From: Sarah Sharp <sarah.a.sharp@linux.intel.com>

In order to giveback a canceled URB, we must ensure that the xHCI
hardware will not access the buffer in an URB.  We can't modify the
buffer pointers on endpoint rings without issuing and waiting for a stop
endpoint command.  Since URBs can be canceled in interrupt context, we
can't wait on that command.  The old code trusted that the host
controller would respond to the command, and would giveback the URBs in
the event handler.  If the hardware never responds to the stop endpoint
command, the URBs will never be completed, and we might hang the USB
subsystem.

Implement a watchdog timer that is spawned whenever a stop endpoint
command is queued.  If a stop endpoint command event is found on the
event ring during an interrupt, we need to stop the watchdog timer with
del_timer().  Since del_timer() can fail if the timer is running and
waiting on the xHCI lock, we need a way to signal to the timer that
everything is fine and it should exit.  If we simply clear
EP_HALT_PENDING, a new stop endpoint command could sneak in and set it
before the watchdog timer can grab the lock.

Instead we use a combination of the EP_HALT_PENDING flag and a counter
for the number of pending stop endpoint commands
(xhci_virt_ep->stop_cmds_pending).  If we need to cancel the watchdog
timer and del_timer() succeeds, we decrement the number of pending stop
endpoint commands.  If del_timer() fails, we leave the number of pending
stop endpoint commands alone.  In either case, we clear the
EP_HALT_PENDING flag.

The timer will decrement the number of pending stop endpoint commands
once it obtains the lock.  If the timer is the tail end of the last stop
endpoint command (xhci_virt_ep->stop_cmds_pending == 0), and the
endpoint's command is still pending (EP_HALT_PENDING is set), we assume
the host is dying.  The watchdog timer will set XHCI_STATE_DYING, try to
halt the xHCI host, and give back all pending URBs.

Various other places in the driver need to check whether the xHCI host
is dying.  If the interrupt handler ever notices, it should immediately
stop processing events.  The URB enqueue function should also return
-ESHUTDOWN.  The URB dequeue function should simply return the value
of usb_hcd_check_unlink_urb() and the watchdog timer will take care of
giving the URB back.  When a device is disconnected, the xHCI hardware
structures should be freed without issuing a disable slot command (since
the hardware probably won't respond to it anyway).  The debugging
polling loop should stop polling if the host is dying.

When a device is disconnected, any pending watchdog timers are killed
with del_timer_sync().  It must be synchronous so that the watchdog
timer doesn't attempt to access the freed endpoint structures.

Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
---
 drivers/usb/host/xhci-hcd.c  |   52 ++++++++++++-
 drivers/usb/host/xhci-mem.c  |   15 +++-
 drivers/usb/host/xhci-ring.c |  170 +++++++++++++++++++++++++++++++++++++++---
 drivers/usb/host/xhci.h      |   22 ++++++
 4 files changed, 243 insertions(+), 16 deletions(-)

diff --git a/drivers/usb/host/xhci-hcd.c b/drivers/usb/host/xhci-hcd.c
index 5839453..0d5a856 100644
--- a/drivers/usb/host/xhci-hcd.c
+++ b/drivers/usb/host/xhci-hcd.c
@@ -246,8 +246,14 @@ static void xhci_work(struct xhci_hcd *xhci)
 	/* Flush posted writes */
 	xhci_readl(xhci, &xhci->ir_set->irq_pending);
 
-	/* FIXME this should be a delayed service routine that clears the EHB */
-	xhci_handle_event(xhci);
+	if (xhci->xhc_state & XHCI_STATE_DYING)
+		xhci_dbg(xhci, "xHCI dying, ignoring interrupt. "
+				"Shouldn't IRQs be disabled?\n");
+	else
+		/* FIXME this should be a delayed service routine
+		 * that clears the EHB.
+		 */
+		xhci_handle_event(xhci);
 
