xref: /freebsd/sys/kern/subr_rman.c (revision 43faedc1339a9624c7acedb7f3e5624e64da5b99)
1 /*-
2  * Copyright 1998 Massachusetts Institute of Technology
3  *
4  * Permission to use, copy, modify, and distribute this software and
5  * its documentation for any purpose and without fee is hereby
6  * granted, provided that both the above copyright notice and this
7  * permission notice appear in all copies, that both the above
8  * copyright notice and this permission notice appear in all
9  * supporting documentation, and that the name of M.I.T. not be used
10  * in advertising or publicity pertaining to distribution of the
11  * software without specific, written prior permission.  M.I.T. makes
12  * no representations about the suitability of this software for any
13  * purpose.  It is provided "as is" without express or implied
14  * warranty.
15  *
16  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
17  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * The kernel resource manager.  This code is responsible for keeping track
32  * of hardware resources which are apportioned out to various drivers.
33  * It does not actually assign those resources, and it is not expected
34  * that end-device drivers will call into this code directly.  Rather,
35  * the code which implements the buses that those devices are attached to,
36  * and the code which manages CPU resources, will call this code, and the
37  * end-device drivers will make upcalls to that code to actually perform
38  * the allocation.
39  *
40  * There are two sorts of resources managed by this code.  The first is
41  * the more familiar array (RMAN_ARRAY) type; resources in this class
42  * consist of a sequence of individually-allocatable objects which have
43  * been numbered in some well-defined order.  Most of the resources
44  * are of this type, as it is the most familiar.  The second type is
45  * called a gauge (RMAN_GAUGE), and models fungible resources (i.e.,
46  * resources in which each instance is indistinguishable from every
47  * other instance).  The principal anticipated application of gauges
48  * is in the context of power consumption, where a bus may have a specific
49  * power budget which all attached devices share.  RMAN_GAUGE is not
50  * implemented yet.
51  *
52  * For array resources, we make one simplifying assumption: two clients
53  * sharing the same resource must use the same range of indices.  That
54  * is to say, sharing of overlapping-but-not-identical regions is not
55  * permitted.
56  */
57 
58 #include "opt_ddb.h"
59 
60 #include <sys/cdefs.h>
61 __FBSDID("$FreeBSD$");
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/kernel.h>
66 #include <sys/limits.h>
67 #include <sys/lock.h>
68 #include <sys/malloc.h>
69 #include <sys/mutex.h>
70 #include <sys/bus.h>		/* XXX debugging */
71 #include <machine/bus.h>
72 #include <sys/rman.h>
73 #include <sys/sysctl.h>
74 
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #endif
78 
79 /*
80  * We use a linked list rather than a bitmap because we need to be able to
81  * represent potentially huge objects (like all of a processor's physical
82  * address space).  That is also why the indices are defined to have type
83  * `unsigned long' -- that being the largest integral type in ISO C (1990).
84  * The 1999 version of C allows `long long'; we may need to switch to that
85  * at some point in the future, particularly if we want to support 36-bit
86  * addresses on IA32 hardware.
87  */
88 struct resource_i {
89 	struct resource		r_r;
90 	TAILQ_ENTRY(resource_i)	r_link;
91 	LIST_ENTRY(resource_i)	r_sharelink;
92 	LIST_HEAD(, resource_i)	*r_sharehead;
93 	rman_res_t	r_start;	/* index of the first entry in this resource */
94 	rman_res_t	r_end;		/* index of the last entry (inclusive) */
95 	u_int	r_flags;
96 	void	*r_virtual;	/* virtual address of this resource */
97 	struct device *r_dev;	/* device which has allocated this resource */
98 	struct rman *r_rm;	/* resource manager from whence this came */
99 	int	r_rid;		/* optional rid for this resource. */
100 };
101 
102 static int rman_debug = 0;
103 SYSCTL_INT(_debug, OID_AUTO, rman_debug, CTLFLAG_RWTUN,
