Thomas Wintergerst reports that when this tsleep went away, certain

cards stopped working.  Specifically the AVM B1 PCMCIA Card no longer
detected.  Its CIS chain read back as all FF's.  Putting the delay
back solves those problems.  I've opted to put in a much shorter delay
because as far as I can tell, no delay is really needed here.  We'll
see how well this works in practice.

1 files changed, 10 insertions, 0 deletions

diff --git a/sys/dev/pccard/pccard_cis.c b/sys/dev/pccard/pccard_cis.c
index ce2c23d..b912a20 100644
--- a/sys/dev/pccard/pccard_cis.c
+++ b/sys/dev/pccard/pccard_cis.c
@@ -94,6 +94,16 @@ pccard_read_cis(struct pccard_softc *sc)
 	STAILQ_INIT(&state.card->pf_head);
 	state.pf = NULL;
 
+	/*
+	 * XXX The following shouldn't be needed, but some slow cards
+	 * XXX seem to need it still.  Need to investigate if there's
+	 * XXX a way to tell if the card is 'ready' or not rather than
+	 * XXX sleeping like this.  We're called just after the power
+	 * XXX up of the socket.  The standard timing diagrams don't
+	 * XXX seem to indicate that a delay is required.  The old
+	 * XXX delay was 1s.  This delay is .1s.
+	 */
+	tsleep(&state, 0, "pccard", hz / 10);
 	if (pccard_scan_cis(device_get_parent(sc->dev), sc->dev,
 	    pccard_parse_cis_tuple, &state) == -1)
 		state.card->error++;