NC-SI driver: Detect OCP module power down!

[Milton Miller II]

January 03, 2021 8:45 AM Thu Nguyen wrote and miltonm at us.ibm.com responded:

In our test platform, BMC boot up with NC-SI module is plugged and 
powered. NC-SI interface (eth0) is up and worked well.

Then the power of NC-SI module is power off. NC-SI driver can't detect 
this state.

BMC console keeps print:

[ 1780.411126] ftgmac100 1e660000.ethernet eth0: NCSI Channel 0 timed out!

[ 1785.579455] ftgmac100 1e660000.ethernet eth0: NCSI: No channel with 
link found, configuring channel 0

[ 1802.253375] ftgmac100 1e660000.ethernet eth0: NCSI Channel 0 timed out!

[ 1807.501456] ftgmac100 1e660000.ethernet eth0: NCSI: No channel with 
link found, configuring channel 0

Look at the ncs-manager.c code, it seems if a channel of NC-SI module is 
time out its' monitor_state will be disabled, the link will be down then 
the driver will go to next channel.

But if all of channels of NC-SI module are down. The driver will reset 
the monitor_state of all channels and check their states again.

Milton> The NCSI specification explicitly says that power to the nics and the controller must be synchronized.   


Do we have any mechanism to detect power off state of NC-SI module?

Milton> No, but the state machines will reset if you ifconfig down the interface.

How about hot plug NC-SI module?


Regards.

Thu Nguyen.







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ozlabs.org/pipermail/openbmc/attachments/20210103/0e3a9ff9/attachment.htm>