Re: [PATCH] virtio-scsi: hotplug suppot for virtio-scsi

[Paolo Bonzini <pbonzini@redhat.com>]

Il 02/07/2012 09:20, mc@linux.vnet.ibm.com ha scritto:
>>> +static void virtscsi_handle_event(struct work_struct *work)
>>> +{
>>> +    struct virtio_scsi_event_node *event_node =
>>> +        container_of(work, struct virtio_scsi_event_node, work);
>>> +    struct virtio_scsi *vscsi = event_node->vscsi;
>>> +    struct virtio_scsi_event *event = &event_node->event;
>>> +
>>> +    switch (event->event) {
>>> +    case VIRTIO_SCSI_T_TRANSPORT_RESET:
>>> +        virtscsi_handle_transport_reset(vscsi, event);
>>> +        break;
>>
>> Please handle VIRTIO_SCSI_T_NO_EVENT too here, and mask out
>> VIRTIO_SCSI_T_EVENTS_MISSED even if you do not handle it by triggering a
>> full rescan.
>>
> Paolo, what's the appropriate action to response VIRTIO_SCSI_T_NO_EVENT
> event? An error message?

No, just do nothing.

> I also have some confusion on VIRTIO_SCSI_T_EVENTS_MISSED:
> Should VIRTIO_SCSI_T_EVENTS_MISSED always be used with other event flag?

Yes.

> For instanse:
> (VIRTIO_SCSI_T_EVENTS_MISSED|VIRTIO_SCSI_T_TRANSPORT_RESET) is used to
> report hotplug event dropped when a buffer available.

The idea in the spec was that QEMU would report VIRTIO_SCSI_T_NO_EVENT |
VIRTIO_SCSI_T_EVENTS_MISSED as soon as a buffer becomes available, but
technically this is indeed possible.  So you need to handle
VIRTIO_SCSI_T_EVENTS_MISSED first, and the other bits separately.

Paolo