When calling the haipin unbind interface, all the hairpin Tx queues
of the port will be unbound from the peer Rx queues. If one of the
Tx queue is working in the auto bind mode, the interface will return
directly.
Only when the Tx and peer Rx ports are the same, the auto bind mode
is supported. In this condition branch, the Tx queue release is
missed and the reference count is not decreased. Then in the port
stop stage, the hardware resources of this Tx queue won't be
freed. There would be some assertion or failure when starting the
port again.
With this commit, the reference count will be operated correctly.
Fixes: 37cd4501e873 ("net/mlx5: support two ports hairpin mode")
Cc: sta...@dpdk.org
Signed-off-by: Bing Zhao <bi...@nvidia.com>
Acked-by: Matan Azrad <ma...@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viachesl...@nvidia.com>
---
drivers/net/mlx5/mlx5_trigger.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 3457bf65d3..bbaa7d2aa0 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -896,11 +896,11 @@ mlx5_hairpin_unbind_single_port(struct rte_eth_dev *dev,
uint16_t rx_port)
}
/* Indeed, only the first used queue needs to be checked. */
if (txq_ctrl->hairpin_conf.manual_bind == 0) {
+ mlx5_txq_release(dev, i);
if (cur_port != rx_port) {
rte_errno = EINVAL;
DRV_LOG(ERR, "port %u and port %u are in"
" auto-bind mode", cur_port, rx_port);
- mlx5_txq_release(dev, i);
return -rte_errno;
} else {
return 0;
--
2.31.1
