)); 209static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 210static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 211 212#define PDSTACKMAX 6 213static vm_offset_t pdstack[PDSTACKMAX]; 214static int pdstackptr; 215 216/* 217 * Bootstrap the system enough to run with virtual memory. 218 * 219 * On the i386 this is called after mapping has already been enabled 220 * and just syncs the pmap module with what has already been done. 221 * [We can't call it easily with mapping off since the kernel is not 222 * mapped with PA == VA, hence we would have to relocate every address 223 * from the linked base (virtual) address "KERNBASE" to the actual 224 * (physical) address starting relative to 0] 225 */ 226void 227pmap_bootstrap(firstaddr, loadaddr) 228 vm_offset_t firstaddr; 229 vm_offset_t loadaddr; 230{ 231 vm_offset_t va; 232 pt_entry_t *pte; 233 234 avail_start = firstaddr; 235 236 /* 237 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 238 * large. It should instead be correctly calculated in locore.s and 239 * not based on 'first' (which is a physical address, not a virtual 240 * address, for the start of unused physical memory). The kernel 241 * page tables are NOT double mapped and thus should not be included 242 * in this calculation. 243 */ 244 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 245 virtual_end = VM_MAX_KERNEL_ADDRESS; 246 247 /* 248 * Initialize protection array. 249 */ 250 i386_protection_init(); 251 252 /* 253 * The kernel's pmap is statically allocated so we don't have to use 254 * pmap_create, which is unlikely to work correctly at this part of 255 * the boot sequence (XXX and which no longer exists). 256 */ 257 kernel_pmap = &kernel_pmap_store; 258 259 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + IdlePTD); 260 261 kernel_pmap->pm_count = 1; 262#if PMAP_PVLIST 263 TAILQ_INIT(&kernel_pmap->pm_pvlist); 264#endif 265 nkpt = NKPT; 266 267 /* 268 * Reserve some special page table entries/VA space for temporary 269 * mapping of pages. 270 */ 271#define SYSMAP(c, p, v, n) \ 272 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 273 274 va = virtual_avail; 275 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 276 277 /* 278 * CMAP1/CMAP2 are used for zeroing and copying pages. 279 */ 280 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 281 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 282 283 /* 284 * ptmmap is used for reading arbitrary physical pages via /dev/mem. 285 */ 286 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 287 288 /* 289 * msgbufmap is used to map the system message buffer. 290 */ 291 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 1) 292 293 /* 294 * ptemap is used for pmap_pte_quick 295 */ 296 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 297 298 virtual_avail = va; 299 300 *(int *) CMAP1 = *(int *) CMAP2 = *(int *) PTD = 0; 301 invltlb(); 302 303} 304 305/* 306 * Initialize the pmap module. 307 * Called by vm_init, to initialize any structures that the pmap 308 * system needs to map virtual memory. 309 * pmap_init has been enhanced to support in a fairly consistant 310 * way, discontiguous physical memory. 311 */ 312void 313pmap_init(phys_start, phys_end) 314 vm_offset_t phys_start, phys_end; 315{ 316 vm_offset_t addr; 317 vm_size_t npg, s; 318 int i; 319 320 /* 321 * calculate the number of pv_entries needed 322 */ 323 vm_first_phys = phys_avail[0]; 324 for (i = 0; phys_avail[i + 1]; i += 2); 325 npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 326 327 /* 328 * Allocate memory for random pmap data structures. Includes the 329 * pv_head_table. 330 */ 331 s = (vm_size_t) (sizeof(pv_table_t) * npg); 332 s = round_page(s); 333 334 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 335 pv_table = (pv_table_t *) addr; 336 for(i=0;i<npg;i++) { 337 vm_offset_t pa; 338 TAILQ_INIT(&pv_table[i].pv_list); 339 pv_table[i].pv_list_count = 0; 340 pa = vm_first_phys + i * PAGE_SIZE; 341 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 342 } 343 TAILQ_INIT(&pv_freelist); 344 345 /* 346 * init the pv free list 347 */ 348 init_pv_entries(npg); 349 /* 350 * Now it is safe to enable pv_table recording. 351 */ 352 pmap_initialized = TRUE; 353} 354 355/* 356 * Used to map a range of physical addresses into kernel 357 * virtual address space. 358 * 359 * For now, VM is already on, we only need to map the 360 * specified memory. 361 */ 362vm_offset_t 363pmap_map(virt, start, end, prot) 364 vm_offset_t virt; 365 vm_offset_t start; 366 vm_offset_t end; 367 int prot; 368{ 369 while (start < end) { 370 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 371 virt += PAGE_SIZE; 372 start += PAGE_SIZE; 373 } 374 return (virt); 375} 376 377 378/*************************************************** 379 * Low level helper routines..... 380 ***************************************************/ 381 382#if defined(PMAP_DIAGNOSTIC) 383 384/* 385 * This code checks for non-writeable/modified pages. 386 * This should be an invalid condition. 387 */ 388static int 389pmap_nw_modified(pt_entry_t ptea) { 390 int pte; 391 392 pte = (int) ptea; 393 394 if ((pte & (PG_M|PG_RW)) == PG_M) 395 return 1; 396 else 397 return 0; 398} 399#endif 400 401 402/* 403 * this routine defines the region(s) of memory that should 404 * not be tested for the modified bit. 405 */ 406static PMAP_INLINE int 407pmap_track_modified( vm_offset_t va) { 408 if ((va < clean_sva) || (va >= clean_eva)) 409 return 1; 410 else 411 return 0; 412} 413 414static PMAP_INLINE void 415invltlb_1pg( vm_offset_t va) { 416#if defined(I386_CPU) 417 if (cpu_class == CPUCLASS_386) { 418 invltlb(); 419 } else 420#endif 421 { 422 invlpg(va); 423 } 424} 425 426static PMAP_INLINE void 427invltlb_2pg( vm_offset_t va1, vm_offset_t va2) { 428#if defined(I386_CPU) 429 if (cpu_class == CPUCLASS_386) { 430 invltlb(); 431 } else 432#endif 433 { 434 invlpg(va1); 435 invlpg(va2); 436 } 437} 438 439static unsigned * 440get_ptbase(pmap) 441 pmap_t pmap; 442{ 443 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 444 445 /* are we current address space or kernel? */ 446 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 447 return (unsigned *) PTmap; 448 } 449 /* otherwise, we are alternate address space */ 450 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 451 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 452 invltlb(); 453 } 454 return (unsigned *) APTmap; 455} 456 457/* 458 * Routine: pmap_pte 459 * Function: 460 * Extract the page table entry associated 461 * with the given map/virtual_address pair. 462 */ 463 464PMAP_INLINE unsigned * 465pmap_pte(pmap, va) 466 register pmap_t pmap; 467 vm_offset_t va; 468{ 469 if (pmap && *pmap_pde(pmap, va)) { 470 return get_ptbase(pmap) + i386_btop(va); 471 } 472 return (0); 473} 474 475/* 476 * Super fast pmap_pte routine best used when scanning 477 * the pv lists. This eliminates many coarse-grained 478 * invltlb calls. Note that many of the pv list 479 * scans are across different pmaps. It is very wasteful 480 * to do an entire invltlb for checking a single mapping. 481 */ 482 483static unsigned * 484pmap_pte_quick(pmap, va) 485 register pmap_t pmap; 486 vm_offset_t va; 487{ 488 unsigned pde, newpf; 489 if (pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) { 490 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 491 unsigned index = i386_btop(va); 492 /* are we current address space or kernel? */ 493 if ((pmap == kernel_pmap) || 494 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 495 return (unsigned *) PTmap + index; 496 } 497 newpf = pde & PG_FRAME; 498 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 499 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 500 invltlb_1pg((vm_offset_t) PADDR1); 501 } 502 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 503 } 504 return (0); 505} 506 507/* 508 * Routine: pmap_extract 509 * Function: 510 * Extract the physical page address associated 511 * with the given map/virtual_address pair. 512 */ 513vm_offset_t 514pmap_extract(pmap, va) 515 register pmap_t pmap; 516 vm_offset_t va; 517{ 518 vm_offset_t rtval; 519 if (pmap && *pmap_pde(pmap, va)) { 520 unsigned *pte; 521 pte = get_ptbase(pmap) + i386_btop(va); 522 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 523 return rtval; 524 } 525 return 0; 526 527} 528 529/* 530 * determine if a page is managed (memory vs. device) 531 */ 532static PMAP_INLINE int 533pmap_is_managed(pa) 534 vm_offset_t pa; 535{ 536 int i; 537 538 if (!pmap_initialized) 539 return 0; 540 541 for (i = 0; phys_avail[i + 1]; i += 2) { 542 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 543 return 1; 544 } 545 return 0; 546} 547 548 549/*************************************************** 550 * Low level mapping routines..... 