platform/upstream/pulseaudio.git
7 years agoremove module-xenpv-sink
Tanu Kaskinen [Mon, 5 Dec 2016 20:20:42 +0000 (22:20 +0200)]
remove module-xenpv-sink

The module doesn't build any more[1], and when starting to investigate
the build failure, I asked the module author if he'd know something
about the breakage. He said that he didn't know about backward
compatibility problems with libxen, but more importantly, he said that
the module probably doesn't have any users[2]. It doesn't make sense to
keep maintaining a module that doesn't have users, so let's drop it.

[1] https://bugs.freedesktop.org/show_bug.cgi?id=98793
[2] https://lists.freedesktop.org/archives/pulseaudio-discuss/2016-November/027172.html

7 years agobluetooth: fix race condition in BlueZ5 device disconnection
Juha Kuikka [Sat, 16 Apr 2016 00:37:22 +0000 (17:37 -0700)]
bluetooth: fix race condition in BlueZ5 device disconnection

SW: Pulseaudio 8.0 / BlueZ 5.39

Symptoms:
While disconnecting/reconnecting a paired bluetooth headset (LG HBS750)
audio fails roughly on every other connection.

On a failed connection "pactl list cards" shows the bluetooth device's
card but "Active Profile: off". Issuing "pacmd set-card-profile X
a2dp_sink" makes audio work immediately.

I realized that when this happened, the previous disconnection did not
remove the card, instead it was only configured for "Active Profile:
off" but otherwise left in place.

Upon looking at PA debug logs I saw that the transport for the a2dp_sink
was first set into disconnected state and then into idle state. In
"device_connection_changed_cb()" this causes the
"pa_bluetooth_device_any_transport_connected()" return true and the
module-bluez5-device is not unloaded.

Further investigation shows that this is caused by a race of
module-bluez5-device.c:thread_func() and
MediaPoint1::ClearConfiguration().

When the FD in thread_func() is closed (POLLHUP) an
BLUETOOTH_MESSAGE_STREAM_FD_HUP message is sent into the main thread.
The handler of this message unconditionally sets the transport into IDLE
state. This is a problem if it has already been set into DISCONNECTED
state.

7 years agoi18n: update Polish translation
Piotr Drąg [Sun, 27 Nov 2016 15:51:49 +0000 (16:51 +0100)]
i18n: update Polish translation

https://bugs.freedesktop.org/show_bug.cgi?id=98872

7 years agoi18n: add Galician translation
Fran Dieguez [Sun, 27 Nov 2016 15:29:16 +0000 (17:29 +0200)]
i18n: add Galician translation

7 years agostream: Frame-align divided audio segments
Ahmed S. Darwish [Tue, 22 Nov 2016 20:18:56 +0000 (22:18 +0200)]
stream: Frame-align divided audio segments

Executing below command will not produce any audio:

  pacat  --channels=3 /dev/urandom

Turns out that pa_stream_write() breaks large audio buffers into
segments of the maximum memblock size available -- a value which
is not necessarily frame aligned.

Meanwhile the server discards any non-aligned client audio, as a
security measure, due to some earlier reported daemon crashes.
Thus divide sent audio to the expected aligned form.

CommitReference-1: 22827a5e1e62
CommitReference-2: 150ace90f380
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=98475
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=77595
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
7 years agopacat: Synchronize STDIN and "write stream ready" events
Ahmed S. Darwish [Tue, 22 Nov 2016 20:16:10 +0000 (22:16 +0200)]
pacat: Synchronize STDIN and "write stream ready" events

Users reported pacat crashes when playing certain multi-channel
audio. For example:

  pacat --channels=2 /dev/zero    works
  pacat --channels=3 /dev/zero    pa_stream_write() failed: EINVAL
  pacat --channels=4 /dev/zero    works
  pacat --channels=5 /dev/zero    pa_stream_write() failed: EINVAL

pacat audio playback is pipe-like, from STDIN to PA write stream.
Meanwhile STDIN "ready to read" events got regularly triggered
before the write stream was even created, or at moments where the
stream could not accept any more audio.

In these out-of-sync cases, the write stream could not report the
appropriate buffer lengths it accepts, thus a default of 4K bytes
was chosen -- compatible by luck with some channel counts and
incompatible with others.

Instead of choosing a faulty default in these scenarios, mute the
the STDIN events until the write stream is available & queriable.

BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=98475
Reported-by: Tanu Kaskinen <tanuk@iki.fi>
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
7 years agoprotocol-native: Don't signal memfd support for 9.0 clients
Ahmed S. Darwish [Fri, 18 Nov 2016 23:54:10 +0000 (01:54 +0200)]
protocol-native: Don't signal memfd support for 9.0 clients

Although such 9.0 clients support memfd transport, they have an
iochannel bug that would break memfd audio if they're run in 32
bit mode over a 64-bit kernel. Influence them to use the POSIX
shared memory model instead.

Also bump the protocol version to exclusively mark such v9.0
libraries. Check commit 451d1d676237c81 for further details.

BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=97769
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
7 years agoiochannel: Strictly specify PF_UNIX ancillary data boundaries
Ahmed S. Darwish [Tue, 15 Nov 2016 16:48:46 +0000 (16:48 +0000)]
iochannel: Strictly specify PF_UNIX ancillary data boundaries

Users reported audio breakage for 32-bit pulse clients connected
to a 64-bit server over memfds. Investigating the issue further,
the problem is twofold:

1. iochannel's file-descriptor passing code is liberal in what it
   issues: produced ancillary data object's "data" section exceeds
   length field. How such an extra space is handled is a grey area
   in the POSIX.1g spec, the IETF RFC #2292 "Advanced Sockets API
   for IPv6" memo, and the cmsg(3) manpage.

2. A 64-bit kernel handling of such extra space differs by whether
   the app is 64-bit or 32-bit. For 64-bit apps, the kernel
   smartly ducks the issue. For 32-bit apps, an -EINVAL is
   directly returned; that's due to a kernel CMSG header traversal
   bug in the networking stack "32-bit sockets emulation layer".

