> No. The server may only assume that the command failed until it gets the > response for the abort command. > Before that the state of the command is undefined, and the server may > not assume anything here.
You're right, but the device will think that the command succeeded, so it may send packets in a queue that is not expected by the server. I'm referring to the VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET command. Don't you think that as a quick fix, adding a big enough timeout is an adequate solution (for virtio net control commands)? The big timeout ensures that when the control command is not answered, it is because of some crash/bug in the network device (which acts as the hypervisor). This way, we can detect the crash/bug and reset/remove the device. _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
