pmap.c revision 49337
1/* 2 * Copyright (c) 1991 Regents of the University of California. 3 * All rights reserved. 4 * Copyright (c) 1994 John S. Dyson 5 * All rights reserved. 6 * Copyright (c) 1994 David Greenman 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * the Systems Programming Group of the University of Utah Computer 11 * Science Department and William Jolitz of UUNET Technologies Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by the University of 24 * California, Berkeley and its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91 42 * $Id: pmap.c,v 1.243 1999/07/31 04:10:31 alc Exp $ 43 */ 44 45/* 46 * Manages physical address maps. 47 * 48 * In addition to hardware address maps, this 49 * module is called upon to provide software-use-only 50 * maps which may or may not be stored in the same 51 * form as hardware maps. These pseudo-maps are 52 * used to store intermediate results from copy 53 * operations to and from address spaces. 54 * 55 * Since the information managed by this module is 56 * also stored by the logical address mapping module, 57 * this module may throw away valid virtual-to-physical 58 * mappings at almost any time. However, invalidations 59 * of virtual-to-physical mappings must be done as 60 * requested. 61 * 62 * In order to cope with hardware architectures which 63 * make virtual-to-physical map invalidates expensive, 64 * this module may delay invalidate or reduced protection 65 * operations until such time as they are actually 66 * necessary. This module is given full information as 67 * to which processors are currently using which maps, 68 * and to when physical maps must be made correct. 69 */ 70 71#include "opt_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 <vm/vm_prot.h> 86#include <sys/lock.h> 87#include <vm/vm_kern.h> 88#include <vm/vm_page.h> 89#include <vm/vm_map.h> 90#include <vm/vm_object.h> 91#include <vm/vm_extern.h> 92#include <vm/vm_pageout.h> 93#include <vm/vm_pager.h> 94#include <vm/vm_zone.h> 95 96#include <sys/user.h> 97 98#include <machine/cputypes.h> 99#include <machine/md_var.h> 100#include <machine/specialreg.h> 101#if defined(SMP) || defined(APIC_IO) 102#include <machine/smp.h> 103#include <machine/apic.h> 104#include <machine/segments.h> 105#include <machine/tss.h> 106#include <machine/globaldata.h> 107#endif /* SMP || APIC_IO */ 108 109#define PMAP_KEEP_PDIRS 110#ifndef PMAP_SHPGPERPROC 111#define PMAP_SHPGPERPROC 200 112#endif 113 114#if defined(DIAGNOSTIC) 115#define PMAP_DIAGNOSTIC 116#endif 117 118#define MINPV 2048 119 120#if !defined(PMAP_DIAGNOSTIC) 121#define PMAP_INLINE __inline 122#else 123#define PMAP_INLINE 124#endif 125 126/* 127 * Get PDEs and PTEs for user/kernel address space 128 */ 129#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 130#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 131 132#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 133#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 134#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 135#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 136#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 137 138#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 139#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 140 141/* 142 * Given a map and a machine independent protection code, 143 * convert to a vax protection code. 144 */ 145#define pte_prot(m, p) (protection_codes[p]) 146static int protection_codes[8]; 147 148#define pa_index(pa) atop((pa) - vm_first_phys) 149#define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) 150 151static struct pmap kernel_pmap_store; 152pmap_t kernel_pmap; 153 154vm_offset_t avail_start; /* PA of first available physical page */ 155vm_offset_t avail_end; /* PA of last available physical page */ 156vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 157vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 158static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 159static vm_offset_t vm_first_phys; 160static int pgeflag; /* PG_G or-in */ 161static int pseflag; /* PG_PS or-in */ 162static int pv_npg; 163 164static vm_object_t kptobj; 165 166static int nkpt; 167vm_offset_t kernel_vm_end; 168 169/* 170 * Data for the pv entry allocation mechanism 171 */ 172static vm_zone_t pvzone; 173static struct vm_zone pvzone_store; 174static struct vm_object pvzone_obj; 175static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0; 176static int pmap_pagedaemon_waken = 0; 177static struct pv_entry *pvinit; 178 179/* 180 * All those kernel PT submaps that BSD is so fond of 181 */ 182pt_entry_t *CMAP1 = 0; 183static pt_entry_t *CMAP2, *ptmmap; 184static pv_table_t *pv_table; 185caddr_t CADDR1 = 0, ptvmmap = 0; 186static caddr_t CADDR2; 187static pt_entry_t *msgbufmap; 188struct msgbuf *msgbufp=0; 189 190#ifdef SMP 191extern pt_entry_t *SMPpt; 192#else 193static pt_entry_t *PMAP1 = 0; 194static unsigned *PADDR1 = 0; 195#endif 196 197static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 198static unsigned * get_ptbase __P((pmap_t pmap)); 199static pv_entry_t get_pv_entry __P((void)); 200static void i386_protection_init __P((void)); 201static __inline void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 202static void pmap_clearbit __P((vm_offset_t pa, int bit)); 203 204static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 205static void pmap_remove_all __P((vm_offset_t pa)); 206static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 207 vm_offset_t pa, vm_page_t mpte)); 208static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 209 vm_offset_t sva)); 210static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 211static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 212 vm_offset_t va)); 213static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 214static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 215 vm_page_t mpte, vm_offset_t pa)); 216 217static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 218 219static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 220static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 221static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 222static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 223static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 224static vm_offset_t pmap_kmem_choose(vm_offset_t addr); 225 226static unsigned pdir4mb; 227 228/* 229 * Routine: pmap_pte 230 * Function: 231 * Extract the page table entry associated 232 * with the given map/virtual_address pair. 233 */ 234 235PMAP_INLINE unsigned * 236pmap_pte(pmap, va) 237 register pmap_t pmap; 238 vm_offset_t va; 239{ 240 unsigned *pdeaddr; 241 242 if (pmap) { 243 pdeaddr = (unsigned *) pmap_pde(pmap, va); 244 if (*pdeaddr & PG_PS) 245 return pdeaddr; 246 if (*pdeaddr) { 247 return get_ptbase(pmap) + i386_btop(va); 248 } 249 } 250 return (0); 251} 252 253/* 254 * Move the kernel virtual free pointer to the next 255 * 4MB. This is used to help improve performance 256 * by using a large (4MB) page for much of the kernel 257 * (.text, .data, .bss) 258 */ 259static vm_offset_t 260pmap_kmem_choose(vm_offset_t addr) { 261 vm_offset_t newaddr = addr; 262#ifndef DISABLE_PSE 263 if (cpu_feature & CPUID_PSE) { 264 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 265 } 266#endif 267 return newaddr; 268} 269 270/* 271 * Bootstrap the system enough to run with virtual memory. 272 * 273 * On the i386 this is called after mapping has already been enabled 274 * and just syncs the pmap module with what has already been done. 275 * [We can't call it easily with mapping off since the kernel is not 276 * mapped with PA == VA, hence we would have to relocate every address 277 * from the linked base (virtual) address "KERNBASE" to the actual 278 * (physical) address starting relative to 0] 279 */ 280void 281pmap_bootstrap(firstaddr, loadaddr) 282 vm_offset_t firstaddr; 283 vm_offset_t loadaddr; 284{ 285 vm_offset_t va; 286 pt_entry_t *pte; 287#ifdef SMP 288 int i, j; 289 struct globaldata *gd; 290#endif 291 292 avail_start = firstaddr; 293 294 /* 295 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 296 * large. It should instead be correctly calculated in locore.s and 297 * not based on 'first' (which is a physical address, not a virtual 298 * address, for the start of unused physical memory). The kernel 299 * page tables are NOT double mapped and thus should not be included 300 * in this calculation. 301 */ 302 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 303 virtual_avail = pmap_kmem_choose(virtual_avail); 304 305 virtual_end = VM_MAX_KERNEL_ADDRESS; 306 307 /* 308 * Initialize protection array. 309 */ 310 i386_protection_init(); 311 312 /* 313 * The kernel's pmap is statically allocated so we don't have to use 314 * pmap_create, which is unlikely to work correctly at this part of 315 * the boot sequence (XXX and which no longer exists). 316 */ 317 kernel_pmap = &kernel_pmap_store; 318 319 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 320 kernel_pmap->pm_count = 1; 321 kernel_pmap->pm_active = -1; /* don't allow deactivation */ 322 TAILQ_INIT(&kernel_pmap->pm_pvlist); 323 nkpt = NKPT; 324 325 /* 326 * Reserve some special page table entries/VA space for temporary 327 * mapping of pages. 328 */ 329#define SYSMAP(c, p, v, n) \ 330 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 331 332 va = virtual_avail; 333 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 334 335 /* 336 * CMAP1/CMAP2 are used for zeroing and copying pages. 337 */ 338 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 339 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 340 341 /* 342 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 343 * XXX ptmmap is not used. 344 */ 345 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 346 347 /* 348 * msgbufp is used to map the system message buffer. 349 * XXX msgbufmap is not used. 350 */ 351 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 352 atop(round_page(MSGBUF_SIZE))) 353 354#if !defined(SMP) 355 /* 356 * ptemap is used for pmap_pte_quick 357 */ 358 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 359#endif 360 361 virtual_avail = va; 362 363 *(int *) CMAP1 = *(int *) CMAP2 = 0; 364 *(int *) PTD = 0; 365 366 367 pgeflag = 0; 368#if !defined(SMP) 369 if (cpu_feature & CPUID_PGE) { 370 pgeflag = PG_G; 371 } 372#endif 373 374/* 375 * Initialize the 4MB page size flag 376 */ 377 pseflag = 0; 378/* 379 * The 4MB page version of the initial 380 * kernel page mapping. 