svn commit: r225324 - user/adrian/if_ath_tx/sys/dev/ath

[Adrian Chadd]

Author: adrian
Date: Fri Sep  2 08:13:07 2011
New Revision: 225324
URL: http://svn.freebsd.org/changeset/base/225324

Log:
  If the RX link pointer is NULL, RX DMA has been halted. This means
  the first buffer re-added to the list should be given to the hardware.
  
  This doesn't restart DMA by itself; I'l follow up with some tinkering
  to try and re-enable it without having to forcibly stop and restart
  the PCU.
  
  Whilst here, add a comment to point out another thing I need to tinker
  with.
  
  Obtained from:	Linux, Atheros

Modified:
  user/adrian/if_ath_tx/sys/dev/ath/if_ath.c

Modified: user/adrian/if_ath_tx/sys/dev/ath/if_ath.c
==============================================================================
--- user/adrian/if_ath_tx/sys/dev/ath/if_ath.c	Fri Sep  2 08:00:57 2011	(r225323)
+++ user/adrian/if_ath_tx/sys/dev/ath/if_ath.c	Fri Sep  2 08:13:07 2011	(r225324)
@@ -3418,7 +3418,9 @@ ath_rxbuf_init(struct ath_softc *sc, str
 		, 0
 	);
 
-	if (sc->sc_rxlink != NULL)
+	if (sc->sc_rxlink == NULL)
+		ath_hal_putrxbuf(sc->sc_ah, bf->bf_daddr);
+	else
 		*sc->sc_rxlink = bf->bf_daddr;
 	sc->sc_rxlink = &ds->ds_link;
 	return 0;
@@ -3931,6 +3933,11 @@ rx_next:
 	 */
 	if (sc->sc_kickpcu) {
 		sc->sc_kickpcu = 0;
+		/*
+		 * XXX this causes a 3ms delay; and shuts down a lof
+		 * XXX is it really needed? Or is it just enough to
+		 * XXX kick the PCU again to continue RXing?
+		 */
 		ath_stoprecv(sc);
 		sc->sc_imask |= (HAL_INT_RXEOL | HAL_INT_RXORN);
 		if (ath_startrecv(sc) != 0) {