[U-Boot] [GIT PULL] Xilinx and SCSI changes

[Michal Simek]

On 7.12.2016 14:10, Tom Rini wrote:
> On Wed, Dec 07, 2016 at 12:21:29PM +0100, Michal Simek wrote:
>> Hi,
>>
>> On 2.12.2016 15:46, Michal Simek wrote:
>>> Hi,
>>>
>>> On 2.12.2016 15:12, Tom Rini wrote:
>>>> On Fri, Dec 02, 2016 at 02:55:00PM +0100, Michal Simek wrote:
>>>>
>>>>> Hi Tom,
>>>>>
>>>>> here are some patches I have collected for Xilinx devices, one miiphy
>>>>> patch and SCSI changes I have made.
>>>>>
>>>>> Simon: I have sent v4 of DM_SCSI + ceva sata driver moved to DM which
>>>>> should go through your tree because you have applied
>>>>> "dm: blk: Fix get_desc to return block device descriptor"
>>>>> which is required. They are based on this branch.
>>>>>
>>>>> Tom: I have tested also travis with zynq qemu but it looks like a
>>>>> problem with qemu timing for zynq model. The rest of python tests look
>>>>> good. https://travis-ci.org/michalsimek-test/u-boot/jobs/180384185
>>>>> Also for running this I had to use CONFIG_OF_EMBED=y to handle dtb.
>>>>> Anyway I will talk to Qemu guys what I can do with it and when this
>>>>> timing issue was solved and which mainline qemu version this requires.
>>>>> Is there arm64 qemu available too?
>>>>
>>>> We have to disable the 'sleep' test on a number of other platforms, you
>>>> just need to add TEST_PY_TEST_SPEC="not sleep"
>>>
>>> ok. Will try.
>>>
>>>>
>>>> For the zynq model, can we not use the -dtb argument to qemu?  Or, how
>>>> does the device tree get passed along on the real platform?
>>>
>>> I was trying it but it didn't work. Will try again next week and will
>>> refresh my mind on Zynq.
>>
>> I have looked at it and discussed with our qemu expect (Edgar) and there
>> is no easy way how to do it with ancient qemu version.
>> In Qemu 2.8 there will be loader functionality which provides an easy way.
>>
>> I was briefly checking current qemu targets and they don't use
>> OF_CONTROL/OF_SEPARATE which based on fdtdec_setup() requires to put dtb
>> to certain location and there is no way around.
>>
>> Also qemu expect elf to be loaded not binary that's why we can't use
>> u-boot-dtb.bin with appended dtb.
>>
>> The only option I use right now and I know it is working is using
>> OF_EMBED for zynq_zc702 target and I can add zynq board to travis testing.
>> The next option is use latest qemu and we can also wire zynqmp platforms.
>>
>> Please let me know what you think.
> 
> OK, I'm leaning towards another round of rework to the travis yml file.
> I think I can make it key building QEMU from sources off of if we're
> doing test.py run or not and then be smart about only doing the extra
> work there.  I'll see how much more time that adds.

Ok. I am just retesting "not sleep" option now but definitely keep in
loop on this because we are using qemu a lot and this can be added to
regression. I will also think about wiring microblaze to regression.

Thanks,
Michal