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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 33 * 34 * 35 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36 * All rights reserved. 37 * 38 * Permission to use, copy, modify and distribute this software and 39 * its documentation is hereby granted, provided that both the copyright 40 * notice and this permission notice appear in all copies of the 41 * software, derivative works or modified versions, and any portions 42 * thereof, and that both notices appear in supporting documentation. 43 * 44 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 45 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 46 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 47 * 48 * Carnegie Mellon requests users of this software to return to 49 * 50 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 51 * School of Computer Science 52 * Carnegie Mellon University 53 * Pittsburgh PA 15213-3890 54 * 55 * any improvements or extensions that they make and grant Carnegie the 56 * rights to redistribute these changes. 57 * 58 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $ 59 */ 60 61 #include "opt_vm.h" 62 63 #include <sys/param.h> 64 #include <sys/systm.h> 65 #include <sys/proc.h> 66 #include <sys/resourcevar.h> 67 #include <sys/buf.h> 68 #include <sys/shm.h> 69 #include <sys/vmmeter.h> 70 #include <sys/sysctl.h> 71 72 #include <sys/kernel.h> 73 #include <sys/unistd.h> 74 75 #include <machine/limits.h> 76 #include <machine/vmm.h> 77 78 #include <vm/vm.h> 79 #include <vm/vm_param.h> 80 #include <sys/lock.h> 81 #include <vm/pmap.h> 82 #include <vm/vm_map.h> 83 #include <vm/vm_page.h> 84 #include <vm/vm_pageout.h> 85 #include <vm/vm_kern.h> 86 #include <vm/vm_extern.h> 87 88 #include <sys/user.h> 89 #include <vm/vm_page2.h> 90 #include <sys/thread2.h> 91 92 /* 93 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 94 * 95 * Note: run scheduling should be divorced from the vm system. 96 */ 97 static void scheduler (void *); 98 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL); 99 100 #ifdef INVARIANTS 101 102 static int swap_debug = 0; 103 SYSCTL_INT(_vm, OID_AUTO, swap_debug, 104 CTLFLAG_RW, &swap_debug, 0, ""); 105 106 #endif 107 108 static int scheduler_notify; 109 110 static void swapout (struct proc *); 111 112 /* 113 * No requirements. 114 */ 115 int 116 kernacc(c_caddr_t addr, int len, int rw) 117 { 118 boolean_t rv; 119 vm_offset_t saddr, eaddr; 120 vm_prot_t prot; 121 122 KASSERT((rw & (~VM_PROT_ALL)) == 0, 123 ("illegal ``rw'' argument to kernacc (%x)", rw)); 124 125 /* 126 * The globaldata space is not part of the kernel_map proper, 127 * check access separately. 128 */ 129 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len))) 130 return (TRUE); 131 132 /* 133 * Nominal kernel memory access - check access via kernel_map. 134 */ 135 if ((vm_offset_t)addr + len > kernel_map.max_offset || 136 (vm_offset_t)addr + len < (vm_offset_t)addr) { 137 return (FALSE); 138 } 139 prot = rw; 140 saddr = trunc_page((vm_offset_t)addr); 141 eaddr = round_page((vm_offset_t)addr + len); 142 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot, FALSE); 143 144 return (rv == TRUE); 145 } 146 147 /* 148 * No requirements. 