vm_machdep.c revision 31322
1/*- 2 * Copyright (c) 1982, 1986 The Regents of the University of California. 3 * Copyright (c) 1989, 1990 William Jolitz 4 * Copyright (c) 1994 John Dyson 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * the Systems Programming Group of the University of Utah Computer 9 * Science Department, and William Jolitz. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 41 * $Id: vm_machdep.c,v 1.91 1997/11/18 09:27:23 bde Exp $ 42 */ 43 44#include "npx.h" 45#include "opt_bounce.h" 46#include "opt_vm86.h" 47 48#include <sys/param.h> 49#include <sys/systm.h> 50#include <sys/proc.h> 51#include <sys/malloc.h> 52#include <sys/buf.h> 53#include <sys/vnode.h> 54#include <sys/vmmeter.h> 55 56#include <machine/clock.h> 57#include <machine/cpu.h> 58#include <machine/md_var.h> 59#ifdef SMP 60#include <machine/smp.h> 61#endif 62 63#include <vm/vm.h> 64#include <vm/vm_param.h> 65#include <vm/vm_prot.h> 66#include <sys/lock.h> 67#include <vm/vm_kern.h> 68#include <vm/vm_page.h> 69#include <vm/vm_map.h> 70#include <vm/vm_extern.h> 71 72#include <sys/user.h> 73 74#ifdef PC98 75#include <pc98/pc98/pc98.h> 76#else 77#include <i386/isa/isa.h> 78#endif 79 80#ifdef BOUNCE_BUFFERS 81static vm_offset_t 82 vm_bounce_kva __P((int size, int waitok)); 83static void vm_bounce_kva_free __P((vm_offset_t addr, vm_offset_t size, 84 int now)); 85static vm_offset_t 86 vm_bounce_page_find __P((int count)); 87static void vm_bounce_page_free __P((vm_offset_t pa, int count)); 88 89static volatile int kvasfreecnt; 90 91caddr_t bouncememory; 92static int bpwait; 93static vm_offset_t *bouncepa; 94static int bmwait, bmfreeing; 95 96#define BITS_IN_UNSIGNED (8*sizeof(unsigned)) 97static int bounceallocarraysize; 98static unsigned *bounceallocarray; 99static int bouncefree; 100 101#if defined(PC98) && defined (EPSON_BOUNCEDMA) 102#define SIXTEENMEG (3840*4096) /* 15MB boundary */ 103#else 104#define SIXTEENMEG (4096*4096) 105#endif 106#define MAXBKVA 1024 107int maxbkva = MAXBKVA*PAGE_SIZE; 108 109/* special list that can be used at interrupt time for eventual kva free */ 110static struct kvasfree { 111 vm_offset_t addr; 112 vm_offset_t size; 113} kvaf[MAXBKVA]; 114 115/* 116 * get bounce buffer pages (count physically contiguous) 117 * (only 1 inplemented now) 118 */ 119static vm_offset_t 120vm_bounce_page_find(count) 121 int count; 122{ 123 int bit; 124 int s,i; 125 126 if (count != 1) 127 panic("vm_bounce_page_find -- no support for > 1 page yet!!!"); 128 129 s = splbio(); 130retry: 131 for (i = 0; i < bounceallocarraysize; i++) { 132 if (bounceallocarray[i] != 0xffffffff) { 133 bit = ffs(~bounceallocarray[i]); 134 if (bit) { 135 bounceallocarray[i] |= 1 << (bit - 1) ; 136 bouncefree -= count; 137 splx(s); 138 return bouncepa[(i * BITS_IN_UNSIGNED + (bit - 1))]; 139 } 140 } 141 } 142 bpwait = 1; 143 tsleep((caddr_t) &bounceallocarray, PRIBIO, "bncwai", 0); 144 goto retry; 145} 146 147static void 148vm_bounce_kva_free(addr, size, now) 149 vm_offset_t addr; 150 vm_offset_t size; 151 int now; 152{ 153 int s = splbio(); 154 kvaf[kvasfreecnt].addr = addr; 155 kvaf[kvasfreecnt].size = size; 156 ++kvasfreecnt; 157 if( now) { 158 /* 159 * this will do wakeups 160 */ 161 vm_bounce_kva(0,0); 162 } else { 163 if (bmwait) { 164 /* 165 * if anyone is waiting on the bounce-map, then wakeup 166 */ 167 wakeup((caddr_t) io_map); 168 bmwait = 0; 169 } 170 } 171 splx(s); 172} 173 174/* 175 * free count bounce buffer pages 176 */ 177static void 178vm_bounce_page_free(pa, count) 179 vm_offset_t pa; 180 int count; 181{ 182 int allocindex; 183 int index; 184 int bit; 185 186 if (count != 1) 187 panic("vm_bounce_page_free -- no support for > 1 page yet!!!"); 188 189 for(index=0;index<bouncepages;index++) { 190 if( pa == bouncepa[index]) 191 break; 192 } 193 194 if( index == bouncepages) 195 panic("vm_bounce_page_free: invalid bounce buffer"); 196 197 allocindex = index / BITS_IN_UNSIGNED; 198 bit = index % BITS_IN_UNSIGNED; 199 200 bounceallocarray[allocindex] &= ~(1 << bit); 201 202 bouncefree += count; 203 if (bpwait) { 204 bpwait = 0; 205 wakeup((caddr_t) &bounceallocarray); 206 } 207} 208 209/* 210 * allocate count bounce buffer kva pages 211 */ 212static vm_offset_t 213vm_bounce_kva(size, waitok) 214 int size; 215 int waitok; 216{ 217 int i; 218 vm_offset_t kva = 0; 219 vm_offset_t off; 220 int s = splbio(); 221more: 222 if (!bmfreeing && kvasfreecnt) { 223 bmfreeing = 1; 224 for (i = 0; i < kvasfreecnt; i++) { 225 for(off=0;off<kvaf[i].size;off+=PAGE_SIZE) { 226 pmap_kremove( kvaf[i].addr + off); 227 } 228 kmem_free_wakeup(io_map, kvaf[i].addr, 229 kvaf[i].size); 230 } 231 kvasfreecnt = 0; 232 bmfreeing = 0; 233 if( bmwait) { 234 bmwait = 0; 235 wakeup( (caddr_t) io_map); 236 } 237 } 238 239 if( size == 0) { 240 splx(s); 241 return 0; 242 } 243 244 if ((kva = kmem_alloc_pageable(io_map, size)) == 0) { 245 if( !waitok) { 246 splx(s); 247 return 0; 248 } 249 bmwait = 1; 250 tsleep((caddr_t) io_map, PRIBIO, "bmwait", 0); 251 goto more; 252 } 253 splx(s); 254 return kva; 255} 256 257/* 258 * same as vm_bounce_kva -- but really allocate (but takes pages as arg) 259 */ 260vm_offset_t 261vm_bounce_kva_alloc(count) 262int count; 263{ 264 int i; 265 vm_offset_t kva; 266 vm_offset_t pa; 267 if( bouncepages == 0) { 268 kva = (vm_offset_t) malloc(count*PAGE_SIZE, M_TEMP, M_WAITOK); 269 return kva; 270 } 271 kva = vm_bounce_kva(count*PAGE_SIZE, 1); 272 for(i=0;i<count;i++) { 273 pa = vm_bounce_page_find(1); 274 pmap_kenter(kva + i * PAGE_SIZE, pa); 275 } 276 return kva; 277} 278 279/* 280 * same as vm_bounce_kva_free -- but really free 281 */ 282void 283vm_bounce_kva_alloc_free(kva, count) 284 vm_offset_t kva; 285 int count; 286{ 287 int i; 288 vm_offset_t pa; 289 if( bouncepages == 0) { 290 free((caddr_t) kva, M_TEMP); 291 return; 292 } 293 for(i = 0; i < count; i++) { 294 pa = pmap_kextract(kva + i * PAGE_SIZE); 295 vm_bounce_page_free(pa, 1); 296 } 297 vm_bounce_kva_free(kva, count*PAGE_SIZE, 0); 298} 299 300/* 301 * do the things necessary to the struct buf to implement 302 * bounce buffers... inserted before the disk sort 303 */ 304void 305vm_bounce_alloc(bp) 306 struct buf *bp; 307{ 308 int countvmpg; 309 vm_offset_t vastart, vaend; 310 vm_offset_t vapstart, vapend; 311 vm_offset_t va, kva; 312 vm_offset_t pa; 313 int dobounceflag = 0; 314 int i; 315 316 if (bouncepages == 0) 317 return; 318 319 if (bp->b_flags & B_BOUNCE) { 320 printf("vm_bounce_alloc: called recursively???\n"); 321 return; 322 } 323 324 if (bp->b_bufsize < bp->b_bcount) { 325 printf( 326 "vm_bounce_alloc: b_bufsize(0x%lx) < b_bcount(0x%lx) !!\n", 327 bp->b_bufsize, bp->b_bcount); 328 panic("vm_bounce_alloc"); 329 } 330 331/* 332 * This is not really necessary 333 * if( bp->b_bufsize != bp->b_bcount) { 334 * printf("size: %d, count: %d\n", bp->b_bufsize, bp->b_bcount); 335 * } 336 */ 337 338 339 vastart = (vm_offset_t) bp->b_data; 340 vaend = (vm_offset_t) bp->b_data + bp->b_bufsize; 341 342 vapstart = trunc_page(vastart); 343 vapend = round_page(vaend); 344 countvmpg = (vapend - vapstart) / PAGE_SIZE; 345 346/* 347 * if any page is above 16MB, then go into bounce-buffer mode 348 */ 349 va = vapstart; 350 for (i = 0; i < countvmpg; i++) { 351 pa = pmap_kextract(va); 352 if (pa >= SIXTEENMEG) 353 ++dobounceflag; 354 if( pa == 0) 355 panic("vm_bounce_alloc: Unmapped page"); 356 va += PAGE_SIZE; 357 } 358 if (dobounceflag == 0) 359 return; 360 361 if (bouncepages < dobounceflag) 362 panic("Not enough bounce buffers!!!"); 363 364/* 365 * allocate a replacement kva for b_addr 366 */ 367 kva = vm_bounce_kva(countvmpg*PAGE_SIZE, 1); 368#if 0 369 printf("%s: vapstart: %x, vapend: %x, countvmpg: %d, kva: %x ", 370 (bp->b_flags & B_READ) ? "read":"write", 371 vapstart, vapend, countvmpg, kva); 372#endif 373 va = vapstart; 374 for (i = 0; i < countvmpg; i++) { 375 pa = pmap_kextract(va); 376 if (pa >= SIXTEENMEG) { 377 /* 378 * allocate a replacement page 379 */ 380 vm_offset_t bpa = vm_bounce_page_find(1); 381 pmap_kenter(kva + (PAGE_SIZE * i), bpa); 382#if 0 383 printf("r(%d): (%x,%x,%x) ", i, va, pa, bpa); 384#endif 385 /* 386 * if we are writing, the copy the data into the page 387 */ 388 if ((bp->b_flags & B_READ) == 0) { 389 bcopy((caddr_t) va, (caddr_t) kva + (PAGE_SIZE * i), PAGE_SIZE); 390 } 391 } else { 392 /* 393 * use original page 394 */ 395 pmap_kenter(kva + (PAGE_SIZE * i), pa); 396 } 397 va += PAGE_SIZE; 398 } 399 400/* 401 * flag the buffer as being bounced 402 */ 403 bp->b_flags |= B_BOUNCE; 404/* 405 * save the original buffer kva 406 */ 407 bp->b_savekva = bp->b_data; 408/* 409 * put our new kva into the buffer (offset by original offset) 410 */ 411 bp->b_data = (caddr_t) (((vm_offset_t) kva) | 412 ((vm_offset_t) bp->b_savekva & PAGE_MASK)); 413#if 0 414 printf("b_savekva: %x, newva: %x\n", bp->b_savekva, bp->b_data); 415#endif 416 return; 417} 418 419/* 420 * hook into biodone to free bounce buffer 421 */ 422void 423vm_bounce_free(bp) 424 struct buf *bp; 425{ 426 int i; 427 vm_offset_t origkva, bouncekva, bouncekvaend; 428 429/* 430 * if this isn't a bounced buffer, then just return 431 */ 432 if ((bp->b_flags & B_BOUNCE) == 0) 433 return; 434 435/* 436 * This check is not necessary 437 * if (bp->b_bufsize != bp->b_bcount) { 438 * printf("vm_bounce_free: b_bufsize=%d, b_bcount=%d\n", 439 * bp->b_bufsize, bp->b_bcount); 440 * } 441 */ 442 443 origkva = (vm_offset_t) bp->b_savekva; 444 bouncekva = (vm_offset_t) bp->b_data; 445/* 446 printf("free: %d ", bp->b_bufsize); 447*/ 448 449/* 450 * check every page in the kva space for b_addr 451 */ 452 for (i = 0; i < bp->b_bufsize; ) { 453 vm_offset_t mybouncepa; 454 vm_offset_t copycount; 455 456 copycount = round_page(bouncekva + 1) - bouncekva; 457 mybouncepa = pmap_kextract(trunc_page(bouncekva)); 458 459/* 460 * if this is a bounced pa, then process as one 461 */ 462 if ( mybouncepa != pmap_kextract( trunc_page( origkva))) { 463 vm_offset_t tocopy = copycount; 464 if (i + tocopy > bp->b_bufsize) 465 tocopy = bp->b_bufsize - i; 466/* 467 * if this is a read, then copy from bounce buffer into original buffer 468 */ 469 if (bp->b_flags & B_READ) 470 bcopy((caddr_t) bouncekva, (caddr_t) origkva, tocopy); 471/* 472 * free the bounce allocation 473 */ 474 475/* 476 printf("(kva: %x, pa: %x)", bouncekva, mybouncepa); 477*/ 478 vm_bounce_page_free(mybouncepa, 1); 479 } 480 481 origkva += copycount; 482 bouncekva += copycount; 483 i += copycount; 484 } 485 486/* 487 printf("\n"); 488*/ 489/* 490 * add the old kva into the "to free" list 491 */ 492 493 bouncekva= trunc_page((vm_offset_t) bp->b_data); 494 bouncekvaend= round_page((vm_offset_t)bp->b_data + bp->b_bufsize); 495 496/* 497 printf("freeva: %d\n", (bouncekvaend - bouncekva) / PAGE_SIZE); 498*/ 499 vm_bounce_kva_free( bouncekva, (bouncekvaend - bouncekva), 0); 500 bp->b_data = bp->b_savekva; 501 bp->b_savekva = 0; 502 bp->b_flags &= ~B_BOUNCE; 503 504 return; 505} 506 507 508/* 509 * init the bounce buffer system 510 */ 511void 512vm_bounce_init() 513{ 514 int i; 515 516 kvasfreecnt = 0; 517 518 if (bouncepages == 0) 519 return; 520 521 bounceallocarraysize = (bouncepages + BITS_IN_UNSIGNED - 1) / BITS_IN_UNSIGNED; 522 bounceallocarray = malloc(bounceallocarraysize * sizeof(unsigned), M_TEMP, M_NOWAIT); 523 524 if (!bounceallocarray) 525 panic("Cannot allocate bounce resource array"); 526 527 bouncepa = malloc(bouncepages * sizeof(vm_offset_t), M_TEMP, M_NOWAIT); 528 if (!bouncepa) 529 panic("Cannot allocate physical memory array"); 530 531 for(i=0;i<bounceallocarraysize;i++) { 532 bounceallocarray[i] = 0xffffffff; 533 } 534 535 for(i=0;i<bouncepages;i++) { 536 vm_offset_t pa; 537 if( (pa = pmap_kextract((vm_offset_t) bouncememory + i * PAGE_SIZE)) >= SIXTEENMEG) 538 panic("bounce memory out of range"); 539 if( pa == 0) 540 panic("bounce memory not resident"); 541 bouncepa[i] = pa; 542 bounceallocarray[i/(8*sizeof(int))] &= ~(1<<(i%(8*sizeof(int)))); 543 } 544 bouncefree = bouncepages; 545 546} 547#endif /* BOUNCE_BUFFERS */ 548 549/* 550 * quick version of vm_fault 551 */ 552void 553vm_fault_quick(v, prot) 554 caddr_t v; 555 int prot; 556{ 557 if (prot & VM_PROT_WRITE) 558 subyte(v, fubyte(v)); 559 else 560 fubyte(v); 561} 562 563/* 564 * Finish a fork operation, with process p2 nearly set up. 565 * Copy and update the pcb, set up the stack so that the child 566 * ready to run and return to user mode. 567 */ 568void 569cpu_fork(p1, p2) 570 register struct proc *p1, *p2; 571{ 572 struct pcb *pcb2 = &p2->p_addr->u_pcb; 573 574 /* Ensure that p1's pcb is up to date. */ 575 if (npxproc == p1) 576 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 577 578 /* Copy p1's pcb. */ 579 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 580 581 /* 582 * Create a new fresh stack for the new process. 583 * Copy the trap frame for the return to user mode as if from a 584 * syscall. This copies the user mode register values. 585 */ 586 p2->p_md.md_regs = (struct trapframe *) 587 ((int)p2->p_addr + UPAGES * PAGE_SIZE) - 1; 588 *p2->p_md.md_regs = *p1->p_md.md_regs; 589 590 /* 591 * Set registers for trampoline to user mode. Leave space for the 592 * return address on stack. These are the kernel mode register values. 593 */ 594 pcb2->pcb_cr3 = vtophys(p2->p_vmspace->vm_pmap.pm_pdir); 595 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 596 pcb2->pcb_esi = (int)fork_return; 597 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 598 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 599 pcb2->pcb_ebx = (int)p2; 600 pcb2->pcb_eip = (int)fork_trampoline; 601 /* 602 * pcb2->pcb_ldt: duplicated below, if necessary. 603 * pcb2->pcb_ldt_len: cloned above. 604 * pcb2->pcb_savefpu: cloned above. 605 * pcb2->pcb_flags: cloned above (always 0 here?). 606 * pcb2->pcb_onfault: cloned above (always NULL here?). 607 */ 608 609#ifdef VM86 610 /* 611 * XXX don't copy the i/o pages. this should probably be fixed. 