[S390] zcrypt: fix request timeout handling

Under very high load zcrypt requests may timeout while waiting on the request queue. Modify zcrypt that timeouts are based on crypto adapter responses. A timeout occurs only if a crypto adapter does not respond within a given time frame to sumitted requests. Signed-off-by: Ralph Wuerthner <rwuerthn@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

[S390] zcrypt: fix request timeout handling
Under very high load zcrypt requests may timeout while waiting on the request queue. Modify zcrypt that timeouts are based on crypto adapter responses. A timeout occurs only if a crypto adapter does not respond within a given time frame to sumitted requests. Signed-off-by: Ralph Wuerthner <rwuerthn@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Ralph Wuerthner · Martin Schwidefsky
1 parent 987ad70a4d
Showing 6 changed files with 140 additions and 90 deletions Side-by-side Diff
drivers/s390/crypto/ap_bus.c
drivers/s390/crypto/ap_bus.h
drivers/s390/crypto/zcrypt_cex2a.c
drivers/s390/crypto/zcrypt_pcica.c
drivers/s390/crypto/zcrypt_pcicc.c
drivers/s390/crypto/zcrypt_pcixcc.c
@@ -43,6 +43,7 @@
 static void ap_poll_timeout(unsigned long);
 static int ap_poll_thread_start(void);
 static void ap_poll_thread_stop(void);
+static void ap_request_timeout(unsigned long);
  
 /**
  * Module description.
@@ -189,6 +190,7 @@
 	case AP_RESPONSE_NORMAL:
 		return 0;
 	case AP_RESPONSE_Q_FULL:
+	case AP_RESPONSE_RESET_IN_PROGRESS:
 		return -EBUSY;
 	default:	/* Device is gone. */
 		return -ENODEV;
@@ -252,6 +254,8 @@
 		if (status.queue_empty)
 			return -ENOENT;
 		return -EBUSY;
+	case AP_RESPONSE_RESET_IN_PROGRESS:
+		return -EBUSY;
 	default:
 		return -ENODEV;
 	}
  
@@ -326,11 +330,12 @@
 			i = AP_MAX_RESET;	/* return with -ENODEV */
 			break;
 		case AP_RESPONSE_RESET_IN_PROGRESS:
+			rc = -EBUSY;
 		case AP_RESPONSE_BUSY:
 		default:
 			break;
 		}
-		if (rc != -ENODEV)
+		if (rc != -ENODEV && rc != -EBUSY)
 			break;
 		if (i < AP_MAX_RESET - 1) {
 			udelay(5);
@@ -341,6 +346,40 @@
 }
  
 /**
+ * Arm request timeout if a AP device was idle and a new request is submitted.
+ */
+static void ap_increase_queue_count(struct ap_device *ap_dev)
+{
+	int timeout = ap_dev->drv->request_timeout;
+
+	ap_dev->queue_count++;
+	if (ap_dev->queue_count == 1) {
+		mod_timer(&ap_dev->timeout, jiffies + timeout);
+		ap_dev->reset = AP_RESET_ARMED;
+	}
+}
+
+/**
+ * AP device is still alive, re-schedule request timeout if there are still
+ * pending requests.
+ */
+static void ap_decrease_queue_count(struct ap_device *ap_dev)
+{
+	int timeout = ap_dev->drv->request_timeout;
+
+	ap_dev->queue_count--;
+	if (ap_dev->queue_count > 0)
+		mod_timer(&ap_dev->timeout, jiffies + timeout);
+	else
+		/**
+		 * The timeout timer should to be disabled now - since
+		 * del_timer_sync() is very expensive, we just tell via the
+		 * reset flag to ignore the pending timeout timer.
+		 */
+		ap_dev->reset = AP_RESET_IGNORE;
+}
+
+/**
  * AP device related attributes.
  */
 static ssize_t ap_hwtype_show(struct device *dev,
@@ -498,6 +537,7 @@
 	struct ap_driver *ap_drv = ap_dev->drv;
  
 	ap_flush_queue(ap_dev);
+	del_timer_sync(&ap_dev->timeout);
 	if (ap_drv->remove)
 		ap_drv->remove(ap_dev);
 	spin_lock_bh(&ap_device_lock);
  
  
  
@@ -759,17 +799,21 @@
 				      __ap_scan_bus);
 		rc = ap_query_queue(qid, &queue_depth, &device_type);
 		if (dev) {
+			if (rc == -EBUSY) {
+				set_current_state(TASK_UNINTERRUPTIBLE);
+				schedule_timeout(AP_RESET_TIMEOUT);
+				rc = ap_query_queue(qid, &queue_depth,
+						    &device_type);
+			}
 			ap_dev = to_ap_dev(dev);
 			spin_lock_bh(&ap_dev->lock);
 			if (rc || ap_dev->unregistered) {
 				spin_unlock_bh(&ap_dev->lock);
-				put_device(dev);
 				device_unregister(dev);
+				put_device(dev);
 				continue;
-			} else
-				spin_unlock_bh(&ap_dev->lock);
-		}
-		if (dev) {
+			}
+			spin_unlock_bh(&ap_dev->lock);
 			put_device(dev);
 			continue;
 		}
@@ -788,6 +832,8 @@
 		INIT_LIST_HEAD(&ap_dev->pendingq);
 		INIT_LIST_HEAD(&ap_dev->requestq);
 		INIT_LIST_HEAD(&ap_dev->list);
+		setup_timer(&ap_dev->timeout, ap_request_timeout,
+			    (unsigned long) ap_dev);
 		if (device_type == 0)
 			ap_probe_device_type(ap_dev);
 		else
@@ -853,7 +899,7 @@
 	switch (status.response_code) {
 	case AP_RESPONSE_NORMAL:
 		atomic_dec(&ap_poll_requests);
-		ap_dev->queue_count--;
+		ap_decrease_queue_count(ap_dev);
 		list_for_each_entry(ap_msg, &ap_dev->pendingq, list) {
 			if (ap_msg->psmid != ap_dev->reply->psmid)
 				continue;
@@ -904,7 +950,7 @@
 	switch (status.response_code) {
 	case AP_RESPONSE_NORMAL:
 		atomic_inc(&ap_poll_requests);
-		ap_dev->queue_count++;
+		ap_increase_queue_count(ap_dev);
 		list_move_tail(&ap_msg->list, &ap_dev->pendingq);
 		ap_dev->requestq_count--;
 		ap_dev->pendingq_count++;
@@ -914,6 +960,7 @@
 		*flags |= 2;
 		break;
 	case AP_RESPONSE_Q_FULL:
+	case AP_RESPONSE_RESET_IN_PROGRESS:
 		*flags |= 2;
 		break;
 	case AP_RESPONSE_MESSAGE_TOO_BIG:
  
@@ -960,10 +1007,11 @@
 			list_add_tail(&ap_msg->list, &ap_dev->pendingq);
 			atomic_inc(&ap_poll_requests);
 			ap_dev->pendingq_count++;
-			ap_dev->queue_count++;
+			ap_increase_queue_count(ap_dev);
 			ap_dev->total_request_count++;
 			break;
 		case AP_RESPONSE_Q_FULL:
+		case AP_RESPONSE_RESET_IN_PROGRESS:
 			list_add_tail(&ap_msg->list, &ap_dev->requestq);
 			ap_dev->requestq_count++;
 			ap_dev->total_request_count++;
@@ -1046,6 +1094,25 @@
 }
  
 /**
+ * Reset a not responding AP device and move all requests from the
+ * pending queue to the request queue.
+ */
+static void ap_reset(struct ap_device *ap_dev)
+{
+	int rc;
+
+	ap_dev->reset = AP_RESET_IGNORE;
+	atomic_sub(ap_dev->queue_count, &ap_poll_requests);
+	ap_dev->queue_count = 0;
+	list_splice_init(&ap_dev->pendingq, &ap_dev->requestq);
+	ap_dev->requestq_count += ap_dev->pendingq_count;
+	ap_dev->pendingq_count = 0;
+	rc = ap_init_queue(ap_dev->qid);
+	if (rc == -ENODEV)
+		ap_dev->unregistered = 1;
+}
+
+/**
  * Poll all AP devices on the bus in a round robin fashion. Continue
  * polling until bit 2^0 of the control flags is not set. If bit 2^1
  * of the control flags has been set arm the poll timer.
@@ -1056,6 +1123,8 @@
 	if (!ap_dev->unregistered) {
 		if (ap_poll_queue(ap_dev, flags))
 			ap_dev->unregistered = 1;
+		if (ap_dev->reset == AP_RESET_DO)
+			ap_reset(ap_dev);
 	}
 	spin_unlock(&ap_dev->lock);
 	return 0;
@@ -1145,6 +1214,17 @@
 		ap_poll_kthread = NULL;
 	}
 	mutex_unlock(&ap_poll_thread_mutex);
+}
+
+/**
+ * Handling of request timeouts
+ */
+static void ap_request_timeout(unsigned long data)
+{
+	struct ap_device *ap_dev = (struct ap_device *) data;
+
+	if (ap_dev->reset == AP_RESET_ARMED)
+		ap_dev->reset = AP_RESET_DO;
 }
  
 static void ap_reset_domain(void)
@@ -33,6 +33,7 @@
 #define AP_DEVICES 64		/* Number of AP devices. */
 #define AP_DOMAINS 16		/* Number of AP domains. */
 #define AP_MAX_RESET 90		/* Maximum number of resets. */
+#define AP_RESET_TIMEOUT (HZ/2)	/* Time in ticks for reset timeouts. */
 #define AP_CONFIG_TIME 30	/* Time in seconds between AP bus rescans. */
 #define AP_POLL_TIME 1		/* Time in ticks between receive polls. */
  
@@ -83,6 +84,13 @@
 #define AP_DEVICE_TYPE_CEX2A	6
 #define AP_DEVICE_TYPE_CEX2C	7
  
+/**
+ * AP reset flag states
+ */
+#define AP_RESET_IGNORE	0	/* request timeout will be ignored */
+#define AP_RESET_ARMED	1	/* request timeout timer is active */
+#define AP_RESET_DO	2	/* AP reset required */
+
 struct ap_device;
 struct ap_message;
  
@@ -95,6 +103,7 @@
 	/* receive is called from tasklet context */
 	void (*receive)(struct ap_device *, struct ap_message *,
 			struct ap_message *);
+	int request_timeout;		/* request timeout in jiffies */
 };
  
 #define to_ap_drv(x) container_of((x), struct ap_driver, driver)
@@ -112,6 +121,8 @@
 	int queue_depth;		/* AP queue depth.*/
 	int device_type;		/* AP device type. */
 	int unregistered;		/* marks AP device as unregistered */
+	struct timer_list timeout;	/* Timer for request timeouts. */
+	int reset;			/* Reset required after req. timeout. */
  
 	int queue_count;		/* # messages currently on AP queue. */
  
@@ -70,6 +70,7 @@
 	.remove = zcrypt_cex2a_remove,
 	.receive = zcrypt_cex2a_receive,
 	.ids = zcrypt_cex2a_ids,
+	.request_timeout = CEX2A_CLEANUP_TIME,
 };
  
 /**
  
  
@@ -306,18 +307,13 @@
 		goto out_free;
 	init_completion(&work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&work, CEX2A_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&work);
+	if (rc == 0)
 		rc = convert_response(zdev, &ap_msg, mex->outputdata,
 				      mex->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	kfree(ap_msg.message);
 	return rc;
  
  
@@ -348,18 +344,13 @@
 		goto out_free;
 	init_completion(&work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&work, CEX2A_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&work);
+	if (rc == 0)
 		rc = convert_response(zdev, &ap_msg, crt->outputdata,
 				      crt->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	kfree(ap_msg.message);
 	return rc;
@@ -70,6 +70,7 @@
 	.remove = zcrypt_pcica_remove,
 	.receive = zcrypt_pcica_receive,
 	.ids = zcrypt_pcica_ids,
+	.request_timeout = PCICA_CLEANUP_TIME,
 };
  
 /**
  
  
@@ -290,18 +291,13 @@
 		goto out_free;
 	init_completion(&work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&work, PCICA_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&work);
+	if (rc == 0)
 		rc = convert_response(zdev, &ap_msg, mex->outputdata,
 				      mex->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	kfree(ap_msg.message);
 	return rc;
  
  
@@ -332,18 +328,13 @@
 		goto out_free;
 	init_completion(&work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&work, PCICA_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&work);
+	if (rc == 0)
 		rc = convert_response(zdev, &ap_msg, crt->outputdata,
 				      crt->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	kfree(ap_msg.message);
 	return rc;
@@ -82,6 +82,7 @@
 	.remove = zcrypt_pcicc_remove,
 	.receive = zcrypt_pcicc_receive,
 	.ids = zcrypt_pcicc_ids,
+	.request_timeout = PCICC_CLEANUP_TIME,
 };
  
 /**
  
  
@@ -501,18 +502,13 @@
 		goto out_free;
 	init_completion(&work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&work, PCICC_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&work);
+	if (rc == 0)
 		rc = convert_response(zdev, &ap_msg, mex->outputdata,
 				      mex->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	free_page((unsigned long) ap_msg.message);
 	return rc;
  
  
@@ -544,18 +540,13 @@
 		goto out_free;
 	init_completion(&work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&work, PCICC_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&work);
+	if (rc == 0)
 		rc = convert_response(zdev, &ap_msg, crt->outputdata,
 				      crt->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	free_page((unsigned long) ap_msg.message);
 	return rc;
@@ -93,6 +93,7 @@
 	.remove = zcrypt_pcixcc_remove,
 	.receive = zcrypt_pcixcc_receive,
 	.ids = zcrypt_pcixcc_ids,
+	.request_timeout = PCIXCC_CLEANUP_TIME,
 };
  
 /**
  
  
@@ -641,18 +642,13 @@
 		goto out_free;
 	init_completion(&resp_type.work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&resp_type.work, PCIXCC_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&resp_type.work);
+	if (rc == 0)
 		rc = convert_response_ica(zdev, &ap_msg, mex->outputdata,
 					  mex->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	free_page((unsigned long) ap_msg.message);
 	return rc;
  
  
@@ -685,18 +681,13 @@
 		goto out_free;
 	init_completion(&resp_type.work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&resp_type.work, PCIXCC_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&resp_type.work);
+	if (rc == 0)
 		rc = convert_response_ica(zdev, &ap_msg, crt->outputdata,
 					  crt->outputdatalength);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	free_page((unsigned long) ap_msg.message);
 	return rc;
  
  
@@ -729,17 +720,12 @@
 		goto out_free;
 	init_completion(&resp_type.work);
 	ap_queue_message(zdev->ap_dev, &ap_msg);
-	rc = wait_for_completion_interruptible_timeout(
-				&resp_type.work, PCIXCC_CLEANUP_TIME);
-	if (rc > 0)
+	rc = wait_for_completion_interruptible(&resp_type.work);
+	if (rc == 0)
 		rc = convert_response_xcrb(zdev, &ap_msg, xcRB);
-	else {
-		/* Signal pending or message timed out. */
+	else
+		/* Signal pending. */
 		ap_cancel_message(zdev->ap_dev, &ap_msg);
-		if (rc == 0)
-			/* Message timed out. */
-			rc = -ETIME;
-	}
 out_free:
 	memset(ap_msg.message, 0x0, ap_msg.length);
 	kfree(ap_msg.message);
...	...	@@ -43,6 +43,7 @@
43	43	static void ap_poll_timeout(unsigned long);
44	44	static int ap_poll_thread_start(void);
45	45	static void ap_poll_thread_stop(void);
	46	+static void ap_request_timeout(unsigned long);
46	47
47	48	/**
48	49	* Module description.
...	...	@@ -189,6 +190,7 @@
189	190	case AP_RESPONSE_NORMAL:
190	191	return 0;
191	192	case AP_RESPONSE_Q_FULL:
	193	+ case AP_RESPONSE_RESET_IN_PROGRESS:
192	194	return -EBUSY;
193	195	default: /* Device is gone. */
194	196	return -ENODEV;
...	...	@@ -252,6 +254,8 @@
252	254	if (status.queue_empty)
253	255	return -ENOENT;
254	256	return -EBUSY;
	257	+ case AP_RESPONSE_RESET_IN_PROGRESS:
	258	+ return -EBUSY;
255	259	default:
256	260	return -ENODEV;
257	261	}
258	262
...	...	@@ -326,11 +330,12 @@
326	330	i = AP_MAX_RESET; /* return with -ENODEV */
327	331	break;
328	332	case AP_RESPONSE_RESET_IN_PROGRESS:
	333	+ rc = -EBUSY;
329	334	case AP_RESPONSE_BUSY:
330	335	default:
331	336	break;
332	337	}
333		- if (rc != -ENODEV)
	338	+ if (rc != -ENODEV && rc != -EBUSY)
334	339	break;
335	340	if (i < AP_MAX_RESET - 1) {
336	341	udelay(5);
...	...	@@ -341,6 +346,40 @@
341	346	}
342	347
343	348	/**
	349	+ * Arm request timeout if a AP device was idle and a new request is submitted.
	350	+ */
	351	+static void ap_increase_queue_count(struct ap_device *ap_dev)
	352	+{
	353	+ int timeout = ap_dev->drv->request_timeout;
	354	+
	355	+ ap_dev->queue_count++;
	356	+ if (ap_dev->queue_count == 1) {
	357	+ mod_timer(&ap_dev->timeout, jiffies + timeout);
	358	+ ap_dev->reset = AP_RESET_ARMED;
	359	+ }
	360	+}
	361	+
	362	+/**
	363	+ * AP device is still alive, re-schedule request timeout if there are still
	364	+ * pending requests.
	365	+ */
	366	+static void ap_decrease_queue_count(struct ap_device *ap_dev)
	367	+{
	368	+ int timeout = ap_dev->drv->request_timeout;
	369	+
	370	+ ap_dev->queue_count--;
	371	+ if (ap_dev->queue_count > 0)
	372	+ mod_timer(&ap_dev->timeout, jiffies + timeout);
	373	+ else
	374	+ /**
	375	+ * The timeout timer should to be disabled now - since
	376	+ * del_timer_sync() is very expensive, we just tell via the
	377	+ * reset flag to ignore the pending timeout timer.
	378	+ */
	379	+ ap_dev->reset = AP_RESET_IGNORE;
	380	+}
	381	+
	382	+/**
344	383	* AP device related attributes.
345	384	*/
346	385	static ssize_t ap_hwtype_show(struct device *dev,
...	...	@@ -498,6 +537,7 @@
498	537	struct ap_driver *ap_drv = ap_dev->drv;
499	538
500	539	ap_flush_queue(ap_dev);
	540	+ del_timer_sync(&ap_dev->timeout);
501	541	if (ap_drv->remove)
502	542	ap_drv->remove(ap_dev);
503	543	spin_lock_bh(&ap_device_lock);
504	544
505	545
506	546
...	...	@@ -759,17 +799,21 @@
759	799	__ap_scan_bus);
760	800	rc = ap_query_queue(qid, &queue_depth, &device_type);
761	801	if (dev) {
	802	+ if (rc == -EBUSY) {
	803	+ set_current_state(TASK_UNINTERRUPTIBLE);
	804	+ schedule_timeout(AP_RESET_TIMEOUT);
	805	+ rc = ap_query_queue(qid, &queue_depth,
	806	+ &device_type);
	807	+ }
762	808	ap_dev = to_ap_dev(dev);
763	809	spin_lock_bh(&ap_dev->lock);
764	810	if (rc \|\| ap_dev->unregistered) {
765	811	spin_unlock_bh(&ap_dev->lock);
766		- put_device(dev);
767	812	device_unregister(dev);
	813	+ put_device(dev);
768	814	continue;
769		- } else
770		- spin_unlock_bh(&ap_dev->lock);
771		- }
772		- if (dev) {
	815	+ }
	816	+ spin_unlock_bh(&ap_dev->lock);
773	817	put_device(dev);
774	818	continue;
775	819	}
...	...	@@ -788,6 +832,8 @@
788	832	INIT_LIST_HEAD(&ap_dev->pendingq);
789	833	INIT_LIST_HEAD(&ap_dev->requestq);
790	834	INIT_LIST_HEAD(&ap_dev->list);
	835	+ setup_timer(&ap_dev->timeout, ap_request_timeout,
	836	+ (unsigned long) ap_dev);
791	837	if (device_type == 0)
792	838	ap_probe_device_type(ap_dev);
793	839	else
...	...	@@ -853,7 +899,7 @@
853	899	switch (status.response_code) {
854	900	case AP_RESPONSE_NORMAL:
855	901	atomic_dec(&ap_poll_requests);
856		- ap_dev->queue_count--;
	902	+ ap_decrease_queue_count(ap_dev);
857	903	list_for_each_entry(ap_msg, &ap_dev->pendingq, list) {
858	904	if (ap_msg->psmid != ap_dev->reply->psmid)
859	905	continue;
...	...	@@ -904,7 +950,7 @@
904	950	switch (status.response_code) {
905	951	case AP_RESPONSE_NORMAL:
906	952	atomic_inc(&ap_poll_requests);
907		- ap_dev->queue_count++;
	953	+ ap_increase_queue_count(ap_dev);
908	954	list_move_tail(&ap_msg->list, &ap_dev->pendingq);
909	955	ap_dev->requestq_count--;
910	956	ap_dev->pendingq_count++;
...	...	@@ -914,6 +960,7 @@
914	960	*flags \|= 2;
915	961	break;
916	962	case AP_RESPONSE_Q_FULL:
	963	+ case AP_RESPONSE_RESET_IN_PROGRESS:
917	964	*flags \|= 2;
918	965	break;
919	966	case AP_RESPONSE_MESSAGE_TOO_BIG:
920	967
...	...	@@ -960,10 +1007,11 @@
960	1007	list_add_tail(&ap_msg->list, &ap_dev->pendingq);
961	1008	atomic_inc(&ap_poll_requests);
962	1009	ap_dev->pendingq_count++;
963		- ap_dev->queue_count++;
	1010	+ ap_increase_queue_count(ap_dev);
964	1011	ap_dev->total_request_count++;
965	1012	break;
966	1013	case AP_RESPONSE_Q_FULL:
	1014	+ case AP_RESPONSE_RESET_IN_PROGRESS:
967	1015	list_add_tail(&ap_msg->list, &ap_dev->requestq);
968	1016	ap_dev->requestq_count++;
969	1017	ap_dev->total_request_count++;
...	...	@@ -1046,6 +1094,25 @@
1046	1094	}
1047	1095
1048	1096	/**
	1097	+ * Reset a not responding AP device and move all requests from the
	1098	+ * pending queue to the request queue.
	1099	+ */
	1100	+static void ap_reset(struct ap_device *ap_dev)
	1101	+{
	1102	+ int rc;
	1103	+
	1104	+ ap_dev->reset = AP_RESET_IGNORE;
	1105	+ atomic_sub(ap_dev->queue_count, &ap_poll_requests);
	1106	+ ap_dev->queue_count = 0;
	1107	+ list_splice_init(&ap_dev->pendingq, &ap_dev->requestq);
	1108	+ ap_dev->requestq_count += ap_dev->pendingq_count;
	1109	+ ap_dev->pendingq_count = 0;
	1110	+ rc = ap_init_queue(ap_dev->qid);
	1111	+ if (rc == -ENODEV)
	1112	+ ap_dev->unregistered = 1;
	1113	+}
	1114	+
	1115	+/**
1049	1116	* Poll all AP devices on the bus in a round robin fashion. Continue
1050	1117	* polling until bit 2^0 of the control flags is not set. If bit 2^1
1051	1118	* of the control flags has been set arm the poll timer.
...	...	@@ -1056,6 +1123,8 @@
1056	1123	if (!ap_dev->unregistered) {
1057	1124	if (ap_poll_queue(ap_dev, flags))
1058	1125	ap_dev->unregistered = 1;
	1126	+ if (ap_dev->reset == AP_RESET_DO)
	1127	+ ap_reset(ap_dev);
1059	1128	}
1060	1129	spin_unlock(&ap_dev->lock);
1061	1130	return 0;
...	...	@@ -1145,6 +1214,17 @@
1145	1214	ap_poll_kthread = NULL;
1146	1215	}
1147	1216	mutex_unlock(&ap_poll_thread_mutex);
	1217	+}
	1218	+
	1219	+/**
	1220	+ * Handling of request timeouts
	1221	+ */
	1222	+static void ap_request_timeout(unsigned long data)
	1223	+{
	1224	+ struct ap_device ap_dev = (struct ap_device ) data;
	1225	+
	1226	+ if (ap_dev->reset == AP_RESET_ARMED)
	1227	+ ap_dev->reset = AP_RESET_DO;
1148	1228	}
1149	1229
1150	1230	static void ap_reset_domain(void)
...	...	@@ -33,6 +33,7 @@
33	33	#define AP_DEVICES 64 /* Number of AP devices. */
34	34	#define AP_DOMAINS 16 /* Number of AP domains. */
35	35	#define AP_MAX_RESET 90 /* Maximum number of resets. */
	36	+#define AP_RESET_TIMEOUT (HZ/2) /* Time in ticks for reset timeouts. */
36	37	#define AP_CONFIG_TIME 30 /* Time in seconds between AP bus rescans. */
37	38	#define AP_POLL_TIME 1 /* Time in ticks between receive polls. */
38	39
...	...	@@ -83,6 +84,13 @@
83	84	#define AP_DEVICE_TYPE_CEX2A 6
84	85	#define AP_DEVICE_TYPE_CEX2C 7
85	86
	87	+/**
	88	+ * AP reset flag states
	89	+ */
	90	+#define AP_RESET_IGNORE 0 /* request timeout will be ignored */
	91	+#define AP_RESET_ARMED 1 /* request timeout timer is active */
	92	+#define AP_RESET_DO 2 /* AP reset required */
	93	+
86	94	struct ap_device;
87	95	struct ap_message;
88	96
...	...	@@ -95,6 +103,7 @@
95	103	/* receive is called from tasklet context */
96	104	void (receive)(struct ap_device , struct ap_message *,
97	105	struct ap_message *);
	106	+ int request_timeout; /* request timeout in jiffies */
98	107	};
99	108
100	109	#define to_ap_drv(x) container_of((x), struct ap_driver, driver)
...	...	@@ -112,6 +121,8 @@
112	121	int queue_depth; /* AP queue depth.*/
113	122	int device_type; /* AP device type. */
114	123	int unregistered; /* marks AP device as unregistered */
	124	+ struct timer_list timeout; /* Timer for request timeouts. */
	125	+ int reset; /* Reset required after req. timeout. */
115	126
116	127	int queue_count; /* # messages currently on AP queue. */
117	128
...	...	@@ -70,6 +70,7 @@
70	70	.remove = zcrypt_cex2a_remove,
71	71	.receive = zcrypt_cex2a_receive,
72	72	.ids = zcrypt_cex2a_ids,
	73	+ .request_timeout = CEX2A_CLEANUP_TIME,
73	74	};
74	75
75	76	/**
76	77
77	78
...	...	@@ -306,18 +307,13 @@
306	307	goto out_free;
307	308	init_completion(&work);
308	309	ap_queue_message(zdev->ap_dev, &ap_msg);
309		- rc = wait_for_completion_interruptible_timeout(
310		- &work, CEX2A_CLEANUP_TIME);
311		- if (rc > 0)
	310	+ rc = wait_for_completion_interruptible(&work);
	311	+ if (rc == 0)
312	312	rc = convert_response(zdev, &ap_msg, mex->outputdata,
313	313	mex->outputdatalength);
314		- else {
315		- /* Signal pending or message timed out. */
	314	+ else
	315	+ /* Signal pending. */
316	316	ap_cancel_message(zdev->ap_dev, &ap_msg);
317		- if (rc == 0)
318		- /* Message timed out. */
319		- rc = -ETIME;
320		- }
321	317	out_free:
322	318	kfree(ap_msg.message);
323	319	return rc;
324	320
325	321
...	...	@@ -348,18 +344,13 @@
348	344	goto out_free;
349	345	init_completion(&work);
350	346	ap_queue_message(zdev->ap_dev, &ap_msg);
351		- rc = wait_for_completion_interruptible_timeout(
352		- &work, CEX2A_CLEANUP_TIME);
353		- if (rc > 0)
	347	+ rc = wait_for_completion_interruptible(&work);
	348	+ if (rc == 0)
354	349	rc = convert_response(zdev, &ap_msg, crt->outputdata,
355	350	crt->outputdatalength);
356		- else {
357		- /* Signal pending or message timed out. */
	351	+ else
	352	+ /* Signal pending. */
358	353	ap_cancel_message(zdev->ap_dev, &ap_msg);
359		- if (rc == 0)
360		- /* Message timed out. */
361		- rc = -ETIME;
362		- }
363	354	out_free:
364	355	kfree(ap_msg.message);
365	356	return rc;
...	...	@@ -70,6 +70,7 @@
70	70	.remove = zcrypt_pcica_remove,
71	71	.receive = zcrypt_pcica_receive,
72	72	.ids = zcrypt_pcica_ids,
	73	+ .request_timeout = PCICA_CLEANUP_TIME,
73	74	};
74	75
75	76	/**
76	77
77	78
...	...	@@ -290,18 +291,13 @@
290	291	goto out_free;
291	292	init_completion(&work);
292	293	ap_queue_message(zdev->ap_dev, &ap_msg);
293		- rc = wait_for_completion_interruptible_timeout(
294		- &work, PCICA_CLEANUP_TIME);
295		- if (rc > 0)
	294	+ rc = wait_for_completion_interruptible(&work);
	295	+ if (rc == 0)
296	296	rc = convert_response(zdev, &ap_msg, mex->outputdata,
297	297	mex->outputdatalength);
298		- else {
299		- /* Signal pending or message timed out. */
	298	+ else
	299	+ /* Signal pending. */
300	300	ap_cancel_message(zdev->ap_dev, &ap_msg);
301		- if (rc == 0)
302		- /* Message timed out. */
303		- rc = -ETIME;
304		- }
305	301	out_free:
306	302	kfree(ap_msg.message);
307	303	return rc;
308	304
309	305
...	...	@@ -332,18 +328,13 @@
332	328	goto out_free;
333	329	init_completion(&work);
334	330	ap_queue_message(zdev->ap_dev, &ap_msg);
335		- rc = wait_for_completion_interruptible_timeout(
336		- &work, PCICA_CLEANUP_TIME);
337		- if (rc > 0)
	331	+ rc = wait_for_completion_interruptible(&work);
	332	+ if (rc == 0)
338	333	rc = convert_response(zdev, &ap_msg, crt->outputdata,
339	334	crt->outputdatalength);
340		- else {
341		- /* Signal pending or message timed out. */
	335	+ else
	336	+ /* Signal pending. */
342	337	ap_cancel_message(zdev->ap_dev, &ap_msg);
343		- if (rc == 0)
344		- /* Message timed out. */
345		- rc = -ETIME;
346		- }
347	338	out_free:
348	339	kfree(ap_msg.message);
349	340	return rc;
...	...	@@ -82,6 +82,7 @@
82	82	.remove = zcrypt_pcicc_remove,
83	83	.receive = zcrypt_pcicc_receive,
84	84	.ids = zcrypt_pcicc_ids,
	85	+ .request_timeout = PCICC_CLEANUP_TIME,
85	86	};
86	87
87	88	/**
88	89
89	90
...	...	@@ -501,18 +502,13 @@
501	502	goto out_free;
502	503	init_completion(&work);
503	504	ap_queue_message(zdev->ap_dev, &ap_msg);
504		- rc = wait_for_completion_interruptible_timeout(
505		- &work, PCICC_CLEANUP_TIME);
506		- if (rc > 0)
	505	+ rc = wait_for_completion_interruptible(&work);
	506	+ if (rc == 0)
507	507	rc = convert_response(zdev, &ap_msg, mex->outputdata,
508	508	mex->outputdatalength);
509		- else {
510		- /* Signal pending or message timed out. */
	509	+ else
	510	+ /* Signal pending. */
