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 $
67 #include <sys/param.h>
68 #include <sys/systm.h>
70 #include <sys/resourcevar.h>
73 #include <sys/vmmeter.h>
74 #include <sys/sysctl.h>
76 #include <sys/kernel.h>
77 #include <sys/unistd.h>
79 #include <machine/limits.h>
82 #include <vm/vm_param.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pageout.h>
88 #include <vm/vm_kern.h>
89 #include <vm/vm_extern.h>
94 * System initialization
96 * Note: proc0 from proc.h
99 static void vm_init_limits __P((void *));
100 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
103 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
105 * Note: run scheduling should be divorced from the vm system.
107 static void scheduler __P((void *));
108 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
111 static void swapout __P((struct proc *));
114 kernacc(addr, len, rw)
119 vm_offset_t saddr, eaddr;
122 KASSERT((rw & (~VM_PROT_ALL)) == 0,
123 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
125 saddr = trunc_page((vm_offset_t)addr);
126 eaddr = round_page((vm_offset_t)addr + len);
127 vm_map_lock_read(kernel_map);
128 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
129 vm_map_unlock_read(kernel_map);
134 useracc(addr, len, rw)
141 vm_map_entry_t save_hint;
143 KASSERT((rw & (~VM_PROT_ALL)) == 0,
144 ("illegal ``rw'' argument to useracc (%x)\n", rw));
147 * XXX - check separately to disallow access to user area and user
148 * page tables - they are in the map.
150 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once
151 * only used (as an end address) in trap.c. Use it as an end address
152 * here too. This bogusness has spread. I just fixed where it was
153 * used as a max in vm_mmap.c.
155 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
156 || (vm_offset_t) addr + len < (vm_offset_t) addr) {
159 map = &curproc->p_vmspace->vm_map;
160 vm_map_lock_read(map);
162 * We save the map hint, and restore it. Useracc appears to distort
163 * the map hint unnecessarily.
165 save_hint = map->hint;
166 rv = vm_map_check_protection(map,
167 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot);
168 map->hint = save_hint;
169 vm_map_unlock_read(map);
179 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
180 round_page((vm_offset_t)addr + len), FALSE);
188 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
189 round_page((vm_offset_t)addr + len), TRUE);
193 * Implement fork's actions on an address space.
194 * Here we arrange for the address space to be copied or referenced,
195 * allocate a user struct (pcb and kernel stack), then call the
196 * machine-dependent layer to fill those in and make the new process
197 * ready to run. The new process is set up so that it returns directly
198 * to user mode to avoid stack copying and relocation problems.
201 vm_fork(p1, p2, flags)
202 register struct proc *p1, *p2;
205 register struct user *up;
207 if ((flags & RFPROC) == 0) {
209 * Divorce the memory, if it is shared, essentially
210 * this changes shared memory amongst threads, into
213 if ((flags & RFMEM) == 0) {
214 if (p1->p_vmspace->vm_refcnt > 1) {
218 cpu_fork(p1, p2, flags);
223 p2->p_vmspace = p1->p_vmspace;
224 p1->p_vmspace->vm_refcnt++;
227 while (vm_page_count_severe()) {
231 if ((flags & RFMEM) == 0) {
232 p2->p_vmspace = vmspace_fork(p1->p_vmspace);
234 pmap_pinit2(vmspace_pmap(p2->p_vmspace));
236 if (p1->p_vmspace->vm_shm)
245 * p_stats currently points at fields in the user struct
246 * but not at &u, instead at p_addr. Copy parts of
247 * p_stats; zero the rest of p_stats (statistics).
249 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need
250 * to share sigacts, so we use the up->u_sigacts.
252 p2->p_stats = &up->u_stats;
253 if (p2->p_sigacts == NULL) {
254 if (p2->p_procsig->ps_refcnt != 1)
255 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid);
256 p2->p_sigacts = &up->u_sigacts;
257 up->u_sigacts = *p1->p_sigacts;
260 bzero(&up->u_stats.pstat_startzero,
261 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
262 (caddr_t) &up->u_stats.pstat_startzero));
263 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
264 ((caddr_t) &up->u_stats.pstat_endcopy -
265 (caddr_t) &up->u_stats.pstat_startcopy));
269 * cpu_fork will copy and update the pcb, set up the kernel stack,
270 * and make the child ready to run.
272 cpu_fork(p1, p2, flags);
276 * Called after process has been wait(2)'ed apon and is being reaped.
277 * The idea is to reclaim resources that we could not reclaim while
278 * 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.
