[dpdk-dev] [PATCH] ethdev: fix applications failure on configure

[Xueming(Steven) Li]

> -----Original Message-----
> From: Ferruh Yigit <ferruh.yigit at intel.com>
> Sent: Tuesday, May 1, 2018 10:32 PM
> To: Thomas Monjalon <thomas at monjalon.net>
> Cc: Xueming(Steven) Li <xuemingl at mellanox.com>; dev at dpdk.org
> Subject: Re: [PATCH] ethdev: fix applications failure on configure
> 
> On 5/1/2018 3:12 PM, Thomas Monjalon wrote:
> > 01/05/2018 16:08, Ferruh Yigit:
> >> On 5/1/2018 3:01 PM, Thomas Monjalon wrote:
> >>> 01/05/2018 15:33, Ferruh Yigit:
> >>>> Many sample applications fail because of
> >>>> dev_info.flow_type_rss_offloads check in rte_eth_dev_configure()
> >>>
> >>> We need to define the API behaviour in doxygen.
> >>>
> >>>> The sample applications need to be fixed/updated before returning
> >>>> error on rte_eth_dev_configure()
> >>>>
> >>>> This patch keeps the error log but removes returning error.
> >>>
> >>> If the doc is updated in 18.05, we can have a deprecation notice to
> >>> insert the error return in 18.08.
> >>
> >> +1 to add a deprecation notice for this release and do the error return on 18.08.
> >>
> >> Is the doc update you mentioned doxygen update?
> >
> > Yes, doxygen update.
> >
> >> Who can do doc update and deprecation notice patches for this release?
> >
> > Either you Ferruh, or Xueming and me together.
> > What do you prefer?
> 
> Or as Xueming suggested, we can take rss_hf config as best effort and not return error at all.
> 
> I think this forces PMDs to have up-to-date flow_type_rss_offloads values, is there any other benefit?
> What was the initial motivation to add error return on this check?

The original idea is to add rss_hf check on mlx5 PMD, while it looks more generic
to move the check to ethdev api from discussion[1].

[1] http://www.dpdk.org/ml/archives/dev/2018-April/095136.html

> 
> >
> >>>> Fixes: 8863a1fbfc66 ("ethdev: add supported hash function check")
> >>>> Cc: xuemingl at mellanox.com
> >>>>
> >>>> Signed-off-by: Ferruh Yigit <ferruh.yigit at intel.com>
> >
> >
> >