pmap.c revision 28808
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 1.159 1997/08/25 21:53:01 bde 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#include <sys/param.h> 74#include <sys/systm.h> 75#include <sys/proc.h> 76#include <sys/malloc.h> 77#include <sys/msgbuf.h> 78#include <sys/vmmeter.h> 79#include <sys/mman.h> 80 81#include <vm/vm.h> 82#include <vm/vm_param.h> 83#include <vm/vm_prot.h> 84#include <sys/lock.h> 85#include <vm/vm_kern.h> 86#include <vm/vm_page.h> 87#include <vm/vm_map.h> 88#include <vm/vm_object.h> 89#include <vm/vm_extern.h> 90#include <vm/vm_pageout.h> 91#include <vm/vm_pager.h> 92#include <vm/vm_zone.h> 93 94#include <sys/user.h> 95 96#include <machine/cpu.h> 97#include <machine/cputypes.h> 98#include <machine/md_var.h> 99#include <machine/specialreg.h> 100#if defined(SMP) || defined(APIC_IO) 101#include <machine/smp.h> 102#include <machine/apic.h> 103#endif /* SMP || APIC_IO */ 104 105#define PMAP_KEEP_PDIRS 106#ifndef PMAP_SHPGPERPROC 107#define PMAP_SHPGPERPROC 200 108#endif 109 110#if defined(DIAGNOSTIC) 111#define PMAP_DIAGNOSTIC 112#endif 113 114#if !defined(PMAP_DIAGNOSTIC) 115#define PMAP_INLINE __inline 116#else 117#define PMAP_INLINE 118#endif 119 120#define PTPHINT 121 122/* 123 * Get PDEs and PTEs for user/kernel address space 124 */ 125#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 126#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 127 128#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 129#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 130#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 131#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 132#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 133 134#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 135#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 136 137/* 138 * Given a map and a machine independent protection code, 139 * convert to a vax protection code. 140 */ 141#define pte_prot(m, p) (protection_codes[p]) 142static int protection_codes[8]; 143 144#define pa_index(pa) atop((pa) - vm_first_phys) 145#define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) 146 147static struct pmap kernel_pmap_store; 148pmap_t kernel_pmap; 149 150vm_offset_t avail_start; /* PA of first available physical page */ 151vm_offset_t avail_end; /* PA of last available physical page */ 152vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 153vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 154static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 155static vm_offset_t vm_first_phys; 156int pgeflag; /* PG_G or-in */ 157int pseflag; /* PG_PS or-in */ 158int pv_npg; 159 160static int nkpt; 161static vm_page_t nkpg; 162vm_offset_t kernel_vm_end; 163 164extern vm_offset_t clean_sva, clean_eva; 165 166/* 167 * Data for the pv entry allocation mechanism 168 */ 169vm_zone_t pvzone; 170struct vm_zone pvzone_store; 171struct vm_object pvzone_obj; 172#define NPVINIT 8192 173struct pv_entry pvinit[NPVINIT]; 174 175/* 176 * All those kernel PT submaps that BSD is so fond of 177 */ 178pt_entry_t *CMAP1 = 0; 179static pt_entry_t *CMAP2, *ptmmap; 180static pv_table_t *pv_table; 181caddr_t CADDR1 = 0, ptvmmap = 0; 182static caddr_t CADDR2; 183static pt_entry_t *msgbufmap; 184struct msgbuf *msgbufp=0; 185 186#ifdef SMP 187extern char prv_CPAGE1[], prv_CPAGE2[], prv_CPAGE3[]; 188extern pt_entry_t *prv_CMAP1, *prv_CMAP2, *prv_CMAP3; 189extern pd_entry_t *IdlePTDS[]; 190extern pt_entry_t SMP_prvpt[]; 191#endif 192 193pt_entry_t *PMAP1 = 0; 194unsigned *PADDR1 = 0; 195 196static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 197static unsigned * get_ptbase __P((pmap_t pmap)); 198static pv_entry_t get_pv_entry __P((void)); 199static void i386_protection_init __P((void)); 200static void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 201 202static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 203static void pmap_remove_all __P((vm_offset_t pa)); 204static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 205 vm_offset_t pa, vm_page_t mpte)); 206static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 207 vm_offset_t sva)); 208static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 209static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 210 vm_offset_t va)); 211static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 212static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 213 vm_page_t mpte, vm_offset_t pa)); 214 215static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 216 217static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 218static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 219static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 220static vm_page_t pmap_page_alloc __P((vm_object_t object, vm_pindex_t pindex)); 221static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 222static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 223vm_offset_t pmap_kmem_choose(vm_offset_t addr) ; 224 225#define PDSTACKMAX 6 226static vm_offset_t pdstack[PDSTACKMAX]; 227static int pdstackptr; 228unsigned pdir4mb; 229 230/* 231 * Routine: pmap_pte 232 * Function: 233 * Extract the page table entry associated 234 * with the given map/virtual_address pair. 235 */ 236 237PMAP_INLINE unsigned * 238pmap_pte(pmap, va) 239 register pmap_t pmap; 240 vm_offset_t va; 241{ 242 unsigned *pdeaddr; 243 244 if (pmap) { 245 pdeaddr = (unsigned *) pmap_pde(pmap, va); 246 if (*pdeaddr & PG_PS) 247 return pdeaddr; 248 if (*pdeaddr) { 249 return get_ptbase(pmap) + i386_btop(va); 250 } 251 } 252 return (0); 253} 254 255/* 256 * Move the kernel virtual free pointer to the next 257 * 4MB. This is used to help improve performance 258 * by using a large (4MB) page for much of the kernel 259 * (.text, .data, .bss) 260 */ 261vm_offset_t 262pmap_kmem_choose(vm_offset_t addr) { 263 vm_offset_t newaddr = addr; 264#ifndef DISABLE_PSE 265 if (cpu_feature & CPUID_PSE) { 266 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 267 } 268#endif 269 return newaddr; 270} 271 272/* 273 * Bootstrap the system enough to run with virtual memory. 274 * 275 * On the i386 this is called after mapping has already been enabled 276 * and just syncs the pmap module with what has already been done. 277 * [We can't call it easily with mapping off since the kernel is not 278 * mapped with PA == VA, hence we would have to relocate every address 279 * from the linked base (virtual) address "KERNBASE" to the actual 280 * (physical) address starting relative to 0] 281 */ 282void 283pmap_bootstrap(firstaddr, loadaddr) 284 vm_offset_t firstaddr; 285 vm_offset_t loadaddr; 286{ 287 vm_offset_t va; 288 pt_entry_t *pte; 289 int i, j; 290 291 avail_start = firstaddr; 292 293 /* 294 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 295 * large. It should instead be correctly calculated in locore.s and 296 * not based on 'first' (which is a physical address, not a virtual 297 * address, for the start of unused physical memory). The kernel 298 * page tables are NOT double mapped and thus should not be included 299 * in this calculation. 300 */ 301 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 302 virtual_avail = pmap_kmem_choose(virtual_avail); 303 304 virtual_end = VM_MAX_KERNEL_ADDRESS; 305 306 /* 307 * Initialize protection array. 308 */ 309 i386_protection_init(); 310 311 /* 312 * The kernel's pmap is statically allocated so we don't have to use 313 * pmap_create, which is unlikely to work correctly at this part of 314 * the boot sequence (XXX and which no longer exists). 315 */ 316 kernel_pmap = &kernel_pmap_store; 317 318 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 319 320 kernel_pmap->pm_count = 1; 321 TAILQ_INIT(&kernel_pmap->pm_pvlist); 322 nkpt = NKPT; 323 324 /* 325 * Reserve some special page table entries/VA space for temporary 326 * mapping of pages. 327 */ 328#define SYSMAP(c, p, v, n) \ 329 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 330 331 va = virtual_avail; 332 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 333 334 /* 335 * CMAP1/CMAP2 are used for zeroing and copying pages. 336 */ 337 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 338 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 339 340 /* 341 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 342 * XXX ptmmap is not used. 343 */ 344 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 345 346 /* 347 * msgbufp is used to map the system message buffer. 348 * XXX msgbufmap is not used. 349 */ 350 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 351 atop(round_page(sizeof(struct msgbuf)))) 352 353 /* 354 * ptemap is used for pmap_pte_quick 355 */ 356 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 357 358 virtual_avail = va; 359 360 *(int *) CMAP1 = *(int *) CMAP2 = 0; 361 *(int *) PTD = 0; 362 363 364 pgeflag = 0; 365#if !defined(SMP) 366 if (cpu_feature & CPUID_PGE) { 367 pgeflag = PG_G; 368 } 369#endif 370 371/* 372 * Initialize the 4MB page size flag 373 */ 374 pseflag = 0; 375/* 376 * The 4MB page version of the initial 377 * kernel page mapping. 378 */ 379 pdir4mb = 0; 380 381#if !defined(DISABLE_PSE) 382 if (cpu_feature & CPUID_PSE) { 383 unsigned ptditmp; 384 /* 385 * Enable the PSE mode 386 */ 387 load_cr4(rcr4() | CR4_PSE); 388 389 /* 390 * Note that we have enabled PSE mode 391 */ 392 pseflag = PG_PS; 393 ptditmp = (unsigned) kernel_pmap->pm_pdir[KPTDI]; 394 ptditmp &= ~(NBPDR - 1); 395 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 396 pdir4mb = ptditmp; 397 /* 398 * We can do the mapping here for the single processor 399 * case. We simply ignore the old page table page from 400 * now on. 401 */ 402#if !defined(SMP) 403 PTD[KPTDI] = (pd_entry_t) ptditmp; 404 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 405 invltlb(); 406#endif 407 } 408#endif 409 410#ifdef SMP 411 if (cpu_apic_address == 0) 412 panic("pmap_bootstrap: no local apic!"); 413 414 /* 0 = private page */ 415 /* 1 = page table page */ 416 /* 2 = local apic */ 417 /* 16-31 = io apics */ 418 SMP_prvpt[2] = (pt_entry_t)(PG_V | PG_RW | pgeflag | ((u_long)cpu_apic_address & PG_FRAME)); 419 420 for (i = 0; i < mp_napics; i++) { 421 for (j = 0; j < 16; j++) { 422 /* same page frame as a previous IO apic? */ 423 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 424 ((u_long)io_apic_address[0] & PG_FRAME)) { 425 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 426 break; 427 } 428 /* use this slot if available */ 429 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 0) { 430 SMP_prvpt[j + 16] = (pt_entry_t)(PG_V | PG_RW | pgeflag | 431 ((u_long)io_apic_address[i] & PG_FRAME)); 432 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 433 break; 434 } 435 } 436 if (j == 16) 437 panic("no space to map IO apic %d!", i); 438 } 439 440 /* BSP does this itself, AP's get it pre-set */ 441 prv_CMAP1 = (pt_entry_t *)&SMP_prvpt[4]; 442 prv_CMAP2 = (pt_entry_t *)&SMP_prvpt[5]; 443 prv_CMAP3 = (pt_entry_t *)&SMP_prvpt[6]; 444#endif 445 446 invltlb(); 447 448} 449 450/* 451 * Set 4mb pdir for mp startup, and global flags 452 */ 453void 454pmap_set_opt(unsigned *pdir) { 455 int i; 456 457 if (pseflag && (cpu_feature & CPUID_PSE)) { 458 load_cr4(rcr4() | CR4_PSE); 459 if (pdir4mb) { 460 (unsigned) pdir[KPTDI] = pdir4mb; 461 } 462 } 463 464 if (pgeflag && (cpu_feature & CPUID_PGE)) { 465 load_cr4(rcr4() | CR4_PGE); 466 for(i = KPTDI; i < KPTDI + nkpt; i++) { 467 if (pdir[i]) { 468 pdir[i] |= PG_G; 469 } 470 } 471 } 472} 473 474/* 475 * Setup the PTD for the boot processor 476 */ 477void 478pmap_set_opt_bsp(void) 479{ 480 pmap_set_opt((unsigned *)kernel_pmap->pm_pdir); 481 pmap_set_opt((unsigned *)PTD); 482 invltlb(); 483} 484 485/* 486 * Initialize the pmap module. 487 * Called by vm_init, to initialize any structures that the pmap 488 * system needs to map virtual memory. 489 * pmap_init has been enhanced to support in a fairly consistant 490 * way, discontiguous physical memory. 491 */ 492void 493pmap_init(phys_start, phys_end) 494 vm_offset_t phys_start, phys_end; 495{ 496 vm_offset_t addr; 497 vm_size_t s; 498 int i; 499 500 /* 501 * calculate the number of pv_entries needed 502 */ 503 vm_first_phys = phys_avail[0]; 504 for (i = 0; phys_avail[i + 1]; i += 2); 505 pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 506 507 /* 508 * Allocate memory for random pmap data structures. Includes the 509 * pv_head_table. 510 */ 511 s = (vm_size_t) (sizeof(pv_table_t) * pv_npg); 512 s = round_page(s); 513 514 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 515 pv_table = (pv_table_t *) addr; 516 for(i = 0; i < pv_npg; i++) { 517 vm_offset_t pa; 518 TAILQ_INIT(&pv_table[i].pv_list); 519 pv_table[i].pv_list_count = 0; 520 pa = vm_first_phys + i * PAGE_SIZE; 521 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 522 } 523 524 /* 525 * init the pv free list 526 */ 527 pvzone = &pvzone_store; 528 zbootinit(pvzone, "PV ENTRY", sizeof(pvinit[0]), pvinit, NPVINIT); 529 530 /* 531 * Now it is safe to enable pv_table recording. 532 */ 533 pmap_initialized = TRUE; 534} 535 536void 537pmap_init2() { 538 zinitna(pvzone, &pvzone_obj, NULL, 0, 539 PMAP_SHPGPERPROC * maxproc + pv_npg, ZONE_INTERRUPT, 4); 540} 541 542/* 543 * Used to map a range of physical addresses into kernel 544 * virtual address space. 545 * 546 * For now, VM is already on, we only need to map the 547 * specified memory. 548 */ 549vm_offset_t 550pmap_map(virt, start, end, prot) 551 vm_offset_t virt; 552 vm_offset_t start; 553 vm_offset_t end; 554 int prot; 555{ 556 while (start < end) { 557 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 558 virt += PAGE_SIZE; 559 start += PAGE_SIZE; 560 } 561 return (virt); 562} 563 564 565/*************************************************** 566 * Low level helper routines..... 567 ***************************************************/ 568 569#if defined(PMAP_DIAGNOSTIC) 570 571/* 572 * This code checks for non-writeable/modified pages. 573 * This should be an invalid condition. 574 */ 575static int 576pmap_nw_modified(pt_entry_t ptea) { 577 int pte; 578 579 pte = (int) ptea; 580 581 if ((pte & (PG_M|PG_RW)) == PG_M) 582 return 1; 583 else 584 return 0; 585} 586#endif 587 588 589/* 590 * this routine defines the region(s) of memory that should 591 * not be tested for the modified bit. 592 */ 593static PMAP_INLINE int 594pmap_track_modified( vm_offset_t va) { 595 if ((va < clean_sva) || (va >= clean_eva)) 596 return 1; 597 else 598 return 0; 599} 600 601static PMAP_INLINE void 602invltlb_1pg( vm_offset_t va) { 603#if defined(I386_CPU) 604 if (cpu_class == CPUCLASS_386) { 605 invltlb(); 606 } else 607#endif 608 { 609 invlpg(va); 610 } 611} 612 613static PMAP_INLINE void 614invltlb_2pg( vm_offset_t va1, vm_offset_t va2) { 615#if defined(I386_CPU) 616 if (cpu_class == CPUCLASS_386) { 617 invltlb(); 618 } else 619#endif 620 { 621 invlpg(va1); 622 invlpg(va2); 623 } 624} 625 626static unsigned * 627get_ptbase(pmap) 628 pmap_t pmap; 629{ 630 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 631 632 /* are we current address space or kernel? */ 633 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 634 return (unsigned *) PTmap; 635 } 636 /* otherwise, we are alternate address space */ 637 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 638 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 639 invltlb(); 640 } 641 return (unsigned *) APTmap; 642} 643 644/* 645 * Super fast pmap_pte routine best used when scanning 646 * the pv lists. This eliminates many coarse-grained 647 * invltlb calls. Note that many of the pv list 648 * scans are across different pmaps. It is very wasteful 649 * to do an entire invltlb for checking a single mapping. 650 */ 651 652static unsigned * 653pmap_pte_quick(pmap, va) 654 register pmap_t pmap; 655 vm_offset_t va; 656{ 657 unsigned pde, newpf; 658 if (pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) { 659 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 660 unsigned index = i386_btop(va); 661 /* are we current address space or kernel? */ 662 if ((pmap == kernel_pmap) || 663 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 664 return (unsigned *) PTmap + index; 665 } 666 newpf = pde & PG_FRAME; 667 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 668 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 669 invltlb_1pg((vm_offset_t) PADDR1); 670 } 671 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 672 } 673 return (0); 674} 675 676/* 677 * Routine: pmap_extract 678 * Function: 679 * Extract the physical page address associated 680 * with the given map/virtual_address pair. 681 */ 682vm_offset_t 683pmap_extract(pmap, va) 684 register pmap_t pmap; 685 vm_offset_t va; 686{ 687 vm_offset_t rtval; 688 vm_offset_t pdirindex; 689 pdirindex = va >> PDRSHIFT; 690 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 691 unsigned *pte; 692 if ((rtval & PG_PS) != 0) { 693 rtval &= ~(NBPDR - 1); 694 rtval |= va & (NBPDR - 1); 695 return rtval; 696 } 697 pte = get_ptbase(pmap) + i386_btop(va); 698 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 699 return rtval; 700 } 701 return 0; 702 703} 704 705/* 706 * determine if a page is managed (memory vs. device) 707 */ 708static PMAP_INLINE int 709pmap_is_managed(pa) 710 vm_offset_t pa; 711{ 712 int i; 713 714 if (!pmap_initialized) 715 return 0; 716 717 for (i = 0; phys_avail[i + 1]; i += 2) { 718 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 719 return 1; 720 } 721 return 0; 722} 723 724 725/*************************************************** 726 * Low level mapping routines..... 727 ***************************************************/ 728 729/* 730 * Add a list of wired pages to the kva 731 * this routine is only used for temporary 732 * kernel mappings that do not need to have 733 * page modification or references recorded. 734 * Note that old mappings are simply written 735 * over. The page *must* be wired. 736 */ 737void 738pmap_qenter(va, m, count) 739 vm_offset_t va; 740 vm_page_t *m; 741 int count; 742{ 743 int i; 744 register unsigned *pte; 745 746 for (i = 0; i < count; i++) { 747 vm_offset_t tva = va + i * PAGE_SIZE; 748 unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | pgeflag; 749 unsigned opte; 750 pte = (unsigned *)vtopte(tva); 751 opte = *pte; 752 *pte = npte; 753 if (opte) 754 invltlb_1pg(tva); 755 } 756} 757 758/* 759 * this routine jerks page mappings from the 760 * kernel -- it is meant only for temporary mappings. 761 */ 762void 763pmap_qremove(va, count) 764 vm_offset_t va; 765 int count; 766{ 767 int i; 768 register unsigned *pte; 769 770 for (i = 0; i < count; i++) { 771 pte = (unsigned *)vtopte(va); 772 *pte = 0; 773 invltlb_1pg(va); 774 va += PAGE_SIZE; 775 } 776} 777 778/* 779 * add a wired page to the kva 780 * note that in order for the mapping to take effect -- you 781 * should do a invltlb after doing the pmap_kenter... 782 */ 783PMAP_INLINE void 784pmap_kenter(va, pa) 785 vm_offset_t va; 786 register vm_offset_t pa; 787{ 788 register unsigned *pte; 789 unsigned npte, opte; 790 791 npte = pa | PG_RW | PG_V | pgeflag; 792 pte = (unsigned *)vtopte(va); 793 opte = *pte; 794 *pte = npte; 795 if (opte) 796 invltlb_1pg(va); 797} 798 799/* 800 * remove a page from the kernel pagetables 801 */ 802PMAP_INLINE void 803pmap_kremove(va) 804 vm_offset_t va; 805{ 806 register unsigned *pte; 807 808 pte = (unsigned *)vtopte(va); 809 *pte = 0; 810 invltlb_1pg(va); 811} 812 813static vm_page_t 814pmap_page_alloc(object, pindex) 815 vm_object_t object; 816 vm_pindex_t pindex; 817{ 818 vm_page_t m; 819 m = vm_page_alloc(object, pindex, VM_ALLOC_ZERO); 820 if (m == NULL) { 821 VM_WAIT; 822 } 823 return m; 824} 825 826static vm_page_t 827pmap_page_lookup(object, pindex) 828 vm_object_t object; 829 vm_pindex_t pindex; 830{ 831 vm_page_t m; 832retry: 833 m = vm_page_lookup(object, pindex); 834 if (m) { 835 if (m->flags & PG_BUSY) { 836 m->flags |= PG_WANTED; 837 tsleep(m, PVM, "pplookp", 0); 838 goto retry; 839 } 840 } 841 842 return m; 843} 844 845/* 846 * Create the UPAGES for a new process. 847 * This routine directly affects the fork perf for a process. 848 */ 849void 850pmap_new_proc(p) 851 struct proc *p; 852{ 853 int i; 854 vm_object_t upobj; 855 vm_page_t m; 856 struct user *up; 857 unsigned *ptek; 858 859 /* 860 * allocate object for the upages 861 */ 862 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 863 p->p_upages_obj = upobj; 864 865 /* get a kernel virtual address for the UPAGES for this proc */ 866 up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * PAGE_SIZE); 867 if (up == NULL) 868 panic("pmap_new_proc: u_map allocation failed"); 869 870 ptek = (unsigned *) vtopte((vm_offset_t) up); 871 872 for(i=0;i<UPAGES;i++) { 873 /* 874 * Get a kernel stack page 875 */ 876 while ((m = vm_page_alloc(upobj, 877 i, VM_ALLOC_NORMAL)) == NULL) { 878 VM_WAIT; 879 } 880 881 /* 882 * Wire the page 883 */ 884 m->wire_count++; 885 ++cnt.v_wire_count; 886 887 /* 888 * Enter the page into the kernel address space. 889 */ 890 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 891 892 m->flags &= ~(PG_ZERO|PG_BUSY); 893 m->flags |= PG_MAPPED|PG_WRITEABLE; 894 m->valid = VM_PAGE_BITS_ALL; 895 } 896 897 p->p_addr = up; 898} 899 900/* 901 * Dispose the UPAGES for a process that has exited. 902 * This routine directly impacts the exit perf of a process. 903 */ 904void 905pmap_dispose_proc(p) 906 struct proc *p; 907{ 908 int i; 909 vm_object_t upobj; 910 vm_page_t m; 911 unsigned *ptek; 912 913 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 914 915 upobj = p->p_upages_obj; 916 917 for(i=0;i<UPAGES;i++) { 918 unsigned oldpte; 919 if ((m = vm_page_lookup(upobj, i)) == NULL) 920 panic("pmap_dispose_proc: upage already missing???"); 921 oldpte = *(ptek + i); 922 *(ptek + i) = 0; 923 if (oldpte & PG_G) 924 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 925 vm_page_unwire(m); 926 vm_page_free(m); 927 } 928 929 vm_object_deallocate(upobj); 930 931 kmem_free(u_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 932} 933 934/* 935 * Allow the UPAGES for a process to be prejudicially paged out. 936 */ 937void 938pmap_swapout_proc(p) 939 struct proc *p; 940{ 941 int i; 942 vm_object_t upobj; 943 vm_page_t m; 944 945 upobj = p->p_upages_obj; 946 /* 947 * let the upages be paged 948 */ 949 for(i=0;i<UPAGES;i++) { 950 if ((m = vm_page_lookup(upobj, i)) == NULL) 951 panic("pmap_swapout_proc: upage already missing???"); 952 m->dirty = VM_PAGE_BITS_ALL; 953 vm_page_unwire(m); 954 vm_page_deactivate(m); 955 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 956 } 957} 958 959/* 960 * Bring the UPAGES for a specified process back in. 961 */ 962void 963pmap_swapin_proc(p) 964 struct proc *p; 965{ 966 int i; 967 vm_object_t upobj; 968 vm_page_t m; 969 unsigned *pte; 970 971 upobj = p->p_upages_obj; 972 for(i=0;i<UPAGES;i++) { 973 int s; 974 s = splvm(); 975retry: 976 if ((m = vm_page_lookup(upobj, i)) == NULL) { 977 if ((m = vm_page_alloc(upobj, i, VM_ALLOC_NORMAL)) == NULL) { 978 VM_WAIT; 979 goto retry; 980 } 981 } else { 982 if ((m->flags & PG_BUSY) || m->busy) { 983 m->flags |= PG_WANTED; 984 tsleep(m, PVM, "swinuw",0); 985 goto retry; 986 } 987 m->flags |= PG_BUSY; 988 } 989 vm_page_wire(m); 990 splx(s); 991 992 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 993 VM_PAGE_TO_PHYS(m)); 994 995 if (m->valid != VM_PAGE_BITS_ALL) { 996 int rv; 997 rv = vm_pager_get_pages(upobj, &m, 1, 0); 998 if (rv != VM_PAGER_OK) 999 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 1000 m->valid = VM_PAGE_BITS_ALL; 1001 } 1002 PAGE_WAKEUP(m); 1003 m->flags |= PG_MAPPED|PG_WRITEABLE; 1004 } 1005} 1006 1007/*************************************************** 1008 * Page table page management routines..... 1009 ***************************************************/ 1010 1011/* 1012 * This routine unholds page table pages, and if the hold count 1013 * drops to zero, then it decrements the wire count. 1014 */ 1015static int 1016_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1017 int s; 1018 1019 if (m->flags & PG_BUSY) { 1020 s = splvm(); 1021 while (m->flags & PG_BUSY) { 1022 m->flags |= PG_WANTED; 1023 tsleep(m, PVM, "pmuwpt", 0); 1024 } 1025 splx(s); 1026 } 1027 1028 if (m->hold_count == 0) { 1029 vm_offset_t pteva; 1030 /* 1031 * unmap the page table page 1032 */ 1033 pmap->pm_pdir[m->pindex] = 0; 1034 --pmap->pm_stats.resident_count; 1035 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1036 (((unsigned) PTDpde) & PG_FRAME)) { 1037 /* 1038 * Do a invltlb to make the invalidated mapping 1039 * take effect immediately. 1040 */ 1041 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1042 invltlb_1pg(pteva); 1043 } 1044 1045#if defined(PTPHINT) 1046 if (pmap->pm_ptphint == m) 1047 pmap->pm_ptphint = NULL; 1048#endif 1049 1050 /* 1051 * If the page is finally unwired, simply free it. 1052 */ 1053 --m->wire_count; 1054 if (m->wire_count == 0) { 1055 1056 if (m->flags & PG_WANTED) { 1057 m->flags &= ~PG_WANTED; 1058 wakeup(m); 1059 } 1060 1061 vm_page_free_zero(m); 1062 --cnt.v_wire_count; 1063 } 1064 return 1; 1065 } 1066 return 0; 1067} 1068 1069__inline static int 1070pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1071 vm_page_unhold(m); 1072 if (m->hold_count == 0) 1073 return _pmap_unwire_pte_hold(pmap, m); 1074 else 1075 return 0; 1076} 1077 1078/* 1079 * After removing a page table entry, this routine is used to 1080 * conditionally free the page, and manage the hold/wire counts. 1081 */ 1082static int 1083pmap_unuse_pt(pmap, va, mpte) 1084 pmap_t pmap; 1085 vm_offset_t va; 1086 vm_page_t mpte; 1087{ 1088 unsigned ptepindex; 1089 if (va >= UPT_MIN_ADDRESS) 1090 return 0; 1091 1092 if (mpte == NULL) { 1093 ptepindex = (va >> PDRSHIFT); 1094#if defined(PTPHINT) 1095 if (pmap->pm_ptphint && 1096 (pmap->pm_ptphint->pindex == ptepindex)) { 1097 mpte = pmap->pm_ptphint; 1098 } else { 1099 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1100 pmap->pm_ptphint = mpte; 1101 } 1102#else 1103 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1104#endif 1105 } 1106 1107 return pmap_unwire_pte_hold(pmap, mpte); 1108} 1109 1110/* 1111 * Initialize a preallocated and zeroed pmap structure, 1112 * such as one in a vmspace structure. 1113 */ 1114void 1115pmap_pinit(pmap) 1116 register struct pmap *pmap; 1117{ 1118 vm_page_t ptdpg; 1119 /* 1120 * No need to allocate page table space yet but we do need a valid 1121 * page directory table. 1122 */ 1123 1124 if (pdstackptr > 0) { 1125 --pdstackptr; 1126 pmap->pm_pdir = (pd_entry_t *)pdstack[pdstackptr]; 1127 } else { 1128 pmap->pm_pdir = 1129 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1130 } 1131 1132 /* 1133 * allocate object for the ptes 1134 */ 1135 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1136 1137 /* 1138 * allocate the page directory page 1139 */ 1140retry: 1141 ptdpg = pmap_page_alloc( pmap->pm_pteobj, PTDPTDI); 1142 if (ptdpg == NULL) 1143 goto retry; 1144 1145 ptdpg->wire_count = 1; 1146 ++cnt.v_wire_count; 1147 1148 ptdpg->flags &= ~(PG_MAPPED|PG_BUSY); /* not mapped normally */ 1149 ptdpg->valid = VM_PAGE_BITS_ALL; 1150 1151 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1152 if ((ptdpg->flags & PG_ZERO) == 0) 1153 bzero(pmap->pm_pdir, PAGE_SIZE); 1154 1155 /* wire in kernel global address entries */ 1156 /* XXX copies current process, does not fill in MPPTDI */ 1157 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1158 1159 /* install self-referential address mapping entry */ 1160 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1161 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW; 1162 1163 pmap->pm_flags = 0; 1164 pmap->pm_count = 1; 1165 pmap->pm_ptphint = NULL; 1166 TAILQ_INIT(&pmap->pm_pvlist); 1167} 1168 1169static int 1170pmap_release_free_page(pmap, p) 1171 struct pmap *pmap; 1172 vm_page_t p; 1173{ 1174 int s; 1175 unsigned *pde = (unsigned *) pmap->pm_pdir; 1176 /* 1177 * This code optimizes the case of freeing non-busy 1178 * page-table pages. Those pages are zero now, and 1179 * might as well be placed directly into the zero queue. 1180 */ 1181 s = splvm(); 1182 if (p->flags & PG_BUSY) { 1183 p->flags |= PG_WANTED; 1184 tsleep(p, PVM, "pmaprl", 0); 1185 splx(s); 1186 return 0; 1187 } 1188 1189 if (p->flags & PG_WANTED) { 1190 p->flags &= ~PG_WANTED; 1191 wakeup(p); 1192 } 1193 1194 /* 1195 * Remove the page table page from the processes address space. 1196 */ 1197 pde[p->pindex] = 0; 1198 --pmap->pm_stats.resident_count; 1199 1200 if (p->hold_count) { 1201 panic("pmap_release: freeing held page table page"); 1202 } 1203 /* 1204 * Page directory pages need to have the kernel 1205 * stuff cleared, so they can go into the zero queue also. 1206 */ 1207 if (p->pindex == PTDPTDI) { 1208 bzero(pde + KPTDI, nkpt * PTESIZE); 1209#ifdef SMP 1210 pde[MPPTDI] = 0; 1211#endif 1212 pde[APTDPTDI] = 0; 1213 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1214 } 1215 1216#if defined(PTPHINT) 1217 if (pmap->pm_ptphint && 1218 (pmap->pm_ptphint->pindex == p->pindex)) 1219 pmap->pm_ptphint = NULL; 1220#endif 1221 1222 vm_page_free_zero(p); 1223 splx(s); 1224 return 1; 1225} 1226 1227/* 1228 * this routine is called if the page table page is not 1229 * mapped correctly. 1230 */ 1231static vm_page_t 1232_pmap_allocpte(pmap, ptepindex) 1233 pmap_t pmap; 1234 unsigned ptepindex; 1235{ 1236 vm_offset_t pteva, ptepa; 1237 vm_page_t m; 1238 int needszero = 0; 1239 1240 /* 1241 * Find or fabricate a new pagetable page 1242 */ 1243retry: 1244 m = vm_page_lookup(pmap->pm_pteobj, ptepindex); 1245 if (m == NULL) { 1246 m = pmap_page_alloc(pmap->pm_pteobj, ptepindex); 1247 if (m == NULL) 1248 goto retry; 1249 if ((m->flags & PG_ZERO) == 0) 1250 needszero = 1; 1251 m->flags &= ~(PG_ZERO|PG_BUSY); 1252 m->valid = VM_PAGE_BITS_ALL; 1253 } else { 1254 if ((m->flags & PG_BUSY) || m->busy) { 1255 m->flags |= PG_WANTED; 1256 tsleep(m, PVM, "ptewai", 0); 1257 goto retry; 1258 } 1259 } 1260 1261 if (m->queue != PQ_NONE) { 1262 int s = splvm(); 1263 vm_page_unqueue(m); 1264 splx(s); 1265 } 1266 1267 if (m->wire_count == 0) 1268 ++cnt.v_wire_count; 1269 ++m->wire_count; 1270 1271 /* 1272 * Increment the hold count for the page table page 1273 * (denoting a new mapping.) 