Hi all,
is there some option to make opensm accept aka activate a link/a port that reports Rate: 10 ?
We have these in a mixed HDR/FDR fabric. It is a known firmware bug in the case of ConnectX-6, but it also happens to ConnectX-4 ports connected to HDR Quantum switches.
The physical links come up without probles, but the State never becomes Active.
My suspicion is that Opensm refuses to deal with such low band widths.
But it would be useful to make it activate the link, so we could install the mft software and try to deal with the port with these tools.
Well, the switch is a QM8790 with FW version: 27.2010.5042,
the ConnectX-4 HCA has FW version: 12.28.2006
We have gone around this issue, though, by connecting the problematic old HCAs to an FDR switch.
What worries me still is that a number of FDR-servers with similar or older firmware states (also ConnectX-3) do not show this behavior - under reboots and resets they have always come up with Rate: 56, although being connected to Quantum HDR switches.
So I expect this phenomenon to appear with these servers only when there is a crash, late in the night, on a weekend, when you would wish for everything to come up automatically instead of having to re-cable the machine…