4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
41 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42 * All rights reserved.
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
54 * Carnegie Mellon requests users of this software to return to
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
64 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $
65 * $DragonFly: src/sys/vm/vm_glue.c,v 1.56 2008/07/01 02:02:56 dillon Exp $
70 #include <sys/param.h>
71 #include <sys/systm.h>
73 #include <sys/resourcevar.h>
76 #include <sys/vmmeter.h>
77 #include <sys/sysctl.h>
79 #include <sys/kernel.h>
80 #include <sys/unistd.h>
82 #include <machine/limits.h>
85 #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_kern.h>
92 #include <vm/vm_extern.h>
95 #include <vm/vm_page2.h>
96 #include <sys/thread2.h>
97 #include <sys/sysref2.h>
100 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
102 * Note: run scheduling should be divorced from the vm system.
104 static void scheduler (void *);
105 SYSINIT(scheduler
, SI_SUB_RUN_SCHEDULER
, SI_ORDER_FIRST
, scheduler
, NULL
)
109 static int swap_debug
= 0;
110 SYSCTL_INT(_vm
, OID_AUTO
, swap_debug
,
111 CTLFLAG_RW
, &swap_debug
, 0, "");
115 static int scheduler_notify
;
117 static void swapout (struct proc
*);
123 kernacc(c_caddr_t addr
, int len
, int rw
)
126 vm_offset_t saddr
, eaddr
;
129 KASSERT((rw
& (~VM_PROT_ALL
)) == 0,
130 ("illegal ``rw'' argument to kernacc (%x)\n", rw
));
133 * The globaldata space is not part of the kernel_map proper,
134 * check access separately.
136 if (is_globaldata_space((vm_offset_t
)addr
, (vm_offset_t
)(addr
+ len
)))
140 * Nominal kernel memory access - check access via kernel_map.
142 if ((vm_offset_t
)addr
+ len
> kernel_map
.max_offset
||
143 (vm_offset_t
)addr
+ len
< (vm_offset_t
)addr
) {
147 saddr
= trunc_page((vm_offset_t
)addr
);
148 eaddr
= round_page((vm_offset_t
)addr
+ len
);
149 rv
= vm_map_check_protection(&kernel_map
, saddr
, eaddr
, prot
, FALSE
);
158 useracc(c_caddr_t addr
, int len
, int rw
)
163 vm_map_entry_t save_hint
;
165 KASSERT((rw
& (~VM_PROT_ALL
)) == 0,
166 ("illegal ``rw'' argument to useracc (%x)\n", rw
));
169 * XXX - check separately to disallow access to user area and user
170 * page tables - they are in the map.
172 * XXX - VM_MAX_USER_ADDRESS is an end address, not a max. It was once
173 * only used (as an end address) in trap.c. Use it as an end address
174 * here too. This bogusness has spread. I just fixed where it was
175 * used as a max in vm_mmap.c.
177 if ((vm_offset_t
) addr
+ len
> /* XXX */ VM_MAX_USER_ADDRESS
178 || (vm_offset_t
) addr
+ len
< (vm_offset_t
) addr
) {
181 map
= &curproc
->p_vmspace
->vm_map
;
182 vm_map_lock_read(map
);
184 * We save the map hint, and restore it. Useracc appears to distort
185 * the map hint unnecessarily.
187 save_hint
= map
->hint
;
188 rv
= vm_map_check_protection(map
, trunc_page((vm_offset_t
)addr
),
189 round_page((vm_offset_t
)addr
+ len
),
191 map
->hint
= save_hint
;
192 vm_map_unlock_read(map
);
201 vslock(caddr_t addr
, u_int len
)
204 vm_map_wire(&curproc
->p_vmspace
->vm_map
,
205 trunc_page((vm_offset_t
)addr
),
206 round_page((vm_offset_t
)addr
+ len
), 0);
214 vsunlock(caddr_t addr
, u_int len
)
217 vm_map_wire(&curproc
->p_vmspace
->vm_map
,
218 trunc_page((vm_offset_t
)addr
),
219 round_page((vm_offset_t
)addr
+ len
),
225 * Implement fork's actions on an address space.
226 * Here we arrange for the address space to be copied or referenced,
227 * allocate a user struct (pcb and kernel stack), then call the
228 * machine-dependent layer to fill those in and make the new process
229 * ready to run. The new process is set up so that it returns directly
230 * to user mode to avoid stack copying and relocation problems.
235 vm_fork(struct proc
*p1
, struct proc
*p2
, int flags
)
237 if ((flags
& RFPROC
) == 0) {
239 * Divorce the memory, if it is shared, essentially
240 * this changes shared memory amongst threads, into
243 if ((flags
& RFMEM
) == 0) {
244 if (p1
->p_vmspace
->vm_sysref
.refcnt
> 1) {
248 cpu_fork(ONLY_LWP_IN_PROC(p1
), NULL
, flags
);
253 p2
->p_vmspace
= p1
->p_vmspace
;
254 sysref_get(&p1
->p_vmspace
->vm_sysref
);
257 while (vm_page_count_severe()) {
261 if ((flags
& RFMEM
) == 0) {
262 p2
->p_vmspace
= vmspace_fork(p1
->p_vmspace
);
264 pmap_pinit2(vmspace_pmap(p2
->p_vmspace
));
266 if (p1
->p_vmspace
->vm_shm
)
274 * Called after process has been wait(2)'ed apon and is being reaped.
275 * The idea is to reclaim resources that we could not reclaim while
276 * the process was still executing.
281 vm_waitproc(struct proc
*p
)
284 vmspace_exitfree(p
); /* and clean-out the vmspace */
288 * Set default limits for VM system. Call during proc0's initialization.
290 * Called from the low level boot code only.
293 vm_init_limits(struct proc
*p
)
298 * Set up the initial limits on process VM. Set the maximum resident
299 * set size to be half of (reasonably) available memory. Since this
300 * is a soft limit, it comes into effect only when the system is out
301 * of memory - half of main memory helps to favor smaller processes,
302 * and reduces thrashing of the object cache.
304 p
->p_rlimit
[RLIMIT_STACK
].rlim_cur
= dflssiz
;
305 p
->p_rlimit
[RLIMIT_STACK
].rlim_max
= maxssiz
;
306 p
->p_rlimit
[RLIMIT_DATA
].rlim_cur
= dfldsiz
;
307 p
->p_rlimit
[RLIMIT_DATA
].rlim_max
= maxdsiz
;
308 /* limit the limit to no less than 2MB */
309 rss_limit
= max(vmstats
.v_free_count
, 512);
310 p
->p_rlimit
[RLIMIT_RSS
].rlim_cur
= ptoa(rss_limit
);
311 p
->p_rlimit
[RLIMIT_RSS
].rlim_max
= RLIM_INFINITY
;
315 * Faultin the specified process. Note that the process can be in any
316 * state. Just clear P_SWAPPEDOUT and call wakeup in case the process is
322 faultin(struct proc
*p
)
324 if (p
->p_flag
& P_SWAPPEDOUT
) {
326 * The process is waiting in the kernel to return to user
327 * mode but cannot until P_SWAPPEDOUT gets cleared.
330 lwkt_gettoken(&proc_token
);
331 p
->p_flag
&= ~(P_SWAPPEDOUT
| P_SWAPWAIT
);
334 kprintf("swapping in %d (%s)\n", p
->p_pid
, p
->p_comm
);
337 lwkt_reltoken(&proc_token
);
343 * Kernel initialization eventually falls through to this function,
344 * which is process 0.
346 * This swapin algorithm attempts to swap-in processes only if there
347 * is enough space for them. Of course, if a process waits for a long
348 * time, it will be swapped in anyway.
350 struct scheduler_info
{
355 static int scheduler_callback(struct proc
*p
, void *data
);
358 scheduler(void *dummy
)
360 struct scheduler_info info
;
363 KKASSERT(!IN_CRITICAL_SECT(curthread
));
365 scheduler_notify
= 0;
367 * Don't try to swap anything in if we are low on memory.
369 if (vm_page_count_severe()) {
375 * Look for a good candidate to wake up
379 allproc_scan(scheduler_callback
, &info
);
382 * Nothing to do, back to sleep for at least 1/10 of a second. If
383 * we are woken up, immediately process the next request. If
384 * multiple requests have built up the first is processed
385 * immediately and the rest are staggered.
387 if ((p
= info
.pp
) == NULL
) {
388 tsleep(&proc0
, 0, "nowork", hz
/ 10);
389 if (scheduler_notify
== 0)
390 tsleep(&scheduler_notify
, 0, "nowork", 0);
395 * Fault the selected process in, then wait for a short period of
398 * XXX we need a heuristic to get a measure of system stress and
399 * then adjust our stagger wakeup delay accordingly.
401 lwkt_gettoken(&proc_token
);
405 lwkt_reltoken(&proc_token
);
406 tsleep(&proc0
, 0, "swapin", hz
/ 10);
411 * The caller must hold proc_token.
414 scheduler_callback(struct proc
*p
, void *data
)
416 struct scheduler_info
*info
= data
;
421 if (p
->p_flag
& P_SWAPWAIT
) {
423 FOREACH_LWP_IN_PROC(lp
, p
) {
424 /* XXX lwp might need a different metric */
425 pri
+= lp
->lwp_slptime
;
427 pri
+= p
->p_swtime
- p
->p_nice
* 8;
430 * The more pages paged out while we were swapped,
431 * the more work we have to do to get up and running
432 * again and the lower our wakeup priority.
434 * Each second of sleep time is worth ~1MB
436 pgs
= vmspace_resident_count(p
->p_vmspace
);
437 if (pgs
< p
->p_vmspace
->vm_swrss
) {
438 pri
-= (p
->p_vmspace
->vm_swrss
- pgs
) /
439 (1024 * 1024 / PAGE_SIZE
);
443 * If this process is higher priority and there is
444 * enough space, then select this process instead of
445 * the previous selection.
447 if (pri
> info
->ppri
) {
465 if (scheduler_notify
== 0) {
466 scheduler_notify
= 1;
467 wakeup(&scheduler_notify
);
473 #define swappable(p) \
474 (((p)->p_lock == 0) && \
475 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_SWAPPEDOUT|P_WEXIT)) == 0)
479 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
481 static int swap_idle_threshold1
= 15;
482 SYSCTL_INT(_vm
, OID_AUTO
, swap_idle_threshold1
,
483 CTLFLAG_RW
, &swap_idle_threshold1
, 0, "Guaranteed process resident time (sec)");
486 * Swap_idle_threshold2 is the time that a process can be idle before
487 * it will be swapped out, if idle swapping is enabled. Default is
490 static int swap_idle_threshold2
= 60;
491 SYSCTL_INT(_vm
, OID_AUTO
, swap_idle_threshold2
,
492 CTLFLAG_RW
, &swap_idle_threshold2
, 0, "Time (sec) a process can idle before being swapped");
495 * Swapout is driven by the pageout daemon. Very simple, we find eligible
496 * procs and mark them as being swapped out. This will cause the kernel
497 * to prefer to pageout those proc's pages first and the procs in question
498 * will not return to user mode until the swapper tells them they can.
500 * If any procs have been sleeping/stopped for at least maxslp seconds,
501 * they are swapped. Else, we swap the longest-sleeping or stopped process,
502 * if any, otherwise the longest-resident process.
505 static int swapout_procs_callback(struct proc
*p
, void *data
);
511 swapout_procs(int action
)
513 lwkt_gettoken(&vmspace_token
);
514 allproc_scan(swapout_procs_callback
, &action
);
515 lwkt_reltoken(&vmspace_token
);
519 * The caller must hold proc_token and vmspace_token.
522 swapout_procs_callback(struct proc
*p
, void *data
)
526 int action
= *(int *)data
;
535 * We only consider active processes.
537 if (p
->p_stat
!= SACTIVE
&& p
->p_stat
!= SSTOP
)
540 FOREACH_LWP_IN_PROC(lp
, p
) {
542 * do not swap out a realtime process
544 if (RTP_PRIO_IS_REALTIME(lp
->lwp_rtprio
.type
))
548 * Guarentee swap_idle_threshold time in memory
550 if (lp
->lwp_slptime
< swap_idle_threshold1
)
554 * If the system is under memory stress, or if we
555 * are swapping idle processes >= swap_idle_threshold2,
556 * then swap the process out.
558 if (((action
& VM_SWAP_NORMAL
) == 0) &&
559 (((action
& VM_SWAP_IDLE
) == 0) ||
560 (lp
->lwp_slptime
< swap_idle_threshold2
))) {
564 if (minslp
== -1 || lp
->lwp_slptime
< minslp
)
565 minslp
= lp
->lwp_slptime
;
568 sysref_get(&vm
->vm_sysref
);
571 * If the process has been asleep for awhile, swap
574 if ((action
& VM_SWAP_NORMAL
) ||
575 ((action
& VM_SWAP_IDLE
) &&
576 (minslp
> swap_idle_threshold2
))) {
581 * cleanup our reference
583 sysref_put(&vm
->vm_sysref
);
589 * The caller must hold proc_token and vmspace_token.
592 swapout(struct proc
*p
)
596 kprintf("swapping out %d (%s)\n", p
->p_pid
, p
->p_comm
);
600 * remember the process resident count
602 p
->p_vmspace
->vm_swrss
= vmspace_resident_count(p
->p_vmspace
);
603 p
->p_flag
|= P_SWAPPEDOUT
;
607 #endif /* !NO_SWAPPING */