Would Intel Spec X86S affect (future) ability to run OS/2 in VM?

[50ShadesOfDirt]

This spec appears to be saying that 16- and 32-bit stuff will be removed from a future version of Intel CPU's. I currently run ArcaOS (OS/2) in a VM under VirtualBox 7.0.x, and it runs perfectly today (& thanks for such support!).

But, what about "tomorrow"? Is there any consideration for what to do for older 32-bit OS's, if we happen to be running an x64 HostOS (Win, Linux) on top of one of these new CPU's, and still want to access old 32-bit OS's like OS/2 in a VM, via VirtualBox?

Who knows when such a CPU would make an appearance, but I'm wondering if there is some "easy" way to answer the folks who now say "OS/2 is finally dead", because virtualization won't be able to run 32-bit OS's (due to this spec).

Intel Spec X86-S: "Envisioning a Simplified Intel Architecture"

I've looked over the Advanced Topics chapter, but can't figure out what piece of the internals would (or might be reworked to) handle the above case.

Thanks!

[fth0]

According to the Intel specification, "Guest is always in long mode" (3.14.3 VMCS Entry Controls). In consequence, (faster) hardware virtualization cannot execute 32-bit guests. But it should be possible to use (slower) hardware emulation instead, and OS/2 might be fast enough for that.

[javispedro]

The situation would be closer to what it was before IntelVT was introduced, i.e. early VMware/VirtualBox versions, where everything 32-bit protected mode user space runs "almost unmodified" directly on the host, but everything else such as 32-bit kernel code and practically all 16-bit code would now need to go back to emulation.

I hope not a lot of the pre-VT code was removed... because it's not an easy task otherwise, and while speed most likely will be fine, certainly compatibility will suffer. I remember VirtualBox was one of the few, if not the only, emulator which could run OS/2 versions back then.

[fth0]

Today, one could try it with QEMU without KVM.

[javispedro]

qemu without kvm would do the trick, but it would not be the same.

I remember one of the BIG benefits of VMware/VirtualBox over qemu back then was precisely the "software" virtualization aspect, i.e. that it was capable of running unmodified user-mode code from the guest on the host, at native speeds, even without Intel VT (which did not exist at the time). It was a big improvement over plain qemu. Even today the difference would be likely noticeable.

In fact qemu did eventually release a kernel module, called kqemu, that was comparable to this software virtualization approach. It was so important that the guy decided to release it as proprietary code (like VirtualBox at the time... ). I would say it was the success of these early purely-software-based virtualizers that lead Intel to design Intel VT, and thus enable hardware-assisted virtualization (and kvm, etc.) .

(Kernel stuff could not be run unmodified on the host for obvious reasons, so it was always emulated or translated. Only after Intel VT showed up you could also natively run that).

[50ShadesOfDirt]

Thanks for all of this, folks!

I'm trying to get the latest ArcaOS to run under the latest QEMU now, just to start understanding the process. I hope to get QEMU to run ArcaOS purely under emulation steam, vs any kind of pass-thru to the CPU, and see what speeds look like. It didn't jump up and run, so there's some fiddling to do ...

Will try to report back when ArcaOS is sorted out ...

Thanks again!