[Intel-gfx] [PATCH 5/5] drm/atomic: Add encoder_mask to crtc_state, v2.

[Maarten Lankhorst]

A diff to make this v3. Found when running this through IGT bat.

- Keeping an encoder but moving it to a different crtc resulted, in encoder_mask not being updated.
- Add some paranoia when best_encoder was already updated to a different encoder in steal_encoder.
    This could happen in theory.
- Relax the WARN_ON(!crtc) when resuming after resume and add an explanation what the WARN_ON is for.
---
diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
index 1ac35072e14f..e5534dc9c4b4 100644
--- a/drivers/gpu/drm/drm_atomic_helper.c
+++ b/drivers/gpu/drm/drm_atomic_helper.c
@@ -137,7 +137,13 @@ set_best_encoder(struct drm_atomic_state *state,
 		/* Unset the encoder_mask in the old crtc state. */
 		crtc = conn_state->connector->state->crtc;
 
-		WARN_ON(!crtc);
+		/* A NULL crtc is an error here because we should have
+		 *  duplicated a NULL best_encoder when crtc was NULL.
+		 * As an exception restoring duplicated atomic state
+		 * during resume is allowed, so don't warn when
+		 * best_encoder is equal to encoder we intend to set.
+		 */
+		WARN_ON(!crtc && encoder != conn_state->best_encoder);
 		if (crtc) {
 			crtc_state = drm_atomic_get_existing_crtc_state(state, crtc);
 
@@ -199,6 +205,9 @@ steal_encoder(struct drm_atomic_state *state,
 		if (IS_ERR(connector_state))
 			return PTR_ERR(connector_state);
 
+		if (connector_state->best_encoder != encoder)
+			continue;
+
 		set_best_encoder(state, connector_state, NULL);
 	}
 
@@ -276,6 +285,8 @@ update_connector_routing(struct drm_atomic_state *state, int conn_idx)
 	}
 
 	if (new_encoder == connector_state->best_encoder) {
+		set_best_encoder(state, connector_state, new_encoder);
+
 		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] keeps [ENCODER:%d:%s], now on [CRTC:%d:%s]\n",
 				 connector->base.id,
 				 connector->name,