[dpdk-dev] [PATCH] net/enic: fix TSO for packets greater than 9208 bytes

[dpdk-dev] [PATCH] net/enic: fix TSO for packets greater than 9208 bytes

[John Daley]

A check was previously added to drop Tx packets greater than what the Nic
is capable of sending since such packets can freeze the send queue. The
check did not account for TSO packets however, so TSO was limited to 9208
bytes.

Check packet length only for non-TSO packets. Also insure that TSO packet
segment size plus the headers do not exceed what the Nic is capable of
since this also can freeze the send queue.

Use the PKT_TX_TCP_SEG ol_flag instead of m->tso_segsz which is the
preferred way to check for TSO.

Fixes: ed6e564c214e ("net/enic: fix memory leak with oversized Tx packets")
Cc: stable@dpdk.org

Signed-off-by: John Daley <johndale@cisco.com>
---

Note that there is some more work to do on enic TSO- the header length is
calculated by looking at the packet instead of just trusting mbuf tso
offload header lengths. The 'tx_oversized' stat is used for more than just
oversized packets- it gets rolled into 'oerrors' so doesn't matter but the
name should be changed. Some TSO tunneling support can be added for newer
hardware. These changes will come in the next relase, but hope that this
patch can be accepted in 17.11 because it solves existing customer problem.

 drivers/net/enic/enic_rxtx.c | 25 +++++++++++++++++++------
 1 file changed, 19 insertions(+), 6 deletions(-)

diff --git a/drivers/net/enic/enic_rxtx.c b/drivers/net/enic/enic_rxtx.c
index a39172f14..e938193b5 100644
--- a/drivers/net/enic/enic_rxtx.c
+++ b/drivers/net/enic/enic_rxtx.c
@@ -546,12 +546,15 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	uint64_t bus_addr;
 	uint8_t offload_mode;
 	uint16_t header_len;
+	uint64_t tso;
+	rte_atomic64_t *tx_oversized;
 
 	enic_cleanup_wq(enic, wq);
 	wq_desc_avail = vnic_wq_desc_avail(wq);
 	head_idx = wq->head_idx;
 	desc_count = wq->ring.desc_count;
 	ol_flags_mask = PKT_TX_VLAN_PKT | PKT_TX_IP_CKSUM | PKT_TX_L4_MASK;
+	tx_oversized = &enic->soft_stats.tx_oversized;
 
 	nb_pkts = RTE_MIN(nb_pkts, ENIC_TX_XMIT_MAX);
 
@@ -561,10 +564,12 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 		data_len = tx_pkt->data_len;
 		ol_flags = tx_pkt->ol_flags;
 		nb_segs = tx_pkt->nb_segs;
+		tso = ol_flags & PKT_TX_TCP_SEG;
 
-		if (pkt_len > ENIC_TX_MAX_PKT_SIZE) {
+		/* drop packet if it's too big to send */
+		if (unlikely(!tso && (pkt_len > ENIC_TX_MAX_PKT_SIZE))) {
 			rte_pktmbuf_free(tx_pkt);
-			rte_atomic64_inc(&enic->soft_stats.tx_oversized);
+			rte_atomic64_inc(tx_oversized);
 			continue;
 		}
 
@@ -587,13 +592,21 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 		offload_mode = WQ_ENET_OFFLOAD_MODE_CSUM;
 		header_len = 0;
 
-		if (tx_pkt->tso_segsz) {
+		if (tso) {
 			header_len = tso_header_len(tx_pkt);
-			if (header_len) {
-				offload_mode = WQ_ENET_OFFLOAD_MODE_TSO;
-				mss = tx_pkt->tso_segsz;
+
+			/* Drop if non-TCP packet or TSO seg size is too big */
+			if (unlikely((header_len == 0) || ((tx_pkt->tso_segsz +
+			    header_len) > ENIC_TX_MAX_PKT_SIZE))) {
+				rte_pktmbuf_free(tx_pkt);
+				rte_atomic64_inc(tx_oversized);
+				continue;
 			}
+
+			offload_mode = WQ_ENET_OFFLOAD_MODE_TSO;
+			mss = tx_pkt->tso_segsz;
 		}
+
 		if ((ol_flags & ol_flags_mask) && (header_len == 0)) {
 			if (ol_flags & PKT_TX_IP_CKSUM)
 				mss |= ENIC_CALC_IP_CKSUM;
-- 
2.12.0

[dpdk-dev] [PATCH v2] net/enic: fix TSO for packets greater than 9208 bytes

[John Daley]

A check was previously added to drop Tx packets greater than what the Nic
is capable of sending since such packets can freeze the send queue. The
check did not account for TSO packets however, so TSO was limited to 9208
bytes.

Check packet length only for non-TSO packets. Also insure that TSO packet
segment size plus the headers do not exceed what the Nic is capable of
since this also can freeze the send queue.

