1/*
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
23 *
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
34 * SUCH DAMAGE.
35 *
36 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
37 *
38 *
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
41 *
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.
47 *
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.
51 *
52 * Carnegie Mellon requests users of this software to return to
53 *
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
58 *
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
61 *
62 * $FreeBSD: head/sys/vm/vm_glue.c 83366 2001-09-12 08:38:13Z julian $
63 */
64
65#include "opt_rlimit.h"
66#include "opt_vm.h"
67
68#include <sys/param.h>
69#include <sys/systm.h>
70#include <sys/lock.h>
71#include <sys/mutex.h>
72#include <sys/proc.h>
73#include <sys/resourcevar.h>
74#include <sys/shm.h>
75#include <sys/vmmeter.h>
76#include <sys/sx.h>
77#include <sys/sysctl.h>
78
79#include <sys/kernel.h>
80#include <sys/ktr.h>
81#include <sys/unistd.h>
82
83#include <machine/limits.h>
84
85#include <vm/vm.h>
86#include <vm/vm_param.h>
87#include <vm/pmap.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>
93
94#include <sys/user.h>
95
96extern int maxslp;
97
98/*
99 * System initialization
100 *
101 * Note: proc0 from proc.h
102 */
103
104static void vm_init_limits __P((void *));
105SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
106
107/*
108 * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
109 *
110 * Note: run scheduling should be divorced from the vm system.
111 */
112static void scheduler __P((void *));
113SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
114
115
116static void swapout __P((struct proc *));
117
118int
119kernacc(addr, len, rw)
120 caddr_t addr;
121 int len, rw;
122{
123 boolean_t rv;
124 vm_offset_t saddr, eaddr;
125 vm_prot_t prot;
126
127 KASSERT((rw & (~VM_PROT_ALL)) == 0,
128 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
129 prot = rw;
130 saddr = trunc_page((vm_offset_t)addr);
131 eaddr = round_page((vm_offset_t)addr + len);
132 vm_map_lock_read(kernel_map);
133 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
134 vm_map_unlock_read(kernel_map);
135 return (rv == TRUE);
136}
137
138int
139useracc(addr, len, rw)
140 caddr_t addr;
141 int len, rw;
142{
143 boolean_t rv;
144 vm_prot_t prot;
145 vm_map_t map;
146 vm_map_entry_t save_hint;
147
148 GIANT_REQUIRED;
149
150 KASSERT((rw & (~VM_PROT_ALL)) == 0,
151 ("illegal ``rw'' argument to useracc (%x)\n", rw));
152 prot = rw;
153 /*
154 * XXX - check separately to disallow access to user area and user
155 * page tables - they are in the map.
156 *
157 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was once
158 * only used (as an end address) in trap.c. Use it as an end address
159 * here too. This bogusness has spread. I just fixed where it was
160 * used as a max in vm_mmap.c.
161 */
162 if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
163 || (vm_offset_t) addr + len < (vm_offset_t) addr) {
164 return (FALSE);
165 }
166 map = &curproc->p_vmspace->vm_map;
167 vm_map_lock_read(map);
168 /*
169 * We save the map hint, and restore it. Useracc appears to distort
170 * the map hint unnecessarily.
171 */
172 save_hint = map->hint;
173 rv = vm_map_check_protection(map,
174 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot);
175 map->hint = save_hint;
176 vm_map_unlock_read(map);
177
178 return (rv == TRUE);
179}
180
181void
182vslock(addr, len)
183 caddr_t addr;
184 u_int len;
185{
186 GIANT_REQUIRED;
187 vm_map_pageable(&curproc->p_vmspace->vm_map,
188 trunc_page((vm_offset_t)addr),
189 round_page((vm_offset_t)addr + len), FALSE);
190}
191
192void
193vsunlock(addr, len)
194 caddr_t addr;
195 u_int len;
196{
197 GIANT_REQUIRED;
198 vm_map_pageable(&curproc->p_vmspace->vm_map,
199 trunc_page((vm_offset_t)addr),
200 round_page((vm_offset_t)addr + len), TRUE);
201}
202
203/*
204 * Implement fork's actions on an address space.
205 * Here we arrange for the address space to be copied or referenced,
206 * allocate a user struct (pcb and kernel stack), then call the
207 * machine-dependent layer to fill those in and make the new process
208 * ready to run. The new process is set up so that it returns directly
209 * to user mode to avoid stack copying and relocation problems.
