[PATCH 0/6] Fix inconsistencies in offload masks

[PATCH 0/6] Fix inconsistencies in offload masks

[Ciara Loftus]

This series fixes some inconsistencies in the path offload masks and
dev_info offload capabilities for the ice, iavf and i40e drivers.
The first 3 patches are concerned with fixing the path offload masks.
The last 3 patches change how offloads are derived for the dev_info offload
capabilities. Instead of defining an explicit list of offloads, reuse the
information that is already available in the path offload masks. It should
help prevent further inconsistencies in the future.

Ciara Loftus (6):
  net/ice: update Rx supported offloads masks
  net/ice: fix Rx vector path offload masks
  net/i40e: drop unimplemented UDP TSO from Tx mask
  net/ice: derive Rx/Tx offload capabilities from path masks
  net/iavf: derive Rx/Tx offload capabilities from path masks
  net/i40e: derive Rx/Tx offload capabilities from path masks

 drivers/net/intel/i40e/i40e_ethdev.c | 29 +++---------------------
 drivers/net/intel/i40e/i40e_rxtx.h   |  1 -
 drivers/net/intel/iavf/iavf_ethdev.c | 34 +++++++++-------------------
 drivers/net/intel/ice/ice_ethdev.c   | 32 ++++++--------------------
 drivers/net/intel/ice/ice_rxtx.h     |  6 +++--
 5 files changed, 25 insertions(+), 77 deletions(-)

-- 
2.43.0

[PATCH 1/6] net/ice: update Rx supported offloads masks

[Ciara Loftus]

All Rx paths support offloading the outer UDP cksum. Advertise this in the
Rx path offload masks and the dev_info Rx offload capabilities.

Fixes: 75c6287f250f ("net/ice: fix outer checksum flags")
Cc: stable@dpdk.org

Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
---
 drivers/net/intel/ice/ice_ethdev.c | 1 +
 drivers/net/intel/ice/ice_rxtx.h   | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/drivers/net/intel/ice/ice_ethdev.c b/drivers/net/intel/ice/ice_ethdev.c
index dfee07c457b..2722f359cb3 100644
--- a/drivers/net/intel/ice/ice_ethdev.c
+++ b/drivers/net/intel/ice/ice_ethdev.c
@@ -4751,6 +4751,7 @@ ice_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 			RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
 			RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
 			RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
+			RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM |
 			RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
 			RTE_ETH_RX_OFFLOAD_RSS_HASH |
 			RTE_ETH_RX_OFFLOAD_TIMESTAMP |
diff --git a/drivers/net/intel/ice/ice_rxtx.h b/drivers/net/intel/ice/ice_rxtx.h
index cd23f7c56f6..b809c8b0e80 100644
--- a/drivers/net/intel/ice/ice_rxtx.h
+++ b/drivers/net/intel/ice/ice_rxtx.h
@@ -87,6 +87,7 @@
 			RTE_ETH_RX_OFFLOAD_TCP_CKSUM |		\
 			RTE_ETH_RX_OFFLOAD_QINQ_STRIP |		\
 			RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |	\
+			RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM |	\
 			RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |	\
 			RTE_ETH_RX_OFFLOAD_RSS_HASH |		\
 			RTE_ETH_RX_OFFLOAD_TIMESTAMP |		\
@@ -100,6 +101,7 @@
 #define ICE_RX_VECTOR_OFFLOAD_OFFLOADS (	\
 		ICE_RX_VECTOR_OFFLOADS |	\
 		RTE_ETH_RX_OFFLOAD_CHECKSUM |	\
+		RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM |	\
 		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |	\
 		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |\
 		RTE_ETH_RX_OFFLOAD_RSS_HASH)
-- 
2.43.0

[PATCH 2/6] net/ice: fix Rx vector path offload masks

[Ciara Loftus]

The offload RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM was advertised as available
on the plain (non-offload) AVX2/AVX512 Rx paths, however only the offload
versions of those paths compute checksum status flags. Move the flag to
the Rx vector offload path mask where it belongs.

Fixes: 872b571787 ("net/ice: use common Rx path selection infrastructure")
Cc: stable@dpdk.org

Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
---
 drivers/net/intel/ice/ice_rxtx.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/net/intel/ice/ice_rxtx.h b/drivers/net/intel/ice/ice_rxtx.h
index b809c8b0e80..e58f1fab376 100644
--- a/drivers/net/intel/ice/ice_rxtx.h
+++ b/drivers/net/intel/ice/ice_rxtx.h
@@ -95,12 +95,12 @@
 /* basic vector paths */
 #define ICE_RX_VECTOR_OFFLOADS (				\
 			RTE_ETH_RX_OFFLOAD_KEEP_CRC |		\
-			RTE_ETH_RX_OFFLOAD_SCATTER |		\
-			RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM)
+			RTE_ETH_RX_OFFLOAD_SCATTER)
 /* vector offload paths */
 #define ICE_RX_VECTOR_OFFLOAD_OFFLOADS (	\
 		ICE_RX_VECTOR_OFFLOADS |	\
 		RTE_ETH_RX_OFFLOAD_CHECKSUM |	\
+		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |	\
 		RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM |	\
 		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |	\
 		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |\
-- 
2.43.0

[PATCH 3/6] net/i40e: drop unimplemented UDP TSO from Tx mask

[Ciara Loftus]

Commit d9550ce576a7 ("net/intel: create common checksum Tx offload
function") added RTE_ETH_TX_OFFLOAD_UDP_TSO to the scalar Tx offload mask
when merging i40e's checksum helper with ice's, but UDP TSO was never
actually wired up for i40e. i40e_calc_context_desc() does not trigger on
RTE_MBUF_F_TX_UDP_SEG, so no context descriptor is built for UDP
segmentation, and get_context_desc() only invokes i40e_set_tso_ctx() for
RTE_MBUF_F_TX_TCP_SEG, so the TSO fields are never programmed for a
UDP-segmented packet either.

Fix this by removing the unimplemented offload from the mask.

Fixes: d9550ce576a7 ("net/intel: create common checksum Tx offload function")
Cc: stable@dpdk.org

Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
---
 drivers/net/intel/i40e/i40e_rxtx.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/net/intel/i40e/i40e_rxtx.h b/drivers/net/intel/i40e/i40e_rxtx.h
index 88d47f261e4..09773420644 100644
--- a/drivers/net/intel/i40e/i40e_rxtx.h
+++ b/drivers/net/intel/i40e/i40e_rxtx.h
@@ -97,7 +97,6 @@ enum i40e_header_split_mode {
 		RTE_ETH_TX_OFFLOAD_SCTP_CKSUM |		\
 		RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM |	\
 		RTE_ETH_TX_OFFLOAD_TCP_TSO |		\
-		RTE_ETH_TX_OFFLOAD_UDP_TSO |		\
 		RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |	\
 		RTE_ETH_TX_OFFLOAD_GRE_TNL_TSO |	\
 		RTE_ETH_TX_OFFLOAD_IPIP_TNL_TSO |	\
-- 
2.43.0

[PATCH 4/6] net/ice: derive Rx/Tx offload capabilities from path masks

[Ciara Loftus]

tx_offload_capa was missing RTE_ETH_TX_OFFLOAD_UDP_TSO even though the
scalar Tx path supports it. Rather than adding the missing flag to the Tx
offload capabilities directly, derive tx_offload_capa from a mask of all of
the Tx path offload masks.

While we are here, do the same for the Rx offload capabilities mask, which
now derives itself from a mask of all of the Rx path offload masks.

A benefit of this is that if an offload is ever added to a path, it is
automatically reflected in the advertised capabilities too, instead of
relying on the capabilities masks in dev_info being kept manually in sync.
Some overlap will be present when masking the masks but it is kept because
relying on one path's offload mask always being a superset of another's is
fragile.

Fixes: 623ca7a15db4 ("net/ice: enable UDP fragmentation offload")
Cc: stable@dpdk.org

Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
---
 drivers/net/intel/ice/ice_ethdev.c | 33 +++++++-----------------------
 1 file changed, 7 insertions(+), 26 deletions(-)

diff --git a/drivers/net/intel/ice/ice_ethdev.c b/drivers/net/intel/ice/ice_ethdev.c
index 2722f359cb3..b8e703c6956 100644
--- a/drivers/net/intel/ice/ice_ethdev.c
+++ b/drivers/net/intel/ice/ice_ethdev.c
@@ -4745,32 +4745,13 @@ ice_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 	dev_info->flow_type_rss_offloads = 0;
 
 	if (!is_safe_mode) {
-		dev_info->rx_offload_capa |=
-			RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
-			RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
-			RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
-			RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
-			RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
-			RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM |
-			RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
-			RTE_ETH_RX_OFFLOAD_RSS_HASH |
-			RTE_ETH_RX_OFFLOAD_TIMESTAMP |
-			RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT;
-		dev_info->tx_offload_capa |=
-			RTE_ETH_TX_OFFLOAD_QINQ_INSERT |
-			RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
-			RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
-			RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
-			RTE_ETH_TX_OFFLOAD_SCTP_CKSUM |
-			RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM |
-			RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM |
-			RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |
-			RTE_ETH_TX_OFFLOAD_GRE_TNL_TSO |
-			RTE_ETH_TX_OFFLOAD_IPIP_TNL_TSO |
-			RTE_ETH_TX_OFFLOAD_GENEVE_TNL_TSO;
-		if (hw->phy_model == ICE_PHY_E830)
-			dev_info->tx_offload_capa |=
-				RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP;
+		dev_info->rx_offload_capa = ICE_RX_SCALAR_OFFLOADS |
+			ICE_RX_VECTOR_OFFLOADS | ICE_RX_VECTOR_OFFLOAD_OFFLOADS;
+		dev_info->tx_offload_capa = ICE_TX_SCALAR_OFFLOADS |
+			CI_TX_VECTOR_OFFLOADS | CI_TX_VEC_OFFLOAD_PATH_OFFLOADS |
+			CI_TX_VEC_CTX_OFFLOAD_PATH_OFFLOADS;
+		if (hw->phy_model != ICE_PHY_E830)
+			dev_info->tx_offload_capa &= ~RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP;
 		dev_info->flow_type_rss_offloads |= ICE_RSS_OFFLOAD_ALL;
 	}
 
-- 
2.43.0

[PATCH 5/6] net/iavf: derive Rx/Tx offload capabilities from path masks

[Ciara Loftus]

rx_offload_capa was missing RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM even though
the scalar Rx path supports it. Rather than adding the missing flag to the
Rx offload capabilities directly, derive rx_offload_capa from a mask of all
of the Rx path offload masks.

Similarly, tx_offload_capa was missing RTE_ETH_TX_OFFLOAD_UDP_TSO despite
the Tx path already supporting and advertising this offload. Derive the
tx offload capabilities mask from a mask of all of the Tx path offload
masks.

A benefit of this is that if an offload is ever added to a path, it is
automatically reflected in the advertised capabilities too, instead of
relying on the capabilities masks in dev_info being kept manually in sync.
Some overlap will be present when masking the masks but it is kept because
relying on one path's offload mask always being a superset of another's is
fragile.

Fixes: 3b8c645afa06 ("net/iavf: fix outer checksum flags")
Fixes: 1e728b01120c ("net/iavf: rework Tx path")
Cc: stable@dpdk.org

Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
---
 drivers/net/intel/iavf/iavf_ethdev.c | 34 +++++++++-------------------
 1 file changed, 11 insertions(+), 23 deletions(-)

diff --git a/drivers/net/intel/iavf/iavf_ethdev.c b/drivers/net/intel/iavf/iavf_ethdev.c
index a8daa507bbb..b8ca1a19e3e 100644
--- a/drivers/net/intel/iavf/iavf_ethdev.c
+++ b/drivers/net/intel/iavf/iavf_ethdev.c
@@ -1202,31 +1202,19 @@ iavf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 			RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 			RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 	dev_info->rx_offload_capa =
-		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
-		RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
-		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
-		RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
-		RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
-		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
-		RTE_ETH_RX_OFFLOAD_SCATTER |
-		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
-		RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
-		RTE_ETH_RX_OFFLOAD_RSS_HASH;
+		(IAVF_RX_SCALAR_OFFLOADS | IAVF_RX_SCALAR_FLEX_OFFLOADS |
+		 IAVF_RX_VECTOR_OFFLOADS | IAVF_RX_VECTOR_OFFLOAD_OFFLOADS |
+		 IAVF_RX_VECTOR_FLEX_OFFLOADS | IAVF_RX_VECTOR_OFFLOAD_FLEX_OFFLOADS) &
+		~(RTE_ETH_RX_OFFLOAD_KEEP_CRC |
+		  RTE_ETH_RX_OFFLOAD_TIMESTAMP |
+		  RTE_ETH_RX_OFFLOAD_SECURITY);
 
 	dev_info->tx_offload_capa =
-		RTE_ETH_TX_OFFLOAD_VLAN_INSERT |
-		RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
-		RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
-		RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
-		RTE_ETH_TX_OFFLOAD_SCTP_CKSUM |
-		RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM |
-		RTE_ETH_TX_OFFLOAD_TCP_TSO |
-		RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |
-		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_MULTI_SEGS |
-		RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
+		(IAVF_TX_SCALAR_OFFLOADS | CI_TX_VECTOR_OFFLOADS |
+		 CI_TX_VEC_OFFLOAD_PATH_OFFLOADS | CI_TX_VEC_CTX_OFFLOAD_PATH_OFFLOADS) &
+		~(RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM |
+		  RTE_ETH_TX_OFFLOAD_QINQ_INSERT |
+		  RTE_ETH_TX_OFFLOAD_SECURITY);
 
 	/* X710 does not support outer udp checksum */
 	if (adapter->hw.mac.type != IAVF_MAC_XL710)
-- 
2.43.0

[PATCH 6/6] net/i40e: derive Rx/Tx offload capabilities from path masks

[Ciara Loftus]

The Rx and Tx offload capability masks duplicated the offload lists already
maintained for tracking the offloads enabled on the various scalar and
vector Rx and Tx paths. Instead of defining explicit lists for the offload
capabilities masks, instead derive the capabilities from the path masks.

