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/eventhandler.h>
81 #include <sys/kernel.h>
83 #include <sys/unistd.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pageout.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_kern.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vm_pager.h>
95 #include <vm/swap_pager.h>
100 * System initialization
102 * Note: proc0 from proc.h
104 static void vm_init_limits(void *);
105 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0);
108 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
110 * Note: run scheduling should be divorced from the vm system.
112 static void scheduler(void *);
113 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL);
116 static int swapout(struct proc *);
117 static void swapclear(struct proc *);
123 * WARNING! This code calls vm_map_check_protection() which only checks
124 * the associated vm_map_entry range. It does not determine whether the
125 * contents of the memory is actually readable or writable. In most cases
126 * just checking the vm_map_entry is sufficient within the kernel's address
130 kernacc(addr, len, rw)
135 vm_offset_t saddr, eaddr;
138 KASSERT((rw & ~VM_PROT_ALL) == 0,
139 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
141 if ((vm_offset_t)addr + len > kernel_map->max_offset ||
142 (vm_offset_t)addr + len < (vm_offset_t)addr)
146 saddr = trunc_page((vm_offset_t)addr);
147 eaddr = round_page((vm_offset_t)addr + len);
148 vm_map_lock_read(kernel_map);
149 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
150 vm_map_unlock_read(kernel_map);
157 * WARNING! This code calls vm_map_check_protection() which only checks
158 * the associated vm_map_entry range. It does not determine whether the
159 * contents of the memory is actually readable or writable. vmapbuf(),
160 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be
161 * used in conjuction with this call.
164 useracc(addr, len, rw)
172 KASSERT((rw & ~VM_PROT_ALL) == 0,
173 ("illegal ``rw'' argument to useracc (%x)\n", rw));
175 map = &curproc->p_vmspace->vm_map;
176 if ((vm_offset_t)addr + len > vm_map_max(map) ||
177 (vm_offset_t)addr + len < (vm_offset_t)addr) {
180 vm_map_lock_read(map);
181 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr),
182 round_page((vm_offset_t)addr + len), prot);
183 vm_map_unlock_read(map);
188 vslock(void *addr, size_t len)
190 vm_offset_t end, last, start;
194 last = (vm_offset_t)addr + len;
195 start = trunc_page((vm_offset_t)addr);
196 end = round_page(last);
197 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr)
199 npages = atop(end - start);
200 if (npages > vm_page_max_wired)
204 pmap_wired_count(vm_map_pmap(&curproc->p_vmspace->vm_map))) >
205 lim_cur(curproc, RLIMIT_MEMLOCK)) {
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);
226 * Return EFAULT on error to match copy{in,out}() behaviour
227 * rather than returning ENOMEM like mlock() would.
229 return (error == KERN_SUCCESS ? 0 : EFAULT);
233 vsunlock(void *addr, size_t len)
236 /* Rely on the parameter sanity checks performed by vslock(). */
237 (void)vm_map_unwire(&curproc->p_vmspace->vm_map,
238 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len),
239 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
243 * Pin the page contained within the given object at the given offset. If the
244 * page is not resident, allocate and load it using the given object's pager.
245 * Return the pinned page if successful; otherwise, return NULL.
248 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset)
254 VM_OBJECT_LOCK(object);
255 pindex = OFF_TO_IDX(offset);
256 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
257 if (m->valid != VM_PAGE_BITS_ALL) {
259 rv = vm_pager_get_pages(object, ma, 1, 0);
260 m = vm_page_lookup(object, pindex);
263 if (rv != VM_PAGER_OK) {
264 vm_page_lock_queues();
266 vm_page_unlock_queues();
271 vm_page_lock_queues();
273 vm_page_unlock_queues();
276 VM_OBJECT_UNLOCK(object);
281 * Return a CPU private mapping to the page at the given offset within the
282 * given object. The page is pinned before it is mapped.
285 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
289 m = vm_imgact_hold_page(object, offset);
293 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
297 * Destroy the given CPU private mapping and unpin the page that it mapped.
300 vm_imgact_unmap_page(struct sf_buf *sf)
307 vm_page_lock_queues();
309 vm_page_unlock_queues();
313 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz)
316 pmap_sync_icache(map->pmap, va, sz);
319 struct kstack_cache_entry {
321 struct kstack_cache_entry *next_ks_entry;
324 static struct kstack_cache_entry *kstack_cache;
325 static int kstack_cache_size = 128;
327 static struct mtx kstack_cache_mtx;
328 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0,
330 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0,
333 #ifndef KSTACK_MAX_PAGES
334 #define KSTACK_MAX_PAGES 32
338 * Create the kernel stack (including pcb for i386) for a new thread.
339 * This routine directly affects the fork perf for a process and
340 * create performance for a thread.
343 vm_thread_new(struct thread *td, int pages)
347 vm_page_t m, ma[KSTACK_MAX_PAGES];
348 struct kstack_cache_entry *ks_ce;
353 pages = KSTACK_PAGES;
354 else if (pages > KSTACK_MAX_PAGES)
355 pages = KSTACK_MAX_PAGES;
357 if (pages == KSTACK_PAGES) {
358 mtx_lock(&kstack_cache_mtx);
359 if (kstack_cache != NULL) {
360 ks_ce = kstack_cache;
361 kstack_cache = ks_ce->next_ks_entry;
362 mtx_unlock(&kstack_cache_mtx);
364 td->td_kstack_obj = ks_ce->ksobj;
365 td->td_kstack = (vm_offset_t)ks_ce;
366 td->td_kstack_pages = KSTACK_PAGES;
369 mtx_unlock(&kstack_cache_mtx);
373 * Allocate an object for the kstack.
375 ksobj = vm_object_allocate(OBJT_DEFAULT, pages);
378 * Get a kernel virtual address for this thread's kstack.
380 #if defined(__mips__)
382 * We need to align the kstack's mapped address to fit within
383 * a single TLB entry.
385 ks = kmem_alloc_nofault_space(kernel_map,
386 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, VMFS_TLB_ALIGNED_SPACE);
388 ks = kmem_alloc_nofault(kernel_map,
389 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
392 printf("vm_thread_new: kstack allocation failed\n");
393 vm_object_deallocate(ksobj);
397 atomic_add_int(&kstacks, 1);
398 if (KSTACK_GUARD_PAGES != 0) {
399 pmap_qremove(ks, KSTACK_GUARD_PAGES);
400 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
402 td->td_kstack_obj = ksobj;
405 * Knowing the number of pages allocated is useful when you
406 * want to deallocate them.
408 td->td_kstack_pages = pages;
410 * For the length of the stack, link in a real page of ram for each
413 VM_OBJECT_LOCK(ksobj);
414 for (i = 0; i < pages; i++) {
416 * Get a kernel stack page.
418 m = vm_page_grab(ksobj, i, VM_ALLOC_NOBUSY |
419 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_WIRED);
421 m->valid = VM_PAGE_BITS_ALL;
423 VM_OBJECT_UNLOCK(ksobj);
424 pmap_qenter(ks, ma, pages);
429 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages)
434 atomic_add_int(&kstacks, -1);
435 pmap_qremove(ks, pages);
436 VM_OBJECT_LOCK(ksobj);
437 for (i = 0; i < pages; i++) {
438 m = vm_page_lookup(ksobj, i);
440 panic("vm_thread_dispose: kstack already missing?");
441 vm_page_lock_queues();
442 vm_page_unwire(m, 0);
444 vm_page_unlock_queues();
446 VM_OBJECT_UNLOCK(ksobj);
447 vm_object_deallocate(ksobj);
448 kmem_free(kernel_map, ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
449 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
453 * Dispose of a thread's kernel stack.
456 vm_thread_dispose(struct thread *td)
460 struct kstack_cache_entry *ks_ce;
463 pages = td->td_kstack_pages;
464 ksobj = td->td_kstack_obj;
467 td->td_kstack_pages = 0;
468 if (pages == KSTACK_PAGES && kstacks <= kstack_cache_size) {
469 ks_ce = (struct kstack_cache_entry *)ks;
470 ks_ce->ksobj = ksobj;
471 mtx_lock(&kstack_cache_mtx);
472 ks_ce->next_ks_entry = kstack_cache;
473 kstack_cache = ks_ce;
474 mtx_unlock(&kstack_cache_mtx);
477 vm_thread_stack_dispose(ksobj, ks, pages);
481 vm_thread_stack_lowmem(void *nulll)
483 struct kstack_cache_entry *ks_ce, *ks_ce1;
485 mtx_lock(&kstack_cache_mtx);
486 ks_ce = kstack_cache;
488 mtx_unlock(&kstack_cache_mtx);
490 while (ks_ce != NULL) {
492 ks_ce = ks_ce->next_ks_entry;
494 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1,
500 kstack_cache_init(void *nulll)
503 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL,
504 EVENTHANDLER_PRI_ANY);
507 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF);
508 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL);
511 * Allow a thread's kernel stack to be paged out.
