drm/amd/display: delay fp2 programming until vactive before lock

[Why]
race condition of programming FP2 wrt pipe locking
and vactive state can cause underflow/black screen

[How]
Enforce the FP2 is only programmed during vactive,
and unlock pipe soon afterwards.
Signed-off-by: Martin Leung <martin.leung@amd.com>
Reviewed-by: Jun Lei <Jun.Lei@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c

@@ -2736,7 +2736,7 @@ static void dcn10_program_all_pipe_in_tree(
 				pipe_ctx->pipe_dlg_param.vupdate_width);
 
 		pipe_ctx->stream_res.tg->funcs->set_vtg_params(
-				pipe_ctx->stream_res.tg, &pipe_ctx->stream->timing);
+				pipe_ctx->stream_res.tg, &pipe_ctx->stream->timing, true);
 
 		if (hws->funcs.setup_vupdate_interrupt)
 			hws->funcs.setup_vupdate_interrupt(dc, pipe_ctx);

drivers/gpu/drm/amd/display/dc/dcn10/dcn10_optc.c

@@ -272,7 +272,7 @@ void optc1_program_timing(
 			vupdate_offset,
 			vupdate_width);
 
-	optc->funcs->set_vtg_params(optc, dc_crtc_timing);
+	optc->funcs->set_vtg_params(optc, dc_crtc_timing, true);
 
 	/* TODO
 	 * patched_crtc_timing.flags.HORZ_COUNT_BY_TWO == 1
@@ -312,7 +312,7 @@ void optc1_program_timing(
 }
 
 void optc1_set_vtg_params(struct timing_generator *optc,
-		const struct dc_crtc_timing *dc_crtc_timing)
+		const struct dc_crtc_timing *dc_crtc_timing, bool program_fp2)
 {
 	struct dc_crtc_timing patched_crtc_timing;
 	uint32_t asic_blank_end;
@@ -348,9 +348,12 @@ void optc1_set_vtg_params(struct timing_generator *optc,
 		}
 	}
 
-	REG_UPDATE_2(CONTROL,
-			VTG0_FP2, v_fp2,
-			VTG0_VCOUNT_INIT, v_init);
+	if (program_fp2)
+		REG_UPDATE_2(CONTROL,
+				VTG0_FP2, v_fp2,
+				VTG0_VCOUNT_INIT, v_init);
+	else
+		REG_UPDATE(CONTROL, VTG0_VCOUNT_INIT, v_init);
 }
 
 void optc1_set_blank_data_double_buffer(struct timing_generator *optc, bool enable)

drivers/gpu/drm/amd/display/dc/dcn10/dcn10_optc.h

@@ -700,6 +700,6 @@ bool optc1_get_crc(struct timing_generator *optc,
 bool optc1_is_two_pixels_per_containter(const struct dc_crtc_timing *timing);
 
 void optc1_set_vtg_params(struct timing_generator *optc,
-		const struct dc_crtc_timing *dc_crtc_timing);
+		const struct dc_crtc_timing *dc_crtc_timing, bool program_fp2);
 
 #endif /* __DC_TIMING_GENERATOR_DCN10_H__ */

drivers/gpu/drm/amd/display/dc/dcn20/dcn20_hwseq.c

@@ -1214,6 +1214,17 @@ void dcn20_pipe_control_lock(
 		!flip_immediate)
 	    dcn20_setup_gsl_group_as_lock(dc, pipe, false);
 
+
+	// If changing VTG FP2: wait until back in vactive to program FP2
+	// Need to ensure that pipe unlock happens soon after to minimize race condition
+	if (!lock && pipe->update_flags.bits.global_sync) {
+		pipe->stream_res.tg->funcs->wait_for_state(pipe->stream_res.tg, CRTC_STATE_VBLANK);
+		pipe->stream_res.tg->funcs->wait_for_state(pipe->stream_res.tg, CRTC_STATE_VACTIVE);
+
+		pipe->stream_res.tg->funcs->set_vtg_params(
+				pipe->stream_res.tg, &pipe->stream->timing, true);
+	}
+
 	if (pipe->stream && should_use_dmub_lock(pipe->stream->link)) {
 		union dmub_hw_lock_flags hw_locks = { 0 };
 		struct dmub_hw_lock_inst_flags inst_flags = { 0 };
@@ -1595,7 +1606,7 @@ static void dcn20_program_pipe(
 				pipe_ctx->pipe_dlg_param.vupdate_width);
 
 		pipe_ctx->stream_res.tg->funcs->set_vtg_params(
-				pipe_ctx->stream_res.tg, &pipe_ctx->stream->timing);
+				pipe_ctx->stream_res.tg, &pipe_ctx->stream->timing, false);
 
 		if (hws->funcs.setup_vupdate_interrupt)
 			hws->funcs.setup_vupdate_interrupt(dc, pipe_ctx);
@@ -1695,14 +1706,6 @@ void dcn20_program_front_end_for_ctx(
 				&& context->res_ctx.pipe_ctx[i].stream)
 			hws->funcs.blank_pixel_data(dc, &context->res_ctx.pipe_ctx[i], true);
 
-	/* wait for outstanding pending changes before adding or removing planes */
-	for (i = 0; i < dc->res_pool->pipe_count; i++) {
-		if (context->res_ctx.pipe_ctx[i].update_flags.bits.disable ||
-				context->res_ctx.pipe_ctx[i].update_flags.bits.enable) {
-			dc->hwss.wait_for_pending_cleared(dc, context);
-			break;
-		}
-	}
 
 	/* Disconnect mpcc */
 	for (i = 0; i < dc->res_pool->pipe_count; i++)
@@ -1856,7 +1859,7 @@ bool dcn20_update_bandwidth(
 					pipe_ctx->pipe_dlg_param.vupdate_width);
 
 			pipe_ctx->stream_res.tg->funcs->set_vtg_params(
-					pipe_ctx->stream_res.tg, &pipe_ctx->stream->timing);
+					pipe_ctx->stream_res.tg, &pipe_ctx->stream->timing, false);
 
 			if (pipe_ctx->prev_odm_pipe == NULL)
 				hws->funcs.blank_pixel_data(dc, pipe_ctx, blank);

drivers/gpu/drm/amd/display/dc/inc/hw/timing_generator.h

@@ -271,7 +271,7 @@ struct timing_generator_funcs {
 			struct dc_crtc_timing *hw_crtc_timing);
 
 	void (*set_vtg_params)(struct timing_generator *optc,
-			const struct dc_crtc_timing *dc_crtc_timing);
+			const struct dc_crtc_timing *dc_crtc_timing, bool program_fp2);
 
 	void (*set_dsc_config)(struct timing_generator *optc,
 			       enum optc_dsc_mode dsc_mode,