pmap.c revision 48677
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 1.240 1999/06/23 21:47:21 luoqi 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 /* wire in kernel global address entries */ 1176 /* XXX copies current process, does not fill in MPPTDI */ 1177 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1178#ifdef SMP 1179 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI]; 1180#endif 1181 1182 /* install self-referential address mapping entry */ 1183 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1184 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1185 1186 pmap->pm_count = 1; 1187 pmap->pm_active = 0; 1188 pmap->pm_ptphint = NULL; 1189 TAILQ_INIT(&pmap->pm_pvlist); 1190 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1191} 1192 1193static int 1194pmap_release_free_page(pmap, p) 1195 struct pmap *pmap; 1196 vm_page_t p; 1197{ 1198 unsigned *pde = (unsigned *) pmap->pm_pdir; 1199 /* 1200 * This code optimizes the case of freeing non-busy 1201 * page-table pages. Those pages are zero now, and 1202 * might as well be placed directly into the zero queue. 1203 */ 1204 if (vm_page_sleep_busy(p, FALSE, "pmaprl")) 1205 return 0; 1206 1207 vm_page_busy(p); 1208 1209 /* 1210 * Remove the page table page from the processes address space. 1211 */ 1212 pde[p->pindex] = 0; 1213 pmap->pm_stats.resident_count--; 1214 1215#if !defined(MAX_PERF) 1216 if (p->hold_count) { 1217 panic("pmap_release: freeing held page table page"); 1218 } 1219#endif 1220 /* 1221 * Page directory pages need to have the kernel 1222 * stuff cleared, so they can go into the zero queue also. 1223 */ 1224 if (p->pindex == PTDPTDI) { 1225 bzero(pde + KPTDI, nkpt * PTESIZE); 1226#ifdef SMP 1227 pde[MPPTDI] = 0; 1228#endif 1229 pde[APTDPTDI] = 0; 1230 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1231 } 1232 1233 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1234 pmap->pm_ptphint = NULL; 1235 1236 p->wire_count--; 1237 cnt.v_wire_count--; 1238 vm_page_free_zero(p); 1239 return 1; 1240} 1241 1242/* 1243 * this routine is called if the page table page is not 1244 * mapped correctly. 1245 */ 1246static vm_page_t 1247_pmap_allocpte(pmap, ptepindex) 1248 pmap_t pmap; 1249 unsigned ptepindex; 1250{ 1251 vm_offset_t pteva, ptepa; 1252 vm_page_t m; 1253 1254 /* 1255 * Find or fabricate a new pagetable page 1256 */ 1257 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1258 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1259 1260 if (m->queue != PQ_NONE) { 1261 int s = splvm(); 1262 vm_page_unqueue(m); 1263 splx(s); 1264 } 1265 1266 if (m->wire_count == 0) 1267 cnt.v_wire_count++; 1268 m->wire_count++; 1269 1270 /* 1271 * Increment the hold count for the page table page 1272 * (denoting a new mapping.) 1273 */ 1274 m->hold_count++; 1275 1276 /* 1277 * Map the pagetable page into the process address space, if 1278 * it isn't already there. 1279 */ 1280 1281 pmap->pm_stats.resident_count++; 1282 1283 ptepa = VM_PAGE_TO_PHYS(m); 1284 pmap->pm_pdir[ptepindex] = 1285 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1286 1287 /* 1288 * Set the page table hint 1289 */ 1290 pmap->pm_ptphint = m; 1291 1292 /* 1293 * Try to use the new mapping, but if we cannot, then 1294 * do it with the routine that maps the page explicitly. 1295 */ 1296 if ((m->flags & PG_ZERO) == 0) { 1297 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1298 (((unsigned) PTDpde) & PG_FRAME)) { 1299 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1300 bzero((caddr_t) pteva, PAGE_SIZE); 1301 } else { 1302 pmap_zero_page(ptepa); 1303 } 1304 } 1305 1306 m->valid = VM_PAGE_BITS_ALL; 1307 vm_page_flag_clear(m, PG_ZERO); 1308 vm_page_flag_set(m, PG_MAPPED); 1309 vm_page_wakeup(m); 1310 1311 return m; 1312} 1313 1314static vm_page_t 1315pmap_allocpte(pmap, va) 1316 pmap_t pmap; 1317 vm_offset_t va; 1318{ 1319 unsigned ptepindex; 1320 vm_offset_t ptepa; 1321 vm_page_t m; 1322 1323 /* 1324 * Calculate pagetable page index 1325 */ 1326 ptepindex = va >> PDRSHIFT; 1327 1328 /* 1329 * Get the page directory entry 1330 */ 1331 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1332 1333 /* 1334 * This supports switching from a 4MB page to a 1335 * normal 4K page. 1336 */ 1337 if (ptepa & PG_PS) { 1338 pmap->pm_pdir[ptepindex] = 0; 1339 ptepa = 0; 1340 invltlb(); 1341 } 1342 1343 /* 1344 * If the page table page is mapped, we just increment the 1345 * hold count, and activate it. 1346 */ 1347 if (ptepa) { 1348 /* 1349 * In order to get the page table page, try the 1350 * hint first. 1351 */ 1352 if (pmap->pm_ptphint && 1353 (pmap->pm_ptphint->pindex == ptepindex)) { 1354 m = pmap->pm_ptphint; 1355 } else { 1356 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1357 pmap->pm_ptphint = m; 1358 } 1359 m->hold_count++; 1360 return m; 1361 } 1362 /* 1363 * Here if the pte page isn't mapped, or if it has been deallocated. 1364 */ 1365 return _pmap_allocpte(pmap, ptepindex); 1366} 1367 1368 1369/*************************************************** 1370* Pmap allocation/deallocation routines. 1371 ***************************************************/ 1372 1373/* 1374 * Release any resources held by the given physical map. 1375 * Called when a pmap initialized by pmap_pinit is being released. 1376 * Should only be called if the map contains no valid mappings. 1377 */ 1378void 1379pmap_release(pmap) 1380 register struct pmap *pmap; 1381{ 1382 vm_page_t p,n,ptdpg; 1383 vm_object_t object = pmap->pm_pteobj; 1384 int curgeneration; 1385 1386#if defined(DIAGNOSTIC) 1387 if (object->ref_count != 1) 1388 panic("pmap_release: pteobj reference count != 1"); 1389#endif 1390 1391 ptdpg = NULL; 1392retry: 1393 curgeneration = object->generation; 1394 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1395 n = TAILQ_NEXT(p, listq); 1396 if (p->pindex == PTDPTDI) { 1397 ptdpg = p; 1398 continue; 1399 } 1400 while (1) { 1401 if (!pmap_release_free_page(pmap, p) && 1402 (object->generation != curgeneration)) 1403 goto retry; 1404 } 1405 } 1406 1407 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1408 goto retry; 1409} 1410 1411/* 1412 * grow the number of kernel page table entries, if needed 1413 */ 1414void 1415pmap_growkernel(vm_offset_t addr) 1416{ 1417 struct proc *p; 1418 struct pmap *pmap; 1419 int s; 1420 vm_offset_t ptppaddr; 1421 vm_page_t nkpg; 1422 pd_entry_t newpdir; 1423 1424 s = splhigh(); 1425 if (kernel_vm_end == 0) { 1426 kernel_vm_end = KERNBASE; 1427 nkpt = 0; 1428 while (pdir_pde(PTD, kernel_vm_end)) { 1429 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1430 nkpt++; 1431 } 1432 } 1433 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1434 while (kernel_vm_end < addr) { 1435 if (pdir_pde(PTD, kernel_vm_end)) { 1436 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1437 continue; 1438 } 1439 1440 /* 1441 * This index is bogus, but out of the way 1442 */ 1443 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1444#if !defined(MAX_PERF) 1445 if (!nkpg) 1446 panic("pmap_growkernel: no memory to grow kernel"); 1447#endif 1448 1449 nkpt++; 1450 1451 vm_page_wire(nkpg); 1452 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1453 pmap_zero_page(ptppaddr); 1454 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1455 pdir_pde(PTD, kernel_vm_end) = newpdir; 1456 1457 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1458 if (p->p_vmspace) { 1459 pmap = vmspace_pmap(p->p_vmspace); 1460 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1461 } 1462 } 1463 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir; 1464 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1465 } 1466 splx(s); 1467} 1468 1469/* 1470 * Retire the given physical map from service. 