[U-Boot] How to support ATF on u-boot

[Andre Przywara]

Hi Andreas,

On 13/07/16 14:25, Andreas Färber wrote:
> Hi,
> 
> Am 13.07.2016 um 14:45 schrieb Andre Przywara:
>> On 13/07/16 13:27, Andreas Färber wrote:
>>> Am 20.06.2016 um 04:59 schrieb Kever Yang:
>>>>     I want to upstream a new SoC named RK3399 from Rockchip which is
>>>> AARCH64/ARMv8, we need to support Arm Trust Firmware base on U-boot.
>>>>
>>>>     Right now we are using a miniloader(just like SPL in U-boot) to load
>>>> ATF/U-boot,
>>>> and PC jump from miniloader to ATF and then to U-boot(with CPU change to
>>>> EL2 mode or nsEL1),
>>>> then U-boot load kernel/rootfs as usual.
>>>>
>>>>     The ATF support for RK3399 has already upstream
>>>>     Could you give your opinion on how to support ATF on U-boot upstream?
>>>> When I asked Simon Glass offline, he suggest if we can build ATF as part
>>>> of the
>>>> U-boot build process, perhaps with a script in U-boot tree,
>>>>
>>>> Perhaps something like:
>>>>
>>>> make rk3399_board_defconfig
>>>> make
>>>> ./scripts/build-atf-image rk3399_board
>>>> ^^ new script
>>
>> I am not sure we should trigger an ATF build on building U-Boot. In my
>> build process for the Pine64 I just point to the compiled binary and
>> leave it up to the user to take care of compiling that. ATF builds are
>> really easy and fast, for the Pine64 it's just: "make PLAT=sun50iw1p1
>> bl31" for instance and takes only a few seconds.
>>
>>>> In any case, a good README would help.
>>>
>>> I've started looking into RK3368 for my GeekBox, which raises a similar
>>> issue. Are you working on that as well or just RK3399?
>>>
>>> Personally I think that the approach the HiKey has taken is the best,
>>> i.e. decouple U-Boot from ATF and just supply a README for how to make
>>> it work with U-Boot as ATF payload.
>>
>> Interestingly ATF itself considers U-Boot a payload, as it provides its
>> own bootstrapping parts which take a similar role as U-Boot's SPL.
>> So the official ATF build process (at least for Juno) lets you specify
>> the location of the U-Boot binary to be included in their FIP image.
>>
>> OTOH, some boards (like the Pine64) only use the runtime component of
>> ATF, so including it in U-Boot makes more sense (see below). I guess
>> this is similar for Rockchip?
> 
> As a distro, like you say above, I wouldn't want to rebuild any
> components of ATF each time U-Boot changes. It also complicates modeling
> package dependencies. So I would want to build ATF and U-Boot separately
> and then combine them in a third step when building an actual board
> image. Supplying a script wouldn't hurt, while integration into `make`
> would be problematic.

I agree, that's why we do for the Pine64, with an off-tree (compiled)
tool for the moment, but with some simple mkimage -f later on.

> The other issue I see is who would take care of mirroring ATF updates
> into U-Boot. By keeping the repos separate it will be easier for users
> to follow arm-trusted-firmware.git master branch or to use forks/patches
> where necessary. Too many vendors are missing in mainline ATF, so a
> single git submodule wouldn't handle the potential diversity.
> 
> Do you have any ATF code from Allwinner that could be integrated into
> mainline? For Amlogic S905 I only have blobs unfortunately and have been
> discussing with Dan Handley how to adapt at least the mainline tools to
> handle their quirks for avoiding per-vendor fip_create tools.

Allwinner provided a tarball with ATF sources, based on the older v1.0
release. I put that into a commit and cleaned it up radically [1]. Next
step would be to do a proper upstream port, which I already started, but
stopped for the time being due to missing time and lack of priority.

As for the tools side, as mentioned we only use bl31 (at least for now)
and let (eventually) SPL do the rest. So I use "mkimage -f pine64.its"
on U-Boot build instead of building a FIP image on the ATF build.

Cheers,
Andre.

[1] https://github.com/apritzel/arm-trusted-firmware/commits/allwinner