104     &rman_debug, 0, "rman debug");
105 
106 #define DPRINTF(params) if (rman_debug) printf params
107 
108 static MALLOC_DEFINE(M_RMAN, "rman", "Resource manager");
109 
110 struct rman_head rman_head;
111 static struct mtx rman_mtx; /* mutex to protect rman_head */
112 static int int_rman_release_resource(struct rman *rm, struct resource_i *r);
113 
114 static __inline struct resource_i *
115 int_alloc_resource(int malloc_flag)
116 {
117 	struct resource_i *r;
118 
119 	r = malloc(sizeof *r, M_RMAN, malloc_flag | M_ZERO);
120 	if (r != NULL) {
121 		r->r_r.__r_i = r;
122 	}
123 	return (r);
124 }
125 
126 int
127 rman_init(struct rman *rm)
128 {
129 	static int once = 0;
130 
131 	if (once == 0) {
132 		once = 1;
133 		TAILQ_INIT(&rman_head);
134 		mtx_init(&rman_mtx, "rman head", NULL, MTX_DEF);
135 	}
136 
137 	if (rm->rm_start == 0 && rm->rm_end == 0)
138 		rm->rm_end = ~0ul;
139 	if (rm->rm_type == RMAN_UNINIT)
140 		panic("rman_init");
141 	if (rm->rm_type == RMAN_GAUGE)
142 		panic("implement RMAN_GAUGE");
143 
144 	TAILQ_INIT(&rm->rm_list);
145 	rm->rm_mtx = malloc(sizeof *rm->rm_mtx, M_RMAN, M_NOWAIT | M_ZERO);
146 	if (rm->rm_mtx == NULL)
147 		return ENOMEM;
148 	mtx_init(rm->rm_mtx, "rman", NULL, MTX_DEF);
149 
150 	mtx_lock(&rman_mtx);
151 	TAILQ_INSERT_TAIL(&rman_head, rm, rm_link);
152 	mtx_unlock(&rman_mtx);
153 	return 0;
154 }
155 
156 int
157 rman_manage_region(struct rman *rm, rman_res_t start, rman_res_t end)
158 {
159 	struct resource_i *r, *s, *t;
160 	int rv = 0;
161 
162 	DPRINTF(("rman_manage_region: <%s> request: start %#lx, end %#lx\n",
163 	    rm->rm_descr, start, end));
164 	if (start < rm->rm_start || end > rm->rm_end)
165 		return EINVAL;
166 	r = int_alloc_resource(M_NOWAIT);
167 	if (r == NULL)
168 		return ENOMEM;
169 	r->r_start = start;
170 	r->r_end = end;
171 	r->r_rm = rm;
172 
173 	mtx_lock(rm->rm_mtx);
174 
175 	/* Skip entries before us. */
176 	TAILQ_FOREACH(s, &rm->rm_list, r_link) {
177 		if (s->r_end == ULONG_MAX)
178 			break;
179 		if (s->r_end + 1 >= r->r_start)
180 			break;
181 	}
182 
183 	/* If we ran off the end of the list, insert at the tail. */
184 	if (s == NULL) {
185 		TAILQ_INSERT_TAIL(&rm->rm_list, r, r_link);
186 	} else {
187 		/* Check for any overlap with the current region. */
188 		if (r->r_start <= s->r_end && r->r_end >= s->r_start) {
189 			rv = EBUSY;
190 			goto out;
191 		}
192 
193 		/* Check for any overlap with the next region. */
194 		t = TAILQ_NEXT(s, r_link);
195 		if (t && r->r_start <= t->r_end && r->r_end >= t->r_start) {
196 			rv = EBUSY;
197 			goto out;
198 		}
199 
200 		/*
201 		 * See if this region can be merged with the next region.  If
202 		 * not, clear the pointer.
203 		 */
204 		if (t && (r->r_end + 1 != t->r_start || t->r_flags != 0))
205 			t = NULL;
206 
207 		/* See if we can merge with the current region. */
208 		if (s->r_end + 1 == r->r_start && s->r_flags == 0) {
209 			/* Can we merge all 3 regions? */
210 			if (t != NULL) {
211 				s->r_end = t->r_end;
212 				TAILQ_REMOVE(&rm->rm_list, t, r_link);
213 				free(r, M_RMAN);
214 				free(t, M_RMAN);
215 			} else {
216 				s->r_end = r->r_end;
217 				free(r, M_RMAN);
218 			}
219 		} else if (t != NULL) {
220 			/* Can we merge with just the next region? */
221 			t->r_start = r->r_start;
222 			free(r, M_RMAN);
223 		} else if (s->r_end < r->r_start) {
224 			TAILQ_INSERT_AFTER(&rm->rm_list, s, r, r_link);
225 		} else {
226 			TAILQ_INSERT_BEFORE(s, r, r_link);
227 		}
228 	}
229 out:
230 	mtx_unlock(rm->rm_mtx);
231 	return rv;
232 }
233 
234 int
235 rman_init_from_resource(struct rman *rm, struct resource *r)
236 {
237 	int rv;
238 
239 	if ((rv = rman_init(rm)) != 0)
240 		return (rv);
241 	return (rman_manage_region(rm, r->__r_i->r_start, r->__r_i->r_end));
242 }
243 
244 int
245 rman_fini(struct rman *rm)
246 {
247 	struct resource_i *r;
248 
249 	mtx_lock(rm->rm_mtx);
250 	TAILQ_FOREACH(r, &rm->rm_list, r_link) {
251 		if (r->r_flags & RF_ALLOCATED) {
252 			mtx_unlock(rm->rm_mtx);
253 			return EBUSY;
254 		}
255 	}
256 
257 	/*
258 	 * There really should only be one of these if we are in this
259 	 * state and the code is working properly, but it can't hurt.
260 	 */
261 	while (!TAILQ_EMPTY(&rm->rm_list)) {
262 		r = TAILQ_FIRST(&rm->rm_list);
263 		TAILQ_REMOVE(&rm->rm_list, r, r_link);
264 		free(r, M_RMAN);
265 	}
266 	mtx_unlock(rm->rm_mtx);
267 	mtx_lock(&rman_mtx);
268 	TAILQ_REMOVE(&rman_head, rm, rm_link);
269 	mtx_unlock(&rman_mtx);
270 	mtx_destroy(rm->rm_mtx);
271 	free(rm->rm_mtx, M_RMAN);
272 
273 	return 0;
274 }
275 
276 int
277 rman_first_free_region(struct rman *rm, rman_res_t *start, rman_res_t *end)
278 {
279 	struct resource_i *r;
280 
281 	mtx_lock(rm->rm_mtx);
282 	TAILQ_FOREACH(r, &rm->rm_list, r_link) {
283 		if (!(r->r_flags & RF_ALLOCATED)) {
284 			*start = r->r_start;
285 			*end = r->r_end;
286 			mtx_unlock(rm->rm_mtx);
287 			return (0);
288 		}
289 	}
290 	mtx_unlock(rm->rm_mtx);
291 	return (ENOENT);
292 }
293 
294 int
295 rman_last_free_region(struct rman *rm, rman_res_t *start, rman_res_t *end)
296 {
297 	struct resource_i *r;
298 
299 	mtx_lock(rm->rm_mtx);
300 	TAILQ_FOREACH_REVERSE(r, &rm->rm_list, resource_head, r_link) {
301 		if (!(r->r_flags & RF_ALLOCATED)) {
302 			*start = r->r_start;
303 			*end = r->r_end;
304 			mtx_unlock(rm->rm_mtx);
305 			return (0);
306 		}
307 	}
308 	mtx_unlock(rm->rm_mtx);
309 	return (ENOENT);
310 }
311 
312 /* Shrink or extend one or both ends of an allocated resource. */
313 int
314 rman_adjust_resource(struct resource *rr, rman_res_t start, rman_res_t end)
315 {
316 	struct resource_i *r, *s, *t, *new;
317 	struct rman *rm;
318 
319 	/* Not supported for shared resources. */
320 	r = rr->__r_i;
321 	if (r->r_flags & RF_SHAREABLE)
322 		return (EINVAL);
323 
324 	/*
325 	 * This does not support wholesale moving of a resource.  At
326 	 * least part of the desired new range must overlap with the
327 	 * existing resource.
328 	 */
329 	if (end < r->r_start || r->r_end < start)
330 		return (EINVAL);
331 
332 	/*
333 	 * Find the two resource regions immediately adjacent to the
334 	 * allocated resource.
335 	 */
336 	rm = r->r_rm;
337 	mtx_lock(rm->rm_mtx);
338 #ifdef INVARIANTS
339 	TAILQ_FOREACH(s, &rm->rm_list, r_link) {
340 		if (s == r)
341 			break;
342 	}
343 	if (s == NULL)
344 		panic("resource not in list");
345 #endif
346 	s = TAILQ_PREV(r, resource_head, r_link);
347 	t = TAILQ_NEXT(r, r_link);
348 	KASSERT(s == NULL || s->r_end + 1 == r->r_start,
349 	    ("prev resource mismatch"));
350 	KASSERT(t == NULL || r->r_end + 1 == t->r_start,
351 	    ("next resource mismatch"));
352 
353 	/*
354 	 * See if the changes are permitted.  Shrinking is always allowed,
355 	 * but growing requires sufficient room in the adjacent region.
356 	 */
357 	if (start < r->r_start && (s == NULL || (s->r_flags & RF_ALLOCATED) ||
358 	    s->r_start > start)) {
359 		mtx_unlock(rm->rm_mtx);
360 		return (EBUSY);
361 	}
362 	if (end > r->r_end && (t == NULL || (t->r_flags & RF_ALLOCATED) ||
363 	    t->r_end < end)) {
364 		mtx_unlock(rm->rm_mtx);
365 		return (EBUSY);
366 	}
367 
368 	/*
369 	 * While holding the lock, grow either end of the resource as
370 	 * needed and shrink either end if the shrinking does not require
371 	 * allocating a new resource.  We can safely drop the lock and then
372 	 * insert a new range to handle the shrinking case afterwards.
373 	 */
374 	if (start < r->r_start ||
375 	    (start > r->r_start && s != NULL && !(s->r_flags & RF_ALLOCATED))) {
376 		KASSERT(s->r_flags == 0, ("prev is busy"));
377 		r->r_start = start;
378 		if (s->r_start == start) {
379 			TAILQ_REMOVE(&rm->rm_list, s, r_link);
380 			free(s, M_RMAN);
381 		} else
382 			s->r_end = start - 1;
383 	}
384 	if (end > r->r_end ||
385 	    (end < r->r_end && t != NULL && !(t->r_flags & RF_ALLOCATED))) {
386 		KASSERT(t->r_flags == 0, ("next is busy"));
387 		r->r_end = end;
388 		if (t->r_end == end) {
389 			TAILQ_REMOVE(&rm->rm_list, t, r_link);
390 			free(t, M_RMAN);
391 		} else
392 			t->r_start = end + 1;
393 	}
394 	mtx_unlock(rm->rm_mtx);
395 
396 	/*
397 	 * Handle the shrinking cases that require allocating a new
398 	 * resource to hold the newly-free region.  We have to recheck
399 	 * if we still need this new region after acquiring the lock.
