vm_machdep.c revision 34569
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 1.102 1998/03/12 09:55:57 bde 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#include <sys/kernel.h> 57#include <sys/sysctl.h> 58 59#include <machine/clock.h> 60#include <machine/cpu.h> 61#include <machine/md_var.h> 62#ifdef SMP 63#include <machine/smp.h> 64#endif 65 66#include <vm/vm.h> 67#include <vm/vm_param.h> 68#include <vm/vm_prot.h> 69#include <sys/lock.h> 70#include <vm/vm_kern.h> 71#include <vm/vm_page.h> 72#include <vm/vm_map.h> 73#include <vm/vm_extern.h> 74 75#include <sys/user.h> 76 77#ifdef PC98 78#include <pc98/pc98/pc98.h> 79#else 80#include <i386/isa/isa.h> 81#endif 82 83#ifdef BOUNCE_BUFFERS 84static vm_offset_t 85 vm_bounce_kva __P((int size, int waitok)); 86static void vm_bounce_kva_free __P((vm_offset_t addr, vm_offset_t size, 87 int now)); 88static vm_offset_t 89 vm_bounce_page_find __P((int count)); 90static void vm_bounce_page_free __P((vm_offset_t pa, int count)); 91 92static volatile int kvasfreecnt; 93 94caddr_t bouncememory; 95static int bpwait; 96static vm_offset_t *bouncepa; 97static int bmwait, bmfreeing; 98 99#define BITS_IN_UNSIGNED (8*sizeof(unsigned)) 100static int bounceallocarraysize; 101static unsigned *bounceallocarray; 102static int bouncefree; 103 104#if defined(PC98) && defined (EPSON_BOUNCEDMA) 105#define SIXTEENMEG (3840*4096) /* 15MB boundary */ 106#else 107#define SIXTEENMEG (4096*4096) 108#endif 109#define MAXBKVA 1024 110int maxbkva = MAXBKVA*PAGE_SIZE; 111 112/* special list that can be used at interrupt time for eventual kva free */ 113static struct kvasfree { 114 vm_offset_t addr; 115 vm_offset_t size; 116} kvaf[MAXBKVA]; 117 118/* 119 * get bounce buffer pages (count physically contiguous) 120 * (only 1 inplemented now) 121 */ 122static vm_offset_t 123vm_bounce_page_find(count) 124 int count; 125{ 126 int bit; 127 int s,i; 128 129 if (count != 1) 130 panic("vm_bounce_page_find -- no support for > 1 page yet!!!"); 131 132 s = splbio(); 133retry: 134 for (i = 0; i < bounceallocarraysize; i++) { 135 if (bounceallocarray[i] != 0xffffffff) { 136 bit = ffs(~bounceallocarray[i]); 137 if (bit) { 138 bounceallocarray[i] |= 1 << (bit - 1) ; 139 bouncefree -= count; 140 splx(s); 141 return bouncepa[(i * BITS_IN_UNSIGNED + (bit - 1))]; 142 } 143 } 144 } 145 bpwait = 1; 146 tsleep((caddr_t) &bounceallocarray, PRIBIO, "bncwai", 0); 147 goto retry; 148} 149 150static void 151vm_bounce_kva_free(addr, size, now) 152 vm_offset_t addr; 153 vm_offset_t size; 154 int now; 155{ 156 int s = splbio(); 157 kvaf[kvasfreecnt].addr = addr; 158 kvaf[kvasfreecnt].size = size; 159 ++kvasfreecnt; 160 if( now) { 161 /* 162 * this will do wakeups 163 */ 164 vm_bounce_kva(0,0); 165 } else { 166 if (bmwait) { 167 /* 168 * if anyone is waiting on the bounce-map, then wakeup 169 */ 170 wakeup((caddr_t) io_map); 171 bmwait = 0; 172 } 173 } 174 splx(s); 175} 176 177/* 178 * free count bounce buffer pages 179 */ 180static void 181vm_bounce_page_free(pa, count) 182 vm_offset_t pa; 183 int count; 184{ 185 int allocindex; 186 int index; 187 int bit; 188 189 if (count != 1) 190 panic("vm_bounce_page_free -- no support for > 1 page yet!!!"); 191 192 for(index=0;index<bouncepages;index++) { 193 if( pa == bouncepa[index]) 194 break; 195 } 196 197 if( index == bouncepages) 198 panic("vm_bounce_page_free: invalid bounce buffer"); 199 200 allocindex = index / BITS_IN_UNSIGNED; 201 bit = index % BITS_IN_UNSIGNED; 202 203 bounceallocarray[allocindex] &= ~(1 << bit); 204 205 bouncefree += count; 206 if (bpwait) { 207 bpwait = 0; 208 wakeup((caddr_t) &bounceallocarray); 209 } 210} 211 212/* 213 * allocate count bounce buffer kva pages 214 */ 215static vm_offset_t 216vm_bounce_kva(size, waitok) 217 int size; 218 int waitok; 219{ 220 int i; 221 vm_offset_t kva = 0; 222 vm_offset_t off; 223 int s = splbio(); 224more: 225 if (!bmfreeing && kvasfreecnt) { 226 bmfreeing = 1; 227 for (i = 0; i < kvasfreecnt; i++) { 228 for(off=0;off<kvaf[i].size;off+=PAGE_SIZE) { 229 pmap_kremove( kvaf[i].addr + off); 230 } 231 kmem_free_wakeup(io_map, kvaf[i].addr, 232 kvaf[i].size); 233 } 234 kvasfreecnt = 0; 235 bmfreeing = 0; 236 if( bmwait) { 237 bmwait = 0; 238 wakeup( (caddr_t) io_map); 239 } 240 } 241 242 if( size == 0) { 243 splx(s); 244 return 0; 245 } 246 247 if ((kva = kmem_alloc_pageable(io_map, size)) == 0) { 248 if( !waitok) { 249 splx(s); 250 return 0; 251 } 252 bmwait = 1; 253 tsleep((caddr_t) io_map, PRIBIO, "bmwait", 0); 254 goto more; 255 } 256 splx(s); 257 return kva; 258} 259 260/* 261 * same as vm_bounce_kva -- but really allocate (but takes pages as arg) 262 */ 263vm_offset_t 264vm_bounce_kva_alloc(count) 265int count; 266{ 267 int i; 268 vm_offset_t kva; 269 vm_offset_t pa; 270 if( bouncepages == 0) { 271 kva = (vm_offset_t) malloc(count*PAGE_SIZE, M_TEMP, M_WAITOK); 272 return kva; 273 } 274 kva = vm_bounce_kva(count*PAGE_SIZE, 1); 275 for(i=0;i<count;i++) { 276 pa = vm_bounce_page_find(1); 277 pmap_kenter(kva + i * PAGE_SIZE, pa); 278 } 279 return kva; 280} 281 282/* 283 * same as vm_bounce_kva_free -- but really free 284 */ 285void 286vm_bounce_kva_alloc_free(kva, count) 287 vm_offset_t kva; 288 int count; 289{ 290 int i; 291 vm_offset_t pa; 292 if( bouncepages == 0) { 293 free((caddr_t) kva, M_TEMP); 294 return; 295 } 296 for(i = 0; i < count; i++) { 297 pa = pmap_kextract(kva + i * PAGE_SIZE); 298 vm_bounce_page_free(pa, 1); 299 } 300 vm_bounce_kva_free(kva, count*PAGE_SIZE, 0); 301} 302 303/* 304 * do the things necessary to the struct buf to implement 305 * bounce buffers... inserted before the disk sort 306 */ 307void 308vm_bounce_alloc(bp) 309 struct buf *bp; 310{ 311 int countvmpg; 312 vm_offset_t vastart, vaend; 313 vm_offset_t vapstart, vapend; 314 vm_offset_t va, kva; 315 vm_offset_t pa; 316 int dobounceflag = 0; 317 int i; 318 319 if (bouncepages == 0) 320 return; 321 322 if (bp->b_flags & B_BOUNCE) { 323 printf("vm_bounce_alloc: called recursively???\n"); 324 return; 325 } 326 327 if (bp->b_bufsize < bp->b_bcount) { 328 printf( 329 "vm_bounce_alloc: b_bufsize(0x%lx) < b_bcount(0x%lx) !!\n", 330 bp->b_bufsize, bp->b_bcount); 331 panic("vm_bounce_alloc"); 332 } 333 334/* 335 * This is not really necessary 336 * if( bp->b_bufsize != bp->b_bcount) { 337 * printf("size: %d, count: %d\n", bp->b_bufsize, bp->b_bcount); 338 * } 339 */ 340 341 342 vastart = (vm_offset_t) bp->b_data; 343 vaend = (vm_offset_t) bp->b_data + bp->b_bufsize; 344 345 vapstart = trunc_page(vastart); 346 vapend = round_page(vaend); 347 countvmpg = (vapend - vapstart) / PAGE_SIZE; 348 349/* 350 * if any page is above 16MB, then go into bounce-buffer mode 351 */ 352 va = vapstart; 353 for (i = 0; i < countvmpg; i++) { 354 pa = pmap_kextract(va); 355 if (pa >= SIXTEENMEG) 356 ++dobounceflag; 357 if( pa == 0) 358 panic("vm_bounce_alloc: Unmapped page"); 359 va += PAGE_SIZE; 360 } 361 if (dobounceflag == 0) 362 return; 363 364 if (bouncepages < dobounceflag) 365 panic("Not enough bounce buffers!!!"); 366 367/* 368 * allocate a replacement kva for b_addr 369 */ 370 kva = vm_bounce_kva(countvmpg*PAGE_SIZE, 1); 371#if 0 372 printf("%s: vapstart: %x, vapend: %x, countvmpg: %d, kva: %x ", 373 (bp->b_flags & B_READ) ? "read":"write", 374 vapstart, vapend, countvmpg, kva); 375#endif 376 va = vapstart; 377 for (i = 0; i < countvmpg; i++) { 378 pa = pmap_kextract(va); 379 if (pa >= SIXTEENMEG) { 380 /* 381 * allocate a replacement page 382 */ 383 vm_offset_t bpa = vm_bounce_page_find(1); 384 pmap_kenter(kva + (PAGE_SIZE * i), bpa); 385#if 0 386 printf("r(%d): (%x,%x,%x) ", i, va, pa, bpa); 387#endif 388 /* 389 * if we are writing, the copy the data into the page 390 */ 391 if ((bp->b_flags & B_READ) == 0) { 392 bcopy((caddr_t) va, (caddr_t) kva + (PAGE_SIZE * i), PAGE_SIZE); 393 } 394 } else { 395 /* 396 * use original page 397 */ 398 pmap_kenter(kva + (PAGE_SIZE * i), pa); 399 } 400 va += PAGE_SIZE; 401 } 402 403/* 404 * flag the buffer as being bounced 405 */ 406 bp->b_flags |= B_BOUNCE; 407/* 408 * save the original buffer kva 409 */ 410 bp->b_savekva = bp->b_data; 411/* 412 * put our new kva into the buffer (offset by original offset) 413 */ 414 bp->b_data = (caddr_t) (((vm_offset_t) kva) | 415 ((vm_offset_t) bp->b_savekva & PAGE_MASK)); 416#if 0 417 printf("b_savekva: %x, newva: %x\n", bp->b_savekva, bp->b_data); 418#endif 419 return; 420} 421 422/* 423 * hook into biodone to free bounce buffer 424 */ 425void 426vm_bounce_free(bp) 427 struct buf *bp; 428{ 429 int i; 430 vm_offset_t origkva, bouncekva, bouncekvaend; 431 432/* 433 * if this isn't a bounced buffer, then just return 434 */ 435 if ((bp->b_flags & B_BOUNCE) == 0) 436 return; 437 438/* 439 * This check is not necessary 440 * if (bp->b_bufsize != bp->b_bcount) { 441 * printf("vm_bounce_free: b_bufsize=%d, b_bcount=%d\n", 442 * bp->b_bufsize, bp->b_bcount); 443 * } 444 */ 445 446 origkva = (vm_offset_t) bp->b_savekva; 447 bouncekva = (vm_offset_t) bp->b_data; 448/* 449 printf("free: %d ", bp->b_bufsize); 450*/ 451 452/* 453 * check every page in the kva space for b_addr 454 */ 455 for (i = 0; i < bp->b_bufsize; ) { 456 vm_offset_t mybouncepa; 457 vm_offset_t copycount; 458 459 copycount = round_page(bouncekva + 1) - bouncekva; 460 mybouncepa = pmap_kextract(trunc_page(bouncekva)); 461 462/* 463 * if this is a bounced pa, then process as one 464 */ 465 if ( mybouncepa != pmap_kextract( trunc_page( origkva))) { 466 vm_offset_t tocopy = copycount; 467 if (i + tocopy > bp->b_bufsize) 468 tocopy = bp->b_bufsize - i; 469/* 470 * if this is a read, then copy from bounce buffer into original buffer 471 */ 472 if (bp->b_flags & B_READ) 473 bcopy((caddr_t) bouncekva, (caddr_t) origkva, tocopy); 474/* 475 * free the bounce allocation 476 */ 477 478/* 479 printf("(kva: %x, pa: %x)", bouncekva, mybouncepa); 480*/ 481 vm_bounce_page_free(mybouncepa, 1); 482 } 483 484 origkva += copycount; 485 bouncekva += copycount; 486 i += copycount; 487 } 488 489/* 490 printf("\n"); 491*/ 492/* 493 * add the old kva into the "to free" list 494 */ 495 496 bouncekva= trunc_page((vm_offset_t) bp->b_data); 497 bouncekvaend= round_page((vm_offset_t)bp->b_data + bp->b_bufsize); 498 499/* 500 printf("freeva: %d\n", (bouncekvaend - bouncekva) / PAGE_SIZE); 501*/ 502 vm_bounce_kva_free( bouncekva, (bouncekvaend - bouncekva), 0); 503 bp->b_data = bp->b_savekva; 504 bp->b_savekva = 0; 505 bp->b_flags &= ~B_BOUNCE; 506 507 return; 508} 509 510 511/* 512 * init the bounce buffer system 513 */ 514void 515vm_bounce_init() 516{ 517 int i; 518 519 kvasfreecnt = 0; 520 521 if (bouncepages == 0) 522 return; 523 524 bounceallocarraysize = (bouncepages + BITS_IN_UNSIGNED - 1) / BITS_IN_UNSIGNED; 525 bounceallocarray = malloc(bounceallocarraysize * sizeof(unsigned), M_TEMP, M_NOWAIT); 526 527 if (!bounceallocarray) 528 panic("Cannot allocate bounce resource array"); 529 530 bouncepa = malloc(bouncepages * sizeof(vm_offset_t), M_TEMP, M_NOWAIT); 531 if (!bouncepa) 532 panic("Cannot allocate physical memory array"); 533 534 for(i=0;i<bounceallocarraysize;i++) { 535 bounceallocarray[i] = 0xffffffff; 536 } 537 538 for(i=0;i<bouncepages;i++) { 539 vm_offset_t pa; 540 if( (pa = pmap_kextract((vm_offset_t) bouncememory + i * PAGE_SIZE)) >= SIXTEENMEG) { 541 printf("vm_bounce_init: bounce memory out of range -- bounce disabled\n"); 542 free(bounceallocarray, M_TEMP); 543 bounceallocarray = NULL; 544 free(bouncepa, M_TEMP); 545 bouncepa = NULL; 546 bouncepages = 0; 547 break; 548 } 549 if( pa == 0) 550 panic("bounce memory not resident"); 551 bouncepa[i] = pa; 552 bounceallocarray[i/(8*sizeof(int))] &= ~(1<<(i%(8*sizeof(int)))); 553 } 554 bouncefree = bouncepages; 555 556} 557#endif /* BOUNCE_BUFFERS */ 558 559/* 560 * quick version of vm_fault 561 */ 562void 563vm_fault_quick(v, prot) 564 caddr_t v; 565 int prot; 566{ 567 if (prot & VM_PROT_WRITE) 568 subyte(v, fubyte(v)); 569 else 570 fubyte(v); 571} 572 573/* 574 * Finish a fork operation, with process p2 nearly set up. 575 * Copy and update the pcb, set up the stack so that the child 576 * ready to run and return to user mode. 577 */ 578void 579cpu_fork(p1, p2) 580 register struct proc *p1, *p2; 581{ 582 struct pcb *pcb2 = &p2->p_addr->u_pcb; 583 584#if NNPX > 0 585 /* Ensure that p1's pcb is up to date. */ 586 if (npxproc == p1) 587 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 588#endif 589 590 /* Copy p1's pcb. */ 591 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 592 593 /* 594 * Create a new fresh stack for the new process. 595 * Copy the trap frame for the return to user mode as if from a 596 * syscall. This copies the user mode register values. 597 */ 598 p2->p_md.md_regs = (struct trapframe *) 599 ((int)p2->p_addr + UPAGES * PAGE_SIZE) - 1; 600 *p2->p_md.md_regs = *p1->p_md.md_regs; 601 602 /* 603 * Set registers for trampoline to user mode. Leave space for the 604 * return address on stack. These are the kernel mode register values. 605 */ 606 pcb2->pcb_cr3 = vtophys(p2->p_vmspace->vm_pmap.pm_pdir); 607 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 608 pcb2->pcb_esi = (int)fork_return; 609 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 610 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 611 pcb2->pcb_ebx = (int)p2; 612 pcb2->pcb_eip = (int)fork_trampoline; 613 /* 614 * pcb2->pcb_ldt: duplicated below, if necessary. 615 * pcb2->pcb_ldt_len: cloned above. 616 * pcb2->pcb_savefpu: cloned above. 617 * pcb2->pcb_flags: cloned above (always 0 here?). 618 * pcb2->pcb_onfault: cloned above (always NULL here?). 619 */ 620 621#ifdef VM86 622 /* 623 * XXX don't copy the i/o pages. this should probably be fixed. 624 */ 625 pcb2->pcb_ext = 0; 626#endif 627 628#ifdef USER_LDT 629 /* Copy the LDT, if necessary. */ 630 if (pcb2->pcb_ldt != 0) { 631 union descriptor *new_ldt; 632 size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 633 634 new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 635 bcopy(pcb2->pcb_ldt, new_ldt, len); 636 pcb2->pcb_ldt = (caddr_t)new_ldt; 637 } 638#endif 639 640 /* 641 * Now, cpu_switch() can schedule the new process. 