This pushes all avahi-client code to a threaded mainloop from the PA
mainloop context. We need to do this because avahi-client makes blocking
D-Bus calls, and we don't want to block the mainloop for that long.
The only exception to this now that I don't see a workaround for is
during module unload time. However, this shouldn't be a huge problem
since in most cases, this will only happen at server shutdown time.
The bulk of the change is partitioning the data so that PA core objects
only (well, mostly) get accessed in the PA mainloop and Avahi calls
happen only in the Avahi threaded mainloop.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=58758
for example:
Profiles:
input:analog-stereo: Analog Stereo Input (sinks: 0, sources: 1, priority. 60)
output:analog-stereo: Analog Stereo Output (sinks: 1, sources: 0, priority. 6000)
it should be "priority: xxx", not "priority. xxx"
Signed-off-by: Peter Meerwald <p.meerwald@bct-electronic.com>
Checking the operation state caused a deadlock, because the state
won't change before my_drain_callback() returns, and it doesn't
return before my_drain_stream_func() calls
pa_threaded_mainloop_accept().
With BlueZ 5, if the remote device suspends the audio, the transport
state will change to "idle" and the endpoint is not required to release
the transport, since this could introduce race conditions. Therefore,
ignore the call to pa_bluetooth_transport_release() if the transport is
not acquired any more.
The new D-Bus API doesn't support access rights, which weren't used by
PulseAudio anyway, but it does solve a race condition: now optional
acquires can be implemented by bluetooth-util atomically using the D-Bus
TryAcquire() method.
BlueZ 5 exposes a 'State' property in the media transport interface.
With regard to PA, this replaces the profile-specific interfaces, since
they were being used to know if the audio was streaming or not.
Add the code to parse the properties of the media transport object when
a PropertiesChanged signal is received.
Note that the transport might have an owner other than BlueZ, and thus
the property changes would be emitted from arbitrary senders. For
performance reasons, the installed match considers the interface name
where the property has changed.
It could be possible to install and remove the D-Bus matches dynamically
when a new owner is registered/unregistered, but filtering based on the
interface name seems good enough already.
Install matches for signals ObjectManager.InterfacesAdded and
ObjectManager.InterfacesRemoved, and process the devices that are
registered and unregistered dynamically.
"pactl subscribe" is running continuously, and without flushing its output is
not usable for "process-on-arrival" per-line tasks, such as grepping. This
patch should fix this. For example, now:
pactl subscribe | grep 'server'
should print only server events as they arrive.
Parse the result of ObjectManager.GetManagedObjects(), which includes
all objects registered, their interfaces and the corresponding
properties per interface.
Any code that runs inside the init() callback sees an invalid module
index. Sometimes init() does things that cause hooks to be fired. This
means that any code that uses hooks may see an invalid module index.
Fix this by assigning the module index before init() is called.
There are no known issues in the upstream code base where an invalid
module index would be used, but an out-of-tree module
(module-murphy-ivi) had a problem with this.
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=63923
The 'Name' property of the Device interface became optional in BlueZ 5
and may not be present anymore (that happens when testing against the
PTS 4.7.0), so it's better not to expose it to clients so they don't
rely on its existence.
pa_write() knows two types of operation:
calling send() and calling write()
there is a flag (a pointer to an int) passed to pa_write()
which can remember which write type was successful
if the pointer is NULL or the int is 0, send() is tried first,
with a fallback to write() if send() resulted in ENOTSOCK
pa_fdsem_post() calls pa_write() with a NULL pointer;
unfortunately (at least with HAVE_SYS_EVENTFD_H #define'd) send()
always fails here and write() is called -- causing an extra syscall
quite frequently
strace:
send(17, "\1\0\0\0\0\0\0\0", 8, MSG_NOSIGNAL) = -1 ENOTSOCK (Socket operation on non-socket)
write(17, "\1\0\0\0\0\0\0\0", 8) = 8
the patch adds a write_type field to pa_fdsem to the successful
pa_write() type can be remembered and unnecessary send() calls are
avoided
Signed-off-by: Peter Meerwald <p.meerwald@bct-electronic.com>
Pressing Ctrl-C in a terminal while pasuspender is running
causes the sinks and sources to stay suspended after
pasuspender has exited, which is very annoying. This patch
fixes that problem, and also a similar problem with fork()
failures.
The old code accepted any word that started with "y", "Y",
"n", "N", "t", "T", "f" or "F". Fix this by having
a whitelist of full strings instead of checking just the
first letter.
the check for NEON so far only checked if -mfpu=neon is understood by the compiler,
however, this is not enough:
(i) #include <arm_neon.h> should be checked
(ii) -mfpu=neon must be passed before CFLAGS because eventually the per-library CFLAGS
for NEON code in src/Makefile.am are passed to the compiler before the global CFLAGS
in case the build environment passes CFLAGS to configure and we try to set conflicting
CFLAGS option, the former take precedence; CFLAGS cannot be overridden
this does not fix
http://lists.freedesktop.org/archives/pulseaudio-discuss/2012-December/015570.html
but at least makes the build fail in configure (and not while compiling stuff)
and gives better diagnostics
Signed-off-by: Peter Meerwald <p.meerwald@bct-electronic.com>
Port creation is now slightly different. It is now similar to how
other objects are created (e.g. sinks/sources/cards).
This should become more useful in the future when we move more stuff to
the ports.
Functionally nothing has changed.
This commit makes the code cleaner, avoiding unnecessary line breaks. It
also changes the debug message elements order, to make it look more
natural ("path, interface, member" instead of "interface, path,
member").
This means that the path names will always correspond to the
path configuration file names, so they will automatically be
unique (in the scope of one card).
Previously the path description was looked up based on the
path name only. Since there can be multiple paths that use
the same description, it had to be possible to have multiple
paths with the same name.
Having the same name with multiple paths makes identifying
the paths more complex than necessary, so the plan is to
make it impossible to have paths with the same name. This
patch prepares for that by retaining the possibility to
still have the same description with multiple paths. Instead
of the path name, the path description is looked up by using
the "path description key" if it is set (path name is still
used as a fallback lookup key).
As an extra, I broke try_to_switch_profile() into smaller
functions, because the two levels of loops with continue
statements inside both were a bit hard to follow.
get_latency_us() used an uninitialized variable, and an incorrect
scope for some of the AudioObjectGetPropertyData() calls. As a result,
audio would randomly not work at all.
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=65122
