Re: NVMM on non-x86?

[Chris Hanson <cmhanson%eschatologist.net@localhost>]

On Feb 16, 2020, at 11:45 PM, Martin Husemann <martin%duskware.de@localhost> wrote:
> 
> On Sun, Feb 16, 2020 at 11:37:50PM -0800, Chris Hanson wrote:
>> My understanding is that the x86 CPUs have generally had poor
>> virtualization features in that it?s difficult to trap and handle all
>> higher privilege instructions accurately. Most other CPU architectures
>> from the mid-1980s on don?t have that problem. (The 68000 had an
>> equivalent issue, it was addressed with the 68010.)
> 
> I think you are mixing up various "virtual" things here.
> 
> I remember the 68010 issue, but it was totaly different (about
> restarting a trapped instruction without side effects)

Not really—while at the time the 68000 issue (fixed with 68010) mainly prevented straightforward implementation of virtual memory, it would have also affected virtual machine implementation too since that also requires instruction restart (or simulation of it).

> than the modern
> virtualization features NVMM uses (or for example the hypervisor on
> sun4v machines).

Sorry, I just don’t see what’s so different between a hypervisor on modern x86 versus what could be done on a 68020 or later, other than the ability to take advantage of specific virtualization “assistance” features in modern x86 designs.

Can you explain in more detail, or point me to someplace I could read up? What’s needed beyond the Popek & Goldberg virtualization requirements?

  — Chris