Virtualization - Feb 2021 - [PATCH iproute2-next v3 0/5] Add vdpa device management tool

On 2021/2/4 ??7:15, Adrian Moreno wrote:
> Sorry I have not followed this work as close as I would have wanted.
> Some questions below.
>
> On 2/4/21 4:16 AM, Jason Wang wrote:
>> On 2021/2/2 ??6:35, Parav Pandit wrote:
>>> Linux vdpa interface allows vdpa device management functionality.
>>> This includes adding, removing, querying vdpa devices.
>>>
>>> vdpa interface also includes showing supported management devices
>>> which support such operations.
>>>
>>> This patchset includes kernel uapi headers and a vdpa tool.
>>>
>>> examples:
>>>
>>> $ vdpa mgmtdev show
>>> vdpasim:
>>>  ?? supported_classes net
>>>
>>> $ vdpa mgmtdev show -jp
>>> {
>>>  ???? "show": {
>>>  ???????? "vdpasim": {
>>>  ???????????? "supported_classes": [ "net" ]
>>>  ???????? }
>>>  ???? }
>>> }
>>>
> How can a user establish the relationship between a mgmtdev and it's
parent
> device (pci vf, sf, etc)?

Parav should know more but I try to answer.

I think there should be BDF information in the mgmtdev show command if 
the parent is a PCI device, or we can simply show the parent here?

>
>>> Create a vdpa device of type networking named as "foo2"
from
>>> the management device vdpasim_net:
>>>
>>> $ vdpa dev add mgmtdev vdpasim_net name foo2
>>>
> I guess this command will accept a 'type' parameter once more
supported_classes
> are added?

This could be extended in the future.

>
> Also, will this tool also handle the vdpa driver binding or will the user
handle
> that through the vdpa bus' sysfs interface?

I think not, it's the configuration below the vdpa bus. The sysfs should 
be the only interface for managing driver binding.

Thanks

>
>>> Show the newly created vdpa device by its name:
>>> $ vdpa dev show foo2
>>> foo2: type network mgmtdev vdpasim_net vendor_id 0 max_vqs 2
max_vq_size 256
>>>
>>> $ vdpa dev show foo2 -jp
>>> {
>>>  ???? "dev": {
>>>  ???????? "foo2": {
>>>  ???????????? "type": "network",
>>>  ???????????? "mgmtdev": "vdpasim_net",
>>>  ???????????? "vendor_id": 0,
>>>  ???????????? "max_vqs": 2,
>>>  ???????????? "max_vq_size": 256
>>>  ???????? }
>>>  ???? }
>>> }
>>>
>>> Delete the vdpa device after its use:
>>> $ vdpa dev del foo2
>>>
>>> Patch summary:
>>> Patch-1 adds kernel headers for vdpa subsystem
>>> Patch-2 adds library routines for indent handling
>>> Patch-3 adds library routines for generic socket communication
>>> PAtch-4 adds library routine for number to string mapping
>>> Patch-5 adds vdpa tool
>>>
>>> Kernel headers are from the vhost kernel tree [1] from branch
linux-next.
>>>
>>> [1] https://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git
>>>
>>> ---
>>
>> Adding Adrian to see if this looks good for k8s integration.
>>
>> Thanks
>>
> Thanks

> From: Jason Wang <jasowang at redhat.com>
> Sent: Friday, February 5, 2021 9:11 AM
> 
> 
> On 2021/2/4 ??7:15, Adrian Moreno wrote:
> > Sorry I have not followed this work as close as I would have wanted.
> > Some questions below.
> >
> > On 2/4/21 4:16 AM, Jason Wang wrote:
> >> On 2021/2/2 ??6:35, Parav Pandit wrote:
> >>> Linux vdpa interface allows vdpa device management
functionality.
> >>> This includes adding, removing, querying vdpa devices.
> >>>
> >>> vdpa interface also includes showing supported management
devices
> >>> which support such operations.
> >>>
> >>> This patchset includes kernel uapi headers and a vdpa tool.
> >>>
> >>> examples:
> >>>
> >>> $ vdpa mgmtdev show
> >>> vdpasim:
> >>>  ?? supported_classes net
> >>>
> >>> $ vdpa mgmtdev show -jp
> >>> {
> >>>  ???? "show": {
> >>>  ???????? "vdpasim": {
> >>>  ???????????? "supported_classes": [ "net"
]
> >>>  ???????? }
> >>>  ???? }
> >>> }
> >>>
> > How can a user establish the relationship between a mgmtdev and
it's
> > parent device (pci vf, sf, etc)?
> 
> 
> Parav should know more but I try to answer.
> 
> I think there should be BDF information in the mgmtdev show command if
> the parent is a PCI device, or we can simply show the parent here?
>
Yes, it is present in the mgmtdev show command.
I should have added the example from the kernel cover letter here.
Link to the kernel cover letter is at 

An example for real PCI PF,VF,SF device looks like below.
I will cover below example to the v4 cover letter while addressing David's
comment for header file relocation.

$ vdpa mgmtdev list
pci/0000:03.00:0
  supported_classes
    net
pci/0000:03.00:4
  supported_classes
    net
auxiliary/mlx5_core.sf.8
  supported_classes
    net

[1] https://lore.kernel.org/netdev/20210105103203.82508-1-parav at nvidia.com/

On 2/5/21 4:40 AM, Jason Wang wrote:
> 
> On 2021/2/4 ??7:15, Adrian Moreno wrote:
>> Sorry I have not followed this work as close as I would have wanted.
>> Some questions below.
>>
>> On 2/4/21 4:16 AM, Jason Wang wrote:
>>> On 2021/2/2 ??6:35, Parav Pandit wrote:
>>>> Linux vdpa interface allows vdpa device management
functionality.
>>>> This includes adding, removing, querying vdpa devices.
>>>>
>>>> vdpa interface also includes showing supported management
devices
>>>> which support such operations.
>>>>
>>>> This patchset includes kernel uapi headers and a vdpa tool.
>>>>
>>>> examples:
>>>>
>>>> $ vdpa mgmtdev show
>>>> vdpasim:
>>>> ??? supported_classes net
>>>>
>>>> $ vdpa mgmtdev show -jp
>>>> {
>>>> ????? "show": {
>>>> ????????? "vdpasim": {
>>>> ????????????? "supported_classes": [ "net"
]
>>>> ????????? }
>>>> ????? }
>>>> }
>>>>
>> How can a user establish the relationship between a mgmtdev and
it's parent
>> device (pci vf, sf, etc)?
> 
> 
> Parav should know more but I try to answer.
> 
> I think there should be BDF information in the mgmtdev show command if the
> parent is a PCI device, or we can simply show the parent here?
> 
> 
>>
>>>> Create a vdpa device of type networking named as
"foo2" from
>>>> the management device vdpasim_net:
>>>>
>>>> $ vdpa dev add mgmtdev vdpasim_net name foo2
>>>>
>> I guess this command will accept a 'type' parameter once more
supported_classes
>> are added?
> 
> 
> This could be extended in the future.
> 
> 
>>
>> Also, will this tool also handle the vdpa driver binding or will the
user handle
>> that through the vdpa bus' sysfs interface?
> 
> 
> I think not, it's the configuration below the vdpa bus. The sysfs
should be the
> only interface for managing driver binding.
> 
Understood, thanks.

-- 
Adri?n Moreno