[Info-vax] Release of OpenVMS V9.2 for x86 Scheduled for July 14, 2022

[Dan Cross]

In article <tahq1i$1ou3j$4 at dont-email.me>,
Simon Clubley  <clubley at remove_me.eisner.decus.org-Earth.UFP> wrote:
>On 2022-07-11, Dave Froble <davef at tsoft-inc.com> wrote:
>> On 7/11/2022 2:03 PM, Simon Clubley wrote:
>>> Rather than one of the above boxes however, I would follow VSI's advice
>>> on what hardware they recommend to host your VMs running VMS. For example,
>>> take note of the fact you can only use Intel CPUs for now and cannot use
>>> AMD CPUs. (Unless that has changed _very_ recently).
>>
>> You maybe getting mixed up.  Running a VM doesn't care about mfg.  Or maybe I'm 
>> mixed up.
>
>Do the VM environments emulate missing hardware instructions with reasonable
>performance levels ?
>
>For example, what happens when the VM software is being run on a box without
>PCID support ?

Typically the Hypervisor (both the VMM and other emulation
components) define a "platform" that they emulate, that consists
of IO devices, CPU features, etc.  This is what they expose to
the guest.  On machines that fully support this virtual platform,
emulation is very efficient; on machines that do not fully
support the platform, trap-and-emulate techniques are used,
which is usually less efficient but very often the use of these
features is rare.

On x86, CPUID (which is often used to probe what CPU features
the machine supports) causes an unconditional trap into the
hypervisor, and the hypervisor emulates the instruction, so the
guest only sees what the HV wants it to see.  Of course, the
guest can ignore that and try to use some features that the
underlying machine exposes without coordinating with the HV,
but this often fails.  It also fails on real hardware, too, so
one doesn't lose much when the guest "cheats", but the error
cases can be abstruse.

Similarly with PCI config space, writes to the APIC's ICR, etc.

	- Dan C.