[PATCH 20/28] drm/amd/display: DML21 Fixes

[Ray Wu <ray.wu@amd.com>]

From: Austin Zheng <Austin.Zheng@amd.com>

- Store state related info inside mode_lib.
- Fix bad DCFCLK deep sleep
- Update FAMS structure in DMUB header

Reviewed-by: Alvin Lee <alvin.lee2@amd.com>
Signed-off-by: Austin Zheng <Austin.Zheng@amd.com>
Signed-off-by: Ray Wu <ray.wu@amd.com>
---
 .../amd/display/dc/dml2/dml21/inc/dml_top.h   |  1 -
 .../dml21/inc/dml_top_soc_parameter_types.h   |  1 +
 .../display/dc/dml2/dml21/inc/dml_top_types.h | 10 ++-
 .../src/dml2_core/dml2_core_dcn4_calcs.c      |  7 +-
 .../src/dml2_core/dml2_core_shared_types.h    | 82 ++++++++++++++++++-
 .../dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.c |  3 -
 .../src/inc/dml2_internal_shared_types.h      | 10 ++-
 .../gpu/drm/amd/display/dmub/inc/dmub_cmd.h   | 18 +++-
 8 files changed, 117 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top.h
index c047d56527c4..a64ec4dcf11a 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top.h
+++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top.h
@@ -43,5 +43,4 @@ bool dml2_build_mode_programming(struct dml2_build_mode_programming_in_out *in_o
  */
 bool dml2_build_mcache_programming(struct dml2_build_mcache_programming_in_out *in_out);
 
-
 #endif
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_soc_parameter_types.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_soc_parameter_types.h
index 5f0bc42d1d2f..047f47e48ec1 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_soc_parameter_types.h
+++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_soc_parameter_types.h
@@ -130,6 +130,7 @@ struct dml2_soc_state_table {
 
 struct dml2_soc_vmin_clock_limits {
 	unsigned long dispclk_khz;
+	unsigned long dcfclk_khz;
 };
 
 struct dml2_soc_bb {
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_types.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_types.h
index 0dbf886d8926..bcb99a155011 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_types.h
+++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_types.h
@@ -279,7 +279,10 @@ struct dml2_per_stream_programming {
 	} phantom_stream;
 
 	union dmub_cmd_fams2_config fams2_base_params;
-	union dmub_cmd_fams2_config fams2_sub_params;
+	union {
+		union dmub_cmd_fams2_config fams2_sub_params;
+		union dmub_fams2_stream_static_sub_state_v2 fams2_sub_params_v2;
+	};
 };
 
 //-----------------
@@ -674,9 +677,14 @@ struct dml2_display_cfg_programming {
 		// unlimited # of mcache
 		struct dml2_mcache_surface_allocation non_optimized_mcache_allocation[DML2_MAX_PLANES];
 
+		bool failed_prefetch;
+		bool failed_uclk_pstate;
 		bool failed_mcache_validation;
 		bool failed_dpmm;
 		bool failed_mode_programming;
+		bool failed_mode_programming_dcfclk;
+		bool failed_mode_programming_prefetch;
+		bool failed_mode_programming_flip;
 		bool failed_map_watermarks;
 	} informative;
 };
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4_calcs.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4_calcs.c
index 6f71b7221384..71789b7783d7 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4_calcs.c
+++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4_calcs.c
@@ -7905,6 +7905,7 @@ static noinline_for_stack void dml_core_ms_prefetch_check(struct dml2_core_inter
 
 }
 
+
 static bool dml_core_mode_support(struct dml2_core_calcs_mode_support_ex *in_out_params)
 {
 	struct dml2_core_internal_display_mode_lib *mode_lib = in_out_params->mode_lib;
@@ -11880,7 +11881,7 @@ static bool dml_core_mode_programming(struct dml2_core_calcs_mode_programming_ex
 		}
 
