Re: [RFC v4 0/3] vhost: introduce mdev based hardware backend

[Tiwei Bie <tiwei.bie@intel.com>]

On Thu, Sep 19, 2019 at 09:08:11PM +0800, Jason Wang wrote:
> On 2019/9/18 下午10:32, Michael S. Tsirkin wrote:
> > > > > So I have some questions:
> > > > > 
> > > > > 1) Compared to method 2, what's the advantage of creating a new vhost char
> > > > > device? I guess it's for keep the API compatibility?
> > > > One benefit is that we can avoid doing vhost ioctls on
> > > > VFIO device fd.
> > > Yes, but any benefit from doing this?
> > It does seem a bit more modular, but it's certainly not a big deal.
> 
> Ok, if we go this way, it could be as simple as provide some callback to
> vhost, then vhost can just forward the ioctl through parent_ops.
> 
> > 
> > > > > 2) For method 2, is there any easy way for user/admin to distinguish e.g
> > > > > ordinary vfio-mdev for vhost from ordinary vfio-mdev?
> > > > I think device-api could be a choice.
> > > Ok.
> > > 
> > > 
> > > > > I saw you introduce
> > > > > ops matching helper but it's not friendly to management.
> > > > The ops matching helper is just to check whether a given
> > > > vfio-device is based on a mdev device.
> > > > 
> > > > > 3) A drawback of 1) and 2) is that it must follow vfio_device_ops that
> > > > > assumes the parameter comes from userspace, it prevents support kernel
> > > > > virtio drivers.
> > > > > 
> > > > > 4) So comes the idea of method 3, since it register a new vhost-mdev driver,
> > > > > we can use device specific ops instead of VFIO ones, then we can have a
> > > > > common API between vDPA parent and vhost-mdev/virtio-mdev drivers.
> > > > As the above draft shows, this requires introducing a new
> > > > VFIO device driver. I think Alex's opinion matters here.
> 
> Just to clarify, a new type of mdev driver but provides dummy
> vfio_device_ops for VFIO to make container DMA ioctl work.

I see. Thanks! IIUC, you mean we can provide a very tiny
VFIO device driver in drivers/vhost/mdev.c, e.g.:

static int vfio_vhost_mdev_open(void *device_data)
{
	if (!try_module_get(THIS_MODULE))
		return -ENODEV;
	return 0;
}

static void vfio_vhost_mdev_release(void *device_data)
{
	module_put(THIS_MODULE);
}

static const struct vfio_device_ops vfio_vhost_mdev_dev_ops = {
	.name		= "vfio-vhost-mdev",
	.open		= vfio_vhost_mdev_open,
	.release	= vfio_vhost_mdev_release,
};

static int vhost_mdev_probe(struct device *dev)
{
	struct mdev_device *mdev = to_mdev_device(dev);

	... Check the mdev device_id proposed in ...
	... https://lkml.org/lkml/2019/9/12/151 ...

	return vfio_add_group_dev(dev, &vfio_vhost_mdev_dev_ops, mdev);
}

static void vhost_mdev_remove(struct device *dev)
{
	vfio_del_group_dev(dev);
}

static struct mdev_driver vhost_mdev_driver = {
	.name	= "vhost_mdev",
	.probe	= vhost_mdev_probe,
	.remove	= vhost_mdev_remove,
};

So we can bind above mdev driver to the virtio-mdev compatible
mdev devices when we want to use vhost-mdev.

After binding above driver to the mdev device, we can setup IOMMU
via VFIO and get VFIO device fd of this mdev device, and pass it
to vhost fd (/dev/vhost-mdev) with a SET_BACKEND ioctl.

Thanks,
Tiwei

> 
> Thanks
> 
> 
> > > Yes, it is.
> > > 
> > > Thanks
> > > 
> > >