pmap.c revision 86439
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 * $FreeBSD: head/sys/i386/i386/pmap.c 86439 2001-11-16 02:17:18Z peter $ 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_disable_pse.h" 72#include "opt_pmap.h" 73#include "opt_msgbuf.h" 74#include "opt_kstack_pages.h" 75 76#include <sys/param.h> 77#include <sys/systm.h> 78#include <sys/kernel.h> 79#include <sys/lock.h> 80#include <sys/mman.h> 81#include <sys/msgbuf.h> 82#include <sys/mutex.h> 83#include <sys/proc.h> 84#include <sys/sx.h> 85#include <sys/user.h> 86#include <sys/vmmeter.h> 87#include <sys/sysctl.h> 88 89#include <vm/vm.h> 90#include <vm/vm_param.h> 91#include <vm/vm_kern.h> 92#include <vm/vm_page.h> 93#include <vm/vm_map.h> 94#include <vm/vm_object.h> 95#include <vm/vm_extern.h> 96#include <vm/vm_pageout.h> 97#include <vm/vm_pager.h> 98#include <vm/vm_zone.h> 99 100#include <machine/cputypes.h> 101#include <machine/md_var.h> 102#include <machine/specialreg.h> 103#if defined(SMP) || defined(APIC_IO) 104#include <machine/smp.h> 105#include <machine/apic.h> 106#include <machine/segments.h> 107#include <machine/tss.h> 108#include <machine/globaldata.h> 109#endif /* SMP || APIC_IO */ 110 111#define PMAP_KEEP_PDIRS 112#ifndef PMAP_SHPGPERPROC 113#define PMAP_SHPGPERPROC 200 114#endif 115 116#if defined(DIAGNOSTIC) 117#define PMAP_DIAGNOSTIC 118#endif 119 120#define MINPV 2048 121 122#if !defined(PMAP_DIAGNOSTIC) 123#define PMAP_INLINE __inline 124#else 125#define PMAP_INLINE 126#endif 127 128/* 129 * Get PDEs and PTEs for user/kernel address space 130 */ 131#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 132#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 133 134#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 135#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 136#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 137#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 138#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 139 140#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 141#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 142 143/* 144 * Given a map and a machine independent protection code, 145 * convert to a vax protection code. 146 */ 147#define pte_prot(m, p) (protection_codes[p]) 148static int protection_codes[8]; 149 150static struct pmap kernel_pmap_store; 151pmap_t kernel_pmap; 152LIST_HEAD(pmaplist, pmap); 153struct pmaplist allpmaps; 154 155vm_offset_t avail_start; /* PA of first available physical page */ 156vm_offset_t avail_end; /* PA of last available physical page */ 157vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 158vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 159static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 160static int pgeflag; /* PG_G or-in */ 161static int pseflag; /* PG_PS or-in */ 162 163static vm_object_t kptobj; 164 165static int nkpt; 166vm_offset_t kernel_vm_end; 167 168/* 169 * Data for the pv entry allocation mechanism 170 */ 171static vm_zone_t pvzone; 172static struct vm_zone pvzone_store; 173static struct vm_object pvzone_obj; 174static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0; 175static int pmap_pagedaemon_waken = 0; 176static struct pv_entry *pvinit; 177 178/* 179 * All those kernel PT submaps that BSD is so fond of 180 */ 181pt_entry_t *CMAP1 = 0; 182static pt_entry_t *CMAP2, *ptmmap; 183caddr_t CADDR1 = 0, ptvmmap = 0; 184static caddr_t CADDR2; 185static pt_entry_t *msgbufmap; 186struct msgbuf *msgbufp = 0; 187 188/* 189 * Crashdump maps. 190 */ 191static pt_entry_t *pt_crashdumpmap; 192static caddr_t crashdumpmap; 193 194#ifdef SMP 195extern pt_entry_t *SMPpt; 196#endif 197static pt_entry_t *PMAP1 = 0; 198static unsigned *PADDR1 = 0; 199 200static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 201static unsigned * get_ptbase __P((pmap_t pmap)); 202static pv_entry_t get_pv_entry __P((void)); 203static void i386_protection_init __P((void)); 204static __inline void pmap_changebit __P((vm_page_t m, int bit, boolean_t setem)); 205 206static void pmap_remove_all __P((vm_page_t m)); 207static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 208 vm_page_t m, vm_page_t mpte)); 209static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 210 vm_offset_t sva)); 211static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 212static int pmap_remove_entry __P((struct pmap *pmap, vm_page_t m, 213 vm_offset_t va)); 214static boolean_t pmap_testbit __P((vm_page_t m, int bit)); 215static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 216 vm_page_t mpte, vm_page_t m)); 217 218static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 219 220static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 221static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 222static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 223static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 224static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 225static vm_offset_t pmap_kmem_choose(vm_offset_t addr); 226 227static unsigned pdir4mb; 228 229/* 230 * Routine: pmap_pte 231 * Function: 232 * Extract the page table entry associated 233 * with the given map/virtual_address pair. 234 */ 235 236PMAP_INLINE unsigned * 237pmap_pte(pmap, va) 238 register pmap_t pmap; 239 vm_offset_t va; 240{ 241 unsigned *pdeaddr; 242 243 if (pmap) { 244 pdeaddr = (unsigned *) pmap_pde(pmap, va); 245 if (*pdeaddr & PG_PS) 246 return pdeaddr; 247 if (*pdeaddr) { 248 return get_ptbase(pmap) + i386_btop(va); 249 } 250 } 251 return (0); 252} 253 254/* 255 * Move the kernel virtual free pointer to the next 256 * 4MB. This is used to help improve performance 257 * by using a large (4MB) page for much of the kernel 258 * (.text, .data, .bss) 259 */ 260static vm_offset_t 261pmap_kmem_choose(vm_offset_t addr) 262{ 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; 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 kernel_pmap->pm_count = 1; 320 kernel_pmap->pm_active = -1; /* don't allow deactivation */ 321 TAILQ_INIT(&kernel_pmap->pm_pvlist); 322 LIST_INIT(&allpmaps); 323 LIST_INSERT_HEAD(&allpmaps, kernel_pmap, pm_list); 324 nkpt = NKPT; 325 326 /* 327 * Reserve some special page table entries/VA space for temporary 328 * mapping of pages. 329 */ 330#define SYSMAP(c, p, v, n) \ 331 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 332 333 va = virtual_avail; 334 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 335 336 /* 337 * CMAP1/CMAP2 are used for zeroing and copying pages. 338 */ 339 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 340 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 341 342 /* 343 * Crashdump maps. 344 */ 345 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS); 346 347 /* 348 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 349 * XXX ptmmap is not used. 350 */ 351 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 352 353 /* 354 * msgbufp is used to map the system message buffer. 355 * XXX msgbufmap is not used. 356 */ 357 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 358 atop(round_page(MSGBUF_SIZE))) 359 360 /* 361 * ptemap is used for pmap_pte_quick 362 */ 363 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 364 365 virtual_avail = va; 366 367 *(int *) CMAP1 = *(int *) CMAP2 = 0; 368 for (i = 0; i < NKPT; i++) 369 PTD[i] = 0; 370 371 pgeflag = 0; 372#if !defined(SMP) /* XXX - see also mp_machdep.c */ 373 if (cpu_feature & CPUID_PGE) { 374 pgeflag = PG_G; 375 } 376#endif 377 378/* 379 * Initialize the 4MB page size flag 380 */ 381 pseflag = 0; 382/* 383 * The 4MB page version of the initial 384 * kernel page mapping. 385 */ 386 pdir4mb = 0; 387 388#if !defined(DISABLE_PSE) 389 if (cpu_feature & CPUID_PSE) { 390 unsigned ptditmp; 391 /* 392 * Note that we have enabled PSE mode 393 */ 394 pseflag = PG_PS; 395 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE)); 396 ptditmp &= ~(NBPDR - 1); 397 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 398 pdir4mb = ptditmp; 399 400#if !defined(SMP) 401 /* 402 * Enable the PSE mode. 403 */ 404 load_cr4(rcr4() | CR4_PSE); 405 406 /* 407 * We can do the mapping here for the single processor 408 * case. We simply ignore the old page table page from 409 * now on. 410 */ 411 /* 412 * For SMP, we still need 4K pages to bootstrap APs, 413 * PSE will be enabled as soon as all APs are up. 414 */ 415 PTD[KPTDI] = (pd_entry_t) ptditmp; 416 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 417 invltlb(); 418#endif 419 } 420#endif 421 422#ifdef SMP 423 if (cpu_apic_address == 0) 424 panic("pmap_bootstrap: no local apic! (non-SMP hardware?)"); 425 426 /* local apic is mapped on last page */ 427 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag | 428 (cpu_apic_address & PG_FRAME)); 429#endif 430 431 invltlb(); 432} 433 434#ifdef SMP 435/* 436 * Set 4mb pdir for mp startup 437 */ 438void 439pmap_set_opt(void) 440{ 441 if (pseflag && (cpu_feature & CPUID_PSE)) { 442 load_cr4(rcr4() | CR4_PSE); 443 if (pdir4mb && PCPU_GET(cpuid) == 0) { /* only on BSP */ 444 kernel_pmap->pm_pdir[KPTDI] = 445 PTD[KPTDI] = (pd_entry_t)pdir4mb; 446 cpu_invltlb(); 447 } 448 } 449} 450#endif 451 452/* 453 * Initialize the pmap module. 454 * Called by vm_init, to initialize any structures that the pmap 455 * system needs to map virtual memory. 456 * pmap_init has been enhanced to support in a fairly consistant 457 * way, discontiguous physical memory. 458 */ 459void 460pmap_init(phys_start, phys_end) 461 vm_offset_t phys_start, phys_end; 462{ 463 int i; 464 int initial_pvs; 465 466 /* 467 * object for kernel page table pages 468 */ 469 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE); 470 471 /* 472 * Allocate memory for random pmap data structures. Includes the 473 * pv_head_table. 474 */ 475 476 for(i = 0; i < vm_page_array_size; i++) { 477 vm_page_t m; 478 479 m = &vm_page_array[i]; 480 TAILQ_INIT(&m->md.pv_list); 481 m->md.pv_list_count = 0; 482 } 483 484 /* 485 * init the pv free list 486 */ 487 initial_pvs = vm_page_array_size; 488 if (initial_pvs < MINPV) 489 initial_pvs = MINPV; 490 pvzone = &pvzone_store; 491 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 492 initial_pvs * sizeof (struct pv_entry)); 493 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, 494 vm_page_array_size); 495 496 /* 497 * Now it is safe to enable pv_table recording. 498 */ 499 pmap_initialized = TRUE; 500} 501 502/* 503 * Initialize the address space (zone) for the pv_entries. Set a 504 * high water mark so that the system can recover from excessive 505 * numbers of pv entries. 506 */ 507void 508pmap_init2() 509{ 510 int shpgperproc = PMAP_SHPGPERPROC; 511 512 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc); 513 pv_entry_max = shpgperproc * maxproc + vm_page_array_size; 514 pv_entry_high_water = 9 * (pv_entry_max / 10); 515 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 516} 517 518 519/*************************************************** 520 * Low level helper routines..... 