vm_glue.c revision 12820
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 1.32 1995/12/07 12:48:11 davidg Exp $ 63 */ 64 65#include <sys/param.h> 66#include <sys/systm.h> 67#include <sys/proc.h> 68#include <sys/resourcevar.h> 69#include <sys/buf.h> 70#include <sys/shm.h> 71#include <sys/vmmeter.h> 72 73#include <sys/kernel.h> 74#include <sys/dkstat.h> 75 76#include <vm/vm.h> 77#include <vm/vm_param.h> 78#include <vm/vm_inherit.h> 79#include <vm/vm_prot.h> 80#include <vm/lock.h> 81#include <vm/pmap.h> 82#include <vm/vm_map.h> 83#include <vm/vm_page.h> 84#include <vm/vm_pageout.h> 85#include <vm/vm_kern.h> 86#include <vm/vm_extern.h> 87 88#include <sys/user.h> 89 90#include <machine/stdarg.h> 91#include <machine/cpu.h> 92 93/* 94 * System initialization 95 * 96 * Note: proc0 from proc.h 97 */ 98 99static void vm_init_limits __P((void *)); 100SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 101 102/* 103 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 104 * 105 * Note: run scheduling should be divorced from the vm system. 106 */ 107static void scheduler __P((void *)); 108SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 109 110 111static void swapout __P((struct proc *)); 112 113extern char kstack[]; 114 115/* vm_map_t upages_map; */ 116 117int 118kernacc(addr, len, rw) 119 caddr_t addr; 120 int len, rw; 121{ 122 boolean_t rv; 123 vm_offset_t saddr, eaddr; 124 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 125 126 saddr = trunc_page(addr); 127 eaddr = round_page(addr + len); 128 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 129 return (rv == TRUE); 130} 131 132int 133useracc(addr, len, rw) 134 caddr_t addr; 135 int len, rw; 136{ 137 boolean_t rv; 138 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 139 140 /* 141 * XXX - check separately to disallow access to user area and user 142 * page tables - they are in the map. 143 * 144 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 145 * only used (as an end address) in trap.c. Use it as an end address 146 * here too. This bogusness has spread. I just fixed where it was 147 * used as a max in vm_mmap.c. 148 */ 149 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 150 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 151 return (FALSE); 152 } 153 rv = vm_map_check_protection(&curproc->p_vmspace->vm_map, 154 trunc_page(addr), round_page(addr + len), prot); 155 return (rv == TRUE); 156} 157 158#ifdef KGDB 159/* 160 * Change protections on kernel pages from addr to addr+len 161 * (presumably so debugger can plant a breakpoint). 162 * All addresses are assumed to reside in the Sysmap, 163 */ 164chgkprot(addr, len, rw) 165 register caddr_t addr; 166 int len, rw; 167{ 168 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 169 170 vm_map_protect(kernel_map, trunc_page(addr), 171 round_page(addr + len), prot, FALSE); 172} 173#endif 174void 175vslock(addr, len) 176 caddr_t addr; 177 u_int len; 178{ 179 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 180 round_page(addr + len), FALSE); 181} 182 183void 184vsunlock(addr, len, dirtied) 185 caddr_t addr; 186 u_int len; 187 int dirtied; 188{ 189#ifdef lint 190 dirtied++; 191#endif /* lint */ 192 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 193 round_page(addr + len), TRUE); 194} 195 196/* 197 * Implement fork's actions on an address space. 198 * Here we arrange for the address space to be copied or referenced, 199 * allocate a user struct (pcb and kernel stack), then call the 200 * machine-dependent layer to fill those in and make the new process 201 * ready to run. 202 * NOTE: the kernel stack may be at a different location in the child 203 * process, and thus addresses of automatic variables may be invalid 204 * after cpu_fork returns in the child process. We do nothing here 205 * after cpu_fork returns. 206 */ 207int 208vm_fork(p1, p2, isvfork) 209 register struct proc *p1, *p2; 210 int isvfork; 211{ 212 register struct user *up; 213 vm_offset_t addr, ptaddr; 214 int error, i; 215 struct vm_map *vp; 216 217 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 218 VM_WAIT; 219 } 220 221 /* 222 * avoid copying any of the parent's pagetables or other per-process 223 * objects that reside in the map by marking all of them 224 * non-inheritable 225 */ 226 (void) vm_map_inherit(&p1->p_vmspace->vm_map, 227 UPT_MIN_ADDRESS - UPAGES * PAGE_SIZE, VM_MAX_ADDRESS, VM_INHERIT_NONE); 228 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 229 230#ifdef SYSVSHM 231 if (p1->p_vmspace->vm_shm) 232 shmfork(p1, p2, isvfork); 233#endif 234 235 /* 236 * Allocate a wired-down (for now) pcb and kernel stack for the 237 * process 238 */ 239 240 addr = (vm_offset_t) kstack; 241 242 vp = &p2->p_vmspace->vm_map; 243 244 /* get new pagetables and kernel stack */ 245 (void) vm_map_find(vp, NULL, 0, &addr, UPT_MAX_ADDRESS - addr, FALSE); 246 247 /* force in the page table encompassing the UPAGES */ 248 ptaddr = trunc_page((u_int) vtopte(addr)); 249 error = vm_map_pageable(vp, ptaddr, ptaddr + PAGE_SIZE, FALSE); 250 if (error) 251 panic("vm_fork: wire of PT failed. error=%d", error); 252 253 /* and force in (demand-zero) the UPAGES */ 254 error = vm_map_pageable(vp, addr, addr + UPAGES * PAGE_SIZE, FALSE); 255 if (error) 256 panic("vm_fork: wire of UPAGES failed. error=%d", error); 257 258 /* get a kernel virtual address for the UPAGES for this proc */ 259 up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * PAGE_SIZE); 260 if (up == NULL) 261 panic("vm_fork: u_map allocation failed"); 262 263 /* and force-map the upages into the kernel pmap */ 264 for (i = 0; i < UPAGES; i++) 265 pmap_kenter(((vm_offset_t) up) + PAGE_SIZE * i, 266 pmap_extract(vp->pmap, addr + PAGE_SIZE * i)); 267 268 p2->p_addr = up; 269 270 /* 271 * p_stats and p_sigacts currently point at fields in the user struct 272 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 273 * p_stats; zero the rest of p_stats (statistics). 274 */ 275 p2->p_stats = &up->u_stats; 276 p2->p_sigacts = &up->u_sigacts; 277 up->u_sigacts = *p1->p_sigacts; 278 bzero(&up->u_stats.pstat_startzero, 279 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 280 (caddr_t) &up->u_stats.pstat_startzero)); 281 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 282 ((caddr_t) &up->u_stats.pstat_endcopy - 283 (caddr_t) &up->u_stats.pstat_startcopy)); 284 285 286 /* 287 * cpu_fork will copy and update the kernel stack and pcb, and make 288 * the child ready to run. It marks the child so that it can return 289 * differently than the parent. It returns twice, once in the parent 290 * process and once in the child. 291 */ 292 return (cpu_fork(p1, p2)); 293} 294 295/* 296 * Set default limits for VM system. 297 * Called for proc 0, and then inherited by all others. 298 * 299 * XXX should probably act directly on proc0. 