 		//Maximum Bandwidth Used
-		s->TotalWRBandwidth = 0;
+		mode_lib->mp.TotalWRBandwidth = 0;
 		for (k = 0; k < display_cfg->num_streams; ++k) {
 			s->WRBandwidth = 0;
 			if (display_cfg->stream_descriptors[k].writeback.active_writebacks_per_stream > 0) {
@@ -11889,7 +11890,7 @@ static bool dml_core_mode_programming(struct dml2_core_calcs_mode_programming_ex
 					(display_cfg->stream_descriptors[k].timing.h_total * display_cfg->stream_descriptors[k].writeback.writeback_stream[0].input_height
 						/ ((double)display_cfg->stream_descriptors[k].timing.pixel_clock_khz / 1000))
 					* (display_cfg->stream_descriptors[k].writeback.writeback_stream[0].pixel_format == dml2_444_32 ? 4.0 : 8.0);
-				s->TotalWRBandwidth = s->TotalWRBandwidth + s->WRBandwidth;
+				mode_lib->mp.TotalWRBandwidth = mode_lib->mp.TotalWRBandwidth + s->WRBandwidth;
 			}
 		}
 
@@ -13243,7 +13244,7 @@ void dml2_core_calcs_get_informative(const struct dml2_core_internal_display_mod
 		out->informative.misc.DisplayPipeLineDeliveryTimeLumaPrefetch[k] = mode_lib->mp.DisplayPipeLineDeliveryTimeLumaPrefetch[k];
 		out->informative.misc.DisplayPipeLineDeliveryTimeChromaPrefetch[k] = mode_lib->mp.DisplayPipeLineDeliveryTimeChromaPrefetch[k];
 
-		out->informative.misc.WritebackRequiredBandwidth = mode_lib->scratch.dml_core_mode_programming_locals.TotalWRBandwidth / 1000.0;
+		out->informative.misc.WritebackRequiredBandwidth = mode_lib->mp.TotalWRBandwidth / 1000.0;
 		out->informative.misc.WritebackAllowDRAMClockChangeEndPosition[k] = mode_lib->mp.WritebackAllowDRAMClockChangeEndPosition[k];
 		out->informative.misc.WritebackAllowFCLKChangeEndPosition[k] = mode_lib->mp.WritebackAllowFCLKChangeEndPosition[k];
 		out->informative.misc.DSCCLK_calculated[k] = mode_lib->mp.DSCCLK[k];
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_shared_types.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_shared_types.h
index bdee6ad7bc59..5f5b15b478e3 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_shared_types.h
+++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_shared_types.h
@@ -329,6 +329,7 @@ struct dml2_core_internal_mode_support_info {
 	bool temp_read_or_ppt_support;
 
 	struct dml2_core_internal_watermarks watermarks;
+	bool dcfclk_support;
 };
 
 struct dml2_core_internal_mode_support {
@@ -350,9 +351,11 @@ struct dml2_core_internal_mode_support {
 	double SOCCLK; /// <brief Basically just the clock freq at the min (or given) state
 	double DCFCLK; /// <brief Basically just the clock freq at the min (or given) state and max combine setting
 	double GlobalDPPCLK; /// <brief the Max DPPCLK freq out of all pipes
+	double GlobalDTBCLK; /// <brief the Max DTBCLK freq out of all pipes
 	double uclk_freq_mhz;
 	double dram_bw_mbps;
 	double max_dram_bw_mbps;
+	double min_available_urgent_bandwidth_MBps; /// <brief Minimum guaranteed available urgent return bandwidth in MBps
 
