1 /* 2 * (MPSAFE) 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 * 60 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $ 61 */ 62 63 #include "opt_vm.h" 64 65 #include <sys/param.h> 66 #include <sys/systm.h> 67 #include <sys/proc.h> 68 #include <sys/resourcevar.h> 69 #include <sys/buf.h> 70 #include <sys/shm.h> 71 #include <sys/vmmeter.h> 72 #include <sys/sysctl.h> 73 74 #include <sys/kernel.h> 75 #include <sys/unistd.h> 76 77 #include <machine/limits.h> 78 79 #include <vm/vm.h> 80 #include <vm/vm_param.h> 81 #include <sys/lock.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_page.h> 85 #include <vm/vm_pageout.h> 86 #include <vm/vm_kern.h> 87 #include <vm/vm_extern.h> 88 89 #include <sys/user.h> 90 #include <vm/vm_page2.h> 91 #include <sys/thread2.h> 92 #include <sys/sysref2.h> 93 94 /* 95 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 96 * 97 * Note: run scheduling should be divorced from the vm system. 98 */ 99 static void scheduler (void *); 100 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 101 102 #ifdef INVARIANTS 103 104 static int swap_debug = 0; 105 SYSCTL_INT(_vm, OID_AUTO, swap_debug, 106 CTLFLAG_RW, &swap_debug, 0, ""); 107 108 #endif 109 110 static int scheduler_notify; 111 112 static void swapout (struct proc *); 113 114 /* 115 * No requirements. 116 */ 117 int 118 kernacc(c_caddr_t addr, int len, int 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)", rw)); 126 127 /* 128 * The globaldata space is not part of the kernel_map proper, 129 * check access separately. 130 */ 131 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len))) 132 return (TRUE); 133 134 /* 135 * Nominal kernel memory access - check access via kernel_map. 136 */ 137 if ((vm_offset_t)addr + len > kernel_map.max_offset || 138 (vm_offset_t)addr + len < (vm_offset_t)addr) { 139 return (FALSE); 140 } 141 prot = rw; 142 saddr = trunc_page((vm_offset_t)addr); 143 eaddr = round_page((vm_offset_t)addr + len); 144 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot, FALSE); 145 146 return (rv == TRUE); 147 } 148 149 /* 150 * No requirements. 151 */ 152 int 153 useracc(c_caddr_t addr, int len, int rw) 154 { 155 boolean_t rv; 156 vm_prot_t prot; 157 vm_map_t map; 158 vm_map_entry_t save_hint; 159 vm_offset_t wrap; 160 161 KASSERT((rw & (~VM_PROT_ALL)) == 0, 162 ("illegal ``rw'' argument to useracc (%x)", rw)); 163 prot = rw; 164 /* 165 * XXX - check separately to disallow access to user area and user 166 * page tables - they are in the map. 167 */ 168 wrap = (vm_offset_t)addr + len; 169 if (wrap > VM_MAX_USER_ADDRESS || wrap < (vm_offset_t)addr) { 170 return (FALSE); 171 } 172 map = &curproc->p_vmspace->vm_map; 173 vm_map_lock_read(map); 174 /* 175 * We save the map hint, and restore it. Useracc appears to distort 176 * the map hint unnecessarily. 177 */ 178 save_hint = map->hint; 179 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 180 round_page(wrap), prot, TRUE); 181 map->hint = save_hint; 182 vm_map_unlock_read(map); 183 184 return (rv == TRUE); 185 } 186 187 /* 188 * No requirements. 189 */ 190 void 191 vslock(caddr_t addr, u_int len) 192 { 193 if (len) { 194 vm_map_wire(&curproc->p_vmspace->vm_map, 195 trunc_page((vm_offset_t)addr), 196 round_page((vm_offset_t)addr + len), 0); 197 } 198 } 199 200 /* 201 * No requirements. 202 */ 203 void 204 vsunlock(caddr_t addr, u_int len) 205 { 206 if (len) { 207 vm_map_wire(&curproc->p_vmspace->vm_map, 208 trunc_page((vm_offset_t)addr), 209 round_page((vm_offset_t)addr + len), 210 KM_PAGEABLE); 211 } 212 } 213 214 /* 215 * Implement fork's actions on an address space. 216 * Here we arrange for the address space to be copied or referenced, 217 * allocate a user struct (pcb and kernel stack), then call the 218 * machine-dependent layer to fill those in and make the new process 219 * ready to run. The new process is set up so that it returns directly 220 * to user mode to avoid stack copying and relocation problems. 221 * 222 * No requirements. 223 */ 224 void 225 vm_fork(struct proc *p1, struct proc *p2, int flags) 226 { 227 if ((flags & RFPROC) == 0) { 228 /* 229 * Divorce the memory, if it is shared, essentially 230 * this changes shared memory amongst threads, into 231 * COW locally. 