Do not refcount the cdevsw, but rather maintain a cdev->si_threadcount

of the number of threads which are inside whatever is behind the
cdevsw for this particular cdev.

Make the device mutex visible through dev_lock() and dev_unlock().
We may want finer granularity later.

Replace spechash_mtx use with dev_lock()/dev_unlock().

5 files changed, 90 insertions, 86 deletions

diff --git a/sys/fs/specfs/spec_vnops.c b/sys/fs/specfs/spec_vnops.c
index 5133496..f1f8075 100644
--- a/sys/fs/specfs/spec_vnops.c
+++ b/sys/fs/specfs/spec_vnops.c
@@ -188,8 +188,9 @@ spec_open(ap)
 		vp->v_vflag |= VV_ISTTY;
 
 	VOP_UNLOCK(vp, 0, td);
-	dev_ref(dev);
-	cdevsw_ref(dsw);
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	if(!(dsw->d_flags & D_NEEDGIANT)) {
 		DROP_GIANT();
 		if (dsw->d_fdopen != NULL)
@@ -201,9 +202,9 @@ spec_open(ap)
 		error = dsw->d_fdopen(dev, ap->a_mode, td, ap->a_fdidx);
 	else
 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
-	cdevsw_rel(dsw);
-	if (error != 0)
-		dev_rel(dev);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
 
 	if (error)
@@ -260,14 +261,18 @@ spec_read(ap)
 	VOP_UNLOCK(vp, 0, td);
 	KASSERT(dev->si_refcount > 0,
 	    ("specread() on un-referenced struct cdev *(%s)", devtoname(dev)));
-	cdevsw_ref(dsw);
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	if (!(dsw->d_flags & D_NEEDGIANT)) {
 		DROP_GIANT();
 		error = dsw->d_read(dev, uio, ap->a_ioflag);
 		PICKUP_GIANT();
 	} else
 		error = dsw->d_read(dev, uio, ap->a_ioflag);
-	cdevsw_rel(dsw);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
 		vfs_timestamp(&dev->si_atime);
@@ -304,14 +309,18 @@ spec_write(ap)
 	VOP_UNLOCK(vp, 0, td);
 	KASSERT(dev->si_refcount > 0,
 	    ("spec_write() on un-referenced struct cdev *(%s)", devtoname(dev)));
-	cdevsw_ref(dsw);
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	if (!(dsw->d_flags & D_NEEDGIANT)) {
 		DROP_GIANT();
 		error = dsw->d_write(dev, uio, ap->a_ioflag);
 		PICKUP_GIANT();
 	} else
 		error = dsw->d_write(dev, uio, ap->a_ioflag);
-	cdevsw_rel(dsw);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
 		vfs_timestamp(&dev->si_ctime);
@@ -343,7 +352,9 @@ spec_ioctl(ap)
 	dsw = devsw(dev);
 	KASSERT(dev->si_refcount > 0,
 	    ("spec_ioctl() on un-referenced struct cdev *(%s)", devtoname(dev)));
-	cdevsw_ref(dsw);
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	if (!(dsw->d_flags & D_NEEDGIANT)) {
 		DROP_GIANT();
 		error = dsw->d_ioctl(dev, ap->a_command,
@@ -352,7 +363,9 @@ spec_ioctl(ap)
 	} else 
 		error = dsw->d_ioctl(dev, ap->a_command,
 		    ap->a_data, ap->a_fflag, ap->a_td);
-	cdevsw_rel(dsw);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 	if (error == ENOIOCTL)
 		error = ENOTTY;
 	return (error);
@@ -376,14 +389,18 @@ spec_poll(ap)
 	dsw = devsw(dev);
 	KASSERT(dev->si_refcount > 0,
 	    ("spec_poll() on un-referenced struct cdev *(%s)", devtoname(dev)));
-	cdevsw_ref(dsw);
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	if (!(dsw->d_flags & D_NEEDGIANT)) {
 		/* XXX: not yet DROP_GIANT(); */
 		error = dsw->d_poll(dev, ap->a_events, ap->a_td);
 		/* XXX: not yet PICKUP_GIANT(); */
 	} else
 		error = dsw->d_poll(dev, ap->a_events, ap->a_td);
-	cdevsw_rel(dsw);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 	return(error);
 }
 
@@ -403,14 +420,18 @@ spec_kqfilter(ap)
 	dsw = devsw(dev);
 	KASSERT(dev->si_refcount > 0,
 	    ("spec_kqfilter() on un-referenced struct cdev *(%s)", devtoname(dev)));
-	cdevsw_ref(dsw);
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	if (!(dsw->d_flags & D_NEEDGIANT)) {
 		DROP_GIANT();
 		error = dsw->d_kqfilter(dev, ap->a_kn);
 		PICKUP_GIANT();
 	} else
 		error = dsw->d_kqfilter(dev, ap->a_kn);
-	cdevsw_rel(dsw);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 	return (error);
 }
 
@@ -595,15 +616,18 @@ spec_close(ap)
 	VI_UNLOCK(vp);
 	KASSERT(dev->si_refcount > 0,
 	    ("spec_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
-	cdevsw_ref(dsw);
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	if (!(dsw->d_flags & D_NEEDGIANT)) {
 		DROP_GIANT();
 		error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
 		PICKUP_GIANT();
 	} else
 		error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
-	cdevsw_rel(dsw);
-	dev_rel(dev);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 	return (error);
 }
 
diff --git a/sys/kern/kern_conf.c b/sys/kern/kern_conf.c
index 2249f04..46fca85 100644
--- a/sys/kern/kern_conf.c
+++ b/sys/kern/kern_conf.c
@@ -73,16 +73,16 @@ static void freedev(struct cdev *dev);
 static struct cdev *newdev(int x, int y);
 
 
-static void
-devlock(void)
+void
+dev_lock(void)
 {
 	if (!mtx_initialized(&devmtx))
 		mtx_init(&devmtx, "cdev", NULL, MTX_DEF);
 	mtx_lock(&devmtx);
 }
 
-static void
-devunlock(void)
+void
+dev_unlock(void)
 {
 	mtx_unlock(&devmtx);
 }
@@ -90,15 +90,15 @@ devunlock(void)
 void
 dev_ref(struct cdev *dev)
 {
-	devlock();
+	dev_lock();
 	dev->si_refcount++;
-	devunlock();
+	dev_unlock();
 }
 
 void
 dev_rel(struct cdev *dev)
 {
-	devlock();
+
 	dev->si_refcount--;
 	KASSERT(dev->si_refcount >= 0,
 	    ("dev_rel(%s) gave negative count", devtoname(dev)));
@@ -106,25 +106,6 @@ dev_rel(struct cdev *dev)
 		LIST_REMOVE(dev, si_list);
 		freedev(dev);
 	}
-	devunlock();
-}
-
-void
-cdevsw_ref(struct cdevsw *csw)
-{
-	devlock();
-	csw->d_refcount++;
-	devunlock();
-}
-
-void
-cdevsw_rel(struct cdevsw *csw)
-{
-	devlock();
-	csw->d_refcount--;
-	KASSERT(csw->d_refcount >= 0,
-	    ("cdevsw_vrel(%s) gave negative count", csw->d_name));
-	devunlock();
 }
 
 int
@@ -396,7 +377,7 @@ static void
 prep_cdevsw(struct cdevsw *devsw)
 {
 
-	devlock();
+	dev_lock();
 
 	if (devsw->d_version != D_VERSION_00) {
 		printf(
@@ -451,7 +432,7 @@ prep_cdevsw(struct cdevsw *devsw)
 			reserved_majors[devsw->d_maj] = devsw->d_maj;
 		}
 	}
-	devunlock();
+	dev_unlock();
 }
 
 struct cdev *
@@ -477,7 +458,7 @@ make_dev(struct cdevsw *devsw, int minornr, uid_t uid, gid_t gid, int perms, con
 		 */
 		return (dev);
 	}
-	devlock();
+	dev_lock();
 	KASSERT(!(dev->si_flags & SI_NAMED),
 	    ("make_dev() by driver %s on pre-existing device (maj=%d, min=%d, name=%s)",
 	    devsw->d_name, major(dev), minor(dev), devtoname(dev)));
@@ -497,7 +478,7 @@ make_dev(struct cdevsw *devsw, int minornr, uid_t uid, gid_t gid, int perms, con
 