1274 */ 1275 ++m->hold_count; 1276 1277 /* 1278 * Map the pagetable page into the process address space, if 1279 * it isn't already there. 1280 */ 1281 1282 pmap->pm_stats.resident_count++; 1283 1284 ptepa = VM_PAGE_TO_PHYS(m); 1285 pmap->pm_pdir[ptepindex] = (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V); 1286 1287#if defined(PTPHINT) 1288 /* 1289 * Set the page table hint 1290 */ 1291 pmap->pm_ptphint = m; 1292#endif 1293 1294 /* 1295 * Try to use the new mapping, but if we cannot, then 1296 * do it with the routine that maps the page explicitly. 1297 */ 1298 if (needszero) { 1299 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1300 (((unsigned) PTDpde) & PG_FRAME)) { 1301 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1302 bzero((caddr_t) pteva, PAGE_SIZE); 1303 } else { 1304 pmap_zero_page(ptepa); 1305 } 1306 } 1307 1308 m->valid = VM_PAGE_BITS_ALL; 1309 m->flags |= PG_MAPPED; 1310 1311 return m; 1312} 1313 1314static vm_page_t 1315pmap_allocpte(pmap, va) 1316 pmap_t pmap; 1317 vm_offset_t va; 1318{ 1319 unsigned ptepindex; 1320 vm_offset_t ptepa; 1321 vm_page_t m; 1322 1323 /* 1324 * Calculate pagetable page index 1325 */ 1326 ptepindex = va >> PDRSHIFT; 1327 1328 /* 1329 * Get the page directory entry 1330 */ 1331 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1332 1333 /* 1334 * This supports switching from a 4MB page to a 1335 * normal 4K page. 1336 */ 1337 if (ptepa & PG_PS) { 1338 pmap->pm_pdir[ptepindex] = 0; 1339 ptepa = 0; 1340 invltlb(); 1341 } 1342 1343 /* 1344 * If the page table page is mapped, we just increment the 1345 * hold count, and activate it. 1346 */ 1347 if (ptepa) { 1348#if defined(PTPHINT) 1349 /* 1350 * In order to get the page table page, try the 1351 * hint first. 1352 */ 1353 if (pmap->pm_ptphint && 1354 (pmap->pm_ptphint->pindex == ptepindex)) { 1355 m = pmap->pm_ptphint; 1356 } else { 1357 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1358 pmap->pm_ptphint = m; 1359 } 1360#else 1361 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1362#endif 1363 ++m->hold_count; 1364 return m; 1365 } 1366 /* 1367 * Here if the pte page isn't mapped, or if it has been deallocated. 1368 */ 1369 return _pmap_allocpte(pmap, ptepindex); 1370} 1371 1372 1373/*************************************************** 1374* Pmap allocation/deallocation routines. 1375 ***************************************************/ 1376 1377/* 1378 * Release any resources held by the given physical map. 1379 * Called when a pmap initialized by pmap_pinit is being released. 1380 * Should only be called if the map contains no valid mappings. 1381 */ 1382void 1383pmap_release(pmap) 1384 register struct pmap *pmap; 1385{ 1386 vm_page_t p,n,ptdpg; 1387 vm_object_t object = pmap->pm_pteobj; 1388 1389#if defined(DIAGNOSTIC) 1390 if (object->ref_count != 1) 1391 panic("pmap_release: pteobj reference count != 1"); 1392#endif 1393 1394 ptdpg = NULL; 1395retry: 1396 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1397 n = TAILQ_NEXT(p, listq); 1398 if (p->pindex == PTDPTDI) { 1399 ptdpg = p; 1400 continue; 1401 } 1402 if (!pmap_release_free_page(pmap, p)) 1403 goto retry; 1404 } 1405 1406 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1407 goto retry; 1408 1409 vm_object_deallocate(object); 1410 if (pdstackptr < PDSTACKMAX) { 1411 pdstack[pdstackptr] = (vm_offset_t) pmap->pm_pdir; 1412 ++pdstackptr; 1413 } else { 1414 int pdstmp = pdstackptr - 1; 1415 kmem_free(kernel_map, pdstack[pdstmp], PAGE_SIZE); 1416 pdstack[pdstmp] = (vm_offset_t) pmap->pm_pdir; 1417 } 1418 pmap->pm_pdir = 0; 1419} 1420 1421/* 1422 * grow the number of kernel page table entries, if needed 1423 */ 1424void 1425pmap_growkernel(vm_offset_t addr) 1426{ 1427 struct proc *p; 1428 struct pmap *pmap; 1429 int s; 1430#ifdef SMP 1431 int i; 1432#endif 1433 1434 s = splhigh(); 1435 if (kernel_vm_end == 0) { 1436 kernel_vm_end = KERNBASE; 1437 nkpt = 0; 1438 while (pdir_pde(PTD, kernel_vm_end)) { 1439 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1440 ++nkpt; 1441 } 1442 } 1443 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1444 while (kernel_vm_end < addr) { 1445 if (pdir_pde(PTD, kernel_vm_end)) { 1446 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1447 continue; 1448 } 1449 ++nkpt; 1450 if (!nkpg) { 1451 vm_offset_t ptpkva = (vm_offset_t) vtopte(addr); 1452 /* 1453 * This index is bogus, but out of the way 1454 */ 1455 vm_pindex_t ptpidx = (ptpkva >> PAGE_SHIFT); 1456 nkpg = vm_page_alloc(kernel_object, 1457 ptpidx, VM_ALLOC_SYSTEM); 1458 if (!nkpg) 1459 panic("pmap_growkernel: no memory to grow kernel"); 1460 vm_page_wire(nkpg); 1461 vm_page_remove(nkpg); 1462 pmap_zero_page(VM_PAGE_TO_PHYS(nkpg)); 1463 } 1464 pdir_pde(PTD, kernel_vm_end) = (pd_entry_t) (VM_PAGE_TO_PHYS(nkpg) | PG_V | PG_RW | pgeflag); 1465 1466#ifdef SMP 1467 for (i = 0; i < mp_naps; i++) { 1468 if (IdlePTDS[i]) 1469 pdir_pde(IdlePTDS[i], kernel_vm_end) = (pd_entry_t) (VM_PAGE_TO_PHYS(nkpg) | PG_V | PG_RW | pgeflag); 1470 } 1471#endif 1472 1473 nkpg = NULL; 1474 1475 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1476 if (p->p_vmspace) { 1477 pmap = &p->p_vmspace->vm_pmap; 1478 *pmap_pde(pmap, kernel_vm_end) = pdir_pde(PTD, kernel_vm_end); 1479 } 1480 } 1481 *pmap_pde(kernel_pmap, kernel_vm_end) = pdir_pde(PTD, kernel_vm_end); 1482 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1483 } 1484 splx(s); 1485} 1486 1487/* 1488 * Retire the given physical map from service. 1489 * Should only be called if the map contains 1490 * no valid mappings. 1491 */ 1492void 1493pmap_destroy(pmap) 1494 register pmap_t pmap; 1495{ 1496 int count; 1497 1498 if (pmap == NULL) 1499 return; 1500 1501 count = --pmap->pm_count; 1502 if (count == 0) { 1503 pmap_release(pmap); 1504 free((caddr_t) pmap, M_VMPMAP); 1505 } 1506} 1507 1508/* 1509 * Add a reference to the specified pmap. 1510 */ 1511void 1512pmap_reference(pmap) 1513 pmap_t pmap; 1514{ 1515 if (pmap != NULL) { 1516 pmap->pm_count++; 1517 } 1518} 1519 1520/*************************************************** 1521* page management routines. 1522 ***************************************************/ 1523 1524/* 1525 * free the pv_entry back to the free list 1526 */ 1527static inline void 1528free_pv_entry(pv) 1529 pv_entry_t pv; 1530{ 1531 zfreei(pvzone, pv); 1532} 1533 1534/* 1535 * get a new pv_entry, allocating a block from the system 1536 * when needed. 1537 * the memory allocation is performed bypassing the malloc code 1538 * because of the possibility of allocations at interrupt time. 1539 */ 1540static inline pv_entry_t 1541get_pv_entry(void) 1542{ 1543 return zalloci(pvzone); 1544} 1545 1546/* 1547 * If it is the first entry on the list, it is actually 1548 * in the header and we must copy the following entry up 1549 * to the header. Otherwise we must search the list for 1550 * the entry. In either case we free the now unused entry. 1551 */ 1552 1553static int 1554pmap_remove_entry(pmap, ppv, va) 1555 struct pmap *pmap; 1556 pv_table_t *ppv; 1557 vm_offset_t va; 1558{ 1559 pv_entry_t pv; 1560 int rtval; 1561 int s; 1562 1563 s = splvm(); 1564 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1565 for (pv = TAILQ_FIRST(&ppv->pv_list); 1566 pv; 1567 pv = TAILQ_NEXT(pv, pv_list)) { 1568 if (pmap == pv->pv_pmap && va == pv->pv_va) 1569 break; 1570 } 1571 } else { 1572 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1573 pv; 1574 pv = TAILQ_NEXT(pv, pv_plist)) { 1575 if (va == pv->pv_va) 1576 break; 1577 } 1578 } 1579 1580 rtval = 0; 1581 if (pv) { 1582 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1583 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1584 --ppv->pv_list_count; 1585 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 1586 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 1587 } 1588 1589 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1590 free_pv_entry(pv); 1591 } 1592 1593 splx(s); 1594 return rtval; 1595} 1596 1597/* 1598 * Create a pv entry for page at pa for 1599 * (pmap, va). 1600 */ 1601static void 1602pmap_insert_entry(pmap, va, mpte, pa) 1603 pmap_t pmap; 1604 vm_offset_t va; 1605 vm_page_t mpte; 1606 vm_offset_t pa; 1607{ 1608 1609 int s; 1610 pv_entry_t pv; 1611 pv_table_t *ppv; 1612 1613 s = splvm(); 1614 pv = get_pv_entry(); 1615 pv->pv_va = va; 1616 pv->pv_pmap = pmap; 1617 pv->pv_ptem = mpte; 1618 1619 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1620 1621 ppv = pa_to_pvh(pa); 1622 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1623 ++ppv->pv_list_count; 1624 1625 splx(s); 1626} 1627 1628/* 1629 * pmap_remove_pte: do the things to unmap a page in a process 1630 */ 1631static int 1632pmap_remove_pte(pmap, ptq, va) 1633 struct pmap *pmap; 1634 unsigned *ptq; 1635 vm_offset_t va; 1636{ 1637 unsigned oldpte; 1638 pv_table_t *ppv; 1639 1640 oldpte = *ptq; 1641 *ptq = 0; 1642 if (oldpte & PG_W) 1643 pmap->pm_stats.wired_count -= 1; 1644 /* 1645 * Machines that don't support invlpg, also don't support 1646 * PG_G. 1647 */ 1648 if (oldpte & PG_G) 1649 invlpg(va); 1650 pmap->pm_stats.resident_count -= 1; 1651 if (oldpte & PG_MANAGED) { 1652 ppv = pa_to_pvh(oldpte); 1653 if (oldpte & PG_M) { 1654#if defined(PMAP_DIAGNOSTIC) 1655 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1656 printf("pmap_remove: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, (int) oldpte); 1657 } 1658#endif 1659 if (pmap_track_modified(va)) 1660 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1661 } 1662 return pmap_remove_entry(pmap, ppv, va); 1663 } else { 1664 return pmap_unuse_pt(pmap, va, NULL); 1665 } 1666 1667 return 0; 1668} 1669 1670/* 1671 * Remove a single page from a process address space 1672 */ 1673static void 1674pmap_remove_page(pmap, va) 1675 struct pmap *pmap; 1676 register vm_offset_t va; 1677{ 1678 register unsigned *ptq; 1679 1680 /* 1681 * if there is no pte for this address, just skip it!!! 1682 */ 1683 if (*pmap_pde(pmap, va) == 0) { 1684 return; 1685 } 1686 1687 /* 1688 * get a local va for mappings for this pmap. 1689 */ 1690 ptq = get_ptbase(pmap) + i386_btop(va); 1691 if (*ptq) { 1692 (void) pmap_remove_pte(pmap, ptq, va); 1693 invltlb_1pg(va); 1694 } 1695 return; 1696} 1697 1698/* 1699 * Remove the given range of addresses from the specified map. 1700 * 1701 * It is assumed that the start and end are properly 1702 * rounded to the page size. 1703 */ 1704void 1705pmap_remove(pmap, sva, eva) 1706 struct pmap *pmap; 1707 register vm_offset_t sva; 1708 register vm_offset_t eva; 1709{ 1710 register unsigned *ptbase; 1711 vm_offset_t pdnxt; 1712 vm_offset_t ptpaddr; 1713 vm_offset_t sindex, eindex; 1714 int anyvalid; 1715 1716 if (pmap == NULL) 1717 return; 1718 1719 if (pmap->pm_stats.resident_count == 0) 1720 return; 1721 1722 /* 1723 * special handling of removing one page. a very 1724 * common operation and easy to short circuit some 1725 * code. 1726 */ 1727 if (((sva + PAGE_SIZE) == eva) && 1728 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1729 pmap_remove_page(pmap, sva); 1730 return; 1731 } 1732 1733 anyvalid = 0; 1734 1735 /* 1736 * Get a local virtual address for the mappings that are being 1737 * worked with. 1738 */ 1739 ptbase = get_ptbase(pmap); 1740 1741 sindex = i386_btop(sva); 1742 eindex = i386_btop(eva); 1743 1744 for (; sindex < eindex; sindex = pdnxt) { 1745 unsigned pdirindex; 1746 1747 /* 1748 * Calculate index for next page table. 1749 */ 1750 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1751 if (pmap->pm_stats.resident_count == 0) 1752 break; 1753 1754 pdirindex = sindex / NPDEPG; 1755 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1756 pmap->pm_pdir[pdirindex] = 0; 1757 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1758 anyvalid++; 1759 continue; 1760 } 1761 1762 /* 1763 * Weed out invalid mappings. Note: we assume that the page 1764 * directory table is always allocated, and in kernel virtual. 1765 */ 1766 if (ptpaddr == 0) 1767 continue; 1768 1769 /* 1770 * Limit our scan to either the end of the va represented 1771 * by the current page table page, or to the end of the 1772 * range being removed. 1773 */ 1774 if (pdnxt > eindex) { 1775 pdnxt = eindex; 1776 } 1777 1778 for ( ;sindex != pdnxt; sindex++) { 1779 vm_offset_t va; 1780 if (ptbase[sindex] == 0) { 1781 continue; 1782 } 1783 va = i386_ptob(sindex); 1784 1785 anyvalid++; 1786 if (pmap_remove_pte(pmap, 1787 ptbase + sindex, va)) 1788 break; 1789 } 1790 } 1791 1792 if (anyvalid) { 1793 invltlb(); 1794 } 1795} 1796 1797/* 1798 * Routine: pmap_remove_all 1799 * Function: 1800 * Removes this physical page from 1801 * all physical maps in which it resides. 1802 * Reflects back modify bits to the pager. 1803 * 1804 * Notes: 1805 * Original versions of this routine were very 1806 * inefficient because they iteratively called 1807 * pmap_remove (slow...) 1808 */ 1809 1810static void 1811pmap_remove_all(pa) 1812 vm_offset_t pa; 1813{ 1814 register pv_entry_t pv; 1815 pv_table_t *ppv; 1816 register unsigned *pte, tpte; 1817 int nmodify; 1818 int update_needed; 1819 int s; 1820 1821 nmodify = 0; 1822 update_needed = 0; 1823#if defined(PMAP_DIAGNOSTIC) 1824 /* 1825 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1826 * pages! 1827 */ 1828 if (!pmap_is_managed(pa)) { 1829 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%lx", pa); 1830 } 1831#endif 1832 1833 s = splvm(); 1834 ppv = pa_to_pvh(pa); 1835 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1836 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1837 1838 pv->pv_pmap->pm_stats.resident_count--; 1839 1840 tpte = *pte; 1841 *pte = 0; 1842 if (tpte & PG_W) 1843 pv->pv_pmap->pm_stats.wired_count--; 1844 /* 1845 * Update the vm_page_t clean and reference bits. 1846 */ 1847 if (tpte & PG_M) { 1848#if defined(PMAP_DIAGNOSTIC) 1849 if (pmap_nw_modified((pt_entry_t) tpte)) { 1850 printf("pmap_remove_all: modified page not writable: va: 0x%lx, pte: 0x%lx\n", pv->pv_va, tpte); 1851 } 1852#endif 1853 if (pmap_track_modified(pv->pv_va)) 1854 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1855 } 1856 if (!update_needed && 1857 ((!curproc || (&curproc->p_vmspace->vm_pmap == pv->pv_pmap)) || 1858 (pv->pv_pmap == kernel_pmap))) { 1859 update_needed = 1; 1860 } 1861 1862 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1863 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1864 --ppv->pv_list_count; 1865 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1866 free_pv_entry(pv); 1867 } 1868 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 1869 1870 1871 if (update_needed) 1872 invltlb(); 1873 splx(s); 1874 return; 1875} 1876 1877/* 1878 * Set the physical protection on the 1879 * specified range of this map as requested. 1880 */ 1881void 1882pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1883{ 1884 register unsigned *ptbase; 1885 vm_offset_t pdnxt; 1886 vm_offset_t ptpaddr; 1887 vm_offset_t sindex, eindex; 1888 int anychanged; 1889 1890 1891 if (pmap == NULL) 1892 return; 1893 1894 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1895 pmap_remove(pmap, sva, eva); 1896 return; 1897 } 1898 1899 anychanged = 0; 1900 1901 ptbase = get_ptbase(pmap); 1902 1903 sindex = i386_btop(sva); 1904 eindex = i386_btop(eva); 1905 1906 for (; sindex < eindex; sindex = pdnxt) { 1907 1908 unsigned pdirindex; 1909 1910 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1911 1912 pdirindex = sindex / NPDEPG; 1913 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1914 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1915 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1916 anychanged++; 1917 continue; 1918 } 1919 1920 /* 1921 * Weed out invalid mappings. Note: we assume that the page 1922 * directory table is always allocated, and in kernel virtual. 1923 */ 1924 if (ptpaddr == 0) 1925 continue; 1926 1927 if (pdnxt > eindex) { 1928 pdnxt = eindex; 1929 } 1930 1931 for (; sindex != pdnxt; sindex++) { 1932 1933 unsigned pbits = ptbase[sindex]; 1934 1935 if (prot & VM_PROT_WRITE) { 1936 if ((pbits & (PG_RW|PG_V)) == PG_V) { 1937 if (pbits & PG_MANAGED) { 1938 vm_page_t m = PHYS_TO_VM_PAGE(pbits); 1939 m->flags |= PG_WRITEABLE; 1940 m->object->flags |= OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY; 1941 } 1942 ptbase[sindex] = pbits | PG_RW; 1943 anychanged = 1; 1944 } 1945 } else if (pbits & PG_RW) { 1946 if (pbits & PG_M) { 1947 vm_offset_t sva = i386_ptob(sindex); 1948 if ((pbits & PG_MANAGED) && pmap_track_modified(sva)) { 1949 vm_page_t m = PHYS_TO_VM_PAGE(pbits); 1950 m->dirty = VM_PAGE_BITS_ALL; 1951 } 1952 } 1953 ptbase[sindex] = pbits & ~(PG_M|PG_RW); 1954 anychanged = 1; 1955 } 1956 } 1957 } 1958 if (anychanged) 1959 invltlb(); 1960} 1961 1962/* 1963 * Insert the given physical page (p) at 1964 * the specified virtual address (v) in the 1965 * target physical map with the protection requested. 1966 * 1967 * If specified, the page will be wired down, meaning 1968 * that the related pte can not be reclaimed. 1969 * 1970 * NB: This is the only routine which MAY NOT lazy-evaluate 1971 * or lose information. That is, this routine must actually 1972 * insert this page into the given map NOW. 1973 */ 1974void 1975pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot, 1976 boolean_t wired) 1977{ 1978 register unsigned *pte; 1979 vm_offset_t opa; 1980 vm_offset_t origpte, newpte; 1981 vm_page_t mpte; 1982 1983 if (pmap == NULL) 1984 return; 1985 1986 va &= PG_FRAME; 1987#ifdef PMAP_DIAGNOSTIC 1988 if (va > VM_MAX_KERNEL_ADDRESS) 1989 panic("pmap_enter: toobig"); 1990 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 1991 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 1992#endif 1993 1994 mpte = NULL; 1995 /* 1996 * In the case that a page table page is not 1997 * resident, we are creating it here. 1998 */ 1999 if (va < UPT_MIN_ADDRESS) 2000 mpte = pmap_allocpte(pmap, va); 2001 2002 pte = pmap_pte(pmap, va); 2003 /* 2004 * Page Directory table entry not valid, we need a new PT page 2005 */ 2006 if (pte == NULL) { 2007 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%lx\n", 2008 pmap->pm_pdir[PTDPTDI], va); 2009 } 2010 2011 origpte = *(vm_offset_t *)pte; 2012 pa &= PG_FRAME; 2013 opa = origpte & PG_FRAME; 2014 if (origpte & PG_PS) 2015 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2016 2017 /* 2018 * Mapping has not changed, must be protection or wiring change. 2019 */ 2020 if (origpte && (opa == pa)) { 2021 /* 2022 * Wiring change, just update stats. We don't worry about 2023 * wiring PT pages as they remain resident as long as there 2024 * are valid mappings in them. Hence, if a user page is wired, 2025 * the PT page will be also. 2026 */ 2027 if (wired && ((origpte & PG_W) == 0)) 2028 pmap->pm_stats.wired_count++; 2029 else if (!wired && (origpte & PG_W)) 2030 pmap->pm_stats.wired_count--; 2031 2032#if defined(PMAP_DIAGNOSTIC) 2033 if (pmap_nw_modified((pt_entry_t) origpte)) { 2034 printf("pmap_enter: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, origpte); 2035 } 2036#endif 2037 2038 /* 2039 * We might be turning off write access to the page, 2040 * so we go ahead and sense modify status. 2041 */ 2042 if (origpte & PG_MANAGED) { 2043 vm_page_t m; 2044 if (origpte & PG_M) { 2045 if (pmap_track_modified(va)) { 2046 m = PHYS_TO_VM_PAGE(pa); 2047 m->dirty = VM_PAGE_BITS_ALL; 2048 } 2049 } 2050 pa |= PG_MANAGED; 2051 } 2052 2053 if (mpte) 2054 --mpte->hold_count; 2055 2056 goto validate; 2057 } 2058 /* 2059 * Mapping has changed, invalidate old range and fall through to 2060 * handle validating new mapping. 