vm_glue.c revision 99563
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 99563 2002-07-08 01:11:10Z peter $ 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_object.h> 91#include <vm/vm_kern.h> 92#include <vm/vm_extern.h> 93#include <vm/vm_pager.h> 94 95#include <sys/user.h> 96 97extern int maxslp; 98 99/* 100 * System initialization 101 * 102 * Note: proc0 from proc.h 103 */ 104static void vm_init_limits(void *); 105SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0) 106 107/* 108 * THIS MUST BE THE LAST INITIALIZATION ITEM!!! 109 * 110 * Note: run scheduling should be divorced from the vm system. 111 */ 112static void scheduler(void *); 113SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL) 114 115#ifndef NO_SWAPPING 116static void swapout(struct proc *); 117static void vm_proc_swapin(struct proc *p); 118static void vm_proc_swapout(struct proc *p); 119#endif 120 121/* 122 * MPSAFE 123 */ 124int 125kernacc(addr, len, rw) 126 caddr_t addr; 127 int len, rw; 128{ 129 boolean_t rv; 130 vm_offset_t saddr, eaddr; 131 vm_prot_t prot; 132 133 KASSERT((rw & ~VM_PROT_ALL) == 0, 134 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 135 prot = rw; 136 saddr = trunc_page((vm_offset_t)addr); 137 eaddr = round_page((vm_offset_t)addr + len); 138 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 139 return (rv == TRUE); 140} 141 142/* 143 * MPSAFE 144 */ 145int 146useracc(addr, len, rw) 147 caddr_t addr; 148 int len, rw; 149{ 150 boolean_t rv; 151 vm_prot_t prot; 152 153 KASSERT((rw & ~VM_PROT_ALL) == 0, 154 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 155 prot = rw; 156 /* 157 * XXX - check separately to disallow access to user area and user 158 * page tables - they are in the map. 159 * 160 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once 161 * only used (as an end address) in trap.c. Use it as an end address 162 * here too. This bogusness has spread. I just fixed where it was 163 * used as a max in vm_mmap.c. 164 */ 165 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS 166 || (vm_offset_t) addr + len < (vm_offset_t) addr) { 167 return (FALSE); 168 } 169 rv = vm_map_check_protection(&curproc->p_vmspace->vm_map, 170 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 171 prot); 172 return (rv == TRUE); 173} 174 175/* 176 * MPSAFE 177 */ 178void 179vslock(addr, len) 180 caddr_t addr; 181 u_int len; 182{ 183 184 vm_map_wire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), 185 round_page((vm_offset_t)addr + len), FALSE); 186} 187 188/* 189 * MPSAFE 190 */ 191void 192vsunlock(addr, len) 193 caddr_t addr; 194 u_int len; 195{ 196 197 vm_map_unwire(&curproc->p_vmspace->vm_map, 198 trunc_page((vm_offset_t)addr), 199 round_page((vm_offset_t)addr + len), FALSE); 200} 201 202/* 203 * Create the U area for a new process. 204 * This routine directly affects the fork perf for a process. 205 */ 206void 207vm_proc_new(struct proc *p) 208{ 209 vm_page_t ma[UAREA_PAGES]; 210 vm_object_t upobj; 211 vm_offset_t up; 212 vm_page_t m; 213 u_int i; 214 215 /* 216 * Allocate object for the upage. 217 */ 218 upobj = vm_object_allocate(OBJT_DEFAULT, UAREA_PAGES); 219 p->p_upages_obj = upobj; 220 221 /* 222 * Get a kernel virtual address for the U area for this process. 223 */ 224 up = kmem_alloc_nofault(kernel_map, UAREA_PAGES * PAGE_SIZE); 225 if (up == 0) 226 panic("vm_proc_new: upage allocation failed"); 227 p->p_uarea = (struct user *)up; 228 229 for (i = 0; i < UAREA_PAGES; i++) { 230 /* 231 * Get a uarea page. 232 */ 233 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 234 ma[i] = m; 235 236 /* 237 * Wire the page. 238 */ 239 m->wire_count++; 240 cnt.v_wire_count++; 241 242 vm_page_wakeup(m); 243 vm_page_flag_clear(m, PG_ZERO); 244 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 245 m->valid = VM_PAGE_BITS_ALL; 246 } 247 248 /* 249 * Enter the pages into the kernel address space. 250 */ 251 pmap_qenter(up, ma, UAREA_PAGES); 252} 253 254/* 255 * Dispose the U area for a process that has exited. 256 * This routine directly impacts the exit perf of a process. 257 * XXX proc_zone is marked UMA_ZONE_NOFREE, so this should never be called. 258 */ 259void 260vm_proc_dispose(struct proc *p) 261{ 262 vm_object_t upobj; 263 vm_offset_t up; 264 vm_page_t m; 265 266 upobj = p->p_upages_obj; 267 if (upobj->resident_page_count != UAREA_PAGES) 268 panic("vm_proc_dispose: incorrect number of pages in upobj"); 269 while ((m = TAILQ_FIRST(&upobj->memq)) != NULL) { 270 vm_page_busy(m); 271 vm_page_unwire(m, 0); 272 vm_page_free(m); 273 } 274 up = (vm_offset_t)p->p_uarea; 275 pmap_qremove(up, UAREA_PAGES); 276 kmem_free(kernel_map, up, UAREA_PAGES * PAGE_SIZE); 277 vm_object_deallocate(upobj); 278} 279 280#ifndef NO_SWAPPING 281/* 282 * Allow the U area for a process to be prejudicially paged out. 283 */ 284void 285vm_proc_swapout(struct proc *p) 286{ 287 vm_object_t upobj; 288 vm_offset_t up; 289 vm_page_t m; 290 291 upobj = p->p_upages_obj; 292 if (upobj->resident_page_count != UAREA_PAGES) 293 panic("vm_proc_dispose: incorrect number of pages in upobj"); 294 TAILQ_FOREACH(m, &upobj->memq, listq) { 295 vm_page_dirty(m); 296 vm_page_unwire(m, 0); 297 } 298 up = (vm_offset_t)p->p_uarea; 299 pmap_qremove(up, UAREA_PAGES); 300} 301 302/* 303 * Bring the U area for a specified process back in. 304 */ 305void 306vm_proc_swapin(struct proc *p) 307{ 308 vm_page_t ma[UAREA_PAGES]; 309 vm_object_t upobj; 310 vm_offset_t up; 311 vm_page_t m; 312 int rv; 313 int i; 314 315 upobj = p->p_upages_obj; 316 for (i = 0; i < UAREA_PAGES; i++) { 317 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 318 if (m->valid != VM_PAGE_BITS_ALL) { 319 rv = vm_pager_get_pages(upobj, &m, 1, 0); 320 if (rv != VM_PAGER_OK) 321 panic("vm_proc_swapin: cannot get upage"); 322 m = vm_page_lookup(upobj, i); 323 m->valid = VM_PAGE_BITS_ALL; 324 } 325 ma[i] = m; 326 vm_page_wire(m); 327 vm_page_wakeup(m); 328 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 329 } 330 up = (vm_offset_t)p->p_uarea; 331 pmap_qenter(up, ma, UAREA_PAGES); 332} 333#endif 334 335/* 336 * Implement fork's actions on an address space. 337 * Here we arrange for the address space to be copied or referenced, 338 * allocate a user struct (pcb and kernel stack), then call the 339 * machine-dependent layer to fill those in and make the new process 340 * ready to run. The new process is set up so that it returns directly 341 * to user mode to avoid stack copying and relocation problems. 342 */ 343void 344vm_forkproc(td, p2, td2, flags) 345 struct thread *td; 346 struct proc *p2; 347 struct thread *td2; 348 int flags; 349{ 350 struct proc *p1 = td->td_proc; 351 struct user *up; 352 353 GIANT_REQUIRED; 354 355 if ((flags & RFPROC) == 0) { 356 /* 357 * Divorce the memory, if it is shared, essentially 358 * this changes shared memory amongst threads, into 359 * COW locally. 360 */ 361 if ((flags & RFMEM) == 0) { 362 if (p1->p_vmspace->vm_refcnt > 1) { 363 vmspace_unshare(p1); 364 } 365 } 366 cpu_fork(td, p2, td2, flags); 367 return; 368 } 369 370 if (flags & RFMEM) { 371 p2->p_vmspace = p1->p_vmspace; 372 p1->p_vmspace->vm_refcnt++; 373 } 374 375 while (vm_page_count_severe()) { 376 VM_WAIT; 377 } 378 379 if ((flags & RFMEM) == 0) { 380 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 381 382 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 383 384 if (p1->p_vmspace->vm_shm) 385 shmfork(p1, p2); 386 } 387 388 /* XXXKSE this is unsatisfactory but should be adequate */ 389 up = p2->p_uarea; 390 391 /* 392 * p_stats currently points at fields in the user struct 393 * but not at &u, instead at p_addr. Copy parts of 394 * p_stats; zero the rest of p_stats (statistics). 395 * 396 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 397 * to share sigacts, so we use the up->u_sigacts. 398 */ 399 p2->p_stats = &up->u_stats; 400 if (p2->p_sigacts == NULL) { 401 if (p2->p_procsig->ps_refcnt != 1) 402 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 403 p2->p_sigacts = &up->u_sigacts; 404 up->u_sigacts = *p1->p_sigacts; 405 } 406 407 bzero(&up->u_stats.pstat_startzero, 408 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 409 (caddr_t) &up->u_stats.pstat_startzero)); 410 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 411 ((caddr_t) &up->u_stats.pstat_endcopy - 412 (caddr_t) &up->u_stats.pstat_startcopy)); 413 414 415 /* 416 * cpu_fork will copy and update the pcb, set up the kernel stack, 417 * and make the child ready to run. 418 */ 419 cpu_fork(td, p2, td2, flags); 420} 421 422/* 423 * Called after process has been wait(2)'ed apon and is being reaped. 424 * The idea is to reclaim resources that we could not reclaim while 425 * the process was still executing. 426 */ 427void 428vm_waitproc(p) 429 struct proc *p; 430{ 431 struct thread *td; 432 433 GIANT_REQUIRED; 434 cpu_wait(p); 435/* XXXKSE by here there should not be any threads left! */ 436 FOREACH_THREAD_IN_PROC(p, td) { 437 panic("vm_waitproc: Survivor thread!"); 438 } 439 vmspace_exitfree(p); /* and clean-out the vmspace */ 440} 441 442/* 443 * Set default limits for VM system. 444 * Called for proc 0, and then inherited by all others. 445 * 446 * XXX should probably act directly on proc0. 447 */ 448static void 449vm_init_limits(udata) 450 void *udata; 451{ 452 struct proc *p = udata; 453 int rss_limit; 454 455 /* 456 * Set up the initial limits on process VM. Set the maximum resident 457 * set size to be half of (reasonably) available memory. Since this 458 * is a soft limit, it comes into effect only when the system is out 459 * of memory - half of main memory helps to favor smaller processes, 460 * and reduces thrashing of the object cache. 461 */ 462 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 463 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 464 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 465 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 466 /* limit the limit to no less than 2MB */ 467 rss_limit = max(cnt.v_free_count, 512); 468 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 469 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 470} 471 472void 473faultin(p) 474 struct proc *p; 475{ 476 477 GIANT_REQUIRED; 478 PROC_LOCK_ASSERT(p, MA_OWNED); 479 mtx_lock_spin(&sched_lock); 480#ifdef NO_SWAPPING 481 if ((p->p_sflag & PS_INMEM) == 0) 482 panic("faultin: proc swapped out with NO_SWAPPING!"); 483#else 484 if ((p->p_sflag & PS_INMEM) == 0) { 485 struct thread *td; 486 487 ++p->p_lock; 488 mtx_unlock_spin(&sched_lock); 489 PROC_UNLOCK(p); 490 491 vm_proc_swapin(p); 492 FOREACH_THREAD_IN_PROC (p, td) 493 pmap_swapin_thread(td); 494 495 PROC_LOCK(p); 496 mtx_lock_spin(&sched_lock); 497 FOREACH_THREAD_IN_PROC (p, td) 498 if (td->td_state == TDS_RUNQ) /* XXXKSE */ 499 setrunqueue(td); 500 501 p->p_sflag |= PS_INMEM; 502 503 /* undo the effect of setting SLOCK above */ 504 --p->p_lock; 505 } 506#endif 507 mtx_unlock_spin(&sched_lock); 508} 509 510/* 511 * This swapin algorithm attempts to swap-in processes only if there 512 * is enough space for them. Of course, if a process waits for a long 513 * time, it will be swapped in anyway. 