In addition to not backing up the existing lilo.conf (see bug 5512), the RH 6.1 upgrade script neither asks where you want lilo installed nor looks at your old configuration. It points at the MBR (eg, /dev/hda) with no chance of setting it to the boot partition instead (eg, /dev/hda5). This rather microsoftish behavior forces a re-install of whatever other boot manager software one might already be using.
This particular bug buggered up my perfectly working lilo setup so badly I couldnt boot off the harddrive after the reboot. I think the problem was exasperated by the fact that I had a windows partition marked bootable but in fact had nothing but old data on it.
I also suffered from this. Please note that some systems *have* to boot via loadlin. The old install systems in expert mode used to request permission to actually run lilo, as opposed to merely upgrading an rpm. Summary *expert upgrade* must not assume lilo. Even when lilo is used, it is likely that an expert user has a later kernel installed with /etc/conf.modules and /etc/lilo.conf customized to match. After all 2.2.13 is already out, not to mention 2.3.*. Please revert as a matter of urgency!
I also had this problem when made upgrade from RH 6 to 6.1. I can add that upgrader (installer) also forgot to offer making boot floppy with new kernel for ver. 6.1, what would be very velcome in case when you reinstall old boot- manager (Select-it in my case), and then Lilo 0.21 (installed in boot sector of Linux part.) works not. And I don't think that it has any relation with what partition are on hard disk.
Assigned to dledford
Reassigned to installer - the lilo package doesn't break lilo.conf. The installer does.
If I remember correctly, this is one of the fixes that appeared on the updates.img for 6.1. The bug is definitely not present in the 6.2 beta