521 ***************************************************/ 522 523#if defined(PMAP_DIAGNOSTIC) 524 525/* 526 * This code checks for non-writeable/modified pages. 527 * This should be an invalid condition. 528 */ 529static int 530pmap_nw_modified(pt_entry_t ptea) 531{ 532 int pte; 533 534 pte = (int) ptea; 535 536 if ((pte & (PG_M|PG_RW)) == PG_M) 537 return 1; 538 else 539 return 0; 540} 541#endif 542 543 544/* 545 * this routine defines the region(s) of memory that should 546 * not be tested for the modified bit. 547 */ 548static PMAP_INLINE int 549pmap_track_modified(vm_offset_t va) 550{ 551 if ((va < kmi.clean_sva) || (va >= kmi.clean_eva)) 552 return 1; 553 else 554 return 0; 555} 556 557static PMAP_INLINE void 558invltlb_1pg(vm_offset_t va) 559{ 560#ifdef I386_CPU 561 invltlb(); 562#else 563 invlpg(va); 564#endif 565} 566 567static __inline void 568pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va) 569{ 570#if defined(SMP) 571 if (pmap->pm_active & (1 << PCPU_GET(cpuid))) 572 cpu_invlpg((void *)va); 573 if (pmap->pm_active & PCPU_GET(other_cpus)) 574 smp_invltlb(); 575#else 576 if (pmap->pm_active) 577 invltlb_1pg(va); 578#endif 579} 580 581static __inline void 582pmap_TLB_invalidate_all(pmap_t pmap) 583{ 584#if defined(SMP) 585 if (pmap->pm_active & (1 << PCPU_GET(cpuid))) 586 cpu_invltlb(); 587 if (pmap->pm_active & PCPU_GET(other_cpus)) 588 smp_invltlb(); 589#else 590 if (pmap->pm_active) 591 invltlb(); 592#endif 593} 594 595/* 596 * Return an address which is the base of the Virtual mapping of 597 * all the PTEs for the given pmap. Note this doesn't say that 598 * all the PTEs will be present or that the pages there are valid. 599 * The PTEs are made available by the recursive mapping trick. 600 * It will map in the alternate PTE space if needed. 601 */ 602static unsigned * 603get_ptbase(pmap) 604 pmap_t pmap; 605{ 606 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 607 608 /* are we current address space or kernel? */ 609 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 610 return (unsigned *) PTmap; 611 } 612 /* otherwise, we are alternate address space */ 613 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 614 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 615#if defined(SMP) 616 /* The page directory is not shared between CPUs */ 617 cpu_invltlb(); 618#else 619 invltlb(); 620#endif 621 } 622 return (unsigned *) APTmap; 623} 624 625/* 626 * Super fast pmap_pte routine best used when scanning 627 * the pv lists. This eliminates many coarse-grained 628 * invltlb calls. Note that many of the pv list 629 * scans are across different pmaps. It is very wasteful 630 * to do an entire invltlb for checking a single mapping. 631 */ 632 633static unsigned * 634pmap_pte_quick(pmap, va) 635 register pmap_t pmap; 636 vm_offset_t va; 637{ 638 unsigned pde, newpf; 639 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) { 640 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 641 unsigned index = i386_btop(va); 642 /* are we current address space or kernel? */ 643 if ((pmap == kernel_pmap) || 644 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 645 return (unsigned *) PTmap + index; 646 } 647 newpf = pde & PG_FRAME; 648 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 649 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 650 invltlb_1pg((vm_offset_t) PADDR1); 651 } 652 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 653 } 654 return (0); 655} 656 657/* 658 * Routine: pmap_extract 659 * Function: 660 * Extract the physical page address associated 661 * with the given map/virtual_address pair. 662 */ 663vm_offset_t 664pmap_extract(pmap, va) 665 register pmap_t pmap; 666 vm_offset_t va; 667{ 668 vm_offset_t rtval; 669 vm_offset_t pdirindex; 670 pdirindex = va >> PDRSHIFT; 671 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 672 unsigned *pte; 673 if ((rtval & PG_PS) != 0) { 674 rtval &= ~(NBPDR - 1); 675 rtval |= va & (NBPDR - 1); 676 return rtval; 677 } 678 pte = get_ptbase(pmap) + i386_btop(va); 679 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 680 return rtval; 681 } 682 return 0; 683 684} 685 686/*************************************************** 687 * Low level mapping routines..... 688 ***************************************************/ 689 690/* 691 * add a wired page to the kva 692 * note that in order for the mapping to take effect -- you 693 * should do a invltlb after doing the pmap_kenter... 694 */ 695PMAP_INLINE void 696pmap_kenter(va, pa) 697 vm_offset_t va; 698 register vm_offset_t pa; 699{ 700 register unsigned *pte; 701 unsigned npte, opte; 702 703 npte = pa | PG_RW | PG_V | pgeflag; 704 pte = (unsigned *)vtopte(va); 705 opte = *pte; 706 *pte = npte; 707 /*if (opte)*/ 708 invltlb_1pg(va); /* XXX what about SMP? */ 709} 710 711/* 712 * remove a page from the kernel pagetables 713 */ 714PMAP_INLINE void 715pmap_kremove(va) 716 vm_offset_t va; 717{ 718 register unsigned *pte; 719 720 pte = (unsigned *)vtopte(va); 721 *pte = 0; 722 invltlb_1pg(va); /* XXX what about SMP? */ 723} 724 725/* 726 * Used to map a range of physical addresses into kernel 727 * virtual address space. 728 * 729 * The value passed in '*virt' is a suggested virtual address for 730 * the mapping. Architectures which can support a direct-mapped 731 * physical to virtual region can return the appropriate address 732 * within that region, leaving '*virt' unchanged. Other 733 * architectures should map the pages starting at '*virt' and 734 * update '*virt' with the first usable address after the mapped 735 * region. 736 */ 737vm_offset_t 738pmap_map(virt, start, end, prot) 739 vm_offset_t *virt; 740 vm_offset_t start; 741 vm_offset_t end; 742 int prot; 743{ 744 vm_offset_t sva = *virt; 745 vm_offset_t va = sva; 746 while (start < end) { 747 pmap_kenter(va, start); 748 va += PAGE_SIZE; 749 start += PAGE_SIZE; 750 } 751 *virt = va; 752 return (sva); 753} 754 755 756/* 757 * Add a list of wired pages to the kva 758 * this routine is only used for temporary 759 * kernel mappings that do not need to have 760 * page modification or references recorded. 761 * Note that old mappings are simply written 762 * over. The page *must* be wired. 763 */ 764void 765pmap_qenter(va, m, count) 766 vm_offset_t va; 767 vm_page_t *m; 768 int count; 769{ 770 vm_offset_t end_va; 771 772 end_va = va + count * PAGE_SIZE; 773 774 while (va < end_va) { 775 unsigned *pte; 776 777 pte = (unsigned *)vtopte(va); 778 *pte = VM_PAGE_TO_PHYS(*m) | PG_RW | PG_V | pgeflag; 779#ifdef SMP 780 cpu_invlpg((void *)va); 781#else 782 invltlb_1pg(va); 783#endif 784 va += PAGE_SIZE; 785 m++; 786 } 787#ifdef SMP 788 smp_invltlb(); 789#endif 790} 791 792/* 793 * this routine jerks page mappings from the 794 * kernel -- it is meant only for temporary mappings. 795 */ 796void 797pmap_qremove(va, count) 798 vm_offset_t va; 799 int count; 800{ 801 vm_offset_t end_va; 802 803 end_va = va + count*PAGE_SIZE; 804 805 while (va < end_va) { 806 unsigned *pte; 807 808 pte = (unsigned *)vtopte(va); 809 *pte = 0; 810#ifdef SMP 811 cpu_invlpg((void *)va); 812#else 813 invltlb_1pg(va); 814#endif 815 va += PAGE_SIZE; 816 } 817#ifdef SMP 818 smp_invltlb(); 819#endif 820} 821 822static vm_page_t 823pmap_page_lookup(object, pindex) 824 vm_object_t object; 825 vm_pindex_t pindex; 826{ 827 vm_page_t m; 828retry: 829 m = vm_page_lookup(object, pindex); 830 if (m && vm_page_sleep_busy(m, FALSE, "pplookp")) 831 goto retry; 832 return m; 833} 834 835/* 836 * Create the Uarea stack for a new process. 837 * This routine directly affects the fork perf for a process. 838 */ 839void 840pmap_new_proc(p) 841 struct proc *p; 842{ 843#ifdef I386_CPU 844 int updateneeded = 0; 845#endif 846 int i; 847 vm_object_t upobj; 848 vm_offset_t up; 849 vm_page_t m; 850 unsigned *ptek, oldpte; 851 852 /* 853 * allocate object for the upages 854 */ 855 upobj = p->p_upages_obj; 856 if (upobj == NULL) { 857 upobj = vm_object_allocate(OBJT_DEFAULT, UAREA_PAGES); 858 p->p_upages_obj = upobj; 859 } 860 861 /* get a kernel virtual address for the U area for this proc */ 862 up = (vm_offset_t)p->p_uarea; 863 if (up == 0) { 864 up = kmem_alloc_nofault(kernel_map, UAREA_PAGES * PAGE_SIZE); 865 if (up == 0) 866 panic("pmap_new_proc: upage allocation failed"); 867 p->p_uarea = (struct user *)up; 868 } 869 870 ptek = (unsigned *)vtopte(up); 871 872 for (i = 0; i < UAREA_PAGES; i++) { 873 /* 874 * Get a kernel stack page 875 */ 876 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 877 878 /* 879 * Wire the page 880 */ 881 m->wire_count++; 882 cnt.v_wire_count++; 883 884 oldpte = *(ptek + i); 885 /* 886 * Enter the page into the kernel address space. 887 */ 888 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 889 if (oldpte) { 890#ifdef I386_CPU 891 updateneeded = 1; 892#else 893 invlpg(up + i * PAGE_SIZE); 894#endif 895 } 896 897 vm_page_wakeup(m); 898 vm_page_flag_clear(m, PG_ZERO); 899 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 900 m->valid = VM_PAGE_BITS_ALL; 901 } 902#ifdef I386_CPU 903 if (updateneeded) 904 invltlb(); 905#endif 906} 907 908/* 909 * Dispose the U-Area for a process that has exited. 910 * This routine directly impacts the exit perf of a process. 911 */ 912void 913pmap_dispose_proc(p) 914 struct proc *p; 915{ 916 int i; 917 vm_object_t upobj; 918 vm_offset_t up; 919 vm_page_t m; 920 unsigned *ptek, oldpte; 921 922 upobj = p->p_upages_obj; 923 up = (vm_offset_t)p->p_uarea; 924 ptek = (unsigned *)vtopte(up); 925 for (i = 0; i < UAREA_PAGES; i++) { 926 m = vm_page_lookup(upobj, i); 927 if (m == NULL) 928 panic("pmap_dispose_proc: upage already missing?"); 929 vm_page_busy(m); 930 oldpte = *(ptek + i); 931 *(ptek + i) = 0; 932#ifndef I386_CPU 933 invlpg(up + i * PAGE_SIZE); 934#endif 935 vm_page_unwire(m, 0); 936 vm_page_free(m); 937 } 938#ifdef I386_CPU 939 invltlb(); 940#endif 941} 942 943/* 944 * Allow the U_AREA for a process to be prejudicially paged out. 945 */ 946void 947pmap_swapout_proc(p) 948 struct proc *p; 949{ 950 int i; 951 vm_object_t upobj; 952 vm_offset_t up; 953 vm_page_t m; 954 955 upobj = p->p_upages_obj; 956 up = (vm_offset_t)p->p_uarea; 957 for (i = 0; i < UAREA_PAGES; i++) { 958 m = vm_page_lookup(upobj, i); 959 if (m == NULL) 960 panic("pmap_swapout_proc: upage already missing?"); 961 vm_page_dirty(m); 962 vm_page_unwire(m, 0); 963 pmap_kremove(up + i * PAGE_SIZE); 964 } 965} 966 967/* 968 * Bring the U-Area for a specified process back in. 969 */ 970void 971pmap_swapin_proc(p) 972 struct proc *p; 973{ 974 int i,rv; 975 vm_object_t upobj; 976 vm_offset_t up; 977 vm_page_t m; 978 979 upobj = p->p_upages_obj; 980 up = (vm_offset_t)p->p_uarea; 981 for (i = 0; i < UAREA_PAGES; i++) { 982 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 983 pmap_kenter(up + i * PAGE_SIZE, VM_PAGE_TO_PHYS(m)); 984 if (m->valid != VM_PAGE_BITS_ALL) { 985 rv = vm_pager_get_pages(upobj, &m, 1, 0); 986 if (rv != VM_PAGER_OK) 987 panic("pmap_swapin_proc: cannot get upage for proc: %d\n", p->p_pid); 988 m = vm_page_lookup(upobj, i); 989 m->valid = VM_PAGE_BITS_ALL; 990 } 991 vm_page_wire(m); 992 vm_page_wakeup(m); 993 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 994 } 995} 996 997/* 998 * Create the kernel stack (including pcb for i386) for a new thread. 999 * This routine directly affects the fork perf for a process and 1000 * create performance for a thread. 1001 */ 1002void 1003pmap_new_thread(td) 1004 struct thread *td; 1005{ 1006#ifdef I386_CPU 1007 int updateneeded = 0; 1008#endif 1009 int i; 1010 vm_object_t ksobj; 1011 vm_page_t m; 1012 vm_offset_t ks; 1013 unsigned *ptek, oldpte; 1014 1015 /* 1016 * allocate object for the kstack 1017 */ 1018 ksobj = td->td_kstack_obj; 1019 if (ksobj == NULL) { 1020 ksobj = vm_object_allocate(OBJT_DEFAULT, KSTACK_PAGES); 1021 td->td_kstack_obj = ksobj; 1022 } 1023 1024#ifdef KSTACK_GUARD 1025 /* get a kernel virtual address for the kstack for this proc */ 1026 ks = (vm_offset_t)td->td_kstack; 1027 if (ks == 0) { 1028 ks = kmem_alloc_nofault(kernel_map, 1029 (KSTACK_PAGES + 1) * PAGE_SIZE); 1030 if (ks == 0) 1031 panic("pmap_new_thread: kstack allocation failed"); 1032 ks += PAGE_SIZE; 1033 td->td_kstack = ks; 1034 } 1035 1036 ptek = (unsigned *)vtopte(ks - PAGE_SIZE); 1037 oldpte = *ptek; 1038 *ptek = 0; 1039 if (oldpte) { 1040#ifdef I386_CPU 1041 updateneeded = 1; 1042#else 1043 invlpg(ks - PAGE_SIZE); 1044#endif 1045 } 1046 ptek++; 1047#else 1048 /* get a kernel virtual address for the kstack for this proc */ 1049 ks = (vm_offset_t)td->td_kstack; 1050 if (ks == 0) { 1051 ks = kmem_alloc_nofault(kernel_map, KSTACK_PAGES * PAGE_SIZE); 1052 if (ks == 0) 1053 panic("pmap_new_thread: kstack allocation failed"); 1054 td->td_kstack = ks; 1055 } 1056#endif 1057 for (i = 0; i < KSTACK_PAGES; i++) { 1058 /* 1059 * Get a kernel stack page 1060 */ 1061 m = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1062 1063 /* 1064 * Wire the page 1065 */ 1066 m->wire_count++; 1067 cnt.v_wire_count++; 1068 1069 oldpte = *(ptek + i); 1070 /* 1071 * Enter the page into the kernel address space. 1072 */ 1073 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 1074 if (oldpte) { 1075#ifdef I386_CPU 1076 updateneeded = 1; 1077#else 1078 invlpg(ks + i * PAGE_SIZE); 1079#endif 1080 } 1081 1082 vm_page_wakeup(m); 1083 vm_page_flag_clear(m, PG_ZERO); 1084 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 1085 m->valid = VM_PAGE_BITS_ALL; 1086 } 1087#ifdef I386_CPU 1088 if (updateneeded) 1089 invltlb(); 1090#endif 1091} 1092 1093/* 1094 * Dispose the kernel stack for a thread that has exited. 1095 * This routine directly impacts the exit perf of a process and thread. 1096 */ 1097void 1098pmap_dispose_thread(td) 1099 struct thread *td; 1100{ 1101 int i; 1102 vm_object_t ksobj; 1103 vm_offset_t ks; 1104 vm_page_t m; 1105 unsigned *ptek, oldpte; 1106 1107 ksobj = td->td_kstack_obj; 1108 ks = td->td_kstack; 1109 ptek = (unsigned *)vtopte(ks); 1110 for (i = 0; i < KSTACK_PAGES; i++) { 1111 m = vm_page_lookup(ksobj, i); 1112 if (m == NULL) 1113 panic("pmap_dispose_thread: kstack already missing?"); 1114 vm_page_busy(m); 1115 oldpte = *(ptek + i); 1116 *(ptek + i) = 0; 1117#ifndef I386_CPU 1118 invlpg(ks + i * PAGE_SIZE); 1119#endif 1120 vm_page_unwire(m, 0); 1121 vm_page_free(m); 1122 } 1123#ifdef I386_CPU 1124 invltlb(); 1125#endif 1126} 1127 1128/* 1129 * Allow the Kernel stack for a thread to be prejudicially paged out. 1130 */ 1131void 1132pmap_swapout_thread(td) 1133 struct thread *td; 1134{ 1135 int i; 1136 vm_object_t ksobj; 1137 vm_offset_t ks; 1138 vm_page_t m; 1139 1140 ksobj = td->td_kstack_obj; 1141 ks = td->td_kstack; 1142 for (i = 0; i < KSTACK_PAGES; i++) { 1143 m = vm_page_lookup(ksobj, i); 1144 if (m == NULL) 1145 panic("pmap_swapout_thread: kstack already missing?"); 1146 vm_page_dirty(m); 1147 vm_page_unwire(m, 0); 1148 pmap_kremove(ks + i * PAGE_SIZE); 1149 } 1150} 1151 1152/* 1153 * Bring the kernel stack for a specified thread back in. 1154 */ 1155void 1156pmap_swapin_thread(td) 1157 struct thread *td; 1158{ 1159 int i, rv; 1160 vm_object_t ksobj; 1161 vm_offset_t ks; 1162 vm_page_t m; 1163 1164 ksobj = td->td_kstack_obj; 1165 ks = td->td_kstack; 1166 for (i = 0; i < KSTACK_PAGES; i++) { 1167 m = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1168 pmap_kenter(ks + i * PAGE_SIZE, VM_PAGE_TO_PHYS(m)); 1169 if (m->valid != VM_PAGE_BITS_ALL) { 1170 rv = vm_pager_get_pages(ksobj, &m, 1, 0); 1171 if (rv != VM_PAGER_OK) 1172 panic("pmap_swapin_thread: cannot get kstack for proc: %d\n", td->td_proc->p_pid); 1173 m = vm_page_lookup(ksobj, i); 1174 m->valid = VM_PAGE_BITS_ALL; 1175 } 1176 vm_page_wire(m); 1177 vm_page_wakeup(m); 1178 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 1179 } 1180} 1181 1182/*************************************************** 1183 * Page table page management routines..... 1184 ***************************************************/ 1185 1186/* 1187 * This routine unholds page table pages, and if the hold count 1188 * drops to zero, then it decrements the wire count. 1189 */ 1190static int 1191_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1192 1193 while (vm_page_sleep_busy(m, FALSE, "pmuwpt")) 1194 ; 1195 1196 if (m->hold_count == 0) { 1197 vm_offset_t pteva; 1198 /* 1199 * unmap the page table page 1200 */ 1201 pmap->pm_pdir[m->pindex] = 0; 1202 --pmap->pm_stats.resident_count; 1203 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1204 (((unsigned) PTDpde) & PG_FRAME)) { 1205 /* 1206 * Do a invltlb to make the invalidated mapping 1207 * take effect immediately. 1208 */ 1209 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1210 pmap_TLB_invalidate(pmap, pteva); 1211 } 1212 1213 if (pmap->pm_ptphint == m) 1214 pmap->pm_ptphint = NULL; 1215 1216 /* 1217 * If the page is finally unwired, simply free it. 1218 */ 1219 --m->wire_count; 1220 if (m->wire_count == 0) { 1221 1222 vm_page_flash(m); 1223 vm_page_busy(m); 1224 vm_page_free_zero(m); 1225 --cnt.v_wire_count; 1226 } 1227 return 1; 1228 } 1229 return 0; 1230} 1231 1232static PMAP_INLINE int 1233pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) 1234{ 1235 vm_page_unhold(m); 1236 if (m->hold_count == 0) 1237 return _pmap_unwire_pte_hold(pmap, m); 1238 else 1239 return 0; 1240} 1241 1242/* 1243 * After removing a page table entry, this routine is used to 1244 * conditionally free the page, and manage the hold/wire counts. 1245 */ 1246static int 1247pmap_unuse_pt(pmap, va, mpte) 1248 pmap_t pmap; 1249 vm_offset_t va; 1250 vm_page_t mpte; 1251{ 1252 unsigned ptepindex; 1253 if (va >= UPT_MIN_ADDRESS) 1254 return 0; 1255 1256 if (mpte == NULL) { 1257 ptepindex = (va >> PDRSHIFT); 1258 if (pmap->pm_ptphint && 1259 (pmap->pm_ptphint->pindex == ptepindex)) { 1260 mpte = pmap->pm_ptphint; 1261 } else { 1262 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1263 pmap->pm_ptphint = mpte; 1264 } 1265 } 1266 1267 return pmap_unwire_pte_hold(pmap, mpte); 1268} 1269 1270void 1271pmap_pinit0(pmap) 1272 struct pmap *pmap; 1273{ 1274 pmap->pm_pdir = 1275 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1276 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1277 pmap->pm_count = 1; 1278 pmap->pm_active = 0; 1279 pmap->pm_ptphint = NULL; 1280 TAILQ_INIT(&pmap->pm_pvlist); 1281 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1282 LIST_INSERT_HEAD(&allpmaps, pmap, pm_list); 1283} 1284 1285/* 1286 * Initialize a preallocated and zeroed pmap structure, 1287 * such as one in a vmspace structure. 1288 */ 1289void 1290pmap_pinit(pmap) 1291 register struct pmap *pmap; 1292{ 1293 vm_page_t ptdpg; 1294 1295 /* 1296 * No need to allocate page table space yet but we do need a valid 1297 * page directory table. 1298 */ 1299 if (pmap->pm_pdir == NULL) 1300 pmap->pm_pdir = 1301 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1302 1303 /* 1304 * allocate object for the ptes 1305 */ 1306 if (pmap->pm_pteobj == NULL) 1307 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1308 1309 /* 1310 * allocate the page directory page 1311 */ 1312 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1313 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1314 1315 ptdpg->wire_count = 1; 1316 ++cnt.v_wire_count; 1317 1318 1319 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/ 1320 ptdpg->valid = VM_PAGE_BITS_ALL; 1321 1322 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1323 if ((ptdpg->flags & PG_ZERO) == 0) 1324 bzero(pmap->pm_pdir, PAGE_SIZE); 1325 1326 LIST_INSERT_HEAD(&allpmaps, pmap, pm_list); 1327 /* Wire in kernel global address entries. */ 1328 /* XXX copies current process, does not fill in MPPTDI */ 1329 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1330#ifdef SMP 1331 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI]; 1332#endif 1333 1334 /* install self-referential address mapping entry */ 1335 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1336 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1337 1338 pmap->pm_count = 1; 1339 pmap->pm_active = 0; 1340 pmap->pm_ptphint = NULL; 1341 TAILQ_INIT(&pmap->pm_pvlist); 1342 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1343} 1344 1345/* 1346 * Wire in kernel global address entries. To avoid a race condition 1347 * between pmap initialization and pmap_growkernel, this procedure 1348 * should be called after the vmspace is attached to the process 1349 * but before this pmap is activated. 1350 */ 1351void 1352pmap_pinit2(pmap) 1353 struct pmap *pmap; 1354{ 1355 /* XXX: Remove this stub when no longer called */ 1356} 1357 1358static int 1359pmap_release_free_page(pmap, p) 1360 struct pmap *pmap; 1361 vm_page_t p; 1362{ 1363 unsigned *pde = (unsigned *) pmap->pm_pdir; 1364 /* 1365 * This code optimizes the case of freeing non-busy 1366 * page-table pages. Those pages are zero now, and 1367 * might as well be placed directly into the zero queue. 1368 */ 1369 if (vm_page_sleep_busy(p, FALSE, "pmaprl")) 1370 return 0; 1371 1372 vm_page_busy(p); 1373 1374 /* 1375 * Remove the page table page from the processes address space. 1376 */ 1377 pde[p->pindex] = 0; 1378 pmap->pm_stats.resident_count--; 1379 1380 if (p->hold_count) { 1381 panic("pmap_release: freeing held page table page"); 1382 } 1383 /* 1384 * Page directory pages need to have the kernel 1385 * stuff cleared, so they can go into the zero queue also. 1386 */ 1387 if (p->pindex == PTDPTDI) { 1388 bzero(pde + KPTDI, nkpt * PTESIZE); 1389#ifdef SMP 1390 pde[MPPTDI] = 0; 1391#endif 1392 pde[APTDPTDI] = 0; 1393 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1394 } 1395 1396 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1397 pmap->pm_ptphint = NULL; 1398 1399 p->wire_count--; 1400 cnt.v_wire_count--; 1401 vm_page_free_zero(p); 1402 return 1; 1403} 1404 1405/* 1406 * this routine is called if the page table page is not 1407 * mapped correctly. 