From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:0.9.2) Gecko/20010809 Description of problem: Since I lost the contents of my disks, I decided to go ahead and get the geometry of the disk changed to the one detected on a clean disk, instead of the geometry recored in the MBR by the installation of Windows 98 done by the computer supplier. The disks used to have CHS=7476/255/63. After wiping out the MBR, they were detected as CHS=119150/16/63, which lets me not waste half a megabyte and gives me more flexibility in choosing the partition boundaries. I had to force the kernel to recognize this geometry for hda, though (hda=...); it would do the right thing for hdc, hde and hdg. hdb is a CD-RW. Also, I have to tell the kernel it's ok to use UDMA5 for the two native IDE controllers of this , so my kernel command-line looks likeAsus A7V133 motherboard; the two additional on-board Promise IDE controllers are already detected as supporting UDMA5. My kernel command-line looks like: ide0=ata66 ide1=ata66 hda=119150,16,63 hdb=ide-scsi Installation proceeded fine. I created a RAID1 /boot on the first 32 MB of every disk. The installer correctly installed grub on hda, and I was happy for a moment. Then, I decided to restore the ability to boot from all other disks (the idea is that, if any of the 4 disks fail, the system can keep running, by using a combination of RAID1 and RAID5 filesystems over the 4 disks). So, I did just like I had done before. I set up the /boot/grub/device.map so that hd0 mapped to hda, hd1 to hdc, hd2 to hde and hd3 to hdg. Then, I proceeded to installing the boot loader on each disk, with `root (hd#,0)' and `setup (hd0)'. To my dismay, none of the disks were bootable after this (stupid me! I should have tested some other disk before overwriting the working hda MBR). Just after loading the boot sector, GRUB would start printing GRUB GRUB GRUB GRUB GRUB etc until I rebooted the machine. After a number of rescue-mode boots, having even considered going back to lilo, I decided to have a look at the code in Anaconda, that had succeeded in installing the boot loader with the new geometry. The difference was that, unlike setup, that ran: install /grub/stage1 d (hd3) (hd3)1+22 p (hd3,0)/grub/stage2 /grub/grub.conf Anaconda would run something like: install /grub/stage1 d (hd3) /grub/e2fs_stage1_5 p /grub/stage2 /grub/grub.conf After I ran install like Anaconda, I was able to boot with Grub again. If I were to guess, I'd say the problem has to do with the embedding of stage1.5 given the smaller number of logical heads. It doesn't make much sense to me, but it was the only difference between the working configuration and the failing one. Version-Release number of selected component (if applicable): How reproducible: Always Steps to Reproduce: 1.Install the system on an IBM Deskstar 60GXP 61.5GB with geometry 119150/16/63. 2.Boot the system. 3.Start grub, and run `root (hd0,0)' and `setup (hd0)'. 4.Reboot. 5.Go find the rescue disk. :-) Actual Results: Grub prints GRUB GRUB GRUB GRUB GRUB and doesn't boot. Expected Results: I'd hope it would just boot. Additional info:
What do you get if from the grub shell you type geometry (hd0) ?
drive 0x80: C/H/S = 53614/16/63, The number of sectors = 120103200, /dev/hda looks reasonable, as far as the 32GB limit on CHS access goes. It shouldn't affect access to the /boot partition, though. Or should it? Or should I force LBA mode enabled or disabled?
If you use lilo with lba32 in /etc/lilo.conf, do you have any problems?
Please rephrase the question bearing in mind that the problem showed up in grub, i.e., without using lilo.
Right, if you install lilo on the system and have the lba32 keyword in /etc/lilo.conf, are you able to boot? Basically, I'm trying to determine if it's the LBA32 stuff causing breakage (as I suspect) as GRUB and LILO both do it similarly
Ok, I've set up lilo with lba32, and indeed it failed to boot too. It would print an infinite sequence of `2 '.
"Great". another broken bios... guess I'll be implementing a --disable-lba32 flag for grub soon *** This bug has been marked as a duplicate of 55168 ***