400 	 */
401 	if (start > r->r_start) {
402 		new = int_alloc_resource(M_WAITOK);
403 		new->r_start = r->r_start;
404 		new->r_end = start - 1;
405 		new->r_rm = rm;
406 		mtx_lock(rm->rm_mtx);
407 		r->r_start = start;
408 		s = TAILQ_PREV(r, resource_head, r_link);
409 		if (s != NULL && !(s->r_flags & RF_ALLOCATED)) {
410 			s->r_end = start - 1;
411 			free(new, M_RMAN);
412 		} else
413 			TAILQ_INSERT_BEFORE(r, new, r_link);
414 		mtx_unlock(rm->rm_mtx);
415 	}
416 	if (end < r->r_end) {
417 		new = int_alloc_resource(M_WAITOK);
418 		new->r_start = end + 1;
419 		new->r_end = r->r_end;
420 		new->r_rm = rm;
421 		mtx_lock(rm->rm_mtx);
422 		r->r_end = end;
423 		t = TAILQ_NEXT(r, r_link);
424 		if (t != NULL && !(t->r_flags & RF_ALLOCATED)) {
425 			t->r_start = end + 1;
426 			free(new, M_RMAN);
427 		} else
428 			TAILQ_INSERT_AFTER(&rm->rm_list, r, new, r_link);
429 		mtx_unlock(rm->rm_mtx);
430 	}
431 	return (0);
432 }
433 
434 #define	SHARE_TYPE(f)	(f & (RF_SHAREABLE | RF_PREFETCHABLE))
435 
436 struct resource *
437 rman_reserve_resource_bound(struct rman *rm, rman_res_t start, rman_res_t end,
438 			    rman_res_t count, rman_res_t bound, u_int flags,
439 			    struct device *dev)
440 {
441 	u_int new_rflags;
442 	struct resource_i *r, *s, *rv;
443 	rman_res_t rstart, rend, amask, bmask;
444 
445 	rv = NULL;
446 
447 	DPRINTF(("rman_reserve_resource_bound: <%s> request: [%#lx, %#lx], "
448 	       "length %#lx, flags %u, device %s\n", rm->rm_descr, start, end,
449 	       count, flags,
450 	       dev == NULL ? "<null>" : device_get_nameunit(dev)));
451 	KASSERT((flags & RF_FIRSTSHARE) == 0,
452 	    ("invalid flags %#x", flags));
453 	new_rflags = (flags & ~RF_FIRSTSHARE) | RF_ALLOCATED;
454 
455 	mtx_lock(rm->rm_mtx);
456 
457 	for (r = TAILQ_FIRST(&rm->rm_list);
458 	     r && r->r_end < start + count - 1;
459 	     r = TAILQ_NEXT(r, r_link))
460 		;
461 
462 	if (r == NULL) {
463 		DPRINTF(("could not find a region\n"));
464 		goto out;
465 	}
466 
467 	amask = (1ul << RF_ALIGNMENT(flags)) - 1;
468 	KASSERT(start <= ULONG_MAX - amask,
469 	    ("start (%#lx) + amask (%#lx) would wrap around", start, amask));
470 
471 	/* If bound is 0, bmask will also be 0 */
472 	bmask = ~(bound - 1);
473 	/*
474 	 * First try to find an acceptable totally-unshared region.
475 	 */
476 	for (s = r; s; s = TAILQ_NEXT(s, r_link)) {
477 		DPRINTF(("considering [%#lx, %#lx]\n", s->r_start, s->r_end));
478 		/*
479 		 * The resource list is sorted, so there is no point in
480 		 * searching further once r_start is too large.
481 		 */
482 		if (s->r_start > end - (count - 1)) {
483 			DPRINTF(("s->r_start (%#lx) + count - 1> end (%#lx)\n",
484 			    s->r_start, end));
485 			break;
486 		}
487 		if (s->r_start > ULONG_MAX - amask) {
488 			DPRINTF(("s->r_start (%#lx) + amask (%#lx) too large\n",
489 			    s->r_start, amask));
490 			break;
491 		}
492 		if (s->r_flags & RF_ALLOCATED) {
493 			DPRINTF(("region is allocated\n"));
494 			continue;
495 		}
496 		rstart = ulmax(s->r_start, start);
497 		/*
498 		 * Try to find a region by adjusting to boundary and alignment
499 		 * until both conditions are satisfied. This is not an optimal
500 		 * algorithm, but in most cases it isn't really bad, either.