 	LIST_INSERT_HEAD(&devsw->d_devs, dev, si_list);
 	devfs_create(dev);
-	devunlock();
+	dev_unlock();
 	return (dev);
 }
 
@@ -518,11 +499,11 @@ void
 dev_depends(struct cdev *pdev, struct cdev *cdev)
 {
 
-	devlock();
+	dev_lock();
 	cdev->si_parent = pdev;
 	cdev->si_flags |= SI_CHILD;
 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
-	devunlock();
+	dev_unlock();
 }
 
 struct cdev *
@@ -533,7 +514,7 @@ make_dev_alias(struct cdev *pdev, const char *fmt, ...)
 	int i;
 
 	dev = allocdev();
-	devlock();
+	dev_lock();
 	dev->si_flags |= SI_ALIAS;
 	dev->si_flags |= SI_NAMED;
 	va_start(ap, fmt);
@@ -545,7 +526,7 @@ make_dev_alias(struct cdev *pdev, const char *fmt, ...)
 	va_end(ap);
 
 	devfs_create(dev);
-	devunlock();
+	dev_unlock();
 	dev_depends(pdev, dev);
 	return (dev);
 }
@@ -606,9 +587,9 @@ void
 destroy_dev(struct cdev *dev)
 {
 
-	devlock();
+	dev_lock();
 	idestroy_dev(dev);
-	devunlock();
+	dev_unlock();
 }
 
 const char *
diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index 063e7d0..b2e73b9 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -3098,18 +3098,24 @@ void
 dev_strategy(struct buf *bp)
 {
 	struct cdevsw *csw;
+	struct cdev *dev;
 
 	if ((!bp->b_iocmd) || (bp->b_iocmd & (bp->b_iocmd - 1)))
 		panic("b_iocmd botch");
 	bp->b_io.bio_done = bufdonebio;
 	bp->b_io.bio_caller2 = bp;
-	csw = devsw(bp->b_io.bio_dev);
-	KASSERT(bp->b_io.bio_dev->si_refcount > 0,
+	dev = bp->b_io.bio_dev;
+	csw = devsw(dev);
+	KASSERT(dev->si_refcount > 0,
 	    ("dev_strategy on un-referenced struct cdev *(%s)",
-	    devtoname(bp->b_io.bio_dev)));
-	cdevsw_ref(csw);
+	    devtoname(dev)));
+	dev_lock();
+	dev->si_threadcount++;
+	dev_unlock();
 	(*devsw(bp->b_io.bio_dev)->d_strategy)(&bp->b_io);
-	cdevsw_rel(csw);
+	dev_lock();
+	dev->si_threadcount--;
+	dev_unlock();
 }
 
 /*
diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
index 5f1f2de..8908ec6 100644
--- a/sys/kern/vfs_subr.c
+++ b/sys/kern/vfs_subr.c
@@ -156,12 +156,6 @@ static struct mtx mntid_mtx;
  */
 static struct mtx vnode_free_list_mtx;
 
-/*
- * For any iteration/modification of dev->si_hlist (linked through
- * v_specnext)
- */
-static struct mtx spechash_mtx;
-
 /* Publicly exported FS */
 struct nfs_public nfs_pub;
 
@@ -283,7 +277,6 @@ vntblinit(void *dummy __unused)
 	minvnodes = desiredvnodes / 4;
 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
-	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
 	TAILQ_INIT(&vnode_free_list);
 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
@@ -1899,9 +1892,9 @@ v_incr_usecount(struct vnode *vp, int delta)
 
 	vp->v_usecount += delta;
 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
-		mtx_lock(&spechash_mtx);
+		dev_lock();
 		vp->v_rdev->si_usecount += delta;
-		mtx_unlock(&spechash_mtx);
+		dev_unlock();
 	}
 }
 
@@ -1975,13 +1968,13 @@ addalias(nvp, dev)
 {
 
 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
-	dev_ref(dev);
-	nvp->v_rdev = dev;
 	VI_LOCK(nvp);
-	mtx_lock(&spechash_mtx);
+	dev_lock();
+	dev->si_refcount++;
+	nvp->v_rdev = dev;
 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
 	dev->si_usecount += nvp->v_usecount;
-	mtx_unlock(&spechash_mtx);
+	dev_unlock();
 	VI_UNLOCK(nvp);
 }
 
@@ -2585,9 +2578,9 @@ vop_revoke(ap)
 	VI_UNLOCK(vp);
 	dev = vp->v_rdev;
 	for (;;) {
-		mtx_lock(&spechash_mtx);
+		dev_lock();
 		vq = SLIST_FIRST(&dev->si_hlist);
-		mtx_unlock(&spechash_mtx);
+		dev_unlock();
 		if (vq == NULL)
 			break;
 		vgone(vq);
@@ -2706,12 +2699,12 @@ vgonel(vp, td)
 	 */
 	VI_LOCK(vp);
 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
-		mtx_lock(&spechash_mtx);
+		dev_lock();
 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
 		vp->v_rdev->si_usecount -= vp->v_usecount;
-		mtx_unlock(&spechash_mtx);
 		dev_rel(vp->v_rdev);
 		vp->v_rdev = NULL;
+		dev_unlock();
 	}
 
 	/*
@@ -2751,13 +2744,13 @@ vfinddev(dev, vpp)
 {
 	struct vnode *vp;
 
-	mtx_lock(&spechash_mtx);
+	dev_lock();
 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
 		*vpp = vp;
-		mtx_unlock(&spechash_mtx);
+		dev_unlock();
 		return (1);
 	}
-	mtx_unlock(&spechash_mtx);
+	dev_unlock();
 	return (0);
 }
 
@@ -2770,9 +2763,9 @@ vcount(vp)
 {
 	int count;
 
-	mtx_lock(&spechash_mtx);
+	dev_lock();
 	count = vp->v_rdev->si_usecount;
-	mtx_unlock(&spechash_mtx);
+	dev_unlock();
 	return (count);
 }
 
@@ -2785,9 +2778,9 @@ count_dev(dev)
 {
 	int count;
 
-	mtx_lock(&spechash_mtx);
+	dev_lock();
 	count = dev->si_usecount;
-	mtx_unlock(&spechash_mtx);
+	dev_unlock();
 	return(count);
 }
 
diff --git a/sys/sys/conf.h b/sys/sys/conf.h
index 20b4326..c714019 100644
--- a/sys/sys/conf.h
+++ b/sys/sys/conf.h
@@ -222,7 +222,7 @@ struct cdevsw {
 	/* These fields should not be messed with by drivers */
 	LIST_ENTRY(cdevsw)	d_list;
 	LIST_HEAD(, cdev)	d_devs;
-	int			d_refcount;
+	int			d_spare3;
 };
 
 #endif /* _KERNEL */
@@ -263,8 +263,6 @@ int clone_create(struct clonedevs **, struct cdevsw *, int *unit, struct cdev **
 int	count_dev(struct cdev *_dev);
 void	destroy_dev(struct cdev *_dev);
 struct cdevsw *devsw(struct cdev *_dev);
-void	cdevsw_ref(struct cdevsw *);
-void	cdevsw_rel(struct cdevsw *);
 const char *devtoname(struct cdev *_dev);
 int	dev_named(struct cdev *_pdev, const char *_name);
 void	dev_depends(struct cdev *_pdev, struct cdev *_cdev);
@@ -276,6 +274,8 @@ struct cdev *make_dev(struct cdevsw *_devsw, int _minor, uid_t _uid, gid_t _gid,
 		int _perms, const char *_fmt, ...) __printflike(6, 7);
 struct cdev *make_dev_alias(struct cdev *_pdev, const char *_fmt, ...) __printflike(2, 3);
 int	dev2unit(struct cdev *_dev);
+void	dev_lock(void);
+void	dev_unlock(void);
 int	unit2minor(int _unit);
 void	setconf(void);
 struct cdev *getdiskbyname(char *_name);