drm/amd/display: Add new message for DF throttling optimization on dcn401

[WHY]
When effective bandwidth from the SoC is enough to perform SubVP
prefetchs, then DF throttling is not required.

[HOW]
Provide SMU the required clocks for which DF throttling is not required.

Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Alvin Lee <alvin.lee2@amd.com>
Signed-off-by: Dillon Varone <dillon.varone@amd.com>
Signed-off-by: Rodrigo Siqueira <rodrigo.siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/display/dc/clk_mgr/dcn401/dcn401_clk_mgr.c

@@ -141,6 +141,20 @@ static bool dcn401_is_ppclk_idle_dpm_enabled(struct clk_mgr_internal *clk_mgr, P
 	return ppclk_idle_dpm_enabled;
 }
 
+static bool dcn401_is_df_throttle_opt_enabled(struct clk_mgr_internal *clk_mgr)
+{
+	bool is_df_throttle_opt_enabled = false;
+
+	if (ASICREV_IS_GC_12_0_1_A0(clk_mgr->base.ctx->asic_id.hw_internal_rev) &&
+			clk_mgr->smu_ver >= 0x663500) {
+		is_df_throttle_opt_enabled = !clk_mgr->base.ctx->dc->debug.force_subvp_df_throttle;
+	}
+
+	is_df_throttle_opt_enabled &= clk_mgr->smu_present;
+
+	return is_df_throttle_opt_enabled;
+}
+
 /* Query SMU for all clock states for a particular clock */
 static void dcn401_init_single_clock(struct clk_mgr_internal *clk_mgr, PPCLK_e clk, unsigned int *entry_0,
 		unsigned int *num_levels)
@@ -869,6 +883,12 @@ static void dcn401_execute_block_sequence(struct clk_mgr *clk_mgr_base, unsigned
 					params->update_idle_hardmin_params.uclk_mhz,
 					params->update_idle_hardmin_params.fclk_mhz);
 			break;
+		case CLK_MGR401_UPDATE_SUBVP_HARDMINS:
+			dcn401_smu_set_subvp_uclk_fclk_hardmin(
+					clk_mgr_internal,
+					params->update_idle_hardmin_params.uclk_mhz,
+					params->update_idle_hardmin_params.fclk_mhz);
+			break;
 		case CLK_MGR401_UPDATE_DEEP_SLEEP_DCFCLK:
 			dcn401_smu_set_min_deep_sleep_dcef_clk(
 					clk_mgr_internal,
@@ -945,15 +965,21 @@ static unsigned int dcn401_build_update_bandwidth_clocks_sequence(
 	bool update_active_uclk = false;
 	bool update_idle_fclk = false;
 	bool update_idle_uclk = false;
+	bool update_subvp_prefetch_dramclk = false;
+	bool update_subvp_prefetch_fclk = false;
 	bool is_idle_dpm_enabled = dcn401_is_ppclk_dpm_enabled(clk_mgr_internal, PPCLK_UCLK) &&
 			dcn401_is_ppclk_dpm_enabled(clk_mgr_internal, PPCLK_FCLK) &&
 			dcn401_is_ppclk_idle_dpm_enabled(clk_mgr_internal, PPCLK_UCLK) &&
 			dcn401_is_ppclk_idle_dpm_enabled(clk_mgr_internal, PPCLK_FCLK);
+	bool is_df_throttle_opt_enabled = is_idle_dpm_enabled &&
+		dcn401_is_df_throttle_opt_enabled(clk_mgr_internal);
 	int total_plane_count = clk_mgr_helper_get_active_plane_cnt(dc, context);
 	int active_uclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.dramclk_khz);
 	int active_fclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.fclk_khz);
 	int idle_uclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.idle_dramclk_khz);
 	int idle_fclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.idle_fclk_khz);
+	int subvp_prefetch_dramclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.subvp_prefetch_dramclk_khz);
+	int subvp_prefetch_fclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.subvp_prefetch_fclk_khz);
 
 	unsigned int num_steps = 0;
 
