gallium/radeon: fix (S)DMA read-after-write hazards

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

src/gallium/drivers/r600/evergreen_hw_context.c

@@ -77,6 +77,7 @@ void evergreen_dma_copy_buffer(struct r600_context *rctx,
 		src_offset += csize << shift;
 		size -= csize;
 	}
+	r600_dma_emit_wait_idle(&rctx->b);
 }
 
 /* The max number of bytes to copy per packet. */

src/gallium/drivers/r600/evergreen_state.c

@@ -3470,6 +3470,7 @@ static void evergreen_dma_copy_tile(struct r600_context *rctx,
 		addr += cheight * pitch;
 		y += cheight;
 	}
+	r600_dma_emit_wait_idle(&rctx->b);
 }
 
 static void evergreen_dma_copy(struct pipe_context *ctx,

src/gallium/drivers/r600/r600_hw_context.c

@@ -484,4 +484,5 @@ void r600_dma_copy_buffer(struct r600_context *rctx,
 		src_offset += csize << 2;
 		size -= csize;
 	}
+	r600_dma_emit_wait_idle(&rctx->b);
 }

src/gallium/drivers/r600/r600_state.c

@@ -2941,6 +2941,7 @@ static boolean r600_dma_copy_tile(struct r600_context *rctx,
 		addr += cheight * pitch;
 		y += cheight;
 	}
+	r600_dma_emit_wait_idle(&rctx->b);
 	return TRUE;
 }
 

src/gallium/drivers/radeon/r600_pipe_common.c

@@ -149,6 +149,30 @@ void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw)
 	}
 }
 
+/* This is required to prevent read-after-write hazards. */
+void r600_dma_emit_wait_idle(struct r600_common_context *rctx)
+{
+	struct radeon_winsys_cs *cs = rctx->dma.cs;
+
+	/* done at the end of DMA calls, so increment this. */
+	rctx->num_dma_calls++;
+
+	r600_need_dma_space(rctx, 1);
+
+	if (cs->cdw == 0) /* empty queue */
+		return;
+
+	/* NOP waits for idle on Evergreen and later. */
+	if (rctx->chip_class >= CIK)
+		radeon_emit(cs, 0x00000000); /* NOP */
+	else if (rctx->chip_class >= EVERGREEN)
+		radeon_emit(cs, 0xf0000000); /* NOP */
+	else {
+		/* TODO: R600-R700 should use the FENCE packet.
+		 * CS checker support is required. */
+	}
+}
+
 static void r600_memory_barrier(struct pipe_context *ctx, unsigned flags)
 {
 }

src/gallium/drivers/radeon/r600_pipe_common.h

@@ -598,6 +598,7 @@ struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
 						  const struct pipe_resource *templ);
 const char *r600_get_llvm_processor_name(enum radeon_family family);
 void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw);
+void r600_dma_emit_wait_idle(struct r600_common_context *rctx);
 
 /* r600_gpu_load.c */
 void r600_gpu_load_kill_thread(struct r600_common_screen *rscreen);

src/gallium/drivers/radeonsi/cik_sdma.c

@@ -87,6 +87,7 @@ static void cik_sdma_copy_buffer(struct si_context *ctx,
 		       dst_offset + size);
 
 	cik_sdma_do_copy_buffer(ctx, dst, src, dst_offset, src_offset, size);
+	r600_dma_emit_wait_idle(&ctx->b);
 }
 
 static void cik_sdma_copy(struct pipe_context *ctx,

src/gallium/drivers/radeonsi/si_dma.c

@@ -82,6 +82,7 @@ static void si_dma_copy_buffer(struct si_context *ctx,
 		src_offset += csize << shift;
 		size -= csize;
 	}
+	r600_dma_emit_wait_idle(&ctx->b);
 }
 
 static void si_dma_copy_tile(struct si_context *ctx,
@@ -188,6 +189,7 @@ static void si_dma_copy_tile(struct si_context *ctx,
 		addr += cheight * pitch;
 		tiled_y += cheight;
 	}
+	r600_dma_emit_wait_idle(&ctx->b);
 }
 
 static void si_dma_copy(struct pipe_context *ctx,