vm_machdep.c revision 30265
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 1.89 1997/09/13 16:12:04 joerg 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 60#include <vm/vm.h> 61#include <vm/vm_param.h> 62#include <vm/vm_prot.h> 63#include <sys/lock.h> 64#include <vm/vm_kern.h> 65#include <vm/vm_page.h> 66#include <vm/vm_map.h> 67#include <vm/vm_extern.h> 68 69#include <sys/user.h> 70 71#ifdef PC98 72#include <pc98/pc98/pc98.h> 73#else 74#include <i386/isa/isa.h> 75#endif 76 77#ifdef BOUNCE_BUFFERS 78static vm_offset_t 79 vm_bounce_kva __P((int size, int waitok)); 80static void vm_bounce_kva_free __P((vm_offset_t addr, vm_offset_t size, 81 int now)); 82static vm_offset_t 83 vm_bounce_page_find __P((int count)); 84static void vm_bounce_page_free __P((vm_offset_t pa, int count)); 85 86static volatile int kvasfreecnt; 87 88caddr_t bouncememory; 89int bouncepages; 90static int bpwait; 91static vm_offset_t *bouncepa; 92static int bmwait, bmfreeing; 93 94#define BITS_IN_UNSIGNED (8*sizeof(unsigned)) 95static int bounceallocarraysize; 96static unsigned *bounceallocarray; 97static int bouncefree; 98 99#if defined(PC98) && defined (EPSON_BOUNCEDMA) 100#define SIXTEENMEG (3840*4096) /* 15MB boundary */ 101#else 102#define SIXTEENMEG (4096*4096) 103#endif 104#define MAXBKVA 1024 105int maxbkva = MAXBKVA*PAGE_SIZE; 106 107/* special list that can be used at interrupt time for eventual kva free */ 108static struct kvasfree { 109 vm_offset_t addr; 110 vm_offset_t size; 111} kvaf[MAXBKVA]; 112 113/* 114 * get bounce buffer pages (count physically contiguous) 115 * (only 1 inplemented now) 116 */ 117static vm_offset_t 118vm_bounce_page_find(count) 119 int count; 120{ 121 int bit; 122 int s,i; 123 124 if (count != 1) 125 panic("vm_bounce_page_find -- no support for > 1 page yet!!!"); 126 127 s = splbio(); 128retry: 129 for (i = 0; i < bounceallocarraysize; i++) { 130 if (bounceallocarray[i] != 0xffffffff) { 131 bit = ffs(~bounceallocarray[i]); 132 if (bit) { 133 bounceallocarray[i] |= 1 << (bit - 1) ; 134 bouncefree -= count; 135 splx(s); 136 return bouncepa[(i * BITS_IN_UNSIGNED + (bit - 1))]; 137 } 138 } 139 } 140 bpwait = 1; 141 tsleep((caddr_t) &bounceallocarray, PRIBIO, "bncwai", 0); 142 goto retry; 143} 144 145static void 146vm_bounce_kva_free(addr, size, now) 147 vm_offset_t addr; 148 vm_offset_t size; 149 int now; 150{ 151 int s = splbio(); 152 kvaf[kvasfreecnt].addr = addr; 153 kvaf[kvasfreecnt].size = size; 154 ++kvasfreecnt; 155 if( now) { 156 /* 157 * this will do wakeups 158 */ 159 vm_bounce_kva(0,0); 160 } else { 161 if (bmwait) { 162 /* 163 * if anyone is waiting on the bounce-map, then wakeup 164 */ 165 wakeup((caddr_t) io_map); 166 bmwait = 0; 167 } 168 } 169 splx(s); 170} 171 172/* 173 * free count bounce buffer pages 174 */ 175static void 176vm_bounce_page_free(pa, count) 177 vm_offset_t pa; 178 int count; 179{ 180 int allocindex; 181 int index; 182 int bit; 183 184 if (count != 1) 185 panic("vm_bounce_page_free -- no support for > 1 page yet!!!"); 186 187 for(index=0;index<bouncepages;index++) { 188 if( pa == bouncepa[index]) 189 break; 190 } 191 192 if( index == bouncepages) 193 panic("vm_bounce_page_free: invalid bounce buffer"); 194 195 allocindex = index / BITS_IN_UNSIGNED; 196 bit = index % BITS_IN_UNSIGNED; 197 198 bounceallocarray[allocindex] &= ~(1 << bit); 199 200 bouncefree += count; 201 if (bpwait) { 202 bpwait = 0; 203 wakeup((caddr_t) &bounceallocarray); 204 } 205} 206 207/* 208 * allocate count bounce buffer kva pages 209 */ 210static vm_offset_t 211vm_bounce_kva(size, waitok) 212 int size; 213 int waitok; 214{ 215 int i; 216 vm_offset_t kva = 0; 217 vm_offset_t off; 218 int s = splbio(); 219more: 220 if (!bmfreeing && kvasfreecnt) { 221 bmfreeing = 1; 222 for (i = 0; i < kvasfreecnt; i++) { 223 for(off=0;off<kvaf[i].size;off+=PAGE_SIZE) { 224 pmap_kremove( kvaf[i].addr + off); 225 } 226 kmem_free_wakeup(io_map, kvaf[i].addr, 227 kvaf[i].size); 228 } 229 kvasfreecnt = 0; 230 bmfreeing = 0; 231 if( bmwait) { 232 bmwait = 0; 233 wakeup( (caddr_t) io_map); 234 } 235 } 236 237 if( size == 0) { 238 splx(s); 239 return 0; 240 } 241 242 if ((kva = kmem_alloc_pageable(io_map, size)) == 0) { 243 if( !waitok) { 244 splx(s); 245 return 0; 246 } 247 bmwait = 1; 248 tsleep((caddr_t) io_map, PRIBIO, "bmwait", 0); 249 goto more; 250 } 251 splx(s); 252 return kva; 253} 254 255/* 256 * same as vm_bounce_kva -- but really allocate (but takes pages as arg) 257 */ 258vm_offset_t 259vm_bounce_kva_alloc(count) 260int count; 261{ 262 int i; 263 vm_offset_t kva; 264 vm_offset_t pa; 265 if( bouncepages == 0) { 266 kva = (vm_offset_t) malloc(count*PAGE_SIZE, M_TEMP, M_WAITOK); 267 return kva; 268 } 269 kva = vm_bounce_kva(count*PAGE_SIZE, 1); 270 for(i=0;i<count;i++) { 271 pa = vm_bounce_page_find(1); 272 pmap_kenter(kva + i * PAGE_SIZE, pa); 273 } 274 return kva; 275} 276 277/* 278 * same as vm_bounce_kva_free -- but really free 279 */ 280void 281vm_bounce_kva_alloc_free(kva, count) 282 vm_offset_t kva; 283 int count; 284{ 285 int i; 286 vm_offset_t pa; 287 if( bouncepages == 0) { 288 free((caddr_t) kva, M_TEMP); 289 return; 290 } 291 for(i = 0; i < count; i++) { 292 pa = pmap_kextract(kva + i * PAGE_SIZE); 293 vm_bounce_page_free(pa, 1); 294 } 295 vm_bounce_kva_free(kva, count*PAGE_SIZE, 0); 296} 297 298/* 299 * do the things necessary to the struct buf to implement 300 * bounce buffers... inserted before the disk sort 301 */ 302void 303vm_bounce_alloc(bp) 304 struct buf *bp; 305{ 306 int countvmpg; 307 vm_offset_t vastart, vaend; 308 vm_offset_t vapstart, vapend; 309 vm_offset_t va, kva; 310 vm_offset_t pa; 311 int dobounceflag = 0; 312 int i; 313 314 if (bouncepages == 0) 315 return; 316 317 if (bp->b_flags & B_BOUNCE) { 318 printf("vm_bounce_alloc: called recursively???\n"); 319 return; 320 } 321 322 if (bp->b_bufsize < bp->b_bcount) { 323 printf( 324 "vm_bounce_alloc: b_bufsize(0x%lx) < b_bcount(0x%lx) !!\n", 325 bp->b_bufsize, bp->b_bcount); 326 panic("vm_bounce_alloc"); 327 } 328 329/* 330 * This is not really necessary 331 * if( bp->b_bufsize != bp->b_bcount) { 332 * printf("size: %d, count: %d\n", bp->b_bufsize, bp->b_bcount); 333 * } 334 */ 335 336 337 vastart = (vm_offset_t) bp->b_data; 338 vaend = (vm_offset_t) bp->b_data + bp->b_bufsize; 339 340 vapstart = trunc_page(vastart); 341 vapend = round_page(vaend); 342 countvmpg = (vapend - vapstart) / PAGE_SIZE; 343 344/* 345 * if any page is above 16MB, then go into bounce-buffer mode 346 */ 347 va = vapstart; 348 for (i = 0; i < countvmpg; i++) { 349 pa = pmap_kextract(va); 350 if (pa >= SIXTEENMEG) 351 ++dobounceflag; 352 if( pa == 0) 353 panic("vm_bounce_alloc: Unmapped page"); 354 va += PAGE_SIZE; 355 } 356 if (dobounceflag == 0) 357 