[PATCH] net/mlx5: store rxq MTU at allocation time

[PATCH] net/mlx5: store rxq MTU at allocation time

[a.schollmeyer]

From: Adrian Schollmeyer <a.schollmeyer@syseleven.de>

For shared Rx queues, equal MTU for all ports sharing queues is enforced
using mlx5_shared_rxq_match() to make sure, the memory allocated in the
Rx buffer is large enough. The check uses the MTU as reported by the
ports' private dev_data structs, which contain the MTU currently set for
the device. In case one port's MTU is altered after Rx queues are
allocated and then a second port joins the shared Rx queue with the old,
yet correct MTU, the check fails despite the fact that the Rx buffer
size is correct for both ports.

This patch adds a new entry to the Rx queue control structure that
captures the MTU at the time the Rx buffer was allocated, since this is
the relevant information that needs to be checked when a port joins a
shared Rx queue.

Fixes: 09c2555303be ("net/mlx5: support shared Rx queue")

Signed-off-by: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
---
 drivers/net/mlx5/mlx5_rx.h  | 1 +
 drivers/net/mlx5/mlx5_rxq.c | 6 +++++-
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 6380895502..58bc2c9f21 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -169,6 +169,7 @@ struct __rte_cache_aligned mlx5_rxq_data {
 /* RX queue control descriptor. */
 struct mlx5_rxq_ctrl {
 	struct mlx5_rxq_data rxq; /* Data path structure. */
+	uint16_t mtu; /* Original MTU that the queue was allocated with. */
 	LIST_ENTRY(mlx5_rxq_ctrl) next; /* Pointer to the next element. */
 	LIST_HEAD(priv, mlx5_rxq_priv) owners; /* Owner rxq list. */
 	struct mlx5_rxq_obj *obj; /* Verbs/DevX elements. */
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 5cf7d4971b..c652204ea8 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -773,7 +773,7 @@ mlx5_shared_rxq_match(struct mlx5_rxq_ctrl *rxq_ctrl, struct rte_eth_dev *dev,
 			dev->data->port_id, idx);
 		return false;
 	}
-	if (priv->mtu != spriv->mtu) {
+	if (priv->mtu != rxq_ctrl->mtu) {
 		DRV_LOG(ERR, "port %u queue index %u failed to join shared group: mtu mismatch",
 			dev->data->port_id, idx);
 		return false;
@@ -1799,6 +1799,10 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 	}
 	LIST_INIT(&tmpl->owners);
 	MLX5_ASSERT(n_seg && n_seg <= MLX5_MAX_RXQ_NSEG);
+	/*
+	 * Save the original MTU to check against for shared rx queues.
+	 */
+	tmpl->mtu = dev->data->mtu;
 	/*
 	 * Save the original segment configuration in the shared queue
 	 * descriptor for the later check on the sibling queue creation.
-- 
2.34.1

Re: [PATCH] net/mlx5: store rxq MTU at allocation time

[Dariusz Sosnowski]

Hi,

Thank you very much for your contribution.

Code looks good to me. Please see a small comment below.

On Tue, Oct 28, 2025 at 11:01:44AM +0100, a.schollmeyer@syseleven.de wrote:
> From: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
> 
> For shared Rx queues, equal MTU for all ports sharing queues is enforced
> using mlx5_shared_rxq_match() to make sure, the memory allocated in the
> Rx buffer is large enough. The check uses the MTU as reported by the
> ports' private dev_data structs, which contain the MTU currently set for
> the device. In case one port's MTU is altered after Rx queues are
> allocated and then a second port joins the shared Rx queue with the old,
> yet correct MTU, the check fails despite the fact that the Rx buffer
> size is correct for both ports.
> 
> This patch adds a new entry to the Rx queue control structure that
> captures the MTU at the time the Rx buffer was allocated, since this is
> the relevant information that needs to be checked when a port joins a
> shared Rx queue.
> 
> Fixes: 09c2555303be ("net/mlx5: support shared Rx queue")

The offending patch was part of 21.11,
so this fix is a candidate for backporting to previously released LTS versions.

Could you send a v2 patch with "Cc: stable@dpdk.org" added after "Fixes" tag?

When sending this change, could you also please send the patch directly to mlx5 maintainers as well?
There's a script in DPDK repository which would add relevant people
based on the changed files - devtools/get-maintainer.sh
It's usage is described in here: https://doc.dpdk.org/guides/contributing/patches.html#sending-patches

> 
> Signed-off-by: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
> ---
>  drivers/net/mlx5/mlx5_rx.h  | 1 +
>  drivers/net/mlx5/mlx5_rxq.c | 6 +++++-
>  2 files changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
> index 6380895502..58bc2c9f21 100644
> --- a/drivers/net/mlx5/mlx5_rx.h
> +++ b/drivers/net/mlx5/mlx5_rx.h
> @@ -169,6 +169,7 @@ struct __rte_cache_aligned mlx5_rxq_data {
>  /* RX queue control descriptor. */
>  struct mlx5_rxq_ctrl {
>  	struct mlx5_rxq_data rxq; /* Data path structure. */
> +	uint16_t mtu; /* Original MTU that the queue was allocated with. */
>  	LIST_ENTRY(mlx5_rxq_ctrl) next; /* Pointer to the next element. */
>  	LIST_HEAD(priv, mlx5_rxq_priv) owners; /* Owner rxq list. */
>  	struct mlx5_rxq_obj *obj; /* Verbs/DevX elements. */
> diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
> index 5cf7d4971b..c652204ea8 100644
> --- a/drivers/net/mlx5/mlx5_rxq.c
> +++ b/drivers/net/mlx5/mlx5_rxq.c
> @@ -773,7 +773,7 @@ mlx5_shared_rxq_match(struct mlx5_rxq_ctrl *rxq_ctrl, struct rte_eth_dev *dev,
>  			dev->data->port_id, idx);
>  		return false;
>  	}
> -	if (priv->mtu != spriv->mtu) {
> +	if (priv->mtu != rxq_ctrl->mtu) {
>  		DRV_LOG(ERR, "port %u queue index %u failed to join shared group: mtu mismatch",
>  			dev->data->port_id, idx);
>  		return false;
> @@ -1799,6 +1799,10 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
>  	}
>  	LIST_INIT(&tmpl->owners);
>  	MLX5_ASSERT(n_seg && n_seg <= MLX5_MAX_RXQ_NSEG);
> +	/*
> +	 * Save the original MTU to check against for shared rx queues.
> +	 */
> +	tmpl->mtu = dev->data->mtu;
>  	/*
>  	 * Save the original segment configuration in the shared queue
>  	 * descriptor for the later check on the sibling queue creation.
> -- 
> 2.34.1
> 

Best regards,
Dariusz Sosnowski

[PATCH v2] net/mlx5: store rxq MTU at allocation time

[a.schollmeyer]

From: Adrian Schollmeyer <a.schollmeyer@syseleven.de>

For shared Rx queues, equal MTU for all ports sharing queues is enforced
using mlx5_shared_rxq_match() to make sure, the memory allocated in the
Rx buffer is large enough. The check uses the MTU as reported by the
ports' private dev_data structs, which contain the MTU currently set for
the device. In case one port's MTU is altered after Rx queues are
allocated and then a second port joins the shared Rx queue with the old,
yet correct MTU, the check fails despite the fact that the Rx buffer
size is correct for both ports.

This patch adds a new entry to the Rx queue control structure that
captures the MTU at the time the Rx buffer was allocated, since this is
the relevant information that needs to be checked when a port joins a
shared Rx queue.

Fixes: 09c2555303be ("net/mlx5: support shared Rx queue")
Cc: stable@dpdk.org

Signed-off-by: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
---
 drivers/net/mlx5/mlx5_rx.h  | 1 +
 drivers/net/mlx5/mlx5_rxq.c | 6 +++++-
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 6380895502..58bc2c9f21 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -169,6 +169,7 @@ struct __rte_cache_aligned mlx5_rxq_data {
 /* RX queue control descriptor. */
 struct mlx5_rxq_ctrl {
 	struct mlx5_rxq_data rxq; /* Data path structure. */
+	uint16_t mtu; /* Original MTU that the queue was allocated with. */
 	LIST_ENTRY(mlx5_rxq_ctrl) next; /* Pointer to the next element. */
 	LIST_HEAD(priv, mlx5_rxq_priv) owners; /* Owner rxq list. */
 	struct mlx5_rxq_obj *obj; /* Verbs/DevX elements. */
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 5cf7d4971b..c652204ea8 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -773,7 +773,7 @@ mlx5_shared_rxq_match(struct mlx5_rxq_ctrl *rxq_ctrl, struct rte_eth_dev *dev,
 			dev->data->port_id, idx);
 		return false;
 	}
-	if (priv->mtu != spriv->mtu) {
+	if (priv->mtu != rxq_ctrl->mtu) {
 		DRV_LOG(ERR, "port %u queue index %u failed to join shared group: mtu mismatch",
 			dev->data->port_id, idx);
 		return false;
@@ -1799,6 +1799,10 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 	}
 	LIST_INIT(&tmpl->owners);
 	MLX5_ASSERT(n_seg && n_seg <= MLX5_MAX_RXQ_NSEG);
+	/*
+	 * Save the original MTU to check against for shared rx queues.
+	 */
+	tmpl->mtu = dev->data->mtu;
 	/*
 	 * Save the original segment configuration in the shared queue
 	 * descriptor for the later check on the sibling queue creation.
-- 
2.34.1

Re: [PATCH v2] net/mlx5: store rxq MTU at allocation time

[Dariusz Sosnowski]

Hi,

Thank you very much for applying the changes from last review.

On Thu, Oct 30, 2025 at 10:13:13AM +0100, a.schollmeyer@syseleven.de wrote:
> From: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
> 
> For shared Rx queues, equal MTU for all ports sharing queues is enforced
> using mlx5_shared_rxq_match() to make sure, the memory allocated in the
> Rx buffer is large enough. The check uses the MTU as reported by the
> ports' private dev_data structs, which contain the MTU currently set for
> the device. In case one port's MTU is altered after Rx queues are
> allocated and then a second port joins the shared Rx queue with the old,
> yet correct MTU, the check fails despite the fact that the Rx buffer
> size is correct for both ports.
> 
> This patch adds a new entry to the Rx queue control structure that
> captures the MTU at the time the Rx buffer was allocated, since this is
> the relevant information that needs to be checked when a port joins a
> shared Rx queue.
> 
> Fixes: 09c2555303be ("net/mlx5: support shared Rx queue")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Adrian Schollmeyer <a.schollmeyer@syseleven.de>

Acked-by: Dariusz Sosnowski <dsosnowski@nvidia.com>

Best regards,
Dariusz Sosnowski

Re: [PATCH v2] net/mlx5: store rxq MTU at allocation time

[Dariusz Sosnowski]

On Thu, Oct 30, 2025 at 11:21:55AM +0100, Dariusz Sosnowski wrote:
> Hi,
> 
> Thank you very much for applying the changes from last review.
> 
> On Thu, Oct 30, 2025 at 10:13:13AM +0100, a.schollmeyer@syseleven.de wrote:
> > From: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
> > 
> > For shared Rx queues, equal MTU for all ports sharing queues is enforced
> > using mlx5_shared_rxq_match() to make sure, the memory allocated in the
> > Rx buffer is large enough. The check uses the MTU as reported by the
> > ports' private dev_data structs, which contain the MTU currently set for
> > the device. In case one port's MTU is altered after Rx queues are
> > allocated and then a second port joins the shared Rx queue with the old,
> > yet correct MTU, the check fails despite the fact that the Rx buffer
> > size is correct for both ports.
> > 
> > This patch adds a new entry to the Rx queue control structure that
> > captures the MTU at the time the Rx buffer was allocated, since this is
> > the relevant information that needs to be checked when a port joins a
> > shared Rx queue.
> > 
> > Fixes: 09c2555303be ("net/mlx5: support shared Rx queue")
> > Cc: stable@dpdk.org
> > 
> > Signed-off-by: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
> 
> Acked-by: Dariusz Sosnowski <dsosnowski@nvidia.com>

One other thing, unrelated to the patch content.

Could you please change the state of previous versions of this patch
to "Superseded" in Patchwork?
This would be very helpful since it reduces the amount of patches to track.

This requires creating the account on Patchwork - https://patches.dpdk.org/

Best regards,
Dariusz Sosnowski

Re: [PATCH v2] net/mlx5: store rxq MTU at allocation time

[Adrian Schollmeyer]

[-- Attachment #1: Type: text/plain, Size: 735 bytes --]

Hi,

Am Donnerstag, dem 30.10.2025 um 11:40 +0100 schrieb Dariusz Sosnowski:
> Could you please change the state of previous versions of this patch
> to "Superseded" in Patchwork?

done. I hope everything should be okay now.

Best regards,
Adrian

-- 
Adrian Schollmeyer

SysEleven GmbH 
Boxhagener Straße 80 
10245 Berlin 

T +49 30 / 23 32 012-0
F +49 30 / 61 67 55 5-0 

https:// www.syseleven.de 
https://www.linkedin.com/company/syseleven-gmbh 

Current system status always at: 
https://www.syseleven-status.net/ 

Company headquarters: Berlin 
Registered court: AG Berlin Charlottenburg, HRB 108571 Berlin 
Managing directors: Andreas Hermann, Jens Ihlenfeld, Jens Plogsties,
Andreas Rückriegel

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v2] net/mlx5: store rxq MTU at allocation time

[Dariusz Sosnowski]

On Thu, Oct 30, 2025 at 12:21:51PM +0100, Adrian Schollmeyer wrote:
> Hi,
> 
> Am Donnerstag, dem 30.10.2025 um 11:40 +0100 schrieb Dariusz Sosnowski:
> > Could you please change the state of previous versions of this patch
> > to "Superseded" in Patchwork?
> 
> done. I hope everything should be okay now.
> 
> Best regards,
> Adrian
> 
> -- 
> Adrian Schollmeyer
> 
> SysEleven GmbH 
> Boxhagener Straße 80 
> 10245 Berlin 
> 
> T +49 30 / 23 32 012-0
> F +49 30 / 61 67 55 5-0 
> 
> https:// www.syseleven.de 
> https://www.linkedin.com/company/syseleven-gmbh 
> 
> Current system status always at: 
> https://www.syseleven-status.net/ 
> 
> Company headquarters: Berlin 
> Registered court: AG Berlin Charlottenburg, HRB 108571 Berlin 
> Managing directors: Andreas Hermann, Jens Ihlenfeld, Jens Plogsties,
> Andreas Rückriegel

Everything's perfect. Thank you very much and thank you for the contribution.

Best regards,
Dariusz Sosnowski

Re: [PATCH v2] net/mlx5: store rxq MTU at allocation time

[Raslan Darawsheh]

Hi,


On 30/10/2025 11:13 AM, a.schollmeyer@syseleven.de wrote:
> From: Adrian Schollmeyer <a.schollmeyer@syseleven.de>
> 
> For shared Rx queues, equal MTU for all ports sharing queues is enforced
> using mlx5_shared_rxq_match() to make sure, the memory allocated in the
> Rx buffer is large enough. The check uses the MTU as reported by the
> ports' private dev_data structs, which contain the MTU currently set for
> the device. In case one port's MTU is altered after Rx queues are
> allocated and then a second port joins the shared Rx queue with the old,
> yet correct MTU, the check fails despite the fact that the Rx buffer
> size is correct for both ports.
> 
> This patch adds a new entry to the Rx queue control structure that
> captures the MTU at the time the Rx buffer was allocated, since this is
> the relevant information that needs to be checked when a port joins a
> shared Rx queue.
> 
> Fixes: 09c2555303be ("net/mlx5: support shared Rx queue")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Adrian Schollmeyer <a.schollmeyer@syseleven.de>

Patch applied to next-net-mlx,

Kindest regards
Raslan Darawsheh