Re: [PATCH v2 1/1] usb: dwc3: gadget: fix IRQ storm on invalid event buffer count

[Thinh Nguyen]

Hi Liu,

On Fri, Aug 28, 2026, Liu Jiazi wrote:
> Hi @Thinh.Nguyen@xxxxxxxxxxxx , Krishna,
> 
> Thanks for the detailed review.
> For the gadget ops protection, I propose adding the check only in
> dwc3_gadget_ep_queue.
> This covers two scenarios:
> 1. The window between error detection and soft_disconnect completing.
> 2. After soft_disconnect, existing checks on
> connected/pullups_connected/endpoint.desc naturally block all ops.
> This is also consistent with how other gadget handles it.
> 

Looking at the change more closely, especially the gadget_ops and
ep_ops. We need to consider the followings:

1) We need a mutex to prevent a race against dwc3_gadget_pullup()
2) ep_enable, ep_queue, ep_set_halt/wedge, wakeup, and function_wakeup
   need to be guarded. The other ones are for cleanup and should be
   left alone.
3) We shouldn't use disable_irq_nosync() as suggested previously because
   we need the event interrupt to drain the GEVENTCOUNT to allow the
   controller to halt after clearing the DCTL.RUN_STOP bit. But we
   should move the synchronize_irq() before the soft_connect.

Can you try the change below (not tested).

Thanks,
Thinh



diff --git a/drivers/usb/dwc3/core.h b/drivers/usb/dwc3/core.h
index 608daeb7ef10..f3dabb37f806 100644
--- a/drivers/usb/dwc3/core.h
+++ b/drivers/usb/dwc3/core.h
@@ -49,6 +49,7 @@
 #define DWC3_ENDPOINTS_NUM	32
 #define DWC3_XHCI_RESOURCES_NUM	2
 #define DWC3_ISOC_MAX_RETRIES	5
+#define DWC3_ERR_RECOVERY_MAX	3
 
 #define DWC3_SCRATCHBUF_SIZE	4096	/* each buffer is assumed to be 4KiB */
 #define DWC3_EVENT_BUFFERS_SIZE	4096
@@ -841,6 +842,12 @@ enum dwc3_link_state {
 	DWC3_LINK_STATE_MASK		= 0x0f,
 };
 
+enum dwc3_err_state {
+	DWC3_ERR_NONE = 0,
+	DWC3_ERR_RECOVERY,
+	DWC3_ERR_UNRECOVERABLE,
+};
+
 /* TRB Length, PCM and Status */
 #define DWC3_TRB_SIZE_MASK	(0x00ffffff)
 #define DWC3_TRB_SIZE_LENGTH(n)	((n) & DWC3_TRB_SIZE_MASK)
@@ -1004,6 +1011,7 @@ struct dwc3_glue_ops {
 /**
  * struct dwc3 - representation of our controller
  * @drd_work: workqueue used for role swapping
+ * @err_recovery_work: workqueue used for controller error recovery
  * @ep0_trb: trb which is used for the ctrl_req
  * @bounce: address of bounce buffer
  * @setup_buf: used while precessing STD USB requests
@@ -1013,6 +1021,7 @@ struct dwc3_glue_ops {
  * @ep0_in_setup: one control transfer is completed and enter setup phase
  * @lock: for synchronizing
  * @mutex: for mode switching
+ * @connect_mutex: for the pull-up and err_recovery_work
  * @dev: pointer to our struct device
  * @sysdev: pointer to the DMA-capable device
  * @xhci: pointer to our xHCI child
@@ -1079,6 +1088,9 @@ struct dwc3_glue_ops {
  * @ep0_next_event: hold the next expected event
  * @ep0state: state of endpoint zero
  * @link_state: link state
+ * @err_state: current error recovery state.
+ * @err_recovery_count: number of consecutive error recovery attempts until
+ *		confirmed healthy and reset to 0 on reset event.
  * @speed: device speed (super, high, full, low)
  * @hwparams: copy of hwparams registers
  * @regset: debugfs pointer to regdump file
@@ -1189,6 +1201,7 @@ struct dwc3_glue_ops {
  */
 struct dwc3 {
 	struct work_struct	drd_work;
+	struct work_struct	err_recovery_work;
 	struct dwc3_trb		*ep0_trb;
 	void			*bounce;
 	u8			*setup_buf;
@@ -1203,6 +1216,9 @@ struct dwc3 {
 	/* mode switching lock */
 	struct mutex		mutex;
 
+	/* serializes pull-up run/stop vs error recovery */
+	struct mutex		connect_mutex;
+
 	struct device		*dev;
 	struct device		*sysdev;
 
@@ -1332,6 +1348,9 @@ struct dwc3 {
 	enum dwc3_ep0_next	ep0_next_event;
 	enum dwc3_ep0_state	ep0state;
 	enum dwc3_link_state	link_state;
+	enum dwc3_err_state	err_state;
+
+	u32			err_recovery_count;
 
 	u16			u2sel;
 	u16			u2pel;
diff --git a/drivers/usb/dwc3/ep0.c b/drivers/usb/dwc3/ep0.c
index bfe616194dfa..fa3d8b9f6f8e 100644
--- a/drivers/usb/dwc3/ep0.c
+++ b/drivers/usb/dwc3/ep0.c
@@ -200,6 +200,11 @@ int dwc3_gadget_ep0_queue(struct usb_ep *ep, struct usb_request *request,
 	int				ret;
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		ret = -ESHUTDOWN;
+		goto out;
+	}
+
 	if (!dep->endpoint.desc || !dwc->pullups_connected || !dwc->connected) {
 		dev_err(dwc->dev, "%s: can't queue to disabled endpoint\n",
 				dep->name);
@@ -271,6 +276,10 @@ int dwc3_gadget_ep0_set_halt(struct usb_ep *ep, int value)
 	int				ret;
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		spin_unlock_irqrestore(&dwc->lock, flags);
+		return -ESHUTDOWN;
+	}
 	ret = __dwc3_gadget_ep0_set_halt(ep, value);
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
index fa944856f956..87b535fb9443 100644
--- a/drivers/usb/dwc3/gadget.c
+++ b/drivers/usb/dwc3/gadget.c
@@ -1149,6 +1149,10 @@ static int dwc3_gadget_ep_enable(struct usb_ep *ep,
 		return 0;
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		spin_unlock_irqrestore(&dwc->lock, flags);
+		return -ESHUTDOWN;
+	}
 	ret = __dwc3_gadget_ep_enable(dep, DWC3_DEPCFG_ACTION_INIT);
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
@@ -2055,6 +2059,10 @@ static int dwc3_gadget_ep_queue(struct usb_ep *ep, struct usb_request *request,
 	int				ret;
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		spin_unlock_irqrestore(&dwc->lock, flags);
+		return -ESHUTDOWN;
+	}
 	ret = __dwc3_gadget_ep_queue(dep, req);
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
@@ -2287,6 +2295,10 @@ static int dwc3_gadget_ep_set_halt(struct usb_ep *ep, int value)
 	int				ret;
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		spin_unlock_irqrestore(&dwc->lock, flags);
+		return -ESHUTDOWN;
+	}
 	ret = __dwc3_gadget_ep_set_halt(dep, value, false);
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
@@ -2301,6 +2313,10 @@ static int dwc3_gadget_ep_set_wedge(struct usb_ep *ep)
 	int				ret;
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		spin_unlock_irqrestore(&dwc->lock, flags);
+		return -ESHUTDOWN;
+	}
 	dep->flags |= DWC3_EP_WEDGE;
 
 	if (dep->number == 0 || dep->number == 1)
@@ -2432,6 +2448,10 @@ static int dwc3_gadget_wakeup(struct usb_gadget *g)
 	}
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		spin_unlock_irqrestore(&dwc->lock, flags);
+		return -ESHUTDOWN;
+	}
 	if (!dwc->gadget->wakeup_armed) {
 		dev_err(dwc->dev, "not armed for remote wakeup\n");
 		spin_unlock_irqrestore(&dwc->lock, flags);
@@ -2459,6 +2479,10 @@ static int dwc3_gadget_func_wakeup(struct usb_gadget *g, int intf_id)
 	}
 
 	spin_lock_irqsave(&dwc->lock, flags);
+	if (dwc->err_state != DWC3_ERR_NONE) {
+		spin_unlock_irqrestore(&dwc->lock, flags);
+		return -ESHUTDOWN;
+	}
 	/*
 	 * If the link is in U3, signal for remote wakeup and wait for the
 	 * link to transition to U0 before sending device notification.
@@ -2828,10 +2852,13 @@ static int dwc3_gadget_pullup(struct usb_gadget *g, int is_on)
 
 	synchronize_irq(dwc->irq_gadget);
 
+	/* Serialize against dwc3_err_recovery_work() */
+	mutex_lock(&dwc->connect_mutex);
 	if (!is_on)
 		ret = dwc3_gadget_soft_disconnect(dwc);
 	else
 		ret = dwc3_gadget_soft_connect(dwc);
+	mutex_unlock(&dwc->connect_mutex);
 
 	pm_runtime_put(dwc->dev);
 
@@ -4148,6 +4175,10 @@ static void dwc3_gadget_reset_interrupt(struct dwc3 *dwc)
 
 	dwc->suspended = false;
 
+	/* The controller is recovered. */
+	if (dwc->err_state == DWC3_ERR_NONE)
+		dwc->err_recovery_count = 0;
+
 	/*
 	 * Ideally, dwc3_reset_gadget() would trigger the function
 	 * drivers to stop any active transfers through ep disable.
@@ -4635,6 +4666,12 @@ static irqreturn_t dwc3_thread_interrupt(int irq, void *_evt)
 	return ret;
 }
 
+static void dwc3_schedule_err_recovery(struct dwc3 *dwc)
+{
+	dwc->err_state = DWC3_ERR_RECOVERY;
+	schedule_work(&dwc->err_recovery_work);
+}
+
 static irqreturn_t dwc3_check_event_buf(struct dwc3_event_buffer *evt)
 {
 	struct dwc3 *dwc = evt->dwc;
@@ -4668,9 +4705,21 @@ static irqreturn_t dwc3_check_event_buf(struct dwc3_event_buffer *evt)
 		return IRQ_NONE;
 
 	if (count > evt->length) {
-		dev_err_ratelimited(dwc->dev, "invalid count(%u) > evt->length(%u)\n",
+		dev_err(dwc->dev, "invalid count(%u) > evt->length(%u)\n",
 			count, evt->length);
-		return IRQ_NONE;
+		/*
+		 * This is a fatal error - the driver and controller are out of
+		 * sync on which event has been consumed. Reinitializing the
+		 * controller is required to recover. Write the bogus count back
+		 * to GEVNTCOUNT to clear the IRQ source, consistent with the
+		 * stale event clearing in dwc3_event_buffers_setup(), then
+		 * schedule error recovery.
+		 */
+		spin_lock(&dwc->lock);
+		dwc3_schedule_err_recovery(dwc);
+		spin_unlock(&dwc->lock);
+		dwc3_writel(dwc, DWC3_GEVNTCOUNT(0), count);
+		return IRQ_HANDLED;
 	}
 
 	evt->count = count;
@@ -4730,6 +4779,53 @@ static void dwc_gadget_release(struct device *dev)
 	kfree(gadget);
 }
 
+static void dwc3_err_recovery_work(struct work_struct *work)
+{
+	struct dwc3 *dwc = container_of(work, struct dwc3, err_recovery_work);
+	unsigned long flags;
+	int ret;
+
+	dwc->err_recovery_count++;
+
+	/* serializes against dwc3_gadget_pullup() */
+	mutex_lock(&dwc->connect_mutex);
+
+	ret = dwc3_gadget_soft_disconnect(dwc);
+	if (ret)
+		goto err_unrecoverable;
+
+	dwc3_disconnect_gadget_sleepable(dwc);
+
+	if (dwc->err_recovery_count > DWC3_ERR_RECOVERY_MAX)
+		goto err_unrecoverable;
+
+	if (dwc->softconnect) {
+		/*
+		 * Wait irq to finish before soft_connect resets evt->lpos and
+		 * the event buffer registers to avoid racing with
+		 * dwc3_process_event_buf().
+		 */
+		synchronize_irq(dwc->irq_gadget);
+
+		ret = dwc3_gadget_soft_connect(dwc);
+		if (ret)
+			goto err_unrecoverable;
+	}
+	mutex_unlock(&dwc->connect_mutex);
+
+	spin_lock_irqsave(&dwc->lock, flags);
+	dwc->err_state = DWC3_ERR_NONE;
+	spin_unlock_irqrestore(&dwc->lock, flags);
+	return;
+
+err_unrecoverable:
+	dev_err(dwc->dev, "Unable to recover the controller\n");
+	mutex_unlock(&dwc->connect_mutex);
+	spin_lock_irqsave(&dwc->lock, flags);
+	dwc->err_state = DWC3_ERR_UNRECOVERABLE;
+	spin_unlock_irqrestore(&dwc->lock, flags);
+}
+
 /**
  * dwc3_gadget_init - initializes gadget related registers
  * @dwc: pointer to our controller context structure
@@ -4773,6 +4869,8 @@ int dwc3_gadget_init(struct dwc3 *dwc)
 	}
 
 	init_completion(&dwc->ep0_in_setup);
+	INIT_WORK(&dwc->err_recovery_work, dwc3_err_recovery_work);
+	mutex_init(&dwc->connect_mutex);
 	dwc->gadget = kzalloc_obj(struct usb_gadget);
 	if (!dwc->gadget) {
 		ret = -ENOMEM;
@@ -4869,6 +4967,8 @@ void dwc3_gadget_exit(struct dwc3 *dwc)
 	if (!dwc->gadget)
 		return;
 
+	cancel_work_sync(&dwc->err_recovery_work);
+	mutex_destroy(&dwc->connect_mutex);
 	dwc3_enable_susphy(dwc, true);
 	usb_del_gadget(dwc->gadget);
 	dwc3_gadget_free_endpoints(dwc);