551 ***************************************************/ 552 553/* 554 * Add a list of wired pages to the kva 555 * this routine is only used for temporary 556 * kernel mappings that do not need to have 557 * page modification or references recorded. 558 * Note that old mappings are simply written 559 * over. The page *must* be wired. 560 */ 561void 562pmap_qenter(va, m, count) 563 vm_offset_t va; 564 vm_page_t *m; 565 int count; 566{ 567 int i; 568 register unsigned *pte; 569 570 for (i = 0; i < count; i++) { 571 vm_offset_t tva = va + i * PAGE_SIZE; 572 unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V; 573 unsigned opte; 574 pte = (unsigned *)vtopte(tva); 575 opte = *pte; 576 *pte = npte; 577 if (opte) 578 invltlb_1pg(tva); 579 } 580} 581 582/* 583 * this routine jerks page mappings from the 584 * kernel -- it is meant only for temporary mappings. 585 */ 586void 587pmap_qremove(va, count) 588 vm_offset_t va; 589 int count; 590{ 591 int i; 592 register unsigned *pte; 593 594 for (i = 0; i < count; i++) { 595 pte = (unsigned *)vtopte(va); 596 *pte = 0; 597 invltlb_1pg(va); 598 va += PAGE_SIZE; 599 } 600} 601 602/* 603 * add a wired page to the kva 604 * note that in order for the mapping to take effect -- you 605 * should do a invltlb after doing the pmap_kenter... 606 */ 607PMAP_INLINE void 608pmap_kenter(va, pa) 609 vm_offset_t va; 610 register vm_offset_t pa; 611{ 612 register unsigned *pte; 613 unsigned npte, opte; 614 615 npte = pa | PG_RW | PG_V; 616 pte = (unsigned *)vtopte(va); 617 opte = *pte; 618 *pte = npte; 619 if (opte) 620 invltlb_1pg(va); 621} 622 623/* 624 * remove a page from the kernel pagetables 625 */ 626PMAP_INLINE void 627pmap_kremove(va) 628 vm_offset_t va; 629{ 630 register unsigned *pte; 631 632 pte = (unsigned *)vtopte(va); 633 *pte = 0; 634 invltlb_1pg(va); 635} 636 637static vm_page_t 638pmap_page_alloc(object, pindex) 639 vm_object_t object; 640 vm_pindex_t pindex; 641{ 642 vm_page_t m; 643 m = vm_page_alloc(object, pindex, VM_ALLOC_ZERO); 644 if (m == NULL) { 645 VM_WAIT; 646 } 647 return m; 648} 649 650static vm_page_t 651pmap_page_lookup(object, pindex) 652 vm_object_t object; 653 vm_pindex_t pindex; 654{ 655 vm_page_t m; 656retry: 657 m = vm_page_lookup(object, pindex); 658 if (m) { 659 if (m->flags & PG_BUSY) { 660 m->flags |= PG_WANTED; 661 tsleep(m, PVM, "pplookp", 0); 662 goto retry; 663 } 664 } 665 666 return m; 667} 668 669 670 671 672/*************************************************** 673 * Page table page management routines..... 674 ***************************************************/ 675 676/* 677 * This routine unholds page table pages, and if the hold count 678 * drops to zero, then it decrements the wire count. 679 */ 680static int 681pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 682 int s; 683 684 vm_page_unhold(m); 685 686 s = splvm(); 687 while (m->flags & PG_BUSY) { 688 m->flags |= PG_WANTED; 689 tsleep(m, PVM, "pmuwpt", 0); 690 } 691 splx(s); 692 693 if (m->hold_count == 0) { 694 vm_offset_t pteva; 695 /* 696 * unmap the page table page 697 */ 698 pmap->pm_pdir[m->pindex] = 0; 699 --pmap->pm_stats.resident_count; 700 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 701 (((unsigned) PTDpde) & PG_FRAME)) { 702 /* 703 * Do a invltlb to make the invalidated mapping 704 * take effect immediately. 705 */ 706 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 707 invltlb_1pg(pteva); 708 } 709 710#if defined(PTPHINT) 711 if (pmap->pm_ptphint == m) 712 pmap->pm_ptphint = NULL; 713#endif 714 715 /* 716 * If the page is finally unwired, simply free it. 717 */ 718 --m->wire_count; 719 if (m->wire_count == 0) { 720 721 if (m->flags & PG_WANTED) { 722 m->flags &= ~PG_WANTED; 723 wakeup(m); 724 } 725 726 vm_page_free_zero(m); 727 --cnt.v_wire_count; 728 } 729 return 1; 730 } 731 return 0; 732} 733 734/* 735 * After removing a page table entry, this routine is used to 736 * conditionally free the page, and manage the hold/wire counts. 737 */ 738static int 739pmap_unuse_pt(pmap, va, mpte) 740 pmap_t pmap; 741 vm_offset_t va; 742 vm_page_t mpte; 743{ 744 unsigned ptepindex; 745 if (va >= UPT_MIN_ADDRESS) 746 return 0; 747 748 if (mpte == NULL) { 749 ptepindex = (va >> PDRSHIFT); 750#if defined(PTPHINT) 751 if (pmap->pm_ptphint && 752 (pmap->pm_ptphint->pindex == ptepindex)) { 753 mpte = pmap->pm_ptphint; 754 } else { 755 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 756 pmap->pm_ptphint = mpte; 757 } 758#else 759 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 760#endif 761 } 762 763 return pmap_unwire_pte_hold(pmap, mpte); 764} 765 766/* 767 * Initialize a preallocated and zeroed pmap structure, 768 * such as one in a vmspace structure. 769 */ 770void 771pmap_pinit(pmap) 772 register struct pmap *pmap; 773{ 774 vm_page_t ptdpg; 775 /* 776 * No need to allocate page table space yet but we do need a valid 777 * page directory table. 778 */ 779 780 if (pdstackptr > 0) { 781 --pdstackptr; 782 pmap->pm_pdir = (pd_entry_t *)pdstack[pdstackptr]; 783 } else { 784 pmap->pm_pdir = 785 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 786 } 787 788 /* 789 * allocate object for the ptes 790 */ 791 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 792 793 /* 794 * allocate the page directory page 795 */ 796retry: 797 ptdpg = pmap_page_alloc( pmap->pm_pteobj, PTDPTDI); 798 if (ptdpg == NULL) 799 goto retry; 800 801 ptdpg->wire_count = 1; 802 ++cnt.v_wire_count; 803 804 ptdpg->flags &= ~(PG_MAPPED|PG_BUSY); /* not mapped normally */ 805 ptdpg->valid = VM_PAGE_BITS_ALL; 806 807 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 808 if ((ptdpg->flags & PG_ZERO) == 0) 809 bzero(pmap->pm_pdir, PAGE_SIZE); 810 811 /* wire in kernel global address entries */ 812 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 813 814 /* install self-referential address mapping entry */ 815 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 816 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW; 817 818 pmap->pm_flags = 0; 819 pmap->pm_count = 1; 820 pmap->pm_ptphint = NULL; 821#if PMAP_PVLIST 822 TAILQ_INIT(&pmap->pm_pvlist); 823#endif 824} 825 826static int 827pmap_release_free_page(pmap, p) 828 struct pmap *pmap; 829 vm_page_t p; 830{ 831 int s; 832 unsigned *pde = (unsigned *) pmap->pm_pdir; 833 /* 834 * This code optimizes the case of freeing non-busy 835 * page-table pages. Those pages are zero now, and 836 * might as well be placed directly into the zero queue. 837 */ 838 s = splvm(); 839 if (p->flags & PG_BUSY) { 840 p->flags |= PG_WANTED; 841 tsleep(p, PVM, "pmaprl", 0); 842 splx(s); 843 return 0; 844 } 845 846 if (p->flags & PG_WANTED) { 847 p->flags &= ~PG_WANTED; 848 wakeup(p); 849 } 850 851 /* 852 * Remove the page table page from the processes address space. 853 */ 854 pde[p->pindex] = 0; 855 --pmap->pm_stats.resident_count; 856 857 if (p->hold_count) { 858 panic("pmap_release: freeing held page table page"); 859 } 860 /* 861 * Page directory pages need to have the kernel 862 * stuff cleared, so they can go into the zero queue also. 863 */ 864 if (p->pindex == PTDPTDI) { 865 bzero(pde + KPTDI, nkpt * PTESIZE); 866 pde[APTDPTDI] = 0; 867 pmap_kremove((vm_offset_t) pmap->pm_pdir); 868 } 869 870#if defined(PTPHINT) 871 if (pmap->pm_ptphint && 872 (pmap->pm_ptphint->pindex == p->pindex)) 873 pmap->pm_ptphint = NULL; 874#endif 875 876 vm_page_free_zero(p); 877 splx(s); 878 return 1; 879} 880 881/* 882 * this routine is called if the page table page is not 883 * mapped correctly. 884 */ 885static vm_page_t 886_pmap_allocpte(pmap, ptepindex) 887 pmap_t pmap; 888 unsigned ptepindex; 889{ 890 vm_offset_t pteva, ptepa; 891 vm_page_t m; 892 int needszero = 0; 893 894 /* 895 * Find or fabricate a new pagetable page 896 */ 897retry: 898 m = vm_page_lookup(pmap->pm_pteobj, ptepindex); 899 if (m == NULL) { 900 m = pmap_page_alloc(pmap->pm_pteobj, ptepindex); 901 if (m == NULL) 902 goto retry; 903 if ((m->flags & PG_ZERO) == 0) 904 needszero = 1; 905 m->flags &= ~(PG_ZERO|PG_BUSY); 906 m->valid = VM_PAGE_BITS_ALL; 907 } else { 908 if ((m->flags & PG_BUSY) || m->busy) { 909 m->flags |= PG_WANTED; 910 tsleep(m, PVM, "ptewai", 0); 911 goto retry; 912 } 913 } 914 915 if (m->queue != PQ_NONE) { 916 int s = splvm(); 917 vm_page_unqueue(m); 918 splx(s); 919 } 920 921 if (m->wire_count == 0) 922 ++cnt.v_wire_count; 923 ++m->wire_count; 924 925 /* 926 * Increment the hold count for the page table page 927 * (denoting a new mapping.) 928 */ 929 ++m->hold_count; 930 931 /* 932 * Map the pagetable page into the process address space, if 933 * it isn't already there. 934 */ 935 936 pmap->pm_stats.resident_count++; 937 938 ptepa = VM_PAGE_TO_PHYS(m); 939 pmap->pm_pdir[ptepindex] = (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V); 940 941#if defined(PTPHINT) 942 /* 943 * Set the page table hint 944 */ 945 pmap->pm_ptphint = m; 946#endif 947 948 /* 949 * Try to use the new mapping, but if we cannot, then 950 * do it with the routine that maps the page explicitly. 951 */ 952 if (needszero) { 953 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 954 (((unsigned) PTDpde) & PG_FRAME)) { 955 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 956 bzero((caddr_t) pteva, PAGE_SIZE); 957 } else { 958 pmap_zero_page(ptepa); 959 } 960 } 961 962 m->valid = VM_PAGE_BITS_ALL; 963 m->flags |= PG_MAPPED; 964 965 return m; 966} 967 968static vm_page_t 969pmap_allocpte(pmap, va) 970 pmap_t pmap; 971 vm_offset_t va; 972{ 973 unsigned ptepindex; 974 vm_offset_t ptepa; 975 vm_page_t m; 976 977 /* 978 * Calculate pagetable page index 979 */ 980 ptepindex = va >> PDRSHIFT; 981 982 /* 983 * Get the page directory entry 984 */ 985 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 986 987 /* 988 * If the page table page is mapped, we just increment the 989 * hold count, and activate it. 990 */ 991 if (ptepa) { 992#if defined(PTPHINT) 993 /* 994 * In order to get the page table page, try the 995 * hint first. 996 */ 997 if (pmap->pm_ptphint && 998 (pmap->pm_ptphint->pindex == ptepindex)) { 999 m = pmap->pm_ptphint; 1000 } else { 1001 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1002 pmap->pm_ptphint = m; 1003 } 1004#else 1005 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1006#endif 1007 ++m->hold_count; 1008 return m; 1009 } 1010 /* 1011 * Here if the pte page isn't mapped, or if it has been deallocated. 1012 */ 1013 return _pmap_allocpte(pmap, ptepindex); 1014} 1015 1016 1017/*************************************************** 1018* Pmap allocation/deallocation routines. 1019 ***************************************************/ 1020 1021/* 1022 * Release any resources held by the given physical map. 1023 * Called when a pmap initialized by pmap_pinit is being released. 1024 * Should only be called if the map contains no valid mappings. 1025 */ 1026void 1027pmap_release(pmap) 1028 register struct pmap *pmap; 1029{ 1030 vm_page_t p,n,ptdpg; 1031 vm_object_t object = pmap->pm_pteobj; 1032 1033#if defined(DIAGNOSTIC) 1034 if (object->ref_count != 1) 1035 panic("pmap_release: pteobj reference count != 1"); 1036#endif 1037 1038 ptdpg = NULL; 1039retry: 1040 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1041 n = TAILQ_NEXT(p, listq); 1042 if (p->pindex == PTDPTDI) { 1043 ptdpg = p; 1044 continue; 1045 } 1046 if (!pmap_release_free_page(pmap, p)) 1047 goto retry; 1048 } 1049 1050 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1051 goto retry; 1052 1053 vm_object_deallocate(object); 1054 if (pdstackptr < PDSTACKMAX) { 1055 pdstack[pdstackptr] = (vm_offset_t) pmap->pm_pdir; 1056 ++pdstackptr; 1057 } else { 1058 kmem_free(kernel_map, (vm_offset_t) pmap->pm_pdir, PAGE_SIZE); 1059 } 1060 pmap->pm_pdir = 0; 1061} 1062 1063/* 1064 * grow the number of kernel page table entries, if needed 1065 */ 1066void 1067pmap_growkernel(vm_offset_t addr) 1068{ 1069 struct proc *p; 1070 struct pmap *pmap; 1071 int s; 1072 1073 s = splhigh(); 1074 if (kernel_vm_end == 0) { 1075 kernel_vm_end = KERNBASE; 1076 nkpt = 0; 1077 while (pdir_pde(PTD, kernel_vm_end)) { 1078 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1079 ++nkpt; 1080 } 1081 } 1082 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1083 while (kernel_vm_end < addr) { 1084 if (pdir_pde(PTD, kernel_vm_end)) { 1085 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1086 continue; 1087 } 1088 ++nkpt; 1089 if (!nkpg) { 1090 vm_offset_t ptpkva = (vm_offset_t) vtopte(addr); 1091 /* 1092 * This index is bogus, but out of the way 1093 */ 1094 vm_pindex_t ptpidx = (ptpkva >> PAGE_SHIFT); 1095 nkpg = vm_page_alloc(kernel_object, 1096 ptpidx, VM_ALLOC_SYSTEM); 1097 if (!nkpg) 1098 panic("pmap_growkernel: no memory to grow kernel"); 1099 vm_page_wire(nkpg); 1100 vm_page_remove(nkpg); 1101 pmap_zero_page(VM_PAGE_TO_PHYS(nkpg)); 1102 } 1103 pdir_pde(PTD, kernel_vm_end) = (pd_entry_t) (VM_PAGE_TO_PHYS(nkpg) | PG_V | PG_RW); 1104 nkpg = NULL; 1105 1106 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1107 if (p->p_vmspace) { 1108 pmap = &p->p_vmspace->vm_pmap; 1109 *pmap_pde(pmap, kernel_vm_end) = pdir_pde(PTD, kernel_vm_end); 1110 } 1111 } 1112 *pmap_pde(kernel_pmap, kernel_vm_end) = pdir_pde(PTD, kernel_vm_end); 1113 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1114 } 1115 splx(s); 1116} 1117 1118/* 1119 * Retire the given physical map from service. 1120 * Should only be called if the map contains 1121 * no valid mappings. 1122 */ 1123void 1124pmap_destroy(pmap) 1125 register pmap_t pmap; 1126{ 1127 int count; 1128 1129 if (pmap == NULL) 1130 return; 1131 1132 count = --pmap->pm_count; 1133 if (count == 0) { 1134 pmap_release(pmap); 1135 free((caddr_t) pmap, M_VMPMAP); 1136 } 1137} 1138 1139/* 1140 * Add a reference to the specified pmap. 1141 */ 1142void 1143pmap_reference(pmap) 1144 pmap_t pmap; 1145{ 1146 if (pmap != NULL) { 1147 pmap->pm_count++; 1148 } 1149} 1150 1151/*************************************************** 1152* page management routines. 1153 ***************************************************/ 1154 1155/* 1156 * free the pv_entry back to the free list 1157 */ 1158static PMAP_INLINE void 1159free_pv_entry(pv) 1160 pv_entry_t pv; 1161{ 1162 ++pv_freelistcnt; 1163 TAILQ_INSERT_HEAD(&pv_freelist, pv, pv_list); 1164} 1165 1166/* 1167 * get a new pv_entry, allocating a block from the system 1168 * when needed. 1169 * the memory allocation is performed bypassing the malloc code 1170 * because of the possibility of allocations at interrupt time. 1171 */ 1172static pv_entry_t 1173get_pv_entry() 1174{ 1175 pv_entry_t tmp; 1176 1177 /* 1178 * get more pv_entry pages if needed 1179 */ 1180 if (pv_freelistcnt < PV_FREELIST_MIN || !TAILQ_FIRST(&pv_freelist)) { 1181 pmap_alloc_pv_entry(); 1182 } 1183 /* 1184 * get a pv_entry off of the free list 1185 */ 1186 --pv_freelistcnt; 1187 tmp = TAILQ_FIRST(&pv_freelist); 1188 TAILQ_REMOVE(&pv_freelist, tmp, pv_list); 1189 return tmp; 1190} 1191 1192/* 1193 * This *strange* allocation routine eliminates the possibility of a malloc 1194 * failure (*FATAL*) for a pv_entry_t data structure. 1195 * also -- this code is MUCH MUCH faster than the malloc equiv... 