If mlx5_dev_link_status_handler() is executed while canceling the alarm,
deadlock can happen because rte_eal_alarm_cancel() waits for all callbackes
to finish execution and both calls are protected by priv->lock.

Fixes: 198a3c339a8f ("mlx5: handle link status interrupts")
Cc: sta...@dpdk.org

Signed-off-by: Yongseok Koh <ys...@mellanox.com>
---
 drivers/net/mlx5/mlx5.h        | 16 ++++++++++++++++
 drivers/net/mlx5/mlx5_ethdev.c | 13 +++++++++----
 2 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index e6a69b823..1a4f444c7 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -165,6 +165,22 @@ priv_lock(struct priv *priv)
 }
 
 /**
+ * Try to lock private structure to protect it from concurrent access in the
+ * control path.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ *
+ * @return
+ *   1 if the lock is successfully taken; 0 otherwise.
+ */
+static inline int
+priv_trylock(struct priv *priv)
+{
+       return rte_spinlock_trylock(&priv->lock);
+}
+
+/**
  * Unlock private structure.
  *
  * @param priv
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 282ef241e..9354dd17f 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -1185,8 +1185,12 @@ mlx5_dev_link_status_handler(void *arg)
        struct priv *priv = dev->data->dev_private;
        int ret;
 
-       priv_lock(priv);
-       assert(priv->pending_alarm == 1);
+       while (!priv_trylock(priv)) {
+               /* Alarm is being canceled. */
+               if (priv->pending_alarm == 0)
+                       return;
+               rte_pause();
+       }
        priv->pending_alarm = 0;
        ret = priv_link_status_update(priv);
        priv_unlock(priv);
@@ -1256,9 +1260,10 @@ priv_dev_interrupt_handler_uninstall(struct priv *priv, 
struct rte_eth_dev *dev)
        if (priv->primary_socket)
                rte_intr_callback_unregister(&priv->intr_handle_socket,
                                             mlx5_dev_handler_socket, dev);
-       if (priv->pending_alarm)
+       if (priv->pending_alarm) {
+               priv->pending_alarm = 0;
                rte_eal_alarm_cancel(mlx5_dev_link_status_handler, dev);
-       priv->pending_alarm = 0;
+       }
        priv->intr_handle.fd = 0;
        priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
        priv->intr_handle_socket.fd = 0;
-- 
2.11.0

Reply via email to