612 */ 613 pcb2->pcb_ext = 0; 614#endif 615 616#ifdef USER_LDT 617 /* Copy the LDT, if necessary. */ 618 if (pcb2->pcb_ldt != 0) { 619 union descriptor *new_ldt; 620 size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 621 622 new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 623 bcopy(pcb2->pcb_ldt, new_ldt, len); 624 pcb2->pcb_ldt = (caddr_t)new_ldt; 625 } 626#endif 627 628 /* 629 * Now, cpu_switch() can schedule the new process. 630 * pcb_esp is loaded pointing to the cpu_switch() stack frame 631 * containing the return address when exiting cpu_switch. 632 * This will normally be to proc_trampoline(), which will have 633 * %ebx loaded with the new proc's pointer. proc_trampoline() 634 * will set up a stack to call fork_return(p, frame); to complete 635 * the return to user-mode. 636 */ 637} 638 639/* 640 * Intercept the return address from a freshly forked process that has NOT 641 * been scheduled yet. 642 * 643 * This is needed to make kernel threads stay in kernel mode. 644 */ 645void 646cpu_set_fork_handler(p, func, arg) 647 struct proc *p; 648 void (*func) __P((void *)); 649 void *arg; 650{ 651 /* 652 * Note that the trap frame follows the args, so the function 653 * is really called like this: func(arg, frame); 654 */ 655 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 656 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 657} 658 659void 660cpu_exit(p) 661 register struct proc *p; 662{ 663#if defined(USER_LDT) || defined(VM86) 664 struct pcb *pcb = &p->p_addr->u_pcb; 665#endif 666 667#if NNPX > 0 668 npxexit(p); 669#endif /* NNPX */ 670#ifdef VM86 671 if (pcb->pcb_ext != 0) { 672 /* 673 * XXX do we need to move the TSS off the allocated pages 674 * before freeing them? (not done here) 675 */ 676 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 677 ctob(IOPAGES + 1)); 678 pcb->pcb_ext = 0; 679 } 680#endif 681#ifdef USER_LDT 682 if (pcb->pcb_ldt != 0) { 683 if (pcb == curpcb) 684 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 685 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 686 pcb->pcb_ldt_len * sizeof(union descriptor)); 687 pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 688 } 689#endif 690 cnt.v_swtch++; 691 cpu_switch(p); 692 panic("cpu_exit"); 693} 694 695void 696cpu_wait(p) 697 struct proc *p; 698{ 699 /* drop per-process resources */ 700 pmap_dispose_proc(p); 701 vmspace_free(p->p_vmspace); 702} 703 704/* 705 * Dump the machine specific header information at the start of a core dump. 706 */ 707int 708cpu_coredump(p, vp, cred) 709 struct proc *p; 710 struct vnode *vp; 711 struct ucred *cred; 712{ 713 714 return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES), 715 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, 716 p)); 717} 718 719#ifdef notyet 720static void 721setredzone(pte, vaddr) 722 u_short *pte; 723 caddr_t vaddr; 724{ 725/* eventually do this by setting up an expand-down stack segment 726 for ss0: selector, allowing stack access down to top of u. 727 this means though that protection violations need to be handled 728 thru a double fault exception that must do an integral task 729 switch to a known good context, within which a dump can be 730 taken. a sensible scheme might be to save the initial context 731 used by sched (that has physical memory mapped 1:1 at bottom) 732 and take the dump while still in mapped mode */ 733} 734#endif 735 736/* 737 * Convert kernel VA to physical address 738 */ 739u_long 740kvtop(void *addr) 741{ 742 vm_offset_t va; 743 744 va = pmap_kextract((vm_offset_t)addr); 745 if (va == 0) 746 panic("kvtop: zero page frame"); 747 return((int)va); 748} 749 750/* 751 * Map an IO request into kernel virtual address space. 