[PATCH] s390: fix endless retry loop in tape driver

If a tape device is assigned to another host, the interrupt for the assign operation comes back with deferred condition code 1. Under some conditions this can lead to an endless loop of retries. Check if the current request is still in IO in deferred condition code handling and prevent retries when the request has already been cancelled. Signed-off-by: Michael Holzheu <holzheu@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] s390: fix endless retry loop in tape driver
If a tape device is assigned to another host, the interrupt for the assign operation comes back with deferred condition code 1. Under some conditions this can lead to an endless loop of retries. Check if the current request is still in IO in deferred condition code handling and prevent retries when the request has already been cancelled. Signed-off-by: Michael Holzheu <holzheu@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Michael Holzheu · Linus Torvalds
1 parent 4cd190a736
Showing 3 changed files with 35 additions and 13 deletions Side-by-side Diff
drivers/s390/char/tape.h
drivers/s390/char/tape_core.c
drivers/s390/char/tape_std.c
@@ -250,6 +250,7 @@
 extern int tape_do_io(struct tape_device *, struct tape_request *);
 extern int tape_do_io_async(struct tape_device *, struct tape_request *);
 extern int tape_do_io_interruptible(struct tape_device *, struct tape_request *);
+extern int tape_cancel_io(struct tape_device *, struct tape_request *);
 void tape_hotplug_event(struct tape_device *, int major, int action);
  
 static inline int
@@ -761,6 +761,13 @@
 		 */
 		if (request->status == TAPE_REQUEST_IN_IO)
 			return;
+		/*
+		 * Request has already been stopped. We have to wait until
+		 * the request is removed from the queue in the interrupt
+		 * handling.
+		 */
+		if (request->status == TAPE_REQUEST_DONE)
+			return;
  
 		/*
 		 * We wanted to cancel the request but the common I/O layer
@@ -1024,6 +1031,20 @@
 }
  
 /*
+ * Stop running ccw.
+ */
+int
+tape_cancel_io(struct tape_device *device, struct tape_request *request)
+{
+	int rc;
+
+	spin_lock_irq(get_ccwdev_lock(device->cdev));
+	rc = __tape_cancel_io(device, request);
+	spin_unlock_irq(get_ccwdev_lock(device->cdev));
+	return rc;
+}
+
+/*
  * Tape interrupt routine, called from the ccw_device layer
  */
 static void
  
@@ -1068,12 +1089,12 @@
 	 * error might still apply. So we just schedule the request to be
 	 * started later.
 	 */
-	if (irb->scsw.cc != 0 && (irb->scsw.fctl & SCSW_FCTL_START_FUNC)) {
-		PRINT_WARN("(%s): deferred cc=%i. restaring\n",
-			cdev->dev.bus_id,
-			irb->scsw.cc);
+	if (irb->scsw.cc != 0 && (irb->scsw.fctl & SCSW_FCTL_START_FUNC) &&
+	    (request->status == TAPE_REQUEST_IN_IO)) {
+		DBF_EVENT(3,"(%08x): deferred cc=%i, fctl=%i. restarting\n",
+			device->cdev_id, irb->scsw.cc, irb->scsw.fctl);
 		request->status = TAPE_REQUEST_QUEUED;
-		schedule_work(&device->tape_dnr);
+		schedule_delayed_work(&device->tape_dnr, HZ);
 		return;
 	}
  
@@ -1287,5 +1308,6 @@
 EXPORT_SYMBOL(tape_do_io);
 EXPORT_SYMBOL(tape_do_io_async);
 EXPORT_SYMBOL(tape_do_io_interruptible);
+EXPORT_SYMBOL(tape_cancel_io);
 EXPORT_SYMBOL(tape_mtop);
@@ -37,20 +37,19 @@
 {
 	struct tape_request *	request;
 	struct tape_device *	device;
+	int rc;
  
 	request = (struct tape_request *) data;
 	if ((device = request->device) == NULL)
 		BUG();
  
-	spin_lock_irq(get_ccwdev_lock(device->cdev));
-	if (request->callback != NULL) {
-		DBF_EVENT(3, "%08x: Assignment timeout. Device busy.\n",
+	DBF_EVENT(3, "%08x: Assignment timeout. Device busy.\n",
 			device->cdev_id);
-		PRINT_ERR("%s: Assignment timeout. Device busy.\n",
-			device->cdev->dev.bus_id);
-		ccw_device_clear(device->cdev, (long) request);
-	}
-	spin_unlock_irq(get_ccwdev_lock(device->cdev));
+	rc = tape_cancel_io(device, request);
+	if(rc)
+		PRINT_ERR("(%s): Assign timeout: Cancel failed with rc = %i\n",
+			device->cdev->dev.bus_id, rc);
+
 }
  
 int
...	...	@@ -250,6 +250,7 @@
250	250	extern int tape_do_io(struct tape_device , struct tape_request );
251	251	extern int tape_do_io_async(struct tape_device , struct tape_request );
252	252	extern int tape_do_io_interruptible(struct tape_device , struct tape_request );
	253	+extern int tape_cancel_io(struct tape_device , struct tape_request );
253	254	void tape_hotplug_event(struct tape_device *, int major, int action);
254	255
255	256	static inline int
...	...	@@ -761,6 +761,13 @@
761	761	*/
762	762	if (request->status == TAPE_REQUEST_IN_IO)
763	763	return;
	764	+ /*
	765	+ * Request has already been stopped. We have to wait until
	766	+ * the request is removed from the queue in the interrupt
	767	+ * handling.
	768	+ */
	769	+ if (request->status == TAPE_REQUEST_DONE)
	770	+ return;
764	771
765	772	/*
766	773	* We wanted to cancel the request but the common I/O layer
...	...	@@ -1024,6 +1031,20 @@
1024	1031	}
1025	1032
1026	1033	/*
	1034	+ * Stop running ccw.
	1035	+ */
	1036	+int
	1037	+tape_cancel_io(struct tape_device device, struct tape_request request)
	1038	+{
	1039	+ int rc;
	1040	+
	1041	+ spin_lock_irq(get_ccwdev_lock(device->cdev));
	1042	+ rc = __tape_cancel_io(device, request);
	1043	+ spin_unlock_irq(get_ccwdev_lock(device->cdev));
	1044	+ return rc;
	1045	+}
	1046	+
	1047	+/*
1027	1048	* Tape interrupt routine, called from the ccw_device layer
1028	1049	*/
1029	1050	static void
1030	1051
...	...	@@ -1068,12 +1089,12 @@
1068	1089	* error might still apply. So we just schedule the request to be
1069	1090	* started later.
1070	1091	*/
1071		- if (irb->scsw.cc != 0 && (irb->scsw.fctl & SCSW_FCTL_START_FUNC)) {
1072		- PRINT_WARN("(%s): deferred cc=%i. restaring\n",
1073		- cdev->dev.bus_id,
1074		- irb->scsw.cc);
	1092	+ if (irb->scsw.cc != 0 && (irb->scsw.fctl & SCSW_FCTL_START_FUNC) &&
	1093	+ (request->status == TAPE_REQUEST_IN_IO)) {
	1094	+ DBF_EVENT(3,"(%08x): deferred cc=%i, fctl=%i. restarting\n",
	1095	+ device->cdev_id, irb->scsw.cc, irb->scsw.fctl);
1075	1096	request->status = TAPE_REQUEST_QUEUED;
1076		- schedule_work(&device->tape_dnr);
	1097	+ schedule_delayed_work(&device->tape_dnr, HZ);
1077	1098	return;
1078	1099	}
1079	1100
...	...	@@ -1287,5 +1308,6 @@
1287	1308	EXPORT_SYMBOL(tape_do_io);
1288	1309	EXPORT_SYMBOL(tape_do_io_async);
1289	1310	EXPORT_SYMBOL(tape_do_io_interruptible);
	1311	+EXPORT_SYMBOL(tape_cancel_io);
1290	1312	EXPORT_SYMBOL(tape_mtop);
...	...	@@ -37,20 +37,19 @@
37	37	{
38	38	struct tape_request * request;
39	39	struct tape_device * device;
	40	+ int rc;
40	41
41	42	request = (struct tape_request *) data;
42	43	if ((device = request->device) == NULL)
43	44	BUG();
44	45
45		- spin_lock_irq(get_ccwdev_lock(device->cdev));
46		- if (request->callback != NULL) {
47		- DBF_EVENT(3, "%08x: Assignment timeout. Device busy.\n",
	46	+ DBF_EVENT(3, "%08x: Assignment timeout. Device busy.\n",
48	47	device->cdev_id);
49		- PRINT_ERR("%s: Assignment timeout. Device busy.\n",
50		- device->cdev->dev.bus_id);
51		- ccw_device_clear(device->cdev, (long) request);
52		- }
53		- spin_unlock_irq(get_ccwdev_lock(device->cdev));
	48	+ rc = tape_cancel_io(device, request);
	49	+ if(rc)
	50	+ PRINT_ERR("(%s): Assign timeout: Cancel failed with rc = %i\n",
	51	+ device->cdev->dev.bus_id, rc);
	52	+
54	53	}
55	54
56	55	int