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 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $ 63 * $DragonFly: src/sys/vm/vm_glue.c,v 1.14 2003/07/26 22:10:02 rob Exp $ 64 */ 65 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 <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 #include <vm/vm_page2.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; 123 124 KASSERT((rw & (~VM_PROT_ALL)) == 0, 125 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 126 prot = rw; 127 saddr = trunc_page((vm_offset_t)addr); 128 eaddr = round_page((vm_offset_t)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 } 134 135 int 136 useracc(addr, len, rw) 137 caddr_t addr; 138 int len, rw; 139 { 140 boolean_t rv; 141 vm_prot_t prot; 142 vm_map_t map; 143 vm_map_entry_t save_hint; 144 145 KASSERT((rw & (~VM_PROT_ALL)) == 0, 146 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 147 prot = rw; 148 /* 149 * XXX - check separately to disallow access to user area and user 150 * page tables - they are in the map. 151 * 152 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 153 * only used (as an end address) in trap.c. Use it as an end address 154 * here too. This bogusness has spread. I just fixed where it was 155 * used as a max in vm_mmap.c. 156 */ 157 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 158 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 159 return (FALSE); 160 } 161 map = &curproc->p_vmspace->vm_map; 162 vm_map_lock_read(map); 163 /* 164 * We save the map hint, and restore it. Useracc appears to distort 165 * the map hint unnecessarily. 166 */ 167 save_hint = map->hint; 168 rv = vm_map_check_protection(map, 169 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 170 map->hint = save_hint; 171 vm_map_unlock_read(map); 172 173 return (rv == TRUE); 174 } 175 176 void 177 vslock(addr, len) 178 caddr_t addr; 179 u_int len; 180 { 181 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 182 round_page((vm_offset_t)addr + len), FALSE); 183 } 184 185 void 186 vsunlock(addr, len) 187 caddr_t addr; 188 u_int len; 189 { 190 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 191 round_page((vm_offset_t)addr + len), TRUE); 192 } 193 194 /* 195 * Implement fork's actions on an address space. 196 * Here we arrange for the address space to be copied or referenced, 197 * allocate a user struct (pcb and kernel stack), then call the 198 * machine-dependent layer to fill those in and make the new process 199 * ready to run. The new process is set up so that it returns directly 200 * to user mode to avoid stack copying and relocation problems. 201 */ 202 void 203 vm_fork(p1, p2, flags) 204 struct proc *p1, *p2; 205 int flags; 206 { 207 struct user *up; 208 struct thread *td2; 209 210 if ((flags & RFPROC) == 0) { 211 /* 212 * Divorce the memory, if it is shared, essentially 213 * this changes shared memory amongst threads, into 214 * COW locally. 215 */ 216 if ((flags & RFMEM) == 0) { 217 if (p1->p_vmspace->vm_refcnt > 1) { 218 vmspace_unshare(p1); 219 } 220 } 221 cpu_fork(p1, p2, flags); 222 return; 223 } 224 225 if (flags & RFMEM) { 226 p2->p_vmspace = p1->p_vmspace; 227 p1->p_vmspace->vm_refcnt++; 228 } 229 230 while (vm_page_count_severe()) { 231 VM_WAIT; 232 } 233 234 if ((flags & RFMEM) == 0) { 235 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 236 237 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 238 239 if (p1->p_vmspace->vm_shm) 240 shmfork(p1, p2); 241 } 242 243 td2 = lwkt_alloc_thread(NULL); 244 pmap_init_proc(p2, td2); 245 lwkt_setpri(td2, TDPRI_KERN_USER); 246 lwkt_set_comm(td2, "%s", p1->p_comm); 247 248 up = p2->p_addr; 249 250 /* 251 * p_stats currently points at fields in the user struct 252 * but not at &u, instead at p_addr. Copy parts of 253 * p_stats; zero the rest of p_stats (statistics). 