LWN: Comments on "Understanding the Jailhouse hypervisor, part 1"

Understanding the Jailhouse hypervisor, part 1

vsinitsyn — Sun, 12 Jan 2014 17:57:57 +0000

I don't remember any specific cache-handling code in Jailhouse neither I remember any benchmarks measuring this trashing's impact on performance. Given the project is quite young, I think this may (or may not) be addressed in future releases.

Understanding the Jailhouse hypervisor, part 1

dchichkov — Sun, 12 Jan 2014 02:01:55 +0000

The CPU L3 cache, is it being managed in any way? Or that Linux cell can still happily trash L3 for all the other cells?

Understanding the Jailhouse hypervisor, part 1

vsinitsyn — Thu, 02 Jan 2014 17:24:14 +0000

It is. The idea is to keep Jailhouse simple but compatible with KVM for those requiring more than Jailhouse provides itself.

Understanding the Jailhouse hypervisor, part 1

ijc — Thu, 02 Jan 2014 17:14:05 +0000

But the "containing" KVM does have emulation, right? In that case what is the overall difference?

Jailhouse sounds interesting but AFAICT it is no more "Xen done right" than it is "KVM done right", in fact I would venture that in terms of functionality it is pretty much orthogonal to the other two.

Understanding the Jailhouse hypervisor, part 1

pbonzini — Thu, 02 Jan 2014 17:03:18 +0000

It is the usual KVM hypervisor, running within the Linux kernel. Jailhouse lets the Linux cell do nested virtualization.

So, unlike Xen, Jailhouse does not have any knowledge of QEMU. All the knowledge of emulation is entirely within the Linux cell.

Understanding the Jailhouse hypervisor, part 1

ijc — Thu, 02 Jan 2014 16:12:41 +0000

How is KVM running in the Linux cell different to the Qemu associated with an HVM guest in an x86 Xen system?

Understanding the Jailhouse hypervisor, part 1

deepfire — Thu, 02 Jan 2014 13:34:38 +0000

I find it hard to reconciliate "Xen done right" with static partitioning.

Understanding the Jailhouse hypervisor, part 1

vsinitsyn — Thu, 02 Jan 2014 13:09:48 +0000

Yes, that's true. Thanks for pointing this out.

In this article I focused more on how Jailhouse does things than on what it does. To that end, the official presentation (easily found through initial announcement link) can be a good starting point.

Understanding the Jailhouse hypervisor, part 1

pbonzini — Thu, 02 Jan 2014 12:21:29 +0000

You haven't mentioned another cool thing that distinguishes Jailhouse from Xen. Not only Jailhouse doesn't care about doing complicated emulation (like Xen does with QEMU in HVM domains); it offloads this to KVM running in the Linux cell.

In this sense, Jailhouse is in my opinion really "Xen done right". It would also be possible, in principle, to add grant table and event channels hypercalls, so that Jailhouse can run the "PVH"-style (PV with an HVM container) domains that Xen recently got support for.