[Buildroot] [PATCH v1 2/2] configs/zynqmp_zcu106_defconfig: transition to tarballs

Luca Ceresoli luca at lucaceresoli.net
Tue May 17 07:42:38 UTC 2022


Hi Neal, Peter,

On 13/05/22 12:48, Neal Frager wrote:
> Hi Luca,
> 
>> Hi Peter, Luca,
>>
>>
>>
>>> Le 12 mai 2022 à 21:33, Peter Korsgaard <peter at korsgaard.com> a écrit :
>>>
>>> CAUTION: This message has originated from an External Source. Please use proper judgment and caution when opening attachments, clicking links, or responding to this email.
>>>
>>>
>>>>>>>> "Neal" == Neal Frager <nealf at xilinx.com> writes:
>>>
>>> Hi,
>>>
>>>>>> I had to also pass the -m* flags in EXTRA_COMPILER_FLAGS to get it 
>>>>>> to link, presumably the petalinux toolchain has incompatible 
>>>>>> defaults. Luca, any specific reasons to not pass CFLAGS there?
>>>>>
>>>>> I'm afraid I have no real answer. :( When I initially wrote the 
>>>>> script I did it to have the same logic as petalinux. Later I didn't 
>>>>> watch for its changes. I would love to be helpful but I have no 
>>>>> access to a zynqmp board now.
>>>
>>>> Would it be possible for you to just implement the changes that 
>>>> Peter has proposed?  Peter appears to have already identified a solution.
>>>
>>>> Both Peter and I have zcu106 boards, so we can both test and verify 
>>>> the update works.
>>>
>>> Sorry, that is not true. I haven't found the magic incarnation to get 
>>> a working pmufw build with the toolchain built by crosstool-ng. My 
>>> diff was just to show what I had to change to adjust 
>>> zynqmp-pmufw-builder to use a petalinux toolchain.
>>>
>>
>> Is there a reason why we cannot change the zynqmp-pmufw-builder to download the same toolchain that petalinux uses as the crosstool-ng toolchain?
> 
>> Can you point to where the petalinux toolchain comes from?
> 
>> If petalinux uses a publicly available prebuilt toolchain, possibly with sources available, then this could be considered.
> 
>> However I suspect it does use a toolchain that is shipped by Vitis. If it is the case, then it's not an option here.
> 
>> The est option would be to fix the ct-ng toolchain of course, but I'm 
>> afraid I cannot be very helpful on this, especially without an 
>> hardware to test. :(
> 
>> Vitis and PetaLinux 2022.1 use GCC compiler 11.2.0 with the following patch set:
>> https://github.com/Xilinx/meta-xilinx/tree/xlnx-rel-v2022.1/meta-micro
>> blaze/recipes-devtools/gcc/gcc-11
> 
>> Feel free to pick your mirror site for the GCC 11.2.0 download.
> 
>> Another alternative is to just download the pre-built compiler directly from the Xilinx website:
>> https://www.xilinx.com/products/design-tools/guest-resources.html#gnu2022
>> https://www.xilinx.com/bin/public/openDownload?filename=gcc-xlnx-11_2_0-release.zip
> 
>> Then you are sure to have the exact same compiler PetaLinux and Vitis are using, and you do not need to worry about building the toolchain.
> 
> Small correction.  You still need to build the compiler after downloading it from the Xilinx release site.
> 
> What do you think?  Could we switch to the same GCC 11.2.0 compiler that PetaLinux uses?

Perhaps it could be done, however I'd like to stay on a standard open
source tool such as ct-ng.

I think these should be done, in this order:

1. try ct-ng 1.25, just released
2. you pointed to the meta-xilinx tree which has 50+ patches;
   use them in our ct-ng build and see if they fix the problem
   -> if they do, isolate to one(s) that do it
3. more options?
9. switch to building the Xilinx toolchain

-- 
Luca



More information about the buildroot mailing list