On Tue, 2014-06-24 at 09:44 +1000, Gavin Shan wrote:

quoted

I much'd prefer see a opal_pci_err_inject that is specific to
IO(D)A errors, which takes a PHB ID and goes via the normal dispatch
to PHB ops inside OPAL. For the rest, especially core specific
injections, we can provide a separate dedicated call.

Thanks, Ben. I'll change the firmware API to have a separate
API (opal_pci_err_inject) for PCI errors.

Also, how do we expose to Linux that the new API is supported ?

Linux shouldn't create the additional files if it isn't...

There are two ways, we can have a property in the DT of the PHB
indicating that it supports error injection or we can check for the
existence of the OPAL token at boot (there's an OPAL call to do that).

Cheers,
Ben.