# HG changeset patch
# User Taylor R Campbell <riastradh@NetBSD.org>
# Date 1590176371 0
#      Fri May 22 19:39:31 2020 +0000
# Branch trunk
# Node ID 13c0c4ce491e63a6c0625dacdadbef70a7b489d7
# Parent  1ed1cd6fab61c6d6fab72eec14f71b46f7f8a352
# EXP-Topic riastradh-sdmmc-task-fixes
Remove unnecessary wait in ldbegindetach.

Like disk_begindetach, ldbegindetach only commits to detaching but
doesn't wait for existing xfers to drain; it is up to the driver to
abort them, once we are committed, and then ldenddetach to wait for
them to drain.

diff -r 1ed1cd6fab61 -r 13c0c4ce491e sys/dev/ld.c
--- a/sys/dev/ld.c	Fri May 22 02:44:03 2020 +0000
+++ b/sys/dev/ld.c	Fri May 22 19:39:31 2020 +0000
@@ -189,26 +189,27 @@ int
 ldbegindetach(struct ld_softc *sc, int flags)
 {
 	struct dk_softc *dksc = &sc->sc_dksc;
-	int rv = 0;
+	int error;
 
+	/* If we never attached properly, no problem with detaching.  */
 	if ((sc->sc_flags & LDF_ENABLED) == 0)
-		return (0);
+		return 0;
 
-	rv = disk_begindetach(&dksc->sc_dkdev, ld_lastclose, dksc->sc_dev, flags);
+	/*
+	 * If the disk is still open, back out before we commit to
+	 * detaching.
+	 */
+	error = disk_begindetach(&dksc->sc_dkdev, ld_lastclose, dksc->sc_dev,
+	    flags);
+	if (error)
+		return error;
 
-	if (rv != 0)
-		return rv;
-
+	/* We are now committed to detaching.  Prevent new xfers.  */
 	mutex_enter(&sc->sc_mutex);
 	sc->sc_maxqueuecnt = 0;
-
-	while (sc->sc_queuecnt > 0) {
-		sc->sc_flags |= LDF_DRAIN;
-		cv_wait(&sc->sc_drain, &sc->sc_mutex);
-	}
 	mutex_exit(&sc->sc_mutex);
 
-	return (rv);
+	return 0;
 }
 
 void
@@ -220,12 +221,17 @@ ldenddetach(struct ld_softc *sc)
 	if ((sc->sc_flags & LDF_ENABLED) == 0)
 		return;
 
+	/* Wait for commands queued with the hardware to complete. */
 	mutex_enter(&sc->sc_mutex);
-
-	/* Wait for commands queued with the hardware to complete. */
-	if (sc->sc_queuecnt != 0) {
-		if (cv_timedwait(&sc->sc_drain, &sc->sc_mutex, 30 * hz))
+	while (sc->sc_queuecnt > 0) {
+		if (cv_timedwait(&sc->sc_drain, &sc->sc_mutex, 30 * hz)) {
+			/*
+			 * XXX This seems like a recipe for crashing on
+			 * use after free...
+			 */
 			printf("%s: not drained\n", dksc->sc_xname);
+			break;
+		}
 	}
 	mutex_exit(&sc->sc_mutex);
 
@@ -467,10 +473,7 @@ lddone(struct ld_softc *sc, struct buf *
 
 	mutex_enter(&sc->sc_mutex);
 	if (--sc->sc_queuecnt <= sc->sc_maxqueuecnt) {
-		if ((sc->sc_flags & LDF_DRAIN) != 0) {
-			sc->sc_flags &= ~LDF_DRAIN;
-			cv_broadcast(&sc->sc_drain);
-		}
+		cv_broadcast(&sc->sc_drain);
 		mutex_exit(&sc->sc_mutex);
 		dk_start(dksc, NULL);
 	} else
diff -r 1ed1cd6fab61 -r 13c0c4ce491e sys/dev/ldvar.h
--- a/sys/dev/ldvar.h	Fri May 22 02:44:03 2020 +0000
+++ b/sys/dev/ldvar.h	Fri May 22 19:39:31 2020 +0000
@@ -67,7 +67,7 @@ struct ld_softc {
 
 /* sc_flags */
 #define	LDF_ENABLED	0x001		/* device enabled */
-#define	LDF_DRAIN	0x020		/* maxqueuecnt has changed; drain */
+#define	LDF_UNUSED0	0x020		/* was LDF_DRAIN */
 #define	LDF_NO_RND	0x040		/* do not attach rnd source */
 #define	LDF_MPSAFE	0x080		/* backend is MPSAFE */