On 23.07.25 10:44, yicongsrfy@163.com wrote:

On Wed, 23 Jul 2025 09:17:02 +0200 Oliver [off-list ref] wrote:

quoted

On 23.07.25 03:29, yicongsrfy@163.com wrote:

quoted

  From these two tests, we can conclude that both full-duplex
and half-duplex modes are supported — the problem is simply
that the duplex status cannot be retrieved in the absence of
MII support.

Sort of. You are asking a generic driver to apply a concept
from ethernet. It cannot. Ethernet even if it is half-duplex
is very much symmetrical in speed. Cable modems do not, just
to give an example.

I think we need to centralize the reaction to stuff that is not ethernet.

Thanks!

I think I understand what you mean now.
You're suggesting to create a unified interface or
framework to retrieve the duplex status of all CDC
protocol-supported devices?

Well no. We have to understand that the difference in duplex
status apply only to a subset of devices. In a way the network
layer is deficient in only having an unknown status rather than
an unknown and an inapplicable status.

If we are to retain a single status for simplicity, then
I'd say that the default being half duplex rather than
unknown is wrong.

This seems like a rather big undertaking, and one of the key
reasons is that the CDC protocol itself does not define anything
related to duplex status — unlike the 802.3 standard, which
clearly defines how to obtain this information via MDIO.

CDC is not a network protocol. CDC is a protocol for talking
to network devices. CDC can define a way to transmit duplex
information. If the protocol under CDC does not have the notion,
this will be of no use.

Coming back to the issue described in this patch,
usbnet_get_link_ksettings_internal is currently only used in
cdc_ether.c and cdc_ncm.c as a callback for ethtool.
Can we assume that this part only concerns Ethernet devices
(and that, at least for now, none of the existing devices can
retrieve the duplex status through this interface)?

Yes. It is not really that important. But strictly speaking the
network layer is using the wrong default.

	Regards
		Oliver