On Thu, Jul 09, 2015 at 04:37:48PM -0700, Stephen Hemminger wrote: > From: Stephen Hemminger <shemming at brocade.com> > > For applications that use m->userdata the initialization can > be a signficant (10%) performance penalty. > > Rather than taking the cache penalty of initializing userdata > in the receive handling, do it in the place where mbuf is > already cache hot and being setup.
Should the management of the userdata field not be the responsibility of the app itself, rather than having the PMD manage it? If the PMD does manage the userdata field, I would suggest taking the approach of having the field cleared by the mbuf library on free, rather than on RX. > > Signed-off-by: Stephen Hemminger <stephen at networkplumber.org> > --- > drivers/net/cxgbe/sge.c | 1 + > drivers/net/e1000/em_rxtx.c | 2 ++ > drivers/net/e1000/igb_rxtx.c | 2 ++ > drivers/net/enic/enic_main.c | 2 ++ > drivers/net/fm10k/fm10k.h | 1 + > drivers/net/i40e/i40e_rxtx.c | 4 ++++ > drivers/net/ixgbe/ixgbe_rxtx.c | 4 ++++ > drivers/net/ixgbe/ixgbe_rxtx_vec.c | 1 + > drivers/net/virtio/virtio_rxtx.c | 3 +++ > drivers/net/vmxnet3/vmxnet3_rxtx.c | 1 + > drivers/net/xenvirt/rte_eth_xenvirt.c | 1 + > lib/librte_mbuf/rte_mbuf.h | 1 + > 12 files changed, 23 insertions(+) > <snip> > diff --git a/drivers/net/ixgbe/ixgbe_rxtx_vec.c > b/drivers/net/ixgbe/ixgbe_rxtx_vec.c > index 912d3b4..d73d8dc 100644 > --- a/drivers/net/ixgbe/ixgbe_rxtx_vec.c > +++ b/drivers/net/ixgbe/ixgbe_rxtx_vec.c > @@ -742,6 +742,7 @@ ixgbe_rxq_vec_setup(struct ixgbe_rx_queue *rxq) > > mb_def.nb_segs = 1; > mb_def.data_off = RTE_PKTMBUF_HEADROOM; > + mb_def.userdata = NULL; > mb_def.port = rxq->port_id; > rte_mbuf_refcnt_set(&mb_def, 1); > This won't actually work for the vector PMD. The userdata field is not covered by the "rearm_data" part of the mbuf. /Bruce
