Skip to contentSkip to conversation

ice: always add legacy 32byte RXDID in supported_rxdids

PATCH Applied Michal Schmidt 4 participants 4 messages Sep 20, 2023 – Sep 28, 2023

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ice: always add legacy 32byte RXDID in supported_rxdids

When the PF and VF drivers both support flexible rx descriptors and have negotiated the VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC capability, the VF driver queries the PF for the list of supported descriptor formats (VIRTCHNL_OP_GET_SUPPORTED_RXDIDS). The PF driver is supposed to set the supported_rxdids bits that correspond to the descriptor formats the firmware implements. The legacy 32-byte rx desc format is always supported, even though it is not expressed in GLFLXP_RXDID_FLAGS.

The ice driver does not advertise the legacy 32-byte rx desc support, which leads to this failure to bring up the VF using the Intel out-of-tree iavf driver:
iavf 0000:41:01.0: PF does not list support for default Rx descriptor format
...
iavf 0000:41:01.0: PF returned error -5 (VIRTCHNL_STATUS_ERR_PARAM) to our request 6

The in-tree iavf driver does not expose this bug, because it does not yet implement VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC.

The ice driver must always set the ICE_RXDID_LEGACY_1 bit in supported_rxdids. The Intel out-of-tree ice driver and the ice driver in DPDK both do this.

I copied this piece of the code and the comment text from the Intel out-of-tree driver.

Fixes e753df8fbca5 ("ice: Add support Flex RXD")
---
 drivers/net/ethernet/intel/ice/ice_virtchnl.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)
drivers/net/ethernet/intel/ice/ice_virtchnl.c +7 -5
index b03426ac932b..db97353efd06 100644
@@ -2617,12 +2617,14 @@ static int ice_vc_query_rxdid(struct ice_vf *vf)
 		goto err;
 	}
 
-	/* Read flexiflag registers to determine whether the
-	 * corresponding RXDID is configured and supported or not.
-	 * Since Legacy 16byte descriptor format is not supported,
-	 * start from Legacy 32byte descriptor.
+	/* RXDIDs supported by DDP package can be read from the register
+	 * to get the supported RXDID bitmap. But the legacy 32byte RXDID
+	 * is not listed in DDP package, add it in the bitmap manually.
+	 * Legacy 16byte descriptor is not supported.
 	 */
-	for (i = ICE_RXDID_LEGACY_1; i < ICE_FLEX_DESC_RXDID_MAX_NUM; i++) {
+	rxdid->supported_rxdids |= BIT(ICE_RXDID_LEGACY_1);
+
+	for (i = ICE_RXDID_FLEX_NIC; i < ICE_FLEX_DESC_RXDID_MAX_NUM; i++) {
 		regval = rd32(hw, GLFLXP_RXDID_FLAGS(i, 0));
 		if ((regval >> GLFLXP_RXDID_FLAGS_FLEXIFLAG_4N_S)
 			& GLFLXP_RXDID_FLAGS_FLEXIFLAG_4N_M)
Signature · 1 line
2.41.0
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 22 lines · Michal Schmidt

When the PF and VF drivers both support flexible rx descriptors and have negotiated the VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC capability, the VF driver queries the PF for the list of supported descriptor formats (VIRTCHNL_OP_GET_SUPPORTED_RXDIDS). The PF driver is supposed to set the supported_rxdids bits that correspond to the descriptor formats the firmware implements. The legacy 32-byte rx desc format is always supported, even though it is not expressed in GLFLXP_RXDID_FLAGS.

The ice driver does not advertise the legacy 32-byte rx desc support, which leads to this failure to bring up the VF using the Intel out-of-tree iavf driver:
iavf 0000:41:01.0: PF does not list support for default Rx descriptor format
...
iavf 0000:41:01.0: PF returned error -5 (VIRTCHNL_STATUS_ERR_PARAM) to our request 6

The in-tree iavf driver does not expose this bug, because it does not yet implement VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC.

The ice driver must always set the ICE_RXDID_LEGACY_1 bit in supported_rxdids. The Intel out-of-tree ice driver and the ice driver in DPDK both do this.

Quoted 8 lines

I copied this piece of the code and the comment text from the Intel out-of-tree driver.

Fixes e753df8fbca5 ("ice: Add support Flex RXD")
---
  drivers/net/ethernet/intel/ice/ice_virtchnl.c | 12 +++++++-----
  1 file changed, 7 insertions(+), 5 deletions(-)

We have tried to upstream that as part of some big series "recently", with no luck so far, so thank for standalone submission!

