Re: [RFC PATCH net-next 1/1] net: Support for switch port configuration

[John Fastabend <john.r.fastabend@intel.com>]

On 12/11/2014 09:41 AM, Roopa Prabhu wrote:
> On 12/11/14, 8:56 AM, Jiri Pirko wrote:
>> Thu, Dec 11, 2014 at 05:37:46PM CET, roopa@cumulusnetworks.com wrote:
>>> On 12/11/14, 3:01 AM, Jiri Pirko wrote:
>>>> Thu, Dec 11, 2014 at 10:59:42AM CET, marco.varlese@intel.com wrote:
>>>>>> -----Original Message-----
>>>>>> From: John Fastabend [mailto:john.fastabend@gmail.com]
>>>>>> Sent: Wednesday, December 10, 2014 5:04 PM
>>>>>> To: Jiri Pirko
>>>>>> Cc: Varlese, Marco; netdev@vger.kernel.org;
>>>>>> stephen@networkplumber.org; Fastabend, John R;
>>>>>> roopa@cumulusnetworks.com; sfeldma@gmail.com; linux-
>>>>>> kernel@vger.kernel.org
>>>>>> Subject: Re: [RFC PATCH net-next 1/1] net: Support for switch port
>>>>>> configuration
>>>>>>
>>>>>> On 12/10/2014 08:50 AM, Jiri Pirko wrote:
>>>>>>> Wed, Dec 10, 2014 at 05:23:40PM CET, marco.varlese@intel.com wrote:
>>>>>>>> From: Marco Varlese <marco.varlese@intel.com>
>>>>>>>>
>>>>>>>> Switch hardware offers a list of attributes that are configurable on
>>>>>>>> a per port basis.
>>>>>>>> This patch provides a mechanism to configure switch ports by adding
>>>>>>>> an NDO for setting specific values to specific attributes.
>>>>>>>> There will be a separate patch that extends iproute2 to call the new
>>>>>>>> NDO.
>>>>>>> What are these attributes? Can you give some examples. I'm asking
>>>>>>> because there is a plan to pass generic attributes to switch ports
>>>>>>> replacing current specific ndo_switch_port_stp_update. In this case,
>>>>>>> bridge is setting that attribute.
>>>>>>>
>>>>>>> Is there need to set something directly from userspace or does it make
>>>>>>> rather sense to use involved bridge/ovs/bond ? I think that both will
>>>>>>> be needed.
>>>>>> +1
>>>>>>
>>>>>> I think for many attributes it would be best to have both. The in kernel callers
>>>>>> and netlink userspace can use the same driver ndo_ops.
>>>>>>
>>>>>> But then we don't _require_ any specific bridge/ovs/etc module. And we
>>>>>> may have some attributes that are not specific to any existing software
>>>>>> module. I'm guessing Marco has some examples of these.
>>>>>>
>>>>>> [...]
>>>>>>
>>>>>>
>>>>>> -- 
>>>>>> John Fastabend         Intel Corporation
>>>>> We do have a need to configure the attributes directly from user-space and I have identified the tool to do that in iproute2.
>>>>>
>>>>> An example of attributes are:
>>>>> * enabling/disabling of learning of source addresses on a given port (you can imagine the attribute called LEARNING for example);
>>>>> * internal loopback control (i.e. LOOPBACK) which will control how the flow of traffic behaves from the switch fabric towards an egress port;
>>>>> * flooding for broadcast/multicast/unicast type of packets (i.e. BFLOODING, MFLOODING, UFLOODING);
>>>>>
>>>>> Some attributes would be of the type enabled/disabled while other will allow specific values to allow the user to configure different behaviours of that feature on that particular port on that platform.
>>>>>
>>>>> One thing to mention - as John stated as well - there might be some attributes that are not specific to any software module but rather have to do with the actual hardware/platform to configure.
>>>>>
>>>>> I hope this clarifies some points.
>>>> It does. Makes sense. We need to expose this attr set/get for both
>>>> in-kernel and userspace use cases.
>>>>
>>>> Please adjust you patch for this. Also, as a second patch, it would be
>>>> great if you can convert ndo_switch_port_stp_update to this new ndo.
>>> Why are we exposing generic switch attribute get/set from userspace ?. We
>>> already have specific attributes for learning/flooding which can be extended
>>> further.
>> Yes, but that is for PF_BRIDGE and bridge specific attributes. There
>> might be another generic attrs, no?
> I cant think of any. And plus, the whole point of switchdev l2 offloads was to map these to existing
> bridge attributes. And we already have a match for some of the attributes that marco wants.
> 
> If there is a need for such attributes, i don't see why it is needed for switch devices only.
> It is needed for any hw (nics etc). And, a precedence to this is to do it via ethtool.

I would prefer to _not_ add more attributes to ethtool. 'ethtool' is in general
harder to work with then netlink for all but the most static attributes.

> 
> Having said that, am sure we will find a need for this in the future. And having a netlink attribute always helps.
> 
> Today, it seems like these can be mapped to existing attributes that are settable via ndo_bridge_setlink/getlink.

Absolutely I view this as an RFC patch noting we may/will need some extensions
in the future. .We can evaluate the attributes on a case by case basis as they
come in. And if they all fit in setlink/getlink that is great. 

> 
>>
>>> And for in kernel api....i had a sample patch in my RFC series (Which i was
>>> going to resubmit, until it was decided that we will use existing api around
>>> ndo_bridge_setlink/ndo_bridge_getlink):
>>> http://www.spinics.net/lists/netdev/msg305473.html
>> Yes, this might become handy for other generic non-bridge attrs.
>>
>>> Thanks,
>>> Roopa
>>>
>>>
>>>
>