The Device Tree has been used for many years on PowerPC, and propagated to several other architectures since then, most notably the ARM architecture. Bringing a number of benefits such as a hardware representation clearly separated from the kernel code, it has also been introduced as a promise of ABI stability by its proponents.
With the Device Tree, a device manufacturer is supposed to be able to write a Device Tree to describe its platform only once, and be sure that any kernel version shipped *after* the kernel version that was originally used to test the platform will continue to work, using the same Device Tree blob.
In this talk, we would like to discuss this ABI stability requirement: does it actually work? what problems is this causing to kernel development? are Device Tree bindings sufficiently reviewed to guarantee such a stability? are SoC and board.. (too short limit)
