pmap.c revision 76166
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 * $FreeBSD: head/sys/i386/i386/pmap.c 76166 2001-05-01 08:13:21Z markm $ 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/lock.h> 78#include <sys/mman.h> 79#include <sys/msgbuf.h> 80#include <sys/proc.h> 81#include <sys/sx.h> 82#include <sys/user.h> 83#include <sys/vmmeter.h> 84 85#include <vm/vm.h> 86#include <vm/vm_param.h> 87#include <vm/vm_kern.h> 88#include <vm/vm_page.h> 89#include <vm/vm_map.h> 90#include <vm/vm_object.h> 91#include <vm/vm_extern.h> 92#include <vm/vm_pageout.h> 93#include <vm/vm_pager.h> 94#include <vm/vm_zone.h> 95 96#include <machine/cputypes.h> 97#include <machine/md_var.h> 98#include <machine/specialreg.h> 99#if defined(SMP) || defined(APIC_IO) 100#include <machine/smp.h> 101#include <machine/apic.h> 102#include <machine/segments.h> 103#include <machine/tss.h> 104#include <machine/globaldata.h> 105#endif /* SMP || APIC_IO */ 106 107#define PMAP_KEEP_PDIRS 108#ifndef PMAP_SHPGPERPROC 109#define PMAP_SHPGPERPROC 200 110#endif 111 112#if defined(DIAGNOSTIC) 113#define PMAP_DIAGNOSTIC 114#endif 115 116#define MINPV 2048 117 118#if !defined(PMAP_DIAGNOSTIC) 119#define PMAP_INLINE __inline 120#else 121#define PMAP_INLINE 122#endif 123 124/* 125 * Get PDEs and PTEs for user/kernel address space 126 */ 127#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 128#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 129 130#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 131#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 132#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 133#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 134#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 135 136#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 137#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 138 139/* 140 * Given a map and a machine independent protection code, 141 * convert to a vax protection code. 142 */ 143#define pte_prot(m, p) (protection_codes[p]) 144static int protection_codes[8]; 145 146static struct pmap kernel_pmap_store; 147pmap_t kernel_pmap; 148LIST_HEAD(pmaplist, pmap); 149struct pmaplist allpmaps; 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 int pgeflag; /* PG_G or-in */ 157static int pseflag; /* PG_PS or-in */ 158 159static vm_object_t kptobj; 160 161static int nkpt; 162vm_offset_t kernel_vm_end; 163 164/* 165 * Data for the pv entry allocation mechanism 166 */ 167static vm_zone_t pvzone; 168static struct vm_zone pvzone_store; 169static struct vm_object pvzone_obj; 170static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0; 171static int pmap_pagedaemon_waken = 0; 172static struct pv_entry *pvinit; 173 174/* 175 * All those kernel PT submaps that BSD is so fond of 176 */ 177pt_entry_t *CMAP1 = 0; 178static pt_entry_t *CMAP2, *ptmmap; 179caddr_t CADDR1 = 0, ptvmmap = 0; 180static caddr_t CADDR2; 181static pt_entry_t *msgbufmap; 182struct msgbuf *msgbufp=0; 183 184/* 185 * Crashdump maps. 186 */ 187static pt_entry_t *pt_crashdumpmap; 188static caddr_t crashdumpmap; 189 190#ifdef SMP 191extern pt_entry_t *SMPpt; 192#endif 193static pt_entry_t *PMAP1 = 0; 194static unsigned *PADDR1 = 0; 195 196static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 197static unsigned * get_ptbase __P((pmap_t pmap)); 198static pv_entry_t get_pv_entry __P((void)); 199static void i386_protection_init __P((void)); 200static __inline void pmap_changebit __P((vm_page_t m, int bit, boolean_t setem)); 201 202static void pmap_remove_all __P((vm_page_t m)); 203static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 204 vm_page_t m, vm_page_t mpte)); 205static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 206 vm_offset_t sva)); 207static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 208static int pmap_remove_entry __P((struct pmap *pmap, vm_page_t m, 209 vm_offset_t va)); 210static boolean_t pmap_testbit __P((vm_page_t m, int bit)); 211static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 212 vm_page_t mpte, vm_page_t m)); 213 214static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 215 216static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 217static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 218static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 219static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 220static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 221static vm_offset_t pmap_kmem_choose(vm_offset_t addr); 222 223static unsigned pdir4mb; 224 225/* 226 * Routine: pmap_pte 227 * Function: 228 * Extract the page table entry associated 229 * with the given map/virtual_address pair. 230 */ 231 232PMAP_INLINE unsigned * 233pmap_pte(pmap, va) 234 register pmap_t pmap; 235 vm_offset_t va; 236{ 237 unsigned *pdeaddr; 238 239 if (pmap) { 240 pdeaddr = (unsigned *) pmap_pde(pmap, va); 241 if (*pdeaddr & PG_PS) 242 return pdeaddr; 243 if (*pdeaddr) { 244 return get_ptbase(pmap) + i386_btop(va); 245 } 246 } 247 return (0); 248} 249 250/* 251 * Move the kernel virtual free pointer to the next 252 * 4MB. This is used to help improve performance 253 * by using a large (4MB) page for much of the kernel 254 * (.text, .data, .bss) 255 */ 256static vm_offset_t 257pmap_kmem_choose(vm_offset_t addr) 258{ 259 vm_offset_t newaddr = addr; 260#ifndef DISABLE_PSE 261 if (cpu_feature & CPUID_PSE) { 262 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 263 } 264#endif 265 return newaddr; 266} 267 268/* 269 * Bootstrap the system enough to run with virtual memory. 270 * 271 * On the i386 this is called after mapping has already been enabled 272 * and just syncs the pmap module with what has already been done. 273 * [We can't call it easily with mapping off since the kernel is not 274 * mapped with PA == VA, hence we would have to relocate every address 275 * from the linked base (virtual) address "KERNBASE" to the actual 276 * (physical) address starting relative to 0] 277 */ 278void 279pmap_bootstrap(firstaddr, loadaddr) 280 vm_offset_t firstaddr; 281 vm_offset_t loadaddr; 282{ 283 vm_offset_t va; 284 pt_entry_t *pte; 285 int i; 286 287 avail_start = firstaddr; 288 289 /* 290 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 291 * large. It should instead be correctly calculated in locore.s and 292 * not based on 'first' (which is a physical address, not a virtual 293 * address, for the start of unused physical memory). The kernel 294 * page tables are NOT double mapped and thus should not be included 295 * in this calculation. 296 */ 297 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 298 virtual_avail = pmap_kmem_choose(virtual_avail); 299 300 virtual_end = VM_MAX_KERNEL_ADDRESS; 301 302 /* 303 * Initialize protection array. 304 */ 305 i386_protection_init(); 306 307 /* 308 * The kernel's pmap is statically allocated so we don't have to use 309 * pmap_create, which is unlikely to work correctly at this part of 310 * the boot sequence (XXX and which no longer exists). 311 */ 312 kernel_pmap = &kernel_pmap_store; 313 314 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 315 kernel_pmap->pm_count = 1; 316 kernel_pmap->pm_active = -1; /* don't allow deactivation */ 317 TAILQ_INIT(&kernel_pmap->pm_pvlist); 318 LIST_INIT(&allpmaps); 319 LIST_INSERT_HEAD(&allpmaps, kernel_pmap, pm_list); 320 nkpt = NKPT; 321 322 /* 323 * Reserve some special page table entries/VA space for temporary 324 * mapping of pages. 325 */ 326#define SYSMAP(c, p, v, n) \ 327 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 328 329 va = virtual_avail; 330 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 331 332 /* 333 * CMAP1/CMAP2 are used for zeroing and copying pages. 334 */ 335 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 336 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 337 338 /* 339 * Crashdump maps. 340 */ 341 SYSMAP(caddr_t, pt_crashdumpmap, crashdumpmap, MAXDUMPPGS); 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 /* 357 * ptemap is used for pmap_pte_quick 358 */ 359 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 360 361 virtual_avail = va; 362 363 *(int *) CMAP1 = *(int *) CMAP2 = 0; 364 for (i = 0; i < NKPT; i++) 365 PTD[i] = 0; 366 367 pgeflag = 0; 368#if !defined(SMP) /* XXX - see also mp_machdep.c */ 369 if (cpu_feature & CPUID_PGE) { 370 pgeflag = PG_G; 371 } 372#endif 373 374/* 375 * Initialize the 4MB page size flag 376 */ 377 pseflag = 0; 378/* 379 * The 4MB page version of the initial 380 * kernel page mapping. 381 */ 382 pdir4mb = 0; 383 384#if !defined(DISABLE_PSE) 385 if (cpu_feature & CPUID_PSE) { 386 unsigned ptditmp; 387 /* 388 * Note that we have enabled PSE mode 389 */ 390 pseflag = PG_PS; 391 ptditmp = *((unsigned *)PTmap + i386_btop(KERNBASE)); 392 ptditmp &= ~(NBPDR - 1); 393 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 394 pdir4mb = ptditmp; 395 396#if !defined(SMP) 397 /* 398 * Enable the PSE mode. 399 */ 400 load_cr4(rcr4() | CR4_PSE); 401 402 /* 403 * We can do the mapping here for the single processor 404 * case. We simply ignore the old page table page from 405 * now on. 406 */ 407 /* 408 * For SMP, we still need 4K pages to bootstrap APs, 409 * PSE will be enabled as soon as all APs are up. 410 */ 411 PTD[KPTDI] = (pd_entry_t) ptditmp; 412 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 413 invltlb(); 414#endif 415 } 416#endif 417 418#ifdef SMP 419 if (cpu_apic_address == 0) 420 panic("pmap_bootstrap: no local apic! (non-SMP hardware?)"); 421 422 /* local apic is mapped on last page */ 423 SMPpt[NPTEPG - 1] = (pt_entry_t)(PG_V | PG_RW | PG_N | pgeflag | 424 (cpu_apic_address & PG_FRAME)); 425#endif 426 427 invltlb(); 428} 429 430#ifdef SMP 431/* 432 * Set 4mb pdir for mp startup 433 */ 434void 435pmap_set_opt(void) 436{ 437 if (pseflag && (cpu_feature & CPUID_PSE)) { 438 load_cr4(rcr4() | CR4_PSE); 439 if (pdir4mb && PCPU_GET(cpuid) == 0) { /* only on BSP */ 440 kernel_pmap->pm_pdir[KPTDI] = 441 PTD[KPTDI] = (pd_entry_t)pdir4mb; 442 cpu_invltlb(); 443 } 444 } 445} 446#endif 447 448/* 449 * Initialize the pmap module. 450 * Called by vm_init, to initialize any structures that the pmap 451 * system needs to map virtual memory. 