Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
Microsoft Virtualization Operating Systems Software Linux

Microsoft Submits Linux Kernel Patches For a 'Complete Virtualization Stack' With Linux and Hyper-V (theregister.com) 105

Microsoft has submitted a series of patches to the Linux kernel with its aim being "to create a complete virtualization stack with Linux and Microsoft Hypervisor." The Register reports: The patches are designated "RFC" (Request for comments) and are a minimal implementation presented for discussion. The key change is that with the patched kernel, Linux will run as the Hyper-V root partition. In the Hyper-V architecture, the root partition has direct access to hardware and creates child partitions for the VMs it hosts. "Just think of it like Xen's Dom0," said Microsoft principal software engineer Wei Liu. Hyper-V's architecture is more similar to Xen than it is to KVM or to VMware's ESXi, and Liu acknowledged that "we drew inspiration from the Xen code in Linux," specifically for code handing interrupts. Until now, the Hyper-V root partition had to run Windows.

Microsoft has also ported Intel's open-source Cloud Hypervisor, a Virtual Machine Monitor (VMM) written in Rust that normally runs on KVM, the hypervisor that is built into the Linux kernel. Cloud Hypervisor itself is currently in "very early pre-alpha stage." Even when Linux is the root partition, it will still run on top of Microsoft's hypervisor, a thin layer running with ring -1 privileges. It will no longer be necessary to run Windows on that hypervisor, though, enabling Microsoft to call the new arrangement "a complete virtualization stack with Linux."

This discussion has been archived. No new comments can be posted.

Microsoft Submits Linux Kernel Patches For a 'Complete Virtualization Stack' With Linux and Hyper-V

Comments Filter:
  • Will be interesting to see what (if any) comments Mr. Torvalds has to say about the quality of these patches.
  • A move long overdue. Although nothing really good has ever been produced by MS and they are sneaky, honor-less, underhanded bastards. So I remain quite suspicious about the long term benefits of this.

    • The benefit is quite simple-
      Shops that use an ass-ton of virtualization (like my own) for them or their clients (Our primary Hyper-V users are clients- we use Citrix internally) are going to give Hyper-V as a type-1 hypervisor a second look if we can manage it from Linux instead of using Windows Server.
      • by Junta ( 36770 )

        I suppose what I'm missing what is particularly attractive about Hyper-V versus either the similarly designed Xen or a KVM strategy if using Linux.

        • by gweihir ( 88907 )

          I suppose what I'm missing what is particularly attractive about Hyper-V versus either the similarly designed Xen or a KVM strategy if using Linux.

          Indeed. They are late to the game. I would be surprised if they have any technological advantages. And it is MS, so the tech is not going to be very good, even if based on Linux.

        • Nothing.
          They're all the same to me (Though KVM really is a different kind of product all together- it's not a bare-metal (Type 1) hypervisor)
          Hyper-V, Xen, ESXi... all just different products doing the same shit.
          Why do some of my customers want their stuff running on Hyper-V? Honestly- I don't have any damn idea.
          I mean it has some pretty great packaged clustering options that integrate will with clustered Windows services, but I imagine those are all operating at the Windows Server primary partition leve
  • How does this improve the linux kernel? If they want to modify the kernel to make a hypervisor for their commercial product, they are free to do so. AFAICT, linux currently has no issues with running Windows VMs in a slim hypervisor, for example, the Xen Dom0 they reference.
    • by hcs_$reboot ( 1536101 ) on Tuesday September 15, 2020 @08:18PM (#60509774)
      The improvement is that when it crashes, the screen is colored in blue
    • Re:Improvement? (Score:5, Interesting)

      by Trongy ( 64652 ) on Tuesday September 15, 2020 @09:14PM (#60509910)

      What I see is developers following the open source process. Most companies who write code for the Linux kernel do so for their own benefit.

      There's probably no value for anyone but Microsoft at this stage. If they never distribute the binaries outside of their own Azure datacentres, they wouldn't even have an obligation under the GPL to distribute the source code.

        We don't know Microsoft's intentions for this code. It might be just an experiment that never goes into production. It's probably an effort to increase reliability of their Azure servers while decreasing the cost of maintaining hardware drivers.

      • by Anonymous Coward

        I agree this is likely a cost thing. It takes a huge amount of effort to maintain all the different pieces, and there is exponentially more code to keep track of these days. It's probably an attempt to negate costs by shoveling some of it over to OSS that will see bug fixes - whereas their own work product not so much.

        • Just because some code becomes part of the Linux kernel doesn't mean that said code will be magically maintained by $someone_else. Microsoft will still need to maintain their own code or it will be removed at some point in the future if it doesn't play nice with other new features introduced into the kernel.

          • by Trongy ( 64652 )

            While that is true, you are missing the point. They have written a relatively small amount of code to interface their hypervisor with the Linux kernel in the root partition. Using the Linux kernel in their Dom0 equivalent means they can rely on a vast number of hardware drivers in the Linux kernel that are maintained by $someone_else. It's a net benefit.

          • by nazg00l ( 699217 )

            That is not necessarily why companies upstream code that addresses mostly or exclusively their own products. Sure they could maintain their own out-of-tree repositories... but then they would have to update the code for any architectural and API changes that get introduced into the kernel. Once they manage to get their code to the mainline tree, any future changes will have to account for that code so that it does not stop to build or work.

      • I think they actually want a lightweight Hyper-V host.
        Installing Linux takes fraction of the time a Windows server install takes. Time saved in deployment is important when you have a high number of servers to deploy.
        • Installing Linux takes fraction of the time a Windows server install takes.

          I don't know how much I agree with that... They're close.
          Biggest advantage I see is never having to use that fucking Hyper-V Manager again.
          With Linux as a host, means linux tools will be available for configuring the hypervisor.
          That's what I look forward to. But then again- I manage thousands of virts.

      • It's probably an effort to increase reliability of their Azure servers

        I manage a good amount of Linux (mostly Centos 5/6/7) and Windows (Mostly Server 2003, but a significant amount of 2018+)
        There isn't any noticeable difference in "reliability" between the 2 classes.
        Only difference that has stuck out to me in the last decade and change, is that Linux servers are vastly more manageable.

        while decreasing the cost of maintaining hardware drivers.

        OK- this belies a fundamental lack of understanding about what a type-1 hypervisor is.
        You should read up and come back, so you can at least formulate some opinions that are informed instead

    • Comment removed (Score:5, Informative)

      by account_deleted ( 4530225 ) on Tuesday September 15, 2020 @11:17PM (#60510108)
      Comment removed based on user account deletion
      • And here is the literal 1-single-person in this thread who knows what they're talking about, amidst the fucking sea of morons trying to sound smart.
    • linux currently has no issues with running Windows VMs in a slim hypervisor

      I take it you've never virtualised. I think there's one thing that's 100% certain is that ${GUEST_OS} always has issues with ${HYPERVISOR} regardless of which products you chose to assign either variables to.

      Virtualisation is not a done concept. It's not "finished". It's ever evolving, ever changing, with new capabilities and incompatibilities continuously being resolved.

  • by Gravis Zero ( 934156 ) on Tuesday September 15, 2020 @08:35PM (#60509824)

    What should really be considered here is if there is any benefit to Linux or FOSS by adding these patches. My understanding is that the Hyper-V Hypervisor is closed source. If they mainline these patches Microsoft will be the only one's benefiting. The closed source nature of the hypervisor could also limit the ability to secure the kernel interface.

    I see no reason to allow patches that only benefit a single party.

    • If you're using Linux on Hyper-V you're a Linux user. Are you arguing only "bare-metal" users should count?
    • Re: (Score:2, Interesting)

      by thegarbz ( 1787294 )

      I see no reason to allow patches that only benefit a single party.

      You may as well strip a large portion of the kernel out then. A metric fuckton of code in the kernel only benefits users of a single party that doesn't have their entire product stack open source.

    • What should really be considered here is if there is any benefit to Linux or FOSS by adding these patches.

      I'm sure there are plenty of Hyper-V users out there who would rather the primary partition be Linux instead of Windows. (Myself for one)

      My understanding is that the Hyper-V Hypervisor is closed source.

      Most machine architectures are "closed source"
      Think of a hypervisor as a modified machine architecture.

      If they mainline these patches Microsoft will be the only one's benefiting.

      That's like saying an ACPI quirk code-around for Phoenix BIOS only benefits Phoenix.
      Have you forgotten that there are... users?

      The closed source nature of the hypervisor could also limit the ability to secure the kernel interface.

      Only in the instance of insecurity in the hypervisor itself- otherwise, the Linux portion of the code is all open-source, and will be maintaine

  • There are many OSes and especially older things like old Linux installs that won't run on Hyper-V. They need special kernel treatment or network adapters that Hyper-V doesn't provide.

    But, all of these same VMs that won't run on Hyper-V will run on any version of ESXi, KVM, Xen, BHyve, openBox

    Microsoft need to fix Hyper-V for kernel issues, and provide a virtual Intel 1Gb NIC. They seem to be focused on extending Linux to run on Hyper-V and forgetting that it is supposed to be hardware virtualization. If it

    • Re: (Score:3, Interesting)

      by realmolo ( 574068 )

      Hyper-V is broken, and it always has been. But it's obvious that Microsoft doesn't really care about it anymore. They want you to run your stuff on Azure.

