diff --git a/drivers/gpu/drm/nouveau/dispnv50/core.h b/drivers/gpu/drm/nouveau/dispnv50/core.h
index df8336b593f7..ff94f3f6f264 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/core.h
+++ b/drivers/gpu/drm/nouveau/dispnv50/core.h
@@ -6,6 +6,7 @@
 struct nv50_core {
 	const struct nv50_core_func *func;
 	struct nv50_dmac chan;
+	bool assign_windows;
 };
 
 int nv50_core_new(struct nouveau_drm *, struct nv50_core **);
@@ -18,6 +19,10 @@ struct nv50_core_func {
 			      struct nvif_device *);
 	void (*update)(struct nv50_core *, u32 *interlock, bool ntfy);
 
+	struct {
+		void (*owner)(struct nv50_core *);
+	} wndw;
+
 	const struct nv50_head_func *head;
 	const struct nv50_outp_func {
 		void (*ctrl)(struct nv50_core *, int or, u32 ctrl,
@@ -48,6 +53,7 @@ int core917d_new(struct nouveau_drm *, s32, struct nv50_core **);
 int corec37d_new(struct nouveau_drm *, s32, struct nv50_core **);
 int corec37d_ntfy_wait_done(struct nouveau_bo *, u32, struct nvif_device *);
 void corec37d_update(struct nv50_core *, u32 *, bool);
+void corec37d_wndw_owner(struct nv50_core *);
 extern const struct nv50_outp_func sorc37d;
 
 int corec57d_new(struct nouveau_drm *, s32, struct nv50_core **);
diff --git a/drivers/gpu/drm/nouveau/dispnv50/corec37d.c b/drivers/gpu/drm/nouveau/dispnv50/corec37d.c
index 40d9b654ab8c..f414171e40b4 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/corec37d.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/corec37d.c
@@ -24,6 +24,20 @@
 
 #include <nouveau_bo.h>
 
+void
+corec37d_wndw_owner(struct nv50_core *core)
+{
+	const u32 windows = 8; /*XXX*/
+	u32 *push, i;
+	if ((push = evo_wait(&core->chan, 2 * windows))) {
+		for (i = 0; i < windows; i++) {
+			evo_mthd(push, 0x1000 + (i * 0x080), 1);
+			evo_data(push, i >> 1);
+		}
+		evo_kick(push, &core->chan);
+	}
+}
+
 void
 corec37d_update(struct nv50_core *core, u32 *interlock, bool ntfy)
 {
@@ -76,12 +90,11 @@ corec37d_init(struct nv50_core *core)
 {
 	const u32 windows = 8; /*XXX*/
 	u32 *push, i;
-	if ((push = evo_wait(&core->chan, 2 + 6 * windows + 2))) {
+	if ((push = evo_wait(&core->chan, 2 + 5 * windows + 2))) {
 		evo_mthd(push, 0x0208, 1);
 		evo_data(push, core->chan.sync.handle);
 		for (i = 0; i < windows; i++) {
-			evo_mthd(push, 0x1000 + (i * 0x080), 3);
-			evo_data(push, i >> 1);
+			evo_mthd(push, 0x1004 + (i * 0x080), 2);
 			evo_data(push, 0x0000001f);
 			evo_data(push, 0x00000000);
 			evo_mthd(push, 0x1010 + (i * 0x080), 1);
@@ -90,6 +103,7 @@ corec37d_init(struct nv50_core *core)
 		evo_mthd(push, 0x0200, 1);
 		evo_data(push, 0x00000001);
 		evo_kick(push, &core->chan);
+		core->assign_windows = true;
 	}
 }
 
@@ -99,6 +113,7 @@ corec37d = {
 	.ntfy_init = corec37d_ntfy_init,
 	.ntfy_wait_done = corec37d_ntfy_wait_done,
 	.update = corec37d_update,
+	.wndw.owner = corec37d_wndw_owner,
 	.head = &headc37d,
 	.sor = &sorc37d,
 };
diff --git a/drivers/gpu/drm/nouveau/dispnv50/corec57d.c b/drivers/gpu/drm/nouveau/dispnv50/corec57d.c
index b606d68cda10..b540606ac052 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/corec57d.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/corec57d.c
@@ -27,12 +27,11 @@ corec57d_init(struct nv50_core *core)
 {
 	const u32 windows = 8; /*XXX*/
 	u32 *push, i;
-	if ((push = evo_wait(&core->chan, 2 + 6 * windows + 2))) {
+	if ((push = evo_wait(&core->chan, 2 + 5 * windows + 2))) {
 		evo_mthd(push, 0x0208, 1);
 		evo_data(push, core->chan.sync.handle);
 		for (i = 0; i < windows; i++) {
-			evo_mthd(push, 0x1000 + (i * 0x080), 3);
-			evo_data(push, i >> 1);
+			evo_mthd(push, 0x1004 + (i * 0x080), 2);
 			evo_data(push, 0x0000000f);
 			evo_data(push, 0x00000000);
 			evo_mthd(push, 0x1010 + (i * 0x080), 1);
@@ -41,6 +40,7 @@ corec57d_init(struct nv50_core *core)
 		evo_mthd(push, 0x0200, 1);
 		evo_data(push, 0x00000001);
 		evo_kick(push, &core->chan);
+		core->assign_windows = true;
 	}
 }
 
@@ -50,6 +50,7 @@ corec57d = {
 	.ntfy_init = corec37d_ntfy_init,
 	.ntfy_wait_done = corec37d_ntfy_wait_done,
 	.update = corec37d_update,
+	.wndw.owner = corec37d_wndw_owner,
 	.head = &headc57d,
 	.sor = &sorc37d,
 };
diff --git a/drivers/gpu/drm/nouveau/dispnv50/disp.c b/drivers/gpu/drm/nouveau/dispnv50/disp.c
index 2f123082c85d..a3dc2ba19fb2 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/disp.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/disp.c
@@ -1933,6 +1933,7 @@ nv50_disp_atomic_commit_tail(struct drm_atomic_state *state)
 	struct nouveau_drm *drm = nouveau_drm(dev);
 	struct nv50_disp *disp = nv50_disp(dev);
 	struct nv50_atom *atom = nv50_atom(state);
+	struct nv50_core *core = disp->core;
 	struct nv50_outp_atom *outp, *outt;
 	u32 interlock[NV50_DISP_INTERLOCK__SIZE] = {};
 	int i;
@@ -2051,6 +2052,21 @@ nv50_disp_atomic_commit_tail(struct drm_atomic_state *state)
 		}
 	}
 
+	/* Update window->head assignment.
+	 *
+	 * This has to happen in an update that's not interlocked with
+	 * any window channels to avoid hitting HW error checks.
+	 *
+	 *TODO: Proper handling of window ownership (Turing apparently
+	 *      supports non-fixed mappings).
+	 */
+	if (core->assign_windows) {
+		core->func->wndw.owner(core);
+		core->func->update(core, interlock, false);
+		core->assign_windows = false;
+		interlock[NV50_DISP_INTERLOCK_CORE] = 0;
+	}
+
 	/* Update plane(s). */
 	for_each_new_plane_in_state(state, plane, new_plane_state, i) {
 		struct nv50_wndw_atom *asyw = nv50_wndw_atom(new_plane_state);