CPU usage spike and unresponsiveness when guest attempts to connect to any port 127.0.0.1 on a MacOS host with TSI

[ggoodman]

When a guest attempts to establish a socket connection to 127.0.0.1:<any_port> on a MacOS host using TSI, the VMM cpu spikes and the VM becomes temporarily unresponsive.

With logging enabled the following lines are repeated hundreds, if not thousands, of times:

[2025-06-05T12:49:53Z DEBUG devices::virtio::vsock::tcp] process_event: HANG_UP
[2025-06-05T12:49:53Z DEBUG devices::virtio::vsock::tcp] push_reset: id: 5562958799897100908, peer_port: 1295227278, local\_port: 620
[2025-06-05T12:49:53Z DEBUG devices::virtio::vsock::muxer_thread] update_polling id=5562958799897100908 fd=157 evset=(empty)
[2025-06-05T12:49:53Z WARN  devices::virtio::vsock::muxer\_thread] deferring proxy removal: 5562958799897100908
[2025-06-05T12:49:53Z DEBUG devices::virtio::vsock::muxer\_thread] signal IRQ
[2025-06-05T12:49:53Z DEBUG devices::virtio::vsock::muxer\_thread] Event: ev.data=5562958799897100908 ev.fd=620

Once this condition is triggered, the VMM process maxes out CPU and remains in this state until killed.

A simple reproducer is:

curl http://127.0.0.1:8080

Screen.Recording.2025-06-05.at.10.29.08.AM.mov

[jeromegn]

I'm facing the same issue. I've had an AI fix it for me, but I'm not sure if the fix is correct. It sounds correct!

I don't think this will apply cleanly, but should give an idea of the changes you may need to make: https://gist.github.com/jeromegn/fc3f6813f1c92ac93a119a47c1f4091d