wayland/src/wayland-client.c

1897 lines
50 KiB
C
Raw Normal View History

2008-12-02 15:15:01 -05:00
/*
* Copyright © 2008-2012 Kristian Høgsberg
* Copyright © 2010-2012 Intel Corporation
2008-12-02 15:15:01 -05:00
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
2008-12-02 15:15:01 -05:00
*
* The above copyright notice and this permission notice (including the
* next paragraph) shall be included in all copies or substantial
* portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
2008-12-02 15:15:01 -05:00
*/
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
#define _GNU_SOURCE
#include <stdlib.h>
#include <stdint.h>
#include <stddef.h>
#include <stdio.h>
#include <stdbool.h>
#include <errno.h>
#include <string.h>
#include <unistd.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <ctype.h>
#include <assert.h>
2011-04-11 09:24:11 -04:00
#include <fcntl.h>
#include <poll.h>
#include <pthread.h>
#include "wayland-util.h"
#include "wayland-os.h"
#include "wayland-client.h"
#include "wayland-private.h"
/** \cond */
enum wl_proxy_flag {
WL_PROXY_FLAG_ID_DELETED = (1 << 0),
WL_PROXY_FLAG_DESTROYED = (1 << 1)
};
struct wl_proxy {
struct wl_object object;
struct wl_display *display;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
struct wl_event_queue *queue;
uint32_t flags;
int refcount;
void *user_data;
wl_dispatcher_func_t dispatcher;
};
struct wl_global {
uint32_t id;
char *interface;
uint32_t version;
struct wl_list link;
};
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
struct wl_event_queue {
struct wl_list event_list;
struct wl_display *display;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
};
struct wl_display {
struct wl_proxy proxy;
struct wl_connection *connection;
/* errno of the last wl_display error */
int last_error;
/* When display gets an error event from some object, it stores
* information about it here, so that client can get this
* information afterwards */
struct {
/* Code of the error. It can be compared to
* the interface's errors enumeration. */
uint32_t code;
/* interface (protocol) in which the error occurred */
const struct wl_interface *interface;
/* id of the proxy that caused the error. There's no warranty
* that the proxy is still valid. It's up to client how it will
* use it */
uint32_t id;
} protocol_error;
int fd;
struct wl_map objects;
struct wl_event_queue display_queue;
struct wl_event_queue default_queue;
pthread_mutex_t mutex;
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
int reader_count;
uint32_t read_serial;
pthread_cond_t reader_cond;
};
/** \endcond */
static int debug_client = 0;
/**
* This helper function wakes up all threads that are
* waiting for display->reader_cond (i. e. when reading is done,
* canceled, or an error occurred)
*
* NOTE: must be called with display->mutex locked
*/
static void
display_wakeup_threads(struct wl_display *display)
{
/* Thread can get sleeping only in read_events(). If we're
* waking it up, it means that the read completed or was
* canceled, so we must increase the read_serial.
* This prevents from indefinite sleeping in read_events().
*/
++display->read_serial;
pthread_cond_broadcast(&display->reader_cond);
}
/**
* This function is called for local errors (no memory, server hung up)
*
* \param display
* \param error error value (EINVAL, EFAULT, ...)
*
* \note this function is called with display mutex locked
*/
static void
display_fatal_error(struct wl_display *display, int error)
{
if (display->last_error)
return;
if (!error)
error = EFAULT;
display->last_error = error;
display_wakeup_threads(display);
}
/**
* This function is called for error events
* and indicates that in some object an error occurred.
* Difference between this function and display_fatal_error()
* is that this one handles errors that will come by wire,
* whereas display_fatal_error() is called for local errors.
*
* \param display
* \param code error code
* \param id id of the object that generated the error
* \param intf protocol interface
*/
static void
display_protocol_error(struct wl_display *display, uint32_t code,
uint32_t id, const struct wl_interface *intf)
{
int err;
if (display->last_error)
return;
/* set correct errno */
if (wl_interface_equal(intf, &wl_display_interface)) {
switch (code) {
case WL_DISPLAY_ERROR_INVALID_OBJECT:
case WL_DISPLAY_ERROR_INVALID_METHOD:
err = EINVAL;
break;
case WL_DISPLAY_ERROR_NO_MEMORY:
err = ENOMEM;
break;
default:
err = EFAULT;
}
} else {
err = EPROTO;
}
pthread_mutex_lock(&display->mutex);
display->last_error = err;
display->protocol_error.code = code;
display->protocol_error.id = id;
display->protocol_error.interface = intf;
/*
* here it is not necessary to wake up threads like in
* display_fatal_error, because this function is called from
* an event handler and that means that read_events() is done
* and woke up all threads. Since wl_display_prepare_read()
* fails when there are events in the queue, no threads
* can sleep in read_events() during dispatching
* (and therefore during calling this function), so this is safe.
*/
pthread_mutex_unlock(&display->mutex);
}
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
static void
wl_event_queue_init(struct wl_event_queue *queue, struct wl_display *display)
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
{
wl_list_init(&queue->event_list);
queue->display = display;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
}
static void
decrease_closure_args_refcount(struct wl_closure *closure)
{
const char *signature;
struct argument_details arg;
int i, count;
struct wl_proxy *proxy;
signature = closure->message->signature;
count = arg_count_for_signature(signature);
for (i = 0; i < count; i++) {
signature = get_next_argument(signature, &arg);
switch (arg.type) {
case 'n':
case 'o':
proxy = (struct wl_proxy *) closure->args[i].o;
if (proxy) {
if (proxy->flags & WL_PROXY_FLAG_DESTROYED)
closure->args[i].o = NULL;
proxy->refcount--;
if (!proxy->refcount)
free(proxy);
}
break;
default:
break;
}
}
}
void
proxy_destroy(struct wl_proxy *proxy);
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
static void
wl_event_queue_release(struct wl_event_queue *queue)
{
struct wl_closure *closure;
struct wl_proxy *proxy;
bool proxy_destroyed;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
while (!wl_list_empty(&queue->event_list)) {
closure = container_of(queue->event_list.next,
struct wl_closure, link);
wl_list_remove(&closure->link);
decrease_closure_args_refcount(closure);
proxy = closure->proxy;
proxy_destroyed = !!(proxy->flags & WL_PROXY_FLAG_DESTROYED);
proxy->refcount--;
if (proxy_destroyed && !proxy->refcount)
free(proxy);
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
wl_closure_destroy(closure);
}
}
/** Destroy an event queue
*
* \param queue The event queue to be destroyed
*
* Destroy the given event queue. Any pending event on that queue is
* discarded.
*
* The \ref wl_display object used to create the queue should not be
* destroyed until all event queues created with it are destroyed with
* this function.
*
* \memberof wl_event_queue
*/
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
WL_EXPORT void
wl_event_queue_destroy(struct wl_event_queue *queue)
{
struct wl_display *display = queue->display;
pthread_mutex_lock(&display->mutex);
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
wl_event_queue_release(queue);
free(queue);
pthread_mutex_unlock(&display->mutex);
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
}
/** Create a new event queue for this display
*
* \param display The display context object
* \return A new event queue associated with this display or NULL on
* failure.