1408 */ 1409static vm_page_t 1410_pmap_allocpte(pmap, ptepindex) 1411 pmap_t pmap; 1412 unsigned ptepindex; 1413{ 1414 vm_offset_t pteva, ptepa; 1415 vm_page_t m; 1416 1417 /* 1418 * Find or fabricate a new pagetable page 1419 */ 1420 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1421 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1422 1423 KASSERT(m->queue == PQ_NONE, 1424 ("_pmap_allocpte: %p->queue != PQ_NONE", m)); 1425 1426 if (m->wire_count == 0) 1427 cnt.v_wire_count++; 1428 m->wire_count++; 1429 1430 /* 1431 * Increment the hold count for the page table page 1432 * (denoting a new mapping.) 1433 */ 1434 m->hold_count++; 1435 1436 /* 1437 * Map the pagetable page into the process address space, if 1438 * it isn't already there. 1439 */ 1440 1441 pmap->pm_stats.resident_count++; 1442 1443 ptepa = VM_PAGE_TO_PHYS(m); 1444 pmap->pm_pdir[ptepindex] = 1445 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1446 1447 /* 1448 * Set the page table hint 1449 */ 1450 pmap->pm_ptphint = m; 1451 1452 /* 1453 * Try to use the new mapping, but if we cannot, then 1454 * do it with the routine that maps the page explicitly. 1455 */ 1456 if ((m->flags & PG_ZERO) == 0) { 1457 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1458 (((unsigned) PTDpde) & PG_FRAME)) { 1459 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1460 bzero((caddr_t) pteva, PAGE_SIZE); 1461 } else { 1462 pmap_zero_page(ptepa); 1463 } 1464 } 1465 1466 m->valid = VM_PAGE_BITS_ALL; 1467 vm_page_flag_clear(m, PG_ZERO); 1468 vm_page_flag_set(m, PG_MAPPED); 1469 vm_page_wakeup(m); 1470 1471 return m; 1472} 1473 1474static vm_page_t 1475pmap_allocpte(pmap, va) 1476 pmap_t pmap; 1477 vm_offset_t va; 1478{ 1479 unsigned ptepindex; 1480 vm_offset_t ptepa; 1481 vm_page_t m; 1482 1483 /* 1484 * Calculate pagetable page index 1485 */ 1486 ptepindex = va >> PDRSHIFT; 1487 1488 /* 1489 * Get the page directory entry 1490 */ 1491 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1492 1493 /* 1494 * This supports switching from a 4MB page to a 1495 * normal 4K page. 1496 */ 1497 if (ptepa & PG_PS) { 1498 pmap->pm_pdir[ptepindex] = 0; 1499 ptepa = 0; 1500 invltlb(); 1501 } 1502 1503 /* 1504 * If the page table page is mapped, we just increment the 1505 * hold count, and activate it. 1506 */ 1507 if (ptepa) { 1508 /* 1509 * In order to get the page table page, try the 1510 * hint first. 1511 */ 1512 if (pmap->pm_ptphint && 1513 (pmap->pm_ptphint->pindex == ptepindex)) { 1514 m = pmap->pm_ptphint; 1515 } else { 1516 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1517 pmap->pm_ptphint = m; 1518 } 1519 m->hold_count++; 1520 return m; 1521 } 1522 /* 1523 * Here if the pte page isn't mapped, or if it has been deallocated. 1524 */ 1525 return _pmap_allocpte(pmap, ptepindex); 1526} 1527 1528 1529/*************************************************** 1530* Pmap allocation/deallocation routines. 1531 ***************************************************/ 1532 1533/* 1534 * Release any resources held by the given physical map. 1535 * Called when a pmap initialized by pmap_pinit is being released. 1536 * Should only be called if the map contains no valid mappings. 1537 */ 1538void 1539pmap_release(pmap) 1540 register struct pmap *pmap; 1541{ 1542 vm_page_t p,n,ptdpg; 1543 vm_object_t object = pmap->pm_pteobj; 1544 int curgeneration; 1545 1546#if defined(DIAGNOSTIC) 1547 if (object->ref_count != 1) 1548 panic("pmap_release: pteobj reference count != 1"); 1549#endif 1550 1551 ptdpg = NULL; 1552 LIST_REMOVE(pmap, pm_list); 1553retry: 1554 curgeneration = object->generation; 1555 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1556 n = TAILQ_NEXT(p, listq); 1557 if (p->pindex == PTDPTDI) { 1558 ptdpg = p; 1559 continue; 1560 } 1561 while (1) { 1562 if (!pmap_release_free_page(pmap, p) && 1563 (object->generation != curgeneration)) 1564 goto retry; 1565 } 1566 } 1567 1568 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1569 goto retry; 1570} 1571 1572static int 1573kvm_size(SYSCTL_HANDLER_ARGS) 1574{ 1575 unsigned long ksize = VM_MAX_KERNEL_ADDRESS - KERNBASE; 1576 1577 return sysctl_handle_long(oidp, &ksize, 0, req); 1578} 1579SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD, 1580 0, 0, kvm_size, "IU", "Size of KVM"); 1581 1582static int 1583kvm_free(SYSCTL_HANDLER_ARGS) 1584{ 1585 unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end; 1586 1587 return sysctl_handle_long(oidp, &kfree, 0, req); 1588} 1589SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD, 1590 0, 0, kvm_free, "IU", "Amount of KVM free"); 1591 1592/* 1593 * grow the number of kernel page table entries, if needed 1594 */ 1595void 1596pmap_growkernel(vm_offset_t addr) 1597{ 1598 struct pmap *pmap; 1599 int s; 1600 vm_offset_t ptppaddr; 1601 vm_page_t nkpg; 1602 pd_entry_t newpdir; 1603 1604 s = splhigh(); 1605 if (kernel_vm_end == 0) { 1606 kernel_vm_end = KERNBASE; 1607 nkpt = 0; 1608 while (pdir_pde(PTD, kernel_vm_end)) { 1609 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1610 nkpt++; 1611 } 1612 } 1613 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1614 while (kernel_vm_end < addr) { 1615 if (pdir_pde(PTD, kernel_vm_end)) { 1616 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1617 continue; 1618 } 1619 1620 /* 1621 * This index is bogus, but out of the way 1622 */ 1623 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1624 if (!nkpg) 1625 panic("pmap_growkernel: no memory to grow kernel"); 1626 1627 nkpt++; 1628 1629 vm_page_wire(nkpg); 1630 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1631 pmap_zero_page(ptppaddr); 1632 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1633 pdir_pde(PTD, kernel_vm_end) = newpdir; 1634 1635 LIST_FOREACH(pmap, &allpmaps, pm_list) { 1636 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1637 } 1638 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1639 } 1640 splx(s); 1641} 1642 1643/* 1644 * Retire the given physical map from service. 1645 * Should only be called if the map contains 1646 * no valid mappings. 1647 */ 1648void 1649pmap_destroy(pmap) 1650 register pmap_t pmap; 1651{ 1652 int count; 1653 1654 if (pmap == NULL) 1655 return; 1656 1657 count = --pmap->pm_count; 1658 if (count == 0) { 1659 pmap_release(pmap); 1660 panic("destroying a pmap is not yet implemented"); 1661 } 1662} 1663 1664/* 1665 * Add a reference to the specified pmap. 1666 */ 1667void 1668pmap_reference(pmap) 1669 pmap_t pmap; 1670{ 1671 if (pmap != NULL) { 1672 pmap->pm_count++; 1673 } 1674} 1675 1676/*************************************************** 1677* page management routines. 1678 ***************************************************/ 1679 1680/* 1681 * free the pv_entry back to the free list 1682 */ 1683static PMAP_INLINE void 1684free_pv_entry(pv) 1685 pv_entry_t pv; 1686{ 1687 pv_entry_count--; 1688 zfree(pvzone, pv); 1689} 1690 1691/* 1692 * get a new pv_entry, allocating a block from the system 1693 * when needed. 1694 * the memory allocation is performed bypassing the malloc code 1695 * because of the possibility of allocations at interrupt time. 1696 */ 1697static pv_entry_t 1698get_pv_entry(void) 1699{ 1700 pv_entry_count++; 1701 if (pv_entry_high_water && 1702 (pv_entry_count > pv_entry_high_water) && 1703 (pmap_pagedaemon_waken == 0)) { 1704 pmap_pagedaemon_waken = 1; 1705 wakeup (&vm_pages_needed); 1706 } 1707 return zalloc(pvzone); 1708} 1709 1710/* 1711 * This routine is very drastic, but can save the system 1712 * in a pinch. 1713 */ 1714void 1715pmap_collect() 1716{ 1717 int i; 1718 vm_page_t m; 1719 static int warningdone = 0; 1720 1721 if (pmap_pagedaemon_waken == 0) 1722 return; 1723 1724 if (warningdone < 5) { 1725 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1726 warningdone++; 1727 } 1728 1729 for(i = 0; i < vm_page_array_size; i++) { 1730 m = &vm_page_array[i]; 1731 if (m->wire_count || m->hold_count || m->busy || 1732 (m->flags & (PG_BUSY | PG_UNMANAGED))) 1733 continue; 1734 pmap_remove_all(m); 1735 } 1736 pmap_pagedaemon_waken = 0; 1737} 1738 1739 1740/* 1741 * If it is the first entry on the list, it is actually 1742 * in the header and we must copy the following entry up 1743 * to the header. Otherwise we must search the list for 1744 * the entry. In either case we free the now unused entry. 1745 */ 1746 1747static int 1748pmap_remove_entry(pmap, m, va) 1749 struct pmap *pmap; 1750 vm_page_t m; 1751 vm_offset_t va; 1752{ 1753 pv_entry_t pv; 1754 int rtval; 1755 int s; 1756 1757 s = splvm(); 1758 if (m->md.pv_list_count < pmap->pm_stats.resident_count) { 1759 for (pv = TAILQ_FIRST(&m->md.pv_list); 1760 pv; 1761 pv = TAILQ_NEXT(pv, pv_list)) { 1762 if (pmap == pv->pv_pmap && va == pv->pv_va) 1763 break; 1764 } 1765 } else { 1766 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1767 pv; 1768 pv = TAILQ_NEXT(pv, pv_plist)) { 1769 if (va == pv->pv_va) 1770 break; 1771 } 1772 } 1773 1774 rtval = 0; 1775 if (pv) { 1776 1777 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1778 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 1779 m->md.pv_list_count--; 1780 if (TAILQ_FIRST(&m->md.pv_list) == NULL) 1781 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 1782 1783 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1784 free_pv_entry(pv); 1785 } 1786 1787 splx(s); 1788 return rtval; 1789} 1790 1791/* 1792 * Create a pv entry for page at pa for 1793 * (pmap, va). 1794 */ 1795static void 1796pmap_insert_entry(pmap, va, mpte, m) 1797 pmap_t pmap; 1798 vm_offset_t va; 1799 vm_page_t mpte; 1800 vm_page_t m; 1801{ 1802 1803 int s; 1804 pv_entry_t pv; 1805 1806 s = splvm(); 1807 pv = get_pv_entry(); 1808 pv->pv_va = va; 1809 pv->pv_pmap = pmap; 1810 pv->pv_ptem = mpte; 1811 1812 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1813 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); 1814 m->md.pv_list_count++; 1815 1816 splx(s); 1817} 1818 1819/* 1820 * pmap_remove_pte: do the things to unmap a page in a process 1821 */ 1822static int 1823pmap_remove_pte(pmap, ptq, va) 1824 struct pmap *pmap; 1825 unsigned *ptq; 1826 vm_offset_t va; 1827{ 1828 unsigned oldpte; 1829 vm_page_t m; 1830 1831 oldpte = atomic_readandclear_int(ptq); 1832 if (oldpte & PG_W) 1833 pmap->pm_stats.wired_count -= 1; 1834 /* 1835 * Machines that don't support invlpg, also don't support 1836 * PG_G. 1837 */ 1838 if (oldpte & PG_G) 1839 invlpg(va); 1840 pmap->pm_stats.resident_count -= 1; 1841 if (oldpte & PG_MANAGED) { 1842 m = PHYS_TO_VM_PAGE(oldpte); 1843 if (oldpte & PG_M) { 1844#if defined(PMAP_DIAGNOSTIC) 1845 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1846 printf( 1847 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n", 1848 va, oldpte); 1849 } 1850#endif 1851 if (pmap_track_modified(va)) 1852 vm_page_dirty(m); 1853 } 1854 if (oldpte & PG_A) 1855 vm_page_flag_set(m, PG_REFERENCED); 1856 return pmap_remove_entry(pmap, m, va); 1857 } else { 1858 return pmap_unuse_pt(pmap, va, NULL); 1859 } 1860 1861 return 0; 1862} 1863 1864/* 1865 * Remove a single page from a process address space 1866 */ 1867static void 1868pmap_remove_page(pmap, va) 1869 struct pmap *pmap; 1870 register vm_offset_t va; 1871{ 1872 register unsigned *ptq; 1873 1874 /* 1875 * if there is no pte for this address, just skip it!!! 1876 */ 1877 if (*pmap_pde(pmap, va) == 0) { 1878 return; 1879 } 1880 1881 /* 1882 * get a local va for mappings for this pmap. 1883 */ 1884 ptq = get_ptbase(pmap) + i386_btop(va); 1885 if (*ptq) { 1886 (void) pmap_remove_pte(pmap, ptq, va); 1887 pmap_TLB_invalidate(pmap, va); 1888 } 1889 return; 1890} 1891 1892/* 1893 * Remove the given range of addresses from the specified map. 1894 * 1895 * It is assumed that the start and end are properly 1896 * rounded to the page size. 1897 */ 1898void 1899pmap_remove(pmap, sva, eva) 1900 struct pmap *pmap; 1901 register vm_offset_t sva; 1902 register vm_offset_t eva; 1903{ 1904 register unsigned *ptbase; 1905 vm_offset_t pdnxt; 1906 vm_offset_t ptpaddr; 1907 vm_offset_t sindex, eindex; 1908 int anyvalid; 1909 1910 if (pmap == NULL) 1911 return; 1912 1913 if (pmap->pm_stats.resident_count == 0) 1914 return; 1915 1916 /* 1917 * special handling of removing one page. a very 1918 * common operation and easy to short circuit some 1919 * code. 1920 */ 1921 if (((sva + PAGE_SIZE) == eva) && 1922 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1923 pmap_remove_page(pmap, sva); 1924 return; 1925 } 1926 1927 anyvalid = 0; 1928 1929 /* 1930 * Get a local virtual address for the mappings that are being 1931 * worked with. 1932 */ 1933 ptbase = get_ptbase(pmap); 1934 1935 sindex = i386_btop(sva); 1936 eindex = i386_btop(eva); 1937 1938 for (; sindex < eindex; sindex = pdnxt) { 1939 unsigned pdirindex; 1940 1941 /* 1942 * Calculate index for next page table. 1943 */ 1944 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1945 if (pmap->pm_stats.resident_count == 0) 1946 break; 1947 1948 pdirindex = sindex / NPDEPG; 1949 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1950 pmap->pm_pdir[pdirindex] = 0; 1951 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1952 anyvalid++; 1953 continue; 1954 } 1955 1956 /* 1957 * Weed out invalid mappings. Note: we assume that the page 1958 * directory table is always allocated, and in kernel virtual. 1959 */ 1960 if (ptpaddr == 0) 1961 continue; 1962 1963 /* 1964 * Limit our scan to either the end of the va represented 1965 * by the current page table page, or to the end of the 1966 * range being removed. 1967 */ 1968 if (pdnxt > eindex) { 1969 pdnxt = eindex; 1970 } 1971 1972 for ( ;sindex != pdnxt; sindex++) { 1973 vm_offset_t va; 1974 if (ptbase[sindex] == 0) { 1975 continue; 1976 } 1977 va = i386_ptob(sindex); 1978 1979 anyvalid++; 1980 if (pmap_remove_pte(pmap, 1981 ptbase + sindex, va)) 1982 break; 1983 } 1984 } 1985 1986 if (anyvalid) 1987 pmap_TLB_invalidate_all(pmap); 1988} 1989 1990/* 1991 * Routine: pmap_remove_all 1992 * Function: 1993 * Removes this physical page from 1994 * all physical maps in which it resides. 1995 * Reflects back modify bits to the pager. 1996 * 1997 * Notes: 1998 * Original versions of this routine were very 1999 * inefficient because they iteratively called 2000 * pmap_remove (slow...) 2001 */ 2002 2003static void 2004pmap_remove_all(m) 2005 vm_page_t m; 2006{ 2007 register pv_entry_t pv; 2008 register unsigned *pte, tpte; 2009 int s; 2010 2011#if defined(PMAP_DIAGNOSTIC) 2012 /* 2013 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 2014 * pages! 2015 */ 2016 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) { 2017 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m)); 2018 } 2019#endif 2020 2021 s = splvm(); 2022 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { 2023 pv->pv_pmap->pm_stats.resident_count--; 2024 2025 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2026 2027 tpte = atomic_readandclear_int(pte); 2028 if (tpte & PG_W) 2029 pv->pv_pmap->pm_stats.wired_count--; 2030 2031 if (tpte & PG_A) 2032 vm_page_flag_set(m, PG_REFERENCED); 2033 2034 /* 2035 * Update the vm_page_t clean and reference bits. 2036 */ 2037 if (tpte & PG_M) { 2038#if defined(PMAP_DIAGNOSTIC) 2039 if (pmap_nw_modified((pt_entry_t) tpte)) { 2040 printf( 2041 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n", 2042 pv->pv_va, tpte); 2043 } 2044#endif 2045 if (pmap_track_modified(pv->pv_va)) 2046 vm_page_dirty(m); 2047 } 2048 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 2049 2050 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2051 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 2052 m->md.pv_list_count--; 2053 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2054 free_pv_entry(pv); 2055 } 2056 2057 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 2058 2059 splx(s); 2060} 2061 2062/* 2063 * Set the physical protection on the 2064 * specified range of this map as requested. 2065 */ 2066void 2067pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 2068{ 2069 register unsigned *ptbase; 2070 vm_offset_t pdnxt, ptpaddr; 2071 vm_pindex_t sindex, eindex; 2072 int anychanged; 2073 2074 if (pmap == NULL) 2075 return; 2076 2077 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 2078 pmap_remove(pmap, sva, eva); 2079 return; 2080 } 2081 2082 if (prot & VM_PROT_WRITE) 2083 return; 2084 2085 anychanged = 0; 2086 2087 ptbase = get_ptbase(pmap); 2088 2089 sindex = i386_btop(sva); 2090 eindex = i386_btop(eva); 2091 2092 for (; sindex < eindex; sindex = pdnxt) { 2093 2094 unsigned pdirindex; 2095 2096 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 2097 2098 pdirindex = sindex / NPDEPG; 2099 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 2100 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 2101 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 2102 anychanged++; 2103 continue; 2104 } 2105 2106 /* 2107 * Weed out invalid mappings. Note: we assume that the page 2108 * directory table is always allocated, and in kernel virtual. 2109 */ 2110 if (ptpaddr == 0) 2111 continue; 2112 2113 if (pdnxt > eindex) { 2114 pdnxt = eindex; 2115 } 2116 2117 for (; sindex != pdnxt; sindex++) { 2118 2119 unsigned pbits; 2120 vm_page_t m; 2121 2122 pbits = ptbase[sindex]; 2123 2124 if (pbits & PG_MANAGED) { 2125 m = NULL; 2126 if (pbits & PG_A) { 2127 m = PHYS_TO_VM_PAGE(pbits); 2128 vm_page_flag_set(m, PG_REFERENCED); 2129 pbits &= ~PG_A; 2130 } 2131 if (pbits & PG_M) { 2132 if (pmap_track_modified(i386_ptob(sindex))) { 2133 if (m == NULL) 2134 m = PHYS_TO_VM_PAGE(pbits); 2135 vm_page_dirty(m); 2136 pbits &= ~PG_M; 2137 } 2138 } 2139 } 2140 2141 pbits &= ~PG_RW; 2142 2143 if (pbits != ptbase[sindex]) { 2144 ptbase[sindex] = pbits; 2145 anychanged = 1; 2146 } 2147 } 2148 } 2149 if (anychanged) 2150 pmap_TLB_invalidate_all(pmap); 2151} 2152 2153/* 2154 * Insert the given physical page (p) at 2155 * the specified virtual address (v) in the 2156 * target physical map with the protection requested. 2157 * 2158 * If specified, the page will be wired down, meaning 2159 * that the related pte can not be reclaimed. 2160 * 2161 * NB: This is the only routine which MAY NOT lazy-evaluate 2162 * or lose information. That is, this routine must actually 2163 * insert this page into the given map NOW. 2164 */ 2165void 2166pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, 2167 boolean_t wired) 2168{ 2169 vm_offset_t pa; 2170 register unsigned *pte; 2171 vm_offset_t opa; 2172 vm_offset_t origpte, newpte; 2173 vm_page_t mpte; 2174 2175 if (pmap == NULL) 2176 return; 2177 2178 va &= PG_FRAME; 2179#ifdef PMAP_DIAGNOSTIC 2180 if (va > VM_MAX_KERNEL_ADDRESS) 2181 panic("pmap_enter: toobig"); 2182 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 2183 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 2184#endif 2185 2186 mpte = NULL; 2187 /* 2188 * In the case that a page table page is not 2189 * resident, we are creating it here. 2190 */ 2191 if (va < UPT_MIN_ADDRESS) { 2192 mpte = pmap_allocpte(pmap, va); 2193 } 2194#if 0 && defined(PMAP_DIAGNOSTIC) 2195 else { 2196 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 2197 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 2198 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 2199 pmap->pm_pdir[PTDPTDI], origpte, va); 2200 } 2201 if (smp_active) { 2202 pdeaddr = (vm_offset_t *) IdlePTDS[PCPU_GET(cpuid)]; 2203 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2204 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2205 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2206 printf("cpuid: %d, pdeaddr: 0x%x\n", PCPU_GET(cpuid), pdeaddr); 2207 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2208 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2209 } 2210 } 2211 } 2212#endif 2213 2214 pte = pmap_pte(pmap, va); 2215 2216 /* 2217 * Page Directory table entry not valid, we need a new PT page 2218 */ 2219 if (pte == NULL) { 2220 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n", 2221 (void *)pmap->pm_pdir[PTDPTDI], va); 2222 } 2223 2224 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME; 2225 origpte = *(vm_offset_t *)pte; 2226 opa = origpte & PG_FRAME; 2227 2228 if (origpte & PG_PS) 2229 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2230 2231 /* 2232 * Mapping has not changed, must be protection or wiring change. 2233 */ 2234 if (origpte && (opa == pa)) { 2235 /* 2236 * Wiring change, just update stats. We don't worry about 2237 * wiring PT pages as they remain resident as long as there 2238 * are valid mappings in them. Hence, if a user page is wired, 2239 * the PT page will be also. 2240 */ 2241 if (wired && ((origpte & PG_W) == 0)) 2242 pmap->pm_stats.wired_count++; 2243 else if (!wired && (origpte & PG_W)) 2244 pmap->pm_stats.wired_count--; 2245 2246#if defined(PMAP_DIAGNOSTIC) 2247 if (pmap_nw_modified((pt_entry_t) origpte)) { 2248 printf( 2249 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n", 2250 va, origpte); 2251 } 2252#endif 2253 2254 /* 2255 * Remove extra pte reference 2256 */ 2257 if (mpte) 2258 mpte->hold_count--; 2259 2260 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2261 if ((origpte & PG_RW) == 0) { 2262 *pte |= PG_RW; 2263#ifdef SMP 2264 cpu_invlpg((void *)va); 2265 if (pmap->pm_active & PCPU_GET(other_cpus)) 2266 smp_invltlb(); 2267#else 2268 invltlb_1pg(va); 2269#endif 2270 } 2271 return; 2272 } 2273 2274 /* 2275 * We might be turning off write access to the page, 2276 * so we go ahead and sense modify status. 2277 */ 2278 if (origpte & PG_MANAGED) { 2279 if ((origpte & PG_M) && pmap_track_modified(va)) { 2280 vm_page_t om; 2281 om = PHYS_TO_VM_PAGE(opa); 2282 vm_page_dirty(om); 2283 } 2284 pa |= PG_MANAGED; 2285 } 2286 goto validate; 2287 } 2288 /* 2289 * Mapping has changed, invalidate old range and fall through to 2290 * handle validating new mapping. 2291 */ 2292 if (opa) { 2293 int err; 2294 err = pmap_remove_pte(pmap, pte, va); 2295 if (err) 2296 panic("pmap_enter: pte vanished, va: 0x%x", va); 2297 } 2298 2299 /* 2300 * Enter on the PV list if part of our managed memory. Note that we 2301 * raise IPL while manipulating pv_table since pmap_enter can be 2302 * called at interrupt time. 2303 */ 2304 if (pmap_initialized && 2305 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) { 2306 pmap_insert_entry(pmap, va, mpte, m); 2307 pa |= PG_MANAGED; 2308 } 2309 2310 /* 2311 * Increment counters 2312 */ 2313 pmap->pm_stats.resident_count++; 2314 if (wired) 2315 pmap->pm_stats.wired_count++; 2316 2317validate: 2318 /* 2319 * Now validate mapping with desired protection/wiring. 2320 */ 2321 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2322 2323 if (wired) 2324 newpte |= PG_W; 2325 if (va < UPT_MIN_ADDRESS) 2326 newpte |= PG_U; 2327 if (pmap == kernel_pmap) 2328 newpte |= pgeflag; 2329 2330 /* 2331 * if the mapping or permission bits are different, we need 2332 * to update the pte. 2333 */ 2334 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2335 *pte = newpte | PG_A; 2336 /*if (origpte)*/ { 2337#ifdef SMP 2338 cpu_invlpg((void *)va); 2339 if (pmap->pm_active & PCPU_GET(other_cpus)) 2340 smp_invltlb(); 2341#else 2342 invltlb_1pg(va); 2343#endif 2344 } 2345 } 2346} 2347 2348/* 2349 * this code makes some *MAJOR* assumptions: 2350 * 1. Current pmap & pmap exists. 2351 * 2. Not wired. 2352 * 3. Read access. 2353 * 4. No page table pages. 2354 * 5. Tlbflush is deferred to calling procedure. 2355 * 6. Page IS managed. 2356 * but is *MUCH* faster than pmap_enter... 2357 */ 2358 2359static vm_page_t 2360pmap_enter_quick(pmap, va, m, mpte) 2361 register pmap_t pmap; 2362 vm_offset_t va; 2363 vm_page_t m; 2364 vm_page_t mpte; 2365{ 2366 unsigned *pte; 2367 vm_offset_t pa; 2368 2369 /* 2370 * In the case that a page table page is not 2371 * resident, we are creating it here. 2372 */ 2373 if (va < UPT_MIN_ADDRESS) { 2374 unsigned ptepindex; 2375 vm_offset_t ptepa; 2376 2377 /* 2378 * Calculate pagetable page index 2379 */ 2380 ptepindex = va >> PDRSHIFT; 2381 if (mpte && (mpte->pindex == ptepindex)) { 2382 mpte->hold_count++; 2383 } else { 2384retry: 2385 /* 2386 * Get the page directory entry 2387 */ 2388 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2389 2390 /* 2391 * If the page table page is mapped, we just increment 2392 * the hold count, and activate it. 2393 */ 2394 if (ptepa) { 2395 if (ptepa & PG_PS) 2396 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2397 if (pmap->pm_ptphint && 2398 (pmap->pm_ptphint->pindex == ptepindex)) { 2399 mpte = pmap->pm_ptphint; 2400 } else { 2401 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2402 pmap->pm_ptphint = mpte; 2403 } 2404 if (mpte == NULL) 2405 goto retry; 2406 mpte->hold_count++; 2407 } else { 2408 mpte = _pmap_allocpte(pmap, ptepindex); 2409 } 2410 } 2411 } else { 2412 mpte = NULL; 2413 } 2414 2415 /* 2416 * This call to vtopte makes the assumption that we are 2417 * entering the page into the current pmap. In order to support 2418 * quick entry into any pmap, one would likely use pmap_pte_quick. 2419 * But that isn't as quick as vtopte. 2420 */ 2421 pte = (unsigned *)vtopte(va); 2422 if (*pte) { 2423 if (mpte) 2424 pmap_unwire_pte_hold(pmap, mpte); 2425 return 0; 2426 } 2427 2428 /* 2429 * Enter on the PV list if part of our managed memory. Note that we 2430 * raise IPL while manipulating pv_table since pmap_enter can be 2431 * called at interrupt time. 2432 */ 2433 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) 2434 pmap_insert_entry(pmap, va, mpte, m); 2435 2436 /* 2437 * Increment counters 2438 */ 2439 pmap->pm_stats.resident_count++; 2440 2441 pa = VM_PAGE_TO_PHYS(m); 2442 2443 /* 2444 * Now validate mapping with RO protection 2445 */ 2446 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) 2447 *pte = pa | PG_V | PG_U; 2448 else 2449 *pte = pa | PG_V | PG_U | PG_MANAGED; 2450 2451 return mpte; 2452} 2453 2454/* 2455 * Make a temporary mapping for a physical address. This is only intended 2456 * to be used for panic dumps. 2457 */ 2458void * 2459pmap_kenter_temporary(vm_offset_t pa, int i) 2460{ 2461 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa); 2462 return ((void *)crashdumpmap); 2463} 2464 2465#define MAX_INIT_PT (96) 2466/* 2467 * pmap_object_init_pt preloads the ptes for a given object 2468 * into the specified pmap. This eliminates the blast of soft 2469 * faults on process startup and immediately after an mmap. 2470 */ 2471void 2472pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2473 pmap_t pmap; 2474 vm_offset_t addr; 2475 vm_object_t object; 2476 vm_pindex_t pindex; 2477 vm_size_t size; 2478 int limit; 2479{ 2480 vm_offset_t tmpidx; 2481 int psize; 2482 vm_page_t p, mpte; 2483 int objpgs; 2484 2485 if (pmap == NULL || object == NULL) 2486 return; 2487 2488 /* 2489 * This code maps large physical mmap regions into the 2490 * processor address space. Note that some shortcuts 2491 * are taken, but the code works. 2492 */ 2493 if (pseflag && 2494 (object->type == OBJT_DEVICE) && 2495 ((addr & (NBPDR - 1)) == 0) && 2496 ((size & (NBPDR - 1)) == 0) ) { 2497 int i; 2498 vm_page_t m[1]; 2499 unsigned int ptepindex; 2500 int npdes; 2501 vm_offset_t ptepa; 2502 2503 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2504 return; 2505 2506retry: 2507 p = vm_page_lookup(object, pindex); 2508 if (p && vm_page_sleep_busy(p, FALSE, "init4p")) 2509 goto retry; 2510 2511 if (p == NULL) { 2512 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2513 if (p == NULL) 2514 return; 2515 m[0] = p; 2516 2517 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2518 vm_page_free(p); 2519 return; 2520 } 2521 2522 p = vm_page_lookup(object, pindex); 2523 vm_page_wakeup(p); 2524 } 2525 2526 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2527 if (ptepa & (NBPDR - 1)) { 2528 return; 2529 } 2530 2531 p->valid = VM_PAGE_BITS_ALL; 2532 2533 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2534 npdes = size >> PDRSHIFT; 2535 for(i = 0; i < npdes; i++) { 2536 pmap->pm_pdir[ptepindex] = 2537 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2538 ptepa += NBPDR; 2539 ptepindex += 1; 2540 } 2541 vm_page_flag_set(p, PG_MAPPED); 2542 invltlb(); 2543 return; 2544 } 2545 2546 psize = i386_btop(size); 2547 2548 if ((object->type != OBJT_VNODE) || 2549 ((limit & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) && 2550 (object->resident_page_count > MAX_INIT_PT))) { 2551 return; 2552 } 2553 2554 if (psize + pindex > object->size) { 2555 if (object->size < pindex) 2556 return; 2557 psize = object->size - pindex; 2558 } 2559 2560 mpte = NULL; 2561 /* 2562 * if we are processing a major portion of the object, then scan the 2563 * entire thing. 2564 */ 2565 if (psize > (object->resident_page_count >> 2)) { 2566 objpgs = psize; 2567 2568 for (p = TAILQ_FIRST(&object->memq); 2569 ((objpgs > 0) && (p != NULL)); 2570 p = TAILQ_NEXT(p, listq)) { 2571 2572 tmpidx = p->pindex; 2573 if (tmpidx < pindex) { 2574 continue; 2575 } 2576 tmpidx -= pindex; 2577 if (tmpidx >= psize) { 2578 continue; 2579 } 2580 /* 2581 * don't allow an madvise to blow away our really 2582 * free pages allocating pv entries. 2583 */ 2584 if ((limit & MAP_PREFAULT_MADVISE) && 2585 cnt.v_free_count < cnt.v_free_reserved) { 2586 break; 2587 } 2588 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2589 (p->busy == 0) && 2590 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2591 if ((p->queue - p->pc) == PQ_CACHE) 2592 vm_page_deactivate(p); 2593 vm_page_busy(p); 2594 mpte = pmap_enter_quick(pmap, 2595 addr + i386_ptob(tmpidx), p, mpte); 2596 vm_page_flag_set(p, PG_MAPPED); 2597 vm_page_wakeup(p); 2598 } 2599 objpgs -= 1; 2600 } 2601 } else { 2602 /* 2603 * else lookup the pages one-by-one. 2604 */ 2605 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2606 /* 2607 * don't allow an madvise to blow away our really 2608 * free pages allocating pv entries. 2609 */ 2610 if ((limit & MAP_PREFAULT_MADVISE) && 2611 cnt.v_free_count < cnt.v_free_reserved) { 2612 break; 2613 } 2614 p = vm_page_lookup(object, tmpidx + pindex); 2615 if (p && 2616 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2617 (p->busy == 0) && 2618 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2619 if ((p->queue - p->pc) == PQ_CACHE) 2620 vm_page_deactivate(p); 2621 vm_page_busy(p); 2622 mpte = pmap_enter_quick(pmap, 2623 addr + i386_ptob(tmpidx), p, mpte); 2624 vm_page_flag_set(p, PG_MAPPED); 2625 vm_page_wakeup(p); 2626 } 2627 } 2628 } 2629 return; 2630} 2631 2632/* 2633 * pmap_prefault provides a quick way of clustering 2634 * pagefaults into a processes address space. It is a "cousin" 2635 * of pmap_object_init_pt, except it runs at page fault time instead 2636 * of mmap time. 2637 */ 2638#define PFBAK 4 2639#define PFFOR 4 2640#define PAGEORDER_SIZE (PFBAK+PFFOR) 2641 2642static int pmap_prefault_pageorder[] = { 2643 -PAGE_SIZE, PAGE_SIZE, 2644 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2645 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2646 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2647}; 2648 2649void 2650pmap_prefault(pmap, addra, entry) 2651 pmap_t pmap; 2652 vm_offset_t addra; 2653 vm_map_entry_t entry; 2654{ 2655 int i; 2656 vm_offset_t starta; 2657 vm_offset_t addr; 2658 vm_pindex_t pindex; 2659 vm_page_t m, mpte; 2660 vm_object_t object; 2661 2662 if (!curthread || (pmap != vmspace_pmap(curthread->td_proc->p_vmspace))) 2663 return; 2664 2665 object = entry->object.vm_object; 2666 2667 starta = addra - PFBAK * PAGE_SIZE; 2668 if (starta < entry->start) { 2669 starta = entry->start; 2670 } else if (starta > addra) { 2671 starta = 0; 2672 } 2673 2674 mpte = NULL; 2675 for (i = 0; i < PAGEORDER_SIZE; i++) { 2676 vm_object_t lobject; 2677 unsigned *pte; 2678 2679 addr = addra + pmap_prefault_pageorder[i]; 2680 if (addr > addra + (PFFOR * PAGE_SIZE)) 2681 addr = 0; 2682 2683 if (addr < starta || addr >= entry->end) 2684 continue; 2685 2686 if ((*pmap_pde(pmap, addr)) == NULL) 2687 continue; 2688 2689 pte = (unsigned *) vtopte(addr); 2690 if (*pte) 2691 continue; 2692 2693 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2694 lobject = object; 2695 for (m = vm_page_lookup(lobject, pindex); 2696 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2697 lobject = lobject->backing_object) { 2698 if (lobject->backing_object_offset & PAGE_MASK) 2699 break; 2700 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2701 m = vm_page_lookup(lobject->backing_object, pindex); 2702 } 2703 2704 /* 2705 * give-up when a page is not in memory 2706 */ 2707 if (m == NULL) 2708 break; 2709 2710 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2711 (m->busy == 0) && 2712 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2713 2714 if ((m->queue - m->pc) == PQ_CACHE) { 2715 vm_page_deactivate(m); 2716 } 2717 vm_page_busy(m); 2718 mpte = pmap_enter_quick(pmap, addr, m, mpte); 2719 vm_page_flag_set(m, PG_MAPPED); 2720 vm_page_wakeup(m); 2721 } 2722 } 2723} 2724 2725/* 2726 * Routine: pmap_change_wiring 2727 * Function: Change the wiring attribute for a map/virtual-address 2728 * pair. 2729 * In/out conditions: 2730 * The mapping must already exist in the pmap. 2731 */ 2732void 2733pmap_change_wiring(pmap, va, wired) 2734 register pmap_t pmap; 2735 vm_offset_t va; 2736 boolean_t wired; 2737{ 2738 register unsigned *pte; 2739 2740 if (pmap == NULL) 2741 return; 2742 2743 pte = pmap_pte(pmap, va); 2744 2745 if (wired && !pmap_pte_w(pte)) 2746 pmap->pm_stats.wired_count++; 2747 else if (!wired && pmap_pte_w(pte)) 2748 pmap->pm_stats.wired_count--; 2749 2750 /* 2751 * Wiring is not a hardware characteristic so there is no need to 2752 * invalidate TLB. 2753 */ 2754 pmap_pte_set_w(pte, wired); 2755} 2756 2757 2758 2759/* 2760 * Copy the range specified by src_addr/len 2761 * from the source map to the range dst_addr/len 2762 * in the destination map. 2763 * 2764 * This routine is only advisory and need not do anything. 2765 */ 2766 2767void 2768pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2769 pmap_t dst_pmap, src_pmap; 2770 vm_offset_t dst_addr; 2771 vm_size_t len; 2772 vm_offset_t src_addr; 2773{ 2774 vm_offset_t addr; 2775 vm_offset_t end_addr = src_addr + len; 2776 vm_offset_t pdnxt; 2777 unsigned src_frame, dst_frame; 2778 vm_page_t m; 2779 pd_entry_t saved_pde; 2780 2781 if (dst_addr != src_addr) 2782 return; 2783 2784 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2785 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2786 return; 2787 } 2788 2789 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2790 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2791 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2792#if defined(SMP) 2793 /* The page directory is not shared between CPUs */ 2794 cpu_invltlb(); 2795#else 2796 invltlb(); 2797#endif 2798 } 2799 saved_pde = (pd_entry_t)((u_int32_t)APTDpde & (PG_FRAME|PG_RW | PG_V)); 2800 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2801 unsigned *src_pte, *dst_pte; 2802 vm_page_t dstmpte, srcmpte; 2803 vm_offset_t srcptepaddr; 2804 unsigned ptepindex; 2805 2806 if (addr >= UPT_MIN_ADDRESS) 2807 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2808 2809 /* 2810 * Don't let optional prefaulting of pages make us go 2811 * way below the low water mark of free pages or way 2812 * above high water mark of used pv entries. 2813 */ 2814 if (cnt.v_free_count < cnt.v_free_reserved || 2815 pv_entry_count > pv_entry_high_water) 2816 break; 2817 2818 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2819 ptepindex = addr >> PDRSHIFT; 2820 2821 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2822 if (srcptepaddr == 0) 2823 continue; 2824 2825 if (srcptepaddr & PG_PS) { 2826 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2827 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2828 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE; 2829 } 2830 continue; 2831 } 2832 2833 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2834 if ((srcmpte == NULL) || 2835 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2836 continue; 2837 2838 if (pdnxt > end_addr) 2839 pdnxt = end_addr; 2840 2841 src_pte = (unsigned *) vtopte(addr); 2842 dst_pte = (unsigned *) avtopte(addr); 2843 while (addr < pdnxt) { 2844 unsigned ptetemp; 2845 ptetemp = *src_pte; 2846 /* 2847 * we only virtual copy managed pages 2848 */ 2849 if ((ptetemp & PG_MANAGED) != 0) { 2850 /* 2851 * We have to check after allocpte for the 2852 * pte still being around... allocpte can 2853 * block. 2854 */ 2855 dstmpte = pmap_allocpte(dst_pmap, addr); 2856 if (((u_int32_t)APTDpde & PG_FRAME) != 2857 ((u_int32_t)saved_pde & PG_FRAME)) { 2858 APTDpde = saved_pde; 2859printf ("IT HAPPENNED!"); 2860#if defined(SMP) 2861 cpu_invltlb(); 2862#else 2863 invltlb(); 2864#endif 2865 } 2866 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2867 /* 2868 * Clear the modified and 2869 * accessed (referenced) bits 2870 * during the copy. 2871 */ 2872 m = PHYS_TO_VM_PAGE(ptetemp); 2873 *dst_pte = ptetemp & ~(PG_M | PG_A); 2874 dst_pmap->pm_stats.resident_count++; 2875 pmap_insert_entry(dst_pmap, addr, 2876 dstmpte, m); 2877 } else { 2878 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2879 } 2880 if (dstmpte->hold_count >= srcmpte->hold_count) 2881 break; 2882 } 2883 addr += PAGE_SIZE; 2884 src_pte++; 2885 dst_pte++; 2886 } 2887 } 2888} 2889 2890/* 2891 * Routine: pmap_kernel 2892 * Function: 2893 * Returns the physical map handle for the kernel. 2894 */ 2895pmap_t 2896pmap_kernel() 2897{ 2898 return (kernel_pmap); 2899} 2900 2901/* 2902 * pmap_zero_page zeros the specified hardware page by mapping 2903 * the page into KVM and using bzero to clear its contents. 2904 */ 2905void 2906pmap_zero_page(phys) 2907 vm_offset_t phys; 2908{ 2909 2910 if (*(int *) CMAP2) 2911 panic("pmap_zero_page: CMAP2 busy"); 2912 2913 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2914 invltlb_1pg((vm_offset_t)CADDR2); 2915 2916#if defined(I686_CPU) 2917 if (cpu_class == CPUCLASS_686) 2918 i686_pagezero(CADDR2); 2919 else 2920#endif 2921 bzero(CADDR2, PAGE_SIZE); 2922 *(int *) CMAP2 = 0; 2923} 2924 2925/* 2926 * pmap_zero_page_area zeros the specified hardware page by mapping 2927 * the page into KVM and using bzero to clear its contents. 2928 * 2929 * off and size may not cover an area beyond a single hardware page. 2930 */ 2931void 2932pmap_zero_page_area(phys, off, size) 2933 vm_offset_t phys; 2934 int off; 2935 int size; 2936{ 2937 2938 if (*(int *) CMAP2) 2939 panic("pmap_zero_page: CMAP2 busy"); 2940 2941 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2942 invltlb_1pg((vm_offset_t)CADDR2); 2943 2944#if defined(I686_CPU) 2945 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2946 i686_pagezero(CADDR2); 2947 else 2948#endif 2949 bzero((char *)CADDR2 + off, size); 2950 *(int *) CMAP2 = 0; 2951} 2952 2953/* 2954 * pmap_copy_page copies the specified (machine independent) 2955 * page by mapping the page into virtual memory and using 2956 * bcopy to copy the page, one machine dependent page at a 2957 * time. 2958 */ 2959void 2960pmap_copy_page(src, dst) 2961 vm_offset_t src; 2962 vm_offset_t dst; 2963{ 2964 2965 if (*(int *) CMAP1) 2966 panic("pmap_copy_page: CMAP1 busy"); 2967 if (*(int *) CMAP2) 2968 panic("pmap_copy_page: CMAP2 busy"); 2969 2970 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2971 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2972#ifdef I386_CPU 2973 invltlb(); 2974#else 2975 invlpg((u_int)CADDR1); 2976 invlpg((u_int)CADDR2); 2977#endif 2978 2979 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2980 2981 *(int *) CMAP1 = 0; 2982 *(int *) CMAP2 = 0; 2983} 2984 2985 2986/* 2987 * Routine: pmap_pageable 2988 * Function: 2989 * Make the specified pages (by pmap, offset) 2990 * pageable (or not) as requested. 2991 * 2992 * A page which is not pageable may not take 2993 * a fault; therefore, its page table entry 2994 * must remain valid for the duration. 2995 * 2996 * This routine is merely advisory; pmap_enter 2997 * will specify that these pages are to be wired 2998 * down (or not) as appropriate. 2999 */ 3000void 3001pmap_pageable(pmap, sva, eva, pageable) 3002 pmap_t pmap; 3003 vm_offset_t sva, eva; 3004 boolean_t pageable; 3005{ 3006} 3007 3008/* 3009 * this routine returns true if a physical page resides 3010 * in the given pmap. 3011 */ 3012boolean_t 3013pmap_page_exists(pmap, m) 3014 pmap_t pmap; 3015 vm_page_t m; 3016{ 3017 register pv_entry_t pv; 3018 int s; 3019 3020 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 3021 return FALSE; 3022 3023 s = splvm(); 3024 3025 /* 3026 * Not found, check current mappings returning immediately if found. 3027 */ 3028 for (pv = TAILQ_FIRST(&m->md.pv_list); 3029 pv; 3030 pv = TAILQ_NEXT(pv, pv_list)) { 3031 if (pv->pv_pmap == pmap) { 3032 splx(s); 3033 return TRUE; 3034 } 3035 } 3036 splx(s); 3037 return (FALSE); 3038} 3039 3040#define PMAP_REMOVE_PAGES_CURPROC_ONLY 3041/* 3042 * Remove all pages from specified address space 3043 * this aids process exit speeds. Also, this code 3044 * is special cased for current process only, but 3045 * can have the more generic (and slightly slower) 3046 * mode enabled. This is much faster than pmap_remove 3047 * in the case of running down an entire address space. 3048 */ 3049void 3050pmap_remove_pages(pmap, sva, eva) 3051 pmap_t pmap; 3052 vm_offset_t sva, eva; 3053{ 3054 unsigned *pte, tpte; 3055 pv_entry_t pv, npv; 3056 int s; 3057 vm_page_t m; 3058 3059#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 3060 if (!curthread || (pmap != vmspace_pmap(curthread->td_proc->p_vmspace))) { 3061 printf("warning: pmap_remove_pages called with non-current pmap\n"); 3062 return; 3063 } 3064#endif 3065 3066 s = splvm(); 3067 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 3068 pv; 3069 pv = npv) { 3070 3071 if (pv->pv_va >= eva || pv->pv_va < sva) { 3072 npv = TAILQ_NEXT(pv, pv_plist); 3073 continue; 3074 } 3075 3076#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 3077 pte = (unsigned *)vtopte(pv->pv_va); 3078#else 3079 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3080#endif 3081 tpte = *pte; 3082 3083 if (tpte == 0) { 3084 printf("TPTE at %p IS ZERO @ VA %08x\n", 3085 pte, pv->pv_va); 3086 panic("bad pte"); 3087 } 3088 3089/* 3090 * We cannot remove wired pages from a process' mapping at this time 3091 */ 3092 if (tpte & PG_W) { 3093 npv = TAILQ_NEXT(pv, pv_plist); 3094 continue; 3095 } 3096 3097 m = PHYS_TO_VM_PAGE(tpte); 3098 KASSERT(m->phys_addr == (tpte & PG_FRAME), 3099 ("vm_page_t %p phys_addr mismatch %08x %08x", 3100 m, m->phys_addr, tpte)); 3101 3102 KASSERT(m < &vm_page_array[vm_page_array_size], 3103 ("pmap_remove_pages: bad tpte %x", tpte)); 3104 3105 pv->pv_pmap->pm_stats.resident_count--; 3106 3107 *pte = 0; 3108 3109 /* 3110 * Update the vm_page_t clean and reference bits. 3111 */ 3112 if (tpte & PG_M) { 3113 vm_page_dirty(m); 3114 } 3115 3116 npv = TAILQ_NEXT(pv, pv_plist); 3117 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 3118 3119 m->md.pv_list_count--; 3120 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 3121 if (TAILQ_FIRST(&m->md.pv_list) == NULL) { 3122 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 3123 } 3124 3125 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 3126 free_pv_entry(pv); 3127 } 3128 splx(s); 3129 pmap_TLB_invalidate_all(pmap); 3130} 3131 3132/* 3133 * pmap_testbit tests bits in pte's 3134 * note that the testbit/changebit routines are inline, 3135 * and a lot of things compile-time evaluate. 3136 */ 3137static boolean_t 3138pmap_testbit(m, bit) 3139 vm_page_t m; 3140 int bit; 3141{ 3142 pv_entry_t pv; 3143 unsigned *pte; 3144 int s; 3145 3146 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 3147 return FALSE; 3148 3149 if (TAILQ_FIRST(&m->md.pv_list) == NULL) 3150 return FALSE; 3151 3152 s = splvm(); 3153 3154 for (pv = TAILQ_FIRST(&m->md.pv_list); 3155 pv; 3156 pv = TAILQ_NEXT(pv, pv_list)) { 3157 3158 /* 3159 * if the bit being tested is the modified bit, then 3160 * mark clean_map and ptes as never 3161 * modified. 