xref: /freebsd/sys/vm/vm_glue.c (revision 0ea3482342b4d7d6e71f3007ce4dafe445c639fd)
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 1.29 1995/10/23 05:35:42 dyson 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 /*
85  * System initialization
86  *
87  * Note: proc0 from proc.h
88  */
89 
90 static void vm_init_limits __P((void *));
91 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
92 
93 /*
94  * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
95  *
96  * Note: run scheduling should be divorced from the vm system.
97  */
98 static void scheduler __P((void *));
99 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
100 
101 
102 extern char kstack[];
103 
104 /* vm_map_t upages_map; */
105 
106 int
107 kernacc(addr, len, rw)
108 	caddr_t addr;
109 	int len, rw;
110 {
111 	boolean_t rv;
112 	vm_offset_t saddr, eaddr;
113 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
114 
115 	saddr = trunc_page(addr);
116 	eaddr = round_page(addr + len);
117 	rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
118 	return (rv == TRUE);
119 }
120 
121 int
122 useracc(addr, len, rw)
123 	caddr_t addr;
124 	int len, rw;
125 {
126 	boolean_t rv;
127 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
128 
129 	/*
130 	 * XXX - check separately to disallow access to user area and user
131 	 * page tables - they are in the map.
132 	 *
133 	 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max.  It was once
134 	 * only used (as an end address) in trap.c.  Use it as an end address
135 	 * here too.  This bogusness has spread.  I just fixed where it was
136 	 * used as a max in vm_mmap.c.
137 	 */
138 	if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
139 	    || (vm_offset_t) addr + len < (vm_offset_t) addr) {
140 		return (FALSE);
141 	}
142 	rv = vm_map_check_protection(&curproc->p_vmspace->vm_map,
143 	    trunc_page(addr), round_page(addr + len), prot);
144 	return (rv == TRUE);
145 }
146 
147 #ifdef KGDB
148 /*
149  * Change protections on kernel pages from addr to addr+len
150  * (presumably so debugger can plant a breakpoint).
151  * All addresses are assumed to reside in the Sysmap,
152  */
153 chgkprot(addr, len, rw)
154 	register caddr_t addr;
155 	int len, rw;
156 {
157 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
158 
159 	vm_map_protect(kernel_map, trunc_page(addr),
160 	    round_page(addr + len), prot, FALSE);
161 }
162 #endif
163 void
164 vslock(addr, len)
165 	caddr_t addr;
166 	u_int len;
167 {
168 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
169 	    round_page(addr + len), FALSE);
170 }
171 
172 void
173 vsunlock(addr, len, dirtied)
174 	caddr_t addr;
175 	u_int len;
176 	int dirtied;
177 {
178 #ifdef	lint
179 	dirtied++;
180 #endif	/* lint */
181 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
182 	    round_page(addr + len), TRUE);
183 }
184 
185 /*
186  * Implement fork's actions on an address space.
187  * Here we arrange for the address space to be copied or referenced,
188  * allocate a user struct (pcb and kernel stack), then call the
189  * machine-dependent layer to fill those in and make the new process
190  * ready to run.
191  * NOTE: the kernel stack may be at a different location in the child
192  * process, and thus addresses of automatic variables may be invalid
193  * after cpu_fork returns in the child process.  We do nothing here
194  * after cpu_fork returns.
195  */
196 int
197 vm_fork(p1, p2, isvfork)
198 	register struct proc *p1, *p2;
199 	int isvfork;
200 {
201 	register struct user *up;
202 	vm_offset_t addr, ptaddr;
203 	int error, i;
204 	struct vm_map *vp;
205 
206 	while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) {
207 		VM_WAIT;
208 	}
209 
210 	/*
211 	 * avoid copying any of the parent's pagetables or other per-process
212 	 * objects that reside in the map by marking all of them
213 	 * non-inheritable
214 	 */
215 	(void) vm_map_inherit(&p1->p_vmspace->vm_map,
216 	    UPT_MIN_ADDRESS - UPAGES * PAGE_SIZE, VM_MAX_ADDRESS, VM_INHERIT_NONE);
217 	p2->p_vmspace = vmspace_fork(p1->p_vmspace);
218 
219 #ifdef SYSVSHM
220 	if (p1->p_vmspace->vm_shm)
221 		shmfork(p1, p2, isvfork);
222 #endif
223 
224 	/*
225 	 * Allocate a wired-down (for now) pcb and kernel stack for the
226 	 * process
227 	 */
228 
229 	addr = (vm_offset_t) kstack;
230 
231 	vp = &p2->p_vmspace->vm_map;
232 
233 	/* get new pagetables and kernel stack */
234 	(void) vm_map_find(vp, NULL, 0, &addr, UPT_MAX_ADDRESS - addr, FALSE);
235 
236 	/* force in the page table encompassing the UPAGES */
237 	ptaddr = trunc_page((u_int) vtopte(addr));
238 	error = vm_map_pageable(vp, ptaddr, ptaddr + PAGE_SIZE, FALSE);
239 	if (error)
240 		panic("vm_fork: wire of PT failed. error=%d", error);
241 
242 	/* and force in (demand-zero) the UPAGES */
243 	error = vm_map_pageable(vp, addr, addr + UPAGES * PAGE_SIZE, FALSE);
244 	if (error)
245 		panic("vm_fork: wire of UPAGES failed. error=%d", error);
246 
247 	/* get a kernel virtual address for the UPAGES for this proc */
248 	up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * PAGE_SIZE);
249 	if (up == NULL)
250 		panic("vm_fork: u_map allocation failed");
251 
252 	/* and force-map the upages into the kernel pmap */
253 	for (i = 0; i < UPAGES; i++)
254 		pmap_kenter(((vm_offset_t) up) + PAGE_SIZE * i,
255 		    pmap_extract(vp->pmap, addr + PAGE_SIZE * i));
256 
257 	p2->p_addr = up;
258 
259 	/*
260 	 * p_stats and p_sigacts currently point at fields in the user struct
261 	 * but not at &u, instead at p_addr. Copy p_sigacts and parts of
262 	 * p_stats; zero the rest of p_stats (statistics).
263 	 */
264 	p2->p_stats = &up->u_stats;
265 	p2->p_sigacts = &up->u_sigacts;
266 	up->u_sigacts = *p1->p_sigacts;
267 	bzero(&up->u_stats.pstat_startzero,
268 	    (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
269 		(caddr_t) &up->u_stats.pstat_startzero));
270 	bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
271 	    ((caddr_t) &up->u_stats.pstat_endcopy -
272 		(caddr_t) &up->u_stats.pstat_startcopy));
273 
274 
275 	/*
276 	 * cpu_fork will copy and update the kernel stack and pcb, and make
277 	 * the child ready to run.  It marks the child so that it can return
278 	 * differently than the parent. It returns twice, once in the parent
279 	 * process and once in the child.
280 	 */
281 	return (cpu_fork(p1, p2));
282 }
283 
284 /*
285  * Set default limits for VM system.
286  * Called for proc 0, and then inherited by all others.
287  *
288  * XXX should probably act directly on proc0.
289  */
290 static void
291 vm_init_limits(udata)
292 	void *udata;
293 {
294 	register struct proc *p = (struct proc *)udata;
295 	int rss_limit;
296 
297 	/*
298 	 * Set up the initial limits on process VM. Set the maximum resident
299 	 * set size to be half of (reasonably) available memory.  Since this
300 	 * is a soft limit, it comes into effect only when the system is out
301 	 * of memory - half of main memory helps to favor smaller processes,
302 	 * and reduces thrashing of the object cache.
303 	 */
304 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
305 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
306 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
307 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
308 	/* limit the limit to no less than 2MB */
309 	rss_limit = max(cnt.v_free_count, 512);
310 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
311 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
312 }
313 
314 void
315 faultin(p)
316 	struct proc *p;
317 {
318 	vm_offset_t i;
319 	vm_offset_t ptaddr;
320 	int s;
321 
322 	if ((p->p_flag & P_INMEM) == 0) {
323 		vm_map_t map;
324 		int error;
325 
326 		++p->p_lock;
327 
328 		map = &p->p_vmspace->vm_map;
329 		/* force the page table encompassing the kernel stack (upages) */
330 		ptaddr = trunc_page((u_int) vtopte(kstack));
331 		error = vm_map_pageable(map, ptaddr, ptaddr + PAGE_SIZE, FALSE);
332 		if (error)
333 			panic("faultin: wire of PT failed. error=%d", error);
334 
335 		/* wire in the UPAGES */
336 		error = vm_map_pageable(map, (vm_offset_t) kstack,
337 		    (vm_offset_t) kstack + UPAGES * PAGE_SIZE, FALSE);
338 		if (error)
339 			panic("faultin: wire of UPAGES failed. error=%d", error);
340 
341 		/* and map them nicely into the kernel pmap */
342 		for (i = 0; i < UPAGES; i++) {
343 			vm_offset_t off = i * PAGE_SIZE;
344 			vm_offset_t pa = (vm_offset_t)
345 				pmap_extract(&p->p_vmspace->vm_pmap,
346 				    (vm_offset_t) kstack + off);
347 
348 			if (pa == 0)
349 				panic("faultin: missing page for UPAGES\n");
350 
351 			pmap_kenter(((vm_offset_t) p->p_addr) + off, pa);
352 		}
353 
354 		s = splhigh();
355 
356 		if (p->p_stat == SRUN)
357 			setrunqueue(p);
358 
359 		p->p_flag |= P_INMEM;
360 
361 		/* undo the effect of setting SLOCK above */
362 		--p->p_lock;
363 		splx(s);
364 
365 	}
366 }
367 
368 /*
369  * This swapin algorithm attempts to swap-in processes only if there
370  * is enough space for them.  Of course, if a process waits for a long
371  * time, it will be swapped in anyway.
372  */
373 /* ARGSUSED*/
374 static void
375 scheduler(udata)
376 	void *udata;		/* not used*/
377 {
378 	register struct proc *p;
379 	register int pri;
380 	struct proc *pp;
381 	int ppri;
382 
383 loop:
384 	while ((cnt.v_free_count + cnt.v_cache_count) < (cnt.v_free_reserved + UPAGES + 2)) {
385 		VM_WAIT;
386 	}
387 
388 	pp = NULL;
389 	ppri = INT_MIN;
390 	for (p = (struct proc *) allproc; p != NULL; p = p->p_next) {
391 		if (p->p_stat == SRUN && (p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
392 			int mempri;
393 
394 			pri = p->p_swtime + p->p_slptime - p->p_nice * 8;
395 			mempri = pri > 0 ? pri : 0;
396 			/*
397 			 * if this process is higher priority and there is
398 			 * enough space, then select this process instead of
399 			 * the previous selection.
400 			 */
401 			if (pri > ppri) {
402 				pp = p;
403 				ppri = pri;
404 			}
405 		}
406 	}
407 
408 	/*
409 	 * Nothing to do, back to sleep
410 	 */
411 	if ((p = pp) == NULL) {
412 		tsleep(&proc0, PVM, "sched", 0);
413 		goto loop;
414 	}
415 	/*
416 	 * We would like to bring someone in. (only if there is space).
417 	 */
418 	faultin(p);
419 	p->p_swtime = 0;
420 	goto loop;
421 }
422 
423 #define	swappable(p) \
424 	(((p)->p_lock == 0) && \
425 		((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM)
426 
427 extern int vm_pageout_free_min;
428 
429 /*
430  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
431  * procs and unwire their u-areas.  We try to always "swap" at least one
432  * process in case we need the room for a swapin.
433  * If any procs have been sleeping/stopped for at least maxslp seconds,
434  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
435  * if any, otherwise the longest-resident process.