   Compare Linux Kernel's socket.h cmsg_nxthdr() code and the
   32-bit emulation layer version of it at net/compat.c
   cmsg_compat_nxthdr() for further info. Notice how the former
   graciously ignores incomplete CMSGs while the latter _directly_
   complains about them -- as of kernel version 4.9-rc5.

   (A kernel patch is to be submitted)

Details:

iochannel typically uses sendmsg() for passing FDs & credentials.
>From RFC 2292, sendmsg() control data is just a heterogeneous
array of embedded ancillary objects that can differ in length.
Linguistically, a "control message" is an ancillary data object.

For example, below is a sendmsg() "msg_control" containing two
ancillary objects:

|<---------------------- msg_controllen---------------------->|
|                                                             |
|<--- ancillary data object -->|<----- ancillary data object->|
|<------- CMSG_SPACE() ------->|<------- CMSG_SPACE() ------->|
|                              |                              |
|<-------- cmsg_len ------->|  |<-------- cmsg_len ------->|  |
|<------- CMSG_LEN() ------>|  |<------- CMSG_LEN() ------>|  |
|                           |  |                           |  |
+-----+-----+-----+--+------+--+-----+-----+-----+--+------+--+
|cmsg_|cmsg_|cmsg_|XX|cmsg_ |XX|cmsg_|cmsg_|cmsg_|XX|cmsg_ |XX|
|len  |level|type |XX|data[]|XX|len  |level|type |XX|data[]|XX|
+-----+-----+-----+--+------+--+-----+-----+-----+--+----+-+--+
 ^^^^^^^ Ancil Object #1        ^^^^^^^ Ancil Object #2
         (control message)              (control message)
^
|
+--- sendmsg() "msg_control" points here

Problem is, while passing FDs, iochannel's code try to avoid
variable-length arrays by creating a single cmsg object that can
fit as much FDs as possible:

  union {
    struct cmsghdr hdr;
    uint8_t data[CMSG_SPACE(sizeof(int) * MAX_ANCIL_DATA_FDS)];
  } cmsg;                                 ^^^^^^^^^^^^^^^^^^

Most of the time though the number of FDs to be passed is less
than the maximum above, thus "cmsg_len" is set to the _actual_ FD
array size:

  cmsg.hdr.cmsg_len = CMSG_LEN(sizeof(int) * nfd);
                                             ^^^
This inconsistency tricks the kernel into thinking that we have 2
ancillay data objects instead of one! First cmsg is valid as
intended, but the second is instantly _corrupt_ since it has a
cmsg_len size of 0 -- thus failing kernel's CMSG_OK() tests.

For 32-bit apps on a 32-bit kernel, and 64-bit apps over a 64-bit
one, the kernel's own CMSG header traversal macros just ignore the
second "incomplete" cmsg. For 32-bit apps over a 64-bit kernel
though, the kernel 32-bit socket emulation macros does not forgive
such incompleteness and directly complains of invalid args (due to
a subtle bug).

Avoid this ugly problem, which can also bite us in a pure 64-bit
environment if MAX_ANCIL_DATA_FDS got extended to 5 FDs, by
setting "cmsg_data[]" array size to "cmsg_len".

BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=97769
Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
7 years agoraop: add compatibility with openssl 1.1.0
Tanu Kaskinen [Fri, 4 Nov 2016 12:43:57 +0000 (14:43 +0200)]
raop: add compatibility with openssl 1.1.0

Openssl 1.1.0 made all structs opaque, which caused a build failure in
raop_client.c. The struct member assignments are now replaced with a
call to RSA_set0_key().

BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=96726
Reviewed-by: Felipe Sateler <fsateler@debian.org>
7 years agoraop: Correct spelling of KTH
Anton Lundin [Tue, 25 Oct 2016 19:31:34 +0000 (21:31 +0200)]
raop: Correct spelling of KTH

KTH is a Swedish institution of higher education, and in its full name
spelled Kungliga Tekniska högskolan.

Signed-off-by: Anton Lundin <glance@acc.umu.se>
7 years agoi18n: update Occitan translation
Cédric Valmary [Thu, 13 Oct 2016 12:28:43 +0000 (15:28 +0300)]
i18n: update Occitan translation

7 years agoi18n: update German translation
Mario Blättermann [Wed, 5 Oct 2016 12:16:20 +0000 (15:16 +0300)]
i18n: update German translation

7 years agobluetooth: Add support for automatic switch between hsp and a2dp profiles also for...
Pali Rohár [Sun, 11 Sep 2016 15:16:38 +0000 (17:16 +0200)]
bluetooth: Add support for automatic switch between hsp and a2dp profiles also for bluez5

Bluez5 uses different profile names as bluez4, so we need to check for
a2dp_sink and headset_head_unit too for bluez5 support.

Signed-off-by: Pali Rohár <pali.rohar@gmail.com>
7 years agozeroconf-discover: fix a memory leak
Philip Withnall [Tue, 20 Sep 2016 17:26:00 +0000 (10:26 -0700)]
zeroconf-discover: fix a memory leak

Coverity CID: #1358700

https://bugs.freedesktop.org/show_bug.cgi?id=97876

7 years agoxen: Fixed possible memory leak.
Deepak Srivastava [Mon, 19 Sep 2016 08:36:10 +0000 (14:06 +0530)]
xen: Fixed possible memory leak.

module-xenpv-sink.c - In pa__init(...), memory for pa_modargs *ma is not released before returning from function.

Signed-off-by: Deepak Srivastava <srivastava.d@samsung.com>
7 years agobluetooth: bluez5: Add profile name to sinks and sources
Pali Rohár [Thu, 15 Sep 2016 15:52:40 +0000 (17:52 +0200)]
bluetooth: bluez5: Add profile name to sinks and sources

See commit 380a7fc2401ea26151e0383adff05314a263ac31.

