vm_machdep.c revision 33307
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 1.99 1998/02/06 12:13:11 eivind Exp $ 42 */ 43 44#include "npx.h" 45#include "opt_bounce.h" 46#include "opt_user_ldt.h" 47#include "opt_vm86.h" 48 49#include <sys/param.h> 50#include <sys/systm.h> 51#include <sys/proc.h> 52#include <sys/malloc.h> 53#include <sys/buf.h> 54#include <sys/vnode.h> 55#include <sys/vmmeter.h> 56 57#include <machine/clock.h> 58#include <machine/cpu.h> 59#include <machine/md_var.h> 60#ifdef SMP 61#include <machine/smp.h> 62#endif 63 64#include <vm/vm.h> 65#include <vm/vm_param.h> 66#include <vm/vm_prot.h> 67#include <sys/lock.h> 68#include <vm/vm_kern.h> 69#include <vm/vm_page.h> 70#include <vm/vm_map.h> 71#include <vm/vm_extern.h> 72 73#include <sys/user.h> 74 75#ifdef PC98 76#include <pc98/pc98/pc98.h> 77#else 78#include <i386/isa/isa.h> 79#endif 80 81#ifdef BOUNCE_BUFFERS 82static vm_offset_t 83 vm_bounce_kva __P((int size, int waitok)); 84static void vm_bounce_kva_free __P((vm_offset_t addr, vm_offset_t size, 85 int now)); 86static vm_offset_t 87 vm_bounce_page_find __P((int count)); 88static void vm_bounce_page_free __P((vm_offset_t pa, int count)); 89 90static volatile int kvasfreecnt; 91 92caddr_t bouncememory; 93static int bpwait; 94static vm_offset_t *bouncepa; 95static int bmwait, bmfreeing; 96 97#define BITS_IN_UNSIGNED (8*sizeof(unsigned)) 98static int bounceallocarraysize; 99static unsigned *bounceallocarray; 100static int bouncefree; 101 102#if defined(PC98) && defined (EPSON_BOUNCEDMA) 103#define SIXTEENMEG (3840*4096) /* 15MB boundary */ 104#else 105#define SIXTEENMEG (4096*4096) 106#endif 107#define MAXBKVA 1024 108int maxbkva = MAXBKVA*PAGE_SIZE; 109 110/* special list that can be used at interrupt time for eventual kva free */ 111static struct kvasfree { 112 vm_offset_t addr; 113 vm_offset_t size; 114} kvaf[MAXBKVA]; 115 116/* 117 * get bounce buffer pages (count physically contiguous) 118 * (only 1 inplemented now) 119 */ 120static vm_offset_t 121vm_bounce_page_find(count) 122 int count; 123{ 124 int bit; 125 int s,i; 126 127 if (count != 1) 128 panic("vm_bounce_page_find -- no support for > 1 page yet!!!"); 129 130 s = splbio(); 131retry: 132 for (i = 0; i < bounceallocarraysize; i++) { 133 if (bounceallocarray[i] != 0xffffffff) { 134 bit = ffs(~bounceallocarray[i]); 135 if (bit) { 136 bounceallocarray[i] |= 1 << (bit - 1) ; 137 bouncefree -= count; 138 splx(s); 139 return bouncepa[(i * BITS_IN_UNSIGNED + (bit - 1))]; 140 } 141 } 142 } 143 bpwait = 1; 144 tsleep((caddr_t) &bounceallocarray, PRIBIO, "bncwai", 0); 145 goto retry; 146} 147 148static void 149vm_bounce_kva_free(addr, size, now) 150 vm_offset_t addr; 151 vm_offset_t size; 152 int now; 153{ 154 int s = splbio(); 155 kvaf[kvasfreecnt].addr = addr; 156 kvaf[kvasfreecnt].size = size; 157 ++kvasfreecnt; 158 if( now) { 159 /* 160 * this will do wakeups 161 */ 162 vm_bounce_kva(0,0); 163 } else { 164 if (bmwait) { 165 /* 166 * if anyone is waiting on the bounce-map, then wakeup 167 */ 168 wakeup((caddr_t) io_map); 169 bmwait = 0; 170 } 171 } 172 splx(s); 173} 174 175/* 176 * free count bounce buffer pages 177 */ 178static void 179vm_bounce_page_free(pa, count) 180 vm_offset_t pa; 181 int count; 182{ 183 int allocindex; 184 int index; 185 int bit; 186 187 if (count != 1) 188 panic("vm_bounce_page_free -- no support for > 1 page yet!!!"); 189 190 for(index=0;index<bouncepages;index++) { 191 if( pa == bouncepa[index]) 192 break; 193 } 194 195 if( index == bouncepages) 196 panic("vm_bounce_page_free: invalid bounce buffer"); 197 198 allocindex = index / BITS_IN_UNSIGNED; 199 bit = index % BITS_IN_UNSIGNED; 200 201 bounceallocarray[allocindex] &= ~(1 << bit); 202 203 bouncefree += count; 204 if (bpwait) { 205 bpwait = 0; 206 wakeup((caddr_t) &bounceallocarray); 207 } 208} 209 210/* 211 * allocate count bounce buffer kva pages 212 */ 213static vm_offset_t 214vm_bounce_kva(size, waitok) 215 int size; 216 int waitok; 217{ 218 int i; 219 vm_offset_t kva = 0; 220 vm_offset_t off; 221 int s = splbio(); 222more: 223 if (!bmfreeing && kvasfreecnt) { 224 bmfreeing = 1; 225 for (i = 0; i < kvasfreecnt; i++) { 226 for(off=0;off<kvaf[i].size;off+=PAGE_SIZE) { 227 pmap_kremove( kvaf[i].addr + off); 228 } 229 kmem_free_wakeup(io_map, kvaf[i].addr, 230 kvaf[i].size); 231 } 232 kvasfreecnt = 0; 233 bmfreeing = 0; 234 if( bmwait) { 235 bmwait = 0; 236 wakeup( (caddr_t) io_map); 237 } 238 } 239 240 if( size == 0) { 241 splx(s); 242 return 0; 243 } 244 245 if ((kva = kmem_alloc_pageable(io_map, size)) == 0) { 246 if( !waitok) { 247 splx(s); 248 return 0; 249 } 250 bmwait = 1; 251 tsleep((caddr_t) io_map, PRIBIO, "bmwait", 0); 252 goto more; 253 } 254 splx(s); 255 return kva; 256} 257 258/* 259 * same as vm_bounce_kva -- but really allocate (but takes pages as arg) 260 */ 261vm_offset_t 262vm_bounce_kva_alloc(count) 263int count; 264{ 265 int i; 266 vm_offset_t kva; 267 vm_offset_t pa; 268 if( bouncepages == 0) { 269 kva = (vm_offset_t) malloc(count*PAGE_SIZE, M_TEMP, M_WAITOK); 270 return kva; 271 } 272 kva = vm_bounce_kva(count*PAGE_SIZE, 1); 273 for(i=0;i<count;i++) { 274 pa = vm_bounce_page_find(1); 275 pmap_kenter(kva + i * PAGE_SIZE, pa); 276 } 277 return kva; 278} 279 280/* 281 * same as vm_bounce_kva_free -- but really free 282 */ 283void 284vm_bounce_kva_alloc_free(kva, count) 285 vm_offset_t kva; 286 int count; 287{ 288 int i; 289 vm_offset_t pa; 290 if( bouncepages == 0) { 291 free((caddr_t) kva, M_TEMP); 292 return; 293 } 294 for(i = 0; i < count; i++) { 295 pa = pmap_kextract(kva + i * PAGE_SIZE); 296 vm_bounce_page_free(pa, 1); 297 } 298 vm_bounce_kva_free(kva, count*PAGE_SIZE, 0); 299} 300 301/* 302 * do the things necessary to the struct buf to implement 303 * bounce buffers... inserted before the disk sort 304 */ 305void 306vm_bounce_alloc(bp) 307 struct buf *bp; 308{ 309 int countvmpg; 310 vm_offset_t vastart, vaend; 311 vm_offset_t vapstart, vapend; 312 vm_offset_t va, kva; 313 vm_offset_t pa; 314 int dobounceflag = 0; 315 int i; 316 317 if (bouncepages == 0) 318 return; 319 320 if (bp->b_flags & B_BOUNCE) { 321 printf("vm_bounce_alloc: called recursively???\n"); 322 return; 323 } 324 325 if (bp->b_bufsize < bp->b_bcount) { 326 printf( 327 "vm_bounce_alloc: b_bufsize(0x%lx) < b_bcount(0x%lx) !!\n", 328 bp->b_bufsize, bp->b_bcount); 329 panic("vm_bounce_alloc"); 330 } 331 332/* 333 * This is not really necessary 334 * if( bp->b_bufsize != bp->b_bcount) { 335 * printf("size: %d, count: %d\n", bp->b_bufsize, bp->b_bcount); 336 * } 337 */ 338 339 340 vastart = (vm_offset_t) bp->b_data; 341 vaend = (vm_offset_t) bp->b_data + bp->b_bufsize; 342 343 vapstart = trunc_page(vastart); 344 vapend = round_page(vaend); 345 countvmpg = (vapend - vapstart) / PAGE_SIZE; 346 347/* 348 * if any page is above 16MB, then go into bounce-buffer mode 349 */ 350 va = vapstart; 351 for (i = 0; i < countvmpg; i++) { 352 pa = pmap_kextract(va); 353 if (pa >= SIXTEENMEG) 354 ++dobounceflag; 355 if( pa == 0) 356 panic("vm_bounce_alloc: Unmapped page"); 357 va += PAGE_SIZE; 358 } 359 if (dobounceflag == 0) 360 return; 361 362 if (bouncepages < dobounceflag) 363 panic("Not enough bounce buffers!!!"); 364 365/* 366 * allocate a replacement kva for b_addr 367 */ 368 kva = vm_bounce_kva(countvmpg*PAGE_SIZE, 1); 369#if 0 370 printf("%s: vapstart: %x, vapend: %x, countvmpg: %d, kva: %x ", 371 (bp->b_flags & B_READ) ? "read":"write", 372 vapstart, vapend, countvmpg, kva); 373#endif 374 va = vapstart; 375 for (i = 0; i < countvmpg; i++) { 376 pa = pmap_kextract(va); 377 if (pa >= SIXTEENMEG) { 378 /* 379 * allocate a replacement page 380 */ 381 vm_offset_t bpa = vm_bounce_page_find(1); 382 pmap_kenter(kva + (PAGE_SIZE * i), bpa); 383#if 0 384 printf("r(%d): (%x,%x,%x) ", i, va, pa, bpa); 385#endif 386 /* 387 * if we are writing, the copy the data into the page 388 */ 389 if ((bp->b_flags & B_READ) == 0) { 390 bcopy((caddr_t) va, (caddr_t) kva + (PAGE_SIZE * i), PAGE_SIZE); 391 } 392 } else { 393 /* 394 * use original page 395 */ 396 pmap_kenter(kva + (PAGE_SIZE * i), pa); 397 } 398 va += PAGE_SIZE; 399 } 400 401/* 402 * flag the buffer as being bounced 403 */ 404 bp->b_flags |= B_BOUNCE; 405/* 406 * save the original buffer kva 407 */ 408 bp->b_savekva = bp->b_data; 409/* 410 * put our new kva into the buffer (offset by original offset) 411 */ 412 bp->b_data = (caddr_t) (((vm_offset_t) kva) | 413 ((vm_offset_t) bp->b_savekva & PAGE_MASK)); 414#if 0 415 printf("b_savekva: %x, newva: %x\n", bp->b_savekva, bp->b_data); 416#endif 417 return; 418} 419 420/* 421 * hook into biodone to free bounce buffer 422 */ 423void 424vm_bounce_free(bp) 425 struct buf *bp; 426{ 427 int i; 428 vm_offset_t origkva, bouncekva, bouncekvaend; 429 430/* 431 * if this isn't a bounced buffer, then just return 432 */ 433 if ((bp->b_flags & B_BOUNCE) == 0) 434 return; 435 436/* 437 * This check is not necessary 438 * if (bp->b_bufsize != bp->b_bcount) { 439 * printf("vm_bounce_free: b_bufsize=%d, b_bcount=%d\n", 440 * bp->b_bufsize, bp->b_bcount); 441 * } 442 */ 443 444 origkva = (vm_offset_t) bp->b_savekva; 445 bouncekva = (vm_offset_t) bp->b_data; 446/* 447 printf("free: %d ", bp->b_bufsize); 448*/ 449 450/* 451 * check every page in the kva space for b_addr 452 */ 453 for (i = 0; i < bp->b_bufsize; ) { 454 vm_offset_t mybouncepa; 455 vm_offset_t copycount; 456 457 copycount = round_page(bouncekva + 1) - bouncekva; 458 mybouncepa = pmap_kextract(trunc_page(bouncekva)); 459 460/* 461 * if this is a bounced pa, then process as one 462 */ 463 if ( mybouncepa != pmap_kextract( trunc_page( origkva))) { 464 vm_offset_t tocopy = copycount; 465 if (i + tocopy > bp->b_bufsize) 466 tocopy = bp->b_bufsize - i; 467/* 468 * if this is a read, then copy from bounce buffer into original buffer 469 */ 470 if (bp->b_flags & B_READ) 471 bcopy((caddr_t) bouncekva, (caddr_t) origkva, tocopy); 472/* 473 * free the bounce allocation 474 */ 475 476/* 477 printf("(kva: %x, pa: %x)", bouncekva, mybouncepa); 478*/ 479 vm_bounce_page_free(mybouncepa, 1); 480 } 481 482 origkva += copycount; 483 bouncekva += copycount; 484 i += copycount; 485 } 486 487/* 488 printf("\n"); 489*/ 490/* 491 * add the old kva into the "to free" list 492 */ 493 494 bouncekva= trunc_page((vm_offset_t) bp->b_data); 495 bouncekvaend= round_page((vm_offset_t)bp->b_data + bp->b_bufsize); 496 497/* 498 printf("freeva: %d\n", (bouncekvaend - bouncekva) / PAGE_SIZE); 499*/ 500 vm_bounce_kva_free( bouncekva, (bouncekvaend - bouncekva), 0); 501 bp->b_data = bp->b_savekva; 502 bp->b_savekva = 0; 503 bp->b_flags &= ~B_BOUNCE; 504 505 return; 506} 507 508 509/* 510 * init the bounce buffer system 511 */ 512void 513vm_bounce_init() 514{ 515 int i; 516 517 kvasfreecnt = 0; 518 519 if (bouncepages == 0) 520 return; 521 522 bounceallocarraysize = (bouncepages + BITS_IN_UNSIGNED - 1) / BITS_IN_UNSIGNED; 523 bounceallocarray = malloc(bounceallocarraysize * sizeof(unsigned), M_TEMP, M_NOWAIT); 524 525 if (!bounceallocarray) 526 panic("Cannot allocate bounce resource array"); 527 528 bouncepa = malloc(bouncepages * sizeof(vm_offset_t), M_TEMP, M_NOWAIT); 529 if (!bouncepa) 530 panic("Cannot allocate physical memory array"); 531 532 for(i=0;i<bounceallocarraysize;i++) { 533 bounceallocarray[i] = 0xffffffff; 534 } 535 536 for(i=0;i<bouncepages;i++) { 537 vm_offset_t pa; 538 if( (pa = pmap_kextract((vm_offset_t) bouncememory + i * PAGE_SIZE)) >= SIXTEENMEG) { 539 printf("vm_bounce_init: bounce memory out of range -- bounce disabled\n"); 540 free(bounceallocarray, M_TEMP); 541 bounceallocarray = NULL; 542 free(bouncepa, M_TEMP); 543 bouncepa = NULL; 544 bouncepages = 0; 545 break; 546 } 547 if( pa == 0) 548 panic("bounce memory not resident"); 549 bouncepa[i] = pa; 550 bounceallocarray[i/(8*sizeof(int))] &= ~(1<<(i%(8*sizeof(int)))); 551 } 552 bouncefree = bouncepages; 553 554} 555#endif /* BOUNCE_BUFFERS */ 556 557/* 558 * quick version of vm_fault 559 */ 560void 561vm_fault_quick(v, prot) 562 caddr_t v; 563 int prot; 564{ 565 if (prot & VM_PROT_WRITE) 566 subyte(v, fubyte(v)); 567 else 568 fubyte(v); 569} 570 571/* 572 * Finish a fork operation, with process p2 nearly set up. 573 * Copy and update the pcb, set up the stack so that the child 574 * ready to run and return to user mode. 575 */ 576void 577cpu_fork(p1, p2) 578 register struct proc *p1, *p2; 579{ 580 struct pcb *pcb2 = &p2->p_addr->u_pcb; 581 582#if NNPX > 0 583 /* Ensure that p1's pcb is up to date. */ 584 if (npxproc == p1) 585 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 586#endif 587 588 /* Copy p1's pcb. */ 589 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 590 591 /* 592 * Create a new fresh stack for the new process. 593 * Copy the trap frame for the return to user mode as if from a 594 * syscall. This copies the user mode register values. 595 */ 596 p2->p_md.md_regs = (struct trapframe *) 597 ((int)p2->p_addr + UPAGES * PAGE_SIZE) - 1; 598 *p2->p_md.md_regs = *p1->p_md.md_regs; 599 600 /* 601 * Set registers for trampoline to user mode. Leave space for the 602 * return address on stack. These are the kernel mode register values. 603 */ 604 pcb2->pcb_cr3 = vtophys(p2->p_vmspace->vm_pmap.pm_pdir); 605 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 606 pcb2->pcb_esi = (int)fork_return; 607 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 608 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 609 pcb2->pcb_ebx = (int)p2; 610 pcb2->pcb_eip = (int)fork_trampoline; 611 /* 612 * pcb2->pcb_ldt: duplicated below, if necessary. 613 * pcb2->pcb_ldt_len: cloned above. 614 * pcb2->pcb_savefpu: cloned above. 615 * pcb2->pcb_flags: cloned above (always 0 here?). 616 * pcb2->pcb_onfault: cloned above (always NULL here?). 617 */ 618 619#ifdef VM86 620 /* 621 * XXX don't copy the i/o pages. this should probably be fixed. 622 */ 623 pcb2->pcb_ext = 0; 624#endif 625 626#ifdef USER_LDT 627 /* Copy the LDT, if necessary. */ 628 if (pcb2->pcb_ldt != 0) { 629 union descriptor *new_ldt; 630 size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 631 632 new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 633 bcopy(pcb2->pcb_ldt, new_ldt, len); 634 pcb2->pcb_ldt = (caddr_t)new_ldt; 635 } 636#endif 637 638 /* 639 * Now, cpu_switch() can schedule the new process. 640 * pcb_esp is loaded pointing to the cpu_switch() stack frame 641 * containing the return address when exiting cpu_switch. 642 * This will normally be to proc_trampoline(), which will have 643 * %ebx loaded with the new proc's pointer. proc_trampoline() 644 * will set up a stack to call fork_return(p, frame); to complete 645 * the return to user-mode. 646 */ 647} 648 649/* 650 * Intercept the return address from a freshly forked process that has NOT 651 * been scheduled yet. 652 * 653 * This is needed to make kernel threads stay in kernel mode. 654 */ 655void 656cpu_set_fork_handler(p, func, arg) 657 struct proc *p; 658 void (*func) __P((void *)); 659 void *arg; 660{ 661 /* 662 * Note that the trap frame follows the args, so the function 663 * is really called like this: func(arg, frame); 664 */ 665 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 666 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 667} 668 669void 670cpu_exit(p) 671 register struct proc *p; 672{ 673#if defined(USER_LDT) || defined(VM86) 674 struct pcb *pcb = &p->p_addr->u_pcb; 675#endif 676 677#if NNPX > 0 678 npxexit(p); 679#endif /* NNPX */ 680#ifdef VM86 681 if (pcb->pcb_ext != 0) { 682 /* 683 * XXX do we need to move the TSS off the allocated pages 684 * before freeing them? (not done here) 685 */ 686 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 687 ctob(IOPAGES + 1)); 688 pcb->pcb_ext = 0; 689 } 690#endif 691#ifdef USER_LDT 692 if (pcb->pcb_ldt != 0) { 693 if (pcb == curpcb) 694 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 695 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 696 pcb->pcb_ldt_len * sizeof(union descriptor)); 697 pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 698 } 699#endif 700 cnt.v_swtch++; 701 cpu_switch(p); 702 panic("cpu_exit"); 703} 704 705void 706cpu_wait(p) 707 struct proc *p; 708{ 709 /* drop per-process resources */ 710 pmap_dispose_proc(p); 711 712 /* and clean-out the vmspace */ 713 vmspace_free(p->p_vmspace); 714} 715 716/* 717 * Dump the machine specific header information at the start of a core dump. 718 */ 719int 720cpu_coredump(p, vp, cred) 721 struct proc *p; 722 struct vnode *vp; 723 struct ucred *cred; 724{ 725 726 return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES), 727 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, 728 p)); 729} 730 731#ifdef notyet 732static void 733setredzone(pte, vaddr) 734 u_short *pte; 735 caddr_t vaddr; 736{ 737/* eventually do this by setting up an expand-down stack segment 738 for ss0: selector, allowing stack access down to top of u. 739 this means though that protection violations need to be handled 740 thru a double fault exception that must do an integral task 741 switch to a known good context, within which a dump can be 742 taken. a sensible scheme might be to save the initial context 743 used by sched (that has physical memory mapped 1:1 at bottom) 744 and take the dump while still in mapped mode */ 745} 746#endif 747 748/* 749 * Convert kernel VA to physical address 750 */ 751u_long 752kvtop(void *addr) 753{ 754 vm_offset_t va; 755 756 va = pmap_kextract((vm_offset_t)addr); 757 if (va == 0) 758 panic("kvtop: zero page frame"); 759 return((int)va); 760} 761 762/* 763 * Map an IO request into kernel virtual address space. 