 	double MaxFabricClock; /// <brief Basically just the clock freq at the min (or given) state
 	double MaxDCFCLK; /// <brief Basically just the clock freq at the min (or given) state and max combine setting
@@ -473,9 +476,19 @@ struct dml2_core_internal_mode_support {
 	bool RequiresDSC[DML2_MAX_PLANES];
 	bool RequiresFEC[DML2_MAX_PLANES];
 	double OutputBpp[DML2_MAX_PLANES];
+	double DesiredOutputBpp[DML2_MAX_PLANES];
+	double PixelClockBackEnd[DML2_MAX_PLANES];
 	unsigned int DSCDelay[DML2_MAX_PLANES];
 	enum dml2_core_internal_output_type OutputType[DML2_MAX_PLANES];
 	enum dml2_core_internal_output_type_rate OutputRate[DML2_MAX_PLANES];
+	bool TotalAvailablePipesSupportNoDSC;
+	bool TotalAvailablePipesSupportDSC;
+	unsigned int NumberOfDPPNoDSC;
+	unsigned int NumberOfDPPDSC;
+	enum dml2_odm_mode ODMModeNoDSC;
+	enum dml2_odm_mode ODMModeDSC;
+	double RequiredDISPCLKPerSurfaceNoDSC;
+	double RequiredDISPCLKPerSurfaceDSC;
 
 	// Bandwidth Related Info
 	double BandwidthAvailableForImmediateFlip;
@@ -538,7 +551,41 @@ struct dml2_core_internal_mode_support {
 	bool mall_comb_mcache_c[DML2_MAX_PLANES];
 	bool lc_comb_mcache[DML2_MAX_PLANES];
 
+	unsigned int vmpg_width_y[DML2_MAX_PLANES];
+	unsigned int vmpg_height_y[DML2_MAX_PLANES];
+	unsigned int vmpg_width_c[DML2_MAX_PLANES];
+	unsigned int vmpg_height_c[DML2_MAX_PLANES];
+
+	unsigned int meta_row_height_luma[DML2_MAX_PLANES];
+	unsigned int meta_row_height_chroma[DML2_MAX_PLANES];
+	unsigned int meta_row_bytes_per_row_ub_l[DML2_MAX_PLANES];
+	unsigned int meta_row_bytes_per_row_ub_c[DML2_MAX_PLANES];
+	unsigned int dpte_row_bytes_per_row_l[DML2_MAX_PLANES];
+	unsigned int dpte_row_bytes_per_row_c[DML2_MAX_PLANES];
+
+	unsigned int pstate_bytes_required_l[DML2_MAX_PLANES];
+	unsigned int pstate_bytes_required_c[DML2_MAX_PLANES];
+	unsigned int cursor_bytes_per_chunk[DML2_MAX_PLANES];
+	unsigned int cursor_bytes_per_line[DML2_MAX_PLANES];
+
+	unsigned int MaximumVStartup[DML2_MAX_PLANES];
+
+	double HostVMInefficiencyFactor;
+	double HostVMInefficiencyFactorPrefetch;
+
+	unsigned int tdlut_pte_bytes_per_frame[DML2_MAX_PLANES];
+	unsigned int tdlut_bytes_per_frame[DML2_MAX_PLANES];
+	double tdlut_opt_time[DML2_MAX_PLANES];
+	double tdlut_drain_time[DML2_MAX_PLANES];
+	unsigned int tdlut_bytes_per_group[DML2_MAX_PLANES];
+
+	double Tvm_trips_flip[DML2_MAX_PLANES];
+	double Tr0_trips_flip[DML2_MAX_PLANES];
+	double Tvm_trips_flip_rounded[DML2_MAX_PLANES];
+	double Tr0_trips_flip_rounded[DML2_MAX_PLANES];
 
+	unsigned int DSTYAfterScaler[DML2_MAX_PLANES];
+	unsigned int DSTXAfterScaler[DML2_MAX_PLANES];
 };
 
 /// @brief A mega structure that houses various info for model programming step.
@@ -548,6 +595,7 @@ struct dml2_core_internal_mode_program {
 	double FabricClock; /// <brief Basically just the clock freq at the min (or given) state
 	//double DCFCLK; /// <brief Basically just the clock freq at the min (or given) state and max combine setting
 	double dram_bw_mbps;
+	double min_available_urgent_bandwidth_MBps; /// <brief Minimum guaranteed available urgent return bandwidth in MBps
 	double uclk_freq_mhz;
 	unsigned int NoOfDPP[DML2_MAX_PLANES];
 	enum dml2_odm_mode ODMMode[DML2_MAX_PLANES];
@@ -684,6 +732,38 @@ struct dml2_core_internal_mode_program {
 	double TCalc;
 	unsigned int TotImmediateFlipBytes;
 
+	unsigned int MaxTotalDETInKByte;
+	unsigned int NomDETInKByte;
+	unsigned int MinCompressedBufferSizeInKByte;
+	double PixelClockBackEnd[DML2_MAX_PLANES];
+	double OutputBpp[DML2_MAX_PLANES];
+	bool dsc_enable[DML2_MAX_PLANES];
+	unsigned int num_dsc_slices[DML2_MAX_PLANES];
+	unsigned int meta_row_bytes_per_row_ub_l[DML2_MAX_PLANES];
+	unsigned int meta_row_bytes_per_row_ub_c[DML2_MAX_PLANES];
+	unsigned int dpte_row_bytes_per_row_l[DML2_MAX_PLANES];
+	unsigned int dpte_row_bytes_per_row_c[DML2_MAX_PLANES];
+	unsigned int cursor_bytes_per_chunk[DML2_MAX_PLANES];
+	unsigned int cursor_bytes_per_line[DML2_MAX_PLANES];
+	unsigned int MaxVStartupLines[DML2_MAX_PLANES]; /// <brief more like vblank for the plane's OTG
+	double HostVMInefficiencyFactor;
+	double HostVMInefficiencyFactorPrefetch;
+	unsigned int tdlut_pte_bytes_per_frame[DML2_MAX_PLANES];
+	unsigned int tdlut_bytes_per_frame[DML2_MAX_PLANES];
+	unsigned int tdlut_groups_per_2row_ub[DML2_MAX_PLANES];
+	double tdlut_opt_time[DML2_MAX_PLANES];
+	double tdlut_drain_time[DML2_MAX_PLANES];
+	unsigned int tdlut_bytes_per_group[DML2_MAX_PLANES];
+	double Tvm_trips_flip[DML2_MAX_PLANES];
+	double Tr0_trips_flip[DML2_MAX_PLANES];
+	double Tvm_trips_flip_rounded[DML2_MAX_PLANES];
+	double Tr0_trips_flip_rounded[DML2_MAX_PLANES];
+	bool immediate_flip_required; // any pipes need immediate flip
+	double SOCCLK; /// <brief Basically just the clock freq at the min (or given) state
+	double TotalWRBandwidth;
+	double max_urgent_latency_us;
+	double df_response_time_us;
+
 	// -------------------
 	// Output
 	// -------------------
@@ -697,6 +777,7 @@ struct dml2_core_internal_mode_program {
 	bool PrefetchModeSupported; // <brief Is the prefetch mode (bandwidth and latency) supported
 	bool ImmediateFlipSupported;
 	bool ImmediateFlipSupportedForPipe[DML2_MAX_PLANES];
+	bool dcfclk_support;
 
 	// Clock
 	double Dcfclk;
@@ -1028,7 +1109,6 @@ struct dml2_core_calcs_mode_programming_locals {
 	double dlg_vblank_start;
 	double LSetup;
 	double blank_lines_remaining;
-	double TotalWRBandwidth;
 	double WRBandwidth;
 	struct dml2_core_internal_DmlPipe myPipe;
 	double PixelClockBackEndFactor;
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.c
index f486b090bbfc..22969a533a7b 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.c
+++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.c
@@ -389,9 +389,6 @@ static bool map_min_clocks_to_dpm(const struct dml2_core_mode_support_result *mo
 	if (result)
 		result = round_up_to_next_dpm(&display_cfg->min_clocks.dcn4x.dispclk_khz, &state_table->dispclk);
 
