Hi,

I thought instead of waiting for mainline kernel support for these boards, if it would be possible to use Allwinner H6 BSP 4.9.56 kernel?

Found at https://github.com/Allwinner-Homlet/H6-BSP4.9-linux

I am able to compile it on Sunxi's Ubuntu 16.04 Beta image, however from the serial output it seems to at least have trouble with the sd card initializing.

I tried to copy over the sd card information in sdmmc@04020000 from Sunix's own /boot/orangepi/OrangePiH6.dtb into Allwinners sun50iw6p1.dtsi but the boot only went a little further ending in a similar kernel panic to when I used the the BSP4.9 with Sunxi's original OrangePiH6.dtb. I am not very skilled at this and do not think I can finish it off, if it's possible at all. I copy the dtb to the same folder with the same filename in my testing.

 

Apparently most of the drivers work, including the cpufreq which is what is notably missing in the current Armbian WPI images. The old BSP 3.10.65 kernel is next to impossible to use any usb wifi dongles for the One Plus.

 

I have attached the output of my experiments.

Cheers,

Rauli

BSP3.10.65-Sunxi-dmesg-normal-boot.txt

BSP4.9-AllwinnerH6.txt

BSP4.9-AllwinnerH6-using-stock-BSP3.10.65-dtb-kernel-panic.txt

5 hours ago, Rauli Kumpulainen said:

I am not very skilled at this and do not think I can finish it off, if it's possible at all. I copy the dtb to the same folder with the same filename in my testing.

As you figured out... This kernel doesn't work 'out of the box'.. And even when "we" solve the first issue (SD-Card) we've no clue what kind of garbage will come up after its. Due to lack of resources it's IMO unlikely that someone dives into a Allwinner BSP adventure. Nevertheless feel free to try it.

It's not done by just bring the kernel up and running, maintenance over time is IMO the harder part. For a mainline/vanilla kernel, this is 'somehow' predictable. You get proper release tags, a commit history and most important, you get some sort of documentation. For a BSP kernel, this differs a lot (e.g. your AW kernel has an initial commit and then nothing.. that's hard to get a clue what's going on there). As soon as Armbian supports a BSP kernel, people expect that this kernel is supported over a longer time. Just have a look at the H3 boards, we still support an EOL 3.4 BSP kernel, due to mainline wasn't mature enough in the past (IMO now, there isn't any reason to stick at the BSP kernel but others might have other opinions on that).

4.9 Vanilla should get EOL in jan 2019, means that after then we'll have the same problems that we have with the 3.4 Kernel. We will then support a Kernel which doesn't get any updates (despite that a BSP kernel may have problems with the Vanilla updates) and have then convince our users that they should switch to a mainline kernel which misses some of the features they had in their BSP kernel. You can either do this by drop a kernel (people will complain) or convince them that 'mainline' is mature enough for their 'pet project'. Both ways needs additional work to do. You can ask yourself if it's worth to do this process for a Kernel which gets EOL in 6 months. There aren't many people working on H6 boards for Armbian. I don't think that they're willing to split their work into 2 kernels, whereas one kernel doesn't have a commit history. 

Ok thank you for clarifying that. Let us wait for the mainline progression instead. If only I had the skills I would help contribute to its development.

1 hour ago, Rauli Kumpulainen said:

If only I had the skills I would help contribute to its development.

Learning by doing...  Nobody starts as a 'perfect' contributor doing everything right from the beginning. Read through this:

and probably this one:

H6 boards have plenty of stuff to step in. Try to fix something and then start to send patches. If you struggle somewhere report what you've tried, where you failed and maybe someone can give you some hints to solve it and maybe not, in case 42 isn't a sufficient answer.   The great thing armbian provides is the build-script which everybody can use, not only @Igor when he provides Images for a *random board*. It saves you in the beginning a lot of headache cause you don't have to deal with patch creation, applying those patches and choose a proper compiler for your tests (you'll understand the scirpt better and better as soon as you start to use it).  

Cheers for the references chwe. I will have to investigate.