[Kimchi-devel] [PATCH V2] Add vfio driver as default for powerkvm systems.

[Aline Manera]

On 27/05/2015 13:31, Jose Ricardo Ziviani wrote:
>
>
> On 27-05-2015 13:20, Aline Manera wrote:
>>
>>
>> On 25/05/2015 13:08, Jose Ricardo Ziviani wrote:
>>>   - ppc hotplug does not work with kvm driver, it must be
>>>     vfio by default.
>>>
>>> Signed-off-by: Jose Ricardo Ziviani <joserz at linux.vnet.ibm.com>
>>> ---
>>>   src/kimchi/model/vmhostdevs.py | 6 ++++++
>>>   1 file changed, 6 insertions(+)
>>>
>>> diff --git a/src/kimchi/model/vmhostdevs.py
>>> b/src/kimchi/model/vmhostdevs.py
>>> index 964ec49..516d48f 100644
>>> --- a/src/kimchi/model/vmhostdevs.py
>>> +++ b/src/kimchi/model/vmhostdevs.py
>>> @@ -31,6 +31,7 @@ from kimchi.model.utils import get_vm_config_flag
>>>   from kimchi.model.vms import DOM_STATE_MAP, VMModel
>>>   from kimchi.rollbackcontext import RollbackContext
>>>   from kimchi.utils import kimchi_log, run_command
>>> +import platform
>>>
>>>
>>>   class VMHostDevsModel(object):
>>> @@ -164,6 +165,11 @@ class VMHostDevsModel(object):
>>>           driver = ('vfio' if DOM_STATE_MAP[dom.info()[0]] ==
>>> "shutoff" and
>>>                     self.caps.kernel_vfio else 'kvm')
>>>
>>> +        # on powerkvm systems it must be vfio driver.
>>> +        distro, _, _ = platform.linux_distribution()
>>> +        if distro == 'IBM_PowerKVM':
>>> +            driver = 'vfio'
>>
>> What happen if the kernel module for vfio is not loaded?
>
> This won't happen because vfio is not a kernel module but a feature 
> built in.
>

So why do we have a feature test (FeatureTests.kernel_support_vfio) to 
verify it?


> # grep -i vfio /boot/config-<kernel-version>
> CONFIG_VFIO_IOMMU_SPAPR_TCE=y
> CONFIG_VFIO_SPAPR_EEH=y
> CONFIG_VFIO=y
> CONFIG_VFIO_PCI=y
> CONFIG_KVM_VFIO=y
>
>
>>
>>> +
>>>           # Attach all PCI devices in the same IOMMU group
>>>           dev_model = DeviceModel(conn=self.conn)
>>>           devs_model = DevicesModel(conn=self.conn)
>>
>