Re: [Qemu-devel] [PATCH 00/10] pci: Partial conversion to realize

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

On Mon, Jan 19, 2015 at 05:00:38PM +0100, Markus Armbruster wrote:
> Andreas Färber <afaerber@suse.de> writes:
> 
> > Hi Markus,
> >
> > Am 19.01.2015 um 15:52 schrieb Markus Armbruster:
> >> I posted this series as RFC back in October, but it depended on
> >> patches then still under review, so I put it aside, and promptly
> >> forgot.  Fortunately, rebasing and updating it wasn't much trouble.
> >
> > A question I had brought up for the RFC was that of testing: Did you
> > check which of your conversions are covered by QTest or where we may
> > first need additional test cases to verify this doesn't regress?
> >
> > As a reminder, when converting ISA devices, I accepted patches that were
> > only tested with one of two serial devices contained in a file, promptly
> > breaking the other in some target. I'd like to avoid that for PCI. ;)
> 
> I got this after I sent my reply to your original question.
> 
> Perhaps we could create a qtest that tries to -device every known PCI
> device: grab them from
> 
>   { "execute": "qom-list-types", "arguments": { "implements": "pci-device" } }
> 
> Some require properties, typically backends.  We could introspect with
> device-list-properties and try to supply common ones, like "drive",
> "chardev", "netdev".
> 
> Doing it as a big device_add / device_del orgy would be cute, except
> device_del requires guest cooperation, complicating the test.  We could
> special-case device_del before guest starts if we wanted.

You can drive the hot-unplug register from qemu monitor
without running guests. Target specific.
Alternatively, e.g. PC completes unplug on reset.
Maybe we should teach all targets to do that, then
you can just reset to complete unplug.

-- 
MST