210 */
211void
212vm_forkproc(td, p2, flags)
213 struct thread *td;
214 struct proc *p2;
215 int flags;
216{
217 struct proc *p1 = td->td_proc;
218 struct user *up;
219
220 GIANT_REQUIRED;
221
222 if ((flags & RFPROC) == 0) {
223 /*
224 * Divorce the memory, if it is shared, essentially
225 * this changes shared memory amongst threads, into
226 * COW locally.
227 */
228 if ((flags & RFMEM) == 0) {
229 if (p1->p_vmspace->vm_refcnt > 1) {
230 vmspace_unshare(p1);
231 }
232 }
233 cpu_fork(td, p2, flags);
234 return;
235 }
236
237 if (flags & RFMEM) {
238 p2->p_vmspace = p1->p_vmspace;
239 p1->p_vmspace->vm_refcnt++;
240 }
241
242 while (vm_page_count_severe()) {
243 VM_WAIT;
244 }
245
246 if ((flags & RFMEM) == 0) {
247 p2->p_vmspace = vmspace_fork(p1->p_vmspace);
248
249 pmap_pinit2(vmspace_pmap(p2->p_vmspace));
250
251 if (p1->p_vmspace->vm_shm)
252 shmfork(p1, p2);
253 }
254
255 pmap_new_proc(p2);
256 pmap_new_thread(&p2->p_thread); /* Initial thread */
257
258 /* XXXKSE this is unsatisfactory but should be adequate */
259 up = p2->p_uarea;
260
261 /*
262 * p_stats currently points at fields in the user struct
263 * but not at &u, instead at p_addr. Copy parts of
264 * p_stats; zero the rest of p_stats (statistics).
265 *
266 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need
267 * to share sigacts, so we use the up->u_sigacts.
268 */
269 p2->p_stats = &up->u_stats;
270 if (p2->p_sigacts == NULL) {
271 if (p2->p_procsig->ps_refcnt != 1)
272 printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid);
273 p2->p_sigacts = &up->u_sigacts;
274 up->u_sigacts = *p1->p_sigacts;
275 }
276
277 bzero(&up->u_stats.pstat_startzero,
278 (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
279 (caddr_t) &up->u_stats.pstat_startzero));
280 bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
281 ((caddr_t) &up->u_stats.pstat_endcopy -
282 (caddr_t) &up->u_stats.pstat_startcopy));
283
284
285 /*
286 * cpu_fork will copy and update the pcb, set up the kernel stack,
287 * and make the child ready to run.
288 */
289 cpu_fork(td, p2, flags);
290}
291
292/*
293 * Called after process has been wait(2)'ed apon and is being reaped.
294 * The idea is to reclaim resources that we could not reclaim while
295 * the process was still executing.
296 */
297void
298vm_waitproc(p)
299 struct proc *p;
300{
301 struct thread *td;
302
303 GIANT_REQUIRED;
304 cpu_wait(p);
305 pmap_dispose_proc(p); /* drop per-process resources */
306 FOREACH_THREAD_IN_PROC(p, td)
307 pmap_dispose_thread(td);
308 vmspace_free(p->p_vmspace); /* and clean-out the vmspace */
309}
310
311/*
312 * Set default limits for VM system.
313 * Called for proc 0, and then inherited by all others.
314 *
315 * XXX should probably act directly on proc0.
316 */
317static void
318vm_init_limits(udata)
319 void *udata;
320{
321 struct proc *p = udata;
322 int rss_limit;
323
324 /*
325 * Set up the initial limits on process VM. Set the maximum resident
326 * set size to be half of (reasonably) available memory. Since this
327 * is a soft limit, it comes into effect only when the system is out
328 * of memory - half of main memory helps to favor smaller processes,
329 * and reduces thrashing of the object cache.
330 */
331 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
332 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
333 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
334 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
335 /* limit the limit to no less than 2MB */
336 rss_limit = max(cnt.v_free_count, 512);
337 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
338 p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
339}
340
341/*
342 * Must be called with the proc struc mutex held.
343 */
344void
345faultin(p)
346 struct proc *p;
347{
348 struct thread *td;
349 GIANT_REQUIRED;
350
351 PROC_LOCK_ASSERT(p, MA_OWNED);
352 mtx_lock_spin(&sched_lock);
353 if ((p->p_sflag & PS_INMEM) == 0) {
354 ++p->p_lock;
355 mtx_unlock_spin(&sched_lock);
356 PROC_UNLOCK(p);
357
358 pmap_swapin_proc(p);
359 FOREACH_THREAD_IN_PROC (p, td)
360 pmap_swapin_thread(td);
361
362 PROC_LOCK(p);
363 mtx_lock_spin(&sched_lock);
364 FOREACH_THREAD_IN_PROC (p, td)
365 if (td->td_proc->p_stat == SRUN) /* XXXKSE */
366 setrunqueue(td);
367
368 p->p_sflag |= PS_INMEM;
369
370 /* undo the effect of setting SLOCK above */
371 --p->p_lock;
372 }
373 mtx_unlock_spin(&sched_lock);
374}
375
376/*
377 * This swapin algorithm attempts to swap-in processes only if there
378 * is enough space for them. Of course, if a process waits for a long
379 * time, it will be swapped in anyway.