1471 * Should only be called if the map contains 1472 * no valid mappings. 1473 */ 1474void 1475pmap_destroy(pmap) 1476 register pmap_t pmap; 1477{ 1478 int count; 1479 1480 if (pmap == NULL) 1481 return; 1482 1483 count = --pmap->pm_count; 1484 if (count == 0) { 1485 pmap_release(pmap); 1486#if !defined(MAX_PERF) 1487 panic("destroying a pmap is not yet implemented"); 1488#endif 1489 } 1490} 1491 1492/* 1493 * Add a reference to the specified pmap. 1494 */ 1495void 1496pmap_reference(pmap) 1497 pmap_t pmap; 1498{ 1499 if (pmap != NULL) { 1500 pmap->pm_count++; 1501 } 1502} 1503 1504/*************************************************** 1505* page management routines. 1506 ***************************************************/ 1507 1508/* 1509 * free the pv_entry back to the free list 1510 */ 1511static PMAP_INLINE void 1512free_pv_entry(pv) 1513 pv_entry_t pv; 1514{ 1515 pv_entry_count--; 1516 zfreei(pvzone, pv); 1517} 1518 1519/* 1520 * get a new pv_entry, allocating a block from the system 1521 * when needed. 1522 * the memory allocation is performed bypassing the malloc code 1523 * because of the possibility of allocations at interrupt time. 1524 */ 1525static pv_entry_t 1526get_pv_entry(void) 1527{ 1528 pv_entry_count++; 1529 if (pv_entry_high_water && 1530 (pv_entry_count > pv_entry_high_water) && 1531 (pmap_pagedaemon_waken == 0)) { 1532 pmap_pagedaemon_waken = 1; 1533 wakeup (&vm_pages_needed); 1534 } 1535 return zalloci(pvzone); 1536} 1537 1538/* 1539 * This routine is very drastic, but can save the system 1540 * in a pinch. 1541 */ 1542void 1543pmap_collect() { 1544 pv_table_t *ppv; 1545 int i; 1546 vm_offset_t pa; 1547 vm_page_t m; 1548 static int warningdone=0; 1549 1550 if (pmap_pagedaemon_waken == 0) 1551 return; 1552 1553 if (warningdone < 5) { 1554 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1555 warningdone++; 1556 } 1557 1558 for(i = 0; i < pv_npg; i++) { 1559 if ((ppv = &pv_table[i]) == 0) 1560 continue; 1561 m = ppv->pv_vm_page; 1562 if ((pa = VM_PAGE_TO_PHYS(m)) == 0) 1563 continue; 1564 if (m->wire_count || m->hold_count || m->busy || 1565 (m->flags & PG_BUSY)) 1566 continue; 1567 pmap_remove_all(pa); 1568 } 1569 pmap_pagedaemon_waken = 0; 1570} 1571 1572 1573/* 1574 * If it is the first entry on the list, it is actually 1575 * in the header and we must copy the following entry up 1576 * to the header. Otherwise we must search the list for 1577 * the entry. In either case we free the now unused entry. 1578 */ 1579 1580static int 1581pmap_remove_entry(pmap, ppv, va) 1582 struct pmap *pmap; 1583 pv_table_t *ppv; 1584 vm_offset_t va; 1585{ 1586 pv_entry_t pv; 1587 int rtval; 1588 int s; 1589 1590 s = splvm(); 1591 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1592 for (pv = TAILQ_FIRST(&ppv->pv_list); 1593 pv; 1594 pv = TAILQ_NEXT(pv, pv_list)) { 1595 if (pmap == pv->pv_pmap && va == pv->pv_va) 1596 break; 1597 } 1598 } else { 1599 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1600 pv; 1601 pv = TAILQ_NEXT(pv, pv_plist)) { 1602 if (va == pv->pv_va) 1603 break; 1604 } 1605 } 1606 1607 rtval = 0; 1608 if (pv) { 1609 1610 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1611 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1612 ppv->pv_list_count--; 1613 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 1614 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 1615 1616 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1617 free_pv_entry(pv); 1618 } 1619 1620 splx(s); 1621 return rtval; 1622} 1623 1624/* 1625 * Create a pv entry for page at pa for 1626 * (pmap, va). 1627 */ 1628static void 1629pmap_insert_entry(pmap, va, mpte, pa) 1630 pmap_t pmap; 1631 vm_offset_t va; 1632 vm_page_t mpte; 1633 vm_offset_t pa; 1634{ 1635 1636 int s; 1637 pv_entry_t pv; 1638 pv_table_t *ppv; 1639 1640 s = splvm(); 1641 pv = get_pv_entry(); 1642 pv->pv_va = va; 1643 pv->pv_pmap = pmap; 1644 pv->pv_ptem = mpte; 1645 1646 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1647 1648 ppv = pa_to_pvh(pa); 1649 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1650 ppv->pv_list_count++; 1651 1652 splx(s); 1653} 1654 1655/* 1656 * pmap_remove_pte: do the things to unmap a page in a process 1657 */ 1658static int 1659pmap_remove_pte(pmap, ptq, va) 1660 struct pmap *pmap; 1661 unsigned *ptq; 1662 vm_offset_t va; 1663{ 1664 unsigned oldpte; 1665 pv_table_t *ppv; 1666 1667 oldpte = loadandclear(ptq); 1668 if (oldpte & PG_W) 1669 pmap->pm_stats.wired_count -= 1; 1670 /* 1671 * Machines that don't support invlpg, also don't support 1672 * PG_G. 1673 */ 1674 if (oldpte & PG_G) 1675 invlpg(va); 1676 pmap->pm_stats.resident_count -= 1; 1677 if (oldpte & PG_MANAGED) { 1678 ppv = pa_to_pvh(oldpte); 1679 if (oldpte & PG_M) { 1680#if defined(PMAP_DIAGNOSTIC) 1681 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1682 printf( 1683 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n", 1684 va, oldpte); 1685 } 1686#endif 1687 if (pmap_track_modified(va)) 1688 vm_page_dirty(ppv->pv_vm_page); 1689 } 1690 if (oldpte & PG_A) 1691 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1692 return pmap_remove_entry(pmap, ppv, va); 1693 } else { 1694 return pmap_unuse_pt(pmap, va, NULL); 1695 } 1696 1697 return 0; 1698} 1699 1700/* 1701 * Remove a single page from a process address space 1702 */ 1703static void 1704pmap_remove_page(pmap, va) 1705 struct pmap *pmap; 1706 register vm_offset_t va; 1707{ 1708 register unsigned *ptq; 1709 1710 /* 1711 * if there is no pte for this address, just skip it!!! 1712 */ 1713 if (*pmap_pde(pmap, va) == 0) { 1714 return; 1715 } 1716 1717 /* 1718 * get a local va for mappings for this pmap. 1719 */ 1720 ptq = get_ptbase(pmap) + i386_btop(va); 1721 if (*ptq) { 1722 (void) pmap_remove_pte(pmap, ptq, va); 1723 pmap_TLB_invalidate(pmap, va); 1724 } 1725 return; 1726} 1727 1728/* 1729 * Remove the given range of addresses from the specified map. 1730 * 1731 * It is assumed that the start and end are properly 1732 * rounded to the page size. 1733 */ 1734void 1735pmap_remove(pmap, sva, eva) 1736 struct pmap *pmap; 1737 register vm_offset_t sva; 1738 register vm_offset_t eva; 1739{ 1740 register unsigned *ptbase; 1741 vm_offset_t pdnxt; 1742 vm_offset_t ptpaddr; 1743 vm_offset_t sindex, eindex; 1744 int anyvalid; 1745 1746 if (pmap == NULL) 1747 return; 1748 1749 if (pmap->pm_stats.resident_count == 0) 1750 return; 1751 1752 /* 1753 * special handling of removing one page. a very 1754 * common operation and easy to short circuit some 1755 * code. 1756 */ 1757 if (((sva + PAGE_SIZE) == eva) && 1758 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1759 pmap_remove_page(pmap, sva); 1760 return; 1761 } 1762 1763 anyvalid = 0; 1764 1765 /* 1766 * Get a local virtual address for the mappings that are being 1767 * worked with. 1768 */ 1769 ptbase = get_ptbase(pmap); 1770 1771 sindex = i386_btop(sva); 1772 eindex = i386_btop(eva); 1773 1774 for (; sindex < eindex; sindex = pdnxt) { 1775 unsigned pdirindex; 1776 1777 /* 1778 * Calculate index for next page table. 1779 */ 1780 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1781 if (pmap->pm_stats.resident_count == 0) 1782 break; 1783 1784 pdirindex = sindex / NPDEPG; 1785 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1786 pmap->pm_pdir[pdirindex] = 0; 1787 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1788 anyvalid++; 1789 continue; 1790 } 1791 1792 /* 1793 * Weed out invalid mappings. Note: we assume that the page 1794 * directory table is always allocated, and in kernel virtual. 1795 */ 1796 if (ptpaddr == 0) 1797 continue; 1798 1799 /* 1800 * Limit our scan to either the end of the va represented 1801 * by the current page table page, or to the end of the 1802 * range being removed. 1803 */ 1804 if (pdnxt > eindex) { 1805 pdnxt = eindex; 1806 } 1807 1808 for ( ;sindex != pdnxt; sindex++) { 1809 vm_offset_t va; 1810 if (ptbase[sindex] == 0) { 1811 continue; 1812 } 1813 va = i386_ptob(sindex); 1814 1815 anyvalid++; 1816 if (pmap_remove_pte(pmap, 1817 ptbase + sindex, va)) 1818 break; 1819 } 1820 } 1821 1822 if (anyvalid) 1823 pmap_TLB_invalidate_all(pmap); 1824} 1825 1826/* 1827 * Routine: pmap_remove_all 1828 * Function: 1829 * Removes this physical page from 1830 * all physical maps in which it resides. 1831 * Reflects back modify bits to the pager. 1832 * 1833 * Notes: 1834 * Original versions of this routine were very 1835 * inefficient because they iteratively called 1836 * pmap_remove (slow...) 1837 */ 1838 1839static void 1840pmap_remove_all(pa) 1841 vm_offset_t pa; 1842{ 1843 register pv_entry_t pv; 1844 pv_table_t *ppv; 1845 register unsigned *pte, tpte; 1846 int s; 1847 1848#if defined(PMAP_DIAGNOSTIC) 1849 /* 1850 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1851 * pages! 1852 */ 1853 if (!pmap_is_managed(pa)) { 1854 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", pa); 1855 } 1856#endif 1857 1858 s = splvm(); 1859 ppv = pa_to_pvh(pa); 1860 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1861 pv->pv_pmap->pm_stats.resident_count--; 1862 1863 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1864 1865 tpte = loadandclear(pte); 1866 if (tpte & PG_W) 1867 pv->pv_pmap->pm_stats.wired_count--; 1868 1869 if (tpte & PG_A) 1870 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1871 1872 /* 1873 * Update the vm_page_t clean and reference bits. 1874 */ 1875 if (tpte & PG_M) { 1876#if defined(PMAP_DIAGNOSTIC) 1877 if (pmap_nw_modified((pt_entry_t) tpte)) { 1878 printf( 1879 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n", 1880 pv->pv_va, tpte); 1881 } 1882#endif 1883 if (pmap_track_modified(pv->pv_va)) 1884 vm_page_dirty(ppv->pv_vm_page); 1885 } 1886 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 1887 1888 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1889 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1890 ppv->pv_list_count--; 1891 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1892 free_pv_entry(pv); 1893 } 1894 1895 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 1896 1897 splx(s); 1898} 1899 1900/* 1901 * Set the physical protection on the 1902 * specified range of this map as requested. 1903 */ 1904void 1905pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1906{ 1907 register unsigned *ptbase; 1908 vm_offset_t pdnxt, ptpaddr; 1909 vm_pindex_t sindex, eindex; 1910 int anychanged; 1911 1912 1913 if (pmap == NULL) 1914 return; 1915 1916 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1917 pmap_remove(pmap, sva, eva); 1918 return; 1919 } 1920 1921 if (prot & VM_PROT_WRITE) 1922 return; 1923 1924 anychanged = 0; 1925 1926 ptbase = get_ptbase(pmap); 1927 1928 sindex = i386_btop(sva); 1929 eindex = i386_btop(eva); 1930 1931 for (; sindex < eindex; sindex = pdnxt) { 1932 1933 unsigned pdirindex; 1934 1935 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1936 1937 pdirindex = sindex / NPDEPG; 1938 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1939 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1940 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1941 anychanged++; 1942 continue; 1943 } 1944 1945 /* 1946 * Weed out invalid mappings. Note: we assume that the page 1947 * directory table is always allocated, and in kernel virtual. 1948 */ 1949 if (ptpaddr == 0) 1950 continue; 1951 1952 if (pdnxt > eindex) { 1953 pdnxt = eindex; 1954 } 1955 1956 for (; sindex != pdnxt; sindex++) { 1957 1958 unsigned pbits; 1959 pv_table_t *ppv; 1960 1961 pbits = ptbase[sindex]; 1962 1963 if (pbits & PG_MANAGED) { 1964 ppv = NULL; 1965 if (pbits & PG_A) { 1966 ppv = pa_to_pvh(pbits); 1967 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1968 pbits &= ~PG_A; 1969 } 1970 if (pbits & PG_M) { 1971 if (pmap_track_modified(i386_ptob(sindex))) { 1972 if (ppv == NULL) 1973 ppv = pa_to_pvh(pbits); 1974 vm_page_dirty(ppv->pv_vm_page); 1975 pbits &= ~PG_M; 1976 } 1977 } 1978 } 1979 1980 pbits &= ~PG_RW; 1981 1982 if (pbits != ptbase[sindex]) { 1983 ptbase[sindex] = pbits; 1984 anychanged = 1; 1985 } 1986 } 1987 } 1988 if (anychanged) 1989 pmap_TLB_invalidate_all(pmap); 1990} 1991 1992/* 1993 * Insert the given physical page (p) at 1994 * the specified virtual address (v) in the 1995 * target physical map with the protection requested. 1996 * 1997 * If specified, the page will be wired down, meaning 1998 * that the related pte can not be reclaimed. 1999 * 2000 * NB: This is the only routine which MAY NOT lazy-evaluate 2001 * or lose information. That is, this routine must actually 2002 * insert this page into the given map NOW. 2003 */ 2004void 2005pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot, 2006 boolean_t wired) 2007{ 2008 register unsigned *pte; 2009 vm_offset_t opa; 2010 vm_offset_t origpte, newpte; 2011 vm_page_t mpte; 2012 2013 if (pmap == NULL) 2014 return; 2015 2016 va &= PG_FRAME; 2017#ifdef PMAP_DIAGNOSTIC 2018 if (va > VM_MAX_KERNEL_ADDRESS) 2019 panic("pmap_enter: toobig"); 2020 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 2021 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 2022#endif 2023 2024 mpte = NULL; 2025 /* 2026 * In the case that a page table page is not 2027 * resident, we are creating it here. 2028 */ 2029 if (va < UPT_MIN_ADDRESS) { 2030 mpte = pmap_allocpte(pmap, va); 2031 } 2032#if 0 && defined(PMAP_DIAGNOSTIC) 2033 else { 2034 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 2035 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 2036 