diff --git a/.pick_status.json b/.pick_status.json
index d16a0cad2d6..db366f2b2fe 100644
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -594,7 +594,7 @@
         "description": "ir3: Align TCS per-patch output to 64 bytes to prevent stale reads",
         "nominated": true,
         "nomination_type": 1,
-        "resolution": 0,
+        "resolution": 1,
         "main_sha": null,
         "because_sha": null,
         "notes": null
diff --git a/src/freedreno/ir3/ir3_nir_lower_tess.c b/src/freedreno/ir3/ir3_nir_lower_tess.c
index 5e3c706079d..9186a1caed9 100644
--- a/src/freedreno/ir3/ir3_nir_lower_tess.c
+++ b/src/freedreno/ir3/ir3_nir_lower_tess.c
@@ -665,7 +665,13 @@ ir3_nir_lower_tess_ctrl(nir_shader *shader, struct ir3_shader_variant *v,
 
    build_primitive_map(shader, &state.map);
    memcpy(v->output_loc, state.map.loc, sizeof(v->output_loc));
-   v->output_size = state.map.stride;
+
+   /* Empirically, TCS outputs have to be aligned to 64 bytes,
+    * otherwise stale data may be read in rare cases. The exact
+    * reason is not clear, but tests and proprietary driver behavior
+    * strongly point at the need for 64 byte alignment.
+    */
+   v->output_size = ALIGN_POT(state.map.stride, 16);
 
    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
    assert(impl);