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/sched.h>
75 #include <sys/sf_buf.h>
77 #include <sys/vmmeter.h>
79 #include <sys/sysctl.h>
81 #include <sys/eventhandler.h>
82 #include <sys/kernel.h>
84 #include <sys/unistd.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pageout.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vm_pager.h>
96 #include <vm/swap_pager.h>
99 * System initialization
101 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
103 * Note: run scheduling should be divorced from the vm system.
105 static void scheduler(void *);
106 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL);
109 static int swapout(struct proc *);
110 static void swapclear(struct proc *);
111 static void vm_thread_swapin(struct thread *td);
112 static void vm_thread_swapout(struct thread *td);
118 * WARNING! This code calls vm_map_check_protection() which only checks
119 * the associated vm_map_entry range. It does not determine whether the
120 * contents of the memory is actually readable or writable. In most cases
121 * just checking the vm_map_entry is sufficient within the kernel's address
125 kernacc(addr, len, rw)
130 vm_offset_t saddr, eaddr;
133 KASSERT((rw & ~VM_PROT_ALL) == 0,
134 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
136 if ((vm_offset_t)addr + len > kernel_map->max_offset ||
137 (vm_offset_t)addr + len < (vm_offset_t)addr)
141 saddr = trunc_page((vm_offset_t)addr);
142 eaddr = round_page((vm_offset_t)addr + len);
143 vm_map_lock_read(kernel_map);
144 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
145 vm_map_unlock_read(kernel_map);
152 * WARNING! This code calls vm_map_check_protection() which only checks
153 * the associated vm_map_entry range. It does not determine whether the
154 * contents of the memory is actually readable or writable. vmapbuf(),
155 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be
156 * used in conjuction with this call.
159 useracc(addr, len, rw)
167 KASSERT((rw & ~VM_PROT_ALL) == 0,
168 ("illegal ``rw'' argument to useracc (%x)\n", rw));
170 map = &curproc->p_vmspace->vm_map;
171 if ((vm_offset_t)addr + len > vm_map_max(map) ||
172 (vm_offset_t)addr + len < (vm_offset_t)addr) {
175 vm_map_lock_read(map);
176 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr),
177 round_page((vm_offset_t)addr + len), prot);
178 vm_map_unlock_read(map);
183 vslock(void *addr, size_t len)
185 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)
199 nsize = ptoa(npages +
200 pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map)));
201 if (nsize > lim_cur(curproc, RLIMIT_MEMLOCK)) {
202 PROC_UNLOCK(curproc);
205 if (racct_set(curproc, RACCT_MEMLOCK, nsize)) {
206 PROC_UNLOCK(curproc);
209 PROC_UNLOCK(curproc);
214 * The limit for transient usage of wired pages should be
215 * larger than for "permanent" wired pages (mlock()).
217 * Also, the sysctl code, which is the only present user
218 * of vslock(), does a hard loop on EAGAIN.
220 if (npages + cnt.v_wire_count > vm_page_max_wired)
223 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end,
224 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
225 if (error != KERN_SUCCESS) {
227 racct_set(curproc, RACCT_MEMLOCK,
228 ptoa(pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map))));
229 PROC_UNLOCK(curproc);
232 * Return EFAULT on error to match copy{in,out}() behaviour
233 * rather than returning ENOMEM like mlock() would.
235 return (error == KERN_SUCCESS ? 0 : EFAULT);
239 vsunlock(void *addr, size_t len)
242 /* Rely on the parameter sanity checks performed by vslock(). */
243 (void)vm_map_unwire(&curproc->p_vmspace->vm_map,
244 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len),
245 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
248 racct_set(curproc, RACCT_MEMLOCK,
249 ptoa(pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map))));
250 PROC_UNLOCK(curproc);
254 * Pin the page contained within the given object at the given offset. If the
255 * page is not resident, allocate and load it using the given object's pager.
256 * Return the pinned page if successful; otherwise, return NULL.
