[U-Boot] [PATCH v3 2/5] board support of arm64

[Stuart Yoder]

On Thu, Aug 22, 2013 at 11:23 AM, Scott Wood <scottwood at freescale.com> wrote:
> On Thu, 2013-08-22 at 11:15 -0500, Stuart Yoder wrote:
>> On Mon, Aug 19, 2013 at 2:59 PM, Scott Wood <scottwood at freescale.com> wrote:
>> > On Fri, 2013-08-16 at 09:14 -0500, Stuart Yoder wrote:
>> >> On Thu, Aug 15, 2013 at 8:47 AM,  <fenghua at phytium.com.cn> wrote:
>> >> > From: David Feng <fenghua at phytium.com.cn>
>> >> >
>> >> > This patch provide u-boot with arm64 support. Currently, it works on
>> >> > Foundation Model for armv8 or Fast Model for armv8.
>> >> >
>> >> > Signed-off-by: David Feng <fenghua at phytium.com.cn>
>> >> > ---
>> >> > Changes for v3:
>> >> >     - rewrite cache.S and exception.S that partly originated from linux kernel,
>> >> >       so the license should be ok.
>> >> >
>> >> >  board/armltd/dts/vexpress64.dts      |  215 ++++++++++++++++++++++++++++++++++
>> >>
>> >> Why is the device tree source in u-boot (instead of in the kernel)?
>> >> Is this temporary?   It
>> >> looks like this device tree is just a copy from somewhere else.
>> >>
>> >> Would suggest removing this from this patch series and keep the dts maintained
>> >> in the Linux kernel.
>> >
>> > U-Boot itself uses the device tree (not just to patch up for Linux) on
>> > some targets.
>> >
>> > Even with the way PPC uses device trees, it doesn't really make sense to
>> > keep them in the kernel given that they're meant to be OS-neutral, and
>> > have ties to U-Boot in terms of what gets fixed up at runtime.
>>
>> It may not make sense, but that is where they are kept currently.
>
> For PPC.

$ find arch/arm/boot/dts | wc -l
425
$ find arch/powerpc/boot/dts | wc -l
315

There are also a handful of device trees under arch/arm64/boot/dts,
including what looks like a vexpress board.

>> It doesn't make sense to maintain 2 copies of a vexpress64.dts device tree in 2 different
>> places...or to maintain 1 lone device tree in u-boot.
>
> Why does it not make sense for there to be one lone device tree in
> U-Boot?

It doesn't make sense to me to keep one device tree in u-boot
and the rest in the kernel.

I don't know where that vexpress64.dts came from, but I'm guessing
it's a copy from the Linux kernel.

> A device tree that is not used with U-Boot may not look the same, since
> U-Boot is (at least on some platforms) responsible for filling in parts
> of the tree, and (again on some platforms) for setting up the address
> map.
>
>> Maybe we need a git repo for device trees that could be included
>> in Linux, u-boot, and other things a submodule.
>
> Submodules can be a pain.  If we don't use them for DTC, why would we
> use them for this?  Since they require extra commands, you'd be
> modifying the workflow of everyone that builds U-Boot and/or Linux for
> affected platforms.

You shouldn't need device trees for building u-boot or the kernel.
I don't think a couple of extra commands is that burdensome.

I agree the DTS files really don't belong in the kernel, but there is
currently no better repository that has been proposed.   I'm not
sure u-boot is a better place.    Device trees should be independent
of any particular bootloader or OS.

Stuart