diff mbox

[dpdk-dev,v4,12/13] testpmd: support TSO in csum forward engine

Message ID 1417014295-29064-13-git-send-email-olivier.matz@6wind.com (mailing list archive)
State Accepted, archived
Headers show

Commit Message

Olivier Matz Nov. 26, 2014, 3:04 p.m. UTC
Add two new commands in testpmd:

- tso set <segsize> <portid>
- tso show <portid>

These commands can be used enable TSO when transmitting TCP packets in
the csum forward engine. Ex:

  set fwd csum
  tx_checksum set ip hw 0
  tso set 800 0
  start

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
---
 app/test-pmd/cmdline.c  | 92 +++++++++++++++++++++++++++++++++++++++++++++++++
 app/test-pmd/csumonly.c | 64 ++++++++++++++++++++++++----------
 app/test-pmd/testpmd.h  |  1 +
 3 files changed, 139 insertions(+), 18 deletions(-)

Comments

Ananyev, Konstantin Nov. 26, 2014, 9:23 p.m. UTC | #1
> From: Olivier Matz [mailto:olivier.matz@6wind.com]
> Sent: Wednesday, November 26, 2014 3:05 PM
> To: dev@dpdk.org
> Cc: olivier.matz@6wind.com; Walukiewicz, Miroslaw; Liu, Jijiang; Liu, Yong; jigsaw@gmail.com; Richardson, Bruce; Ananyev, Konstantin
> Subject: [PATCH v4 12/13] testpmd: support TSO in csum forward engine
> 
> Add two new commands in testpmd:
> 
> - tso set <segsize> <portid>
> - tso show <portid>
> 
> These commands can be used enable TSO when transmitting TCP packets in
> the csum forward engine. Ex:
> 
>   set fwd csum
>   tx_checksum set ip hw 0
>   tso set 800 0
>   start
> 
> Signed-off-by: Olivier Matz <olivier.matz@6wind.com>

Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>

> ---
>  app/test-pmd/cmdline.c  | 92 +++++++++++++++++++++++++++++++++++++++++++++++++
>  app/test-pmd/csumonly.c | 64 ++++++++++++++++++++++++----------
>  app/test-pmd/testpmd.h  |  1 +
>  3 files changed, 139 insertions(+), 18 deletions(-)
> 
> diff --git a/app/test-pmd/cmdline.c b/app/test-pmd/cmdline.c
> index 722cd76..2a8c260 100644
> --- a/app/test-pmd/cmdline.c
> +++ b/app/test-pmd/cmdline.c
> @@ -329,6 +329,14 @@ static void cmd_help_long_parsed(void *parsed_result,
>  			"tx_checksum show (port_id)\n"
>  			"    Display tx checksum offload configuration\n\n"
> 
> +			"tso set (segsize) (portid)\n"
> +			"    Enable TCP Segmentation Offload in csum forward"
> +			" engine.\n"
> +			"    Please check the NIC datasheet for HW limits.\n\n"
> +
> +			"tso show (portid)"
> +			"    Display the status of TCP Segmentation Offload.\n\n"
> +
>  			"set fwd (%s)\n"
>  			"    Set packet forwarding mode.\n\n"
> 
> @@ -2984,6 +2992,88 @@ cmdline_parse_inst_t cmd_tx_cksum_show = {
>  	},
>  };
> 
> +/* *** ENABLE HARDWARE SEGMENTATION IN TX PACKETS *** */
> +struct cmd_tso_set_result {
> +	cmdline_fixed_string_t tso;
> +	cmdline_fixed_string_t mode;
> +	uint16_t tso_segsz;
> +	uint8_t port_id;
> +};
> +
> +static void
> +cmd_tso_set_parsed(void *parsed_result,
> +		       __attribute__((unused)) struct cmdline *cl,
> +		       __attribute__((unused)) void *data)
> +{
> +	struct cmd_tso_set_result *res = parsed_result;
> +	struct rte_eth_dev_info dev_info;
> +
> +	if (port_id_is_invalid(res->port_id))
> +		return;
> +
> +	if (!strcmp(res->mode, "set"))
> +		ports[res->port_id].tso_segsz = res->tso_segsz;
> +
> +	if (ports[res->port_id].tso_segsz == 0)
> +		printf("TSO is disabled\n");
> +	else
> +		printf("TSO segment size is %d\n",
> +			ports[res->port_id].tso_segsz);
> +
> +	/* display warnings if configuration is not supported by the NIC */
> +	rte_eth_dev_info_get(res->port_id, &dev_info);
> +	if ((ports[res->port_id].tso_segsz != 0) &&
> +		(dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) == 0) {
> +		printf("Warning: TSO enabled but not "
> +			"supported by port %d\n", res->port_id);
> +	}
> +}
> +
> +cmdline_parse_token_string_t cmd_tso_set_tso =
> +	TOKEN_STRING_INITIALIZER(struct cmd_tso_set_result,
> +				tso, "tso");
> +cmdline_parse_token_string_t cmd_tso_set_mode =
> +	TOKEN_STRING_INITIALIZER(struct cmd_tso_set_result,
> +				mode, "set");
> +cmdline_parse_token_num_t cmd_tso_set_tso_segsz =
> +	TOKEN_NUM_INITIALIZER(struct cmd_tso_set_result,
> +				tso_segsz, UINT16);
> +cmdline_parse_token_num_t cmd_tso_set_portid =
> +	TOKEN_NUM_INITIALIZER(struct cmd_tso_set_result,
> +				port_id, UINT8);
> +
> +cmdline_parse_inst_t cmd_tso_set = {
> +	.f = cmd_tso_set_parsed,
> +	.data = NULL,
> +	.help_str = "Set TSO segment size for csum engine (0 to disable): "
> +	"tso set <tso_segsz> <port>",
> +	.tokens = {
> +		(void *)&cmd_tso_set_tso,
> +		(void *)&cmd_tso_set_mode,
> +		(void *)&cmd_tso_set_tso_segsz,
> +		(void *)&cmd_tso_set_portid,
> +		NULL,
> +	},
> +};
> +
> +cmdline_parse_token_string_t cmd_tso_show_mode =
> +	TOKEN_STRING_INITIALIZER(struct cmd_tso_set_result,
> +				mode, "show");
> +
> +
> +cmdline_parse_inst_t cmd_tso_show = {
> +	.f = cmd_tso_set_parsed,
> +	.data = NULL,
> +	.help_str = "Show TSO segment size for csum engine: "
> +	"tso show <port>",
> +	.tokens = {
> +		(void *)&cmd_tso_set_tso,
> +		(void *)&cmd_tso_show_mode,
> +		(void *)&cmd_tso_set_portid,
> +		NULL,
> +	},
> +};
> +
>  /* *** ENABLE/DISABLE FLUSH ON RX STREAMS *** */
>  struct cmd_set_flush_rx {
>  	cmdline_fixed_string_t set;
> @@ -8660,6 +8750,8 @@ cmdline_parse_ctx_t main_ctx[] = {
>  	(cmdline_parse_inst_t *)&cmd_tx_vlan_set_pvid,
>  	(cmdline_parse_inst_t *)&cmd_tx_cksum_set,
>  	(cmdline_parse_inst_t *)&cmd_tx_cksum_show,
> +	(cmdline_parse_inst_t *)&cmd_tso_set,
> +	(cmdline_parse_inst_t *)&cmd_tso_show,
>  	(cmdline_parse_inst_t *)&cmd_link_flow_control_set,
>  	(cmdline_parse_inst_t *)&cmd_link_flow_control_set_rx,
>  	(cmdline_parse_inst_t *)&cmd_link_flow_control_set_tx,
> diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c
> index 9a5408e..490342f 100644
> --- a/app/test-pmd/csumonly.c
> +++ b/app/test-pmd/csumonly.c
> @@ -88,12 +88,12 @@
>  #endif
> 
>  static uint16_t
> -get_psd_sum(void *l3_hdr, uint16_t ethertype)
> +get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
>  {
>  	if (ethertype == _htons(ETHER_TYPE_IPv4))
> -		return rte_ipv4_phdr_cksum(l3_hdr);
> +		return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
>  	else /* assume ethertype == ETHER_TYPE_IPv6 */
> -		return rte_ipv6_phdr_cksum(l3_hdr);
> +		return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
>  }
> 
>  static uint16_t
> @@ -108,14 +108,15 @@ get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
>  /*
>   * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
>   * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
> - * header.
> + * header. The l4_len argument is only set in case of TCP (useful for TSO).
>   */
>  static void
>  parse_ethernet(struct ether_hdr *eth_hdr, uint16_t *ethertype, uint16_t *l2_len,
> -	uint16_t *l3_len, uint8_t *l4_proto)
> +	uint16_t *l3_len, uint8_t *l4_proto, uint16_t *l4_len)
>  {
>  	struct ipv4_hdr *ipv4_hdr;
>  	struct ipv6_hdr *ipv6_hdr;
> +	struct tcp_hdr *tcp_hdr;
> 
>  	*l2_len = sizeof(struct ether_hdr);
>  	*ethertype = eth_hdr->ether_type;
> @@ -143,6 +144,14 @@ parse_ethernet(struct ether_hdr *eth_hdr, uint16_t *ethertype, uint16_t *l2_len,
>  		*l4_proto = 0;
>  		break;
>  	}
> +
> +	if (*l4_proto == IPPROTO_TCP) {
> +		tcp_hdr = (struct tcp_hdr *)((char *)eth_hdr +
> +			*l2_len + *l3_len);
> +		*l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
> +	}
> +	else
> +		*l4_len = 0;
>  }
> 
>  /* modify the IPv4 or IPv4 source address of a packet */
> @@ -165,7 +174,7 @@ change_ip_addresses(void *l3_hdr, uint16_t ethertype)
>   * depending on the testpmd command line configuration */
>  static uint64_t
>  process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
> -	uint8_t l4_proto, uint16_t testpmd_ol_flags)
> +	uint8_t l4_proto, uint16_t tso_segsz, uint16_t testpmd_ol_flags)
>  {
>  	struct ipv4_hdr *ipv4_hdr = l3_hdr;
>  	struct udp_hdr *udp_hdr;
> @@ -177,11 +186,16 @@ process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
>  		ipv4_hdr = l3_hdr;
>  		ipv4_hdr->hdr_checksum = 0;
> 
> -		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
> +		if (tso_segsz != 0 && l4_proto == IPPROTO_TCP) {
>  			ol_flags |= PKT_TX_IP_CKSUM;
> -		else
> -			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
> -
> +		}
> +		else {
> +			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
> +				ol_flags |= PKT_TX_IP_CKSUM;
> +			else
> +				ipv4_hdr->hdr_checksum =
> +					rte_ipv4_cksum(ipv4_hdr);
> +		}
>  		ol_flags |= PKT_TX_IPV4;
>  	}
>  	else if (ethertype == _htons(ETHER_TYPE_IPv6))
> @@ -197,7 +211,7 @@ process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
>  			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
>  				ol_flags |= PKT_TX_UDP_CKSUM;
>  				udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
> -					ethertype);
> +					ethertype, ol_flags);
>  			}
>  			else {
>  				udp_hdr->dgram_cksum =
> @@ -209,9 +223,13 @@ process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
>  	else if (l4_proto == IPPROTO_TCP) {
>  		tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + l3_len);
>  		tcp_hdr->cksum = 0;
> -		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
> +		if (tso_segsz != 0) {
> +			ol_flags |= PKT_TX_TCP_SEG;
> +			tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype, ol_flags);
> +		}
> +		else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
>  			ol_flags |= PKT_TX_TCP_CKSUM;
> -			tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype);
> +			tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype, ol_flags);
>  		}
>  		else {
>  			tcp_hdr->cksum =
> @@ -282,6 +300,8 @@ process_outer_cksums(void *outer_l3_hdr, uint16_t outer_ethertype,
>   *  - modify the IPs in inner headers and in outer headers if any
>   *  - reprocess the checksum of all supported layers. This is done in SW
>   *    or HW, depending on testpmd command line configuration
> + *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
> + *    segmentation offload (this implies HW TCP checksum)
>   * Then transmit packets on the output port.
>   *
>   * (1) Supported packets are:
> @@ -312,7 +332,9 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
>  	uint16_t testpmd_ol_flags;
>  	uint8_t l4_proto;
>  	uint16_t ethertype = 0, outer_ethertype = 0;
> -	uint16_t  l2_len = 0, l3_len = 0, outer_l2_len = 0, outer_l3_len = 0;
> +	uint16_t l2_len = 0, l3_len = 0, l4_len = 0;
> +	uint16_t outer_l2_len = 0, outer_l3_len = 0;
> +	uint16_t tso_segsz;
>  	int tunnel = 0;
>  	uint32_t rx_bad_ip_csum;
>  	uint32_t rx_bad_l4_csum;
> @@ -342,6 +364,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
> 
>  	txp = &ports[fs->tx_port];
>  	testpmd_ol_flags = txp->tx_ol_flags;
> +	tso_segsz = txp->tso_segsz;
> 
>  	for (i = 0; i < nb_rx; i++) {
> 
> @@ -357,7 +380,8 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
>  		 * and inner headers */
> 
>  		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
> -		parse_ethernet(eth_hdr, &ethertype, &l2_len, &l3_len, &l4_proto);
> +		parse_ethernet(eth_hdr, &ethertype, &l2_len, &l3_len,
> +			&l4_proto, &l4_len);
>  		l3_hdr = (char *)eth_hdr + l2_len;
> 
>  		/* check if it's a supported tunnel (only vxlan for now) */
> @@ -385,7 +409,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
>  					sizeof(struct vxlan_hdr));
> 
>  				parse_ethernet(eth_hdr, &ethertype, &l2_len,
> -					&l3_len, &l4_proto);
> +					&l3_len, &l4_proto, &l4_len);
>  				l3_hdr = (char *)eth_hdr + l2_len;
>  			}
>  		}
> @@ -399,11 +423,12 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
> 
>  		/* step 3: depending on user command line configuration,
>  		 * recompute checksum either in software or flag the
> -		 * mbuf to offload the calculation to the NIC */
> +		 * mbuf to offload the calculation to the NIC. If TSO
> +		 * is configured, prepare the mbuf for TCP segmentation. */
> 
>  		/* process checksums of inner headers first */
>  		ol_flags |= process_inner_cksums(l3_hdr, ethertype,
> -			l3_len, l4_proto, testpmd_ol_flags);
> +			l3_len, l4_proto, tso_segsz, testpmd_ol_flags);
> 
>  		/* Then process outer headers if any. Note that the software
>  		 * checksum will be wrong if one of the inner checksums is
> @@ -432,6 +457,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
>  					sizeof(struct udp_hdr) +
>  					sizeof(struct vxlan_hdr) + l2_len;
>  				m->l3_len = l3_len;
> +				m->l4_len = l4_len;
>  			}
>  		} else {
>  			/* this is only useful if an offload flag is
> @@ -439,7 +465,9 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
>  			 * case */
>  			m->l2_len = l2_len;
>  			m->l3_len = l3_len;
> +			m->l4_len = l4_len;
>  		}
> +		m->tso_segsz = tso_segsz;
>  		m->ol_flags = ol_flags;
> 
>  	}
> diff --git a/app/test-pmd/testpmd.h b/app/test-pmd/testpmd.h
> index 83311fa..16c6fbf 100644
> --- a/app/test-pmd/testpmd.h
> +++ b/app/test-pmd/testpmd.h
> @@ -149,6 +149,7 @@ struct rte_port {
>  	struct fwd_stream       *tx_stream; /**< Port TX stream, if unique */
>  	unsigned int            socket_id;  /**< For NUMA support */
>  	uint16_t                tx_ol_flags;/**< TX Offload Flags (TESTPMD_TX_OFFLOAD...). */
> +	uint16_t                tso_segsz;  /**< MSS for segmentation offload. */
>  	uint16_t                tx_vlan_id; /**< Tag Id. in TX VLAN packets. */
>  	void                    *fwd_ctx;   /**< Forwarding mode context */
>  	uint64_t                rx_bad_ip_csum; /**< rx pkts with bad ip checksum  */
> --
> 2.1.0
diff mbox

Patch

diff --git a/app/test-pmd/cmdline.c b/app/test-pmd/cmdline.c
index 722cd76..2a8c260 100644
--- a/app/test-pmd/cmdline.c
+++ b/app/test-pmd/cmdline.c
@@ -329,6 +329,14 @@  static void cmd_help_long_parsed(void *parsed_result,
 			"tx_checksum show (port_id)\n"
 			"    Display tx checksum offload configuration\n\n"
 
+			"tso set (segsize) (portid)\n"
+			"    Enable TCP Segmentation Offload in csum forward"
+			" engine.\n"
+			"    Please check the NIC datasheet for HW limits.\n\n"
+
+			"tso show (portid)"
+			"    Display the status of TCP Segmentation Offload.\n\n"
+
 			"set fwd (%s)\n"
 			"    Set packet forwarding mode.\n\n"
 
@@ -2984,6 +2992,88 @@  cmdline_parse_inst_t cmd_tx_cksum_show = {
 	},
 };
 
+/* *** ENABLE HARDWARE SEGMENTATION IN TX PACKETS *** */
+struct cmd_tso_set_result {
+	cmdline_fixed_string_t tso;
+	cmdline_fixed_string_t mode;
+	uint16_t tso_segsz;
+	uint8_t port_id;
+};
+
+static void
+cmd_tso_set_parsed(void *parsed_result,
+		       __attribute__((unused)) struct cmdline *cl,
+		       __attribute__((unused)) void *data)
+{
+	struct cmd_tso_set_result *res = parsed_result;
+	struct rte_eth_dev_info dev_info;
+
+	if (port_id_is_invalid(res->port_id))
+		return;
+
+	if (!strcmp(res->mode, "set"))
+		ports[res->port_id].tso_segsz = res->tso_segsz;
+
+	if (ports[res->port_id].tso_segsz == 0)
+		printf("TSO is disabled\n");
+	else
+		printf("TSO segment size is %d\n",
+			ports[res->port_id].tso_segsz);
+
+	/* display warnings if configuration is not supported by the NIC */
+	rte_eth_dev_info_get(res->port_id, &dev_info);
+	if ((ports[res->port_id].tso_segsz != 0) &&
+		(dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) == 0) {
+		printf("Warning: TSO enabled but not "
+			"supported by port %d\n", res->port_id);
+	}
+}
+
+cmdline_parse_token_string_t cmd_tso_set_tso =
+	TOKEN_STRING_INITIALIZER(struct cmd_tso_set_result,
+				tso, "tso");
+cmdline_parse_token_string_t cmd_tso_set_mode =
+	TOKEN_STRING_INITIALIZER(struct cmd_tso_set_result,
+				mode, "set");
+cmdline_parse_token_num_t cmd_tso_set_tso_segsz =
+	TOKEN_NUM_INITIALIZER(struct cmd_tso_set_result,
+				tso_segsz, UINT16);
+cmdline_parse_token_num_t cmd_tso_set_portid =
+	TOKEN_NUM_INITIALIZER(struct cmd_tso_set_result,
+				port_id, UINT8);
+
+cmdline_parse_inst_t cmd_tso_set = {
+	.f = cmd_tso_set_parsed,
+	.data = NULL,
+	.help_str = "Set TSO segment size for csum engine (0 to disable): "
+	"tso set <tso_segsz> <port>",
+	.tokens = {
+		(void *)&cmd_tso_set_tso,
+		(void *)&cmd_tso_set_mode,
+		(void *)&cmd_tso_set_tso_segsz,
+		(void *)&cmd_tso_set_portid,
+		NULL,
+	},
+};
+
+cmdline_parse_token_string_t cmd_tso_show_mode =
+	TOKEN_STRING_INITIALIZER(struct cmd_tso_set_result,
+				mode, "show");
+
+
+cmdline_parse_inst_t cmd_tso_show = {
+	.f = cmd_tso_set_parsed,
+	.data = NULL,
+	.help_str = "Show TSO segment size for csum engine: "
+	"tso show <port>",
+	.tokens = {
+		(void *)&cmd_tso_set_tso,
+		(void *)&cmd_tso_show_mode,
+		(void *)&cmd_tso_set_portid,
+		NULL,
+	},
+};
+
 /* *** ENABLE/DISABLE FLUSH ON RX STREAMS *** */
 struct cmd_set_flush_rx {
 	cmdline_fixed_string_t set;
@@ -8660,6 +8750,8 @@  cmdline_parse_ctx_t main_ctx[] = {
 	(cmdline_parse_inst_t *)&cmd_tx_vlan_set_pvid,
 	(cmdline_parse_inst_t *)&cmd_tx_cksum_set,
 	(cmdline_parse_inst_t *)&cmd_tx_cksum_show,
+	(cmdline_parse_inst_t *)&cmd_tso_set,
+	(cmdline_parse_inst_t *)&cmd_tso_show,
 	(cmdline_parse_inst_t *)&cmd_link_flow_control_set,
 	(cmdline_parse_inst_t *)&cmd_link_flow_control_set_rx,
 	(cmdline_parse_inst_t *)&cmd_link_flow_control_set_tx,
diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c
index 9a5408e..490342f 100644
--- a/app/test-pmd/csumonly.c
+++ b/app/test-pmd/csumonly.c
@@ -88,12 +88,12 @@ 
 #endif
 
 static uint16_t
-get_psd_sum(void *l3_hdr, uint16_t ethertype)
+get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
 {
 	if (ethertype == _htons(ETHER_TYPE_IPv4))
-		return rte_ipv4_phdr_cksum(l3_hdr);
+		return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
 	else /* assume ethertype == ETHER_TYPE_IPv6 */
-		return rte_ipv6_phdr_cksum(l3_hdr);
+		return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
 }
 
 static uint16_t
@@ -108,14 +108,15 @@  get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
 /*
  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
- * header.
+ * header. The l4_len argument is only set in case of TCP (useful for TSO).
  */
 static void
 parse_ethernet(struct ether_hdr *eth_hdr, uint16_t *ethertype, uint16_t *l2_len,
-	uint16_t *l3_len, uint8_t *l4_proto)
+	uint16_t *l3_len, uint8_t *l4_proto, uint16_t *l4_len)
 {
 	struct ipv4_hdr *ipv4_hdr;
 	struct ipv6_hdr *ipv6_hdr;
+	struct tcp_hdr *tcp_hdr;
 
 	*l2_len = sizeof(struct ether_hdr);
 	*ethertype = eth_hdr->ether_type;
@@ -143,6 +144,14 @@  parse_ethernet(struct ether_hdr *eth_hdr, uint16_t *ethertype, uint16_t *l2_len,
 		*l4_proto = 0;
 		break;
 	}
+
+	if (*l4_proto == IPPROTO_TCP) {
+		tcp_hdr = (struct tcp_hdr *)((char *)eth_hdr +
+			*l2_len + *l3_len);
+		*l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
+	}
+	else
+		*l4_len = 0;
 }
 
 /* modify the IPv4 or IPv4 source address of a packet */
@@ -165,7 +174,7 @@  change_ip_addresses(void *l3_hdr, uint16_t ethertype)
  * depending on the testpmd command line configuration */
 static uint64_t
 process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
-	uint8_t l4_proto, uint16_t testpmd_ol_flags)
+	uint8_t l4_proto, uint16_t tso_segsz, uint16_t testpmd_ol_flags)
 {
 	struct ipv4_hdr *ipv4_hdr = l3_hdr;
 	struct udp_hdr *udp_hdr;
@@ -177,11 +186,16 @@  process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
 		ipv4_hdr = l3_hdr;
 		ipv4_hdr->hdr_checksum = 0;
 
-		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
+		if (tso_segsz != 0 && l4_proto == IPPROTO_TCP) {
 			ol_flags |= PKT_TX_IP_CKSUM;
-		else
-			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
-
+		}
+		else {
+			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
+				ol_flags |= PKT_TX_IP_CKSUM;
+			else
+				ipv4_hdr->hdr_checksum =
+					rte_ipv4_cksum(ipv4_hdr);
+		}
 		ol_flags |= PKT_TX_IPV4;
 	}
 	else if (ethertype == _htons(ETHER_TYPE_IPv6))
@@ -197,7 +211,7 @@  process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
 				ol_flags |= PKT_TX_UDP_CKSUM;
 				udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
-					ethertype);
+					ethertype, ol_flags);
 			}
 			else {
 				udp_hdr->dgram_cksum =
@@ -209,9 +223,13 @@  process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
 	else if (l4_proto == IPPROTO_TCP) {
 		tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + l3_len);
 		tcp_hdr->cksum = 0;
-		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
+		if (tso_segsz != 0) {
+			ol_flags |= PKT_TX_TCP_SEG;
+			tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype, ol_flags);
+		}
+		else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
 			ol_flags |= PKT_TX_TCP_CKSUM;
-			tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype);
+			tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype, ol_flags);
 		}
 		else {
 			tcp_hdr->cksum =
@@ -282,6 +300,8 @@  process_outer_cksums(void *outer_l3_hdr, uint16_t outer_ethertype,
  *  - modify the IPs in inner headers and in outer headers if any
  *  - reprocess the checksum of all supported layers. This is done in SW
  *    or HW, depending on testpmd command line configuration
+ *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
+ *    segmentation offload (this implies HW TCP checksum)
  * Then transmit packets on the output port.
  *
  * (1) Supported packets are:
@@ -312,7 +332,9 @@  pkt_burst_checksum_forward(struct fwd_stream *fs)
 	uint16_t testpmd_ol_flags;
 	uint8_t l4_proto;
 	uint16_t ethertype = 0, outer_ethertype = 0;
-	uint16_t  l2_len = 0, l3_len = 0, outer_l2_len = 0, outer_l3_len = 0;
+	uint16_t l2_len = 0, l3_len = 0, l4_len = 0;
+	uint16_t outer_l2_len = 0, outer_l3_len = 0;
+	uint16_t tso_segsz;
 	int tunnel = 0;
 	uint32_t rx_bad_ip_csum;
 	uint32_t rx_bad_l4_csum;
@@ -342,6 +364,7 @@  pkt_burst_checksum_forward(struct fwd_stream *fs)
 
 	txp = &ports[fs->tx_port];
 	testpmd_ol_flags = txp->tx_ol_flags;
+	tso_segsz = txp->tso_segsz;
 
 	for (i = 0; i < nb_rx; i++) {
 
@@ -357,7 +380,8 @@  pkt_burst_checksum_forward(struct fwd_stream *fs)
 		 * and inner headers */
 
 		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
-		parse_ethernet(eth_hdr, &ethertype, &l2_len, &l3_len, &l4_proto);
+		parse_ethernet(eth_hdr, &ethertype, &l2_len, &l3_len,
+			&l4_proto, &l4_len);
 		l3_hdr = (char *)eth_hdr + l2_len;
 
 		/* check if it's a supported tunnel (only vxlan for now) */
@@ -385,7 +409,7 @@  pkt_burst_checksum_forward(struct fwd_stream *fs)
 					sizeof(struct vxlan_hdr));
 
 				parse_ethernet(eth_hdr, &ethertype, &l2_len,
-					&l3_len, &l4_proto);
+					&l3_len, &l4_proto, &l4_len);
 				l3_hdr = (char *)eth_hdr + l2_len;
 			}
 		}
@@ -399,11 +423,12 @@  pkt_burst_checksum_forward(struct fwd_stream *fs)
 
 		/* step 3: depending on user command line configuration,
 		 * recompute checksum either in software or flag the
-		 * mbuf to offload the calculation to the NIC */
+		 * mbuf to offload the calculation to the NIC. If TSO
+		 * is configured, prepare the mbuf for TCP segmentation. */
 
 		/* process checksums of inner headers first */
 		ol_flags |= process_inner_cksums(l3_hdr, ethertype,
-			l3_len, l4_proto, testpmd_ol_flags);
+			l3_len, l4_proto, tso_segsz, testpmd_ol_flags);
 
 		/* Then process outer headers if any. Note that the software
 		 * checksum will be wrong if one of the inner checksums is
@@ -432,6 +457,7 @@  pkt_burst_checksum_forward(struct fwd_stream *fs)
 					sizeof(struct udp_hdr) +
 					sizeof(struct vxlan_hdr) + l2_len;
 				m->l3_len = l3_len;
+				m->l4_len = l4_len;
 			}
 		} else {
 			/* this is only useful if an offload flag is
@@ -439,7 +465,9 @@  pkt_burst_checksum_forward(struct fwd_stream *fs)
 			 * case */
 			m->l2_len = l2_len;
 			m->l3_len = l3_len;
+			m->l4_len = l4_len;
 		}
+		m->tso_segsz = tso_segsz;
 		m->ol_flags = ol_flags;
 
 	}
diff --git a/app/test-pmd/testpmd.h b/app/test-pmd/testpmd.h
index 83311fa..16c6fbf 100644
--- a/app/test-pmd/testpmd.h
+++ b/app/test-pmd/testpmd.h
@@ -149,6 +149,7 @@  struct rte_port {
 	struct fwd_stream       *tx_stream; /**< Port TX stream, if unique */
 	unsigned int            socket_id;  /**< For NUMA support */
 	uint16_t                tx_ol_flags;/**< TX Offload Flags (TESTPMD_TX_OFFLOAD...). */
+	uint16_t                tso_segsz;  /**< MSS for segmentation offload. */
 	uint16_t                tx_vlan_id; /**< Tag Id. in TX VLAN packets. */
 	void                    *fwd_ctx;   /**< Forwarding mode context */
 	uint64_t                rx_bad_ip_csum; /**< rx pkts with bad ip checksum  */