Re: [Qemu-devel] Crash when deleting the diag288 watchdog

[Cornelia Huck]

On Wed, 16 Aug 2017 15:36:42 +0200
Thomas Huth <address@hidden> wrote:

> On 16.08.2017 11:23, Cornelia Huck wrote:
> > On Wed, 16 Aug 2017 07:05:37 +0200
> > Thomas Huth <address@hidden> wrote:
> >   
> >>  Hi,
> >>
> >> I recently noticed that QEMU abort()s if you try to remove the diag288
> >> watchdog. For example:
> >>
> >> $ qemu-system-s390x -nographic -nodefaults -S -monitor stdio
> >> QEMU 2.9.92 monitor - type 'help' for more information
> >> (qemu) device_add diag288,id=x
> >> (qemu) device_del x
> >> **
> >> ERROR:/home/thuth/devel/qemu/qdev-monitor.c:872:qdev_unplug: assertion
> >> failed: (hotplug_ctrl)
> >> Aborted (core dumped)
> >>
> >> This is ugly, can we fix this somehow? For example, should the diag288
> >> device be hot-pluggable at all, or can it only be used via the
> >> "-watchdog" parameter instead? In the latter case, we could simply mark
> >> the device with "user_creatable = false", I guess?  
> > 
> > I don't think the diag288 watchdog should be hotpluggable. IIUC, it is
> > simply present on z/VM (and I don't think it's different on LPAR, but I
> > could not find docs for that). So yes, user_creatable = false sounds
> > like the right thing to do.  
> 
> I tried that now, but it does not work - when I specify the -watchdog
> parameter, I get a "Parameter 'driver' expects pluggable device type"
> with that modification.
> I think the best option is to use "hotpluggable = false" instead...

OK, let's try that.

> 
> > Related: We currently only handle diag288 via kvm if I did not miss
> > something. It probably makes sense to wire it up for tcg as well.  
> 
> Sounds like a good idea, yes... are you going to send a patch, or shall
> I put it on my todo list?

I'll deal with the diag288 stuff, I started to look into it anyway.