Signed-off-by: Pali Rohár <pali.rohar@gmail.com>
7 years agocore-util: log error if we hit file descriptors limit
Marcin Lewandowski [Thu, 30 Jun 2016 10:38:14 +0000 (12:38 +0200)]
core-util: log error if we hit file descriptors limit

7 years agolaunch: Remove the already implicit After=pulseaudio.socket
Jan Alexander Steffens (heftig) [Fri, 9 Sep 2016 20:09:35 +0000 (22:09 +0200)]
launch: Remove the already implicit After=pulseaudio.socket

Also clarify the comment as to what can actually happen here.

7 years agodaemon-conf: enable memfd by default
Tanu Kaskinen [Thu, 8 Sep 2016 13:20:04 +0000 (16:20 +0300)]
daemon-conf: enable memfd by default

memfd support was introduced in 9.0, but disabled by default. No issues
have been reported - now is a good time to enable it by default.

7 years agozeroconf-publish: unref D-Bus connection
Tanu Kaskinen [Mon, 5 Sep 2016 15:57:17 +0000 (18:57 +0300)]
zeroconf-publish: unref D-Bus connection

pa_dbus_bus_get() increments the bus connection refcount, but unreffing
the connection was never done.

7 years agozeroconf-publish: fix uninitialized variable
Tanu Kaskinen [Mon, 5 Sep 2016 15:55:37 +0000 (18:55 +0300)]
zeroconf-publish: fix uninitialized variable

get_icon_name() returns the icon_name variable, and without this
change the function might exit before icon_name is initialized.

7 years agozeroconf: use local icon for shared devices
Sylvain Baubeau [Thu, 1 Sep 2016 09:14:05 +0000 (11:14 +0200)]
zeroconf: use local icon for shared devices

systemd-hostnamed provides an icon for the machine it is running on.
If it is running, module-zeroconf-publish uses this icon for the
'icon-name' attribute in the Avahi properties. module-zeroconf-discover
passes this icon to module-tunnel using the module parameter
{sink/source}_properties.

This allows to display a portable, desktop or phone instead of
the generic sound card icon.

7 years agosample: Assert validity of sample_spec
Peter Meerwald-Stadler [Thu, 18 Aug 2016 14:08:45 +0000 (16:08 +0200)]
sample: Assert validity of sample_spec

passing an invalid sample_spec to
pa_sample_size_of_format(),
pa_frame_size(),
pa_bytes_per_second(),
pa_bytes_to_usec(),
pa_usec_to_bytes()
currently gives a result of 0

this is problematic as
(a) it leads to many potential divide-by-zero issues flagged by Coverity,
(b) pa_sample_spec_valid() is called often and the mostly unnecessary validation
of the sample_spec cannot be optimized away due to pa_return_val_if_fail()
(c) nobody checks the result for 0 and the behaviour is not documented

this patch replaces pa_return_val_if_fail() with pa_assert()

note that this commit changes the API!
note that pa_return_val_if_fail() strangely logs an assertion, but then happily
continues...

fixes numerious CIDs

7 years agotests: Assert granularity range in stripnul.c
Peter Meerwald-Stadler [Thu, 18 Aug 2016 07:24:41 +0000 (09:24 +0200)]
tests: Assert granularity range in stripnul.c

granularity must not be larger than buffer size

CID 1138482

7 years agocore: Replace PA_PAGE_SIZE with pa_page_size()
Peter Meerwald-Stadler [Wed, 17 Aug 2016 23:06:47 +0000 (01:06 +0200)]
core: Replace PA_PAGE_SIZE with pa_page_size()

PA_PAGE_SIZE using sysconf() may return a negative number

CID 1137925, CID 1137926, CID 1138485

instead of calling sysconf() directly, add function pa_page_size()
which uses the guestimate 4096 in case sysconf(_SC_PAGE_SIZE) fails

using PA_ONCE to only evaluate sysconf() once

7 years agopadsp: Fix flush and improve error handling
Peter Meerwald-Stadler [Wed, 17 Aug 2016 21:52:17 +0000 (23:52 +0200)]
padsp: Fix flush and improve error handling

read() can return a number of bytes read less than k

in addition, handle EAGAIN and EOF

CID 1137981

7 years agoi18n: fix errors and warnings in Belarusian and Korean translations
Piotr Drąg [Fri, 26 Aug 2016 14:26:03 +0000 (16:26 +0200)]
i18n: fix errors and warnings in Belarusian and Korean translations

7 years agoi18n: update Hungarian translation
Balázs Úr [Wed, 24 Aug 2016 05:34:08 +0000 (07:34 +0200)]
i18n: update Hungarian translation

7 years agoi18n: add Belarusian translation
Viktar Vaŭčkievič [Wed, 24 Aug 2016 14:35:03 +0000 (17:35 +0300)]
i18n: add Belarusian translation

7 years agobluetooth: Fix negative array index write
Peter Meerwald-Stadler [Tue, 16 Aug 2016 13:56:40 +0000 (15:56 +0200)]
bluetooth: Fix negative array index write

CID 1533121

7 years agobluetooth: Reorganize code to avoid Coverity NULL dereference warning
Peter Meerwald-Stadler [Tue, 16 Aug 2016 13:33:24 +0000 (15:33 +0200)]
bluetooth: Reorganize code to avoid Coverity NULL dereference warning

CID 1353122

this is a false-positive because

   if (dbus_message_has_interface(p->message, "org.bluez.Manager") ||
        dbus_message_has_interface(p->message, "org.bluez.Adapter"))
        d = NULL;
    else if (!(d = pa_hashmap_get(y->devices, dbus_message_get_path(p->message)))) {
        pa_log_warn("Received GetProperties() reply from unknown device: %s (device removed?)",
dbus_message_get_path(p->message));
        goto finish2;
    }

d can be NULL only if p->message interface is org.bluez.Manager or
org.bluez.Adapter. If

    dbus_message_is_method_call(p->message, "org.bluez.Device", "GetProperties")

returns true, we know that the interface is org.bluez.Device.

thanks, Tanu!

7 years agobluetooth: Fix dead code
Peter Meerwald-Stadler [Tue, 16 Aug 2016 05:42:16 +0000 (07:42 +0200)]
bluetooth: Fix dead code

CID 1353115

Signed-off-by: Peter Meerwald-Stadler <pmeerw@pmeerw.net>
7 years agobluetooth: Don't free modargs twice
Peter Meerwald-Stadler [Tue, 16 Aug 2016 05:16:31 +0000 (07:16 +0200)]
bluetooth: Don't free modargs twice

CID1353139

Signed-off-by: Peter Meerwald-Stadler <pmeerw@pmeerw.net>
7 years agoRemove newline at end of log messages
Peter Meerwald-Stadler [Tue, 16 Aug 2016 05:03:25 +0000 (07:03 +0200)]
Remove newline at end of log messages

Signed-off-by: Peter Meerwald-Stadler <pmeerw@pmeerw.net>
7 years agomodules: Use pa_assert_se() to check return value of dbus_message_iter_close_container()
Peter Meerwald-Stadler [Tue, 15 Sep 2015 14:42:58 +0000 (16:42 +0200)]
modules: Use pa_assert_se() to check return value of dbus_message_iter_close_container()

CID 1140353

... as it is done everythere else

7 years agotests: Check pa_rtpoll_run() return value
Peter Meerwald-Stadler [Tue, 15 Sep 2015 20:34:48 +0000 (22:34 +0200)]
tests: Check pa_rtpoll_run() return value

CID 1138499

7 years agomodules: Check pa_threaded_mainloop_start() return value
Peter Meerwald-Stadler [Tue, 15 Sep 2015 14:49:14 +0000 (16:49 +0200)]
modules: Check pa_threaded_mainloop_start() return value

CID 1138500

7 years agostream: Check pa_tagstruct_get_format_info() retval in pa_create_stream_callback()
Peter Meerwald-Stadler [Tue, 15 Sep 2015 21:11:29 +0000 (23:11 +0200)]
stream: Check pa_tagstruct_get_format_info() retval in pa_create_stream_callback()

CID 1137984

7 years agotests: Assert fillrate > 0 in alsa-time-test
Peter Meerwald-Stadler [Mon, 14 Sep 2015 14:48:10 +0000 (16:48 +0200)]
tests: Assert fillrate > 0 in alsa-time-test

CID 1323592

assert that fillrate is strictly positive

7 years agoalsa: Check pa_modargs_get_value_boolean() retval for use_ucm
Peter Meerwald-Stadler [Mon, 15 Aug 2016 21:11:50 +0000 (23:11 +0200)]
alsa: Check pa_modargs_get_value_boolean() retval for use_ucm

CID 1137983

7 years agosink-input,source-output: Fix logging, don't overwrite old_value when value == 0
Peter Meerwald-Stadler [Mon, 15 Aug 2016 17:08:49 +0000 (19:08 +0200)]
sink-input,source-output: Fix logging, don't overwrite old_value when value == 0

7 years agocore-util: Use _SC_NPROCESSORS_ONLN instead of _SC_NPROCESSORS_CONF
John Paul Adrian Glaubitz [Mon, 4 Jul 2016 18:23:17 +0000 (20:23 +0200)]
core-util: Use _SC_NPROCESSORS_ONLN instead of _SC_NPROCESSORS_CONF

pa_ncpu() is supposed to report the number of processors available on
the system. For that, it currently calls sysconf(_SC_NPROCESSORS_CONF).
However, since the operating system can disable individual processors,
we should call sysconf(_SC_NPROCESSORS_ONLN) to determine the number
of processors currently available [1]. Consequently, the once-test will
fail since pthread_setaffinity_np() is called with CPUs that are
currently not available.

It might also be advisable to change the code in the future to use CPU
sets on Linux as even the suggested change is not 100% safe but at least
it improves over the existing code. If PulseAudio was to be run in a CPU
set [2], the number of processors available to PulseAudio could be even
less than the number of CPUs currently online (_SC_NPROCESSORS_CONF).

[1] https://www.gnu.org/software/libc/manual/html_node/Processor-Resources.html
[2] http://man7.org/linux/man-pages/man7/cpuset.7.html

BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=96809
Signed-off-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
7 years agopipe-sink: align buffer
Nils Schneider [Thu, 11 Aug 2016 15:58:24 +0000 (17:58 +0200)]
pipe-sink: align buffer

The pipe buffer is likely to be a power of 2 (e.g. 4096 bytes). This
works nicely for 16 bit stereo samples but breaks when using 24 bit
samples.

This patch aligns the buffer using pa_frame_align().

7 years agobluetooth: don't create the HSP/HFP profile twice
Tanu Kaskinen [Sat, 30 Jul 2016 23:44:55 +0000 (02:44 +0300)]
bluetooth: don't create the HSP/HFP profile twice

create_card_profile() used to get called separately for HSP and HFP,
so if a headset supports both profiles, a profile named
"headset_head_unit" would get created twice. The second instance would
get immediately freed, so that wasn't a particularly serious problem.
However, I think it makes more sense to create the profile only once.
This patch makes things so that before a profile is created, we check
what name that profile would have, and if a profile with that name
already exists, we don't create the profile.

A couple of Yocto releases (jethro and krogoth) have non-upstream
patches that suffer from this double creation. The patches add
associations between profiles and ports, and those associations use
the profile name as the key. When the second profile gets freed, the
associations between the profile and its ports get removed, and since
the profile name is used as the key, this erroneously affects the
first profile too. Crashing ensues.

BugLink: https://bugzilla.yoctoproject.org/show_bug.cgi?id=10018
7 years agobluetooth: refactor BlueZ 4 transport state setting
Tanu Kaskinen [Sun, 7 Aug 2016 15:45:54 +0000 (18:45 +0300)]
bluetooth: refactor BlueZ 4 transport state setting

Add transport_set_state() that encapsulates changing the variable,
logging and firing the change hook.

I also made a cosmetic change to the corresponding BlueZ 5 log
message so that both messages have the format that I like.

7 years agobluetooth: unify BlueZ 4 and BlueZ 5 profile constant names
Tanu Kaskinen [Sun, 7 Aug 2016 15:45:53 +0000 (18:45 +0300)]
bluetooth: unify BlueZ 4 and BlueZ 5 profile constant names

This should make it slightly easier to copy code between BlueZ 4 and
BlueZ 5.

7 years agobluetooth: unify BlueZ 4 and BlueZ 5 UUID handling
Tanu Kaskinen [Sun, 7 Aug 2016 15:45:52 +0000 (18:45 +0300)]
bluetooth: unify BlueZ 4 and BlueZ 5 UUID handling

A hashmap is more convenient than a linked list for storing the UUIDs,
so change the BlueZ 4 code accordingly.

Rename the BlueZ 4 UUID constants to match the BlueZ 5 naming.

The only changes to the BlueZ 5 code are the addition of one comment
and making another comment a bit clearer.

7 years agobluetooth: update device's valid flag after parsing properties
Tanu Kaskinen [Sun, 7 Aug 2016 15:45:51 +0000 (18:45 +0300)]
bluetooth: update device's valid flag after parsing properties

The properties_received flag affects whether the device should be
considered valid, so let's update the valid flag after setting the
properties_received flag.

There's a call to device_update_valid() anyway later when setting
the device adapters, so this change isn't strictly necessary, but
this makes it more obvious that the code is correct (and less
fragile).

7 years agobluetooth: remove a redundant assignment
Tanu Kaskinen [Sun, 7 Aug 2016 15:45:50 +0000 (18:45 +0300)]
bluetooth: remove a redundant assignment

pa_bluetooth_transport_put() assigns the transport to the device's
transports array, so the caller doesn't have to do that.

7 years agocard-restore: don't switch profiles when availability changes
Tanu Kaskinen [Sun, 7 Aug 2016 15:45:49 +0000 (18:45 +0300)]
card-restore: don't switch profiles when availability changes

module-card-restore should only restore the initial state of new
cards, but profile_available_changed_callback() changed the profile
whenever the saved profile became available. That caused interference
with module-bluetooth-policy, which also sets card profiles based on
the availability changes.

The original reason for having this code was to work around the
problem that bluetooth cards used to be created with only one profile
available, and other profiles would become available soon after the
card creation. Now the bluetooth card creation is delayed until all
profiles are available, so this bad workaround can be removed.

Discussion:
https://lists.freedesktop.org/archives/pulseaudio-discuss/2016-August/026575.html

7 years agobluetooth: wait for all profiles to connect before creating card
Tanu Kaskinen [Sun, 7 Aug 2016 15:45:48 +0000 (18:45 +0300)]
bluetooth: wait for all profiles to connect before creating card

The CONNECTION_CHANGED hook is used to notify the discovery module
about new and removed devices. When a bluetooth device connects, the
hook used to be called immediately when the first profile connected.
That meant that only one profile was marked as available during the
card creation, other profiles would get marked as available later.

That makes it hard for module-card-restore to restore the saved
profile, if the saved profile becomes available with some delay.
module-card-restore has a workaround for this problem, but that turned
out to interfere with module-bluetooth-policy, so the workaround will
be removed in the next patch.

The BlueZ 4 code doesn't need changes, because we use the
org.bluez.Audio interface to get a notification when all profiles are
connected.

7 years agostream-interaction: Fix a memory leak
Arun Raghavan [Wed, 10 Aug 2016 15:53:44 +0000 (21:23 +0530)]
stream-interaction: Fix a memory leak

CID: 1352053

7 years agoshm: Wrap memfd-specific code in relevant ifdef
Arun Raghavan [Wed, 10 Aug 2016 16:13:32 +0000 (21:43 +0530)]
shm: Wrap memfd-specific code in relevant ifdef

Doesn't really affect logic, but Coverity reports this as dead-code, and
I figure it makes sense to be consistent about our use of HAVE_MEMFD.

CID: 1352045

7 years agosource-output: Fix copy-pasto
Arun Raghavan [Wed, 10 Aug 2016 16:07:12 +0000 (21:37 +0530)]
source-output: Fix copy-pasto

CID: 1352047

7 years agosink-input,source-output: Fix a leak during property change logging
Arun Raghavan [Wed, 10 Aug 2016 15:59:56 +0000 (21:29 +0530)]
sink-input,source-output: Fix a leak during property change logging

CID: 1352052

7 years agobuild: Check version of check library, require >= 0.9.10
Peter Meerwald [Wed, 10 Aug 2016 13:22:16 +0000 (15:22 +0200)]
build: Check version of check library, require >= 0.9.10

tests/core-util-test.c uses ck_assert_int_lt() which was introduced
in check 0.9.10

make this dependency (with --enable-tests) explicit in configure.ac

Signed-off-by: Peter Meerwald-Stadler <pmeerw@pmeerw.net>
7 years agotests: Include signal.h in core-util-test.c
Peter Meerwald [Wed, 10 Aug 2016 13:22:15 +0000 (15:22 +0200)]
tests: Include signal.h in core-util-test.c

on oldish Ubuntu 12.04:
tests/core-util-test.c: In function ‘main’:
tests/core-util-test.c:269:66: error: ‘SIGABRT’ undeclared (first use in this function)
     tcase_add_test_raise_signal(tc, modargs_test_replace_fail_1, SIGABRT);

Signed-off-by: Peter Meerwald-Stadler <pmeerw@pmeerw.net>
7 years agoi18n: Update Polish translation
Dušan Kazik Kazik [Thu, 28 Jul 2016 06:50:40 +0000 (12:20 +0530)]
i18n: Update Polish translation

7 years agobluetooth: Add support for automatic switch between hsp and a2dp profiles
Pali Rohár [Sat, 2 Jul 2016 20:26:17 +0000 (22:26 +0200)]
bluetooth: Add support for automatic switch between hsp and a2dp profiles

With this patch module-bluetooth-policy automatically switch from a2dp profile
to hsp profile if some VOIP application with media.role=phone wants to start
recording audio.

By default a2dp profile is used for listening music, but for VOIP calls is
needed profile with microphone support (hsp). So this patch will switch to
hsp profile if some application want to use microphone (and specify it in
media.role as "phone). After recording is stopped profile is switched back
to a2dp. So this patch allows to use bluetooth microphone for VOIP applications
with media.role=phone automatically without need of user interaction.

Signed-off-by: Pali Rohár <pali.rohar@gmail.com>
7 years agomemblockq: remove internal "missing" state variable
Pierre Ossman [Thu, 19 May 2016 13:54:08 +0000 (15:54 +0200)]
memblockq: remove internal "missing" state variable

It was a very confusing state variable that required a lot of
fiddling. It was also redundant in that it can be computed from
the other variables, removing any risk of it getting out of sync.
In the same spirit, make sure "requested" also always contains a
sane value, even though it may not be used by every caller.

7 years agobuild-sys: C++11 is only required for WebRTC support
Romain Naour [Wed, 6 Jul 2016 14:04:27 +0000 (16:04 +0200)]
build-sys: C++11 is only required for WebRTC support

Make C++11 support optional and explicitly check for gnu++11.

Stop the build only if no C++11 support has been found and WebRTC
support has been requested.

Fixes:
http://autobuild.buildroot.net/results/d89/d897a94a8b6a52eba03d04c24536f554a7643000

Signed-off-by: Romain Naour <romain.naour@gmail.com>
7 years agomemblockq-test: Rename test to something more relevant
Arun Raghavan [Fri, 22 Jul 2016 09:51:20 +0000 (15:21 +0530)]
memblockq-test: Rename test to something more relevant

We now have a separate test for _pop_missing().

7 years agomemblockq-test: Add a test for missing data behaviour
Ulrich Eckhardt [Sun, 26 Jun 2016 19:15:44 +0000 (21:15 +0200)]
memblockq-test: Add a test for missing data behaviour

This involves in particular pa_memblockq_missing() and
pa_memblockq_pop_missing(). The test demonstrates that the latter
doesn't work as expected. It should report whenever queue level is
drained below target level. Instead, it reports any case that the queue
level is drained, even when it is still above target level.

7 years agomemblockq-test: Add test changing the parameters of a queue
Ulrich Eckhardt [Thu, 19 May 2016 19:34:22 +0000 (21:34 +0200)]
memblockq-test: Add test changing the parameters of a queue

When changing e.g. the maximum length, the target length may may have to
be adjusted, too.

7 years agomemblockq-test: Add a test that checks initial properties
Ulrich Eckhardt [Thu, 19 May 2016 20:06:12 +0000 (22:06 +0200)]
memblockq-test: Add a test that checks initial properties

Make sure that e.g. the initial size is zero and the initially missing
bytes are exactly the target length etc.

7 years agomemblockq-test: Some refactoring
Ulrich Eckhardt [Thu, 19 May 2016 18:00:46 +0000 (20:00 +0200)]
memblockq-test: Some refactoring

 - Set the loglevel once in the main entry code instead of in each test function.
 - Check pool allocation succeeded.
 - Reduce code by using utility function to allocate chunks.
 - Improve coverage by using utility function to validate queue invariants.

7 years agomemblockq-test: Utility function to validate queue invariants
Ulrich Eckhardt [Sat, 28 May 2016 09:15:49 +0000 (11:15 +0200)]
memblockq-test: Utility function to validate queue invariants

In particular, the relations between base, minreq, tlength, length,
missing, maxlength follow certain rules. On change, these invariants can
be violated, which requires additional code to restore them. Setting one
value can thus cause a cascade of changes. This utility function can
assert those invariants after changing something.

7 years agomemblockq-test: Utility function to alloc chunks
Ulrich Eckhardt [Thu, 19 May 2016 19:32:22 +0000 (21:32 +0200)]
memblockq-test: Utility function to alloc chunks

This eases creating a chunk from a string for writing tests.

7 years agotests: add test to verify correct minreq behaviour
Pierre Ossman [Sun, 26 Jun 2016 21:39:11 +0000 (23:39 +0200)]
tests: add test to verify correct minreq behaviour

Fixed up by Ulrich Eckhardt <ulrich.eckhardt@base-42.de>.

Signed-off-by: Arun Raghavan <arun@arunraghavan.net>
7 years agomemblockq: move minreq handling in to memblockq
Pierre Ossman [Thu, 19 May 2016 13:26:08 +0000 (15:26 +0200)]
memblockq: move minreq handling in to memblockq

Having it handled in the callers proved to be a poor fit as it
became difficult to handle a shrinking minreq sanely. It could end
up in a state where the request was never sent downstream to the
client.

8 years agolaunch: explain why .service depends on .socket
Tanu Kaskinen [Wed, 20 Jul 2016 20:39:59 +0000 (23:39 +0300)]
launch: explain why .service depends on .socket

The reason for depending on the socket unit is rather unobvious, so
let's add a comment to help people reading the service unit file. Felipe
Sateler explained the rationale well in the commit message of
7cb524a77b89, so I just copied the same text into the comment.

8 years agolaunch: make pulseaudio.service properly order and require the socket
Felipe Sateler [Wed, 20 Jul 2016 13:57:13 +0000 (09:57 -0400)]
launch: make pulseaudio.service properly order and require the socket

This commit fixes two problems:

1. Because there are no implicit dependencies between sockets and services,
   the socket, as set up by systemd will race with the socket, as set up
   by the pulseaudio daemon. This can cause the pulseaudio.socket unit to
   fail (even though the pulseaudio service started just fine), which can
   confuse users.
2. While it is possible to use the service without the socket, it is not
   clear why it would be desirable. And a user installing pulseaudio and
   doing `systemctl --user start pulseaudio` will not get the socket
   started, which might be confusing and problematic if the server is to
   be restarted later on, as the client autospawn feature might kick in.

8 years agodaemon: Tone down system mode snarkiness
Arun Raghavan [Thu, 14 Jul 2016 02:38:35 +0000 (11:38 +0900)]
daemon: Tone down system mode snarkiness

We do support system mode for the cases where it makes sense, so it's
really not sensible to be unconditionally snarky at our users for doing
it.

