From: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
To: dev@dpdk.org
Cc: nirranjan@chelsio.com
Subject: [dpdk-dev] [PATCH 09/12] net/cxgbe: add devarg to control Tx coalescing
Date: Sat, 7 Sep 2019 03:22:30 +0530 [thread overview]
Message-ID: <1c5340dcd919f94af9fd63a9f6ef951eb5544a1e.1567799552.git.rahul.lakkireddy@chelsio.com> (raw)
In-Reply-To: <cover.1567799552.git.rahul.lakkireddy@chelsio.com>
In-Reply-To: <cover.1567799552.git.rahul.lakkireddy@chelsio.com>
Remove compile time option to control Tx coalescing Latency vs
Throughput behavior. Add tx_mode_latency devarg instead, to
dynamically control Tx coalescing behavior.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
---
config/common_base | 1 -
doc/guides/nics/cxgbe.rst | 13 +++++++++----
drivers/net/cxgbe/base/adapter.h | 1 +
drivers/net/cxgbe/cxgbe.h | 1 +
drivers/net/cxgbe/cxgbe_ethdev.c | 3 ++-
drivers/net/cxgbe/cxgbe_main.c | 3 +++
drivers/net/cxgbe/cxgbevf_ethdev.c | 1 +
drivers/net/cxgbe/sge.c | 18 ++++++++----------
8 files changed, 25 insertions(+), 16 deletions(-)
diff --git a/config/common_base b/config/common_base
index 43964de6d..48b134521 100644
--- a/config/common_base
+++ b/config/common_base
@@ -217,7 +217,6 @@ CONFIG_RTE_LIBRTE_BNXT_PMD=y
# Compile burst-oriented Chelsio Terminator (CXGBE) PMD
#
CONFIG_RTE_LIBRTE_CXGBE_PMD=y
-CONFIG_RTE_LIBRTE_CXGBE_TPUT=y
# NXP DPAA Bus
CONFIG_RTE_LIBRTE_DPAA_BUS=n
diff --git a/doc/guides/nics/cxgbe.rst b/doc/guides/nics/cxgbe.rst
index 6e6767194..f15e0b90a 100644
--- a/doc/guides/nics/cxgbe.rst
+++ b/doc/guides/nics/cxgbe.rst
@@ -104,10 +104,6 @@ enabling debugging options may affect system performance.
This controls compilation of both CXGBE and CXGBEVF PMD.
-- ``CONFIG_RTE_LIBRTE_CXGBE_TPUT`` (default **y**)
-
- Toggle behavior to prefer Throughput or Latency.
-
Runtime Options
~~~~~~~~~~~~~~~
@@ -127,6 +123,15 @@ Common Runtime Options
enabled, the outer VLAN tag is preserved in Q-in-Q packets. Otherwise,
the outer VLAN tag is stripped in Q-in-Q packets.
+- ``tx_mode_latency`` (default **0**)
+
+ When set to 1, Tx doesn't wait for max number of packets to get
+ coalesced and sends the packets immediately at the end of the
+ current Tx burst. When set to 0, Tx waits across multiple Tx bursts
+ until the max number of packets have been coalesced. In this case,
+ Tx only sends the coalesced packets to hardware once the max
+ coalesce limit has been reached.
+
CXGBE VF Only Runtime Options
-----------------------------
diff --git a/drivers/net/cxgbe/base/adapter.h b/drivers/net/cxgbe/base/adapter.h
index 68bd5a9cc..6a931ce97 100644
--- a/drivers/net/cxgbe/base/adapter.h
+++ b/drivers/net/cxgbe/base/adapter.h
@@ -302,6 +302,7 @@ TAILQ_HEAD(mbox_list, mbox_entry);
struct adapter_devargs {
bool keep_ovlan;
bool force_link_up;
+ bool tx_mode_latency;
};
struct adapter {
diff --git a/drivers/net/cxgbe/cxgbe.h b/drivers/net/cxgbe/cxgbe.h
index 3a50502b7..ed1be3559 100644
--- a/drivers/net/cxgbe/cxgbe.h
+++ b/drivers/net/cxgbe/cxgbe.h
@@ -52,6 +52,7 @@
/* Common PF and VF devargs */
#define CXGBE_DEVARG_CMN_KEEP_OVLAN "keep_ovlan"
+#define CXGBE_DEVARG_CMN_TX_MODE_LATENCY "tx_mode_latency"
/* VF only devargs */
#define CXGBE_DEVARG_VF_FORCE_LINK_UP "force_link_up"
diff --git a/drivers/net/cxgbe/cxgbe_ethdev.c b/drivers/net/cxgbe/cxgbe_ethdev.c
index 8a2c2ca11..e99e5d057 100644
--- a/drivers/net/cxgbe/cxgbe_ethdev.c
+++ b/drivers/net/cxgbe/cxgbe_ethdev.c
@@ -1238,7 +1238,8 @@ RTE_PMD_REGISTER_PCI(net_cxgbe, rte_cxgbe_pmd);
RTE_PMD_REGISTER_PCI_TABLE(net_cxgbe, cxgb4_pci_tbl);
RTE_PMD_REGISTER_KMOD_DEP(net_cxgbe, "* igb_uio | uio_pci_generic | vfio-pci");
RTE_PMD_REGISTER_PARAM_STRING(net_cxgbe,
- CXGBE_DEVARG_CMN_KEEP_OVLAN "=<0|1> ");
+ CXGBE_DEVARG_CMN_KEEP_OVLAN "=<0|1> "
+ CXGBE_DEVARG_CMN_TX_MODE_LATENCY "=<0|1> ");
RTE_INIT(cxgbe_init_log)
{
diff --git a/drivers/net/cxgbe/cxgbe_main.c b/drivers/net/cxgbe/cxgbe_main.c
index 6a6137f06..23b74c754 100644
--- a/drivers/net/cxgbe/cxgbe_main.c
+++ b/drivers/net/cxgbe/cxgbe_main.c
@@ -672,6 +672,7 @@ void cxgbe_print_port_info(struct adapter *adap)
static int check_devargs_handler(const char *key, const char *value, void *p)
{
if (!strncmp(key, CXGBE_DEVARG_CMN_KEEP_OVLAN, strlen(key)) ||
+ !strncmp(key, CXGBE_DEVARG_CMN_TX_MODE_LATENCY, strlen(key)) ||
!strncmp(key, CXGBE_DEVARG_VF_FORCE_LINK_UP, strlen(key))) {
if (!strncmp(value, "1", 1)) {
bool *dst_val = (bool *)p;
@@ -728,6 +729,8 @@ void cxgbe_process_devargs(struct adapter *adap)
{
cxgbe_get_devargs_int(adap, &adap->devargs.keep_ovlan,
CXGBE_DEVARG_CMN_KEEP_OVLAN, 0);
+ cxgbe_get_devargs_int(adap, &adap->devargs.tx_mode_latency,
+ CXGBE_DEVARG_CMN_TX_MODE_LATENCY, 0);
cxgbe_get_devargs_int(adap, &adap->devargs.force_link_up,
CXGBE_DEVARG_VF_FORCE_LINK_UP, 0);
}
diff --git a/drivers/net/cxgbe/cxgbevf_ethdev.c b/drivers/net/cxgbe/cxgbevf_ethdev.c
index cc0938b43..4165ba986 100644
--- a/drivers/net/cxgbe/cxgbevf_ethdev.c
+++ b/drivers/net/cxgbe/cxgbevf_ethdev.c
@@ -213,4 +213,5 @@ RTE_PMD_REGISTER_PCI_TABLE(net_cxgbevf, cxgb4vf_pci_tbl);
RTE_PMD_REGISTER_KMOD_DEP(net_cxgbevf, "* igb_uio | vfio-pci");
RTE_PMD_REGISTER_PARAM_STRING(net_cxgbevf,
CXGBE_DEVARG_CMN_KEEP_OVLAN "=<0|1> "
+ CXGBE_DEVARG_CMN_TX_MODE_LATENCY "=<0|1> "
CXGBE_DEVARG_VF_FORCE_LINK_UP "=<0|1> ");
diff --git a/drivers/net/cxgbe/sge.c b/drivers/net/cxgbe/sge.c
index bf3190211..0df870a41 100644
--- a/drivers/net/cxgbe/sge.c
+++ b/drivers/net/cxgbe/sge.c
@@ -1007,10 +1007,6 @@ static inline int tx_do_packet_coalesce(struct sge_eth_txq *txq,
unsigned int max_coal_pkt_num = is_pf4(adap) ? ETH_COALESCE_PKT_NUM :
ETH_COALESCE_VF_PKT_NUM;
-#ifdef RTE_LIBRTE_CXGBE_TPUT
- RTE_SET_USED(nb_pkts);
-#endif
-
if (q->coalesce.type == 0) {
mc = (struct ulp_txpkt *)q->coalesce.ptr;
mc->cmd_dest = htonl(V_ULPTX_CMD(4) | V_ULP_TXPKT_DEST(0) |
@@ -1082,13 +1078,15 @@ static inline int tx_do_packet_coalesce(struct sge_eth_txq *txq,
sd->coalesce.sgl[idx & 1] = (struct ulptx_sgl *)(cpl + 1);
sd->coalesce.idx = (idx & 1) + 1;
- /* send the coaelsced work request if max reached */
- if (++q->coalesce.idx == max_coal_pkt_num
-#ifndef RTE_LIBRTE_CXGBE_TPUT
- || q->coalesce.idx >= nb_pkts
-#endif
- )
+ /* Send the coalesced work request, only if max reached. However,
+ * if lower latency is preferred over throughput, then don't wait
+ * for coalescing the next Tx burst and send the packets now.
+ */
+ q->coalesce.idx++;
+ if (q->coalesce.idx == max_coal_pkt_num ||
+ (adap->devargs.tx_mode_latency && q->coalesce.idx >= nb_pkts))
ship_tx_pkt_coalesce_wr(adap, txq);
+
return 0;
}
--
2.18.0
next prev parent reply other threads:[~2019-09-06 22:01 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-06 21:52 [dpdk-dev] [PATCH 00/12] net/cxgbe: bug fixes and updates for CXGBE/CXGBEVF PMD Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 01/12] net/cxgbe: add cxgbe_ prefix to global functions Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 02/12] net/cxgbe: fix NULL access when allocating CLIP entry Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 03/12] net/cxgbe: fix slot allocation for IPv6 flows Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 04/12] net/cxgbe: fix parsing VLAN ID rewrite action Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 05/12] net/cxgbe: fix prefetch for non-coalesced Tx packets Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 06/12] net/cxgbe: avoid polling link status before device start Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 07/12] net/cxgbe: use dynamic logging for debug prints Rahul Lakkireddy
2019-09-27 14:37 ` Ferruh Yigit
2019-09-27 19:55 ` Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 08/12] net/cxgbe: separate VF only devargs Rahul Lakkireddy
2019-09-06 21:52 ` Rahul Lakkireddy [this message]
2019-09-06 21:52 ` [dpdk-dev] [PATCH 10/12] net/cxgbe: fetch max Tx coalesce limit from firmware Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 11/12] net/cxgbe: add rte_flow support for matching VLAN Rahul Lakkireddy
2019-09-06 21:52 ` [dpdk-dev] [PATCH 12/12] net/cxgbe: add rte_flow support for setting VLAN PCP Rahul Lakkireddy
2019-09-27 14:41 ` [dpdk-dev] [PATCH 00/12] net/cxgbe: bug fixes and updates for CXGBE/CXGBEVF PMD Ferruh Yigit
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 " Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 01/12] net/cxgbe: add cxgbe_ prefix to global functions Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 02/12] net/cxgbe: fix NULL access when allocating CLIP entry Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 03/12] net/cxgbe: fix slot allocation for IPv6 flows Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 04/12] net/cxgbe: fix parsing VLAN ID rewrite action Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 05/12] net/cxgbe: fix prefetch for non-coalesced Tx packets Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 06/12] net/cxgbe: avoid polling link status before device start Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 07/12] net/cxgbe: use dynamic logging for debug prints Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 08/12] net/cxgbe: separate VF only devargs Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 09/12] net/cxgbe: add devarg to control Tx coalescing Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 10/12] net/cxgbe: fetch max Tx coalesce limit from firmware Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 11/12] net/cxgbe: add rte_flow support for matching VLAN Rahul Lakkireddy
2019-09-27 20:30 ` [dpdk-dev] [PATCH v2 12/12] net/cxgbe: add rte_flow support for setting VLAN PCP Rahul Lakkireddy
2019-09-30 12:34 ` [dpdk-dev] [PATCH v2 00/12] net/cxgbe: bug fixes and updates for CXGBE/CXGBEVF PMD 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=1c5340dcd919f94af9fd63a9f6ef951eb5544a1e.1567799552.git.rahul.lakkireddy@chelsio.com \
--to=rahul.lakkireddy@chelsio.com \
--cc=dev@dpdk.org \
--cc=nirranjan@chelsio.com \
/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).