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/malloc.h>
71 #include <sys/mutex.h>
73 #include <sys/racct.h>
74 #include <sys/resourcevar.h>
75 #include <sys/rwlock.h>
76 #include <sys/sched.h>
77 #include <sys/sf_buf.h>
79 #include <sys/vmmeter.h>
82 #include <sys/sysctl.h>
83 #include <sys/_kstack_cache.h>
84 #include <sys/eventhandler.h>
85 #include <sys/kernel.h>
87 #include <sys/unistd.h>
90 #include <vm/vm_param.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_kern.h>
97 #include <vm/vm_extern.h>
98 #include <vm/vm_pager.h>
99 #include <vm/swap_pager.h>
102 static int swapout(struct proc *);
103 static void swapclear(struct proc *);
104 static void vm_thread_swapin(struct thread *td);
105 static void vm_thread_swapout(struct thread *td);
111 * WARNING! This code calls vm_map_check_protection() which only checks
112 * the associated vm_map_entry range. It does not determine whether the
113 * contents of the memory is actually readable or writable. In most cases
114 * just checking the vm_map_entry is sufficient within the kernel's address
118 kernacc(addr, len, rw)
123 vm_offset_t saddr, eaddr;
126 KASSERT((rw & ~VM_PROT_ALL) == 0,
127 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
129 if ((vm_offset_t)addr + len > kernel_map->max_offset ||
130 (vm_offset_t)addr + len < (vm_offset_t)addr)
134 saddr = trunc_page((vm_offset_t)addr);
135 eaddr = round_page((vm_offset_t)addr + len);
136 vm_map_lock_read(kernel_map);
137 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
138 vm_map_unlock_read(kernel_map);
145 * WARNING! This code calls vm_map_check_protection() which only checks
146 * the associated vm_map_entry range. It does not determine whether the
147 * contents of the memory is actually readable or writable. vmapbuf(),
148 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be
149 * used in conjuction with this call.
152 useracc(addr, len, rw)
160 KASSERT((rw & ~VM_PROT_ALL) == 0,
161 ("illegal ``rw'' argument to useracc (%x)\n", rw));
163 map = &curproc->p_vmspace->vm_map;
164 if ((vm_offset_t)addr + len > vm_map_max(map) ||
165 (vm_offset_t)addr + len < (vm_offset_t)addr) {
168 vm_map_lock_read(map);
169 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr),
170 round_page((vm_offset_t)addr + len), prot);
171 vm_map_unlock_read(map);
176 vslock(void *addr, size_t len)
178 vm_offset_t end, last, start;
182 last = (vm_offset_t)addr + len;
183 start = trunc_page((vm_offset_t)addr);
184 end = round_page(last);
185 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr)
187 npages = atop(end - start);
188 if (npages > vm_page_max_wired)
194 * The limit for transient usage of wired pages should be
195 * larger than for "permanent" wired pages (mlock()).
197 * Also, the sysctl code, which is the only present user
198 * of vslock(), does a hard loop on EAGAIN.
200 if (npages + cnt.v_wire_count > vm_page_max_wired)
203 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end,
204 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
206 * Return EFAULT on error to match copy{in,out}() behaviour
207 * rather than returning ENOMEM like mlock() would.
209 return (error == KERN_SUCCESS ? 0 : EFAULT);
213 vsunlock(void *addr, size_t len)
216 /* Rely on the parameter sanity checks performed by vslock(). */
217 (void)vm_map_unwire(&curproc->p_vmspace->vm_map,
218 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len),
219 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
223 * Pin the page contained within the given object at the given offset. If the
224 * page is not resident, allocate and load it using the given object's pager.
225 * Return the pinned page if successful; otherwise, return NULL.
228 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset)
234 VM_OBJECT_WLOCK(object);
235 pindex = OFF_TO_IDX(offset);
236 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL);
237 if (m->valid != VM_PAGE_BITS_ALL) {
239 rv = vm_pager_get_pages(object, ma, 1, 0);
240 m = vm_page_lookup(object, pindex);
243 if (rv != VM_PAGER_OK) {
256 VM_OBJECT_WUNLOCK(object);
261 * Return a CPU private mapping to the page at the given offset within the
262 * given object. The page is pinned before it is mapped.
265 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
269 m = vm_imgact_hold_page(object, offset);
273 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
277 * Destroy the given CPU private mapping and unpin the page that it mapped.
280 vm_imgact_unmap_page(struct sf_buf *sf)
293 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz)
296 pmap_sync_icache(map->pmap, va, sz);
299 struct kstack_cache_entry *kstack_cache;
300 static int kstack_cache_size = 128;
302 static struct mtx kstack_cache_mtx;
303 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF);
305 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0,
307 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0,
310 #ifndef KSTACK_MAX_PAGES
311 #define KSTACK_MAX_PAGES 32
315 * Create the kernel stack (including pcb for i386) for a new thread.
316 * This routine directly affects the fork perf for a process and
317 * create performance for a thread.
320 vm_thread_new(struct thread *td, int pages)
324 vm_page_t m, ma[KSTACK_MAX_PAGES];
325 struct kstack_cache_entry *ks_ce;
330 pages = KSTACK_PAGES;
331 else if (pages > KSTACK_MAX_PAGES)
332 pages = KSTACK_MAX_PAGES;
334 if (pages == KSTACK_PAGES) {
335 mtx_lock(&kstack_cache_mtx);
336 if (kstack_cache != NULL) {
337 ks_ce = kstack_cache;
338 kstack_cache = ks_ce->next_ks_entry;
339 mtx_unlock(&kstack_cache_mtx);
341 td->td_kstack_obj = ks_ce->ksobj;
342 td->td_kstack = (vm_offset_t)ks_ce;
343 td->td_kstack_pages = KSTACK_PAGES;
346 mtx_unlock(&kstack_cache_mtx);
350 * Allocate an object for the kstack.
352 ksobj = vm_object_allocate(OBJT_DEFAULT, pages);
355 * Get a kernel virtual address for this thread's kstack.
357 #if defined(__mips__)
359 * We need to align the kstack's mapped address to fit within
360 * a single TLB entry.
362 if (vmem_xalloc(kernel_arena, (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE,
363 PAGE_SIZE * 2, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX,
364 M_BESTFIT | M_NOWAIT, &ks)) {
368 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
371 printf("vm_thread_new: kstack allocation failed\n");
372 vm_object_deallocate(ksobj);
376 atomic_add_int(&kstacks, 1);
377 if (KSTACK_GUARD_PAGES != 0) {
378 pmap_qremove(ks, KSTACK_GUARD_PAGES);
379 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
381 td->td_kstack_obj = ksobj;
384 * Knowing the number of pages allocated is useful when you
385 * want to deallocate them.
387 td->td_kstack_pages = pages;
389 * For the length of the stack, link in a real page of ram for each
392 VM_OBJECT_WLOCK(ksobj);
393 for (i = 0; i < pages; i++) {
395 * Get a kernel stack page.
397 m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY |
398 VM_ALLOC_NORMAL | VM_ALLOC_WIRED);
400 m->valid = VM_PAGE_BITS_ALL;
402 VM_OBJECT_WUNLOCK(ksobj);
403 pmap_qenter(ks, ma, pages);
408 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages)
413 atomic_add_int(&kstacks, -1);
414 pmap_qremove(ks, pages);
415 VM_OBJECT_WLOCK(ksobj);
416 for (i = 0; i < pages; i++) {
417 m = vm_page_lookup(ksobj, i);
419 panic("vm_thread_dispose: kstack already missing?");
421 vm_page_unwire(m, 0);
425 VM_OBJECT_WUNLOCK(ksobj);
426 vm_object_deallocate(ksobj);
427 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
428 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
432 * Dispose of a thread's kernel stack.