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 2037 pmap->pm_pdir[PTDPTDI], origpte, va); 2038 } 2039 if (smp_active) { 2040 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid]; 2041 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2042 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2043 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2044 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr); 2045 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2046 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2047 } 2048 } 2049 } 2050#endif 2051 2052 pte = pmap_pte(pmap, va); 2053 2054#if !defined(MAX_PERF) 2055 /* 2056 * Page Directory table entry not valid, we need a new PT page 2057 */ 2058 if (pte == NULL) { 2059 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n", 2060 (void *)pmap->pm_pdir[PTDPTDI], va); 2061 } 2062#endif 2063 2064 origpte = *(vm_offset_t *)pte; 2065 pa &= PG_FRAME; 2066 opa = origpte & PG_FRAME; 2067 2068#if !defined(MAX_PERF) 2069 if (origpte & PG_PS) 2070 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2071#endif 2072 2073 /* 2074 * Mapping has not changed, must be protection or wiring change. 2075 */ 2076 if (origpte && (opa == pa)) { 2077 /* 2078 * Wiring change, just update stats. We don't worry about 2079 * wiring PT pages as they remain resident as long as there 2080 * are valid mappings in them. Hence, if a user page is wired, 2081 * the PT page will be also. 2082 */ 2083 if (wired && ((origpte & PG_W) == 0)) 2084 pmap->pm_stats.wired_count++; 2085 else if (!wired && (origpte & PG_W)) 2086 pmap->pm_stats.wired_count--; 2087 2088#if defined(PMAP_DIAGNOSTIC) 2089 if (pmap_nw_modified((pt_entry_t) origpte)) { 2090 printf( 2091 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n", 2092 va, origpte); 2093 } 2094#endif 2095 2096 /* 2097 * Remove extra pte reference 2098 */ 2099 if (mpte) 2100 mpte->hold_count--; 2101 2102 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2103 if ((origpte & PG_RW) == 0) { 2104 *pte |= PG_RW; 2105#ifdef SMP 2106 cpu_invlpg((void *)va); 2107 if (pmap->pm_active & other_cpus) 2108 smp_invltlb(); 2109#else 2110 invltlb_1pg(va); 2111#endif 2112 } 2113 return; 2114 } 2115 2116 /* 2117 * We might be turning off write access to the page, 2118 * so we go ahead and sense modify status. 2119 */ 2120 if (origpte & PG_MANAGED) { 2121 if ((origpte & PG_M) && pmap_track_modified(va)) { 2122 pv_table_t *ppv; 2123 ppv = pa_to_pvh(opa); 2124 vm_page_dirty(ppv->pv_vm_page); 2125 } 2126 pa |= PG_MANAGED; 2127 } 2128 goto validate; 2129 } 2130 /* 2131 * Mapping has changed, invalidate old range and fall through to 2132 * handle validating new mapping. 2133 */ 2134 if (opa) { 2135 int err; 2136 err = pmap_remove_pte(pmap, pte, va); 2137#if !defined(MAX_PERF) 2138 if (err) 2139 panic("pmap_enter: pte vanished, va: 0x%x", va); 2140#endif 2141 } 2142 2143 /* 2144 * Enter on the PV list if part of our managed memory Note that we 2145 * raise IPL while manipulating pv_table since pmap_enter can be 2146 * called at interrupt time. 2147 */ 2148 if (pmap_is_managed(pa)) { 2149 pmap_insert_entry(pmap, va, mpte, pa); 2150 pa |= PG_MANAGED; 2151 } 2152 2153 /* 2154 * Increment counters 2155 */ 2156 pmap->pm_stats.resident_count++; 2157 if (wired) 2158 pmap->pm_stats.wired_count++; 2159 2160validate: 2161 /* 2162 * Now validate mapping with desired protection/wiring. 2163 */ 2164 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2165 2166 if (wired) 2167 newpte |= PG_W; 2168 if (va < UPT_MIN_ADDRESS) 2169 newpte |= PG_U; 2170 if (pmap == kernel_pmap) 2171 newpte |= pgeflag; 2172 2173 /* 2174 * if the mapping or permission bits are different, we need 2175 * to update the pte. 2176 */ 2177 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2178 *pte = newpte | PG_A; 2179 if (origpte) { 2180#ifdef SMP 2181 cpu_invlpg((void *)va); 2182 if (pmap->pm_active & other_cpus) 2183 smp_invltlb(); 2184#else 2185 invltlb_1pg(va); 2186#endif 2187 } 2188 } 2189} 2190 2191/* 2192 * this code makes some *MAJOR* assumptions: 2193 * 1. Current pmap & pmap exists. 2194 * 2. Not wired. 2195 * 3. Read access. 2196 * 4. No page table pages. 2197 * 5. Tlbflush is deferred to calling procedure. 2198 * 6. Page IS managed. 2199 * but is *MUCH* faster than pmap_enter... 2200 */ 2201 2202static vm_page_t 2203pmap_enter_quick(pmap, va, pa, mpte) 2204 register pmap_t pmap; 2205 vm_offset_t va; 2206 register vm_offset_t pa; 2207 vm_page_t mpte; 2208{ 2209 register unsigned *pte; 2210 2211 /* 2212 * In the case that a page table page is not 2213 * resident, we are creating it here. 2214 */ 2215 if (va < UPT_MIN_ADDRESS) { 2216 unsigned ptepindex; 2217 vm_offset_t ptepa; 2218 2219 /* 2220 * Calculate pagetable page index 2221 */ 2222 ptepindex = va >> PDRSHIFT; 2223 if (mpte && (mpte->pindex == ptepindex)) { 2224 mpte->hold_count++; 2225 } else { 2226retry: 2227 /* 2228 * Get the page directory entry 2229 */ 2230 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2231 2232 /* 2233 * If the page table page is mapped, we just increment 2234 * the hold count, and activate it. 2235 */ 2236 if (ptepa) { 2237#if !defined(MAX_PERF) 2238 if (ptepa & PG_PS) 2239 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2240#endif 2241 if (pmap->pm_ptphint && 2242 (pmap->pm_ptphint->pindex == ptepindex)) { 2243 mpte = pmap->pm_ptphint; 2244 } else { 2245 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2246 pmap->pm_ptphint = mpte; 2247 } 2248 if (mpte == NULL) 2249 goto retry; 2250 mpte->hold_count++; 2251 } else { 2252 mpte = _pmap_allocpte(pmap, ptepindex); 2253 } 2254 } 2255 } else { 2256 mpte = NULL; 2257 } 2258 2259 /* 2260 * This call to vtopte makes the assumption that we are 2261 * entering the page into the current pmap. In order to support 2262 * quick entry into any pmap, one would likely use pmap_pte_quick. 2263 * But that isn't as quick as vtopte. 2264 */ 2265 pte = (unsigned *)vtopte(va); 2266 if (*pte) { 2267 if (mpte) 2268 pmap_unwire_pte_hold(pmap, mpte); 2269 return 0; 2270 } 2271 2272 /* 2273 * Enter on the PV list if part of our managed memory Note that we 2274 * raise IPL while manipulating pv_table since pmap_enter can be 2275 * called at interrupt time. 2276 */ 2277 pmap_insert_entry(pmap, va, mpte, pa); 2278 2279 /* 2280 * Increment counters 2281 */ 2282 pmap->pm_stats.resident_count++; 2283 2284 /* 2285 * Now validate mapping with RO protection 2286 */ 2287 *pte = pa | PG_V | PG_U | PG_MANAGED; 2288 2289 return mpte; 2290} 2291 2292#define MAX_INIT_PT (96) 2293/* 2294 * pmap_object_init_pt preloads the ptes for a given object 2295 * into the specified pmap. This eliminates the blast of soft 2296 * faults on process startup and immediately after an mmap. 2297 */ 2298void 2299pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2300 pmap_t pmap; 2301 vm_offset_t addr; 2302 vm_object_t object; 2303 vm_pindex_t pindex; 2304 vm_size_t size; 2305 int limit; 2306{ 2307 vm_offset_t tmpidx; 2308 int psize; 2309 vm_page_t p, mpte; 2310 int objpgs; 2311 2312 if (!pmap) 2313 return; 2314 2315 /* 2316 * This code maps large physical mmap regions into the 2317 * processor address space. Note that some shortcuts 2318 * are taken, but the code works. 2319 */ 2320 if (pseflag && 2321 (object->type == OBJT_DEVICE) && 2322 ((addr & (NBPDR - 1)) == 0) && 2323 ((size & (NBPDR - 1)) == 0) ) { 2324 int i; 2325 vm_page_t m[1]; 2326 unsigned int ptepindex; 2327 int npdes; 2328 vm_offset_t ptepa; 2329 2330 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2331 return; 2332 2333retry: 2334 p = vm_page_lookup(object, pindex); 2335 if (p && vm_page_sleep_busy(p, FALSE, "init4p")) 2336 goto retry; 2337 2338 if (p == NULL) { 2339 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2340 if (p == NULL) 2341 return; 2342 m[0] = p; 2343 2344 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2345 vm_page_free(p); 2346 return; 2347 } 2348 2349 p = vm_page_lookup(object, pindex); 2350 vm_page_wakeup(p); 2351 } 2352 2353 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2354 if (ptepa & (NBPDR - 1)) { 2355 return; 2356 } 2357 2358 p->valid = VM_PAGE_BITS_ALL; 2359 2360 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2361 npdes = size >> PDRSHIFT; 2362 for(i=0;i<npdes;i++) { 2363 pmap->pm_pdir[ptepindex] = 2364 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2365 ptepa += NBPDR; 2366 ptepindex += 1; 2367 } 2368 vm_page_flag_set(p, PG_MAPPED); 2369 invltlb(); 2370 return; 2371 } 2372 2373 psize = i386_btop(size); 2374 2375 if ((object->type != OBJT_VNODE) || 2376 (limit && (psize > MAX_INIT_PT) && 2377 (object->resident_page_count > MAX_INIT_PT))) { 2378 return; 2379 } 2380 2381 if (psize + pindex > object->size) 2382 psize = object->size - pindex; 2383 2384 mpte = NULL; 2385 /* 2386 * if we are processing a major portion of the object, then scan the 2387 * entire thing. 2388 */ 2389 if (psize > (object->resident_page_count >> 2)) { 2390 objpgs = psize; 2391 2392 for (p = TAILQ_FIRST(&object->memq); 2393 ((objpgs > 0) && (p != NULL)); 2394 p = TAILQ_NEXT(p, listq)) { 2395 2396 tmpidx = p->pindex; 2397 if (tmpidx < pindex) { 2398 continue; 2399 } 2400 tmpidx -= pindex; 2401 if (tmpidx >= psize) { 2402 continue; 2403 } 2404 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2405 (p->busy == 0) && 2406 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2407 if ((p->queue - p->pc) == PQ_CACHE) 2408 vm_page_deactivate(p); 2409 vm_page_busy(p); 2410 mpte = pmap_enter_quick(pmap, 2411 addr + i386_ptob(tmpidx), 2412 VM_PAGE_TO_PHYS(p), mpte); 2413 vm_page_flag_set(p, PG_MAPPED); 2414 vm_page_wakeup(p); 2415 } 2416 objpgs -= 1; 2417 } 2418 } else { 2419 /* 2420 * else lookup the pages one-by-one. 2421 */ 2422 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2423 p = vm_page_lookup(object, tmpidx + pindex); 2424 if (p && 2425 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2426 (p->busy == 0) && 2427 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2428 if ((p->queue - p->pc) == PQ_CACHE) 2429 vm_page_deactivate(p); 2430 vm_page_busy(p); 2431 mpte = pmap_enter_quick(pmap, 2432 addr + i386_ptob(tmpidx), 2433 VM_PAGE_TO_PHYS(p), mpte); 2434 vm_page_flag_set(p, PG_MAPPED); 2435 vm_page_wakeup(p); 2436 } 2437 } 2438 } 2439 return; 2440} 2441 2442/* 2443 * pmap_prefault provides a quick way of clustering 2444 * pagefaults into a processes address space. It is a "cousin" 2445 * of pmap_object_init_pt, except it runs at page fault time instead 2446 * of mmap time. 2447 */ 2448#define PFBAK 4 2449#define PFFOR 4 2450#define PAGEORDER_SIZE (PFBAK+PFFOR) 2451 2452static int pmap_prefault_pageorder[] = { 2453 -PAGE_SIZE, PAGE_SIZE, 2454 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2455 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2456 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2457}; 2458 2459void 2460pmap_prefault(pmap, addra, entry) 2461 pmap_t pmap; 2462 vm_offset_t addra; 2463 vm_map_entry_t entry; 2464{ 2465 int i; 2466 vm_offset_t starta; 2467 vm_offset_t addr; 2468 vm_pindex_t pindex; 2469 vm_page_t m, mpte; 2470 vm_object_t object; 2471 2472 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) 2473 return; 2474 2475 object = entry->object.vm_object; 2476 2477 starta = addra - PFBAK * PAGE_SIZE; 2478 if (starta < entry->start) { 2479 starta = entry->start; 2480 } else if (starta > addra) { 2481 starta = 0; 2482 } 2483 2484 mpte = NULL; 2485 for (i = 0; i < PAGEORDER_SIZE; i++) { 2486 vm_object_t lobject; 2487 unsigned *pte; 2488 2489 addr = addra + pmap_prefault_pageorder[i]; 2490 if (addr > addra + (PFFOR * PAGE_SIZE)) 2491 addr = 0; 2492 2493 if (addr < starta || addr >= entry->end) 2494 continue; 2495 2496 if ((*pmap_pde(pmap, addr)) == NULL) 2497 continue; 2498 2499 pte = (unsigned *) vtopte(addr); 2500 if (*pte) 2501 continue; 2502 2503 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2504 lobject = object; 2505 for (m = vm_page_lookup(lobject, pindex); 2506 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2507 lobject = lobject->backing_object) { 2508 if (lobject->backing_object_offset & PAGE_MASK) 2509 break; 2510 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2511 m = vm_page_lookup(lobject->backing_object, pindex); 2512 } 2513 2514 /* 2515 * give-up when a page is not in memory 2516 */ 2517 if (m == NULL) 2518 break; 2519 2520 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2521 (m->busy == 0) && 2522 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2523 2524 if ((m->queue - m->pc) == PQ_CACHE) { 2525 vm_page_deactivate(m); 2526 } 2527 vm_page_busy(m); 2528 mpte = pmap_enter_quick(pmap, addr, 2529 VM_PAGE_TO_PHYS(m), mpte); 2530 vm_page_flag_set(m, PG_MAPPED); 2531 vm_page_wakeup(m); 2532 } 2533 } 2534} 2535 2536/* 2537 * Routine: pmap_change_wiring 2538 * Function: Change the wiring attribute for a map/virtual-address 2539 * pair. 2540 * In/out conditions: 2541 * The mapping must already exist in the pmap. 2542 */ 2543void 2544pmap_change_wiring(pmap, va, wired) 2545 register pmap_t pmap; 2546 vm_offset_t va; 2547 boolean_t wired; 2548{ 2549 register unsigned *pte; 2550 2551 if (pmap == NULL) 2552 return; 2553 2554 pte = pmap_pte(pmap, va); 2555 2556 if (wired && !pmap_pte_w(pte)) 2557 pmap->pm_stats.wired_count++; 2558 else if (!wired && pmap_pte_w(pte)) 2559 pmap->pm_stats.wired_count--; 2560 2561 /* 2562 * Wiring is not a hardware characteristic so there is no need to 2563 * invalidate TLB. 2564 */ 2565 pmap_pte_set_w(pte, wired); 2566} 2567 2568 2569 2570/* 2571 * Copy the range specified by src_addr/len 2572 * from the source map to the range dst_addr/len 2573 * in the destination map. 2574 * 2575 * This routine is only advisory and need not do anything. 