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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 39 * 40 * 41 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 42 * All rights reserved. 43 * 44 * Permission to use, copy, modify and distribute this software and 45 * its documentation is hereby granted, provided that both the copyright 46 * notice and this permission notice appear in all copies of the 47 * software, derivative works or modified versions, and any portions 48 * thereof, and that both notices appear in supporting documentation. 49 * 50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 53 * 54 * Carnegie Mellon requests users of this software to return to 55 * 56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57 * School of Computer Science 58 * Carnegie Mellon University 59 * Pittsburgh PA 15213-3890 60 * 61 * any improvements or extensions that they make and grant Carnegie the 62 * rights to redistribute these changes. 63 * 64 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $ 65 */ 66 67 #include "opt_vm.h" 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/proc.h> 72 #include <sys/resourcevar.h> 73 #include <sys/buf.h> 74 #include <sys/shm.h> 75 #include <sys/vmmeter.h> 76 #include <sys/sysctl.h> 77 78 #include <sys/kernel.h> 79 #include <sys/unistd.h> 80 81 #include <machine/limits.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_param.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 #include <vm/vm_page2.h> 95 #include <sys/thread2.h> 96 #include <sys/sysref2.h> 97 98 /* 99 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 100 * 101 * Note: run scheduling should be divorced from the vm system. 102 */ 103 static void scheduler (void *); 104 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 105 106 #ifdef INVARIANTS 107 108 static int swap_debug = 0; 109 SYSCTL_INT(_vm, OID_AUTO, swap_debug, 110 CTLFLAG_RW, &swap_debug, 0, ""); 111 112 #endif 113 114 static int scheduler_notify; 115 116 static void swapout (struct proc *); 117 118 /* 119 * No requirements. 120 */ 121 int 122 kernacc(c_caddr_t addr, int len, int rw) 123 { 124 boolean_t rv; 125 vm_offset_t saddr, eaddr; 126 vm_prot_t prot; 127 128 KASSERT((rw & (~VM_PROT_ALL)) == 0, 129 ("illegal ``rw'' argument to kernacc (%x)", rw)); 130 131 /* 132 * The globaldata space is not part of the kernel_map proper, 133 * check access separately. 134 */ 135 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len))) 136 return (TRUE); 137 138 /* 139 * Nominal kernel memory access - check access via kernel_map. 140 */ 141 if ((vm_offset_t)addr + len > kernel_map.max_offset || 142 (vm_offset_t)addr + len < (vm_offset_t)addr) { 143 return (FALSE); 144 } 145 prot = rw; 146 saddr = trunc_page((vm_offset_t)addr); 147 eaddr = round_page((vm_offset_t)addr + len); 148 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot, FALSE); 149 150 return (rv == TRUE); 151 } 152 153 /* 154 * No requirements. 155 */ 156 int 157 useracc(c_caddr_t addr, int len, int rw) 158 { 159 boolean_t rv; 160 vm_prot_t prot; 161 vm_map_t map; 162 vm_map_entry_t save_hint; 163 vm_offset_t wrap; 164 165 KASSERT((rw & (~VM_PROT_ALL)) == 0, 166 ("illegal ``rw'' argument to useracc (%x)", rw)); 167 prot = rw; 168 /* 169 * XXX - check separately to disallow access to user area and user 170 * page tables - they are in the map. 171 */ 172 wrap = (vm_offset_t)addr + len; 173 if (wrap > VM_MAX_USER_ADDRESS || wrap < (vm_offset_t)addr) { 174 return (FALSE); 175 } 176 map = &curproc->p_vmspace->vm_map; 177 vm_map_lock_read(map); 178 /* 179 * We save the map hint, and restore it. Useracc appears to distort 180 * the map hint unnecessarily. 