* [PATCH net-next V2 01/11] net/mlx5: Introduce device queue counters
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
@ 2016-04-17 21:31 ` Saeed Mahameed
2016-04-17 21:31 ` [PATCH net-next V2 02/11] net/mlx5e: Allocate set of queue counters per netdev Saeed Mahameed
` (9 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:31 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Rana Shahout, Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
A queue counter can collect several statistics for one or more
hardware queues (QPs, RQs, etc ..) that the counter is attached to.
For Ethernet it will provide an "out of buffer" counter which collects
the number of all packets that are dropped due to lack of software
buffers.
Here we add device commands to alloc/query/dealloc queue counters.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/qp.c | 68 ++++++++++++++++++++++++++
include/linux/mlx5/qp.h | 6 ++
2 files changed, 74 insertions(+), 0 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/qp.c b/drivers/net/ethernet/mellanox/mlx5/core/qp.c
index def2893..b720a27 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/qp.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/qp.c
@@ -538,3 +538,71 @@ void mlx5_core_destroy_sq_tracked(struct mlx5_core_dev *dev,
mlx5_core_destroy_sq(dev, sq->qpn);
}
EXPORT_SYMBOL(mlx5_core_destroy_sq_tracked);
+
+int mlx5_core_alloc_q_counter(struct mlx5_core_dev *dev, u16 *counter_id)
+{
+ u32 in[MLX5_ST_SZ_DW(alloc_q_counter_in)];
+ u32 out[MLX5_ST_SZ_DW(alloc_q_counter_out)];
+ int err;
+
+ memset(in, 0, sizeof(in));
+ memset(out, 0, sizeof(out));
+
+ MLX5_SET(alloc_q_counter_in, in, opcode, MLX5_CMD_OP_ALLOC_Q_COUNTER);
+ err = mlx5_cmd_exec_check_status(dev, in, sizeof(in), out, sizeof(out));
+ if (!err)
+ *counter_id = MLX5_GET(alloc_q_counter_out, out,
+ counter_set_id);
+ return err;
+}
+EXPORT_SYMBOL_GPL(mlx5_core_alloc_q_counter);
+
+int mlx5_core_dealloc_q_counter(struct mlx5_core_dev *dev, u16 counter_id)
+{
+ u32 in[MLX5_ST_SZ_DW(dealloc_q_counter_in)];
+ u32 out[MLX5_ST_SZ_DW(dealloc_q_counter_out)];
+
+ memset(in, 0, sizeof(in));
+ memset(out, 0, sizeof(out));
+
+ MLX5_SET(dealloc_q_counter_in, in, opcode,
+ MLX5_CMD_OP_DEALLOC_Q_COUNTER);
+ MLX5_SET(dealloc_q_counter_in, in, counter_set_id, counter_id);
+ return mlx5_cmd_exec_check_status(dev, in, sizeof(in), out,
+ sizeof(out));
+}
+EXPORT_SYMBOL_GPL(mlx5_core_dealloc_q_counter);
+
+int mlx5_core_query_q_counter(struct mlx5_core_dev *dev, u16 counter_id,
+ int reset, void *out, int out_size)
+{
+ u32 in[MLX5_ST_SZ_DW(query_q_counter_in)];
+
+ memset(in, 0, sizeof(in));
+
+ MLX5_SET(query_q_counter_in, in, opcode, MLX5_CMD_OP_QUERY_Q_COUNTER);
+ MLX5_SET(query_q_counter_in, in, clear, reset);
+ MLX5_SET(query_q_counter_in, in, counter_set_id, counter_id);
+ return mlx5_cmd_exec_check_status(dev, in, sizeof(in), out, out_size);
+}
+EXPORT_SYMBOL_GPL(mlx5_core_query_q_counter);
+
+int mlx5_core_query_out_of_buffer(struct mlx5_core_dev *dev, u16 counter_id,
+ u32 *out_of_buffer)
+{
+ int outlen = MLX5_ST_SZ_BYTES(query_q_counter_out);
+ void *out;
+ int err;
+
+ out = mlx5_vzalloc(outlen);
+ if (!out)
+ return -ENOMEM;
+
+ err = mlx5_core_query_q_counter(dev, counter_id, 0, out, outlen);
+ if (!err)
+ *out_of_buffer = MLX5_GET(query_q_counter_out, out,
+ out_of_buffer);
+
+ kfree(out);
+ return err;
+}
diff --git a/include/linux/mlx5/qp.h b/include/linux/mlx5/qp.h
index cf031a3..6422102 100644
--- a/include/linux/mlx5/qp.h
+++ b/include/linux/mlx5/qp.h
@@ -668,6 +668,12 @@ int mlx5_core_create_sq_tracked(struct mlx5_core_dev *dev, u32 *in, int inlen,
struct mlx5_core_qp *sq);
void mlx5_core_destroy_sq_tracked(struct mlx5_core_dev *dev,
struct mlx5_core_qp *sq);
+int mlx5_core_alloc_q_counter(struct mlx5_core_dev *dev, u16 *counter_id);
+int mlx5_core_dealloc_q_counter(struct mlx5_core_dev *dev, u16 counter_id);
+int mlx5_core_query_q_counter(struct mlx5_core_dev *dev, u16 counter_id,
+ int reset, void *out, int out_size);
+int mlx5_core_query_out_of_buffer(struct mlx5_core_dev *dev, u16 counter_id,
+ u32 *out_of_buffer);
static inline const char *mlx5_qp_type_str(int type)
{
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 02/11] net/mlx5e: Allocate set of queue counters per netdev
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
2016-04-17 21:31 ` [PATCH net-next V2 01/11] net/mlx5: Introduce device queue counters Saeed Mahameed
@ 2016-04-17 21:31 ` Saeed Mahameed
2016-04-17 21:31 ` [PATCH net-next V2 03/11] net/mlx5e: Use only close NUMA node for default RSS Saeed Mahameed
` (8 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:31 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Rana Shahout, Saeed Mahameed
From: Rana Shahout <ranas@mellanox.com>
Connect all netdev RQs to this set of queue counters.
Also, add an "rx_out_of_buffer" counter to ethtool,
which indicates RX packet drops due to lack of receive
buffers.
Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 11 +++++
.../net/ethernet/mellanox/mlx5/core/en_ethtool.c | 11 +++++
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 42 +++++++++++++++++++-
3 files changed, 62 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 879e627..c4ddbe8 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -236,6 +236,15 @@ struct mlx5e_pport_stats {
__be64 RFC_2819_counters[NUM_RFC_2819_COUNTERS];
};
+static const char qcounter_stats_strings[][ETH_GSTRING_LEN] = {
+ "rx_out_of_buffer",
+};
+
+struct mlx5e_qcounter_stats {
+ u32 rx_out_of_buffer;
+#define NUM_Q_COUNTERS 1
+};
+
static const char rq_stats_strings[][ETH_GSTRING_LEN] = {
"packets",
"bytes",
@@ -293,6 +302,7 @@ struct mlx5e_sq_stats {
struct mlx5e_stats {
struct mlx5e_vport_stats vport;
struct mlx5e_pport_stats pport;
+ struct mlx5e_qcounter_stats qcnt;
};
struct mlx5e_params {
@@ -575,6 +585,7 @@ struct mlx5e_priv {
struct net_device *netdev;
struct mlx5e_stats stats;
struct mlx5e_tstamp tstamp;
+ u16 q_counter;
};
#define MLX5E_NET_IP_ALIGN 2
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
index 68834b7..39c1902 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
@@ -165,6 +165,8 @@ static const struct {
},
};
+#define MLX5E_NUM_Q_CNTRS(priv) (NUM_Q_COUNTERS * (!!priv->q_counter))
+
static int mlx5e_get_sset_count(struct net_device *dev, int sset)
{
struct mlx5e_priv *priv = netdev_priv(dev);
@@ -172,6 +174,7 @@ static int mlx5e_get_sset_count(struct net_device *dev, int sset)
switch (sset) {
case ETH_SS_STATS:
return NUM_VPORT_COUNTERS + NUM_PPORT_COUNTERS +
+ MLX5E_NUM_Q_CNTRS(priv) +
priv->params.num_channels * NUM_RQ_STATS +
priv->params.num_channels * priv->params.num_tc *
NUM_SQ_STATS;
@@ -200,6 +203,11 @@ static void mlx5e_get_strings(struct net_device *dev,
strcpy(data + (idx++) * ETH_GSTRING_LEN,
vport_strings[i]);
+ /* Q counters */
+ for (i = 0; i < MLX5E_NUM_Q_CNTRS(priv); i++)
+ strcpy(data + (idx++) * ETH_GSTRING_LEN,
+ qcounter_stats_strings[i]);
+
/* PPORT counters */
for (i = 0; i < NUM_PPORT_COUNTERS; i++)
strcpy(data + (idx++) * ETH_GSTRING_LEN,
@@ -240,6 +248,9 @@ static void mlx5e_get_ethtool_stats(struct net_device *dev,
for (i = 0; i < NUM_VPORT_COUNTERS; i++)
data[idx++] = ((u64 *)&priv->stats.vport)[i];
+ for (i = 0; i < MLX5E_NUM_Q_CNTRS(priv); i++)
+ data[idx++] = ((u32 *)&priv->stats.qcnt)[i];
+
for (i = 0; i < NUM_PPORT_COUNTERS; i++)
data[idx++] = be64_to_cpu(((__be64 *)&priv->stats.pport)[i]);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index e0adb60..7fbe1ba 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -129,6 +129,17 @@ free_out:
kvfree(out);
}
+static void mlx5e_update_q_counter(struct mlx5e_priv *priv)
+{
+ struct mlx5e_qcounter_stats *qcnt = &priv->stats.qcnt;
+
+ if (!priv->q_counter)
+ return;
+
+ mlx5_core_query_out_of_buffer(priv->mdev, priv->q_counter,
+ &qcnt->rx_out_of_buffer);
+}
+
void mlx5e_update_stats(struct mlx5e_priv *priv)
{
struct mlx5_core_dev *mdev = priv->mdev;
@@ -250,6 +261,8 @@ void mlx5e_update_stats(struct mlx5e_priv *priv)
s->rx_csum_sw;
mlx5e_update_pport_counters(priv);
+ mlx5e_update_q_counter(priv);
+
free_out:
kvfree(out);
}
@@ -1055,6 +1068,7 @@ static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
MLX5_SET(wq, wq, pd, priv->pdn);
+ MLX5_SET(rqc, rqc, counter_set_id, priv->q_counter);
param->wq.buf_numa_node = dev_to_node(&priv->mdev->pdev->dev);
param->wq.linear = 1;
@@ -2442,6 +2456,26 @@ static int mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn,
return err;
}
+static void mlx5e_create_q_counter(struct mlx5e_priv *priv)
+{
+ struct mlx5_core_dev *mdev = priv->mdev;
+ int err;
+
+ err = mlx5_core_alloc_q_counter(mdev, &priv->q_counter);
+ if (err) {
+ mlx5_core_warn(mdev, "alloc queue counter failed, %d\n", err);
+ priv->q_counter = 0;
+ }
+}
+
+static void mlx5e_destroy_q_counter(struct mlx5e_priv *priv)
+{
+ if (!priv->q_counter)
+ return;
+
+ mlx5_core_dealloc_q_counter(priv->mdev, priv->q_counter);
+}
+
static void *mlx5e_create_netdev(struct mlx5_core_dev *mdev)
{
struct net_device *netdev;
@@ -2527,13 +2561,15 @@ static void *mlx5e_create_netdev(struct mlx5_core_dev *mdev)
goto err_destroy_tirs;
}
+ mlx5e_create_q_counter(priv);
+
mlx5e_init_eth_addr(priv);
mlx5e_vxlan_init(priv);
err = mlx5e_tc_init(priv);
if (err)
- goto err_destroy_flow_tables;
+ goto err_dealloc_q_counters;
#ifdef CONFIG_MLX5_CORE_EN_DCB
mlx5e_dcbnl_ieee_setets_core(priv, &priv->params.ets);
@@ -2556,7 +2592,8 @@ static void *mlx5e_create_netdev(struct mlx5_core_dev *mdev)
err_tc_cleanup:
mlx5e_tc_cleanup(priv);
-err_destroy_flow_tables:
+err_dealloc_q_counters:
+ mlx5e_destroy_q_counter(priv);
mlx5e_destroy_flow_tables(priv);
err_destroy_tirs:
@@ -2605,6 +2642,7 @@ static void mlx5e_destroy_netdev(struct mlx5_core_dev *mdev, void *vpriv)
unregister_netdev(netdev);
mlx5e_tc_cleanup(priv);
mlx5e_vxlan_cleanup(priv);
+ mlx5e_destroy_q_counter(priv);
mlx5e_destroy_flow_tables(priv);
mlx5e_destroy_tirs(priv);
mlx5e_destroy_rqt(priv, MLX5E_SINGLE_RQ_RQT);
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 03/11] net/mlx5e: Use only close NUMA node for default RSS
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
2016-04-17 21:31 ` [PATCH net-next V2 01/11] net/mlx5: Introduce device queue counters Saeed Mahameed
2016-04-17 21:31 ` [PATCH net-next V2 02/11] net/mlx5e: Allocate set of queue counters per netdev Saeed Mahameed
@ 2016-04-17 21:31 ` Saeed Mahameed
2016-04-17 21:31 ` [PATCH net-next V2 04/11] net/mlx5e: Use function pointers for RX data path handling Saeed Mahameed
` (7 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:31 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
Distribute default RSS table uniformly over the rings of the
close NUMA node, instead of all available channels.
This way we enforce the preference of close rings over far ones.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 3 ++-
.../net/ethernet/mellanox/mlx5/core/en_ethtool.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 15 +++++++++++++--
3 files changed, 16 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index c4ddbe8..7f19644 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -671,7 +671,8 @@ void mlx5e_build_tir_ctx_hash(void *tirc, struct mlx5e_priv *priv);
int mlx5e_open_locked(struct net_device *netdev);
int mlx5e_close_locked(struct net_device *netdev);
-void mlx5e_build_default_indir_rqt(u32 *indirection_rqt, int len,
+void mlx5e_build_default_indir_rqt(struct mlx5_core_dev *mdev,
+ u32 *indirection_rqt, int len,
int num_channels);
static inline void mlx5e_tx_notify_hw(struct mlx5e_sq *sq,
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
index 39c1902..6f40ba4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
@@ -397,7 +397,7 @@ static int mlx5e_set_channels(struct net_device *dev,
mlx5e_close_locked(dev);
priv->params.num_channels = count;
- mlx5e_build_default_indir_rqt(priv->params.indirection_rqt,
+ mlx5e_build_default_indir_rqt(priv->mdev, priv->params.indirection_rqt,
MLX5E_INDIR_RQT_SIZE, count);
if (was_opened)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 7fbe1ba..9b58ef6 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2297,11 +2297,22 @@ static void mlx5e_ets_init(struct mlx5e_priv *priv)
}
#endif
-void mlx5e_build_default_indir_rqt(u32 *indirection_rqt, int len,
+void mlx5e_build_default_indir_rqt(struct mlx5_core_dev *mdev,
+ u32 *indirection_rqt, int len,
int num_channels)
{
+ int node = mdev->priv.numa_node;
+ int node_num_of_cores;
int i;
+ if (node == -1)
+ node = first_online_node;
+
+ node_num_of_cores = cpumask_weight(cpumask_of_node(node));
+
+ if (node_num_of_cores)
+ num_channels = min_t(int, num_channels, node_num_of_cores);
+
for (i = 0; i < len; i++)
indirection_rqt[i] = i % num_channels;
}
@@ -2333,7 +2344,7 @@ static void mlx5e_build_netdev_priv(struct mlx5_core_dev *mdev,
netdev_rss_key_fill(priv->params.toeplitz_hash_key,
sizeof(priv->params.toeplitz_hash_key));
- mlx5e_build_default_indir_rqt(priv->params.indirection_rqt,
+ mlx5e_build_default_indir_rqt(mdev, priv->params.indirection_rqt,
MLX5E_INDIR_RQT_SIZE, num_channels);
priv->params.lro_wqe_sz =
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 04/11] net/mlx5e: Use function pointers for RX data path handling
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (2 preceding siblings ...)
2016-04-17 21:31 ` [PATCH net-next V2 03/11] net/mlx5e: Use only close NUMA node for default RSS Saeed Mahameed
@ 2016-04-17 21:31 ` Saeed Mahameed
2016-04-17 21:31 ` [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ) Saeed Mahameed
` (6 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:31 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Achiad Shochat, Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
In preparation for Striding RQ feature, which will need its own
RX handlers.
This patch does not change any functionality.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 33 ++++++----
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 +
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 74 +++++++++++----------
3 files changed, 62 insertions(+), 47 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 7f19644..61e249d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -72,6 +72,17 @@
#define MLX5E_SQ_BF_BUDGET 16
#define MLX5E_NUM_MAIN_GROUPS 9
+#define MLX5E_NET_IP_ALIGN 2
+
+struct mlx5e_tx_wqe {
+ struct mlx5_wqe_ctrl_seg ctrl;
+ struct mlx5_wqe_eth_seg eth;
+};
+
+struct mlx5e_rx_wqe {
+ struct mlx5_wqe_srq_next_seg next;
+ struct mlx5_wqe_data_seg data;
+};
#ifdef CONFIG_MLX5_CORE_EN_DCB
#define MLX5E_MAX_BW_ALLOC 100 /* Max percentage of BW allocation */
@@ -357,6 +368,12 @@ struct mlx5e_cq {
struct mlx5_wq_ctrl wq_ctrl;
} ____cacheline_aligned_in_smp;
+struct mlx5e_rq;
+typedef void (*mlx5e_fp_handle_rx_cqe)(struct mlx5e_rq *rq,
+ struct mlx5_cqe64 *cqe);
+typedef int (*mlx5e_fp_alloc_wqe)(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe,
+ u16 ix);
+
struct mlx5e_rq {
/* data path */
struct mlx5_wq_ll wq;
@@ -368,6 +385,8 @@ struct mlx5e_rq {
struct mlx5e_tstamp *tstamp;
struct mlx5e_rq_stats stats;
struct mlx5e_cq cq;
+ mlx5e_fp_handle_rx_cqe handle_rx_cqe;
+ mlx5e_fp_alloc_wqe alloc_wqe;
unsigned long state;
int ix;
@@ -588,18 +607,6 @@ struct mlx5e_priv {
u16 q_counter;
};
-#define MLX5E_NET_IP_ALIGN 2
-
-struct mlx5e_tx_wqe {
- struct mlx5_wqe_ctrl_seg ctrl;
- struct mlx5_wqe_eth_seg eth;
-};
-
-struct mlx5e_rx_wqe {
- struct mlx5_wqe_srq_next_seg next;
- struct mlx5_wqe_data_seg data;
-};
-
enum mlx5e_link_mode {
MLX5E_1000BASE_CX_SGMII = 0,
MLX5E_1000BASE_KX = 1,
@@ -642,7 +649,9 @@ void mlx5e_cq_error_event(struct mlx5_core_cq *mcq, enum mlx5_event event);
int mlx5e_napi_poll(struct napi_struct *napi, int budget);
bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq, int napi_budget);
int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget);
+void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe);
bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq);
+int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix);
struct mlx5_cqe64 *mlx5e_get_cqe(struct mlx5e_cq *cq);
void mlx5e_update_stats(struct mlx5e_priv *priv);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 9b58ef6..23ba12c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -357,6 +357,8 @@ static int mlx5e_create_rq(struct mlx5e_channel *c,
cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
}
+ rq->handle_rx_cqe = mlx5e_handle_rx_cqe;
+ rq->alloc_wqe = mlx5e_alloc_rx_wqe;
rq->pdev = c->pdev;
rq->netdev = c->netdev;
rq->tstamp = &priv->tstamp;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 58d4e2f..d7ccced 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -42,8 +42,7 @@ static inline bool mlx5e_rx_hw_stamp(struct mlx5e_tstamp *tstamp)
return tstamp->hwtstamp_config.rx_filter == HWTSTAMP_FILTER_ALL;
}
-static inline int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq,
- struct mlx5e_rx_wqe *wqe, u16 ix)
+int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
{
struct sk_buff *skb;
dma_addr_t dma_addr;
@@ -87,7 +86,7 @@ bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
while (!mlx5_wq_ll_is_full(wq)) {
struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(wq, wq->head);
- if (unlikely(mlx5e_alloc_rx_wqe(rq, wqe, wq->head)))
+ if (unlikely(rq->alloc_wqe(rq, wqe, wq->head)))
break;
mlx5_wq_ll_push(wq, be16_to_cpu(wqe->next.next_wqe_index));
@@ -229,50 +228,55 @@ static inline void mlx5e_build_rx_skb(struct mlx5_cqe64 *cqe,
skb->mark = be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK;
}
+void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
+{
+ struct mlx5e_rx_wqe *wqe;
+ struct sk_buff *skb;
+ __be16 wqe_counter_be;
+ u16 wqe_counter;
+
+ wqe_counter_be = cqe->wqe_counter;
+ wqe_counter = be16_to_cpu(wqe_counter_be);
+ wqe = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
+ skb = rq->skb[wqe_counter];
+ prefetch(skb->data);
+ rq->skb[wqe_counter] = NULL;
+
+ dma_unmap_single(rq->pdev,
+ *((dma_addr_t *)skb->cb),
+ rq->wqe_sz,
+ DMA_FROM_DEVICE);
+
+ if (unlikely((cqe->op_own >> 4) != MLX5_CQE_RESP_SEND)) {
+ rq->stats.wqe_err++;
+ dev_kfree_skb(skb);
+ goto wq_ll_pop;
+ }
+
+ mlx5e_build_rx_skb(cqe, rq, skb);
+ rq->stats.packets++;
+ rq->stats.bytes += be32_to_cpu(cqe->byte_cnt);
+ napi_gro_receive(rq->cq.napi, skb);
+
+wq_ll_pop:
+ mlx5_wq_ll_pop(&rq->wq, wqe_counter_be,
+ &wqe->next.next_wqe_index);
+}
+
int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
{
struct mlx5e_rq *rq = container_of(cq, struct mlx5e_rq, cq);
int work_done;
for (work_done = 0; work_done < budget; work_done++) {
- struct mlx5e_rx_wqe *wqe;
- struct mlx5_cqe64 *cqe;
- struct sk_buff *skb;
- __be16 wqe_counter_be;
- u16 wqe_counter;
+ struct mlx5_cqe64 *cqe = mlx5e_get_cqe(cq);
- cqe = mlx5e_get_cqe(cq);
if (!cqe)
break;
mlx5_cqwq_pop(&cq->wq);
- wqe_counter_be = cqe->wqe_counter;
- wqe_counter = be16_to_cpu(wqe_counter_be);
- wqe = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
- skb = rq->skb[wqe_counter];
- prefetch(skb->data);
- rq->skb[wqe_counter] = NULL;
-
- dma_unmap_single(rq->pdev,
- *((dma_addr_t *)skb->cb),
- rq->wqe_sz,
- DMA_FROM_DEVICE);
-
- if (unlikely((cqe->op_own >> 4) != MLX5_CQE_RESP_SEND)) {
- rq->stats.wqe_err++;
- dev_kfree_skb(skb);
- goto wq_ll_pop;
- }
-
- mlx5e_build_rx_skb(cqe, rq, skb);
- rq->stats.packets++;
- rq->stats.bytes += be32_to_cpu(cqe->byte_cnt);
- napi_gro_receive(cq->napi, skb);
-
-wq_ll_pop:
- mlx5_wq_ll_pop(&rq->wq, wqe_counter_be,
- &wqe->next.next_wqe_index);
+ rq->handle_rx_cqe(rq, cqe);
}
mlx5_cqwq_update_db_record(&cq->wq);
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (3 preceding siblings ...)
2016-04-17 21:31 ` [PATCH net-next V2 04/11] net/mlx5e: Use function pointers for RX data path handling Saeed Mahameed
@ 2016-04-17 21:31 ` Saeed Mahameed
2016-04-18 0:29 ` Eric Dumazet
2016-04-17 21:32 ` [PATCH net-next V2 06/11] net/mlx5e: Added ICO SQs Saeed Mahameed
` (5 subsequent siblings)
10 siblings, 1 reply; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:31 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Achiad Shochat, Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
Introduce the feature of multi-packet WQE (RX Work Queue Element)
referred to as (MPWQE or Striding RQ), in which WQEs are larger
and serve multiple packets each.
Every WQE consists of many strides of the same size, every received
packet is aligned to a beginning of a stride and is written to
consecutive strides within a WQE.
In the regular approach, each regular WQE is big enough to be capable
of serving one received packet of any size up to MTU or 64K in case of
device LRO is enabled, making it very wasteful when dealing with
small packets or device LRO is enabled.
For its flexibility, MPWQE allows a better memory utilization and lower
memory footprint (implying improvements in CPU utilization and packet
rate) as packets consume strides according to their size, preserving
the rest of the WQE to be available for other packets.
MPWQE default configuration:
Num of WQEs = 16
Strides Per WQE = 2048
Stride Size = 64 byte
The default WQEs memory footprint went from 1024*mtu (~1.5MB) to
16 * 2048 * 64 = 2MB per ring.
However, HW LRO can now be supported at no additional cost in memory
footprint, and hence we turn it on by default and get an even better
performance.
Performance tested on ConnectX4-Lx 50G.
To isolate the feature under test, the numbers below were measured with
HW LRO turned off. We verified that the performance just improves when
LRO is turned back on.
* Netperf single TCP stream:
- BW raised by 10-15% for representative packet sizes:
default, 64B, 1024B, 1478B, 65536B.
* Netperf multi TCP stream:
- No degradation, line rate reached.
* Pktgen: packet rate raised by 5-10% for traffic of different message
sizes: 64B, 128B, 256B, 1024B, and 1500B.
* Pktgen: packet loss in bursts of small messages (64byte),
single stream:
- | num packets | packets loss before | packets loss after
| 2K | ~ 1K | 0
| 8K | ~ 6K | 0
| 16K | ~13K | 0
| 32K | ~28K | 0
| 64K | ~57K | ~24K
As expected as the driver can receive as many small packets (<=64B) as
the number of total strides in the ring (default = 2048 * 16) vs. 1024
(default ring size regardless of packets size) before this feature.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 74 +++++++++++-
.../net/ethernet/mellanox/mlx5/core/en_ethtool.c | 15 ++-
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 109 +++++++++++++----
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 128 ++++++++++++++++++--
include/linux/mlx5/device.h | 39 ++++++-
5 files changed, 321 insertions(+), 44 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 61e249d..bb6aa77 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -57,12 +57,27 @@
#define MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE 0xa
#define MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE 0xd
+#define MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE_MPW 0x1
+#define MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE_MPW 0x4
+#define MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE_MPW 0x6
+
+#define MLX5_MPWRQ_LOG_NUM_STRIDES 11 /* >= 9, HW restriction */
+#define MLX5_MPWRQ_LOG_STRIDE_SIZE 6 /* >= 6, HW restriction */
+#define MLX5_MPWRQ_NUM_STRIDES BIT(MLX5_MPWRQ_LOG_NUM_STRIDES)
+#define MLX5_MPWRQ_STRIDE_SIZE BIT(MLX5_MPWRQ_LOG_STRIDE_SIZE)
+#define MLX5_MPWRQ_LOG_WQE_SZ (MLX5_MPWRQ_LOG_NUM_STRIDES +\
+ MLX5_MPWRQ_LOG_STRIDE_SIZE)
+#define MLX5_MPWRQ_WQE_PAGE_ORDER (max_t(int, 0, \
+ MLX5_MPWRQ_LOG_WQE_SZ - PAGE_SHIFT))
+#define MLX5_MPWRQ_SMALL_PACKET_THRESHOLD (128)
+
#define MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ (64 * 1024)
#define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC 0x10
#define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS 0x20
#define MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC 0x10
#define MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS 0x20
#define MLX5E_PARAMS_DEFAULT_MIN_RX_WQES 0x80
+#define MLX5E_PARAMS_DEFAULT_MIN_RX_WQES_MPW 0x2
#define MLX5E_LOG_INDIR_RQT_SIZE 0x7
#define MLX5E_INDIR_RQT_SIZE BIT(MLX5E_LOG_INDIR_RQT_SIZE)
@@ -74,6 +89,38 @@
#define MLX5E_NUM_MAIN_GROUPS 9
#define MLX5E_NET_IP_ALIGN 2
+static inline u16 mlx5_min_rx_wqes(int wq_type, u32 wq_size)
+{
+ switch (wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ return min_t(u16, MLX5E_PARAMS_DEFAULT_MIN_RX_WQES_MPW,
+ wq_size / 2);
+ default:
+ return min_t(u16, MLX5E_PARAMS_DEFAULT_MIN_RX_WQES,
+ wq_size / 2);
+ }
+}
+
+static inline int mlx5_min_log_rq_size(int wq_type)
+{
+ switch (wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ return MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE_MPW;
+ default:
+ return MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE;
+ }
+}
+
+static inline int mlx5_max_log_rq_size(int wq_type)
+{
+ switch (wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ return MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE_MPW;
+ default:
+ return MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE;
+ }
+}
+
struct mlx5e_tx_wqe {
struct mlx5_wqe_ctrl_seg ctrl;
struct mlx5_wqe_eth_seg eth;
@@ -128,6 +175,7 @@ static const char vport_strings[][ETH_GSTRING_LEN] = {
"tx_queue_wake",
"tx_queue_dropped",
"rx_wqe_err",
+ "rx_mpwqe_filler",
};
struct mlx5e_vport_stats {
@@ -169,8 +217,9 @@ struct mlx5e_vport_stats {
u64 tx_queue_wake;
u64 tx_queue_dropped;
u64 rx_wqe_err;
+ u64 rx_mpwqe_filler;
-#define NUM_VPORT_COUNTERS 35
+#define NUM_VPORT_COUNTERS 36
};
static const char pport_strings[][ETH_GSTRING_LEN] = {
@@ -263,7 +312,8 @@ static const char rq_stats_strings[][ETH_GSTRING_LEN] = {
"csum_sw",
"lro_packets",
"lro_bytes",
- "wqe_err"
+ "wqe_err",
+ "mpwqe_filler",
};
struct mlx5e_rq_stats {
@@ -274,7 +324,8 @@ struct mlx5e_rq_stats {
u64 lro_packets;
u64 lro_bytes;
u64 wqe_err;
-#define NUM_RQ_STATS 7
+ u64 mpwqe_filler;
+#define NUM_RQ_STATS 8
};
static const char sq_stats_strings[][ETH_GSTRING_LEN] = {
@@ -318,6 +369,7 @@ struct mlx5e_stats {
struct mlx5e_params {
u8 log_sq_size;
+ u8 rq_wq_type;
u8 log_rq_size;
u16 num_channels;
u8 num_tc;
@@ -374,11 +426,23 @@ typedef void (*mlx5e_fp_handle_rx_cqe)(struct mlx5e_rq *rq,
typedef int (*mlx5e_fp_alloc_wqe)(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe,
u16 ix);
+struct mlx5e_dma_info {
+ struct page *page;
+ dma_addr_t addr;
+};
+
+struct mlx5e_mpw_info {
+ struct mlx5e_dma_info dma_info;
+ u16 consumed_strides;
+ u16 skbs_frags;
+};
+
struct mlx5e_rq {
/* data path */
struct mlx5_wq_ll wq;
u32 wqe_sz;
struct sk_buff **skb;
+ struct mlx5e_mpw_info *wqe_info;
struct device *pdev;
struct net_device *netdev;
@@ -393,6 +457,7 @@ struct mlx5e_rq {
/* control */
struct mlx5_wq_ctrl wq_ctrl;
+ u8 wq_type;
u32 rqn;
struct mlx5e_channel *channel;
struct mlx5e_priv *priv;
@@ -649,9 +714,12 @@ void mlx5e_cq_error_event(struct mlx5_core_cq *mcq, enum mlx5_event event);
int mlx5e_napi_poll(struct napi_struct *napi, int budget);
bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq, int napi_budget);
int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget);
+
void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe);
+void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe);
bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq);
int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix);
+int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix);
struct mlx5_cqe64 *mlx5e_get_cqe(struct mlx5e_cq *cq);
void mlx5e_update_stats(struct mlx5e_priv *priv);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
index 6f40ba4..4077856 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
@@ -273,8 +273,9 @@ static void mlx5e_get_ringparam(struct net_device *dev,
struct ethtool_ringparam *param)
{
struct mlx5e_priv *priv = netdev_priv(dev);
+ int rq_wq_type = priv->params.rq_wq_type;
- param->rx_max_pending = 1 << MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE;
+ param->rx_max_pending = 1 << mlx5_max_log_rq_size(rq_wq_type);
param->tx_max_pending = 1 << MLX5E_PARAMS_MAXIMUM_LOG_SQ_SIZE;
param->rx_pending = 1 << priv->params.log_rq_size;
param->tx_pending = 1 << priv->params.log_sq_size;
@@ -285,6 +286,7 @@ static int mlx5e_set_ringparam(struct net_device *dev,
{
struct mlx5e_priv *priv = netdev_priv(dev);
bool was_opened;
+ int rq_wq_type = priv->params.rq_wq_type;
u16 min_rx_wqes;
u8 log_rq_size;
u8 log_sq_size;
@@ -300,16 +302,16 @@ static int mlx5e_set_ringparam(struct net_device *dev,
__func__);
return -EINVAL;
}
- if (param->rx_pending < (1 << MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE)) {
+ if (param->rx_pending < (1 << mlx5_min_log_rq_size(rq_wq_type))) {
netdev_info(dev, "%s: rx_pending (%d) < min (%d)\n",
__func__, param->rx_pending,
- 1 << MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE);
+ 1 << mlx5_min_log_rq_size(rq_wq_type));
return -EINVAL;
}
- if (param->rx_pending > (1 << MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE)) {
+ if (param->rx_pending > (1 << mlx5_max_log_rq_size(rq_wq_type))) {
netdev_info(dev, "%s: rx_pending (%d) > max (%d)\n",
__func__, param->rx_pending,
- 1 << MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE);
+ 1 << mlx5_max_log_rq_size(rq_wq_type));
return -EINVAL;
}
if (param->tx_pending < (1 << MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE)) {
@@ -327,8 +329,7 @@ static int mlx5e_set_ringparam(struct net_device *dev,
log_rq_size = order_base_2(param->rx_pending);
log_sq_size = order_base_2(param->tx_pending);
- min_rx_wqes = min_t(u16, param->rx_pending - 1,
- MLX5E_PARAMS_DEFAULT_MIN_RX_WQES);
+ min_rx_wqes = mlx5_min_rx_wqes(rq_wq_type, param->rx_pending);
if (log_rq_size == priv->params.log_rq_size &&
log_sq_size == priv->params.log_sq_size &&
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 23ba12c..871f3af 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -175,6 +175,7 @@ void mlx5e_update_stats(struct mlx5e_priv *priv)
s->rx_csum_none = 0;
s->rx_csum_sw = 0;
s->rx_wqe_err = 0;
+ s->rx_mpwqe_filler = 0;
for (i = 0; i < priv->params.num_channels; i++) {
rq_stats = &priv->channel[i]->rq.stats;
@@ -185,6 +186,7 @@ void mlx5e_update_stats(struct mlx5e_priv *priv)
s->rx_csum_none += rq_stats->csum_none;
s->rx_csum_sw += rq_stats->csum_sw;
s->rx_wqe_err += rq_stats->wqe_err;
+ s->rx_mpwqe_filler += rq_stats->mpwqe_filler;
for (j = 0; j < priv->params.num_tc; j++) {
sq_stats = &priv->channel[i]->sq[j].stats;
@@ -323,6 +325,7 @@ static int mlx5e_create_rq(struct mlx5e_channel *c,
struct mlx5_core_dev *mdev = priv->mdev;
void *rqc = param->rqc;
void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
+ u32 byte_count;
int wq_sz;
int err;
int i;
@@ -337,28 +340,47 @@ static int mlx5e_create_rq(struct mlx5e_channel *c,
rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
wq_sz = mlx5_wq_ll_get_size(&rq->wq);
- rq->skb = kzalloc_node(wq_sz * sizeof(*rq->skb), GFP_KERNEL,
- cpu_to_node(c->cpu));
- if (!rq->skb) {
- err = -ENOMEM;
- goto err_rq_wq_destroy;
- }
- rq->wqe_sz = (priv->params.lro_en) ? priv->params.lro_wqe_sz :
- MLX5E_SW2HW_MTU(priv->netdev->mtu);
- rq->wqe_sz = SKB_DATA_ALIGN(rq->wqe_sz + MLX5E_NET_IP_ALIGN);
+ switch (priv->params.rq_wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ rq->wqe_info = kzalloc_node(wq_sz * sizeof(*rq->wqe_info),
+ GFP_KERNEL, cpu_to_node(c->cpu));
+ if (!rq->wqe_info) {
+ err = -ENOMEM;
+ goto err_rq_wq_destroy;
+ }
+ rq->handle_rx_cqe = mlx5e_handle_rx_cqe_mpwrq;
+ rq->alloc_wqe = mlx5e_alloc_rx_mpwqe;
+
+ rq->wqe_sz = MLX5_MPWRQ_NUM_STRIDES * MLX5_MPWRQ_STRIDE_SIZE;
+ byte_count = rq->wqe_sz;
+ break;
+ default: /* MLX5_WQ_TYPE_LINKED_LIST */
+ rq->skb = kzalloc_node(wq_sz * sizeof(*rq->skb), GFP_KERNEL,
+ cpu_to_node(c->cpu));
+ if (!rq->skb) {
+ err = -ENOMEM;
+ goto err_rq_wq_destroy;
+ }
+ rq->handle_rx_cqe = mlx5e_handle_rx_cqe;
+ rq->alloc_wqe = mlx5e_alloc_rx_wqe;
+
+ rq->wqe_sz = (priv->params.lro_en) ?
+ priv->params.lro_wqe_sz :
+ MLX5E_SW2HW_MTU(priv->netdev->mtu);
+ rq->wqe_sz = SKB_DATA_ALIGN(rq->wqe_sz + MLX5E_NET_IP_ALIGN);
+ byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
+ byte_count |= MLX5_HW_START_PADDING;
+ }
for (i = 0; i < wq_sz; i++) {
struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
- u32 byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
wqe->data.lkey = c->mkey_be;
- wqe->data.byte_count =
- cpu_to_be32(byte_count | MLX5_HW_START_PADDING);
+ wqe->data.byte_count = cpu_to_be32(byte_count);
}
- rq->handle_rx_cqe = mlx5e_handle_rx_cqe;
- rq->alloc_wqe = mlx5e_alloc_rx_wqe;
+ rq->wq_type = priv->params.rq_wq_type;
rq->pdev = c->pdev;
rq->netdev = c->netdev;
rq->tstamp = &priv->tstamp;
@@ -376,7 +398,14 @@ err_rq_wq_destroy:
static void mlx5e_destroy_rq(struct mlx5e_rq *rq)
{
- kfree(rq->skb);
+ switch (rq->wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ kfree(rq->wqe_info);
+ break;
+ default: /* MLX5_WQ_TYPE_LINKED_LIST */
+ kfree(rq->skb);
+ }
+
mlx5_wq_destroy(&rq->wq_ctrl);
}
@@ -1065,7 +1094,18 @@ static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
void *rqc = param->rqc;
void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
- MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
+ switch (priv->params.rq_wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ MLX5_SET(wq, wq, log_wqe_num_of_strides,
+ MLX5_MPWRQ_LOG_NUM_STRIDES - 9);
+ MLX5_SET(wq, wq, log_wqe_stride_size,
+ MLX5_MPWRQ_LOG_STRIDE_SIZE - 6);
+ MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ);
+ break;
+ default: /* MLX5_WQ_TYPE_LINKED_LIST */
+ MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
+ }
+
MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size);
@@ -1111,8 +1151,18 @@ static void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
struct mlx5e_cq_param *param)
{
void *cqc = param->cqc;
+ u8 log_cq_size;
- MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size);
+ switch (priv->params.rq_wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ log_cq_size = priv->params.log_rq_size +
+ MLX5_MPWRQ_LOG_NUM_STRIDES;
+ break;
+ default: /* MLX5_WQ_TYPE_LINKED_LIST */
+ log_cq_size = priv->params.log_rq_size;
+ }
+
+ MLX5_SET(cqc, cqc, log_cq_size, log_cq_size);
mlx5e_build_common_cq_param(priv, param);
}
@@ -1983,7 +2033,8 @@ static int mlx5e_set_features(struct net_device *netdev,
if (changes & NETIF_F_LRO) {
bool was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
- if (was_opened)
+ if (was_opened && (priv->params.rq_wq_type ==
+ MLX5_WQ_TYPE_LINKED_LIST))
mlx5e_close_locked(priv->netdev);
priv->params.lro_en = !!(features & NETIF_F_LRO);
@@ -1992,7 +2043,8 @@ static int mlx5e_set_features(struct net_device *netdev,
mlx5_core_warn(priv->mdev, "lro modify failed, %d\n",
err);
- if (was_opened)
+ if (was_opened && (priv->params.rq_wq_type ==
+ MLX5_WQ_TYPE_LINKED_LIST))
err = mlx5e_open_locked(priv->netdev);
}
@@ -2327,8 +2379,21 @@ static void mlx5e_build_netdev_priv(struct mlx5_core_dev *mdev,
priv->params.log_sq_size =
MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
- priv->params.log_rq_size =
- MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
+ priv->params.rq_wq_type = MLX5_CAP_GEN(mdev, striding_rq) ?
+ MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ :
+ MLX5_WQ_TYPE_LINKED_LIST;
+
+ switch (priv->params.rq_wq_type) {
+ case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+ priv->params.log_rq_size = MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE_MPW;
+ priv->params.lro_en = true;
+ break;
+ default: /* MLX5_WQ_TYPE_LINKED_LIST */
+ priv->params.log_rq_size = MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
+ }
+
+ priv->params.min_rx_wqes = mlx5_min_rx_wqes(priv->params.rq_wq_type,
+ BIT(priv->params.log_rq_size));
priv->params.rx_cq_moderation_usec =
MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
priv->params.rx_cq_moderation_pkts =
@@ -2338,8 +2403,6 @@ static void mlx5e_build_netdev_priv(struct mlx5_core_dev *mdev,
priv->params.tx_cq_moderation_pkts =
MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev);
- priv->params.min_rx_wqes =
- MLX5E_PARAMS_DEFAULT_MIN_RX_WQES;
priv->params.num_tc = 1;
priv->params.rss_hfunc = ETH_RSS_HASH_XOR;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index d7ccced..e652604 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -76,6 +76,34 @@ err_free_skb:
return -ENOMEM;
}
+int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
+{
+ struct mlx5e_mpw_info *wi = &rq->wqe_info[ix];
+ int ret = 0;
+
+ wi->dma_info.page = dev_alloc_pages(MLX5_MPWRQ_WQE_PAGE_ORDER);
+ if (unlikely(!wi->dma_info.page))
+ return -ENOMEM;
+
+ wi->dma_info.addr = dma_map_page(rq->pdev, wi->dma_info.page, 0,
+ rq->wqe_sz, PCI_DMA_FROMDEVICE);
+ if (unlikely(dma_mapping_error(rq->pdev, wi->dma_info.addr))) {
+ ret = -ENOMEM;
+ goto err_put_page;
+ }
+
+ atomic_add(MLX5_MPWRQ_NUM_STRIDES, &wi->dma_info.page->_count);
+ wi->skbs_frags = 0;
+ wi->consumed_strides = 0;
+ wqe->data.addr = cpu_to_be64(wi->dma_info.addr);
+
+ return 0;
+
+err_put_page:
+ put_page(wi->dma_info.page);
+ return ret;
+}
+
bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
{
struct mlx5_wq_ll *wq = &rq->wq;
@@ -100,7 +128,8 @@ bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
return !mlx5_wq_ll_is_full(wq);
}
-static void mlx5e_lro_update_hdr(struct sk_buff *skb, struct mlx5_cqe64 *cqe)
+static void mlx5e_lro_update_hdr(struct sk_buff *skb, struct mlx5_cqe64 *cqe,
+ u32 cqe_bcnt)
{
struct ethhdr *eth = (struct ethhdr *)(skb->data);
struct iphdr *ipv4 = (struct iphdr *)(skb->data + ETH_HLEN);
@@ -111,7 +140,7 @@ static void mlx5e_lro_update_hdr(struct sk_buff *skb, struct mlx5_cqe64 *cqe)
int tcp_ack = ((CQE_L4_HDR_TYPE_TCP_ACK_NO_DATA == l4_hdr_type) ||
(CQE_L4_HDR_TYPE_TCP_ACK_AND_DATA == l4_hdr_type));
- u16 tot_len = be32_to_cpu(cqe->byte_cnt) - ETH_HLEN;
+ u16 tot_len = cqe_bcnt - ETH_HLEN;
if (eth->h_proto == htons(ETH_P_IP)) {
tcp = (struct tcphdr *)(skb->data + ETH_HLEN +
@@ -191,19 +220,17 @@ csum_none:
}
static inline void mlx5e_build_rx_skb(struct mlx5_cqe64 *cqe,
+ u32 cqe_bcnt,
struct mlx5e_rq *rq,
struct sk_buff *skb)
{
struct net_device *netdev = rq->netdev;
- u32 cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
struct mlx5e_tstamp *tstamp = rq->tstamp;
int lro_num_seg;
- skb_put(skb, cqe_bcnt);
-
lro_num_seg = be32_to_cpu(cqe->srqn) >> 24;
if (lro_num_seg > 1) {
- mlx5e_lro_update_hdr(skb, cqe);
+ mlx5e_lro_update_hdr(skb, cqe, cqe_bcnt);
skb_shinfo(skb)->gso_size = DIV_ROUND_UP(cqe_bcnt, lro_num_seg);
rq->stats.lro_packets++;
rq->stats.lro_bytes += cqe_bcnt;
@@ -228,12 +255,24 @@ static inline void mlx5e_build_rx_skb(struct mlx5_cqe64 *cqe,
skb->mark = be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK;
}
+static inline void mlx5e_complete_rx_cqe(struct mlx5e_rq *rq,
+ struct mlx5_cqe64 *cqe,
+ u32 cqe_bcnt,
+ struct sk_buff *skb)
+{
+ rq->stats.packets++;
+ rq->stats.bytes += cqe_bcnt;
+ mlx5e_build_rx_skb(cqe, cqe_bcnt, rq, skb);
+ napi_gro_receive(rq->cq.napi, skb);
+}
+
void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
{
struct mlx5e_rx_wqe *wqe;
struct sk_buff *skb;
__be16 wqe_counter_be;
u16 wqe_counter;
+ u32 cqe_bcnt;
wqe_counter_be = cqe->wqe_counter;
wqe_counter = be16_to_cpu(wqe_counter_be);
@@ -253,16 +292,85 @@ void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
goto wq_ll_pop;
}
- mlx5e_build_rx_skb(cqe, rq, skb);
- rq->stats.packets++;
- rq->stats.bytes += be32_to_cpu(cqe->byte_cnt);
- napi_gro_receive(rq->cq.napi, skb);
+ cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
+ skb_put(skb, cqe_bcnt);
+
+ mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
wq_ll_pop:
mlx5_wq_ll_pop(&rq->wq, wqe_counter_be,
&wqe->next.next_wqe_index);
}
+void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
+{
+ u16 cstrides = mpwrq_get_cqe_consumed_strides(cqe);
+ u16 stride_ix = mpwrq_get_cqe_stride_index(cqe);
+ u32 consumed_bytes = cstrides * MLX5_MPWRQ_STRIDE_SIZE;
+ u32 data_offset = stride_ix * MLX5_MPWRQ_STRIDE_SIZE;
+ u16 wqe_id = be16_to_cpu(cqe->wqe_id);
+ struct mlx5e_mpw_info *wi = &rq->wqe_info[wqe_id];
+ struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, wqe_id);
+ struct sk_buff *skb;
+ u16 byte_cnt;
+ u16 cqe_bcnt;
+ u16 headlen;
+
+ wi->consumed_strides += cstrides;
+
+ if (unlikely((cqe->op_own >> 4) != MLX5_CQE_RESP_SEND)) {
+ rq->stats.wqe_err++;
+ goto mpwrq_cqe_out;
+ }
+
+ if (unlikely(mpwrq_is_filler_cqe(cqe))) {
+ rq->stats.mpwqe_filler++;
+ goto mpwrq_cqe_out;
+ }
+
+ skb = netdev_alloc_skb(rq->netdev,
+ ALIGN(MLX5_MPWRQ_SMALL_PACKET_THRESHOLD,
+ sizeof(long)));
+ if (unlikely(!skb))
+ goto mpwrq_cqe_out;
+
+ prefetch(skb->data);
+ dma_sync_single_for_cpu(rq->pdev, wi->dma_info.addr + data_offset,
+ consumed_bytes, DMA_FROM_DEVICE);
+
+ cqe_bcnt = mpwrq_get_cqe_byte_cnt(cqe);
+ headlen = min_t(u16, MLX5_MPWRQ_SMALL_PACKET_THRESHOLD, cqe_bcnt);
+
+ byte_cnt = cqe_bcnt - headlen;
+ if (byte_cnt) {
+ wi->skbs_frags++;
+ skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
+ wi->dma_info.page, data_offset + headlen,
+ byte_cnt,
+ ALIGN(byte_cnt, MLX5_MPWRQ_STRIDE_SIZE));
+ }
+
+ skb_copy_to_linear_data(skb,
+ page_address(wi->dma_info.page) + data_offset,
+ ALIGN(headlen, sizeof(long)));
+ /* skb linear part was allocated with headlen and aligned to long */
+ skb->tail += headlen;
+ skb->len += headlen;
+
+ mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
+
+mpwrq_cqe_out:
+ if (likely(wi->consumed_strides < MLX5_MPWRQ_NUM_STRIDES))
+ return;
+
+ dma_unmap_page(rq->pdev, wi->dma_info.addr, rq->wqe_sz,
+ PCI_DMA_FROMDEVICE);
+ atomic_sub(MLX5_MPWRQ_NUM_STRIDES - wi->skbs_frags,
+ &wi->dma_info.page->_count);
+ put_page(wi->dma_info.page);
+ mlx5_wq_ll_pop(&rq->wq, cqe->wqe_id, &wqe->next.next_wqe_index);
+}
+
int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
{
struct mlx5e_rq *rq = container_of(cq, struct mlx5e_rq, cq);
diff --git a/include/linux/mlx5/device.h b/include/linux/mlx5/device.h
index 8156e3c..03f8d71 100644
--- a/include/linux/mlx5/device.h
+++ b/include/linux/mlx5/device.h
@@ -644,7 +644,8 @@ struct mlx5_err_cqe {
};
struct mlx5_cqe64 {
- u8 rsvd0[4];
+ u8 rsvd0[2];
+ __be16 wqe_id;
u8 lro_tcppsh_abort_dupack;
u8 lro_min_ttl;
__be16 lro_tcp_win;
@@ -696,6 +697,42 @@ static inline u64 get_cqe_ts(struct mlx5_cqe64 *cqe)
return (u64)lo | ((u64)hi << 32);
}
+struct mpwrq_cqe_bc {
+ __be16 filler_consumed_strides;
+ __be16 byte_cnt;
+};
+
+static inline u16 mpwrq_get_cqe_byte_cnt(struct mlx5_cqe64 *cqe)
+{
+ struct mpwrq_cqe_bc *bc = (struct mpwrq_cqe_bc *)&cqe->byte_cnt;
+
+ return be16_to_cpu(bc->byte_cnt);
+}
+
+static inline u16 mpwrq_get_cqe_bc_consumed_strides(struct mpwrq_cqe_bc *bc)
+{
+ return 0x7fff & be16_to_cpu(bc->filler_consumed_strides);
+}
+
+static inline u16 mpwrq_get_cqe_consumed_strides(struct mlx5_cqe64 *cqe)
+{
+ struct mpwrq_cqe_bc *bc = (struct mpwrq_cqe_bc *)&cqe->byte_cnt;
+
+ return mpwrq_get_cqe_bc_consumed_strides(bc);
+}
+
+static inline bool mpwrq_is_filler_cqe(struct mlx5_cqe64 *cqe)
+{
+ struct mpwrq_cqe_bc *bc = (struct mpwrq_cqe_bc *)&cqe->byte_cnt;
+
+ return 0x8000 & be16_to_cpu(bc->filler_consumed_strides);
+}
+
+static inline u16 mpwrq_get_cqe_stride_index(struct mlx5_cqe64 *cqe)
+{
+ return be16_to_cpu(cqe->wqe_counter);
+}
+
enum {
CQE_L4_HDR_TYPE_NONE = 0x0,
CQE_L4_HDR_TYPE_TCP_NO_ACK = 0x1,
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-17 21:31 ` [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ) Saeed Mahameed
@ 2016-04-18 0:29 ` Eric Dumazet
2016-04-18 12:48 ` Eric Dumazet
0 siblings, 1 reply; 20+ messages in thread
From: Eric Dumazet @ 2016-04-18 0:29 UTC (permalink / raw)
To: Saeed Mahameed
Cc: David S. Miller, netdev, Or Gerlitz, Tal Alon, Tariq Toukan,
Eran Ben Elisha, Achiad Shochat
On Mon, 2016-04-18 at 00:31 +0300, Saeed Mahameed wrote:
> Performance tested on ConnectX4-Lx 50G.
> To isolate the feature under test, the numbers below were measured with
> HW LRO turned off. We verified that the performance just improves when
> LRO is turned back on.
>
> * Netperf single TCP stream:
> - BW raised by 10-15% for representative packet sizes:
> default, 64B, 1024B, 1478B, 65536B.
>
> * Netperf multi TCP stream:
> - No degradation, line rate reached.
>
> * Pktgen: packet rate raised by 5-10% for traffic of different message
> sizes: 64B, 128B, 256B, 1024B, and 1500B.
>
> * Pktgen: packet loss in bursts of small messages (64byte),
> single stream:
> - | num packets | packets loss before | packets loss after
> | 2K | ~ 1K | 0
> | 8K | ~ 6K | 0
> | 16K | ~13K | 0
> | 32K | ~28K | 0
> | 64K | ~57K | ~24K
As I already mentioned, allocated order-5 pages and hoping host only
receives friendly traffic is very optimistic.
A 192 bytes frame, is claiming to consume 192 bytes frag with your new
allocation strategy. (skb->truesize is kind of minimal)
In reality, it can prevent a whole 131072 bytes of memory from being
reclaimed/freed. TCP stack will not consider such skb has a candidate
for collapsing in case of memory pressure or hostile peer.
Your tests are obviously run on a freshly booted host, where all
physical memory can be consumed for networking buffers.
Even with order-3 pages, we have problems (at Facebook and Google) on
hosts that we do not reboot every day.
At the time order-5 allocations fail, it is already too late, as maybe
thousands of out-of-order TCP packets might have consumed all the memory
and the host will die.
/proc/sys/net/ipv4/tcp_mem by default allows TCP to use up to 10% of
hysical memory, assuming skb->truesize is true.
In your schem, TCP might never notice it uses 100% of the ram for
packets stored in out or order queues, since a frag will hold 32 times
more pages than really announced.
If really you need to allocate physically contiguous memory, have you
considered converting the order-5 pages into 32 order-0 ones ?
This way, a 192 bytes frame sitting in one socket would hold one order-0
page in the worst case, and TCP wont be allowed to use all physical
memory.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-18 0:29 ` Eric Dumazet
@ 2016-04-18 12:48 ` Eric Dumazet
2016-04-18 13:05 ` Saeed Mahameed
0 siblings, 1 reply; 20+ messages in thread
From: Eric Dumazet @ 2016-04-18 12:48 UTC (permalink / raw)
To: Saeed Mahameed
Cc: David S. Miller, netdev, Or Gerlitz, Tal Alon, Tariq Toukan,
Eran Ben Elisha, Achiad Shochat
On Sun, 2016-04-17 at 17:29 -0700, Eric Dumazet wrote:
>
> If really you need to allocate physically contiguous memory, have you
> considered converting the order-5 pages into 32 order-0 ones ?
Search for split_page() call sites for examples.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-18 12:48 ` Eric Dumazet
@ 2016-04-18 13:05 ` Saeed Mahameed
2016-04-18 14:17 ` Eric Dumazet
0 siblings, 1 reply; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-18 13:05 UTC (permalink / raw)
To: Eric Dumazet
Cc: Saeed Mahameed, David S. Miller, Linux Netdev List, Or Gerlitz,
Tal Alon, Tariq Toukan, Eran Ben Elisha, Achiad Shochat
On Mon, Apr 18, 2016 at 3:48 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> On Sun, 2016-04-17 at 17:29 -0700, Eric Dumazet wrote:
>
>>
>> If really you need to allocate physically contiguous memory, have you
>> considered converting the order-5 pages into 32 order-0 ones ?
>
> Search for split_page() call sites for examples.
>
>
Thanks Eric, we are already evaluating split_page as we speak.
We did look but could not find any specific alloc_pages API that
allocates many physically contiguous pages with order0 ! so we assume
it is ok to use split_page.
BTW our MPWQE solution doesn't totally rely on huge physically
contiguous memory, as you see in the next two patches we introduce a
fragmented MPWQE approach as a fallback, but we do understand your
concern for the normal flow.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-18 13:05 ` Saeed Mahameed
@ 2016-04-18 14:17 ` Eric Dumazet
2016-04-19 16:25 ` Jesper Dangaard Brouer
0 siblings, 1 reply; 20+ messages in thread
From: Eric Dumazet @ 2016-04-18 14:17 UTC (permalink / raw)
To: Saeed Mahameed
Cc: Saeed Mahameed, David S. Miller, Linux Netdev List, Or Gerlitz,
Tal Alon, Tariq Toukan, Eran Ben Elisha, Achiad Shochat
On Mon, 2016-04-18 at 16:05 +0300, Saeed Mahameed wrote:
> On Mon, Apr 18, 2016 at 3:48 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > On Sun, 2016-04-17 at 17:29 -0700, Eric Dumazet wrote:
> >
> >>
> >> If really you need to allocate physically contiguous memory, have you
> >> considered converting the order-5 pages into 32 order-0 ones ?
> >
> > Search for split_page() call sites for examples.
> >
> >
>
> Thanks Eric, we are already evaluating split_page as we speak.
>
> We did look but could not find any specific alloc_pages API that
> allocates many physically contiguous pages with order0 ! so we assume
> it is ok to use split_page.
Note: I have no idea of split_page() performance :
Buddy page allocator has to aggregate pages into order-5, then we would
undo the work, touching 32 cache lines.
You might first benchmark a simple loop doing
loop 10,000,000 times
Order-5 allocation
split into 32 order-0
free 32 pages
Another idea would be to have a way to control max number of order-5
pages that a port would be using.
Since driver always own a ref on a order-5 pages, idea would be to
maintain a circular ring of up to XXX such pages, so that we can detect
an abnormal use and fallback to order-0 immediately.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-18 14:17 ` Eric Dumazet
@ 2016-04-19 16:25 ` Jesper Dangaard Brouer
2016-04-19 17:39 ` Mel Gorman
2016-04-19 18:30 ` Saeed Mahameed
0 siblings, 2 replies; 20+ messages in thread
From: Jesper Dangaard Brouer @ 2016-04-19 16:25 UTC (permalink / raw)
To: Eric Dumazet
Cc: brouer, Saeed Mahameed, Saeed Mahameed, David S. Miller,
Linux Netdev List, Or Gerlitz, Tal Alon, Tariq Toukan,
Eran Ben Elisha, Achiad Shochat, Mel Gorman, linux-mm
On Mon, 18 Apr 2016 07:17:13 -0700
Eric Dumazet <eric.dumazet@gmail.com> wrote:
> On Mon, 2016-04-18 at 16:05 +0300, Saeed Mahameed wrote:
> > On Mon, Apr 18, 2016 at 3:48 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > > On Sun, 2016-04-17 at 17:29 -0700, Eric Dumazet wrote:
> > >
> > >>
> > >> If really you need to allocate physically contiguous memory, have you
> > >> considered converting the order-5 pages into 32 order-0 ones ?
> > >
> > > Search for split_page() call sites for examples.
> > >
> > >
> >
> > Thanks Eric, we are already evaluating split_page as we speak.
> >
> > We did look but could not find any specific alloc_pages API that
> > allocates many physically contiguous pages with order0 ! so we assume
> > it is ok to use split_page.
>
> Note: I have no idea of split_page() performance :
Maybe Mel knows? And maybe Mel have an opinion about if this is a good
or bad approach, e.g. will this approach stress the page allocator in a
bad way?
> Buddy page allocator has to aggregate pages into order-5, then we would
> undo the work, touching 32 cache lines.
>
> You might first benchmark a simple loop doing
>
> loop 10,000,000 times
> Order-5 allocation
> split into 32 order-0
> free 32 pages
>
>
> Another idea would be to have a way to control max number of order-5
> pages that a port would be using.
>
> Since driver always own a ref on a order-5 pages, idea would be to
> maintain a circular ring of up to XXX such pages, so that we can detect
> an abnormal use and fallback to order-0 immediately.
That is part of my idea with my page-pool proposal. In the page-pool I
want to have some watermark counter that can block/stop the OOM issue at
this RX ring level.
See slide 12 of presentation:
http://people.netfilter.org/hawk/presentations/MM-summit2016/generic_page_pool_mm_summit2016.pdf
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
Author of http://www.iptv-analyzer.org
LinkedIn: http://www.linkedin.com/in/brouer
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-19 16:25 ` Jesper Dangaard Brouer
@ 2016-04-19 17:39 ` Mel Gorman
2016-04-20 16:46 ` Saeed Mahameed
2016-04-19 18:30 ` Saeed Mahameed
1 sibling, 1 reply; 20+ messages in thread
From: Mel Gorman @ 2016-04-19 17:39 UTC (permalink / raw)
To: Jesper Dangaard Brouer
Cc: Eric Dumazet, Saeed Mahameed, Saeed Mahameed, David S. Miller,
Linux Netdev List, Or Gerlitz, Tal Alon, Tariq Toukan,
Eran Ben Elisha, Achiad Shochat, linux-mm
On Tue, Apr 19, 2016 at 06:25:32PM +0200, Jesper Dangaard Brouer wrote:
> On Mon, 18 Apr 2016 07:17:13 -0700
> Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> > On Mon, 2016-04-18 at 16:05 +0300, Saeed Mahameed wrote:
> > > On Mon, Apr 18, 2016 at 3:48 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > > > On Sun, 2016-04-17 at 17:29 -0700, Eric Dumazet wrote:
> > > >
> > > >>
> > > >> If really you need to allocate physically contiguous memory, have you
> > > >> considered converting the order-5 pages into 32 order-0 ones ?
> > > >
> > > > Search for split_page() call sites for examples.
> > > >
> > > >
> > >
> > > Thanks Eric, we are already evaluating split_page as we speak.
> > >
> > > We did look but could not find any specific alloc_pages API that
alloc_pages_exact()
> > > allocates many physically contiguous pages with order0 ! so we assume
> > > it is ok to use split_page.
> >
> > Note: I have no idea of split_page() performance :
>
> Maybe Mel knows?
Irrelevant in comparison to the cost of allocating an order-5 pages if
one is not already available.
> And maybe Mel have an opinion about if this is a good
> or bad approach, e.g. will this approach stress the page allocator in a
> bad way?
>
It'll contend on the zone lock minimally but again, irrelevant in
comparison to having to reclaim/compact an order-5 page if one is not
already free.
It'll appear to work well in benchmarks and then fall apart when the
system is running for long enough.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-19 17:39 ` Mel Gorman
@ 2016-04-20 16:46 ` Saeed Mahameed
0 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-20 16:46 UTC (permalink / raw)
To: Mel Gorman
Cc: Jesper Dangaard Brouer, Eric Dumazet, Saeed Mahameed,
David S. Miller, Linux Netdev List, Or Gerlitz, Tal Alon,
Tariq Toukan, Eran Ben Elisha, Achiad Shochat, linux-mm
On Tue, Apr 19, 2016 at 8:39 PM, Mel Gorman <mgorman@techsingularity.net> wrote:
> On Tue, Apr 19, 2016 at 06:25:32PM +0200, Jesper Dangaard Brouer wrote:
>> On Mon, 18 Apr 2016 07:17:13 -0700
>> Eric Dumazet <eric.dumazet@gmail.com> wrote:
>>
>
> alloc_pages_exact()
>
We want to allocate 32 order-0 physically contiguous pages and to free
each one of them individually.
the documentation states "Memory allocated by this function must be
released by free_pages_exact()"
Also it returns a pointer to the memory and we need pointers to pages.