289 * Called for proc 0, and then inherited by all others.
291 * XXX should probably act directly on proc0.
294 vm_init_limits(udata)
297 register struct proc *p = udata;
301 * Set up the initial limits on process VM. Set the maximum resident
302 * set size to be half of (reasonably) available memory. Since this
303 * is a soft limit, it comes into effect only when the system is out
304 * of memory - half of main memory helps to favor smaller processes,
305 * and reduces thrashing of the object cache.
307 p->p_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
308 p->p_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
309 p->p_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
310 p->p_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
311 /* limit the limit to no less than 2MB */
312 rss_limit = max(cnt.v_free_count, 512);
313 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
314 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
323 if ((p->p_flag & P_INMEM) == 0) {
331 if (p->p_stat == SRUN)
334 p->p_flag |= P_INMEM;
336 /* undo the effect of setting SLOCK above */
344 * This swapin algorithm attempts to swap-in processes only if there
345 * is enough space for them. Of course, if a process waits for a long
346 * time, it will be swapped in anyway.
353 register struct proc *p;
359 if (vm_page_count_min()) {
366 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
367 if (p->p_stat == SRUN &&
368 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
370 pri = p->p_swtime + p->p_slptime;
371 if ((p->p_flag & P_SWAPINREQ) == 0) {
372 pri -= p->p_nice * 8;
376 * if this process is higher priority and there is
377 * enough space, then select this process instead of
378 * the previous selection.
388 * Nothing to do, back to sleep.
390 if ((p = pp) == NULL) {
391 tsleep(&proc0, PVM, "sched", 0);
394 p->p_flag &= ~P_SWAPINREQ;
397 * We would like to bring someone in. (only if there is space).
406 #define swappable(p) \
407 (((p)->p_lock == 0) && \
408 ((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM)
412 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
414 static int swap_idle_threshold1 = 2;
415 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
416 CTLFLAG_RW, &swap_idle_threshold1, 0, "");
419 * Swap_idle_threshold2 is the time that a process can be idle before
420 * it will be swapped out, if idle swapping is enabled.
422 static int swap_idle_threshold2 = 10;
423 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
424 CTLFLAG_RW, &swap_idle_threshold2, 0, "");
427 * Swapout is driven by the pageout daemon. Very simple, we find eligible
428 * procs and unwire their u-areas. We try to always "swap" at least one
429 * process in case we need the room for a swapin.
430 * If any procs have been sleeping/stopped for at least maxslp seconds,
431 * they are swapped. Else, we swap the longest-sleeping or stopped process,
432 * if any, otherwise the longest-resident process.
435 swapout_procs(action)
438 register struct proc *p;
439 struct proc *outp, *outp2;
444 outpri = outpri2 = INT_MIN;
446 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
460 * do not swapout a realtime process
462 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type))
466 * Do not swapout a process waiting on a critical
467 * event of some kind. Also guarantee swap_idle_threshold1
470 if (((p->p_priority & 0x7f) < PSOCK) ||
471 (p->p_slptime < swap_idle_threshold1))
475 * If the system is under memory stress, or if we are swapping
476 * idle processes >= swap_idle_threshold2, then swap the process
479 if (((action & VM_SWAP_NORMAL) == 0) &&
480 (((action & VM_SWAP_IDLE) == 0) ||
481 (p->p_slptime < swap_idle_threshold2)))
486 * do not swapout a process that is waiting for VM
487 * data structures there is a possible deadlock.
489 if (lockmgr(&vm->vm_map.lock,
490 LK_EXCLUSIVE | LK_NOWAIT,
491 (void *)0, curproc)) {
495 vm_map_unlock(&vm->vm_map);
497 * If the process has been asleep for awhile and had
498 * most of its pages taken away already, swap it out.
500 if ((action & VM_SWAP_NORMAL) ||
501 ((action & VM_SWAP_IDLE) &&
502 (p->p_slptime > swap_idle_threshold2))) {
510 * cleanup our reference
516 * If we swapped something out, and another process needed memory,
517 * then wakeup the sched process.
525 register struct proc *p;
528 #if defined(SWAP_DEBUG)
529 printf("swapping out %d\n", p->p_pid);
531 ++p->p_stats->p_ru.ru_nswap;
533 * remember the process resident count
535 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
538 p->p_flag &= ~P_INMEM;
539 p->p_flag |= P_SWAPPING;
540 if (p->p_stat == SRUN)
544 pmap_swapout_proc(p);
546 p->p_flag &= ~P_SWAPPING;
549 #endif /* !NO_SWAPPING */