[GIT PULL] make mach-omap2 boot with device tree only for v3.14

[Tony Lindgren]

* Russell King - ARM Linux <linux at arm.linux.org.uk> [131210 06:37]:
> On Mon, Dec 09, 2013 at 06:42:27PM -0800, Tony Lindgren wrote:
> > We can now finally make mach-omap2 to boot with device tree only and get
> > rid of over 20k lines of platform init code that way.
> > 
> > Most basic devices already work using device tree based initialization
> > and the remaining devices can be initialized using platform data
> > with pdata-quirks.c.
> > 
> > So for most missing boards it's just a question of adding a .dts file
> > that should be fairly similar to one of the existing .dts files as
> > we've tried to cover all basic omap3 board types.
> > 
> > For people getting started updating their board files to for device
> > tree, there are some basic instructions in commit 8dc8b3ddf5d7
> > (ARM: OMAP2+: Remove legacy omap3 board-*.c files and make mach-omap2
> > DT only for booting).
> > 
> > Please also note that there are also some related fixes making their way
> > into the mainline kernel that are needed for some use cases:
> > 
> > PM off-idle for omap3 and wake-up events need the following two patches:
> 
> While this is good, what seems to be happening is a complete switch
> from one method of booting to a different method of booting with no
> compatibility inbetween.  We're going from only supporting the ATAG
> booting protocol for these platforms in v3.12 to only supporting the
> DT booting protocol in v3.13.  There's no forward or backwards
> compatibility - it's one or the other.

Hmm hey this is for v3.14, not for v3.13.

The reason why I've been sending fixes all over the place is to have them
out of the way for v3.13 so we have v3.13 work reasonably the same way
with both legacy booting and device tree based booting. Of course some
.dts files will not be there for v3.13 though.

> This is bad: this is the kind of "flag day" change which Linus hates -
> and I hate it because it means I need to update my build test system
> at the same time that arm-soc merges this stuff.

Unfortunately there's no way around that unless we carry the legacy
booting setup for longer. We can certainly postpone the flip over if
if there's a real neeed to do it.

But if the issue is just building an appended DTB image, I'd rather just
get done with it as that problem we're always going to have no matter
how much we delay the flip over.

> This is utterly crap - it's crap not only from the point of view of the
> user perspective, but also from the development methodology point of
> view.

The crappy part here is the fact that we need to build the kernel with
appended DTB. Maybe there's something more we can do to make it easier.

Regards,

Tony