DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Yao, Lei A" <lei.a.yao@intel.com>
To: "Hu, Jiayu" <jiayu.hu@intel.com>, "dev@dpdk.org" <dev@dpdk.org>
Cc: "Yigit, Ferruh" <ferruh.yigit@intel.com>,
	"Tan, Jianfeng" <jianfeng.tan@intel.com>,
	"Ananyev, Konstantin" <konstantin.ananyev@intel.com>,
	"thomas@monjalon.net" <thomas@monjalon.net>,
	"Wu, Jingjing" <jingjing.wu@intel.com>
Subject: Re: [dpdk-dev] [PATCH v4] app/testpmd: enable the heavyweight mode TCP/IPv4 GRO
Date: Wed, 27 Sep 2017 07:23:50 +0000	[thread overview]
Message-ID: <2DBBFF226F7CF64BAFCA79B681719D953A2965D1@shsmsx102.ccr.corp.intel.com> (raw)
In-Reply-To: <1506407202-29843-1-git-send-email-jiayu.hu@intel.com>



> -----Original Message-----
> From: Hu, Jiayu
> Sent: Tuesday, September 26, 2017 2:27 PM
> To: dev@dpdk.org
> Cc: Yigit, Ferruh <ferruh.yigit@intel.com>; Tan, Jianfeng
> <jianfeng.tan@intel.com>; Ananyev, Konstantin
> <konstantin.ananyev@intel.com>; thomas@monjalon.net; Wu, Jingjing
> <jingjing.wu@intel.com>; Yao, Lei A <lei.a.yao@intel.com>; Hu, Jiayu
> <jiayu.hu@intel.com>
> Subject: [PATCH v4] app/testpmd: enable the heavyweight mode TCP/IPv4
> GRO
> 
> The GRO library provides two modes to reassemble packets. Currently, the
> csum forwarding engine has supported to use the lightweight mode to
> reassemble TCP/IPv4 packets. This patch introduces the heavyweight mode
> for TCP/IPv4 GRO in the csum forwarding engine.
> 
> With the command "set port <port_id> gro on|off", users can enable
> TCP/IPv4 GRO for a given port. With the command "set gro flush <cycles>",
> users can determine when the GROed TCP/IPv4 packets are flushed from
> reassembly tables. With the command "show port <port_id> gro", users can
> display GRO configuration.
> 
> The GRO library doesn't re-calculate checksums for merged packets. If
> users want the merged packets to have correct IP and TCP checksums,
> please select HW IP checksum calculation and HW TCP checksum calculation
> for the port which the merged packets are transmitted to.
> 
> Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Tested-by: Yao Lei<lei.a.yao@intel.com>

This patch has beed tested on my bench. The following
is the performance data got from iperf test with single flow
No GRO: 9.5 Gbps
Kernel GRO: 13.6 Gbps
DPDK GRO with flush cycle=1 : 25.9 Gbps
DPDK GRO with flush cycle=2 : 27.9 Gbps

Note: When use DPDK GRO with flush cycle=2, I set 
the vhost rx_queue_size to 1024, if use default number
256, sometimes I met the date stall. 
OS: Ubuntu 16.04
CPU: Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz  

> ---
> changes in v4:
> - fix unchecking the min value of 'cycle' bug in setup_gro_flush_cycles
> - update the context of the testpmd document and commit logs
> changes in v3:
> - remove "heavyweight mode" and "lightweight mode" from GRO
> commands
> - combine two patches into one
> - use consistent help string for GRO commands
> - remove the unnecessary command "gro set (max_flow_num)
> 	(max_item_num_per_flow) (port_id)"
> changes in v2:
> - use "set" and "show" as the root level command
> - add a new command to show GRO configuration
> - fix l2_len/l3_len/l4_len unset etc. bugs
> 
>  app/test-pmd/cmdline.c                      | 206 ++++++++++++++++------------
>  app/test-pmd/config.c                       |  68 +++++++--
>  app/test-pmd/csumonly.c                     |  31 ++++-
>  app/test-pmd/testpmd.c                      |  19 ++-
>  app/test-pmd/testpmd.h                      |  16 ++-
>  doc/guides/testpmd_app_ug/testpmd_funcs.rst |  50 +++++--
>  6 files changed, 270 insertions(+), 120 deletions(-)
> 
> diff --git a/app/test-pmd/cmdline.c b/app/test-pmd/cmdline.c
> index ccdf239..e44c02e 100644
> --- a/app/test-pmd/cmdline.c
> +++ b/app/test-pmd/cmdline.c
> @@ -423,13 +423,16 @@ static void cmd_help_long_parsed(void
> *parsed_result,
>  			"tso show (portid)"
>  			"    Display the status of TCP Segmentation
> Offload.\n\n"
> 
> -			"gro (on|off) (port_id)"
> +			"set port (port_id) gro on|off\n"
>  			"    Enable or disable Generic Receive Offload in"
>  			" csum forwarding engine.\n\n"
> 
> -			"gro set (max_flow_num)
> (max_item_num_per_flow) (port_id)\n"
> -			"    Set max flow number and max packet number
> per-flow"
> -			" for GRO.\n\n"
> +			"show port (port_id) gro\n"
> +			"    Display GRO configuration.\n\n"
> +
> +			"set gro flush (cycles)\n"
> +			"    Set the cycle to flush GROed packets from"
> +			" reassembly tables.\n\n"
> 
>  			"set fwd (%s)\n"
>  			"    Set packet forwarding mode.\n\n"
> @@ -3854,115 +3857,145 @@ cmdline_parse_inst_t cmd_tunnel_tso_show
> = {
>  };
> 
>  /* *** SET GRO FOR A PORT *** */
> -struct cmd_gro_result {
> +struct cmd_gro_enable_result {
> +	cmdline_fixed_string_t cmd_set;
> +	cmdline_fixed_string_t cmd_port;
>  	cmdline_fixed_string_t cmd_keyword;
> -	cmdline_fixed_string_t mode;
> -	uint8_t port_id;
> +	cmdline_fixed_string_t cmd_onoff;
> +	uint8_t cmd_pid;
>  };
> 
>  static void
> -cmd_enable_gro_parsed(void *parsed_result,
> +cmd_gro_enable_parsed(void *parsed_result,
>  		__attribute__((unused)) struct cmdline *cl,
>  		__attribute__((unused)) void *data)
>  {
> -	struct cmd_gro_result *res;
> +	struct cmd_gro_enable_result *res;
> 
>  	res = parsed_result;
> -	setup_gro(res->mode, res->port_id);
> -}
> -
> -cmdline_parse_token_string_t cmd_gro_keyword =
> -	TOKEN_STRING_INITIALIZER(struct cmd_gro_result,
> +	if (!strcmp(res->cmd_keyword, "gro"))
> +		setup_gro(res->cmd_onoff, res->cmd_pid);
> +}
> +
> +cmdline_parse_token_string_t cmd_gro_enable_set =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_enable_result,
> +			cmd_set, "set");
> +cmdline_parse_token_string_t cmd_gro_enable_port =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_enable_result,
> +			cmd_keyword, "port");
> +cmdline_parse_token_num_t cmd_gro_enable_pid =
> +	TOKEN_NUM_INITIALIZER(struct cmd_gro_enable_result,
> +			cmd_pid, UINT8);
> +cmdline_parse_token_string_t cmd_gro_enable_keyword =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_enable_result,
>  			cmd_keyword, "gro");
> -cmdline_parse_token_string_t cmd_gro_mode =