514 * 515 * XXXKSE - process with the thread with highest priority counts.. 516 * 517 * Giant is still held at this point, to be released in tsleep. 518 */ 519/* ARGSUSED*/ 520static void 521scheduler(dummy) 522 void *dummy; 523{ 524 struct proc *p; 525 struct thread *td; 526 int pri; 527 struct proc *pp; 528 int ppri; 529 530 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 531 /* GIANT_REQUIRED */ 532 533loop: 534 if (vm_page_count_min()) { 535 VM_WAIT; 536 goto loop; 537 } 538 539 pp = NULL; 540 ppri = INT_MIN; 541 sx_slock(&allproc_lock); 542 FOREACH_PROC_IN_SYSTEM(p) { 543 struct ksegrp *kg; 544 if (p->p_sflag & (PS_INMEM | PS_SWAPPING)) { 545 continue; 546 } 547 mtx_lock_spin(&sched_lock); 548 FOREACH_THREAD_IN_PROC(p, td) { 549 /* Only consider runnable threads */ 550 if (td->td_state == TDS_RUNQ) { 551 kg = td->td_ksegrp; 552 pri = p->p_swtime + kg->kg_slptime; 553 if ((p->p_sflag & PS_SWAPINREQ) == 0) { 554 pri -= kg->kg_nice * 8; 555 } 556 557 /* 558 * if this ksegrp is higher priority 559 * and there is enough space, then select 560 * this process instead of the previous 561 * selection. 562 */ 563 if (pri > ppri) { 564 pp = p; 565 ppri = pri; 566 } 567 } 568 } 569 mtx_unlock_spin(&sched_lock); 570 } 571 sx_sunlock(&allproc_lock); 572 573 /* 574 * Nothing to do, back to sleep. 575 */ 576 if ((p = pp) == NULL) { 577 tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 578 goto loop; 579 } 580 mtx_lock_spin(&sched_lock); 581 p->p_sflag &= ~PS_SWAPINREQ; 582 mtx_unlock_spin(&sched_lock); 583 584 /* 585 * We would like to bring someone in. (only if there is space). 586 * [What checks the space? ] 587 */ 588 PROC_LOCK(p); 589 faultin(p); 590 PROC_UNLOCK(p); 591 mtx_lock_spin(&sched_lock); 592 p->p_swtime = 0; 593 mtx_unlock_spin(&sched_lock); 594 goto loop; 595} 596 597#ifndef NO_SWAPPING 598 599/* 600 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 601 */ 602static int swap_idle_threshold1 = 2; 603SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 604 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 605 606/* 607 * Swap_idle_threshold2 is the time that a process can be idle before 608 * it will be swapped out, if idle swapping is enabled. 609 */ 610static int swap_idle_threshold2 = 10; 611SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 612 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 613 614/* 615 * Swapout is driven by the pageout daemon. Very simple, we find eligible 616 * procs and unwire their u-areas. We try to always "swap" at least one 617 * process in case we need the room for a swapin. 618 * If any procs have been sleeping/stopped for at least maxslp seconds, 619 * they are swapped. Else, we swap the longest-sleeping or stopped process, 620 * if any, otherwise the longest-resident process. 