xref: /netbsd-src/sys/uvm/uvm_glue.c (revision c71562d660be5e4ad22016bce45e96f08af190cc)
1 /*	$NetBSD: uvm_glue.c,v 1.93 2006/03/15 18:09:25 drochner 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 1.93 2006/03/15 18:09:25 drochner 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, access_type, 0);
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_PROT_READ | VM_PROT_WRITE, 0);
253 		if (error)
254 			panic("uvm_lwp_fork: uvm_fault_wire failed: %d", error);
255 #ifdef PMAP_UAREA
256 		/* Tell the pmap this is a u-area mapping */
257 		PMAP_UAREA((vaddr_t)up);
258 #endif
259 		l2->l_flag |= L_INMEM;
260 	}
261 
262 #ifdef KSTACK_CHECK_MAGIC
263 	/*
264 	 * fill stack with magic number
265 	 */
266 	kstack_setup_magic(l2);
267 #endif
268 
269 	/*
270 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
271  	 * to run.  If this is a normal user fork, the child will exit
272 	 * directly to user mode via child_return() on its first time
273 	 * slice and will not return here.  If this is a kernel thread,
274 	 * the specified entry point will be executed.
275 	 */
276 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
277 }
278 
279 /*
280  * uvm_uarea_alloc: allocate a u-area
281  */
282 
283 boolean_t
284 uvm_uarea_alloc(vaddr_t *uaddrp)
285 {
286 	vaddr_t uaddr;
287 
288 #ifndef USPACE_ALIGN
289 #define USPACE_ALIGN    0
290 #endif
291 
292 	simple_lock(&uvm_uareas_slock);
293 	if (uvm_nuarea > 0) {
294 		uaddr = (vaddr_t)uvm_uareas;
295 		uvm_uareas = *(void **)uvm_uareas;
296 		uvm_nuarea--;
297 		simple_unlock(&uvm_uareas_slock);
298 		*uaddrp = uaddr;
299 		return TRUE;
300 	} else {
301 		simple_unlock(&uvm_uareas_slock);
302 		*uaddrp = uvm_km_alloc(kernel_map, USPACE, USPACE_ALIGN,
303 		    UVM_KMF_PAGEABLE);
304 		return FALSE;
305 	}
306 }
307 
308 /*
309  * uvm_uarea_free: free a u-area; never blocks
310  */
311 
312 static inline void
313 uvm_uarea_free(vaddr_t uaddr)
314 {
315 	simple_lock(&uvm_uareas_slock);
316 	*(void **)uaddr = uvm_uareas;
317 	uvm_uareas = (void *)uaddr;
318 	uvm_nuarea++;
319 	simple_unlock(&uvm_uareas_slock);
320 }
321 
322 /*
323  * uvm_uarea_drain: return memory of u-areas over limit
324  * back to system
325  */
326 
327 void
328 uvm_uarea_drain(boolean_t empty)
329 {
330 	int leave = empty ? 0 : UVM_NUAREA_MAX;
331 	vaddr_t uaddr;
332 
333 	if (uvm_nuarea <= leave)
334 		return;
335 
336 	simple_lock(&uvm_uareas_slock);
337 	while(uvm_nuarea > leave) {
338 		uaddr = (vaddr_t)uvm_uareas;
339 		uvm_uareas = *(void **)uvm_uareas;
340 		uvm_nuarea--;
341 		simple_unlock(&uvm_uareas_slock);
342 		uvm_km_free(kernel_map, uaddr, USPACE, UVM_KMF_PAGEABLE);
343 		simple_lock(&uvm_uareas_slock);
344 	}
345 	simple_unlock(&uvm_uareas_slock);
346 }
347 
348 /*
349  * uvm_exit: exit a virtual address space
350  *
351  * - the process passed to us is a dead (pre-zombie) process; we
352  *   are running on a different context now (the reaper).
353  * - borrow proc0's address space because freeing the vmspace
354  *   of the dead process may block.
355  */
356 
357 void
358 uvm_proc_exit(struct proc *p)
359 {
360 	struct lwp *l = curlwp; /* XXX */
361 	struct vmspace *ovm;
362 
363 	KASSERT(p == l->l_proc);
364 	ovm = p->p_vmspace;
365 
366 	/*
367 	 * borrow proc0's address space.
368 	 */
369 	pmap_deactivate(l);
370 	p->p_vmspace = proc0.p_vmspace;
371 	pmap_activate(l);
372 
373 	uvmspace_free(ovm);
374 }
375 
376 void
377 uvm_lwp_exit(struct lwp *l)
378 {
379 	vaddr_t va = (vaddr_t)l->l_addr;
380 
381 	l->l_flag &= ~L_INMEM;
382 	uvm_uarea_free(va);
383 	l->l_addr = NULL;
384 }
385 
386 /*
387  * uvm_init_limit: init per-process VM limits
388  *
389  * - called for process 0 and then inherited by all others.
390  */
391 
392 void
393 uvm_init_limits(struct proc *p)
394 {
395 
396 	/*
397 	 * Set up the initial limits on process VM.  Set the maximum
398 	 * resident set size to be all of (reasonably) available memory.
399 	 * This causes any single, large process to start random page
400 	 * replacement once it fills memory.
401 	 */
402 
403 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
404 	p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
405 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
406 	p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
407 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
408 }
409 
410 #ifdef DEBUG
411 int	enableswap = 1;
412 int	swapdebug = 0;
413 #define	SDB_FOLLOW	1
414 #define SDB_SWAPIN	2
415 #define SDB_SWAPOUT	4
416 #endif
417 
418 /*
419  * uvm_swapin: swap in a process's u-area.
420  */
421 
422 void
423 uvm_swapin(struct lwp *l)
424 {
425 	vaddr_t addr;
426 	int s, error;
427 
428 	addr = (vaddr_t)l->l_addr;
429 	/* make L_INMEM true */
430 	error = uvm_fault_wire(kernel_map, addr, addr + USPACE,
431 	    VM_PROT_READ | VM_PROT_WRITE, 0);
432 	if (error) {
433 		panic("uvm_swapin: rewiring stack failed: %d", error);
434 	}
435 
436 	/*
437 	 * Some architectures need to be notified when the user area has
438 	 * moved to new physical page(s) (e.g.  see mips/mips/vm_machdep.c).
439 	 */
440 	cpu_swapin(l);
441 	SCHED_LOCK(s);
442 	if (l->l_stat == LSRUN)
443 		setrunqueue(l);
444 	l->l_flag |= L_INMEM;
445 	SCHED_UNLOCK(s);
446 	l->l_swtime = 0;
447 	++uvmexp.swapins;
448 }
449 
450 /*
451  * uvm_scheduler: process zero main loop
452  *
453  * - attempt to swapin every swaped-out, runnable process in order of
454  *	priority.
455  * - if not enough memory, wake the pagedaemon and let it clear space.
456  */
457 
458 void
459 uvm_scheduler(void)
460 {
461 	struct lwp *l, *ll;
462 	int pri;
463 	int ppri;
464 
465 loop:
466 #ifdef DEBUG
467 	while (!enableswap)
468 		tsleep(&proc0, PVM, "noswap", 0);
469 #endif
470 	ll = NULL;		/* process to choose */
471 	ppri = INT_MIN;	/* its priority */
472 	proclist_lock_read();
473 
474 	LIST_FOREACH(l, &alllwp, l_list) {
475 		/* is it a runnable swapped out process? */
476 		if (l->l_stat == LSRUN && (l->l_flag & L_INMEM) == 0) {
477 			pri = l->l_swtime + l->l_slptime -
478 			    (l->l_proc->p_nice - NZERO) * 8;
479 			if (pri > ppri) {   /* higher priority?  remember it. */
480 				ll = l;
481 				ppri = pri;
482 			}
483 		}
484 	}
485 	/*
486 	 * XXXSMP: possible unlock/sleep race between here and the
487 	 * "scheduler" tsleep below..
488 	 */
489 	proclist_unlock_read();
490 
491 #ifdef DEBUG
492 	if (swapdebug & SDB_FOLLOW)
493 		printf("scheduler: running, procp %p pri %d\n", ll, ppri);
494 #endif
495 	/*
496 	 * Nothing to do, back to sleep
497 	 */
498 	if ((l = ll) == NULL) {
499 		tsleep(&proc0, PVM, "scheduler", 0);
500 		goto loop;
501 	}
502 
503 	/*
504 	 * we have found swapped out process which we would like to bring
505 	 * back in.
506 	 *
507 	 * XXX: this part is really bogus cuz we could deadlock on memory
508 	 * despite our feeble check
509 	 */
510 	if (uvmexp.free > atop(USPACE)) {
511 #ifdef DEBUG
512 		if (swapdebug & SDB_SWAPIN)
513 			printf("swapin: pid %d(%s)@%p, pri %d free %d\n",
514 	     l->l_proc->p_pid, l->l_proc->p_comm, l->l_addr, ppri, uvmexp.free);
515 #endif
516 		uvm_swapin(l);
517 		goto loop;
518 	}
519 	/*
520 	 * not enough memory, jab the pageout daemon and wait til the coast
521 	 * is clear
522 	 */
523 #ifdef DEBUG
524 	if (swapdebug & SDB_FOLLOW)
525 		printf("scheduler: no room for pid %d(%s), free %d\n",
526 	   l->l_proc->p_pid, l->l_proc->p_comm, uvmexp.free);
527 #endif
528 	uvm_wait("schedpwait");
529 #ifdef DEBUG
530 	if (swapdebug & SDB_FOLLOW)
531 		printf("scheduler: room again, free %d\n", uvmexp.free);
532 #endif
533 	goto loop;
534 }
535 
536 /*
537  * swappable: is LWP "l" swappable?
