pmap.c revision 31709
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 1.173 1997/11/20 19:30:31 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 <sys/param.h> 72#include <sys/systm.h> 73#include <sys/proc.h> 74#include <sys/msgbuf.h> 75#include <sys/vmmeter.h> 76#include <sys/mman.h> 77 78#include <vm/vm.h> 79#include <vm/vm_param.h> 80#include <vm/vm_prot.h> 81#include <sys/lock.h> 82#include <vm/vm_kern.h> 83#include <vm/vm_page.h> 84#include <vm/vm_map.h> 85#include <vm/vm_object.h> 86#include <vm/vm_extern.h> 87#include <vm/vm_pageout.h> 88#include <vm/vm_pager.h> 89#include <vm/vm_zone.h> 90 91#include <sys/user.h> 92 93#include <machine/cputypes.h> 94#include <machine/md_var.h> 95#include <machine/specialreg.h> 96#if defined(SMP) || defined(APIC_IO) 97#include <machine/smp.h> 98#include <machine/apic.h> 99#endif /* SMP || APIC_IO */ 100 101#define PMAP_KEEP_PDIRS 102#ifndef PMAP_SHPGPERPROC 103#define PMAP_SHPGPERPROC 200 104#endif 105 106#if defined(DIAGNOSTIC) 107#define PMAP_DIAGNOSTIC 108#endif 109 110#define MINPV 2048 111 112#if !defined(PMAP_DIAGNOSTIC) 113#define PMAP_INLINE __inline 114#else 115#define PMAP_INLINE 116#endif 117 118#define PTPHINT 119 120/* 121 * Get PDEs and PTEs for user/kernel address space 122 */ 123#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 124#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 125 126#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 127#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 128#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 129#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 130#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 131 132#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 133#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 134 135/* 136 * Given a map and a machine independent protection code, 137 * convert to a vax protection code. 138 */ 139#define pte_prot(m, p) (protection_codes[p]) 140static int protection_codes[8]; 141 142#define pa_index(pa) atop((pa) - vm_first_phys) 143#define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) 144 145static struct pmap kernel_pmap_store; 146pmap_t kernel_pmap; 147extern pd_entry_t my_idlePTD; 148 149vm_offset_t avail_start; /* PA of first available physical page */ 150vm_offset_t avail_end; /* PA of last available physical page */ 151vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 152vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 153static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 154static vm_offset_t vm_first_phys; 155int pgeflag; /* PG_G or-in */ 156int pseflag; /* PG_PS or-in */ 157int pv_npg; 158 159int nkpt; 160vm_offset_t kernel_vm_end; 161 162/* 163 * Data for the pv entry allocation mechanism 164 */ 165vm_zone_t pvzone; 166struct vm_zone pvzone_store; 167struct vm_object pvzone_obj; 168int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0; 169int pmap_pagedaemon_waken = 0; 170struct pv_entry *pvinit; 171 172/* 173 * All those kernel PT submaps that BSD is so fond of 174 */ 175pt_entry_t *CMAP1 = 0; 176static pt_entry_t *CMAP2, *ptmmap; 177static pv_table_t *pv_table; 178caddr_t CADDR1 = 0, ptvmmap = 0; 179static caddr_t CADDR2; 180static pt_entry_t *msgbufmap; 181struct msgbuf *msgbufp=0; 182 183#ifdef SMP 184extern char prv_CPAGE1[], prv_CPAGE2[], prv_CPAGE3[]; 185extern pt_entry_t *prv_CMAP1, *prv_CMAP2, *prv_CMAP3; 186extern pd_entry_t *IdlePTDS[]; 187extern pt_entry_t SMP_prvpt[]; 188#endif 189 190pt_entry_t *PMAP1 = 0; 191unsigned *PADDR1 = 0; 192 193static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 194static unsigned * get_ptbase __P((pmap_t pmap)); 195static pv_entry_t get_pv_entry __P((void)); 196static void i386_protection_init __P((void)); 197static void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 198 199static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 200static void pmap_remove_all __P((vm_offset_t pa)); 201static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 202 vm_offset_t pa, vm_page_t mpte)); 203static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 204 vm_offset_t sva)); 205static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 206static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 207 vm_offset_t va)); 208static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 209static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 210 vm_page_t mpte, vm_offset_t pa)); 211 212static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 213 214static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 215static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 216static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 217static vm_page_t pmap_page_alloc __P((vm_object_t object, vm_pindex_t pindex)); 218static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 219static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 220vm_offset_t pmap_kmem_choose(vm_offset_t addr) ; 221void pmap_collect(void); 222 223#define PDSTACKMAX 6 224static vm_offset_t pdstack[PDSTACKMAX]; 225static int pdstackptr; 226unsigned pdir4mb; 227 228/* 229 * Routine: pmap_pte 230 * Function: 231 * Extract the page table entry associated 232 * with the given map/virtual_address pair. 233 */ 234 235PMAP_INLINE unsigned * 236pmap_pte(pmap, va) 237 register pmap_t pmap; 238 vm_offset_t va; 239{ 240 unsigned *pdeaddr; 241 242 if (pmap) { 243 pdeaddr = (unsigned *) pmap_pde(pmap, va); 244 if (*pdeaddr & PG_PS) 245 return pdeaddr; 246 if (*pdeaddr) { 247 return get_ptbase(pmap) + i386_btop(va); 248 } 249 } 250 return (0); 251} 252 253/* 254 * Move the kernel virtual free pointer to the next 255 * 4MB. This is used to help improve performance 256 * by using a large (4MB) page for much of the kernel 257 * (.text, .data, .bss) 258 */ 259vm_offset_t 260pmap_kmem_choose(vm_offset_t addr) { 261 vm_offset_t newaddr = addr; 262#ifndef DISABLE_PSE 263 if (cpu_feature & CPUID_PSE) { 264 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 265 } 266#endif 267 return newaddr; 268} 269 270/* 271 * Bootstrap the system enough to run with virtual memory. 272 * 273 * On the i386 this is called after mapping has already been enabled 274 * and just syncs the pmap module with what has already been done. 275 * [We can't call it easily with mapping off since the kernel is not 276 * mapped with PA == VA, hence we would have to relocate every address 277 * from the linked base (virtual) address "KERNBASE" to the actual 278 * (physical) address starting relative to 0] 279 */ 280void 281pmap_bootstrap(firstaddr, loadaddr) 282 vm_offset_t firstaddr; 283 vm_offset_t loadaddr; 284{ 285 vm_offset_t va; 286 pt_entry_t *pte; 287 int i, j; 288 289 avail_start = firstaddr; 290 291 /* 292 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 293 * large. It should instead be correctly calculated in locore.s and 294 * not based on 'first' (which is a physical address, not a virtual 295 * address, for the start of unused physical memory). The kernel 296 * page tables are NOT double mapped and thus should not be included 297 * in this calculation. 298 */ 299 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 300 virtual_avail = pmap_kmem_choose(virtual_avail); 301 302 virtual_end = VM_MAX_KERNEL_ADDRESS; 303 304 /* 305 * Initialize protection array. 306 */ 307 i386_protection_init(); 308 309 /* 310 * The kernel's pmap is statically allocated so we don't have to use 311 * pmap_create, which is unlikely to work correctly at this part of 312 * the boot sequence (XXX and which no longer exists). 313 */ 314 kernel_pmap = &kernel_pmap_store; 315 316 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 317 318 kernel_pmap->pm_count = 1; 319 TAILQ_INIT(&kernel_pmap->pm_pvlist); 320 nkpt = NKPT; 321 322 /* 323 * Reserve some special page table entries/VA space for temporary 324 * mapping of pages. 325 */ 326#define SYSMAP(c, p, v, n) \ 327 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 328 329 va = virtual_avail; 330 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 331 332 /* 333 * CMAP1/CMAP2 are used for zeroing and copying pages. 334 */ 335 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 336 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 337 338 /* 339 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 340 * XXX ptmmap is not used. 341 */ 342 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 343 344 /* 345 * msgbufp is used to map the system message buffer. 346 * XXX msgbufmap is not used. 347 */ 348 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 349 atop(round_page(sizeof(struct msgbuf)))) 350 351 /* 352 * ptemap is used for pmap_pte_quick 353 */ 354 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 355 356 virtual_avail = va; 357 358 *(int *) CMAP1 = *(int *) CMAP2 = 0; 359 *(int *) PTD = 0; 360 361 362 pgeflag = 0; 363#if !defined(SMP) 364 if (cpu_feature & CPUID_PGE) { 365 pgeflag = PG_G; 366 } 367#endif 368 369/* 370 * Initialize the 4MB page size flag 371 */ 372 pseflag = 0; 373/* 374 * The 4MB page version of the initial 375 * kernel page mapping. 376 */ 377 pdir4mb = 0; 378 379#if !defined(DISABLE_PSE) 380 if (cpu_feature & CPUID_PSE) { 381 unsigned ptditmp; 382 /* 383 * Enable the PSE mode 384 */ 385 load_cr4(rcr4() | CR4_PSE); 386 387 /* 388 * Note that we have enabled PSE mode 389 */ 390 pseflag = PG_PS; 391 ptditmp = (unsigned) kernel_pmap->pm_pdir[KPTDI]; 392 ptditmp &= ~(NBPDR - 1); 393 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 394 pdir4mb = ptditmp; 395 /* 396 * We can do the mapping here for the single processor 397 * case. We simply ignore the old page table page from 398 * now on. 399 */ 400#if !defined(SMP) 401 PTD[KPTDI] = (pd_entry_t) ptditmp; 402 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 403 invltlb(); 404#endif 405 } 406#endif 407 408#ifdef SMP 409 if (cpu_apic_address == 0) 410 panic("pmap_bootstrap: no local apic!"); 411 412 /* 0 = private page */ 413 /* 1 = page table page */ 414 /* 2 = local apic */ 415 /* 16-31 = io apics */ 416 SMP_prvpt[2] = (pt_entry_t)(PG_V | PG_RW | pgeflag | ((u_long)cpu_apic_address & PG_FRAME)); 417 418 for (i = 0; i < mp_napics; i++) { 419 for (j = 0; j < 16; j++) { 420 /* same page frame as a previous IO apic? */ 421 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 422 ((u_long)io_apic_address[0] & PG_FRAME)) { 423 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 424 break; 425 } 426 /* use this slot if available */ 427 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 0) { 428 SMP_prvpt[j + 16] = (pt_entry_t)(PG_V | PG_RW | pgeflag | 429 ((u_long)io_apic_address[i] & PG_FRAME)); 430 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 431 break; 432 } 433 } 434 if (j == 16) 435 panic("no space to map IO apic %d!", i); 436 } 437 438 /* BSP does this itself, AP's get it pre-set */ 439 prv_CMAP1 = (pt_entry_t *)&SMP_prvpt[3 + UPAGES]; 440 prv_CMAP2 = (pt_entry_t *)&SMP_prvpt[4 + UPAGES]; 441 prv_CMAP3 = (pt_entry_t *)&SMP_prvpt[5 + UPAGES]; 442#endif 443 444 invltlb(); 445 446} 447 448/* 449 * Set 4mb pdir for mp startup, and global flags 450 */ 451void 452pmap_set_opt(unsigned *pdir) { 453 int i; 454 455 if (pseflag && (cpu_feature & CPUID_PSE)) { 456 load_cr4(rcr4() | CR4_PSE); 457 if (pdir4mb) { 458 (unsigned) pdir[KPTDI] = pdir4mb; 459 } 460 } 461 462 if (pgeflag && (cpu_feature & CPUID_PGE)) { 463 load_cr4(rcr4() | CR4_PGE); 464 for(i = KPTDI; i < KPTDI + nkpt; i++) { 465 if (pdir[i]) { 466 pdir[i] |= PG_G; 467 } 468 } 469 } 470} 471 472/* 473 * Setup the PTD for the boot processor 474 */ 475void 476pmap_set_opt_bsp(void) 477{ 478 pmap_set_opt((unsigned *)kernel_pmap->pm_pdir); 479 pmap_set_opt((unsigned *)PTD); 480 invltlb(); 481} 482 483/* 484 * Initialize the pmap module. 485 * Called by vm_init, to initialize any structures that the pmap 486 * system needs to map virtual memory. 487 * pmap_init has been enhanced to support in a fairly consistant 488 * way, discontiguous physical memory. 489 */ 490void 491pmap_init(phys_start, phys_end) 492 vm_offset_t phys_start, phys_end; 493{ 494 vm_offset_t addr; 495 vm_size_t s; 496 int i; 497 int initial_pvs; 498 499 /* 500 * calculate the number of pv_entries needed 501 */ 502 vm_first_phys = phys_avail[0]; 503 for (i = 0; phys_avail[i + 1]; i += 2); 504 pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 505 506 /* 507 * Allocate memory for random pmap data structures. Includes the 508 * pv_head_table. 509 */ 510 s = (vm_size_t) (sizeof(pv_table_t) * pv_npg); 511 s = round_page(s); 512 513 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 514 pv_table = (pv_table_t *) addr; 515 for(i = 0; i < pv_npg; i++) { 516 vm_offset_t pa; 517 TAILQ_INIT(&pv_table[i].pv_list); 518 pv_table[i].pv_list_count = 0; 519 pa = vm_first_phys + i * PAGE_SIZE; 520 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 521 } 522 523 /* 524 * init the pv free list 525 */ 526 initial_pvs = pv_npg; 527 if (initial_pvs < MINPV) 528 initial_pvs = MINPV; 529 pvzone = &pvzone_store; 530 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 531 initial_pvs * sizeof (struct pv_entry)); 532 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, pv_npg); 533 534 /* 535 * Now it is safe to enable pv_table recording. 536 */ 537 pmap_initialized = TRUE; 538} 539 540/* 541 * Initialize the address space (zone) for the pv_entries. Set a 542 * high water mark so that the system can recover from excessive 543 * numbers of pv entries. 544 */ 545void 546pmap_init2() { 547 pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg; 548 pv_entry_high_water = 9 * (pv_entry_max / 10); 549 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 550} 551 552/* 553 * Used to map a range of physical addresses into kernel 554 * virtual address space. 555 * 556 * For now, VM is already on, we only need to map the 557 * specified memory. 558 */ 559vm_offset_t 560pmap_map(virt, start, end, prot) 561 vm_offset_t virt; 562 vm_offset_t start; 563 vm_offset_t end; 564 int prot; 565{ 566 while (start < end) { 567 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 568 virt += PAGE_SIZE; 569 start += PAGE_SIZE; 570 } 571 return (virt); 572} 573 574 575/*************************************************** 576 * Low level helper routines..... 577 ***************************************************/ 578 579#if defined(PMAP_DIAGNOSTIC) 580 581/* 582 * This code checks for non-writeable/modified pages. 583 * This should be an invalid condition. 584 */ 585static int 586pmap_nw_modified(pt_entry_t ptea) { 587 int pte; 588 589 pte = (int) ptea; 590 591 if ((pte & (PG_M|PG_RW)) == PG_M) 592 return 1; 593 else 594 return 0; 595} 596#endif 597 598 599/* 600 * this routine defines the region(s) of memory that should 601 * not be tested for the modified bit. 602 */ 603static PMAP_INLINE int 604pmap_track_modified( vm_offset_t va) { 605 if ((va < clean_sva) || (va >= clean_eva)) 606 return 1; 607 else 608 return 0; 609} 610 611static PMAP_INLINE void 612invltlb_1pg( vm_offset_t va) { 613#if defined(I386_CPU) 614 if (cpu_class == CPUCLASS_386) { 615 invltlb(); 616 } else 617#endif 618 { 619 invlpg(va); 620 } 621} 622 623static PMAP_INLINE void 624invltlb_2pg( vm_offset_t va1, vm_offset_t va2) { 625#if defined(I386_CPU) 626 if (cpu_class == CPUCLASS_386) { 627 invltlb(); 628 } else 629#endif 630 { 631 invlpg(va1); 632 invlpg(va2); 633 } 634} 635 636static unsigned * 637get_ptbase(pmap) 638 pmap_t pmap; 639{ 640 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 641 642 /* are we current address space or kernel? */ 643 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 644 return (unsigned *) PTmap; 645 } 646 /* otherwise, we are alternate address space */ 647 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 648 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 649 invltlb(); 650 } 651 return (unsigned *) APTmap; 652} 653 654/* 655 * Super fast pmap_pte routine best used when scanning 656 * the pv lists. This eliminates many coarse-grained 657 * invltlb calls. Note that many of the pv list 658 * scans are across different pmaps. It is very wasteful 659 * to do an entire invltlb for checking a single mapping. 660 */ 661 662static unsigned * 663pmap_pte_quick(pmap, va) 664 register pmap_t pmap; 665 vm_offset_t va; 666{ 667 unsigned pde, newpf; 668 if (pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) { 669 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 670 unsigned index = i386_btop(va); 671 /* are we current address space or kernel? */ 672 if ((pmap == kernel_pmap) || 673 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 674 return (unsigned *) PTmap + index; 675 } 676 newpf = pde & PG_FRAME; 677 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 678 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 679 invltlb_1pg((vm_offset_t) PADDR1); 680 } 681 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 682 } 683 return (0); 684} 685 686/* 687 * Routine: pmap_extract 688 * Function: 689 * Extract the physical page address associated 690 * with the given map/virtual_address pair. 691 */ 692vm_offset_t 693pmap_extract(pmap, va) 694 register pmap_t pmap; 695 vm_offset_t va; 696{ 697 vm_offset_t rtval; 698 vm_offset_t pdirindex; 699 pdirindex = va >> PDRSHIFT; 700 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 701 unsigned *pte; 702 if ((rtval & PG_PS) != 0) { 703 rtval &= ~(NBPDR - 1); 704 rtval |= va & (NBPDR - 1); 705 return rtval; 706 } 707 pte = get_ptbase(pmap) + i386_btop(va); 708 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 709 return rtval; 710 } 711 return 0; 712 713} 714 715/* 716 * determine if a page is managed (memory vs. device) 717 */ 718static PMAP_INLINE int 719pmap_is_managed(pa) 720 vm_offset_t pa; 721{ 722 int i; 723 724 if (!pmap_initialized) 725 return 0; 726 727 for (i = 0; phys_avail[i + 1]; i += 2) { 728 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 729 return 1; 730 } 731 return 0; 732} 733 734 735/*************************************************** 736 * Low level mapping routines..... 737 ***************************************************/ 738 739/* 740 * Add a list of wired pages to the kva 741 * this routine is only used for temporary 742 * kernel mappings that do not need to have 743 * page modification or references recorded. 744 * Note that old mappings are simply written 745 * over. The page *must* be wired. 746 */ 747void 748pmap_qenter(va, m, count) 749 vm_offset_t va; 750 vm_page_t *m; 751 int count; 752{ 753 int i; 754 register unsigned *pte; 755 756 for (i = 0; i < count; i++) { 757 vm_offset_t tva = va + i * PAGE_SIZE; 758 unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | pgeflag; 759 unsigned opte; 760 pte = (unsigned *)vtopte(tva); 761 opte = *pte; 762 *pte = npte; 763 if (opte) 764 invltlb_1pg(tva); 765 } 766} 767 768/* 769 * this routine jerks page mappings from the 770 * kernel -- it is meant only for temporary mappings. 771 */ 772void 773pmap_qremove(va, count) 774 vm_offset_t va; 775 int count; 776{ 777 int i; 778 register unsigned *pte; 779 780 for (i = 0; i < count; i++) { 781 pte = (unsigned *)vtopte(va); 782 *pte = 0; 783 invltlb_1pg(va); 784 va += PAGE_SIZE; 785 } 786} 787 788/* 789 * add a wired page to the kva 790 * note that in order for the mapping to take effect -- you 791 * should do a invltlb after doing the pmap_kenter... 792 */ 793PMAP_INLINE void 794pmap_kenter(va, pa) 795 vm_offset_t va; 796 register vm_offset_t pa; 797{ 798 register unsigned *pte; 799 unsigned npte, opte; 800 801 npte = pa | PG_RW | PG_V | pgeflag; 802 pte = (unsigned *)vtopte(va); 803 opte = *pte; 804 *pte = npte; 805 if (opte) 806 invltlb_1pg(va); 807} 808 809/* 810 * remove a page from the kernel pagetables 811 */ 812PMAP_INLINE void 813pmap_kremove(va) 814 vm_offset_t va; 815{ 816 register unsigned *pte; 817 818 pte = (unsigned *)vtopte(va); 819 *pte = 0; 820 invltlb_1pg(va); 821} 822 823static vm_page_t 824pmap_page_alloc(object, pindex) 825 vm_object_t object; 826 vm_pindex_t pindex; 827{ 828 vm_page_t m; 829 m = vm_page_alloc(object, pindex, VM_ALLOC_ZERO); 830 if (m == NULL) { 831 VM_WAIT; 832 } 833 return m; 834} 835 836static vm_page_t 837pmap_page_lookup(object, pindex) 838 vm_object_t object; 839 vm_pindex_t pindex; 840{ 841 vm_page_t m; 842retry: 843 m = vm_page_lookup(object, pindex); 844 if (m) { 845 if (m->flags & PG_BUSY) { 846 m->flags |= PG_WANTED; 847 tsleep(m, PVM, "pplookp", 0); 848 goto retry; 849 } 850 } 851 852 return m; 853} 854 855/* 856 * Create the UPAGES for a new process. 