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_page2.h> 85 #include <vm/vm_pageout.h> 86 #include <vm/vm_kern.h> 87 #include <vm/vm_extern.h> 88 89 /* 90 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 91 * 92 * Note: run scheduling should be divorced from the vm system. 93 */ 94 static void scheduler (void *); 95 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL); 96 97 #ifdef INVARIANTS 98 99 static int swap_debug = 0; 100 SYSCTL_INT(_vm, OID_AUTO, swap_debug, 101 CTLFLAG_RW, &swap_debug, 0, ""); 102 103 #endif 104 105 static int scheduler_notify; 106 107 static void swapout (struct proc *); 108 109 /* 110 * No requirements. 111 */ 112 int 113 kernacc(c_caddr_t addr, int len, int rw) 114 { 115 boolean_t rv; 116 vm_offset_t saddr, eaddr; 117 vm_prot_t prot; 118 119 KASSERT((rw & (~VM_PROT_ALL)) == 0, 120 ("illegal ``rw'' argument to kernacc (%x)", rw)); 121 122 /* 123 * The globaldata space is not part of the kernel_map proper, 124 * check access separately. 125 */ 126 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len))) 127 return (TRUE); 128 129 /* 130 * Nominal kernel memory access - check access via kernel_map. 131 */ 132 if ((vm_offset_t)addr + len > vm_map_max(&kernel_map) || 133 (vm_offset_t)addr + len < (vm_offset_t)addr) { 134 return (FALSE); 135 } 136 prot = rw; 137 saddr = trunc_page((vm_offset_t)addr); 138 eaddr = round_page((vm_offset_t)addr + len); 139 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot, FALSE); 140 141 return (rv == TRUE); 142 } 143 144 /* 145 * No requirements. 146 */ 147 int 148 useracc(c_caddr_t addr, int len, int rw) 149 { 150 boolean_t rv; 151 vm_prot_t prot; 152 vm_map_t map; 153 vm_offset_t wrap; 154 vm_offset_t gpa; 155 156 KASSERT((rw & (~VM_PROT_ALL)) == 0, 157 ("illegal ``rw'' argument to useracc (%x)", rw)); 158 prot = rw; 159 160 if (curthread->td_vmm) { 161 if (vmm_vm_get_gpa(curproc, (register_t *)&gpa, (register_t) addr)) 162 panic("%s: could not get GPA\n", __func__); 163 addr = (c_caddr_t) gpa; 164 } 165 166 /* 167 * XXX - check separately to disallow access to user area and user 168 * page tables - they are in the map. 169 */ 170 wrap = (vm_offset_t)addr + len; 171 if (wrap > VM_MAX_USER_ADDRESS || wrap < (vm_offset_t)addr) { 172 return (FALSE); 173 } 174 map = &curproc->p_vmspace->vm_map; 175 vm_map_lock_read(map); 176 177 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 178 round_page(wrap), prot, TRUE); 179 vm_map_unlock_read(map); 180 181 return (rv == TRUE); 182 } 183 184 /* 185 * No requirements. 186 */ 187 void 188 vslock(caddr_t addr, u_int len) 189 { 190 if (len) { 191 vm_map_wire(&curproc->p_vmspace->vm_map, 192 trunc_page((vm_offset_t)addr), 193 round_page((vm_offset_t)addr + len), 0); 194 } 195 } 196 197 /* 198 * No requirements. 199 */ 200 void 201 vsunlock(caddr_t addr, u_int len) 202 { 203 if (len) { 204 vm_map_wire(&curproc->p_vmspace->vm_map, 205 trunc_page((vm_offset_t)addr), 206 round_page((vm_offset_t)addr + len), 207 KM_PAGEABLE); 208 } 209 } 210 211 /* 212 * Implement fork's actions on an address space. 213 * Here we arrange for the address space to be copied or referenced, 214 * allocate a user struct (pcb and kernel stack), then call the 215 * machine-dependent layer to fill those in and make the new process 216 * ready to run. The new process is set up so that it returns directly 217 * to user mode to avoid stack copying and relocation problems. 218 * 219 * No requirements. 220 */ 221 void 222 vm_fork(struct proc *p1, struct proc *p2, int flags) 223 { 224 if ((flags & RFPROC) == 0) { 225 /* 226 * Divorce the memory, if it is shared, essentially 227 * this changes shared memory amongst threads, into 228 * COW locally. 229 */ 230 if ((flags & RFMEM) == 0) { 231 if (vmspace_getrefs(p1->p_vmspace) > 1) { 232 vmspace_unshare(p1); 233 } 234 } 235 cpu_fork(ONLY_LWP_IN_PROC(p1), NULL, flags); 236 return; 237 } 238 239 if (flags & RFMEM) { 240 vmspace_ref(p1->p_vmspace); 241 p2->p_vmspace = p1->p_vmspace; 242 } 243 244 while (vm_page_count_severe()) { 245 vm_wait(0); 246 } 247 248 if ((flags & RFMEM) == 0) { 249 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 250 251 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 252 253 if (p1->p_vmspace->vm_shm) 254 shmfork(p1, p2); 255 } 256 257 pmap_init_proc(p2); 258 } 259 260 /* 261 * Set default limits for VM system. Call during proc0's initialization. 