> -	TOKEN_STRING_INITIALIZER(struct cmd_gro_result,
> -			mode, "on#off");
> -cmdline_parse_token_num_t cmd_gro_pid =
> -	TOKEN_NUM_INITIALIZER(struct cmd_gro_result,
> -			port_id, UINT8);
> +cmdline_parse_token_string_t cmd_gro_enable_onoff =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_enable_result,
> +			cmd_onoff, "on#off");
> 
> -cmdline_parse_inst_t cmd_enable_gro = {
> -	.f = cmd_enable_gro_parsed,
> +cmdline_parse_inst_t cmd_gro_enable = {
> +	.f = cmd_gro_enable_parsed,
>  	.data = NULL,
> -	.help_str = "gro (on|off) (port_id)",
> +	.help_str = "set port <port_id> gro on|off",
>  	.tokens = {
> -		(void *)&cmd_gro_keyword,
> -		(void *)&cmd_gro_mode,
> -		(void *)&cmd_gro_pid,
> +		(void *)&cmd_gro_enable_set,
> +		(void *)&cmd_gro_enable_port,
> +		(void *)&cmd_gro_enable_pid,
> +		(void *)&cmd_gro_enable_keyword,
> +		(void *)&cmd_gro_enable_onoff,
>  		NULL,
>  	},
>  };
> 
> -/* *** SET MAX FLOW NUMBER AND ITEM NUM PER FLOW FOR GRO *** */
> -struct cmd_gro_set_result {
> -	cmdline_fixed_string_t gro;
> -	cmdline_fixed_string_t mode;
> -	uint16_t flow_num;
> -	uint16_t item_num_per_flow;
> -	uint8_t port_id;
> +/* *** DISPLAY GRO CONFIGURATION *** */
> +struct cmd_gro_show_result {
> +	cmdline_fixed_string_t cmd_show;
> +	cmdline_fixed_string_t cmd_port;
> +	cmdline_fixed_string_t cmd_keyword;
> +	uint8_t cmd_pid;
>  };
> 
>  static void
> -cmd_gro_set_parsed(void *parsed_result,
> -		       __attribute__((unused)) struct cmdline *cl,
> -		       __attribute__((unused)) void *data)
> +cmd_gro_show_parsed(void *parsed_result,
> +		__attribute__((unused)) struct cmdline *cl,
> +		__attribute__((unused)) void *data)
>  {
> -	struct cmd_gro_set_result *res = parsed_result;
> +	struct cmd_gro_show_result *res;
> 
> -	if (port_id_is_invalid(res->port_id, ENABLED_WARN))
> -		return;
> -	if (test_done == 0) {
> -		printf("Before set GRO flow_num and item_num_per_flow,"
> -				" please stop forwarding first\n");
> -		return;
> -	}
> +	res = parsed_result;
> +	if (!strcmp(res->cmd_keyword, "gro"))
> +		show_gro(res->cmd_pid);
> +}
> +
> +cmdline_parse_token_string_t cmd_gro_show_show =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_show_result,
> +			cmd_show, "show");
> +cmdline_parse_token_string_t cmd_gro_show_port =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_show_result,
> +			cmd_port, "port");
> +cmdline_parse_token_num_t cmd_gro_show_pid =
> +	TOKEN_NUM_INITIALIZER(struct cmd_gro_show_result,
> +			cmd_pid, UINT8);
> +cmdline_parse_token_string_t cmd_gro_show_keyword =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_show_result,
> +			cmd_keyword, "gro");
> 
> -	if (!strcmp(res->mode, "set")) {
> -		if (res->flow_num == 0)
> -			printf("Invalid flow number. Revert to default value:"
> -					" %u.\n",
> GRO_DEFAULT_FLOW_NUM);
> -		else
> -			gro_ports[res->port_id].param.max_flow_num =
> -				res->flow_num;
> +cmdline_parse_inst_t cmd_gro_show = {
> +	.f = cmd_gro_show_parsed,
> +	.data = NULL,
> +	.help_str = "show port <port_id> gro",
> +	.tokens = {
> +		(void *)&cmd_gro_show_show,
> +		(void *)&cmd_gro_show_port,
> +		(void *)&cmd_gro_show_pid,
> +		(void *)&cmd_gro_show_keyword,
> +		NULL,
> +	},
> +};
> 
> -		if (res->item_num_per_flow == 0)
> -			printf("Invalid item number per-flow. Revert"
> -					" to default value:%u.\n",
> -
> 	GRO_DEFAULT_ITEM_NUM_PER_FLOW);
> -		else
> -			gro_ports[res->port_id].param.max_item_per_flow
> =
> -				res->item_num_per_flow;
> -	}
> +/* *** SET FLUSH CYCLES FOR GRO *** */
> +struct cmd_gro_flush_result {
> +	cmdline_fixed_string_t cmd_set;
> +	cmdline_fixed_string_t cmd_keyword;
> +	cmdline_fixed_string_t cmd_flush;
> +	uint8_t cmd_cycles;
> +};
> +
> +static void
> +cmd_gro_flush_parsed(void *parsed_result,
> +		__attribute__((unused)) struct cmdline *cl,
> +		__attribute__((unused)) void *data)
> +{
> +	struct cmd_gro_flush_result *res;
> +
> +	res = parsed_result;
> +	if ((!strcmp(res->cmd_keyword, "gro")) &&
> +			(!strcmp(res->cmd_flush, "flush")))
> +		setup_gro_flush_cycles(res->cmd_cycles);
>  }
> 
> -cmdline_parse_token_string_t cmd_gro_set_gro =
> -	TOKEN_STRING_INITIALIZER(struct cmd_gro_set_result,
> -				gro, "gro");
> -cmdline_parse_token_string_t cmd_gro_set_mode =
> -	TOKEN_STRING_INITIALIZER(struct cmd_gro_set_result,
> -				mode, "set");
> -cmdline_parse_token_num_t cmd_gro_set_flow_num =
> -	TOKEN_NUM_INITIALIZER(struct cmd_gro_set_result,
> -				flow_num, UINT16);
> -cmdline_parse_token_num_t cmd_gro_set_item_num_per_flow =
> -	TOKEN_NUM_INITIALIZER(struct cmd_gro_set_result,
> -				item_num_per_flow, UINT16);
> -cmdline_parse_token_num_t cmd_gro_set_portid =
> -	TOKEN_NUM_INITIALIZER(struct cmd_gro_set_result,
> -				port_id, UINT8);
> +cmdline_parse_token_string_t cmd_gro_flush_set =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_flush_result,
> +			cmd_set, "set");
> +cmdline_parse_token_string_t cmd_gro_flush_keyword =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_flush_result,
> +			cmd_keyword, "gro");
> +cmdline_parse_token_string_t cmd_gro_flush_flush =
> +	TOKEN_STRING_INITIALIZER(struct cmd_gro_flush_result,
> +			cmd_flush, "flush");
> +cmdline_parse_token_num_t cmd_gro_flush_cycles =
> +	TOKEN_NUM_INITIALIZER(struct cmd_gro_flush_result,
> +			cmd_cycles, UINT8);
> 
> -cmdline_parse_inst_t cmd_gro_set = {
> -	.f = cmd_gro_set_parsed,
> +cmdline_parse_inst_t cmd_gro_flush = {
> +	.f = cmd_gro_flush_parsed,
>  	.data = NULL,
> -	.help_str = "gro set <max_flow_num> <max_item_num_per_flow>
> "
> -		"<port_id>: set max flow number and max packet number
> per-flow "
> -		"for GRO",
> +	.help_str = "set gro flush <cycles>",
>  	.tokens = {
> -		(void *)&cmd_gro_set_gro,
> -		(void *)&cmd_gro_set_mode,
> -		(void *)&cmd_gro_set_flow_num,
> -		(void *)&cmd_gro_set_item_num_per_flow,
> -		(void *)&cmd_gro_set_portid,
> +		(void *)&cmd_gro_flush_set,
> +		(void *)&cmd_gro_flush_keyword,
> +		(void *)&cmd_gro_flush_flush,
> +		(void *)&cmd_gro_flush_cycles,
>  		NULL,
>  	},
>  };
> @@ -14253,8 +14286,9 @@ cmdline_parse_ctx_t main_ctx[] = {
>  	(cmdline_parse_inst_t *)&cmd_tso_show,
>  	(cmdline_parse_inst_t *)&cmd_tunnel_tso_set,
>  	(cmdline_parse_inst_t *)&cmd_tunnel_tso_show,
> -	(cmdline_parse_inst_t *)&cmd_enable_gro,
> -	(cmdline_parse_inst_t *)&cmd_gro_set,
> +	(cmdline_parse_inst_t *)&cmd_gro_enable,
> +	(cmdline_parse_inst_t *)&cmd_gro_flush,
> +	(cmdline_parse_inst_t *)&cmd_gro_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/config.c b/app/test-pmd/config.c
> index 3ae3e1c..92220b1 100644
> --- a/app/test-pmd/config.c
> +++ b/app/test-pmd/config.c
> @@ -2420,7 +2420,7 @@ set_tx_pkt_segments(unsigned *seg_lengths,
> unsigned nb_segs)
>  }
> 
>  void
> -setup_gro(const char *mode, uint8_t port_id)
> +setup_gro(const char *onoff, uint8_t port_id)
>  {
>  	if (!rte_eth_dev_is_valid_port(port_id)) {
>  		printf("invalid port id %u\n", port_id);
> @@ -2431,29 +2431,77 @@ setup_gro(const char *mode, uint8_t port_id)
>  				" please stop forwarding first\n");
>  		return;
>  	}
> -	if (strcmp(mode, "on") == 0) {
> -		if (gro_ports[port_id].enable) {
> -			printf("port %u has enabled GRO\n", port_id);
> +	if (strcmp(onoff, "on") == 0) {
> +		if (gro_ports[port_id].enable != 0) {
> +			printf("Port %u has enabled GRO. Please"
> +					" disable GRO first\n", port_id);
>  			return;
>  		}
> -		gro_ports[port_id].enable = 1;
> -		gro_ports[port_id].param.gro_types = RTE_GRO_TCP_IPV4;
> -
> -		if (gro_ports[port_id].param.max_flow_num == 0)
> +		if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
> +			gro_ports[port_id].param.gro_types =
> RTE_GRO_TCP_IPV4;
>  			gro_ports[port_id].param.max_flow_num =
>  				GRO_DEFAULT_FLOW_NUM;
> -		if (gro_ports[port_id].param.max_item_per_flow == 0)
>  			gro_ports[port_id].param.max_item_per_flow =
>  				GRO_DEFAULT_ITEM_NUM_PER_FLOW;
> +		}
> +		gro_ports[port_id].enable = 1;
>  	} else {
>  		if (gro_ports[port_id].enable == 0) {
> -			printf("port %u has disabled GRO\n", port_id);
> +			printf("Port %u has disabled GRO\n", port_id);
>  			return;
>  		}
>  		gro_ports[port_id].enable = 0;
>  	}
>  }
> 
> +void
> +setup_gro_flush_cycles(uint8_t cycles)
> +{
> +	if (test_done == 0) {
> +		printf("Before change flush interval for GRO,"
> +				" please stop forwarding first.\n");
> +		return;
> +	}
> +
> +	if (cycles > GRO_MAX_FLUSH_CYCLES || cycles <
> +			GRO_DEFAULT_FLUSH_CYCLES) {
> +		printf("The flushing cycle be in the range"
> +				" of 1 to %u. Revert to the default"
> +				" value %u.\n",
> +				GRO_MAX_FLUSH_CYCLES,
> +				GRO_DEFAULT_FLUSH_CYCLES);
> +		cycles = GRO_DEFAULT_FLUSH_CYCLES;
> +	}
> +
> +	gro_flush_cycles = cycles;
> +}
> +
> +void
> +show_gro(uint8_t port_id)
> +{
> +	struct rte_gro_param *param;
> +	uint32_t max_pkts_num;
> +
> +	param = &gro_ports[port_id].param;
> +
> +	if (!rte_eth_dev_is_valid_port(port_id)) {
> +		printf("Invalid port id %u.\n", port_id);
> +		return;
> +	}
> +	if (gro_ports[port_id].enable) {
> +		printf("GRO type: TCP/IPv4\n");
> +		if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
> +			max_pkts_num = param->max_flow_num *
> +				param->max_item_per_flow;
> +		} else
> +			max_pkts_num = MAX_PKT_BURST *
> GRO_MAX_FLUSH_CYCLES;
> +		printf("Max number of packets to perform GRO: %u\n",
> +				max_pkts_num);
> +		printf("Flushing cycles: %u\n", gro_flush_cycles);
> +	} else
> +		printf("Port %u doesn't enable GRO.\n", port_id);
> +}
> +
>  char*
>  list_pkt_forwarding_modes(void)
>  {
> diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c
> index 90c8119..ca50ab7 100644
> --- a/app/test-pmd/csumonly.c
> +++ b/app/test-pmd/csumonly.c
> @@ -631,6 +631,9 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
>  	struct rte_mbuf *m, *p;
>  	struct ether_hdr *eth_hdr;
>  	void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
> +	void **gro_ctx;
> +	uint16_t gro_pkts_num;
> +	uint8_t gro_enable;
>  	uint16_t nb_rx;
>  	uint16_t nb_tx;
>  	uint16_t nb_prep;
> @@ -657,17 +660,13 @@ pkt_burst_checksum_forward(struct fwd_stream
> *fs)
>  				 nb_pkt_per_burst);
>  	if (unlikely(nb_rx == 0))
>  		return;
> -	if (unlikely(gro_ports[fs->rx_port].enable))
> -		nb_rx = rte_gro_reassemble_burst(pkts_burst,
> -				nb_rx,
> -				&(gro_ports[fs->rx_port].param));
> -
>  #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
>  	fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
>  #endif
>  	fs->rx_packets += nb_rx;
>  	rx_bad_ip_csum = 0;
>  	rx_bad_l4_csum = 0;
> +	gro_enable = gro_ports[fs->rx_port].enable;
> 
>  	txp = &ports[fs->tx_port];
>  	testpmd_ol_flags = txp->tx_ol_flags;
> @@ -851,6 +850,28 @@ pkt_burst_checksum_forward(struct fwd_stream
> *fs)
>  		}
>  	}
> 
> +	if (unlikely(gro_enable)) {
> +		if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
> +			nb_rx = rte_gro_reassemble_burst(pkts_burst,
> nb_rx,
> +					&(gro_ports[fs->rx_port].param));
> +		} else {
> +			gro_ctx = current_fwd_lcore()->gro_ctx;
> +			nb_rx = rte_gro_reassemble(pkts_burst, nb_rx,
> gro_ctx);
> +
> +			if (++fs->gro_times >= gro_flush_cycles) {
> +				gro_pkts_num =
> rte_gro_get_pkt_count(gro_ctx);
> +				if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
> +					gro_pkts_num = MAX_PKT_BURST -
> nb_rx;
> +
> +				nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
> +						RTE_GRO_TCP_IPV4,
> +						&pkts_burst[nb_rx],
> +						gro_pkts_num);
> +				fs->gro_times = 0;
> +			}
> +		}
> +	}
> +
>  	nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
>  			pkts_burst, nb_rx);
>  	if (nb_prep != nb_rx)
> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> index e097ee0..c9d988e 100644
> --- a/app/test-pmd/testpmd.c
> +++ b/app/test-pmd/testpmd.c
> @@ -90,7 +90,6 @@
>  #ifdef RTE_LIBRTE_LATENCY_STATS
>  #include <rte_latencystats.h>
>  #endif
> -#include <rte_gro.h>
> 
>  #include "testpmd.h"
> 
> @@ -386,6 +385,7 @@ uint8_t bitrate_enabled;
>  #endif
> 
>  struct gro_status gro_ports[RTE_MAX_ETHPORTS];
> +uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
> 
>  /* Forward function declarations */
>  static void map_port_queue_stats_mapping_registers(uint8_t pi, struct
> rte_port *port);
> @@ -570,6 +570,7 @@ init_config(void)
>  	unsigned int nb_mbuf_per_pool;
>  	lcoreid_t  lc_id;
>  	uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
> +	struct rte_gro_param gro_param;
> 
>  	memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
> 
> @@ -671,6 +672,20 @@ init_config(void)
>  		rte_exit(EXIT_FAILURE, "FAIL from init_fwd_streams()\n");
> 
>  	fwd_config_setup();
> +
> +	/* create a gro context for each lcore */
> +	gro_param.gro_types = RTE_GRO_TCP_IPV4;
> +	gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
> +	gro_param.max_item_per_flow = MAX_PKT_BURST;
> +	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
> +		gro_param.socket_id = rte_lcore_to_socket_id(
> +				fwd_lcores_cpuids[lc_id]);
> +		fwd_lcores[lc_id]->gro_ctx =
> rte_gro_ctx_create(&gro_param);
> +		if (fwd_lcores[lc_id]->gro_ctx == NULL) {
> +			rte_exit(EXIT_FAILURE,
> +					"rte_gro_ctx_create() failed\n");
> +		}
> +	}
>  }
> 
> 
> @@ -1217,6 +1232,7 @@ stop_packet_forwarding(void)
>  #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
>  	uint64_t fwd_cycles;
>  #endif
> +
>  	static const char *acc_stats_border = "+++++++++++++++";
> 
>  	if (test_done) {
> @@ -1307,6 +1323,7 @@ stop_packet_forwarding(void)
> 
>  		fwd_port_stats_display(pt_id, &stats);
>  	}
> +
>  	printf("\n  %s Accumulated forward statistics for all ports"
>  	       "%s\n",
>  	       acc_stats_border, acc_stats_border);
> diff --git a/app/test-pmd/testpmd.h b/app/test-pmd/testpmd.h
> index 1d1ee75..9433eae 100644
> --- a/app/test-pmd/testpmd.h
> +++ b/app/test-pmd/testpmd.h
> @@ -120,6 +120,7 @@ struct fwd_stream {
>  	unsigned int fwd_dropped; /**< received packets not forwarded */
>  	unsigned int rx_bad_ip_csum ; /**< received packets has bad ip
> checksum */
>  	unsigned int rx_bad_l4_csum ; /**< received packets has bad l4
> checksum */
> +	unsigned int gro_times;	/**< GRO operation times */
>  #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
>  	uint64_t     core_cycles; /**< used for RX and TX processing */
>  #endif
> @@ -206,6 +207,7 @@ struct rte_port {
>   */
>  struct fwd_lcore {
>  	struct rte_mempool *mbp; /**< The mbuf pool to use by this core */
> +	void *gro_ctx;		/**< GRO context */
>  	streamid_t stream_idx;   /**< index of 1st stream in "fwd_streams"
> */
>  	streamid_t stream_nb;    /**< number of streams in "fwd_streams"
