Re: [Qemu-devel] [PATCH v1 for-2.12 5/5] s390x/tcg: wire up STORE CHANNEL REPORT WORD

[Cornelia Huck <cohuck@redhat.com>]

On Mon, 4 Dec 2017 19:37:26 +0100
David Hildenbrand <david@redhat.com> wrote:

> On 04.12.2017 18:58, Cornelia Huck wrote:
> > On Mon, 4 Dec 2017 18:56:00 +0100
> > David Hildenbrand <david@redhat.com> wrote:
> >   
> >> On 04.12.2017 18:53, Cornelia Huck wrote:  
> >>> On Mon, 4 Dec 2017 18:34:36 +0100
> >>> David Hildenbrand <david@redhat.com> wrote:
> >>>     
> >>>> On 04.12.2017 18:22, Cornelia Huck wrote:    
> >>>>> On Mon,  4 Dec 2017 13:55:05 +0100
> >>>>> David Hildenbrand <david@redhat.com> wrote:
> >>>>>       
> >>>>>> We somehow missed that, new kernels require it.      
> >>>>>
> >>>>> Why _new_ kernels? I think we have unconditionally issued stcrw since
> >>>>> back in 2.2?      
> >>>>
> >>>> Okay, the problem is then rather related to my setup. No ccw devices ->
> >>>> no stcrw.
> >>>>
> >>>> How could we ever add ccw devices to a TCG guest then?    
> >>>
> >>> Hotplug never worked before your patches. Coldplug does not create
> >>> machine checks and thus does not trigger the guest to do STCRW. (Only
> >>> STSCH, in keeping with "all channel I/O acronyms look the same".)
> >>>     
> >>
> >> But booting Fedora 26/27 without any involved hotplugs (therefore
> >> machine checks) triggers a STCRW.
> >>
> >> That's how I originally found it. (I started playing with machine checks
> >> after I had fedora 26/27 running)
> >>  
> > 
> > Confused. Do you have a backtrace? (Guest/host)
> >   
> 
> Unfortunately not. Strange, I just tried to reproduce but can't trigger
> it. Maybe aside effect from the other BUGs I have been fixing :)
> 
> So I'll just rephrase this to "CRW machine check handling requires STCRW."
> 

Yeah, let's do that.