pmap.c revision 43314
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 1.221 1999/01/24 06:04:51 dillon 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 75#include <sys/param.h> 76#include <sys/systm.h> 77#include <sys/proc.h> 78#include <sys/msgbuf.h> 79#include <sys/vmmeter.h> 80#include <sys/mman.h> 81 82#include <vm/vm.h> 83#include <vm/vm_param.h> 84#include <vm/vm_prot.h> 85#include <sys/lock.h> 86#include <vm/vm_kern.h> 87#include <vm/vm_page.h> 88#include <vm/vm_map.h> 89#include <vm/vm_object.h> 90#include <vm/vm_extern.h> 91#include <vm/vm_pageout.h> 92#include <vm/vm_pager.h> 93#include <vm/vm_zone.h> 94 95#include <sys/user.h> 96 97#include <machine/cputypes.h> 98#include <machine/md_var.h> 99#include <machine/specialreg.h> 100#if defined(SMP) || defined(APIC_IO) 101#include <machine/smp.h> 102#include <machine/apic.h> 103#endif /* SMP || APIC_IO */ 104 105#define PMAP_KEEP_PDIRS 106#ifndef PMAP_SHPGPERPROC 107#define PMAP_SHPGPERPROC 200 108#endif 109 110#if defined(DIAGNOSTIC) 111#define PMAP_DIAGNOSTIC 112#endif 113 114#define MINPV 2048 115 116#if !defined(PMAP_DIAGNOSTIC) 117#define PMAP_INLINE __inline 118#else 119#define PMAP_INLINE 120#endif 121 122/* 123 * Get PDEs and PTEs for user/kernel address space 124 */ 125#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 126#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 127 128#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 129#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 130#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 131#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 132#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 133 134#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 135#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 136 137/* 138 * Given a map and a machine independent protection code, 139 * convert to a vax protection code. 140 */ 141#define pte_prot(m, p) (protection_codes[p]) 142static int protection_codes[8]; 143 144#define pa_index(pa) atop((pa) - vm_first_phys) 145#define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) 146 147static struct pmap kernel_pmap_store; 148pmap_t kernel_pmap; 149extern pd_entry_t my_idlePTD; 150 151vm_offset_t avail_start; /* PA of first available physical page */ 152vm_offset_t avail_end; /* PA of last available physical page */ 153vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 154vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 155static boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 156static vm_offset_t vm_first_phys; 157static int pgeflag; /* PG_G or-in */ 158static int pseflag; /* PG_PS or-in */ 159static int pv_npg; 160 161static vm_object_t kptobj; 162 163static int nkpt; 164vm_offset_t kernel_vm_end; 165 166/* 167 * Data for the pv entry allocation mechanism 168 */ 169static vm_zone_t pvzone; 170static struct vm_zone pvzone_store; 171static struct vm_object pvzone_obj; 172static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0; 173static int pmap_pagedaemon_waken = 0; 174static struct pv_entry *pvinit; 175 176/* 177 * All those kernel PT submaps that BSD is so fond of 178 */ 179pt_entry_t *CMAP1 = 0; 180static pt_entry_t *CMAP2, *ptmmap; 181static pv_table_t *pv_table; 182caddr_t CADDR1 = 0, ptvmmap = 0; 183static caddr_t CADDR2; 184static pt_entry_t *msgbufmap; 185struct msgbuf *msgbufp=0; 186 187/* 188 * PPro_vmtrr 189 */ 190struct ppro_vmtrr PPro_vmtrr[NPPROVMTRR]; 191 192/* AIO support */ 193extern struct vmspace *aiovmspace; 194 195#ifdef SMP 196extern char prv_CPAGE1[], prv_CPAGE2[], prv_CPAGE3[]; 197extern pt_entry_t *prv_CMAP1, *prv_CMAP2, *prv_CMAP3; 198extern pd_entry_t *IdlePTDS[]; 199extern pt_entry_t SMP_prvpt[]; 200#endif 201 202#ifdef SMP 203extern unsigned int prv_PPAGE1[]; 204extern pt_entry_t *prv_PMAP1; 205#else 206static pt_entry_t *PMAP1 = 0; 207static unsigned *PADDR1 = 0; 208#endif 209 210static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 211static unsigned * get_ptbase __P((pmap_t pmap)); 212static pv_entry_t get_pv_entry __P((void)); 213static void i386_protection_init __P((void)); 214static void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 215 216static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 217static void pmap_remove_all __P((vm_offset_t pa)); 218static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 219 vm_offset_t pa, vm_page_t mpte)); 220static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 221 vm_offset_t sva)); 222static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 223static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 224 vm_offset_t va)); 225static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 226static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 227 vm_page_t mpte, vm_offset_t pa)); 228 229static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 230 231static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 232static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 233static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 234static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 235static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 236static vm_offset_t pmap_kmem_choose(vm_offset_t addr); 237void pmap_collect(void); 238 239static unsigned pdir4mb; 240 241/* 242 * Routine: pmap_pte 243 * Function: 244 * Extract the page table entry associated 245 * with the given map/virtual_address pair. 246 */ 247 248PMAP_INLINE unsigned * 249pmap_pte(pmap, va) 250 register pmap_t pmap; 251 vm_offset_t va; 252{ 253 unsigned *pdeaddr; 254 255 if (pmap) { 256 pdeaddr = (unsigned *) pmap_pde(pmap, va); 257 if (*pdeaddr & PG_PS) 258 return pdeaddr; 259 if (*pdeaddr) { 260 return get_ptbase(pmap) + i386_btop(va); 261 } 262 } 263 return (0); 264} 265 266/* 267 * Move the kernel virtual free pointer to the next 268 * 4MB. This is used to help improve performance 269 * by using a large (4MB) page for much of the kernel 270 * (.text, .data, .bss) 271 */ 272static vm_offset_t 273pmap_kmem_choose(vm_offset_t addr) { 274 vm_offset_t newaddr = addr; 275#ifndef DISABLE_PSE 276 if (cpu_feature & CPUID_PSE) { 277 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 278 } 279#endif 280 return newaddr; 281} 282 283/* 284 * Bootstrap the system enough to run with virtual memory. 285 * 286 * On the i386 this is called after mapping has already been enabled 287 * and just syncs the pmap module with what has already been done. 288 * [We can't call it easily with mapping off since the kernel is not 289 * mapped with PA == VA, hence we would have to relocate every address 290 * from the linked base (virtual) address "KERNBASE" to the actual 291 * (physical) address starting relative to 0] 292 */ 293void 294pmap_bootstrap(firstaddr, loadaddr) 295 vm_offset_t firstaddr; 296 vm_offset_t loadaddr; 297{ 298 vm_offset_t va; 299 pt_entry_t *pte; 300#ifdef SMP 301 int i, j; 302#endif 303 304 avail_start = firstaddr; 305 306 /* 307 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 308 * large. It should instead be correctly calculated in locore.s and 309 * not based on 'first' (which is a physical address, not a virtual 310 * address, for the start of unused physical memory). The kernel 311 * page tables are NOT double mapped and thus should not be included 312 * in this calculation. 313 */ 314 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 315 virtual_avail = pmap_kmem_choose(virtual_avail); 316 317 virtual_end = VM_MAX_KERNEL_ADDRESS; 318 319 /* 320 * Initialize protection array. 321 */ 322 i386_protection_init(); 323 324 /* 325 * The kernel's pmap is statically allocated so we don't have to use 326 * pmap_create, which is unlikely to work correctly at this part of 327 * the boot sequence (XXX and which no longer exists). 328 */ 329 kernel_pmap = &kernel_pmap_store; 330 331 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 332 333 kernel_pmap->pm_count = 1; 334 TAILQ_INIT(&kernel_pmap->pm_pvlist); 335 nkpt = NKPT; 336 337 /* 338 * Reserve some special page table entries/VA space for temporary 339 * mapping of pages. 340 */ 341#define SYSMAP(c, p, v, n) \ 342 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 343 344 va = virtual_avail; 345 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 346 347 /* 348 * CMAP1/CMAP2 are used for zeroing and copying pages. 349 */ 350 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 351 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 352 353 /* 354 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 355 * XXX ptmmap is not used. 356 */ 357 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 358 359 /* 360 * msgbufp is used to map the system message buffer. 361 * XXX msgbufmap is not used. 362 */ 363 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 364 atop(round_page(MSGBUF_SIZE))) 365 366#if !defined(SMP) 367 /* 368 * ptemap is used for pmap_pte_quick 369 */ 370 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 371#endif 372 373 virtual_avail = va; 374 375 *(int *) CMAP1 = *(int *) CMAP2 = 0; 376 *(int *) PTD = 0; 377 378 379 pgeflag = 0; 380#if !defined(SMP) 381 if (cpu_feature & CPUID_PGE) { 382 pgeflag = PG_G; 383 } 384#endif 385 386/* 387 * Initialize the 4MB page size flag 388 */ 389 pseflag = 0; 390/* 391 * The 4MB page version of the initial 392 * kernel page mapping. 