501 		 */
502 		do {
503 			rstart = (rstart + amask) & ~amask;
504 			if (((rstart ^ (rstart + count - 1)) & bmask) != 0)
505 				rstart += bound - (rstart & ~bmask);
506 		} while ((rstart & amask) != 0 && rstart < end &&
507 		    rstart < s->r_end);
508 		rend = ulmin(s->r_end, ulmax(rstart + count - 1, end));
509 		if (rstart > rend) {
510 			DPRINTF(("adjusted start exceeds end\n"));
511 			continue;
512 		}
513 		DPRINTF(("truncated region: [%#lx, %#lx]; size %#lx (requested %#lx)\n",
514 		       rstart, rend, (rend - rstart + 1), count));
515 
516 		if ((rend - rstart + 1) >= count) {
517 			DPRINTF(("candidate region: [%#lx, %#lx], size %#lx\n",
518 			       rstart, rend, (rend - rstart + 1)));
519 			if ((s->r_end - s->r_start + 1) == count) {
520 				DPRINTF(("candidate region is entire chunk\n"));
521 				rv = s;
522 				rv->r_flags = new_rflags;
523 				rv->r_dev = dev;
524 				goto out;
525 			}
526 
527 			/*
528 			 * If s->r_start < rstart and
529 			 *    s->r_end > rstart + count - 1, then
530 			 * we need to split the region into three pieces
531 			 * (the middle one will get returned to the user).
532 			 * Otherwise, we are allocating at either the
533 			 * beginning or the end of s, so we only need to
534 			 * split it in two.  The first case requires
535 			 * two new allocations; the second requires but one.
536 			 */
537 			rv = int_alloc_resource(M_NOWAIT);
538 			if (rv == NULL)
539 				goto out;
540 			rv->r_start = rstart;
541 			rv->r_end = rstart + count - 1;
542 			rv->r_flags = new_rflags;
543 			rv->r_dev = dev;
544 			rv->r_rm = rm;
545 
546 			if (s->r_start < rv->r_start && s->r_end > rv->r_end) {
547 				DPRINTF(("splitting region in three parts: "
548 				       "[%#lx, %#lx]; [%#lx, %#lx]; [%#lx, %#lx]\n",
549 				       s->r_start, rv->r_start - 1,
550 				       rv->r_start, rv->r_end,
551 				       rv->r_end + 1, s->r_end));
552 				/*
553 				 * We are allocating in the middle.
554 				 */
555 				r = int_alloc_resource(M_NOWAIT);
556 				if (r == NULL) {
557 					free(rv, M_RMAN);
558 					rv = NULL;
559 					goto out;
560 				}
561 				r->r_start = rv->r_end + 1;
562 				r->r_end = s->r_end;
563 				r->r_flags = s->r_flags;
564 				r->r_rm = rm;
565 				s->r_end = rv->r_start - 1;
566 				TAILQ_INSERT_AFTER(&rm->rm_list, s, rv,
567 						     r_link);
568 				TAILQ_INSERT_AFTER(&rm->rm_list, rv, r,
569 						     r_link);
570 			} else if (s->r_start == rv->r_start) {
571 				DPRINTF(("allocating from the beginning\n"));
572 				/*
573 				 * We are allocating at the beginning.
574 				 */
575 				s->r_start = rv->r_end + 1;
576 				TAILQ_INSERT_BEFORE(s, rv, r_link);
577 			} else {
578 				DPRINTF(("allocating at the end\n"));
579 				/*
580 				 * We are allocating at the end.
581 				 */
582 				s->r_end = rv->r_start - 1;
583 				TAILQ_INSERT_AFTER(&rm->rm_list, s, rv,
584 						     r_link);
585 			}
586 			goto out;
587 		}
588 	}
589 
590 	/*
591 	 * Now find an acceptable shared region, if the client's requirements
592 	 * allow sharing.  By our implementation restriction, a candidate
593 	 * region must match exactly by both size and sharing type in order
594 	 * to be considered compatible with the client's request.  (The
595 	 * former restriction could probably be lifted without too much
596 	 * additional work, but this does not seem warranted.)
597 	 */
598 	DPRINTF(("no unshared regions found\n"));
599 	if ((flags & RF_SHAREABLE) == 0)
600 		goto out;
601 
602 	for (s = r; s && s->r_end <= end; s = TAILQ_NEXT(s, r_link)) {
603 		if (SHARE_TYPE(s->r_flags) == SHARE_TYPE(flags) &&
604 		    s->r_start >= start &&
605 		    (s->r_end - s->r_start + 1) == count &&
606 		    (s->r_start & amask) == 0 &&
607 		    ((s->r_start ^ s->r_end) & bmask) == 0) {
608 			rv = int_alloc_resource(M_NOWAIT);
609 			if (rv == NULL)
610 				goto out;
611 			rv->r_start = s->r_start;
612 			rv->r_end = s->r_end;
613 			rv->r_flags = new_rflags;
614 			rv->r_dev = dev;
615 			rv->r_rm = rm;
616 			if (s->r_sharehead == NULL) {
617 				s->r_sharehead = malloc(sizeof *s->r_sharehead,
618 						M_RMAN, M_NOWAIT | M_ZERO);
619 				if (s->r_sharehead == NULL) {
620 					free(rv, M_RMAN);
621 					rv = NULL;
622 					goto out;
623 				}
624 				LIST_INIT(s->r_sharehead);
625 				LIST_INSERT_HEAD(s->r_sharehead, s,
626 						 r_sharelink);
627 				s->r_flags |= RF_FIRSTSHARE;
628 			}
629 			rv->r_sharehead = s->r_sharehead;
630 			LIST_INSERT_HEAD(s->r_sharehead, rv, r_sharelink);
631 			goto out;
632 		}
633 	}
634 	/*
635 	 * We couldn't find anything.
