pmap.c revision 47292
1234449Sobrien/* 268349Sobrien * Copyright (c) 1991 Regents of the University of California. 3234449Sobrien * All rights reserved. 468349Sobrien * Copyright (c) 1994 John S. Dyson 568349Sobrien * All rights reserved. 668349Sobrien * Copyright (c) 1994 David Greenman 768349Sobrien * All rights reserved. 868349Sobrien * 968349Sobrien * This code is derived from software contributed to Berkeley by 1068349Sobrien * the Systems Programming Group of the University of Utah Computer 1168349Sobrien * Science Department and William Jolitz of UUNET Technologies Inc. 1268349Sobrien * 13186691Sobrien * Redistribution and use in source and binary forms, with or without 1468349Sobrien * modification, are permitted provided that the following conditions 1568349Sobrien * are met: 16186691Sobrien * 1. Redistributions of source code must retain the above copyright 1768349Sobrien * notice, this list of conditions and the following disclaimer. 1868349Sobrien * 2. Redistributions in binary form must reproduce the above copyright 1968349Sobrien * notice, this list of conditions and the following disclaimer in the 20186691Sobrien * documentation and/or other materials provided with the distribution. 2168349Sobrien * 3. All advertising materials mentioning features or use of this software 2268349Sobrien * must display the following acknowledgement: 2368349Sobrien * This product includes software developed by the University of 24186691Sobrien * California, Berkeley and its contributors. 2568349Sobrien * 4. Neither the name of the University nor the names of its contributors 2668349Sobrien * may be used to endorse or promote products derived from this software 2768349Sobrien * without specific prior written permission. 28186691Sobrien * 29186691Sobrien * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 3068349Sobrien * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31186691Sobrien * 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 1.234 1999/04/28 01:03:23 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_vm86.h" 75#include "opt_user_ldt.h" 76 77#include <sys/param.h> 78#include <sys/systm.h> 79#include <sys/proc.h> 80#include <sys/msgbuf.h> 81#include <sys/vmmeter.h> 82#include <sys/mman.h> 83 84#include <vm/vm.h> 85#include <vm/vm_param.h> 86#include <vm/vm_prot.h> 87#include <sys/lock.h> 88#include <vm/vm_kern.h> 89#include <vm/vm_page.h> 90#include <vm/vm_map.h> 91#include <vm/vm_object.h> 92#include <vm/vm_extern.h> 93#include <vm/vm_pageout.h> 94#include <vm/vm_pager.h> 95#include <vm/vm_zone.h> 96 97#include <sys/user.h> 98 99#include <machine/cputypes.h> 100#include <machine/md_var.h> 101#include <machine/specialreg.h> 102#if defined(SMP) || defined(APIC_IO) 103#include <machine/smp.h> 104#include <machine/apic.h> 105#include <machine/segments.h> 106#include <machine/tss.h> 107#include <machine/globaldata.h> 108#endif /* SMP || APIC_IO */ 109 110#define PMAP_KEEP_PDIRS 111#ifndef PMAP_SHPGPERPROC 112#define PMAP_SHPGPERPROC 200 113#endif 114 115#if defined(DIAGNOSTIC) 116#define PMAP_DIAGNOSTIC 117#endif 118 119#define MINPV 2048 120 121#if !defined(PMAP_DIAGNOSTIC) 122#define PMAP_INLINE __inline 123#else 124#define PMAP_INLINE 125#endif 126 127/* 128 * Get PDEs and PTEs for user/kernel address space 129 */ 130#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 131#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 132 133#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 134#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 135#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 136#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 137#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 138 139#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 140#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 141 142/* 143 * Given a map and a machine independent protection code, 144 * convert to a vax protection code. 145 */ 146#define pte_prot(m, p) (protection_codes[p]) 147static int protection_codes[8]; 148 149#define pa_index(pa) atop((pa) - vm_first_phys) 150#define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) 151 152static struct pmap kernel_pmap_store; 153pmap_t kernel_pmap; 154 155vm_offset_t avail_start; /* PA of first available physical page */ 156vm_offset_t avail_end; /* PA of last available physical page */ 157vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 158vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 159static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 160static vm_offset_t vm_first_phys; 161static int pgeflag; /* PG_G or-in */ 162static int pseflag; /* PG_PS or-in */ 163static int pv_npg; 164 165static vm_object_t kptobj; 166 167static int nkpt; 168vm_offset_t kernel_vm_end; 169 170/* 171 * Data for the pv entry allocation mechanism 172 */ 173static vm_zone_t pvzone; 174static struct vm_zone pvzone_store; 175static struct vm_object pvzone_obj; 176static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0; 177static int pmap_pagedaemon_waken = 0; 178static struct pv_entry *pvinit; 179 180/* 181 * All those kernel PT submaps that BSD is so fond of 182 */ 183pt_entry_t *CMAP1 = 0; 184static pt_entry_t *CMAP2, *ptmmap; 185static pv_table_t *pv_table; 186caddr_t CADDR1 = 0, ptvmmap = 0; 187static caddr_t CADDR2; 188static pt_entry_t *msgbufmap; 189struct msgbuf *msgbufp=0; 190 191#ifdef SMP 192extern pt_entry_t *SMPpt; 193#else 194static pt_entry_t *PMAP1 = 0; 195static unsigned *PADDR1 = 0; 196#endif 197 198static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 199static unsigned * get_ptbase __P((pmap_t pmap)); 200static pv_entry_t get_pv_entry __P((void)); 201static void i386_protection_init __P((void)); 202static __inline void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 203static void pmap_clearbit __P((vm_offset_t pa, int bit)); 204 205static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 206static void pmap_remove_all __P((vm_offset_t pa)); 207static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 208 vm_offset_t pa, vm_page_t mpte)); 209static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 210 vm_offset_t sva)); 211static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 212static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 213 vm_offset_t va)); 214static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 215static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 216 vm_page_t mpte, vm_offset_t pa)); 217 218static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 219 220static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 221static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 222static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 223static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 224static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 225static vm_offset_t pmap_kmem_choose(vm_offset_t addr); 226 227static unsigned pdir4mb; 228 229/* 230 * Routine: pmap_pte 231 * Function: 232 * Extract the page table entry associated 233 * with the given map/virtual_address pair. 234 */ 235 236PMAP_INLINE unsigned * 237pmap_pte(pmap, va) 238 register pmap_t pmap; 239 vm_offset_t va; 240{ 241 unsigned *pdeaddr; 242 243 if (pmap) { 244 pdeaddr = (unsigned *) pmap_pde(pmap, va); 245 if (*pdeaddr & PG_PS) 246 return pdeaddr; 247 if (*pdeaddr) { 248 return get_ptbase(pmap) + i386_btop(va); 249 } 250 } 251 return (0); 252} 253 254/* 255 * Move the kernel virtual free pointer to the next 256 * 4MB. This is used to help improve performance 257 * by using a large (4MB) page for much of the kernel 258 * (.text, .data, .bss) 259 */ 260static vm_offset_t 261pmap_kmem_choose(vm_offset_t addr) { 262 vm_offset_t newaddr = addr; 263#ifndef DISABLE_PSE 264 if (cpu_feature & CPUID_PSE) { 265 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 266 } 267#endif 268 return newaddr; 269} 270 271/* 272 * Bootstrap the system enough to run with virtual memory. 273 * 274 * On the i386 this is called after mapping has already been enabled 275 * and just syncs the pmap module with what has already been done. 276 * [We can't call it easily with mapping off since the kernel is not 277 * mapped with PA == VA, hence we would have to relocate every address 278 * from the linked base (virtual) address "KERNBASE" to the actual 279 * (physical) address starting relative to 0] 280 */ 281void 282pmap_bootstrap(firstaddr, loadaddr) 283 vm_offset_t firstaddr; 284 vm_offset_t loadaddr; 285{ 286 vm_offset_t va; 287 pt_entry_t *pte; 288#ifdef SMP 289 int i, j; 290 struct globaldata *gd; 291#endif 292 293 avail_start = firstaddr; 294 295 /* 296 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 297 * large. It should instead be correctly calculated in locore.s and 298 * not based on 'first' (which is a physical address, not a virtual 299 * address, for the start of unused physical memory). The kernel 300 * page tables are NOT double mapped and thus should not be included 301 * in this calculation. 302 */ 303 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 304 virtual_avail = pmap_kmem_choose(virtual_avail); 305 306 virtual_end = VM_MAX_KERNEL_ADDRESS; 307 308 /* 309 * Initialize protection array. 310 */ 311 i386_protection_init(); 312 313 /* 314 * The kernel's pmap is statically allocated so we don't have to use 315 * pmap_create, which is unlikely to work correctly at this part of 316 * the boot sequence (XXX and which no longer exists). 317 */ 318 kernel_pmap = &kernel_pmap_store; 319 320 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 321 kernel_pmap->pm_count = 1; 322 kernel_pmap->pm_active = -1; /* don't allow deactivation */ 323 TAILQ_INIT(&kernel_pmap->pm_pvlist); 324 nkpt = NKPT; 325 326 /* 327 * Reserve some special page table entries/VA space for temporary 328 * mapping of pages. 329 */ 330#define SYSMAP(c, p, v, n) \ 331 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 332 333 va = virtual_avail; 334 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 335 336 /* 337 * CMAP1/CMAP2 are used for zeroing and copying pages. 338 */ 339 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 340 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 341 342 /* 343 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 344 * XXX ptmmap is not used. 345 */ 346 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 347 348 /* 349 * msgbufp is used to map the system message buffer. 350 * XXX msgbufmap is not used. 351 */ 352 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 353 atop(round_page(MSGBUF_SIZE))) 354 355#if !defined(SMP) 356 /* 357 * ptemap is used for pmap_pte_quick 358 */ 359 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 360#endif 361 362 virtual_avail = va; 363 364 *(int *) CMAP1 = *(int *) CMAP2 = 0; 365 *(int *) PTD = 0; 366 367 368 pgeflag = 0; 369#if !defined(SMP) 370 if (cpu_feature & CPUID_PGE) { 371 pgeflag = PG_G; 372 } 373#endif 374 375/* 376 * Initialize the 4MB page size flag 377 */ 378 pseflag = 0; 379/* 380 * The 4MB page version of the initial 381 * kernel page mapping. 382 */ 383 pdir4mb = 0; 384 385#if !defined(DISABLE_PSE) 386 if (cpu_feature & CPUID_PSE) { 387 unsigned ptditmp; 388 /* 389 * Enable the PSE mode 390 */ 391 load_cr4(rcr4() | CR4_PSE); 392 393 /* 394 * Note that we have enabled PSE mode 395 */ 396 pseflag = PG_PS; 397 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE)); 398 ptditmp &= ~(NBPDR - 1); 399 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 400 pdir4mb = ptditmp; 401 402#if !defined(SMP) 403 /* 404 * We can do the mapping here for the single processor 405 * case. We simply ignore the old page table page from 406 * now on. 407 */ 408 /* 409 * For SMP, we still need 4K pages to bootstrap APs, 410 * PSE will be enabled as soon as all APs are up. 411 */ 412 PTD[KPTDI] = (pd_entry_t) ptditmp; 413 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 414 invltlb(); 415#endif 416 } 417#endif 418 419#ifdef SMP 420 if (cpu_apic_address == 0) 421 panic("pmap_bootstrap: no local apic!"); 422 423 /* local apic is mapped on last page */ 424 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag | 425 (cpu_apic_address & PG_FRAME)); 426 427 for (i = 0; i < mp_napics; i++) { 428 for (j = 0; j < mp_napics; j++) { 429 /* same page frame as a previous IO apic? */ 430 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 431 (io_apic_address[0] & PG_FRAME)) { 432 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 433 + (NPTEPG-2-j)*PAGE_SIZE); 434 break; 435 } 436 /* use this slot if available */ 437 if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 0) { 438 SMPpt[NPTEPG-2-j] = (pt_entry_t)(PG_V | PG_RW | 439 pgeflag | (io_apic_address[i] & PG_FRAME)); 440 ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace 441 + (NPTEPG-2-j)*PAGE_SIZE); 442 break; 443 } 444 } 445 } 446 447 /* BSP does this itself, AP's get it pre-set */ 448 gd = &SMP_prvspace[0].globaldata; 449 gd->gd_prv_CMAP1 = &SMPpt[1]; 450 gd->gd_prv_CMAP2 = &SMPpt[2]; 451 gd->gd_prv_CMAP3 = &SMPpt[3]; 452 gd->gd_prv_PMAP1 = &SMPpt[4]; 453 gd->gd_prv_CADDR1 = SMP_prvspace[0].CPAGE1; 454 gd->gd_prv_CADDR2 = SMP_prvspace[0].CPAGE2; 455 gd->gd_prv_CADDR3 = SMP_prvspace[0].CPAGE3; 456 gd->gd_prv_PADDR1 = (unsigned *)SMP_prvspace[0].PPAGE1; 457#endif 458 459 invltlb(); 460} 461 462#ifdef SMP 463/* 464 * Set 4mb pdir for mp startup, and global flags 465 */ 466void 467pmap_set_opt(unsigned *pdir) { 468 int i; 469 470 if (pseflag && (cpu_feature & CPUID_PSE)) { 471 load_cr4(rcr4() | CR4_PSE); 472 if (pdir4mb) { 473 pdir[KPTDI] = pdir4mb; 474 } 475 } 476 477 if (pgeflag && (cpu_feature & CPUID_PGE)) { 478 load_cr4(rcr4() | CR4_PGE); 479 for(i = KPTDI; i < KPTDI + nkpt; i++) { 480 if (pdir[i]) { 481 pdir[i] |= PG_G; 482 } 483 } 484 } 485} 486 487/* 488 * Setup the PTD for the boot processor 489 */ 490void 491pmap_set_opt_bsp(void) 492{ 493 pmap_set_opt((unsigned *)kernel_pmap->pm_pdir); 494 pmap_set_opt((unsigned *)PTD); 495 invltlb(); 496} 497#endif 498 499/* 500 * Initialize the pmap module. 501 * Called by vm_init, to initialize any structures that the pmap 502 * system needs to map virtual memory. 503 * pmap_init has been enhanced to support in a fairly consistant 504 * way, discontiguous physical memory. 505 */ 506void 507pmap_init(phys_start, phys_end) 508 vm_offset_t phys_start, phys_end; 509{ 510 vm_offset_t addr; 511 vm_size_t s; 512 int i; 513 int initial_pvs; 514 515 /* 516 * object for kernel page table pages 517 */ 518 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE); 519 520 /* 521 * calculate the number of pv_entries needed 522 */ 523 vm_first_phys = phys_avail[0]; 524 for (i = 0; phys_avail[i + 1]; i += 2); 525 pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 526 527 /* 528 * Allocate memory for random pmap data structures. Includes the 529 * pv_head_table. 530 */ 531 s = (vm_size_t) (sizeof(pv_table_t) * pv_npg); 532 s = round_page(s); 533 534 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 535 pv_table = (pv_table_t *) addr; 536 for(i = 0; i < pv_npg; i++) { 537 vm_offset_t pa; 538 TAILQ_INIT(&pv_table[i].pv_list); 539 pv_table[i].pv_list_count = 0; 540 pa = vm_first_phys + i * PAGE_SIZE; 541 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 542 } 543 544 /* 545 * init the pv free list 546 */ 547 initial_pvs = pv_npg; 548 if (initial_pvs < MINPV) 549 initial_pvs = MINPV; 550 pvzone = &pvzone_store; 551 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 552 initial_pvs * sizeof (struct pv_entry)); 553 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, pv_npg); 554 555 /* 556 * Now it is safe to enable pv_table recording. 557 */ 558 pmap_initialized = TRUE; 559} 560 561/* 562 * Initialize the address space (zone) for the pv_entries. Set a 563 * high water mark so that the system can recover from excessive 564 * numbers of pv entries. 565 */ 566void 567pmap_init2() { 568 pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg; 569 pv_entry_high_water = 9 * (pv_entry_max / 10); 570 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 571} 572 573/* 574 * Used to map a range of physical addresses into kernel 575 * virtual address space. 576 * 577 * For now, VM is already on, we only need to map the 578 * specified memory. 579 */ 580vm_offset_t 581pmap_map(virt, start, end, prot) 582 vm_offset_t virt; 583 vm_offset_t start; 584 vm_offset_t end; 585 int prot; 586{ 587 while (start < end) { 588 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 589 virt += PAGE_SIZE; 590 start += PAGE_SIZE; 591 } 592 return (virt); 593} 594 595 596/*************************************************** 597 * Low level helper routines..... 598 ***************************************************/ 599 600#if defined(PMAP_DIAGNOSTIC) 601 602/* 603 * This code checks for non-writeable/modified pages. 604 * This should be an invalid condition. 605 */ 606static int 607pmap_nw_modified(pt_entry_t ptea) { 608 int pte; 609 610 pte = (int) ptea; 611 612 if ((pte & (PG_M|PG_RW)) == PG_M) 613 return 1; 614 else 615 return 0; 616} 617#endif 618 619 620/* 621 * this routine defines the region(s) of memory that should 622 * not be tested for the modified bit. 623 */ 624static PMAP_INLINE int 625pmap_track_modified( vm_offset_t va) { 626 if ((va < clean_sva) || (va >= clean_eva)) 627 return 1; 628 else 629 return 0; 630} 631 632static PMAP_INLINE void 633invltlb_1pg( vm_offset_t va) { 634#if defined(I386_CPU) 635 if (cpu_class == CPUCLASS_386) { 636 invltlb(); 637 } else 638#endif 639 { 640 invlpg(va); 641 } 642} 643 644static __inline void 645pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va) 646{ 647#if defined(SMP) 648 if (pmap->pm_active & (1 << cpuid)) 649 cpu_invlpg((void *)va); 650 if (pmap->pm_active & other_cpus) 651 smp_invltlb(); 652#else 653 if (pmap->pm_active) 654 invltlb_1pg(va); 655#endif 656} 657 658static __inline void 659pmap_TLB_invalidate_all(pmap_t pmap) 660{ 661#if defined(SMP) 662 if (pmap->pm_active & (1 << cpuid)) 663 cpu_invltlb(); 664 if (pmap->pm_active & other_cpus) 665 smp_invltlb(); 666#else 667 if (pmap->pm_active) 668 invltlb(); 669#endif 670} 671 672static unsigned * 673get_ptbase(pmap) 674 pmap_t pmap; 675{ 676 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 677 678 /* are we current address space or kernel? */ 679 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 680 return (unsigned *) PTmap; 681 } 682 /* otherwise, we are alternate address space */ 683 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 684 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 685#if defined(SMP) 686 /* The page directory is not shared between CPUs */ 687 cpu_invltlb(); 688#else 689 invltlb(); 690#endif 691 } 692 return (unsigned *) APTmap; 693} 694 695/* 696 * Super fast pmap_pte routine best used when scanning 697 * the pv lists. This eliminates many coarse-grained 698 * invltlb calls. Note that many of the pv list 699 * scans are across different pmaps. It is very wasteful 700 * to do an entire invltlb for checking a single mapping. 701 */ 702 703static unsigned * 704pmap_pte_quick(pmap, va) 705 register pmap_t pmap; 706 vm_offset_t va; 707{ 708 unsigned pde, newpf; 709 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) { 710 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 711 unsigned index = i386_btop(va); 712 /* are we current address space or kernel? */ 713 if ((pmap == kernel_pmap) || 714 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 715 return (unsigned *) PTmap + index; 716 } 717 newpf = pde & PG_FRAME; 718#ifdef SMP 719 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) { 720 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V; 721 cpu_invlpg(prv_PADDR1); 722 } 723 return prv_PADDR1 + ((unsigned) index & (NPTEPG - 1)); 724#else 725 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 726 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 727 invltlb_1pg((vm_offset_t) PADDR1); 728 } 729 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 730#endif 731 } 732 return (0); 733} 734 735/* 736 * Routine: pmap_extract 737 * Function: 738 * Extract the physical page address associated 739 * with the given map/virtual_address pair. 740 */ 741vm_offset_t 742pmap_extract(pmap, va) 743 register pmap_t pmap; 744 vm_offset_t va; 745{ 746 vm_offset_t rtval; 747 vm_offset_t pdirindex; 748 pdirindex = va >> PDRSHIFT; 749 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 750 unsigned *pte; 751 if ((rtval & PG_PS) != 0) { 752 rtval &= ~(NBPDR - 1); 753 rtval |= va & (NBPDR - 1); 754 return rtval; 755 } 756 pte = get_ptbase(pmap) + i386_btop(va); 757 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 758 return rtval; 759 } 760 return 0; 761 762} 763 764/* 765 * determine if a page is managed (memory vs. device) 766 */ 767static PMAP_INLINE int 768pmap_is_managed(pa) 769 vm_offset_t pa; 770{ 771 int i; 772 773 if (!pmap_initialized) 774 return 0; 775 776 for (i = 0; phys_avail[i + 1]; i += 2) { 777 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 778 return 1; 779 } 780 return 0; 781} 782 783 784/*************************************************** 785 * Low level mapping routines..... 786 ***************************************************/ 787 788/* 789 * add a wired page to the kva 790 * note that in order for the mapping to take effect -- you 791 * should do a invltlb after doing the pmap_kenter... 