A modern Music Player Daemon based on Rockbox open source high quality audio player
libadwaita
audio
rust
zig
deno
mpris
rockbox
mpd
1/***************************************************************************
2 * __________ __ ___.
3 * Open \______ \ ____ ____ | | _\_ |__ _______ ___
4 * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
7 * \/ \/ \/ \/ \/
8 * $Id$
9 *
10 * Copyright (C) 2002 by Björn Stenberg
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version 2
15 * of the License, or (at your option) any later version.
16 *
17 * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
18 * KIND, either express or implied.
19 *
20 ****************************************************************************/
21#include <string.h>
22#include "kernel-internal.h"
23#include "queue.h"
24#include "general.h"
25
26/* This array holds all queues that are initiated. It is used for broadcast. */
27static struct
28{
29 struct event_queue *queues[MAX_NUM_QUEUES+1];
30#ifdef HAVE_CORELOCK_OBJECT
31 struct corelock cl;
32#endif
33} all_queues SHAREDBSS_ATTR;
34
35/****************************************************************************
36 * Queue handling stuff
37 ****************************************************************************/
38
39#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
40/****************************************************************************
41 * Sender thread queue structure that aids implementation of priority
42 * inheritance on queues because the send list structure is the same as
43 * for all other kernel objects:
44 *
45 * Example state:
46 * E0 added with queue_send and removed by thread via queue_wait(_w_tmo)
47 * E3 was posted with queue_post
48 * 4 events remain enqueued (E1-E4)
49 *
50 * rd wr
51 * q->events[]: | XX | E1 | E2 | E3 | E4 | XX |
52 * q->send->senders[]: | NULL | T1 | T2 | NULL | T3 | NULL |
53 * \/ \/ \/
54 * q->send->list: 0<-|T0|<->|T1|<->|T2|<-------->|T3|->0
55 * q->send->curr_sender: /\
56 *
57 * Thread has E0 in its own struct queue_event.
58 *
59 ****************************************************************************/
60
61/* Puts the specified return value in the waiting thread's return value
62 * and wakes the thread.
63 *
64 * A sender should be confirmed to exist before calling which makes it
65 * more efficent to reject the majority of cases that don't need this
66 * called.
67 */
68static void queue_release_sender_inner(
69 struct thread_entry * volatile * sender, intptr_t retval)
70{
71 struct thread_entry *thread = *sender;
72 *sender = NULL; /* Clear slot. */
73 thread->retval = retval; /* Assign thread-local return value. */
74 wakeup_thread(thread, WAKEUP_RELEASE);
75}
76
77static inline void queue_release_sender(
78 struct thread_entry * volatile * sender, intptr_t retval)
79{
80 if(UNLIKELY(*sender))
81 queue_release_sender_inner(sender, retval);
82}
83
84/* Releases any waiting threads that are queued with queue_send -
85 * reply with 0.
86 */
87static void queue_release_all_senders(struct event_queue *q)
88{
89 if(q->send)
90 {
91 unsigned int i;
92 for(i = q->read; i != q->write; i++)
93 {
94 struct thread_entry **spp =
95 &q->send->senders[i & QUEUE_LENGTH_MASK];
96 queue_release_sender(spp, 0);
97 }
98 }
99}
100
101/* Enables queue_send on the specified queue - caller allocates the extra
102 * data structure. Only queues which are taken to be owned by a thread should
103 * enable this however an official owner is not compulsory but must be
104 * specified for priority inheritance to operate.
105 *
106 * Use of queue_wait(_w_tmo) by multiple threads on a queue using synchronous
107 * messages results in an undefined order of message replies or possible default
108 * replies if two or more waits happen before a reply is done.
