Hi, [Please Cc me as I am not subscribed.] We're getting lots of NETDEV WATCHDOG: ... transmit timed out when under heavy load, despite using driver version 5.2.30.1. This: http://www.intel.com/support/network/sb/cs-009865-prd38.htm says that there's some sort of issue when the card is in a PCI-X slot and doing jumbo frames, which is exactly our case. PCI-X does seem to have something to do with this: we tried putting another card on the same bus, an unrelated card that is NOT PCI-X, and the NETDEV WATCHDOG timeouts went away. I would like to ask whether there might be a way to workaround this problem until a real fix is found: is there a way to make the card non PCI-X somehow, perhaps with setpci or by tweaking the driver? More details...: NIC: 82546EB Gigabit Ethernet Controller (rev 1) (that's a dual port). Although I believe we had similar problems with other models. Machine: Dell PowerEdge 2650 and 4600. Kernel: 2.4.x, x from 20 to 25... derived from kernel.org. E1000 driver version: tried a bunch including 5.2.30.1, 5.2.22, 5.2.20, 4.4.19... tried with and without NAPI. With our current setup with the latest versions of stuff, we consistently see the WATCHDOG timeout errors start to appear after about half an hour of heavy load. In some cases the timeouts will repeatedly occur as often as one or two minutes apart. Our typical test load is file serving over NFS (UDP) plus samba. However for some reason, very regular traffic patterns don't seem to trigger the problem: uni-directional UDP or TCP streams don't do it, serving files with most operations being all reads or all writes doesn't trigger it, it seems to require a mix of operations creating a mix of heavy in and out-bound traffic. Thanks for any hints - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
