Re: kvm-kmod.git

[Ryan Harper <ryanh@us.ibm.com>]

* Avi Kivity <avi@redhat.com> [2009-06-01 12:04]:
> Ryan Harper wrote:
> >>>I also wonder what happens if one tries to build on
> >>>a machine with kvm built into kernel. Ideally one would get
> >>>a clear error message.
> >>> 
> >>>      
> >>kvm-kmod is really designed for those running on pre-kvm distro kernels, 
> >>and for those testing newer kvm versions on distro kernels.  If you can 
> >>compile your own kernel, download kvm.git and run that.
> >>    
> >
> >So no way of having kvm-kmod sync in kvm.git bits and still build
> >against the current running kernel?
> >  
> 
> We must be miscommunicating, what you describe is kvm-kmod.git's sole 
> purpose in life.
> 
> >I wanted to test out 2.6.29-maint kernel modules and in
> >kvm-userspace/kernel, I could checkout maint/2.6.29 in kvm.git and then
> >make LINUX= sync.  I couldn't quite figure out how to do that with
> >kvm-kmod since when I do a ./configure --kerneldir -- it's building
> >modules *for* whatever kernel is at kerneldir rather than just syncing
> >in the kvm bits from that kerneldir and then building modules against
> >the running kernel.
> >  
> 
> 'make sync' will copy the kvm bits from the linux-2.6 directory, or 
> $(LINUX) if you specify that to make.
> 
> So:
>  LINUX= is for the kvm sources (further controlled by whatever branch 
> is checked out)
>  --kerneldir= is for the host kernel

Ah, excellent, I didn't see it documented on the Code page of the wiki
and I blindly assumed that it went away.  Thanks for correction.
> 
> >The reason I'm digging is that I'm seeing 64-bit migration failing on
> >maint/2.6.29 kvm modules, but not on upstream kvm kernel bits and I
> >wanted to bisect to find where 64-bit migration is fixed so I can
> >suggest what to pull into maint/2.6.29.
> >  
> 
> Thanks, that's helpful.  How does it fail?  maybe I can supply an 
> educated guess.

Migration succeeds, but source and target are frozen (though monitor of
guest is still interactive. 

Nothing interesting in host dmesg, and of course no output from either
guest (they seemed locked).  Reproduced this sort of hang for any number
of guests (RHEL4u7/8, SLES10 SP2, SLES11, Win2k3-r2, win2k8, etc.) as
long as they are 64-bit

I think Gleb was working on a irq-chip fix for migration, so I need to
try with -no-kvm-irqchip as well to see if
that helps.

-- 
Ryan Harper
Software Engineer; Linux Technology Center
IBM Corp., Austin, Tx
ryanh@us.ibm.com