Description of problem: Installer produces a perfectly valid system that is inaccessible to GRUB, on account of BIOS (even LBA) not being able to retrieve config file. This leads to a 3-hour turnaround time before a usable installation can be built. Version-Release number of selected component (if applicable): How reproducible: 100%. But it requires a target disc with non-Linux partition using up the first 2GB of space. This is pretty easy to do with a 40GB dis. Steps to Reproduce: 1. Use a 1999 BIOS (lots of them still around) 2. Fill first 2GB with FAT partition 3. Build Fedora Core 5, letting it take default actions about disc space assignment. Actual results: GRUB> does not boot GRUB> root (hd0,0) Error 15 (I think) Expected results: Additional info:
What did you do to fix this? I'm not sure we can query the bios to see if it is a 1999 bios as you say to work around it...
I was lazy, so to get around this, I deleted all the partitions on the disc, and created /boot first, so that it would be addressable (no matter how small the LBA field was). I recommend that, when the installer comes up with a place to put the GRUB config file, it check the block address by asking the BIOS to seek there, and see if an error comes back. This would save the unsuspecting user from having to spend three hours wading through all the build screens (and CDs) before discovering that the system would be unbootable. Thanks for the quick reply.
Moving over to anaconda. Not sure if this is doable.
Unfortunately, there's no reliable way to do this -- we used to actually try to check but there were at least as many false hits as actual helpful ones. And given that it's not a problem for modern BIOS'es (which still trigger the false hit in some cases), keeping the code around was causing more problems than help.