452 * pmap_init has been enhanced to support in a fairly consistant 453 * way, discontiguous physical memory. 454 */ 455void 456pmap_init(phys_start, phys_end) 457 vm_offset_t phys_start, phys_end; 458{ 459 int i; 460 int initial_pvs; 461 462 /* 463 * object for kernel page table pages 464 */ 465 kptobj = vm_object_allocate(OBJT_DEFAULT, NKPDE); 466 467 /* 468 * Allocate memory for random pmap data structures. Includes the 469 * pv_head_table. 470 */ 471 472 for(i = 0; i < vm_page_array_size; i++) { 473 vm_page_t m; 474 475 m = &vm_page_array[i]; 476 TAILQ_INIT(&m->md.pv_list); 477 m->md.pv_list_count = 0; 478 } 479 480 /* 481 * init the pv free list 482 */ 483 initial_pvs = vm_page_array_size; 484 if (initial_pvs < MINPV) 485 initial_pvs = MINPV; 486 pvzone = &pvzone_store; 487 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 488 initial_pvs * sizeof (struct pv_entry)); 489 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, 490 vm_page_array_size); 491 492 /* 493 * Now it is safe to enable pv_table recording. 494 */ 495 pmap_initialized = TRUE; 496} 497 498/* 499 * Initialize the address space (zone) for the pv_entries. Set a 500 * high water mark so that the system can recover from excessive 501 * numbers of pv entries. 502 */ 503void 504pmap_init2() 505{ 506 pv_entry_max = PMAP_SHPGPERPROC * maxproc + vm_page_array_size; 507 pv_entry_high_water = 9 * (pv_entry_max / 10); 508 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 509} 510 511 512/*************************************************** 513 * Low level helper routines..... 514 ***************************************************/ 515 516#if defined(PMAP_DIAGNOSTIC) 517 518/* 519 * This code checks for non-writeable/modified pages. 520 * This should be an invalid condition. 521 */ 522static int 523pmap_nw_modified(pt_entry_t ptea) 524{ 525 int pte; 526 527 pte = (int) ptea; 528 529 if ((pte & (PG_M|PG_RW)) == PG_M) 530 return 1; 531 else 532 return 0; 533} 534#endif 535 536 537/* 538 * this routine defines the region(s) of memory that should 539 * not be tested for the modified bit. 540 */ 541static PMAP_INLINE int 542pmap_track_modified(vm_offset_t va) 543{ 544 if ((va < clean_sva) || (va >= clean_eva)) 545 return 1; 546 else 547 return 0; 548} 549 550static PMAP_INLINE void 551invltlb_1pg(vm_offset_t va) 552{ 553#ifdef I386_CPU 554 invltlb(); 555#else 556 invlpg(va); 557#endif 558} 559 560static __inline void 561pmap_TLB_invalidate(pmap_t pmap, vm_offset_t va) 562{ 563#if defined(SMP) 564 if (pmap->pm_active & (1 << PCPU_GET(cpuid))) 565 cpu_invlpg((void *)va); 566 if (pmap->pm_active & PCPU_GET(other_cpus)) 567 smp_invltlb(); 568#else 569 if (pmap->pm_active) 570 invltlb_1pg(va); 571#endif 572} 573 574static __inline void 575pmap_TLB_invalidate_all(pmap_t pmap) 576{ 577#if defined(SMP) 578 if (pmap->pm_active & (1 << PCPU_GET(cpuid))) 579 cpu_invltlb(); 580 if (pmap->pm_active & PCPU_GET(other_cpus)) 581 smp_invltlb(); 582#else 583 if (pmap->pm_active) 584 invltlb(); 585#endif 586} 587 588static unsigned * 589get_ptbase(pmap) 590 pmap_t pmap; 591{ 592 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 593 594 /* are we current address space or kernel? */ 595 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 596 return (unsigned *) PTmap; 597 } 598 /* otherwise, we are alternate address space */ 599 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 600 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 601#if defined(SMP) 602 /* The page directory is not shared between CPUs */ 603 cpu_invltlb(); 604#else 605 invltlb(); 606#endif 607 } 608 return (unsigned *) APTmap; 609} 610 611/* 612 * Super fast pmap_pte routine best used when scanning 613 * the pv lists. This eliminates many coarse-grained 614 * invltlb calls. Note that many of the pv list 615 * scans are across different pmaps. It is very wasteful 616 * to do an entire invltlb for checking a single mapping. 617 */ 618 619static unsigned * 620pmap_pte_quick(pmap, va) 621 register pmap_t pmap; 622 vm_offset_t va; 623{ 624 unsigned pde, newpf; 625 if ((pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) != 0) { 626 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 627 unsigned index = i386_btop(va); 628 /* are we current address space or kernel? */ 629 if ((pmap == kernel_pmap) || 630 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 631 return (unsigned *) PTmap + index; 632 } 633 newpf = pde & PG_FRAME; 634 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 635 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 636 invltlb_1pg((vm_offset_t) PADDR1); 637 } 638 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 639 } 640 return (0); 641} 642 643/* 644 * Routine: pmap_extract 645 * Function: 646 * Extract the physical page address associated 647 * with the given map/virtual_address pair. 648 */ 649vm_offset_t 650pmap_extract(pmap, va) 651 register pmap_t pmap; 652 vm_offset_t va; 653{ 654 vm_offset_t rtval; 655 vm_offset_t pdirindex; 656 pdirindex = va >> PDRSHIFT; 657 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 658 unsigned *pte; 659 if ((rtval & PG_PS) != 0) { 660 rtval &= ~(NBPDR - 1); 661 rtval |= va & (NBPDR - 1); 662 return rtval; 663 } 664 pte = get_ptbase(pmap) + i386_btop(va); 665 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 666 return rtval; 667 } 668 return 0; 669 670} 671 672/*************************************************** 673 * Low level mapping routines..... 674 ***************************************************/ 675 676/* 677 * add a wired page to the kva 678 * note that in order for the mapping to take effect -- you 679 * should do a invltlb after doing the pmap_kenter... 680 */ 681PMAP_INLINE void 682pmap_kenter(va, pa) 683 vm_offset_t va; 684 register vm_offset_t pa; 685{ 686 register unsigned *pte; 687 unsigned npte, opte; 688 689 npte = pa | PG_RW | PG_V | pgeflag; 690 pte = (unsigned *)vtopte(va); 691 opte = *pte; 692 *pte = npte; 693 /*if (opte)*/ 694 invltlb_1pg(va); /* XXX what about SMP? */ 695} 696 697/* 698 * remove a page from the kernel pagetables 699 */ 700PMAP_INLINE void 701pmap_kremove(va) 702 vm_offset_t va; 703{ 704 register unsigned *pte; 705 706 pte = (unsigned *)vtopte(va); 707 *pte = 0; 708 invltlb_1pg(va); /* XXX what about SMP? */ 709} 710 711/* 712 * Used to map a range of physical addresses into kernel 713 * virtual address space. 714 * 715 * The value passed in '*virt' is a suggested virtual address for 716 * the mapping. Architectures which can support a direct-mapped 717 * physical to virtual region can return the appropriate address 718 * within that region, leaving '*virt' unchanged. Other 719 * architectures should map the pages starting at '*virt' and 720 * update '*virt' with the first usable address after the mapped 721 * region. 722 */ 723vm_offset_t 724pmap_map(virt, start, end, prot) 725 vm_offset_t *virt; 726 vm_offset_t start; 727 vm_offset_t end; 728 int prot; 729{ 730 vm_offset_t sva = *virt; 731 vm_offset_t va = sva; 732 while (start < end) { 733 pmap_kenter(va, start); 734 va += PAGE_SIZE; 735 start += PAGE_SIZE; 736 } 737 *virt = va; 738 return (sva); 739} 740 741 742/* 743 * Add a list of wired pages to the kva 744 * this routine is only used for temporary 745 * kernel mappings that do not need to have 746 * page modification or references recorded. 747 * Note that old mappings are simply written 748 * over. The page *must* be wired. 749 */ 750void 751pmap_qenter(va, m, count) 752 vm_offset_t va; 753 vm_page_t *m; 754 int count; 755{ 756 int i; 757 758 for (i = 0; i < count; i++) { 759 vm_offset_t tva = va + i * PAGE_SIZE; 760 pmap_kenter(tva, VM_PAGE_TO_PHYS(m[i])); 761 } 762} 763 764/* 765 * this routine jerks page mappings from the 766 * kernel -- it is meant only for temporary mappings. 767 */ 768void 769pmap_qremove(va, count) 770 vm_offset_t va; 771 int count; 772{ 773 vm_offset_t end_va; 774 775 end_va = va + count*PAGE_SIZE; 776 777 while (va < end_va) { 778 unsigned *pte; 779 780 pte = (unsigned *)vtopte(va); 781 *pte = 0; 782#ifdef SMP 783 cpu_invlpg((void *)va); 784#else 785 invltlb_1pg(va); 786#endif 787 va += PAGE_SIZE; 788 } 789#ifdef SMP 790 smp_invltlb(); 791#endif 792} 793 794static vm_page_t 795pmap_page_lookup(object, pindex) 796 vm_object_t object; 797 vm_pindex_t pindex; 798{ 799 vm_page_t m; 800retry: 801 m = vm_page_lookup(object, pindex); 802 if (m && vm_page_sleep_busy(m, FALSE, "pplookp")) 803 goto retry; 804 return m; 805} 806 807/* 808 * Create the UPAGES for a new process. 809 * This routine directly affects the fork perf for a process. 810 */ 811void 812pmap_new_proc(p) 813 struct proc *p; 814{ 815#ifdef I386_CPU 816 int updateneeded; 817#endif 818 int i; 819 vm_object_t upobj; 820 vm_page_t m; 821 struct user *up; 822 unsigned *ptek, oldpte; 823 824 /* 825 * allocate object for the upages 826 */ 827 if ((upobj = p->p_upages_obj) == NULL) { 828 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 829 p->p_upages_obj = upobj; 830 } 831 832 /* get a kernel virtual address for the UPAGES for this proc */ 833 if ((up = p->p_addr) == NULL) { 834 up = (struct user *) kmem_alloc_nofault(kernel_map, 835 UPAGES * PAGE_SIZE); 836 if (up == NULL) 837 panic("pmap_new_proc: u_map allocation failed"); 838 p->p_addr = up; 839 } 840 841 ptek = (unsigned *) vtopte((vm_offset_t) up); 842 843#ifdef I386_CPU 844 updateneeded = 0; 845#endif 846 for(i=0;i<UPAGES;i++) { 847 /* 848 * Get a kernel stack page 849 */ 850 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 851 852 /* 853 * Wire the page 854 */ 855 m->wire_count++; 856 cnt.v_wire_count++; 857 858 oldpte = *(ptek + i); 859 /* 860 * Enter the page into the kernel address space. 861 */ 862 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 863 if (oldpte) { 864#ifdef I386_CPU 865 updateneeded = 1; 866#else 867 invlpg((vm_offset_t) up + i * PAGE_SIZE); 868#endif 869 } 870 871 vm_page_wakeup(m); 872 vm_page_flag_clear(m, PG_ZERO); 873 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 874 m->valid = VM_PAGE_BITS_ALL; 875 } 876#ifdef I386_CPU 877 if (updateneeded) 878 invltlb(); 879#endif 880} 881 882/* 883 * Dispose the UPAGES for a process that has exited. 884 * This routine directly impacts the exit perf of a process. 885 */ 886void 887pmap_dispose_proc(p) 888 struct proc *p; 889{ 890 int i; 891 vm_object_t upobj; 892 vm_page_t m; 893 unsigned *ptek, oldpte; 894 895 upobj = p->p_upages_obj; 896 897 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 898 for(i=0;i<UPAGES;i++) { 899 900 if ((m = vm_page_lookup(upobj, i)) == NULL) 901 panic("pmap_dispose_proc: upage already missing???"); 902 903 vm_page_busy(m); 904 905 oldpte = *(ptek + i); 906 *(ptek + i) = 0; 907#ifndef I386_CPU 908 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 909#endif 910 vm_page_unwire(m, 0); 911 vm_page_free(m); 912 } 913#ifdef I386_CPU 914 invltlb(); 915#endif 916} 917 918/* 919 * Allow the UPAGES for a process to be prejudicially paged out. 920 */ 921void 922pmap_swapout_proc(p) 923 struct proc *p; 924{ 925 int i; 926 vm_object_t upobj; 927 vm_page_t m; 928 929 upobj = p->p_upages_obj; 930 /* 931 * let the upages be paged 932 */ 933 for(i=0;i<UPAGES;i++) { 934 if ((m = vm_page_lookup(upobj, i)) == NULL) 935 panic("pmap_swapout_proc: upage already missing???"); 936 vm_page_dirty(m); 937 vm_page_unwire(m, 0); 938 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 939 } 940} 941 942/* 943 * Bring the UPAGES for a specified process back in. 944 */ 945void 946pmap_swapin_proc(p) 947 struct proc *p; 948{ 949 int i,rv; 950 vm_object_t upobj; 951 vm_page_t m; 952 953 upobj = p->p_upages_obj; 954 for(i=0;i<UPAGES;i++) { 955 956 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 957 958 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 959 VM_PAGE_TO_PHYS(m)); 960 961 if (m->valid != VM_PAGE_BITS_ALL) { 962 rv = vm_pager_get_pages(upobj, &m, 1, 0); 963 if (rv != VM_PAGER_OK) 964 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 965 m = vm_page_lookup(upobj, i); 966 m->valid = VM_PAGE_BITS_ALL; 967 } 968 969 vm_page_wire(m); 970 vm_page_wakeup(m); 971 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 972 } 973} 974 975/*************************************************** 976 * Page table page management routines..... 977 ***************************************************/ 978 979/* 980 * This routine unholds page table pages, and if the hold count 981 * drops to zero, then it decrements the wire count. 982 */ 983static int 984_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 985 986 while (vm_page_sleep_busy(m, FALSE, "pmuwpt")) 987 ; 988 989 if (m->hold_count == 0) { 990 vm_offset_t pteva; 991 /* 992 * unmap the page table page 993 */ 994 pmap->pm_pdir[m->pindex] = 0; 995 --pmap->pm_stats.resident_count; 996 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 997 (((unsigned) PTDpde) & PG_FRAME)) { 998 /* 999 * Do a invltlb to make the invalidated mapping 1000 * take effect immediately. 1001 */ 1002 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1003 pmap_TLB_invalidate(pmap, pteva); 1004 } 1005 1006 if (pmap->pm_ptphint == m) 1007 pmap->pm_ptphint = NULL; 1008 1009 /* 1010 * If the page is finally unwired, simply free it. 1011 */ 1012 --m->wire_count; 1013 if (m->wire_count == 0) { 1014 1015 vm_page_flash(m); 1016 vm_page_busy(m); 1017 vm_page_free_zero(m); 1018 --cnt.v_wire_count; 1019 } 1020 return 1; 1021 } 1022 return 0; 1023} 1024 1025static PMAP_INLINE int 1026pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) 1027{ 1028 vm_page_unhold(m); 1029 if (m->hold_count == 0) 1030 return _pmap_unwire_pte_hold(pmap, m); 1031 else 1032 return 0; 1033} 1034 1035/* 1036 * After removing a page table entry, this routine is used to 1037 * conditionally free the page, and manage the hold/wire counts. 1038 */ 1039static int 1040pmap_unuse_pt(pmap, va, mpte) 1041 pmap_t pmap; 1042 vm_offset_t va; 1043 vm_page_t mpte; 1044{ 1045 unsigned ptepindex; 1046 if (va >= UPT_MIN_ADDRESS) 1047 return 0; 1048 1049 if (mpte == NULL) { 1050 ptepindex = (va >> PDRSHIFT); 1051 if (pmap->pm_ptphint && 1052 (pmap->pm_ptphint->pindex == ptepindex)) { 1053 mpte = pmap->pm_ptphint; 1054 } else { 1055 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1056 pmap->pm_ptphint = mpte; 1057 } 1058 } 1059 1060 return pmap_unwire_pte_hold(pmap, mpte); 1061} 1062 1063void 1064pmap_pinit0(pmap) 1065 struct pmap *pmap; 1066{ 1067 pmap->pm_pdir = 1068 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1069 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1070 pmap->pm_count = 1; 1071 pmap->pm_active = 0; 1072 pmap->pm_ptphint = NULL; 1073 TAILQ_INIT(&pmap->pm_pvlist); 1074 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1075 LIST_INSERT_HEAD(&allpmaps, pmap, pm_list); 1076} 1077 1078/* 1079 * Initialize a preallocated and zeroed pmap structure, 1080 * such as one in a vmspace structure. 1081 */ 1082void 1083pmap_pinit(pmap) 1084 register struct pmap *pmap; 1085{ 1086 vm_page_t ptdpg; 1087 1088 /* 1089 * No need to allocate page table space yet but we do need a valid 1090 * page directory table. 1091 */ 1092 if (pmap->pm_pdir == NULL) 1093 pmap->pm_pdir = 1094 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1095 1096 /* 1097 * allocate object for the ptes 1098 */ 1099 if (pmap->pm_pteobj == NULL) 1100 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1101 1102 /* 1103 * allocate the page directory page 1104 */ 1105 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1106 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1107 1108 ptdpg->wire_count = 1; 1109 ++cnt.v_wire_count; 1110 1111 1112 vm_page_flag_clear(ptdpg, PG_MAPPED | PG_BUSY); /* not usually mapped*/ 1113 ptdpg->valid = VM_PAGE_BITS_ALL; 1114 1115 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1116 if ((ptdpg->flags & PG_ZERO) == 0) 1117 bzero(pmap->pm_pdir, PAGE_SIZE); 1118 1119 LIST_INSERT_HEAD(&allpmaps, pmap, pm_list); 1120 /* Wire in kernel global address entries. */ 1121 /* XXX copies current process, does not fill in MPPTDI */ 1122 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1123#ifdef SMP 1124 pmap->pm_pdir[MPPTDI] = PTD[MPPTDI]; 1125#endif 1126 1127 /* install self-referential address mapping entry */ 1128 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1129 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW | PG_A | PG_M; 1130 1131 pmap->pm_count = 1; 1132 pmap->pm_active = 0; 1133 pmap->pm_ptphint = NULL; 1134 TAILQ_INIT(&pmap->pm_pvlist); 1135 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1136} 1137 1138/* 1139 * Wire in kernel global address entries. To avoid a race condition 1140 * between pmap initialization and pmap_growkernel, this procedure 1141 * should be called after the vmspace is attached to the process 1142 * but before this pmap is activated. 1143 */ 1144void 1145pmap_pinit2(pmap) 1146 struct pmap *pmap; 1147{ 1148 /* XXX: Remove this stub when no longer called */ 1149} 1150 1151static int 1152pmap_release_free_page(pmap, p) 1153 struct pmap *pmap; 1154 vm_page_t p; 1155{ 1156 unsigned *pde = (unsigned *) pmap->pm_pdir; 1157 /* 1158 * This code optimizes the case of freeing non-busy 1159 * page-table pages. Those pages are zero now, and 1160 * might as well be placed directly into the zero queue. 1161 */ 1162 if (vm_page_sleep_busy(p, FALSE, "pmaprl")) 1163 return 0; 1164 1165 vm_page_busy(p); 1166 1167 /* 1168 * Remove the page table page from the processes address space. 1169 */ 1170 pde[p->pindex] = 0; 1171 pmap->pm_stats.resident_count--; 1172 1173 if (p->hold_count) { 1174 panic("pmap_release: freeing held page table page"); 1175 } 1176 /* 1177 * Page directory pages need to have the kernel 1178 * stuff cleared, so they can go into the zero queue also. 1179 */ 1180 if (p->pindex == PTDPTDI) { 1181 bzero(pde + KPTDI, nkpt * PTESIZE); 1182#ifdef SMP 1183 pde[MPPTDI] = 0; 1184#endif 1185 pde[APTDPTDI] = 0; 1186 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1187 } 1188 1189 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1190 pmap->pm_ptphint = NULL; 1191 1192 p->wire_count--; 1193 cnt.v_wire_count--; 1194 vm_page_free_zero(p); 1195 return 1; 1196} 1197 1198/* 1199 * this routine is called if the page table page is not 1200 * mapped correctly. 1201 */ 1202static vm_page_t 1203_pmap_allocpte(pmap, ptepindex) 1204 pmap_t pmap; 1205 unsigned ptepindex; 1206{ 1207 vm_offset_t pteva, ptepa; 1208 vm_page_t m; 1209 1210 /* 1211 * Find or fabricate a new pagetable page 1212 */ 1213 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1214 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1215 1216 KASSERT(m->queue == PQ_NONE, 1217 ("_pmap_allocpte: %p->queue != PQ_NONE", m)); 1218 1219 if (m->wire_count == 0) 1220 cnt.v_wire_count++; 1221 m->wire_count++; 1222 1223 /* 1224 * Increment the hold count for the page table page 1225 * (denoting a new mapping.) 1226 */ 1227 m->hold_count++; 1228 1229 /* 1230 * Map the pagetable page into the process address space, if 1231 * it isn't already there. 1232 */ 1233 1234 pmap->pm_stats.resident_count++; 1235 1236 ptepa = VM_PAGE_TO_PHYS(m); 1237 pmap->pm_pdir[ptepindex] = 1238 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A | PG_M); 1239 1240 /* 1241 * Set the page table hint 1242 */ 1243 pmap->pm_ptphint = m; 1244 1245 /* 1246 * Try to use the new mapping, but if we cannot, then 1247 * do it with the routine that maps the page explicitly. 1248 */ 1249 if ((m->flags & PG_ZERO) == 0) { 1250 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1251 (((unsigned) PTDpde) & PG_FRAME)) { 1252 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1253 bzero((caddr_t) pteva, PAGE_SIZE); 1254 } else { 1255 pmap_zero_page(ptepa); 1256 } 1257 } 1258 1259 m->valid = VM_PAGE_BITS_ALL; 1260 vm_page_flag_clear(m, PG_ZERO); 1261 vm_page_flag_set(m, PG_MAPPED); 1262 vm_page_wakeup(m); 1263 1264 return m; 1265} 1266 1267static vm_page_t 1268pmap_allocpte(pmap, va) 1269 pmap_t pmap; 1270 vm_offset_t va; 1271{ 1272 unsigned ptepindex; 1273 vm_offset_t ptepa; 1274 vm_page_t m; 1275 1276 /* 1277 * Calculate pagetable page index 1278 */ 1279 ptepindex = va >> PDRSHIFT; 1280 1281 /* 1282 * Get the page directory entry 1283 */ 1284 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1285 1286 /* 1287 * This supports switching from a 4MB page to a 1288 * normal 4K page. 1289 */ 1290 if (ptepa & PG_PS) { 1291 pmap->pm_pdir[ptepindex] = 0; 1292 ptepa = 0; 1293 invltlb(); 1294 } 1295 1296 /* 1297 * If the page table page is mapped, we just increment the 1298 * hold count, and activate it. 1299 */ 1300 if (ptepa) { 1301 /* 1302 * In order to get the page table page, try the 1303 * hint first. 1304 */ 1305 if (pmap->pm_ptphint && 1306 (pmap->pm_ptphint->pindex == ptepindex)) { 1307 m = pmap->pm_ptphint; 1308 } else { 1309 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1310 pmap->pm_ptphint = m; 1311 } 1312 m->hold_count++; 1313 return m; 1314 } 1315 /* 1316 * Here if the pte page isn't mapped, or if it has been deallocated. 1317 */ 1318 return _pmap_allocpte(pmap, ptepindex); 1319} 1320 1321 1322/*************************************************** 1323* Pmap allocation/deallocation routines. 1324 ***************************************************/ 1325 1326/* 1327 * Release any resources held by the given physical map. 1328 * Called when a pmap initialized by pmap_pinit is being released. 1329 * Should only be called if the map contains no valid mappings. 1330 */ 1331void 1332pmap_release(pmap) 1333 register struct pmap *pmap; 1334{ 1335 vm_page_t p,n,ptdpg; 1336 vm_object_t object = pmap->pm_pteobj; 1337 int curgeneration; 1338 1339#if defined(DIAGNOSTIC) 1340 if (object->ref_count != 1) 1341 panic("pmap_release: pteobj reference count != 1"); 1342#endif 1343 1344 ptdpg = NULL; 1345 LIST_REMOVE(pmap, pm_list); 1346retry: 1347 curgeneration = object->generation; 1348 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1349 n = TAILQ_NEXT(p, listq); 1350 if (p->pindex == PTDPTDI) { 1351 ptdpg = p; 1352 continue; 1353 } 1354 while (1) { 1355 if (!pmap_release_free_page(pmap, p) && 1356 (object->generation != curgeneration)) 1357 goto retry; 1358 } 1359 } 1360 1361 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1362 goto retry; 1363} 1364 1365/* 1366 * grow the number of kernel page table entries, if needed 1367 */ 1368void 1369pmap_growkernel(vm_offset_t addr) 1370{ 1371 struct pmap *pmap; 1372 int s; 1373 vm_offset_t ptppaddr; 1374 vm_page_t nkpg; 1375 pd_entry_t newpdir; 1376 1377 s = splhigh(); 1378 if (kernel_vm_end == 0) { 1379 kernel_vm_end = KERNBASE; 1380 nkpt = 0; 1381 while (pdir_pde(PTD, kernel_vm_end)) { 1382 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1383 nkpt++; 1384 } 1385 } 1386 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1387 while (kernel_vm_end < addr) { 1388 if (pdir_pde(PTD, kernel_vm_end)) { 1389 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1390 continue; 1391 } 1392 1393 /* 1394 * This index is bogus, but out of the way 1395 */ 1396 nkpg = vm_page_alloc(kptobj, nkpt, VM_ALLOC_SYSTEM); 1397 if (!nkpg) 1398 panic("pmap_growkernel: no memory to grow kernel"); 1399 1400 nkpt++; 1401 1402 vm_page_wire(nkpg); 1403 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1404 pmap_zero_page(ptppaddr); 1405 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW | PG_A | PG_M); 1406 pdir_pde(PTD, kernel_vm_end) = newpdir; 1407 1408 LIST_FOREACH(pmap, &allpmaps, pm_list) { 1409 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1410 } 1411 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1412 } 1413 splx(s); 1414} 1415 1416/* 1417 * Retire the given physical map from service. 1418 * Should only be called if the map contains 1419 * no valid mappings. 1420 */ 1421void 1422pmap_destroy(pmap) 1423 register pmap_t pmap; 1424{ 1425 int count; 1426 1427 if (pmap == NULL) 1428 return; 1429 1430 count = --pmap->pm_count; 1431 if (count == 0) { 1432 pmap_release(pmap); 1433 panic("destroying a pmap is not yet implemented"); 1434 } 1435} 1436 1437/* 1438 * Add a reference to the specified pmap. 1439 */ 1440void 1441pmap_reference(pmap) 1442 pmap_t pmap; 1443{ 1444 if (pmap != NULL) { 1445 pmap->pm_count++; 1446 } 1447} 1448 1449/*************************************************** 1450* page management routines. 1451 ***************************************************/ 1452 1453/* 1454 * free the pv_entry back to the free list 1455 */ 1456static PMAP_INLINE void 1457free_pv_entry(pv) 1458 pv_entry_t pv; 1459{ 1460 pv_entry_count--; 1461 zfree(pvzone, pv); 1462} 1463 1464/* 1465 * get a new pv_entry, allocating a block from the system 1466 * when needed. 1467 * the memory allocation is performed bypassing the malloc code 1468 * because of the possibility of allocations at interrupt time. 1469 */ 1470static pv_entry_t 1471get_pv_entry(void) 1472{ 1473 pv_entry_count++; 1474 if (pv_entry_high_water && 1475 (pv_entry_count > pv_entry_high_water) && 1476 (pmap_pagedaemon_waken == 0)) { 1477 pmap_pagedaemon_waken = 1; 1478 wakeup (&vm_pages_needed); 1479 } 1480 return zalloc(pvzone); 1481} 1482 1483/* 1484 * This routine is very drastic, but can save the system 1485 * in a pinch. 1486 */ 1487void 1488pmap_collect() 1489{ 1490 int i; 1491 vm_page_t m; 1492 static int warningdone=0; 1493 1494 if (pmap_pagedaemon_waken == 0) 1495 return; 1496 1497 if (warningdone < 5) { 1498 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1499 warningdone++; 1500 } 1501 1502 for(i = 0; i < vm_page_array_size; i++) { 1503 m = &vm_page_array[i]; 1504 if (m->wire_count || m->hold_count || m->busy || 1505 (m->flags & PG_BUSY)) 1506 continue; 1507 pmap_remove_all(m); 1508 } 1509 pmap_pagedaemon_waken = 0; 1510} 1511 1512 1513/* 1514 * If it is the first entry on the list, it is actually 1515 * in the header and we must copy the following entry up 1516 * to the header. Otherwise we must search the list for 1517 * the entry. In either case we free the now unused entry. 1518 */ 1519 1520static int 1521pmap_remove_entry(pmap, m, va) 1522 struct pmap *pmap; 1523 vm_page_t m; 1524 vm_offset_t va; 1525{ 1526 pv_entry_t pv; 1527 int rtval; 1528 int s; 1529 1530 s = splvm(); 1531 if (m->md.pv_list_count < pmap->pm_stats.resident_count) { 1532 for (pv = TAILQ_FIRST(&m->md.pv_list); 1533 pv; 1534 pv = TAILQ_NEXT(pv, pv_list)) { 1535 if (pmap == pv->pv_pmap && va == pv->pv_va) 1536 break; 1537 } 1538 } else { 1539 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1540 pv; 1541 pv = TAILQ_NEXT(pv, pv_plist)) { 1542 if (va == pv->pv_va) 1543 break; 1544 } 1545 } 1546 1547 rtval = 0; 1548 if (pv) { 1549 1550 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1551 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 1552 m->md.pv_list_count--; 1553 if (TAILQ_FIRST(&m->md.pv_list) == NULL) 1554 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 1555 1556 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1557 free_pv_entry(pv); 1558 } 1559 1560 splx(s); 1561 return rtval; 1562} 1563 1564/* 1565 * Create a pv entry for page at pa for 1566 * (pmap, va). 1567 */ 1568static void 1569pmap_insert_entry(pmap, va, mpte, m) 1570 pmap_t pmap; 1571 vm_offset_t va; 1572 vm_page_t mpte; 1573 vm_page_t m; 1574{ 1575 1576 int s; 1577 pv_entry_t pv; 1578 1579 s = splvm(); 1580 pv = get_pv_entry(); 1581 pv->pv_va = va; 1582 pv->pv_pmap = pmap; 1583 pv->pv_ptem = mpte; 1584 1585 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1586 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); 1587 m->md.pv_list_count++; 1588 1589 splx(s); 1590} 1591 1592/* 1593 * pmap_remove_pte: do the things to unmap a page in a process 1594 */ 1595static int 1596pmap_remove_pte(pmap, ptq, va) 1597 struct pmap *pmap; 1598 unsigned *ptq; 1599 vm_offset_t va; 1600{ 1601 unsigned oldpte; 1602 vm_page_t m; 1603 1604 oldpte = atomic_readandclear_int(ptq); 1605 if (oldpte & PG_W) 1606 pmap->pm_stats.wired_count -= 1; 1607 /* 1608 * Machines that don't support invlpg, also don't support 1609 * PG_G. 1610 */ 1611 if (oldpte & PG_G) 1612 invlpg(va); 1613 pmap->pm_stats.resident_count -= 1; 1614 if (oldpte & PG_MANAGED) { 1615 m = PHYS_TO_VM_PAGE(oldpte); 1616 if (oldpte & PG_M) { 1617#if defined(PMAP_DIAGNOSTIC) 1618 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1619 printf( 1620 "pmap_remove: modified page not writable: va: 0x%x, pte: 0x%x\n", 1621 va, oldpte); 1622 } 1623#endif 1624 if (pmap_track_modified(va)) 1625 vm_page_dirty(m); 1626 } 1627 if (oldpte & PG_A) 1628 vm_page_flag_set(m, PG_REFERENCED); 1629 return pmap_remove_entry(pmap, m, va); 1630 } else { 1631 return pmap_unuse_pt(pmap, va, NULL); 1632 } 1633 1634 return 0; 1635} 1636 1637/* 1638 * Remove a single page from a process address space 1639 */ 1640static void 1641pmap_remove_page(pmap, va) 1642 struct pmap *pmap; 1643 register vm_offset_t va; 1644{ 1645 register unsigned *ptq; 1646 1647 /* 1648 * if there is no pte for this address, just skip it!!! 1649 */ 1650 if (*pmap_pde(pmap, va) == 0) { 1651 return; 1652 } 1653 1654 /* 1655 * get a local va for mappings for this pmap. 1656 */ 1657 ptq = get_ptbase(pmap) + i386_btop(va); 1658 if (*ptq) { 1659 (void) pmap_remove_pte(pmap, ptq, va); 1660 pmap_TLB_invalidate(pmap, va); 1661 } 1662 return; 1663} 1664 1665/* 1666 * Remove the given range of addresses from the specified map. 1667 * 1668 * It is assumed that the start and end are properly 1669 * rounded to the page size. 1670 */ 1671void 1672pmap_remove(pmap, sva, eva) 1673 struct pmap *pmap; 1674 register vm_offset_t sva; 1675 register vm_offset_t eva; 1676{ 1677 register unsigned *ptbase; 1678 vm_offset_t pdnxt; 1679 vm_offset_t ptpaddr; 1680 vm_offset_t sindex, eindex; 1681 int anyvalid; 1682 1683 if (pmap == NULL) 1684 return; 1685 1686 if (pmap->pm_stats.resident_count == 0) 1687 return; 1688 1689 /* 1690 * special handling of removing one page. a very 1691 * common operation and easy to short circuit some 1692 * code. 1693 */ 1694 if (((sva + PAGE_SIZE) == eva) && 1695 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1696 pmap_remove_page(pmap, sva); 1697 return; 1698 } 1699 1700 anyvalid = 0; 1701 1702 /* 1703 * Get a local virtual address for the mappings that are being 1704 * worked with. 1705 */ 1706 ptbase = get_ptbase(pmap); 1707 1708 sindex = i386_btop(sva); 1709 eindex = i386_btop(eva); 1710 1711 for (; sindex < eindex; sindex = pdnxt) { 1712 unsigned pdirindex; 1713 1714 /* 1715 * Calculate index for next page table. 1716 */ 1717 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1718 if (pmap->pm_stats.resident_count == 0) 1719 break; 1720 1721 pdirindex = sindex / NPDEPG; 1722 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1723 pmap->pm_pdir[pdirindex] = 0; 1724 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1725 anyvalid++; 1726 continue; 1727 } 1728 1729 /* 1730 * Weed out invalid mappings. Note: we assume that the page 1731 * directory table is always allocated, and in kernel virtual. 1732 */ 1733 if (ptpaddr == 0) 1734 continue; 1735 1736 /* 1737 * Limit our scan to either the end of the va represented 1738 * by the current page table page, or to the end of the 1739 * range being removed. 1740 */ 1741 if (pdnxt > eindex) { 1742 pdnxt = eindex; 1743 } 1744 1745 for ( ;sindex != pdnxt; sindex++) { 1746 vm_offset_t va; 1747 if (ptbase[sindex] == 0) { 1748 continue; 1749 } 1750 va = i386_ptob(sindex); 1751 1752 anyvalid++; 1753 if (pmap_remove_pte(pmap, 1754 ptbase + sindex, va)) 1755 break; 1756 } 1757 } 1758 1759 if (anyvalid) 1760 pmap_TLB_invalidate_all(pmap); 1761} 1762 1763/* 1764 * Routine: pmap_remove_all 1765 * Function: 1766 * Removes this physical page from 1767 * all physical maps in which it resides. 1768 * Reflects back modify bits to the pager. 1769 * 1770 * Notes: 1771 * Original versions of this routine were very 1772 * inefficient because they iteratively called 1773 * pmap_remove (slow...) 1774 */ 1775 1776static void 1777pmap_remove_all(m) 1778 vm_page_t m; 1779{ 1780 register pv_entry_t pv; 1781 register unsigned *pte, tpte; 1782 int s; 1783 1784#if defined(PMAP_DIAGNOSTIC) 1785 /* 1786 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1787 * pages! 1788 */ 1789 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) { 1790 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%x", VM_PAGE_TO_PHYS(m)); 1791 } 1792#endif 1793 1794 s = splvm(); 1795 while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { 1796 pv->pv_pmap->pm_stats.resident_count--; 1797 1798 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1799 1800 tpte = atomic_readandclear_int(pte); 1801 if (tpte & PG_W) 1802 pv->pv_pmap->pm_stats.wired_count--; 1803 1804 if (tpte & PG_A) 1805 vm_page_flag_set(m, PG_REFERENCED); 1806 1807 /* 1808 * Update the vm_page_t clean and reference bits. 1809 */ 1810 if (tpte & PG_M) { 1811#if defined(PMAP_DIAGNOSTIC) 1812 if (pmap_nw_modified((pt_entry_t) tpte)) { 1813 printf( 1814 "pmap_remove_all: modified page not writable: va: 0x%x, pte: 0x%x\n", 1815 pv->pv_va, tpte); 1816 } 1817#endif 1818 if (pmap_track_modified(pv->pv_va)) 1819 vm_page_dirty(m); 1820 } 1821 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 1822 1823 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1824 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 1825 m->md.pv_list_count--; 1826 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1827 free_pv_entry(pv); 1828 } 1829 1830 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 1831 1832 splx(s); 1833} 1834 1835/* 1836 * Set the physical protection on the 1837 * specified range of this map as requested. 1838 */ 1839void 1840pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1841{ 1842 register unsigned *ptbase; 1843 vm_offset_t pdnxt, ptpaddr; 1844 vm_pindex_t sindex, eindex; 1845 int anychanged; 1846 1847 if (pmap == NULL) 1848 return; 1849 1850 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1851 pmap_remove(pmap, sva, eva); 1852 return; 1853 } 1854 1855 if (prot & VM_PROT_WRITE) 1856 return; 1857 1858 anychanged = 0; 1859 1860 ptbase = get_ptbase(pmap); 1861 1862 sindex = i386_btop(sva); 1863 eindex = i386_btop(eva); 1864 1865 for (; sindex < eindex; sindex = pdnxt) { 1866 1867 unsigned pdirindex; 1868 1869 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1870 1871 pdirindex = sindex / NPDEPG; 1872 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1873 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1874 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1875 anychanged++; 1876 continue; 1877 } 1878 1879 /* 1880 * Weed out invalid mappings. Note: we assume that the page 1881 * directory table is always allocated, and in kernel virtual. 1882 */ 1883 if (ptpaddr == 0) 1884 continue; 1885 1886 if (pdnxt > eindex) { 1887 pdnxt = eindex; 1888 } 1889 1890 for (; sindex != pdnxt; sindex++) { 1891 1892 unsigned pbits; 1893 vm_page_t m; 1894 1895 pbits = ptbase[sindex]; 1896 1897 if (pbits & PG_MANAGED) { 1898 m = NULL; 1899 if (pbits & PG_A) { 1900 m = PHYS_TO_VM_PAGE(pbits); 1901 vm_page_flag_set(m, PG_REFERENCED); 1902 pbits &= ~PG_A; 1903 } 1904 if (pbits & PG_M) { 1905 if (pmap_track_modified(i386_ptob(sindex))) { 1906 if (m == NULL) 1907 m = PHYS_TO_VM_PAGE(pbits); 1908 vm_page_dirty(m); 1909 pbits &= ~PG_M; 1910 } 1911 } 1912 } 1913 1914 pbits &= ~PG_RW; 1915 1916 if (pbits != ptbase[sindex]) { 1917 ptbase[sindex] = pbits; 1918 anychanged = 1; 1919 } 1920 } 1921 } 1922 if (anychanged) 1923 pmap_TLB_invalidate_all(pmap); 1924} 1925 1926/* 1927 * Insert the given physical page (p) at 1928 * the specified virtual address (v) in the 1929 * target physical map with the protection requested. 1930 * 1931 * If specified, the page will be wired down, meaning 1932 * that the related pte can not be reclaimed. 1933 * 1934 * NB: This is the only routine which MAY NOT lazy-evaluate 1935 * or lose information. That is, this routine must actually 1936 * insert this page into the given map NOW. 1937 */ 1938void 1939pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, 1940 boolean_t wired) 1941{ 1942 vm_offset_t pa; 1943 register unsigned *pte; 1944 vm_offset_t opa; 1945 vm_offset_t origpte, newpte; 1946 vm_page_t mpte; 1947 1948 if (pmap == NULL) 1949 return; 1950 1951 va &= PG_FRAME; 1952#ifdef PMAP_DIAGNOSTIC 1953 if (va > VM_MAX_KERNEL_ADDRESS) 1954 panic("pmap_enter: toobig"); 1955 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 1956 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 1957#endif 1958 1959 mpte = NULL; 1960 /* 1961 * In the case that a page table page is not 1962 * resident, we are creating it here. 1963 */ 1964 if (va < UPT_MIN_ADDRESS) { 1965 mpte = pmap_allocpte(pmap, va); 1966 } 1967#if 0 && defined(PMAP_DIAGNOSTIC) 1968 else { 1969 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 1970 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 1971 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 1972 pmap->pm_pdir[PTDPTDI], origpte, va); 1973 } 1974 if (smp_active) { 1975 pdeaddr = (vm_offset_t *) IdlePTDS[PCPU_GET(cpuid)]; 1976 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 1977 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 1978 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 1979 printf("cpuid: %d, pdeaddr: 0x%x\n", PCPU_GET(cpuid), pdeaddr); 1980 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 1981 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 1982 } 1983 } 1984 } 1985#endif 1986 1987 pte = pmap_pte(pmap, va); 1988 1989 /* 1990 * Page Directory table entry not valid, we need a new PT page 1991 */ 1992 if (pte == NULL) { 1993 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%x\n", 1994 (void *)pmap->pm_pdir[PTDPTDI], va); 1995 } 1996 1997 pa = VM_PAGE_TO_PHYS(m) & PG_FRAME; 1998 origpte = *(vm_offset_t *)pte; 1999 opa = origpte & PG_FRAME; 2000 2001 if (origpte & PG_PS) 2002 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2003 2004 /* 2005 * Mapping has not changed, must be protection or wiring change. 2006 */ 2007 if (origpte && (opa == pa)) { 2008 /* 2009 * Wiring change, just update stats. We don't worry about 2010 * wiring PT pages as they remain resident as long as there 2011 * are valid mappings in them. Hence, if a user page is wired, 2012 * the PT page will be also. 2013 */ 2014 if (wired && ((origpte & PG_W) == 0)) 2015 pmap->pm_stats.wired_count++; 2016 else if (!wired && (origpte & PG_W)) 2017 pmap->pm_stats.wired_count--; 2018 2019#if defined(PMAP_DIAGNOSTIC) 2020 if (pmap_nw_modified((pt_entry_t) origpte)) { 2021 printf( 2022 "pmap_enter: modified page not writable: va: 0x%x, pte: 0x%x\n", 2023 va, origpte); 2024 } 2025#endif 2026 2027 /* 2028 * Remove extra pte reference 2029 */ 2030 if (mpte) 2031 mpte->hold_count--; 2032 2033 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2034 if ((origpte & PG_RW) == 0) { 2035 *pte |= PG_RW; 2036#ifdef SMP 2037 cpu_invlpg((void *)va); 2038 if (pmap->pm_active & PCPU_GET(other_cpus)) 2039 smp_invltlb(); 2040#else 2041 invltlb_1pg(va); 2042#endif 2043 } 2044 return; 2045 } 2046 2047 /* 2048 * We might be turning off write access to the page, 2049 * so we go ahead and sense modify status. 2050 */ 2051 if (origpte & PG_MANAGED) { 2052 if ((origpte & PG_M) && pmap_track_modified(va)) { 2053 vm_page_t om; 2054 om = PHYS_TO_VM_PAGE(opa); 2055 vm_page_dirty(om); 2056 } 2057 pa |= PG_MANAGED; 2058 } 2059 goto validate; 2060 } 2061 /* 2062 * Mapping has changed, invalidate old range and fall through to 2063 * handle validating new mapping. 2064 */ 2065 if (opa) { 2066 int err; 2067 err = pmap_remove_pte(pmap, pte, va); 2068 if (err) 2069 panic("pmap_enter: pte vanished, va: 0x%x", va); 2070 } 2071 2072 /* 2073 * Enter on the PV list if part of our managed memory. Note that we 2074 * raise IPL while manipulating pv_table since pmap_enter can be 2075 * called at interrupt time. 2076 */ 2077 if (pmap_initialized && 2078 (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) { 2079 pmap_insert_entry(pmap, va, mpte, m); 2080 pa |= PG_MANAGED; 2081 } 2082 2083 /* 2084 * Increment counters 2085 */ 2086 pmap->pm_stats.resident_count++; 2087 if (wired) 2088 pmap->pm_stats.wired_count++; 2089 2090validate: 2091 /* 2092 * Now validate mapping with desired protection/wiring. 2093 */ 2094 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2095 2096 if (wired) 2097 newpte |= PG_W; 2098 if (va < UPT_MIN_ADDRESS) 2099 newpte |= PG_U; 2100 if (pmap == kernel_pmap) 2101 newpte |= pgeflag; 2102 2103 /* 2104 * if the mapping or permission bits are different, we need 2105 * to update the pte. 2106 */ 2107 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2108 *pte = newpte | PG_A; 2109 /*if (origpte)*/ { 2110#ifdef SMP 2111 cpu_invlpg((void *)va); 2112 if (pmap->pm_active & PCPU_GET(other_cpus)) 2113 smp_invltlb(); 2114#else 2115 invltlb_1pg(va); 2116#endif 2117 } 2118 } 2119} 2120 2121/* 2122 * this code makes some *MAJOR* assumptions: 2123 * 1. Current pmap & pmap exists. 2124 * 2. Not wired. 2125 * 3. Read access. 2126 * 4. No page table pages. 2127 * 5. Tlbflush is deferred to calling procedure. 2128 * 6. Page IS managed. 2129 * but is *MUCH* faster than pmap_enter... 2130 */ 2131 2132static vm_page_t 2133pmap_enter_quick(pmap, va, m, mpte) 2134 register pmap_t pmap; 2135 vm_offset_t va; 2136 vm_page_t m; 2137 vm_page_t mpte; 2138{ 2139 unsigned *pte; 2140 vm_offset_t pa; 2141 2142 /* 2143 * In the case that a page table page is not 2144 * resident, we are creating it here. 2145 */ 2146 if (va < UPT_MIN_ADDRESS) { 2147 unsigned ptepindex; 2148 vm_offset_t ptepa; 2149 2150 /* 2151 * Calculate pagetable page index 2152 */ 2153 ptepindex = va >> PDRSHIFT; 2154 if (mpte && (mpte->pindex == ptepindex)) { 2155 mpte->hold_count++; 2156 } else { 2157retry: 2158 /* 2159 * Get the page directory entry 2160 */ 2161 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2162 2163 /* 2164 * If the page table page is mapped, we just increment 2165 * the hold count, and activate it. 2166 */ 2167 if (ptepa) { 2168 if (ptepa & PG_PS) 2169 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2170 if (pmap->pm_ptphint && 2171 (pmap->pm_ptphint->pindex == ptepindex)) { 2172 mpte = pmap->pm_ptphint; 2173 } else { 2174 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2175 pmap->pm_ptphint = mpte; 2176 } 2177 if (mpte == NULL) 2178 goto retry; 2179 mpte->hold_count++; 2180 } else { 2181 mpte = _pmap_allocpte(pmap, ptepindex); 2182 } 2183 } 2184 } else { 2185 mpte = NULL; 2186 } 2187 2188 /* 2189 * This call to vtopte makes the assumption that we are 2190 * entering the page into the current pmap. In order to support 2191 * quick entry into any pmap, one would likely use pmap_pte_quick. 2192 * But that isn't as quick as vtopte. 2193 */ 2194 pte = (unsigned *)vtopte(va); 2195 if (*pte) { 2196 if (mpte) 2197 pmap_unwire_pte_hold(pmap, mpte); 2198 return 0; 2199 } 2200 2201 /* 2202 * Enter on the PV list if part of our managed memory. Note that we 2203 * raise IPL while manipulating pv_table since pmap_enter can be 2204 * called at interrupt time. 2205 */ 2206 if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) 2207 pmap_insert_entry(pmap, va, mpte, m); 2208 2209 /* 2210 * Increment counters 2211 */ 2212 pmap->pm_stats.resident_count++; 2213 2214 pa = VM_PAGE_TO_PHYS(m); 2215 2216 /* 2217 * Now validate mapping with RO protection 2218 */ 2219 if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) 2220 *pte = pa | PG_V | PG_U; 2221 else 2222 *pte = pa | PG_V | PG_U | PG_MANAGED; 2223 2224 return mpte; 2225} 2226 2227/* 2228 * Make a temporary mapping for a physical address. This is only intended 2229 * to be used for panic dumps. 2230 */ 2231void * 2232pmap_kenter_temporary(vm_offset_t pa, int i) 2233{ 2234 pmap_kenter((vm_offset_t)crashdumpmap + (i * PAGE_SIZE), pa); 2235 return ((void *)crashdumpmap); 2236} 2237 2238#define MAX_INIT_PT (96) 2239/* 2240 * pmap_object_init_pt preloads the ptes for a given object 2241 * into the specified pmap. This eliminates the blast of soft 2242 * faults on process startup and immediately after an mmap. 2243 */ 2244void 2245pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2246 pmap_t pmap; 2247 vm_offset_t addr; 2248 vm_object_t object; 2249 vm_pindex_t pindex; 2250 vm_size_t size; 2251 int limit; 2252{ 2253 vm_offset_t tmpidx; 2254 int psize; 2255 vm_page_t p, mpte; 2256 int objpgs; 2257 2258 if (pmap == NULL || object == NULL) 2259 return; 2260 2261 /* 2262 * This code maps large physical mmap regions into the 2263 * processor address space. Note that some shortcuts 2264 * are taken, but the code works. 2265 */ 2266 if (pseflag && 2267 (object->type == OBJT_DEVICE) && 2268 ((addr & (NBPDR - 1)) == 0) && 2269 ((size & (NBPDR - 1)) == 0) ) { 2270 int i; 2271 vm_page_t m[1]; 2272 unsigned int ptepindex; 2273 int npdes; 2274 vm_offset_t ptepa; 2275 2276 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2277 return; 2278 2279retry: 2280 p = vm_page_lookup(object, pindex); 2281 if (p && vm_page_sleep_busy(p, FALSE, "init4p")) 2282 goto retry; 2283 2284 if (p == NULL) { 2285 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2286 if (p == NULL) 2287 return; 2288 m[0] = p; 2289 2290 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2291 vm_page_free(p); 2292 return; 2293 } 2294 2295 p = vm_page_lookup(object, pindex); 2296 vm_page_wakeup(p); 2297 } 2298 2299 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2300 if (ptepa & (NBPDR - 1)) { 2301 return; 2302 } 2303 2304 p->valid = VM_PAGE_BITS_ALL; 2305 2306 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2307 npdes = size >> PDRSHIFT; 2308 for(i=0;i<npdes;i++) { 2309 pmap->pm_pdir[ptepindex] = 2310 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2311 ptepa += NBPDR; 2312 ptepindex += 1; 2313 } 2314 vm_page_flag_set(p, PG_MAPPED); 2315 invltlb(); 2316 return; 2317 } 2318 2319 psize = i386_btop(size); 2320 2321 if ((object->type != OBJT_VNODE) || 2322 (limit && (psize > MAX_INIT_PT) && 2323 (object->resident_page_count > MAX_INIT_PT))) { 2324 return; 2325 } 2326 2327 if (psize + pindex > object->size) { 2328 if (object->size < pindex) 2329 return; 2330 psize = object->size - pindex; 2331 } 2332 2333 mpte = NULL; 2334 /* 2335 * if we are processing a major portion of the object, then scan the 2336 * entire thing. 2337 */ 2338 if (psize > (object->resident_page_count >> 2)) { 2339 objpgs = psize; 2340 2341 for (p = TAILQ_FIRST(&object->memq); 2342 ((objpgs > 0) && (p != NULL)); 2343 p = TAILQ_NEXT(p, listq)) { 2344 2345 tmpidx = p->pindex; 2346 if (tmpidx < pindex) { 2347 continue; 2348 } 2349 tmpidx -= pindex; 2350 if (tmpidx >= psize) { 2351 continue; 2352 } 2353 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2354 (p->busy == 0) && 2355 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2356 if ((p->queue - p->pc) == PQ_CACHE) 2357 vm_page_deactivate(p); 2358 vm_page_busy(p); 2359 mpte = pmap_enter_quick(pmap, 2360 addr + i386_ptob(tmpidx), p, mpte); 2361 vm_page_flag_set(p, PG_MAPPED); 2362 vm_page_wakeup(p); 2363 } 2364 objpgs -= 1; 2365 } 2366 } else { 2367 /* 2368 * else lookup the pages one-by-one. 2369 */ 2370 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2371 p = vm_page_lookup(object, tmpidx + pindex); 2372 if (p && 2373 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2374 (p->busy == 0) && 2375 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2376 if ((p->queue - p->pc) == PQ_CACHE) 2377 vm_page_deactivate(p); 2378 vm_page_busy(p); 2379 mpte = pmap_enter_quick(pmap, 2380 addr + i386_ptob(tmpidx), p, mpte); 2381 vm_page_flag_set(p, PG_MAPPED); 2382 vm_page_wakeup(p); 2383 } 2384 } 2385 } 2386 return; 2387} 2388 2389/* 2390 * pmap_prefault provides a quick way of clustering 2391 * pagefaults into a processes address space. It is a "cousin" 2392 * of pmap_object_init_pt, except it runs at page fault time instead 2393 * of mmap time. 2394 */ 2395#define PFBAK 4 2396#define PFFOR 4 2397#define PAGEORDER_SIZE (PFBAK+PFFOR) 2398 2399static int pmap_prefault_pageorder[] = { 2400 -PAGE_SIZE, PAGE_SIZE, 2401 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2402 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2403 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2404}; 2405 2406void 2407pmap_prefault(pmap, addra, entry) 2408 pmap_t pmap; 2409 vm_offset_t addra; 2410 vm_map_entry_t entry; 2411{ 2412 int i; 2413 vm_offset_t starta; 2414 vm_offset_t addr; 2415 vm_pindex_t pindex; 2416 vm_page_t m, mpte; 2417 vm_object_t object; 2418 2419 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) 2420 return; 2421 2422 object = entry->object.vm_object; 2423 2424 starta = addra - PFBAK * PAGE_SIZE; 2425 if (starta < entry->start) { 2426 starta = entry->start; 2427 } else if (starta > addra) { 2428 starta = 0; 2429 } 2430 2431 mpte = NULL; 2432 for (i = 0; i < PAGEORDER_SIZE; i++) { 2433 vm_object_t lobject; 2434 unsigned *pte; 2435 2436 addr = addra + pmap_prefault_pageorder[i]; 2437 if (addr > addra + (PFFOR * PAGE_SIZE)) 2438 addr = 0; 2439 2440 if (addr < starta || addr >= entry->end) 2441 continue; 2442 2443 if ((*pmap_pde(pmap, addr)) == NULL) 2444 continue; 2445 2446 pte = (unsigned *) vtopte(addr); 2447 if (*pte) 2448 continue; 2449 2450 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2451 lobject = object; 2452 for (m = vm_page_lookup(lobject, pindex); 2453 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2454 lobject = lobject->backing_object) { 2455 if (lobject->backing_object_offset & PAGE_MASK) 2456 break; 2457 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2458 m = vm_page_lookup(lobject->backing_object, pindex); 2459 } 2460 2461 /* 2462 * give-up when a page is not in memory 2463 */ 2464 if (m == NULL) 2465 break; 2466 2467 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2468 (m->busy == 0) && 2469 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2470 2471 if ((m->queue - m->pc) == PQ_CACHE) { 2472 vm_page_deactivate(m); 2473 } 2474 vm_page_busy(m); 2475 mpte = pmap_enter_quick(pmap, addr, m, mpte); 2476 vm_page_flag_set(m, PG_MAPPED); 2477 vm_page_wakeup(m); 2478 } 2479 } 2480} 2481 2482/* 2483 * Routine: pmap_change_wiring 2484 * Function: Change the wiring attribute for a map/virtual-address 2485 * pair. 2486 * In/out conditions: 2487 * The mapping must already exist in the pmap. 2488 */ 2489void 2490pmap_change_wiring(pmap, va, wired) 2491 register pmap_t pmap; 2492 vm_offset_t va; 2493 boolean_t wired; 2494{ 2495 register unsigned *pte; 2496 2497 if (pmap == NULL) 2498 return; 2499 2500 pte = pmap_pte(pmap, va); 2501 2502 if (wired && !pmap_pte_w(pte)) 2503 pmap->pm_stats.wired_count++; 2504 else if (!wired && pmap_pte_w(pte)) 2505 pmap->pm_stats.wired_count--; 2506 2507 /* 2508 * Wiring is not a hardware characteristic so there is no need to 2509 * invalidate TLB. 2510 */ 2511 pmap_pte_set_w(pte, wired); 2512} 2513 2514 2515 2516/* 2517 * Copy the range specified by src_addr/len 2518 * from the source map to the range dst_addr/len 2519 * in the destination map. 2520 * 2521 * This routine is only advisory and need not do anything. 2522 */ 2523 2524void 2525pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2526 pmap_t dst_pmap, src_pmap; 2527 vm_offset_t dst_addr; 2528 vm_size_t len; 2529 vm_offset_t src_addr; 2530{ 2531 vm_offset_t addr; 2532 vm_offset_t end_addr = src_addr + len; 2533 vm_offset_t pdnxt; 2534 unsigned src_frame, dst_frame; 2535 vm_page_t m; 2536 2537 if (dst_addr != src_addr) 2538 return; 2539 2540 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2541 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2542 return; 2543 } 2544 2545 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2546 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2547 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2548#if defined(SMP) 2549 /* The page directory is not shared between CPUs */ 2550 cpu_invltlb(); 2551#else 2552 invltlb(); 2553#endif 2554 } 2555 2556 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2557 unsigned *src_pte, *dst_pte; 2558 vm_page_t dstmpte, srcmpte; 2559 vm_offset_t srcptepaddr; 2560 unsigned ptepindex; 2561 2562 if (addr >= UPT_MIN_ADDRESS) 2563 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2564 2565 /* 2566 * Don't let optional prefaulting of pages make us go 2567 * way below the low water mark of free pages or way 2568 * above high water mark of used pv entries. 2569 */ 2570 if (cnt.v_free_count < cnt.v_free_reserved || 2571 pv_entry_count > pv_entry_high_water) 2572 break; 2573 2574 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2575 ptepindex = addr >> PDRSHIFT; 2576 2577 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2578 if (srcptepaddr == 0) 2579 continue; 2580 2581 if (srcptepaddr & PG_PS) { 2582 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2583 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2584 dst_pmap->pm_stats.resident_count += NBPDR / PAGE_SIZE; 2585 } 2586 continue; 2587 } 2588 2589 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2590 if ((srcmpte == NULL) || 2591 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2592 continue; 2593 2594 if (pdnxt > end_addr) 2595 pdnxt = end_addr; 2596 2597 src_pte = (unsigned *) vtopte(addr); 2598 dst_pte = (unsigned *) avtopte(addr); 2599 while (addr < pdnxt) { 2600 unsigned ptetemp; 2601 ptetemp = *src_pte; 2602 /* 2603 * we only virtual copy managed pages 2604 */ 2605 if ((ptetemp & PG_MANAGED) != 0) { 2606 /* 2607 * We have to check after allocpte for the 2608 * pte still being around... allocpte can 2609 * block. 2610 */ 2611 dstmpte = pmap_allocpte(dst_pmap, addr); 2612 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2613 /* 2614 * Clear the modified and 2615 * accessed (referenced) bits 2616 * during the copy. 2617 */ 2618 m = PHYS_TO_VM_PAGE(ptetemp); 2619 *dst_pte = ptetemp & ~(PG_M | PG_A); 2620 dst_pmap->pm_stats.resident_count++; 2621 pmap_insert_entry(dst_pmap, addr, 2622 dstmpte, m); 2623 } else { 2624 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2625 } 2626 if (dstmpte->hold_count >= srcmpte->hold_count) 2627 break; 2628 } 2629 addr += PAGE_SIZE; 2630 src_pte++; 2631 dst_pte++; 2632 } 2633 } 2634} 2635 2636/* 2637 * Routine: pmap_kernel 2638 * Function: 2639 * Returns the physical map handle for the kernel. 2640 */ 2641pmap_t 2642pmap_kernel() 2643{ 2644 return (kernel_pmap); 2645} 2646 2647/* 2648 * pmap_zero_page zeros the specified hardware page by mapping 2649 * the page into KVM and using bzero to clear its contents. 2650 */ 2651void 2652pmap_zero_page(phys) 2653 vm_offset_t phys; 2654{ 2655 2656 if (*(int *) CMAP2) 2657 panic("pmap_zero_page: CMAP2 busy"); 2658 2659 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2660 invltlb_1pg((vm_offset_t)CADDR2); 2661 2662#if defined(I686_CPU) 2663 if (cpu_class == CPUCLASS_686) 2664 i686_pagezero(CADDR2); 2665 else 2666#endif 2667 bzero(CADDR2, PAGE_SIZE); 2668 *(int *) CMAP2 = 0; 2669} 2670 2671/* 2672 * pmap_zero_page_area zeros the specified hardware page by mapping 2673 * the page into KVM and using bzero to clear its contents. 2674 * 2675 * off and size may not cover an area beyond a single hardware page. 2676 */ 2677void 2678pmap_zero_page_area(phys, off, size) 2679 vm_offset_t phys; 2680 int off; 2681 int size; 2682{ 2683 2684 if (*(int *) CMAP2) 2685 panic("pmap_zero_page: CMAP2 busy"); 2686 2687 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2688 invltlb_1pg((vm_offset_t)CADDR2); 2689 2690#if defined(I686_CPU) 2691 if (cpu_class == CPUCLASS_686 && off == 0 && size == PAGE_SIZE) 2692 i686_pagezero(CADDR2); 2693 else 2694#endif 2695 bzero((char *)CADDR2 + off, size); 2696 *(int *) CMAP2 = 0; 2697} 2698 2699/* 2700 * pmap_copy_page copies the specified (machine independent) 2701 * page by mapping the page into virtual memory and using 2702 * bcopy to copy the page, one machine dependent page at a 2703 * time. 2704 */ 2705void 2706pmap_copy_page(src, dst) 2707 vm_offset_t src; 2708 vm_offset_t dst; 2709{ 2710 2711 if (*(int *) CMAP1) 2712 panic("pmap_copy_page: CMAP1 busy"); 2713 if (*(int *) CMAP2) 2714 panic("pmap_copy_page: CMAP2 busy"); 2715 2716 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2717 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2718#ifdef I386_CPU 2719 invltlb(); 2720#else 2721 invlpg((u_int)CADDR1); 2722 invlpg((u_int)CADDR2); 2723#endif 2724 2725 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2726 2727 *(int *) CMAP1 = 0; 2728 *(int *) CMAP2 = 0; 2729} 2730 2731 2732/* 2733 * Routine: pmap_pageable 2734 * Function: 2735 * Make the specified pages (by pmap, offset) 2736 * pageable (or not) as requested. 2737 * 2738 * A page which is not pageable may not take 2739 * a fault; therefore, its page table entry 2740 * must remain valid for the duration. 2741 * 2742 * This routine is merely advisory; pmap_enter 2743 * will specify that these pages are to be wired 2744 * down (or not) as appropriate. 2745 */ 2746void 2747pmap_pageable(pmap, sva, eva, pageable) 2748 pmap_t pmap; 2749 vm_offset_t sva, eva; 2750 boolean_t pageable; 2751{ 2752} 2753 2754/* 2755 * this routine returns true if a physical page resides 2756 * in the given pmap. 2757 */ 2758boolean_t 2759pmap_page_exists(pmap, m) 2760 pmap_t pmap; 2761 vm_page_t m; 2762{ 2763 register pv_entry_t pv; 2764 int s; 2765 2766 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 2767 return FALSE; 2768 2769 s = splvm(); 2770 2771 /* 2772 * Not found, check current mappings returning immediately if found. 2773 */ 2774 for (pv = TAILQ_FIRST(&m->md.pv_list); 2775 pv; 2776 pv = TAILQ_NEXT(pv, pv_list)) { 2777 if (pv->pv_pmap == pmap) { 2778 splx(s); 2779 return TRUE; 2780 } 2781 } 2782 splx(s); 2783 return (FALSE); 2784} 2785 2786#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2787/* 2788 * Remove all pages from specified address space 2789 * this aids process exit speeds. Also, this code 2790 * is special cased for current process only, but 2791 * can have the more generic (and slightly slower) 2792 * mode enabled. This is much faster than pmap_remove 2793 * in the case of running down an entire address space. 2794 */ 2795void 2796pmap_remove_pages(pmap, sva, eva) 2797 pmap_t pmap; 2798 vm_offset_t sva, eva; 2799{ 2800 unsigned *pte, tpte; 2801 pv_entry_t pv, npv; 2802 int s; 2803 vm_page_t m; 2804 2805#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2806 if (!curproc || (pmap != vmspace_pmap(curproc->p_vmspace))) { 2807 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2808 return; 2809 } 2810#endif 2811 2812 s = splvm(); 2813 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2814 pv; 2815 pv = npv) { 2816 2817 if (pv->pv_va >= eva || pv->pv_va < sva) { 2818 npv = TAILQ_NEXT(pv, pv_plist); 2819 continue; 2820 } 2821 2822#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2823 pte = (unsigned *)vtopte(pv->pv_va); 2824#else 2825 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2826#endif 2827 tpte = *pte; 2828 2829/* 2830 * We cannot remove wired pages from a process' mapping at this time 2831 */ 2832 if (tpte & PG_W) { 2833 npv = TAILQ_NEXT(pv, pv_plist); 2834 continue; 2835 } 2836 *pte = 0; 2837 2838 m = PHYS_TO_VM_PAGE(tpte); 2839 2840 KASSERT(m < &vm_page_array[vm_page_array_size], 2841 ("pmap_remove_pages: bad tpte %x", tpte)); 2842 2843 pv->pv_pmap->pm_stats.resident_count--; 2844 2845 /* 2846 * Update the vm_page_t clean and reference bits. 2847 */ 2848 if (tpte & PG_M) { 2849 vm_page_dirty(m); 2850 } 2851 2852 2853 npv = TAILQ_NEXT(pv, pv_plist); 2854 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2855 2856 m->md.pv_list_count--; 2857 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 2858 if (TAILQ_FIRST(&m->md.pv_list) == NULL) { 2859 vm_page_flag_clear(m, PG_MAPPED | PG_WRITEABLE); 2860 } 2861 2862 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2863 free_pv_entry(pv); 2864 } 2865 splx(s); 2866 pmap_TLB_invalidate_all(pmap); 2867} 2868 2869/* 2870 * pmap_testbit tests bits in pte's 2871 * note that the testbit/changebit routines are inline, 2872 * and a lot of things compile-time evaluate. 2873 */ 2874static boolean_t 2875pmap_testbit(m, bit) 2876 vm_page_t m; 2877 int bit; 2878{ 2879 pv_entry_t pv; 2880 unsigned *pte; 2881 int s; 2882 2883 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 2884 return FALSE; 2885 2886 if (TAILQ_FIRST(&m->md.pv_list) == NULL) 2887 return FALSE; 2888 2889 s = splvm(); 2890 2891 for (pv = TAILQ_FIRST(&m->md.pv_list); 2892 pv; 2893 pv = TAILQ_NEXT(pv, pv_list)) { 2894 2895 /* 2896 * if the bit being tested is the modified bit, then 2897 * mark clean_map and ptes as never 2898 * modified. 