[libvirt] [PATCH 4/5] qemu: Check if PCI passhtrough is available in the host when starting

[Laine Stump]

On 09/23/2013 02:39 AM, Peter Krempa wrote:
> On 09/20/13 13:57, Daniel P. Berrange wrote:
>> On Fri, Sep 20, 2013 at 11:06:59AM +0200, Peter Krempa wrote:
>>> Check the presence of the selected PCI passthrough option when starting
>>> a VM.
>>> ---
>>>  src/qemu/qemu_process.c | 41 +++++++++++++++++++++++++++++++++++++++++
>>>  1 file changed, 41 insertions(+)
>>>
>>> diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c
>>> index dd16f6c..e36ab99 100644
>>> --- a/src/qemu/qemu_process.c
>>> +++ b/src/qemu/qemu_process.c
>>> @@ -3481,6 +3481,8 @@ int qemuProcessStart(virConnectPtr conn,
>>>      unsigned int stop_flags;
>>>      virQEMUDriverConfigPtr cfg;
>>>      virCapsPtr caps = NULL;
>>> +    int supportsPassthroughKVM = -1;
>>> +    int supportsPassthroughVFIO = -1;
>>>
>>>      /* Okay, these are just internal flags,
>>>       * but doesn't hurt to check */
>>> @@ -3713,6 +3715,45 @@ int qemuProcessStart(virConnectPtr conn,
>>>          goto cleanup;
>>>      }
>>>
>>> +    /* assign defaults for hostdev passthrough */
>>> +    for (i = 0; i < vm->def->nhostdevs; i++) {
>>> +        virDomainHostdevDefPtr hostdev = vm->def->hostdevs[i];
>>> +
>>> +        if (hostdev->mode == VIR_DOMAIN_HOSTDEV_MODE_SUBSYS &&
>>> +            hostdev->source.subsys.type == VIR_DOMAIN_HOSTDEV_SUBSYS_TYPE_PCI) {
>>> +            int *backend = &hostdev->source.subsys.u.pci.backend;
>>> +
>>> +            /* cache host state of passthrough support */
>>> +            if (supportsPassthroughKVM == -1 || supportsPassthroughVFIO == -1) {
>>> +                supportsPassthroughKVM = qemuHostdevHostSupportsPassthroughLegacy();
>>> +                supportsPassthroughVFIO = qemuHostdevHostSupportsPassthroughVFIO();
>>> +            }
>>> +
>>> +            switch ((virDomainHostdevSubsysPciBackendType) *backend) {
>>> +            case VIR_DOMAIN_HOSTDEV_PCI_BACKEND_VFIO:
>>> +                if (!supportsPassthroughVFIO) {
>>> +                    virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
>>> +                                   _("host doesn't support VFIO PCI passthrough"));
>>> +                    goto cleanup;
>>> +                }
>>> +                break;
>>> +
>>> +            case VIR_DOMAIN_HOSTDEV_PCI_BACKEND_DEFAULT:
>>> +            case VIR_DOMAIN_HOSTDEV_PCI_BACKEND_KVM:
>>> +                if (!supportsPassthroughKVM) {
>>> +                    virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
>>> +                                   _("host doesn't support legacy PCI passthrough"));
>>> +                    goto cleanup;
>>> +                }
>>> +
>>> +                break;
>>> +
>>> +            case VIR_DOMAIN_HOSTDEV_PCI_BACKEND_TYPE_LAST:
>>> +                break;
>>> +            }
>>> +        }
>>> +    }
>> It seems like the body of this loop ought to be shared with the hotplug
>> code, in a helper function, so we don't have so much extra code in this
>> start method
> Yeah, the basic idea of the code is the same between those two places.
> The issue with the hotplug code path is that it's using only one device,
> whereas here we iterate through a list. I will change this to a common
> func where the hotplug code will pass a fake array of just the single
> device.

That's exactly how qemuPrepareHostdevPCIDevices() is used. Maybe this
should just be another part of that function?