vm_glue.c revision 32702
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 1.69 1997/12/11 02:10:55 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#include <sys/sysctl.h> 75 76#include <sys/kernel.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_prot.h> 84#include <sys/lock.h> 85#include <vm/pmap.h> 86#include <vm/vm_map.h> 87#include <vm/vm_page.h> 88#include <vm/vm_pageout.h> 89#include <vm/vm_kern.h> 90#include <vm/vm_extern.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 vm_map_lock_read(kernel_map); 130 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 131 vm_map_unlock_read(kernel_map); 132 return (rv == TRUE); 133} 134 135int 136useracc(addr, len, rw) 137 caddr_t addr; 138 int len, rw; 139{ 140 boolean_t rv; 141 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 142 vm_map_t map; 143 vm_map_entry_t save_hint; 144 145 /* 146 * XXX - check separately to disallow access to user area and user 147 * page tables - they are in the map. 148 * 149 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 150 * only used (as an end address) in trap.c. Use it as an end address 151 * here too. This bogusness has spread. I just fixed where it was 152 * used as a max in vm_mmap.c. 153 */ 154 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 155 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 156 return (FALSE); 157 } 158 map = &curproc->p_vmspace->vm_map; 159 vm_map_lock_read(map); 160 /* 161 * We save the map hint, and restore it. Useracc appears to distort 162 * the map hint unnecessarily. 163 */ 164 save_hint = map->hint; 165 rv = vm_map_check_protection(map, 166 trunc_page(addr), round_page(addr + len), prot); 167 map->hint = save_hint; 168 vm_map_unlock_read(map); 169 170 return (rv == TRUE); 171} 172 173void 174vslock(addr, len) 175 caddr_t addr; 176 u_int len; 177{ 178 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 179 round_page(addr + len), FALSE); 180} 181 182void 183vsunlock(addr, len, dirtied) 184 caddr_t addr; 185 u_int len; 186 int dirtied; 187{ 188#ifdef lint 189 dirtied++; 190#endif /* lint */ 191 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), 192 round_page(addr + len), TRUE); 193} 194 195/* 196 * Implement fork's actions on an address space. 197 * Here we arrange for the address space to be copied or referenced, 198 * allocate a user struct (pcb and kernel stack), then call the 199 * machine-dependent layer to fill those in and make the new process 200 * ready to run. The new process is set up so that it returns directly 201 * to user mode to avoid stack copying and relocation problems. 202 */ 203void 204vm_fork(p1, p2, flags) 205 register struct proc *p1, *p2; 206 int flags; 207{ 208 register struct user *up; 209 210 if (flags & RFMEM) { 211 p2->p_vmspace = p1->p_vmspace; 212 p1->p_vmspace->vm_refcnt++; 213 } 214 215 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 216 VM_WAIT; 217 } 218 219 if ((flags & RFMEM) == 0) { 220 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 221 222 if (p1->p_vmspace->vm_shm) 223 shmfork(p1, p2); 224 } 225 226 pmap_new_proc(p2); 227 228 up = p2->p_addr; 229 230 /* 231 * p_stats and p_sigacts currently point at fields in the user struct 232 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 233 * p_stats; zero the rest of p_stats (statistics). 234 */ 235 p2->p_stats = &up->u_stats; 236 p2->p_sigacts = &up->u_sigacts; 237 up->u_sigacts = *p1->p_sigacts; 238 bzero(&up->u_stats.pstat_startzero, 239 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 240 (caddr_t) &up->u_stats.pstat_startzero)); 241 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 242 ((caddr_t) &up->u_stats.pstat_endcopy - 243 (caddr_t) &up->u_stats.pstat_startcopy)); 244 245 246 /* 247 * cpu_fork will copy and update the pcb, set up the kernel stack, 248 * and make the child ready to run. 249 */ 250 cpu_fork(p1, p2); 251} 252 253/* 254 * Set default limits for VM system. 255 * Called for proc 0, and then inherited by all others. 256 * 257 * XXX should probably act directly on proc0. 258 */ 259static void 260vm_init_limits(udata) 261 void *udata; 262{ 263 register struct proc *p = udata; 264 int rss_limit; 265 266 /* 267 * Set up the initial limits on process VM. Set the maximum resident 268 * set size to be half of (reasonably) available memory. Since this 269 * is a soft limit, it comes into effect only when the system is out 270 * of memory - half of main memory helps to favor smaller processes, 271 * and reduces thrashing of the object cache. 272 */ 273 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 274 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 275 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 276 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 277 /* limit the limit to no less than 2MB */ 278 rss_limit = max(cnt.v_free_count, 512); 279 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 280 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 281} 282 283void 284faultin(p) 285 struct proc *p; 286{ 287 int s; 288 289 if ((p->p_flag & P_INMEM) == 0) { 290 291 ++p->p_lock; 292 293 pmap_swapin_proc(p); 294 295 s = splhigh(); 296 297 if (p->p_stat == SRUN) 298 setrunqueue(p); 299 300 p->p_flag |= P_INMEM; 301 302 /* undo the effect of setting SLOCK above */ 303 --p->p_lock; 304 splx(s); 305 306 } 307} 308 309/* 310 * This swapin algorithm attempts to swap-in processes only if there 311 * is enough space for them. Of course, if a process waits for a long 312 * time, it will be swapped in anyway. 