* [PATCH net-next 1/4] net/dim: Rename *_get_profile() functions to *_get_rx_moderation()
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 ` Tal Gilboa
2018-03-30 6:37 ` [PATCH net-next 2/4] net/dim: Add "enabled" field to net_dim struct Tal Gilboa
` (4 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Tal Gilboa @ 2018-03-30 6:37 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Tariq Toukan, Tal Gilboa, Andy Gospodarek, Saeed Mahameed
Preparation for introducing adaptive TX to net DIM.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
---
drivers/net/ethernet/broadcom/bcmsysport.c | 2 +-
drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c | 8 ++++----
drivers/net/ethernet/broadcom/genet/bcmgenet.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/en_dim.c | 6 +++---
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 6 ++++--
include/linux/net_dim.h | 12 ++++++------
6 files changed, 19 insertions(+), 17 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bcmsysport.c b/drivers/net/ethernet/broadcom/bcmsysport.c
index 4a75b1d..53c035f 100644
--- a/drivers/net/ethernet/broadcom/bcmsysport.c
+++ b/drivers/net/ethernet/broadcom/bcmsysport.c
@@ -1064,7 +1064,7 @@ static void bcm_sysport_dim_work(struct work_struct *work)
struct bcm_sysport_priv *priv =
container_of(ndim, struct bcm_sysport_priv, dim);
struct net_dim_cq_moder cur_profile =
- net_dim_get_profile(dim->mode, dim->profile_ix);
+ net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
bcm_sysport_set_rx_coalesce(priv, cur_profile.usec, cur_profile.pkts);
dim->state = NET_DIM_START_MEASURE;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
index 408dd19..afa97c8 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c
@@ -21,11 +21,11 @@ void bnxt_dim_work(struct work_struct *work)
struct bnxt_napi *bnapi = container_of(cpr,
struct bnxt_napi,
cp_ring);
- struct net_dim_cq_moder cur_profile = net_dim_get_profile(dim->mode,
- dim->profile_ix);
+ struct net_dim_cq_moder cur_moder =
+ net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
- cpr->rx_ring_coal.coal_ticks = cur_profile.usec;
- cpr->rx_ring_coal.coal_bufs = cur_profile.pkts;
+ cpr->rx_ring_coal.coal_ticks = cur_moder.usec;
+ cpr->rx_ring_coal.coal_bufs = cur_moder.pkts;
bnxt_hwrm_set_ring_coal(bnapi->bp, bnapi);
dim->state = NET_DIM_START_MEASURE;
diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
index f8af472..1c9c75b 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.c
+++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
@@ -1924,7 +1924,7 @@ static void bcmgenet_dim_work(struct work_struct *work)
struct bcmgenet_rx_ring *ring =
container_of(ndim, struct bcmgenet_rx_ring, dim);
struct net_dim_cq_moder cur_profile =
- net_dim_get_profile(dim->mode, dim->profile_ix);
+ net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
bcmgenet_set_rx_coalesce(ring, cur_profile.usec, cur_profile.pkts);
dim->state = NET_DIM_START_MEASURE;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
index 602851a..1b286e1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_dim.c
@@ -38,11 +38,11 @@ void mlx5e_rx_dim_work(struct work_struct *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_profile = net_dim_get_profile(dim->mode,
- dim->profile_ix);
+ 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_profile.usec, cur_profile.pkts);
+ cur_moder.usec, cur_moder.pkts);
dim->state = NET_DIM_START_MEASURE;
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 4a675f1..8a9670c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -4087,12 +4087,14 @@ void mlx5e_set_rx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
switch (cq_period_mode) {
case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
params->rx_cq_moderation =
- net_dim_get_def_profile(NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE);
+ net_dim_get_def_rx_moderation(
+ NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE);
break;
case MLX5_CQ_PERIOD_MODE_START_FROM_EQE:
default:
params->rx_cq_moderation =
- net_dim_get_def_profile(NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE);
+ net_dim_get_def_rx_moderation(
+ NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE);
}
}
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index bebeaad..e1cabaf 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -129,17 +129,17 @@ enum {
NET_DIM_CQE_PROFILES,
};
-static inline struct net_dim_cq_moder net_dim_get_profile(u8 cq_period_mode,
- int ix)
+static inline struct net_dim_cq_moder
+net_dim_get_rx_moderation(u8 cq_period_mode, int ix)
{
- struct net_dim_cq_moder cq_moder;
+ struct net_dim_cq_moder cq_moder = profile[cq_period_mode][ix];
- cq_moder = profile[cq_period_mode][ix];
cq_moder.cq_period_mode = cq_period_mode;
return cq_moder;
}
-static inline struct net_dim_cq_moder net_dim_get_def_profile(u8 rx_cq_period_mode)
+static inline struct net_dim_cq_moder
+net_dim_get_def_rx_moderation(u8 rx_cq_period_mode)
{
int default_profile_ix;
@@ -148,7 +148,7 @@ static inline struct net_dim_cq_moder net_dim_get_def_profile(u8 rx_cq_period_mo
else /* NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE */
default_profile_ix = NET_DIM_DEF_PROFILE_EQE;
- return net_dim_get_profile(rx_cq_period_mode, default_profile_ix);
+ return net_dim_get_rx_moderation(rx_cq_period_mode, default_profile_ix);
}
static inline bool net_dim_on_top(struct net_dim *dim)
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH net-next 2/4] net/dim: Add "enabled" field to net_dim struct
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 ` Tal Gilboa
2018-03-30 6:37 ` [PATCH net-next 3/4] net/dim: Support adaptive TX moderation Tal Gilboa
` (3 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Tal Gilboa @ 2018-03-30 6:37 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Tariq Toukan, Tal Gilboa, Andy Gospodarek, Saeed Mahameed
Preparation for introducing adaptive TX to net DIM.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 1 -
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c | 10 +++++++---
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 6 +++---
drivers/net/ethernet/mellanox/mlx5/core/en_rep.c | 2 +-
include/linux/net_dim.h | 2 ++
5 files changed, 13 insertions(+), 8 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 0c2fdf0..3e24864 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -250,7 +250,6 @@ struct mlx5e_params {
u32 indirection_rqt[MLX5E_INDIR_RQT_SIZE];
bool vlan_strip_disable;
bool scatter_fcs_en;
- bool rx_dim_enabled;
u32 lro_timeout;
u32 pflags;
struct bpf_prog *xdp_prog;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
index c57c929..4b5d022 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
@@ -459,9 +459,10 @@ int mlx5e_ethtool_get_coalesce(struct mlx5e_priv *priv,
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;
- coal->use_adaptive_rx_coalesce = priv->channels.params.rx_dim_enabled;
return 0;
}
@@ -515,7 +516,8 @@ int mlx5e_ethtool_set_coalesce(struct mlx5e_priv *priv,
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_dim_enabled = !!coal->use_adaptive_rx_coalesce;
+ new_channels.params.rx_cq_moderation.enabled =
+ !!coal->use_adaptive_rx_coalesce;
if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
priv->channels.params = new_channels.params;
@@ -523,7 +525,9 @@ int mlx5e_ethtool_set_coalesce(struct mlx5e_priv *priv,
}
/* we are opened */
- reset = !!coal->use_adaptive_rx_coalesce != priv->channels.params.rx_dim_enabled;
+ reset = !!coal->use_adaptive_rx_coalesce !=
+ priv->channels.params.rx_cq_moderation.enabled;
+
if (!reset) {
mlx5e_set_priv_channels_coalesce(priv, coal);
priv->channels.params = new_channels.params;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 8a9670c..e16a705 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -781,7 +781,7 @@ static int mlx5e_open_rq(struct mlx5e_channel *c,
if (err)
goto err_destroy_rq;
- if (params->rx_dim_enabled)
+ if (params->rx_cq_moderation.enabled)
c->rq.state |= BIT(MLX5E_RQ_STATE_AM);
return 0;
@@ -4083,7 +4083,7 @@ void mlx5e_set_rx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
params->rx_cq_moderation.usec =
MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE;
- if (params->rx_dim_enabled) {
+ if (params->rx_cq_moderation.enabled) {
switch (cq_period_mode) {
case MLX5_CQ_PERIOD_MODE_START_FROM_CQE:
params->rx_cq_moderation =
@@ -4155,7 +4155,7 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
- params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
+ params->rx_cq_moderation.enabled = MLX5_CAP_GEN(mdev, cq_moderation);
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_rep.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
index dd32f3e..f238d4c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
@@ -881,7 +881,7 @@ static void mlx5e_build_rep_params(struct mlx5_core_dev *mdev,
params->rq_wq_type = MLX5_WQ_TYPE_LINKED_LIST;
params->log_rq_size = MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE;
- params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
+ params->rx_cq_moderation.enabled = MLX5_CAP_GEN(mdev, cq_moderation);
mlx5e_set_rx_cq_mode_params(params, cq_period_mode);
params->num_tc = 1;
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index e1cabaf..d34fbfe 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -40,6 +40,7 @@ struct net_dim_cq_moder {
u16 usec;
u16 pkts;
u8 cq_period_mode;
+ bool enabled;
};
struct net_dim_sample {
@@ -135,6 +136,7 @@ enum {
struct net_dim_cq_moder cq_moder = profile[cq_period_mode][ix];
cq_moder.cq_period_mode = cq_period_mode;
+ cq_moder.enabled = true;
return cq_moder;
}
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH net-next 3/4] net/dim: Support adaptive TX moderation
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 ` Tal Gilboa
2018-03-30 6:37 ` [PATCH net-next 4/4] net/mlx5e: Enable adaptive-TX moderation Tal Gilboa
` (2 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Tal Gilboa @ 2018-03-30 6:37 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Tariq Toukan, Tal Gilboa, Andy Gospodarek, Saeed Mahameed
Interrupt moderation for TX traffic requires different profiles than RX
interrupt moderation. The main goal here is to reduce interrupt rate and
allow better payload aggregation by keeping SKBs in the TX queue a bit
longer. Ping-pong behavior would get a profile with a short timer, so
latency wouldn't increase for these scenarios. There's a slight degradation
in bandwidth for single stream with large message sizes, since
net.ipv4.tcp_limit_output_bytes is limiting the allowed TX traffic, but
with many streams performance is always improved.
Performance improvements (ConnectX-5 100GbE)
Bandwidth: increased up to 40% (1024B with 10s of streams).
Interrupt rate: reduced up to 50% (1024B with 1000s of streams).
Performance degradation (ConnectX-5 100GbE)
Bandwidth: up to 10% decrease single stream TCP (1MB message size from
51Gb/s to 47Gb/s).
*Both cases with TX EQE based moderation enabled.
Signed-off-by: Tal Gilboa <talgi@mellanox.com>
---
include/linux/net_dim.h | 64 +++++++++++++++++++++++++++++++++++++++----------
1 file changed, 51 insertions(+), 13 deletions(-)
diff --git a/include/linux/net_dim.h b/include/linux/net_dim.h
index d34fbfe..9449a61 100644
--- a/include/linux/net_dim.h
+++ b/include/linux/net_dim.h
@@ -104,11 +104,12 @@ enum {
#define NET_DIM_PARAMS_NUM_PROFILES 5
/* Adaptive moderation profiles */
#define NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE 256
+#define NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE 128
#define NET_DIM_DEF_PROFILE_CQE 1
#define NET_DIM_DEF_PROFILE_EQE 1
/* All profiles sizes must be NET_PARAMS_DIM_NUM_PROFILES */
-#define NET_DIM_EQE_PROFILES { \
+#define NET_DIM_RX_EQE_PROFILES { \
{1, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
{8, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
{64, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
@@ -116,7 +117,7 @@ enum {
{256, NET_DIM_DEFAULT_RX_CQ_MODERATION_PKTS_FROM_EQE}, \
}
-#define NET_DIM_CQE_PROFILES { \
+#define NET_DIM_RX_CQE_PROFILES { \
{2, 256}, \
{8, 128}, \
{16, 64}, \
@@ -124,16 +125,38 @@ enum {
{64, 64} \
}
+#define NET_DIM_TX_EQE_PROFILES { \
+ {1, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
+ {8, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
+ {32, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
+ {64, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE}, \
+ {128, NET_DIM_DEFAULT_TX_CQ_MODERATION_PKTS_FROM_EQE} \
+}
+
+#define NET_DIM_TX_CQE_PROFILES { \
+ {5, 128}, \
+ {8, 64}, \
+ {16, 32}, \
+ {32, 32}, \
+ {64, 32} \
+}
+
+static const struct net_dim_cq_moder
+rx_profile[NET_DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
+ NET_DIM_RX_EQE_PROFILES,
+ NET_DIM_RX_CQE_PROFILES,
+};
+
static const struct net_dim_cq_moder
-profile[NET_DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
- NET_DIM_EQE_PROFILES,
- NET_DIM_CQE_PROFILES,
+tx_profile[NET_DIM_CQ_PERIOD_NUM_MODES][NET_DIM_PARAMS_NUM_PROFILES] = {
+ NET_DIM_TX_EQE_PROFILES,
+ NET_DIM_TX_CQE_PROFILES,
};
static inline struct net_dim_cq_moder
net_dim_get_rx_moderation(u8 cq_period_mode, int ix)
{
- struct net_dim_cq_moder cq_moder = profile[cq_period_mode][ix];
+ struct net_dim_cq_moder cq_moder = rx_profile[cq_period_mode][ix];
cq_moder.cq_period_mode = cq_period_mode;
cq_moder.enabled = true;
@@ -141,16 +164,31 @@ enum {
}
static inline struct net_dim_cq_moder
-net_dim_get_def_rx_moderation(u8 rx_cq_period_mode)
+net_dim_get_def_rx_moderation(u8 cq_period_mode)
{
- int default_profile_ix;
+ u8 profile_ix = cq_period_mode == NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+ NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
- if (rx_cq_period_mode == NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE)
- default_profile_ix = NET_DIM_DEF_PROFILE_CQE;
- else /* NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE */
- default_profile_ix = NET_DIM_DEF_PROFILE_EQE;
+ return net_dim_get_rx_moderation(cq_period_mode, profile_ix);
+}
+
+static inline struct net_dim_cq_moder
+net_dim_get_tx_moderation(u8 cq_period_mode, int ix)
+{
+ struct net_dim_cq_moder cq_moder = tx_profile[cq_period_mode][ix];
+
+ cq_moder.cq_period_mode = cq_period_mode;
+ cq_moder.enabled = true;
+ return cq_moder;
+}
+
+static inline struct net_dim_cq_moder
+net_dim_get_def_tx_moderation(u8 cq_period_mode)
+{
+ u8 profile_ix = cq_period_mode == NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE ?
+ NET_DIM_DEF_PROFILE_CQE : NET_DIM_DEF_PROFILE_EQE;
- return net_dim_get_rx_moderation(rx_cq_period_mode, default_profile_ix);
+ return net_dim_get_tx_moderation(cq_period_mode, profile_ix);
}
static inline bool net_dim_on_top(struct net_dim *dim)
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH net-next 4/4] net/mlx5e: Enable adaptive-TX moderation
2018-03-30 6:37 [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM Tal Gilboa
` (2 preceding siblings ...)
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
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
5 siblings, 0 replies; 9+ messages in thread
From: Tal Gilboa @ 2018-03-30 6:37 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Tariq Toukan, Tal Gilboa, Andy Gospodarek, Saeed Mahameed
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
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM
2018-03-30 6:37 [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM Tal Gilboa
` (3 preceding siblings ...)
2018-03-30 6:37 ` [PATCH net-next 4/4] net/mlx5e: Enable adaptive-TX moderation Tal Gilboa
@ 2018-03-30 6:43 ` Tal Gilboa
2018-04-01 2:02 ` David Miller
5 siblings, 0 replies; 9+ messages in thread
From: Tal Gilboa @ 2018-03-30 6:43 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Tariq Toukan, Andy Gospodarek, Saeed Mahameed, Florian Fainelli
On 3/30/2018 9:37 AM, Tal Gilboa wrote:
> Net DIM is a library designed for dynamic interrupt moderation. It was
> implemented and optimized with receive side interrupts in mind, since these
> are usually the CPU expensive ones. This patch-set introduces adaptive transmit
> interrupt moderation to net DIM, complete with a usage in the mlx5e driver.
> Using adaptive TX behavior would reduce interrupt rate for multiple scenarios.
> Furthermore, it is essential for increasing bandwidth on cases where payload
> aggregation is required.
Important note: In order to avoid conflicts, this patch-set is rebased
over Florian Fainelli's "[PATCH net-next v2 3/3] net: bcmgenet: Fix
coalescing settings handling" patch.
>
> Tal Gilboa (4):
> net/dim: Rename *_get_profile() functions to *_get_rx_moderation()
> net/dim: Add "enabled" field to net_dim struct
> net/dim: Support adaptive TX moderation
> net/mlx5e: Enable adaptive-TX moderation
>
> drivers/net/ethernet/broadcom/bcmsysport.c | 2 +-
> drivers/net/ethernet/broadcom/bnxt/bnxt_dim.c | 8 +--
> drivers/net/ethernet/broadcom/genet/bcmgenet.c | 2 +-
> drivers/net/ethernet/mellanox/mlx5/core/en.h | 5 +-
> drivers/net/ethernet/mellanox/mlx5/core/en_dim.c | 28 ++++++---
> .../net/ethernet/mellanox/mlx5/core/en_ethtool.c | 35 +++++++----
> drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 34 ++++++++--
> drivers/net/ethernet/mellanox/mlx5/core/en_rep.c | 2 +-
> drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 37 ++++++++---
> include/linux/net_dim.h | 72 +++++++++++++++++-----
> 10 files changed, 169 insertions(+), 56 deletions(-)
>
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM
2018-03-30 6:37 [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM Tal Gilboa
` (4 preceding siblings ...)
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
5 siblings, 1 reply; 9+ messages in thread
From: David Miller @ 2018-04-01 2:02 UTC (permalink / raw)
To: talgi; +Cc: netdev, tariqt, andrew.gospodarek, saeedm
From: Tal Gilboa <talgi@mellanox.com>
Date: Fri, 30 Mar 2018 09:37:29 +0300
> Net DIM is a library designed for dynamic interrupt moderation. It was
> implemented and optimized with receive side interrupts in mind, since these
> are usually the CPU expensive ones. This patch-set introduces adaptive transmit
> interrupt moderation to net DIM, complete with a usage in the mlx5e driver.
> Using adaptive TX behavior would reduce interrupt rate for multiple scenarios.
> Furthermore, it is essential for increasing bandwidth on cases where payload
> aggregation is required.
Series applied.
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM
2018-04-01 2:02 ` David Miller
@ 2018-04-01 3:19 ` David Miller
2018-04-01 6:49 ` Tal Gilboa
0 siblings, 1 reply; 9+ messages in thread
From: David Miller @ 2018-04-01 3:19 UTC (permalink / raw)
To: talgi; +Cc: netdev, tariqt, andrew.gospodarek, saeedm
From: David Miller <davem@davemloft.net>
Date: Sat, 31 Mar 2018 22:02:55 -0400 (EDT)
> From: Tal Gilboa <talgi@mellanox.com>
> Date: Fri, 30 Mar 2018 09:37:29 +0300
>
>> Net DIM is a library designed for dynamic interrupt moderation. It was
>> implemented and optimized with receive side interrupts in mind, since these
>> are usually the CPU expensive ones. This patch-set introduces adaptive transmit
>> interrupt moderation to net DIM, complete with a usage in the mlx5e driver.
>> Using adaptive TX behavior would reduce interrupt rate for multiple scenarios.
>> Furthermore, it is essential for increasing bandwidth on cases where payload
>> aggregation is required.
>
> Series applied.
This breaks the build:
drivers/net/ethernet/broadcom/bcmsysport.c: In function ‘bcm_sysport_set_coalesce’:
drivers/net/ethernet/broadcom/bcmsysport.c:657:11: error: implicit declaration of function ‘net_dim_get_def_profile’; did you mean ‘net_dim_exit_parking’? [-Werror=implicit-function-declaration]
moder = net_dim_get_def_profile(priv->dim.dim.mode);
^~~~~~~~~~~~~~~~~~~~~~~
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH net-next 0/4] Introduce adaptive TX interrupt moderation to net DIM
2018-04-01 3:19 ` David Miller
@ 2018-04-01 6:49 ` Tal Gilboa
0 siblings, 0 replies; 9+ messages in thread
From: Tal Gilboa @ 2018-04-01 6:49 UTC (permalink / raw)
To: David Miller; +Cc: netdev, tariqt, andrew.gospodarek, saeedm
On 4/1/2018 6:19 AM, David Miller wrote:
> From: David Miller <davem@davemloft.net>
> Date: Sat, 31 Mar 2018 22:02:55 -0400 (EDT)
>
>> From: Tal Gilboa <talgi@mellanox.com>
>> Date: Fri, 30 Mar 2018 09:37:29 +0300
>>
>>> Net DIM is a library designed for dynamic interrupt moderation. It was
>>> implemented and optimized with receive side interrupts in mind, since these
>>> are usually the CPU expensive ones. This patch-set introduces adaptive transmit
>>> interrupt moderation to net DIM, complete with a usage in the mlx5e driver.
>>> Using adaptive TX behavior would reduce interrupt rate for multiple scenarios.
>>> Furthermore, it is essential for increasing bandwidth on cases where payload
>>> aggregation is required.
>>
>> Series applied.
>
> This breaks the build:
>
> drivers/net/ethernet/broadcom/bcmsysport.c: In function ‘bcm_sysport_set_coalesce’:
> drivers/net/ethernet/broadcom/bcmsysport.c:657:11: error: implicit declaration of function ‘net_dim_get_def_profile’; did you mean ‘net_dim_exit_parking’? [-Werror=implicit-function-declaration]
> moder = net_dim_get_def_profile(priv->dim.dim.mode);
> ^~~~~~~~~~~~~~~~~~~~~~~
>
Rebase issues, sorry.
Sent a fixup patch.
^ permalink raw reply [flat|nested] 9+ messages in thread