pmap.c revision 66069
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 66069 2000-09-19 10:40:52Z eivind $ 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_user_ldt.h" 75 76#include <sys/param.h> 77#include <sys/systm.h> 78#include <sys/proc.h> 79#include <sys/msgbuf.h> 80#include <sys/vmmeter.h> 81#include <sys/mman.h> 82 83#include <vm/vm.h> 84#include <vm/vm_param.h> 85#include <sys/lock.h> 86#include <vm/vm_kern.h> 87#include <vm/vm_page.h> 88#include <vm/vm_map.h> 89#include <vm/vm_object.h> 90#include <vm/vm_extern.h> 91#include <vm/vm_pageout.h> 92#include <vm/vm_pager.h> 93#include <vm/vm_zone.h> 94 95#include <sys/user.h> 96 97#include <machine/cputypes.h> 98#include <machine/md_var.h> 99#include <machine/specialreg.h> 100#if defined(SMP) || defined(APIC_IO) 101#include <machine/smp.h> 102#include <machine/apic.h> 103#include <machine/segments.h> 104#include <machine/tss.h> 105#include <machine/globaldata.h> 106#endif /* SMP || APIC_IO */ 107 108#define PMAP_KEEP_PDIRS 109#ifndef PMAP_SHPGPERPROC 110#define PMAP_SHPGPERPROC 200 111#endif 112 113#if defined(DIAGNOSTIC) 114#define PMAP_DIAGNOSTIC 115#endif 116 117#define MINPV 2048 118 119#if !defined(PMAP_DIAGNOSTIC) 120#define PMAP_INLINE __inline 121#else 122#define PMAP_INLINE 123#endif 124 125/* 126 * Get PDEs and PTEs for user/kernel address space 127 */ 128#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 129#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 130 131#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 132#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 133#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 134#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 135#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 136 137#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 138#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 139 140/* 141 * Given a map and a machine independent protection code, 142 * convert to a vax protection code. 143 */ 144#define pte_prot(m, p) (protection_codes[p]) 145static int protection_codes[8]; 146 147static struct pmap kernel_pmap_store; 148pmap_t kernel_pmap; 149LIST_HEAD(pmaplist, pmap); 150struct pmaplist allpmaps; 151 152vm_offset_t avail_start; /* PA of first available physical page */ 153vm_offset_t avail_end; /* PA of last available physical page */ 154vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 155vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 156static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 157static int pgeflag; /* PG_G or-in */ 158static int pseflag; /* PG_PS or-in */ 159 160static vm_object_t kptobj; 161 162static int nkpt; 163vm_offset_t kernel_vm_end; 164 165/* 166 * Data for the pv entry allocation mechanism 167 */ 168static vm_zone_t pvzone; 169static struct vm_zone pvzone_store; 170static struct vm_object pvzone_obj; 171static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0; 172static int pmap_pagedaemon_waken = 0; 173static struct pv_entry *pvinit; 174 175/* 176 * All those kernel PT submaps that BSD is so fond of 177 */ 178pt_entry_t *CMAP1 = 0; 179static pt_entry_t *CMAP2, *ptmmap; 180caddr_t CADDR1 = 0, ptvmmap = 0; 181static caddr_t CADDR2; 182static pt_entry_t *msgbufmap; 183struct msgbuf *msgbufp=0; 184 185#ifdef SMP 186extern pt_entry_t *SMPpt; 187#else 188static pt_entry_t *PMAP1 = 0; 189static unsigned *PADDR1 = 0; 190#endif 191 192static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 193static unsigned * get_ptbase __P((pmap_t pmap)); 194static pv_entry_t get_pv_entry __P((void)); 195static void i386_protection_init __P((void)); 196static __inline void pmap_changebit __P((vm_page_t m, int bit, boolean_t setem)); 197 198static void pmap_remove_all __P((vm_page_t m)); 199static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 200 vm_page_t m, vm_page_t mpte)); 201static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 202 vm_offset_t sva)); 203static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 204static int pmap_remove_entry __P((struct pmap *pmap, vm_page_t m, 205 vm_offset_t va)); 206static boolean_t pmap_testbit __P((vm_page_t m, int bit)); 207static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 208 vm_page_t mpte, vm_page_t m)); 209 210static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 211 212static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 213static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 214static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 215static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 216static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 217static vm_offset_t pmap_kmem_choose(vm_offset_t addr); 218 219static unsigned pdir4mb; 220 221/* 222 * Routine: pmap_pte 223 * Function: 224 * Extract the page table entry associated 225 * with the given map/virtual_address pair. 226 */ 227 228PMAP_INLINE unsigned * 229pmap_pte(pmap, va) 230 register pmap_t pmap; 231 vm_offset_t va; 232{ 233 unsigned *pdeaddr; 234 235 if (pmap) { 236 pdeaddr = (unsigned *) pmap_pde(pmap, va); 237 if (*pdeaddr & PG_PS) 238 return pdeaddr; 239 if (*pdeaddr) { 240 return get_ptbase(pmap) + i386_btop(va); 241 } 242 } 243 return (0); 244} 245 246/* 247 * Move the kernel virtual free pointer to the next 248 * 4MB. This is used to help improve performance 249 * by using a large (4MB) page for much of the kernel 250 * (.text, .data, .bss) 251 */ 252static vm_offset_t 253pmap_kmem_choose(vm_offset_t addr) 254{ 255 vm_offset_t newaddr = addr; 256#ifndef DISABLE_PSE 257 if (cpu_feature & CPUID_PSE) { 258 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 259 } 260#endif 261 return newaddr; 262} 263 264/* 265 * Bootstrap the system enough to run with virtual memory. 266 * 267 * On the i386 this is called after mapping has already been enabled 268 * and just syncs the pmap module with what has already been done. 269 * [We can't call it easily with mapping off since the kernel is not 270 * mapped with PA == VA, hence we would have to relocate every address 271 * from the linked base (virtual) address "KERNBASE" to the actual 272 * (physical) address starting relative to 0] 273 */ 274void 275pmap_bootstrap(firstaddr, loadaddr) 276 vm_offset_t firstaddr; 277 vm_offset_t loadaddr; 278{ 279 vm_offset_t va; 280 pt_entry_t *pte; 281#ifdef SMP 282 int i, j; 283 struct globaldata *gd; 284#endif 285 286 avail_start = firstaddr; 287 288 /* 289 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 290 * large. It should instead be correctly calculated in locore.s and 291 * not based on 'first' (which is a physical address, not a virtual 292 * address, for the start of unused physical memory). The kernel 293 * page tables are NOT double mapped and thus should not be included 294 * in this calculation. 295 */ 296 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 297 virtual_avail = pmap_kmem_choose(virtual_avail); 298 299 virtual_end = VM_MAX_KERNEL_ADDRESS; 300 301 /* 302 * Initialize protection array. 303 */ 304 i386_protection_init(); 305 306 /* 307 * The kernel's pmap is statically allocated so we don't have to use 308 * pmap_create, which is unlikely to work correctly at this part of 309 * the boot sequence (XXX and which no longer exists). 310 */ 311 kernel_pmap = &kernel_pmap_store; 312 313 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 314 kernel_pmap->pm_count = 1; 315 kernel_pmap->pm_active = -1; /* don't allow deactivation */ 316 TAILQ_INIT(&kernel_pmap->pm_pvlist); 317 LIST_INIT(&allpmaps); 318 LIST_INSERT_HEAD(&allpmaps, kernel_pmap, pm_list); 319 nkpt = NKPT; 320 321 /* 322 * Reserve some special page table entries/VA space for temporary 323 * mapping of pages. 324 */ 325#define SYSMAP(c, p, v, n) \ 326 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 327 328 va = virtual_avail; 329 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 330 331 /* 332 * CMAP1/CMAP2 are used for zeroing and copying pages. 333 */ 334 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 335 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 336 337 /* 338 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 339 * XXX ptmmap is not used. 340 */ 341 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 342 343 /* 344 * msgbufp is used to map the system message buffer. 345 * XXX msgbufmap is not used. 346 */ 347 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 348 atop(round_page(MSGBUF_SIZE))) 349 350#if !defined(SMP) 351 /* 352 * ptemap is used for pmap_pte_quick 353 */ 354 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 355#endif 356 357 virtual_avail = va; 358 359 *(int *) CMAP1 = *(int *) CMAP2 = 0; 360 *(int *) PTD = 0; 361 362 363 pgeflag = 0; 364#if !defined(SMP) 365 if (cpu_feature & CPUID_PGE) { 366 pgeflag = PG_G; 367 } 368#endif 369 370/* 371 * Initialize the 4MB page size flag 372 */ 373 pseflag = 0; 374/* 375 * The 4MB page version of the initial 376 * kernel page mapping. 377 */ 378 pdir4mb = 0; 379 380#if !defined(DISABLE_PSE) 381 if (cpu_feature & CPUID_PSE) { 382 unsigned ptditmp; 383 /* 384 * Note that we have enabled PSE mode 385 */ 386 pseflag = PG_PS; 387 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE)); 388 ptditmp &= ~(NBPDR - 1); 389 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 390 pdir4mb = ptditmp; 391 392#if !defined(SMP) 393 /* 394 * Enable the PSE mode. 395 */ 396 load_cr4(rcr4() | CR4_PSE); 397 398 /* 399 * We can do the mapping here for the single processor 400 * case. We simply ignore the old page table page from 401 * now on. 402 */ 403 /* 404 * For SMP, we still need 4K pages to bootstrap APs, 405 * PSE will be enabled as soon as all APs are up. 406 */ 407 PTD[KPTDI] = (pd_entry_t) ptditmp; 408 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 409 invltlb(); 410#endif 411 } 412#endif 413 414#ifdef SMP 415 if (cpu_apic_address == 0) 416 panic("pmap_bootstrap: no local apic! (non-SMP hardware?)"); 417 418 /* local apic is mapped on last page */ 419 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag | 420 (cpu_apic_address & PG_FRAME)); 421 422 for (i = 0; i < mp_napics; i++) { 423 for (j = 0; j < mp_napics; j++) { 424 /* same page frame as a previous IO apic? */ 425 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 426 (io_apic_address[i] & PG_FRAME)) { 427 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 428 + (NPTEPG-2-j) * PAGE_SIZE 429 + (io_apic_address[i] & PAGE_MASK)); 430 break; 431 } 432 /* use this slot if available */ 433 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 0) { 434 SMPpt[NPTEPG-2-j] = (pt_entry_t)(PG_V | PG_RW | 435 pgeflag | (io_apic_address[i] & PG_FRAME)); 436 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 437 + (NPTEPG-2-j) * PAGE_SIZE 438 + (io_apic_address[i] & PAGE_MASK)); 439 break; 440 } 441 } 442 } 443 444 /* BSP does this itself, AP's get it pre-set */ 445 gd = &SMP_prvspace[0].globaldata; 446 gd->gd_prv_CMAP1 = &SMPpt[1]; 447 gd->gd_prv_CMAP2 = &SMPpt[2]; 448 gd->gd_prv_CMAP3 = &SMPpt[3]; 449 gd->gd_prv_PMAP1 = &SMPpt[4]; 450 gd->gd_prv_CADDR1 = SMP_prvspace[0].CPAGE1; 451 gd->gd_prv_CADDR2 = SMP_prvspace[0].CPAGE2; 452 gd->gd_prv_CADDR3 = SMP_prvspace[0].CPAGE3; 453 gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[0].PPAGE1; 454#endif 455 456 invltlb(); 457} 458 459#ifdef SMP 460/* 461 * Set 4mb pdir for mp startup 462 */ 463void 464pmap_set_opt(void) 465{ 466 if (pseflag && (cpu_feature & CPUID_PSE)) { 467 load_cr4(rcr4() | CR4_PSE); 468 if (pdir4mb && cpuid == 0) { /* only on BSP */ 469 kernel_pmap->pm_pdir[KPTDI] = 470 PTD[KPTDI] = (pd_entry_t)pdir4mb; 471 cpu_invltlb(); 472 } 473 } 474} 475#endif 476 477/* 478 * Initialize the pmap module. 479 * Called by vm_init, to initialize any structures that the pmap 480 * system needs to map virtual memory. 481 * pmap_init has been enhanced to support in a fairly consistant 482 * way, discontiguous physical memory. 483 */ 484void 485pmap_init(phys_start, phys_end) 486 vm_offset_t phys_start, phys_end; 487{ 488 int i; 489 int initial_pvs; 490 491 /* 492 * object for kernel page table pages 493 */ 494 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE); 495 496 /* 497 * Allocate memory for random pmap data structures. Includes the 498 * pv_head_table. 499 */ 500 501 for(i = 0; i < vm_page_array_size; i++) { 502 vm_page_t m; 503 504 m = &vm_page_array[i]; 505 TAILQ_INIT(&m->md.pv_list); 506 m->md.pv_list_count = 0; 507 } 508 509 /* 510 * init the pv free list 511 */ 512 initial_pvs = vm_page_array_size; 513 if (initial_pvs < MINPV) 514 initial_pvs = MINPV; 515 pvzone = &pvzone_store; 516 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 517 initial_pvs * sizeof (struct pv_entry)); 518 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, 519 vm_page_array_size); 520 521 /* 522 * Now it is safe to enable pv_table recording. 523 */ 524 pmap_initialized = TRUE; 525} 526 527/* 528 * Initialize the address space (zone) for the pv_entries. Set a 529 * high water mark so that the system can recover from excessive 530 * numbers of pv entries. 531 */ 532void 533pmap_init2() 534{ 535 pv_entry_max = PMAP_SHPGPERPROC * maxproc + vm_page_array_size; 536 pv_entry_high_water = 9 * (pv_entry_max / 10); 537 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 538} 539 540 541/*************************************************** 542 * Low level helper routines..... 543 ***************************************************/ 544 545#if defined(PMAP_DIAGNOSTIC) 546 547/* 548 * This code checks for non-writeable/modified pages. 549 * This should be an invalid condition. 550 */ 551static int 552pmap_nw_modified(pt_entry_t ptea) 553{ 554 int pte; 555 556 pte = (int) ptea; 557 558 if ((pte & (PG_M|PG_RW)) == PG_M) 559 return 1; 560 else 561 return 0; 562} 563#endif 564 565 566/* 567 * this routine defines the region(s) of memory that should 568 * not be tested for the modified bit. 569 */ 570static PMAP_INLINE int 571pmap_track_modified(vm_offset_t va) 572{ 573 if ((va < clean_sva) || (va >= clean_eva)) 574 return 1; 575 else 576 return 0; 577} 578 579static PMAP_INLINE void 580invltlb_1pg(vm_offset_t va) 581{ 582#if defined(I386_CPU) 583 if (cpu_class == CPUCLASS_386) { 584 invltlb(); 585 } else 586#endif 587 { 588 invlpg(va); 589 } 590} 591 592static __inline void 593pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va) 594{ 595#if defined(SMP) 596 if (pmap->pm_active & (1 << cpuid)) 597 cpu_invlpg((void *)va); 598 if (pmap->pm_active & other_cpus) 599 smp_invltlb(); 600#else 601 if (pmap->pm_active) 602 invltlb_1pg(va); 603#endif 604} 605 606static __inline void 607pmap_TLB_invalidate_all(pmap_t pmap) 608{ 609#if defined(SMP) 610 if (pmap->pm_active & (1 << cpuid)) 611 cpu_invltlb(); 612 if (pmap->pm_active & other_cpus) 613 smp_invltlb(); 614#else 615 if (pmap->pm_active) 616 invltlb(); 617#endif 618} 619 620static unsigned * 621get_ptbase(pmap) 622 pmap_t pmap; 623{ 624 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 625 626 /* are we current address space or kernel? */ 627 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 628 return (unsigned *) PTmap; 629 } 630 /* otherwise, we are alternate address space */ 631 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 632 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 633#if defined(SMP) 634 /* The page directory is not shared between CPUs */ 635 cpu_invltlb(); 636#else 637 invltlb(); 638#endif 639 } 640 return (unsigned *) APTmap; 641} 642 643/* 644 * Super fast pmap_pte routine best used when scanning 645 * the pv lists. This eliminates many coarse-grained 646 * invltlb calls. Note that many of the pv list 647 * scans are across different pmaps. It is very wasteful 648 * to do an entire invltlb for checking a single mapping. 649 */ 650 651static unsigned * 652pmap_pte_quick(pmap, va) 653 register pmap_t pmap; 654 vm_offset_t va; 655{ 656 unsigned pde, newpf; 657 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) { 658 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 659 unsigned index = i386_btop(va); 660 /* are we current address space or kernel? */ 661 if ((pmap == kernel_pmap) || 662 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 663 return (unsigned *) PTmap + index; 664 } 665 newpf = pde & PG_FRAME; 666#ifdef SMP 667 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) { 668 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V; 669 cpu_invlpg(prv_PADDR1); 670 } 671 return (unsigned *)(prv_PADDR1 + (index & (NPTEPG - 1))); 672#else 673 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 674 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 675 invltlb_1pg((vm_offset_t) PADDR1); 676 } 677 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 678#endif 679 } 680 return (0); 681} 682 683/* 684 * Routine: pmap_extract 685 * Function: 686 * Extract the physical page address associated 687 * with the given map/virtual_address pair. 688 */ 689vm_offset_t 690pmap_extract(pmap, va) 691 register pmap_t pmap; 692 vm_offset_t va; 693{ 694 vm_offset_t rtval; 695 vm_offset_t pdirindex; 696 pdirindex = va >> PDRSHIFT; 697 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 698 unsigned *pte; 699 if ((rtval & PG_PS) != 0) { 700 rtval &= ~(NBPDR - 1); 701 rtval |= va & (NBPDR - 1); 702 return rtval; 703 } 704 pte = get_ptbase(pmap) + i386_btop(va); 705 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 706 return rtval; 707 } 708 return 0; 709 710} 711 712/*************************************************** 713 * Low level mapping routines..... 714 ***************************************************/ 715 716/* 717 * add a wired page to the kva 718 * note that in order for the mapping to take effect -- you 719 * should do a invltlb after doing the pmap_kenter... 720 */ 721PMAP_INLINE void 722pmap_kenter(va, pa) 723 vm_offset_t va; 724 register vm_offset_t pa; 725{ 726 register unsigned *pte; 727 unsigned npte, opte; 728 729 npte = pa | PG_RW | PG_V | pgeflag; 730 pte = (unsigned *)vtopte(va); 731 opte = *pte; 732 *pte = npte; 733 /*if (opte)*/ 734 invltlb_1pg(va); /* XXX what about SMP? */ 735} 736 737/* 738 * remove a page from the kernel pagetables 739 */ 740PMAP_INLINE void 741pmap_kremove(va) 742 vm_offset_t va; 743{ 744 register unsigned *pte; 745 746 pte = (unsigned *)vtopte(va); 747 *pte = 0; 748 invltlb_1pg(va); /* XXX what about SMP? */ 749} 750 751/* 752 * Used to map a range of physical addresses into kernel 753 * virtual address space. 754 * 755 * For now, VM is already on, we only need to map the 756 * specified memory. 757 */ 758vm_offset_t 759pmap_map(virt, start, end, prot) 760 vm_offset_t virt; 761 vm_offset_t start; 762 vm_offset_t end; 763 int prot; 764{ 765 while (start < end) { 766 pmap_kenter(virt, start); 767 virt += PAGE_SIZE; 768 start += PAGE_SIZE; 769 } 770 return (virt); 771} 772 773 774/* 775 * Add a list of wired pages to the kva 776 * this routine is only used for temporary 777 * kernel mappings that do not need to have 778 * page modification or references recorded. 779 * Note that old mappings are simply written 780 * over. The page *must* be wired. 781 */ 782void 783pmap_qenter(va, m, count) 784 vm_offset_t va; 785 vm_page_t *m; 786 int count; 787{ 788 int i; 789 790 for (i = 0; i < count; i++) { 791 vm_offset_t tva = va + i * PAGE_SIZE; 792 pmap_kenter(tva, VM_PAGE_TO_PHYS(m[i])); 793 } 794} 795 796/* 797 * this routine jerks page mappings from the 798 * kernel -- it is meant only for temporary mappings. 799 */ 800void 801pmap_qremove(va, count) 802 vm_offset_t va; 803 int count; 804{ 805 vm_offset_t end_va; 806 807 end_va = va + count*PAGE_SIZE; 808 809 while (va < end_va) { 810 unsigned *pte; 811 812 pte = (unsigned *)vtopte(va); 813 *pte = 0; 814#ifdef SMP 815 cpu_invlpg((void *)va); 816#else 817 invltlb_1pg(va); 818#endif 819 va += PAGE_SIZE; 820 } 821#ifdef SMP 822 smp_invltlb(); 823#endif 824} 825 826static vm_page_t 827pmap_page_lookup(object, pindex) 828 vm_object_t object; 829 vm_pindex_t pindex; 830{ 831 vm_page_t m; 832retry: 833 m = vm_page_lookup(object, pindex); 834 if (m && vm_page_sleep_busy(m, FALSE, "pplookp")) 835 goto retry; 836 return m; 837} 838 839/* 840 * Create the UPAGES for a new process. 841 * This routine directly affects the fork perf for a process. 842 */ 843void 844pmap_new_proc(p) 845 struct proc *p; 846{ 847 int i, updateneeded; 848 vm_object_t upobj; 849 vm_page_t m; 850 struct user *up; 851 unsigned *ptek, oldpte; 852 853 /* 854 * allocate object for the upages 855 */ 856 if ((upobj = p->p_upages_obj) == NULL) { 857 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 858 p->p_upages_obj = upobj; 859 } 860 861 /* get a kernel virtual address for the UPAGES for this proc */ 862 if ((up = p->p_addr) == NULL) { 863 up = (struct user *) kmem_alloc_nofault(kernel_map, 864 UPAGES * PAGE_SIZE); 865 if (up == NULL) 866 panic("pmap_new_proc: u_map allocation failed"); 867 p->p_addr = up; 868 } 869 870 ptek = (unsigned *) vtopte((vm_offset_t) up); 871 872 updateneeded = 0; 873 for(i=0;i<UPAGES;i++) { 874 /* 875 * Get a kernel stack page 876 */ 877 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 878 879 /* 880 * Wire the page 881 */ 882 m->wire_count++; 883 cnt.v_wire_count++; 884 885 oldpte = *(ptek + i); 886 /* 887 * Enter the page into the kernel address space. 888 */ 889 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 890 if (oldpte) { 891 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) { 892 invlpg((vm_offset_t) up + i * PAGE_SIZE); 893 } else { 894 updateneeded = 1; 895 } 896 } 897 898 vm_page_wakeup(m); 899 vm_page_flag_clear(m, PG_ZERO); 900 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 901 m->valid = VM_PAGE_BITS_ALL; 902 } 903 if (updateneeded) 904 invltlb(); 905} 906 907/* 908 * Dispose the UPAGES for a process that has exited. 909 * This routine directly impacts the exit perf of a process. 910 */ 911void 912pmap_dispose_proc(p) 913 struct proc *p; 914{ 915 int i; 916 vm_object_t upobj; 917 vm_page_t m; 918 unsigned *ptek, oldpte; 919 920 upobj = p->p_upages_obj; 921 922 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 923 for(i=0;i<UPAGES;i++) { 924 925 if ((m = vm_page_lookup(upobj, i)) == NULL) 926 panic("pmap_dispose_proc: upage already missing???"); 927 928 vm_page_busy(m); 929 930 oldpte = *(ptek + i); 931 *(ptek + i) = 0; 932 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) 933 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 934 vm_page_unwire(m, 0); 935 vm_page_free(m); 936 } 937#if defined(I386_CPU) 938 if (cpu_class <= CPUCLASS_386) 939 invltlb(); 940#endif 941} 942 943/* 944 * Allow the UPAGES 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_page_t m; 953 954 upobj = p->p_upages_obj; 955 /* 956 * let the upages be paged 957 */ 958 for(i=0;i<UPAGES;i++) { 959 if ((m = vm_page_lookup(upobj, i)) == NULL) 960 panic("pmap_swapout_proc: upage already missing???"); 961 vm_page_dirty(m); 962 vm_page_unwire(m, 0); 963 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 964 } 965} 966 967/* 968 * Bring the UPAGES 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_page_t m; 977 978 upobj = p->p_upages_obj; 979 for(i=0;i<UPAGES;i++) { 980 981 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 982 983 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 984 VM_PAGE_TO_PHYS(m)); 985 986 if (m->valid != VM_PAGE_BITS_ALL) { 987 rv = vm_pager_get_pages(upobj, &m, 1, 0); 988 if (rv != VM_PAGER_OK) 989 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 990 m = vm_page_lookup(upobj, i); 991 m->valid = VM_PAGE_BITS_ALL; 992 } 993 994 vm_page_wire(m); 995 vm_page_wakeup(m); 996 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 997 } 998} 999 1000/*************************************************** 1001 * Page table page management routines..... 1002 ***************************************************/ 1003 1004/* 1005 * This routine unholds page table pages, and if the hold count 1006 * drops to zero, then it decrements the wire count. 1007 */ 1008static int 1009_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1010 1011 while (vm_page_sleep_busy(m, FALSE, "pmuwpt")) 1012 ; 1013 1014 if (m->hold_count == 0) { 1015 vm_offset_t pteva; 1016 /* 1017 * unmap the page table page 1018 */ 1019 pmap->pm_pdir[m->pindex] = 0; 1020 --pmap->pm_stats.resident_count; 1021 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1022 (((unsigned) PTDpde) & PG_FRAME)) { 1023 /* 1024 * Do a invltlb to make the invalidated mapping 1025 * take effect immediately. 1026 */ 1027 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1028 pmap_TLB_invalidate(pmap, pteva); 1029 } 1030 1031 if (pmap->pm_ptphint == m) 1032 pmap->pm_ptphint = NULL; 1033 1034 /* 1035 * If the page is finally unwired, simply free it. 1036 */ 1037 --m->wire_count; 1038 if (m->wire_count == 0) { 1039 1040 vm_page_flash(m); 1041 vm_page_busy(m); 1042 vm_page_free_zero(m); 1043 --cnt.v_wire_count; 1044 } 1045 return 1; 1046 } 1047 return 0; 1048} 1049 1050static PMAP_INLINE int 1051pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) 1052{ 1053 vm_page_unhold(m); 1054 if (m->hold_count == 0) 1055 return _pmap_unwire_pte_hold(pmap, m); 1056 else 1057 return 0; 1058} 1059 1060/* 1061 * After removing a page table entry, this routine is used to 1062 * conditionally free the page, and manage the hold/wire counts. 1063 */ 1064static int 1065pmap_unuse_pt(pmap, va, mpte) 1066 pmap_t pmap; 1067 vm_offset_t va; 1068 vm_page_t mpte; 1069{ 1070 unsigned ptepindex; 1071 if (va >= UPT_MIN_ADDRESS) 1072 return 0; 1073 1074 if (mpte == NULL) { 1075 ptepindex = (va >> PDRSHIFT); 1076 if (pmap->pm_ptphint && 1077 (pmap->pm_ptphint->pindex == ptepindex)) { 1078 mpte = pmap->pm_ptphint; 1079 } else { 1080 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1081 pmap->pm_ptphint = mpte; 1082 } 1083 } 1084 1085 return pmap_unwire_pte_hold(pmap, mpte); 1086} 1087 1088void 1089pmap_pinit0(pmap) 1090 struct pmap *pmap; 1091{ 1092 pmap->pm_pdir = 1093 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1094 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1095 pmap->pm_count = 1; 1096 pmap->pm_active = 0; 1097 pmap->pm_ptphint = NULL; 1098 TAILQ_INIT(&pmap->pm_pvlist); 1099 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1100 LIST_INSERT_HEAD(&allpmaps, pmap, pm_list); 1101} 1102 1103/* 1104 * Initialize a preallocated and zeroed pmap structure, 1105 * such as one in a vmspace structure. 1106 */ 1107void 1108pmap_pinit(pmap) 1109 register struct pmap *pmap; 1110{ 1111 vm_page_t ptdpg; 1112 1113 /* 1114 * No need to allocate page table space yet but we do need a valid 1115 * page directory table. 1116 */ 1117 if (pmap->pm_pdir == NULL) 1118 pmap->pm_pdir = 1119 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1120 1121 /* 1122 * allocate object for the ptes 1123 */ 1124 if (pmap->pm_pteobj == NULL) 1125 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1126 1127 /* 1128 * allocate the page directory page 1129 */ 1130 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1131 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1132 1133 ptdpg->wire_count = 1; 1134 ++cnt.v_wire_count; 1135 1136 1137 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/ 1138 ptdpg->valid = VM_PAGE_BITS_ALL; 1139 1140 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1141 if ((ptdpg->flags & PG_ZERO) == 0) 1142 bzero(pmap->pm_pdir, PAGE_SIZE); 1143 1144 LIST_INSERT_HEAD(&allpmaps, pmap, pm_list); 1145 /* Wire in kernel global address entries. */ 1146 /* XXX copies current process, does not fill in MPPTDI */ 1147 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1148#ifdef SMP 1149 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI]; 1150#endif 1151 1152 /* install self-referential address mapping entry */ 1153 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1154 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1155 1156 pmap->pm_count = 1; 1157 pmap->pm_active = 0; 1158 pmap->pm_ptphint = NULL; 1159 TAILQ_INIT(&pmap->pm_pvlist); 1160 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1161} 1162 1163/* 1164 * Wire in kernel global address entries. To avoid a race condition 1165 * between pmap initialization and pmap_growkernel, this procedure 1166 * should be called after the vmspace is attached to the process 1167 * but before this pmap is activated. 1168 */ 1169void 1170pmap_pinit2(pmap) 1171 struct pmap *pmap; 1172{ 1173 /* XXX: Remove this stub when no longer called */ 1174} 1175 1176static int 1177pmap_release_free_page(pmap, p) 1178 struct pmap *pmap; 1179 vm_page_t p; 1180{ 1181 unsigned *pde = (unsigned *) pmap->pm_pdir; 1182 /* 1183 * This code optimizes the case of freeing non-busy 1184 * page-table pages. Those pages are zero now, and 1185 * might as well be placed directly into the zero queue. 1186 */ 1187 if (vm_page_sleep_busy(p, FALSE, "pmaprl")) 1188 return 0; 1189 1190 vm_page_busy(p); 1191 1192 /* 1193 * Remove the page table page from the processes address space. 1194 */ 1195 pde[p->pindex] = 0; 1196 pmap->pm_stats.resident_count--; 1197 1198 if (p->hold_count) { 1199 panic("pmap_release: freeing held page table page"); 1200 } 1201 /* 1202 * Page directory pages need to have the kernel 1203 * stuff cleared, so they can go into the zero queue also. 1204 */ 1205 if (p->pindex == PTDPTDI) { 1206 bzero(pde + KPTDI, nkpt * PTESIZE); 1207#ifdef SMP 1208 pde[MPPTDI] = 0; 1209#endif 1210 pde[APTDPTDI] = 0; 1211 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1212 } 1213 1214 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1215 pmap->pm_ptphint = NULL; 1216 1217 p->wire_count--; 1218 cnt.v_wire_count--; 1219 vm_page_free_zero(p); 1220 return 1; 1221} 1222 1223/* 1224 * this routine is called if the page table page is not 1225 * mapped correctly. 1226 */ 1227static vm_page_t 1228_pmap_allocpte(pmap, ptepindex) 1229 pmap_t pmap; 1230 unsigned ptepindex; 1231{ 1232 vm_offset_t pteva, ptepa; 1233 vm_page_t m; 1234 1235 /* 1236 * Find or fabricate a new pagetable page 1237 */ 1238 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1239 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1240 1241 KASSERT(m->queue == PQ_NONE, 1242 ("_pmap_allocpte: %p->queue != PQ_NONE", m)); 1243 1244 if (m->wire_count == 0) 1245 cnt.v_wire_count++; 1246 m->wire_count++; 1247 1248 /* 1249 * Increment the hold count for the page table page 1250 * (denoting a new mapping.) 1251 */ 1252 m->hold_count++; 1253 1254 /* 1255 * Map the pagetable page into the process address space, if 1256 * it isn't already there. 1257 */ 1258 1259 pmap->pm_stats.resident_count++; 1260 1261 ptepa = VM_PAGE_TO_PHYS(m); 1262 pmap->pm_pdir[ptepindex] = 1263 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1264 1265 /* 1266 * Set the page table hint 1267 */ 1268 pmap->pm_ptphint = m; 1269 1270 /* 1271 * Try to use the new mapping, but if we cannot, then 1272 * do it with the routine that maps the page explicitly. 1273 */ 1274 if ((m->flags & PG_ZERO) == 0) { 1275 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1276 (((unsigned) PTDpde) & PG_FRAME)) { 1277 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1278 bzero((caddr_t) pteva, PAGE_SIZE); 1279 } else { 1280 pmap_zero_page(ptepa); 1281 } 1282 } 1283 1284 m->valid = VM_PAGE_BITS_ALL; 1285 vm_page_flag_clear(m, PG_ZERO); 1286 vm_page_flag_set(m, PG_MAPPED); 1287 vm_page_wakeup(m); 1288 1289 return m; 1290} 1291 1292static vm_page_t 1293pmap_allocpte(pmap, va) 1294 pmap_t pmap; 1295 vm_offset_t va; 1296{ 1297 unsigned ptepindex; 1298 vm_offset_t ptepa; 1299 vm_page_t m; 1300 1301 /* 1302 * Calculate pagetable page index 1303 */ 1304 ptepindex = va >> PDRSHIFT; 1305 1306 /* 1307 * Get the page directory entry 1308 */ 1309 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1310 1311 /* 1312 * This supports switching from a 4MB page to a 1313 * normal 4K page. 1314 */ 1315 if (ptepa & PG_PS) { 1316 pmap->pm_pdir[ptepindex] = 0; 1317 ptepa = 0; 1318 invltlb(); 1319 } 1320 1321 /* 1322 * If the page table page is mapped, we just increment the 1323 * hold count, and activate it. 1324 */ 1325 if (ptepa) { 1326 /* 1327 * In order to get the page table page, try the 1328 * hint first. 1329 */ 1330 if (pmap->pm_ptphint && 1331 (pmap->pm_ptphint->pindex == ptepindex)) { 1332 m = pmap->pm_ptphint; 1333 } else { 1334 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1335 pmap->pm_ptphint = m; 1336 } 1337 m->hold_count++; 1338 return m; 1339 } 1340 /* 1341 * Here if the pte page isn't mapped, or if it has been deallocated. 1342 */ 1343 return _pmap_allocpte(pmap, ptepindex); 1344} 1345 1346 1347/*************************************************** 1348* Pmap allocation/deallocation routines. 1349 ***************************************************/ 1350 1351/* 1352 * Release any resources held by the given physical map. 1353 * Called when a pmap initialized by pmap_pinit is being released. 1354 * Should only be called if the map contains no valid mappings. 1355 */ 1356void 1357pmap_release(pmap) 1358 register struct pmap *pmap; 1359{ 1360 vm_page_t p,n,ptdpg; 1361 vm_object_t object = pmap->pm_pteobj; 1362 int curgeneration; 1363 1364#if defined(DIAGNOSTIC) 1365 if (object->ref_count != 1) 1366 panic("pmap_release: pteobj reference count != 1"); 1367#endif 1368 1369 ptdpg = NULL; 1370 LIST_REMOVE(pmap, pm_list); 1371retry: 1372 curgeneration = object->generation; 1373 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1374 n = TAILQ_NEXT(p, listq); 1375 if (p->pindex == PTDPTDI) { 1376 ptdpg = p; 1377 continue; 1378 } 1379 while (1) { 1380 if (!pmap_release_free_page(pmap, p) && 1381 (object->generation != curgeneration)) 1382 goto retry; 1383 } 1384 } 1385 1386 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1387 goto retry; 1388} 1389 1390/* 1391 * grow the number of kernel page table entries, if needed 1392 */ 1393void 1394pmap_growkernel(vm_offset_t addr) 1395{ 1396 struct pmap *pmap; 1397 int s; 1398 vm_offset_t ptppaddr; 1399 vm_page_t nkpg; 1400 pd_entry_t newpdir; 1401 1402 s = splhigh(); 1403 if (kernel_vm_end == 0) { 1404 kernel_vm_end = KERNBASE; 1405 nkpt = 0; 1406 while (pdir_pde(PTD, kernel_vm_end)) { 1407 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1408 nkpt++; 1409 } 1410 } 1411 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1412 while (kernel_vm_end < addr) { 1413 if (pdir_pde(PTD, kernel_vm_end)) { 1414 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1415 continue; 1416 } 1417 1418 /* 1419 * This index is bogus, but out of the way 1420 */ 1421 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1422 if (!nkpg) 1423 panic("pmap_growkernel: no memory to grow kernel"); 1424 1425 nkpt++; 1426 1427 vm_page_wire(nkpg); 1428 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1429 pmap_zero_page(ptppaddr); 1430 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1431 pdir_pde(PTD, kernel_vm_end) = newpdir; 1432 1433 LIST_FOREACH(pmap, &allpmaps, pm_list) { 1434 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1435 } 1436 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1437 } 1438 splx(s); 1439} 1440 1441/* 1442 * Retire the given physical map from service. 1443 * Should only be called if the map contains 1444 * no valid mappings. 1445 */ 1446void 1447pmap_destroy(pmap) 1448 register pmap_t pmap; 1449{ 1450 int count; 1451 1452 if (pmap == NULL) 1453 return; 1454 1455 count = --pmap->pm_count; 1456 if (count == 0) { 1457 pmap_release(pmap); 1458 panic("destroying a pmap is not yet implemented"); 1459 } 1460} 1461 1462/* 1463 * Add a reference to the specified pmap. 1464 */ 1465void 1466pmap_reference(pmap) 1467 pmap_t pmap; 1468{ 1469 if (pmap != NULL) { 1470 pmap->pm_count++; 1471 } 1472} 1473 1474/*************************************************** 1475* page management routines. 1476 ***************************************************/ 1477 1478/* 1479 * free the pv_entry back to the free list 1480 */ 1481static PMAP_INLINE void 1482free_pv_entry(pv) 1483 pv_entry_t pv; 1484{ 1485 pv_entry_count--; 1486 zfreei(pvzone, pv); 1487} 1488 1489/* 1490 * get a new pv_entry, allocating a block from the system 1491 * when needed. 1492 * the memory allocation is performed bypassing the malloc code 1493 * because of the possibility of allocations at interrupt time. 1494 */ 1495static pv_entry_t 1496get_pv_entry(void) 1497{ 1498 pv_entry_count++; 1499 if (pv_entry_high_water && 1500 (pv_entry_count > pv_entry_high_water) && 1501 (pmap_pagedaemon_waken == 0)) { 1502 pmap_pagedaemon_waken = 1; 1503 wakeup (&vm_pages_needed); 1504 } 1505 return zalloci(pvzone); 1506} 1507 1508/* 1509 * This routine is very drastic, but can save the system 1510 * in a pinch. 1511 */ 1512void 1513pmap_collect() 1514{ 1515 int i; 1516 vm_page_t m; 1517 static int warningdone=0; 1518 1519 if (pmap_pagedaemon_waken == 0) 1520 return; 1521 1522 if (warningdone < 5) { 1523 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1524 warningdone++; 1525 } 1526 1527 for(i = 0; i < vm_page_array_size; i++) { 1528 m = &vm_page_array[i]; 1529 if (m->wire_count || m->hold_count || m->busy || 1530 (m->flags & PG_BUSY)) 1531 continue; 1532 pmap_remove_all(m); 1533 } 1534 pmap_pagedaemon_waken = 0; 1535} 1536 1537 1538/* 1539 * If it is the first entry on the list, it is actually 1540 * in the header and we must copy the following entry up 1541 * to the header. Otherwise we must search the list for 1542 * the entry. In either case we free the now unused entry. 1543 */ 1544 1545static int 1546pmap_remove_entry(pmap, m, va) 1547 struct pmap *pmap; 1548 vm_page_t m; 1549 vm_offset_t va; 1550{ 1551 pv_entry_t pv; 1552 int rtval; 1553 int s; 1554 1555 s = splvm(); 1556 if (m->md.pv_list_count < pmap->pm_stats.resident_count) { 1557 for (pv = TAILQ_FIRST(&m->md.pv_list); 1558 pv; 1559 pv = TAILQ_NEXT(pv, pv_list)) { 1560 if (pmap == pv->pv_pmap && va == pv->pv_va) 1561 break; 1562 } 1563 } else { 1564 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1565 pv; 1566 pv = TAILQ_NEXT(pv, pv_plist)) { 1567 if (va == pv->pv_va) 1568 break; 1569 } 1570 } 1571 1572 rtval = 0; 1573 if (pv) { 1574 1575 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1576 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 1577 m->md.pv_list_count--; 1578 if (TAILQ_FIRST(&m->md.pv_list) == NULL) 1579 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 1580 1581 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1582 free_pv_entry(pv); 1583 } 1584 1585 splx(s); 1586 return rtval; 1587} 1588 1589/* 1590 * Create a pv entry for page at pa for 1591 * (pmap, va). 1592 */ 1593static void 1594pmap_insert_entry(pmap, va, mpte, m) 1595 pmap_t pmap; 1596 vm_offset_t va; 1597 vm_page_t mpte; 1598 vm_page_t m; 1599{ 1600 1601 int s; 1602 pv_entry_t pv; 1603 1604 s = splvm(); 1605 pv = get_pv_entry(); 1606 pv->pv_va = va; 1607 pv->pv_pmap = pmap; 1608 pv->pv_ptem = mpte; 1609 1610 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1611 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); 1612 m->md.pv_list_count++; 1613 1614 splx(s); 1615} 1616 1617/* 1618 * pmap_remove_pte: do the things to unmap a page in a process 1619 */ 1620static int 1621pmap_remove_pte(pmap, ptq, va) 1622 struct pmap *pmap; 1623 unsigned *ptq; 1624 vm_offset_t va; 1625{ 1626 unsigned oldpte; 1627 vm_page_t m; 1628 1629 oldpte = loadandclear(ptq); 1630 if (oldpte & PG_W) 1631 pmap->pm_stats.wired_count -= 1; 1632 /* 1633 * Machines that don't support invlpg, also don't support 1634 * PG_G. 1635 */ 1636 if (oldpte & PG_G) 1637 invlpg(va); 1638 pmap->pm_stats.resident_count -= 1; 1639 if (oldpte & PG_MANAGED) { 1640 m = PHYS_TO_VM_PAGE(oldpte); 1641 if (oldpte & PG_M) { 1642#if defined(PMAP_DIAGNOSTIC) 1643 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1644 printf( 1645 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n", 1646 va, oldpte); 1647 } 1648#endif 1649 if (pmap_track_modified(va)) 1650 vm_page_dirty(m); 1651 } 1652 if (oldpte & PG_A) 1653 vm_page_flag_set(m, PG_REFERENCED); 1654 return pmap_remove_entry(pmap, m, va); 1655 } else { 1656 return pmap_unuse_pt(pmap, va, NULL); 1657 } 1658 1659 return 0; 1660} 1661 1662/* 1663 * Remove a single page from a process address space 1664 */ 1665static void 1666pmap_remove_page(pmap, va) 1667 struct pmap *pmap; 1668 register vm_offset_t va; 1669{ 1670 register unsigned *ptq; 1671 1672 /* 1673 * if there is no pte for this address, just skip it!!! 1674 */ 1675 if (*pmap_pde(pmap, va) == 0) { 1676 return; 1677 } 1678 1679 /* 1680 * get a local va for mappings for this pmap. 1681 */ 1682 ptq = get_ptbase(pmap) + i386_btop(va); 1683 if (*ptq) { 1684 (void) pmap_remove_pte(pmap, ptq, va); 1685 pmap_TLB_invalidate(pmap, va); 1686 } 1687 return; 1688} 1689 1690/* 1691 * Remove the given range of addresses from the specified map. 1692 * 1693 * It is assumed that the start and end are properly 1694 * rounded to the page size. 1695 */ 1696void 1697pmap_remove(pmap, sva, eva) 1698 struct pmap *pmap; 1699 register vm_offset_t sva; 1700 register vm_offset_t eva; 1701{ 1702 register unsigned *ptbase; 1703 vm_offset_t pdnxt; 1704 vm_offset_t ptpaddr; 1705 vm_offset_t sindex, eindex; 1706 int anyvalid; 1707 1708 if (pmap == NULL) 1709 return; 1710 1711 if (pmap->pm_stats.resident_count == 0) 1712 return; 1713 1714 /* 1715 * special handling of removing one page. a very 1716 * common operation and easy to short circuit some 1717 * code. 1718 */ 1719 if (((sva + PAGE_SIZE) == eva) && 1720 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1721 pmap_remove_page(pmap, sva); 1722 return; 1723 } 1724 1725 anyvalid = 0; 1726 1727 /* 1728 * Get a local virtual address for the mappings that are being 1729 * worked with. 1730 */ 1731 ptbase = get_ptbase(pmap); 1732 1733 sindex = i386_btop(sva); 1734 eindex = i386_btop(eva); 1735 1736 for (; sindex < eindex; sindex = pdnxt) { 1737 unsigned pdirindex; 1738 1739 /* 1740 * Calculate index for next page table. 1741 */ 1742 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1743 if (pmap->pm_stats.resident_count == 0) 1744 break; 1745 1746 pdirindex = sindex / NPDEPG; 1747 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1748 pmap->pm_pdir[pdirindex] = 0; 1749 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1750 anyvalid++; 1751 continue; 1752 } 1753 1754 /* 1755 * Weed out invalid mappings. Note: we assume that the page 1756 * directory table is always allocated, and in kernel virtual. 1757 */ 1758 if (ptpaddr == 0) 1759 continue; 1760 1761 /* 1762 * Limit our scan to either the end of the va represented 1763 * by the current page table page, or to the end of the 1764 * range being removed. 1765 */ 1766 if (pdnxt > eindex) { 1767 pdnxt = eindex; 1768 } 1769 1770 for ( ;sindex != pdnxt; sindex++) { 1771 vm_offset_t va; 1772 if (ptbase[sindex] == 0) { 1773 continue; 1774 } 1775 va = i386_ptob(sindex); 1776 1777 anyvalid++; 1778 if (pmap_remove_pte(pmap, 1779 ptbase + sindex, va)) 1780 break; 1781 } 1782 } 1783 1784 if (anyvalid) 1785 pmap_TLB_invalidate_all(pmap); 1786} 1787 1788/* 1789 * Routine: pmap_remove_all 1790 * Function: 1791 * Removes this physical page from 1792 * all physical maps in which it resides. 1793 * Reflects back modify bits to the pager. 1794 * 1795 * Notes: 1796 * Original versions of this routine were very 1797 * inefficient because they iteratively called 1798 * pmap_remove (slow...) 1799 */ 1800 1801static void 1802pmap_remove_all(m) 1803 vm_page_t m; 1804{ 1805 register pv_entry_t pv; 1806 register unsigned *pte, tpte; 1807 int s; 1808 1809#if defined(PMAP_DIAGNOSTIC) 1810 /* 1811 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1812 * pages! 1813 */ 1814 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) { 1815 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m)); 1816 } 1817#endif 1818 1819 s = splvm(); 1820 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { 1821 pv->pv_pmap->pm_stats.resident_count--; 1822 1823 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1824 1825 tpte = loadandclear(pte); 1826 if (tpte & PG_W) 1827 pv->pv_pmap->pm_stats.wired_count--; 1828 1829 if (tpte & PG_A) 1830 vm_page_flag_set(m, PG_REFERENCED); 1831 1832 /* 1833 * Update the vm_page_t clean and reference bits. 1834 */ 1835 if (tpte & PG_M) { 1836#if defined(PMAP_DIAGNOSTIC) 1837 if (pmap_nw_modified((pt_entry_t) tpte)) { 1838 printf( 1839 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n", 1840 pv->pv_va, tpte); 1841 } 1842#endif 1843 if (pmap_track_modified(pv->pv_va)) 1844 vm_page_dirty(m); 1845 } 1846 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 1847 1848 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1849 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 1850 m->md.pv_list_count--; 1851 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1852 free_pv_entry(pv); 1853 } 1854 1855 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 1856 1857 splx(s); 1858} 1859 1860/* 1861 * Set the physical protection on the 1862 * specified range of this map as requested. 1863 */ 1864void 1865pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1866{ 1867 register unsigned *ptbase; 1868 vm_offset_t pdnxt, ptpaddr; 1869 vm_pindex_t sindex, eindex; 1870 int anychanged; 1871 1872 if (pmap == NULL) 1873 return; 1874 1875 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1876 pmap_remove(pmap, sva, eva); 1877 return; 1878 } 1879 1880 if (prot & VM_PROT_WRITE) 1881 return; 1882 1883 anychanged = 0; 1884 1885 ptbase = get_ptbase(pmap); 1886 1887 sindex = i386_btop(sva); 1888 eindex = i386_btop(eva); 1889 1890 for (; sindex < eindex; sindex = pdnxt) { 1891 1892 unsigned pdirindex; 1893 1894 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1895 1896 pdirindex = sindex / NPDEPG; 1897 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1898 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1899 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1900 anychanged++; 1901 continue; 1902 } 1903 1904 /* 1905 * Weed out invalid mappings. Note: we assume that the page 1906 * directory table is always allocated, and in kernel virtual. 1907 */ 1908 if (ptpaddr == 0) 1909 continue; 1910 1911 if (pdnxt > eindex) { 1912 pdnxt = eindex; 1913 } 1914 1915 for (; sindex != pdnxt; sindex++) { 1916 1917 unsigned pbits; 1918 vm_page_t m; 1919 1920 pbits = ptbase[sindex]; 1921 1922 if (pbits & PG_MANAGED) { 1923 m = NULL; 1924 if (pbits & PG_A) { 1925 m = PHYS_TO_VM_PAGE(pbits); 1926 vm_page_flag_set(m, PG_REFERENCED); 1927 pbits &= ~PG_A; 1928 } 1929 if (pbits & PG_M) { 1930 if (pmap_track_modified(i386_ptob(sindex))) { 1931 if (m == NULL) 1932 m = PHYS_TO_VM_PAGE(pbits); 1933 vm_page_dirty(m); 1934 pbits &= ~PG_M; 1935 } 1936 } 1937 } 1938 1939 pbits &= ~PG_RW; 1940 1941 if (pbits != ptbase[sindex]) { 1942 ptbase[sindex] = pbits; 1943 anychanged = 1; 1944 } 1945 } 1946 } 1947 if (anychanged) 1948 pmap_TLB_invalidate_all(pmap); 1949} 1950 1951/* 1952 * Insert the given physical page (p) at 1953 * the specified virtual address (v) in the 1954 * target physical map with the protection requested. 1955 * 1956 * If specified, the page will be wired down, meaning 1957 * that the related pte can not be reclaimed. 1958 * 1959 * NB: This is the only routine which MAY NOT lazy-evaluate 1960 * or lose information. That is, this routine must actually 1961 * insert this page into the given map NOW. 1962 */ 1963void 1964pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, 1965 boolean_t wired) 1966{ 1967 vm_offset_t pa; 1968 register unsigned *pte; 1969 vm_offset_t opa; 1970 vm_offset_t origpte, newpte; 1971 vm_page_t mpte; 1972 1973 if (pmap == NULL) 1974 return; 1975 1976 va &= PG_FRAME; 1977#ifdef PMAP_DIAGNOSTIC 1978 if (va > VM_MAX_KERNEL_ADDRESS) 1979 panic("pmap_enter: toobig"); 1980 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 1981 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 1982#endif 1983 1984 mpte = NULL; 1985 /* 1986 * In the case that a page table page is not 1987 * resident, we are creating it here. 1988 */ 1989 if (va < UPT_MIN_ADDRESS) { 1990 mpte = pmap_allocpte(pmap, va); 1991 } 1992#if 0 && defined(PMAP_DIAGNOSTIC) 1993 else { 1994 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 1995 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 1996 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 1997 pmap->pm_pdir[PTDPTDI], origpte, va); 1998 } 1999 if (smp_active) { 2000 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid]; 2001 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2002 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2003 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2004 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr); 2005 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2006 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2007 } 2008 } 2009 } 2010#endif 2011 2012 pte = pmap_pte(pmap, va); 2013 2014 /* 2015 * Page Directory table entry not valid, we need a new PT page 2016 */ 2017 if (pte == NULL) { 2018 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n", 2019 (void *)pmap->pm_pdir[PTDPTDI], va); 2020 } 2021 2022 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME; 2023 origpte = *(vm_offset_t *)pte; 2024 opa = origpte & PG_FRAME; 2025 2026 if (origpte & PG_PS) 2027 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2028 2029 /* 2030 * Mapping has not changed, must be protection or wiring change. 2031 */ 2032 if (origpte && (opa == pa)) { 2033 /* 2034 * Wiring change, just update stats. We don't worry about 2035 * wiring PT pages as they remain resident as long as there 2036 * are valid mappings in them. Hence, if a user page is wired, 2037 * the PT page will be also. 2038 */ 2039 if (wired && ((origpte & PG_W) == 0)) 2040 pmap->pm_stats.wired_count++; 2041 else if (!wired && (origpte & PG_W)) 2042 pmap->pm_stats.wired_count--; 2043 2044#if defined(PMAP_DIAGNOSTIC) 2045 if (pmap_nw_modified((pt_entry_t) origpte)) { 2046 printf( 2047 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n", 2048 va, origpte); 2049 } 2050#endif 2051 2052 /* 2053 * Remove extra pte reference 2054 */ 2055 if (mpte) 2056 mpte->hold_count--; 2057 2058 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2059 if ((origpte & PG_RW) == 0) { 2060 *pte |= PG_RW; 2061#ifdef SMP 2062 cpu_invlpg((void *)va); 2063 if (pmap->pm_active & other_cpus) 2064 smp_invltlb(); 2065#else 2066 invltlb_1pg(va); 2067#endif 2068 } 2069 return; 2070 } 2071 2072 /* 2073 * We might be turning off write access to the page, 2074 * so we go ahead and sense modify status. 2075 */ 2076 if (origpte & PG_MANAGED) { 2077 if ((origpte & PG_M) && pmap_track_modified(va)) { 2078 vm_page_t om; 2079 om = PHYS_TO_VM_PAGE(opa); 2080 vm_page_dirty(om); 2081 } 2082 pa |= PG_MANAGED; 2083 } 2084 goto validate; 2085 } 2086 /* 2087 * Mapping has changed, invalidate old range and fall through to 2088 * handle validating new mapping. 2089 */ 2090 if (opa) { 2091 int err; 2092 err = pmap_remove_pte(pmap, pte, va); 2093 if (err) 2094 panic("pmap_enter: pte vanished, va: 0x%x", va); 2095 } 2096 2097 /* 2098 * Enter on the PV list if part of our managed memory. Note that we 2099 * raise IPL while manipulating pv_table since pmap_enter can be 2100 * called at interrupt time. 2101 */ 2102 if (pmap_initialized && 2103 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) { 2104 pmap_insert_entry(pmap, va, mpte, m); 2105 pa |= PG_MANAGED; 2106 } 2107 2108 /* 2109 * Increment counters 2110 */ 2111 pmap->pm_stats.resident_count++; 2112 if (wired) 2113 pmap->pm_stats.wired_count++; 2114 2115validate: 2116 /* 2117 * Now validate mapping with desired protection/wiring. 2118 */ 2119 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2120 2121 if (wired) 2122 newpte |= PG_W; 2123 if (va < UPT_MIN_ADDRESS) 2124 newpte |= PG_U; 2125 if (pmap == kernel_pmap) 2126 newpte |= pgeflag; 2127 2128 /* 2129 * if the mapping or permission bits are different, we need 2130 * to update the pte. 2131 */ 2132 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2133 *pte = newpte | PG_A; 2134 /*if (origpte)*/ { 2135#ifdef SMP 2136 cpu_invlpg((void *)va); 2137 if (pmap->pm_active & other_cpus) 2138 smp_invltlb(); 2139#else 2140 invltlb_1pg(va); 2141#endif 2142 } 2143 } 2144} 2145 2146/* 2147 * this code makes some *MAJOR* assumptions: 2148 * 1. Current pmap & pmap exists. 2149 * 2. Not wired. 2150 * 3. Read access. 2151 * 4. No page table pages. 2152 * 5. Tlbflush is deferred to calling procedure. 2153 * 6. Page IS managed. 2154 * but is *MUCH* faster than pmap_enter... 2155 */ 2156 2157static vm_page_t 2158pmap_enter_quick(pmap, va, m, mpte) 2159 register pmap_t pmap; 2160 vm_offset_t va; 2161 vm_page_t m; 2162 vm_page_t mpte; 2163{ 2164 unsigned *pte; 2165 vm_offset_t pa; 2166 2167 /* 2168 * In the case that a page table page is not 2169 * resident, we are creating it here. 2170 */ 2171 if (va < UPT_MIN_ADDRESS) { 2172 unsigned ptepindex; 2173 vm_offset_t ptepa; 2174 2175 /* 2176 * Calculate pagetable page index 2177 */ 2178 ptepindex = va >> PDRSHIFT; 2179 if (mpte && (mpte->pindex == ptepindex)) { 2180 mpte->hold_count++; 2181 } else { 2182retry: 2183 /* 2184 * Get the page directory entry 2185 */ 2186 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2187 2188 /* 2189 * If the page table page is mapped, we just increment 2190 * the hold count, and activate it. 2191 */ 2192 if (ptepa) { 2193 if (ptepa & PG_PS) 2194 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2195 if (pmap->pm_ptphint && 2196 (pmap->pm_ptphint->pindex == ptepindex)) { 2197 mpte = pmap->pm_ptphint; 2198 } else { 2199 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2200 pmap->pm_ptphint = mpte; 2201 } 2202 if (mpte == NULL) 2203 goto retry; 2204 mpte->hold_count++; 2205 } else { 2206 mpte = _pmap_allocpte(pmap, ptepindex); 2207 } 2208 } 2209 } else { 2210 mpte = NULL; 2211 } 2212 2213 /* 2214 * This call to vtopte makes the assumption that we are 2215 * entering the page into the current pmap. In order to support 2216 * quick entry into any pmap, one would likely use pmap_pte_quick. 2217 * But that isn't as quick as vtopte. 2218 */ 2219 pte = (unsigned *)vtopte(va); 2220 if (*pte) { 2221 if (mpte) 2222 pmap_unwire_pte_hold(pmap, mpte); 2223 return 0; 2224 } 2225 2226 /* 2227 * Enter on the PV list if part of our managed memory. Note that we 2228 * raise IPL while manipulating pv_table since pmap_enter can be 2229 * called at interrupt time. 2230 */ 2231 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) 2232 pmap_insert_entry(pmap, va, mpte, m); 2233 2234 /* 2235 * Increment counters 2236 */ 2237 pmap->pm_stats.resident_count++; 2238 2239 pa = VM_PAGE_TO_PHYS(m); 2240 2241 /* 2242 * Now validate mapping with RO protection 2243 */ 2244 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) 2245 *pte = pa | PG_V | PG_U; 2246 else 2247 *pte = pa | PG_V | PG_U | PG_MANAGED; 2248 2249 return mpte; 2250} 2251 2252/* 2253 * Make a temporary mapping for a physical address. This is only intended 2254 * to be used for panic dumps. 2255 */ 2256void * 2257pmap_kenter_temporary(vm_offset_t pa) 2258{ 2259 pmap_kenter((vm_offset_t)CADDR1, pa); 2260 return ((void *)CADDR1); 2261} 2262 2263#define MAX_INIT_PT (96) 2264/* 2265 * pmap_object_init_pt preloads the ptes for a given object 2266 * into the specified pmap. This eliminates the blast of soft 2267 * faults on process startup and immediately after an mmap. 2268 */ 2269void 2270pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2271 pmap_t pmap; 2272 vm_offset_t addr; 2273 vm_object_t object; 2274 vm_pindex_t pindex; 2275 vm_size_t size; 2276 int limit; 2277{ 2278 vm_offset_t tmpidx; 2279 int psize; 2280 vm_page_t p, mpte; 2281 int objpgs; 2282 2283 if (pmap == NULL || object == NULL) 2284 return; 2285 2286 /* 2287 * This code maps large physical mmap regions into the 2288 * processor address space. Note that some shortcuts 2289 * are taken, but the code works. 2290 */ 2291 if (pseflag && 2292 (object->type == OBJT_DEVICE) && 2293 ((addr & (NBPDR - 1)) == 0) && 2294 ((size & (NBPDR - 1)) == 0) ) { 2295 int i; 2296 vm_page_t m[1]; 2297 unsigned int ptepindex; 2298 int npdes; 2299 vm_offset_t ptepa; 2300 2301 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2302 return; 2303 2304retry: 2305 p = vm_page_lookup(object, pindex); 2306 if (p && vm_page_sleep_busy(p, FALSE, "init4p")) 2307 goto retry; 2308 2309 if (p == NULL) { 2310 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2311 if (p == NULL) 2312 return; 2313 m[0] = p; 2314 2315 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2316 vm_page_free(p); 2317 return; 2318 } 2319 2320 p = vm_page_lookup(object, pindex); 2321 vm_page_wakeup(p); 2322 } 2323 2324 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2325 if (ptepa & (NBPDR - 1)) { 2326 return; 2327 } 2328 2329 p->valid = VM_PAGE_BITS_ALL; 2330 2331 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2332 npdes = size >> PDRSHIFT; 2333 for(i=0;i<npdes;i++) { 2334 pmap->pm_pdir[ptepindex] = 2335 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2336 ptepa += NBPDR; 2337 ptepindex += 1; 2338 } 2339 vm_page_flag_set(p, PG_MAPPED); 2340 invltlb(); 2341 return; 2342 } 2343 2344 psize = i386_btop(size); 2345 2346 if ((object->type != OBJT_VNODE) || 2347 (limit && (psize > MAX_INIT_PT) && 2348 (object->resident_page_count > MAX_INIT_PT))) { 2349 return; 2350 } 2351 2352 if (psize + pindex > object->size) 2353 psize = object->size - pindex; 2354 2355 mpte = NULL; 2356 /* 2357 * if we are processing a major portion of the object, then scan the 2358 * entire thing. 2359 */ 2360 if (psize > (object->resident_page_count >> 2)) { 2361 objpgs = psize; 2362 2363 for (p = TAILQ_FIRST(&object->memq); 2364 ((objpgs > 0) && (p != NULL)); 2365 p = TAILQ_NEXT(p, listq)) { 2366 2367 tmpidx = p->pindex; 2368 if (tmpidx < pindex) { 2369 continue; 2370 } 2371 tmpidx -= pindex; 2372 if (tmpidx >= psize) { 2373 continue; 2374 } 2375 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2376 (p->busy == 0) && 2377 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2378 if ((p->queue - p->pc) == PQ_CACHE) 2379 vm_page_deactivate(p); 2380 vm_page_busy(p); 2381 mpte = pmap_enter_quick(pmap, 2382 addr + i386_ptob(tmpidx), p, mpte); 2383 vm_page_flag_set(p, PG_MAPPED); 2384 vm_page_wakeup(p); 2385 } 2386 objpgs -= 1; 2387 } 2388 } else { 2389 /* 2390 * else lookup the pages one-by-one. 2391 */ 2392 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2393 p = vm_page_lookup(object, tmpidx + pindex); 2394 if (p && 2395 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2396 (p->busy == 0) && 2397 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2398 if ((p->queue - p->pc) == PQ_CACHE) 2399 vm_page_deactivate(p); 2400 vm_page_busy(p); 2401 mpte = pmap_enter_quick(pmap, 2402 addr + i386_ptob(tmpidx), p, mpte); 2403 vm_page_flag_set(p, PG_MAPPED); 2404 vm_page_wakeup(p); 2405 } 2406 } 2407 } 2408 return; 2409} 2410 2411/* 2412 * pmap_prefault provides a quick way of clustering 2413 * pagefaults into a processes address space. It is a "cousin" 2414 * of pmap_object_init_pt, except it runs at page fault time instead 2415 * of mmap time. 2416 */ 2417#define PFBAK 4 2418#define PFFOR 4 2419#define PAGEORDER_SIZE (PFBAK+PFFOR) 2420 2421static int pmap_prefault_pageorder[] = { 2422 -PAGE_SIZE, PAGE_SIZE, 2423 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2424 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2425 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2426}; 2427 2428void 2429pmap_prefault(pmap, addra, entry) 2430 pmap_t pmap; 2431 vm_offset_t addra; 2432 vm_map_entry_t entry; 2433{ 2434 int i; 2435 vm_offset_t starta; 2436 vm_offset_t addr; 2437 vm_pindex_t pindex; 2438 vm_page_t m, mpte; 2439 vm_object_t object; 2440 2441 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) 2442 return; 2443 2444 object = entry->object.vm_object; 2445 2446 starta = addra - PFBAK * PAGE_SIZE; 2447 if (starta < entry->start) { 2448 starta = entry->start; 2449 } else if (starta > addra) { 2450 starta = 0; 2451 } 2452 2453 mpte = NULL; 2454 for (i = 0; i < PAGEORDER_SIZE; i++) { 2455 vm_object_t lobject; 2456 unsigned *pte; 2457 2458 addr = addra + pmap_prefault_pageorder[i]; 2459 if (addr > addra + (PFFOR * PAGE_SIZE)) 2460 addr = 0; 2461 2462 if (addr < starta || addr >= entry->end) 2463 continue; 2464 2465 if ((*pmap_pde(pmap, addr)) == NULL) 2466 continue; 2467 2468 pte = (unsigned *) vtopte(addr); 2469 if (*pte) 2470 continue; 2471 2472 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2473 lobject = object; 2474 for (m = vm_page_lookup(lobject, pindex); 2475 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2476 lobject = lobject->backing_object) { 2477 if (lobject->backing_object_offset & PAGE_MASK) 2478 break; 2479 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2480 m = vm_page_lookup(lobject->backing_object, pindex); 2481 } 2482 2483 /* 2484 * give-up when a page is not in memory 2485 */ 2486 if (m == NULL) 2487 break; 2488 2489 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2490 (m->busy == 0) && 2491 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2492 2493 if ((m->queue - m->pc) == PQ_CACHE) { 2494 vm_page_deactivate(m); 2495 } 2496 vm_page_busy(m); 2497 mpte = pmap_enter_quick(pmap, addr, m, mpte); 2498 vm_page_flag_set(m, PG_MAPPED); 2499 vm_page_wakeup(m); 2500 } 2501 } 2502} 2503 2504/* 2505 * Routine: pmap_change_wiring 2506 * Function: Change the wiring attribute for a map/virtual-address 2507 * pair. 2508 * In/out conditions: 2509 * The mapping must already exist in the pmap. 2510 */ 2511void 2512pmap_change_wiring(pmap, va, wired) 2513 register pmap_t pmap; 2514 vm_offset_t va; 2515 boolean_t wired; 2516{ 2517 register unsigned *pte; 2518 2519 if (pmap == NULL) 2520 return; 2521 2522 pte = pmap_pte(pmap, va); 2523 2524 if (wired && !pmap_pte_w(pte)) 2525 pmap->pm_stats.wired_count++; 2526 else if (!wired && pmap_pte_w(pte)) 2527 pmap->pm_stats.wired_count--; 2528 2529 /* 2530 * Wiring is not a hardware characteristic so there is no need to 2531 * invalidate TLB. 2532 */ 2533 pmap_pte_set_w(pte, wired); 2534} 2535 2536 2537 2538/* 2539 * Copy the range specified by src_addr/len 2540 * from the source map to the range dst_addr/len 2541 * in the destination map. 2542 * 2543 * This routine is only advisory and need not do anything. 2544 */ 2545 2546void 2547pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2548 pmap_t dst_pmap, src_pmap; 2549 vm_offset_t dst_addr; 2550 vm_size_t len; 2551 vm_offset_t src_addr; 2552{ 2553 vm_offset_t addr; 2554 vm_offset_t end_addr = src_addr + len; 2555 vm_offset_t pdnxt; 2556 unsigned src_frame, dst_frame; 2557 vm_page_t m; 2558 2559 if (dst_addr != src_addr) 2560 return; 2561 2562 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2563 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2564 return; 2565 } 2566 2567 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2568 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2569 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2570#if defined(SMP) 2571 /* The page directory is not shared between CPUs */ 2572 cpu_invltlb(); 2573#else 2574 invltlb(); 2575#endif 2576 } 2577 2578 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2579 unsigned *src_pte, *dst_pte; 2580 vm_page_t dstmpte, srcmpte; 2581 vm_offset_t srcptepaddr; 2582 unsigned ptepindex; 2583 2584 if (addr >= UPT_MIN_ADDRESS) 2585 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2586 2587 /* 2588 * Don't let optional prefaulting of pages make us go 2589 * way below the low water mark of free pages or way 2590 * above high water mark of used pv entries. 2591 */ 2592 if (cnt.v_free_count < cnt.v_free_reserved || 2593 pv_entry_count > pv_entry_high_water) 2594 break; 2595 2596 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2597 ptepindex = addr >> PDRSHIFT; 2598 2599 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2600 if (srcptepaddr == 0) 2601 continue; 2602 2603 if (srcptepaddr & PG_PS) { 2604 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2605 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2606 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE; 2607 } 2608 continue; 2609 } 2610 2611 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2612 if ((srcmpte == NULL) || 2613 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2614 continue; 2615 2616 if (pdnxt > end_addr) 2617 pdnxt = end_addr; 2618 2619 src_pte = (unsigned *) vtopte(addr); 2620 dst_pte = (unsigned *) avtopte(addr); 2621 while (addr < pdnxt) { 2622 unsigned ptetemp; 2623 ptetemp = *src_pte; 2624 /* 2625 * we only virtual copy managed pages 2626 */ 2627 if ((ptetemp & PG_MANAGED) != 0) { 2628 /* 2629 * We have to check after allocpte for the 2630 * pte still being around... allocpte can 2631 * block. 2632 */ 2633 dstmpte = pmap_allocpte(dst_pmap, addr); 2634 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2635 /* 2636 * Clear the modified and 2637 * accessed (referenced) bits 2638 * during the copy. 2639 */ 2640 m = PHYS_TO_VM_PAGE(ptetemp); 2641 *dst_pte = ptetemp & ~(PG_M | PG_A); 2642 dst_pmap->pm_stats.resident_count++; 2643 pmap_insert_entry(dst_pmap, addr, 2644 dstmpte, m); 2645 } else { 2646 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2647 } 2648 if (dstmpte->hold_count >= srcmpte->hold_count) 2649 break; 2650 } 2651 addr += PAGE_SIZE; 2652 src_pte++; 2653 dst_pte++; 2654 } 2655 } 2656} 2657 2658/* 2659 * Routine: pmap_kernel 2660 * Function: 2661 * Returns the physical map handle for the kernel. 2662 */ 2663pmap_t 2664pmap_kernel() 2665{ 2666 return (kernel_pmap); 2667} 2668 2669/* 2670 * pmap_zero_page zeros the specified hardware page by mapping 2671 * the page into KVM and using bzero to clear its contents. 2672 */ 2673void 2674pmap_zero_page(phys) 2675 vm_offset_t phys; 2676{ 2677#ifdef SMP 2678 if (*(int *) prv_CMAP3) 2679 panic("pmap_zero_page: prv_CMAP3 busy"); 2680 2681 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2682 cpu_invlpg(prv_CADDR3); 2683 2684#if defined(I686_CPU) 2685 if (cpu_class == CPUCLASS_686) 2686 i686_pagezero(prv_CADDR3); 2687 else 2688#endif 2689 bzero(prv_CADDR3, PAGE_SIZE); 2690 2691 *(int *) prv_CMAP3 = 0; 2692#else 2693 if (*(int *) CMAP2) 2694 panic("pmap_zero_page: CMAP2 busy"); 2695 2696 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2697 invltlb_1pg((vm_offset_t)CADDR2); 2698 2699#if defined(I686_CPU) 2700 if (cpu_class == CPUCLASS_686) 2701 i686_pagezero(CADDR2); 2702 else 2703#endif 2704 bzero(CADDR2, PAGE_SIZE); 2705 *(int *) CMAP2 = 0; 2706#endif 2707} 2708 2709/* 2710 * pmap_zero_page_area zeros the specified hardware page by mapping 2711 * the page into KVM and using bzero to clear its contents. 2712 * 2713 * off and size may not cover an area beyond a single hardware page. 2714 */ 2715void 2716pmap_zero_page_area(phys, off, size) 2717 vm_offset_t phys; 2718 int off; 2719 int size; 2720{ 2721#ifdef SMP 2722 if (*(int *) prv_CMAP3) 2723 panic("pmap_zero_page: prv_CMAP3 busy"); 2724 2725 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2726 cpu_invlpg(prv_CADDR3); 2727 2728#if defined(I686_CPU) 2729 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2730 i686_pagezero(prv_CADDR3); 2731 else 2732#endif 2733 bzero((char *)prv_CADDR3 + off, size); 2734 2735 *(int *) prv_CMAP3 = 0; 2736#else 2737 if (*(int *) CMAP2) 2738 panic("pmap_zero_page: CMAP2 busy"); 2739 2740 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2741 invltlb_1pg((vm_offset_t)CADDR2); 2742 2743#if defined(I686_CPU) 2744 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2745 i686_pagezero(CADDR2); 2746 else 2747#endif 2748 bzero((char *)CADDR2 + off, size); 2749 *(int *) CMAP2 = 0; 2750#endif 2751} 2752 2753/* 2754 * pmap_copy_page copies the specified (machine independent) 2755 * page by mapping the page into virtual memory and using 2756 * bcopy to copy the page, one machine dependent page at a 2757 * time. 2758 */ 2759void 2760pmap_copy_page(src, dst) 2761 vm_offset_t src; 2762 vm_offset_t dst; 2763{ 2764#ifdef SMP 2765 if (*(int *) prv_CMAP1) 2766 panic("pmap_copy_page: prv_CMAP1 busy"); 2767 if (*(int *) prv_CMAP2) 2768 panic("pmap_copy_page: prv_CMAP2 busy"); 2769 2770 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2771 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2772 2773 cpu_invlpg(prv_CADDR1); 2774 cpu_invlpg(prv_CADDR2); 2775 2776 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE); 2777 2778 *(int *) prv_CMAP1 = 0; 2779 *(int *) prv_CMAP2 = 0; 2780#else 2781 if (*(int *) CMAP1 || *(int *) CMAP2) 2782 panic("pmap_copy_page: CMAP busy"); 2783 2784 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2785 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2786#if defined(I386_CPU) 2787 if (cpu_class == CPUCLASS_386) { 2788 invltlb(); 2789 } else 2790#endif 2791 { 2792 invlpg((u_int)CADDR1); 2793 invlpg((u_int)CADDR2); 2794 } 2795 2796 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2797 2798 *(int *) CMAP1 = 0; 2799 *(int *) CMAP2 = 0; 2800#endif 2801} 2802 2803 2804/* 2805 * Routine: pmap_pageable 2806 * Function: 2807 * Make the specified pages (by pmap, offset) 2808 * pageable (or not) as requested. 2809 * 2810 * A page which is not pageable may not take 2811 * a fault; therefore, its page table entry 2812 * must remain valid for the duration. 2813 * 2814 * This routine is merely advisory; pmap_enter 2815 * will specify that these pages are to be wired 2816 * down (or not) as appropriate. 2817 */ 2818void 2819pmap_pageable(pmap, sva, eva, pageable) 2820 pmap_t pmap; 2821 vm_offset_t sva, eva; 2822 boolean_t pageable; 2823{ 2824} 2825 2826/* 2827 * this routine returns true if a physical page resides 2828 * in the given pmap. 2829 */ 2830boolean_t 2831pmap_page_exists(pmap, m) 2832 pmap_t pmap; 2833 vm_page_t m; 2834{ 2835 register pv_entry_t pv; 2836 int s; 2837 2838 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 2839 return FALSE; 2840 2841 s = splvm(); 2842 2843 /* 2844 * Not found, check current mappings returning immediately if found. 2845 */ 2846 for (pv = TAILQ_FIRST(&m->md.pv_list); 2847 pv; 2848 pv = TAILQ_NEXT(pv, pv_list)) { 2849 if (pv->pv_pmap == pmap) { 2850 splx(s); 2851 return TRUE; 2852 } 2853 } 2854 splx(s); 2855 return (FALSE); 2856} 2857 2858#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2859/* 2860 * Remove all pages from specified address space 2861 * this aids process exit speeds. Also, this code 2862 * is special cased for current process only, but 2863 * can have the more generic (and slightly slower) 2864 * mode enabled. This is much faster than pmap_remove 2865 * in the case of running down an entire address space. 2866 */ 2867void 2868pmap_remove_pages(pmap, sva, eva) 2869 pmap_t pmap; 2870 vm_offset_t sva, eva; 2871{ 2872 unsigned *pte, tpte; 2873 pv_entry_t pv, npv; 2874 int s; 2875 vm_page_t m; 2876 2877#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2878 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) { 2879 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2880 return; 2881 } 2882#endif 2883 2884 s = splvm(); 2885 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2886 pv; 2887 pv = npv) { 2888 2889 if (pv->pv_va >= eva || pv->pv_va < sva) { 2890 npv = TAILQ_NEXT(pv, pv_plist); 2891 continue; 2892 } 2893 2894#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2895 pte = (unsigned *)vtopte(pv->pv_va); 2896#else 2897 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2898#endif 2899 tpte = *pte; 2900 2901/* 2902 * We cannot remove wired pages from a process' mapping at this time 2903 */ 2904 if (tpte & PG_W) { 2905 npv = TAILQ_NEXT(pv, pv_plist); 2906 continue; 2907 } 2908 *pte = 0; 2909 2910 m = PHYS_TO_VM_PAGE(tpte); 2911 2912 KASSERT(m < &vm_page_array[vm_page_array_size], 2913 ("pmap_remove_pages: bad tpte %x", tpte)); 2914 2915 pv->pv_pmap->pm_stats.resident_count--; 2916 2917 /* 2918 * Update the vm_page_t clean and reference bits. 2919 */ 2920 if (tpte & PG_M) { 2921 vm_page_dirty(m); 2922 } 2923 2924 2925 npv = TAILQ_NEXT(pv, pv_plist); 2926 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2927 2928 m->md.pv_list_count--; 2929 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 2930 if (TAILQ_FIRST(&m->md.pv_list) == NULL) { 2931 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 2932 } 2933 2934 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2935 free_pv_entry(pv); 2936 } 2937 splx(s); 2938 pmap_TLB_invalidate_all(pmap); 2939} 2940 2941/* 2942 * pmap_testbit tests bits in pte's 2943 * note that the testbit/changebit routines are inline, 2944 * and a lot of things compile-time evaluate. 2945 */ 2946static boolean_t 2947pmap_testbit(m, bit) 2948 vm_page_t m; 2949 int bit; 2950{ 2951 pv_entry_t pv; 2952 unsigned *pte; 2953 int s; 2954 2955 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 2956 return FALSE; 2957 2958 if (TAILQ_FIRST(&m->md.pv_list) == NULL) 2959 return FALSE; 2960 2961 s = splvm(); 2962 2963 for (pv = TAILQ_FIRST(&m->md.pv_list); 2964 pv; 2965 pv = TAILQ_NEXT(pv, pv_list)) { 2966 2967 /* 2968 * if the bit being tested is the modified bit, then 2969 * mark clean_map and ptes as never 2970 * modified. 2971 */ 2972 if (bit & (PG_A|PG_M)) { 2973 if (!pmap_track_modified(pv->pv_va)) 2974 continue; 2975 } 2976 2977#if defined(PMAP_DIAGNOSTIC) 2978 if (!pv->pv_pmap) { 2979 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 2980 continue; 2981 } 2982#endif 2983 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2984 if (*pte & bit) { 2985 splx(s); 2986 return TRUE; 2987 } 2988 } 2989 splx(s); 2990 return (FALSE); 2991} 2992 2993/* 2994 * this routine is used to modify bits in ptes 2995 */ 2996static __inline void 2997pmap_changebit(m, bit, setem) 2998 vm_page_t m; 2999 int bit; 3000 boolean_t setem; 3001{ 3002 register pv_entry_t pv; 3003 register unsigned *pte; 3004 int s; 3005 3006 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 3007 return; 3008 3009 s = splvm(); 3010 3011 /* 3012 * Loop over all current mappings setting/clearing as appropos If 3013 * setting RO do we need to clear the VAC? 3014 */ 3015 for (pv = TAILQ_FIRST(&m->md.pv_list); 3016 pv; 3017 pv = TAILQ_NEXT(pv, pv_list)) { 3018 3019 /* 3020 * don't write protect pager mappings 3021 */ 3022 if (!setem && (bit == PG_RW)) { 3023 if (!pmap_track_modified(pv->pv_va)) 3024 continue; 3025 } 3026 3027#if defined(PMAP_DIAGNOSTIC) 3028 if (!pv->pv_pmap) { 3029 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 3030 continue; 3031 } 3032#endif 3033 3034 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3035 3036 if (setem) { 3037 *(int *)pte |= bit; 3038 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3039 } else { 3040 vm_offset_t pbits = *(vm_offset_t *)pte; 3041 if (pbits & bit) { 3042 if (bit == PG_RW) { 3043 if (pbits & PG_M) { 3044 vm_page_dirty(m); 3045 } 3046 *(int *)pte = pbits & ~(PG_M|PG_RW); 3047 } else { 3048 *(int *)pte = pbits & ~bit; 3049 } 3050 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3051 } 3052 } 3053 } 3054 splx(s); 3055} 3056 3057/* 3058 * pmap_page_protect: 3059 * 3060 * Lower the permission for all mappings to a given page. 3061 */ 3062void 3063pmap_page_protect(vm_page_t m, vm_prot_t prot) 3064{ 3065 if ((prot & VM_PROT_WRITE) == 0) { 3066 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3067 pmap_changebit(m, PG_RW, FALSE); 3068 } else { 3069 pmap_remove_all(m); 3070 } 3071 } 3072} 3073 3074vm_offset_t 3075pmap_phys_address(ppn) 3076 int ppn; 3077{ 3078 return (i386_ptob(ppn)); 3079} 3080 3081/* 3082 * pmap_ts_referenced: 3083 * 3084 * Return the count of reference bits for a page, clearing all of them. 3085 */ 3086int 3087pmap_ts_referenced(vm_page_t m) 3088{ 3089 register pv_entry_t pv, pvf, pvn; 3090 unsigned *pte; 3091 int s; 3092 int rtval = 0; 3093 3094 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 3095 return (rtval); 3096 3097 s = splvm(); 3098 3099 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { 3100 3101 pvf = pv; 3102 3103 do { 3104 pvn = TAILQ_NEXT(pv, pv_list); 3105 3106 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 3107 3108 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); 3109 3110 if (!pmap_track_modified(pv->pv_va)) 3111 continue; 3112 3113 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3114 3115 if (pte && (*pte & PG_A)) { 3116 *pte &= ~PG_A; 3117 3118 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3119 3120 rtval++; 3121 if (rtval > 4) { 3122 break; 3123 } 3124 } 3125 } while ((pv = pvn) != NULL && pv != pvf); 3126 } 3127 splx(s); 3128 3129 return (rtval); 3130} 3131 3132/* 3133 * pmap_is_modified: 3134 * 3135 * Return whether or not the specified physical page was modified 3136 * in any physical maps. 3137 */ 3138boolean_t 3139pmap_is_modified(vm_page_t m) 3140{ 3141 return pmap_testbit(m, PG_M); 3142} 3143 3144/* 3145 * Clear the modify bits on the specified physical page. 3146 */ 3147void 3148pmap_clear_modify(vm_page_t m) 3149{ 3150 pmap_changebit(m, PG_M, FALSE); 3151} 3152 3153/* 3154 * pmap_clear_reference: 3155 * 3156 * Clear the reference bit on the specified physical page. 3157 */ 3158void 3159pmap_clear_reference(vm_page_t m) 3160{ 3161 pmap_changebit(m, PG_A, FALSE); 3162} 3163 3164/* 3165 * Miscellaneous support routines follow 3166 */ 3167 3168static void 3169i386_protection_init() 3170{ 3171 register int *kp, prot; 3172 3173 kp = protection_codes; 3174 for (prot = 0; prot < 8; prot++) { 3175 switch (prot) { 3176 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3177 /* 3178 * Read access is also 0. There isn't any execute bit, 3179 * so just make it readable. 3180 */ 3181 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3182 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3183 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3184 *kp++ = 0; 3185 break; 3186 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3187 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3188 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3189 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3190 *kp++ = PG_RW; 3191 break; 3192 } 3193 } 3194} 3195 3196/* 3197 * Map a set of physical memory pages into the kernel virtual 3198 * address space. Return a pointer to where it is mapped. This 3199 * routine is intended to be used for mapping device memory, 3200 * NOT real memory. 3201 */ 3202void * 3203pmap_mapdev(pa, size) 3204 vm_offset_t pa; 3205 vm_size_t size; 3206{ 3207 vm_offset_t va, tmpva, offset; 3208 unsigned *pte; 3209 3210 offset = pa & PAGE_MASK; 3211 size = roundup(offset + size, PAGE_SIZE); 3212 3213 va = kmem_alloc_pageable(kernel_map, size); 3214 if (!va) 3215 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3216 3217 pa = pa & PG_FRAME; 3218 for (tmpva = va; size > 0;) { 3219 pte = (unsigned *)vtopte(tmpva); 3220 *pte = pa | PG_RW | PG_V | pgeflag; 3221 size -= PAGE_SIZE; 3222 tmpva += PAGE_SIZE; 3223 pa += PAGE_SIZE; 3224 } 3225 invltlb(); 3226 3227 return ((void *)(va + offset)); 3228} 3229 3230void 3231pmap_unmapdev(va, size) 3232 vm_offset_t va; 3233 vm_size_t size; 3234{ 3235 vm_offset_t base, offset; 3236 3237 base = va & PG_FRAME; 3238 offset = va & PAGE_MASK; 3239 size = roundup(offset + size, PAGE_SIZE); 3240 kmem_free(kernel_map, base, size); 3241} 3242 3243/* 3244 * perform the pmap work for mincore 3245 */ 3246int 3247pmap_mincore(pmap, addr) 3248 pmap_t pmap; 3249 vm_offset_t addr; 3250{ 3251 3252 unsigned *ptep, pte; 3253 vm_page_t m; 3254 int val = 0; 3255 3256 ptep = pmap_pte(pmap, addr); 3257 if (ptep == 0) { 3258 return 0; 3259 } 3260 3261 if ((pte = *ptep) != 0) { 3262 vm_offset_t pa; 3263 3264 val = MINCORE_INCORE; 3265 if ((pte & PG_MANAGED) == 0) 3266 return val; 3267 3268 pa = pte & PG_FRAME; 3269 3270 m = PHYS_TO_VM_PAGE(pa); 3271 3272 /* 3273 * Modified by us 3274 */ 3275 if (pte & PG_M) 3276 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3277 /* 3278 * Modified by someone 3279 */ 3280 else if (m->dirty || pmap_is_modified(m)) 3281 val |= MINCORE_MODIFIED_OTHER; 3282 /* 3283 * Referenced by us 3284 */ 3285 if (pte & PG_A) 3286 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3287 3288 /* 3289 * Referenced by someone 3290 */ 3291 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) { 3292 val |= MINCORE_REFERENCED_OTHER; 3293 vm_page_flag_set(m, PG_REFERENCED); 3294 } 3295 } 3296 return val; 3297} 3298 3299void 3300pmap_activate(struct proc *p) 3301{ 3302 pmap_t pmap; 3303 3304 pmap = vmspace_pmap(p->p_vmspace); 3305#if defined(SMP) 3306 pmap->pm_active |= 1 << cpuid; 3307#else 3308 pmap->pm_active |= 1; 3309#endif 3310#if defined(SWTCH_OPTIM_STATS) 3311 tlb_flush_count++; 3312#endif 3313 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir)); 3314} 3315 3316vm_offset_t 3317pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) 3318{ 3319 3320 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3321 return addr; 3322 } 3323 3324 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3325 return addr; 3326} 3327 3328 3329#if defined(PMAP_DEBUG) 3330pmap_pid_dump(int pid) 3331{ 3332 pmap_t pmap; 3333 struct proc *p; 3334 int npte = 0; 3335 int index; 3336 LIST_FOREACH(p, &allproc, p_list) { 3337 if (p->p_pid != pid) 3338 continue; 3339 3340 if (p->p_vmspace) { 3341 int i,j; 3342 index = 0; 3343 pmap = vmspace_pmap(p->p_vmspace); 3344 for(i=0;i<1024;i++) { 3345 pd_entry_t *pde; 3346 unsigned *pte; 3347 unsigned base = i << PDRSHIFT; 3348 3349 pde = &pmap->pm_pdir[i]; 3350 if (pde && pmap_pde_v(pde)) { 3351 for(j=0;j<1024;j++) { 3352 unsigned va = base + (j << PAGE_SHIFT); 3353 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3354 if (index) { 3355 index = 0; 3356 printf("\n"); 3357 } 3358 return npte; 3359 } 3360 pte = pmap_pte_quick( pmap, va); 3361 if (pte && pmap_pte_v(pte)) { 3362 vm_offset_t pa; 3363 vm_page_t m; 3364 pa = *(int *)pte; 3365 m = PHYS_TO_VM_PAGE(pa); 3366 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3367 va, pa, m->hold_count, m->wire_count, m->flags); 3368 npte++; 3369 index++; 3370 if (index >= 2) { 3371 index = 0; 3372 printf("\n"); 3373 } else { 3374 printf(" "); 3375 } 3376 } 3377 } 3378 } 3379 } 3380 } 3381 } 3382 return npte; 3383} 3384#endif 3385 3386#if defined(DEBUG) 3387 3388static void pads __P((pmap_t pm)); 3389void pmap_pvdump __P((vm_offset_t pa)); 3390 3391/* print address space of pmap*/ 3392static void 3393pads(pm) 3394 pmap_t pm; 3395{ 3396 unsigned va, i, j; 3397 unsigned *ptep; 3398 3399 if (pm == kernel_pmap) 3400 return; 3401 for (i = 0; i < 1024; i++) 3402 if (pm->pm_pdir[i]) 3403 for (j = 0; j < 1024; j++) { 3404 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3405 if (pm == kernel_pmap && va < KERNBASE) 3406 continue; 3407 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3408 continue; 3409 ptep = pmap_pte_quick(pm, va); 3410 if (pmap_pte_v(ptep)) 3411 printf("%x:%x ", va, *(int *) ptep); 3412 }; 3413 3414} 3415 3416void 3417pmap_pvdump(pa) 3418 vm_offset_t pa; 3419{ 3420 register pv_entry_t pv; 3421 vm_page_t m; 3422 3423 printf("pa %x", pa); 3424 m = PHYS_TO_VM_PAGE(pa); 3425 for (pv = TAILQ_FIRST(&m->md.pv_list); 3426 pv; 3427 pv = TAILQ_NEXT(pv, pv_list)) { 3428#ifdef used_to_be 3429 printf(" -> pmap %p, va %x, flags %x", 3430 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags); 3431#endif 3432 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3433 pads(pv->pv_pmap); 3434 } 3435 printf(" "); 3436} 3437#endif 3438