*
* \memberof wl_event_queue
*/
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
WL_EXPORT struct wl_event_queue *
wl_display_create_queue(struct wl_display *display)
{
struct wl_event_queue *queue;
queue = malloc(sizeof *queue);
if (queue == NULL)
return NULL;
wl_event_queue_init(queue, display);
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
return queue;
}
2013-11-14 21:29:06 -08:00
static struct wl_proxy *
proxy_create(struct wl_proxy *factory, const struct wl_interface *interface)
{
struct wl_proxy *proxy;
struct wl_display *display = factory->display;
proxy = malloc(sizeof *proxy);
if (proxy == NULL)
return NULL;
memset(proxy, 0, sizeof *proxy);
2013-11-14 21:29:06 -08:00
proxy->object.interface = interface;
proxy->display = display;
proxy->queue = factory->queue;
proxy->refcount = 1;
proxy->object.id = wl_map_insert_new(&display->objects, 0, proxy);
return proxy;
}
/** Create a proxy object with a given interface
*
* \param factory Factory proxy object
* \param interface Interface the proxy object should use
* \return A newly allocated proxy object or NULL on failure
*
* This function creates a new proxy object with the supplied interface. The
* proxy object will have an id assigned from the client id space. The id
* should be created on the compositor side by sending an appropriate request
* with \ref wl_proxy_marshal().
*
* The proxy will inherit the display and event queue of the factory object.
*
* \note This should not normally be used by non-generated code.
*
* \sa wl_display, wl_event_queue, wl_proxy_marshal()
*
* \memberof wl_proxy
*/
WL_EXPORT struct wl_proxy *
wl_proxy_create(struct wl_proxy *factory, const struct wl_interface *interface)
{
struct wl_display *display = factory->display;
2013-11-14 21:29:06 -08:00
struct wl_proxy *proxy;
pthread_mutex_lock(&display->mutex);
2013-11-14 21:29:06 -08:00
proxy = proxy_create(factory, interface);
pthread_mutex_unlock(&display->mutex);
return proxy;
}
/* The caller should hold the display lock */
static struct wl_proxy *
wl_proxy_create_for_id(struct wl_proxy *factory,
uint32_t id, const struct wl_interface *interface)
{
struct wl_proxy *proxy;
struct wl_display *display = factory->display;
proxy = malloc(sizeof *proxy);
if (proxy == NULL)
return NULL;
memset(proxy, 0, sizeof *proxy);
proxy->object.interface = interface;
proxy->object.id = id;
proxy->display = display;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
proxy->queue = factory->queue;
proxy->refcount = 1;
wl_map_insert_at(&display->objects, 0, id, proxy);
return proxy;
}
void
proxy_destroy(struct wl_proxy *proxy)
{
if (proxy->flags & WL_PROXY_FLAG_ID_DELETED)
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
wl_map_remove(&proxy->display->objects, proxy->object.id);
else if (proxy->object.id < WL_SERVER_ID_START)
wl_map_insert_at(&proxy->display->objects, 0,
proxy->object.id, WL_ZOMBIE_OBJECT);
else
wl_map_insert_at(&proxy->display->objects, 0,
proxy->object.id, NULL);
proxy->flags |= WL_PROXY_FLAG_DESTROYED;
proxy->refcount--;
if (!proxy->refcount)
free(proxy);
}
/** Destroy a proxy object
*
* \param proxy The proxy to be destroyed
*
* \memberof wl_proxy
*/
WL_EXPORT void
wl_proxy_destroy(struct wl_proxy *proxy)
{
struct wl_display *display = proxy->display;
pthread_mutex_lock(&display->mutex);
proxy_destroy(proxy);
pthread_mutex_unlock(&display->mutex);
}
/** Set a proxy's listener
*
* \param proxy The proxy object
* \param implementation The listener to be added to proxy
* \param data User data to be associated with the proxy
* \return 0 on success or -1 on failure
*
* Set proxy's listener to \c implementation and its user data to
* \c data. If a listener has already been set, this function
* fails and nothing is changed.
*
* \c implementation is a vector of function pointers. For an opcode
* \c n, \c implementation[n] should point to the handler of \c n for
* the given object.
*
* \memberof wl_proxy
*/
WL_EXPORT int
wl_proxy_add_listener(struct wl_proxy *proxy,
void (**implementation)(void), void *data)
{
if (proxy->object.implementation || proxy->dispatcher) {
wl_log("proxy %p already has listener\n", proxy);
return -1;
2011-02-18 15:28:54 -05:00
}
2011-02-18 15:28:54 -05:00
proxy->object.implementation = implementation;
proxy->user_data = data;
return 0;
}
/** Get a proxy's listener
*
* \param proxy The proxy object
* \return The address of the proxy's listener or NULL if no listener is set
*
* Gets the address to the proxy's listener; which is the listener set with
* \ref wl_proxy_add_listener.
*
* This function is useful in clients with multiple listeners on the same
* interface to allow the identification of which code to execute.
*
* \memberof wl_proxy
*/
WL_EXPORT const void *
wl_proxy_get_listener(struct wl_proxy *proxy)
{
return proxy->object.implementation;
}
/** Set a proxy's listener (with dispatcher)
*
* \param proxy The proxy object
* \param dispatcher The dispatcher to be used for this proxy
* \param implementation The dispatcher-specific listener implementation
* \param data User data to be associated with the proxy
* \return 0 on success or -1 on failure
*
* Set proxy's listener to use \c dispatcher_func as its dispatcher and \c
* dispatcher_data as its dispatcher-specific implementation and its user data
* to \c data. If a listener has already been set, this function
* fails and nothing is changed.
*
* The exact details of dispatcher_data depend on the dispatcher used. This
* function is intended to be used by language bindings, not user code.
*
* \memberof wl_proxy
*/
WL_EXPORT int
wl_proxy_add_dispatcher(struct wl_proxy *proxy,
wl_dispatcher_func_t dispatcher,
const void *implementation, void *data)
{
if (proxy->object.implementation || proxy->dispatcher) {
wl_log("proxy %p already has listener\n");
return -1;
}
proxy->object.implementation = implementation;
proxy->dispatcher = dispatcher;
proxy->user_data = data;
return 0;
}
2013-11-14 21:29:06 -08:00
static struct wl_proxy *
create_outgoing_proxy(struct wl_proxy *proxy, const struct wl_message *message,
union wl_argument *args,
const struct wl_interface *interface)
{
int i, count;
const char *signature;
struct argument_details arg;
struct wl_proxy *new_proxy = NULL;
signature = message->signature;
count = arg_count_for_signature(signature);
for (i = 0; i < count; i++) {
signature = get_next_argument(signature, &arg);
switch (arg.type) {
case 'n':
new_proxy = proxy_create(proxy, interface);
if (new_proxy == NULL)
return NULL;
args[i].o = &new_proxy->object;
break;
}
}
return new_proxy;
}
/** Prepare a request to be sent to the compositor
*
* \param proxy The proxy object
* \param opcode Opcode of the request to be sent
2013-11-14 21:29:06 -08:00
* \param args Extra arguments for the given request
* \param interface The interface to use for the new proxy
*
* Translates the request given by opcode and the extra arguments into the
2013-11-14 21:29:06 -08:00
* wire format and write it to the connection buffer. This version takes an
* array of the union type wl_argument.
*
* For new-id arguments, this function will allocate a new wl_proxy
* and send the ID to the server. The new wl_proxy will be returned
* on success or NULL on errror with errno set accordingly.
*
* \note This is intended to be used by language bindings and not in
* non-generated code.
*
* \sa wl_proxy_marshal()
*
* \memberof wl_proxy
*/
WL_EXPORT struct wl_proxy *
wl_proxy_marshal_array_constructor(struct wl_proxy *proxy,
uint32_t opcode, union wl_argument *args,
const struct wl_interface *interface)
{
struct wl_closure *closure;
struct wl_proxy *new_proxy = NULL;
const struct wl_message *message;
pthread_mutex_lock(&proxy->display->mutex);
message = &proxy->object.interface->methods[opcode];
if (interface) {
new_proxy = create_outgoing_proxy(proxy, message,
args, interface);
if (new_proxy == NULL)
goto err_unlock;
}
closure = wl_closure_marshal(&proxy->object, opcode, args, message);
if (closure == NULL) {
wl_log("Error marshalling request: %m\n");
2013-11-14 21:29:06 -08:00
abort();
}
if (debug_client)
2013-11-14 21:29:06 -08:00
wl_closure_print(closure, &proxy->object, true);
if (wl_closure_send(closure, proxy->display->connection)) {
wl_log("Error sending request: %m\n");
2013-11-14 21:29:06 -08:00
abort();
}
wl_closure_destroy(closure);
err_unlock:
pthread_mutex_unlock(&proxy->display->mutex);
return new_proxy;
}
/** Prepare a request to be sent to the compositor
*
2013-11-14 21:29:06 -08:00
* \param proxy The proxy object
* \param opcode Opcode of the request to be sent
* \param ... Extra arguments for the given request
*
2013-11-14 21:29:06 -08:00
* This function is similar to wl_proxy_marshal_constructor(), except
* it doesn't create proxies for new-id arguments.
*
* \note This should not normally be used by non-generated code.
*
* \sa wl_proxy_create()
*
* \memberof wl_proxy
*/
WL_EXPORT void
wl_proxy_marshal(struct wl_proxy *proxy, uint32_t opcode, ...)
{
union wl_argument args[WL_CLOSURE_MAX_ARGS];
va_list ap;
va_start(ap, opcode);
wl_argument_from_va_list(proxy->object.interface->methods[opcode].signature,
args, WL_CLOSURE_MAX_ARGS, ap);
va_end(ap);
2013-11-14 21:29:06 -08:00
wl_proxy_marshal_array_constructor(proxy, opcode, args, NULL);
}
/** Prepare a request to be sent to the compositor
*
* \param proxy The proxy object
* \param opcode Opcode of the request to be sent
* \param interface The interface to use for the new proxy
2013-11-14 21:29:06 -08:00
* \param ... Extra arguments for the given request
* \return A new wl_proxy for the new_id argument or NULL on error
*
* Translates the request given by opcode and the extra arguments into the
2013-11-14 21:29:06 -08:00
* wire format and write it to the connection buffer.
*
* For new-id arguments, this function will allocate a new wl_proxy
* and send the ID to the server. The new wl_proxy will be returned
* on success or NULL on errror with errno set accordingly.
*
* \note This should not normally be used by non-generated code.
*
* \memberof wl_proxy
*/
WL_EXPORT struct wl_proxy *
wl_proxy_marshal_constructor(struct wl_proxy *proxy, uint32_t opcode,
const struct wl_interface *interface, ...)
{
union wl_argument args[WL_CLOSURE_MAX_ARGS];
va_list ap;
va_start(ap, interface);
wl_argument_from_va_list(proxy->object.interface->methods[opcode].signature,
args, WL_CLOSURE_MAX_ARGS, ap);
va_end(ap);
return wl_proxy_marshal_array_constructor(proxy, opcode,
args, interface);
}
/** Prepare a request to be sent to the compositor
*
* \param proxy The proxy object
* \param opcode Opcode of the request to be sent
* \param args Extra arguments for the given request
*
* This function is similar to wl_proxy_marshal_array_constructor(), except
* it doesn't create proxies for new-id arguments.
*
* \note This is intended to be used by language bindings and not in
* non-generated code.
*
* \sa wl_proxy_marshal()
*
* \memberof wl_proxy
*/
WL_EXPORT void
wl_proxy_marshal_array(struct wl_proxy *proxy, uint32_t opcode,
union wl_argument *args)
{
2013-11-14 21:29:06 -08:00
wl_proxy_marshal_array_constructor(proxy, opcode, args, NULL);
}
static void
display_handle_error(void *data,
struct wl_display *display, void *object,
uint32_t code, const char *message)
{
struct wl_proxy *proxy = object;
wl_log("%s@%u: error %d: %s\n",
proxy->object.interface->name, proxy->object.id, code, message);
display_protocol_error(display, code, proxy->object.id,
proxy->object.interface);
}
static void
display_handle_delete_id(void *data, struct wl_display *display, uint32_t id)
{
struct wl_proxy *proxy;
pthread_mutex_lock(&display->mutex);
proxy = wl_map_lookup(&display->objects, id);
if (!proxy)
wl_log("error: received delete_id for unknown id (%u)\n", id);
if (proxy && proxy != WL_ZOMBIE_OBJECT)
proxy->flags |= WL_PROXY_FLAG_ID_DELETED;
else
wl_map_remove(&display->objects, id);
pthread_mutex_unlock(&display->mutex);
}
static const struct wl_display_listener display_listener = {
display_handle_error,
display_handle_delete_id
};
static int
connect_to_socket(const char *name)
{
struct sockaddr_un addr;
socklen_t size;
const char *runtime_dir;
int name_size, fd;
runtime_dir = getenv("XDG_RUNTIME_DIR");
if (!runtime_dir) {
wl_log("error: XDG_RUNTIME_DIR not set in the environment.\n");
/* to prevent programs reporting
* "failed to create display: Success" */
errno = ENOENT;
return -1;
}
if (name == NULL)
name = getenv("WAYLAND_DISPLAY");
client: require WAYLAND_DISPLAY to be set Although defaulting to wayland-0 seems convenient, it has an undesirable side effect: clients may unintentionally connect to the wrong compositor. Generally, it's safer to fail instead. Here's a real example: In Fedora 22, Gtk+ prefers Wayland over X11, though the default session is still a normal X11 Gnome session. When you launch a Gtk+ app, it will try Wayland, fail, then try X11, and succesfully start up. That works fine. Now suppose you launch Weston while running the Gnome session. Suddenly, all of the Gtk+ apps launched from Gnome will show up inside Weston instead. That's unexpected. There's also no good way to prevent that from happening (other than perhaps setting WAYLAND_DISPLAY to an invalid value when launching an app). Not using wayland-0 as the default will solve that problem: an app launched from the X11 Gnome session will use the X11 backend regardless of whether there's a wayland compositor running at the same time. Everything else should work as before. The compositor already sets the WAYLAND_DISPLAY when starting the session, so the lack of the default value should not make a difference to the user. Signed-off-by: Dima Ryazanov <dima@gmail.com> Acked-by: Pekka Paalanen <ppaalanen@gmail.com> Acked-by: Giulio Camuffo <giuliocamuffo@gmail.com> Acked-by: Daniel Stone <daniel@fooishbar.org> Acked-by: Jasper St. Pierre <jstpierre@mecheye.net> Reviewed-by: Ryo Munakata <ryomnktml@gmail.com> [Pekka: dropped the wayland-server.c hunk, adjusted summary] Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
2015-08-12 19:34:31 -07:00
if (name == NULL) {
wl_log("error: WAYLAND_DISPLAY not set in the environment.\n");
errno = ENOENT;
return -1;
}
fd = wl_os_socket_cloexec(PF_LOCAL, SOCK_STREAM, 0);
if (fd < 0)
return -1;
memset(&addr, 0, sizeof addr);
addr.sun_family = AF_LOCAL;
name_size =
snprintf(addr.sun_path, sizeof addr.sun_path,
"%s/%s", runtime_dir, name) + 1;
assert(name_size > 0);
if (name_size > (int)sizeof addr.sun_path) {
wl_log("error: socket path \"%s/%s\" plus null terminator"
" exceeds 108 bytes\n", runtime_dir, name);
close(fd);
/* to prevent programs reporting
* "failed to add socket: Success" */
errno = ENAMETOOLONG;
return -1;
};
size = offsetof (struct sockaddr_un, sun_path) + name_size;
if (connect(fd, (struct sockaddr *) &addr, size) < 0) {
close(fd);
return -1;
}
return fd;
}
/** Connect to Wayland display on an already open fd
*
* \param fd The fd to use for the connection
* \return A \ref wl_display object or \c NULL on failure
*
* The wl_display takes ownership of the fd and will close it when the
* display is destroyed. The fd will also be closed in case of
* failure.
*
* \memberof wl_display
*/
WL_EXPORT struct wl_display *
wl_display_connect_to_fd(int fd)
{
struct wl_display *display;
const char *debug;
debug = getenv("WAYLAND_DEBUG");
if (debug && (strstr(debug, "client") || strstr(debug, "1")))
debug_client = 1;
display = malloc(sizeof *display);
if (display == NULL) {
close(fd);
return NULL;
}
memset(display, 0, sizeof *display);
display->fd = fd;
wl_map_init(&display->objects, WL_MAP_CLIENT_SIDE);
wl_event_queue_init(&display->default_queue, display);
wl_event_queue_init(&display->display_queue, display);
pthread_mutex_init(&display->mutex, NULL);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
pthread_cond_init(&display->reader_cond, NULL);
display->reader_count = 0;
wl_map_insert_new(&display->objects, 0, NULL);
display->proxy.object.interface = &wl_display_interface;
display->proxy.object.id =
wl_map_insert_new(&display->objects, 0, display);
display->proxy.display = display;
display->proxy.object.implementation = (void(**)(void)) &display_listener;
2011-02-18 15:28:54 -05:00
display->proxy.user_data = display;
display->proxy.queue = &display->default_queue;
display->proxy.flags = 0;
display->proxy.refcount = 1;
display->connection = wl_connection_create(display->fd);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
if (display->connection == NULL)
goto err_connection;
return display;
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
err_connection:
pthread_mutex_destroy(&display->mutex);
pthread_cond_destroy(&display->reader_cond);
wl_map_release(&display->objects);
close(display->fd);
free(display);
return NULL;
}
/** Connect to a Wayland display
*
* \param name Name of the Wayland display to connect to
* \return A \ref wl_display object or \c NULL on failure
*
* Connect to the Wayland display named \c name. If \c name is \c NULL,
client: require WAYLAND_DISPLAY to be set Although defaulting to wayland-0 seems convenient, it has an undesirable side effect: clients may unintentionally connect to the wrong compositor. Generally, it's safer to fail instead. Here's a real example: In Fedora 22, Gtk+ prefers Wayland over X11, though the default session is still a normal X11 Gnome session. When you launch a Gtk+ app, it will try Wayland, fail, then try X11, and succesfully start up. That works fine. Now suppose you launch Weston while running the Gnome session. Suddenly, all of the Gtk+ apps launched from Gnome will show up inside Weston instead. That's unexpected. There's also no good way to prevent that from happening (other than perhaps setting WAYLAND_DISPLAY to an invalid value when launching an app). Not using wayland-0 as the default will solve that problem: an app launched from the X11 Gnome session will use the X11 backend regardless of whether there's a wayland compositor running at the same time. Everything else should work as before. The compositor already sets the WAYLAND_DISPLAY when starting the session, so the lack of the default value should not make a difference to the user. Signed-off-by: Dima Ryazanov <dima@gmail.com> Acked-by: Pekka Paalanen <ppaalanen@gmail.com> Acked-by: Giulio Camuffo <giuliocamuffo@gmail.com> Acked-by: Daniel Stone <daniel@fooishbar.org> Acked-by: Jasper St. Pierre <jstpierre@mecheye.net> Reviewed-by: Ryo Munakata <ryomnktml@gmail.com> [Pekka: dropped the wayland-server.c hunk, adjusted summary] Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
2015-08-12 19:34:31 -07:00
* its value will be replaced with the WAYLAND_DISPLAY environment variable.
*
* \memberof wl_display
*/
WL_EXPORT struct wl_display *
wl_display_connect(const char *name)
{
char *connection, *end;
int flags, fd;
connection = getenv("WAYLAND_SOCKET");
if (connection) {
int prev_errno = errno;
errno = 0;
fd = strtol(connection, &end, 0);
if (errno != 0 || connection == end || *end != '\0')
return NULL;
errno = prev_errno;
flags = fcntl(fd, F_GETFD);
if (flags != -1)
fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
unsetenv("WAYLAND_SOCKET");
} else {
fd = connect_to_socket(name);
if (fd < 0)
return NULL;
}
return wl_display_connect_to_fd(fd);
}
/** Close a connection to a Wayland display
*
* \param display The display context object
*
* Close the connection to \c display and free all resources associated
* with it.
*
* \memberof wl_display
*/
WL_EXPORT void
wl_display_disconnect(struct wl_display *display)
{
wl_connection_destroy(display->connection);
wl_map_release(&display->objects);
wl_event_queue_release(&display->default_queue);
wl_event_queue_release(&display->display_queue);
pthread_mutex_destroy(&display->mutex);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
pthread_cond_destroy(&display->reader_cond);
close(display->fd);
free(display);
}
/** Get a display context's file descriptor
*
* \param display The display context object
* \return Display object file descriptor
*
* Return the file descriptor associated with a display so it can be
* integrated into the client's main loop.
*
* \memberof wl_display
*/
WL_EXPORT int
wl_display_get_fd(struct wl_display *display)
{
return display->fd;
}
static void
sync_callback(void *data, struct wl_callback *callback, uint32_t serial)
{
2014-08-05 15:21:36 -04:00
int *done = data;
2014-08-05 15:21:36 -04:00
*done = 1;
wl_callback_destroy(callback);
}
static const struct wl_callback_listener sync_listener = {
sync_callback
};
/** Block until all pending request are processed by the server
*
* \param display The display context object
* \param queue The queue on which to run the roundtrip
* \return The number of dispatched events on success or -1 on failure
*
* Blocks until the server processes all currently issued requests and
* sends out pending events on the event queue.
