Heavy iptables traffic causes reboot?

We’ve got a Wi-i.MX53 board serving as, among other things, an Ethernet bridge. Most of the network traffic we’ve had in and out of (and across) the board has been encrypted or has involved local reads/writes so traffic rates have topped out in the 4-5 MB/s range which the board seems to handle with no problems.

In the last week or so, we added a few iptables rules to make life easier in the lab that allow iptables to forward directly across the board with no encryption or other local interaction, and that traffic goes much faster - up to 10MB/s - but only briefly, then the board reboots.

It’s 100% reproducible - if we try to scp large files across the board with fast network connections all around (e.g. all relevant machines are in the same building), we’ll hit around 10MB/s for a few seconds and then the board will reboot. If we’re running traffic between buildings, or over the encrypted VPN, or anything involving local storage on the board, transfer rates are slower and there don’t seem to be any problems. So the problem appears to be related to transfer speed.

The only clue I’ve found so far that seems possibly relevant is that with high traffic, there is a huge flood of interrupts on IRQ 87 (huge ~ 30,000/second) so I’m wondering if it’s an issue with handling the rate of hardware interrupts from the FEC (eth1 does not generate a similar rate of interrupts, and we haven’t had the problem when going eth0 <-> wlan0).

There are several variables I haven’t had a chance to test yet (e.g. we’re also bonding one wired and wireless interface, though in this case the wireless interface is inactive) but I’m wondering if anyone has seen anything like this.


1 Like