On 26 Oct 2018, at 11:53, Simon Horman wrote:
> On Fri, 26 Oct 2018 at 08:58, Eelco Chaudron <[email protected]> wrote: > <SNIP> >>>> I have a general comment, don't know where to put it, so I put it >>>> here. >>>> Some hardware might have multiple tables. If one type of table is >>>> full >>>> the ENOSPC might be returned, but it does not mean all type of flows >>>> can >>>> no longer be offloaded. This might be a situation to think about. >>> >>> Ok, thanks for bringing it up. Currently from OvS daemon's perspective >>> a >>> request to add/delete a flow is issued on a netdev and the failure >>> indicates >>> that the particular netdev is out of resources. If we need to handle >>> the >>> condition where HW has different tables, we need to further extend >>> this >>> design and the tc interfaces to propagate this fine grained >>> information. >> >> Would be good if other hardware vendors can comment here? >> > > There was a discussion in another forum involving at least Mellanox, > Broadcom and Netronome. > From a Netronome point of view this scheme is satisfactory and my > recollection is that > was the agreement of those involved in the discussion. Thanks for the clarification… _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
