error 38086101 and spanning tree/port-fast

[shchoi]

Hi,
so we are using ipxe in our environment but more recently we have been running into problems with systems randomly hitting

waiting for link-up on net0......... Down (http://ipxe.org/38086101).

Now this error shows up after the server has already gotten an DHCP IP successfully but hits it after it chainloads ipxe. However this error shows up randomly and sometimes finds the link down but other times sees link up and continues. I will also note that we are seeing this more and more with newer servers overall.

Now we had done some exhaustive troubleshooting and we think we have figured out the problem but was hoping the ipxe team might test, validate, and maybe put a fix in. The problem I believe is actually caused if you are using switches and have spanning-tree on with port-fast on. Every time a pxe image is chainloaded it is actually shutting down the network interface, then turns it on and then requests for a DHCP address. The port on the switch the server is connected on sees the network interface go down and up, and due to portfast it correspondingly turns off/on the port which we actually observe by watching the port lights on the switch. The turning off and on the switch port probably takes a millisecond to 1 second to come back online but I think the chainloaded pxe images loads so fast that it actually gets to checking link-up on net0 while the switch is still resetting that port and thus sees the link as down. Hence a race condition is created and sometimes shows up based on what finishes first (is the switch still in the process of resetting the port while the ipxe loads and gets to checking link on net0).

I am hoping this makes sense and if it does I'm hoping someone can look into this and maybe add a simple sleep command for about 2 seconds before ipxe checks for link on network interfaces. Thanks.