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 1.76 1998/09/29 17:33:59 abial 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 */ 100 101 static void vm_init_limits __P((void *)); 102 SYSINIT(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 */ 109 static void scheduler __P((void *)); 110 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 111 112 113 static void swapout __P((struct proc *)); 114 115 extern char kstack[]; 116 117 /* vm_map_t upages_map; */ 118 119 int 120 kernacc(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 } 135 136 int 137 useracc(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 } 173 174 void 175 vslock(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 } 182 183 void 184 vsunlock(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 */ 204 void 205 vm_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 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 217 vm_pageout_deficit += (UPAGES + VM_INITIAL_PAGEIN); 218 VM_WAIT; 219 } 220 221 if ((flags & RFMEM) == 0) { 222 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 223 224 if (p1->p_vmspace->vm_shm) 225 shmfork(p1, p2); 226 } 227 228 pmap_new_proc(p2); 229 230 up = p2->p_addr; 231 232 /* 233 * p_stats and p_sigacts currently point at fields in the user struct 234 * but not at &u, instead at p_addr. Copy p_sigacts and parts of 235 * p_stats; zero the rest of p_stats (statistics). 236 */ 237 p2->p_stats = &up->u_stats; 238 p2->p_sigacts = &up->u_sigacts; 239 up->u_sigacts = *p1->p_sigacts; 240 bzero(&up->u_stats.pstat_startzero, 241 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 242 (caddr_t) &up->u_stats.pstat_startzero)); 243 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 244 ((caddr_t) &up->u_stats.pstat_endcopy - 245 (caddr_t) &up->u_stats.pstat_startcopy)); 246 247 248 /* 249 * cpu_fork will copy and update the pcb, set up the kernel stack, 250 * and make the child ready to run. 251 */ 252 cpu_fork(p1, p2); 253 } 254 255 /* 256 * Set default limits for VM system. 257 * Called for proc 0, and then inherited by all others. 258 * 259 * XXX should probably act directly on proc0. 260 */ 261 static void 262 vm_init_limits(udata) 263 void *udata; 264 { 265 register struct proc *p = udata; 266 int rss_limit; 267 268 /* 269 * Set up the initial limits on process VM. Set the maximum resident 270 * set size to be half of (reasonably) available memory. Since this 271 * is a soft limit, it comes into effect only when the system is out 272 * of memory - half of main memory helps to favor smaller processes, 273 * and reduces thrashing of the object cache. 274 */ 275 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 276 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 277 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 278 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 279 /* limit the limit to no less than 2MB */ 280 rss_limit = max(cnt.v_free_count, 512); 281 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 282 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 283 } 284 285 void 286 faultin(p) 287 struct proc *p; 288 { 289 int s; 290 291 if ((p->p_flag & P_INMEM) == 0) { 292 293 ++p->p_lock; 294 295 pmap_swapin_proc(p); 296 297 s = splhigh(); 298 299 if (p->p_stat == SRUN) 300 setrunqueue(p); 301 302 p->p_flag |= P_INMEM; 303 304 /* undo the effect of setting SLOCK above */ 305 --p->p_lock; 306 splx(s); 307 308 } 309 } 310 311 /* 312 * This swapin algorithm attempts to swap-in processes only if there 313 * is enough space for them. Of course, if a process waits for a long 314 * time, it will be swapped in anyway. 315 */ 316 /* ARGSUSED*/ 317 static void 318 scheduler(dummy) 319 void *dummy; 320 { 321 register struct proc *p; 322 register int pri; 323 struct proc *pp; 324 int ppri; 325 326 loop: 327 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 328 VM_WAIT; 329 } 330 331 pp = NULL; 332 ppri = INT_MIN; 333 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 334 if (p->p_stat == SRUN && 335 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 336 337 pri = p->p_swtime + p->p_slptime; 338 if ((p->p_flag & P_SWAPINREQ) == 0) { 339 pri -= p->p_nice * 8; 340 } 341 342 /* 343 * if this process is higher priority and there is 344 * enough space, then select this process instead of 345 * the previous selection. 346 */ 347 if (pri > ppri) { 348 pp = p; 349 ppri = pri; 350 } 351 } 352 } 353 354 /* 355 * Nothing to do, back to sleep. 356 */ 357 if ((p = pp) == NULL) { 358 tsleep(&proc0, PVM, "sched", 0); 359 goto loop; 360 } 361 p->p_flag &= ~P_SWAPINREQ; 362 363 /* 364 * We would like to bring someone in. (only if there is space). 365 */ 366 faultin(p); 367 p->p_swtime = 0; 368 goto loop; 369 } 370 371 #ifndef NO_SWAPPING 372 373 #define swappable(p) \ 374 (((p)->p_lock == 0) && \ 375 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM) 376 377 378 /* 379 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 380 */ 381 static int swap_idle_threshold1 = 2; 382 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 383 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 384 385 /* 386 * Swap_idle_threshold2 is the time that a process can be idle before 387 * it will be swapped out, if idle swapping is enabled. 388 */ 389 static int swap_idle_threshold2 = 10; 390 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 391 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 392 393 /* 394 * Swapout is driven by the pageout daemon. Very simple, we find eligible 395 * procs and unwire their u-areas. We try to always "swap" at least one 396 * process in case we need the room for a swapin. 397 * If any procs have been sleeping/stopped for at least maxslp seconds, 398 * they are swapped. Else, we swap the longest-sleeping or stopped process, 399 * if any, otherwise the longest-resident process. 400 */ 401 void 402 swapout_procs(action) 403 int action; 404 { 405 register struct proc *p; 406 struct proc *outp, *outp2; 407 int outpri, outpri2; 408 int didswap = 0; 409 410 outp = outp2 = NULL; 411 outpri = outpri2 = INT_MIN; 412 retry: 413 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 414 struct vmspace *vm; 415 if (!swappable(p)) 416 continue; 417 418 vm = p->p_vmspace; 419 420 switch (p->p_stat) { 421 default: 422 continue; 423 424 case SSLEEP: 425 case SSTOP: 426 /* 427 * do not swapout a realtime process 428 */ 429 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type)) 430 continue; 431 432 /* 433 * Do not swapout a process waiting on a critical 434 * event of some kind. Also guarantee swap_idle_threshold1 435 * time in memory. 436 */ 437 if (((p->p_priority & 0x7f) < PSOCK) || 438 (p->p_slptime < swap_idle_threshold1)) 439 continue; 440 441 /* 442 * If the system is under memory stress, or if we are swapping 443 * idle processes >= swap_idle_threshold2, then swap the process 444 * out. 445 */ 446 if (((action & VM_SWAP_NORMAL) == 0) && 447 (((action & VM_SWAP_IDLE) == 0) || 448 (p->p_slptime < swap_idle_threshold2))) 449 continue; 450 451 ++vm->vm_refcnt; 452 /* 453 * do not swapout a process that is waiting for VM 454 * data structures there is a possible deadlock. 455 */ 456 if (lockmgr(&vm->vm_map.lock, 457 LK_EXCLUSIVE | LK_NOWAIT, 458 (void *)0, curproc)) { 459 vmspace_free(vm); 460 continue; 461 } 462 vm_map_unlock(&vm->vm_map); 463 /* 464 * If the process has been asleep for awhile and had 465 * most of its pages taken away already, swap it out. 466 */ 467 if ((action & VM_SWAP_NORMAL) || 468 ((action & VM_SWAP_IDLE) && 469 (p->p_slptime > swap_idle_threshold2))) { 470 swapout(p); 471 vmspace_free(vm); 472 didswap++; 473 goto retry; 474 } 475 } 476 } 477 /* 478 * If we swapped something out, and another process needed memory, 479 * then wakeup the sched process. 480 */ 481 if (didswap) 482 wakeup(&proc0); 483 } 484 485 static void 486 swapout(p) 487 register struct proc *p; 488 { 489 490 #if defined(SWAP_DEBUG) 491 printf("swapping out %d\n", p->p_pid); 492 #endif 493 ++p->p_stats->p_ru.ru_nswap; 494 /* 495 * remember the process resident count 496 */ 497 p->p_vmspace->vm_swrss = 498 p->p_vmspace->vm_pmap.pm_stats.resident_count; 499 500 (void) splhigh(); 501 p->p_flag &= ~P_INMEM; 502 p->p_flag |= P_SWAPPING; 503 if (p->p_stat == SRUN) 504 remrq(p); 505 (void) spl0(); 506 507 pmap_swapout_proc(p); 508 509 p->p_flag &= ~P_SWAPPING; 510 p->p_swtime = 0; 511 } 512 #endif /* !NO_SWAPPING */ 513