vm_glue.c revision 20821
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 1.55 1996/10/17 02:58:20 dyson 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 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/* 95 * System initialization 96 * 97 * Note: proc0 from proc.h 98 */ 99 100static void vm_init_limits __P((void *)); 101SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 102 103/* 104 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 105 * 106 * Note: run scheduling should be divorced from the vm system. 107 */ 108static void scheduler __P((void *)); 109SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 110 111 112static void swapout __P((struct proc *)); 113 114extern char kstack[]; 115 116/* vm_map_t upages_map; */ 117 118int 119kernacc(addr, len, rw) 120 caddr_t addr; 121 int len, rw; 122{ 123 boolean_t rv; 124 vm_offset_t saddr, eaddr; 125 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 126 127 saddr = trunc_page(addr); 128 eaddr = round_page(addr + len); 129 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 130 return (rv == TRUE); 131} 132 133int 134useracc(addr, len, rw) 135 caddr_t addr; 136 int len, rw; 137{ 138 boolean_t rv; 139 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 140 141 /* 142 * XXX - check separately to disallow access to user area and user 143 * page tables - they are in the map. 144 * 145 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 146 * only used (as an end address) in trap.c. Use it as an end address 147 * here too. This bogusness has spread. I just fixed where it was 148 * used as a max in vm_mmap.c. 149 */ 150 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 151 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 152 return (FALSE); 153 } 154 rv = vm_map_check_protection(&curproc->p_vmspace->vm_map, 155 trunc_page(addr), round_page(addr + len), prot); 156 return (rv == TRUE); 157} 158 159void 160vslock(addr, len) 161 caddr_t addr; 162 u_int len; 163{ 164 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 165 round_page(addr + len), FALSE); 166} 167 168void 169vsunlock(addr, len, dirtied) 170 caddr_t addr; 171 u_int len; 172 int dirtied; 173{ 174#ifdef lint 175 dirtied++; 176#endif /* lint */ 177 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 178 round_page(addr + len), TRUE); 179} 180 181/* 182 * Implement fork's actions on an address space. 183 * Here we arrange for the address space to be copied or referenced, 184 * allocate a user struct (pcb and kernel stack), then call the 185 * machine-dependent layer to fill those in and make the new process 186 * ready to run. 187 * NOTE: the kernel stack may be at a different location in the child 188 * process, and thus addresses of automatic variables may be invalid 189 * after cpu_fork returns in the child process. We do nothing here 190 * after cpu_fork returns. 191 */ 192int 193vm_fork(p1, p2) 194 register struct proc *p1, *p2; 195{ 196 register struct user *up; 197 int i; 198 pmap_t pvp; 199 vm_object_t upobj; 200 201 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 202 VM_WAIT; 203 } 204 205 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 206 207 if (p1->p_vmspace->vm_shm) 208 shmfork(p1, p2); 209 210 pmap_new_proc(p2); 211 212 up = p2->p_addr; 213 214 /* 215 * p_stats and p_sigacts currently point at fields in the user struct 216 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 217 * p_stats; zero the rest of p_stats (statistics). 218 */ 219 p2->p_stats = &up->u_stats; 220 p2->p_sigacts = &up->u_sigacts; 221 up->u_sigacts = *p1->p_sigacts; 222 bzero(&up->u_stats.pstat_startzero, 223 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 224 (caddr_t) &up->u_stats.pstat_startzero)); 225 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 226 ((caddr_t) &up->u_stats.pstat_endcopy - 227 (caddr_t) &up->u_stats.pstat_startcopy)); 228 229 230 /* 231 * cpu_fork will copy and update the kernel stack and pcb, and make 232 * the child ready to run. It marks the child so that it can return 233 * differently than the parent. It returns twice, once in the parent 234 * process and once in the child. 235 */ 236 return (cpu_fork(p1, p2)); 237} 238 239/* 240 * Set default limits for VM system. 241 * Called for proc 0, and then inherited by all others. 242 * 243 * XXX should probably act directly on proc0. 244 */ 245static void 246vm_init_limits(udata) 247 void *udata; 248{ 249 register struct proc *p = udata; 250 int rss_limit; 251 252 /* 253 * Set up the initial limits on process VM. Set the maximum resident 254 * set size to be half of (reasonably) available memory. Since this 255 * is a soft limit, it comes into effect only when the system is out 256 * of memory - half of main memory helps to favor smaller processes, 257 * and reduces thrashing of the object cache. 258 */ 259 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 260 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 261 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 262 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 263 /* limit the limit to no less than 2MB */ 264 rss_limit = max(cnt.v_free_count, 512); 265 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 266 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 267} 268 269void 270faultin(p) 271 struct proc *p; 272{ 273 vm_offset_t i; 274 int s; 275 276 if ((p->p_flag & P_INMEM) == 0) { 277 278 ++p->p_lock; 279 280 pmap_swapin_proc(p); 281 282 s = splhigh(); 283 284 if (p->p_stat == SRUN) 285 setrunqueue(p); 286 287 p->p_flag |= P_INMEM; 288 289 /* undo the effect of setting SLOCK above */ 290 --p->p_lock; 291 splx(s); 292 293 } 294} 295 296/* 297 * This swapin algorithm attempts to swap-in processes only if there 298 * is enough space for them. Of course, if a process waits for a long 299 * time, it will be swapped in anyway. 300 */ 301/* ARGSUSED*/ 302static void 303scheduler(dummy) 304 void *dummy; 305{ 306 register struct proc *p; 307 register int pri; 308 struct proc *pp; 309 int ppri; 310 311 spl0(); 312loop: 313 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 314 VM_WAIT; 315 } 316 317 pp = NULL; 318 ppri = INT_MIN; 319 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 320 if (p->p_stat == SRUN && 321 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 322 int mempri; 323 324 pri = p->p_swtime + p->p_slptime; 325 if ((p->p_flag & P_SWAPINREQ) == 0) { 326 pri -= p->p_nice * 8; 327 } 328 mempri = pri > 0 ? pri : 0; 329 /* 330 * if this process is higher priority and there is 331 * enough space, then select this process instead of 332 * the previous selection. 333 */ 334 if (pri > ppri) { 335 pp = p; 336 ppri = pri; 337 } 338 } 339 } 340 341 /* 342 * Nothing to do, back to sleep. 343 */ 344 if ((p = pp) == NULL) { 345 tsleep(&proc0, PVM, "sched", 0); 346 goto loop; 347 } 348 p->p_flag &= ~P_SWAPINREQ; 349 350 /* 351 * We would like to bring someone in. (only if there is space). 352 */ 353 faultin(p); 354 p->p_swtime = 0; 355 goto loop; 356} 357 358#ifndef NO_SWAPPING 359 360#define swappable(p) \ 361 (((p)->p_lock == 0) && \ 362 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 363 364/* 365 * Swapout is driven by the pageout daemon. Very simple, we find eligible 366 * procs and unwire their u-areas. We try to always "swap" at least one 367 * process in case we need the room for a swapin. 368 * If any procs have been sleeping/stopped for at least maxslp seconds, 369 * they are swapped. Else, we swap the longest-sleeping or stopped process, 370 * if any, otherwise the longest-resident process. 371 */ 372void 373swapout_procs() 374{ 375 register struct proc *p; 376 struct proc *outp, *outp2; 377 int outpri, outpri2; 378 int didswap = 0; 379 380 outp = outp2 = NULL; 381 outpri = outpri2 = INT_MIN; 382retry: 383 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 384 struct vmspace *vm; 385 if (!swappable(p)) 386 continue; 387 388 vm = p->p_vmspace; 389 390 switch (p->p_stat) { 391 default: 392 continue; 393 394 case SSLEEP: 395 case SSTOP: 396 /* 397 * do not swapout a realtime process 398 */ 399 if (p->p_rtprio.type == RTP_PRIO_REALTIME) 400 continue; 401 402 /* 403 * do not swapout a process waiting on a critical 404 * event of some kind 405 */ 406 if (((p->p_priority & 0x7f) < PSOCK) || 407 (p->p_slptime <= 10)) 408 continue; 409 410 ++vm->vm_refcnt; 411 vm_map_reference(&vm->vm_map); 412 /* 413 * do not swapout a process that is waiting for VM 414 * datastructures there is a possible deadlock. 415 */ 416 if (!lock_try_write(&vm->vm_map.lock)) { 417 vm_map_deallocate(&vm->vm_map); 418 vmspace_free(vm); 419 continue; 420 } 421 vm_map_unlock(&vm->vm_map); 422 /* 423 * If the process has been asleep for awhile and had 424 * most of its pages taken away already, swap it out. 425 */ 426 swapout(p); 427 vm_map_deallocate(&vm->vm_map); 428 vmspace_free(vm); 429 didswap++; 430 goto retry; 431 } 432 } 433 /* 434 * If we swapped something out, and another process needed memory, 435 * then wakeup the sched process. 436 */ 437 if (didswap) 438 wakeup(&proc0); 439} 440 441static void 442swapout(p) 443 register struct proc *p; 444{ 445 pmap_t pmap = &p->p_vmspace->vm_pmap; 446 int i; 447 448#if defined(SWAP_DEBUG) 449 printf("swapping out %d\n", p->p_pid); 450#endif 451 ++p->p_stats->p_ru.ru_nswap; 452 /* 453 * remember the process resident count 454 */ 455 p->p_vmspace->vm_swrss = 456 p->p_vmspace->vm_pmap.pm_stats.resident_count; 457 458 (void) splhigh(); 459 p->p_flag &= ~P_INMEM; 460 p->p_flag |= P_SWAPPING; 461 if (p->p_stat == SRUN) 462 remrq(p); 463 (void) spl0(); 464 465 pmap_swapout_proc(p); 466 467 p->p_flag &= ~P_SWAPPING; 468 p->p_swtime = 0; 469} 470#endif /* !NO_SWAPPING */ 471