On 10. 12. 19 16:30, Tom Rini wrote:
> On Tue, Dec 10, 2019 at 04:24:09PM +0100, Michal Simek wrote:
>> On 10. 12. 19 14:56, Tom Rini wrote:
>>> On Tue, Dec 10, 2019 at 01:40:42PM +0100, Michal Simek wrote:
>>>> Hi Tom,
>>>>
>>>> On 09. 12. 19 16:19, Michal Simek wrote:
>>>>> Hi,
>>>>>
>>>>> over years a lot of new Xilinx ZynqMP board have been added to U-Boot with
>>>>> corresponding defconfigs. Also a lot of drivers have been moved to DM that
>>>>> we can make one generic configuration with one defconfig.
>>>>> Nand still needs to be validated that's why dc2/dc3 are not moved yet.
>>>>>
>>>>> Boards can be build like this:
>>>>> export DEVICE_TREE="avnet-ultra96-rev1"
>>>>> make xilinx_zynqmp_virt_defconfig
>>>>> make -j
>>>>>
>>>>> Series depends on patches sent before that's why here is full tree:
>>>>> https://github.com/michalsimek/u-boot/tree/20191209-mainline
>>>>
>>>> This series will require some changes in CI loops because right now
>>>> I didn't setup default device tree (CONFIG_DEFAULT_DEVICE_TREE) to
>>>> "force" everybody to properly pass DEVICE_TREE via command line.
>>>> I can't use OF_BOARD because then SPL is built without DT at all.
>>>>
>>>> How do you think I should handle it for CI loops?
>>>> 1. I can remove this configuration but it will be only one at the end
>>>> that's why not a good solution.
>>>> 2. Add generic option to run some commands before tests like export
>>>> DEVICE_TREE above
>>>> 3. provide different options for SPL/full u-boot how
>>>> OF_SEPARATE/OF_BOARD is handled.
>>>
>>> So, for CI are you wanting to test most device trees, or just one?
>>
>> All zynqmp dtses are built by default.
> 
> Right, but for what I thought you're saying the real use is, you pass
> just a single device tree, right?  If so, do you think we should loop
> over each, or just build one?
> 
>>>  Are
>>> we able to run one of these device trees via QEMU? 
>>
>> zynqmp is covered just by buildman not by pytest. I have this on my todo
>> list for some time but there will be other issues with mainline qemu to
>> do so.
> 
> OK, so something to improve for the future, and after we handle this
> "today" problem.
> 
>>> If we can run the
>>> virt defconfig via qemu, we should just update/extend that stanza in the
>>> CI files to set DEVICE_TREE and exclude building it from the normal
>>> jobs.
>>
>> Based on next generation Xilinx Versal where we use OF_BOARD qemu is
>> generated DT for model self to ensure that only stuff which are emulated
>> are provided to SW. Logic for dt selection is quite simply.
>> https://github.com/Xilinx/u-boot-xlnx/blob/master/arch/arm/mach-versal/cpu.c#L112
>> But Versal is not using SPL and SPL needs initial DT. Also standard
>> Xilinx boot flow on zynq/zynqmp is not using SPL and SPL is community
>> driven effort.
>>
>> At the end of the day I would like to use the same functionality across
>> boards. It means full u-boot should check one fixed location for DT
>> first with priority. For this OF_SEPARATE can be also used because
>> board_fdt_blob_setup can be overwritten for these cases too.
>> https://gitlab.denx.de/u-boot/u-boot/blob/master/lib/fdtdec.c#L1190
>>
>>>  If we can't run via qemu then yes, something like option #2 and
>>> we set DEVICE_TREE in one job and export it if set in the build job.
>>
>> It means for qemu there is no real need to build dts from source tree at
>> all.
>> Let me look at #2 for CI.
> 
> OK, thanks!
> 

I have look at it again and there are other strange things. When you
select CONFIG_SPL_FIT_GENERATOR then there is no reason to generate
u-boot.img (because you want to replace this by your generator) or
append dtb behind u-boot.bin (because default option is already setup in
fit format).

  CAT     u-boot-dtb.bin
  MKIMAGE u-boot.img
  MKIMAGE u-boot-dtb.img
  COPY    u-boot.bin
  LD      u-boot.elf
  MKIMAGE u-boot.itb



Regarding travis. just c&p. What about just this?

diff --git a/.travis.yml b/.travis.yml
index 5da046ca7ed5..4cef12e18370 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -118,6 +118,9 @@ script:
  # use clang only do one configuration.
  - if [[ "${BUILDMAN}" != "" ]]; then
      ret=0;
+     if [[ "${ENV}" != "" ]]; then
+        export eval `echo ${ENV}`;
+     fi;
      tools/buildman/buildman -P -E ${BUILDMAN} ${OVERRIDE}|| ret=$?;
      if [[ $ret -ne 0 && $ret -ne 129 ]]; then
        tools/buildman/buildman -sdeP ${BUILDMAN};

And then job like this
    - name: "buildman ZynqMP Virt"
      env:
        - ENV="DEVICE_TREE=zynqmp-zcu100-revC"
        - BUILDMAN="xilinx_zynqmp_virt"

Or do you see any smarter way how this can be done?

Thanks,
Michal


Reply via email to