792 */ 793PMAP_INLINE void 794pmap_kenter(va, pa) 795 vm_offset_t va; 796 register vm_offset_t pa; 797{ 798 register unsigned *pte; 799 unsigned npte, opte; 800 801 npte = pa | PG_RW | PG_V | pgeflag; 802 pte = (unsigned *)vtopte(va); 803 opte = *pte; 804 *pte = npte; 805 if (opte) 806 invltlb_1pg(va); 807} 808 809/* 810 * remove a page from the kernel pagetables 811 */ 812PMAP_INLINE void 813pmap_kremove(va) 814 vm_offset_t va; 815{ 816 register unsigned *pte; 817 818 pte = (unsigned *)vtopte(va); 819 *pte = 0; 820 invltlb_1pg(va); 821} 822 823/* 824 * Add a list of wired pages to the kva 825 * this routine is only used for temporary 826 * kernel mappings that do not need to have 827 * page modification or references recorded. 828 * Note that old mappings are simply written 829 * over. The page *must* be wired. 830 */ 831void 832pmap_qenter(va, m, count) 833 vm_offset_t va; 834 vm_page_t *m; 835 int count; 836{ 837 int i; 838 839 for (i = 0; i < count; i++) { 840 vm_offset_t tva = va + i * PAGE_SIZE; 841 pmap_kenter(tva, VM_PAGE_TO_PHYS(m[i])); 842 } 843} 844 845/* 846 * this routine jerks page mappings from the 847 * kernel -- it is meant only for temporary mappings. 848 */ 849void 850pmap_qremove(va, count) 851 vm_offset_t va; 852 int count; 853{ 854 vm_offset_t end_va; 855 856 end_va = va + count*PAGE_SIZE; 857 858 while (va < end_va) { 859 unsigned *pte; 860 861 pte = (unsigned *)vtopte(va); 862 *pte = 0; 863#ifdef SMP 864 cpu_invlpg((void *)va); 865#else 866 invltlb_1pg(va); 867#endif 868 va += PAGE_SIZE; 869 } 870#ifdef SMP 871 smp_invltlb(); 872#endif 873} 874 875static vm_page_t 876pmap_page_lookup(object, pindex) 877 vm_object_t object; 878 vm_pindex_t pindex; 879{ 880 vm_page_t m; 881retry: 882 m = vm_page_lookup(object, pindex); 883 if (m && vm_page_sleep_busy(m, FALSE, "pplookp")) 884 goto retry; 885 return m; 886} 887 888/* 889 * Create the UPAGES for a new process. 890 * This routine directly affects the fork perf for a process. 891 */ 892void 893pmap_new_proc(p) 894 struct proc *p; 895{ 896 int i, updateneeded; 897 vm_object_t upobj; 898 vm_page_t m; 899 struct user *up; 900 unsigned *ptek, oldpte; 901 902 /* 903 * allocate object for the upages 904 */ 905 if ((upobj = p->p_upages_obj) == NULL) { 906 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 907 p->p_upages_obj = upobj; 908 } 909 910 /* get a kernel virtual address for the UPAGES for this proc */ 911 if ((up = p->p_addr) == NULL) { 912 up = (struct user *) kmem_alloc_pageable(kernel_map, 913 UPAGES * PAGE_SIZE); 914#if !defined(MAX_PERF) 915 if (up == NULL) 916 panic("pmap_new_proc: u_map allocation failed"); 917#endif 918 p->p_addr = up; 919 } 920 921 ptek = (unsigned *) vtopte((vm_offset_t) up); 922 923 updateneeded = 0; 924 for(i=0;i<UPAGES;i++) { 925 /* 926 * Get a kernel stack page 927 */ 928 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 929 930 /* 931 * Wire the page 932 */ 933 m->wire_count++; 934 cnt.v_wire_count++; 935 936 oldpte = *(ptek + i); 937 /* 938 * Enter the page into the kernel address space. 939 */ 940 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 941 if (oldpte) { 942 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) { 943 invlpg((vm_offset_t) up + i * PAGE_SIZE); 944 } else { 945 updateneeded = 1; 946 } 947 } 948 949 vm_page_wakeup(m); 950 vm_page_flag_clear(m, PG_ZERO); 951 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 952 m->valid = VM_PAGE_BITS_ALL; 953 } 954 if (updateneeded) 955 invltlb(); 956} 957 958/* 959 * Dispose the UPAGES for a process that has exited. 960 * This routine directly impacts the exit perf of a process. 961 */ 962void 963pmap_dispose_proc(p) 964 struct proc *p; 965{ 966 int i; 967 vm_object_t upobj; 968 vm_page_t m; 969 unsigned *ptek, oldpte; 970 971 upobj = p->p_upages_obj; 972 973 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 974 for(i=0;i<UPAGES;i++) { 975 976 if ((m = vm_page_lookup(upobj, i)) == NULL) 977 panic("pmap_dispose_proc: upage already missing???"); 978 979 vm_page_busy(m); 980 981 oldpte = *(ptek + i); 982 *(ptek + i) = 0; 983 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) 984 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 985 vm_page_unwire(m, 0); 986 vm_page_free(m); 987 } 988#if defined(I386_CPU) 989 if (cpu_class <= CPUCLASS_386) 990 invltlb(); 991#endif 992} 993 994/* 995 * Allow the UPAGES for a process to be prejudicially paged out. 996 */ 997void 998pmap_swapout_proc(p) 999 struct proc *p; 1000{ 1001 int i; 1002 vm_object_t upobj; 1003 vm_page_t m; 1004 1005 upobj = p->p_upages_obj; 1006 /* 1007 * let the upages be paged 1008 */ 1009 for(i=0;i<UPAGES;i++) { 1010 if ((m = vm_page_lookup(upobj, i)) == NULL) 1011 panic("pmap_swapout_proc: upage already missing???"); 1012 vm_page_dirty(m); 1013 vm_page_unwire(m, 0); 1014 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 1015 } 1016} 1017 1018/* 1019 * Bring the UPAGES for a specified process back in. 1020 */ 1021void 1022pmap_swapin_proc(p) 1023 struct proc *p; 1024{ 1025 int i,rv; 1026 vm_object_t upobj; 1027 vm_page_t m; 1028 1029 upobj = p->p_upages_obj; 1030 for(i=0;i<UPAGES;i++) { 1031 1032 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1033 1034 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 1035 VM_PAGE_TO_PHYS(m)); 1036 1037 if (m->valid != VM_PAGE_BITS_ALL) { 1038 rv = vm_pager_get_pages(upobj, &m, 1, 0); 1039#if !defined(MAX_PERF) 1040 if (rv != VM_PAGER_OK) 1041 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 1042#endif 1043 m = vm_page_lookup(upobj, i); 1044 m->valid = VM_PAGE_BITS_ALL; 1045 } 1046 1047 vm_page_wire(m); 1048 vm_page_wakeup(m); 1049 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 1050 } 1051} 1052 1053/*************************************************** 1054 * Page table page management routines..... 1055 ***************************************************/ 1056 1057/* 1058 * This routine unholds page table pages, and if the hold count 1059 * drops to zero, then it decrements the wire count. 1060 */ 1061static int 1062_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1063 1064 while (vm_page_sleep_busy(m, FALSE, "pmuwpt")) 1065 ; 1066 1067 if (m->hold_count == 0) { 1068 vm_offset_t pteva; 1069 /* 1070 * unmap the page table page 1071 */ 1072 pmap->pm_pdir[m->pindex] = 0; 1073 --pmap->pm_stats.resident_count; 1074 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1075 (((unsigned) PTDpde) & PG_FRAME)) { 1076 /* 1077 * Do a invltlb to make the invalidated mapping 1078 * take effect immediately. 1079 */ 1080 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1081 pmap_TLB_invalidate(pmap, pteva); 1082 } 1083 1084 if (pmap->pm_ptphint == m) 1085 pmap->pm_ptphint = NULL; 1086 1087 /* 1088 * If the page is finally unwired, simply free it. 1089 */ 1090 --m->wire_count; 1091 if (m->wire_count == 0) { 1092 1093 vm_page_flash(m); 1094 vm_page_busy(m); 1095 vm_page_free_zero(m); 1096 --cnt.v_wire_count; 1097 } 1098 return 1; 1099 } 1100 return 0; 1101} 1102 1103static PMAP_INLINE int 1104pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1105 vm_page_unhold(m); 1106 if (m->hold_count == 0) 1107 return _pmap_unwire_pte_hold(pmap, m); 1108 else 1109 return 0; 1110} 1111 1112/* 1113 * After removing a page table entry, this routine is used to 1114 * conditionally free the page, and manage the hold/wire counts. 1115 */ 1116static int 1117pmap_unuse_pt(pmap, va, mpte) 1118 pmap_t pmap; 1119 vm_offset_t va; 1120 vm_page_t mpte; 1121{ 1122 unsigned ptepindex; 1123 if (va >= UPT_MIN_ADDRESS) 1124 return 0; 1125 1126 if (mpte == NULL) { 1127 ptepindex = (va >> PDRSHIFT); 1128 if (pmap->pm_ptphint && 1129 (pmap->pm_ptphint->pindex == ptepindex)) { 1130 mpte = pmap->pm_ptphint; 1131 } else { 1132 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1133 pmap->pm_ptphint = mpte; 1134 } 1135 } 1136 1137 return pmap_unwire_pte_hold(pmap, mpte); 1138} 1139 1140void 1141pmap_pinit0(pmap) 1142 struct pmap *pmap; 1143{ 1144 pmap->pm_pdir = 1145 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1146 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1147 pmap->pm_count = 1; 1148 pmap->pm_active = 0; 1149 pmap->pm_ptphint = NULL; 1150 TAILQ_INIT(&pmap->pm_pvlist); 1151 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1152} 1153 1154/* 1155 * Initialize a preallocated and zeroed pmap structure, 1156 * such as one in a vmspace structure. 1157 */ 1158void 1159pmap_pinit(pmap) 1160 register struct pmap *pmap; 1161{ 1162 vm_page_t ptdpg; 1163 1164 /* 1165 * No need to allocate page table space yet but we do need a valid 1166 * page directory table. 1167 */ 1168 if (pmap->pm_pdir == NULL) 1169 pmap->pm_pdir = 1170 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1171 1172 /* 1173 * allocate object for the ptes 1174 */ 1175 if (pmap->pm_pteobj == NULL) 1176 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1177 1178 /* 1179 * allocate the page directory page 1180 */ 1181 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1182 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1183 1184 ptdpg->wire_count = 1; 1185 ++cnt.v_wire_count; 1186 1187 1188 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/ 1189 ptdpg->valid = VM_PAGE_BITS_ALL; 1190 1191 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1192 if ((ptdpg->flags & PG_ZERO) == 0) 1193 bzero(pmap->pm_pdir, PAGE_SIZE); 1194 1195 /* wire in kernel global address entries */ 1196 /* XXX copies current process, does not fill in MPPTDI */ 1197 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1198#ifdef SMP 1199 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI]; 1200#endif 1201 1202 /* install self-referential address mapping entry */ 1203 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1204 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1205 1206 pmap->pm_count = 1; 1207 pmap->pm_active = 0; 1208 pmap->pm_ptphint = NULL; 1209 TAILQ_INIT(&pmap->pm_pvlist); 1210 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1211} 1212 1213static int 1214pmap_release_free_page(pmap, p) 1215 struct pmap *pmap; 1216 vm_page_t p; 1217{ 1218 unsigned *pde = (unsigned *) pmap->pm_pdir; 1219 /* 1220 * This code optimizes the case of freeing non-busy 1221 * page-table pages. Those pages are zero now, and 1222 * might as well be placed directly into the zero queue. 1223 */ 1224 if (vm_page_sleep_busy(p, FALSE, "pmaprl")) 1225 return 0; 1226 1227 vm_page_busy(p); 1228 1229 /* 1230 * Remove the page table page from the processes address space. 1231 */ 1232 pde[p->pindex] = 0; 1233 pmap->pm_stats.resident_count--; 1234 1235#if !defined(MAX_PERF) 1236 if (p->hold_count) { 1237 panic("pmap_release: freeing held page table page"); 1238 } 1239#endif 1240 /* 1241 * Page directory pages need to have the kernel 1242 * stuff cleared, so they can go into the zero queue also. 1243 */ 1244 if (p->pindex == PTDPTDI) { 1245 bzero(pde + KPTDI, nkpt * PTESIZE); 1246#ifdef SMP 1247 pde[MPPTDI] = 0; 1248#endif 1249 pde[APTDPTDI] = 0; 1250 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1251 } 1252 1253 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1254 pmap->pm_ptphint = NULL; 1255 1256 p->wire_count--; 1257 cnt.v_wire_count--; 1258 vm_page_free_zero(p); 1259 return 1; 1260} 1261 1262/* 1263 * this routine is called if the page table page is not 1264 * mapped correctly. 1265 */ 1266static vm_page_t 1267_pmap_allocpte(pmap, ptepindex) 1268 pmap_t pmap; 1269 unsigned ptepindex; 1270{ 1271 vm_offset_t pteva, ptepa; 1272 vm_page_t m; 1273 1274 /* 1275 * Find or fabricate a new pagetable page 1276 */ 1277 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1278 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1279 1280 if (m->queue != PQ_NONE) { 1281 int s = splvm(); 1282 vm_page_unqueue(m); 1283 splx(s); 1284 } 1285 1286 if (m->wire_count == 0) 1287 cnt.v_wire_count++; 1288 m->wire_count++; 1289 1290 /* 1291 * Increment the hold count for the page table page 1292 * (denoting a new mapping.) 1293 */ 1294 m->hold_count++; 1295 1296 /* 1297 * Map the pagetable page into the process address space, if 1298 * it isn't already there. 1299 */ 1300 1301 pmap->pm_stats.resident_count++; 1302 1303 ptepa = VM_PAGE_TO_PHYS(m); 1304 pmap->pm_pdir[ptepindex] = 1305 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1306 1307 /* 1308 * Set the page table hint 1309 */ 1310 pmap->pm_ptphint = m; 1311 1312 /* 1313 * Try to use the new mapping, but if we cannot, then 1314 * do it with the routine that maps the page explicitly. 1315 */ 1316 if ((m->flags & PG_ZERO) == 0) { 1317 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1318 (((unsigned) PTDpde) & PG_FRAME)) { 1319 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1320 bzero((caddr_t) pteva, PAGE_SIZE); 1321 } else { 1322 pmap_zero_page(ptepa); 1323 } 1324 } 1325 1326 m->valid = VM_PAGE_BITS_ALL; 1327 vm_page_flag_clear(m, PG_ZERO); 1328 vm_page_flag_set(m, PG_MAPPED); 1329 vm_page_wakeup(m); 1330 1331 return m; 1332} 1333 1334static vm_page_t 1335pmap_allocpte(pmap, va) 1336 pmap_t pmap; 1337 vm_offset_t va; 1338{ 1339 unsigned ptepindex; 1340 vm_offset_t ptepa; 1341 vm_page_t m; 1342 1343 /* 1344 * Calculate pagetable page index 1345 */ 1346 ptepindex = va >> PDRSHIFT; 1347 1348 /* 1349 * Get the page directory entry 1350 */ 1351 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1352 1353 /* 1354 * This supports switching from a 4MB page to a 1355 * normal 4K page. 1356 */ 1357 if (ptepa & PG_PS) { 1358 pmap->pm_pdir[ptepindex] = 0; 1359 ptepa = 0; 1360 invltlb(); 1361 } 1362 1363 /* 1364 * If the page table page is mapped, we just increment the 1365 * hold count, and activate it. 1366 */ 1367 if (ptepa) { 1368 /* 1369 * In order to get the page table page, try the 1370 * hint first. 1371 */ 1372 if (pmap->pm_ptphint && 1373 (pmap->pm_ptphint->pindex == ptepindex)) { 1374 m = pmap->pm_ptphint; 1375 } else { 1376 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1377 pmap->pm_ptphint = m; 1378 } 1379 m->hold_count++; 1380 return m; 1381 } 1382 /* 1383 * Here if the pte page isn't mapped, or if it has been deallocated. 1384 */ 1385 return _pmap_allocpte(pmap, ptepindex); 1386} 1387 1388 1389/*************************************************** 1390* Pmap allocation/deallocation routines. 1391 ***************************************************/ 1392 1393/* 1394 * Release any resources held by the given physical map. 1395 * Called when a pmap initialized by pmap_pinit is being released. 1396 * Should only be called if the map contains no valid mappings. 1397 */ 1398void 1399pmap_release(pmap) 1400 register struct pmap *pmap; 1401{ 1402 vm_page_t p,n,ptdpg; 1403 vm_object_t object = pmap->pm_pteobj; 1404 int curgeneration; 1405 1406#if defined(DIAGNOSTIC) 1407 if (object->ref_count != 1) 1408 panic("pmap_release: pteobj reference count != 1"); 1409#endif 1410 1411 ptdpg = NULL; 1412retry: 1413 curgeneration = object->generation; 1414 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1415 n = TAILQ_NEXT(p, listq); 1416 if (p->pindex == PTDPTDI) { 1417 ptdpg = p; 1418 continue; 1419 } 1420 while (1) { 1421 if (!pmap_release_free_page(pmap, p) && 1422 (object->generation != curgeneration)) 1423 goto retry; 1424 } 1425 } 1426 1427 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1428 goto retry; 1429} 1430 1431/* 1432 * grow the number of kernel page table entries, if needed 1433 */ 1434void 1435pmap_growkernel(vm_offset_t addr) 1436{ 1437 struct proc *p; 1438 struct pmap *pmap; 1439 int s; 1440 vm_offset_t ptppaddr; 1441 vm_page_t nkpg; 1442 pd_entry_t newpdir; 1443 1444 s = splhigh(); 1445 if (kernel_vm_end == 0) { 1446 kernel_vm_end = KERNBASE; 1447 nkpt = 0; 1448 while (pdir_pde(PTD, kernel_vm_end)) { 1449 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1450 nkpt++; 1451 } 1452 } 1453 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1454 while (kernel_vm_end < addr) { 1455 if (pdir_pde(PTD, kernel_vm_end)) { 1456 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1457 continue; 1458 } 1459 1460 /* 1461 * This index is bogus, but out of the way 1462 */ 1463 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1464#if !defined(MAX_PERF) 1465 if (!nkpg) 1466 panic("pmap_growkernel: no memory to grow kernel"); 1467#endif 1468 1469 nkpt++; 1470 1471 vm_page_wire(nkpg); 1472 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1473 pmap_zero_page(ptppaddr); 1474 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1475 pdir_pde(PTD, kernel_vm_end) = newpdir; 1476 1477 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1478 if (p->p_vmspace) { 1479 pmap = vmspace_pmap(p->p_vmspace); 1480 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1481 } 1482 } 1483 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir; 1484 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1485 } 1486 splx(s); 1487} 1488 1489/* 1490 * Retire the given physical map from service. 1491 * Should only be called if the map contains 1492 * no valid mappings. 1493 */ 1494void 1495pmap_destroy(pmap) 1496 register pmap_t pmap; 1497{ 1498 int count; 1499 1500 if (pmap == NULL) 1501 return; 1502 1503 count = --pmap->pm_count; 1504 if (count == 0) { 1505 pmap_release(pmap); 1506#if !defined(MAX_PERF) 1507 panic("destroying a pmap is not yet implemented"); 1508#endif 1509 } 1510} 1511 1512/* 1513 * Add a reference to the specified pmap. 1514 */ 1515void 1516pmap_reference(pmap) 1517 pmap_t pmap; 1518{ 1519 if (pmap != NULL) { 1520 pmap->pm_count++; 1521 } 1522} 1523 1524/*************************************************** 1525* page management routines. 1526 ***************************************************/ 1527 1528/* 1529 * free the pv_entry back to the free list 1530 */ 1531static PMAP_INLINE void 1532free_pv_entry(pv) 1533 pv_entry_t pv; 1534{ 1535 pv_entry_count--; 1536 zfreei(pvzone, pv); 1537} 1538 1539/* 1540 * get a new pv_entry, allocating a block from the system 1541 * when needed. 1542 * the memory allocation is performed bypassing the malloc code 1543 * because of the possibility of allocations at interrupt time. 1544 */ 1545static pv_entry_t 1546get_pv_entry(void) 1547{ 1548 pv_entry_count++; 1549 if (pv_entry_high_water && 1550 (pv_entry_count > pv_entry_high_water) && 1551 (pmap_pagedaemon_waken == 0)) { 1552 pmap_pagedaemon_waken = 1; 1553 wakeup (&vm_pages_needed); 1554 } 1555 return zalloci(pvzone); 1556} 1557 1558/* 1559 * This routine is very drastic, but can save the system 1560 * in a pinch. 1561 */ 1562void 1563pmap_collect() { 1564 pv_table_t *ppv; 1565 int i; 1566 vm_offset_t pa; 1567 vm_page_t m; 1568 static int warningdone=0; 1569 1570 if (pmap_pagedaemon_waken == 0) 1571 return; 1572 1573 if (warningdone < 5) { 1574 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1575 warningdone++; 1576 } 1577 1578 for(i = 0; i < pv_npg; i++) { 1579 if ((ppv = &pv_table[i]) == 0) 1580 continue; 1581 m = ppv->pv_vm_page; 1582 if ((pa = VM_PAGE_TO_PHYS(m)) == 0) 1583 continue; 1584 if (m->wire_count || m->hold_count || m->busy || 1585 (m->flags & PG_BUSY)) 1586 continue; 1587 pmap_remove_all(pa); 1588 } 1589 pmap_pagedaemon_waken = 0; 1590} 1591 1592 1593/* 1594 * If it is the first entry on the list, it is actually 1595 * in the header and we must copy the following entry up 1596 * to the header. Otherwise we must search the list for 1597 * the entry. In either case we free the now unused entry. 1598 */ 1599 1600static int 1601pmap_remove_entry(pmap, ppv, va) 1602 struct pmap *pmap; 1603 pv_table_t *ppv; 1604 vm_offset_t va; 1605{ 1606 pv_entry_t pv; 1607 int rtval; 1608 int s; 1609 1610 s = splvm(); 1611 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1612 for (pv = TAILQ_FIRST(&ppv->pv_list); 1613 pv; 1614 pv = TAILQ_NEXT(pv, pv_list)) { 1615 if (pmap == pv->pv_pmap && va == pv->pv_va) 1616 break; 1617 } 1618 } else { 1619 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1620 pv; 1621 pv = TAILQ_NEXT(pv, pv_plist)) { 1622 if (va == pv->pv_va) 1623 break; 1624 } 1625 } 1626 1627 rtval = 0; 1628 if (pv) { 1629 1630 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1631 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1632 ppv->pv_list_count--; 1633 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 1634 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 1635 1636 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1637 free_pv_entry(pv); 1638 } 1639 1640 splx(s); 1641 return rtval; 1642} 1643 1644/* 1645 * Create a pv entry for page at pa for 1646 * (pmap, va). 