xref: /freebsd/sys/vm/vm_glue.c (revision 23f282aa31e9b6fceacd449020e936e98d6f2298)
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/shm.h>
73 #include <sys/vmmeter.h>
74 #include <sys/sysctl.h>
75 
76 #include <sys/kernel.h>
77 #include <sys/unistd.h>
78 
79 #include <machine/limits.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <sys/lock.h>
84 #include <vm/pmap.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>
90 
91 #include <sys/user.h>
92 
93 /*
94  * System initialization
95  *
96  * Note: proc0 from proc.h
97  */
98 
99 static void vm_init_limits __P((void *));
100 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
101 
102 /*
103  * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
104  *
105  * Note: run scheduling should be divorced from the vm system.
106  */
107 static void scheduler __P((void *));
108 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
109 
110 
111 static void swapout __P((struct proc *));
112 
113 int
114 kernacc(addr, len, rw)
115 	caddr_t addr;
116 	int len, rw;
117 {
118 	boolean_t rv;
119 	vm_offset_t saddr, eaddr;
120 	vm_prot_t prot;
121 
122 	KASSERT((rw & (~VM_PROT_ALL)) == 0,
123 	    ("illegal ``rw'' argument to kernacc (%x)\n", rw));
124 	prot = 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);
130 	return (rv == TRUE);
131 }
132 
133 int
134 useracc(addr, len, rw)
135 	caddr_t addr;
136 	int len, rw;
137 {
138 	boolean_t rv;
139 	vm_prot_t prot;
140 	vm_map_t map;
141 	vm_map_entry_t save_hint;
142 
143 	KASSERT((rw & (~VM_PROT_ALL)) == 0,
144 	    ("illegal ``rw'' argument to useracc (%x)\n", rw));
145 	prot = rw;
146 	/*
147 	 * XXX - check separately to disallow access to user area and user
148 	 * page tables - they are in the map.
149 	 *
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.
154 	 */
155 	if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
156 	    || (vm_offset_t) addr + len < (vm_offset_t) addr) {
157 		return (FALSE);
158 	}
159 	map = &curproc->p_vmspace->vm_map;
160 	vm_map_lock_read(map);
161 	/*
162 	 * We save the map hint, and restore it.  Useracc appears to distort
163 	 * the map hint unnecessarily.
164 	 */
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);
170 
171 	return (rv == TRUE);
172 }
173 
174 void
175 vslock(addr, len)
176 	caddr_t addr;
177 	u_int len;
178 {
179 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
180 	    round_page((vm_offset_t)addr + len), FALSE);
181 }
182 
183 void
184 vsunlock(addr, len)
185 	caddr_t addr;
186 	u_int len;
187 {
188 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
189 	    round_page((vm_offset_t)addr + len), TRUE);
190 }
191 
192 /*
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.
199  */
200 void
201 vm_fork(p1, p2, flags)
202 	register struct proc *p1, *p2;
203 	int flags;
204 {
205 	register struct user *up;
206 
207 	if ((flags & RFPROC) == 0) {
208 		/*
209 		 * Divorce the memory, if it is shared, essentially
210 		 * this changes shared memory amongst threads, into
211 		 * COW locally.
212 		 */
213 		if ((flags & RFMEM) == 0) {
214 			if (p1->p_vmspace->vm_refcnt > 1) {
215 				vmspace_unshare(p1);
216 			}
217 		}
218 		cpu_fork(p1, p2, flags);
219 		return;
220 	}
221 
222 	if (flags & RFMEM) {
223 		p2->p_vmspace = p1->p_vmspace;
224 		p1->p_vmspace->vm_refcnt++;
225 	}
226 
227 	while (vm_page_count_severe()) {
228 		VM_WAIT;
229 	}
230 
231 	if ((flags & RFMEM) == 0) {
232 		p2->p_vmspace = vmspace_fork(p1->p_vmspace);
233 
234 		pmap_pinit2(vmspace_pmap(p2->p_vmspace));
235 
236 		if (p1->p_vmspace->vm_shm)
237 			shmfork(p1, p2);
238 	}
239 
240 	pmap_new_proc(p2);
241 
242 	up = p2->p_addr;
243 
244 	/*
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).
