1
0
mirror of https://github.com/mpv-player/mpv synced 2024-12-28 01:52:19 +00:00
mpv/misc/dispatch.c

398 lines
16 KiB
C
Raw Normal View History

/*
* This file is part of mpv.
*
Relicense some non-MPlayer source files to LGPL 2.1 or later This covers source files which were added in mplayer2 and mpv times only, and where all code is covered by LGPL relicensing agreements. There are probably more files to which this applies, but I'm being conservative here. A file named ao_sdl.c exists in MPlayer too, but the mpv one is a complete rewrite, and was added some time after the original ao_sdl.c was removed. The same applies to vo_sdl.c, for which the SDL2 API is radically different in addition (MPlayer supports SDL 1.2 only). common.c contains only code written by me. But common.h is a strange case: although it originally was named mp_common.h and exists in MPlayer too, by now it contains only definitions written by uau and me. The exceptions are the CONTROL_ defines - thus not changing the license of common.h yet. codec_tags.c contained once large tables generated from MPlayer's codecs.conf, but all of these tables were removed. From demux_playlist.c I'm removing a code fragment from someone who was not asked; this probably could be done later (see commit 15dccc37). misc.c is a bit complicated to reason about (it was split off mplayer.c and thus contains random functions out of this file), but actually all functions have been added post-MPlayer. Except get_relative_time(), which was written by uau, but looks similar to 3 different versions of something similar in each of the Unix/win32/OSX timer source files. I'm not sure what that means in regards to copyright, so I've just moved it into another still-GPL source file for now. screenshot.c once had some minor parts of MPlayer's vf_screenshot.c, but they're all gone.
2016-01-19 17:36:06 +00:00
* mpv is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* mpv is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Relicense some non-MPlayer source files to LGPL 2.1 or later This covers source files which were added in mplayer2 and mpv times only, and where all code is covered by LGPL relicensing agreements. There are probably more files to which this applies, but I'm being conservative here. A file named ao_sdl.c exists in MPlayer too, but the mpv one is a complete rewrite, and was added some time after the original ao_sdl.c was removed. The same applies to vo_sdl.c, for which the SDL2 API is radically different in addition (MPlayer supports SDL 1.2 only). common.c contains only code written by me. But common.h is a strange case: although it originally was named mp_common.h and exists in MPlayer too, by now it contains only definitions written by uau and me. The exceptions are the CONTROL_ defines - thus not changing the license of common.h yet. codec_tags.c contained once large tables generated from MPlayer's codecs.conf, but all of these tables were removed. From demux_playlist.c I'm removing a code fragment from someone who was not asked; this probably could be done later (see commit 15dccc37). misc.c is a bit complicated to reason about (it was split off mplayer.c and thus contains random functions out of this file), but actually all functions have been added post-MPlayer. Except get_relative_time(), which was written by uau, but looks similar to 3 different versions of something similar in each of the Unix/win32/OSX timer source files. I'm not sure what that means in regards to copyright, so I've just moved it into another still-GPL source file for now. screenshot.c once had some minor parts of MPlayer's vf_screenshot.c, but they're all gone.
2016-01-19 17:36:06 +00:00
* GNU Lesser General Public License for more details.
*
Relicense some non-MPlayer source files to LGPL 2.1 or later This covers source files which were added in mplayer2 and mpv times only, and where all code is covered by LGPL relicensing agreements. There are probably more files to which this applies, but I'm being conservative here. A file named ao_sdl.c exists in MPlayer too, but the mpv one is a complete rewrite, and was added some time after the original ao_sdl.c was removed. The same applies to vo_sdl.c, for which the SDL2 API is radically different in addition (MPlayer supports SDL 1.2 only). common.c contains only code written by me. But common.h is a strange case: although it originally was named mp_common.h and exists in MPlayer too, by now it contains only definitions written by uau and me. The exceptions are the CONTROL_ defines - thus not changing the license of common.h yet. codec_tags.c contained once large tables generated from MPlayer's codecs.conf, but all of these tables were removed. From demux_playlist.c I'm removing a code fragment from someone who was not asked; this probably could be done later (see commit 15dccc37). misc.c is a bit complicated to reason about (it was split off mplayer.c and thus contains random functions out of this file), but actually all functions have been added post-MPlayer. Except get_relative_time(), which was written by uau, but looks similar to 3 different versions of something similar in each of the Unix/win32/OSX timer source files. I'm not sure what that means in regards to copyright, so I've just moved it into another still-GPL source file for now. screenshot.c once had some minor parts of MPlayer's vf_screenshot.c, but they're all gone.