A benefit of this is that if an offload is ever added to a path, it is
automatically reflected in the advertised capabilities too, instead of
relying on the capabilities masks in dev_info being kept manually in sync.
Some overlap will be present when masking the masks but it is kept because
relying on one path's offload mask always being a superset of another's is
fragile.

Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
---
 drivers/net/intel/i40e/i40e_ethdev.c | 29 +++-------------------------
 1 file changed, 3 insertions(+), 26 deletions(-)

diff --git a/drivers/net/intel/i40e/i40e_ethdev.c b/drivers/net/intel/i40e/i40e_ethdev.c
index defdff0378f..5d6102cae7b 100644
--- a/drivers/net/intel/i40e/i40e_ethdev.c
+++ b/drivers/net/intel/i40e/i40e_ethdev.c
@@ -3865,35 +3865,12 @@ i40e_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 	dev_info->max_mtu = dev_info->max_rx_pktlen - I40E_ETH_OVERHEAD;
 	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
 	dev_info->rx_queue_offload_capa = 0;
-	dev_info->rx_offload_capa =
-		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
-		RTE_ETH_RX_OFFLOAD_QINQ_STRIP |
-		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
-		RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
-		RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
-		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
-		RTE_ETH_RX_OFFLOAD_KEEP_CRC |
-		RTE_ETH_RX_OFFLOAD_SCATTER |
-		RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
-		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
-		RTE_ETH_RX_OFFLOAD_RSS_HASH;
+	dev_info->rx_offload_capa = I40E_RX_SCALAR_OFFLOADS | I40E_RX_VECTOR_OFFLOADS;
 
 	dev_info->tx_queue_offload_capa = RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
 	dev_info->tx_offload_capa =
-		RTE_ETH_TX_OFFLOAD_VLAN_INSERT |
-		RTE_ETH_TX_OFFLOAD_QINQ_INSERT |
-		RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
-		RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
-		RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
-		RTE_ETH_TX_OFFLOAD_SCTP_CKSUM |
-		RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM |
-		RTE_ETH_TX_OFFLOAD_TCP_TSO |
-		RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |
-		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_MULTI_SEGS |
-		dev_info->tx_queue_offload_capa;
+		(I40E_TX_SCALAR_OFFLOADS | I40E_TX_VECTOR_OFFLOADS) &
+		~RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM;
 	if (hw->mac.type == I40E_MAC_X722) {
 		dev_info->tx_offload_capa |=
 			RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM;
-- 
2.43.0

Re: [PATCH 0/6] Fix inconsistencies in offload masks

[Burakov, Anatoly]

On 9/24/2026 3:13 PM, Ciara Loftus wrote:
> This series fixes some inconsistencies in the path offload masks and
> dev_info offload capabilities for the ice, iavf and i40e drivers.
> The first 3 patches are concerned with fixing the path offload masks.
> The last 3 patches change how offloads are derived for the dev_info offload
> capabilities. Instead of defining an explicit list of offloads, reuse the
> information that is already available in the path offload masks. It should
> help prevent further inconsistencies in the future.
> 

Series-acked-by: Anatoly Burakov <anatoly.burakov@intel.com>

-- 
Thanks,
Anatoly

Re: [PATCH 0/6] Fix inconsistencies in offload masks

[Bruce Richardson]

On Fri, Sep 25, 2026 at 12:50:14PM +0200, Burakov, Anatoly wrote:
> On 9/24/2026 3:13 PM, Ciara Loftus wrote:
> > This series fixes some inconsistencies in the path offload masks and
> > dev_info offload capabilities for the ice, iavf and i40e drivers.
> > The first 3 patches are concerned with fixing the path offload masks.
> > The last 3 patches change how offloads are derived for the dev_info offload
> > capabilities. Instead of defining an explicit list of offloads, reuse the
> > information that is already available in the path offload masks. It should
> > help prevent further inconsistencies in the future.
> > 
> 
> Series-acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
> 
Series applied to dpdk-next-net-intel

Thanks,
/Bruce