return; 358 359 if (bouncepages < dobounceflag) 360 panic("Not enough bounce buffers!!!"); 361 362/* 363 * allocate a replacement kva for b_addr 364 */ 365 kva = vm_bounce_kva(countvmpg*PAGE_SIZE, 1); 366#if 0 367 printf("%s: vapstart: %x, vapend: %x, countvmpg: %d, kva: %x ", 368 (bp->b_flags & B_READ) ? "read":"write", 369 vapstart, vapend, countvmpg, kva); 370#endif 371 va = vapstart; 372 for (i = 0; i < countvmpg; i++) { 373 pa = pmap_kextract(va); 374 if (pa >= SIXTEENMEG) { 375 /* 376 * allocate a replacement page 377 */ 378 vm_offset_t bpa = vm_bounce_page_find(1); 379 pmap_kenter(kva + (PAGE_SIZE * i), bpa); 380#if 0 381 printf("r(%d): (%x,%x,%x) ", i, va, pa, bpa); 382#endif 383 /* 384 * if we are writing, the copy the data into the page 385 */ 386 if ((bp->b_flags & B_READ) == 0) { 387 bcopy((caddr_t) va, (caddr_t) kva + (PAGE_SIZE * i), PAGE_SIZE); 388 } 389 } else { 390 /* 391 * use original page 392 */ 393 pmap_kenter(kva + (PAGE_SIZE * i), pa); 394 } 395 va += PAGE_SIZE; 396 } 397 398/* 399 * flag the buffer as being bounced 400 */ 401 bp->b_flags |= B_BOUNCE; 402/* 403 * save the original buffer kva 404 */ 405 bp->b_savekva = bp->b_data; 406/* 407 * put our new kva into the buffer (offset by original offset) 408 */ 409 bp->b_data = (caddr_t) (((vm_offset_t) kva) | 410 ((vm_offset_t) bp->b_savekva & PAGE_MASK)); 411#if 0 412 printf("b_savekva: %x, newva: %x\n", bp->b_savekva, bp->b_data); 413#endif 414 return; 415} 416 417/* 418 * hook into biodone to free bounce buffer 419 */ 420void 421vm_bounce_free(bp) 422 struct buf *bp; 423{ 424 int i; 425 vm_offset_t origkva, bouncekva, bouncekvaend; 426 427/* 428 * if this isn't a bounced buffer, then just return 429 */ 430 if ((bp->b_flags & B_BOUNCE) == 0) 431 return; 432 433/* 434 * This check is not necessary 435 * if (bp->b_bufsize != bp->b_bcount) { 436 * printf("vm_bounce_free: b_bufsize=%d, b_bcount=%d\n", 437 * bp->b_bufsize, bp->b_bcount); 438 * } 439 */ 440 441 origkva = (vm_offset_t) bp->b_savekva; 442 bouncekva = (vm_offset_t) bp->b_data; 443/* 444 printf("free: %d ", bp->b_bufsize); 445*/ 446 447/* 448 * check every page in the kva space for b_addr 449 */ 450 for (i = 0; i < bp->b_bufsize; ) { 451 vm_offset_t mybouncepa; 452 vm_offset_t copycount; 453 454 copycount = round_page(bouncekva + 1) - bouncekva; 455 mybouncepa = pmap_kextract(trunc_page(bouncekva)); 456 457/* 458 * if this is a bounced pa, then process as one 459 */ 460 if ( mybouncepa != pmap_kextract( trunc_page( origkva))) { 461 vm_offset_t tocopy = copycount; 462 if (i + tocopy > bp->b_bufsize) 463 tocopy = bp->b_bufsize - i; 464/* 465 * if this is a read, then copy from bounce buffer into original buffer 466 */ 467 if (bp->b_flags & B_READ) 468 bcopy((caddr_t) bouncekva, (caddr_t) origkva, tocopy); 469/* 470 * free the bounce allocation 471 */ 472 473/* 474 printf("(kva: %x, pa: %x)", bouncekva, mybouncepa); 475*/ 476 vm_bounce_page_free(mybouncepa, 1); 477 } 478 479 origkva += copycount; 480 bouncekva += copycount; 481 i += copycount; 482 } 483 484/* 485 printf("\n"); 486*/ 487/* 488 * add the old kva into the "to free" list 489 */ 490 491 bouncekva= trunc_page((vm_offset_t) bp->b_data); 492 bouncekvaend= round_page((vm_offset_t)bp->b_data + bp->b_bufsize); 493 494/* 495 printf("freeva: %d\n", (bouncekvaend - bouncekva) / PAGE_SIZE); 496*/ 497 vm_bounce_kva_free( bouncekva, (bouncekvaend - bouncekva), 0); 498 bp->b_data = bp->b_savekva; 499 bp->b_savekva = 0; 500 bp->b_flags &= ~B_BOUNCE; 501 502 return; 503} 504 505 506/* 507 * init the bounce buffer system 508 */ 509void 510vm_bounce_init() 511{ 512 int i; 513 514 kvasfreecnt = 0; 515 516 if (bouncepages == 0) 517 return; 518 519 bounceallocarraysize = (bouncepages + BITS_IN_UNSIGNED - 1) / BITS_IN_UNSIGNED; 520 bounceallocarray = malloc(bounceallocarraysize * sizeof(unsigned), M_TEMP, M_NOWAIT); 521 522 if (!bounceallocarray) 523 panic("Cannot allocate bounce resource array"); 524 525 bouncepa = malloc(bouncepages * sizeof(vm_offset_t), M_TEMP, M_NOWAIT); 526 if (!bouncepa) 527 panic("Cannot allocate physical memory array"); 528 529 for(i=0;i<bounceallocarraysize;i++) { 530 bounceallocarray[i] = 0xffffffff; 531 } 532 533 for(i=0;i<bouncepages;i++) { 534 vm_offset_t pa; 535 if( (pa = pmap_kextract((vm_offset_t) bouncememory + i * PAGE_SIZE)) >= SIXTEENMEG) 536 panic("bounce memory out of range"); 537 if( pa == 0) 538 panic("bounce memory not resident"); 539 bouncepa[i] = pa; 540 bounceallocarray[i/(8*sizeof(int))] &= ~(1<<(i%(8*sizeof(int)))); 541 } 542 bouncefree = bouncepages; 543 544} 545#endif /* BOUNCE_BUFFERS */ 546 547/* 548 * quick version of vm_fault 549 */ 550void 551vm_fault_quick(v, prot) 552 caddr_t v; 553 int prot; 554{ 555 if (prot & VM_PROT_WRITE) 556 subyte(v, fubyte(v)); 557 else 558 fubyte(v); 559} 560 561/* 562 * Finish a fork operation, with process p2 nearly set up. 563 * Copy and update the pcb, set up the stack so that the child 564 * ready to run and return to user mode. 565 */ 566void 567cpu_fork(p1, p2) 568 register struct proc *p1, *p2; 569{ 570 struct pcb *pcb2 = &p2->p_addr->u_pcb; 571 572 /* Ensure that p1's pcb is up to date. */ 573 if (npxproc == p1) 574 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 575 576 /* Copy p1's pcb. */ 577 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 578 579 /* 580 * Create a new fresh stack for the new process. 581 * Copy the trap frame for the return to user mode as if from a 582 * syscall. This copies the user mode register values. 583 */ 584 p2->p_md.md_regs = (struct trapframe *) 585 ((int)p2->p_addr + UPAGES * PAGE_SIZE) - 1; 586 *p2->p_md.md_regs = *p1->p_md.md_regs; 587 588 /* 589 * Set registers for trampoline to user mode. Leave space for the 590 * return address on stack. These are the kernel mode register values. 591 */ 592 pcb2->pcb_cr3 = vtophys(p2->p_vmspace->vm_pmap.pm_pdir); 593 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 594 pcb2->pcb_esi = (int)fork_return; 595 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 596 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 597 pcb2->pcb_ebx = (int)p2; 598 pcb2->pcb_eip = (int)fork_trampoline; 599 /* 600 * pcb2->pcb_ldt: duplicated below, if necessary. 601 * pcb2->pcb_ldt_len: cloned above. 602 * pcb2->pcb_savefpu: cloned above. 603 * pcb2->pcb_flags: cloned above (always 0 here?). 604 * pcb2->pcb_onfault: cloned above (always NULL here?). 605 */ 606 607#ifdef VM86 608 /* 609 * XXX don't copy the i/o pages. this should probably be fixed. 610 */ 611 pcb2->pcb_ext = 0; 612#endif 613 614#ifdef USER_LDT 615 /* Copy the LDT, if necessary. */ 616 if (pcb2->pcb_ldt != 0) { 617 union descriptor *new_ldt; 618 size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 619 620 new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 621 bcopy(pcb2->pcb_ldt, new_ldt, len); 622 pcb2->pcb_ldt = (caddr_t)new_ldt; 623 } 624#endif 625 626 /* 627 * Now, cpu_switch() can schedule the new process. 628 * pcb_esp is loaded pointing to the cpu_switch() stack frame 629 * containing the return address when exiting cpu_switch. 630 * This will normally be to proc_trampoline(), which will have 631 * %ebx loaded with the new proc's pointer. proc_trampoline() 632 * will set up a stack to call fork_return(p, frame); to complete 633 * the return to user-mode. 634 */ 635} 636 637/* 638 * Intercept the return address from a freshly forked process that has NOT 639 * been scheduled yet. 640 * 641 * This is needed to make kernel threads stay in kernel mode. 642 */ 643void 644cpu_set_fork_handler(p, func, arg) 645 struct proc *p; 646 void (*func) __P((void *)); 647 void *arg; 648{ 649 /* 650 * Note that the trap frame follows the args, so the function 651 * is really called like this: func(arg, frame); 652 */ 653 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 654 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 655} 656 657void 658cpu_exit(p) 659 register struct proc *p; 660{ 661#if defined(USER_LDT) || defined(VM86) 662 struct pcb *pcb = &p->p_addr->u_pcb; 663#endif 664 665#if NNPX > 0 666 npxexit(p); 667#endif /* NNPX */ 668#ifdef VM86 669 if (pcb->pcb_ext != 0) { 670 /* 671 * XXX do we need to move the TSS off the allocated pages 672 * before freeing them? (not done here) 673 */ 674 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 675 ctob(IOPAGES + 1)); 676 pcb->pcb_ext = 0; 677 } 678#endif 679#ifdef USER_LDT 680 if (pcb->pcb_ldt != 0) { 681 if (pcb == curpcb) 682 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 683 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 684 pcb->pcb_ldt_len * sizeof(union descriptor)); 685 pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 686 } 687#endif 688 cnt.v_swtch++; 689 cpu_switch(p); 690 panic("cpu_exit"); 691} 692 693void 694cpu_wait(p) 695 struct proc *p; 696{ 697 /* drop per-process resources */ 698 pmap_dispose_proc(p); 699 vmspace_free(p->p_vmspace); 700} 701 702/* 703 * Dump the machine specific header information at the start of a core dump. 704 */ 705int 706cpu_coredump(p, vp, cred) 707 struct proc *p; 708 struct vnode *vp; 709 struct ucred *cred; 710{ 711 712 return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES), 713 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, 714 p)); 715} 716 717#ifdef notyet 718static void 719setredzone(pte, vaddr) 720 u_short *pte; 721 caddr_t vaddr; 722{ 723/* eventually do this by setting up an expand-down stack segment 724 for ss0: selector, allowing stack access down to top of u. 725 this means though that protection violations need to be handled 726 thru a double fault exception that must do an integral task 727 switch to a known good context, within which a dump can be 728 taken. a sensible scheme might be to save the initial context 729 used by sched (that has physical memory mapped 1:1 at bottom) 730 and take the dump while still in mapped mode */ 731} 732#endif 733 734/* 735 * Convert kernel VA to physical address 736 */ 737u_long 738kvtop(void *addr) 739{ 740 vm_offset_t va; 741 742 va = pmap_kextract((vm_offset_t)addr); 743 if (va == 0) 744 panic("kvtop: zero page frame"); 745 return((int)va); 746} 747 748/* 749 * Map an IO request into kernel virtual address space. 750 * 751 * All requests are (re)mapped into kernel VA space. 752 * Notice that we use b_bufsize for the size of the buffer 753 * to be mapped. b_bcount might be modified by the driver. 754 */ 755void 756vmapbuf(bp) 757 register struct buf *bp; 758{ 759 register caddr_t addr, v, kva; 760 vm_offset_t pa; 761 762 if ((bp->b_flags & B_PHYS) == 0) 763 panic("vmapbuf"); 764 765 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page(bp->b_data); 766 addr < bp->b_data + bp->b_bufsize; 767 addr += PAGE_SIZE, v += PAGE_SIZE) { 768 /* 769 * Do the vm_fault if needed; do the copy-on-write thing 770 * when reading stuff off device into memory. 771 */ 772 vm_fault_quick(addr, 773 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 774 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 775 if (pa == 0) 776 panic("vmapbuf: page not present"); 777 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 778 pmap_kenter((vm_offset_t) v, pa); 779 } 780 781 kva = bp->b_saveaddr; 782 bp->b_saveaddr = bp->b_data; 783 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 784} 785 786/* 787 * Free the io map PTEs associated with this IO operation. 788 * We also invalidate the TLB entries and restore the original b_addr. 789 */ 790void 791vunmapbuf(bp) 792 register struct buf *bp; 793{ 794 register caddr_t addr; 795 vm_offset_t pa; 796 797 if ((bp->b_flags & B_PHYS) == 0) 798 panic("vunmapbuf"); 799 800 for (addr = (caddr_t)trunc_page(bp->b_data); 801 addr < bp->b_data + bp->b_bufsize; 802 addr += PAGE_SIZE) { 803 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 804 pmap_kremove((vm_offset_t) addr); 805 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 806 } 807 808 bp->b_data = bp->b_saveaddr; 809} 810 811/* 812 * Force reset the processor by invalidating the entire address space! 813 */ 814void 815cpu_reset() { 816#ifdef PC98 817 /* 818 * Attempt to do a CPU reset via CPU reset port. 819 */ 820 asm("cli"); 821 outb(0x37, 0x0f); /* SHUT0 = 0. */ 822 outb(0x37, 0x0b); /* SHUT1 = 0. */ 823 outb(0xf0, 0x00); /* Reset. */ 824#else 825 /* 826 * Attempt to do a CPU reset via the keyboard controller, 827 * do not turn of the GateA20, as any machine that fails 828 * to do the reset here would then end up in no man's land. 829 */ 830 831#if !defined(BROKEN_KEYBOARD_RESET) 832 outb(IO_KBD + 4, 0xFE); 833 DELAY(500000); /* wait 0.5 sec to see if that did it */ 834 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 835 DELAY(1000000); /* wait 1 sec for printf to complete */ 836#endif 837#endif /* PC98 */ 838 /* force a shutdown by unmapping entire address space ! */ 839 bzero((caddr_t) PTD, PAGE_SIZE); 840 841 /* "good night, sweet prince .... <THUNK!>" */ 842 invltlb(); 843 /* NOTREACHED */ 844 while(1); 845} 846 847/* 848 * Grow the user stack to allow for 'sp'. This version grows the stack in 849 * chunks of SGROWSIZ. 850 */ 851int 852grow(p, sp) 853 struct proc *p; 854 u_int sp; 855{ 856 unsigned int nss; 857 caddr_t v; 858 struct vmspace *vm = p->p_vmspace; 859 860 if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK) 861 return (1); 862 863 nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE); 864 865 if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur) 866 return (0); 867 868 if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT, 869 SGROWSIZ) < nss) { 870 int grow_amount; 871 /* 872 * If necessary, grow the VM that the stack occupies 873 * to allow for the rlimit. This allows us to not have 874 * to allocate all of the VM up-front in execve (which 875 * is expensive). 