254 * 255 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 256 * to share sigacts, so we use the up->u_sigacts. 257 */ 258 p2->p_stats = &up->u_stats; 259 if (p2->p_sigacts == NULL) { 260 if (p2->p_procsig->ps_refcnt != 1) 261 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 262 p2->p_sigacts = &up->u_sigacts; 263 up->u_sigacts = *p1->p_sigacts; 264 } 265 266 bzero(&up->u_stats.pstat_startzero, 267 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 268 (caddr_t) &up->u_stats.pstat_startzero)); 269 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 270 ((caddr_t) &up->u_stats.pstat_endcopy - 271 (caddr_t) &up->u_stats.pstat_startcopy)); 272 273 274 /* 275 * cpu_fork will copy and update the pcb, set up the kernel stack, 276 * and make the child ready to run. 277 */ 278 cpu_fork(p1, p2, flags); 279 } 280 281 /* 282 * Called after process has been wait(2)'ed apon and is being reaped. 283 * The idea is to reclaim resources that we could not reclaim while 284 * the process was still executing. 285 */ 286 void 287 vm_waitproc(struct proc *p) 288 { 289 cpu_proc_wait(p); 290 vmspace_exitfree(p); /* and clean-out the vmspace */ 291 } 292 293 /* 294 * Set default limits for VM system. 295 * Called for proc 0, and then inherited by all others. 296 * 297 * XXX should probably act directly on proc0. 298 */ 299 static void 300 vm_init_limits(udata) 301 void *udata; 302 { 303 struct proc *p = udata; 304 int rss_limit; 305 306 /* 307 * Set up the initial limits on process VM. Set the maximum resident 308 * set size to be half of (reasonably) available memory. Since this 309 * is a soft limit, it comes into effect only when the system is out 310 * of memory - half of main memory helps to favor smaller processes, 311 * and reduces thrashing of the object cache. 312 */ 313 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 314 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 315 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 316 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 317 /* limit the limit to no less than 2MB */ 318 rss_limit = max(vmstats.v_free_count, 512); 319 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 320 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 321 } 322 323 void 324 faultin(p) 325 struct proc *p; 326 { 327 int s; 328 329 if ((p->p_flag & P_INMEM) == 0) { 330 331 ++p->p_lock; 332 333 pmap_swapin_proc(p); 334 335 s = splhigh(); 336 337 if (p->p_stat == SRUN) 338 setrunqueue(p); 339 340 p->p_flag |= P_INMEM; 341 342 /* undo the effect of setting SLOCK above */ 343 --p->p_lock; 344 splx(s); 345 346 } 347 } 348 349 /* 350 * This swapin algorithm attempts to swap-in processes only if there 351 * is enough space for them. Of course, if a process waits for a long 352 * time, it will be swapped in anyway. 353 */ 354 /* ARGSUSED*/ 355 static void 356 scheduler(dummy) 357 void *dummy; 358 { 359 struct proc *p; 360 int pri; 361 struct proc *pp; 362 int ppri; 363 364 loop: 365 if (vm_page_count_min()) { 366 VM_WAIT; 367 goto loop; 368 } 369 370 pp = NULL; 371 ppri = INT_MIN; 372 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 373 if (p->p_stat == SRUN && 374 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) { 375 376 pri = p->p_swtime + p->p_slptime; 377 if ((p->p_flag & P_SWAPINREQ) == 0) { 378 pri -= p->p_nice * 8; 379 } 380 381 /* 382 * if this process is higher priority and there is 383 * enough space, then select this process instead of 384 * the previous selection. 385 */ 386 if (pri > ppri) { 387 pp = p; 388 ppri = pri; 389 } 390 } 391 } 392 393 /* 394 * Nothing to do, back to sleep. 395 */ 396 if ((p = pp) == NULL) { 397 tsleep(&proc0, 0, "sched", 0); 398 goto loop; 399 } 400 p->p_flag &= ~P_SWAPINREQ; 401 402 /* 403 * We would like to bring someone in. (only if there is space). 404 */ 405 faultin(p); 406 p->p_swtime = 0; 407 goto loop; 408 } 409 410 #ifndef NO_SWAPPING 411 412 #define swappable(p) \ 413 (((p)->p_lock == 0) && \ 414 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM) 415 416 417 /* 418 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 419 */ 420 static int swap_idle_threshold1 = 2; 421 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 422 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 423 424 /* 425 * Swap_idle_threshold2 is the time that a process can be idle before 426 * it will be swapped out, if idle swapping is enabled. 427 */ 428 static int swap_idle_threshold2 = 10; 429 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 430 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 431 432 /* 433 * Swapout is driven by the pageout daemon. Very simple, we find eligible 434 * procs and unwire their u-areas. We try to always "swap" at least one 435 * process in case we need the room for a swapin. 436 * If any procs have been sleeping/stopped for at least maxslp seconds, 437 * they are swapped. Else, we swap the longest-sleeping or stopped process, 438 * if any, otherwise the longest-resident process. 439 */ 440 void 441 swapout_procs(action) 442 int action; 443 { 444 struct proc *p; 445 struct proc *outp, *outp2; 446 int outpri, outpri2; 447 int didswap = 0; 448 449 outp = outp2 = NULL; 450 outpri = outpri2 = INT_MIN; 451 retry: 452 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 453 struct vmspace *vm; 454 if (!swappable(p)) 455 continue; 456 457 vm = p->p_vmspace; 458 459 switch (p->p_stat) { 460 default: 461 continue; 462 463 case SSLEEP: 464 case SSTOP: 465 /* 466 * do not swapout a realtime process 467 */ 468 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type)) 469 continue; 470 471 /* 472 * YYY do not swapout a proc waiting on a critical 473 * event. 474 * 475 * Guarentee swap_idle_threshold time in memory 476 */ 477 if (p->p_slptime < swap_idle_threshold1) 478 continue; 479 480 /* 481 * If the system is under memory stress, or if we 482 * are swapping idle processes >= swap_idle_threshold2, 483 * then swap the process out. 484 */ 485 if (((action & VM_SWAP_NORMAL) == 0) && 486 (((action & VM_SWAP_IDLE) == 0) || 487 (p->p_slptime < swap_idle_threshold2))) 488 continue; 489 490 ++vm->vm_refcnt; 491 /* 492 * do not swapout a process that is waiting for VM 493 * data structures there is a possible deadlock. 494 */ 495 if (lockmgr(&vm->vm_map.lock, 496 LK_EXCLUSIVE | LK_NOWAIT, 497 (void *)0, curthread)) { 498 vmspace_free(vm); 499 continue; 500 } 501 vm_map_unlock(&vm->vm_map); 502 /* 503 * If the process has been asleep for awhile and had 504 * most of its pages taken away already, swap it out. 505 */ 506 if ((action & VM_SWAP_NORMAL) || 507 ((action & VM_SWAP_IDLE) && 508 (p->p_slptime > swap_idle_threshold2))) { 509 swapout(p); 510 vmspace_free(vm); 511 didswap++; 512 goto retry; 513 } 514 515 /* 516 * cleanup our reference 517 */ 518 vmspace_free(vm); 519 } 520 } 521 /* 522 * If we swapped something out, and another process needed memory, 523 * then wakeup the sched process. 524 */ 525 if (didswap) 526 wakeup(&proc0); 527 } 528 529 static void 530 swapout(p) 531 struct proc *p; 532 { 533 534 #if defined(SWAP_DEBUG) 535 printf("swapping out %d\n", p->p_pid); 536 #endif 537 ++p->p_stats->p_ru.ru_nswap; 538 /* 539 * remember the process resident count 540 */ 541 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 542 543 (void) splhigh(); 544 p->p_flag &= ~P_INMEM; 545 p->p_flag |= P_SWAPPING; 546 if (p->p_flag & P_ONRUNQ) 547 remrunqueue(p); 548 (void) spl0(); 549 550 pmap_swapout_proc(p); 551 552 p->p_flag &= ~P_SWAPPING; 553 p->p_swtime = 0; 554 } 555 #endif /* !NO_SWAPPING */ 556