vm_glue.c revision 15117
1/* 2 * Copyright (c) 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * The Mach Operating System project at Carnegie-Mellon University. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 37 * 38 * 39 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40 * All rights reserved. 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 * 62 * $Id: vm_glue.c,v 1.44 1996/04/03 05:23:36 dyson Exp $ 63 */ 64 65#include "opt_ddb.h" 66 67#include <sys/param.h> 68#include <sys/systm.h> 69#include <sys/proc.h> 70#include <sys/resourcevar.h> 71#include <sys/buf.h> 72#include <sys/shm.h> 73#include <sys/vmmeter.h> 74 75#include <sys/kernel.h> 76#include <sys/dkstat.h> 77 78#include <vm/vm.h> 79#include <vm/vm_param.h> 80#include <vm/vm_inherit.h> 81#include <vm/vm_prot.h> 82#include <vm/lock.h> 83#include <vm/pmap.h> 84#include <vm/vm_map.h> 85#include <vm/vm_page.h> 86#include <vm/vm_pageout.h> 87#include <vm/vm_kern.h> 88#include <vm/vm_extern.h> 89#include <vm/vm_object.h> 90#include <vm/vm_pager.h> 91 92#include <sys/user.h> 93 94#include <machine/stdarg.h> 95 96/* 97 * System initialization 98 * 99 * Note: proc0 from proc.h 100 */ 101 102static void vm_init_limits __P((void *)); 103SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 104 105/* 106 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 107 * 108 * Note: run scheduling should be divorced from the vm system. 109 */ 110static void scheduler __P((void *)); 111SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 112 113 114static void swapout __P((struct proc *)); 115 116extern char kstack[]; 117 118/* vm_map_t upages_map; */ 119 120int 121kernacc(addr, len, rw) 122 caddr_t addr; 123 int len, rw; 124{ 125 boolean_t rv; 126 vm_offset_t saddr, eaddr; 127 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 128 129 saddr = trunc_page(addr); 130 eaddr = round_page(addr + len); 131 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 132 return (rv == TRUE); 133} 134 135int 136useracc(addr, len, rw) 137 caddr_t addr; 138 int len, rw; 139{ 140 boolean_t rv; 141 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 142 143 /* 144 * XXX - check separately to disallow access to user area and user 145 * page tables - they are in the map. 146 * 147 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 148 * only used (as an end address) in trap.c. Use it as an end address 149 * here too. This bogusness has spread. I just fixed where it was 150 * used as a max in vm_mmap.c. 151 */ 152 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 153 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 154 return (FALSE); 155 } 156 rv = vm_map_check_protection(&curproc->p_vmspace->vm_map, 157 trunc_page(addr), round_page(addr + len), prot); 158 return (rv == TRUE); 159} 160 161#ifdef KGDB 162/* 163 * Change protections on kernel pages from addr to addr+len 164 * (presumably so debugger can plant a breakpoint). 165 * All addresses are assumed to reside in the Sysmap, 166 */ 167chgkprot(addr, len, rw) 168 register caddr_t addr; 169 int len, rw; 170{ 171 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 172 173 vm_map_protect(kernel_map, trunc_page(addr), 174 round_page(addr + len), prot, FALSE); 175} 176#endif 177void 178vslock(addr, len) 179 caddr_t addr; 180 u_int len; 181{ 182 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 183 round_page(addr + len), FALSE); 184} 185 186void 187vsunlock(addr, len, dirtied) 188 caddr_t addr; 189 u_int len; 190 int dirtied; 191{ 192#ifdef lint 193 dirtied++; 194#endif /* lint */ 195 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 196 round_page(addr + len), TRUE); 197} 198 199/* 200 * Implement fork's actions on an address space. 201 * Here we arrange for the address space to be copied or referenced, 202 * allocate a user struct (pcb and kernel stack), then call the 203 * machine-dependent layer to fill those in and make the new process 204 * ready to run. 205 * NOTE: the kernel stack may be at a different location in the child 206 * process, and thus addresses of automatic variables may be invalid 207 * after cpu_fork returns in the child process. We do nothing here 208 * after cpu_fork returns. 