xref: /netbsd-src/sys/uvm/uvm_glue.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: uvm_glue.c,v 1.95 2006/06/13 13:22:06 yamt 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 1.95 2006/06/13 13:22:06 yamt 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 static vaddr_t uvm_uareas;
99 static int uvm_nuarea;
100 static struct simplelock uvm_uareas_slock = SIMPLELOCK_INITIALIZER;
101 #define	UAREA_NEXTFREE(uarea)	(*(vaddr_t *)(UAREA_TO_USER(uarea)))
102 
103 static void uvm_uarea_free(vaddr_t);
104 
105 /*
106  * XXXCDC: do these really belong here?
107  */
108 
109 /*
110  * uvm_kernacc: can the kernel access a region of memory
111  *
112  * - used only by /dev/kmem driver (mem.c)
113  */
114 
115 boolean_t
116 uvm_kernacc(caddr_t addr, size_t len, int rw)
117 {
118 	boolean_t rv;
119 	vaddr_t saddr, eaddr;
120 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
121 
122 	saddr = trunc_page((vaddr_t)addr);
123 	eaddr = round_page((vaddr_t)addr + len);
124 	vm_map_lock_read(kernel_map);
125 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
126 	vm_map_unlock_read(kernel_map);
127 
128 	return(rv);
129 }
130 
131 #ifdef KGDB
132 /*
133  * Change protections on kernel pages from addr to addr+len
134  * (presumably so debugger can plant a breakpoint).
135  *
136  * We force the protection change at the pmap level.  If we were
137  * to use vm_map_protect a change to allow writing would be lazily-
138  * applied meaning we would still take a protection fault, something
139  * we really don't want to do.  It would also fragment the kernel
140  * map unnecessarily.  We cannot use pmap_protect since it also won't
141  * enforce a write-enable request.  Using pmap_enter is the only way
142  * we can ensure the change takes place properly.
143  */
144 void
145 uvm_chgkprot(caddr_t addr, size_t len, int rw)
146 {
147 	vm_prot_t prot;
148 	paddr_t pa;
149 	vaddr_t sva, eva;
150 
151 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
152 	eva = round_page((vaddr_t)addr + len);
153 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
154 		/*
155 		 * Extract physical address for the page.
156 		 */
157 		if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE)
158 			panic("chgkprot: invalid page");
159 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
160 	}
161 	pmap_update(pmap_kernel());
162 }
163 #endif
164 
165 /*
166  * uvm_vslock: wire user memory for I/O
167  *
168  * - called from physio and sys___sysctl
169  * - XXXCDC: consider nuking this (or making it a macro?)
170  */
171 
172 int
173 uvm_vslock(struct proc *p, caddr_t addr, size_t len, vm_prot_t access_type)
174 {
175 	struct vm_map *map;
176 	vaddr_t start, end;
177 	int error;
178 
179 	map = &p->p_vmspace->vm_map;
180 	start = trunc_page((vaddr_t)addr);
181 	end = round_page((vaddr_t)addr + len);
182 	error = uvm_fault_wire(map, start, end, access_type, 0);
183 	return error;
184 }
185 
186 /*
187  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
188  *
189  * - called from physio and sys___sysctl
190  * - XXXCDC: consider nuking this (or making it a macro?)
191  */
192 
193 void
194 uvm_vsunlock(struct proc *p, caddr_t addr, size_t len)
195 {
196 	uvm_fault_unwire(&p->p_vmspace->vm_map, trunc_page((vaddr_t)addr),
197 		round_page((vaddr_t)addr + len));
198 }
199 
200 /*
201  * uvm_proc_fork: fork a virtual address space
202  *
203  * - the address space is copied as per parent map's inherit values
204  */
205 void
206 uvm_proc_fork(struct proc *p1, struct proc *p2, boolean_t shared)
207 {
208 
209 	if (shared == TRUE) {
210 		p2->p_vmspace = NULL;
211 		uvmspace_share(p1, p2);
212 	} else {
213 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
214 	}
215 
216 	cpu_proc_fork(p1, p2);
217 }
218 
219 
220 /*
221  * uvm_lwp_fork: fork a thread
222  *
223  * - a new "user" structure is allocated for the child process
224  *	[filled in by MD layer...]
225  * - if specified, the child gets a new user stack described by
226  *	stack and stacksize
227  * - NOTE: the kernel stack may be at a different location in the child
228  *	process, and thus addresses of automatic variables may be invalid
229  *	after cpu_lwp_fork returns in the child process.  We do nothing here
230  *	after cpu_lwp_fork returns.
231  * - XXXCDC: we need a way for this to return a failure value rather
232  *   than just hang
233  */
234 void
235 uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
236     void (*func)(void *), void *arg)
237 {
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 		vaddr_t uarea = USER_TO_UAREA(l2->l_addr);
252 
253 		error = uvm_fault_wire(kernel_map, uarea,
254 		    uarea + USPACE, VM_PROT_READ | VM_PROT_WRITE, 0);
255 		if (error)
256 			panic("uvm_lwp_fork: uvm_fault_wire failed: %d", error);
257 #ifdef PMAP_UAREA
258 		/* Tell the pmap this is a u-area mapping */
259 		PMAP_UAREA(uarea);
260 #endif
261 		l2->l_flag |= L_INMEM;
262 	}
263 
264 #ifdef KSTACK_CHECK_MAGIC
265 	/*
266 	 * fill stack with magic number
267 	 */
268 	kstack_setup_magic(l2);
269 #endif
270 
271 	/*
272 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
273  	 * to run.  If this is a normal user fork, the child will exit
274 	 * directly to user mode via child_return() on its first time
275 	 * slice and will not return here.  If this is a kernel thread,
276 	 * the specified entry point will be executed.
277 	 */
278 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
279 }
280 
281 /*
282  * uvm_uarea_alloc: allocate a u-area
283  */
284 
285 boolean_t
286 uvm_uarea_alloc(vaddr_t *uaddrp)
287 {
288 	vaddr_t uaddr;
289 
290 #ifndef USPACE_ALIGN
291 #define USPACE_ALIGN    0
292 #endif
293 
294 	simple_lock(&uvm_uareas_slock);
295 	if (uvm_nuarea > 0) {
296 		uaddr = uvm_uareas;
297 		uvm_uareas = UAREA_NEXTFREE(uaddr);
298 		uvm_nuarea--;
299 		simple_unlock(&uvm_uareas_slock);
300 		*uaddrp = uaddr;
301 		return TRUE;
302 	} else {
303 		simple_unlock(&uvm_uareas_slock);
304 		*uaddrp = uvm_km_alloc(kernel_map, USPACE, USPACE_ALIGN,
305 		    UVM_KMF_PAGEABLE);
306 		return FALSE;
307 	}
308 }
309 
310 /*
311  * uvm_uarea_free: free a u-area; never blocks
312  */
313 
314 static inline void
315 uvm_uarea_free(vaddr_t uaddr)
316 {
317 	simple_lock(&uvm_uareas_slock);
318 	UAREA_NEXTFREE(uaddr) = uvm_uareas;
319 	uvm_uareas = uaddr;
320 	uvm_nuarea++;
321 	simple_unlock(&uvm_uareas_slock);
322 }
323 
324 /*
325  * uvm_uarea_drain: return memory of u-areas over limit
326  * back to system
327  */
328 
329 void
330 uvm_uarea_drain(boolean_t empty)
331 {
332 	int leave = empty ? 0 : UVM_NUAREA_MAX;
333 	vaddr_t uaddr;
334 
335 	if (uvm_nuarea <= leave)
336 		return;
337 
338 	simple_lock(&uvm_uareas_slock);
339 	while(uvm_nuarea > leave) {
340 		uaddr = uvm_uareas;
341 		uvm_uareas = UAREA_NEXTFREE(uaddr);
342 		uvm_nuarea--;
343 		simple_unlock(&uvm_uareas_slock);
344 		uvm_km_free(kernel_map, uaddr, USPACE, UVM_KMF_PAGEABLE);
345 		simple_lock(&uvm_uareas_slock);
346 	}
347 	simple_unlock(&uvm_uareas_slock);
348 }
349 
350 /*
351  * uvm_exit: exit a virtual address space
352  *
353  * - the process passed to us is a dead (pre-zombie) process; we
354  *   are running on a different context now (the reaper).
