开云体育

Tony Harminc wrote:

> What it does do is make the architecture no longer virtualizable
> (without some feature like SIE).

Eh? How so?


> This 100% virtualizable feature (even for privileged states,
> and for that matter I/O) was to my knowledge unique among the
> architectures of the day, and may be still true.

I'm not following you.


> Look at the early kluges with i86 and VMWare and such, that
> had to resort to paravirtualization, where the guest has to
> cooperate (or the host zaps the guest code on the fly so that
> it does).

What does that have to do with your claim that the introduction of new instructions that weren't defined in the original published version of the architecture's manuals somehow makes the architecture "longer virtualizable without some feature like SIE"? I'm not understanding your argument/reasoning.

The architecture should still be virtualizable (even *without* SIE) with or without the introduction of the new instructions. And if none of the new instructions are privileged instructions, the architecture should still be virtualizable (without SIE) without any modification at all to the hypervisor. Only if support for a new privileged instruction was needed would the hypervisor then need to be changed (to add support to simulate the privileged instruction).

Is that maybe what you're referring to?

> I'm not sure that even today i64 is 100% virtualizable, is it?

I have no idea. I don't know i64.