directdisk images - replace the image type with a script?

directdisk images - replace the image type with a script?

[Darren Hart]

I recently discovered that the directdisk images are malformed:

http://bugzilla.pokylinux.org/show_bug.cgi?id=1028

While reviewing the directdisk recipe, I'm wondering if this image type
has any real usage model. These are x86 specific. They partition a file
image so when written to a real disk you get goofy partition tables
(partitions do not end on cylinder boundaries).

The current recipes mangle the .cfg file (writing binary data over the
text file - I'm guessing the dd commands do this).

We could patch this up, but is it worth it? All the other systems create
ext3 images which are then installed on a disk per the instructions in
the README.hardware document. The live image provides a simple
dd-to-disk-and-boot solution. It seems to me the directdisk idea could
be replaced with a script "mksyslinuximage.sh" or similar that took the
image and kernel to use and the drive to partition, install syslinux on,
and copy the filesystem to.

Thoughts?

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

Re: directdisk images - replace the image type with a script?

[Richard Purdie]

On Wed, 2011-05-11 at 16:20 -0700, Darren Hart wrote:
> I recently discovered that the directdisk images are malformed:
> 
> http://bugzilla.pokylinux.org/show_bug.cgi?id=1028
> 
> While reviewing the directdisk recipe, I'm wondering if this image type
> has any real usage model. These are x86 specific. They partition a file
> image so when written to a real disk you get goofy partition tables
> (partitions do not end on cylinder boundaries).
> 
> The current recipes mangle the .cfg file (writing binary data over the
> text file - I'm guessing the dd commands do this).
> 
> We could patch this up, but is it worth it? All the other systems create
> ext3 images which are then installed on a disk per the instructions in
> the README.hardware document. The live image provides a simple
> dd-to-disk-and-boot solution. It seems to me the directdisk idea could
> be replaced with a script "mksyslinuximage.sh" or similar that took the
> image and kernel to use and the drive to partition, install syslinux on,
> and copy the filesystem to.
> 
> Thoughts?

We did get requests for that image type but I agree the implementation
is flawed. The trouble with the ext3 approach is that you depend on the
user having a script, syslinux installed, maybe knowing the location of
some syslinux files (vary by distro?), a set of tools, possibly a kernel
and that they get all the commands/parameters to the script right.

I strongly dislike the way its currently implemented since it should be
an image type, not a separate recipe (the same complaint applies to
-live images).

Since you've looked at it, is it possible to fix it or are there
variables that can only be determined when you know the final disk
geometry?

The trouble with -live images is that you have to write them somewhere
secondary to make them persistent or you hit the overhead of ramdisks.
Its a fairly ugly format and not what some end users want.

Cheers,

Richard

Re: directdisk images - replace the image type with a script?

[Darren Hart]

On 05/12/2011 02:20 AM, Richard Purdie wrote:
> On Wed, 2011-05-11 at 16:20 -0700, Darren Hart wrote:
>> I recently discovered that the directdisk images are malformed:
>>
>> http://bugzilla.pokylinux.org/show_bug.cgi?id=1028
>>
>> While reviewing the directdisk recipe, I'm wondering if this image type
>> has any real usage model. These are x86 specific. They partition a file
>> image so when written to a real disk you get goofy partition tables
>> (partitions do not end on cylinder boundaries).
>>
>> The current recipes mangle the .cfg file (writing binary data over the
>> text file - I'm guessing the dd commands do this).
>>
>> We could patch this up, but is it worth it? All the other systems create
>> ext3 images which are then installed on a disk per the instructions in
>> the README.hardware document. The live image provides a simple
>> dd-to-disk-and-boot solution. It seems to me the directdisk idea could
>> be replaced with a script "mksyslinuximage.sh" or similar that took the
>> image and kernel to use and the drive to partition, install syslinux on,
>> and copy the filesystem to.
>>
>> Thoughts?
> 
> We did get requests for that image type but I agree the implementation
> is flawed. The trouble with the ext3 approach is that you depend on the
> user having a script, syslinux installed, maybe knowing the location of
> some syslinux files (vary by distro?), a set of tools, possibly a kernel
> and that they get all the commands/parameters to the script right.
> 
> I strongly dislike the way its currently implemented since it should be
> an image type, not a separate recipe (the same complaint applies to
> -live images).
> 
> Since you've looked at it, is it possible to fix it or are there
> variables that can only be determined when you know the final disk
> geometry?

I could spend some time looking at it. I was hoping to get an idea first
if it even made sense to keep the image type at all. I still believe it
doesn't make sense conceptually and people wanting such images are
better off using a script to write it to a real disk and then creating a
file image of that disk that they can reuse to create identical images
on identical disks.

> 
> The trouble with -live images is that you have to write them somewhere
> secondary to make them persistent or you hit the overhead of ramdisks.
> Its a fairly ugly format and not what some end users want.

I see it as a quick test image. Do others use it for anything else?

> 
> Cheers,
> 
> Richard
> 
> 

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel