Skip to contentSkip to conversation

usb: dwc3: fix race of usb_gadget_driver operation

PATCH In review Linyu Yuan 2 participants 8 messages Jun 25, 2021 – Jun 28, 2021

Subsystems (2)
lore
Share & tools
lore

Linyu Yuan

author
#0 lore raw

PATCH 1/1 usb: dwc3: fix race of usb_gadget_driver operation

there is following race condition,

CPU1 CPU2
dwc3_runtime_suspend() dwc3_gadget_stop()
spin_lock(&dwc->lock)
dwc3_gadget_suspend()
dwc3_disconnect_gadget()
dwc->gadget_driver != NULL
spin_unlock(&dwc->lock)

                              spin_lock(&dwc->lock)
                              dwc->gadget_driver = NULL
                              spin_unlock(&dwc->lock)

dwc->gadget_driver->disconnect(dwc->gadget);

system will crash when access NULL gadget_driver.

7dc0c55e9f30 ('USB: UDC core: Add udc_async_callbacks gadget op') suggest a common way to avoid such kind of race.

this change implment the callback in dwc3 and change related functions which have callback to UDC core.

---
 drivers/usb/dwc3/core.h   |  1 +
 drivers/usb/dwc3/ep0.c    |  5 +++--
 drivers/usb/dwc3/gadget.c | 39 +++++++++++++++++++++++----------------
 3 files changed, 27 insertions(+), 18 deletions(-)
drivers/usb/dwc3/core.h +1 -0
index dccdf13b5f9e..5991766239ba 100644
@@ -1279,6 +1279,7 @@ struct dwc3 {
 	unsigned		dis_metastability_quirk:1;
 
 	unsigned		dis_split_quirk:1;
+	unsigned		async_callbacks:1;
 
 	u16			imod_interval;
 };
drivers/usb/dwc3/ep0.c +3 -2
index 3cd294264372..26419077c7c9 100644
@@ -597,10 +597,11 @@ static int dwc3_ep0_set_address(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl)
 
 static int dwc3_ep0_delegate_req(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl)
 {
-	int ret;
+	int ret = 0;
 
 	spin_unlock(&dwc->lock);
-	ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
+	if (dwc->async_callbacks)
+		ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
 	spin_lock(&dwc->lock);
 	return ret;
 }
drivers/usb/dwc3/gadget.c +23 -16
index af6d7f157989..d868f30007cc 100644
@@ -2585,6 +2585,16 @@ static int dwc3_gadget_vbus_draw(struct usb_gadget *g, unsigned int mA)
 	return ret;
 }
 
+static void dwc3_gadget_async_callbacks(struct usb_gadget *g, bool enable)
+{
+	struct dwc3		*dwc = gadget_to_dwc(g);
+	unsigned long		flags;
+
+	spin_lock_irqsave(&dwc->lock, flags);
+	dwc->async_callbacks = enable;
+	spin_unlock_irqrestore(&dwc->lock, flags);
+}
+
 static const struct usb_gadget_ops dwc3_gadget_ops = {
 	.get_frame		= dwc3_gadget_get_frame,
 	.wakeup			= dwc3_gadget_wakeup,
@@ -2596,6 +2606,7 @@ static const struct usb_gadget_ops dwc3_gadget_ops = {
 	.udc_set_ssp_rate	= dwc3_gadget_set_ssp_rate,
 	.get_config_params	= dwc3_gadget_config_params,
 	.vbus_draw		= dwc3_gadget_vbus_draw,
+	.udc_async_callbacks	= dwc3_gadget_async_callbacks,
 };
 
 /* -------------------------------------------------------------------------- */
@@ -3231,29 +3242,26 @@ static void dwc3_endpoint_interrupt(struct dwc3 *dwc,
 
 static void dwc3_disconnect_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->disconnect) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->disconnect)
 		dwc->gadget_driver->disconnect(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_suspend_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->suspend) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->suspend)
 		dwc->gadget_driver->suspend(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_resume_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->resume) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->resume)
 		dwc->gadget_driver->resume(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_reset_gadget(struct dwc3 *dwc)
@@ -3585,11 +3593,10 @@ static void dwc3_gadget_wakeup_interrupt(struct dwc3 *dwc)
 	 * implemented.
 	 */
 
-	if (dwc->gadget_driver && dwc->gadget_driver->resume) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->resume)
 		dwc->gadget_driver->resume(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_gadget_linksts_change_interrupt(struct dwc3 *dwc,
Signature · 1 line
2.25.1
#1 lore raw
Quoted 47 lines · Linyu Yuan

there is following race condition,

CPU1 CPU2
dwc3_runtime_suspend() dwc3_gadget_stop()
spin_lock(&dwc->lock)
dwc3_gadget_suspend()
dwc3_disconnect_gadget()
dwc->gadget_driver != NULL
spin_unlock(&dwc->lock)

                              spin_lock(&dwc->lock)
                              dwc->gadget_driver = NULL
                              spin_unlock(&dwc->lock)

dwc->gadget_driver->disconnect(dwc->gadget);

system will crash when access NULL gadget_driver.

7dc0c55e9f30 ('USB: UDC core: Add udc_async_callbacks gadget op') suggest a common way to avoid such kind of race.

this change implment the callback in dwc3 and change related functions which have callback to UDC core.

---
 drivers/usb/dwc3/core.h   |  1 +
 drivers/usb/dwc3/ep0.c    |  5 +++--
 drivers/usb/dwc3/gadget.c | 39 +++++++++++++++++++++++----------------
 3 files changed, 27 insertions(+), 18 deletions(-)
drivers/usb/dwc3/core.h +1 -0
index dccdf13b5f9e..5991766239ba 100644
@@ -1279,6 +1279,7 @@ struct dwc3 {
 	unsigned		dis_metastability_quirk:1;
 
 	unsigned		dis_split_quirk:1;
+	unsigned		async_callbacks:1;
 
 	u16			imod_interval;
 };
drivers/usb/dwc3/ep0.c +0 -0
index 3cd294264372..26419077c7c9 100644
@@ -597,10 +597,11 @@ static int dwc3_ep0_set_address(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl)
 

drivers/usb/dwc3/ep0.c static int dwc3_ep0_set_address(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl) · line 605

Quoted 10 lines
drivers/usb/dwc3/ep0.c +3 -2
 static int dwc3_ep0_delegate_req(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl)
 {
-	int ret;
+	int ret = 0;
 
 	spin_unlock(&dwc->lock);
-	ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
+	if (dwc->async_callbacks)
+		ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
 	spin_lock(&dwc->lock);

Here and in the other places, you should test dwc->async_callbacks _before_ dropping the spinlock. Otherwise there is a race (the flag could be written at about the same time it is checked).

Alan Stern

Quoted 88 lines
drivers/usb/dwc3/ep0.c +0 -0
 	return ret;
 }
drivers/usb/dwc3/gadget.c +23 -16
index af6d7f157989..d868f30007cc 100644
@@ -2585,6 +2585,16 @@ static int dwc3_gadget_vbus_draw(struct usb_gadget *g, unsigned int mA)
 	return ret;
 }
 
+static void dwc3_gadget_async_callbacks(struct usb_gadget *g, bool enable)
+{
+	struct dwc3		*dwc = gadget_to_dwc(g);
+	unsigned long		flags;
+
+	spin_lock_irqsave(&dwc->lock, flags);
+	dwc->async_callbacks = enable;
+	spin_unlock_irqrestore(&dwc->lock, flags);
+}
+
 static const struct usb_gadget_ops dwc3_gadget_ops = {
 	.get_frame		= dwc3_gadget_get_frame,
 	.wakeup			= dwc3_gadget_wakeup,
@@ -2596,6 +2606,7 @@ static const struct usb_gadget_ops dwc3_gadget_ops = {
 	.udc_set_ssp_rate	= dwc3_gadget_set_ssp_rate,
 	.get_config_params	= dwc3_gadget_config_params,
 	.vbus_draw		= dwc3_gadget_vbus_draw,
+	.udc_async_callbacks	= dwc3_gadget_async_callbacks,
 };
 
 /* -------------------------------------------------------------------------- */
@@ -3231,29 +3242,26 @@ static void dwc3_endpoint_interrupt(struct dwc3 *dwc,
 
 static void dwc3_disconnect_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->disconnect) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->disconnect)
 		dwc->gadget_driver->disconnect(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_suspend_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->suspend) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->suspend)
 		dwc->gadget_driver->suspend(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_resume_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->resume) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->resume)
 		dwc->gadget_driver->resume(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_reset_gadget(struct dwc3 *dwc)
@@ -3585,11 +3593,10 @@ static void dwc3_gadget_wakeup_interrupt(struct dwc3 *dwc)
 	 * implemented.
 	 */
 
-	if (dwc->gadget_driver && dwc->gadget_driver->resume) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->resume)
 		dwc->gadget_driver->resume(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_gadget_linksts_change_interrupt(struct dwc3 *dwc,
Signature · 1 line
2.25.1
#2 lore raw
Quoted 56 lines · Alan Stern
Quoted 55 lines · Linyu Yuan

there is following race condition,

CPU1 CPU2
dwc3_runtime_suspend() dwc3_gadget_stop()
spin_lock(&dwc->lock)
dwc3_gadget_suspend()
dwc3_disconnect_gadget()
dwc->gadget_driver != NULL
spin_unlock(&dwc->lock)

                              spin_lock(&dwc->lock)
                              dwc->gadget_driver = NULL
                              spin_unlock(&dwc->lock)

dwc->gadget_driver->disconnect(dwc->gadget);

system will crash when access NULL gadget_driver.

7dc0c55e9f30 ('USB: UDC core: Add udc_async_callbacks gadget op') suggest a common way to avoid such kind of race.

this change implment the callback in dwc3 and change related functions which have callback to UDC core.

---
 drivers/usb/dwc3/core.h   |  1 +
 drivers/usb/dwc3/ep0.c    |  5 +++--
 drivers/usb/dwc3/gadget.c | 39 
+++++++++++++++++++++++----------------
 3 files changed, 27 insertions(+), 18 deletions(-)
drivers/usb/dwc3/core.h +1 -0
index dccdf13b5f9e..5991766239ba 100644
@@ -1279,6 +1279,7 @@ struct dwc3 {
 	unsigned		dis_metastability_quirk:1;
 
 	unsigned		dis_split_quirk:1;
+	unsigned		async_callbacks:1;
 
 	u16			imod_interval;
 };
drivers/usb/dwc3/ep0.c +1 -1
index 3cd294264372..26419077c7c9 100644
@@ -597,10 +597,11 @@ static int dwc3_ep0_set_address(struct dwc3 
 *dwc, struct usb_ctrlrequest *ctrl)
 
 static int dwc3_ep0_delegate_req(struct dwc3 *dwc, struct 
 usb_ctrlrequest *ctrl)
 {
-	int ret;
+	int ret = 0;
 
Quoted 9 lines

drivers/usb/dwc3/ep0.c static int dwc3_ep0_set_address(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl) · line 605

Quoted 5 lines
drivers/usb/dwc3/ep0.c +2 -1
 	spin_unlock(&dwc->lock);
-	ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
+	if (dwc->async_callbacks)
+		ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
 	spin_lock(&dwc->lock);

Here and in the other places, you should test dwc->async_callbacks _before_ dropping the spinlock. Otherwise there is a race (the flag could be written at about the same time it is checked).

thanks for your comments,

if you think there is race here, how to make sure gadget_driver pointer is safe, this is closest place where we can confirm it is non-NULL by checking async_callbacks ?

