diff --git a/src/gallium/drivers/freedreno/freedreno_context.c b/src/gallium/drivers/freedreno/freedreno_context.c
index 52557d16068..65c38913c22 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.c
+++ b/src/gallium/drivers/freedreno/freedreno_context.c
@@ -244,6 +244,8 @@ fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
 	 */
 	ctx->sample_mask = 0xffff;
 
+	ctx->stage = FD_STAGE_NULL;
+
 	pctx = &ctx->base;
 	pctx->screen = pscreen;
 	pctx->priv = priv;
diff --git a/src/gallium/drivers/freedreno/freedreno_context.h b/src/gallium/drivers/freedreno/freedreno_context.h
index 438c2e27035..53b4e1dbaa7 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.h
+++ b/src/gallium/drivers/freedreno/freedreno_context.h
@@ -116,18 +116,19 @@ struct fd_vertex_state {
  * is active across IB's (or between tile IB and draw IB)
  */
 enum fd_render_stage {
-	FD_STAGE_NULL     = 0x00,
-	FD_STAGE_DRAW     = 0x01,
-	FD_STAGE_CLEAR    = 0x02,
+	FD_STAGE_NULL     = 0x01,
+	FD_STAGE_DRAW     = 0x02,
+	FD_STAGE_CLEAR    = 0x04,
 	/* TODO before queries which include MEM2GMEM or GMEM2MEM will
 	 * work we will need to call fd_hw_query_prepare() from somewhere
 	 * appropriate so that queries in the tiling IB get backed with
 	 * memory to write results to.
 	 */
-	FD_STAGE_MEM2GMEM = 0x04,
-	FD_STAGE_GMEM2MEM = 0x08,
+	FD_STAGE_MEM2GMEM = 0x08,
+	FD_STAGE_GMEM2MEM = 0x10,
 	/* used for driver internal draws (ie. util_blitter_blit()): */
-	FD_STAGE_BLIT     = 0x10,
+	FD_STAGE_BLIT     = 0x20,
+	FD_STAGE_ALL      = 0xff,
 };
 
 #define MAX_HW_SAMPLE_PROVIDERS 4
diff --git a/src/gallium/drivers/freedreno/freedreno_query_hw.c b/src/gallium/drivers/freedreno/freedreno_query_hw.c
index 95fff3f108e..141dc8a28fe 100644
--- a/src/gallium/drivers/freedreno/freedreno_query_hw.c
+++ b/src/gallium/drivers/freedreno/freedreno_query_hw.c
@@ -66,6 +66,7 @@ get_sample(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	if (!ctx->sample_cache[idx]) {
 		ctx->sample_cache[idx] =
 			ctx->sample_providers[idx]->get_sample(ctx, ring);
+		ctx->needs_flush = true;
 	}
 
 	fd_hw_sample_reference(ctx, &samp, ctx->sample_cache[idx]);