262 * 263 * Called from the low level boot code only. 264 */ 265 void 266 vm_init_limits(struct proc *p) 267 { 268 int rss_limit; 269 270 /* 271 * Set up the initial limits on process VM. Set the maximum resident 272 * set size to be half of (reasonably) available memory. Since this 273 * is a soft limit, it comes into effect only when the system is out 274 * of memory - half of main memory helps to favor smaller processes, 275 * and reduces thrashing of the object cache. 276 */ 277 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 278 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 279 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 280 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 281 /* limit the limit to no less than 2MB */ 282 rss_limit = max(vmstats.v_free_count, 512); 283 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 284 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 285 } 286 287 /* 288 * Faultin the specified process. Note that the process can be in any 289 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is 290 * sleeping. 291 * 292 * No requirements. 293 */ 294 void 295 faultin(struct proc *p) 296 { 297 if (p->p_flags & P_SWAPPEDOUT) { 298 /* 299 * The process is waiting in the kernel to return to user 300 * mode but cannot until P_SWAPPEDOUT gets cleared. 301 */ 302 lwkt_gettoken(&p->p_token); 303 p->p_flags &= ~(P_SWAPPEDOUT | P_SWAPWAIT); 304 #ifdef INVARIANTS 305 if (swap_debug) 306 kprintf("swapping in %d (%s)\n", p->p_pid, p->p_comm); 307 #endif 308 wakeup(p); 309 lwkt_reltoken(&p->p_token); 310 } 311 } 312 313 /* 314 * Kernel initialization eventually falls through to this function, 315 * which is process 0. 316 * 317 * This swapin algorithm attempts to swap-in processes only if there 318 * is enough space for them. Of course, if a process waits for a long 319 * time, it will be swapped in anyway. 320 */ 321 struct scheduler_info { 322 struct proc *pp; 323 int ppri; 324 }; 325 326 static int scheduler_callback(struct proc *p, void *data); 327 328 static void 329 scheduler(void *dummy) 330 { 331 struct scheduler_info info; 332 struct proc *p; 333 334 KKASSERT(!IN_CRITICAL_SECT(curthread)); 335 loop: 336 scheduler_notify = 0; 337 /* 338 * Don't try to swap anything in if we are low on memory. 339 */ 340 if (vm_page_count_severe()) { 341 vm_wait(0); 342 goto loop; 343 } 344 345 /* 346 * Look for a good candidate to wake up 347 * 348 * XXX we should make the schedule thread pcpu and then use a 349 * segmented allproc scan. 350 */ 351 info.pp = NULL; 352 info.ppri = INT_MIN; 353 allproc_scan(scheduler_callback, &info, 0); 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(&p->p_token); 376 faultin(p); 377 p->p_swtime = 0; 378 lwkt_reltoken(&p->p_token); 379 PRELE(p); 380 tsleep(&proc0, 0, "swapin", hz / 10); 381 goto loop; 382 } 383 384 /* 385 * Process only has its hold count bumped, we need the token 386 * to safely scan the LWPs 387 */ 388 static int 389 scheduler_callback(struct proc *p, void *data) 390 { 391 struct scheduler_info *info = data; 392 struct vmspace *vm; 393 struct lwp *lp; 394 segsz_t pgs; 395 int pri; 396 397 /* 398 * We only care about processes in swap-wait. Interlock test with 399 * token if the flag is found set. 400 */ 401 if ((p->p_flags & P_SWAPWAIT) == 0) 402 return 0; 403 lwkt_gettoken_shared(&p->p_token); 404 if ((p->p_flags & P_SWAPWAIT) == 0) { 405 lwkt_reltoken(&p->p_token); 406 return 0; 407 } 408 409 /* 410 * Calculate priority for swap-in 411 */ 412 pri = 0; 413 FOREACH_LWP_IN_PROC(lp, p) { 414 /* XXX lwp might need a different metric */ 415 pri += lp->lwp_slptime; 416 } 417 pri += p->p_swtime - p->p_nice * 8; 418 419 /* 420 * The more pages paged out while we were swapped, 421 * the more work we have to do to get up and running 422 * again and the lower our wakeup priority. 