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 91 /* 92 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 93 * 94 * Note: run scheduling should be divorced from the vm system. 95 */ 96 static void scheduler (void *); 97 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL); 98 99 #ifdef INVARIANTS 100 101 static int swap_debug = 0; 102 SYSCTL_INT(_vm, OID_AUTO, swap_debug, 103 CTLFLAG_RW, &swap_debug, 0, ""); 104 105 #endif 106 107 static int scheduler_notify; 108 109 static void swapout (struct proc *); 110 111 /* 112 * No requirements. 113 */ 114 int 115 kernacc(c_caddr_t addr, int len, int rw) 116 { 117 boolean_t rv; 118 vm_offset_t saddr, eaddr; 119 vm_prot_t prot; 120 121 KASSERT((rw & (~VM_PROT_ALL)) == 0, 122 ("illegal ``rw'' argument to kernacc (%x)", rw)); 123 124 /* 125 * The globaldata space is not part of the kernel_map proper, 126 * check access separately. 127 */ 128 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len))) 129 return (TRUE); 130 131 /* 132 * Nominal kernel memory access - check access via kernel_map. 133 */ 134 if ((vm_offset_t)addr + len > vm_map_max(&kernel_map) || 135 (vm_offset_t)addr + len < (vm_offset_t)addr) { 136 return (FALSE); 137 } 138 prot = rw; 139 saddr = trunc_page((vm_offset_t)addr); 140 eaddr = round_page((vm_offset_t)addr + len); 141 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot, FALSE); 142 143 return (rv == TRUE); 144 } 145 146 /* 147 * No requirements. 148 */ 149 int 150 useracc(c_caddr_t addr, int len, int rw) 151 { 152 boolean_t rv; 153 vm_prot_t prot; 154 vm_map_t map; 155 vm_offset_t wrap; 156 vm_offset_t gpa; 157 158 KASSERT((rw & (~VM_PROT_ALL)) == 0, 159 ("illegal ``rw'' argument to useracc (%x)", rw)); 160 prot = rw; 161 162 if (curthread->td_vmm) { 163 if (vmm_vm_get_gpa(curproc, (register_t *)&gpa, (register_t) addr)) 164 panic("%s: could not get GPA\n", __func__); 165 addr = (c_caddr_t) gpa; 166 } 167 168 /* 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 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 180 round_page(wrap), prot, TRUE); 181 vm_map_unlock_read(map); 182 183 return (rv == TRUE); 184 } 185 186 /* 187 * No requirements. 188 */ 189 void 190 vslock(caddr_t addr, u_int len) 191 { 192 if (len) { 193 vm_map_wire(&curproc->p_vmspace->vm_map, 194 trunc_page((vm_offset_t)addr), 195 round_page((vm_offset_t)addr + len), 0); 196 } 197 } 198 199 /* 200 * No requirements. 201 */ 202 void 203 vsunlock(caddr_t addr, u_int len) 204 { 205 if (len) { 206 vm_map_wire(&curproc->p_vmspace->vm_map, 207 trunc_page((vm_offset_t)addr), 208 round_page((vm_offset_t)addr + len), 209 KM_PAGEABLE); 210 } 211 } 212 213 /* 214 * Implement fork's actions on an address space. 215 * Here we arrange for the address space to be copied or referenced, 216 * allocate a user struct (pcb and kernel stack), then call the 217 * machine-dependent layer to fill those in and make the new process 218 * ready to run. The new process is set up so that it returns directly 219 * to user mode to avoid stack copying and relocation problems. 220 * 221 * No requirements. 222 */ 223 void 224 vm_fork(struct proc *p1, struct proc *p2, int flags) 225 { 226 if ((flags & RFPROC) == 0) { 227 /* 228 * Divorce the memory, if it is shared, essentially 229 * this changes shared memory amongst threads, into 230 * COW locally. 231 */ 232 if ((flags & RFMEM) == 0) { 233 if (vmspace_getrefs(p1->p_vmspace) > 1) { 234 vmspace_unshare(p1); 235 } 236 } 237 cpu_fork(ONLY_LWP_IN_PROC(p1), NULL, flags); 238 return; 239 } 240 241 if (flags & RFMEM) { 242 vmspace_ref(p1->p_vmspace); 243 p2->p_vmspace = p1->p_vmspace; 244 } 245 246 while (vm_page_count_severe()) { 247 vm_wait(0); 248 } 249 250 if ((flags & RFMEM) == 0) { 251 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 252 253 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 254 255 if (p1->p_vmspace->vm_shm) 256 shmfork(p1, p2); 257 } 258 259 pmap_init_proc(p2); 260 } 261 262 /* 263 * Set default limits for VM system. Call during proc0's initialization. 264 * 265 * Called from the low level boot code only. 266 */ 267 void 268 vm_init_limits(struct proc *p) 269 { 270 int rss_limit; 271 272 /* 273 * Set up the initial limits on process VM. Set the maximum resident 274 * set size to be half of (reasonably) available memory. Since this 275 * is a soft limit, it comes into effect only when the system is out 276 * of memory - half of main memory helps to favor smaller processes, 277 * and reduces thrashing of the object cache. 