380 *
381 * XXXKSE - KSEGRP with highest priority counts..
382 *
383 * Giant is still held at this point, to be released in tsleep.
384 */
385/* ARGSUSED*/
386static void
387scheduler(dummy)
388 void *dummy;
389{
390 struct proc *p;
391 int pri;
392 struct proc *pp;
393 int ppri;
394
395 mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
396 /* GIANT_REQUIRED */
397
398loop:
399 if (vm_page_count_min()) {
400 VM_WAIT;
401 goto loop;
402 }
403
404 pp = NULL;
405 ppri = INT_MIN;
406 sx_slock(&allproc_lock);
407 FOREACH_PROC_IN_SYSTEM(p) {
408 struct ksegrp *kg;
409 mtx_lock_spin(&sched_lock);
410 if (p->p_stat == SRUN
411 && (p->p_sflag & (PS_INMEM | PS_SWAPPING)) == 0) {
412 /* Find the minimum sleeptime for the process */
413 FOREACH_KSEGRP_IN_PROC(p, kg) {
414 pri = p->p_swtime + kg->kg_slptime;
415 if ((p->p_sflag & PS_SWAPINREQ) == 0) {
416 pri -= kg->kg_nice * 8;
417 }
418
419
420 /*
421 * if this ksegrp is higher priority
422 * and there is enough space, then select
423 * this process instead of the previous
424 * selection.
425 */
426 if (pri > ppri) {
427 pp = p;
428 ppri = pri;
429 }
430 }
431 }
432 mtx_unlock_spin(&sched_lock);
433 }
434 sx_sunlock(&allproc_lock);
435
436 /*
437 * Nothing to do, back to sleep.
438 */
439 if ((p = pp) == NULL) {
440 tsleep(&proc0, PVM, "sched", maxslp * hz / 2);
441 goto loop;
442 }
443 mtx_lock_spin(&sched_lock);
444 p->p_sflag &= ~PS_SWAPINREQ;
445 mtx_unlock_spin(&sched_lock);
446
447 /*
448 * We would like to bring someone in. (only if there is space).
449 */
450 PROC_LOCK(p);
451 faultin(p);
452 PROC_UNLOCK(p);
453 mtx_lock_spin(&sched_lock);
454 p->p_swtime = 0;
455 mtx_unlock_spin(&sched_lock);
456 goto loop;
457}
458
459#ifndef NO_SWAPPING
460
461/*
462 * Swap_idle_threshold1 is the guaranteed swapped in time for a process
463 */
464static int swap_idle_threshold1 = 2;
465SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
466 CTLFLAG_RW, &swap_idle_threshold1, 0, "");
467
468/*
469 * Swap_idle_threshold2 is the time that a process can be idle before
470 * it will be swapped out, if idle swapping is enabled.
471 */
472static int swap_idle_threshold2 = 10;
473SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
474 CTLFLAG_RW, &swap_idle_threshold2, 0, "");
475
476/*
477 * Swapout is driven by the pageout daemon. Very simple, we find eligible
478 * procs and unwire their u-areas. We try to always "swap" at least one
479 * process in case we need the room for a swapin.
480 * If any procs have been sleeping/stopped for at least maxslp seconds,
481 * they are swapped. Else, we swap the longest-sleeping or stopped process,
482 * if any, otherwise the longest-resident process.
483 */
484void
485swapout_procs(action)
486int action;
487{
488 struct proc *p;
489 struct ksegrp *kg;
490 struct proc *outp, *outp2;
491 int outpri, outpri2;
492 int didswap = 0;
493
494 GIANT_REQUIRED;
495
496 outp = outp2 = NULL;
497 outpri = outpri2 = INT_MIN;
498retry:
499 sx_slock(&allproc_lock);
500 LIST_FOREACH(p, &allproc, p_list) {
501 struct vmspace *vm;
502 int minslptime = 100000;
503
504 PROC_LOCK(p);
505 if (p->p_lock != 0 ||
506 (p->p_flag & (P_TRACED|P_SYSTEM|P_WEXIT)) != 0) {
507 PROC_UNLOCK(p);
508 continue;
509 }
510 /*
511 * only aiod changes vmspace, however it will be
512 * skipped because of the if statement above checking
513 * for P_SYSTEM
514 */
515 vm = p->p_vmspace;
516 mtx_lock_spin(&sched_lock);
517 if ((p->p_sflag & (PS_INMEM|PS_SWAPPING)) != PS_INMEM) {
518 mtx_unlock_spin(&sched_lock);
519 PROC_UNLOCK(p);
520 continue;
521 }
522
523 switch (p->p_stat) {
524 default:
525 mtx_unlock_spin(&sched_lock);
526 PROC_UNLOCK(p);
527 continue;
528
529 case SSLEEP:
530 case SSTOP:
531 /*
532 * do not swapout a realtime process
533 * Check all the thread groups..
534 */
535 FOREACH_KSEGRP_IN_PROC(p, kg) {
536 if (PRI_IS_REALTIME(kg->kg_pri.pri_class)) {
537 mtx_unlock_spin(&sched_lock);
538 PROC_UNLOCK(p);
539 goto nextproc;
540 }
541
542 /*
543 * Do not swapout a process waiting
544 * on a critical event of some kind.
545 * Also guarantee swap_idle_threshold1
546 * time in memory.
547 */
548 if (((kg->kg_pri.pri_level) < PSOCK) ||
549 (kg->kg_slptime < swap_idle_threshold1)) {
550 mtx_unlock_spin(&sched_lock);
551 PROC_UNLOCK(p);
552 goto nextproc;
553 }
554
555 /*
556 * If the system is under memory stress,
557 * or if we are swapping
558 * idle processes >= swap_idle_threshold2,
559 * then swap the process out.
560 */
561 if (((action & VM_SWAP_NORMAL) == 0) &&
562 (((action & VM_SWAP_IDLE) == 0) ||
563 (kg->kg_slptime < swap_idle_threshold2))) {
564 mtx_unlock_spin(&sched_lock);
565 PROC_UNLOCK(p);
566 goto nextproc;
567 }
568 if (minslptime > kg->kg_slptime)
569 minslptime = kg->kg_slptime;
570 }
571
572 mtx_unlock_spin(&sched_lock);
573 ++vm->vm_refcnt;
574 /*
575 * do not swapout a process that
576 * is waiting for VM
577 * data structures there is a
578 * possible deadlock.
579 */
580 if (lockmgr(&vm->vm_map.lock,
581 LK_EXCLUSIVE | LK_NOWAIT,
582 NULL, curthread)) {
583 vmspace_free(vm);
584 PROC_UNLOCK(p);
585 goto nextproc;
586 }
587 vm_map_unlock(&vm->vm_map);
588 /*
589 * If the process has been asleep for awhile and had
590 * most of its pages taken away already, swap it out.
591 */
592 if ((action & VM_SWAP_NORMAL) ||
593 ((action & VM_SWAP_IDLE) &&
594 (minslptime > swap_idle_threshold2))) {
595 sx_sunlock(&allproc_lock);
596 swapout(p);
597 vmspace_free(vm);
598 didswap++;
599 goto retry;
600 }
601 PROC_UNLOCK(p);
602 vmspace_free(vm);
603 }
604nextproc:
605 }
606 sx_sunlock(&allproc_lock);
607 /*
608 * If we swapped something out, and another process needed memory,
609 * then wakeup the sched process.
610 */
611 if (didswap)
612 wakeup(&proc0);
613}
614
615static void
616swapout(p)
617 struct proc *p;
618{
619 struct thread *td;
620
621 PROC_LOCK_ASSERT(p, MA_OWNED);
622#if defined(SWAP_DEBUG)
623 printf("swapping out %d\n", p->p_pid);
624#endif
625 ++p->p_stats->p_ru.ru_nswap;
626 /*
627 * remember the process resident count
628 */
629 p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
630
631 mtx_lock_spin(&sched_lock);
632 p->p_sflag &= ~PS_INMEM;
633 p->p_sflag |= PS_SWAPPING;
634 PROC_UNLOCK_NOSWITCH(p);
635 FOREACH_THREAD_IN_PROC (p, td)
636 if (td->td_proc->p_stat == SRUN) /* XXXKSE */
637 remrunqueue(td); /* XXXKSE */
638 mtx_unlock_spin(&sched_lock);
639
640 pmap_swapout_proc(p);
641 FOREACH_THREAD_IN_PROC(p, td)
642 pmap_swapout_thread(td);
643
644 mtx_lock_spin(&sched_lock);
645 p->p_sflag &= ~PS_SWAPPING;
646 p->p_swtime = 0;
647 mtx_unlock_spin(&sched_lock);
648}
649#endif /* !NO_SWAPPING */