 	/* Clear the event handler busy flag (RW1C); the event ring should be empty. */
 	temp_64 = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
@@ -320,7 +326,7 @@ void xhci_event_ring_work(unsigned long arg)
 	spin_lock_irqsave(&xhci->lock, flags);
 	temp = xhci_readl(xhci, &xhci->op_regs->status);
 	xhci_dbg(xhci, "op reg status = 0x%x\n", temp);
-	if (temp == 0xffffffff) {
+	if (temp == 0xffffffff || (xhci->xhc_state & XHCI_STATE_DYING)) {
 		xhci_dbg(xhci, "HW died, polling stopped.\n");
 		spin_unlock_irqrestore(&xhci->lock, flags);
 		return;
@@ -710,16 +716,22 @@ int xhci_urb_enqueue(struct usb_hcd *hcd, struct urb *urb, gfp_t mem_flags)
 		 * atomic context to this function, which may allocate memory.
 		 */
 		spin_lock_irqsave(&xhci->lock, flags);
+		if (xhci->xhc_state & XHCI_STATE_DYING)
+			goto dying;
 		ret = xhci_queue_ctrl_tx(xhci, GFP_ATOMIC, urb,
 				slot_id, ep_index);
 		spin_unlock_irqrestore(&xhci->lock, flags);
 	} else if (usb_endpoint_xfer_bulk(&urb->ep->desc)) {
 		spin_lock_irqsave(&xhci->lock, flags);
+		if (xhci->xhc_state & XHCI_STATE_DYING)
+			goto dying;
 		ret = xhci_queue_bulk_tx(xhci, GFP_ATOMIC, urb,
 				slot_id, ep_index);
 		spin_unlock_irqrestore(&xhci->lock, flags);
 	} else if (usb_endpoint_xfer_int(&urb->ep->desc)) {
 		spin_lock_irqsave(&xhci->lock, flags);
+		if (xhci->xhc_state & XHCI_STATE_DYING)
+			goto dying;
 		ret = xhci_queue_intr_tx(xhci, GFP_ATOMIC, urb,
 				slot_id, ep_index);
 		spin_unlock_irqrestore(&xhci->lock, flags);
@@ -728,6 +740,12 @@ int xhci_urb_enqueue(struct usb_hcd *hcd, struct urb *urb, gfp_t mem_flags)
 	}
 exit:
 	return ret;
+dying:
+	xhci_dbg(xhci, "Ep 0x%x: URB %p submitted for "
+			"non-responsive xHCI host.\n",
+			urb->ep->desc.bEndpointAddress, urb);
+	spin_unlock_irqrestore(&xhci->lock, flags);
+	return -ESHUTDOWN;
 }
 
 /*
@@ -789,6 +807,17 @@ int xhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
 		kfree(td);
 		return ret;
 	}
+	if (xhci->xhc_state & XHCI_STATE_DYING) {
+		xhci_dbg(xhci, "Ep 0x%x: URB %p to be canceled on "
+				"non-responsive xHCI host.\n",
+				urb->ep->desc.bEndpointAddress, urb);
+		/* Let the stop endpoint command watchdog timer (which set this
+		 * state) finish cleaning up the endpoint TD lists.  We must
+		 * have caught it in the middle of dropping a lock and giving
+		 * back an URB.
+		 */
+		goto done;
+	}
 
 	xhci_dbg(xhci, "Cancel URB %p\n", urb);
 	xhci_dbg(xhci, "Event ring:\n");
@@ -806,6 +835,10 @@ int xhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
 	 */
 	if (!(ep->ep_state & EP_HALT_PENDING)) {
 		ep->ep_state |= EP_HALT_PENDING;
+		ep->stop_cmds_pending++;
+		ep->stop_cmd_timer.expires = jiffies +
+			XHCI_STOP_EP_CMD_TIMEOUT * HZ;
+		add_timer(&ep->stop_cmd_timer);
 		xhci_queue_stop_endpoint(xhci, urb->dev->slot_id, ep_index);
 		xhci_ring_cmd_db(xhci);
 	}
@@ -1410,16 +1443,27 @@ void xhci_endpoint_reset(struct usb_hcd *hcd,
 void xhci_free_dev(struct usb_hcd *hcd, struct usb_device *udev)
 {
 	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
+	struct xhci_virt_device *virt_dev;
 	unsigned long flags;
 	u32 state;
+	int i;
 
 	if (udev->slot_id == 0)
 		return;
+	virt_dev = xhci->devs[udev->slot_id];
+	if (!virt_dev)
+		return;
+
+	/* Stop any wayward timer functions (which may grab the lock) */
+	for (i = 0; i < 31; ++i) {
+		virt_dev->eps[i].ep_state &= ~EP_HALT_PENDING;
+		del_timer_sync(&virt_dev->eps[i].stop_cmd_timer);
+	}
 
 	spin_lock_irqsave(&xhci->lock, flags);
 	/* Don't disable the slot if the host controller is dead. */
 	state = xhci_readl(xhci, &xhci->op_regs->status);
-	if (state == 0xffffffff) {
+	if (state == 0xffffffff || (xhci->xhc_state & XHCI_STATE_DYING)) {
 		xhci_free_virt_device(xhci, udev->slot_id);
 		spin_unlock_irqrestore(&xhci->lock, flags);
 		return;
diff --git a/drivers/usb/host/xhci-mem.c b/drivers/usb/host/xhci-mem.c
index b8fd270..fd05247 100644
--- a/drivers/usb/host/xhci-mem.c
+++ b/drivers/usb/host/xhci-mem.c
@@ -248,6 +248,15 @@ struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
 		(ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
 }
 
+static void xhci_init_endpoint_timer(struct xhci_hcd *xhci,
+		struct xhci_virt_ep *ep)
+{
+	init_timer(&ep->stop_cmd_timer);
+	ep->stop_cmd_timer.data = (unsigned long) ep;
+	ep->stop_cmd_timer.function = xhci_stop_endpoint_command_watchdog;
+	ep->xhci = xhci;
+}
+
 /* All the xhci_tds in the ring's TD list should be freed at this point */
 void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
 {
@@ -309,9 +318,11 @@ int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
 	xhci_dbg(xhci, "Slot %d input ctx = 0x%llx (dma)\n", slot_id,
 			(unsigned long long)dev->in_ctx->dma);
 
-	/* Initialize the cancellation list for each endpoint */
-	for (i = 0; i < 31; i++)
+	/* Initialize the cancellation list and watchdog timers for each ep */
+	for (i = 0; i < 31; i++) {
+		xhci_init_endpoint_timer(xhci, &dev->eps[i]);
 		INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
+	}
 
 	/* Allocate endpoint 0 ring */
 	dev->eps[0].ring = xhci_ring_alloc(xhci, 1, true, flags);
diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
index 184e8b6..9541e88 100644
--- a/drivers/usb/host/xhci-ring.c
+++ b/drivers/usb/host/xhci-ring.c
@@ -475,6 +475,35 @@ void xhci_queue_new_dequeue_state(struct xhci_hcd *xhci,
 	ep->ep_state |= SET_DEQ_PENDING;
 }
 
+static inline void xhci_stop_watchdog_timer_in_irq(struct xhci_hcd *xhci,
+		struct xhci_virt_ep *ep)
+{
+	ep->ep_state &= ~EP_HALT_PENDING;
+	/* Can't del_timer_sync in interrupt, so we attempt to cancel.  If the
+	 * timer is running on another CPU, we don't decrement stop_cmds_pending
+	 * (since we didn't successfully stop the watchdog timer).
+	 */
+	if (del_timer(&ep->stop_cmd_timer))
+		ep->stop_cmds_pending--;
+}
+
+/* Must be called with xhci->lock held in interrupt context */
+static void xhci_giveback_urb_in_irq(struct xhci_hcd *xhci,
+		struct xhci_td *cur_td, int status, char *adjective)
+{
+	struct usb_hcd *hcd = xhci_to_hcd(xhci);
+
+	cur_td->urb->hcpriv = NULL;
+	usb_hcd_unlink_urb_from_ep(hcd, cur_td->urb);
+	xhci_dbg(xhci, "Giveback %s URB %p\n", adjective, cur_td->urb);
+
+	spin_unlock(&xhci->lock);
+	usb_hcd_giveback_urb(hcd, cur_td->urb, status);
+	kfree(cur_td);
+	spin_lock(&xhci->lock);
+	xhci_dbg(xhci, "%s URB given back\n", adjective);
+}
+
 /*
  * When we get a command completion for a Stop Endpoint Command, we need to
  * unlink any cancelled TDs from the ring.  There are two ways to do that:
@@ -508,7 +537,7 @@ static void handle_stopped_endpoint(struct xhci_hcd *xhci,
 	ep_ring = ep->ring;
 
 	if (list_empty(&ep->cancelled_td_list)) {
-		ep->ep_state &= ~EP_HALT_PENDING;
+		xhci_stop_watchdog_timer_in_irq(xhci, ep);
 		ring_ep_doorbell(xhci, slot_id, ep_index);
 		return;
 	}
@@ -540,7 +569,7 @@ static void handle_stopped_endpoint(struct xhci_hcd *xhci,
 		list_del(&cur_td->td_list);
 	}
 	last_unlinked_td = cur_td;
-	ep->ep_state &= ~EP_HALT_PENDING;
+	xhci_stop_watchdog_timer_in_irq(xhci, ep);
 
 	/* If necessary, queue a Set Transfer Ring Dequeue Pointer command */
 	if (deq_state.new_deq_ptr && deq_state.new_deq_seg) {
@@ -568,23 +597,136 @@ static void handle_stopped_endpoint(struct xhci_hcd *xhci,
 		hcd_stat_update(xhci->tp_stat, cur_td->urb->actual_length,
 				ktime_sub(stop_time, cur_td->start_time));
 #endif
-		cur_td->urb->hcpriv = NULL;
-		usb_hcd_unlink_urb_from_ep(xhci_to_hcd(xhci), cur_td->urb);
-
-		xhci_dbg(xhci, "Giveback cancelled URB %p\n", cur_td->urb);
-		spin_unlock(&xhci->lock);
 		/* Doesn't matter what we pass for status, since the core will
 		 * just overwrite it (because the URB has been unlinked).
 		 */
-		usb_hcd_giveback_urb(xhci_to_hcd(xhci), cur_td->urb, 0);
-		kfree(cur_td);
+		xhci_giveback_urb_in_irq(xhci, cur_td, 0, "cancelled");
 
-		spin_lock(&xhci->lock);
+		/* Stop processing the cancelled list if the watchdog timer is
+		 * running.
+		 */
+		if (xhci->xhc_state & XHCI_STATE_DYING)
+			return;
 	} while (cur_td != last_unlinked_td);
 
 	/* Return to the event handler with xhci->lock re-acquired */
 }
 
+/* Watchdog timer function for when a stop endpoint command fails to complete.
+ * In this case, we assume the host controller is broken or dying or dead.  The
+ * host may still be completing some other events, so we have to be careful to
+ * let the event ring handler and the URB dequeueing/enqueueing functions know
+ * through xhci->state.
+ *
+ * The timer may also fire if the host takes a very long time to respond to the
+ * command, and the stop endpoint command completion handler cannot delete the
+ * timer before the timer function is called.  Another endpoint cancellation may
+ * sneak in before the timer function can grab the lock, and that may queue
+ * another stop endpoint command and add the timer back.  So we cannot use a
+ * simple flag to say whether there is a pending stop endpoint command for a
+ * particular endpoint.
+ *
+ * Instead we use a combination of that flag and a counter for the number of
+ * pending stop endpoint commands.  If the timer is the tail end of the last
+ * stop endpoint command, and the endpoint's command is still pending, we assume
+ * the host is dying.
+ */
+void xhci_stop_endpoint_command_watchdog(unsigned long arg)
+{
+	struct xhci_hcd *xhci;
+	struct xhci_virt_ep *ep;
+	struct xhci_virt_ep *temp_ep;
+	struct xhci_ring *ring;
+	struct xhci_td *cur_td;
+	int ret, i, j;
+
+	ep = (struct xhci_virt_ep *) arg;
+	xhci = ep->xhci;
+
+	spin_lock(&xhci->lock);
+
+	ep->stop_cmds_pending--;
+	if (xhci->xhc_state & XHCI_STATE_DYING) {
+		xhci_dbg(xhci, "Stop EP timer ran, but another timer marked "
+				"xHCI as DYING, exiting.\n");
+		spin_unlock(&xhci->lock);
+		return;
+	}
+	if (!(ep->stop_cmds_pending == 0 && (ep->ep_state & EP_HALT_PENDING))) {
+		xhci_dbg(xhci, "Stop EP timer ran, but no command pending, "
+				"exiting.\n");
+		spin_unlock(&xhci->lock);
+		return;
+	}
+
+	xhci_warn(xhci, "xHCI host not responding to stop endpoint command.\n");
+	xhci_warn(xhci, "Assuming host is dying, halting host.\n");
+	/* Oops, HC is dead or dying or at least not responding to the stop
+	 * endpoint command.
+	 */
+	xhci->xhc_state |= XHCI_STATE_DYING;
+	/* Disable interrupts from the host controller and start halting it */
+	xhci_quiesce(xhci);
+	spin_unlock(&xhci->lock);
+
+	ret = xhci_halt(xhci);
+
+	spin_lock(&xhci->lock);
+	if (ret < 0) {
+		/* This is bad; the host is not responding to commands and it's
+		 * not allowing itself to be halted.  At least interrupts are
+		 * disabled, so we can set HC_STATE_HALT and notify the
+		 * USB core.  But if we call usb_hc_died(), it will attempt to
+		 * disconnect all device drivers under this host.  Those
+		 * disconnect() methods will wait for all URBs to be unlinked,
+		 * so we must complete them.
+		 */
+		xhci_warn(xhci, "Non-responsive xHCI host is not halting.\n");
+		xhci_warn(xhci, "Completing active URBs anyway.\n");
+		/* We could turn all TDs on the rings to no-ops.  This won't
+		 * help if the host has cached part of the ring, and is slow if
+		 * we want to preserve the cycle bit.  Skip it and hope the host
+		 * doesn't touch the memory.
+		 */
+	}
+	for (i = 0; i < MAX_HC_SLOTS; i++) {
+		if (!xhci->devs[i])
+			continue;
+		for (j = 0; j < 31; j++) {
+			temp_ep = &xhci->devs[i]->eps[j];
+			ring = temp_ep->ring;
+			if (!ring)
+				continue;
+			xhci_dbg(xhci, "Killing URBs for slot ID %u, "
+					"ep index %u\n", i, j);
+			while (!list_empty(&ring->td_list)) {
+				cur_td = list_first_entry(&ring->td_list,
+						struct xhci_td,
+						td_list);
+				list_del(&cur_td->td_list);
+				if (!list_empty(&cur_td->cancelled_td_list))
+					list_del(&cur_td->cancelled_td_list);
+				xhci_giveback_urb_in_irq(xhci, cur_td,
+						-ESHUTDOWN, "killed");
+			}
+			while (!list_empty(&temp_ep->cancelled_td_list)) {
+				cur_td = list_first_entry(
+						&temp_ep->cancelled_td_list,
+						struct xhci_td,
+						cancelled_td_list);
+				list_del(&cur_td->cancelled_td_list);
+				xhci_giveback_urb_in_irq(xhci, cur_td,
+						-ESHUTDOWN, "killed");
+			}
+		}
+	}
+	spin_unlock(&xhci->lock);
+	xhci_to_hcd(xhci)->state = HC_STATE_HALT;
+	xhci_dbg(xhci, "Calling usb_hc_died()\n");
+	usb_hc_died(xhci_to_hcd(xhci));
+	xhci_dbg(xhci, "xHCI host controller is dead.\n");
+}
+
 /*
  * When we get a completion for a Set Transfer Ring Dequeue Pointer command,
  * we need to clear the set deq pending flag in the endpoint ring state, so that
@@ -1333,6 +1475,14 @@ void xhci_handle_event(struct xhci_hcd *xhci)
 	default:
 		xhci->error_bitmask |= 1 << 3;
 	}
+	/* Any of the above functions may drop and re-acquire the lock, so check
+	 * to make sure a watchdog timer didn't mark the host as non-responsive.
+	 */
+	if (xhci->xhc_state & XHCI_STATE_DYING) {
+		xhci_dbg(xhci, "xHCI host dying, returning from "
+				"event handler.\n");
+		return;
+	}
 