381 */ 382 pdir4mb = 0; 383 384#if !defined(DISABLE_PSE) 385 if (cpu_feature & CPUID_PSE) { 386 unsigned ptditmp; 387 /* 388 * Note that we have enabled PSE mode 389 */ 390 pseflag = PG_PS; 391 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE)); 392 ptditmp &= ~(NBPDR - 1); 393 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 394 pdir4mb = ptditmp; 395 396#if !defined(SMP) 397 /* 398 * Enable the PSE mode. 399 */ 400 load_cr4(rcr4() | CR4_PSE); 401 402 /* 403 * We can do the mapping here for the single processor 404 * case. We simply ignore the old page table page from 405 * now on. 406 */ 407 /* 408 * For SMP, we still need 4K pages to bootstrap APs, 409 * PSE will be enabled as soon as all APs are up. 410 */ 411 PTD[KPTDI] = (pd_entry_t) ptditmp; 412 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 413 invltlb(); 414#endif 415 } 416#endif 417 418#ifdef SMP 419 if (cpu_apic_address == 0) 420 panic("pmap_bootstrap: no local apic!"); 421 422 /* local apic is mapped on last page */ 423 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag | 424 (cpu_apic_address & PG_FRAME)); 425 426 for (i = 0; i < mp_napics; i++) { 427 for (j = 0; j < mp_napics; j++) { 428 /* same page frame as a previous IO apic? */ 429 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 430 (io_apic_address[0] & PG_FRAME)) { 431 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 432 + (NPTEPG-2-j)*PAGE_SIZE); 433 break; 434 } 435 /* use this slot if available */ 436 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 0) { 437 SMPpt[NPTEPG-2-j] = (pt_entry_t)(PG_V | PG_RW | 438 pgeflag | (io_apic_address[i] & PG_FRAME)); 439 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 440 + (NPTEPG-2-j)*PAGE_SIZE); 441 break; 442 } 443 } 444 } 445 446 /* BSP does this itself, AP's get it pre-set */ 447 gd = &SMP_prvspace[0].globaldata; 448 gd->gd_prv_CMAP1 = &SMPpt[1]; 449 gd->gd_prv_CMAP2 = &SMPpt[2]; 450 gd->gd_prv_CMAP3 = &SMPpt[3]; 451 gd->gd_prv_PMAP1 = &SMPpt[4]; 452 gd->gd_prv_CADDR1 = SMP_prvspace[0].CPAGE1; 453 gd->gd_prv_CADDR2 = SMP_prvspace[0].CPAGE2; 454 gd->gd_prv_CADDR3 = SMP_prvspace[0].CPAGE3; 455 gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[0].PPAGE1; 456#endif 457 458 invltlb(); 459} 460 461#ifdef SMP 462/* 463 * Set 4mb pdir for mp startup 464 */ 465void 466pmap_set_opt(void) 467{ 468 if (pseflag && (cpu_feature & CPUID_PSE)) { 469 load_cr4(rcr4() | CR4_PSE); 470 if (pdir4mb && cpuid == 0) { /* only on BSP */ 471 kernel_pmap->pm_pdir[KPTDI] = 472 PTD[KPTDI] = (pd_entry_t)pdir4mb; 473 cpu_invltlb(); 474 } 475 } 476} 477#endif 478 479/* 480 * Initialize the pmap module. 481 * Called by vm_init, to initialize any structures that the pmap 482 * system needs to map virtual memory. 483 * pmap_init has been enhanced to support in a fairly consistant 484 * way, discontiguous physical memory. 485 */ 486void 487pmap_init(phys_start, phys_end) 488 vm_offset_t phys_start, phys_end; 489{ 490 vm_offset_t addr; 491 vm_size_t s; 492 int i; 493 int initial_pvs; 494 495 /* 496 * object for kernel page table pages 497 */ 498 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE); 499 500 /* 501 * calculate the number of pv_entries needed 502 */ 503 vm_first_phys = phys_avail[0]; 504 for (i = 0; phys_avail[i + 1]; i += 2); 505 pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 506 507 /* 508 * Allocate memory for random pmap data structures. Includes the 509 * pv_head_table. 510 */ 511 s = (vm_size_t) (sizeof(pv_table_t) * pv_npg); 512 s = round_page(s); 513 514 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 515 pv_table = (pv_table_t *) addr; 516 for(i = 0; i < pv_npg; i++) { 517 vm_offset_t pa; 518 TAILQ_INIT(&pv_table[i].pv_list); 519 pv_table[i].pv_list_count = 0; 520 pa = vm_first_phys + i * PAGE_SIZE; 521 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 522 } 523 524 /* 525 * init the pv free list 526 */ 527 initial_pvs = pv_npg; 528 if (initial_pvs < MINPV) 529 initial_pvs = MINPV; 530 pvzone = &pvzone_store; 531 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 532 initial_pvs * sizeof (struct pv_entry)); 533 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, pv_npg); 534 535 /* 536 * Now it is safe to enable pv_table recording. 537 */ 538 pmap_initialized = TRUE; 539} 540 541/* 542 * Initialize the address space (zone) for the pv_entries. Set a 543 * high water mark so that the system can recover from excessive 544 * numbers of pv entries. 545 */ 546void 547pmap_init2() { 548 pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg; 549 pv_entry_high_water = 9 * (pv_entry_max / 10); 550 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 551} 552 553/* 554 * Used to map a range of physical addresses into kernel 555 * virtual address space. 556 * 557 * For now, VM is already on, we only need to map the 558 * specified memory. 559 */ 560vm_offset_t 561pmap_map(virt, start, end, prot) 562 vm_offset_t virt; 563 vm_offset_t start; 564 vm_offset_t end; 565 int prot; 566{ 567 while (start < end) { 568 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 569 virt += PAGE_SIZE; 570 start += PAGE_SIZE; 571 } 572 return (virt); 573} 574 575 576/*************************************************** 577 * Low level helper routines..... 578 ***************************************************/ 579 580#if defined(PMAP_DIAGNOSTIC) 581 582/* 583 * This code checks for non-writeable/modified pages. 584 * This should be an invalid condition. 585 */ 586static int 587pmap_nw_modified(pt_entry_t ptea) { 588 int pte; 589 590 pte = (int) ptea; 591 592 if ((pte & (PG_M|PG_RW)) == PG_M) 593 return 1; 594 else 595 return 0; 596} 597#endif 598 599 600/* 601 * this routine defines the region(s) of memory that should 602 * not be tested for the modified bit. 603 */ 604static PMAP_INLINE int 605pmap_track_modified( vm_offset_t va) { 606 if ((va < clean_sva) || (va >= clean_eva)) 607 return 1; 608 else 609 return 0; 610} 611 612static PMAP_INLINE void 613invltlb_1pg( vm_offset_t va) { 614#if defined(I386_CPU) 615 if (cpu_class == CPUCLASS_386) { 616 invltlb(); 617 } else 618#endif 619 { 620 invlpg(va); 621 } 622} 623 624static __inline void 625pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va) 626{ 627#if defined(SMP) 628 if (pmap->pm_active & (1 << cpuid)) 629 cpu_invlpg((void *)va); 630 if (pmap->pm_active & other_cpus) 631 smp_invltlb(); 632#else 633 if (pmap->pm_active) 634 invltlb_1pg(va); 635#endif 636} 637 638static __inline void 639pmap_TLB_invalidate_all(pmap_t pmap) 640{ 641#if defined(SMP) 642 if (pmap->pm_active & (1 << cpuid)) 643 cpu_invltlb(); 644 if (pmap->pm_active & other_cpus) 645 smp_invltlb(); 646#else 647 if (pmap->pm_active) 648 invltlb(); 649#endif 650} 651 652static unsigned * 653get_ptbase(pmap) 654 pmap_t pmap; 655{ 656 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 657 658 /* are we current address space or kernel? */ 659 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 660 return (unsigned *) PTmap; 661 } 662 /* otherwise, we are alternate address space */ 663 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 664 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 665#if defined(SMP) 666 /* The page directory is not shared between CPUs */ 667 cpu_invltlb(); 668#else 669 invltlb(); 670#endif 671 } 672 return (unsigned *) APTmap; 673} 674 675/* 676 * Super fast pmap_pte routine best used when scanning 677 * the pv lists. This eliminates many coarse-grained 678 * invltlb calls. Note that many of the pv list 679 * scans are across different pmaps. It is very wasteful 680 * to do an entire invltlb for checking a single mapping. 681 */ 682 683static unsigned * 684pmap_pte_quick(pmap, va) 685 register pmap_t pmap; 686 vm_offset_t va; 687{ 688 unsigned pde, newpf; 689 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) { 690 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 691 unsigned index = i386_btop(va); 692 /* are we current address space or kernel? */ 693 if ((pmap == kernel_pmap) || 694 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 695 return (unsigned *) PTmap + index; 696 } 697 newpf = pde & PG_FRAME; 698#ifdef SMP 699 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) { 700 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V; 701 cpu_invlpg(prv_PADDR1); 702 } 703 return prv_PADDR1 + ((unsigned) index & (NPTEPG - 1)); 704#else 705 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 706 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 707 invltlb_1pg((vm_offset_t) PADDR1); 708 } 709 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 710#endif 711 } 712 return (0); 713} 714 715/* 716 * Routine: pmap_extract 717 * Function: 718 * Extract the physical page address associated 719 * with the given map/virtual_address pair. 720 */ 721vm_offset_t 722pmap_extract(pmap, va) 723 register pmap_t pmap; 724 vm_offset_t va; 725{ 726 vm_offset_t rtval; 727 vm_offset_t pdirindex; 728 pdirindex = va >> PDRSHIFT; 729 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 730 unsigned *pte; 731 if ((rtval & PG_PS) != 0) { 732 rtval &= ~(NBPDR - 1); 733 rtval |= va & (NBPDR - 1); 734 return rtval; 735 } 736 pte = get_ptbase(pmap) + i386_btop(va); 737 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 738 return rtval; 739 } 740 return 0; 741 742} 743 744/* 745 * determine if a page is managed (memory vs. device) 746 */ 747static PMAP_INLINE int 748pmap_is_managed(pa) 749 vm_offset_t pa; 750{ 751 int i; 752 753 if (!pmap_initialized) 754 return 0; 755 756 for (i = 0; phys_avail[i + 1]; i += 2) { 757 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 758 return 1; 759 } 760 return 0; 761} 762 763 764/*************************************************** 765 * Low level mapping routines..... 766 ***************************************************/ 767 768/* 769 * add a wired page to the kva 770 * note that in order for the mapping to take effect -- you 771 * should do a invltlb after doing the pmap_kenter... 772 */ 773PMAP_INLINE void 774pmap_kenter(va, pa) 775 vm_offset_t va; 776 register vm_offset_t pa; 777{ 778 register unsigned *pte; 779 unsigned npte, opte; 780 781 npte = pa | PG_RW | PG_V | pgeflag; 782 pte = (unsigned *)vtopte(va); 783 opte = *pte; 784 *pte = npte; 785 if (opte) 786 invltlb_1pg(va); 787} 788 789/* 790 * remove a page from the kernel pagetables 791 */ 792PMAP_INLINE void 793pmap_kremove(va) 794 vm_offset_t va; 795{ 796 register unsigned *pte; 797 798 pte = (unsigned *)vtopte(va); 799 *pte = 0; 800 invltlb_1pg(va); 801} 802 803/* 804 * Add a list of wired pages to the kva 805 * this routine is only used for temporary 806 * kernel mappings that do not need to have 807 * page modification or references recorded. 808 * Note that old mappings are simply written 809 * over. The page *must* be wired. 810 */ 811void 812pmap_qenter(va, m, count) 813 vm_offset_t va; 814 vm_page_t *m; 815 int count; 816{ 817 int i; 818 819 for (i = 0; i < count; i++) { 820 vm_offset_t tva = va + i * PAGE_SIZE; 821 pmap_kenter(tva, VM_PAGE_TO_PHYS(m[i])); 822 } 823} 824 825/* 826 * this routine jerks page mappings from the 827 * kernel -- it is meant only for temporary mappings. 