[PATCH] ethdev: introduce ethdev dump API

[Ferruh Yigit]

On 2/8/2022 11:14 AM, Min Hu (Connor) wrote:
> Hi, Ferruh,
> 
> 在 2022/2/8 18:21, Ferruh Yigit 写道:
>> On 2/8/2022 12:39 AM, Min Hu (Connor) wrote:
>>> Hi, Ferruh,
>>>
>>> 在 2022/2/7 23:35, Ferruh Yigit 写道:
>>>> On 2/7/2022 12:56 PM, Morten Brørup wrote:
>>>>>> From: Ferruh Yigit [mailto:ferruh.yigit at intel.com]
>>>>>> Sent: Monday, 7 February 2022 13.36
>>>>>>
>>>>>> On 2/7/2022 12:18 PM, Morten Brørup wrote:
>>>>>>>> From: Ferruh Yigit [mailto:ferruh.yigit at intel.com]
>>>>>>>> Sent: Monday, 7 February 2022 12.46
>>>>>>>>
>>>>>>>> On 2/7/2022 1:47 AM, Min Hu (Connor) wrote:
>>>>>>>>> Added the ethdev dump API which provides functions for query
>>>>>> private
>>>>>>>> info
>>>>>>>>
>>>>>>>> Isn't API and function are same thing in this contexts?
>>>>>>>>
>>>>>>>>> from device. There exists many private properties in different PMD
>>>>>>>> drivers,
>>>>>>>>> such as adapter state, Rx/Tx func algorithm in hns3 PMD. The
>>>>>>>> information of
>>>>>>>>> these properties is important for debug. As the information is
>>>>>>>> private,
>>>>>>>>> the new API is introduced.>
>>>>>>>>
>>>>>>>> In the patch title 'ethdev' is duplicated, can you fix it?
>>>>>>>>
>>>>>>>>
>>>>>>>>> Signed-off-by: Min Hu (Connor) <humin29 at huawei.com>
>>>>>>>>> Acked-by: Morten Brørup <mb at smartsharesystems.com>
>>>>>>>>> Acked-by: Ray Kinsella <mdr at ashroe.eu>
>>>>>>>>> Acked-by: Ajit Khaparde <ajit.khaparde at broadcom.com>
>>>>>>>
>>>>>>> [...]
>>>>>>>
>>>>>>>>> @@ -990,6 +990,20 @@ typedef int
>>>>>> (*eth_representor_info_get_t)(struct
>>>>>>>> rte_eth_dev *dev,
>>>>>>>>>     typedef int (*eth_rx_metadata_negotiate_t)(struct rte_eth_dev
>>>>>> *dev,
>>>>>>>>>                            uint64_t *features);
>>>>>>>>>
>>>>>>>>> +/**
>>>>>>>>> + * @internal
>>>>>>>>> + * Dump ethdev private info to a file.
>>>>>>>>> + *
>>>>>>>>
>>>>>>>> It doesn't dump the 'ethdev' private info, it dumps the private info
>>>>>>>> from device.
>>>>>>>
>>>>>>> It seems perfectly clear to me. How would you prefer it phrased
>>>>>> instead?
>>>>>>>
>>>>>>
>>>>>> What described in the document is more accurate,
>>>>>> "query private info from device".
>>>>>>
>>>>>> What we are dumping here is not ethdev private info, it is device
>>>>>> private info,
>>>
>>> what is the difference between ethdev and device?
>>
>> It is not very clear, but for me 'ethdev' is refers to device abstract
>> layer (ethdev library) specific private data 
> Could you give an example for 'ethdev'specific private data ?
> 

I think 'struct rte_eth_dev' content can be a sample.

But I hear you, diff is not clear, it is subtle as Morten said,
when doc and commit log refers it as "private info from device",
I think we can use the same in the API documentation as well.

> and device refers to ethdev
>> device (PMD) private data. ethdev is common for all drivers.
> OK, we could treat it as convention in future.
>>
>>>>>> and we really don't know what that data may be in the ethdev layer.
>>>>>>
>>>>>> Also there is a chance that 'ethdev private info' can be confused with
>>>>>> 'ethdev->data->dev_private'
>>> what I want to dump is exactly the 'ethdev->data->dev_private'.
>>> 'ethdev private info' means 'ethdev->data->dev_private'.
>>> why confused?
>>
>> What I understand was this API can return any device private information,
>> it is not limited to 'ethdev->data->dev_private', (although most of the 
> I think this API is limited to 'ethdev->data->dev_private'.
>> data
>> is represented in this struct), like if you want to dump queue state,
>> this is out of 'ethdev->data->dev_private'.
> Queue state can be dumped using API 'rte_eth_tx_queue_info_get'.
> 

Yes it can be. But as far as I can see there is nothing prevents the dump()
API to provide the same, it is up to PMD.

If the intention is to limit what can be dump to 'ethdev->data->dev_private',
it is not clear from API documentation/implementation.

>>
>>>>>
>>>>> OK. Now I got your point! The difference is very subtle.
>>>>>
>>>>>>
>>>>>>> [...]
>>>>>>>
>>>>>>>>
>>>>>>>>> + */
>>>>>>>>> +__rte_experimental
>>>>>>>>> +int rte_eth_dev_priv_dump(FILE *file, uint16_t port_id);
>>>>>>>>> +
>>>>>>>>
>>>>>>>> What do you think to have the 'port_id' as first argument to be
>>>>>>>> consistent
>>>>>>>> with the other APIs?
>>>>>>>
>>>>>>> The _dump APIs in other libraries have the file pointer as the first
>>>>>> parameter, so let's follow that convention here too. No need to move
>>>>>> the port_id parameter here.
>>>>>>>
>>>>>>
>>>>>> Yes, for most of the _dump() APIs, file pointer seems is the first
>>>>>> argument,
>>>>>> bu they are from various libraries.
>>>>>>
>>>>>> Within the ethdev APIs, I think it makes sense that all APIs start with
>>>>>> 'port_id' parameter for consistency, like done in:
>>>>>> rte_flow_dev_dump(uint16_t port_id, ...)
>>>>>>
>>>>>>> Only rte_dma_dump() has the file pointer last, and I didn't catch it
>>>>>> when the function was defined.
>>>>>>>
>>>>>
>>>>> OK. Then I agree with you about following the convention like rte_flow_dev_dump() with the port_id first.
>>>>>
>>>>> I even think Connor got it right the first time, and I proposed following the other convention.
>>>>>
>>>>
>>>> Ahh, may bad I missed that, sorry for not commenting on time.
>>>>
>>>>
>>>>> It's not easy when there are two opposite conventions. :-)
>>>>>
>>>>
>>>> Yep, that is the main issue.
>>>>
>>>> .
>>
>> .