514 vm_thread_swapout(struct thread *td)
520 cpu_thread_swapout(td);
521 pages = td->td_kstack_pages;
522 ksobj = td->td_kstack_obj;
523 pmap_qremove(td->td_kstack, pages);
524 VM_OBJECT_LOCK(ksobj);
525 for (i = 0; i < pages; i++) {
526 m = vm_page_lookup(ksobj, i);
528 panic("vm_thread_swapout: kstack already missing?");
529 vm_page_lock_queues();
531 vm_page_unwire(m, 0);
532 vm_page_unlock_queues();
534 VM_OBJECT_UNLOCK(ksobj);
538 * Bring the kernel stack for a specified thread back in.
541 vm_thread_swapin(struct thread *td)
544 vm_page_t ma[KSTACK_MAX_PAGES];
545 int i, j, k, pages, rv;
547 pages = td->td_kstack_pages;
548 ksobj = td->td_kstack_obj;
549 VM_OBJECT_LOCK(ksobj);
550 for (i = 0; i < pages; i++)
551 ma[i] = vm_page_grab(ksobj, i, VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
553 for (i = 0; i < pages; i++) {
554 if (ma[i]->valid != VM_PAGE_BITS_ALL) {
555 KASSERT(ma[i]->oflags & VPO_BUSY,
557 vm_object_pip_add(ksobj, 1);
558 for (j = i + 1; j < pages; j++) {
559 KASSERT(ma[j]->valid == VM_PAGE_BITS_ALL ||
560 (ma[j]->oflags & VPO_BUSY),
562 if (ma[j]->valid == VM_PAGE_BITS_ALL)
565 rv = vm_pager_get_pages(ksobj, ma + i, j - i, 0);
566 if (rv != VM_PAGER_OK)
567 panic("vm_thread_swapin: cannot get kstack for proc: %d",
569 vm_object_pip_wakeup(ksobj);
570 for (k = i; k < j; k++)
571 ma[k] = vm_page_lookup(ksobj, k);
572 vm_page_wakeup(ma[i]);
573 } else if (ma[i]->oflags & VPO_BUSY)
574 vm_page_wakeup(ma[i]);
576 VM_OBJECT_UNLOCK(ksobj);
577 pmap_qenter(td->td_kstack, ma, pages);
578 cpu_thread_swapin(td);
582 * Implement fork's actions on an address space.
583 * Here we arrange for the address space to be copied or referenced,
584 * allocate a user struct (pcb and kernel stack), then call the
585 * machine-dependent layer to fill those in and make the new process
586 * ready to run. The new process is set up so that it returns directly
587 * to user mode to avoid stack copying and relocation problems.
590 vm_forkproc(td, p2, td2, vm2, flags)
597 struct proc *p1 = td->td_proc;
600 if ((flags & RFPROC) == 0) {
602 * Divorce the memory, if it is shared, essentially
603 * this changes shared memory amongst threads, into
606 if ((flags & RFMEM) == 0) {
607 if (p1->p_vmspace->vm_refcnt > 1) {
608 error = vmspace_unshare(p1);
613 cpu_fork(td, p2, td2, flags);
618 p2->p_vmspace = p1->p_vmspace;
619 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1);
622 while (vm_page_count_severe()) {
626 if ((flags & RFMEM) == 0) {
628 if (p1->p_vmspace->vm_shm)
633 * cpu_fork will copy and update the pcb, set up the kernel stack,
634 * and make the child ready to run.
636 cpu_fork(td, p2, td2, flags);
641 * Called after process has been wait(2)'ed apon and is being reaped.
642 * The idea is to reclaim resources that we could not reclaim while
643 * the process was still executing.
650 vmspace_exitfree(p); /* and clean-out the vmspace */
654 * Set default limits for VM system.
655 * Called for proc 0, and then inherited by all others.
657 * XXX should probably act directly on proc0.
660 vm_init_limits(udata)
663 struct proc *p = udata;
668 * Set up the initial limits on process VM. Set the maximum resident
669 * set size to be half of (reasonably) available memory. Since this
670 * is a soft limit, it comes into effect only when the system is out
671 * of memory - half of main memory helps to favor smaller processes,
672 * and reduces thrashing of the object cache.
675 limp->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
676 limp->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
677 limp->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
678 limp->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
679 /* limit the limit to no less than 2MB */
680 rss_limit = max(cnt.v_free_count, 512);
681 limp->pl_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
682 limp->pl_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
691 PROC_LOCK_ASSERT(p, MA_OWNED);
692 if ((p->p_flag & P_INMEM) == 0)
693 panic("faultin: proc swapped out with NO_SWAPPING!");
694 #else /* !NO_SWAPPING */
697 PROC_LOCK_ASSERT(p, MA_OWNED);
699 * If another process is swapping in this process,
700 * just wait until it finishes.
702 if (p->p_flag & P_SWAPPINGIN) {
703 while (p->p_flag & P_SWAPPINGIN)
704 msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0);
707 if ((p->p_flag & P_INMEM) == 0) {
709 * Don't let another thread swap process p out while we are
710 * busy swapping it in.
713 p->p_flag |= P_SWAPPINGIN;
717 * We hold no lock here because the list of threads
718 * can not change while all threads in the process are
721 FOREACH_THREAD_IN_PROC(p, td)
722 vm_thread_swapin(td);
729 /* Allow other threads to swap p out now. */
732 #endif /* NO_SWAPPING */
736 * This swapin algorithm attempts to swap-in processes only if there
737 * is enough space for them. Of course, if a process waits for a long
738 * time, it will be swapped in anyway.
740 * Giant is held on entry.
755 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
759 if (vm_page_count_min()) {
766 sx_slock(&allproc_lock);
767 FOREACH_PROC_IN_SYSTEM(p) {
769 if (p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) {
773 swtime = (ticks - p->p_swtick) / hz;
774 FOREACH_THREAD_IN_PROC(p, td) {
776 * An otherwise runnable thread of a process
777 * swapped out has only the TDI_SWAPPED bit set.
781 if (td->td_inhibitors == TDI_SWAPPED) {
782 slptime = (ticks - td->td_slptick) / hz;
783 pri = swtime + slptime;
784 if ((td->td_flags & TDF_SWAPINREQ) == 0)
785 pri -= p->p_nice * 8;
787 * if this thread is higher priority
788 * and there is enough space, then select
789 * this process instead of the previous
801 sx_sunlock(&allproc_lock);
804 * Nothing to do, back to sleep.
806 if ((p = pp) == NULL) {
807 tsleep(&proc0, PVM, "sched", maxslp * hz / 2);
813 * Another process may be bringing or may have already
814 * brought this process in while we traverse all threads.
815 * Or, this process may even be being swapped out again.
817 if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) {
823 * We would like to bring someone in. (only if there is space).
824 * [What checks the space? ]
841 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
843 static int swap_idle_threshold1 = 2;
844 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1, CTLFLAG_RW,
845 &swap_idle_threshold1, 0, "Guaranteed swapped in time for a process");
848 * Swap_idle_threshold2 is the time that a process can be idle before
849 * it will be swapped out, if idle swapping is enabled.
851 static int swap_idle_threshold2 = 10;
852 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW,
853 &swap_idle_threshold2, 0, "Time before a process will be swapped out");
856 * Swapout is driven by the pageout daemon. Very simple, we find eligible
857 * procs and swap out their stacks. We try to always "swap" at least one
858 * process in case we need the room for a swapin.
859 * If any procs have been sleeping/stopped for at least maxslp seconds,
860 * they are swapped. Else, we swap the longest-sleeping or stopped process,
861 * if any, otherwise the longest-resident process.
864 swapout_procs(action)
872 sx_slock(&allproc_lock);
873 FOREACH_PROC_IN_SYSTEM(p) {
875 int minslptime = 100000;
879 * Watch out for a process in
880 * creation. It may have no
881 * address space or lock yet.
883 if (p->p_state == PRS_NEW)
886 * An aio daemon switches its
887 * address space while running.
888 * Perform a quick check whether
889 * a process has P_SYSTEM.
891 if ((p->p_flag & P_SYSTEM) != 0)
894 * Do not swapout a process that
895 * is waiting for VM data
896 * structures as there is a possible
897 * deadlock. Test this first as
900 * Lock the map until swapout
901 * finishes, or a thread of this
902 * process may attempt to alter
905 vm = vmspace_acquire_ref(p);
908 if (!vm_map_trylock(&vm->vm_map))
912 if (p->p_lock != 0 ||
913 (p->p_flag & (P_STOPPED_SINGLE|P_TRACED|P_SYSTEM|P_WEXIT)
918 * only aiod changes vmspace, however it will be
919 * skipped because of the if statement above checking
922 if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM)
925 switch (p->p_state) {
927 /* Don't swap out processes in any sort
928 * of 'special' state. */
933 * do not swapout a realtime process
934 * Check all the thread groups..
936 FOREACH_THREAD_IN_PROC(p, td) {
938 if (PRI_IS_REALTIME(td->td_pri_class)) {
942 slptime = (ticks - td->td_slptick) / hz;
944 * Guarantee swap_idle_threshold1
947 if (slptime < swap_idle_threshold1) {
953 * Do not swapout a process if it is
954 * waiting on a critical event of some
955 * kind or there is a thread whose
956 * pageable memory may be accessed.
958 * This could be refined to support
959 * swapping out a thread.
961 if (!thread_safetoswapout(td)) {
966 * If the system is under memory stress,
967 * or if we are swapping
968 * idle processes >= swap_idle_threshold2,
969 * then swap the process out.
971 if (((action & VM_SWAP_NORMAL) == 0) &&
972 (((action & VM_SWAP_IDLE) == 0) ||
973 (slptime < swap_idle_threshold2))) {
978 if (minslptime > slptime)
979 minslptime = slptime;
984 * If the pageout daemon didn't free enough pages,
985 * or if this process is idle and the system is
986 * configured to swap proactively, swap it out.
988 if ((action & VM_SWAP_NORMAL) ||
989 ((action & VM_SWAP_IDLE) &&
990 (minslptime > swap_idle_threshold2))) {
994 vm_map_unlock(&vm->vm_map);
996 sx_sunlock(&allproc_lock);
1002 vm_map_unlock(&vm->vm_map);
1007 sx_sunlock(&allproc_lock);
1009 * If we swapped something out, and another process needed memory,
1010 * then wakeup the sched process.
1022 PROC_LOCK_ASSERT(p, MA_OWNED);
1024 FOREACH_THREAD_IN_PROC(p, td) {
1026 td->td_flags |= TDF_INMEM;
1027 td->td_flags &= ~TDF_SWAPINREQ;
1030 if (setrunnable(td)) {
1033 * XXX: We just cleared TDI_SWAPPED
1034 * above and set TDF_INMEM, so this
1035 * should never happen.
1037 panic("not waking up swapper");
1042 p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT);
1043 p->p_flag |= P_INMEM;
1052 PROC_LOCK_ASSERT(p, MA_OWNED);
1053 #if defined(SWAP_DEBUG)
1054 printf("swapping out %d\n", p->p_pid);
1058 * The states of this process and its threads may have changed
1059 * by now. Assuming that there is only one pageout daemon thread,
1060 * this process should still be in memory.
1062 KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM,
1063 ("swapout: lost a swapout race?"));
1066 * remember the process resident count
1068 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
1070 * Check and mark all threads before we proceed.
1072 p->p_flag &= ~P_INMEM;
1073 p->p_flag |= P_SWAPPINGOUT;
1074 FOREACH_THREAD_IN_PROC(p, td) {
1076 if (!thread_safetoswapout(td)) {
1081 td->td_flags &= ~TDF_INMEM;
1085 td = FIRST_THREAD_IN_PROC(p);
1086 ++td->td_ru.ru_nswap;
1090 * This list is stable because all threads are now prevented from
1091 * running. The list is only modified in the context of a running
1092 * thread in this process.
1094 FOREACH_THREAD_IN_PROC(p, td)
1095 vm_thread_swapout(td);
1098 p->p_flag &= ~P_SWAPPINGOUT;
1099 p->p_swtick = ticks;
1102 #endif /* !NO_SWAPPING */