1196 * We really need to do the slab allocator thingie here. 1197 */ 1198static void 1199pmap_alloc_pv_entry() 1200{ 1201 /* 1202 * do we have any pre-allocated map-pages left? 1203 */ 1204 if (npvvapg) { 1205 vm_page_t m; 1206 1207 /* 1208 * allocate a physical page out of the vm system 1209 */ 1210 m = vm_page_alloc(kernel_object, 1211 OFF_TO_IDX(pvva - vm_map_min(kernel_map)), 1212 VM_ALLOC_INTERRUPT); 1213 if (m) { 1214 int newentries; 1215 int i; 1216 pv_entry_t entry; 1217 1218 newentries = (PAGE_SIZE / sizeof(struct pv_entry)); 1219 /* 1220 * wire the page 1221 */ 1222 vm_page_wire(m); 1223 m->flags &= ~PG_BUSY; 1224 /* 1225 * let the kernel see it 1226 */ 1227 pmap_kenter(pvva, VM_PAGE_TO_PHYS(m)); 1228 1229 entry = (pv_entry_t) pvva; 1230 /* 1231 * update the allocation pointers 1232 */ 1233 pvva += PAGE_SIZE; 1234 --npvvapg; 1235 1236 /* 1237 * free the entries into the free list 1238 */ 1239 for (i = 0; i < newentries; i++) { 1240 free_pv_entry(entry); 1241 entry++; 1242 } 1243 } 1244 } 1245 if (!TAILQ_FIRST(&pv_freelist)) 1246 panic("get_pv_entry: cannot get a pv_entry_t"); 1247} 1248 1249/* 1250 * init the pv_entry allocation system 1251 */ 1252#define PVSPERPAGE 64 1253void 1254init_pv_entries(npg) 1255 int npg; 1256{ 1257 /* 1258 * allocate enough kvm space for PVSPERPAGE entries per page (lots) 1259 * kvm space is fairly cheap, be generous!!! (the system can panic if 1260 * this is too small.) 1261 */ 1262 npvvapg = ((npg * PVSPERPAGE) * sizeof(struct pv_entry) 1263 + PAGE_SIZE - 1) / PAGE_SIZE; 1264 pvva = kmem_alloc_pageable(kernel_map, npvvapg * PAGE_SIZE); 1265 /* 1266 * get the first batch of entries 1267 */ 1268 pmap_alloc_pv_entry(); 1269} 1270 1271/* 1272 * If it is the first entry on the list, it is actually 1273 * in the header and we must copy the following entry up 1274 * to the header. Otherwise we must search the list for 1275 * the entry. In either case we free the now unused entry. 1276 */ 1277 1278static int 1279pmap_remove_entry(pmap, ppv, va) 1280 struct pmap *pmap; 1281 pv_table_t *ppv; 1282 vm_offset_t va; 1283{ 1284 pv_entry_t pv; 1285 int rtval; 1286 int s; 1287 1288 s = splvm(); 1289#if PMAP_PVLIST 1290 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1291#endif 1292 for (pv = TAILQ_FIRST(&ppv->pv_list); 1293 pv; 1294 pv = TAILQ_NEXT(pv, pv_list)) { 1295 if (pmap == pv->pv_pmap && va == pv->pv_va) 1296 break; 1297 } 1298#if PMAP_PVLIST 1299 } else { 1300 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1301 pv; 1302 pv = TAILQ_NEXT(pv, pv_plist)) { 1303 if (va == pv->pv_va) 1304 break; 1305 } 1306 } 1307#endif 1308 1309 rtval = 0; 1310 if (pv) { 1311 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1312 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1313 --ppv->pv_list_count; 1314 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 1315 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 1316 } 1317 1318#if PMAP_PVLIST 1319 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1320#endif 1321 free_pv_entry(pv); 1322 } 1323 1324 splx(s); 1325 return rtval; 1326} 1327 1328/* 1329 * Create a pv entry for page at pa for 1330 * (pmap, va). 1331 */ 1332static void 1333pmap_insert_entry(pmap, va, mpte, pa) 1334 pmap_t pmap; 1335 vm_offset_t va; 1336 vm_page_t mpte; 1337 vm_offset_t pa; 1338{ 1339 1340 int s; 1341 pv_entry_t pv; 1342 pv_table_t *ppv; 1343 1344 s = splvm(); 1345 pv = get_pv_entry(); 1346 pv->pv_va = va; 1347 pv->pv_pmap = pmap; 1348 pv->pv_ptem = mpte; 1349 1350#if PMAP_PVLIST 1351 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1352#endif 1353 1354 ppv = pa_to_pvh(pa); 1355 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1356 ++ppv->pv_list_count; 1357 1358 splx(s); 1359} 1360 1361/* 1362 * pmap_remove_pte: do the things to unmap a page in a process 1363 */ 1364static int 1365pmap_remove_pte(pmap, ptq, va) 1366 struct pmap *pmap; 1367 unsigned *ptq; 1368 vm_offset_t va; 1369{ 1370 unsigned oldpte; 1371 pv_table_t *ppv; 1372 1373 oldpte = *ptq; 1374 *ptq = 0; 1375 if (oldpte & PG_W) 1376 pmap->pm_stats.wired_count -= 1; 1377 pmap->pm_stats.resident_count -= 1; 1378 if (oldpte & PG_MANAGED) { 1379 ppv = pa_to_pvh(oldpte); 1380 if (oldpte & PG_M) { 1381#if defined(PMAP_DIAGNOSTIC) 1382 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1383 printf("pmap_remove: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, (int) oldpte); 1384 } 1385#endif 1386 if (pmap_track_modified(va)) 1387 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1388 } 1389 return pmap_remove_entry(pmap, ppv, va); 1390 } else { 1391 return pmap_unuse_pt(pmap, va, NULL); 1392 } 1393 1394 return 0; 1395} 1396 1397/* 1398 * Remove a single page from a process address space 1399 */ 1400static void 1401pmap_remove_page(pmap, va) 1402 struct pmap *pmap; 1403 register vm_offset_t va; 1404{ 1405 register unsigned *ptq; 1406 1407 /* 1408 * if there is no pte for this address, just skip it!!! 1409 */ 1410 if (*pmap_pde(pmap, va) == 0) { 1411 return; 1412 } 1413 1414 /* 1415 * get a local va for mappings for this pmap. 1416 */ 1417 ptq = get_ptbase(pmap) + i386_btop(va); 1418 if (*ptq) { 1419 (void) pmap_remove_pte(pmap, ptq, va); 1420 invltlb_1pg(va); 1421 } 1422 return; 1423} 1424 1425/* 1426 * Remove the given range of addresses from the specified map. 1427 * 1428 * It is assumed that the start and end are properly 1429 * rounded to the page size. 1430 */ 1431void 1432pmap_remove(pmap, sva, eva) 1433 struct pmap *pmap; 1434 register vm_offset_t sva; 1435 register vm_offset_t eva; 1436{ 1437 register unsigned *ptbase; 1438 vm_offset_t pdnxt; 1439 vm_offset_t ptpaddr; 1440 vm_offset_t sindex, eindex; 1441 int anyvalid; 1442 1443 if (pmap == NULL) 1444 return; 1445 1446 if (pmap->pm_stats.resident_count == 0) 1447 return; 1448 1449 /* 1450 * special handling of removing one page. a very 1451 * common operation and easy to short circuit some 1452 * code. 1453 */ 1454 if ((sva + PAGE_SIZE) == eva) { 1455 pmap_remove_page(pmap, sva); 1456 return; 1457 } 1458 1459 anyvalid = 0; 1460 1461 /* 1462 * Get a local virtual address for the mappings that are being 1463 * worked with. 1464 */ 1465 ptbase = get_ptbase(pmap); 1466 1467 sindex = i386_btop(sva); 1468 eindex = i386_btop(eva); 1469 1470 for (; sindex < eindex; sindex = pdnxt) { 1471 1472 /* 1473 * Calculate index for next page table. 1474 */ 1475 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1476 if (pmap->pm_stats.resident_count == 0) 1477 break; 1478 ptpaddr = (vm_offset_t) *pmap_pde(pmap, i386_ptob(sindex)); 1479 1480 /* 1481 * Weed out invalid mappings. Note: we assume that the page 1482 * directory table is always allocated, and in kernel virtual. 1483 */ 1484 if (ptpaddr == 0) 1485 continue; 1486 1487 /* 1488 * Limit our scan to either the end of the va represented 1489 * by the current page table page, or to the end of the 1490 * range being removed. 1491 */ 1492 if (pdnxt > eindex) { 1493 pdnxt = eindex; 1494 } 1495 1496 for ( ;sindex != pdnxt; sindex++) { 1497 vm_offset_t va; 1498 if (ptbase[sindex] == 0) { 1499 continue; 1500 } 1501 va = i386_ptob(sindex); 1502 1503 anyvalid++; 1504 if (pmap_remove_pte(pmap, 1505 ptbase + sindex, va)) 1506 break; 1507 } 1508 } 1509 1510 if (anyvalid) { 1511 invltlb(); 1512 } 1513} 1514 1515/* 1516 * Routine: pmap_remove_all 1517 * Function: 1518 * Removes this physical page from 1519 * all physical maps in which it resides. 1520 * Reflects back modify bits to the pager. 1521 * 1522 * Notes: 1523 * Original versions of this routine were very 1524 * inefficient because they iteratively called 1525 * pmap_remove (slow...) 1526 */ 1527 1528static void 1529pmap_remove_all(pa) 1530 vm_offset_t pa; 1531{ 1532 register pv_entry_t pv; 1533 pv_table_t *ppv; 1534 register unsigned *pte, tpte; 1535 int nmodify; 1536 int update_needed; 1537 int s; 1538 1539 nmodify = 0; 1540 update_needed = 0; 1541#if defined(PMAP_DIAGNOSTIC) 1542 /* 1543 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1544 * pages! 