RDMA/IPoIB: Add tx timeout work to recover queue stop situation

As we sometime run into TX timeout from IPoIB, queue seems stopped
and can't recover. Diff with Mellanox OFED show Mellanox driver
has timeout work to recover in such case.

Add TX timeout work/NAPI work to recover such case.

Also increase the watchdog_timeo to 10 seconds, so more tolerant to
error.

Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Link: https://lore.kernel.org/r/20231121130316.126364-3-jinpu.wang@ionos.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

3 files changed, 60 insertions, 3 deletions

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 35e9c8a330e25..963e936da5e3a 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -351,10 +351,12 @@ struct ipoib_dev_priv {
 	struct workqueue_struct *wq;
 	struct delayed_work mcast_task;
 	struct work_struct carrier_on_task;
+	struct work_struct reschedule_napi_work;
 	struct work_struct flush_light;
 	struct work_struct flush_normal;
 	struct work_struct flush_heavy;
 	struct work_struct restart_task;
+	struct work_struct tx_timeout_work;
 	struct delayed_work ah_reap_task;
 	struct delayed_work neigh_reap_task;
 	struct ib_device *ca;
@@ -499,6 +501,7 @@ int ipoib_send(struct net_device *dev, struct sk_buff *skb,
 	       struct ib_ah *address, u32 dqpn);
 void ipoib_reap_ah(struct work_struct *work);
 
+void ipoib_napi_schedule_work(struct work_struct *work);
 struct ipoib_path *__path_find(struct net_device *dev, void *gid);
 void ipoib_mark_paths_invalid(struct net_device *dev);
 void ipoib_flush_paths(struct net_device *dev);
@@ -510,6 +513,7 @@ void ipoib_ib_tx_timer_func(struct timer_list *t);
 void ipoib_ib_dev_flush_light(struct work_struct *work);
 void ipoib_ib_dev_flush_normal(struct work_struct *work);
 void ipoib_ib_dev_flush_heavy(struct work_struct *work);
+void ipoib_ib_tx_timeout_work(struct work_struct *work);
 void ipoib_pkey_event(struct work_struct *work);
 void ipoib_ib_dev_cleanup(struct net_device *dev);
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 7f84d9866ceff..5cde275daa941 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -531,11 +531,35 @@ void ipoib_ib_rx_completion(struct ib_cq *cq, void *ctx_ptr)
 	napi_schedule(&priv->recv_napi);
 }
 
+/* The function will force napi_schedule */
+void ipoib_napi_schedule_work(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, reschedule_napi_work);
+	bool ret;
+
+	do {
+		ret = napi_schedule(&priv->send_napi);
+		if (!ret)
+			msleep(3);
+	} while (!ret && netif_queue_stopped(priv->dev) &&
+		 test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags));
+}
+
 void ipoib_ib_tx_completion(struct ib_cq *cq, void *ctx_ptr)
 {
 	struct ipoib_dev_priv *priv = ctx_ptr;
+	bool ret;
 
-	napi_schedule(&priv->send_napi);
+	ret = napi_schedule(&priv->send_napi);
+	/*
+	 * if the queue is closed the driver must be able to schedule napi,
+	 * otherwise we can end with closed queue forever, because no new
+	 * packets to send and napi callback might not get new event after
+	 * its re-arm of the napi.
+	 */
+	if (!ret && netif_queue_stopped(priv->dev))
+		schedule_work(&priv->reschedule_napi_work);
 }
 
 static inline int post_send(struct ipoib_dev_priv *priv,
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 967004ccad987..7a5be705d7183 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -1200,7 +1200,34 @@ static void ipoib_timeout(struct net_device *dev, unsigned int txqueue)
 		   netif_queue_stopped(dev), priv->tx_head, priv->tx_tail,
 		   priv->global_tx_head, priv->global_tx_tail);
 
-	/* XXX reset QP, etc. */
+
+	schedule_work(&priv->tx_timeout_work);
+}
+
+void ipoib_ib_tx_timeout_work(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv = container_of(work,
+						   struct ipoib_dev_priv,
+						   tx_timeout_work);
+	int err;
+
+	rtnl_lock();
+
+	if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
+		goto unlock;
+
+	ipoib_stop(priv->dev);
+	err = ipoib_open(priv->dev);
+	if (err) {
+		ipoib_warn(priv, "ipoib_open failed recovering from a tx_timeout, err(%d).\n",
+				err);
+		goto unlock;
+	}
+
+	netif_tx_wake_all_queues(priv->dev);
+unlock:
+	rtnl_unlock();
+
 }
 
 static int ipoib_hard_header(struct sk_buff *skb,
@@ -2112,7 +2139,7 @@ void ipoib_setup_common(struct net_device *dev)
 
 	ipoib_set_ethtool_ops(dev);
 
-	dev->watchdog_timeo	 = HZ;
+	dev->watchdog_timeo	 = 10 * HZ;
 
 	dev->flags		|= IFF_BROADCAST | IFF_MULTICAST;
 
@@ -2150,10 +2177,12 @@ static void ipoib_build_priv(struct net_device *dev)
 
 	INIT_DELAYED_WORK(&priv->mcast_task,   ipoib_mcast_join_task);
 	INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
+	INIT_WORK(&priv->reschedule_napi_work, ipoib_napi_schedule_work);
 	INIT_WORK(&priv->flush_light,   ipoib_ib_dev_flush_light);
 	INIT_WORK(&priv->flush_normal,   ipoib_ib_dev_flush_normal);
 	INIT_WORK(&priv->flush_heavy,   ipoib_ib_dev_flush_heavy);
 	INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
+	INIT_WORK(&priv->tx_timeout_work, ipoib_ib_tx_timeout_work);
 	INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
 	INIT_DELAYED_WORK(&priv->neigh_reap_task, ipoib_reap_neigh);
 }