248 	 *
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.
251 	 */
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;
258 	}
259 
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));
266 
267 
268 	/*
269 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
270 	 * and make the child ready to run.
271 	 */
272 	cpu_fork(p1, p2, flags);
273 }
274 
275 /*
276  * Set default limits for VM system.
277  * Called for proc 0, and then inherited by all others.
278  *
279  * XXX should probably act directly on proc0.
280  */
281 static void
282 vm_init_limits(udata)
283 	void *udata;
284 {
285 	register struct proc *p = udata;
286 	int rss_limit;
287 
288 	/*
289 	 * Set up the initial limits on process VM. Set the maximum resident
290 	 * set size to be half of (reasonably) available memory.  Since this
291 	 * is a soft limit, it comes into effect only when the system is out
292 	 * of memory - half of main memory helps to favor smaller processes,
293 	 * and reduces thrashing of the object cache.
294 	 */
295 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
296 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
297 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
298 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
299 	/* limit the limit to no less than 2MB */
300 	rss_limit = max(cnt.v_free_count, 512);
301 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
302 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
303 }
304 
305 void
306 faultin(p)
307 	struct proc *p;
308 {
309 	int s;
310 
311 	if ((p->p_flag & P_INMEM) == 0) {
312 
313 		++p->p_lock;
314 
315 		pmap_swapin_proc(p);
316 
317 		s = splhigh();
318 
319 		if (p->p_stat == SRUN)
320 			setrunqueue(p);
321 
322 		p->p_flag |= P_INMEM;
323 
324 		/* undo the effect of setting SLOCK above */
325 		--p->p_lock;
326 		splx(s);
327 
328 	}
329 }
330 
331 /*
332  * This swapin algorithm attempts to swap-in processes only if there
333  * is enough space for them.  Of course, if a process waits for a long
334  * time, it will be swapped in anyway.
335  */
336 /* ARGSUSED*/
337 static void
338 scheduler(dummy)
339 	void *dummy;
340 {
341 	register struct proc *p;
342 	register int pri;
343 	struct proc *pp;
344 	int ppri;
345 
346 loop:
347 	if (vm_page_count_min()) {
348 		VM_WAIT;
349 		goto loop;
350 	}
351 
352 	pp = NULL;
353 	ppri = INT_MIN;
354 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
355 		if (p->p_stat == SRUN &&
356 			(p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
357 
358 			pri = p->p_swtime + p->p_slptime;
359 			if ((p->p_flag & P_SWAPINREQ) == 0) {
360 				pri -= p->p_nice * 8;
361 			}
362 
363 			/*
364 			 * if this process is higher priority and there is
365 			 * enough space, then select this process instead of
366 			 * the previous selection.
367 			 */
368 			if (pri > ppri) {
369 				pp = p;
370 				ppri = pri;
371 			}
372 		}
373 	}
374 
375 	/*
376 	 * Nothing to do, back to sleep.
377 	 */
378 	if ((p = pp) == NULL) {
379 		tsleep(&proc0, PVM, "sched", 0);
380 		goto loop;
381 	}
382 	p->p_flag &= ~P_SWAPINREQ;
383 
384 	/*
385 	 * We would like to bring someone in. (only if there is space).
386 	 */
387 	faultin(p);
388 	p->p_swtime = 0;
389 	goto loop;
390 }
391 
392 #ifndef NO_SWAPPING
393 
394 #define	swappable(p) \
395 	(((p)->p_lock == 0) && \
396 		((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM)
397 
398 
399 /*
400  * Swap_idle_threshold1 is the guaranteed swapped in time for a process
401  */
402 static int swap_idle_threshold1 = 2;
403 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
404 	CTLFLAG_RW, &swap_idle_threshold1, 0, "");
405 
406 /*
407  * Swap_idle_threshold2 is the time that a process can be idle before
408  * it will be swapped out, if idle swapping is enabled.
