Re: [dpdk-dev] [PATCH] Revert "app/testpmd: set fixed flag for exact link speed"

[Andrew Rybchenko <arybchenko@solarflare.com>]

On 5/4/19 11:45 PM, Thomas Monjalon wrote:
> 02/05/2019 22:27, Thomas Monjalon:
>> 02/05/2019 20:31, Ferruh Yigit:
>>> On 4/29/2019 10:52 AM, Thomas Monjalon wrote:
>>>> 25/04/2019 17:27, Ferruh Yigit:
>>>>> On 4/25/2019 2:29 PM, Thomas Monjalon wrote:
>>>>>> 25/04/2019 13:47, Ferruh Yigit:
>>>>>>> On 4/25/2019 9:19 AM, Thomas Monjalon wrote:
>>>>>>>> 25/04/2019 00:03, Ferruh Yigit:
>>>>>>>>> This reverts commit bdca79053b6aea504d02691d9319fa976062457f.
>>>>>>>>>
>>>>>>>>> Not all PMDs support the fixed link speed set, and link speed can be set
>>>>>>>>> even with auto negotiation enabled. Reverting the patch to not break
>>>>>>>>> existing usage.
>>>>>>>> Which PMDs do not support this flag?
>>>>>>>> Why not fixing the PMDs?
>>>>>>>>
>>>>>>> At least ixgbe and i40e is not supporting setting a fixed speed.
>>>>>>> But I am not sure if this is something to fix, the command in testpmd is to set
>>>>>>> the link speed, what is the problem with setting the link speed without
>>>>>>> disabling the auto-negotiation?
>>>>>> It means it will negotiate with only one speed proposed.
>>>>> Yes.
>>>>>
>>>>>> The real issue is to not support the fixed flag.
>>>>> I don't know if this is a real issue but
>>>>> even it is, is it an issue in the scope of this testpmd command?
>>>>>
>>>>> right now we are first updating the command to set fixed speed flag, and
>>>>> requesting PMDs to fix for it, I am suggesting not to update the command at all.
>>>> I understand. But this change shows a broken behaviour.
>>>> This is the intent of testpmd to show what works or not in PMDs.
>>>> How hard is it to fix the PMDs in your opinion?
>>>>
>>> As far as I can see the the fixed link speed set is not supported in the PMD.
>>>
>>> It may be easy to add perhaps, I don't know, but is it really a "broken
>>> behavior" to not have this support?
>>> What defines that setting speed has to be "fixed speed", if this demand is not
>>> there, should testpmd enforce it?
>> I think a PMD should support both: fixed or not.
>>
>>> In mail thread we have talked that this testpmd command can get an extra
>>> argument to define the speed fixed or not, this can be used to test fixed speed
>>> by who wants to test/use fixed speed.
>>>
>>> I am for reverting this for the release, and adding a new version next release
>>> with fixed speed argument, otherwise testpmd won't be used to set the speed for
>>> some PMDs.
>> OK
> We could have an option in testpmd to test ETH_LINK_SPEED_FIXED.
>
> Revert applied.

I agree that revert is the best option for the release, but not long term.

Typical options are (in order):
  1. Auto-negotiation of whatever port supports
  2. Fixed exactly one speed with autoneg disabled
  3. Auto-negotiation with limitations (and it is really the third 
option which makes sense if and only if interface allows to specify more 
than one speed to be negotiated)

Right now testpmd supports (1) and (3) with limitation to only one speed 
to be negotiated. I think it is wrong.

Andrew.