Since 14.0, filter streams would be killed if the master sink or source disappeared. This
patch changes the behavior to allow rescuing of filter streams.
This patch moves the code for the virtual sink callbacks and initialization to
a separate file. The code is re-factored, extended and built as library, so that
it can be used by other virtual sinks as well. The suspend-virtual-on-master-suspend
fix for the ladspa sink (!68) was incorporated into the common code as well as the
sink part of !78 which fixes a crash with stacked virtual sinks. !68 has a bug
which leaves a virtual sink unavailable when the master sink disappears. This bug
is also fixed.
Additionally, fixed block size filters, fixed window size filters and updating filter
parameters are supported by the library. Fixed window size filters always deliver the
same number of frames to the filter, padding new data with history frames if necessary.
Rewinding can be disabled or or limited to the number of frames that the virtual sink
supports. Thanks to Alexander E. Patrakov for pointing out how to rewind fixed block
size filters.
The library implements following command line arguments if they are enabled in
valid_modargs: sink_name, sink_properties, sink_input_properties, force_flat_volume,
remix, resample_method and autoloaded.
This patch introduces the vsink structure which holds all data needed
by virtual sinks. This is in preparation of the consolidation of the
virtual sink code. The input_to_master field of the sink will be moved
to the vsink structure after the consolidation. Until all virtual sinks
are converted, sink->input_to_master and sink->vsink->input_to_master
must both be supported.
If pa_memblockq_push needs to write into the middle of a chunk, target chunk
is split into head and tail sharing the same memblock. Size of head and
tail chunks is adjusted correctly, head chunk pointer into memblock remains
unchanged from target chunk.
The problem is with tail chunk offset into memblock which should be advanced
past write region of memblock, but currently it is left as 0.
This is causing an issue where seeking a few frames back into the middle of
memblock and writing a frame there ends up with tail chunk referencing frames
from very beginning of memblock causing corrupted output from memblockq.
Fix this by adjusting tail chunk offset into memblock past write region and
add a test case.
Fixes#3789
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/798>
ref: https://lore.kernel.org/lkml/20221207154939.2532830-4-jeffxu@google.com/
The new MFD_NOEXEC_SEAL and MFD_EXEC flags allows application to
set executable bit at creation time (memfd_create).
When MFD_NOEXEC_SEAL is set, memfd is created without executable bit
(mode:0666), and sealed with F_SEAL_EXEC, so it can't be chmod to
be executable (mode: 0777) after creation.
when MFD_EXEC flag is set, memfd is created with executable bit
(mode:0777), this is the same as the old behavior of memfd_create.
Signed-off-by: Rudi Heitbaum <rudi@heitbaum.com>
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/792>
Currently there is no way to unset the default sink or source once it was
configured manually by the user.
This patch introduces the special name @NONE@, which can be used with the pacmd
or pactl set-default-sink and set-default-source commands to unset the user
configured default. When the default is unset, pulseaudio will return to the
standard default sink or source selection mechanism based on priority.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/785>
Currently module-switch-on-connect overwrites the default sink or source that
the user has configured. This means that when the overwritten default sink or
source becomes unavailable, the new default will be chosen based on priority
and the default will not return to the originally configured value.
This patch solves the issue by introducing new core variables for the sink
or source chosen by the policy module which have higher priority than the
user configured defaults.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/784>
There is no requirement for chunk index to be aligned, we only need chunk length
to be multiple of sample frame size.
Fixes: 6434853b0 ("memblockq: Do not allow non-frame indices in the memblock queue")
Fixes: 22827a5e1 ("protocol-native: Fail if trying to push unaligned memblock into queue")
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/779>
`pa_pstream_send_memblock()` would split incoming memblock into parts not
exceeding maximum pool block size.
To make sure split parts of memblock are still frame-aligned add new `align` arg
to `pa_pstream_send_memblock`, find out required alignment from stream sample
format and pass it there. Bump default alignment to 256 which is good up to
32bit 64ch frames.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/780>
The assumption that the format enum is ordered by size is not valid for quite
some time, since 24bit formats were appended to format enum later than 32bit
formats. This causes resampler to produce properly aligned memblock of size
larger than maximum mempool block size if input format is 24bit and output
format is 32bit.
Oversized block is getting split by `pa_pstream_send_memblock()` into parts of
size not exceeding maximum mempool block size. This usually works well but for
32ch 32bit 48000Hz stream the frame alignment is 128 bytes and maximum mempool
block size value is multiple of 64 but not 128 bytes, therefore resulting parts
are misaligned.
On receiving side this causes extra allocation of 128 byte chunk while `mcalign`
helper reassembles properly aligned frame out of second block of misaligned
size. While first and second properly aligned frames are retrieved successfully
from `mcalign` helper, third retrieved frame would end up with properly aligned
size but misaligned memblock index (in this example, that would be 64 bytes.)
Attempt to push a chunk with misaligned memblock index causes assertion failure
Assertion 'uchunk->index % bq->base == 0' failed at memblockq.c:289,
function pa_memblockq_push(). Aborting.
Fix oversized block issue by checking proper size of format instead of enum
value.
Fixes: a67c21f09 ("merge 'lennart' branch back into trunk.")
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/778>
This fixes the rare case of resume_time being bigger than time_stamp. Which
happens sometimes when a gstreamer client is quickly seeking through a
media file. The resulting integer underflow then causes a huge value in
current_time which will break the playback.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/764>
* Enable macOS specific modules (module-bonjour-publish,
module-coreaudio-detect and module-coreaudio-device)
* Correctly set `PA_SOEXT` (.so, .dylib and .dll)
* Build `poll-posix.c` and `semaphore-osx.c`
* Drop linker flag `-Wl,-z,nodelete` on Darwin
* Drop linker flag `-Wl,--no-undefined` on Darwin
* Prefer to `clock_gettime` over compat impl for old Darwin
* Disable SCM credential on Darwin
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/746>
When a stream is started but has not yet called smoother_2_put(), pa_smoother_2_get()
returns the time since the start of the stream even if the stream was started paused.
When the stream is started paused, pa_smoother_2_get() should return 0 instead. This
patch fixes the problem.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/745>
Add libatomic_ops dependencies to libraries/modules that showed a
failure on an arch that does not have native atomic operations support.
Not all optional dependencies were tested, so it is possible that
some optional modules are still missing libatomic_ops dependencies.
Signed-off-by: Nicolas Cavallari <nicolas.cavallari@green-communications.fr>
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/732>
Add complementary functions to the existing idxset iterate(),
steal_first(), first(), next() functions that work in the reverse
direction: reverse_iterate(), steal_last(), last() and previous().
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/596>
This is functionally equivalent to get_by_data(s, p, NULL) == p, but
with a more obvious name and form because some existing code is instead
manually iterating through idxsets to check for existence of an item.
Signed-off-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/596>
A recent commit added i->origin sink for the sink inputs of the combine sinks.
Therefore pa_sink_process_input_underruns() treated the combine sink like
filter sinks. pa_sink_process_input_underruns() calls itself with the
origin sink, which is only correct for filter sinks because they run in the
thread context of the origin sink. The combine sink however has its own
thread context, so pa_sink_process_input_underruns() was executed in the
wrong context.
This patch fixes the issue by skipping the section for module-combine-sink.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/722>
pulseaudio crash occurred when I play a file using pacmd play-file command.
The file is not aligned with its frame size and the last rendering size
is also not aligned. Thus, an assertion was generated at the end of the
file as the following.
memblockq.c: Assertion 'uchunk->length % bq->base == 0' failed at
../src/pulsecore/memblockq.c:288, function pa_memblockq_push(). Aborting.
When I play the file using paplay, it works good. So, I changed to
pa_memblockq_push_align instead of pa_memblockq_push to prevent the
assertion.
Signed-off-by: Jaechul Lee <jcsing.lee@samsung.com>
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/717>
Taking addresses of fields in a packed struct are not guaranteed to be
aligned, resulting in warnings such as:
../src/pulsecore/shm.c: In function 'sharedmem_create':
../src/pulsecore/shm.c:198:25: error: taking address of packed member of 'struct shm_marker' may result in an unaligned pointer value [-Werror=address-of-packed-member]
198 | pa_atomic_store(&marker->pid, (int) getpid());
| ^~~~~~~~~~~~
The struct already has its fields and types laid out in such a way that
the desired packing (without padding) is guaranteed - enforce this with
a `static_assert` to get rid of the unaligned pointer warning.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/653>
This patch adds an alternative time smoother implementation based on the theory
found at https://www.freedesktop.org/software/pulseaudio/misc/rate_estimator.odt.
The functions were written to replace the current smoother functions nearly on
a one-to-one basis, though there are a few differences:
- The smoother_2_put() function takes a byte count instead of a sound card
time as argument. This was changed because in most places a sample count
was converted to a time before passing it to the smoother.
- The smoother needs to know sample rate and frame size to convert byte
counts to time.
- A smoother_2_get_delay() function was added to directly retrieve the stream
delay from the smoother.
- A hack for USB devices was added which works around an issue in the alsa
latency reports for USB devices.
The smoother delivers much better precision than the current implementation.
For results, see the document referenced above.
The new functions are still unused. The following patches will convert all
callers of the smoother functions so that they can use both smoother
implementations, depending on a configure option.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/55>
This patch is in preparation of allowing virtual sinks to specify their own
max_rewind limit.
Currently pa_sink_set_max_rewind_within_thread() simply sets the value of
max_rewind and informs the sink inputs about the new value. Virtual sinks
may however provide their own limit on max_rewind.
This patch allows to query the active sink inputs for the max_rewind value
they support and sets max_rewind to the minimum supported value. This way,
the max_rewind value from the virtual sinks can be communicated to the master
sink.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/120>
This patch is in preparation of allowing virtual sinks to specify their own
max_rewind limit.
Currently, virtual sinks cannot specify their max_rewind limit, but just copy
the value from the master sink. This may not be correct, if the DSP code of the
virtual sink has limited (or no) rewinding capability.
Because the DSP code of the virtual sink is rewound in the process_rewind()
callback of the sink input, it must be ensured, that rewinding a sink input
to the master of a virtual sink is limited similar to rewinding a sink.
There are two remaining exceptions:
1) If an underrun is detected. In that case, the filter should be reset anyway.
2) When the sink input of the filter is moved and attached to the destination
sink.
The move case is handled without involvement of the implementer, so the implementer
can only receive a rewind larger than max_rewind when the filter should be reset
anyway.
All existing virtual sinks do not distinguish between reset and rewind of the
filter.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/120>
If the output implements a process_rewind() callback, the resampler delay is
not taken into account. This leads to glitches during volume changes when
source and source output rates differ.
This patch fixes the problem.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/120>
The introduction of the history queue makes it possible to implement moving
of streams without involving the implementer. Instead of dropping all data
from the render memblockq and requesting the implementer to rewrite the
data, the render memblockq is now reconstructed from the history queue.
Additionally, the render queue will be filled with silence matching the
amount of audio that is left playing on the old sink to avoid playing
the same audio twice.
This patch slightly breaks moving for virtual sinks because they do not
yet include the resampler delay in their latency reports. This will be
fixed in a different patch set.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/120>
This patch uses the two previous patches to implemnt pseudo-rewinding for the
resamplers by feeding some old data into the resampler after a reset. This is
necessary because PA is using external resamplers that do not implement
rewinding.
To get exactly the same output data from the resampler after a rewind if possible,
the matching period is calculated. This is the number of input samples that produces
an integral number of output samples. After the matching period, the resampler state
repeats. If the matching period is not too large, feeding history into the resampler
will start at a point that is a multiple of the matching period back in time. Then
the resampler will produce exactly the same samples.
The PA_RESAMPLER_MAX_HISTORY value has been replaced by PA_RESAMPLER_MAX_DELAY_USEC
and the required number of history samples is calculated from the sink input sample
rate. The number of history samples can be as large as about 12500.
This fixes glitches during volume changes when the sink runs on a rate different
from the sink input rate.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/120>
The pa_resampler_get_delay() function allows to retrieve the current resampler
delay in input samples for all supported resamplers. The return value is a double
to maintain precision when using variable rate resamplers. Because in many places
the delay is needed in usec, pa_resampler_get_delay_usec() was also supplied.
The speex resampler now skips leading zero samples to provide meaningful delay values.
In the next patch, the pa_resampler_prepare() function will be used to train the
resampler after a rewind. It takes data from a history memblockq and runs it through
the resampler. The output data is discarded.
To make this logic possible, the soxr resampler had to be converted to use variable
rate. The fixed rate version has a variable delay, therefore the logic above could
not be applied. Additionally, with fixed rate, the delay is larger than 150ms in
some situations, while with variable rate the delay is fixed and comparable to the
other resamplers.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/120>
A new memblockq is added to the sink input code to keep some history of the
input data. The queue is kept in sync with the render memblockq. The old input
data will be used to prepare the resampler after a rewind.
pa_resampler_request() and pa_resampler_result() have been changed to round
as good as possible to avoid loosing or duplicating samples during rewinds.
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/120>
Requiring user to invoke send-message with correctly quoted parameters string
is not good for usability. Wrap parameters string into JSON string and try
again. If that works, log a warning use wrapped JSON string with parameters.
As an example these two commands will now invoke the same action:
pactl send-message /card/bluez_card... switch-codec '"CODECNAME"'
pactl send-message /card/bluez_card... switch-codec CODECNAME
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/648>
Recently we found an issue of output volume on speaker and headphone,
they should have their own volume but in practice they share one
output volume.
This issue happens on the laptops which use the ucm2 sof-hda-dsp,
originally the speaker has output volume A while the headphone has the
output volume B, suppose the speaker is the active port at the moment
and the output volume is A, users plug a headphone to the jack and the
headphone becomes the active port, in this process, ucm_set_port()
calls _disdev/_enadev which triggers the io_mixer_callback(), in the
meanwhile, the module_device_restore will restore the headphone's
volume to B, it will call set_volume_cb() to set the volume to B, but
this value is not written to hw immediately, during the time of
waiting for the B to be written to the hw, the io_mixer_callback()
calls get_volume_cb(), it reads hw volume and gets the volume A, then
it overrides the output volume to A, this results in the headphone
gets the volume A instead of B.
If a machine doesn't use the ucm, this issue will not happen since the
set_port_cb() will not trigger the io_mixer_callback(). If the ports
don't belong to the same sink/source, this issue also doesn't happen.
BugLink: http://bugs.launchpad.net/bugs/1930188
Signed-off-by: Hui Wang <hui.wang@canonical.com>
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/577>
* Minimal implementation of --system on win32.
* Wrap main with a Windows Service on win32 (with a fallback to
running it directly).
* Update PA_SYSTEM_{RUNTIME,STATE,CONFIG}_PATH and HOME dynamically
on Windows (overrides the build config, similar to the existing
config path replacement logic).
Part-of: <https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/549>