1545 */ 1546 if (!pmap_is_managed(pa)) { 1547 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%lx", pa); 1548 } 1549#endif 1550 1551 s = splvm(); 1552 ppv = pa_to_pvh(pa); 1553 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1554 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1555 1556 pv->pv_pmap->pm_stats.resident_count--; 1557 1558 tpte = *pte; 1559 *pte = 0; 1560 if (tpte & PG_W) 1561 pv->pv_pmap->pm_stats.wired_count--; 1562 /* 1563 * Update the vm_page_t clean and reference bits. 1564 */ 1565 if (tpte & PG_M) { 1566#if defined(PMAP_DIAGNOSTIC) 1567 if (pmap_nw_modified((pt_entry_t) tpte)) { 1568 printf("pmap_remove_all: modified page not writable: va: 0x%lx, pte: 0x%lx\n", pv->pv_va, tpte); 1569 } 1570#endif 1571 if (pmap_track_modified(pv->pv_va)) 1572 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1573 } 1574 if (!update_needed && 1575 ((!curproc || (&curproc->p_vmspace->vm_pmap == pv->pv_pmap)) || 1576 (pv->pv_pmap == kernel_pmap))) { 1577 update_needed = 1; 1578 } 1579 1580#if PMAP_PVLIST 1581 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1582#endif 1583 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1584 --ppv->pv_list_count; 1585 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1586 free_pv_entry(pv); 1587 } 1588 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 1589 1590 1591 if (update_needed) 1592 invltlb(); 1593 splx(s); 1594 return; 1595} 1596 1597/* 1598 * Set the physical protection on the 1599 * specified range of this map as requested. 1600 */ 1601void 1602pmap_protect(pmap, sva, eva, prot) 1603 register pmap_t pmap; 1604 vm_offset_t sva, eva; 1605 vm_prot_t prot; 1606{ 1607 register unsigned *ptbase; 1608 vm_offset_t pdnxt; 1609 vm_offset_t ptpaddr; 1610 vm_offset_t sindex, eindex; 1611 int anychanged; 1612 1613 1614 if (pmap == NULL) 1615 return; 1616 1617 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1618 pmap_remove(pmap, sva, eva); 1619 return; 1620 } 1621 if (prot & VM_PROT_WRITE) { 1622 return; 1623 } 1624 1625 anychanged = 0; 1626 1627 ptbase = get_ptbase(pmap); 1628 1629 sindex = i386_btop(sva); 1630 eindex = i386_btop(eva); 1631 1632 for (; sindex < eindex; sindex = pdnxt) { 1633 1634 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1635 ptpaddr = (vm_offset_t) *pmap_pde(pmap, i386_ptob(sindex)); 1636 1637 /* 1638 * Weed out invalid mappings. Note: we assume that the page 1639 * directory table is always allocated, and in kernel virtual. 1640 */ 1641 if (ptpaddr == 0) 1642 continue; 1643 1644 if (pdnxt > eindex) { 1645 pdnxt = eindex; 1646 } 1647 1648 for (; sindex != pdnxt; sindex++) { 1649 1650 unsigned pbits = ptbase[sindex]; 1651 1652 if (pbits & PG_RW) { 1653 if (pbits & PG_M) { 1654 vm_offset_t sva = i386_ptob(sindex); 1655 if (pmap_track_modified(sva)) { 1656 vm_page_t m = PHYS_TO_VM_PAGE(pbits); 1657 m->dirty = VM_PAGE_BITS_ALL; 1658 } 1659 } 1660 ptbase[sindex] = pbits & ~(PG_M|PG_RW); 1661 anychanged = 1; 1662 } 1663 } 1664 } 1665 if (anychanged) 1666 invltlb(); 1667} 1668 1669/* 1670 * Insert the given physical page (p) at 1671 * the specified virtual address (v) in the 1672 * target physical map with the protection requested. 1673 * 1674 * If specified, the page will be wired down, meaning 1675 * that the related pte can not be reclaimed. 1676 * 1677 * NB: This is the only routine which MAY NOT lazy-evaluate 1678 * or lose information. That is, this routine must actually 1679 * insert this page into the given map NOW. 1680 */ 1681void 1682pmap_enter(pmap, va, pa, prot, wired) 1683 register pmap_t pmap; 1684 vm_offset_t va; 1685 register vm_offset_t pa; 1686 vm_prot_t prot; 1687 boolean_t wired; 1688{ 1689 register unsigned *pte; 1690 vm_offset_t opa; 1691 vm_offset_t origpte, newpte; 1692 vm_page_t mpte; 1693 1694 if (pmap == NULL) 1695 return; 1696 1697 va &= PG_FRAME; 1698#ifdef PMAP_DIAGNOSTIC 1699 if (va > VM_MAX_KERNEL_ADDRESS) 1700 panic("pmap_enter: toobig"); 1701 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 1702 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 1703#endif 1704 1705 mpte = NULL; 1706 /* 1707 * In the case that a page table page is not 1708 * resident, we are creating it here. 1709 */ 1710 if (va < UPT_MIN_ADDRESS) 1711 mpte = pmap_allocpte(pmap, va); 1712 1713 pte = pmap_pte(pmap, va); 1714 /* 1715 * Page Directory table entry not valid, we need a new PT page 1716 */ 1717 if (pte == NULL) { 1718 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%lx\n", 1719 pmap->pm_pdir[PTDPTDI], va); 1720 } 1721 1722 origpte = *(vm_offset_t *)pte; 1723 pa &= PG_FRAME; 1724 opa = origpte & PG_FRAME; 1725 1726 /* 1727 * Mapping has not changed, must be protection or wiring change. 1728 */ 1729 if (origpte && (opa == pa)) { 1730 /* 1731 * Wiring change, just update stats. We don't worry about 1732 * wiring PT pages as they remain resident as long as there 1733 * are valid mappings in them. Hence, if a user page is wired, 1734 * the PT page will be also. 1735 */ 1736 if (wired && ((origpte & PG_W) == 0)) 1737 pmap->pm_stats.wired_count++; 1738 else if (!wired && (origpte & PG_W)) 1739 pmap->pm_stats.wired_count--; 1740 1741#if defined(PMAP_DIAGNOSTIC) 1742 if (pmap_nw_modified((pt_entry_t) origpte)) { 1743 printf("pmap_enter: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, origpte); 1744 } 1745#endif 1746 1747 /* 1748 * We might be turning off write access to the page, 1749 * so we go ahead and sense modify status. 1750 */ 1751 if (origpte & PG_MANAGED) { 1752 vm_page_t m; 1753 if (origpte & PG_M) { 1754 if (pmap_track_modified(va)) { 1755 m = PHYS_TO_VM_PAGE(pa); 1756 m->dirty = VM_PAGE_BITS_ALL; 1757 } 1758 } 1759 pa |= PG_MANAGED; 1760 } 1761 1762 if (mpte) 1763 --mpte->hold_count; 1764 1765 goto validate; 1766 } 1767 /* 1768 * Mapping has changed, invalidate old range and fall through to 1769 * handle validating new mapping. 1770 */ 1771 if (opa) { 1772 int err; 1773 err = pmap_remove_pte(pmap, pte, va); 1774 if (err) 1775 panic("pmap_enter: pte vanished, va: 0x%x", va); 1776 } 1777 1778 /* 1779 * Enter on the PV list if part of our managed memory Note that we 1780 * raise IPL while manipulating pv_table since pmap_enter can be 1781 * called at interrupt time. 1782 */ 1783 if (pmap_is_managed(pa)) { 1784 pmap_insert_entry(pmap, va, mpte, pa); 1785 pa |= PG_MANAGED; 1786 } 1787 1788 /* 1789 * Increment counters 1790 */ 1791 pmap->pm_stats.resident_count++; 1792 if (wired) 1793 pmap->pm_stats.wired_count++; 1794 1795validate: 1796 /* 1797 * Now validate mapping with desired protection/wiring. 1798 */ 1799 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 1800 1801 if (wired) 1802 newpte |= PG_W; 1803 if (va < UPT_MIN_ADDRESS) 1804 newpte |= PG_U; 1805 1806 /* 1807 * if the mapping or permission bits are different, we need 1808 * to update the pte. 1809 */ 1810 if ((origpte & ~(PG_M|PG_A)) != newpte) { 1811 *pte = newpte; 1812 if (origpte) 1813 invltlb_1pg(va); 1814 } 1815} 1816 1817/* 1818 * this code makes some *MAJOR* assumptions: 1819 * 1. Current pmap & pmap exists. 1820 * 2. Not wired. 1821 * 3. Read access. 1822 * 4. No page table pages. 1823 * 5. Tlbflush is deferred to calling procedure. 1824 * 6. Page IS managed. 1825 * but is *MUCH* faster than pmap_enter... 1826 */ 1827 1828static vm_page_t 1829pmap_enter_quick(pmap, va, pa, mpte) 1830 register pmap_t pmap; 1831 vm_offset_t va; 1832 register vm_offset_t pa; 1833 vm_page_t mpte; 1834{ 1835 register unsigned *pte; 1836 1837 /* 1838 * In the case that a page table page is not 1839 * resident, we are creating it here. 1840 */ 1841 if (va < UPT_MIN_ADDRESS) { 1842 unsigned ptepindex; 1843 vm_offset_t ptepa; 1844 1845 /* 1846 * Calculate pagetable page index 1847 */ 1848 ptepindex = va >> PDRSHIFT; 1849 if (mpte && (mpte->pindex == ptepindex)) { 1850 ++mpte->hold_count; 1851 } else { 1852retry: 1853 /* 1854 * Get the page directory entry 1855 */ 1856 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1857 1858 /* 1859 * If the page table page is mapped, we just increment 1860 * the hold count, and activate it. 1861 */ 1862 if (ptepa) { 1863#if defined(PTPHINT) 1864 if (pmap->pm_ptphint && 1865 (pmap->pm_ptphint->pindex == ptepindex)) { 1866 mpte = pmap->pm_ptphint; 1867 } else { 1868 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1869 pmap->pm_ptphint = mpte; 1870 } 1871#else 1872 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1873#endif 1874 if (mpte == NULL) 1875 goto retry; 1876 ++mpte->hold_count; 1877 } else { 1878 mpte = _pmap_allocpte(pmap, ptepindex); 1879 } 1880 } 1881 } else { 1882 mpte = NULL; 1883 } 1884 1885 /* 1886 * This call to vtopte makes the assumption that we are 1887 * entering the page into the current pmap. In order to support 1888 * quick entry into any pmap, one would likely use pmap_pte_quick. 1889 * But that isn't as quick as vtopte. 