[pulseaudio-discuss] Best Case Latency

[Patrick Shirkey]

On Thu, October 3, 2013 11:41 pm, Thomas Martitz wrote:
> Am 03.10.2013 10:20, schrieb Patrick Shirkey:
>> On Thu, October 3, 2013 5:14 pm, Tanu Kaskinen wrote:
>>>
>>> That doesn't look healthy. The message is printed when
>>> pa_asyncmsgq_post() is called and the message queue is full. The
>>> message
>>> queue can store 256 messages before this starts to happen, so some
>>> queue
>>> consumer is having serious trouble keeping up with the producer. It
>>> would be nice to know which pa_asyncmsgq_post() call this is (you could
>>> set a breakpoint on the line that prints "q overrun", and then get a
>>> backtrace).
>>>
>> Sorry, if this is dense but how do I set a breakpoint on this line in PA
>> while it is running?
>
> $ gdb --pid=$(pidof pulseaudio)
>
> If you're on debian you need to install debug symbols ($ sudo apt-get
> install pulseaudio-dbg). For other distros I can't help. Compiling from
> source in debug mode of course also works.
>

Thanks. FYI, installing "pulseaudio-dbg" hosed my NVIDIA video driver
which was a bit confusing.

Looking at /var/log/messages I spotted this when starting PA:

----
Oct  4 17:30:49 xxx pulseaudio[6571]: [pulseaudio] sink.c: Default and
alternate sample rates are the same.
Oct  4 17:30:49 xxx pulseaudio[6571]: [pulseaudio] module-jack-sink.c:
JACK error >Cannot use real-time scheduling (RR/5)(1: Operation not
permitted)<
Oct  4 17:30:49 xxx pulseaudio[6571]: [pulseaudio] module-jack-sink.c:
JACK error >JackClient::AcquireSelfRealTime error<
Oct  4 17:30:49 xxx pulseaudio[6571]: [pulseaudio] source.c: Default and
alternate sample rates are the same.
Oct  4 17:30:49 xxx pulseaudio[6571]: [pulseaudio] module-jack-source.c:
JACK error >Cannot use real-time scheduling (RR/5)(1: Operation not
permitted)<
Oct  4 17:30:49 xxx pulseaudio[6571]: [pulseaudio] module-jack-source.c:
JACK error >JackClient::AcquireSelfRealTime error<
----

- Not sure why it is being printed twice. I have enabled realtime in
/etc/pulse/daemon.conf and jack is running in realtime mode so not sure
why PA is unhappy. It could be a bug that is already fixed though.

- I'm running pulseaudio 4.0 in debian  pulseaudio-dbg:amd64/unstable
4.0-6+b1

- Here's the backtrace based on the line number that Tanu provided. Not
sure if I nailed it though.

(gdb) break asyncq.c:211
Breakpoint 1 at 0x7f4d4572ff90: file pulsecore/asyncq.c, line 211.
(gdb) continue
Continuing.
[Switching to Thread 0x7f4d39b2b700 (LWP 6686)]

Breakpoint 1, 0x00007f4d4572ff90 in pa_asyncq_post () from
/usr/lib/libpulsecore-4.0.so
(gdb) bt
#0  0x00007f4d4572ff90 in pa_asyncq_post () from /usr/lib/libpulsecore-4.0.so
#1  0x00007f4d4572ee7b in pa_asyncmsgq_post () at pulsecore/asyncmsgq.c:137
#2  0x00007f4d3bf374f3 in source_output_push_cb () at
pulsecore/protocol-native.c:1832
#3  0x00007f4d4576d3b3 in pa_source_output_push () at
pulsecore/source-output.c:822
#4  0x00007f4d45773e5b in pa_source_post () at pulsecore/source.c:935
#5  0x00007f4d39baf820 in source_process_msg () at
modules/jack/module-jack-source.c:115
#6  0x00007f4d4574d084 in asyncmsgq_read_work () from
/usr/lib/libpulsecore-4.0.so
#7  0x00007f4d4574c607 in pa_rtpoll_run () from /usr/lib/libpulsecore-4.0.so
#8  0x00007f4d39baf953 in thread_func () at
modules/jack/module-jack-source.c:201
#9  0x00007f4d454f5d08 in ?? () from
/usr/lib/x86_64-linux-gnu/pulseaudio/libpulsecommon-4.0.so
#10 0x00007f4d449fae0e in start_thread () from
/lib/x86_64-linux-gnu/libpthread.so.0
#11 0x00007f4d4402593d in clone () from /lib/x86_64-linux-gnu/libc.so.6





--
Patrick Shirkey
Boost Hardware Ltd