181 */ 182 save_hint = map->hint; 183 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 184 round_page(wrap), prot, TRUE); 185 map->hint = save_hint; 186 vm_map_unlock_read(map); 187 188 return (rv == TRUE); 189 } 190 191 /* 192 * No requirements. 193 */ 194 void 195 vslock(caddr_t addr, u_int len) 196 { 197 if (len) { 198 vm_map_wire(&curproc->p_vmspace->vm_map, 199 trunc_page((vm_offset_t)addr), 200 round_page((vm_offset_t)addr + len), 0); 201 } 202 } 203 204 /* 205 * No requirements. 206 */ 207 void 208 vsunlock(caddr_t addr, u_int len) 209 { 210 if (len) { 211 vm_map_wire(&curproc->p_vmspace->vm_map, 212 trunc_page((vm_offset_t)addr), 213 round_page((vm_offset_t)addr + len), 214 KM_PAGEABLE); 215 } 216 } 217 218 /* 219 * Implement fork's actions on an address space. 220 * Here we arrange for the address space to be copied or referenced, 221 * allocate a user struct (pcb and kernel stack), then call the 222 * machine-dependent layer to fill those in and make the new process 223 * ready to run. The new process is set up so that it returns directly 224 * to user mode to avoid stack copying and relocation problems. 225 * 226 * No requirements. 227 */ 228 void 229 vm_fork(struct proc *p1, struct proc *p2, int flags) 230 { 231 if ((flags & RFPROC) == 0) { 232 /* 233 * Divorce the memory, if it is shared, essentially 234 * this changes shared memory amongst threads, into 235 * COW locally. 236 */ 237 if ((flags & RFMEM) == 0) { 238 if (p1->p_vmspace->vm_sysref.refcnt > 1) { 239 vmspace_unshare(p1); 240 } 241 } 242 cpu_fork(ONLY_LWP_IN_PROC(p1), NULL, flags); 243 return; 244 } 245 246 if (flags & RFMEM) { 247 vmspace_ref(p1->p_vmspace); 248 p2->p_vmspace = p1->p_vmspace; 249 } 250 251 while (vm_page_count_severe()) { 252 vm_wait(0); 253 } 254 255 if ((flags & RFMEM) == 0) { 256 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 257 258 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 259 260 if (p1->p_vmspace->vm_shm) 261 shmfork(p1, p2); 262 } 263 264 pmap_init_proc(p2); 265 } 266 267 /* 268 * Set default limits for VM system. Call during proc0's initialization. 269 * 270 * Called from the low level boot code only. 271 */ 272 void 273 vm_init_limits(struct proc *p) 274 { 275 int rss_limit; 276 277 /* 278 * Set up the initial limits on process VM. Set the maximum resident 279 * set size to be half of (reasonably) available memory. Since this 280 * is a soft limit, it comes into effect only when the system is out 281 * of memory - half of main memory helps to favor smaller processes, 282 * and reduces thrashing of the object cache. 283 */ 284 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 285 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 286 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 287 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 288 /* limit the limit to no less than 2MB */ 289 rss_limit = max(vmstats.v_free_count, 512); 290 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 291 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 292 } 293 294 /* 295 * Faultin the specified process. Note that the process can be in any 296 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is 297 * sleeping. 298 * 299 * No requirements. 300 */ 301 void 302 faultin(struct proc *p) 303 { 304 if (p->p_flags & P_SWAPPEDOUT) { 305 /* 306 * The process is waiting in the kernel to return to user 307 * mode but cannot until P_SWAPPEDOUT gets cleared. 308 */ 309 lwkt_gettoken(&p->p_token); 310 p->p_flags &= ~(P_SWAPPEDOUT | P_SWAPWAIT); 311 #ifdef INVARIANTS 312 if (swap_debug) 313 kprintf("swapping in %d (%s)\n", p->p_pid, p->p_comm); 314 #endif 315 wakeup(p); 316 lwkt_reltoken(&p->p_token); 317 } 318 } 319 320 /* 321 * Kernel initialization eventually falls through to this function, 322 * which is process 0. 323 * 324 * This swapin algorithm attempts to swap-in processes only if there 325 * is enough space for them. Of course, if a process waits for a long 326 * time, it will be swapped in anyway. 327 */ 328 struct scheduler_info { 329 struct proc *pp; 330 int ppri; 331 }; 332 333 static int scheduler_callback(struct proc *p, void *data); 334 335 static void 336 scheduler(void *dummy) 337 { 338 struct scheduler_info info; 339 struct proc *p; 340 341 KKASSERT(!IN_CRITICAL_SECT(curthread)); 342 loop: 343 scheduler_notify = 0; 344 /* 345 * Don't try to swap anything in if we are low on memory. 346 */ 347 if (vm_page_count_severe()) { 348 vm_wait(0); 349 goto loop; 350 } 351 352 /* 353 * Look for a good candidate to wake up 354 */ 355 info.pp = NULL; 356 info.ppri = INT_MIN; 357 allproc_scan(scheduler_callback, &info); 358 359 /* 360 * Nothing to do, back to sleep for at least 1/10 of a second. If 361 * we are woken up, immediately process the next request. If 362 * multiple requests have built up the first is processed 363 * immediately and the rest are staggered. 364 */ 365 if ((p = info.pp) == NULL) { 366 tsleep(&proc0, 0, "nowork", hz / 10); 367 if (scheduler_notify == 0) 368 tsleep(&scheduler_notify, 0, "nowork", 0); 369 goto loop; 370 } 371 372 /* 373 * Fault the selected process in, then wait for a short period of 374 * time and loop up. 375 * 376 * XXX we need a heuristic to get a measure of system stress and 377 * then adjust our stagger wakeup delay accordingly. 378 */ 379 lwkt_gettoken(&proc_token); 380 faultin(p); 381 p->p_swtime = 0; 382 PRELE(p); 383 lwkt_reltoken(&proc_token); 384 tsleep(&proc0, 0, "swapin", hz / 10); 385 goto loop; 386 } 387 388 /* 389 * The caller must hold proc_token. 390 */ 391 static int 392 scheduler_callback(struct proc *p, void *data) 393 { 394 struct scheduler_info *info = data; 395 struct lwp *lp; 396 segsz_t pgs; 397 int pri; 398 399 if (p->p_flags & P_SWAPWAIT) { 400 pri = 0; 401 FOREACH_LWP_IN_PROC(lp, p) { 402 /* XXX lwp might need a different metric */ 403 pri += lp->lwp_slptime; 404 } 405 pri += p->p_swtime - p->p_nice * 8; 406 407 /* 408 * The more pages paged out while we were swapped, 409 * the more work we have to do to get up and running 410 * again and the lower our wakeup priority. 411 * 412 * Each second of sleep time is worth ~1MB 413 */ 414 lwkt_gettoken(&p->p_vmspace->vm_map.token); 415 pgs = vmspace_resident_count(p->p_vmspace); 416 if (pgs < p->p_vmspace->vm_swrss) { 417 pri -= (p->p_vmspace->vm_swrss - pgs) / 418 (1024 * 1024 / PAGE_SIZE); 419 } 420 lwkt_reltoken(&p->p_vmspace->vm_map.token); 421 422 /* 423 * If this process is higher priority and there is 424 * enough space, then select this process instead of 425 * the previous selection. 426 */ 427 if (pri > info->ppri) { 428 if (info->pp) 429 PRELE(info->pp); 430 PHOLD(p); 431 info->pp = p; 432 info->ppri = pri; 433 } 434 } 435 return(0); 436 } 437 438 /* 439 * SMP races ok. 440 * No requirements. 