Use the PKT_TX_TCP_SEG ol_flag instead of m->tso_segsz which is the
preferred way to check for TSO.

Fixes: ed6e564c214e ("net/enic: fix memory leak with oversized Tx packets")
Cc: stable@dpdk.org

Signed-off-by: John Daley <johndale@cisco.com>
---

Note that there is some more work to do on enic TSO- the header length is
calculated by looking at the packet instead of just trusting mbuf TSO
offload header lengths. The 'tx_oversized' stat is used for more than just
oversized packets- it gets rolled into 'oerrors' so doesn't matter but the
name should be changed. Some TSO tunneling support can be added for newer
hardware. These changes will come in the next relase, but hope that this
patch can be accepted in 17.11 because it solves existing customer problem.

v2: remeved extra parens, found by patchworks checkpatch, but not mine.

 drivers/net/enic/enic_rxtx.c | 25 +++++++++++++++++++------
 1 file changed, 19 insertions(+), 6 deletions(-)

diff --git a/drivers/net/enic/enic_rxtx.c b/drivers/net/enic/enic_rxtx.c
index 1d43bde9a..8291865c6 100644
--- a/drivers/net/enic/enic_rxtx.c
+++ b/drivers/net/enic/enic_rxtx.c
@@ -546,12 +546,15 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	uint64_t bus_addr;
 	uint8_t offload_mode;
 	uint16_t header_len;
+	uint64_t tso;
+	rte_atomic64_t *tx_oversized;
 
 	enic_cleanup_wq(enic, wq);
 	wq_desc_avail = vnic_wq_desc_avail(wq);
 	head_idx = wq->head_idx;
 	desc_count = wq->ring.desc_count;
 	ol_flags_mask = PKT_TX_VLAN_PKT | PKT_TX_IP_CKSUM | PKT_TX_L4_MASK;
+	tx_oversized = &enic->soft_stats.tx_oversized;
 
 	nb_pkts = RTE_MIN(nb_pkts, ENIC_TX_XMIT_MAX);
 
@@ -561,10 +564,12 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 		data_len = tx_pkt->data_len;
 		ol_flags = tx_pkt->ol_flags;
 		nb_segs = tx_pkt->nb_segs;
+		tso = ol_flags & PKT_TX_TCP_SEG;
 
-		if (pkt_len > ENIC_TX_MAX_PKT_SIZE) {
+		/* drop packet if it's too big to send */
+		if (unlikely(!tso && pkt_len > ENIC_TX_MAX_PKT_SIZE)) {
 			rte_pktmbuf_free(tx_pkt);
-			rte_atomic64_inc(&enic->soft_stats.tx_oversized);
+			rte_atomic64_inc(tx_oversized);
 			continue;
 		}
 
@@ -587,13 +592,21 @@ uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 		offload_mode = WQ_ENET_OFFLOAD_MODE_CSUM;
 		header_len = 0;
 
-		if (tx_pkt->tso_segsz) {
+		if (tso) {
 			header_len = tso_header_len(tx_pkt);
-			if (header_len) {
-				offload_mode = WQ_ENET_OFFLOAD_MODE_TSO;
-				mss = tx_pkt->tso_segsz;
+
+			/* Drop if non-TCP packet or TSO seg size is too big */
+			if (unlikely(header_len == 0 || ((tx_pkt->tso_segsz +
+			    header_len) > ENIC_TX_MAX_PKT_SIZE))) {
+				rte_pktmbuf_free(tx_pkt);
+				rte_atomic64_inc(tx_oversized);
+				continue;
 			}
+
+			offload_mode = WQ_ENET_OFFLOAD_MODE_TSO;
+			mss = tx_pkt->tso_segsz;
 		}
+
 		if ((ol_flags & ol_flags_mask) && (header_len == 0)) {
 			if (ol_flags & PKT_TX_IP_CKSUM)
 				mss |= ENIC_CALC_IP_CKSUM;
-- 
2.12.0

Re: [dpdk-dev] [PATCH v2] net/enic: fix TSO for packets greater than 9208 bytes

[Ferruh Yigit]

On 11/1/2017 10:47 PM, John Daley wrote:
> A check was previously added to drop Tx packets greater than what the Nic
> is capable of sending since such packets can freeze the send queue. The
> check did not account for TSO packets however, so TSO was limited to 9208
> bytes.
> 
> Check packet length only for non-TSO packets. Also insure that TSO packet
> segment size plus the headers do not exceed what the Nic is capable of
> since this also can freeze the send queue.
> 
> Use the PKT_TX_TCP_SEG ol_flag instead of m->tso_segsz which is the
> preferred way to check for TSO.
> 
> Fixes: ed6e564c214e ("net/enic: fix memory leak with oversized Tx packets")
> Cc: stable@dpdk.org
> 
> Signed-off-by: John Daley <johndale@cisco.com>

Applied to dpdk-next-net/master, thanks.