[Linux-kernel-commits] [6626] trunk/arch/blackfin/mach-common/smp.c: Fix bug[#5226] IPI handler should circle all messges in one loop.

[sonicz]

Title: [6626] trunk/arch/blackfin/mach-common/smp.c: Fix bug[#5226] IPI handler should circle all messges in one loop.

Revision

6626

Author

sonicz

Date

2009-06-10 03:42:41 -0500 (Wed, 10 Jun 2009)

Log Message

Fix bug[#5226] IPI handler should circle all messges in one loop.

A low priority ethernet interrupt is incorrectly triggered between 2
consequent high priority IPI messages, which creates a deadlock in
disable_irq() invoked in smc91x net poll controller by the second IPI
handler. The second IPI message is queued within the first IPI handler.
They should be handled in first IPI handler, but current IPI handler
doesn't circle all messges, which may increase in this loop, in the
queue properly before return.

Modified Paths

• trunk/arch/blackfin/mach-common/smp.c

Diff

Modified: trunk/arch/blackfin/mach-common/smp.c (6625 => 6626)

--- trunk/arch/blackfin/mach-common/smp.c	2009-06-10 08:36:55 UTC (rev 6625)
+++ trunk/arch/blackfin/mach-common/smp.c	2009-06-10 08:42:41 UTC (rev 6626)
@@ -144,7 +144,7 @@
 
 static irqreturn_t ipi_handler(int irq, void *dev_instance)
 {
-	struct ipi_message *msg, *mg;
+	struct ipi_message *msg;
 	struct ipi_message_queue *msg_queue;
 	unsigned int cpu = smp_processor_id();
 
@@ -154,7 +154,8 @@
 	msg_queue->count++;
 
 	spin_lock(&msg_queue->lock);
-	list_for_each_entry_safe(msg, mg, &msg_queue->head, list) {
+	while (!list_empty(&msg_queue->head)) {
+		msg = list_entry(msg_queue->head.next, typeof(*msg), list);
 		list_del(&msg->list);
 		switch (msg->type) {
 		case BFIN_IPI_RESCHEDULE:
@@ -221,7 +222,7 @@
 	for_each_cpu_mask(cpu, callmap) {
 		msg_queue = &per_cpu(ipi_msg_queue, cpu);
 		spin_lock_irqsave(&msg_queue->lock, flags);
-		list_add(&msg->list, &msg_queue->head);
+		list_add_tail(&msg->list, &msg_queue->head);
 		spin_unlock_irqrestore(&msg_queue->lock, flags);
 		platform_send_ipi_cpu(cpu);
 	}
@@ -261,7 +262,7 @@
 
 	msg_queue = &per_cpu(ipi_msg_queue, cpu);
 	spin_lock_irqsave(&msg_queue->lock, flags);
-	list_add(&msg->list, &msg_queue->head);
+	list_add_tail(&msg->list, &msg_queue->head);
 	spin_unlock_irqrestore(&msg_queue->lock, flags);
 	platform_send_ipi_cpu(cpu);
 
@@ -292,7 +293,7 @@
 
 	msg_queue = &per_cpu(ipi_msg_queue, cpu);
 	spin_lock_irqsave(&msg_queue->lock, flags);
-	list_add(&msg->list, &msg_queue->head);
+	list_add_tail(&msg->list, &msg_queue->head);
 	spin_unlock_irqrestore(&msg_queue->lock, flags);
 	platform_send_ipi_cpu(cpu);
 
@@ -320,7 +321,7 @@
 	for_each_cpu_mask(cpu, callmap) {
 		msg_queue = &per_cpu(ipi_msg_queue, cpu);
 		spin_lock_irqsave(&msg_queue->lock, flags);
-		list_add(&msg->list, &msg_queue->head);
+		list_add_tail(&msg->list, &msg_queue->head);
 		spin_unlock_irqrestore(&msg_queue->lock, flags);
 		platform_send_ipi_cpu(cpu);
 	}

_______________________________________________
Linux-kernel-commits mailing list
Linux-kernel-commits@blackfin.uclinux.org
https://blackfin.uclinux.org/mailman/listinfo/linux-kernel-commits