636 	 */
637 
638 out:
639 	mtx_unlock(rm->rm_mtx);
640 	return (rv == NULL ? NULL : &rv->r_r);
641 }
642 
643 struct resource *
644 rman_reserve_resource(struct rman *rm, rman_res_t start, rman_res_t end,
645 		      rman_res_t count, u_int flags, struct device *dev)
646 {
647 
648 	return (rman_reserve_resource_bound(rm, start, end, count, 0, flags,
649 	    dev));
650 }
651 
652 int
653 rman_activate_resource(struct resource *re)
654 {
655 	struct resource_i *r;
656 	struct rman *rm;
657 
658 	r = re->__r_i;
659 	rm = r->r_rm;
660 	mtx_lock(rm->rm_mtx);
661 	r->r_flags |= RF_ACTIVE;
662 	mtx_unlock(rm->rm_mtx);
663 	return 0;
664 }
665 
666 int
667 rman_deactivate_resource(struct resource *r)
668 {
669 	struct rman *rm;
670 
671 	rm = r->__r_i->r_rm;
672 	mtx_lock(rm->rm_mtx);
673 	r->__r_i->r_flags &= ~RF_ACTIVE;
674 	mtx_unlock(rm->rm_mtx);
675 	return 0;
676 }
677 
678 static int
679 int_rman_release_resource(struct rman *rm, struct resource_i *r)
680 {
681 	struct resource_i *s, *t;
682 
683 	if (r->r_flags & RF_ACTIVE)
684 		r->r_flags &= ~RF_ACTIVE;
685 
686 	/*
687 	 * Check for a sharing list first.  If there is one, then we don't
688 	 * have to think as hard.
689 	 */
690 	if (r->r_sharehead) {
691 		/*
692 		 * If a sharing list exists, then we know there are at
693 		 * least two sharers.
694 		 *
695 		 * If we are in the main circleq, appoint someone else.
696 		 */
697 		LIST_REMOVE(r, r_sharelink);
698 		s = LIST_FIRST(r->r_sharehead);
699 		if (r->r_flags & RF_FIRSTSHARE) {
700 			s->r_flags |= RF_FIRSTSHARE;
701 			TAILQ_INSERT_BEFORE(r, s, r_link);
702 			TAILQ_REMOVE(&rm->rm_list, r, r_link);
703 		}
704 
705 		/*
706 		 * Make sure that the sharing list goes away completely
707 		 * if the resource is no longer being shared at all.
708 		 */
709 		if (LIST_NEXT(s, r_sharelink) == NULL) {
710 			free(s->r_sharehead, M_RMAN);
711 			s->r_sharehead = NULL;
712 			s->r_flags &= ~RF_FIRSTSHARE;
713 		}
714 		goto out;
715 	}
716 
717 	/*
718 	 * Look at the adjacent resources in the list and see if our
719 	 * segment can be merged with any of them.  If either of the
720 	 * resources is allocated or is not exactly adjacent then they
721 	 * cannot be merged with our segment.
722 	 */
723 	s = TAILQ_PREV(r, resource_head, r_link);
724 	if (s != NULL && ((s->r_flags & RF_ALLOCATED) != 0 ||
725 	    s->r_end + 1 != r->r_start))
726 		s = NULL;
727 	t = TAILQ_NEXT(r, r_link);
728 	if (t != NULL && ((t->r_flags & RF_ALLOCATED) != 0 ||
729 	    r->r_end + 1 != t->r_start))
730 		t = NULL;
731 
732 	if (s != NULL && t != NULL) {
733 		/*
734 		 * Merge all three segments.
735 		 */
736 		s->r_end = t->r_end;
737 		TAILQ_REMOVE(&rm->rm_list, r, r_link);
738 		TAILQ_REMOVE(&rm->rm_list, t, r_link);
739 		free(t, M_RMAN);
740 	} else if (s != NULL) {
741 		/*
742 		 * Merge previous segment with ours.
743 		 */
744 		s->r_end = r->r_end;
745 		TAILQ_REMOVE(&rm->rm_list, r, r_link);
746 	} else if (t != NULL) {
747 		/*
748 		 * Merge next segment with ours.
749 		 */
750 		t->r_start = r->r_start;
751 		TAILQ_REMOVE(&rm->rm_list, r, r_link);
752 	} else {
753 		/*
754 		 * At this point, we know there is nothing we
755 		 * can potentially merge with, because on each
756 		 * side, there is either nothing there or what is
757 		 * there is still allocated.  In that case, we don't
758 		 * want to remove r from the list; we simply want to
759 		 * change it to an unallocated region and return
760 		 * without freeing anything.
