On 1/11/24 4:43 AM, Cédric Le Goater wrote: > OK. this condition is considered serious enough to be reported to a > management level. This seems a bit excessive since the recovery can be > handled by software, but manually. Are there any plans to address this > problem ?
Yes, eventually. Currently for drivers other than QEMU+vfio-pci we rely on automated PCI recovery to handle this event, but today we do not forward the event to QEMU when using vfio-pci because the code in vfio_err_notifier_handler will halt the guest. This leaves the device in an error state to be recovered manually. The intent is to eventually address this by triggering the guest to initiate the recovery action rather than halting the guest for s390.