vm_machdep.c revision 1307
1/*- 2 * Copyright (c) 1982, 1986 The Regents of the University of California. 3 * Copyright (c) 1989, 1990 William Jolitz 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * the Systems Programming Group of the University of Utah Computer 8 * Science Department, and William Jolitz. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 39 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 40 * $Id: vm_machdep.c,v 1.14 1994/03/23 09:15:06 davidg Exp $ 41 */ 42 43#include "npx.h" 44#include "param.h" 45#include "systm.h" 46#include "proc.h" 47#include "malloc.h" 48#include "buf.h" 49#include "user.h" 50 51#include "../include/cpu.h" 52 53#include "vm/vm.h" 54#include "vm/vm_kern.h" 55 56#ifndef NOBOUNCE 57 58caddr_t bouncememory; 59vm_offset_t bouncepa, bouncepaend; 60int bouncepages, bpwait; 61vm_map_t bounce_map; 62int bmwait, bmfreeing; 63 64#define BITS_IN_UNSIGNED (8*sizeof(unsigned)) 65int bounceallocarraysize; 66unsigned *bounceallocarray; 67int bouncefree; 68 69#define SIXTEENMEG (4096*4096) 70#define MAXBKVA 512 71 72/* special list that can be used at interrupt time for eventual kva free */ 73struct kvasfree { 74 vm_offset_t addr; 75 vm_offset_t size; 76} kvaf[MAXBKVA]; 77 78int kvasfreecnt; 79 80/* 81 * get bounce buffer pages (count physically contiguous) 82 * (only 1 inplemented now) 83 */ 84vm_offset_t 85vm_bounce_page_find(count) 86 int count; 87{ 88 int bit; 89 int s,i; 90 91 if (count != 1) 92 panic("vm_bounce_page_find -- no support for > 1 page yet!!!"); 93 94 s = splbio(); 95retry: 96 for (i = 0; i < bounceallocarraysize; i++) { 97 if (bounceallocarray[i] != 0xffffffff) { 98 if (bit = ffs(~bounceallocarray[i])) { 99 bounceallocarray[i] |= 1 << (bit - 1) ; 100 bouncefree -= count; 101 splx(s); 102 return bouncepa + (i * BITS_IN_UNSIGNED + (bit - 1)) * NBPG; 103 } 104 } 105 } 106 bpwait = 1; 107 tsleep((caddr_t) &bounceallocarray, PRIBIO, "bncwai", 0); 108 goto retry; 109} 110 111/* 112 * free count bounce buffer pages 113 */ 114void 115vm_bounce_page_free(pa, count) 116 vm_offset_t pa; 117 int count; 118{ 119 int allocindex; 120 int index; 121 int bit; 122 123 if (count != 1) 124 panic("vm_bounce_page_free -- no support for > 1 page yet!!!\n"); 125 126 index = (pa - bouncepa) / NBPG; 127 128 if ((index < 0) || (index >= bouncepages)) 129 panic("vm_bounce_page_free -- bad index\n"); 130 131 allocindex = index / BITS_IN_UNSIGNED; 132 bit = index % BITS_IN_UNSIGNED; 133 134 bounceallocarray[allocindex] &= ~(1 << bit); 135 136 bouncefree += count; 137 if (bpwait) { 138 bpwait = 0; 139 wakeup((caddr_t) &bounceallocarray); 140 } 141} 142 143/* 144 * allocate count bounce buffer kva pages 145 */ 146vm_offset_t 147vm_bounce_kva(count) 148 int count; 149{ 150 int tofree; 151 int i; 152 int startfree; 153 vm_offset_t kva; 154 int s = splbio(); 155 startfree = 0; 156more: 157 if (!bmfreeing && (tofree = kvasfreecnt)) { 158 bmfreeing = 1; 159more1: 160 for (i = startfree; i < kvasfreecnt; i++) { 161 pmap_remove(kernel_pmap, 162 kvaf[i].addr, kvaf[i].addr + kvaf[i].size); 163 kmem_free_wakeup(bounce_map, kvaf[i].addr, 164 kvaf[i].size); 165 } 166 if (kvasfreecnt != tofree) { 167 startfree = i; 168 bmfreeing = 0; 169 goto more; 170 } 171 kvasfreecnt = 0; 172 bmfreeing = 0; 173 } 174 175 if (!(kva = kmem_alloc_pageable(bounce_map, count * NBPG))) { 176 bmwait = 1; 177 tsleep((caddr_t) bounce_map, PRIBIO, "bmwait", 0); 178 goto more; 179 } 180 181 splx(s); 182 183 return kva; 184} 185 186/* 187 * init the bounce buffer system 188 */ 189void 190vm_bounce_init() 191{ 192 vm_offset_t minaddr, maxaddr; 193 194 if (bouncepages == 0) 195 return; 196 197 bounceallocarraysize = (bouncepages + BITS_IN_UNSIGNED - 1) / BITS_IN_UNSIGNED; 198 bounceallocarray = malloc(bounceallocarraysize * sizeof(unsigned), M_TEMP, M_NOWAIT); 199 200 if (!bounceallocarray) 201 panic("Cannot allocate bounce resource array\n"); 202 203 bzero(bounceallocarray, bounceallocarraysize * sizeof(long)); 204 205 bounce_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr, MAXBKVA * NBPG, FALSE); 206 207 bouncepa = pmap_kextract((vm_offset_t) bouncememory); 208 bouncepaend = bouncepa + bouncepages * NBPG; 209 bouncefree = bouncepages; 210 kvasfreecnt = 0; 211} 212 213/* 214 * do the things necessary to the struct buf to implement 215 * bounce buffers... inserted before the disk sort 216 */ 217void 218vm_bounce_alloc(bp) 219 struct buf *bp; 220{ 221 int countvmpg; 222 vm_offset_t vastart, vaend; 223 vm_offset_t vapstart, vapend; 224 vm_offset_t va, kva; 225 vm_offset_t pa; 226 int dobounceflag = 0; 227 int bounceindex; 228 int i; 229 int s; 230 231 if (bouncepages == 0) 232 return; 233 234 vastart = (vm_offset_t) bp->b_un.b_addr; 235 vaend = (vm_offset_t) bp->b_un.b_addr + bp->b_bcount; 236 237 vapstart = i386_trunc_page(vastart); 238 vapend = i386_round_page(vaend); 239 countvmpg = (vapend - vapstart) / NBPG; 240 241/* 242 * if any page is above 16MB, then go into bounce-buffer mode 243 */ 244 va = vapstart; 245 for (i = 0; i < countvmpg; i++) { 246 pa = pmap_kextract(va); 247 if (pa >= SIXTEENMEG) 248 ++dobounceflag; 249 va += NBPG; 250 } 251 if (dobounceflag == 0) 252 return; 253 254 if (bouncepages < dobounceflag) 255 panic("Not enough bounce buffers!!!"); 256 257/* 258 * allocate a replacement kva for b_addr 259 */ 260 kva = vm_bounce_kva(countvmpg); 261 va = vapstart; 262 for (i = 0; i < countvmpg; i++) { 263 pa = pmap_kextract(va); 264 if (pa >= SIXTEENMEG) { 265 /* 266 * allocate a replacement page 267 */ 268 vm_offset_t bpa = vm_bounce_page_find(1); 269 pmap_enter(kernel_pmap, kva + (NBPG * i), bpa, VM_PROT_DEFAULT, 270 TRUE); 271 /* 272 * if we are writing, the copy the data into the page 273 */ 274 if ((bp->b_flags & B_READ) == 0) 275 bcopy((caddr_t) va, (caddr_t) kva + (NBPG * i), NBPG); 276 } else { 277 /* 278 * use original page 279 */ 280 pmap_enter(kernel_pmap, kva + (NBPG * i), pa, VM_PROT_DEFAULT, 281 TRUE); 282 } 283 va += NBPG; 284 } 285 286/* 287 * flag the buffer as being bounced 288 */ 289 bp->b_flags |= B_BOUNCE; 290/* 291 * save the original buffer kva 292 */ 293 bp->b_savekva = bp->b_un.b_addr; 294/* 295 * put our new kva into the buffer (offset by original offset) 296 */ 297 bp->b_un.b_addr = (caddr_t) (((vm_offset_t) kva) | 298 ((vm_offset_t) bp->b_savekva & (NBPG - 1))); 299 return; 300} 301 302/* 303 * hook into biodone to free bounce buffer 304 */ 305void 306vm_bounce_free(bp) 307 struct buf *bp; 308{ 309 int i; 310 vm_offset_t origkva, bouncekva; 311 vm_offset_t vastart, vaend; 312 vm_offset_t vapstart, vapend; 313 int countbounce = 0; 314 vm_offset_t firstbouncepa = 0; 315 int firstbounceindex; 316 int countvmpg; 317 vm_offset_t bcount; 318 int s; 319 320/* 321 * if this isn't a bounced buffer, then just return 322 */ 323 if ((bp->b_flags & B_BOUNCE) == 0) 324 return; 325 326 origkva = (vm_offset_t) bp->b_savekva; 327 bouncekva = (vm_offset_t) bp->b_un.b_addr; 328 329 vastart = bouncekva; 330 vaend = bouncekva + bp->b_bcount; 331 bcount = bp->b_bcount; 332 333 vapstart = i386_trunc_page(vastart); 334 vapend = i386_round_page(vaend); 335 336 countvmpg = (vapend - vapstart) / NBPG; 337 338/* 339 * check every page in the kva space for b_addr 340 */ 341 for (i = 0; i < countvmpg; i++) { 342 vm_offset_t mybouncepa; 343 vm_offset_t copycount; 344 345 copycount = i386_round_page(bouncekva + 1) - bouncekva; 346 mybouncepa = pmap_kextract(i386_trunc_page(bouncekva)); 347 348/* 349 * if this is a bounced pa, then process as one 350 */ 351 if ((mybouncepa >= bouncepa) && (mybouncepa < bouncepaend)) { 352 if (copycount > bcount) 353 copycount = bcount; 354/* 355 * if this is a read, then copy from bounce buffer into original buffer 356 */ 357 if (bp->b_flags & B_READ) 358 bcopy((caddr_t) bouncekva, (caddr_t) origkva, copycount); 359/* 360 * free the bounce allocation 361 */ 362 vm_bounce_page_free(i386_trunc_page(mybouncepa), 1); 363 } 364 365 origkva += copycount; 366 bouncekva += copycount; 367 bcount -= copycount; 368 } 369 370/* 371 * add the old kva into the "to free" list 372 */ 373 bouncekva = i386_trunc_page((vm_offset_t) bp->b_un.b_addr); 374 kvaf[kvasfreecnt].addr = bouncekva; 375 kvaf[kvasfreecnt++].size = countvmpg * NBPG; 376 if (bmwait) { 377 /* 378 * if anyone is waiting on the bounce-map, then wakeup 379 */ 380 wakeup((caddr_t) bounce_map); 381 bmwait = 0; 382 } 383 384 bp->b_un.b_addr = bp->b_savekva; 385 bp->b_savekva = 0; 386 bp->b_flags &= ~B_BOUNCE; 387 388 return; 389} 390 391#endif /* NOBOUNCE */ 392 393/* 394 * Finish a fork operation, with process p2 nearly set up. 395 * Copy and update the kernel stack and pcb, making the child 396 * ready to run, and marking it so that it can return differently 397 * than the parent. Returns 1 in the child process, 0 in the parent. 398 * We currently double-map the user area so that the stack is at the same 399 * address in each process; in the future we will probably relocate 400 * the frame pointers on the stack after copying. 401 */ 402int 403cpu_fork(p1, p2) 404 register struct proc *p1, *p2; 405{ 406 register struct user *up = p2->p_addr; 407 int foo, offset, addr, i; 408 extern char kstack[]; 409 extern int mvesp(); 410 411 /* 412 * Copy pcb and stack from proc p1 to p2. 413 * We do this as cheaply as possible, copying only the active 414 * part of the stack. The stack and pcb need to agree; 415 * this is tricky, as the final pcb is constructed by savectx, 416 * but its frame isn't yet on the stack when the stack is copied. 417 * swtch compensates for this when the child eventually runs. 418 * This should be done differently, with a single call 419 * that copies and updates the pcb+stack, 420 * replacing the bcopy and savectx. 421 */ 422 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 423 offset = mvesp() - (int)kstack; 424 bcopy((caddr_t)kstack + offset, (caddr_t)p2->p_addr + offset, 425 (unsigned) ctob(UPAGES) - offset); 426 p2->p_regs = p1->p_regs; 427 428 /* 429 * Wire top of address space of child to it's kstack. 430 * First, fault in a page of pte's to map it. 431 */ 432#if 0 433 addr = trunc_page((u_int)vtopte(kstack)); 434 vm_map_pageable(&p2->p_vmspace->vm_map, addr, addr+NBPG, FALSE); 435 for (i=0; i < UPAGES; i++) 436 pmap_enter(&p2->p_vmspace->vm_pmap, kstack+i*NBPG, 437 pmap_extract(kernel_pmap, ((int)p2->p_addr)+i*NBPG), 438 /* 439 * The user area has to be mapped writable because 440 * it contains the kernel stack (when CR0_WP is on 441 * on a 486 there is no user-read/kernel-write 442 * mode). It is protected from user mode access 443 * by the segment limits. 444 */ 445 VM_PROT_READ|VM_PROT_WRITE, TRUE); 446#endif 447 pmap_activate(&p2->p_vmspace->vm_pmap, &up->u_pcb); 448 449 /* 450 * 451 * Arrange for a non-local goto when the new process 452 * is started, to resume here, returning nonzero from setjmp. 453 */ 454 if (savectx(up, 1)) { 455 /* 456 * Return 1 in child. 457 */ 458 return (1); 459 } 460 return (0); 461} 462 463#ifdef notyet 464/* 465 * cpu_exit is called as the last action during exit. 466 * 467 * We change to an inactive address space and a "safe" stack, 468 * passing thru an argument to the new stack. Now, safely isolated 469 * from the resources we're shedding, we release the address space 470 * and any remaining machine-dependent resources, including the 471 * memory for the user structure and kernel stack. 472 * 473 * Next, we assign a dummy context to be written over by swtch, 474 * calling it to send this process off to oblivion. 475 * [The nullpcb allows us to minimize cost in swtch() by not having 476 * a special case]. 477 */ 478struct proc *swtch_to_inactive(); 479volatile void 480cpu_exit(p) 481 register struct proc *p; 482{ 483 static struct pcb nullpcb; /* pcb to overwrite on last swtch */ 484 485#if NNPX > 0 486 npxexit(p); 487#endif /* NNPX */ 488 489 /* move to inactive space and stack, passing arg accross */ 490 p = swtch_to_inactive(p); 491 492 /* drop per-process resources */ 493 vmspace_free(p->p_vmspace); 494 kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 495 496 p->p_addr = (struct user *) &nullpcb; 497 splclock(); 498 swtch(); 499 /* NOTREACHED */ 500} 501#else 502void 503cpu_exit(p) 504 register struct proc *p; 505{ 506 507#if NNPX > 0 508 npxexit(p); 509#endif /* NNPX */ 510 splclock(); 511 curproc = 0; 512 swtch(); 513 /* 514 * This is to shutup the compiler, and if swtch() failed I suppose 515 * this would be a good thing. This keeps gcc happy because panic 516 * is a volatile void function as well. 517 */ 518 panic("cpu_exit"); 519} 520 521void 522cpu_wait(p) struct proc *p; { 523/* extern vm_map_t upages_map; */ 524 extern char kstack[]; 525 526 /* drop per-process resources */ 527 pmap_remove(vm_map_pmap(kernel_map), (vm_offset_t) p->p_addr, 528 ((vm_offset_t) p->p_addr) + ctob(UPAGES)); 529 kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 530 vmspace_free(p->p_vmspace); 531} 532#endif 533 534/* 535 * Set a red zone in the kernel stack after the u. area. 536 */ 537void 538setredzone(pte, vaddr) 539 u_short *pte; 540 caddr_t vaddr; 541{ 542/* eventually do this by setting up an expand-down stack segment 543 for ss0: selector, allowing stack access down to top of u. 544 this means though that protection violations need to be handled 545 thru a double fault exception that must do an integral task 546 switch to a known good context, within which a dump can be 547 taken. a sensible scheme might be to save the initial context 548 used by sched (that has physical memory mapped 1:1 at bottom) 549 and take the dump while still in mapped mode */ 550} 551 552/* 553 * Convert kernel VA to physical address 554 */ 555u_long 556kvtop(void *addr) 557{ 558 vm_offset_t va; 559 560 va = pmap_kextract((vm_offset_t)addr); 561 if (va == 0) 562 panic("kvtop: zero page frame"); 563 return((int)va); 564} 565 566extern vm_map_t phys_map; 567 568/* 569 * Map an IO request into kernel virtual address space. Requests fall into 570 * one of five catagories: 571 * 572 * B_PHYS|B_UAREA: User u-area swap. 573 * Address is relative to start of u-area (p_addr). 574 * B_PHYS|B_PAGET: User page table swap. 575 * Address is a kernel VA in usrpt (Usrptmap). 576 * B_PHYS|B_DIRTY: Dirty page push. 577 * Address is a VA in proc2's address space. 578 * B_PHYS|B_PGIN: Kernel pagein of user pages. 579 * Address is VA in user's address space. 580 * B_PHYS: User "raw" IO request. 581 * Address is VA in user's address space. 582 * 583 * All requests are (re)mapped into kernel VA space via the useriomap 584 * (a name with only slightly more meaning than "kernelmap") 585 */ 586void 587vmapbuf(bp) 588 register struct buf *bp; 589{ 590 register int npf; 591 register caddr_t addr; 592 register long flags = bp->b_flags; 593 struct proc *p; 594 int off; 595 vm_offset_t kva; 596 register vm_offset_t pa; 597 598 if ((flags & B_PHYS) == 0) 599 panic("vmapbuf"); 600 addr = bp->b_saveaddr = bp->b_un.b_addr; 601 off = (int)addr & PGOFSET; 602 p = bp->b_proc; 603 npf = btoc(round_page(bp->b_bcount + off)); 604 kva = kmem_alloc_wait(phys_map, ctob(npf)); 605 bp->b_un.b_addr = (caddr_t) (kva + off); 606 while (npf--) { 607 pa = pmap_extract(&p->p_vmspace->vm_pmap, (vm_offset_t)addr); 608 if (pa == 0) 609 panic("vmapbuf: null page frame"); 610 pmap_enter(vm_map_pmap(phys_map), kva, trunc_page(pa), 611 VM_PROT_READ|VM_PROT_WRITE, TRUE); 612 addr += PAGE_SIZE; 613 kva += PAGE_SIZE; 614 } 615} 616 617/* 618 * Free the io map PTEs associated with this IO operation. 619 * We also invalidate the TLB entries and restore the original b_addr. 620 */ 621void 622vunmapbuf(bp) 623 register struct buf *bp; 624{ 625 register int npf; 626 register caddr_t addr = bp->b_un.b_addr; 627 vm_offset_t kva; 628 629 if ((bp->b_flags & B_PHYS) == 0) 630 panic("vunmapbuf"); 631 npf = btoc(round_page(bp->b_bcount + ((int)addr & PGOFSET))); 632 kva = (vm_offset_t)((int)addr & ~PGOFSET); 633 kmem_free_wakeup(phys_map, kva, ctob(npf)); 634 bp->b_un.b_addr = bp->b_saveaddr; 635 bp->b_saveaddr = NULL; 636} 637 638/* 639 * Force reset the processor by invalidating the entire address space! 640 */ 641void 642cpu_reset() { 643 644 /* force a shutdown by unmapping entire address space ! */ 645 bzero((caddr_t) PTD, NBPG); 646 647 /* "good night, sweet prince .... <THUNK!>" */ 648 tlbflush(); 649 /* NOTREACHED */ 650 while(1); 651} 652 653/* 654 * Grow the user stack to allow for 'sp'. This version grows the stack in 655 * chunks of SGROWSIZ. 656 */ 657int 658grow(p, sp) 659 struct proc *p; 660 int sp; 661{ 662 unsigned int nss; 663 caddr_t v; 664 struct vmspace *vm = p->p_vmspace; 665 666 if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK) 667 return (1); 668 669 nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE); 670 671 if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur) 672 return (0); 673 674 if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT, 675 SGROWSIZ) < nss) { 676 int grow_amount; 677 /* 678 * If necessary, grow the VM that the stack occupies 679 * to allow for the rlimit. This allows us to not have 680 * to allocate all of the VM up-front in execve (which 681 * is expensive). 682 * Grow the VM by the amount requested rounded up to 683 * the nearest SGROWSIZ to provide for some hysteresis. 684 */ 685 grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ); 686 v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT, 687 SGROWSIZ) - grow_amount; 688 /* 689 * If there isn't enough room to extend by SGROWSIZ, then 690 * just extend to the maximum size 691 */ 692 if (v < vm->vm_maxsaddr) { 693 v = vm->vm_maxsaddr; 694 grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT); 695 } 696 if (vm_allocate(&vm->vm_map, (vm_offset_t *)&v, 697 grow_amount, FALSE) != KERN_SUCCESS) { 698 return (0); 699 } 700 vm->vm_ssize += grow_amount >> PAGE_SHIFT; 701 } 702 703 return (1); 704} 705