109 */
110void queue_enable_queue_send(struct event_queue *q,
111 struct queue_sender_list *send,
112 unsigned int owner_id)
113{
114 int oldlevel = disable_irq_save();
115 corelock_lock(&q->cl);
116
117 if(send != NULL && q->send == NULL)
118 {
119 memset(send, 0, sizeof(*send));
120 wait_queue_init(&send->list);
121#ifdef HAVE_PRIORITY_SCHEDULING
122 blocker_init(&send->blocker);
123 if(owner_id != 0)
124 {
125 send->blocker.thread = __thread_id_entry(owner_id);
126 q->blocker_p = &send->blocker;
127 }
128#endif
129 q->send = send;
130 }
131
132 corelock_unlock(&q->cl);
133 restore_irq(oldlevel);
134
135 (void)owner_id;
136}
137
138/* Unblock a blocked thread at a given event index */
139static inline void queue_do_unblock_sender(struct queue_sender_list *send,
140 unsigned int i)
141{
142 if(send)
143 queue_release_sender(&send->senders[i], 0);
144}
145
146/* Perform the auto-reply sequence */
147static inline void queue_do_auto_reply(struct queue_sender_list *send)
148{
149 if(send)
150 queue_release_sender(&send->curr_sender, 0);
151}
152
153/* Moves waiting thread's refrence from the senders array to the
154 * current_sender which represents the thread waiting for a reponse to the
155 * last message removed from the queue. This also protects the thread from
156 * being bumped due to overflow which would not be a valid action since its
157 * message _is_ being processed at this point. */
158static inline void queue_do_fetch_sender(struct queue_sender_list *send,
159 unsigned int rd)
160{
161 if(send)
162 {
163 struct thread_entry **spp = &send->senders[rd];
164
165 if(*spp)
166 {
167 /* Move thread reference from array to the next thread
168 that queue_reply will release */
169 send->curr_sender = *spp;
170 *spp = NULL;
171 }
172 /* else message was posted asynchronously with queue_post */
173 }
174}
175#else
176/* Empty macros for when synchoronous sending is not made */
177#define queue_release_all_senders(q)
178#define queue_do_unblock_sender(send, i)
179#define queue_do_auto_reply(send)
180#define queue_do_fetch_sender(send, rd)
181#endif /* HAVE_EXTENDED_MESSAGING_AND_NAME */
182
183static void queue_wake_waiter_inner(struct thread_entry *thread)
184{
185 wakeup_thread(thread, WAKEUP_DEFAULT);
186}
187
188static inline void queue_wake_waiter(struct event_queue *q)
189{
190 struct thread_entry *thread = WQ_THREAD_FIRST(&q->queue);
191 if(thread != NULL)
192 queue_wake_waiter_inner(thread);
193}
194
195/* Queue must not be available for use during this call */
196void queue_init(struct event_queue *q, bool register_queue)
197{
198 int oldlevel = disable_irq_save();
199
200 if(register_queue)
201 corelock_lock(&all_queues.cl);
202
203 corelock_init(&q->cl);
204 wait_queue_init(&q->queue);
205 /* What garbage is in write is irrelevant because of the masking design-
206 * any other functions the empty the queue do this as well so that
207 * queue_count and queue_empty return sane values in the case of a
208 * concurrent change without locking inside them. */
209 q->read = q->write;
210#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
211 q->send = NULL; /* No message sending by default */
212 IF_PRIO( q->blocker_p = NULL; )
213#endif
214
215 if(register_queue)
216 {
217 void **queues = (void **)all_queues.queues;
218 void **p = find_array_ptr(queues, q);
219
220 if(p - queues >= MAX_NUM_QUEUES)
221 {
222 panicf("queue_init->out of queues");
223 }
224
225 if(*p == NULL)
226 {
227 /* Add it to the all_queues array */
228 *p = q;
229 corelock_unlock(&all_queues.cl);
230 }
231 }
232
233 restore_irq(oldlevel);
234}
235
236/* Queue must not be available for use during this call */
237void queue_delete(struct event_queue *q)
238{
239 int oldlevel = disable_irq_save();
240 corelock_lock(&all_queues.cl);
241 corelock_lock(&q->cl);
242
243 /* Remove the queue if registered */
244 remove_array_ptr((void **)all_queues.queues, q);
245
246 corelock_unlock(&all_queues.cl);
247
248 /* Release thread(s) waiting on queue head */
249 wait_queue_wake(&q->queue);
250
251#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
252 if(q->send)
253 {
254 /* Release threads waiting for replies */
255 queue_release_all_senders(q);
256
257 /* Reply to any dequeued message waiting for one */
258 queue_do_auto_reply(q->send);
259
260 q->send = NULL;
261 IF_PRIO( q->blocker_p = NULL; )
262 }
263#endif
264
265 q->read = q->write;
266
267 corelock_unlock(&q->cl);
268 restore_irq(oldlevel);
269}
270
271/* NOTE: multiple threads waiting on a queue head cannot have a well-
272 defined release order if timeouts are used. If multiple threads must
273 access the queue head, use a dispatcher or queue_wait only. */
274void queue_wait(struct event_queue *q, struct queue_event *ev)
275{
276 int oldlevel;
277 unsigned int rd;
278
279#ifdef HAVE_PRIORITY_SCHEDULING
280 KERNEL_ASSERT(QUEUE_GET_THREAD(q) == NULL ||
281 QUEUE_GET_THREAD(q) == __running_self_entry(),
282 "queue_wait->wrong thread\n");
283#endif
284
285 oldlevel = disable_irq_save();
286
287 ASSERT_CPU_MODE(CPU_MODE_THREAD_CONTEXT, oldlevel);
288
289 corelock_lock(&q->cl);
290
291#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
292 /* Auto-reply (even if ev is NULL to avoid stalling a waiting thread) */
293 queue_do_auto_reply(q->send);
294#endif
295
296 while(1)
297 {
298 rd = q->read;
299 if (rd != q->write) /* A waking message could disappear */
300 break;
301
302 struct thread_entry *current = __running_self_entry();
303 block_thread(current, TIMEOUT_BLOCK, &q->queue, NULL);
304
305 corelock_unlock(&q->cl);
306 switch_thread();
307
308 disable_irq();
309 corelock_lock(&q->cl);
310 }
311
312#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
313 if(ev)
314#endif
315 {
316 q->read = rd + 1;
317 rd &= QUEUE_LENGTH_MASK;
318 *ev = q->events[rd];
319
320 /* Get data for a waiting thread if one */
321 queue_do_fetch_sender(q->send, rd);
322 }
323 /* else just waiting on non-empty */
324
325 corelock_unlock(&q->cl);
326 restore_irq(oldlevel);
327}
328
329void queue_wait_w_tmo(struct event_queue *q, struct queue_event *ev, int ticks)
330{
331 int oldlevel;
332 unsigned int rd, wr;
333
334#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
335 KERNEL_ASSERT(QUEUE_GET_THREAD(q) == NULL ||
336 QUEUE_GET_THREAD(q) == __running_self_entry(),
337 "queue_wait_w_tmo->wrong thread\n");
338#endif
339
340 oldlevel = disable_irq_save();
341
342 corelock_lock(&q->cl);
343
344#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
345 /* Auto-reply (even if ev is NULL to avoid stalling a waiting thread) */
346 queue_do_auto_reply(q->send);
347#endif
348
349 rd = q->read;
350 wr = q->write;
351
352 if(rd != wr || ticks == 0)
353 ; /* no block */
354 else while(1)
355 {
356 ASSERT_CPU_MODE(CPU_MODE_THREAD_CONTEXT, oldlevel);
357
358 struct thread_entry *current = __running_self_entry();
359 block_thread(current, ticks, &q->queue, NULL);
360
361 corelock_unlock(&q->cl);
362 switch_thread();
363
364 disable_irq();
365 corelock_lock(&q->cl);
366
367 rd = q->read;
368 wr = q->write;
369
370 if(rd != wr)
371 break;
372
373 if(ticks < 0)
374 continue; /* empty again, infinite block */
375
376 /* timeout is legit if thread is still queued and awake */
377 if(LIKELY(wait_queue_try_remove(current)))
378 break;
379
380 /* we mustn't return earlier than expected wait time */
381 ticks = get_tmo_tick(current) - current_tick;
382 if(ticks <= 0)
383 break;
384 }
385
386#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
387 if(UNLIKELY(!ev))
