DPDK patches and discussions
 help / color / mirror / Atom feed
From: Manish Kurup <manish.kurup@broadcom.com>
To: dev@dpdk.org
Cc: ajit.khaparde@broadcom.com,
	Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Subject: [PATCH 22/54] net/bnxt: fix default rss config
Date: Mon, 29 Sep 2025 20:35:32 -0400	[thread overview]
Message-ID: <20250930003604.87108-23-manish.kurup@broadcom.com> (raw)
In-Reply-To: <20250930003604.87108-1-manish.kurup@broadcom.com>

From: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>

On Thor2, the set of supported RSS hash types reported in
dev_info->flow_type_rss_offloads includes both IPV6 and
IPV6_FLOW_LABEL. However, hardware support is mutually exculsive.
That is, either hash type IPV6 or IPV6_FLOW_LABEL can be
specified, but not both.

But, an application (such as testpmd) that sets the default
hash types based on dev_info could end up specifying both bits
in rss_conf->rss_hf field. This results in an error when both
are set in HWRM_VNIC_RSS_CFG command.

testpmd> port config all rss default
bnxt_hwrm_vnic_rss_cfg_p5(): error 2:0:00000000:0000
Configuration of RSS hash at ethernet port 0 failed with
		error (22): Invalid argument.
bnxt_hwrm_vnic_rss_cfg_p5(): error 2:0:00000000:0000
Configuration of RSS hash at ethernet port 1 failed with
		error (22): Invalid argument.

Fix this issue by specifying only hash type IPV6_FLOW_LABEL in
the hwrm, when both are specified in the rte command. This works
since the hw includes IPV6 src and dst fields also in its hash
function when the type is IPV6_FLOW_LABEL.

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
---
 drivers/net/bnxt/bnxt_ethdev.c | 11 +++++++++++
 drivers/net/bnxt/bnxt_vnic.c   | 10 +++++++++-
 2 files changed, 20 insertions(+), 1 deletion(-)

diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
index 7edfd8ca89..7177941e09 100644
--- a/drivers/net/bnxt/bnxt_ethdev.c
+++ b/drivers/net/bnxt/bnxt_ethdev.c
@@ -2485,6 +2485,17 @@ static int bnxt_rss_hash_conf_get_op(struct rte_eth_dev *eth_dev,
 			memcpy(rss_conf->rss_key, vnic->rss_hash_key, len);
 		}
 		bnxt_hwrm_rss_to_rte_hash_conf(vnic, &rss_conf->rss_hf);
+		/* HASH_TYPE_IPV6_FLOW_LABEL and HASH_TYPE_IPV6 are mutually
+		 * exclusive in hardware. See related comments in
+		 * bnxt_rte_to_hwrm_hash_types(). If the cached user config
+		 * has both bits enabled, make sure both are reported in
+		 * conf_get_op().
+		 */
+		if (bp->rss_conf.rss_hf &&
+		    (bp->rss_conf.rss_hf &
+		    (RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_IPV6_FLOW_LABEL)) &&
+		    (rss_conf->rss_hf & RTE_ETH_RSS_IPV6_FLOW_LABEL))
+			rss_conf->rss_hf |= RTE_ETH_RSS_IPV6;
 		rss_conf->rss_hf |=
 			bnxt_hwrm_to_rte_rss_level(bp, vnic->hash_mode);
 	} else {
diff --git a/drivers/net/bnxt/bnxt_vnic.c b/drivers/net/bnxt/bnxt_vnic.c
index 637e9b9aa4..f0ec87198f 100644
--- a/drivers/net/bnxt/bnxt_vnic.c
+++ b/drivers/net/bnxt/bnxt_vnic.c
@@ -285,8 +285,16 @@ uint32_t bnxt_rte_to_hwrm_hash_types(uint64_t rte_type)
 		hwrm_type |= HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6;
 	if (rte_type & RTE_ETH_RSS_NONFRAG_IPV6_UDP)
 		hwrm_type |= HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6;
-	if (rte_type & RTE_ETH_RSS_IPV6_FLOW_LABEL)
+	if (rte_type & RTE_ETH_RSS_IPV6_FLOW_LABEL) {
+		/* HASH_TYPE_IPV6_FLOW_LABEL and HASH_TYPE_IPV6 are mutually
+		 * exclusive. If both bits are specified in rte_type, set only
+		 * HASH_TYPE_IPV6_FLOW_LABEL in hardware since it subsumes
+		 * HASH_TYPE_IPV6 (see comments for HASH_TYPE_IPV6_FLOW_LABEL
+		 * in hsi.h).
+		 */
 		hwrm_type |= HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6_FLOW_LABEL;
+		hwrm_type &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6;
+	}
 	if (rte_type & RTE_ETH_RSS_ESP)
 		hwrm_type |= HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_ESP_SPI_IPV4 |
 			     HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_ESP_SPI_IPV6;
-- 
2.39.5 (Apple Git-154)


  parent reply	other threads:[~2025-09-30  7:08 UTC|newest]

Thread overview: 55+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-09-30  0:35 [PATCH 00/54] bnxt patchset Manish Kurup
2025-09-30  0:35 ` [PATCH 01/54] net/bnxt/tf_ulp: add bnxt app data for 25.11 Manish Kurup
2025-09-30  0:35 ` [PATCH 02/54] net/bnxt: fix a NULL pointer dereference in bnxt_rep funcs Manish Kurup
2025-09-30  0:35 ` [PATCH 03/54] net/bnxt: enable vector mode processing Manish Kurup
2025-09-30  0:35 ` [PATCH 04/54] net/bnxt/tf_ulp: add meter stats support for Thor2 Manish Kurup
2025-09-30  0:35 ` [PATCH 05/54] net/bnxt/tf_core: dynamic UPAR support for THOR2 Manish Kurup
2025-09-30  0:35 ` [PATCH 06/54] net/bnxt/tf_core: fix the miscalculation of the lkup table pool Manish Kurup
2025-09-30  0:35 ` [PATCH 07/54] net/bnxt/tf_core: thor2 TF table scope sizing adjustments Manish Kurup
2025-09-30  0:35 ` [PATCH 08/54] net/bnxt/tf_ulp: add support for global identifiers Manish Kurup
2025-09-30  0:35 ` [PATCH 09/54] net/bnxt/tf_core: add support for multi instance Manish Kurup
2025-09-30  0:35 ` [PATCH 10/54] net/bnxt/tf_core: fix table scope free Manish Kurup
2025-09-30  0:35 ` [PATCH 11/54] net/bnxt/tf_core: fix vfr clean up and stats lockup Manish Kurup
2025-09-30  0:35 ` [PATCH 12/54] net/bnxt/tf_ulp: add support for special vxlan Manish Kurup
2025-09-30  0:35 ` [PATCH 13/54] net/bnxt/tf_ulp: increase shared pool size to 32 Manish Kurup
2025-09-30  0:35 ` [PATCH 14/54] next/bnxt/tf_ulp: truflow fixes for meter and mac_addr cache Manish Kurup
2025-09-30  0:35 ` [PATCH 15/54] net/bnxt/tf_ulp: add support for tcam priority update Manish Kurup
2025-09-30  0:35 ` [PATCH 16/54] net/bnxt/tf_ulp: hot upgrade support Manish Kurup
2025-09-30  0:35 ` [PATCH 17/54] net/bnxt/tf_core: tcam manager logical id free Manish Kurup
2025-09-30  0:35 ` [PATCH 18/54] net/bnxt/tf_ulp: fix stats counter memory initialization Manish Kurup
2025-09-30  0:35 ` [PATCH 19/54] net/bnxt: fix max VFs count for thor2 Manish Kurup
2025-09-30  0:35 ` [PATCH 20/54] net/bnxt/tf_ulp: ovs-dpdk packet drop observed with thor2 Manish Kurup
2025-09-30  0:35 ` [PATCH 21/54] net/bnxt/tf_ulp: fix seg fault when devargs argument missing Manish Kurup
2025-09-30  0:35 ` Manish Kurup [this message]
2025-09-30  0:35 ` [PATCH 23/54] net/bnxt/tf_ulp: enable support for global index table Manish Kurup
2025-09-30  0:35 ` [PATCH 24/54] net/bnxt/tf_core: fix build failure with flow scale option Manish Kurup
2025-09-30  0:35 ` [PATCH 25/54] net/bnxt: truflow remove redundant code for mpc init Manish Kurup
2025-09-30  0:35 ` [PATCH 26/54] net/bnxt/tf_ulp: optimize template enums Manish Kurup
2025-09-30  0:35 ` [PATCH 27/54] net/bnxt/tf_core: thor2 hot upgrade ungraceful quit crash Manish Kurup
2025-09-30  0:35 ` [PATCH 28/54] net/bnxt/tf_ulp: support MPLS packets Manish Kurup
2025-09-30  0:35 ` [PATCH 29/54] net/bnxt/tf_core: add backing store debug to dpdk Manish Kurup
2025-09-30  0:35 ` [PATCH 30/54] net/bnxt/tf_core: truflow global table scope Manish Kurup
2025-09-30  0:35 ` [PATCH 31/54] net/bnxt/tf_ulp: ulp parser support to handle gre key Manish Kurup
2025-09-30  0:35 ` [PATCH 32/54] net/bnxt/tf_core: handle out of order MPC completions Manish Kurup
2025-09-30  0:35 ` [PATCH 33/54] net/bnxt/tf_ulp: socket direct enable Manish Kurup
2025-09-30  0:35 ` [PATCH 34/54] net/bnxt: fix adding udp_tunnel_port Manish Kurup
2025-09-30  0:35 ` [PATCH 35/54] net/bnxt/tf_ulp: add non vfr mode capability Manish Kurup
2025-09-30  0:35 ` [PATCH 36/54] net/bnxt: avoid iova range check when external memory is used Manish Kurup
2025-09-30  0:35 ` [PATCH 37/54] net/bnxt: avoid potential segfault in VFR handling Manish Kurup
2025-09-30  0:35 ` [PATCH 38/54] net/bnxt/tf_ulp: change rte_mem_virt2iova to rte_mem_virt2phys Manish Kurup
2025-09-30  0:35 ` [PATCH 39/54] net/bnxt: thor2 truflow memory manager bug Manish Kurup
2025-09-30  0:35 ` [PATCH 40/54] net/bnxt: fix stats collection when rx queue is not set Manish Kurup
2025-09-30  0:35 ` [PATCH 41/54] net/bnxt: fix rss configuration when set to none Manish Kurup
2025-09-30  0:35 ` [PATCH 42/54] net/bnxt: packet drop after port stop and start Manish Kurup
2025-09-30  0:35 ` [PATCH 43/54] net/bnxt/tf_core: fix truflow crash on memory allocation failure Manish Kurup
2025-09-30  0:35 ` [PATCH 44/54] net/bnxt: truflow remove RTE devarg processing for mpc=1 Manish Kurup
2025-09-30  0:35 ` [PATCH 45/54] net/bnxt: add meson build options for TruFlow Manish Kurup
2025-09-30  0:35 ` [PATCH 46/54] net/bnxt: truflow HSI struct fixes Manish Kurup
2025-09-30  0:35 ` [PATCH 47/54] net/bnxt/tf_ulp: truflow add pf action handler Manish Kurup
2025-09-30  0:35 ` [PATCH 48/54] net/bnxt/tf_ulp: add support for unicast only feature Manish Kurup
2025-09-30  0:35 ` [PATCH 49/54] net/bnxt/tf_core: remove excessive debug logging Manish Kurup
2025-09-30  0:36 ` [PATCH 50/54] net/bnxt/tf_core: fix truflow PF init failure on sriov disabled Manish Kurup
2025-09-30  0:36 ` [PATCH 51/54] net/bnxt/tf_ulp: fixes to enable TF functionality Manish Kurup
2025-09-30  0:36 ` [PATCH 52/54] net/bnxt/tf_ulp: add feature bit rx miss handling Manish Kurup
2025-09-30  0:36 ` [PATCH 53/54] net/bnxt: add support for truflow promiscuous mode Manish Kurup
2025-09-30  0:36 ` [PATCH 54/54] net/bnxt/tf_ulp: remove Truflow DEBUG code Manish Kurup

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20250930003604.87108-23-manish.kurup@broadcom.com \
    --to=manish.kurup@broadcom.com \
    --cc=ajit.khaparde@broadcom.com \
    --cc=dev@dpdk.org \
    --cc=sriharsha.basavapatna@broadcom.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).