Re: [RFC 1/2] vhost: IFC VF hardware operation layer

[Jason Wang <jasowang@redhat.com>]

On 2019/10/21 下午5:57, Zhu, Lingshan wrote:
>
> On 10/16/2019 4:45 PM, Jason Wang wrote:
>>
>> On 2019/10/16 上午9:30, Zhu Lingshan wrote:
>>> + */
>>> +#define IFCVF_TRANSPORT_F_START 28
>>> +#define IFCVF_TRANSPORT_F_END   34
>>> +
>>> +#define IFC_SUPPORTED_FEATURES \
>>> +        ((1ULL << VIRTIO_NET_F_MAC)            | \
>>> +         (1ULL << VIRTIO_F_ANY_LAYOUT)            | \
>>> +         (1ULL << VIRTIO_F_VERSION_1) | \
>>> +         (1ULL << VHOST_F_LOG_ALL)            | \
>>
>>
>> Let's avoid using VHOST_F_LOG_ALL, using the get_mdev_features() 
>> instead.
> Thanks, I will remove VHOST_F_LOG_ALL
>>
>>
>>> +         (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE)        | \
>>> +         (1ULL << VIRTIO_NET_F_CTRL_VQ)            | \
>>> +         (1ULL << VIRTIO_NET_F_STATUS)            | \
>>> +         (1ULL << VIRTIO_NET_F_MRG_RXBUF)) /* not fully supported */
>>
>>
>> Why not having VIRTIO_F_IOMMU_PLATFORM and VIRTIO_F_ORDER_PLATFORM?
>
> I will add VIRTIO_F_ORDER_PLATFORM, for VIRTIO_F_IOMMU_PLATFORM, if we 
> add this bit, QEMU may enable viommu, can cause troubles in LM


Qemu has mature support of vIOMMU support for VFIO device, it can shadow 
IO page tables and setup them through DMA ioctl of vfio containers. Any 
issue you saw here?

Btw, to test them quickly, you can implement set_config/get_config and 
test them through virtio-mdev/kernel drivers as well.

Thanks


> (through we don't support LM in this version driver)
>
> Thanks,
> BR
> Zhu Lingshan
>>
>> Thanks
>>