diff --git a/src/sna/gen3_render.c b/src/sna/gen3_render.c
index 06c4e784..6b9a2f81 100644
--- a/src/sna/gen3_render.c
+++ b/src/sna/gen3_render.c
@@ -1900,6 +1900,9 @@ static bool gen3_rectangle_begin(struct sna *sna,
 	struct gen3_render_state *state = &sna->render_state.gen3;
 	int ndwords, i1_cmd = 0, i1_len = 0;
 
+	if (sna_vertex_wait__locked(&sna->render) && sna->render.vertex_offset)
+		return true;
+
 	ndwords = 2;
 	if (op->need_magic_ca_pass)
 		ndwords += 100;
@@ -1939,6 +1942,13 @@ static bool gen3_rectangle_begin(struct sna *sna,
 static int gen3_get_rectangles__flush(struct sna *sna,
 				      const struct sna_composite_op *op)
 {
+	/* Preventing discarding new vbo after lock contention */
+	if (sna_vertex_wait__locked(&sna->render)) {
+		int rem = vertex_space(sna);
+		if (rem > op->floats_per_rect)
+			return rem;
+	}
+
 	if (!kgem_check_batch(&sna->kgem, op->need_magic_ca_pass ? 105: 5))
 		return 0;
 	if (!kgem_check_reloc_and_exec(&sna->kgem, 1))
@@ -1955,17 +1965,6 @@ static int gen3_get_rectangles__flush(struct sna *sna,
 		}
 	}
 
-	/* Preventing discarding new vbo after lock contention */
-	if (sna->render.active) {
-		int rem;
-
-		sna_vertex_wait__locked(&sna->render);
-
-		rem = vertex_space(sna);
-		if (rem > op->floats_per_rect)
-			return rem;
-	}
-
 	return gen3_vertex_finish(sna);
 }
 
diff --git a/src/sna/gen4_render.c b/src/sna/gen4_render.c
index 829f4e9a..7fc85209 100644
--- a/src/sna/gen4_render.c
+++ b/src/sna/gen4_render.c
@@ -587,6 +587,9 @@ static bool gen4_rectangle_begin(struct sna *sna,
 	int id = op->u.gen4.ve_id;
 	int ndwords;
 
+	if (sna_vertex_wait__locked(&sna->render) && sna->render.vertex_offset)
+		return true;
+
 	/* 7xpipelined pointers + 6xprimitive + 1xflush */
 	ndwords = op->need_magic_ca_pass? 20 : 6;
 	if ((sna->render.vb_id & (1 << id)) == 0)
@@ -606,6 +609,13 @@ static bool gen4_rectangle_begin(struct sna *sna,
 static int gen4_get_rectangles__flush(struct sna *sna,
 				      const struct sna_composite_op *op)
 {
+	/* Preventing discarding new vbo after lock contention */
+	if (sna_vertex_wait__locked(&sna->render)) {
+		int rem = vertex_space(sna);
+		if (rem > op->floats_per_rect)
+			return rem;
+	}
+
 	if (!kgem_check_batch(&sna->kgem, op->need_magic_ca_pass ? 25 : 6))
 		return 0;
 	if (!kgem_check_reloc_and_exec(&sna->kgem, 2))
@@ -621,17 +631,6 @@ static int gen4_get_rectangles__flush(struct sna *sna,
 						     op->u.gen4.wm_kernel);
 	}
 
-	/* Preventing discarding new vbo after lock contention */
-	if (sna->render.active) {
-		int rem;
-
-		sna_vertex_wait__locked(&sna->render);
-
-		rem = vertex_space(sna);
-		if (rem > op->floats_per_rect)
-			return rem;
-	}
-
 	return gen4_vertex_finish(sna);
 }
 
@@ -644,7 +643,7 @@ inline static int gen4_get_rectangles(struct sna *sna,
 
 start:
 	rem = vertex_space(sna);
-	if (rem < op->floats_per_rect) {
+	if (unlikely(rem < op->floats_per_rect)) {
 		DBG(("flushing vbo for %s: %d < %d\n",
 		     __FUNCTION__, rem, op->floats_per_rect));
 		rem = gen4_get_rectangles__flush(sna, op);
diff --git a/src/sna/gen5_render.c b/src/sna/gen5_render.c
index 07e95712..35f70f83 100644
--- a/src/sna/gen5_render.c
+++ b/src/sna/gen5_render.c
@@ -577,6 +577,9 @@ static bool gen5_rectangle_begin(struct sna *sna,
 	int id = op->u.gen5.ve_id;
 	int ndwords;
 
+	if (sna_vertex_wait__locked(&sna->render) && sna->render.vertex_offset)
+		return true;
+
 	ndwords = op->need_magic_ca_pass ? 20 : 6;
 	if ((sna->render.vb_id & (1 << id)) == 0)
 		ndwords += 5;
@@ -595,6 +598,13 @@ static bool gen5_rectangle_begin(struct sna *sna,
 static int gen5_get_rectangles__flush(struct sna *sna,
 				      const struct sna_composite_op *op)
 {
+	/* Preventing discarding new vbo after lock contention */
+	if (sna_vertex_wait__locked(&sna->render)) {
+		int rem = vertex_space(sna);
+		if (rem > op->floats_per_rect)
+			return rem;
+	}
+
 	if (!kgem_check_batch(&sna->kgem, op->need_magic_ca_pass ? 20 : 6))
 		return 0;
 	if (!kgem_check_reloc_and_exec(&sna->kgem, 2))
@@ -607,17 +617,6 @@ static int gen5_get_rectangles__flush(struct sna *sna,
 						     op->u.gen5.wm_kernel);
 	}
 
-	/* Preventing discarding new vbo after lock contention */
-	if (sna->render.active) {
-		int rem;
-
-		sna_vertex_wait__locked(&sna->render);
-
-		rem = vertex_space(sna);
-		if (rem > op->floats_per_rect)
-			return rem;
-	}
-
 	return gen4_vertex_finish(sna);
 }
 
@@ -631,7 +630,7 @@ inline static int gen5_get_rectangles(struct sna *sna,
 
 start:
 	rem = vertex_space(sna);
-	if (rem < op->floats_per_rect) {
+	if (unlikely(rem < op->floats_per_rect)) {
 		DBG(("flushing vbo for %s: %d < %d\n",
 		     __FUNCTION__, rem, op->floats_per_rect));
 		rem = gen5_get_rectangles__flush(sna, op);
diff --git a/src/sna/gen6_render.c b/src/sna/gen6_render.c
index a9d81114..ec642504 100644
--- a/src/sna/gen6_render.c
+++ b/src/sna/gen6_render.c
@@ -1125,6 +1125,9 @@ static bool gen6_rectangle_begin(struct sna *sna,
 	int id = 1 << GEN6_VERTEX(op->u.gen6.flags);
 	int ndwords;
 
+	if (sna_vertex_wait__locked(&sna->render) && sna->render.vertex_offset)
+		return true;
+
 	ndwords = op->need_magic_ca_pass ? 60 : 6;
 	if ((sna->render.vb_id & id) == 0)
 		ndwords += 5;
@@ -1141,6 +1144,13 @@ static bool gen6_rectangle_begin(struct sna *sna,
 static int gen6_get_rectangles__flush(struct sna *sna,
 				      const struct sna_composite_op *op)
 {
+	/* Preventing discarding new vbo after lock contention */
+	if (sna_vertex_wait__locked(&sna->render)) {
+		int rem = vertex_space(sna);
+		if (rem > op->floats_per_rect)
+			return rem;
+	}
+
 	if (!kgem_check_batch(&sna->kgem, op->need_magic_ca_pass ? 65 : 5))
 		return 0;
 	if (!kgem_check_reloc_and_exec(&sna->kgem, 2))
@@ -1157,17 +1167,6 @@ static int gen6_get_rectangles__flush(struct sna *sna,
 		}
 	}
 
-	/* Preventing discarding new vbo after lock contention */
-	if (sna->render.active) {
-		int rem;
-
-		sna_vertex_wait__locked(&sna->render);
-
-		rem = vertex_space(sna);
-		if (rem > op->floats_per_rect)
-			return rem;
-	}
-
 	return gen4_vertex_finish(sna);
 }
 
@@ -1180,7 +1179,7 @@ inline static int gen6_get_rectangles(struct sna *sna,
 
 start:
 	rem = vertex_space(sna);
-	if (rem < op->floats_per_rect) {
+	if (unlikely(rem < op->floats_per_rect)) {
 		DBG(("flushing vbo for %s: %d < %d\n",
 		     __FUNCTION__, rem, op->floats_per_rect));
 		rem = gen6_get_rectangles__flush(sna, op);
diff --git a/src/sna/gen7_render.c b/src/sna/gen7_render.c
index dc704dde..59b38f6c 100644
--- a/src/sna/gen7_render.c
+++ b/src/sna/gen7_render.c
@@ -1270,6 +1270,9 @@ static bool gen7_rectangle_begin(struct sna *sna,
 	int id = 1 << GEN7_VERTEX(op->u.gen7.flags);
 	int ndwords;
 
+	if (sna_vertex_wait__locked(&sna->render) && sna->render.vertex_offset)
+		return true;
+
 	ndwords = op->need_magic_ca_pass ? 60 : 6;
 	if ((sna->render.vb_id & id) == 0)
 		ndwords += 5;
@@ -1286,6 +1289,13 @@ static bool gen7_rectangle_begin(struct sna *sna,
 static int gen7_get_rectangles__flush(struct sna *sna,
 				      const struct sna_composite_op *op)
 {
+	/* Preventing discarding new vbo after lock contention */
+	if (sna_vertex_wait__locked(&sna->render)) {
+		int rem = vertex_space(sna);
+		if (rem > op->floats_per_rect)
+			return rem;
+	}
+
 	if (!kgem_check_batch(&sna->kgem, op->need_magic_ca_pass ? 65 : 6))
 		return 0;
 	if (!kgem_check_reloc_and_exec(&sna->kgem, 2))
@@ -1300,17 +1310,6 @@ static int gen7_get_rectangles__flush(struct sna *sna,
 		}
 	}
 
-	/* Preventing discarding new vbo after lock contention */
-	if (sna->render.active) {
-		int rem;
-
-		sna_vertex_wait__locked(&sna->render);
-
-		rem = vertex_space(sna);
-		if (rem > op->floats_per_rect)
-			return rem;
-	}
-
 	return gen4_vertex_finish(sna);
 }
 
@@ -1323,7 +1322,7 @@ inline static int gen7_get_rectangles(struct sna *sna,
 
 start:
 	rem = vertex_space(sna);
-	if (rem < op->floats_per_rect) {
+	if (unlikely(rem < op->floats_per_rect)) {
 		DBG(("flushing vbo for %s: %d < %d\n",
 		     __FUNCTION__, rem, op->floats_per_rect));
 		rem = gen7_get_rectangles__flush(sna, op);
diff --git a/src/sna/sna_render.h b/src/sna/sna_render.h
index c953e50c..6cf90eb9 100644
--- a/src/sna/sna_render.h
+++ b/src/sna/sna_render.h
@@ -763,10 +763,12 @@ static inline void sna_vertex_release__locked(struct sna_render *r)
 		pthread_cond_signal(&r->wait);
 }
 
-static inline void sna_vertex_wait__locked(struct sna_render *r)
+static inline bool sna_vertex_wait__locked(struct sna_render *r)
 {
+	bool was_active = r->active;
 	while (r->active)
 		pthread_cond_wait(&r->wait, &r->lock);
+	return was_active;
 }
 
 #endif /* SNA_RENDER_H */