pmap.c revision 33221
14Srgrimes/* 24Srgrimes * Copyright (c) 1991 Regents of the University of California. 34Srgrimes * All rights reserved. 44Srgrimes * Copyright (c) 1994 John S. Dyson 54Srgrimes * All rights reserved. 64Srgrimes * Copyright (c) 1994 David Greenman 74Srgrimes * All rights reserved. 84Srgrimes * 94Srgrimes * This code is derived from software contributed to Berkeley by 104Srgrimes * the Systems Programming Group of the University of Utah Computer 114Srgrimes * Science Department and William Jolitz of UUNET Technologies Inc. 124Srgrimes * 134Srgrimes * Redistribution and use in source and binary forms, with or without 144Srgrimes * modification, are permitted provided that the following conditions 154Srgrimes * are met: 164Srgrimes * 1. Redistributions of source code must retain the above copyright 174Srgrimes * notice, this list of conditions and the following disclaimer. 184Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 194Srgrimes * notice, this list of conditions and the following disclaimer in the 204Srgrimes * documentation and/or other materials provided with the distribution. 214Srgrimes * 3. All advertising materials mentioning features or use of this software 224Srgrimes * must display the following acknowledgement: 234Srgrimes * This product includes software developed by the University of 244Srgrimes * California, Berkeley and its contributors. 254Srgrimes * 4. Neither the name of the University nor the names of its contributors 264Srgrimes * may be used to endorse or promote products derived from this software 274Srgrimes * without specific prior written permission. 284Srgrimes * 294Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 304Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 314Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32621Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 3350477Speter * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 344Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 354Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 364Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37122292Speter * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 384Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 394Srgrimes * SUCH DAMAGE. 404Srgrimes * 41122292Speter * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91 42122292Speter * $Id: pmap.c,v 1.184 1998/02/09 06:08:16 eivind Exp $ 434Srgrimes */ 44114349Speter 4583047Sobrien/* 46230503Skib * Manages physical address maps. 474Srgrimes * 484Srgrimes * In addition to hardware address maps, this 49114349Speter * module is called upon to provide software-use-only 50114349Speter * maps which may or may not be stored in the same 51230503Skib * form as hardware maps. These pseudo-maps are 524Srgrimes * used to store intermediate results from copy 534Srgrimes * operations to and from address spaces. 54230503Skib * 55230503Skib * Since the information managed by this module is 56230503Skib * also stored by the logical address mapping module, 57230503Skib * this module may throw away valid virtual-to-physical 58230503Skib * mappings at almost any time. However, invalidations 5979609Speter * of virtual-to-physical mappings must be done as 60230503Skib * requested. 61230503Skib * 62230503Skib * In order to cope with hardware architectures which 63230503Skib * make virtual-to-physical map invalidates expensive, 64230503Skib * this module may delay invalidate or reduced protection 65230503Skib * operations until such time as they are actually 66230503Skib * necessary. This module is given full information as 67230503Skib * to which processors are currently using which maps, 68230503Skib * and to when physical maps must be made correct. 6979609Speter */ 7079609Speter 71114349Speter#include "opt_disable_pse.h" 7279609Speter 7379609Speter#include <sys/param.h> 7479609Speter#include <sys/systm.h> 75230503Skib#include <sys/proc.h> 7679609Speter#include <sys/msgbuf.h> 77114349Speter#include <sys/vmmeter.h> 78230503Skib#include <sys/mman.h> 79103834Speter 8079609Speter#include <vm/vm.h> 81230504Skib#include <vm/vm_param.h> 82230504Skib#include <vm/vm_prot.h> 83230504Skib#include <sys/lock.h> 84230504Skib#include <vm/vm_kern.h> 85230504Skib#include <vm/vm_page.h> 86230504Skib#include <vm/vm_map.h> 87230504Skib#include <vm/vm_object.h> 88230504Skib#include <vm/vm_extern.h> 89230504Skib#include <vm/vm_pageout.h> 90230504Skib#include <vm/vm_pager.h> 91230504Skib#include <vm/vm_zone.h> 92230504Skib 93230504Skib#include <sys/user.h> 94230504Skib 95230504Skib#include <machine/cputypes.h> 96230504Skib#include <machine/md_var.h> 97230504Skib#include <machine/specialreg.h> 98230504Skib#if defined(SMP) || defined(APIC_IO) 99230504Skib#include <machine/smp.h> 100230504Skib#include <machine/apic.h> 101230504Skib#endif /* SMP || APIC_IO */ 102230504Skib 103208833Skib#define PMAP_KEEP_PDIRS 104208833Skib#ifndef PMAP_SHPGPERPROC 105208833Skib#define PMAP_SHPGPERPROC 200 106208833Skib#endif 107208833Skib 108208833Skib#if defined(DIAGNOSTIC) 109208833Skib#define PMAP_DIAGNOSTIC 110208833Skib#endif 111208833Skib 112208833Skib#define MINPV 2048 113208833Skib 1144Srgrimes#if !defined(PMAP_DIAGNOSTIC) 11557890Scracauer#define PMAP_INLINE __inline 11657890Scracauer#else 1174Srgrimes#define PMAP_INLINE 1184Srgrimes#endif 1194Srgrimes 1204Srgrimes/* 1214Srgrimes * Get PDEs and PTEs for user/kernel address space 122117865Speter */ 123117865Speter#define pmap_pde(m, v) (&((m)->pm_pdir[(vm_offset_t)(v) >> PDRSHIFT])) 124117865Speter#define pdir_pde(m, v) (m[(vm_offset_t)(v) >> PDRSHIFT]) 125117865Speter 126117865Speter#define pmap_pde_v(pte) ((*(int *)pte & PG_V) != 0) 127117865Speter#define pmap_pte_w(pte) ((*(int *)pte & PG_W) != 0) 1284Srgrimes#define pmap_pte_m(pte) ((*(int *)pte & PG_M) != 0) 129117865Speter#define pmap_pte_u(pte) ((*(int *)pte & PG_A) != 0) 130117865Speter#define pmap_pte_v(pte) ((*(int *)pte & PG_V) != 0) 1314Srgrimes 132122292Speter#define pmap_pte_set_w(pte, v) ((v)?(*(int *)pte |= PG_W):(*(int *)pte &= ~PG_W)) 133187867Sjhb#define pmap_pte_set_prot(pte, v) ((*(int *)pte &= ~PG_PROT), (*(int *)pte |= (v))) 134114859Speter 135114859Speter/* 1364Srgrimes * Given a map and a machine independent protection code, 13755205Speter * convert to a vax protection code. 138189412Sjhb */ 139122292Speter#define pte_prot(m, p) (protection_codes[p]) 140122292Speterstatic int protection_codes[8]; 141122292Speter 142215865Skib#define pa_index(pa) atop((pa) - vm_first_phys) 143122295Speter#define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) 144122292Speter 145122292Speterstatic struct pmap kernel_pmap_store; 146215865Skibpmap_t kernel_pmap; 147208833Skibextern pd_entry_t my_idlePTD; 148208833Skib 149208833Skibvm_offset_t avail_start; /* PA of first available physical page */ 150208833Skibvm_offset_t avail_end; /* PA of last available physical page */ 151208833Skibvm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ 152208833Skibvm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ 153208833Skibstatic boolean_t pmap_initialized = FALSE; /* Has pmap_init completed? */ 154208833Skibstatic vm_offset_t vm_first_phys; 155208833Skibstatic int pgeflag; /* PG_G or-in */ 156208833Skibstatic int pseflag; /* PG_PS or-in */ 157208833Skibstatic int pv_npg; 1584175Sbde 1594175Sbdestatic int nkpt; 160122292Spetervm_offset_t kernel_vm_end; 161 162/* 163 * Data for the pv entry allocation mechanism 164 */ 165static vm_zone_t pvzone; 166static struct vm_zone pvzone_store; 167static struct vm_object pvzone_obj; 168static int pv_entry_count=0, pv_entry_max=0, pv_entry_high_water=0; 169static int pmap_pagedaemon_waken = 0; 170static struct pv_entry *pvinit; 171 172/* 173 * All those kernel PT submaps that BSD is so fond of 174 */ 175pt_entry_t *CMAP1 = 0; 176static pt_entry_t *CMAP2, *ptmmap; 177static pv_table_t *pv_table; 178caddr_t CADDR1 = 0, ptvmmap = 0; 179static caddr_t CADDR2; 180static pt_entry_t *msgbufmap; 181struct msgbuf *msgbufp=0; 182 183#ifdef SMP 184extern char prv_CPAGE1[], prv_CPAGE2[], prv_CPAGE3[]; 185extern pt_entry_t *prv_CMAP1, *prv_CMAP2, *prv_CMAP3; 186extern pd_entry_t *IdlePTDS[]; 187extern pt_entry_t SMP_prvpt[]; 188#endif 189 190static pt_entry_t *PMAP1 = 0; 191static unsigned *PADDR1 = 0; 192 193static PMAP_INLINE void free_pv_entry __P((pv_entry_t pv)); 194static unsigned * get_ptbase __P((pmap_t pmap)); 195static pv_entry_t get_pv_entry __P((void)); 196static void i386_protection_init __P((void)); 197static void pmap_changebit __P((vm_offset_t pa, int bit, boolean_t setem)); 198 199static PMAP_INLINE int pmap_is_managed __P((vm_offset_t pa)); 200static void pmap_remove_all __P((vm_offset_t pa)); 201static vm_page_t pmap_enter_quick __P((pmap_t pmap, vm_offset_t va, 202 vm_offset_t pa, vm_page_t mpte)); 203static int pmap_remove_pte __P((struct pmap *pmap, unsigned *ptq, 204 vm_offset_t sva)); 205static void pmap_remove_page __P((struct pmap *pmap, vm_offset_t va)); 206static int pmap_remove_entry __P((struct pmap *pmap, pv_table_t *pv, 207 vm_offset_t va)); 208static boolean_t pmap_testbit __P((vm_offset_t pa, int bit)); 209static void pmap_insert_entry __P((pmap_t pmap, vm_offset_t va, 210 vm_page_t mpte, vm_offset_t pa)); 211 212static vm_page_t pmap_allocpte __P((pmap_t pmap, vm_offset_t va)); 213 214static int pmap_release_free_page __P((pmap_t pmap, vm_page_t p)); 215static vm_page_t _pmap_allocpte __P((pmap_t pmap, unsigned ptepindex)); 216static unsigned * pmap_pte_quick __P((pmap_t pmap, vm_offset_t va)); 217static vm_page_t pmap_page_lookup __P((vm_object_t object, vm_pindex_t pindex)); 218static int pmap_unuse_pt __P((pmap_t, vm_offset_t, vm_page_t)); 219static vm_offset_t pmap_kmem_choose(vm_offset_t addr) ; 220void pmap_collect(void); 221 222static unsigned pdir4mb; 223 224/* 225 * Routine: pmap_pte 226 * Function: 227 * Extract the page table entry associated 228 * with the given map/virtual_address pair. 229 */ 230 231PMAP_INLINE unsigned * 232pmap_pte(pmap, va) 233 register pmap_t pmap; 234 vm_offset_t va; 235{ 236 unsigned *pdeaddr; 237 238 if (pmap) { 239 pdeaddr = (unsigned *) pmap_pde(pmap, va); 240 if (*pdeaddr & PG_PS) 241 return pdeaddr; 242 if (*pdeaddr) { 243 return get_ptbase(pmap) + i386_btop(va); 244 } 245 } 246 return (0); 247} 248 249/* 250 * Move the kernel virtual free pointer to the next 251 * 4MB. This is used to help improve performance 252 * by using a large (4MB) page for much of the kernel 253 * (.text, .data, .bss) 254 */ 255static vm_offset_t 256pmap_kmem_choose(vm_offset_t addr) { 257 vm_offset_t newaddr = addr; 258#ifndef DISABLE_PSE 259 if (cpu_feature & CPUID_PSE) { 260 newaddr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 261 } 262#endif 263 return newaddr; 264} 265 266/* 267 * Bootstrap the system enough to run with virtual memory. 268 * 269 * On the i386 this is called after mapping has already been enabled 270 * and just syncs the pmap module with what has already been done. 271 * [We can't call it easily with mapping off since the kernel is not 272 * mapped with PA == VA, hence we would have to relocate every address 273 * from the linked base (virtual) address "KERNBASE" to the actual 274 * (physical) address starting relative to 0] 275 */ 276void 277pmap_bootstrap(firstaddr, loadaddr) 278 vm_offset_t firstaddr; 279 vm_offset_t loadaddr; 280{ 281 vm_offset_t va; 282 pt_entry_t *pte; 283 int i, j; 284 285 avail_start = firstaddr; 286 287 /* 288 * XXX The calculation of virtual_avail is wrong. It's NKPT*PAGE_SIZE too 289 * large. It should instead be correctly calculated in locore.s and 290 * not based on 'first' (which is a physical address, not a virtual 291 * address, for the start of unused physical memory). The kernel 292 * page tables are NOT double mapped and thus should not be included 293 * in this calculation. 294 */ 295 virtual_avail = (vm_offset_t) KERNBASE + firstaddr; 296 virtual_avail = pmap_kmem_choose(virtual_avail); 297 298 virtual_end = VM_MAX_KERNEL_ADDRESS; 299 300 /* 301 * Initialize protection array. 302 */ 303 i386_protection_init(); 304 305 /* 306 * The kernel's pmap is statically allocated so we don't have to use 307 * pmap_create, which is unlikely to work correctly at this part of 308 * the boot sequence (XXX and which no longer exists). 309 */ 310 kernel_pmap = &kernel_pmap_store; 311 312 kernel_pmap->pm_pdir = (pd_entry_t *) (KERNBASE + (u_int)IdlePTD); 313 314 kernel_pmap->pm_count = 1; 315 TAILQ_INIT(&kernel_pmap->pm_pvlist); 316 nkpt = NKPT; 317 318 /* 319 * Reserve some special page table entries/VA space for temporary 320 * mapping of pages. 321 */ 322#define SYSMAP(c, p, v, n) \ 323 v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); 324 325 va = virtual_avail; 326 pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); 327 328 /* 329 * CMAP1/CMAP2 are used for zeroing and copying pages. 330 */ 331 SYSMAP(caddr_t, CMAP1, CADDR1, 1) 332 SYSMAP(caddr_t, CMAP2, CADDR2, 1) 333 334 /* 335 * ptvmmap is used for reading arbitrary physical pages via /dev/mem. 336 * XXX ptmmap is not used. 337 */ 338 SYSMAP(caddr_t, ptmmap, ptvmmap, 1) 339 340 /* 341 * msgbufp is used to map the system message buffer. 342 * XXX msgbufmap is not used. 343 */ 344 SYSMAP(struct msgbuf *, msgbufmap, msgbufp, 345 atop(round_page(sizeof(struct msgbuf)))) 346 347 /* 348 * ptemap is used for pmap_pte_quick 349 */ 350 SYSMAP(unsigned *, PMAP1, PADDR1, 1); 351 352 virtual_avail = va; 353 354 *(int *) CMAP1 = *(int *) CMAP2 = 0; 355 *(int *) PTD = 0; 356 357 358 pgeflag = 0; 359#if !defined(SMP) 360 if (cpu_feature & CPUID_PGE) { 361 pgeflag = PG_G; 362 } 363#endif 364 365/* 366 * Initialize the 4MB page size flag 367 */ 368 pseflag = 0; 369/* 370 * The 4MB page version of the initial 371 * kernel page mapping. 372 */ 373 pdir4mb = 0; 374 375#if !defined(DISABLE_PSE) 376 if (cpu_feature & CPUID_PSE) { 377 unsigned ptditmp; 378 /* 379 * Enable the PSE mode 380 */ 381 load_cr4(rcr4() | CR4_PSE); 382 383 /* 384 * Note that we have enabled PSE mode 385 */ 386 pseflag = PG_PS; 387 ptditmp = (unsigned) kernel_pmap->pm_pdir[KPTDI]; 388 ptditmp &= ~(NBPDR - 1); 389 ptditmp |= PG_V | PG_RW | PG_PS | PG_U | pgeflag; 390 pdir4mb = ptditmp; 391 /* 392 * We can do the mapping here for the single processor 393 * case. We simply ignore the old page table page from 394 * now on. 395 */ 396#if !defined(SMP) 397 PTD[KPTDI] = (pd_entry_t) ptditmp; 398 kernel_pmap->pm_pdir[KPTDI] = (pd_entry_t) ptditmp; 399 invltlb(); 400#endif 401 } 402#endif 403 404#ifdef SMP 405 if (cpu_apic_address == 0) 406 panic("pmap_bootstrap: no local apic!"); 407 408 /* 0 = private page */ 409 /* 1 = page table page */ 410 /* 2 = local apic */ 411 /* 16-31 = io apics */ 412 SMP_prvpt[2] = (pt_entry_t)(PG_V | PG_RW | pgeflag | ((u_long)cpu_apic_address & PG_FRAME)); 413 414 for (i = 0; i < mp_napics; i++) { 415 for (j = 0; j < 16; j++) { 416 /* same page frame as a previous IO apic? */ 417 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 418 ((u_long)io_apic_address[0] & PG_FRAME)) { 419 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 420 break; 421 } 422 /* use this slot if available */ 423 if (((u_long)SMP_prvpt[j + 16] & PG_FRAME) == 0) { 424 SMP_prvpt[j + 16] = (pt_entry_t)(PG_V | PG_RW | pgeflag | 425 ((u_long)io_apic_address[i] & PG_FRAME)); 426 ioapic[i] = (ioapic_t *)&SMP_ioapic[j * PAGE_SIZE]; 427 break; 428 } 429 } 430 if (j == 16) 431 panic("no space to map IO apic %d!", i); 432 } 433 434 /* BSP does this itself, AP's get it pre-set */ 435 prv_CMAP1 = (pt_entry_t *)&SMP_prvpt[3 + UPAGES]; 436 prv_CMAP2 = (pt_entry_t *)&SMP_prvpt[4 + UPAGES]; 437 prv_CMAP3 = (pt_entry_t *)&SMP_prvpt[5 + UPAGES]; 438#endif 439 440 invltlb(); 441 442} 443 444/* 445 * Set 4mb pdir for mp startup, and global flags 446 */ 447void 448pmap_set_opt(unsigned *pdir) { 449 int i; 450 451 if (pseflag && (cpu_feature & CPUID_PSE)) { 452 load_cr4(rcr4() | CR4_PSE); 453 if (pdir4mb) { 454 (unsigned) pdir[KPTDI] = pdir4mb; 455 } 456 } 457 458 if (pgeflag && (cpu_feature & CPUID_PGE)) { 459 load_cr4(rcr4() | CR4_PGE); 460 for(i = KPTDI; i < KPTDI + nkpt; i++) { 461 if (pdir[i]) { 462 pdir[i] |= PG_G; 463 } 464 } 465 } 466} 467 468/* 469 * Setup the PTD for the boot processor 470 */ 471void 472pmap_set_opt_bsp(void) 473{ 474 pmap_set_opt((unsigned *)kernel_pmap->pm_pdir); 475 pmap_set_opt((unsigned *)PTD); 476 invltlb(); 477} 478 479/* 480 * Initialize the pmap module. 481 * Called by vm_init, to initialize any structures that the pmap 482 * system needs to map virtual memory. 483 * pmap_init has been enhanced to support in a fairly consistant 484 * way, discontiguous physical memory. 485 */ 486void 487pmap_init(phys_start, phys_end) 488 vm_offset_t phys_start, phys_end; 489{ 490 vm_offset_t addr; 491 vm_size_t s; 492 int i; 493 int initial_pvs; 494 495 /* 496 * calculate the number of pv_entries needed 497 */ 498 vm_first_phys = phys_avail[0]; 499 for (i = 0; phys_avail[i + 1]; i += 2); 500 pv_npg = (phys_avail[(i - 2) + 1] - vm_first_phys) / PAGE_SIZE; 501 502 /* 503 * Allocate memory for random pmap data structures. Includes the 504 * pv_head_table. 505 */ 506 s = (vm_size_t) (sizeof(pv_table_t) * pv_npg); 507 s = round_page(s); 508 509 addr = (vm_offset_t) kmem_alloc(kernel_map, s); 510 pv_table = (pv_table_t *) addr; 511 for(i = 0; i < pv_npg; i++) { 512 vm_offset_t pa; 513 TAILQ_INIT(&pv_table[i].pv_list); 514 pv_table[i].pv_list_count = 0; 515 pa = vm_first_phys + i * PAGE_SIZE; 516 pv_table[i].pv_vm_page = PHYS_TO_VM_PAGE(pa); 517 } 518 519 /* 520 * init the pv free list 521 */ 522 initial_pvs = pv_npg; 523 if (initial_pvs < MINPV) 524 initial_pvs = MINPV; 525 pvzone = &pvzone_store; 526 pvinit = (struct pv_entry *) kmem_alloc(kernel_map, 527 initial_pvs * sizeof (struct pv_entry)); 528 zbootinit(pvzone, "PV ENTRY", sizeof (struct pv_entry), pvinit, pv_npg); 529 530 /* 531 * Now it is safe to enable pv_table recording. 532 */ 533 pmap_initialized = TRUE; 534} 535 536/* 537 * Initialize the address space (zone) for the pv_entries. Set a 538 * high water mark so that the system can recover from excessive 539 * numbers of pv entries. 540 */ 541void 542pmap_init2() { 543 pv_entry_max = PMAP_SHPGPERPROC * maxproc + pv_npg; 544 pv_entry_high_water = 9 * (pv_entry_max / 10); 545 zinitna(pvzone, &pvzone_obj, NULL, 0, pv_entry_max, ZONE_INTERRUPT, 1); 546} 547 548/* 549 * Used to map a range of physical addresses into kernel 550 * virtual address space. 551 * 552 * For now, VM is already on, we only need to map the 553 * specified memory. 554 */ 555vm_offset_t 556pmap_map(virt, start, end, prot) 557 vm_offset_t virt; 558 vm_offset_t start; 559 vm_offset_t end; 560 int prot; 561{ 562 while (start < end) { 563 pmap_enter(kernel_pmap, virt, start, prot, FALSE); 564 virt += PAGE_SIZE; 565 start += PAGE_SIZE; 566 } 567 return (virt); 568} 569 570 571/*************************************************** 572 * Low level helper routines..... 573 ***************************************************/ 574 575#if defined(PMAP_DIAGNOSTIC) 576 577/* 578 * This code checks for non-writeable/modified pages. 579 * This should be an invalid condition. 580 */ 581static int 582pmap_nw_modified(pt_entry_t ptea) { 583 int pte; 584 585 pte = (int) ptea; 586 587 if ((pte & (PG_M|PG_RW)) == PG_M) 588 return 1; 589 else 590 return 0; 591} 592#endif 593 594 595/* 596 * this routine defines the region(s) of memory that should 597 * not be tested for the modified bit. 598 */ 599static PMAP_INLINE int 600pmap_track_modified( vm_offset_t va) { 601 if ((va < clean_sva) || (va >= clean_eva)) 602 return 1; 603 else 604 return 0; 605} 606 607static PMAP_INLINE void 608invltlb_1pg( vm_offset_t va) { 609#if defined(I386_CPU) 610 if (cpu_class == CPUCLASS_386) { 611 invltlb(); 612 } else 613#endif 614 { 615 invlpg(va); 616 } 617} 618 619static PMAP_INLINE void 620invltlb_2pg( vm_offset_t va1, vm_offset_t va2) { 621#if defined(I386_CPU) 622 if (cpu_class == CPUCLASS_386) { 623 invltlb(); 624 } else 625#endif 626 { 627 invlpg(va1); 628 invlpg(va2); 629 } 630} 631 632static unsigned * 633get_ptbase(pmap) 634 pmap_t pmap; 635{ 636 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 637 638 /* are we current address space or kernel? */ 639 if (pmap == kernel_pmap || frame == (((unsigned) PTDpde) & PG_FRAME)) { 640 return (unsigned *) PTmap; 641 } 642 /* otherwise, we are alternate address space */ 643 if (frame != (((unsigned) APTDpde) & PG_FRAME)) { 644 APTDpde = (pd_entry_t) (frame | PG_RW | PG_V); 645 invltlb(); 646 } 647 return (unsigned *) APTmap; 648} 649 650/* 651 * Super fast pmap_pte routine best used when scanning 652 * the pv lists. This eliminates many coarse-grained 653 * invltlb calls. Note that many of the pv list 654 * scans are across different pmaps. It is very wasteful 655 * to do an entire invltlb for checking a single mapping. 656 */ 657 658static unsigned * 659pmap_pte_quick(pmap, va) 660 register pmap_t pmap; 661 vm_offset_t va; 662{ 663 unsigned pde, newpf; 664 if (pde = (unsigned) pmap->pm_pdir[va >> PDRSHIFT]) { 665 unsigned frame = (unsigned) pmap->pm_pdir[PTDPTDI] & PG_FRAME; 666 unsigned index = i386_btop(va); 667 /* are we current address space or kernel? */ 668 if ((pmap == kernel_pmap) || 669 (frame == (((unsigned) PTDpde) & PG_FRAME))) { 670 return (unsigned *) PTmap + index; 671 } 672 newpf = pde & PG_FRAME; 673 if ( ((* (unsigned *) PMAP1) & PG_FRAME) != newpf) { 674 * (unsigned *) PMAP1 = newpf | PG_RW | PG_V; 675 invltlb_1pg((vm_offset_t) PADDR1); 676 } 677 return PADDR1 + ((unsigned) index & (NPTEPG - 1)); 678 } 679 return (0); 680} 681 682/* 683 * Routine: pmap_extract 684 * Function: 685 * Extract the physical page address associated 686 * with the given map/virtual_address pair. 687 */ 688vm_offset_t 689pmap_extract(pmap, va) 690 register pmap_t pmap; 691 vm_offset_t va; 692{ 693 vm_offset_t rtval; 694 vm_offset_t pdirindex; 695 pdirindex = va >> PDRSHIFT; 696 if (pmap && (rtval = (unsigned) pmap->pm_pdir[pdirindex])) { 697 unsigned *pte; 698 if ((rtval & PG_PS) != 0) { 699 rtval &= ~(NBPDR - 1); 700 rtval |= va & (NBPDR - 1); 701 return rtval; 702 } 703 pte = get_ptbase(pmap) + i386_btop(va); 704 rtval = ((*pte & PG_FRAME) | (va & PAGE_MASK)); 705 return rtval; 706 } 707 return 0; 708 709} 710 711/* 712 * determine if a page is managed (memory vs. device) 713 */ 714static PMAP_INLINE int 715pmap_is_managed(pa) 716 vm_offset_t pa; 717{ 718 int i; 719 720 if (!pmap_initialized) 721 return 0; 722 723 for (i = 0; phys_avail[i + 1]; i += 2) { 724 if (pa < phys_avail[i + 1] && pa >= phys_avail[i]) 725 return 1; 726 } 727 return 0; 728} 729 730 731/*************************************************** 732 * Low level mapping routines..... 733 ***************************************************/ 734 735/* 736 * Add a list of wired pages to the kva 737 * this routine is only used for temporary 738 * kernel mappings that do not need to have 739 * page modification or references recorded. 740 * Note that old mappings are simply written 741 * over. The page *must* be wired. 742 */ 743void 744pmap_qenter(va, m, count) 745 vm_offset_t va; 746 vm_page_t *m; 747 int count; 748{ 749 int i; 750 register unsigned *pte; 751 752 for (i = 0; i < count; i++) { 753 vm_offset_t tva = va + i * PAGE_SIZE; 754 unsigned npte = VM_PAGE_TO_PHYS(m[i]) | PG_RW | PG_V | pgeflag; 755 unsigned opte; 756 pte = (unsigned *)vtopte(tva); 757 opte = *pte; 758 *pte = npte; 759 if (opte) 760 invltlb_1pg(tva); 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 int i; 774 register unsigned *pte; 775 776 for (i = 0; i < count; i++) { 777 pte = (unsigned *)vtopte(va); 778 *pte = 0; 779 invltlb_1pg(va); 780 va += PAGE_SIZE; 781 } 782} 783 784/* 785 * add a wired page to the kva 786 * note that in order for the mapping to take effect -- you 787 * should do a invltlb after doing the pmap_kenter... 788 */ 789PMAP_INLINE void 790pmap_kenter(va, pa) 791 vm_offset_t va; 792 register vm_offset_t pa; 793{ 794 register unsigned *pte; 795 unsigned npte, opte; 796 797 npte = pa | PG_RW | PG_V | pgeflag; 798 pte = (unsigned *)vtopte(va); 799 opte = *pte; 800 *pte = npte; 801 if (opte) 802 invltlb_1pg(va); 803} 804 805/* 806 * remove a page from the kernel pagetables 807 */ 808PMAP_INLINE void 809pmap_kremove(va) 810 vm_offset_t va; 811{ 812 register unsigned *pte; 813 814 pte = (unsigned *)vtopte(va); 815 *pte = 0; 816 invltlb_1pg(va); 817} 818 819static vm_page_t 820pmap_page_lookup(object, pindex) 821 vm_object_t object; 822 vm_pindex_t pindex; 823{ 824 vm_page_t m; 825retry: 826 m = vm_page_lookup(object, pindex); 827 if (m) { 828 if (m->flags & PG_BUSY) { 829 m->flags |= PG_WANTED; 830 tsleep(m, PVM, "pplookp", 0); 831 goto retry; 832 } 833 } 834 835 return m; 836} 837 838/* 839 * Create the UPAGES for a new process. 840 * This routine directly affects the fork perf for a process. 841 */ 842void 843pmap_new_proc(p) 844 struct proc *p; 845{ 846 int i, updateneeded; 847 vm_object_t upobj; 848 vm_page_t m; 849 struct user *up; 850 unsigned *ptek, oldpte; 851 852 /* 853 * allocate object for the upages 854 */ 855 if ((upobj = p->p_upages_obj) == NULL) { 856 upobj = vm_object_allocate( OBJT_DEFAULT, UPAGES); 857 p->p_upages_obj = upobj; 858 } 859 860 /* get a kernel virtual address for the UPAGES for this proc */ 861 if ((up = p->p_addr) == NULL) { 862 up = (struct user *) kmem_alloc_pageable(kernel_map, 863 UPAGES * PAGE_SIZE); 864#if !defined(MAX_PERF) 865 if (up == NULL) 866 panic("pmap_new_proc: u_map allocation failed"); 867#endif 868 p->p_addr = up; 869 } 870 871 ptek = (unsigned *) vtopte((vm_offset_t) up); 872 873 updateneeded = 0; 874 for(i=0;i<UPAGES;i++) { 875 /* 876 * Get a kernel stack page 877 */ 878 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 879 880 /* 881 * Wire the page 882 */ 883 m->wire_count++; 884 cnt.v_wire_count++; 885 886 oldpte = *(ptek + i); 887 /* 888 * Enter the page into the kernel address space. 889 */ 890 *(ptek + i) = VM_PAGE_TO_PHYS(m) | PG_RW | PG_V | pgeflag; 891 if (oldpte) { 892 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) { 893 invlpg((vm_offset_t) up + i * PAGE_SIZE); 894 } else { 895 updateneeded = 1; 896 } 897 } 898 899 PAGE_WAKEUP(m); 900 m->flags &= ~PG_ZERO; 901 m->flags |= PG_MAPPED | PG_WRITEABLE; 902 m->valid = VM_PAGE_BITS_ALL; 903 } 904 if (updateneeded) 905 invltlb(); 906} 907 908/* 909 * Dispose the UPAGES for a process that has exited. 910 * This routine directly impacts the exit perf of a process. 911 */ 912void 913pmap_dispose_proc(p) 914 struct proc *p; 915{ 916 int i; 917 vm_object_t upobj; 918 vm_page_t m; 919 unsigned *ptek, oldpte; 920 921 upobj = p->p_upages_obj; 922 923 ptek = (unsigned *) vtopte((vm_offset_t) p->p_addr); 924 for(i=0;i<UPAGES;i++) { 925 926 if ((m = vm_page_lookup(upobj, i)) == NULL) 927 panic("pmap_dispose_proc: upage already missing???"); 928 929 m->flags |= PG_BUSY; 930 931 oldpte = *(ptek + i); 932 *(ptek + i) = 0; 933 if ((oldpte & PG_G) || (cpu_class > CPUCLASS_386)) 934 invlpg((vm_offset_t) p->p_addr + i * PAGE_SIZE); 935 vm_page_unwire(m); 936 vm_page_free(m); 937 } 938 939 if (cpu_class <= CPUCLASS_386) 940 invltlb(); 941} 942 943/* 944 * Allow the UPAGES for a process to be prejudicially paged out. 945 */ 946void 947pmap_swapout_proc(p) 948 struct proc *p; 949{ 950 int i; 951 vm_object_t upobj; 952 vm_page_t m; 953 954 upobj = p->p_upages_obj; 955 /* 956 * let the upages be paged 957 */ 958 for(i=0;i<UPAGES;i++) { 959 if ((m = vm_page_lookup(upobj, i)) == NULL) 960 panic("pmap_swapout_proc: upage already missing???"); 961 m->dirty = VM_PAGE_BITS_ALL; 962 vm_page_unwire(m); 963 vm_page_deactivate(m); 964 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 965 } 966} 967 968/* 969 * Bring the UPAGES for a specified process back in. 970 */ 971void 972pmap_swapin_proc(p) 973 struct proc *p; 974{ 975 int i,rv; 976 vm_object_t upobj; 977 vm_page_t m; 978 979 upobj = p->p_upages_obj; 980 for(i=0;i<UPAGES;i++) { 981 982 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 983 984 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 985 VM_PAGE_TO_PHYS(m)); 986 987 if (m->valid != VM_PAGE_BITS_ALL) { 988 rv = vm_pager_get_pages(upobj, &m, 1, 0); 989#if !defined(MAX_PERF) 990 if (rv != VM_PAGER_OK) 991 panic("pmap_swapin_proc: cannot get upages for proc: %d\n", p->p_pid); 992#endif 993 m = vm_page_lookup(upobj, i); 994 m->valid = VM_PAGE_BITS_ALL; 995 } 996 997 vm_page_wire(m); 998 PAGE_WAKEUP(m); 999 m->flags |= PG_MAPPED | PG_WRITEABLE; 1000 } 1001} 1002 1003/*************************************************** 1004 * Page table page management routines..... 1005 ***************************************************/ 1006 1007/* 1008 * This routine unholds page table pages, and if the hold count 1009 * drops to zero, then it decrements the wire count. 1010 */ 1011static int 1012_pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1013 int s; 1014 1015 if (m->flags & PG_BUSY) { 1016 s = splvm(); 1017 while (m->flags & PG_BUSY) { 1018 m->flags |= PG_WANTED; 1019 tsleep(m, PVM, "pmuwpt", 0); 1020 } 1021 splx(s); 1022 } 1023 1024 if (m->hold_count == 0) { 1025 vm_offset_t pteva; 1026 /* 1027 * unmap the page table page 1028 */ 1029 pmap->pm_pdir[m->pindex] = 0; 1030 --pmap->pm_stats.resident_count; 1031 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1032 (((unsigned) PTDpde) & PG_FRAME)) { 1033 /* 1034 * Do a invltlb to make the invalidated mapping 1035 * take effect immediately. 1036 */ 1037 pteva = UPT_MIN_ADDRESS + i386_ptob(m->pindex); 1038 invltlb_1pg(pteva); 1039 } 1040 1041 if (pmap->pm_ptphint == m) 1042 pmap->pm_ptphint = NULL; 1043 1044 /* 1045 * If the page is finally unwired, simply free it. 1046 */ 1047 --m->wire_count; 1048 if (m->wire_count == 0) { 1049 1050 if (m->flags & PG_WANTED) { 1051 m->flags &= ~PG_WANTED; 1052 wakeup(m); 1053 } 1054 1055 m->flags |= PG_BUSY; 1056 vm_page_free_zero(m); 1057 --cnt.v_wire_count; 1058 } 1059 return 1; 1060 } 1061 return 0; 1062} 1063 1064__inline static int 1065pmap_unwire_pte_hold(pmap_t pmap, vm_page_t m) { 1066 vm_page_unhold(m); 1067 if (m->hold_count == 0) 1068 return _pmap_unwire_pte_hold(pmap, m); 1069 else 1070 return 0; 1071} 1072 1073/* 1074 * After removing a page table entry, this routine is used to 1075 * conditionally free the page, and manage the hold/wire counts. 1076 */ 1077static int 1078pmap_unuse_pt(pmap, va, mpte) 1079 pmap_t pmap; 1080 vm_offset_t va; 1081 vm_page_t mpte; 1082{ 1083 unsigned ptepindex; 1084 if (va >= UPT_MIN_ADDRESS) 1085 return 0; 1086 1087 if (mpte == NULL) { 1088 ptepindex = (va >> PDRSHIFT); 1089 if (pmap->pm_ptphint && 1090 (pmap->pm_ptphint->pindex == ptepindex)) { 1091 mpte = pmap->pm_ptphint; 1092 } else { 1093 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1094 pmap->pm_ptphint = mpte; 1095 } 1096 } 1097 1098 return pmap_unwire_pte_hold(pmap, mpte); 1099} 1100 1101#if !defined(SMP) 1102void 1103pmap_pinit0(pmap) 1104 struct pmap *pmap; 1105{ 1106 pmap->pm_pdir = 1107 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1108 pmap_kenter((vm_offset_t) pmap->pm_pdir, (vm_offset_t) IdlePTD); 1109 pmap->pm_flags = 0; 1110 pmap->pm_count = 1; 1111 pmap->pm_ptphint = NULL; 1112 TAILQ_INIT(&pmap->pm_pvlist); 1113 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1114} 1115#else 1116void 1117pmap_pinit0(pmap) 1118 struct pmap *pmap; 1119{ 1120 pmap_pinit(pmap); 1121} 1122#endif 1123 1124/* 1125 * Initialize a preallocated and zeroed pmap structure, 1126 * such as one in a vmspace structure. 1127 */ 1128void 1129pmap_pinit(pmap) 1130 register struct pmap *pmap; 1131{ 1132 vm_page_t ptdpg; 1133 1134 /* 1135 * No need to allocate page table space yet but we do need a valid 1136 * page directory table. 1137 */ 1138 if (pmap->pm_pdir == NULL) 1139 pmap->pm_pdir = 1140 (pd_entry_t *)kmem_alloc_pageable(kernel_map, PAGE_SIZE); 1141 1142 /* 1143 * allocate object for the ptes 1144 */ 1145 if (pmap->pm_pteobj == NULL) 1146 pmap->pm_pteobj = vm_object_allocate( OBJT_DEFAULT, PTDPTDI + 1); 1147 1148 /* 1149 * allocate the page directory page 1150 */ 1151retry: 1152 ptdpg = vm_page_grab( pmap->pm_pteobj, PTDPTDI, 1153 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 1154 1155 ptdpg->wire_count = 1; 1156 ++cnt.v_wire_count; 1157 1158 ptdpg->flags &= ~(PG_MAPPED|PG_BUSY); /* not mapped normally */ 1159 ptdpg->valid = VM_PAGE_BITS_ALL; 1160 1161 pmap_kenter((vm_offset_t) pmap->pm_pdir, VM_PAGE_TO_PHYS(ptdpg)); 1162 if ((ptdpg->flags & PG_ZERO) == 0) 1163 bzero(pmap->pm_pdir, PAGE_SIZE); 1164 1165 /* wire in kernel global address entries */ 1166 /* XXX copies current process, does not fill in MPPTDI */ 1167 bcopy(PTD + KPTDI, pmap->pm_pdir + KPTDI, nkpt * PTESIZE); 1168 1169 /* install self-referential address mapping entry */ 1170 *(unsigned *) (pmap->pm_pdir + PTDPTDI) = 1171 VM_PAGE_TO_PHYS(ptdpg) | PG_V | PG_RW; 1172 1173 pmap->pm_flags = 0; 1174 pmap->pm_count = 1; 1175 pmap->pm_ptphint = NULL; 1176 TAILQ_INIT(&pmap->pm_pvlist); 1177 bzero(&pmap->pm_stats, sizeof pmap->pm_stats); 1178} 1179 1180static int 1181pmap_release_free_page(pmap, p) 1182 struct pmap *pmap; 1183 vm_page_t p; 1184{ 1185 int s; 1186 unsigned *pde = (unsigned *) pmap->pm_pdir; 1187 /* 1188 * This code optimizes the case of freeing non-busy 1189 * page-table pages. Those pages are zero now, and 1190 * might as well be placed directly into the zero queue. 1191 */ 1192 s = splvm(); 1193 if (p->flags & PG_BUSY) { 1194 p->flags |= PG_WANTED; 1195 tsleep(p, PVM, "pmaprl", 0); 1196 splx(s); 1197 return 0; 1198 } 1199 1200 p->flags |= PG_BUSY; 1201 splx(s); 1202 1203 /* 1204 * Remove the page table page from the processes address space. 1205 */ 1206 pde[p->pindex] = 0; 1207 pmap->pm_stats.resident_count--; 1208 1209#if !defined(MAX_PERF) 1210 if (p->hold_count) { 1211 panic("pmap_release: freeing held page table page"); 1212 } 1213#endif 1214 /* 1215 * Page directory pages need to have the kernel 1216 * stuff cleared, so they can go into the zero queue also. 1217 */ 1218 if (p->pindex == PTDPTDI) { 1219 bzero(pde + KPTDI, nkpt * PTESIZE); 1220#ifdef SMP 1221 pde[MPPTDI] = 0; 1222#endif 1223 pde[APTDPTDI] = 0; 1224 pmap_kremove((vm_offset_t) pmap->pm_pdir); 1225 } 1226 1227 if (pmap->pm_ptphint && (pmap->pm_ptphint->pindex == p->pindex)) 1228 pmap->pm_ptphint = NULL; 1229 1230 vm_page_free_zero(p); 1231 return 1; 1232} 1233 1234/* 1235 * this routine is called if the page table page is not 1236 * mapped correctly. 1237 */ 1238static vm_page_t 1239_pmap_allocpte(pmap, ptepindex) 1240 pmap_t pmap; 1241 unsigned ptepindex; 1242{ 1243 vm_offset_t pteva, ptepa; 1244 vm_page_t m; 1245 1246 /* 1247 * Find or fabricate a new pagetable page 1248 */ 1249 m = vm_page_grab(pmap->pm_pteobj, ptepindex, 1250 VM_ALLOC_ZERO | VM_ALLOC_RETRY); 1251 1252 if (m->queue != PQ_NONE) { 1253 int s = splvm(); 1254 vm_page_unqueue(m); 1255 splx(s); 1256 } 1257 1258 if (m->wire_count == 0) 1259 cnt.v_wire_count++; 1260 m->wire_count++; 1261 1262 /* 1263 * Increment the hold count for the page table page 1264 * (denoting a new mapping.) 1265 */ 1266 m->hold_count++; 1267 1268 /* 1269 * Map the pagetable page into the process address space, if 1270 * it isn't already there. 1271 */ 1272 1273 pmap->pm_stats.resident_count++; 1274 1275 ptepa = VM_PAGE_TO_PHYS(m); 1276 pmap->pm_pdir[ptepindex] = 1277 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_A); 1278 1279 /* 1280 * Set the page table hint 1281 */ 1282 pmap->pm_ptphint = m; 1283 1284 /* 1285 * Try to use the new mapping, but if we cannot, then 1286 * do it with the routine that maps the page explicitly. 1287 */ 1288 if ((m->flags & PG_ZERO) == 0) { 1289 if ((((unsigned)pmap->pm_pdir[PTDPTDI]) & PG_FRAME) == 1290 (((unsigned) PTDpde) & PG_FRAME)) { 1291 pteva = UPT_MIN_ADDRESS + i386_ptob(ptepindex); 1292 bzero((caddr_t) pteva, PAGE_SIZE); 1293 } else { 1294 pmap_zero_page(ptepa); 1295 } 1296 } 1297 1298 m->valid = VM_PAGE_BITS_ALL; 1299 m->flags &= ~(PG_ZERO|PG_BUSY); 1300 m->flags |= PG_MAPPED; 1301 1302 return m; 1303} 1304 1305static vm_page_t 1306pmap_allocpte(pmap, va) 1307 pmap_t pmap; 1308 vm_offset_t va; 1309{ 1310 unsigned ptepindex; 1311 vm_offset_t ptepa; 1312 vm_page_t m; 1313 1314 /* 1315 * Calculate pagetable page index 1316 */ 1317 ptepindex = va >> PDRSHIFT; 1318 1319 /* 1320 * Get the page directory entry 1321 */ 1322 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 1323 1324 /* 1325 * This supports switching from a 4MB page to a 1326 * normal 4K page. 1327 */ 1328 if (ptepa & PG_PS) { 1329 pmap->pm_pdir[ptepindex] = 0; 1330 ptepa = 0; 1331 invltlb(); 1332 } 1333 1334 /* 1335 * If the page table page is mapped, we just increment the 1336 * hold count, and activate it. 1337 */ 1338 if (ptepa) { 1339 /* 1340 * In order to get the page table page, try the 1341 * hint first. 1342 */ 1343 if (pmap->pm_ptphint && 1344 (pmap->pm_ptphint->pindex == ptepindex)) { 1345 m = pmap->pm_ptphint; 1346 } else { 1347 m = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 1348 pmap->pm_ptphint = m; 1349 } 1350 m->hold_count++; 1351 return m; 1352 } 1353 /* 1354 * Here if the pte page isn't mapped, or if it has been deallocated. 1355 */ 1356 return _pmap_allocpte(pmap, ptepindex); 1357} 1358 1359 1360/*************************************************** 1361* Pmap allocation/deallocation routines. 1362 ***************************************************/ 1363 1364/* 1365 * Release any resources held by the given physical map. 1366 * Called when a pmap initialized by pmap_pinit is being released. 1367 * Should only be called if the map contains no valid mappings. 1368 */ 1369void 1370pmap_release(pmap) 1371 register struct pmap *pmap; 1372{ 1373 vm_page_t p,n,ptdpg; 1374 vm_object_t object = pmap->pm_pteobj; 1375 int curgeneration; 1376 1377#if defined(DIAGNOSTIC) 1378 if (object->ref_count != 1) 1379 panic("pmap_release: pteobj reference count != 1"); 1380#endif 1381 1382 ptdpg = NULL; 1383retry: 1384 curgeneration = object->generation; 1385 for (p = TAILQ_FIRST(&object->memq); p != NULL; p = n) { 1386 n = TAILQ_NEXT(p, listq); 1387 if (p->pindex == PTDPTDI) { 1388 ptdpg = p; 1389 continue; 1390 } 1391 while (1) { 1392 if (!pmap_release_free_page(pmap, p) && 1393 (object->generation != curgeneration)) 1394 goto retry; 1395 } 1396 } 1397 1398 if (ptdpg && !pmap_release_free_page(pmap, ptdpg)) 1399 goto retry; 1400} 1401 1402/* 1403 * grow the number of kernel page table entries, if needed 1404 */ 1405void 1406pmap_growkernel(vm_offset_t addr) 1407{ 1408 struct proc *p; 1409 struct pmap *pmap; 1410 int s; 1411 vm_offset_t ptpkva, ptppaddr; 1412 vm_page_t nkpg; 1413#ifdef SMP 1414 int i; 1415#endif 1416 pd_entry_t newpdir; 1417 vm_pindex_t ptpidx; 1418 1419 s = splhigh(); 1420 if (kernel_vm_end == 0) { 1421 kernel_vm_end = KERNBASE; 1422 nkpt = 0; 1423 while (pdir_pde(PTD, kernel_vm_end)) { 1424 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1425 nkpt++; 1426 } 1427 } 1428 addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1429 while (kernel_vm_end < addr) { 1430 if (pdir_pde(PTD, kernel_vm_end)) { 1431 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1432 continue; 1433 } 1434 nkpt++; 1435 ptpkva = (vm_offset_t) vtopte(addr); 1436 ptpidx = (ptpkva >> PAGE_SHIFT); 1437 /* 1438 * This index is bogus, but out of the way 1439 */ 1440 nkpg = vm_page_alloc(kernel_object, ptpidx, VM_ALLOC_SYSTEM); 1441#if !defined(MAX_PERF) 1442 if (!nkpg) 1443 panic("pmap_growkernel: no memory to grow kernel"); 1444#endif 1445 1446 vm_page_wire(nkpg); 1447 vm_page_remove(nkpg); 1448 ptppaddr = VM_PAGE_TO_PHYS(nkpg); 1449 pmap_zero_page(ptppaddr); 1450 newpdir = (pd_entry_t) (ptppaddr | PG_V | PG_RW); 1451 pdir_pde(PTD, kernel_vm_end) = newpdir; 1452 1453#ifdef SMP 1454 for (i = 0; i < mp_ncpus; i++) { 1455 if (IdlePTDS[i]) 1456 pdir_pde(IdlePTDS[i], kernel_vm_end) = newpdir; 1457 } 1458#endif 1459 1460 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 1461 if (p->p_vmspace) { 1462 pmap = &p->p_vmspace->vm_pmap; 1463 *pmap_pde(pmap, kernel_vm_end) = newpdir; 1464 } 1465 } 1466 *pmap_pde(kernel_pmap, kernel_vm_end) = newpdir; 1467 kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); 1468 } 1469 splx(s); 1470} 1471 1472/* 1473 * Retire the given physical map from service. 1474 * Should only be called if the map contains 1475 * no valid mappings. 1476 */ 1477void 1478pmap_destroy(pmap) 1479 register pmap_t pmap; 1480{ 1481 int count; 1482 1483 if (pmap == NULL) 1484 return; 1485 1486 count = --pmap->pm_count; 1487 if (count == 0) { 1488 pmap_release(pmap); 1489#if !defined(MAX_PERF) 1490 panic("destroying a pmap is not yet implemented"); 1491#endif 1492 } 1493} 1494 1495/* 1496 * Add a reference to the specified pmap. 1497 */ 1498void 1499pmap_reference(pmap) 1500 pmap_t pmap; 1501{ 1502 if (pmap != NULL) { 1503 pmap->pm_count++; 1504 } 1505} 1506 1507/*************************************************** 1508* page management routines. 1509 ***************************************************/ 1510 1511/* 1512 * free the pv_entry back to the free list 1513 */ 1514static inline void 1515free_pv_entry(pv) 1516 pv_entry_t pv; 1517{ 1518 pv_entry_count--; 1519 zfreei(pvzone, pv); 1520} 1521 1522/* 1523 * get a new pv_entry, allocating a block from the system 1524 * when needed. 1525 * the memory allocation is performed bypassing the malloc code 1526 * because of the possibility of allocations at interrupt time. 1527 */ 1528static pv_entry_t 1529get_pv_entry(void) 1530{ 1531 pv_entry_count++; 1532 if (pv_entry_high_water && 1533 (pv_entry_count > pv_entry_high_water) && 1534 (pmap_pagedaemon_waken == 0)) { 1535 pmap_pagedaemon_waken = 1; 1536 wakeup (&vm_pages_needed); 1537 } 1538 return zalloci(pvzone); 1539} 1540 1541/* 1542 * This routine is very drastic, but can save the system 1543 * in a pinch. 1544 */ 1545void 1546pmap_collect() { 1547 pv_table_t *ppv; 1548 int i; 1549 vm_offset_t pa; 1550 vm_page_t m; 1551 static int warningdone=0; 1552 1553 if (pmap_pagedaemon_waken == 0) 1554 return; 1555 1556 if (warningdone < 5) { 1557 printf("pmap_collect: collecting pv entries -- suggest increasing PMAP_SHPGPERPROC\n"); 1558 warningdone++; 1559 } 1560 1561 for(i = 0; i < pv_npg; i++) { 1562 if ((ppv = &pv_table[i]) == 0) 1563 continue; 1564 m = ppv->pv_vm_page; 1565 if ((pa = VM_PAGE_TO_PHYS(m)) == 0) 1566 continue; 1567 if (m->wire_count || m->hold_count || m->busy || (m->flags & PG_BUSY)) 1568 continue; 1569 pmap_remove_all(pa); 1570 } 1571 pmap_pagedaemon_waken = 0; 1572} 1573 1574 1575/* 1576 * If it is the first entry on the list, it is actually 1577 * in the header and we must copy the following entry up 1578 * to the header. Otherwise we must search the list for 1579 * the entry. In either case we free the now unused entry. 1580 */ 1581 1582static int 1583pmap_remove_entry(pmap, ppv, va) 1584 struct pmap *pmap; 1585 pv_table_t *ppv; 1586 vm_offset_t va; 1587{ 1588 pv_entry_t pv; 1589 int rtval; 1590 int s; 1591 1592 s = splvm(); 1593 if (ppv->pv_list_count < pmap->pm_stats.resident_count) { 1594 for (pv = TAILQ_FIRST(&ppv->pv_list); 1595 pv; 1596 pv = TAILQ_NEXT(pv, pv_list)) { 1597 if (pmap == pv->pv_pmap && va == pv->pv_va) 1598 break; 1599 } 1600 } else { 1601 for (pv = TAILQ_FIRST(&pmap->pm_pvlist); 1602 pv; 1603 pv = TAILQ_NEXT(pv, pv_plist)) { 1604 if (va == pv->pv_va) 1605 break; 1606 } 1607 } 1608 1609 rtval = 0; 1610 if (pv) { 1611 1612 rtval = pmap_unuse_pt(pmap, va, pv->pv_ptem); 1613 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1614 ppv->pv_list_count--; 1615 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 1616 ppv->pv_vm_page->flags &= ~(PG_MAPPED | PG_WRITEABLE); 1617 1618 TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); 1619 free_pv_entry(pv); 1620 } 1621 1622 splx(s); 1623 return rtval; 1624} 1625 1626/* 1627 * Create a pv entry for page at pa for 1628 * (pmap, va). 