(In reply to Alex Deucher from comment #8) > https://patchwork.kernel.org/patch/10889269/ I think it's actually a problem with runtime pm and some pci state. I may ask you to help debug that when I get a chance.