[dpdk-dev] [PATCH v4 21/23] ethdev: Move filling of rte_eth_dev_info->pci_dev to dev_infos_get()

[Shreyansh Jain]

On Friday 23 December 2016 04:20 PM, Jan Blunck wrote:
> On Thu, Dec 22, 2016 at 9:11 AM, Shreyansh Jain <shreyansh.jain at nxp.com> wrote:
>> On Thursday 22 December 2016 01:39 AM, Stephen Hemminger wrote:
>>>
>>> On Wed, 21 Dec 2016 16:09:44 +0100
>>> Jan Blunck <jblunck at infradead.org> wrote:
>>>
>>>> Only the device itself can decide its PCI or not.
>>>>
>>>> Signed-off-by: Jan Blunck <jblunck at infradead.org>
>>>> Acked-by: Shreyansh Jain <shreyansh.jain at nxp.com>
>>>> ---
>>>
>>>
>>> I would still like to kill dev_pci from the dev_info API.
>>>
>
> I'm fine with that too.
>
>>
>> +1. It should be rte_dev reference instead.
>>
>
> Only if you can give use-cases for what users should be able to do
> with it. If that is the case we need to clearly define what that
> means. Do we want to enable users to control the low-level EAL device
> directly and shortcut the ethdev driver? If that is necessary we need
> to give control to the driver first to decide if it is safe to do so.
>

An ethernet device is not necessarily a PCI device. With planned removal 
of rte_pci_device from rte_eth_device, this will be realized.
Similarly, the info is also not PCI device specific.

With the '+1', my intention was not to say we should do it in this 
patch. We should prepare eth_dev_info in similar manner as done for 
pci_dev of rte_eth_dev (ETH_DEV_PCI_DEV() style macro, or inline).

And now for whether we should expose lower level device details or not, 
I was of the view that keeping pci_dev linked to this structure exposes 
more lower level info than keeping rte_dev. Another view point could be 
to completely do away with pci_info within eth_dev_info - but, I am not 
sure of dependencies on it.

-
Shreyansh