Since the CPU selected by schedule_work is uncertain, multiple link_event
callbacks may be executed at same time. For example, after peer's link
is up, it is down quickly before local link_work completed. If link_cleanup
is added to the workqueue of another CPU, then link_work and link_cleanup
may be executed at the same time. So add a mutex to prevent them from being
executed concurrently.

Signed-off-by: fuyuanli <fuyuanli@didiglobal.com>
---
 drivers/ntb/ntb_transport.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
index 4f775c3e218f..902968e24c7a 100644
--- a/drivers/ntb/ntb_transport.c
+++ b/drivers/ntb/ntb_transport.c
@@ -59,6 +59,7 @@
 #include <linux/slab.h>
 #include <linux/types.h>
 #include <linux/uaccess.h>
+#include <linux/mutex.h>
 #include "linux/ntb.h"
 #include "linux/ntb_transport.h"
 
@@ -241,6 +242,9 @@ struct ntb_transport_ctx {
 	struct work_struct link_cleanup;
 
 	struct dentry *debugfs_node_dir;
+
+	/* Make sure workq of link event be executed serially */
+	struct mutex link_event_lock;
 };
 
 enum {
@@ -1024,7 +1028,9 @@ static void ntb_transport_link_cleanup_work(struct work_struct *work)
 	struct ntb_transport_ctx *nt =
 		container_of(work, struct ntb_transport_ctx, link_cleanup);
 
+	mutex_lock(&nt->link_event_lock);
 	ntb_transport_link_cleanup(nt);
+	mutex_unlock(&nt->link_event_lock);
 }
 
 static void ntb_transport_event_callback(void *data)
@@ -1047,6 +1053,8 @@ static void ntb_transport_link_work(struct work_struct *work)
 	u32 val;
 	int rc = 0, i, spad;
 
+	mutex_lock(&nt->link_event_lock);
+
 	/* send the local info, in the opposite order of the way we read it */
 
 	if (nt->use_msi) {
@@ -1125,6 +1133,7 @@ static void ntb_transport_link_work(struct work_struct *work)
 			schedule_delayed_work(&qp->link_work, 0);
 	}
 
+	mutex_unlock(&nt->link_event_lock);
 	return;
 
 out1:
@@ -1132,10 +1141,13 @@ static void ntb_transport_link_work(struct work_struct *work)
 		ntb_free_mw(nt, i);
 
 	/* if there's an actual failure, we should just bail */
-	if (rc < 0)
+	if (rc < 0) {
+		mutex_unlock(&nt->link_event_lock);
 		return;
+	}
 
 out:
+	mutex_unlock(&nt->link_event_lock);
 	if (ntb_link_is_up(ndev, NULL, NULL) == 1)
 		schedule_delayed_work(&nt->link_work,
 				      msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
-- 
2.34.1

On 8/22/25 7:41 AM, fuyuanli wrote:
> Since the CPU selected by schedule_work is uncertain, multiple link_event
> callbacks may be executed at same time. For example, after peer's link
> is up, it is down quickly before local link_work completed. If link_cleanup
> is added to the workqueue of another CPU, then link_work and link_cleanup
> may be executed at the same time. So add a mutex to prevent them from being
> executed concurrently.
> 
> Signed-off-by: fuyuanli <fuyuanli@didiglobal.com>
> ---
>  drivers/ntb/ntb_transport.c | 14 +++++++++++++-
>  1 file changed, 13 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
> index 4f775c3e218f..902968e24c7a 100644
> --- a/drivers/ntb/ntb_transport.c
> +++ b/drivers/ntb/ntb_transport.c
> @@ -59,6 +59,7 @@
>  #include <linux/slab.h>
>  #include <linux/types.h>
>  #include <linux/uaccess.h>
> +#include <linux/mutex.h>
>  #include "linux/ntb.h"
>  #include "linux/ntb_transport.h"
>  
> @@ -241,6 +242,9 @@ struct ntb_transport_ctx {
>  	struct work_struct link_cleanup;
>  
>  	struct dentry *debugfs_node_dir;
> +
> +	/* Make sure workq of link event be executed serially */
> +	struct mutex link_event_lock;
>  };
>  
>  enum {
> @@ -1024,7 +1028,9 @@ static void ntb_transport_link_cleanup_work(struct work_struct *work)
>  	struct ntb_transport_ctx *nt =
>  		container_of(work, struct ntb_transport_ctx, link_cleanup);
>  
> +	mutex_lock(&nt->link_event_lock);

Can you please use guard() instead? Should produce less code and not worry about calling all the unlocks.

DJ

>  	ntb_transport_link_cleanup(nt);
> +	mutex_unlock(&nt->link_event_lock);
>  }
>  
>  static void ntb_transport_event_callback(void *data)
> @@ -1047,6 +1053,8 @@ static void ntb_transport_link_work(struct work_struct *work)
>  	u32 val;
>  	int rc = 0, i, spad;
>  
> +	mutex_lock(&nt->link_event_lock);
> +
>  	/* send the local info, in the opposite order of the way we read it */
>  
>  	if (nt->use_msi) {
> @@ -1125,6 +1133,7 @@ static void ntb_transport_link_work(struct work_struct *work)
>  			schedule_delayed_work(&qp->link_work, 0);
>  	}
>  
> +	mutex_unlock(&nt->link_event_lock);
>  	return;
>  
>  out1:
> @@ -1132,10 +1141,13 @@ static void ntb_transport_link_work(struct work_struct *work)
>  		ntb_free_mw(nt, i);
>  
>  	/* if there's an actual failure, we should just bail */
> -	if (rc < 0)
> +	if (rc < 0) {
> +		mutex_unlock(&nt->link_event_lock);
>  		return;
> +	}
>  
>  out:
> +	mutex_unlock(&nt->link_event_lock);
>  	if (ntb_link_is_up(ndev, NULL, NULL) == 1)
>  		schedule_delayed_work(&nt->link_work,
>  				      msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));