-	if (result)
-		result = round_up_to_next_dpm(&display_cfg->min_clocks.dcn4x.deepsleep_dcfclk_khz, &state_table->dcfclk);
-
 	for (i = 0; i < DML2_MAX_DCN_PIPES; i++) {
 		if (result)
 			result = round_up_to_next_dpm(&display_cfg->plane_programming[i].min_clocks.dcn4x.dppclk_khz, &state_table->dppclk);
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/inc/dml2_internal_shared_types.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/inc/dml2_internal_shared_types.h
index 00688b9f1df4..d52aa82283b3 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/inc/dml2_internal_shared_types.h
+++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/inc/dml2_internal_shared_types.h
@@ -202,6 +202,8 @@ struct dml2_core_mode_support_result {
 		} active;
 
 		unsigned int dispclk_khz;
+		unsigned int dpprefclk_khz;
+		unsigned int dtbrefclk_khz;
 		unsigned int dcfclk_deepsleep_khz;
 		unsigned int socclk_khz;
 
@@ -446,13 +448,17 @@ struct dml2_core_internal_state_intermediates {
 };
 
 struct dml2_core_mode_support_locals {
-	struct dml2_core_calcs_mode_support_ex mode_support_ex_params;
+	union {
+		struct dml2_core_calcs_mode_support_ex mode_support_ex_params;
+	};
 	struct dml2_display_cfg svp_expanded_display_cfg;
 	struct dml2_calculate_mcache_allocation_in_out calc_mcache_allocation_params;
 };
 
 struct dml2_core_mode_programming_locals {
-	struct dml2_core_calcs_mode_programming_ex mode_programming_ex_params;
+	union {
+		struct dml2_core_calcs_mode_programming_ex mode_programming_ex_params;
+	};
 	struct dml2_display_cfg svp_expanded_display_cfg;
 };
 
diff --git a/drivers/gpu/drm/amd/display/dmub/inc/dmub_cmd.h b/drivers/gpu/drm/amd/display/dmub/inc/dmub_cmd.h
index 76e9dcc15466..52dc731fa094 100644
--- a/drivers/gpu/drm/amd/display/dmub/inc/dmub_cmd.h
+++ b/drivers/gpu/drm/amd/display/dmub/inc/dmub_cmd.h
@@ -104,6 +104,14 @@
  */
 #define DMUB_MAX_FPO_STREAMS 4
 
+/* Define to ensure that the "common" members always appear in the same
+ * order in different structs for back compat purposes
+ */
+#define COMMON_STREAM_STATIC_SUB_STATE \
+    struct dmub_fams2_cmd_legacy_stream_static_state legacy; \
+    struct dmub_fams2_cmd_subvp_stream_static_state subvp; \
+    struct dmub_fams2_cmd_drr_stream_static_state drr;
+
 /* Maximum number of streams on any ASIC. */
 #define DMUB_MAX_STREAMS 6
 
@@ -1998,11 +2006,13 @@ union dmub_fams2_stream_static_sub_state {
 }; //v0
 
 union dmub_fams2_cmd_stream_static_sub_state {
-	struct dmub_fams2_cmd_legacy_stream_static_state legacy;
-	struct dmub_fams2_cmd_subvp_stream_static_state subvp;
-	struct dmub_fams2_cmd_drr_stream_static_state drr;
+	COMMON_STREAM_STATIC_SUB_STATE
 }; //v1
 
+union dmub_fams2_stream_static_sub_state_v2 {
+	COMMON_STREAM_STATIC_SUB_STATE
+}; //v2
+
 struct dmub_fams2_stream_static_state {
 	enum fams2_stream_type type;
 	uint32_t otg_vline_time_ns;
@@ -2068,7 +2078,7 @@ struct dmub_fams2_cmd_stream_static_base_state {
 
 struct dmub_fams2_stream_static_state_v1 {
 	struct dmub_fams2_cmd_stream_static_base_state base;
-	union dmub_fams2_cmd_stream_static_sub_state sub_state;
+	union dmub_fams2_stream_static_sub_state_v2 sub_state;
 }; //v1
 
 /**
-- 
2.43.0