2016-01-19 17:36:06 +00:00
* You should have received a copy of the GNU Lesser General Public
* License along with mpv. If not, see <http://www.gnu.org/licenses/>.
*/
#include <stdbool.h>
#include <assert.h>
#include "common/common.h"
#include "osdep/threads.h"
#include "osdep/timer.h"
#include "dispatch.h"
struct mp_dispatch_queue {
struct mp_dispatch_item *head, *tail;
pthread_mutex_t lock;
pthread_cond_t cond;
void (*wakeup_fn)(void *wakeup_ctx);
void *wakeup_ctx;
// Time at which mp_dispatch_queue_process() should return.
int64_t wait;
// Make mp_dispatch_queue_process() exit if it's idle.
bool interrupted;
// The target thread is in mp_dispatch_queue_process() (and either idling,
// locked, or running a dispatch callback).
bool in_process;
pthread_t in_process_thread;
// The target thread is in mp_dispatch_queue_process(), and currently
// something has exclusive access to it (e.g. running a dispatch callback,
// or a different thread got it with mp_dispatch_lock()).
bool locked;
// A mp_dispatch_lock() call is requesting an exclusive lock.
size_t lock_requests;
// locked==true is due to a mp_dispatch_lock() call (for debugging).
bool locked_explicit;
pthread_t locked_explicit_thread;
};
struct mp_dispatch_item {
mp_dispatch_fn fn;
void *fn_data;
bool asynchronous;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
bool mergeable;
bool completed;
struct mp_dispatch_item *next;
};
static void queue_dtor(void *p)
{
struct mp_dispatch_queue *queue = p;
assert(!queue->head);
assert(!queue->in_process);
assert(!queue->lock_requests);
assert(!queue->locked);
pthread_cond_destroy(&queue->cond);
pthread_mutex_destroy(&queue->lock);
}
2016-02-26 22:28:02 +00:00
// A dispatch queue lets other threads run callbacks in a target thread.
// The target thread is the thread which calls mp_dispatch_queue_process().
// Free the dispatch queue with talloc_free(). At the time of destruction,
// the queue must be empty. The easiest way to guarantee this is to
// terminate all potential senders, then call mp_dispatch_run() with a
// function that e.g. makes the target thread exit, then pthread_join() the
// target thread, and finally destroy the queue. Another way is calling
// mp_dispatch_queue_process() after terminating all potential senders, and
// then destroying the queue.
struct mp_dispatch_queue *mp_dispatch_create(void *ta_parent)
{
struct mp_dispatch_queue *queue = talloc_ptrtype(ta_parent, queue);
*queue = (struct mp_dispatch_queue){0};
talloc_set_destructor(queue, queue_dtor);
pthread_mutex_init(&queue->lock, NULL);
pthread_cond_init(&queue->cond, NULL);
return queue;
}
// Set a custom function that should be called to guarantee that the target
// thread wakes up. This is intended for use with code that needs to block
// on non-pthread primitives, such as e.g. select(). In the case of select(),
// the wakeup_fn could for example write a byte into a "wakeup" pipe in order
// to unblock the select(). The wakeup_fn is called from the dispatch queue
// when there are new dispatch items, and the target thread should then enter
// mp_dispatch_queue_process() as soon as possible. Note that wakeup_fn is
// called under no lock, so you might have to do synchronization yourself.
