Bug#825931: s390-netdevice virtio interface choice misleading
On 26.10.2016 22:53, Philipp Kern wrote:
> On 10/26/2016 05:13 PM, Dimitri John Ledkov wrote:
>> Currently s390-netdevice in the code supports following
>> networktypes: * qeth * ctc * iucv * virtio
>>
>> virtio is, in fact, a no-op.
>
> Yup.
>
>> iucv refers to "IUCV network device" which has been deprecated
>> at least since October 2005 (linux v2.6+)
>
> I guess IUCV is only deprecated for networking, not for console
> access, right? I would not mind dropping it then.
>
>> ctc refers to channel-to-channel devices which has also been
>> deprecated at least since October 2005 (linux v2.6+)
>
> CTC is used by default by Hercules. I suppose it can also do qeth
> now? Or maybe that's just in a fork? Last I checked it still
> worked, as well.
>
>> The migration path from above to is onto Hipersockets or QDIO.
>> Both migration paths are handled on the linux side by the qeth
>> kernel module as fas as I understand.
>>
>> Hence, the only option of the 4 that may do anything is qeth, if
>> there are any qeth devices detected that need activation.
>>
>> Furthermore, I'm not sure if iucv network device works correctly
>> - confirm_iucv function returns WANT_ERROR, unless after
>> activating iucv one must continue with the ctc code path.
>>
>> Imho, we should simply skip the whole d-i module if there are no
>> qeth devices to activate. If there are any qeth devices to
>> activate, display multi select question of the qeth devices to
>> activate. Multi-select question, such that continue without
>> selecting any is a valid option (for example if one wants to
>> install using the PCIe interfaces).
>
> I very much sympathize with this idea in general. If we can also
> detect if CTC devices are on the bus, only showing the prompt if
> devices can be found would be useful. Did the kernel start off not
> offering all devices in sysfs because LPARs can have a huge amount
> of them? Or did it always rely on blacklisting in this case? I
> suppose that the original reason was that you actually had to enter
> device numbers because it couldn't probe, but then on the other
> hand the selection for qeth was shown for a long time now.
Whether (and which) CCW devices are probed and exposed via sysfs is
controlled by the cio_ignore kernel parameter, actually black and
white listing at the same time. And the reason to add this was that
LPARs can have really many devices.
The last time I looked the installer modules were not equipped to deal
with cio_ignore and the installer image didn't utilize the cio_ignore
parameter, so a Debian installer will see all CCW devices.
>
> Kind regards and thanks Philipp Kern
>
--
Mit freundlichen Grüßen/Kind Regards
Viktor Mihajlovski
IBM Deutschland Research & Development GmbH
Vorsitzender des Aufsichtsrats: Martina Köderitz
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen
Registergericht: Amtsgericht Stuttgart, HRB 243294
Reply to: