pmap.c revision 37557
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 1.203 1998/07/11 07:45:32 bde 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#ifdef SMP 188extern char prv_CPAGE1[], prv_CPAGE2[], prv_CPAGE3[]; 189extern pt_entry_t *prv_CMAP1, *prv_CMAP2, *prv_CMAP3; 190extern pd_entry_t *IdlePTDS[]; 191extern pt_entry_t SMP_prvpt[]; 192#endif 193 194#ifdef SMP 195extern unsigned int prv_PPAGE1[]; 196extern pt_entry_t *prv_PMAP1; 197#else 198static pt_entry_t *PMAP1 = 0; 199static unsigned *PADDR1 = 0; 200#endif 201 202static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 203static unsigned * get_ptbase __P((pmap_t pmap)); 204static pv_entry_t get_pv_entry __P((void)); 205static void i386_protection_init __P((void)); 206static void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 207 208static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 209static void pmap_remove_all __P((vm_offset_t pa)); 210static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 211 vm_offset_t pa, vm_page_t mpte)); 212static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 213 vm_offset_t sva)); 214static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 215static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 216 vm_offset_t va)); 217static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 218static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 219 vm_page_t mpte, vm_offset_t pa)); 220 221static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 222 223static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 224static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 225static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 226static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 227static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 228static vm_offset_t pmap_kmem_choose(vm_offset_t addr); 229void pmap_collect(void); 230 231static unsigned pdir4mb; 232 233/* 234 * Routine: pmap_pte 235 * Function: 236 * Extract the page table entry associated 237 * with the given map/virtual_address pair. 238 */ 239 240PMAP_INLINE unsigned * 241pmap_pte(pmap, va) 242 register pmap_t pmap; 243 vm_offset_t va; 244{ 245 unsigned *pdeaddr; 246 247 if (pmap) { 248 pdeaddr = (unsigned *) pmap_pde(pmap, va); 249 if (*pdeaddr & PG_PS) 250 return pdeaddr; 251 if (*pdeaddr) { 252 return get_ptbase(pmap) + i386_btop(va); 253 } 254 } 255 return (0); 256} 257 258/* 259 * Move the kernel virtual free pointer to the next 260 * 4MB. This is used to help improve performance 261 * by using a large (4MB) page for much of the kernel 262 * (.text, .data, .bss) 263 */ 264static vm_offset_t 265pmap_kmem_choose(vm_offset_t addr) { 266 vm_offset_t newaddr = addr; 267#ifndef DISABLE_PSE 268 if (cpu_feature & CPUID_PSE) { 269 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 270 } 271#endif 272 return newaddr; 273} 274 275/* 276 * Bootstrap the system enough to run with virtual memory. 277 * 278 * On the i386 this is called after mapping has already been enabled 279 * and just syncs the pmap module with what has already been done. 280 * [We can't call it easily with mapping off since the kernel is not 281 * mapped with PA == VA, hence we would have to relocate every address 282 * from the linked base (virtual) address "KERNBASE" to the actual 283 * (physical) address starting relative to 0] 284 */ 285void 286pmap_bootstrap(firstaddr, loadaddr) 287 vm_offset_t firstaddr; 288 vm_offset_t loadaddr; 289{ 290 vm_offset_t va; 291 pt_entry_t *pte; 292 int i, j; 293 294 avail_start = firstaddr; 295 296 /* 297 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 298 * large. It should instead be correctly calculated in locore.s and 299 * not based on 'first' (which is a physical address, not a virtual 300 * address, for the start of unused physical memory). The kernel 301 * page tables are NOT double mapped and thus should not be included 302 * in this calculation. 303 */ 304 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 305 virtual_avail = pmap_kmem_choose(virtual_avail); 306 307 virtual_end = VM_MAX_KERNEL_ADDRESS; 308 309 /* 310 * Initialize protection array. 311 */ 312 i386_protection_init(); 313 314 /* 315 * The kernel's pmap is statically allocated so we don't have to use 316 * pmap_create, which is unlikely to work correctly at this part of 317 * the boot sequence (XXX and which no longer exists). 318 */ 319 kernel_pmap = &kernel_pmap_store; 320 321 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 322 323 kernel_pmap->pm_count = 1; 324 TAILQ_INIT(&kernel_pmap->pm_pvlist); 325 nkpt = NKPT; 326 327 /* 328 * Reserve some special page table entries/VA space for temporary 329 * mapping of pages. 330 */ 331#define SYSMAP(c, p, v, n) \ 332 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 333 334 va = virtual_avail; 335 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 336 337 /* 338 * CMAP1/CMAP2 are used for zeroing and copying pages. 339 */ 340 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 341 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 342 343 /* 344 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 345 * XXX ptmmap is not used. 346 */ 347 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 348 349 /* 350 * msgbufp is used to map the system message buffer. 351 * XXX msgbufmap is not used. 352 */ 353 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 354 atop(round_page(MSGBUF_SIZE))) 355 356#if !defined(SMP) 357 /* 358 * ptemap is used for pmap_pte_quick 359 */ 360 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 361#endif 362 363 virtual_avail = va; 364 365 *(int *) CMAP1 = *(int *) CMAP2 = 0; 366 *(int *) PTD = 0; 367 368 369 pgeflag = 0; 370#if !defined(SMP) 371 if (cpu_feature & CPUID_PGE) { 372 pgeflag = PG_G; 373 } 374#endif 375 376/* 377 * Initialize the 4MB page size flag 378 */ 379 pseflag = 0; 380/* 381 * The 4MB page version of the initial 382 * kernel page mapping. 383 */ 384 pdir4mb = 0; 385 386#if !defined(DISABLE_PSE) 387 if (cpu_feature & CPUID_PSE) { 388 unsigned ptditmp; 389 /* 390 * Enable the PSE mode 391 */ 392 load_cr4(rcr4() | CR4_PSE); 393 394 /* 395 * Note that we have enabled PSE mode 396 */ 397 pseflag = PG_PS; 398 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE)); 399 ptditmp &= ~(NBPDR - 1); 400 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 401 pdir4mb = ptditmp; 402 /* 403 * We can do the mapping here for the single processor 404 * case. We simply ignore the old page table page from 405 * now on. 406 */ 407#if !defined(SMP) 408 PTD[KPTDI] = (pd_entry_t) ptditmp; 409 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 410 invltlb(); 411#endif 412 } 413#endif 414 415#ifdef SMP 416 if (cpu_apic_address == 0) 417 panic("pmap_bootstrap: no local apic!"); 418 419 /* 0 = private page */ 420 /* 1 = page table page */ 421 /* 2 = local apic */ 422 /* 16-31 = io apics */ 423 SMP_prvpt[2] = (pt_entry_t)(PG_V | PG_RW | pgeflag | ((u_long)cpu_apic_address & PG_FRAME)); 424 425 for (i = 0; i < mp_napics; i++) { 426 for (j = 0; j < 16; j++) { 427 /* same page frame as a previous IO apic? */ 428 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 429 ((u_long)io_apic_address[0] & PG_FRAME)) { 430 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 431 break; 432 } 433 /* use this slot if available */ 434 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 0) { 435 SMP_prvpt[j + 16] = (pt_entry_t)(PG_V | PG_RW | pgeflag | 436 ((u_long)io_apic_address[i] & PG_FRAME)); 437 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 438 break; 439 } 440 } 441 if (j == 16) 442 panic("no space to map IO apic %d!", i); 443 } 444 445 /* BSP does this itself, AP's get it pre-set */ 446 prv_CMAP1 = &SMP_prvpt[3 + UPAGES]; 447 prv_CMAP2 = &SMP_prvpt[4 + UPAGES]; 448 prv_CMAP3 = &SMP_prvpt[5 + UPAGES]; 449 prv_PMAP1 = &SMP_prvpt[6 + UPAGES]; 450#endif 451 452 invltlb(); 453 454} 455 456void 457getmtrr() 458{ 459 int i; 460 461 if (cpu == CPU_686) { 462 for(i = 0; i < NPPROVMTRR; i++) { 463 PPro_vmtrr[i].base = rdmsr(PPRO_VMTRRphysBase0 + i * 2); 464 PPro_vmtrr[i].mask = rdmsr(PPRO_VMTRRphysMask0 + i * 2); 465 } 466 } 467} 468 469void 470putmtrr() 471{ 472 int i; 473 474 if (cpu == CPU_686) { 475 wbinvd(); 476 for(i = 0; i < NPPROVMTRR; i++) { 477 wrmsr(PPRO_VMTRRphysBase0 + i * 2, PPro_vmtrr[i].base); 478 wrmsr(PPRO_VMTRRphysMask0 + i * 2, PPro_vmtrr[i].mask); 479 } 480 } 481} 482 483void 484pmap_setvidram(void) 485{ 486#if 0 487 if (cpu == CPU_686) { 488 wbinvd(); 489 /* 490 * Set memory between 0-640K to be WB 491 */ 492 wrmsr(0x250, 0x0606060606060606LL); 493 wrmsr(0x258, 0x0606060606060606LL); 494 /* 495 * Set normal, PC video memory to be WC 496 */ 497 wrmsr(0x259, 0x0101010101010101LL); 498 } 499#endif 500} 501 502void 503pmap_setdevram(unsigned long long basea, vm_offset_t sizea) 504{ 505 int i, free, skip; 506 unsigned basepage, basepaget; 507 unsigned long long base; 508 unsigned long long mask; 509 510 if (cpu != CPU_686) 511 return; 512 513 free = -1; 514 skip = 0; 515 basea &= ~0xfff; 516 base = basea | 0x1; 517 mask = (long long) (0xfffffffffLL - ((long) sizea - 1)) | (long long) 0x800; 518 mask &= ~0x7ff; 519 520 basepage = (long long) (base >> 12); 521 for(i = 0; i < NPPROVMTRR; i++) { 522 PPro_vmtrr[i].base = rdmsr(PPRO_VMTRRphysBase0 + i * 2); 523 PPro_vmtrr[i].mask = rdmsr(PPRO_VMTRRphysMask0 + i * 2); 524 basepaget = (long long) (PPro_vmtrr[i].base >> 12); 525 if (basepage == basepaget) 526 skip = 1; 527 if ((PPro_vmtrr[i].mask & 0x800) == 0) { 528 if (free == -1) 529 free = i; 530 } 531 } 532 533 if (!skip && free != -1) { 534 wbinvd(); 535 PPro_vmtrr[free].base = base; 536 PPro_vmtrr[free].mask = mask; 537 wrmsr(PPRO_VMTRRphysBase0 + free * 2, base); 538 wrmsr(PPRO_VMTRRphysMask0 + free * 2, mask); 539 printf( 540 "pmap: added WC mapping at page: 0x%x %x, size: %u mask: 0x%x %x\n", 541 (u_int)(base >> 32), (u_int)base, sizea, 542 (u_int)(mask >> 32), (u_int)mask); 543 } 544} 545 546/* 547 * Set 4mb pdir for mp startup, and global flags 548 */ 549void 550pmap_set_opt(unsigned *pdir) { 551 int i; 552 553 if (pseflag && (cpu_feature & CPUID_PSE)) { 554 load_cr4(rcr4() | CR4_PSE); 555 if (pdir4mb) { 556 pdir[KPTDI] = pdir4mb; 557 } 558 } 559 560 if (pgeflag && (cpu_feature & CPUID_PGE)) { 561 load_cr4(rcr4() | CR4_PGE); 562 for(i = KPTDI; i < KPTDI + nkpt; i++) { 563 if (pdir[i]) { 564 pdir[i] |= PG_G; 565 } 566 } 567 } 568} 569 570/* 571 * Setup the PTD for the boot processor 572 */ 573void 574pmap_set_opt_bsp(void) 575{ 576 pmap_set_opt((unsigned *)kernel_pmap->pm_pdir); 577 pmap_set_opt((unsigned *)PTD); 578 invltlb(); 579} 580 581/* 582 * Initialize the pmap module. 583 * Called by vm_init, to initialize any structures that the pmap 584 * system needs to map virtual memory. 585 * pmap_init has been enhanced to support in a fairly consistant 586 * way, discontiguous physical memory. 587 */ 588void 589pmap_init(phys_start, phys_end) 590 vm_offset_t phys_start, phys_end; 591{ 592 vm_offset_t addr; 593 vm_size_t s; 594 int i; 595 int initial_pvs; 596 597 /* 598 * calculate the number of pv_entries needed 599 */ 600 vm_first_phys = phys_avail[0]; 601 for (i = 0; phys_avail[i + 1]; i += 2); 602 pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 603 604 /* 605 * Allocate memory for random pmap data structures. Includes the 606 * pv_head_table. 607 */ 608 s = (vm_size_t) (sizeof(pv_table_t) * pv_npg); 609 s = round_page(s); 610 611 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 612 pv_table = (pv_table_t *) addr; 613 for(i = 0; i < pv_npg; i++) { 614 vm_offset_t pa; 615 TAILQ_INIT(&pv_table[i].pv_list); 616 pv_table[i].pv_list_count = 0; 617 pa = vm_first_phys + i * PAGE_SIZE; 618 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 619 } 620 621 /* 622 * init the pv free list 623 */ 624 initial_pvs = pv_npg; 625 if (initial_pvs < MINPV) 626 initial_pvs = MINPV; 627 pvzone = &pvzone_store; 628 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 629 initial_pvs * sizeof (struct pv_entry)); 630 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, pv_npg); 631 /* 632 * object for kernel page table pages 633 */ 634 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE); 635 636 /* 637 * Now it is safe to enable pv_table recording. 638 */ 639 pmap_initialized = TRUE; 640} 641 642/* 643 * Initialize the address space (zone) for the pv_entries. Set a 644 * high water mark so that the system can recover from excessive 645 * numbers of pv entries. 646 */ 647void 648pmap_init2() { 649 pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg; 650 pv_entry_high_water = 9 * (pv_entry_max / 10); 651 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 652} 653 654/* 655 * Used to map a range of physical addresses into kernel 656 * virtual address space. 657 * 658 * For now, VM is already on, we only need to map the 659 * specified memory. 660 */ 661vm_offset_t 662pmap_map(virt, start, end, prot) 663 vm_offset_t virt; 664 vm_offset_t start; 665 vm_offset_t end; 666 int prot; 667{ 668 while (start < end) { 669 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 670 virt += PAGE_SIZE; 671 start += PAGE_SIZE; 672 } 673 return (virt); 674} 675 676 677/*************************************************** 678 * Low level helper routines..... 679 ***************************************************/ 680 681#if defined(PMAP_DIAGNOSTIC) 682 683/* 684 * This code checks for non-writeable/modified pages. 685 * This should be an invalid condition. 686 */ 687static int 688pmap_nw_modified(pt_entry_t ptea) { 689 int pte; 690 691 pte = (int) ptea; 692 693 if ((pte & (PG_M|PG_RW)) == PG_M) 694 return 1; 695 else 696 return 0; 697} 698#endif 699 700 701/* 702 * this routine defines the region(s) of memory that should 703 * not be tested for the modified bit. 704 */ 705static PMAP_INLINE int 706pmap_track_modified( vm_offset_t va) { 707 if ((va < clean_sva) || (va >= clean_eva)) 708 return 1; 709 else 710 return 0; 711} 712 713static PMAP_INLINE void 714invltlb_1pg( vm_offset_t va) { 715#if defined(I386_CPU) 716 if (cpu_class == CPUCLASS_386) { 717 invltlb(); 718 } else 719#endif 720 { 721 invlpg(va); 722 } 723} 724 725static PMAP_INLINE void 726invltlb_2pg( vm_offset_t va1, vm_offset_t va2) { 727#if defined(I386_CPU) 728 if (cpu_class == CPUCLASS_386) { 729 invltlb(); 730 } else 731#endif 732 { 733 invlpg(va1); 734 invlpg(va2); 735 } 736} 737 738static unsigned * 739get_ptbase(pmap) 740 pmap_t pmap; 741{ 742 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 743 744 /* are we current address space or kernel? */ 745 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 746 return (unsigned *) PTmap; 747 } 748 /* otherwise, we are alternate address space */ 749 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 750 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 751#if defined(SMP) 752 /* The page directory is not shared between CPUs */ 753 cpu_invltlb(); 754#else 755 invltlb(); 756#endif 757 } 758 return (unsigned *) APTmap; 759} 760 761/* 762 * Super fast pmap_pte routine best used when scanning 763 * the pv lists. This eliminates many coarse-grained 764 * invltlb calls. Note that many of the pv list 765 * scans are across different pmaps. It is very wasteful 766 * to do an entire invltlb for checking a single mapping. 767 */ 768 769static unsigned * 770pmap_pte_quick(pmap, va) 771 register pmap_t pmap; 772 vm_offset_t va; 773{ 774 unsigned pde, newpf; 775 if (pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) { 776 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 777 unsigned index = i386_btop(va); 778 /* are we current address space or kernel? */ 779 if ((pmap == kernel_pmap) || 780 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 781 return (unsigned *) PTmap + index; 782 } 783 newpf = pde & PG_FRAME; 784#ifdef SMP 785 if ( ((* (unsigned *) prv_PMAP1) & PG_FRAME) != newpf) { 786 * (unsigned *) prv_PMAP1 = newpf | PG_RW | PG_V; 787 cpu_invlpg(&prv_PPAGE1); 788 } 789 return prv_PPAGE1 + ((unsigned) index & (NPTEPG - 1)); 790#else 791 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 792 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 793 invltlb_1pg((vm_offset_t) PADDR1); 794 } 795 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 796#endif 797 } 798 return (0); 799} 800 801/* 802 * Routine: pmap_extract 803 * Function: 804 * Extract the physical page address associated 805 * with the given map/virtual_address pair. 806 */ 807vm_offset_t 808pmap_extract(pmap, va) 809 register pmap_t pmap; 810 vm_offset_t va; 811{ 812 vm_offset_t rtval; 813 vm_offset_t pdirindex; 814 pdirindex = va >> PDRSHIFT; 815 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 816 unsigned *pte; 817 if ((rtval & PG_PS) != 0) { 818 rtval &= ~(NBPDR - 1); 819 rtval |= va & (NBPDR - 1); 820 return rtval; 821 } 822 pte = get_ptbase(pmap) + i386_btop(va); 823 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 824 return rtval; 825 } 826 return 0; 827 828} 829 830/* 831 * determine if a page is managed (memory vs. device) 832 */ 833static PMAP_INLINE int 834pmap_is_managed(pa) 835 vm_offset_t pa; 836{ 837 int i; 838 839 if (!pmap_initialized) 840 return 0; 841 842 for (i = 0; phys_avail[i + 1]; i += 2) { 843 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 844 return 1; 845 } 846 return 0; 847} 848 849 850/*************************************************** 851 * Low level mapping routines..... 852 ***************************************************/ 853 854/* 855 * Add a list of wired pages to the kva 856 * this routine is only used for temporary 857 * kernel mappings that do not need to have 858 * page modification or references recorded. 859 * Note that old mappings are simply written 860 * over. The page *must* be wired. 861 */ 862void 863pmap_qenter(va, m, count) 864 vm_offset_t va; 865 vm_page_t *m; 866 int count; 867{ 868 int i; 869 register unsigned *pte; 870 871 for (i = 0; i < count; i++) { 872 vm_offset_t tva = va + i * PAGE_SIZE; 873 unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | pgeflag; 874 unsigned opte; 875 pte = (unsigned *)vtopte(tva); 876 opte = *pte; 877 *pte = npte; 878 if (opte) 879 invltlb_1pg(tva); 880 } 881} 882 883/* 884 * this routine jerks page mappings from the 885 * kernel -- it is meant only for temporary mappings. 886 */ 887void 888pmap_qremove(va, count) 889 vm_offset_t va; 890 int count; 891{ 892 int i; 893 register unsigned *pte; 894 895 for (i = 0; i < count; i++) { 896 pte = (unsigned *)vtopte(va); 897 *pte = 0; 898 invltlb_1pg(va); 899 va += PAGE_SIZE; 900 } 901} 902 903/* 904 * add a wired page to the kva 905 * note that in order for the mapping to take effect -- you 906 * should do a invltlb after doing the pmap_kenter... 907 */ 908PMAP_INLINE void 909pmap_kenter(va, pa) 910 vm_offset_t va; 911 register vm_offset_t pa; 912{ 913 register unsigned *pte; 914 unsigned npte, opte; 915 916 npte = pa | PG_RW | PG_V | pgeflag; 917 pte = (unsigned *)vtopte(va); 918 opte = *pte; 919 *pte = npte; 920 if (opte) 921 invltlb_1pg(va); 922} 923 924/* 925 * remove a page from the kernel pagetables 926 */ 927PMAP_INLINE void 928pmap_kremove(va) 929 vm_offset_t va; 930{ 931 register unsigned *pte; 932 933 pte = (unsigned *)vtopte(va); 934 *pte = 0; 935 invltlb_1pg(va); 936} 937 938static vm_page_t 939pmap_page_lookup(object, pindex) 940 vm_object_t object; 941 vm_pindex_t pindex; 942{ 943 vm_page_t m; 944retry: 945 m = vm_page_lookup(object, pindex); 946 if (m && vm_page_sleep(m, "pplookp", NULL)) 947 goto retry; 948 return m; 949} 950 951/* 952 * Create the UPAGES for a new process. 953 * This routine directly affects the fork perf for a process. 954 */ 955void 956pmap_new_proc(p) 957 struct proc *p; 958{ 959 int i, updateneeded; 960 vm_object_t upobj; 961 vm_page_t m; 962 struct user *up; 963 unsigned *ptek, oldpte; 964 965 /* 966 * allocate object for the upages 967 */ 968 if ((upobj = p->p_upages_obj) == NULL) { 969 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 970 p->p_upages_obj = upobj; 971 } 972 973 /* get a kernel virtual address for the UPAGES for this proc */ 974 if ((up = p->p_addr) == NULL) { 975 up = (struct user *) kmem_alloc_pageable(kernel_map, 976 UPAGES * PAGE_SIZE); 977#if !defined(MAX_PERF) 978 if (up == NULL) 979 panic("pmap_new_proc: u_map allocation failed"); 980#endif 981 p->p_addr = up; 982 } 983 984 ptek = (unsigned *) vtopte((vm_offset_t) up); 985 986 updateneeded = 0; 987 for(i=0;i<UPAGES;i++) { 988 /* 989 * Get a kernel stack page 990 */ 991 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 992 993 /* 994 * Wire the page 995 */ 996 m->wire_count++; 997 cnt.v_wire_count++; 998 999 oldpte = *(ptek + i); 1000 /* 1001 * Enter the page into the kernel address space. 1002 */ 1003 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 1004 if (oldpte) { 1005 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) { 1006 invlpg((vm_offset_t) up + i * PAGE_SIZE); 1007 } else { 1008 updateneeded = 1; 1009 } 1010 } 1011 1012 PAGE_WAKEUP(m); 1013 m->flags &= ~PG_ZERO; 1014 m->flags |= PG_MAPPED | PG_WRITEABLE; 1015 m->valid = VM_PAGE_BITS_ALL; 1016 } 1017 if (updateneeded) 1018 invltlb(); 1019} 1020 1021/* 1022 * Dispose the UPAGES for a process that has exited. 1023 * This routine directly impacts the exit perf of a process. 1024 */ 1025void 1026pmap_dispose_proc(p) 1027 struct proc *p; 1028{ 1029 int i; 1030 vm_object_t upobj; 1031 vm_page_t m; 1032 unsigned *ptek, oldpte; 1033 1034 upobj = p->p_upages_obj; 1035 1036 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 1037 for(i=0;i<UPAGES;i++) { 1038 1039 if ((m = vm_page_lookup(upobj, i)) == NULL) 1040 panic("pmap_dispose_proc: upage already missing???"); 1041 1042 m->flags |= PG_BUSY; 1043 1044 oldpte = *(ptek + i); 1045 *(ptek + i) = 0; 1046 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) 1047 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 1048 vm_page_unwire(m); 1049 vm_page_free(m); 1050 } 1051 1052 if (cpu_class <= CPUCLASS_386) 1053 invltlb(); 1054} 1055 1056/* 1057 * Allow the UPAGES for a process to be prejudicially paged out. 1058 */ 1059void 1060pmap_swapout_proc(p) 1061 struct proc *p; 1062{ 1063 int i; 1064 vm_object_t upobj; 1065 vm_page_t m; 1066 1067 upobj = p->p_upages_obj; 1068 /* 1069 * let the upages be paged 1070 */ 1071 for(i=0;i<UPAGES;i++) { 1072 if ((m = vm_page_lookup(upobj, i)) == NULL) 1073 panic("pmap_swapout_proc: upage already missing???"); 1074 m->dirty = VM_PAGE_BITS_ALL; 1075 vm_page_unwire(m); 1076 vm_page_deactivate(m); 1077 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 1078 } 1079} 1080 1081/* 1082 * Bring the UPAGES for a specified process back in. 1083 */ 1084void 1085pmap_swapin_proc(p) 1086 struct proc *p; 1087{ 1088 int i,rv; 1089 vm_object_t upobj; 1090 vm_page_t m; 1091 1092 upobj = p->p_upages_obj; 1093 for(i=0;i<UPAGES;i++) { 1094 1095 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1096 1097 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 1098 VM_PAGE_TO_PHYS(m)); 1099 1100 if (m->valid != VM_PAGE_BITS_ALL) { 1101 rv = vm_pager_get_pages(upobj, &m, 1, 0); 1102#if !defined(MAX_PERF) 1103 if (rv != VM_PAGER_OK) 1104 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 1105#endif 1106 m = vm_page_lookup(upobj, i); 1107 m->valid = VM_PAGE_BITS_ALL; 1108 } 1109 1110 vm_page_wire(m); 1111 PAGE_WAKEUP(m); 1112 m->flags |= PG_MAPPED | PG_WRITEABLE; 1113 } 1114} 1115 1116/*************************************************** 1117 * Page table page management routines..... 1118 ***************************************************/ 1119 1120/* 1121 * This routine unholds page table pages, and if the hold count 1122 * drops to zero, then it decrements the wire count. 1123 */ 1124static int 1125_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1126 int s; 1127 1128 while (vm_page_sleep(m, "pmuwpt", NULL)); 1129 1130 if (m->hold_count == 0) { 1131 vm_offset_t pteva; 1132 /* 1133 * unmap the page table page 1134 */ 1135 pmap->pm_pdir[m->pindex] = 0; 1136 --pmap->pm_stats.resident_count; 1137 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1138 (((unsigned) PTDpde) & PG_FRAME)) { 1139 /* 1140 * Do a invltlb to make the invalidated mapping 1141 * take effect immediately. 1142 */ 1143 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1144 invltlb_1pg(pteva); 1145 } 1146 1147 if (pmap->pm_ptphint == m) 1148 pmap->pm_ptphint = NULL; 1149 1150 /* 1151 * If the page is finally unwired, simply free it. 1152 */ 1153 --m->wire_count; 1154 if (m->wire_count == 0) { 1155 1156 if (m->flags & PG_WANTED) { 1157 m->flags &= ~PG_WANTED; 1158 wakeup(m); 1159 } 1160 1161 m->flags |= PG_BUSY; 1162 vm_page_free_zero(m); 1163 --cnt.v_wire_count; 1164 } 1165 return 1; 1166 } 1167 return 0; 1168} 1169 1170static PMAP_INLINE int 1171pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1172 vm_page_unhold(m); 1173 if (m->hold_count == 0) 1174 return _pmap_unwire_pte_hold(pmap, m); 1175 else 1176 return 0; 1177} 1178 1179/* 1180 * After removing a page table entry, this routine is used to 1181 * conditionally free the page, and manage the hold/wire counts. 1182 */ 1183static int 1184pmap_unuse_pt(pmap, va, mpte) 1185 pmap_t pmap; 1186 vm_offset_t va; 1187 vm_page_t mpte; 1188{ 1189 unsigned ptepindex; 1190 if (va >= UPT_MIN_ADDRESS) 1191 return 0; 1192 1193 if (mpte == NULL) { 1194 ptepindex = (va >> PDRSHIFT); 1195 if (pmap->pm_ptphint && 1196 (pmap->pm_ptphint->pindex == ptepindex)) { 1197 mpte = pmap->pm_ptphint; 1198 } else { 1199 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1200 pmap->pm_ptphint = mpte; 1201 } 1202 } 1203 1204 return pmap_unwire_pte_hold(pmap, mpte); 1205} 1206 1207#if !defined(SMP) 1208void 1209pmap_pinit0(pmap) 1210 struct pmap *pmap; 1211{ 1212 pmap->pm_pdir = 1213 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1214 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1215 pmap->pm_flags = 0; 1216 pmap->pm_count = 1; 1217 pmap->pm_ptphint = NULL; 1218 TAILQ_INIT(&pmap->pm_pvlist); 1219 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1220} 1221#else 1222void 1223pmap_pinit0(pmap) 1224 struct pmap *pmap; 1225{ 1226 pmap_pinit(pmap); 1227} 1228#endif 1229 1230/* 1231 * Initialize a preallocated and zeroed pmap structure, 1232 * such as one in a vmspace structure. 1233 */ 1234void 1235pmap_pinit(pmap) 1236 register struct pmap *pmap; 1237{ 1238 vm_page_t ptdpg; 1239 1240 /* 1241 * No need to allocate page table space yet but we do need a valid 1242 * page directory table. 1243 */ 1244 if (pmap->pm_pdir == NULL) 1245 pmap->pm_pdir = 1246 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1247 1248 /* 1249 * allocate object for the ptes 1250 */ 1251 if (pmap->pm_pteobj == NULL) 1252 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1253 1254 /* 1255 * allocate the page directory page 1256 */ 1257retry: 1258 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1259 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1260 1261 ptdpg->wire_count = 1; 1262 ++cnt.v_wire_count; 1263 1264 ptdpg->flags &= ~(PG_MAPPED | PG_BUSY); /* not mapped normally */ 1265 ptdpg->valid = VM_PAGE_BITS_ALL; 1266 1267 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1268 if ((ptdpg->flags & PG_ZERO) == 0) 1269 bzero(pmap->pm_pdir, PAGE_SIZE); 1270 1271 /* wire in kernel global address entries */ 1272 /* XXX copies current process, does not fill in MPPTDI */ 1273 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1274 1275 /* install self-referential address mapping entry */ 1276 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1277 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1278 1279 pmap->pm_flags = 0; 1280 pmap->pm_count = 1; 1281 pmap->pm_ptphint = NULL; 1282 TAILQ_INIT(&pmap->pm_pvlist); 1283 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1284} 1285 1286static int 1287pmap_release_free_page(pmap, p) 1288 struct pmap *pmap; 1289 vm_page_t p; 1290{ 1291 int s; 1292 unsigned *pde = (unsigned *) pmap->pm_pdir; 1293 /* 1294 * This code optimizes the case of freeing non-busy 1295 * page-table pages. Those pages are zero now, and 1296 * might as well be placed directly into the zero queue. 1297 */ 1298 if (vm_page_sleep(p, "pmaprl", NULL)) 1299 return 0; 1300 1301 p->flags |= PG_BUSY; 1302 1303 /* 1304 * Remove the page table page from the processes address space. 1305 */ 1306 pde[p->pindex] = 0; 1307 pmap->pm_stats.resident_count--; 1308 1309#if !defined(MAX_PERF) 1310 if (p->hold_count) { 1311 panic("pmap_release: freeing held page table page"); 1312 } 1313#endif 1314 /* 1315 * Page directory pages need to have the kernel 1316 * stuff cleared, so they can go into the zero queue also. 1317 */ 1318 if (p->pindex == PTDPTDI) { 1319 bzero(pde + KPTDI, nkpt * PTESIZE); 1320#ifdef SMP 1321 pde[MPPTDI] = 0; 1322#endif 1323 pde[APTDPTDI] = 0; 1324 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1325 } 1326 1327 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1328 pmap->pm_ptphint = NULL; 1329 1330 vm_page_free_zero(p); 1331 return 1; 1332} 1333 1334/* 1335 * this routine is called if the page table page is not 1336 * mapped correctly. 1337 */ 1338static vm_page_t 1339_pmap_allocpte(pmap, ptepindex) 1340 pmap_t pmap; 1341 unsigned ptepindex; 1342{ 1343 vm_offset_t pteva, ptepa; 1344 vm_page_t m; 1345 1346 /* 1347 * Find or fabricate a new pagetable page 1348 */ 1349 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1350 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1351 1352 if (m->queue != PQ_NONE) { 1353 int s = splvm(); 1354 vm_page_unqueue(m); 1355 splx(s); 1356 } 1357 1358 if (m->wire_count == 0) 1359 cnt.v_wire_count++; 1360 m->wire_count++; 1361 1362 /* 1363 * Increment the hold count for the page table page 1364 * (denoting a new mapping.) 1365 */ 1366 m->hold_count++; 1367 1368 /* 1369 * Map the pagetable page into the process address space, if 1370 * it isn't already there. 1371 */ 1372 1373 pmap->pm_stats.resident_count++; 1374 1375 ptepa = VM_PAGE_TO_PHYS(m); 1376 pmap->pm_pdir[ptepindex] = 1377 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1378 1379 /* 1380 * Set the page table hint 1381 */ 1382 pmap->pm_ptphint = m; 1383 1384 /* 1385 * Try to use the new mapping, but if we cannot, then 1386 * do it with the routine that maps the page explicitly. 1387 */ 1388 if ((m->flags & PG_ZERO) == 0) { 1389 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1390 (((unsigned) PTDpde) & PG_FRAME)) { 1391 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1392 bzero((caddr_t) pteva, PAGE_SIZE); 1393 } else { 1394 pmap_zero_page(ptepa); 1395 } 1396 } 1397 1398 m->valid = VM_PAGE_BITS_ALL; 1399 m->flags &= ~(PG_ZERO | PG_BUSY); 1400 m->flags |= PG_MAPPED; 1401 1402 return m; 1403} 1404 1405static vm_page_t 1406pmap_allocpte(pmap, va) 1407 pmap_t pmap; 1408 vm_offset_t va; 1409{ 1410 unsigned ptepindex; 1411 vm_offset_t ptepa; 1412 vm_page_t m; 1413 1414 /* 1415 * Calculate pagetable page index 1416 */ 1417 ptepindex = va >> PDRSHIFT; 1418 1419 /* 1420 * Get the page directory entry 1421 */ 1422 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1423 1424 /* 1425 * This supports switching from a 4MB page to a 1426 * normal 4K page. 1427 */ 1428 if (ptepa & PG_PS) { 1429 pmap->pm_pdir[ptepindex] = 0; 1430 ptepa = 0; 1431 invltlb(); 1432 } 1433 1434 /* 1435 * If the page table page is mapped, we just increment the 1436 * hold count, and activate it. 1437 */ 1438 if (ptepa) { 1439 /* 1440 * In order to get the page table page, try the 1441 * hint first. 1442 */ 1443 if (pmap->pm_ptphint && 1444 (pmap->pm_ptphint->pindex == ptepindex)) { 1445 m = pmap->pm_ptphint; 1446 } else { 1447 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1448 pmap->pm_ptphint = m; 1449 } 1450 m->hold_count++; 1451 return m; 1452 } 1453 /* 1454 * Here if the pte page isn't mapped, or if it has been deallocated. 1455 */ 1456 return _pmap_allocpte(pmap, ptepindex); 1457} 1458 1459 1460/*************************************************** 1461* Pmap allocation/deallocation routines. 1462 ***************************************************/ 1463 1464/* 1465 * Release any resources held by the given physical map. 1466 * Called when a pmap initialized by pmap_pinit is being released. 1467 * Should only be called if the map contains no valid mappings. 1468 */ 1469void 1470pmap_release(pmap) 1471 register struct pmap *pmap; 1472{ 1473 vm_page_t p,n,ptdpg; 1474 vm_object_t object = pmap->pm_pteobj; 1475 int curgeneration; 1476 1477#if defined(DIAGNOSTIC) 1478 if (object->ref_count != 1) 1479 panic("pmap_release: pteobj reference count != 1"); 1480#endif 1481 1482 ptdpg = NULL; 1483retry: 1484 curgeneration = object->generation; 1485 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1486 n = TAILQ_NEXT(p, listq); 1487 if (p->pindex == PTDPTDI) { 1488 ptdpg = p; 1489 continue; 1490 } 1491 while (1) { 1492 if (!pmap_release_free_page(pmap, p) && 1493 (object->generation != curgeneration)) 1494 goto retry; 1495 } 1496 } 1497 1498 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1499 goto retry; 1500} 1501 1502/* 1503 * grow the number of kernel page table entries, if needed 1504 */ 1505void 1506pmap_growkernel(vm_offset_t addr) 1507{ 1508 struct proc *p; 1509 struct pmap *pmap; 1510 int s; 1511 vm_offset_t ptppaddr; 1512 vm_page_t nkpg; 1513#ifdef SMP 1514 int i; 1515#endif 1516 pd_entry_t newpdir; 1517 1518 s = splhigh(); 1519 if (kernel_vm_end == 0) { 1520 kernel_vm_end = KERNBASE; 1521 nkpt = 0; 1522 while (pdir_pde(PTD, kernel_vm_end)) { 1523 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1524 nkpt++; 1525 } 1526 } 1527 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1528 while (kernel_vm_end < addr) { 1529 if (pdir_pde(PTD, kernel_vm_end)) { 1530 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1531 continue; 1532 } 1533 1534 /* 1535 * This index is bogus, but out of the way 1536 */ 1537 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1538#if !defined(MAX_PERF) 1539 if (!nkpg) 1540 panic("pmap_growkernel: no memory to grow kernel"); 1541#endif 1542 1543 nkpt++; 1544 1545 vm_page_wire(nkpg); 1546 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1547 pmap_zero_page(ptppaddr); 1548 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1549 pdir_pde(PTD, kernel_vm_end) = newpdir; 1550 1551#ifdef SMP 1552 for (i = 0; i < mp_ncpus; i++) { 1553 if (IdlePTDS[i]) 1554 pdir_pde(IdlePTDS[i], kernel_vm_end) = newpdir; 1555 } 1556#endif 1557 1558 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1559 if (p->p_vmspace) { 1560 pmap = &p->p_vmspace->vm_pmap; 1561 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1562 } 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 ppv->pv_vm_page->flags &= ~(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 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1791 } 1792 if (oldpte & PG_A) 1793 ppv->pv_vm_page->flags |= 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%lx", 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 ppv->pv_vm_page->flags |= 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 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 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 ppv->pv_vm_page->flags &= ~(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 ppv->pv_vm_page->flags |= 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 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 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%lx\n", 2176 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 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 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 int s; 2429 vm_page_t m[1]; 2430 unsigned int ptepindex; 2431 int npdes; 2432 vm_offset_t ptepa; 2433 2434 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2435 return; 2436 2437retry: 2438 p = vm_page_lookup(object, pindex); 2439 if (p && vm_page_sleep(p, "init4p", NULL)) 2440 goto retry; 2441 2442 if (p == NULL) { 2443 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2444 if (p == NULL) 2445 return; 2446 m[0] = p; 2447 2448 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2449 vm_page_free(p); 2450 return; 2451 } 2452 2453 p = vm_page_lookup(object, pindex); 2454 PAGE_WAKEUP(p); 2455 } 2456 2457 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2458 if (ptepa & (NBPDR - 1)) { 2459 return; 2460 } 2461 2462 p->valid = VM_PAGE_BITS_ALL; 2463 2464 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2465 npdes = size >> PDRSHIFT; 2466 for(i=0;i<npdes;i++) { 2467 pmap->pm_pdir[ptepindex] = 2468 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2469 ptepa += NBPDR; 2470 ptepindex += 1; 2471 } 2472 p->flags |= PG_MAPPED; 2473 invltlb(); 2474 return; 2475 } 2476 2477 psize = i386_btop(size); 2478 2479 if ((object->type != OBJT_VNODE) || 2480 (limit && (psize > MAX_INIT_PT) && 2481 (object->resident_page_count > MAX_INIT_PT))) { 2482 return; 2483 } 2484 2485 if (psize + pindex > object->size) 2486 psize = object->size - pindex; 2487 2488 mpte = NULL; 2489 /* 2490 * if we are processing a major portion of the object, then scan the 2491 * entire thing. 2492 */ 2493 if (psize > (object->size >> 2)) { 2494 objpgs = psize; 2495 2496 for (p = TAILQ_FIRST(&object->memq); 2497 ((objpgs > 0) && (p != NULL)); 2498 p = TAILQ_NEXT(p, listq)) { 2499 2500 tmpidx = p->pindex; 2501 if (tmpidx < pindex) { 2502 continue; 2503 } 2504 tmpidx -= pindex; 2505 if (tmpidx >= psize) { 2506 continue; 2507 } 2508 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2509 (p->busy == 0) && 2510 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2511 if ((p->queue - p->pc) == PQ_CACHE) 2512 vm_page_deactivate(p); 2513 p->flags |= PG_BUSY; 2514 mpte = pmap_enter_quick(pmap, 2515 addr + i386_ptob(tmpidx), 2516 VM_PAGE_TO_PHYS(p), mpte); 2517 p->flags |= PG_MAPPED; 2518 PAGE_WAKEUP(p); 2519 } 2520 objpgs -= 1; 2521 } 2522 } else { 2523 /* 2524 * else lookup the pages one-by-one. 2525 */ 2526 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2527 p = vm_page_lookup(object, tmpidx + pindex); 2528 if (p && 2529 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2530 (p->busy == 0) && 2531 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2532 if ((p->queue - p->pc) == PQ_CACHE) 2533 vm_page_deactivate(p); 2534 p->flags |= PG_BUSY; 2535 mpte = pmap_enter_quick(pmap, 2536 addr + i386_ptob(tmpidx), 2537 VM_PAGE_TO_PHYS(p), mpte); 2538 p->flags |= PG_MAPPED; 2539 PAGE_WAKEUP(p); 2540 } 2541 } 2542 } 2543 return; 2544} 2545 2546/* 2547 * pmap_prefault provides a quick way of clustering 2548 * pagefaults into a processes address space. It is a "cousin" 2549 * of pmap_object_init_pt, except it runs at page fault time instead 2550 * of mmap time. 2551 */ 2552#define PFBAK 4 2553#define PFFOR 4 2554#define PAGEORDER_SIZE (PFBAK+PFFOR) 2555 2556static int pmap_prefault_pageorder[] = { 2557 -PAGE_SIZE, PAGE_SIZE, 2558 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2559 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2560 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2561}; 2562 2563void 2564pmap_prefault(pmap, addra, entry) 2565 pmap_t pmap; 2566 vm_offset_t addra; 2567 vm_map_entry_t entry; 2568{ 2569 int i; 2570 vm_offset_t starta; 2571 vm_offset_t addr; 2572 vm_pindex_t pindex; 2573 vm_page_t m, mpte; 2574 vm_object_t object; 2575 2576 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) 2577 return; 2578 2579 object = entry->object.vm_object; 2580 2581 starta = addra - PFBAK * PAGE_SIZE; 2582 if (starta < entry->start) { 2583 starta = entry->start; 2584 } else if (starta > addra) { 2585 starta = 0; 2586 } 2587 2588 mpte = NULL; 2589 for (i = 0; i < PAGEORDER_SIZE; i++) { 2590 vm_object_t lobject; 2591 unsigned *pte; 2592 2593 addr = addra + pmap_prefault_pageorder[i]; 2594 if (addr > addra + (PFFOR * PAGE_SIZE)) 2595 addr = 0; 2596 2597 if (addr < starta || addr >= entry->end) 2598 continue; 2599 2600 if ((*pmap_pde(pmap, addr)) == NULL) 2601 continue; 2602 2603 pte = (unsigned *) vtopte(addr); 2604 if (*pte) 2605 continue; 2606 2607 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2608 lobject = object; 2609 for (m = vm_page_lookup(lobject, pindex); 2610 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2611 lobject = lobject->backing_object) { 2612 if (lobject->backing_object_offset & PAGE_MASK) 2613 break; 2614 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2615 m = vm_page_lookup(lobject->backing_object, pindex); 2616 } 2617 2618 /* 2619 * give-up when a page is not in memory 2620 */ 2621 if (m == NULL) 2622 break; 2623 2624 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2625 (m->busy == 0) && 2626 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2627 2628 if ((m->queue - m->pc) == PQ_CACHE) { 2629 vm_page_deactivate(m); 2630 } 2631 m->flags |= PG_BUSY; 2632 mpte = pmap_enter_quick(pmap, addr, 2633 VM_PAGE_TO_PHYS(m), mpte); 2634 m->flags |= PG_MAPPED; 2635 PAGE_WAKEUP(m); 2636 } 2637 } 2638} 2639 2640/* 2641 * Routine: pmap_change_wiring 2642 * Function: Change the wiring attribute for a map/virtual-address 2643 * pair. 2644 * In/out conditions: 2645 * The mapping must already exist in the pmap. 2646 */ 2647void 2648pmap_change_wiring(pmap, va, wired) 2649 register pmap_t pmap; 2650 vm_offset_t va; 2651 boolean_t wired; 2652{ 2653 register unsigned *pte; 2654 2655 if (pmap == NULL) 2656 return; 2657 2658 pte = pmap_pte(pmap, va); 2659 2660 if (wired && !pmap_pte_w(pte)) 2661 pmap->pm_stats.wired_count++; 2662 else if (!wired && pmap_pte_w(pte)) 2663 pmap->pm_stats.wired_count--; 2664 2665 /* 2666 * Wiring is not a hardware characteristic so there is no need to 2667 * invalidate TLB. 2668 */ 2669 pmap_pte_set_w(pte, wired); 2670} 2671 2672 2673 2674/* 2675 * Copy the range specified by src_addr/len 2676 * from the source map to the range dst_addr/len 2677 * in the destination map. 2678 * 2679 * This routine is only advisory and need not do anything. 2680 */ 2681 2682void 2683pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2684 pmap_t dst_pmap, src_pmap; 2685 vm_offset_t dst_addr; 2686 vm_size_t len; 2687 vm_offset_t src_addr; 2688{ 2689 vm_offset_t addr; 2690 vm_offset_t end_addr = src_addr + len; 2691 vm_offset_t pdnxt; 2692 unsigned src_frame, dst_frame; 2693 2694 if (dst_addr != src_addr) 2695 return; 2696 2697 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2698 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2699 return; 2700 } 2701 2702 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2703 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2704 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2705 invltlb(); 2706 } 2707 2708 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2709 unsigned *src_pte, *dst_pte; 2710 vm_page_t dstmpte, srcmpte; 2711 vm_offset_t srcptepaddr; 2712 unsigned ptepindex; 2713 2714#if !defined(MAX_PERF) 2715 if (addr >= UPT_MIN_ADDRESS) 2716 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2717#endif 2718 2719 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2720 ptepindex = addr >> PDRSHIFT; 2721 2722 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2723 if (srcptepaddr == 0) 2724 continue; 2725 2726 if (srcptepaddr & PG_PS) { 2727 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2728 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2729 dst_pmap->pm_stats.resident_count += NBPDR; 2730 } 2731 continue; 2732 } 2733 2734 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2735 if ((srcmpte == NULL) || 2736 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2737 continue; 2738 2739 if (pdnxt > end_addr) 2740 pdnxt = end_addr; 2741 2742 src_pte = (unsigned *) vtopte(addr); 2743 dst_pte = (unsigned *) avtopte(addr); 2744 while (addr < pdnxt) { 2745 unsigned ptetemp; 2746 ptetemp = *src_pte; 2747 /* 2748 * we only virtual copy managed pages 2749 */ 2750 if ((ptetemp & PG_MANAGED) != 0) { 2751 /* 2752 * We have to check after allocpte for the 2753 * pte still being around... allocpte can 2754 * block. 2755 */ 2756 dstmpte = pmap_allocpte(dst_pmap, addr); 2757 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2758 /* 2759 * Clear the modified and 2760 * accessed (referenced) bits 2761 * during the copy. 2762 */ 2763 *dst_pte = ptetemp & ~(PG_M | PG_A); 2764 dst_pmap->pm_stats.resident_count++; 2765 pmap_insert_entry(dst_pmap, addr, 2766 dstmpte, 2767 (ptetemp & PG_FRAME)); 2768 } else { 2769 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2770 } 2771 if (dstmpte->hold_count >= srcmpte->hold_count) 2772 break; 2773 } 2774 addr += PAGE_SIZE; 2775 src_pte++; 2776 dst_pte++; 2777 } 2778 } 2779} 2780 2781/* 2782 * Routine: pmap_kernel 2783 * Function: 2784 * Returns the physical map handle for the kernel. 2785 */ 2786pmap_t 2787pmap_kernel() 2788{ 2789 return (kernel_pmap); 2790} 2791 2792/* 2793 * pmap_zero_page zeros the specified (machine independent) 2794 * page by mapping the page into virtual memory and using 2795 * bzero to clear its contents, one machine dependent page 2796 * at a time. 2797 */ 2798void 2799pmap_zero_page(phys) 2800 vm_offset_t phys; 2801{ 2802#ifdef SMP 2803#if !defined(MAX_PERF) 2804 if (*(int *) prv_CMAP3) 2805 panic("pmap_zero_page: prv_CMAP3 busy"); 2806#endif 2807 2808 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2809 cpu_invlpg(&prv_CPAGE3); 2810 2811#if defined(I686_CPU) 2812 if (cpu == CPU_686) 2813 i686_pagezero(&prv_CPAGE3); 2814 else 2815#endif 2816 bzero(&prv_CPAGE3, PAGE_SIZE); 2817 2818 *(int *) prv_CMAP3 = 0; 2819#else 2820#if !defined(MAX_PERF) 2821 if (*(int *) CMAP2) 2822 panic("pmap_zero_page: CMAP2 busy"); 2823#endif 2824 2825 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2826 if (cpu_class == CPUCLASS_386) { 2827 invltlb(); 2828 } else { 2829 invlpg((u_int)CADDR2); 2830 } 2831 2832#if defined(I686_CPU) 2833 if (cpu == CPU_686) 2834 i686_pagezero(CADDR2); 2835 else 2836#endif 2837 bzero(CADDR2, PAGE_SIZE); 2838 *(int *) CMAP2 = 0; 2839#endif 2840} 2841 2842/* 2843 * pmap_copy_page copies the specified (machine independent) 2844 * page by mapping the page into virtual memory and using 2845 * bcopy to copy the page, one machine dependent page at a 2846 * time. 2847 */ 2848void 2849pmap_copy_page(src, dst) 2850 vm_offset_t src; 2851 vm_offset_t dst; 2852{ 2853#ifdef SMP 2854#if !defined(MAX_PERF) 2855 if (*(int *) prv_CMAP1) 2856 panic("pmap_copy_page: prv_CMAP1 busy"); 2857 if (*(int *) prv_CMAP2) 2858 panic("pmap_copy_page: prv_CMAP2 busy"); 2859#endif 2860 2861 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2862 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2863 2864 cpu_invlpg(&prv_CPAGE1); 2865 cpu_invlpg(&prv_CPAGE2); 2866 2867 bcopy(&prv_CPAGE1, &prv_CPAGE2, PAGE_SIZE); 2868 2869 *(int *) prv_CMAP1 = 0; 2870 *(int *) prv_CMAP2 = 0; 2871#else 2872#if !defined(MAX_PERF) 2873 if (*(int *) CMAP1 || *(int *) CMAP2) 2874 panic("pmap_copy_page: CMAP busy"); 2875#endif 2876 2877 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2878 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2879 if (cpu_class == CPUCLASS_386) { 2880 invltlb(); 2881 } else { 2882 invlpg((u_int)CADDR1); 2883 invlpg((u_int)CADDR2); 2884 } 2885 2886 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2887 2888 *(int *) CMAP1 = 0; 2889 *(int *) CMAP2 = 0; 2890#endif 2891} 2892 2893 2894/* 2895 * Routine: pmap_pageable 2896 * Function: 2897 * Make the specified pages (by pmap, offset) 2898 * pageable (or not) as requested. 2899 * 2900 * A page which is not pageable may not take 2901 * a fault; therefore, its page table entry 2902 * must remain valid for the duration. 2903 * 2904 * This routine is merely advisory; pmap_enter 2905 * will specify that these pages are to be wired 2906 * down (or not) as appropriate. 2907 */ 2908void 2909pmap_pageable(pmap, sva, eva, pageable) 2910 pmap_t pmap; 2911 vm_offset_t sva, eva; 2912 boolean_t pageable; 2913{ 2914} 2915 2916/* 2917 * this routine returns true if a physical page resides 2918 * in the given pmap. 2919 */ 2920boolean_t 2921pmap_page_exists(pmap, pa) 2922 pmap_t pmap; 2923 vm_offset_t pa; 2924{ 2925 register pv_entry_t pv; 2926 pv_table_t *ppv; 2927 int s; 2928 2929 if (!pmap_is_managed(pa)) 2930 return FALSE; 2931 2932 s = splvm(); 2933 2934 ppv = pa_to_pvh(pa); 2935 /* 2936 * Not found, check current mappings returning immediately if found. 2937 */ 2938 for (pv = TAILQ_FIRST(&ppv->pv_list); 2939 pv; 2940 pv = TAILQ_NEXT(pv, pv_list)) { 2941 if (pv->pv_pmap == pmap) { 2942 splx(s); 2943 return TRUE; 2944 } 2945 } 2946 splx(s); 2947 return (FALSE); 2948} 2949 2950#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2951/* 2952 * Remove all pages from specified address space 2953 * this aids process exit speeds. Also, this code 2954 * is special cased for current process only, but 2955 * can have the more generic (and slightly slower) 2956 * mode enabled. This is much faster than pmap_remove 2957 * in the case of running down an entire address space. 2958 */ 2959void 2960pmap_remove_pages(pmap, sva, eva) 2961 pmap_t pmap; 2962 vm_offset_t sva, eva; 2963{ 2964 unsigned *pte, tpte; 2965 pv_table_t *ppv; 2966 pv_entry_t pv, npv; 2967 int s; 2968 2969#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2970 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) { 2971 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2972 return; 2973 } 2974#endif 2975 2976 s = splvm(); 2977 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2978 pv; 2979 pv = npv) { 2980 2981 if (pv->pv_va >= eva || pv->pv_va < sva) { 2982 npv = TAILQ_NEXT(pv, pv_plist); 2983 continue; 2984 } 2985 2986#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2987 pte = (unsigned *)vtopte(pv->pv_va); 2988#else 2989 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2990#endif 2991 tpte = *pte; 2992 2993/* 2994 * We cannot remove wired pages from a process' mapping at this time 2995 */ 2996 if (tpte & PG_W) { 2997 npv = TAILQ_NEXT(pv, pv_plist); 2998 continue; 2999 } 3000 *pte = 0; 3001 3002 ppv = pa_to_pvh(tpte); 3003 3004 pv->pv_pmap->pm_stats.resident_count--; 3005 3006 /* 3007 * Update the vm_page_t clean and reference bits. 3008 */ 3009 if (tpte & PG_M) { 3010 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 3011 } 3012 3013 3014 npv = TAILQ_NEXT(pv, pv_plist); 3015 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 3016 3017 ppv->pv_list_count--; 3018 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 3019 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 3020 ppv->pv_vm_page->flags &= ~(PG_MAPPED | PG_WRITEABLE); 3021 } 3022 3023 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 3024 free_pv_entry(pv); 3025 } 3026 splx(s); 3027 invltlb(); 3028} 3029 3030/* 3031 * pmap_testbit tests bits in pte's 3032 * note that the testbit/changebit routines are inline, 3033 * and a lot of things compile-time evaluate. 3034 */ 3035static boolean_t 3036pmap_testbit(pa, bit) 3037 register vm_offset_t pa; 3038 int bit; 3039{ 3040 register pv_entry_t pv; 3041 pv_table_t *ppv; 3042 unsigned *pte; 3043 int s; 3044 3045 if (!pmap_is_managed(pa)) 3046 return FALSE; 3047 3048 ppv = pa_to_pvh(pa); 3049 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 3050 return FALSE; 3051 3052 s = splvm(); 3053 3054 for (pv = TAILQ_FIRST(&ppv->pv_list); 3055 pv; 3056 pv = TAILQ_NEXT(pv, pv_list)) { 3057 3058 /* 3059 * if the bit being tested is the modified bit, then 3060 * mark clean_map and ptes as never 3061 * modified. 3062 */ 3063 if (bit & (PG_A|PG_M)) { 3064 if (!pmap_track_modified(pv->pv_va)) 3065 continue; 3066 } 3067 3068#if defined(PMAP_DIAGNOSTIC) 3069 if (!pv->pv_pmap) { 3070 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 3071 continue; 3072 } 3073#endif 3074 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3075 if (*pte & bit) { 3076 splx(s); 3077 return TRUE; 3078 } 3079 } 3080 splx(s); 3081 return (FALSE); 3082} 3083 3084/* 3085 * this routine is used to modify bits in ptes 3086 */ 3087static void 3088pmap_changebit(pa, bit, setem) 3089 vm_offset_t pa; 3090 int bit; 3091 boolean_t setem; 3092{ 3093 register pv_entry_t pv; 3094 pv_table_t *ppv; 3095 register unsigned *pte; 3096 int changed; 3097 int s; 3098 3099 if (!pmap_is_managed(pa)) 3100 return; 3101 3102 s = splvm(); 3103 changed = 0; 3104 ppv = pa_to_pvh(pa); 3105 3106 /* 3107 * Loop over all current mappings setting/clearing as appropos If 3108 * setting RO do we need to clear the VAC? 3109 */ 3110 for (pv = TAILQ_FIRST(&ppv->pv_list); 3111 pv; 3112 pv = TAILQ_NEXT(pv, pv_list)) { 3113 3114 /* 3115 * don't write protect pager mappings 3116 */ 3117 if (!setem && (bit == PG_RW)) { 3118 if (!pmap_track_modified(pv->pv_va)) 3119 continue; 3120 } 3121 3122#if defined(PMAP_DIAGNOSTIC) 3123 if (!pv->pv_pmap) { 3124 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 3125 continue; 3126 } 3127#endif 3128 3129 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3130 3131 if (setem) { 3132 *(int *)pte |= bit; 3133 changed = 1; 3134 } else { 3135 vm_offset_t pbits = *(vm_offset_t *)pte; 3136 if (pbits & bit) { 3137 changed = 1; 3138 if (bit == PG_RW) { 3139 if (pbits & PG_M) { 3140 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 3141 } 3142 *(int *)pte = pbits & ~(PG_M|PG_RW); 3143 } else { 3144 *(int *)pte = pbits & ~bit; 3145 } 3146 } 3147 } 3148 } 3149 splx(s); 3150 if (changed) 3151 invltlb(); 3152} 3153 3154/* 3155 * pmap_page_protect: 3156 * 3157 * Lower the permission for all mappings to a given page. 3158 */ 3159void 3160pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 3161{ 3162 if ((prot & VM_PROT_WRITE) == 0) { 3163 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3164 pmap_changebit(phys, PG_RW, FALSE); 3165 } else { 3166 pmap_remove_all(phys); 3167 } 3168 } 3169} 3170 3171vm_offset_t 3172pmap_phys_address(ppn) 3173 int ppn; 3174{ 3175 return (i386_ptob(ppn)); 3176} 3177 3178/* 3179 * pmap_ts_referenced: 3180 * 3181 * Return the count of reference bits for a page, clearing all of them. 3182 * 3183 */ 3184int 3185pmap_ts_referenced(vm_offset_t pa) 3186{ 3187 register pv_entry_t pv; 3188 pv_table_t *ppv; 3189 unsigned *pte; 3190 int s; 3191 int rtval = 0; 3192 3193 if (!pmap_is_managed(pa)) 3194 return FALSE; 3195 3196 s = splvm(); 3197 3198 ppv = pa_to_pvh(pa); 3199 3200 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 3201 splx(s); 3202 return 0; 3203 } 3204 3205 /* 3206 * Not found, check current mappings returning immediately if found. 