Re: [PATCH v2 1/4] x86/MSI-X: be more careful during teardown

[Ian Campbell <ian.campbell@citrix.com>]

On Mon, 2015-04-13 at 10:11 +0100, Jan Beulich wrote:
> >>> On 02.04.15 at 18:49, <stefano.stabellini@eu.citrix.com> wrote:
> > In any case we should make it clear somewhere who is supposed to write
> > to the command register (and other PCI reigsters) at any given time,
> > otherwise it would be very easy for a new kernel update to break the
> > hypervisor and we wouldn't even know why it happened.
> 
> We should, but at this point in time this is going to be rather
> problematic. Such a separation of responsibilities should have been
> done before all the pass-through code got written.

Wasn't Stefano just asking to write down the current semantics, problems
and all?

I think that would be a wonderful idea even if for some parts the
documentation is things like:
      * "we don't really know";
      * "there are multiple parties X, Y and Z involved and it's a
        mess";
      * "currently qemu, but really ought to be Xen";
      * etc.

I think half the problem with coming up with a plan to move forward is
that exactly where we are now isn't very clear, so having a thing to
refer to there might be helpful above and beyond it being a good idea to
write these things down on principal.

Ian.