1890 */ 1891 pte = (unsigned *)vtopte(va); 1892 if (*pte) { 1893 if (mpte) 1894 pmap_unwire_pte_hold(pmap, mpte); 1895 return 0; 1896 } 1897 1898 /* 1899 * Enter on the PV list if part of our managed memory Note that we 1900 * raise IPL while manipulating pv_table since pmap_enter can be 1901 * called at interrupt time. 1902 */ 1903 pmap_insert_entry(pmap, va, mpte, pa); 1904 1905 /* 1906 * Increment counters 1907 */ 1908 pmap->pm_stats.resident_count++; 1909 1910 /* 1911 * Now validate mapping with RO protection 1912 */ 1913 *pte = pa | PG_V | PG_U | PG_MANAGED; 1914 1915 return mpte; 1916} 1917 1918#define MAX_INIT_PT (96) 1919/* 1920 * pmap_object_init_pt preloads the ptes for a given object 1921 * into the specified pmap. This eliminates the blast of soft 1922 * faults on process startup and immediately after an mmap. 1923 */ 1924void 1925pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 1926 pmap_t pmap; 1927 vm_offset_t addr; 1928 vm_object_t object; 1929 vm_pindex_t pindex; 1930 vm_size_t size; 1931 int limit; 1932{ 1933 vm_offset_t tmpidx; 1934 int psize; 1935 vm_page_t p, mpte; 1936 int objpgs; 1937 1938 psize = i386_btop(size); 1939 1940 if (!pmap || (object->type != OBJT_VNODE) || 1941 (limit && (psize > MAX_INIT_PT) && 1942 (object->resident_page_count > MAX_INIT_PT))) { 1943 return; 1944 } 1945 1946 if (psize + pindex > object->size) 1947 psize = object->size - pindex; 1948 1949 mpte = NULL; 1950 /* 1951 * if we are processing a major portion of the object, then scan the 1952 * entire thing. 1953 */ 1954 if (psize > (object->size >> 2)) { 1955 objpgs = psize; 1956 1957 for (p = TAILQ_FIRST(&object->memq); 1958 ((objpgs > 0) && (p != NULL)); 1959 p = TAILQ_NEXT(p, listq)) { 1960 1961 tmpidx = p->pindex; 1962 if (tmpidx < pindex) { 1963 continue; 1964 } 1965 tmpidx -= pindex; 1966 if (tmpidx >= psize) { 1967 continue; 1968 } 1969 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 1970 (p->busy == 0) && 1971 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 1972 if ((p->queue - p->pc) == PQ_CACHE) 1973 vm_page_deactivate(p); 1974 p->flags |= PG_BUSY; 1975 mpte = pmap_enter_quick(pmap, 1976 addr + i386_ptob(tmpidx), 1977 VM_PAGE_TO_PHYS(p), mpte); 1978 p->flags |= PG_MAPPED; 1979 PAGE_WAKEUP(p); 1980 } 1981 objpgs -= 1; 1982 } 1983 } else { 1984 /* 1985 * else lookup the pages one-by-one. 1986 */ 1987 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 1988 p = vm_page_lookup(object, tmpidx + pindex); 1989 if (p && 1990 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 1991 (p->busy == 0) && 1992 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 1993 if ((p->queue - p->pc) == PQ_CACHE) 1994 vm_page_deactivate(p); 1995 p->flags |= PG_BUSY; 1996 mpte = pmap_enter_quick(pmap, 1997 addr + i386_ptob(tmpidx), 1998 VM_PAGE_TO_PHYS(p), mpte); 1999 p->flags |= PG_MAPPED; 2000 PAGE_WAKEUP(p); 2001 } 2002 } 2003 } 2004 return; 2005} 2006 2007/* 2008 * pmap_prefault provides a quick way of clustering 2009 * pagefaults into a processes address space. It is a "cousin" 2010 * of pmap_object_init_pt, except it runs at page fault time instead 2011 * of mmap time. 2012 */ 2013#define PFBAK 2 2014#define PFFOR 2 2015#define PAGEORDER_SIZE (PFBAK+PFFOR) 2016 2017static int pmap_prefault_pageorder[] = { 2018 -PAGE_SIZE, PAGE_SIZE, -2 * PAGE_SIZE, 2 * PAGE_SIZE 2019}; 2020 2021void 2022pmap_prefault(pmap, addra, entry, object) 2023 pmap_t pmap; 2024 vm_offset_t addra; 2025 vm_map_entry_t entry; 2026 vm_object_t object; 2027{ 2028 int i; 2029 vm_offset_t starta; 2030 vm_offset_t addr; 2031 vm_pindex_t pindex; 2032 vm_page_t m, mpte; 2033 2034 if (entry->object.vm_object != object) 2035 return; 2036 2037 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) 2038 return; 2039 2040 starta = addra - PFBAK * PAGE_SIZE; 2041 if (starta < entry->start) { 2042 starta = entry->start; 2043 } else if (starta > addra) { 2044 starta = 0; 2045 } 2046 2047 mpte = NULL; 2048 for (i = 0; i < PAGEORDER_SIZE; i++) { 2049 vm_object_t lobject; 2050 unsigned *pte; 2051 2052 addr = addra + pmap_prefault_pageorder[i]; 2053 if (addr < starta || addr >= entry->end) 2054 continue; 2055 2056 if ((*pmap_pde(pmap, addr)) == NULL) 2057 continue; 2058 2059 pte = (unsigned *) vtopte(addr); 2060 if (*pte) 2061 continue; 2062 2063 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2064 lobject = object; 2065 for (m = vm_page_lookup(lobject, pindex); 2066 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2067 lobject = lobject->backing_object) { 2068 if (lobject->backing_object_offset & PAGE_MASK) 2069 break; 2070 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2071 m = vm_page_lookup(lobject->backing_object, pindex); 2072 } 2073 2074 /* 2075 * give-up when a page is not in memory 2076 */ 2077 if (m == NULL) 2078 break; 2079 2080 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2081 (m->busy == 0) && 2082 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2083 2084 if ((m->queue - m->pc) == PQ_CACHE) { 2085 vm_page_deactivate(m); 2086 } 2087 m->flags |= PG_BUSY; 2088 mpte = pmap_enter_quick(pmap, addr, 2089 VM_PAGE_TO_PHYS(m), mpte); 2090 m->flags |= PG_MAPPED; 2091 PAGE_WAKEUP(m); 2092 } 2093 } 2094} 2095 2096/* 2097 * Routine: pmap_change_wiring 2098 * Function: Change the wiring attribute for a map/virtual-address 2099 * pair. 2100 * In/out conditions: 2101 * The mapping must already exist in the pmap. 2102 */ 2103void 2104pmap_change_wiring(pmap, va, wired) 2105 register pmap_t pmap; 2106 vm_offset_t va; 2107 boolean_t wired; 2108{ 2109 register unsigned *pte; 2110 2111 if (pmap == NULL) 2112 return; 2113 2114 pte = pmap_pte(pmap, va); 2115 2116 if (wired && !pmap_pte_w(pte)) 2117 pmap->pm_stats.wired_count++; 2118 else if (!wired && pmap_pte_w(pte)) 2119 pmap->pm_stats.wired_count--; 2120 2121 /* 2122 * Wiring is not a hardware characteristic so there is no need to 2123 * invalidate TLB. 2124 */ 2125 pmap_pte_set_w(pte, wired); 2126} 2127 2128 2129 2130/* 2131 * Copy the range specified by src_addr/len 2132 * from the source map to the range dst_addr/len 2133 * in the destination map. 2134 * 2135 * This routine is only advisory and need not do anything. 2136 */ 2137 2138void 2139pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2140 pmap_t dst_pmap, src_pmap; 2141 vm_offset_t dst_addr; 2142 vm_size_t len; 2143 vm_offset_t src_addr; 2144{ 2145 vm_offset_t addr; 2146 vm_offset_t end_addr = src_addr + len; 2147 vm_offset_t pdnxt; 2148 unsigned src_frame, dst_frame; 2149 2150 if (dst_addr != src_addr) 2151 return; 2152 2153 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2154 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2155 return; 2156 } 2157 2158 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2159 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2160 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2161 invltlb(); 2162 } 2163 2164 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2165 unsigned *src_pte, *dst_pte; 2166 vm_page_t dstmpte, srcmpte; 2167 vm_offset_t srcptepaddr; 2168 unsigned ptepindex; 2169 2170 if (addr >= UPT_MIN_ADDRESS) 2171 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2172 2173 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2174 ptepindex = addr >> PDRSHIFT; 2175 2176 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2177 if (srcptepaddr == 0) 2178 continue; 2179 2180 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2181 if ((srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2182 continue; 2183 2184 if (pdnxt > end_addr) 2185 pdnxt = end_addr; 2186 2187 src_pte = (unsigned *) vtopte(addr); 2188 dst_pte = (unsigned *) avtopte(addr); 2189 while (addr < pdnxt) { 2190 unsigned ptetemp; 2191 ptetemp = *src_pte; 2192 /* 2193 * we only virtual copy managed pages 2194 */ 2195 if ((ptetemp & PG_MANAGED) != 0) { 2196 /* 2197 * We have to check after allocpte for the 2198 * pte still being around... allocpte can 2199 * block. 