ARM has such a mess. They don't actually make chips. And they don't make a sizable % of the design that goes into chips. There's all kinds of other IP, from power management to USB controllers, all manners of things they don't do.
So then the task falls to the chip makers. Who each are trying to figure out what to do themselves.
These folks don't usually want to do that or have the chops. There's usually one or many different folks taking off the shelf open source and porting it to this particular platform. These folks have a strong strong anti-incentive to do the right thing to upstream support: no one's gonna keep buying sdk's from these software vendors if they upstream support. And it is a pain to upstream support, to spend possibly years figuring out the long term way to do something.
(Notably some good players who focus on mainlining have emerged: Bootlin, Collabora.)
It's all so terrible. Theres some attempts to mature the platform, to make some standards so at least folks can boot something maybe (SystemReady and an array of neighboring acronyms). But man, it's so bad, your question is so searing, so obvious. This whole world systematically seems unable to do the right obvious good thing for itself, has resolutely remained a shitty backwater for 3+ decades versus x86.
I wanted to add an exception to this: MNT Research.
These guys are always doing the right thing, and they care so much about it that they even go as far as designing their own keyboards, touchpads, usb controllers and other things that would contribute to the amount of proprietary blobs otherwise.
Though their Rockchip based platform is more a low end performance product at the moment. I really hope they can compete with high end x86 laptops in the future because they're the good ones.
> to make some standards so at least folks can boot something maybe
There's still no incentive to do this outside of server hardware (Red Hat doesn't support rebuilds of RHEL). Why bother making a firmware stack that allows anything to boot when you can just modify the Linux kernel to work on your non-standard hardware and work around any bugs, publish an Ubuntu image that you'll only update a few times and then call it a day. Then, maybe, just maybe, support for your SBC will be upstreamed a few years later, but long after it's actually useful.
Look how messed up and fragmented Android is still, even Windows Phone had a UEFI stack that in theory could allow all Windows phones of a certain architecture to have a shared image (although I don't believe this was fully achieved).
Imagine if Windows or Fedora for x64 had to be repacked for every single motherboard/CPU combination, it'd be insane, yet somehow it's fine for ARM, what a joke it's been allowed to become. I think electronic waste regulation is going to be needed to sort it out.
The Radxa Orion O6 we're talking about is not server hardware, but they did implement ServerReady UEFI firmware anyway, and it does allow anything to boot. In the article they booted unmodified, upstream Windows and Linux distro ISOs.
Someone told me once, yes we set breakpoint on the CPU, doesn't do anything to the AMBA bus on SoC it's a peripheral you know, the bus clock keeps going and DMA keeps going in and out if you stop the CPU, that's how we debug stuffs young man.
So then the task falls to the chip makers. Who each are trying to figure out what to do themselves.
These folks don't usually want to do that or have the chops. There's usually one or many different folks taking off the shelf open source and porting it to this particular platform. These folks have a strong strong anti-incentive to do the right thing to upstream support: no one's gonna keep buying sdk's from these software vendors if they upstream support. And it is a pain to upstream support, to spend possibly years figuring out the long term way to do something.
(Notably some good players who focus on mainlining have emerged: Bootlin, Collabora.)
It's all so terrible. Theres some attempts to mature the platform, to make some standards so at least folks can boot something maybe (SystemReady and an array of neighboring acronyms). But man, it's so bad, your question is so searing, so obvious. This whole world systematically seems unable to do the right obvious good thing for itself, has resolutely remained a shitty backwater for 3+ decades versus x86.