Re: [Qemu-devel] [PATCH] migration: notify runstate immediately before vcpu stops

[Yan Zhao <yan.y.zhao@intel.com>]

On Tue, Jul 16, 2019 at 06:33:41PM +0800, Peter Xu wrote:
> On Tue, Jul 16, 2019 at 03:57:49AM -0400, Yan Zhao wrote:
> > On Tue, Jul 16, 2019 at 03:50:25PM +0800, Peter Xu wrote:
> > > On Tue, Jul 16, 2019 at 03:29:19AM -0400, Yan Zhao wrote:
> > > > On Tue, Jul 16, 2019 at 03:23:16PM +0800, Peter Xu wrote:
> > > > > On Tue, Jul 16, 2019 at 03:10:42PM +0800, Yan Zhao wrote:
> > > > > > for some devices to do live migration, it is needed to do something
> > > > > > immediately before vcpu stops. add a notification here.
> > > > > 
> > > > > Hi, Yan,
> > > > > 
> > > > > Could I ask for a more detailed commit message here?  E.g., what is
> > > > > "some devices"?  And, what's the problem behind?
> > > > >
> > > > hi Peter,
> > > > 
> > > > Some devices refer to assigned devices, like NICs.
> > > > For assigned devices to do live migration, it is sometimes required that
> > > > source device is stopped before stopping source vcpus. vcpus can do some
> > > > final cleanups (like handling interrupt) in that case.
> > > 
> > > I see, so this is a prerequisite of another work?
> > 
> > Yes.
> > > 
> > > IMHO it would make more sense to have this patch to be with that
> > > patchset, then it'll justify itself with reasoning.  Unless I
> > > misunderstood - this single patch seems to help nothing if as a
> > > standalone one.
> > 
> > It would be better. but this patch alone is also somewhat general,
> > as it only adds an extra notification and wouldn't impact others.
> > 
> > Only after this patch is upstreamed, can VFIO live migration have a
> > second choice for those devices of special requirements.
> > 
> > Hope it can get understanding from you:)
> 
> No you don't need my understanding, as long as the maintainer likes it
> it's good enough, so you probably only need to persuade the
> maintainers. :)
> 
> But again for me I would prefer patch like this to simply be the first
> patch of your live migration series.  There can be some exceptions
> like that when the prerequisite is too big so we'd better split them
> out to do things step by step, but this patch (which is a oneliner)
> should not be the case.
> 
> Thanks,
>

ok. Thank you, Peter :)

> > 
> > Or do you think I need to change the commit message a little to unbind
> > from migration?
> > 
> > Thanks
> > Yan
> > 
> 
> -- 
> Peter Xu