259 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset)
265 VM_OBJECT_LOCK(object);
266 pindex = OFF_TO_IDX(offset);
267 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
268 if (m->valid != VM_PAGE_BITS_ALL) {
270 rv = vm_pager_get_pages(object, ma, 1, 0);
271 m = vm_page_lookup(object, pindex);
274 if (rv != VM_PAGER_OK) {
287 VM_OBJECT_UNLOCK(object);
292 * Return a CPU private mapping to the page at the given offset within the
293 * given object. The page is pinned before it is mapped.
296 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
300 m = vm_imgact_hold_page(object, offset);
304 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
308 * Destroy the given CPU private mapping and unpin the page that it mapped.
311 vm_imgact_unmap_page(struct sf_buf *sf)
324 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz)
327 pmap_sync_icache(map->pmap, va, sz);
330 struct kstack_cache_entry {
332 struct kstack_cache_entry *next_ks_entry;
335 static struct kstack_cache_entry *kstack_cache;
336 static int kstack_cache_size = 128;
338 static struct mtx kstack_cache_mtx;
339 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0,
341 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0,
344 #ifndef KSTACK_MAX_PAGES
345 #define KSTACK_MAX_PAGES 32
349 * Create the kernel stack (including pcb for i386) for a new thread.
350 * This routine directly affects the fork perf for a process and
351 * create performance for a thread.
354 vm_thread_new(struct thread *td, int pages)
358 vm_page_t m, ma[KSTACK_MAX_PAGES];
359 struct kstack_cache_entry *ks_ce;
364 pages = KSTACK_PAGES;
365 else if (pages > KSTACK_MAX_PAGES)
366 pages = KSTACK_MAX_PAGES;
368 if (pages == KSTACK_PAGES) {
369 mtx_lock(&kstack_cache_mtx);
370 if (kstack_cache != NULL) {
371 ks_ce = kstack_cache;
372 kstack_cache = ks_ce->next_ks_entry;
373 mtx_unlock(&kstack_cache_mtx);
375 td->td_kstack_obj = ks_ce->ksobj;
376 td->td_kstack = (vm_offset_t)ks_ce;
377 td->td_kstack_pages = KSTACK_PAGES;
380 mtx_unlock(&kstack_cache_mtx);
384 * Allocate an object for the kstack.
386 ksobj = vm_object_allocate(OBJT_DEFAULT, pages);
389 * Get a kernel virtual address for this thread's kstack.
391 #if defined(__mips__)
393 * We need to align the kstack's mapped address to fit within
394 * a single TLB entry.
396 ks = kmem_alloc_nofault_space(kernel_map,
397 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, VMFS_TLB_ALIGNED_SPACE);
399 ks = kmem_alloc_nofault(kernel_map,
400 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
403 printf("vm_thread_new: kstack allocation failed\n");
404 vm_object_deallocate(ksobj);
408 atomic_add_int(&kstacks, 1);
409 if (KSTACK_GUARD_PAGES != 0) {
410 pmap_qremove(ks, KSTACK_GUARD_PAGES);
411 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
413 td->td_kstack_obj = ksobj;
416 * Knowing the number of pages allocated is useful when you
417 * want to deallocate them.
419 td->td_kstack_pages = pages;
421 * For the length of the stack, link in a real page of ram for each
424 VM_OBJECT_LOCK(ksobj);
425 for (i = 0; i < pages; i++) {
427 * Get a kernel stack page.
429 m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY |
430 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED);
432 m->valid = VM_PAGE_BITS_ALL;
434 VM_OBJECT_UNLOCK(ksobj);
435 pmap_qenter(ks, ma, pages);
440 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages)
445 atomic_add_int(&kstacks, -1);
446 pmap_qremove(ks, pages);
447 VM_OBJECT_LOCK(ksobj);
448 for (i = 0; i < pages; i++) {
449 m = vm_page_lookup(ksobj, i);
451 panic("vm_thread_dispose: kstack already missing?");
453 vm_page_unwire(m, 0);
457 VM_OBJECT_UNLOCK(ksobj);
458 vm_object_deallocate(ksobj);
459 kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
460 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
464 * Dispose of a thread's kernel stack.
