Hello guys,

Our preliminary tests seems to indicate this solves the issue with the ixgbe driver. My fear is now that other drivers might show the same behavior. Patching ovs to fallback to ethtool is a good way to avoid this imho.


Thank you very much,

David Hill

On 08/31/2016 03:46 PM, Jeff Kirsher wrote:
On Tue, 2016-08-30 at 17:35 -0400, David Hill wrote:
Hello sir,

     I do not understand either but that's what's happening.  It's a
fiber adapter so maybe the driver says it's supported and when trying to
request values from the nic itself it fails because it's no longer
supported... but that's only a though .

One of our ixgbe maintainers has created a patch http://patchwork.ozlabs.or
g/patch/664633/ to fix the issue.  If you could have your customer please
try the patch in the above link, to verify that it fixes the issue would be
greatly appreciated.

On 08/30/2016 04:33 PM, Joe Stringer wrote:
On 30 August 2016 at 12:49, David Hill <dh...@redhat.com> wrote:
Hello sir,

     We have a customer hitting an issue where ixgbe will not return
an error
at "SIOCGMIIPHY" but will hit it later at "SIOCGMIIREG" so instead of
failling at miimon and never doing the failover, we'd like it to use
ethtool
which is properly returning the link state.
Hm, I don't understand why ixgbe would support getting the address of
the MII PHY in use, but not to fetch the IMM_BMSR "basic mode status
register"
from it. Is this supposed to happen? Have you tried discussing this
with the upstream driver developers?

_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev

Reply via email to