[coyotos-dev] Drop multiboot module support?

[Jonathan S. Shapiro]

On Wed, 2008-03-05 at 08:17 -0500, Christopher Nelson wrote:
> The one use I have had for them involves loading user-space drivers at
> boot time.  Probably you have some other method to do this, but it
> worked well to load them this way for me.  Especially the disk driver.
> It avoids chicken and the egg problems.

That is how we intended to use it as well. Note, however, that the
initrd (driver set) image and the kernel image must share a common
version number, so the two are closely tied. The main advantage to
separating initrd is that you can rebuild the initrd file without
touching the kernel file.

I think this can be done sufficiently well with coymerge. Let me explain
how coymerge works, and let's see if you agree:

1. At kernel link time, we perform a full (static) link of the kernel
image to check for any unresolved symbols. This is done for debugging
only. Having done it, we then perform a relocatable link (using ld -r).
The resulting kernel image is a .o rather than an executable.

2. We bundle the mkimage.out file into a loadable section in a second .o

3. When we are ready to merge them, we simply link these two .o files to
form a combined image. The resulting image is relocated, and is directly
executable.


So the main difference seems to be the last step where we combine the
files. From a functional perspective, I don't see anything that the
initrd mechanism can do that we cannot do here.


What do you think?


shap