355  * - borrow proc0's address space because freeing the vmspace
356  *   of the dead process may block.
357  */
358 
359 void
360 uvm_proc_exit(struct proc *p)
361 {
362 	struct lwp *l = curlwp; /* XXX */
363 	struct vmspace *ovm;
364 
365 	KASSERT(p == l->l_proc);
366 	ovm = p->p_vmspace;
367 
368 	/*
369 	 * borrow proc0's address space.
370 	 */
371 	pmap_deactivate(l);
372 	p->p_vmspace = proc0.p_vmspace;
373 	pmap_activate(l);
374 
375 	uvmspace_free(ovm);
376 }
377 
378 void
379 uvm_lwp_exit(struct lwp *l)
380 {
381 	vaddr_t va = USER_TO_UAREA(l->l_addr);
382 
383 	l->l_flag &= ~L_INMEM;
384 	uvm_uarea_free(va);
385 	l->l_addr = NULL;
386 }
387 
388 /*
389  * uvm_init_limit: init per-process VM limits
390  *
391  * - called for process 0 and then inherited by all others.
392  */
393 
394 void
395 uvm_init_limits(struct proc *p)
396 {
397 
398 	/*
399 	 * Set up the initial limits on process VM.  Set the maximum
400 	 * resident set size to be all of (reasonably) available memory.
401 	 * This causes any single, large process to start random page
402 	 * replacement once it fills memory.
403 	 */
404 
405 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
406 	p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
407 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
408 	p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
409 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
410 }
411 
412 #ifdef DEBUG
413 int	enableswap = 1;
414 int	swapdebug = 0;
415 #define	SDB_FOLLOW	1
416 #define SDB_SWAPIN	2
417 #define SDB_SWAPOUT	4
418 #endif
419 
420 /*
421  * uvm_swapin: swap in an lwp's u-area.
422  */
423 
424 void
425 uvm_swapin(struct lwp *l)
426 {
427 	vaddr_t addr;
428 	int s, error;
429 
430 	addr = USER_TO_UAREA(l->l_addr);
431 	/* make L_INMEM true */
432 	error = uvm_fault_wire(kernel_map, addr, addr + USPACE,
433 	    VM_PROT_READ | VM_PROT_WRITE, 0);
434 	if (error) {
435 		panic("uvm_swapin: rewiring stack failed: %d", error);
436 	}
437 
438 	/*
439 	 * Some architectures need to be notified when the user area has
440 	 * moved to new physical page(s) (e.g.  see mips/mips/vm_machdep.c).
441 	 */
442 	cpu_swapin(l);
443 	SCHED_LOCK(s);
444 	if (l->l_stat == LSRUN)
445 		setrunqueue(l);
446 	l->l_flag |= L_INMEM;
447 	SCHED_UNLOCK(s);
448 	l->l_swtime = 0;
449 	++uvmexp.swapins;
450 }
451 
452 /*
453  * uvm_scheduler: process zero main loop
454  *
455  * - attempt to swapin every swaped-out, runnable process in order of
456  *	priority.
457  * - if not enough memory, wake the pagedaemon and let it clear space.
458  */
459 
460 void
461 uvm_scheduler(void)
462 {
463 	struct lwp *l, *ll;
464 	int pri;
465 	int ppri;
466 
467 loop:
468 #ifdef DEBUG
469 	while (!enableswap)
470 		tsleep(&proc0, PVM, "noswap", 0);
471 #endif
472 	ll = NULL;		/* process to choose */
473 	ppri = INT_MIN;	/* its priority */
474 	proclist_lock_read();
475 
476 	LIST_FOREACH(l, &alllwp, l_list) {
477 		/* is it a runnable swapped out process? */
478 		if (l->l_stat == LSRUN && (l->l_flag & L_INMEM) == 0) {
479 			pri = l->l_swtime + l->l_slptime -
480 			    (l->l_proc->p_nice - NZERO) * 8;
481 			if (pri > ppri) {   /* higher priority?  remember it. */
482 				ll = l;
483 				ppri = pri;
484 			}
485 		}
486 	}
487 	/*
488 	 * XXXSMP: possible unlock/sleep race between here and the
489 	 * "scheduler" tsleep below..
490 	 */
491 	proclist_unlock_read();
492 
493 #ifdef DEBUG
494 	if (swapdebug & SDB_FOLLOW)
495 		printf("scheduler: running, procp %p pri %d\n", ll, ppri);
496 #endif
497 	/*
498 	 * Nothing to do, back to sleep
499 	 */
500 	if ((l = ll) == NULL) {
501 		tsleep(&proc0, PVM, "scheduler", 0);
502 		goto loop;
503 	}
504 
505 	/*
506 	 * we have found swapped out process which we would like to bring
507 	 * back in.
508 	 *
509 	 * XXX: this part is really bogus cuz we could deadlock on memory
510 	 * despite our feeble check
511 	 */
512 	if (uvmexp.free > atop(USPACE)) {
513 #ifdef DEBUG
514 		if (swapdebug & SDB_SWAPIN)
515 			printf("swapin: pid %d(%s)@%p, pri %d free %d\n",
516 	     l->l_proc->p_pid, l->l_proc->p_comm, l->l_addr, ppri, uvmexp.free);
517 #endif
518 		uvm_swapin(l);
519 		goto loop;
520 	}
521 	/*
522 	 * not enough memory, jab the pageout daemon and wait til the coast
523 	 * is clear
524 	 */
525 #ifdef DEBUG
526 	if (swapdebug & SDB_FOLLOW)
527 		printf("scheduler: no room for pid %d(%s), free %d\n",
528 	   l->l_proc->p_pid, l->l_proc->p_comm, uvmexp.free);
529 #endif
530 	uvm_wait("schedpwait");
531 #ifdef DEBUG
532 	if (swapdebug & SDB_FOLLOW)
533 		printf("scheduler: room again, free %d\n", uvmexp.free);
534 #endif
535 	goto loop;
536 }
537 
538 /*
539  * swappable: is LWP "l" swappable?
540  */
541 
542 #define	swappable(l)							\
543 	(((l)->l_flag & (L_INMEM)) &&					\
544 	 ((((l)->l_proc->p_flag) & (P_SYSTEM | P_WEXIT)) == 0) &&	\
545 	 (l)->l_holdcnt == 0)
546 
547 /*
548  * swapout_threads: find threads that can be swapped and unwire their
549  *	u-areas.
550  *
551  * - called by the pagedaemon
552  * - try and swap at least one processs
553  * - processes that are sleeping or stopped for maxslp or more seconds
554  *   are swapped... otherwise the longest-sleeping or stopped process
555  *   is swapped, otherwise the longest resident process...
556  */
557 
558 void
559 uvm_swapout_threads(void)
560 {
561 	struct lwp *l;
562 	struct lwp *outl, *outl2;
563 	int outpri, outpri2;
564 	int didswap = 0;
565 	extern int maxslp;
566 	/* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
567 
568 #ifdef DEBUG
569 	if (!enableswap)
570 		return;
571 #endif
572 
573 	/*
574 	 * outl/outpri  : stop/sleep thread with largest sleeptime < maxslp
575 	 * outl2/outpri2: the longest resident thread (its swap time)
576 	 */
577 	outl = outl2 = NULL;
578 	outpri = outpri2 = 0;
579 	proclist_lock_read();
580 	LIST_FOREACH(l, &alllwp, l_list) {
581 		KASSERT(l->l_proc != NULL);
582 		if (!swappable(l))
583 			continue;
584 		switch (l->l_stat) {
585 		case LSONPROC:
586 			continue;
587 
588 		case LSRUN:
589 			if (l->l_swtime > outpri2) {
590 				outl2 = l;
591 				outpri2 = l->l_swtime;
592 			}
593 			continue;
594 
595 		case LSSLEEP:
596 		case LSSTOP:
597 			if (l->l_slptime >= maxslp) {
598 				uvm_swapout(l);
599 				didswap++;
600 			} else if (l->l_slptime > outpri) {
601 				outl = l;
602 				outpri = l->l_slptime;
603 			}
604 			continue;
605 		}
606 	}
607 	proclist_unlock_read();
608 
609 	/*
610 	 * If we didn't get rid of any real duds, toss out the next most
611 	 * likely sleeping/stopped or running candidate.  We only do this
612 	 * if we are real low on memory since we don't gain much by doing
613 	 * it (USPACE bytes).
614 	 */
615 	if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
616 		if ((l = outl) == NULL)
617 			l = outl2;
618 #ifdef DEBUG
619 		if (swapdebug & SDB_SWAPOUT)
620 			printf("swapout_threads: no duds, try procp %p\n", l);
621 #endif
622 		if (l)
623 			uvm_swapout(l);
624 	}
625 }
626 
627 /*
628  * uvm_swapout: swap out lwp "l"
629  *
630  * - currently "swapout" means "unwire U-area" and "pmap_collect()"
631  *   the pmap.
632  * - XXXCDC: should deactivate all process' private anonymous memory
633  */
634 
635 static void
636 uvm_swapout(struct lwp *l)
637 {
638 	vaddr_t addr;
639 	int s;
640 	struct proc *p = l->l_proc;
641 
642 #ifdef DEBUG
643 	if (swapdebug & SDB_SWAPOUT)
644 		printf("swapout: lid %d.%d(%s)@%p, stat %x pri %d free %d\n",
645 	   p->p_pid, l->l_lid, p->p_comm, l->l_addr, l->l_stat,
646 	   l->l_slptime, uvmexp.free);
647 #endif
648 
649 	/*
650 	 * Mark it as (potentially) swapped out.
651 	 */
652 	SCHED_LOCK(s);
653 	if (l->l_stat == LSONPROC) {
654 		KDASSERT(l->l_cpu != curcpu());
655 		SCHED_UNLOCK(s);
656 		return;
657 	}
658 	l->l_flag &= ~L_INMEM;
659 	if (l->l_stat == LSRUN)
660 		remrunqueue(l);
661 	SCHED_UNLOCK(s);
662 	l->l_swtime = 0;
663 	p->p_stats->p_ru.ru_nswap++;
664 	++uvmexp.swapouts;
665 
666 	/*
667 	 * Do any machine-specific actions necessary before swapout.
668 	 * This can include saving floating point state, etc.
669 	 */
670 	cpu_swapout(l);
671 
672 	/*
673 	 * Unwire the to-be-swapped process's user struct and kernel stack.
674 	 */
675 	addr = USER_TO_UAREA(l->l_addr);
676 	uvm_fault_unwire(kernel_map, addr, addr + USPACE); /* !L_INMEM */
677 	pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
678 }
679 
680 /*
681  * uvm_coredump_walkmap: walk a process's map for the purpose of dumping
682  * a core file.
