The RC1 release incorrectly sets up the partition table. This results in
not being able to boot the OS after install.
When the driver loads it shows 3 partitions as expected. However, after
moving the drive to another system and running fdisk shows a corrupted
partition table. The / filesystem can be mounted but the EFI partition
Create the partitions using the Beta 1 release, then install RC1. Or, on a disk
that already has the desired partitons use disk druid to set mount points only.
Don't use fdisk. Disk Druid uses EFI GPT partitioning. This should work
fine with the firmware, AFAIK.
The partition table is corrupt using disk druid or fdisk. We tried both with
the same results. I noticed that the partition created with druid was type EFI
GPT, however it also showed me a number of blocks which greatly exceeded the
capacity of my 9GB drive.
So just using disk druid rather than fdisk does not work.
This wouldn't happen to have been attached to a smartarray in the ia64, was it?
I don't think you can do that.
Compaq's system has a ROC embedded. This even happened on a SCSI card.
Notes from one of my engineers:
I tested with a blank disk hooked to the qlogic controller. The same thing
happens. Thier installer is NOT updating the partition table on the disk.
If you run fdisk /tmp/sda just before the system reboots (this is where the OS
is installed) it said I still had an invalid partion table.
The problem is that fdisk and diskdruid are incorrectly setting up the
partition tables. The problem is seen if you use a clean disk drive that has
not been sliced up yet. If you have a drive that is already set up, then you
don't see the problem.
This worked with beta3. Workarounds are to go to a i386 system and setup
partition table with i386 software or run beta 3 to setup partitions and then
This is a must fix.
As stated before, fdisk does not read the partition table format created
by disk druid; it doesn't matter when during the install you run it. You can
read the GPT partition table with
- the kernel
- disk druid
Can you get Mike.firstname.lastname@example.org added to the permissions of this bug? Or
talk to him directly? He is the engineer with the most information for you to
get this bug resolved.
He should be on the CC: list now.
FYI: we did manage to reproduce this in-house yesterday, and are now
working towards a fix.
A fix has been made available on the beta site, and should be in the next