OHCI: implement new semantics for URB_ISO_ASAP

This patch (as1614) updates the isochronous scheduling in ohci-hcd to
match the new semantics for URB_ISO_ASAP.  Testing revealed a hardware
bug in the way my OHCI controller handles expired isochronous TDs;
consequently the patch tries hard to avoid creating them (unlike the
ehci-hcd and uhci-hcd drivers).

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

drivers/usb/host/ohci-hcd.c

@@ -231,13 +231,41 @@ static int ohci_urb_enqueue (
 			frame &= ~(ed->interval - 1);
 			frame |= ed->branch;
 			urb->start_frame = frame;
+		}
+	} else if (ed->type == PIPE_ISOCHRONOUS) {
+		u16	next = ohci_frame_no(ohci) + 2;
+		u16	frame = ed->last_iso + ed->interval;
+
+		/* Behind the scheduling threshold? */
+		if (unlikely(tick_before(frame, next))) {
 
-			/* yes, only URB_ISO_ASAP is supported, and
-			 * urb->start_frame is never used as input.
+			/* USB_ISO_ASAP: Round up to the first available slot */
+			if (urb->transfer_flags & URB_ISO_ASAP)
+				frame += (next - frame + ed->interval - 1) &
+						-ed->interval;
+
+			/*
+			 * Not ASAP: Use the next slot in the stream.  If
+			 * the entire URB falls before the threshold, fail.
 			 */
+			else if (tick_before(frame + ed->interval *
+					(urb->number_of_packets - 1), next)) {
+				retval = -EXDEV;
+				usb_hcd_unlink_urb_from_ep(hcd, urb);
+				goto fail;
+			}
+
+			/*
+			 * Some OHCI hardware doesn't handle late TDs
+			 * correctly.  After retiring them it proceeds to
+			 * the next ED instead of the next TD.  Therefore
+			 * we have to omit the late TDs entirely.
+			 */
+			urb_priv->td_cnt = DIV_ROUND_UP(next - frame,
+					ed->interval);
 		}
-	} else if (ed->type == PIPE_ISOCHRONOUS)
-		urb->start_frame = ed->last_iso + ed->interval;
+		urb->start_frame = frame;
+	}
 
 	/* fill the TDs and link them to the ed; and
 	 * enable that part of the schedule, if needed

drivers/usb/host/ohci-q.c

@@ -596,7 +596,6 @@ static void td_submit_urb (
 		urb_priv->ed->hwHeadP &= ~cpu_to_hc32 (ohci, ED_C);
 	}
 
-	urb_priv->td_cnt = 0;
 	list_add (&urb_priv->pending, &ohci->pending);
 
 	if (data_len)
@@ -672,7 +671,8 @@ static void td_submit_urb (
 	 * we could often reduce the number of TDs here.
 	 */
 	case PIPE_ISOCHRONOUS:
-		for (cnt = 0; cnt < urb->number_of_packets; cnt++) {
+		for (cnt = urb_priv->td_cnt; cnt < urb->number_of_packets;
+				cnt++) {
 			int	frame = urb->start_frame;
 
 			// FIXME scheduling should handle frame counter