USB: EHCI: unlink multiple async QHs together

This patch (as1582) changes ehci-hcd's strategy for unlinking async
QHs.  Currently the driver never unlinks more than one QH at a time.
This can be inefficient and cause unnecessary delays, since a QH
cannot be reused while it is waiting to be unlinked.

The new strategy unlinks all the waiting QHs at once.  In practice the
improvement won't be very big, because it's somewhat uncommon to have
two or more QHs waiting to be unlinked at any time.  But it does
happen, and in any case, doing things this way makes more sense IMO.

The change requires the async unlinking code to be refactored
slightly.  Now in addition to the routines for starting and ending an
unlink, there are new routines for unlinking a single QH and starting
an IAA cycle.  This approach is needed because there are two separate
paths for unlinking async QHs:

	When a transfer error occurs or an URB is cancelled, the QH
	must be unlinked right away;

	When a QH has been idle sufficiently long, it is unlinked
	to avoid consuming DMA bandwidth uselessly.

In the first case we want the unlink to proceed as quickly as
possible, whereas in the second case we can afford to batch several
QHs together and unlink them all at once.  Hence the division of
labor.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

1 files changed, 89 insertions, 60 deletions

diff --git a/drivers/usb/host/ehci-q.c b/drivers/usb/host/ehci-q.c
index 7b35933d91a..181832921c5 100644
--- a/drivers/usb/host/ehci-q.c
+++ b/drivers/usb/host/ehci-q.c
@@ -293,9 +293,6 @@ __acquires(ehci->lock)
 	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 int qh_schedule (struct ehci_hcd *ehci, struct ehci_qh *qh);
 
 /*
@@ -1166,82 +1163,114 @@ submit_async (
 
 /*-------------------------------------------------------------------------*/
 
-/* the async qh for the qtds being unlinked are now gone from the HC */
-
-static void end_unlink_async (struct ehci_hcd *ehci)
+static void single_unlink_async(struct ehci_hcd *ehci, struct ehci_qh *qh)
 {
-	struct ehci_qh		*qh = ehci->async_unlink;
-	struct ehci_qh		*next;
+	struct ehci_qh		*prev;
 
-	// qh->hw_next = cpu_to_hc32(qh->qh_dma);
-	qh->qh_state = QH_STATE_IDLE;
-	qh->qh_next.qh = NULL;
+	/* Add to the end of the list of QHs waiting for the next IAAD */
+	qh->qh_state = QH_STATE_UNLINK;
+	if (ehci->async_unlink)
+		ehci->async_unlink_last->unlink_next = qh;
+	else
+		ehci->async_unlink = qh;
+	ehci->async_unlink_last = qh;
+
+	/* Unlink it from the schedule */
+	prev = ehci->async;
+	while (prev->qh_next.qh != qh)
+		prev = prev->qh_next.qh;
+
+	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;
+}
 
-	/* other unlink(s) may be pending (in QH_STATE_UNLINK_WAIT) */
-	next = qh->unlink_next;
-	ehci->async_unlink = next;
-	qh->unlink_next = NULL;
+static void start_iaa_cycle(struct ehci_hcd *ehci, bool nested)
+{
+	/*
+	 * Do nothing if an IAA cycle is already running or
+	 * if one will be started shortly.
+	 */
+	if (ehci->async_iaa || ehci->async_unlinking)
+		return;
 
-	qh_completions (ehci, qh);
+	/* Do all the waiting QHs at once */
+	ehci->async_iaa = ehci->async_unlink;
+	ehci->async_unlink = NULL;
 
-	if (!list_empty(&qh->qtd_list) && ehci->rh_state == EHCI_RH_RUNNING)
-		qh_link_async (ehci, qh);
+	/* If the controller isn't running, we don't have to wait for it */
+	if (unlikely(ehci->rh_state < EHCI_RH_RUNNING)) {
+		if (!nested)		/* Avoid recursion */
+			end_unlink_async(ehci);
 
-	disable_async(ehci);
+	/* Otherwise start a new IAA cycle */
+	} else {
+		/* Make sure the unlinks are all visible to the hardware */
+		wmb();
 
-	if (next) {
-		ehci->async_unlink = NULL;
-		start_unlink_async (ehci, next);
+		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);
 	}
+}
+
+/* the async qh for the qtds being unlinked are now gone from the HC */
+
+static void end_unlink_async(struct ehci_hcd *ehci)
+{
+	struct ehci_qh		*qh;
 
 	if (ehci->has_synopsys_hc_bug)
 		ehci_writel(ehci, (u32) ehci->async->qh_dma,
 			    &ehci->regs->async_next);
+
+	/* Process the idle QHs */
+ restart:
+	ehci->async_unlinking = true;
+	while (ehci->async_iaa) {
+		qh = ehci->async_iaa;
+		ehci->async_iaa = qh->unlink_next;
+		qh->unlink_next = NULL;
+
+		qh->qh_state = QH_STATE_IDLE;
+		qh->qh_next.qh = NULL;
+
+		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;
+
+	/* Start a new IAA cycle if any QHs are waiting for it */
+	if (ehci->async_unlink) {
+		start_iaa_cycle(ehci, true);
+		if (unlikely(ehci->rh_state < EHCI_RH_RUNNING))
+			goto restart;
+	}
 }
 
 /* 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)
 {
-	struct ehci_qh	*prev;
-
-#ifdef DEBUG
-	assert_spin_locked(&ehci->lock);
-	if (ehci->async_unlink
-			|| (qh->qh_state != QH_STATE_LINKED
-				&& qh->qh_state != QH_STATE_UNLINK_WAIT)
-			)
-		BUG ();
-#endif
-
-	qh->qh_state = QH_STATE_UNLINK;
-	ehci->async_unlink = qh;
-	if (!qh->unlink_next)
-		ehci->async_unlink_last = qh;
-
-	prev = ehci->async;
-	while (prev->qh_next.qh != qh)
-		prev = prev->qh_next.qh;
-
-	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;
-	wmb ();
-
-	/* If the controller isn't running, we don't have to wait for it */
-	if (unlikely(ehci->rh_state < EHCI_RH_RUNNING)) {
-		/* if (unlikely (qh->unlink_next != 0))
-		 *	this will recurse, probably not much
-		 */
-		end_unlink_async (ehci);
+	/*
+	 * If the QH isn't linked then there's nothing we can do
+	 * unless we were called during a giveback, in which case
+	 * qh_completions() has to deal with it.
+	 */
+	if (qh->qh_state != QH_STATE_LINKED) {
+		if (qh->qh_state == QH_STATE_COMPLETING)
+			qh->needs_rescan = 1;
 		return;
 	}
 
-	ehci_writel(ehci, ehci->command | CMD_IAAD, &ehci->regs->command);
-	(void)ehci_readl(ehci, &ehci->regs->command);
-	ehci_enable_event(ehci, EHCI_HRTIMER_IAA_WATCHDOG, true);
+	single_unlink_async(ehci, qh);
+	start_iaa_cycle(ehci, false);
 }
 
 /*-------------------------------------------------------------------------*/
@@ -1269,11 +1298,11 @@ static void scan_async (struct ehci_hcd *ehci)
 			 * 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 start_unlink_async().
+			 * in single_unlink_async().
 			 */
 			temp = qh_completions(ehci, qh);
 			if (qh->needs_rescan)
-				unlink_async(ehci, qh);
+				start_unlink_async(ehci, qh);
 			qh->unlink_time = jiffies + EHCI_SHRINK_JIFFIES;
 			if (temp != 0)
 				goto rescan;