vm_glue.c revision 90361
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. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 37 * 38 * 39 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 40 * All rights reserved. 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 * 62 * $FreeBSD: head/sys/vm/vm_glue.c 90361 2002-02-07 20:58:47Z julian $ 63 */ 64 65#include "opt_vm.h" 66 67#include <sys/param.h> 68#include <sys/systm.h> 69#include <sys/lock.h> 70#include <sys/mutex.h> 71#include <sys/proc.h> 72#include <sys/resourcevar.h> 73#include <sys/shm.h> 74#include <sys/vmmeter.h> 75#include <sys/sx.h> 76#include <sys/sysctl.h> 77 78#include <sys/kernel.h> 79#include <sys/ktr.h> 80#include <sys/unistd.h> 81 82#include <machine/limits.h> 83 84#include <vm/vm.h> 85#include <vm/vm_param.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 95extern int maxslp; 96 97/* 98 * System initialization 99 * 100 * Note: proc0 from proc.h 101 */ 102 103static void vm_init_limits __P((void *)); 104SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 105 106/* 107 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 108 * 109 * Note: run scheduling should be divorced from the vm system. 110 */ 111static void scheduler __P((void *)); 112SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 113 114#ifndef NO_SWAPPING 115static void swapout __P((struct proc *)); 116#endif 117 118int 119kernacc(addr, len, rw) 120 caddr_t addr; 121 int len, rw; 122{ 123 boolean_t rv; 124 vm_offset_t saddr, eaddr; 125 vm_prot_t prot; 126 127 KASSERT((rw & ~VM_PROT_ALL) == 0, 128 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 129 prot = rw; 130 saddr = trunc_page((vm_offset_t)addr); 131 eaddr = round_page((vm_offset_t)addr + len); 132 vm_map_lock_read(kernel_map); 133 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 134 vm_map_unlock_read(kernel_map); 135 return (rv == TRUE); 136} 137 138int 139useracc(addr, len, rw) 140 caddr_t addr; 141 int len, rw; 142{ 143 boolean_t rv; 144 vm_prot_t prot; 145 vm_map_t map; 146 vm_map_entry_t save_hint; 147 148 GIANT_REQUIRED; 149 150 KASSERT((rw & ~VM_PROT_ALL) == 0, 151 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 152 prot = rw; 153 /* 154 * XXX - check separately to disallow access to user area and user 155 * page tables - they are in the map. 156 * 157 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 158 * only used (as an end address) in trap.c. Use it as an end address 159 * here too. This bogusness has spread. I just fixed where it was 160 * used as a max in vm_mmap.c. 161 */ 162 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 163 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 164 return (FALSE); 165 } 166 map = &curproc->p_vmspace->vm_map; 167 vm_map_lock_read(map); 168 /* 169 * We save the map hint, and restore it. Useracc appears to distort 170 * the map hint unnecessarily. 171 */ 172 save_hint = map->hint; 173 rv = vm_map_check_protection(map, 174 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); 175 map->hint = save_hint; 176 vm_map_unlock_read(map); 177 178 return (rv == TRUE); 179} 180 181void 182vslock(addr, len) 183 caddr_t addr; 184 u_int len; 185{ 186 GIANT_REQUIRED; 187 vm_map_pageable(&curproc->p_vmspace->vm_map, 188 trunc_page((vm_offset_t)addr), 189 round_page((vm_offset_t)addr + len), FALSE); 190} 191 192void 193vsunlock(addr, len) 194 caddr_t addr; 195 u_int len; 196{ 197 GIANT_REQUIRED; 198 vm_map_pageable(&curproc->p_vmspace->vm_map, 199 trunc_page((vm_offset_t)addr), 200 round_page((vm_offset_t)addr + len), TRUE); 201} 202 203/* 204 * Implement fork's actions on an address space. 205 * Here we arrange for the address space to be copied or referenced, 206 * allocate a user struct (pcb and kernel stack), then call the 207 * machine-dependent layer to fill those in and make the new process 208 * ready to run. The new process is set up so that it returns directly 209 * to user mode to avoid stack copying and relocation problems. 210 */ 211void 212vm_forkproc(td, p2, td2, flags) 213 struct thread *td; 214 struct proc *p2; 215 struct thread *td2; 216 int flags; 217{ 218 struct proc *p1 = td->td_proc; 219 struct user *up; 220 221 GIANT_REQUIRED; 222 223 if ((flags & RFPROC) == 0) { 224 /* 225 * Divorce the memory, if it is shared, essentially 226 * this changes shared memory amongst threads, into 227 * COW locally. 228 */ 229 if ((flags & RFMEM) == 0) { 230 if (p1->p_vmspace->vm_refcnt > 1) { 231 vmspace_unshare(p1); 232 } 233 } 234 cpu_fork(td, p2, td2, flags); 235 return; 236 } 237 238 if (flags & RFMEM) { 239 p2->p_vmspace = p1->p_vmspace; 240 p1->p_vmspace->vm_refcnt++; 241 } 242 243 while (vm_page_count_severe()) { 244 VM_WAIT; 245 } 246 247 if ((flags & RFMEM) == 0) { 248 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 249 250 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 251 252 if (p1->p_vmspace->vm_shm) 253 shmfork(p1, p2); 254 } 255 256 pmap_new_proc(p2); 257 pmap_new_thread(td2); /* Initial thread */ 258 259 /* XXXKSE this is unsatisfactory but should be adequate */ 260 up = p2->p_uarea; 261 262 /* 263 * p_stats currently points at fields in the user struct 264 * but not at &u, instead at p_addr. Copy parts of 265 * p_stats; zero the rest of p_stats (statistics). 266 * 267 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 268 * to share sigacts, so we use the up->u_sigacts. 269 */ 270 p2->p_stats = &up->u_stats; 271 if (p2->p_sigacts == NULL) { 272 if (p2->p_procsig->ps_refcnt != 1) 273 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 274 p2->p_sigacts = &up->u_sigacts; 275 up->u_sigacts = *p1->p_sigacts; 276 } 277 278 bzero(&up->u_stats.pstat_startzero, 279 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 280 (caddr_t) &up->u_stats.pstat_startzero)); 281 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 282 ((caddr_t) &up->u_stats.pstat_endcopy - 283 (caddr_t) &up->u_stats.pstat_startcopy)); 284 285 286 /* 287 * cpu_fork will copy and update the pcb, set up the kernel stack, 288 * and make the child ready to run. 289 */ 290 cpu_fork(td, p2, td2, flags); 291} 292 293/* 294 * Called after process has been wait(2)'ed apon and is being reaped. 295 * The idea is to reclaim resources that we could not reclaim while 296 * the process was still executing. 297 */ 298void 299vm_waitproc(p) 300 struct proc *p; 301{ 302 struct thread *td; 303 304 GIANT_REQUIRED; 305 cpu_wait(p); 306 pmap_dispose_proc(p); /* drop per-process resources */ 307 FOREACH_THREAD_IN_PROC(p, td) 308 pmap_dispose_thread(td); 309 vmspace_exitfree(p); /* and clean-out the vmspace */ 310} 311 312/* 313 * Set default limits for VM system. 314 * Called for proc 0, and then inherited by all others. 315 * 316 * XXX should probably act directly on proc0. 317 */ 318static void 319vm_init_limits(udata) 320 void *udata; 321{ 322 struct proc *p = udata; 323 int rss_limit; 324 325 /* 326 * Set up the initial limits on process VM. Set the maximum resident 327 * set size to be half of (reasonably) available memory. Since this 328 * is a soft limit, it comes into effect only when the system is out 329 * of memory - half of main memory helps to favor smaller processes, 330 * and reduces thrashing of the object cache. 331 */ 332 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 333 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 334 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 335 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 336 /* limit the limit to no less than 2MB */ 337 rss_limit = max(cnt.v_free_count, 512); 338 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 339 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 340} 341 342void 343faultin(p) 344 struct proc *p; 345{ 346 struct thread *td; 347 GIANT_REQUIRED; 348 349 PROC_LOCK_ASSERT(p, MA_OWNED); 350 mtx_lock_spin(&sched_lock); 351 if ((p->p_sflag & PS_INMEM) == 0) { 352 ++p->p_lock; 353 mtx_unlock_spin(&sched_lock); 354 PROC_UNLOCK(p); 355 356 pmap_swapin_proc(p); 357 FOREACH_THREAD_IN_PROC (p, td) 358 pmap_swapin_thread(td); 359 360 PROC_LOCK(p); 361 mtx_lock_spin(&sched_lock); 362 FOREACH_THREAD_IN_PROC (p, td) 363 if (td->td_proc->p_stat == SRUN) /* XXXKSE */ 364 setrunqueue(td); 365 366 p->p_sflag |= PS_INMEM; 367 368 /* undo the effect of setting SLOCK above */ 369 --p->p_lock; 370 } 371 mtx_unlock_spin(&sched_lock); 372} 373 374/* 375 * This swapin algorithm attempts to swap-in processes only if there 376 * is enough space for them. Of course, if a process waits for a long 377 * time, it will be swapped in anyway. 378 * 379 * XXXKSE - KSEGRP with highest priority counts.. 380 * 381 * Giant is still held at this point, to be released in tsleep. 382 */ 383/* ARGSUSED*/ 384static void 385scheduler(dummy) 386 void *dummy; 387{ 388 struct proc *p; 389 int pri; 390 struct proc *pp; 391 int ppri; 392 393 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 394 /* GIANT_REQUIRED */ 395 396loop: 397 if (vm_page_count_min()) { 398 VM_WAIT; 399 goto loop; 400 } 401 402 pp = NULL; 403 ppri = INT_MIN; 404 sx_slock(&allproc_lock); 405 FOREACH_PROC_IN_SYSTEM(p) { 406 struct ksegrp *kg; 407 mtx_lock_spin(&sched_lock); 408 if (p->p_stat == SRUN 409 && (p->p_sflag & (PS_INMEM | PS_SWAPPING)) == 0) { 410 /* Find the minimum sleeptime for the process */ 411 FOREACH_KSEGRP_IN_PROC(p, kg) { 412 pri = p->p_swtime + kg->kg_slptime; 413 if ((p->p_sflag & PS_SWAPINREQ) == 0) { 414 pri -= kg->kg_nice * 8; 415 } 416 417 418 /* 419 * if this ksegrp is higher priority 420 * and there is enough space, then select 421 * this process instead of the previous 422 * selection. 423 */ 424 if (pri > ppri) { 425 pp = p; 426 ppri = pri; 427 } 428 } 429 } 430 mtx_unlock_spin(&sched_lock); 431 } 432 sx_sunlock(&allproc_lock); 433 434 /* 435 * Nothing to do, back to sleep. 436 */ 437 if ((p = pp) == NULL) { 438 tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 439 goto loop; 440 } 441 mtx_lock_spin(&sched_lock); 442 p->p_sflag &= ~PS_SWAPINREQ; 443 mtx_unlock_spin(&sched_lock); 444 445 /* 446 * We would like to bring someone in. (only if there is space). 447 */ 448 PROC_LOCK(p); 449 faultin(p); 450 PROC_UNLOCK(p); 451 mtx_lock_spin(&sched_lock); 452 p->p_swtime = 0; 453 mtx_unlock_spin(&sched_lock); 454 goto loop; 455} 456 457#ifndef NO_SWAPPING 458 459/* 460 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 461 */ 462static int swap_idle_threshold1 = 2; 463SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 464 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 465 466/* 467 * Swap_idle_threshold2 is the time that a process can be idle before 468 * it will be swapped out, if idle swapping is enabled. 469 */ 470static int swap_idle_threshold2 = 10; 471SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 472 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 473 474/* 475 * Swapout is driven by the pageout daemon. Very simple, we find eligible 476 * procs and unwire their u-areas. We try to always "swap" at least one 477 * process in case we need the room for a swapin. 478 * If any procs have been sleeping/stopped for at least maxslp seconds, 479 * they are swapped. Else, we swap the longest-sleeping or stopped process, 480 * if any, otherwise the longest-resident process. 481 */ 482void 483swapout_procs(action) 484int action; 485{ 486 struct proc *p; 487 struct ksegrp *kg; 488 struct proc *outp, *outp2; 489 int outpri, outpri2; 490 int didswap = 0; 491 492 GIANT_REQUIRED; 493 494 outp = outp2 = NULL; 495 outpri = outpri2 = INT_MIN; 496retry: 497 sx_slock(&allproc_lock); 498 LIST_FOREACH(p, &allproc, p_list) { 499 struct vmspace *vm; 500 int minslptime = 100000; 501 502 PROC_LOCK(p); 503 if (p->p_lock != 0 || 504 (p->p_flag & (P_TRACED|P_SYSTEM|P_WEXIT)) != 0) { 505 PROC_UNLOCK(p); 506 continue; 507 } 508 /* 509 * only aiod changes vmspace, however it will be 510 * skipped because of the if statement above checking 511 * for P_SYSTEM 512 */ 513 vm = p->p_vmspace; 514 mtx_lock_spin(&sched_lock); 515 if ((p->p_sflag & (PS_INMEM|PS_SWAPPING)) != PS_INMEM) { 516 mtx_unlock_spin(&sched_lock); 517 PROC_UNLOCK(p); 518 continue; 519 } 520 521 switch (p->p_stat) { 522 default: 523 mtx_unlock_spin(&sched_lock); 524 PROC_UNLOCK(p); 525 continue; 526 527 case SSLEEP: 528 case SSTOP: 529 /* 530 * do not swapout a realtime process 531 * Check all the thread groups.. 532 */ 533 FOREACH_KSEGRP_IN_PROC(p, kg) { 534 if (PRI_IS_REALTIME(kg->kg_pri.pri_class)) { 535 mtx_unlock_spin(&sched_lock); 536 PROC_UNLOCK(p); 537 goto nextproc; 538 } 539 540 /* 541 * Do not swapout a process waiting 542 * on a critical event of some kind. 543 * Also guarantee swap_idle_threshold1 544 * time in memory. 545 */ 546 if (((kg->kg_pri.pri_level) < PSOCK) || 547 (kg->kg_slptime < swap_idle_threshold1)) { 548 mtx_unlock_spin(&sched_lock); 549 PROC_UNLOCK(p); 550 goto nextproc; 551 } 552 553 /* 554 * If the system is under memory stress, 555 * or if we are swapping 556 * idle processes >= swap_idle_threshold2, 557 * then swap the process out. 558 */ 559 if (((action & VM_SWAP_NORMAL) == 0) && 560 (((action & VM_SWAP_IDLE) == 0) || 561 (kg->kg_slptime < swap_idle_threshold2))) { 562 mtx_unlock_spin(&sched_lock); 563 PROC_UNLOCK(p); 564 goto nextproc; 565 } 566 if (minslptime > kg->kg_slptime) 567 minslptime = kg->kg_slptime; 568 } 569 570 mtx_unlock_spin(&sched_lock); 571 ++vm->vm_refcnt; 572 /* 573 * do not swapout a process that 574 * is waiting for VM 575 * data structures there is a 576 * possible deadlock. 577 */ 578 if (lockmgr(&vm->vm_map.lock, 579 LK_EXCLUSIVE | LK_NOWAIT, 580 NULL, curthread)) { 581 vmspace_free(vm); 582 PROC_UNLOCK(p); 583 goto nextproc; 584 } 585 vm_map_unlock(&vm->vm_map); 586 /* 587 * If the process has been asleep for awhile and had 588 * most of its pages taken away already, swap it out. 589 */ 590 if ((action & VM_SWAP_NORMAL) || 591 ((action & VM_SWAP_IDLE) && 592 (minslptime > swap_idle_threshold2))) { 593 sx_sunlock(&allproc_lock); 594 swapout(p); 595 vmspace_free(vm); 596 didswap++; 597 goto retry; 598 } 599 PROC_UNLOCK(p); 600 vmspace_free(vm); 601 } 602nextproc: 603 } 604 sx_sunlock(&allproc_lock); 605 /* 606 * If we swapped something out, and another process needed memory, 607 * then wakeup the sched process. 608 */ 609 if (didswap) 610 wakeup(&proc0); 611} 612 613static void 614swapout(p) 615 struct proc *p; 616{ 617 struct thread *td; 618 619 PROC_LOCK_ASSERT(p, MA_OWNED); 620#if defined(SWAP_DEBUG) 621 printf("swapping out %d\n", p->p_pid); 622#endif 623 ++p->p_stats->p_ru.ru_nswap; 624 /* 625 * remember the process resident count 626 */ 627 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 628 629 mtx_lock_spin(&sched_lock); 630 p->p_sflag &= ~PS_INMEM; 631 p->p_sflag |= PS_SWAPPING; 632 PROC_UNLOCK(p); 633 FOREACH_THREAD_IN_PROC (p, td) 634 if (td->td_proc->p_stat == SRUN) /* XXXKSE */ 635 remrunqueue(td); /* XXXKSE */ 636 mtx_unlock_spin(&sched_lock); 637 638 pmap_swapout_proc(p); 639 FOREACH_THREAD_IN_PROC(p, td) 640 pmap_swapout_thread(td); 641 642 mtx_lock_spin(&sched_lock); 643 p->p_sflag &= ~PS_SWAPPING; 644 p->p_swtime = 0; 645 mtx_unlock_spin(&sched_lock); 646} 647#endif /* !NO_SWAPPING */ 648