467 vm_thread_dispose(struct thread *td)
471 struct kstack_cache_entry *ks_ce;
474 pages = td->td_kstack_pages;
475 ksobj = td->td_kstack_obj;
478 td->td_kstack_pages = 0;
479 if (pages == KSTACK_PAGES && kstacks <= kstack_cache_size) {
480 ks_ce = (struct kstack_cache_entry *)ks;
481 ks_ce->ksobj = ksobj;
482 mtx_lock(&kstack_cache_mtx);
483 ks_ce->next_ks_entry = kstack_cache;
484 kstack_cache = ks_ce;
485 mtx_unlock(&kstack_cache_mtx);
488 vm_thread_stack_dispose(ksobj, ks, pages);
492 vm_thread_stack_lowmem(void *nulll)
494 struct kstack_cache_entry *ks_ce, *ks_ce1;
496 mtx_lock(&kstack_cache_mtx);
497 ks_ce = kstack_cache;
499 mtx_unlock(&kstack_cache_mtx);
501 while (ks_ce != NULL) {
503 ks_ce = ks_ce->next_ks_entry;
505 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1,
511 kstack_cache_init(void *nulll)
514 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL,
515 EVENTHANDLER_PRI_ANY);
518 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF);
519 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL);
523 * Allow a thread's kernel stack to be paged out.
526 vm_thread_swapout(struct thread *td)
532 cpu_thread_swapout(td);
533 pages = td->td_kstack_pages;
534 ksobj = td->td_kstack_obj;
535 pmap_qremove(td->td_kstack, pages);
536 VM_OBJECT_LOCK(ksobj);
537 for (i = 0; i < pages; i++) {
538 m = vm_page_lookup(ksobj, i);
540 panic("vm_thread_swapout: kstack already missing?");
543 vm_page_unwire(m, 0);
546 VM_OBJECT_UNLOCK(ksobj);
550 * Bring the kernel stack for a specified thread back in.
553 vm_thread_swapin(struct thread *td)
556 vm_page_t ma[KSTACK_MAX_PAGES];
557 int i, j, k, pages, rv;
559 pages = td->td_kstack_pages;
560 ksobj = td->td_kstack_obj;
561 VM_OBJECT_LOCK(ksobj);
562 for (i = 0; i < pages; i++)
563 ma[i] = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
565 for (i = 0; i < pages; i++) {
566 if (ma[i]->valid != VM_PAGE_BITS_ALL) {
567 KASSERT(ma[i]->oflags & VPO_BUSY,
569 vm_object_pip_add(ksobj, 1);
570 for (j = i + 1; j < pages; j++) {
571 KASSERT(ma[j]->valid == VM_PAGE_BITS_ALL ||
572 (ma[j]->oflags & VPO_BUSY),
574 if (ma[j]->valid == VM_PAGE_BITS_ALL)
577 rv = vm_pager_get_pages(ksobj, ma + i, j - i, 0);
578 if (rv != VM_PAGER_OK)
579 panic("vm_thread_swapin: cannot get kstack for proc: %d",
581 vm_object_pip_wakeup(ksobj);
582 for (k = i; k < j; k++)
583 ma[k] = vm_page_lookup(ksobj, k);
584 vm_page_wakeup(ma[i]);
585 } else if (ma[i]->oflags & VPO_BUSY)
586 vm_page_wakeup(ma[i]);
588 VM_OBJECT_UNLOCK(ksobj);
589 pmap_qenter(td->td_kstack, ma, pages);
590 cpu_thread_swapin(td);
592 #endif /* !NO_SWAPPING */
595 * Implement fork's actions on an address space.
596 * Here we arrange for the address space to be copied or referenced,
597 * allocate a user struct (pcb and kernel stack), then call the
598 * machine-dependent layer to fill those in and make the new process
599 * ready to run. The new process is set up so that it returns directly
600 * to user mode to avoid stack copying and relocation problems.
603 vm_forkproc(td, p2, td2, vm2, flags)
610 struct proc *p1 = td->td_proc;
613 if ((flags & RFPROC) == 0) {
615 * Divorce the memory, if it is shared, essentially
616 * this changes shared memory amongst threads, into
619 if ((flags & RFMEM) == 0) {
620 if (p1->p_vmspace->vm_refcnt > 1) {
621 error = vmspace_unshare(p1);
626 cpu_fork(td, p2, td2, flags);
631 p2->p_vmspace = p1->p_vmspace;
632 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1);
635 while (vm_page_count_severe()) {
639 if ((flags & RFMEM) == 0) {
641 if (p1->p_vmspace->vm_shm)
646 * cpu_fork will copy and update the pcb, set up the kernel stack,
647 * and make the child ready to run.
649 cpu_fork(td, p2, td2, flags);
654 * Called after process has been wait(2)'ed apon and is being reaped.
655 * The idea is to reclaim resources that we could not reclaim while
656 * the process was still executing.
663 vmspace_exitfree(p); /* and clean-out the vmspace */
672 PROC_LOCK_ASSERT(p, MA_OWNED);
673 if ((p->p_flag & P_INMEM) == 0)
674 panic("faultin: proc swapped out with NO_SWAPPING!");
675 #else /* !NO_SWAPPING */
678 PROC_LOCK_ASSERT(p, MA_OWNED);
680 * If another process is swapping in this process,
681 * just wait until it finishes.
683 if (p->p_flag & P_SWAPPINGIN) {
684 while (p->p_flag & P_SWAPPINGIN)
685 msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0);
688 if ((p->p_flag & P_INMEM) == 0) {
690 * Don't let another thread swap process p out while we are
691 * busy swapping it in.
694 p->p_flag |= P_SWAPPINGIN;
698 * We hold no lock here because the list of threads
699 * can not change while all threads in the process are
702 FOREACH_THREAD_IN_PROC(p, td)
703 vm_thread_swapin(td);
710 /* Allow other threads to swap p out now. */
713 #endif /* NO_SWAPPING */
717 * This swapin algorithm attempts to swap-in processes only if there
718 * is enough space for them. Of course, if a process waits for a long
719 * time, it will be swapped in anyway.
721 * Giant is held on entry.
736 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
740 if (vm_page_count_min()) {
747 sx_slock(&allproc_lock);
748 FOREACH_PROC_IN_SYSTEM(p) {
750 if (p->p_state == PRS_NEW ||
751 p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) {
755 swtime = (ticks - p->p_swtick) / hz;
756 FOREACH_THREAD_IN_PROC(p, td) {
758 * An otherwise runnable thread of a process
759 * swapped out has only the TDI_SWAPPED bit set.
763 if (td->td_inhibitors == TDI_SWAPPED) {
764 slptime = (ticks - td->td_slptick) / hz;
765 pri = swtime + slptime;
766 if ((td->td_flags & TDF_SWAPINREQ) == 0)
767 pri -= p->p_nice * 8;
769 * if this thread is higher priority
770 * and there is enough space, then select
771 * this process instead of the previous
783 sx_sunlock(&allproc_lock);
786 * Nothing to do, back to sleep.
788 if ((p = pp) == NULL) {
789 tsleep(&proc0, PVM, "sched", MAXSLP * hz / 2);
795 * Another process may be bringing or may have already
796 * brought this process in while we traverse all threads.
797 * Or, this process may even be being swapped out again.
799 if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) {
805 * We would like to bring someone in. (only if there is space).
806 * [What checks the space? ]
823 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
825 static int swap_idle_threshold1 = 2;
826 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW,
827 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process");
830 * Swap_idle_threshold2 is the time that a process can be idle before
831 * it will be swapped out, if idle swapping is enabled.
833 static int swap_idle_threshold2 = 10;
834 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW,
835 &swap_idle_threshold2, 0, "Time before a process will be swapped out");
838 * First, if any processes have been sleeping or stopped for at least
839 * "swap_idle_threshold1" seconds, they are swapped out. If, however,
840 * no such processes exist, then the longest-sleeping or stopped
841 * process is swapped out. Finally, and only as a last resort, if
842 * there are no sleeping or stopped processes, the longest-resident
843 * process is swapped out.
846 swapout_procs(action)
854 sx_slock(&allproc_lock);
855 FOREACH_PROC_IN_SYSTEM(p) {
857 int minslptime = 100000;
861 * Watch out for a process in
862 * creation. It may have no
863 * address space or lock yet.
865 if (p->p_state == PRS_NEW)
868 * An aio daemon switches its
869 * address space while running.
