xref: /freebsd/sys/vm/vm_glue.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
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  * $FreeBSD$
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/proc.h>
71 #include <sys/resourcevar.h>
72 #include <sys/buf.h>
73 #include <sys/shm.h>
74 #include <sys/vmmeter.h>
75 #include <sys/sysctl.h>
76 
77 #include <sys/kernel.h>
78 #include <sys/unistd.h>
79 
80 #include <machine/limits.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <sys/lock.h>
85 #include <vm/pmap.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>
91 
92 #include <sys/user.h>
93 
94 /*
95  * System initialization
96  *
97  * Note: proc0 from proc.h
98  */
99 
100 static void vm_init_limits __P((void *));
101 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
102 
103 /*
104  * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
105  *
106  * Note: run scheduling should be divorced from the vm system.
107  */
108 static void scheduler __P((void *));
109 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
110 
111 
112 static void swapout __P((struct proc *));
113 
114 int
115 kernacc(addr, len, rw)
116 	caddr_t addr;
117 	int len, rw;
118 {
119 	boolean_t rv;
120 	vm_offset_t saddr, eaddr;
121 	vm_prot_t prot;
122 
123 	KASSERT((rw & (~VM_PROT_ALL)) == 0,
124 	    ("illegal ``rw'' argument to kernacc (%x)\n", rw));
125 	prot = 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);
131 	return (rv == TRUE);
132 }
133 
134 int
135 useracc(addr, len, rw)
136 	caddr_t addr;
137 	int len, rw;
138 {
139 	boolean_t rv;
140 	vm_prot_t prot;
141 	vm_map_t map;
142 	vm_map_entry_t save_hint;
143 
144 	KASSERT((rw & (~VM_PROT_ALL)) == 0,
145 	    ("illegal ``rw'' argument to useracc (%x)\n", rw));
146 	prot = rw;
147 	/*
148 	 * XXX - check separately to disallow access to user area and user
149 	 * page tables - they are in the map.
150 	 *
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.
155 	 */
156 	if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
157 	    || (vm_offset_t) addr + len < (vm_offset_t) addr) {
158 		return (FALSE);
159 	}
160 	map = &curproc->p_vmspace->vm_map;
161 	vm_map_lock_read(map);
162 	/*
163 	 * We save the map hint, and restore it.  Useracc appears to distort
164 	 * the map hint unnecessarily.
165 	 */
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);
171 
172 	return (rv == TRUE);
173 }
174 
175 void
176 vslock(addr, len)
177 	caddr_t addr;
178 	u_int len;
179 {
180 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
181 	    round_page((vm_offset_t)addr + len), FALSE);
182 }
183 
184 void
185 vsunlock(addr, len, dirtied)
186 	caddr_t addr;
187 	u_int len;
188 	int dirtied;
189 {
190 #ifdef	lint
191 	dirtied++;
192 #endif	/* lint */
193 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
194 	    round_page((vm_offset_t)addr + len), TRUE);
195 }
196 
197 /*
198  * Implement fork's actions on an address space.
199  * Here we arrange for the address space to be copied or referenced,
200  * allocate a user struct (pcb and kernel stack), then call the
201  * machine-dependent layer to fill those in and make the new process
202  * ready to run.  The new process is set up so that it returns directly
203  * to user mode to avoid stack copying and relocation problems.
204  */
205 void
206 vm_fork(p1, p2, flags)
207 	register struct proc *p1, *p2;
208 	int flags;
209 {
210 	register struct user *up;
211 
212 	if ((flags & RFPROC) == 0) {
213 		/*
214 		 * Divorce the memory, if it is shared, essentially
215 		 * this changes shared memory amongst threads, into
216 		 * COW locally.
217 		 */
218 		if ((flags & RFMEM) == 0) {
219 			if (p1->p_vmspace->vm_refcnt > 1) {
220 				vmspace_unshare(p1);
221 			}
222 		}
223 		cpu_fork(p1, p2, flags);
224 		return;
225 	}
226 
227 	if (flags & RFMEM) {
228 		p2->p_vmspace = p1->p_vmspace;
229 		p1->p_vmspace->vm_refcnt++;
230 	}
231 
232 	while (vm_page_count_severe()) {
233 		VM_WAIT;
234 	}
235 
236 	if ((flags & RFMEM) == 0) {
237 		p2->p_vmspace = vmspace_fork(p1->p_vmspace);
238 
239 		pmap_pinit2(vmspace_pmap(p2->p_vmspace));
240 
241 		if (p1->p_vmspace->vm_shm)
242 			shmfork(p1, p2);
243 	}
244 
245 	pmap_new_proc(p2);
246 
247 	up = p2->p_addr;
248 
249 	/*
250 	 * p_stats currently points at fields in the user struct
251 	 * but not at &u, instead at p_addr. Copy parts of
252 	 * p_stats; zero the rest of p_stats (statistics).