300 */ 301static void 302vm_init_limits(udata) 303 void *udata; 304{ 305 register struct proc *p = udata; 306 int rss_limit; 307 308 /* 309 * Set up the initial limits on process VM. Set the maximum resident 310 * set size to be half of (reasonably) available memory. Since this 311 * is a soft limit, it comes into effect only when the system is out 312 * of memory - half of main memory helps to favor smaller processes, 313 * and reduces thrashing of the object cache. 314 */ 315 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 316 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 317 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 318 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 319 /* limit the limit to no less than 2MB */ 320 rss_limit = max(cnt.v_free_count, 512); 321 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 322 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 323} 324 325void 326faultin(p) 327 struct proc *p; 328{ 329 vm_offset_t i; 330 vm_offset_t ptaddr; 331 int s; 332 333 if ((p->p_flag & P_INMEM) == 0) { 334 vm_map_t map; 335 int error; 336 337 ++p->p_lock; 338 339 map = &p->p_vmspace->vm_map; 340 /* force the page table encompassing the kernel stack (upages) */ 341 ptaddr = trunc_page((u_int) vtopte(kstack)); 342 error = vm_map_pageable(map, ptaddr, ptaddr + PAGE_SIZE, FALSE); 343 if (error) 344 panic("faultin: wire of PT failed. error=%d", error); 345 346 /* wire in the UPAGES */ 347 error = vm_map_pageable(map, (vm_offset_t) kstack, 348 (vm_offset_t) kstack + UPAGES * PAGE_SIZE, FALSE); 349 if (error) 350 panic("faultin: wire of UPAGES failed. error=%d", error); 351 352 /* and map them nicely into the kernel pmap */ 353 for (i = 0; i < UPAGES; i++) { 354 vm_offset_t off = i * PAGE_SIZE; 355 vm_offset_t pa = (vm_offset_t) 356 pmap_extract(&p->p_vmspace->vm_pmap, 357 (vm_offset_t) kstack + off); 358 359 if (pa == 0) 360 panic("faultin: missing page for UPAGES\n"); 361 362 pmap_kenter(((vm_offset_t) p->p_addr) + off, pa); 363 } 364 365 s = splhigh(); 366 367 if (p->p_stat == SRUN) 368 setrunqueue(p); 369 370 p->p_flag |= P_INMEM; 371 372 /* undo the effect of setting SLOCK above */ 373 --p->p_lock; 374 splx(s); 375 376 } 377} 378 379/* 380 * This swapin algorithm attempts to swap-in processes only if there 381 * is enough space for them. Of course, if a process waits for a long 382 * time, it will be swapped in anyway. 383 */ 384/* ARGSUSED*/ 385static void 386scheduler(dummy) 387 void *dummy; 388{ 389 register struct proc *p; 390 register int pri; 391 struct proc *pp; 392 int ppri; 393 394loop: 395 while ((cnt.v_free_count + cnt.v_cache_count) < (cnt.v_free_reserved + UPAGES + 2)) { 396 VM_WAIT; 397 } 398 399 pp = NULL; 400 ppri = INT_MIN; 401 for (p = (struct proc *) allproc; p != NULL; p = p->p_next) { 402 if (p->p_stat == SRUN && (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 403 int mempri; 404 405 pri = p->p_swtime + p->p_slptime - p->p_nice * 8; 406 mempri = pri > 0 ? pri : 0; 407 /* 408 * if this process is higher priority and there is 409 * enough space, then select this process instead of 410 * the previous selection. 411 */ 412 if (pri > ppri) { 413 pp = p; 414 ppri = pri; 415 } 416 } 417 } 418 419 /* 420 * Nothing to do, back to sleep 421 */ 422 if ((p = pp) == NULL) { 423 tsleep(&proc0, PVM, "sched", 0); 424 goto loop; 425 } 426 /* 427 * We would like to bring someone in. (only if there is space). 428 */ 429 faultin(p); 430 p->p_swtime = 0; 431 goto loop; 432} 433 434#define swappable(p) \ 435 (((p)->p_lock == 0) && \ 436 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 437 438extern int vm_pageout_free_min; 439 440/* 441 * Swapout is driven by the pageout daemon. Very simple, we find eligible 442 * procs and unwire their u-areas. We try to always "swap" at least one 443 * process in case we need the room for a swapin. 444 * If any procs have been sleeping/stopped for at least maxslp seconds, 445 * they are swapped. Else, we swap the longest-sleeping or stopped process, 446 * if any, otherwise the longest-resident process. 447 */ 448void 449swapout_procs() 450{ 451 register struct proc *p; 452 struct proc *outp, *outp2; 453 int outpri, outpri2; 454 int didswap = 0; 455 456 outp = outp2 = NULL; 457 outpri = outpri2 = INT_MIN; 458retry: 459 for (p = (struct proc *) allproc; p != NULL; p = p->p_next) { 460 if (!swappable(p)) 461 continue; 462 switch (p->p_stat) { 463 default: 464 continue; 465 466 case SSLEEP: 467 case SSTOP: 468 /* 469 * do not swapout a realtime process 470 */ 471 if (p->p_rtprio.type == RTP_PRIO_REALTIME) 472 continue; 473 474 /* 475 * do not swapout a process waiting on a critical 476 * event of some kind 477 */ 478 if (((p->p_priority & 0x7f) < PSOCK) || 479 (p->p_slptime <= 4)) 480 continue; 481 482 vm_map_reference(&p->p_vmspace->vm_map); 483 /* 484 * do not swapout a process that is waiting for VM 485 * datastructures there is a possible deadlock. 486 */ 487 if (!lock_try_write(&p->p_vmspace->vm_map.lock)) { 488 vm_map_deallocate(&p->p_vmspace->vm_map); 489 continue; 490 } 491 vm_map_unlock(&p->p_vmspace->vm_map); 492 /* 493 * If the process has been asleep for awhile and had 494 * most of its pages taken away already, swap it out. 495 */ 496 swapout(p); 497 vm_map_deallocate(&p->p_vmspace->vm_map); 498 didswap++; 499 goto retry; 500 } 501 } 502 /* 503 * If we swapped something out, and another process needed memory, 504 * then wakeup the sched process. 505 */ 506 if (didswap) 507 wakeup(&proc0); 508} 509 510static void 511swapout(p) 512 register struct proc *p; 513{ 514 vm_map_t map = &p->p_vmspace->vm_map; 515 vm_offset_t ptaddr; 516 int i; 517 518 ++p->p_stats->p_ru.ru_nswap; 519 /* 520 * remember the process resident count 521 */ 522 p->p_vmspace->vm_swrss = 523 p->p_vmspace->vm_pmap.pm_stats.resident_count; 524 525 (void) splhigh(); 526 p->p_flag &= ~P_INMEM; 527 p->p_flag |= P_SWAPPING; 528 if (p->p_stat == SRUN) 529 remrq(p); 530 (void) spl0(); 531 532 /* 533 * let the upages be paged 534 */ 535 for(i=0;i<UPAGES;i++) 536 pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i); 537 538 vm_map_pageable(map, (vm_offset_t) kstack, 539 (vm_offset_t) kstack + UPAGES * PAGE_SIZE, TRUE); 540 541 ptaddr = trunc_page((u_int) vtopte(kstack)); 542 vm_map_pageable(map, ptaddr, ptaddr + PAGE_SIZE, TRUE); 543 544 p->p_flag &= ~P_SWAPPING; 545 p->p_swtime = 0; 546} 547 548#ifdef DDB 549/* 550 * DEBUG stuff 551 */ 552 553int indent; 554 555#include <machine/stdarg.h> /* see subr_prf.c */ 556 557/*ARGSUSED2*/ 558void 559#if __STDC__ 560iprintf(const char *fmt,...) 561#else 562iprintf(fmt /* , va_alist */ ) 563 char *fmt; 564 565 /* va_dcl */ 566#endif 567{ 568 register int i; 569 va_list ap; 570 571 for (i = indent; i >= 8; i -= 8) 572 printf("\t"); 573 while (--i >= 0) 574 printf(" "); 575 va_start(ap, fmt); 576 printf("%r", fmt, ap); 577 va_end(ap); 578} 579#endif /* DDB */ 580