423 * 424 * Each second of sleep time is worth ~1MB 425 */ 426 if ((vm = p->p_vmspace) != NULL) { 427 vmspace_hold(vm); 428 pgs = vmspace_resident_count(vm); 429 if (pgs < vm->vm_swrss) { 430 pri -= (vm->vm_swrss - pgs) / 431 (1024 * 1024 / PAGE_SIZE); 432 } 433 vmspace_drop(vm); 434 } 435 lwkt_reltoken(&p->p_token); 436 437 /* 438 * If this process is higher priority and there is 439 * enough space, then select this process instead of 440 * the previous selection. 441 */ 442 if (pri > info->ppri) { 443 if (info->pp) 444 PRELE(info->pp); 445 PHOLD(p); 446 info->pp = p; 447 info->ppri = pri; 448 } 449 return(0); 450 } 451 452 /* 453 * SMP races ok. 454 * No requirements. 455 */ 456 void 457 swapin_request(void) 458 { 459 if (scheduler_notify == 0) { 460 scheduler_notify = 1; 461 wakeup(&scheduler_notify); 462 } 463 } 464 465 #ifndef NO_SWAPPING 466 467 #define swappable(p) \ 468 (((p)->p_lock == 0) && \ 469 ((p)->p_flags & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0) 470 471 472 /* 473 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 474 */ 475 static int swap_idle_threshold1 = 15; 476 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 477 CTLFLAG_RW, &swap_idle_threshold1, 0, "Guaranteed process resident time (sec)"); 478 479 /* 480 * Swap_idle_threshold2 is the time that a process can be idle before 481 * it will be swapped out, if idle swapping is enabled. Default is 482 * one minute. 483 */ 484 static int swap_idle_threshold2 = 60; 485 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 486 CTLFLAG_RW, &swap_idle_threshold2, 0, "Time (sec) a process can idle before being swapped"); 487 488 /* 489 * Swapout is driven by the pageout daemon. Very simple, we find eligible 490 * procs and mark them as being swapped out. This will cause the kernel 491 * to prefer to pageout those proc's pages first and the procs in question 492 * will not return to user mode until the swapper tells them they can. 493 * 494 * If any procs have been sleeping/stopped for at least maxslp seconds, 495 * they are swapped. Else, we swap the longest-sleeping or stopped process, 496 * if any, otherwise the longest-resident process. 497 */ 498 499 static int swapout_procs_callback(struct proc *p, void *data); 500 501 /* 502 * No requirements. 503 */ 504 void 505 swapout_procs(int action) 506 { 507 allproc_scan(swapout_procs_callback, &action, 0); 508 } 509 510 static int 511 swapout_procs_callback(struct proc *p, void *data) 512 { 513 struct lwp *lp; 514 int action = *(int *)data; 515 int minslp = -1; 516 517 if (!swappable(p)) 518 return(0); 519 520 lwkt_gettoken(&p->p_token); 521 522 /* 523 * We only consider active processes. 524 */ 525 if (p->p_stat != SACTIVE && p->p_stat != SSTOP) { 526 lwkt_reltoken(&p->p_token); 527 return(0); 528 } 529 530 FOREACH_LWP_IN_PROC(lp, p) { 531 /* 532 * do not swap out a realtime process 533 */ 534 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type)) { 535 lwkt_reltoken(&p->p_token); 536 return(0); 537 } 538 539 /* 540 * Guarentee swap_idle_threshold time in memory 541 */ 542 if (lp->lwp_slptime < swap_idle_threshold1) { 543 lwkt_reltoken(&p->p_token); 544 return(0); 545 } 546 547 /* 548 * If the system is under memory stress, or if we 549 * are swapping idle processes >= swap_idle_threshold2, 550 * then swap the process out. 551 */ 552 if (((action & VM_SWAP_NORMAL) == 0) && 553 (((action & VM_SWAP_IDLE) == 0) || 554 (lp->lwp_slptime < swap_idle_threshold2))) { 555 lwkt_reltoken(&p->p_token); 556 return(0); 557 } 558 559 if (minslp == -1 || lp->lwp_slptime < minslp) 560 minslp = lp->lwp_slptime; 561 } 562 563 /* 564 * If the process has been asleep for awhile, swap 565 * it out. 566 */ 567 if ((action & VM_SWAP_NORMAL) || 568 ((action & VM_SWAP_IDLE) && 569 (minslp > swap_idle_threshold2))) { 570 swapout(p); 571 } 572 573 /* 574 * cleanup our reference 575 */ 576 lwkt_reltoken(&p->p_token); 577 578 return(0); 579 } 580 581 /* 582 * The caller must hold p->p_token 583 */ 584 static void 585 swapout(struct proc *p) 586 { 587 #ifdef INVARIANTS 588 if (swap_debug) 589 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm); 590 #endif 591 ++p->p_ru.ru_nswap; 592 593 /* 594 * remember the process resident count 595 */ 596 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 597 p->p_flags |= P_SWAPPEDOUT; 598 p->p_swtime = 0; 599 } 600 601 #endif /* !NO_SWAPPING */ 602 603