vm_glue.c revision 28551
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 1.64 1997/04/14 03:40:42 peter Exp $ 63 */ 64 65#include "opt_rlimit.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#include <sys/unistd.h> 78 79#include <machine/limits.h> 80 81#include <vm/vm.h> 82#include <vm/vm_param.h> 83#include <vm/vm_inherit.h> 84#include <vm/vm_prot.h> 85#include <sys/lock.h> 86#include <vm/pmap.h> 87#include <vm/vm_map.h> 88#include <vm/vm_page.h> 89#include <vm/vm_pageout.h> 90#include <vm/vm_kern.h> 91#include <vm/vm_extern.h> 92#include <vm/vm_object.h> 93#include <vm/vm_pager.h> 94 95#include <sys/user.h> 96 97/* 98 * System initialization 99 * 100 * Note: proc0 from proc.h 101 */ 102 103static void vm_init_limits __P((void *)); 104SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 105 106/* 107 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 108 * 109 * Note: run scheduling should be divorced from the vm system. 110 */ 111static void scheduler __P((void *)); 112SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 113 114 115static void swapout __P((struct proc *)); 116 117extern char kstack[]; 118 119/* vm_map_t upages_map; */ 120 121int 122kernacc(addr, len, rw) 123 caddr_t addr; 124 int len, rw; 125{ 126 boolean_t rv; 127 vm_offset_t saddr, eaddr; 128 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 129 130 saddr = trunc_page(addr); 131 eaddr = round_page(addr + len); 132 vm_map_lock_read(kernel_map); 133 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 134 vm_map_unlock_read(kernel_map); 135 return (rv == TRUE); 136} 137 138int 139useracc(addr, len, rw) 140 caddr_t addr; 141 int len, rw; 142{ 143 boolean_t rv; 144 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 145 vm_map_t map; 146 vm_map_entry_t save_hint; 147 148 /* 149 * XXX - check separately to disallow access to user area and user 150 * page tables - they are in the map. 151 * 152 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 153 * only used (as an end address) in trap.c. Use it as an end address 154 * here too. This bogusness has spread. I just fixed where it was 155 * used as a max in vm_mmap.c. 156 */ 157 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 158 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 159 return (FALSE); 160 } 161 map = &curproc->p_vmspace->vm_map; 162 vm_map_lock_read(map); 163 /* 164 * We save the map hint, and restore it. Useracc appears to distort 165 * the map hint unnecessarily. 166 */ 167 save_hint = map->hint; 168 rv = vm_map_check_protection(map, 169 trunc_page(addr), round_page(addr + len), prot); 170 map->hint = save_hint; 171 vm_map_unlock_read(map); 172 173 return (rv == TRUE); 174} 175 176void 177vslock(addr, len) 178 caddr_t addr; 179 u_int len; 180{ 181 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 182 round_page(addr + len), FALSE); 183} 184 185void 186vsunlock(addr, len, dirtied) 187 caddr_t addr; 188 u_int len; 189 int dirtied; 190{ 191#ifdef lint 192 dirtied++; 193#endif /* lint */ 194 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 195 round_page(addr + len), TRUE); 196} 197 198/* 199 * Implement fork's actions on an address space. 200 * Here we arrange for the address space to be copied or referenced, 201 * allocate a user struct (pcb and kernel stack), then call the 202 * machine-dependent layer to fill those in and make the new process 203 * ready to run. The new process is set up so that it returns directly 204 * to user mode to avoid stack copying and relocation problems. 205 */ 206void 207vm_fork(p1, p2, flags) 208 register struct proc *p1, *p2; 209 int flags; 210{ 211 register struct user *up; 212 int i; 213 pmap_t pvp; 214 215 if (flags & RFMEM) { 216 p2->p_vmspace = p1->p_vmspace; 217 p1->p_vmspace->vm_refcnt++; 218 } 219 220 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 221 VM_WAIT; 222 } 223 224 if ((flags & RFMEM) == 0) { 225 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 226 227 if (p1->p_vmspace->vm_shm) 228 shmfork(p1, p2); 229 } 230 231 pmap_new_proc(p2); 232 233 up = p2->p_addr; 234 235 /* 236 * p_stats and p_sigacts currently point at fields in the user struct 237 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 238 * p_stats; zero the rest of p_stats (statistics). 239 */ 240 p2->p_stats = &up->u_stats; 241 p2->p_sigacts = &up->u_sigacts; 242 up->u_sigacts = *p1->p_sigacts; 243 bzero(&up->u_stats.pstat_startzero, 244 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 245 (caddr_t) &up->u_stats.pstat_startzero)); 246 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 247 ((caddr_t) &up->u_stats.pstat_endcopy - 248 (caddr_t) &up->u_stats.pstat_startcopy)); 249 250 251 /* 252 * cpu_fork will copy and update the pcb, set up the kernel stack, 253 * and make the child ready to run. 254 */ 255 cpu_fork(p1, p2); 256} 257 258/* 259 * Set default limits for VM system. 260 * Called for proc 0, and then inherited by all others. 261 * 262 * XXX should probably act directly on proc0. 