1 files changed, 697 insertions, 343 deletions

diff --git a/drivers/usb/host/ehci-q.c b/drivers/usb/host/ehci-q.c
index 2e49de820b1..54f5332f814 100644
--- a/drivers/usb/host/ehci-q.c
+++ b/drivers/usb/host/ehci-q.c
@@ -87,31 +87,31 @@ qtd_fill(struct ehci_hcd *ehci, struct ehci_qtd *qtd, dma_addr_t buf,
 static inline void
 qh_update (struct ehci_hcd *ehci, struct ehci_qh *qh, struct ehci_qtd *qtd)
 {
+	struct ehci_qh_hw *hw = qh->hw;
+
 	/* writes to an active overlay are unsafe */
-	BUG_ON(qh->qh_state != QH_STATE_IDLE);
+	WARN_ON(qh->qh_state != QH_STATE_IDLE);
 
-	qh->hw_qtd_next = QTD_NEXT(ehci, qtd->qtd_dma);
-	qh->hw_alt_next = EHCI_LIST_END(ehci);
+	hw->hw_qtd_next = QTD_NEXT(ehci, qtd->qtd_dma);
+	hw->hw_alt_next = EHCI_LIST_END(ehci);
 
 	/* Except for control endpoints, we make hardware maintain data
 	 * toggle (like OHCI) ... here (re)initialize the toggle in the QH,
 	 * and set the pseudo-toggle in udev. Only usb_clear_halt() will
 	 * ever clear it.
 	 */
-	if (!(qh->hw_info1 & cpu_to_hc32(ehci, 1 << 14))) {
+	if (!(hw->hw_info1 & cpu_to_hc32(ehci, QH_TOGGLE_CTL))) {
 		unsigned	is_out, epnum;
 
-		is_out = !(qtd->hw_token & cpu_to_hc32(ehci, 1 << 8));
-		epnum = (hc32_to_cpup(ehci, &qh->hw_info1) >> 8) & 0x0f;
-		if (unlikely (!usb_gettoggle (qh->dev, epnum, is_out))) {
-			qh->hw_token &= ~cpu_to_hc32(ehci, QTD_TOGGLE);
-			usb_settoggle (qh->dev, epnum, is_out, 1);
+		is_out = qh->is_out;
+		epnum = (hc32_to_cpup(ehci, &hw->hw_info1) >> 8) & 0x0f;
+		if (unlikely(!usb_gettoggle(qh->ps.udev, epnum, is_out))) {
+			hw->hw_token &= ~cpu_to_hc32(ehci, QTD_TOGGLE);
+			usb_settoggle(qh->ps.udev, epnum, is_out, 1);
 		}
 	}
 
-	/* HC must see latest qtd and qh data before we clear ACTIVE+HALT */
-	wmb ();
-	qh->hw_token &= cpu_to_hc32(ehci, QTD_TOGGLE | QTD_STS_PING);
+	hw->hw_token &= cpu_to_hc32(ehci, QTD_TOGGLE | QTD_STS_PING);
 }
 
 /* if it weren't for a common silicon quirk (writing the dummy into the qh
@@ -123,22 +123,72 @@ qh_refresh (struct ehci_hcd *ehci, struct ehci_qh *qh)
 {
 	struct ehci_qtd *qtd;
 
-	if (list_empty (&qh->qtd_list))
-		qtd = qh->dummy;
-	else {
-		qtd = list_entry (qh->qtd_list.next,
-				struct ehci_qtd, qtd_list);
-		/* first qtd may already be partially processed */
-		if (cpu_to_hc32(ehci, qtd->qtd_dma) == qh->hw_current)
-			qtd = NULL;
-	}
+	qtd = list_entry(qh->qtd_list.next, struct ehci_qtd, qtd_list);
 
-	if (qtd)
-		qh_update (ehci, qh, qtd);
+	/*
+	 * first qtd may already be partially processed.
+	 * If we come here during unlink, the QH overlay region
+	 * might have reference to the just unlinked qtd. The
+	 * qtd is updated in qh_completions(). Update the QH
+	 * overlay here.
+	 */
+	if (qh->hw->hw_token & ACTIVE_BIT(ehci))
+		qh->hw->hw_qtd_next = qtd->hw_next;
+	else
+		qh_update(ehci, qh, qtd);
 }
 
 /*-------------------------------------------------------------------------*/
 
+static void qh_link_async(struct ehci_hcd *ehci, struct ehci_qh *qh);
+
+static void ehci_clear_tt_buffer_complete(struct usb_hcd *hcd,
+		struct usb_host_endpoint *ep)
+{
+	struct ehci_hcd		*ehci = hcd_to_ehci(hcd);
+	struct ehci_qh		*qh = ep->hcpriv;
+	unsigned long		flags;
+
+	spin_lock_irqsave(&ehci->lock, flags);
+	qh->clearing_tt = 0;
+	if (qh->qh_state == QH_STATE_IDLE && !list_empty(&qh->qtd_list)
+			&& ehci->rh_state == EHCI_RH_RUNNING)
+		qh_link_async(ehci, qh);
+	spin_unlock_irqrestore(&ehci->lock, flags);
+}
+
+static void ehci_clear_tt_buffer(struct ehci_hcd *ehci, struct ehci_qh *qh,
+		struct urb *urb, u32 token)
+{
+
+	/* If an async split transaction gets an error or is unlinked,
+	 * the TT buffer may be left in an indeterminate state.  We
+	 * have to clear the TT buffer.
+	 *
+	 * Note: this routine is never called for Isochronous transfers.
+	 */
+	if (urb->dev->tt && !usb_pipeint(urb->pipe) && !qh->clearing_tt) {
+#ifdef CONFIG_DYNAMIC_DEBUG
+		struct usb_device *tt = urb->dev->tt->hub;
+		dev_dbg(&tt->dev,
+			"clear tt buffer port %d, a%d ep%d t%08x\n",
+			urb->dev->ttport, urb->dev->devnum,
+			usb_pipeendpoint(urb->pipe), token);
+#endif /* CONFIG_DYNAMIC_DEBUG */
+		if (!ehci_is_TDI(ehci)
+				|| urb->dev->tt->hub !=
+				   ehci_to_hcd(ehci)->self.root_hub) {
+			if (usb_hub_clear_tt_buffer(urb) == 0)
+				qh->clearing_tt = 1;
+		} else {
+
+			/* REVISIT ARC-derived cores don't clear the root
+			 * hub TT buffer in this way...
+			 */
+		}
+	}
+}
+
 static int qtd_copy_status (
 	struct ehci_hcd *ehci,
 	struct urb *urb,
@@ -165,6 +215,14 @@ static int qtd_copy_status (
 		if (token & QTD_STS_BABBLE) {
 			/* FIXME "must" disable babbling device's port too */
 			status = -EOVERFLOW;
+		/* CERR nonzero + halt --> stall */
+		} else if (QTD_CERR(token)) {
+			status = -EPIPE;
+
+		/* In theory, more than one of the following bits can be set
+		 * since they are sticky and the transaction is retried.
+		 * Which to test first is rather arbitrary.
+		 */
 		} else if (token & QTD_STS_MMF) {
 			/* fs/ls interrupt xfer missed the complete-split */
 			status = -EPROTO;
@@ -173,49 +231,14 @@ static int qtd_copy_status (
 				? -ENOSR  /* hc couldn't read data */
 				: -ECOMM; /* hc couldn't write data */
 		} else if (token & QTD_STS_XACT) {
-			/* timeout, bad crc, wrong PID, etc; retried */
-			if (QTD_CERR (token))
-				status = -EPIPE;
-			else {
-				ehci_dbg (ehci, "devpath %s ep%d%s 3strikes\n",
-					urb->dev->devpath,
-					usb_pipeendpoint (urb->pipe),
-					usb_pipein (urb->pipe) ? "in" : "out");
-				status = -EPROTO;
-			}
-		/* CERR nonzero + no errors + halt --> stall */
-		} else if (QTD_CERR (token))
-			status = -EPIPE;
-		else	/* unknown */
+			/* timeout, bad CRC, wrong PID, etc */
+			ehci_dbg(ehci, "devpath %s ep%d%s 3strikes\n",
+				urb->dev->devpath,
+				usb_pipeendpoint(urb->pipe),
+				usb_pipein(urb->pipe) ? "in" : "out");
+			status = -EPROTO;
+		} else {	/* unknown */
 			status = -EPROTO;
-
-		ehci_vdbg (ehci,
-			"dev%d ep%d%s qtd token %08x --> status %d\n",
-			usb_pipedevice (urb->pipe),
-			usb_pipeendpoint (urb->pipe),
-			usb_pipein (urb->pipe) ? "in" : "out",
-			token, status);
-
-		/* if async CSPLIT failed, try cleaning out the TT buffer */
-		if (status != -EPIPE
-				&& urb->dev->tt
-				&& !usb_pipeint(urb->pipe)
-				&& ((token & QTD_STS_MMF) != 0
-					|| QTD_CERR(token) == 0)
-				&& (!ehci_is_TDI(ehci)
-			                || urb->dev->tt->hub !=
-					   ehci_to_hcd(ehci)->self.root_hub)) {
-#ifdef DEBUG
-			struct usb_device *tt = urb->dev->tt->hub;
-			dev_dbg (&tt->dev,
-				"clear tt buffer port %d, a%d ep%d t%08x\n",
-				urb->dev->ttport, urb->dev->devnum,
-				usb_pipeendpoint (urb->pipe), token);
-#endif /* DEBUG */
-			/* REVISIT ARC-derived cores don't clear the root
-			 * hub TT buffer in this way...
-			 */
-			usb_hub_tt_clear_buffer (urb->dev, urb->pipe);
 		}
 	}
 
@@ -224,25 +247,17 @@ static int qtd_copy_status (
 
 static void
 ehci_urb_done(struct ehci_hcd *ehci, struct urb *urb, int status)
-__releases(ehci->lock)
-__acquires(ehci->lock)
 {
-	if (likely (urb->hcpriv != NULL)) {
-		struct ehci_qh	*qh = (struct ehci_qh *) urb->hcpriv;
-
-		/* S-mask in a QH means it's an interrupt urb */
-		if ((qh->hw_info2 & cpu_to_hc32(ehci, QH_SMASK)) != 0) {
-
-			/* ... update hc-wide periodic stats (for usbfs) */
-			ehci_to_hcd(ehci)->self.bandwidth_int_reqs--;
-		}
-		qh_put (qh);
+	if (usb_pipetype(urb->pipe) == PIPE_INTERRUPT) {
+		/* ... update hc-wide periodic stats */
+		ehci_to_hcd(ehci)->self.bandwidth_int_reqs--;
 	}
 
 	if (unlikely(urb->unlinked)) {
 		COUNT(ehci->stats.unlink);
 	} else {
-		if (likely(status == -EINPROGRESS))
+		/* report non-error and short read status as zero */
+		if (status == -EINPROGRESS || status == -EREMOTEIO)
 			status = 0;
 		COUNT(ehci->stats.complete);
 	}
@@ -250,56 +265,53 @@ __acquires(ehci->lock)
 #ifdef EHCI_URB_TRACE
 	ehci_dbg (ehci,
 		"%s %s urb %p ep%d%s status %d len %d/%d\n",
-		__FUNCTION__, urb->dev->devpath, urb,
+		__func__, urb->dev->devpath, urb,
 		usb_pipeendpoint (urb->pipe),
 		usb_pipein (urb->pipe) ? "in" : "out",
 		status,
 		urb->actual_length, urb->transfer_buffer_length);
 #endif
 
-	/* complete() can reenter this HCD */
 	usb_hcd_unlink_urb_from_ep(ehci_to_hcd(ehci), urb);
-	spin_unlock (&ehci->lock);
 	usb_hcd_giveback_urb(ehci_to_hcd(ehci), urb, status);
-	spin_lock (&ehci->lock);
 }
 
-static void start_unlink_async (struct ehci_hcd *ehci, struct ehci_qh *qh);
-static void unlink_async (struct ehci_hcd *ehci, struct ehci_qh *qh);
-
-static void intr_deschedule (struct ehci_hcd *ehci, struct ehci_qh *qh);
 static int qh_schedule (struct ehci_hcd *ehci, struct ehci_qh *qh);
 
 /*
  * Process and free completed qtds for a qh, returning URBs to drivers.
- * Chases up to qh->hw_current.  Returns number of completions called,
- * indicating how much "real" work we did.
+ * Chases up to qh->hw_current.  Returns nonzero if the caller should
+ * unlink qh.
  */
 static unsigned
 qh_completions (struct ehci_hcd *ehci, struct ehci_qh *qh)
 {
-	struct ehci_qtd		*last = NULL, *end = qh->dummy;
+	struct ehci_qtd		*last, *end = qh->dummy;
 	struct list_head	*entry, *tmp;
-	int			last_status = -EINPROGRESS;
+	int			last_status;
 	int			stopped;
-	unsigned		count = 0;
-	int			do_status = 0;
 	u8			state;
-	u32			halt = HALT_BIT(ehci);
-
-	if (unlikely (list_empty (&qh->qtd_list)))
-		return count;
+	struct ehci_qh_hw	*hw = qh->hw;
 
 	/* completions (or tasks on other cpus) must never clobber HALT
 	 * till we've gone through and cleaned everything up, even when
 	 * they add urbs to this qh's queue or mark them for unlinking.
 	 *
 	 * NOTE:  unlinking expects to be done in queue order.
+	 *
+	 * It's a bug for qh->qh_state to be anything other than
+	 * QH_STATE_IDLE, unless our caller is scan_async() or
+	 * scan_intr().
 	 */
 	state = qh->qh_state;
 	qh->qh_state = QH_STATE_COMPLETING;
 	stopped = (state == QH_STATE_IDLE);
 
+ rescan:
+	last = NULL;
+	last_status = -EINPROGRESS;
+	qh->dequeue_during_giveback = 0;
+
 	/* remove de-activated QTDs from front of queue.
 	 * after faults (including short reads), cleanup this urb
 	 * then let the queue advance.
@@ -309,7 +321,6 @@ qh_completions (struct ehci_hcd *ehci, struct ehci_qh *qh)
 		struct ehci_qtd	*qtd;
 		struct urb	*urb;
 		u32		token = 0;
-		int		qtd_status;
 
 		qtd = list_entry (entry, struct ehci_qtd, qtd_list);
 		urb = qtd->urb;
@@ -318,7 +329,6 @@ qh_completions (struct ehci_hcd *ehci, struct ehci_qh *qh)
 		if (last) {
 			if (likely (last->urb != urb)) {
 				ehci_urb_done(ehci, last->urb, last_status);
-				count++;
 				last_status = -EINPROGRESS;
 			}
 			ehci_qtd_free (ehci, last);
@@ -334,118 +344,199 @@ qh_completions (struct ehci_hcd *ehci, struct ehci_qh *qh)
 		token = hc32_to_cpu(ehci, qtd->hw_token);
 
 		/* always clean up qtds the hc de-activated */
+ retry_xacterr:
 		if ((token & QTD_STS_ACTIVE) == 0) {
 
+			/* Report Data Buffer Error: non-fatal but useful */
+			if (token & QTD_STS_DBE)
+				ehci_dbg(ehci,
+					"detected DataBufferErr for urb %p ep%d%s len %d, qtd %p [qh %p]\n",
+					urb,
+					usb_endpoint_num(&urb->ep->desc),
+					usb_endpoint_dir_in(&urb->ep->desc) ? "in" : "out",
+					urb->transfer_buffer_length,
+					qtd,
+					qh);
+
+			/* on STALL, error, and short reads this urb must
+			 * complete and all its qtds must be recycled.
+			 */
 			if ((token & QTD_STS_HALT) != 0) {
+
+				/* retry transaction errors until we
+				 * reach the software xacterr limit
+				 */
+				if ((token & QTD_STS_XACT) &&
+						QTD_CERR(token) == 0 &&
+						++qh->xacterrs < QH_XACTERR_MAX &&
+						!urb->unlinked) {
+					ehci_dbg(ehci,
+	"detected XactErr len %zu/%zu retry %d\n",
+	qtd->length - QTD_LENGTH(token), qtd->length, qh->xacterrs);
+
+					/* reset the token in the qtd and the
+					 * qh overlay (which still contains
+					 * the qtd) so that we pick up from
+					 * where we left off
+					 */
+					token &= ~QTD_STS_HALT;
+					token |= QTD_STS_ACTIVE |
+							(EHCI_TUNE_CERR << 10);
+					qtd->hw_token = cpu_to_hc32(ehci,
+							token);
+					wmb();
+					hw->hw_token = cpu_to_hc32(ehci,
+							token);
+					goto retry_xacterr;
+				}
 				stopped = 1;
 
 			/* magic dummy for some short reads; qh won't advance.
 			 * that silicon quirk can kick in with this dummy too.
+			 *
+			 * other short reads won't stop the queue, including
+			 * control transfers (status stage handles that) or
+			 * most other single-qtd reads ... the queue stops if
+			 * URB_SHORT_NOT_OK was set so the driver submitting
+			 * the urbs could clean it up.
 			 */
 			} else if (IS_SHORT_READ (token)
 					&& !(qtd->hw_alt_next
 						& EHCI_LIST_END(ehci))) {
 				stopped = 1;
-				goto halt;
 			}
 
 		/* stop scanning when we reach qtds the hc is using */
 		} else if (likely (!stopped
-				&& HC_IS_RUNNING (ehci_to_hcd(ehci)->state))) {
+				&& ehci->rh_state >= EHCI_RH_RUNNING)) {
 			break;
 
+		/* scan the whole queue for unlinks whenever it stops */
 		} else {
 			stopped = 1;
 
-			if (unlikely (!HC_IS_RUNNING (ehci_to_hcd(ehci)->state)))
+			/* cancel everything if we halt, suspend, etc */
+			if (ehci->rh_state < EHCI_RH_RUNNING)
 				last_status = -ESHUTDOWN;
 
-			/* ignore active urbs unless some previous qtd
-			 * for the urb faulted (including short read) or
-			 * its urb was canceled.  we may patch qh or qtds.
+			/* this qtd is active; skip it unless a previous qtd
+			 * for its urb faulted, or its urb was canceled.
 			 */
-			if (likely(last_status == -EINPROGRESS &&
-					!urb->unlinked))
+			else if (last_status == -EINPROGRESS && !urb->unlinked)
 				continue;
 
-			/* issue status after short control reads */
-			if (unlikely (do_status != 0)
-					&& QTD_PID (token) == 0 /* OUT */) {
-				do_status = 0;
-				continue;
+			/*
+			 * If this was the active qtd when the qh was unlinked
+			 * and the overlay's token is active, then the overlay
+			 * hasn't been written back to the qtd yet so use its
+			 * token instead of the qtd's.  After the qtd is
+			 * processed and removed, the overlay won't be valid
+			 * any more.
+			 */
+			if (state == QH_STATE_IDLE &&
+					qh->qtd_list.next == &qtd->qtd_list &&
+					(hw->hw_token & ACTIVE_BIT(ehci))) {
+				token = hc32_to_cpu(ehci, hw->hw_token);
+				hw->hw_token &= ~ACTIVE_BIT(ehci);
+
+				/* An unlink may leave an incomplete
+				 * async transaction in the TT buffer.
+				 * We have to clear it.
+				 */
+				ehci_clear_tt_buffer(ehci, qh, urb, token);
 			}
+		}
 
-			/* token in overlay may be most current */
-			if (state == QH_STATE_IDLE
-					&& cpu_to_hc32(ehci, qtd->qtd_dma)
-						== qh->hw_current)
-				token = hc32_to_cpu(ehci, qh->hw_token);
+		/* unless we already know the urb's status, collect qtd status
+		 * and update count of bytes transferred.  in common short read
+		 * cases with only one data qtd (including control transfers),
+		 * queue processing won't halt.  but with two or more qtds (for
+		 * example, with a 32 KB transfer), when the first qtd gets a
+		 * short read the second must be removed by hand.
+		 */
+		if (last_status == -EINPROGRESS) {
+			last_status = qtd_copy_status(ehci, urb,
+					qtd->length, token);
+			if (last_status == -EREMOTEIO
+					&& (qtd->hw_alt_next
+						& EHCI_LIST_END(ehci)))
+				last_status = -EINPROGRESS;
 
-			/* force halt for unlinked or blocked qh, so we'll
-			 * patch the qh later and so that completions can't
-			 * activate it while we "know" it's stopped.
+			/* As part of low/full-speed endpoint-halt processing
+			 * we must clear the TT buffer (11.17.5).
 			 */
-			if ((halt & qh->hw_token) == 0) {
-halt:
-				qh->hw_token |= halt;
-				wmb ();
+			if (unlikely(last_status != -EINPROGRESS &&
+					last_status != -EREMOTEIO)) {
+				/* The TT's in some hubs malfunction when they
+				 * receive this request following a STALL (they
+				 * stop sending isochronous packets).  Since a
+				 * STALL can't leave the TT buffer in a busy
+				 * state (if you believe Figures 11-48 - 11-51
+				 * in the USB 2.0 spec), we won't clear the TT
+				 * buffer in this case.  Strictly speaking this
+				 * is a violation of the spec.
+				 */
+				if (last_status != -EPIPE)
+					ehci_clear_tt_buffer(ehci, qh, urb,
+							token);
 			}
 		}
 
-		/* remove it from the queue */
-		qtd_status = qtd_copy_status(ehci, urb, qtd->length, token);
-		if (unlikely(qtd_status == -EREMOTEIO)) {
-			do_status = (!urb->unlinked &&
-					usb_pipecontrol(urb->pipe));
-			qtd_status = 0;
-		}
-		if (likely(last_status == -EINPROGRESS))
-			last_status = qtd_status;
-
+		/* if we're removing something not at the queue head,
+		 * patch the hardware queue pointer.
+		 */
 		if (stopped && qtd->qtd_list.prev != &qh->qtd_list) {
 			last = list_entry (qtd->qtd_list.prev,
 					struct ehci_qtd, qtd_list);
 			last->hw_next = qtd->hw_next;
 		}
+
+		/* remove qtd; it's recycled after possible urb completion */
 		list_del (&qtd->qtd_list);
 		last = qtd;
+
+		/* reinit the xacterr counter for the next qtd */
+		qh->xacterrs = 0;
 	}
 
 	/* last urb's completion might still need calling */
 	if (likely (last != NULL)) {
 		ehci_urb_done(ehci, last->urb, last_status);
-		count++;
 		ehci_qtd_free (ehci, last);
 	}
 
+	/* Do we need to rescan for URBs dequeued during a giveback? */
+	if (unlikely(qh->dequeue_during_giveback)) {
+		/* If the QH is already unlinked, do the rescan now. */
+		if (state == QH_STATE_IDLE)
+			goto rescan;
+
+		/* Otherwise the caller must unlink the QH. */
+	}
+
 	/* restore original state; caller must unlink or relink */
 	qh->qh_state = state;
 
 	/* be sure the hardware's done with the qh before refreshing
 	 * it after fault cleanup, or recovering from silicon wrongly
 	 * overlaying the dummy qtd (which reduces DMA chatter).
+	 *
+	 * We won't refresh a QH that's linked (after the HC
+	 * stopped the queue).  That avoids a race:
+	 *  - HC reads first part of QH;
+	 *  - CPU updates that first part and the token;
+	 *  - HC reads rest of that QH, including token
+	 * Result:  HC gets an inconsistent image, and then
+	 * DMAs to/from the wrong memory (corrupting it).
+	 *
+	 * That should be rare for interrupt transfers,
+	 * except maybe high bandwidth ...
 	 */
-	if (stopped != 0 || qh->hw_qtd_next == EHCI_LIST_END(ehci)) {
-		switch (state) {
-		case QH_STATE_IDLE:
-			qh_refresh(ehci, qh);
-			break;
-		case QH_STATE_LINKED:
-			/* should be rare for periodic transfers,
-			 * except maybe high bandwidth ...
-			 */
-			if ((cpu_to_hc32(ehci, QH_SMASK)
-					& qh->hw_info2) != 0) {
-				intr_deschedule (ehci, qh);
-				(void) qh_schedule (ehci, qh);
-			} else
-				unlink_async (ehci, qh);
-			break;
-		/* otherwise, unlink already started */
-		}
-	}
+	if (stopped != 0 || hw->hw_qtd_next == EHCI_LIST_END(ehci))
+		qh->exception = 1;
 
-	return count;
+	/* Let the caller know if the QH needs to be unlinked. */
+	return qh->exception;
 }
 
 /*-------------------------------------------------------------------------*/
@@ -487,9 +578,11 @@ qh_urb_transaction (
 ) {
 	struct ehci_qtd		*qtd, *qtd_prev;
 	dma_addr_t		buf;
-	int			len, maxpacket;
+	int			len, this_sg_len, maxpacket;
 	int			is_input;
 	u32			token;
+	int			i;
+	struct scatterlist	*sg;
 
 	/*
 	 * URBs map to sequences of QTDs:  one logical transaction
@@ -530,7 +623,20 @@ qh_urb_transaction (
 	/*
 	 * data transfer stage:  buffer setup
 	 */
-	buf = urb->transfer_dma;
+	i = urb->num_mapped_sgs;
+	if (len > 0 && i > 0) {
+		sg = urb->sg;
+		buf = sg_dma_address(sg);
+
+		/* urb->transfer_buffer_length may be smaller than the
+		 * size of the scatterlist (or vice versa)
+		 */
+		this_sg_len = min_t(int, sg_dma_len(sg), len);
+	} else {
+		sg = NULL;
+		buf = urb->transfer_dma;
+		this_sg_len = len;
+	}
 
 	if (is_input)
 		token |= (1 /* "in" */ << 8);
@@ -546,18 +652,31 @@ qh_urb_transaction (
 	for (;;) {
 		int this_qtd_len;
 
-		this_qtd_len = qtd_fill(ehci, qtd, buf, len, token, maxpacket);
+		this_qtd_len = qtd_fill(ehci, qtd, buf, this_sg_len, token,
+				maxpacket);
+		this_sg_len -= this_qtd_len;
 		len -= this_qtd_len;
 		buf += this_qtd_len;
+
+		/*
+		 * short reads advance to a "magic" dummy instead of the next
+		 * qtd ... that forces the queue to stop, for manual cleanup.
+		 * (this will usually be overridden later.)
+		 */
 		if (is_input)
-			qtd->hw_alt_next = ehci->async->hw_alt_next;
+			qtd->hw_alt_next = ehci->async->hw->hw_alt_next;
 
 		/* qh makes control packets use qtd toggle; maybe switch it */
 		if ((maxpacket & (this_qtd_len + (maxpacket - 1))) == 0)
 			token ^= QTD_TOGGLE;
 
-		if (likely (len <= 0))
-			break;
+		if (likely(this_sg_len <= 0)) {
+			if (--i <= 0 || len <= 0)
+				break;
+			sg = sg_next(sg);
+			buf = sg_dma_address(sg);
+			this_sg_len = min_t(int, sg_dma_len(sg), len);
+		}
 
 		qtd_prev = qtd;
 		qtd = ehci_qtd_alloc (ehci, flags);
@@ -568,8 +687,10 @@ qh_urb_transaction (
 		list_add_tail (&qtd->qtd_list, head);
 	}
 
-	/* unless the bulk/interrupt caller wants a chance to clean
-	 * up after short reads, hc should advance qh past this urb
+	/*
+	 * unless the caller requires manual cleanup after short reads,
+	 * have the alt_next mechanism keep the queue running after the
+	 * last data qtd (the only one, for control and most other cases).
 	 */
 	if (likely ((urb->transfer_flags & URB_SHORT_NOT_OK) == 0
 				|| usb_pipecontrol (urb->pipe)))
@@ -577,7 +698,8 @@ qh_urb_transaction (
 
 	/*
 	 * control requests may need a terminating data "status" ack;
-	 * bulk ones may need a terminating short packet (zero length).
+	 * other OUT ones may need a terminating short packet
+	 * (zero length).
 	 */
 	if (likely (urb->transfer_buffer_length != 0)) {
 		int	one_more = 0;
@@ -586,7 +708,7 @@ qh_urb_transaction (
 			one_more = 1;
 			token ^= 0x0100;	/* "in" <--> "out"  */
 			token |= QTD_TOGGLE;	/* force DATA1 */
-		} else if (usb_pipebulk (urb->pipe)
+		} else if (usb_pipeout(urb->pipe)
 				&& (urb->transfer_flags & URB_ZERO_PACKET)
 				&& !(urb->transfer_buffer_length % maxpacket)) {
 			one_more = 1;
@@ -643,6 +765,7 @@ qh_make (
 	int			is_input, type;
 	int			maxp = 0;
 	struct usb_tt		*tt = urb->dev->tt;
+	struct ehci_qh_hw	*hw;
 
 	if (!qh)
 		return qh;
@@ -657,6 +780,14 @@ qh_make (
 	type = usb_pipetype (urb->pipe);
 	maxp = usb_maxpacket (urb->dev, urb->pipe, !is_input);
 
+	/* 1024 byte maxpacket is a hardware ceiling.  High bandwidth
+	 * acts like up to 3KB, but is built from smaller packets.
+	 */
+	if (max_packet(maxp) > 1024) {
+		ehci_dbg(ehci, "bogus qh maxpacket %d\n", max_packet(maxp));
+		goto done;
+	}
+
 	/* Compute interrupt scheduling parameters just once, and save.
 	 * - allowing for high bandwidth, how many nsec/uframe are used?
 	 * - split transactions need a second CSPLIT uframe; same question
@@ -666,25 +797,35 @@ qh_make (
 	 * For control/bulk requests, the HC or TT handles these.
 	 */
 	if (type == PIPE_INTERRUPT) {
-		qh->usecs = NS_TO_US(usb_calc_bus_time(USB_SPEED_HIGH,
+		unsigned	tmp;
+
+		qh->ps.usecs = NS_TO_US(usb_calc_bus_time(USB_SPEED_HIGH,
 				is_input, 0,
 				hb_mult(maxp) * max_packet(maxp)));
-		qh->start = NO_FRAME;
+		qh->ps.phase = NO_FRAME;
 
 		if (urb->dev->speed == USB_SPEED_HIGH) {
-			qh->c_usecs = 0;
+			qh->ps.c_usecs = 0;
 			qh->gap_uf = 0;
 
-			qh->period = urb->interval >> 3;
-			if (qh->period == 0 && urb->interval != 1) {
+			if (urb->interval > 1 && urb->interval < 8) {
 				/* NOTE interval 2 or 4 uframes could work.
 				 * But interval 1 scheduling is simpler, and
 				 * includes high bandwidth.
 				 */
-				dbg ("intr period %d uframes, NYET!",
-						urb->interval);
-				goto done;
+				urb->interval = 1;
+			} else if (urb->interval > ehci->periodic_size << 3) {
+				urb->interval = ehci->periodic_size << 3;
 			}
+			qh->ps.period = urb->interval >> 3;
+
+			/* period for bandwidth allocation */
+			tmp = min_t(unsigned, EHCI_BANDWIDTH_SIZE,
+					1 << (urb->ep->desc.bInterval - 1));
+
+			/* Allow urb->interval to override */
+			qh->ps.bw_uperiod = min_t(unsigned, tmp, urb->interval);
+			qh->ps.bw_period = qh->ps.bw_uperiod >> 3;
 		} else {
 			int		think_time;
 
@@ -694,28 +835,40 @@ qh_make (
 
 			/* FIXME this just approximates SPLIT/CSPLIT times */
 			if (is_input) {		// SPLIT, gap, CSPLIT+DATA
-				qh->c_usecs = qh->usecs + HS_USECS (0);
-				qh->usecs = HS_USECS (1);
+				qh->ps.c_usecs = qh->ps.usecs + HS_USECS(0);
+				qh->ps.usecs = HS_USECS(1);
 			} else {		// SPLIT+DATA, gap, CSPLIT
-				qh->usecs += HS_USECS (1);
-				qh->c_usecs = HS_USECS (0);
+				qh->ps.usecs += HS_USECS(1);
+				qh->ps.c_usecs = HS_USECS(0);
 			}
 
 			think_time = tt ? tt->think_time : 0;
-			qh->tt_usecs = NS_TO_US (think_time +
+			qh->ps.tt_usecs = NS_TO_US(think_time +
 					usb_calc_bus_time (urb->dev->speed,
 					is_input, 0, max_packet (maxp)));
-			qh->period = urb->interval;
+			if (urb->interval > ehci->periodic_size)
+				urb->interval = ehci->periodic_size;
+			qh->ps.period = urb->interval;
+
+			/* period for bandwidth allocation */
+			tmp = min_t(unsigned, EHCI_BANDWIDTH_FRAMES,
+					urb->ep->desc.bInterval);
+			tmp = rounddown_pow_of_two(tmp);
+
+			/* Allow urb->interval to override */
+			qh->ps.bw_period = min_t(unsigned, tmp, urb->interval);
+			qh->ps.bw_uperiod = qh->ps.bw_period << 3;
 		}
 	}
 
 	/* support for tt scheduling, and access to toggles */
-	qh->dev = urb->dev;
+	qh->ps.udev = urb->dev;
+	qh->ps.ep = urb->ep;
 
 	/* using TT? */
 	switch (urb->dev->speed) {
 	case USB_SPEED_LOW:
-		info1 |= (1 << 12);	/* EPS "low" */
+		info1 |= QH_LOW_SPEED;
 		/* FALL THROUGH */
 
 	case USB_SPEED_FULL:
@@ -723,8 +876,8 @@ qh_make (
 		if (type != PIPE_INTERRUPT)
 			info1 |= (EHCI_TUNE_RL_TT << 28);
 		if (type == PIPE_CONTROL) {
-			info1 |= (1 << 27);	/* for TT */
-			info1 |= 1 << 14;	/* toggle from qtd */
+			info1 |= QH_CONTROL_EP;		/* for TT */
+			info1 |= QH_TOGGLE_CTL;		/* toggle from qtd */
 		}
 		info1 |= maxp << 16;
 
@@ -749,15 +902,21 @@ qh_make (
 		break;
 
 	case USB_SPEED_HIGH:		/* no TT involved */
-		info1 |= (2 << 12);	/* EPS "high" */
+		info1 |= QH_HIGH_SPEED;
 		if (type == PIPE_CONTROL) {
 			info1 |= (EHCI_TUNE_RL_HS << 28);
 			info1 |= 64 << 16;	/* usb2 fixed maxpacket */
-			info1 |= 1 << 14;	/* toggle from qtd */
+			info1 |= QH_TOGGLE_CTL;	/* toggle from qtd */
 			info2 |= (EHCI_TUNE_MULT_HS << 30);
 		} else if (type == PIPE_BULK) {
 			info1 |= (EHCI_TUNE_RL_HS << 28);
-			info1 |= 512 << 16;	/* usb2 fixed maxpacket */
+			/* The USB spec says that high speed bulk endpoints
+			 * always use 512 byte maxpacket.  But some device
+			 * vendors decided to ignore that, and MSFT is happy
+			 * to help them do so.  So now people expect to use
+			 * such nonconformant devices with Linux too; sigh.
+			 */
+			info1 |= max_packet(maxp) << 16;
 			info2 |= (EHCI_TUNE_MULT_HS << 30);
 		} else {		/* PIPE_INTERRUPT */
 			info1 |= max_packet (maxp) << 16;
@@ -765,25 +924,53 @@ qh_make (
 		}
 		break;
 	default:
-		dbg ("bogus dev %p speed %d", urb->dev, urb->dev->speed);
+		ehci_dbg(ehci, "bogus dev %p speed %d\n", urb->dev,
+			urb->dev->speed);
 done:
-		qh_put (qh);
+		qh_destroy(ehci, qh);
 		return NULL;
 	}
 
 	/* NOTE:  if (PIPE_INTERRUPT) { scheduler sets s-mask } */
 
-	/* init as live, toggle clear, advance to dummy */
+	/* init as live, toggle clear */
 	qh->qh_state = QH_STATE_IDLE;
-	qh->hw_info1 = cpu_to_hc32(ehci, info1);
-	qh->hw_info2 = cpu_to_hc32(ehci, info2);
+	hw = qh->hw;
+	hw->hw_info1 = cpu_to_hc32(ehci, info1);
+	hw->hw_info2 = cpu_to_hc32(ehci, info2);
+	qh->is_out = !is_input;
 	usb_settoggle (urb->dev, usb_pipeendpoint (urb->pipe), !is_input, 1);
-	qh_refresh (ehci, qh);
 	return qh;
 }
 
 /*-------------------------------------------------------------------------*/
 
+static void enable_async(struct ehci_hcd *ehci)
+{
+	if (ehci->async_count++)
+		return;
+
+	/* Stop waiting to turn off the async schedule */
+	ehci->enabled_hrtimer_events &= ~BIT(EHCI_HRTIMER_DISABLE_ASYNC);
+
+	/* Don't start the schedule until ASS is 0 */
+	ehci_poll_ASS(ehci);
+	turn_on_io_watchdog(ehci);
+}
+
+static void disable_async(struct ehci_hcd *ehci)
+{
+	if (--ehci->async_count)
+		return;
+
+	/* The async schedule and unlink lists are supposed to be empty */
+	WARN_ON(ehci->async->qh_next.qh || !list_empty(&ehci->async_unlink) ||
+			!list_empty(&ehci->async_idle));
+
+	/* Don't turn off the schedule until ASS is 1 */
+	ehci_poll_ASS(ehci);
+}
+
 /* move qh (and its qtds) onto async queue; maybe enable queue.  */
 
 static void qh_link_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
@@ -791,37 +978,30 @@ static void qh_link_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
 	__hc32		dma = QH_NEXT(ehci, qh->qh_dma);
 	struct ehci_qh	*head;
 
-	/* (re)start the async schedule? */
-	head = ehci->async;
-	timer_action_done (ehci, TIMER_ASYNC_OFF);
-	if (!head->qh_next.qh) {
-		u32	cmd = ehci_readl(ehci, &ehci->regs->command);
-
-		if (!(cmd & CMD_ASE)) {
-			/* in case a clear of CMD_ASE didn't take yet */
-			(void)handshake(ehci, &ehci->regs->status,
-					STS_ASS, 0, 150);
-			cmd |= CMD_ASE | CMD_RUN;
-			ehci_writel(ehci, cmd, &ehci->regs->command);
-			ehci_to_hcd(ehci)->state = HC_STATE_RUNNING;
-			/* posted write need not be known to HC yet ... */
-		}
-	}
+	/* Don't link a QH if there's a Clear-TT-Buffer pending */
+	if (unlikely(qh->clearing_tt))
+		return;
+
+	WARN_ON(qh->qh_state != QH_STATE_IDLE);
 
 	/* clear halt and/or toggle; and maybe recover from silicon quirk */
-	if (qh->qh_state == QH_STATE_IDLE)
-		qh_refresh (ehci, qh);
+	qh_refresh(ehci, qh);
 
 	/* splice right after start */
+	head = ehci->async;
 	qh->qh_next = head->qh_next;
-	qh->hw_next = head->hw_next;
+	qh->hw->hw_next = head->hw->hw_next;
 	wmb ();
 
 	head->qh_next.qh = qh;
-	head->hw_next = dma;
+	head->hw->hw_next = dma;
 
 	qh->qh_state = QH_STATE_LINKED;
+	qh->xacterrs = 0;
+	qh->exception = 0;
 	/* qtd completions reported later by interrupt */
+
+	enable_async(ehci);
 }
 
 /*-------------------------------------------------------------------------*/
@@ -841,7 +1021,7 @@ static struct ehci_qh *qh_append_tds (
 )
 {
 	struct ehci_qh		*qh = NULL;
-	u32			qh_addr_mask = cpu_to_hc32(ehci, 0x7f);
+	__hc32			qh_addr_mask = cpu_to_hc32(ehci, 0x7f);
 
 	qh = (struct ehci_qh *) *ptr;
 	if (unlikely (qh == NULL)) {
@@ -863,7 +1043,7 @@ static struct ehci_qh *qh_append_tds (
 
                         /* usb_reset_device() briefly reverts to address 0 */
                         if (usb_pipedevice (urb->pipe) == 0)
-                                qh->hw_info1 &= ~qh_addr_mask;
+				qh->hw->hw_info1 &= ~qh_addr_mask;
 		}
 
 		/* just one way to queue requests: swap with the dummy qtd.
@@ -881,7 +1061,7 @@ static struct ehci_qh *qh_append_tds (
 			 */
 			token = qtd->hw_token;
 			qtd->hw_token = HALT_BIT(ehci);
-			wmb ();
+
 			dummy = qh->dummy;
 
 			dma = dummy->qtd_dma;
@@ -890,7 +1070,7 @@ static struct ehci_qh *qh_append_tds (
 
 			list_del (&qtd->qtd_list);
 			list_add (&dummy->qtd_list, qtd_list);
-			__list_splice (qtd_list, qh->qtd_list.prev);
+			list_splice_tail(qtd_list, &qh->qtd_list);
 
 			ehci_qtd_init(ehci, qtd, qtd->qtd_dma);
 			qh->dummy = qtd;
@@ -905,7 +1085,7 @@ static struct ehci_qh *qh_append_tds (
 			wmb ();
 			dummy->hw_token = token;
 
-			urb->hcpriv = qh_get (qh);
+			urb->hcpriv = qh;
 		}
 	}
 	return qh;
@@ -920,27 +1100,28 @@ submit_async (
 	struct list_head	*qtd_list,
 	gfp_t			mem_flags
 ) {
-	struct ehci_qtd		*qtd;
 	int			epnum;
 	unsigned long		flags;
 	struct ehci_qh		*qh = NULL;
 	int			rc;
 
-	qtd = list_entry (qtd_list->next, struct ehci_qtd, qtd_list);
 	epnum = urb->ep->desc.bEndpointAddress;
 
 #ifdef EHCI_URB_TRACE
-	ehci_dbg (ehci,
-		"%s %s urb %p ep%d%s len %d, qtd %p [qh %p]\n",
-		__FUNCTION__, urb->dev->devpath, urb,
-		epnum & 0x0f, (epnum & USB_DIR_IN) ? "in" : "out",
-		urb->transfer_buffer_length,
-		qtd, urb->ep->hcpriv);
+	{
+		struct ehci_qtd *qtd;
+		qtd = list_entry(qtd_list->next, struct ehci_qtd, qtd_list);
+		ehci_dbg(ehci,
+			 "%s %s urb %p ep%d%s len %d, qtd %p [qh %p]\n",
+			 __func__, urb->dev->devpath, urb,
+			 epnum & 0x0f, (epnum & USB_DIR_IN) ? "in" : "out",
+			 urb->transfer_buffer_length,
+			 qtd, urb->ep->hcpriv);
+	}
 #endif
 
 	spin_lock_irqsave (&ehci->lock, flags);
-	if (unlikely(!test_bit(HCD_FLAG_HW_ACCESSIBLE,
-			       &ehci_to_hcd(ehci)->flags))) {
+	if (unlikely(!HCD_HW_ACCESSIBLE(ehci_to_hcd(ehci)))) {
 		rc = -ESHUTDOWN;
 		goto done;
 	}
@@ -959,7 +1140,7 @@ submit_async (
 	 * the HC and TT handle it when the TT has a buffer ready.
 	 */
 	if (likely (qh->qh_state == QH_STATE_IDLE))
-		qh_link_async (ehci, qh_get (qh));
+		qh_link_async(ehci, qh);
  done:
 	spin_unlock_irqrestore (&ehci->lock, flags);
 	if (unlikely (qh == NULL))
@@ -968,103 +1149,282 @@ submit_async (
 }
 
 /*-------------------------------------------------------------------------*/
+#ifdef CONFIG_USB_HCD_TEST_MODE
+/*
+ * This function creates the qtds and submits them for the
+ * SINGLE_STEP_SET_FEATURE Test.
+ * This is done in two parts: first SETUP req for GetDesc is sent then
+ * 15 seconds later, the IN stage for GetDesc starts to req data from dev
+ *
+ * is_setup : i/p arguement decides which of the two stage needs to be
+ * performed; TRUE - SETUP and FALSE - IN+STATUS
+ * Returns 0 if success
+ */
+static int submit_single_step_set_feature(
+	struct usb_hcd  *hcd,
+	struct urb      *urb,
+	int             is_setup
+) {
+	struct ehci_hcd		*ehci = hcd_to_ehci(hcd);
+	struct list_head	qtd_list;
+	struct list_head	*head;
+
+	struct ehci_qtd		*qtd, *qtd_prev;
+	dma_addr_t		buf;
+	int			len, maxpacket;
+	u32			token;
+
+	INIT_LIST_HEAD(&qtd_list);
+	head = &qtd_list;
+
+	/* URBs map to sequences of QTDs:  one logical transaction */
+	qtd = ehci_qtd_alloc(ehci, GFP_KERNEL);
+	if (unlikely(!qtd))
+		return -1;
+	list_add_tail(&qtd->qtd_list, head);
+	qtd->urb = urb;
+
+	token = QTD_STS_ACTIVE;
+	token |= (EHCI_TUNE_CERR << 10);
+
+	len = urb->transfer_buffer_length;
+	/*
+	 * Check if the request is to perform just the SETUP stage (getDesc)
+	 * as in SINGLE_STEP_SET_FEATURE test, DATA stage (IN) happens
+	 * 15 secs after the setup
+	 */
+	if (is_setup) {
+		/* SETUP pid */
+		qtd_fill(ehci, qtd, urb->setup_dma,
+				sizeof(struct usb_ctrlrequest),
+				token | (2 /* "setup" */ << 8), 8);
+
+		submit_async(ehci, urb, &qtd_list, GFP_ATOMIC);
+		return 0; /*Return now; we shall come back after 15 seconds*/
+	}
+
+	/*
+	 * IN: data transfer stage:  buffer setup : start the IN txn phase for
+	 * the get_Desc SETUP which was sent 15seconds back
+	 */
+	token ^= QTD_TOGGLE;   /*We need to start IN with DATA-1 Pid-sequence*/
+	buf = urb->transfer_dma;
+
+	token |= (1 /* "in" */ << 8);  /*This is IN stage*/
+
+	maxpacket = max_packet(usb_maxpacket(urb->dev, urb->pipe, 0));
+
+	qtd_fill(ehci, qtd, buf, len, token, maxpacket);
+
+	/*
+	 * Our IN phase shall always be a short read; so keep the queue running
+	 * and let it advance to the next qtd which zero length OUT status
+	 */
+	qtd->hw_alt_next = EHCI_LIST_END(ehci);
+
+	/* STATUS stage for GetDesc control request */
+	token ^= 0x0100;        /* "in" <--> "out"  */
+	token |= QTD_TOGGLE;    /* force DATA1 */
+
+	qtd_prev = qtd;
+	qtd = ehci_qtd_alloc(ehci, GFP_ATOMIC);
+	if (unlikely(!qtd))
+		goto cleanup;
+	qtd->urb = urb;
+	qtd_prev->hw_next = QTD_NEXT(ehci, qtd->qtd_dma);
+	list_add_tail(&qtd->qtd_list, head);
+
+	/* dont fill any data in such packets */
+	qtd_fill(ehci, qtd, 0, 0, token, 0);
+
+	/* by default, enable interrupt on urb completion */
+	if (likely(!(urb->transfer_flags & URB_NO_INTERRUPT)))
+		qtd->hw_token |= cpu_to_hc32(ehci, QTD_IOC);
+
+	submit_async(ehci, urb, &qtd_list, GFP_KERNEL);
+
+	return 0;
+
+cleanup:
+	qtd_list_free(ehci, urb, head);
+	return -1;
+}
+#endif /* CONFIG_USB_HCD_TEST_MODE */
+
+/*-------------------------------------------------------------------------*/
+
+static void single_unlink_async(struct ehci_hcd *ehci, struct ehci_qh *qh)
+{
+	struct ehci_qh		*prev;
+
+	/* Add to the end of the list of QHs waiting for the next IAAD */
+	qh->qh_state = QH_STATE_UNLINK_WAIT;
+	list_add_tail(&qh->unlink_node, &ehci->async_unlink);
+
+	/* Unlink it from the schedule */
+	prev = ehci->async;
+	while (prev->qh_next.qh != qh)
+		prev = prev->qh_next.qh;
 
-/* the async qh for the qtds being reclaimed are now unlinked from the HC */
+	prev->hw->hw_next = qh->hw->hw_next;
+	prev->qh_next = qh->qh_next;
+	if (ehci->qh_scan_next == qh)
+		ehci->qh_scan_next = qh->qh_next.qh;
+}
 
-static void end_unlink_async (struct ehci_hcd *ehci)
+static void start_iaa_cycle(struct ehci_hcd *ehci)
 {
-	struct ehci_qh		*qh = ehci->reclaim;
-	struct ehci_qh		*next;
+	/* Do nothing if an IAA cycle is already running */
+	if (ehci->iaa_in_progress)
+		return;
+	ehci->iaa_in_progress = true;
 
-	iaa_watchdog_done(ehci);
+	/* If the controller isn't running, we don't have to wait for it */
+	if (unlikely(ehci->rh_state < EHCI_RH_RUNNING)) {
+		end_unlink_async(ehci);
 
-	// qh->hw_next = cpu_to_hc32(qh->qh_dma);
-	qh->qh_state = QH_STATE_IDLE;
-	qh->qh_next.qh = NULL;
-	qh_put (qh);			// refcount from reclaim
+	/* Otherwise start a new IAA cycle */
+	} else if (likely(ehci->rh_state == EHCI_RH_RUNNING)) {
 
-	/* other unlink(s) may be pending (in QH_STATE_UNLINK_WAIT) */
-	next = qh->reclaim;
-	ehci->reclaim = next;
-	qh->reclaim = NULL;
+		/* Make sure the unlinks are all visible to the hardware */
+		wmb();
 
-	qh_completions (ehci, qh);
+		ehci_writel(ehci, ehci->command | CMD_IAAD,
+				&ehci->regs->command);
+		ehci_readl(ehci, &ehci->regs->command);
+		ehci_enable_event(ehci, EHCI_HRTIMER_IAA_WATCHDOG, true);
+	}
+}
 
-	if (!list_empty (&qh->qtd_list)
-			&& HC_IS_RUNNING (ehci_to_hcd(ehci)->state))
-		qh_link_async (ehci, qh);
-	else {
-		qh_put (qh);		// refcount from async list
+/* the async qh for the qtds being unlinked are now gone from the HC */
 
-		/* it's not free to turn the async schedule on/off; leave it
-		 * active but idle for a while once it empties.
-		 */
-		if (HC_IS_RUNNING (ehci_to_hcd(ehci)->state)
-				&& ehci->async->qh_next.qh == NULL)
-			timer_action (ehci, TIMER_ASYNC_OFF);
+static void end_unlink_async(struct ehci_hcd *ehci)
+{
+	struct ehci_qh		*qh;
+	bool			early_exit;
+
+	if (ehci->has_synopsys_hc_bug)
+		ehci_writel(ehci, (u32) ehci->async->qh_dma,
+			    &ehci->regs->async_next);
+
+	/* The current IAA cycle has ended */
+	ehci->iaa_in_progress = false;
+
+	if (list_empty(&ehci->async_unlink))
+		return;
+	qh = list_first_entry(&ehci->async_unlink, struct ehci_qh,
+			unlink_node);	/* QH whose IAA cycle just ended */
+
+	/*
+	 * If async_unlinking is set then this routine is already running,
+	 * either on the stack or on another CPU.
+	 */
+	early_exit = ehci->async_unlinking;
+
+	/* If the controller isn't running, process all the waiting QHs */
+	if (ehci->rh_state < EHCI_RH_RUNNING)
+		list_splice_tail_init(&ehci->async_unlink, &ehci->async_idle);
+
+	/*
+	 * Intel (?) bug: The HC can write back the overlay region even
+	 * after the IAA interrupt occurs.  In self-defense, always go
+	 * through two IAA cycles for each QH.
+	 */
+	else if (qh->qh_state == QH_STATE_UNLINK_WAIT) {
+		qh->qh_state = QH_STATE_UNLINK;
+		early_exit = true;
 	}
 
-	if (next) {
-		ehci->reclaim = NULL;
-		start_unlink_async (ehci, next);
+	/* Otherwise process only the first waiting QH (NVIDIA bug?) */
+	else
+		list_move_tail(&qh->unlink_node, &ehci->async_idle);
+
+	/* Start a new IAA cycle if any QHs are waiting for it */
+	if (!list_empty(&ehci->async_unlink))
+		start_iaa_cycle(ehci);
+
+	/*
+	 * Don't allow nesting or concurrent calls,
+	 * or wait for the second IAA cycle for the next QH.
+	 */
+	if (early_exit)
+		return;
+
+	/* Process the idle QHs */
+	ehci->async_unlinking = true;
+	while (!list_empty(&ehci->async_idle)) {
+		qh = list_first_entry(&ehci->async_idle, struct ehci_qh,
+				unlink_node);
+		list_del(&qh->unlink_node);
+
+		qh->qh_state = QH_STATE_IDLE;
+		qh->qh_next.qh = NULL;
+
+		if (!list_empty(&qh->qtd_list))
+			qh_completions(ehci, qh);
+		if (!list_empty(&qh->qtd_list) &&
+				ehci->rh_state == EHCI_RH_RUNNING)
+			qh_link_async(ehci, qh);
+		disable_async(ehci);
 	}
+	ehci->async_unlinking = false;
 }
 
-/* makes sure the async qh will become idle */
-/* caller must own ehci->lock */
+static void start_unlink_async(struct ehci_hcd *ehci, struct ehci_qh *qh);
 
-static void start_unlink_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
+static void unlink_empty_async(struct ehci_hcd *ehci)
 {
-	int		cmd = ehci_readl(ehci, &ehci->regs->command);
-	struct ehci_qh	*prev;
-
-#ifdef DEBUG
-	assert_spin_locked(&ehci->lock);
-	if (ehci->reclaim
-			|| (qh->qh_state != QH_STATE_LINKED
-				&& qh->qh_state != QH_STATE_UNLINK_WAIT)
-			)
-		BUG ();
-#endif
-
-	/* stop async schedule right now? */
-	if (unlikely (qh == ehci->async)) {
-		/* can't get here without STS_ASS set */
-		if (ehci_to_hcd(ehci)->state != HC_STATE_HALT
-				&& !ehci->reclaim) {
-			/* ... and CMD_IAAD clear */
-			ehci_writel(ehci, cmd & ~CMD_ASE,
-				    &ehci->regs->command);
-			wmb ();
-			// handshake later, if we need to
-			timer_action_done (ehci, TIMER_ASYNC_OFF);
+	struct ehci_qh		*qh;
+	struct ehci_qh		*qh_to_unlink = NULL;
+	int			count = 0;
+
+	/* Find the last async QH which has been empty for a timer cycle */
+	for (qh = ehci->async->qh_next.qh; qh; qh = qh->qh_next.qh) {
+		if (list_empty(&qh->qtd_list) &&
+				qh->qh_state == QH_STATE_LINKED) {
+			++count;
+			if (qh->unlink_cycle != ehci->async_unlink_cycle)
+				qh_to_unlink = qh;
 		}
-		return;
 	}
 
-	qh->qh_state = QH_STATE_UNLINK;
-	ehci->reclaim = qh = qh_get (qh);
+	/* If nothing else is being unlinked, unlink the last empty QH */
+	if (list_empty(&ehci->async_unlink) && qh_to_unlink) {
+		start_unlink_async(ehci, qh_to_unlink);
+		--count;
+	}
 
-	prev = ehci->async;
-	while (prev->qh_next.qh != qh)
-		prev = prev->qh_next.qh;
+	/* Other QHs will be handled later */
+	if (count > 0) {
+		ehci_enable_event(ehci, EHCI_HRTIMER_ASYNC_UNLINKS, true);
+		++ehci->async_unlink_cycle;
+	}
+}
 
-	prev->hw_next = qh->hw_next;
-	prev->qh_next = qh->qh_next;
-	wmb ();
+/* The root hub is suspended; unlink all the async QHs */
+static void __maybe_unused unlink_empty_async_suspended(struct ehci_hcd *ehci)
+{
+	struct ehci_qh		*qh;
 
-	if (unlikely (ehci_to_hcd(ehci)->state == HC_STATE_HALT)) {
-		/* if (unlikely (qh->reclaim != 0))
-		 *	this will recurse, probably not much
-		 */
-		end_unlink_async (ehci);
-		return;
+	while (ehci->async->qh_next.qh) {
+		qh = ehci->async->qh_next.qh;
+		WARN_ON(!list_empty(&qh->qtd_list));
+		single_unlink_async(ehci, qh);
 	}
+	start_iaa_cycle(ehci);
+}
+
+/* makes sure the async qh will become idle */
+/* caller must own ehci->lock */
+
+static void start_unlink_async(struct ehci_hcd *ehci, struct ehci_qh *qh)
+{
+	/* If the QH isn't linked then there's nothing we can do. */
+	if (qh->qh_state != QH_STATE_LINKED)
+		return;
 
-	cmd |= CMD_IAAD;
-	ehci_writel(ehci, cmd, &ehci->regs->command);
-	(void)ehci_readl(ehci, &ehci->regs->command);
-	iaa_watchdog_start(ehci);
+	single_unlink_async(ehci, qh);
+	start_iaa_cycle(ehci);
 }
 
 /*-------------------------------------------------------------------------*/
@@ -1072,51 +1432,45 @@ static void start_unlink_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
 static void scan_async (struct ehci_hcd *ehci)
 {
 	struct ehci_qh		*qh;
-	enum ehci_timer_action	action = TIMER_IO_WATCHDOG;
-
-	if (!++(ehci->stamp))
-		ehci->stamp++;
-	timer_action_done (ehci, TIMER_ASYNC_SHRINK);
-rescan:
-	qh = ehci->async->qh_next.qh;
-	if (likely (qh != NULL)) {
-		do {
-			/* clean any finished work for this qh */
-			if (!list_empty (&qh->qtd_list)
-					&& qh->stamp != ehci->stamp) {
-				int temp;
-
-				/* unlinks could happen here; completion
-				 * reporting drops the lock.  rescan using
-				 * the latest schedule, but don't rescan
-				 * qhs we already finished (no looping).
-				 */
-				qh = qh_get (qh);
-				qh->stamp = ehci->stamp;
-				temp = qh_completions (ehci, qh);
-				qh_put (qh);
-				if (temp != 0) {
-					goto rescan;
-				}
-			}
-
-			/* unlink idle entries, reducing HC PCI usage as well
-			 * as HCD schedule-scanning costs.  delay for any qh
-			 * we just scanned, there's a not-unusual case that it
-			 * doesn't stay idle for long.
-			 * (plus, avoids some kind of re-activation race.)
+	bool			check_unlinks_later = false;
+
+	ehci->qh_scan_next = ehci->async->qh_next.qh;
+	while (ehci->qh_scan_next) {
+		qh = ehci->qh_scan_next;
+		ehci->qh_scan_next = qh->qh_next.qh;
+
+		/* clean any finished work for this qh */
+		if (!list_empty(&qh->qtd_list)) {
+			int temp;
+
+			/*
+			 * Unlinks could happen here; completion reporting
+			 * drops the lock.  That's why ehci->qh_scan_next
+			 * always holds the next qh to scan; if the next qh
+			 * gets unlinked then ehci->qh_scan_next is adjusted
+			 * in single_unlink_async().
 			 */
-			if (list_empty (&qh->qtd_list)) {
-				if (qh->stamp == ehci->stamp)
-					action = TIMER_ASYNC_SHRINK;
-				else if (!ehci->reclaim
-					    && qh->qh_state == QH_STATE_LINKED)
-					start_unlink_async (ehci, qh);
+			temp = qh_completions(ehci, qh);
+			if (unlikely(temp)) {
+				start_unlink_async(ehci, qh);
+			} else if (list_empty(&qh->qtd_list)
+					&& qh->qh_state == QH_STATE_LINKED) {
+				qh->unlink_cycle = ehci->async_unlink_cycle;
+				check_unlinks_later = true;
 			}
+		}
+	}
 
-			qh = qh->qh_next.qh;
-		} while (qh);
+	/*
+	 * Unlink empty entries, reducing DMA usage as well
+	 * as HCD schedule-scanning costs.  Delay for any qh
+	 * we just scanned, there's a not-unusual case that it
+	 * doesn't stay idle for long.
+	 */
+	if (check_unlinks_later && ehci->rh_state == EHCI_RH_RUNNING &&
+			!(ehci->enabled_hrtimer_events &
+				BIT(EHCI_HRTIMER_ASYNC_UNLINKS))) {
+		ehci_enable_event(ehci, EHCI_HRTIMER_ASYNC_UNLINKS, true);
+		++ehci->async_unlink_cycle;
 	}
-	if (action == TIMER_ASYNC_SHRINK)
-		timer_action (ehci, TIMER_ASYNC_SHRINK);
 }