Hello. We have an embedded system (x86) running Linux 2.6.15 with three (3) 82574L NICs. The system is an Advantech SOM-4455Z ETX module (AMD Geode LX800 500MHz processor, 512MB DDR).
Since it is a System-On-Module in the ETX format, it has only PCI not PCIe. Therefore, we developed a custom PCI network card that has a PCI-to-PCIe bridge that connects to a PCIe switch (1to3) which connects to three 82574L chips.
Our system does not have ACPI nor APM enabled in the BIOS nor the Linux kernel.
Our system does not support MSI or MSI-X, it uses legacy interrupts.
At one of our customer sites where they have just one of the 82574L ports connected to their network (eth1), the system will send/receive traffic on eth1 normally for anywhere from a few hours to a couple weeks then it goes into a state where all RX packets are dropped. If the interface is brought down and then back up, it will begin receiving packets once again until eventually it gets back into that same state and drops all RX packets.
This problem was first noticed in version 1.1.2 of the e1000e driver. I then updated to the 2.0.0 version of the driver and the problem still occurs. I then compiled the driver without support for NAPI, but the problem still occurs. I have since built the driver with DEBUG enabled, hoping to get some useful log messages when the problem occurs, but there aren't any useful log messages displayed when the problem occurs.
I am attaching the results of 'lspci -vvv', 'lspci -s 03:00.0 -vvv -xxxx', ifconfig (which was done a few times to display the RX dropped count increasing but RX count not increasing), 'ethtool -e eth1', 'ethtool -S eth1', and also a dump of the debug log messages printed int he kernel log.
Any help here will be appreciated.