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 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
62 * $FreeBSD: src/sys/vm/vm_glue.c,v 1.94.2.4 2003/01/13 22:51:17 dillon Exp $
63 * $DragonFly: src/sys/vm/vm_glue.c,v 1.4 2003/06/19 06:26:10 dillon Exp $
68 #include <sys/param.h>
69 #include <sys/systm.h>
71 #include <sys/resourcevar.h>
74 #include <sys/vmmeter.h>
75 #include <sys/sysctl.h>
77 #include <sys/kernel.h>
78 #include <sys/unistd.h>
80 #include <machine/limits.h>
83 #include <vm/vm_param.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pageout.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_extern.h>
95 * System initialization
97 * Note: proc0 from proc.h
100 static void vm_init_limits __P((void *));
101 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
104 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
106 * Note: run scheduling should be divorced from the vm system.
108 static void scheduler __P((void *));
109 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
112 static void swapout __P((struct proc *));
115 kernacc(addr, len, rw)
120 vm_offset_t saddr, eaddr;
123 KASSERT((rw & (~VM_PROT_ALL)) == 0,
124 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
126 saddr = trunc_page((vm_offset_t)addr);
127 eaddr = round_page((vm_offset_t)addr + len);
128 vm_map_lock_read(kernel_map);
129 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
130 vm_map_unlock_read(kernel_map);
135 useracc(addr, len, rw)
142 vm_map_entry_t save_hint;
144 KASSERT((rw & (~VM_PROT_ALL)) == 0,
145 ("illegal ``rw'' argument to useracc (%x)\n", rw));
148 * XXX - check separately to disallow access to user area and user
149 * page tables - they are in the map.
151 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once
152 * only used (as an end address) in trap.c. Use it as an end address
153 * here too. This bogusness has spread. I just fixed where it was
154 * used as a max in vm_mmap.c.
156 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
157 || (vm_offset_t) addr + len < (vm_offset_t) addr) {
160 map = &curproc->p_vmspace->vm_map;
161 vm_map_lock_read(map);
163 * We save the map hint, and restore it. Useracc appears to distort
164 * the map hint unnecessarily.
166 save_hint = map->hint;
167 rv = vm_map_check_protection(map,
168 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot);
169 map->hint = save_hint;
170 vm_map_unlock_read(map);
180 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
181 round_page((vm_offset_t)addr + len), FALSE);
189 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
190 round_page((vm_offset_t)addr + len), TRUE);
194 * Implement fork's actions on an address space.
195 * Here we arrange for the address space to be copied or referenced,
196 * allocate a user struct (pcb and kernel stack), then call the
197 * machine-dependent layer to fill those in and make the new process
198 * ready to run. The new process is set up so that it returns directly
199 * to user mode to avoid stack copying and relocation problems.
202 vm_fork(p1, p2, flags)
203 register struct proc *p1, *p2;
206 register struct user *up;
209 if ((flags & RFPROC) == 0) {
211 * Divorce the memory, if it is shared, essentially
212 * this changes shared memory amongst threads, into
215 if ((flags & RFMEM) == 0) {
216 if (p1->p_vmspace->vm_refcnt > 1) {
220 cpu_fork(p1, p2, flags);
225 p2->p_vmspace = p1->p_vmspace;
226 p1->p_vmspace->vm_refcnt++;
229 while (vm_page_count_severe()) {
233 if ((flags & RFMEM) == 0) {
234 p2->p_vmspace = vmspace_fork(p1->p_vmspace);
236 pmap_pinit2(vmspace_pmap(p2->p_vmspace));
238 if (p1->p_vmspace->vm_shm)
242 td2 = pmap_new_thread();
243 pmap_new_proc(p2, td2);
248 * p_stats currently points at fields in the user struct
249 * but not at &u, instead at p_addr. Copy parts of
250 * p_stats; zero the rest of p_stats (statistics).
252 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need
253 * to share sigacts, so we use the up->u_sigacts.
255 p2->p_stats = &up->u_stats;
256 if (p2->p_sigacts == NULL) {
257 if (p2->p_procsig->ps_refcnt != 1)
258 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid);
259 p2->p_sigacts = &up->u_sigacts;
260 up->u_sigacts = *p1->p_sigacts;
263 bzero(&up->u_stats.pstat_startzero,
264 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
265 (caddr_t) &up->u_stats.pstat_startzero));
266 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
267 ((caddr_t) &up->u_stats.pstat_endcopy -
268 (caddr_t) &up->u_stats.pstat_startcopy));
272 * cpu_fork will copy and update the pcb, set up the kernel stack,
273 * and make the child ready to run.
275 cpu_fork(p1, p2, flags);
279 * Called after process has been wait(2)'ed apon and is being reaped.
280 * The idea is to reclaim resources that we could not reclaim while
281 * the process was still executing.
284 vm_waitproc(struct proc *p)
287 vmspace_exitfree(p); /* and clean-out the vmspace */
291 * Set default limits for VM system.
292 * Called for proc 0, and then inherited by all others.
294 * XXX should probably act directly on proc0.
297 vm_init_limits(udata)
300 register struct proc *p = udata;
304 * Set up the initial limits on process VM. Set the maximum resident
305 * set size to be half of (reasonably) available memory. Since this
306 * is a soft limit, it comes into effect only when the system is out
307 * of memory - half of main memory helps to favor smaller processes,
308 * and reduces thrashing of the object cache.
310 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
311 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
312 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
313 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
314 /* limit the limit to no less than 2MB */
315 rss_limit = max(cnt.v_free_count, 512);
316 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
317 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
326 if ((p->p_flag & P_INMEM) == 0) {
334 if (p->p_stat == SRUN)
337 p->p_flag |= P_INMEM;
339 /* undo the effect of setting SLOCK above */
347 * This swapin algorithm attempts to swap-in processes only if there
348 * is enough space for them. Of course, if a process waits for a long
349 * time, it will be swapped in anyway.
356 register struct proc *p;
362 if (vm_page_count_min()) {
369 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
370 if (p->p_stat == SRUN &&
371 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
373 pri = p->p_swtime + p->p_slptime;
374 if ((p->p_flag & P_SWAPINREQ) == 0) {
375 pri -= p->p_nice * 8;
379 * if this process is higher priority and there is
380 * enough space, then select this process instead of
381 * the previous selection.
391 * Nothing to do, back to sleep.
393 if ((p = pp) == NULL) {
394 tsleep(&proc0, PVM, "sched", 0);
397 p->p_flag &= ~P_SWAPINREQ;
400 * We would like to bring someone in. (only if there is space).
409 #define swappable(p) \
410 (((p)->p_lock == 0) && \
411 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM)
415 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
417 static int swap_idle_threshold1 = 2;
418 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
419 CTLFLAG_RW, &swap_idle_threshold1, 0, "");
422 * Swap_idle_threshold2 is the time that a process can be idle before
423 * it will be swapped out, if idle swapping is enabled.
425 static int swap_idle_threshold2 = 10;
426 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
427 CTLFLAG_RW, &swap_idle_threshold2, 0, "");
430 * Swapout is driven by the pageout daemon. Very simple, we find eligible
431 * procs and unwire their u-areas. We try to always "swap" at least one
432 * process in case we need the room for a swapin.
433 * If any procs have been sleeping/stopped for at least maxslp seconds,
434 * they are swapped. Else, we swap the longest-sleeping or stopped process,
435 * if any, otherwise the longest-resident process.
438 swapout_procs(action)
441 register struct proc *p;
442 struct proc *outp, *outp2;
447 outpri = outpri2 = INT_MIN;
449 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
463 * do not swapout a realtime process
465 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type))
469 * Do not swapout a process waiting on a critical
470 * event of some kind. Also guarantee swap_idle_threshold1
473 if (((p->p_priority & 0x7f) < PSOCK) ||
474 (p->p_slptime < swap_idle_threshold1))
478 * If the system is under memory stress, or if we are swapping
479 * idle processes >= swap_idle_threshold2, then swap the process
482 if (((action & VM_SWAP_NORMAL) == 0) &&
483 (((action & VM_SWAP_IDLE) == 0) ||
484 (p->p_slptime < swap_idle_threshold2)))
489 * do not swapout a process that is waiting for VM
490 * data structures there is a possible deadlock.
492 if (lockmgr(&vm->vm_map.lock,
493 LK_EXCLUSIVE | LK_NOWAIT,
494 (void *)0, curproc)) {
498 vm_map_unlock(&vm->vm_map);
500 * If the process has been asleep for awhile and had
501 * most of its pages taken away already, swap it out.
503 if ((action & VM_SWAP_NORMAL) ||
504 ((action & VM_SWAP_IDLE) &&
505 (p->p_slptime > swap_idle_threshold2))) {
513 * cleanup our reference
519 * If we swapped something out, and another process needed memory,
520 * then wakeup the sched process.
528 register struct proc *p;
531 #if defined(SWAP_DEBUG)
532 printf("swapping out %d\n", p->p_pid);
534 ++p->p_stats->p_ru.ru_nswap;
536 * remember the process resident count
538 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
541 p->p_flag &= ~P_INMEM;
542 p->p_flag |= P_SWAPPING;
543 if (p->p_stat == SRUN)
547 pmap_swapout_proc(p);
549 p->p_flag &= ~P_SWAPPING;
552 #endif /* !NO_SWAPPING */