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 1.85 1999/04/06 03:11:34 peter 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 int 116 kernacc(addr, len, rw) 117 caddr_t addr; 118 int len, rw; 119 { 120 boolean_t rv; 121 vm_offset_t saddr, eaddr; 122 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 123 124 saddr = trunc_page((vm_offset_t)addr); 125 eaddr = round_page((vm_offset_t)addr + len); 126 vm_map_lock_read(kernel_map); 127 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 128 vm_map_unlock_read(kernel_map); 129 return (rv == TRUE); 130 } 131 132 int 133 useracc(addr, len, rw) 134 caddr_t addr; 135 int len, rw; 136 { 137 boolean_t rv; 138 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; 139 vm_map_t map; 140 vm_map_entry_t save_hint; 141 142 /* 143 * XXX - check separately to disallow access to user area and user 144 * page tables - they are in the map. 145 * 146 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 147 * only used (as an end address) in trap.c. Use it as an end address 148 * here too. This bogusness has spread. I just fixed where it was 149 * used as a max in vm_mmap.c. 150 */ 151 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 152 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 153 return (FALSE); 154 } 155 map = &curproc->p_vmspace->vm_map; 156 vm_map_lock_read(map); 157 /* 158 * We save the map hint, and restore it. Useracc appears to distort 159 * the map hint unnecessarily. 160 */ 161 save_hint = map->hint; 162 rv = vm_map_check_protection(map, 163 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 164 map->hint = save_hint; 165 vm_map_unlock_read(map); 166 167 return (rv == TRUE); 168 } 169 170 void 171 vslock(addr, len) 172 caddr_t addr; 173 u_int len; 174 { 175 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 176 round_page((vm_offset_t)addr + len), FALSE); 177 } 178 179 void 180 vsunlock(addr, len, dirtied) 181 caddr_t addr; 182 u_int len; 183 int dirtied; 184 { 185 #ifdef lint 186 dirtied++; 187 #endif /* lint */ 188 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 189 round_page((vm_offset_t)addr + len), TRUE); 190 } 191 192 /* 193 * Implement fork's actions on an address space. 194 * Here we arrange for the address space to be copied or referenced, 195 * allocate a user struct (pcb and kernel stack), then call the 196 * machine-dependent layer to fill those in and make the new process 197 * ready to run. The new process is set up so that it returns directly 198 * to user mode to avoid stack copying and relocation problems. 199 */ 200 void 201 vm_fork(p1, p2, flags) 202 register struct proc *p1, *p2; 203 int flags; 204 { 205 register struct user *up; 206 207 if (flags & RFMEM) { 208 p2->p_vmspace = p1->p_vmspace; 209 p1->p_vmspace->vm_refcnt++; 210 } 211 212 /* 213 * Great, so we have a memory-heavy process and the 214 * entire machine comes to a screaching halt because 215 * nobody can fork/exec anything. What we really need 216 * to do is fix the process swapper so it swaps out the right 217 * processes. 218 */ 219 #if 0 220 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 221 vm_pageout_deficit += (UPAGES + VM_INITIAL_PAGEIN); 222 VM_WAIT; 223 } 224 #endif 225 226 if ((flags & RFMEM) == 0) { 227 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 228 229 if (p1->p_vmspace->vm_shm) 230 shmfork(p1, p2); 231 } 232 233 pmap_new_proc(p2); 234 235 up = p2->p_addr; 236 237 /* 238 * p_stats currently points at fields in the user struct 239 * but not at &u, instead at p_addr. Copy parts of 240 * p_stats; zero the rest of p_stats (statistics). 241 * 242 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 243 * to share sigacts, so we use the up->u_sigacts. 244 */ 245 p2->p_stats = &up->u_stats; 246 if (p2->p_sigacts == NULL) { 247 if (p2->p_procsig->ps_refcnt != 1) 248 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 249 p2->p_sigacts = &up->u_sigacts; 250 up->u_sigacts = *p1->p_sigacts; 251 } 252 253 bzero(&up->u_stats.pstat_startzero, 254 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 255 (caddr_t) &up->u_stats.pstat_startzero)); 256 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 257 ((caddr_t) &up->u_stats.pstat_endcopy - 258 (caddr_t) &up->u_stats.pstat_startcopy)); 259 260 261 /* 262 * cpu_fork will copy and update the pcb, set up the kernel stack, 263 * and make the child ready to run. 264 */ 265 cpu_fork(p1, p2); 266 } 267 268 /* 269 * Set default limits for VM system. 270 * Called for proc 0, and then inherited by all others. 271 * 272 * XXX should probably act directly on proc0. 273 */ 274 static void 275 vm_init_limits(udata) 276 void *udata; 277 { 278 register struct proc *p = udata; 279 int rss_limit; 280 281 /* 282 * Set up the initial limits on process VM. Set the maximum resident 283 * set size to be half of (reasonably) available memory. Since this 284 * is a soft limit, it comes into effect only when the system is out 285 * of memory - half of main memory helps to favor smaller processes, 286 * and reduces thrashing of the object cache. 