On Fri, Apr 10, 2020 at 2:26 PM Thomas Monjalon <tho...@monjalon.net> wrote: > > 10/04/2020 10:44, Ferruh Yigit: > > On 10/25/2019 2:39 PM, Jerin Jacob wrote: > > > On Fri, Oct 25, 2019 at 6:56 PM Thomas Monjalon <tho...@monjalon.net> > > > wrote: > > >> > > >> 25/10/2019 14:51, Ferruh Yigit: > > >>> "Flow API" is a method/API to implement various filtering features, on > > >>> its own it doesn't give much context on what features are provided. And > > >>> it is not really a feature, so doesn't fit into feature table. > > >>> > > >>> Also since other filtering related APIs, 'filter_ctrl', has been > > >>> deprecated, flow API is the only supported way in the DPDK to implement > > >>> filtering options, if related filter options announced by PMDs, listing > > >>> "Flow API" as implemented is redundant information. > > >> > > >> I fully agree with this explanation. > > >> rte_flow is the only supported API for flow offloads. > > >> That's why we must remove the legacy API. > > >> > > >>> Signed-off-by: Ferruh Yigit <ferruh.yi...@intel.com> > > >>> --- > > >>> --- a/doc/guides/nics/features/default.ini > > >>> +++ b/doc/guides/nics/features/default.ini > > >>> -Flow API = > > >> > > >> Acked-by: Thomas Monjalon <tho...@monjalon.net> > > > > > > # Need to remove "Flow API" from doc/guides/nics/features.rst > > > > +1 > > > > > # Need to remove refference of "Flow API" from "doc/guides/nics/*" as > > > well. > > > > "Flow API" is the implementation of the filtering, it may exist in the nic > > documentation, only it is not a feature on itself. I will scan the docs for > > usage. > > > > > > > > Not specific to this patch, > > > Probably we need to add a new matrix to enumerate PATTERN and ACTIONS > > > supported by each PMD as a rte_flow feature matrix. > > > That some else can take it up if everyone agrees the semantics. > > > > > > > +1, there needs a way to figure out which filtering is supported by a > > device/driver. It is not documented and it is very hard to got it from the > > code. > > > > Not sure if a new matrix is the good way to go, but I agree we need some > > way to > > clarify it. > > I think we should split the matrix. > Adding a new matrix for flow offloads looks the way to go. > I suggest 3 matrices: > - port-level features > - queue-level features > - flow-level features
Not sure what will be the details in "flow-level features". IMO, We need to have a separate matrix for subdomain features for rte_flow, rte_tm, rte_mtr, etc which part of ethdev. For instance, rte_flow features can be translated into a matrix of supported PATTERN and ACTIONS. > >
