usb: dwc2: host: Add a delay before releasing periodic bandwidth

We'd like to be able to use HCD_BH in order to speed up the dwc2 host
interrupt handler quite a bit.  However, according to the kernel doc for
usb_submit_urb() (specifically the part about "Reserved Bandwidth
Transfers"), we need to keep a reservation active as long as a device
driver keeps submitting.  That was easy to do when we gave back the URB
in the interrupt context: we just looked at when our queue was empty and
released the reserved bandwidth then.  ...but now we need a little more
complexity.

We'll follow EHCI's lead in commit 9118f9eb4f1e ("USB: EHCI: improve
interrupt qh unlink") and add a 5ms delay.  Since we don't have a whole
timer infrastructure in dwc2, we'll just add a timer per QH.  The
overhead for this is very small.

Note that the dwc2 scheduler is pretty broken (see future patches to fix
it).  This patch attempts to replicate all old behavior and just add the
proper delay.

Acked-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Tested-by: Heiko Stuebner <heiko@sntech.de>
Tested-by: Stefan Wahren <stefan.wahren@i2se.com>
Signed-off-by: Felipe Balbi <balbi@kernel.org>

1 files changed, 181 insertions, 56 deletions

diff --git a/drivers/usb/dwc2/hcd_queue.c b/drivers/usb/dwc2/hcd_queue.c
index 0e9faa75593c..b9e4867e1afd 100644
--- a/drivers/usb/dwc2/hcd_queue.c
+++ b/drivers/usb/dwc2/hcd_queue.c
@@ -53,6 +53,94 @@
 #include "core.h"
 #include "hcd.h"
 
+/* Wait this long before releasing periodic reservation */
+#define DWC2_UNRESERVE_DELAY (msecs_to_jiffies(5))
+
+/**
+ * dwc2_do_unreserve() - Actually release the periodic reservation
+ *
+ * This function actually releases the periodic bandwidth that was reserved
+ * by the given qh.
+ *
+ * @hsotg: The HCD state structure for the DWC OTG controller
+ * @qh:    QH for the periodic transfer.
+ */
+static void dwc2_do_unreserve(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh)
+{
+	assert_spin_locked(&hsotg->lock);
+
+	WARN_ON(!qh->unreserve_pending);
+
+	/* No more unreserve pending--we're doing it */
+	qh->unreserve_pending = false;
+
+	if (WARN_ON(!list_empty(&qh->qh_list_entry)))
+		list_del_init(&qh->qh_list_entry);
+
+	/* Update claimed usecs per (micro)frame */
+	hsotg->periodic_usecs -= qh->usecs;
+
+	if (hsotg->core_params->uframe_sched > 0) {
+		int i;
+
+		for (i = 0; i < 8; i++) {
+			hsotg->frame_usecs[i] += qh->frame_usecs[i];
+			qh->frame_usecs[i] = 0;
+		}
+	} else {
+		/* Release periodic channel reservation */
+		hsotg->periodic_channels--;
+	}
+}
+
+/**
+ * dwc2_unreserve_timer_fn() - Timer function to release periodic reservation
+ *
+ * According to the kernel doc for usb_submit_urb() (specifically the part about
+ * "Reserved Bandwidth Transfers"), we need to keep a reservation active as
+ * long as a device driver keeps submitting.  Since we're using HCD_BH to give
+ * back the URB we need to give the driver a little bit of time before we
+ * release the reservation.  This worker is called after the appropriate
+ * delay.
+ *
+ * @work: Pointer to a qh unreserve_work.
+ */
+static void dwc2_unreserve_timer_fn(unsigned long data)
+{
+	struct dwc2_qh *qh = (struct dwc2_qh *)data;
+	struct dwc2_hsotg *hsotg = qh->hsotg;
+	unsigned long flags;
+
+	/*
+	 * Wait for the lock, or for us to be scheduled again.  We
+	 * could be scheduled again if:
+	 * - We started executing but didn't get the lock yet.
+	 * - A new reservation came in, but cancel didn't take effect
+	 *   because we already started executing.
+	 * - The timer has been kicked again.
+	 * In that case cancel and wait for the next call.
+	 */
+	while (!spin_trylock_irqsave(&hsotg->lock, flags)) {
+		if (timer_pending(&qh->unreserve_timer))
+			return;
+	}
+
+	/*
+	 * Might be no more unreserve pending if:
+	 * - We started executing but didn't get the lock yet.
+	 * - A new reservation came in, but cancel didn't take effect
+	 *   because we already started executing.
+	 *
+	 * We can't put this in the loop above because unreserve_pending needs
+	 * to be accessed under lock, so we can only check it once we got the
+	 * lock.
+	 */
+	if (qh->unreserve_pending)
+		dwc2_do_unreserve(hsotg, qh);
+
+	spin_unlock_irqrestore(&hsotg->lock, flags);
+}
+
 /**
  * dwc2_qh_init() - Initializes a QH structure
  *
@@ -71,6 +159,9 @@ static void dwc2_qh_init(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh,
 	dev_vdbg(hsotg->dev, "%s()\n", __func__);
 
