xref: /freebsd/sys/vm/vm_glue.c (revision 0de89efe5c443f213c7ea28773ef2dc6cf3af2ed)
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 1.65 1997/08/21 20:33:42 bde Exp $
63  */
64 
65 #include "opt_rlimit.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/unistd.h>
77 
78 #include <machine/limits.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/vm_prot.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 extern char kstack[];
114 
115 /* vm_map_t upages_map; */
116 
117 int
118 kernacc(addr, len, rw)
119 	caddr_t addr;
120 	int len, rw;
121 {
122 	boolean_t rv;
123 	vm_offset_t saddr, eaddr;
124 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
125 
126 	saddr = trunc_page(addr);
127 	eaddr = round_page(addr + len);
128 	vm_map_lock_read(kernel_map);
129 	rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
130 	vm_map_unlock_read(kernel_map);
131 	return (rv == TRUE);
132 }
133 
134 int
135 useracc(addr, len, rw)
136 	caddr_t addr;
137 	int len, rw;
138 {
139 	boolean_t rv;
140 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
141 	vm_map_t map;
142 	vm_map_entry_t save_hint;
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 	map = &curproc->p_vmspace->vm_map;
158 	vm_map_lock_read(map);
159 	/*
160 	 * We save the map hint, and restore it.  Useracc appears to distort
161 	 * the map hint unnecessarily.
162 	 */
163 	save_hint = map->hint;
164 	rv = vm_map_check_protection(map,
165 	    trunc_page(addr), round_page(addr + len), prot);
166 	map->hint = save_hint;
167 	vm_map_unlock_read(map);
168 
169 	return (rv == TRUE);
170 }
171 
172 void
173 vslock(addr, len)
174 	caddr_t addr;
175 	u_int len;
176 {
177 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
178 	    round_page(addr + len), FALSE);
179 }
180 
181 void
182 vsunlock(addr, len, dirtied)
183 	caddr_t addr;
184 	u_int len;
185 	int dirtied;
186 {
187 #ifdef	lint
188 	dirtied++;
189 #endif	/* lint */
190 	vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr),
191 	    round_page(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 	int i;
209 	pmap_t pvp;
210 
211 	if (flags & RFMEM) {
212 		p2->p_vmspace = p1->p_vmspace;
213 		p1->p_vmspace->vm_refcnt++;
214 	}
215 
216 	while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) {
217 		VM_WAIT;
218 	}
219 
220 	if ((flags & RFMEM) == 0) {
221 		p2->p_vmspace = vmspace_fork(p1->p_vmspace);
222 
223 		if (p1->p_vmspace->vm_shm)
224 			shmfork(p1, p2);
225 	}
226 
227 	pmap_new_proc(p2);
228 
229 	up = p2->p_addr;
230 
231 	/*
232 	 * p_stats and p_sigacts currently point at fields in the user struct
233 	 * but not at &u, instead at p_addr. Copy p_sigacts and parts of
234 	 * p_stats; zero the rest of p_stats (statistics).
235 	 */
236 	p2->p_stats = &up->u_stats;
237 	p2->p_sigacts = &up->u_sigacts;
238 	up->u_sigacts = *p1->p_sigacts;
239 	bzero(&up->u_stats.pstat_startzero,
240 	    (unsigned) ((caddr_t) &up->u_stats.pstat_endzero -
241 		(caddr_t) &up->u_stats.pstat_startzero));
242 	bcopy(&p1->p_stats->pstat_startcopy, &up->u_stats.pstat_startcopy,
243 	    ((caddr_t) &up->u_stats.pstat_endcopy -
244 		(caddr_t) &up->u_stats.pstat_startcopy));
245 
246 
247 	/*
248 	 * cpu_fork will copy and update the pcb, set up the kernel stack,
249 	 * and make the child ready to run.
250 	 */
251 	cpu_fork(p1, p2);
252 }
253 
254 /*
255  * Set default limits for VM system.
256  * Called for proc 0, and then inherited by all others.
257  *
258  * XXX should probably act directly on proc0.
259  */
260 static void
261 vm_init_limits(udata)
262 	void *udata;
263 {
264 	register struct proc *p = udata;
265 	int rss_limit;
266 
267 	/*
268 	 * Set up the initial limits on process VM. Set the maximum resident
269 	 * set size to be half of (reasonably) available memory.  Since this
270 	 * is a soft limit, it comes into effect only when the system is out
271 	 * of memory - half of main memory helps to favor smaller processes,
272 	 * and reduces thrashing of the object cache.
273 	 */
274 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
275 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
276 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
277 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
278 	/* limit the limit to no less than 2MB */
279 	rss_limit = max(cnt.v_free_count, 512);
280 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(rss_limit);
281 	p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY;
282 }
283 
284 void
285 faultin(p)
286 	struct proc *p;
287 {
288 	vm_offset_t i;
289 	int s;
290 
291 	if ((p->p_flag & P_INMEM) == 0) {
292 
293 		++p->p_lock;
294 
295 		pmap_swapin_proc(p);
296 
297 		s = splhigh();
298 
299 		if (p->p_stat == SRUN)
300 			setrunqueue(p);
301 
302 		p->p_flag |= P_INMEM;
303 
304 		/* undo the effect of setting SLOCK above */
305 		--p->p_lock;
306 		splx(s);
307 
308 	}
309 }
310 
311 /*
312  * This swapin algorithm attempts to swap-in processes only if there
313  * is enough space for them.  Of course, if a process waits for a long
314  * time, it will be swapped in anyway.
315  */
316 /* ARGSUSED*/
317 static void
318 scheduler(dummy)
319 	void *dummy;
320 {
321 	register struct proc *p;
322 	register int pri;
323 	struct proc *pp;
324 	int ppri;
325 
326 loop:
327 	while ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_free_min) {
328 		VM_WAIT;
329 	}
330 
331 	pp = NULL;
332 	ppri = INT_MIN;
333 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
334 		if (p->p_stat == SRUN &&
335 			(p->p_flag & (P_INMEM | P_SWAPPING)) == 0) {
336 			int mempri;
337 
338 			pri = p->p_swtime + p->p_slptime;
339 			if ((p->p_flag & P_SWAPINREQ) == 0) {
340 				pri -= p->p_nice * 8;
341 			}
342 			mempri = pri > 0 ? pri : 0;
343 			/*
344 			 * if this process is higher priority and there is
345 			 * enough space, then select this process instead of
346 			 * the previous selection.
347 			 */
348 			if (pri > ppri) {
349 				pp = p;
350 				ppri = pri;
351 			}
352 		}
353 	}
354 
355 	/*
356 	 * Nothing to do, back to sleep.
357 	 */
358 	if ((p = pp) == NULL) {
359 		tsleep(&proc0, PVM, "sched", 0);
360 		goto loop;
361 	}
362 	p->p_flag &= ~P_SWAPINREQ;
363 
364 	/*
365 	 * We would like to bring someone in. (only if there is space).
366 	 */
367 	faultin(p);
368 	p->p_swtime = 0;
369 	goto loop;
370 }
371 
372 #ifndef NO_SWAPPING
373 
374 #define	swappable(p) \
375 	(((p)->p_lock == 0) && \
376 		((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO|P_SWAPPING)) == P_INMEM)
377 
378 /*
379  * Swapout is driven by the pageout daemon.  Very simple, we find eligible
380  * procs and unwire their u-areas.  We try to always "swap" at least one
381  * process in case we need the room for a swapin.
382  * If any procs have been sleeping/stopped for at least maxslp seconds,
383  * they are swapped.  Else, we swap the longest-sleeping or stopped process,
384  * if any, otherwise the longest-resident process.
385  */
386 void
387 swapout_procs()
388 {
389 	register struct proc *p;
390 	struct proc *outp, *outp2;
391 	int outpri, outpri2;
392 	int didswap = 0;
393 
394 	outp = outp2 = NULL;
395 	outpri = outpri2 = INT_MIN;
396 retry:
397 	for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
398 		struct vmspace *vm;
399 		if (!swappable(p))
400 			continue;
401 
402 		vm = p->p_vmspace;
403 
404 		switch (p->p_stat) {
405 		default:
406 			continue;
407 
408 		case SSLEEP:
409 		case SSTOP:
410 			/*
411 			 * do not swapout a realtime process
412 			 */
413 			if (p->p_rtprio.type == RTP_PRIO_REALTIME)
414 				continue;
415 
416 			/*
417 			 * do not swapout a process waiting on a critical
418 			 * event of some kind
419 			 */
420 			if (((p->p_priority & 0x7f) < PSOCK) ||
421 				(p->p_slptime <= 10))
422 				continue;
423 
424 			++vm->vm_refcnt;
425 			vm_map_reference(&vm->vm_map);
426 			/*
427 			 * do not swapout a process that is waiting for VM
428 			 * data structures there is a possible deadlock.
429 			 */
430 			if (lockmgr(&vm->vm_map.lock,
431 					LK_EXCLUSIVE | LK_NOWAIT,
432 					(void *)0, curproc)) {
433 				vm_map_deallocate(&vm->vm_map);
434 				vmspace_free(vm);
435 				continue;
436 			}
437 			vm_map_unlock(&vm->vm_map);
438 			/*
439 			 * If the process has been asleep for awhile and had
440 			 * most of its pages taken away already, swap it out.
441 			 */
442 			swapout(p);
443 			vm_map_deallocate(&vm->vm_map);
444 			vmspace_free(vm);
445 			didswap++;
446 			goto retry;
447 		}
448 	}
449 	/*
450 	 * If we swapped something out, and another process needed memory,
451 	 * then wakeup the sched process.
452 	 */
453 	if (didswap)
454 		wakeup(&proc0);
455 }
456 
457 static void
458 swapout(p)
459 	register struct proc *p;
460 {
461 	pmap_t pmap = &p->p_vmspace->vm_pmap;
462 	int i;
463 
464 #if defined(SWAP_DEBUG)
465 	printf("swapping out %d\n", p->p_pid);
466 #endif
467 	++p->p_stats->p_ru.ru_nswap;
468 	/*
469 	 * remember the process resident count
470 	 */
471 	p->p_vmspace->vm_swrss =
472 	    p->p_vmspace->vm_pmap.pm_stats.resident_count;
473 
474 	(void) splhigh();
475 	p->p_flag &= ~P_INMEM;
476 	p->p_flag |= P_SWAPPING;
477 	if (p->p_stat == SRUN)
478 		remrq(p);
479 	(void) spl0();
480 
481 	pmap_swapout_proc(p);
482 
483 	p->p_flag &= ~P_SWAPPING;
484 	p->p_swtime = 0;
485 }
486 #endif /* !NO_SWAPPING */
487