287 */ 288 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 289 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 290 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 291 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 292 /* limit the limit to no less than 2MB */ 293 rss_limit = max(cnt.v_free_count, 512); 294 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 295 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 296 } 297 298 void 299 faultin(p) 300 struct proc *p; 301 { 302 int s; 303 304 if ((p->p_flag & P_INMEM) == 0) { 305 306 ++p->p_lock; 307 308 pmap_swapin_proc(p); 309 310 s = splhigh(); 311 312 if (p->p_stat == SRUN) 313 setrunqueue(p); 314 315 p->p_flag |= P_INMEM; 316 317 /* undo the effect of setting SLOCK above */ 318 --p->p_lock; 319 splx(s); 320 321 } 322 } 323 324 /* 325 * This swapin algorithm attempts to swap-in processes only if there 326 * is enough space for them. Of course, if a process waits for a long 327 * time, it will be swapped in anyway. 328 */ 329 /* ARGSUSED*/ 330 static void 331 scheduler(dummy) 332 void *dummy; 333 { 334 register struct proc *p; 335 register int pri; 336 struct proc *pp; 337 int ppri; 338 339 loop: 340 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) { 341 VM_WAIT; 342 } 343 344 pp = NULL; 345 ppri = INT_MIN; 346 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 347 if (p->p_stat == SRUN && 348 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 349 350 pri = p->p_swtime + p->p_slptime; 351 if ((p->p_flag & P_SWAPINREQ) == 0) { 352 pri -= p->p_nice * 8; 353 } 354 355 /* 356 * if this process is higher priority and there is 357 * enough space, then select this process instead of 358 * the previous selection. 359 */ 360 if (pri > ppri) { 361 pp = p; 362 ppri = pri; 363 } 364 } 365 } 366 367 /* 368 * Nothing to do, back to sleep. 369 */ 370 if ((p = pp) == NULL) { 371 tsleep(&proc0, PVM, "sched", 0); 372 goto loop; 373 } 374 p->p_flag &= ~P_SWAPINREQ; 375 376 /* 377 * We would like to bring someone in. (only if there is space). 378 */ 379 faultin(p); 380 p->p_swtime = 0; 381 goto loop; 382 } 383 384 #ifndef NO_SWAPPING 385 386 #define swappable(p) \ 387 (((p)->p_lock == 0) && \ 388 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM) 389 390 391 /* 392 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 393 */ 394 static int swap_idle_threshold1 = 2; 395 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 396 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 397 398 /* 399 * Swap_idle_threshold2 is the time that a process can be idle before 400 * it will be swapped out, if idle swapping is enabled. 401 */ 402 static int swap_idle_threshold2 = 10; 403 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 404 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 405 406 /* 407 * Swapout is driven by the pageout daemon. Very simple, we find eligible 408 * procs and unwire their u-areas. We try to always "swap" at least one 409 * process in case we need the room for a swapin. 410 * If any procs have been sleeping/stopped for at least maxslp seconds, 411 * they are swapped. Else, we swap the longest-sleeping or stopped process, 412 * if any, otherwise the longest-resident process. 413 */ 414 void 415 swapout_procs(action) 416 int action; 417 { 418 register struct proc *p; 419 struct proc *outp, *outp2; 420 int outpri, outpri2; 421 int didswap = 0; 422 423 outp = outp2 = NULL; 424 outpri = outpri2 = INT_MIN; 425 retry: 426 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 427 struct vmspace *vm; 428 if (!swappable(p)) 429 continue; 430 431 vm = p->p_vmspace; 432 433 switch (p->p_stat) { 434 default: 435 continue; 436 437 case SSLEEP: 438 case SSTOP: 439 /* 440 * do not swapout a realtime process 441 */ 442 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type)) 443 continue; 444 445 /* 446 * Do not swapout a process waiting on a critical 447 * event of some kind. Also guarantee swap_idle_threshold1 448 * time in memory. 449 */ 450 if (((p->p_priority & 0x7f) < PSOCK) || 451 (p->p_slptime < swap_idle_threshold1)) 452 continue; 453 454 /* 455 * If the system is under memory stress, or if we are swapping 456 * idle processes >= swap_idle_threshold2, then swap the process 457 * out. 458 */ 459 if (((action & VM_SWAP_NORMAL) == 0) && 460 (((action & VM_SWAP_IDLE) == 0) || 461 (p->p_slptime < swap_idle_threshold2))) 462 continue; 463 464 ++vm->vm_refcnt; 465 /* 466 * do not swapout a process that is waiting for VM 467 * data structures there is a possible deadlock. 468 */ 469 if (lockmgr(&vm->vm_map.lock, 470 LK_EXCLUSIVE | LK_NOWAIT, 471 (void *)0, curproc)) { 472 vmspace_free(vm); 473 continue; 474 } 475 vm_map_unlock(&vm->vm_map); 476 /* 477 * If the process has been asleep for awhile and had 478 * most of its pages taken away already, swap it out. 479 */ 480 if ((action & VM_SWAP_NORMAL) || 481 ((action & VM_SWAP_IDLE) && 482 (p->p_slptime > swap_idle_threshold2))) { 483 swapout(p); 484 vmspace_free(vm); 485 didswap++; 486 goto retry; 487 } 488 } 489 } 490 /* 491 * If we swapped something out, and another process needed memory, 492 * then wakeup the sched process. 493 */ 494 if (didswap) 495 wakeup(&proc0); 496 } 497 498 static void 499 swapout(p) 500 register struct proc *p; 501 { 502 503 #if defined(SWAP_DEBUG) 504 printf("swapping out %d\n", p->p_pid); 505 #endif 506 ++p->p_stats->p_ru.ru_nswap; 507 /* 508 * remember the process resident count 509 */ 510 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 511 512 (void) splhigh(); 513 p->p_flag &= ~P_INMEM; 514 p->p_flag |= P_SWAPPING; 515 if (p->p_stat == SRUN) 516 remrq(p); 517 (void) spl0(); 518 519 pmap_swapout_proc(p); 520 521 p->p_flag &= ~P_SWAPPING; 522 p->p_swtime = 0; 523 } 524 #endif /* !NO_SWAPPING */ 525