I have an RM502Q-AE (FW RM502QAEAAR13A04M4G_01.203.01.203) in PCIe root complex mode with an RTL8125B, r8125 driver with the IPA offload enabled (default).
When a lot of new connections arrive at once (in my case a router failing over to the module) and there is traffic, after 20 to 50 seconds the IPA stops completely: the RTK pipes freeze, immediate commands are never acknowledged, ipacm gets stuck in D state in ipa3_send_cmd, and the WAN dies too (“data stall in UL” on rmnet_ipa0). No error in dmesg or in the ipc logs, no SSR. Only a reboot recovers it. It happens every time.
With enable_ipa_offload=0 it does not happen, but then the CPU is saturated and throughput is too low.
Is this a known issue? Is there a firmware with a fix? I can provide logs and reproduce on request.
Dear @SuperKali
Currently, there is no newer version and this project is EOL.
There is high failure rate on your side, right? Is there so many scenario about high loading?
I’m using this module as a backup connection. When my main fiber connection goes down, mwan3 switches to the backup connection provided by this module.
However, after a few packets, the driver stops working and I have to unplug and reconnect the Ethernet cable to regain access. After a few minutes, the connection goes down again and I receive the same error.
I’d really appreciate it if there were a way to fix this issue. The newer module using the r8125_ioss kernel module does not seem to have this problem.
And yes, I’m aware that this module is EOL. However, here in Italy, it doesn’t make much sense for me to buy a new module because our infrastructure is not yet ready to take advantage of the newer modules.
Dear @SuperKali
Due to there is no newer version, you can limit the connection and don’t allow a lot of connection arrive at the same time or configure enable_ipa_offload=0 .