642 * pcb_esp is loaded pointing to the cpu_switch() stack frame 643 * containing the return address when exiting cpu_switch. 644 * This will normally be to proc_trampoline(), which will have 645 * %ebx loaded with the new proc's pointer. proc_trampoline() 646 * will set up a stack to call fork_return(p, frame); to complete 647 * the return to user-mode. 648 */ 649} 650 651/* 652 * Intercept the return address from a freshly forked process that has NOT 653 * been scheduled yet. 654 * 655 * This is needed to make kernel threads stay in kernel mode. 656 */ 657void 658cpu_set_fork_handler(p, func, arg) 659 struct proc *p; 660 void (*func) __P((void *)); 661 void *arg; 662{ 663 /* 664 * Note that the trap frame follows the args, so the function 665 * is really called like this: func(arg, frame); 666 */ 667 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 668 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 669} 670 671void 672cpu_exit(p) 673 register struct proc *p; 674{ 675#if defined(USER_LDT) || defined(VM86) 676 struct pcb *pcb = &p->p_addr->u_pcb; 677#endif 678 679#if NNPX > 0 680 npxexit(p); 681#endif /* NNPX */ 682#ifdef VM86 683 if (pcb->pcb_ext != 0) { 684 /* 685 * XXX do we need to move the TSS off the allocated pages 686 * before freeing them? (not done here) 687 */ 688 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 689 ctob(IOPAGES + 1)); 690 pcb->pcb_ext = 0; 691 } 692#endif 693#ifdef USER_LDT 694 if (pcb->pcb_ldt != 0) { 695 if (pcb == curpcb) 696 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 697 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 698 pcb->pcb_ldt_len * sizeof(union descriptor)); 699 pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 700 } 701#endif 702 cnt.v_swtch++; 703 cpu_switch(p); 704 panic("cpu_exit"); 705} 706 707void 708cpu_wait(p) 709 struct proc *p; 710{ 711 /* drop per-process resources */ 712 pmap_dispose_proc(p); 713 714 /* and clean-out the vmspace */ 715 vmspace_free(p->p_vmspace); 716} 717 718/* 719 * Dump the machine specific header information at the start of a core dump. 720 */ 721int 722cpu_coredump(p, vp, cred) 723 struct proc *p; 724 struct vnode *vp; 725 struct ucred *cred; 726{ 727 728 return (vn_rdwr(UIO_WRITE, vp, (caddr_t) p->p_addr, ctob(UPAGES), 729 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, 730 p)); 731} 732 733#ifdef notyet 734static void 735setredzone(pte, vaddr) 736 u_short *pte; 737 caddr_t vaddr; 738{ 739/* eventually do this by setting up an expand-down stack segment 740 for ss0: selector, allowing stack access down to top of u. 741 this means though that protection violations need to be handled 742 thru a double fault exception that must do an integral task 743 switch to a known good context, within which a dump can be 744 taken. a sensible scheme might be to save the initial context 745 used by sched (that has physical memory mapped 1:1 at bottom) 746 and take the dump while still in mapped mode */ 747} 748#endif 749 750/* 751 * Convert kernel VA to physical address 752 */ 753u_long 754kvtop(void *addr) 755{ 756 vm_offset_t va; 757 758 va = pmap_kextract((vm_offset_t)addr); 759 if (va == 0) 760 panic("kvtop: zero page frame"); 761 return((int)va); 762} 763 764/* 765 * Map an IO request into kernel virtual address space. 766 * 767 * All requests are (re)mapped into kernel VA space. 768 * Notice that we use b_bufsize for the size of the buffer 769 * to be mapped. b_bcount might be modified by the driver. 770 */ 771void 772vmapbuf(bp) 773 register struct buf *bp; 774{ 775 register caddr_t addr, v, kva; 776 vm_offset_t pa; 777 778 if ((bp->b_flags & B_PHYS) == 0) 779 panic("vmapbuf"); 780 781 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page(bp->b_data); 782 addr < bp->b_data + bp->b_bufsize; 783 addr += PAGE_SIZE, v += PAGE_SIZE) { 784 /* 785 * Do the vm_fault if needed; do the copy-on-write thing 786 * when reading stuff off device into memory. 787 */ 788 vm_fault_quick(addr, 789 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 790 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 791 if (pa == 0) 792 panic("vmapbuf: page not present"); 793 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 794 pmap_kenter((vm_offset_t) v, pa); 795 } 796 797 kva = bp->b_saveaddr; 798 bp->b_saveaddr = bp->b_data; 799 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 800} 801 802/* 803 * Free the io map PTEs associated with this IO operation. 804 * We also invalidate the TLB entries and restore the original b_addr. 805 */ 806void 807vunmapbuf(bp) 808 register struct buf *bp; 809{ 810 register caddr_t addr; 811 vm_offset_t pa; 812 813 if ((bp->b_flags & B_PHYS) == 0) 814 panic("vunmapbuf"); 815 816 for (addr = (caddr_t)trunc_page(bp->b_data); 817 addr < bp->b_data + bp->b_bufsize; 818 addr += PAGE_SIZE) { 819 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 820 pmap_kremove((vm_offset_t) addr); 821 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 822 } 823 824 bp->b_data = bp->b_saveaddr; 825} 826 827/* 828 * Force reset the processor by invalidating the entire address space! 829 */ 830void 831cpu_reset() 832{ 833 834#ifdef PC98 835 /* 836 * Attempt to do a CPU reset via CPU reset port. 837 */ 838 disable_intr(); 839 outb(0x37, 0x0f); /* SHUT0 = 0. */ 840 outb(0x37, 0x0b); /* SHUT1 = 0. */ 841 outb(0xf0, 0x00); /* Reset. */ 842#else 843 /* 844 * Attempt to do a CPU reset via the keyboard controller, 845 * do not turn of the GateA20, as any machine that fails 846 * to do the reset here would then end up in no man's land. 847 */ 848 849#if !defined(BROKEN_KEYBOARD_RESET) 850 outb(IO_KBD + 4, 0xFE); 851 DELAY(500000); /* wait 0.5 sec to see if that did it */ 852 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 853 DELAY(1000000); /* wait 1 sec for printf to complete */ 854#endif 855#endif /* PC98 */ 856 /* force a shutdown by unmapping entire address space ! */ 857 bzero((caddr_t) PTD, PAGE_SIZE); 858 859 /* "good night, sweet prince .... <THUNK!>" */ 860 invltlb(); 861 /* NOTREACHED */ 862 while(1); 863} 864 865/* 866 * Grow the user stack to allow for 'sp'. This version grows the stack in 867 * chunks of SGROWSIZ. 868 */ 869int 870grow(p, sp) 871 struct proc *p; 872 u_int sp; 873{ 874 unsigned int nss; 875 caddr_t v; 876 struct vmspace *vm = p->p_vmspace; 877 878 if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK) 879 return (1); 880 881 nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE); 882 883 if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur) 884 return (0); 885 886 if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT, 887 SGROWSIZ) < nss) { 888 int grow_amount; 889 /* 890 * If necessary, grow the VM that the stack occupies 891 * to allow for the rlimit. This allows us to not have 892 * to allocate all of the VM up-front in execve (which 893 * is expensive). 