388 ; /* just waiting for something */
389 else
390#endif
391 if(rd != wr)
392 {
393 q->read = rd + 1;
394 rd &= QUEUE_LENGTH_MASK;
395 *ev = q->events[rd];
396
397 /* Get data for a waiting thread if one */
398 queue_do_fetch_sender(q->send, rd);
399 }
400 else
401 {
402 ev->id = SYS_TIMEOUT;
403 ev->data = 0;
404 }
405
406 corelock_unlock(&q->cl);
407 restore_irq(oldlevel);
408}
409
410void queue_post(struct event_queue *q, long id, intptr_t data)
411{
412 int oldlevel;
413 unsigned int wr;
414
415 oldlevel = disable_irq_save();
416 corelock_lock(&q->cl);
417
418 wr = q->write++ & QUEUE_LENGTH_MASK;
419
420 KERNEL_ASSERT((q->write - q->read) <= QUEUE_LENGTH,
421 "queue_post ovf q=%p", q);
422
423 q->events[wr].id = id;
424 q->events[wr].data = data;
425
426 /* overflow protect - unblock any thread waiting at this index */
427 queue_do_unblock_sender(q->send, wr);
428
429 /* Wakeup a waiting thread if any */
430 queue_wake_waiter(q);
431
432 corelock_unlock(&q->cl);
433 restore_irq(oldlevel);
434}
435
436#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
437/* IRQ handlers are not allowed use of this function - we only aim to
438 protect the queue integrity by turning them off. */
439intptr_t queue_send(struct event_queue *q, long id, intptr_t data)
440{
441 int oldlevel;
442 unsigned int wr;
443
444 oldlevel = disable_irq_save();
445
446 ASSERT_CPU_MODE(CPU_MODE_THREAD_CONTEXT, oldlevel);
447
448 corelock_lock(&q->cl);
449
450 wr = q->write++ & QUEUE_LENGTH_MASK;
451
452 KERNEL_ASSERT((q->write - q->read) <= QUEUE_LENGTH,
453 "queue_send ovf q=%p", q);
454
455 q->events[wr].id = id;
456 q->events[wr].data = data;
457
458 if(LIKELY(q->send))
459 {
460 struct queue_sender_list *send = q->send;
461 struct thread_entry **spp = &send->senders[wr];
462 struct thread_entry *current = __running_self_entry();
463
464 /* overflow protect - unblock any thread waiting at this index */
465 queue_release_sender(spp, 0);
466
467 /* Wakeup a waiting thread if any */
468 queue_wake_waiter(q);
469
470 /* Save thread in slot, add to list and wait for reply */
471 *spp = current;
472 block_thread(current, TIMEOUT_BLOCK, &send->list, q->blocker_p);
473
474 corelock_unlock(&q->cl);
475 switch_thread();
476
477 return current->retval;
478 }
479
480 /* Function as queue_post if sending is not enabled */
481 queue_wake_waiter(q);
482
483 corelock_unlock(&q->cl);
484 restore_irq(oldlevel);
485
486 return 0;
487}
488
489#if 0 /* not used now but probably will be later */
490/* Query if the last message dequeued was added by queue_send or not */
491bool queue_in_queue_send(struct event_queue *q)
492{
493 bool in_send;
494
495#if NUM_CORES > 1
496 int oldlevel = disable_irq_save();
497 corelock_lock(&q->cl);
498#endif
499
500 in_send = q->send && q->send->curr_sender;
501
502#if NUM_CORES > 1
503 corelock_unlock(&q->cl);
504 restore_irq(oldlevel);
505#endif
506
507 return in_send;
508}
509#endif
510
511/* Replies with retval to the last dequeued message sent with queue_send */
512void queue_reply(struct event_queue *q, intptr_t retval)
513{
514 if(q->send && q->send->curr_sender)
515 {
516 int oldlevel = disable_irq_save();
517 corelock_lock(&q->cl);
518
519 struct queue_sender_list *send = q->send;
520 if(send)
521 queue_release_sender(&send->curr_sender, retval);
522
523 corelock_unlock(&q->cl);
524 restore_irq(oldlevel);
525 }
526}
527#endif /* HAVE_EXTENDED_MESSAGING_AND_NAME */
528
529#ifdef HAVE_EXTENDED_MESSAGING_AND_NAME
530/* Scan the even queue from head to tail, returning any event from the
531 filter list that was found, optionally removing the event. If an
532 event is returned, synchronous events are handled in the same manner as
533 with queue_wait(_w_tmo); if discarded, then as queue_clear.
534 If filters are NULL, any event matches. If filters exist, the default
535 is to search the full queue depth.
536 Earlier filters take precedence.
537
538 Return true if an event was found, false otherwise. */
539bool queue_peek_ex(struct event_queue *q, struct queue_event *ev,
540 unsigned int flags, const long (*filters)[2])
541{
542 bool have_msg;
543 unsigned int rd, wr;
544 int oldlevel;
545
546 if(LIKELY(q->read == q->write))
547 return false; /* Empty: do nothing further */
548
549 have_msg = false;
550
551 oldlevel = disable_irq_save();
552 corelock_lock(&q->cl);
553
554 /* Starting at the head, find first match */
555 for(rd = q->read, wr = q->write; rd != wr; rd++)
556 {
557 struct queue_event *e = &q->events[rd & QUEUE_LENGTH_MASK];
558
559 if(filters)
560 {
561 /* Have filters - find the first thing that passes */
562 const long (* f)[2] = filters;
563 const long (* const f_last)[2] =
564 &filters[flags & QPEEK_FILTER_COUNT_MASK];
565 long id = e->id;
566
567 do
568 {
569 if(UNLIKELY(id >= (*f)[0] && id <= (*f)[1]))
570 goto passed_filter;
571 }
572 while(++f <= f_last);
573
574 if(LIKELY(!(flags & QPEEK_FILTER_HEAD_ONLY)))
575 continue; /* No match; test next event */
576 else
577 break; /* Only check the head */
578 }
579 /* else - anything passes */
580
581 passed_filter:
582
583 /* Found a matching event */
584 have_msg = true;
585
586 if(ev)
587 *ev = *e; /* Caller wants the event */
588
589 if(flags & QPEEK_REMOVE_EVENTS)
590 {
591 /* Do event removal */
592 unsigned int r = q->read;
593 q->read = r + 1; /* Advance head */
594
595 if(ev)
596 {
597 /* Auto-reply */
598 queue_do_auto_reply(q->send);
599 /* Get the thread waiting for reply, if any */
600 queue_do_fetch_sender(q->send, rd & QUEUE_LENGTH_MASK);
601 }
602 else
603 {
604 /* Release any thread waiting on this message */
605 queue_do_unblock_sender(q->send, rd & QUEUE_LENGTH_MASK);
606 }
607
608 /* Slide messages forward into the gap if not at the head */
609 while(rd != r)
610 {
611 unsigned int dst = rd & QUEUE_LENGTH_MASK;
612 unsigned int src = --rd & QUEUE_LENGTH_MASK;
613
614 q->events[dst] = q->events[src];
615 /* Keep sender wait list in sync */
616 if(q->send)
617 q->send->senders[dst] = q->send->senders[src];
618 }
619 }
620
621 break;
622 }
623
624 corelock_unlock(&q->cl);
625 restore_irq(oldlevel);
626
627 return have_msg;
628}
629
630bool queue_peek(struct event_queue *q, struct queue_event *ev)
631{
632 return queue_peek_ex(q, ev, 0, NULL);
633}
634
635void queue_remove_from_head(struct event_queue *q, long id)
636{
637 const long f[2] = { id, id };
638 while (queue_peek_ex(q, NULL,
639 QPEEK_FILTER_HEAD_ONLY | QPEEK_REMOVE_EVENTS, &f));
640}
641#else /* !HAVE_EXTENDED_MESSAGING_AND_NAME */
