Software VT-x

[groveborn]

I know that there are plenty of CPUs out there these days that support hardware VT-x and similar functions, but wouldn't it be nice if there were a software equivalent? Some of us can't upgrade to those CPUs due to system restrictions... Something that sits between Vbox and the system (or simply integrated) that grants the virtualizer full CPU access. Then again, I don't know that such a thing is possible.

[Sasquatch]

Even if it would be possible, it's going to be so slow, you really don't want it. But it's not possible, because hardware virtualisation is actually a bug in the x86 architecture that got fixed. The processors were never made to work with virtualisation in this manner. Now they found a way to make it work better without loosing or altering the x86 architecture. There is more information about this on Wikipedia and other sites.

[Technologov]

The Good News is that VirtualBox supports software virtualization, so you can run your VMs on your CPU.

[Sasquatch]

I think he's suggesting the VT-x requirement for 64 bit and SMP to be emulated in software.

[chaer]

Any progress on Software VT-x ?

[mpack]

Progress how? The notion is ridiculous. VT-x is hardware assisted virtualization. You can't do "hardware assisted" in software. Plus a VT-x enabled CPU exposes certain processor features that in the past had to be worked around using code patches: that can't be simulated either.

If the question is, are the devs ever going to do 64bit support in software only, I'd have thought the answer was no, because there's no market for it. Almost all modern CPUs that can run a 64bit OS already have VT-x (or AMD-v).

[michaln]

There is about zero likelihood that VT-x would be fully emulated in software in VirtualBox. It makes very little sense for general use.

Some hypervisors (VMware, KVM) have the ability to expose VT-x/AMD-V to guest software, but only if the host hardware already supports it. It is possible that something like bochs emulates VT-x fully in software. This means much lower performance than e.g. the software virtualization that VirtualBox uses.

As usual, one problem is that for the amount of money Oracle would have to spend developing this, you can buy a mountain of VT-x capable hardware.

[chaer]

thanks for your reply.
attually, software virtulization is what i concern. could you please talk about the story "PATM_EMULATE_SYSENTER" ?

[Hugh2]

Some hypervisors (VMware, KVM) have the ability to expose VT-x/AMD-V to guest software, but only if the host hardware already supports it.

That's cool! How well does it work?

I always bought into the argument that you shouldn't try to nest VMs because performance will suck and you don't need to anyway because you can run any number of guests side by side instead of nested.

But I can think of a couple of use cases for it.

• Debugging a hypervisor by running it in a VM.
• Trying out VM software that you don't trust. (eg. don't trust it not to crash your PC, or just don't trust it to uninstall cleanly)

[michaln]

I always bought into the argument that you shouldn't try to nest VMs because performance will suck and you don't need to anyway because you can run any number of guests side by side instead of nested.

That's still true. A non-nested VM will always perform better than a nested one.

Debugging a hypervisor by running it in a VM.

That is true, but it assumes the software implementation is bug-free (possibly true) and behaves exactly like hardware (not true--think TLB management for example).

Trying out VM software that you don't trust. (eg. don't trust it not to crash your PC, or just don't trust it to uninstall cleanly)

That would be the typical use case. Probably not something a significant percentage of users is interested in, although it would be interesting to know what exactly the percentage is.

[ErikW]

The System Management Mode of Intel CPUs would certainly be capable of supporting software emulation of the VT-x features. The issue is definitely performance. The result would be so slow that the guest OS would be unusable, if it even did successfully boot.

A great analogy is floating-point instruction emulation. Intel no longer sells x86 CPUs without floating point hardware because modern software relies on floating-point calculations too much for emulated floating-point to have adequate performance.

VirtualBox uses other techniques to make up for the lack of hardware virtualization, including translating the binary instructions to include the necessary hooks when software is loaded into memory. Those techniques could be applied to SMP as well, though the complexity of the problem goes up dramatically when emulating multiple CPUs.

It is unfortunate that Intel and AMD chose to make hardware virtualization an optional feature. My solution is to only buy CPUs that provide the hardware features for virtualization and support enough memory and performance. Now that we've hit the 5 Ghz. clock rate ceiling it makes more sense to buy CPUs less often and spend the money on features to improve the performance of a particular application. Buying mediocre CPUs more frequently no longer provides significant performance increases.

Microsoft is pushing virtualization quite heavily, but only for server applications. We are probably going to see consumer Desktop PCs become less powerful and maybe disappear entirely. My guess is that gaming PCs and workstations will stay around, but remain in a higher price bracket, while many companies will bail out of the less expensive consumer desktop PC market.

I remember the days when those who chose to run NT 4.0 had limited choices for hardware and software applicable to personal computing, multimedia and gaming. We may be returning to that same situation, where "multimedia" and "consumer" devices will have to be used for those activities rather than a desktop PC. The good news is that the life cycle for desktop PCs may actually increase, and the quality may increase as well.

It makes little sense to invest effort into software emulation of hardware features that will likely become a standard part of the hardware. It's likely that the less expensive CPUs will disappear entirely or become embedded into non desktop devices.

[Pattyyivi]

Almost all modern CPUs that can run a 64bit OS already have VT-x (or AMD-v).

You mean desktop hardware, however what about some laptops, modern as well, some of them seem have the virtualization feature missing or switched off in the wired manner.

[mpack]

What about them? If you need VT-x then make sure you buy a device that supports it. Just like if you need lots of RAM etc. As has already been made clear above, a software VT-x isn't going to happen for multiple reasons.

[Pattyyivi]

You know, it is always easier and less cumbersome to make software alterations than hardware restructuring, replacing CPU can damage motherboard and CPU itself.
Besides, some laptop manufacturers disabled VT-x intentionally, anybody knows why?

[socratis]

You know, it is always easier and less cumbersome to make software alterations than hardware restructuring

You're joking, right? Either that, or you don't honestly understand how things work in the computer world.

Pattyyivi, I believe the answer has been given more than once in a very clear manner. Especially the one from michaln on the other thread that you were complaining about the same exactly thing (w10 without any hardware virtualization, why not?). Give it a rest. It's not going to happen. It doesn't make sense for Oracle's business model.

On the other hand, it is an open source project, so contributions are welcomed (https://www.virtualbox.org/wiki/Contributor_information). Or you can hire a developer to do this for you. Expect to pay at least 6 figures.