KS2010: Development process [LWN.net]

KS2010: Development process

Posted Nov 3, 2010 23:22 UTC (Wed) by vomlehn (subscriber, #45588)
Parent article: KS2010: Development process

Hearing that Linux doesn't like device trees leaves me concerned that he is less focused on the embedded world than I might like. Embedded developers can't afford to stick self-describing buses on all of our hardware and having a standard way to describe hardware is a big plus. Also, seems PCs use something rather like the device tree to pass information from the bootloader to the kernel; is there a conceptual difference to justify the dislike?

It's true we have a lot of work to do on the device tree, but the need is strong.

(Log in to post comments)

KS2010: Development process

Posted Nov 4, 2010 2:30 UTC (Thu) by neilbrown (subscriber, #359) [Link]

As I understood it, Linus doesn't like device trees primarily because discoverable buses are just soooo much better. So in overall design decisions, discoverable buses should be considered the norm and device tree is the second cousin that is used when there is no alternative.

There is no suggestion that we must not have device trees, just that they be kept clearly in their place, which is secondary to discoverable buses.

KS2010: Development process

Posted Nov 4, 2010 12:36 UTC (Thu) by etienne (subscriber, #25256) [Link]

IMHO, most of the device tree use I have seen is to define addresses of subcompoments, and I still think that is a job for the linker - having a special extra file as linker command to contain all those addresses simplify the process.
It should be possible to have late relocation for very few fields in the bootloader (at least kernel relocations is possible on ia32/amd64) if you really want a single kernel for different "fat" microcontrollers.