H3 was already quite old back when we looked. The support offered by Allwinner for old chips (like the H3) back then was basically "If you're buying a ton, sure we'll support!", which was very off-putting.
H5's mainline support is still lacking in key areas. I2C, I2S, and cpufreq are deal-breakers for our product. Of course, combined with the fact that it will soon be out of support for a not-huge-buyer like us, it is out-of-question.
The status of their latest offering, the H6, in that matrix shows how bleak it was when we were looking at H5 for our board design. It was when the latest release of Linux was 4.4. You can see how few H5 features were supported with 4.4, almost exactly like how few are supported on the H6 today.
Sure, none of these chips are new, and the older models (which are easily & cheaply accessible) have mainline support. I wouldn't expect jack from Allwinner besides their crummy kernel 3.4 branch, but that is the beauty in mainline support, you aren't limited to what Allwinner supports!
> Sure, none of these chips are new, and the older models (which are easily & cheaply accessible...
The then-older chips were not available with the future volume we needed, unless we increased our volumes a few times. We had little choice but to build on their newer chips for the volume bracket we fell into. Perhaps the situation has changed in two years.
> I wouldn't expect jack from Allwinner besides their crummy kernel 3.4 branch, but that is the beauty in mainline support, you aren't limited to what Allwinner supports!
You can see how that is off-putting, especially when other vendors (e.g., the Raspberry Pi Foundation) themselves support the latest LTS (and non-LTS) kernel with maintained patches, and upstream their patches. And their code is not utter garbage. New RPis have a much easier time getting mainlined precisely because of these reasons.
I'd have thought they'd still be better than the chip in the Pi, which last I checked wasn't available at all to anyone but the Pi Foundation (and the proprietary bootloader license didn't allow you to run it on anything non-Pi anyway).
It might be better in principle — though shitty code is only so much different from unmodifiable code — but it turned out to be the best compromise available in practice. The boards put out by the Foundation were cheap and good enough for our purpose. Of course we'd have liked to go with our own boards. We actually designed and commissioned a small batch of builds using the H5. But then we saw the software provided by Allwinner. It was so fragile, it wouldn't even boot correctly a second time. Looking at the code itself revealed it to be a nightmare. And at that time (over a period of three months) there was no software freedom to be had since the only software that would even boot at all was the shitty dump. We'd have had to wait a year or two for the code to get usable or clean, and by that time both Allwinner would have moved on from the H5 (as we were told) and our product would have been too late to market.
The licensing and supply for the RasPi were pretty off-putting to us. Sure, we aren't making a custom board, but I don't want to be held captive to a single supplier that can stop producing boards at any time, whether that is due to a lack of chips or a lack of desire.
We were afraid of that too. But Allwinner wasn't willing to give us any guarantees they'd even continue to make the H5 two years later. Nor was the H5 backwards compatible with either of its predecessors H3 and A64, giving us more reason to worry about the future.
The Raspberry Pi Foundation (and Broadcom, in contract partnership), on the other hand, had a proven track record of both continuing to manufacture RPis 1 and 2 and to maintain backwards compatibility when putting out new RPi SoCs and boards.
Perhaps the situation with Allwinner has changed now. Perhaps they're willing to support their products for longer for smaller clients. We'd have to look into that again. But, of course, without mainline support anytime soon (and the quality of their past code doesn't help a bit) it's very hard for us to actually build a good product with it.
H5's mainline support is still lacking in key areas. I2C, I2S, and cpufreq are deal-breakers for our product. Of course, combined with the fact that it will soon be out of support for a not-huge-buyer like us, it is out-of-question.
The status of their latest offering, the H6, in that matrix shows how bleak it was when we were looking at H5 for our board design. It was when the latest release of Linux was 4.4. You can see how few H5 features were supported with 4.4, almost exactly like how few are supported on the H6 today.