Fix module loading for vkernel.
[dragonfly.git] / sys / kern / subr_rman.c
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  * $FreeBSD: src/sys/kern/subr_rman.c,v 1.10.2.1 2001/06/05 08:06:08 imp Exp $
30  * $DragonFly: src/sys/kern/subr_rman.c,v 1.12 2007/05/20 07:43:24 y0netan1 Exp $
31  */
32
33 /*
34  * The kernel resource manager.  This code is responsible for keeping track
35  * of hardware resources which are apportioned out to various drivers.
36  * It does not actually assign those resources, and it is not expected
37  * that end-device drivers will call into this code directly.  Rather,
38  * the code which implements the buses that those devices are attached to,
39  * and the code which manages CPU resources, will call this code, and the
40  * end-device drivers will make upcalls to that code to actually perform
41  * the allocation.
42  *
43  * There are two sorts of resources managed by this code.  The first is
44  * the more familiar array (RMAN_ARRAY) type; resources in this class
45  * consist of a sequence of individually-allocatable objects which have
46  * been numbered in some well-defined order.  Most of the resources
47  * are of this type, as it is the most familiar.  The second type is
48  * called a gauge (RMAN_GAUGE), and models fungible resources (i.e.,
49  * resources in which each instance is indistinguishable from every
50  * other instance).  The principal anticipated application of gauges
51  * is in the context of power consumption, where a bus may have a specific
52  * power budget which all attached devices share.  RMAN_GAUGE is not
53  * implemented yet.
54  *
55  * For array resources, we make one simplifying assumption: two clients
56  * sharing the same resource must use the same range of indices.  That
57  * is to say, sharing of overlapping-but-not-identical regions is not
58  * permitted.
59  */
60
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/kernel.h>
64 #include <sys/lock.h>
65 #include <sys/malloc.h>
66 #include <sys/bus.h>            /* XXX debugging */
67 #include <sys/rman.h>
68
69 #ifdef RMAN_DEBUG
70 #define DPRINTF(params) kprintf params
71 #else
72 #define DPRINTF(params)
73 #endif
74
75 static MALLOC_DEFINE(M_RMAN, "rman", "Resource manager");
76
77 struct  rman_head rman_head;
78 static  struct lwkt_token rman_tok; /* mutex to protect rman_head */
79 static  int int_rman_activate_resource(struct rman *rm, struct resource *r,
80                                        struct resource **whohas);
81 static  int int_rman_deactivate_resource(struct resource *r);
82 static  int int_rman_release_resource(struct rman *rm, struct resource *r);
83
84 #define CIRCLEQ_TERMCOND(var, head)     (var == (void *)&(head))
85
86 int
87 rman_init(struct rman *rm)
88 {
89         static int once;
90         lwkt_tokref ilock;
91
92         if (once == 0) {
93                 once = 1;
94                 TAILQ_INIT(&rman_head);
95                 lwkt_token_init(&rman_tok);
96         }
97
98         if (rm->rm_type == RMAN_UNINIT)
99                 panic("rman_init");
100         if (rm->rm_type == RMAN_GAUGE)
101                 panic("implement RMAN_GAUGE");
102
103         CIRCLEQ_INIT(&rm->rm_list);
104         rm->rm_slock = kmalloc(sizeof *rm->rm_slock, M_RMAN, M_NOWAIT);
105         if (rm->rm_slock == NULL)
106                 return ENOMEM;
107         lwkt_token_init(rm->rm_slock);
108
109         lwkt_gettoken(&ilock, &rman_tok);
110         TAILQ_INSERT_TAIL(&rman_head, rm, rm_link);
111         lwkt_reltoken(&ilock);
112         return 0;
113 }
114
115 /*
116  * NB: this interface is not robust against programming errors which
117  * add multiple copies of the same region.
118  */
119 int
120 rman_manage_region(struct rman *rm, u_long start, u_long end)
121 {
122         struct resource *r, *s;
123         lwkt_tokref ilock;
124
125         DPRINTF(("rman_manage_region: <%s> request: start %#lx, end %#lx\n",
126             rm->rm_descr, start, end));
127         r = kmalloc(sizeof *r, M_RMAN, M_NOWAIT);
128         if (r == 0)
129                 return ENOMEM;
130         bzero(r, sizeof *r);
131         r->r_sharehead = 0;
132         r->r_start = start;
133         r->r_end = end;
134         r->r_flags = 0;
135         r->r_dev = 0;
136         r->r_rm = rm;
137
138         lwkt_gettoken(&ilock, rm->rm_slock);
139         for (s = CIRCLEQ_FIRST(&rm->rm_list);   
140              !CIRCLEQ_TERMCOND(s, rm->rm_list) && s->r_end < r->r_start;
141              s = CIRCLEQ_NEXT(s, r_link))
142                 ;
143
144         if (CIRCLEQ_TERMCOND(s, rm->rm_list)) {
145                 CIRCLEQ_INSERT_TAIL(&rm->rm_list, r, r_link);
146         } else {
147                 CIRCLEQ_INSERT_BEFORE(&rm->rm_list, s, r, r_link);
148         }
149
150         lwkt_reltoken(&ilock);
151         return 0;
152 }
153
154 int
155 rman_fini(struct rman *rm)
156 {
157         struct resource *r;
158         lwkt_tokref ilock;
159
160         lwkt_gettoken(&ilock, rm->rm_slock);
161         CIRCLEQ_FOREACH(r, &rm->rm_list, r_link) {
162                 if (r->r_flags & RF_ALLOCATED) {
163                         lwkt_reltoken(&ilock);
164                         return EBUSY;
165                 }
166         }
167
168         /*
169          * There really should only be one of these if we are in this
170          * state and the code is working properly, but it can't hurt.
171          */
172         while (!CIRCLEQ_EMPTY(&rm->rm_list)) {
173                 r = CIRCLEQ_FIRST(&rm->rm_list);
174                 CIRCLEQ_REMOVE(&rm->rm_list, r, r_link);
175                 kfree(r, M_RMAN);
176         }
177         lwkt_reltoken(&ilock);
178         /* XXX what's the point of this if we are going to free the struct? */
179         lwkt_gettoken(&ilock, &rman_tok);
180         TAILQ_REMOVE(&rman_head, rm, rm_link);
181         lwkt_reltoken(&ilock);
182         kfree(rm->rm_slock, M_RMAN);
183
184         return 0;
185 }
186
187 struct resource *
188 rman_reserve_resource(struct rman *rm, u_long start, u_long end, u_long count,
189                       u_int flags, struct device *dev)
190 {
191         u_int   want_activate;
192         struct  resource *r, *s, *rv;
193         u_long  rstart, rend;
194         lwkt_tokref ilock;
195
196         rv = 0;
197
198         DPRINTF(("rman_reserve_resource: <%s> request: [%#lx, %#lx], length "
199                "%#lx, flags %u, device %s\n", rm->rm_descr, start, end,
200                count, flags,
201                dev == NULL ? "<null>" : device_get_nameunit(dev)));
202         want_activate = (flags & RF_ACTIVE);
203         flags &= ~RF_ACTIVE;
204
205         lwkt_gettoken(&ilock, rm->rm_slock);
206
207         for (r = CIRCLEQ_FIRST(&rm->rm_list); 
208              !CIRCLEQ_TERMCOND(r, rm->rm_list) && r->r_end < start;
209              r = CIRCLEQ_NEXT(r, r_link))
210                 ;
211
212         if (CIRCLEQ_TERMCOND(r, rm->rm_list)) {
213                 DPRINTF(("could not find a region\n"));
214                 goto out;
215         }
216
217         /*
218          * First try to find an acceptable totally-unshared region.
219          */
220         for (s = r; !CIRCLEQ_TERMCOND(s, rm->rm_list);
221              s = CIRCLEQ_NEXT(s, r_link)) {
222                 DPRINTF(("considering [%#lx, %#lx]\n", s->r_start, s->r_end));
223                 if (s->r_start > end) {
224                         DPRINTF(("s->r_start (%#lx) > end (%#lx)\n",
225                             s->r_start, end));
226                         break;
227                 }
228                 if (s->r_flags & RF_ALLOCATED) {
229                         DPRINTF(("region is allocated\n"));
230                         continue;
231                 }
232                 rstart = max(s->r_start, start);
233                 rstart = (rstart + ((1ul << RF_ALIGNMENT(flags))) - 1) &
234                     ~((1ul << RF_ALIGNMENT(flags)) - 1);
235                 rend = min(s->r_end, max(start + count, end));
236                 DPRINTF(("truncated region: [%#lx, %#lx]; size %#lx (requested %#lx)\n",
237                        rstart, rend, (rend - rstart + 1), count));
238
239                 if ((rend - rstart + 1) >= count) {
240                         DPRINTF(("candidate region: [%#lx, %#lx], size %#lx\n",
241                                rstart, rend, (rend - rstart + 1)));
242                         if ((s->r_end - s->r_start + 1) == count) {
243                                 DPRINTF(("candidate region is entire chunk\n"));
244                                 rv = s;
245                                 rv->r_flags |= RF_ALLOCATED | flags;
246                                 rv->r_dev = dev;
247                                 goto out;
248                         }
249
250                         /*
251                          * If s->r_start < rstart and
252                          *    s->r_end > rstart + count - 1, then
253                          * we need to split the region into three pieces
254                          * (the middle one will get returned to the user).
255                          * Otherwise, we are allocating at either the
256                          * beginning or the end of s, so we only need to
257                          * split it in two.  The first case requires
258                          * two new allocations; the second requires but one.
259                          */
260                         rv = kmalloc(sizeof *rv, M_RMAN, M_NOWAIT);
261                         if (rv == 0)
262                                 goto out;
263                         bzero(rv, sizeof *rv);
264                         rv->r_start = rstart;
265                         rv->r_end = rstart + count - 1;
266                         rv->r_flags = flags | RF_ALLOCATED;
267                         rv->r_dev = dev;
268                         rv->r_sharehead = 0;
269                         rv->r_rm = rm;
270                         
271                         if (s->r_start < rv->r_start && s->r_end > rv->r_end) {
272                                 DPRINTF(("splitting region in three parts: "
273                                        "[%#lx, %#lx]; [%#lx, %#lx]; [%#lx, %#lx]\n",
274                                        s->r_start, rv->r_start - 1,
275                                        rv->r_start, rv->r_end,
276                                        rv->r_end + 1, s->r_end));
277                                 /*
278                                  * We are allocating in the middle.
279                                  */
280                                 r = kmalloc(sizeof *r, M_RMAN, M_NOWAIT);
281                                 if (r == 0) {
282                                         kfree(rv, M_RMAN);
283                                         rv = 0;
284                                         goto out;
285                                 }
286                                 bzero(r, sizeof *r);
287                                 r->r_start = rv->r_end + 1;
288                                 r->r_end = s->r_end;
289                                 r->r_flags = s->r_flags;
290                                 r->r_dev = 0;
291                                 r->r_sharehead = 0;
292                                 r->r_rm = rm;
293                                 s->r_end = rv->r_start - 1;
294                                 CIRCLEQ_INSERT_AFTER(&rm->rm_list, s, rv,
295                                                      r_link);
296                                 CIRCLEQ_INSERT_AFTER(&rm->rm_list, rv, r,
297                                                      r_link);
298                         } else if (s->r_start == rv->r_start) {
299                                 DPRINTF(("allocating from the beginning\n"));
300                                 /*
301                                  * We are allocating at the beginning.
302                                  */
303                                 s->r_start = rv->r_end + 1;
304                                 CIRCLEQ_INSERT_BEFORE(&rm->rm_list, s, rv,
305                                                       r_link);
306                         } else {
307                                 DPRINTF(("allocating at the end\n"));
308                                 /*
309                                  * We are allocating at the end.
310                                  */
311                                 s->r_end = rv->r_start - 1;
312                                 CIRCLEQ_INSERT_AFTER(&rm->rm_list, s, rv,
313                                                      r_link);
314                         }
315                         goto out;
316                 }
317         }
318
319         /*
320          * Now find an acceptable shared region, if the client's requirements
321          * allow sharing.  By our implementation restriction, a candidate
322          * region must match exactly by both size and sharing type in order
323          * to be considered compatible with the client's request.  (The
324          * former restriction could probably be lifted without too much
325          * additional work, but this does not seem warranted.)
326          */
327         DPRINTF(("no unshared regions found\n"));
328         if ((flags & (RF_SHAREABLE | RF_TIMESHARE)) == 0)
329                 goto out;
330
331         for (s = r; !CIRCLEQ_TERMCOND(s, rm->rm_list);
332              s = CIRCLEQ_NEXT(s, r_link)) {
333                 if (s->r_start > end)
334                         break;
335                 if ((s->r_flags & flags) != flags)
336                         continue;
337                 rstart = max(s->r_start, start);
338                 rend = min(s->r_end, max(start + count, end));
339                 if (s->r_start >= start && s->r_end <= end
340                     && (s->r_end - s->r_start + 1) == count) {
341                         rv = kmalloc(sizeof *rv, M_RMAN, M_NOWAIT);
342                         if (rv == 0)
343                                 goto out;
344                         bzero(rv, sizeof *rv);
345                         rv->r_start = s->r_start;
346                         rv->r_end = s->r_end;
347                         rv->r_flags = s->r_flags & 
348                                 (RF_ALLOCATED | RF_SHAREABLE | RF_TIMESHARE);
349                         rv->r_dev = dev;
350                         rv->r_rm = rm;
351                         if (s->r_sharehead == 0) {
352                                 s->r_sharehead = kmalloc(sizeof *s->r_sharehead,
353                                                         M_RMAN, M_NOWAIT);
354                                 if (s->r_sharehead == 0) {
355                                         kfree(rv, M_RMAN);
356                                         rv = 0;
357                                         goto out;
358                                 }
359                                 bzero(s->r_sharehead, sizeof *s->r_sharehead);
360                                 LIST_INIT(s->r_sharehead);
361                                 LIST_INSERT_HEAD(s->r_sharehead, s, 
362                                                  r_sharelink);
363                                 s->r_flags |= RF_FIRSTSHARE;
364                         }
365                         rv->r_sharehead = s->r_sharehead;
366                         LIST_INSERT_HEAD(s->r_sharehead, rv, r_sharelink);
367                         goto out;
368                 }
369         }
370
371         /*
372          * We couldn't find anything.
