2023-09-13 09:55:26

by Wen Gu

[permalink] [raw]
Subject: Re: [RFC PATCH net-next] net/smc: Introduce SMC-related proc files



On 2023/9/11 19:54, Wenjia Zhang wrote:
>
>
>
> Hi Wen,
>
> I can understand your problem and frustration. However, there are two reasons I'm not really convinced by the proc file
> method:
> 1) AFAI, the proc method could consume many CPU time especially in case with a log of sockets to read the pseudo files.
> 2) We have already implemented the complex netlink method on the same purpose. I see the double expense to main the code.
>
> Then the question is if the lack of dependency issue can be handle somehow, or the proc method is the only way to
> achieve this purpose?
>
> Any opinion is welcome!
>
> Thanks,
> Wenjia

Hi, Wenjia. I agree with your concerns.

My initial intention is to make these proc files serve as a supplement to netlink to conveniently
check smc connections in an environment where smc-tools cannot be easily obtained.

Yes, proc files won't be the first choice for diagnosis, but can be a convenient backup.

Thanks,
Wen Gu


2023-09-14 19:39:09

by Alexandra Winter

[permalink] [raw]
Subject: Re: [RFC PATCH net-next] net/smc: Introduce SMC-related proc files



On 13.09.23 11:53, Wen Gu wrote:
>
>
> On 2023/9/11 19:54, Wenjia Zhang wrote:
>>
>>
>>
>> Hi Wen,
>>
>> I can understand your problem and frustration. However, there are two reasons I'm not really convinced by the proc file method:
>> 1) AFAI, the proc method could consume many CPU time especially in case with a log of sockets to read the pseudo files.
>> 2) We have already implemented the complex netlink method on the same purpose. I see the double expense to main the code.
>>
>> Then the question is if the lack of dependency issue can be handle somehow, or the proc method is the only way to achieve this purpose?
>>
>> Any opinion is welcome!
>>
>> Thanks,
>> Wenjia
>
> Hi, Wenjia. I agree with your concerns.
>
> My initial intention is to make these proc files serve as a supplement to netlink to conveniently
> check smc connections in an environment where smc-tools cannot be easily obtained.
>
> Yes, proc files won't be the first choice for diagnosis, but can be a convenient backup.
>
> Thanks,
> Wen Gu
>
>
As /proc is an interface to userface, we would have to maintain the 2 redundant methods basically forever.
I personally don't think we should implement another interface without a very strong reason.

2023-09-15 01:59:34

by Wen Gu

[permalink] [raw]
Subject: Re: [RFC PATCH net-next] net/smc: Introduce SMC-related proc files



On 2023/9/14 18:29, Alexandra Winter wrote:
>
>
> On 13.09.23 11:53, Wen Gu wrote:
>>
>>
>> On 2023/9/11 19:54, Wenjia Zhang wrote:
>>>
>>>
>>>
>>> Hi Wen,
>>>
>>> I can understand your problem and frustration. However, there are two reasons I'm not really convinced by the proc file method:
>>> 1) AFAI, the proc method could consume many CPU time especially in case with a log of sockets to read the pseudo files.
>>> 2) We have already implemented the complex netlink method on the same purpose. I see the double expense to main the code.
>>>
>>> Then the question is if the lack of dependency issue can be handle somehow, or the proc method is the only way to achieve this purpose?
>>>
>>> Any opinion is welcome!
>>>
>>> Thanks,
>>> Wenjia
>>
>> Hi, Wenjia. I agree with your concerns.
>>
>> My initial intention is to make these proc files serve as a supplement to netlink to conveniently
>> check smc connections in an environment where smc-tools cannot be easily obtained.
>>
>> Yes, proc files won't be the first choice for diagnosis, but can be a convenient backup.
>>
>> Thanks,
>> Wen Gu
>>
>>
> As /proc is an interface to userface, we would have to maintain the 2 redundant methods basically forever.
> I personally don't think we should implement another interface without a very strong reason.


I understand, thank you Alexandra and Wenjia.

Regards,
Wen Gu