On 06/26/2009 09:19 AM, Kedar Sovani wrote:
How should we go about the multiple kernels approach:
1. create multiple kernel rpms for multiple boards? kernel-sheevaplug, kernel-beagle,
etc.?
2. create a single kernel rpm with multiple images stored within it?
3. forget the kernel rpm, let each board have its own pre-built kernel binary available?
Since anyway most of the people will probably burn the kernel separately on the flash?
The kernel does also have this thing where you can build against
multiple machine IDs in a single image, and it decides on its
personality at runtime based on the machine ID that's passed in by the
bootloader.
The problem is it would be bloated for any individual case... there's
patchsets floating around as well for many projects, some take in
out-of-tree things, the patchsets can easily conflict...
I guess 1 is the scalable way, where each variant package is maintained
by someone who builds for the device anyway, it can have whatever
patchset is needed just for that, the kernel image is no bigger than it
has to be either. It doesn't stop the board project people having their
own as well etc.
-Andy