> */
>  	lcoreid_t  cpuid_idx;    /**< index of logical core in CPU id table */
> @@ -434,13 +436,19 @@ extern struct ether_addr
> peer_eth_addrs[RTE_MAX_ETHPORTS];
>  extern uint32_t burst_tx_delay_time; /**< Burst tx delay time(us) for mac-
> retry. */
>  extern uint32_t burst_tx_retry_num;  /**< Burst tx retry number for mac-
> retry. */
> 
> -#define GRO_DEFAULT_FLOW_NUM 4
> -#define GRO_DEFAULT_ITEM_NUM_PER_FLOW DEF_PKT_BURST
> +#define GRO_DEFAULT_ITEM_NUM_PER_FLOW 32
> +#define GRO_DEFAULT_FLOW_NUM (RTE_GRO_MAX_BURST_ITEM_NUM
> / \
> +		GRO_DEFAULT_ITEM_NUM_PER_FLOW)
> +
> +#define GRO_DEFAULT_FLUSH_CYCLES 1
> +#define GRO_MAX_FLUSH_CYCLES 4
> +
>  struct gro_status {
>  	struct rte_gro_param param;
>  	uint8_t enable;
>  };
>  extern struct gro_status gro_ports[RTE_MAX_ETHPORTS];
> +extern uint8_t gro_flush_cycles;
> 
>  static inline unsigned int
>  lcore_num(void)
> @@ -641,7 +649,9 @@ void get_2tuple_filter(uint8_t port_id, uint16_t
> index);
>  void get_5tuple_filter(uint8_t port_id, uint16_t index);
>  int rx_queue_id_is_invalid(queueid_t rxq_id);
>  int tx_queue_id_is_invalid(queueid_t txq_id);
> -void setup_gro(const char *mode, uint8_t port_id);
> +void setup_gro(const char *onoff, uint8_t port_id);
> +void setup_gro_flush_cycles(uint8_t cycles);
> +void show_gro(uint8_t port_id);
> 
>  /* Functions to manage the set of filtered Multicast MAC addresses */
>  void mcast_addr_add(uint8_t port_id, struct ether_addr *mc_addr);
> diff --git a/doc/guides/testpmd_app_ug/testpmd_funcs.rst
> b/doc/guides/testpmd_app_ug/testpmd_funcs.rst
> index 2ed62f5..74a1fb4 100644
> --- a/doc/guides/testpmd_app_ug/testpmd_funcs.rst
> +++ b/doc/guides/testpmd_app_ug/testpmd_funcs.rst
> @@ -898,12 +898,12 @@ Display the status of TCP Segmentation Offload::
> 
>     testpmd> tso show (port_id)
> 
> -gro
> -~~~
> +set port - gro
> +~~~~~~~~~~~~~~
> 
>  Enable or disable GRO in ``csum`` forwarding engine::
> 
> -   testpmd> gro (on|off) (port_id)
> +   testpmd> set port <port_id> gro on|off
> 
>  If enabled, the csum forwarding engine will perform GRO on the TCP/IPv4
>  packets received from the given port.
> @@ -914,23 +914,43 @@ GRO. By default, GRO is disabled for all ports.
>  .. note::
> 
>     When enable GRO for a port, TCP/IPv4 packets received from the port
> -   will be performed GRO. After GRO, the merged packets are multi-
> segments.
> -   But csum forwarding engine doesn't support to calculate TCP checksum
> -   for multi-segment packets in SW. So please select TCP HW checksum
> -   calculation for the port which GROed packets are transmitted to.
> +   will be performed GRO. After GRO, all merged packets have bad
> +   checksums, since the GRO library doesn't re-calculate checksums for
> +   the merged packets. Therefore, if users want the merged packets to
> +   have correct checksums, please select HW IP checksum calculation and
> +   HW TCP checksum calculation for the port which the merged packets are
> +   transmitted to.
> +
> +show port - gro
> +~~~~~~~~~~~~~~~
> 
> -gro set
> -~~~~~~~
> +Display GRO configuration for a given port::
> +
> +   testpmd> show port <port_id> gro
> +
> +set gro flush
> +~~~~~~~~~~~~~
> +
> +Set the cycle to flush the GROed packets from reassembly tables::
> +
> +   testpmd> set gro flush <cycles>
> 
> -Set max flow number and max packet number per-flow for GRO::
> +When enable GRO, the csum forwarding engine performs GRO on received
> +packets, and the GROed packets are stored in reassembly tables. Users
> +can use this command to determine when the GROed packets are flushed
> +from the reassembly tables.