 	/* Initialize QH */
+	qh->hsotg = hsotg;
+	setup_timer(&qh->unreserve_timer, dwc2_unreserve_timer_fn,
+		    (unsigned long)qh);
 	qh->ep_type = dwc2_hcd_get_pipe_type(&urb->pipe_info);
 	qh->ep_is_in = dwc2_hcd_is_pipe_in(&urb->pipe_info) ? 1 : 0;
 
@@ -240,6 +331,15 @@ struct dwc2_qh *dwc2_hcd_qh_create(struct dwc2_hsotg *hsotg,
  */
 void dwc2_hcd_qh_free(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh)
 {
+	/* Make sure any unreserve work is finished. */
+	if (del_timer_sync(&qh->unreserve_timer)) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&hsotg->lock, flags);
+		dwc2_do_unreserve(hsotg, qh);
+		spin_unlock_irqrestore(&hsotg->lock, flags);
+	}
+
 	if (qh->desc_list)
 		dwc2_hcd_qh_free_ddma(hsotg, qh);
 	kfree(qh);
@@ -477,51 +577,74 @@ static int dwc2_schedule_periodic(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh)
 {
 	int status;
 
-	if (hsotg->core_params->uframe_sched > 0) {
-		int frame = -1;
-
-		status = dwc2_find_uframe(hsotg, qh);
-		if (status == 0)
-			frame = 7;
-		else if (status > 0)
-			frame = status - 1;
-
-		/* Set the new frame up */
-		if (frame >= 0) {
-			qh->sched_frame &= ~0x7;
-			qh->sched_frame |= (frame & 7);
-			dwc2_sch_dbg(hsotg, "QH=%p sched_p sch=%04x, uf=%d\n",
-				     qh, qh->sched_frame, frame);
+	status = dwc2_check_max_xfer_size(hsotg, qh);
+	if (status) {
+		dev_dbg(hsotg->dev,
+			"%s: Channel max transfer size too small for periodic transfer\n",
+			__func__);
+		return status;
+	}
+
+	/* Cancel pending unreserve; if canceled OK, unreserve was pending */
+	if (del_timer(&qh->unreserve_timer))
+		WARN_ON(!qh->unreserve_pending);
+
+	/*
+	 * Only need to reserve if there's not an unreserve pending, since if an
+	 * unreserve is pending then by definition our old reservation is still
+	 * valid.  Unreserve might still be pending even if we didn't cancel if
+	 * dwc2_unreserve_timer_fn() already started.  Code in the timer handles
+	 * that case.
+	 */
+	if (!qh->unreserve_pending) {
+		if (hsotg->core_params->uframe_sched > 0) {
+			int frame = -1;
+
+			status = dwc2_find_uframe(hsotg, qh);
+			if (status == 0)
+				frame = 7;
+			else if (status > 0)
+				frame = status - 1;
+
+			/* Set the new frame up */
+			if (frame >= 0) {
+				qh->sched_frame &= ~0x7;
+				qh->sched_frame |= (frame & 7);
+				dwc2_sch_dbg(hsotg,
+					     "QH=%p sched_p sch=%04x, uf=%d\n",
+					     qh, qh->sched_frame, frame);
+			}
+
+			if (status > 0)
+				status = 0;
+		} else {
+			status = dwc2_periodic_channel_available(hsotg);
+			if (status) {
+				dev_info(hsotg->dev,
+					"%s: No host channel available for periodic transfer\n",
+					__func__);
+				return status;
+			}
+
+			status = dwc2_check_periodic_bandwidth(hsotg, qh);
 		}
 
-		if (status > 0)
-			status = 0;
-	} else {
-		status = dwc2_periodic_channel_available(hsotg);
 		if (status) {
-			dev_info(hsotg->dev,
-				 "%s: No host channel available for periodic transfer\n",
-				 __func__);
+			dev_dbg(hsotg->dev,
+				"%s: Insufficient periodic bandwidth for periodic transfer\n",
+				__func__);
 			return status;
 		}
 
-		status = dwc2_check_periodic_bandwidth(hsotg, qh);
-	}
+		if (hsotg->core_params->uframe_sched <= 0)
+			/* Reserve periodic channel */
+			hsotg->periodic_channels++;
 
-	if (status) {
-		dev_dbg(hsotg->dev,
-			"%s: Insufficient periodic bandwidth for periodic transfer\n",
-			__func__);
-		return status;
+		/* Update claimed usecs per (micro)frame */
+		hsotg->periodic_usecs += qh->usecs;
 	}
 
-	status = dwc2_check_max_xfer_size(hsotg, qh);
-	if (status) {
-		dev_dbg(hsotg->dev,
-			"%s: Channel max transfer size too small for periodic transfer\n",
-			__func__);
-		return status;
-	}
+	qh->unreserve_pending = 0;
 
 	if (hsotg->core_params->dma_desc_enable > 0)
 		/* Don't rely on SOF and start in ready schedule */
@@ -531,13 +654,6 @@ static int dwc2_schedule_periodic(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh)
 		list_add_tail(&qh->qh_list_entry,
 			      &hsotg->periodic_sched_inactive);
 
-	if (hsotg->core_params->uframe_sched <= 0)
-		/* Reserve periodic channel */
-		hsotg->periodic_channels++;
-
-	/* Update claimed usecs per (micro)frame */
-	hsotg->periodic_usecs += qh->usecs;
-
 	return status;
 }
 
@@ -551,22 +667,31 @@ static int dwc2_schedule_periodic(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh)
 static void dwc2_deschedule_periodic(struct dwc2_hsotg *hsotg,
 				     struct dwc2_qh *qh)
 {
-	int i;
+	bool did_modify;
 
-	list_del_init(&qh->qh_list_entry);
+	assert_spin_locked(&hsotg->lock);
 
-	/* Update claimed usecs per (micro)frame */
-	hsotg->periodic_usecs -= qh->usecs;
+	/*
+	 * Schedule the unreserve to happen in a little bit.  Cases here:
+	 * - Unreserve worker might be sitting there waiting to grab the lock.
+	 *   In this case it will notice it's been schedule again and will
+	 *   quit.
+	 * - Unreserve worker might not be scheduled.
+	 *
+	 * We should never already be scheduled since dwc2_schedule_periodic()
+	 * should have canceled the scheduled unreserve timer (hence the
+	 * warning on did_modify).
+	 *
+	 * We add + 1 to the timer to guarantee that at least 1 jiffy has
+	 * passed (otherwise if the jiffy counter might tick right after we
+	 * read it and we'll get no delay).
+	 */
+	did_modify = mod_timer(&qh->unreserve_timer,
+			       jiffies + DWC2_UNRESERVE_DELAY + 1);
+	WARN_ON(did_modify);
+	qh->unreserve_pending = 1;
 
-	if (hsotg->core_params->uframe_sched > 0) {
-		for (i = 0; i < 8; i++) {
-			hsotg->frame_usecs[i] += qh->frame_usecs[i];
-			qh->frame_usecs[i] = 0;
-		}
-	} else {
-		/* Release periodic channel reservation */
-		hsotg->periodic_channels--;
-	}
+	list_del_init(&qh->qh_list_entry);
 }
 
 /**