1629 */ 1630static void 1631pmap_insert_entry(pmap, va, mpte, pa) 1632 pmap_t pmap; 1633 vm_offset_t va; 1634 vm_page_t mpte; 1635 vm_offset_t pa; 1636{ 1637 1638 int s; 1639 pv_entry_t pv; 1640 pv_table_t *ppv; 1641 1642 s = splvm(); 1643 pv = get_pv_entry(); 1644 pv->pv_va = va; 1645 pv->pv_pmap = pmap; 1646 pv->pv_ptem = mpte; 1647 1648 TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); 1649 1650 ppv = pa_to_pvh(pa); 1651 TAILQ_INSERT_TAIL(&ppv->pv_list, pv, pv_list); 1652 ppv->pv_list_count++; 1653 1654 splx(s); 1655} 1656 1657/* 1658 * pmap_remove_pte: do the things to unmap a page in a process 1659 */ 1660static int 1661pmap_remove_pte(pmap, ptq, va) 1662 struct pmap *pmap; 1663 unsigned *ptq; 1664 vm_offset_t va; 1665{ 1666 unsigned oldpte; 1667 pv_table_t *ppv; 1668 1669 oldpte = *ptq; 1670 *ptq = 0; 1671 if (oldpte & PG_W) 1672 pmap->pm_stats.wired_count -= 1; 1673 /* 1674 * Machines that don't support invlpg, also don't support 1675 * PG_G. 1676 */ 1677 if (oldpte & PG_G) 1678 invlpg(va); 1679 pmap->pm_stats.resident_count -= 1; 1680 if (oldpte & PG_MANAGED) { 1681 ppv = pa_to_pvh(oldpte); 1682 if (oldpte & PG_M) { 1683#if defined(PMAP_DIAGNOSTIC) 1684 if (pmap_nw_modified((pt_entry_t) oldpte)) { 1685 printf("pmap_remove: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, (int) oldpte); 1686 } 1687#endif 1688 if (pmap_track_modified(va)) 1689 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1690 } 1691 if (oldpte & PG_A) 1692 ppv->pv_vm_page->flags |= PG_REFERENCED; 1693 return pmap_remove_entry(pmap, ppv, va); 1694 } else { 1695 return pmap_unuse_pt(pmap, va, NULL); 1696 } 1697 1698 return 0; 1699} 1700 1701/* 1702 * Remove a single page from a process address space 1703 */ 1704static void 1705pmap_remove_page(pmap, va) 1706 struct pmap *pmap; 1707 register vm_offset_t va; 1708{ 1709 register unsigned *ptq; 1710 1711 /* 1712 * if there is no pte for this address, just skip it!!! 1713 */ 1714 if (*pmap_pde(pmap, va) == 0) { 1715 return; 1716 } 1717 1718 /* 1719 * get a local va for mappings for this pmap. 1720 */ 1721 ptq = get_ptbase(pmap) + i386_btop(va); 1722 if (*ptq) { 1723 (void) pmap_remove_pte(pmap, ptq, va); 1724 invltlb_1pg(va); 1725 } 1726 return; 1727} 1728 1729/* 1730 * Remove the given range of addresses from the specified map. 1731 * 1732 * It is assumed that the start and end are properly 1733 * rounded to the page size. 1734 */ 1735void 1736pmap_remove(pmap, sva, eva) 1737 struct pmap *pmap; 1738 register vm_offset_t sva; 1739 register vm_offset_t eva; 1740{ 1741 register unsigned *ptbase; 1742 vm_offset_t pdnxt; 1743 vm_offset_t ptpaddr; 1744 vm_offset_t sindex, eindex; 1745 int anyvalid; 1746 1747 if (pmap == NULL) 1748 return; 1749 1750 if (pmap->pm_stats.resident_count == 0) 1751 return; 1752 1753 /* 1754 * special handling of removing one page. a very 1755 * common operation and easy to short circuit some 1756 * code. 1757 */ 1758 if (((sva + PAGE_SIZE) == eva) && 1759 (((unsigned) pmap->pm_pdir[(sva >> PDRSHIFT)] & PG_PS) == 0)) { 1760 pmap_remove_page(pmap, sva); 1761 return; 1762 } 1763 1764 anyvalid = 0; 1765 1766 /* 1767 * Get a local virtual address for the mappings that are being 1768 * worked with. 1769 */ 1770 ptbase = get_ptbase(pmap); 1771 1772 sindex = i386_btop(sva); 1773 eindex = i386_btop(eva); 1774 1775 for (; sindex < eindex; sindex = pdnxt) { 1776 unsigned pdirindex; 1777 1778 /* 1779 * Calculate index for next page table. 1780 */ 1781 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1782 if (pmap->pm_stats.resident_count == 0) 1783 break; 1784 1785 pdirindex = sindex / NPDEPG; 1786 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1787 pmap->pm_pdir[pdirindex] = 0; 1788 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1789 anyvalid++; 1790 continue; 1791 } 1792 1793 /* 1794 * Weed out invalid mappings. Note: we assume that the page 1795 * directory table is always allocated, and in kernel virtual. 1796 */ 1797 if (ptpaddr == 0) 1798 continue; 1799 1800 /* 1801 * Limit our scan to either the end of the va represented 1802 * by the current page table page, or to the end of the 1803 * range being removed. 1804 */ 1805 if (pdnxt > eindex) { 1806 pdnxt = eindex; 1807 } 1808 1809 for ( ;sindex != pdnxt; sindex++) { 1810 vm_offset_t va; 1811 if (ptbase[sindex] == 0) { 1812 continue; 1813 } 1814 va = i386_ptob(sindex); 1815 1816 anyvalid++; 1817 if (pmap_remove_pte(pmap, 1818 ptbase + sindex, va)) 1819 break; 1820 } 1821 } 1822 1823 if (anyvalid) { 1824 invltlb(); 1825 } 1826} 1827 1828/* 1829 * Routine: pmap_remove_all 1830 * Function: 1831 * Removes this physical page from 1832 * all physical maps in which it resides. 1833 * Reflects back modify bits to the pager. 1834 * 1835 * Notes: 1836 * Original versions of this routine were very 1837 * inefficient because they iteratively called 1838 * pmap_remove (slow...) 1839 */ 1840 1841static void 1842pmap_remove_all(pa) 1843 vm_offset_t pa; 1844{ 1845 register pv_entry_t pv; 1846 pv_table_t *ppv; 1847 register unsigned *pte, tpte; 1848 int nmodify; 1849 int update_needed; 1850 int s; 1851 1852 nmodify = 0; 1853 update_needed = 0; 1854#if defined(PMAP_DIAGNOSTIC) 1855 /* 1856 * XXX this makes pmap_page_protect(NONE) illegal for non-managed 1857 * pages! 1858 */ 1859 if (!pmap_is_managed(pa)) { 1860 panic("pmap_page_protect: illegal for unmanaged page, va: 0x%lx", pa); 1861 } 1862#endif 1863 1864 s = splvm(); 1865 ppv = pa_to_pvh(pa); 1866 while ((pv = TAILQ_FIRST(&ppv->pv_list)) != NULL) { 1867 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 1868 1869 pv->pv_pmap->pm_stats.resident_count--; 1870 1871 tpte = *pte; 1872 *pte = 0; 1873 if (tpte & PG_W) 1874 pv->pv_pmap->pm_stats.wired_count--; 1875 1876 if (tpte & PG_A) 1877 ppv->pv_vm_page->flags |= PG_REFERENCED; 1878 1879 /* 1880 * Update the vm_page_t clean and reference bits. 1881 */ 1882 if (tpte & PG_M) { 1883#if defined(PMAP_DIAGNOSTIC) 1884 if (pmap_nw_modified((pt_entry_t) tpte)) { 1885 printf("pmap_remove_all: modified page not writable: va: 0x%lx, pte: 0x%lx\n", pv->pv_va, tpte); 1886 } 1887#endif 1888 if (pmap_track_modified(pv->pv_va)) 1889 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1890 } 1891 if (!update_needed && 1892 ((!curproc || (&curproc->p_vmspace->vm_pmap == pv->pv_pmap)) || 1893 (pv->pv_pmap == kernel_pmap))) { 1894 update_needed = 1; 1895 } 1896 1897 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 1898 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 1899 ppv->pv_list_count--; 1900 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 1901 free_pv_entry(pv); 1902 } 1903 1904 ppv->pv_vm_page->flags &= ~(PG_MAPPED | PG_WRITEABLE); 1905 1906 if (update_needed) 1907 invltlb(); 1908 1909 splx(s); 1910 return; 1911} 1912 1913/* 1914 * Set the physical protection on the 1915 * specified range of this map as requested. 1916 */ 1917void 1918pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) 1919{ 1920 register unsigned *ptbase; 1921 vm_offset_t pdnxt, ptpaddr; 1922 vm_pindex_t sindex, eindex; 1923 int anychanged; 1924 1925 1926 if (pmap == NULL) 1927 return; 1928 1929 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1930 pmap_remove(pmap, sva, eva); 1931 return; 1932 } 1933 1934 if (prot & VM_PROT_WRITE) 1935 return; 1936 1937 anychanged = 0; 1938 1939 ptbase = get_ptbase(pmap); 1940 1941 sindex = i386_btop(sva); 1942 eindex = i386_btop(eva); 1943 1944 for (; sindex < eindex; sindex = pdnxt) { 1945 1946 unsigned pdirindex; 1947 1948 pdnxt = ((sindex + NPTEPG) & ~(NPTEPG - 1)); 1949 1950 pdirindex = sindex / NPDEPG; 1951 if (((ptpaddr = (unsigned) pmap->pm_pdir[pdirindex]) & PG_PS) != 0) { 1952 (unsigned) pmap->pm_pdir[pdirindex] &= ~(PG_M|PG_RW); 1953 pmap->pm_stats.resident_count -= NBPDR / PAGE_SIZE; 1954 anychanged++; 1955 continue; 1956 } 1957 1958 /* 1959 * Weed out invalid mappings. Note: we assume that the page 1960 * directory table is always allocated, and in kernel virtual. 1961 */ 1962 if (ptpaddr == 0) 1963 continue; 1964 1965 if (pdnxt > eindex) { 1966 pdnxt = eindex; 1967 } 1968 1969 for (; sindex != pdnxt; sindex++) { 1970 1971 unsigned pbits; 1972 pv_table_t *ppv; 1973 1974 pbits = ptbase[sindex]; 1975 1976 if (pbits & PG_MANAGED) { 1977 ppv = NULL; 1978 if (pbits & PG_A) { 1979 ppv = pa_to_pvh(pbits); 1980 ppv->pv_vm_page->flags |= PG_REFERENCED; 1981 pbits &= ~PG_A; 1982 } 1983 if (pbits & PG_M) { 1984 if (pmap_track_modified(i386_ptob(sindex))) { 1985 if (ppv == NULL) 1986 ppv = pa_to_pvh(pbits); 1987 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 1988 pbits &= ~PG_M; 1989 } 1990 } 1991 } 1992 1993 pbits &= ~PG_RW; 1994 1995 if (pbits != ptbase[sindex]) { 1996 ptbase[sindex] = pbits; 1997 anychanged = 1; 1998 } 1999 } 2000 } 2001 if (anychanged) 2002 invltlb(); 2003} 2004 2005/* 2006 * Insert the given physical page (p) at 2007 * the specified virtual address (v) in the 2008 * target physical map with the protection requested. 2009 * 2010 * If specified, the page will be wired down, meaning 2011 * that the related pte can not be reclaimed. 2012 * 2013 * NB: This is the only routine which MAY NOT lazy-evaluate 2014 * or lose information. That is, this routine must actually 2015 * insert this page into the given map NOW. 2016 */ 2017void 2018pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot, 2019 boolean_t wired) 2020{ 2021 register unsigned *pte; 2022 vm_offset_t opa; 2023 vm_offset_t origpte, newpte; 2024 vm_page_t mpte; 2025 2026 if (pmap == NULL) 2027 return; 2028 2029 va &= PG_FRAME; 2030#ifdef PMAP_DIAGNOSTIC 2031 if (va > VM_MAX_KERNEL_ADDRESS) 2032 panic("pmap_enter: toobig"); 2033 if ((va >= UPT_MIN_ADDRESS) && (va < UPT_MAX_ADDRESS)) 2034 panic("pmap_enter: invalid to pmap_enter page table pages (va: 0x%x)", va); 2035#endif 2036 2037 mpte = NULL; 2038 /* 2039 * In the case that a page table page is not 2040 * resident, we are creating it here. 2041 */ 2042 if (va < UPT_MIN_ADDRESS) { 2043 mpte = pmap_allocpte(pmap, va); 2044 } 2045#if 0 && defined(PMAP_DIAGNOSTIC) 2046 else { 2047 vm_offset_t *pdeaddr = (vm_offset_t *)pmap_pde(pmap, va); 2048 if (((origpte = (vm_offset_t) *pdeaddr) & PG_V) == 0) { 2049 panic("pmap_enter: invalid kernel page table page(0), pdir=%p, pde=%p, va=%p\n", 2050 pmap->pm_pdir[PTDPTDI], origpte, va); 2051 } 2052 if (smp_active) { 2053 pdeaddr = (vm_offset_t *) IdlePTDS[cpuid]; 2054 if (((newpte = pdeaddr[va >> PDRSHIFT]) & PG_V) == 0) { 2055 if ((vm_offset_t) my_idlePTD != (vm_offset_t) vtophys(pdeaddr)) 2056 printf("pde mismatch: %x, %x\n", my_idlePTD, pdeaddr); 2057 printf("cpuid: %d, pdeaddr: 0x%x\n", cpuid, pdeaddr); 2058 panic("pmap_enter: invalid kernel page table page(1), pdir=%p, npde=%p, pde=%p, va=%p\n", 2059 pmap->pm_pdir[PTDPTDI], newpte, origpte, va); 2060 } 2061 } 2062 } 2063#endif 2064 2065 pte = pmap_pte(pmap, va); 2066 2067#if !defined(MAX_PERF) 2068 /* 2069 * Page Directory table entry not valid, we need a new PT page 2070 */ 2071 if (pte == NULL) { 2072 panic("pmap_enter: invalid page directory, pdir=%p, va=0x%lx\n", 2073 pmap->pm_pdir[PTDPTDI], va); 2074 } 2075#endif 2076 2077 origpte = *(vm_offset_t *)pte; 2078 pa &= PG_FRAME; 2079 opa = origpte & PG_FRAME; 2080 2081#if !defined(MAX_PERF) 2082 if (origpte & PG_PS) 2083 panic("pmap_enter: attempted pmap_enter on 4MB page"); 2084#endif 2085 2086 /* 2087 * Mapping has not changed, must be protection or wiring change. 