393 */ 394 pdir4mb = 0; 395 396#if !defined(DISABLE_PSE) 397 if (cpu_feature & CPUID_PSE) { 398 unsigned ptditmp; 399 /* 400 * Enable the PSE mode 401 */ 402 load_cr4(rcr4() | CR4_PSE); 403 404 /* 405 * Note that we have enabled PSE mode 406 */ 407 pseflag = PG_PS; 408 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE)); 409 ptditmp &= ~(NBPDR - 1); 410 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 411 pdir4mb = ptditmp; 412 /* 413 * We can do the mapping here for the single processor 414 * case. We simply ignore the old page table page from 415 * now on. 416 */ 417#if !defined(SMP) 418 PTD[KPTDI] = (pd_entry_t) ptditmp; 419 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 420 invltlb(); 421#endif 422 } 423#endif 424 425#ifdef SMP 426 if (cpu_apic_address == 0) 427 panic("pmap_bootstrap: no local apic!"); 428 429 /* 0 = private page */ 430 /* 1 = page table page */ 431 /* 2 = local apic */ 432 /* 16-31 = io apics */ 433 SMP_prvpt[2] = (pt_entry_t)(PG_V | PG_RW | pgeflag | 434 (cpu_apic_address & PG_FRAME)); 435 436 for (i = 0; i < mp_napics; i++) { 437 for (j = 0; j < 16; j++) { 438 /* same page frame as a previous IO apic? */ 439 if (((vm_offset_t)SMP_prvpt[j + 16] & PG_FRAME) == 440 (io_apic_address[0] & PG_FRAME)) { 441 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 442 break; 443 } 444 /* use this slot if available */ 445 if (((vm_offset_t)SMP_prvpt[j + 16] & PG_FRAME) == 0) { 446 SMP_prvpt[j + 16] = (pt_entry_t)(PG_V | PG_RW | 447 pgeflag | (io_apic_address[i] & PG_FRAME)); 448 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 449 break; 450 } 451 } 452 if (j == 16) 453 panic("no space to map IO apic %d!", i); 454 } 455 456 /* BSP does this itself, AP's get it pre-set */ 457 prv_CMAP1 = &SMP_prvpt[3 + UPAGES]; 458 prv_CMAP2 = &SMP_prvpt[4 + UPAGES]; 459 prv_CMAP3 = &SMP_prvpt[5 + UPAGES]; 460 prv_PMAP1 = &SMP_prvpt[6 + UPAGES]; 461#endif 462 463 invltlb(); 464 465} 466 467void 468getmtrr() 469{ 470 int i; 471 472 if (cpu_class == CPUCLASS_686) { 473 for(i = 0; i < NPPROVMTRR; i++) { 474 PPro_vmtrr[i].base = rdmsr(PPRO_VMTRRphysBase0 + i * 2); 475 PPro_vmtrr[i].mask = rdmsr(PPRO_VMTRRphysMask0 + i * 2); 476 } 477 } 478} 479 480void 481putmtrr() 482{ 483 int i; 484 485 if (cpu_class == CPUCLASS_686) { 486 wbinvd(); 487 for(i = 0; i < NPPROVMTRR; i++) { 488 wrmsr(PPRO_VMTRRphysBase0 + i * 2, PPro_vmtrr[i].base); 489 wrmsr(PPRO_VMTRRphysMask0 + i * 2, PPro_vmtrr[i].mask); 490 } 491 } 492} 493 494void 495pmap_setvidram(void) 496{ 497#if 0 498 if (cpu_class == CPUCLASS_686) { 499 wbinvd(); 500 /* 501 * Set memory between 0-640K to be WB 502 */ 503 wrmsr(0x250, 0x0606060606060606LL); 504 wrmsr(0x258, 0x0606060606060606LL); 505 /* 506 * Set normal, PC video memory to be WC 507 */ 508 wrmsr(0x259, 0x0101010101010101LL); 509 } 510#endif 511} 512 513void 514pmap_setdevram(unsigned long long basea, vm_offset_t sizea) 515{ 516 int i, free, skip; 517 unsigned basepage, basepaget; 518 unsigned long long base; 519 unsigned long long mask; 520 521 if (cpu_class != CPUCLASS_686) 522 return; 523 524 free = -1; 525 skip = 0; 526 basea &= ~0xfff; 527 base = basea | 0x1; 528 mask = (long long) (0xfffffffffLL - ((long) sizea - 1)) | (long long) 0x800; 529 mask &= ~0x7ff; 530 531 basepage = (long long) (base >> 12); 532 for(i = 0; i < NPPROVMTRR; i++) { 533 PPro_vmtrr[i].base = rdmsr(PPRO_VMTRRphysBase0 + i * 2); 534 PPro_vmtrr[i].mask = rdmsr(PPRO_VMTRRphysMask0 + i * 2); 535 basepaget = (long long) (PPro_vmtrr[i].base >> 12); 536 if (basepage == basepaget) 537 skip = 1; 538 if ((PPro_vmtrr[i].mask & 0x800) == 0) { 539 if (free == -1) 540 free = i; 541 } 542 } 543 544 if (!skip && free != -1) { 545 wbinvd(); 546 PPro_vmtrr[free].base = base; 547 PPro_vmtrr[free].mask = mask; 548 wrmsr(PPRO_VMTRRphysBase0 + free * 2, base); 549 wrmsr(PPRO_VMTRRphysMask0 + free * 2, mask); 550 printf( 551 "pmap: added WC mapping at page: 0x%x %x, size: %u mask: 0x%x %x\n", 552 (u_int)(base >> 32), (u_int)base, sizea, 553 (u_int)(mask >> 32), (u_int)mask); 554 } 555} 556 557/* 558 * Set 4mb pdir for mp startup, and global flags 559 */ 560void 561pmap_set_opt(unsigned *pdir) { 562 int i; 563 564 if (pseflag && (cpu_feature & CPUID_PSE)) { 565 load_cr4(rcr4() | CR4_PSE); 566 if (pdir4mb) { 567 pdir[KPTDI] = pdir4mb; 568 } 569 } 570 571 if (pgeflag && (cpu_feature & CPUID_PGE)) { 572 load_cr4(rcr4() | CR4_PGE); 573 for(i = KPTDI; i < KPTDI + nkpt; i++) { 574 if (pdir[i]) { 575 pdir[i] |= PG_G; 576 } 577 } 578 } 579} 580 581/* 582 * Setup the PTD for the boot processor 583 */ 584void 585pmap_set_opt_bsp(void) 586{ 587 pmap_set_opt((unsigned *)kernel_pmap->pm_pdir); 588 pmap_set_opt((unsigned *)PTD); 589 invltlb(); 590} 591 592/* 593 * Initialize the pmap module. 594 * Called by vm_init, to initialize any structures that the pmap 595 * system needs to map virtual memory. 596 * pmap_init has been enhanced to support in a fairly consistant 597 * way, discontiguous physical memory. 598 */ 599void 600pmap_init(phys_start, phys_end) 601 vm_offset_t phys_start, phys_end; 602{ 603 vm_offset_t addr; 604 vm_size_t s; 605 int i; 606 int initial_pvs; 607 608 /* 609 * object for kernel page table pages 610 */ 611 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE); 612 613 /* 614 * calculate the number of pv_entries needed 615 */ 616 vm_first_phys = phys_avail[0]; 617 for (i = 0; phys_avail[i + 1]; i += 2); 618 pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 619 620 /* 621 * Allocate memory for random pmap data structures. Includes the 622 * pv_head_table. 623 */ 624 s = (vm_size_t) (sizeof(pv_table_t) * pv_npg); 625 s = round_page(s); 626 627 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 628 pv_table = (pv_table_t *) addr; 629 for(i = 0; i < pv_npg; i++) { 630 vm_offset_t pa; 631 TAILQ_INIT(&pv_table[i].pv_list); 632 pv_table[i].pv_list_count = 0; 633 pa = vm_first_phys + i * PAGE_SIZE; 634 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 635 } 636 637 /* 638 * init the pv free list 639 */ 640 initial_pvs = pv_npg; 641 if (initial_pvs < MINPV) 642 initial_pvs = MINPV; 643 pvzone = &pvzone_store; 644 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 645 initial_pvs * sizeof (struct pv_entry)); 646 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, pv_npg); 647 648 /* 649 * Now it is safe to enable pv_table recording. 650 */ 651 pmap_initialized = TRUE; 652} 653 654/* 655 * Initialize the address space (zone) for the pv_entries. Set a 656 * high water mark so that the system can recover from excessive 657 * numbers of pv entries. 658 */ 659void 660pmap_init2() { 661 pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg; 662 pv_entry_high_water = 9 * (pv_entry_max / 10); 663 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 664} 665 666/* 667 * Used to map a range of physical addresses into kernel 668 * virtual address space. 669 * 670 * For now, VM is already on, we only need to map the 671 * specified memory. 672 */ 673vm_offset_t 674pmap_map(virt, start, end, prot) 675 vm_offset_t virt; 676 vm_offset_t start; 677 vm_offset_t end; 678 int prot; 679{ 680 while (start < end) { 681 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 682 virt += PAGE_SIZE; 683 start += PAGE_SIZE; 684 } 685 return (virt); 686} 687 688 689/*************************************************** 690 * Low level helper routines..... 691 ***************************************************/ 692 693#if defined(PMAP_DIAGNOSTIC) 694 695/* 696 * This code checks for non-writeable/modified pages. 697 * This should be an invalid condition. 698 */ 699static int 700pmap_nw_modified(pt_entry_t ptea) { 701 int pte; 702 703 pte = (int) ptea; 704 705 if ((pte & (PG_M|PG_RW)) == PG_M) 706 return 1; 707 else 708 return 0; 709} 710#endif 711 712 713/* 714 * this routine defines the region(s) of memory that should 715 * not be tested for the modified bit. 716 */ 717static PMAP_INLINE int 718pmap_track_modified( vm_offset_t va) { 719 if ((va < clean_sva) || (va >= clean_eva)) 720 return 1; 721 else 722 return 0; 723} 724 725static PMAP_INLINE void 726invltlb_1pg( vm_offset_t va) { 727#if defined(I386_CPU) 728 if (cpu_class == CPUCLASS_386) { 729 invltlb(); 730 } else 731#endif 732 { 733 invlpg(va); 734 } 735} 736 737static unsigned * 738get_ptbase(pmap) 739 pmap_t pmap; 740{ 741 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 742 743 /* are we current address space or kernel? */ 744 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 745 return (unsigned *) PTmap; 746 } 747 /* otherwise, we are alternate address space */ 748 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 749 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 750#if defined(SMP) 751 /* The page directory is not shared between CPUs */ 752 cpu_invltlb(); 753#else 754 invltlb(); 755#endif 756 } 757 return (unsigned *) APTmap; 758} 759 760/* 761 * Super fast pmap_pte routine best used when scanning 762 * the pv lists. This eliminates many coarse-grained 763 * invltlb calls. Note that many of the pv list 764 * scans are across different pmaps. It is very wasteful 765 * to do an entire invltlb for checking a single mapping. 766 */ 767 768static unsigned * 769pmap_pte_quick(pmap, va) 770 register pmap_t pmap; 771 vm_offset_t va; 772{ 773 unsigned pde, newpf; 774 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) { 775 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 776 unsigned index = i386_btop(va); 777 /* are we current address space or kernel? */ 778 if ((pmap == kernel_pmap) || 779 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 780 return (unsigned *) PTmap + index; 781 } 782 newpf = pde & PG_FRAME; 783#ifdef SMP 784 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) { 785 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V; 786 cpu_invlpg(&prv_PPAGE1); 787 } 788 return prv_PPAGE1 + ((unsigned) index & (NPTEPG - 1)); 789#else 790 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 791 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 792 invltlb_1pg((vm_offset_t) PADDR1); 793 } 794 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 795#endif 796 } 797 return (0); 798} 799 800/* 801 * Routine: pmap_extract 802 * Function: 803 * Extract the physical page address associated 804 * with the given map/virtual_address pair. 805 */ 806vm_offset_t 807pmap_extract(pmap, va) 808 register pmap_t pmap; 809 vm_offset_t va; 810{ 811 vm_offset_t rtval; 812 vm_offset_t pdirindex; 813 pdirindex = va >> PDRSHIFT; 814 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 815 unsigned *pte; 816 if ((rtval & PG_PS) != 0) { 817 rtval &= ~(NBPDR - 1); 818 rtval |= va & (NBPDR - 1); 819 return rtval; 820 } 821 pte = get_ptbase(pmap) + i386_btop(va); 822 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 823 return rtval; 824 } 825 return 0; 826 827} 828 829/* 830 * determine if a page is managed (memory vs. device) 831 */ 832static PMAP_INLINE int 833pmap_is_managed(pa) 834 vm_offset_t pa; 835{ 836 int i; 837 838 if (!pmap_initialized) 839 return 0; 840 841 for (i = 0; phys_avail[i + 1]; i += 2) { 842 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 843 return 1; 844 } 845 return 0; 846} 847 848 849/*************************************************** 850 * Low level mapping routines..... 851 ***************************************************/ 852 853/* 854 * Add a list of wired pages to the kva 855 * this routine is only used for temporary 856 * kernel mappings that do not need to have 857 * page modification or references recorded. 858 * Note that old mappings are simply written 859 * over. The page *must* be wired. 860 */ 861void 862pmap_qenter(va, m, count) 863 vm_offset_t va; 864 vm_page_t *m; 865 int count; 866{ 867 int i; 868 register unsigned *pte; 869 870 for (i = 0; i < count; i++) { 871 vm_offset_t tva = va + i * PAGE_SIZE; 872 unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | pgeflag; 873 unsigned opte; 874 pte = (unsigned *)vtopte(tva); 875 opte = *pte; 876 *pte = npte; 877 if (opte) 878 invltlb_1pg(tva); 879 } 880} 881 882/* 883 * this routine jerks page mappings from the 884 * kernel -- it is meant only for temporary mappings. 885 */ 886void 887pmap_qremove(va, count) 888 vm_offset_t va; 889 int count; 890{ 891 int i; 892 register unsigned *pte; 893 894 for (i = 0; i < count; i++) { 895 pte = (unsigned *)vtopte(va); 896 *pte = 0; 897 invltlb_1pg(va); 898 va += PAGE_SIZE; 899 } 900} 901 902/* 903 * add a wired page to the kva 904 * note that in order for the mapping to take effect -- you 905 * should do a invltlb after doing the pmap_kenter... 906 */ 907PMAP_INLINE void 908pmap_kenter(va, pa) 909 vm_offset_t va; 910 register vm_offset_t pa; 911{ 912 register unsigned *pte; 913 unsigned npte, opte; 914 915 npte = pa | PG_RW | PG_V | pgeflag; 916 pte = (unsigned *)vtopte(va); 917 opte = *pte; 918 *pte = npte; 919 if (opte) 920 invltlb_1pg(va); 921} 922 923/* 924 * remove a page from the kernel pagetables 925 */ 926PMAP_INLINE void 927pmap_kremove(va) 928 vm_offset_t va; 929{ 930 register unsigned *pte; 931 932 pte = (unsigned *)vtopte(va); 933 *pte = 0; 934 invltlb_1pg(va); 935} 936 937static vm_page_t 938pmap_page_lookup(object, pindex) 939 vm_object_t object; 940 vm_pindex_t pindex; 941{ 942 vm_page_t m; 943retry: 944 m = vm_page_lookup(object, pindex); 945 if (m && vm_page_sleep_busy(m, FALSE, "pplookp")) 946 goto retry; 947 return m; 948} 949 950/* 951 * Create the UPAGES for a new process. 952 * This routine directly affects the fork perf for a process. 953 */ 954void 955pmap_new_proc(p) 956 struct proc *p; 957{ 958 int i, updateneeded; 959 vm_object_t upobj; 960 vm_page_t m; 961 struct user *up; 962 unsigned *ptek, oldpte; 963 964 /* 965 * allocate object for the upages 966 */ 967 if ((upobj = p->p_upages_obj) == NULL) { 968 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 969 p->p_upages_obj = upobj; 970 } 971 972 /* get a kernel virtual address for the UPAGES for this proc */ 973 if ((up = p->p_addr) == NULL) { 974 up = (struct user *) kmem_alloc_pageable(kernel_map, 975 UPAGES * PAGE_SIZE); 976#if !defined(MAX_PERF) 977 if (up == NULL) 978 panic("pmap_new_proc: u_map allocation failed"); 979#endif 980 p->p_addr = up; 981 } 982 983 ptek = (unsigned *) vtopte((vm_offset_t) up); 984 985 updateneeded = 0; 986 for(i=0;i<UPAGES;i++) { 987 /* 988 * Get a kernel stack page 989 */ 990 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 991 992 /* 993 * Wire the page 994 */ 995 m->wire_count++; 996 cnt.v_wire_count++; 997 998 oldpte = *(ptek + i); 999 /* 1000 * Enter the page into the kernel address space. 1001 */ 1002 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 1003 if (oldpte) { 1004 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) { 1005 invlpg((vm_offset_t) up + i * PAGE_SIZE); 1006 } else { 1007 updateneeded = 1; 1008 } 1009 } 1010 1011 vm_page_wakeup(m); 1012 vm_page_flag_clear(m, PG_ZERO); 1013 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 1014 m->valid = VM_PAGE_BITS_ALL; 1015 } 1016 if (updateneeded) 1017 invltlb(); 1018} 1019 1020/* 1021 * Dispose the UPAGES for a process that has exited. 1022 * This routine directly impacts the exit perf of a process. 1023 */ 1024void 1025pmap_dispose_proc(p) 1026 struct proc *p; 1027{ 1028 int i; 1029 vm_object_t upobj; 1030 vm_page_t m; 1031 unsigned *ptek, oldpte; 1032 1033 upobj = p->p_upages_obj; 1034 1035 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 1036 for(i=0;i<UPAGES;i++) { 1037 1038 if ((m = vm_page_lookup(upobj, i)) == NULL) 1039 panic("pmap_dispose_proc: upage already missing???"); 1040 1041 vm_page_busy(m); 1042 1043 oldpte = *(ptek + i); 1044 *(ptek + i) = 0; 1045 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) 1046 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 1047 vm_page_unwire(m, 0); 1048 vm_page_free(m); 1049 } 1050 1051 if (cpu_class <= CPUCLASS_386) 1052 invltlb(); 1053} 1054 1055/* 1056 * Allow the UPAGES for a process to be prejudicially paged out. 1057 */ 1058void 1059pmap_swapout_proc(p) 1060 struct proc *p; 1061{ 1062 int i; 1063 vm_object_t upobj; 1064 vm_page_t m; 1065 1066 upobj = p->p_upages_obj; 1067 /* 1068 * let the upages be paged 1069 */ 1070 for(i=0;i<UPAGES;i++) { 1071 if ((m = vm_page_lookup(upobj, i)) == NULL) 1072 panic("pmap_swapout_proc: upage already missing???"); 1073 vm_page_dirty(m); 1074 vm_page_unwire(m, 0); 1075 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 1076 } 1077} 1078 1079/* 1080 * Bring the UPAGES for a specified process back in. 1081 */ 1082void 1083pmap_swapin_proc(p) 1084 struct proc *p; 1085{ 1086 int i,rv; 1087 vm_object_t upobj; 1088 vm_page_t m; 1089 1090 upobj = p->p_upages_obj; 1091 for(i=0;i<UPAGES;i++) { 1092 1093 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1094 1095 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 1096 VM_PAGE_TO_PHYS(m)); 1097 1098 if (m->valid != VM_PAGE_BITS_ALL) { 1099 rv = vm_pager_get_pages(upobj, &m, 1, 0); 1100#if !defined(MAX_PERF) 1101 if (rv != VM_PAGER_OK) 1102 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 1103#endif 1104 m = vm_page_lookup(upobj, i); 1105 m->valid = VM_PAGE_BITS_ALL; 1106 } 1107 1108 vm_page_wire(m); 1109 vm_page_wakeup(m); 1110 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 1111 } 1112} 1113 1114/*************************************************** 1115 * Page table page management routines..... 1116 ***************************************************/ 1117 1118/* 1119 * This routine unholds page table pages, and if the hold count 1120 * drops to zero, then it decrements the wire count. 1121 */ 1122static int 1123_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1124 1125 while (vm_page_sleep_busy(m, FALSE, "pmuwpt")) 1126 ; 1127 1128 if (m->hold_count == 0) { 1129 vm_offset_t pteva; 1130 /* 1131 * unmap the page table page 1132 */ 1133 pmap->pm_pdir[m->pindex] = 0; 1134 --pmap->pm_stats.resident_count; 1135 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1136 (((unsigned) PTDpde) & PG_FRAME)) { 1137 /* 1138 * Do a invltlb to make the invalidated mapping 1139 * take effect immediately. 1140 */ 1141 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1142 invltlb_1pg(pteva); 1143 } 1144 1145 if (pmap->pm_ptphint == m) 1146 pmap->pm_ptphint = NULL; 1147 1148 /* 1149 * If the page is finally unwired, simply free it. 1150 */ 1151 --m->wire_count; 1152 if (m->wire_count == 0) { 1153 1154 vm_page_flash(m); 1155 vm_page_busy(m); 1156 vm_page_free_zero(m); 1157 --cnt.v_wire_count; 1158 } 1159 return 1; 1160 } 1161 return 0; 1162} 1163 1164static PMAP_INLINE int 1165pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1166 vm_page_unhold(m); 1167 if (m->hold_count == 0) 1168 return _pmap_unwire_pte_hold(pmap, m); 1169 else 1170 return 0; 1171} 1172 1173/* 1174 * After removing a page table entry, this routine is used to 1175 * conditionally free the page, and manage the hold/wire counts. 1176 */ 1177static int 1178pmap_unuse_pt(pmap, va, mpte) 1179 pmap_t pmap; 1180 vm_offset_t va; 1181 vm_page_t mpte; 1182{ 1183 unsigned ptepindex; 1184 if (va >= UPT_MIN_ADDRESS) 1185 return 0; 1186 1187 if (mpte == NULL) { 1188 ptepindex = (va >> PDRSHIFT); 1189 if (pmap->pm_ptphint && 1190 (pmap->pm_ptphint->pindex == ptepindex)) { 1191 mpte = pmap->pm_ptphint; 1192 } else { 1193 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1194 pmap->pm_ptphint = mpte; 1195 } 1196 } 1197 1198 return pmap_unwire_pte_hold(pmap, mpte); 1199} 1200 1201#if !defined(SMP) 1202void 1203pmap_pinit0(pmap) 1204 struct pmap *pmap; 1205{ 1206 pmap->pm_pdir = 1207 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1208 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1209 pmap->pm_flags = 0; 1210 pmap->pm_count = 1; 1211 pmap->pm_ptphint = NULL; 1212 TAILQ_INIT(&pmap->pm_pvlist); 1213 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1214} 1215#else 1216void 1217pmap_pinit0(pmap) 1218 struct pmap *pmap; 1219{ 1220 pmap_pinit(pmap); 1221} 1222#endif 1223 1224/* 1225 * Initialize a preallocated and zeroed pmap structure, 1226 * such as one in a vmspace structure. 1227 */ 1228void 1229pmap_pinit(pmap) 1230 register struct pmap *pmap; 1231{ 1232 vm_page_t ptdpg; 1233 1234 /* 1235 * No need to allocate page table space yet but we do need a valid 1236 * page directory table. 1237 */ 1238 if (pmap->pm_pdir == NULL) 1239 pmap->pm_pdir = 1240 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1241 1242 /* 1243 * allocate object for the ptes 1244 */ 1245 if (pmap->pm_pteobj == NULL) 1246 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1247 1248 /* 1249 * allocate the page directory page 1250 */ 1251 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1252 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1253 1254 ptdpg->wire_count = 1; 1255 ++cnt.v_wire_count; 1256 1257 1258 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/ 1259 ptdpg->valid = VM_PAGE_BITS_ALL; 1260 1261 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1262 if ((ptdpg->flags & PG_ZERO) == 0) 1263 bzero(pmap->pm_pdir, PAGE_SIZE); 1264 1265 /* wire in kernel global address entries */ 1266 /* XXX copies current process, does not fill in MPPTDI */ 1267 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1268 1269 /* install self-referential address mapping entry */ 1270 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1271 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1272 1273 pmap->pm_flags = 0; 1274 pmap->pm_count = 1; 1275 pmap->pm_ptphint = NULL; 1276 TAILQ_INIT(&pmap->pm_pvlist); 1277 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1278} 1279 1280static int 1281pmap_release_free_page(pmap, p) 1282 struct pmap *pmap; 1283 vm_page_t p; 1284{ 1285 unsigned *pde = (unsigned *) pmap->pm_pdir; 1286 /* 1287 * This code optimizes the case of freeing non-busy 1288 * page-table pages. Those pages are zero now, and 1289 * might as well be placed directly into the zero queue. 1290 */ 1291 if (vm_page_sleep_busy(p, FALSE, "pmaprl")) 1292 return 0; 1293 1294 vm_page_busy(p); 1295 1296 /* 1297 * Remove the page table page from the processes address space. 1298 */ 1299 pde[p->pindex] = 0; 1300 pmap->pm_stats.resident_count--; 1301 1302#if !defined(MAX_PERF) 1303 if (p->hold_count) { 1304 panic("pmap_release: freeing held page table page"); 1305 } 1306#endif 1307 /* 1308 * Page directory pages need to have the kernel 1309 * stuff cleared, so they can go into the zero queue also. 1310 */ 1311 if (p->pindex == PTDPTDI) { 1312 bzero(pde + KPTDI, nkpt * PTESIZE); 1313#ifdef SMP 1314 pde[MPPTDI] = 0; 1315#endif 1316 pde[APTDPTDI] = 0; 1317 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1318 } 1319 1320 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1321 pmap->pm_ptphint = NULL; 1322 1323 p->wire_count--; 1324 cnt.v_wire_count--; 1325 vm_page_free_zero(p); 1326 return 1; 1327} 1328 1329/* 1330 * this routine is called if the page table page is not 1331 * mapped correctly. 