3207 */ 3208 for (pv = TAILQ_FIRST(&ppv->pv_list); 3209 pv; 3210 pv = TAILQ_NEXT(pv, pv_list)) { 3211 3212 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 3213 /* 3214 * if the bit being tested is the modified bit, then 3215 * mark clean_map and ptes as never 3216 * modified. 3217 */ 3218 if (!pmap_track_modified(pv->pv_va)) { 3219 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3220 continue; 3221 } 3222 3223 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3224 if (pte == NULL) { 3225 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3226 continue; 3227 } 3228 3229 if (*pte & PG_A) { 3230 rtval++; 3231 *pte &= ~PG_A; 3232 if (rtval > 4) { 3233 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3234 break; 3235 } 3236 } 3237 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 3238 } 3239 3240 splx(s); 3241 if (rtval) { 3242 invltlb(); 3243 } 3244 return (rtval); 3245} 3246 3247/* 3248 * pmap_is_modified: 3249 * 3250 * Return whether or not the specified physical page was modified 3251 * in any physical maps. 3252 */ 3253boolean_t 3254pmap_is_modified(vm_offset_t pa) 3255{ 3256 return pmap_testbit((pa), PG_M); 3257} 3258 3259/* 3260 * Clear the modify bits on the specified physical page. 3261 */ 3262void 3263pmap_clear_modify(vm_offset_t pa) 3264{ 3265 pmap_changebit((pa), PG_M, FALSE); 3266} 3267 3268/* 3269 * pmap_clear_reference: 3270 * 3271 * Clear the reference bit on the specified physical page. 3272 */ 3273void 3274pmap_clear_reference(vm_offset_t pa) 3275{ 3276 pmap_changebit((pa), PG_A, FALSE); 3277} 3278 3279/* 3280 * Miscellaneous support routines follow 3281 */ 3282 3283static void 3284i386_protection_init() 3285{ 3286 register int *kp, prot; 3287 3288 kp = protection_codes; 3289 for (prot = 0; prot < 8; prot++) { 3290 switch (prot) { 3291 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3292 /* 3293 * Read access is also 0. There isn't any execute bit, 3294 * so just make it readable. 3295 */ 3296 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3297 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3298 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3299 *kp++ = 0; 3300 break; 3301 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3302 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3303 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3304 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3305 *kp++ = PG_RW; 3306 break; 3307 } 3308 } 3309} 3310 3311/* 3312 * Map a set of physical memory pages into the kernel virtual 3313 * address space. Return a pointer to where it is mapped. This 3314 * routine is intended to be used for mapping device memory, 3315 * NOT real memory. 3316 */ 3317void * 3318pmap_mapdev(pa, size) 3319 vm_offset_t pa; 3320 vm_size_t size; 3321{ 3322 vm_offset_t va, tmpva; 3323 unsigned *pte; 3324 3325 size = roundup(size, PAGE_SIZE); 3326 3327 va = kmem_alloc_pageable(kernel_map, size); 3328#if !defined(MAX_PERF) 3329 if (!va) 3330 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3331#endif 3332 3333 pa = pa & PG_FRAME; 3334 for (tmpva = va; size > 0;) { 3335 pte = (unsigned *)vtopte(tmpva); 3336 *pte = pa | PG_RW | PG_V | pgeflag; 3337 size -= PAGE_SIZE; 3338 tmpva += PAGE_SIZE; 3339 pa += PAGE_SIZE; 3340 } 3341 invltlb(); 3342 3343 return ((void *) va); 3344} 3345 3346/* 3347 * perform the pmap work for mincore 3348 */ 3349int 3350pmap_mincore(pmap, addr) 3351 pmap_t pmap; 3352 vm_offset_t addr; 3353{ 3354 3355 unsigned *ptep, pte; 3356 vm_page_t m; 3357 int val = 0; 3358 3359 ptep = pmap_pte(pmap, addr); 3360 if (ptep == 0) { 3361 return 0; 3362 } 3363 3364 if (pte = *ptep) { 3365 pv_table_t *ppv; 3366 vm_offset_t pa; 3367 3368 val = MINCORE_INCORE; 3369 if ((pte & PG_MANAGED) == 0) 3370 return val; 3371 3372 pa = pte & PG_FRAME; 3373 3374 ppv = pa_to_pvh((pa & PG_FRAME)); 3375 m = ppv->pv_vm_page; 3376 3377 /* 3378 * Modified by us 3379 */ 3380 if (pte & PG_M) 3381 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3382 /* 3383 * Modified by someone 3384 */ 3385 else if (m->dirty || pmap_is_modified(pa)) 3386 val |= MINCORE_MODIFIED_OTHER; 3387 /* 3388 * Referenced by us 3389 */ 3390 if (pte & PG_A) 3391 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3392 3393 /* 3394 * Referenced by someone 3395 */ 3396 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(pa)) { 3397 val |= MINCORE_REFERENCED_OTHER; 3398 m->flags |= PG_REFERENCED; 3399 } 3400 } 3401 return val; 3402} 3403 3404void 3405pmap_activate(struct proc *p) 3406{ 3407#if defined(SWTCH_OPTIM_STATS) 3408 tlb_flush_count++; 3409#endif 3410 load_cr3(p->p_addr->u_pcb.pcb_cr3 = 3411 vtophys(p->p_vmspace->vm_pmap.pm_pdir)); 3412} 3413 3414vm_offset_t 3415pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3416 3417 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3418 return addr; 3419 } 3420 3421 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3422 return addr; 3423} 3424 3425 3426#if defined(PMAP_DEBUG) 3427pmap_pid_dump(int pid) { 3428 pmap_t pmap; 3429 struct proc *p; 3430 int npte = 0; 3431 int index; 3432 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3433 if (p->p_pid != pid) 3434 continue; 3435 3436 if (p->p_vmspace) { 3437 int i,j; 3438 index = 0; 3439 pmap = &p->p_vmspace->vm_pmap; 3440 for(i=0;i<1024;i++) { 3441 pd_entry_t *pde; 3442 unsigned *pte; 3443 unsigned base = i << PDRSHIFT; 3444 3445 pde = &pmap->pm_pdir[i]; 3446 if (pde && pmap_pde_v(pde)) { 3447 for(j=0;j<1024;j++) { 3448 unsigned va = base + (j << PAGE_SHIFT); 3449 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3450 if (index) { 3451 index = 0; 3452 printf("\n"); 3453 } 3454 return npte; 3455 } 3456 pte = pmap_pte_quick( pmap, va); 3457 if (pte && pmap_pte_v(pte)) { 3458 vm_offset_t pa; 3459 vm_page_t m; 3460 pa = *(int *)pte; 3461 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3462 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3463 va, pa, m->hold_count, m->wire_count, m->flags); 3464 npte++; 3465 index++; 3466 if (index >= 2) { 3467 index = 0; 3468 printf("\n"); 3469 } else { 3470 printf(" "); 3471 } 3472 } 3473 } 3474 } 3475 } 3476 } 3477 } 3478 return npte; 3479} 3480#endif 3481 3482#if defined(DEBUG) 3483 3484static void pads __P((pmap_t pm)); 3485static void pmap_pvdump __P((vm_offset_t pa)); 3486 3487/* print address space of pmap*/ 3488static void 3489pads(pm) 3490 pmap_t pm; 3491{ 3492 unsigned va, i, j; 3493 unsigned *ptep; 3494 3495 if (pm == kernel_pmap) 3496 return; 3497 for (i = 0; i < 1024; i++) 3498 if (pm->pm_pdir[i]) 3499 for (j = 0; j < 1024; j++) { 3500 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3501 if (pm == kernel_pmap && va < KERNBASE) 3502 continue; 3503 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3504 continue; 3505 ptep = pmap_pte_quick(pm, va); 3506 if (pmap_pte_v(ptep)) 3507 printf("%x:%x ", va, *(int *) ptep); 3508 }; 3509 3510} 3511 3512static void 3513pmap_pvdump(pa) 3514 vm_offset_t pa; 3515{ 3516 pv_table_t *ppv; 3517 register pv_entry_t pv; 3518 3519 printf("pa %x", pa); 3520 ppv = pa_to_pvh(pa); 3521 for (pv = TAILQ_FIRST(&ppv->pv_list); 3522 pv; 3523 pv = TAILQ_NEXT(pv, pv_list)) { 3524#ifdef used_to_be 3525 printf(" -> pmap %p, va %x, flags %x", 3526 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags); 3527#endif 3528 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3529 pads(pv->pv_pmap); 3530 } 3531 printf(" "); 3532} 3533#endif 3534