232 */ 233 if ((flags & RFMEM) == 0) { 234 if (p1->p_vmspace->vm_sysref.refcnt > 1) { 235 vmspace_unshare(p1); 236 } 237 } 238 cpu_fork(ONLY_LWP_IN_PROC(p1), NULL, flags); 239 return; 240 } 241 242 if (flags & RFMEM) { 243 vmspace_ref(p1->p_vmspace); 244 p2->p_vmspace = p1->p_vmspace; 245 } 246 247 while (vm_page_count_severe()) { 248 vm_wait(0); 249 } 250 251 if ((flags & RFMEM) == 0) { 252 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 253 254 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 255 256 if (p1->p_vmspace->vm_shm) 257 shmfork(p1, p2); 258 } 259 260 pmap_init_proc(p2); 261 } 262 263 /* 264 * Set default limits for VM system. Call during proc0's initialization. 265 * 266 * Called from the low level boot code only. 267 */ 268 void 269 vm_init_limits(struct proc *p) 270 { 271 int rss_limit; 272 273 /* 274 * Set up the initial limits on process VM. Set the maximum resident 275 * set size to be half of (reasonably) available memory. Since this 276 * is a soft limit, it comes into effect only when the system is out 277 * of memory - half of main memory helps to favor smaller processes, 278 * and reduces thrashing of the object cache. 279 */ 280 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 281 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 282 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 283 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 284 /* limit the limit to no less than 2MB */ 285 rss_limit = max(vmstats.v_free_count, 512); 286 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 287 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 288 } 289 290 /* 291 * Faultin the specified process. Note that the process can be in any 292 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is 293 * sleeping. 294 * 295 * No requirements. 296 */ 297 void 298 faultin(struct proc *p) 299 { 300 if (p->p_flags & P_SWAPPEDOUT) { 301 /* 302 * The process is waiting in the kernel to return to user 303 * mode but cannot until P_SWAPPEDOUT gets cleared. 304 */ 305 lwkt_gettoken(&p->p_token); 306 p->p_flags &= ~(P_SWAPPEDOUT | P_SWAPWAIT); 307 #ifdef INVARIANTS 308 if (swap_debug) 309 kprintf("swapping in %d (%s)\n", p->p_pid, p->p_comm); 310 #endif 311 wakeup(p); 312 lwkt_reltoken(&p->p_token); 313 } 314 } 315 316 /* 317 * Kernel initialization eventually falls through to this function, 318 * which is process 0. 319 * 320 * This swapin algorithm attempts to swap-in processes only if there 321 * is enough space for them. Of course, if a process waits for a long 322 * time, it will be swapped in anyway. 323 */ 324 struct scheduler_info { 325 struct proc *pp; 326 int ppri; 327 }; 328 329 static int scheduler_callback(struct proc *p, void *data); 330 331 static void 332 scheduler(void *dummy) 333 { 334 struct scheduler_info info; 335 struct proc *p; 336 337 KKASSERT(!IN_CRITICAL_SECT(curthread)); 338 loop: 339 scheduler_notify = 0; 340 /* 341 * Don't try to swap anything in if we are low on memory. 342 */ 343 if (vm_page_count_severe()) { 344 vm_wait(0); 345 goto loop; 346 } 347 348 /* 349 * Look for a good candidate to wake up 350 */ 351 info.pp = NULL; 352 info.ppri = INT_MIN; 353 allproc_scan(scheduler_callback, &info); 354 355 /* 356 * Nothing to do, back to sleep for at least 1/10 of a second. If 357 * we are woken up, immediately process the next request. If 358 * multiple requests have built up the first is processed 359 * immediately and the rest are staggered. 360 */ 361 if ((p = info.pp) == NULL) { 362 tsleep(&proc0, 0, "nowork", hz / 10); 363 if (scheduler_notify == 0) 364 tsleep(&scheduler_notify, 0, "nowork", 0); 365 goto loop; 366 } 367 368 /* 369 * Fault the selected process in, then wait for a short period of 370 * time and loop up. 371 * 372 * XXX we need a heuristic to get a measure of system stress and 373 * then adjust our stagger wakeup delay accordingly. 374 */ 375 lwkt_gettoken(&proc_token); 376 faultin(p); 377 p->p_swtime = 0; 378 PRELE(p); 379 lwkt_reltoken(&proc_token); 380 tsleep(&proc0, 0, "swapin", hz / 10); 381 goto loop; 382 } 383 384 /* 385 * The caller must hold proc_token. 386 */ 387 static int 388 scheduler_callback(struct proc *p, void *data) 389 { 390 struct scheduler_info *info = data; 391 struct lwp *lp; 392 segsz_t pgs; 393 int pri; 394 395 if (p->p_flags & P_SWAPWAIT) { 396 pri = 0; 397 FOREACH_LWP_IN_PROC(lp, p) { 398 /* XXX lwp might need a different metric */ 399 pri += lp->lwp_slptime; 400 } 401 pri += p->p_swtime - p->p_nice * 8; 402 403 /* 404 * The more pages paged out while we were swapped, 405 * the more work we have to do to get up and running 406 * again and the lower our wakeup priority. 407 * 408 * Each second of sleep time is worth ~1MB 409 */ 410 lwkt_gettoken(&p->p_vmspace->vm_map.token); 411 pgs = vmspace_resident_count(p->p_vmspace); 412 if (pgs < p->p_vmspace->vm_swrss) { 413 pri -= (p->p_vmspace->vm_swrss - pgs) / 414 (1024 * 1024 / PAGE_SIZE); 415 } 416 lwkt_reltoken(&p->p_vmspace->vm_map.token); 417 418 /* 419 * If this process is higher priority and there is 420 * enough space, then select this process instead of 421 * the previous selection. 422 */ 423 if (pri > info->ppri) { 424 if (info->pp) 425 PRELE(info->pp); 426 PHOLD(p); 427 info->pp = p; 428 info->ppri = pri; 429 } 430 } 431 return(0); 432 } 433 434 /* 435 * SMP races ok. 436 * No requirements. 437 */ 438 void 439 swapin_request(void) 440 { 441 if (scheduler_notify == 0) { 442 scheduler_notify = 1; 443 wakeup(&scheduler_notify); 444 } 445 } 446 447 #ifndef NO_SWAPPING 448 449 #define swappable(p) \ 450 (((p)->p_lock == 0) && \ 451 ((p)->p_flags & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0) 452 453 454 /* 455 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 456 */ 457 static int swap_idle_threshold1 = 15; 458 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 459 CTLFLAG_RW, &swap_idle_threshold1, 0, "Guaranteed process resident time (sec)"); 460 461 /* 462 * Swap_idle_threshold2 is the time that a process can be idle before 463 * it will be swapped out, if idle swapping is enabled. Default is 464 * one minute. 465 */ 466 static int swap_idle_threshold2 = 60; 467 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 468 CTLFLAG_RW, &swap_idle_threshold2, 0, "Time (sec) a process can idle before being swapped"); 469 470 /* 471 * Swapout is driven by the pageout daemon. Very simple, we find eligible 472 * procs and mark them as being swapped out. This will cause the kernel 473 * to prefer to pageout those proc's pages first and the procs in question 474 * will not return to user mode until the swapper tells them they can. 475 * 476 * If any procs have been sleeping/stopped for at least maxslp seconds, 477 * they are swapped. Else, we swap the longest-sleeping or stopped process, 478 * if any, otherwise the longest-resident process. 479 */ 480 481 static int swapout_procs_callback(struct proc *p, void *data); 482 483 /* 484 * No requirements. 485 */ 486 void 487 swapout_procs(int action) 488 { 489 allproc_scan(swapout_procs_callback, &action); 490 } 491 492 /* 493 * The caller must hold proc_token 494 */ 495 static int 496 swapout_procs_callback(struct proc *p, void *data) 497 { 498 struct lwp *lp; 499 int action = *(int *)data; 500 int minslp = -1; 501 502 if (!swappable(p)) 503 return(0); 504 505 lwkt_gettoken(&p->p_token); 506 507 /* 508 * We only consider active processes. 509 */ 510 if (p->p_stat != SACTIVE && p->p_stat != SSTOP) { 511 lwkt_reltoken(&p->p_token); 512 return(0); 513 } 514 515 FOREACH_LWP_IN_PROC(lp, p) { 516 /* 517 * do not swap out a realtime process 518 */ 519 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type)) { 520 lwkt_reltoken(&p->p_token); 521 return(0); 522 } 523 524 /* 525 * Guarentee swap_idle_threshold time in memory 526 */ 527 if (lp->lwp_slptime < swap_idle_threshold1) { 528 lwkt_reltoken(&p->p_token); 529 return(0); 530 } 531 532 /* 533 * If the system is under memory stress, or if we 534 * are swapping idle processes >= swap_idle_threshold2, 535 * then swap the process out. 536 */ 537 if (((action & VM_SWAP_NORMAL) == 0) && 538 (((action & VM_SWAP_IDLE) == 0) || 539 (lp->lwp_slptime < swap_idle_threshold2))) { 540 lwkt_reltoken(&p->p_token); 541 return(0); 542 } 543 544 if (minslp == -1 || lp->lwp_slptime < minslp) 545 minslp = lp->lwp_slptime; 546 } 547 548 /* 549 * If the process has been asleep for awhile, swap 550 * it out. 551 */ 552 if ((action & VM_SWAP_NORMAL) || 553 ((action & VM_SWAP_IDLE) && 554 (minslp > swap_idle_threshold2))) { 555 swapout(p); 556 } 557 558 /* 559 * cleanup our reference 560 */ 561 lwkt_reltoken(&p->p_token); 562 563 return(0); 564 } 565 566 /* 567 * The caller must hold proc_token and p->p_token 568 */ 569 static void 570 swapout(struct proc *p) 571 { 572 #ifdef INVARIANTS 573 if (swap_debug) 574 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm); 575 #endif 576 ++p->p_ru.ru_nswap; 577 578 /* 579 * remember the process resident count 580 */ 581 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 582 p->p_flags |= P_SWAPPEDOUT; 583 p->p_swtime = 0; 584 } 585 586 #endif /* !NO_SWAPPING */ 587 588