538  */
539 
540 #define	swappable(l)							\
541 	(((l)->l_flag & (L_INMEM)) &&					\
542 	 ((((l)->l_proc->p_flag) & (P_SYSTEM | P_WEXIT)) == 0) &&	\
543 	 (l)->l_holdcnt == 0)
544 
545 /*
546  * swapout_threads: find threads that can be swapped and unwire their
547  *	u-areas.
548  *
549  * - called by the pagedaemon
550  * - try and swap at least one processs
551  * - processes that are sleeping or stopped for maxslp or more seconds
552  *   are swapped... otherwise the longest-sleeping or stopped process
553  *   is swapped, otherwise the longest resident process...
554  */
555 
556 void
557 uvm_swapout_threads(void)
558 {
559 	struct lwp *l;
560 	struct lwp *outl, *outl2;
561 	int outpri, outpri2;
562 	int didswap = 0;
563 	extern int maxslp;
564 	/* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
565 
566 #ifdef DEBUG
567 	if (!enableswap)
568 		return;
569 #endif
570 
571 	/*
572 	 * outl/outpri  : stop/sleep thread with largest sleeptime < maxslp
573 	 * outl2/outpri2: the longest resident thread (its swap time)
574 	 */
575 	outl = outl2 = NULL;
576 	outpri = outpri2 = 0;
577 	proclist_lock_read();
578 	LIST_FOREACH(l, &alllwp, l_list) {
579 		KASSERT(l->l_proc != NULL);
580 		if (!swappable(l))
581 			continue;
582 		switch (l->l_stat) {
583 		case LSONPROC:
584 			continue;
585 
586 		case LSRUN:
587 			if (l->l_swtime > outpri2) {
588 				outl2 = l;
589 				outpri2 = l->l_swtime;
590 			}
591 			continue;
592 
593 		case LSSLEEP:
594 		case LSSTOP:
595 			if (l->l_slptime >= maxslp) {
596 				uvm_swapout(l);
597 				didswap++;
598 			} else if (l->l_slptime > outpri) {
599 				outl = l;
600 				outpri = l->l_slptime;
601 			}
602 			continue;
603 		}
604 	}
605 	proclist_unlock_read();
606 
607 	/*
608 	 * If we didn't get rid of any real duds, toss out the next most
609 	 * likely sleeping/stopped or running candidate.  We only do this
610 	 * if we are real low on memory since we don't gain much by doing
611 	 * it (USPACE bytes).
612 	 */
613 	if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
614 		if ((l = outl) == NULL)
615 			l = outl2;
616 #ifdef DEBUG
617 		if (swapdebug & SDB_SWAPOUT)
618 			printf("swapout_threads: no duds, try procp %p\n", l);
619 #endif
620 		if (l)
621 			uvm_swapout(l);
622 	}
623 }
624 
625 /*
626  * uvm_swapout: swap out lwp "l"
627  *
628  * - currently "swapout" means "unwire U-area" and "pmap_collect()"
629  *   the pmap.
630  * - XXXCDC: should deactivate all process' private anonymous memory
631  */
632 
633 static void
634 uvm_swapout(struct lwp *l)
635 {
636 	vaddr_t addr;
637 	int s;
638 	struct proc *p = l->l_proc;
639 
640 #ifdef DEBUG
641 	if (swapdebug & SDB_SWAPOUT)
642 		printf("swapout: lid %d.%d(%s)@%p, stat %x pri %d free %d\n",
643 	   p->p_pid, l->l_lid, p->p_comm, l->l_addr, l->l_stat,
644 	   l->l_slptime, uvmexp.free);
645 #endif
646 
647 	/*
648 	 * Mark it as (potentially) swapped out.
649 	 */
650 	SCHED_LOCK(s);
651 	if (l->l_stat == LSONPROC) {
652 		KDASSERT(l->l_cpu != curcpu());
653 		SCHED_UNLOCK(s);
654 		return;
655 	}
656 	l->l_flag &= ~L_INMEM;
657 	if (l->l_stat == LSRUN)
658 		remrunqueue(l);
659 	SCHED_UNLOCK(s);
660 	l->l_swtime = 0;
661 	p->p_stats->p_ru.ru_nswap++;
662 	++uvmexp.swapouts;
663 
664 	/*
665 	 * Do any machine-specific actions necessary before swapout.
