Skip to contentSkip to conversation

qed: Fix race condition between scheduling and destroying the slowpath workqueue

PATCH Applied Yuval Basson 2 participants 2 messages Mar 25, 2020 – Mar 30, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 qed: Fix race condition between scheduling and destroying the slowpath workqueue

Calling queue_delayed_work concurrently with destroy_workqueue might race to an unexpected outcome - scheduled task after wq is destroyed or other resources (like ptt_pool) are freed (yields NULL pointer dereference). cancel_delayed_work prevents the race by cancelling the timer triggered for scheduling a new task.

Fixes 59ccf86fe ("qed: Add driver infrastucture for handling mfw requests")

---

Changes in v2:
- This patch is sent as a stand-alone after being part of a patch series where the first two patches were dropped https://www.spinics.net/lists/netdev/msg639940.html.

 drivers/net/ethernet/qlogic/qed/qed_main.c | 13 ++-----------
 1 file changed, 2 insertions(+), 11 deletions(-)
drivers/net/ethernet/qlogic/qed/qed_main.c +2 -11
index 2c189c6..96356e8 100644
@@ -1087,9 +1087,6 @@ static void qed_update_pf_params(struct qed_dev *cdev,
 #define QED_PERIODIC_DB_REC_INTERVAL_MS		100
 #define QED_PERIODIC_DB_REC_INTERVAL \
 	msecs_to_jiffies(QED_PERIODIC_DB_REC_INTERVAL_MS)
-#define QED_PERIODIC_DB_REC_WAIT_COUNT		10
-#define QED_PERIODIC_DB_REC_WAIT_INTERVAL \
-	(QED_PERIODIC_DB_REC_INTERVAL_MS / QED_PERIODIC_DB_REC_WAIT_COUNT)
 
 static int qed_slowpath_delayed_work(struct qed_hwfn *hwfn,
 				     enum qed_slowpath_wq_flag wq_flag,
@@ -1123,7 +1120,7 @@ void qed_periodic_db_rec_start(struct qed_hwfn *p_hwfn)
 
 static void qed_slowpath_wq_stop(struct qed_dev *cdev)
 {
-	int i, sleep_count = QED_PERIODIC_DB_REC_WAIT_COUNT;
+	int i;
 
 	if (IS_VF(cdev))
 		return;
@@ -1135,13 +1132,7 @@ static void qed_slowpath_wq_stop(struct qed_dev *cdev)
 		/* Stop queuing new delayed works */
 		cdev->hwfns[i].slowpath_wq_active = false;
 
-		/* Wait until the last periodic doorbell recovery is executed */
-		while (test_bit(QED_SLOWPATH_PERIODIC_DB_REC,
-				&cdev->hwfns[i].slowpath_task_flags) &&
-		       sleep_count--)
-			msleep(QED_PERIODIC_DB_REC_WAIT_INTERVAL);
-
-		flush_workqueue(cdev->hwfns[i].slowpath_wq);
+		cancel_delayed_work(&cdev->hwfns[i].slowpath_task);
 		destroy_workqueue(cdev->hwfns[i].slowpath_wq);
 	}
 }
Signature · 1 line
1.8.3.1
maintainer Applied
#1 lore raw

From: Yuval Basson <[email protected]>
Date: Wed, 25 Mar 2020 22:50:43 +0200

Quoted 11 lines

Calling queue_delayed_work concurrently with destroy_workqueue might race to an unexpected outcome - scheduled task after wq is destroyed or other resources (like ptt_pool) are freed (yields NULL pointer dereference). cancel_delayed_work prevents the race by cancelling the timer triggered for scheduling a new task.

Fixes 59ccf86fe ("qed: Add driver infrastucture for handling mfw requests")

Applied.