511	511	ap_cancel_message(zdev->ap_dev, &ap_msg);
512		- if (rc == 0)
513		- /* Message timed out. */
514		- rc = -ETIME;
515		- }
516	512	out_free:
517	513	free_page((unsigned long) ap_msg.message);
518	514	return rc;
519	515
520	516
...	...	@@ -544,18 +540,13 @@
544	540	goto out_free;
545	541	init_completion(&work);
546	542	ap_queue_message(zdev->ap_dev, &ap_msg);
547		- rc = wait_for_completion_interruptible_timeout(
548		- &work, PCICC_CLEANUP_TIME);
549		- if (rc > 0)
	543	+ rc = wait_for_completion_interruptible(&work);
	544	+ if (rc == 0)
550	545	rc = convert_response(zdev, &ap_msg, crt->outputdata,
551	546	crt->outputdatalength);
552		- else {
553		- /* Signal pending or message timed out. */
	547	+ else
	548	+ /* Signal pending. */
554	549	ap_cancel_message(zdev->ap_dev, &ap_msg);
555		- if (rc == 0)
556		- /* Message timed out. */
557		- rc = -ETIME;
558		- }
559	550	out_free:
560	551	free_page((unsigned long) ap_msg.message);
561	552	return rc;
...	...	@@ -93,6 +93,7 @@
93	93	.remove = zcrypt_pcixcc_remove,
94	94	.receive = zcrypt_pcixcc_receive,
95	95	.ids = zcrypt_pcixcc_ids,
	96	+ .request_timeout = PCIXCC_CLEANUP_TIME,
96	97	};
97	98
98	99	/**
99	100
100	101
...	...	@@ -641,18 +642,13 @@
641	642	goto out_free;
642	643	init_completion(&resp_type.work);
643	644	ap_queue_message(zdev->ap_dev, &ap_msg);
644		- rc = wait_for_completion_interruptible_timeout(
645		- &resp_type.work, PCIXCC_CLEANUP_TIME);
646		- if (rc > 0)
	645	+ rc = wait_for_completion_interruptible(&resp_type.work);
	646	+ if (rc == 0)
647	647	rc = convert_response_ica(zdev, &ap_msg, mex->outputdata,
648	648	mex->outputdatalength);
649		- else {
650		- /* Signal pending or message timed out. */
	649	+ else
	650	+ /* Signal pending. */
651	651	ap_cancel_message(zdev->ap_dev, &ap_msg);
652		- if (rc == 0)
653		- /* Message timed out. */
654		- rc = -ETIME;
655		- }
656	652	out_free:
657	653	free_page((unsigned long) ap_msg.message);
658	654	return rc;
659	655
660	656
...	...	@@ -685,18 +681,13 @@
685	681	goto out_free;
686	682	init_completion(&resp_type.work);
687	683	ap_queue_message(zdev->ap_dev, &ap_msg);
688		- rc = wait_for_completion_interruptible_timeout(
689		- &resp_type.work, PCIXCC_CLEANUP_TIME);
690		- if (rc > 0)
	684	+ rc = wait_for_completion_interruptible(&resp_type.work);
	685	+ if (rc == 0)
691	686	rc = convert_response_ica(zdev, &ap_msg, crt->outputdata,
692	687	crt->outputdatalength);
693		- else {
694		- /* Signal pending or message timed out. */
	688	+ else
	689	+ /* Signal pending. */
695	690	ap_cancel_message(zdev->ap_dev, &ap_msg);
696		- if (rc == 0)
697		- /* Message timed out. */
698		- rc = -ETIME;
699		- }
700	691	out_free:
701	692	free_page((unsigned long) ap_msg.message);
702	693	return rc;
703	694
704	695
...	...	@@ -729,17 +720,12 @@
729	720	goto out_free;
730	721	init_completion(&resp_type.work);
731	722	ap_queue_message(zdev->ap_dev, &ap_msg);
732		- rc = wait_for_completion_interruptible_timeout(
733		- &resp_type.work, PCIXCC_CLEANUP_TIME);
734		- if (rc > 0)
	723	+ rc = wait_for_completion_interruptible(&resp_type.work);
	724	+ if (rc == 0)
735	725	rc = convert_response_xcrb(zdev, &ap_msg, xcRB);
736		- else {
737		- /* Signal pending or message timed out. */
	726	+ else
	727	+ /* Signal pending. */
738	728	ap_cancel_message(zdev->ap_dev, &ap_msg);
739		- if (rc == 0)
740		- /* Message timed out. */
741		- rc = -ETIME;
742		- }
743	729	out_free:
744	730	memset(ap_msg.message, 0x0, ap_msg.length);
745	731	kfree(ap_msg.message);