[dpdk-dev] [PATCH 0/6] Enable VF RSS for Niantic

[Vladislav Zolotarov]

On Dec 19, 2014 3:35 AM, "Ouyang, Changchun" <changchun.ouyang at intel.com>
wrote:
>
> My response as below.
>
>
>
> From: Vlad Zolotarov [mailto:vladz at cloudius-systems.com]
> Sent: Thursday, December 18, 2014 9:06 PM
> To: dev at dpdk.org; Ouyang, Changchun
> Subject: Re: [PATCH 0/6] Enable VF RSS for Niantic
>
>
>
>
>
> On 12/18/14 12:11, Vlad Zolotarov wrote:
>>
>> From: Changchun Ouyang <changchun.ouyang at intel.com>
>>
>>
>>
>> This patch enables VF RSS for Niantic, which allow each VF having at
most 4 queues.
>>
>> The actual queue number per VF depends on the number of VF:
>>
>> VF number from 1~32: 4 queues per VF;
>>
>> VF number from 33~max vf num: 2 queues per VF;
>>
>>
>>
>> On host, to enable VF RSS functionality, mq mode should be set as
ETH_MQ_RX_VMDQ_RSS
>>
>> or ETH_MQ_RX_RSS mode, and SRIOV mode should be activated.
>>
>> It also needs config VF RSS information like hash function, RSS key, RSS
key length.
>
>
> This patch series is missing a few things:
>
> Taking into the consideration the number of Rx queues requested by a user
in the rte_eth_dev_configure().
>
> Changchun: yes, will have a v2 to consider it, as I say in my previous
response.

I must have joined the mailing list after your v2 series. I'll take a look
in the web for it.

>
> dev->dev_ops->reta_query used by a rte_eth_dev_rss_reta_query() is still
not initialized for a VF. Thus there is no way to query the RSS table
>
> Changchun: do you mean query rss table on guest?

Of course.

Niantic doesn’t have separate reta for a specific vf, the reta is shared by
pf and all vf.

AFAIK the reta is statically divided between all available functions.

So, first of all we need a way to know that is the size of a single VF
partition in order to know how many RSS queues may be configured.

Then we need an ability to query/set the contents of the corresponding reta
partition like we can for PF.

>
> Do you think of returning all reta contents for each vf make sense?  Or
any other insight here?
>
> rte_eth_dev_info_get() returns reta_size == 0 when called for a VF
function.
>
> Changchun:  same as above.
>
> Your insight here are welcome here.
>
> thanks
>
> Changchun
>
>