All of lore.kernel.org
 help / color / mirror / Atom feed
From: Tal Gilboa <talgi@mellanox.com>
To: "David S. Miller" <davem@davemloft.net>
Cc: netdev@vger.kernel.org, Tariq Toukan <tariqt@mellanox.com>,
	Tal Gilboa <talgi@mellanox.com>,
	Andy Gospodarek <andrew.gospodarek@broadcom.com>,
	Saeed Mahameed <saeedm@mellanox.com>
Subject: [PATCH net-next 4/4] net/mlx5e: Enable adaptive-TX moderation
Date: Fri, 30 Mar 2018 09:37:33 +0300	[thread overview]
Message-ID: <1522391853-37298-5-git-send-email-talgi@mellanox.com> (raw)
In-Reply-To: <1522391853-37298-1-git-send-email-talgi@mellanox.com>

Add support for adaptive TX moderation. This greatly reduces TX interrupt
rate and increases bandwidth, mostly for TCP bandwidth over ARM
architecture (below). There is a slight single stream TCP with very large
message sizes degradation (x86). In this case if there's any moderation on
transmitted packets the bandwidth would reduce due to hitting TCP output
limit. Since this is a synthetic case, this is still worth doing.

Performance improvement (ConnectX-4Lx 40GbE, ARM)
TCP 64B bandwidth with 1-50 streams increased 6-35%.
TCP 64B bandwidth with 100-500 streams increased 20-70%.

Performance improvement (ConnectX-5 100GbE, x86)
Bandwidth: increased up to 40% (1024B with 10s of streams).
Interrupt rate: reduced up to 50% (1024B with 1000s of streams).

Signed-off-by: Tal Gilboa <talgi@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h       |  4 +++
 drivers/net/ethernet/mellanox/mlx5/core/en_dim.c   | 24 +++++++++++---
 .../net/ethernet/mellanox/mlx5/core/en_ethtool.c   | 37 ++++++++++++++--------
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c  | 22 +++++++++++++
 drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c  | 37 ++++++++++++++++------
 5 files changed, 96 insertions(+), 28 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 3e24864..e18574d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -338,6 +338,7 @@ enum {
 	MLX5E_SQ_STATE_IPSEC,
 	MLX5E_SQ_STATE_TLS,
 	MLX5E_SQ_STATE_RECOVERING,
+	MLX5E_SQ_STATE_AM,
 };
 
 struct mlx5e_sq_wqe_info {
@@ -350,6 +351,7 @@ struct mlx5e_txqsq {
 	/* dirtied @completion */
 	u16                        cc;
 	u32                        dma_fifo_cc;
+	struct net_dim             dim; /* Adaptive Moderation */
 
 	/* dirtied @xmit */
 	u16                        pc ____cacheline_aligned_in_smp;
@@ -387,6 +389,7 @@ struct mlx5e_txqsq {
 		struct work_struct         recover_work;
 		u64                        last_recover;
 	} recover;
+
 } ____cacheline_aligned_in_smp;
 
 struct mlx5e_xdpsq {
@@ -1136,4 +1139,5 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
 			    u16 max_channels);
 u8 mlx5e_params_calculate_tx_min_inline(struct mlx5_core_dev *mdev);
 void mlx5e_rx_dim_work(struct work_struct *work);
+void mlx5e_tx_dim_work(struct work_struct *work);
 #endif /* __MLX5_EN_H__ */
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
index 1b286e1..9cec351 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
@@ -33,16 +33,30 @@
 #include <linux/net_dim.h>
 #include "en.h"
 
+static inline void
+mlx5e_complete_dim_work(struct net_dim *dim, struct net_dim_cq_moder moder,
+			struct mlx5_core_dev *mdev, struct mlx5_core_cq *mcq)
+{
+	mlx5_core_modify_cq_moderation(mdev, mcq, moder.usec, moder.pkts);
+	dim->state = NET_DIM_START_MEASURE;
+}
+
 void mlx5e_rx_dim_work(struct work_struct *work)
 {
-	struct net_dim *dim = container_of(work, struct net_dim,
-					   work);
+	struct net_dim *dim = container_of(work, struct net_dim, work);
 	struct mlx5e_rq *rq = container_of(dim, struct mlx5e_rq, dim);
 	struct net_dim_cq_moder cur_moder =
 		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
 
-	mlx5_core_modify_cq_moderation(rq->mdev, &rq->cq.mcq,
-				       cur_moder.usec, cur_moder.pkts);
+	mlx5e_complete_dim_work(dim, cur_moder, rq->mdev, &rq->cq.mcq);
+}
 
-	dim->state = NET_DIM_START_MEASURE;
+void mlx5e_tx_dim_work(struct work_struct *work)
+{
+	struct net_dim *dim = container_of(work, struct net_dim, work);
+	struct mlx5e_txqsq *sq = container_of(dim, struct mlx5e_txqsq, dim);
+	struct net_dim_cq_moder cur_moder =
+		net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
+
+	mlx5e_complete_dim_work(dim, cur_moder, sq->cq.mdev, &sq->cq.mcq);
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
index 4b5d022..05faaf7 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
@@ -454,15 +454,20 @@ static int mlx5e_set_channels(struct net_device *dev,
 int mlx5e_ethtool_get_coalesce(struct mlx5e_priv *priv,
 			       struct ethtool_coalesce *coal)
 {
+	struct net_dim_cq_moder *rx_moder, *tx_moder;
+
 	if (!MLX5_CAP_GEN(priv->mdev, cq_moderation))
 		return -EOPNOTSUPP;
 
-	coal->rx_coalesce_usecs       = priv->channels.params.rx_cq_moderation.usec;
-	coal->rx_max_coalesced_frames = priv->channels.params.rx_cq_moderation.pkts;
-	coal->use_adaptive_rx_coalesce =
-		priv->channels.params.rx_cq_moderation.enabled;
-	coal->tx_coalesce_usecs       = priv->channels.params.tx_cq_moderation.usec;
-	coal->tx_max_coalesced_frames = priv->channels.params.tx_cq_moderation.pkts;
+	rx_moder = &priv->channels.params.rx_cq_moderation;
+	coal->rx_coalesce_usecs		= rx_moder->usec;
+	coal->rx_max_coalesced_frames	= rx_moder->pkts;
+	coal->use_adaptive_rx_coalesce	= rx_moder->enabled;
+
+	tx_moder = &priv->channels.params.tx_cq_moderation;
+	coal->tx_coalesce_usecs		= tx_moder->usec;
+	coal->tx_max_coalesced_frames	= tx_moder->pkts;
+	coal->use_adaptive_tx_coalesce	= tx_moder->enabled;
 
 	return 0;
 }
@@ -501,6 +506,7 @@ static int mlx5e_get_coalesce(struct net_device *netdev,
 int mlx5e_ethtool_set_coalesce(struct mlx5e_priv *priv,
 			       struct ethtool_coalesce *coal)
 {
+	struct net_dim_cq_moder *rx_moder, *tx_moder;
 	struct mlx5_core_dev *mdev = priv->mdev;
 	struct mlx5e_channels new_channels = {};
 	int err = 0;
@@ -512,12 +518,15 @@ int mlx5e_ethtool_set_coalesce(struct mlx5e_priv *priv,
 	mutex_lock(&priv->state_lock);
 	new_channels.params = priv->channels.params;
 
-	new_channels.params.tx_cq_moderation.usec = coal->tx_coalesce_usecs;
-	new_channels.params.tx_cq_moderation.pkts = coal->tx_max_coalesced_frames;
-	new_channels.params.rx_cq_moderation.usec = coal->rx_coalesce_usecs;
-	new_channels.params.rx_cq_moderation.pkts = coal->rx_max_coalesced_frames;
-	new_channels.params.rx_cq_moderation.enabled =
-		!!coal->use_adaptive_rx_coalesce;
+	rx_moder          = &new_channels.params.rx_cq_moderation;
+	rx_moder->usec    = coal->rx_coalesce_usecs;
+	rx_moder->pkts    = coal->rx_max_coalesced_frames;
+	rx_moder->enabled = !!coal->use_adaptive_rx_coalesce;
+
+	tx_moder          = &new_channels.params.tx_cq_moderation;
+	tx_moder->usec    = coal->tx_coalesce_usecs;
+	tx_moder->pkts    = coal->tx_max_coalesced_frames;
+	tx_moder->enabled = !!coal->use_adaptive_tx_coalesce;
 
 	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
 		priv->channels.params = new_channels.params;
@@ -525,8 +534,8 @@ int mlx5e_ethtool_set_coalesce(struct mlx5e_priv *priv,
 	}
 	/* we are opened */
 
-	reset = !!coal->use_adaptive_rx_coalesce !=
-			priv->channels.params.rx_cq_moderation.enabled;
+	reset = (!!rx_moder->enabled != priv->channels.params.rx_cq_moderation.enabled) ||
+		(!!tx_moder->enabled != priv->channels.params.tx_cq_moderation.enabled);
 
 	if (!reset) {
 		mlx5e_set_priv_channels_coalesce(priv, coal);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index e16a705..651f029 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -993,6 +993,9 @@ static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
 	if (err)
 		goto err_sq_wq_destroy;
 
+	INIT_WORK(&sq->dim.work, mlx5e_tx_dim_work);
+	sq->dim.mode = params->tx_cq_moderation.cq_period_mode;
+
 	sq->edge = (sq->wq.sz_m1 + 1) - MLX5_SEND_WQE_MAX_WQEBBS;
 
 	return 0;
@@ -1156,6 +1159,9 @@ static int mlx5e_open_txqsq(struct mlx5e_channel *c,
 	if (tx_rate)
 		mlx5e_set_sq_maxrate(c->netdev, sq, tx_rate);
 
+	if (params->tx_cq_moderation.enabled)
+		sq->state |= BIT(MLX5E_SQ_STATE_AM);
+
 	return 0;
 
 err_free_txqsq:
@@ -4065,6 +4071,21 @@ void mlx5e_set_tx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
 		params->tx_cq_moderation.usec =
 			MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC_FROM_CQE;
 
+	if (params->tx_cq_moderation.enabled) {
+		switch (cq_period_mode) {
+		case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
+			params->tx_cq_moderation =
+				net_dim_get_def_tx_moderation(
+					NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE);
+			break;
+		case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
+		default:
+			params->tx_cq_moderation =
+				net_dim_get_def_tx_moderation(
+					NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE);
+		}
+	}
+
 	MLX5E_SET_PFLAG(params, MLX5E_PFLAG_TX_CQE_BASED_MODER,
 			params->tx_cq_moderation.cq_period_mode ==
 				MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
@@ -4156,6 +4177,7 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
 			MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
 			MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
 	params->rx_cq_moderation.enabled = MLX5_CAP_GEN(mdev, cq_moderation);
+	params->tx_cq_moderation.enabled = params->rx_cq_moderation.enabled;
 	mlx5e_set_rx_cq_mode_params(params, cq_period_mode);
 	mlx5e_set_tx_cq_mode_params(params, cq_period_mode);
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index f292bb3..ff1d5fe 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
@@ -44,6 +44,30 @@ static inline bool mlx5e_channel_no_affinity_change(struct mlx5e_channel *c)
 	return cpumask_test_cpu(current_cpu, aff);
 }
 
+static inline void mlx5e_handle_tx_dim(struct mlx5e_txqsq *sq)
+{
+	struct net_dim_sample dim_sample;
+
+	if (unlikely(!MLX5E_TEST_BIT(sq->state, MLX5E_SQ_STATE_AM)))
+		return;
+
+	net_dim_sample(sq->cq.event_ctr, sq->stats.packets, sq->stats.bytes,
+		       &dim_sample);
+	net_dim(&sq->dim, dim_sample);
+}
+
+static inline void mlx5e_handle_rx_dim(struct mlx5e_rq *rq)
+{
+	struct net_dim_sample dim_sample;
+
+	if (unlikely(!MLX5E_TEST_BIT(rq->state, MLX5E_RQ_STATE_AM)))
+		return;
+
+	net_dim_sample(rq->cq.event_ctr, rq->stats.packets, rq->stats.bytes,
+		       &dim_sample);
+	net_dim(&rq->dim, dim_sample);
+}
+
 int mlx5e_napi_poll(struct napi_struct *napi, int budget)
 {
 	struct mlx5e_channel *c = container_of(napi, struct mlx5e_channel,
@@ -75,18 +99,13 @@ int mlx5e_napi_poll(struct napi_struct *napi, int budget)
 	if (unlikely(!napi_complete_done(napi, work_done)))
 		return work_done;
 
-	for (i = 0; i < c->num_tc; i++)
+	for (i = 0; i < c->num_tc; i++) {
+		mlx5e_handle_tx_dim(&c->sq[i]);
 		mlx5e_cq_arm(&c->sq[i].cq);
-
-	if (MLX5E_TEST_BIT(c->rq.state, MLX5E_RQ_STATE_AM)) {
-		struct net_dim_sample dim_sample;
-		net_dim_sample(c->rq.cq.event_ctr,
-			       c->rq.stats.packets,
-			       c->rq.stats.bytes,
-			       &dim_sample);
-		net_dim(&c->rq.dim, dim_sample);
 	}
 
+	mlx5e_handle_rx_dim(&c->rq);
+
 	mlx5e_cq_arm(&c->rq.cq);
 	mlx5e_cq_arm(&c->icosq.cq);
 
-- 
1.8.3.1

  parent reply	other threads:[~2018-03-30  6:38 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-03-30  6:37 [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM Tal Gilboa
2018-03-30  6:37 ` [PATCH net-next 1/4] net/dim: Rename *_get_profile() functions to *_get_rx_moderation() Tal Gilboa
2018-03-30  6:37 ` [PATCH net-next 2/4] net/dim: Add "enabled" field to net_dim struct Tal Gilboa
2018-03-30  6:37 ` [PATCH net-next 3/4] net/dim: Support adaptive TX moderation Tal Gilboa
2018-03-30  6:37 ` Tal Gilboa [this message]
2018-03-30  6:43 ` [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM Tal Gilboa
2018-04-01  2:02 ` David Miller
2018-04-01  3:19   ` David Miller
2018-04-01  6:49     ` Tal Gilboa

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=1522391853-37298-5-git-send-email-talgi@mellanox.com \
    --to=talgi@mellanox.com \
    --cc=andrew.gospodarek@broadcom.com \
    --cc=davem@davemloft.net \
    --cc=netdev@vger.kernel.org \
    --cc=saeedm@mellanox.com \
    --cc=tariqt@mellanox.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.