373          */
374 out:
375         /*
376          * If the user specified RF_ACTIVE in the initial flags,
377          * which is reflected in `want_activate', we attempt to atomically
378          * activate the resource.  If this fails, we release the resource
379          * and indicate overall failure.  (This behavior probably doesn't
380          * make sense for RF_TIMESHARE-type resources.)
381          */
382         if (rv && want_activate) {
383                 struct resource *whohas;
384                 if (int_rman_activate_resource(rm, rv, &whohas)) {
385                         int_rman_release_resource(rm, rv);
386                         rv = 0;
387                 }
388         }
389         lwkt_reltoken(&ilock);
390         return (rv);
391 }
392
393 static int
394 int_rman_activate_resource(struct rman *rm, struct resource *r,
395                            struct resource **whohas)
396 {
397         struct resource *s;
398         int ok;
399
400         /*
401          * If we are not timesharing, then there is nothing much to do.
402          * If we already have the resource, then there is nothing at all to do.
403          * If we are not on a sharing list with anybody else, then there is
404          * little to do.
405          */
406         if ((r->r_flags & RF_TIMESHARE) == 0
407             || (r->r_flags & RF_ACTIVE) != 0
408             || r->r_sharehead == 0) {
409                 r->r_flags |= RF_ACTIVE;
410                 return 0;
411         }
412
413         ok = 1;
414         for (s = LIST_FIRST(r->r_sharehead); s && ok;
415              s = LIST_NEXT(s, r_sharelink)) {
416                 if ((s->r_flags & RF_ACTIVE) != 0) {
417                         ok = 0;
418                         *whohas = s;
419                 }
420         }
421         if (ok) {
422                 r->r_flags |= RF_ACTIVE;
423                 return 0;
424         }
425         return EBUSY;
426 }
427
428 int
429 rman_activate_resource(struct resource *r)
430 {
431         int rv;
432         struct resource *whohas;
433         lwkt_tokref ilock;
434         struct rman *rm;
435
436         rm = r->r_rm;
437         lwkt_gettoken(&ilock, rm->rm_slock);
438         rv = int_rman_activate_resource(rm, r, &whohas);
439         lwkt_reltoken(&ilock);
440         return rv;
441 }
442
443 #if 0
444
445 /* XXX */
446 int
447 rman_await_resource(struct resource *r, lwkt_tokref_t ilock, int slpflags, int timo)
448 {
449         int     rv;
450         struct  resource *whohas;
451         struct  rman *rm;
452
453         rm = r->r_rm;
454         for (;;) {
455                 lwkt_gettoken(ilock, rm->rm_slock);
456                 rv = int_rman_activate_resource(rm, r, &whohas);
457                 if (rv != EBUSY)
458                         return (rv);    /* returns with ilock held */
459
460                 if (r->r_sharehead == 0)
461                         panic("rman_await_resource");
462                 /*
463                  * A critical section will hopefully will prevent a race 
464                  * between lwkt_reltoken and tsleep where a process
465                  * could conceivably get in and release the resource
466                  * before we have a chance to sleep on it. YYY
467                  */
468                 crit_enter();
469                 whohas->r_flags |= RF_WANTED;
470                 rv = tsleep(r->r_sharehead, slpflags, "rmwait", timo);
471                 if (rv) {
472                         lwkt_reltoken(ilock);
473                         crit_exit();
474                         return rv;
475                 }
476                 crit_exit();
477         }
478 }
479
480 #endif
481
482 static int
483 int_rman_deactivate_resource(struct resource *r)
484 {
485         struct  rman *rm;
486
487         rm = r->r_rm;
488         r->r_flags &= ~RF_ACTIVE;
489         if (r->r_flags & RF_WANTED) {
490                 r->r_flags &= ~RF_WANTED;
491                 wakeup(r->r_sharehead);
492         }
493         return 0;
494 }
495
496 int
497 rman_deactivate_resource(struct resource *r)
498 {
499         lwkt_tokref ilock;
500         struct rman *rm;
501
502         rm = r->r_rm;
503         lwkt_gettoken(&ilock, rm->rm_slock);
504         int_rman_deactivate_resource(r);
505         lwkt_reltoken(&ilock);
506         return 0;
507 }
508
509 static int
510 int_rman_release_resource(struct rman *rm, struct resource *r)
511 {
512         struct  resource *s, *t;
513
514         if (r->r_flags & RF_ACTIVE)
515                 int_rman_deactivate_resource(r);
516
517         /*
518          * Check for a sharing list first.  If there is one, then we don't
519          * have to think as hard.