1332 */ 1333static vm_page_t 1334_pmap_allocpte(pmap, ptepindex) 1335 pmap_t pmap; 1336 unsigned ptepindex; 1337{ 1338 vm_offset_t pteva, ptepa; 1339 vm_page_t m; 1340 1341 /* 1342 * Find or fabricate a new pagetable page 1343 */ 1344 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1345 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1346 1347 if (m->queue != PQ_NONE) { 1348 int s = splvm(); 1349 vm_page_unqueue(m); 1350 splx(s); 1351 } 1352 1353 if (m->wire_count == 0) 1354 cnt.v_wire_count++; 1355 m->wire_count++; 1356 1357 /* 1358 * Increment the hold count for the page table page 1359 * (denoting a new mapping.) 1360 */ 1361 m->hold_count++; 1362 1363 /* 1364 * Map the pagetable page into the process address space, if 1365 * it isn't already there. 1366 */ 1367 1368 pmap->pm_stats.resident_count++; 1369 1370 ptepa = VM_PAGE_TO_PHYS(m); 1371 pmap->pm_pdir[ptepindex] = 1372 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1373 1374 /* 1375 * Set the page table hint 1376 */ 1377 pmap->pm_ptphint = m; 1378 1379 /* 1380 * Try to use the new mapping, but if we cannot, then 1381 * do it with the routine that maps the page explicitly. 1382 */ 1383 if ((m->flags & PG_ZERO) == 0) { 1384 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1385 (((unsigned) PTDpde) & PG_FRAME)) { 1386 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1387 bzero((caddr_t) pteva, PAGE_SIZE); 1388 } else { 1389 pmap_zero_page(ptepa); 1390 } 1391 } 1392 1393 m->valid = VM_PAGE_BITS_ALL; 1394 vm_page_flag_clear(m, PG_ZERO); 1395 vm_page_flag_set(m, PG_MAPPED); 1396 vm_page_wakeup(m); 1397 1398 return m; 1399} 1400 1401static vm_page_t 1402pmap_allocpte(pmap, va) 1403 pmap_t pmap; 1404 vm_offset_t va; 1405{ 1406 unsigned ptepindex; 1407 vm_offset_t ptepa; 1408 vm_page_t m; 1409 1410 /* 1411 * Calculate pagetable page index 1412 */ 1413 ptepindex = va >> PDRSHIFT; 1414 1415 /* 1416 * Get the page directory entry 1417 */ 1418 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1419 1420 /* 1421 * This supports switching from a 4MB page to a 1422 * normal 4K page. 1423 */ 1424 if (ptepa & PG_PS) { 1425 pmap->pm_pdir[ptepindex] = 0; 1426 ptepa = 0; 1427 invltlb(); 1428 } 1429 1430 /* 1431 * If the page table page is mapped, we just increment the 1432 * hold count, and activate it. 1433 */ 1434 if (ptepa) { 1435 /* 1436 * In order to get the page table page, try the 1437 * hint first. 1438 */ 1439 if (pmap->pm_ptphint && 1440 (pmap->pm_ptphint->pindex == ptepindex)) { 1441 m = pmap->pm_ptphint; 1442 } else { 1443 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1444 pmap->pm_ptphint = m; 1445 } 1446 m->hold_count++; 1447 return m; 1448 } 1449 /* 1450 * Here if the pte page isn't mapped, or if it has been deallocated. 1451 */ 1452 return _pmap_allocpte(pmap, ptepindex); 1453} 1454 1455 1456/*************************************************** 1457* Pmap allocation/deallocation routines. 1458 ***************************************************/ 1459 1460/* 1461 * Release any resources held by the given physical map. 1462 * Called when a pmap initialized by pmap_pinit is being released. 1463 * Should only be called if the map contains no valid mappings. 1464 */ 1465void 1466pmap_release(pmap) 1467 register struct pmap *pmap; 1468{ 1469 vm_page_t p,n,ptdpg; 1470 vm_object_t object = pmap->pm_pteobj; 1471 int curgeneration; 1472 1473#if defined(DIAGNOSTIC) 1474 if (object->ref_count != 1) 1475 panic("pmap_release: pteobj reference count != 1"); 1476#endif 1477 1478 ptdpg = NULL; 1479retry: 1480 curgeneration = object->generation; 1481 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1482 n = TAILQ_NEXT(p, listq); 1483 if (p->pindex == PTDPTDI) { 1484 ptdpg = p; 1485 continue; 1486 } 1487 while (1) { 1488 if (!pmap_release_free_page(pmap, p) && 1489 (object->generation != curgeneration)) 1490 goto retry; 1491 } 1492 } 1493 1494 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1495 goto retry; 1496} 1497 1498/* 1499 * grow the number of kernel page table entries, if needed 1500 */ 1501void 1502pmap_growkernel(vm_offset_t addr) 1503{ 1504 struct proc *p; 1505 struct pmap *pmap; 1506 int s; 1507 vm_offset_t ptppaddr; 1508 vm_page_t nkpg; 1509#ifdef SMP 1510 int i; 1511#endif 1512 pd_entry_t newpdir; 1513 1514 s = splhigh(); 1515 if (kernel_vm_end == 0) { 1516 kernel_vm_end = KERNBASE; 1517 nkpt = 0; 1518 while (pdir_pde(PTD, kernel_vm_end)) { 1519 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1520 nkpt++; 1521 } 1522 } 1523 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1524 while (kernel_vm_end < addr) { 1525 if (pdir_pde(PTD, kernel_vm_end)) { 1526 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1527 continue; 1528 } 1529 1530 /* 1531 * This index is bogus, but out of the way 1532 */ 1533 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1534#if !defined(MAX_PERF) 1535 if (!nkpg) 1536 panic("pmap_growkernel: no memory to grow kernel"); 1537#endif 1538 1539 nkpt++; 1540 1541 vm_page_wire(nkpg); 1542 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1543 pmap_zero_page(ptppaddr); 1544 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1545 pdir_pde(PTD, kernel_vm_end) = newpdir; 1546 1547#ifdef SMP 1548 for (i = 0; i < mp_ncpus; i++) { 1549 if (IdlePTDS[i]) 1550 pdir_pde(IdlePTDS[i], kernel_vm_end) = newpdir; 1551 } 1552#endif 1553 1554 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1555 if (p->p_vmspace) { 1556 pmap = &p->p_vmspace->vm_pmap; 1557 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1558 } 1559 } 1560 if (aiovmspace != NULL) { 1561 pmap = &aiovmspace->vm_pmap; 1562 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1563 } 1564 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir; 1565 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1566 } 1567 splx(s); 1568} 1569 1570/* 1571 * Retire the given physical map from service. 1572 * Should only be called if the map contains 1573 * no valid mappings. 1574 */ 1575void 1576pmap_destroy(pmap) 1577 register pmap_t pmap; 1578{ 1579 int count; 1580 1581 if (pmap == NULL) 1582 return; 1583 1584 count = --pmap->pm_count; 1585 if (count == 0) { 1586 pmap_release(pmap); 1587#if !defined(MAX_PERF) 1588 panic("destroying a pmap is not yet implemented"); 1589#endif 1590 } 1591} 1592 1593/* 1594 * Add a reference to the specified pmap. 1595 */ 1596void 1597pmap_reference(pmap) 1598 pmap_t pmap; 1599{ 1600 if (pmap != NULL) { 1601 pmap->pm_count++; 1602 } 1603} 1604 1605/*************************************************** 1606* page management routines. 1607 ***************************************************/ 1608 1609/* 1610 * free the pv_entry back to the free list 1611 */ 1612static PMAP_INLINE void 1613free_pv_entry(pv) 1614 pv_entry_t pv; 1615{ 1616 pv_entry_count--; 1617 zfreei(pvzone, pv); 1618} 1619 1620/* 1621 * get a new pv_entry, allocating a block from the system 1622 * when needed. 1623 * the memory allocation is performed bypassing the malloc code 1624 * because of the possibility of allocations at interrupt time. 1625 */ 1626static pv_entry_t 1627get_pv_entry(void) 1628{ 1629 pv_entry_count++; 1630 if (pv_entry_high_water && 1631 (pv_entry_count > pv_entry_high_water) && 1632 (pmap_pagedaemon_waken == 0)) { 1633 pmap_pagedaemon_waken = 1; 1634 wakeup (&vm_pages_needed); 1635 } 1636 return zalloci(pvzone); 1637} 1638 1639/* 1640 * This routine is very drastic, but can save the system 1641 * in a pinch. 1642 */ 1643void 1644pmap_collect() { 1645 pv_table_t *ppv; 1646 int i; 1647 vm_offset_t pa; 1648 vm_page_t m; 1649 static int warningdone=0; 1650 1651 if (pmap_pagedaemon_waken == 0) 1652 return; 1653 1654 if (warningdone < 5) { 1655 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1656 warningdone++; 1657 } 1658 1659 for(i = 0; i < pv_npg; i++) { 1660 if ((ppv = &pv_table[i]) == 0) 1661 continue; 1662 m = ppv->pv_vm_page; 1663 if ((pa = VM_PAGE_TO_PHYS(m)) == 0) 1664 continue; 1665 if (m->wire_count || m->hold_count || m->busy || 1666 (m->flags & PG_BUSY)) 1667 continue; 1668 pmap_remove_all(pa); 1669 } 1670 pmap_pagedaemon_waken = 0; 1671} 1672 1673 1674/* 1675 * If it is the first entry on the list, it is actually 1676 * in the header and we must copy the following entry up 1677 * to the header. Otherwise we must search the list for 1678 * the entry. In either case we free the now unused entry. 1679 */ 1680 1681static int 1682pmap_remove_entry(pmap, ppv, va) 1683 struct pmap *pmap; 1684 pv_table_t *ppv; 1685 vm_offset_t va; 1686{ 1687 pv_entry_t pv; 1688 int rtval; 1689 int s; 1690 1691 s = splvm(); 1692 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1693 for (pv = TAILQ_FIRST(&ppv->pv_list); 1694 pv; 1695 pv = TAILQ_NEXT(pv, pv_list)) { 1696 if (pmap == pv->pv_pmap && va == pv->pv_va) 1697 break; 1698 } 1699 } else { 1700 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1701 pv; 1702 pv = TAILQ_NEXT(pv, pv_plist)) { 1703 if (va == pv->pv_va) 1704 break; 1705 } 1706 } 1707 1708 rtval = 0; 1709 if (pv) { 1710 1711 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1712 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1713 ppv->pv_list_count--; 1714 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 1715 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 1716 1717 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1718 free_pv_entry(pv); 1719 } 1720 1721 splx(s); 1722 return rtval; 1723} 1724 1725/* 1726 * Create a pv entry for page at pa for 1727 * (pmap, va). 1728 */ 1729static void 1730pmap_insert_entry(pmap, va, mpte, pa) 1731 pmap_t pmap; 1732 vm_offset_t va; 1733 vm_page_t mpte; 1734 vm_offset_t pa; 1735{ 1736 1737 int s; 1738 pv_entry_t pv; 1739 pv_table_t *ppv; 1740 1741 s = splvm(); 1742 pv = get_pv_entry(); 1743 pv->pv_va = va; 1744 pv->pv_pmap = pmap; 1745 pv->pv_ptem = mpte; 1746 1747 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1748 1749 ppv = pa_to_pvh(pa); 1750 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1751 ppv->pv_list_count++; 1752 1753 splx(s); 1754} 1755 1756/* 1757 * pmap_remove_pte: do the things to unmap a page in a process 1758 */ 1759static int 1760pmap_remove_pte(pmap, ptq, va) 1761 struct pmap *pmap; 1762 unsigned *ptq; 1763 vm_offset_t va; 1764{ 1765 unsigned oldpte; 1766 pv_table_t *ppv; 1767 1768 oldpte = *ptq; 1769 *ptq = 0; 1770 if (oldpte & PG_W) 1771 pmap->pm_stats.wired_count -= 1; 1772 /* 1773 * Machines that don't support invlpg, also don't support 1774 * PG_G. 1775 */ 1776 if (oldpte & PG_G) 1777 invlpg(va); 1778 pmap->pm_stats.resident_count -= 1; 1779 if (oldpte & PG_MANAGED) { 1780 ppv = pa_to_pvh(oldpte); 1781 if (oldpte & PG_M) { 1782#if defined(PMAP_DIAGNOSTIC) 1783 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1784 printf( 1785 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n", 1786 va, oldpte); 1787 } 1788#endif 1789 if (pmap_track_modified(va)) 1790 vm_page_dirty(ppv->pv_vm_page); 1791 } 1792 if (oldpte & PG_A) 1793 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1794 return pmap_remove_entry(pmap, ppv, va); 1795 } else { 1796 return pmap_unuse_pt(pmap, va, NULL); 1797 } 1798 1799 return 0; 1800} 1801 1802/* 1803 * Remove a single page from a process address space 1804 */ 1805static void 1806pmap_remove_page(pmap, va) 1807 struct pmap *pmap; 1808 register vm_offset_t va; 1809{ 1810 register unsigned *ptq; 1811 1812 /* 1813 * if there is no pte for this address, just skip it!!! 1814 */ 1815 if (*pmap_pde(pmap, va) == 0) { 1816 return; 1817 } 1818 1819 /* 1820 * get a local va for mappings for this pmap. 1821 */ 1822 ptq = get_ptbase(pmap) + i386_btop(va); 1823 if (*ptq) { 1824 (void) pmap_remove_pte(pmap, ptq, va); 1825 invltlb_1pg(va); 1826 } 1827 return; 1828} 1829 1830/* 1831 * Remove the given range of addresses from the specified map. 1832 * 1833 * It is assumed that the start and end are properly 1834 * rounded to the page size. 1835 */ 1836void 1837pmap_remove(pmap, sva, eva) 1838 struct pmap *pmap; 1839 register vm_offset_t sva; 1840 register vm_offset_t eva; 1841{ 1842 register unsigned *ptbase; 1843 vm_offset_t pdnxt; 1844 vm_offset_t ptpaddr; 1845 vm_offset_t sindex, eindex; 1846 int anyvalid; 1847 1848 if (pmap == NULL) 1849 return; 1850 1851 if (pmap->pm_stats.resident_count == 0) 1852 return; 1853 1854 /* 1855 * special handling of removing one page. a very 1856 * common operation and easy to short circuit some 1857 * code. 1858 */ 1859 if (((sva + PAGE_SIZE) == eva) && 1860 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1861 pmap_remove_page(pmap, sva); 1862 return; 1863 } 1864 1865 anyvalid = 0; 1866 1867 /* 1868 * Get a local virtual address for the mappings that are being 1869 * worked with. 1870 */ 1871 ptbase = get_ptbase(pmap); 1872 1873 sindex = i386_btop(sva); 1874 eindex = i386_btop(eva); 1875 1876 for (; sindex < eindex; sindex = pdnxt) { 1877 unsigned pdirindex; 1878 1879 /* 1880 * Calculate index for next page table. 1881 */ 1882 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1883 if (pmap->pm_stats.resident_count == 0) 1884 break; 1885 1886 pdirindex = sindex / NPDEPG; 1887 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1888 pmap->pm_pdir[pdirindex] = 0; 1889 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1890 anyvalid++; 1891 continue; 1892 } 1893 1894 /* 1895 * Weed out invalid mappings. Note: we assume that the page 1896 * directory table is always allocated, and in kernel virtual. 1897 */ 1898 if (ptpaddr == 0) 1899 continue; 1900 1901 /* 1902 * Limit our scan to either the end of the va represented 1903 * by the current page table page, or to the end of the 1904 * range being removed. 1905 */ 1906 if (pdnxt > eindex) { 1907 pdnxt = eindex; 1908 } 1909 1910 for ( ;sindex != pdnxt; sindex++) { 1911 vm_offset_t va; 1912 if (ptbase[sindex] == 0) { 1913 continue; 1914 } 1915 va = i386_ptob(sindex); 1916 1917 anyvalid++; 1918 if (pmap_remove_pte(pmap, 1919 ptbase + sindex, va)) 1920 break; 1921 } 1922 } 1923 1924 if (anyvalid) { 1925 invltlb(); 1926 } 1927} 1928 1929/* 1930 * Routine: pmap_remove_all 1931 * Function: 1932 * Removes this physical page from 1933 * all physical maps in which it resides. 1934 * Reflects back modify bits to the pager. 1935 * 1936 * Notes: 1937 * Original versions of this routine were very 1938 * inefficient because they iteratively called 1939 * pmap_remove (slow...) 1940 */ 1941 1942static void 1943pmap_remove_all(pa) 1944 vm_offset_t pa; 1945{ 1946 register pv_entry_t pv; 1947 pv_table_t *ppv; 1948 register unsigned *pte, tpte; 1949 int nmodify; 1950 int update_needed; 1951 int s; 1952 1953 nmodify = 0; 1954 update_needed = 0; 1955#if defined(PMAP_DIAGNOSTIC) 1956 /* 1957 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1958 * pages! 1959 */ 1960 if (!pmap_is_managed(pa)) { 1961 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", pa); 1962 } 1963#endif 1964 1965 s = splvm(); 1966 ppv = pa_to_pvh(pa); 1967 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1968 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1969 1970 pv->pv_pmap->pm_stats.resident_count--; 1971 1972 tpte = *pte; 1973 *pte = 0; 1974 if (tpte & PG_W) 1975 pv->pv_pmap->pm_stats.wired_count--; 1976 1977 if (tpte & PG_A) 1978 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 1979 1980 /* 1981 * Update the vm_page_t clean and reference bits. 1982 */ 1983 if (tpte & PG_M) { 1984#if defined(PMAP_DIAGNOSTIC) 1985 if (pmap_nw_modified((pt_entry_t) tpte)) { 1986 printf( 1987 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n", 1988 pv->pv_va, tpte); 1989 } 1990#endif 1991 if (pmap_track_modified(pv->pv_va)) 1992 vm_page_dirty(ppv->pv_vm_page); 1993 } 1994 if (!update_needed && 1995 ((!curproc || (&curproc->p_vmspace->vm_pmap == pv->pv_pmap)) || 1996 (pv->pv_pmap == kernel_pmap))) { 1997 update_needed = 1; 1998 } 1999 2000 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2001 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2002 ppv->pv_list_count--; 2003 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2004 free_pv_entry(pv); 2005 } 2006 2007 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 2008 2009 if (update_needed) 2010 invltlb(); 2011 2012 splx(s); 2013 return; 2014} 2015 2016/* 2017 * Set the physical protection on the 2018 * specified range of this map as requested. 2019 */ 2020void 2021pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 2022{ 2023 register unsigned *ptbase; 2024 vm_offset_t pdnxt, ptpaddr; 2025 vm_pindex_t sindex, eindex; 2026 int anychanged; 2027 2028 2029 if (pmap == NULL) 2030 return; 2031 2032 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 2033 pmap_remove(pmap, sva, eva); 2034 return; 2035 } 2036 2037 if (prot & VM_PROT_WRITE) 2038 return; 2039 2040 anychanged = 0; 2041 2042 ptbase = get_ptbase(pmap); 2043 2044 sindex = i386_btop(sva); 2045 eindex = i386_btop(eva); 2046 2047 for (; sindex < eindex; sindex = pdnxt) { 2048 2049 unsigned pdirindex; 2050 2051 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 2052 2053 pdirindex = sindex / NPDEPG; 2054 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 2055 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 2056 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 2057 anychanged++; 2058 continue; 2059 } 2060 2061 /* 2062 * Weed out invalid mappings. Note: we assume that the page 2063 * directory table is always allocated, and in kernel virtual. 2064 */ 2065 if (ptpaddr == 0) 2066 continue; 2067 2068 if (pdnxt > eindex) { 2069 pdnxt = eindex; 2070 } 2071 2072 for (; sindex != pdnxt; sindex++) { 2073 2074 unsigned pbits; 2075 pv_table_t *ppv; 2076 2077 pbits = ptbase[sindex]; 2078 2079 if (pbits & PG_MANAGED) { 2080 ppv = NULL; 2081 if (pbits & PG_A) { 2082 ppv = pa_to_pvh(pbits); 2083 vm_page_flag_set(ppv->pv_vm_page, PG_REFERENCED); 2084 pbits &= ~PG_A; 2085 } 2086 if (pbits & PG_M) { 2087 if (pmap_track_modified(i386_ptob(sindex))) { 2088 if (ppv == NULL) 2089 ppv = pa_to_pvh(pbits); 2090 vm_page_dirty(ppv->pv_vm_page); 2091 pbits &= ~PG_M; 2092 } 2093 } 2094 } 2095 2096 pbits &= ~PG_RW; 2097 2098 if (pbits != ptbase[sindex]) { 2099 ptbase[sindex] = pbits; 2100 anychanged = 1; 2101 } 2102 } 2103 } 2104 if (anychanged) 2105 invltlb(); 2106} 2107 2108/* 2109 * Insert the given physical page (p) at 2110 * the specified virtual address (v) in the 2111 * target physical map with the protection requested. 2112 * 2113 * If specified, the page will be wired down, meaning 2114 * that the related pte can not be reclaimed. 2115 * 2116 * NB: This is the only routine which MAY NOT lazy-evaluate 2117 * or lose information. That is, this routine must actually 2118 * insert this page into the given map NOW. 2119 */ 2120void 2121pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot, 2122 boolean_t wired) 2123{ 2124 register unsigned *pte; 2125 vm_offset_t opa; 2126 vm_offset_t origpte, newpte; 2127 vm_page_t mpte; 2128 2129 if (pmap == NULL) 2130 return; 2131 2132 va &= PG_FRAME; 2133#ifdef PMAP_DIAGNOSTIC 2134 if (va > VM_MAX_KERNEL_ADDRESS) 2135 panic("pmap_enter: toobig"); 2136 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 2137 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 2138#endif 2139 2140 mpte = NULL; 2141 /* 2142 * In the case that a page table page is not 2143 * resident, we are creating it here. 2144 */ 2145 if (va < UPT_MIN_ADDRESS) { 2146 mpte = pmap_allocpte(pmap, va); 2147 } 2148#if 0 && defined(PMAP_DIAGNOSTIC) 2149 else { 2150 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 2151 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 2152 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 2153 pmap->pm_pdir[PTDPTDI], origpte, va); 2154 } 2155 if (smp_active) { 2156 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid]; 2157 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2158 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2159 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2160 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr); 2161 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2162 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2163 } 2164 } 2165 } 2166#endif 2167 2168 pte = pmap_pte(pmap, va); 2169 2170#if !defined(MAX_PERF) 2171 /* 2172 * Page Directory table entry not valid, we need a new PT page 2173 */ 2174 if (pte == NULL) { 2175 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n", 2176 (void *)pmap->pm_pdir[PTDPTDI], va); 2177 } 2178#endif 2179 2180 origpte = *(vm_offset_t *)pte; 2181 pa &= PG_FRAME; 2182 opa = origpte & PG_FRAME; 2183 2184#if !defined(MAX_PERF) 2185 if (origpte & PG_PS) 2186 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2187#endif 2188 2189 /* 2190 * Mapping has not changed, must be protection or wiring change. 