Re: [PATCH v2] vhost: introduce mdev based hardware backend

From: Jason Wang
Date: Thu Oct 24 2019 - 04:03:44 EST



On 2019/10/24 äå12:21, Tiwei Bie wrote:
On Wed, Oct 23, 2019 at 06:29:21PM +0800, Jason Wang wrote:
On 2019/10/23 äå6:11, Tiwei Bie wrote:
On Wed, Oct 23, 2019 at 03:25:00PM +0800, Jason Wang wrote:
On 2019/10/23 äå3:07, Tiwei Bie wrote:
On Wed, Oct 23, 2019 at 01:46:23PM +0800, Jason Wang wrote:
On 2019/10/23 äå11:02, Tiwei Bie wrote:
On Tue, Oct 22, 2019 at 09:30:16PM +0800, Jason Wang wrote:
On 2019/10/22 äå5:52, Tiwei Bie wrote:
This patch introduces a mdev based hardware vhost backend.
This backend is built on top of the same abstraction used
in virtio-mdev and provides a generic vhost interface for
userspace to accelerate the virtio devices in guest.

This backend is implemented as a mdev device driver on top
of the same mdev device ops used in virtio-mdev but using
a different mdev class id, and it will register the device
as a VFIO device for userspace to use. Userspace can setup
the IOMMU with the existing VFIO container/group APIs and
then get the device fd with the device name. After getting
the device fd of this device, userspace can use vhost ioctls
to setup the backend.

Signed-off-by: Tiwei Bie <tiwei.bie@xxxxxxxxx>
---
This patch depends on below series:
https://lkml.org/lkml/2019/10/17/286

v1 -> v2:
- Replace _SET_STATE with _SET_STATUS (MST);
- Check status bits at each step (MST);
- Report the max ring size and max number of queues (MST);
- Add missing MODULE_DEVICE_TABLE (Jason);
- Only support the network backend w/o multiqueue for now;
Any idea on how to extend it to support devices other than net? I think we
want a generic API or an API that could be made generic in the future.

Do we want to e.g having a generic vhost mdev for all kinds of devices or
introducing e.g vhost-net-mdev and vhost-scsi-mdev?
One possible way is to do what vhost-user does. I.e. Apart from
the generic ring, features, ... related ioctls, we also introduce
device specific ioctls when we need them. As vhost-mdev just needs
to forward configs between parent and userspace and even won't
cache any info when possible,
So it looks to me this is only possible if we expose e.g set_config and
get_config to userspace.
The set_config and get_config interface isn't really everything
of device specific settings. We also have ctrlq in virtio-net.
Yes, but it could be processed by the exist API. Isn't it? Just set ctrl vq
address and let parent to deal with that.
I mean how to expose ctrlq related settings to userspace?

I think it works like:

1) userspace find ctrl_vq is supported

2) then it can allocate memory for ctrl vq and set its address through
vhost-mdev

3) userspace can populate ctrl vq itself
I see. That is to say, userspace e.g. QEMU will program the
ctrl vq with the existing VHOST_*_VRING_* ioctls, and parent
drivers should know that the addresses used in ctrl vq are
host virtual addresses in vhost-mdev's case.


That's really good point. And that means parent needs to differ vhost from virtio. It should work. But is there any chance to use DMA address? I'm asking since the API then tends to be device specific.




I think it might be better to do
this in one generic vhost-mdev module.
Looking at definitions of VhostUserRequest in qemu, it mixed generic API
with device specific API. If we want go this ways (a generic vhost-mdev),
more questions needs to be answered:

1) How could userspace know which type of vhost it would use? Do we need to
expose virtio subsystem device in for userspace this case?

2) That generic vhost-mdev module still need to filter out unsupported
ioctls for a specific type. E.g if it probes a net device, it should refuse
API for other type. This in fact a vhost-mdev-net but just not modularize it
on top of vhost-mdev.


- Some minor fixes and improvements;
- Rebase on top of virtio-mdev series v4;
[...]
+
+static long vhost_mdev_get_features(struct vhost_mdev *m, u64 __user *featurep)
+{
+ if (copy_to_user(featurep, &m->features, sizeof(m->features)))
+ return -EFAULT;
As discussed in previous version do we need to filter out MQ feature here?
I think it's more straightforward to let the parent drivers to
filter out the unsupported features. Otherwise it would be tricky
when we want to add more features in vhost-mdev module,
It's as simple as remove the feature from blacklist?
It's not really that easy. It may break the old drivers.
I'm not sure I understand here, we do feature negotiation anyhow. For old
drivers do you mean the guest drivers without MQ?
For old drivers I mean old parent drivers. It's possible
to compile old drivers on new kernels.

Yes, but if old parent driver itself can not support MQ it should just not
advertise that feature.


I'm not quite sure how will we implement MQ support in
vhost-mdev.

Yes, that's why I ask here. I think we want the vhost-mdev to be generic
which means it's better not let vhost-mdev to know anything which is device
specific. So this is a question that should be considered.
+1


If we need to introduce new virtio_mdev_device_ops
callbacks and an old driver exposed the MQ feature,
then the new vhost-mdev will see this old driver expose
MQ feature but not provide corresponding callbacks.ean

That's exact the issue which current API can not handle, so that's why I
suggest to filter MQ out for vhost-mdev.

And in the future, we can:

1) invent new ioctls and convert them to config access or

2) just exposing config for userspace to access (then vhost-mdev work much
more similar to virtio-mdev).


i.e. if
the parent drivers may expose unsupported features and relay on
vhost-mdev to filter them out, these features will be exposed
to userspace automatically when they are enabled in vhost-mdev
in the future.
The issue is, it's only that vhost-mdev knows its own limitation. E.g in
this patch, vhost-mdev only implements a subset of transport API, but parent
doesn't know about that.

Still MQ as an example, there's no way (or no need) for parent to know that
vhost-mdev does not support MQ.
The mdev is a MDEV_CLASS_ID_VHOST mdev device. When the parent
is being developed, it should know the currently supported features
of vhost-mdev.
How can parent know MQ is not supported by vhost-mdev?
Good point. I agree vhost-mdev should filter out the unsupported
features. But in the meantime, I think drivers also shouldn't
expose unsupported features.

Exactly. But there's a case in the middle, e.g parent drivers support MQ and
virtio-mdev can do that but not vhost-mdev.
As we have different mdev class IDs between virtio-mdev and
vhost-mdev, maybe parent can leverage it to return different
sets of supported features for virtio-mdev and vhost-mdev
respectively.


Yes, that should work.

Thanks




And this allows old kenrel to work with new
parent drivers.
The new drivers should provide things like VIRTIO_MDEV_F_VERSION_1
to be compatible with the old kernels. When VIRTIO_MDEV_F_VERSION_1
is provided/negotiated, the behaviours should be consistent.
To be clear, I didn't mean a change in virtio-mdev API, I meant:

1) old vhost-mdev kernel driver that filters out MQ

2) new parent driver that support MQ


So basically we have three choices here:

1) Implement what vhost-user did and implement a generic vhost-mdev (but may
still have lots of device specific code). To support advanced feature which
requires the access to config, still lots of API that needs to be added.

2) Implement what vhost-kernel did, have a generic vhost-mdev driver and a
vhost bus on top for match a device specific API e.g vhost-mdev-net. We
still have device specific API but limit them only to device specific
module. Still require new ioctls for advanced feature like MQ.

3) Simply expose all virtio-mdev transport to userspace.
Currently, virtio-mdev transport is a set of function callbacks
defined in kernel. How to simply expose virtio-mdev transport to
userspace?
The most straightforward way is to have an 1:1 mapping between ioctl and
virito_mdev_device_ops.
Seems we are already trying to do 1:1 mapping between ioctl
and virtio_mdev_device_ops in vhost-mdev now (the major piece
missing is get_device_id/get_config/set_config).

Yes, with this we can have a device independent API. Do you think this is
better?
Yeah, I agree.

Thanks,
Tiwei

Thanks



Thanks


A generic module
without any type specific code (like virtio-mdev). No need dedicated API for
e.g MQ. But then the API will look much different than current vhost did.

Consider the limitation of 1) I tend to choose 2 or 3. What's you opinion?