209 */ 210int 211vm_fork(p1, p2) 212 register struct proc *p1, *p2; 213{ 214 register struct user *up; 215 vm_offset_t addr, ptaddr, ptpa; 216 int error, i; 217 vm_map_t map; 218 pmap_t pvp; 219 vm_page_t stkm; 220 221 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 222 VM_WAIT; 223 } 224 225 /* 226 * avoid copying any of the parent's pagetables or other per-process 227 * objects that reside in the map by marking all of them 228 * non-inheritable 229 */ 230 (void) vm_map_inherit(&p1->p_vmspace->vm_map, 231 UPT_MIN_ADDRESS - UPAGES * PAGE_SIZE, VM_MAX_ADDRESS, VM_INHERIT_NONE); 232 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 233 234 if (p1->p_vmspace->vm_shm) 235 shmfork(p1, p2); 236 237 /* 238 * Allocate a wired-down (for now) pcb and kernel stack for the 239 * process 240 */ 241 242 addr = (vm_offset_t) kstack; 243 244 map = &p2->p_vmspace->vm_map; 245 pvp = &p2->p_vmspace->vm_pmap; 246 247 /* 248 * allocate object for the upages 249 */ 250 p2->p_vmspace->vm_upages_obj = vm_object_allocate( OBJT_DEFAULT, 251 UPAGES); 252 253 /* 254 * put upages into the address space 255 */ 256 error = vm_map_find(map, p2->p_vmspace->vm_upages_obj, 0, 257 &addr, UPT_MIN_ADDRESS - addr, FALSE, VM_PROT_ALL, 258 VM_PROT_ALL, 0); 259 if (error != KERN_SUCCESS) 260 panic("vm_fork: vm_map_find (UPAGES) failed, addr=0x%x, error=%d", addr, error); 261 262 addr += UPAGES * PAGE_SIZE; 263 /* allocate space for page tables */ 264 error = vm_map_find(map, NULL, 0, &addr, UPT_MAX_ADDRESS - addr, FALSE, 265 VM_PROT_ALL, VM_PROT_ALL, 0); 266 if (error != KERN_SUCCESS) 267 panic("vm_fork: vm_map_find (PTES) failed, addr=0x%x, error=%d", addr, error); 268 269 /* get a kernel virtual address for the UPAGES for this proc */ 270 up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * PAGE_SIZE); 271 if (up == NULL) 272 panic("vm_fork: u_map allocation failed"); 273 274 /* 275 * create a pagetable page for the UPAGES in the process address space 276 */ 277 ptaddr = trunc_page((u_int) vtopte(kstack)); 278 (void) vm_fault(map, ptaddr, VM_PROT_READ|VM_PROT_WRITE, FALSE); 279 ptpa = pmap_extract(pvp, ptaddr); 280 if (ptpa == 0) { 281 panic("vm_fork: no pte for UPAGES"); 282 } 283 284 /* 285 * hold the page table page for the kernel stack, and fault them in 286 */ 287 stkm = PHYS_TO_VM_PAGE(ptpa); 288 vm_page_hold(stkm); 289 290 for(i=0;i<UPAGES;i++) { 291 vm_page_t m; 292 293 /* 294 * Get a kernel stack page 295 */ 296 while ((m = vm_page_alloc(p2->p_vmspace->vm_upages_obj, 297 i, VM_ALLOC_NORMAL)) == NULL) { 298 VM_WAIT; 299 } 300 301 /* 302 * Wire the page 303 */ 304 vm_page_wire(m); 305 m->flags &= ~PG_BUSY; 306 307 /* 308 * Enter the page into both the kernel and the process 309 * address space. 310 */ 311 pmap_enter( pvp, (vm_offset_t) kstack + i * PAGE_SIZE, 312 VM_PAGE_TO_PHYS(m), VM_PROT_READ|VM_PROT_WRITE, 1); 313 pmap_kenter(((vm_offset_t) up) + i * PAGE_SIZE, 314 VM_PAGE_TO_PHYS(m)); 315 m->flags &= ~PG_ZERO; 316 m->valid = VM_PAGE_BITS_ALL; 317 } 318 /* 319 * The page table page for the kernel stack should be held in memory 320 * now. 321 */ 322 vm_page_unhold(stkm); 323 324 p2->p_addr = up; 325 326 /* 327 * p_stats and p_sigacts currently point at fields in the user struct 328 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 329 * p_stats; zero the rest of p_stats (statistics). 330 */ 331 p2->p_stats = &up->u_stats; 332 p2->p_sigacts = &up->u_sigacts; 333 up->u_sigacts = *p1->p_sigacts; 334 bzero(&up->u_stats.pstat_startzero, 335 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 336 (caddr_t) &up->u_stats.pstat_startzero)); 337 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 338 ((caddr_t) &up->u_stats.pstat_endcopy - 339 (caddr_t) &up->u_stats.pstat_startcopy)); 340 341 342 /* 343 * cpu_fork will copy and update the kernel stack and pcb, and make 344 * the child ready to run. It marks the child so that it can return 345 * differently than the parent. It returns twice, once in the parent 346 * process and once in the child. 347 */ 348 return (cpu_fork(p1, p2)); 349} 350 351/* 352 * Set default limits for VM system. 353 * Called for proc 0, and then inherited by all others. 354 * 355 * XXX should probably act directly on proc0. 356 */ 357static void 358vm_init_limits(udata) 359 void *udata; 360{ 361 register struct proc *p = udata; 362 int rss_limit; 363 364 /* 365 * Set up the initial limits on process VM. Set the maximum resident 366 * set size to be half of (reasonably) available memory. Since this 367 * is a soft limit, it comes into effect only when the system is out 368 * of memory - half of main memory helps to favor smaller processes, 369 * and reduces thrashing of the object cache. 370 */ 371 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 372 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 373 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 374 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 375 /* limit the limit to no less than 2MB */ 376 rss_limit = max(cnt.v_free_count, 512); 377 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 378 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 379} 380 381void 382faultin(p) 383 struct proc *p; 384{ 385 vm_offset_t i; 386 vm_offset_t ptaddr; 387 int s; 388 389 if ((p->p_flag & P_INMEM) == 0) { 390 vm_map_t map = &p->p_vmspace->vm_map; 391 pmap_t pmap = &p->p_vmspace->vm_pmap; 392 vm_page_t stkm, m; 393 vm_offset_t ptpa; 394 int error; 395 396 ++p->p_lock; 397 398 ptaddr = trunc_page((u_int) vtopte(kstack)); 399 (void) vm_fault(map, ptaddr, VM_PROT_READ|VM_PROT_WRITE, FALSE); 400 ptpa = pmap_extract(&p->p_vmspace->vm_pmap, ptaddr); 401 if (ptpa == 0) { 402 panic("vm_fork: no pte for UPAGES"); 403 } 404 stkm = PHYS_TO_VM_PAGE(ptpa); 405 vm_page_hold(stkm); 406 407 for(i=0;i<UPAGES;i++) { 408 int s; 409 s = splhigh(); 410 411retry: 412 if ((m = vm_page_lookup(p->p_vmspace->vm_upages_obj, i)) == NULL) { 413 if ((m = vm_page_alloc(p->p_vmspace->vm_upages_obj, i, VM_ALLOC_NORMAL)) == NULL) { 414 VM_WAIT; 415 goto retry; 416 } 417 } else { 418 if ((m->flags & PG_BUSY) || m->busy) { 419 m->flags |= PG_WANTED; 420 tsleep(m, PVM, "swinuw",0); 421 goto retry; 422 } 423 } 424 vm_page_wire(m); 425 if (m->valid == VM_PAGE_BITS_ALL) 426 m->flags &= ~PG_BUSY; 427 splx(s); 428 429 pmap_enter( pmap, (vm_offset_t) kstack + i * PAGE_SIZE, 430 VM_PAGE_TO_PHYS(m), VM_PROT_READ|VM_PROT_WRITE, TRUE); 431 pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE, 432 VM_PAGE_TO_PHYS(m)); 433 if (m->valid != VM_PAGE_BITS_ALL) { 434 int rv; 435 rv = vm_pager_get_pages(p->p_vmspace->vm_upages_obj, 436 &m, 1, 0); 437 if (rv != VM_PAGER_OK) 438 panic("faultin: cannot get upages for proc: %d\n", p->p_pid); 439 m->valid = VM_PAGE_BITS_ALL; 440 m->flags &= ~PG_BUSY; 441 } 442 } 443 vm_page_unhold(stkm); 444 445 446 s = splhigh(); 447 448 if (p->p_stat == SRUN) 449 setrunqueue(p); 450 451 p->p_flag |= P_INMEM; 452 453 /* undo the effect of setting SLOCK above */ 454 --p->p_lock; 455 splx(s); 456 457 } 458} 459 460/* 461 * This swapin algorithm attempts to swap-in processes only if there 462 * is enough space for them. Of course, if a process waits for a long 463 * time, it will be swapped in anyway. 464 */ 465/* ARGSUSED*/ 466static void 467scheduler(dummy) 468 void *dummy; 469{ 470 register struct proc *p; 471 register int pri; 472 struct proc *pp; 473 int ppri; 474 475loop: 476 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 477 VM_WAIT; 478 } 479 480 pp = NULL; 481 ppri = INT_MIN; 482 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 483 if (p->p_stat == SRUN && 484 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 485 int mempri; 486 487 pri = p->p_swtime + p->p_slptime - p->p_nice * 8; 488 mempri = pri > 0 ? pri : 0; 489 /* 490 * if this process is higher priority and there is 491 * enough space, then select this process instead of 492 * the previous selection. 493 */ 494 if (pri > ppri) { 495 pp = p; 496 ppri = pri; 497 } 498 } 499 } 500 501 /* 502 * Nothing to do, back to sleep 503 */ 504 if ((p = pp) == NULL) { 505 tsleep(&proc0, PVM, "sched", 0); 506 goto loop; 507 } 508 /* 509 * We would like to bring someone in. (only if there is space). 510 */ 511 faultin(p); 512 p->p_swtime = 0; 513 goto loop; 514} 515 516#ifndef NO_SWAPPING 517 518#define swappable(p) \ 519 (((p)->p_lock == 0) && \ 520 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 521 522/* 523 * Swapout is driven by the pageout daemon. Very simple, we find eligible 524 * procs and unwire their u-areas. We try to always "swap" at least one 525 * process in case we need the room for a swapin. 526 * If any procs have been sleeping/stopped for at least maxslp seconds, 527 * they are swapped. Else, we swap the longest-sleeping or stopped process, 528 * if any, otherwise the longest-resident process. 529 */ 530void 531swapout_procs() 532{ 533 register struct proc *p; 534 struct proc *outp, *outp2; 535 int outpri, outpri2; 536 int didswap = 0; 537 538 outp = outp2 = NULL; 539 outpri = outpri2 = INT_MIN; 540retry: 541 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 542 if (!swappable(p)) 543 continue; 544 switch (p->p_stat) { 545 default: 546 continue; 547 548 case SSLEEP: 549 case SSTOP: 550 /* 551 * do not swapout a realtime process 552 */ 553 if (p->p_rtprio.type == RTP_PRIO_REALTIME) 554 continue; 555 556 /* 557 * do not swapout a process waiting on a critical 558 * event of some kind 559 */ 560 if (((p->p_priority & 0x7f) < PSOCK) || 561 (p->p_slptime <= 4)) 562 continue; 563 564 vm_map_reference(&p->p_vmspace->vm_map); 565 /* 566 * do not swapout a process that is waiting for VM 567 * datastructures there is a possible deadlock. 568 */ 569 if (!lock_try_write(&p->p_vmspace->vm_map.lock)) { 570 vm_map_deallocate(&p->p_vmspace->vm_map); 571 continue; 572 } 573 vm_map_unlock(&p->p_vmspace->vm_map); 574 /* 575 * If the process has been asleep for awhile and had 576 * most of its pages taken away already, swap it out. 577 */ 578 swapout(p); 579 vm_map_deallocate(&p->p_vmspace->vm_map); 580 didswap++; 581 goto retry; 582 } 583 } 584 /* 585 * If we swapped something out, and another process needed memory, 586 * then wakeup the sched process. 587 */ 588 if (didswap) 589 wakeup(&proc0); 590} 591 592static void 593swapout(p) 594 register struct proc *p; 595{ 596 vm_map_t map = &p->p_vmspace->vm_map; 597 pmap_t pmap = &p->p_vmspace->vm_pmap; 598 vm_offset_t ptaddr; 599 int i; 600 601 ++p->p_stats->p_ru.ru_nswap; 602 /* 603 * remember the process resident count 604 */ 605 p->p_vmspace->vm_swrss = 606 p->p_vmspace->vm_pmap.pm_stats.resident_count; 607 608 (void) splhigh(); 609 p->p_flag &= ~P_INMEM; 610 p->p_flag |= P_SWAPPING; 611 if (p->p_stat == SRUN) 612 remrq(p); 613 (void) spl0(); 614 615 /* 616 * let the upages be paged 617 */ 618 for(i=0;i<UPAGES;i++) { 619 vm_page_t m; 620 if ((m = vm_page_lookup(p->p_vmspace->vm_upages_obj, i)) == NULL) 621 panic("swapout: upage already missing???"); 622 m->dirty = VM_PAGE_BITS_ALL; 623 vm_page_unwire(m); 624 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 625 } 626 pmap_remove(pmap, (vm_offset_t) kstack, 627 (vm_offset_t) kstack + PAGE_SIZE * UPAGES); 628 629 p->p_flag &= ~P_SWAPPING; 630 p->p_swtime = 0; 631} 632#endif /* !NO_SWAPPING */ 633 634#ifdef DDB 635/* 636 * DEBUG stuff 637 */ 638 639int indent; 640 641#include <machine/stdarg.h> /* see subr_prf.c */ 642 643/*ARGSUSED2*/ 644void 645#if __STDC__ 646iprintf(const char *fmt,...) 647#else 648iprintf(fmt /* , va_alist */ ) 649 char *fmt; 650 651 /* va_dcl */ 652#endif 653{ 654 register int i; 655 va_list ap; 656 657 for (i = indent; i >= 8; i -= 8) 658 printf("\t"); 659 while (--i >= 0) 660 printf(" "); 661 va_start(ap, fmt); 662 vprintf(fmt, ap); 663 va_end(ap); 664} 665#endif /* DDB */ 666