409  */
410 static int swap_idle_threshold2 = 10;
411 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
412 	CTLFLAG_RW, &swap_idle_threshold2, 0, "");
413 
414 /*
415  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
416  * procs and unwire their u-areas.  We try to always "swap" at least one
417  * process in case we need the room for a swapin.
418  * If any procs have been sleeping/stopped for at least maxslp seconds,
419  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
420  * if any, otherwise the longest-resident process.
421  */
422 void
423 swapout_procs(action)
424 int action;
425 {
426 	register struct proc *p;
427 	struct proc *outp, *outp2;
428 	int outpri, outpri2;
429 	int didswap = 0;
430 
431 	outp = outp2 = NULL;
432 	outpri = outpri2 = INT_MIN;
433 retry:
434 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
435 		struct vmspace *vm;
436 		if (!swappable(p))
437 			continue;
438 
439 		vm = p->p_vmspace;
440 
441 		switch (p->p_stat) {
442 		default:
443 			continue;
444 
445 		case SSLEEP:
446 		case SSTOP:
447 			/*
448 			 * do not swapout a realtime process
449 			 */
450 			if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type))
451 				continue;
452 
453 			/*
454 			 * Do not swapout a process waiting on a critical
455 			 * event of some kind.  Also guarantee swap_idle_threshold1
456 			 * time in memory.
457 			 */
458 			if (((p->p_priority & 0x7f) < PSOCK) ||
459 				(p->p_slptime < swap_idle_threshold1))
460 				continue;
461 
462 			/*
463 			 * If the system is under memory stress, or if we are swapping
464 			 * idle processes >= swap_idle_threshold2, then swap the process
465 			 * out.
466 			 */
467 			if (((action & VM_SWAP_NORMAL) == 0) &&
468 				(((action & VM_SWAP_IDLE) == 0) ||
469 				  (p->p_slptime < swap_idle_threshold2)))
470 				continue;
471 
472 			++vm->vm_refcnt;
473 			/*
474 			 * do not swapout a process that is waiting for VM
475 			 * data structures there is a possible deadlock.
476 			 */
477 			if (lockmgr(&vm->vm_map.lock,
478 					LK_EXCLUSIVE | LK_NOWAIT,
479 					(void *)0, curproc)) {
480 				vmspace_free(vm);
481 				continue;
482 			}
483 			vm_map_unlock(&vm->vm_map);
484 			/*
485 			 * If the process has been asleep for awhile and had
486 			 * most of its pages taken away already, swap it out.
487 			 */
488 			if ((action & VM_SWAP_NORMAL) ||
489 				((action & VM_SWAP_IDLE) &&
490 				 (p->p_slptime > swap_idle_threshold2))) {
491 				swapout(p);
492 				vmspace_free(vm);
493 				didswap++;
494 				goto retry;
495 			}
496 		}
497 	}
498 	/*
499 	 * If we swapped something out, and another process needed memory,
500 	 * then wakeup the sched process.
501 	 */
502 	if (didswap)
503 		wakeup(&proc0);
504 }
505 
506 static void
507 swapout(p)
508 	register struct proc *p;
509 {
510 
511 #if defined(SWAP_DEBUG)
512 	printf("swapping out %d\n", p->p_pid);
513 #endif
514 	++p->p_stats->p_ru.ru_nswap;
515 	/*
516 	 * remember the process resident count
517 	 */
518 	p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
519 
520 	(void) splhigh();
521 	p->p_flag &= ~P_INMEM;
522 	p->p_flag |= P_SWAPPING;
523 	if (p->p_stat == SRUN)
524 		remrunqueue(p);
525 	(void) spl0();
526 
527 	pmap_swapout_proc(p);
528 
529 	p->p_flag &= ~P_SWAPPING;
530 	p->p_swtime = 0;
531 }
532 #endif /* !NO_SWAPPING */
533