xref: /netbsd-src/sys/uvm/uvm_glue.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: uvm_glue.c,v 1.92 2005/12/24 23:41:34 perry Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1991, 1993, The Regents of the University of California.
6  *
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * The Mach Operating System project at Carnegie-Mellon University.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by Charles D. Cranor,
23  *      Washington University, the University of California, Berkeley and
24  *      its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)vm_glue.c	8.6 (Berkeley) 1/5/94
42  * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
43  *
44  *
45  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46  * All rights reserved.
47  *
48  * Permission to use, copy, modify and distribute this software and
49  * its documentation is hereby granted, provided that both the copyright
50  * notice and this permission notice appear in all copies of the
51  * software, derivative works or modified versions, and any portions
52  * thereof, and that both notices appear in supporting documentation.
53  *
54  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57  *
58  * Carnegie Mellon requests users of this software to return to
59  *
60  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
61  *  School of Computer Science
62  *  Carnegie Mellon University
63  *  Pittsburgh PA 15213-3890
64  *
65  * any improvements or extensions that they make and grant Carnegie the
66  * rights to redistribute these changes.
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.92 2005/12/24 23:41:34 perry Exp $");
71 
72 #include "opt_kgdb.h"
73 #include "opt_kstack.h"
74 #include "opt_uvmhist.h"
75 
76 /*
77  * uvm_glue.c: glue functions
78  */
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/proc.h>
83 #include <sys/resourcevar.h>
84 #include <sys/buf.h>
85 #include <sys/user.h>
86 
87 #include <uvm/uvm.h>
88 
89 #include <machine/cpu.h>
90 
91 /*
92  * local prototypes
93  */
94 
95 static void uvm_swapout(struct lwp *);
96 
97 #define UVM_NUAREA_MAX 16
98 void *uvm_uareas;
99 int uvm_nuarea;
100 struct simplelock uvm_uareas_slock = SIMPLELOCK_INITIALIZER;
101 
102 static void uvm_uarea_free(vaddr_t);
103 
104 /*
105  * XXXCDC: do these really belong here?
106  */
107 
108 /*
109  * uvm_kernacc: can the kernel access a region of memory
110  *
111  * - used only by /dev/kmem driver (mem.c)
112  */
113 
114 boolean_t
115 uvm_kernacc(caddr_t addr, size_t len, int rw)
116 {
117 	boolean_t rv;
118 	vaddr_t saddr, eaddr;
119 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
120 
121 	saddr = trunc_page((vaddr_t)addr);
122 	eaddr = round_page((vaddr_t)addr + len);
123 	vm_map_lock_read(kernel_map);
124 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
125 	vm_map_unlock_read(kernel_map);
126 
127 	return(rv);
128 }
129 
130 #ifdef KGDB
131 /*
132  * Change protections on kernel pages from addr to addr+len
133  * (presumably so debugger can plant a breakpoint).
134  *
135  * We force the protection change at the pmap level.  If we were
136  * to use vm_map_protect a change to allow writing would be lazily-
137  * applied meaning we would still take a protection fault, something
138  * we really don't want to do.  It would also fragment the kernel
139  * map unnecessarily.  We cannot use pmap_protect since it also won't
140  * enforce a write-enable request.  Using pmap_enter is the only way
141  * we can ensure the change takes place properly.
142  */
143 void
144 uvm_chgkprot(caddr_t addr, size_t len, int rw)
145 {
146 	vm_prot_t prot;
147 	paddr_t pa;
148 	vaddr_t sva, eva;
149 
150 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
151 	eva = round_page((vaddr_t)addr + len);
152 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
153 		/*
154 		 * Extract physical address for the page.
155 		 */
156 		if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE)
157 			panic("chgkprot: invalid page");
158 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
159 	}
160 	pmap_update(pmap_kernel());
161 }
162 #endif
163 
164 /*
165  * uvm_vslock: wire user memory for I/O
166  *
167  * - called from physio and sys___sysctl
168  * - XXXCDC: consider nuking this (or making it a macro?)
169  */
170 
171 int
172 uvm_vslock(struct proc *p, caddr_t addr, size_t len, vm_prot_t access_type)
173 {
174 	struct vm_map *map;
175 	vaddr_t start, end;
176 	int error;
177 
178 	map = &p->p_vmspace->vm_map;
179 	start = trunc_page((vaddr_t)addr);
180 	end = round_page((vaddr_t)addr + len);
181 	error = uvm_fault_wire(map, start, end, VM_FAULT_WIRE, access_type);
182 	return error;
183 }
184 
185 /*
186  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
187  *
188  * - called from physio and sys___sysctl
189  * - XXXCDC: consider nuking this (or making it a macro?)
190  */
191 
192 void
193 uvm_vsunlock(struct proc *p, caddr_t addr, size_t len)
194 {
195 	uvm_fault_unwire(&p->p_vmspace->vm_map, trunc_page((vaddr_t)addr),
196 		round_page((vaddr_t)addr + len));
197 }
198 
199 /*
200  * uvm_proc_fork: fork a virtual address space
201  *
202  * - the address space is copied as per parent map's inherit values
203  */
204 void
205 uvm_proc_fork(struct proc *p1, struct proc *p2, boolean_t shared)
206 {
207 
208 	if (shared == TRUE) {
209 		p2->p_vmspace = NULL;
210 		uvmspace_share(p1, p2);
211 	} else {
212 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
213 	}
214 
215 	cpu_proc_fork(p1, p2);
216 }
217 
218 
219 /*
220  * uvm_lwp_fork: fork a thread
221  *
222  * - a new "user" structure is allocated for the child process
223  *	[filled in by MD layer...]
224  * - if specified, the child gets a new user stack described by
225  *	stack and stacksize
226  * - NOTE: the kernel stack may be at a different location in the child
227  *	process, and thus addresses of automatic variables may be invalid
228  *	after cpu_lwp_fork returns in the child process.  We do nothing here
229  *	after cpu_lwp_fork returns.
230  * - XXXCDC: we need a way for this to return a failure value rather
231  *   than just hang
232  */
233 void
234 uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
235     void (*func)(void *), void *arg)
236 {
237 	struct user *up = l2->l_addr;
238 	int error;
239 
240 	/*
241 	 * Wire down the U-area for the process, which contains the PCB
242 	 * and the kernel stack.  Wired state is stored in l->l_flag's
243 	 * L_INMEM bit rather than in the vm_map_entry's wired count
244 	 * to prevent kernel_map fragmentation.  If we reused a cached U-area,
245 	 * L_INMEM will already be set and we don't need to do anything.
246 	 *
247 	 * Note the kernel stack gets read/write accesses right off the bat.
248 	 */
249 
250 	if ((l2->l_flag & L_INMEM) == 0) {
251 		error = uvm_fault_wire(kernel_map, (vaddr_t)up,
252 		    (vaddr_t)up + USPACE, VM_FAULT_WIRE,
253 		    VM_PROT_READ | VM_PROT_WRITE);
254 		if (error)
255 			panic("uvm_lwp_fork: uvm_fault_wire failed: %d", error);
256 #ifdef PMAP_UAREA
257 		/* Tell the pmap this is a u-area mapping */
258 		PMAP_UAREA((vaddr_t)up);
259 #endif
260 		l2->l_flag |= L_INMEM;
261 	}
262 
263 #ifdef KSTACK_CHECK_MAGIC
264 	/*
265 	 * fill stack with magic number
266 	 */
267 	kstack_setup_magic(l2);
268 #endif
269 
270 	/*
271 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
272  	 * to run.  If this is a normal user fork, the child will exit
273 	 * directly to user mode via child_return() on its first time
274 	 * slice and will not return here.  If this is a kernel thread,
275 	 * the specified entry point will be executed.
276 	 */
277 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
278 }
279 
280 /*
281  * uvm_uarea_alloc: allocate a u-area
282  */
283 
284 boolean_t
285 uvm_uarea_alloc(vaddr_t *uaddrp)
286 {
287 	vaddr_t uaddr;
288 
289 #ifndef USPACE_ALIGN
290 #define USPACE_ALIGN    0
291 #endif
292 
293 	simple_lock(&uvm_uareas_slock);
294 	if (uvm_nuarea > 0) {
295 		uaddr = (vaddr_t)uvm_uareas;
296 		uvm_uareas = *(void **)uvm_uareas;
297 		uvm_nuarea--;
298 		simple_unlock(&uvm_uareas_slock);
299 		*uaddrp = uaddr;
300 		return TRUE;
301 	} else {
302 		simple_unlock(&uvm_uareas_slock);
303 		*uaddrp = uvm_km_alloc(kernel_map, USPACE, USPACE_ALIGN,
304 		    UVM_KMF_PAGEABLE);
305 		return FALSE;
306 	}
307 }
308 
309 /*
310  * uvm_uarea_free: free a u-area; never blocks
311  */
312 
313 static inline void
314 uvm_uarea_free(vaddr_t uaddr)
315 {
316 	simple_lock(&uvm_uareas_slock);
317 	*(void **)uaddr = uvm_uareas;
318 	uvm_uareas = (void *)uaddr;
319 	uvm_nuarea++;
320 	simple_unlock(&uvm_uareas_slock);
321 }
322 
323 /*
324  * uvm_uarea_drain: return memory of u-areas over limit
325  * back to system
326  */
327 
328 void
329 uvm_uarea_drain(boolean_t empty)
330 {
331 	int leave = empty ? 0 : UVM_NUAREA_MAX;
332 	vaddr_t uaddr;
333 
334 	if (uvm_nuarea <= leave)
335 		return;
336 
337 	simple_lock(&uvm_uareas_slock);
338 	while(uvm_nuarea > leave) {
339 		uaddr = (vaddr_t)uvm_uareas;
340 		uvm_uareas = *(void **)uvm_uareas;
341 		uvm_nuarea--;
342 		simple_unlock(&uvm_uareas_slock);
343 		uvm_km_free(kernel_map, uaddr, USPACE, UVM_KMF_PAGEABLE);
344 		simple_lock(&uvm_uareas_slock);
345 	}
346 	simple_unlock(&uvm_uareas_slock);
347 }
348 
349 /*
350  * uvm_exit: exit a virtual address space
351  *
352  * - the process passed to us is a dead (pre-zombie) process; we
353  *   are running on a different context now (the reaper).
354  * - borrow proc0's address space because freeing the vmspace
355  *   of the dead process may block.
356  */
357 
358 void
359 uvm_proc_exit(struct proc *p)
360 {
361 	struct lwp *l = curlwp; /* XXX */
362 	struct vmspace *ovm;
363 
364 	KASSERT(p == l->l_proc);
365 	ovm = p->p_vmspace;
366 
367 	/*
368 	 * borrow proc0's address space.
369 	 */
370 	pmap_deactivate(l);
371 	p->p_vmspace = proc0.p_vmspace;
372 	pmap_activate(l);
373 
374 	uvmspace_free(ovm);
375 }
376 
377 void
378 uvm_lwp_exit(struct lwp *l)
379 {
380 	vaddr_t va = (vaddr_t)l->l_addr;
381 
382 	l->l_flag &= ~L_INMEM;
383 	uvm_uarea_free(va);
384 	l->l_addr = NULL;
385 }
386 
387 /*
388  * uvm_init_limit: init per-process VM limits
389  *
390  * - called for process 0 and then inherited by all others.
391  */
392 
393 void
394 uvm_init_limits(struct proc *p)
395 {
396 
397 	/*
398 	 * Set up the initial limits on process VM.  Set the maximum
399 	 * resident set size to be all of (reasonably) available memory.
400 	 * This causes any single, large process to start random page
401 	 * replacement once it fills memory.
402 	 */
403 
404 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
405 	p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
406 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
407 	p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
408 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
409 }
410 
411 #ifdef DEBUG
412 int	enableswap = 1;
413 int	swapdebug = 0;
414 #define	SDB_FOLLOW	1
415 #define SDB_SWAPIN	2
416 #define SDB_SWAPOUT	4
417 #endif
418 
419 /*
420  * uvm_swapin: swap in a process's u-area.
421  */
422 
423 void
424 uvm_swapin(struct lwp *l)
425 {
426 	vaddr_t addr;
427 	int s, error;
428 
429 	addr = (vaddr_t)l->l_addr;
430 	/* make L_INMEM true */
431 	error = uvm_fault_wire(kernel_map, addr, addr + USPACE, VM_FAULT_WIRE,
432 	    VM_PROT_READ | VM_PROT_WRITE);
433 	if (error) {
434 		panic("uvm_swapin: rewiring stack failed: %d", error);
435 	}
436 
437 	/*
438 	 * Some architectures need to be notified when the user area has
439 	 * moved to new physical page(s) (e.g.  see mips/mips/vm_machdep.c).
440 	 */
441 	cpu_swapin(l);
442 	SCHED_LOCK(s);
443 	if (l->l_stat == LSRUN)
444 		setrunqueue(l);
445 	l->l_flag |= L_INMEM;
446 	SCHED_UNLOCK(s);
447 	l->l_swtime = 0;
448 	++uvmexp.swapins;
449 }
450 
451 /*
452  * uvm_scheduler: process zero main loop
453  *
454  * - attempt to swapin every swaped-out, runnable process in order of
455  *	priority.
456  * - if not enough memory, wake the pagedaemon and let it clear space.
457  */
458 
459 void
460 uvm_scheduler(void)
461 {
462 	struct lwp *l, *ll;
463 	int pri;
464 	int ppri;
465 
466 loop:
467 #ifdef DEBUG
468 	while (!enableswap)
469 		tsleep(&proc0, PVM, "noswap", 0);
470 #endif
471 	ll = NULL;		/* process to choose */
472 	ppri = INT_MIN;	/* its priority */
473 	proclist_lock_read();
474 
475 	LIST_FOREACH(l, &alllwp, l_list) {
476 		/* is it a runnable swapped out process? */
477 		if (l->l_stat == LSRUN && (l->l_flag & L_INMEM) == 0) {
478 			pri = l->l_swtime + l->l_slptime -
479 			    (l->l_proc->p_nice - NZERO) * 8;
480 			if (pri > ppri) {   /* higher priority?  remember it. */
481 				ll = l;
482 				ppri = pri;
483 			}
484 		}
485 	}
486 	/*
487 	 * XXXSMP: possible unlock/sleep race between here and the
488 	 * "scheduler" tsleep below..
489 	 */
490 	proclist_unlock_read();
491 
492 #ifdef DEBUG
493 	if (swapdebug & SDB_FOLLOW)
494 		printf("scheduler: running, procp %p pri %d\n", ll, ppri);
495 #endif
496 	/*
497 	 * Nothing to do, back to sleep
498 	 */
499 	if ((l = ll) == NULL) {
500 		tsleep(&proc0, PVM, "scheduler", 0);
501 		goto loop;
502 	}
503 
504 	/*
505 	 * we have found swapped out process which we would like to bring
506 	 * back in.
507 	 *
508 	 * XXX: this part is really bogus cuz we could deadlock on memory
509 	 * despite our feeble check
510 	 */
511 	if (uvmexp.free > atop(USPACE)) {
512 #ifdef DEBUG
513 		if (swapdebug & SDB_SWAPIN)
514 			printf("swapin: pid %d(%s)@%p, pri %d free %d\n",
515 	     l->l_proc->p_pid, l->l_proc->p_comm, l->l_addr, ppri, uvmexp.free);
516 #endif
517 		uvm_swapin(l);
518 		goto loop;
519 	}
520 	/*
521 	 * not enough memory, jab the pageout daemon and wait til the coast
522 	 * is clear
523 	 */
524 #ifdef DEBUG
525 	if (swapdebug & SDB_FOLLOW)
526 		printf("scheduler: no room for pid %d(%s), free %d\n",
527 	   l->l_proc->p_pid, l->l_proc->p_comm, uvmexp.free);
528 #endif
529 	uvm_wait("schedpwait");
530 #ifdef DEBUG
531 	if (swapdebug & SDB_FOLLOW)
532 		printf("scheduler: room again, free %d\n", uvmexp.free);
533 #endif
534 	goto loop;
535 }
536 
537 /*
538  * swappable: is LWP "l" swappable?
539  */
540 
541 #define	swappable(l)							\
542 	(((l)->l_flag & (L_INMEM)) &&					\
543 	 ((((l)->l_proc->p_flag) & (P_SYSTEM | P_WEXIT)) == 0) &&	\
544 	 (l)->l_holdcnt == 0)
545 
546 /*
547  * swapout_threads: find threads that can be swapped and unwire their
548  *	u-areas.
549  *
550  * - called by the pagedaemon
551  * - try and swap at least one processs
552  * - processes that are sleeping or stopped for maxslp or more seconds
553  *   are swapped... otherwise the longest-sleeping or stopped process
554  *   is swapped, otherwise the longest resident process...
555  */
556 
557 void
558 uvm_swapout_threads(void)
559 {
560 	struct lwp *l;
561 	struct lwp *outl, *outl2;
562 	int outpri, outpri2;
563 	int didswap = 0;
564 	extern int maxslp;
565 	/* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
566 
567 #ifdef DEBUG
568 	if (!enableswap)
569 		return;
570 #endif
571 
572 	/*
573 	 * outl/outpri  : stop/sleep thread with largest sleeptime < maxslp
574 	 * outl2/outpri2: the longest resident thread (its swap time)
575 	 */
576 	outl = outl2 = NULL;
577 	outpri = outpri2 = 0;
578 	proclist_lock_read();
579 	LIST_FOREACH(l, &alllwp, l_list) {
580 		KASSERT(l->l_proc != NULL);
581 		if (!swappable(l))
582 			continue;
583 		switch (l->l_stat) {
584 		case LSONPROC:
585 			continue;
586 
587 		case LSRUN:
588 			if (l->l_swtime > outpri2) {
589 				outl2 = l;
590 				outpri2 = l->l_swtime;
591 			}
592 			continue;
593 
594 		case LSSLEEP:
595 		case LSSTOP:
596 			if (l->l_slptime >= maxslp) {
597 				uvm_swapout(l);
598 				didswap++;
599 			} else if (l->l_slptime > outpri) {
600 				outl = l;
601 				outpri = l->l_slptime;
602 			}
603 			continue;
604 		}
605 	}
606 	proclist_unlock_read();
607 
608 	/*
609 	 * If we didn't get rid of any real duds, toss out the next most
610 	 * likely sleeping/stopped or running candidate.  We only do this
611 	 * if we are real low on memory since we don't gain much by doing
612 	 * it (USPACE bytes).
613 	 */
614 	if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
615 		if ((l = outl) == NULL)
616 			l = outl2;
617 #ifdef DEBUG
618 		if (swapdebug & SDB_SWAPOUT)
619 			printf("swapout_threads: no duds, try procp %p\n", l);
620 #endif
621 		if (l)
622 			uvm_swapout(l);
623 	}
624 }
625 
626 /*
627  * uvm_swapout: swap out lwp "l"
628  *
629  * - currently "swapout" means "unwire U-area" and "pmap_collect()"
630  *   the pmap.
631  * - XXXCDC: should deactivate all process' private anonymous memory
632  */
633 
634 static void
635 uvm_swapout(struct lwp *l)
636 {
637 	vaddr_t addr;
638 	int s;
639 	struct proc *p = l->l_proc;
640 
641 #ifdef DEBUG
642 	if (swapdebug & SDB_SWAPOUT)
643 		printf("swapout: lid %d.%d(%s)@%p, stat %x pri %d free %d\n",
644 	   p->p_pid, l->l_lid, p->p_comm, l->l_addr, l->l_stat,
645 	   l->l_slptime, uvmexp.free);
646 #endif
647 
648 	/*
649 	 * Mark it as (potentially) swapped out.
650 	 */
651 	SCHED_LOCK(s);
652 	if (l->l_stat == LSONPROC) {
653 		KDASSERT(l->l_cpu != curcpu());
654 		SCHED_UNLOCK(s);
655 		return;
656 	}
657 	l->l_flag &= ~L_INMEM;
658 	if (l->l_stat == LSRUN)
659 		remrunqueue(l);
660 	SCHED_UNLOCK(s);
661 	l->l_swtime = 0;
662 	p->p_stats->p_ru.ru_nswap++;
663 	++uvmexp.swapouts;
664 
665 	/*
666 	 * Do any machine-specific actions necessary before swapout.
667 	 * This can include saving floating point state, etc.
668 	 */
669 	cpu_swapout(l);
670 
671 	/*
672 	 * Unwire the to-be-swapped process's user struct and kernel stack.