1647 */ 1648static void 1649pmap_insert_entry(pmap, va, mpte, pa) 1650 pmap_t pmap; 1651 vm_offset_t va; 1652 vm_page_t mpte; 1653 vm_offset_t pa; 1654{ 1655 1656 int s; 1657 pv_entry_t pv; 1658 pv_table_t *ppv; 1659 1660 s = splvm(); 1661 pv = get_pv_entry(); 1662 pv->pv_va = va; 1663 pv->pv_pmap = pmap; 1664 pv->pv_ptem = mpte; 1665 1666 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1667 1668 ppv = pa_to_pvh(pa); 1669 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1670 ppv->pv_list_count++; 1671 1672 splx(s); 1673} 1674 1675/* 1676 * pmap_remove_pte: do the things to unmap a page in a process 1677 */ 1678static int 1679pmap_remove_pte(pmap, ptq, va) 1680 struct pmap *pmap; 1681 unsigned *ptq; 1682 vm_offset_t va; 1683{ 1684 unsigned oldpte; 1685 pv_table_t *ppv; 1686 1687 oldpte = loadandclear(ptq); 1688 if (oldpte & PG_W) 1689 pmap->pm_stats.wired_count -= 1; 1690 /* 1691 * Machines that don't support invlpg, also don't support 1692 * PG_G. 1693 */ 1694 if (oldpte & PG_G) 1695 invlpg(va); 1696 pmap->pm_stats.resident_count -= 1; 1697 if (oldpte & PG_MANAGED) { 1698 ppv = pa_to_pvh(oldpte); 1699 if (oldpte & PG_M) { 1700#if defined(PMAP_DIAGNOSTIC) 1701 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1702 printf( 1703 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n", 1704 va, oldpte); 1705 } 1706#endif 1707 if (pmap_track_modified(va)) 1708 vm_page_dirty(ppv->pv_vm_page); 1709 } 1710 if (oldpte & PG_A) 1711 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1712 return pmap_remove_entry(pmap, ppv, va); 1713 } else { 1714 return pmap_unuse_pt(pmap, va, NULL); 1715 } 1716 1717 return 0; 1718} 1719 1720/* 1721 * Remove a single page from a process address space 1722 */ 1723static void 1724pmap_remove_page(pmap, va) 1725 struct pmap *pmap; 1726 register vm_offset_t va; 1727{ 1728 register unsigned *ptq; 1729 1730 /* 1731 * if there is no pte for this address, just skip it!!! 1732 */ 1733 if (*pmap_pde(pmap, va) == 0) { 1734 return; 1735 } 1736 1737 /* 1738 * get a local va for mappings for this pmap. 1739 */ 1740 ptq = get_ptbase(pmap) + i386_btop(va); 1741 if (*ptq) { 1742 (void) pmap_remove_pte(pmap, ptq, va); 1743 pmap_TLB_invalidate(pmap, va); 1744 } 1745 return; 1746} 1747 1748/* 1749 * Remove the given range of addresses from the specified map. 1750 * 1751 * It is assumed that the start and end are properly 1752 * rounded to the page size. 1753 */ 1754void 1755pmap_remove(pmap, sva, eva) 1756 struct pmap *pmap; 1757 register vm_offset_t sva; 1758 register vm_offset_t eva; 1759{ 1760 register unsigned *ptbase; 1761 vm_offset_t pdnxt; 1762 vm_offset_t ptpaddr; 1763 vm_offset_t sindex, eindex; 1764 int anyvalid; 1765 1766 if (pmap == NULL) 1767 return; 1768 1769 if (pmap->pm_stats.resident_count == 0) 1770 return; 1771 1772 /* 1773 * special handling of removing one page. a very 1774 * common operation and easy to short circuit some 1775 * code. 1776 */ 1777 if (((sva + PAGE_SIZE) == eva) && 1778 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1779 pmap_remove_page(pmap, sva); 1780 return; 1781 } 1782 1783 anyvalid = 0; 1784 1785 /* 1786 * Get a local virtual address for the mappings that are being 1787 * worked with. 1788 */ 1789 ptbase = get_ptbase(pmap); 1790 1791 sindex = i386_btop(sva); 1792 eindex = i386_btop(eva); 1793 1794 for (; sindex < eindex; sindex = pdnxt) { 1795 unsigned pdirindex; 1796 1797 /* 1798 * Calculate index for next page table. 1799 */ 1800 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1801 if (pmap->pm_stats.resident_count == 0) 1802 break; 1803 1804 pdirindex = sindex / NPDEPG; 1805 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1806 pmap->pm_pdir[pdirindex] = 0; 1807 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1808 anyvalid++; 1809 continue; 1810 } 1811 1812 /* 1813 * Weed out invalid mappings. Note: we assume that the page 1814 * directory table is always allocated, and in kernel virtual. 1815 */ 1816 if (ptpaddr == 0) 1817 continue; 1818 1819 /* 1820 * Limit our scan to either the end of the va represented 1821 * by the current page table page, or to the end of the 1822 * range being removed. 1823 */ 1824 if (pdnxt > eindex) { 1825 pdnxt = eindex; 1826 } 1827 1828 for ( ;sindex != pdnxt; sindex++) { 1829 vm_offset_t va; 1830 if (ptbase[sindex] == 0) { 1831 continue; 1832 } 1833 va = i386_ptob(sindex); 1834 1835 anyvalid++; 1836 if (pmap_remove_pte(pmap, 1837 ptbase + sindex, va)) 1838 break; 1839 } 1840 } 1841 1842 if (anyvalid) 1843 pmap_TLB_invalidate_all(pmap); 1844} 1845 1846/* 1847 * Routine: pmap_remove_all 1848 * Function: 1849 * Removes this physical page from 1850 * all physical maps in which it resides. 1851 * Reflects back modify bits to the pager. 1852 * 1853 * Notes: 1854 * Original versions of this routine were very 1855 * inefficient because they iteratively called 1856 * pmap_remove (slow...) 1857 */ 1858 1859static void 1860pmap_remove_all(pa) 1861 vm_offset_t pa; 1862{ 1863 register pv_entry_t pv; 1864 pv_table_t *ppv; 1865 register unsigned *pte, tpte; 1866 int s; 1867 1868#if defined(PMAP_DIAGNOSTIC) 1869 /* 1870 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1871 * pages! 1872 */ 1873 if (!pmap_is_managed(pa)) { 1874 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", pa); 1875 } 1876#endif 1877 1878 s = splvm(); 1879 ppv = pa_to_pvh(pa); 1880 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1881 pv->pv_pmap->pm_stats.resident_count--; 1882 1883 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1884 1885 tpte = loadandclear(pte); 1886 if (tpte & PG_W) 1887 pv->pv_pmap->pm_stats.wired_count--; 1888 1889 if (tpte & PG_A) 1890 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1891 1892 /* 1893 * Update the vm_page_t clean and reference bits. 1894 */ 1895 if (tpte & PG_M) { 1896#if defined(PMAP_DIAGNOSTIC) 1897 if (pmap_nw_modified((pt_entry_t) tpte)) { 1898 printf( 1899 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n", 1900 pv->pv_va, tpte); 1901 } 1902#endif 1903 if (pmap_track_modified(pv->pv_va)) 1904 vm_page_dirty(ppv->pv_vm_page); 1905 } 1906 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 1907 1908 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1909 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1910 ppv->pv_list_count--; 1911 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1912 free_pv_entry(pv); 1913 } 1914 1915 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 1916 1917 splx(s); 1918} 1919 1920/* 1921 * Set the physical protection on the 1922 * specified range of this map as requested. 1923 */ 1924void 1925pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1926{ 1927 register unsigned *ptbase; 1928 vm_offset_t pdnxt, ptpaddr; 1929 vm_pindex_t sindex, eindex; 1930 int anychanged; 1931 1932 1933 if (pmap == NULL) 1934 return; 1935 1936 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1937 pmap_remove(pmap, sva, eva); 1938 return; 1939 } 1940 1941 if (prot & VM_PROT_WRITE) 1942 return; 1943 1944 anychanged = 0; 1945 1946 ptbase = get_ptbase(pmap); 1947 1948 sindex = i386_btop(sva); 1949 eindex = i386_btop(eva); 1950 1951 for (; sindex < eindex; sindex = pdnxt) { 1952 1953 unsigned pdirindex; 1954 1955 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1956 1957 pdirindex = sindex / NPDEPG; 1958 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1959 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1960 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1961 anychanged++; 1962 continue; 1963 } 1964 1965 /* 1966 * Weed out invalid mappings. Note: we assume that the page 1967 * directory table is always allocated, and in kernel virtual. 1968 */ 1969 if (ptpaddr == 0) 1970 continue; 1971 1972 if (pdnxt > eindex) { 1973 pdnxt = eindex; 1974 } 1975 1976 for (; sindex != pdnxt; sindex++) { 1977 1978 unsigned pbits; 1979 pv_table_t *ppv; 1980 1981 pbits = ptbase[sindex]; 1982 1983 if (pbits & PG_MANAGED) { 1984 ppv = NULL; 1985 if (pbits & PG_A) { 1986 ppv = pa_to_pvh(pbits); 1987 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1988 pbits &= ~PG_A; 1989 } 1990 if (pbits & PG_M) { 1991 if (pmap_track_modified(i386_ptob(sindex))) { 1992 if (ppv == NULL) 1993 ppv = pa_to_pvh(pbits); 1994 vm_page_dirty(ppv->pv_vm_page); 1995 pbits &= ~PG_M; 1996 } 1997 } 1998 } 1999 2000 pbits &= ~PG_RW; 2001 2002 if (pbits != ptbase[sindex]) { 2003 ptbase[sindex] = pbits; 2004 anychanged = 1; 2005 } 2006 } 2007 } 2008 if (anychanged) 2009 pmap_TLB_invalidate_all(pmap); 2010} 2011 2012/* 2013 * Insert the given physical page (p) at 2014 * the specified virtual address (v) in the 2015 * target physical map with the protection requested. 2016 * 2017 * If specified, the page will be wired down, meaning 2018 * that the related pte can not be reclaimed. 2019 * 2020 * NB: This is the only routine which MAY NOT lazy-evaluate 2021 * or lose information. That is, this routine must actually 2022 * insert this page into the given map NOW. 2023 */ 2024void 2025pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot, 2026 boolean_t wired) 2027{ 2028 register unsigned *pte; 2029 vm_offset_t opa; 2030 vm_offset_t origpte, newpte; 2031 vm_page_t mpte; 2032 2033 if (pmap == NULL) 2034 return; 2035 2036 va &= PG_FRAME; 2037#ifdef PMAP_DIAGNOSTIC 2038 if (va > VM_MAX_KERNEL_ADDRESS) 2039 panic("pmap_enter: toobig"); 2040 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 2041 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 2042#endif 2043 2044 mpte = NULL; 2045 /* 2046 * In the case that a page table page is not 2047 * resident, we are creating it here. 2048 */ 2049 if (va < UPT_MIN_ADDRESS) { 2050 mpte = pmap_allocpte(pmap, va); 2051 } 2052#if 0 && defined(PMAP_DIAGNOSTIC) 2053 else { 2054 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 2055 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 2056 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 2057 pmap->pm_pdir[PTDPTDI], origpte, va); 2058 } 2059 if (smp_active) { 2060 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid]; 2061 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2062 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2063 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2064 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr); 2065 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2066 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2067 } 2068 } 2069 } 2070#endif 2071 2072 pte = pmap_pte(pmap, va); 2073 2074#if !defined(MAX_PERF) 2075 /* 2076 * Page Directory table entry not valid, we need a new PT page 2077 */ 2078 if (pte == NULL) { 2079 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n", 2080 (void *)pmap->pm_pdir[PTDPTDI], va); 2081 } 2082#endif 2083 2084 origpte = *(vm_offset_t *)pte; 2085 pa &= PG_FRAME; 2086 opa = origpte & PG_FRAME; 2087 2088#if !defined(MAX_PERF) 2089 if (origpte & PG_PS) 2090 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2091#endif 2092 2093 /* 2094 * Mapping has not changed, must be protection or wiring change. 2095 */ 2096 if (origpte && (opa == pa)) { 2097 /* 2098 * Wiring change, just update stats. We don't worry about 2099 * wiring PT pages as they remain resident as long as there 2100 * are valid mappings in them. Hence, if a user page is wired, 2101 * the PT page will be also. 2102 */ 2103 if (wired && ((origpte & PG_W) == 0)) 2104 pmap->pm_stats.wired_count++; 2105 else if (!wired && (origpte & PG_W)) 2106 pmap->pm_stats.wired_count--; 2107 2108#if defined(PMAP_DIAGNOSTIC) 2109 if (pmap_nw_modified((pt_entry_t) origpte)) { 2110 printf( 2111 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n", 2112 va, origpte); 2113 } 2114#endif 2115 2116 /* 2117 * Remove extra pte reference 2118 */ 2119 if (mpte) 2120 mpte->hold_count--; 2121 2122 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2123 if ((origpte & PG_RW) == 0) { 2124 *pte |= PG_RW; 2125#ifdef SMP 2126 cpu_invlpg((void *)va); 2127 if (pmap->pm_active & other_cpus) 2128 smp_invltlb(); 2129#else 2130 invltlb_1pg(va); 2131#endif 2132 } 2133 return; 2134 } 2135 2136 /* 2137 * We might be turning off write access to the page, 2138 * so we go ahead and sense modify status. 2139 */ 2140 if (origpte & PG_MANAGED) { 2141 if ((origpte & PG_M) && pmap_track_modified(va)) { 2142 pv_table_t *ppv; 2143 ppv = pa_to_pvh(opa); 2144 vm_page_dirty(ppv->pv_vm_page); 2145 } 2146 pa |= PG_MANAGED; 2147 } 2148 goto validate; 2149 } 2150 /* 2151 * Mapping has changed, invalidate old range and fall through to 2152 * handle validating new mapping. 