      Which is fine. If you are serious about running VMs on your own hardware, you should be using vSphere anyway. There really is no substitute for it in the enterprise. *Everything* works with vSphere, the management tools are the best, and there is an entire third-party industry built around supporting/extending vSphere.

      • by kriston ( 7886 )

        But it's obvious that Microsoft doesn't really care about it anymore. They want you to run your stuff on Azure.

        I wouldn't say that. They backported the same redesigned and rewritten Hyper-V that Azure uses to Windows 10 and Windows Server.

    • by kriston ( 7886 )

      They haven't yet bothered to provide accelerated 2D VGA graphics after all these years and they expect us to seriously consider Hyper-V?

      • 2d vga graphics?!! Hold on, I have 1993 on the other line...

      • You? No. They don't give one shit about you. They care about me.
        You see, I have to work with Hyper-V, Citrix, and ESXi on a daily basis due to being responsible for solutions built by customers. High-paying customers. Customers paying their hypervisor providers a shit ton of money too.

        If you asked me "What could I do to make Hyper-V better?"
        The first first thing I would say would be: "Give me a fucking console."
        If you wanted to one-up my request- give me a Linux primary partition.
        This is smart. It's j
    • There are many OSes and especially older things like old Linux installs that won't run on Hyper-V. They need special kernel treatment or network adapters that Hyper-V doesn't provide.

      Oh boy. I can see we're wading into a rather deep pool of ignorance, here.

      But, all of these same VMs that won't run on Hyper-V will run on any version of ESXi, KVM, Xen, BHyve, openBox

      Depends. Hypervisors that fully support paravirt-ops tend to do well with Linux guest compatibility. Otherwise, whatever free BIOS they're using is the big determining factor, and it's a mine-field.
      Overall- your statement is just false, though.
      There's a lot more to virtualization than hardware virtualization (these days, anyway)
      It doesn't work like bochs or old VMWare anymore. Everything deals with some level of paravirtualization.

      • by Junta ( 36770 )

        As for the CPU-specific virtualization, why would you want to emulate hardware when a pure-software interface can be achieved that makes the entire process thousands of times more efficient in terms of used CPU cycles and context switches?

        Taken to its logical conclusion, this would seem to advocate a more container mindset with common kernel seems to be the path then and eschewing the virtualization paradigm in general...

        • Definitely.
          Not that there aren't still advantages to paravirtualization.
          You can have a need for custom kernels that can't be containerized, but that you still want to operate efficiently. A fully paravirtualized kernel that can virtualize all of its x86 harware through direct hypervisor calls instead of actual hardware emulation actually has pretty low overhead.

          Other than that, most benefits of (para)virtualized instances vs. containerized are just the serious drawbacks to current container technologies
      • I provide 5 example hypervisors that can all do what Hyper-V cannot.

        But, my statement is false and the problem is my ignorance, and not Hyper-V being intentionally broken?

        What a thoroughly Microsofty response. I hope they pay you well for that.

        • I provide 5 example hypervisors that can all do what Hyper-V cannot.

          Well first off, you provided an unsubstantiated claim that I know to be a complete fabrication- but let's skip past that.
          I too can provide 5 example 4-wheel vehicles that can't do what a truck can.
          As I pointed out, you simply do not understand the market. You think hypervisors exist to emulate legacy hardware. That's not even what they do. You think the market for Hyper-V is you. It's not.

          Your understanding of the topic is infantile enough that you simply don't matter into any calculations as far as to

    • by Junta ( 36770 )

      While it is a valid criticism, if you have older things on your network that are too old to have drivers for a current hypervisor, you probably also have a security timebomb waiting to go off with your no-patches-in-ages images. It's probably gogod to pause and ask how good is it really to be running end of life systems that participate on a network.

      The other facet is that making Linux a valid dom0 might help them address those concerns. Every virtualization technology employes a 'fast path' where the guest

  • Several people I up-modded only show a single -1 Troll downmod. I know I didn't fat-finger it that many times. Hopefully posting un-does it.

    • by gweihir ( 88907 )

      You know, I have some suspicion there is something shady going on with moderation recently. Has /. been bought by MS and nobody noticed?

      • I don't know but my post it didn't undo it. It must have been someone else who modded all the anti-MS comments as troll and my up-mods never applied.

  • Microsoft, Microsoft never changes.
    The Gates waged war to gather desktop dominance and wealth,
    Ballmer, built an empire from his lust for throwing chairs and selling office,
    Nadella, shaped a battered Microsoft into cloud super power,

    But Microsoft never changes.
    They destroy Nokia for their flawed Mobile strategy

    And Microsoft never changes.

Think of it! With VLSI we can pack 100 ENIACs in 1 sq. cm.!

Working...