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/racct.h>
73 #include <sys/resourcevar.h>
74 #include <sys/rwlock.h>
75 #include <sys/sched.h>
76 #include <sys/sf_buf.h>
78 #include <sys/vmmeter.h>
80 #include <sys/sysctl.h>
81 #include <sys/_kstack_cache.h>
82 #include <sys/eventhandler.h>
83 #include <sys/kernel.h>
85 #include <sys/unistd.h>
88 #include <vm/vm_param.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_pageout.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_extern.h>
96 #include <vm/vm_pager.h>
97 #include <vm/swap_pager.h>
100 * System initialization
102 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
104 * Note: run scheduling should be divorced from the vm system.
106 static void scheduler(void *);
107 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL);
110 static int swapout(struct proc *);
111 static void swapclear(struct proc *);
112 static void vm_thread_swapin(struct thread *td);
113 static void vm_thread_swapout(struct thread *td);
119 * WARNING! This code calls vm_map_check_protection() which only checks
120 * the associated vm_map_entry range. It does not determine whether the
121 * contents of the memory is actually readable or writable. In most cases
122 * just checking the vm_map_entry is sufficient within the kernel's address
126 kernacc(addr, len, rw)
131 vm_offset_t saddr, eaddr;
134 KASSERT((rw & ~VM_PROT_ALL) == 0,
135 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
137 if ((vm_offset_t)addr + len > kernel_map->max_offset ||
138 (vm_offset_t)addr + len < (vm_offset_t)addr)
142 saddr = trunc_page((vm_offset_t)addr);
143 eaddr = round_page((vm_offset_t)addr + len);
144 vm_map_lock_read(kernel_map);
145 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
146 vm_map_unlock_read(kernel_map);
153 * WARNING! This code calls vm_map_check_protection() which only checks
154 * the associated vm_map_entry range. It does not determine whether the
155 * contents of the memory is actually readable or writable. vmapbuf(),
156 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be
157 * used in conjuction with this call.
160 useracc(addr, len, rw)
168 KASSERT((rw & ~VM_PROT_ALL) == 0,
169 ("illegal ``rw'' argument to useracc (%x)\n", rw));
171 map = &curproc->p_vmspace->vm_map;
172 if ((vm_offset_t)addr + len > vm_map_max(map) ||
173 (vm_offset_t)addr + len < (vm_offset_t)addr) {
176 vm_map_lock_read(map);
177 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr),
178 round_page((vm_offset_t)addr + len), prot);
179 vm_map_unlock_read(map);
184 vslock(void *addr, size_t len)
186 vm_offset_t end, last, start;
190 last = (vm_offset_t)addr + len;
191 start = trunc_page((vm_offset_t)addr);
192 end = round_page(last);
193 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr)
195 npages = atop(end - start);
196 if (npages > vm_page_max_wired)
202 * The limit for transient usage of wired pages should be
203 * larger than for "permanent" wired pages (mlock()).
205 * Also, the sysctl code, which is the only present user
206 * of vslock(), does a hard loop on EAGAIN.
208 if (npages + cnt.v_wire_count > vm_page_max_wired)
211 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end,
212 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
214 * Return EFAULT on error to match copy{in,out}() behaviour
215 * rather than returning ENOMEM like mlock() would.
217 return (error == KERN_SUCCESS ? 0 : EFAULT);
221 vsunlock(void *addr, size_t len)
224 /* Rely on the parameter sanity checks performed by vslock(). */
225 (void)vm_map_unwire(&curproc->p_vmspace->vm_map,
226 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len),
227 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
231 * Pin the page contained within the given object at the given offset. If the
232 * page is not resident, allocate and load it using the given object's pager.
233 * Return the pinned page if successful; otherwise, return NULL.
236 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset)
242 VM_OBJECT_WLOCK(object);
243 pindex = OFF_TO_IDX(offset);
244 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
246 if (m->valid != VM_PAGE_BITS_ALL) {
249 rv = vm_pager_get_pages(object, ma, 1, 0);
250 m = vm_page_lookup(object, pindex);
253 if (rv != VM_PAGER_OK) {
266 VM_OBJECT_WUNLOCK(object);
271 * Return a CPU private mapping to the page at the given offset within the
272 * given object. The page is pinned before it is mapped.
275 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
279 m = vm_imgact_hold_page(object, offset);
283 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
287 * Destroy the given CPU private mapping and unpin the page that it mapped.
290 vm_imgact_unmap_page(struct sf_buf *sf)
303 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz)
306 pmap_sync_icache(map->pmap, va, sz);
309 struct kstack_cache_entry *kstack_cache;
310 static int kstack_cache_size = 128;
312 static struct mtx kstack_cache_mtx;
313 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF);
315 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0,
317 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0,
320 #ifndef KSTACK_MAX_PAGES
321 #define KSTACK_MAX_PAGES 32
325 * Create the kernel stack (including pcb for i386) for a new thread.
326 * This routine directly affects the fork perf for a process and
327 * create performance for a thread.
330 vm_thread_new(struct thread *td, int pages)
334 vm_page_t m, ma[KSTACK_MAX_PAGES];
335 struct kstack_cache_entry *ks_ce;
340 pages = KSTACK_PAGES;
341 else if (pages > KSTACK_MAX_PAGES)
342 pages = KSTACK_MAX_PAGES;
344 if (pages == KSTACK_PAGES) {
345 mtx_lock(&kstack_cache_mtx);
346 if (kstack_cache != NULL) {
347 ks_ce = kstack_cache;
348 kstack_cache = ks_ce->next_ks_entry;
349 mtx_unlock(&kstack_cache_mtx);
351 td->td_kstack_obj = ks_ce->ksobj;
352 td->td_kstack = (vm_offset_t)ks_ce;
353 td->td_kstack_pages = KSTACK_PAGES;
356 mtx_unlock(&kstack_cache_mtx);
360 * Allocate an object for the kstack.
362 ksobj = vm_object_allocate(OBJT_DEFAULT, pages);
365 * Get a kernel virtual address for this thread's kstack.
367 #if defined(__mips__)
369 * We need to align the kstack's mapped address to fit within
370 * a single TLB entry.
372 ks = kmem_alloc_nofault_space(kernel_map,
373 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, VMFS_TLB_ALIGNED_SPACE);
375 ks = kmem_alloc_nofault(kernel_map,
376 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
379 printf("vm_thread_new: kstack allocation failed\n");
380 vm_object_deallocate(ksobj);
384 atomic_add_int(&kstacks, 1);
385 if (KSTACK_GUARD_PAGES != 0) {
386 pmap_qremove(ks, KSTACK_GUARD_PAGES);
387 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
389 td->td_kstack_obj = ksobj;
392 * Knowing the number of pages allocated is useful when you
393 * want to deallocate them.
395 td->td_kstack_pages = pages;
397 * For the length of the stack, link in a real page of ram for each
400 VM_OBJECT_WLOCK(ksobj);
401 for (i = 0; i < pages; i++) {
403 * Get a kernel stack page.
405 m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY |
406 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED);
408 m->valid = VM_PAGE_BITS_ALL;
410 VM_OBJECT_WUNLOCK(ksobj);
411 pmap_qenter(ks, ma, pages);
416 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages)
421 atomic_add_int(&kstacks, -1);
422 pmap_qremove(ks, pages);
423 VM_OBJECT_WLOCK(ksobj);
424 for (i = 0; i < pages; i++) {
425 m = vm_page_lookup(ksobj, i);
427 panic("vm_thread_dispose: kstack already missing?");
429 vm_page_unwire(m, 0);
433 VM_OBJECT_WUNLOCK(ksobj);
434 vm_object_deallocate(ksobj);
435 kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
436 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
440 * Dispose of a thread's kernel stack.