> 
> -   testpmd> gro set (max_flow_num) (max_item_num_per_flow) (port_id)
> +The ``cycles`` is measured in GRO operation times. The csum forwarding
> +engine flushes the GROed packets from the tables every ``cycles`` GRO
> +operations.
> 
> -The product of ``max_flow_num`` and ``max_item_num_per_flow`` is the
> max
> -number of packets a GRO table can store.
> +By default, the value of ``cycles`` is 1, which means flush GROed packets
> +from the reassembly tables as soon as one GRO operation finishes. The
> value
> +of ``cycles`` should be in the range of 1 to ``GRO_MAX_FLUSH_CYCLES``.
> 
> -If current packet number is greater than or equal to the max value, GRO
> -will stop processing incoming packets.
> +Please note that the large value of ``cycles`` may cause the poor TCP/IP
> +stack performance. Because the GROed packets are delayed to arrive the
> +stack, thus causing more duplicated ACKs and TCP retransmissions.
> 
>  mac_addr add
>  ~~~~~~~~~~~~
> --
> 2.7.4

  reply	other threads:[~2017-09-27  7:23 UTC|newest]

Thread overview: 27+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-08-10  2:50 [dpdk-dev] [PATCH 0/2] Support the Heavyweight Mode GRO in Testpmd Jiayu Hu
2017-08-10  2:50 ` [dpdk-dev] [PATCH 1/2] app/testpmd: support the heavywight mode GRO Jiayu Hu
2017-08-10  9:50   ` Ferruh Yigit
2017-08-15  6:01     ` Jiayu Hu
2017-08-10  2:50 ` [dpdk-dev] [PATCH 2/2] doc: update testpmd user guide for the heavyweight " Jiayu Hu
2017-08-17  9:08 ` [dpdk-dev] [PATCH v2 0/2] Support the Heavyweight Mode GRO in Testpmd Jiayu Hu
2017-08-17  9:08   ` [dpdk-dev] [PATCH v2 1/2] app/testpmd: support the heavywight mode GRO Jiayu Hu
2017-08-21 11:03     ` Ferruh Yigit
2017-08-22  1:00       ` Hu, Jiayu
     [not found]         ` <671112a6-6ac6-7476-4270-be1a0258f06e@intel.com>
2017-08-22 14:27           ` Jiayu Hu
2017-08-21 11:16     ` Ferruh Yigit
2017-08-22  0:53       ` Hu, Jiayu
2017-08-17  9:08   ` [dpdk-dev] [PATCH v2 2/2] doc: update testpmd user guide for the heavyweight " Jiayu Hu
2017-08-21 11:03     ` Ferruh Yigit
2017-08-22  0:52       ` Hu, Jiayu
2017-09-03  6:30   ` [dpdk-dev] [PATCH v3] app/testpmd: enable the heavyweight mode TCP/IPv4 GRO Jiayu Hu
2017-09-20  7:00     ` Yao, Lei A
2017-09-25 10:20       ` Hu, Jiayu
2017-09-26  3:54         ` Hu, Jiayu
2017-09-25 11:11     ` Ferruh Yigit
2017-09-26  6:26     ` [dpdk-dev] [PATCH v4] " Jiayu Hu
2017-09-27  7:23       ` Yao, Lei A [this message]
2017-10-06 23:04         ` Ferruh Yigit
2017-10-06 23:26           ` Ferruh Yigit
2017-10-07  0:51             ` Hu, Jiayu
2017-10-07  7:45       ` [dpdk-dev] [PATCH v5] " Jiayu Hu
2017-10-07  7:58         ` Ferruh Yigit

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=2DBBFF226F7CF64BAFCA79B681719D953A2965D1@shsmsx102.ccr.corp.intel.com \
    --to=lei.a.yao@intel.com \
    --cc=dev@dpdk.org \
    --cc=ferruh.yigit@intel.com \
    --cc=jianfeng.tan@intel.com \
    --cc=jiayu.hu@intel.com \
    --cc=jingjing.wu@intel.com \
    --cc=konstantin.ananyev@intel.com \
    --cc=thomas@monjalon.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).