857 * This routine directly affects the fork perf for a process. 858 */ 859void 860pmap_new_proc(p) 861 struct proc *p; 862{ 863 int i; 864 vm_object_t upobj; 865 vm_page_t m; 866 struct user *up; 867 unsigned *ptek; 868 869 /* 870 * allocate object for the upages 871 */ 872 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 873 p->p_upages_obj = upobj; 874 875 /* get a kernel virtual address for the UPAGES for this proc */ 876 up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * PAGE_SIZE); 877 if (up == NULL) 878 panic("pmap_new_proc: u_map allocation failed"); 879 880 ptek = (unsigned *) vtopte((vm_offset_t) up); 881 882 for(i=0;i<UPAGES;i++) { 883 /* 884 * Get a kernel stack page 885 */ 886 while ((m = vm_page_alloc(upobj, 887 i, VM_ALLOC_NORMAL)) == NULL) { 888 VM_WAIT; 889 } 890 891 /* 892 * Wire the page 893 */ 894 m->wire_count++; 895 ++cnt.v_wire_count; 896 897 /* 898 * Enter the page into the kernel address space. 899 */ 900 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 901 902 m->flags &= ~(PG_ZERO|PG_BUSY); 903 m->flags |= PG_MAPPED|PG_WRITEABLE; 904 m->valid = VM_PAGE_BITS_ALL; 905 } 906 907 p->p_addr = up; 908} 909 910/* 911 * Dispose the UPAGES for a process that has exited. 912 * This routine directly impacts the exit perf of a process. 913 */ 914void 915pmap_dispose_proc(p) 916 struct proc *p; 917{ 918 int i; 919 vm_object_t upobj; 920 vm_page_t m; 921 unsigned *ptek; 922 923 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 924 925 upobj = p->p_upages_obj; 926 927 for(i=0;i<UPAGES;i++) { 928 unsigned oldpte; 929 if ((m = vm_page_lookup(upobj, i)) == NULL) 930 panic("pmap_dispose_proc: upage already missing???"); 931 oldpte = *(ptek + i); 932 *(ptek + i) = 0; 933 if (oldpte & PG_G) 934 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 935 vm_page_unwire(m); 936 vm_page_free(m); 937 } 938 939 vm_object_deallocate(upobj); 940 941 kmem_free(u_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 942} 943 944/* 945 * Allow the UPAGES for a process to be prejudicially paged out. 946 */ 947void 948pmap_swapout_proc(p) 949 struct proc *p; 950{ 951 int i; 952 vm_object_t upobj; 953 vm_page_t m; 954 955 upobj = p->p_upages_obj; 956 /* 957 * let the upages be paged 958 */ 959 for(i=0;i<UPAGES;i++) { 960 if ((m = vm_page_lookup(upobj, i)) == NULL) 961 panic("pmap_swapout_proc: upage already missing???"); 962 m->dirty = VM_PAGE_BITS_ALL; 963 vm_page_unwire(m); 964 vm_page_deactivate(m); 965 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 966 } 967} 968 969/* 970 * Bring the UPAGES for a specified process back in. 971 */ 972void 973pmap_swapin_proc(p) 974 struct proc *p; 975{ 976 int i; 977 vm_object_t upobj; 978 vm_page_t m; 979 980 upobj = p->p_upages_obj; 981 for(i=0;i<UPAGES;i++) { 982 int s; 983 s = splvm(); 984retry: 985 if ((m = vm_page_lookup(upobj, i)) == NULL) { 986 if ((m = vm_page_alloc(upobj, i, VM_ALLOC_NORMAL)) == NULL) { 987 VM_WAIT; 988 goto retry; 989 } 990 } else { 991 if ((m->flags & PG_BUSY) || m->busy) { 992 m->flags |= PG_WANTED; 993 tsleep(m, PVM, "swinuw",0); 994 goto retry; 995 } 996 m->flags |= PG_BUSY; 997 } 998 vm_page_wire(m); 999 splx(s); 1000 1001 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 1002 VM_PAGE_TO_PHYS(m)); 1003 1004 if (m->valid != VM_PAGE_BITS_ALL) { 1005 int rv; 1006 rv = vm_pager_get_pages(upobj, &m, 1, 0); 1007 if (rv != VM_PAGER_OK) 1008 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 1009 m->valid = VM_PAGE_BITS_ALL; 1010 } 1011 PAGE_WAKEUP(m); 1012 m->flags |= PG_MAPPED|PG_WRITEABLE; 1013 } 1014} 1015 1016/*************************************************** 1017 * Page table page management routines..... 1018 ***************************************************/ 1019 1020/* 1021 * This routine unholds page table pages, and if the hold count 1022 * drops to zero, then it decrements the wire count. 1023 */ 1024static int 1025_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1026 int s; 1027 1028 if (m->flags & PG_BUSY) { 1029 s = splvm(); 1030 while (m->flags & PG_BUSY) { 1031 m->flags |= PG_WANTED; 1032 tsleep(m, PVM, "pmuwpt", 0); 1033 } 1034 splx(s); 1035 } 1036 1037 if (m->hold_count == 0) { 1038 vm_offset_t pteva; 1039 /* 1040 * unmap the page table page 1041 */ 1042 pmap->pm_pdir[m->pindex] = 0; 1043 --pmap->pm_stats.resident_count; 1044 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1045 (((unsigned) PTDpde) & PG_FRAME)) { 1046 /* 1047 * Do a invltlb to make the invalidated mapping 1048 * take effect immediately. 1049 */ 1050 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1051 invltlb_1pg(pteva); 1052 } 1053 1054#if defined(PTPHINT) 1055 if (pmap->pm_ptphint == m) 1056 pmap->pm_ptphint = NULL; 1057#endif 1058 1059 /* 1060 * If the page is finally unwired, simply free it. 1061 */ 1062 --m->wire_count; 1063 if (m->wire_count == 0) { 1064 1065 if (m->flags & PG_WANTED) { 1066 m->flags &= ~PG_WANTED; 1067 wakeup(m); 1068 } 1069 1070 vm_page_free_zero(m); 1071 --cnt.v_wire_count; 1072 } 1073 return 1; 1074 } 1075 return 0; 1076} 1077 1078__inline static int 1079pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1080 vm_page_unhold(m); 1081 if (m->hold_count == 0) 1082 return _pmap_unwire_pte_hold(pmap, m); 1083 else 1084 return 0; 1085} 1086 1087/* 1088 * After removing a page table entry, this routine is used to 1089 * conditionally free the page, and manage the hold/wire counts. 1090 */ 1091static int 1092pmap_unuse_pt(pmap, va, mpte) 1093 pmap_t pmap; 1094 vm_offset_t va; 1095 vm_page_t mpte; 1096{ 1097 unsigned ptepindex; 1098 if (va >= UPT_MIN_ADDRESS) 1099 return 0; 1100 1101 if (mpte == NULL) { 1102 ptepindex = (va >> PDRSHIFT); 1103#if defined(PTPHINT) 1104 if (pmap->pm_ptphint && 1105 (pmap->pm_ptphint->pindex == ptepindex)) { 1106 mpte = pmap->pm_ptphint; 1107 } else { 1108 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1109 pmap->pm_ptphint = mpte; 1110 } 1111#else 1112 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1113#endif 1114 } 1115 1116 return pmap_unwire_pte_hold(pmap, mpte); 1117} 1118 1119#if !defined(SMP) 1120void 1121pmap_pinit0(pmap) 1122 struct pmap *pmap; 1123{ 1124 pmap->pm_pdir = 1125 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1126 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1127 pmap->pm_flags = 0; 1128 pmap->pm_count = 1; 1129 pmap->pm_ptphint = NULL; 1130 TAILQ_INIT(&pmap->pm_pvlist); 1131} 1132#else 1133void 1134pmap_pinit0(pmap) 1135 struct pmap *pmap; 1136{ 1137 pmap_pinit(pmap); 1138} 1139#endif 1140 1141/* 1142 * Initialize a preallocated and zeroed pmap structure, 1143 * such as one in a vmspace structure. 1144 */ 1145void 1146pmap_pinit(pmap) 1147 register struct pmap *pmap; 1148{ 1149 vm_page_t ptdpg; 1150 /* 1151 * No need to allocate page table space yet but we do need a valid 1152 * page directory table. 1153 */ 1154 1155 if (pdstackptr > 0) { 1156 --pdstackptr; 1157 pmap->pm_pdir = (pd_entry_t *)pdstack[pdstackptr]; 1158 } else { 1159 pmap->pm_pdir = 1160 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1161 } 1162 1163 /* 1164 * allocate object for the ptes 1165 */ 1166 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1167 1168 /* 1169 * allocate the page directory page 1170 */ 1171retry: 1172 ptdpg = pmap_page_alloc( pmap->pm_pteobj, PTDPTDI); 1173 if (ptdpg == NULL) 1174 goto retry; 1175 1176 ptdpg->wire_count = 1; 1177 ++cnt.v_wire_count; 1178 1179 ptdpg->flags &= ~(PG_MAPPED|PG_BUSY); /* not mapped normally */ 1180 ptdpg->valid = VM_PAGE_BITS_ALL; 1181 1182 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1183 if ((ptdpg->flags & PG_ZERO) == 0) 1184 bzero(pmap->pm_pdir, PAGE_SIZE); 1185 1186 /* wire in kernel global address entries */ 1187 /* XXX copies current process, does not fill in MPPTDI */ 1188 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1189 1190 /* install self-referential address mapping entry */ 1191 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1192 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW; 1193 1194 pmap->pm_flags = 0; 1195 pmap->pm_count = 1; 1196 pmap->pm_ptphint = NULL; 1197 TAILQ_INIT(&pmap->pm_pvlist); 1198} 1199 1200static int 1201pmap_release_free_page(pmap, p) 1202 struct pmap *pmap; 1203 vm_page_t p; 1204{ 1205 int s; 1206 unsigned *pde = (unsigned *) pmap->pm_pdir; 1207 /* 1208 * This code optimizes the case of freeing non-busy 1209 * page-table pages. Those pages are zero now, and 1210 * might as well be placed directly into the zero queue. 1211 */ 1212 s = splvm(); 1213 if (p->flags & PG_BUSY) { 1214 p->flags |= PG_WANTED; 1215 tsleep(p, PVM, "pmaprl", 0); 1216 splx(s); 1217 return 0; 1218 } 1219 1220 if (p->flags & PG_WANTED) { 1221 p->flags &= ~PG_WANTED; 1222 wakeup(p); 1223 } 1224 1225 /* 1226 * Remove the page table page from the processes address space. 1227 */ 1228 pde[p->pindex] = 0; 1229 --pmap->pm_stats.resident_count; 1230 1231 if (p->hold_count) { 1232 panic("pmap_release: freeing held page table page"); 1233 } 1234 /* 1235 * Page directory pages need to have the kernel 1236 * stuff cleared, so they can go into the zero queue also. 1237 */ 1238 if (p->pindex == PTDPTDI) { 1239 bzero(pde + KPTDI, nkpt * PTESIZE); 1240#ifdef SMP 1241 pde[MPPTDI] = 0; 1242#endif 1243 pde[APTDPTDI] = 0; 1244 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1245 } 1246 1247#if defined(PTPHINT) 1248 if (pmap->pm_ptphint && 1249 (pmap->pm_ptphint->pindex == p->pindex)) 1250 pmap->pm_ptphint = NULL; 1251#endif 1252 1253 vm_page_free_zero(p); 1254 splx(s); 1255 return 1; 1256} 1257 1258/* 1259 * this routine is called if the page table page is not 1260 * mapped correctly. 1261 */ 1262static vm_page_t 1263_pmap_allocpte(pmap, ptepindex) 1264 pmap_t pmap; 1265 unsigned ptepindex; 1266{ 1267 vm_offset_t pteva, ptepa; 1268 vm_page_t m; 1269 int needszero = 0; 1270 1271 /* 1272 * Find or fabricate a new pagetable page 1273 */ 1274retry: 1275 m = vm_page_lookup(pmap->pm_pteobj, ptepindex); 1276 if (m == NULL) { 1277 m = pmap_page_alloc(pmap->pm_pteobj, ptepindex); 1278 if (m == NULL) 1279 goto retry; 1280 if ((m->flags & PG_ZERO) == 0) 1281 needszero = 1; 1282 m->flags &= ~(PG_ZERO|PG_BUSY); 1283 m->valid = VM_PAGE_BITS_ALL; 1284 } else { 1285 if ((m->flags & PG_BUSY) || m->busy) { 1286 m->flags |= PG_WANTED; 1287 tsleep(m, PVM, "ptewai", 0); 1288 goto retry; 1289 } 1290 } 1291 1292 if (m->queue != PQ_NONE) { 1293 int s = splvm(); 1294 vm_page_unqueue(m); 1295 splx(s); 1296 } 1297 1298 if (m->wire_count == 0) 1299 ++cnt.v_wire_count; 1300 ++m->wire_count; 1301 1302 /* 1303 * Increment the hold count for the page table page 1304 * (denoting a new mapping.) 1305 */ 1306 ++m->hold_count; 1307 1308 /* 1309 * Map the pagetable page into the process address space, if 1310 * it isn't already there. 1311 */ 1312 1313 pmap->pm_stats.resident_count++; 1314 1315 ptepa = VM_PAGE_TO_PHYS(m); 1316 pmap->pm_pdir[ptepindex] = (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V); 1317 1318#if defined(PTPHINT) 1319 /* 1320 * Set the page table hint 1321 */ 1322 pmap->pm_ptphint = m; 1323#endif 1324 1325 /* 1326 * Try to use the new mapping, but if we cannot, then 1327 * do it with the routine that maps the page explicitly. 1328 */ 1329 if (needszero) { 1330 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1331 (((unsigned) PTDpde) & PG_FRAME)) { 1332 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1333 bzero((caddr_t) pteva, PAGE_SIZE); 1334 } else { 1335 pmap_zero_page(ptepa); 1336 } 1337 } 1338 1339 m->valid = VM_PAGE_BITS_ALL; 1340 m->flags |= PG_MAPPED; 1341 1342 return m; 1343} 1344 1345static vm_page_t 1346pmap_allocpte(pmap, va) 1347 pmap_t pmap; 1348 vm_offset_t va; 1349{ 1350 unsigned ptepindex; 1351 vm_offset_t ptepa; 1352 vm_page_t m; 1353 1354 /* 1355 * Calculate pagetable page index 1356 */ 1357 ptepindex = va >> PDRSHIFT; 1358 1359 /* 1360 * Get the page directory entry 1361 */ 1362 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1363 1364 /* 1365 * This supports switching from a 4MB page to a 1366 * normal 4K page. 1367 */ 1368 if (ptepa & PG_PS) { 1369 pmap->pm_pdir[ptepindex] = 0; 1370 ptepa = 0; 1371 invltlb(); 1372 } 1373 1374 /* 1375 * If the page table page is mapped, we just increment the 1376 * hold count, and activate it. 1377 */ 1378 if (ptepa) { 1379#if defined(PTPHINT) 1380 /* 1381 * In order to get the page table page, try the 1382 * hint first. 1383 */ 1384 if (pmap->pm_ptphint && 1385 (pmap->pm_ptphint->pindex == ptepindex)) { 1386 m = pmap->pm_ptphint; 1387 } else { 1388 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1389 pmap->pm_ptphint = m; 1390 } 1391#else 1392 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1393#endif 1394 ++m->hold_count; 1395 return m; 1396 } 1397 /* 1398 * Here if the pte page isn't mapped, or if it has been deallocated. 1399 */ 1400 return _pmap_allocpte(pmap, ptepindex); 1401} 1402 1403 1404/*************************************************** 1405* Pmap allocation/deallocation routines. 1406 ***************************************************/ 1407 1408/* 1409 * Release any resources held by the given physical map. 1410 * Called when a pmap initialized by pmap_pinit is being released. 1411 * Should only be called if the map contains no valid mappings. 1412 */ 1413void 1414pmap_release(pmap) 1415 register struct pmap *pmap; 1416{ 1417 vm_page_t p,n,ptdpg; 1418 vm_object_t object = pmap->pm_pteobj; 1419 1420#if defined(DIAGNOSTIC) 1421 if (object->ref_count != 1) 1422 panic("pmap_release: pteobj reference count != 1"); 1423#endif 1424 1425 ptdpg = NULL; 1426retry: 1427 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1428 n = TAILQ_NEXT(p, listq); 1429 if (p->pindex == PTDPTDI) { 1430 ptdpg = p; 1431 continue; 1432 } 1433 if (!pmap_release_free_page(pmap, p)) 1434 goto retry; 1435 } 1436 1437 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1438 goto retry; 1439 1440 vm_object_deallocate(object); 1441 if (pdstackptr < PDSTACKMAX) { 1442 pdstack[pdstackptr] = (vm_offset_t) pmap->pm_pdir; 1443 ++pdstackptr; 1444 } else { 1445 int pdstmp = pdstackptr - 1; 1446 kmem_free(kernel_map, pdstack[pdstmp], PAGE_SIZE); 1447 pdstack[pdstmp] = (vm_offset_t) pmap->pm_pdir; 1448 } 1449 pmap->pm_pdir = 0; 1450} 1451 1452/* 1453 * grow the number of kernel page table entries, if needed 1454 */ 1455void 1456pmap_growkernel(vm_offset_t addr) 1457{ 1458 struct proc *p; 1459 struct pmap *pmap; 1460 int s; 1461 vm_offset_t ptpkva, ptppaddr; 1462 vm_page_t nkpg; 1463#ifdef SMP 1464 int i; 1465#endif 1466 pd_entry_t newpdir; 1467 vm_pindex_t ptpidx; 1468 1469 s = splhigh(); 1470 if (kernel_vm_end == 0) { 1471 kernel_vm_end = KERNBASE; 1472 nkpt = 0; 1473 while (pdir_pde(PTD, kernel_vm_end)) { 1474 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1475 ++nkpt; 1476 } 1477 } 1478 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1479 while (kernel_vm_end < addr) { 1480 if (pdir_pde(PTD, kernel_vm_end)) { 1481 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1482 continue; 1483 } 1484 ++nkpt; 1485 ptpkva = (vm_offset_t) vtopte(addr); 1486 ptpidx = (ptpkva >> PAGE_SHIFT); 1487 /* 1488 * This index is bogus, but out of the way 1489 */ 1490 nkpg = vm_page_alloc(kernel_object, 1491 ptpidx, VM_ALLOC_SYSTEM); 1492 if (!nkpg) 1493 panic("pmap_growkernel: no memory to grow kernel"); 1494 1495 vm_page_wire(nkpg); 1496 vm_page_remove(nkpg); 1497 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1498 pmap_zero_page(ptppaddr); 1499 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW); 1500 pdir_pde(PTD, kernel_vm_end) = newpdir; 1501 1502#ifdef SMP 1503 for (i = 0; i < mp_ncpus; i++) { 1504 if (IdlePTDS[i]) 1505 pdir_pde(IdlePTDS[i], kernel_vm_end) = newpdir; 1506 } 1507#endif 1508 1509 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1510 if (p->p_vmspace) { 1511 pmap = &p->p_vmspace->vm_pmap; 1512 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1513 } 1514 } 1515 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir; 1516 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1517 } 1518 splx(s); 1519} 1520 1521/* 1522 * Retire the given physical map from service. 1523 * Should only be called if the map contains 1524 * no valid mappings. 1525 */ 1526void 1527pmap_destroy(pmap) 1528 register pmap_t pmap; 1529{ 1530 int count; 1531 1532 if (pmap == NULL) 1533 return; 1534 1535 count = --pmap->pm_count; 1536 if (count == 0) { 1537 pmap_release(pmap); 1538 panic("destroying a pmap is not yet implemented"); 1539 /* free((caddr_t) pmap, M_VMPMAP); */ 1540 } 1541} 1542 1543/* 1544 * Add a reference to the specified pmap. 1545 */ 1546void 1547pmap_reference(pmap) 1548 pmap_t pmap; 1549{ 1550 if (pmap != NULL) { 1551 pmap->pm_count++; 1552 } 1553} 1554 1555/*************************************************** 1556* page management routines. 1557 ***************************************************/ 1558 1559/* 1560 * free the pv_entry back to the free list 1561 */ 1562static inline void 1563free_pv_entry(pv) 1564 pv_entry_t pv; 1565{ 1566 pv_entry_count--; 1567 zfreei(pvzone, pv); 1568} 1569 1570/* 1571 * get a new pv_entry, allocating a block from the system 1572 * when needed. 1573 * the memory allocation is performed bypassing the malloc code 1574 * because of the possibility of allocations at interrupt time. 1575 */ 1576static pv_entry_t 1577get_pv_entry(void) 1578{ 1579 pv_entry_count++; 1580 if (pv_entry_high_water && 1581 (pv_entry_count > pv_entry_high_water) && 1582 (pmap_pagedaemon_waken == 0)) { 1583 pmap_pagedaemon_waken = 1; 1584 wakeup (&vm_pages_needed); 1585 } 1586 return zalloci(pvzone); 1587} 1588 1589/* 1590 * This routine is very drastic, but can save the system 1591 * in a pinch. 1592 */ 1593void 1594pmap_collect() { 1595 pv_table_t *ppv; 1596 int i; 1597 vm_offset_t pa; 1598 vm_page_t m; 1599 static int warningdone=0; 1600 1601 if (pmap_pagedaemon_waken == 0) 1602 return; 1603 1604 if (warningdone < 5) { 1605 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1606 warningdone++; 1607 } 1608 1609 for(i = 0; i < pv_npg; i++) { 1610 if ((ppv = &pv_table[i]) == 0) 1611 continue; 1612 m = ppv->pv_vm_page; 1613 if ((pa = VM_PAGE_TO_PHYS(m)) == 0) 1614 continue; 1615 if (m->wire_count || m->hold_count || m->busy || (m->flags & PG_BUSY)) 1616 continue; 1617 pmap_remove_all(pa); 1618 } 1619 pmap_pagedaemon_waken = 0; 1620} 1621 1622 1623/* 1624 * If it is the first entry on the list, it is actually 1625 * in the header and we must copy the following entry up 1626 * to the header. Otherwise we must search the list for 1627 * the entry. In either case we free the now unused entry. 1628 */ 1629 1630static int 1631pmap_remove_entry(pmap, ppv, va) 1632 struct pmap *pmap; 1633 pv_table_t *ppv; 1634 vm_offset_t va; 1635{ 1636 pv_entry_t pv; 1637 int rtval; 1638 int s; 1639 1640 s = splvm(); 1641 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1642 for (pv = TAILQ_FIRST(&ppv->pv_list); 1643 pv; 1644 pv = TAILQ_NEXT(pv, pv_list)) { 1645 if (pmap == pv->pv_pmap && va == pv->pv_va) 1646 break; 1647 } 1648 } else { 1649 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1650 pv; 1651 pv = TAILQ_NEXT(pv, pv_plist)) { 1652 if (va == pv->pv_va) 1653 break; 1654 } 1655 } 1656 1657 rtval = 0; 1658 if (pv) { 1659 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1660 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1661 --ppv->pv_list_count; 1662 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 1663 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 1664 } 1665 1666 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1667 free_pv_entry(pv); 1668 } 1669 1670 splx(s); 1671 return rtval; 1672} 1673 1674/* 1675 * Create a pv entry for page at pa for 1676 * (pmap, va). 1677 */ 1678static void 1679pmap_insert_entry(pmap, va, mpte, pa) 1680 pmap_t pmap; 1681 vm_offset_t va; 1682 vm_page_t mpte; 1683 vm_offset_t pa; 1684{ 1685 1686 int s; 1687 pv_entry_t pv; 1688 pv_table_t *ppv; 1689 1690 s = splvm(); 1691 pv = get_pv_entry(); 1692 pv->pv_va = va; 1693 pv->pv_pmap = pmap; 1694 pv->pv_ptem = mpte; 1695 1696 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1697 1698 ppv = pa_to_pvh(pa); 1699 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1700 ++ppv->pv_list_count; 1701 1702 splx(s); 1703} 1704 1705/* 1706 * pmap_remove_pte: do the things to unmap a page in a process 1707 */ 1708static int 1709pmap_remove_pte(pmap, ptq, va) 1710 struct pmap *pmap; 1711 unsigned *ptq; 1712 vm_offset_t va; 1713{ 1714 unsigned oldpte; 1715 pv_table_t *ppv; 1716 1717 oldpte = *ptq; 1718 *ptq = 0; 1719 if (oldpte & PG_W) 1720 pmap->pm_stats.wired_count -= 1; 1721 /* 1722 * Machines that don't support invlpg, also don't support 1723 * PG_G. 1724 */ 1725 if (oldpte & PG_G) 1726 invlpg(va); 1727 pmap->pm_stats.resident_count -= 1; 1728 if (oldpte & PG_MANAGED) { 1729 ppv = pa_to_pvh(oldpte); 1730 if (oldpte & PG_M) { 1731#if defined(PMAP_DIAGNOSTIC) 1732 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1733 printf("pmap_remove: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, (int) oldpte); 1734 } 1735#endif 1736 if (pmap_track_modified(va)) 1737 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1738 } 1739 return pmap_remove_entry(pmap, ppv, va); 1740 } else { 1741 return pmap_unuse_pt(pmap, va, NULL); 1742 } 1743 1744 return 0; 1745} 1746 1747/* 1748 * Remove a single page from a process address space 1749 */ 1750static void 1751pmap_remove_page(pmap, va) 1752 struct pmap *pmap; 1753 register vm_offset_t va; 1754{ 1755 register unsigned *ptq; 1756 1757 /* 1758 * if there is no pte for this address, just skip it!!! 1759 */ 1760 if (*pmap_pde(pmap, va) == 0) { 1761 return; 1762 } 1763 1764 /* 1765 * get a local va for mappings for this pmap. 1766 */ 1767 ptq = get_ptbase(pmap) + i386_btop(va); 1768 if (*ptq) { 1769 (void) pmap_remove_pte(pmap, ptq, va); 1770 invltlb_1pg(va); 1771 } 1772 return; 1773} 1774 1775/* 1776 * Remove the given range of addresses from the specified map. 1777 * 1778 * It is assumed that the start and end are properly 1779 * rounded to the page size. 1780 */ 1781void 1782pmap_remove(pmap, sva, eva) 1783 struct pmap *pmap; 1784 register vm_offset_t sva; 1785 register vm_offset_t eva; 1786{ 1787 register unsigned *ptbase; 1788 vm_offset_t pdnxt; 1789 vm_offset_t ptpaddr; 1790 vm_offset_t sindex, eindex; 1791 int anyvalid; 1792 1793 if (pmap == NULL) 1794 return; 1795 1796 if (pmap->pm_stats.resident_count == 0) 1797 return; 1798 1799 /* 1800 * special handling of removing one page. a very 1801 * common operation and easy to short circuit some 1802 * code. 1803 */ 1804 if (((sva + PAGE_SIZE) == eva) && 1805 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1806 pmap_remove_page(pmap, sva); 1807 return; 1808 } 1809 1810 anyvalid = 0; 1811 1812 /* 1813 * Get a local virtual address for the mappings that are being 1814 * worked with. 1815 */ 1816 ptbase = get_ptbase(pmap); 1817 1818 sindex = i386_btop(sva); 1819 eindex = i386_btop(eva); 1820 1821 for (; sindex < eindex; sindex = pdnxt) { 1822 unsigned pdirindex; 1823 1824 /* 1825 * Calculate index for next page table. 1826 */ 1827 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1828 if (pmap->pm_stats.resident_count == 0) 1829 break; 1830 1831 pdirindex = sindex / NPDEPG; 1832 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1833 pmap->pm_pdir[pdirindex] = 0; 1834 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1835 anyvalid++; 1836 continue; 1837 } 1838 1839 /* 1840 * Weed out invalid mappings. Note: we assume that the page 1841 * directory table is always allocated, and in kernel virtual. 1842 */ 1843 if (ptpaddr == 0) 1844 continue; 1845 1846 /* 1847 * Limit our scan to either the end of the va represented 1848 * by the current page table page, or to the end of the 1849 * range being removed. 1850 */ 1851 if (pdnxt > eindex) { 1852 pdnxt = eindex; 1853 } 1854 1855 for ( ;sindex != pdnxt; sindex++) { 1856 vm_offset_t va; 1857 if (ptbase[sindex] == 0) { 1858 continue; 1859 } 1860 va = i386_ptob(sindex); 1861 1862 anyvalid++; 1863 if (pmap_remove_pte(pmap, 1864 ptbase + sindex, va)) 1865 break; 1866 } 1867 } 1868 1869 if (anyvalid) { 1870 invltlb(); 1871 } 1872} 1873 1874/* 1875 * Routine: pmap_remove_all 1876 * Function: 1877 * Removes this physical page from 1878 * all physical maps in which it resides. 1879 * Reflects back modify bits to the pager. 1880 * 1881 * Notes: 1882 * Original versions of this routine were very 1883 * inefficient because they iteratively called 1884 * pmap_remove (slow...) 1885 */ 1886 1887static void 1888pmap_remove_all(pa) 1889 vm_offset_t pa; 1890{ 1891 register pv_entry_t pv; 1892 pv_table_t *ppv; 1893 register unsigned *pte, tpte; 1894 int nmodify; 1895 int update_needed; 1896 int s; 1897 1898 nmodify = 0; 1899 update_needed = 0; 1900#if defined(PMAP_DIAGNOSTIC) 1901 /* 1902 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1903 * pages! 1904 */ 1905 if (!pmap_is_managed(pa)) { 1906 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%lx", pa); 1907 } 1908#endif 1909 1910 s = splvm(); 1911 ppv = pa_to_pvh(pa); 1912 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1913 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1914 1915 pv->pv_pmap->pm_stats.resident_count--; 1916 1917 tpte = *pte; 1918 *pte = 0; 1919 if (tpte & PG_W) 1920 pv->pv_pmap->pm_stats.wired_count--; 1921 /* 1922 * Update the vm_page_t clean and reference bits. 1923 */ 1924 if (tpte & PG_M) { 1925#if defined(PMAP_DIAGNOSTIC) 1926 if (pmap_nw_modified((pt_entry_t) tpte)) { 1927 printf("pmap_remove_all: modified page not writable: va: 0x%lx, pte: 0x%lx\n", pv->pv_va, tpte); 1928 } 1929#endif 1930 if (pmap_track_modified(pv->pv_va)) 1931 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1932 } 1933 if (!update_needed && 1934 ((!curproc || (&curproc->p_vmspace->vm_pmap == pv->pv_pmap)) || 1935 (pv->pv_pmap == kernel_pmap))) { 1936 update_needed = 1; 1937 } 1938 1939 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1940 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1941 --ppv->pv_list_count; 1942 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1943 free_pv_entry(pv); 1944 } 1945 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 1946 1947 1948 if (update_needed) 1949 invltlb(); 1950 splx(s); 1951 return; 1952} 1953 1954/* 1955 * Set the physical protection on the 1956 * specified range of this map as requested. 1957 */ 1958void 1959pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1960{ 1961 register unsigned *ptbase; 1962 vm_offset_t pdnxt; 1963 vm_offset_t ptpaddr; 1964 vm_offset_t sindex, eindex; 1965 int anychanged; 1966 1967 1968 if (pmap == NULL) 1969 return; 1970 1971 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1972 pmap_remove(pmap, sva, eva); 1973 return; 1974 } 1975 1976 anychanged = 0; 1977 1978 ptbase = get_ptbase(pmap); 1979 1980 sindex = i386_btop(sva); 1981 eindex = i386_btop(eva); 1982 1983 for (; sindex < eindex; sindex = pdnxt) { 1984 1985 unsigned pdirindex; 1986 1987 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1988 1989 pdirindex = sindex / NPDEPG; 1990 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1991 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1992 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1993 anychanged++; 1994 continue; 1995 } 1996 1997 /* 1998 * Weed out invalid mappings. Note: we assume that the page 1999 * directory table is always allocated, and in kernel virtual. 2000 */ 2001 if (ptpaddr == 0) 2002 continue; 2003 2004 if (pdnxt > eindex) { 2005 pdnxt = eindex; 2006 } 2007 2008 for (; sindex != pdnxt; sindex++) { 2009 2010 unsigned pbits = ptbase[sindex]; 2011 2012 if (prot & VM_PROT_WRITE) { 2013 if ((pbits & (PG_RW|PG_V)) == PG_V) { 2014 if (pbits & PG_MANAGED) { 2015 vm_page_t m = PHYS_TO_VM_PAGE(pbits); 2016 m->flags |= PG_WRITEABLE; 2017 m->object->flags |= OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY; 2018 } 2019 ptbase[sindex] = pbits | PG_RW; 2020 anychanged = 1; 2021 } 2022 } else if (pbits & PG_RW) { 2023 if (pbits & PG_M) { 2024 vm_offset_t sva1 = i386_ptob(sindex); 2025 if ((pbits & PG_MANAGED) && pmap_track_modified(sva1)) { 2026 vm_page_t m = PHYS_TO_VM_PAGE(pbits); 2027 m->dirty = VM_PAGE_BITS_ALL; 2028 } 2029 } 2030 ptbase[sindex] = pbits & ~(PG_M|PG_RW); 2031 anychanged = 1; 2032 } 2033 } 2034 } 2035 if (anychanged) 2036 invltlb(); 2037} 2038 2039/* 2040 * Insert the given physical page (p) at 2041 * the specified virtual address (v) in the 2042 * target physical map with the protection requested. 2043 * 2044 * If specified, the page will be wired down, meaning 2045 * that the related pte can not be reclaimed. 2046 * 2047 * NB: This is the only routine which MAY NOT lazy-evaluate 2048 * or lose information. That is, this routine must actually 2049 * insert this page into the given map NOW. 2050 */ 2051void 2052pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot, 2053 boolean_t wired) 2054{ 2055 register unsigned *pte; 2056 vm_offset_t opa; 2057 vm_offset_t origpte, newpte; 2058 vm_page_t mpte; 2059 2060 if (pmap == NULL) 2061 return; 2062 2063 va &= PG_FRAME; 2064#ifdef PMAP_DIAGNOSTIC 2065 if (va > VM_MAX_KERNEL_ADDRESS) 2066 panic("pmap_enter: toobig"); 2067 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 2068 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 2069#endif 2070 2071 mpte = NULL; 2072 /* 2073 * In the case that a page table page is not 2074 * resident, we are creating it here. 2075 */ 2076 if (va < UPT_MIN_ADDRESS) { 2077 mpte = pmap_allocpte(pmap, va); 2078 } 2079#if 0 && defined(PMAP_DIAGNOSTIC) 2080 else { 2081 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 2082 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 2083 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 2084 pmap->pm_pdir[PTDPTDI], origpte, va); 2085 } 2086 if (smp_active) { 2087 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid]; 2088 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2089 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2090 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2091 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr); 2092 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2093 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2094 } 2095 } 2096 } 2097#endif 2098 2099 pte = pmap_pte(pmap, va); 2100 /* 2101 * Page Directory table entry not valid, we need a new PT page 2102 */ 2103 if (pte == NULL) { 2104 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%lx\n", 2105 pmap->pm_pdir[PTDPTDI], va); 2106 } 2107 2108 origpte = *(vm_offset_t *)pte; 2109 pa &= PG_FRAME; 2110 opa = origpte & PG_FRAME; 2111 if (origpte & PG_PS) 2112 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2113 2114 /* 2115 * Mapping has not changed, must be protection or wiring change. 2116 */ 2117 if (origpte && (opa == pa)) { 2118 /* 2119 * Wiring change, just update stats. We don't worry about 2120 * wiring PT pages as they remain resident as long as there 2121 * are valid mappings in them. Hence, if a user page is wired, 2122 * the PT page will be also. 2123 */ 2124 if (wired && ((origpte & PG_W) == 0)) 2125 pmap->pm_stats.wired_count++; 2126 else if (!wired && (origpte & PG_W)) 2127 pmap->pm_stats.wired_count--; 2128 2129#if defined(PMAP_DIAGNOSTIC) 2130 if (pmap_nw_modified((pt_entry_t) origpte)) { 2131 printf("pmap_enter: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, origpte); 2132 } 2133#endif 2134 2135 /* 2136 * We might be turning off write access to the page, 2137 * so we go ahead and sense modify status. 2138 */ 2139 if (origpte & PG_MANAGED) { 2140 vm_page_t m; 2141 if (origpte & PG_M) { 2142 if (pmap_track_modified(va)) { 2143 m = PHYS_TO_VM_PAGE(pa); 2144 m->dirty = VM_PAGE_BITS_ALL; 2145 } 2146 } 2147 pa |= PG_MANAGED; 2148 } 2149 2150 if (mpte) 2151 --mpte->hold_count; 2152 2153 goto validate; 2154 } 2155 /* 2156 * Mapping has changed, invalidate old range and fall through to 2157 * handle validating new mapping. 2158 */ 2159 if (opa) { 2160 int err; 2161 err = pmap_remove_pte(pmap, pte, va); 2162 if (err) 2163 panic("pmap_enter: pte vanished, va: 0x%x", va); 2164 } 2165 2166 /* 2167 * Enter on the PV list if part of our managed memory Note that we 2168 * raise IPL while manipulating pv_table since pmap_enter can be 2169 * called at interrupt time. 2170 */ 2171 if (pmap_is_managed(pa)) { 2172 pmap_insert_entry(pmap, va, mpte, pa); 2173 pa |= PG_MANAGED; 2174 } 2175 2176 /* 2177 * Increment counters 2178 */ 2179 pmap->pm_stats.resident_count++; 2180 if (wired) 2181 pmap->pm_stats.wired_count++; 2182 2183validate: 2184 /* 2185 * Now validate mapping with desired protection/wiring. 2186 */ 2187 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2188 2189 if (wired) 2190 newpte |= PG_W; 2191 if (va < UPT_MIN_ADDRESS) 2192 newpte |= PG_U; 2193 if (pmap == kernel_pmap) 2194 newpte |= pgeflag; 2195 2196 /* 2197 * if the mapping or permission bits are different, we need 2198 * to update the pte. 2199 */ 2200 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2201 *pte = newpte; 2202 if (origpte) 2203 invltlb_1pg(va); 2204 } 2205} 2206 2207/* 2208 * this code makes some *MAJOR* assumptions: 2209 * 1. Current pmap & pmap exists. 2210 * 2. Not wired. 2211 * 3. Read access. 2212 * 4. No page table pages. 2213 * 5. Tlbflush is deferred to calling procedure. 2214 * 6. Page IS managed. 2215 * but is *MUCH* faster than pmap_enter... 2216 */ 2217 2218static vm_page_t 2219pmap_enter_quick(pmap, va, pa, mpte) 2220 register pmap_t pmap; 2221 vm_offset_t va; 2222 register vm_offset_t pa; 2223 vm_page_t mpte; 2224{ 2225 register unsigned *pte; 2226 2227 /* 2228 * In the case that a page table page is not 2229 * resident, we are creating it here. 2230 */ 2231 if (va < UPT_MIN_ADDRESS) { 2232 unsigned ptepindex; 2233 vm_offset_t ptepa; 2234 2235 /* 2236 * Calculate pagetable page index 2237 */ 2238 ptepindex = va >> PDRSHIFT; 2239 if (mpte && (mpte->pindex == ptepindex)) { 2240 ++mpte->hold_count; 2241 } else { 2242retry: 2243 /* 2244 * Get the page directory entry 2245 */ 2246 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2247 2248 /* 2249 * If the page table page is mapped, we just increment 2250 * the hold count, and activate it. 2251 */ 2252 if (ptepa) { 2253 if (ptepa & PG_PS) 2254 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2255#if defined(PTPHINT) 2256 if (pmap->pm_ptphint && 2257 (pmap->pm_ptphint->pindex == ptepindex)) { 2258 mpte = pmap->pm_ptphint; 2259 } else { 2260 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2261 pmap->pm_ptphint = mpte; 2262 } 2263#else 2264 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2265#endif 2266 if (mpte == NULL) 2267 goto retry; 2268 ++mpte->hold_count; 2269 } else { 2270 mpte = _pmap_allocpte(pmap, ptepindex); 2271 } 2272 } 2273 } else { 2274 mpte = NULL; 2275 } 2276 2277 /* 2278 * This call to vtopte makes the assumption that we are 2279 * entering the page into the current pmap. In order to support 2280 * quick entry into any pmap, one would likely use pmap_pte_quick. 2281 * But that isn't as quick as vtopte. 2282 */ 2283 pte = (unsigned *)vtopte(va); 2284 if (*pte) { 2285 if (mpte) 2286 pmap_unwire_pte_hold(pmap, mpte); 2287 return 0; 2288 } 2289 2290 /* 2291 * Enter on the PV list if part of our managed memory Note that we 2292 * raise IPL while manipulating pv_table since pmap_enter can be 2293 * called at interrupt time. 2294 */ 2295 pmap_insert_entry(pmap, va, mpte, pa); 2296 2297 /* 2298 * Increment counters 2299 */ 2300 pmap->pm_stats.resident_count++; 2301 2302 /* 2303 * Now validate mapping with RO protection 2304 */ 2305 *pte = pa | PG_V | PG_U | PG_MANAGED; 2306 2307 return mpte; 2308} 2309 2310#define MAX_INIT_PT (96) 2311/* 2312 * pmap_object_init_pt preloads the ptes for a given object 2313 * into the specified pmap. This eliminates the blast of soft 2314 * faults on process startup and immediately after an mmap. 2315 */ 2316void 2317pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2318 pmap_t pmap; 2319 vm_offset_t addr; 2320 vm_object_t object; 2321 vm_pindex_t pindex; 2322 vm_size_t size; 2323 int limit; 2324{ 2325 vm_offset_t tmpidx; 2326 int psize; 2327 vm_page_t p, mpte; 2328 int objpgs; 2329 2330 if (!pmap) 2331 return; 2332 2333 /* 2334 * This code maps large physical mmap regions into the 2335 * processor address space. Note that some shortcuts 2336 * are taken, but the code works. 2337 */ 2338 if (pseflag && 2339 (object->type == OBJT_DEVICE) && 2340 ((addr & (NBPDR - 1)) == 0) && 2341 ((size & (NBPDR - 1)) == 0) ) { 2342 int i; 2343 int s; 2344 vm_page_t m[1]; 2345 unsigned int ptepindex; 2346 int npdes; 2347 vm_offset_t ptepa; 2348 2349 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2350 return; 2351 2352 s = splhigh(); 2353retry: 2354 p = vm_page_lookup(object, pindex); 2355 if (p && (p->flags & PG_BUSY)) { 2356 tsleep(p, PVM, "init4p", 0); 2357 goto retry; 2358 } 2359 splx(s); 2360 2361 if (p == NULL) { 2362 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2363 if (p == NULL) 2364 return; 2365 m[0] = p; 2366 2367 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2368 PAGE_WAKEUP(p); 2369 vm_page_free(p); 2370 return; 2371 } 2372 2373 p = vm_page_lookup(object, pindex); 2374 PAGE_WAKEUP(p); 2375 } 2376 2377 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2378 if (ptepa & (NBPDR - 1)) { 2379 return; 2380 } 2381 2382 p->valid = VM_PAGE_BITS_ALL; 2383 2384 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2385 npdes = size >> PDRSHIFT; 2386 for(i=0;i<npdes;i++) { 2387 pmap->pm_pdir[ptepindex] = 2388 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2389 ptepa += NBPDR; 2390 ptepindex += 1; 2391 } 2392 p->flags |= PG_MAPPED; 2393 invltlb(); 2394 return; 2395 } 2396 2397 psize = i386_btop(size); 2398 2399 if ((object->type != OBJT_VNODE) || 2400 (limit && (psize > MAX_INIT_PT) && 2401 (object->resident_page_count > MAX_INIT_PT))) { 2402 return; 2403 } 2404 2405 if (psize + pindex > object->size) 2406 psize = object->size - pindex; 2407 2408 mpte = NULL; 2409 /* 2410 * if we are processing a major portion of the object, then scan the 2411 * entire thing. 2412 */ 2413 if (psize > (object->size >> 2)) { 2414 objpgs = psize; 2415 2416 for (p = TAILQ_FIRST(&object->memq); 2417 ((objpgs > 0) && (p != NULL)); 2418 p = TAILQ_NEXT(p, listq)) { 2419 2420 tmpidx = p->pindex; 2421 if (tmpidx < pindex) { 2422 continue; 2423 } 2424 tmpidx -= pindex; 2425 if (tmpidx >= psize) { 2426 continue; 2427 } 2428 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2429 (p->busy == 0) && 2430 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2431 if ((p->queue - p->pc) == PQ_CACHE) 2432 vm_page_deactivate(p); 2433 p->flags |= PG_BUSY; 2434 mpte = pmap_enter_quick(pmap, 2435 addr + i386_ptob(tmpidx), 2436 VM_PAGE_TO_PHYS(p), mpte); 2437 p->flags |= PG_MAPPED; 2438 PAGE_WAKEUP(p); 2439 } 2440 objpgs -= 1; 2441 } 2442 } else { 2443 /* 2444 * else lookup the pages one-by-one. 2445 */ 2446 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2447 p = vm_page_lookup(object, tmpidx + pindex); 2448 if (p && 2449 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2450 (p->busy == 0) && 2451 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2452 if ((p->queue - p->pc) == PQ_CACHE) 2453 vm_page_deactivate(p); 2454 p->flags |= PG_BUSY; 2455 mpte = pmap_enter_quick(pmap, 2456 addr + i386_ptob(tmpidx), 2457 VM_PAGE_TO_PHYS(p), mpte); 2458 p->flags |= PG_MAPPED; 2459 PAGE_WAKEUP(p); 2460 } 2461 } 2462 } 2463 return; 2464} 2465 2466/* 2467 * pmap_prefault provides a quick way of clustering 2468 * pagefaults into a processes address space. It is a "cousin" 2469 * of pmap_object_init_pt, except it runs at page fault time instead 2470 * of mmap time. 2471 */ 2472#define PFBAK 2 2473#define PFFOR 2 2474#define PAGEORDER_SIZE (PFBAK+PFFOR) 2475 2476static int pmap_prefault_pageorder[] = { 2477 -PAGE_SIZE, PAGE_SIZE, -2 * PAGE_SIZE, 2 * PAGE_SIZE 2478}; 2479 2480void 2481pmap_prefault(pmap, addra, entry, object) 2482 pmap_t pmap; 2483 vm_offset_t addra; 2484 vm_map_entry_t entry; 2485 vm_object_t object; 2486{ 2487 int i; 2488 vm_offset_t starta; 2489 vm_offset_t addr; 2490 vm_pindex_t pindex; 2491 vm_page_t m, mpte; 2492 2493 if (entry->object.vm_object != object) 2494 return; 2495 2496 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) 2497 return; 2498 2499 starta = addra - PFBAK * PAGE_SIZE; 2500 if (starta < entry->start) { 2501 starta = entry->start; 2502 } else if (starta > addra) { 2503 starta = 0; 2504 } 2505 2506 mpte = NULL; 2507 for (i = 0; i < PAGEORDER_SIZE; i++) { 2508 vm_object_t lobject; 2509 unsigned *pte; 2510 2511 addr = addra + pmap_prefault_pageorder[i]; 2512 if (addr < starta || addr >= entry->end) 2513 continue; 2514 2515 if ((*pmap_pde(pmap, addr)) == NULL) 2516 continue; 2517 2518 pte = (unsigned *) vtopte(addr); 2519 if (*pte) 2520 continue; 2521 2522 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2523 lobject = object; 2524 for (m = vm_page_lookup(lobject, pindex); 2525 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2526 lobject = lobject->backing_object) { 2527 if (lobject->backing_object_offset & PAGE_MASK) 2528 break; 2529 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2530 m = vm_page_lookup(lobject->backing_object, pindex); 2531 } 2532 2533 /* 2534 * give-up when a page is not in memory 2535 */ 2536 if (m == NULL) 2537 break; 2538 2539 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2540 (m->busy == 0) && 2541 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2542 2543 if ((m->queue - m->pc) == PQ_CACHE) { 2544 vm_page_deactivate(m); 2545 } 2546 m->flags |= PG_BUSY; 2547 mpte = pmap_enter_quick(pmap, addr, 2548 VM_PAGE_TO_PHYS(m), mpte); 2549 m->flags |= PG_MAPPED; 2550 PAGE_WAKEUP(m); 2551 } 2552 } 2553} 2554 2555/* 2556 * Routine: pmap_change_wiring 2557 * Function: Change the wiring attribute for a map/virtual-address 2558 * pair. 2559 * In/out conditions: 2560 * The mapping must already exist in the pmap. 2561 */ 2562void 2563pmap_change_wiring(pmap, va, wired) 2564 register pmap_t pmap; 2565 vm_offset_t va; 2566 boolean_t wired; 2567{ 2568 register unsigned *pte; 2569 2570 if (pmap == NULL) 2571 return; 2572 2573 pte = pmap_pte(pmap, va); 2574 2575 if (wired && !pmap_pte_w(pte)) 2576 pmap->pm_stats.wired_count++; 2577 else if (!wired && pmap_pte_w(pte)) 2578 pmap->pm_stats.wired_count--; 2579 2580 /* 2581 * Wiring is not a hardware characteristic so there is no need to 2582 * invalidate TLB. 2583 */ 2584 pmap_pte_set_w(pte, wired); 2585} 2586 2587 2588 2589/* 2590 * Copy the range specified by src_addr/len 2591 * from the source map to the range dst_addr/len 2592 * in the destination map. 2593 * 2594 * This routine is only advisory and need not do anything. 2595 */ 2596 2597void 2598pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2599 pmap_t dst_pmap, src_pmap; 2600 vm_offset_t dst_addr; 2601 vm_size_t len; 2602 vm_offset_t src_addr; 2603{ 2604 vm_offset_t addr; 2605 vm_offset_t end_addr = src_addr + len; 2606 vm_offset_t pdnxt; 2607 unsigned src_frame, dst_frame; 2608 2609 if (dst_addr != src_addr) 2610 return; 2611 2612 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2613 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2614 return; 2615 } 2616 2617 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2618 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2619 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2620 invltlb(); 2621 } 2622 2623 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2624 unsigned *src_pte, *dst_pte; 2625 vm_page_t dstmpte, srcmpte; 2626 vm_offset_t srcptepaddr; 2627 unsigned ptepindex; 2628 2629 if (addr >= UPT_MIN_ADDRESS) 2630 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2631 2632 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2633 ptepindex = addr >> PDRSHIFT; 2634 2635 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2636 if (srcptepaddr == 0) 2637 continue; 2638 2639 if (srcptepaddr & PG_PS) { 2640 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2641 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2642 dst_pmap->pm_stats.resident_count += NBPDR; 2643 } 2644 continue; 2645 } 2646 2647 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2648 if ((srcmpte == NULL) || 2649 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2650 continue; 2651 2652 if (pdnxt > end_addr) 2653 pdnxt = end_addr; 2654 2655 src_pte = (unsigned *) vtopte(addr); 2656 dst_pte = (unsigned *) avtopte(addr); 2657 while (addr < pdnxt) { 2658 unsigned ptetemp; 2659 ptetemp = *src_pte; 2660 /* 2661 * we only virtual copy managed pages 2662 */ 2663 if ((ptetemp & PG_MANAGED) != 0) { 2664 /* 2665 * We have to check after allocpte for the 2666 * pte still being around... allocpte can 2667 * block. 2668 */ 2669 dstmpte = pmap_allocpte(dst_pmap, addr); 2670 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2671 /* 2672 * Clear the modified and 2673 * accessed (referenced) bits 2674 * during the copy. 2675 */ 2676 *dst_pte = ptetemp & ~(PG_M|PG_A); 2677 dst_pmap->pm_stats.resident_count++; 2678 pmap_insert_entry(dst_pmap, addr, 2679 dstmpte, 2680 (ptetemp & PG_FRAME)); 2681 } else { 2682 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2683 } 2684 if (dstmpte->hold_count >= srcmpte->hold_count) 2685 break; 2686 } 2687 addr += PAGE_SIZE; 2688 ++src_pte; 2689 ++dst_pte; 2690 } 2691 } 2692} 2693 2694/* 2695 * Routine: pmap_kernel 2696 * Function: 2697 * Returns the physical map handle for the kernel. 2698 */ 2699pmap_t 2700pmap_kernel() 2701{ 2702 return (kernel_pmap); 2703} 2704 2705/* 2706 * pmap_zero_page zeros the specified (machine independent) 2707 * page by mapping the page into virtual memory and using 2708 * bzero to clear its contents, one machine dependent page 2709 * at a time. 2710 */ 2711void 2712pmap_zero_page(phys) 2713 vm_offset_t phys; 2714{ 2715#ifdef SMP 2716 if (*(int *) prv_CMAP3) 2717 panic("pmap_zero_page: prv_CMAP3 busy"); 2718 2719 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME); 2720 invltlb_1pg((vm_offset_t) &prv_CPAGE3); 2721 2722 bzero(&prv_CPAGE3, PAGE_SIZE); 2723 2724 *(int *) prv_CMAP3 = 0; 2725 invltlb_1pg((vm_offset_t) &prv_CPAGE3); 2726#else 2727 if (*(int *) CMAP2) 2728 panic("pmap_zero_page: CMAP busy"); 2729 2730 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME); 2731 bzero(CADDR2, PAGE_SIZE); 2732 *(int *) CMAP2 = 0; 2733 invltlb_1pg((vm_offset_t) CADDR2); 2734#endif 2735} 2736 2737/* 2738 * pmap_copy_page copies the specified (machine independent) 2739 * page by mapping the page into virtual memory and using 2740 * bcopy to copy the page, one machine dependent page at a 2741 * time. 2742 */ 2743void 2744pmap_copy_page(src, dst) 2745 vm_offset_t src; 2746 vm_offset_t dst; 2747{ 2748#ifdef SMP 2749 if (*(int *) prv_CMAP1) 2750 panic("pmap_copy_page: prv_CMAP1 busy"); 2751 if (*(int *) prv_CMAP2) 2752 panic("pmap_copy_page: prv_CMAP2 busy"); 2753 2754 *(int *) prv_CMAP1 = PG_V | PG_RW | (src & PG_FRAME); 2755 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME); 2756 2757 invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2); 2758 2759 bcopy(&prv_CPAGE1, &prv_CPAGE2, PAGE_SIZE); 2760 2761 *(int *) prv_CMAP1 = 0; 2762 *(int *) prv_CMAP2 = 0; 2763 invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2); 2764#else 2765 if (*(int *) CMAP1 || *(int *) CMAP2) 2766 panic("pmap_copy_page: CMAP busy"); 2767 2768 *(int *) CMAP1 = PG_V | PG_RW | (src & PG_FRAME); 2769 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME); 2770 2771 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2772 2773 *(int *) CMAP1 = 0; 2774 *(int *) CMAP2 = 0; 2775 invltlb_2pg( (vm_offset_t) CADDR1, (vm_offset_t) CADDR2); 2776#endif 2777} 2778 2779 2780/* 2781 * Routine: pmap_pageable 2782 * Function: 2783 * Make the specified pages (by pmap, offset) 2784 * pageable (or not) as requested. 2785 * 2786 * A page which is not pageable may not take 2787 * a fault; therefore, its page table entry 2788 * must remain valid for the duration. 2789 * 2790 * This routine is merely advisory; pmap_enter 2791 * will specify that these pages are to be wired 2792 * down (or not) as appropriate. 2793 */ 2794void 2795pmap_pageable(pmap, sva, eva, pageable) 2796 pmap_t pmap; 2797 vm_offset_t sva, eva; 2798 boolean_t pageable; 2799{ 2800} 2801 2802/* 2803 * this routine returns true if a physical page resides 2804 * in the given pmap. 2805 */ 2806boolean_t 2807pmap_page_exists(pmap, pa) 2808 pmap_t pmap; 2809 vm_offset_t pa; 2810{ 2811 register pv_entry_t pv; 2812 pv_table_t *ppv; 2813 int s; 2814 2815 if (!pmap_is_managed(pa)) 2816 return FALSE; 2817 2818 s = splvm(); 2819 2820 ppv = pa_to_pvh(pa); 2821 /* 2822 * Not found, check current mappings returning immediately if found. 2823 */ 2824 for (pv = TAILQ_FIRST(&ppv->pv_list); 2825 pv; 2826 pv = TAILQ_NEXT(pv, pv_list)) { 2827 if (pv->pv_pmap == pmap) { 2828 splx(s); 2829 return TRUE; 2830 } 2831 } 2832 splx(s); 2833 return (FALSE); 2834} 2835 2836#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2837/* 2838 * Remove all pages from specified address space 2839 * this aids process exit speeds. Also, this code 2840 * is special cased for current process only, but 2841 * can have the more generic (and slightly slower) 2842 * mode enabled. This is much faster than pmap_remove 2843 * in the case of running down an entire address space. 2844 */ 2845void 2846pmap_remove_pages(pmap, sva, eva) 2847 pmap_t pmap; 2848 vm_offset_t sva, eva; 2849{ 2850 unsigned *pte, tpte; 2851 pv_table_t *ppv; 2852 pv_entry_t pv, npv; 2853 int s; 2854 2855#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2856 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) { 2857 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2858 return; 2859 } 2860#endif 2861 2862 s = splvm(); 2863 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2864 pv; 2865 pv = npv) { 2866 2867 if (pv->pv_va >= eva || pv->pv_va < sva) { 2868 npv = TAILQ_NEXT(pv, pv_plist); 2869 continue; 2870 } 2871 2872#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2873 pte = (unsigned *)vtopte(pv->pv_va); 2874#else 2875 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2876#endif 2877 tpte = *pte; 2878 2879/* 2880 * We cannot remove wired pages from a process' mapping at this time 2881 */ 2882 if (tpte & PG_W) { 2883 npv = TAILQ_NEXT(pv, pv_plist); 2884 continue; 2885 } 2886 *pte = 0; 2887 2888 ppv = pa_to_pvh(tpte); 2889 2890 pv->pv_pmap->pm_stats.resident_count--; 2891 2892 /* 2893 * Update the vm_page_t clean and reference bits. 2894 */ 2895 if (tpte & PG_M) { 2896 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 2897 } 2898 2899 2900 npv = TAILQ_NEXT(pv, pv_plist); 2901 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2902 2903 --ppv->pv_list_count; 2904 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2905 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2906 ppv->pv_vm_page->flags &= ~(PG_MAPPED|PG_WRITEABLE); 2907 } 2908 2909 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2910 free_pv_entry(pv); 2911 } 2912 splx(s); 2913 invltlb(); 2914} 2915 2916/* 2917 * pmap_testbit tests bits in pte's 2918 * note that the testbit/changebit routines are inline, 2919 * and a lot of things compile-time evaluate. 2920 */ 2921static boolean_t 2922pmap_testbit(pa, bit) 2923 register vm_offset_t pa; 2924 int bit; 2925{ 2926 register pv_entry_t pv; 2927 pv_table_t *ppv; 2928 unsigned *pte; 2929 int s; 2930 2931 if (!pmap_is_managed(pa)) 2932 return FALSE; 2933 2934 ppv = pa_to_pvh(pa); 2935 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 2936 return FALSE; 2937 2938 s = splvm(); 2939 2940 for (pv = TAILQ_FIRST(&ppv->pv_list); 2941 pv; 2942 pv = TAILQ_NEXT(pv, pv_list)) { 2943 2944 /* 2945 * if the bit being tested is the modified bit, then 2946 * mark clean_map and ptes as never 2947 * modified. 2948 */ 2949 if (bit & (PG_A|PG_M)) { 2950 if (!pmap_track_modified(pv->pv_va)) 2951 continue; 2952 } 2953 2954#if defined(PMAP_DIAGNOSTIC) 2955 if (!pv->pv_pmap) { 2956 printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va); 2957 continue; 2958 } 2959#endif 2960 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2961 if (*pte & bit) { 2962 splx(s); 2963 return TRUE; 2964 } 2965 } 2966 splx(s); 2967 return (FALSE); 2968} 2969 2970/* 2971 * this routine is used to modify bits in ptes 2972 */ 2973static void 2974pmap_changebit(pa, bit, setem) 2975 vm_offset_t pa; 2976 int bit; 2977 boolean_t setem; 2978{ 2979 register pv_entry_t pv; 2980 pv_table_t *ppv; 2981 register unsigned *pte; 2982 int changed; 2983 int s; 2984 2985 if (!pmap_is_managed(pa)) 2986 return; 2987 2988 s = splvm(); 2989 changed = 0; 2990 ppv = pa_to_pvh(pa); 2991 2992 /* 2993 * Loop over all current mappings setting/clearing as appropos If 2994 * setting RO do we need to clear the VAC? 2995 */ 2996 for (pv = TAILQ_FIRST(&ppv->pv_list); 2997 pv; 2998 pv = TAILQ_NEXT(pv, pv_list)) { 2999 3000 /* 3001 * don't write protect pager mappings 3002 */ 3003 if (!setem && (bit == PG_RW)) { 3004 if (!pmap_track_modified(pv->pv_va)) 3005 continue; 3006 } 3007 3008#if defined(PMAP_DIAGNOSTIC) 3009 if (!pv->pv_pmap) { 3010 printf("Null pmap (cb) at va: 0x%lx\n", pv->pv_va); 3011 continue; 3012 } 3013#endif 3014 3015 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3016 3017 if (setem) { 3018 *(int *)pte |= bit; 3019 changed = 1; 3020 } else { 3021 vm_offset_t pbits = *(vm_offset_t *)pte; 3022 if (pbits & bit) { 3023 changed = 1; 3024 if (bit == PG_RW) { 3025 if (pbits & PG_M) { 3026 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 3027 } 3028 *(int *)pte = pbits & ~(PG_M|PG_RW); 3029 } else { 3030 *(int *)pte = pbits & ~bit; 3031 } 3032 } 3033 } 3034 } 3035 splx(s); 3036 if (changed) 3037 invltlb(); 3038} 3039 3040/* 3041 * pmap_page_protect: 3042 * 3043 * Lower the permission for all mappings to a given page. 3044 */ 3045void 3046pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 3047{ 3048 if ((prot & VM_PROT_WRITE) == 0) { 3049 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3050 pmap_changebit(phys, PG_RW, FALSE); 3051 } else { 3052 pmap_remove_all(phys); 3053 } 3054 } 3055} 3056 3057vm_offset_t 3058pmap_phys_address(ppn) 3059 int ppn; 3060{ 3061 return (i386_ptob(ppn)); 3062} 3063 3064/* 3065 * pmap_ts_referenced: 3066 * 3067 * Return the count of reference bits for a page, clearing all of them. 3068 * 3069 */ 3070int 3071pmap_ts_referenced(vm_offset_t pa) 3072{ 3073 register pv_entry_t pv; 3074 pv_table_t *ppv; 3075 unsigned *pte; 3076 int s; 3077 int rtval = 0; 3078 3079 if (!pmap_is_managed(pa)) 3080 return FALSE; 3081 3082 s = splvm(); 3083 3084 ppv = pa_to_pvh(pa); 3085 3086 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 3087 splx(s); 3088 return 0; 3089 } 3090 3091 /* 3092 * Not found, check current mappings returning immediately if found. 3093 */ 3094 for (pv = TAILQ_FIRST(&ppv->pv_list); 3095 pv; 3096 pv = TAILQ_NEXT(pv, pv_list)) { 3097 /* 3098 * if the bit being tested is the modified bit, then 3099 * mark clean_map and ptes as never 3100 * modified. 3101 */ 3102 if (!pmap_track_modified(pv->pv_va)) 3103 continue; 3104 3105 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3106 if (pte == NULL) { 3107 continue; 3108 } 3109 if (*pte & PG_A) { 3110 rtval++; 3111 *pte &= ~PG_A; 3112 } 3113 } 3114 splx(s); 3115 if (rtval) { 3116 invltlb(); 3117 } 3118 return (rtval); 3119} 3120 3121/* 3122 * pmap_is_modified: 3123 * 3124 * Return whether or not the specified physical page was modified 3125 * in any physical maps. 3126 */ 3127boolean_t 3128pmap_is_modified(vm_offset_t pa) 3129{ 3130 return pmap_testbit((pa), PG_M); 3131} 3132 3133/* 3134 * Clear the modify bits on the specified physical page. 3135 */ 3136void 3137pmap_clear_modify(vm_offset_t pa) 3138{ 3139 pmap_changebit((pa), PG_M, FALSE); 3140} 3141 3142/* 3143 * pmap_clear_reference: 3144 * 3145 * Clear the reference bit on the specified physical page. 3146 */ 3147void 3148pmap_clear_reference(vm_offset_t pa) 3149{ 3150 pmap_changebit((pa), PG_A, FALSE); 3151} 3152 3153/* 3154 * Miscellaneous support routines follow 3155 */ 3156 3157static void 3158i386_protection_init() 3159{ 3160 register int *kp, prot; 3161 3162 kp = protection_codes; 3163 for (prot = 0; prot < 8; prot++) { 3164 switch (prot) { 3165 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3166 /* 3167 * Read access is also 0. There isn't any execute bit, 3168 * so just make it readable. 3169 */ 3170 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3171 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3172 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3173 *kp++ = 0; 3174 break; 3175 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3176 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3177 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3178 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3179 *kp++ = PG_RW; 3180 break; 3181 } 3182 } 3183} 3184 3185/* 3186 * Map a set of physical memory pages into the kernel virtual 3187 * address space. Return a pointer to where it is mapped. This 3188 * routine is intended to be used for mapping device memory, 3189 * NOT real memory. 3190 */ 3191void * 3192pmap_mapdev(pa, size) 3193 vm_offset_t pa; 3194 vm_size_t size; 3195{ 3196 vm_offset_t va, tmpva; 3197 unsigned *pte; 3198 3199 size = roundup(size, PAGE_SIZE); 3200 3201 va = kmem_alloc_pageable(kernel_map, size); 3202 if (!va) 3203 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3204 3205 pa = pa & PG_FRAME; 3206 for (tmpva = va; size > 0;) { 3207 pte = (unsigned *)vtopte(tmpva); 3208 *pte = pa | PG_RW | PG_V | pgeflag; 3209 size -= PAGE_SIZE; 3210 tmpva += PAGE_SIZE; 3211 pa += PAGE_SIZE; 3212 } 3213 invltlb(); 3214 3215 return ((void *) va); 3216} 3217 3218/* 3219 * perform the pmap work for mincore 3220 */ 3221int 3222pmap_mincore(pmap, addr) 3223 pmap_t pmap; 3224 vm_offset_t addr; 3225{ 3226 3227 unsigned *ptep, pte; 3228 int val = 0; 3229 3230 ptep = pmap_pte(pmap, addr); 3231 if (ptep == 0) { 3232 return 0; 3233 } 3234 3235 if (pte = *ptep) { 3236 vm_offset_t pa; 3237 val = MINCORE_INCORE; 3238 pa = pte & PG_FRAME; 3239 3240 /* 3241 * Modified by us 3242 */ 3243 if (pte & PG_M) 3244 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3245 /* 3246 * Modified by someone 3247 */ 3248 else if (PHYS_TO_VM_PAGE(pa)->dirty || 3249 pmap_is_modified(pa)) 3250 val |= MINCORE_MODIFIED_OTHER; 3251 /* 3252 * Referenced by us 3253 */ 3254 if (pte & PG_U) 3255 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3256 3257 /* 3258 * Referenced by someone 3259 */ 3260 else if ((PHYS_TO_VM_PAGE(pa)->flags & PG_REFERENCED) || 3261 pmap_ts_referenced(pa)) { 3262 val |= MINCORE_REFERENCED_OTHER; 3263 PHYS_TO_VM_PAGE(pa)->flags |= PG_REFERENCED; 3264 } 3265 } 3266 return val; 3267} 3268 3269void 3270pmap_activate(struct proc *p) 3271{ 3272#if defined(SWTCH_OPTIM_STATS) 3273 ++tlb_flush_count; 3274#endif 3275 load_cr3(p->p_addr->u_pcb.pcb_cr3 = 3276 vtophys(p->p_vmspace->vm_pmap.pm_pdir)); 3277} 3278 3279vm_offset_t 3280pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3281 3282 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3283 return addr; 3284 } 3285 3286 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3287 return addr; 3288} 3289 3290 3291#if defined(PMAP_DEBUG) 3292pmap_pid_dump(int pid) { 3293 pmap_t pmap; 3294 struct proc *p; 3295 int npte = 0; 3296 int index; 3297 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3298 if (p->p_pid != pid) 3299 continue; 3300 3301 if (p->p_vmspace) { 3302 int i,j; 3303 index = 0; 3304 pmap = &p->p_vmspace->vm_pmap; 3305 for(i=0;i<1024;i++) { 3306 pd_entry_t *pde; 3307 unsigned *pte; 3308 unsigned base = i << PDRSHIFT; 3309 3310 pde = &pmap->pm_pdir[i]; 3311 if (pde && pmap_pde_v(pde)) { 3312 for(j=0;j<1024;j++) { 3313 unsigned va = base + (j << PAGE_SHIFT); 3314 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3315 if (index) { 3316 index = 0; 3317 printf("\n"); 3318 } 3319 return npte; 3320 } 3321 pte = pmap_pte_quick( pmap, va); 3322 if (pte && pmap_pte_v(pte)) { 3323 vm_offset_t pa; 3324 vm_page_t m; 3325 pa = *(int *)pte; 3326 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3327 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3328 va, pa, m->hold_count, m->wire_count, m->flags); 3329 npte++; 3330 index++; 3331 if (index >= 2) { 3332 index = 0; 3333 printf("\n"); 3334 } else { 3335 printf(" "); 3336 } 3337 } 3338 } 3339 } 3340 } 3341 } 3342 } 3343 return npte; 3344} 3345#endif 3346 3347#if defined(DEBUG) 3348 3349static void pads __P((pmap_t pm)); 3350static void pmap_pvdump __P((vm_offset_t pa)); 3351 3352/* print address space of pmap*/ 3353static void 3354pads(pm) 3355 pmap_t pm; 3356{ 3357 unsigned va, i, j; 3358 unsigned *ptep; 3359 3360 if (pm == kernel_pmap) 3361 return; 3362 for (i = 0; i < 1024; i++) 3363 if (pm->pm_pdir[i]) 3364 for (j = 0; j < 1024; j++) { 3365 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3366 if (pm == kernel_pmap && va < KERNBASE) 3367 continue; 3368 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3369 continue; 3370 ptep = pmap_pte_quick(pm, va); 3371 if (pmap_pte_v(ptep)) 3372 printf("%x:%x ", va, *(int *) ptep); 3373 }; 3374 3375} 3376 3377static void 3378pmap_pvdump(pa) 3379 vm_offset_t pa; 3380{ 3381 pv_table_t *ppv; 3382 register pv_entry_t pv; 3383 3384 printf("pa %x", pa); 3385 ppv = pa_to_pvh(pa); 3386 for (pv = TAILQ_FIRST(&ppv->pv_list); 3387 pv; 3388 pv = TAILQ_NEXT(pv, pv_list)) { 3389#ifdef used_to_be 3390 printf(" -> pmap %x, va %x, flags %x", 3391 pv->pv_pmap, pv->pv_va, pv->pv_flags); 3392#endif 3393 printf(" -> pmap %x, va %x", 3394 pv->pv_pmap, pv->pv_va); 3395 pads(pv->pv_pmap); 3396 } 3397 printf(" "); 3398} 3399#endif 3400