Virtualization - Oct 2019 - [PATCH V3 0/7] mdev based hardware virtio offloading support

On 2019/10/15 ??1:49, Stefan Hajnoczi wrote:
> On Fri, Oct 11, 2019 at 04:15:50PM +0800, Jason Wang wrote:
>> There are hardware that can do virtio datapath offloading while having
>> its own control path. This path tries to implement a mdev based
>> unified API to support using kernel virtio driver to drive those
>> devices. This is done by introducing a new mdev transport for virtio
>> (virtio_mdev) and register itself as a new kind of mdev driver. Then
>> it provides a unified way for kernel virtio driver to talk with mdev
>> device implementation.
>>
>> Though the series only contains kernel driver support, the goal is to
>> make the transport generic enough to support userspace drivers. This
>> means vhost-mdev[1] could be built on top as well by resuing the
>> transport.
>>
>> A sample driver is also implemented which simulate a virito-net
>> loopback ethernet device on top of vringh + workqueue. This could be
>> used as a reference implementation for real hardware driver.
>>
>> Consider mdev framework only support VFIO device and driver right now,
>> this series also extend it to support other types. This is done
>> through introducing class id to the device and pairing it with
>> id_talbe claimed by the driver. On top, this seris also decouple
>> device specific parents ops out of the common ones.
> I was curious so I took a quick look and posted comments.
>
> I guess this driver runs inside the guest since it registers virtio
> devices?

It could run in either guest or host. But the main focus is to run in 
the host then we can use virtio drivers in containers.

>
> If this is used with physical PCI devices that support datapath
> offloading then how are physical devices presented to the guest without
> SR-IOV?

We will do control path meditation through vhost-mdev[1] and 
vhost-vfio[2]. Then we will present a full virtio compatible ethernet 
device for guest.

SR-IOV is not a must, any mdev device that implements the API defined in 
patch 5 can be used by this framework.

Thanks

[1] https://lkml.org/lkml/2019/9/26/15

[2] https://patchwork.ozlabs.org/cover/984763/

>
> Stefan

On Tue, Oct 15, 2019 at 11:37:17AM +0800, Jason Wang
wrote:
> 
> On 2019/10/15 ??1:49, Stefan Hajnoczi wrote:
> > On Fri, Oct 11, 2019 at 04:15:50PM +0800, Jason Wang wrote:
> > > There are hardware that can do virtio datapath offloading while
having
> > > its own control path. This path tries to implement a mdev based
> > > unified API to support using kernel virtio driver to drive those
> > > devices. This is done by introducing a new mdev transport for
virtio
> > > (virtio_mdev) and register itself as a new kind of mdev driver.
Then
> > > it provides a unified way for kernel virtio driver to talk with
mdev
> > > device implementation.
> > > 
> > > Though the series only contains kernel driver support, the goal
is to
> > > make the transport generic enough to support userspace drivers.
This
> > > means vhost-mdev[1] could be built on top as well by resuing the
> > > transport.
> > > 
> > > A sample driver is also implemented which simulate a virito-net
> > > loopback ethernet device on top of vringh + workqueue. This could
be
> > > used as a reference implementation for real hardware driver.
> > > 
> > > Consider mdev framework only support VFIO device and driver right
now,
> > > this series also extend it to support other types. This is done
> > > through introducing class id to the device and pairing it with
> > > id_talbe claimed by the driver. On top, this seris also decouple
> > > device specific parents ops out of the common ones.
> > I was curious so I took a quick look and posted comments.
> > 
> > I guess this driver runs inside the guest since it registers virtio
> > devices?
> 
> 
> It could run in either guest or host. But the main focus is to run in the
> host then we can use virtio drivers in containers.
> 
> 
> > 
> > If this is used with physical PCI devices that support datapath
> > offloading then how are physical devices presented to the guest
without
> > SR-IOV?
> 
> 
> We will do control path meditation through vhost-mdev[1] and vhost-vfio[2].
> Then we will present a full virtio compatible ethernet device for guest.
> 
> SR-IOV is not a must, any mdev device that implements the API defined in
> patch 5 can be used by this framework.
What I'm trying to understand is: if you want to present a virtio-pci
device to the guest (e.g. using vhost-mdev or vhost-vfio), then how is
that related to this patch series?

Does this mean this patch series is useful mostly for presenting virtio
devices to containers or the host?

Stefan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL:
<http://lists.linuxfoundation.org/pipermail/virtualization/attachments/20191015/cc4a5d13/attachment.sig>

On 2019/10/15 ??10:37, Stefan Hajnoczi wrote:
> On Tue, Oct 15, 2019 at 11:37:17AM +0800, Jason Wang wrote:
>> On 2019/10/15 ??1:49, Stefan Hajnoczi wrote:
>>> On Fri, Oct 11, 2019 at 04:15:50PM +0800, Jason Wang wrote:
>>>> There are hardware that can do virtio datapath offloading while
having
>>>> its own control path. This path tries to implement a mdev based
>>>> unified API to support using kernel virtio driver to drive
those
>>>> devices. This is done by introducing a new mdev transport for
virtio
>>>> (virtio_mdev) and register itself as a new kind of mdev driver.
Then
>>>> it provides a unified way for kernel virtio driver to talk with
mdev
>>>> device implementation.
>>>>
>>>> Though the series only contains kernel driver support, the goal
is to
>>>> make the transport generic enough to support userspace drivers.
This
>>>> means vhost-mdev[1] could be built on top as well by resuing
the
>>>> transport.
>>>>
>>>> A sample driver is also implemented which simulate a virito-net
>>>> loopback ethernet device on top of vringh + workqueue. This
could be
>>>> used as a reference implementation for real hardware driver.
>>>>
>>>> Consider mdev framework only support VFIO device and driver
right now,
>>>> this series also extend it to support other types. This is done
>>>> through introducing class id to the device and pairing it with
>>>> id_talbe claimed by the driver. On top, this seris also
decouple
>>>> device specific parents ops out of the common ones.
>>> I was curious so I took a quick look and posted comments.
>>>
>>> I guess this driver runs inside the guest since it registers virtio
>>> devices?
>>
>> It could run in either guest or host. But the main focus is to run in
the
>> host then we can use virtio drivers in containers.
>>
>>
>>> If this is used with physical PCI devices that support datapath
>>> offloading then how are physical devices presented to the guest
without
>>> SR-IOV?
>>
>> We will do control path meditation through vhost-mdev[1] and
vhost-vfio[2].
>> Then we will present a full virtio compatible ethernet device for
guest.
>>
>> SR-IOV is not a must, any mdev device that implements the API defined
in
>> patch 5 can be used by this framework.
> What I'm trying to understand is: if you want to present a virtio-pci
> device to the guest (e.g. using vhost-mdev or vhost-vfio), then how is
> that related to this patch series?

This series introduce some infrastructure that would be used by vhost-mdev:

1) allow new type of mdev devices/drivers other than vfio (through 
class_id and device ops)

2) a set of virtio specific callbacks that will be used by both 
vhost-mdev and virtio-mdev defined in patch 5

Then vhost-mdev can be implemented on top: a new mdev class id but reuse 
the callback defined in 2. Through this way the parent can provides a 
single set of callbacks (device ops) for both kernel virtio driver 
(through virtio-mdev) or userspace virtio driver (through vhost-mdev).

>
> Does this mean this patch series is useful mostly for presenting virtio
> devices to containers or the host?

Patch 6 is mainly for bare metal or container use case, through it could 
be used in guest as well. Patch 7 is a sample virtio mdev device 
implementation. Patch 1 - 5 was the infrastructure for implementing 
types other than vfio, the first user is virito-mdev, then Tiwei's 
vhost-mdev and Parav's mlx5 mdev.

Thanks

>
> Stefan