Re: [RFC PATCH v3 net-next] Documentation: devlink: Add devlink-sd

[William Tu <witu@nvidia.com>]

On 2/1/24 7:30 PM, Jakub Kicinski wrote:
> External email: Use caution opening links or attachments
>
>
> On Thu, 1 Feb 2024 11:16:22 -0800 William Tu wrote:
>>>> I guess bnxt, ice, nfp are doing tx buffer sharing?
>>> I'm not familiar with ice. I'm 90% sure bnxt shares both Rx and Tx.
>>> I'm 99.9% sure nfp does.
>>>
>>> It'd be great if you could do the due diligence rather than guessing
>>> given that you're proposing uAPI extension :(
>>>
>> *
>>
>> (sorry again, html is detected in previous email)
>>
>> due diligence here:
> Thanks for collecting the deets!
>
> If I'm reading this right we have 3 drivers which straight up share
> queues. I don't remember anyone complaining about the queue sharing
> in my time at Netro. Meaning we can probably find reasonable defaults
> and not start with the full API? Just have the
>
>    devlink dev eswitch set DEV mode switchdev shared-descs enable
>
> extension to begin with?
yes, that's what I began with months ago, by just adding new devlink params.
We need at least two new parameters: enable/disable, and size of the pool.
I have right now is s.t like:
$ devlink dev eswitch set pci/0000:08:00.0 mode switchdev 
shared-descs-counts  1024
$ devlink dev eswitch set pci/0000:08:00.0 mode switchdev shared-descs 
enable
Once enabled, the default sharing scheme, described in the RFC, applies 
to all REP queues, and no way to customize the queue-to-pool mapping.

Then discuss with a couple people internally, and come up with this more 
flexible and generic API.

William