752 * 753 * All requests are (re)mapped into kernel VA space. 754 * Notice that we use b_bufsize for the size of the buffer 755 * to be mapped. b_bcount might be modified by the driver. 756 */ 757void 758vmapbuf(bp) 759 register struct buf *bp; 760{ 761 register caddr_t addr, v, kva; 762 vm_offset_t pa; 763 764 if ((bp->b_flags & B_PHYS) == 0) 765 panic("vmapbuf"); 766 767 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page(bp->b_data); 768 addr < bp->b_data + bp->b_bufsize; 769 addr += PAGE_SIZE, v += PAGE_SIZE) { 770 /* 771 * Do the vm_fault if needed; do the copy-on-write thing 772 * when reading stuff off device into memory. 773 */ 774 vm_fault_quick(addr, 775 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 776 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 777 if (pa == 0) 778 panic("vmapbuf: page not present"); 779 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 780 pmap_kenter((vm_offset_t) v, pa); 781 } 782 783 kva = bp->b_saveaddr; 784 bp->b_saveaddr = bp->b_data; 785 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 786} 787 788/* 789 * Free the io map PTEs associated with this IO operation. 790 * We also invalidate the TLB entries and restore the original b_addr. 791 */ 792void 793vunmapbuf(bp) 794 register struct buf *bp; 795{ 796 register caddr_t addr; 797 vm_offset_t pa; 798 799 if ((bp->b_flags & B_PHYS) == 0) 800 panic("vunmapbuf"); 801 802 for (addr = (caddr_t)trunc_page(bp->b_data); 803 addr < bp->b_data + bp->b_bufsize; 804 addr += PAGE_SIZE) { 805 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 806 pmap_kremove((vm_offset_t) addr); 807 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 808 } 809 810 bp->b_data = bp->b_saveaddr; 811} 812 813/* 814 * Force reset the processor by invalidating the entire address space! 815 */ 816void 817cpu_reset() 818{ 819 820#ifdef PC98 821 /* 822 * Attempt to do a CPU reset via CPU reset port. 823 */ 824 disable_intr(); 825 outb(0x37, 0x0f); /* SHUT0 = 0. */ 826 outb(0x37, 0x0b); /* SHUT1 = 0. */ 827 outb(0xf0, 0x00); /* Reset. */ 828#else 829 /* 830 * Attempt to do a CPU reset via the keyboard controller, 831 * do not turn of the GateA20, as any machine that fails 832 * to do the reset here would then end up in no man's land. 833 */ 834 835#if !defined(BROKEN_KEYBOARD_RESET) 836 outb(IO_KBD + 4, 0xFE); 837 DELAY(500000); /* wait 0.5 sec to see if that did it */ 838 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 839 DELAY(1000000); /* wait 1 sec for printf to complete */ 840#endif 841#endif /* PC98 */ 842 /* force a shutdown by unmapping entire address space ! */ 843 bzero((caddr_t) PTD, PAGE_SIZE); 844 845 /* "good night, sweet prince .... <THUNK!>" */ 846 invltlb(); 847 /* NOTREACHED */ 848 while(1); 849} 850 851/* 852 * Grow the user stack to allow for 'sp'. This version grows the stack in 853 * chunks of SGROWSIZ. 