2061 */ 2062 if (opa) { 2063 int err; 2064 err = pmap_remove_pte(pmap, pte, va); 2065 if (err) 2066 panic("pmap_enter: pte vanished, va: 0x%x", va); 2067 } 2068 2069 /* 2070 * Enter on the PV list if part of our managed memory Note that we 2071 * raise IPL while manipulating pv_table since pmap_enter can be 2072 * called at interrupt time. 2073 */ 2074 if (pmap_is_managed(pa)) { 2075 pmap_insert_entry(pmap, va, mpte, pa); 2076 pa |= PG_MANAGED; 2077 } 2078 2079 /* 2080 * Increment counters 2081 */ 2082 pmap->pm_stats.resident_count++; 2083 if (wired) 2084 pmap->pm_stats.wired_count++; 2085 2086validate: 2087 /* 2088 * Now validate mapping with desired protection/wiring. 2089 */ 2090 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2091 2092 if (wired) 2093 newpte |= PG_W; 2094 if (va < UPT_MIN_ADDRESS) 2095 newpte |= PG_U; 2096 if (pmap == kernel_pmap) 2097 newpte |= pgeflag; 2098 2099 /* 2100 * if the mapping or permission bits are different, we need 2101 * to update the pte. 2102 */ 2103 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2104 *pte = newpte; 2105 if (origpte) 2106 invltlb_1pg(va); 2107 } 2108} 2109 2110/* 2111 * this code makes some *MAJOR* assumptions: 2112 * 1. Current pmap & pmap exists. 2113 * 2. Not wired. 2114 * 3. Read access. 2115 * 4. No page table pages. 2116 * 5. Tlbflush is deferred to calling procedure. 2117 * 6. Page IS managed. 2118 * but is *MUCH* faster than pmap_enter... 2119 */ 2120 2121static vm_page_t 2122pmap_enter_quick(pmap, va, pa, mpte) 2123 register pmap_t pmap; 2124 vm_offset_t va; 2125 register vm_offset_t pa; 2126 vm_page_t mpte; 2127{ 2128 register unsigned *pte; 2129 2130 /* 2131 * In the case that a page table page is not 2132 * resident, we are creating it here. 2133 */ 2134 if (va < UPT_MIN_ADDRESS) { 2135 unsigned ptepindex; 2136 vm_offset_t ptepa; 2137 2138 /* 2139 * Calculate pagetable page index 2140 */ 2141 ptepindex = va >> PDRSHIFT; 2142 if (mpte && (mpte->pindex == ptepindex)) { 2143 ++mpte->hold_count; 2144 } else { 2145retry: 2146 /* 2147 * Get the page directory entry 2148 */ 2149 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2150 2151 /* 2152 * If the page table page is mapped, we just increment 2153 * the hold count, and activate it. 2154 */ 2155 if (ptepa) { 2156 if (ptepa & PG_PS) 2157 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2158#if defined(PTPHINT) 2159 if (pmap->pm_ptphint && 2160 (pmap->pm_ptphint->pindex == ptepindex)) { 2161 mpte = pmap->pm_ptphint; 2162 } else { 2163 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2164 pmap->pm_ptphint = mpte; 2165 } 2166#else 2167 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2168#endif 2169 if (mpte == NULL) 2170 goto retry; 2171 ++mpte->hold_count; 2172 } else { 2173 mpte = _pmap_allocpte(pmap, ptepindex); 2174 } 2175 } 2176 } else { 2177 mpte = NULL; 2178 } 2179 2180 /* 2181 * This call to vtopte makes the assumption that we are 2182 * entering the page into the current pmap. In order to support 2183 * quick entry into any pmap, one would likely use pmap_pte_quick. 2184 * But that isn't as quick as vtopte. 2185 */ 2186 pte = (unsigned *)vtopte(va); 2187 if (*pte) { 2188 if (mpte) 2189 pmap_unwire_pte_hold(pmap, mpte); 2190 return 0; 2191 } 2192 2193 /* 2194 * Enter on the PV list if part of our managed memory Note that we 2195 * raise IPL while manipulating pv_table since pmap_enter can be 2196 * called at interrupt time. 2197 */ 2198 pmap_insert_entry(pmap, va, mpte, pa); 2199 2200 /* 2201 * Increment counters 2202 */ 2203 pmap->pm_stats.resident_count++; 2204 2205 /* 2206 * Now validate mapping with RO protection 2207 */ 2208 *pte = pa | PG_V | PG_U | PG_MANAGED; 2209 2210 return mpte; 2211} 2212 2213#define MAX_INIT_PT (96) 2214/* 2215 * pmap_object_init_pt preloads the ptes for a given object 2216 * into the specified pmap. This eliminates the blast of soft 2217 * faults on process startup and immediately after an mmap. 2218 */ 2219void 2220pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2221 pmap_t pmap; 2222 vm_offset_t addr; 2223 vm_object_t object; 2224 vm_pindex_t pindex; 2225 vm_size_t size; 2226 int limit; 2227{ 2228 vm_offset_t tmpidx; 2229 int psize; 2230 vm_page_t p, mpte; 2231 int objpgs; 2232 2233 if (!pmap) 2234 return; 2235 2236 /* 2237 * This code maps large physical mmap regions into the 2238 * processor address space. Note that some shortcuts 2239 * are taken, but the code works. 2240 */ 2241 if (pseflag && 2242 (object->type == OBJT_DEVICE) && 2243 ((addr & (NBPDR - 1)) == 0) && 2244 ((size & (NBPDR - 1)) == 0) ) { 2245 int i; 2246 int s; 2247 vm_page_t m[1]; 2248 unsigned int ptepindex; 2249 int npdes; 2250 vm_offset_t ptepa; 2251 2252 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2253 return; 2254 2255 s = splhigh(); 2256retry: 2257 p = vm_page_lookup(object, pindex); 2258 if (p && (p->flags & PG_BUSY)) { 2259 tsleep(p, PVM, "init4p", 0); 2260 goto retry; 2261 } 2262 splx(s); 2263 2264 if (p == NULL) { 2265 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2266 if (p == NULL) 2267 return; 2268 m[0] = p; 2269 2270 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2271 PAGE_WAKEUP(p); 2272 vm_page_free(p); 2273 return; 2274 } 2275 2276 p = vm_page_lookup(object, pindex); 2277 PAGE_WAKEUP(p); 2278 } 2279 2280 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2281 if (ptepa & (NBPDR - 1)) { 2282 return; 2283 } 2284 2285 p->valid = VM_PAGE_BITS_ALL; 2286 2287 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2288 npdes = size >> PDRSHIFT; 2289 for(i=0;i<npdes;i++) { 2290 pmap->pm_pdir[ptepindex] = 2291 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2292 ptepa += NBPDR; 2293 ptepindex += 1; 2294 } 2295 p->flags |= PG_MAPPED; 2296 invltlb(); 2297 return; 2298 } 2299 2300 psize = i386_btop(size); 2301 2302 if ((object->type != OBJT_VNODE) || 2303 (limit && (psize > MAX_INIT_PT) && 2304 (object->resident_page_count > MAX_INIT_PT))) { 2305 return; 2306 } 2307 2308 if (psize + pindex > object->size) 2309 psize = object->size - pindex; 2310 2311 mpte = NULL; 2312 /* 2313 * if we are processing a major portion of the object, then scan the 2314 * entire thing. 2315 */ 2316 if (psize > (object->size >> 2)) { 2317 objpgs = psize; 2318 2319 for (p = TAILQ_FIRST(&object->memq); 2320 ((objpgs > 0) && (p != NULL)); 2321 p = TAILQ_NEXT(p, listq)) { 2322 2323 tmpidx = p->pindex; 2324 if (tmpidx < pindex) { 2325 continue; 2326 } 2327 tmpidx -= pindex; 2328 if (tmpidx >= psize) { 2329 continue; 2330 } 2331 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2332 (p->busy == 0) && 2333 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2334 if ((p->queue - p->pc) == PQ_CACHE) 2335 vm_page_deactivate(p); 2336 p->flags |= PG_BUSY; 2337 mpte = pmap_enter_quick(pmap, 2338 addr + i386_ptob(tmpidx), 2339 VM_PAGE_TO_PHYS(p), mpte); 2340 p->flags |= PG_MAPPED; 2341 PAGE_WAKEUP(p); 2342 } 2343 objpgs -= 1; 2344 } 2345 } else { 2346 /* 2347 * else lookup the pages one-by-one. 2348 */ 2349 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2350 p = vm_page_lookup(object, tmpidx + pindex); 2351 if (p && 2352 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2353 (p->busy == 0) && 2354 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2355 if ((p->queue - p->pc) == PQ_CACHE) 2356 vm_page_deactivate(p); 2357 p->flags |= PG_BUSY; 2358 mpte = pmap_enter_quick(pmap, 2359 addr + i386_ptob(tmpidx), 2360 VM_PAGE_TO_PHYS(p), mpte); 2361 p->flags |= PG_MAPPED; 2362 PAGE_WAKEUP(p); 2363 } 2364 } 2365 } 2366 return; 2367} 2368 2369/* 2370 * pmap_prefault provides a quick way of clustering 2371 * pagefaults into a processes address space. It is a "cousin" 2372 * of pmap_object_init_pt, except it runs at page fault time instead 2373 * of mmap time. 2374 */ 2375#define PFBAK 2 2376#define PFFOR 2 2377#define PAGEORDER_SIZE (PFBAK+PFFOR) 2378 2379static int pmap_prefault_pageorder[] = { 2380 -PAGE_SIZE, PAGE_SIZE, -2 * PAGE_SIZE, 2 * PAGE_SIZE 2381}; 2382 2383void 2384pmap_prefault(pmap, addra, entry, object) 2385 pmap_t pmap; 2386 vm_offset_t addra; 2387 vm_map_entry_t entry; 2388 vm_object_t object; 2389{ 2390 int i; 2391 vm_offset_t starta; 2392 vm_offset_t addr; 2393 vm_pindex_t pindex; 2394 vm_page_t m, mpte; 2395 2396 if (entry->object.vm_object != object) 2397 return; 2398 2399 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) 2400 return; 2401 2402 starta = addra - PFBAK * PAGE_SIZE; 2403 if (starta < entry->start) { 2404 starta = entry->start; 2405 } else if (starta > addra) { 2406 starta = 0; 2407 } 2408 2409 mpte = NULL; 2410 for (i = 0; i < PAGEORDER_SIZE; i++) { 2411 vm_object_t lobject; 2412 unsigned *pte; 2413 2414 addr = addra + pmap_prefault_pageorder[i]; 2415 if (addr < starta || addr >= entry->end) 2416 continue; 2417 2418 if ((*pmap_pde(pmap, addr)) == NULL) 2419 continue; 2420 2421 pte = (unsigned *) vtopte(addr); 2422 if (*pte) 2423 continue; 2424 2425 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2426 lobject = object; 2427 for (m = vm_page_lookup(lobject, pindex); 2428 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2429 lobject = lobject->backing_object) { 2430 if (lobject->backing_object_offset & PAGE_MASK) 2431 break; 2432 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2433 m = vm_page_lookup(lobject->backing_object, pindex); 2434 } 2435 2436 /* 2437 * give-up when a page is not in memory 2438 */ 2439 if (m == NULL) 2440 break; 2441 2442 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2443 (m->busy == 0) && 2444 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2445 2446 if ((m->queue - m->pc) == PQ_CACHE) { 2447 vm_page_deactivate(m); 2448 } 2449 m->flags |= PG_BUSY; 2450 mpte = pmap_enter_quick(pmap, addr, 2451 VM_PAGE_TO_PHYS(m), mpte); 2452 m->flags |= PG_MAPPED; 2453 PAGE_WAKEUP(m); 2454 } 2455 } 2456} 2457 2458/* 2459 * Routine: pmap_change_wiring 2460 * Function: Change the wiring attribute for a map/virtual-address 2461 * pair. 2462 * In/out conditions: 2463 * The mapping must already exist in the pmap. 2464 */ 2465void 2466pmap_change_wiring(pmap, va, wired) 2467 register pmap_t pmap; 2468 vm_offset_t va; 2469 boolean_t wired; 2470{ 2471 register unsigned *pte; 2472 2473 if (pmap == NULL) 2474 return; 2475 2476 pte = pmap_pte(pmap, va); 2477 2478 if (wired && !pmap_pte_w(pte)) 2479 pmap->pm_stats.wired_count++; 2480 else if (!wired && pmap_pte_w(pte)) 2481 pmap->pm_stats.wired_count--; 2482 2483 /* 2484 * Wiring is not a hardware characteristic so there is no need to 2485 * invalidate TLB. 2486 */ 2487 pmap_pte_set_w(pte, wired); 2488} 2489 2490 2491 2492/* 2493 * Copy the range specified by src_addr/len 2494 * from the source map to the range dst_addr/len 2495 * in the destination map. 2496 * 2497 * This routine is only advisory and need not do anything. 2498 */ 2499 2500void 2501pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2502 pmap_t dst_pmap, src_pmap; 2503 vm_offset_t dst_addr; 2504 vm_size_t len; 2505 vm_offset_t src_addr; 2506{ 2507 vm_offset_t addr; 2508 vm_offset_t end_addr = src_addr + len; 2509 vm_offset_t pdnxt; 2510 unsigned src_frame, dst_frame; 2511 2512 if (dst_addr != src_addr) 2513 return; 2514 2515 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2516 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2517 return; 2518 } 2519 2520 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2521 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2522 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2523 invltlb(); 2524 } 2525 2526 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2527 unsigned *src_pte, *dst_pte; 2528 vm_page_t dstmpte, srcmpte; 2529 vm_offset_t srcptepaddr; 2530 unsigned ptepindex; 2531 2532 if (addr >= UPT_MIN_ADDRESS) 2533 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2534 2535 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2536 ptepindex = addr >> PDRSHIFT; 2537 2538 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2539 if (srcptepaddr == 0) 2540 continue; 2541 2542 if (srcptepaddr & PG_PS) { 2543 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2544 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2545 dst_pmap->pm_stats.resident_count += NBPDR; 2546 } 2547 continue; 2548 } 2549 2550 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2551 if ((srcmpte == NULL) || 2552 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2553 continue; 2554 2555 if (pdnxt > end_addr) 2556 pdnxt = end_addr; 2557 2558 src_pte = (unsigned *) vtopte(addr); 2559 dst_pte = (unsigned *) avtopte(addr); 2560 while (addr < pdnxt) { 2561 unsigned ptetemp; 2562 ptetemp = *src_pte; 2563 /* 2564 * we only virtual copy managed pages 2565 */ 2566 if ((ptetemp & PG_MANAGED) != 0) { 2567 /* 2568 * We have to check after allocpte for the 2569 * pte still being around... allocpte can 2570 * block. 2571 */ 2572 dstmpte = pmap_allocpte(dst_pmap, addr); 2573 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2574 /* 2575 * Clear the modified and 2576 * accessed (referenced) bits 2577 * during the copy. 2578 */ 2579 *dst_pte = ptetemp & ~(PG_M|PG_A); 2580 dst_pmap->pm_stats.resident_count++; 2581 pmap_insert_entry(dst_pmap, addr, 2582 dstmpte, 2583 (ptetemp & PG_FRAME)); 2584 } else { 2585 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2586 } 2587 if (dstmpte->hold_count >= srcmpte->hold_count) 2588 break; 2589 } 2590 addr += PAGE_SIZE; 2591 ++src_pte; 2592 ++dst_pte; 2593 } 2594 } 2595} 2596 2597/* 2598 * Routine: pmap_kernel 2599 * Function: 2600 * Returns the physical map handle for the kernel. 2601 */ 2602pmap_t 2603pmap_kernel() 2604{ 2605 return (kernel_pmap); 2606} 2607 2608/* 2609 * pmap_zero_page zeros the specified (machine independent) 2610 * page by mapping the page into virtual memory and using 2611 * bzero to clear its contents, one machine dependent page 2612 * at a time. 2613 */ 2614void 2615pmap_zero_page(phys) 2616 vm_offset_t phys; 2617{ 2618#ifdef SMP 2619 if (*(int *) prv_CMAP3) 2620 panic("pmap_zero_page: prv_CMAP3 busy"); 2621 2622 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME); 2623 invltlb_1pg((vm_offset_t) &prv_CPAGE3); 2624 2625 bzero(&prv_CPAGE3, PAGE_SIZE); 2626 2627 *(int *) prv_CMAP3 = 0; 2628 invltlb_1pg((vm_offset_t) &prv_CPAGE3); 2629#else 2630 if (*(int *) CMAP2) 2631 panic("pmap_zero_page: CMAP busy"); 2632 2633 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME); 2634 bzero(CADDR2, PAGE_SIZE); 2635 *(int *) CMAP2 = 0; 2636 invltlb_1pg((vm_offset_t) CADDR2); 2637#endif 2638} 2639 2640/* 2641 * pmap_copy_page copies the specified (machine independent) 2642 * page by mapping the page into virtual memory and using 2643 * bcopy to copy the page, one machine dependent page at a 2644 * time. 2645 */ 2646void 2647pmap_copy_page(src, dst) 2648 vm_offset_t src; 2649 vm_offset_t dst; 2650{ 2651#ifdef SMP 2652 if (*(int *) prv_CMAP1) 2653 panic("pmap_copy_page: prv_CMAP1 busy"); 2654 if (*(int *) prv_CMAP2) 2655 panic("pmap_copy_page: prv_CMAP2 busy"); 2656 2657 *(int *) prv_CMAP1 = PG_V | PG_RW | (src & PG_FRAME); 2658 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME); 2659 2660 invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2); 2661 2662 bcopy(&prv_CPAGE1, &prv_CPAGE2, PAGE_SIZE); 2663 2664 *(int *) prv_CMAP1 = 0; 2665 *(int *) prv_CMAP2 = 0; 2666 invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2); 2667#else 2668 if (*(int *) CMAP1 || *(int *) CMAP2) 2669 panic("pmap_copy_page: CMAP busy"); 2670 2671 *(int *) CMAP1 = PG_V | PG_RW | (src & PG_FRAME); 2672 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME); 2673 2674 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2675 2676 *(int *) CMAP1 = 0; 2677 *(int *) CMAP2 = 0; 2678 invltlb_2pg( (vm_offset_t) CADDR1, (vm_offset_t) CADDR2); 2679#endif 2680} 2681 2682 2683/* 2684 * Routine: pmap_pageable 2685 * Function: 2686 * Make the specified pages (by pmap, offset) 2687 * pageable (or not) as requested. 2688 * 2689 * A page which is not pageable may not take 2690 * a fault; therefore, its page table entry 2691 * must remain valid for the duration. 2692 * 2693 * This routine is merely advisory; pmap_enter 2694 * will specify that these pages are to be wired 2695 * down (or not) as appropriate. 2696 */ 2697void 2698pmap_pageable(pmap, sva, eva, pageable) 2699 pmap_t pmap; 2700 vm_offset_t sva, eva; 2701 boolean_t pageable; 2702{ 2703} 2704 2705/* 2706 * this routine returns true if a physical page resides 2707 * in the given pmap. 2708 */ 2709boolean_t 2710pmap_page_exists(pmap, pa) 2711 pmap_t pmap; 2712 vm_offset_t pa; 2713{ 2714 register pv_entry_t pv; 2715 pv_table_t *ppv; 2716 int s; 2717 2718 if (!pmap_is_managed(pa)) 2719 return FALSE; 2720 2721 s = splvm(); 2722 2723 ppv = pa_to_pvh(pa); 2724 /* 2725 * Not found, check current mappings returning immediately if found. 2726 */ 2727 for (pv = TAILQ_FIRST(&ppv->pv_list); 2728 pv; 2729 pv = TAILQ_NEXT(pv, pv_list)) { 2730 if (pv->pv_pmap == pmap) { 2731 splx(s); 2732 return TRUE; 2733 } 2734 } 2735 splx(s); 2736 return (FALSE); 2737} 2738 2739#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2740/* 2741 * Remove all pages from specified address space 2742 * this aids process exit speeds. Also, this code 2743 * is special cased for current process only, but 2744 * can have the more generic (and slightly slower) 2745 * mode enabled. This is much faster than pmap_remove 2746 * in the case of running down an entire address space. 2747 */ 2748void 2749pmap_remove_pages(pmap, sva, eva) 2750 pmap_t pmap; 2751 vm_offset_t sva, eva; 2752{ 2753 unsigned *pte, tpte; 2754 pv_table_t *ppv; 2755 pv_entry_t pv, npv; 2756 int s; 2757 2758#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2759 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) { 2760 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2761 return; 2762 } 2763#endif 2764 2765 s = splvm(); 2766 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2767 pv; 2768 pv = npv) { 2769 2770 if (pv->pv_va >= eva || pv->pv_va < sva) { 2771 npv = TAILQ_NEXT(pv, pv_plist); 2772 continue; 2773 } 2774 2775#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2776 pte = (unsigned *)vtopte(pv->pv_va); 2777#else 2778 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2779#endif 2780 tpte = *pte; 2781 2782/* 2783 * We cannot remove wired pages from a process' mapping at this time 2784 */ 2785 if (tpte & PG_W) { 2786 npv = TAILQ_NEXT(pv, pv_plist); 2787 continue; 2788 } 2789 *pte = 0; 2790 2791 ppv = pa_to_pvh(tpte); 2792 2793 pv->pv_pmap->pm_stats.resident_count--; 2794 2795 /* 2796 * Update the vm_page_t clean and reference bits. 2797 */ 2798 if (tpte & PG_M) { 2799 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 2800 } 2801 2802 2803 npv = TAILQ_NEXT(pv, pv_plist); 2804 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2805 2806 --ppv->pv_list_count; 2807 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2808 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2809 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 2810 } 2811 2812 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2813 free_pv_entry(pv); 2814 } 2815 splx(s); 2816 invltlb(); 2817} 2818 2819/* 2820 * pmap_testbit tests bits in pte's 2821 * note that the testbit/changebit routines are inline, 2822 * and a lot of things compile-time evaluate. 2823 */ 2824static boolean_t 2825pmap_testbit(pa, bit) 2826 register vm_offset_t pa; 2827 int bit; 2828{ 2829 register pv_entry_t pv; 2830 pv_table_t *ppv; 2831 unsigned *pte; 2832 int s; 2833 2834 if (!pmap_is_managed(pa)) 2835 return FALSE; 2836 2837 ppv = pa_to_pvh(pa); 2838 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 2839 return FALSE; 2840 2841 s = splvm(); 2842 2843 for (pv = TAILQ_FIRST(&ppv->pv_list); 2844 pv; 2845 pv = TAILQ_NEXT(pv, pv_list)) { 2846 2847 /* 2848 * if the bit being tested is the modified bit, then 2849 * mark clean_map and ptes as never 2850 * modified. 