vm_glue.c revision 24848
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 1.62 1997/04/07 07:16:04 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 <vm/vm.h> 80#include <vm/vm_param.h> 81#include <vm/vm_inherit.h> 82#include <vm/vm_prot.h> 83#include <sys/lock.h> 84#include <vm/pmap.h> 85#include <vm/vm_map.h> 86#include <vm/vm_page.h> 87#include <vm/vm_pageout.h> 88#include <vm/vm_kern.h> 89#include <vm/vm_extern.h> 90#include <vm/vm_object.h> 91#include <vm/vm_pager.h> 92 93#include <sys/user.h> 94 95/* 96 * System initialization 97 * 98 * Note: proc0 from proc.h 99 */ 100 101static void vm_init_limits __P((void *)); 102SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 103 104/* 105 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 106 * 107 * Note: run scheduling should be divorced from the vm system. 108 */ 109static void scheduler __P((void *)); 110SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 111 112 113static void swapout __P((struct proc *)); 114 115extern char kstack[]; 116 117/* vm_map_t upages_map; */ 118 119int 120kernacc(addr, len, rw) 121 caddr_t addr; 122 int len, rw; 123{ 124 boolean_t rv; 125 vm_offset_t saddr, eaddr; 126 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 127 128 saddr = trunc_page(addr); 129 eaddr = round_page(addr + len); 130 vm_map_lock_read(kernel_map); 131 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 132 vm_map_unlock_read(kernel_map); 133 return (rv == TRUE); 134} 135 136int 137useracc(addr, len, rw) 138 caddr_t addr; 139 int len, rw; 140{ 141 boolean_t rv; 142 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 143 vm_map_t map; 144 vm_map_entry_t save_hint; 145 146 /* 147 * XXX - check separately to disallow access to user area and user 148 * page tables - they are in the map. 149 * 150 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 151 * only used (as an end address) in trap.c. Use it as an end address 152 * here too. This bogusness has spread. I just fixed where it was 153 * used as a max in vm_mmap.c. 154 */ 155 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 156 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 157 return (FALSE); 158 } 159 map = &curproc->p_vmspace->vm_map; 160 vm_map_lock_read(map); 161 /* 162 * We save the map hint, and restore it. Useracc appears to distort 163 * the map hint unnecessarily. 164 */ 165 save_hint = map->hint; 166 rv = vm_map_check_protection(map, 167 trunc_page(addr), round_page(addr + len), prot); 168 map->hint = save_hint; 169 vm_map_unlock_read(map); 170 171 return (rv == TRUE); 172} 173 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. The new process is set up so that it returns directly 202 * to user mode to avoid stack copying and relocation problems. 203 */ 204void 205vm_fork(p1, p2, flags) 206 register struct proc *p1, *p2; 207 int flags; 208{ 209 register struct user *up; 210 int i; 211 pmap_t pvp; 212 vm_object_t upobj; 213 214 if (flags & RFMEM) { 215 p2->p_vmspace = p1->p_vmspace; 216 p1->p_vmspace->vm_refcnt++; 217 } 218 219 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 220 VM_WAIT; 221 } 222 223 if ((flags & RFMEM) == 0) { 224 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 225 226 if (p1->p_vmspace->vm_shm) 227 shmfork(p1, p2); 228 } 229 230 pmap_new_proc(p2); 231 232 up = p2->p_addr; 233 234 /* 235 * p_stats and p_sigacts currently point at fields in the user struct 236 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 237 * p_stats; zero the rest of p_stats (statistics). 238 */ 239 p2->p_stats = &up->u_stats; 240 p2->p_sigacts = &up->u_sigacts; 241 up->u_sigacts = *p1->p_sigacts; 242 bzero(&up->u_stats.pstat_startzero, 243 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 244 (caddr_t) &up->u_stats.pstat_startzero)); 245 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 246 ((caddr_t) &up->u_stats.pstat_endcopy - 247 (caddr_t) &up->u_stats.pstat_startcopy)); 248 249 250 /* 251 * cpu_fork will copy and update the pcb, set up the kernel stack, 252 * and make the child ready to run. 253 */ 254 cpu_fork(p1, p2); 255} 256 257/* 258 * Set default limits for VM system. 259 * Called for proc 0, and then inherited by all others. 260 * 261 * XXX should probably act directly on proc0. 262 */ 263static void 264vm_init_limits(udata) 265 void *udata; 266{ 267 register struct proc *p = udata; 268 int rss_limit; 269 270 /* 271 * Set up the initial limits on process VM. Set the maximum resident 272 * set size to be half of (reasonably) available memory. Since this 273 * is a soft limit, it comes into effect only when the system is out 274 * of memory - half of main memory helps to favor smaller processes, 275 * and reduces thrashing of the object cache. 276 */ 277 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 278 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 279 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 280 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 281 /* limit the limit to no less than 2MB */ 282 rss_limit = max(cnt.v_free_count, 512); 283 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 284 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 285} 286 287void 288faultin(p) 289 struct proc *p; 290{ 291 vm_offset_t i; 292 int s; 293 294 if ((p->p_flag & P_INMEM) == 0) { 295 296 ++p->p_lock; 297 298 pmap_swapin_proc(p); 299 300 s = splhigh(); 301 302 if (p->p_stat == SRUN) 303 setrunqueue(p); 304 305 p->p_flag |= P_INMEM; 306 307 /* undo the effect of setting SLOCK above */ 308 --p->p_lock; 309 splx(s); 310 311 } 312} 313 314/* 315 * This swapin algorithm attempts to swap-in processes only if there 316 * is enough space for them. Of course, if a process waits for a long 317 * time, it will be swapped in anyway. 318 */ 319/* ARGSUSED*/ 320static void 321scheduler(dummy) 322 void *dummy; 323{ 324 register struct proc *p; 325 register int pri; 326 struct proc *pp; 327 int ppri; 328 329loop: 330 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 331 VM_WAIT; 332 } 333 334 pp = NULL; 335 ppri = INT_MIN; 336 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 337 if (p->p_stat == SRUN && 338 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 339 int mempri; 340 341 pri = p->p_swtime + p->p_slptime; 342 if ((p->p_flag & P_SWAPINREQ) == 0) { 343 pri -= p->p_nice * 8; 344 } 345 mempri = pri > 0 ? pri : 0; 346 /* 347 * if this process is higher priority and there is 348 * enough space, then select this process instead of 349 * the previous selection. 350 */ 351 if (pri > ppri) { 352 pp = p; 353 ppri = pri; 354 } 355 } 356 } 357 358 /* 359 * Nothing to do, back to sleep. 360 */ 361 if ((p = pp) == NULL) { 362 tsleep(&proc0, PVM, "sched", 0); 363 goto loop; 364 } 365 p->p_flag &= ~P_SWAPINREQ; 366 367 /* 368 * We would like to bring someone in. (only if there is space). 369 */ 370 faultin(p); 371 p->p_swtime = 0; 372 goto loop; 373} 374 375#ifndef NO_SWAPPING 376 377#define swappable(p) \ 378 (((p)->p_lock == 0) && \ 379 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 380 381/* 382 * Swapout is driven by the pageout daemon. Very simple, we find eligible 383 * procs and unwire their u-areas. We try to always "swap" at least one 384 * process in case we need the room for a swapin. 385 * If any procs have been sleeping/stopped for at least maxslp seconds, 386 * they are swapped. Else, we swap the longest-sleeping or stopped process, 387 * if any, otherwise the longest-resident process. 388 */ 389void 390swapout_procs() 391{ 392 register struct proc *p; 393 struct proc *outp, *outp2; 394 int outpri, outpri2; 395 int didswap = 0; 396 397 outp = outp2 = NULL; 398 outpri = outpri2 = INT_MIN; 399retry: 400 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 401 struct vmspace *vm; 402 if (!swappable(p)) 403 continue; 404 405 vm = p->p_vmspace; 406 407 switch (p->p_stat) { 408 default: 409 continue; 410 411 case SSLEEP: 412 case SSTOP: 413 /* 414 * do not swapout a realtime process 415 */ 416 if (p->p_rtprio.type == RTP_PRIO_REALTIME) 417 continue; 418 419 /* 420 * do not swapout a process waiting on a critical 421 * event of some kind 422 */ 423 if (((p->p_priority & 0x7f) < PSOCK) || 424 (p->p_slptime <= 10)) 425 continue; 426 427 ++vm->vm_refcnt; 428 vm_map_reference(&vm->vm_map); 429 /* 430 * do not swapout a process that is waiting for VM 431 * data structures there is a possible deadlock. 432 */ 433 if (lockmgr(&vm->vm_map.lock, 434 LK_EXCLUSIVE | LK_NOWAIT, 435 (void *)0, curproc)) { 436 vm_map_deallocate(&vm->vm_map); 437 vmspace_free(vm); 438 continue; 439 } 440 vm_map_unlock(&vm->vm_map); 441 /* 442 * If the process has been asleep for awhile and had 443 * most of its pages taken away already, swap it out. 444 */ 445 swapout(p); 446 vm_map_deallocate(&vm->vm_map); 447 vmspace_free(vm); 448 didswap++; 449 goto retry; 450 } 451 } 452 /* 453 * If we swapped something out, and another process needed memory, 454 * then wakeup the sched process. 455 */ 456 if (didswap) 457 wakeup(&proc0); 458} 459 460static void 461swapout(p) 462 register struct proc *p; 463{ 464 pmap_t pmap = &p->p_vmspace->vm_pmap; 465 int i; 466 467#if defined(SWAP_DEBUG) 468 printf("swapping out %d\n", p->p_pid); 469#endif 470 ++p->p_stats->p_ru.ru_nswap; 471 /* 472 * remember the process resident count 473 */ 474 p->p_vmspace->vm_swrss = 475 p->p_vmspace->vm_pmap.pm_stats.resident_count; 476 477 (void) splhigh(); 478 p->p_flag &= ~P_INMEM; 479 p->p_flag |= P_SWAPPING; 480 if (p->p_stat == SRUN) 481 remrq(p); 482 (void) spl0(); 483 484 pmap_swapout_proc(p); 485 486 p->p_flag &= ~P_SWAPPING; 487 p->p_swtime = 0; 488} 489#endif /* !NO_SWAPPING */ 490