@@ -1109,6 +1135,12 @@ static unsigned int dcn401_build_update_bandwidth_clocks_sequence(
 		}
 	}
 
+	if (should_set_clock(safe_to_lower, new_clocks->subvp_prefetch_dramclk_khz, clk_mgr_base->clks.subvp_prefetch_dramclk_khz)) {
+		clk_mgr_base->clks.subvp_prefetch_dramclk_khz = new_clocks->subvp_prefetch_dramclk_khz;
+		update_subvp_prefetch_dramclk = true;
+		subvp_prefetch_dramclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.subvp_prefetch_dramclk_khz);
+	}
+
 	/* FCLK */
 	/* Always update saved value, even if new value not set due to P-State switching unsupported */
 	if (should_set_clock(safe_to_lower, new_clocks->fclk_khz, clk_mgr_base->clks.fclk_khz)) {
@@ -1129,6 +1161,12 @@ static unsigned int dcn401_build_update_bandwidth_clocks_sequence(
 		}
 	}
 
+	if (should_set_clock(safe_to_lower, new_clocks->subvp_prefetch_fclk_khz, clk_mgr_base->clks.subvp_prefetch_fclk_khz)) {
+		clk_mgr_base->clks.subvp_prefetch_fclk_khz = new_clocks->subvp_prefetch_fclk_khz;
+		update_subvp_prefetch_fclk = true;
+		subvp_prefetch_fclk_mhz = khz_to_mhz_ceil(clk_mgr_base->clks.subvp_prefetch_fclk_khz);
+	}
+
 	/* When idle DPM is enabled, need to send active and idle hardmins separately */
 	/* CLK_MGR401_UPDATE_ACTIVE_HARDMINS */
 	if ((update_active_uclk || update_active_fclk) && is_idle_dpm_enabled) {
@@ -1146,6 +1184,14 @@ static unsigned int dcn401_build_update_bandwidth_clocks_sequence(
 		num_steps++;
 	}
 
+	/* CLK_MGR401_UPDATE_SUBVP_HARDMINS */
+	if ((update_subvp_prefetch_dramclk || update_subvp_prefetch_fclk) && is_df_throttle_opt_enabled) {
+		block_sequence[num_steps].params.update_idle_hardmin_params.uclk_mhz = subvp_prefetch_dramclk_mhz;
+		block_sequence[num_steps].params.update_idle_hardmin_params.fclk_mhz = subvp_prefetch_fclk_mhz;
+		block_sequence[num_steps].func = CLK_MGR401_UPDATE_SUBVP_HARDMINS;
+		num_steps++;
+	}
+
 	/* set UCLK to requested value if P-State switching is supported, or to re-enable P-State switching */
 	if (update_active_uclk || update_idle_uclk) {
 		if (!is_idle_dpm_enabled) {

drivers/gpu/drm/amd/display/dc/clk_mgr/dcn401/dcn401_clk_mgr.h

@@ -90,6 +90,7 @@ enum dcn401_clk_mgr_block_sequence_func {
 	CLK_MGR401_UPDATE_DTBCLK_DTO,
 	CLK_MGR401_UPDATE_DENTIST,
 	CLK_MGR401_UPDATE_PSR_WAIT_LOOP,
+	CLK_MGR401_UPDATE_SUBVP_HARDMINS,
 };
 
 struct dcn401_clk_mgr_block_sequence {

drivers/gpu/drm/amd/display/dc/clk_mgr/dcn401/dcn401_clk_mgr_smu_msg.c

@@ -21,6 +21,11 @@
 
 #define smu_print(str, ...) {DC_LOG_SMU(str, ##__VA_ARGS__); }
 
+/* temporary define */
+#ifndef DALSMC_MSG_SubvpUclkFclk
+#define DALSMC_MSG_SubvpUclkFclk 0x1B
+#endif
+
 /*
  * Function to be used instead of REG_WAIT macro because the wait ends when
  * the register is NOT EQUAL to zero, and because the translation in msg_if.h
@@ -296,6 +301,24 @@ bool dcn401_smu_set_active_uclk_fclk_hardmin(struct clk_mgr_internal *clk_mgr,
 	return success;
 }
 
+bool dcn401_smu_set_subvp_uclk_fclk_hardmin(struct clk_mgr_internal *clk_mgr,
+		uint16_t uclk_freq_mhz,
+		uint16_t fclk_freq_mhz)
+{
+	uint32_t response = 0;
+	bool success;
+
+	/* 15:0 for uclk, 32:16 for fclk */
+	uint32_t param = (fclk_freq_mhz << 16) | uclk_freq_mhz;
+
+	smu_print("SMU Set active hardmin by freq: uclk_freq_mhz = %d MHz, fclk_freq_mhz = %d MHz\n", uclk_freq_mhz, fclk_freq_mhz);
+
+	success = dcn401_smu_send_msg_with_param(clk_mgr,
+			DALSMC_MSG_SubvpUclkFclk, param, &response);
+
+	return success;
+}
+
 void dcn401_smu_set_min_deep_sleep_dcef_clk(struct clk_mgr_internal *clk_mgr, uint32_t freq_mhz)
 {
 	smu_print("SMU Set min deep sleep dcef clk: freq_mhz = %d MHz\n", freq_mhz);

drivers/gpu/drm/amd/display/dc/clk_mgr/dcn401/dcn401_clk_mgr_smu_msg.h

@@ -23,6 +23,9 @@ bool dcn401_smu_set_idle_uclk_fclk_hardmin(struct clk_mgr_internal *clk_mgr,
 bool dcn401_smu_set_active_uclk_fclk_hardmin(struct clk_mgr_internal *clk_mgr,
 		uint16_t uclk_freq_mhz,
 		uint16_t fclk_freq_mhz);
+bool dcn401_smu_set_subvp_uclk_fclk_hardmin(struct clk_mgr_internal *clk_mgr,
+		uint16_t uclk_freq_mhz,
+		uint16_t fclk_freq_mhz);
 void dcn401_smu_set_min_deep_sleep_dcef_clk(struct clk_mgr_internal *clk_mgr, uint32_t freq_mhz);
 void dcn401_smu_set_num_of_displays(struct clk_mgr_internal *clk_mgr, uint32_t num_displays);
 

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

@@ -629,6 +629,8 @@ struct dc_clocks {
 	int bw_dispclk_khz;
 	int idle_dramclk_khz;
 	int idle_fclk_khz;
+	int subvp_prefetch_dramclk_khz;
+	int subvp_prefetch_fclk_khz;
 };
 
 struct dc_bw_validation_profile {
@@ -1072,6 +1074,7 @@ struct dc_debug_options {
 	bool skip_full_updated_if_possible;
 	unsigned int enable_oled_edp_power_up_opt;
 	bool enable_hblank_borrow;
+	bool force_subvp_df_throttle;
 };
 
 

drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.c

@@ -1077,6 +1077,8 @@ void dml21_copy_clocks_to_dc_state(struct dml2_context *in_ctx, struct dc_state
 	context->bw_ctx.bw.dcn.clk.dtbclk_en = in_ctx->v21.mode_programming.programming->min_clocks.dcn4x.dtbrefclk_khz > 0;
 	context->bw_ctx.bw.dcn.clk.ref_dtbclk_khz = in_ctx->v21.mode_programming.programming->min_clocks.dcn4x.dtbrefclk_khz;
 	context->bw_ctx.bw.dcn.clk.socclk_khz = in_ctx->v21.mode_programming.programming->min_clocks.dcn4x.socclk_khz;
+	context->bw_ctx.bw.dcn.clk.subvp_prefetch_dramclk_khz = in_ctx->v21.mode_programming.programming->min_clocks.dcn4x.svp_prefetch_no_throttle.uclk_khz;
+	context->bw_ctx.bw.dcn.clk.subvp_prefetch_fclk_khz = in_ctx->v21.mode_programming.programming->min_clocks.dcn4x.svp_prefetch_no_throttle.fclk_khz;
 }
 
 void dml21_extract_legacy_watermark_set(const struct dc *in_dc, struct dcn_watermarks *watermark, enum dml2_dchub_watermark_reg_set_index reg_set_idx, struct dml2_context *in_ctx)

drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_types.h

@@ -387,6 +387,11 @@ struct dml2_display_cfg_programming {
 				unsigned long fclk_khz;
 				unsigned long dcfclk_khz;
 			} svp_prefetch;
+			struct {
+				unsigned long uclk_khz;
+				unsigned long fclk_khz;
+				unsigned long dcfclk_khz;
+			} svp_prefetch_no_throttle;
 
 			unsigned long deepsleep_dcfclk_khz;
 			unsigned long dispclk_khz;

drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.c

@@ -96,6 +96,7 @@ static void calculate_svp_prefetch_minimums(struct dml2_dpmm_map_mode_to_soc_dpm
 	double min_uclk_latency;
 	const struct dml2_core_mode_support_result *mode_support_result = &in_out->display_cfg->mode_support_result;
 
+	/* assumes DF throttling is enabled */
 	min_uclk_avg = dram_bw_kbps_to_uclk_khz(mode_support_result->global.svp_prefetch.average_bw_dram_kbps, &in_out->soc_bb->clk_table.dram_config);
 	min_uclk_avg = (double)min_uclk_avg / ((double)in_out->soc_bb->qos_parameters.derate_table.dcn_mall_prefetch_average.dram_derate_percent_pixel / 100);
 
@@ -125,6 +126,37 @@ static void calculate_svp_prefetch_minimums(struct dml2_dpmm_map_mode_to_soc_dpm
 	in_out->programming->min_clocks.dcn4x.svp_prefetch.uclk_khz = dml_round_up(min_uclk_bw > min_uclk_latency ? min_uclk_bw : min_uclk_latency);
 	in_out->programming->min_clocks.dcn4x.svp_prefetch.fclk_khz = dml_round_up(min_fclk_bw > min_fclk_latency ? min_fclk_bw : min_fclk_latency);
 	in_out->programming->min_clocks.dcn4x.svp_prefetch.dcfclk_khz = dml_round_up(min_dcfclk_bw > min_dcfclk_latency ? min_dcfclk_bw : min_dcfclk_latency);
+
+	/* assumes DF throttling is disabled */
+	min_uclk_avg = dram_bw_kbps_to_uclk_khz(mode_support_result->global.svp_prefetch.average_bw_dram_kbps, &in_out->soc_bb->clk_table.dram_config);
+	min_uclk_avg = (double)min_uclk_avg / ((double)in_out->soc_bb->qos_parameters.derate_table.system_active_average.dram_derate_percent_pixel / 100);
+
+	min_uclk_urgent = dram_bw_kbps_to_uclk_khz(mode_support_result->global.svp_prefetch.urgent_bw_dram_kbps, &in_out->soc_bb->clk_table.dram_config);
+	min_uclk_urgent = (double)min_uclk_urgent / ((double)in_out->soc_bb->qos_parameters.derate_table.system_active_urgent.dram_derate_percent_pixel / 100);
+
+	min_uclk_bw = min_uclk_urgent > min_uclk_avg ? min_uclk_urgent : min_uclk_avg;
+
+	min_fclk_avg = (double)mode_support_result->global.svp_prefetch.average_bw_sdp_kbps / in_out->soc_bb->fabric_datapath_to_dcn_data_return_bytes;
+	min_fclk_avg = (double)min_fclk_avg / ((double)in_out->soc_bb->qos_parameters.derate_table.system_active_average.fclk_derate_percent / 100);
+
+	min_fclk_urgent = (double)mode_support_result->global.svp_prefetch.urgent_bw_sdp_kbps / in_out->soc_bb->fabric_datapath_to_dcn_data_return_bytes;
+	min_fclk_urgent = (double)min_fclk_urgent / ((double)in_out->soc_bb->qos_parameters.derate_table.system_active_urgent.fclk_derate_percent / 100);
+
+	min_fclk_bw = min_fclk_urgent > min_fclk_avg ? min_fclk_urgent : min_fclk_avg;
+
+	min_dcfclk_avg = (double)mode_support_result->global.svp_prefetch.average_bw_sdp_kbps / in_out->soc_bb->return_bus_width_bytes;
+	min_dcfclk_avg = (double)min_dcfclk_avg / ((double)in_out->soc_bb->qos_parameters.derate_table.system_active_average.dcfclk_derate_percent / 100);
+
+	min_dcfclk_urgent = (double)mode_support_result->global.svp_prefetch.urgent_bw_sdp_kbps / in_out->soc_bb->return_bus_width_bytes;
+	min_dcfclk_urgent = (double)min_dcfclk_urgent / ((double)in_out->soc_bb->qos_parameters.derate_table.system_active_urgent.dcfclk_derate_percent / 100);
+
+	min_dcfclk_bw = min_dcfclk_urgent > min_dcfclk_avg ? min_dcfclk_urgent : min_dcfclk_avg;
+
+	get_minimum_clocks_for_latency(in_out, &min_uclk_latency, &min_fclk_latency, &min_dcfclk_latency);
+
+	in_out->programming->min_clocks.dcn4x.svp_prefetch_no_throttle.uclk_khz = dml_round_up(min_uclk_bw > min_uclk_latency ? min_uclk_bw : min_uclk_latency);
+	in_out->programming->min_clocks.dcn4x.svp_prefetch_no_throttle.fclk_khz = dml_round_up(min_fclk_bw > min_fclk_latency ? min_fclk_bw : min_fclk_latency);
+	in_out->programming->min_clocks.dcn4x.svp_prefetch_no_throttle.dcfclk_khz = dml_round_up(min_dcfclk_bw > min_dcfclk_latency ? min_dcfclk_bw : min_dcfclk_latency);
 }
 
 static void calculate_idle_minimums(struct dml2_dpmm_map_mode_to_soc_dpm_params_in_out *in_out)
@@ -272,6 +304,17 @@ static bool map_soc_min_clocks_to_dpm_fine_grained(struct dml2_display_cfg_progr
 	if (result)
 		result = round_up_to_next_dpm(&display_cfg->min_clocks.dcn4x.idle.uclk_khz, &state_table->uclk);
 
+	/* these clocks are optional, so they can fail to map, in which case map all to 0 */
+	if (result) {
+		if (!round_up_to_next_dpm(&display_cfg->min_clocks.dcn4x.svp_prefetch_no_throttle.dcfclk_khz, &state_table->dcfclk) ||
+				!round_up_to_next_dpm(&display_cfg->min_clocks.dcn4x.svp_prefetch_no_throttle.fclk_khz, &state_table->fclk) ||
+				!round_up_to_next_dpm(&display_cfg->min_clocks.dcn4x.svp_prefetch_no_throttle.uclk_khz, &state_table->uclk)) {
+			display_cfg->min_clocks.dcn4x.svp_prefetch_no_throttle.dcfclk_khz = 0;
+			display_cfg->min_clocks.dcn4x.svp_prefetch_no_throttle.fclk_khz = 0;
+			display_cfg->min_clocks.dcn4x.svp_prefetch_no_throttle.uclk_khz = 0;
+		}
+	}
+
 	return result;
 }