Now it's getting more interesting. I could reproduce the same kernel panic on the 4.10-rc2 kernel I had. And the bug doesn't appear on 4.10.6. So there was a fix somewhere between them. A good hope.