263 */ 264static void 265vm_init_limits(udata) 266 void *udata; 267{ 268 register struct proc *p = udata; 269 int rss_limit; 270 271 /* 272 * Set up the initial limits on process VM. Set the maximum resident 273 * set size to be half of (reasonably) available memory. Since this 274 * is a soft limit, it comes into effect only when the system is out 275 * of memory - half of main memory helps to favor smaller processes, 276 * and reduces thrashing of the object cache. 277 */ 278 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 279 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 280 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 281 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 282 /* limit the limit to no less than 2MB */ 283 rss_limit = max(cnt.v_free_count, 512); 284 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 285 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 286} 287 288void 289faultin(p) 290 struct proc *p; 291{ 292 vm_offset_t i; 293 int s; 294 295 if ((p->p_flag & P_INMEM) == 0) { 296 297 ++p->p_lock; 298 299 pmap_swapin_proc(p); 300 301 s = splhigh(); 302 303 if (p->p_stat == SRUN) 304 setrunqueue(p); 305 306 p->p_flag |= P_INMEM; 307 308 /* undo the effect of setting SLOCK above */ 309 --p->p_lock; 310 splx(s); 311 312 } 313} 314 315/* 316 * This swapin algorithm attempts to swap-in processes only if there 317 * is enough space for them. Of course, if a process waits for a long 318 * time, it will be swapped in anyway. 319 */ 320/* ARGSUSED*/ 321static void 322scheduler(dummy) 323 void *dummy; 324{ 325 register struct proc *p; 326 register int pri; 327 struct proc *pp; 328 int ppri; 329 330loop: 331 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 332 VM_WAIT; 333 } 334 335 pp = NULL; 336 ppri = INT_MIN; 337 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 338 if (p->p_stat == SRUN && 339 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 340 int mempri; 341 342 pri = p->p_swtime + p->p_slptime; 343 if ((p->p_flag & P_SWAPINREQ) == 0) { 344 pri -= p->p_nice * 8; 345 } 346 mempri = pri > 0 ? pri : 0; 347 /* 348 * if this process is higher priority and there is 349 * enough space, then select this process instead of 350 * the previous selection. 351 */ 352 if (pri > ppri) { 353 pp = p; 354 ppri = pri; 355 } 356 } 357 } 358 359 /* 360 * Nothing to do, back to sleep. 361 */ 362 if ((p = pp) == NULL) { 363 tsleep(&proc0, PVM, "sched", 0); 364 goto loop; 365 } 366 p->p_flag &= ~P_SWAPINREQ; 367 368 /* 369 * We would like to bring someone in. (only if there is space). 370 */ 371 faultin(p); 372 p->p_swtime = 0; 373 goto loop; 374} 375 376#ifndef NO_SWAPPING 377 378#define swappable(p) \ 379 (((p)->p_lock == 0) && \ 380 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 381 382/* 383 * Swapout is driven by the pageout daemon. Very simple, we find eligible 384 * procs and unwire their u-areas. We try to always "swap" at least one 385 * process in case we need the room for a swapin. 386 * If any procs have been sleeping/stopped for at least maxslp seconds, 387 * they are swapped. Else, we swap the longest-sleeping or stopped process, 388 * if any, otherwise the longest-resident process. 389 */ 390void 391swapout_procs() 392{ 393 register struct proc *p; 394 struct proc *outp, *outp2; 395 int outpri, outpri2; 396 int didswap = 0; 397 398 outp = outp2 = NULL; 399 outpri = outpri2 = INT_MIN; 400retry: 401 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 402 struct vmspace *vm; 403 if (!swappable(p)) 404 continue; 405 406 vm = p->p_vmspace; 407 408 switch (p->p_stat) { 409 default: 410 continue; 411 412 case SSLEEP: 413 case SSTOP: 414 /* 415 * do not swapout a realtime process 416 */ 417 if (p->p_rtprio.type == RTP_PRIO_REALTIME) 418 continue; 419 420 /* 421 * do not swapout a process waiting on a critical 422 * event of some kind 423 */ 424 if (((p->p_priority & 0x7f) < PSOCK) || 425 (p->p_slptime <= 10)) 426 continue; 427 428 ++vm->vm_refcnt; 429 vm_map_reference(&vm->vm_map); 430 /* 431 * do not swapout a process that is waiting for VM 432 * data structures there is a possible deadlock. 433 */ 434 if (lockmgr(&vm->vm_map.lock, 435 LK_EXCLUSIVE | LK_NOWAIT, 436 (void *)0, curproc)) { 437 vm_map_deallocate(&vm->vm_map); 438 vmspace_free(vm); 439 continue; 440 } 441 vm_map_unlock(&vm->vm_map); 442 /* 443 * If the process has been asleep for awhile and had 444 * most of its pages taken away already, swap it out. 445 */ 446 swapout(p); 447 vm_map_deallocate(&vm->vm_map); 448 vmspace_free(vm); 449 didswap++; 450 goto retry; 451 } 452 } 453 /* 454 * If we swapped something out, and another process needed memory, 455 * then wakeup the sched process. 456 */ 457 if (didswap) 458 wakeup(&proc0); 459} 460 461static void 462swapout(p) 463 register struct proc *p; 464{ 465 pmap_t pmap = &p->p_vmspace->vm_pmap; 466 int i; 467 468#if defined(SWAP_DEBUG) 469 printf("swapping out %d\n", p->p_pid); 470#endif 471 ++p->p_stats->p_ru.ru_nswap; 472 /* 473 * remember the process resident count 474 */ 475 p->p_vmspace->vm_swrss = 476 p->p_vmspace->vm_pmap.pm_stats.resident_count; 477 478 (void) splhigh(); 479 p->p_flag &= ~P_INMEM; 480 p->p_flag |= P_SWAPPING; 481 if (p->p_stat == SRUN) 482 remrq(p); 483 (void) spl0(); 484 485 pmap_swapout_proc(p); 486 487 p->p_flag &= ~P_SWAPPING; 488 p->p_swtime = 0; 489} 490#endif /* !NO_SWAPPING */ 491