>> > > allocates many physically contiguous pages with order0 ! so we assume
>> > > it is ok to use split_page.
>> >
>> > Note: I have no idea of split_page() performance :
>>
>> Maybe Mel knows?
>
> Irrelevant in comparison to the cost of allocating an order-5 pages if
> one is not already available.
>
we still allocate order-5 pages but now we split them to 32 order-0 pages.
the split adds extra few cpu cycles but it is lookless and
straightforward, and it does the job in terms of better memory
utilization.
now in scenarios where small packets can hold a ref on pages for too
long they would hold a ref on order-0 pages rather than order-5.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ)
2016-04-19 16:25 ` Jesper Dangaard Brouer
2016-04-19 17:39 ` Mel Gorman
@ 2016-04-19 18:30 ` Saeed Mahameed
1 sibling, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-19 18:30 UTC (permalink / raw)
To: Jesper Dangaard Brouer
Cc: Eric Dumazet, Saeed Mahameed, David S. Miller, Linux Netdev List,
Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Achiad Shochat, Mel Gorman, linux-mm
On Tue, Apr 19, 2016 at 7:25 PM, Jesper Dangaard Brouer
<brouer@redhat.com> wrote:
> On Mon, 18 Apr 2016 07:17:13 -0700
> Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
>> Another idea would be to have a way to control max number of order-5
>> pages that a port would be using.
>>
>> Since driver always own a ref on a order-5 pages, idea would be to
>> maintain a circular ring of up to XXX such pages, so that we can detect
>> an abnormal use and fallback to order-0 immediately.
>
> That is part of my idea with my page-pool proposal. In the page-pool I
> want to have some watermark counter that can block/stop the OOM issue at
> this RX ring level.
>
> See slide 12 of presentation:
> http://people.netfilter.org/hawk/presentations/MM-summit2016/generic_page_pool_mm_summit2016.pdf
>
Cool Idea guys, and we already tested our own version of it,
we tried to recycle our own driver pages but we saw that the stack
took too long to release them, we had to work with 2X and sometimes 4X
pages pool per ring to be able to reuse recycled pages on every RX
packet on 50Gb line rate, but we dropped the Idea since 2X is too
much.
but definitely, this is the best way to go for all drivers, reusing
already dma mapped pages and significantly reducing dma operations for
the driver is a big win !
we are still considering such option as future optimization.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 20+ messages in thread
* [PATCH net-next V2 06/11] net/mlx5e: Added ICO SQs
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (4 preceding siblings ...)
2016-04-17 21:31 ` [PATCH net-next V2 05/11] net/mlx5e: Support RX multi-packet WQE (Striding RQ) Saeed Mahameed
@ 2016-04-17 21:32 ` Saeed Mahameed
2016-04-17 21:32 ` [PATCH net-next V2 07/11] net/mlx5e: Add fragmented memory support for RX multi packet WQE Saeed Mahameed
` (4 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:32 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
Added ICO (Internal Control Operations) SQ per channel to be used
for driver internal operations such as memory registration for
fragmented memory and nop requests upon ifconfig up.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 7 +
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 135 +++++++++++++++++----
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 55 +++++++++
4 files changed, 174 insertions(+), 25 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index bb6aa77..f62495f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -485,6 +485,11 @@ enum {
MLX5E_SQ_STATE_BF_ENABLE,
};
+struct mlx5e_ico_wqe_info {
+ u8 opcode;
+ u8 num_wqebbs;
+};
+
struct mlx5e_sq {
/* data path */
@@ -526,6 +531,7 @@ struct mlx5e_sq {
struct mlx5_uar uar;
struct mlx5e_channel *channel;
int tc;
+ struct mlx5e_ico_wqe_info *ico_wqe_info;
} ____cacheline_aligned_in_smp;
static inline bool mlx5e_sq_has_room_for(struct mlx5e_sq *sq, u16 n)
@@ -542,6 +548,7 @@ struct mlx5e_channel {
/* data path */
struct mlx5e_rq rq;
struct mlx5e_sq sq[MLX5E_MAX_NUM_TC];
+ struct mlx5e_sq icosq; /* internal control operations */
struct napi_struct napi;
struct device *pdev;
struct net_device *netdev;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 871f3af..b25b429 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -48,6 +48,7 @@ struct mlx5e_sq_param {
u32 sqc[MLX5_ST_SZ_DW(sqc)];
struct mlx5_wq_param wq;
u16 max_inline;
+ bool icosq;
};
struct mlx5e_cq_param {
@@ -59,8 +60,10 @@ struct mlx5e_cq_param {
struct mlx5e_channel_param {
struct mlx5e_rq_param rq;
struct mlx5e_sq_param sq;
+ struct mlx5e_sq_param icosq;
struct mlx5e_cq_param rx_cq;
struct mlx5e_cq_param tx_cq;
+ struct mlx5e_cq_param icosq_cq;
};
static void mlx5e_update_carrier(struct mlx5e_priv *priv)
@@ -502,6 +505,8 @@ static int mlx5e_open_rq(struct mlx5e_channel *c,
struct mlx5e_rq_param *param,
struct mlx5e_rq *rq)
{
+ struct mlx5e_sq *sq = &c->icosq;
+ u16 pi = sq->pc & sq->wq.sz_m1;
int err;
err = mlx5e_create_rq(c, param, rq);
@@ -517,7 +522,10 @@ static int mlx5e_open_rq(struct mlx5e_channel *c,
goto err_disable_rq;
set_bit(MLX5E_RQ_STATE_POST_WQES_ENABLE, &rq->state);
- mlx5e_send_nop(&c->sq[0], true); /* trigger mlx5e_post_rx_wqes() */
+
+ sq->ico_wqe_info[pi].opcode = MLX5_OPCODE_NOP;
+ sq->ico_wqe_info[pi].num_wqebbs = 1;
+ mlx5e_send_nop(sq, true); /* trigger mlx5e_post_rx_wqes() */
return 0;
@@ -583,7 +591,6 @@ static int mlx5e_create_sq(struct mlx5e_channel *c,
void *sqc = param->sqc;
void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq);
- int txq_ix;
int err;
err = mlx5_alloc_map_uar(mdev, &sq->uar, true);
@@ -611,8 +618,24 @@ static int mlx5e_create_sq(struct mlx5e_channel *c,
if (err)
goto err_sq_wq_destroy;
- txq_ix = c->ix + tc * priv->params.num_channels;
- sq->txq = netdev_get_tx_queue(priv->netdev, txq_ix);
+ if (param->icosq) {
+ u8 wq_sz = mlx5_wq_cyc_get_size(&sq->wq);
+
+ sq->ico_wqe_info = kzalloc_node(sizeof(*sq->ico_wqe_info) *
+ wq_sz,
+ GFP_KERNEL,
+ cpu_to_node(c->cpu));
+ if (!sq->ico_wqe_info) {
+ err = -ENOMEM;
+ goto err_free_sq_db;
+ }
+ } else {
+ int txq_ix;
+
+ txq_ix = c->ix + tc * priv->params.num_channels;
+ sq->txq = netdev_get_tx_queue(priv->netdev, txq_ix);
+ priv->txq_to_sq_map[txq_ix] = sq;
+ }
sq->pdev = c->pdev;
sq->tstamp = &priv->tstamp;
@@ -621,10 +644,12 @@ static int mlx5e_create_sq(struct mlx5e_channel *c,
sq->tc = tc;
sq->edge = (sq->wq.sz_m1 + 1) - MLX5_SEND_WQE_MAX_WQEBBS;
sq->bf_budget = MLX5E_SQ_BF_BUDGET;
- priv->txq_to_sq_map[txq_ix] = sq;
return 0;
+err_free_sq_db:
+ mlx5e_free_sq_db(sq);
+
err_sq_wq_destroy:
mlx5_wq_destroy(&sq->wq_ctrl);
@@ -639,6 +664,7 @@ static void mlx5e_destroy_sq(struct mlx5e_sq *sq)
struct mlx5e_channel *c = sq->channel;
struct mlx5e_priv *priv = c->priv;
+ kfree(sq->ico_wqe_info);
mlx5e_free_sq_db(sq);
mlx5_wq_destroy(&sq->wq_ctrl);
mlx5_unmap_free_uar(priv->mdev, &sq->uar);
@@ -667,10 +693,10 @@ static int mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param)
memcpy(sqc, param->sqc, sizeof(param->sqc));
- MLX5_SET(sqc, sqc, tis_num_0, priv->tisn[sq->tc]);
- MLX5_SET(sqc, sqc, cqn, c->sq[sq->tc].cq.mcq.cqn);
+ MLX5_SET(sqc, sqc, tis_num_0, param->icosq ? 0 : priv->tisn[sq->tc]);
+ MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn);
MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST);
- MLX5_SET(sqc, sqc, tis_lst_sz, 1);
+ MLX5_SET(sqc, sqc, tis_lst_sz, param->icosq ? 0 : 1);
MLX5_SET(sqc, sqc, flush_in_error_en, 1);
MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
@@ -745,9 +771,11 @@ static int mlx5e_open_sq(struct mlx5e_channel *c,
if (err)
goto err_disable_sq;
- set_bit(MLX5E_SQ_STATE_WAKE_TXQ_ENABLE, &sq->state);
- netdev_tx_reset_queue(sq->txq);
- netif_tx_start_queue(sq->txq);
+ if (sq->txq) {
+ set_bit(MLX5E_SQ_STATE_WAKE_TXQ_ENABLE, &sq->state);
+ netdev_tx_reset_queue(sq->txq);
+ netif_tx_start_queue(sq->txq);
+ }
return 0;
@@ -768,15 +796,19 @@ static inline void netif_tx_disable_queue(struct netdev_queue *txq)
static void mlx5e_close_sq(struct mlx5e_sq *sq)
{
- clear_bit(MLX5E_SQ_STATE_WAKE_TXQ_ENABLE, &sq->state);
- napi_synchronize(&sq->channel->napi); /* prevent netif_tx_wake_queue */
- netif_tx_disable_queue(sq->txq);
+ if (sq->txq) {
+ clear_bit(MLX5E_SQ_STATE_WAKE_TXQ_ENABLE, &sq->state);
+ /* prevent netif_tx_wake_queue */
+ napi_synchronize(&sq->channel->napi);
+ netif_tx_disable_queue(sq->txq);
- /* ensure hw is notified of all pending wqes */
- if (mlx5e_sq_has_room_for(sq, 1))
- mlx5e_send_nop(sq, true);
+ /* ensure hw is notified of all pending wqes */
+ if (mlx5e_sq_has_room_for(sq, 1))
+ mlx5e_send_nop(sq, true);
+
+ mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
+ }
- mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR);
while (sq->cc != sq->pc) /* wait till sq is empty */
msleep(20);
@@ -1030,10 +1062,14 @@ static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
netif_napi_add(netdev, &c->napi, mlx5e_napi_poll, 64);
- err = mlx5e_open_tx_cqs(c, cparam);
+ err = mlx5e_open_cq(c, &cparam->icosq_cq, &c->icosq.cq, 0, 0);
if (err)
goto err_napi_del;
+ err = mlx5e_open_tx_cqs(c, cparam);
+ if (err)
+ goto err_close_icosq_cq;
+
err = mlx5e_open_cq(c, &cparam->rx_cq, &c->rq.cq,
priv->params.rx_cq_moderation_usec,
priv->params.rx_cq_moderation_pkts);
@@ -1042,10 +1078,14 @@ static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
napi_enable(&c->napi);
- err = mlx5e_open_sqs(c, cparam);
+ err = mlx5e_open_sq(c, 0, &cparam->icosq, &c->icosq);
if (err)
goto err_disable_napi;
+ err = mlx5e_open_sqs(c, cparam);
+ if (err)
+ goto err_close_icosq;
+
err = mlx5e_open_rq(c, &cparam->rq, &c->rq);
if (err)
goto err_close_sqs;
@@ -1058,6 +1098,9 @@ static int mlx5e_open_channel(struct mlx5e_priv *priv, int ix,
err_close_sqs:
mlx5e_close_sqs(c);
+err_close_icosq:
+ mlx5e_close_sq(&c->icosq);
+
err_disable_napi:
napi_disable(&c->napi);
mlx5e_close_cq(&c->rq.cq);
@@ -1065,6 +1108,9 @@ err_disable_napi:
err_close_tx_cqs:
mlx5e_close_tx_cqs(c);
+err_close_icosq_cq:
+ mlx5e_close_cq(&c->icosq.cq);
+
err_napi_del:
netif_napi_del(&c->napi);
napi_hash_del(&c->napi);
@@ -1077,9 +1123,11 @@ static void mlx5e_close_channel(struct mlx5e_channel *c)
{
mlx5e_close_rq(&c->rq);
mlx5e_close_sqs(c);
+ mlx5e_close_sq(&c->icosq);
napi_disable(&c->napi);
mlx5e_close_cq(&c->rq.cq);
mlx5e_close_tx_cqs(c);
+ mlx5e_close_cq(&c->icosq.cq);
netif_napi_del(&c->napi);
napi_hash_del(&c->napi);
@@ -1125,17 +1173,27 @@ static void mlx5e_build_drop_rq_param(struct mlx5e_rq_param *param)
MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe)));
}
-static void mlx5e_build_sq_param(struct mlx5e_priv *priv,
- struct mlx5e_sq_param *param)
+static void mlx5e_build_sq_param_common(struct mlx5e_priv *priv,
+ struct mlx5e_sq_param *param)
{
void *sqc = param->sqc;
void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
- MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
MLX5_SET(wq, wq, pd, priv->pdn);
param->wq.buf_numa_node = dev_to_node(&priv->mdev->pdev->dev);
+}
+
+static void mlx5e_build_sq_param(struct mlx5e_priv *priv,
+ struct mlx5e_sq_param *param)
+{
+ void *sqc = param->sqc;
+ void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
+
+ mlx5e_build_sq_param_common(priv, param);
+ MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size);
+
param->max_inline = priv->params.tx_max_inline;
}
@@ -1172,20 +1230,49 @@ static void mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
{
void *cqc = param->cqc;
- MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
+ MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size);
mlx5e_build_common_cq_param(priv, param);
}
+static void mlx5e_build_ico_cq_param(struct mlx5e_priv *priv,
+ struct mlx5e_cq_param *param,
+ u8 log_wq_size)
+{
+ void *cqc = param->cqc;
+
+ MLX5_SET(cqc, cqc, log_cq_size, log_wq_size);
+
+ mlx5e_build_common_cq_param(priv, param);
+}
+
+static void mlx5e_build_icosq_param(struct mlx5e_priv *priv,
+ struct mlx5e_sq_param *param,
+ u8 log_wq_size)
+{
+ void *sqc = param->sqc;
+ void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
+
+ mlx5e_build_sq_param_common(priv, param);
+
+ MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
+
+ param->icosq = true;
+}
+
static void mlx5e_build_channel_param(struct mlx5e_priv *priv,
struct mlx5e_channel_param *cparam)
{
+ u8 icosq_log_wq_sz = 0;
+
memset(cparam, 0, sizeof(*cparam));
mlx5e_build_rq_param(priv, &cparam->rq);
mlx5e_build_sq_param(priv, &cparam->sq);
+ mlx5e_build_icosq_param(priv, &cparam->icosq, icosq_log_wq_sz);
mlx5e_build_rx_cq_param(priv, &cparam->rx_cq);
mlx5e_build_tx_cq_param(priv, &cparam->tx_cq);
+ mlx5e_build_ico_cq_param(priv, &cparam->icosq_cq, icosq_log_wq_sz);
}
static int mlx5e_open_channels(struct mlx5e_priv *priv)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 1ffc7cb..a8d2935 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -54,6 +54,7 @@ void mlx5e_send_nop(struct mlx5e_sq *sq, bool notify_hw)
sq->skb[pi] = NULL;
sq->pc++;
+ sq->stats.nop++;
if (notify_hw) {
cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
@@ -387,7 +388,6 @@ bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq, int napi_budget)
wi = &sq->wqe_info[ci];
if (unlikely(!skb)) { /* nop */
- sq->stats.nop++;
sqcc++;
continue;
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index 9bb4395..ad624cb 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
@@ -49,6 +49,57 @@ struct mlx5_cqe64 *mlx5e_get_cqe(struct mlx5e_cq *cq)
return cqe;
}
+static void mlx5e_poll_ico_cq(struct mlx5e_cq *cq)
+{
+ struct mlx5_wq_cyc *wq;
+ struct mlx5_cqe64 *cqe;
+ struct mlx5e_sq *sq;
+ u16 sqcc;
+
+ cqe = mlx5e_get_cqe(cq);
+ if (likely(!cqe))
+ return;
+
+ sq = container_of(cq, struct mlx5e_sq, cq);
+ wq = &sq->wq;
+
+ /* sq->cc must be updated only after mlx5_cqwq_update_db_record(),
+ * otherwise a cq overrun may occur
+ */
+ sqcc = sq->cc;
+
+ do {
+ u16 ci = be16_to_cpu(cqe->wqe_counter) & wq->sz_m1;
+ struct mlx5e_ico_wqe_info *icowi = &sq->ico_wqe_info[ci];
+
+ mlx5_cqwq_pop(&cq->wq);
+ sqcc += icowi->num_wqebbs;
+
+ if (unlikely((cqe->op_own >> 4) != MLX5_CQE_REQ)) {
+ WARN_ONCE(true, "mlx5e: Bad OP in ICOSQ CQE: 0x%x\n",
+ cqe->op_own);
+ break;
+ }
+
+ switch (icowi->opcode) {
+ case MLX5_OPCODE_NOP:
+ break;
+ default:
+ WARN_ONCE(true,
+ "mlx5e: Bad OPCODE in ICOSQ WQE info: 0x%x\n",
+ icowi->opcode);
+ }
+
+ } while ((cqe = mlx5e_get_cqe(cq)));
+
+ mlx5_cqwq_update_db_record(&cq->wq);
+
+ /* ensure cq space is freed before enabling more cqes */
+ wmb();
+
+ sq->cc = sqcc;
+}
+
int mlx5e_napi_poll(struct napi_struct *napi, int budget)
{
struct mlx5e_channel *c = container_of(napi, struct mlx5e_channel,
@@ -64,6 +115,9 @@ int mlx5e_napi_poll(struct napi_struct *napi, int budget)
work_done = mlx5e_poll_rx_cq(&c->rq.cq, budget);
busy |= work_done == budget;
+
+ mlx5e_poll_ico_cq(&c->icosq.cq);
+
busy |= mlx5e_post_rx_wqes(&c->rq);
if (busy)
@@ -80,6 +134,7 @@ int mlx5e_napi_poll(struct napi_struct *napi, int budget)
for (i = 0; i < c->num_tc; i++)
mlx5e_cq_arm(&c->sq[i].cq);
mlx5e_cq_arm(&c->rq.cq);
+ mlx5e_cq_arm(&c->icosq.cq);
return work_done;
}
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 07/11] net/mlx5e: Add fragmented memory support for RX multi packet WQE
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (5 preceding siblings ...)
2016-04-17 21:32 ` [PATCH net-next V2 06/11] net/mlx5e: Added ICO SQs Saeed Mahameed
@ 2016-04-17 21:32 ` Saeed Mahameed
2016-04-17 21:32 ` [PATCH net-next V2 08/11] net/mlx5e: Use napi_alloc_skb for RX SKB allocations Saeed Mahameed
` (3 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:32 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
If the allocation of a linear (physically continuous) MPWQE fails,
we allocate a fragmented MPWQE.
This is implemented via device's UMR (User Memory Registration)
which allows to register multiple memory fragments into ConnectX
hardware as a continuous buffer.
UMR registration is an asynchronous operation and is done via
ICO SQs.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 78 ++++-
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 64 ++++-
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 357 +++++++++++++++++++--
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c | 4 +-
drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 3 +
5 files changed, 458 insertions(+), 48 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index f62495f..1d4e8fe 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -69,6 +69,10 @@
MLX5_MPWRQ_LOG_STRIDE_SIZE)
#define MLX5_MPWRQ_WQE_PAGE_ORDER (max_t(int, 0, \
MLX5_MPWRQ_LOG_WQE_SZ - PAGE_SHIFT))
+#define MLX5_MPWRQ_WQE_NUM_PAGES BIT(MLX5_MPWRQ_WQE_PAGE_ORDER)
+#define MLX5_CHANNEL_MAX_NUM_MTTS (ALIGN(MLX5_MPWRQ_WQE_NUM_PAGES, 8) * \
+ BIT(MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE_MPW))
+#define MLX5_UMR_ALIGN (2048)
#define MLX5_MPWRQ_SMALL_PACKET_THRESHOLD (128)
#define MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ (64 * 1024)
@@ -131,6 +135,13 @@ struct mlx5e_rx_wqe {
struct mlx5_wqe_data_seg data;
};
+struct mlx5e_umr_wqe {
+ struct mlx5_wqe_ctrl_seg ctrl;
+ struct mlx5_wqe_umr_ctrl_seg uctrl;
+ struct mlx5_mkey_seg mkc;
+ struct mlx5_wqe_data_seg data;
+};
+
#ifdef CONFIG_MLX5_CORE_EN_DCB
#define MLX5E_MAX_BW_ALLOC 100 /* Max percentage of BW allocation */
#define MLX5E_MIN_BW_ALLOC 1 /* Min percentage of BW allocation */
@@ -176,6 +187,7 @@ static const char vport_strings[][ETH_GSTRING_LEN] = {
"tx_queue_dropped",
"rx_wqe_err",
"rx_mpwqe_filler",
+ "rx_mpwqe_frag",
};
struct mlx5e_vport_stats {
@@ -218,8 +230,9 @@ struct mlx5e_vport_stats {
u64 tx_queue_dropped;
u64 rx_wqe_err;
u64 rx_mpwqe_filler;
+ u64 rx_mpwqe_frag;
-#define NUM_VPORT_COUNTERS 36
+#define NUM_VPORT_COUNTERS 37
};
static const char pport_strings[][ETH_GSTRING_LEN] = {
@@ -314,6 +327,7 @@ static const char rq_stats_strings[][ETH_GSTRING_LEN] = {
"lro_bytes",
"wqe_err",
"mpwqe_filler",
+ "mpwqe_frag",
};
struct mlx5e_rq_stats {
@@ -325,7 +339,8 @@ struct mlx5e_rq_stats {
u64 lro_bytes;
u64 wqe_err;
u64 mpwqe_filler;
-#define NUM_RQ_STATS 8
+ u64 mpwqe_frag;
+#define NUM_RQ_STATS 9
};
static const char sq_stats_strings[][ETH_GSTRING_LEN] = {
@@ -404,6 +419,7 @@ struct mlx5e_tstamp {
enum {
MLX5E_RQ_STATE_POST_WQES_ENABLE,
+ MLX5E_RQ_STATE_UMR_WQE_IN_PROGRESS,
};
struct mlx5e_cq {
@@ -431,18 +447,14 @@ struct mlx5e_dma_info {
dma_addr_t addr;
};
-struct mlx5e_mpw_info {
- struct mlx5e_dma_info dma_info;
- u16 consumed_strides;
- u16 skbs_frags;
-};
-
struct mlx5e_rq {
/* data path */
struct mlx5_wq_ll wq;
u32 wqe_sz;
struct sk_buff **skb;
struct mlx5e_mpw_info *wqe_info;
+ __be32 mkey_be;
+ __be32 umr_mkey_be;
struct device *pdev;
struct net_device *netdev;
@@ -463,6 +475,29 @@ struct mlx5e_rq {
struct mlx5e_priv *priv;
} ____cacheline_aligned_in_smp;
+struct mlx5e_umr_dma_info {
+ __be64 *mtt;
+ __be64 *mtt_no_align;
+ dma_addr_t mtt_addr;
+ struct mlx5e_dma_info *dma_info;
+};
+
+struct mlx5e_mpw_info {
+ union {
+ struct mlx5e_dma_info dma_info;
+ struct mlx5e_umr_dma_info umr;
+ };
+ u16 consumed_strides;
+ u16 skbs_frags;
+
+ void (*complete_wqe)(struct mlx5e_rq *rq,
+ struct mlx5_cqe64 *cqe,
+ u16 byte_cnt,
+ struct mlx5e_mpw_info *wi,
+ struct sk_buff *skb);
+ void (*free_wqe)(struct mlx5e_rq *rq, struct mlx5e_mpw_info *wi);
+};
+
struct mlx5e_tx_wqe_info {
u32 num_bytes;
u8 num_wqebbs;
@@ -655,6 +690,7 @@ struct mlx5e_priv {
u32 pdn;
u32 tdn;
struct mlx5_core_mkey mkey;
+ struct mlx5_core_mkey umr_mkey;
struct mlx5e_rq drop_rq;
struct mlx5e_channel **channel;
@@ -727,6 +763,21 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe);
bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq);
int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix);
int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix);
+void mlx5e_post_rx_fragmented_mpwqe(struct mlx5e_rq *rq);
+void mlx5e_complete_rx_linear_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5_cqe64 *cqe,
+ u16 byte_cnt,
+ struct mlx5e_mpw_info *wi,
+ struct sk_buff *skb);
+void mlx5e_complete_rx_fragmented_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5_cqe64 *cqe,
+ u16 byte_cnt,
+ struct mlx5e_mpw_info *wi,
+ struct sk_buff *skb);
+void mlx5e_free_rx_linear_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5e_mpw_info *wi);
+void mlx5e_free_rx_fragmented_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5e_mpw_info *wi);
struct mlx5_cqe64 *mlx5e_get_cqe(struct mlx5e_cq *cq);
void mlx5e_update_stats(struct mlx5e_priv *priv);
@@ -760,7 +811,7 @@ void mlx5e_build_default_indir_rqt(struct mlx5_core_dev *mdev,
int num_channels);
static inline void mlx5e_tx_notify_hw(struct mlx5e_sq *sq,
- struct mlx5e_tx_wqe *wqe, int bf_sz)
+ struct mlx5_wqe_ctrl_seg *ctrl, int bf_sz)
{
u16 ofst = MLX5_BF_OFFSET + sq->bf_offset;
@@ -774,9 +825,9 @@ static inline void mlx5e_tx_notify_hw(struct mlx5e_sq *sq,
*/
wmb();
if (bf_sz)
- __iowrite64_copy(sq->uar_map + ofst, &wqe->ctrl, bf_sz);
+ __iowrite64_copy(sq->uar_map + ofst, ctrl, bf_sz);
else
- mlx5_write64((__be32 *)&wqe->ctrl, sq->uar_map + ofst, NULL);
+ mlx5_write64((__be32 *)ctrl, sq->uar_map + ofst, NULL);
/* flush the write-combining mapped buffer */
wmb();
@@ -797,6 +848,11 @@ static inline int mlx5e_get_max_num_channels(struct mlx5_core_dev *mdev)
MLX5E_MAX_NUM_CHANNELS);
}
+static inline int mlx5e_get_mtt_octw(int npages)
+{
+ return ALIGN(npages, 8) / 2;
+}
+
extern const struct ethtool_ops mlx5e_ethtool_ops;
#ifdef CONFIG_MLX5_CORE_EN_DCB
extern const struct dcbnl_rtnl_ops mlx5e_dcbnl_ops;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index b25b429..942829e 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -179,6 +179,7 @@ void mlx5e_update_stats(struct mlx5e_priv *priv)
s->rx_csum_sw = 0;
s->rx_wqe_err = 0;
s->rx_mpwqe_filler = 0;
+ s->rx_mpwqe_frag = 0;
for (i = 0; i < priv->params.num_channels; i++) {
rq_stats = &priv->channel[i]->rq.stats;
@@ -190,6 +191,7 @@ void mlx5e_update_stats(struct mlx5e_priv *priv)
s->rx_csum_sw += rq_stats->csum_sw;
s->rx_wqe_err += rq_stats->wqe_err;
s->rx_mpwqe_filler += rq_stats->mpwqe_filler;
+ s->rx_mpwqe_frag += rq_stats->mpwqe_frag;
for (j = 0; j < priv->params.num_tc; j++) {
sq_stats = &priv->channel[i]->sq[j].stats;
@@ -379,7 +381,6 @@ static int mlx5e_create_rq(struct mlx5e_channel *c,
for (i = 0; i < wq_sz; i++) {
struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
- wqe->data.lkey = c->mkey_be;
wqe->data.byte_count = cpu_to_be32(byte_count);
}
@@ -390,6 +391,8 @@ static int mlx5e_create_rq(struct mlx5e_channel *c,
rq->channel = c;
rq->ix = c->ix;
rq->priv = c->priv;
+ rq->mkey_be = c->mkey_be;
+ rq->umr_mkey_be = cpu_to_be32(c->priv->umr_mkey.key);
return 0;
@@ -1256,6 +1259,7 @@ static void mlx5e_build_icosq_param(struct mlx5e_priv *priv,
mlx5e_build_sq_param_common(priv, param);
MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
+ MLX5_SET(sqc, sqc, reg_umr, MLX5_CAP_ETH(priv->mdev, reg_umr_sq));
param->icosq = true;
}
@@ -1263,7 +1267,7 @@ static void mlx5e_build_icosq_param(struct mlx5e_priv *priv,
static void mlx5e_build_channel_param(struct mlx5e_priv *priv,
struct mlx5e_channel_param *cparam)
{
- u8 icosq_log_wq_sz = 0;
+ u8 icosq_log_wq_sz = MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE;
memset(cparam, 0, sizeof(*cparam));
@@ -2458,6 +2462,13 @@ void mlx5e_build_default_indir_rqt(struct mlx5_core_dev *mdev,
indirection_rqt[i] = i % num_channels;
}
+static bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
+{
+ return MLX5_CAP_GEN(mdev, striding_rq) &&
+ MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
+ MLX5_CAP_ETH(mdev, reg_umr_sq);
+}
+
static void mlx5e_build_netdev_priv(struct mlx5_core_dev *mdev,
struct net_device *netdev,
int num_channels)
@@ -2466,7 +2477,7 @@ static void mlx5e_build_netdev_priv(struct mlx5_core_dev *mdev,
priv->params.log_sq_size =
MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE;
- priv->params.rq_wq_type = MLX5_CAP_GEN(mdev, striding_rq) ?
+ priv->params.rq_wq_type = mlx5e_check_fragmented_striding_rq_cap(mdev) ?
MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ :
MLX5_WQ_TYPE_LINKED_LIST;
@@ -2639,6 +2650,41 @@ static void mlx5e_destroy_q_counter(struct mlx5e_priv *priv)
mlx5_core_dealloc_q_counter(priv->mdev, priv->q_counter);
}
+static int mlx5e_create_umr_mkey(struct mlx5e_priv *priv)
+{
+ struct mlx5_core_dev *mdev = priv->mdev;
+ struct mlx5_create_mkey_mbox_in *in;
+ struct mlx5_mkey_seg *mkc;
+ int inlen = sizeof(*in);
+ u64 npages =
+ mlx5e_get_max_num_channels(mdev) * MLX5_CHANNEL_MAX_NUM_MTTS;
+ int err;
+
+ in = mlx5_vzalloc(inlen);
+ if (!in)
+ return -ENOMEM;
+
+ mkc = &in->seg;
+ mkc->status = MLX5_MKEY_STATUS_FREE;
+ mkc->flags = MLX5_PERM_UMR_EN |
+ MLX5_PERM_LOCAL_READ |
+ MLX5_PERM_LOCAL_WRITE |
+ MLX5_ACCESS_MODE_MTT;
+
+ mkc->qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+ mkc->flags_pd = cpu_to_be32(priv->pdn);
+ mkc->len = cpu_to_be64(npages << PAGE_SHIFT);
+ mkc->xlt_oct_size = cpu_to_be32(mlx5e_get_mtt_octw(npages));
+ mkc->log2_page_size = PAGE_SHIFT;
+
+ err = mlx5_core_create_mkey(mdev, &priv->umr_mkey, in, inlen, NULL,
+ NULL, NULL);
+
+ kvfree(in);
+
+ return err;
+}
+
static void *mlx5e_create_netdev(struct mlx5_core_dev *mdev)
{
struct net_device *netdev;
@@ -2688,10 +2734,16 @@ static void *mlx5e_create_netdev(struct mlx5_core_dev *mdev)
goto err_dealloc_transport_domain;
}
+ err = mlx5e_create_umr_mkey(priv);
+ if (err) {
+ mlx5_core_err(mdev, "create umr mkey failed, %d\n", err);
+ goto err_destroy_mkey;
+ }
+
err = mlx5e_create_tises(priv);
if (err) {
mlx5_core_warn(mdev, "create tises failed, %d\n", err);
- goto err_destroy_mkey;
+ goto err_destroy_umr_mkey;
}
err = mlx5e_open_drop_rq(priv);
@@ -2774,6 +2826,9 @@ err_close_drop_rq:
err_destroy_tises:
mlx5e_destroy_tises(priv);
+err_destroy_umr_mkey:
+ mlx5_core_destroy_mkey(mdev, &priv->umr_mkey);
+
err_destroy_mkey:
mlx5_core_destroy_mkey(mdev, &priv->mkey);
@@ -2812,6 +2867,7 @@ static void mlx5e_destroy_netdev(struct mlx5_core_dev *mdev, void *vpriv)
mlx5e_destroy_rqt(priv, MLX5E_INDIRECTION_RQT);
mlx5e_close_drop_rq(priv);
mlx5e_destroy_tises(priv);
+ mlx5_core_destroy_mkey(priv->mdev, &priv->umr_mkey);
mlx5_core_destroy_mkey(priv->mdev, &priv->mkey);
mlx5_core_dealloc_transport_domain(priv->mdev, priv->tdn);
mlx5_core_dealloc_pd(priv->mdev, priv->pdn);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index e652604..22c2812 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -65,6 +65,7 @@ int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
*((dma_addr_t *)skb->cb) = dma_addr;
wqe->data.addr = cpu_to_be64(dma_addr + MLX5E_NET_IP_ALIGN);
+ wqe->data.lkey = rq->mkey_be;
rq->skb[ix] = skb;
@@ -76,7 +77,190 @@ err_free_skb:
return -ENOMEM;
}
-int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
+static u16 mlx5e_get_wqe_mtt_offset(u16 rq_ix, u16 wqe_ix)
+{
+ return rq_ix * MLX5_CHANNEL_MAX_NUM_MTTS +
+ wqe_ix * ALIGN(MLX5_MPWRQ_WQE_NUM_PAGES, 8);
+}
+
+static void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,
+ struct mlx5e_sq *sq,
+ struct mlx5e_umr_wqe *wqe,
+ u16 ix)
+{
+ struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
+ struct mlx5_wqe_umr_ctrl_seg *ucseg = &wqe->uctrl;
+ struct mlx5_wqe_data_seg *dseg = &wqe->data;
+ struct mlx5e_mpw_info *wi = &rq->wqe_info[ix];
+ u8 ds_cnt = DIV_ROUND_UP(sizeof(*wqe), MLX5_SEND_WQE_DS);
+ u16 umr_wqe_mtt_offset = mlx5e_get_wqe_mtt_offset(rq->ix, ix);
+
+ memset(wqe, 0, sizeof(*wqe));
+ cseg->opmod_idx_opcode =
+ cpu_to_be32((sq->pc << MLX5_WQE_CTRL_WQE_INDEX_SHIFT) |
+ MLX5_OPCODE_UMR);
+ cseg->qpn_ds = cpu_to_be32((sq->sqn << MLX5_WQE_CTRL_QPN_SHIFT) |
+ ds_cnt);
+ cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
+ cseg->imm = rq->umr_mkey_be;
+
+ ucseg->flags = MLX5_UMR_TRANSLATION_OFFSET_EN;
+ ucseg->klm_octowords =
+ cpu_to_be16(mlx5e_get_mtt_octw(MLX5_MPWRQ_WQE_NUM_PAGES));
+ ucseg->bsf_octowords =
+ cpu_to_be16(mlx5e_get_mtt_octw(umr_wqe_mtt_offset));
+ ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
+
+ dseg->lkey = sq->mkey_be;
+ dseg->addr = cpu_to_be64(wi->umr.mtt_addr);
+}
+
+static void mlx5e_post_umr_wqe(struct mlx5e_rq *rq, u16 ix)
+{
+ struct mlx5e_sq *sq = &rq->channel->icosq;
+ struct mlx5_wq_cyc *wq = &sq->wq;
+ struct mlx5e_umr_wqe *wqe;
+ u8 num_wqebbs = DIV_ROUND_UP(sizeof(*wqe), MLX5_SEND_WQE_BB);
+ u16 pi;
+
+ /* fill sq edge with nops to avoid wqe wrap around */
+ while ((pi = (sq->pc & wq->sz_m1)) > sq->edge) {
+ sq->ico_wqe_info[pi].opcode = MLX5_OPCODE_NOP;
+ sq->ico_wqe_info[pi].num_wqebbs = 1;
+ mlx5e_send_nop(sq, true);
+ }
+
+ wqe = mlx5_wq_cyc_get_wqe(wq, pi);
+ mlx5e_build_umr_wqe(rq, sq, wqe, ix);
+ sq->ico_wqe_info[pi].opcode = MLX5_OPCODE_UMR;
+ sq->ico_wqe_info[pi].num_wqebbs = num_wqebbs;
+ sq->pc += num_wqebbs;
+ mlx5e_tx_notify_hw(sq, &wqe->ctrl, 0);
+}
+
+static inline int mlx5e_get_wqe_mtt_sz(void)
+{
+ /* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes.
+ * To avoid copying garbage after the mtt array, we allocate
+ * a little more.
+ */
+ return ALIGN(MLX5_MPWRQ_WQE_NUM_PAGES * sizeof(__be64),
+ MLX5_UMR_MTT_ALIGNMENT);
+}
+
+static int mlx5e_alloc_and_map_page(struct mlx5e_rq *rq,
+ struct mlx5e_mpw_info *wi,
+ int i)
+{
+ struct page *page;
+
+ page = dev_alloc_page();
+ if (unlikely(!page))
+ return -ENOMEM;
+
+ wi->umr.dma_info[i].page = page;
+ wi->umr.dma_info[i].addr = dma_map_page(rq->pdev, page, 0, PAGE_SIZE,
+ PCI_DMA_FROMDEVICE);
+ if (unlikely(dma_mapping_error(rq->pdev, wi->umr.dma_info[i].addr))) {
+ put_page(page);
+ return -ENOMEM;
+ }
+ wi->umr.mtt[i] = cpu_to_be64(wi->umr.dma_info[i].addr | MLX5_EN_WR);
+
+ return 0;
+}
+
+static int mlx5e_alloc_rx_fragmented_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5e_rx_wqe *wqe,
+ u16 ix)
+{
+ struct mlx5e_mpw_info *wi = &rq->wqe_info[ix];
+ int mtt_sz = mlx5e_get_wqe_mtt_sz();
+ u32 dma_offset = mlx5e_get_wqe_mtt_offset(rq->ix, ix) << PAGE_SHIFT;
+ int i;
+
+ wi->umr.dma_info = kmalloc(sizeof(*wi->umr.dma_info) *
+ MLX5_MPWRQ_WQE_NUM_PAGES,
+ GFP_ATOMIC);
+ if (unlikely(!wi->umr.dma_info))
+ goto err_out;
+
+ /* We allocate more than mtt_sz as we will align the pointer */
+ wi->umr.mtt_no_align = kzalloc(mtt_sz + MLX5_UMR_ALIGN - 1,
+ GFP_ATOMIC);
+ if (unlikely(!wi->umr.mtt_no_align))
+ goto err_free_umr;
+
+ wi->umr.mtt = PTR_ALIGN(wi->umr.mtt_no_align, MLX5_UMR_ALIGN);
+ wi->umr.mtt_addr = dma_map_single(rq->pdev, wi->umr.mtt, mtt_sz,
+ PCI_DMA_TODEVICE);
+ if (unlikely(dma_mapping_error(rq->pdev, wi->umr.mtt_addr)))
+ goto err_free_mtt;
+
+ for (i = 0; i < MLX5_MPWRQ_WQE_NUM_PAGES; i++)
+ if (unlikely(mlx5e_alloc_and_map_page(rq, wi, i)))
+ goto err_unmap;
+
+ wi->consumed_strides = 0;
+ wi->complete_wqe = mlx5e_complete_rx_fragmented_mpwqe;
+ wi->free_wqe = mlx5e_free_rx_fragmented_mpwqe;
+ wqe->data.lkey = rq->umr_mkey_be;
+ wqe->data.addr = cpu_to_be64(dma_offset);
+
+ return 0;
+
+err_unmap:
+ while (--i >= 0) {
+ dma_unmap_page(rq->pdev, wi->umr.dma_info[i].addr, PAGE_SIZE,
+ PCI_DMA_FROMDEVICE);
+ put_page(wi->umr.dma_info[i].page);
+ }
+ dma_unmap_single(rq->pdev, wi->umr.mtt_addr, mtt_sz, PCI_DMA_TODEVICE);
+
+err_free_mtt:
+ kfree(wi->umr.mtt_no_align);
+
+err_free_umr:
+ kfree(wi->umr.dma_info);
+
+err_out:
+ return -ENOMEM;
+}
+
+void mlx5e_free_rx_fragmented_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5e_mpw_info *wi)
+{
+ int mtt_sz = mlx5e_get_wqe_mtt_sz();
+ int i;
+
+ for (i = 0; i < MLX5_MPWRQ_WQE_NUM_PAGES; i++) {
+ dma_unmap_page(rq->pdev, wi->umr.dma_info[i].addr, PAGE_SIZE,
+ PCI_DMA_FROMDEVICE);
+ put_page(wi->umr.dma_info[i].page);
+ }
+ dma_unmap_single(rq->pdev, wi->umr.mtt_addr, mtt_sz, PCI_DMA_TODEVICE);
+ kfree(wi->umr.mtt_no_align);
+ kfree(wi->umr.dma_info);
+}
+
+void mlx5e_post_rx_fragmented_mpwqe(struct mlx5e_rq *rq)
+{
+ struct mlx5_wq_ll *wq = &rq->wq;
+ struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(wq, wq->head);
+
+ clear_bit(MLX5E_RQ_STATE_UMR_WQE_IN_PROGRESS, &rq->state);
+ mlx5_wq_ll_push(wq, be16_to_cpu(wqe->next.next_wqe_index));
+ rq->stats.mpwqe_frag++;
+
+ /* ensure wqes are visible to device before updating doorbell record */
+ dma_wmb();
+
+ mlx5_wq_ll_update_db_record(wq);
+}
+
+static int mlx5e_alloc_rx_linear_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5e_rx_wqe *wqe,
+ u16 ix)
{
struct mlx5e_mpw_info *wi = &rq->wqe_info[ix];
int ret = 0;
@@ -95,7 +279,10 @@ int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
atomic_add(MLX5_MPWRQ_NUM_STRIDES, &wi->dma_info.page->_count);
wi->skbs_frags = 0;
wi->consumed_strides = 0;
- wqe->data.addr = cpu_to_be64(wi->dma_info.addr);
+ wi->complete_wqe = mlx5e_complete_rx_linear_mpwqe;
+ wi->free_wqe = mlx5e_free_rx_linear_mpwqe;
+ wqe->data.lkey = rq->mkey_be;
+ wqe->data.addr = cpu_to_be64(wi->dma_info.addr);
return 0;
@@ -104,11 +291,42 @@ err_put_page:
return ret;
}
+void mlx5e_free_rx_linear_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5e_mpw_info *wi)
+{
+ dma_unmap_page(rq->pdev, wi->dma_info.addr, rq->wqe_sz,
+ PCI_DMA_FROMDEVICE);
+ atomic_sub(MLX5_MPWRQ_NUM_STRIDES - wi->skbs_frags,
+ &wi->dma_info.page->_count);
+ put_page(wi->dma_info.page);
+}
+
+int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
+{
+ int err;
+
+ err = mlx5e_alloc_rx_linear_mpwqe(rq, wqe, ix);
+ if (unlikely(err)) {
+ err = mlx5e_alloc_rx_fragmented_mpwqe(rq, wqe, ix);
+ if (unlikely(err))
+ return err;
+ set_bit(MLX5E_RQ_STATE_UMR_WQE_IN_PROGRESS, &rq->state);
+ mlx5e_post_umr_wqe(rq, ix);
+ return -EBUSY;
+ }
+
+ return 0;
+}
+
+#define RQ_CANNOT_POST(rq) \
+ (!test_bit(MLX5E_RQ_STATE_POST_WQES_ENABLE, &rq->state) || \
+ test_bit(MLX5E_RQ_STATE_UMR_WQE_IN_PROGRESS, &rq->state))
+
bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
{
struct mlx5_wq_ll *wq = &rq->wq;
- if (unlikely(!test_bit(MLX5E_RQ_STATE_POST_WQES_ENABLE, &rq->state)))
+ if (unlikely(RQ_CANNOT_POST(rq)))
return false;
while (!mlx5_wq_ll_is_full(wq)) {
@@ -302,19 +520,119 @@ wq_ll_pop:
&wqe->next.next_wqe_index);
}
-void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
+static inline void mlx5e_sync_copy_linear_data(struct mlx5e_rq *rq,
+ struct sk_buff *skb,
+ struct mlx5e_dma_info *dma_info,
+ unsigned int pg_offset,
+ unsigned int skb_offset,
+ unsigned int len)
+{
+ /* Aligning len to sizeof(long) optimizes memcpy performance */
+ len = ALIGN(len, sizeof(long));
+ dma_sync_single_for_cpu(rq->pdev, dma_info->addr + pg_offset, len,
+ DMA_FROM_DEVICE);
+ skb_copy_to_linear_data_offset(skb, skb_offset,
+ page_address(dma_info->page) + pg_offset,
+ len);
+}
+
+void mlx5e_complete_rx_fragmented_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5_cqe64 *cqe,
+ u16 byte_cnt,
+ struct mlx5e_mpw_info *wi,
+ struct sk_buff *skb)
+{
+ u16 stride_ix = mpwrq_get_cqe_stride_index(cqe);
+ u32 wqe_offset = stride_ix * MLX5_MPWRQ_STRIDE_SIZE;
+ u32 head_offset = wqe_offset & (PAGE_SIZE - 1);
+ u32 page_idx = wqe_offset >> PAGE_SHIFT;
+ struct mlx5e_dma_info *head_dma_info;
+ struct mlx5e_dma_info *dma_info;
+ u16 headlen = min_t(u16, MLX5_MPWRQ_SMALL_PACKET_THRESHOLD, byte_cnt);
+ u16 headlen_pg = min_t(u32, headlen, PAGE_SIZE - head_offset);
+ u32 frag_offset = head_offset + headlen;
+
+ dma_info = &wi->umr.dma_info[page_idx];
+ head_dma_info = dma_info;
+#if (MLX5_MPWRQ_SMALL_PACKET_THRESHOLD >= MLX5_MPWRQ_STRIDE_SIZE)
+ if (unlikely(frag_offset >= PAGE_SIZE)) {
+ dma_info++;
+ frag_offset = headlen - headlen_pg;
+ }
+#endif
+
+ byte_cnt -= headlen;
+
+ while (byte_cnt) {
+ u32 pg_consumed_bytes =
+ min_t(u32, PAGE_SIZE - frag_offset, byte_cnt);
+ unsigned int truesize;
+
+ dma_sync_single_for_cpu(rq->pdev,
+ dma_info->addr + frag_offset,
+ pg_consumed_bytes,
+ DMA_FROM_DEVICE);
+ truesize = ALIGN(pg_consumed_bytes, MLX5_MPWRQ_STRIDE_SIZE);
+ get_page(dma_info->page);
+ skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, dma_info->page,
+ frag_offset, pg_consumed_bytes, truesize);
+ byte_cnt -= pg_consumed_bytes;
+ dma_info++;
+ frag_offset = 0;
+ }
+ /* copy header */
+ mlx5e_sync_copy_linear_data(rq, skb, head_dma_info, head_offset, 0,
+ headlen_pg);
+#if (MLX5_MPWRQ_SMALL_PACKET_THRESHOLD >= MLX5_MPWRQ_STRIDE_SIZE)
+ if (unlikely(headlen - headlen_pg > 0)) {
+ mlx5e_sync_copy_linear_data(rq, skb, ++head_dma_info, 0,
+ headlen_pg, headlen - headlen_pg);
+ }
+#endif
+ /* skb linear part was allocated with headlen and aligned to long */
+ skb->tail += headlen;
+ skb->len += headlen;
+}
+
+void mlx5e_complete_rx_linear_mpwqe(struct mlx5e_rq *rq,
+ struct mlx5_cqe64 *cqe,
+ u16 byte_cnt,
+ struct mlx5e_mpw_info *wi,
+ struct sk_buff *skb)
{
u16 cstrides = mpwrq_get_cqe_consumed_strides(cqe);
u16 stride_ix = mpwrq_get_cqe_stride_index(cqe);
u32 consumed_bytes = cstrides * MLX5_MPWRQ_STRIDE_SIZE;
u32 data_offset = stride_ix * MLX5_MPWRQ_STRIDE_SIZE;
+ u16 headlen = min_t(u16, MLX5_MPWRQ_SMALL_PACKET_THRESHOLD, byte_cnt);
+ struct mlx5e_dma_info *dma_info;
+
+ dma_info = &wi->dma_info;
+ dma_sync_single_for_cpu(rq->pdev, dma_info->addr + data_offset,
+ consumed_bytes, DMA_FROM_DEVICE);
+
+ byte_cnt -= headlen;
+ if (byte_cnt) {
+ wi->skbs_frags++;
+ skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, dma_info->page,
+ data_offset + headlen, byte_cnt,
+ ALIGN(byte_cnt, MLX5_MPWRQ_STRIDE_SIZE));
+ }
+ skb_copy_to_linear_data(skb, page_address(dma_info->page) + data_offset,
+ ALIGN(headlen, sizeof(long)));
+ /* skb linear part was allocated with headlen and aligned to long */
+ skb->tail += headlen;
+ skb->len += headlen;
+}
+
+void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
+{
+ u16 cstrides = mpwrq_get_cqe_consumed_strides(cqe);
u16 wqe_id = be16_to_cpu(cqe->wqe_id);
struct mlx5e_mpw_info *wi = &rq->wqe_info[wqe_id];
struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, wqe_id);
struct sk_buff *skb;
- u16 byte_cnt;
u16 cqe_bcnt;
- u16 headlen;
wi->consumed_strides += cstrides;
@@ -335,27 +653,8 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
goto mpwrq_cqe_out;
prefetch(skb->data);
- dma_sync_single_for_cpu(rq->pdev, wi->dma_info.addr + data_offset,
- consumed_bytes, DMA_FROM_DEVICE);
-
cqe_bcnt = mpwrq_get_cqe_byte_cnt(cqe);
- headlen = min_t(u16, MLX5_MPWRQ_SMALL_PACKET_THRESHOLD, cqe_bcnt);
-
- byte_cnt = cqe_bcnt - headlen;
- if (byte_cnt) {
- wi->skbs_frags++;
- skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
- wi->dma_info.page, data_offset + headlen,
- byte_cnt,
- ALIGN(byte_cnt, MLX5_MPWRQ_STRIDE_SIZE));
- }
-
- skb_copy_to_linear_data(skb,
- page_address(wi->dma_info.page) + data_offset,
- ALIGN(headlen, sizeof(long)));
- /* skb linear part was allocated with headlen and aligned to long */
- skb->tail += headlen;
- skb->len += headlen;
+ wi->complete_wqe(rq, cqe, cqe_bcnt, wi, skb);
mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
@@ -363,11 +662,7 @@ mpwrq_cqe_out:
if (likely(wi->consumed_strides < MLX5_MPWRQ_NUM_STRIDES))
return;
- dma_unmap_page(rq->pdev, wi->dma_info.addr, rq->wqe_sz,
- PCI_DMA_FROMDEVICE);
- atomic_sub(MLX5_MPWRQ_NUM_STRIDES - wi->skbs_frags,
- &wi->dma_info.page->_count);
- put_page(wi->dma_info.page);
+ wi->free_wqe(rq, wi);
mlx5_wq_ll_pop(&rq->wq, cqe->wqe_id, &wqe->next.next_wqe_index);
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index a8d2935..229ab16 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -58,7 +58,7 @@ void mlx5e_send_nop(struct mlx5e_sq *sq, bool notify_hw)
if (notify_hw) {
cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
- mlx5e_tx_notify_hw(sq, wqe, 0);
+ mlx5e_tx_notify_hw(sq, &wqe->ctrl, 0);
}
}
@@ -310,7 +310,7 @@ static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_sq *sq, struct sk_buff *skb)
bf_sz = wi->num_wqebbs << 3;
cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
- mlx5e_tx_notify_hw(sq, wqe, bf_sz);
+ mlx5e_tx_notify_hw(sq, &wqe->ctrl, bf_sz);
}
/* fill sq edge with nops to avoid wqe wrap around */
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index ad624cb..a3fd0f5 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
@@ -84,6 +84,9 @@ static void mlx5e_poll_ico_cq(struct mlx5e_cq *cq)
switch (icowi->opcode) {
case MLX5_OPCODE_NOP:
break;
+ case MLX5_OPCODE_UMR:
+ mlx5e_post_rx_fragmented_mpwqe(&sq->channel->rq);
+ break;
default:
WARN_ONCE(true,
"mlx5e: Bad OPCODE in ICOSQ WQE info: 0x%x\n",
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 08/11] net/mlx5e: Use napi_alloc_skb for RX SKB allocations
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (6 preceding siblings ...)
2016-04-17 21:32 ` [PATCH net-next V2 07/11] net/mlx5e: Add fragmented memory support for RX multi packet WQE Saeed Mahameed
@ 2016-04-17 21:32 ` Saeed Mahameed
2016-04-17 21:32 ` [PATCH net-next V2 09/11] net/mlx5e: Remove redundant barrier Saeed Mahameed
` (2 subsequent siblings)
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:32 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
Instead of netdev_alloc_skb, we use the napi_alloc_skb function
which is designated to allocate skbuff's for RX in a
channel-specific NAPI instance, and implies the IP packet alignment.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 1 -
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 4 ++--
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 12 +++++-------
3 files changed, 7 insertions(+), 10 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 1d4e8fe..6127bc7 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -91,7 +91,6 @@
#define MLX5E_SQ_BF_BUDGET 16
#define MLX5E_NUM_MAIN_GROUPS 9
-#define MLX5E_NET_IP_ALIGN 2
static inline u16 mlx5_min_rx_wqes(int wq_type, u32 wq_size)
{
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 942829e..9b17bc0 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -373,8 +373,8 @@ static int mlx5e_create_rq(struct mlx5e_channel *c,
rq->wqe_sz = (priv->params.lro_en) ?
priv->params.lro_wqe_sz :
MLX5E_SW2HW_MTU(priv->netdev->mtu);
- rq->wqe_sz = SKB_DATA_ALIGN(rq->wqe_sz + MLX5E_NET_IP_ALIGN);
- byte_count = rq->wqe_sz - MLX5E_NET_IP_ALIGN;
+ rq->wqe_sz = SKB_DATA_ALIGN(rq->wqe_sz);
+ byte_count = rq->wqe_sz;
byte_count |= MLX5_HW_START_PADDING;
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 22c2812..ea90e6f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -47,7 +47,7 @@ int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
struct sk_buff *skb;
dma_addr_t dma_addr;
- skb = netdev_alloc_skb(rq->netdev, rq->wqe_sz);
+ skb = napi_alloc_skb(rq->cq.napi, rq->wqe_sz);
if (unlikely(!skb))
return -ENOMEM;
@@ -61,10 +61,8 @@ int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
if (unlikely(dma_mapping_error(rq->pdev, dma_addr)))
goto err_free_skb;
- skb_reserve(skb, MLX5E_NET_IP_ALIGN);
-
*((dma_addr_t *)skb->cb) = dma_addr;
- wqe->data.addr = cpu_to_be64(dma_addr + MLX5E_NET_IP_ALIGN);
+ wqe->data.addr = cpu_to_be64(dma_addr);
wqe->data.lkey = rq->mkey_be;
rq->skb[ix] = skb;
@@ -646,9 +644,9 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
goto mpwrq_cqe_out;
}
- skb = netdev_alloc_skb(rq->netdev,
- ALIGN(MLX5_MPWRQ_SMALL_PACKET_THRESHOLD,
- sizeof(long)));
+ skb = napi_alloc_skb(rq->cq.napi,
+ ALIGN(MLX5_MPWRQ_SMALL_PACKET_THRESHOLD,
+ sizeof(long)));
if (unlikely(!skb))
goto mpwrq_cqe_out;
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 09/11] net/mlx5e: Remove redundant barrier
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (7 preceding siblings ...)
2016-04-17 21:32 ` [PATCH net-next V2 08/11] net/mlx5e: Use napi_alloc_skb for RX SKB allocations Saeed Mahameed
@ 2016-04-17 21:32 ` Saeed Mahameed
2016-04-17 21:32 ` [PATCH net-next V2 10/11] net/mlx5e: Delay skb->data access Saeed Mahameed
2016-04-17 21:32 ` [PATCH net-next V2 11/11] net/mlx5e: Add ethtool counter for RX buffer allocation failures Saeed Mahameed
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:32 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
The bit-op operation one line before is an explicit barrier
by itself.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c | 1 -
1 files changed, 0 insertions(+), 1 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
index a3fd0f5..c38781f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c
@@ -147,7 +147,6 @@ void mlx5e_completion_event(struct mlx5_core_cq *mcq)
struct mlx5e_cq *cq = container_of(mcq, struct mlx5e_cq, mcq);
set_bit(MLX5E_CHANNEL_NAPI_SCHED, &cq->channel->flags);
- barrier();
napi_schedule(cq->napi);
}
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 10/11] net/mlx5e: Delay skb->data access
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (8 preceding siblings ...)
2016-04-17 21:32 ` [PATCH net-next V2 09/11] net/mlx5e: Remove redundant barrier Saeed Mahameed
@ 2016-04-17 21:32 ` Saeed Mahameed
2016-04-17 21:32 ` [PATCH net-next V2 11/11] net/mlx5e: Add ethtool counter for RX buffer allocation failures Saeed Mahameed
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:32 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Saeed Mahameed
Move mlx5e_handle_csum and eth_type_trans to the end of
mlx5e_build_rx_skb to gain some more time before accessing
skb->data, to reduce cache misses.
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 7 +++----
1 files changed, 3 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index ea90e6f..ade858b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -455,10 +455,6 @@ static inline void mlx5e_build_rx_skb(struct mlx5_cqe64 *cqe,
if (unlikely(mlx5e_rx_hw_stamp(tstamp)))
mlx5e_fill_hwstamp(tstamp, get_cqe_ts(cqe), skb_hwtstamps(skb));
- mlx5e_handle_csum(netdev, cqe, rq, skb, !!lro_num_seg);
-
- skb->protocol = eth_type_trans(skb, netdev);
-
skb_record_rx_queue(skb, rq->ix);
if (likely(netdev->features & NETIF_F_RXHASH))
@@ -469,6 +465,9 @@ static inline void mlx5e_build_rx_skb(struct mlx5_cqe64 *cqe,
be16_to_cpu(cqe->vlan_info));
skb->mark = be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK;
+
+ mlx5e_handle_csum(netdev, cqe, rq, skb, !!lro_num_seg);
+ skb->protocol = eth_type_trans(skb, netdev);
}
static inline void mlx5e_complete_rx_cqe(struct mlx5e_rq *rq,
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread* [PATCH net-next V2 11/11] net/mlx5e: Add ethtool counter for RX buffer allocation failures
2016-04-17 21:31 [PATCH net-next V2 00/11] Mellanox 100G mlx5 driver receive path optimizations Saeed Mahameed
` (9 preceding siblings ...)
2016-04-17 21:32 ` [PATCH net-next V2 10/11] net/mlx5e: Delay skb->data access Saeed Mahameed
@ 2016-04-17 21:32 ` Saeed Mahameed
10 siblings, 0 replies; 20+ messages in thread
From: Saeed Mahameed @ 2016-04-17 21:32 UTC (permalink / raw)
To: David S. Miller
Cc: netdev, Or Gerlitz, Tal Alon, Tariq Toukan, Eran Ben Elisha,
Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
Counts the number of RX buffer allocation failures and shows it
in ethtool statistics.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 8 ++++++--
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 ++
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 11 +++++++++--
3 files changed, 17 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 6127bc7..9aaa23d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -187,6 +187,7 @@ static const char vport_strings[][ETH_GSTRING_LEN] = {
"rx_wqe_err",
"rx_mpwqe_filler",
"rx_mpwqe_frag",
+ "rx_buff_alloc_err",
};
struct mlx5e_vport_stats {
@@ -230,8 +231,9 @@ struct mlx5e_vport_stats {
u64 rx_wqe_err;
u64 rx_mpwqe_filler;
u64 rx_mpwqe_frag;
+ u64 rx_buff_alloc_err;
-#define NUM_VPORT_COUNTERS 37
+#define NUM_VPORT_COUNTERS 38
};
static const char pport_strings[][ETH_GSTRING_LEN] = {
@@ -327,6 +329,7 @@ static const char rq_stats_strings[][ETH_GSTRING_LEN] = {
"wqe_err",
"mpwqe_filler",
"mpwqe_frag",
+ "buff_alloc_err",
};
struct mlx5e_rq_stats {
@@ -339,7 +342,8 @@ struct mlx5e_rq_stats {
u64 wqe_err;
u64 mpwqe_filler;
u64 mpwqe_frag;
-#define NUM_RQ_STATS 9
+ u64 buff_alloc_err;
+#define NUM_RQ_STATS 10
};
static const char sq_stats_strings[][ETH_GSTRING_LEN] = {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 9b17bc0..d485d1e 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -180,6 +180,7 @@ void mlx5e_update_stats(struct mlx5e_priv *priv)
s->rx_wqe_err = 0;
s->rx_mpwqe_filler = 0;
s->rx_mpwqe_frag = 0;
+ s->rx_buff_alloc_err = 0;
for (i = 0; i < priv->params.num_channels; i++) {
rq_stats = &priv->channel[i]->rq.stats;
@@ -192,6 +193,7 @@ void mlx5e_update_stats(struct mlx5e_priv *priv)
s->rx_wqe_err += rq_stats->wqe_err;
s->rx_mpwqe_filler += rq_stats->mpwqe_filler;
s->rx_mpwqe_frag += rq_stats->mpwqe_frag;
+ s->rx_buff_alloc_err += rq_stats->buff_alloc_err;
for (j = 0; j < priv->params.num_tc; j++) {
sq_stats = &priv->channel[i]->sq[j].stats;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index ade858b..e4e8c37 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -329,9 +329,14 @@ bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
while (!mlx5_wq_ll_is_full(wq)) {
struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(wq, wq->head);
+ int err;
- if (unlikely(rq->alloc_wqe(rq, wqe, wq->head)))
+ err = rq->alloc_wqe(rq, wqe, wq->head);
+ if (unlikely(err)) {
+ if (err != -EBUSY)
+ rq->stats.buff_alloc_err++;
break;
+ }
mlx5_wq_ll_push(wq, be16_to_cpu(wqe->next.next_wqe_index));
}
@@ -646,8 +651,10 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
skb = napi_alloc_skb(rq->cq.napi,
ALIGN(MLX5_MPWRQ_SMALL_PACKET_THRESHOLD,
sizeof(long)));
- if (unlikely(!skb))
+ if (unlikely(!skb)) {
+ rq->stats.buff_alloc_err++;
goto mpwrq_cqe_out;
+ }
prefetch(skb->data);
cqe_bcnt = mpwrq_get_cqe_byte_cnt(cqe);
--
1.7.1
^ permalink raw reply related [flat|nested] 20+ messages in thread