Quoted 25 lines
drivers/net/ethernet/intel/ice/ice_virtchnl.c +7 -5
index b03426ac932b..db97353efd06 100644
@@ -2617,12 +2617,14 @@ static int ice_vc_query_rxdid(struct ice_vf *vf)
  		goto err;
  	}
 
-	/* Read flexiflag registers to determine whether the
-	 * corresponding RXDID is configured and supported or not.
-	 * Since Legacy 16byte descriptor format is not supported,
-	 * start from Legacy 32byte descriptor.
+	/* RXDIDs supported by DDP package can be read from the register
+	 * to get the supported RXDID bitmap. But the legacy 32byte RXDID
+	 * is not listed in DDP package, add it in the bitmap manually.
+	 * Legacy 16byte descriptor is not supported.
  	 */
-	for (i = ICE_RXDID_LEGACY_1; i < ICE_FLEX_DESC_RXDID_MAX_NUM; i++) {
+	rxdid->supported_rxdids |= BIT(ICE_RXDID_LEGACY_1);
+
+	for (i = ICE_RXDID_FLEX_NIC; i < ICE_FLEX_DESC_RXDID_MAX_NUM; i++) {
  		regval = rd32(hw, GLFLXP_RXDID_FLAGS(i, 0));
  		if ((regval >> GLFLXP_RXDID_FLAGS_FLEXIFLAG_4N_S)
  			& GLFLXP_RXDID_FLAGS_FLEXIFLAG_4N_M)
#2 lore raw
In reply to Michal Schmidt
Quoted 22 lines · Michal Schmidt

When the PF and VF drivers both support flexible rx descriptors and have negotiated the VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC capability, the VF driver queries the PF for the list of supported descriptor formats (VIRTCHNL_OP_GET_SUPPORTED_RXDIDS). The PF driver is supposed to set the supported_rxdids bits that correspond to the descriptor formats the firmware implements. The legacy 32-byte rx desc format is always supported, even though it is not expressed in GLFLXP_RXDID_FLAGS.

The ice driver does not advertise the legacy 32-byte rx desc support, which leads to this failure to bring up the VF using the Intel out-of-tree iavf driver:
iavf 0000:41:01.0: PF does not list support for default Rx descriptor format
...
iavf 0000:41:01.0: PF returned error -5 (VIRTCHNL_STATUS_ERR_PARAM) to our request 6

The in-tree iavf driver does not expose this bug, because it does not yet implement VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC.

The ice driver must always set the ICE_RXDID_LEGACY_1 bit in supported_rxdids. The Intel out-of-tree ice driver and the ice driver in DPDK both do this.

Quoted 5 lines

I copied this piece of the code and the comment text from the Intel out-of-tree driver.

Fixes e753df8fbca5 ("ice: Add support Flex RXD")

Yep, this happens because of the way the feature interacts with VIRTHCNL_VF_OFFLOAD_RX_FLEX_DESC and how that wasn't enabled in upstream yet. Not sure how this got missed with the mentioned fixed commit but probably a gap because the upstream iAVF didn't implement changing descriptor format.

Thanks for fixing this!

-Jake

Quoted 28 lines
---
 drivers/net/ethernet/intel/ice/ice_virtchnl.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)
drivers/net/ethernet/intel/ice/ice_virtchnl.c +7 -5
index b03426ac932b..db97353efd06 100644
@@ -2617,12 +2617,14 @@ static int ice_vc_query_rxdid(struct ice_vf *vf)
 		goto err;
 	}
 
-	/* Read flexiflag registers to determine whether the
-	 * corresponding RXDID is configured and supported or not.
-	 * Since Legacy 16byte descriptor format is not supported,
-	 * start from Legacy 32byte descriptor.
+	/* RXDIDs supported by DDP package can be read from the register
+	 * to get the supported RXDID bitmap. But the legacy 32byte RXDID
+	 * is not listed in DDP package, add it in the bitmap manually.
+	 * Legacy 16byte descriptor is not supported.
 	 */
-	for (i = ICE_RXDID_LEGACY_1; i < ICE_FLEX_DESC_RXDID_MAX_NUM; i++) {
+	rxdid->supported_rxdids |= BIT(ICE_RXDID_LEGACY_1);
+
+	for (i = ICE_RXDID_FLEX_NIC; i < ICE_FLEX_DESC_RXDID_MAX_NUM; i++) {
 		regval = rd32(hw, GLFLXP_RXDID_FLAGS(i, 0));
 		if ((regval >> GLFLXP_RXDID_FLAGS_FLEXIFLAG_4N_S)
 			& GLFLXP_RXDID_FLAGS_FLEXIFLAG_4N_M)
#3 lore raw
In reply to Michal Schmidt
This patch was applied to netdev/net.git (main) by Paolo Abeni <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by Paolo Abeni <[email protected]>:

Quoted 9 lines · On Wed, 20 Sep 2023 13:54:38 +0200 you wrote:

When the PF and VF drivers both support flexible rx descriptors and have negotiated the VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC capability, the VF driver queries the PF for the list of supported descriptor formats (VIRTCHNL_OP_GET_SUPPORTED_RXDIDS). The PF driver is supposed to set the supported_rxdids bits that correspond to the descriptor formats the firmware implements. The legacy 32-byte rx desc format is always supported, even though it is not expressed in GLFLXP_RXDID_FLAGS.

[...]

Here is the summary with links:
- [net] ice: always add legacy 32byte RXDID in supported_rxdids

    https://git.kernel.org/netdev/net/c/c070e51db5e2

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html