2153 */ 2154 if (opa) { 2155 int err; 2156 err = pmap_remove_pte(pmap, pte, va); 2157#if !defined(MAX_PERF) 2158 if (err) 2159 panic("pmap_enter: pte vanished, va: 0x%x", va); 2160#endif 2161 } 2162 2163 /* 2164 * Enter on the PV list if part of our managed memory Note that we 2165 * raise IPL while manipulating pv_table since pmap_enter can be 2166 * called at interrupt time. 2167 */ 2168 if (pmap_is_managed(pa)) { 2169 pmap_insert_entry(pmap, va, mpte, pa); 2170 pa |= PG_MANAGED; 2171 } 2172 2173 /* 2174 * Increment counters 2175 */ 2176 pmap->pm_stats.resident_count++; 2177 if (wired) 2178 pmap->pm_stats.wired_count++; 2179 2180validate: 2181 /* 2182 * Now validate mapping with desired protection/wiring. 2183 */ 2184 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2185 2186 if (wired) 2187 newpte |= PG_W; 2188 if (va < UPT_MIN_ADDRESS) 2189 newpte |= PG_U; 2190 if (pmap == kernel_pmap) 2191 newpte |= pgeflag; 2192 2193 /* 2194 * if the mapping or permission bits are different, we need 2195 * to update the pte. 2196 */ 2197 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2198 *pte = newpte | PG_A; 2199 if (origpte) { 2200#ifdef SMP 2201 cpu_invlpg((void *)va); 2202 if (pmap->pm_active & other_cpus) 2203 smp_invltlb(); 2204#else 2205 invltlb_1pg(va); 2206#endif 2207 } 2208 } 2209} 2210 2211/* 2212 * this code makes some *MAJOR* assumptions: 2213 * 1. Current pmap & pmap exists. 2214 * 2. Not wired. 2215 * 3. Read access. 2216 * 4. No page table pages. 2217 * 5. Tlbflush is deferred to calling procedure. 2218 * 6. Page IS managed. 2219 * but is *MUCH* faster than pmap_enter... 2220 */ 2221 2222static vm_page_t 2223pmap_enter_quick(pmap, va, pa, mpte) 2224 register pmap_t pmap; 2225 vm_offset_t va; 2226 register vm_offset_t pa; 2227 vm_page_t mpte; 2228{ 2229 register unsigned *pte; 2230 2231 /* 2232 * In the case that a page table page is not 2233 * resident, we are creating it here. 2234 */ 2235 if (va < UPT_MIN_ADDRESS) { 2236 unsigned ptepindex; 2237 vm_offset_t ptepa; 2238 2239 /* 2240 * Calculate pagetable page index 2241 */ 2242 ptepindex = va >> PDRSHIFT; 2243 if (mpte && (mpte->pindex == ptepindex)) { 2244 mpte->hold_count++; 2245 } else { 2246retry: 2247 /* 2248 * Get the page directory entry 2249 */ 2250 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2251 2252 /* 2253 * If the page table page is mapped, we just increment 2254 * the hold count, and activate it. 2255 */ 2256 if (ptepa) { 2257#if !defined(MAX_PERF) 2258 if (ptepa & PG_PS) 2259 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2260#endif 2261 if (pmap->pm_ptphint && 2262 (pmap->pm_ptphint->pindex == ptepindex)) { 2263 mpte = pmap->pm_ptphint; 2264 } else { 2265 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2266 pmap->pm_ptphint = mpte; 2267 } 2268 if (mpte == NULL) 2269 goto retry; 2270 mpte->hold_count++; 2271 } else { 2272 mpte = _pmap_allocpte(pmap, ptepindex); 2273 } 2274 } 2275 } else { 2276 mpte = NULL; 2277 } 2278 2279 /* 2280 * This call to vtopte makes the assumption that we are 2281 * entering the page into the current pmap. In order to support 2282 * quick entry into any pmap, one would likely use pmap_pte_quick. 2283 * But that isn't as quick as vtopte. 2284 */ 2285 pte = (unsigned *)vtopte(va); 2286 if (*pte) { 2287 if (mpte) 2288 pmap_unwire_pte_hold(pmap, mpte); 2289 return 0; 2290 } 2291 2292 /* 2293 * Enter on the PV list if part of our managed memory Note that we 2294 * raise IPL while manipulating pv_table since pmap_enter can be 2295 * called at interrupt time. 2296 */ 2297 pmap_insert_entry(pmap, va, mpte, pa); 2298 2299 /* 2300 * Increment counters 2301 */ 2302 pmap->pm_stats.resident_count++; 2303 2304 /* 2305 * Now validate mapping with RO protection 2306 */ 2307 *pte = pa | PG_V | PG_U | PG_MANAGED; 2308 2309 return mpte; 2310} 2311 2312#define MAX_INIT_PT (96) 2313/* 2314 * pmap_object_init_pt preloads the ptes for a given object 2315 * into the specified pmap. This eliminates the blast of soft 2316 * faults on process startup and immediately after an mmap. 2317 */ 2318void 2319pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2320 pmap_t pmap; 2321 vm_offset_t addr; 2322 vm_object_t object; 2323 vm_pindex_t pindex; 2324 vm_size_t size; 2325 int limit; 2326{ 2327 vm_offset_t tmpidx; 2328 int psize; 2329 vm_page_t p, mpte; 2330 int objpgs; 2331 2332 if (!pmap) 2333 return; 2334 2335 /* 2336 * This code maps large physical mmap regions into the 2337 * processor address space. Note that some shortcuts 2338 * are taken, but the code works. 2339 */ 2340 if (pseflag && 2341 (object->type == OBJT_DEVICE) && 2342 ((addr & (NBPDR - 1)) == 0) && 2343 ((size & (NBPDR - 1)) == 0) ) { 2344 int i; 2345 vm_page_t m[1]; 2346 unsigned int ptepindex; 2347 int npdes; 2348 vm_offset_t ptepa; 2349 2350 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2351 return; 2352 2353retry: 2354 p = vm_page_lookup(object, pindex); 2355 if (p && vm_page_sleep_busy(p, FALSE, "init4p")) 2356 goto retry; 2357 2358 if (p == NULL) { 2359 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2360 if (p == NULL) 2361 return; 2362 m[0] = p; 2363 2364 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2365 vm_page_free(p); 2366 return; 2367 } 2368 2369 p = vm_page_lookup(object, pindex); 2370 vm_page_wakeup(p); 2371 } 2372 2373 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2374 if (ptepa & (NBPDR - 1)) { 2375 return; 2376 } 2377 2378 p->valid = VM_PAGE_BITS_ALL; 2379 2380 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2381 npdes = size >> PDRSHIFT; 2382 for(i=0;i<npdes;i++) { 2383 pmap->pm_pdir[ptepindex] = 2384 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2385 ptepa += NBPDR; 2386 ptepindex += 1; 2387 } 2388 vm_page_flag_set(p, PG_MAPPED); 2389 invltlb(); 2390 return; 2391 } 2392 2393 psize = i386_btop(size); 2394 2395 if ((object->type != OBJT_VNODE) || 2396 (limit && (psize > MAX_INIT_PT) && 2397 (object->resident_page_count > MAX_INIT_PT))) { 2398 return; 2399 } 2400 2401 if (psize + pindex > object->size) 2402 psize = object->size - pindex; 2403 2404 mpte = NULL; 2405 /* 2406 * if we are processing a major portion of the object, then scan the 2407 * entire thing. 2408 */ 2409 if (psize > (object->size >> 2)) { 2410 objpgs = psize; 2411 2412 for (p = TAILQ_FIRST(&object->memq); 2413 ((objpgs > 0) && (p != NULL)); 2414 p = TAILQ_NEXT(p, listq)) { 2415 2416 tmpidx = p->pindex; 2417 if (tmpidx < pindex) { 2418 continue; 2419 } 2420 tmpidx -= pindex; 2421 if (tmpidx >= psize) { 2422 continue; 2423 } 2424 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2425 (p->busy == 0) && 2426 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2427 if ((p->queue - p->pc) == PQ_CACHE) 2428 vm_page_deactivate(p); 2429 vm_page_busy(p); 2430 mpte = pmap_enter_quick(pmap, 2431 addr + i386_ptob(tmpidx), 2432 VM_PAGE_TO_PHYS(p), mpte); 2433 vm_page_flag_set(p, PG_MAPPED); 2434 vm_page_wakeup(p); 2435 } 2436 objpgs -= 1; 2437 } 2438 } else { 2439 /* 2440 * else lookup the pages one-by-one. 2441 */ 2442 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2443 p = vm_page_lookup(object, tmpidx + pindex); 2444 if (p && 2445 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2446 (p->busy == 0) && 2447 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2448 if ((p->queue - p->pc) == PQ_CACHE) 2449 vm_page_deactivate(p); 2450 vm_page_busy(p); 2451 mpte = pmap_enter_quick(pmap, 2452 addr + i386_ptob(tmpidx), 2453 VM_PAGE_TO_PHYS(p), mpte); 2454 vm_page_flag_set(p, PG_MAPPED); 2455 vm_page_wakeup(p); 2456 } 2457 } 2458 } 2459 return; 2460} 2461 2462/* 2463 * pmap_prefault provides a quick way of clustering 2464 * pagefaults into a processes address space. It is a "cousin" 2465 * of pmap_object_init_pt, except it runs at page fault time instead 2466 * of mmap time. 2467 */ 2468#define PFBAK 4 2469#define PFFOR 4 2470#define PAGEORDER_SIZE (PFBAK+PFFOR) 2471 2472static int pmap_prefault_pageorder[] = { 2473 -PAGE_SIZE, PAGE_SIZE, 2474 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2475 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2476 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2477}; 2478 2479void 2480pmap_prefault(pmap, addra, entry) 2481 pmap_t pmap; 2482 vm_offset_t addra; 2483 vm_map_entry_t entry; 2484{ 2485 int i; 2486 vm_offset_t starta; 2487 vm_offset_t addr; 2488 vm_pindex_t pindex; 2489 vm_page_t m, mpte; 2490 vm_object_t object; 2491 2492 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) 2493 return; 2494 2495 object = entry->object.vm_object; 2496 2497 starta = addra - PFBAK * PAGE_SIZE; 2498 if (starta < entry->start) { 2499 starta = entry->start; 2500 } else if (starta > addra) { 2501 starta = 0; 2502 } 2503 2504 mpte = NULL; 2505 for (i = 0; i < PAGEORDER_SIZE; i++) { 2506 vm_object_t lobject; 2507 unsigned *pte; 2508 2509 addr = addra + pmap_prefault_pageorder[i]; 2510 if (addr > addra + (PFFOR * PAGE_SIZE)) 2511 addr = 0; 2512 2513 if (addr < starta || addr >= entry->end) 2514 continue; 2515 2516 if ((*pmap_pde(pmap, addr)) == NULL) 2517 continue; 2518 2519 pte = (unsigned *) vtopte(addr); 2520 if (*pte) 2521 continue; 2522 2523 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2524 lobject = object; 2525 for (m = vm_page_lookup(lobject, pindex); 2526 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2527 lobject = lobject->backing_object) { 2528 if (lobject->backing_object_offset & PAGE_MASK) 2529 break; 2530 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2531 m = vm_page_lookup(lobject->backing_object, pindex); 2532 } 2533 2534 /* 2535 * give-up when a page is not in memory 2536 */ 2537 if (m == NULL) 2538 break; 2539 2540 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2541 (m->busy == 0) && 2542 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2543 2544 if ((m->queue - m->pc) == PQ_CACHE) { 2545 vm_page_deactivate(m); 2546 } 2547 vm_page_busy(m); 2548 mpte = pmap_enter_quick(pmap, addr, 2549 VM_PAGE_TO_PHYS(m), mpte); 2550 vm_page_flag_set(m, PG_MAPPED); 2551 vm_page_wakeup(m); 2552 } 2553 } 2554} 2555 2556/* 2557 * Routine: pmap_change_wiring 2558 * Function: Change the wiring attribute for a map/virtual-address 2559 * pair. 2560 * In/out conditions: 2561 * The mapping must already exist in the pmap. 2562 */ 2563void 2564pmap_change_wiring(pmap, va, wired) 2565 register pmap_t pmap; 2566 vm_offset_t va; 2567 boolean_t wired; 2568{ 2569 register unsigned *pte; 2570 2571 if (pmap == NULL) 2572 return; 2573 2574 pte = pmap_pte(pmap, va); 2575 2576 if (wired && !pmap_pte_w(pte)) 2577 pmap->pm_stats.wired_count++; 2578 else if (!wired && pmap_pte_w(pte)) 2579 pmap->pm_stats.wired_count--; 2580 2581 /* 2582 * Wiring is not a hardware characteristic so there is no need to 2583 * invalidate TLB. 2584 */ 2585 pmap_pte_set_w(pte, wired); 2586} 2587 2588 2589 2590/* 2591 * Copy the range specified by src_addr/len 2592 * from the source map to the range dst_addr/len 2593 * in the destination map. 2594 * 2595 * This routine is only advisory and need not do anything. 