764 * 765 * All requests are (re)mapped into kernel VA space. 766 * Notice that we use b_bufsize for the size of the buffer 767 * to be mapped. b_bcount might be modified by the driver. 768 */ 769void 770vmapbuf(bp) 771 register struct buf *bp; 772{ 773 register caddr_t addr, v, kva; 774 vm_offset_t pa; 775 776 if ((bp->b_flags & B_PHYS) == 0) 777 panic("vmapbuf"); 778 779 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page(bp->b_data); 780 addr < bp->b_data + bp->b_bufsize; 781 addr += PAGE_SIZE, v += PAGE_SIZE) { 782 /* 783 * Do the vm_fault if needed; do the copy-on-write thing 784 * when reading stuff off device into memory. 785 */ 786 vm_fault_quick(addr, 787 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 788 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 789 if (pa == 0) 790 panic("vmapbuf: page not present"); 791 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 792 pmap_kenter((vm_offset_t) v, pa); 793 } 794 795 kva = bp->b_saveaddr; 796 bp->b_saveaddr = bp->b_data; 797 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 798} 799 800/* 801 * Free the io map PTEs associated with this IO operation. 802 * We also invalidate the TLB entries and restore the original b_addr. 803 */ 804void 805vunmapbuf(bp) 806 register struct buf *bp; 807{ 808 register caddr_t addr; 809 vm_offset_t pa; 810 811 if ((bp->b_flags & B_PHYS) == 0) 812 panic("vunmapbuf"); 813 814 for (addr = (caddr_t)trunc_page(bp->b_data); 815 addr < bp->b_data + bp->b_bufsize; 816 addr += PAGE_SIZE) { 817 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 818 pmap_kremove((vm_offset_t) addr); 819 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 820 } 821 822 bp->b_data = bp->b_saveaddr; 823} 824 825/* 826 * Force reset the processor by invalidating the entire address space! 827 */ 828void 829cpu_reset() 830{ 831 832#ifdef PC98 833 /* 834 * Attempt to do a CPU reset via CPU reset port. 835 */ 836 disable_intr(); 837 outb(0x37, 0x0f); /* SHUT0 = 0. */ 838 outb(0x37, 0x0b); /* SHUT1 = 0. */ 839 outb(0xf0, 0x00); /* Reset. */ 840#else 841 /* 842 * Attempt to do a CPU reset via the keyboard controller, 843 * do not turn of the GateA20, as any machine that fails 844 * to do the reset here would then end up in no man's land. 845 */ 846 847#if !defined(BROKEN_KEYBOARD_RESET) 848 outb(IO_KBD + 4, 0xFE); 849 DELAY(500000); /* wait 0.5 sec to see if that did it */ 850 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 851 DELAY(1000000); /* wait 1 sec for printf to complete */ 852#endif 853#endif /* PC98 */ 854 /* force a shutdown by unmapping entire address space ! */ 855 bzero((caddr_t) PTD, PAGE_SIZE); 856 857 /* "good night, sweet prince .... <THUNK!>" */ 858 invltlb(); 859 /* NOTREACHED */ 860 while(1); 861} 862 863/* 864 * Grow the user stack to allow for 'sp'. This version grows the stack in 865 * chunks of SGROWSIZ. 866 */ 867int 868grow(p, sp) 869 struct proc *p; 870 u_int sp; 871{ 872 unsigned int nss; 873 caddr_t v; 874 struct vmspace *vm = p->p_vmspace; 875 876 if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK) 877 return (1); 878 879 nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE); 880 881 if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur) 882 return (0); 883 884 if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT, 885 SGROWSIZ) < nss) { 886 int grow_amount; 887 /* 888 * If necessary, grow the VM that the stack occupies 889 * to allow for the rlimit. This allows us to not have 890 * to allocate all of the VM up-front in execve (which 891 * is expensive). 892 * Grow the VM by the amount requested rounded up to 893 * the nearest SGROWSIZ to provide for some hysteresis. 894 */ 895 grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ); 896 v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT, 897 SGROWSIZ) - grow_amount; 898 /* 899 * If there isn't enough room to extend by SGROWSIZ, then 900 * just extend to the maximum size 901 */ 902 if (v < vm->vm_maxsaddr) { 903 v = vm->vm_maxsaddr; 904 grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT); 905 } 906 if ((grow_amount == 0) || (vm_map_find(&vm->vm_map, NULL, 0, (vm_offset_t *)&v, 907 grow_amount, FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != KERN_SUCCESS)) { 908 return (0); 909 } 910 vm->vm_ssize += grow_amount >> PAGE_SHIFT; 911 } 912 913 return (1); 914} 915 916/* 917 * Implement the pre-zeroed page mechanism. 918 * This routine is called from the idle loop. 919 */ 920int 921vm_page_zero_idle() 922{ 923 static int free_rover; 924 vm_page_t m; 925 int s; 926 927#ifdef WRONG 928 if (cnt.v_free_count <= cnt.v_interrupt_free_min) 929 return (0); 930#endif 931 /* 932 * XXX 933 * We stop zeroing pages when there are sufficent prezeroed pages. 934 * This threshold isn't really needed, except we want to 935 * bypass unneeded calls to vm_page_list_find, and the 936 * associated cache flush and latency. The pre-zero will 937 * still be called when there are significantly more 938 * non-prezeroed pages than zeroed pages. The threshold 939 * of half the number of reserved pages is arbitrary, but 940 * approximately the right amount. Eventually, we should 941 * perhaps interrupt the zero operation when a process 942 * is found to be ready to run. 943 */ 944 if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2) 945 return (0); 946#ifdef SMP 947 get_mplock(); 948#endif 949 s = splvm(); 950 enable_intr(); 951 m = vm_page_list_find(PQ_FREE, free_rover); 952 if (m != NULL) { 953 --(*vm_page_queues[m->queue].lcnt); 954 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 955 m->queue = PQ_NONE; 956 splx(s); 957#ifdef SMP 958 rel_mplock(); 959#endif 960 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 961#ifdef SMP 962 get_mplock(); 963#endif 964 (void)splvm(); 965 m->queue = PQ_ZERO + m->pc; 966 ++(*vm_page_queues[m->queue].lcnt); 967 TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, pageq); 968 free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK; 969 ++vm_page_zero_count; 970 } 971 splx(s); 972 disable_intr(); 973#ifdef SMP 974 rel_mplock(); 975#endif 976 return (1); 977} 978 979/* 980 * Software interrupt handler for queued VM system processing. 981 */ 982void 983swi_vm() 984{ 985 if (busdma_swi_pending != 0) 986 busdma_swi(); 987} 988 989/* 990 * Tell whether this address is in some physical memory region. 991 * Currently used by the kernel coredump code in order to avoid 992 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 993 * or other unpredictable behaviour. 994 */ 995 996#include "isa.h" 997 998int 999is_physical_memory(addr) 1000 vm_offset_t addr; 1001{ 1002 1003#if NISA > 0 1004 /* The ISA ``memory hole''. */ 1005 if (addr >= 0xa0000 && addr < 0x100000) 1006 return 0; 1007#endif 1008 1009 /* 1010 * stuff other tests for known memory-mapped devices (PCI?) 1011 * here 1012 */ 1013 1014 return 1; 1015} 1016