Re: [Qemu-devel] [PATCH 09/10] piix: do not raise irq while loading vmstate

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

On Wed, Sep 10, 2014 at 11:05:39AM +0200, Paolo Bonzini wrote:
> Il 10/09/2014 10:51, Peter Maydell ha scritto:
> > > What is not okay (and I think it should be a rule) is to touch other
> > > devices from post_load, unless you know that they are deserialized
> > > first.  For example it's okay for a PCI device to talk to the parent
> > > bridge in its post_load function.
> > 
> > I don't think it's right to talk to another device even if you do
> > know it's deserialized first. Talking to it might make it change
> > its state, which would be wrong (since its correct state is
> > the state it's just deserialized). I would suggest the rule should
> > be "never do something that can change the state of another
> > device in post-load".
> 
> That's harder to do, but if it is possible to do it, it would be great
> as well.
> 
> It would not surprise me to find a case where the parent device actually
> _expects_ the children's post_load to inform it about something, instead
> of serializing that part of state on its own.
> 
> Paolo

Absolutely, I don't think we can require that.

For example, at the moment, for PCI bridges, we serialize the
state of all interrupt lines, but that's just a function
of all devices connected to each line.
So we are transmitting redundant information, and I have plans
to discard that and recompute parent state based on child state.



> > (We have similar issues with reset, except worse in that we
> > don't have a coherent rule to cause everything to come out
> > of reset in the right state.)