[feature-request] PCI pass-through (VT-d) in GUI

[michaln]

In PCI-passthrough, the guest gets the PCI device totally. It is yanked from the host, and guest gets total control of it.
So driver works ONLY in the guest, and performance is at hardware level. Host OS gets bypassed completely. Host OS won't even detect bypassed PCI devices, the moment you attach them to guest.

Wrong. You clearly have no idea how it actually works The host OS is still in control to some extent (how do you think the VMM would manage the device if it completely disappeared from the host?), and most importantly any interrupts from the device are received by the host OS and must be forwarded to the VM. Only the very latest CPUs might possibly be able to handle interrupts without host OS intervention, though I'm not entirely sure about that.

For register/MMIO/memory access, yes, the performance penalty should be negligible or nonexistent. For interrupts definitely not.

[Armando]

Priorities depend on customer interest. There is no customer interest in PCI passthrough.

Well, worse for me, then. :]

As to the "delay or not delay" issue, I actually did not expect VT-d PCI passthrough to totally remove any differences between host and guest. I expected it, however, to reduce them significantly AND to let me install the proper audio drivers inside the VM.
Which could maybe prove to be enough for my needs.

Thanks, anyway, for letting me clearly understand I'd better stop waiting for something that won't happen.

[michaln]

Thanks, anyway, for letting me clearly understand I'd better stop waiting for something that won't happen.

That's not quite what I said -- but your takeaway is correct. If you want a solution in the foreseeable future, don't wait for VirtualBox to improve PCI passthrough support.

[ToryPhnx]

Hello,

So....... First, to the claim that there is no interest in PCI Pass-through.... That is UTTER BS. Seriously, how many 40 year old virgins to I need to sacrifice on the alter of VirtualBox PCI Passthrough to get the feature?

The entire reason I first looked into Virtualization was for PCI passthrough. I have to support dozens of older systems that require PCI cards such as GPIB/HPIB adapters and other more esoteric connection types. This is due to the fact that software on those systems needed to run instrumentation and such will not run on new OS's. No, Win7 "Run-As" does not work. The hardware gets harder and harder to find when you need to replace hardware for a Windows 98/NT/2000 system. Virtualization is becoming one of my only options, and yet, without PCI passthrough I am forced in most cases, if a USB converter is not available, to beg, borrow, and bleed in order to keep said systems running.

I can pass USB and Serial ports, even serial ports on a PCI card, Network connections, etc..., to a VM all day long, every day, and twice on Sunday. But if I need to pass the PCI card itself, I'm screwed and don't even get to enjoy it. Shared interrupts on the PCI bus kill that no matter what I do in BIOS to kill/shutdown/eliminate other devices on the bus and therefore the VM will not start.

I am begging you for this feature on any Host, Windows preferred (AD environment), but if I have to work with Linux I will. Just give me PCI passthrough that actually works.

[loukingjr]

Hello,
So....... First, to the claim that there is no interest in PCI Pass-through.... That is UTTER BS.

Actually what was said was "customer interest". You don't pay for VirtualBox hence you are not a customer.

[michaln]

Seriously, how many 40 year old virgins to I need to sacrifice on the alter of VirtualBox PCI Passthrough to get the feature?

No need to sacrifice virgins, just contact your Oracle sales representative.

[ucrasher]

Hi,
I agree to "not needed in GUI" for now.
I disagree there is no user interest. Many guys are trying it and finally give up. Intel is most guilty do to the Vt-d policy and poor information. Followed by the mainboard makers that do not care.

Actually I had it working for quite some time, but then a kernel update from 3.2 to newer broke it. (there is a thread here).
Found this thread after trying kernel 3.19 (Ubuntu 14.04) without success.

If I can help, let me know - at least as a tester.

[mpack]

I agree to "not needed in GUI" for now.
I disagree there is no user interest.

For the n'th time, that is not what was said.

What was said was that there was no customer interest. I.e. there is not enough people willing to pay for this work. Taking advantage of a free download does not make you a customer.

[pabzum]

I’ve read this thread through, and it sounds like pci-passthrough is already possible but hard to figure out for each individual case. Have I interpreted it correctly? The feature is not being pushed through development quickly enough because very few paying customer want it.

But look at my case: I’m trying to get a Soundblaster Audigy audio card pass through into a WinXP guest from an Ubuntu host. The reason for this is that WinXP already has everything I need to do my job as musician and music transcriber. Why would I keep wasting money on newer and newer versions of Windows, Sibelius and Sonar? So I have installed Ubuntu as host, isolated WinXP inside a VM (from the internet, I mean), and am working with my music software just as before (with a little latency on audio recording, yes). The SB audio card might add a few useful features, though, but I haven’t yet figured it out.

I bet that, as the customer base of current software gets older, customers will start moving more and more into virtual machines inside Linux hosts. Who wants to spend a lifetime of learning tons of useless new ways of doing things everytime Windows changes everything around? To hell with it. I still use Word 97. It’s got all my macros and it’s more than enough for me. It’s now inside my VM. So I predict that, though there’s little customer support for pci-passthrough right now, it will come. It’s in the wind already.

[Intrepid04]

For sure. It IS coming. I was looking for GPU pass through as well and went with KVM for now. Can't post the link yet but google for Multiheaded-NVIDIA-Gaming-using-Ubuntu-14-04-KVM
here is a demo - https://www.youtube.com/watch?v=37D2bRsthfI
It's also a choice, whether to lead the pack or play catch-up with the technology.

[Hwrd69]

The entire reason I first looked into Virtualization was for PCI passthrough. I have to support dozens of older systems that require PCI cards such as GPIB/HPIB adapters and other more esoteric connection types. This is due to the fact that software on those systems needed to run instrumentation and such will not run on new OS's. No, Win7 "Run-As" does not work. The hardware gets harder and harder to find when you need to replace hardware for a Windows 98/NT/2000 system. Virtualization is becoming one of my only options, and yet, without PCI passthrough I am forced in most cases, if a USB converter is not available, to beg, borrow, and bleed in order to keep said systems running.

Though I am not as knowledgeable as some of the others on this forum, I too have a need for PCI pass-thrus on Windows. We are being "forced" to upgrade our systems where some are still running older OSes (Win2k & WinXP) to newer OSes; yet this hardware will not run with the newer OSes. Being able to run WinXP within Win 7 would allow me to upgrade these systems yet maintain compatibility with the hardware. These systems are also using PCI-based instrumentation which requires that PCI-pass thrus to work. I too am a lone voice, but believe me, there is a need. Even if I purchased VirtualBox as a "paying customer", I still would be a lone voice and not sufficient to encourage development.

Just saying....

[mpack]

Frankly, if you are thinking of using a virtual machine to control physical hardware then I have my doubts that you understand what a virtual machine is. Almost any other solution would be better IMHO.

[Intrepid04]

That doesn't sound quite humble. What is not understood is there are various different ways the Virtualization can be used and non of it is invalid per say. There are different ways to solve the same problem obviously but having the capability to use available hardware to achieve performance benefits is not an invalid use case but you can obviously choose to ignore it.

[mpack]

Being humble not a trait I value. Being accurate is.

Sorry, but we all go on our experience: and my experience is that controlling physical hardware is not something which VMs do well, nor ever will. Use your wits and come up with a better solution. Don't hold your breath waiting for this to be practical with VirtualBox.