*
* \note This function uses wl_display_dispatch_queue() internally. If you
* are using wl_display_read_events() from more threads, don't use this function
* (or make sure that calling wl_display_roundtrip_queue() doesn't interfere
* with calling wl_display_prepare_read() and wl_display_read_events())
*
* \sa wl_display_roundtrip()
* \memberof wl_event_queue
*/
WL_EXPORT int
wl_display_roundtrip_queue(struct wl_display *display, struct wl_event_queue *queue)
{
struct wl_callback *callback;
int done, ret = 0;
done = 0;
callback = wl_display_sync(display);
if (callback == NULL)
return -1;
wl_proxy_set_queue((struct wl_proxy *) callback, queue);
wl_callback_add_listener(callback, &sync_listener, &done);
while (!done && ret >= 0)
ret = wl_display_dispatch_queue(display, queue);
if (ret == -1 && !done)
wl_callback_destroy(callback);
return ret;
}
/** Block until all pending request are processed by the server
*
* \param display The display context object
* \return The number of dispatched events on success or -1 on failure
*
* Blocks until the server process all currently issued requests and
* sends out pending events on the default event queue.
*
* \note This function uses wl_display_dispatch_queue() internally. If you
* are using wl_display_read_events() from more threads, don't use this function
* (or make sure that calling wl_display_roundtrip() doesn't interfere
* with calling wl_display_prepare_read() and wl_display_read_events())
*
* \memberof wl_display
*/
WL_EXPORT int
wl_display_roundtrip(struct wl_display *display)
{
return wl_display_roundtrip_queue(display, &display->default_queue);
}
static int
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
create_proxies(struct wl_proxy *sender, struct wl_closure *closure)
{
struct wl_proxy *proxy;
const char *signature;
struct argument_details arg;
uint32_t id;
int i;
int count;
signature = closure->message->signature;
count = arg_count_for_signature(signature);
for (i = 0; i < count; i++) {
signature = get_next_argument(signature, &arg);
switch (arg.type) {
case 'n':
id = closure->args[i].n;
if (id == 0) {
closure->args[i].o = NULL;
break;
}
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
proxy = wl_proxy_create_for_id(sender, id,
closure->message->types[i]);
if (proxy == NULL)
return -1;
closure->args[i].o = (struct wl_object *)proxy;
break;
default:
break;
}
}
return 0;
}
static void
increase_closure_args_refcount(struct wl_closure *closure)
{
const char *signature;
struct argument_details arg;
int i, count;
struct wl_proxy *proxy;
signature = closure->message->signature;
count = arg_count_for_signature(signature);
for (i = 0; i < count; i++) {
signature = get_next_argument(signature, &arg);
switch (arg.type) {
case 'n':
case 'o':
proxy = (struct wl_proxy *) closure->args[i].o;
if (proxy)
proxy->refcount++;
break;
default:
break;
}
}
}
static int
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
queue_event(struct wl_display *display, int len)
{
uint32_t p[2], id;
int opcode, size;
struct wl_proxy *proxy;
struct wl_closure *closure;
2010-09-01 17:18:33 -04:00
const struct wl_message *message;
struct wl_event_queue *queue;
wl_connection_copy(display->connection, p, sizeof p);
id = p[0];
opcode = p[1] & 0xffff;
size = p[1] >> 16;
if (len < size)
return 0;
proxy = wl_map_lookup(&display->objects, id);
if (proxy == WL_ZOMBIE_OBJECT) {
wl_connection_consume(display->connection, size);
return size;
} else if (proxy == NULL) {
wl_connection_consume(display->connection, size);
return size;
}
message = &proxy->object.interface->events[opcode];
closure = wl_connection_demarshal(display->connection, size,
&display->objects, message);
if (!closure)
return -1;
if (create_proxies(proxy, closure) < 0) {
wl_closure_destroy(closure);
return -1;
}
if (wl_closure_lookup_objects(closure, &display->objects) != 0) {
wl_closure_destroy(closure);
return -1;
}
increase_closure_args_refcount(closure);
proxy->refcount++;
closure->proxy = proxy;
if (proxy == &display->proxy)
queue = &display->display_queue;
else
queue = proxy->queue;
wl_list_insert(queue->event_list.prev, &closure->link);
return size;
}
static void
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
dispatch_event(struct wl_display *display, struct wl_event_queue *queue)
{
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
struct wl_closure *closure;
struct wl_proxy *proxy;
int opcode;
bool proxy_destroyed;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
closure = container_of(queue->event_list.next,
struct wl_closure, link);
wl_list_remove(&closure->link);
opcode = closure->opcode;
/* Verify that the receiving object is still valid by checking if has
* been destroyed by the application. */
decrease_closure_args_refcount(closure);
proxy = closure->proxy;
proxy_destroyed = !!(proxy->flags & WL_PROXY_FLAG_DESTROYED);
proxy->refcount--;
if (proxy_destroyed) {
if (!proxy->refcount)
free(proxy);
wl_closure_destroy(closure);
return;
}
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
pthread_mutex_unlock(&display->mutex);
if (proxy->dispatcher) {
if (debug_client)
wl_closure_print(closure, &proxy->object, false);
wl_closure_dispatch(closure, proxy->dispatcher,
&proxy->object, opcode);
} else if (proxy->object.implementation) {
if (debug_client)
wl_closure_print(closure, &proxy->object, false);
wl_closure_invoke(closure, WL_CLOSURE_INVOKE_CLIENT,
&proxy->object, opcode, proxy->user_data);
}
wl_closure_destroy(closure);
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
pthread_mutex_lock(&display->mutex);
}
static int
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
read_events(struct wl_display *display)
{
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
int total, rem, size;
uint32_t serial;
display->reader_count--;
if (display->reader_count == 0) {
total = wl_connection_read(display->connection);
if (total == -1) {
if (errno == EAGAIN) {
/* we must wake up threads whenever
* the reader_count dropped to 0 */
display_wakeup_threads(display);
return 0;
}
display_fatal_error(display, errno);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
return -1;
} else if (total == 0) {
/* The compositor has closed the socket. This
* should be considered an error so we'll fake
* an errno */
errno = EPIPE;
display_fatal_error(display, errno);
return -1;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
}
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
for (rem = total; rem >= 8; rem -= size) {
size = queue_event(display, rem);
if (size == -1) {
display_fatal_error(display, errno);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
return -1;
} else if (size == 0) {
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
break;
}
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
}
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
display_wakeup_threads(display);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
} else {
serial = display->read_serial;
while (display->read_serial == serial)
pthread_cond_wait(&display->reader_cond,
&display->mutex);
if (display->last_error) {
errno = display->last_error;
return -1;
}
}
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
return 0;
}
static void
cancel_read(struct wl_display *display)
{
display->reader_count--;
if (display->reader_count == 0)
display_wakeup_threads(display);
}
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
/** Read events from display file descriptor
*
* \param display The display context object
* \return 0 on success or -1 on error. In case of error errno will
* be set accordingly
*
* This will read events from the file descriptor for the display.
* This function does not dispatch events, it only reads and queues
* events into their corresponding event queues. If no data is
* available on the file descriptor, wl_display_read_events() returns
* immediately. To dispatch events that may have been queued, call
* wl_display_dispatch_pending() or wl_display_dispatch_queue_pending().
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* Before calling this function, wl_display_prepare_read() must be
* called first. When running in more threads (which is the usual
* case, since we'd use wl_display_dispatch() otherwise), every thread
* must call wl_display_prepare_read() before calling this function.
*
* After calling wl_display_prepare_read() there can be some extra code
* before calling wl_display_read_events(), for example poll() or alike.
* Example of code in a thread:
*
* \code
*
* while (wl_display_prepare_read(display) < 0)
* wl_display_dispatch_pending(display);
* wl_display_flush(display);
*
* ... some code ...
*
* fds[0].fd = wl_display_get_fd(display);
* fds[0].events = POLLIN;
* poll(fds, 1, -1);
*
* if (!everything_ok()) {
* wl_display_cancel_read(display);
* handle_error();
* }
*
* if (wl_display_read_events(display) < 0)
* handle_error();
*
* ...
* \endcode
*
* After wl_display_prepare_read() succeeds, other threads that enter
* wl_display_read_events() will sleep until the very last thread enters
* it too or cancels. Therefore when the display fd becomes (or already
* is) readable, wl_display_read_events() should be called as soon as
* possible to unblock all threads. If wl_display_read_events() will not
* be called, then wl_display_cancel_read() must be called instead to let
* the other threads continue.
*
* This function must not be called simultaneously with wl_display_dispatch().
* It may lead to deadlock. If programmer wants, for some reason, use
* wl_display_dispatch() in one thread and wl_display_prepare_read() with
* wl_display_read_events() in another, extra care must be taken to serialize
* these calls, i. e. use mutexes or similar (on whole prepare + read sequence)
*
* \sa wl_display_prepare_read(), wl_display_cancel_read(),
* wl_display_dispatch_pending(), wl_display_dispatch()
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* \memberof wl_display
*/
WL_EXPORT int
wl_display_read_events(struct wl_display *display)
{
int ret;
pthread_mutex_lock(&display->mutex);
if (display->last_error) {
cancel_read(display);
pthread_mutex_unlock(&display->mutex);
errno = display->last_error;
return -1;
}
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
ret = read_events(display);
pthread_mutex_unlock(&display->mutex);
return ret;
}
static int
dispatch_queue(struct wl_display *display, struct wl_event_queue *queue)
{
int count;
if (display->last_error)
goto err;
count = 0;
while (!wl_list_empty(&display->display_queue.event_list)) {
dispatch_event(display, &display->display_queue);
if (display->last_error)
goto err;
count++;
}
while (!wl_list_empty(&queue->event_list)) {
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
dispatch_event(display, queue);
if (display->last_error)
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
goto err;
count++;
}
return count;
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
err:
errno = display->last_error;
return -1;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
}
/** Prepare to read events from the display to this queue
*
* \param display The display context object
* \param queue The event queue to use
* \return 0 on success or -1 if event queue was not empty
*
* Atomically makes sure the queue is empty and stops any other thread
* from placing events into this (or any) queue. Caller must
* eventually call either wl_display_cancel_read() or
* wl_display_read_events(), usually after waiting for the
* display fd to become ready for reading, to release the lock.
*
* \sa wl_display_prepare_read
* \memberof wl_event_queue
*/
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
WL_EXPORT int
wl_display_prepare_read_queue(struct wl_display *display,
struct wl_event_queue *queue)
{
int ret;
pthread_mutex_lock(&display->mutex);
if (!wl_list_empty(&queue->event_list)) {
errno = EAGAIN;
ret = -1;
} else {
display->reader_count++;
ret = 0;
}
pthread_mutex_unlock(&display->mutex);
return ret;
}
/** Prepare to read events from the display's file descriptor
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* \param display The display context object
* \return 0 on success or -1 if event queue was not empty
*
* This function must be called before reading from the file
* descriptor using wl_display_read_events(). Calling
* wl_display_prepare_read() announces the calling thread's intention
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
* to read and ensures that until the thread is ready to read and
* calls wl_display_read_events(), no other thread will read from the
* file descriptor. This only succeeds if the event queue is empty
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
* though, and if there are undispatched events in the queue, -1 is
* returned and errno set to EAGAIN.
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* If a thread successfully calls wl_display_prepare_read(), it must
* either call wl_display_read_events() when it's ready or cancel the
* read intention by calling wl_display_cancel_read().
*
* Use this function before polling on the display fd or to integrate
* the fd into a toolkit event loop in a race-free way.
* A correct usage would be (we left out most of error checking):
*
* \code
* while (wl_display_prepare_read(display) != 0)
* wl_display_dispatch_pending(display);
* wl_display_flush(display);
*
* ret = poll(fds, nfds, -1);
* if (has_error(ret))
* wl_display_cancel_read(display);
* else
* wl_display_read_events(display);
*
* wl_display_dispatch_pending(display);
* \endcode
*
* Here we call wl_display_prepare_read(), which ensures that between
* returning from that call and eventually calling
* wl_display_read_events(), no other thread will read from the fd and
* queue events in our queue. If the call to wl_display_prepare_read() fails,
* we dispatch the pending events and try again until we're successful.
*
* When using wl_display_dispatch() we'd have something like:
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* \code
* wl_display_dispatch_pending(display);
* wl_display_flush(display);
* poll(fds, nfds, -1);
* wl_display_dispatch(display);
* \endcode
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* This sequence in not thread-safe. The race is immediately after poll(),
* where one thread could preempt and read events before the other thread calls
* wl_display_dispatch(). This call now blocks and starves the other
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
* fds in the event loop.
*
* Another race would be when using more event queues.
* When one thread calls wl_display_dispatch(_queue)(), then it
* reads all events from display's fd and queues them in appropriate
* queues. Then it dispatches only its own queue and the other events
* are sitting in their queues, waiting for dispatching. If that happens
* before the other thread managed to call poll(), it will
* block with events queued.
*
* wl_display_prepare_read() function doesn't acquire exclusive access
* to the display's fd. It only registers that the thread calling this function
* has intention to read from fd.
* When all registered readers call wl_display_read_events(),
* only one (at random) eventually reads and queues the events and the
* others are sleeping meanwhile. This way we avoid races and still
* can read from more threads.
*
* If the relevant queue is not the default queue, then
* wl_display_prepare_read_queue() and wl_display_dispatch_queue_pending()
* need to be used instead.
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* \sa wl_display_cancel_read(), wl_display_read_events()
*
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
* \memberof wl_display
*/
WL_EXPORT int
wl_display_prepare_read(struct wl_display *display)
{
return wl_display_prepare_read_queue(display, &display->default_queue);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
}
/** Cancel read intention on display's fd
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* \param display The display context object
*
* After a thread successfully called wl_display_prepare_read() it must
* either call wl_display_read_events() or wl_display_cancel_read().
* If the threads do not follow this rule it will lead to deadlock.
*
* \sa wl_display_prepare_read(), wl_display_read_events()
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
*
* \memberof wl_display
*/
WL_EXPORT void
wl_display_cancel_read(struct wl_display *display)
{
pthread_mutex_lock(&display->mutex);
cancel_read(display);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
pthread_mutex_unlock(&display->mutex);
}
/** Dispatch events in an event queue
*
* \param display The display context object
* \param queue The event queue to dispatch
* \return The number of dispatched events on success or -1 on failure
*
* Dispatch all incoming events for objects assigned to the given
* event queue. On failure -1 is returned and errno set appropriately.
