xref: /freebsd/sys/vm/vm_glue.c (revision 5ebc7e6281887681c3a348a5a4c902e262ccd656)
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  * $Id: vm_glue.c,v 1.18 1995/03/28 07:58:53 bde Exp $
63  */
64 
65 #include <sys/param.h>
66 #include <sys/systm.h>
67 #include <sys/proc.h>
68 #include <sys/resourcevar.h>
69 #include <sys/buf.h>
70 #include <sys/shm.h>
71 #include <sys/user.h>
72 
73 #include <sys/kernel.h>
74 #include <sys/dkstat.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_pageout.h>
79 #include <vm/vm_kern.h>
80 
81 #include <machine/stdarg.h>
82 #include <machine/cpu.h>
83 
84 extern char kstack[];
85 
86 /* vm_map_t upages_map; */
87 
88 int
89 kernacc(addr, len, rw)
90 	caddr_t addr;
91 	int len, rw;
92 {
93 	boolean_t rv;
94 	vm_offset_t saddr, eaddr;
95 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
96 
97 	saddr = trunc_page(addr);
98 	eaddr = round_page(addr + len);
99 	rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
100 	return (rv == TRUE);
101 }
102 
103 int
104 useracc(addr, len, rw)
105 	caddr_t addr;
106 	int len, rw;
107 {
108 	boolean_t rv;
109 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
110 
111 	/*
112 	 * XXX - check separately to disallow access to user area and user
113 	 * page tables - they are in the map.
114 	 *
115 	 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max.  It was once
116 	 * only used (as an end address) in trap.c.  Use it as an end address
117 	 * here too.  This bogusness has spread.  I just fixed where it was
118 	 * used as a max in vm_mmap.c.
119 	 */
120 	if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
121 	    || (vm_offset_t) addr + len < (vm_offset_t) addr) {
122 		return (FALSE);
123 	}
124 	rv = vm_map_check_protection(&curproc->p_vmspace->vm_map,
125 	    trunc_page(addr), round_page(addr + len), prot);
126 	return (rv == TRUE);
127 }
128 
129 #ifdef KGDB
130 /*
131  * Change protections on kernel pages from addr to addr+len
132  * (presumably so debugger can plant a breakpoint).
133  * All addresses are assumed to reside in the Sysmap,
134  */
135 chgkprot(addr, len, rw)
136 	register caddr_t addr;
137 	int len, rw;
138 {
139 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
140 
141 	vm_map_protect(kernel_map, trunc_page(addr),
142 	    round_page(addr + len), prot, FALSE);
143 }
144 #endif
145 void
146 vslock(addr, len)
147 	caddr_t addr;
148 	u_int len;
149 {
150 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
151 	    round_page(addr + len), FALSE);
152 }
153 
154 void
155 vsunlock(addr, len, dirtied)
156 	caddr_t addr;
157 	u_int len;
158 	int dirtied;
159 {
160 #ifdef	lint
161 	dirtied++;
162 #endif	/* lint */
163 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
164 	    round_page(addr + len), TRUE);
165 }
166 
167 /*
168  * Implement fork's actions on an address space.
169  * Here we arrange for the address space to be copied or referenced,
170  * allocate a user struct (pcb and kernel stack), then call the
171  * machine-dependent layer to fill those in and make the new process
172  * ready to run.
173  * NOTE: the kernel stack may be at a different location in the child
174  * process, and thus addresses of automatic variables may be invalid
175  * after cpu_fork returns in the child process.  We do nothing here
176  * after cpu_fork returns.
177  */
178 int
179 vm_fork(p1, p2, isvfork)
180 	register struct proc *p1, *p2;
181 	int isvfork;
182 {
183 	register struct user *up;
184 	vm_offset_t addr, ptaddr;
185 	int i;
186 	struct vm_map *vp;
187 
188 	while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) {
189 		VM_WAIT;
190 	}
191 
192 	/*
193 	 * avoid copying any of the parent's pagetables or other per-process
194 	 * objects that reside in the map by marking all of them
195 	 * non-inheritable
196 	 */
197 	(void) vm_map_inherit(&p1->p_vmspace->vm_map,
198 	    UPT_MIN_ADDRESS - UPAGES * NBPG, VM_MAX_ADDRESS, VM_INHERIT_NONE);
199 	p2->p_vmspace = vmspace_fork(p1->p_vmspace);
200 
201 #ifdef SYSVSHM
202 	if (p1->p_vmspace->vm_shm)
203 		shmfork(p1, p2, isvfork);
204 #endif
205 
206 	/*
207 	 * Allocate a wired-down (for now) pcb and kernel stack for the
208 	 * process
209 	 */
210 
211 	addr = (vm_offset_t) kstack;
212 
213 	vp = &p2->p_vmspace->vm_map;
214 
215 	/* get new pagetables and kernel stack */
216 	(void) vm_map_find(vp, NULL, 0, &addr, UPT_MAX_ADDRESS - addr, FALSE);
217 
218 	/* force in the page table encompassing the UPAGES */
219 	ptaddr = trunc_page((u_int) vtopte(addr));
220 	vm_map_pageable(vp, ptaddr, ptaddr + NBPG, FALSE);
221 
222 	/* and force in (demand-zero) the UPAGES */
223 	vm_map_pageable(vp, addr, addr + UPAGES * NBPG, FALSE);
224 
225 	/* get a kernel virtual address for the UPAGES for this proc */
226 	up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * NBPG);
227 	if (up == NULL)
228 		panic("vm_fork: u_map allocation failed");
229 
230 	/* and force-map the upages into the kernel pmap */
231 	for (i = 0; i < UPAGES; i++)
232 		pmap_enter(vm_map_pmap(u_map),
233 		    ((vm_offset_t) up) + NBPG * i,
234 		    pmap_extract(vp->pmap, addr + NBPG * i),
235 		    VM_PROT_READ | VM_PROT_WRITE, 1);
236 
237 	p2->p_addr = up;
238 
239 	/*
240 	 * p_stats and p_sigacts currently point at fields in the user struct
241 	 * but not at &u, instead at p_addr. Copy p_sigacts and parts of
242 	 * p_stats; zero the rest of p_stats (statistics).
243 	 */
244 	p2->p_stats = &up->u_stats;
245 	p2->p_sigacts = &up->u_sigacts;
246 	up->u_sigacts = *p1->p_sigacts;
247 	bzero(&up->u_stats.pstat_startzero,
248 	    (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
249 		(caddr_t) &up->u_stats.pstat_startzero));
250 	bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
251 	    ((caddr_t) &up->u_stats.pstat_endcopy -
252 		(caddr_t) &up->u_stats.pstat_startcopy));
253 
254 
255 	/*
256 	 * cpu_fork will copy and update the kernel stack and pcb, and make
257 	 * the child ready to run.  It marks the child so that it can return
258 	 * differently than the parent. It returns twice, once in the parent
259 	 * process and once in the child.
260 	 */
261 	return (cpu_fork(p1, p2));
262 }
263 
264 /*
265  * Set default limits for VM system.
266  * Called for proc 0, and then inherited by all others.
267  */
268 void
269 vm_init_limits(p)
270 	register struct proc *p;
271 {
272 	int rss_limit;
273 
274 	/*
275 	 * Set up the initial limits on process VM. Set the maximum resident
276 	 * set size to be half of (reasonably) available memory.  Since this
277 	 * is a soft limit, it comes into effect only when the system is out
278 	 * of memory - half of main memory helps to favor smaller processes,
279 	 * and reduces thrashing of the object cache.
280 	 */
281 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
282 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
283 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
284 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
285 	/* limit the limit to no less than 2MB */
286 	rss_limit = max(cnt.v_free_count / 2, 512);
287 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
288 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
289 }
290 
291 void
292 faultin(p)
293 	struct proc *p;
294 {
295 	vm_offset_t i;
296 	vm_offset_t ptaddr;
297 	int s;
298 
299 	if ((p->p_flag & P_INMEM) == 0) {
300 		vm_map_t map;
301 
302 		++p->p_lock;
303 
304 		map = &p->p_vmspace->vm_map;
305 		/* force the page table encompassing the kernel stack (upages) */
306 		ptaddr = trunc_page((u_int) vtopte(kstack));
307 		vm_map_pageable(map, ptaddr, ptaddr + NBPG, FALSE);
308 
309 		/* wire in the UPAGES */
310 		vm_map_pageable(map, (vm_offset_t) kstack,
311 		    (vm_offset_t) kstack + UPAGES * NBPG, FALSE);
312 
313 		/* and map them nicely into the kernel pmap */
314 		for (i = 0; i < UPAGES; i++) {
315 			vm_offset_t off = i * NBPG;
316 			vm_offset_t pa = (vm_offset_t)
317 			pmap_extract(&p->p_vmspace->vm_pmap,
318 			    (vm_offset_t) kstack + off);
319 
320 			pmap_enter(vm_map_pmap(u_map),
321 			    ((vm_offset_t) p->p_addr) + off,
322 			    pa, VM_PROT_READ | VM_PROT_WRITE, 1);
323 		}
324 
325 		s = splhigh();
326 
327 		if (p->p_stat == SRUN)
328 			setrunqueue(p);
329 
330 		p->p_flag |= P_INMEM;
331 
332 		/* undo the effect of setting SLOCK above */
333 		--p->p_lock;
334 		splx(s);
335 
336 	}
337 }
338 
339 /*
340  * This swapin algorithm attempts to swap-in processes only if there
341  * is enough space for them.  Of course, if a process waits for a long
342  * time, it will be swapped in anyway.
343  */
344 void
345 scheduler()
346 {
347 	register struct proc *p;
348 	register int pri;
349 	struct proc *pp;
350 	int ppri;
351 
352 loop:
353 	while ((cnt.v_free_count + cnt.v_cache_count) < (cnt.v_free_reserved + UPAGES + 2)) {
354 		VM_WAIT;
355 		tsleep((caddr_t) &proc0, PVM, "schedm", 0);
356 	}
357 
358 	pp = NULL;
359 	ppri = INT_MIN;
360 	for (p = (struct proc *) allproc; p != NULL; p = p->p_next) {
361 		if (p->p_stat == SRUN && (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
362 			int mempri;
363 
364 			pri = p->p_swtime + p->p_slptime - p->p_nice * 8;
365 			mempri = pri > 0 ? pri : 0;
366 			/*
367 			 * if this process is higher priority and there is
368 			 * enough space, then select this process instead of
369 			 * the previous selection.
370 			 */
371 			if (pri > ppri) {
372 				pp = p;
373 				ppri = pri;
374 			}
375 		}
376 	}
377 
378 	/*
379 	 * Nothing to do, back to sleep
380 	 */
381 	if ((p = pp) == NULL) {
382 		tsleep((caddr_t) &proc0, PVM, "sched", 0);
383 		goto loop;
384 	}
385 	/*
386 	 * We would like to bring someone in. (only if there is space).
387 	 */
388 	faultin(p);
389 	p->p_swtime = 0;
390 	goto loop;
391 }
392 
393 #define	swappable(p) \
394 	(((p)->p_lock == 0) && \
395 		((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM)
396 
397 extern int vm_pageout_free_min;
398 
399 /*
400  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
401  * procs and unwire their u-areas.  We try to always "swap" at least one
402  * process in case we need the room for a swapin.
403  * If any procs have been sleeping/stopped for at least maxslp seconds,
404  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
405  * if any, otherwise the longest-resident process.
406  */
407 void
408 swapout_threads()
409 {
410 	register struct proc *p;
411 	struct proc *outp, *outp2;
412 	int outpri, outpri2;
413 	int didswap = 0;
414 
415 	outp = outp2 = NULL;
416 	outpri = outpri2 = INT_MIN;
417 retry:
418 	for (p = (struct proc *) allproc; p != NULL; p = p->p_next) {
419 		if (!swappable(p))
420 			continue;
421 		switch (p->p_stat) {
422 		default:
423 			continue;
424 
425 		case SSLEEP:
426 		case SSTOP:
427 			/*
428 			 * do not swapout a realtime process
429 			 */
430 			if (p->p_rtprio.type == RTP_PRIO_REALTIME)
431 				continue;
432 
433 			/*
434 			 * do not swapout a process waiting on a critical
435 			 * event of some kind
436 			 */
437 			if ((p->p_priority & 0x7f) < PSOCK)
438 				continue;
439 
440 			vm_map_reference(&p->p_vmspace->vm_map);
441 			/*
442 			 * do not swapout a process that is waiting for VM
443 			 * datastructures there is a possible deadlock.
444 			 */
445 			if (!lock_try_write(&p->p_vmspace->vm_map.lock)) {
446 				vm_map_deallocate(&p->p_vmspace->vm_map);
447 				continue;
448 			}
449 			vm_map_unlock(&p->p_vmspace->vm_map);
450 			/*
451 			 * If the process has been asleep for awhile and had
452 			 * most of its pages taken away already, swap it out.
453 			 */
454 			if (p->p_slptime > 4) {
455 				swapout(p);
456 				vm_map_deallocate(&p->p_vmspace->vm_map);
457 				didswap++;
458 				goto retry;
459 			}
460 			vm_map_deallocate(&p->p_vmspace->vm_map);
461 		}
462 	}
463 	/*
464 	 * If we swapped something out, and another process needed memory,
465 	 * then wakeup the sched process.
466 	 */
467 	if (didswap)
468 		wakeup((caddr_t) &proc0);
469 }
470 
471 void
472 swapout(p)
473 	register struct proc *p;
474 {
475 	vm_map_t map = &p->p_vmspace->vm_map;
476 	vm_offset_t ptaddr;
477 
478 	++p->p_stats->p_ru.ru_nswap;
479 	/*
480 	 * remember the process resident count
481 	 */
482 	p->p_vmspace->vm_swrss =
483 	    p->p_vmspace->vm_pmap.pm_stats.resident_count;
484 
485 	(void) splhigh();
486 	p->p_flag &= ~P_INMEM;
487 	if (p->p_stat == SRUN)
488 		remrq(p);
489 	(void) spl0();
490 
491 	p->p_flag |= P_SWAPPING;
492 	/*
493 	 * let the upages be paged
494 	 */
495 	pmap_remove(vm_map_pmap(u_map),
496 	    (vm_offset_t) p->p_addr, ((vm_offset_t) p->p_addr) + UPAGES * NBPG);
497 
498 	vm_map_pageable(map, (vm_offset_t) kstack,
499 	    (vm_offset_t) kstack + UPAGES * NBPG, TRUE);
500 
501 	ptaddr = trunc_page((u_int) vtopte(kstack));
502 	vm_map_pageable(map, ptaddr, ptaddr + NBPG, TRUE);
503 
504 	p->p_flag &= ~P_SWAPPING;
505 	p->p_swtime = 0;
506 }
507 
508 /*
509  * The rest of these routines fake thread handling
510  */
511 
512 #ifndef assert_wait
513 void
514 assert_wait(event, ruptible)
515 	int event;
516 	boolean_t ruptible;
517 {
518 #ifdef lint
519 	ruptible++;
520 #endif
521 	curproc->p_thread = event;
522 }
523 #endif
524 
525 void
526 thread_block(char *msg)
527 {
528 	if (curproc->p_thread)
529 		tsleep((caddr_t) curproc->p_thread, PVM, msg, 0);
530 }
531 
532 
533 void
534 thread_sleep_(event, lock, wmesg)
535 	int event;
536 	simple_lock_t lock;
537 	char *wmesg;
538 {
539 
540 	curproc->p_thread = event;
541 	simple_unlock(lock);
542 	if (curproc->p_thread) {
543 		tsleep((caddr_t) event, PVM, wmesg, 0);
544 	}
545 }
546 
547 #ifndef thread_wakeup
548 void
549 thread_wakeup(event)
550 	int event;
551 {
552 	wakeup((caddr_t) event);
553 }
554 #endif
555 
556 #ifdef DDB
557 
558 /*
559  * DEBUG stuff
560  */
561 
562 int indent;
563 
564 #include <machine/stdarg.h>	/* see subr_prf.c */
565 
566 /*ARGSUSED2*/
567 void
568 #if __STDC__
569 iprintf(const char *fmt,...)
570 #else
571 iprintf(fmt /* , va_alist */ )
572 	char *fmt;
573 
574  /* va_dcl */
575 #endif
576 {
577 	register int i;
578 	va_list ap;
579 
580 	for (i = indent; i >= 8; i -= 8)
581 		printf("\t");
582 	while (--i >= 0)
583 		printf(" ");
584 	va_start(ap, fmt);
585 	printf("%r", fmt, ap);
586 	va_end(ap);
587 }
588 #endif /* DDB */
589