Skip to contentSkip to conversation

UIO: Fix concurrency issue

PATCH In review Vitalii Demianets 2 participants 2 messages Jun 20, 2013 – Jun 21, 2013 Userspace I/O (UIO)

lore
Share & tools
lore

Vitalii Demianets

author
#0 lore raw

PATCH 1/1 UIO: Fix concurrency issue

In a SMP case there was a race condition issue between uio_pdrv_genirq_irqcontrol() running on one CPU and irq handler on another CPU. Fix it by spin_locking shared resources access inside irq handler. Also:
- Change disable_irq to disable_irq_nosync to avoid deadlock, because
disable_irq waits for the completion of the irq handler;
- Change atomic bit-manipulation routines to their non-atomic counterparts
as we already are guarding the code by spinlock.

Signed-off-by Vitalii Demianets <[email protected]>
---
 drivers/uio/uio_pdrv_genirq.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)
drivers/uio/uio_pdrv_genirq.c +13 -5
index 9741549..bcd72f3 100644
@@ -37,6 +37,11 @@ struct uio_pdrv_genirq_platdata {
 	struct platform_device *pdev;
 };
 
+/* Bits in uio_pdrv_genirq_platdata.flags */
+enum {
+	UIO_IRQ_DISABLED = 0,
+};
+
 static int uio_pdrv_genirq_open(struct uio_info *info, struct inode *inode)
 {
 	struct uio_pdrv_genirq_platdata *priv = info->priv;

@@ -63,8 +68,10 @@ static irqreturn_t uio_pdrv_genirq_handler(int irq, struct uio_info *dev_info)
 	 * remember the state so we can allow user space to enable it later.
 	 */
 
-	if (!test_and_set_bit(0, &priv->flags))
+	spin_lock(&priv->lock);
+	if (!__test_and_set_bit(UIO_IRQ_DISABLED, &priv->flags))
 		disable_irq_nosync(irq);
+	spin_unlock(&priv->lock);
 
 	return IRQ_HANDLED;
 }

@@ -78,16 +85,17 @@ static int uio_pdrv_genirq_irqcontrol(struct uio_info *dev_info, s32 irq_on)
 	 * in the interrupt controller, but keep track of the
 	 * state to prevent per-irq depth damage.
 	 *
-	 * Serialize this operation to support multiple tasks.
+	 * Serialize this operation to support multiple tasks and concurrency
+	 * with irq handler on SMP systems.
 	 */
 
 	spin_lock_irqsave(&priv->lock, flags);
 	if (irq_on) {
-		if (test_and_clear_bit(0, &priv->flags))
+		if (__test_and_clear_bit(UIO_IRQ_DISABLED, &priv->flags))
 			enable_irq(dev_info->irq);
 	} else {
-		if (!test_and_set_bit(0, &priv->flags))
-			disable_irq(dev_info->irq);
+		if (!__test_and_set_bit(UIO_IRQ_DISABLED, &priv->flags))
+			disable_irq_nosync(dev_info->irq);
 	}
 	spin_unlock_irqrestore(&priv->lock, flags);
Signature · 4 lines
1.7.12.4
-- 
With Best Regards,
Vitalii Demianets
Reviewed-by on v1 1/1
#1 lore raw
Quoted 10 lines · Vitalii Demianets

In a SMP case there was a race condition issue between uio_pdrv_genirq_irqcontrol() running on one CPU and irq handler on another CPU. Fix it by spin_locking shared resources access inside irq handler. Also:
- Change disable_irq to disable_irq_nosync to avoid deadlock, because
disable_irq waits for the completion of the irq handler;
- Change atomic bit-manipulation routines to their non-atomic counterparts
as we already are guarding the code by spinlock.

Signed-off-by Vitalii Demianets <[email protected]>

Sounds good.

Reviewed-by Pavel Machek <[email protected]>
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html