2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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 * 4. 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.
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
32 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
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.
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.
48 * Carnegie Mellon requests users of this software to return to
50 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
51 * School of Computer Science
52 * Carnegie Mellon University
53 * Pittsburgh PA 15213-3890
55 * any improvements or extensions that they make and grant Carnegie the
56 * rights to redistribute these changes.
59 #include <sys/cdefs.h>
60 __FBSDID("$FreeBSD$");
63 #include "opt_kstack_pages.h"
64 #include "opt_kstack_max_pages.h"
66 #include <sys/param.h>
67 #include <sys/systm.h>
68 #include <sys/limits.h>
70 #include <sys/mutex.h>
72 #include <sys/resourcevar.h>
73 #include <sys/sched.h>
74 #include <sys/sf_buf.h>
76 #include <sys/vmmeter.h>
78 #include <sys/sysctl.h>
80 #include <sys/kernel.h>
82 #include <sys/unistd.h>
85 #include <vm/vm_param.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 #include <vm/swap_pager.h>
99 * System initialization
101 * Note: proc0 from proc.h
103 static void vm_init_limits(void *);
104 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0);
107 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
109 * Note: run scheduling should be divorced from the vm system.
111 static void scheduler(void *);
112 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL);
115 static int swapout(struct proc *);
116 static void swapclear(struct proc *);
122 * WARNING! This code calls vm_map_check_protection() which only checks
123 * the associated vm_map_entry range. It does not determine whether the
124 * contents of the memory is actually readable or writable. In most cases
125 * just checking the vm_map_entry is sufficient within the kernel's address
129 kernacc(addr, len, rw)
134 vm_offset_t saddr, eaddr;
137 KASSERT((rw & ~VM_PROT_ALL) == 0,
138 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
140 if ((vm_offset_t)addr + len > kernel_map->max_offset ||
141 (vm_offset_t)addr + len < (vm_offset_t)addr)
145 saddr = trunc_page((vm_offset_t)addr);
146 eaddr = round_page((vm_offset_t)addr + len);
147 vm_map_lock_read(kernel_map);
148 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
149 vm_map_unlock_read(kernel_map);
156 * WARNING! This code calls vm_map_check_protection() which only checks
157 * the associated vm_map_entry range. It does not determine whether the
158 * contents of the memory is actually readable or writable. vmapbuf(),
159 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be
160 * used in conjuction with this call.
163 useracc(addr, len, rw)
171 KASSERT((rw & ~VM_PROT_ALL) == 0,
172 ("illegal ``rw'' argument to useracc (%x)\n", rw));
174 map = &curproc->p_vmspace->vm_map;
175 if ((vm_offset_t)addr + len > vm_map_max(map) ||
176 (vm_offset_t)addr + len < (vm_offset_t)addr) {
179 vm_map_lock_read(map);
180 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr),
181 round_page((vm_offset_t)addr + len), prot);
182 vm_map_unlock_read(map);
187 vslock(void *addr, size_t len)
189 vm_offset_t end, last, start;
193 last = (vm_offset_t)addr + len;
194 start = trunc_page((vm_offset_t)addr);
195 end = round_page(last);
196 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr)
198 npages = atop(end - start);
199 if (npages > vm_page_max_wired)
203 pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map))) >
204 lim_cur(curproc, RLIMIT_MEMLOCK)) {
205 PROC_UNLOCK(curproc);
208 PROC_UNLOCK(curproc);
213 * The limit for transient usage of wired pages should be
214 * larger than for "permanent" wired pages (mlock()).
216 * Also, the sysctl code, which is the only present user
217 * of vslock(), does a hard loop on EAGAIN.
219 if (npages + cnt.v_wire_count > vm_page_max_wired)
222 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end,
223 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
225 * Return EFAULT on error to match copy{in,out}() behaviour
226 * rather than returning ENOMEM like mlock() would.
228 return (error == KERN_SUCCESS ? 0 : EFAULT);
232 vsunlock(void *addr, size_t len)
235 /* Rely on the parameter sanity checks performed by vslock(). */
236 (void)vm_map_unwire(&curproc->p_vmspace->vm_map,
237 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len),
238 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
242 * Pin the page contained within the given object at the given offset. If the
243 * page is not resident, allocate and load it using the given object's pager.
244 * Return the pinned page if successful; otherwise, return NULL.
247 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset)
253 VM_OBJECT_LOCK(object);
254 pindex = OFF_TO_IDX(offset);
255 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
256 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
258 rv = vm_pager_get_pages(object, ma, 1, 0);
259 m = vm_page_lookup(object, pindex);
262 if (m->valid == 0 || rv != VM_PAGER_OK) {
263 vm_page_lock_queues();
265 vm_page_unlock_queues();
270 vm_page_lock_queues();
272 vm_page_unlock_queues();
275 VM_OBJECT_UNLOCK(object);
280 * Return a CPU private mapping to the page at the given offset within the
281 * given object. The page is pinned before it is mapped.
284 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
288 m = vm_imgact_hold_page(object, offset);
292 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
296 * Destroy the given CPU private mapping and unpin the page that it mapped.
299 vm_imgact_unmap_page(struct sf_buf *sf)
306 vm_page_lock_queues();
308 vm_page_unlock_queues();
311 #ifndef KSTACK_MAX_PAGES
312 #define KSTACK_MAX_PAGES 32
316 * Create the kernel stack (including pcb for i386) for a new thread.
317 * This routine directly affects the fork perf for a process and
318 * create performance for a thread.
321 vm_thread_new(struct thread *td, int pages)
325 vm_page_t m, ma[KSTACK_MAX_PAGES];
330 pages = KSTACK_PAGES;
331 else if (pages > KSTACK_MAX_PAGES)
332 pages = KSTACK_MAX_PAGES;
334 * Allocate an object for the kstack.
336 ksobj = vm_object_allocate(OBJT_DEFAULT, pages);
338 * Get a kernel virtual address for this thread's kstack.
340 ks = kmem_alloc_nofault(kernel_map,
341 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
343 printf("vm_thread_new: kstack allocation failed\n");
344 vm_object_deallocate(ksobj);
348 if (KSTACK_GUARD_PAGES != 0) {
349 pmap_qremove(ks, KSTACK_GUARD_PAGES);
350 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
352 td->td_kstack_obj = ksobj;
355 * Knowing the number of pages allocated is useful when you
356 * want to deallocate them.
358 td->td_kstack_pages = pages;
360 * For the length of the stack, link in a real page of ram for each
363 VM_OBJECT_LOCK(ksobj);
364 for (i = 0; i < pages; i++) {
366 * Get a kernel stack page.
368 m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY |
369 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED);
371 m->valid = VM_PAGE_BITS_ALL;
373 VM_OBJECT_UNLOCK(ksobj);
374 pmap_qenter(ks, ma, pages);
379 * Dispose of a thread's kernel stack.
382 vm_thread_dispose(struct thread *td)
389 pages = td->td_kstack_pages;
390 ksobj = td->td_kstack_obj;
392 pmap_qremove(ks, pages);
393 VM_OBJECT_LOCK(ksobj);
394 for (i = 0; i < pages; i++) {
395 m = vm_page_lookup(ksobj, i);
397 panic("vm_thread_dispose: kstack already missing?");
398 vm_page_lock_queues();
399 vm_page_unwire(m, 0);
401 vm_page_unlock_queues();
403 VM_OBJECT_UNLOCK(ksobj);
404 vm_object_deallocate(ksobj);
405 kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
406 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
411 * Allow a thread's kernel stack to be paged out.
414 vm_thread_swapout(struct thread *td)
420 cpu_thread_swapout(td);
421 pages = td->td_kstack_pages;
422 ksobj = td->td_kstack_obj;
423 pmap_qremove(td->td_kstack, pages);
424 VM_OBJECT_LOCK(ksobj);
425 for (i = 0; i < pages; i++) {
426 m = vm_page_lookup(ksobj, i);
428 panic("vm_thread_swapout: kstack already missing?");
429 vm_page_lock_queues();
431 vm_page_unwire(m, 0);
432 vm_page_unlock_queues();
434 VM_OBJECT_UNLOCK(ksobj);
438 * Bring the kernel stack for a specified thread back in.
441 vm_thread_swapin(struct thread *td)
444 vm_page_t m, ma[KSTACK_MAX_PAGES];
447 pages = td->td_kstack_pages;
448 ksobj = td->td_kstack_obj;
449 VM_OBJECT_LOCK(ksobj);
450 for (i = 0; i < pages; i++) {
451 m = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
452 if (m->valid != VM_PAGE_BITS_ALL) {
453 rv = vm_pager_get_pages(ksobj, &m, 1, 0);
454 if (rv != VM_PAGER_OK)
455 panic("vm_thread_swapin: cannot get kstack for proc: %d", td->td_proc->p_pid);
456 m = vm_page_lookup(ksobj, i);
457 m->valid = VM_PAGE_BITS_ALL;
460 vm_page_lock_queues();
462 vm_page_unlock_queues();
465 VM_OBJECT_UNLOCK(ksobj);
466 pmap_qenter(td->td_kstack, ma, pages);
467 cpu_thread_swapin(td);
471 * Set up a variable-sized alternate kstack.
474 vm_thread_new_altkstack(struct thread *td, int pages)
477 td->td_altkstack = td->td_kstack;
478 td->td_altkstack_obj = td->td_kstack_obj;
479 td->td_altkstack_pages = td->td_kstack_pages;
481 return (vm_thread_new(td, pages));
485 * Restore the original kstack.
488 vm_thread_dispose_altkstack(struct thread *td)
491 vm_thread_dispose(td);
493 td->td_kstack = td->td_altkstack;
494 td->td_kstack_obj = td->td_altkstack_obj;
495 td->td_kstack_pages = td->td_altkstack_pages;
496 td->td_altkstack = 0;
497 td->td_altkstack_obj = NULL;
498 td->td_altkstack_pages = 0;
502 * Implement fork's actions on an address space.
503 * Here we arrange for the address space to be copied or referenced,
504 * allocate a user struct (pcb and kernel stack), then call the
505 * machine-dependent layer to fill those in and make the new process
506 * ready to run. The new process is set up so that it returns directly
507 * to user mode to avoid stack copying and relocation problems.
510 vm_forkproc(td, p2, td2, vm2, flags)
517 struct proc *p1 = td->td_proc;
520 if ((flags & RFPROC) == 0) {
522 * Divorce the memory, if it is shared, essentially
523 * this changes shared memory amongst threads, into
526 if ((flags & RFMEM) == 0) {
527 if (p1->p_vmspace->vm_refcnt > 1) {
528 error = vmspace_unshare(p1);
533 cpu_fork(td, p2, td2, flags);
538 p2->p_vmspace = p1->p_vmspace;
539 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1);
542 while (vm_page_count_severe()) {
546 if ((flags & RFMEM) == 0) {
548 if (p1->p_vmspace->vm_shm)
553 * cpu_fork will copy and update the pcb, set up the kernel stack,
554 * and make the child ready to run.
556 cpu_fork(td, p2, td2, flags);
561 * Called after process has been wait(2)'ed apon and is being reaped.
562 * The idea is to reclaim resources that we could not reclaim while
563 * the process was still executing.
570 vmspace_exitfree(p); /* and clean-out the vmspace */
574 * Set default limits for VM system.
575 * Called for proc 0, and then inherited by all others.
577 * XXX should probably act directly on proc0.
580 vm_init_limits(udata)
583 struct proc *p = udata;
588 * Set up the initial limits on process VM. Set the maximum resident
589 * set size to be half of (reasonably) available memory. Since this
590 * is a soft limit, it comes into effect only when the system is out
591 * of memory - half of main memory helps to favor smaller processes,
592 * and reduces thrashing of the object cache.
595 limp->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
596 limp->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
597 limp->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
598 limp->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
599 /* limit the limit to no less than 2MB */
600 rss_limit = max(cnt.v_free_count, 512);
601 limp->pl_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
602 limp->pl_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
611 PROC_LOCK_ASSERT(p, MA_OWNED);
612 if ((p->p_flag & P_INMEM) == 0)
613 panic("faultin: proc swapped out with NO_SWAPPING!");
614 #else /* !NO_SWAPPING */
617 PROC_LOCK_ASSERT(p, MA_OWNED);
619 * If another process is swapping in this process,
620 * just wait until it finishes.
622 if (p->p_flag & P_SWAPPINGIN) {
623 while (p->p_flag & P_SWAPPINGIN)
624 msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0);
627 if ((p->p_flag & P_INMEM) == 0) {
629 * Don't let another thread swap process p out while we are
630 * busy swapping it in.
633 p->p_flag |= P_SWAPPINGIN;
637 * We hold no lock here because the list of threads
638 * can not change while all threads in the process are
641 FOREACH_THREAD_IN_PROC(p, td)
642 vm_thread_swapin(td);
651 /* Allow other threads to swap p out now. */
654 #endif /* NO_SWAPPING */
658 * This swapin algorithm attempts to swap-in processes only if there
659 * is enough space for them. Of course, if a process waits for a long
660 * time, it will be swapped in anyway.
662 * XXXKSE - process with the thread with highest priority counts..
664 * Giant is held on entry.
679 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
683 if (vm_page_count_min()) {
690 sx_slock(&allproc_lock);
691 FOREACH_PROC_IN_SYSTEM(p) {
693 if (p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) {
697 swtime = (ticks - p->p_swtick) / hz;
699 FOREACH_THREAD_IN_PROC(p, td) {
701 * An otherwise runnable thread of a process
702 * swapped out has only the TDI_SWAPPED bit set.
706 if (td->td_inhibitors == TDI_SWAPPED) {
707 slptime = (ticks - td->td_slptick) / hz;
708 pri = swtime + slptime;
709 if ((td->td_flags & TDF_SWAPINREQ) == 0)
710 pri -= p->p_nice * 8;
712 * if this thread is higher priority
713 * and there is enough space, then select
714 * this process instead of the previous
727 sx_sunlock(&allproc_lock);
730 * Nothing to do, back to sleep.
732 if ((p = pp) == NULL) {
733 tsleep(&proc0, PVM, "sched", maxslp * hz / 2);
739 * Another process may be bringing or may have already
740 * brought this process in while we traverse all threads.
741 * Or, this process may even be being swapped out again.
743 if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) {
749 * We would like to bring someone in. (only if there is space).
750 * [What checks the space? ]
767 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
769 static int swap_idle_threshold1 = 2;
770 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW,
771 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process");
774 * Swap_idle_threshold2 is the time that a process can be idle before
775 * it will be swapped out, if idle swapping is enabled.
777 static int swap_idle_threshold2 = 10;
778 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW,
779 &swap_idle_threshold2, 0, "Time before a process will be swapped out");
782 * Swapout is driven by the pageout daemon. Very simple, we find eligible
783 * procs and swap out their stacks. We try to always "swap" at least one
784 * process in case we need the room for a swapin.
785 * If any procs have been sleeping/stopped for at least maxslp seconds,
786 * they are swapped. Else, we swap the longest-sleeping or stopped process,
787 * if any, otherwise the longest-resident process.
790 swapout_procs(action)
798 sx_slock(&allproc_lock);
799 FOREACH_PROC_IN_SYSTEM(p) {
801 int minslptime = 100000;
805 * Watch out for a process in
806 * creation. It may have no
807 * address space or lock yet.
809 if (p->p_state == PRS_NEW)
812 * An aio daemon switches its
813 * address space while running.
814 * Perform a quick check whether
815 * a process has P_SYSTEM.
817 if ((p->p_flag & P_SYSTEM) != 0)
820 * Do not swapout a process that
821 * is waiting for VM data
822 * structures as there is a possible
823 * deadlock. Test this first as
826 * Lock the map until swapout
827 * finishes, or a thread of this
828 * process may attempt to alter
831 vm = vmspace_acquire_ref(p);
834 if (!vm_map_trylock(&vm->vm_map))
838 if (p->p_lock != 0 ||
839 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT)
844 * only aiod changes vmspace, however it will be
845 * skipped because of the if statement above checking
848 if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM)
851 switch (p->p_state) {
853 /* Don't swap out processes in any sort
854 * of 'special' state. */
860 * do not swapout a realtime process
861 * Check all the thread groups..
863 FOREACH_THREAD_IN_PROC(p, td) {
865 if (PRI_IS_REALTIME(td->td_pri_class)) {
869 slptime = (ticks - td->td_slptick) / hz;
871 * Guarantee swap_idle_threshold1
874 if (slptime < swap_idle_threshold1) {
880 * Do not swapout a process if it is
881 * waiting on a critical event of some
882 * kind or there is a thread whose
883 * pageable memory may be accessed.
885 * This could be refined to support
886 * swapping out a thread.
888 if ((td->td_priority) < PSOCK ||
889 !thread_safetoswapout(td)) {
894 * If the system is under memory stress,
895 * or if we are swapping
896 * idle processes >= swap_idle_threshold2,
897 * then swap the process out.
899 if (((action & VM_SWAP_NORMAL) == 0) &&
900 (((action & VM_SWAP_IDLE) == 0) ||
901 (slptime < swap_idle_threshold2))) {
906 if (minslptime > slptime)
907 minslptime = slptime;
912 * If the pageout daemon didn't free enough pages,
913 * or if this process is idle and the system is
914 * configured to swap proactively, swap it out.
916 if ((action & VM_SWAP_NORMAL) ||
917 ((action & VM_SWAP_IDLE) &&
918 (minslptime > swap_idle_threshold2))) {
923 vm_map_unlock(&vm->vm_map);
925 sx_sunlock(&allproc_lock);
933 vm_map_unlock(&vm->vm_map);
938 sx_sunlock(&allproc_lock);
940 * If we swapped something out, and another process needed memory,
941 * then wakeup the sched process.
953 PROC_LOCK_ASSERT(p, MA_OWNED);
954 PROC_SLOCK_ASSERT(p, MA_OWNED);
956 FOREACH_THREAD_IN_PROC(p, td) {
958 td->td_flags |= TDF_INMEM;
959 td->td_flags &= ~TDF_SWAPINREQ;
962 if (setrunnable(td)) {
965 * XXX: We just cleared TDI_SWAPPED
966 * above and set TDF_INMEM, so this
967 * should never happen.
969 panic("not waking up swapper");
974 p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT);
975 p->p_flag |= P_INMEM;
984 PROC_LOCK_ASSERT(p, MA_OWNED);
985 PROC_SLOCK_ASSERT(p, MA_OWNED | MA_NOTRECURSED);
986 #if defined(SWAP_DEBUG)
987 printf("swapping out %d\n", p->p_pid);
991 * The states of this process and its threads may have changed
992 * by now. Assuming that there is only one pageout daemon thread,
993 * this process should still be in memory.
995 KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM,
996 ("swapout: lost a swapout race?"));
999 * remember the process resident count
1001 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
1003 * Check and mark all threads before we proceed.
1005 p->p_flag &= ~P_INMEM;
1006 p->p_flag |= P_SWAPPINGOUT;
1007 FOREACH_THREAD_IN_PROC(p, td) {
1009 if (!thread_safetoswapout(td)) {
1014 td->td_flags &= ~TDF_INMEM;
1018 td = FIRST_THREAD_IN_PROC(p);
1019 ++td->td_ru.ru_nswap;
1024 * This list is stable because all threads are now prevented from
1025 * running. The list is only modified in the context of a running
1026 * thread in this process.
1028 FOREACH_THREAD_IN_PROC(p, td)
1029 vm_thread_swapout(td);
1032 p->p_flag &= ~P_SWAPPINGOUT;
1034 p->p_swtick = ticks;
1037 #endif /* !NO_SWAPPING */