Re: [Xen-devel] [PATCH RFC v1 00/14] xen: arm: support for save restore and dead migration

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On Wed, 2015-12-09 at 15:51 +0000, Andrew Cooper wrote:
> On 09/12/15 14:31, Ian Campbell wrote:
> > For arm32 + gicv2 systems the following supports apparently successful
> > save/restore as well dead migrate a domain.
> > 
> > There are several caveats/blockers, hence RFC.
> > 
> > Â* GIC v2 support only, no GIC v3 at all
> > Â* ARM32 only. Doesn't even build for ARM64 (vfp state handling needs
> > ÂÂÂadjustment)
> > Â* No live migration, only "dead" (no logdirty support yet)
> > Â* Have not fully audited the set of state which needs to be saved
> > (i.e. I
> > ÂÂÂmight have missed some registers or something)
> > 
> > On the flip side, thanks to migration v2 the toolstack side of this
> > practically trivial.
> 
> Glad to hear this.

BTW, when I said it was trivial in patch #13 I was referring to this aspect
of adding the ARM side, not the overall effort of migration v2 (which was
clearly a non-trivial effort)...

> > This is (of course) using the current Xen hvm blob handling strata,
> > which I
> > believe there are plans afoot to rework along the lines of the
> > libxc/libxl
> > migration v2 protocol.
> 
> Yes - I have plans as part of fixing cpuid handling for x86.
> 
> > 
> > For now I have no problem waiting and reworking in that vein as things
> > progress, clearly the blockers/caveats (logdirty support in particular)
> > are
> > going to take some time to sort out.
> > This is all based (very distantly) on some original patches from Evgeny
> > Fedotov and Junghyun Yoo (both of Samsung) however very little of that
> > code
> > remains here.
> > 
> > As well as the Xen series there will be a single Linux patch too, I'll
> > only
> > CC ARM folks on that one.
> 
> Something towards the end of this series should update
> docs/features/migration.pandoc to indicate the presence of ARM support.

I held off on doing so for now due to all the caveats and because I don't
think the series should be applied at all as it currently is. I suppose I
could put the caveats into the doc.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel