On Wed, Jan 02, 2019 at 08:00:41PM -0800, Philip Langdale wrote: > > 0e157e5 PCI/PME: Implement runtime PM callbacks > > Yes, if I revert this change it also works. And I am reminded that when > I tested the hash from the usb branch that was merged for 4.20, but on > the original branch, it didn't fail either. So I guess it's the > interaction between the two changes that causes the breakage. Right, if only the xHCI is runtime suspended but not the PCIe switch it is part of, things should work. We definitely want to make sure the whole hierarhcy goes into D3cold whenever possible (to save energy) so xHCI runtime PM should be enabled. The above commit tries to fix another issue but it seems it is not correct and causes issues like what you are seeing and another one reported here: https://bugzilla.kernel.org/show_bug.cgi?id=202103 I'm currently investigating it.
