Android openSLES mal_device_uninit deadlock about miniaudio HOT 14 CLOSED

What thread are you calling mal_device_uninit() from? What version of Android? Hardware or emulator?

EDIT: Is it happening consistently, or randomly?

from miniaudio.

both the last version of master and dev, Android 7.0 , Hardware , Xiaomi redmi , consistently.
last master line.18239
last dev line.18250

the log pthread 0x7f6a27a450 is the caller thread, a object locked by self ? or the mutexlock need a PTHREAD_MUTEX_RECURSIVE lock ?
or need call some functions to release the object

i found a android issue there:
https://grokbase.com/t/gg/android-ndk/156tayqf1c/opensl-deadlock-issue

https://groups.google.com/forum/#!searchin/android-ndk/deadlock/android-ndk/G7dLKAGGL28/xwlmjpdv2ssJ

from miniaudio.

I'm not sure what's going on. This is the first report I've had with this. I'm wondering if it's something to do with the callback being running at the same time as Destroy() is called. Not 100% sure, though... Are you calling mal_device_uninit() from the same thread as the call to mal_device_init()? Also, make sure you're not calling it from the callback thread.

from miniaudio.

Is it happening on any other devices or emulators?

from miniaudio.

i have found lots of this issue of opensl, it's not mini_al issue , i create and destroy the device by same thread , and not call it from callback .
i test it only one device, let me test it from other device

from miniaudio.

I test it in huawei 7x / android 8.0, same as xiaomi redmi4x, playback success, but when uninit device , it crash on libwilhelm.so(OpenSLES) pthread_mutex_trylock, error : SIGSEGV (signal SIGSEGV: invalid address (fault address: 0x55555555555591))
xiaomi redmi note4x : deadlock screenshot

from miniaudio.

Test in Xiaomi mi6 / Android 8.0, same as Xiaomi redmi note4x, deadlok.

seems ,it's not the random occue event and in less class of device , is there a solution to avoid openSLES issue?

from miniaudio.

OpenSL|ES is the only supported backend for Android. An AAudio backend is planned, but this is only supported on later versions of Android.

This feels like a bug in mini_al somewhere. I will need to do a few experiments and get back to you.

from miniaudio.

and i test in android emulator 7.0 (X86_64 without google api) ,it would deadlock too.

from miniaudio.

OK something I just saw in your screenshot is that the variable g_malOpenSLInitCounter is set to 0. That means the mal_context object has either been uninitialized before uninitializing the device (you must uninitialize all devices before attempting to uninitialize the context) or you have not initialized the context at all. Can you please check that the context is not getting uninitialized before the device.

Some basic things to consider:

• If you are using the onStop callback, make sure you're not uninitializing the context in onStop and then again outside of it.
• Your tear down should look something like this:

mal_device_uninit(device); // <-- Must do this _before_ mal_context_uninit()
mal_context_uninit(context);

Meanwhile, I have pushed an update to the dev branch which checks for this and throws an assert. Can you please let me know what the outcome of that is?

from miniaudio.

Yes, it's my fault, i init mal_context before init mal_device , should be uninit mal_context after uninit mal_device , but before that , thank you point out the error .i changed that it's work. :) :) :)

from miniaudio.

That's good to hear! So everything is working now?

from miniaudio.

all ok, please close the thread.

from miniaudio.

Great! Closing.

from miniaudio.