673 	 */
674 	addr = (vaddr_t)l->l_addr;
675 	uvm_fault_unwire(kernel_map, addr, addr + USPACE); /* !L_INMEM */
676 	pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
677 }
678 
679 /*
680  * uvm_coredump_walkmap: walk a process's map for the purpose of dumping
681  * a core file.
682  */
683 
684 int
685 uvm_coredump_walkmap(struct proc *p, void *iocookie,
686     int (*func)(struct proc *, void *, struct uvm_coredump_state *),
687     void *cookie)
688 {
689 	struct uvm_coredump_state state;
690 	struct vmspace *vm = p->p_vmspace;
691 	struct vm_map *map = &vm->vm_map;
692 	struct vm_map_entry *entry;
693 	int error;
694 
695 	entry = NULL;
696 	vm_map_lock_read(map);
697 	state.end = 0;
698 	for (;;) {
699 		if (entry == NULL)
700 			entry = map->header.next;
701 		else if (!uvm_map_lookup_entry(map, state.end, &entry))
702 			entry = entry->next;
703 		if (entry == &map->header)
704 			break;
705 
706 		state.cookie = cookie;
707 		if (state.end > entry->start) {
708 			state.start = state.end;
709 		} else {
710 			state.start = entry->start;
711 		}
712 		state.realend = entry->end;
713 		state.end = entry->end;
714 		state.prot = entry->protection;
715 		state.flags = 0;
716 
717 		/*
718 		 * Dump the region unless one of the following is true:
719 		 *
720 		 * (1) the region has neither object nor amap behind it
721 		 *     (ie. it has never been accessed).
722 		 *
723 		 * (2) the region has no amap and is read-only
724 		 *     (eg. an executable text section).
725 		 *
726 		 * (3) the region's object is a device.
727 		 *
728 		 * (4) the region is unreadable by the process.
729 		 */
730 
731 		KASSERT(!UVM_ET_ISSUBMAP(entry));
732 		KASSERT(state.start < VM_MAXUSER_ADDRESS);
733 		KASSERT(state.end <= VM_MAXUSER_ADDRESS);
734 		if (entry->object.uvm_obj == NULL &&
735 		    entry->aref.ar_amap == NULL) {
736 			state.realend = state.start;
737 		} else if ((entry->protection & VM_PROT_WRITE) == 0 &&
738 		    entry->aref.ar_amap == NULL) {
739 			state.realend = state.start;
740 		} else if (entry->object.uvm_obj != NULL &&
741 		    UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) {
742 			state.realend = state.start;
743 		} else if ((entry->protection & VM_PROT_READ) == 0) {
744 			state.realend = state.start;
745 		} else {
746 			if (state.start >= (vaddr_t)vm->vm_maxsaddr)
747 				state.flags |= UVM_COREDUMP_STACK;
748 
749 			/*
750 			 * If this an anonymous entry, only dump instantiated
751 			 * pages.
752 			 */
753 			if (entry->object.uvm_obj == NULL) {
754 				vaddr_t end;
755 
756 				amap_lock(entry->aref.ar_amap);
757 				for (end = state.start;
758 				     end < state.end; end += PAGE_SIZE) {
759 					struct vm_anon *anon;
760 					anon = amap_lookup(&entry->aref,
761 					    end - entry->start);
762 					/*
763 					 * If we have already encountered an
764 					 * uninstantiated page, stop at the
765 					 * first instantied page.
766 					 */
767 					if (anon != NULL &&
768 					    state.realend != state.end) {
769 						state.end = end;
770 						break;
771 					}
772 
773 					/*
774 					 * If this page is the first
775 					 * uninstantiated page, mark this as
776 					 * the real ending point.  Continue to
777 					 * counting uninstantiated pages.
778 					 */
779 					if (anon == NULL &&
780 					    state.realend == state.end) {
781 						state.realend = end;
782 					}
783 				}
784 				amap_unlock(entry->aref.ar_amap);
785 			}
786 		}
787 
788 
789 		vm_map_unlock_read(map);
790 		error = (*func)(p, iocookie, &state);
791 		if (error)
792 			return (error);
793 		vm_map_lock_read(map);
794 	}
795 	vm_map_unlock_read(map);
796 
797 	return (0);
798 }
799