2596 */ 2597 2598void 2599pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2600 pmap_t dst_pmap, src_pmap; 2601 vm_offset_t dst_addr; 2602 vm_size_t len; 2603 vm_offset_t src_addr; 2604{ 2605 vm_offset_t addr; 2606 vm_offset_t end_addr = src_addr + len; 2607 vm_offset_t pdnxt; 2608 unsigned src_frame, dst_frame; 2609 2610 if (dst_addr != src_addr) 2611 return; 2612 2613 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2614 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2615 return; 2616 } 2617 2618 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2619 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2620 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2621#if defined(SMP) 2622 /* The page directory is not shared between CPUs */ 2623 cpu_invltlb(); 2624#else 2625 invltlb(); 2626#endif 2627 } 2628 2629 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2630 unsigned *src_pte, *dst_pte; 2631 vm_page_t dstmpte, srcmpte; 2632 vm_offset_t srcptepaddr; 2633 unsigned ptepindex; 2634 2635#if !defined(MAX_PERF) 2636 if (addr >= UPT_MIN_ADDRESS) 2637 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2638#endif 2639 2640 /* 2641 * Don't let optional prefaulting of pages make us go 2642 * way below the low water mark of free pages or way 2643 * above high water mark of used pv entries. 2644 */ 2645 if (cnt.v_free_count < cnt.v_free_reserved || 2646 pv_entry_count > pv_entry_high_water) 2647 break; 2648 2649 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2650 ptepindex = addr >> PDRSHIFT; 2651 2652 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2653 if (srcptepaddr == 0) 2654 continue; 2655 2656 if (srcptepaddr & PG_PS) { 2657 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2658 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2659 dst_pmap->pm_stats.resident_count += NBPDR; 2660 } 2661 continue; 2662 } 2663 2664 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2665 if ((srcmpte == NULL) || 2666 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2667 continue; 2668 2669 if (pdnxt > end_addr) 2670 pdnxt = end_addr; 2671 2672 src_pte = (unsigned *) vtopte(addr); 2673 dst_pte = (unsigned *) avtopte(addr); 2674 while (addr < pdnxt) { 2675 unsigned ptetemp; 2676 ptetemp = *src_pte; 2677 /* 2678 * we only virtual copy managed pages 2679 */ 2680 if ((ptetemp & PG_MANAGED) != 0) { 2681 /* 2682 * We have to check after allocpte for the 2683 * pte still being around... allocpte can 2684 * block. 2685 */ 2686 dstmpte = pmap_allocpte(dst_pmap, addr); 2687 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2688 /* 2689 * Clear the modified and 2690 * accessed (referenced) bits 2691 * during the copy. 2692 */ 2693 *dst_pte = ptetemp & ~(PG_M | PG_A); 2694 dst_pmap->pm_stats.resident_count++; 2695 pmap_insert_entry(dst_pmap, addr, 2696 dstmpte, 2697 (ptetemp & PG_FRAME)); 2698 } else { 2699 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2700 } 2701 if (dstmpte->hold_count >= srcmpte->hold_count) 2702 break; 2703 } 2704 addr += PAGE_SIZE; 2705 src_pte++; 2706 dst_pte++; 2707 } 2708 } 2709} 2710 2711/* 2712 * Routine: pmap_kernel 2713 * Function: 2714 * Returns the physical map handle for the kernel. 2715 */ 2716pmap_t 2717pmap_kernel() 2718{ 2719 return (kernel_pmap); 2720} 2721 2722/* 2723 * pmap_zero_page zeros the specified hardware page by mapping 2724 * the page into KVM and using bzero to clear its contents. 2725 */ 2726void 2727pmap_zero_page(phys) 2728 vm_offset_t phys; 2729{ 2730#ifdef SMP 2731#if !defined(MAX_PERF) 2732 if (*(int *) prv_CMAP3) 2733 panic("pmap_zero_page: prv_CMAP3 busy"); 2734#endif 2735 2736 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2737 cpu_invlpg(prv_CADDR3); 2738 2739#if defined(I686_CPU) 2740 if (cpu_class == CPUCLASS_686) 2741 i686_pagezero(prv_CADDR3); 2742 else 2743#endif 2744 bzero(prv_CADDR3, PAGE_SIZE); 2745 2746 *(int *) prv_CMAP3 = 0; 2747#else 2748#if !defined(MAX_PERF) 2749 if (*(int *) CMAP2) 2750 panic("pmap_zero_page: CMAP2 busy"); 2751#endif 2752 2753 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2754 invltlb_1pg((vm_offset_t)CADDR2); 2755 2756#if defined(I686_CPU) 2757 if (cpu_class == CPUCLASS_686) 2758 i686_pagezero(CADDR2); 2759 else 2760#endif 2761 bzero(CADDR2, PAGE_SIZE); 2762 *(int *) CMAP2 = 0; 2763#endif 2764} 2765 2766/* 2767 * pmap_zero_page_area zeros the specified hardware page by mapping 2768 * the page into KVM and using bzero to clear its contents. 2769 * 2770 * off and size may not cover an area beyond a single hardware page. 2771 */ 2772void 2773pmap_zero_page_area(phys, off, size) 2774 vm_offset_t phys; 2775 int off; 2776 int size; 2777{ 2778#ifdef SMP 2779#if !defined(MAX_PERF) 2780 if (*(int *) prv_CMAP3) 2781 panic("pmap_zero_page: prv_CMAP3 busy"); 2782#endif 2783 2784 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2785 cpu_invlpg(prv_CADDR3); 2786 2787#if defined(I686_CPU) 2788 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2789 i686_pagezero(prv_CADDR3); 2790 else 2791#endif 2792 bzero((char *)prv_CADDR3 + off, size); 2793 2794 *(int *) prv_CMAP3 = 0; 2795#else 2796#if !defined(MAX_PERF) 2797 if (*(int *) CMAP2) 2798 panic("pmap_zero_page: CMAP2 busy"); 2799#endif 2800 2801 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2802 invltlb_1pg((vm_offset_t)CADDR2); 2803 2804#if defined(I686_CPU) 2805 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2806 i686_pagezero(CADDR2); 2807 else 2808#endif 2809 bzero((char *)CADDR2 + off, size); 2810 *(int *) CMAP2 = 0; 2811#endif 2812} 2813 2814/* 2815 * pmap_copy_page copies the specified (machine independent) 2816 * page by mapping the page into virtual memory and using 2817 * bcopy to copy the page, one machine dependent page at a 2818 * time. 2819 */ 2820void 2821pmap_copy_page(src, dst) 2822 vm_offset_t src; 2823 vm_offset_t dst; 2824{ 2825#ifdef SMP 2826#if !defined(MAX_PERF) 2827 if (*(int *) prv_CMAP1) 2828 panic("pmap_copy_page: prv_CMAP1 busy"); 2829 if (*(int *) prv_CMAP2) 2830 panic("pmap_copy_page: prv_CMAP2 busy"); 2831#endif 2832 2833 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2834 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2835 2836 cpu_invlpg(prv_CADDR1); 2837 cpu_invlpg(prv_CADDR2); 2838 2839 bcopy(prv_CADDR1, prv_CADDR2, PAGE_SIZE); 2840 2841 *(int *) prv_CMAP1 = 0; 2842 *(int *) prv_CMAP2 = 0; 2843#else 2844#if !defined(MAX_PERF) 2845 if (*(int *) CMAP1 || *(int *) CMAP2) 2846 panic("pmap_copy_page: CMAP busy"); 2847#endif 2848 2849 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2850 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2851#if defined(I386_CPU) 2852 if (cpu_class == CPUCLASS_386) { 2853 invltlb(); 2854 } else 2855#endif 2856 { 2857 invlpg((u_int)CADDR1); 2858 invlpg((u_int)CADDR2); 2859 } 2860 2861 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2862 2863 *(int *) CMAP1 = 0; 2864 *(int *) CMAP2 = 0; 2865#endif 2866} 2867 2868 2869/* 2870 * Routine: pmap_pageable 2871 * Function: 2872 * Make the specified pages (by pmap, offset) 2873 * pageable (or not) as requested. 2874 * 2875 * A page which is not pageable may not take 2876 * a fault; therefore, its page table entry 2877 * must remain valid for the duration. 2878 * 2879 * This routine is merely advisory; pmap_enter 2880 * will specify that these pages are to be wired 2881 * down (or not) as appropriate. 2882 */ 2883void 2884pmap_pageable(pmap, sva, eva, pageable) 2885 pmap_t pmap; 2886 vm_offset_t sva, eva; 2887 boolean_t pageable; 2888{ 2889} 2890 2891/* 2892 * this routine returns true if a physical page resides 2893 * in the given pmap. 2894 */ 2895boolean_t 2896pmap_page_exists(pmap, pa) 2897 pmap_t pmap; 2898 vm_offset_t pa; 2899{ 2900 register pv_entry_t pv; 2901 pv_table_t *ppv; 2902 int s; 2903 2904 if (!pmap_is_managed(pa)) 2905 return FALSE; 2906 2907 s = splvm(); 2908 2909 ppv = pa_to_pvh(pa); 2910 /* 2911 * Not found, check current mappings returning immediately if found. 2912 */ 2913 for (pv = TAILQ_FIRST(&ppv->pv_list); 2914 pv; 2915 pv = TAILQ_NEXT(pv, pv_list)) { 2916 if (pv->pv_pmap == pmap) { 2917 splx(s); 2918 return TRUE; 2919 } 2920 } 2921 splx(s); 2922 return (FALSE); 2923} 2924 2925#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2926/* 2927 * Remove all pages from specified address space 2928 * this aids process exit speeds. Also, this code 2929 * is special cased for current process only, but 2930 * can have the more generic (and slightly slower) 2931 * mode enabled. This is much faster than pmap_remove 2932 * in the case of running down an entire address space. 2933 */ 2934void 2935pmap_remove_pages(pmap, sva, eva) 2936 pmap_t pmap; 2937 vm_offset_t sva, eva; 2938{ 2939 unsigned *pte, tpte; 2940 pv_table_t *ppv; 2941 pv_entry_t pv, npv; 2942 int s; 2943 2944#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2945 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) { 2946 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2947 return; 2948 } 2949#endif 2950 2951 s = splvm(); 2952 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2953 pv; 2954 pv = npv) { 2955 2956 if (pv->pv_va >= eva || pv->pv_va < sva) { 2957 npv = TAILQ_NEXT(pv, pv_plist); 2958 continue; 2959 } 2960 2961#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2962 pte = (unsigned *)vtopte(pv->pv_va); 2963#else 2964 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2965#endif 2966 tpte = *pte; 2967 2968/* 2969 * We cannot remove wired pages from a process' mapping at this time 2970 */ 2971 if (tpte & PG_W) { 2972 npv = TAILQ_NEXT(pv, pv_plist); 2973 continue; 2974 } 2975 *pte = 0; 2976 2977 ppv = pa_to_pvh(tpte); 2978 2979 pv->pv_pmap->pm_stats.resident_count--; 2980 2981 /* 2982 * Update the vm_page_t clean and reference bits. 2983 */ 2984 if (tpte & PG_M) { 2985 vm_page_dirty(ppv->pv_vm_page); 2986 } 2987 2988 2989 npv = TAILQ_NEXT(pv, pv_plist); 2990 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2991 2992 ppv->pv_list_count--; 2993 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2994 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2995 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 2996 } 2997 2998 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2999 free_pv_entry(pv); 3000 } 3001 splx(s); 3002 pmap_TLB_invalidate_all(pmap); 3003} 3004 3005/* 3006 * pmap_testbit tests bits in pte's 3007 * note that the testbit/changebit routines are inline, 3008 * and a lot of things compile-time evaluate. 