Re: [PATCH v2 0/6] support setting lanes

[Ferruh Yigit <ferruh.yigit@amd.com>]

On 5/16/2024 1:48 PM, huangdengdui wrote:
> Hi, Ferruh
> Sorry for replying your email very late.
> The answers to your questions are as follows. Please correct me if I am wrong.
> 
> On 2024/4/4 21:58, Ferruh Yigit wrote:
>>
>> Hi Dengdui, Damodharam,
>>
>> As details of the implementation under discussion, I have a high level
>> question.
>>
>> Why/when an application need to configure the lanes explicitly?
>>
>> In above description, it mentions if one port is configured as 4x25G and
>> other 2x50G, port can't be up. How do you end up being in this situation?
>>
> 
> According to the Ethernet standard document[1], speed and lanes need to be configured to the PHY together.
> Currently, the application just set one speed like 100G to the driver.
> And this doesn't matter with lane number.
> Currently, the lane number of one NIC for a specified speed depand on their default behavior.
> As a result, this situation will be happened.
> 

I guess we come to agreement that with three new APIs for lane (get,
set, get_capabilities), link speed and lane APIs can be separated, this
cause more clear APIs.

>> Lets assume first port is configured as 100G, and FW configured it as
>> 4x25G, and again user configured second port as 100G, why FW can't
>> detect this and configure ports with correct lane configuration?
> 
> When the auto-negotiation is disabled, FW of the local port never know the configuration of
> the peer port.
> 
> After all, not all NICs support auto-negotiation feature.
> 

Ack

>>
>> In this case, if we push the responsibility to the user, when user is
>> configuring the second port how she will know what is the lane
>> configuration for first port, and what is the proper lane configuration
>> for the second port?
> 
> So we need to support the lane query function for above reason.
> 

Ack

>>
>> Instead of pushing this configuration to user, why it can't be handled
>> internally?
>>
>> As long as user requested speed configured by device, the lane
>> configuration has no impact to the user, right?> Is there a case/reason user need to explicitly set, lets say PAM4
>> against NRZ?
>>
> Sorry, I can not understand what you mean.
> 

Indeed you already answered this one, explicit lane configuration is
required two align configuration of both peers when auto-negotiation is off.


> [1]
> https://lore.kernel.org/netdev/20201010154119.3537085-1-idosch@idosch.org/T/