Host Platform: Ubuntu 24.04
DPU: Bluefield BF1500
Driver: MLNX_OFED (MLNX_OFED_LINUX-24.10-0.7.0.0-ubuntu24.04-x86_64)
Has anyone had this problem:
- Works: The /dev/rshim0/console works,
- Works: the tmfifo_net0 exists, and 192.168.100.2 can be pinged
- Problem: ssh to 192.168.100.2 sends the data sent by host ssh client to the console as input. Same for http connection to 192.168.100.2.
The problem with this beyond the obvious inability to make an ssh connection is that it totally freezes up the DPU even from pinging. Presumably the poor processor is fully engaged with processing garbage on its login: prompt.
Background:
We recently upgraded the OS on our machine from Ubuntu 20 to 24.04. Prior to the
upgrade the Bluefield adapter was was working fine. After the upgrade, it was still using Bluefield software 2.0.0.10829 but the DKMS build was failing on a header from OFED, so I decided to upgrade for something for Ubuntu 24.04. What I found was MLNX_OFED_LINUX-24.10-0.7.0.0-ubuntu24.04-x86_64 and I ran the mlnxofedinstall script and it installed things including rshim_2.1.5-0.g8dfd6ea_amd64.deb as far as I can tell, and that made the /dev/rshim0/* appear.
This replaced rshim 2.0.6-1.ga97dc5d which seems to work fine.
My DPU is still running Ubuntu 20.04, and I am wondering if what I need to do is upgrade the DPU’s OS/firmware also to match a new protocol the rshim is using?
My only hints of problem are
rshim[1301753]: Create rshim pcie-0000:65:00.2
—> rshim[1301753]: rshim0 failed to enable INTx
rshim[1301753]: pcie-0000:65:00.2 enable
rshim[1301753]: rshim0 attached
—> (rshim)[1302823]: rshim.service: Referenced but unset environment variable evaluates to an empty string: OPTIONS
Has anyone else experienced this problem?