2200 */ 2201 dstmpte = pmap_allocpte(dst_pmap, addr); 2202 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2203 /* 2204 * Clear the modified and 2205 * accessed (referenced) bits 2206 * during the copy. 2207 */ 2208 *dst_pte = ptetemp & ~(PG_M|PG_A); 2209 dst_pmap->pm_stats.resident_count++; 2210 pmap_insert_entry(dst_pmap, addr, 2211 dstmpte, 2212 (ptetemp & PG_FRAME)); 2213 } else { 2214 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2215 } 2216 if (dstmpte->hold_count >= srcmpte->hold_count) 2217 break; 2218 } 2219 addr += PAGE_SIZE; 2220 ++src_pte; 2221 ++dst_pte; 2222 } 2223 } 2224} 2225 2226/* 2227 * Routine: pmap_kernel 2228 * Function: 2229 * Returns the physical map handle for the kernel. 2230 */ 2231pmap_t 2232pmap_kernel() 2233{ 2234 return (kernel_pmap); 2235} 2236 2237/* 2238 * pmap_zero_page zeros the specified (machine independent) 2239 * page by mapping the page into virtual memory and using 2240 * bzero to clear its contents, one machine dependent page 2241 * at a time. 2242 */ 2243void 2244pmap_zero_page(phys) 2245 vm_offset_t phys; 2246{ 2247 if (*(int *) CMAP2) 2248 panic("pmap_zero_page: CMAP busy"); 2249 2250 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME); 2251 bzero(CADDR2, PAGE_SIZE); 2252 *(int *) CMAP2 = 0; 2253 invltlb_1pg((vm_offset_t) CADDR2); 2254} 2255 2256/* 2257 * pmap_copy_page copies the specified (machine independent) 2258 * page by mapping the page into virtual memory and using 2259 * bcopy to copy the page, one machine dependent page at a 2260 * time. 2261 */ 2262void 2263pmap_copy_page(src, dst) 2264 vm_offset_t src; 2265 vm_offset_t dst; 2266{ 2267 if (*(int *) CMAP1 || *(int *) CMAP2) 2268 panic("pmap_copy_page: CMAP busy"); 2269 2270 *(int *) CMAP1 = PG_V | PG_RW | (src & PG_FRAME); 2271 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME); 2272 2273 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2274 2275 *(int *) CMAP1 = 0; 2276 *(int *) CMAP2 = 0; 2277 invltlb_2pg( (vm_offset_t) CADDR1, (vm_offset_t) CADDR2); 2278} 2279 2280 2281/* 2282 * Routine: pmap_pageable 2283 * Function: 2284 * Make the specified pages (by pmap, offset) 2285 * pageable (or not) as requested. 2286 * 2287 * A page which is not pageable may not take 2288 * a fault; therefore, its page table entry 2289 * must remain valid for the duration. 2290 * 2291 * This routine is merely advisory; pmap_enter 2292 * will specify that these pages are to be wired 2293 * down (or not) as appropriate. 2294 */ 2295void 2296pmap_pageable(pmap, sva, eva, pageable) 2297 pmap_t pmap; 2298 vm_offset_t sva, eva; 2299 boolean_t pageable; 2300{ 2301} 2302 2303/* 2304 * this routine returns true if a physical page resides 2305 * in the given pmap. 2306 */ 2307boolean_t 2308pmap_page_exists(pmap, pa) 2309 pmap_t pmap; 2310 vm_offset_t pa; 2311{ 2312 register pv_entry_t pv; 2313 pv_table_t *ppv; 2314 int s; 2315 2316 if (!pmap_is_managed(pa)) 2317 return FALSE; 2318 2319 s = splvm(); 2320 2321 ppv = pa_to_pvh(pa); 2322 /* 2323 * Not found, check current mappings returning immediately if found. 2324 */ 2325 for (pv = TAILQ_FIRST(&ppv->pv_list); 2326 pv; 2327 pv = TAILQ_NEXT(pv, pv_list)) { 2328 if (pv->pv_pmap == pmap) { 2329 splx(s); 2330 return TRUE; 2331 } 2332 } 2333 splx(s); 2334 return (FALSE); 2335} 2336 2337#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2338/* 2339 * Remove all pages from specified address space 2340 * this aids process exit speeds. Also, this code 2341 * is special cased for current process only, but 2342 * can have the more generic (and slightly slower) 2343 * mode enabled. This is much faster than pmap_remove 2344 * in the case of running down an entire address space. 2345 */ 2346void 2347pmap_remove_pages(pmap, sva, eva) 2348 pmap_t pmap; 2349 vm_offset_t sva, eva; 2350{ 2351 unsigned *pte, tpte; 2352 pv_table_t *ppv; 2353 pv_entry_t pv, npv; 2354 int s; 2355 2356#if PMAP_PVLIST 2357 2358#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2359 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) { 2360 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2361 return; 2362 } 2363#endif 2364 2365 s = splvm(); 2366 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2367 pv; 2368 pv = npv) { 2369 2370 if (pv->pv_va >= eva || pv->pv_va < sva) { 2371 npv = TAILQ_NEXT(pv, pv_plist); 2372 continue; 2373 } 2374 2375#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2376 pte = (unsigned *)vtopte(pv->pv_va); 2377#else 2378 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2379#endif 2380 tpte = *pte; 2381 *pte = 0; 2382 2383 ppv = pa_to_pvh(tpte); 2384 2385 pv->pv_pmap->pm_stats.resident_count--; 2386 if (tpte & PG_W) 2387 pv->pv_pmap->pm_stats.wired_count--; 2388 /* 2389 * Update the vm_page_t clean and reference bits. 2390 */ 2391 if (tpte & PG_M) { 2392 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 2393 } 2394 2395 2396 npv = TAILQ_NEXT(pv, pv_plist); 2397 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2398 2399 --ppv->pv_list_count; 2400 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2401 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2402 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 2403 } 2404 2405 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2406 free_pv_entry(pv); 2407 } 2408 splx(s); 2409 invltlb(); 2410#endif 2411} 2412 2413/* 2414 * pmap_testbit tests bits in pte's 2415 * note that the testbit/changebit routines are inline, 2416 * and a lot of things compile-time evaluate. 2417 */ 2418static boolean_t 2419pmap_testbit(pa, bit) 2420 register vm_offset_t pa; 2421 int bit; 2422{ 2423 register pv_entry_t pv; 2424 pv_table_t *ppv; 2425 unsigned *pte; 2426 int s; 2427 2428 if (!pmap_is_managed(pa)) 2429 return FALSE; 2430 2431 ppv = pa_to_pvh(pa); 2432 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 2433 return FALSE; 2434 2435 s = splvm(); 2436 2437 for (pv = TAILQ_FIRST(&ppv->pv_list); 2438 pv; 2439 pv = TAILQ_NEXT(pv, pv_list)) { 2440 2441 /* 2442 * if the bit being tested is the modified bit, then 2443 * mark clean_map and ptes as never 2444 * modified. 2445 */ 2446 if (bit & (PG_A|PG_M)) { 2447 if (!pmap_track_modified(pv->pv_va)) 2448 continue; 2449 } 2450 2451#if defined(PMAP_DIAGNOSTIC) 2452 if (!pv->pv_pmap) { 2453 printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va); 2454 continue; 2455 } 2456#endif 2457 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2458 if (*pte & bit) { 2459 splx(s); 2460 return TRUE; 2461 } 2462 } 2463 splx(s); 2464 return (FALSE); 2465} 2466 2467/* 2468 * this routine is used to modify bits in ptes 2469 */ 2470static void 2471pmap_changebit(pa, bit, setem) 2472 vm_offset_t pa; 2473 int bit; 2474 boolean_t setem; 2475{ 2476 register pv_entry_t pv; 2477 pv_table_t *ppv; 2478 register unsigned *pte; 2479 int changed; 2480 int s; 2481 2482 if (!pmap_is_managed(pa)) 2483 return; 2484 2485 s = splvm(); 2486 changed = 0; 2487 ppv = pa_to_pvh(pa); 2488 2489 /* 2490 * Loop over all current mappings setting/clearing as appropos If 2491 * setting RO do we need to clear the VAC? 2492 */ 2493 for (pv = TAILQ_FIRST(&ppv->pv_list); 2494 pv; 2495 pv = TAILQ_NEXT(pv, pv_list)) { 2496 2497 /* 2498 * don't write protect pager mappings 2499 */ 2500 if (!setem && (bit == PG_RW)) { 2501 if (!pmap_track_modified(pv->pv_va)) 2502 continue; 2503 } 2504 2505#if defined(PMAP_DIAGNOSTIC) 2506 if (!pv->pv_pmap) { 2507 printf("Null pmap (cb) at va: 0x%lx\n", pv->pv_va); 2508 continue; 2509 } 2510#endif 2511 2512 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2513 2514 if (setem) { 2515 *(int *)pte |= bit; 2516 changed = 1; 2517 } else { 2518 vm_offset_t pbits = *(vm_offset_t *)pte; 2519 if (pbits & bit) { 2520 changed = 1; 2521 if (bit == PG_RW) { 2522 if (pbits & PG_M) { 2523 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 2524 } 2525 *(int *)pte = pbits & ~(PG_M|PG_RW); 2526 } else { 2527 *(int *)pte = pbits & ~bit; 2528 } 2529 } 2530 } 2531 } 2532 splx(s); 2533 if (changed) 2534 invltlb(); 2535} 2536 2537/* 2538 * pmap_page_protect: 2539 * 2540 * Lower the permission for all mappings to a given page. 2541 */ 2542void 2543pmap_page_protect(phys, prot) 2544 vm_offset_t phys; 2545 vm_prot_t prot; 2546{ 2547 if ((prot & VM_PROT_WRITE) == 0) { 2548 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 2549 pmap_changebit(phys, PG_RW, FALSE); 2550 } else { 2551 pmap_remove_all(phys); 2552 } 2553 } 2554} 2555 2556vm_offset_t 2557pmap_phys_address(ppn) 2558 int ppn; 2559{ 2560 return (i386_ptob(ppn)); 2561} 2562 2563/* 2564 * pmap_is_referenced: 2565 * 2566 * Return whether or not the specified physical page was referenced 2567 * by any physical maps. 2568 */ 2569boolean_t 2570pmap_is_referenced(vm_offset_t pa) 2571{ 2572 register pv_entry_t pv; 2573 pv_table_t *ppv; 2574 unsigned *pte; 2575 int s; 2576 2577 if (!pmap_is_managed(pa)) 2578 return FALSE; 2579 2580 ppv = pa_to_pvh(pa); 2581 2582 s = splvm(); 2583 /* 2584 * Not found, check current mappings returning immediately if found. 2585 */ 2586 for (pv = TAILQ_FIRST(&ppv->pv_list); 2587 pv; 2588 pv = TAILQ_NEXT(pv, pv_list)) { 2589 2590 /* 2591 * if the bit being tested is the modified bit, then 2592 * mark clean_map and ptes as never 2593 * modified. 2594 */ 2595 if (!pmap_track_modified(pv->pv_va)) 2596 continue; 2597 2598 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2599 if ((int) *pte & PG_A) { 2600 splx(s); 2601 return TRUE; 2602 } 2603 } 2604 splx(s); 2605 return (FALSE); 2606} 2607 2608/* 2609 * pmap_ts_referenced: 2610 * 2611 * Return the count of reference bits for a page, clearing all of them. 2612 * 2613 */ 2614int 2615pmap_ts_referenced(vm_offset_t pa) 2616{ 2617 register pv_entry_t pv; 2618 pv_table_t *ppv; 2619 unsigned *pte; 2620 int s; 2621 int rtval = 0; 2622 2623 if (!pmap_is_managed(pa)) 2624 return FALSE; 2625 2626 s = splvm(); 2627 2628 ppv = pa_to_pvh(pa); 2629 2630 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2631 splx(s); 2632 return 0; 2633 } 2634 2635 /* 2636 * Not found, check current mappings returning immediately if found. 2637 */ 2638 for (pv = TAILQ_FIRST(&ppv->pv_list); 2639 pv; 2640 pv = TAILQ_NEXT(pv, pv_list)) { 2641 /* 2642 * if the bit being tested is the modified bit, then 2643 * mark clean_map and ptes as never 2644 * modified. 2645 */ 2646 if (!pmap_track_modified(pv->pv_va)) 2647 continue; 2648 2649 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2650 if (pte == NULL) { 2651 continue; 2652 } 2653 if (*pte & PG_A) { 2654 rtval++; 2655 *pte &= ~PG_A; 2656 } 2657 } 2658 splx(s); 2659 if (rtval) { 2660 invltlb(); 2661 } 2662 return (rtval); 2663} 2664 2665/* 2666 * pmap_is_modified: 2667 * 2668 * Return whether or not the specified physical page was modified 2669 * in any physical maps. 2670 */ 2671boolean_t 2672pmap_is_modified(vm_offset_t pa) 2673{ 2674 return pmap_testbit((pa), PG_M); 2675} 2676 2677/* 2678 * Clear the modify bits on the specified physical page. 2679 */ 2680void 2681pmap_clear_modify(vm_offset_t pa) 2682{ 2683 pmap_changebit((pa), PG_M, FALSE); 2684} 2685 2686/* 2687 * pmap_clear_reference: 2688 * 2689 * Clear the reference bit on the specified physical page. 2690 */ 2691void 2692pmap_clear_reference(vm_offset_t pa) 2693{ 2694 pmap_changebit((pa), PG_A, FALSE); 2695} 2696 2697/* 2698 * Miscellaneous support routines follow 2699 */ 2700 2701static void 2702i386_protection_init() 2703{ 2704 register int *kp, prot; 2705 2706 kp = protection_codes; 2707 for (prot = 0; prot < 8; prot++) { 2708 switch (prot) { 2709 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 2710 /* 2711 * Read access is also 0. There isn't any execute bit, 2712 * so just make it readable. 2713 */ 2714 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 2715 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 2716 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 2717 *kp++ = 0; 2718 break; 2719 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 2720 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 2721 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 2722 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 2723 *kp++ = PG_RW; 2724 break; 2725 } 2726 } 2727} 2728 2729/* 2730 * Map a set of physical memory pages into the kernel virtual 2731 * address space. Return a pointer to where it is mapped. This 2732 * routine is intended to be used for mapping device memory, 2733 * NOT real memory. The non-cacheable bits are set on each 2734 * mapped page. 2735 */ 2736void * 2737pmap_mapdev(pa, size) 2738 vm_offset_t pa; 2739 vm_size_t size; 2740{ 2741 vm_offset_t va, tmpva; 2742 unsigned *pte; 2743 2744 size = roundup(size, PAGE_SIZE); 2745 2746 va = kmem_alloc_pageable(kernel_map, size); 2747 if (!va) 2748 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 2749 2750 pa = pa & PG_FRAME; 2751 for (tmpva = va; size > 0;) { 2752 pte = (unsigned *)vtopte(tmpva); 2753 *pte = pa | PG_RW | PG_V | PG_N; 2754 size -= PAGE_SIZE; 2755 tmpva += PAGE_SIZE; 2756 pa += PAGE_SIZE; 2757 } 2758 invltlb(); 2759 2760 return ((void *) va); 2761} 2762 2763/* 2764 * perform the pmap work for mincore 2765 */ 2766int 2767pmap_mincore(pmap, addr) 2768 pmap_t pmap; 2769 vm_offset_t addr; 2770{ 2771 2772 unsigned *ptep, pte; 2773 int val = 0; 2774 2775 ptep = pmap_pte(pmap, addr); 2776 if (ptep == 0) { 2777 return 0; 2778 } 2779 2780 if (pte = *ptep) { 2781 vm_offset_t pa; 2782 val = MINCORE_INCORE; 2783 pa = pte & PG_FRAME; 2784 2785 /* 2786 * Modified by us 2787 */ 2788 if (pte & PG_M) 2789 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 2790 /* 2791 * Modified by someone 2792 */ 2793 else if (PHYS_TO_VM_PAGE(pa)->dirty || 2794 pmap_is_modified(pa)) 2795 val |= MINCORE_MODIFIED_OTHER; 2796 /* 2797 * Referenced by us 2798 */ 2799 if (pte & PG_U) 2800 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 2801 2802 /* 2803 * Referenced by someone 2804 */ 2805 else if ((PHYS_TO_VM_PAGE(pa)->flags & PG_REFERENCED) || 2806 pmap_is_referenced(pa)) 2807 val |= MINCORE_REFERENCED_OTHER; 2808 } 2809 return val; 2810} 2811 2812#if defined(PMAP_DEBUG) 2813pmap_pid_dump(int pid) { 2814 pmap_t pmap; 2815 struct proc *p; 2816 int npte = 0; 2817 int index; 2818 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 2819 if (p->p_pid != pid) 2820 continue; 2821 2822 if (p->p_vmspace) { 2823 int i,j; 2824 index = 0; 2825 pmap = &p->p_vmspace->vm_pmap; 2826 for(i=0;i<1024;i++) { 2827 pd_entry_t *pde; 2828 unsigned *pte; 2829 unsigned base = i << PDRSHIFT; 2830 2831 pde = &pmap->pm_pdir[i]; 2832 if (pde && pmap_pde_v(pde)) { 2833 for(j=0;j<1024;j++) { 2834 unsigned va = base + (j << PAGE_SHIFT); 2835 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 2836 if (index) { 2837 index = 0; 2838 printf("\n"); 2839 } 2840 return npte; 2841 } 2842 pte = pmap_pte_quick( pmap, va); 2843 if (pte && pmap_pte_v(pte)) { 2844 vm_offset_t pa; 2845 vm_page_t m; 2846 pa = *(int *)pte; 2847 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 2848 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 2849 va, pa, m->hold_count, m->wire_count, m->flags); 2850 npte++; 2851 index++; 2852 if (index >= 2) { 2853 index = 0; 2854 printf("\n"); 2855 } else { 2856 printf(" "); 2857 } 2858 } 2859 } 2860 } 2861 } 2862 } 2863 } 2864 return npte; 2865} 2866#endif 2867 2868#if defined(DEBUG) 2869 2870static void pads __P((pmap_t pm)); 2871static void pmap_pvdump __P((vm_offset_t pa)); 2872 2873/* print address space of pmap*/ 2874static void 2875pads(pm) 2876 pmap_t pm; 2877{ 2878 unsigned va, i, j; 2879 unsigned *ptep; 2880 2881 if (pm == kernel_pmap) 2882 return; 2883 for (i = 0; i < 1024; i++) 2884 if (pm->pm_pdir[i]) 2885 for (j = 0; j < 1024; j++) { 2886 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 2887 if (pm == kernel_pmap && va < KERNBASE) 2888 continue; 2889 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 2890 continue; 2891 ptep = pmap_pte_quick(pm, va); 2892 if (pmap_pte_v(ptep)) 2893 printf("%x:%x ", va, *(int *) ptep); 2894 }; 2895 2896} 2897 2898static void 2899pmap_pvdump(pa) 2900 vm_offset_t pa; 2901{ 2902 pv_table_t *ppv; 2903 register pv_entry_t pv; 2904 2905 printf("pa %x", pa); 2906 ppv = pa_to_pvh(pa); 2907 for (pv = TAILQ_FIRST(&ppv->pv_list); 2908 pv; 2909 pv = TAILQ_NEXT(pv, pv_list)) { 2910#ifdef used_to_be 2911 printf(" -> pmap %x, va %x, flags %x", 2912 pv->pv_pmap, pv->pv_va, pv->pv_flags); 2913#endif 2914 printf(" -> pmap %x, va %x", 2915 pv->pv_pmap, pv->pv_va); 2916 pads(pv->pv_pmap); 2917 } 2918 printf(" "); 2919} 2920#endif 2921