net/mlx5: Fix error handling in irq_pool_request_irq

In case mlx5_irq_alloc fails, the previously allocated index remains
in the XArray, which could lead to inconsistencies.

Fix it by adding error handling that erases the allocated index
from the XArray if mlx5_irq_alloc returns an error.

Fixes: c36326d ("net/mlx5: Round-Robin EQs over IRQs")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Maher Sanalla <msanalla@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
Signed-off-by: NipaLocal <nipa@local>

Author: shayshyi

drivers/net/ethernet/mellanox/mlx5/core/irq_affinity.c

@@ -48,6 +48,7 @@ static struct mlx5_irq *
 irq_pool_request_irq(struct mlx5_irq_pool *pool, struct irq_affinity_desc *af_desc)
 {
 	struct irq_affinity_desc auto_desc = {};
+	struct mlx5_irq *irq;
 	u32 irq_index;
 	int err;
 
@@ -64,9 +65,12 @@ irq_pool_request_irq(struct mlx5_irq_pool *pool, struct irq_affinity_desc *af_de
 		else
 			cpu_get(pool, cpumask_first(&af_desc->mask));
 	}
-	return mlx5_irq_alloc(pool, irq_index,
-			      cpumask_empty(&auto_desc.mask) ? af_desc : &auto_desc,
-			      NULL);
+	irq = mlx5_irq_alloc(pool, irq_index,
+			     cpumask_empty(&auto_desc.mask) ? af_desc : &auto_desc,
+			     NULL);
+	if (IS_ERR(irq))
+		xa_erase(&pool->irqs, irq_index);
+	return irq;
 }
 
 /* Looking for the IRQ with the smallest refcount that fits req_mask.