441 */ 442 void 443 swapin_request(void) 444 { 445 if (scheduler_notify == 0) { 446 scheduler_notify = 1; 447 wakeup(&scheduler_notify); 448 } 449 } 450 451 #ifndef NO_SWAPPING 452 453 #define swappable(p) \ 454 (((p)->p_lock == 0) && \ 455 ((p)->p_flags & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0) 456 457 458 /* 459 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 460 */ 461 static int swap_idle_threshold1 = 15; 462 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 463 CTLFLAG_RW, &swap_idle_threshold1, 0, "Guaranteed process resident time (sec)"); 464 465 /* 466 * Swap_idle_threshold2 is the time that a process can be idle before 467 * it will be swapped out, if idle swapping is enabled. Default is 468 * one minute. 469 */ 470 static int swap_idle_threshold2 = 60; 471 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 472 CTLFLAG_RW, &swap_idle_threshold2, 0, "Time (sec) a process can idle before being swapped"); 473 474 /* 475 * Swapout is driven by the pageout daemon. Very simple, we find eligible 476 * procs and mark them as being swapped out. This will cause the kernel 477 * to prefer to pageout those proc's pages first and the procs in question 478 * will not return to user mode until the swapper tells them they can. 479 * 480 * If any procs have been sleeping/stopped for at least maxslp seconds, 481 * they are swapped. Else, we swap the longest-sleeping or stopped process, 482 * if any, otherwise the longest-resident process. 483 */ 484 485 static int swapout_procs_callback(struct proc *p, void *data); 486 487 /* 488 * No requirements. 489 */ 490 void 491 swapout_procs(int action) 492 { 493 allproc_scan(swapout_procs_callback, &action); 494 } 495 496 /* 497 * The caller must hold proc_token 498 */ 499 static int 500 swapout_procs_callback(struct proc *p, void *data) 501 { 502 struct lwp *lp; 503 int action = *(int *)data; 504 int minslp = -1; 505 506 if (!swappable(p)) 507 return(0); 508 509 lwkt_gettoken(&p->p_token); 510 511 /* 512 * We only consider active processes. 513 */ 514 if (p->p_stat != SACTIVE && p->p_stat != SSTOP) { 515 lwkt_reltoken(&p->p_token); 516 return(0); 517 } 518 519 FOREACH_LWP_IN_PROC(lp, p) { 520 /* 521 * do not swap out a realtime process 522 */ 523 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type)) { 524 lwkt_reltoken(&p->p_token); 525 return(0); 526 } 527 528 /* 529 * Guarentee swap_idle_threshold time in memory 530 */ 531 if (lp->lwp_slptime < swap_idle_threshold1) { 532 lwkt_reltoken(&p->p_token); 533 return(0); 534 } 535 536 /* 537 * If the system is under memory stress, or if we 538 * are swapping idle processes >= swap_idle_threshold2, 539 * then swap the process out. 540 */ 541 if (((action & VM_SWAP_NORMAL) == 0) && 542 (((action & VM_SWAP_IDLE) == 0) || 543 (lp->lwp_slptime < swap_idle_threshold2))) { 544 lwkt_reltoken(&p->p_token); 545 return(0); 546 } 547 548 if (minslp == -1 || lp->lwp_slptime < minslp) 549 minslp = lp->lwp_slptime; 550 } 551 552 /* 553 * If the process has been asleep for awhile, swap 554 * it out. 555 */ 556 if ((action & VM_SWAP_NORMAL) || 557 ((action & VM_SWAP_IDLE) && 558 (minslp > swap_idle_threshold2))) { 559 swapout(p); 560 } 561 562 /* 563 * cleanup our reference 564 */ 565 lwkt_reltoken(&p->p_token); 566 567 return(0); 568 } 569 570 /* 571 * The caller must hold proc_token and p->p_token 572 */ 573 static void 574 swapout(struct proc *p) 575 { 576 #ifdef INVARIANTS 577 if (swap_debug) 578 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm); 579 #endif 580 ++p->p_ru.ru_nswap; 581 582 /* 583 * remember the process resident count 584 */ 585 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 586 p->p_flags |= P_SWAPPEDOUT; 587 p->p_swtime = 0; 588 } 589 590 #endif /* !NO_SWAPPING */ 591 592