diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c
index ceb3b3a841d..3bb3895bf73 100644
--- a/src/gallium/drivers/r600/evergreen_compute.c
+++ b/src/gallium/drivers/r600/evergreen_compute.c
@@ -376,7 +376,7 @@ static void compute_emit_cs(struct r600_context *ctx)
 	}
 #endif
 
-	ctx->ws->cs_flush(ctx->cs, RADEON_FLUSH_ASYNC);
+	ctx->ws->cs_flush(ctx->cs, RADEON_FLUSH_ASYNC | RADEON_FLUSH_COMPUTE);
 
 	ctx->pm4_dirty_cdwords = 0;
 	ctx->flags = 0;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index b598e6cf281..186c5bce9ca 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -81,6 +81,8 @@
 
 /* The first dword of RADEON_CHUNK_ID_FLAGS is a uint32 of these flags: */
 #define RADEON_CS_KEEP_TILING_FLAGS 0x01
+
+
 #endif
 
 #ifndef RADEON_CS_USE_VM
@@ -118,7 +120,7 @@ static boolean radeon_init_cs_context(struct radeon_cs_context *csc,
     csc->chunks[1].length_dw = 0;
     csc->chunks[1].chunk_data = (uint64_t)(uintptr_t)csc->relocs;
     csc->chunks[2].chunk_id = RADEON_CHUNK_ID_FLAGS;
-    csc->chunks[2].length_dw = 1;
+    csc->chunks[2].length_dw = 2;
     csc->chunks[2].chunk_data = (uint64_t)(uintptr_t)&csc->flags;
 
     csc->chunk_array[0] = (uint64_t)(uintptr_t)&csc->chunks[0];
@@ -454,15 +456,20 @@ static void radeon_drm_cs_flush(struct radeon_winsys_cs *rcs, unsigned flags)
             p_atomic_inc(&cs->cst->relocs_bo[i]->num_active_ioctls);
         }
 
-        cs->cst->flags = 0;
+        cs->cst->flags[0] = 0;
+        cs->cst->flags[1] = RADEON_CS_RING_GFX;
         cs->cst->cs.num_chunks = 2;
         if (flags & RADEON_FLUSH_KEEP_TILING_FLAGS) {
-            cs->cst->flags |= RADEON_CS_KEEP_TILING_FLAGS;
+            cs->cst->flags[0] |= RADEON_CS_KEEP_TILING_FLAGS;
             cs->cst->cs.num_chunks = 3;
         }
         if (cs->ws->info.r600_virtual_address) {
+            cs->cst->flags[0] |= RADEON_CS_USE_VM;
+            cs->cst->cs.num_chunks = 3;
+        }
+        if (flags & RADEON_FLUSH_COMPUTE) {
+            cs->cst->flags[1] = RADEON_CS_RING_COMPUTE;
             cs->cst->cs.num_chunks = 3;
-            cs->cst->flags |= RADEON_CS_USE_VM;
         }
 
         if (cs->thread &&
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.h b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
index 05b9a487645..5ce000215e1 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
@@ -37,7 +37,7 @@ struct radeon_cs_context {
     struct drm_radeon_cs        cs;
     struct drm_radeon_cs_chunk  chunks[3];
     uint64_t                    chunk_array[3];
-    uint32_t                    flags;
+    uint32_t                    flags[2];
 
     /* Relocs. */
     unsigned                    nrelocs;
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index 73160b63a27..4aa7cfc5f61 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -45,8 +45,9 @@
 
 #define RADEON_MAX_CMDBUF_DWORDS (16 * 1024)
 
-#define RADEON_FLUSH_ASYNC             (1 << 0)
-#define RADEON_FLUSH_KEEP_TILING_FLAGS (1 << 1) /* needs DRM 2.12.0 */
+#define RADEON_FLUSH_ASYNC		(1 << 0)
+#define RADEON_FLUSH_KEEP_TILING_FLAGS	(1 << 1) /* needs DRM 2.12.0 */
+#define RADEON_FLUSH_COMPUTE		(1 << 2)
 
 /* Tiling flags. */
 enum radeon_bo_layout {