xref: /freebsd/sys/vm/vm_glue.c (revision a14a0223ae1b172e96dd2a1d849e22026a98b692)
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 & RFMEM) {
213 		p2->p_vmspace = p1->p_vmspace;
214 		p1->p_vmspace->vm_refcnt++;
215 	}
216 
217 	while (vm_page_count_severe()) {
218 		VM_WAIT;
219 	}
220 
221 	if ((flags & RFMEM) == 0) {
222 		p2->p_vmspace = vmspace_fork(p1->p_vmspace);
223 
224 		pmap_pinit2(vmspace_pmap(p2->p_vmspace));
225 
226 		if (p1->p_vmspace->vm_shm)
227 			shmfork(p1, p2);
228 	}
229 
230 	pmap_new_proc(p2);
231 
232 	up = p2->p_addr;
233 
234 	/*
235 	 * p_stats currently points at fields in the user struct
236 	 * but not at &u, instead at p_addr. Copy parts of
237 	 * p_stats; zero the rest of p_stats (statistics).
238 	 *
239 	 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need
240 	 * to share sigacts, so we use the up->u_sigacts.
241 	 */
242 	p2->p_stats = &up->u_stats;
243 	if (p2->p_sigacts == NULL) {
244 		if (p2->p_procsig->ps_refcnt != 1)
245 			printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid);
246 		p2->p_sigacts = &up->u_sigacts;
247 		up->u_sigacts = *p1->p_sigacts;
248 	}
249 
250 	bzero(&up->u_stats.pstat_startzero,
251 	    (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
252 		(caddr_t) &up->u_stats.pstat_startzero));
253 	bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
254 	    ((caddr_t) &up->u_stats.pstat_endcopy -
255 		(caddr_t) &up->u_stats.pstat_startcopy));
256 
257 
258 	/*
259 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
260 	 * and make the child ready to run.
261 	 */
262 	cpu_fork(p1, p2);
263 }
264 
265 /*
266  * Set default limits for VM system.
267  * Called for proc 0, and then inherited by all others.
268  *
269  * XXX should probably act directly on proc0.
270  */
271 static void
272 vm_init_limits(udata)
273 	void *udata;
274 {
275 	register struct proc *p = udata;
276 	int rss_limit;
277 
278 	/*
279 	 * Set up the initial limits on process VM. Set the maximum resident
280 	 * set size to be half of (reasonably) available memory.  Since this
281 	 * is a soft limit, it comes into effect only when the system is out
282 	 * of memory - half of main memory helps to favor smaller processes,
283 	 * and reduces thrashing of the object cache.
284 	 */
285 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
286 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
287 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
288 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
289 	/* limit the limit to no less than 2MB */
290 	rss_limit = max(cnt.v_free_count, 512);
291 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
292 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
293 }
294 
295 void
296 faultin(p)
297 	struct proc *p;
298 {
299 	int s;
300 
301 	if ((p->p_flag & P_INMEM) == 0) {
302 
303 		++p->p_lock;
304 
305 		pmap_swapin_proc(p);
306 
307 		s = splhigh();
308 
309 		if (p->p_stat == SRUN)
310 			setrunqueue(p);
311 
312 		p->p_flag |= P_INMEM;
313 
314 		/* undo the effect of setting SLOCK above */
315 		--p->p_lock;
316 		splx(s);
317 
318 	}
319 }
320 
321 /*
322  * This swapin algorithm attempts to swap-in processes only if there
323  * is enough space for them.  Of course, if a process waits for a long
324  * time, it will be swapped in anyway.
325  */
326 /* ARGSUSED*/
327 static void
328 scheduler(dummy)
329 	void *dummy;
330 {
331 	register struct proc *p;
332 	register int pri;
333 	struct proc *pp;
334 	int ppri;
335 
336 loop:
337 	if (vm_page_count_min()) {
338 		VM_WAIT;
339 		goto loop;
340 	}
341 
342 	pp = NULL;
343 	ppri = INT_MIN;
344 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
345 		if (p->p_stat == SRUN &&
346 			(p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
347 
348 			pri = p->p_swtime + p->p_slptime;
349 			if ((p->p_flag & P_SWAPINREQ) == 0) {
350 				pri -= p->p_nice * 8;
351 			}
352 
353 			/*
354 			 * if this process is higher priority and there is
355 			 * enough space, then select this process instead of
356 			 * the previous selection.
357 			 */
358 			if (pri > ppri) {
359 				pp = p;
360 				ppri = pri;
361 			}
362 		}
363 	}
364 
365 	/*
366 	 * Nothing to do, back to sleep.
367 	 */
368 	if ((p = pp) == NULL) {
369 		tsleep(&proc0, PVM, "sched", 0);
370 		goto loop;
371 	}
372 	p->p_flag &= ~P_SWAPINREQ;
373 
374 	/*
375 	 * We would like to bring someone in. (only if there is space).
376 	 */
377 	faultin(p);
378 	p->p_swtime = 0;
379 	goto loop;
380 }
381 
382 #ifndef NO_SWAPPING
383 
384 #define	swappable(p) \
385 	(((p)->p_lock == 0) && \
386 		((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM)
387 
388 
389 /*
390  * Swap_idle_threshold1 is the guaranteed swapped in time for a process
391  */
392 static int swap_idle_threshold1 = 2;
393 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
394 	CTLFLAG_RW, &swap_idle_threshold1, 0, "");
395 
396 /*
397  * Swap_idle_threshold2 is the time that a process can be idle before
398  * it will be swapped out, if idle swapping is enabled.
399  */
400 static int swap_idle_threshold2 = 10;
401 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
402 	CTLFLAG_RW, &swap_idle_threshold2, 0, "");
403 
404 /*
405  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
406  * procs and unwire their u-areas.  We try to always "swap" at least one
407  * process in case we need the room for a swapin.
408  * If any procs have been sleeping/stopped for at least maxslp seconds,
409  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
410  * if any, otherwise the longest-resident process.
411  */
412 void
413 swapout_procs(action)
414 int action;
415 {
416 	register struct proc *p;
417 	struct proc *outp, *outp2;
418 	int outpri, outpri2;
419 	int didswap = 0;
420 
421 	outp = outp2 = NULL;
422 	outpri = outpri2 = INT_MIN;
423 retry:
424 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
425 		struct vmspace *vm;
426 		if (!swappable(p))
427 			continue;
428 
429 		vm = p->p_vmspace;
430 
431 		switch (p->p_stat) {
432 		default:
433 			continue;
434 
435 		case SSLEEP:
436 		case SSTOP:
437 			/*
438 			 * do not swapout a realtime process
439 			 */
440 			if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type))
441 				continue;
442 
443 			/*
444 			 * Do not swapout a process waiting on a critical
445 			 * event of some kind.  Also guarantee swap_idle_threshold1
446 			 * time in memory.
447 			 */
448 			if (((p->p_priority & 0x7f) < PSOCK) ||
449 				(p->p_slptime < swap_idle_threshold1))
450 				continue;
451 
452 			/*
453 			 * If the system is under memory stress, or if we are swapping
454 			 * idle processes >= swap_idle_threshold2, then swap the process
455 			 * out.
456 			 */
457 			if (((action & VM_SWAP_NORMAL) == 0) &&
458 				(((action & VM_SWAP_IDLE) == 0) ||
459 				  (p->p_slptime < swap_idle_threshold2)))
460 				continue;
461 
462 			++vm->vm_refcnt;
463 			/*
464 			 * do not swapout a process that is waiting for VM
465 			 * data structures there is a possible deadlock.
466 			 */
467 			if (lockmgr(&vm->vm_map.lock,
468 					LK_EXCLUSIVE | LK_NOWAIT,
469 					(void *)0, curproc)) {
470 				vmspace_free(vm);
471 				continue;
472 			}
473 			vm_map_unlock(&vm->vm_map);
474 			/*
475 			 * If the process has been asleep for awhile and had
476 			 * most of its pages taken away already, swap it out.
477 			 */
478 			if ((action & VM_SWAP_NORMAL) ||
479 				((action & VM_SWAP_IDLE) &&
480 				 (p->p_slptime > swap_idle_threshold2))) {
481 				swapout(p);
482 				vmspace_free(vm);
483 				didswap++;
484 				goto retry;
485 			}
486 		}
487 	}
488 	/*
489 	 * If we swapped something out, and another process needed memory,
490 	 * then wakeup the sched process.
491 	 */
492 	if (didswap)
493 		wakeup(&proc0);
494 }
495 
496 static void
497 swapout(p)
498 	register struct proc *p;
499 {
500 
501 #if defined(SWAP_DEBUG)
502 	printf("swapping out %d\n", p->p_pid);
503 #endif
504 	++p->p_stats->p_ru.ru_nswap;
505 	/*
506 	 * remember the process resident count
507 	 */
508 	p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
509 
510 	(void) splhigh();
511 	p->p_flag &= ~P_INMEM;
512 	p->p_flag |= P_SWAPPING;
513 	if (p->p_stat == SRUN)
514 		remrunqueue(p);
515 	(void) spl0();
516 
517 	pmap_swapout_proc(p);
518 
519 	p->p_flag &= ~P_SWAPPING;
520 	p->p_swtime = 0;
521 }
522 #endif /* !NO_SWAPPING */
523