642/* The more powerful routines aren't required */
643bool queue_peek(struct event_queue *q, struct queue_event *ev)
644{
645 unsigned int rd;
646
647 if(q->read == q->write)
648 return false;
649
650 bool have_msg = false;
651
652 int oldlevel = disable_irq_save();
653 corelock_lock(&q->cl);
654
655 rd = q->read;
656 if(rd != q->write)
657 {
658 *ev = q->events[rd & QUEUE_LENGTH_MASK];
659 have_msg = true;
660 }
661
662 corelock_unlock(&q->cl);
663 restore_irq(oldlevel);
664
665 return have_msg;
666}
667
668void queue_remove_from_head(struct event_queue *q, long id)
669{
670 int oldlevel;
671
672 oldlevel = disable_irq_save();
673 corelock_lock(&q->cl);
674
675 while(q->read != q->write)
676 {
677 unsigned int rd = q->read & QUEUE_LENGTH_MASK;
678
679 if(q->events[rd].id != id)
680 {
681 break;
682 }
683
684 /* Release any thread waiting on this message */
685 queue_do_unblock_sender(q->send, rd);
686
687 q->read++;
688 }
689
690 corelock_unlock(&q->cl);
691 restore_irq(oldlevel);
692}
693#endif /* HAVE_EXTENDED_MESSAGING_AND_NAME */
694
695/* Poll queue to see if a message exists - careful in using the result if
696 * queue_remove_from_head is called when messages are posted - possibly use
697 * queue_wait_w_tmo(&q, 0) in that case or else a removed message that
698 * unsignals the queue may cause an unwanted block */
699bool queue_empty(const struct event_queue* q)
700{
701 return ( q->read == q->write );
702}
703
704/* Poll queue to see if it is full */
705bool queue_full(const struct event_queue* q)
706{
707 return ((q->write - q->read) >= QUEUE_LENGTH);
708}
709
710void queue_clear(struct event_queue* q)
711{
712 int oldlevel;
713
714 oldlevel = disable_irq_save();
715 corelock_lock(&q->cl);
716
717 /* Release all threads waiting in the queue for a reply -
718 dequeued sent message will be handled by owning thread */
719 queue_release_all_senders(q);
720
721 q->read = q->write;
722
723 corelock_unlock(&q->cl);
724 restore_irq(oldlevel);
725}
726
727/**
728 * The number of events waiting in the queue.
729 *
730 * @param struct of event_queue
731 * @return number of events in the queue
732 */
733int queue_count(const struct event_queue *q)
734{
735 return q->write - q->read;
736}
737
738int queue_broadcast(long id, intptr_t data)
739{
740 struct event_queue **p = all_queues.queues;
741 struct event_queue *q;
742
743#if NUM_CORES > 1
744 int oldlevel = disable_irq_save();
745 corelock_lock(&all_queues.cl);
746#endif
747
748 for(q = *p; q != NULL; q = *(++p))
749 {
750 queue_post(q, id, data);
751 }
752
753#if NUM_CORES > 1
754 corelock_unlock(&all_queues.cl);
755 restore_irq(oldlevel);
756#endif
757
758 return p - all_queues.queues;
759}
760
761void init_queues(void)
762{
763 corelock_init(&all_queues.cl);
764}