The Illumos' family is an interesting one. I wish it was easier to get it installed on modern hardware. Any of my attempts with distros like OpenIndiana, Tribblix and OmniOS didn't go further than the boot menu.

I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability. Maybe the Oxide folks know more of what would be involved in such an effort.

▲

JdeBP 7 days ago | parent | next [-]

That's the wrong solution to the problem, as it means learning a second kernel and all of its stuff and making compatibility shims, whilst still facing the real problem that is not software at all.

The right solution is actually explained the headlined WWW site, where Peter Tribble points out (in the About page and in the Use guide) that the significant constraint is that xe does not own the actual physical hardware to develop against. It's the usual story with small projects: good donated hardware, and developer time (and workspace, and food, and water, and housing, and electricity supply (-:), needed.

	▲	gr4vityWall 6 days ago \| parent [-]
		I agree with you in principle, and I understand the existing constraint. But I don't see developer time scaling up without those operating systems reaching (again) critical mass, and then having upstream wanting to write drivers for even more operating systems. Which is fine per se. But I wish people could buy relatively new hardware (say, last 3 years) and be able to run FreeBSD or Illumos with most things working.

▲

sunshowers 6 days ago | parent | prev | next [-]

I run our (Oxide's) Helios distribution on a headless Ryzen 7900x with an ASRock Rack B650D4U motherboard. I use an Intel X520 NIC for 10GbE.

In general I'd expect Zen 3 and 4 to work, and Zen 5 either right now or very soon.

See https://github.com/oxidecomputer/helios-engvm

▲

gr4vityWall 6 days ago | parent [-]

Thanks for the reply, I appreciate the link! Do you know if people are using it for workstations?

▲

sunshowers 6 days ago | parent [-]

Not entirely sure what you mean (do you mean Threadripper, which some of use?), but this is purely a work machine for me. I regularly build and test our software on my Helios box.

It isn't exactly the same as our production machines (as you can imagine we have shared environments for this), but we've written in-memory simulators for most of our hardware components so it's good enough for most use cases.

▲

gr4vityWall 6 days ago | parent [-]

Oh my bad, I meant "workstation" as in, a full desktop with a working desktop environment like XFCE, audio support, GPU acceleration for AMD iGPUs, etc.

Thanks again for the informative reply.

	▲	sunshowers 6 days ago \| parent [-]
		Ah gotcha! I asked internally, and the initial answer is "no", but if someone says yes I'll let you know.

▲

yjftsjthsd-h 6 days ago | parent | prev | next [-]

> I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability.

It might be easier to take them from NetBSD; it wouldn't introduce the GPL licensing issue, and courtesy of their rump kernel system they're actually kind of designed for it.

	▲	yellowapple 6 days ago \| parent [-]
		Or OpenBSD, for at least one of those two reasons. I know Haiku reuses OpenBSD driver code in some places, and I'm pretty sure 9front does, too.

▲

ori_b 6 days ago | parent | prev [-]

> I wonder how far a compatibility layer for Linux drivers could go to help other UNIX kernels' usability

This isn't very interesting. It means you're constrained to the Linux design decisions, and you're wasting time debugging mismatches and poor design decisions.

▲

msgodel 6 days ago | parent [-]

It also creates a licensing issue. Part of the reason Linux has so many drivers is because the GPL requires all of the kernel code be provided to users under the GPL as well. If you link to that in your OS/distribution now all of your code must be published in the same way.

▲

gr4vityWall 6 days ago | parent [-]

Can't an hypothetical cross platform driver be MIT licensed and be part of both?

Although I'd be more than happy with other OSes/distributions defaulting to GPL.

	▲	yjftsjthsd-h 6 days ago \| parent [-]
		> Can't an hypothetical cross platform driver be MIT licensed and be part of both? Yes, but if the goal is to reuse existing code that already exists in Linux then there's no reason to expect it to be under a permissive license. Some of it actually already is if I recall correctly, but you shouldn't expect that. > Although I'd be more than happy with other OSes/distributions defaulting to GPL. Illumos isn't GPL and and in the legal views of many people can't be compatible with it; CDDL/GPL (in)compatibility is a very long running issue, usually in the other direction with CDDL ZFS drivers in the GPL Linux kernel. (IANAL; I'm not asserting specifically that it is or is not actually compatible, just that a lot of people consider it to be incompatible)