Re: [Qemu-devel] [PATCH for-1.4 0/2] fix migration failure from 1.3 due to SeaBIOS size change

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Apr 30, 2013 at 11:22:41AM -0500, Anthony Liguori wrote:
> "Michael S. Tsirkin" <mst@redhat.com> writes:
> 
> > On Tue, Feb 05, 2013 at 09:08:20PM -0600, Anthony Liguori wrote:
> >> Michael Roth <mdroth@linux.vnet.ibm.com> writes:
> >> 
> >> > Migration from 1.3 currently fails due to a mismatch between the expected size
> >> > of 256KB and the received size of 128KB for seabios. This series attempts to
> >> > fix the issue by allowing a resize function to be registered to handle resizing
> >> > memory blocks to accomodate the source's, and using that to re-initialize
> >> > "pc.bios" and "isa-bios" MemoryRegions based on the size of the incoming block.
> >> >
> >> > There are also 2 potential alternatives to this patch set:
> >> >
> >> > 1) When I compile seabios 1.7.2 locally I end up with a 128KB binary.
> >> 
> >> So do I on Fedora 18.  Unless someone has a reason why this is a bad
> >> thing, I'll commit a 128kb binary and we can avoid this problem for 1.4.
> >> 
> >> Obviously, we need to figure it out but that at least buys us some time...
> >> 
> >> Regards,
> >> 
> >> Anthony Liguori
> >
> > So we never bothered with this patch, and so it's still out.
> > How about we apply this for 1.5?
> 
> This problem occurred because Gerd built SeaBIOS on RHEL6.x which
> resulted in a 256k binary.
> 
> On F18, we're still fitting in 128k.  There is no reason to worry about
> BIOS size right now.
> 
> At some point, we will need to split into two separate BIOSes--a 128Kb
> and a 256Kb.  This can be handled by having two SeaBIOS configuration
> files where one has a reduced feature set.
> 
> We then would create an i440fx property that selects either the 128Kb or
> 256Kb BIOS.  This property can then be setup as part of machine globals.
> 
> So -M 1.4  or -M 1.5 will always use a 128Kb BIOS.  If we need to, 1.6
> may end up with a 256Kb BIOS.
> 
> > This way we, at least, won't have the problem with -M1.5 in the future.
> > Barring that - early in 1.6 cycle?
> 
> The problem with Mike's patch is that it only solves older -> newer.
> The solution I outlined above also works for newer -> older.
> 
> Regards,
> 
> Anthony Liguori

Why doesn't the same approach work for new->old?
I am guessing you are talking about migration to 1.4 and older?

> >
> > -- 
> > MST