void mp_dispatch_set_wakeup_fn(struct mp_dispatch_queue *queue,
void (*wakeup_fn)(void *wakeup_ctx),
void *wakeup_ctx)
{
queue->wakeup_fn = wakeup_fn;
queue->wakeup_ctx = wakeup_ctx;
}
static void mp_dispatch_append(struct mp_dispatch_queue *queue,
struct mp_dispatch_item *item)
{
pthread_mutex_lock(&queue->lock);
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
if (item->mergeable) {
for (struct mp_dispatch_item *cur = queue->head; cur; cur = cur->next) {
if (cur->mergeable && cur->fn == item->fn &&
cur->fn_data == item->fn_data)
{
talloc_free(item);
pthread_mutex_unlock(&queue->lock);
return;
}
}
}
if (queue->tail) {
queue->tail->next = item;
} else {
queue->head = item;
}
queue->tail = item;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
// Wake up the main thread; note that other threads might wait on this
// condition for reasons, so broadcast the condition.
pthread_cond_broadcast(&queue->cond);
// No wakeup callback -> assume mp_dispatch_queue_process() needs to be
// interrupted instead.
if (!queue->wakeup_fn)
queue->interrupted = true;
pthread_mutex_unlock(&queue->lock);
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
if (queue->wakeup_fn)
queue->wakeup_fn(queue->wakeup_ctx);
}
// Enqueue a callback to run it on the target thread asynchronously. The target
// thread will run fn(fn_data) as soon as it enter mp_dispatch_queue_process.
// Note that mp_dispatch_enqueue() will usually return long before that happens.
// It's up to the user to signal completion of the callback. It's also up to
// the user to guarantee that the context fn_data has correct lifetime, i.e.
// lives until the callback is run, and is freed after that.
void mp_dispatch_enqueue(struct mp_dispatch_queue *queue,
mp_dispatch_fn fn, void *fn_data)
{
struct mp_dispatch_item *item = talloc_ptrtype(NULL, item);
*item = (struct mp_dispatch_item){
.fn = fn,
.fn_data = fn_data,
.asynchronous = true,
};
mp_dispatch_append(queue, item);
}
// Like mp_dispatch_enqueue(), but the queue code will call talloc_free(fn_data)
// after the fn callback has been run. (The callback could trivially do that
// itself, but it makes it easier to implement synchronous and asynchronous
// requests with the same callback implementation.)
void mp_dispatch_enqueue_autofree(struct mp_dispatch_queue *queue,
mp_dispatch_fn fn, void *fn_data)
{
struct mp_dispatch_item *item = talloc_ptrtype(NULL, item);
*item = (struct mp_dispatch_item){
.fn = fn,
.fn_data = talloc_steal(item, fn_data),
.asynchronous = true,
};
mp_dispatch_append(queue, item);
}
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
// Like mp_dispatch_enqueue(), but
void mp_dispatch_enqueue_notify(struct mp_dispatch_queue *queue,
mp_dispatch_fn fn, void *fn_data)
{
struct mp_dispatch_item *item = talloc_ptrtype(NULL, item);
*item = (struct mp_dispatch_item){
.fn = fn,
.fn_data = fn_data,
.mergeable = true,
.asynchronous = true,
};
mp_dispatch_append(queue, item);
}
// Remove already queued item. Only items enqueued with the following functions
// can be canceled:
// - mp_dispatch_enqueue()
// - mp_dispatch_enqueue_notify()
// Items which were enqueued, and which are currently executing, can not be
// canceled anymore. This function is mostly for being called from the same
// context as mp_dispatch_queue_process(), where the "currently executing" case
// can be excluded.
void mp_dispatch_cancel_fn(struct mp_dispatch_queue *queue,
mp_dispatch_fn fn, void *fn_data)
{
pthread_mutex_lock(&queue->lock);
struct mp_dispatch_item **pcur = &queue->head;
queue->tail = NULL;
while (*pcur) {
struct mp_dispatch_item *cur = *pcur;
if (cur->fn == fn && cur->fn_data == fn_data) {
*pcur = cur->next;
talloc_free(cur);
} else {
queue->tail = cur;
pcur = &cur->next;
}
}
pthread_mutex_unlock(&queue->lock);
}
// Run fn(fn_data) on the target thread synchronously. This function enqueues
// the callback and waits until the target thread is done doing this.
// This is redundant to calling the function inside mp_dispatch_[un]lock(),
// but can be helpful with code that relies on TLS (such as OpenGL).
void mp_dispatch_run(struct mp_dispatch_queue *queue,
mp_dispatch_fn fn, void *fn_data)
{
struct mp_dispatch_item item = {
.fn = fn,
.fn_data = fn_data,
};
mp_dispatch_append(queue, &item);
pthread_mutex_lock(&queue->lock);
while (!item.completed)
pthread_cond_wait(&queue->cond, &queue->lock);
pthread_mutex_unlock(&queue->lock);
}
// Process any outstanding dispatch items in the queue. This also handles
// suspending or locking the this thread from another thread via
// mp_dispatch_lock().
// The timeout specifies the minimum wait time. The actual time spent in this
// function can be much higher if the suspending/locking functions are used, or
// if executing the dispatch items takes time. On the other hand, this function
// can return much earlier than the timeout due to sporadic wakeups.
// Note that this will strictly return only after:
// - timeout has passed,
// - all queue items were processed,
// - the possibly acquired lock has been released
// It's possible to cancel the timeout by calling mp_dispatch_interrupt().
// Reentrant calls are not allowed. There can be only 1 thread calling
// mp_dispatch_queue_process() at a time. In addition, mp_dispatch_lock() can
// not be called from a thread that is calling mp_dispatch_queue_process() (i.e.
// no enqueued callback can call the lock/unlock functions).
void mp_dispatch_queue_process(struct mp_dispatch_queue *queue, double timeout)
{
pthread_mutex_lock(&queue->lock);
queue->wait = timeout > 0 ? mp_add_timeout(mp_time_us(), timeout) : 0;
assert(!queue->in_process); // recursion not allowed
queue->in_process = true;
queue->in_process_thread = pthread_self();
// Wake up thread which called mp_dispatch_lock().
if (queue->lock_requests)
pthread_cond_broadcast(&queue->cond);
while (1) {
if (queue->lock_requests) {
// Block due to something having called mp_dispatch_lock().
pthread_cond_wait(&queue->cond, &queue->lock);
} else if (queue->head) {
struct mp_dispatch_item *item = queue->head;
queue->head = item->next;
if (!queue->head)
queue->tail = NULL;
item->next = NULL;
// Unlock, because we want to allow other threads to queue items
// while the dispatch item is processed.
// At the same time, we must prevent other threads from returning
// from mp_dispatch_lock(), which is done by locked=true.
2018-05-05 18:19:34 +00:00
assert(!queue->locked);
queue->locked = true;
pthread_mutex_unlock(&queue->lock);
item->fn(item->fn_data);
pthread_mutex_lock(&queue->lock);
assert(queue->locked);
queue->locked = false;
// Wakeup mp_dispatch_run(), also mp_dispatch_lock().
pthread_cond_broadcast(&queue->cond);
if (item->asynchronous) {
talloc_free(item);
} else {
item->completed = true;
}
} else if (queue->wait > 0 && !queue->interrupted) {
struct timespec ts = mp_time_us_to_timespec(queue->wait);
if (pthread_cond_timedwait(&queue->cond, &queue->lock, &ts))
queue->wait = 0;
} else {
break;
}
}
assert(!queue->locked);
queue->in_process = false;
queue->interrupted = false;
pthread_mutex_unlock(&queue->lock);
}
// If the queue is inside of mp_dispatch_queue_process(), make it return as
// soon as all work items have been run, without waiting for the timeout. This
// does not make it return early if it's blocked by a mp_dispatch_lock().
