xref: /freebsd/sys/vm/vm_glue.c (revision 817420dc8eac7df799c78f5309b75092b7f7cd40)
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/ktr.h>
78 #include <sys/mutex.h>
79 #include <sys/unistd.h>
80 
81 #include <machine/limits.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_param.h>
85 #include <sys/lock.h>
86 #include <vm/pmap.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>
92 
93 #include <sys/user.h>
94 
95 /*
96  * System initialization
97  *
98  * Note: proc0 from proc.h
99  */
100 
101 static void vm_init_limits __P((void *));
102 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
103 
104 /*
105  * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
106  *
107  * Note: run scheduling should be divorced from the vm system.
108  */
109 static void scheduler __P((void *));
110 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
111 
112 
113 static void swapout __P((struct proc *));
114 
115 int
116 kernacc(addr, len, rw)
117 	caddr_t addr;
118 	int len, rw;
119 {
120 	boolean_t rv;
121 	vm_offset_t saddr, eaddr;
122 	vm_prot_t prot;
123 
124 	KASSERT((rw & (~VM_PROT_ALL)) == 0,
125 	    ("illegal ``rw'' argument to kernacc (%x)\n", rw));
126 	prot = rw;
127 	saddr = trunc_page((vm_offset_t)addr);
128 	eaddr = round_page((vm_offset_t)addr + len);
129 	vm_map_lock_read(kernel_map);
130 	rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
131 	vm_map_unlock_read(kernel_map);
132 	return (rv == TRUE);
133 }
134 
135 int
136 useracc(addr, len, rw)
137 	caddr_t addr;
138 	int len, rw;
139 {
140 	boolean_t rv;
141 	vm_prot_t prot;
142 	vm_map_t map;
143 	vm_map_entry_t save_hint;
144 
145 	KASSERT((rw & (~VM_PROT_ALL)) == 0,
146 	    ("illegal ``rw'' argument to useracc (%x)\n", rw));
147 	prot = rw;
148 	/*
149 	 * XXX - check separately to disallow access to user area and user
150 	 * page tables - they are in the map.
151 	 *
152 	 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max.  It was once
153 	 * only used (as an end address) in trap.c.  Use it as an end address
154 	 * here too.  This bogusness has spread.  I just fixed where it was
155 	 * used as a max in vm_mmap.c.
156 	 */
157 	if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
158 	    || (vm_offset_t) addr + len < (vm_offset_t) addr) {
159 		return (FALSE);
160 	}
161 	map = &curproc->p_vmspace->vm_map;
162 	vm_map_lock_read(map);
163 	/*
164 	 * We save the map hint, and restore it.  Useracc appears to distort
165 	 * the map hint unnecessarily.
166 	 */
167 	save_hint = map->hint;
168 	rv = vm_map_check_protection(map,
169 	    trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot);
170 	map->hint = save_hint;
171 	vm_map_unlock_read(map);
172 
173 	return (rv == TRUE);
174 }
175 
176 void
177 vslock(addr, len)
178 	caddr_t addr;
179 	u_int len;
180 {
181 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
182 	    round_page((vm_offset_t)addr + len), FALSE);
183 }
184 
185 void
186 vsunlock(addr, len)
187 	caddr_t addr;
188 	u_int len;
189 {
190 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr),
191 	    round_page((vm_offset_t)addr + len), TRUE);
192 }
193 
194 /*
195  * Implement fork's actions on an address space.
196  * Here we arrange for the address space to be copied or referenced,
197  * allocate a user struct (pcb and kernel stack), then call the
198  * machine-dependent layer to fill those in and make the new process
199  * ready to run.  The new process is set up so that it returns directly
200  * to user mode to avoid stack copying and relocation problems.
201  */
202 void
203 vm_fork(p1, p2, flags)
204 	register struct proc *p1, *p2;
205 	int flags;
206 {
207 	register struct user *up;
208 
209 	if ((flags & RFPROC) == 0) {
210 		/*
211 		 * Divorce the memory, if it is shared, essentially
212 		 * this changes shared memory amongst threads, into
213 		 * COW locally.
214 		 */
215 		if ((flags & RFMEM) == 0) {
216 			if (p1->p_vmspace->vm_refcnt > 1) {
217 				vmspace_unshare(p1);
218 			}
219 		}
220 		cpu_fork(p1, p2, flags);
221 		return;
222 	}
223 
224 	if (flags & RFMEM) {
225 		p2->p_vmspace = p1->p_vmspace;
226 		p1->p_vmspace->vm_refcnt++;
227 	}
228 
229 	while (vm_page_count_severe()) {
230 		VM_WAIT;
231 	}
232 
233 	if ((flags & RFMEM) == 0) {
234 		p2->p_vmspace = vmspace_fork(p1->p_vmspace);
235 
236 		pmap_pinit2(vmspace_pmap(p2->p_vmspace));
237 
238 		if (p1->p_vmspace->vm_shm)
239 			shmfork(p1, p2);
240 	}
241 
242 	pmap_new_proc(p2);
243 
244 	up = p2->p_addr;
245 
246 	/*
247 	 * p_stats currently points at fields in the user struct
248 	 * but not at &u, instead at p_addr. Copy parts of
249 	 * p_stats; zero the rest of p_stats (statistics).
250 	 *
251 	 * If procsig->ps_refcnt is 1 and p2->p_sigacts is NULL we dont' need
252 	 * to share sigacts, so we use the up->u_sigacts.
253 	 */
254 	p2->p_stats = &up->u_stats;
255 	if (p2->p_sigacts == NULL) {
256 		if (p2->p_procsig->ps_refcnt != 1)
257 			printf ("PID:%d NULL sigacts with refcnt not 1!\n",p2->p_pid);
258 		p2->p_sigacts = &up->u_sigacts;
259 		up->u_sigacts = *p1->p_sigacts;
260 	}
261 
262 	bzero(&up->u_stats.pstat_startzero,
263 	    (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
264 		(caddr_t) &up->u_stats.pstat_startzero));
265 	bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
266 	    ((caddr_t) &up->u_stats.pstat_endcopy -
267 		(caddr_t) &up->u_stats.pstat_startcopy));
268 
269 
270 	/*
271 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
272 	 * and make the child ready to run.
273 	 */
274 	cpu_fork(p1, p2, flags);
275 }
276 
277 /*
278  * Set default limits for VM system.
279  * Called for proc 0, and then inherited by all others.
280  *
281  * XXX should probably act directly on proc0.
282  */
283 static void
284 vm_init_limits(udata)
285 	void *udata;
286 {
287 	register struct proc *p = udata;
288 	int rss_limit;
289 
290 	/*
291 	 * Set up the initial limits on process VM. Set the maximum resident
292 	 * set size to be half of (reasonably) available memory.  Since this
293 	 * is a soft limit, it comes into effect only when the system is out
294 	 * of memory - half of main memory helps to favor smaller processes,
295 	 * and reduces thrashing of the object cache.
296 	 */
297 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
298 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
299 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
300 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
301 	/* limit the limit to no less than 2MB */
302 	rss_limit = max(cnt.v_free_count, 512);
303 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
304 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
305 }
306 
307 void
308 faultin(p)
309 	struct proc *p;
310 {
311 	int s;
312 
313 	if ((p->p_flag & P_INMEM) == 0) {
314 
315 		++p->p_lock;
316 
317 		pmap_swapin_proc(p);
318 
319 		s = splhigh();
320 
321 		mtx_enter(&sched_lock, MTX_SPIN);
322 		if (p->p_stat == SRUN) {
323 			setrunqueue(p);
324 		}
325 		mtx_exit(&sched_lock, MTX_SPIN);
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  * Giant is still held at this point, to be released in tsleep.
342  */
343 /* ARGSUSED*/
344 static void
345 scheduler(dummy)
346 	void *dummy;
347 {
348 	register struct proc *p;
349 	register int pri;
350 	struct proc *pp;
351 	int ppri;
352 
353 	mtx_assert(&Giant, MA_OWNED);
354 
355 loop:
356 	if (vm_page_count_min()) {
357 		VM_WAIT;
358 		goto loop;
359 	}
360 
361 	pp = NULL;
362 	ppri = INT_MIN;
363 	lockmgr(&allproc_lock, LK_SHARED, NULL, CURPROC);
364 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
365 		if (p->p_stat == SRUN &&
366 			(p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
367 
368 			pri = p->p_swtime + p->p_slptime;
369 			if ((p->p_flag & P_SWAPINREQ) == 0) {
370 				pri -= p->p_nice * 8;
371 			}
372 
373 			/*
374 			 * if this process is higher priority and there is
375 			 * enough space, then select this process instead of
376 			 * the previous selection.
377 			 */
378 			if (pri > ppri) {
379 				pp = p;
380 				ppri = pri;
381 			}
382 		}
383 	}
384 	lockmgr(&allproc_lock, LK_RELEASE, NULL, CURPROC);
385 
386 	/*
387 	 * Nothing to do, back to sleep.
388 	 */
389 	if ((p = pp) == NULL) {
390 		tsleep(&proc0, PVM, "sched", 0);
391 		goto loop;
392 	}
393 	p->p_flag &= ~P_SWAPINREQ;
394 
395 	/*
396 	 * We would like to bring someone in. (only if there is space).
