xref: /freebsd/sys/vm/vm_glue.c (revision ef5d438ed4bc17ad7ece3e40fe4d1f9baf3aadf7)
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 1.39 1996/02/22 10:57:36 davidg Exp $
63  */
64 
65 #include "opt_ddb.h"
66 
67 #include <sys/param.h>
68 #include <sys/systm.h>
69 #include <sys/proc.h>
70 #include <sys/resourcevar.h>
71 #include <sys/buf.h>
72 #include <sys/shm.h>
73 #include <sys/vmmeter.h>
74 
75 #include <sys/kernel.h>
76 #include <sys/dkstat.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_inherit.h>
81 #include <vm/vm_prot.h>
82 #include <vm/lock.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pageout.h>
87 #include <vm/vm_kern.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_pager.h>
91 
92 #include <sys/user.h>
93 
94 #include <machine/stdarg.h>
95 #include <machine/cpu.h>
96 
97 /*
98  * System initialization
99  *
100  * Note: proc0 from proc.h
101  */
102 
103 static void vm_init_limits __P((void *));
104 SYSINIT(vm_limits, SI_SUB_VM_CONF, SI_ORDER_FIRST, vm_init_limits, &proc0)
105 
106 /*
107  * THIS MUST BE THE LAST INITIALIZATION ITEM!!!
108  *
109  * Note: run scheduling should be divorced from the vm system.
110  */
111 static void scheduler __P((void *));
112 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, scheduler, NULL)
113 
114 
115 static void swapout __P((struct proc *));
116 
117 extern char kstack[];
118 
119 /* vm_map_t upages_map; */
120 
121 int
122 kernacc(addr, len, rw)
123 	caddr_t addr;
124 	int len, rw;
125 {
126 	boolean_t rv;
127 	vm_offset_t saddr, eaddr;
128 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
129 
130 	saddr = trunc_page(addr);
131 	eaddr = round_page(addr + len);
132 	rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
133 	return (rv == TRUE);
134 }
135 
136 int
137 useracc(addr, len, rw)
138 	caddr_t addr;
139 	int len, rw;
140 {
141 	boolean_t rv;
142 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
143 
144 	/*
145 	 * XXX - check separately to disallow access to user area and user
146 	 * page tables - they are in the map.
147 	 *
148 	 * XXX - VM_MAXUSER_ADDRESS is an end address, not a max.  It was once
149 	 * only used (as an end address) in trap.c.  Use it as an end address
150 	 * here too.  This bogusness has spread.  I just fixed where it was
151 	 * used as a max in vm_mmap.c.
152 	 */
153 	if ((vm_offset_t) addr + len > /* XXX */ VM_MAXUSER_ADDRESS
154 	    || (vm_offset_t) addr + len < (vm_offset_t) addr) {
155 		return (FALSE);
156 	}
157 	rv = vm_map_check_protection(&curproc->p_vmspace->vm_map,
158 	    trunc_page(addr), round_page(addr + len), prot);
159 	return (rv == TRUE);
160 }
161 
162 #ifdef KGDB
163 /*
164  * Change protections on kernel pages from addr to addr+len
165  * (presumably so debugger can plant a breakpoint).
166  * All addresses are assumed to reside in the Sysmap,
167  */
168 chgkprot(addr, len, rw)
169 	register caddr_t addr;
170 	int len, rw;
171 {
172 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
173 
174 	vm_map_protect(kernel_map, trunc_page(addr),
175 	    round_page(addr + len), prot, FALSE);
176 }
177 #endif
178 void
179 vslock(addr, len)
180 	caddr_t addr;
181 	u_int len;
182 {
183 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
184 	    round_page(addr + len), FALSE);
185 }
186 
187 void
188 vsunlock(addr, len, dirtied)
189 	caddr_t addr;
190 	u_int len;
191 	int dirtied;
192 {
193 #ifdef	lint
194 	dirtied++;
195 #endif	/* lint */
196 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
197 	    round_page(addr + len), TRUE);
198 }
199 
200 /*
201  * Implement fork's actions on an address space.
202  * Here we arrange for the address space to be copied or referenced,
203  * allocate a user struct (pcb and kernel stack), then call the
204  * machine-dependent layer to fill those in and make the new process
205  * ready to run.
206  * NOTE: the kernel stack may be at a different location in the child
207  * process, and thus addresses of automatic variables may be invalid
208  * after cpu_fork returns in the child process.  We do nothing here
209  * after cpu_fork returns.
210  */
211 int
212 vm_fork(p1, p2)
213 	register struct proc *p1, *p2;
214 {
215 	register struct user *up;
216 	vm_offset_t addr, ptaddr, ptpa;
217 	int error, i;
218 	vm_map_t map;
219 	pmap_t pvp;
220 	vm_page_t stkm;
221 
222 	while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) {
223 		VM_WAIT;
224 	}
225 
226 	/*
227 	 * avoid copying any of the parent's pagetables or other per-process
228 	 * objects that reside in the map by marking all of them
229 	 * non-inheritable
230 	 */
231 	(void) vm_map_inherit(&p1->p_vmspace->vm_map,
232 	    UPT_MIN_ADDRESS - UPAGES * PAGE_SIZE, VM_MAX_ADDRESS, VM_INHERIT_NONE);
233 	p2->p_vmspace = vmspace_fork(p1->p_vmspace);
234 
235 	if (p1->p_vmspace->vm_shm)
236 		shmfork(p1, p2);
237 
238 	/*
239 	 * Allocate a wired-down (for now) pcb and kernel stack for the
240 	 * process
241 	 */
242 
243 	addr = (vm_offset_t) kstack;
244 
245 	map = &p2->p_vmspace->vm_map;
246 	pvp = &p2->p_vmspace->vm_pmap;
247 
248 	/* get new pagetables and kernel stack */
249 	error = vm_map_find(map, NULL, 0, &addr, UPT_MAX_ADDRESS - addr, FALSE,
250 		VM_PROT_ALL, VM_PROT_ALL, 0);
251 	if (error != KERN_SUCCESS)
252 		panic("vm_fork: vm_map_find failed, addr=0x%x, error=%d", addr, error);
253 
254 	/* get a kernel virtual address for the UPAGES for this proc */
255 	up = (struct user *) kmem_alloc_pageable(u_map, UPAGES * PAGE_SIZE);
256 	if (up == NULL)
257 		panic("vm_fork: u_map allocation failed");
258 
259 	p2->p_vmspace->vm_upages_obj = vm_object_allocate( OBJT_DEFAULT,
260 		UPAGES);
261 
262 	ptaddr = trunc_page((u_int) vtopte(kstack));
263 	(void) vm_fault(map, ptaddr, VM_PROT_READ|VM_PROT_WRITE, FALSE);
264 	ptpa = pmap_extract(pvp, ptaddr);
265 	if (ptpa == 0) {
266 		panic("vm_fork: no pte for UPAGES");
267 	}
268 	stkm = PHYS_TO_VM_PAGE(ptpa);
269 	vm_page_hold(stkm);
270 
271 	for(i=0;i<UPAGES;i++) {
272 		vm_page_t m;
273 
274 		while ((m = vm_page_alloc(p2->p_vmspace->vm_upages_obj, i, VM_ALLOC_ZERO)) == NULL) {
275 			VM_WAIT;
276 		}
277 
278 		vm_page_wire(m);
279 		m->flags &= ~PG_BUSY;
280 		pmap_enter( pvp, (vm_offset_t) kstack + i * PAGE_SIZE,
281 			VM_PAGE_TO_PHYS(m), VM_PROT_READ|VM_PROT_WRITE, 1);
282 		pmap_kenter(((vm_offset_t) up) + i * PAGE_SIZE,
283 			VM_PAGE_TO_PHYS(m));
284 		if ((m->flags & PG_ZERO) == 0)
285 			bzero(((caddr_t) up) + i * PAGE_SIZE, PAGE_SIZE);
286 		m->flags &= ~PG_ZERO;
287 		m->valid = VM_PAGE_BITS_ALL;
288 	}
289 	vm_page_unhold(stkm);
290 
291 	p2->p_addr = up;
292 
293 	/*
294 	 * p_stats and p_sigacts currently point at fields in the user struct
295 	 * but not at &u, instead at p_addr. Copy p_sigacts and parts of
296 	 * p_stats; zero the rest of p_stats (statistics).
297 	 */
298 	p2->p_stats = &up->u_stats;
299 	p2->p_sigacts = &up->u_sigacts;
300 	up->u_sigacts = *p1->p_sigacts;
301 	bzero(&up->u_stats.pstat_startzero,
302 	    (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
303 		(caddr_t) &up->u_stats.pstat_startzero));
304 	bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
305 	    ((caddr_t) &up->u_stats.pstat_endcopy -
306 		(caddr_t) &up->u_stats.pstat_startcopy));
307 
308 
309 	/*
310 	 * cpu_fork will copy and update the kernel stack and pcb, and make
311 	 * the child ready to run.  It marks the child so that it can return
312 	 * differently than the parent. It returns twice, once in the parent
313 	 * process and once in the child.
314 	 */
315 	return (cpu_fork(p1, p2));
316 }
317 
318 /*
319  * Set default limits for VM system.
320  * Called for proc 0, and then inherited by all others.
321  *
322  * XXX should probably act directly on proc0.
323  */
324 static void
325 vm_init_limits(udata)
326 	void *udata;
327 {
328 	register struct proc *p = udata;
329 	int rss_limit;
330 
331 	/*
332 	 * Set up the initial limits on process VM. Set the maximum resident
333 	 * set size to be half of (reasonably) available memory.  Since this
334 	 * is a soft limit, it comes into effect only when the system is out
335 	 * of memory - half of main memory helps to favor smaller processes,
336 	 * and reduces thrashing of the object cache.
337 	 */
338 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
339 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
340 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
341 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
342 	/* limit the limit to no less than 2MB */
343 	rss_limit = max(cnt.v_free_count, 512);
344 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
345 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
346 }
347 
348 void
349 faultin(p)
350 	struct proc *p;
351 {
352 	vm_offset_t i;
353 	vm_offset_t ptaddr;
354 	int s;
355 
356 	if ((p->p_flag & P_INMEM) == 0) {
357 		vm_map_t map = &p->p_vmspace->vm_map;
358 		pmap_t pmap = &p->p_vmspace->vm_pmap;
359 		vm_page_t stkm, m;
360 		vm_offset_t ptpa;
361 		int error;
362 
363 		++p->p_lock;
364 
365 		ptaddr = trunc_page((u_int) vtopte(kstack));
366 		(void) vm_fault(map, ptaddr, VM_PROT_READ|VM_PROT_WRITE, FALSE);
367 		ptpa = pmap_extract(&p->p_vmspace->vm_pmap, ptaddr);
368 		if (ptpa == 0) {
369 			panic("vm_fork: no pte for UPAGES");
370 		}
371 		stkm = PHYS_TO_VM_PAGE(ptpa);
372 		vm_page_hold(stkm);
373 
374 		for(i=0;i<UPAGES;i++) {
375 			int s;
376 			s = splhigh();
377 
378 retry:
379 			if ((m = vm_page_lookup(p->p_vmspace->vm_upages_obj, i)) == NULL) {
380 				if ((m = vm_page_alloc(p->p_vmspace->vm_upages_obj, i, VM_ALLOC_NORMAL)) == NULL) {
381 					VM_WAIT;
382 					goto retry;
383 				}
384 			} else {
385 				if ((m->flags & PG_BUSY) || m->busy) {
386 					m->flags |= PG_WANTED;
387 					tsleep(m, PVM, "swinuw",0);
388 					goto retry;
389 				}
390 			}
391 			vm_page_wire(m);
392 			if (m->valid == VM_PAGE_BITS_ALL)
393 				m->flags &= ~PG_BUSY;
394 			splx(s);
395 
396 			pmap_enter( pmap, (vm_offset_t) kstack + i * PAGE_SIZE,
397 				VM_PAGE_TO_PHYS(m), VM_PROT_READ|VM_PROT_WRITE, TRUE);
398 			pmap_kenter(((vm_offset_t) p->p_addr) + i * PAGE_SIZE,
399 				VM_PAGE_TO_PHYS(m));
400 			if (m->valid != VM_PAGE_BITS_ALL) {
401 				int rv;
402 				rv = vm_pager_get_pages(p->p_vmspace->vm_upages_obj,
403 					&m, 1, 0);
404 				if (rv != VM_PAGER_OK)
405 					panic("faultin: cannot get upages for proc: %d\n", p->p_pid);
406 				m->valid = VM_PAGE_BITS_ALL;
407 				m->flags &= ~PG_BUSY;
408 			}
409 		}
410 		vm_page_unhold(stkm);
411 
412 
413 		s = splhigh();
414 
415 		if (p->p_stat == SRUN)
416 			setrunqueue(p);
417 
418 		p->p_flag |= P_INMEM;
419 
420 		/* undo the effect of setting SLOCK above */
421 		--p->p_lock;
422 		splx(s);
423 
424 	}
425 }
426 
427 /*
428  * This swapin algorithm attempts to swap-in processes only if there
429  * is enough space for them.  Of course, if a process waits for a long
430  * time, it will be swapped in anyway.
431  */
432 /* ARGSUSED*/
433 static void
434 scheduler(dummy)
435 	void *dummy;
436 {
437 	register struct proc *p;
438 	register int pri;
439 	struct proc *pp;
440 	int ppri;
441 
442 loop:
443 	while ((cnt.v_free_count + cnt.v_cache_count) < (cnt.v_free_reserved + UPAGES + 2)) {
444 		VM_WAIT;
445 	}
446 
447 	pp = NULL;
448 	ppri = INT_MIN;
449 	for (p = (struct proc *) allproc; p != NULL; p = p->p_next) {
450 		if (p->p_stat == SRUN &&
451 			(p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
452 			int mempri;
453 
454 			pri = p->p_swtime + p->p_slptime - p->p_nice * 8;
455 			mempri = pri > 0 ? pri : 0;
456 			/*
457 			 * if this process is higher priority and there is
458 			 * enough space, then select this process instead of
459 			 * the previous selection.
460 			 */
461 			if (pri > ppri) {
462 				pp = p;
463 				ppri = pri;
464 			}
465 		}
466 	}
467 
468 	/*
469 	 * Nothing to do, back to sleep
470 	 */
471 	if ((p = pp) == NULL) {
472 		tsleep(&proc0, PVM, "sched", 0);
473 		goto loop;
474 	}
475 	/*
476 	 * We would like to bring someone in. (only if there is space).
477 	 */
478 	faultin(p);
479 	p->p_swtime = 0;
480 	goto loop;
481 }
482 
483 #ifndef NO_SWAPPING
484 
485 #define	swappable(p) \
486 	(((p)->p_lock == 0) && \
487 		((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM)
488 
489 /*
490  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
491  * procs and unwire their u-areas.  We try to always "swap" at least one
492  * process in case we need the room for a swapin.
493  * If any procs have been sleeping/stopped for at least maxslp seconds,
494  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
495  * if any, otherwise the longest-resident process.
496  */
497 void
498 swapout_procs()
499 {
500 	register struct proc *p;
501 	struct proc *outp, *outp2;
502 	int outpri, outpri2;
503 	int didswap = 0;
504 
505 	outp = outp2 = NULL;
506 	outpri = outpri2 = INT_MIN;
507 retry:
508 	for (p = (struct proc *) allproc; p != NULL; p = p->p_next) {
509 		if (!swappable(p))
510 			continue;
511 		switch (p->p_stat) {
512 		default:
513 			continue;
514 
515 		case SSLEEP:
516 		case SSTOP:
517 			/*
518 			 * do not swapout a realtime process
519 			 */
520 			if (p->p_rtprio.type == RTP_PRIO_REALTIME)
521 				continue;
522 
523 			/*
524 			 * do not swapout a process waiting on a critical
525 			 * event of some kind
526 			 */
527 			if (((p->p_priority & 0x7f) < PSOCK) ||
528 				(p->p_slptime <= 4))
529 				continue;
530 
531 			vm_map_reference(&p->p_vmspace->vm_map);
532 			/*
533 			 * do not swapout a process that is waiting for VM
534 			 * datastructures there is a possible deadlock.
535 			 */
536 			if (!lock_try_write(&p->p_vmspace->vm_map.lock)) {
537 				vm_map_deallocate(&p->p_vmspace->vm_map);
538 				continue;
539 			}
540 			vm_map_unlock(&p->p_vmspace->vm_map);
541 			/*
542 			 * If the process has been asleep for awhile and had
543 			 * most of its pages taken away already, swap it out.
544 			 */
545 			swapout(p);
546 			vm_map_deallocate(&p->p_vmspace->vm_map);
547 			didswap++;
548 			goto retry;
549 		}
550 	}
551 	/*
552 	 * If we swapped something out, and another process needed memory,
553 	 * then wakeup the sched process.
554 	 */
555 	if (didswap)
556 		wakeup(&proc0);
557 }
558 
559 static void
560 swapout(p)
561 	register struct proc *p;
562 {
563 	vm_map_t map = &p->p_vmspace->vm_map;
564 	pmap_t pmap = &p->p_vmspace->vm_pmap;
565 	vm_offset_t ptaddr;
566 	int i;
567 
568 	++p->p_stats->p_ru.ru_nswap;
569 	/*
570 	 * remember the process resident count
571 	 */
572 	p->p_vmspace->vm_swrss =
573 	    p->p_vmspace->vm_pmap.pm_stats.resident_count;
574 
575 	(void) splhigh();
576 	p->p_flag &= ~P_INMEM;
577 	p->p_flag |= P_SWAPPING;
578 	if (p->p_stat == SRUN)
579 		remrq(p);
580 	(void) spl0();
581 
582 	/*
583 	 * let the upages be paged
584 	 */
585 	for(i=0;i<UPAGES;i++) {
586 		vm_page_t m;
587 		if ((m = vm_page_lookup(p->p_vmspace->vm_upages_obj, i)) == NULL)
588 			panic("swapout: upage already missing???");
589 		m->dirty = VM_PAGE_BITS_ALL;
590 		vm_page_unwire(m);
591 		pmap_kremove( (vm_offset_t) p->p_addr + PAGE_SIZE * i);
592 	}
593 	pmap_remove(pmap, (vm_offset_t) kstack,
594 		(vm_offset_t) kstack + PAGE_SIZE * UPAGES);
595 
596 	p->p_flag &= ~P_SWAPPING;
597 	p->p_swtime = 0;
598 }
599 #endif /* !NO_SWAPPING */
600 
601 #ifdef DDB
602 /*
603  * DEBUG stuff
604  */
605 
606 int indent;
607 
608 #include <machine/stdarg.h>	/* see subr_prf.c */
609 
610 /*ARGSUSED2*/
611 void
612 #if __STDC__
613 iprintf(const char *fmt,...)
614 #else
615 iprintf(fmt /* , va_alist */ )
616 	char *fmt;
617 
618  /* va_dcl */
619 #endif
620 {
621 	register int i;
622 	va_list ap;
623 
624 	for (i = indent; i >= 8; i -= 8)
625 		printf("\t");
626 	while (--i >= 0)
627 		printf(" ");
628 	va_start(ap, fmt);
629 	vprintf(fmt, ap);
630 	va_end(ap);
631 }
632 #endif /* DDB */
633