question about vdpa_mgmt_dev

question about vdpa_mgmt_dev

[Liu,Feng(INF)]

Hi Experts
	I have read the upstream code about vdpa, and I notice that there is a new struct vdpa_mgmt_dev have been added.
	Every driver implements dev_add() interface, such as mlx5 use mlx5_vdpa_dev_add(). This interface completes the creation of the vdpa device, hardware initialization and system registration. And I found that only netlink will call the dev_add() interface, in vdpa_nl_cmd_dev_add_set_doit(). While the probe() function just register vdpa_mgmt_dev without call dev_add()
	So, my questions are:
	Does it create a vdpa device only through the netlink interface? The system scans the PCI device and does not automatically create the corresponding vdpa device?
	

Thanks
Feng

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: question about vdpa_mgmt_dev

[Jason Wang]

On Mon, Nov 15, 2021 at 2:08 PM Liu,Feng(INF) <liufeng32@baidu.com> wrote:
>
> Hi Experts
>         I have read the upstream code about vdpa, and I notice that there is a new struct vdpa_mgmt_dev have been added.
>         Every driver implements dev_add() interface, such as mlx5 use mlx5_vdpa_dev_add(). This interface completes the creation of the vdpa device, hardware initialization and system registration. And I found that only netlink will call the dev_add() interface, in vdpa_nl_cmd_dev_add_set_doit(). While the probe() function just register vdpa_mgmt_dev without call dev_add()
>         So, my questions are:
>         Does it create a vdpa device only through the netlink interface?

Yes, this is the plan, and we're working on converting all the drivers
to support that.

> The system scans the PCI device and does not automatically create the corresponding vdpa device?

It's a choice for the vDPA parent driver but we suggest creating vDPA
via netlink(), this gives a persistent method for the management
layer.

Thanks

>
>
> Thanks
> Feng
>

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

答复: question about vdpa_mgmt_dev

[Liu,Feng(INF)]

Jason
	Thanks for your comments

Thanks
Feng

> -----邮件原件-----
> 发件人: Jason Wang <jasowang@redhat.com>
> 发送时间: 2021年11月15日 14:31
> 收件人: Liu,Feng(INF) <liufeng32@baidu.com>
> 抄送: virtualization@lists.linux-foundation.org
> 主题: Re: question about vdpa_mgmt_dev
> 
> On Mon, Nov 15, 2021 at 2:08 PM Liu,Feng(INF) <liufeng32@baidu.com>
> wrote:
> >
> > Hi Experts
> >         I have read the upstream code about vdpa, and I notice that
> there is a new struct vdpa_mgmt_dev have been added.
> >         Every driver implements dev_add() interface, such as mlx5 use
> mlx5_vdpa_dev_add(). This interface completes the creation of the vdpa
> device, hardware initialization and system registration. And I found that only
> netlink will call the dev_add() interface, in vdpa_nl_cmd_dev_add_set_doit().
> While the probe() function just register vdpa_mgmt_dev without call
> dev_add()
> >         So, my questions are:
> >         Does it create a vdpa device only through the netlink interface?
> 
> Yes, this is the plan, and we're working on converting all the drivers to
> support that.
> 
> > The system scans the PCI device and does not automatically create the
> corresponding vdpa device?
> 
> It's a choice for the vDPA parent driver but we suggest creating vDPA via
> netlink(), this gives a persistent method for the management layer.
> 
> Thanks
> 
> >
> >
> > Thanks
> > Feng
> >

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

答复: question about vdpa_mgmt_dev

[Liu,Feng(INF)]

Jason
	It's clear, thanks for your comments

Thanks
Feng

> -----邮件原件-----
> 发件人: Jason Wang <jasowang@redhat.com>
> 发送时间: 2021年11月15日 14:31
> 收件人: Liu,Feng(INF) <liufeng32@baidu.com>
> 抄送: virtualization@lists.linux-foundation.org
> 主题: Re: question about vdpa_mgmt_dev
> 
> On Mon, Nov 15, 2021 at 2:08 PM Liu,Feng(INF) <liufeng32@baidu.com>
> wrote:
> >
> > Hi Experts
> >         I have read the upstream code about vdpa, and I notice that
> there is a new struct vdpa_mgmt_dev have been added.
> >         Every driver implements dev_add() interface, such as mlx5 use
> mlx5_vdpa_dev_add(). This interface completes the creation of the vdpa
> device, hardware initialization and system registration. And I found that only
> netlink will call the dev_add() interface, in vdpa_nl_cmd_dev_add_set_doit().
> While the probe() function just register vdpa_mgmt_dev without call
> dev_add()
> >         So, my questions are:
> >         Does it create a vdpa device only through the netlink interface?
> 
> Yes, this is the plan, and we're working on converting all the drivers to
> support that.
> 
> > The system scans the PCI device and does not automatically create the
> corresponding vdpa device?
> 
> It's a choice for the vDPA parent driver but we suggest creating vDPA via
> netlink(), this gives a persistent method for the management layer.
> 
> Thanks
> 
> >
> >
> > Thanks
> > Feng
> >

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization