You are not logged in.
I just got into 10g networking and I had two AQC107 based devices at hand:
1. Integrated NIC in Asrock x470 Taichi Ultimate motherboard
2. Asus XG-C100C PCie NIC.
Both had old firmware initially, I think 1.5 version.
For both of these interfaces, the same problem manifested. After being connected for a while with a short ethernet cable to a 10g switch, with reasonably heavy traffic (constant torrenting at ~200-300 mbps, maybe 2500 connections total) they would lose all connectivity. Network indicator would still say connected, but there was no traffic, not even the local gateway could be pinged. On doing ifdown/ifup it would resume working, but I noticed it would take a significantly less time before it would freeze again, once it started doing this.
I updated firmware to 3.1.121 on the integrated nic and 4.2.46 on the PCIe one.
I used the method described here: https://www.rashedtalukder.com/updating … d-drivers/
with sourcing the actual firmware from here: https://station-drivers.com/index.php/e … ang,en-gb/
(no idea how to get it from the marvell site).
After firmware update the problem persisted. I changed thermal pad and put a fan on the PCie one, to eliminate thermal issues as the cause, and used only that one to test further.
Kernels were 6.16 and 6.17 now.
Googling found lots of issues with these interfaces, and some suggestion were disabling some features like gro and lro among others.
I started disabling just the GRO, which at first seemed to fix it, but after a day and half or so it froze again. Then I also disabled LRO and it worked for a few days with no issues, still working.
There was no relevant dmesg messages except the initial one at booting, saying the interface has "suspect GRO implementation".
Any idea who can help with this ? I tried emailing someone with marvell.com that was maintainer for the aquantia, but no reply.
Offline