Put the clearing of the socket cd interrupt and setting of it under

protection of sc->mtx.  Sometimes multiple insert/ejections could
cause further insertion/removals to not happen due to racing an ISR...

1 files changed, 8 insertions, 10 deletions

diff --git a/sys/dev/pccbb/pccbb.c b/sys/dev/pccbb/pccbb.c
index 8df8efc..a3435b4 100644
--- a/sys/dev/pccbb/pccbb.c
+++ b/sys/dev/pccbb/pccbb.c
@@ -1011,23 +1011,21 @@ cbb_event_thread(void *arg)
 		mtx_unlock(&Giant);
 
 		/*
-		 * In our ISR, we turn off the card changed interrupt.  Turn
-		 * them back on here before we wait for them to happen.  We
-		 * turn them on/off so that we can tolerate a large latency
-		 * between the time we signal cbb_event_thread and it gets
-		 * a chance to run.
-		 */
-		cbb_setb(sc, CBB_SOCKET_MASK, CBB_SOCKET_MASK_CD);
-
-		/*
 		 * Wait until it has been 1s since the last time we
 		 * get an interrupt.  We handle the rest of the interrupt
 		 * at the top of the loop.  Although we clear the bit in the
 		 * ISR, we signal sc->cv from the detach path after we've
 		 * set the CBB_KTHREAD_DONE bit, so we can't do a simple
 		 * 1s sleep here.
+		 *
+		 * In our ISR, we turn off the card changed interrupt.  Turn
+		 * them back on here before we wait for them to happen.  We
+		 * turn them on/off so that we can tolerate a large latency
+		 * between the time we signal cbb_event_thread and it gets
+		 * a chance to run.
 		 */
 		mtx_lock(&sc->mtx);
+		cbb_setb(sc, CBB_SOCKET_MASK, CBB_SOCKET_MASK_CD);
 		cv_wait(&sc->cv, &sc->mtx);
 		err = 0;
 		while (err != EWOULDBLOCK &&
@@ -1125,8 +1123,8 @@ cbb_intr(void *arg)
 		 * excellent results.
 		 */
 		if (sockevent & CBB_SOCKET_EVENT_CD) {
-			cbb_clrb(sc, CBB_SOCKET_MASK, CBB_SOCKET_MASK_CD);
 			mtx_lock(&sc->mtx);
+			cbb_setb(sc, CBB_SOCKET_MASK, CBB_SOCKET_MASK_CD);
 			sc->flags &= ~CBB_CARD_OK;
 			cbb_disable_func_intr(sc);
 			DPRINTF(("Waking up thread\n"));