3009 */ 3010static boolean_t 3011pmap_testbit(pa, bit) 3012 register vm_offset_t pa; 3013 int bit; 3014{ 3015 register pv_entry_t pv; 3016 pv_table_t *ppv; 3017 unsigned *pte; 3018 int s; 3019 3020 if (!pmap_is_managed(pa)) 3021 return FALSE; 3022 3023 ppv = pa_to_pvh(pa); 3024 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 3025 return FALSE; 3026 3027 s = splvm(); 3028 3029 for (pv = TAILQ_FIRST(&ppv->pv_list); 3030 pv; 3031 pv = TAILQ_NEXT(pv, pv_list)) { 3032 3033 /* 3034 * if the bit being tested is the modified bit, then 3035 * mark clean_map and ptes as never 3036 * modified. 3037 */ 3038 if (bit & (PG_A|PG_M)) { 3039 if (!pmap_track_modified(pv->pv_va)) 3040 continue; 3041 } 3042 3043#if defined(PMAP_DIAGNOSTIC) 3044 if (!pv->pv_pmap) { 3045 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 3046 continue; 3047 } 3048#endif 3049 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3050 if (*pte & bit) { 3051 splx(s); 3052 return TRUE; 3053 } 3054 } 3055 splx(s); 3056 return (FALSE); 3057} 3058 3059/* 3060 * this routine is used to modify bits in ptes 3061 */ 3062static __inline void 3063pmap_changebit(pa, bit, setem) 3064 vm_offset_t pa; 3065 int bit; 3066 boolean_t setem; 3067{ 3068 register pv_entry_t pv; 3069 pv_table_t *ppv; 3070 register unsigned *pte; 3071 int s; 3072 3073 if (!pmap_is_managed(pa)) 3074 return; 3075 3076 s = splvm(); 3077 ppv = pa_to_pvh(pa); 3078 3079 /* 3080 * Loop over all current mappings setting/clearing as appropos If 3081 * setting RO do we need to clear the VAC? 3082 */ 3083 for (pv = TAILQ_FIRST(&ppv->pv_list); 3084 pv; 3085 pv = TAILQ_NEXT(pv, pv_list)) { 3086 3087 /* 3088 * don't write protect pager mappings 3089 */ 3090 if (!setem && (bit == PG_RW)) { 3091 if (!pmap_track_modified(pv->pv_va)) 3092 continue; 3093 } 3094 3095#if defined(PMAP_DIAGNOSTIC) 3096 if (!pv->pv_pmap) { 3097 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 3098 continue; 3099 } 3100#endif 3101 3102 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3103 3104 if (setem) { 3105 *(int *)pte |= bit; 3106 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3107 } else { 3108 vm_offset_t pbits = *(vm_offset_t *)pte; 3109 if (pbits & bit) { 3110 if (bit == PG_RW) { 3111 if (pbits & PG_M) { 3112 vm_page_dirty(ppv->pv_vm_page); 3113 } 3114 *(int *)pte = pbits & ~(PG_M|PG_RW); 3115 } else { 3116 *(int *)pte = pbits & ~bit; 3117 } 3118 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3119 } 3120 } 3121 } 3122 splx(s); 3123} 3124 3125/* 3126 * pmap_clearbit: 3127 * 3128 * Clear a bit/bits in every pte mapping a given physical page. 3129 */ 3130static void 3131pmap_clearbit( 3132 vm_offset_t pa, 3133 int bit) 3134{ 3135 pmap_changebit(pa, bit, FALSE); 3136} 3137 3138/* 3139 * pmap_page_protect: 3140 * 3141 * Lower the permission for all mappings to a given page. 3142 */ 3143void 3144pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 3145{ 3146 if ((prot & VM_PROT_WRITE) == 0) { 3147 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3148 pmap_clearbit(phys, PG_RW); 3149 } else { 3150 pmap_remove_all(phys); 3151 } 3152 } 3153} 3154 3155vm_offset_t 3156pmap_phys_address(ppn) 3157 int ppn; 3158{ 3159 return (i386_ptob(ppn)); 3160} 3161 3162/* 3163 * pmap_ts_referenced: 3164 * 3165 * Return the count of reference bits for a page, clearing all of them. 3166 */ 3167int 3168pmap_ts_referenced(vm_offset_t pa) 3169{ 3170 register pv_entry_t pv, pvf, pvn; 3171 pv_table_t *ppv; 3172 unsigned *pte; 3173 int s; 3174 int rtval = 0; 3175 3176 if (!pmap_is_managed(pa)) 3177 return (rtval); 3178 3179 s = splvm(); 3180 3181 ppv = pa_to_pvh(pa); 3182 3183 if ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 3184 3185 pvf = pv; 3186 3187 do { 3188 pvn = TAILQ_NEXT(pv, pv_list); 3189 3190 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 3191 3192 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3193 3194 if (!pmap_track_modified(pv->pv_va)) 3195 continue; 3196 3197 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3198 3199 if (pte && *pte & PG_A) { 3200 *pte &= ~PG_A; 3201 3202 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3203 3204 rtval++; 3205 if (rtval > 4) { 3206 break; 3207 } 3208 } 3209 } while ((pv = pvn) != NULL && pv != pvf); 3210 } 3211 splx(s); 3212 3213 return (rtval); 3214} 3215 3216/* 3217 * pmap_is_modified: 3218 * 3219 * Return whether or not the specified physical page was modified 3220 * in any physical maps. 3221 */ 3222boolean_t 3223pmap_is_modified(vm_offset_t pa) 3224{ 3225 return pmap_testbit((pa), PG_M); 3226} 3227 3228/* 3229 * Clear the modify bits on the specified physical page. 3230 */ 3231void 3232pmap_clear_modify(vm_offset_t pa) 3233{ 3234 pmap_clearbit(pa, PG_M); 3235} 3236 3237/* 3238 * pmap_clear_reference: 3239 * 3240 * Clear the reference bit on the specified physical page. 3241 */ 3242void 3243pmap_clear_reference(vm_offset_t pa) 3244{ 3245 pmap_clearbit(pa, PG_A); 3246} 3247 3248/* 3249 * Miscellaneous support routines follow 3250 */ 3251 3252static void 3253i386_protection_init() 3254{ 3255 register int *kp, prot; 3256 3257 kp = protection_codes; 3258 for (prot = 0; prot < 8; prot++) { 3259 switch (prot) { 3260 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3261 /* 3262 * Read access is also 0. There isn't any execute bit, 3263 * so just make it readable. 3264 */ 3265 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3266 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3267 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3268 *kp++ = 0; 3269 break; 3270 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3271 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3272 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3273 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3274 *kp++ = PG_RW; 3275 break; 3276 } 3277 } 3278} 3279 3280/* 3281 * Map a set of physical memory pages into the kernel virtual 3282 * address space. Return a pointer to where it is mapped. This 3283 * routine is intended to be used for mapping device memory, 3284 * NOT real memory. 3285 */ 3286void * 3287pmap_mapdev(pa, size) 3288 vm_offset_t pa; 3289 vm_size_t size; 3290{ 3291 vm_offset_t va, tmpva; 3292 unsigned *pte; 3293 3294 size = roundup(size, PAGE_SIZE); 3295 3296 va = kmem_alloc_pageable(kernel_map, size); 3297#if !defined(MAX_PERF) 3298 if (!va) 3299 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3300#endif 3301 3302 pa = pa & PG_FRAME; 3303 for (tmpva = va; size > 0;) { 3304 pte = (unsigned *)vtopte(tmpva); 3305 *pte = pa | PG_RW | PG_V | pgeflag; 3306 size -= PAGE_SIZE; 3307 tmpva += PAGE_SIZE; 3308 pa += PAGE_SIZE; 3309 } 3310 invltlb(); 3311 3312 return ((void *) va); 3313} 3314 3315/* 3316 * perform the pmap work for mincore 3317 */ 3318int 3319pmap_mincore(pmap, addr) 3320 pmap_t pmap; 3321 vm_offset_t addr; 3322{ 3323 3324 unsigned *ptep, pte; 3325 vm_page_t m; 3326 int val = 0; 3327 3328 ptep = pmap_pte(pmap, addr); 3329 if (ptep == 0) { 3330 return 0; 3331 } 3332 3333 if ((pte = *ptep) != 0) { 3334 pv_table_t *ppv; 3335 vm_offset_t pa; 3336 3337 val = MINCORE_INCORE; 3338 if ((pte & PG_MANAGED) == 0) 3339 return val; 3340 3341 pa = pte & PG_FRAME; 3342 3343 ppv = pa_to_pvh((pa & PG_FRAME)); 3344 m = ppv->pv_vm_page; 3345 3346 /* 3347 * Modified by us 3348 */ 3349 if (pte & PG_M) 3350 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3351 /* 3352 * Modified by someone 3353 */ 3354 else if (m->dirty || pmap_is_modified(pa)) 3355 val |= MINCORE_MODIFIED_OTHER; 3356 /* 3357 * Referenced by us 3358 */ 3359 if (pte & PG_A) 3360 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3361 3362 /* 3363 * Referenced by someone 3364 */ 3365 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(pa)) { 3366 val |= MINCORE_REFERENCED_OTHER; 3367 vm_page_flag_set(m, PG_REFERENCED); 3368 } 3369 } 3370 return val; 3371} 3372 3373void 3374pmap_activate(struct proc *p) 3375{ 3376 pmap_t pmap; 3377 3378 pmap = vmspace_pmap(p->p_vmspace); 3379#if defined(SMP) 3380 pmap->pm_active |= 1 << cpuid; 3381#else 3382 pmap->pm_active |= 1; 3383#endif 3384#if defined(SWTCH_OPTIM_STATS) 3385 tlb_flush_count++; 3386#endif 3387 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir)); 3388} 3389 3390vm_offset_t 3391pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3392 3393 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3394 return addr; 3395 } 3396 3397 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3398 return addr; 3399} 3400 3401 3402#if defined(PMAP_DEBUG) 3403pmap_pid_dump(int pid) { 3404 pmap_t pmap; 3405 struct proc *p; 3406 int npte = 0; 3407 int index; 3408 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3409 if (p->p_pid != pid) 3410 continue; 3411 3412 if (p->p_vmspace) { 3413 int i,j; 3414 index = 0; 3415 pmap = vmspace_pmap(p->p_vmspace); 3416 for(i=0;i<1024;i++) { 3417 pd_entry_t *pde; 3418 unsigned *pte; 3419 unsigned base = i << PDRSHIFT; 3420 3421 pde = &pmap->pm_pdir[i]; 3422 if (pde && pmap_pde_v(pde)) { 3423 for(j=0;j<1024;j++) { 3424 unsigned va = base + (j << PAGE_SHIFT); 3425 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3426 if (index) { 3427 index = 0; 3428 printf("\n"); 3429 } 3430 return npte; 3431 } 3432 pte = pmap_pte_quick( pmap, va); 3433 if (pte && pmap_pte_v(pte)) { 3434 vm_offset_t pa; 3435 vm_page_t m; 3436 pa = *(int *)pte; 3437 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3438 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3439 va, pa, m->hold_count, m->wire_count, m->flags); 3440 npte++; 3441 index++; 3442 if (index >= 2) { 3443 index = 0; 3444 printf("\n"); 3445 } else { 3446 printf(" "); 3447 } 3448 } 3449 } 3450 } 3451 } 3452 } 3453 } 3454 return npte; 3455} 3456#endif 3457 3458#if defined(DEBUG) 3459 3460static void pads __P((pmap_t pm)); 3461void pmap_pvdump __P((vm_offset_t pa)); 3462 3463/* print address space of pmap*/ 3464static void 3465pads(pm) 3466 pmap_t pm; 3467{ 3468 unsigned va, i, j; 3469 unsigned *ptep; 3470 3471 if (pm == kernel_pmap) 3472 return; 3473 for (i = 0; i < 1024; i++) 3474 if (pm->pm_pdir[i]) 3475 for (j = 0; j < 1024; j++) { 3476 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3477 if (pm == kernel_pmap && va < KERNBASE) 3478 continue; 3479 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3480 continue; 3481 ptep = pmap_pte_quick(pm, va); 3482 if (pmap_pte_v(ptep)) 3483 printf("%x:%x ", va, *(int *) ptep); 3484 }; 3485 3486} 3487 3488void 3489pmap_pvdump(pa) 3490 vm_offset_t pa; 3491{ 3492 pv_table_t *ppv; 3493 register pv_entry_t pv; 3494 3495 printf("pa %x", pa); 3496 ppv = pa_to_pvh(pa); 3497 for (pv = TAILQ_FIRST(&ppv->pv_list); 3498 pv; 3499 pv = TAILQ_NEXT(pv, pv_list)) { 3500#ifdef used_to_be 3501 printf(" -> pmap %p, va %x, flags %x", 3502 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags); 3503#endif 3504 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3505 pads(pv->pv_pmap); 3506 } 3507 printf(" "); 3508} 3509#endif 3510