[parted-devel] Using parted to create a loopback disk with cylinders of a non-default size.

[Mills, James]

On Mon, Oct 11, 2010 at 1:01 PM, Brian C. Lane <bcl at redhat.com> wrote:
> On Mon, Oct 11, 2010 at 12:02:22PM -0500, Mills, James wrote:
>>
>> Replying to my own post, I have a bit more information around this
>> behavior.  Hopefully it will help someone on the list point me in the
>> right direction.
>>
>> I have 4 machines:
>>
>> Ubuntu 9.10 x86 running parted 1.8.8.1.159-1e0e
>> Ubuntu 10.04 amd64 running parted 2.2
>> Ubuntu 10.04 x86 running parted 2.2
>> Fedora Core 13 x86 running parted 1.9
>
> The Fedora version should be v2.1

It was.  I downgraded it to test a >1.88<2.x version.  Sorry for the confusion.

>
>>
>> On all four boxes, I am using the following steps (as root)
>>
>> losetup -d /dev/loop1
>> dd if=/dev/zero of=test bs=1M count=100
>> losetup /dev/loop1 test
>> parted  -s /dev/loop1 mklabel msdos
>>
>> At this point, I run both "fdisk -l /dev/loop1" and "parted /dev/loop1
>> -s unit chs p"
>>
>> Version 1.8.8 reports the same CHS numbers using both fdisk and parted
>> (12,255,63). All other versions report:
>>
>> fdisk - 12,255,63
>> parted - 1600,4,32
>>
>> Can anyone help me understand what has changed that would cause me to
>> get such disparate behavior?
>
> When using a file or loop device the heads and sectors are set to 4,32
> and cylinders is set based on the length using these defaults.
>
> See init_file() in libparted/arch/linux.c
>
> As for why this is different, I have no idea. When dealing with real
> devices it appears to set it to 255,63 as a default, unless the kernel
> returns useful info with HDIO_GETGEO ioctl call (which according to the
> comments isn't useful with 2.6.x kernels).

Interesting.  I wonder what prompted the change, as the loopback
behavior in 1.8.8 *seemed* to work more like an actual device.

>
> See _device_probe_geometry() in libparted/arch/linux.c
>
> Does C,H,S really matter for any real usage, or is this just cosmetic?

Unfortunately it does matter for this particular application.I am
working with a piece of embedded hardware that requires a 2-stage
bootloader, and the first stage is basically hardcoded to expecting
that particular HS configuration.  Anything else, and the machine
won't boot. :(

Thanks for the information!  I'm guessing it is probably time to
investigate integrating another disk partitioning/labeling tool into
the image creation tools we are using...

~james

>
> --
> Brian C. Lane / Anaconda Team
> Port Orchard, WA (PST8PDT)

I miss Port Orchard! I lived in Bremerton for several years... :)

>