Device tree home

Hi Kim,

Back to the fdt/of methodology (I started a new thread since it is kind of a new topic)... I have a hot project that I have not quite gotten to yet to integrate the fdt more tightly into u-boot rather than it being some magic wedged into bootm.

The ultimate goal (for me) is to have an option where the u-boot env gets stored in a fdt rather than the current env so that the fdt really is native to u-boot. The goal would be that all the current env commands would Just Work[tm].

For compatibility with architectures that don't use the fdt (i.e. everything other than PPC :-), the current env would have to be supported and and mongrel system of both traditional env and fdt would need to be supported as well for backwards compatibility.

I've been seriously looking at David Gibson's libfdt library which looks a lot more straightforward than the kernel's implementation. (I've CC:ed him on this email since I don't believe he is subscribed to the u-boot list.) For more information, see:
<http://www.archivesat.com/Linux_on_PowerPC_Developers/thread2163413.htm>

I have not thought of any reason this won't work. I have not started implementing it yet, but it is #1 in my job jar. (That would be more meaningful if there weren't 20 other things also at #1. :-/)

Best regards,
gvb

I'm not sure I understand the benefit of this - I thought the only parts of the device tree likely to get changed on the command line are the mac addresses and the chosen node's bootargs property, which today's u-boot handles well.

My belief is merely that the device tree data should belong in the u-boot image; they are both equally board specific.

Kim

It also changes several properties: timebase-frequency, clock-frequency, and bus-frequency, although someone reported a possible bug in this code yesterday on linuxppc-embedded.

Timur Tabi
Linux Kernel Developer @ Freescale