313 */ 314/* ARGSUSED*/ 315static void 316scheduler(dummy) 317 void *dummy; 318{ 319 register struct proc *p; 320 register int pri; 321 struct proc *pp; 322 int ppri; 323 324loop: 325 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 326 VM_WAIT; 327 } 328 329 pp = NULL; 330 ppri = INT_MIN; 331 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 332 if (p->p_stat == SRUN && 333 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 334 int mempri; 335 336 pri = p->p_swtime + p->p_slptime; 337 if ((p->p_flag & P_SWAPINREQ) == 0) { 338 pri -= p->p_nice * 8; 339 } 340 mempri = pri > 0 ? pri : 0; 341 /* 342 * if this process is higher priority and there is 343 * enough space, then select this process instead of 344 * the previous selection. 345 */ 346 if (pri > ppri) { 347 pp = p; 348 ppri = pri; 349 } 350 } 351 } 352 353 /* 354 * Nothing to do, back to sleep. 355 */ 356 if ((p = pp) == NULL) { 357 tsleep(&proc0, PVM, "sched", 0); 358 goto loop; 359 } 360 p->p_flag &= ~P_SWAPINREQ; 361 362 /* 363 * We would like to bring someone in. (only if there is space). 364 */ 365 faultin(p); 366 p->p_swtime = 0; 367 goto loop; 368} 369 370#ifndef NO_SWAPPING 371 372#define swappable(p) \ 373 (((p)->p_lock == 0) && \ 374 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 375 376 377/* 378 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 379 */ 380int swap_idle_threshold1 = 2; 381SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 382 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 383 384/* 385 * Swap_idle_threshold2 is the time that a process can be idle before 386 * it will be swapped out, if idle swapping is enabled. 387 */ 388int swap_idle_threshold2 = 10; 389SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 390 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 391 392/* 393 * Swapout is driven by the pageout daemon. Very simple, we find eligible 394 * procs and unwire their u-areas. We try to always "swap" at least one 395 * process in case we need the room for a swapin. 396 * If any procs have been sleeping/stopped for at least maxslp seconds, 397 * they are swapped. Else, we swap the longest-sleeping or stopped process, 398 * if any, otherwise the longest-resident process. 399 */ 400void 401swapout_procs(action) 402int action; 403{ 404 register struct proc *p; 405 struct proc *outp, *outp2; 406 int outpri, outpri2; 407 int didswap = 0; 408 409 outp = outp2 = NULL; 410 outpri = outpri2 = INT_MIN; 411retry: 412 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 413 struct vmspace *vm; 414 if (!swappable(p)) 415 continue; 416 417 vm = p->p_vmspace; 418 419 switch (p->p_stat) { 420 default: 421 continue; 422 423 case SSLEEP: 424 case SSTOP: 425 /* 426 * do not swapout a realtime process 427 */ 428 if (p->p_rtprio.type == RTP_PRIO_REALTIME) 429 continue; 430 431 /* 432 * Do not swapout a process waiting on a critical 433 * event of some kind. Also guarantee swap_idle_threshold1 434 * time in memory. 435 */ 436 if (((p->p_priority & 0x7f) < PSOCK) || 437 (p->p_slptime < swap_idle_threshold1)) 438 continue; 439 440 /* 441 * If the system is under memory stress, or if we are swapping 442 * idle processes >= swap_idle_threshold2, then swap the process 443 * out. 444 */ 445 if (((action & VM_SWAP_NORMAL) == 0) && 446 (((action & VM_SWAP_IDLE) == 0) || 447 (p->p_slptime < swap_idle_threshold2))) 448 continue; 449 450 ++vm->vm_refcnt; 451 /* 452 * do not swapout a process that is waiting for VM 453 * data structures there is a possible deadlock. 454 */ 455 if (lockmgr(&vm->vm_map.lock, 456 LK_EXCLUSIVE | LK_NOWAIT, 457 (void *)0, curproc)) { 458 vmspace_free(vm); 459 continue; 460 } 461 vm_map_unlock(&vm->vm_map); 462 /* 463 * If the process has been asleep for awhile and had 464 * most of its pages taken away already, swap it out. 465 */ 466 if ((action & VM_SWAP_NORMAL) || 467 ((action & VM_SWAP_IDLE) && 468 (p->p_slptime > swap_idle_threshold2))) { 469 swapout(p); 470 vmspace_free(vm); 471 didswap++; 472 goto retry; 473 } 474 } 475 } 476 /* 477 * If we swapped something out, and another process needed memory, 478 * then wakeup the sched process. 479 */ 480 if (didswap) 481 wakeup(&proc0); 482} 483 484static void 485swapout(p) 486 register struct proc *p; 487{ 488 489#if defined(SWAP_DEBUG) 490 printf("swapping out %d\n", p->p_pid); 491#endif 492 ++p->p_stats->p_ru.ru_nswap; 493 /* 494 * remember the process resident count 495 */ 496 p->p_vmspace->vm_swrss = 497 p->p_vmspace->vm_pmap.pm_stats.resident_count; 498 499 (void) splhigh(); 500 p->p_flag &= ~P_INMEM; 501 p->p_flag |= P_SWAPPING; 502 if (p->p_stat == SRUN) 503 remrq(p); 504 (void) spl0(); 505 506 pmap_swapout_proc(p); 507 508 p->p_flag &= ~P_SWAPPING; 509 p->p_swtime = 0; 510} 511#endif /* !NO_SWAPPING */ 512