Signed-off-by: Arun Raghavan <arun@arunraghavan.net>
8 years agoloopback: Stop tracking max_request and min_memblockq_length
Georg Chini [Sun, 5 Jun 2016 19:05:11 +0000 (21:05 +0200)]
loopback: Stop tracking max_request and min_memblockq_length

They are not needed any longer because the regulation is now based on the
difference between current and configured latency.

8 years agoloopback: Adjust rate based on latency difference
Georg Chini [Sun, 5 Jun 2016 19:05:10 +0000 (21:05 +0200)]
loopback: Adjust rate based on latency difference

Replace the current latency controller with a modified P-controller. For
better readability separate the controller function. For small latency
differences, the controller forms a classical P-controller while it saturates
at 1% deviation from the base rate for large latency differences.

8 years agoloopback: Restart timer after 1/3 second
Georg Chini [Sun, 5 Jun 2016 19:05:09 +0000 (21:05 +0200)]
loopback: Restart timer after 1/3 second

After switching source or sink, call adjust_rates after a third of a second
instead of waiting one full adjust time. This will ensure that latency regulation
starts as soon as possible.

8 years agoloopback: Move timer restart and snapshots to timer callback
Georg Chini [Sun, 5 Jun 2016 19:05:08 +0000 (21:05 +0200)]
loopback: Move timer restart and snapshots to timer callback

Restaring the timer and obtaining the latency snapshots belong to the timer callback.
To maintain an adjust time as near as possible to the configured value, the timer is
now restarted immediately at the beginning of the timer callback.

8 years agoloopback: Change memblockq length
Georg Chini [Sun, 5 Jun 2016 19:05:07 +0000 (21:05 +0200)]
loopback: Change memblockq length

The size of the memblockq must be increased to allow for long latencies
at high sample rates.

8 years agoloopback: Improve latency estimation
Georg Chini [Sun, 5 Jun 2016 19:05:06 +0000 (21:05 +0200)]
loopback: Improve latency estimation

To improve the overall latency estimation, the delay between the two snapshots
is taken into account. To minimize the snapshot delay, the order of the snapshots
is reverted. Additionally the latency at the base rate is calculated. It will be
used later as the input to the latency controller.

8 years agoloopback: Improve latency estimation in snapshots
Georg Chini [Sun, 5 Jun 2016 19:05:05 +0000 (21:05 +0200)]
loopback: Improve latency estimation in snapshots

The delay and render memblockq are using the source and sink sample specs,
so using pa_bytes_to_usec() will produce better estimates of the delays than
using pa_resmpler_result(). Because the delays are considered to be part of
the sink or source latency, they are added to them. source_output_buffer
becomes obsolete.

8 years agoloopback: Save time stamps during latency snapshots
Georg Chini [Sun, 5 Jun 2016 19:05:04 +0000 (21:05 +0200)]
loopback: Save time stamps during latency snapshots

Saving the time stamps is necessary to account for the delay between the two
latency snapshots. Time stamps will be used in later patches.

8 years agomodargs: Document behaviour on missing arguments
Ulrich Eckhardt [Wed, 6 Jul 2016 18:00:54 +0000 (20:00 +0200)]
modargs: Document behaviour on missing arguments

The behaviour is to leave the value unchanged. The idea is to init the value
with a default before the call and not treat a missing value as error. That
way, only parsing errors or validating errors actually return error codes.

8 years agoprotocol-native: DRY: Remove pdispatch callbacks declarations
Ahmed S. Darwish [Sat, 2 Jul 2016 22:08:06 +0000 (00:08 +0200)]
protocol-native: DRY: Remove pdispatch callbacks declarations

Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
8 years agoalsa: set availability for (some) unavailable profiles
Tanu Kaskinen [Fri, 23 Oct 2015 10:37:11 +0000 (13:37 +0300)]
alsa: set availability for (some) unavailable profiles

The alsa card hasn't so far set any availability for profiles. That
caused an issue with some HDMI hardware: the sound card has two HDMI
outputs, but only the second of them is actually usable. The
unavailable port is marked as unavailable and the available port is
marked as available, but this information isn't propagated to the
profile availability. Without profile availability information, the
initial profile policy picks the unavailable one, since it has a
higher priority value.

This patch adds simple logic for marking some profiles unavailable:
if the profile only contains unavailable ports, the profile is
unavailable too. This can be improved in the future so that if a
profile contains sinks or sources that only contain unavailable ports,
the profile should be marked as unavailable. Implementing that
requires adding more information about the sinks and sources to
pa_card_profile, however.

BugLink: https://bugzilla.yoctoproject.org/show_bug.cgi?id=8448
8 years agocard: remove pa_card_new_data.active_profile
Tanu Kaskinen [Fri, 23 Oct 2015 10:50:35 +0000 (13:50 +0300)]
card: remove pa_card_new_data.active_profile

It's not being used any more.

8 years agocard: move profile selection after pa_card_new()
Tanu Kaskinen [Fri, 23 Oct 2015 09:59:53 +0000 (12:59 +0300)]
card: move profile selection after pa_card_new()

I want module-alsa-card to set the availability of unavailable
profiles before the initial card profile gets selected, so that the
selection logic can use correct availability information.
module-alsa-card initializes the jack state after calling
pa_card_new(), however, and the profile selection happens in
pa_card_new(). This patch solves that by moving parts of pa_card_new()
to pa_card_choose_initial_profile() and pa_card_put().

pa_card_choose_initial_profile() applies the profile selection policy,
so module-alsa-card can first call pa_card_new(), then initialize the
jack state, and then call pa_card_choose_initial_profile(). After that
module-alsa-card can still override the profile selection policy, in
case module-alsa-card was loaded with the "profile" argument. Finally,
pa_card_put() finalizes the card creation.

An alternative solution would have been to move the jack
initialization to happen before pa_card_new() and use pa_card_new_data
instead of pa_card in the jack initialization code, but I disliked
that idea (I want to get rid of the "new data" pattern eventually).

The order in which the initial profile policy is applied is reversed
in this patch. Previously the first one to set it won, now the last
one to set it wins. I think this is better, because if you have N
parties that want to set the profile, we avoid checking N times
whether someone else has already set the profile.

8 years agocard: don't allow the CARD_NEW hook to fail
Tanu Kaskinen [Tue, 7 Jun 2016 13:51:00 +0000 (16:51 +0300)]
card: don't allow the CARD_NEW hook to fail

There is currently no use for allowing modules to cancel card creation,
and I don't see need for that in the future either. Let's simplify
things by removing the failure handling code.

8 years agoalsa, bluetooth: fail if user-requested profile doesn't exist
Tanu Kaskinen [Fri, 23 Oct 2015 09:36:34 +0000 (12:36 +0300)]
alsa, bluetooth: fail if user-requested profile doesn't exist

If we can't fulfill the user request fully, I think we shouldn't
fulfill it at all, to make it clear that the requested operation
didn't succeed.

8 years agopstream: Add rationale for pa_cmsg_ancil_data_close_fds()
Ahmed S. Darwish [Thu, 16 Jun 2016 08:27:37 +0000 (10:27 +0200)]
pstream: Add rationale for pa_cmsg_ancil_data_close_fds()

Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
8 years agomodargs: Add a mechanism to append modargs
Arun Raghavan [Tue, 3 May 2016 12:56:51 +0000 (18:26 +0530)]
modargs: Add a mechanism to append modargs

This allows us to parse an extra set of modargs to tack on to an
existing set. Duplicates in the second set are ignored, since this fits
our use best. In the future, this could be extended to support different
merge modes (ignore dupes vs. replace with dupes), but I've left this
out since there isn't a clear need and it would be dead code for now.

8 years agoalsa: Use pa_device_port->impl_free() for freeing port data
Arun Raghavan [Tue, 3 May 2016 12:52:09 +0000 (18:22 +0530)]
alsa: Use pa_device_port->impl_free() for freeing port data

This allows us to clean up ucm port data associated with a port during
port clean up, instead of having to track this separately using a
dynarray.

8 years agodevice-port: Add mechanism to free implementation data
Arun Raghavan [Tue, 3 May 2016 12:52:08 +0000 (18:22 +0530)]
device-port: Add mechanism to free implementation data

This will be needed if the implementation data stores pointers to
additional data that needs to be freed as well.

8 years agosink, source, device-port: renames to distinguish latency offsets
Chris Billington [Sat, 23 Jan 2016 01:31:34 +0000 (12:31 +1100)]
sink, source, device-port: renames to distinguish latency offsets

Renamed all variables pertaining to latency offsets of sinks and sources,
calling them "port_latency_offset" or similar instead. All of these variables
refer to latency offsets inherited from ports, rather than being unique to
the sinks or sources themselves.

This change is to pave the way for additional functionality for setting
latency offsets on sources and sinks independenly from the value they inherit
from their port. In order to implement them we first need this rename so that
the two latency offsets can be stored individually and summed when reporting
the total latency of the source or sink.

The renames made are:

pa_sink_set_latency_offset() -> pa_sink_set_port_latency_offset()
pa_source_set_latency_offset() -> pa_source_set_port_latency_offset()
sink->latency_offset -> sink->port_latency_offset
sink->thread_info.latency_offset -> sink->thread_info.port_latency_offset
source->latency_offset -> source->port_latency_offset
source->thread_info.latency_offset -> source->thread_info.port_latency_offset
PA_SINK_MESSAGE_SET_LATENCY_OFFSET -> PA_SINK_MESSAGE_SET_PORT_LATENCY_OFFSET
PA_SOURCE_MESSAGE_SET_LATENCY_OFFSET -> PA_SOURCE_MESSAGE_SET_PORT_LATENCY_OFFSET

8 years agowebrtc: improve comment about mic geometry
Tanu Kaskinen [Thu, 2 Jun 2016 13:00:51 +0000 (16:00 +0300)]
webrtc: improve comment about mic geometry

The first mic channel position is not relevant for the target
direction definition.

8 years agosource: Fixed bug: pa_source_set_port() did not update the latency_offset.
Chris Billington [Sat, 23 Jan 2016 01:31:33 +0000 (12:31 +1100)]
source: Fixed bug: pa_source_set_port() did not update the latency_offset.

Unlike pa_sink_set_port(), which calls pa_sink_set_latency_offset() to update
the latency offset of the sink to match that of its newly set port,
pa_source_set_port() did not do so. This patch adds the appropriate call to
pa_source_set_latency_offset() in pa_source_set_port() to fix this.

8 years agojson: Drop refcounting of json objects
Arun Raghavan [Wed, 1 Jun 2016 11:48:39 +0000 (17:18 +0530)]
json: Drop refcounting of json objects

We don't actually use the refcounting bits.

Signed-off-by: Arun Raghavan <arun@arunraghavan.net>
8 years agojson: Add some more negative test cases
Arun Raghavan [Wed, 1 Jun 2016 11:48:38 +0000 (17:18 +0530)]
json: Add some more negative test cases

Signed-off-by: Arun Raghavan <arun@arunraghavan.net>
8 years agojson: Add a positive test for nested objects
Arun Raghavan [Wed, 1 Jun 2016 11:48:37 +0000 (17:18 +0530)]
json: Add a positive test for nested objects

Signed-off-by: Arun Raghavan <arun@arunraghavan.net>
8 years agojson: Error out for objects and arrays that are nested too deep
Arun Raghavan [Wed, 1 Jun 2016 11:48:36 +0000 (17:18 +0530)]
json: Error out for objects and arrays that are nested too deep

Signed-off-by: Arun Raghavan <arun@arunraghavan.net>
8 years agojson: Handle error cases while parsing numbers
Arun Raghavan [Wed, 1 Jun 2016 11:48:35 +0000 (17:18 +0530)]
json: Handle error cases while parsing numbers

Signed-off-by: Arun Raghavan <arun@arunraghavan.net>