when i try to kickstart a compaq 1850r (which has the tlan ethernet chip), the dhcp request issued from 'loader' times out. the problem is the tlan driver takes ~40 seconds to initialize -- the pump dhcp request times out much earlier. i hacked the loader code and increased the timeout which fixed the problem. since kickstart is installing a system, i don't see any danger in setting large timeouts (e.g., on the order of 2 or 3 minutes). as a sanity check, the dhcp client from ISC defaults it's timeout to 60 seconds (and it's configurable). i'd like to see this change make it's way back into the red hat loader source code.
I have the same problem (and request), though I am not sure exactly where the fault lies. When using the 3c59xdriver which reports the following: 3c59x.c:v0.99H 27May00 Donald Becker http://cesdis.gsfc.nasa.gov/linux/drivers/v ortex.html eth0: 3Com 3c905B Cyclone 100baseTx at 0xec00, 00:50:04:3d:58:26, IRQ 10 8K byte-wide RAM 5:3 Rx:Tx split, autoselect/MII interface. connected via a Cisco Catalyst 2900 Series XL switch it seems that the switch takes about 45 seconds to recover (sense speed?) whenever the interface is initialised. To get dhcp to work at all I had to make the following change: *** dhcp.c Tue Aug 22 18:38:55 2000 --- /t/ctcmw/pump-0.8.2/dhcp.c Mon Aug 7 22:57:58 2000 *************** *** 1159,1166 **** return chptr; } - sleep(45); /* wait for ethernet card to autosense speed etc */ - if ((chptr = prepareRequest(&breq, s, intf->device, startTime))) { close(s); pumpDisableInterface(intf->device); --- 1159,1164 ---- This may actually be a problem with the card initialisation, as I have also noticed that there can be other problems during a boot without dhcp, suggesting that the first packet through the interface is garbled. Note that it does have something to do with the switch, as the delay is not required if I connect two computers together directly.
This appears to be a very isolated problem and we have not received other reports so it doesn't require any changes.