520          */
521         if (r->r_sharehead) {
522                 /*
523                  * If a sharing list exists, then we know there are at
524                  * least two sharers.
525                  *
526                  * If we are in the main circleq, appoint someone else.
527                  */
528                 LIST_REMOVE(r, r_sharelink);
529                 s = LIST_FIRST(r->r_sharehead);
530                 if (r->r_flags & RF_FIRSTSHARE) {
531                         s->r_flags |= RF_FIRSTSHARE;
532                         CIRCLEQ_INSERT_BEFORE(&rm->rm_list, r, s, r_link);
533                         CIRCLEQ_REMOVE(&rm->rm_list, r, r_link);
534                 }
535
536                 /*
537                  * Make sure that the sharing list goes away completely
538                  * if the resource is no longer being shared at all.
539                  */
540                 if (LIST_NEXT(s, r_sharelink) == 0) {
541                         kfree(s->r_sharehead, M_RMAN);
542                         s->r_sharehead = 0;
543                         s->r_flags &= ~RF_FIRSTSHARE;
544                 }
545                 goto out;
546         }
547
548         /*
549          * Look at the adjacent resources in the list and see if our
550          * segment can be merged with any of them.
551          */
552         s = CIRCLEQ_PREV(r, r_link);
553         t = CIRCLEQ_NEXT(r, r_link);
554
555         if (s != (void *)&rm->rm_list && (s->r_flags & RF_ALLOCATED) == 0
556             && t != (void *)&rm->rm_list && (t->r_flags & RF_ALLOCATED) == 0) {
557                 /*
558                  * Merge all three segments.
559                  */
560                 s->r_end = t->r_end;
561                 CIRCLEQ_REMOVE(&rm->rm_list, r, r_link);
562                 CIRCLEQ_REMOVE(&rm->rm_list, t, r_link);
563                 kfree(t, M_RMAN);
564         } else if (s != (void *)&rm->rm_list
565                    && (s->r_flags & RF_ALLOCATED) == 0) {
566                 /*
567                  * Merge previous segment with ours.
568                  */
569                 s->r_end = r->r_end;
570                 CIRCLEQ_REMOVE(&rm->rm_list, r, r_link);
571         } else if (t != (void *)&rm->rm_list
572                    && (t->r_flags & RF_ALLOCATED) == 0) {
573                 /*
574                  * Merge next segment with ours.
575                  */
576                 t->r_start = r->r_start;
577                 CIRCLEQ_REMOVE(&rm->rm_list, r, r_link);
578         } else {
579                 /*
580                  * At this point, we know there is nothing we
581                  * can potentially merge with, because on each
582                  * side, there is either nothing there or what is
583                  * there is still allocated.  In that case, we don't
584                  * want to remove r from the list; we simply want to
585                  * change it to an unallocated region and return
586                  * without freeing anything.
587                  */
588                 r->r_flags &= ~RF_ALLOCATED;
589                 return 0;
590         }
591
592 out:
593         kfree(r, M_RMAN);
594         return 0;
595 }
596
597 int
598 rman_release_resource(struct resource *r)
599 {
600         struct  rman *rm = r->r_rm;
601         lwkt_tokref ilock;
602         int     rv;
603
604         lwkt_gettoken(&ilock, rm->rm_slock);
605         rv = int_rman_release_resource(rm, r);
606         lwkt_reltoken(&ilock);
607         return (rv);
608 }
609
610 uint32_t
611 rman_make_alignment_flags(uint32_t size)
612 {
613         int     i;
614
615         /*
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.
618          */
619         for (i = 32; i > 0; i--)
620                 if ((1 << i) & size)
621                         break;
622         if (~(1 << i) & size)
623                 i++;
624
625         return(RF_ALIGNMENT_LOG2(i));
626 }