On Sun, Jan 27, 2019 at 09:43:45AM +0200, Kamal Heib wrote: > > > On 1/24/19 9:10 PM, Jason Gunthorpe wrote: > > On Thu, Jan 24, 2019 at 09:02:56PM +0200, Yuval Shaia wrote: > >> On Thu, Jan 24, 2019 at 11:59:38AM -0700, Jason Gunthorpe wrote: > >>> On Thu, Jan 24, 2019 at 08:50:47PM +0200, Yuval Shaia wrote: > >>>> On Thu, Jan 24, 2019 at 11:30:48AM -0700, Jason Gunthorpe wrote: > >>>>> On Thu, Jan 24, 2019 at 04:44:34PM +0200, Kamal Heib wrote: > >>>>>> Currently a packet is marked for loopback only if the source and > >>>>>> destination addresses match, This is not enough when multiple gids are > >>>>>> present in rxe device's gid table and the traffic is from one gid to > >>>>>> another, this patch fixes this by marking the packet for loopback if the > >>>>>> destination address found in the rxe's netdevice address list. > >>>>> Shouldn't loop back detection be based on the DMAC/SMAC being equal > >>>>> and nothing else? > >>>>> > >>>>> Jason > >>>> No because even if DMAC != SMAC (and in our case dgid != sgid) we can still > >>>> can end up on the same rxe device. > >>>> > >>>> As the commit message trying to explain, take for example one rxe device > >>>> with two gids one send to the second, this should be processed as loopback, > >>>> right? > >>> rxe sits on top of a netdev - so it has *EXACTLY ONE* mac > >>> address. Anything else is a bug. > >>> > >>> Jason > >> ok, i think see your point. > >> So you are saying that instead of traveling the routing table we just need > >> to check if dmac is the same as smac, correct? > > dmac should be set to the 'exactly one' mac, so yes. > > > > Jsaon > > Thanks Yuval and Jason. > I'll prepare a new fix that will check if smac == dmac. > > Thanks, > Kamal Thanks, I was under the (probably wrong) assumption that dmac is yet unknown at this stage. Yuval