3162 */ 3163 if (bit & (PG_A|PG_M)) { 3164 if (!pmap_track_modified(pv->pv_va)) 3165 continue; 3166 } 3167 3168#if defined(PMAP_DIAGNOSTIC) 3169 if (!pv->pv_pmap) { 3170 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 3171 continue; 3172 } 3173#endif 3174 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3175 if (*pte & bit) { 3176 splx(s); 3177 return TRUE; 3178 } 3179 } 3180 splx(s); 3181 return (FALSE); 3182} 3183 3184/* 3185 * this routine is used to modify bits in ptes 3186 */ 3187static __inline void 3188pmap_changebit(m, bit, setem) 3189 vm_page_t m; 3190 int bit; 3191 boolean_t setem; 3192{ 3193 register pv_entry_t pv; 3194 register unsigned *pte; 3195 int s; 3196 3197 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 3198 return; 3199 3200 s = splvm(); 3201 3202 /* 3203 * Loop over all current mappings setting/clearing as appropos If 3204 * setting RO do we need to clear the VAC? 3205 */ 3206 for (pv = TAILQ_FIRST(&m->md.pv_list); 3207 pv; 3208 pv = TAILQ_NEXT(pv, pv_list)) { 3209 3210 /* 3211 * don't write protect pager mappings 3212 */ 3213 if (!setem && (bit == PG_RW)) { 3214 if (!pmap_track_modified(pv->pv_va)) 3215 continue; 3216 } 3217 3218#if defined(PMAP_DIAGNOSTIC) 3219 if (!pv->pv_pmap) { 3220 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 3221 continue; 3222 } 3223#endif 3224 3225 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3226 3227 if (setem) { 3228 *(int *)pte |= bit; 3229 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3230 } else { 3231 vm_offset_t pbits = *(vm_offset_t *)pte; 3232 if (pbits & bit) { 3233 if (bit == PG_RW) { 3234 if (pbits & PG_M) { 3235 vm_page_dirty(m); 3236 } 3237 *(int *)pte = pbits & ~(PG_M|PG_RW); 3238 } else { 3239 *(int *)pte = pbits & ~bit; 3240 } 3241 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3242 } 3243 } 3244 } 3245 splx(s); 3246} 3247 3248/* 3249 * pmap_page_protect: 3250 * 3251 * Lower the permission for all mappings to a given page. 3252 */ 3253void 3254pmap_page_protect(vm_page_t m, vm_prot_t prot) 3255{ 3256 if ((prot & VM_PROT_WRITE) == 0) { 3257 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3258 pmap_changebit(m, PG_RW, FALSE); 3259 } else { 3260 pmap_remove_all(m); 3261 } 3262 } 3263} 3264 3265vm_offset_t 3266pmap_phys_address(ppn) 3267 int ppn; 3268{ 3269 return (i386_ptob(ppn)); 3270} 3271 3272/* 3273 * pmap_ts_referenced: 3274 * 3275 * Return the count of reference bits for a page, clearing all of them. 3276 */ 3277int 3278pmap_ts_referenced(vm_page_t m) 3279{ 3280 register pv_entry_t pv, pvf, pvn; 3281 unsigned *pte; 3282 int s; 3283 int rtval = 0; 3284 3285 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 3286 return (rtval); 3287 3288 s = splvm(); 3289 3290 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { 3291 3292 pvf = pv; 3293 3294 do { 3295 pvn = TAILQ_NEXT(pv, pv_list); 3296 3297 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 3298 3299 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); 3300 3301 if (!pmap_track_modified(pv->pv_va)) 3302 continue; 3303 3304 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3305 3306 if (pte && (*pte & PG_A)) { 3307 *pte &= ~PG_A; 3308 3309 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3310 3311 rtval++; 3312 if (rtval > 4) { 3313 break; 3314 } 3315 } 3316 } while ((pv = pvn) != NULL && pv != pvf); 3317 } 3318 splx(s); 3319 3320 return (rtval); 3321} 3322 3323/* 3324 * pmap_is_modified: 3325 * 3326 * Return whether or not the specified physical page was modified 3327 * in any physical maps. 3328 */ 3329boolean_t 3330pmap_is_modified(vm_page_t m) 3331{ 3332 return pmap_testbit(m, PG_M); 3333} 3334 3335/* 3336 * Clear the modify bits on the specified physical page. 3337 */ 3338void 3339pmap_clear_modify(vm_page_t m) 3340{ 3341 pmap_changebit(m, PG_M, FALSE); 3342} 3343 3344/* 3345 * pmap_clear_reference: 3346 * 3347 * Clear the reference bit on the specified physical page. 3348 */ 3349void 3350pmap_clear_reference(vm_page_t m) 3351{ 3352 pmap_changebit(m, PG_A, FALSE); 3353} 3354 3355/* 3356 * Miscellaneous support routines follow 3357 */ 3358 3359static void 3360i386_protection_init() 3361{ 3362 register int *kp, prot; 3363 3364 kp = protection_codes; 3365 for (prot = 0; prot < 8; prot++) { 3366 switch (prot) { 3367 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3368 /* 3369 * Read access is also 0. There isn't any execute bit, 3370 * so just make it readable. 3371 */ 3372 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3373 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3374 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3375 *kp++ = 0; 3376 break; 3377 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3378 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3379 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3380 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3381 *kp++ = PG_RW; 3382 break; 3383 } 3384 } 3385} 3386 3387/* 3388 * Map a set of physical memory pages into the kernel virtual 3389 * address space. Return a pointer to where it is mapped. This 3390 * routine is intended to be used for mapping device memory, 3391 * NOT real memory. 3392 */ 3393void * 3394pmap_mapdev(pa, size) 3395 vm_offset_t pa; 3396 vm_size_t size; 3397{ 3398 vm_offset_t va, tmpva, offset; 3399 unsigned *pte; 3400 3401 offset = pa & PAGE_MASK; 3402 size = roundup(offset + size, PAGE_SIZE); 3403 3404 GIANT_REQUIRED; 3405 3406 va = kmem_alloc_pageable(kernel_map, size); 3407 if (!va) 3408 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3409 3410 pa = pa & PG_FRAME; 3411 for (tmpva = va; size > 0;) { 3412 pte = (unsigned *)vtopte(tmpva); 3413 *pte = pa | PG_RW | PG_V | pgeflag; 3414 size -= PAGE_SIZE; 3415 tmpva += PAGE_SIZE; 3416 pa += PAGE_SIZE; 3417 } 3418 invltlb(); 3419 3420 return ((void *)(va + offset)); 3421} 3422 3423void 3424pmap_unmapdev(va, size) 3425 vm_offset_t va; 3426 vm_size_t size; 3427{ 3428 vm_offset_t base, offset; 3429 3430 base = va & PG_FRAME; 3431 offset = va & PAGE_MASK; 3432 size = roundup(offset + size, PAGE_SIZE); 3433 kmem_free(kernel_map, base, size); 3434} 3435 3436/* 3437 * perform the pmap work for mincore 3438 */ 3439int 3440pmap_mincore(pmap, addr) 3441 pmap_t pmap; 3442 vm_offset_t addr; 3443{ 3444 3445 unsigned *ptep, pte; 3446 vm_page_t m; 3447 int val = 0; 3448 3449 ptep = pmap_pte(pmap, addr); 3450 if (ptep == 0) { 3451 return 0; 3452 } 3453 3454 if ((pte = *ptep) != 0) { 3455 vm_offset_t pa; 3456 3457 val = MINCORE_INCORE; 3458 if ((pte & PG_MANAGED) == 0) 3459 return val; 3460 3461 pa = pte & PG_FRAME; 3462 3463 m = PHYS_TO_VM_PAGE(pa); 3464 3465 /* 3466 * Modified by us 3467 */ 3468 if (pte & PG_M) 3469 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3470 /* 3471 * Modified by someone 3472 */ 3473 else if (m->dirty || pmap_is_modified(m)) 3474 val |= MINCORE_MODIFIED_OTHER; 3475 /* 3476 * Referenced by us 3477 */ 3478 if (pte & PG_A) 3479 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3480 3481 /* 3482 * Referenced by someone 3483 */ 3484 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) { 3485 val |= MINCORE_REFERENCED_OTHER; 3486 vm_page_flag_set(m, PG_REFERENCED); 3487 } 3488 } 3489 return val; 3490} 3491 3492void 3493pmap_activate(struct thread *td) 3494{ 3495 struct proc *p = td->td_proc; 3496 pmap_t pmap; 3497 u_int32_t cr3; 3498 3499 pmap = vmspace_pmap(td->td_proc->p_vmspace); 3500#if defined(SMP) 3501 pmap->pm_active |= 1 << PCPU_GET(cpuid); 3502#else 3503 pmap->pm_active |= 1; 3504#endif 3505#if defined(SWTCH_OPTIM_STATS) 3506 tlb_flush_count++; 3507#endif 3508 cr3 = vtophys(pmap->pm_pdir); 3509 /* XXXKSE this is wrong. 3510 * pmap_activate is for the current thread on the current cpu 3511 */ 3512 if (p->p_flag & P_KSES) { 3513 /* Make sure all other cr3 entries are updated. */ 3514 /* what if they are running? XXXKSE (maybe abort them) */ 3515 FOREACH_THREAD_IN_PROC(p, td) { 3516 td->td_pcb->pcb_cr3 = cr3; 3517 } 3518 } else { 3519 td->td_pcb->pcb_cr3 = cr3; 3520 } 3521 load_cr3(cr3); 3522} 3523 3524vm_offset_t 3525pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) 3526{ 3527 3528 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3529 return addr; 3530 } 3531 3532 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3533 return addr; 3534} 3535 3536 3537#if defined(PMAP_DEBUG) 3538pmap_pid_dump(int pid) 3539{ 3540 pmap_t pmap; 3541 struct proc *p; 3542 int npte = 0; 3543 int index; 3544 3545 sx_slock(&allproc_lock); 3546 LIST_FOREACH(p, &allproc, p_list) { 3547 if (p->p_pid != pid) 3548 continue; 3549 3550 if (p->p_vmspace) { 3551 int i,j; 3552 index = 0; 3553 pmap = vmspace_pmap(p->p_vmspace); 3554 for (i = 0; i < NPDEPG; i++) { 3555 pd_entry_t *pde; 3556 pt_entry_t *pte; 3557 vm_offset_t base = i << PDRSHIFT; 3558 3559 pde = &pmap->pm_pdir[i]; 3560 if (pde && pmap_pde_v(pde)) { 3561 for (j = 0; j < NPTEPG; j++) { 3562 vm_offset_t va = base + (j << PAGE_SHIFT); 3563 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3564 if (index) { 3565 index = 0; 3566 printf("\n"); 3567 } 3568 sx_sunlock(&allproc_lock); 3569 return npte; 3570 } 3571 pte = pmap_pte_quick( pmap, va); 3572 if (pte && pmap_pte_v(pte)) { 3573 vm_offset_t pa; 3574 vm_page_t m; 3575 pa = *(int *)pte; 3576 m = PHYS_TO_VM_PAGE(pa); 3577 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3578 va, pa, m->hold_count, m->wire_count, m->flags); 3579 npte++; 3580 index++; 3581 if (index >= 2) { 3582 index = 0; 3583 printf("\n"); 3584 } else { 3585 printf(" "); 3586 } 3587 } 3588 } 3589 } 3590 } 3591 } 3592 } 3593 sx_sunlock(&allproc_lock); 3594 return npte; 3595} 3596#endif 3597 3598#if defined(DEBUG) 3599 3600static void pads __P((pmap_t pm)); 3601void pmap_pvdump __P((vm_offset_t pa)); 3602 3603/* print address space of pmap*/ 3604static void 3605pads(pm) 3606 pmap_t pm; 3607{ 3608 int i, j; 3609 vm_offset_t va; 3610 pt_entry_t *ptep; 3611 3612 if (pm == kernel_pmap) 3613 return; 3614 for (i = 0; i < NPDEPG; i++) 3615 if (pm->pm_pdir[i]) 3616 for (j = 0; j < NPTEPG; j++) { 3617 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3618 if (pm == kernel_pmap && va < KERNBASE) 3619 continue; 3620 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3621 continue; 3622 ptep = pmap_pte_quick(pm, va); 3623 if (pmap_pte_v(ptep)) 3624 printf("%x:%x ", va, *(int *) ptep); 3625 }; 3626 3627} 3628 3629void 3630pmap_pvdump(pa) 3631 vm_offset_t pa; 3632{ 3633 pv_entry_t pv; 3634 vm_page_t m; 3635 3636 printf("pa %x", pa); 3637 m = PHYS_TO_VM_PAGE(pa); 3638 for (pv = TAILQ_FIRST(&m->md.pv_list); 3639 pv; 3640 pv = TAILQ_NEXT(pv, pv_list)) { 3641 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3642 pads(pv->pv_pmap); 3643 } 3644 printf(" "); 3645} 3646#endif 3647