 	if (update_ptrs) {
 		/* Update SW and HC event ring dequeue pointer */
diff --git a/drivers/usb/host/xhci.h b/drivers/usb/host/xhci.h
index af3c563..c92f841 100644
--- a/drivers/usb/host/xhci.h
+++ b/drivers/usb/host/xhci.h
@@ -659,6 +659,10 @@ struct xhci_virt_ep {
 	/* The TRB that was last reported in a stopped endpoint ring */
 	union xhci_trb		*stopped_trb;
 	struct xhci_td		*stopped_td;
+	/* Watchdog timer for stop endpoint command to cancel URBs */
+	struct timer_list	stop_cmd_timer;
+	int			stop_cmds_pending;
+	struct xhci_hcd		*xhci;
 };
 
 struct xhci_virt_device {
@@ -1022,6 +1026,8 @@ struct xhci_scratchpad {
 #define	ERST_ENTRIES	1
 /* Poll every 60 seconds */
 #define	POLL_TIMEOUT	60
+/* Stop endpoint command timeout (secs) for URB cancellation watchdog timer */
+#define XHCI_STOP_EP_CMD_TIMEOUT	5
 /* XXX: Make these module parameters */
 
 
@@ -1083,6 +1089,21 @@ struct xhci_hcd {
 	struct timer_list	event_ring_timer;
 	int			zombie;
 #endif
+	/* Host controller watchdog timer structures */
+	unsigned int		xhc_state;
+/* Host controller is dying - not responding to commands. "I'm not dead yet!"
+ *
+ * xHC interrupts have been disabled and a watchdog timer will (or has already)
+ * halt the xHCI host, and complete all URBs with an -ESHUTDOWN code.  Any code
+ * that sees this status (other than the timer that set it) should stop touching
+ * hardware immediately.  Interrupt handlers should return immediately when
+ * they see this status (any time they drop and re-acquire xhci->lock).
+ * xhci_urb_dequeue() should call usb_hcd_check_unlink_urb() and return without
+ * putting the TD on the canceled list, etc.
+ *
+ * There are no reports of xHCI host controllers that display this issue.
+ */
+#define XHCI_STATE_DYING	(1 << 0)
 	/* Statistics */
 	int			noops_submitted;
 	int			noops_handled;
@@ -1279,6 +1300,7 @@ void xhci_cleanup_stalled_ring(struct xhci_hcd *xhci,
 void xhci_queue_config_ep_quirk(struct xhci_hcd *xhci,
 		unsigned int slot_id, unsigned int ep_index,
 		struct xhci_dequeue_state *deq_state);
+void xhci_stop_endpoint_command_watchdog(unsigned long arg);
 
 /* xHCI roothub code */
 int xhci_hub_control(struct usb_hcd *hcd, u16 typeReq, u16 wValue, u16 wIndex,
-- 
1.6.5.5

--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html