vm_glue.c revision 99559
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 99559 2002-07-07 23:05:27Z 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 int i; 266 267 upobj = p->p_upages_obj; 268 up = (vm_offset_t)p->p_uarea; 269 for (i = 0; i < UAREA_PAGES; i++) { 270 m = vm_page_lookup(upobj, i); 271 if (m == NULL) 272 panic("vm_proc_dispose: upage already missing?"); 273 vm_page_busy(m); 274 vm_page_unwire(m, 0); 275 vm_page_free(m); 276 } 277 pmap_qremove(up, UAREA_PAGES); 278 kmem_free(kernel_map, up, UAREA_PAGES * PAGE_SIZE); 279 p->p_upages_obj = NULL; 280 vm_object_deallocate(upobj); 281} 282 283#ifndef NO_SWAPPING 284/* 285 * Allow the U area for a process to be prejudicially paged out. 286 */ 287void 288vm_proc_swapout(struct proc *p) 289{ 290 vm_object_t upobj; 291 vm_offset_t up; 292 vm_page_t m; 293 int i; 294 295 upobj = p->p_upages_obj; 296 up = (vm_offset_t)p->p_uarea; 297 for (i = 0; i < UAREA_PAGES; i++) { 298 m = vm_page_lookup(upobj, i); 299 if (m == NULL) 300 panic("vm_proc_swapout: upage already missing?"); 301 vm_page_dirty(m); 302 vm_page_unwire(m, 0); 303 } 304 pmap_qremove(up, UAREA_PAGES); 305} 306 307/* 308 * Bring the U area for a specified process back in. 309 */ 310void 311vm_proc_swapin(struct proc *p) 312{ 313 vm_page_t ma[UAREA_PAGES]; 314 vm_object_t upobj; 315 vm_offset_t up; 316 vm_page_t m; 317 int rv; 318 int i; 319 320 upobj = p->p_upages_obj; 321 up = (vm_offset_t)p->p_uarea; 322 for (i = 0; i < UAREA_PAGES; i++) { 323 m = vm_page_grab(upobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 324 if (m->valid != VM_PAGE_BITS_ALL) { 325 rv = vm_pager_get_pages(upobj, &m, 1, 0); 326 if (rv != VM_PAGER_OK) 327 panic("vm_proc_swapin: cannot get upage"); 328 m = vm_page_lookup(upobj, i); 329 m->valid = VM_PAGE_BITS_ALL; 330 } 331 ma[i] = m; 332 vm_page_wire(m); 333 vm_page_wakeup(m); 334 vm_page_flag_set(m, PG_MAPPED | PG_WRITEABLE); 335 } 336 pmap_qenter(up, ma, UAREA_PAGES); 337} 338#endif 339 340/* 341 * Implement fork's actions on an address space. 342 * Here we arrange for the address space to be copied or referenced, 343 * allocate a user struct (pcb and kernel stack), then call the 344 * machine-dependent layer to fill those in and make the new process 345 * ready to run. The new process is set up so that it returns directly 346 * to user mode to avoid stack copying and relocation problems. 347 */ 348void 349vm_forkproc(td, p2, td2, flags) 350 struct thread *td; 351 struct proc *p2; 352 struct thread *td2; 353 int flags; 354{ 355 struct proc *p1 = td->td_proc; 356 struct user *up; 357 358 GIANT_REQUIRED; 359 360 if ((flags & RFPROC) == 0) { 361 /* 362 * Divorce the memory, if it is shared, essentially 363 * this changes shared memory amongst threads, into 364 * COW locally. 365 */ 366 if ((flags & RFMEM) == 0) { 367 if (p1->p_vmspace->vm_refcnt > 1) { 368 vmspace_unshare(p1); 369 } 370 } 371 cpu_fork(td, p2, td2, flags); 372 return; 373 } 374 375 if (flags & RFMEM) { 376 p2->p_vmspace = p1->p_vmspace; 377 p1->p_vmspace->vm_refcnt++; 378 } 379 380 while (vm_page_count_severe()) { 381 VM_WAIT; 382 } 383 384 if ((flags & RFMEM) == 0) { 385 p2->p_vmspace = vmspace_fork(p1->p_vmspace); 386 387 pmap_pinit2(vmspace_pmap(p2->p_vmspace)); 388 389 if (p1->p_vmspace->vm_shm) 390 shmfork(p1, p2); 391 } 392 393 /* XXXKSE this is unsatisfactory but should be adequate */ 394 up = p2->p_uarea; 395 396 /* 397 * p_stats currently points at fields in the user struct 398 * but not at &u, instead at p_addr. Copy parts of 399 * p_stats; zero the rest of p_stats (statistics). 400 * 401 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need 402 * to share sigacts, so we use the up->u_sigacts. 403 */ 404 p2->p_stats = &up->u_stats; 405 if (p2->p_sigacts == NULL) { 406 if (p2->p_procsig->ps_refcnt != 1) 407 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid); 408 p2->p_sigacts = &up->u_sigacts; 409 up->u_sigacts = *p1->p_sigacts; 410 } 411 412 bzero(&up->u_stats.pstat_startzero, 413 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero - 414 (caddr_t) &up->u_stats.pstat_startzero)); 415 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy, 416 ((caddr_t) &up->u_stats.pstat_endcopy - 417 (caddr_t) &up->u_stats.pstat_startcopy)); 418 419 420 /* 421 * cpu_fork will copy and update the pcb, set up the kernel stack, 422 * and make the child ready to run. 423 */ 424 cpu_fork(td, p2, td2, flags); 425} 426 427/* 428 * Called after process has been wait(2)'ed apon and is being reaped. 429 * The idea is to reclaim resources that we could not reclaim while 430 * the process was still executing. 431 */ 432void 433vm_waitproc(p) 434 struct proc *p; 435{ 436 struct thread *td; 437 438 GIANT_REQUIRED; 439 cpu_wait(p); 440/* XXXKSE by here there should not be any threads left! */ 441 FOREACH_THREAD_IN_PROC(p, td) { 442 panic("vm_waitproc: Survivor thread!"); 443 } 444 vmspace_exitfree(p); /* and clean-out the vmspace */ 445} 446 447/* 448 * Set default limits for VM system. 449 * Called for proc 0, and then inherited by all others. 450 * 451 * XXX should probably act directly on proc0. 452 */ 453static void 454vm_init_limits(udata) 455 void *udata; 456{ 457 struct proc *p = udata; 458 int rss_limit; 459 460 /* 461 * Set up the initial limits on process VM. Set the maximum resident 462 * set size to be half of (reasonably) available memory. Since this 463 * is a soft limit, it comes into effect only when the system is out 464 * of memory - half of main memory helps to favor smaller processes, 465 * and reduces thrashing of the object cache. 466 */ 467 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz; 468 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz; 469 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz; 470 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz; 471 /* limit the limit to no less than 2MB */ 472 rss_limit = max(cnt.v_free_count, 512); 473 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit); 474 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; 475} 476 477void 478faultin(p) 479 struct proc *p; 480{ 481 482 GIANT_REQUIRED; 483 PROC_LOCK_ASSERT(p, MA_OWNED); 484 mtx_lock_spin(&sched_lock); 485#ifdef NO_SWAPPING 486 if ((p->p_sflag & PS_INMEM) == 0) 487 panic("faultin: proc swapped out with NO_SWAPPING!"); 488#else 489 if ((p->p_sflag & PS_INMEM) == 0) { 490 struct thread *td; 491 492 ++p->p_lock; 493 mtx_unlock_spin(&sched_lock); 494 PROC_UNLOCK(p); 495 496 vm_proc_swapin(p); 497 FOREACH_THREAD_IN_PROC (p, td) 498 pmap_swapin_thread(td); 499 500 PROC_LOCK(p); 501 mtx_lock_spin(&sched_lock); 502 FOREACH_THREAD_IN_PROC (p, td) 503 if (td->td_state == TDS_RUNQ) /* XXXKSE */ 504 setrunqueue(td); 505 506 p->p_sflag |= PS_INMEM; 507 508 /* undo the effect of setting SLOCK above */ 509 --p->p_lock; 510 } 511#endif 512 mtx_unlock_spin(&sched_lock); 513} 514 515/* 516 * This swapin algorithm attempts to swap-in processes only if there 517 * is enough space for them. Of course, if a process waits for a long 518 * time, it will be swapped in anyway. 519 * 520 * XXXKSE - process with the thread with highest priority counts.. 521 * 522 * Giant is still held at this point, to be released in tsleep. 523 */ 524/* ARGSUSED*/ 525static void 526scheduler(dummy) 527 void *dummy; 528{ 529 struct proc *p; 530 struct thread *td; 531 int pri; 532 struct proc *pp; 533 int ppri; 534 535 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED); 536 /* GIANT_REQUIRED */ 537 538loop: 539 if (vm_page_count_min()) { 540 VM_WAIT; 541 goto loop; 542 } 543 544 pp = NULL; 545 ppri = INT_MIN; 546 sx_slock(&allproc_lock); 547 FOREACH_PROC_IN_SYSTEM(p) { 548 struct ksegrp *kg; 549 if (p->p_sflag & (PS_INMEM | PS_SWAPPING)) { 550 continue; 551 } 552 mtx_lock_spin(&sched_lock); 553 FOREACH_THREAD_IN_PROC(p, td) { 554 /* Only consider runnable threads */ 555 if (td->td_state == TDS_RUNQ) { 556 kg = td->td_ksegrp; 557 pri = p->p_swtime + kg->kg_slptime; 558 if ((p->p_sflag & PS_SWAPINREQ) == 0) { 559 pri -= kg->kg_nice * 8; 560 } 561 562 /* 563 * if this ksegrp is higher priority 564 * and there is enough space, then select 565 * this process instead of the previous 566 * selection. 567 */ 568 if (pri > ppri) { 569 pp = p; 570 ppri = pri; 571 } 572 } 573 } 574 mtx_unlock_spin(&sched_lock); 575 } 576 sx_sunlock(&allproc_lock); 577 578 /* 579 * Nothing to do, back to sleep. 580 */ 581 if ((p = pp) == NULL) { 582 tsleep(&proc0, PVM, "sched", maxslp * hz / 2); 583 goto loop; 584 } 585 mtx_lock_spin(&sched_lock); 586 p->p_sflag &= ~PS_SWAPINREQ; 587 mtx_unlock_spin(&sched_lock); 588 589 /* 590 * We would like to bring someone in. (only if there is space). 591 * [What checks the space? ] 592 */ 593 PROC_LOCK(p); 594 faultin(p); 595 PROC_UNLOCK(p); 596 mtx_lock_spin(&sched_lock); 597 p->p_swtime = 0; 598 mtx_unlock_spin(&sched_lock); 599 goto loop; 600} 601 602#ifndef NO_SWAPPING 603 604/* 605 * Swap_idle_threshold1 is the guaranteed swapped in time for a process 606 */ 607static int swap_idle_threshold1 = 2; 608SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, 609 CTLFLAG_RW, &swap_idle_threshold1, 0, ""); 610 611/* 612 * Swap_idle_threshold2 is the time that a process can be idle before 613 * it will be swapped out, if idle swapping is enabled. 614 */ 615static int swap_idle_threshold2 = 10; 616SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, 617 CTLFLAG_RW, &swap_idle_threshold2, 0, ""); 618 619/* 620 * Swapout is driven by the pageout daemon. Very simple, we find eligible 621 * procs and unwire their u-areas. We try to always "swap" at least one 622 * process in case we need the room for a swapin. 623 * If any procs have been sleeping/stopped for at least maxslp seconds, 624 * they are swapped. Else, we swap the longest-sleeping or stopped process, 625 * if any, otherwise the longest-resident process. 626 */ 627void 628swapout_procs(action) 629int action; 630{ 631 struct proc *p; 632 struct thread *td; 633 struct ksegrp *kg; 634 struct proc *outp, *outp2; 635 int outpri, outpri2; 636 int didswap = 0; 637 638 GIANT_REQUIRED; 639 640 outp = outp2 = NULL; 641 outpri = outpri2 = INT_MIN; 642retry: 643 sx_slock(&allproc_lock); 644 FOREACH_PROC_IN_SYSTEM(p) { 645 struct vmspace *vm; 646 int minslptime = 100000; 647 648 PROC_LOCK(p); 649 if (p->p_lock != 0 || 650 (p->p_flag & (P_STOPPED_SNGL|P_TRACED|P_SYSTEM|P_WEXIT)) != 0) { 651 PROC_UNLOCK(p); 652 continue; 653 } 654 /* 655 * only aiod changes vmspace, however it will be 656 * skipped because of the if statement above checking 657 * for P_SYSTEM 658 */ 659 vm = p->p_vmspace; 660 mtx_lock_spin(&sched_lock); 661 if ((p->p_sflag & (PS_INMEM|PS_SWAPPING)) != PS_INMEM) { 662 mtx_unlock_spin(&sched_lock); 663 PROC_UNLOCK(p); 664 continue; 665 } 666 667 switch (p->p_state) { 668 default: 669 /* Don't swap out processes in any sort 670 * of 'special' state. */ 671 mtx_unlock_spin(&sched_lock); 672 PROC_UNLOCK(p); 673 continue; 674 675 case PRS_NORMAL: 676 /* 677 * do not swapout a realtime process 678 * Check all the thread groups.. 679 */ 680 FOREACH_KSEGRP_IN_PROC(p, kg) { 681 if (PRI_IS_REALTIME(kg->kg_pri_class)) { 682 mtx_unlock_spin(&sched_lock); 683 PROC_UNLOCK(p); 684 goto nextproc; 685 } 686 687 /* 688 * Do not swapout a process waiting 689 * on a critical event of some kind. 690 * Also guarantee swap_idle_threshold1 691 * time in memory. 692 */ 693 if (kg->kg_slptime < swap_idle_threshold1) { 694 mtx_unlock_spin(&sched_lock); 695 PROC_UNLOCK(p); 696 goto nextproc; 697 } 698 FOREACH_THREAD_IN_PROC(p, td) { 699 if ((td->td_priority) < PSOCK) { 700 mtx_unlock_spin(&sched_lock); 701 PROC_UNLOCK(p); 702 goto nextproc; 703 } 704 } 705 /* 706 * If the system is under memory stress, 707 * or if we are swapping 708 * idle processes >= swap_idle_threshold2, 709 * then swap the process out. 710 */ 711 if (((action & VM_SWAP_NORMAL) == 0) && 712 (((action & VM_SWAP_IDLE) == 0) || 713 (kg->kg_slptime < swap_idle_threshold2))) { 714 mtx_unlock_spin(&sched_lock); 715 PROC_UNLOCK(p); 716 goto nextproc; 717 } 718 if (minslptime > kg->kg_slptime) 719 minslptime = kg->kg_slptime; 720 } 721 722 mtx_unlock_spin(&sched_lock); 723 ++vm->vm_refcnt; 724 /* 725 * do not swapout a process that 726 * is waiting for VM 727 * data structures there is a 728 * possible deadlock. 729 */ 730 if (!vm_map_trylock(&vm->vm_map)) { 731 vmspace_free(vm); 732 PROC_UNLOCK(p); 733 goto nextproc; 734 } 735 vm_map_unlock(&vm->vm_map); 736 /* 737 * If the process has been asleep for awhile and had 738 * most of its pages taken away already, swap it out. 739 */ 740 if ((action & VM_SWAP_NORMAL) || 741 ((action & VM_SWAP_IDLE) && 742 (minslptime > swap_idle_threshold2))) { 743 sx_sunlock(&allproc_lock); 744 swapout(p); 745 vmspace_free(vm); 746 didswap++; 747 goto retry; 748 } 749 PROC_UNLOCK(p); 750 vmspace_free(vm); 751 } 752nextproc: 753 continue; 754 } 755 sx_sunlock(&allproc_lock); 756 /* 757 * If we swapped something out, and another process needed memory, 758 * then wakeup the sched process. 759 */ 760 if (didswap) 761 wakeup(&proc0); 762} 763 764static void 765swapout(p) 766 struct proc *p; 767{ 768 struct thread *td; 769 770 PROC_LOCK_ASSERT(p, MA_OWNED); 771#if defined(SWAP_DEBUG) 772 printf("swapping out %d\n", p->p_pid); 773#endif 774 ++p->p_stats->p_ru.ru_nswap; 775 /* 776 * remember the process resident count 777 */ 778 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace); 779 780 mtx_lock_spin(&sched_lock); 781 p->p_sflag &= ~PS_INMEM; 782 p->p_sflag |= PS_SWAPPING; 783 PROC_UNLOCK(p); 784 FOREACH_THREAD_IN_PROC (p, td) 785 if (td->td_state == TDS_RUNQ) /* XXXKSE */ 786 remrunqueue(td); /* XXXKSE */ 787 mtx_unlock_spin(&sched_lock); 788 789 vm_proc_swapout(p); 790 FOREACH_THREAD_IN_PROC(p, td) 791 pmap_swapout_thread(td); 792 mtx_lock_spin(&sched_lock); 793 p->p_sflag &= ~PS_SWAPPING; 794 p->p_swtime = 0; 795 mtx_unlock_spin(&sched_lock); 796} 797#endif /* !NO_SWAPPING */ 798