2851 */ 2852 if (bit & (PG_A|PG_M)) { 2853 if (!pmap_track_modified(pv->pv_va)) 2854 continue; 2855 } 2856 2857#if defined(PMAP_DIAGNOSTIC) 2858 if (!pv->pv_pmap) { 2859 printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va); 2860 continue; 2861 } 2862#endif 2863 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2864 if (*pte & bit) { 2865 splx(s); 2866 return TRUE; 2867 } 2868 } 2869 splx(s); 2870 return (FALSE); 2871} 2872 2873/* 2874 * this routine is used to modify bits in ptes 2875 */ 2876static void 2877pmap_changebit(pa, bit, setem) 2878 vm_offset_t pa; 2879 int bit; 2880 boolean_t setem; 2881{ 2882 register pv_entry_t pv; 2883 pv_table_t *ppv; 2884 register unsigned *pte; 2885 int changed; 2886 int s; 2887 2888 if (!pmap_is_managed(pa)) 2889 return; 2890 2891 s = splvm(); 2892 changed = 0; 2893 ppv = pa_to_pvh(pa); 2894 2895 /* 2896 * Loop over all current mappings setting/clearing as appropos If 2897 * setting RO do we need to clear the VAC? 2898 */ 2899 for (pv = TAILQ_FIRST(&ppv->pv_list); 2900 pv; 2901 pv = TAILQ_NEXT(pv, pv_list)) { 2902 2903 /* 2904 * don't write protect pager mappings 2905 */ 2906 if (!setem && (bit == PG_RW)) { 2907 if (!pmap_track_modified(pv->pv_va)) 2908 continue; 2909 } 2910 2911#if defined(PMAP_DIAGNOSTIC) 2912 if (!pv->pv_pmap) { 2913 printf("Null pmap (cb) at va: 0x%lx\n", pv->pv_va); 2914 continue; 2915 } 2916#endif 2917 2918 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2919 2920 if (setem) { 2921 *(int *)pte |= bit; 2922 changed = 1; 2923 } else { 2924 vm_offset_t pbits = *(vm_offset_t *)pte; 2925 if (pbits & bit) { 2926 changed = 1; 2927 if (bit == PG_RW) { 2928 if (pbits & PG_M) { 2929 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 2930 } 2931 *(int *)pte = pbits & ~(PG_M|PG_RW); 2932 } else { 2933 *(int *)pte = pbits & ~bit; 2934 } 2935 } 2936 } 2937 } 2938 splx(s); 2939 if (changed) 2940 invltlb(); 2941} 2942 2943/* 2944 * pmap_page_protect: 2945 * 2946 * Lower the permission for all mappings to a given page. 2947 */ 2948void 2949pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 2950{ 2951 if ((prot & VM_PROT_WRITE) == 0) { 2952 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 2953 pmap_changebit(phys, PG_RW, FALSE); 2954 } else { 2955 pmap_remove_all(phys); 2956 } 2957 } 2958} 2959 2960vm_offset_t 2961pmap_phys_address(ppn) 2962 int ppn; 2963{ 2964 return (i386_ptob(ppn)); 2965} 2966 2967/* 2968 * pmap_ts_referenced: 2969 * 2970 * Return the count of reference bits for a page, clearing all of them. 2971 * 2972 */ 2973int 2974pmap_ts_referenced(vm_offset_t pa) 2975{ 2976 register pv_entry_t pv; 2977 pv_table_t *ppv; 2978 unsigned *pte; 2979 int s; 2980 int rtval = 0; 2981 2982 if (!pmap_is_managed(pa)) 2983 return FALSE; 2984 2985 s = splvm(); 2986 2987 ppv = pa_to_pvh(pa); 2988 2989 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2990 splx(s); 2991 return 0; 2992 } 2993 2994 /* 2995 * Not found, check current mappings returning immediately if found. 2996 */ 2997 for (pv = TAILQ_FIRST(&ppv->pv_list); 2998 pv; 2999 pv = TAILQ_NEXT(pv, pv_list)) { 3000 /* 3001 * if the bit being tested is the modified bit, then 3002 * mark clean_map and ptes as never 3003 * modified. 3004 */ 3005 if (!pmap_track_modified(pv->pv_va)) 3006 continue; 3007 3008 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3009 if (pte == NULL) { 3010 continue; 3011 } 3012 if (*pte & PG_A) { 3013 rtval++; 3014 *pte &= ~PG_A; 3015 } 3016 } 3017 splx(s); 3018 if (rtval) { 3019 invltlb(); 3020 } 3021 return (rtval); 3022} 3023 3024/* 3025 * pmap_is_modified: 3026 * 3027 * Return whether or not the specified physical page was modified 3028 * in any physical maps. 3029 */ 3030boolean_t 3031pmap_is_modified(vm_offset_t pa) 3032{ 3033 return pmap_testbit((pa), PG_M); 3034} 3035 3036/* 3037 * Clear the modify bits on the specified physical page. 3038 */ 3039void 3040pmap_clear_modify(vm_offset_t pa) 3041{ 3042 pmap_changebit((pa), PG_M, FALSE); 3043} 3044 3045/* 3046 * pmap_clear_reference: 3047 * 3048 * Clear the reference bit on the specified physical page. 3049 */ 3050void 3051pmap_clear_reference(vm_offset_t pa) 3052{ 3053 pmap_changebit((pa), PG_A, FALSE); 3054} 3055 3056/* 3057 * Miscellaneous support routines follow 3058 */ 3059 3060static void 3061i386_protection_init() 3062{ 3063 register int *kp, prot; 3064 3065 kp = protection_codes; 3066 for (prot = 0; prot < 8; prot++) { 3067 switch (prot) { 3068 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3069 /* 3070 * Read access is also 0. There isn't any execute bit, 3071 * so just make it readable. 3072 */ 3073 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3074 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3075 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3076 *kp++ = 0; 3077 break; 3078 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3079 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3080 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3081 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3082 *kp++ = PG_RW; 3083 break; 3084 } 3085 } 3086} 3087 3088/* 3089 * Map a set of physical memory pages into the kernel virtual 3090 * address space. Return a pointer to where it is mapped. This 3091 * routine is intended to be used for mapping device memory, 3092 * NOT real memory. 3093 */ 3094void * 3095pmap_mapdev(pa, size) 3096 vm_offset_t pa; 3097 vm_size_t size; 3098{ 3099 vm_offset_t va, tmpva; 3100 unsigned *pte; 3101 3102 size = roundup(size, PAGE_SIZE); 3103 3104 va = kmem_alloc_pageable(kernel_map, size); 3105 if (!va) 3106 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3107 3108 pa = pa & PG_FRAME; 3109 for (tmpva = va; size > 0;) { 3110 pte = (unsigned *)vtopte(tmpva); 3111 *pte = pa | PG_RW | PG_V | pgeflag; 3112 size -= PAGE_SIZE; 3113 tmpva += PAGE_SIZE; 3114 pa += PAGE_SIZE; 3115 } 3116 invltlb(); 3117 3118 return ((void *) va); 3119} 3120 3121/* 3122 * perform the pmap work for mincore 3123 */ 3124int 3125pmap_mincore(pmap, addr) 3126 pmap_t pmap; 3127 vm_offset_t addr; 3128{ 3129 3130 unsigned *ptep, pte; 3131 int val = 0; 3132 3133 ptep = pmap_pte(pmap, addr); 3134 if (ptep == 0) { 3135 return 0; 3136 } 3137 3138 if (pte = *ptep) { 3139 vm_offset_t pa; 3140 val = MINCORE_INCORE; 3141 pa = pte & PG_FRAME; 3142 3143 /* 3144 * Modified by us 3145 */ 3146 if (pte & PG_M) 3147 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3148 /* 3149 * Modified by someone 3150 */ 3151 else if (PHYS_TO_VM_PAGE(pa)->dirty || 3152 pmap_is_modified(pa)) 3153 val |= MINCORE_MODIFIED_OTHER; 3154 /* 3155 * Referenced by us 3156 */ 3157 if (pte & PG_U) 3158 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3159 3160 /* 3161 * Referenced by someone 3162 */ 3163 else if ((PHYS_TO_VM_PAGE(pa)->flags & PG_REFERENCED) || 3164 pmap_ts_referenced(pa)) { 3165 val |= MINCORE_REFERENCED_OTHER; 3166 PHYS_TO_VM_PAGE(pa)->flags |= PG_REFERENCED; 3167 } 3168 } 3169 return val; 3170} 3171 3172void 3173pmap_activate(struct proc *p) 3174{ 3175 load_cr3(p->p_addr->u_pcb.pcb_cr3 = 3176 vtophys(p->p_vmspace->vm_pmap.pm_pdir)); 3177} 3178 3179vm_offset_t 3180pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3181 3182 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3183 return addr; 3184 } 3185 3186 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3187 return addr; 3188} 3189 3190 3191#if defined(PMAP_DEBUG) 3192pmap_pid_dump(int pid) { 3193 pmap_t pmap; 3194 struct proc *p; 3195 int npte = 0; 3196 int index; 3197 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3198 if (p->p_pid != pid) 3199 continue; 3200 3201 if (p->p_vmspace) { 3202 int i,j; 3203 index = 0; 3204 pmap = &p->p_vmspace->vm_pmap; 3205 for(i=0;i<1024;i++) { 3206 pd_entry_t *pde; 3207 unsigned *pte; 3208 unsigned base = i << PDRSHIFT; 3209 3210 pde = &pmap->pm_pdir[i]; 3211 if (pde && pmap_pde_v(pde)) { 3212 for(j=0;j<1024;j++) { 3213 unsigned va = base + (j << PAGE_SHIFT); 3214 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3215 if (index) { 3216 index = 0; 3217 printf("\n"); 3218 } 3219 return npte; 3220 } 3221 pte = pmap_pte_quick( pmap, va); 3222 if (pte && pmap_pte_v(pte)) { 3223 vm_offset_t pa; 3224 vm_page_t m; 3225 pa = *(int *)pte; 3226 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3227 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3228 va, pa, m->hold_count, m->wire_count, m->flags); 3229 npte++; 3230 index++; 3231 if (index >= 2) { 3232 index = 0; 3233 printf("\n"); 3234 } else { 3235 printf(" "); 3236 } 3237 } 3238 } 3239 } 3240 } 3241 } 3242 } 3243 return npte; 3244} 3245#endif 3246 3247#if defined(DEBUG) 3248 3249static void pads __P((pmap_t pm)); 3250static void pmap_pvdump __P((vm_offset_t pa)); 3251 3252/* print address space of pmap*/ 3253static void 3254pads(pm) 3255 pmap_t pm; 3256{ 3257 unsigned va, i, j; 3258 unsigned *ptep; 3259 3260 if (pm == kernel_pmap) 3261 return; 3262 for (i = 0; i < 1024; i++) 3263 if (pm->pm_pdir[i]) 3264 for (j = 0; j < 1024; j++) { 3265 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3266 if (pm == kernel_pmap && va < KERNBASE) 3267 continue; 3268 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3269 continue; 3270 ptep = pmap_pte_quick(pm, va); 3271 if (pmap_pte_v(ptep)) 3272 printf("%x:%x ", va, *(int *) ptep); 3273 }; 3274 3275} 3276 3277static void 3278pmap_pvdump(pa) 3279 vm_offset_t pa; 3280{ 3281 pv_table_t *ppv; 3282 register pv_entry_t pv; 3283 3284 printf("pa %x", pa); 3285 ppv = pa_to_pvh(pa); 3286 for (pv = TAILQ_FIRST(&ppv->pv_list); 3287 pv; 3288 pv = TAILQ_NEXT(pv, pv_list)) { 3289#ifdef used_to_be 3290 printf(" -> pmap %x, va %x, flags %x", 3291 pv->pv_pmap, pv->pv_va, pv->pv_flags); 3292#endif 3293 printf(" -> pmap %x, va %x", 3294 pv->pv_pmap, pv->pv_va); 3295 pads(pv->pv_pmap); 3296 } 3297 printf(" "); 3298} 3299#endif 3300