usbip: vudc: Don't enable IRQs prematurely

PATCH In review Dan Carpenter 2 participants 2 messages May 25, 2022

Subsystems (3)

USB Kernel janitors USB over IP driver

lore

Share & tools

lore

Dan Carpenter

author

2022-05-25 12:30 UTC #0 lore raw

PATCH 1/1 usbip: vudc: Don't enable IRQs prematurely

This code does:

	spin_unlock_irq(&udc->ud.lock);
	spin_unlock_irqrestore(&udc->lock, flags);

which does not make sense. In theory, the first unlock could enable IRQs and then the second _irqrestore could disable them again. There would be a brief momemt where IRQs were enabled improperly.

In real life, however, this function is always called with IRQs enabled and the bug does not affect runtime.

Signed-off-by Dan Carpenter <[email protected]>

---
An alternative fix would be to make changes below and also change the outside _irqsave/_irqrestore() to just _irq(). However, that's more invasive and riskier with no upside.

 drivers/usb/usbip/vudc_sysfs.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

drivers/usb/usbip/vudc_sysfs.c +7 -7

index d1cf6b51bf85..c95e6b2bfd32 100644
@@ -128,7 +128,7 @@ static ssize_t usbip_sockfd_store(struct device *dev,
 			goto unlock;
 		}
 
-		spin_lock_irq(&udc->ud.lock);
+		spin_lock(&udc->ud.lock);
 
 		if (udc->ud.status != SDEV_ST_AVAILABLE) {
 			ret = -EINVAL;

@@ -150,7 +150,7 @@ static ssize_t usbip_sockfd_store(struct device *dev,
 		}
 
 		/* unlock and create threads and get tasks */
-		spin_unlock_irq(&udc->ud.lock);
+		spin_unlock(&udc->ud.lock);
 		spin_unlock_irqrestore(&udc->lock, flags);
 
 		tcp_rx = kthread_create(&v_rx_loop, &udc->ud, "vudc_rx");

@@ -173,14 +173,14 @@ static ssize_t usbip_sockfd_store(struct device *dev,
 
 		/* lock and update udc->ud state */
 		spin_lock_irqsave(&udc->lock, flags);
-		spin_lock_irq(&udc->ud.lock);
+		spin_lock(&udc->ud.lock);
 
 		udc->ud.tcp_socket = socket;
 		udc->ud.tcp_rx = tcp_rx;
 		udc->ud.tcp_tx = tcp_tx;
 		udc->ud.status = SDEV_ST_USED;
 
-		spin_unlock_irq(&udc->ud.lock);
+		spin_unlock(&udc->ud.lock);
 
 		ktime_get_ts64(&udc->start_time);
 		v_start_timer(udc);

@@ -201,12 +201,12 @@ static ssize_t usbip_sockfd_store(struct device *dev,
 			goto unlock;
 		}
 
-		spin_lock_irq(&udc->ud.lock);
+		spin_lock(&udc->ud.lock);
 		if (udc->ud.status != SDEV_ST_USED) {
 			ret = -EINVAL;
 			goto unlock_ud;
 		}
-		spin_unlock_irq(&udc->ud.lock);
+		spin_unlock(&udc->ud.lock);
 
 		usbip_event_add(&udc->ud, VUDC_EVENT_DOWN);
 	}

@@ -219,7 +219,7 @@ static ssize_t usbip_sockfd_store(struct device *dev,
 sock_err:
 	sockfd_put(socket);
 unlock_ud:
-	spin_unlock_irq(&udc->ud.lock);
+	spin_unlock(&udc->ud.lock);
 unlock:
 	spin_unlock_irqrestore(&udc->lock, flags);
 	mutex_unlock(&udc->ud.sysfs_lock);

Signature · 1 line

2.35.1

Shuah Khan

maintainer Reviewed-by on v1 1/1

2022-05-25 15:13 UTC #1 lore raw

Quoted 9 lines · Dan Carpenter

This code does:
	spin_unlock_irq(&udc->ud.lock);
	spin_unlock_irqrestore(&udc->lock, flags);
which does not make sense. In theory, the first unlock could enable IRQs and then the second _irqrestore could disable them again. There would be a brief momemt where IRQs were enabled improperly.

Quoted 9 lines

In real life, however, this function is always called with IRQs enabled and the bug does not affect runtime.
Signed-off-by Dan Carpenter <[email protected]>
---
An alternative fix would be to make changes below and also change the outside _irqsave/_irqrestore() to just _irq(). However, that's more invasive and riskier with no upside.

Thank you for the patch.

Reviewed-by Shuah Khan <[email protected]>

Greg, please pick this up.

thanks,
-- Shuah