894 * Grow the VM by the amount requested rounded up to 895 * the nearest SGROWSIZ to provide for some hysteresis. 896 */ 897 grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ); 898 v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT, 899 SGROWSIZ) - grow_amount; 900 /* 901 * If there isn't enough room to extend by SGROWSIZ, then 902 * just extend to the maximum size 903 */ 904 if (v < vm->vm_maxsaddr) { 905 v = vm->vm_maxsaddr; 906 grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT); 907 } 908 if ((grow_amount == 0) || (vm_map_find(&vm->vm_map, NULL, 0, (vm_offset_t *)&v, 909 grow_amount, FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != KERN_SUCCESS)) { 910 return (0); 911 } 912 vm->vm_ssize += grow_amount >> PAGE_SHIFT; 913 } 914 915 return (1); 916} 917 918static int cnt_prezero; 919 920SYSCTL_INT(_machdep, OID_AUTO, cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 921 922/* 923 * Implement the pre-zeroed page mechanism. 924 * This routine is called from the idle loop. 925 */ 926int 927vm_page_zero_idle() 928{ 929 static int free_rover; 930 vm_page_t m; 931 int s; 932 933 /* 934 * XXX 935 * We stop zeroing pages when there are sufficent prezeroed pages. 936 * This threshold isn't really needed, except we want to 937 * bypass unneeded calls to vm_page_list_find, and the 938 * associated cache flush and latency. The pre-zero will 939 * still be called when there are significantly more 940 * non-prezeroed pages than zeroed pages. The threshold 941 * of half the number of reserved pages is arbitrary, but 942 * approximately the right amount. Eventually, we should 943 * perhaps interrupt the zero operation when a process 944 * is found to be ready to run. 945 */ 946 if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2) 947 return (0); 948#ifdef SMP 949 if (try_mplock()) { 950#endif 951 s = splvm(); 952 __asm __volatile("sti" : : : "memory"); 953 m = vm_page_list_find(PQ_FREE, free_rover); 954 if (m != NULL) { 955 --(*vm_page_queues[m->queue].lcnt); 956 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 957 m->queue = PQ_NONE; 958 splx(s); 959#if 0 960 rel_mplock(); 961#endif 962 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 963#if 0 964 get_mplock(); 965#endif 966 (void)splvm(); 967 m->queue = PQ_ZERO + m->pc; 968 ++(*vm_page_queues[m->queue].lcnt); 969 TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, 970 pageq); 971 free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK; 972 ++vm_page_zero_count; 973 ++cnt_prezero; 974 } 975 splx(s); 976 __asm __volatile("cli" : : : "memory"); 977#ifdef SMP 978 rel_mplock(); 979#endif 980 return (1); 981#ifdef SMP 982 } 983#endif 984 return (0); 985} 986 987/* 988 * Software interrupt handler for queued VM system processing. 989 */ 990void 991swi_vm() 992{ 993 if (busdma_swi_pending != 0) 994 busdma_swi(); 995} 996 997/* 998 * Tell whether this address is in some physical memory region. 999 * Currently used by the kernel coredump code in order to avoid 1000 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 1001 * or other unpredictable behaviour. 1002 */ 1003 1004#include "isa.h" 1005 1006int 1007is_physical_memory(addr) 1008 vm_offset_t addr; 1009{ 1010 1011#if NISA > 0 1012 /* The ISA ``memory hole''. */ 1013 if (addr >= 0xa0000 && addr < 0x100000) 1014 return 0; 1015#endif 1016 1017 /* 1018 * stuff other tests for known memory-mapped devices (PCI?) 1019 * here 1020 */ 1021 1022 return 1; 1023} 1024