Jim Fehlig wrote:
Daniel P. Berrange wrote:
> On Wed, Apr 10, 2013 at 04:44:43AM +0200, Marek Marczykowski wrote:
>
>
>> ---
>> src/libxl/libxl_conf.c | 72 ++++++++++++++++++++++++++++++++++++++++++++++++++
>> src/libxl/libxl_conf.h | 2 ++
>> 2 files changed, 74 insertions(+)
>>
>>
> This needs todo more than just create the config. You need to use
> the libvirt PCI APIs to validate that it is safe to assign the
> devices requested, and track which devices are given to which
> guest, to avoid concurrent assignemnt to 2 guests. Basically I'd
> expect this to borrow all the code currently in the QMEU driver
> related to the 'activePciHostdevs' and 'inactivePciHostdevs'
> data structures.
>
>
Right. Chunyan already posted such a patch series [1], but Laine noted
that it would be useful to maintain the state of PCI device assignment
for coordination among multiple drivers. Chunyan posted a design [2]
for a "hostdev passthrough driver", and IIRC she has started coding
after receiving feedback from the list.
I see that Chunyan has now posted patches for this work
https://www.redhat.com/archives/libvir-list/2013-May/msg01162.html
Could you can help review and test those?
Regards,
Jim