683  */
684 
685 int
686 uvm_coredump_walkmap(struct proc *p, void *iocookie,
687     int (*func)(struct proc *, void *, struct uvm_coredump_state *),
688     void *cookie)
689 {
690 	struct uvm_coredump_state state;
691 	struct vmspace *vm = p->p_vmspace;
692 	struct vm_map *map = &vm->vm_map;
693 	struct vm_map_entry *entry;
694 	int error;
695 
696 	entry = NULL;
697 	vm_map_lock_read(map);
698 	state.end = 0;
699 	for (;;) {
700 		if (entry == NULL)
701 			entry = map->header.next;
702 		else if (!uvm_map_lookup_entry(map, state.end, &entry))
703 			entry = entry->next;
704 		if (entry == &map->header)
705 			break;
706 
707 		state.cookie = cookie;
708 		if (state.end > entry->start) {
709 			state.start = state.end;
710 		} else {
711 			state.start = entry->start;
712 		}
713 		state.realend = entry->end;
714 		state.end = entry->end;
715 		state.prot = entry->protection;
716 		state.flags = 0;
717 
718 		/*
719 		 * Dump the region unless one of the following is true:
720 		 *
721 		 * (1) the region has neither object nor amap behind it
722 		 *     (ie. it has never been accessed).
723 		 *
724 		 * (2) the region has no amap and is read-only
725 		 *     (eg. an executable text section).
726 		 *
727 		 * (3) the region's object is a device.
728 		 *
729 		 * (4) the region is unreadable by the process.
730 		 */
731 
732 		KASSERT(!UVM_ET_ISSUBMAP(entry));
733 		KASSERT(state.start < VM_MAXUSER_ADDRESS);
734 		KASSERT(state.end <= VM_MAXUSER_ADDRESS);
735 		if (entry->object.uvm_obj == NULL &&
736 		    entry->aref.ar_amap == NULL) {
737 			state.realend = state.start;
738 		} else if ((entry->protection & VM_PROT_WRITE) == 0 &&
739 		    entry->aref.ar_amap == NULL) {
740 			state.realend = state.start;
741 		} else if (entry->object.uvm_obj != NULL &&
742 		    UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) {
743 			state.realend = state.start;
744 		} else if ((entry->protection & VM_PROT_READ) == 0) {
745 			state.realend = state.start;
746 		} else {
747 			if (state.start >= (vaddr_t)vm->vm_maxsaddr)
748 				state.flags |= UVM_COREDUMP_STACK;
749 
750 			/*
751 			 * If this an anonymous entry, only dump instantiated
752 			 * pages.
753 			 */
754 			if (entry->object.uvm_obj == NULL) {
755 				vaddr_t end;
756 
757 				amap_lock(entry->aref.ar_amap);
758 				for (end = state.start;
759 				     end < state.end; end += PAGE_SIZE) {
760 					struct vm_anon *anon;
761 					anon = amap_lookup(&entry->aref,
762 					    end - entry->start);
763 					/*
764 					 * If we have already encountered an
765 					 * uninstantiated page, stop at the
766 					 * first instantied page.
767 					 */
768 					if (anon != NULL &&
769 					    state.realend != state.end) {
770 						state.end = end;
771 						break;
772 					}
773 
774 					/*
775 					 * If this page is the first
776 					 * uninstantiated page, mark this as
777 					 * the real ending point.  Continue to
778 					 * counting uninstantiated pages.
779 					 */
780 					if (anon == NULL &&
781 					    state.realend == state.end) {
782 						state.realend = end;
783 					}
784 				}
785 				amap_unlock(entry->aref.ar_amap);
786 			}
787 		}
788 
789 
790 		vm_map_unlock_read(map);
791 		error = (*func)(p, iocookie, &state);
792 		if (error)
793 			return (error);
794 		vm_map_lock_read(map);
795 	}
796 	vm_map_unlock_read(map);
797 
798 	return (0);
799 }
800