2 * Copyright 1998 Massachusetts Institute of Technology
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
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
29 * $FreeBSD: src/sys/kern/subr_rman.c,v 1.10.2.1 2001/06/05 08:06:08 imp Exp $
33 * The kernel resource manager. This code is responsible for keeping track
34 * of hardware resources which are apportioned out to various drivers.
35 * It does not actually assign those resources, and it is not expected
36 * that end-device drivers will call into this code directly. Rather,
37 * the code which implements the buses that those devices are attached to,
38 * and the code which manages CPU resources, will call this code, and the
39 * end-device drivers will make upcalls to that code to actually perform
42 * There are two sorts of resources managed by this code. The first is
43 * the more familiar array (RMAN_ARRAY) type; resources in this class
44 * consist of a sequence of individually-allocatable objects which have
45 * been numbered in some well-defined order. Most of the resources
46 * are of this type, as it is the most familiar. The second type is
47 * called a gauge (RMAN_GAUGE), and models fungible resources (i.e.,
48 * resources in which each instance is indistinguishable from every
49 * other instance). The principal anticipated application of gauges
50 * is in the context of power consumption, where a bus may have a specific
51 * power budget which all attached devices share. RMAN_GAUGE is not
54 * For array resources, we make one simplifying assumption: two clients
55 * sharing the same resource must use the same range of indices. That
56 * is to say, sharing of overlapping-but-not-identical regions is not
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
64 #include <sys/malloc.h>
65 #include <sys/bus.h> /* XXX debugging */
67 #include <sys/sysctl.h>
69 static int rman_debug
= 0;
70 TUNABLE_INT("debug.rman_debug", &rman_debug
);
71 SYSCTL_INT(_debug
, OID_AUTO
, rman_debug
, CTLFLAG_RW
,
72 &rman_debug
, 0, "rman debug");
74 #define DPRINTF(params) if (rman_debug) kprintf params
76 static MALLOC_DEFINE(M_RMAN
, "rman", "Resource manager");
78 TAILQ_HEAD(rman_head
, rman
);
79 static struct rman_head rman_head
;
80 static struct lwkt_token rman_tok
; /* mutex to protect rman_head */
81 static int int_rman_activate_resource(struct rman
*rm
, struct resource
*r
,
82 struct resource
**whohas
);
83 static int int_rman_deactivate_resource(struct resource
*r
);
84 static int int_rman_release_resource(struct rman
*rm
, struct resource
*r
);
87 rman_init(struct rman
*rm
, int cpuid
)
93 TAILQ_INIT(&rman_head
);
94 lwkt_token_init(&rman_tok
, "rman");
97 if (rm
->rm_type
== RMAN_UNINIT
)
99 if (rm
->rm_type
== RMAN_GAUGE
)
100 panic("implement RMAN_GAUGE");
102 TAILQ_INIT(&rm
->rm_list
);
103 rm
->rm_slock
= kmalloc(sizeof *rm
->rm_slock
, M_RMAN
, M_NOWAIT
);
104 if (rm
->rm_slock
== NULL
)
106 lwkt_token_init(rm
->rm_slock
, "rmanslock");
108 rm
->rm_cpuid
= cpuid
;
111 lwkt_gettoken(&rman_tok
);
112 TAILQ_INSERT_TAIL(&rman_head
, rm
, rm_link
);
113 lwkt_reltoken(&rman_tok
);
119 * NB: this interface is not robust against programming errors which
120 * add multiple copies of the same region.
123 rman_manage_region(struct rman
*rm
, u_long start
, u_long end
)
125 struct resource
*r
, *s
;
127 DPRINTF(("rman_manage_region: <%s> request: start %#lx, end %#lx\n",
128 rm
->rm_descr
, start
, end
));
129 r
= kmalloc(sizeof *r
, M_RMAN
, M_NOWAIT
| M_ZERO
);
139 lwkt_gettoken(rm
->rm_slock
);
140 for (s
= TAILQ_FIRST(&rm
->rm_list
);
141 s
&& s
->r_end
< r
->r_start
;
142 s
= TAILQ_NEXT(s
, r_link
))
146 TAILQ_INSERT_TAIL(&rm
->rm_list
, r
, r_link
);
148 TAILQ_INSERT_BEFORE(s
, r
, r_link
);
150 lwkt_reltoken(rm
->rm_slock
);
155 rman_fini(struct rman
*rm
)
160 * All resources must already have been deallocated.
162 lwkt_gettoken(rm
->rm_slock
);
163 TAILQ_FOREACH(r
, &rm
->rm_list
, r_link
) {
164 if (r
->r_flags
& RF_ALLOCATED
) {
165 lwkt_reltoken(rm
->rm_slock
);
171 * Protected list removal. Once removed, wait for any temporary
172 * holds to be dropped before actually destroying the resource.
174 lwkt_gettoken(&rman_tok
);
175 TAILQ_REMOVE(&rman_head
, rm
, rm_link
);
176 lwkt_reltoken(&rman_tok
);
179 kprintf("debug: rman_fini(): rm_hold race fixed on %s\n",
182 tsleep(rm
, 0, "rmfree", 2);
186 * Destroy all elements remaining on rm_list
188 while ((r
= TAILQ_FIRST(&rm
->rm_list
)) != NULL
) {
189 TAILQ_REMOVE(&rm
->rm_list
, r
, r_link
);
192 lwkt_reltoken(rm
->rm_slock
);
197 lwkt_token_uninit(rm
->rm_slock
);
198 kfree(rm
->rm_slock
, M_RMAN
);
205 rman_reserve_resource(struct rman
*rm
, u_long start
, u_long end
, u_long count
,
206 u_int flags
, device_t dev
)
209 struct resource
*r
, *s
, *rv
;
214 DPRINTF(("rman_reserve_resource: <%s> request: [%#lx, %#lx], length "
215 "%#lx, flags %u, device %s\n", rm
->rm_descr
, start
, end
,
217 dev
== NULL
? "<null>" : device_get_nameunit(dev
)));
218 want_activate
= (flags
& RF_ACTIVE
);
221 lwkt_gettoken(rm
->rm_slock
);
223 for (r
= TAILQ_FIRST(&rm
->rm_list
);
224 r
&& r
->r_end
< start
+ count
- 1;
225 r
= TAILQ_NEXT(r
, r_link
))
229 DPRINTF(("could not find a region\n"));
234 * First try to find an acceptable totally-unshared region.
236 for (s
= r
; s
; s
= TAILQ_NEXT(s
, r_link
)) {
237 DPRINTF(("considering [%#lx, %#lx]\n", s
->r_start
, s
->r_end
));
238 if (s
->r_start
> end
- (count
- 1)) {
239 DPRINTF(("s->r_start (%#lx) > end (%#lx)\n",
243 if (s
->r_flags
& RF_ALLOCATED
) {
244 DPRINTF(("region is allocated\n"));
247 rstart
= ulmax(s
->r_start
, start
);
248 rstart
= rounddown2(rstart
+ (1ul << RF_ALIGNMENT(flags
)) - 1,
249 1ul << RF_ALIGNMENT(flags
));
250 rend
= ulmin(s
->r_end
, ulmax(start
+ count
- 1, end
));
251 DPRINTF(("truncated region: [%#lx, %#lx]; size %#lx (requested %#lx)\n",
252 rstart
, rend
, (rend
- rstart
+ 1), count
));
254 if ((rend
- rstart
+ 1) >= count
) {
255 DPRINTF(("candidate region: [%#lx, %#lx], size %#lx\n",
256 rstart
, rend
, (rend
- rstart
+ 1)));
257 if ((s
->r_end
- s
->r_start
+ 1) == count
) {
258 DPRINTF(("candidate region is entire chunk\n"));
260 rv
->r_flags
|= RF_ALLOCATED
| flags
;
266 * If s->r_start < rstart and
267 * s->r_end > rstart + count - 1, then
268 * we need to split the region into three pieces
269 * (the middle one will get returned to the user).
270 * Otherwise, we are allocating at either the
271 * beginning or the end of s, so we only need to
272 * split it in two. The first case requires
273 * two new allocations; the second requires but one.
