diff --git a/libdrm/nouveau/nouveau_bo.c b/libdrm/nouveau/nouveau_bo.c
index b7e6d86..85fc14f 100644
--- a/libdrm/nouveau/nouveau_bo.c
+++ b/libdrm/nouveau/nouveau_bo.c
@@ -607,6 +607,7 @@ nouveau_bo_emit_buffer(struct nouveau_channel *chan, struct nouveau_bo *bo)
 	pbbo = nvpb->buffers + nvpb->nr_buffers++;
 	nvbo->pending = pbbo;
 	nvbo->pending_channel = chan;
+	nvbo->pending_refcnt = 0;
 
 	nouveau_bo_ref(bo, &ref);
 	pbbo->user_priv = (uint64_t)(unsigned long)ref;
diff --git a/libdrm/nouveau/nouveau_private.h b/libdrm/nouveau/nouveau_private.h
index 9ce87fb..784afc9 100644
--- a/libdrm/nouveau/nouveau_private.h
+++ b/libdrm/nouveau/nouveau_private.h
@@ -52,6 +52,9 @@ struct nouveau_pushbuf_priv {
 	unsigned *pushbuf;
 	unsigned  size;
 
+	unsigned marker;
+	unsigned marker_relocs;
+
 	struct drm_nouveau_gem_pushbuf_bo *buffers;
 	unsigned nr_buffers;
 	struct drm_nouveau_gem_pushbuf_reloc *relocs;
@@ -99,6 +102,7 @@ struct nouveau_bo_priv {
 	/* Tracking */
 	struct drm_nouveau_gem_pushbuf_bo *pending;
 	struct nouveau_channel *pending_channel;
+	int pending_refcnt;
 	int write_marker;
 
 	/* Userspace object */
diff --git a/libdrm/nouveau/nouveau_pushbuf.c b/libdrm/nouveau/nouveau_pushbuf.c
index d434a5f..90250c0 100644
--- a/libdrm/nouveau/nouveau_pushbuf.c
+++ b/libdrm/nouveau/nouveau_pushbuf.c
@@ -60,17 +60,17 @@ nouveau_pushbuf_emit_reloc(struct nouveau_channel *chan, void *ptr,
 			   uint32_t flags, uint32_t vor, uint32_t tor)
 {
 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(chan->pushbuf);
+	struct nouveau_bo_priv *nvbo = nouveau_bo(bo);
 	struct drm_nouveau_gem_pushbuf_reloc *r;
 	struct drm_nouveau_gem_pushbuf_bo *pbbo;
 	uint32_t domains = 0;
 
 	if (nvpb->nr_relocs >= NOUVEAU_GEM_MAX_RELOCS) {
 		fprintf(stderr, "too many relocs!!\n");
-		assert(0);
 		return -ENOMEM;
 	}
 
-	if (nouveau_bo(bo)->user && (flags & NOUVEAU_BO_WR)) {
+	if (nvbo->user && (flags & NOUVEAU_BO_WR)) {
 		fprintf(stderr, "write to user buffer!!\n");
 		return -EINVAL;
 	}
@@ -78,16 +78,21 @@ nouveau_pushbuf_emit_reloc(struct nouveau_channel *chan, void *ptr,
 	pbbo = nouveau_bo_emit_buffer(chan, bo);
 	if (!pbbo) {
 		fprintf(stderr, "buffer emit fail :(\n");
-		assert(0);
 		return -ENOMEM;
 	}
 
+	nvbo->pending_refcnt++;
+
 	if (flags & NOUVEAU_BO_VRAM)
 		domains |= NOUVEAU_GEM_DOMAIN_VRAM;
 	if (flags & NOUVEAU_BO_GART)
 		domains |= NOUVEAU_GEM_DOMAIN_GART;
+
+	if (!(pbbo->valid_domains & domains)) {
+		fprintf(stderr, "no valid domains remain!\n");
+		return -EINVAL;
+	}
 	pbbo->valid_domains &= domains;
-	assert(pbbo->valid_domains);
 
 	assert(flags & NOUVEAU_BO_RDWR);
 	if (flags & NOUVEAU_BO_RD) {
@@ -95,7 +100,7 @@ nouveau_pushbuf_emit_reloc(struct nouveau_channel *chan, void *ptr,
 	}
 	if (flags & NOUVEAU_BO_WR) {
 		pbbo->write_domains |= domains;
-		nouveau_bo(bo)->write_marker = 1;
+		nvbo->write_marker = 1;
 	}
 
 	r = nvpb->relocs + nvpb->nr_relocs++;
@@ -322,18 +327,25 @@ restart_push:
 	/* Update presumed offset/domain for any buffers that moved.
 	 * Dereference all buffers on validate list
 	 */
-	for (i = 0; i < nvpb->nr_buffers; i++) {
-		struct drm_nouveau_gem_pushbuf_bo *pbbo = &nvpb->buffers[i];
+	for (i = 0; i < nvpb->nr_relocs; i++) {
+		struct drm_nouveau_gem_pushbuf_reloc *r = &nvpb->relocs[i];
+		struct drm_nouveau_gem_pushbuf_bo *pbbo =
+			&nvpb->buffers[r->bo_index];
 		struct nouveau_bo *bo = (void *)(unsigned long)pbbo->user_priv;
+		struct nouveau_bo_priv *nvbo = nouveau_bo(bo);
+
+		if (--nvbo->pending_refcnt)
+			continue;
 
 		if (pbbo->presumed_ok == 0) {
-			nouveau_bo(bo)->domain = pbbo->presumed_domain;
-			nouveau_bo(bo)->offset = pbbo->presumed_offset;
+			nvbo->domain = pbbo->presumed_domain;
+			nvbo->offset = pbbo->presumed_offset;
 		}
 
-		nouveau_bo(bo)->pending = NULL;
+		nvbo->pending = NULL;
 		nouveau_bo_ref(NULL, &bo);
 	}
+
 	nvpb->nr_buffers = 0;
 	nvpb->nr_relocs = 0;
 
@@ -343,6 +355,57 @@ restart_push:
 	if (chan->flush_notify)
 		chan->flush_notify(chan);
 
+	nvpb->marker = 0;
 	return ret;
 }
 
+int
+nouveau_pushbuf_marker_emit(struct nouveau_channel *chan,
+			    unsigned wait_dwords, unsigned wait_relocs)
+{
+	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(chan->pushbuf);
+
+	if (AVAIL_RING(chan) < wait_dwords)
+		return nouveau_pushbuf_flush(chan, wait_dwords);
+
+	if (nvpb->nr_relocs + wait_relocs >= NOUVEAU_GEM_MAX_RELOCS)
+		return nouveau_pushbuf_flush(chan, wait_dwords);
+
+	nvpb->marker = nvpb->base.cur - nvpb->pushbuf;
+	nvpb->marker_relocs = nvpb->nr_relocs;
+	return 0;
+}
+
+void
+nouveau_pushbuf_marker_undo(struct nouveau_channel *chan)
+{
+	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(chan->pushbuf);
+	unsigned i;
+
+	if (!nvpb->marker)
+		return;
+
+	/* undo any relocs/buffers added to the list since last marker */
+	for (i = nvpb->marker_relocs; i < nvpb->nr_relocs; i++) {
+		struct drm_nouveau_gem_pushbuf_reloc *r = &nvpb->relocs[i];
+		struct drm_nouveau_gem_pushbuf_bo *pbbo =
+			&nvpb->buffers[r->bo_index];
+		struct nouveau_bo *bo = (void *)(unsigned long)pbbo->user_priv;
+		struct nouveau_bo_priv *nvbo = nouveau_bo(bo);
+
+		if (--nvbo->pending_refcnt)
+			continue;
+
+		nvbo->pending = NULL;
+		nouveau_bo_ref(NULL, &bo);
+		nvpb->nr_buffers--;
+	}
+	nvpb->nr_relocs = nvpb->marker_relocs;
+
+	/* reset pushbuf back to last marker */
+	nvpb->base.cur = nvpb->pushbuf + nvpb->marker;
+	nvpb->base.remaining = nvpb->size - nvpb->marker;
+	nvpb->marker = 0;
+}
+
+
diff --git a/libdrm/nouveau/nouveau_pushbuf.h b/libdrm/nouveau/nouveau_pushbuf.h
index 3c746ed..c7ac8c4 100644
--- a/libdrm/nouveau/nouveau_pushbuf.h
+++ b/libdrm/nouveau/nouveau_pushbuf.h
@@ -40,11 +40,30 @@ int
 nouveau_pushbuf_flush(struct nouveau_channel *, unsigned min);
 
 int
+nouveau_pushbuf_marker_emit(struct nouveau_channel *chan,
+			    unsigned wait_dwords, unsigned wait_relocs);
+
+void
+nouveau_pushbuf_marker_undo(struct nouveau_channel *chan);
+
+int
 nouveau_pushbuf_emit_reloc(struct nouveau_channel *, void *ptr,
 			   struct nouveau_bo *, uint32_t data, uint32_t data2,
 			   uint32_t flags, uint32_t vor, uint32_t tor);
 
 /* Push buffer access macros */
+static __inline__ int
+MARK_RING(struct nouveau_channel *chan, unsigned dwords, unsigned relocs)
+{
+	return nouveau_pushbuf_marker_emit(chan, dwords, relocs);
+}
+
+static __inline__ void
+MARK_UNDO(struct nouveau_channel *chan)
+{
+	nouveau_pushbuf_marker_undo(chan);
+}
+
 static __inline__ void
 OUT_RING(struct nouveau_channel *chan, unsigned data)
 {
@@ -116,62 +135,62 @@ BIND_RING(struct nouveau_channel *chan, struct nouveau_grobj *gr, unsigned sc)
 	OUT_RING  (chan, gr->handle);
 }
 
-static __inline__ void
+static __inline__ int
 OUT_RELOC(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	  unsigned data, unsigned flags, unsigned vor, unsigned tor)
 {
-	nouveau_pushbuf_emit_reloc(chan, chan->pushbuf->cur++, bo,
-				   data, 0, flags, vor, tor);
+	return nouveau_pushbuf_emit_reloc(chan, chan->pushbuf->cur++, bo,
+					  data, 0, flags, vor, tor);
 }
 
-static __inline__ void
+static __inline__ int
 OUT_RELOC2(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	   unsigned data, unsigned data2, unsigned flags,
 	   unsigned vor, unsigned tor)
 {
-	nouveau_pushbuf_emit_reloc(chan, chan->pushbuf->cur++, bo,
-				   data, data2, flags, vor, tor);
+	return nouveau_pushbuf_emit_reloc(chan, chan->pushbuf->cur++, bo,
+					  data, data2, flags, vor, tor);
 }
 
 /* Raw data + flags depending on FB/TT buffer */
-static __inline__ void
+static __inline__ int
 OUT_RELOCd(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	   unsigned data, unsigned flags, unsigned vor, unsigned tor)
 {
-	OUT_RELOC(chan, bo, data, flags | NOUVEAU_BO_OR, vor, tor);
+	return OUT_RELOC(chan, bo, data, flags | NOUVEAU_BO_OR, vor, tor);
 }
 
 /* FB/TT object handle */
-static __inline__ void
+static __inline__ int
 OUT_RELOCo(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	   unsigned flags)
 {
-	OUT_RELOC(chan, bo, 0, flags | NOUVEAU_BO_OR,
-		  chan->vram->handle, chan->gart->handle);
+	return OUT_RELOC(chan, bo, 0, flags | NOUVEAU_BO_OR,
+			 chan->vram->handle, chan->gart->handle);
 }
 
 /* Low 32-bits of offset */
-static __inline__ void
+static __inline__ int
 OUT_RELOCl(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	   unsigned delta, unsigned flags)
 {
-	OUT_RELOC(chan, bo, delta, flags | NOUVEAU_BO_LOW, 0, 0);
+	return OUT_RELOC(chan, bo, delta, flags | NOUVEAU_BO_LOW, 0, 0);
 }
 
 /* Low 32-bits of offset + GPU linear access range info */
-static __inline__ void
+static __inline__ int
 OUT_RELOCr(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	   unsigned delta, unsigned size, unsigned flags)
 {
-	OUT_RELOC2(chan, bo, delta, size, flags | NOUVEAU_BO_LOW, 0, 0);
+	return OUT_RELOC2(chan, bo, delta, size, flags | NOUVEAU_BO_LOW, 0, 0);
 }
 
 /* High 32-bits of offset */
-static __inline__ void
+static __inline__ int
 OUT_RELOCh(struct nouveau_channel *chan, struct nouveau_bo *bo,
 	   unsigned delta, unsigned flags)
 {
-	OUT_RELOC(chan, bo, delta, flags | NOUVEAU_BO_HIGH, 0, 0);
+	return OUT_RELOC(chan, bo, delta, flags | NOUVEAU_BO_HIGH, 0, 0);
 }
 
 #endif