[RFC PATCH 0/4] Add device-tree support to kexec-tools for ARM

[Simon Horman]

On Thu, Sep 06, 2012 at 10:14:06AM +0100, Will Deacon wrote:
> On Thu, Sep 06, 2012 at 10:09:14AM +0100, Simon Horman wrote:
> > Hi Will,
> > 
> > On Thu, Sep 06, 2012 at 10:01:28AM +0100, Will Deacon wrote:
> > > Hi guys,
> > > 
> > > On Thu, Sep 06, 2012 at 04:29:02AM +0100, Simon Horman wrote:
> > > > On Wed, Sep 05, 2012 at 03:34:09PM +0100, Matthew Leach wrote:
> > > > > Also, I use a
> > > > > different segment for the dtb rather than appending it to the
> > > > > zImage; I think this approach would be better as it is less
> > > > > restrictive, however a kernel patch is required to set r2 to the
> > > > > appropriate address on entry to the new kernel. What are your
> > > > > thoughts?
> > > > 
> > > > I would prefer to avoid requiring kernel changes unless necessary -
> > > > the kernels some of the boards I work with require DT since 3.5.
> > > > However, I am happy to discuss this further, there certainly is
> > > > merit to a clean implementation.
> > > 
> > > I had a quick look at both the approaches and it looks like Matthew requires
> > > changes to the host kernel (to load the dtb correctly) and Simon requires
> > > changes to the target kernel (to pick up the dtb correctly).
> > 
> > Could you explain a little why you feel my patch requires a change to
> > the target kernel?
> 
> Sure, I may be misunderstanding something, so please shout if that's the
> case! Anyway, I was under the impression that you required
> CONFIG_ARM_APPENDED_DTB to be enabled in the target so that it can pick up
> the new DT. If that's not the case, then you'll need to change the host
> kernel to fix up r2 (rather than point it at the KEXEC_ARM_ATAGS_OFFSET,
> which is too small for a dtb).

Thanks. It was not my intention that  CONFIG_ARM_APPENDED_DTB is required,
though due to limitations of the bootloader on the board I was using I did
have that option enabled for the host kernel.

Thanks for pointing out the size issue with using KEXEC_ARM_ATAGS_OFFSET,
I had overlooked that.

I think the best way forward, as both you and Matt suggested earlier,
is to modify the host kernel.