2191 */ 2192 if (origpte && (opa == pa)) { 2193 /* 2194 * Wiring change, just update stats. We don't worry about 2195 * wiring PT pages as they remain resident as long as there 2196 * are valid mappings in them. Hence, if a user page is wired, 2197 * the PT page will be also. 2198 */ 2199 if (wired && ((origpte & PG_W) == 0)) 2200 pmap->pm_stats.wired_count++; 2201 else if (!wired && (origpte & PG_W)) 2202 pmap->pm_stats.wired_count--; 2203 2204#if defined(PMAP_DIAGNOSTIC) 2205 if (pmap_nw_modified((pt_entry_t) origpte)) { 2206 printf( 2207 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n", 2208 va, origpte); 2209 } 2210#endif 2211 2212 /* 2213 * Remove extra pte reference 2214 */ 2215 if (mpte) 2216 mpte->hold_count--; 2217 2218 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2219 if ((origpte & PG_RW) == 0) { 2220 *pte |= PG_RW; 2221 invltlb_1pg(va); 2222 } 2223 return; 2224 } 2225 2226 /* 2227 * We might be turning off write access to the page, 2228 * so we go ahead and sense modify status. 2229 */ 2230 if (origpte & PG_MANAGED) { 2231 if ((origpte & PG_M) && pmap_track_modified(va)) { 2232 pv_table_t *ppv; 2233 ppv = pa_to_pvh(opa); 2234 vm_page_dirty(ppv->pv_vm_page); 2235 } 2236 pa |= PG_MANAGED; 2237 } 2238 goto validate; 2239 } 2240 /* 2241 * Mapping has changed, invalidate old range and fall through to 2242 * handle validating new mapping. 2243 */ 2244 if (opa) { 2245 int err; 2246 err = pmap_remove_pte(pmap, pte, va); 2247#if !defined(MAX_PERF) 2248 if (err) 2249 panic("pmap_enter: pte vanished, va: 0x%x", va); 2250#endif 2251 } 2252 2253 /* 2254 * Enter on the PV list if part of our managed memory Note that we 2255 * raise IPL while manipulating pv_table since pmap_enter can be 2256 * called at interrupt time. 2257 */ 2258 if (pmap_is_managed(pa)) { 2259 pmap_insert_entry(pmap, va, mpte, pa); 2260 pa |= PG_MANAGED; 2261 } 2262 2263 /* 2264 * Increment counters 2265 */ 2266 pmap->pm_stats.resident_count++; 2267 if (wired) 2268 pmap->pm_stats.wired_count++; 2269 2270validate: 2271 /* 2272 * Now validate mapping with desired protection/wiring. 2273 */ 2274 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2275 2276 if (wired) 2277 newpte |= PG_W; 2278 if (va < UPT_MIN_ADDRESS) 2279 newpte |= PG_U; 2280 if (pmap == kernel_pmap) 2281 newpte |= pgeflag; 2282 2283 /* 2284 * if the mapping or permission bits are different, we need 2285 * to update the pte. 2286 */ 2287 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2288 *pte = newpte | PG_A; 2289 if (origpte) 2290 invltlb_1pg(va); 2291 } 2292} 2293 2294/* 2295 * this code makes some *MAJOR* assumptions: 2296 * 1. Current pmap & pmap exists. 2297 * 2. Not wired. 2298 * 3. Read access. 2299 * 4. No page table pages. 2300 * 5. Tlbflush is deferred to calling procedure. 2301 * 6. Page IS managed. 2302 * but is *MUCH* faster than pmap_enter... 2303 */ 2304 2305static vm_page_t 2306pmap_enter_quick(pmap, va, pa, mpte) 2307 register pmap_t pmap; 2308 vm_offset_t va; 2309 register vm_offset_t pa; 2310 vm_page_t mpte; 2311{ 2312 register unsigned *pte; 2313 2314 /* 2315 * In the case that a page table page is not 2316 * resident, we are creating it here. 2317 */ 2318 if (va < UPT_MIN_ADDRESS) { 2319 unsigned ptepindex; 2320 vm_offset_t ptepa; 2321 2322 /* 2323 * Calculate pagetable page index 2324 */ 2325 ptepindex = va >> PDRSHIFT; 2326 if (mpte && (mpte->pindex == ptepindex)) { 2327 mpte->hold_count++; 2328 } else { 2329retry: 2330 /* 2331 * Get the page directory entry 2332 */ 2333 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2334 2335 /* 2336 * If the page table page is mapped, we just increment 2337 * the hold count, and activate it. 2338 */ 2339 if (ptepa) { 2340#if !defined(MAX_PERF) 2341 if (ptepa & PG_PS) 2342 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2343#endif 2344 if (pmap->pm_ptphint && 2345 (pmap->pm_ptphint->pindex == ptepindex)) { 2346 mpte = pmap->pm_ptphint; 2347 } else { 2348 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2349 pmap->pm_ptphint = mpte; 2350 } 2351 if (mpte == NULL) 2352 goto retry; 2353 mpte->hold_count++; 2354 } else { 2355 mpte = _pmap_allocpte(pmap, ptepindex); 2356 } 2357 } 2358 } else { 2359 mpte = NULL; 2360 } 2361 2362 /* 2363 * This call to vtopte makes the assumption that we are 2364 * entering the page into the current pmap. In order to support 2365 * quick entry into any pmap, one would likely use pmap_pte_quick. 2366 * But that isn't as quick as vtopte. 2367 */ 2368 pte = (unsigned *)vtopte(va); 2369 if (*pte) { 2370 if (mpte) 2371 pmap_unwire_pte_hold(pmap, mpte); 2372 return 0; 2373 } 2374 2375 /* 2376 * Enter on the PV list if part of our managed memory Note that we 2377 * raise IPL while manipulating pv_table since pmap_enter can be 2378 * called at interrupt time. 2379 */ 2380 pmap_insert_entry(pmap, va, mpte, pa); 2381 2382 /* 2383 * Increment counters 2384 */ 2385 pmap->pm_stats.resident_count++; 2386 2387 /* 2388 * Now validate mapping with RO protection 2389 */ 2390 *pte = pa | PG_V | PG_U | PG_MANAGED; 2391 2392 return mpte; 2393} 2394 2395#define MAX_INIT_PT (96) 2396/* 2397 * pmap_object_init_pt preloads the ptes for a given object 2398 * into the specified pmap. This eliminates the blast of soft 2399 * faults on process startup and immediately after an mmap. 2400 */ 2401void 2402pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2403 pmap_t pmap; 2404 vm_offset_t addr; 2405 vm_object_t object; 2406 vm_pindex_t pindex; 2407 vm_size_t size; 2408 int limit; 2409{ 2410 vm_offset_t tmpidx; 2411 int psize; 2412 vm_page_t p, mpte; 2413 int objpgs; 2414 2415 if (!pmap) 2416 return; 2417 2418 /* 2419 * This code maps large physical mmap regions into the 2420 * processor address space. Note that some shortcuts 2421 * are taken, but the code works. 2422 */ 2423 if (pseflag && 2424 (object->type == OBJT_DEVICE) && 2425 ((addr & (NBPDR - 1)) == 0) && 2426 ((size & (NBPDR - 1)) == 0) ) { 2427 int i; 2428 vm_page_t m[1]; 2429 unsigned int ptepindex; 2430 int npdes; 2431 vm_offset_t ptepa; 2432 2433 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2434 return; 2435 2436retry: 2437 p = vm_page_lookup(object, pindex); 2438 if (p && vm_page_sleep_busy(p, FALSE, "init4p")) 2439 goto retry; 2440 2441 if (p == NULL) { 2442 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2443 if (p == NULL) 2444 return; 2445 m[0] = p; 2446 2447 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2448 vm_page_free(p); 2449 return; 2450 } 2451 2452 p = vm_page_lookup(object, pindex); 2453 vm_page_wakeup(p); 2454 } 2455 2456 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2457 if (ptepa & (NBPDR - 1)) { 2458 return; 2459 } 2460 2461 p->valid = VM_PAGE_BITS_ALL; 2462 2463 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2464 npdes = size >> PDRSHIFT; 2465 for(i=0;i<npdes;i++) { 2466 pmap->pm_pdir[ptepindex] = 2467 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2468 ptepa += NBPDR; 2469 ptepindex += 1; 2470 } 2471 vm_page_flag_set(p, PG_MAPPED); 2472 invltlb(); 2473 return; 2474 } 2475 2476 psize = i386_btop(size); 2477 2478 if ((object->type != OBJT_VNODE) || 2479 (limit && (psize > MAX_INIT_PT) && 2480 (object->resident_page_count > MAX_INIT_PT))) { 2481 return; 2482 } 2483 2484 if (psize + pindex > object->size) 2485 psize = object->size - pindex; 2486 2487 mpte = NULL; 2488 /* 2489 * if we are processing a major portion of the object, then scan the 2490 * entire thing. 2491 */ 2492 if (psize > (object->size >> 2)) { 2493 objpgs = psize; 2494 2495 for (p = TAILQ_FIRST(&object->memq); 2496 ((objpgs > 0) && (p != NULL)); 2497 p = TAILQ_NEXT(p, listq)) { 2498 2499 tmpidx = p->pindex; 2500 if (tmpidx < pindex) { 2501 continue; 2502 } 2503 tmpidx -= pindex; 2504 if (tmpidx >= psize) { 2505 continue; 2506 } 2507 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2508 (p->busy == 0) && 2509 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2510 if ((p->queue - p->pc) == PQ_CACHE) 2511 vm_page_deactivate(p); 2512 vm_page_busy(p); 2513 mpte = pmap_enter_quick(pmap, 2514 addr + i386_ptob(tmpidx), 2515 VM_PAGE_TO_PHYS(p), mpte); 2516 vm_page_flag_set(p, PG_MAPPED); 2517 vm_page_wakeup(p); 2518 } 2519 objpgs -= 1; 2520 } 2521 } else { 2522 /* 2523 * else lookup the pages one-by-one. 2524 */ 2525 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2526 p = vm_page_lookup(object, tmpidx + pindex); 2527 if (p && 2528 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2529 (p->busy == 0) && 2530 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2531 if ((p->queue - p->pc) == PQ_CACHE) 2532 vm_page_deactivate(p); 2533 vm_page_busy(p); 2534 mpte = pmap_enter_quick(pmap, 2535 addr + i386_ptob(tmpidx), 2536 VM_PAGE_TO_PHYS(p), mpte); 2537 vm_page_flag_set(p, PG_MAPPED); 2538 vm_page_wakeup(p); 2539 } 2540 } 2541 } 2542 return; 2543} 2544 2545/* 2546 * pmap_prefault provides a quick way of clustering 2547 * pagefaults into a processes address space. It is a "cousin" 2548 * of pmap_object_init_pt, except it runs at page fault time instead 2549 * of mmap time. 2550 */ 2551#define PFBAK 4 2552#define PFFOR 4 2553#define PAGEORDER_SIZE (PFBAK+PFFOR) 2554 2555static int pmap_prefault_pageorder[] = { 2556 -PAGE_SIZE, PAGE_SIZE, 2557 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2558 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2559 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2560}; 2561 2562void 2563pmap_prefault(pmap, addra, entry) 2564 pmap_t pmap; 2565 vm_offset_t addra; 2566 vm_map_entry_t entry; 2567{ 2568 int i; 2569 vm_offset_t starta; 2570 vm_offset_t addr; 2571 vm_pindex_t pindex; 2572 vm_page_t m, mpte; 2573 vm_object_t object; 2574 2575 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) 2576 return; 2577 2578 object = entry->object.vm_object; 2579 2580 starta = addra - PFBAK * PAGE_SIZE; 2581 if (starta < entry->start) { 2582 starta = entry->start; 2583 } else if (starta > addra) { 2584 starta = 0; 2585 } 2586 2587 mpte = NULL; 2588 for (i = 0; i < PAGEORDER_SIZE; i++) { 2589 vm_object_t lobject; 2590 unsigned *pte; 2591 2592 addr = addra + pmap_prefault_pageorder[i]; 2593 if (addr > addra + (PFFOR * PAGE_SIZE)) 2594 addr = 0; 2595 2596 if (addr < starta || addr >= entry->end) 2597 continue; 2598 2599 if ((*pmap_pde(pmap, addr)) == NULL) 2600 continue; 2601 2602 pte = (unsigned *) vtopte(addr); 2603 if (*pte) 2604 continue; 2605 2606 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2607 lobject = object; 2608 for (m = vm_page_lookup(lobject, pindex); 2609 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2610 lobject = lobject->backing_object) { 2611 if (lobject->backing_object_offset & PAGE_MASK) 2612 break; 2613 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2614 m = vm_page_lookup(lobject->backing_object, pindex); 2615 } 2616 2617 /* 2618 * give-up when a page is not in memory 2619 */ 2620 if (m == NULL) 2621 break; 2622 2623 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2624 (m->busy == 0) && 2625 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2626 2627 if ((m->queue - m->pc) == PQ_CACHE) { 2628 vm_page_deactivate(m); 2629 } 2630 vm_page_busy(m); 2631 mpte = pmap_enter_quick(pmap, addr, 2632 VM_PAGE_TO_PHYS(m), mpte); 2633 vm_page_flag_set(m, PG_MAPPED); 2634 vm_page_wakeup(m); 2635 } 2636 } 2637} 2638 2639/* 2640 * Routine: pmap_change_wiring 2641 * Function: Change the wiring attribute for a map/virtual-address 2642 * pair. 2643 * In/out conditions: 2644 * The mapping must already exist in the pmap. 2645 */ 2646void 2647pmap_change_wiring(pmap, va, wired) 2648 register pmap_t pmap; 2649 vm_offset_t va; 2650 boolean_t wired; 2651{ 2652 register unsigned *pte; 2653 2654 if (pmap == NULL) 2655 return; 2656 2657 pte = pmap_pte(pmap, va); 2658 2659 if (wired && !pmap_pte_w(pte)) 2660 pmap->pm_stats.wired_count++; 2661 else if (!wired && pmap_pte_w(pte)) 2662 pmap->pm_stats.wired_count--; 2663 2664 /* 2665 * Wiring is not a hardware characteristic so there is no need to 2666 * invalidate TLB. 2667 */ 2668 pmap_pte_set_w(pte, wired); 2669} 2670 2671 2672 2673/* 2674 * Copy the range specified by src_addr/len 2675 * from the source map to the range dst_addr/len 2676 * in the destination map. 2677 * 2678 * This routine is only advisory and need not do anything. 2679 */ 2680 2681void 2682pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2683 pmap_t dst_pmap, src_pmap; 2684 vm_offset_t dst_addr; 2685 vm_size_t len; 2686 vm_offset_t src_addr; 2687{ 2688 vm_offset_t addr; 2689 vm_offset_t end_addr = src_addr + len; 2690 vm_offset_t pdnxt; 2691 unsigned src_frame, dst_frame; 2692 2693 if (dst_addr != src_addr) 2694 return; 2695 2696 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2697 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2698 return; 2699 } 2700 2701 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2702 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2703 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2704 invltlb(); 2705 } 2706 2707 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2708 unsigned *src_pte, *dst_pte; 2709 vm_page_t dstmpte, srcmpte; 2710 vm_offset_t srcptepaddr; 2711 unsigned ptepindex; 2712 2713#if !defined(MAX_PERF) 2714 if (addr >= UPT_MIN_ADDRESS) 2715 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2716#endif 2717 2718 /* 2719 * Don't let optional prefaulting of pages make us go 2720 * way below the low water mark of free pages or way 2721 * above high water mark of used pv entries. 2722 */ 2723 if (cnt.v_free_count < cnt.v_free_reserved || 2724 pv_entry_count > pv_entry_high_water) 2725 break; 2726 2727 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2728 ptepindex = addr >> PDRSHIFT; 2729 2730 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2731 if (srcptepaddr == 0) 2732 continue; 2733 2734 if (srcptepaddr & PG_PS) { 2735 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2736 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2737 dst_pmap->pm_stats.resident_count += NBPDR; 2738 } 2739 continue; 2740 } 2741 2742 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2743 if ((srcmpte == NULL) || 2744 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2745 continue; 2746 2747 if (pdnxt > end_addr) 2748 pdnxt = end_addr; 2749 2750 src_pte = (unsigned *) vtopte(addr); 2751 dst_pte = (unsigned *) avtopte(addr); 2752 while (addr < pdnxt) { 2753 unsigned ptetemp; 2754 ptetemp = *src_pte; 2755 /* 2756 * we only virtual copy managed pages 2757 */ 2758 if ((ptetemp & PG_MANAGED) != 0) { 2759 /* 2760 * We have to check after allocpte for the 2761 * pte still being around... allocpte can 2762 * block. 2763 */ 2764 dstmpte = pmap_allocpte(dst_pmap, addr); 2765 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2766 /* 2767 * Clear the modified and 2768 * accessed (referenced) bits 2769 * during the copy. 2770 */ 2771 *dst_pte = ptetemp & ~(PG_M | PG_A); 2772 dst_pmap->pm_stats.resident_count++; 2773 pmap_insert_entry(dst_pmap, addr, 2774 dstmpte, 2775 (ptetemp & PG_FRAME)); 2776 } else { 2777 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2778 } 2779 if (dstmpte->hold_count >= srcmpte->hold_count) 2780 break; 2781 } 2782 addr += PAGE_SIZE; 2783 src_pte++; 2784 dst_pte++; 2785 } 2786 } 2787} 2788 2789/* 2790 * Routine: pmap_kernel 2791 * Function: 2792 * Returns the physical map handle for the kernel. 2793 */ 2794pmap_t 2795pmap_kernel() 2796{ 2797 return (kernel_pmap); 2798} 2799 2800/* 2801 * pmap_zero_page zeros the specified (machine independent) 2802 * page by mapping the page into virtual memory and using 2803 * bzero to clear its contents, one machine dependent page 2804 * at a time. 2805 */ 2806void 2807pmap_zero_page(phys) 2808 vm_offset_t phys; 2809{ 2810#ifdef SMP 2811#if !defined(MAX_PERF) 2812 if (*(int *) prv_CMAP3) 2813 panic("pmap_zero_page: prv_CMAP3 busy"); 2814#endif 2815 2816 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2817 cpu_invlpg(&prv_CPAGE3); 2818 2819#if defined(I686_CPU) 2820 if (cpu_class == CPUCLASS_686) 2821 i686_pagezero(&prv_CPAGE3); 2822 else 2823#endif 2824 bzero(&prv_CPAGE3, PAGE_SIZE); 2825 2826 *(int *) prv_CMAP3 = 0; 2827#else 2828#if !defined(MAX_PERF) 2829 if (*(int *) CMAP2) 2830 panic("pmap_zero_page: CMAP2 busy"); 2831#endif 2832 2833 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2834 if (cpu_class == CPUCLASS_386) { 2835 invltlb(); 2836 } else { 2837 invlpg((u_int)CADDR2); 2838 } 2839 2840#if defined(I686_CPU) 2841 if (cpu_class == CPUCLASS_686) 2842 i686_pagezero(CADDR2); 2843 else 2844#endif 2845 bzero(CADDR2, PAGE_SIZE); 2846 *(int *) CMAP2 = 0; 2847#endif 2848} 2849 2850/* 2851 * pmap_copy_page copies the specified (machine independent) 2852 * page by mapping the page into virtual memory and using 2853 * bcopy to copy the page, one machine dependent page at a 2854 * time. 2855 */ 2856void 2857pmap_copy_page(src, dst) 2858 vm_offset_t src; 2859 vm_offset_t dst; 2860{ 2861#ifdef SMP 2862#if !defined(MAX_PERF) 2863 if (*(int *) prv_CMAP1) 2864 panic("pmap_copy_page: prv_CMAP1 busy"); 2865 if (*(int *) prv_CMAP2) 2866 panic("pmap_copy_page: prv_CMAP2 busy"); 2867#endif 2868 2869 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2870 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2871 2872 cpu_invlpg(&prv_CPAGE1); 2873 cpu_invlpg(&prv_CPAGE2); 2874 2875 bcopy(&prv_CPAGE1, &prv_CPAGE2, PAGE_SIZE); 2876 2877 *(int *) prv_CMAP1 = 0; 2878 *(int *) prv_CMAP2 = 0; 2879#else 2880#if !defined(MAX_PERF) 2881 if (*(int *) CMAP1 || *(int *) CMAP2) 2882 panic("pmap_copy_page: CMAP busy"); 2883#endif 2884 2885 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2886 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2887 if (cpu_class == CPUCLASS_386) { 2888 invltlb(); 2889 } else { 2890 invlpg((u_int)CADDR1); 2891 invlpg((u_int)CADDR2); 2892 } 2893 2894 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2895 2896 *(int *) CMAP1 = 0; 2897 *(int *) CMAP2 = 0; 2898#endif 2899} 2900 2901 2902/* 2903 * Routine: pmap_pageable 2904 * Function: 2905 * Make the specified pages (by pmap, offset) 2906 * pageable (or not) as requested. 2907 * 2908 * A page which is not pageable may not take 2909 * a fault; therefore, its page table entry 2910 * must remain valid for the duration. 2911 * 2912 * This routine is merely advisory; pmap_enter 2913 * will specify that these pages are to be wired 2914 * down (or not) as appropriate. 2915 */ 2916void 2917pmap_pageable(pmap, sva, eva, pageable) 2918 pmap_t pmap; 2919 vm_offset_t sva, eva; 2920 boolean_t pageable; 2921{ 2922} 2923 2924/* 2925 * this routine returns true if a physical page resides 2926 * in the given pmap. 2927 */ 2928boolean_t 2929pmap_page_exists(pmap, pa) 2930 pmap_t pmap; 2931 vm_offset_t pa; 2932{ 2933 register pv_entry_t pv; 2934 pv_table_t *ppv; 2935 int s; 2936 2937 if (!pmap_is_managed(pa)) 2938 return FALSE; 2939 2940 s = splvm(); 2941 2942 ppv = pa_to_pvh(pa); 2943 /* 2944 * Not found, check current mappings returning immediately if found. 2945 */ 2946 for (pv = TAILQ_FIRST(&ppv->pv_list); 2947 pv; 2948 pv = TAILQ_NEXT(pv, pv_list)) { 2949 if (pv->pv_pmap == pmap) { 2950 splx(s); 2951 return TRUE; 2952 } 2953 } 2954 splx(s); 2955 return (FALSE); 2956} 2957 2958#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2959/* 2960 * Remove all pages from specified address space 2961 * this aids process exit speeds. Also, this code 2962 * is special cased for current process only, but 2963 * can have the more generic (and slightly slower) 2964 * mode enabled. This is much faster than pmap_remove 2965 * in the case of running down an entire address space. 2966 */ 2967void 2968pmap_remove_pages(pmap, sva, eva) 2969 pmap_t pmap; 2970 vm_offset_t sva, eva; 2971{ 2972 unsigned *pte, tpte; 2973 pv_table_t *ppv; 2974 pv_entry_t pv, npv; 2975 int s; 2976 2977#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2978 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) { 2979 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2980 return; 2981 } 2982#endif 2983 2984 s = splvm(); 2985 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2986 pv; 2987 pv = npv) { 2988 2989 if (pv->pv_va >= eva || pv->pv_va < sva) { 2990 npv = TAILQ_NEXT(pv, pv_plist); 2991 continue; 2992 } 2993 2994#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2995 pte = (unsigned *)vtopte(pv->pv_va); 2996#else 2997 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2998#endif 2999 tpte = *pte; 3000 3001/* 3002 * We cannot remove wired pages from a process' mapping at this time 3003 */ 3004 if (tpte & PG_W) { 3005 npv = TAILQ_NEXT(pv, pv_plist); 3006 continue; 3007 } 3008 *pte = 0; 3009 3010 ppv = pa_to_pvh(tpte); 3011 3012 pv->pv_pmap->pm_stats.resident_count--; 3013 3014 /* 3015 * Update the vm_page_t clean and reference bits. 3016 */ 3017 if (tpte & PG_M) { 3018 vm_page_dirty(ppv->pv_vm_page); 3019 } 3020 3021 3022 npv = TAILQ_NEXT(pv, pv_plist); 3023 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 3024 3025 ppv->pv_list_count--; 3026 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 3027 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 3028 vm_page_flag_clear(ppv->pv_vm_page, PG_MAPPED | PG_WRITEABLE); 3029 } 3030 3031 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 3032 free_pv_entry(pv); 3033 } 3034 splx(s); 3035 invltlb(); 3036} 3037 3038/* 3039 * pmap_testbit tests bits in pte's 3040 * note that the testbit/changebit routines are inline, 3041 * and a lot of things compile-time evaluate. 3042 */ 3043static boolean_t 3044pmap_testbit(pa, bit) 3045 register vm_offset_t pa; 3046 int bit; 3047{ 3048 register pv_entry_t pv; 3049 pv_table_t *ppv; 3050 unsigned *pte; 3051 int s; 3052 3053 if (!pmap_is_managed(pa)) 3054 return FALSE; 3055 3056 ppv = pa_to_pvh(pa); 3057 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 3058 return FALSE; 3059 3060 s = splvm(); 3061 3062 for (pv = TAILQ_FIRST(&ppv->pv_list); 3063 pv; 3064 pv = TAILQ_NEXT(pv, pv_list)) { 3065 3066 /* 3067 * if the bit being tested is the modified bit, then 3068 * mark clean_map and ptes as never 3069 * modified. 