News on Rosetta for Linux VM's

[ColoradoMarmot]

https://arstechnica.com/gadgets/2022/06/macos-ventura-will-extend-rosetta-support-to-linux-virtual-m...

Interesting.  I suspect it requires using their hypervisor, which is, well, not as robust as Fusion.  Wonder if Fusion could use the same API's though.

[Technogeezer]

This stuff Is seriously cool.

If you have an Apple Developer account (even the free one) I suggest taking a look at the WWDC session on virtualization they posted today. They not only now support Linux with things such as virtio devices, and EFI support for VMs but they have extended Rosetta 2 to allow Intel x86 apps to run in ARM Linux VMs. Also seems like they’ve built a folder sharing mechanism that works as a virtio device. 

I would hope that VMware would be working on incorporating these sooner rather than later.  don’t give Parallels another 2 year head start or else Fusion will descend into irrelevance. VMware, please show us that you’re into supporting macOS rather than Fusion looking like a red headed step child of ESXi and Workstation. 
(No flames here about red heads - I’m a ginger). 

[paddytb]

I'm scratching my head wondering why I spent £166 renewing my Fusion Pro licence - I naively it seems expected VMWare to fully support M1s in Fusion but they have produced nothing apart from the tech preview sitting on my Mac mini - which is pretty awful compared to Parallels.

I used an Intel Mac to run VMs for work but have been unable to since switching to M1 - the stack I work with does have a Linux x86/64 variant though so this could potentially be what I've been waiting for... but will VMWare get off their butts and do anything?

Very disappointed by the lack of interest - there have not even been any new blog posts for 9 months 😞

[Technogeezer]

I’ve been willing to give VMware the benefit of the doubt here but not much longer.  I too am disappointed in VMware’s inaction. Their legendary silence on upcoming releases is deafening. They still have no viable product on any Mac that Apple announced starting in late 2020. Almost 2 years without a product.

it is also now mid-June and WWDC has come and gone. Not a peep about a Tech Preview update like they hinted here. It is desperately needed as the current Tech Preview has not been able to run ARM Linux kernels since March (3 months ago). I suspect many are losing patience.

VMware owes their Mac customers a lot more than what they’ve done to date. They’re almost encouraging their Mac customers to move to Parallels.  

I  would not hold my breath for VMware to include Intel chip emulation in their hypervisors. Nor would I expect it out of Parallels. Emulation is not a core competency of either. CPU chipset emulation is too slow to be useful. @ColoradoMarmot has said it before here. Use an Intel system if you need to run X86_64 architecture operating systems. 

Unfortunately if you still need to run Intel operating systems on todays Macs, Apple is the one that sunk that boat. Parallels and VMware in the same boat. 

[ColoradoMarmot]

Their conference is the end of August, so hopefully that's production release date.  I'm surprised we haven't gotten an update for the linux kernel issue.  Perhaps there's something else coming and they're bundling it together, but all too often marketing vs engineering vs product stars don't align the right way.  Delaying a critical patch to make a bigger splash may look good in one department, but it does raise issues.

I do wonder though if the acquisition has scrambled things a bit (or if it will do so going forward). 

[Technogeezer]

The optimist in me is hoping that the delay may be due to VMware wanting to make sure the TP works on the Ventura beta. 🤞

TheUber-optimist in me hopes they are building in some of the goodies announced at WWDC and are going to build VMware Tools for Windows 11 ARM  (hah to you Microsoft). 

I can wish, can’t I?

[fabienmagagnosc]

actually, based on how Parallel, VMware fusion and Qemu works, they all use the hypervisor API from MacOS Monterey (https://developer.apple.com/documentation/hypervisor) which allow to run the VM in user space instead of as a kernel level extension (I did check, and didn't find any any Vmware, Parallel extension ...)

So, I suppose it can be "easy" to implement/migrate feature(s) from VMware ...

But, there is a clear thing going on, as VMware is not an independeant company (spawn off Dell/EMC) : they do have priorities, and clearly, based on my exposure of Vmware stack : they care more on VMware Tanzu, totake over the full building pipeline from dev to cloud hosting, then probably VMware Fusion ... it's less of a priority for them, and they obviously stay silent about it ... that's it ... but as we did mentioned : Parallel is a great stable alternative, as Qemu, and maybe it's time to look at those too.

