Re: [Qemu-devel] [PATCH] Migration compatibility for serial

[Michael S. Tsirkin]

On Wed, Jun 17, 2015 at 02:20:30PM +0200, Paolo Bonzini wrote:
> 
> 
> On 17/06/2015 13:58, Michael S. Tsirkin wrote:
> > > No, it doesn't.  The guest misbehaves maybe, but the migration format is
> > > not affected.
> > 
> > I just tried, set prog_if to different values, sure it failed.
> 
> How so?  It's just a byte in config space.  But even then, fixing
> migration is just a side effect of keeping config space consistent for a
> given machine type (i.e. not changing hardware type under the guest's feet).

David's patches are also guest visible, are they not?
We are losing state guest can indirectly observe, right?

> > Here's another one, at random:
> > 
> > Author: Michael S. Tsirkin <address@hidden>
> > Date:   Thu Feb 14 19:11:27 2013 +0200
> > 
> >     e1000: unbreak the guest network migration to 1.3
> >     
> >     QEMU 1.3 does not emulate the link auto negotiation, so if migrate to a
> >     1.3 machine during link auto negotiation, the guest link will be set to 
> > down.
> >     Fix this by just disabling auto negotiation for 1.3 and older.
> >     
> >     Signed-off-by: Michael S. Tsirkin <address@hidden>
> 
> Okay, that's an interesting one, and there's a similar one for e1000
> interrupt mitigation.
> 
> The interesting point is that in both cases the bug compatibility
> extends to other behavior of the device, i.e. more than just migration.

But the guest registers are exactly the same. It is only guest-visible
indirectly, as timing of link up events.
So why not keep auto-negotiation running correctly?  Because
we can't keep it running across migration.

> I would even say that bug-compatibility of migration is just a side
> effect, not the primary end.  For interrupt mitigation, it was not
> enabled on older machine types in the first place, because it could
> break guests.  Keeping backwards migration working was just a side
> effect; simply, checking "s->compat_flags & E1000_FLAG_MIT" is the only
> sensible way to write e1000_mit_state_needed.  Auto negotiation should
> have been done the same way, which is what your patch did.
> 
> Paolo

True, David's patches only trigger if migration happens.
I am guessing there is no need to touch other paths,
but I am not very familiar with the hardware in question.

-- 
MST