443 vm_thread_dispose(struct thread *td)
447 struct kstack_cache_entry *ks_ce;
450 pages = td->td_kstack_pages;
451 ksobj = td->td_kstack_obj;
454 td->td_kstack_pages = 0;
455 if (pages == KSTACK_PAGES && kstacks <= kstack_cache_size) {
456 ks_ce = (struct kstack_cache_entry *)ks;
457 ks_ce->ksobj = ksobj;
458 mtx_lock(&kstack_cache_mtx);
459 ks_ce->next_ks_entry = kstack_cache;
460 kstack_cache = ks_ce;
461 mtx_unlock(&kstack_cache_mtx);
464 vm_thread_stack_dispose(ksobj, ks, pages);
468 vm_thread_stack_lowmem(void *nulll)
470 struct kstack_cache_entry *ks_ce, *ks_ce1;
472 mtx_lock(&kstack_cache_mtx);
473 ks_ce = kstack_cache;
475 mtx_unlock(&kstack_cache_mtx);
477 while (ks_ce != NULL) {
479 ks_ce = ks_ce->next_ks_entry;
481 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1,
487 kstack_cache_init(void *nulll)
490 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL,
491 EVENTHANDLER_PRI_ANY);
494 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL);
498 * Allow a thread's kernel stack to be paged out.
501 vm_thread_swapout(struct thread *td)
507 cpu_thread_swapout(td);
508 pages = td->td_kstack_pages;
509 ksobj = td->td_kstack_obj;
510 pmap_qremove(td->td_kstack, pages);
511 VM_OBJECT_WLOCK(ksobj);
512 for (i = 0; i < pages; i++) {
513 m = vm_page_lookup(ksobj, i);
515 panic("vm_thread_swapout: kstack already missing?");
518 vm_page_unwire(m, 0);
521 VM_OBJECT_WUNLOCK(ksobj);
525 * Bring the kernel stack for a specified thread back in.
528 vm_thread_swapin(struct thread *td)
531 vm_page_t ma[KSTACK_MAX_PAGES];
532 int i, j, k, pages, rv;
534 pages = td->td_kstack_pages;
535 ksobj = td->td_kstack_obj;
536 VM_OBJECT_WLOCK(ksobj);
537 for (i = 0; i < pages; i++)
538 ma[i] = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
540 for (i = 0; i < pages; i++) {
541 if (ma[i]->valid != VM_PAGE_BITS_ALL) {
542 KASSERT(ma[i]->oflags & VPO_BUSY,
544 vm_object_pip_add(ksobj, 1);
545 for (j = i + 1; j < pages; j++) {
546 KASSERT(ma[j]->valid == VM_PAGE_BITS_ALL ||
547 (ma[j]->oflags & VPO_BUSY),
549 if (ma[j]->valid == VM_PAGE_BITS_ALL)
552 rv = vm_pager_get_pages(ksobj, ma + i, j - i, 0);
553 if (rv != VM_PAGER_OK)
554 panic("vm_thread_swapin: cannot get kstack for proc: %d",
556 vm_object_pip_wakeup(ksobj);
557 for (k = i; k < j; k++)
558 ma[k] = vm_page_lookup(ksobj, k);
559 vm_page_wakeup(ma[i]);
560 } else if (ma[i]->oflags & VPO_BUSY)
561 vm_page_wakeup(ma[i]);
563 VM_OBJECT_WUNLOCK(ksobj);
564 pmap_qenter(td->td_kstack, ma, pages);
565 cpu_thread_swapin(td);
567 #endif /* !NO_SWAPPING */
570 * Implement fork's actions on an address space.
571 * Here we arrange for the address space to be copied or referenced,
572 * allocate a user struct (pcb and kernel stack), then call the
573 * machine-dependent layer to fill those in and make the new process
574 * ready to run. The new process is set up so that it returns directly
575 * to user mode to avoid stack copying and relocation problems.
578 vm_forkproc(td, p2, td2, vm2, flags)
585 struct proc *p1 = td->td_proc;
588 if ((flags & RFPROC) == 0) {
590 * Divorce the memory, if it is shared, essentially
591 * this changes shared memory amongst threads, into
594 if ((flags & RFMEM) == 0) {
595 if (p1->p_vmspace->vm_refcnt > 1) {
596 error = vmspace_unshare(p1);
601 cpu_fork(td, p2, td2, flags);
606 p2->p_vmspace = p1->p_vmspace;
607 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1);
610 while (vm_page_count_severe()) {
614 if ((flags & RFMEM) == 0) {
616 if (p1->p_vmspace->vm_shm)
621 * cpu_fork will copy and update the pcb, set up the kernel stack,
622 * and make the child ready to run.
624 cpu_fork(td, p2, td2, flags);
629 * Called after process has been wait(2)'ed apon and is being reaped.
630 * The idea is to reclaim resources that we could not reclaim while
631 * the process was still executing.
638 vmspace_exitfree(p); /* and clean-out the vmspace */
647 PROC_LOCK_ASSERT(p, MA_OWNED);
648 if ((p->p_flag & P_INMEM) == 0)
649 panic("faultin: proc swapped out with NO_SWAPPING!");
650 #else /* !NO_SWAPPING */
653 PROC_LOCK_ASSERT(p, MA_OWNED);
655 * If another process is swapping in this process,
656 * just wait until it finishes.
658 if (p->p_flag & P_SWAPPINGIN) {
659 while (p->p_flag & P_SWAPPINGIN)
660 msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0);
663 if ((p->p_flag & P_INMEM) == 0) {
665 * Don't let another thread swap process p out while we are
666 * busy swapping it in.
669 p->p_flag |= P_SWAPPINGIN;
673 * We hold no lock here because the list of threads
674 * can not change while all threads in the process are
677 FOREACH_THREAD_IN_PROC(p, td)
678 vm_thread_swapin(td);
685 /* Allow other threads to swap p out now. */
688 #endif /* NO_SWAPPING */
692 * This swapin algorithm attempts to swap-in processes only if there
693 * is enough space for them. Of course, if a process waits for a long
694 * time, it will be swapped in anyway.
696 * Giant is held on entry.
711 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
715 if (vm_page_count_min()) {
722 sx_slock(&allproc_lock);
723 FOREACH_PROC_IN_SYSTEM(p) {
725 if (p->p_state == PRS_NEW ||
726 p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) {
730 swtime = (ticks - p->p_swtick) / hz;
731 FOREACH_THREAD_IN_PROC(p, td) {
733 * An otherwise runnable thread of a process
734 * swapped out has only the TDI_SWAPPED bit set.
738 if (td->td_inhibitors == TDI_SWAPPED) {
739 slptime = (ticks - td->td_slptick) / hz;
740 pri = swtime + slptime;
741 if ((td->td_flags & TDF_SWAPINREQ) == 0)
742 pri -= p->p_nice * 8;
744 * if this thread is higher priority
745 * and there is enough space, then select
746 * this process instead of the previous
758 sx_sunlock(&allproc_lock);
761 * Nothing to do, back to sleep.
763 if ((p = pp) == NULL) {
764 tsleep(&proc0, PVM, "sched", MAXSLP * hz / 2);
770 * Another process may be bringing or may have already
771 * brought this process in while we traverse all threads.
772 * Or, this process may even be being swapped out again.
774 if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) {
780 * We would like to bring someone in. (only if there is space).
781 * [What checks the space? ]
798 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
800 static int swap_idle_threshold1 = 2;
801 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW,
802 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process");
805 * Swap_idle_threshold2 is the time that a process can be idle before
806 * it will be swapped out, if idle swapping is enabled.
808 static int swap_idle_threshold2 = 10;
809 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW,
810 &swap_idle_threshold2, 0, "Time before a process will be swapped out");
813 * First, if any processes have been sleeping or stopped for at least
814 * "swap_idle_threshold1" seconds, they are swapped out. If, however,
815 * no such processes exist, then the longest-sleeping or stopped
816 * process is swapped out. Finally, and only as a last resort, if
817 * there are no sleeping or stopped processes, the longest-resident
818 * process is swapped out.
821 swapout_procs(action)
829 sx_slock(&allproc_lock);
830 FOREACH_PROC_IN_SYSTEM(p) {
832 int minslptime = 100000;
836 * Watch out for a process in
837 * creation. It may have no
838 * address space or lock yet.
840 if (p->p_state == PRS_NEW)
843 * An aio daemon switches its
844 * address space while running.
845 * Perform a quick check whether
846 * a process has P_SYSTEM.