870 * Perform a quick check whether
871 * a process has P_SYSTEM.
873 if ((p->p_flag & P_SYSTEM) != 0)
876 * Do not swapout a process that
877 * is waiting for VM data
878 * structures as there is a possible
879 * deadlock. Test this first as
882 * Lock the map until swapout
883 * finishes, or a thread of this
884 * process may attempt to alter
887 vm = vmspace_acquire_ref(p);
890 if (!vm_map_trylock(&vm->vm_map))
894 if (p->p_lock != 0 ||
895 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT)
900 * only aiod changes vmspace, however it will be
901 * skipped because of the if statement above checking
904 if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM)
907 switch (p->p_state) {
909 /* Don't swap out processes in any sort
910 * of 'special' state. */
915 * do not swapout a realtime process
916 * Check all the thread groups..
918 FOREACH_THREAD_IN_PROC(p, td) {
920 if (PRI_IS_REALTIME(td->td_pri_class)) {
924 slptime = (ticks - td->td_slptick) / hz;
926 * Guarantee swap_idle_threshold1
929 if (slptime < swap_idle_threshold1) {
935 * Do not swapout a process if it is
936 * waiting on a critical event of some
937 * kind or there is a thread whose
938 * pageable memory may be accessed.
940 * This could be refined to support
941 * swapping out a thread.
943 if (!thread_safetoswapout(td)) {
948 * If the system is under memory stress,
949 * or if we are swapping
950 * idle processes >= swap_idle_threshold2,
951 * then swap the process out.
953 if (((action & VM_SWAP_NORMAL) == 0) &&
954 (((action & VM_SWAP_IDLE) == 0) ||
955 (slptime < swap_idle_threshold2))) {
960 if (minslptime > slptime)
961 minslptime = slptime;
966 * If the pageout daemon didn't free enough pages,
967 * or if this process is idle and the system is
968 * configured to swap proactively, swap it out.
970 if ((action & VM_SWAP_NORMAL) ||
971 ((action & VM_SWAP_IDLE) &&
972 (minslptime > swap_idle_threshold2))) {
976 vm_map_unlock(&vm->vm_map);
978 sx_sunlock(&allproc_lock);
984 vm_map_unlock(&vm->vm_map);
989 sx_sunlock(&allproc_lock);
991 * If we swapped something out, and another process needed memory,
992 * then wakeup the sched process.
1004 PROC_LOCK_ASSERT(p, MA_OWNED);
1006 FOREACH_THREAD_IN_PROC(p, td) {
1008 td->td_flags |= TDF_INMEM;
1009 td->td_flags &= ~TDF_SWAPINREQ;
1012 if (setrunnable(td)) {
1015 * XXX: We just cleared TDI_SWAPPED
1016 * above and set TDF_INMEM, so this
1017 * should never happen.
1019 panic("not waking up swapper");
1024 p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT);
1025 p->p_flag |= P_INMEM;
1034 PROC_LOCK_ASSERT(p, MA_OWNED);
1035 #if defined(SWAP_DEBUG)
1036 printf("swapping out %d\n", p->p_pid);
1040 * The states of this process and its threads may have changed
1041 * by now. Assuming that there is only one pageout daemon thread,
1042 * this process should still be in memory.
1044 KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM,
1045 ("swapout: lost a swapout race?"));
1048 * remember the process resident count
1050 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
1052 * Check and mark all threads before we proceed.
1054 p->p_flag &= ~P_INMEM;
1055 p->p_flag |= P_SWAPPINGOUT;
1056 FOREACH_THREAD_IN_PROC(p, td) {
1058 if (!thread_safetoswapout(td)) {
1063 td->td_flags &= ~TDF_INMEM;
1067 td = FIRST_THREAD_IN_PROC(p);
1068 ++td->td_ru.ru_nswap;
1072 * This list is stable because all threads are now prevented from
1073 * running. The list is only modified in the context of a running
1074 * thread in this process.
1076 FOREACH_THREAD_IN_PROC(p, td)
1077 vm_thread_swapout(td);
1080 p->p_flag &= ~P_SWAPPINGOUT;
1081 p->p_swtick = ticks;
1084 #endif /* !NO_SWAPPING */