828 */ 829void 830pmap_qremove(va, count) 831 vm_offset_t va; 832 int count; 833{ 834 vm_offset_t end_va; 835 836 end_va = va + count*PAGE_SIZE; 837 838 while (va < end_va) { 839 unsigned *pte; 840 841 pte = (unsigned *)vtopte(va); 842 *pte = 0; 843#ifdef SMP 844 cpu_invlpg((void *)va); 845#else 846 invltlb_1pg(va); 847#endif 848 va += PAGE_SIZE; 849 } 850#ifdef SMP 851 smp_invltlb(); 852#endif 853} 854 855static vm_page_t 856pmap_page_lookup(object, pindex) 857 vm_object_t object; 858 vm_pindex_t pindex; 859{ 860 vm_page_t m; 861retry: 862 m = vm_page_lookup(object, pindex); 863 if (m && vm_page_sleep_busy(m, FALSE, "pplookp")) 864 goto retry; 865 return m; 866} 867 868/* 869 * Create the UPAGES for a new process. 870 * This routine directly affects the fork perf for a process. 871 */ 872void 873pmap_new_proc(p) 874 struct proc *p; 875{ 876 int i, updateneeded; 877 vm_object_t upobj; 878 vm_page_t m; 879 struct user *up; 880 unsigned *ptek, oldpte; 881 882 /* 883 * allocate object for the upages 884 */ 885 if ((upobj = p->p_upages_obj) == NULL) { 886 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 887 p->p_upages_obj = upobj; 888 } 889 890 /* get a kernel virtual address for the UPAGES for this proc */ 891 if ((up = p->p_addr) == NULL) { 892 up = (struct user *) kmem_alloc_nofault(kernel_map, 893 UPAGES * PAGE_SIZE); 894#if !defined(MAX_PERF) 895 if (up == NULL) 896 panic("pmap_new_proc: u_map allocation failed"); 897#endif 898 p->p_addr = up; 899 } 900 901 ptek = (unsigned *) vtopte((vm_offset_t) up); 902 903 updateneeded = 0; 904 for(i=0;i<UPAGES;i++) { 905 /* 906 * Get a kernel stack page 907 */ 908 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 909 910 /* 911 * Wire the page 912 */ 913 m->wire_count++; 914 cnt.v_wire_count++; 915 916 oldpte = *(ptek + i); 917 /* 918 * Enter the page into the kernel address space. 919 */ 920 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 921 if (oldpte) { 922 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) { 923 invlpg((vm_offset_t) up + i * PAGE_SIZE); 924 } else { 925 updateneeded = 1; 926 } 927 } 928 929 vm_page_wakeup(m); 930 vm_page_flag_clear(m, PG_ZERO); 931 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 932 m->valid = VM_PAGE_BITS_ALL; 933 } 934 if (updateneeded) 935 invltlb(); 936} 937 938/* 939 * Dispose the UPAGES for a process that has exited. 940 * This routine directly impacts the exit perf of a process. 941 */ 942void 943pmap_dispose_proc(p) 944 struct proc *p; 945{ 946 int i; 947 vm_object_t upobj; 948 vm_page_t m; 949 unsigned *ptek, oldpte; 950 951 upobj = p->p_upages_obj; 952 953 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 954 for(i=0;i<UPAGES;i++) { 955 956 if ((m = vm_page_lookup(upobj, i)) == NULL) 957 panic("pmap_dispose_proc: upage already missing???"); 958 959 vm_page_busy(m); 960 961 oldpte = *(ptek + i); 962 *(ptek + i) = 0; 963 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) 964 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 965 vm_page_unwire(m, 0); 966 vm_page_free(m); 967 } 968#if defined(I386_CPU) 969 if (cpu_class <= CPUCLASS_386) 970 invltlb(); 971#endif 972} 973 974/* 975 * Allow the UPAGES for a process to be prejudicially paged out. 976 */ 977void 978pmap_swapout_proc(p) 979 struct proc *p; 980{ 981 int i; 982 vm_object_t upobj; 983 vm_page_t m; 984 985 upobj = p->p_upages_obj; 986 /* 987 * let the upages be paged 988 */ 989 for(i=0;i<UPAGES;i++) { 990 if ((m = vm_page_lookup(upobj, i)) == NULL) 991 panic("pmap_swapout_proc: upage already missing???"); 992 vm_page_dirty(m); 993 vm_page_unwire(m, 0); 994 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 995 } 996} 997 998/* 999 * Bring the UPAGES for a specified process back in. 1000 */ 1001void 1002pmap_swapin_proc(p) 1003 struct proc *p; 1004{ 1005 int i,rv; 1006 vm_object_t upobj; 1007 vm_page_t m; 1008 1009 upobj = p->p_upages_obj; 1010 for(i=0;i<UPAGES;i++) { 1011 1012 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1013 1014 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 1015 VM_PAGE_TO_PHYS(m)); 1016 1017 if (m->valid != VM_PAGE_BITS_ALL) { 1018 rv = vm_pager_get_pages(upobj, &m, 1, 0); 1019#if !defined(MAX_PERF) 1020 if (rv != VM_PAGER_OK) 1021 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 1022#endif 1023 m = vm_page_lookup(upobj, i); 1024 m->valid = VM_PAGE_BITS_ALL; 1025 } 1026 1027 vm_page_wire(m); 1028 vm_page_wakeup(m); 1029 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 1030 } 1031} 1032 1033/*************************************************** 1034 * Page table page management routines..... 1035 ***************************************************/ 1036 1037/* 1038 * This routine unholds page table pages, and if the hold count 1039 * drops to zero, then it decrements the wire count. 1040 */ 1041static int 1042_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1043 1044 while (vm_page_sleep_busy(m, FALSE, "pmuwpt")) 1045 ; 1046 1047 if (m->hold_count == 0) { 1048 vm_offset_t pteva; 1049 /* 1050 * unmap the page table page 1051 */ 1052 pmap->pm_pdir[m->pindex] = 0; 1053 --pmap->pm_stats.resident_count; 1054 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1055 (((unsigned) PTDpde) & PG_FRAME)) { 1056 /* 1057 * Do a invltlb to make the invalidated mapping 1058 * take effect immediately. 1059 */ 1060 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1061 pmap_TLB_invalidate(pmap, pteva); 1062 } 1063 1064 if (pmap->pm_ptphint == m) 1065 pmap->pm_ptphint = NULL; 1066 1067 /* 1068 * If the page is finally unwired, simply free it. 1069 */ 1070 --m->wire_count; 1071 if (m->wire_count == 0) { 1072 1073 vm_page_flash(m); 1074 vm_page_busy(m); 1075 vm_page_free_zero(m); 1076 --cnt.v_wire_count; 1077 } 1078 return 1; 1079 } 1080 return 0; 1081} 1082 1083static PMAP_INLINE int 1084pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1085 vm_page_unhold(m); 1086 if (m->hold_count == 0) 1087 return _pmap_unwire_pte_hold(pmap, m); 1088 else 1089 return 0; 1090} 1091 1092/* 1093 * After removing a page table entry, this routine is used to 1094 * conditionally free the page, and manage the hold/wire counts. 1095 */ 1096static int 1097pmap_unuse_pt(pmap, va, mpte) 1098 pmap_t pmap; 1099 vm_offset_t va; 1100 vm_page_t mpte; 1101{ 1102 unsigned ptepindex; 1103 if (va >= UPT_MIN_ADDRESS) 1104 return 0; 1105 1106 if (mpte == NULL) { 1107 ptepindex = (va >> PDRSHIFT); 1108 if (pmap->pm_ptphint && 1109 (pmap->pm_ptphint->pindex == ptepindex)) { 1110 mpte = pmap->pm_ptphint; 1111 } else { 1112 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1113 pmap->pm_ptphint = mpte; 1114 } 1115 } 1116 1117 return pmap_unwire_pte_hold(pmap, mpte); 1118} 1119 1120void 1121pmap_pinit0(pmap) 1122 struct pmap *pmap; 1123{ 1124 pmap->pm_pdir = 1125 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1126 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1127 pmap->pm_count = 1; 1128 pmap->pm_active = 0; 1129 pmap->pm_ptphint = NULL; 1130 TAILQ_INIT(&pmap->pm_pvlist); 1131 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1132} 1133 1134/* 1135 * Initialize a preallocated and zeroed pmap structure, 1136 * such as one in a vmspace structure. 1137 */ 1138void 1139pmap_pinit(pmap) 1140 register struct pmap *pmap; 1141{ 1142 vm_page_t ptdpg; 1143 1144 /* 1145 * No need to allocate page table space yet but we do need a valid 1146 * page directory table. 1147 */ 1148 if (pmap->pm_pdir == NULL) 1149 pmap->pm_pdir = 1150 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1151 1152 /* 1153 * allocate object for the ptes 1154 */ 1155 if (pmap->pm_pteobj == NULL) 1156 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1157 1158 /* 1159 * allocate the page directory page 1160 */ 1161 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1162 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1163 1164 ptdpg->wire_count = 1; 1165 ++cnt.v_wire_count; 1166 1167 1168 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/ 1169 ptdpg->valid = VM_PAGE_BITS_ALL; 1170 1171 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1172 if ((ptdpg->flags & PG_ZERO) == 0) 1173 bzero(pmap->pm_pdir, PAGE_SIZE); 1174 1175#ifdef SMP 1176 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI]; 1177#endif 1178 1179 /* install self-referential address mapping entry */ 1180 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1181 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1182 1183 pmap->pm_count = 1; 1184 pmap->pm_active = 0; 1185 pmap->pm_ptphint = NULL; 1186 TAILQ_INIT(&pmap->pm_pvlist); 1187 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1188} 1189 1190/* 1191 * Wire in kernel global address entries. To avoid a race condition 1192 * between pmap initialization and pmap_growkernel, this procedure 1193 * should be called after the vmspace is attached to the process 1194 * but before this pmap is activated. 1195 */ 1196void 1197pmap_pinit2(pmap) 1198 struct pmap *pmap; 1199{ 1200 /* XXX copies current process, does not fill in MPPTDI */ 1201 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1202} 1203 1204static int 1205pmap_release_free_page(pmap, p) 1206 struct pmap *pmap; 1207 vm_page_t p; 1208{ 1209 unsigned *pde = (unsigned *) pmap->pm_pdir; 1210 /* 1211 * This code optimizes the case of freeing non-busy 1212 * page-table pages. Those pages are zero now, and 1213 * might as well be placed directly into the zero queue. 1214 */ 1215 if (vm_page_sleep_busy(p, FALSE, "pmaprl")) 1216 return 0; 1217 1218 vm_page_busy(p); 1219 1220 /* 1221 * Remove the page table page from the processes address space. 1222 */ 1223 pde[p->pindex] = 0; 1224 pmap->pm_stats.resident_count--; 1225 1226#if !defined(MAX_PERF) 1227 if (p->hold_count) { 1228 panic("pmap_release: freeing held page table page"); 1229 } 1230#endif 1231 /* 1232 * Page directory pages need to have the kernel 1233 * stuff cleared, so they can go into the zero queue also. 1234 */ 1235 if (p->pindex == PTDPTDI) { 1236 bzero(pde + KPTDI, nkpt * PTESIZE); 1237#ifdef SMP 1238 pde[MPPTDI] = 0; 1239#endif 1240 pde[APTDPTDI] = 0; 1241 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1242 } 1243 1244 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1245 pmap->pm_ptphint = NULL; 1246 1247 p->wire_count--; 1248 cnt.v_wire_count--; 1249 vm_page_free_zero(p); 1250 return 1; 1251} 1252 1253/* 1254 * this routine is called if the page table page is not 1255 * mapped correctly. 1256 */ 1257static vm_page_t 1258_pmap_allocpte(pmap, ptepindex) 1259 pmap_t pmap; 1260 unsigned ptepindex; 1261{ 1262 vm_offset_t pteva, ptepa; 1263 vm_page_t m; 1264 1265 /* 1266 * Find or fabricate a new pagetable page 1267 */ 1268 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1269 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1270 1271 if (m->queue != PQ_NONE) { 1272 int s = splvm(); 1273 vm_page_unqueue(m); 1274 splx(s); 1275 } 1276 1277 if (m->wire_count == 0) 1278 cnt.v_wire_count++; 1279 m->wire_count++; 1280 1281 /* 1282 * Increment the hold count for the page table page 1283 * (denoting a new mapping.) 1284 */ 1285 m->hold_count++; 1286 1287 /* 1288 * Map the pagetable page into the process address space, if 1289 * it isn't already there. 1290 */ 1291 1292 pmap->pm_stats.resident_count++; 1293 1294 ptepa = VM_PAGE_TO_PHYS(m); 1295 pmap->pm_pdir[ptepindex] = 1296 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1297 1298 /* 1299 * Set the page table hint 1300 */ 1301 pmap->pm_ptphint = m; 1302 1303 /* 1304 * Try to use the new mapping, but if we cannot, then 1305 * do it with the routine that maps the page explicitly. 1306 */ 1307 if ((m->flags & PG_ZERO) == 0) { 1308 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1309 (((unsigned) PTDpde) & PG_FRAME)) { 1310 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1311 bzero((caddr_t) pteva, PAGE_SIZE); 1312 } else { 1313 pmap_zero_page(ptepa); 1314 } 1315 } 1316 1317 m->valid = VM_PAGE_BITS_ALL; 1318 vm_page_flag_clear(m, PG_ZERO); 1319 vm_page_flag_set(m, PG_MAPPED); 1320 vm_page_wakeup(m); 1321 1322 return m; 1323} 1324 1325static vm_page_t 1326pmap_allocpte(pmap, va) 1327 pmap_t pmap; 1328 vm_offset_t va; 1329{ 1330 unsigned ptepindex; 1331 vm_offset_t ptepa; 1332 vm_page_t m; 1333 1334 /* 1335 * Calculate pagetable page index 1336 */ 1337 ptepindex = va >> PDRSHIFT; 1338 1339 /* 1340 * Get the page directory entry 1341 */ 1342 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1343 1344 /* 1345 * This supports switching from a 4MB page to a 1346 * normal 4K page. 1347 */ 1348 if (ptepa & PG_PS) { 1349 pmap->pm_pdir[ptepindex] = 0; 1350 ptepa = 0; 1351 invltlb(); 1352 } 1353 1354 /* 1355 * If the page table page is mapped, we just increment the 1356 * hold count, and activate it. 1357 */ 1358 if (ptepa) { 1359 /* 1360 * In order to get the page table page, try the 1361 * hint first. 1362 */ 1363 if (pmap->pm_ptphint && 1364 (pmap->pm_ptphint->pindex == ptepindex)) { 1365 m = pmap->pm_ptphint; 1366 } else { 1367 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1368 pmap->pm_ptphint = m; 1369 } 1370 m->hold_count++; 1371 return m; 1372 } 1373 /* 1374 * Here if the pte page isn't mapped, or if it has been deallocated. 1375 */ 1376 return _pmap_allocpte(pmap, ptepindex); 1377} 1378 1379 1380/*************************************************** 1381* Pmap allocation/deallocation routines. 1382 ***************************************************/ 1383 1384/* 1385 * Release any resources held by the given physical map. 1386 * Called when a pmap initialized by pmap_pinit is being released. 1387 * Should only be called if the map contains no valid mappings. 1388 */ 1389void 1390pmap_release(pmap) 1391 register struct pmap *pmap; 1392{ 1393 vm_page_t p,n,ptdpg; 1394 vm_object_t object = pmap->pm_pteobj; 1395 int curgeneration; 1396 1397#if defined(DIAGNOSTIC) 1398 if (object->ref_count != 1) 1399 panic("pmap_release: pteobj reference count != 1"); 1400#endif 1401 1402 ptdpg = NULL; 1403retry: 1404 curgeneration = object->generation; 1405 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1406 n = TAILQ_NEXT(p, listq); 1407 if (p->pindex == PTDPTDI) { 1408 ptdpg = p; 1409 continue; 1410 } 1411 while (1) { 1412 if (!pmap_release_free_page(pmap, p) && 1413 (object->generation != curgeneration)) 1414 goto retry; 1415 } 1416 } 1417 1418 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1419 goto retry; 1420} 1421 1422/* 1423 * grow the number of kernel page table entries, if needed 1424 */ 1425void 1426pmap_growkernel(vm_offset_t addr) 1427{ 1428 struct proc *p; 1429 struct pmap *pmap; 1430 int s; 1431 vm_offset_t ptppaddr; 1432 vm_page_t nkpg; 1433 pd_entry_t newpdir; 1434 1435 s = splhigh(); 1436 if (kernel_vm_end == 0) { 1437 kernel_vm_end = KERNBASE; 1438 nkpt = 0; 1439 while (pdir_pde(PTD, kernel_vm_end)) { 1440 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1441 nkpt++; 1442 } 1443 } 1444 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1445 while (kernel_vm_end < addr) { 1446 if (pdir_pde(PTD, kernel_vm_end)) { 1447 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1448 continue; 1449 } 1450 1451 /* 1452 * This index is bogus, but out of the way 1453 */ 1454 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1455#if !defined(MAX_PERF) 1456 if (!nkpg) 1457 panic("pmap_growkernel: no memory to grow kernel"); 1458#endif 1459 1460 nkpt++; 1461 1462 vm_page_wire(nkpg); 1463 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1464 pmap_zero_page(ptppaddr); 1465 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1466 pdir_pde(PTD, kernel_vm_end) = newpdir; 1467 1468 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1469 if (p->p_vmspace) { 1470 pmap = vmspace_pmap(p->p_vmspace); 1471 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1472 } 1473 } 1474 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir; 1475 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1476 } 1477 splx(s); 1478} 1479 1480/* 1481 * Retire the given physical map from service. 1482 * Should only be called if the map contains 1483 * no valid mappings. 1484 */ 1485void 1486pmap_destroy(pmap) 1487 register pmap_t pmap; 1488{ 1489 int count; 1490 1491 if (pmap == NULL) 1492 return; 1493 1494 count = --pmap->pm_count; 1495 if (count == 0) { 1496 pmap_release(pmap); 1497#if !defined(MAX_PERF) 1498 panic("destroying a pmap is not yet implemented"); 1499#endif 1500 } 1501} 1502 1503/* 1504 * Add a reference to the specified pmap. 1505 */ 1506void 1507pmap_reference(pmap) 1508 pmap_t pmap; 1509{ 1510 if (pmap != NULL) { 1511 pmap->pm_count++; 1512 } 1513} 1514 1515/*************************************************** 1516* page management routines. 1517 ***************************************************/ 1518 1519/* 1520 * free the pv_entry back to the free list 1521 */ 1522static PMAP_INLINE void 1523free_pv_entry(pv) 1524 pv_entry_t pv; 1525{ 1526 pv_entry_count--; 1527 zfreei(pvzone, pv); 1528} 1529 1530/* 1531 * get a new pv_entry, allocating a block from the system 1532 * when needed. 1533 * the memory allocation is performed bypassing the malloc code 1534 * because of the possibility of allocations at interrupt time. 1535 */ 1536static pv_entry_t 1537get_pv_entry(void) 1538{ 1539 pv_entry_count++; 1540 if (pv_entry_high_water && 1541 (pv_entry_count > pv_entry_high_water) && 1542 (pmap_pagedaemon_waken == 0)) { 1543 pmap_pagedaemon_waken = 1; 1544 wakeup (&vm_pages_needed); 1545 } 1546 return zalloci(pvzone); 1547} 1548 1549/* 1550 * This routine is very drastic, but can save the system 1551 * in a pinch. 1552 */ 1553void 1554pmap_collect() { 1555 pv_table_t *ppv; 1556 int i; 1557 vm_offset_t pa; 1558 vm_page_t m; 1559 static int warningdone=0; 1560 1561 if (pmap_pagedaemon_waken == 0) 1562 return; 1563 1564 if (warningdone < 5) { 1565 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1566 warningdone++; 1567 } 1568 1569 for(i = 0; i < pv_npg; i++) { 1570 if ((ppv = &pv_table[i]) == 0) 1571 continue; 1572 m = ppv->pv_vm_page; 1573 if ((pa = VM_PAGE_TO_PHYS(m)) == 0) 1574 continue; 1575 if (m->wire_count || m->hold_count || m->busy || 1576 (m->flags & PG_BUSY)) 1577 continue; 1578 pmap_remove_all(pa); 1579 } 1580 pmap_pagedaemon_waken = 0; 1581} 1582 1583 1584/* 1585 * If it is the first entry on the list, it is actually 1586 * in the header and we must copy the following entry up 1587 * to the header. Otherwise we must search the list for 1588 * the entry. In either case we free the now unused entry. 1589 */ 1590 1591static int 1592pmap_remove_entry(pmap, ppv, va) 1593 struct pmap *pmap; 1594 pv_table_t *ppv; 1595 vm_offset_t va; 1596{ 1597 pv_entry_t pv; 1598 int rtval; 1599 int s; 1600 1601 s = splvm(); 1602 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1603 for (pv = TAILQ_FIRST(&ppv->pv_list); 1604 pv; 1605 pv = TAILQ_NEXT(pv, pv_list)) { 1606 if (pmap == pv->pv_pmap && va == pv->pv_va) 1607 break; 1608 } 1609 } else { 1610 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1611 pv; 1612 pv = TAILQ_NEXT(pv, pv_plist)) { 1613 if (va == pv->pv_va) 1614 break; 1615 } 1616 } 1617 1618 rtval = 0; 1619 if (pv) { 1620 1621 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1622 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1623 ppv->pv_list_count--; 1624 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 1625 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 1626 1627 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1628 free_pv_entry(pv); 1629 } 1630 1631 splx(s); 1632 return rtval; 1633} 1634 1635/* 1636 * Create a pv entry for page at pa for 1637 * (pmap, va). 1638 */ 1639static void 1640pmap_insert_entry(pmap, va, mpte, pa) 1641 pmap_t pmap; 1642 vm_offset_t va; 1643 vm_page_t mpte; 1644 vm_offset_t pa; 1645{ 1646 1647 int s; 1648 pv_entry_t pv; 1649 pv_table_t *ppv; 1650 1651 s = splvm(); 1652 pv = get_pv_entry(); 1653 pv->pv_va = va; 1654 pv->pv_pmap = pmap; 1655 pv->pv_ptem = mpte; 1656 1657 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1658 1659 ppv = pa_to_pvh(pa); 1660 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1661 ppv->pv_list_count++; 1662 1663 splx(s); 1664} 1665 1666/* 1667 * pmap_remove_pte: do the things to unmap a page in a process 1668 */ 1669static int 1670pmap_remove_pte(pmap, ptq, va) 1671 struct pmap *pmap; 1672 unsigned *ptq; 1673 vm_offset_t va; 1674{ 1675 unsigned oldpte; 1676 pv_table_t *ppv; 1677 1678 oldpte = loadandclear(ptq); 1679 if (oldpte & PG_W) 1680 pmap->pm_stats.wired_count -= 1; 1681 /* 1682 * Machines that don't support invlpg, also don't support 1683 * PG_G. 1684 */ 1685 if (oldpte & PG_G) 1686 invlpg(va); 1687 pmap->pm_stats.resident_count -= 1; 1688 if (oldpte & PG_MANAGED) { 1689 ppv = pa_to_pvh(oldpte); 1690 if (oldpte & PG_M) { 1691#if defined(PMAP_DIAGNOSTIC) 1692 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1693 printf( 1694 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n", 1695 va, oldpte); 1696 } 1697#endif 1698 if (pmap_track_modified(va)) 1699 vm_page_dirty(ppv->pv_vm_page); 1700 } 1701 if (oldpte & PG_A) 1702 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1703 return pmap_remove_entry(pmap, ppv, va); 1704 } else { 1705 return pmap_unuse_pt(pmap, va, NULL); 1706 } 1707 1708 return 0; 1709} 1710 1711/* 1712 * Remove a single page from a process address space 1713 */ 1714static void 1715pmap_remove_page(pmap, va) 1716 struct pmap *pmap; 1717 register vm_offset_t va; 1718{ 1719 register unsigned *ptq; 1720 1721 /* 1722 * if there is no pte for this address, just skip it!!! 1723 */ 1724 if (*pmap_pde(pmap, va) == 0) { 1725 return; 1726 } 1727 1728 /* 1729 * get a local va for mappings for this pmap. 1730 */ 1731 ptq = get_ptbase(pmap) + i386_btop(va); 1732 if (*ptq) { 1733 (void) pmap_remove_pte(pmap, ptq, va); 1734 pmap_TLB_invalidate(pmap, va); 1735 } 1736 return; 1737} 1738 1739/* 1740 * Remove the given range of addresses from the specified map. 1741 * 1742 * It is assumed that the start and end are properly 1743 * rounded to the page size. 1744 */ 1745void 1746pmap_remove(pmap, sva, eva) 1747 struct pmap *pmap; 1748 register vm_offset_t sva; 1749 register vm_offset_t eva; 1750{ 1751 register unsigned *ptbase; 1752 vm_offset_t pdnxt; 1753 vm_offset_t ptpaddr; 1754 vm_offset_t sindex, eindex; 1755 int anyvalid; 1756 1757 if (pmap == NULL) 1758 return; 1759 1760 if (pmap->pm_stats.resident_count == 0) 1761 return; 1762 1763 /* 1764 * special handling of removing one page. a very 1765 * common operation and easy to short circuit some 1766 * code. 1767 */ 1768 if (((sva + PAGE_SIZE) == eva) && 1769 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1770 pmap_remove_page(pmap, sva); 1771 return; 1772 } 1773 1774 anyvalid = 0; 1775 1776 /* 1777 * Get a local virtual address for the mappings that are being 1778 * worked with. 1779 */ 1780 ptbase = get_ptbase(pmap); 1781 1782 sindex = i386_btop(sva); 1783 eindex = i386_btop(eva); 1784 1785 for (; sindex < eindex; sindex = pdnxt) { 1786 unsigned pdirindex; 1787 1788 /* 1789 * Calculate index for next page table. 1790 */ 1791 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1792 if (pmap->pm_stats.resident_count == 0) 1793 break; 1794 1795 pdirindex = sindex / NPDEPG; 1796 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1797 pmap->pm_pdir[pdirindex] = 0; 1798 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1799 anyvalid++; 1800 continue; 1801 } 1802 1803 /* 1804 * Weed out invalid mappings. Note: we assume that the page 1805 * directory table is always allocated, and in kernel virtual. 1806 */ 1807 if (ptpaddr == 0) 1808 continue; 1809 1810 /* 1811 * Limit our scan to either the end of the va represented 1812 * by the current page table page, or to the end of the 1813 * range being removed. 1814 */ 1815 if (pdnxt > eindex) { 1816 pdnxt = eindex; 1817 } 1818 1819 for ( ;sindex != pdnxt; sindex++) { 1820 vm_offset_t va; 1821 if (ptbase[sindex] == 0) { 1822 continue; 1823 } 1824 va = i386_ptob(sindex); 1825 1826 anyvalid++; 1827 if (pmap_remove_pte(pmap, 1828 ptbase + sindex, va)) 1829 break; 1830 } 1831 } 1832 1833 if (anyvalid) 1834 pmap_TLB_invalidate_all(pmap); 1835} 1836 1837/* 1838 * Routine: pmap_remove_all 1839 * Function: 1840 * Removes this physical page from 1841 * all physical maps in which it resides. 1842 * Reflects back modify bits to the pager. 1843 * 1844 * Notes: 1845 * Original versions of this routine were very 1846 * inefficient because they iteratively called 1847 * pmap_remove (slow...) 1848 */ 1849 1850static void 1851pmap_remove_all(pa) 1852 vm_offset_t pa; 1853{ 1854 register pv_entry_t pv; 1855 pv_table_t *ppv; 1856 register unsigned *pte, tpte; 1857 int s; 1858 1859#if defined(PMAP_DIAGNOSTIC) 1860 /* 1861 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1862 * pages! 1863 */ 1864 if (!pmap_is_managed(pa)) { 1865 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", pa); 1866 } 1867#endif 1868 1869 s = splvm(); 1870 ppv = pa_to_pvh(pa); 1871 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1872 pv->pv_pmap->pm_stats.resident_count--; 1873 1874 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1875 1876 tpte = loadandclear(pte); 1877 if (tpte & PG_W) 1878 pv->pv_pmap->pm_stats.wired_count--; 1879 1880 if (tpte & PG_A) 1881 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1882 1883 /* 1884 * Update the vm_page_t clean and reference bits. 1885 */ 1886 if (tpte & PG_M) { 1887#if defined(PMAP_DIAGNOSTIC) 1888 if (pmap_nw_modified((pt_entry_t) tpte)) { 1889 printf( 1890 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n", 1891 pv->pv_va, tpte); 1892 } 1893#endif 1894 if (pmap_track_modified(pv->pv_va)) 1895 vm_page_dirty(ppv->pv_vm_page); 1896 } 1897 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 1898 1899 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1900 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1901 ppv->pv_list_count--; 1902 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1903 free_pv_entry(pv); 1904 } 1905 1906 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 1907 1908 splx(s); 1909} 1910 1911/* 1912 * Set the physical protection on the 1913 * specified range of this map as requested. 1914 */ 1915void 1916pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1917{ 1918 register unsigned *ptbase; 1919 vm_offset_t pdnxt, ptpaddr; 1920 vm_pindex_t sindex, eindex; 1921 int anychanged; 1922 1923 1924 if (pmap == NULL) 1925 return; 1926 1927 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1928 pmap_remove(pmap, sva, eva); 1929 return; 1930 } 1931 1932 if (prot & VM_PROT_WRITE) 1933 return; 1934 1935 anychanged = 0; 1936 1937 ptbase = get_ptbase(pmap); 1938 1939 sindex = i386_btop(sva); 1940 eindex = i386_btop(eva); 1941 1942 for (; sindex < eindex; sindex = pdnxt) { 1943 1944 unsigned pdirindex; 1945 1946 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1947 1948 pdirindex = sindex / NPDEPG; 1949 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1950 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1951 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1952 anychanged++; 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 if (pdnxt > eindex) { 1964 pdnxt = eindex; 1965 } 1966 1967 for (; sindex != pdnxt; sindex++) { 1968 1969 unsigned pbits; 1970 pv_table_t *ppv; 1971 1972 pbits = ptbase[sindex]; 1973 1974 if (pbits & PG_MANAGED) { 1975 ppv = NULL; 1976 if (pbits & PG_A) { 1977 ppv = pa_to_pvh(pbits); 1978 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1979 pbits &= ~PG_A; 1980 } 1981 if (pbits & PG_M) { 1982 if (pmap_track_modified(i386_ptob(sindex))) { 1983 if (ppv == NULL) 1984 ppv = pa_to_pvh(pbits); 1985 vm_page_dirty(ppv->pv_vm_page); 1986 pbits &= ~PG_M; 1987 } 1988 } 1989 } 1990 1991 pbits &= ~PG_RW; 1992 1993 if (pbits != ptbase[sindex]) { 1994 ptbase[sindex] = pbits; 1995 anychanged = 1; 1996 } 1997 } 1998 } 1999 if (anychanged) 2000 pmap_TLB_invalidate_all(pmap); 2001} 2002 2003/* 2004 * Insert the given physical page (p) at 2005 * the specified virtual address (v) in the 2006 * target physical map with the protection requested. 2007 * 2008 * If specified, the page will be wired down, meaning 2009 * that the related pte can not be reclaimed. 2010 * 2011 * NB: This is the only routine which MAY NOT lazy-evaluate 2012 * or lose information. That is, this routine must actually 2013 * insert this page into the given map NOW. 2014 */ 2015void 2016pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot, 2017 boolean_t wired) 2018{ 2019 register unsigned *pte; 2020 vm_offset_t opa; 2021 vm_offset_t origpte, newpte; 2022 vm_page_t mpte; 2023 2024 if (pmap == NULL) 2025 return; 2026 2027 va &= PG_FRAME; 2028#ifdef PMAP_DIAGNOSTIC 2029 if (va > VM_MAX_KERNEL_ADDRESS) 2030 panic("pmap_enter: toobig"); 2031 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 2032 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 2033#endif 2034 2035 mpte = NULL; 2036 /* 2037 * In the case that a page table page is not 2038 * resident, we are creating it here. 2039 */ 2040 if (va < UPT_MIN_ADDRESS) { 2041 mpte = pmap_allocpte(pmap, va); 2042 } 2043#if 0 && defined(PMAP_DIAGNOSTIC) 2044 else { 2045 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 2046 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 2047 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 2048 pmap->pm_pdir[PTDPTDI], origpte, va); 2049 } 2050 if (smp_active) { 2051 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid]; 2052 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2053 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2054 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2055 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr); 2056 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2057 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2058 } 2059 } 2060 } 2061#endif 2062 2063 pte = pmap_pte(pmap, va); 2064 2065#if !defined(MAX_PERF) 2066 /* 2067 * Page Directory table entry not valid, we need a new PT page 2068 */ 2069 if (pte == NULL) { 2070 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n", 2071 (void *)pmap->pm_pdir[PTDPTDI], va); 2072 } 2073#endif 2074 2075 origpte = *(vm_offset_t *)pte; 2076 pa &= PG_FRAME; 2077 opa = origpte & PG_FRAME; 2078 2079#if !defined(MAX_PERF) 2080 if (origpte & PG_PS) 2081 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2082#endif 2083 2084 /* 2085 * Mapping has not changed, must be protection or wiring change. 2086 */ 2087 if (origpte && (opa == pa)) { 2088 /* 2089 * Wiring change, just update stats. We don't worry about 2090 * wiring PT pages as they remain resident as long as there 2091 * are valid mappings in them. Hence, if a user page is wired, 2092 * the PT page will be also. 2093 */ 2094 if (wired && ((origpte & PG_W) == 0)) 2095 pmap->pm_stats.wired_count++; 2096 else if (!wired && (origpte & PG_W)) 2097 pmap->pm_stats.wired_count--; 2098 2099#if defined(PMAP_DIAGNOSTIC) 2100 if (pmap_nw_modified((pt_entry_t) origpte)) { 2101 printf( 2102 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n", 2103 va, origpte); 2104 } 2105#endif 2106 2107 /* 2108 * Remove extra pte reference 2109 */ 2110 if (mpte) 2111 mpte->hold_count--; 2112 2113 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2114 if ((origpte & PG_RW) == 0) { 2115 *pte |= PG_RW; 2116#ifdef SMP 2117 cpu_invlpg((void *)va); 2118 if (pmap->pm_active & other_cpus) 2119 smp_invltlb(); 2120#else 2121 invltlb_1pg(va); 2122#endif 2123 } 2124 return; 2125 } 2126 2127 /* 2128 * We might be turning off write access to the page, 2129 * so we go ahead and sense modify status. 2130 */ 2131 if (origpte & PG_MANAGED) { 2132 if ((origpte & PG_M) && pmap_track_modified(va)) { 2133 pv_table_t *ppv; 2134 ppv = pa_to_pvh(opa); 2135 vm_page_dirty(ppv->pv_vm_page); 2136 } 2137 pa |= PG_MANAGED; 2138 } 2139 goto validate; 2140 } 2141 /* 2142 * Mapping has changed, invalidate old range and fall through to 2143 * handle validating new mapping. 2144 */ 2145 if (opa) { 2146 int err; 2147 err = pmap_remove_pte(pmap, pte, va); 2148#if !defined(MAX_PERF) 2149 if (err) 2150 panic("pmap_enter: pte vanished, va: 0x%x", va); 2151#endif 2152 } 2153 2154 /* 2155 * Enter on the PV list if part of our managed memory Note that we 2156 * raise IPL while manipulating pv_table since pmap_enter can be 2157 * called at interrupt time. 2158 */ 2159 if (pmap_is_managed(pa)) { 2160 pmap_insert_entry(pmap, va, mpte, pa); 2161 pa |= PG_MANAGED; 2162 } 2163 2164 /* 2165 * Increment counters 2166 */ 2167 pmap->pm_stats.resident_count++; 2168 if (wired) 2169 pmap->pm_stats.wired_count++; 2170 2171validate: 2172 /* 2173 * Now validate mapping with desired protection/wiring. 2174 */ 2175 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2176 2177 if (wired) 2178 newpte |= PG_W; 2179 if (va < UPT_MIN_ADDRESS) 2180 newpte |= PG_U; 2181 if (pmap == kernel_pmap) 2182 newpte |= pgeflag; 2183 2184 /* 2185 * if the mapping or permission bits are different, we need 2186 * to update the pte. 2187 */ 2188 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2189 *pte = newpte | PG_A; 2190 if (origpte) { 2191#ifdef SMP 2192 cpu_invlpg((void *)va); 2193 if (pmap->pm_active & other_cpus) 2194 smp_invltlb(); 2195#else 2196 invltlb_1pg(va); 2197#endif 2198 } 2199 } 2200} 2201 2202/* 2203 * this code makes some *MAJOR* assumptions: 2204 * 1. Current pmap & pmap exists. 2205 * 2. Not wired. 2206 * 3. Read access. 2207 * 4. No page table pages. 2208 * 5. Tlbflush is deferred to calling procedure. 2209 * 6. Page IS managed. 2210 * but is *MUCH* faster than pmap_enter... 2211 */ 2212 2213static vm_page_t 2214pmap_enter_quick(pmap, va, pa, mpte) 2215 register pmap_t pmap; 2216 vm_offset_t va; 2217 register vm_offset_t pa; 2218 vm_page_t mpte; 2219{ 2220 register unsigned *pte; 2221 2222 /* 2223 * In the case that a page table page is not 2224 * resident, we are creating it here. 2225 */ 2226 if (va < UPT_MIN_ADDRESS) { 2227 unsigned ptepindex; 2228 vm_offset_t ptepa; 2229 2230 /* 2231 * Calculate pagetable page index 2232 */ 2233 ptepindex = va >> PDRSHIFT; 2234 if (mpte && (mpte->pindex == ptepindex)) { 2235 mpte->hold_count++; 2236 } else { 2237retry: 2238 /* 2239 * Get the page directory entry 2240 */ 2241 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2242 2243 /* 2244 * If the page table page is mapped, we just increment 2245 * the hold count, and activate it. 2246 */ 2247 if (ptepa) { 2248#if !defined(MAX_PERF) 2249 if (ptepa & PG_PS) 2250 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2251#endif 2252 if (pmap->pm_ptphint && 2253 (pmap->pm_ptphint->pindex == ptepindex)) { 2254 mpte = pmap->pm_ptphint; 2255 } else { 2256 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2257 pmap->pm_ptphint = mpte; 2258 } 2259 if (mpte == NULL) 2260 goto retry; 2261 mpte->hold_count++; 2262 } else { 2263 mpte = _pmap_allocpte(pmap, ptepindex); 2264 } 2265 } 2266 } else { 2267 mpte = NULL; 2268 } 2269 2270 /* 2271 * This call to vtopte makes the assumption that we are 2272 * entering the page into the current pmap. In order to support 2273 * quick entry into any pmap, one would likely use pmap_pte_quick. 2274 * But that isn't as quick as vtopte. 2275 */ 2276 pte = (unsigned *)vtopte(va); 2277 if (*pte) { 2278 if (mpte) 2279 pmap_unwire_pte_hold(pmap, mpte); 2280 return 0; 2281 } 2282 2283 /* 2284 * Enter on the PV list if part of our managed memory Note that we 2285 * raise IPL while manipulating pv_table since pmap_enter can be 2286 * called at interrupt time. 2287 */ 2288 pmap_insert_entry(pmap, va, mpte, pa); 2289 2290 /* 2291 * Increment counters 2292 */ 2293 pmap->pm_stats.resident_count++; 2294 2295 /* 2296 * Now validate mapping with RO protection 2297 */ 2298 *pte = pa | PG_V | PG_U | PG_MANAGED; 2299 2300 return mpte; 2301} 2302 2303#define MAX_INIT_PT (96) 2304/* 2305 * pmap_object_init_pt preloads the ptes for a given object 2306 * into the specified pmap. This eliminates the blast of soft 2307 * faults on process startup and immediately after an mmap. 2308 */ 2309void 2310pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2311 pmap_t pmap; 2312 vm_offset_t addr; 2313 vm_object_t object; 2314 vm_pindex_t pindex; 2315 vm_size_t size; 2316 int limit; 2317{ 2318 vm_offset_t tmpidx; 2319 int psize; 2320 vm_page_t p, mpte; 2321 int objpgs; 2322 2323 if (pmap == NULL || object == NULL) 2324 return; 2325 2326 /* 2327 * This code maps large physical mmap regions into the 2328 * processor address space. Note that some shortcuts 2329 * are taken, but the code works. 2330 */ 2331 if (pseflag && 2332 (object->type == OBJT_DEVICE) && 2333 ((addr & (NBPDR - 1)) == 0) && 2334 ((size & (NBPDR - 1)) == 0) ) { 2335 int i; 2336 vm_page_t m[1]; 2337 unsigned int ptepindex; 2338 int npdes; 2339 vm_offset_t ptepa; 2340 2341 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2342 return; 2343 2344retry: 2345 p = vm_page_lookup(object, pindex); 2346 if (p && vm_page_sleep_busy(p, FALSE, "init4p")) 2347 goto retry; 2348 2349 if (p == NULL) { 2350 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2351 if (p == NULL) 2352 return; 2353 m[0] = p; 2354 2355 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2356 vm_page_free(p); 2357 return; 2358 } 2359 2360 p = vm_page_lookup(object, pindex); 2361 vm_page_wakeup(p); 2362 } 2363 2364 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2365 if (ptepa & (NBPDR - 1)) { 2366 return; 2367 } 2368 2369 p->valid = VM_PAGE_BITS_ALL; 2370 2371 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2372 npdes = size >> PDRSHIFT; 2373 for(i=0;i<npdes;i++) { 2374 pmap->pm_pdir[ptepindex] = 2375 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2376 ptepa += NBPDR; 2377 ptepindex += 1; 2378 } 2379 vm_page_flag_set(p, PG_MAPPED); 2380 invltlb(); 2381 return; 2382 } 2383 2384 psize = i386_btop(size); 2385 2386 if ((object->type != OBJT_VNODE) || 2387 (limit && (psize > MAX_INIT_PT) && 2388 (object->resident_page_count > MAX_INIT_PT))) { 2389 return; 2390 } 2391 2392 if (psize + pindex > object->size) 2393 psize = object->size - pindex; 2394 2395 mpte = NULL; 2396 /* 2397 * if we are processing a major portion of the object, then scan the 2398 * entire thing. 2399 */ 2400 if (psize > (object->resident_page_count >> 2)) { 2401 objpgs = psize; 2402 2403 for (p = TAILQ_FIRST(&object->memq); 2404 ((objpgs > 0) && (p != NULL)); 2405 p = TAILQ_NEXT(p, listq)) { 2406 2407 tmpidx = p->pindex; 2408 if (tmpidx < pindex) { 2409 continue; 2410 } 2411 tmpidx -= pindex; 2412 if (tmpidx >= psize) { 2413 continue; 2414 } 2415 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2416 (p->busy == 0) && 2417 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2418 if ((p->queue - p->pc) == PQ_CACHE) 2419 vm_page_deactivate(p); 2420 vm_page_busy(p); 2421 mpte = pmap_enter_quick(pmap, 2422 addr + i386_ptob(tmpidx), 2423 VM_PAGE_TO_PHYS(p), mpte); 2424 vm_page_flag_set(p, PG_MAPPED); 2425 vm_page_wakeup(p); 2426 } 2427 objpgs -= 1; 2428 } 2429 } else { 2430 /* 2431 * else lookup the pages one-by-one. 2432 */ 2433 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2434 p = vm_page_lookup(object, tmpidx + pindex); 2435 if (p && 2436 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2437 (p->busy == 0) && 2438 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2439 if ((p->queue - p->pc) == PQ_CACHE) 2440 vm_page_deactivate(p); 2441 vm_page_busy(p); 2442 mpte = pmap_enter_quick(pmap, 2443 addr + i386_ptob(tmpidx), 2444 VM_PAGE_TO_PHYS(p), mpte); 2445 vm_page_flag_set(p, PG_MAPPED); 2446 vm_page_wakeup(p); 2447 } 2448 } 2449 } 2450 return; 2451} 2452 2453/* 2454 * pmap_prefault provides a quick way of clustering 2455 * pagefaults into a processes address space. It is a "cousin" 2456 * of pmap_object_init_pt, except it runs at page fault time instead 2457 * of mmap time. 2458 */ 2459#define PFBAK 4 2460#define PFFOR 4 2461#define PAGEORDER_SIZE (PFBAK+PFFOR) 2462 2463static int pmap_prefault_pageorder[] = { 2464 -PAGE_SIZE, PAGE_SIZE, 2465 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2466 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2467 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2468}; 2469 2470void 2471pmap_prefault(pmap, addra, entry) 2472 pmap_t pmap; 2473 vm_offset_t addra; 2474 vm_map_entry_t entry; 2475{ 2476 int i; 2477 vm_offset_t starta; 2478 vm_offset_t addr; 2479 vm_pindex_t pindex; 2480 vm_page_t m, mpte; 2481 vm_object_t object; 2482 2483 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) 2484 return; 2485 2486 object = entry->object.vm_object; 2487 2488 starta = addra - PFBAK * PAGE_SIZE; 2489 if (starta < entry->start) { 2490 starta = entry->start; 2491 } else if (starta > addra) { 2492 starta = 0; 2493 } 2494 2495 mpte = NULL; 2496 for (i = 0; i < PAGEORDER_SIZE; i++) { 2497 vm_object_t lobject; 2498 unsigned *pte; 2499 2500 addr = addra + pmap_prefault_pageorder[i]; 2501 if (addr > addra + (PFFOR * PAGE_SIZE)) 2502 addr = 0; 2503 2504 if (addr < starta || addr >= entry->end) 2505 continue; 2506 2507 if ((*pmap_pde(pmap, addr)) == NULL) 2508 continue; 2509 2510 pte = (unsigned *) vtopte(addr); 2511 if (*pte) 2512 continue; 2513 2514 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2515 lobject = object; 2516 for (m = vm_page_lookup(lobject, pindex); 2517 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2518 lobject = lobject->backing_object) { 2519 if (lobject->backing_object_offset & PAGE_MASK) 2520 break; 2521 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2522 m = vm_page_lookup(lobject->backing_object, pindex); 2523 } 2524 2525 /* 2526 * give-up when a page is not in memory 2527 */ 2528 if (m == NULL) 2529 break; 2530 2531 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2532 (m->busy == 0) && 2533 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2534 2535 if ((m->queue - m->pc) == PQ_CACHE) { 2536 vm_page_deactivate(m); 2537 } 2538 vm_page_busy(m); 2539 mpte = pmap_enter_quick(pmap, addr, 2540 VM_PAGE_TO_PHYS(m), mpte); 2541 vm_page_flag_set(m, PG_MAPPED); 2542 vm_page_wakeup(m); 2543 } 2544 } 2545} 2546 2547/* 2548 * Routine: pmap_change_wiring 2549 * Function: Change the wiring attribute for a map/virtual-address 2550 * pair. 2551 * In/out conditions: 2552 * The mapping must already exist in the pmap. 2553 */ 2554void 2555pmap_change_wiring(pmap, va, wired) 2556 register pmap_t pmap; 2557 vm_offset_t va; 2558 boolean_t wired; 2559{ 2560 register unsigned *pte; 2561 2562 if (pmap == NULL) 2563 return; 2564 2565 pte = pmap_pte(pmap, va); 2566 2567 if (wired && !pmap_pte_w(pte)) 2568 pmap->pm_stats.wired_count++; 2569 else if (!wired && pmap_pte_w(pte)) 2570 pmap->pm_stats.wired_count--; 2571 2572 /* 2573 * Wiring is not a hardware characteristic so there is no need to 2574 * invalidate TLB. 2575 */ 2576 pmap_pte_set_w(pte, wired); 2577} 2578 2579 2580 2581/* 2582 * Copy the range specified by src_addr/len 2583 * from the source map to the range dst_addr/len 2584 * in the destination map. 2585 * 2586 * This routine is only advisory and need not do anything. 2587 */ 2588 2589void 2590pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2591 pmap_t dst_pmap, src_pmap; 2592 vm_offset_t dst_addr; 2593 vm_size_t len; 2594 vm_offset_t src_addr; 2595{ 2596 vm_offset_t addr; 2597 vm_offset_t end_addr = src_addr + len; 2598 vm_offset_t pdnxt; 2599 unsigned src_frame, dst_frame; 2600 2601 if (dst_addr != src_addr) 2602 return; 2603 2604 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2605 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2606 return; 2607 } 2608 2609 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2610 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2611 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2612#if defined(SMP) 2613 /* The page directory is not shared between CPUs */ 2614 cpu_invltlb(); 2615#else 2616 invltlb(); 2617#endif 2618 } 2619 2620 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2621 unsigned *src_pte, *dst_pte; 2622 vm_page_t dstmpte, srcmpte; 2623 vm_offset_t srcptepaddr; 2624 unsigned ptepindex; 2625 2626#if !defined(MAX_PERF) 2627 if (addr >= UPT_MIN_ADDRESS) 2628 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2629#endif 2630 2631 /* 2632 * Don't let optional prefaulting of pages make us go 2633 * way below the low water mark of free pages or way 2634 * above high water mark of used pv entries. 2635 */ 2636 if (cnt.v_free_count < cnt.v_free_reserved || 2637 pv_entry_count > pv_entry_high_water) 2638 break; 2639 2640 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2641 ptepindex = addr >> PDRSHIFT; 2642 2643 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2644 if (srcptepaddr == 0) 2645 continue; 2646 2647 if (srcptepaddr & PG_PS) { 2648 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2649 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2650 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE; 2651 } 2652 continue; 2653 } 2654 2655 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2656 if ((srcmpte == NULL) || 2657 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2658 continue; 2659 2660 if (pdnxt > end_addr) 2661 pdnxt = end_addr; 2662 2663 src_pte = (unsigned *) vtopte(addr); 2664 dst_pte = (unsigned *) avtopte(addr); 2665 while (addr < pdnxt) { 2666 unsigned ptetemp; 2667 ptetemp = *src_pte; 2668 /* 2669 * we only virtual copy managed pages 2670 */ 2671 if ((ptetemp & PG_MANAGED) != 0) { 2672 /* 2673 * We have to check after allocpte for the 2674 * pte still being around... allocpte can 2675 * block. 2676 */ 2677 dstmpte = pmap_allocpte(dst_pmap, addr); 2678 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2679 /* 2680 * Clear the modified and 2681 * accessed (referenced) bits 2682 * during the copy. 2683 */ 2684 *dst_pte = ptetemp & ~(PG_M | PG_A); 2685 dst_pmap->pm_stats.resident_count++; 2686 pmap_insert_entry(dst_pmap, addr, 2687 dstmpte, 2688 (ptetemp & PG_FRAME)); 2689 } else { 2690 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2691 } 2692 if (dstmpte->hold_count >= srcmpte->hold_count) 2693 break; 2694 } 2695 addr += PAGE_SIZE; 2696 src_pte++; 2697 dst_pte++; 2698 } 2699 } 2700} 2701 2702/* 2703 * Routine: pmap_kernel 2704 * Function: 2705 * Returns the physical map handle for the kernel. 2706 */ 2707pmap_t 2708pmap_kernel() 2709{ 2710 return (kernel_pmap); 2711} 2712 2713/* 2714 * pmap_zero_page zeros the specified hardware page by mapping 2715 * the page into KVM and using bzero to clear its contents. 2716 */ 2717void 2718pmap_zero_page(phys) 2719 vm_offset_t phys; 2720{ 2721#ifdef SMP 2722#if !defined(MAX_PERF) 2723 if (*(int *) prv_CMAP3) 2724 panic("pmap_zero_page: prv_CMAP3 busy"); 2725#endif 2726 2727 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2728 cpu_invlpg(prv_CADDR3); 2729 2730#if defined(I686_CPU) 2731 if (cpu_class == CPUCLASS_686) 2732 i686_pagezero(prv_CADDR3); 2733 else 2734#endif 2735 bzero(prv_CADDR3, PAGE_SIZE); 2736 2737 *(int *) prv_CMAP3 = 0; 2738#else 2739#if !defined(MAX_PERF) 2740 if (*(int *) CMAP2) 2741 panic("pmap_zero_page: CMAP2 busy"); 2742#endif 2743 2744 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2745 invltlb_1pg((vm_offset_t)CADDR2); 2746 2747#if defined(I686_CPU) 2748 if (cpu_class == CPUCLASS_686) 2749 i686_pagezero(CADDR2); 2750 else 2751#endif 2752 bzero(CADDR2, PAGE_SIZE); 2753 *(int *) CMAP2 = 0; 2754#endif 2755} 2756 2757/* 2758 * pmap_zero_page_area zeros the specified hardware page by mapping 2759 * the page into KVM and using bzero to clear its contents. 2760 * 2761 * off and size may not cover an area beyond a single hardware page. 2762 */ 2763void 2764pmap_zero_page_area(phys, off, size) 2765 vm_offset_t phys; 2766 int off; 2767 int size; 2768{ 2769#ifdef SMP 2770#if !defined(MAX_PERF) 2771 if (*(int *) prv_CMAP3) 2772 panic("pmap_zero_page: prv_CMAP3 busy"); 2773#endif 2774 2775 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2776 cpu_invlpg(prv_CADDR3); 2777 2778#if defined(I686_CPU) 2779 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2780 i686_pagezero(prv_CADDR3); 2781 else 2782#endif 2783 bzero((char *)prv_CADDR3 + off, size); 2784 2785 *(int *) prv_CMAP3 = 0; 2786#else 2787#if !defined(MAX_PERF) 2788 if (*(int *) CMAP2) 2789 panic("pmap_zero_page: CMAP2 busy"); 2790#endif 2791 2792 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2793 invltlb_1pg((vm_offset_t)CADDR2); 2794 2795#if defined(I686_CPU) 2796 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2797 i686_pagezero(CADDR2); 2798 else 2799#endif 2800 bzero((char *)CADDR2 + off, size); 2801 *(int *) CMAP2 = 0; 2802#endif 2803} 2804 2805/* 2806 * pmap_copy_page copies the specified (machine independent) 2807 * page by mapping the page into virtual memory and using 2808 * bcopy to copy the page, one machine dependent page at a 2809 * time. 2810 */ 2811void 2812pmap_copy_page(src, dst) 2813 vm_offset_t src; 2814 vm_offset_t dst; 2815{ 2816#ifdef SMP 2817#if !defined(MAX_PERF) 2818 if (*(int *) prv_CMAP1) 2819 panic("pmap_copy_page: prv_CMAP1 busy"); 2820 if (*(int *) prv_CMAP2) 2821 panic("pmap_copy_page: prv_CMAP2 busy"); 2822#endif 2823 2824 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2825 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2826 2827 cpu_invlpg(prv_CADDR1); 2828 cpu_invlpg(prv_CADDR2); 2829 2830 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE); 2831 2832 *(int *) prv_CMAP1 = 0; 2833 *(int *) prv_CMAP2 = 0; 2834#else 2835#if !defined(MAX_PERF) 2836 if (*(int *) CMAP1 || *(int *) CMAP2) 2837 panic("pmap_copy_page: CMAP busy"); 2838#endif 2839 2840 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2841 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2842#if defined(I386_CPU) 2843 if (cpu_class == CPUCLASS_386) { 2844 invltlb(); 2845 } else 2846#endif 2847 { 2848 invlpg((u_int)CADDR1); 2849 invlpg((u_int)CADDR2); 2850 } 2851 2852 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2853 2854 *(int *) CMAP1 = 0; 2855 *(int *) CMAP2 = 0; 2856#endif 2857} 2858 2859 2860/* 2861 * Routine: pmap_pageable 2862 * Function: 2863 * Make the specified pages (by pmap, offset) 2864 * pageable (or not) as requested. 2865 * 2866 * A page which is not pageable may not take 2867 * a fault; therefore, its page table entry 2868 * must remain valid for the duration. 2869 * 2870 * This routine is merely advisory; pmap_enter 2871 * will specify that these pages are to be wired 2872 * down (or not) as appropriate. 2873 */ 2874void 2875pmap_pageable(pmap, sva, eva, pageable) 2876 pmap_t pmap; 2877 vm_offset_t sva, eva; 2878 boolean_t pageable; 2879{ 2880} 2881 2882/* 2883 * this routine returns true if a physical page resides 2884 * in the given pmap. 2885 */ 2886boolean_t 2887pmap_page_exists(pmap, pa) 2888 pmap_t pmap; 2889 vm_offset_t pa; 2890{ 2891 register pv_entry_t pv; 2892 pv_table_t *ppv; 2893 int s; 2894 2895 if (!pmap_is_managed(pa)) 2896 return FALSE; 2897 2898 s = splvm(); 2899 2900 ppv = pa_to_pvh(pa); 2901 /* 2902 * Not found, check current mappings returning immediately if found. 2903 */ 2904 for (pv = TAILQ_FIRST(&ppv->pv_list); 2905 pv; 2906 pv = TAILQ_NEXT(pv, pv_list)) { 2907 if (pv->pv_pmap == pmap) { 2908 splx(s); 2909 return TRUE; 2910 } 2911 } 2912 splx(s); 2913 return (FALSE); 2914} 2915 2916#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2917/* 2918 * Remove all pages from specified address space 2919 * this aids process exit speeds. Also, this code 2920 * is special cased for current process only, but 2921 * can have the more generic (and slightly slower) 2922 * mode enabled. This is much faster than pmap_remove 2923 * in the case of running down an entire address space. 2924 */ 2925void 2926pmap_remove_pages(pmap, sva, eva) 2927 pmap_t pmap; 2928 vm_offset_t sva, eva; 2929{ 2930 unsigned *pte, tpte; 2931 pv_table_t *ppv; 2932 pv_entry_t pv, npv; 2933 int s; 2934 2935#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2936 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) { 2937 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2938 return; 2939 } 2940#endif 2941 2942 s = splvm(); 2943 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2944 pv; 2945 pv = npv) { 2946 2947 if (pv->pv_va >= eva || pv->pv_va < sva) { 2948 npv = TAILQ_NEXT(pv, pv_plist); 2949 continue; 2950 } 2951 2952#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2953 pte = (unsigned *)vtopte(pv->pv_va); 2954#else 2955 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2956#endif 2957 tpte = *pte; 2958 2959/* 2960 * We cannot remove wired pages from a process' mapping at this time 2961 */ 2962 if (tpte & PG_W) { 2963 npv = TAILQ_NEXT(pv, pv_plist); 2964 continue; 2965 } 2966 *pte = 0; 2967 2968 ppv = pa_to_pvh(tpte); 2969 2970 pv->pv_pmap->pm_stats.resident_count--; 2971 2972 /* 2973 * Update the vm_page_t clean and reference bits. 2974 */ 2975 if (tpte & PG_M) { 2976 vm_page_dirty(ppv->pv_vm_page); 2977 } 2978 2979 2980 npv = TAILQ_NEXT(pv, pv_plist); 2981 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2982 2983 ppv->pv_list_count--; 2984 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2985 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2986 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 2987 } 2988 2989 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2990 free_pv_entry(pv); 2991 } 2992 splx(s); 2993 pmap_TLB_invalidate_all(pmap); 2994} 2995 2996/* 2997 * pmap_testbit tests bits in pte's 2998 * note that the testbit/changebit routines are inline, 2999 * and a lot of things compile-time evaluate. 3000 */ 3001static boolean_t 3002pmap_testbit(pa, bit) 3003 register vm_offset_t pa; 3004 int bit; 3005{ 3006 register pv_entry_t pv; 3007 pv_table_t *ppv; 3008 unsigned *pte; 3009 int s; 3010 3011 if (!pmap_is_managed(pa)) 3012 return FALSE; 3013 3014 ppv = pa_to_pvh(pa); 3015 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 3016 return FALSE; 3017 3018 s = splvm(); 3019 3020 for (pv = TAILQ_FIRST(&ppv->pv_list); 3021 pv; 3022 pv = TAILQ_NEXT(pv, pv_list)) { 3023 3024 /* 3025 * if the bit being tested is the modified bit, then 3026 * mark clean_map and ptes as never 3027 * modified. 3028 */ 3029 if (bit & (PG_A|PG_M)) { 3030 if (!pmap_track_modified(pv->pv_va)) 3031 continue; 3032 } 3033 3034#if defined(PMAP_DIAGNOSTIC) 3035 if (!pv->pv_pmap) { 3036 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 3037 continue; 3038 } 3039#endif 3040 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3041 if (*pte & bit) { 3042 splx(s); 3043 return TRUE; 3044 } 3045 } 3046 splx(s); 3047 return (FALSE); 3048} 3049 3050/* 3051 * this routine is used to modify bits in ptes 3052 */ 3053static __inline void 3054pmap_changebit(pa, bit, setem) 3055 vm_offset_t pa; 3056 int bit; 3057 boolean_t setem; 3058{ 3059 register pv_entry_t pv; 3060 pv_table_t *ppv; 3061 register unsigned *pte; 3062 int s; 3063 3064 if (!pmap_is_managed(pa)) 3065 return; 3066 3067 s = splvm(); 3068 ppv = pa_to_pvh(pa); 3069 3070 /* 3071 * Loop over all current mappings setting/clearing as appropos If 3072 * setting RO do we need to clear the VAC? 3073 */ 3074 for (pv = TAILQ_FIRST(&ppv->pv_list); 3075 pv; 3076 pv = TAILQ_NEXT(pv, pv_list)) { 3077 3078 /* 3079 * don't write protect pager mappings 3080 */ 3081 if (!setem && (bit == PG_RW)) { 3082 if (!pmap_track_modified(pv->pv_va)) 3083 continue; 3084 } 3085 3086#if defined(PMAP_DIAGNOSTIC) 3087 if (!pv->pv_pmap) { 3088 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 3089 continue; 3090 } 3091#endif 3092 3093 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3094 3095 if (setem) { 3096 *(int *)pte |= bit; 3097 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3098 } else { 3099 vm_offset_t pbits = *(vm_offset_t *)pte; 3100 if (pbits & bit) { 3101 if (bit == PG_RW) { 3102 if (pbits & PG_M) { 3103 vm_page_dirty(ppv->pv_vm_page); 3104 } 3105 *(int *)pte = pbits & ~(PG_M|PG_RW); 3106 } else { 3107 *(int *)pte = pbits & ~bit; 3108 } 3109 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3110 } 3111 } 3112 } 3113 splx(s); 3114} 3115 3116/* 3117 * pmap_clearbit: 3118 * 3119 * Clear a bit/bits in every pte mapping a given physical page. Making 3120 * this inline allows the pmap_changebit inline to be well optimized. 3121 */ 3122static __inline void 3123pmap_clearbit( 3124 vm_offset_t pa, 3125 int bit) 3126{ 3127 pmap_changebit(pa, bit, FALSE); 3128} 3129 3130/* 3131 * pmap_page_protect: 3132 * 3133 * Lower the permission for all mappings to a given page. 3134 */ 3135void 3136pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 3137{ 3138 if ((prot & VM_PROT_WRITE) == 0) { 3139 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3140 pmap_clearbit(phys, PG_RW); 3141 } else { 3142 pmap_remove_all(phys); 3143 } 3144 } 3145} 3146 3147vm_offset_t 3148pmap_phys_address(ppn) 3149 int ppn; 3150{ 3151 return (i386_ptob(ppn)); 3152} 3153 3154/* 3155 * pmap_ts_referenced: 3156 * 3157 * Return the count of reference bits for a page, clearing all of them. 3158 */ 3159int 3160pmap_ts_referenced(vm_offset_t pa) 3161{ 3162 register pv_entry_t pv, pvf, pvn; 3163 pv_table_t *ppv; 3164 unsigned *pte; 3165 int s; 3166 int rtval = 0; 3167 3168 if (!pmap_is_managed(pa)) 3169 return (rtval); 3170 3171 s = splvm(); 3172 3173 ppv = pa_to_pvh(pa); 3174 3175 if ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 3176 3177 pvf = pv; 3178 3179 do { 3180 pvn = TAILQ_NEXT(pv, pv_list); 3181 3182 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 3183 3184 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3185 3186 if (!pmap_track_modified(pv->pv_va)) 3187 continue; 3188 3189 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3190 3191 if (pte && (*pte & PG_A)) { 3192 *pte &= ~PG_A; 3193 3194 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3195 3196 rtval++; 3197 if (rtval > 4) { 3198 break; 3199 } 3200 } 3201 } while ((pv = pvn) != NULL && pv != pvf); 3202 } 3203 splx(s); 3204 3205 return (rtval); 3206} 3207 3208/* 3209 * pmap_is_modified: 3210 * 3211 * Return whether or not the specified physical page was modified 3212 * in any physical maps. 3213 */ 3214boolean_t 3215pmap_is_modified(vm_offset_t pa) 3216{ 3217 return pmap_testbit((pa), PG_M); 3218} 3219 3220/* 3221 * Clear the modify bits on the specified physical page. 3222 */ 3223void 3224pmap_clear_modify(vm_offset_t pa) 3225{ 3226 pmap_clearbit(pa, PG_M); 3227} 3228 3229/* 3230 * pmap_clear_reference: 3231 * 3232 * Clear the reference bit on the specified physical page. 3233 */ 3234void 3235pmap_clear_reference(vm_offset_t pa) 3236{ 3237 pmap_clearbit(pa, PG_A); 3238} 3239 3240/* 3241 * Miscellaneous support routines follow 3242 */ 3243 3244static void 3245i386_protection_init() 3246{ 3247 register int *kp, prot; 3248 3249 kp = protection_codes; 3250 for (prot = 0; prot < 8; prot++) { 3251 switch (prot) { 3252 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3253 /* 3254 * Read access is also 0. There isn't any execute bit, 3255 * so just make it readable. 3256 */ 3257 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3258 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3259 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3260 *kp++ = 0; 3261 break; 3262 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3263 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3264 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3265 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3266 *kp++ = PG_RW; 3267 break; 3268 } 3269 } 3270} 3271 3272/* 3273 * Map a set of physical memory pages into the kernel virtual 3274 * address space. Return a pointer to where it is mapped. This 3275 * routine is intended to be used for mapping device memory, 3276 * NOT real memory. 3277 */ 3278void * 3279pmap_mapdev(pa, size) 3280 vm_offset_t pa; 3281 vm_size_t size; 3282{ 3283 vm_offset_t va, tmpva; 3284 unsigned *pte; 3285 3286 size = roundup(size, PAGE_SIZE); 3287 3288 va = kmem_alloc_pageable(kernel_map, size); 3289#if !defined(MAX_PERF) 3290 if (!va) 3291 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3292#endif 3293 3294 pa = pa & PG_FRAME; 3295 for (tmpva = va; size > 0;) { 3296 pte = (unsigned *)vtopte(tmpva); 3297 *pte = pa | PG_RW | PG_V | pgeflag; 3298 size -= PAGE_SIZE; 3299 tmpva += PAGE_SIZE; 3300 pa += PAGE_SIZE; 3301 } 3302 invltlb(); 3303 3304 return ((void *) va); 3305} 3306 3307/* 3308 * perform the pmap work for mincore 3309 */ 3310int 3311pmap_mincore(pmap, addr) 3312 pmap_t pmap; 3313 vm_offset_t addr; 3314{ 3315 3316 unsigned *ptep, pte; 3317 vm_page_t m; 3318 int val = 0; 3319 3320 ptep = pmap_pte(pmap, addr); 3321 if (ptep == 0) { 3322 return 0; 3323 } 3324 3325 if ((pte = *ptep) != 0) { 3326 pv_table_t *ppv; 3327 vm_offset_t pa; 3328 3329 val = MINCORE_INCORE; 3330 if ((pte & PG_MANAGED) == 0) 3331 return val; 3332 3333 pa = pte & PG_FRAME; 3334 3335 ppv = pa_to_pvh((pa & PG_FRAME)); 3336 m = ppv->pv_vm_page; 3337 3338 /* 3339 * Modified by us 3340 */ 3341 if (pte & PG_M) 3342 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3343 /* 3344 * Modified by someone 3345 */ 3346 else if (m->dirty || pmap_is_modified(pa)) 3347 val |= MINCORE_MODIFIED_OTHER; 3348 /* 3349 * Referenced by us 3350 */ 3351 if (pte & PG_A) 3352 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3353 3354 /* 3355 * Referenced by someone 3356 */ 3357 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(pa)) { 3358 val |= MINCORE_REFERENCED_OTHER; 3359 vm_page_flag_set(m, PG_REFERENCED); 3360 } 3361 } 3362 return val; 3363} 3364 3365void 3366pmap_activate(struct proc *p) 3367{ 3368 pmap_t pmap; 3369 3370 pmap = vmspace_pmap(p->p_vmspace); 3371#if defined(SMP) 3372 pmap->pm_active |= 1 << cpuid; 3373#else 3374 pmap->pm_active |= 1; 3375#endif 3376#if defined(SWTCH_OPTIM_STATS) 3377 tlb_flush_count++; 3378#endif 3379 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir)); 3380} 3381 3382vm_offset_t 3383pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3384 3385 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3386 return addr; 3387 } 3388 3389 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3390 return addr; 3391} 3392 3393 3394#if defined(PMAP_DEBUG) 3395pmap_pid_dump(int pid) { 3396 pmap_t pmap; 3397 struct proc *p; 3398 int npte = 0; 3399 int index; 3400 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3401 if (p->p_pid != pid) 3402 continue; 3403 3404 if (p->p_vmspace) { 3405 int i,j; 3406 index = 0; 3407 pmap = vmspace_pmap(p->p_vmspace); 3408 for(i=0;i<1024;i++) { 3409 pd_entry_t *pde; 3410 unsigned *pte; 3411 unsigned base = i << PDRSHIFT; 3412 3413 pde = &pmap->pm_pdir[i]; 3414 if (pde && pmap_pde_v(pde)) { 3415 for(j=0;j<1024;j++) { 3416 unsigned va = base + (j << PAGE_SHIFT); 3417 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3418 if (index) { 3419 index = 0; 3420 printf("\n"); 3421 } 3422 return npte; 3423 } 3424 pte = pmap_pte_quick( pmap, va); 3425 if (pte && pmap_pte_v(pte)) { 3426 vm_offset_t pa; 3427 vm_page_t m; 3428 pa = *(int *)pte; 3429 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3430 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3431 va, pa, m->hold_count, m->wire_count, m->flags); 3432 npte++; 3433 index++; 3434 if (index >= 2) { 3435 index = 0; 3436 printf("\n"); 3437 } else { 3438 printf(" "); 3439 } 3440 } 3441 } 3442 } 3443 } 3444 } 3445 } 3446 return npte; 3447} 3448#endif 3449 3450#if defined(DEBUG) 3451 3452static void pads __P((pmap_t pm)); 3453void pmap_pvdump __P((vm_offset_t pa)); 3454 3455/* print address space of pmap*/ 3456static void 3457pads(pm) 3458 pmap_t pm; 3459{ 3460 unsigned va, i, j; 3461 unsigned *ptep; 3462 3463 if (pm == kernel_pmap) 3464 return; 3465 for (i = 0; i < 1024; i++) 3466 if (pm->pm_pdir[i]) 3467 for (j = 0; j < 1024; j++) { 3468 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3469 if (pm == kernel_pmap && va < KERNBASE) 3470 continue; 3471 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3472 continue; 3473 ptep = pmap_pte_quick(pm, va); 3474 if (pmap_pte_v(ptep)) 3475 printf("%x:%x ", va, *(int *) ptep); 3476 }; 3477 3478} 3479 3480void 3481pmap_pvdump(pa) 3482 vm_offset_t pa; 3483{ 3484 pv_table_t *ppv; 3485 register pv_entry_t pv; 3486 3487 printf("pa %x", pa); 3488 ppv = pa_to_pvh(pa); 3489 for (pv = TAILQ_FIRST(&ppv->pv_list); 3490 pv; 3491 pv = TAILQ_NEXT(pv, pv_list)) { 3492#ifdef used_to_be 3493 printf(" -> pmap %p, va %x, flags %x", 3494 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags); 3495#endif 3496 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3497 pads(pv->pv_pmap); 3498 } 3499 printf(" "); 3500} 3501#endif 3502