Status of live migration on ARMv7

Status of live migration on ARMv7

[ib10evp0]

Hi,

I was wondering what the status of live migration for ARMv7 with Virtualization
Extensions is. I saw somebody was able to live migrate on ARM [1] but the 
repository he linked [2] has not been updated for 2 years. I'd like to know if
live migration support for ARM is already in the mainstream kernel or if there
are any custom patches required and if so if there is any newer code than [2].

BTW: The Migration page [3] claims the migration feature is hardware independent
but it remains unclear if this expands to CPU architectures and if this applies
to live migration or only offline migration. Some clarification in this regard
would be nice to have.

Regards


[1] https://sites.google.com/site/itriccmahypervisor/project/demo
[2] https://github.com/ITRI-CCMA-hypervisor/linux-kvm-arm/tree/live-migration-v8
[3] http://www.linux-kvm.org/page/Migration

Re: Status of live migration on ARMv7

[Marc Zyngier]

On Sun, 12 Jun 2016 14:40:40 +0200
ib10evp0@gmx.org wrote:

> Hi,
> 
> I was wondering what the status of live migration for ARMv7 with
> Virtualization Extensions is. I saw somebody was able to live migrate
> on ARM [1] but the repository he linked [2] has not been updated for
> 2 years. I'd like to know if live migration support for ARM is
> already in the mainstream kernel or if there are any custom patches
> required and if so if there is any newer code than [2].

Everything that's needed for a typical ARMv7 is already in mainline,
both for the kernel and qemu. And since the code is almost entirely
shared with ARMv8, it is probably in the same state. If it not working,
then it is a bug that should be addressed.

> BTW: The Migration page [3] claims the migration feature is hardware
> independent but it remains unclear if this expands to CPU
> architectures and if this applies to live migration or only offline
> migration. Some clarification in this regard would be nice to have.

We do have support for live migration (migrating the VM whilst it is
still running, tracking memory that gets dirtied).

	M.
-- 
Jazz is not dead. It just smells funny.

Re: Re: Status of live migration on ARMv7

[ib10evp0]

Hi Marc,

> Everything that's needed for a typical ARMv7 is already in mainline,
> both for the kernel and qemu. And since the code is almost entirely
> shared with ARMv8, it is probably in the same state. If it not working,
> then it is a bug that should be addressed.
> 
> > BTW: The Migration page [3] claims the migration feature is hardware
> > independent but it remains unclear if this expands to CPU
> > architectures and if this applies to live migration or only offline
> > migration. Some clarification in this regard would be nice to have.
> 
> We do have support for live migration (migrating the VM whilst it is
> still running, tracking memory that gets dirtied).

After reading this it's still not entirely clear to me if live migration using
your definition is also available for ARMv7 in the mainline kernel. Could you
confirm this with one sentence, please? Sorry for the inconvenience.

If it's available I'll build a KVM virtualization cluster over the following
weeks and report any bugs I encounter.

Thank you.

Re: Status of live migration on ARMv7

[Marc Zyngier]

On Tue, 14 Jun 2016 22:52:41 +0200
<ib10evp0@gmx.org> wrote:

> Hi Marc,
> 
> > Everything that's needed for a typical ARMv7 is already in mainline,
> > both for the kernel and qemu. And since the code is almost entirely
> > shared with ARMv8, it is probably in the same state. If it not working,
> > then it is a bug that should be addressed.
> > 
> > > BTW: The Migration page [3] claims the migration feature is hardware
> > > independent but it remains unclear if this expands to CPU
> > > architectures and if this applies to live migration or only offline
> > > migration. Some clarification in this regard would be nice to have.
> > 
> > We do have support for live migration (migrating the VM whilst it is
> > still running, tracking memory that gets dirtied).
> 
> After reading this it's still not entirely clear to me if live migration using
> your definition is also available for ARMv7 in the mainline kernel. Could you
> confirm this with one sentence, please? Sorry for the inconvenience.

I know it is available because I did review the code. Look for commits

7276030 KVM: arm/arm64: Enable Dirty Page logging for ARMv8
15a49a4 KVM: arm: page logging 2nd stage fault handling
53c810c KVM: arm: dirty logging write protect support
c647355 KVM: arm: Add initial dirty page locking support

as well as the generic infrastructure that precede them.

	M.
-- 
Jazz is not dead. It just smells funny.

Re: Status of live migration on ARMv7

[ib10evp0]

Hi Marc.

> I know it is available because I did review the code. Look for commits
> 
> 7276030 KVM: arm/arm64: Enable Dirty Page logging for ARMv8
> 15a49a4 KVM: arm: page logging 2nd stage fault handling
> 53c810c KVM: arm: dirty logging write protect support
> c647355 KVM: arm: Add initial dirty page locking support
> 
> as well as the generic infrastructure that precede them.

Thanks for your detailed response. It's all clear now.

Regards