RE: [PATCH] ethdev: introduce ethdev dump API

["Morten Brørup" <mb@smartsharesystems.com>]

> From: Ferruh Yigit [mailto:ferruh.yigit@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@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@huawei.com>
> >>> Acked-by: Morten Brørup <mb@smartsharesystems.com>
> >>> Acked-by: Ray Kinsella <mdr@ashroe.eu>
> >>> Acked-by: Ajit Khaparde <ajit.khaparde@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,
> 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'

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.

It's not easy when there are two opposite conventions. :-)

-Morten