621 */ 622void 623swapout_procs(action) 624int action; 625{ 626 struct proc *p; 627 struct thread *td; 628 struct ksegrp *kg; 629 struct proc *outp, *outp2; 630 int outpri, outpri2; 631 int didswap = 0; 632 633 GIANT_REQUIRED; 634 635 outp = outp2 = NULL; 636 outpri = outpri2 = INT_MIN; 637retry: 638 sx_slock(&allproc_lock); 639 FOREACH_PROC_IN_SYSTEM(p) { 640 struct vmspace *vm; 641 int minslptime = 100000; 642 643 PROC_LOCK(p); 644 if (p->p_lock != 0 || 645 (p->p_flag & (P_STOPPED_SNGL|P_TRACED|P_SYSTEM|P_WEXIT)) != 0) { 646 PROC_UNLOCK(p); 647 continue; 648 } 649 /* 650 * only aiod changes vmspace, however it will be 651 * skipped because of the if statement above checking 652 * for P_SYSTEM 653 */ 654 vm = p->p_vmspace; 655 mtx_lock_spin(&sched_lock); 656 if ((p->p_sflag & (PS_INMEM|PS_SWAPPING)) != PS_INMEM) { 657 mtx_unlock_spin(&sched_lock); 658 PROC_UNLOCK(p); 659 continue; 660 } 661 662 switch (p->p_state) { 663 default: 664 /* Don't swap out processes in any sort 665 * of 'special' state. */ 666 mtx_unlock_spin(&sched_lock); 667 PROC_UNLOCK(p); 668 continue; 669 670 case PRS_NORMAL: 671 /* 672 * do not swapout a realtime process 673 * Check all the thread groups.. 674 */ 675 FOREACH_KSEGRP_IN_PROC(p, kg) { 676 if (PRI_IS_REALTIME(kg->kg_pri_class)) { 677 mtx_unlock_spin(&sched_lock); 678 PROC_UNLOCK(p); 679 goto nextproc; 680 } 681 682 /* 683 * Do not swapout a process waiting 684 * on a critical event of some kind. 685 * Also guarantee swap_idle_threshold1 686 * time in memory. 687 */ 688 if (kg->kg_slptime < swap_idle_threshold1) { 689 mtx_unlock_spin(&sched_lock); 690 PROC_UNLOCK(p); 691 goto nextproc; 692 } 693 FOREACH_THREAD_IN_PROC(p, td) { 694 if ((td->td_priority) < PSOCK) { 695 mtx_unlock_spin(&sched_lock); 696 PROC_UNLOCK(p); 697 goto nextproc; 698 } 699 } 700 /* 701 * If the system is under memory stress, 702 * or if we are swapping 703 * idle processes >= swap_idle_threshold2, 704 * then swap the process out. 705 */ 706 if (((action & VM_SWAP_NORMAL) == 0) && 707 (((action & VM_SWAP_IDLE) == 0) || 708 (kg->kg_slptime < swap_idle_threshold2))) { 709 mtx_unlock_spin(&sched_lock); 710 PROC_UNLOCK(p); 711 goto nextproc; 712 } 713 if (minslptime > kg->kg_slptime) 714 minslptime = kg->kg_slptime; 715 } 716 717 mtx_unlock_spin(&sched_lock); 718 ++vm->vm_refcnt; 719 /* 720 * do not swapout a process that 721 * is waiting for VM 722 * data structures there is a 723 * possible deadlock. 724 */ 725 if (!vm_map_trylock(&vm->vm_map)) { 726 vmspace_free(vm); 727 PROC_UNLOCK(p); 728 goto nextproc; 729 } 730 vm_map_unlock(&vm->vm_map); 731 /* 732 * If the process has been asleep for awhile and had 733 * most of its pages taken away already, swap it out. 734 */ 735 if ((action & VM_SWAP_NORMAL) || 736 ((action & VM_SWAP_IDLE) && 737 (minslptime > swap_idle_threshold2))) { 738 sx_sunlock(&allproc_lock); 739 swapout(p); 740 vmspace_free(vm); 741 didswap++; 742 goto retry; 743 } 744 PROC_UNLOCK(p); 745 vmspace_free(vm); 746 } 747nextproc: 748 continue; 749 } 750 sx_sunlock(&allproc_lock); 751 /* 752 * If we swapped something out, and another process needed memory, 753 * then wakeup the sched process. 754 */ 755 if (didswap) 756 wakeup(&proc0); 757} 758 759static void 760swapout(p) 761 struct proc *p; 762{ 763 struct thread *td; 764 765 PROC_LOCK_ASSERT(p, MA_OWNED); 766#if defined(SWAP_DEBUG) 767 printf("swapping out %d\n", p->p_pid); 768#endif 769 ++p->p_stats->p_ru.ru_nswap; 770 /* 771 * remember the process resident count 772 */ 773 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 774 775 mtx_lock_spin(&sched_lock); 776 p->p_sflag &= ~PS_INMEM; 777 p->p_sflag |= PS_SWAPPING; 778 PROC_UNLOCK(p); 779 FOREACH_THREAD_IN_PROC (p, td) 780 if (td->td_state == TDS_RUNQ) /* XXXKSE */ 781 remrunqueue(td); /* XXXKSE */ 782 mtx_unlock_spin(&sched_lock); 783 784 vm_proc_swapout(p); 785 FOREACH_THREAD_IN_PROC(p, td) 786 pmap_swapout_thread(td); 787 mtx_lock_spin(&sched_lock); 788 p->p_sflag &= ~PS_SWAPPING; 789 p->p_swtime = 0; 790 mtx_unlock_spin(&sched_lock); 791} 792#endif /* !NO_SWAPPING */ 793