patches for DPDK stable branches
 help / color / mirror / Atom feed
* [PATCH 19.11] net/af_packet: fix ignoring full ring on Tx
@ 2021-12-01 10:32 Tudor Cornea
  2021-12-01 15:38 ` Christian Ehrhardt
  0 siblings, 1 reply; 2+ messages in thread
From: Tudor Cornea @ 2021-12-01 10:32 UTC (permalink / raw)
  To: stable; +Cc: christian.ehrhardt, Tudor Cornea, Mihai Pogonaru

[ upstream commit f86d553cc180f9a65e115edb5641a49bbf2cf2f0 ]

The poll call can return POLLERR which is ignored, or it can return
POLLOUT, even if there are no free frames in the mmap-ed area.

We can account for both of these cases by re-checking if the next
frame is empty before writing into it.

We have attempted to reproduce this issue with pktgen-dpdk, using
the following configuration.

pktgen -l 1-4 -n 4 --proc-type=primary --no-pci --no-telemetry \
    --no-huge -m 512 \
    --vdev=net_af_packet0,iface=eth1,blocksz=16384,framesz=8192, \
    framecnt=2048,qpairs=1,qdisc_bypass=0 \
    -- \
    -P \
    -T \
    -m "3.0" \
    -f themes/black-yellow.theme

We configure a low tx rate (~ 335 packets / second) and a small
packet size, of about 300 Bytes from the pktgen CLI.

set 0 size 300
set 0 rate 0.008
set 0 burst 1
start 0

After bringing the interface down, and up again, we seem to arrive
in a state in which the tx rate is inconsistent, and does not recover.

ifconfig eth1 down; sleep 7; ifconfig eth1 up

[1] http://code.dpdk.org/pktgen-dpdk/pktgen-20.11.2/source/INSTALL.md

Fixes: 364e08f2bbc0 ("af_packet: add PMD for AF_PACKET-based virtual devices")

Signed-off-by: Mihai Pogonaru <pogonarumihai@gmail.com>
Signed-off-by: Tudor Cornea <tudor.cornea@gmail.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
---
 drivers/net/af_packet/rte_eth_af_packet.c | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)

diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
index 00387ed..dbbe8e5 100644
--- a/drivers/net/af_packet/rte_eth_af_packet.c
+++ b/drivers/net/af_packet/rte_eth_af_packet.c
@@ -213,8 +213,30 @@ eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 		}
 
 		/* point at the next incoming frame */
-		if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
-		    (poll(&pfd, 1, -1) < 0))
+		if (ppd->tp_status != TP_STATUS_AVAILABLE) {
+			if (poll(&pfd, 1, -1) < 0)
+				break;
+
+			/* poll() can return POLLERR if the interface is down */
+			if (pfd.revents & POLLERR)
+				break;
+		}
+
+		/*
+		 * poll() will almost always return POLLOUT, even if there
+		 * are no extra buffers available
+		 *
+		 * This happens, because packet_poll() calls datagram_poll()
+		 * which checks the space left in the socket buffer and,
+		 * in the case of packet_mmap, the default socket buffer length
+		 * doesn't match the requested size for the tx_ring.
+		 * As such, there is almost always space left in socket buffer,
+		 * which doesn't seem to be correlated to the requested size
+		 * for the tx_ring in packet_mmap.
+		 *
+		 * This results in poll() returning POLLOUT.
+		 */
+		if (ppd->tp_status != TP_STATUS_AVAILABLE)
 			break;
 
 		/* copy the tx frame data */
-- 
2.7.4


^ permalink raw reply	[flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-12-01 15:38 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-12-01 10:32 [PATCH 19.11] net/af_packet: fix ignoring full ring on Tx Tudor Cornea
2021-12-01 15:38 ` Christian Ehrhardt

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).