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 * $Id: vm_glue.c,v 1.76 1998/09/29 17:33:59 abial Exp $
65 #include "opt_rlimit.h"
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>
84 #include <vm/vm_prot.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pageout.h>
90 #include <vm/vm_kern.h>
91 #include <vm/vm_extern.h>
96 * System initialization
98 * Note: proc0 from proc.h
101 static void vm_init_limits __P((void *));
102 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
105 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
107 * Note: run scheduling should be divorced from the vm system.
109 static void scheduler __P((void *));
110 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
113 static void swapout __P((struct proc *));
115 extern char kstack[];
117 /* vm_map_t upages_map; */
120 kernacc(addr, len, rw)
125 vm_offset_t saddr, eaddr;
126 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
128 saddr = trunc_page((vm_offset_t)addr);
129 eaddr = round_page((vm_offset_t)addr + len);
130 vm_map_lock_read(kernel_map);
131 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
132 vm_map_unlock_read(kernel_map);
137 useracc(addr, len, rw)
142 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
144 vm_map_entry_t save_hint;
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);
184 vsunlock(addr, len, dirtied)
192 vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
193 round_page((vm_offset_t)addr + len), TRUE);
197 * Implement fork's actions on an address space.
198 * Here we arrange for the address space to be copied or referenced,
199 * allocate a user struct (pcb and kernel stack), then call the
200 * machine-dependent layer to fill those in and make the new process
201 * ready to run. The new process is set up so that it returns directly
202 * to user mode to avoid stack copying and relocation problems.
205 vm_fork(p1, p2, flags)
206 register struct proc *p1, *p2;
209 register struct user *up;
212 p2->p_vmspace = p1->p_vmspace;
213 p1->p_vmspace->vm_refcnt++;
216 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) {
217 vm_pageout_deficit += (UPAGES + VM_INITIAL_PAGEIN);
221 if ((flags & RFMEM) == 0) {
222 p2->p_vmspace = vmspace_fork(p1->p_vmspace);
224 if (p1->p_vmspace->vm_shm)
233 * p_stats and p_sigacts currently point at fields in the user struct
234 * but not at &u, instead at p_addr. Copy p_sigacts and parts of
235 * p_stats; zero the rest of p_stats (statistics).
237 p2->p_stats = &up->u_stats;
238 p2->p_sigacts = &up->u_sigacts;
239 up->u_sigacts = *p1->p_sigacts;
240 bzero(&up->u_stats.pstat_startzero,
241 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
242 (caddr_t) &up->u_stats.pstat_startzero));
243 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
244 ((caddr_t) &up->u_stats.pstat_endcopy -
245 (caddr_t) &up->u_stats.pstat_startcopy));
249 * cpu_fork will copy and update the pcb, set up the kernel stack,
250 * and make the child ready to run.
256 * Set default limits for VM system.
257 * Called for proc 0, and then inherited by all others.
259 * XXX should probably act directly on proc0.
262 vm_init_limits(udata)
265 register struct proc *p = udata;
269 * Set up the initial limits on process VM. Set the maximum resident
270 * set size to be half of (reasonably) available memory. Since this
271 * is a soft limit, it comes into effect only when the system is out
272 * of memory - half of main memory helps to favor smaller processes,
273 * and reduces thrashing of the object cache.
275 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
276 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
277 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
278 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
279 /* limit the limit to no less than 2MB */
280 rss_limit = max(cnt.v_free_count, 512);
281 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
282 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
291 if ((p->p_flag & P_INMEM) == 0) {
299 if (p->p_stat == SRUN)
302 p->p_flag |= P_INMEM;
304 /* undo the effect of setting SLOCK above */
312 * This swapin algorithm attempts to swap-in processes only if there
313 * is enough space for them. Of course, if a process waits for a long
314 * time, it will be swapped in anyway.
321 register struct proc *p;
327 while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) {
333 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
334 if (p->p_stat == SRUN &&
335 (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
337 pri = p->p_swtime + p->p_slptime;
338 if ((p->p_flag & P_SWAPINREQ) == 0) {
339 pri -= p->p_nice * 8;
343 * if this process is higher priority and there is
344 * enough space, then select this process instead of
345 * the previous selection.
355 * Nothing to do, back to sleep.
357 if ((p = pp) == NULL) {
358 tsleep(&proc0, PVM, "sched", 0);
361 p->p_flag &= ~P_SWAPINREQ;
364 * We would like to bring someone in. (only if there is space).
373 #define swappable(p) \
374 (((p)->p_lock == 0) && \
375 ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM)
379 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
381 static int swap_idle_threshold1 = 2;
382 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
383 CTLFLAG_RW, &swap_idle_threshold1, 0, "");
386 * Swap_idle_threshold2 is the time that a process can be idle before
387 * it will be swapped out, if idle swapping is enabled.
389 static int swap_idle_threshold2 = 10;
390 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
391 CTLFLAG_RW, &swap_idle_threshold2, 0, "");
394 * Swapout is driven by the pageout daemon. Very simple, we find eligible
395 * procs and unwire their u-areas. We try to always "swap" at least one
396 * process in case we need the room for a swapin.
397 * If any procs have been sleeping/stopped for at least maxslp seconds,
398 * they are swapped. Else, we swap the longest-sleeping or stopped process,
399 * if any, otherwise the longest-resident process.
402 swapout_procs(action)
405 register struct proc *p;
406 struct proc *outp, *outp2;
411 outpri = outpri2 = INT_MIN;
413 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
427 * do not swapout a realtime process
429 if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type))
433 * Do not swapout a process waiting on a critical
434 * event of some kind. Also guarantee swap_idle_threshold1
437 if (((p->p_priority & 0x7f) < PSOCK) ||
438 (p->p_slptime < swap_idle_threshold1))
442 * If the system is under memory stress, or if we are swapping
443 * idle processes >= swap_idle_threshold2, then swap the process
446 if (((action & VM_SWAP_NORMAL) == 0) &&
447 (((action & VM_SWAP_IDLE) == 0) ||
448 (p->p_slptime < swap_idle_threshold2)))
453 * do not swapout a process that is waiting for VM
454 * data structures there is a possible deadlock.
456 if (lockmgr(&vm->vm_map.lock,
457 LK_EXCLUSIVE | LK_NOWAIT,
458 (void *)0, curproc)) {
462 vm_map_unlock(&vm->vm_map);
464 * If the process has been asleep for awhile and had
465 * most of its pages taken away already, swap it out.
467 if ((action & VM_SWAP_NORMAL) ||
468 ((action & VM_SWAP_IDLE) &&
469 (p->p_slptime > swap_idle_threshold2))) {
478 * If we swapped something out, and another process needed memory,
479 * then wakeup the sched process.
487 register struct proc *p;
490 #if defined(SWAP_DEBUG)
491 printf("swapping out %d\n", p->p_pid);
493 ++p->p_stats->p_ru.ru_nswap;
495 * remember the process resident count
497 p->p_vmspace->vm_swrss =
498 p->p_vmspace->vm_pmap.pm_stats.resident_count;
501 p->p_flag &= ~P_INMEM;
502 p->p_flag |= P_SWAPPING;
503 if (p->p_stat == SRUN)
507 pmap_swapout_proc(p);
509 p->p_flag &= ~P_SWAPPING;
512 #endif /* !NO_SWAPPING */