397 	 */
398 	faultin(p);
399 	p->p_swtime = 0;
400 	goto loop;
401 }
402 
403 #ifndef NO_SWAPPING
404 
405 #define	swappable(p) \
406 	(((p)->p_lock == 0) && \
407 		((p)->p_flag & (P_TRACED|P_SYSTEM|P_INMEM|P_WEXIT|P_SWAPPING)) == P_INMEM)
408 
409 
410 /*
411  * Swap_idle_threshold1 is the guaranteed swapped in time for a process
412  */
413 static int swap_idle_threshold1 = 2;
414 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold1,
415 	CTLFLAG_RW, &swap_idle_threshold1, 0, "");
416 
417 /*
418  * Swap_idle_threshold2 is the time that a process can be idle before
419  * it will be swapped out, if idle swapping is enabled.
420  */
421 static int swap_idle_threshold2 = 10;
422 SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2,
423 	CTLFLAG_RW, &swap_idle_threshold2, 0, "");
424 
425 /*
426  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
427  * procs and unwire their u-areas.  We try to always "swap" at least one
428  * process in case we need the room for a swapin.
429  * If any procs have been sleeping/stopped for at least maxslp seconds,
430  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
431  * if any, otherwise the longest-resident process.
432  */
433 void
434 swapout_procs(action)
435 int action;
436 {
437 	register struct proc *p;
438 	struct proc *outp, *outp2;
439 	int outpri, outpri2;
440 	int didswap = 0;
441 
442 	outp = outp2 = NULL;
443 	outpri = outpri2 = INT_MIN;
444 	lockmgr(&allproc_lock, LK_SHARED, NULL, CURPROC);
445 retry:
446 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
447 		struct vmspace *vm;
448 		if (!swappable(p))
449 			continue;
450 
451 		vm = p->p_vmspace;
452 
453 		switch (p->p_stat) {
454 		default:
455 			continue;
456 
457 		case SSLEEP:
458 		case SSTOP:
459 			/*
460 			 * do not swapout a realtime process
461 			 */
462 			if (RTP_PRIO_IS_REALTIME(p->p_rtprio.type))
463 				continue;
464 
465 			/*
466 			 * Do not swapout a process waiting on a critical
467 			 * event of some kind.  Also guarantee swap_idle_threshold1
468 			 * time in memory.
469 			 */
470 			if (((p->p_priority & 0x7f) < PSOCK) ||
471 				(p->p_slptime < swap_idle_threshold1))
472 				continue;
473 
474 			/*
475 			 * If the system is under memory stress, or if we are swapping
476 			 * idle processes >= swap_idle_threshold2, then swap the process
477 			 * out.
478 			 */
479 			if (((action & VM_SWAP_NORMAL) == 0) &&
480 				(((action & VM_SWAP_IDLE) == 0) ||
481 				  (p->p_slptime < swap_idle_threshold2)))
482 				continue;
483 
484 			++vm->vm_refcnt;
485 			/*
486 			 * do not swapout a process that is waiting for VM
487 			 * data structures there is a possible deadlock.
488 			 */
489 			if (lockmgr(&vm->vm_map.lock,
490 					LK_EXCLUSIVE | LK_NOWAIT,
491 					(void *)0, curproc)) {
492 				vmspace_free(vm);
493 				continue;
494 			}
495 			vm_map_unlock(&vm->vm_map);
496 			/*
497 			 * If the process has been asleep for awhile and had
498 			 * most of its pages taken away already, swap it out.
499 			 */
500 			if ((action & VM_SWAP_NORMAL) ||
501 				((action & VM_SWAP_IDLE) &&
502 				 (p->p_slptime > swap_idle_threshold2))) {
503 				swapout(p);
504 				vmspace_free(vm);
505 				didswap++;
506 				goto retry;
507 			}
508 		}
509 	}
510 	lockmgr(&allproc_lock, LK_RELEASE, NULL, CURPROC);
511 	/*
512 	 * If we swapped something out, and another process needed memory,
513 	 * then wakeup the sched process.
514 	 */
515 	if (didswap)
516 		wakeup(&proc0);
517 }
518 
519 static void
520 swapout(p)
521 	register struct proc *p;
522 {
523 
524 #if defined(SWAP_DEBUG)
525 	printf("swapping out %d\n", p->p_pid);
526 #endif
527 	++p->p_stats->p_ru.ru_nswap;
528 	/*
529 	 * remember the process resident count
530 	 */
531 	p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
532 
533 	(void) splhigh();
534 	p->p_flag &= ~P_INMEM;
535 	p->p_flag |= P_SWAPPING;
536 	if (p->p_stat == SRUN)
537 		remrunqueue(p);
538 	(void) spl0();
539 
540 	pmap_swapout_proc(p);
541 
542 	p->p_flag &= ~P_SWAPPING;
543 	p->p_swtime = 0;
544 }
545 #endif /* !NO_SWAPPING */
546