Quoted 98 lines

Alan Stern

Quoted 95 lines
drivers/usb/dwc3/ep0.c +0 -0
 	return ret;
 }
drivers/usb/dwc3/gadget.c +10 -0
index af6d7f157989..d868f30007cc 100644
@@ -2585,6 +2585,16 @@ static int dwc3_gadget_vbus_draw(struct 
 usb_gadget *g, unsigned int mA)
 	return ret;
 }
 
+static void dwc3_gadget_async_callbacks(struct usb_gadget *g, bool 
 enable)
+{
+	struct dwc3		*dwc = gadget_to_dwc(g);
+	unsigned long		flags;
+
+	spin_lock_irqsave(&dwc->lock, flags);
+	dwc->async_callbacks = enable;
+	spin_unlock_irqrestore(&dwc->lock, flags);
+}
+
 static const struct usb_gadget_ops dwc3_gadget_ops = {
drivers/usb/dwc3/gadget.c +1 -0
 	.get_frame		= dwc3_gadget_get_frame,
 	.wakeup			= dwc3_gadget_wakeup,
@@ -2596,6 +2606,7 @@ static const struct usb_gadget_ops 
 dwc3_gadget_ops = {
 	.udc_set_ssp_rate	= dwc3_gadget_set_ssp_rate,
 	.get_config_params	= dwc3_gadget_config_params,
 	.vbus_draw		= dwc3_gadget_vbus_draw,
+	.udc_async_callbacks	= dwc3_gadget_async_callbacks,
 };
 

/*
--------------------------------------------------------------------------
*/

drivers/usb/dwc3/gadget.c +9 -12
@@ -3231,29 +3242,26 @@ static void dwc3_endpoint_interrupt(struct 
 dwc3 *dwc,
 
 static void dwc3_disconnect_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->disconnect) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->disconnect)
 		dwc->gadget_driver->disconnect(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_suspend_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->suspend) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->suspend)
 		dwc->gadget_driver->suspend(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 
 static void dwc3_resume_gadget(struct dwc3 *dwc)
 {
-	if (dwc->gadget_driver && dwc->gadget_driver->resume) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->resume)
 		dwc->gadget_driver->resume(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 

static void dwc3_reset_gadget(struct dwc3 *dwc)

drivers/usb/dwc3/gadget.c +3 -4
@@ -3585,11 +3593,10 @@ static void 
 dwc3_gadget_wakeup_interrupt(struct dwc3 *dwc)
 	 * implemented.
 	 */
 
-	if (dwc->gadget_driver && dwc->gadget_driver->resume) {
-		spin_unlock(&dwc->lock);
+	spin_unlock(&dwc->lock);
+	if (dwc->async_callbacks && dwc->gadget_driver->resume)
 		dwc->gadget_driver->resume(dwc->gadget);
-		spin_lock(&dwc->lock);
-	}
+	spin_lock(&dwc->lock);
 }
 

static void dwc3_gadget_linksts_change_interrupt(struct dwc3 *dwc,
--
2.25.1

#3 lore raw
Quoted 23 lines · On Sat, Jun 26, 2021 at 09:16:25AM +0800, [email protected] wrote:
Quoted 13 lines · Alan Stern
Quoted 11 lines · Linyu Yuan
drivers/usb/dwc3/ep0.c +1 -1
@@ -597,10 +597,11 @@ static int dwc3_ep0_set_address(struct dwc3
 *dwc, struct usb_ctrlrequest *ctrl)
 
 static int dwc3_ep0_delegate_req(struct dwc3 *dwc, struct
 usb_ctrlrequest *ctrl)
 {
-	int ret;
+	int ret = 0;
 
Quoted 9 lines

drivers/usb/dwc3/ep0.c static int dwc3_ep0_set_address(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl) · line 605

Quoted 5 lines
drivers/usb/dwc3/ep0.c +2 -1
 	spin_unlock(&dwc->lock);
-	ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
+	if (dwc->async_callbacks)
+		ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
 	spin_lock(&dwc->lock);

Here and in the other places, you should test dwc->async_callbacks _before_ dropping the spinlock. Otherwise there is a race (the flag could be written at about the same time it is checked).

Quoted 6 lines

thanks for your comments,

if you think there is race here, how to make sure gadget_driver pointer is safe,
this is closest place where we can confirm it is non-NULL by checking async_callbacks ?

I explained this twice already: We know that gadget_driver is not NULL because usb_gadget_remove_driver calls synchronize_irq before doing usb_gadget_udc_stop.

Look at this timing diagram:

	CPU0				CPU1
	----				----
	IRQ happens for setup packet
	  Handler sees async_callbacks
	    is enabled
	  Handler unlocks dwc->lock
					usb_gadget_remove_driver runs
					  Disables async callbacks
					  Calls synchronize_irq
	  Handler calls dwc->		  . waits for IRQ handler to 
	    gadget_driver->setup	  .   return
	  Handler locks dwc-lock	  .
	  ...				  .
	  Handler returns		  .
					  . synchronize_irq returns
					  Calls usb_gadget_udc_stop
					    dwc->gadget_driver is
					      set to NULL

As you can see, dwc->gadget_driver is non-NULL when CPU0 uses it, even though async_callbacks gets cleared during the time when the lock is released.

Alan Stern

#4 lore raw
Quoted 56 lines · Alan Stern
Quoted 23 lines · On Sat, Jun 26, 2021 at 09:16:25AM +0800, [email protected] wrote:
Quoted 13 lines · Alan Stern
Quoted 11 lines · Linyu Yuan
drivers/usb/dwc3/ep0.c +1 -1
@@ -597,10 +597,11 @@ static int dwc3_ep0_set_address(struct dwc3
 *dwc, struct usb_ctrlrequest *ctrl)
 
 static int dwc3_ep0_delegate_req(struct dwc3 *dwc, struct
 usb_ctrlrequest *ctrl)
 {
-	int ret;
+	int ret = 0;
 
Quoted 9 lines

drivers/usb/dwc3/ep0.c static int dwc3_ep0_set_address(struct dwc3 *dwc, struct usb_ctrlrequest *ctrl) · line 605

Quoted 5 lines
drivers/usb/dwc3/ep0.c +2 -1
 	spin_unlock(&dwc->lock);
-	ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
+	if (dwc->async_callbacks)
+		ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
 	spin_lock(&dwc->lock);

Here and in the other places, you should test dwc->async_callbacks _before_ dropping the spinlock. Otherwise there is a race (the flag could be written at about the same time it is checked).

Quoted 6 lines

thanks for your comments,

if you think there is race here, how to make sure gadget_driver pointer is safe,
this is closest place where we can confirm it is non-NULL by checking async_callbacks ?

I explained this twice already: We know that gadget_driver is not NULL because usb_gadget_remove_driver calls synchronize_irq before doing usb_gadget_udc_stop.

Look at this timing diagram:

	CPU0				CPU1
	----				----
	IRQ happens for setup packet
	  Handler sees async_callbacks
	    is enabled
	  Handler unlocks dwc->lock
					usb_gadget_remove_driver runs
					  Disables async callbacks
					  Calls synchronize_irq
	  Handler calls dwc->		  . waits for IRQ handler to
	    gadget_driver->setup	  .   return
	  Handler locks dwc-lock	  .
	  ...				  .
	  Handler returns		  .
					  . synchronize_irq returns
					  Calls usb_gadget_udc_stop
					    dwc->gadget_driver is
					      set to NULL
Quoted 3 lines

As you can see, dwc->gadget_driver is non-NULL when CPU0 uses it, even though async_callbacks gets cleared during the time when the lock is released.

thanks for your patient explanation,
but from this part, seem it is synchronize_irq() help to avoid NULL pointer crash.

can you also explain how async_callbacks flag help here ?

Quoted 2 lines

Alan Stern

#5 lore raw
Quoted 40 lines · On Sun, Jun 27, 2021 at 10:48:56AM +0800, [email protected] wrote:
Quoted 39 lines · Alan Stern
Quoted 5 lines · On Sat, Jun 26, 2021 at 09:16:25AM +0800, [email protected] wrote:
Quoted 3 lines · Alan Stern

Here and in the other places, you should test dwc->async_callbacks _before_ dropping the spinlock. Otherwise there is a race (the flag could be written at about the same time it is checked).

Quoted 7 lines

thanks for your comments,

if you think there is race here, how to make sure gadget_driver pointer is
safe,
this is closest place where we can confirm it is non-NULL by checking async_callbacks ?

I explained this twice already: We know that gadget_driver is not NULL because usb_gadget_remove_driver calls synchronize_irq before doing usb_gadget_udc_stop.

Look at this timing diagram:

	CPU0				CPU1
	----				----
	IRQ happens for setup packet
	  Handler sees async_callbacks
	    is enabled
	  Handler unlocks dwc->lock
					usb_gadget_remove_driver runs
					  Disables async callbacks
					  Calls synchronize_irq
	  Handler calls dwc->		  . waits for IRQ handler to
	    gadget_driver->setup	  .   return
	  Handler locks dwc-lock	  .
	  ...				  .
	  Handler returns		  .
					  . synchronize_irq returns
					  Calls usb_gadget_udc_stop
					    dwc->gadget_driver is
					      set to NULL
Quoted 6 lines
Quoted 3 lines

As you can see, dwc->gadget_driver is non-NULL when CPU0 uses it, even though async_callbacks gets cleared during the time when the lock is released.

thanks for your patient explanation,
but from this part, seem it is synchronize_irq() help to avoid NULL pointer crash.

That's right.

Quoted 1 line

can you also explain how async_callbacks flag help here ?

It doesn't help in the situation shown above, but it does help in other situations. Consider this timing diagram:

CPU0 CPU1
---- ----
usb_gadget_remove_driver runs
Disables async callbacks Calls synchronize_irq
synchronize_irq returns
Calls udc_driver_unbind
IRQ happens for disconnect
Handler sees async_callbacks
is disabled
Handler returns
Calls usb_gadget_udc_stop
dwc->gadget_driver is
set to NULL

With the async_callbacks check, everything works okay. But now look at what would happen without the async_callbacks mechanism:

	CPU0				CPU1
	----				----
					usb_gadget_remove_driver runs
					  Calls synchronize_irq
					    synchronize_irq returns
					  Calls udc_driver_unbind
	IRQ happens for disconnect
	  Handler unlocks dwc->lock
	  Calls dwc->gadget_driver->disconnect
	    Gadget driver has already been unbound
	      and is not prepared to handle a
	      callback, so it crashes
					  Calls usb_gadget_udc_stop
					    dwc->gadget_driver is
					      set to NULL

Without the async_callbacks mechanism, the gadget driver can get a callback at the wrong time (after it has been unbound), which might cause it to crash.

Alan Stern

#6 lore raw
Quoted 89 lines · Alan Stern
Quoted 40 lines · On Sun, Jun 27, 2021 at 10:48:56AM +0800, [email protected] wrote:
Quoted 39 lines · Alan Stern
Quoted 5 lines · On Sat, Jun 26, 2021 at 09:16:25AM +0800, [email protected] wrote:
Quoted 3 lines · Alan Stern

Here and in the other places, you should test dwc->async_callbacks _before_ dropping the spinlock. Otherwise there is a race (the flag could be written at about the same time it is checked).

Quoted 7 lines

thanks for your comments,

if you think there is race here, how to make sure gadget_driver pointer is
safe,
this is closest place where we can confirm it is non-NULL by checking async_callbacks ?

I explained this twice already: We know that gadget_driver is not NULL because usb_gadget_remove_driver calls synchronize_irq before doing usb_gadget_udc_stop.

Look at this timing diagram:

	CPU0				CPU1
	----				----
	IRQ happens for setup packet
	  Handler sees async_callbacks
	    is enabled
	  Handler unlocks dwc->lock
					usb_gadget_remove_driver runs
					  Disables async callbacks
					  Calls synchronize_irq
	  Handler calls dwc->		  . waits for IRQ handler to
	    gadget_driver->setup	  .   return
	  Handler locks dwc-lock	  .
	  ...				  .
	  Handler returns		  .
					  . synchronize_irq returns
					  Calls usb_gadget_udc_stop
					    dwc->gadget_driver is
					      set to NULL
Quoted 6 lines
Quoted 3 lines

As you can see, dwc->gadget_driver is non-NULL when CPU0 uses it, even though async_callbacks gets cleared during the time when the lock is released.

thanks for your patient explanation,
but from this part, seem it is synchronize_irq() help to avoid NULL pointer crash.

That's right.

Quoted 1 line

can you also explain how async_callbacks flag help here ?

It doesn't help in the situation shown above, but it does help in other situations. Consider this timing diagram:

CPU0 CPU1
---- ----
usb_gadget_remove_driver runs
Disables async callbacks Calls synchronize_irq
synchronize_irq returns
Calls udc_driver_unbind
IRQ happens for disconnect
Handler sees async_callbacks
is disabled
Handler returns
Calls usb_gadget_udc_stop
dwc->gadget_driver is
set to NULL

With the async_callbacks check, everything works okay. But now look at what would happen without the async_callbacks mechanism:

	CPU0				CPU1
	----				----
					usb_gadget_remove_driver runs
					  Calls synchronize_irq
					    synchronize_irq returns
					  Calls udc_driver_unbind
	IRQ happens for disconnect
	  Handler unlocks dwc->lock
	  Calls dwc->gadget_driver->disconnect
	    Gadget driver has already been unbound
	      and is not prepared to handle a
	      callback, so it crashes
					  Calls usb_gadget_udc_stop
					    dwc->gadget_driver is
					      set to NULL
Quoted 3 lines

Without the async_callbacks mechanism, the gadget driver can get a callback at the wrong time (after it has been unbound), which might cause it to crash.

1. do you think we need to back to my original patch,
https://lore.kernel.org/linux-usb/[email protected]/T/#t

i think we can add the spin lock or mutex lock to protect this kind of race from UDC layer, it will be easy understanding.

2. if you insist this kind of change, how to change following code in dwc3 ? if (dwc->gadget_driver && dwc->gadget_driver->disconnect) {

2.1 if (dwc->async_callbacks && dwc->gadget_driver->disconnect) {
or
2.2 if (dwc->async_callbacks && vdwc->gadget_driver && dwc->gadget_driver->disconnect) {

Quoted 2 lines

Alan Stern

#7 lore raw
Quoted 18 lines · On Mon, Jun 28, 2021 at 05:36:22PM +0800, [email protected] wrote:
Quoted 17 lines · Alan Stern
	CPU0				CPU1
	----				----
					usb_gadget_remove_driver runs
					  Calls synchronize_irq
					    synchronize_irq returns
					  Calls udc_driver_unbind
	IRQ happens for disconnect
	  Handler unlocks dwc->lock
	  Calls dwc->gadget_driver->disconnect
	    Gadget driver has already been unbound
	      and is not prepared to handle a
	      callback, so it crashes
					  Calls usb_gadget_udc_stop
					    dwc->gadget_driver is
					      set to NULL
Quoted 5 lines
Quoted 3 lines

Without the async_callbacks mechanism, the gadget driver can get a callback at the wrong time (after it has been unbound), which might cause it to crash.

1. do you think we need to back to my original patch,
https://lore.kernel.org/linux-usb/[email protected]/T/#t

No, I think the async_callbacks approach is slightly better.

Quoted 2 lines

i think we can add the spin lock or mutex lock to protect this kind of race from UDC layer, it will be easy understanding.

We don't actually need a lock or mutex to fix this problem. We just need to make the remove_driver sequence issue two calls to the UDC driver rather than just one: async_callbacks and udc_stop.

Quoted 7 lines

2. if you insist this kind of change, how to change following code in dwc3 ? if (dwc->gadget_driver && dwc->gadget_driver->disconnect) {

2.1 if (dwc->async_callbacks && dwc->gadget_driver->disconnect) { or
2.2 if (dwc->async_callbacks && vdwc->gadget_driver && dwc->gadget_driver->disconnect) {

Either one would be okay. If async_callbacks is enabled then dwc->gadget_driver should never be NULL, but it won't hurt to check. After all, disconnect does not occur often and it doesn't need to run extremely quickly.

Alan Stern