876 * Grow the VM by the amount requested rounded up to 877 * the nearest SGROWSIZ to provide for some hysteresis. 878 */ 879 grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ); 880 v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT, 881 SGROWSIZ) - grow_amount; 882 /* 883 * If there isn't enough room to extend by SGROWSIZ, then 884 * just extend to the maximum size 885 */ 886 if (v < vm->vm_maxsaddr) { 887 v = vm->vm_maxsaddr; 888 grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT); 889 } 890 if ((grow_amount == 0) || (vm_map_find(&vm->vm_map, NULL, 0, (vm_offset_t *)&v, 891 grow_amount, FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != KERN_SUCCESS)) { 892 return (0); 893 } 894 vm->vm_ssize += grow_amount >> PAGE_SHIFT; 895 } 896 897 return (1); 898} 899 900/* 901 * Implement the pre-zeroed page mechanism. 902 * This routine is called from the idle loop. 903 */ 904int 905vm_page_zero_idle() 906{ 907 static int free_rover; 908 vm_page_t m; 909 int s; 910 911#ifdef WRONG 912 if (cnt.v_free_count <= cnt.v_interrupt_free_min) 913 return (0); 914#endif 915 /* 916 * XXX 917 * We stop zeroing pages when there are sufficent prezeroed pages. 918 * This threshold isn't really needed, except we want to 919 * bypass unneeded calls to vm_page_list_find, and the 920 * associated cache flush and latency. The pre-zero will 921 * still be called when there are significantly more 922 * non-prezeroed pages than zeroed pages. The threshold 923 * of half the number of reserved pages is arbitrary, but 924 * approximately the right amount. Eventually, we should 925 * perhaps interrupt the zero operation when a process 926 * is found to be ready to run. 927 */ 928 if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2) 929 return (0); 930#ifdef SMP 931 get_mplock(); 932#endif 933 s = splvm(); 934 enable_intr(); 935 m = vm_page_list_find(PQ_FREE, free_rover); 936 if (m != NULL) { 937 --(*vm_page_queues[m->queue].lcnt); 938 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 939 splx(s); 940#ifdef SMP 941 rel_mplock(); 942#endif 943 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 944#ifdef SMP 945 get_mplock(); 946#endif 947 (void)splvm(); 948 m->queue = PQ_ZERO + m->pc; 949 ++(*vm_page_queues[m->queue].lcnt); 950 TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq); 951 free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK; 952 ++vm_page_zero_count; 953 } 954 splx(s); 955 disable_intr(); 956#ifdef SMP 957 rel_mplock(); 958#endif 959 return (1); 960} 961 962/* 963 * Tell whether this address is in some physical memory region. 964 * Currently used by the kernel coredump code in order to avoid 965 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 966 * or other unpredictable behaviour. 967 */ 968 969#include "isa.h" 970 971int 972is_physical_memory(addr) 973 vm_offset_t addr; 974{ 975 976#if NISA > 0 977 /* The ISA ``memory hole''. */ 978 if (addr >= 0xa0000 && addr < 0x100000) 979 return 0; 980#endif 981 982 /* 983 * stuff other tests for known memory-mapped devices (PCI?) 984 * here 985 */ 986 987 return 1; 988} 989