Message ID | 201006151735.17258.ossthema@de.ibm.com (mailing list archive) |
---|---|
State | Not Applicable |
Headers | show |
Jan-Bernd Themann <ossthema@de.ibm.com> wrote: >In the eHEA poll function an rmb() is required. Without that some packets >on the receive queue are not seen and thus delayed until the next interrupt >is handled for the same receive queue. > >Signed-off-by: Jan-Bernd Themann <themann@de.ibm.com> To add a bit of background, this could manifest during a netperf TCP_RR or UDP_RR on an otherwise idle network. TCP would occasionally retransmit, but then both the original segment and the retransmission would simultaneously appear at the receiver. For UDP_RR, message sizes in excess of the mtu would occasionally "lose" an IP fragment, and eventually IP reassembly would time out. -J Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> >--- >Patch created against 2.6.35-rc3 > > drivers/net/ehea/ehea_main.c | 1 + > 1 files changed, 1 insertions(+), 0 deletions(-) > >diff --git a/drivers/net/ehea/ehea_main.c b/drivers/net/ehea/ehea_main.c >index f547894..fd890fa 100644 >--- a/drivers/net/ehea/ehea_main.c >+++ b/drivers/net/ehea/ehea_main.c >@@ -867,6 +867,7 @@ static int ehea_poll(struct napi_struct *napi, int budget) > ehea_reset_cq_ep(pr->send_cq); > ehea_reset_cq_n1(pr->recv_cq); > ehea_reset_cq_n1(pr->send_cq); >+ rmb(); > cqe = ehea_poll_rq1(pr->qp, &wqe_index); > cqe_skb = ehea_poll_cq(pr->send_cq); > >-- >1.7.0 >-- >To unsubscribe from this list: send the line "unsubscribe netdev" in >the body of a message to majordomo@vger.kernel.org >More majordomo info at http://vger.kernel.org/majordomo-info.html
From: Jay Vosburgh <fubar@us.ibm.com> Date: Tue, 15 Jun 2010 09:45:47 -0700 > Jan-Bernd Themann <ossthema@de.ibm.com> wrote: > >>In the eHEA poll function an rmb() is required. Without that some packets >>on the receive queue are not seen and thus delayed until the next interrupt >>is handled for the same receive queue. >> >>Signed-off-by: Jan-Bernd Themann <themann@de.ibm.com> > > To add a bit of background, this could manifest during a netperf > TCP_RR or UDP_RR on an otherwise idle network. TCP would occasionally > retransmit, but then both the original segment and the retransmission > would simultaneously appear at the receiver. For UDP_RR, message sizes > in excess of the mtu would occasionally "lose" an IP fragment, and > eventually IP reassembly would time out. > > -J > > Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Applied.
diff --git a/drivers/net/ehea/ehea_main.c b/drivers/net/ehea/ehea_main.c index f547894..fd890fa 100644 --- a/drivers/net/ehea/ehea_main.c +++ b/drivers/net/ehea/ehea_main.c @@ -867,6 +867,7 @@ static int ehea_poll(struct napi_struct *napi, int budget) ehea_reset_cq_ep(pr->send_cq); ehea_reset_cq_n1(pr->recv_cq); ehea_reset_cq_n1(pr->send_cq); + rmb(); cqe = ehea_poll_rq1(pr->qp, &wqe_index); cqe_skb = ehea_poll_cq(pr->send_cq);
In the eHEA poll function an rmb() is required. Without that some packets on the receive queue are not seen and thus delayed until the next interrupt is handled for the same receive queue. Signed-off-by: Jan-Bernd Themann <themann@de.ibm.com> --- Patch created against 2.6.35-rc3 drivers/net/ehea/ehea_main.c | 1 + 1 files changed, 1 insertions(+), 0 deletions(-)