[PATCH v1 1/3] net/ice: restrict testpmd to scalar path for TxPP

[PATCH v1 1/3] net/ice: restrict testpmd to scalar path for TxPP

[Soumyadeep Hore]

ICE PMD supports TxPP feature only in scalar path. Hence restricted
testpmd to scalar path when the feature is enabled.

Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
Cc: stable@dpdk.org

Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
---
 drivers/net/intel/ice/ice_rxtx.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/net/intel/ice/ice_rxtx.c b/drivers/net/intel/ice/ice_rxtx.c
index fd0b3a7532..33d82cd46d 100644
--- a/drivers/net/intel/ice/ice_rxtx.c
+++ b/drivers/net/intel/ice/ice_rxtx.c
@@ -4103,8 +4103,11 @@ ice_set_tx_function(struct rte_eth_dev *dev)
 	struct ci_tx_queue *txq;
 	int i;
 	int tx_check_ret = -1;
+	uint64_t offloads;
 
-	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
+	offloads = dev->data->dev_conf.txmode.offloads;
+	if ((offloads & RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP) == 0 &&
+		rte_eal_process_type() == RTE_PROC_PRIMARY) {
 		ad->tx_simd_width = RTE_VECT_SIMD_DISABLED;
 		tx_check_ret = ice_tx_vec_dev_check(dev);
 		ad->tx_simd_width = ice_get_max_simd_bitwidth();
-- 
2.47.1

RE: [PATCH v1 1/3] net/ice: restrict testpmd to scalar path for TxPP

[Loftus, Ciara]

> 
> ICE PMD supports TxPP feature only in scalar path. Hence restricted
> testpmd to scalar path when the feature is enabled.
> 
> Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
> ---
>  drivers/net/intel/ice/ice_rxtx.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/intel/ice/ice_rxtx.c b/drivers/net/intel/ice/ice_rxtx.c
> index fd0b3a7532..33d82cd46d 100644
> --- a/drivers/net/intel/ice/ice_rxtx.c
> +++ b/drivers/net/intel/ice/ice_rxtx.c
> @@ -4103,8 +4103,11 @@ ice_set_tx_function(struct rte_eth_dev *dev)
>  	struct ci_tx_queue *txq;
>  	int i;
>  	int tx_check_ret = -1;
> +	uint64_t offloads;
> 
> -	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
> +	offloads = dev->data->dev_conf.txmode.offloads;
> +	if ((offloads & RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP) == 0
> &&
> +		rte_eal_process_type() == RTE_PROC_PRIMARY) {
>  		ad->tx_simd_width = RTE_VECT_SIMD_DISABLED;
>  		tx_check_ret = ice_tx_vec_dev_check(dev);
>  		ad->tx_simd_width = ice_get_max_simd_bitwidth();
> --
> 2.47.1

I think you can achieve the desired behaviour by adding
RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP to
ICE_TX_NO_VECTOR_FLAGS. Also you should remove the reference
to testpmd in the commit message. The patch is just a change
to the ice driver so should not reference any specific
application.

[PATCH v1 2/3] net/ice: fix PTP clock corruption with TxPP

[Soumyadeep Hore]

On enabling TxPP PTP clock was getting corrupted as timesync was enabled
on during setup. Currently timesync will be enabled during start of tx
queue, hence enabling PHC clock to get updated on starting and stopping
of ports.

Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
Cc: stable@dpdk.org

Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
---
 drivers/net/intel/ice/ice_rxtx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/intel/ice/ice_rxtx.c b/drivers/net/intel/ice/ice_rxtx.c
index 33d82cd46d..7473d45260 100644
--- a/drivers/net/intel/ice/ice_rxtx.c
+++ b/drivers/net/intel/ice/ice_rxtx.c
@@ -908,6 +908,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 			rte_free(txq_elem);
 			return err;
 		}
+		dev->dev_ops->timesync_enable(dev);
 	} else {
 		txq->qtx_tail = hw->hw_addr + QTX_COMM_DBELL(txq->reg_idx);
 
@@ -1671,7 +1672,6 @@ ice_tx_queue_setup(struct rte_eth_dev *dev,
 			PMD_INIT_LOG(ERR, "Cannot register Tx mbuf field/flag for timestamp");
 			return -EINVAL;
 		}
-		dev->dev_ops->timesync_enable(dev);
 
 		txq->tsq->nb_ts_desc = ice_calc_ts_ring_count(ICE_VSI_TO_HW(vsi), txq->nb_tx_desc);
 		ring_size = sizeof(struct ice_ts_desc) * txq->tsq->nb_ts_desc;
-- 
2.47.1

Re: [PATCH v1 2/3] net/ice: fix PTP clock corruption with TxPP

[Bruce Richardson]

On Mon, Oct 27, 2025 at 02:28:00PM -0400, Soumyadeep Hore wrote:
> On enabling TxPP PTP clock was getting corrupted as timesync was enabled
> on during setup. Currently timesync will be enabled during start of tx
> queue, hence enabling PHC clock to get updated on starting and stopping
> of ports.

Can you clarify in a bit more detail, how did enabling the timesync during
setup corrupt the clock, and how does enabling it only on start fix it?

> 
> Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
> ---
>  drivers/net/intel/ice/ice_rxtx.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/net/intel/ice/ice_rxtx.c b/drivers/net/intel/ice/ice_rxtx.c
> index 33d82cd46d..7473d45260 100644
> --- a/drivers/net/intel/ice/ice_rxtx.c
> +++ b/drivers/net/intel/ice/ice_rxtx.c
> @@ -908,6 +908,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
>  			rte_free(txq_elem);
>  			return err;
>  		}
> +		dev->dev_ops->timesync_enable(dev);
>  	} else {
>  		txq->qtx_tail = hw->hw_addr + QTX_COMM_DBELL(txq->reg_idx);
>  
> @@ -1671,7 +1672,6 @@ ice_tx_queue_setup(struct rte_eth_dev *dev,
>  			PMD_INIT_LOG(ERR, "Cannot register Tx mbuf field/flag for timestamp");
>  			return -EINVAL;
>  		}
> -		dev->dev_ops->timesync_enable(dev);
>  
>  		txq->tsq->nb_ts_desc = ice_calc_ts_ring_count(ICE_VSI_TO_HW(vsi), txq->nb_tx_desc);
>  		ring_size = sizeof(struct ice_ts_desc) * txq->tsq->nb_ts_desc;
> -- 
> 2.47.1
> 

[PATCH v1 3/3] doc: update TxPP documentation

[Soumyadeep Hore]

TxPP feature works only when DPDK library is compiled with
PTP feature enabled and application is run on PF.

Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
Cc: stable@dpdk.org

Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
---
 doc/guides/nics/ice.rst | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/doc/guides/nics/ice.rst b/doc/guides/nics/ice.rst
index 7e9ba23102..a15304fb34 100644
--- a/doc/guides/nics/ice.rst
+++ b/doc/guides/nics/ice.rst
@@ -457,6 +457,10 @@ This feature is currently supported only in E830 adapters.
 The flag ``RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP`` is used to enable the feature.
 In order to deliver timestamps internally ``set txtimes`` is used,
 where inter burst and intra burst time interval in nsecs is provided.
+
+Note that dpdk library should be compiled using PTP support enabled and testpmd
+application should run on PF.
+
 For example:
 
 .. code-block:: console
-- 
2.47.1

[PATCH v2 1/3] net/ice: restrict ice PMD to scalar path for TxPP

[Soumyadeep Hore]

ICE PMD supports TxPP feature only in scalar path. Hence restricted
testpmd to scalar path when the feature is enabled.

Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
Cc: stable@dpdk.org

Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
---
 drivers/net/intel/ice/ice_rxtx_vec_common.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/intel/ice/ice_rxtx_vec_common.h b/drivers/net/intel/ice/ice_rxtx_vec_common.h
index a7cc4736cf..a24694c0b1 100644
--- a/drivers/net/intel/ice/ice_rxtx_vec_common.h
+++ b/drivers/net/intel/ice/ice_rxtx_vec_common.h
@@ -59,7 +59,8 @@ _ice_rx_queue_release_mbufs_vec(struct ci_rx_queue *rxq)
 		RTE_ETH_TX_OFFLOAD_GRE_TNL_TSO |    \
 		RTE_ETH_TX_OFFLOAD_IPIP_TNL_TSO |    \
 		RTE_ETH_TX_OFFLOAD_GENEVE_TNL_TSO |    \
-		RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM)
+		RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM |	\
+		RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP)
 
 #define ICE_TX_VECTOR_OFFLOAD (				\
 		RTE_ETH_TX_OFFLOAD_VLAN_INSERT |		\
-- 
2.47.1

[PATCH v2 2/3] net/ice: fix PTP clock corruption with TxPP

[Soumyadeep Hore]

On enabling TxPP PTP clock was getting corrupted as timesync was enabled
on during setup. Currently timesync will be enabled during start of tx
queue, hence enabling PHC clock to get updated on starting and stopping
of ports.

Earlier when timesync was enabled in tx queue setup, on stopping ports
time will reset back to 0. Currently after every port restart PHC clock
will be set to current system CLOCK_REALTIME.

Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
Cc: stable@dpdk.org

Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
---
 drivers/net/intel/ice/ice_rxtx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/intel/ice/ice_rxtx.c b/drivers/net/intel/ice/ice_rxtx.c
index fd0b3a7532..752e52b8cd 100644
--- a/drivers/net/intel/ice/ice_rxtx.c
+++ b/drivers/net/intel/ice/ice_rxtx.c
@@ -908,6 +908,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 			rte_free(txq_elem);
 			return err;
 		}
+		dev->dev_ops->timesync_enable(dev);
 	} else {
 		txq->qtx_tail = hw->hw_addr + QTX_COMM_DBELL(txq->reg_idx);
 
@@ -1671,7 +1672,6 @@ ice_tx_queue_setup(struct rte_eth_dev *dev,
 			PMD_INIT_LOG(ERR, "Cannot register Tx mbuf field/flag for timestamp");
 			return -EINVAL;
 		}
-		dev->dev_ops->timesync_enable(dev);
 
 		txq->tsq->nb_ts_desc = ice_calc_ts_ring_count(ICE_VSI_TO_HW(vsi), txq->nb_tx_desc);
 		ring_size = sizeof(struct ice_ts_desc) * txq->tsq->nb_ts_desc;
-- 
2.47.1

[PATCH v2 3/3] doc: update TxPP documentation

[Soumyadeep Hore]

TxPP feature works only when DPDK library is compiled with
PTP feature enabled and application is run on PF.

Fixes: 0b6ff09a1f19 ("net/intel: support Tx packet pacing for E830")
Cc: stable@dpdk.org

Signed-off-by: Soumyadeep Hore <soumyadeep.hore@intel.com>
---
 doc/guides/nics/ice.rst | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/doc/guides/nics/ice.rst b/doc/guides/nics/ice.rst
index 7e9ba23102..a15304fb34 100644
--- a/doc/guides/nics/ice.rst
+++ b/doc/guides/nics/ice.rst
@@ -457,6 +457,10 @@ This feature is currently supported only in E830 adapters.
 The flag ``RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP`` is used to enable the feature.
 In order to deliver timestamps internally ``set txtimes`` is used,
 where inter burst and intra burst time interval in nsecs is provided.
+
+Note that dpdk library should be compiled using PTP support enabled and testpmd
+application should run on PF.
+
 For example:
 
 .. code-block:: console
-- 
2.47.1