[dpdk-dev] [PATCH 0/5] net/nfp logging fixes

Ferruh Yigit ferruh.yigit at intel.com
Thu Apr 26 23:52:48 CEST 2018


On 4/26/2018 7:14 PM, Alejandro Lucero wrote:
> On Thu, Apr 26, 2018 at 4:42 PM, Stephen Hemminger <
> stephen at networkplumber.org> wrote:
> 
>> On Thu, 26 Apr 2018 13:52:53 +0100
>> Alejandro Lucero <alejandro.lucero at netronome.com> wrote:
>>
>>> Hi Stephen,
>>>
>>> Thanks for this patch set.
>>>
>>> I'm happy with it although I have some concerns regarding how the dynamic
>>> logs work, or maybe I have a wrong understanding about it. I have tried
>> to
>>> read some doc about how it works, and I found the original patch from
>>> Olivier the best source, so maybe things have changed a bit and my
>> concerns
>>> are unfounded.
>>>
>>> I think it is OK to specifically add something like
>>>
>>> --log-level='pmd\.i40e.*,8'
>>>
>>> if you want to debug a PMD, but if you are an user and you just want to
>>> know why the app is not finding any port, finding out the right string is
>>> not trivial. For example, with an PF, the NFP PMD goes through a process
>>> where the NFP device (no the NIC) is accessed first through a complex
>>> interface, then firmware is uploaded, DPDK ports created (for multiport
>>> devices), etc. I think any error in that process should be output if the
>>> right loglevel is there and not just if the right log type was
>> specifically
>>> enabled. Is this what would happen with your patchset?
>>
>> Most drivers set default log level to NOTICE. Then if they see something
>> obviously wrong it will show up if the right log level is used.
> 
> 
>> For the case of finding out why no drivers are found then doing
>> something like
>>         --log-level='pmd.*:info'
>> would be useful.
>>
>> Latest version makes regex optional and allows symbolic levels.
>>
>>
>>> I have suffered silent configuration problems, like the NFP card being in
>>> the wrong NUMA socket, and although I can solve that quickly because I
>> have
>>> the knowledge, other people using NFP with DPDK require someone to help
>>> because they do not know what is going on. And this is usually bad
>> because
>>> they have another NIC card in the same host (in the right NUMA socket)
>> and
>>> the app just works smoothly then, leaving our NIC with a bad press. So I
>>> think, some errors should always appear with the right loglevel
>> configured.
>>
>> Driver should definitely use level > INFO for things that are wrong.
>>
>>
> Uhmm, yes. I think I need to submit some changes to the level of most of
> the PMD messages.
> 
> Thanks for the heads up.
> 
> I have reviewed and tested the patches and they all seem all right.
> 
> Acked-by: Alejandro Lucero <alejandro.lucero at netronome.com>
> Tested-by: Alejandro Lucero <alejandro.lucero at netronome.com>

Series applied to dpdk-next-net/master, thanks.


More information about the dev mailing list