848 if ((p->p_flag & P_SYSTEM) != 0)
851 * Do not swapout a process that
852 * is waiting for VM data
853 * structures as there is a possible
854 * deadlock. Test this first as
857 * Lock the map until swapout
858 * finishes, or a thread of this
859 * process may attempt to alter
862 vm = vmspace_acquire_ref(p);
865 if (!vm_map_trylock(&vm->vm_map))
869 if (p->p_lock != 0 ||
870 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT)
875 * only aiod changes vmspace, however it will be
876 * skipped because of the if statement above checking
879 if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM)
882 switch (p->p_state) {
884 /* Don't swap out processes in any sort
885 * of 'special' state. */
890 * do not swapout a realtime process
891 * Check all the thread groups..
893 FOREACH_THREAD_IN_PROC(p, td) {
895 if (PRI_IS_REALTIME(td->td_pri_class)) {
899 slptime = (ticks - td->td_slptick) / hz;
901 * Guarantee swap_idle_threshold1
904 if (slptime < swap_idle_threshold1) {
910 * Do not swapout a process if it is
911 * waiting on a critical event of some
912 * kind or there is a thread whose
913 * pageable memory may be accessed.
915 * This could be refined to support
916 * swapping out a thread.
918 if (!thread_safetoswapout(td)) {
923 * If the system is under memory stress,
924 * or if we are swapping
925 * idle processes >= swap_idle_threshold2,
926 * then swap the process out.
928 if (((action & VM_SWAP_NORMAL) == 0) &&
929 (((action & VM_SWAP_IDLE) == 0) ||
930 (slptime < swap_idle_threshold2))) {
935 if (minslptime > slptime)
936 minslptime = slptime;
941 * If the pageout daemon didn't free enough pages,
942 * or if this process is idle and the system is
943 * configured to swap proactively, swap it out.
945 if ((action & VM_SWAP_NORMAL) ||
946 ((action & VM_SWAP_IDLE) &&
947 (minslptime > swap_idle_threshold2))) {
951 vm_map_unlock(&vm->vm_map);
953 sx_sunlock(&allproc_lock);
959 vm_map_unlock(&vm->vm_map);
964 sx_sunlock(&allproc_lock);
966 * If we swapped something out, and another process needed memory,
967 * then wakeup the sched process.
979 PROC_LOCK_ASSERT(p, MA_OWNED);
981 FOREACH_THREAD_IN_PROC(p, td) {
983 td->td_flags |= TDF_INMEM;
984 td->td_flags &= ~TDF_SWAPINREQ;
987 if (setrunnable(td)) {
990 * XXX: We just cleared TDI_SWAPPED
991 * above and set TDF_INMEM, so this
992 * should never happen.
994 panic("not waking up swapper");
999 p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT);
1000 p->p_flag |= P_INMEM;
1009 PROC_LOCK_ASSERT(p, MA_OWNED);
1010 #if defined(SWAP_DEBUG)
1011 printf("swapping out %d\n", p->p_pid);
1015 * The states of this process and its threads may have changed
1016 * by now. Assuming that there is only one pageout daemon thread,
1017 * this process should still be in memory.
1019 KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM,
1020 ("swapout: lost a swapout race?"));
1023 * remember the process resident count
1025 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
1027 * Check and mark all threads before we proceed.
1029 p->p_flag &= ~P_INMEM;
1030 p->p_flag |= P_SWAPPINGOUT;
1031 FOREACH_THREAD_IN_PROC(p, td) {
1033 if (!thread_safetoswapout(td)) {
1038 td->td_flags &= ~TDF_INMEM;
1042 td = FIRST_THREAD_IN_PROC(p);
1043 ++td->td_ru.ru_nswap;
1047 * This list is stable because all threads are now prevented from
1048 * running. The list is only modified in the context of a running
1049 * thread in this process.
1051 FOREACH_THREAD_IN_PROC(p, td)
1052 vm_thread_swapout(td);
1055 p->p_flag &= ~P_SWAPPINGOUT;
1056 p->p_swtick = ticks;
1059 #endif /* !NO_SWAPPING */