[libvirt] [PATCH v2 RESEND 03/12] conf: Introduce a new PCI address extension flag

[Yi Min Zhao]

在 2018/7/25 下午9:31, Andrea Bolognani 写道:
> On Wed, 2018-07-25 at 16:58 +0800, Yi Min Zhao wrote:
> [...]
>>>> +    case VIR_DOMAIN_DEVICE_CONTROLLER:
>>>> +    case VIR_DOMAIN_DEVICE_DISK:
>>>> +    case VIR_DOMAIN_DEVICE_LEASE:
>>>> +    case VIR_DOMAIN_DEVICE_FS:
>>>> +    case VIR_DOMAIN_DEVICE_NET:
>>>> +    case VIR_DOMAIN_DEVICE_INPUT:
>>>> +    case VIR_DOMAIN_DEVICE_SOUND:
>>>> +    case VIR_DOMAIN_DEVICE_VIDEO:
>>>> +    case VIR_DOMAIN_DEVICE_HOSTDEV:
>>>> +    case VIR_DOMAIN_DEVICE_WATCHDOG:
>>>> +    case VIR_DOMAIN_DEVICE_GRAPHICS:
>>>> +    case VIR_DOMAIN_DEVICE_HUB:
>>>> +    case VIR_DOMAIN_DEVICE_REDIRDEV:
>>>> +    case VIR_DOMAIN_DEVICE_SMARTCARD:
>>>> +    case VIR_DOMAIN_DEVICE_MEMBALLOON:
>>>> +    case VIR_DOMAIN_DEVICE_NVRAM:
>>>> +    case VIR_DOMAIN_DEVICE_RNG:
>>>> +    case VIR_DOMAIN_DEVICE_SHMEM:
>>>> +    case VIR_DOMAIN_DEVICE_TPM:
>>>> +    case VIR_DOMAIN_DEVICE_PANIC:
>>>> +    case VIR_DOMAIN_DEVICE_MEMORY:
>>>> +    case VIR_DOMAIN_DEVICE_IOMMU:
>>>> +    case VIR_DOMAIN_DEVICE_VSOCK:
>>> Did you validate that all of the above can be used with zPCI?
>> Yes, I did. But how far can zPCI be used is a question. I make sure that
>> if their address
>> type can be PCI type zPCI address could be expanded. But we don't
>> guarantee they can
>> be really used in qemu. Like RNG device can't be used because it doesn't
>> support MSIx
>> which is required on S390.
> If that's the case, I'm not sure the current solution is what we
> want: if someone creates a guest and includes a RNG device in the
> configuration, they probably expect it to, well, work. IIUC, with
> the current implementation they would get a non-working RNG device
> instead, which certainly feels suboptimal.
Actually the guest can't startup with RNG device instead of a running guest
with a non-working RNG device and qemu would report error. The user can
get the information regarding startup failure. I think this is proper.
If RNG could support MSIx in future, we won't need to do any thing in 
Libvirt.
>
>>> [...]
>>>> +static virDomainPCIAddressExtensionFlags
>>>> +qemuDomainDeviceCalculatePCIAddressExtensionFlags(virQEMUCapsPtr qemuCaps,
>>>> +                                                  virDomainDeviceDefPtr dev)
>>>> +{
>>>> +    virDomainPCIAddressExtensionFlags extFlags = 0;
>>>> +
>>>> +    if (virQEMUCapsGet(qemuCaps, QEMU_CAPS_DEVICE_ZPCI) &&
>>>> +        qemuDomainDeviceSupportZPCI(dev))
>>>> +        extFlags |= VIR_PCI_ADDRESS_EXTENSION_ZPCI;
>>> The libvirt code style guidelines[1] state that this should be
>>> formatted as
>>>
>>>     if (virQEMUCapsGet(qemuCaps, QEMU_CAPS_DEVICE_ZPCI) &&
>>>         qemuDomainDeviceSupportZPCI(dev)) {
>>>         extFlags |= VIR_PCI_ADDRESS_EXTENSION_ZPCI;
>>>     }
>>>
>>> [1] https://libvirt.org/hacking.html
>> I see a lot of code in libvirt use this style. Is it new guideline?
> It's certainly been around for the past 3+ years. There's a lot of
> code in libvirt that's *way* older than that, though :)
>