vm_glue.c revision 42968
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 1.81 1999/01/21 08:29:10 dillon Exp $ 63 */ 64 65#include "opt_rlimit.h" 66#include "opt_vm.h" 67 68#include <sys/param.h> 69#include <sys/systm.h> 70#include <sys/proc.h> 71#include <sys/resourcevar.h> 72#include <sys/buf.h> 73#include <sys/shm.h> 74#include <sys/vmmeter.h> 75#include <sys/sysctl.h> 76 77#include <sys/kernel.h> 78#include <sys/unistd.h> 79 80#include <machine/limits.h> 81 82#include <vm/vm.h> 83#include <vm/vm_param.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 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((vm_offset_t)addr); 129 eaddr = round_page((vm_offset_t)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((vm_offset_t)addr), round_page((vm_offset_t)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((vm_offset_t)addr), 180 round_page((vm_offset_t)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((vm_offset_t)addr), 193 round_page((vm_offset_t)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; 210 211 if (flags & RFMEM) { 212 p2->p_vmspace = p1->p_vmspace; 213 p1->p_vmspace->vm_refcnt++; 214 } 215 216 /* 217 * Great, so we have a memory-heavy process and the 218 * entire machine comes to a screaching halt because 219 * nobody can fork/exec anything. What we really need 220 * to do is fix the process swapper so it swaps out the right 221 * processes. 222 */ 223#if 0 224 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 225 vm_pageout_deficit += (UPAGES + VM_INITIAL_PAGEIN); 226 VM_WAIT; 227 } 228#endif 229 230 if ((flags & RFMEM) == 0) { 231 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 232 233 if (p1->p_vmspace->vm_shm) 234 shmfork(p1, p2); 235 } 236 237 pmap_new_proc(p2); 238 239 up = p2->p_addr; 240 241#ifndef COMPAT_LINUX_THREADS 242 /* 243 * p_stats and p_sigacts currently point at fields in the user struct 244 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 245 * p_stats; zero the rest of p_stats (statistics). 246 */ 247 p2->p_stats = &up->u_stats; 248 p2->p_sigacts = &up->u_sigacts; 249 up->u_sigacts = *p1->p_sigacts; 250#else 251 /* 252 * p_stats currently points at fields in the user struct 253 * but not at &u, instead at p_addr. Copy parts of 254 * p_stats; zero the rest of p_stats (statistics). 255 * 256 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 257 * to share sigacts, so we use the up->u_sigacts. 258 */ 259 p2->p_stats = &up->u_stats; 260 if (p2->p_sigacts == NULL) { 261 if (p2->p_procsig->ps_refcnt != 1) 262 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 263 p2->p_sigacts = &up->u_sigacts; 264 up->u_sigacts = *p1->p_sigacts; 265 } 266#endif /* COMPAT_LINUX_THREADS */ 267 bzero(&up->u_stats.pstat_startzero, 268 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 269 (caddr_t) &up->u_stats.pstat_startzero)); 270 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 271 ((caddr_t) &up->u_stats.pstat_endcopy - 272 (caddr_t) &up->u_stats.pstat_startcopy)); 273 274 275 /* 276 * cpu_fork will copy and update the pcb, set up the kernel stack, 277 * and make the child ready to run. 278 */ 279 cpu_fork(p1, p2); 280} 281 282/* 283 * Set default limits for VM system. 284 * Called for proc 0, and then inherited by all others. 285 * 286 * XXX should probably act directly on proc0. 287 */ 288static void 289vm_init_limits(udata) 290 void *udata; 291{ 292 register struct proc *p = udata; 293 int rss_limit; 294 295 /* 296 * Set up the initial limits on process VM. Set the maximum resident 297 * set size to be half of (reasonably) available memory. Since this 298 * is a soft limit, it comes into effect only when the system is out 299 * of memory - half of main memory helps to favor smaller processes, 300 * and reduces thrashing of the object cache. 301 */ 302 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 303 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 304 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 305 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 306 /* limit the limit to no less than 2MB */ 307 rss_limit = max(cnt.v_free_count, 512); 308 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 309 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 310} 311 312void 313faultin(p) 314 struct proc *p; 315{ 316 int s; 317 318 if ((p->p_flag & P_INMEM) == 0) { 319 320 ++p->p_lock; 321 322 pmap_swapin_proc(p); 323 324 s = splhigh(); 325 326 if (p->p_stat == SRUN) 327 setrunqueue(p); 328 329 p->p_flag |= P_INMEM; 330 331 /* undo the effect of setting SLOCK above */ 332 --p->p_lock; 333 splx(s); 334 335 } 336} 337 338/* 339 * This swapin algorithm attempts to swap-in processes only if there 340 * is enough space for them. Of course, if a process waits for a long 341 * time, it will be swapped in anyway. 