435 vm_thread_dispose(struct thread *td)
439 struct kstack_cache_entry *ks_ce;
442 pages = td->td_kstack_pages;
443 ksobj = td->td_kstack_obj;
446 td->td_kstack_pages = 0;
447 if (pages == KSTACK_PAGES && kstacks <= kstack_cache_size) {
448 ks_ce = (struct kstack_cache_entry *)ks;
449 ks_ce->ksobj = ksobj;
450 mtx_lock(&kstack_cache_mtx);
451 ks_ce->next_ks_entry = kstack_cache;
452 kstack_cache = ks_ce;
453 mtx_unlock(&kstack_cache_mtx);
456 vm_thread_stack_dispose(ksobj, ks, pages);
460 vm_thread_stack_lowmem(void *nulll)
462 struct kstack_cache_entry *ks_ce, *ks_ce1;
464 mtx_lock(&kstack_cache_mtx);
465 ks_ce = kstack_cache;
467 mtx_unlock(&kstack_cache_mtx);
469 while (ks_ce != NULL) {
471 ks_ce = ks_ce->next_ks_entry;
473 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1,
479 kstack_cache_init(void *nulll)
482 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL,
483 EVENTHANDLER_PRI_ANY);
486 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL);
490 * Allow a thread's kernel stack to be paged out.
493 vm_thread_swapout(struct thread *td)
499 cpu_thread_swapout(td);
500 pages = td->td_kstack_pages;
501 ksobj = td->td_kstack_obj;
502 pmap_qremove(td->td_kstack, pages);
503 VM_OBJECT_WLOCK(ksobj);
504 for (i = 0; i < pages; i++) {
505 m = vm_page_lookup(ksobj, i);
507 panic("vm_thread_swapout: kstack already missing?");
510 vm_page_unwire(m, 0);
513 VM_OBJECT_WUNLOCK(ksobj);
517 * Bring the kernel stack for a specified thread back in.
520 vm_thread_swapin(struct thread *td)
523 vm_page_t ma[KSTACK_MAX_PAGES];
524 int i, j, k, pages, rv;
526 pages = td->td_kstack_pages;
527 ksobj = td->td_kstack_obj;
528 VM_OBJECT_WLOCK(ksobj);
529 for (i = 0; i < pages; i++)
530 ma[i] = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL |
532 for (i = 0; i < pages; i++) {
533 if (ma[i]->valid != VM_PAGE_BITS_ALL) {
534 vm_page_assert_xbusied(ma[i]);
535 vm_object_pip_add(ksobj, 1);
536 for (j = i + 1; j < pages; j++) {
537 if (ma[j]->valid != VM_PAGE_BITS_ALL)
538 vm_page_assert_xbusied(ma[j]);
539 if (ma[j]->valid == VM_PAGE_BITS_ALL)
542 rv = vm_pager_get_pages(ksobj, ma + i, j - i, 0);
543 if (rv != VM_PAGER_OK)
544 panic("vm_thread_swapin: cannot get kstack for proc: %d",
546 vm_object_pip_wakeup(ksobj);
547 for (k = i; k < j; k++)
548 ma[k] = vm_page_lookup(ksobj, k);
549 vm_page_xunbusy(ma[i]);
550 } else if (vm_page_xbusied(ma[i]))
551 vm_page_xunbusy(ma[i]);
553 VM_OBJECT_WUNLOCK(ksobj);
554 pmap_qenter(td->td_kstack, ma, pages);
555 cpu_thread_swapin(td);
557 #endif /* !NO_SWAPPING */
560 * Implement fork's actions on an address space.
561 * Here we arrange for the address space to be copied or referenced,
562 * allocate a user struct (pcb and kernel stack), then call the
563 * machine-dependent layer to fill those in and make the new process
564 * ready to run. The new process is set up so that it returns directly
565 * to user mode to avoid stack copying and relocation problems.
568 vm_forkproc(td, p2, td2, vm2, flags)
575 struct proc *p1 = td->td_proc;
578 if ((flags & RFPROC) == 0) {
580 * Divorce the memory, if it is shared, essentially
581 * this changes shared memory amongst threads, into
584 if ((flags & RFMEM) == 0) {
585 if (p1->p_vmspace->vm_refcnt > 1) {
586 error = vmspace_unshare(p1);
591 cpu_fork(td, p2, td2, flags);
596 p2->p_vmspace = p1->p_vmspace;
597 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1);
600 while (vm_page_count_severe()) {
604 if ((flags & RFMEM) == 0) {
606 if (p1->p_vmspace->vm_shm)
611 * cpu_fork will copy and update the pcb, set up the kernel stack,
612 * and make the child ready to run.
614 cpu_fork(td, p2, td2, flags);
619 * Called after process has been wait(2)'ed apon and is being reaped.
620 * The idea is to reclaim resources that we could not reclaim while
621 * the process was still executing.
628 vmspace_exitfree(p); /* and clean-out the vmspace */
637 PROC_LOCK_ASSERT(p, MA_OWNED);
638 if ((p->p_flag & P_INMEM) == 0)
639 panic("faultin: proc swapped out with NO_SWAPPING!");
640 #else /* !NO_SWAPPING */
643 PROC_LOCK_ASSERT(p, MA_OWNED);
645 * If another process is swapping in this process,
646 * just wait until it finishes.
648 if (p->p_flag & P_SWAPPINGIN) {
649 while (p->p_flag & P_SWAPPINGIN)
650 msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0);
653 if ((p->p_flag & P_INMEM) == 0) {
655 * Don't let another thread swap process p out while we are
656 * busy swapping it in.
659 p->p_flag |= P_SWAPPINGIN;
663 * We hold no lock here because the list of threads
664 * can not change while all threads in the process are
667 FOREACH_THREAD_IN_PROC(p, td)
668 vm_thread_swapin(td);
675 /* Allow other threads to swap p out now. */
678 #endif /* NO_SWAPPING */
682 * This swapin algorithm attempts to swap-in processes only if there
683 * is enough space for them. Of course, if a process waits for a long
684 * time, it will be swapped in anyway.
686 * Giant is held on entry.
700 if (vm_page_count_min()) {
707 sx_slock(&allproc_lock);
708 FOREACH_PROC_IN_SYSTEM(p) {
710 if (p->p_state == PRS_NEW ||
711 p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) {
715 swtime = (ticks - p->p_swtick) / hz;
716 FOREACH_THREAD_IN_PROC(p, td) {
718 * An otherwise runnable thread of a process
719 * swapped out has only the TDI_SWAPPED bit set.
723 if (td->td_inhibitors == TDI_SWAPPED) {
724 slptime = (ticks - td->td_slptick) / hz;
725 pri = swtime + slptime;
726 if ((td->td_flags & TDF_SWAPINREQ) == 0)
727 pri -= p->p_nice * 8;
729 * if this thread is higher priority
730 * and there is enough space, then select
731 * this process instead of the previous
743 sx_sunlock(&allproc_lock);
746 * Nothing to do, back to sleep.
748 if ((p = pp) == NULL) {
749 tsleep(&proc0, PVM, "swapin", MAXSLP * hz / 2);
755 * Another process may be bringing or may have already
756 * brought this process in while we traverse all threads.
757 * Or, this process may even be being swapped out again.
759 if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) {
765 * We would like to bring someone in. (only if there is space).
766 * [What checks the space? ]
783 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
785 static int swap_idle_threshold1 = 2;
786 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW,
787 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process");
790 * Swap_idle_threshold2 is the time that a process can be idle before
791 * it will be swapped out, if idle swapping is enabled.
793 static int swap_idle_threshold2 = 10;
794 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW,
795 &swap_idle_threshold2, 0, "Time before a process will be swapped out");
798 * First, if any processes have been sleeping or stopped for at least
799 * "swap_idle_threshold1" seconds, they are swapped out. If, however,
800 * no such processes exist, then the longest-sleeping or stopped
801 * process is swapped out. Finally, and only as a last resort, if
802 * there are no sleeping or stopped processes, the longest-resident
803 * process is swapped out.
806 swapout_procs(action)
814 sx_slock(&allproc_lock);
815 FOREACH_PROC_IN_SYSTEM(p) {
817 int minslptime = 100000;
821 * Watch out for a process in
822 * creation. It may have no
823 * address space or lock yet.
825 if (p->p_state == PRS_NEW)
828 * An aio daemon switches its
829 * address space while running.
830 * Perform a quick check whether
831 * a process has P_SYSTEM.
833 if ((p->p_flag & P_SYSTEM) != 0)
836 * Do not swapout a process that
837 * is waiting for VM data
838 * structures as there is a possible
839 * deadlock. Test this first as
842 * Lock the map until swapout
843 * finishes, or a thread of this
844 * process may attempt to alter
847 vm = vmspace_acquire_ref(p);
850 if (!vm_map_trylock(&vm->vm_map))
854 if (p->p_lock != 0 ||
855 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT)
860 * only aiod changes vmspace, however it will be
861 * skipped because of the if statement above checking
864 if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM)
867 switch (p->p_state) {
869 /* Don't swap out processes in any sort
870 * of 'special' state. */
875 * do not swapout a realtime process
876 * Check all the thread groups..
878 FOREACH_THREAD_IN_PROC(p, td) {
880 if (PRI_IS_REALTIME(td->td_pri_class)) {
884 slptime = (ticks - td->td_slptick) / hz;
886 * Guarantee swap_idle_threshold1
889 if (slptime < swap_idle_threshold1) {
895 * Do not swapout a process if it is
896 * waiting on a critical event of some
897 * kind or there is a thread whose
898 * pageable memory may be accessed.
900 * This could be refined to support
901 * swapping out a thread.
903 if (!thread_safetoswapout(td)) {
908 * If the system is under memory stress,
909 * or if we are swapping
910 * idle processes >= swap_idle_threshold2,
911 * then swap the process out.
913 if (((action & VM_SWAP_NORMAL) == 0) &&
914 (((action & VM_SWAP_IDLE) == 0) ||
915 (slptime < swap_idle_threshold2))) {
920 if (minslptime > slptime)
921 minslptime = slptime;
926 * If the pageout daemon didn't free enough pages,
927 * or if this process is idle and the system is
928 * configured to swap proactively, swap it out.
930 if ((action & VM_SWAP_NORMAL) ||
931 ((action & VM_SWAP_IDLE) &&
932 (minslptime > swap_idle_threshold2))) {
936 vm_map_unlock(&vm->vm_map);
938 sx_sunlock(&allproc_lock);
944 vm_map_unlock(&vm->vm_map);
949 sx_sunlock(&allproc_lock);
951 * If we swapped something out, and another process needed memory,
952 * then wakeup the sched process.
964 PROC_LOCK_ASSERT(p, MA_OWNED);
966 FOREACH_THREAD_IN_PROC(p, td) {
968 td->td_flags |= TDF_INMEM;
969 td->td_flags &= ~TDF_SWAPINREQ;
972 if (setrunnable(td)) {
975 * XXX: We just cleared TDI_SWAPPED
976 * above and set TDF_INMEM, so this
977 * should never happen.
979 panic("not waking up swapper");
984 p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT);
985 p->p_flag |= P_INMEM;
994 PROC_LOCK_ASSERT(p, MA_OWNED);
995 #if defined(SWAP_DEBUG)
996 printf("swapping out %d\n", p->p_pid);
1000 * The states of this process and its threads may have changed
1001 * by now. Assuming that there is only one pageout daemon thread,
1002 * this process should still be in memory.
1004 KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM,
1005 ("swapout: lost a swapout race?"));
1008 * remember the process resident count
1010 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
1012 * Check and mark all threads before we proceed.
1014 p->p_flag &= ~P_INMEM;
1015 p->p_flag |= P_SWAPPINGOUT;
1016 FOREACH_THREAD_IN_PROC(p, td) {
1018 if (!thread_safetoswapout(td)) {
1023 td->td_flags &= ~TDF_INMEM;
1027 td = FIRST_THREAD_IN_PROC(p);
1028 ++td->td_ru.ru_nswap;
1032 * This list is stable because all threads are now prevented from
1033 * running. The list is only modified in the context of a running
1034 * thread in this process.
1036 FOREACH_THREAD_IN_PROC(p, td)
1037 vm_thread_swapout(td);
1040 p->p_flag &= ~P_SWAPPINGOUT;
1041 p->p_swtick = ticks;
1044 #endif /* !NO_SWAPPING */