5.1.24 regression: Windows 10 guest microphone stops working after high host CPU load

[ocecaco]

In VirtualBox 5.1.24, when the host CPU usage gets very high (i.e., close to 100% on all cores), then the microphone in a Windows 10 guest stops receiving sound. Note that it continues to receive no sound even after the high host CPU usage has stopped. The microphone works just fine before the high CPU load in the host. This issue does not occur in VirtualBox 5.1.22, and therefore appears to be a regression (the changelog mentions audio-related changes for better Windows 10 support, so it would make sense).

This issue occurs with both a Linux host (Ubuntu xenial) and a Windows 10 host. The guest was Windows 10 in both cases.

Steps to reproduce this issue:

1. Create an up-to-date Windows 10 guest using Intel HDA as the sound device. The host can be either Windows or Linux since the issue occurs with both. The host should be using VirtualBox 5.1.24.
2. In the guest, right-click the audio icon next to the clock and choose Recording Devices. This shows a volume meter for the microphone. Check to see that the microphone works before proceeding. Do not switch to another tab in the window during the test since that seems to reinitialize the microphone "connection."
3. In the host, cause a high CPU load. I did this by running 5 instances of a simple Python script that calculates Fibonacci numbers (attached).
4. After about 10-20 seconds, the volume meter either doesn't respond anymore or stays stuck on a particular volume. Killing/waiting for the Python scripts to terminate doesn't cause the volume meter to start working again. Only switching to another tab and back seems to reinitialize the microphone and fixes the issue temporarily. In 5.1.22, the volume meter never freezes.

It is also possible to test using the Voice Recorder, there I noticed that only a static noise is received after several seconds of high host CPU usage. The easiest way to reproduce, however, is to use the Recording Devices panel.

I have tested using VirtualBox 5.2.0 as well, and the same issue exists there. However, 5.1.24 is the earliest version that has this issue.

[ocecaco]

Fibonacci number calculation used in host to cause high CPU usage

[ocecaco]

I tried to reproduce on a different computer, and in step 3 I had to run more of the scripts to cause the bug to occur, in that case about 12 instances of it. It is not clear to me what caused the difference.

[Pi Guy]

+1. Replicated on Mac OS High sierra, and El Capitan host with windows 7 VM, virtualbox 5.1.28, CoreAudio host audio driver, IntelHdAudio Audio Controller

[pentagonik]

Does the issue still persist with the latest version of VirtualBox?

[Pi Guy]

On Version 5.2.2 r119230 (Qt5.6.3) on MacOS (with same details as my other comment), no the same issue occurs.

[pentagonik]

So the bug is still there?

[Pi Guy]

Sorry for slow reply.

Yes the bug is still thereon the version i tested

[pentagonik]

Please have a try with the latest 5.2 test build 120822, which is located here: Testbuilds. Thank you!

[Pi Guy]

Does this include Mac OS X 5.2.x revision 120866?

[pentagonik]

Yes, it does -- revision 120822 or any newer one.

[Pi Guy]

I have not had this bug happen to me on that build, macos 120866. However , i have noticed that there is a delay between me speaking into the microphone and the virtual machine registering a sound. (checked using the 'Listen to this device ' option in the sound settings on windows 7)

[Pi Guy]

Is any investigation still being done on this? It would suck to see this bug persist