854 */ 855int 856grow(p, sp) 857 struct proc *p; 858 u_int sp; 859{ 860 unsigned int nss; 861 caddr_t v; 862 struct vmspace *vm = p->p_vmspace; 863 864 if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK) 865 return (1); 866 867 nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE); 868 869 if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur) 870 return (0); 871 872 if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT, 873 SGROWSIZ) < nss) { 874 int grow_amount; 875 /* 876 * If necessary, grow the VM that the stack occupies 877 * to allow for the rlimit. This allows us to not have 878 * to allocate all of the VM up-front in execve (which 879 * is expensive). 880 * Grow the VM by the amount requested rounded up to 881 * the nearest SGROWSIZ to provide for some hysteresis. 882 */ 883 grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ); 884 v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT, 885 SGROWSIZ) - grow_amount; 886 /* 887 * If there isn't enough room to extend by SGROWSIZ, then 888 * just extend to the maximum size 889 */ 890 if (v < vm->vm_maxsaddr) { 891 v = vm->vm_maxsaddr; 892 grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT); 893 } 894 if ((grow_amount == 0) || (vm_map_find(&vm->vm_map, NULL, 0, (vm_offset_t *)&v, 895 grow_amount, FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != KERN_SUCCESS)) { 896 return (0); 897 } 898 vm->vm_ssize += grow_amount >> PAGE_SHIFT; 899 } 900 901 return (1); 902} 903 904/* 905 * Implement the pre-zeroed page mechanism. 906 * This routine is called from the idle loop. 907 */ 908int 909vm_page_zero_idle() 910{ 911 static int free_rover; 912 vm_page_t m; 913 int s; 914 915#ifdef WRONG 916 if (cnt.v_free_count <= cnt.v_interrupt_free_min) 917 return (0); 918#endif 919 /* 920 * XXX 921 * We stop zeroing pages when there are sufficent prezeroed pages. 922 * This threshold isn't really needed, except we want to 923 * bypass unneeded calls to vm_page_list_find, and the 924 * associated cache flush and latency. The pre-zero will 925 * still be called when there are significantly more 926 * non-prezeroed pages than zeroed pages. The threshold 927 * of half the number of reserved pages is arbitrary, but 928 * approximately the right amount. Eventually, we should 929 * perhaps interrupt the zero operation when a process 930 * is found to be ready to run. 931 */ 932 if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2) 933 return (0); 934#ifdef SMP 935 get_mplock(); 936#endif 937 s = splvm(); 938 enable_intr(); 939 m = vm_page_list_find(PQ_FREE, free_rover); 940 if (m != NULL) { 941 --(*vm_page_queues[m->queue].lcnt); 942 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 943 splx(s); 944#ifdef SMP 945 rel_mplock(); 946#endif 947 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 948#ifdef SMP 949 get_mplock(); 950#endif 951 (void)splvm(); 952 m->queue = PQ_ZERO + m->pc; 953 ++(*vm_page_queues[m->queue].lcnt); 954 TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq); 955 free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK; 956 ++vm_page_zero_count; 957 } 958 splx(s); 959 disable_intr(); 960#ifdef SMP 961 rel_mplock(); 962#endif 963 return (1); 964} 965 966/* 967 * Tell whether this address is in some physical memory region. 968 * Currently used by the kernel coredump code in order to avoid 969 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 970 * or other unpredictable behaviour. 971 */ 972 973#include "isa.h" 974 975int 976is_physical_memory(addr) 977 vm_offset_t addr; 978{ 979 980#if NISA > 0 981 /* The ISA ``memory hole''. */ 982 if (addr >= 0xa0000 && addr < 0x100000) 983 return 0; 984#endif 985 986 /* 987 * stuff other tests for known memory-mapped devices (PCI?) 988 * here 989 */ 990 991 return 1; 992} 993