Re: [Etherboot-developers] [COMMIT] 5.1.2+ boot from disk.

[<ebi...@ln...>]

ke...@us... (Ken Yap) writes:

> >Was it a compressed rom image?
> 
> Yes.
> 
> >At the moment I would only expect non-compressed images which still
> >run at RELOCADDR to work reliably.  The compressed case relies on
> >being relocatable, and decompreses from RELOCADDR to a little further
> >down.   
> 
> Ok, a .rom image works, but unfortunately it can't load the NFL menu
> which uses 0x20000 also. Which would also be a problem for zImage
> kernels. I thought you were going to relocate Etherboot to the top of
> the 32-bit address space?

Right the code is there to do that but it takes a while to weed
out the dependencies on code that uses absolute addresses.  0x20000 is
just the initial load address. 

If you specify -DRELOCATE etherboot will relocate itself up to the top
of memory.  I have just left the default set so people can build
drivers and then see if there are relocation fixes that must be
handled.

Relocation fixes.  Mostly it is a matter of adding:
virt_to_bus, bus_to_virt, and ioremap in the drivers.
In some of the other code calling  phys_to_virt, and virt_to_phys
are appropriate.

For the most part the pointers just need to be fixed up and the code
will work.  Straight forward but work that needs to be done a small
piece at a time.

We should probably compile a list of drivers that have not
yet been converted to work when relocated.  Then walk through fixing
and testing them.  I'd like to add the ability to test multicast at
the same time so my next generic step will be to get the multicast
sever going.  Then I can convert as many drivers as possible, but I
don't have a lot of the hardware so there are a lot of cases I can't
test.

If you can think of a better way to handle it.  I am all ears.

Eric