149 */ 150 int 151 useracc(c_caddr_t addr, int len, int rw) 152 { 153 boolean_t rv; 154 vm_prot_t prot; 155 vm_map_t map; 156 vm_offset_t wrap; 157 vm_offset_t gpa; 158 159 KASSERT((rw & (~VM_PROT_ALL)) == 0, 160 ("illegal ``rw'' argument to useracc (%x)", rw)); 161 prot = rw; 162 163 if (curthread->td_vmm) { 164 if (vmm_vm_get_gpa(curproc, (register_t *)&gpa, (register_t) addr)) 165 panic("%s: could not get GPA\n", __func__); 166 addr = (c_caddr_t) gpa; 167 } 168 169 /* 170 * XXX - check separately to disallow access to user area and user 171 * page tables - they are in the map. 172 */ 173 wrap = (vm_offset_t)addr + len; 174 if (wrap > VM_MAX_USER_ADDRESS || wrap < (vm_offset_t)addr) { 175 return (FALSE); 176 } 177 map = &curproc->p_vmspace->vm_map; 178 vm_map_lock_read(map); 179 180 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 181 round_page(wrap), prot, TRUE); 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 (vmspace_getrefs(p1->p_vmspace) > 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 * XXX we should make the schedule thread pcpu and then use a 352 * segmented allproc scan. 353 */ 354 info.pp = NULL; 355 info.ppri = INT_MIN; 356 allproc_scan(scheduler_callback, &info, 0); 357 358 /* 359 * Nothing to do, back to sleep for at least 1/10 of a second. If 360 * we are woken up, immediately process the next request. If 361 * multiple requests have built up the first is processed 362 * immediately and the rest are staggered. 363 */ 364 if ((p = info.pp) == NULL) { 365 tsleep(&proc0, 0, "nowork", hz / 10); 366 if (scheduler_notify == 0) 367 tsleep(&scheduler_notify, 0, "nowork", 0); 368 goto loop; 369 } 370 371 /* 372 * Fault the selected process in, then wait for a short period of 373 * time and loop up. 374 * 375 * XXX we need a heuristic to get a measure of system stress and 376 * then adjust our stagger wakeup delay accordingly. 377 */ 378 lwkt_gettoken(&p->p_token); 379 faultin(p); 380 p->p_swtime = 0; 381 lwkt_reltoken(&p->p_token); 382 PRELE(p); 383 tsleep(&proc0, 0, "swapin", hz / 10); 384 goto loop; 385 } 386 387 /* 388 * Process only has its hold count bumped, we need the token 389 * to safely scan the LWPs 390 */ 391 static int 392 scheduler_callback(struct proc *p, void *data) 393 { 394 struct scheduler_info *info = data; 395 struct vmspace *vm; 396 struct lwp *lp; 397 segsz_t pgs; 398 int pri; 399 400 /* 401 * We only care about processes in swap-wait. Interlock test with 402 * token if the flag is found set. 403 */ 404 if ((p->p_flags & P_SWAPWAIT) == 0) 405 return 0; 406 lwkt_gettoken_shared(&p->p_token); 407 if ((p->p_flags & P_SWAPWAIT) == 0) { 408 lwkt_reltoken(&p->p_token); 409 return 0; 410 } 411 412 /* 413 * Calculate priority for swap-in 414 */ 415 pri = 0; 416 FOREACH_LWP_IN_PROC(lp, p) { 417 /* XXX lwp might need a different metric */ 418 pri += lp->lwp_slptime; 419 } 420 pri += p->p_swtime - p->p_nice * 8; 421 422 /* 423 * The more pages paged out while we were swapped, 424 * the more work we have to do to get up and running 425 * again and the lower our wakeup priority. 426 * 427 * Each second of sleep time is worth ~1MB 428 */ 429 if ((vm = p->p_vmspace) != NULL) { 430 vmspace_hold(vm); 431 pgs = vmspace_resident_count(vm); 432 if (pgs < vm->vm_swrss) { 433 pri -= (vm->vm_swrss - pgs) / 434 (1024 * 1024 / PAGE_SIZE); 435 } 436 vmspace_drop(vm); 437 } 438 lwkt_reltoken(&p->p_token); 439 440 /* 441 * If this process is higher priority and there is 442 * enough space, then select this process instead of 443 * the previous selection. 444 */ 445 if (pri > info->ppri) { 446 if (info->pp) 447 PRELE(info->pp); 448 PHOLD(p); 449 info->pp = p; 450 info->ppri = pri; 451 } 452 return(0); 453 } 454 455 /* 456 * SMP races ok. 457 * No requirements. 458 */ 459 void 460 swapin_request(void) 461 { 462 if (scheduler_notify == 0) { 463 scheduler_notify = 1; 464 wakeup(&scheduler_notify); 465 } 466 } 467 468 #ifndef NO_SWAPPING 469 470 #define swappable(p) \ 471 (((p)->p_lock == 0) && \ 472 ((p)->p_flags & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0) 473 474 475 /* 476 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 477 */ 478 static int swap_idle_threshold1 = 15; 479 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 480 CTLFLAG_RW, &swap_idle_threshold1, 0, "Guaranteed process resident time (sec)"); 481 482 /* 483 * Swap_idle_threshold2 is the time that a process can be idle before 484 * it will be swapped out, if idle swapping is enabled. Default is 485 * one minute. 486 */ 487 static int swap_idle_threshold2 = 60; 488 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 489 CTLFLAG_RW, &swap_idle_threshold2, 0, "Time (sec) a process can idle before being swapped"); 490 491 /* 492 * Swapout is driven by the pageout daemon. Very simple, we find eligible 493 * procs and mark them as being swapped out. This will cause the kernel 494 * to prefer to pageout those proc's pages first and the procs in question 495 * will not return to user mode until the swapper tells them they can. 496 * 497 * If any procs have been sleeping/stopped for at least maxslp seconds, 498 * they are swapped. Else, we swap the longest-sleeping or stopped process, 499 * if any, otherwise the longest-resident process. 500 */ 501 502 static int swapout_procs_callback(struct proc *p, void *data); 503 504 /* 505 * No requirements. 506 */ 507 void 508 swapout_procs(int action) 509 { 510 allproc_scan(swapout_procs_callback, &action, 0); 511 } 512 513 static int 514 swapout_procs_callback(struct proc *p, void *data) 515 { 516 struct lwp *lp; 517 int action = *(int *)data; 518 int minslp = -1; 519 520 if (!swappable(p)) 521 return(0); 522 523 lwkt_gettoken(&p->p_token); 524 525 /* 526 * We only consider active processes. 527 */ 528 if (p->p_stat != SACTIVE && p->p_stat != SSTOP) { 529 lwkt_reltoken(&p->p_token); 530 return(0); 531 } 532 533 FOREACH_LWP_IN_PROC(lp, p) { 534 /* 535 * do not swap out a realtime process 536 */ 537 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type)) { 538 lwkt_reltoken(&p->p_token); 539 return(0); 540 } 541 542 /* 543 * Guarentee swap_idle_threshold time in memory 544 */ 545 if (lp->lwp_slptime < swap_idle_threshold1) { 546 lwkt_reltoken(&p->p_token); 547 return(0); 548 } 549 550 /* 551 * If the system is under memory stress, or if we 552 * are swapping idle processes >= swap_idle_threshold2, 553 * then swap the process out. 554 */ 555 if (((action & VM_SWAP_NORMAL) == 0) && 556 (((action & VM_SWAP_IDLE) == 0) || 557 (lp->lwp_slptime < swap_idle_threshold2))) { 558 lwkt_reltoken(&p->p_token); 559 return(0); 560 } 561 562 if (minslp == -1 || lp->lwp_slptime < minslp) 563 minslp = lp->lwp_slptime; 564 } 565 566 /* 567 * If the process has been asleep for awhile, swap 568 * it out. 569 */ 570 if ((action & VM_SWAP_NORMAL) || 571 ((action & VM_SWAP_IDLE) && 572 (minslp > swap_idle_threshold2))) { 573 swapout(p); 574 } 575 576 /* 577 * cleanup our reference 578 */ 579 lwkt_reltoken(&p->p_token); 580 581 return(0); 582 } 583 584 /* 585 * The caller must hold p->p_token 586 */ 587 static void 588 swapout(struct proc *p) 589 { 590 #ifdef INVARIANTS 591 if (swap_debug) 592 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm); 593 #endif 594 ++p->p_ru.ru_nswap; 595 596 /* 597 * remember the process resident count 598 */ 599 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 600 p->p_flags |= P_SWAPPEDOUT; 601 p->p_swtime = 0; 602 } 603 604 #endif /* !NO_SWAPPING */ 605 606