musb: support disconnect after HNP roleswitch

Adjust HNP state machines in MUSB driver so that they handle the
case where the cable is disconnected.  The A-side machine was
very wrong (unrecoverable); the B-Side was much less so.

 - A_PERIPHERAL ... as usual, the non-observability of the ID
   pin through Mentor's registers makes trouble.  We can't go
   directly to A_WAIT_VFALL to end the session and start the
   disconnect processing.  We can however sense link suspending,
   go to A_WAIT_BCON, and from there use OTG timeouts to finally
   trigger that A_WAIT_VFALL transition.  (Hoping that nobody
   reconnects quickly to that port and notices the wrong state.)

 - B_HOST ... actually clear the Host Request (HR) bit as the
   messages say, disconnect the peripheral from the root hub,
   and don't detour through a suspend state.  (In some cases
   this would eventually have cleaned up.)

Also adjust the A_SUSPEND transition to respect the A_AIDL_BDIS
timeout, so if HNP doesn't trigger quickly enough the A_WAIT_VFALL
transition happens as it should.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

drivers/usb/musb/musb_gadget.c

@@ -1962,9 +1962,11 @@ void musb_g_disconnect(struct musb *musb)
 		DBG(2, "Unhandled disconnect %s, setting a_idle\n",
 			otg_state_string(musb));
 		musb->xceiv->state = OTG_STATE_A_IDLE;
+		MUSB_HST_MODE(musb);
 		break;
 	case OTG_STATE_A_PERIPHERAL:
 		musb->xceiv->state = OTG_STATE_A_WAIT_BCON;
+		MUSB_HST_MODE(musb);
 		break;
 	case OTG_STATE_B_WAIT_ACON:
 	case OTG_STATE_B_HOST: