Skip to contentSkip to conversation

net/mlx5: fix skipping PF representors

PATCH In review Dariusz Sosnowski 2 participants 2 messages Apr 7 – Apr 9 Network drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/mlx5: fix skipping PF representors

Offending patch changed logic of matching IB ports to requested representors in mlx5 driver.
Each found IB port was matched against all requested representors. Whenever:

- requested representor was VF or SF
- PF was not ignored
- IB port was physical port
- physical port index matched requested PF index

the physical port representor was probed.
If any of the above is false, mlx5 driver should have continue testing other requested PF representors.
In the offending patch, the representor matching loop was stopped. As a result, if mlx5 device with MPESW enabled was probed with the following devargs:

	-a 08:00.0,dv_flow_en=2,representor=pf[0-1]vf[0,1]

Only 5 ports were probed (physical port 0 and all VF representors), instead of 6 ports (physical port 1 is missing).

This patch fixes that by continuing representor matching loop on physical port to PF index mismatch as described above.

Fixes 2f7cdd821b1b ("net/mlx5: fix probing to allow BlueField Socket Direct")
---
 drivers/net/mlx5/linux/mlx5_os.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/mlx5/linux/mlx5_os.c +1 -1
index a717191002..10b947116c 100644
@@ -1186,7 +1186,7 @@ representor_match_port(const struct mlx5_dev_spawn_data *spawn,
 
 		/* Uplink ports should not be matched against representor_ports. */
 		if (spawn->info.name_type == MLX5_PHYS_PORT_NAME_TYPE_UPLINK)
-			return false;
+			continue;
 
 		for (uint16_t f = 0; f < eth_da->nb_representor_ports; ++f) {
 			uint16_t port_num = eth_da->representor_ports[f];
Signature · 1 line
2.47.3
#1 lore raw

Hi,

Quoted 22 lines · Dariusz Sosnowski

Offending patch changed logic of matching IB ports to requested representors in mlx5 driver.
Each found IB port was matched against all requested representors. Whenever:

- requested representor was VF or SF
- PF was not ignored
- IB port was physical port
- physical port index matched requested PF index

the physical port representor was probed.
If any of the above is false, mlx5 driver should have continue testing other requested PF representors.
In the offending patch, the representor matching loop was stopped. As a result, if mlx5 device with MPESW enabled was probed with the following devargs:

	-a 08:00.0,dv_flow_en=2,representor=pf[0-1]vf[0,1]

Only 5 ports were probed (physical port 0 and all VF representors), instead of 6 ports (physical port 1 is missing).

Quoted 8 lines

This patch fixes that by continuing representor matching loop on physical port to PF index mismatch as described above.

Fixes 2f7cdd821b1b ("net/mlx5: fix probing to allow BlueField Socket Direct")

Patch applied to next-net-mlx,

Kindest regards
Raslan Darawsheh