RFC: blow Ethernet-over-USB into a module

Andy Green andy at openmoko.com
Wed Jul 30 20:56:51 CEST 2008


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Somebody in the thread at some point said:
| Sean McNeil wrote:
|> Andy Green wrote:
|>> Somebody in the thread at some point said:
|>> | Andy Green wrote:
|>> |> This patch stops making Ethernet-over-USB as a built-in and makes
|>> it a
|>> |> module instead.  If you use NFS like that you will hate it.
|>> |
|>> | Hmm, I understand the goal. Unfortunately, this also means that having
|>> | bad modules can now prevent you from ssh'ing in even more easily :-(
|>>
|>> I realize it's contentious.  You are right, it is less convenient if you
|>> were in the situation you had userspace up and got an IP allocated, etc,
|>> but your modules were broken.
|>>
|>> Personally, I don't use our dist config and use one with many more
|>> drivers as built-in anyway.  Someone could recover by DFU-ing this kind
|>> of kernel in.
|>>
|>> Bad modules is a repeated problem in the wild we need to solve anyway.
|>> And if your modules are bad, currently they can trash your kernel making
|>> an even worse problem.
|>>
|>> Is anyone using NFS over USB Ethernet?
|>>
|>> -Andy
|> I use it for development until a certain point. What you are talking
|> about, however, is the default config and that should be whatever you
|> determine is the best fit for the end users not us developers.
|
| P.S. I don't see the benefit of modules in an embedded device. You know

This is on the edge of what an embedded device is an what a weak PC is,
but that isn't the issue.

| all the drivers and functionality you need to support the device
| already. What do you gain by making them modules as opposed to compiling
| them into the kernel? initialization ordering?

Some of the drivers are mutually exclusive, like the gadgets... hey OK
we can do two but we can't do five.

I like my stuff to be built-in, but if everything we enable was built-in
it would start to get a bit crazy... for all kernel users.

I have a radical patchset coming which largely resolves init ordering
and suspend / resume ordering stuff that has been around since the
start, if I can get it finalized.

- -Andy
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org

iEYEARECAAYFAkiQuXMACgkQOjLpvpq7dMp1IQCffXJmlMj8m4k0J1gEGPz+796T
tMwAniP4NNiOHADJLAu+ws05U6xyHiR6
=fq3E
-----END PGP SIGNATURE-----




More information about the openmoko-kernel mailing list