On Wed, Feb 27, 2013 at 08:25:19PM +0000, Arnd Bergmann wrote:

On Wednesday 27 February 2013, Will Deacon wrote:

quoted

As long as we remain compatible with "linux,dummy-virt" then I'm not fussed
what happens to the support code. I don't envisage adding anything else to
it that can't live under drivers/ (e.g. new virtio devices).

The idea with my patch is that you don't actually have to check anything
against "linux,dummy-virt" or another top-level "compatible value.

Great, I've got no problem killing of the short-lived mach-virt then!

Basically any device tree would boot as long as all device you want to
use (including timer, irqchip, smp CPU setup method, ...) can be brought
up using the DT information. The SMP stuff is the only bit missing
right now, any other field of the machine_desc structure is already
optional with my patch, and if all fields are NULL, you can leave
out the entire structure. The best part of this is that an empty
machine_desc is exactly what we have on arm64, so any machine that
is supported on arm64 could also be supported on arch/arm without
any machine specific code.

Keeping on the same topic: using a PSCI implementation as the default SMP
ops would extend the arm/arm64 compatibility to SMP booting, as both are
PSCI-capable.

Will