2088 */ 2089 if (origpte && (opa == pa)) { 2090 /* 2091 * Wiring change, just update stats. We don't worry about 2092 * wiring PT pages as they remain resident as long as there 2093 * are valid mappings in them. Hence, if a user page is wired, 2094 * the PT page will be also. 2095 */ 2096 if (wired && ((origpte & PG_W) == 0)) 2097 pmap->pm_stats.wired_count++; 2098 else if (!wired && (origpte & PG_W)) 2099 pmap->pm_stats.wired_count--; 2100 2101#if defined(PMAP_DIAGNOSTIC) 2102 if (pmap_nw_modified((pt_entry_t) origpte)) { 2103 printf("pmap_enter: modified page not writable: va: 0x%lx, pte: 0x%lx\n", va, origpte); 2104 } 2105#endif 2106 2107 /* 2108 * Remove extra pte reference 2109 */ 2110 if (mpte) 2111 mpte->hold_count--; 2112 2113 if ((prot & VM_PROT_WRITE) && (origpte & PG_V)) { 2114 if ((origpte & PG_RW) == 0) { 2115 *pte |= PG_RW; 2116 invltlb_1pg(va); 2117 } 2118 return; 2119 } 2120 2121 /* 2122 * We might be turning off write access to the page, 2123 * so we go ahead and sense modify status. 2124 */ 2125 if (origpte & PG_MANAGED) { 2126 if ((origpte & PG_M) && pmap_track_modified(va)) { 2127 pv_table_t *ppv; 2128 ppv = pa_to_pvh(opa); 2129 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 2130 } 2131 pa |= PG_MANAGED; 2132 } 2133 goto validate; 2134 } 2135 /* 2136 * Mapping has changed, invalidate old range and fall through to 2137 * handle validating new mapping. 2138 */ 2139 if (opa) { 2140 int err; 2141 err = pmap_remove_pte(pmap, pte, va); 2142#if !defined(MAX_PERF) 2143 if (err) 2144 panic("pmap_enter: pte vanished, va: 0x%x", va); 2145#endif 2146 } 2147 2148 /* 2149 * Enter on the PV list if part of our managed memory Note that we 2150 * raise IPL while manipulating pv_table since pmap_enter can be 2151 * called at interrupt time. 2152 */ 2153 if (pmap_is_managed(pa)) { 2154 pmap_insert_entry(pmap, va, mpte, pa); 2155 pa |= PG_MANAGED; 2156 } 2157 2158 /* 2159 * Increment counters 2160 */ 2161 pmap->pm_stats.resident_count++; 2162 if (wired) 2163 pmap->pm_stats.wired_count++; 2164 2165validate: 2166 /* 2167 * Now validate mapping with desired protection/wiring. 2168 */ 2169 newpte = (vm_offset_t) (pa | pte_prot(pmap, prot) | PG_V); 2170 2171 if (wired) 2172 newpte |= PG_W; 2173 if (va < UPT_MIN_ADDRESS) 2174 newpte |= PG_U; 2175 if (pmap == kernel_pmap) 2176 newpte |= pgeflag; 2177 2178 /* 2179 * if the mapping or permission bits are different, we need 2180 * to update the pte. 2181 */ 2182 if ((origpte & ~(PG_M|PG_A)) != newpte) { 2183 *pte = newpte; 2184 if (origpte) 2185 invltlb_1pg(va); 2186 } 2187} 2188 2189/* 2190 * this code makes some *MAJOR* assumptions: 2191 * 1. Current pmap & pmap exists. 2192 * 2. Not wired. 2193 * 3. Read access. 2194 * 4. No page table pages. 2195 * 5. Tlbflush is deferred to calling procedure. 2196 * 6. Page IS managed. 2197 * but is *MUCH* faster than pmap_enter... 2198 */ 2199 2200static vm_page_t 2201pmap_enter_quick(pmap, va, pa, mpte) 2202 register pmap_t pmap; 2203 vm_offset_t va; 2204 register vm_offset_t pa; 2205 vm_page_t mpte; 2206{ 2207 register unsigned *pte; 2208 2209 /* 2210 * In the case that a page table page is not 2211 * resident, we are creating it here. 2212 */ 2213 if (va < UPT_MIN_ADDRESS) { 2214 unsigned ptepindex; 2215 vm_offset_t ptepa; 2216 2217 /* 2218 * Calculate pagetable page index 2219 */ 2220 ptepindex = va >> PDRSHIFT; 2221 if (mpte && (mpte->pindex == ptepindex)) { 2222 mpte->hold_count++; 2223 } else { 2224retry: 2225 /* 2226 * Get the page directory entry 2227 */ 2228 ptepa = (vm_offset_t) pmap->pm_pdir[ptepindex]; 2229 2230 /* 2231 * If the page table page is mapped, we just increment 2232 * the hold count, and activate it. 2233 */ 2234 if (ptepa) { 2235#if !defined(MAX_PERF) 2236 if (ptepa & PG_PS) 2237 panic("pmap_enter_quick: unexpected mapping into 4MB page"); 2238#endif 2239 if (pmap->pm_ptphint && 2240 (pmap->pm_ptphint->pindex == ptepindex)) { 2241 mpte = pmap->pm_ptphint; 2242 } else { 2243 mpte = pmap_page_lookup( pmap->pm_pteobj, ptepindex); 2244 pmap->pm_ptphint = mpte; 2245 } 2246 if (mpte == NULL) 2247 goto retry; 2248 mpte->hold_count++; 2249 } else { 2250 mpte = _pmap_allocpte(pmap, ptepindex); 2251 } 2252 } 2253 } else { 2254 mpte = NULL; 2255 } 2256 2257 /* 2258 * This call to vtopte makes the assumption that we are 2259 * entering the page into the current pmap. In order to support 2260 * quick entry into any pmap, one would likely use pmap_pte_quick. 2261 * But that isn't as quick as vtopte. 2262 */ 2263 pte = (unsigned *)vtopte(va); 2264 if (*pte) { 2265 if (mpte) 2266 pmap_unwire_pte_hold(pmap, mpte); 2267 return 0; 2268 } 2269 2270 /* 2271 * Enter on the PV list if part of our managed memory Note that we 2272 * raise IPL while manipulating pv_table since pmap_enter can be 2273 * called at interrupt time. 2274 */ 2275 pmap_insert_entry(pmap, va, mpte, pa); 2276 2277 /* 2278 * Increment counters 2279 */ 2280 pmap->pm_stats.resident_count++; 2281 2282 /* 2283 * Now validate mapping with RO protection 2284 */ 2285 *pte = pa | PG_V | PG_U | PG_MANAGED; 2286 2287 return mpte; 2288} 2289 2290#define MAX_INIT_PT (96) 2291/* 2292 * pmap_object_init_pt preloads the ptes for a given object 2293 * into the specified pmap. This eliminates the blast of soft 2294 * faults on process startup and immediately after an mmap. 2295 */ 2296void 2297pmap_object_init_pt(pmap, addr, object, pindex, size, limit) 2298 pmap_t pmap; 2299 vm_offset_t addr; 2300 vm_object_t object; 2301 vm_pindex_t pindex; 2302 vm_size_t size; 2303 int limit; 2304{ 2305 vm_offset_t tmpidx; 2306 int psize; 2307 vm_page_t p, mpte; 2308 int objpgs; 2309 2310 if (!pmap) 2311 return; 2312 2313 /* 2314 * This code maps large physical mmap regions into the 2315 * processor address space. Note that some shortcuts 2316 * are taken, but the code works. 2317 */ 2318 if (pseflag && 2319 (object->type == OBJT_DEVICE) && 2320 ((addr & (NBPDR - 1)) == 0) && 2321 ((size & (NBPDR - 1)) == 0) ) { 2322 int i; 2323 int s; 2324 vm_page_t m[1]; 2325 unsigned int ptepindex; 2326 int npdes; 2327 vm_offset_t ptepa; 2328 2329 if (pmap->pm_pdir[ptepindex = (addr >> PDRSHIFT)]) 2330 return; 2331 2332 s = splhigh(); 2333retry: 2334 p = vm_page_lookup(object, pindex); 2335 if (p && (p->flags & PG_BUSY)) { 2336 tsleep(p, PVM, "init4p", 0); 2337 goto retry; 2338 } 2339 splx(s); 2340 2341 if (p == NULL) { 2342 p = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL); 2343 if (p == NULL) 2344 return; 2345 m[0] = p; 2346 2347 if (vm_pager_get_pages(object, m, 1, 0) != VM_PAGER_OK) { 2348 vm_page_free(p); 2349 return; 2350 } 2351 2352 p = vm_page_lookup(object, pindex); 2353 PAGE_WAKEUP(p); 2354 } 2355 2356 ptepa = (vm_offset_t) VM_PAGE_TO_PHYS(p); 2357 if (ptepa & (NBPDR - 1)) { 2358 return; 2359 } 2360 2361 p->valid = VM_PAGE_BITS_ALL; 2362 2363 pmap->pm_stats.resident_count += size >> PAGE_SHIFT; 2364 npdes = size >> PDRSHIFT; 2365 for(i=0;i<npdes;i++) { 2366 pmap->pm_pdir[ptepindex] = 2367 (pd_entry_t) (ptepa | PG_U | PG_RW | PG_V | PG_PS); 2368 ptepa += NBPDR; 2369 ptepindex += 1; 2370 } 2371 p->flags |= PG_MAPPED; 2372 invltlb(); 2373 return; 2374 } 2375 2376 psize = i386_btop(size); 2377 2378 if ((object->type != OBJT_VNODE) || 2379 (limit && (psize > MAX_INIT_PT) && 2380 (object->resident_page_count > MAX_INIT_PT))) { 2381 return; 2382 } 2383 2384 if (psize + pindex > object->size) 2385 psize = object->size - pindex; 2386 2387 mpte = NULL; 2388 /* 2389 * if we are processing a major portion of the object, then scan the 2390 * entire thing. 2391 */ 2392 if (psize > (object->size >> 2)) { 2393 objpgs = psize; 2394 2395 for (p = TAILQ_FIRST(&object->memq); 2396 ((objpgs > 0) && (p != NULL)); 2397 p = TAILQ_NEXT(p, listq)) { 2398 2399 tmpidx = p->pindex; 2400 if (tmpidx < pindex) { 2401 continue; 2402 } 2403 tmpidx -= pindex; 2404 if (tmpidx >= psize) { 2405 continue; 2406 } 2407 if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2408 (p->busy == 0) && 2409 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2410 if ((p->queue - p->pc) == PQ_CACHE) 2411 vm_page_deactivate(p); 2412 p->flags |= PG_BUSY; 2413 mpte = pmap_enter_quick(pmap, 2414 addr + i386_ptob(tmpidx), 2415 VM_PAGE_TO_PHYS(p), mpte); 2416 p->flags |= PG_MAPPED; 2417 PAGE_WAKEUP(p); 2418 } 2419 objpgs -= 1; 2420 } 2421 } else { 2422 /* 2423 * else lookup the pages one-by-one. 2424 */ 2425 for (tmpidx = 0; tmpidx < psize; tmpidx += 1) { 2426 p = vm_page_lookup(object, tmpidx + pindex); 2427 if (p && 2428 ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2429 (p->busy == 0) && 2430 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2431 if ((p->queue - p->pc) == PQ_CACHE) 2432 vm_page_deactivate(p); 2433 p->flags |= PG_BUSY; 2434 mpte = pmap_enter_quick(pmap, 2435 addr + i386_ptob(tmpidx), 2436 VM_PAGE_TO_PHYS(p), mpte); 2437 p->flags |= PG_MAPPED; 2438 PAGE_WAKEUP(p); 2439 } 2440 } 2441 } 2442 return; 2443} 2444 2445/* 2446 * pmap_prefault provides a quick way of clustering 2447 * pagefaults into a processes address space. It is a "cousin" 2448 * of pmap_object_init_pt, except it runs at page fault time instead 2449 * of mmap time. 2450 */ 2451#define PFBAK 4 2452#define PFFOR 4 2453#define PAGEORDER_SIZE (PFBAK+PFFOR) 2454 2455static int pmap_prefault_pageorder[] = { 2456 -PAGE_SIZE, PAGE_SIZE, 2457 -2 * PAGE_SIZE, 2 * PAGE_SIZE, 2458 -3 * PAGE_SIZE, 3 * PAGE_SIZE 2459 -4 * PAGE_SIZE, 4 * PAGE_SIZE 2460}; 2461 2462void 2463pmap_prefault(pmap, addra, entry) 2464 pmap_t pmap; 2465 vm_offset_t addra; 2466 vm_map_entry_t entry; 2467{ 2468 int i; 2469 vm_offset_t starta; 2470 vm_offset_t addr; 2471 vm_pindex_t pindex; 2472 vm_page_t m, mpte; 2473 vm_object_t object; 2474 2475 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) 2476 return; 2477 2478 object = entry->object.vm_object; 2479 2480 starta = addra - PFBAK * PAGE_SIZE; 2481 if (starta < entry->start) { 2482 starta = entry->start; 2483 } else if (starta > addra) { 2484 starta = 0; 2485 } 2486 2487 mpte = NULL; 2488 for (i = 0; i < PAGEORDER_SIZE; i++) { 2489 vm_object_t lobject; 2490 unsigned *pte; 2491 2492 addr = addra + pmap_prefault_pageorder[i]; 2493 if (addr > addra + (PFFOR * PAGE_SIZE)) 2494 addr = 0; 2495 2496 if (addr < starta || addr >= entry->end) 2497 continue; 2498 2499 if ((*pmap_pde(pmap, addr)) == NULL) 2500 continue; 2501 2502 pte = (unsigned *) vtopte(addr); 2503 if (*pte) 2504 continue; 2505 2506 pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT; 2507 lobject = object; 2508 for (m = vm_page_lookup(lobject, pindex); 2509 (!m && (lobject->type == OBJT_DEFAULT) && (lobject->backing_object)); 2510 lobject = lobject->backing_object) { 2511 if (lobject->backing_object_offset & PAGE_MASK) 2512 break; 2513 pindex += (lobject->backing_object_offset >> PAGE_SHIFT); 2514 m = vm_page_lookup(lobject->backing_object, pindex); 2515 } 2516 2517 /* 2518 * give-up when a page is not in memory 2519 */ 2520 if (m == NULL) 2521 break; 2522 2523 if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) && 2524 (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 2525 2526 if ((m->queue - m->pc) == PQ_CACHE) { 2527 vm_page_deactivate(m); 2528 } 2529 m->flags |= PG_BUSY; 2530 mpte = pmap_enter_quick(pmap, addr, 2531 VM_PAGE_TO_PHYS(m), mpte); 2532 m->flags |= PG_MAPPED; 2533 PAGE_WAKEUP(m); 2534 } 2535 } 2536} 2537 2538/* 2539 * Routine: pmap_change_wiring 2540 * Function: Change the wiring attribute for a map/virtual-address 2541 * pair. 2542 * In/out conditions: 2543 * The mapping must already exist in the pmap. 2544 */ 2545void 2546pmap_change_wiring(pmap, va, wired) 2547 register pmap_t pmap; 2548 vm_offset_t va; 2549 boolean_t wired; 2550{ 2551 register unsigned *pte; 2552 2553 if (pmap == NULL) 2554 return; 2555 2556 pte = pmap_pte(pmap, va); 2557 2558 if (wired && !pmap_pte_w(pte)) 2559 pmap->pm_stats.wired_count++; 2560 else if (!wired && pmap_pte_w(pte)) 2561 pmap->pm_stats.wired_count--; 2562 2563 /* 2564 * Wiring is not a hardware characteristic so there is no need to 2565 * invalidate TLB. 2566 */ 2567 pmap_pte_set_w(pte, wired); 2568} 2569 2570 2571 2572/* 2573 * Copy the range specified by src_addr/len 2574 * from the source map to the range dst_addr/len 2575 * in the destination map. 2576 * 2577 * This routine is only advisory and need not do anything. 2578 */ 2579 2580void 2581pmap_copy(dst_pmap, src_pmap, dst_addr, len, src_addr) 2582 pmap_t dst_pmap, src_pmap; 2583 vm_offset_t dst_addr; 2584 vm_size_t len; 2585 vm_offset_t src_addr; 2586{ 2587 vm_offset_t addr; 2588 vm_offset_t end_addr = src_addr + len; 2589 vm_offset_t pdnxt; 2590 unsigned src_frame, dst_frame; 2591 2592 if (dst_addr != src_addr) 2593 return; 2594 2595 src_frame = ((unsigned) src_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2596 if (src_frame != (((unsigned) PTDpde) & PG_FRAME)) { 2597 return; 2598 } 2599 2600 dst_frame = ((unsigned) dst_pmap->pm_pdir[PTDPTDI]) & PG_FRAME; 2601 if (dst_frame != (((unsigned) APTDpde) & PG_FRAME)) { 2602 APTDpde = (pd_entry_t) (dst_frame | PG_RW | PG_V); 2603 invltlb(); 2604 } 2605 2606 for(addr = src_addr; addr < end_addr; addr = pdnxt) { 2607 unsigned *src_pte, *dst_pte; 2608 vm_page_t dstmpte, srcmpte; 2609 vm_offset_t srcptepaddr; 2610 unsigned ptepindex; 2611 2612#if !defined(MAX_PERF) 2613 if (addr >= UPT_MIN_ADDRESS) 2614 panic("pmap_copy: invalid to pmap_copy page tables\n"); 2615#endif 2616 2617 pdnxt = ((addr + PAGE_SIZE*NPTEPG) & ~(PAGE_SIZE*NPTEPG - 1)); 2618 ptepindex = addr >> PDRSHIFT; 2619 2620 srcptepaddr = (vm_offset_t) src_pmap->pm_pdir[ptepindex]; 2621 if (srcptepaddr == 0) 2622 continue; 2623 2624 if (srcptepaddr & PG_PS) { 2625 if (dst_pmap->pm_pdir[ptepindex] == 0) { 2626 dst_pmap->pm_pdir[ptepindex] = (pd_entry_t) srcptepaddr; 2627 dst_pmap->pm_stats.resident_count += NBPDR; 2628 } 2629 continue; 2630 } 2631 2632 srcmpte = vm_page_lookup(src_pmap->pm_pteobj, ptepindex); 2633 if ((srcmpte == NULL) || 2634 (srcmpte->hold_count == 0) || (srcmpte->flags & PG_BUSY)) 2635 continue; 2636 2637 if (pdnxt > end_addr) 2638 pdnxt = end_addr; 2639 2640 src_pte = (unsigned *) vtopte(addr); 2641 dst_pte = (unsigned *) avtopte(addr); 2642 while (addr < pdnxt) { 2643 unsigned ptetemp; 2644 ptetemp = *src_pte; 2645 /* 2646 * we only virtual copy managed pages 2647 */ 2648 if ((ptetemp & PG_MANAGED) != 0) { 2649 /* 2650 * We have to check after allocpte for the 2651 * pte still being around... allocpte can 2652 * block. 2653 */ 2654 dstmpte = pmap_allocpte(dst_pmap, addr); 2655 if ((*dst_pte == 0) && (ptetemp = *src_pte)) { 2656 /* 2657 * Clear the modified and 2658 * accessed (referenced) bits 2659 * during the copy. 2660 */ 2661 *dst_pte = ptetemp & ~(PG_M | PG_A); 2662 dst_pmap->pm_stats.resident_count++; 2663 pmap_insert_entry(dst_pmap, addr, 2664 dstmpte, 2665 (ptetemp & PG_FRAME)); 2666 } else { 2667 pmap_unwire_pte_hold(dst_pmap, dstmpte); 2668 } 2669 if (dstmpte->hold_count >= srcmpte->hold_count) 2670 break; 2671 } 2672 addr += PAGE_SIZE; 2673 src_pte++; 2674 dst_pte++; 2675 } 2676 } 2677} 2678 2679/* 2680 * Routine: pmap_kernel 2681 * Function: 2682 * Returns the physical map handle for the kernel. 2683 */ 2684pmap_t 2685pmap_kernel() 2686{ 2687 return (kernel_pmap); 2688} 2689 2690/* 2691 * pmap_zero_page zeros the specified (machine independent) 2692 * page by mapping the page into virtual memory and using 2693 * bzero to clear its contents, one machine dependent page 2694 * at a time. 2695 */ 2696void 2697pmap_zero_page(phys) 2698 vm_offset_t phys; 2699{ 2700#ifdef SMP 2701#if !defined(MAX_PERF) 2702 if (*(int *) prv_CMAP3) 2703 panic("pmap_zero_page: prv_CMAP3 busy"); 2704#endif 2705 2706 *(int *) prv_CMAP3 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2707 invltlb_1pg((vm_offset_t) &prv_CPAGE3); 2708 2709 bzero(&prv_CPAGE3, PAGE_SIZE); 2710 2711 *(int *) prv_CMAP3 = 0; 2712 invltlb_1pg((vm_offset_t) &prv_CPAGE3); 2713#else 2714#if !defined(MAX_PERF) 2715 if (*(int *) CMAP2) 2716 panic("pmap_zero_page: CMAP busy"); 2717#endif 2718 2719 *(int *) CMAP2 = PG_V | PG_RW | (phys & PG_FRAME) | PG_A | PG_M; 2720 bzero(CADDR2, PAGE_SIZE); 2721 *(int *) CMAP2 = 0; 2722 invltlb_1pg((vm_offset_t) CADDR2); 2723#endif 2724} 2725 2726/* 2727 * pmap_copy_page copies the specified (machine independent) 2728 * page by mapping the page into virtual memory and using 2729 * bcopy to copy the page, one machine dependent page at a 2730 * time. 2731 */ 2732void 2733pmap_copy_page(src, dst) 2734 vm_offset_t src; 2735 vm_offset_t dst; 2736{ 2737#ifdef SMP 2738#if !defined(MAX_PERF) 2739 if (*(int *) prv_CMAP1) 2740 panic("pmap_copy_page: prv_CMAP1 busy"); 2741 if (*(int *) prv_CMAP2) 2742 panic("pmap_copy_page: prv_CMAP2 busy"); 2743#endif 2744 2745 *(int *) prv_CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2746 *(int *) prv_CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2747 2748 invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2); 2749 2750 bcopy(&prv_CPAGE1, &prv_CPAGE2, PAGE_SIZE); 2751 2752 *(int *) prv_CMAP1 = 0; 2753 *(int *) prv_CMAP2 = 0; 2754 invltlb_2pg( (vm_offset_t) &prv_CPAGE1, (vm_offset_t) &prv_CPAGE2); 2755#else 2756#if !defined(MAX_PERF) 2757 if (*(int *) CMAP1 || *(int *) CMAP2) 2758 panic("pmap_copy_page: CMAP busy"); 2759#endif 2760 2761 *(int *) CMAP1 = PG_V | (src & PG_FRAME) | PG_A; 2762 *(int *) CMAP2 = PG_V | PG_RW | (dst & PG_FRAME) | PG_A | PG_M; 2763 2764 bcopy(CADDR1, CADDR2, PAGE_SIZE); 2765 2766 *(int *) CMAP1 = 0; 2767 *(int *) CMAP2 = 0; 2768 invltlb_2pg( (vm_offset_t) CADDR1, (vm_offset_t) CADDR2); 2769#endif 2770} 2771 2772 2773/* 2774 * Routine: pmap_pageable 2775 * Function: 2776 * Make the specified pages (by pmap, offset) 2777 * pageable (or not) as requested. 2778 * 2779 * A page which is not pageable may not take 2780 * a fault; therefore, its page table entry 2781 * must remain valid for the duration. 2782 * 2783 * This routine is merely advisory; pmap_enter 2784 * will specify that these pages are to be wired 2785 * down (or not) as appropriate. 2786 */ 2787void 2788pmap_pageable(pmap, sva, eva, pageable) 2789 pmap_t pmap; 2790 vm_offset_t sva, eva; 2791 boolean_t pageable; 2792{ 2793} 2794 2795/* 2796 * this routine returns true if a physical page resides 2797 * in the given pmap. 2798 */ 2799boolean_t 2800pmap_page_exists(pmap, pa) 2801 pmap_t pmap; 2802 vm_offset_t pa; 2803{ 2804 register pv_entry_t pv; 2805 pv_table_t *ppv; 2806 int s; 2807 2808 if (!pmap_is_managed(pa)) 2809 return FALSE; 2810 2811 s = splvm(); 2812 2813 ppv = pa_to_pvh(pa); 2814 /* 2815 * Not found, check current mappings returning immediately if found. 2816 */ 2817 for (pv = TAILQ_FIRST(&ppv->pv_list); 2818 pv; 2819 pv = TAILQ_NEXT(pv, pv_list)) { 2820 if (pv->pv_pmap == pmap) { 2821 splx(s); 2822 return TRUE; 2823 } 2824 } 2825 splx(s); 2826 return (FALSE); 2827} 2828 2829#define PMAP_REMOVE_PAGES_CURPROC_ONLY 2830/* 2831 * Remove all pages from specified address space 2832 * this aids process exit speeds. Also, this code 2833 * is special cased for current process only, but 2834 * can have the more generic (and slightly slower) 2835 * mode enabled. This is much faster than pmap_remove 2836 * in the case of running down an entire address space. 2837 */ 2838void 2839pmap_remove_pages(pmap, sva, eva) 2840 pmap_t pmap; 2841 vm_offset_t sva, eva; 2842{ 2843 unsigned *pte, tpte; 2844 pv_table_t *ppv; 2845 pv_entry_t pv, npv; 2846 int s; 2847 2848#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2849 if (!curproc || (pmap != &curproc->p_vmspace->vm_pmap)) { 2850 printf("warning: pmap_remove_pages called with non-current pmap\n"); 2851 return; 2852 } 2853#endif 2854 2855 s = splvm(); 2856 for(pv = TAILQ_FIRST(&pmap->pm_pvlist); 2857 pv; 2858 pv = npv) { 2859 2860 if (pv->pv_va >= eva || pv->pv_va < sva) { 2861 npv = TAILQ_NEXT(pv, pv_plist); 2862 continue; 2863 } 2864 2865#ifdef PMAP_REMOVE_PAGES_CURPROC_ONLY 2866 pte = (unsigned *)vtopte(pv->pv_va); 2867#else 2868 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2869#endif 2870 tpte = *pte; 2871 2872/* 2873 * We cannot remove wired pages from a process' mapping at this time 2874 */ 2875 if (tpte & PG_W) { 2876 npv = TAILQ_NEXT(pv, pv_plist); 2877 continue; 2878 } 2879 *pte = 0; 2880 2881 ppv = pa_to_pvh(tpte); 2882 2883 pv->pv_pmap->pm_stats.resident_count--; 2884 2885 /* 2886 * Update the vm_page_t clean and reference bits. 2887 */ 2888 if (tpte & PG_M) { 2889 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 2890 } 2891 2892 2893 npv = TAILQ_NEXT(pv, pv_plist); 2894 TAILQ_REMOVE(&pv->pv_pmap->pm_pvlist, pv, pv_plist); 2895 2896 ppv->pv_list_count--; 2897 TAILQ_REMOVE(&ppv->pv_list, pv, pv_list); 2898 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 2899 ppv->pv_vm_page->flags &= ~(PG_MAPPED | PG_WRITEABLE); 2900 } 2901 2902 pmap_unuse_pt(pv->pv_pmap, pv->pv_va, pv->pv_ptem); 2903 free_pv_entry(pv); 2904 } 2905 splx(s); 2906 invltlb(); 2907} 2908 2909/* 2910 * pmap_testbit tests bits in pte's 2911 * note that the testbit/changebit routines are inline, 2912 * and a lot of things compile-time evaluate. 