CC'ing follow up conversation which I accidentally took of list. -------- Forwarded Message -------- Subject: Re: [dpdk-dev] [PATCH] bond: inherit maximum rx packet length Date: Wed, 4 May 2016 14:11:53 -0400 From: Eric Kinzie <ehkin...@gmail.com> To: Declan Doherty <declan.doherty at intel.com>
On Wed May 04 14:53:26 +0100 2016, Declan Doherty wrote: > On 29/04/16 22:36, Eric Kinzie wrote: > >On Tue Apr 26 11:51:53 +0100 2016, Declan Doherty wrote: > >>On 14/04/16 18:23, Eric Kinzie wrote: > >>> Instead of a hard-coded maximum receive length, allow the bond interface > >>> to inherit this limit from the first slave added. This allows > >>> an application that uses jumbo frames to pass realistic values to > >>> rte_eth_dev_configure without causing an error. > >>> > >>>Signed-off-by: Eric Kinzie <ehkinzie at gmail.com> > >>>--- > >>... > >>> > >> > >>Hey Eric, just one small thing, I think it probably makes sense to > >>return the max rx pktlen for all slaves, so as we add each slave > >>just check if that the slave being value is larger than the current > >>value. > >> > >>@@ -385,6 +389,10 @@ __eth_bond_slave_add_lock_free(uint8_t > >>bonded_port_id, uint8_t slave_port_id) > >> internals->tx_offload_capa &= dev_info.tx_offload_capa; > >> internals->flow_type_rss_offloads &= > >>dev_info.flow_type_rss_offloads; > >> > >>+ /* If new slave's max rx packet size is larger than > >>current value then override */ > >>+ if (dev_info.max_rx_pktlen > internals->max_rx_pktlen) > >>+ internals->max_rx_pktlen = dev_info.max_rx_pktlen; > >>+ > >> > >>Declan > > > >Declan, I sent an updated patch but now release that I mis-read your > >comments. Is it a good idea to change the value once it's been set? > >My patch now refuses to add a slave with a pktlen value that's smaller > >than that of the first slave. > > > >Eric > > > > Hey Eric, > > actually I think you're right, we can't change the value dynamically > once the bonded device has been configured (maybe gating on > start/stop would be sufficient), but I think we shouldn't explicitly > gate on the first slave added, as we may be bonding multiple slaves > each of which could have different max_rx_pktlens. Prior to calling > dev_configure on the bonded device it should be possible to add any > slave with any max_rx_pktlen and inherit the minimum value, but once > the bonded device has been configured we would refuse a slave with a > max_rx_pktlen value which is smaller than the current value. I think > this should then satisfy that all slaves meet the minimum > requirements published by the bonded device? > Hi, Declan. I think that would work out ok. I'll write something to that effect and send another version of the patch.