On 05/19/2010 08:18 PM, David Gibson wrote:
On Wed, May 19, 2010 at 07:03:17PM -0500, Timur Tabi wrote:
On Wed, May 19, 2010 at 5:44 PM, Benjamin Herrenschmidt
The padding in the kernel built is intended to
make space for DT changes done by the zImage wrapper.
Well, okay. I think it would be nice if we expanded that to handle
general usage.
Don't we have dynamic expansion in the zImage wrapper?
Maybe we could add to libfdt a way to provide a realloc() callback to it
when it hits the max size, and uboot can then move things around (or
fail).
The problem is that the code which allocates a block for the fdt is
completely distinct from the code that manipulates the fdt. We'd need
to put in either some kind of funky callback mechanism, or insist that
every fdt exist in a block of memory allocated by some specific method
(e.g. lmb).
I'm stuck between a rock and a hard place, it seems. No one is
willing to compromise on any of my ideas. It's hard to convince our
BSP developers that they should be pushing more code upstream when I
get so much resistance for a such a mundane change.
Couldn't you use the configurable padding, but put the stuff to do it
into u-boot. i.e. repad the dtb at u-boot build time, rather than
u-boot runtime.
Currently generating a dtb is not done as part of the u-boot build
process -- and apparently some people are using the Linux makefile
target to do it.
I think the right thing is to do whatever we need to do in u-boot to get
dynamic expansion fully working (it already does it in some
circumstances, just not always, or necessarily safely). Static padding
is too fragile.
-Scott
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev
