The latter is more straightforward atm, but imagine the Linux kernel is equipped with GPT. So whenever you run into a problem you can simply ask it for solutions! That would help with the adoption of Linux IMO.

What features are you imagining that would otherwise be impossible in userspace?

maybe their bar isn't as high as "impossible"

You should see Linus Torvalds' bar. If you think I'm asking hardball questions, you're in for a treat.

is his bar higher than "impossible"

His bar is that he doesn't merge bad or redundant code. He's turned down major features for this before (eg. NTFS support) and would certainly do it again if someone suggested a 10,000 SLOC patch for LLM support that works fine in Ring 2.

Of course, nothing stops you from forking the kernel and doing the heavy lifting yourself. That's why I asked which features they need first, though.

Like when there’s a kernel panic or your graphics drivers are not working, you should still be able to talk to kernel and find a solution.

A kernel panic probably halts the machine for a good reason. A user space log analyzer can search for a solution after the reboot, ChatGPT or other traditional technologies.

A not working graphic driver would result in a black screen. Again, a user space log analysis after a reboot or maybe a ALT Fn?

Anyway, why baking that functionality inside the kernel and how would it interact with the user? Over the console? A user space log analyzer could detect issues before the previous shutdown or reboot and offer to find a solution.

> Like when there’s a kernel panic or your graphics drivers are not working, you should still be able to talk to kernel and find a solution.

Even if GPT was in the kernel, you wouldn’t be able to talk to it after a kernel panic.

Why do you need GPT in the kernel for this? If the relevant data is exposed to userspace, you can use whatever LLM or other tool, local or remote, to consume and interact based on the data, while not putting unnecessary heavyweight features in the kernel.