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 #include <sys/sysref2.h> 92 93 /* 94 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 95 * 96 * Note: run scheduling should be divorced from the vm system. 97 */ 98 static void scheduler (void *); 99 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL); 100 101 #ifdef INVARIANTS 102 103 static int swap_debug = 0; 104 SYSCTL_INT(_vm, OID_AUTO, swap_debug, 105 CTLFLAG_RW, &swap_debug, 0, ""); 106 107 #endif 108 109 static int scheduler_notify; 110 111 static void swapout (struct proc *); 112 113 /* 114 * No requirements. 115 */ 116 int 117 kernacc(c_caddr_t addr, int len, int rw) 118 { 119 boolean_t rv; 120 vm_offset_t saddr, eaddr; 121 vm_prot_t prot; 122 123 KASSERT((rw & (~VM_PROT_ALL)) == 0, 124 ("illegal ``rw'' argument to kernacc (%x)", rw)); 125 126 /* 127 * The globaldata space is not part of the kernel_map proper, 128 * check access separately. 129 */ 130 if (is_globaldata_space((vm_offset_t)addr, (vm_offset_t)(addr + len))) 131 return (TRUE); 132 133 /* 134 * Nominal kernel memory access - check access via kernel_map. 135 */ 136 if ((vm_offset_t)addr + len > kernel_map.max_offset || 137 (vm_offset_t)addr + len < (vm_offset_t)addr) { 138 return (FALSE); 139 } 140 prot = rw; 141 saddr = trunc_page((vm_offset_t)addr); 142 eaddr = round_page((vm_offset_t)addr + len); 143 rv = vm_map_check_protection(&kernel_map, saddr, eaddr, prot, FALSE); 144 145 return (rv == TRUE); 146 } 147 148 /* 149 * No requirements. 150 */ 151 int 152 useracc(c_caddr_t addr, int len, int rw) 153 { 154 boolean_t rv; 155 vm_prot_t prot; 156 vm_map_t map; 157 vm_offset_t wrap; 158 vm_offset_t gpa; 159 160 KASSERT((rw & (~VM_PROT_ALL)) == 0, 161 ("illegal ``rw'' argument to useracc (%x)", rw)); 162 prot = rw; 163 164 if (curthread->td_vmm) { 165 if (vmm_vm_get_gpa(curproc, (register_t *)&gpa, (register_t) addr)) 166 panic("%s: could not get GPA\n", __func__); 167 addr = (c_caddr_t) gpa; 168 } 169 170 /* 171 * XXX - check separately to disallow access to user area and user 172 * page tables - they are in the map. 173 */ 174 wrap = (vm_offset_t)addr + len; 175 if (wrap > VM_MAX_USER_ADDRESS || wrap < (vm_offset_t)addr) { 176 return (FALSE); 177 } 178 map = &curproc->p_vmspace->vm_map; 179 vm_map_lock_read(map); 180 181 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 182 round_page(wrap), prot, TRUE); 183 vm_map_unlock_read(map); 184 185 return (rv == TRUE); 186 } 187 188 /* 189 * No requirements. 190 */ 191 void 192 vslock(caddr_t addr, u_int len) 193 { 194 if (len) { 195 vm_map_wire(&curproc->p_vmspace->vm_map, 196 trunc_page((vm_offset_t)addr), 197 round_page((vm_offset_t)addr + len), 0); 198 } 199 } 200 201 /* 202 * No requirements. 203 */ 204 void 205 vsunlock(caddr_t addr, u_int len) 206 { 207 if (len) { 208 vm_map_wire(&curproc->p_vmspace->vm_map, 209 trunc_page((vm_offset_t)addr), 210 round_page((vm_offset_t)addr + len), 211 KM_PAGEABLE); 212 } 213 } 214 215 /* 216 * Implement fork's actions on an address space. 217 * Here we arrange for the address space to be copied or referenced, 218 * allocate a user struct (pcb and kernel stack), then call the 219 * machine-dependent layer to fill those in and make the new process 220 * ready to run. The new process is set up so that it returns directly 221 * to user mode to avoid stack copying and relocation problems. 222 * 223 * No requirements. 224 */ 225 void 226 vm_fork(struct proc *p1, struct proc *p2, int flags) 227 { 228 if ((flags & RFPROC) == 0) { 229 /* 230 * Divorce the memory, if it is shared, essentially 231 * this changes shared memory amongst threads, into 232 * COW locally. 233 */ 234 if ((flags & RFMEM) == 0) { 235 if (vmspace_getrefs(p1->p_vmspace) > 1) { 236 vmspace_unshare(p1); 237 } 238 } 239 cpu_fork(ONLY_LWP_IN_PROC(p1), NULL, flags); 240 return; 241 } 242 243 if (flags & RFMEM) { 244 vmspace_ref(p1->p_vmspace); 245 p2->p_vmspace = p1->p_vmspace; 246 } 247 248 while (vm_page_count_severe()) { 249 vm_wait(0); 250 } 251 252 if ((flags & RFMEM) == 0) { 253 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 254 255 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 256 257 if (p1->p_vmspace->vm_shm) 258 shmfork(p1, p2); 259 } 260 261 pmap_init_proc(p2); 262 } 263 264 /* 265 * Set default limits for VM system. Call during proc0's initialization. 266 * 267 * Called from the low level boot code only. 268 */ 269 void 270 vm_init_limits(struct proc *p) 271 { 272 int rss_limit; 273 274 /* 275 * Set up the initial limits on process VM. Set the maximum resident 276 * set size to be half of (reasonably) available memory. Since this 277 * is a soft limit, it comes into effect only when the system is out 278 * of memory - half of main memory helps to favor smaller processes, 279 * and reduces thrashing of the object cache. 280 */ 281 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 282 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 283 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 284 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 285 /* limit the limit to no less than 2MB */ 286 rss_limit = max(vmstats.v_free_count, 512); 287 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 288 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 289 } 290 291 /* 292 * Faultin the specified process. Note that the process can be in any 293 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is 294 * sleeping. 295 * 296 * No requirements. 297 */ 298 void 299 faultin(struct proc *p) 300 { 301 if (p->p_flags & P_SWAPPEDOUT) { 302 /* 303 * The process is waiting in the kernel to return to user 304 * mode but cannot until P_SWAPPEDOUT gets cleared. 305 */ 306 lwkt_gettoken(&p->p_token); 307 p->p_flags &= ~(P_SWAPPEDOUT | P_SWAPWAIT); 308 #ifdef INVARIANTS 309 if (swap_debug) 310 kprintf("swapping in %d (%s)\n", p->p_pid, p->p_comm); 311 #endif 312 wakeup(p); 313 lwkt_reltoken(&p->p_token); 314 } 315 } 316 317 /* 318 * Kernel initialization eventually falls through to this function, 319 * which is process 0. 320 * 321 * This swapin algorithm attempts to swap-in processes only if there 322 * is enough space for them. Of course, if a process waits for a long 323 * time, it will be swapped in anyway. 324 */ 325 struct scheduler_info { 326 struct proc *pp; 327 int ppri; 328 }; 329 330 static int scheduler_callback(struct proc *p, void *data); 331 332 static void 333 scheduler(void *dummy) 334 { 335 struct scheduler_info info; 336 struct proc *p; 337 338 KKASSERT(!IN_CRITICAL_SECT(curthread)); 339 loop: 340 scheduler_notify = 0; 341 /* 342 * Don't try to swap anything in if we are low on memory. 343 */ 344 if (vm_page_count_severe()) { 345 vm_wait(0); 346 goto loop; 347 } 348 349 /* 350 * Look for a good candidate to wake up 351 * 352 * XXX we should make the schedule thread pcpu and then use a 353 * segmented allproc scan. 354 */ 355 info.pp = NULL; 356 info.ppri = INT_MIN; 357 allproc_scan(scheduler_callback, &info, 0); 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(&p->p_token); 380 faultin(p); 381 p->p_swtime = 0; 382 lwkt_reltoken(&p->p_token); 383 PRELE(p); 384 tsleep(&proc0, 0, "swapin", hz / 10); 385 goto loop; 386 } 387 388 /* 389 * Process only has its hold count bumped, we need the token 390 * to safely scan the LWPs 391 */ 392 static int 393 scheduler_callback(struct proc *p, void *data) 394 { 395 struct scheduler_info *info = data; 396 struct vmspace *vm; 397 struct lwp *lp; 398 segsz_t pgs; 399 int pri; 400 401 /* 402 * We only care about processes in swap-wait. Interlock test with 403 * token if the flag is found set. 404 */ 405 if ((p->p_flags & P_SWAPWAIT) == 0) 406 return 0; 407 lwkt_gettoken_shared(&p->p_token); 408 if ((p->p_flags & P_SWAPWAIT) == 0) { 409 lwkt_reltoken(&p->p_token); 410 return 0; 411 } 412 413 /* 414 * Calculate priority for swap-in 415 */ 416 pri = 0; 417 FOREACH_LWP_IN_PROC(lp, p) { 418 /* XXX lwp might need a different metric */ 419 pri += lp->lwp_slptime; 420 } 421 pri += p->p_swtime - p->p_nice * 8; 422 423 /* 424 * The more pages paged out while we were swapped, 425 * the more work we have to do to get up and running 426 * again and the lower our wakeup priority. 