mvebu marvell espressobin fails to build, possible typo ?

[xogium]

Hi,
I've been trying to build the ATF for the marvell espressobin board. The make command I use is the following:

make DEBUG=0 USE_COHERENT_MEM=0 LOG_LEVEL=20 SECURE=0 CLOCKSPRESET=CPU_1000_DDR_800 DDR_TOPOLOGY=7 BOOTDEV=SPINOR PARTNUM=0 WTP=../a3700-utils/ PLAT=a3700 all fip

Topology 7 is one I added myself to support 2 gb ddr3 espressobin.

I get the following output:

http://ix.io/1ylY

Why is it trying to build something called DDR_TYPE=DDR3 which to my knowledge is a build flag ? I also tested by using another, already included topology (topology 2) with a clean copy of the repository, same issue.
Can anyone help with this ?
Thanks

[ghost]

CC: @kostapr

For a quick test I have removed the WTP=../a3700-utils/ part, I don't have that repository. In my machine it builds that way.

[xogium]

Edit:

Here's what it does when I have not set the WTP path. Not working either, it seems:

http://ix.io/1yo8

Hi,
I'm a bit confused now. I thought a3700-utils was a requirement to even attempt to build ATF for armada 3700 SoC family ?

Regardless, I'll try as well.

I also gave a go to the marvell ATF's repository: the last branch that worked was atf-v1.3-armada-17.10:
http://ix.io/1ymZ

Beyond that branch, 18.06 has FIP build problem, and both 18.09 and 18.12 have the exact same issue as we have here with DDR_TYPE=DDR3.

I'm not exactly sure how this work, but I suppose marvell might be backporting current ATF code into its own fork ?
I hope this helps, anyways.
Thanks !

[ghost]

Well, it is a requirement to build the final image that you flash to the board, I guess. I don't have a board to test, so I can't tell.

We regularly get patches from the Marvell maintainer, so we can say that at the very least they keep up with master and upstream their changes. I've taken a quick look at https://github.com/MarvellEmbeddedProcessors/atf-marvell and it looks like they indeed keep their own branches with their changes. I guess this is for reference.

Also, all we test in our CI is whether the firmware is built or not, we don't test with the a3700-utils tools.

[xogium]

Hello,
well as you can see by my edited comment, it isn't working without setting the wtp path here, either. It's treating clean as a file to compile, but of course there's no such file so it fails.

I also tried with several distro to make sure it was not an archlinux issue: ubuntu 18.04 and debian 9 both also do the same thing.
Thanks

[xogium]

Hi,
after struggling like crazy, I finally got it to work. I explain how if it can help people in the future, it was really something stupid:

1. I needed mv_ddr from marvell,

and 2. The build needs full path, like so:

make DEBUG=0 USE_COHERENT_MEM=0 LOG_LEVEL=20 SECURE=0 CLOCKSPRESET=CPU_1000_DDR_800 DDR_TOPOLOGY=7 BOOTDEV=SPINOR PARTNUM=0 WTP=$PWD/../a3700-utils/ MV_DDR_PATH=$PWD/../mv-ddr-marvell PLAT=a3700 all fip 2>&1 |tee build.log

Setting the paths to ../a3700-utils or similar will make the build fail, same for mv_ddr, only way found to solve that was by using full paths.

Thanks for your help, in all case, and I hope this will be useful to someone if they stumble on the same problems as I did.
:)

[ghost]

That is surprising, actually. Thanks for telling us the fix, we'll keep it in mind. I don't know if we should close this issue or not. Maybe this needs clarification in the documentation and we can use this issue to track that.