436  */
437 void
438 swapout_procs()
439 {
440 	register struct proc *p;
441 	struct proc *outp, *outp2;
442 	int outpri, outpri2;
443 	int didswap = 0;
444 
445 	outp = outp2 = NULL;
446 	outpri = outpri2 = INT_MIN;
447 retry:
448 	for (p = (struct proc *) allproc; p != NULL; p = p->p_next) {
449 		if (!swappable(p))
450 			continue;
451 		switch (p->p_stat) {
452 		default:
453 			continue;
454 
455 		case SSLEEP:
456 		case SSTOP:
457 			/*
458 			 * do not swapout a realtime process
459 			 */
460 			if (p->p_rtprio.type == RTP_PRIO_REALTIME)
461 				continue;
462 
463 			/*
464 			 * do not swapout a process waiting on a critical
465 			 * event of some kind
466 			 */
467 			if (((p->p_priority & 0x7f) < PSOCK) ||
468 				(p->p_slptime <= 4))
469 				continue;
470 
471 			vm_map_reference(&p->p_vmspace->vm_map);
472 			/*
473 			 * do not swapout a process that is waiting for VM
474 			 * datastructures there is a possible deadlock.
475 			 */
476 			if (!lock_try_write(&p->p_vmspace->vm_map.lock)) {
477 				vm_map_deallocate(&p->p_vmspace->vm_map);
478 				continue;
479 			}
480 			vm_map_unlock(&p->p_vmspace->vm_map);
481 			/*
482 			 * If the process has been asleep for awhile and had
483 			 * most of its pages taken away already, swap it out.
484 			 */
485 			swapout(p);
486 			vm_map_deallocate(&p->p_vmspace->vm_map);
487 			didswap++;
488 			goto retry;
489 		}
490 	}
491 	/*
492 	 * If we swapped something out, and another process needed memory,
493 	 * then wakeup the sched process.
494 	 */
495 	if (didswap)
496 		wakeup(&proc0);
497 }
498 
499 void
500 swapout(p)
501 	register struct proc *p;
502 {
503 	vm_map_t map = &p->p_vmspace->vm_map;
504 	vm_offset_t ptaddr;
505 	int i;
506 
507 	++p->p_stats->p_ru.ru_nswap;
508 	/*
509 	 * remember the process resident count
510 	 */
511 	p->p_vmspace->vm_swrss =
512 	    p->p_vmspace->vm_pmap.pm_stats.resident_count;
513 
514 	(void) splhigh();
515 	p->p_flag &= ~P_INMEM;
516 	p->p_flag |= P_SWAPPING;
517 	if (p->p_stat == SRUN)
518 		remrq(p);
519 	(void) spl0();
520 
521 	/*
522 	 * let the upages be paged
523 	 */
524 	for(i=0;i<UPAGES;i++)
525 		pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
526 
527 	vm_map_pageable(map, (vm_offset_t) kstack,
528 	    (vm_offset_t) kstack + UPAGES * PAGE_SIZE, TRUE);
529 
530 	ptaddr = trunc_page((u_int) vtopte(kstack));
531 	vm_map_pageable(map, ptaddr, ptaddr + PAGE_SIZE, TRUE);
532 
533 	p->p_flag &= ~P_SWAPPING;
534 	p->p_swtime = 0;
535 }
536 
537 #ifdef DDB
538 /*
539  * DEBUG stuff
540  */
541 
542 int indent;
543 
544 #include <machine/stdarg.h>	/* see subr_prf.c */
545 
546 /*ARGSUSED2*/
547 void
548 #if __STDC__
549 iprintf(const char *fmt,...)
550 #else
551 iprintf(fmt /* , va_alist */ )
552 	char *fmt;
553 
554  /* va_dcl */
555 #endif
556 {
557 	register int i;
558 	va_list ap;
559 
560 	for (i = indent; i >= 8; i -= 8)
561 		printf("\t");
562 	while (--i >= 0)
563 		printf(" ");
564 	va_start(ap, fmt);
565 	printf("%r", fmt, ap);
566 	va_end(ap);
567 }
568 #endif /* DDB */
569