On Thu, Jun 23, 2005 at 10:28:07PM -0400, John W. Linville wrote: > On Thu, Jun 23, 2005 at 03:14:52PM -0400, John W. Linville wrote: > > > This issue regarding D3hot->D0 state transitions seems like a piece > > of minutiae that we should not force individual drivers to address. > > After some thought, I'm inclined to think that the patch below is > the right one. It unconditionally saves and restores the PCI config > registers when pci_enable_device is called. Although this is often > (usually?) unnecessary, it seems like a safe thing to do and it should > not be a performance-sensitive path. The code to check for whether > or not this is necessary would be a little harder to read IMHO, > so I think this is warranted. But how does this solve your problem with the state change? > The comment block at the head of pci_enable_device says "Initialize > device before it's used by a driver" which implies that saving and > restoring the PCI config should be safe if the function is used > as intended. All pci drivers must call pci_enable_device() before they start to use it. thanks, greg k-h