On a dual port MCP55 10de:0373 (rev a3) NIC with both ports connected, we identified a configuration that does freeze the whole NIC: having autoneg & TX pause turned on while one port is physically connected but interface is down (eg. eth1) eventually causes the whole NIC to freeze (eth1 and... eth0). This triggers TX timeouts on the UP interface and, more generally, an unreachable network.
In order to avoid the bug, all we have to do is make sure not to configure TX pause on the hardware while NIC is down. This is what the 2nd patch of the series does (details included). And, in case the NIC is in a bad state at reboot (should not happen anymore thanks to patch above), third patch basically always makes sure to fix the NIC when module is loaded. I could only test this with a MCP55 10de:0373 (rev a3) PCI device on a x86_64 host. Any feedback on these patches welcome! In particular, please let me know if this should not apply to other hardware. ############################################ # Patch Set Summary: David Decotigny (3): forcedeth: fix buffer overflow forcedeth: fix TX timeout caused by TX pause on down link forcedeth: prevent TX timeouts after reboot drivers/net/ethernet/nvidia/forcedeth.c | 16 ++++++++++++---- 1 file changed, 12 insertions(+), 4 deletions(-) -- 1.7.10.2.5.g20d7bc9 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