666 	 * This can include saving floating point state, etc.
667 	 */
668 	cpu_swapout(l);
669 
670 	/*
671 	 * Unwire the to-be-swapped process's user struct and kernel stack.
672 	 */
673 	addr = (vaddr_t)l->l_addr;
674 	uvm_fault_unwire(kernel_map, addr, addr + USPACE); /* !L_INMEM */
675 	pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
676 }
677 
678 /*
679  * uvm_coredump_walkmap: walk a process's map for the purpose of dumping
680  * a core file.
681  */
682 
683 int
684 uvm_coredump_walkmap(struct proc *p, void *iocookie,
685     int (*func)(struct proc *, void *, struct uvm_coredump_state *),
686     void *cookie)
687 {
688 	struct uvm_coredump_state state;
689 	struct vmspace *vm = p->p_vmspace;
690 	struct vm_map *map = &vm->vm_map;
691 	struct vm_map_entry *entry;
692 	int error;
693 
694 	entry = NULL;
695 	vm_map_lock_read(map);
696 	state.end = 0;
697 	for (;;) {
698 		if (entry == NULL)
699 			entry = map->header.next;
700 		else if (!uvm_map_lookup_entry(map, state.end, &entry))
701 			entry = entry->next;
702 		if (entry == &map->header)
703 			break;
704 
705 		state.cookie = cookie;
706 		if (state.end > entry->start) {
707 			state.start = state.end;
708 		} else {
709 			state.start = entry->start;
710 		}
711 		state.realend = entry->end;
712 		state.end = entry->end;
713 		state.prot = entry->protection;
714 		state.flags = 0;
715 
716 		/*
717 		 * Dump the region unless one of the following is true:
718 		 *
719 		 * (1) the region has neither object nor amap behind it
720 		 *     (ie. it has never been accessed).
721 		 *
722 		 * (2) the region has no amap and is read-only
723 		 *     (eg. an executable text section).
724 		 *
725 		 * (3) the region's object is a device.
726 		 *
727 		 * (4) the region is unreadable by the process.
728 		 */
729 
730 		KASSERT(!UVM_ET_ISSUBMAP(entry));
731 		KASSERT(state.start < VM_MAXUSER_ADDRESS);
732 		KASSERT(state.end <= VM_MAXUSER_ADDRESS);
733 		if (entry->object.uvm_obj == NULL &&
734 		    entry->aref.ar_amap == NULL) {
735 			state.realend = state.start;
736 		} else if ((entry->protection & VM_PROT_WRITE) == 0 &&
737 		    entry->aref.ar_amap == NULL) {
738 			state.realend = state.start;
739 		} else if (entry->object.uvm_obj != NULL &&
740 		    UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) {
741 			state.realend = state.start;
742 		} else if ((entry->protection & VM_PROT_READ) == 0) {
743 			state.realend = state.start;
744 		} else {
745 			if (state.start >= (vaddr_t)vm->vm_maxsaddr)
746 				state.flags |= UVM_COREDUMP_STACK;
747 
748 			/*
749 			 * If this an anonymous entry, only dump instantiated
750 			 * pages.
751 			 */
752 			if (entry->object.uvm_obj == NULL) {
753 				vaddr_t end;
754 
755 				amap_lock(entry->aref.ar_amap);
756 				for (end = state.start;
757 				     end < state.end; end += PAGE_SIZE) {
758 					struct vm_anon *anon;
759 					anon = amap_lookup(&entry->aref,
760 					    end - entry->start);
761 					/*
762 					 * If we have already encountered an
763 					 * uninstantiated page, stop at the
764 					 * first instantied page.
765 					 */
766 					if (anon != NULL &&
767 					    state.realend != state.end) {
768 						state.end = end;
769 						break;
770 					}
771 
772 					/*
773 					 * If this page is the first
774 					 * uninstantiated page, mark this as
775 					 * the real ending point.  Continue to
776 					 * counting uninstantiated pages.
777 					 */
778 					if (anon == NULL &&
779 					    state.realend == state.end) {
780 						state.realend = end;
781 					}
782 				}
783 				amap_unlock(entry->aref.ar_amap);
784 			}
785 		}
786 
787 
788 		vm_map_unlock_read(map);
789 		error = (*func)(p, iocookie, &state);
790 		if (error)
791 			return (error);
792 		vm_map_lock_read(map);
793 	}
794 	vm_map_unlock_read(map);
795 
796 	return (0);
797 }
798