3070 */ 3071 if (bit & (PG_A|PG_M)) { 3072 if (!pmap_track_modified(pv->pv_va)) 3073 continue; 3074 } 3075 3076#if defined(PMAP_DIAGNOSTIC) 3077 if (!pv->pv_pmap) { 3078 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 3079 continue; 3080 } 3081#endif 3082 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3083 if (*pte & bit) { 3084 splx(s); 3085 return TRUE; 3086 } 3087 } 3088 splx(s); 3089 return (FALSE); 3090} 3091 3092/* 3093 * this routine is used to modify bits in ptes 3094 */ 3095static void 3096pmap_changebit(pa, bit, setem) 3097 vm_offset_t pa; 3098 int bit; 3099 boolean_t setem; 3100{ 3101 register pv_entry_t pv; 3102 pv_table_t *ppv; 3103 register unsigned *pte; 3104 int changed; 3105 int s; 3106 3107 if (!pmap_is_managed(pa)) 3108 return; 3109 3110 s = splvm(); 3111 changed = 0; 3112 ppv = pa_to_pvh(pa); 3113 3114 /* 3115 * Loop over all current mappings setting/clearing as appropos If 3116 * setting RO do we need to clear the VAC? 3117 */ 3118 for (pv = TAILQ_FIRST(&ppv->pv_list); 3119 pv; 3120 pv = TAILQ_NEXT(pv, pv_list)) { 3121 3122 /* 3123 * don't write protect pager mappings 3124 */ 3125 if (!setem && (bit == PG_RW)) { 3126 if (!pmap_track_modified(pv->pv_va)) 3127 continue; 3128 } 3129 3130#if defined(PMAP_DIAGNOSTIC) 3131 if (!pv->pv_pmap) { 3132 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 3133 continue; 3134 } 3135#endif 3136 3137 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3138 3139 if (setem) { 3140 *(int *)pte |= bit; 3141 changed = 1; 3142 } else { 3143 vm_offset_t pbits = *(vm_offset_t *)pte; 3144 if (pbits & bit) { 3145 changed = 1; 3146 if (bit == PG_RW) { 3147 if (pbits & PG_M) { 3148 vm_page_dirty(ppv->pv_vm_page); 3149 } 3150 *(int *)pte = pbits & ~(PG_M|PG_RW); 3151 } else { 3152 *(int *)pte = pbits & ~bit; 3153 } 3154 } 3155 } 3156 } 3157 splx(s); 3158 if (changed) 3159 invltlb(); 3160} 3161 3162/* 3163 * pmap_page_protect: 3164 * 3165 * Lower the permission for all mappings to a given page. 3166 */ 3167void 3168pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 3169{ 3170 if ((prot & VM_PROT_WRITE) == 0) { 3171 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3172 pmap_changebit(phys, PG_RW, FALSE); 3173 } else { 3174 pmap_remove_all(phys); 3175 } 3176 } 3177} 3178 3179vm_offset_t 3180pmap_phys_address(ppn) 3181 int ppn; 3182{ 3183 return (i386_ptob(ppn)); 3184} 3185 3186/* 3187 * pmap_ts_referenced: 3188 * 3189 * Return the count of reference bits for a page, clearing all of them. 3190 * 3191 */ 3192int 3193pmap_ts_referenced(vm_offset_t pa) 3194{ 3195 register pv_entry_t pv, pvf, pvn; 3196 pv_table_t *ppv; 3197 unsigned *pte; 3198 int s; 3199 int rtval = 0; 3200 3201 if (!pmap_is_managed(pa)) 3202 return FALSE; 3203 3204 s = splvm(); 3205 3206 ppv = pa_to_pvh(pa); 3207 3208 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 3209 splx(s); 3210 return 0; 3211 } 3212 3213 /* 3214 * Not found, check current mappings returning immediately if found. 3215 */ 3216 pvf = 0; 3217 for (pv = TAILQ_FIRST(&ppv->pv_list); pv && pv != pvf; pv = pvn) { 3218 if (!pvf) 3219 pvf = pv; 3220 pvn = TAILQ_NEXT(pv, pv_list); 3221 3222 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 3223 /* 3224 * if the bit being tested is the modified bit, then 3225 * mark clean_map and ptes as never 3226 * modified. 3227 */ 3228 if (!pmap_track_modified(pv->pv_va)) { 3229 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3230 continue; 3231 } 3232 3233 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3234 if (pte == NULL) { 3235 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3236 continue; 3237 } 3238 3239 if (*pte & PG_A) { 3240 rtval++; 3241 *pte &= ~PG_A; 3242 if (rtval > 4) { 3243 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3244 break; 3245 } 3246 } 3247 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3248 } 3249 3250 splx(s); 3251 if (rtval) { 3252 invltlb(); 3253 } 3254 return (rtval); 3255} 3256 3257/* 3258 * pmap_is_modified: 3259 * 3260 * Return whether or not the specified physical page was modified 3261 * in any physical maps. 3262 */ 3263boolean_t 3264pmap_is_modified(vm_offset_t pa) 3265{ 3266 return pmap_testbit((pa), PG_M); 3267} 3268 3269/* 3270 * Clear the modify bits on the specified physical page. 3271 */ 3272void 3273pmap_clear_modify(vm_offset_t pa) 3274{ 3275 pmap_changebit((pa), PG_M, FALSE); 3276} 3277 3278/* 3279 * pmap_clear_reference: 3280 * 3281 * Clear the reference bit on the specified physical page. 3282 */ 3283void 3284pmap_clear_reference(vm_offset_t pa) 3285{ 3286 pmap_changebit((pa), PG_A, FALSE); 3287} 3288 3289/* 3290 * Miscellaneous support routines follow 3291 */ 3292 3293static void 3294i386_protection_init() 3295{ 3296 register int *kp, prot; 3297 3298 kp = protection_codes; 3299 for (prot = 0; prot < 8; prot++) { 3300 switch (prot) { 3301 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3302 /* 3303 * Read access is also 0. There isn't any execute bit, 3304 * so just make it readable. 3305 */ 3306 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3307 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3308 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3309 *kp++ = 0; 3310 break; 3311 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3312 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3313 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3314 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3315 *kp++ = PG_RW; 3316 break; 3317 } 3318 } 3319} 3320 3321/* 3322 * Map a set of physical memory pages into the kernel virtual 3323 * address space. Return a pointer to where it is mapped. This 3324 * routine is intended to be used for mapping device memory, 3325 * NOT real memory. 3326 */ 3327void * 3328pmap_mapdev(pa, size) 3329 vm_offset_t pa; 3330 vm_size_t size; 3331{ 3332 vm_offset_t va, tmpva; 3333 unsigned *pte; 3334 3335 size = roundup(size, PAGE_SIZE); 3336 3337 va = kmem_alloc_pageable(kernel_map, size); 3338#if !defined(MAX_PERF) 3339 if (!va) 3340 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3341#endif 3342 3343 pa = pa & PG_FRAME; 3344 for (tmpva = va; size > 0;) { 3345 pte = (unsigned *)vtopte(tmpva); 3346 *pte = pa | PG_RW | PG_V | pgeflag; 3347 size -= PAGE_SIZE; 3348 tmpva += PAGE_SIZE; 3349 pa += PAGE_SIZE; 3350 } 3351 invltlb(); 3352 3353 return ((void *) va); 3354} 3355 3356/* 3357 * perform the pmap work for mincore 3358 */ 3359int 3360pmap_mincore(pmap, addr) 3361 pmap_t pmap; 3362 vm_offset_t addr; 3363{ 3364 3365 unsigned *ptep, pte; 3366 vm_page_t m; 3367 int val = 0; 3368 3369 ptep = pmap_pte(pmap, addr); 3370 if (ptep == 0) { 3371 return 0; 3372 } 3373 3374 if ((pte = *ptep) != 0) { 3375 pv_table_t *ppv; 3376 vm_offset_t pa; 3377 3378 val = MINCORE_INCORE; 3379 if ((pte & PG_MANAGED) == 0) 3380 return val; 3381 3382 pa = pte & PG_FRAME; 3383 3384 ppv = pa_to_pvh((pa & PG_FRAME)); 3385 m = ppv->pv_vm_page; 3386 3387 /* 3388 * Modified by us 3389 */ 3390 if (pte & PG_M) 3391 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3392 /* 3393 * Modified by someone 3394 */ 3395 else if (m->dirty || pmap_is_modified(pa)) 3396 val |= MINCORE_MODIFIED_OTHER; 3397 /* 3398 * Referenced by us 3399 */ 3400 if (pte & PG_A) 3401 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3402 3403 /* 3404 * Referenced by someone 3405 */ 3406 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(pa)) { 3407 val |= MINCORE_REFERENCED_OTHER; 3408 vm_page_flag_set(m, PG_REFERENCED); 3409 } 3410 } 3411 return val; 3412} 3413 3414void 3415pmap_activate(struct proc *p) 3416{ 3417#if defined(SWTCH_OPTIM_STATS) 3418 tlb_flush_count++; 3419#endif 3420 load_cr3(p->p_addr->u_pcb.pcb_cr3 = 3421 vtophys(p->p_vmspace->vm_pmap.pm_pdir)); 3422} 3423 3424vm_offset_t 3425pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3426 3427 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3428 return addr; 3429 } 3430 3431 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3432 return addr; 3433} 3434 3435 3436#if defined(PMAP_DEBUG) 3437pmap_pid_dump(int pid) { 3438 pmap_t pmap; 3439 struct proc *p; 3440 int npte = 0; 3441 int index; 3442 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3443 if (p->p_pid != pid) 3444 continue; 3445 3446 if (p->p_vmspace) { 3447 int i,j; 3448 index = 0; 3449 pmap = &p->p_vmspace->vm_pmap; 3450 for(i=0;i<1024;i++) { 3451 pd_entry_t *pde; 3452 unsigned *pte; 3453 unsigned base = i << PDRSHIFT; 3454 3455 pde = &pmap->pm_pdir[i]; 3456 if (pde && pmap_pde_v(pde)) { 3457 for(j=0;j<1024;j++) { 3458 unsigned va = base + (j << PAGE_SHIFT); 3459 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3460 if (index) { 3461 index = 0; 3462 printf("\n"); 3463 } 3464 return npte; 3465 } 3466 pte = pmap_pte_quick( pmap, va); 3467 if (pte && pmap_pte_v(pte)) { 3468 vm_offset_t pa; 3469 vm_page_t m; 3470 pa = *(int *)pte; 3471 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3472 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3473 va, pa, m->hold_count, m->wire_count, m->flags); 3474 npte++; 3475 index++; 3476 if (index >= 2) { 3477 index = 0; 3478 printf("\n"); 3479 } else { 3480 printf(" "); 3481 } 3482 } 3483 } 3484 } 3485 } 3486 } 3487 } 3488 return npte; 3489} 3490#endif 3491 3492#if defined(DEBUG) 3493 3494static void pads __P((pmap_t pm)); 3495void pmap_pvdump __P((vm_offset_t pa)); 3496 3497/* print address space of pmap*/ 3498static void 3499pads(pm) 3500 pmap_t pm; 3501{ 3502 unsigned va, i, j; 3503 unsigned *ptep; 3504 3505 if (pm == kernel_pmap) 3506 return; 3507 for (i = 0; i < 1024; i++) 3508 if (pm->pm_pdir[i]) 3509 for (j = 0; j < 1024; j++) { 3510 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3511 if (pm == kernel_pmap && va < KERNBASE) 3512 continue; 3513 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3514 continue; 3515 ptep = pmap_pte_quick(pm, va); 3516 if (pmap_pte_v(ptep)) 3517 printf("%x:%x ", va, *(int *) ptep); 3518 }; 3519 3520} 3521 3522void 3523pmap_pvdump(pa) 3524 vm_offset_t pa; 3525{ 3526 pv_table_t *ppv; 3527 register pv_entry_t pv; 3528 3529 printf("pa %x", pa); 3530 ppv = pa_to_pvh(pa); 3531 for (pv = TAILQ_FIRST(&ppv->pv_list); 3532 pv; 3533 pv = TAILQ_NEXT(pv, pv_list)) { 3534#ifdef used_to_be 3535 printf(" -> pmap %p, va %x, flags %x", 3536 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags); 3537#endif 3538 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3539 pads(pv->pv_pmap); 3540 } 3541 printf(" "); 3542} 3543#endif 3544