2913 */ 2914static boolean_t 2915pmap_testbit(pa, bit) 2916 register vm_offset_t pa; 2917 int bit; 2918{ 2919 register pv_entry_t pv; 2920 pv_table_t *ppv; 2921 unsigned *pte; 2922 int s; 2923 2924 if (!pmap_is_managed(pa)) 2925 return FALSE; 2926 2927 ppv = pa_to_pvh(pa); 2928 if (TAILQ_FIRST(&ppv->pv_list) == NULL) 2929 return FALSE; 2930 2931 s = splvm(); 2932 2933 for (pv = TAILQ_FIRST(&ppv->pv_list); 2934 pv; 2935 pv = TAILQ_NEXT(pv, pv_list)) { 2936 2937 /* 2938 * if the bit being tested is the modified bit, then 2939 * mark clean_map and ptes as never 2940 * modified. 2941 */ 2942 if (bit & (PG_A|PG_M)) { 2943 if (!pmap_track_modified(pv->pv_va)) 2944 continue; 2945 } 2946 2947#if defined(PMAP_DIAGNOSTIC) 2948 if (!pv->pv_pmap) { 2949 printf("Null pmap (tb) at va: 0x%lx\n", pv->pv_va); 2950 continue; 2951 } 2952#endif 2953 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 2954 if (*pte & bit) { 2955 splx(s); 2956 return TRUE; 2957 } 2958 } 2959 splx(s); 2960 return (FALSE); 2961} 2962 2963/* 2964 * this routine is used to modify bits in ptes 2965 */ 2966static void 2967pmap_changebit(pa, bit, setem) 2968 vm_offset_t pa; 2969 int bit; 2970 boolean_t setem; 2971{ 2972 register pv_entry_t pv; 2973 pv_table_t *ppv; 2974 register unsigned *pte; 2975 int changed; 2976 int s; 2977 2978 if (!pmap_is_managed(pa)) 2979 return; 2980 2981 s = splvm(); 2982 changed = 0; 2983 ppv = pa_to_pvh(pa); 2984 2985 /* 2986 * Loop over all current mappings setting/clearing as appropos If 2987 * setting RO do we need to clear the VAC? 2988 */ 2989 for (pv = TAILQ_FIRST(&ppv->pv_list); 2990 pv; 2991 pv = TAILQ_NEXT(pv, pv_list)) { 2992 2993 /* 2994 * don't write protect pager mappings 2995 */ 2996 if (!setem && (bit == PG_RW)) { 2997 if (!pmap_track_modified(pv->pv_va)) 2998 continue; 2999 } 3000 3001#if defined(PMAP_DIAGNOSTIC) 3002 if (!pv->pv_pmap) { 3003 printf("Null pmap (cb) at va: 0x%lx\n", pv->pv_va); 3004 continue; 3005 } 3006#endif 3007 3008 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3009 3010 if (setem) { 3011 *(int *)pte |= bit; 3012 changed = 1; 3013 } else { 3014 vm_offset_t pbits = *(vm_offset_t *)pte; 3015 if (pbits & bit) { 3016 changed = 1; 3017 if (bit == PG_RW) { 3018 if (pbits & PG_M) { 3019 ppv->pv_vm_page->dirty = VM_PAGE_BITS_ALL; 3020 } 3021 *(int *)pte = pbits & ~(PG_M|PG_RW); 3022 } else { 3023 *(int *)pte = pbits & ~bit; 3024 } 3025 } 3026 } 3027 } 3028 splx(s); 3029 if (changed) 3030 invltlb(); 3031} 3032 3033/* 3034 * pmap_page_protect: 3035 * 3036 * Lower the permission for all mappings to a given page. 3037 */ 3038void 3039pmap_page_protect(vm_offset_t phys, vm_prot_t prot) 3040{ 3041 if ((prot & VM_PROT_WRITE) == 0) { 3042 if (prot & (VM_PROT_READ | VM_PROT_EXECUTE)) { 3043 pmap_changebit(phys, PG_RW, FALSE); 3044 } else { 3045 pmap_remove_all(phys); 3046 } 3047 } 3048} 3049 3050vm_offset_t 3051pmap_phys_address(ppn) 3052 int ppn; 3053{ 3054 return (i386_ptob(ppn)); 3055} 3056 3057/* 3058 * pmap_ts_referenced: 3059 * 3060 * Return the count of reference bits for a page, clearing all of them. 3061 * 3062 */ 3063int 3064pmap_ts_referenced(vm_offset_t pa) 3065{ 3066 register pv_entry_t pv; 3067 pv_table_t *ppv; 3068 unsigned *pte; 3069 int s; 3070 int rtval = 0; 3071 3072 if (!pmap_is_managed(pa)) 3073 return FALSE; 3074 3075 s = splvm(); 3076 3077 ppv = pa_to_pvh(pa); 3078 3079 if (TAILQ_FIRST(&ppv->pv_list) == NULL) { 3080 splx(s); 3081 return 0; 3082 } 3083 3084 /* 3085 * Not found, check current mappings returning immediately if found. 3086 */ 3087 for (pv = TAILQ_FIRST(&ppv->pv_list); 3088 pv; 3089 pv = TAILQ_NEXT(pv, pv_list)) { 3090 3091 /* 3092 * if the bit being tested is the modified bit, then 3093 * mark clean_map and ptes as never 3094 * modified. 3095 */ 3096 if (!pmap_track_modified(pv->pv_va)) 3097 continue; 3098 3099 pte = pmap_pte_quick(pv->pv_pmap, pv->pv_va); 3100 if (pte == NULL) { 3101 continue; 3102 } 3103 3104 if (*pte & PG_A) { 3105 rtval++; 3106 *pte &= ~PG_A; 3107 if (rtval > 16) 3108 break; 3109 } 3110 } 3111 3112 splx(s); 3113 if (rtval) { 3114 invltlb(); 3115 } 3116 return (rtval); 3117} 3118 3119/* 3120 * pmap_is_modified: 3121 * 3122 * Return whether or not the specified physical page was modified 3123 * in any physical maps. 3124 */ 3125boolean_t 3126pmap_is_modified(vm_offset_t pa) 3127{ 3128 return pmap_testbit((pa), PG_M); 3129} 3130 3131/* 3132 * Clear the modify bits on the specified physical page. 3133 */ 3134void 3135pmap_clear_modify(vm_offset_t pa) 3136{ 3137 pmap_changebit((pa), PG_M, FALSE); 3138} 3139 3140/* 3141 * pmap_clear_reference: 3142 * 3143 * Clear the reference bit on the specified physical page. 3144 */ 3145void 3146pmap_clear_reference(vm_offset_t pa) 3147{ 3148 pmap_changebit((pa), PG_A, FALSE); 3149} 3150 3151/* 3152 * Miscellaneous support routines follow 3153 */ 3154 3155static void 3156i386_protection_init() 3157{ 3158 register int *kp, prot; 3159 3160 kp = protection_codes; 3161 for (prot = 0; prot < 8; prot++) { 3162 switch (prot) { 3163 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_NONE: 3164 /* 3165 * Read access is also 0. There isn't any execute bit, 3166 * so just make it readable. 3167 */ 3168 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_NONE: 3169 case VM_PROT_READ | VM_PROT_NONE | VM_PROT_EXECUTE: 3170 case VM_PROT_NONE | VM_PROT_NONE | VM_PROT_EXECUTE: 3171 *kp++ = 0; 3172 break; 3173 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_NONE: 3174 case VM_PROT_NONE | VM_PROT_WRITE | VM_PROT_EXECUTE: 3175 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_NONE: 3176 case VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE: 3177 *kp++ = PG_RW; 3178 break; 3179 } 3180 } 3181} 3182 3183/* 3184 * Map a set of physical memory pages into the kernel virtual 3185 * address space. Return a pointer to where it is mapped. This 3186 * routine is intended to be used for mapping device memory, 3187 * NOT real memory. 3188 */ 3189void * 3190pmap_mapdev(pa, size) 3191 vm_offset_t pa; 3192 vm_size_t size; 3193{ 3194 vm_offset_t va, tmpva; 3195 unsigned *pte; 3196 3197 size = roundup(size, PAGE_SIZE); 3198 3199 va = kmem_alloc_pageable(kernel_map, size); 3200#if !defined(MAX_PERF) 3201 if (!va) 3202 panic("pmap_mapdev: Couldn't alloc kernel virtual memory"); 3203#endif 3204 3205 pa = pa & PG_FRAME; 3206 for (tmpva = va; size > 0;) { 3207 pte = (unsigned *)vtopte(tmpva); 3208 *pte = pa | PG_RW | PG_V | pgeflag; 3209 size -= PAGE_SIZE; 3210 tmpva += PAGE_SIZE; 3211 pa += PAGE_SIZE; 3212 } 3213 invltlb(); 3214 3215 return ((void *) va); 3216} 3217 3218/* 3219 * perform the pmap work for mincore 3220 */ 3221int 3222pmap_mincore(pmap, addr) 3223 pmap_t pmap; 3224 vm_offset_t addr; 3225{ 3226 3227 unsigned *ptep, pte; 3228 vm_page_t m; 3229 int val = 0; 3230 3231 ptep = pmap_pte(pmap, addr); 3232 if (ptep == 0) { 3233 return 0; 3234 } 3235 3236 if (pte = *ptep) { 3237 pv_table_t *ppv; 3238 vm_offset_t pa; 3239 3240 val = MINCORE_INCORE; 3241 if ((pte & PG_MANAGED) == 0) 3242 return val; 3243 3244 pa = pte & PG_FRAME; 3245 3246 ppv = pa_to_pvh((pa & PG_FRAME)); 3247 m = ppv->pv_vm_page; 3248 3249 /* 3250 * Modified by us 3251 */ 3252 if (pte & PG_M) 3253 val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; 3254 /* 3255 * Modified by someone 3256 */ 3257 else if (m->dirty || pmap_is_modified(pa)) 3258 val |= MINCORE_MODIFIED_OTHER; 3259 /* 3260 * Referenced by us 3261 */ 3262 if (pte & PG_A) 3263 val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; 3264 3265 /* 3266 * Referenced by someone 3267 */ 3268 else if ((m->flags & PG_REFERENCED) || pmap_ts_referenced(pa)) { 3269 val |= MINCORE_REFERENCED_OTHER; 3270 m->flags |= PG_REFERENCED; 3271 } 3272 } 3273 return val; 3274} 3275 3276void 3277pmap_activate(struct proc *p) 3278{ 3279#if defined(SWTCH_OPTIM_STATS) 3280 tlb_flush_count++; 3281#endif 3282 load_cr3(p->p_addr->u_pcb.pcb_cr3 = 3283 vtophys(p->p_vmspace->vm_pmap.pm_pdir)); 3284} 3285 3286vm_offset_t 3287pmap_addr_hint(vm_object_t obj, vm_offset_t addr, vm_size_t size) { 3288 3289 if ((obj == NULL) || (size < NBPDR) || (obj->type != OBJT_DEVICE)) { 3290 return addr; 3291 } 3292 3293 addr = (addr + (NBPDR - 1)) & ~(NBPDR - 1); 3294 return addr; 3295} 3296 3297 3298#if defined(PMAP_DEBUG) 3299pmap_pid_dump(int pid) { 3300 pmap_t pmap; 3301 struct proc *p; 3302 int npte = 0; 3303 int index; 3304 for (p = allproc.lh_first; p != NULL; p = p->p_list.le_next) { 3305 if (p->p_pid != pid) 3306 continue; 3307 3308 if (p->p_vmspace) { 3309 int i,j; 3310 index = 0; 3311 pmap = &p->p_vmspace->vm_pmap; 3312 for(i=0;i<1024;i++) { 3313 pd_entry_t *pde; 3314 unsigned *pte; 3315 unsigned base = i << PDRSHIFT; 3316 3317 pde = &pmap->pm_pdir[i]; 3318 if (pde && pmap_pde_v(pde)) { 3319 for(j=0;j<1024;j++) { 3320 unsigned va = base + (j << PAGE_SHIFT); 3321 if (va >= (vm_offset_t) VM_MIN_KERNEL_ADDRESS) { 3322 if (index) { 3323 index = 0; 3324 printf("\n"); 3325 } 3326 return npte; 3327 } 3328 pte = pmap_pte_quick( pmap, va); 3329 if (pte && pmap_pte_v(pte)) { 3330 vm_offset_t pa; 3331 vm_page_t m; 3332 pa = *(int *)pte; 3333 m = PHYS_TO_VM_PAGE((pa & PG_FRAME)); 3334 printf("va: 0x%x, pt: 0x%x, h: %d, w: %d, f: 0x%x", 3335 va, pa, m->hold_count, m->wire_count, m->flags); 3336 npte++; 3337 index++; 3338 if (index >= 2) { 3339 index = 0; 3340 printf("\n"); 3341 } else { 3342 printf(" "); 3343 } 3344 } 3345 } 3346 } 3347 } 3348 } 3349 } 3350 return npte; 3351} 3352#endif 3353 3354#if defined(DEBUG) 3355 3356static void pads __P((pmap_t pm)); 3357static void pmap_pvdump __P((vm_offset_t pa)); 3358 3359/* print address space of pmap*/ 3360static void 3361pads(pm) 3362 pmap_t pm; 3363{ 3364 unsigned va, i, j; 3365 unsigned *ptep; 3366 3367 if (pm == kernel_pmap) 3368 return; 3369 for (i = 0; i < 1024; i++) 3370 if (pm->pm_pdir[i]) 3371 for (j = 0; j < 1024; j++) { 3372 va = (i << PDRSHIFT) + (j << PAGE_SHIFT); 3373 if (pm == kernel_pmap && va < KERNBASE) 3374 continue; 3375 if (pm != kernel_pmap && va > UPT_MAX_ADDRESS) 3376 continue; 3377 ptep = pmap_pte_quick(pm, va); 3378 if (pmap_pte_v(ptep)) 3379 printf("%x:%x ", va, *(int *) ptep); 3380 }; 3381 3382} 3383 3384static void 3385pmap_pvdump(pa) 3386 vm_offset_t pa; 3387{ 3388 pv_table_t *ppv; 3389 register pv_entry_t pv; 3390 3391 printf("pa %x", pa); 3392 ppv = pa_to_pvh(pa); 3393 for (pv = TAILQ_FIRST(&ppv->pv_list); 3394 pv; 3395 pv = TAILQ_NEXT(pv, pv_list)) { 3396#ifdef used_to_be 3397 printf(" -> pmap %x, va %x, flags %x", 3398 pv->pv_pmap, pv->pv_va, pv->pv_flags); 3399#endif 3400 printf(" -> pmap %x, va %x", 3401 pv->pv_pmap, pv->pv_va); 3402 pads(pv->pv_pmap); 3403 } 3404 printf(" "); 3405} 3406#endif 3407