427 * 428 * Each second of sleep time is worth ~1MB 429 */ 430 if ((vm = p->p_vmspace) != NULL) { 431 vmspace_hold(vm); 432 pgs = vmspace_resident_count(vm); 433 if (pgs < vm->vm_swrss) { 434 pri -= (vm->vm_swrss - pgs) / 435 (1024 * 1024 / PAGE_SIZE); 436 } 437 vmspace_drop(vm); 438 } 439 lwkt_reltoken(&p->p_token); 440 441 /* 442 * If this process is higher priority and there is 443 * enough space, then select this process instead of 444 * the previous selection. 445 */ 446 if (pri > info->ppri) { 447 if (info->pp) 448 PRELE(info->pp); 449 PHOLD(p); 450 info->pp = p; 451 info->ppri = pri; 452 } 453 return(0); 454 } 455 456 /* 457 * SMP races ok. 458 * No requirements. 459 */ 460 void 461 swapin_request(void) 462 { 463 if (scheduler_notify == 0) { 464 scheduler_notify = 1; 465 wakeup(&scheduler_notify); 466 } 467 } 468 469 #ifndef NO_SWAPPING 470 471 #define swappable(p) \ 472 (((p)->p_lock == 0) && \ 473 ((p)->p_flags & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0) 474 475 476 /* 477 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 478 */ 479 static int swap_idle_threshold1 = 15; 480 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 481 CTLFLAG_RW, &swap_idle_threshold1, 0, "Guaranteed process resident time (sec)"); 482 483 /* 484 * Swap_idle_threshold2 is the time that a process can be idle before 485 * it will be swapped out, if idle swapping is enabled. Default is 486 * one minute. 487 */ 488 static int swap_idle_threshold2 = 60; 489 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 490 CTLFLAG_RW, &swap_idle_threshold2, 0, "Time (sec) a process can idle before being swapped"); 491 492 /* 493 * Swapout is driven by the pageout daemon. Very simple, we find eligible 494 * procs and mark them as being swapped out. This will cause the kernel 495 * to prefer to pageout those proc's pages first and the procs in question 496 * will not return to user mode until the swapper tells them they can. 497 * 498 * If any procs have been sleeping/stopped for at least maxslp seconds, 499 * they are swapped. Else, we swap the longest-sleeping or stopped process, 500 * if any, otherwise the longest-resident process. 501 */ 502 503 static int swapout_procs_callback(struct proc *p, void *data); 504 505 /* 506 * No requirements. 507 */ 508 void 509 swapout_procs(int action) 510 { 511 allproc_scan(swapout_procs_callback, &action, 0); 512 } 513 514 static int 515 swapout_procs_callback(struct proc *p, void *data) 516 { 517 struct lwp *lp; 518 int action = *(int *)data; 519 int minslp = -1; 520 521 if (!swappable(p)) 522 return(0); 523 524 lwkt_gettoken(&p->p_token); 525 526 /* 527 * We only consider active processes. 528 */ 529 if (p->p_stat != SACTIVE && p->p_stat != SSTOP) { 530 lwkt_reltoken(&p->p_token); 531 return(0); 532 } 533 534 FOREACH_LWP_IN_PROC(lp, p) { 535 /* 536 * do not swap out a realtime process 537 */ 538 if (RTP_PRIO_IS_REALTIME(lp->lwp_rtprio.type)) { 539 lwkt_reltoken(&p->p_token); 540 return(0); 541 } 542 543 /* 544 * Guarentee swap_idle_threshold time in memory 545 */ 546 if (lp->lwp_slptime < swap_idle_threshold1) { 547 lwkt_reltoken(&p->p_token); 548 return(0); 549 } 550 551 /* 552 * If the system is under memory stress, or if we 553 * are swapping idle processes >= swap_idle_threshold2, 554 * then swap the process out. 555 */ 556 if (((action & VM_SWAP_NORMAL) == 0) && 557 (((action & VM_SWAP_IDLE) == 0) || 558 (lp->lwp_slptime < swap_idle_threshold2))) { 559 lwkt_reltoken(&p->p_token); 560 return(0); 561 } 562 563 if (minslp == -1 || lp->lwp_slptime < minslp) 564 minslp = lp->lwp_slptime; 565 } 566 567 /* 568 * If the process has been asleep for awhile, swap 569 * it out. 570 */ 571 if ((action & VM_SWAP_NORMAL) || 572 ((action & VM_SWAP_IDLE) && 573 (minslp > swap_idle_threshold2))) { 574 swapout(p); 575 } 576 577 /* 578 * cleanup our reference 579 */ 580 lwkt_reltoken(&p->p_token); 581 582 return(0); 583 } 584 585 /* 586 * The caller must hold p->p_token 587 */ 588 static void 589 swapout(struct proc *p) 590 { 591 #ifdef INVARIANTS 592 if (swap_debug) 593 kprintf("swapping out %d (%s)\n", p->p_pid, p->p_comm); 594 #endif 595 ++p->p_ru.ru_nswap; 596 597 /* 598 * remember the process resident count 599 */ 600 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 601 p->p_flags |= P_SWAPPEDOUT; 602 p->p_swtime = 0; 603 } 604 605 #endif /* !NO_SWAPPING */ 606 607