278 */ 279 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 280 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 281 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 282 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 283 /* limit the limit to no less than 2MB */ 284 rss_limit = max(vmstats.v_free_count, 512); 285 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 286 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 287 } 288 289 /* 290 * Faultin the specified process. Note that the process can be in any 291 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is 292 * sleeping. 293 * 294 * No requirements. 295 */ 296 void 297 faultin(struct proc *p) 298 { 299 if (p->p_flags & P_SWAPPEDOUT) { 300 /* 301 * The process is waiting in the kernel to return to user 302 * mode but cannot until P_SWAPPEDOUT gets cleared. 303 */ 304 lwkt_gettoken(&p->p_token); 305 p->p_flags &= ~(P_SWAPPEDOUT | P_SWAPWAIT); 306 #ifdef INVARIANTS 307 if (swap_debug) 308 kprintf("swapping in %d (%s)\n", p->p_pid, p->p_comm); 309 #endif 310 wakeup(p); 311 lwkt_reltoken(&p->p_token); 312 } 313 } 314 315 /* 316 * Kernel initialization eventually falls through to this function, 317 * which is process 0. 318 * 319 * This swapin algorithm attempts to swap-in processes only if there 320 * is enough space for them. Of course, if a process waits for a long 321 * time, it will be swapped in anyway. 322 */ 323 struct scheduler_info { 324 struct proc *pp; 325 int ppri; 326 }; 327 328 static int scheduler_callback(struct proc *p, void *data); 329 330 static void 331 scheduler(void *dummy) 332 { 333 struct scheduler_info info; 334 struct proc *p; 335 336 KKASSERT(!IN_CRITICAL_SECT(curthread)); 337 loop: 338 scheduler_notify = 0; 339 /* 340 * Don't try to swap anything in if we are low on memory. 341 */ 342 if (vm_page_count_severe()) { 343 vm_wait(0); 344 goto loop; 345 } 346 347 /* 348 * Look for a good candidate to wake up 349 * 350 * XXX we should make the schedule thread pcpu and then use a 351 * segmented allproc scan. 352 */ 353 info.pp = NULL; 354 info.ppri = INT_MIN; 355 allproc_scan(scheduler_callback, &info, 0); 356 357 /* 358 * Nothing to do, back to sleep for at least 1/10 of a second. If 359 * we are woken up, immediately process the next request. If 360 * multiple requests have built up the first is processed 361 * immediately and the rest are staggered. 362 */ 363 if ((p = info.pp) == NULL) { 364 tsleep(&proc0, 0, "nowork", hz / 10); 365 if (scheduler_notify == 0) 366 tsleep(&scheduler_notify, 0, "nowork", 0); 367 goto loop; 368 } 369 370 /* 371 * Fault the selected process in, then wait for a short period of 372 * time and loop up. 373 * 374 * XXX we need a heuristic to get a measure of system stress and 375 * then adjust our stagger wakeup delay accordingly. 376 */ 377 lwkt_gettoken(&p->p_token); 378 faultin(p); 379 p->p_swtime = 0; 380 lwkt_reltoken(&p->p_token); 381 PRELE(p); 382 tsleep(&proc0, 0, "swapin", hz / 10); 383 goto loop; 384 } 385 386 /* 387 * Process only has its hold count bumped, we need the token 388 * to safely scan the LWPs 389 */ 390 static int 391 scheduler_callback(struct proc *p, void *data) 392 { 393 struct scheduler_info *info = data; 394 struct vmspace *vm; 395 struct lwp *lp; 396 segsz_t pgs; 397 int pri; 398 399 /* 400 * We only care about processes in swap-wait. Interlock test with 401 * token if the flag is found set. 402 */ 403 if ((p->p_flags & P_SWAPWAIT) == 0) 404 return 0; 405 lwkt_gettoken_shared(&p->p_token); 406 if ((p->p_flags & P_SWAPWAIT) == 0) { 407 lwkt_reltoken(&p->p_token); 408 return 0; 409 } 410 411 /* 412 * Calculate priority for swap-in 413 */ 414 pri = 0; 415 FOREACH_LWP_IN_PROC(lp, p) { 416 /* XXX lwp might need a different metric */ 417 pri += lp->lwp_slptime; 418 } 419 pri += p->p_swtime - p->p_nice * 8; 420 421 /* 422 * The more pages paged out while we were swapped, 423 * the more work we have to do to get up and running 424 * again and the lower our wakeup priority. 425 * 426 * Each second of sleep time is worth ~1MB 427 */ 428 if ((vm = p->p_vmspace) != NULL) { 429 vmspace_hold(vm); 430 pgs = vmspace_resident_count(vm); 431 if (pgs < vm->vm_swrss) { 432 pri -= (vm->vm_swrss - pgs) / 433 (1024 * 1024 / PAGE_SIZE); 434 } 435 vmspace_drop(vm); 436 } 437 lwkt_reltoken(&p->p_token); 438 439 /* 440 * If this process is higher priority and there is 441 * enough space, then select this process instead of 442 * the previous selection. 443 */ 444 if (pri > info->ppri) { 445 if (info->pp) 446 PRELE(info->pp); 447 PHOLD(p); 448 info->pp = p; 449 info->ppri = pri; 450 } 451 return(0); 452 } 453 454 /* 455 * SMP races ok. 456 * No requirements. 457 */ 458 void 459 swapin_request(void) 460 { 461 if (scheduler_notify == 0) { 462 scheduler_notify = 1; 463 wakeup(&scheduler_notify); 464 } 465 } 466 467 #ifndef NO_SWAPPING 468 469 #define swappable(p) \ 470 (((p)->p_lock == 0) && \ 471 ((p)->p_flags & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0) 472 473 474 /* 475 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 476 */ 477 static int swap_idle_threshold1 = 15; 478 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 479 CTLFLAG_RW, &swap_idle_threshold1, 0, "Guaranteed process resident time (sec)"); 480 481 /* 482 * Swap_idle_threshold2 is the time that a process can be idle before 483 * it will be swapped out, if idle swapping is enabled. Default is 484 * one minute. 485 */ 486 static int swap_idle_threshold2 = 60; 487 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 488 CTLFLAG_RW, &swap_idle_threshold2, 0, "Time (sec) a process can idle before being swapped"); 489 490 /* 491 * Swapout is driven by the pageout daemon. Very simple, we find eligible 492 * procs and mark them as being swapped out. This will cause the kernel 493 * to prefer to pageout those proc's pages first and the procs in question 494 * will not return to user mode until the swapper tells them they can. 495 * 496 * If any procs have been sleeping/stopped for at least maxslp seconds, 497 * they are swapped. Else, we swap the longest-sleeping or stopped process, 498 * if any, otherwise the longest-resident process. 499 */ 500 501 static int swapout_procs_callback(struct proc *p, void *data); 502 503 /* 504 * No requirements. 505 */ 506 void 507 swapout_procs(int action) 508 { 509 allproc_scan(swapout_procs_callback, &action, 0); 510 } 511 512 static int 513 swapout_procs_callback(struct proc *p, void *data) 514 { 515 struct lwp *lp; 516 int action = *(int *)data; 517 int minslp = -1; 518 519 if (!swappable(p)) 520 return(0); 521 522 lwkt_gettoken(&p->p_token); 523 524 /* 525 * We only consider active processes. 526 */ 527 if (p->p_stat != SACTIVE && p->p_stat != SSTOP) { 528 lwkt_reltoken(&p->p_token); 529 return(0); 530 } 531 532 FOREACH_LWP_IN_PROC(lp, p) { 533 /* 534 * do not swap out a realtime process 535 */ 536 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type)) { 537 lwkt_reltoken(&p->p_token); 538 return(0); 539 } 540 541 /* 542 * Guarentee swap_idle_threshold time in memory 543 */ 544 if (lp->lwp_slptime < swap_idle_threshold1) { 545 lwkt_reltoken(&p->p_token); 546 return(0); 547 } 548 549 /* 550 * If the system is under memory stress, or if we 551 * are swapping idle processes >= swap_idle_threshold2, 552 * then swap the process out. 553 */ 554 if (((action & VM_SWAP_NORMAL) == 0) && 555 (((action & VM_SWAP_IDLE) == 0) || 556 (lp->lwp_slptime < swap_idle_threshold2))) { 557 lwkt_reltoken(&p->p_token); 558 return(0); 559 } 560 561 if (minslp == -1 || lp->lwp_slptime < minslp) 562 minslp = lp->lwp_slptime; 563 } 564 565 /* 566 * If the process has been asleep for awhile, swap 567 * it out. 568 */ 569 if ((action & VM_SWAP_NORMAL) || 570 ((action & VM_SWAP_IDLE) && 571 (minslp > swap_idle_threshold2))) { 572 swapout(p); 573 } 574 575 /* 576 * cleanup our reference 577 */ 578 lwkt_reltoken(&p->p_token); 579 580 return(0); 581 } 582 583 /* 584 * The caller must hold p->p_token 585 */ 586 static void 587 swapout(struct proc *p) 588 { 589 #ifdef INVARIANTS 590 if (swap_debug) 591 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm); 592 #endif 593 ++p->p_ru.ru_nswap; 594 595 /* 596 * remember the process resident count 597 */ 598 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 599 p->p_flags |= P_SWAPPEDOUT; 600 p->p_swtime = 0; 601 } 602 603 #endif /* !NO_SWAPPING */ 604 605