2576 */ 2577 2578void 2579pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2580 pmap_t dst_pmap, src_pmap; 2581 vm_offset_t dst_addr; 2582 vm_size_t len; 2583 vm_offset_t src_addr; 2584{ 2585 vm_offset_t addr; 2586 vm_offset_t end_addr = src_addr + len; 2587 vm_offset_t pdnxt; 2588 unsigned src_frame, dst_frame; 2589 2590 if (dst_addr != src_addr) 2591 return; 2592 2593 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2594 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2595 return; 2596 } 2597 2598 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2599 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2600 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2601#if defined(SMP) 2602 /* The page directory is not shared between CPUs */ 2603 cpu_invltlb(); 2604#else 2605 invltlb(); 2606#endif 2607 } 2608 2609 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2610 unsigned *src_pte, *dst_pte; 2611 vm_page_t dstmpte, srcmpte; 2612 vm_offset_t srcptepaddr; 2613 unsigned ptepindex; 2614 2615#if !defined(MAX_PERF) 2616 if (addr >= UPT_MIN_ADDRESS) 2617 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2618#endif 2619 2620 /* 2621 * Don't let optional prefaulting of pages make us go 2622 * way below the low water mark of free pages or way 2623 * above high water mark of used pv entries. 2624 */ 2625 if (cnt.v_free_count < cnt.v_free_reserved || 2626 pv_entry_count > pv_entry_high_water) 2627 break; 2628 2629 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2630 ptepindex = addr >> PDRSHIFT; 2631 2632 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2633 if (srcptepaddr == 0) 2634 continue; 2635 2636 if (srcptepaddr & PG_PS) { 2637 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2638 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2639 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE; 2640 } 2641 continue; 2642 } 2643 2644 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2645 if ((srcmpte == NULL) || 2646 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2647 continue; 2648 2649 if (pdnxt > end_addr) 2650 pdnxt = end_addr; 2651 2652 src_pte = (unsigned *) vtopte(addr); 2653 dst_pte = (unsigned *) avtopte(addr); 2654 while (addr < pdnxt) { 2655 unsigned ptetemp; 2656 ptetemp = *src_pte; 2657 /* 2658 * we only virtual copy managed pages 2659 */ 2660 if ((ptetemp & PG_MANAGED) != 0) { 2661 /* 2662 * We have to check after allocpte for the 2663 * pte still being around... allocpte can 2664 * block. 2665 */ 2666 dstmpte = pmap_allocpte(dst_pmap, addr); 2667 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2668 /* 2669 * Clear the modified and 2670 * accessed (referenced) bits 2671 * during the copy. 2672 */ 2673 *dst_pte = ptetemp & ~(PG_M | PG_A); 2674 dst_pmap->pm_stats.resident_count++; 2675 pmap_insert_entry(dst_pmap, addr, 2676 dstmpte, 2677 (ptetemp & PG_FRAME)); 2678 } else { 2679 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2680 } 2681 if (dstmpte->hold_count >= srcmpte->hold_count) 2682 break; 2683 } 2684 addr += PAGE_SIZE; 2685 src_pte++; 2686 dst_pte++; 2687 } 2688 } 2689} 2690 2691/* 2692 * Routine: pmap_kernel 2693 * Function: 2694 * Returns the physical map handle for the kernel. 2695 */ 2696pmap_t 2697pmap_kernel() 2698{ 2699 return (kernel_pmap); 2700} 2701 2702/* 2703 * pmap_zero_page zeros the specified hardware page by mapping 2704 * the page into KVM and using bzero to clear its contents. 2705 */ 2706void 2707pmap_zero_page(phys) 2708 vm_offset_t phys; 2709{ 2710#ifdef SMP 2711#if !defined(MAX_PERF) 2712 if (*(int *) prv_CMAP3) 2713 panic("pmap_zero_page: prv_CMAP3 busy"); 2714#endif 2715 2716 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2717 cpu_invlpg(prv_CADDR3); 2718 2719#if defined(I686_CPU) 2720 if (cpu_class == CPUCLASS_686) 2721 i686_pagezero(prv_CADDR3); 2722 else 2723#endif 2724 bzero(prv_CADDR3, PAGE_SIZE); 2725 2726 *(int *) prv_CMAP3 = 0; 2727#else 2728#if !defined(MAX_PERF) 2729 if (*(int *) CMAP2) 2730 panic("pmap_zero_page: CMAP2 busy"); 2731#endif 2732 2733 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2734 invltlb_1pg((vm_offset_t)CADDR2); 2735 2736#if defined(I686_CPU) 2737 if (cpu_class == CPUCLASS_686) 2738 i686_pagezero(CADDR2); 2739 else 2740#endif 2741 bzero(CADDR2, PAGE_SIZE); 2742 *(int *) CMAP2 = 0; 2743#endif 2744} 2745 2746/* 2747 * pmap_zero_page_area zeros the specified hardware page by mapping 2748 * the page into KVM and using bzero to clear its contents. 2749 * 2750 * off and size may not cover an area beyond a single hardware page. 2751 */ 2752void 2753pmap_zero_page_area(phys, off, size) 2754 vm_offset_t phys; 2755 int off; 2756 int size; 2757{ 2758#ifdef SMP 2759#if !defined(MAX_PERF) 2760 if (*(int *) prv_CMAP3) 2761 panic("pmap_zero_page: prv_CMAP3 busy"); 2762#endif 2763 2764 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2765 cpu_invlpg(prv_CADDR3); 2766 2767#if defined(I686_CPU) 2768 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2769 i686_pagezero(prv_CADDR3); 2770 else 2771#endif 2772 bzero((char *)prv_CADDR3 + off, size); 2773 2774 *(int *) prv_CMAP3 = 0; 2775#else 2776#if !defined(MAX_PERF) 2777 if (*(int *) CMAP2) 2778 panic("pmap_zero_page: CMAP2 busy"); 2779#endif 2780 2781 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2782 invltlb_1pg((vm_offset_t)CADDR2); 2783 2784#if defined(I686_CPU) 2785 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2786 i686_pagezero(CADDR2); 2787 else 2788#endif 2789 bzero((char *)CADDR2 + off, size); 2790 *(int *) CMAP2 = 0; 2791#endif 2792} 2793 2794/* 2795 * pmap_copy_page copies the specified (machine independent) 2796 * page by mapping the page into virtual memory and using 2797 * bcopy to copy the page, one machine dependent page at a 2798 * time. 2799 */ 2800void 2801pmap_copy_page(src, dst) 2802 vm_offset_t src; 2803 vm_offset_t dst; 2804{ 2805#ifdef SMP 2806#if !defined(MAX_PERF) 2807 if (*(int *) prv_CMAP1) 2808 panic("pmap_copy_page: prv_CMAP1 busy"); 2809 if (*(int *) prv_CMAP2) 2810 panic("pmap_copy_page: prv_CMAP2 busy"); 2811#endif 2812 2813 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2814 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2815 2816 cpu_invlpg(prv_CADDR1); 2817 cpu_invlpg(prv_CADDR2); 2818 2819 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE); 2820 2821 *(int *) prv_CMAP1 = 0; 2822 *(int *) prv_CMAP2 = 0; 2823#else 2824#if !defined(MAX_PERF) 2825 if (*(int *) CMAP1 || *(int *) CMAP2) 2826 panic("pmap_copy_page: CMAP busy"); 2827#endif 2828 2829 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2830 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2831#if defined(I386_CPU) 2832 if (cpu_class == CPUCLASS_386) { 2833 invltlb(); 2834 } else 2835#endif 2836 { 2837 invlpg((u_int)CADDR1); 2838 invlpg((u_int)CADDR2); 2839 } 2840 2841 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2842 2843 *(int *) CMAP1 = 0; 2844 *(int *) CMAP2 = 0; 2845#endif 2846} 2847 2848 2849/* 2850 * Routine: pmap_pageable 2851 * Function: 2852 * Make the specified pages (by pmap, offset) 2853 * pageable (or not) as requested. 2854 * 2855 * A page which is not pageable may not take 2856 * a fault; therefore, its page table entry 2857 * must remain valid for the duration. 2858 * 2859 * This routine is merely advisory; pmap_enter 2860 * will specify that these pages are to be wired 2861 * down (or not) as appropriate. 2862 */ 2863void 2864pmap_pageable(pmap, sva, eva, pageable) 2865 pmap_t pmap; 2866 vm_offset_t sva, eva; 2867 boolean_t pageable; 2868{ 2869} 2870 2871/* 2872 * this routine returns true if a physical page resides 2873 * in the given pmap. 2874 */ 2875boolean_t 2876pmap_page_exists(pmap, pa) 2877 pmap_t pmap; 2878 vm_offset_t pa; 2879{ 2880 register pv_entry_t pv; 2881 pv_table_t *ppv; 2882 int s; 2883 2884 if (!pmap_is_managed(pa)) 2885 return FALSE; 2886 2887 s = splvm(); 2888 2889 ppv = pa_to_pvh(pa); 2890 /* 2891 * Not found, check current mappings returning immediately if found. 2892 */ 2893 for (pv = TAILQ_FIRST(&ppv->pv_list); 2894 pv; 2895 pv = TAILQ_NEXT(pv, pv_list)) { 2896 if (pv->pv_pmap == pmap) { 2897 splx(s); 2898 return TRUE; 2899 } 2900 } 2901 splx(s); 2902 return (FALSE); 2903} 2904 2905#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2906/* 2907 * Remove all pages from specified address space 2908 * this aids process exit speeds. Also, this code 2909 * is special cased for current process only, but 2910 * can have the more generic (and slightly slower) 2911 * mode enabled. This is much faster than pmap_remove 2912 * in the case of running down an entire address space. 2913 */ 2914void 2915pmap_remove_pages(pmap, sva, eva) 2916 pmap_t pmap; 2917 vm_offset_t sva, eva; 2918{ 2919 unsigned *pte, tpte; 2920 pv_table_t *ppv; 2921 pv_entry_t pv, npv; 2922 int s; 2923 2924#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2925 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) { 2926 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2927 return; 2928 } 2929#endif 2930 2931 s = splvm(); 2932 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2933 pv; 2934 pv = npv) { 2935 2936 if (pv->pv_va >= eva || pv->pv_va < sva) { 2937 npv = TAILQ_NEXT(pv, pv_plist); 2938 continue; 2939 } 2940 2941#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2942 pte = (unsigned *)vtopte(pv->pv_va); 2943#else 2944 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2945#endif 2946 tpte = *pte; 2947 2948/* 2949 * We cannot remove wired pages from a process' mapping at this time 2950 */ 2951 if (tpte & PG_W) { 2952 npv = TAILQ_NEXT(pv, pv_plist); 2953 continue; 2954 } 2955 *pte = 0; 2956 2957 ppv = pa_to_pvh(tpte); 2958 2959 pv->pv_pmap->pm_stats.resident_count--; 2960 2961 /* 2962 * Update the vm_page_t clean and reference bits. 2963 */ 2964 if (tpte & PG_M) { 2965 vm_page_dirty(ppv->pv_vm_page); 2966 } 2967 2968 2969 npv = TAILQ_NEXT(pv, pv_plist); 2970 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2971 2972 ppv->pv_list_count--; 2973 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2974 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2975 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 2976 } 2977 2978 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2979 free_pv_entry(pv); 2980 } 2981 splx(s); 2982 pmap_TLB_invalidate_all(pmap); 2983} 2984 2985/* 2986 * pmap_testbit tests bits in pte's 2987 * note that the testbit/changebit routines are inline, 2988 * and a lot of things compile-time evaluate. 2989 */ 2990static boolean_t 2991pmap_testbit(pa, bit) 2992 register vm_offset_t pa; 2993 int bit; 2994{ 2995 register pv_entry_t pv; 2996 pv_table_t *ppv; 2997 unsigned *pte; 2998 int s; 2999 3000 if (!pmap_is_managed(pa)) 3001 return FALSE; 3002 3003 ppv = pa_to_pvh(pa); 3004 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 3005 return FALSE; 3006 3007 s = splvm(); 3008 3009 for (pv = TAILQ_FIRST(&ppv->pv_list); 3010 pv; 3011 pv = TAILQ_NEXT(pv, pv_list)) { 3012 3013 /* 3014 * if the bit being tested is the modified bit, then 3015 * mark clean_map and ptes as never 3016 * modified. 3017 */ 3018 if (bit & (PG_A|PG_M)) { 3019 if (!pmap_track_modified(pv->pv_va)) 3020 continue; 3021 } 3022 3023#if defined(PMAP_DIAGNOSTIC) 3024 if (!pv->pv_pmap) { 3025 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 3026 continue; 3027 } 3028#endif 3029 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3030 if (*pte & bit) { 3031 splx(s); 3032 return TRUE; 3033 } 3034 } 3035 splx(s); 3036 return (FALSE); 3037} 3038 3039/* 3040 * this routine is used to modify bits in ptes 3041 */ 3042static __inline void 3043pmap_changebit(pa, bit, setem) 3044 vm_offset_t pa; 3045 int bit; 3046 boolean_t setem; 3047{ 3048 register pv_entry_t pv; 3049 pv_table_t *ppv; 3050 register unsigned *pte; 3051 int s; 3052 3053 if (!pmap_is_managed(pa)) 3054 return; 3055 3056 s = splvm(); 3057 ppv = pa_to_pvh(pa); 3058 3059 /* 3060 * Loop over all current mappings setting/clearing as appropos If 3061 * setting RO do we need to clear the VAC? 3062 */ 3063 for (pv = TAILQ_FIRST(&ppv->pv_list); 3064 pv; 3065 pv = TAILQ_NEXT(pv, pv_list)) { 3066 3067 /* 3068 * don't write protect pager mappings 3069 */ 3070 if (!setem && (bit == PG_RW)) { 3071 if (!pmap_track_modified(pv->pv_va)) 3072 continue; 3073 } 3074 3075#if defined(PMAP_DIAGNOSTIC) 3076 if (!pv->pv_pmap) { 3077 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 3078 continue; 3079 } 3080#endif 3081 3082 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3083 3084 if (setem) { 3085 *(int *)pte |= bit; 3086 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3087 } else { 3088 vm_offset_t pbits = *(vm_offset_t *)pte; 3089 if (pbits & bit) { 3090 if (bit == PG_RW) { 3091 if (pbits & PG_M) { 3092 vm_page_dirty(ppv->pv_vm_page); 3093 } 3094 *(int *)pte = pbits & ~(PG_M|PG_RW); 3095 } else { 3096 *(int *)pte = pbits & ~bit; 3097 } 3098 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3099 } 3100 } 3101 } 3102 splx(s); 3103} 3104 3105/* 3106 * pmap_clearbit: 3107 * 3108 * Clear a bit/bits in every pte mapping a given physical page. Making 3109 * this inline allows the pmap_changebit inline to be well optimized. 3110 */ 3111static __inline void 3112pmap_clearbit( 3113 vm_offset_t pa, 3114 int bit) 3115{ 3116 pmap_changebit(pa, bit, FALSE); 3117} 3118 3119/* 3120 * pmap_page_protect: 3121 * 3122 * Lower the permission for all mappings to a given page. 3123 */ 3124void 3125pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 3126{ 3127 if ((prot & VM_PROT_WRITE) == 0) { 3128 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3129 pmap_clearbit(phys, PG_RW); 3130 } else { 3131 pmap_remove_all(phys); 3132 } 3133 } 3134} 3135 3136vm_offset_t 3137pmap_phys_address(ppn) 3138 int ppn; 3139{ 3140 return (i386_ptob(ppn)); 3141} 3142 3143/* 3144 * pmap_ts_referenced: 3145 * 3146 * Return the count of reference bits for a page, clearing all of them. 3147 */ 3148int 3149pmap_ts_referenced(vm_offset_t pa) 3150{ 3151 register pv_entry_t pv, pvf, pvn; 3152 pv_table_t *ppv; 3153 unsigned *pte; 3154 int s; 3155 int rtval = 0; 3156 3157 if (!pmap_is_managed(pa)) 3158 return (rtval); 3159 3160 s = splvm(); 3161 3162 ppv = pa_to_pvh(pa); 3163 3164 if ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 3165 3166 pvf = pv; 3167 3168 do { 3169 pvn = TAILQ_NEXT(pv, pv_list); 3170 3171 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 3172 3173 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3174 3175 if (!pmap_track_modified(pv->pv_va)) 3176 continue; 3177 3178 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3179 3180 if (pte && *pte & PG_A) { 3181 *pte &= ~PG_A; 3182 3183 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3184 3185 rtval++; 3186 if (rtval > 4) { 3187 break; 3188 } 3189 } 3190 } while ((pv = pvn) != NULL && pv != pvf); 3191 } 3192 splx(s); 3193 3194 return (rtval); 3195} 3196 3197/* 3198 * pmap_is_modified: 3199 * 3200 * Return whether or not the specified physical page was modified 3201 * in any physical maps. 3202 */ 3203boolean_t 3204pmap_is_modified(vm_offset_t pa) 3205{ 3206 return pmap_testbit((pa), PG_M); 3207} 3208 3209/* 3210 * Clear the modify bits on the specified physical page. 3211 */ 3212void 3213pmap_clear_modify(vm_offset_t pa) 3214{ 3215 pmap_clearbit(pa, PG_M); 3216} 3217 3218/* 3219 * pmap_clear_reference: 3220 * 3221 * Clear the reference bit on the specified physical page. 3222 */ 3223void 3224pmap_clear_reference(vm_offset_t pa) 3225{ 3226 pmap_clearbit(pa, PG_A); 3227} 3228 3229/* 3230 * Miscellaneous support routines follow 3231 */ 3232 3233static void 3234i386_protection_init() 3235{ 3236 register int *kp, prot; 3237 3238 kp = protection_codes; 3239 for (prot = 0; prot < 8; prot++) { 3240 switch (prot) { 3241 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3242 /* 3243 * Read access is also 0. There isn't any execute bit, 3244 * so just make it readable. 3245 */ 3246 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3247 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3248 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3249 *kp++ = 0; 3250 break; 3251 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3252 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3253 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3254 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3255 *kp++ = PG_RW; 3256 break; 3257 } 3258 } 3259} 3260 3261/* 3262 * Map a set of physical memory pages into the kernel virtual 3263 * address space. Return a pointer to where it is mapped. This 3264 * routine is intended to be used for mapping device memory, 3265 * NOT real memory. 3266 */ 3267void * 3268pmap_mapdev(pa, size) 3269 vm_offset_t pa; 3270 vm_size_t size; 3271{ 3272 vm_offset_t va, tmpva; 3273 unsigned *pte; 3274 3275 size = roundup(size, PAGE_SIZE); 3276 3277 va = kmem_alloc_pageable(kernel_map, size); 3278#if !defined(MAX_PERF) 3279 if (!va) 3280 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3281#endif 3282 3283 pa = pa & PG_FRAME; 3284 for (tmpva = va; size > 0;) { 3285 pte = (unsigned *)vtopte(tmpva); 3286 *pte = pa | PG_RW | PG_V | pgeflag; 3287 size -= PAGE_SIZE; 3288 tmpva += PAGE_SIZE; 3289 pa += PAGE_SIZE; 3290 } 3291 invltlb(); 3292 3293 return ((void *) va); 3294} 3295 3296/* 3297 * perform the pmap work for mincore 3298 */ 3299int 3300pmap_mincore(pmap, addr) 3301 pmap_t pmap; 3302 vm_offset_t addr; 3303{ 3304 3305 unsigned *ptep, pte; 3306 vm_page_t m; 3307 int val = 0; 3308 3309 ptep = pmap_pte(pmap, addr); 3310 if (ptep == 0) { 3311 return 0; 3312 } 3313 3314 if ((pte = *ptep) != 0) { 3315 pv_table_t *ppv; 3316 vm_offset_t pa; 3317 3318 val = MINCORE_INCORE; 3319 if ((pte & PG_MANAGED) == 0) 3320 return val; 3321 3322 pa = pte & PG_FRAME; 3323 3324 ppv = pa_to_pvh((pa & PG_FRAME)); 3325 m = ppv->pv_vm_page; 3326 3327 /* 3328 * Modified by us 3329 */ 3330 if (pte & PG_M) 3331 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3332 /* 3333 * Modified by someone 3334 */ 3335 else if (m->dirty || pmap_is_modified(pa)) 3336 val |= MINCORE_MODIFIED_OTHER; 3337 /* 3338 * Referenced by us 3339 */ 3340 if (pte & PG_A) 3341 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3342 3343 /* 3344 * Referenced by someone 3345 */ 3346 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(pa)) { 3347 val |= MINCORE_REFERENCED_OTHER; 3348 vm_page_flag_set(m, PG_REFERENCED); 3349 } 3350 } 3351 return val; 3352} 3353 3354void 3355pmap_activate(struct proc *p) 3356{ 3357 pmap_t pmap; 3358 3359 pmap = vmspace_pmap(p->p_vmspace); 3360#if defined(SMP) 3361 pmap->pm_active |= 1 << cpuid; 3362#else 3363 pmap->pm_active |= 1; 3364#endif 3365#if defined(SWTCH_OPTIM_STATS) 3366 tlb_flush_count++; 3367#endif 3368 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir)); 3369} 3370 3371vm_offset_t 3372pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3373 3374 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3375 return addr; 3376 } 3377 3378 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3379 return addr; 3380} 3381 3382 3383#if defined(PMAP_DEBUG) 3384pmap_pid_dump(int pid) { 3385 pmap_t pmap; 3386 struct proc *p; 3387 int npte = 0; 3388 int index; 3389 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3390 if (p->p_pid != pid) 3391 continue; 3392 3393 if (p->p_vmspace) { 3394 int i,j; 3395 index = 0; 3396 pmap = vmspace_pmap(p->p_vmspace); 3397 for(i=0;i<1024;i++) { 3398 pd_entry_t *pde; 3399 unsigned *pte; 3400 unsigned base = i << PDRSHIFT; 3401 3402 pde = &pmap->pm_pdir[i]; 3403 if (pde && pmap_pde_v(pde)) { 3404 for(j=0;j<1024;j++) { 3405 unsigned va = base + (j << PAGE_SHIFT); 3406 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3407 if (index) { 3408 index = 0; 3409 printf("\n"); 3410 } 3411 return npte; 3412 } 3413 pte = pmap_pte_quick( pmap, va); 3414 if (pte && pmap_pte_v(pte)) { 3415 vm_offset_t pa; 3416 vm_page_t m; 3417 pa = *(int *)pte; 3418 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3419 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3420 va, pa, m->hold_count, m->wire_count, m->flags); 3421 npte++; 3422 index++; 3423 if (index >= 2) { 3424 index = 0; 3425 printf("\n"); 3426 } else { 3427 printf(" "); 3428 } 3429 } 3430 } 3431 } 3432 } 3433 } 3434 } 3435 return npte; 3436} 3437#endif 3438 3439#if defined(DEBUG) 3440 3441static void pads __P((pmap_t pm)); 3442void pmap_pvdump __P((vm_offset_t pa)); 3443 3444/* print address space of pmap*/ 3445static void 3446pads(pm) 3447 pmap_t pm; 3448{ 3449 unsigned va, i, j; 3450 unsigned *ptep; 3451 3452 if (pm == kernel_pmap) 3453 return; 3454 for (i = 0; i < 1024; i++) 3455 if (pm->pm_pdir[i]) 3456 for (j = 0; j < 1024; j++) { 3457 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3458 if (pm == kernel_pmap && va < KERNBASE) 3459 continue; 3460 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3461 continue; 3462 ptep = pmap_pte_quick(pm, va); 3463 if (pmap_pte_v(ptep)) 3464 printf("%x:%x ", va, *(int *) ptep); 3465 }; 3466 3467} 3468 3469void 3470pmap_pvdump(pa) 3471 vm_offset_t pa; 3472{ 3473 pv_table_t *ppv; 3474 register pv_entry_t pv; 3475 3476 printf("pa %x", pa); 3477 ppv = pa_to_pvh(pa); 3478 for (pv = TAILQ_FIRST(&ppv->pv_list); 3479 pv; 3480 pv = TAILQ_NEXT(pv, pv_list)) { 3481#ifdef used_to_be 3482 printf(" -> pmap %p, va %x, flags %x", 3483 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags); 3484#endif 3485 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3486 pads(pv->pv_pmap); 3487 } 3488 printf(" "); 3489} 3490#endif 3491