761 		 */
762 		r->r_flags &= ~RF_ALLOCATED;
763 		r->r_dev = NULL;
764 		return 0;
765 	}
766 
767 out:
768 	free(r, M_RMAN);
769 	return 0;
770 }
771 
772 int
773 rman_release_resource(struct resource *re)
774 {
775 	int rv;
776 	struct resource_i *r;
777 	struct rman *rm;
778 
779 	r = re->__r_i;
780 	rm = r->r_rm;
781 	mtx_lock(rm->rm_mtx);
782 	rv = int_rman_release_resource(rm, r);
783 	mtx_unlock(rm->rm_mtx);
784 	return (rv);
785 }
786 
787 uint32_t
788 rman_make_alignment_flags(uint32_t size)
789 {
790 	int i;
791 
792 	/*
793 	 * Find the hightest bit set, and add one if more than one bit
794 	 * set.  We're effectively computing the ceil(log2(size)) here.
795 	 */
796 	for (i = 31; i > 0; i--)
797 		if ((1 << i) & size)
798 			break;
799 	if (~(1 << i) & size)
800 		i++;
801 
802 	return(RF_ALIGNMENT_LOG2(i));
803 }
804 
805 void
806 rman_set_start(struct resource *r, rman_res_t start)
807 {
808 
809 	r->__r_i->r_start = start;
810 }
811 
812 rman_res_t
813 rman_get_start(struct resource *r)
814 {
815 
816 	return (r->__r_i->r_start);
817 }
818 
819 void
820 rman_set_end(struct resource *r, rman_res_t end)
821 {
822 
823 	r->__r_i->r_end = end;
824 }
825 
826 rman_res_t
827 rman_get_end(struct resource *r)
828 {
829 
830 	return (r->__r_i->r_end);
831 }
832 
833 rman_res_t
834 rman_get_size(struct resource *r)
835 {
836 
837 	return (r->__r_i->r_end - r->__r_i->r_start + 1);
838 }
839 
840 u_int
841 rman_get_flags(struct resource *r)
842 {
843 
844 	return (r->__r_i->r_flags);
845 }
846 
847 void
848 rman_set_virtual(struct resource *r, void *v)
849 {
850 
851 	r->__r_i->r_virtual = v;
852 }
853 
854 void *
855 rman_get_virtual(struct resource *r)
856 {
857 
858 	return (r->__r_i->r_virtual);
859 }
860 
861 void
862 rman_set_bustag(struct resource *r, bus_space_tag_t t)
863 {
864 
865 	r->r_bustag = t;
866 }
867 
868 bus_space_tag_t
869 rman_get_bustag(struct resource *r)
870 {
871 
872 	return (r->r_bustag);
873 }
874 
875 void
876 rman_set_bushandle(struct resource *r, bus_space_handle_t h)
877 {
878 
879 	r->r_bushandle = h;
880 }
881 
882 bus_space_handle_t
883 rman_get_bushandle(struct resource *r)
884 {
885 
886 	return (r->r_bushandle);
887 }
888 
889 void
890 rman_set_rid(struct resource *r, int rid)
891 {
892 
893 	r->__r_i->r_rid = rid;
894 }
895 
896 int
897 rman_get_rid(struct resource *r)
898 {
899 
900 	return (r->__r_i->r_rid);
901 }
902 
903 void
904 rman_set_device(struct resource *r, struct device *dev)
905 {
906 
907 	r->__r_i->r_dev = dev;
908 }
909 
910 struct device *
911 rman_get_device(struct resource *r)
912 {
913 
914 	return (r->__r_i->r_dev);
915 }
916 
917 int
918 rman_is_region_manager(struct resource *r, struct rman *rm)
919 {
920 
921 	return (r->__r_i->r_rm == rm);
922 }
923 
924 /*
925  * Sysctl interface for scanning the resource lists.
926  *
927  * We take two input parameters; the index into the list of resource
928  * managers, and the resource offset into the list.
929  */
930 static int
931 sysctl_rman(SYSCTL_HANDLER_ARGS)
932 {
933 	int			*name = (int *)arg1;
934 	u_int			namelen = arg2;
935 	int			rman_idx, res_idx;
936 	struct rman		*rm;
937 	struct resource_i	*res;
938 	struct resource_i	*sres;
939 	struct u_rman		urm;
940 	struct u_resource	ures;
941 	int			error;
942 
943 	if (namelen != 3)
944 		return (EINVAL);
945 
946 	if (bus_data_generation_check(name[0]))
947 		return (EINVAL);
948 	rman_idx = name[1];
949 	res_idx = name[2];
950 
951 	/*
952 	 * Find the indexed resource manager
953 	 */
954 	mtx_lock(&rman_mtx);
955 	TAILQ_FOREACH(rm, &rman_head, rm_link) {
956 		if (rman_idx-- == 0)
957 			break;
958 	}
959 	mtx_unlock(&rman_mtx);
960 	if (rm == NULL)
961 		return (ENOENT);
962 
963 	/*
964 	 * If the resource index is -1, we want details on the
965 	 * resource manager.