*
* The behaviour of this function is exactly the same as the behaviour of
* wl_display_dispatch(), but it dispatches events on given queue,
* not on the default queue.
*
* This function blocks if there are no events to dispatch (if there are,
* it only dispatches these events and returns immediately).
* When this function returns after blocking, it means that it read events
* from display's fd and queued them to appropriate queues.
* If among the incoming events were some events assigned to the given queue,
* they are dispatched by this moment.
*
* \note Since Wayland 1.5 the display has an extra queue
* for its own events (i. e. delete_id). This queue is dispatched always,
* no matter what queue we passed as an argument to this function.
* That means that this function can return non-0 value even when it
* haven't dispatched any event for the given queue.
*
* This function has the same constrains for using in multi-threaded apps
* as \ref wl_display_dispatch().
*
* \sa wl_display_dispatch(), wl_display_dispatch_pending(),
* wl_display_dispatch_queue_pending()
*
* \memberof wl_event_queue
*/
WL_EXPORT int
wl_display_dispatch_queue(struct wl_display *display,
struct wl_event_queue *queue)
{
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
struct pollfd pfd[2];
int ret;
pthread_mutex_lock(&display->mutex);
ret = dispatch_queue(display, queue);
if (ret == -1)
goto err_unlock;
if (ret > 0) {
pthread_mutex_unlock(&display->mutex);
return ret;
}
/* We ignore EPIPE here, so that we try to read events before
* returning an error. When the compositor sends an error it
* will close the socket, and if we bail out here we don't get
* a chance to process the error. */
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
ret = wl_connection_flush(display->connection);
if (ret < 0 && errno != EAGAIN && errno != EPIPE) {
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
display_fatal_error(display, errno);
goto err_unlock;
}
display->reader_count++;
pthread_mutex_unlock(&display->mutex);
pfd[0].fd = display->fd;
pfd[0].events = POLLIN;
do {
ret = poll(pfd, 1, -1);
} while (ret == -1 && errno == EINTR);
if (ret == -1) {
wl_display_cancel_read(display);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
return -1;
}
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
pthread_mutex_lock(&display->mutex);
if (read_events(display) == -1)
goto err_unlock;
ret = dispatch_queue(display, queue);
if (ret == -1)
goto err_unlock;
pthread_mutex_unlock(&display->mutex);
return ret;
err_unlock:
pthread_mutex_unlock(&display->mutex);
return -1;
}
/** Dispatch pending events in an event queue
*
* \param display The display context object
* \param queue The event queue to dispatch
* \return The number of dispatched events on success or -1 on failure
*
* Dispatch all incoming events for objects assigned to the given
* event queue. On failure -1 is returned and errno set appropriately.
* If there are no events queued, this function returns immediately.
*
* \memberof wl_event_queue
* \since 1.0.2
*/
WL_EXPORT int
wl_display_dispatch_queue_pending(struct wl_display *display,
struct wl_event_queue *queue)
{
int ret;
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
pthread_mutex_lock(&display->mutex);
ret = dispatch_queue(display, queue);
client: Add wl_display_prepare_read() API to relax thread model assumptions The current thread model assumes that the application or toolkit will have one thread that either polls the display fd and dispatches events or just dispatches in a loop. Only this main thread will read from the fd while all other threads will block on a pthread condition and expect the main thread to deliver events to them. This turns out to be too restrictive. We can't assume that there always will be a thread like that. Qt QML threaded rendering will block the main thread on a condition that's signaled by a rendering thread after it finishes rendering. This leads to a deadlock when the rendering threads blocks in eglSwapBuffers(), and the main thread is waiting on the condition. Another problematic use case is with games that has a rendering thread for a splash screen while the main thread is busy loading game data or compiling shaders. The main thread isn't responsive and ends up blocking eglSwapBuffers() in the rendering thread. We also can't assume that there will be only one thread polling on the file descriptor. A valid use case is a thread receiving data from a custom wayland interface as well as a device fd or network socket. The thread may want to wait on either events from the wayland interface or data from the fd, in which case it needs to poll on both the wayland display fd and the device/network fd. The solution seems pretty straightforward: just let all threads read from the fd. However, the main-thread restriction was introduced to avoid a race. Simplified, main loops will do something like this: wl_display_dispatch_pending(display); /* Race here if other thread reads from fd and places events * in main eent queue. We go to sleep in poll while sitting on * events that may stall the application if not dispatched. */ poll(fds, nfds, -1); /* Race here if other thread reads and doesn't queue any * events for main queue. wl_display_dispatch() below will block * trying to read from the fd, while other fds in the mainloop * are ignored. */ wl_display_dispatch(display); The restriction that only the main thread can read from the fd avoids these races, but has the problems described above. This patch introduces new API to solve both problems. We add int wl_display_prepare_read(struct wl_display *display); and int wl_display_read_events(struct wl_display *display); wl_display_prepare_read() registers the calling thread as a potential reader of events. Once data is available on the fd, all reader threads must call wl_display_read_events(), at which point one of the threads will read from the fd and distribute the events to event queues. When that is done, all threads return from wl_display_read_events(). From the point of view of a single thread, this ensures that between calling wl_display_prepare_read() and wl_display_read_events(), no other thread will read from the fd and queue events in its event queue. This avoids the race conditions described above, and we avoid relying on any one thread to be available to read events.
2013-03-17 14:21:48 -04:00
pthread_mutex_unlock(&display->mutex);
return ret;
}
/** Process incoming events
*
* \param display The display context object
* \return The number of dispatched events on success or -1 on failure
*
* Dispatch the display's default event queue.
*
* If the default event queue is empty, this function blocks until there are
* events to be read from the display fd. Events are read and queued on
* the appropriate event queues. Finally, events on the default event queue
* are dispatched.
*
* In multi-threaded environment, programmer may want to use
* wl_display_read_events(). However, use of wl_display_read_events()
* must not be mixed with wl_display_dispatch(). See wl_display_read_events()
* and wl_display_prepare_read() for more details.
*
* \note It is not possible to check if there are events on the queue
* or not. For dispatching default queue events without blocking, see \ref
* wl_display_dispatch_pending().
*
* \sa wl_display_dispatch_pending(), wl_display_dispatch_queue(),
* wl_display_read_events()
*
* \memberof wl_display
*/
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
WL_EXPORT int
wl_display_dispatch(struct wl_display *display)
{
return wl_display_dispatch_queue(display, &display->default_queue);
}
/** Dispatch default queue events without reading from the display fd
*
* \param display The display context object
* \return The number of dispatched events or -1 on failure
*
* This function dispatches events on the main event queue. It does not
* attempt to read the display fd and simply returns zero if the main
* queue is empty, i.e., it doesn't block.
*
* This is necessary when a client's main loop wakes up on some fd other
* than the display fd (network socket, timer fd, etc) and calls \ref
* wl_display_dispatch_queue() from that callback. This may queue up
* events in other queues while reading all data from the display fd.
* When the main loop returns from the handler, the display fd
* no longer has data, causing a call to \em poll(2) (or similar
* functions) to block indefinitely, even though there are events ready
* to dispatch.
*
* To proper integrate the wayland display fd into a main loop, the
* client should always call wl_display_dispatch_pending() and then
* \ref wl_display_flush() prior to going back to sleep. At that point,
* the fd typically doesn't have data so attempting I/O could block, but
* events queued up on the default queue should be dispatched.
*
* A real-world example is a main loop that wakes up on a timerfd (or a
* sound card fd becoming writable, for example in a video player), which
* then triggers GL rendering and eventually eglSwapBuffers().
* eglSwapBuffers() may call wl_display_dispatch_queue() if it didn't
* receive the frame event for the previous frame, and as such queue
* events in the default queue.
*
* \sa wl_display_dispatch(), wl_display_dispatch_queue(),
* wl_display_flush()
*
* \memberof wl_display
*/
WL_EXPORT int
wl_display_dispatch_pending(struct wl_display *display)
{
return wl_display_dispatch_queue_pending(display,
&display->default_queue);
}
/** Retrieve the last error that occurred on a display
*
* \param display The display context object
* \return The last error that occurred on \c display or 0 if no error occurred
*
* Return the last error that occurred on the display. This may be an error sent
* by the server or caused by the local client.
*
* \note Errors are \b fatal. If this function returns non-zero the display
* can no longer be used.
*
* \memberof wl_display
*/
WL_EXPORT int
wl_display_get_error(struct wl_display *display)
{
int ret;
pthread_mutex_lock(&display->mutex);
ret = display->last_error;
pthread_mutex_unlock(&display->mutex);
return ret;
}
/** Retrieves the information about a protocol error:
*
* \param display The Wayland display
* \param interface if not NULL, stores the interface where the error occurred
* \param id if not NULL, stores the object id that generated
* the error. There's no guarantee the object is
* still valid; the client must know if it deleted the object.
* \return The error code as defined in the interface specification.
*
* \code
* int err = wl_display_get_error(display);
*
* if (err == EPROTO) {
* code = wl_display_get_protocol_error(display, &interface, &id);
* handle_error(code, interface, id);
* }
*
* ...
* \endcode
* \memberof wl_display
*/
WL_EXPORT uint32_t
wl_display_get_protocol_error(struct wl_display *display,
const struct wl_interface **interface,
uint32_t *id)
{
uint32_t ret;
pthread_mutex_lock(&display->mutex);
ret = display->protocol_error.code;
if (interface)
*interface = display->protocol_error.interface;
if (id)
*id = display->protocol_error.id;
pthread_mutex_unlock(&display->mutex);
return ret;
}
/** Send all buffered requests on the display to the server
*
* \param display The display context object
* \return The number of bytes sent on success or -1 on failure
*
* Send all buffered data on the client side to the server. Clients
* should call this function before blocking. On success, the number
* of bytes sent to the server is returned. On failure, this
* function returns -1 and errno is set appropriately.
*
* wl_display_flush() never blocks. It will write as much data as
* possible, but if all data could not be written, errno will be set
* to EAGAIN and -1 returned. In that case, use poll on the display
* file descriptor to wait for it to become writable again.
*
* \memberof wl_display
*/
WL_EXPORT int
wl_display_flush(struct wl_display *display)
{
int ret;
pthread_mutex_lock(&display->mutex);
if (display->last_error) {
errno = display->last_error;
ret = -1;
} else {
ret = wl_connection_flush(display->connection);
if (ret < 0 && errno != EAGAIN)
display_fatal_error(display, errno);
}
pthread_mutex_unlock(&display->mutex);
return ret;
}
/** Set the user data associated with a proxy
*
* \param proxy The proxy object
* \param user_data The data to be associated with proxy
*
* Set the user data associated with \c proxy. When events for this
* proxy are received, \c user_data will be supplied to its listener.
*
* \memberof wl_proxy
*/
WL_EXPORT void
wl_proxy_set_user_data(struct wl_proxy *proxy, void *user_data)
{
proxy->user_data = user_data;
}
/** Get the user data associated with a proxy
*
* \param proxy The proxy object
* \return The user data associated with proxy
*
* \memberof wl_proxy
*/
WL_EXPORT void *
wl_proxy_get_user_data(struct wl_proxy *proxy)
{
return proxy->user_data;
}
2012-04-27 11:31:07 -04:00
/** Get the id of a proxy object
*
* \param proxy The proxy object
* \return The id the object associated with the proxy
*
* \memberof wl_proxy
*/
2012-04-27 11:31:07 -04:00
WL_EXPORT uint32_t
wl_proxy_get_id(struct wl_proxy *proxy)
{
return proxy->object.id;
}
/** Get the interface name (class) of a proxy object
*
* \param proxy The proxy object
* \return The interface name of the object associated with the proxy
*
* \memberof wl_proxy
*/
WL_EXPORT const char *
wl_proxy_get_class(struct wl_proxy *proxy)
{
return proxy->object.interface->name;
}
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
/** Assign a proxy to an event queue
*
* \param proxy The proxy object
* \param queue The event queue that will handle this proxy or NULL
*
* Assign proxy to event queue. Events coming from \c proxy will be
* queued in \c queue from now. If queue is NULL, then the display's
* default queue is set to the proxy.
*
* \note By default, the queue set in proxy is the one inherited from parent.
*
* \sa wl_display_dispatch_queue()
*
* \memberof wl_proxy
*/
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
WL_EXPORT void
wl_proxy_set_queue(struct wl_proxy *proxy, struct wl_event_queue *queue)
{
if (queue)
proxy->queue = queue;
else
proxy->queue = &proxy->display->default_queue;
client: Add wl_event_queue for multi-thread dispatching This introduces wl_event_queue, which is what will make multi-threaded wayland clients possible and useful. The driving use case is that of a GL rendering thread that renders and calls eglSwapBuffer independently of a "main thread" that owns the wl_display and handles input events and everything else. In general, the EGL and GL APIs have a threading model that requires the wayland client library to be usable from several threads. Finally, the current callback model gets into trouble even in a single threaded scenario: if we have to block in eglSwapBuffers, we may end up doing unrelated callbacks from within EGL. The wl_event_queue mechanism lets the application (or middleware such as EGL or toolkits) assign a proxy to an event queue. Only events from objects associated with the queue will be put in the queue, and conversely, events from objects associated with the queue will not be queue up anywhere else. The wl_display struct has a built-in event queue, which is considered the main and default event queue. New proxies are associated with the same queue as the object that created them (either the object that a request with a new-id argument was sent to or the object that sent an event with a new-id argument). A proxy can be moved to a different event queue by calling wl_proxy_set_queue(). A subsystem, such as EGL, will then create its own event queue and associate the objects it expects to receive events from with that queue. If EGL needs to block and wait for a certain event, it can keep dispatching event from its queue until that events comes in. This wont call out to unrelated code with an EGL lock held. Similarly, we don't risk the main thread handling an event from an EGL object and then calling into EGL from a different thread without the lock held.
2012-10-05 13:49:48 -04:00
}
WL_EXPORT void
wl_log_set_handler_client(wl_log_func_t handler)
{
wl_log_handler = handler;
}