2899 */ 2900 if (bit & (PG_A|PG_M)) { 2901 if (!pmap_track_modified(pv->pv_va)) 2902 continue; 2903 } 2904 2905#if defined(PMAP_DIAGNOSTIC) 2906 if (!pv->pv_pmap) { 2907 printf("Null pmap (tb) at va: 0x%x\n", pv->pv_va); 2908 continue; 2909 } 2910#endif 2911 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2912 if (*pte & bit) { 2913 splx(s); 2914 return TRUE; 2915 } 2916 } 2917 splx(s); 2918 return (FALSE); 2919} 2920 2921/* 2922 * this routine is used to modify bits in ptes 2923 */ 2924static __inline void 2925pmap_changebit(m, bit, setem) 2926 vm_page_t m; 2927 int bit; 2928 boolean_t setem; 2929{ 2930 register pv_entry_t pv; 2931 register unsigned *pte; 2932 int s; 2933 2934 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 2935 return; 2936 2937 s = splvm(); 2938 2939 /* 2940 * Loop over all current mappings setting/clearing as appropos If 2941 * setting RO do we need to clear the VAC? 2942 */ 2943 for (pv = TAILQ_FIRST(&m->md.pv_list); 2944 pv; 2945 pv = TAILQ_NEXT(pv, pv_list)) { 2946 2947 /* 2948 * don't write protect pager mappings 2949 */ 2950 if (!setem && (bit == PG_RW)) { 2951 if (!pmap_track_modified(pv->pv_va)) 2952 continue; 2953 } 2954 2955#if defined(PMAP_DIAGNOSTIC) 2956 if (!pv->pv_pmap) { 2957 printf("Null pmap (cb) at va: 0x%x\n", pv->pv_va); 2958 continue; 2959 } 2960#endif 2961 2962 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2963 2964 if (setem) { 2965 *(int *)pte |= bit; 2966 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 2967 } else { 2968 vm_offset_t pbits = *(vm_offset_t *)pte; 2969 if (pbits & bit) { 2970 if (bit == PG_RW) { 2971 if (pbits & PG_M) { 2972 vm_page_dirty(m); 2973 } 2974 *(int *)pte = pbits & ~(PG_M|PG_RW); 2975 } else { 2976 *(int *)pte = pbits & ~bit; 2977 } 2978 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 2979 } 2980 } 2981 } 2982 splx(s); 2983} 2984 2985/* 2986 * pmap_page_protect: 2987 * 2988 * Lower the permission for all mappings to a given page. 2989 */ 2990void 2991pmap_page_protect(vm_page_t m, vm_prot_t prot) 2992{ 2993 if ((prot & VM_PROT_WRITE) == 0) { 2994 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 2995 pmap_changebit(m, PG_RW, FALSE); 2996 } else { 2997 pmap_remove_all(m); 2998 } 2999 } 3000} 3001 3002vm_offset_t 3003pmap_phys_address(ppn) 3004 int ppn; 3005{ 3006 return (i386_ptob(ppn)); 3007} 3008 3009/* 3010 * pmap_ts_referenced: 3011 * 3012 * Return the count of reference bits for a page, clearing all of them. 3013 */ 3014int 3015pmap_ts_referenced(vm_page_t m) 3016{ 3017 register pv_entry_t pv, pvf, pvn; 3018 unsigned *pte; 3019 int s; 3020 int rtval = 0; 3021 3022 if (!pmap_initialized || (m->flags & PG_FICTITIOUS)) 3023 return (rtval); 3024 3025 s = splvm(); 3026 3027 if ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { 3028 3029 pvf = pv; 3030 3031 do { 3032 pvn = TAILQ_NEXT(pv, pv_list); 3033 3034 TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); 3035 3036 TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); 3037 3038 if (!pmap_track_modified(pv->pv_va)) 3039 continue; 3040 3041 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3042 3043 if (pte && (*pte & PG_A)) { 3044 *pte &= ~PG_A; 3045 3046 pmap_TLB_invalidate(pv->pv_pmap, pv->pv_va); 3047 3048 rtval++; 3049 if (rtval > 4) { 3050 break; 3051 } 3052 } 3053 } while ((pv = pvn) != NULL && pv != pvf); 3054 } 3055 splx(s); 3056 3057 return (rtval); 3058} 3059 3060/* 3061 * pmap_is_modified: 3062 * 3063 * Return whether or not the specified physical page was modified 3064 * in any physical maps. 3065 */ 3066boolean_t 3067pmap_is_modified(vm_page_t m) 3068{ 3069 return pmap_testbit(m, PG_M); 3070} 3071 3072/* 3073 * Clear the modify bits on the specified physical page. 3074 */ 3075void 3076pmap_clear_modify(vm_page_t m) 3077{ 3078 pmap_changebit(m, PG_M, FALSE); 3079} 3080 3081/* 3082 * pmap_clear_reference: 3083 * 3084 * Clear the reference bit on the specified physical page. 3085 */ 3086void 3087pmap_clear_reference(vm_page_t m) 3088{ 3089 pmap_changebit(m, PG_A, FALSE); 3090} 3091 3092/* 3093 * Miscellaneous support routines follow 3094 */ 3095 3096static void 3097i386_protection_init() 3098{ 3099 register int *kp, prot; 3100 3101 kp = protection_codes; 3102 for (prot = 0; prot < 8; prot++) { 3103 switch (prot) { 3104 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3105 /* 3106 * Read access is also 0. There isn't any execute bit, 3107 * so just make it readable. 3108 */ 3109 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3110 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3111 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3112 *kp++ = 0; 3113 break; 3114 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3115 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3116 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3117 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3118 *kp++ = PG_RW; 3119 break; 3120 } 3121 } 3122} 3123 3124/* 3125 * Map a set of physical memory pages into the kernel virtual 3126 * address space. Return a pointer to where it is mapped. This 3127 * routine is intended to be used for mapping device memory, 3128 * NOT real memory. 3129 */ 3130void * 3131pmap_mapdev(pa, size) 3132 vm_offset_t pa; 3133 vm_size_t size; 3134{ 3135 vm_offset_t va, tmpva, offset; 3136 unsigned *pte; 3137 3138 offset = pa & PAGE_MASK; 3139 size = roundup(offset + size, PAGE_SIZE); 3140 3141 va = kmem_alloc_pageable(kernel_map, size); 3142 if (!va) 3143 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3144 3145 pa = pa & PG_FRAME; 3146 for (tmpva = va; size > 0;) { 3147 pte = (unsigned *)vtopte(tmpva); 3148 *pte = pa | PG_RW | PG_V | pgeflag; 3149 size -= PAGE_SIZE; 3150 tmpva += PAGE_SIZE; 3151 pa += PAGE_SIZE; 3152 } 3153 invltlb(); 3154 3155 return ((void *)(va + offset)); 3156} 3157 3158void 3159pmap_unmapdev(va, size) 3160 vm_offset_t va; 3161 vm_size_t size; 3162{ 3163 vm_offset_t base, offset; 3164 3165 base = va & PG_FRAME; 3166 offset = va & PAGE_MASK; 3167 size = roundup(offset + size, PAGE_SIZE); 3168 kmem_free(kernel_map, base, size); 3169} 3170 3171/* 3172 * perform the pmap work for mincore 3173 */ 3174int 3175pmap_mincore(pmap, addr) 3176 pmap_t pmap; 3177 vm_offset_t addr; 3178{ 3179 3180 unsigned *ptep, pte; 3181 vm_page_t m; 3182 int val = 0; 3183 3184 ptep = pmap_pte(pmap, addr); 3185 if (ptep == 0) { 3186 return 0; 3187 } 3188 3189 if ((pte = *ptep) != 0) { 3190 vm_offset_t pa; 3191 3192 val = MINCORE_INCORE; 3193 if ((pte & PG_MANAGED) == 0) 3194 return val; 3195 3196 pa = pte & PG_FRAME; 3197 3198 m = PHYS_TO_VM_PAGE(pa); 3199 3200 /* 3201 * Modified by us 3202 */ 3203 if (pte & PG_M) 3204 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3205 /* 3206 * Modified by someone 3207 */ 3208 else if (m->dirty || pmap_is_modified(m)) 3209 val |= MINCORE_MODIFIED_OTHER; 3210 /* 3211 * Referenced by us 3212 */ 3213 if (pte & PG_A) 3214 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3215 3216 /* 3217 * Referenced by someone 3218 */ 3219 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(m)) { 3220 val |= MINCORE_REFERENCED_OTHER; 3221 vm_page_flag_set(m, PG_REFERENCED); 3222 } 3223 } 3224 return val; 3225} 3226 3227void 3228pmap_activate(struct proc *p) 3229{ 3230 pmap_t pmap; 3231 3232 pmap = vmspace_pmap(p->p_vmspace); 3233#if defined(SMP) 3234 pmap->pm_active |= 1 << PCPU_GET(cpuid); 3235#else 3236 pmap->pm_active |= 1; 3237#endif 3238#if defined(SWTCH_OPTIM_STATS) 3239 tlb_flush_count++; 3240#endif 3241 load_cr3(p->p_addr->u_pcb.pcb_cr3 = vtophys(pmap->pm_pdir)); 3242} 3243 3244vm_offset_t 3245pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) 3246{ 3247 3248 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3249 return addr; 3250 } 3251 3252 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3253 return addr; 3254} 3255 3256 3257#if defined(PMAP_DEBUG) 3258pmap_pid_dump(int pid) 3259{ 3260 pmap_t pmap; 3261 struct proc *p; 3262 int npte = 0; 3263 int index; 3264 3265 sx_slock(&allproc_lock); 3266 LIST_FOREACH(p, &allproc, p_list) { 3267 if (p->p_pid != pid) 3268 continue; 3269 3270 if (p->p_vmspace) { 3271 int i,j; 3272 index = 0; 3273 pmap = vmspace_pmap(p->p_vmspace); 3274 for(i=0;i<1024;i++) { 3275 pd_entry_t *pde; 3276 unsigned *pte; 3277 unsigned base = i << PDRSHIFT; 3278 3279 pde = &pmap->pm_pdir[i]; 3280 if (pde && pmap_pde_v(pde)) { 3281 for(j=0;j<1024;j++) { 3282 unsigned va = base + (j << PAGE_SHIFT); 3283 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3284 if (index) { 3285 index = 0; 3286 printf("\n"); 3287 } 3288 sx_sunlock(&allproc_lock); 3289 return npte; 3290 } 3291 pte = pmap_pte_quick( pmap, va); 3292 if (pte && pmap_pte_v(pte)) { 3293 vm_offset_t pa; 3294 vm_page_t m; 3295 pa = *(int *)pte; 3296 m = PHYS_TO_VM_PAGE(pa); 3297 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3298 va, pa, m->hold_count, m->wire_count, m->flags); 3299 npte++; 3300 index++; 3301 if (index >= 2) { 3302 index = 0; 3303 printf("\n"); 3304 } else { 3305 printf(" "); 3306 } 3307 } 3308 } 3309 } 3310 } 3311 } 3312 } 3313 sx_sunlock(&allproc_lock); 3314 return npte; 3315} 3316#endif 3317 3318#if defined(DEBUG) 3319 3320static void pads __P((pmap_t pm)); 3321void pmap_pvdump __P((vm_offset_t pa)); 3322 3323/* print address space of pmap*/ 3324static void 3325pads(pm) 3326 pmap_t pm; 3327{ 3328 unsigned va, i, j; 3329 unsigned *ptep; 3330 3331 if (pm == kernel_pmap) 3332 return; 3333 for (i = 0; i < 1024; i++) 3334 if (pm->pm_pdir[i]) 3335 for (j = 0; j < 1024; j++) { 3336 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3337 if (pm == kernel_pmap && va < KERNBASE) 3338 continue; 3339 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3340 continue; 3341 ptep = pmap_pte_quick(pm, va); 3342 if (pmap_pte_v(ptep)) 3343 printf("%x:%x ", va, *(int *) ptep); 3344 }; 3345 3346} 3347 3348void 3349pmap_pvdump(pa) 3350 vm_offset_t pa; 3351{ 3352 register pv_entry_t pv; 3353 vm_page_t m; 3354 3355 printf("pa %x", pa); 3356 m = PHYS_TO_VM_PAGE(pa); 3357 for (pv = TAILQ_FIRST(&m->md.pv_list); 3358 pv; 3359 pv = TAILQ_NEXT(pv, pv_list)) { 3360#ifdef used_to_be 3361 printf(" -> pmap %p, va %x, flags %x", 3362 (void *)pv->pv_pmap, pv->pv_va, pv->pv_flags); 3363#endif 3364 printf(" -> pmap %p, va %x", (void *)pv->pv_pmap, pv->pv_va); 3365 pads(pv->pv_pmap); 3366 } 3367 printf(" "); 3368} 3369#endif 3370