[ozsmacd]

Sadly I doubt they are going to push out the feature set, it seems to be a desire to do the minimum work to keep something running.

What I’m most surprised about is the total radio silence on the SIGNIFICANT security risks anyone running the most recent preview (19431034), in that users are forced to use older Linux kernels without key security fixes/patches.

Come on VMware, it’s been months now, this is now beyond a joke.

[yfx01]

Yes, it is very bad, but I have to use it again. I hope to release an updated version earlier to solve the current problem. It has been two years and it is still the same. I am very frustrated.

[Idontwanttobeab]

I find the total lack of information about the future of Fusion utterly disappointing.

Having a preview sitting there dangling is not very reassuring.

[ColoradoMarmot]

They've hinted something is coming in June, which is more than they used to be able to do.  VMWare, like most public companies, has to be careful about what they state because it can impact earnings.  I suspect that's even more true now with the pending acquisition.

Hopefully we'll see something in the next couple of weeks.

[Idontwanttobeab]

I frankly cannot see they have to be more cautious or secretive than Apple who can disclose quite a bit of future plans and betas both to developers and customers. 

It is a pretty bad excuse to hide behind. 

[Technogeezer]

To be fair, Apple only seems to disclose futures at their WWDC conference. They do not otherwise discuss futures with the public. They tend to be as closed lipped as VMware. 

[Idontwanttobeab]

Apple is tight lipped about HW development. For software not so much so. Once a product is out in (public) beta there is a steady flow of information both to devs and the public.

The Fusion tech preview is for all practical purposes a public beta so progress information is important.

This is particularly the case now in the middle of a processor transition imposed on the Fusion product by Apple, which leaves many users stranded in terms of covering their needs for running other operating systems on Apple hardware. 

Many Fusion users are cross platform devs, and the lack of the ability to run virtual machines on the Macs can influence anything from change of workflows, investment in more kit to cover their needs to even abandoning one of the platforms they used to develop for. 

If there are issues that needs to be sorted, tell us. No information is absolutely the worst.

[yfx01]

I agree with you very much.  @Idontwanttobeab 

[Mikero]

Okay...

So, off the bat we can't use this really intersting sounding Rosetta feature because it's limited to virtualization.framework, whereas we use hypervisor.framework becuase we already wrote the code to do 'all the things a VM does'. Virt framework is a super-high-level API that basically says 'give me VM, here is my ISO' and it does that.

We're thinking about what makes sense there, but we also want to keep features that are uniquely ours like Snapshots, Clones, 3D hardware accelerated graphics, the network editor, and connecting to + compatibility with vSphere.

Re: the delay... No one is more disappointed about this than I am. We expected the current TP to be healthy for a while, but we didn't anticipate the ID_AA64ISAR2_EL1 bug. We have that fixed internally, and we're just working out when to ship it as we weren't planning an interim release on the current TP code branch. 

We're a little behind where we wanted to be this year for a bunch of reasons I can't get into, but I'm seeing really great progress internally.

IIRC, we've already checked in our 3D vmwgfx drivers into Linux, just waiting for distros to pick that up.

Likewise, there's an Ubuntu bug which got fixed _IN JANUARY_ that Ubuntu still hasn't pulled in for some reason:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/video/fbdev/core/f... 

You can tell it's not there by looking at Ubuntu's fbmem.c:

https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/focal/tree/drivers/video/fbdev/co...

https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/kinetic/tree/drivers/video/fbdev/...

Kali picked up this change, and it works _SUPER GOOD_ internally. (once you get around the _EL1 bug from above of course.)

So it's really a confluence of unexpected bugs that got us down, and it's all outside of what should have been our normal release process, so it's taking time to cut a new branch and pull all the now-fixed bits together.

Not to mention we're doing big changes to support Windows 😉

As for the 'VMware being Silient'... I'm pretty vocal and I'm not trying to hide anything (within the limits of what I'm allowed to disclose of course).... I'm just SUPER BUSY trying to test all this stuff and make sure it meets the community's expectations, and am probably the only one authorized to really talk about it publicly.

The last thing I'd want is for us to rush out a broken build or something that eats your VM if you click the wrong window or something.

Anyway... Long story short... I'm _REALLY_ encouraged with what we're seeing internally... I know I said June for a TP refresh, and that's some egg on my face which I accept... but the benefit of us waiting a _little_ longer is that we'll have a vastly superior Windows and Linux experience vs what we're seeing today.

[Idontwanttobeab]

Thanks for the update! 👍

[Technogeezer]

Michael, appreciate that you guys don't want to ship out anything that's half baked. And I appreciate you being able to share as much as you are within the corporate disclosure mandates. I'm going to rant a bit here, but trying to keep it professional.

It's simply not been enough.

The big issue is not 3D or the graphics frame buffer stuff. It's the ID_AA64ISAR2_EL1 "bug" - your bug, not Linux - that's ground the TP to a halt because NO distribution that has this register check (and that's now pretty much all of them because of the kernel security implications) can install latest releases or boot after upgrade. QEMU and Parallels fixed it within a couple of weeks of it being found. Over 3 months later and the Tech Preview still has the bug.

Yes, I understand it's a tech preview. But it's the only thing we have from VMware that works on M1 Macs, and like it or not it's being used for more than just testing because of that. 

You guys do this for a living and are the "standard" in virtualization. We expect more from you because of that. 

I hope that what you are going to ship "soon" blows our minds and is worth the wait. 

[fabienmagagnosc]

I do understand the pressure to deliver "something" when in the mean time there is probably Fusion 13 to look for, as,we all understand this is a commercial product after all !

But, when you realize that the day you got an issue on a similar product, Qemu, and you got a properly describer situation and a comprehensive patch http://patchwork.ozlabs.org/project/qemu-devel/patch/20220207225211.15281-1-agraf@csgraf.de/, then you realize that paying, now or later for a product which is now 4 months lagging behind is maybe not a solution.

We are now using Qemu and Parallel, as we ended up completely stuck for now 4 months with VMware Fusion, when basically everybody else was moving forward ... (no centos 9, no fedora 36 except via upgrade from an old fedora 35, and the device drivers are not working "perfectly, try to boot .. 15 to 20 minutes each time !, no ubuntu 22 ...)

As per the graphical interface, I do think there is clearly something I don't understand here : there is not even a single message in the forum mentioning it .. and this is clearly understood by everybody that the priority is to support the ISA : the CPU mainly, NOT THE GPU ... so I'm wondering if the priorities are managed right in the VMware fusion branch of VMware, which is maybe now only few developers as VMware workstation and ESXi are probably taking much of the resources (those are VMware bread and butter) !

Anyway, I clearly appreciate the explanations, but for some of us, it's too late, and not too little, but the wrong focus I feel, causing a 4 months "completely stuck" and for some of us working on kernel, and products where schedulers, drivers and others features are important, then this is too much to wait .. it's nearly 5 development sprints/cycles (of 3 weeks each), too much to have developers stuck and not progressing on an architecture which showed massive improvements over X86/ADM64 in term of performance per watt opening the door for datacenter changes in a massive way ... and Qemu allowed us to do it (some use Parallel)

[Technogeezer]

---

@fabienmagagnosc wrote:

As per the graphical interface, I do think there is clearly something I don't understand here : there is not even a single message in the forum mentioning it .. and this is clearly understood by everybody that the priority is to support the ISA : the CPU mainly, NOT THE GPU ..

To be fair, @Mikero has stated that 3D support for Linux is upcoming in the Fusion for Apple Silicon - they have checked in changes to the open source community that are required for Linux to support it. However your point still stands that they have not released a update to the tech preview that takes advantage of that.

My feeling is that they'll support it as they have on Intel - you won't get direct access to the GPU on the host, but they will have their VMware SVGA 3D support that will map to Apple's Metal graphics APIs, and you'll need a recent version of Mesa in the guest.