342 */ 343/* ARGSUSED*/ 344static void 345scheduler(dummy) 346 void *dummy; 347{ 348 register struct proc *p; 349 register int pri; 350 struct proc *pp; 351 int ppri; 352 353loop: 354 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 355 VM_WAIT; 356 } 357 358 pp = NULL; 359 ppri = INT_MIN; 360 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 361 if (p->p_stat == SRUN && 362 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 363 364 pri = p->p_swtime + p->p_slptime; 365 if ((p->p_flag & P_SWAPINREQ) == 0) { 366 pri -= p->p_nice * 8; 367 } 368 369 /* 370 * if this process is higher priority and there is 371 * enough space, then select this process instead of 372 * the previous selection. 373 */ 374 if (pri > ppri) { 375 pp = p; 376 ppri = pri; 377 } 378 } 379 } 380 381 /* 382 * Nothing to do, back to sleep. 383 */ 384 if ((p = pp) == NULL) { 385 tsleep(&proc0, PVM, "sched", 0); 386 goto loop; 387 } 388 p->p_flag &= ~P_SWAPINREQ; 389 390 /* 391 * We would like to bring someone in. (only if there is space). 392 */ 393 faultin(p); 394 p->p_swtime = 0; 395 goto loop; 396} 397 398#ifndef NO_SWAPPING 399 400#define swappable(p) \ 401 (((p)->p_lock == 0) && \ 402 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 403 404 405/* 406 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 407 */ 408static int swap_idle_threshold1 = 2; 409SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 410 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 411 412/* 413 * Swap_idle_threshold2 is the time that a process can be idle before 414 * it will be swapped out, if idle swapping is enabled. 415 */ 416static int swap_idle_threshold2 = 10; 417SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 418 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 419 420/* 421 * Swapout is driven by the pageout daemon. Very simple, we find eligible 422 * procs and unwire their u-areas. We try to always "swap" at least one 423 * process in case we need the room for a swapin. 424 * If any procs have been sleeping/stopped for at least maxslp seconds, 425 * they are swapped. Else, we swap the longest-sleeping or stopped process, 426 * if any, otherwise the longest-resident process. 427 */ 428void 429swapout_procs(action) 430int action; 431{ 432 register struct proc *p; 433 struct proc *outp, *outp2; 434 int outpri, outpri2; 435 int didswap = 0; 436 437 outp = outp2 = NULL; 438 outpri = outpri2 = INT_MIN; 439retry: 440 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 441 struct vmspace *vm; 442 if (!swappable(p)) 443 continue; 444 445 vm = p->p_vmspace; 446 447 switch (p->p_stat) { 448 default: 449 continue; 450 451 case SSLEEP: 452 case SSTOP: 453 /* 454 * do not swapout a realtime process 455 */ 456 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type)) 457 continue; 458 459 /* 460 * Do not swapout a process waiting on a critical 461 * event of some kind. Also guarantee swap_idle_threshold1 462 * time in memory. 463 */ 464 if (((p->p_priority & 0x7f) < PSOCK) || 465 (p->p_slptime < swap_idle_threshold1)) 466 continue; 467 468 /* 469 * If the system is under memory stress, or if we are swapping 470 * idle processes >= swap_idle_threshold2, then swap the process 471 * out. 472 */ 473 if (((action & VM_SWAP_NORMAL) == 0) && 474 (((action & VM_SWAP_IDLE) == 0) || 475 (p->p_slptime < swap_idle_threshold2))) 476 continue; 477 478 ++vm->vm_refcnt; 479 /* 480 * do not swapout a process that is waiting for VM 481 * data structures there is a possible deadlock. 482 */ 483 if (lockmgr(&vm->vm_map.lock, 484 LK_EXCLUSIVE | LK_NOWAIT, 485 (void *)0, curproc)) { 486 vmspace_free(vm); 487 continue; 488 } 489 vm_map_unlock(&vm->vm_map); 490 /* 491 * If the process has been asleep for awhile and had 492 * most of its pages taken away already, swap it out. 493 */ 494 if ((action & VM_SWAP_NORMAL) || 495 ((action & VM_SWAP_IDLE) && 496 (p->p_slptime > swap_idle_threshold2))) { 497 swapout(p); 498 vmspace_free(vm); 499 didswap++; 500 goto retry; 501 } 502 } 503 } 504 /* 505 * If we swapped something out, and another process needed memory, 506 * then wakeup the sched process. 507 */ 508 if (didswap) 509 wakeup(&proc0); 510} 511 512static void 513swapout(p) 514 register struct proc *p; 515{ 516 517#if defined(SWAP_DEBUG) 518 printf("swapping out %d\n", p->p_pid); 519#endif 520 ++p->p_stats->p_ru.ru_nswap; 521 /* 522 * remember the process resident count 523 */ 524 p->p_vmspace->vm_swrss = 525 p->p_vmspace->vm_pmap.pm_stats.resident_count; 526 527 (void) splhigh(); 528 p->p_flag &= ~P_INMEM; 529 p->p_flag |= P_SWAPPING; 530 if (p->p_stat == SRUN) 531 remrq(p); 532 (void) spl0(); 533 534 pmap_swapout_proc(p); 535 536 p->p_flag &= ~P_SWAPPING; 537 p->p_swtime = 0; 538} 539#endif /* !NO_SWAPPING */ 540