// If the queue is _not_ inside of mp_dispatch_queue_process(), make the next
// call of it use a timeout of 0 (this is useful behavior if you need to
// wakeup the main thread from another thread in a race free way).
void mp_dispatch_interrupt(struct mp_dispatch_queue *queue)
{
pthread_mutex_lock(&queue->lock);
queue->interrupted = true;
pthread_cond_broadcast(&queue->cond);
pthread_mutex_unlock(&queue->lock);
}
// If a mp_dispatch_queue_process() call is in progress, then adjust the maximum
// time it blocks due to its timeout argument. Otherwise does nothing. (It
// makes sense to call this in code that uses both mp_dispatch_[un]lock() and
// a normal event loop.)
// Does not work correctly with queues that have mp_dispatch_set_wakeup_fn()
// called on them, because this implies you actually do waiting via
// mp_dispatch_queue_process(), while wakeup callbacks are used when you need
// to wait in external APIs.
void mp_dispatch_adjust_timeout(struct mp_dispatch_queue *queue, int64_t until)
{
pthread_mutex_lock(&queue->lock);
if (queue->in_process && queue->wait > until) {
queue->wait = until;
pthread_cond_broadcast(&queue->cond);
}
pthread_mutex_unlock(&queue->lock);
}
// Grant exclusive access to the target thread's state. While this is active,
// no other thread can return from mp_dispatch_lock() (i.e. it behaves like
// a pthread mutex), and no other thread can get dispatch items completed.
// Other threads can still queue asynchronous dispatch items without waiting,
// and the mutex behavior applies to this function and dispatch callbacks only.
// The lock is non-recursive, and dispatch callback functions can be thought of
// already holding the dispatch lock.
void mp_dispatch_lock(struct mp_dispatch_queue *queue)
{
pthread_mutex_lock(&queue->lock);
// Must not be called recursively from dispatched callbacks.
if (queue->in_process)
assert(!pthread_equal(queue->in_process_thread, pthread_self()));
// Must not be called recursively at all.
if (queue->locked_explicit)
assert(!pthread_equal(queue->locked_explicit_thread, pthread_self()));
queue->lock_requests += 1;
// And now wait until the target thread gets "trapped" within the
// mp_dispatch_queue_process() call, which will mean we get exclusive
// access to the target's thread state.
while (!queue->in_process) {
pthread_mutex_unlock(&queue->lock);
if (queue->wakeup_fn)
queue->wakeup_fn(queue->wakeup_ctx);
pthread_mutex_lock(&queue->lock);
if (queue->in_process)
break;
pthread_cond_wait(&queue->cond, &queue->lock);
}
// Wait until we can get the lock.
while (!queue->in_process || queue->locked)
pthread_cond_wait(&queue->cond, &queue->lock);
// "Lock".
assert(queue->lock_requests);
assert(!queue->locked);
assert(!queue->locked_explicit);
queue->locked = true;
queue->locked_explicit = true;
queue->locked_explicit_thread = pthread_self();
pthread_mutex_unlock(&queue->lock);
}
// Undo mp_dispatch_lock().
void mp_dispatch_unlock(struct mp_dispatch_queue *queue)
{
pthread_mutex_lock(&queue->lock);
assert(queue->locked);
// Must be called after a mp_dispatch_lock(), from the same thread.
assert(queue->locked_explicit);
assert(pthread_equal(queue->locked_explicit_thread, pthread_self()));
// "Unlock".
queue->locked = false;
queue->locked_explicit = false;
queue->lock_requests -= 1;
// Wakeup mp_dispatch_queue_process(), and maybe other mp_dispatch_lock()s.
// (Would be nice to wake up only 1 other locker if lock_requests>0.)
pthread_cond_broadcast(&queue->cond);
pthread_mutex_unlock(&queue->lock);
}