253 	 *
254 	 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need
255 	 * to share sigacts, so we use the up->u_sigacts.
256 	 */
257 	p2->p_stats = &up->u_stats;
258 	if (p2->p_sigacts == NULL) {
259 		if (p2->p_procsig->ps_refcnt != 1)
260 			printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid);
261 		p2->p_sigacts = &up->u_sigacts;
262 		up->u_sigacts = *p1->p_sigacts;
263 	}
264 
265 	bzero(&up->u_stats.pstat_startzero,
266 	    (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
267 		(caddr_t) &up->u_stats.pstat_startzero));
268 	bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
269 	    ((caddr_t) &up->u_stats.pstat_endcopy -
270 		(caddr_t) &up->u_stats.pstat_startcopy));
271 
272 
273 	/*
274 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
275 	 * and make the child ready to run.
276 	 */
277 	cpu_fork(p1, p2, flags);
278 }
279 
280 /*
281  * Set default limits for VM system.
282  * Called for proc 0, and then inherited by all others.
283  *
284  * XXX should probably act directly on proc0.
285  */
286 static void
287 vm_init_limits(udata)
288 	void *udata;
289 {
290 	register struct proc *p = udata;
291 	int rss_limit;
292 
293 	/*
294 	 * Set up the initial limits on process VM. Set the maximum resident
295 	 * set size to be half of (reasonably) available memory.  Since this
296 	 * is a soft limit, it comes into effect only when the system is out
297 	 * of memory - half of main memory helps to favor smaller processes,
298 	 * and reduces thrashing of the object cache.
299 	 */
300 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
301 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
302 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
303 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
304 	/* limit the limit to no less than 2MB */
305 	rss_limit = max(cnt.v_free_count, 512);
306 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
307 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
308 }
309 
310 void
311 faultin(p)
312 	struct proc *p;
313 {
314 	int s;
315 
316 	if ((p->p_flag & P_INMEM) == 0) {
317 
318 		++p->p_lock;
319 
320 		pmap_swapin_proc(p);
321 
322 		s = splhigh();
323 
324 		if (p->p_stat == SRUN)
325 			setrunqueue(p);
326 
327 		p->p_flag |= P_INMEM;
328 
329 		/* undo the effect of setting SLOCK above */
330 		--p->p_lock;
331 		splx(s);
332 
333 	}
334 }
335 
336 /*
337  * This swapin algorithm attempts to swap-in processes only if there
338  * is enough space for them.  Of course, if a process waits for a long
339  * time, it will be swapped in anyway.
340  */
341 /* ARGSUSED*/
342 static void
343 scheduler(dummy)
344 	void *dummy;
345 {
346 	register struct proc *p;
347 	register int pri;
348 	struct proc *pp;
349 	int ppri;
350 
351 loop:
352 	if (vm_page_count_min()) {
353 		VM_WAIT;
354 		goto loop;
355 	}
356 
357 	pp = NULL;
358 	ppri = INT_MIN;
359 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
360 		if (p->p_stat == SRUN &&
361 			(p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
362 
363 			pri = p->p_swtime + p->p_slptime;
364 			if ((p->p_flag & P_SWAPINREQ) == 0) {
365 				pri -= p->p_nice * 8;
366 			}
367 
368 			/*
369 			 * if this process is higher priority and there is
370 			 * enough space, then select this process instead of
371 			 * the previous selection.
372 			 */
373 			if (pri > ppri) {
374 				pp = p;
375 				ppri = pri;
376 			}
377 		}
378 	}
379 
380 	/*
381 	 * Nothing to do, back to sleep.
382 	 */
383 	if ((p = pp) == NULL) {
384 		tsleep(&proc0, PVM, "sched", 0);
385 		goto loop;
386 	}
387 	p->p_flag &= ~P_SWAPINREQ;
388 
389 	/*
390 	 * We would like to bring someone in. (only if there is space).
391 	 */
392 	faultin(p);
393 	p->p_swtime = 0;
394 	goto loop;
395 }
396 
397 #ifndef NO_SWAPPING
398 
399 #define	swappable(p) \
400 	(((p)->p_lock == 0) && \
401 		((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM)
402 
403 
404 /*
405  * Swap_idle_threshold1 is the guaranteed swapped in time for a process
406  */
407 static int swap_idle_threshold1 = 2;
408 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
409 	CTLFLAG_RW, &swap_idle_threshold1, 0, "");
410 
411 /*
412  * Swap_idle_threshold2 is the time that a process can be idle before
413  * it will be swapped out, if idle swapping is enabled.
414  */
415 static int swap_idle_threshold2 = 10;
416 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
417 	CTLFLAG_RW, &swap_idle_threshold2, 0, "");
418 
419 /*
420  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
421  * procs and unwire their u-areas.  We try to always "swap" at least one
422  * process in case we need the room for a swapin.
423  * If any procs have been sleeping/stopped for at least maxslp seconds,
424  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
425  * if any, otherwise the longest-resident process.
426  */
427 void
428 swapout_procs(action)
429 int action;
430 {
431 	register struct proc *p;
432 	struct proc *outp, *outp2;
433 	int outpri, outpri2;
434 	int didswap = 0;
435 
436 	outp = outp2 = NULL;
437 	outpri = outpri2 = INT_MIN;
438 retry:
439 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
440 		struct vmspace *vm;
441 		if (!swappable(p))
442 			continue;
443 
444 		vm = p->p_vmspace;
445 
446 		switch (p->p_stat) {
447 		default:
448 			continue;
449 
450 		case SSLEEP:
451 		case SSTOP:
452 			/*
453 			 * do not swapout a realtime process
454 			 */
455 			if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type))
456 				continue;
457 
458 			/*
459 			 * Do not swapout a process waiting on a critical
460 			 * event of some kind.  Also guarantee swap_idle_threshold1
461 			 * time in memory.
462 			 */
463 			if (((p->p_priority & 0x7f) < PSOCK) ||
464 				(p->p_slptime < swap_idle_threshold1))
465 				continue;
466 
467 			/*
468 			 * If the system is under memory stress, or if we are swapping
469 			 * idle processes >= swap_idle_threshold2, then swap the process
470 			 * out.
471 			 */
472 			if (((action & VM_SWAP_NORMAL) == 0) &&
473 				(((action & VM_SWAP_IDLE) == 0) ||
474 				  (p->p_slptime < swap_idle_threshold2)))
475 				continue;
476 
477 			++vm->vm_refcnt;
478 			/*
479 			 * do not swapout a process that is waiting for VM
480 			 * data structures there is a possible deadlock.
481 			 */
482 			if (lockmgr(&vm->vm_map.lock,
483 					LK_EXCLUSIVE | LK_NOWAIT,
484 					(void *)0, curproc)) {
485 				vmspace_free(vm);
486 				continue;
487 			}
488 			vm_map_unlock(&vm->vm_map);
489 			/*
490 			 * If the process has been asleep for awhile and had
491 			 * most of its pages taken away already, swap it out.
492 			 */
493 			if ((action & VM_SWAP_NORMAL) ||
494 				((action & VM_SWAP_IDLE) &&
495 				 (p->p_slptime > swap_idle_threshold2))) {
496 				swapout(p);
497 				vmspace_free(vm);
498 				didswap++;
499 				goto retry;
500 			}
501 		}
502 	}
503 	/*
504 	 * If we swapped something out, and another process needed memory,
505 	 * then wakeup the sched process.
506 	 */
507 	if (didswap)
508 		wakeup(&proc0);
509 }
510 
511 static void
512 swapout(p)
513 	register struct proc *p;
514 {
515 
516 #if defined(SWAP_DEBUG)
517 	printf("swapping out %d\n", p->p_pid);
518 #endif
519 	++p->p_stats->p_ru.ru_nswap;
520 	/*
521 	 * remember the process resident count
522 	 */
523 	p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
524 
525 	(void) splhigh();
526 	p->p_flag &= ~P_INMEM;
527 	p->p_flag |= P_SWAPPING;
528 	if (p->p_stat == SRUN)
529 		remrunqueue(p);
530 	(void) spl0();
531 
532 	pmap_swapout_proc(p);
533 
534 	p->p_flag &= ~P_SWAPPING;
535 	p->p_swtime = 0;
536 }
537 #endif /* !NO_SWAPPING */
538