275 rv
= kmalloc(sizeof *rv
, M_RMAN
, M_NOWAIT
| M_ZERO
);
278 rv
->r_start
= rstart
;
279 rv
->r_end
= rstart
+ count
- 1;
280 rv
->r_flags
= flags
| RF_ALLOCATED
;
285 if (s
->r_start
< rv
->r_start
&& s
->r_end
> rv
->r_end
) {
286 DPRINTF(("splitting region in three parts: "
287 "[%#lx, %#lx]; [%#lx, %#lx]; [%#lx, %#lx]\n",
288 s
->r_start
, rv
->r_start
- 1,
289 rv
->r_start
, rv
->r_end
,
290 rv
->r_end
+ 1, s
->r_end
));
292 * We are allocating in the middle.
294 r
= kmalloc(sizeof *r
, M_RMAN
,
301 r
->r_start
= rv
->r_end
+ 1;
303 r
->r_flags
= s
->r_flags
;
307 s
->r_end
= rv
->r_start
- 1;
308 TAILQ_INSERT_AFTER(&rm
->rm_list
, s
, rv
,
310 TAILQ_INSERT_AFTER(&rm
->rm_list
, rv
, r
,
312 } else if (s
->r_start
== rv
->r_start
) {
313 DPRINTF(("allocating from the beginning\n"));
315 * We are allocating at the beginning.
317 s
->r_start
= rv
->r_end
+ 1;
318 TAILQ_INSERT_BEFORE(s
, rv
, r_link
);
320 DPRINTF(("allocating at the end\n"));
322 * We are allocating at the end.
324 s
->r_end
= rv
->r_start
- 1;
325 TAILQ_INSERT_AFTER(&rm
->rm_list
, s
, rv
,
333 * Now find an acceptable shared region, if the client's requirements
334 * allow sharing. By our implementation restriction, a candidate
335 * region must match exactly by both size and sharing type in order
336 * to be considered compatible with the client's request. (The
337 * former restriction could probably be lifted without too much
338 * additional work, but this does not seem warranted.)
340 DPRINTF(("no unshared regions found\n"));
341 if ((flags
& (RF_SHAREABLE
| RF_TIMESHARE
)) == 0)
344 for (s
= r
; s
; s
= TAILQ_NEXT(s
, r_link
)) {
345 if (s
->r_start
> end
)
347 if ((s
->r_flags
& flags
) != flags
)
349 rstart
= ulmax(s
->r_start
, start
);
350 rend
= ulmin(s
->r_end
, ulmax(start
+ count
, end
));
351 if (s
->r_start
>= start
&& s
->r_end
<= end
352 && (s
->r_end
- s
->r_start
+ 1) == count
) {
353 rv
= kmalloc(sizeof *rv
, M_RMAN
, M_NOWAIT
| M_ZERO
);
356 rv
->r_start
= s
->r_start
;
357 rv
->r_end
= s
->r_end
;
358 rv
->r_flags
= s
->r_flags
&
359 (RF_ALLOCATED
| RF_SHAREABLE
| RF_TIMESHARE
);
362 if (s
->r_sharehead
== 0) {
363 s
->r_sharehead
= kmalloc(sizeof *s
->r_sharehead
,
366 if (s
->r_sharehead
== 0) {
371 LIST_INIT(s
->r_sharehead
);
372 LIST_INSERT_HEAD(s
->r_sharehead
, s
,
374 s
->r_flags
|= RF_FIRSTSHARE
;
376 rv
->r_sharehead
= s
->r_sharehead
;
377 LIST_INSERT_HEAD(s
->r_sharehead
, rv
, r_sharelink
);
383 * We couldn't find anything.
385 DPRINTF(("no region found\n"));
388 * If the user specified RF_ACTIVE in the initial flags,
389 * which is reflected in `want_activate', we attempt to atomically
390 * activate the resource. If this fails, we release the resource
391 * and indicate overall failure. (This behavior probably doesn't
392 * make sense for RF_TIMESHARE-type resources.)
394 if (rv
&& want_activate
) {
395 struct resource
*whohas
;
396 DPRINTF(("activating region\n"));
397 if (int_rman_activate_resource(rm
, rv
, &whohas
)) {
398 int_rman_release_resource(rm
, rv
);
402 lwkt_reltoken(rm
->rm_slock
);
407 int_rman_activate_resource(struct rman
*rm
, struct resource
*r
,
408 struct resource
**whohas
)
414 * If we are not timesharing, then there is nothing much to do.
415 * If we already have the resource, then there is nothing at all to do.
416 * If we are not on a sharing list with anybody else, then there is
419 if ((r
->r_flags
& RF_TIMESHARE
) == 0
420 || (r
->r_flags
& RF_ACTIVE
) != 0
421 || r
->r_sharehead
== 0) {
422 r
->r_flags
|= RF_ACTIVE
;
427 for (s
= LIST_FIRST(r
->r_sharehead
); s
&& ok
;
428 s
= LIST_NEXT(s
, r_sharelink
)) {
429 if ((s
->r_flags
& RF_ACTIVE
) != 0) {
435 r
->r_flags
|= RF_ACTIVE
;
442 rman_activate_resource(struct resource
*r
)
445 struct resource
*whohas
;
449 lwkt_gettoken(rm
->rm_slock
);
450 rv
= int_rman_activate_resource(rm
, r
, &whohas
);
451 lwkt_reltoken(rm
->rm_slock
);
459 rman_await_resource(struct resource
*r
, int slpflags
, int timo
)
462 struct resource
*whohas
;
467 lwkt_gettoken(rm
->rm_slock
);
468 rv
= int_rman_activate_resource(rm
, r
, &whohas
);
470 return (rv
); /* returns with ilock held */
472 if (r
->r_sharehead
== 0)
473 panic("rman_await_resource");
475 * A critical section will hopefully will prevent a race
476 * between lwkt_reltoken and tsleep where a process
477 * could conceivably get in and release the resource
478 * before we have a chance to sleep on it. YYY
481 whohas
->r_flags
|= RF_WANTED
;
482 rv
= tsleep(r
->r_sharehead
, slpflags
, "rmwait", timo
);
484 lwkt_reltoken(rm
->rm_slock
);
495 int_rman_deactivate_resource(struct resource
*r
)
497 r
->r_flags
&= ~RF_ACTIVE
;
498 if (r
->r_flags
& RF_WANTED
) {
499 r
->r_flags
&= ~RF_WANTED
;
500 wakeup(r
->r_sharehead
);
506 rman_deactivate_resource(struct resource
*r
)
511 lwkt_gettoken(rm
->rm_slock
);
512 int_rman_deactivate_resource(r
);
513 lwkt_reltoken(rm
->rm_slock
);
518 int_rman_release_resource(struct rman
*rm
, struct resource
*r
)
520 struct resource
*s
, *t
;
522 if (r
->r_flags
& RF_ACTIVE
)
523 int_rman_deactivate_resource(r
);
526 * Check for a sharing list first. If there is one, then we don't
527 * have to think as hard.
529 if (r
->r_sharehead
) {
531 * If a sharing list exists, then we know there are at
534 * If we are in the main circleq, appoint someone else.
536 LIST_REMOVE(r
, r_sharelink
);
537 s
= LIST_FIRST(r
->r_sharehead
);
538 if (r
->r_flags
& RF_FIRSTSHARE
) {
539 s
->r_flags
|= RF_FIRSTSHARE
;
540 TAILQ_INSERT_BEFORE(r
, s
, r_link
);
541 TAILQ_REMOVE(&rm
->rm_list
, r
, r_link
);
545 * Make sure that the sharing list goes away completely
546 * if the resource is no longer being shared at all.
548 if (LIST_NEXT(s
, r_sharelink
) == 0) {
549 kfree(s
->r_sharehead
, M_RMAN
);
551 s
->r_flags
&= ~RF_FIRSTSHARE
;
557 * Look at the adjacent resources in the list and see if our
558 * segment can be merged with any of them.
560 s
= TAILQ_PREV(r
, resource_head
, r_link
);
561 t
= TAILQ_NEXT(r
, r_link
);
563 if (s
!= NULL
&& (s
->r_flags
& RF_ALLOCATED
) == 0
564 && t
!= NULL
&& (t
->r_flags
& RF_ALLOCATED
) == 0) {
566 * Merge all three segments.
569 TAILQ_REMOVE(&rm
->rm_list
, r
, r_link
);
570 TAILQ_REMOVE(&rm
->rm_list
, t
, r_link
);
572 } else if (s
!= NULL
&& (s
->r_flags
& RF_ALLOCATED
) == 0) {
574 * Merge previous segment with ours.
577 TAILQ_REMOVE(&rm
->rm_list
, r
, r_link
);
578 } else if (t
!= NULL
&& (t
->r_flags
& RF_ALLOCATED
) == 0) {
580 * Merge next segment with ours.
582 t
->r_start
= r
->r_start
;
583 TAILQ_REMOVE(&rm
->rm_list
, r
, r_link
);
586 * At this point, we know there is nothing we
587 * can potentially merge with, because on each
588 * side, there is either nothing there or what is
589 * there is still allocated. In that case, we don't
590 * want to remove r from the list; we simply want to
591 * change it to an unallocated region and return
592 * without freeing anything.
594 r
->r_flags
&= ~RF_ALLOCATED
;
604 rman_release_resource(struct resource
*r
)
606 struct rman
*rm
= r
->r_rm
;
609 lwkt_gettoken(rm
->rm_slock
);
610 rv
= int_rman_release_resource(rm
, r
);
611 lwkt_reltoken(rm
->rm_slock
);
616 * Find the hightest bit set, and add one if more than one bit
617 * set. We're effectively computing the ceil(log2(size)) here.
619 * This function cannot compute alignments above (1LU<<63)+1
620 * as this would require returning '64' which will not fit in
621 * the flags field and doesn't work well for calculations either.
624 rman_make_alignment_flags(size_t size
)
628 for (i
= 63; i
; --i
) {
629 if ((1LU << i
) & size
)
632 if (~(1LU << i
) & size
)
636 return(RF_ALIGNMENT_LOG2(i
));
640 * Sysctl interface for scanning the resource lists.
642 * We take two input parameters; the index into the list of resource
643 * managers, and the resource offset into the list.
646 sysctl_rman(SYSCTL_HANDLER_ARGS
)
648 int *name
= (int *)arg1
;
649 u_int namelen
= arg2
;
650 int rman_idx
, res_idx
;
652 struct resource
*res
;
654 struct u_resource ures
;
660 if (bus_data_generation_check(name
[0]))
666 * Find the indexed resource manager
669 lwkt_gettoken(&rman_tok
);
671 TAILQ_FOREACH(rm
, &rman_head
, rm_link
) {
679 * If the resource index is -1, we want details on the
683 urm
.rm_handle
= (uintptr_t)rm
;
684 strlcpy(urm
.rm_descr
, rm
->rm_descr
, RM_TEXTLEN
);
685 urm
.rm_start
= rm
->rm_start
;
686 urm
.rm_size
= rm
->rm_end
- rm
->rm_start
+ 1;
687 urm
.rm_type
= rm
->rm_type
;
689 error
= SYSCTL_OUT(req
, &urm
, sizeof(urm
));
694 * Find the indexed resource and return it.
696 atomic_add_int(&rm
->rm_hold
, 1); /* temp prevent destruction */
697 lwkt_gettoken(rm
->rm_slock
);
699 TAILQ_FOREACH(res
, &rm
->rm_list
, r_link
) {
700 if (res_idx
-- == 0) {
701 ures
.r_handle
= (uintptr_t)res
;
702 ures
.r_parent
= (uintptr_t)res
->r_rm
;
703 ures
.r_device
= (uintptr_t)res
->r_dev
;
704 if (res
->r_dev
!= NULL
) {
705 if (device_get_name(res
->r_dev
) != NULL
) {
706 ksnprintf(ures
.r_devname
, RM_TEXTLEN
,
708 device_get_name(res
->r_dev
),
709 device_get_unit(res
->r_dev
));
711 strlcpy(ures
.r_devname
, "nomatch",
715 ures
.r_devname
[0] = '\0';
717 ures
.r_start
= res
->r_start
;
718 ures
.r_size
= res
->r_end
- res
->r_start
+ 1;
719 ures
.r_flags
= res
->r_flags
;
721 error
= SYSCTL_OUT(req
, &ures
, sizeof(ures
));
725 lwkt_reltoken(rm
->rm_slock
);
726 atomic_add_int(&rm
->rm_hold
, -1);
728 lwkt_reltoken(&rman_tok
);
733 SYSCTL_NODE(_hw_bus
, OID_AUTO
, rman
, CTLFLAG_RD
, sysctl_rman
,
734 "kernel resource manager");