966 	 */
967 	if (res_idx == -1) {
968 		bzero(&urm, sizeof(urm));
969 		urm.rm_handle = (uintptr_t)rm;
970 		if (rm->rm_descr != NULL)
971 			strlcpy(urm.rm_descr, rm->rm_descr, RM_TEXTLEN);
972 		urm.rm_start = rm->rm_start;
973 		urm.rm_size = rm->rm_end - rm->rm_start + 1;
974 		urm.rm_type = rm->rm_type;
975 
976 		error = SYSCTL_OUT(req, &urm, sizeof(urm));
977 		return (error);
978 	}
979 
980 	/*
981 	 * Find the indexed resource and return it.
982 	 */
983 	mtx_lock(rm->rm_mtx);
984 	TAILQ_FOREACH(res, &rm->rm_list, r_link) {
985 		if (res->r_sharehead != NULL) {
986 			LIST_FOREACH(sres, res->r_sharehead, r_sharelink)
987 				if (res_idx-- == 0) {
988 					res = sres;
989 					goto found;
990 				}
991 		}
992 		else if (res_idx-- == 0)
993 				goto found;
994 	}
995 	mtx_unlock(rm->rm_mtx);
996 	return (ENOENT);
997 
998 found:
999 	bzero(&ures, sizeof(ures));
1000 	ures.r_handle = (uintptr_t)res;
1001 	ures.r_parent = (uintptr_t)res->r_rm;
1002 	ures.r_device = (uintptr_t)res->r_dev;
1003 	if (res->r_dev != NULL) {
1004 		if (device_get_name(res->r_dev) != NULL) {
1005 			snprintf(ures.r_devname, RM_TEXTLEN,
1006 			    "%s%d",
1007 			    device_get_name(res->r_dev),
1008 			    device_get_unit(res->r_dev));
1009 		} else {
1010 			strlcpy(ures.r_devname, "nomatch",
1011 			    RM_TEXTLEN);
1012 		}
1013 	} else {
1014 		ures.r_devname[0] = '\0';
1015 	}
1016 	ures.r_start = res->r_start;
1017 	ures.r_size = res->r_end - res->r_start + 1;
1018 	ures.r_flags = res->r_flags;
1019 
1020 	mtx_unlock(rm->rm_mtx);
1021 	error = SYSCTL_OUT(req, &ures, sizeof(ures));
1022 	return (error);
1023 }
1024 
1025 static SYSCTL_NODE(_hw_bus, OID_AUTO, rman, CTLFLAG_RD, sysctl_rman,
1026     "kernel resource manager");
1027 
1028 #ifdef DDB
1029 static void
1030 dump_rman_header(struct rman *rm)
1031 {
1032 
1033 	if (db_pager_quit)
1034 		return;
1035 	db_printf("rman %p: %s (0x%lx-0x%lx full range)\n",
1036 	    rm, rm->rm_descr, rm->rm_start, rm->rm_end);
1037 }
1038 
1039 static void
1040 dump_rman(struct rman *rm)
1041 {
1042 	struct resource_i *r;
1043 	const char *devname;
1044 
1045 	if (db_pager_quit)
1046 		return;
1047 	TAILQ_FOREACH(r, &rm->rm_list, r_link) {
1048 		if (r->r_dev != NULL) {
1049 			devname = device_get_nameunit(r->r_dev);
1050 			if (devname == NULL)
1051 				devname = "nomatch";
1052 		} else
1053 			devname = NULL;
1054 		db_printf("    0x%lx-0x%lx (RID=%d) ",
1055 		    r->r_start, r->r_end, r->r_rid);
1056 		if (devname != NULL)
1057 			db_printf("(%s)\n", devname);
1058 		else
1059 			db_printf("----\n");
1060 		if (db_pager_quit)
1061 			return;
1062 	}
1063 }
1064 
1065 DB_SHOW_COMMAND(rman, db_show_rman)
1066 {
1067 
1068 	if (have_addr) {
1069 		dump_rman_header((struct rman *)addr);
1070 		dump_rman((struct rman *)addr);
1071 	}
1072 }
1073 
1074 DB_SHOW_COMMAND(rmans, db_show_rmans)
1075 {
1076 	struct rman *rm;
1077 
1078 	TAILQ_FOREACH(rm, &rman_head, rm_link) {
1079 		dump_rman_header(rm);
1080 	}
1081 }
1082 
1083 DB_SHOW_ALL_COMMAND(rman, db_show_all_rman)
1084 {
1085 	struct rman *rm;
1086 
1087 	TAILQ_FOREACH(rm, &rman_head, rm_link) {
1088 		dump_rman_header(rm);
1089 		dump_rman(rm);
1090 	}
1091 }
1092 DB_SHOW_ALIAS(allrman, db_show_all_rman);
1093 #endif
1094