diff options
Diffstat (limited to 'drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c')
-rw-r--r-- | drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c | 285 |
1 files changed, 245 insertions, 40 deletions
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c b/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c index 8e4c9d0887ce..0571700f53f9 100644 --- a/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c +++ b/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c @@ -121,8 +121,8 @@ struct _vcs_dpi_soc_bounding_box_st dcn3_2_soc = { }, }, .num_states = 1, - .sr_exit_time_us = 20.16, - .sr_enter_plus_exit_time_us = 27.13, + .sr_exit_time_us = 42.97, + .sr_enter_plus_exit_time_us = 49.94, .sr_exit_z8_time_us = 285.0, .sr_enter_plus_exit_z8_time_us = 320, .writeback_latency_us = 12.0, @@ -244,6 +244,50 @@ void dcn32_build_wm_range_table_fpu(struct clk_mgr_internal *clk_mgr) } /** + * Finds dummy_latency_index when MCLK switching using firmware based + * vblank stretch is enabled. This function will iterate through the + * table of dummy pstate latencies until the lowest value that allows + * dm_allow_self_refresh_and_mclk_switch to happen is found + */ +int dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(struct dc *dc, + struct dc_state *context, + display_e2e_pipe_params_st *pipes, + int pipe_cnt, + int vlevel) +{ + const int max_latency_table_entries = 4; + const struct vba_vars_st *vba = &context->bw_ctx.dml.vba; + int dummy_latency_index = 0; + + dc_assert_fp_enabled(); + + while (dummy_latency_index < max_latency_table_entries) { + context->bw_ctx.dml.soc.dram_clock_change_latency_us = + dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us; + dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false); + + if (vlevel < context->bw_ctx.dml.vba.soc.num_states && + vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported) + break; + + dummy_latency_index++; + } + + if (dummy_latency_index == max_latency_table_entries) { + ASSERT(dummy_latency_index != max_latency_table_entries); + /* If the execution gets here, it means dummy p_states are + * not possible. This should never happen and would mean + * something is severely wrong. + * Here we reset dummy_latency_index to 3, because it is + * better to have underflows than system crashes. + */ + dummy_latency_index = max_latency_table_entries - 1; + } + + return dummy_latency_index; +} + +/** * dcn32_helper_populate_phantom_dlg_params - Get DLG params for phantom pipes * and populate pipe_ctx with those params. * @@ -286,41 +330,92 @@ void dcn32_helper_populate_phantom_dlg_params(struct dc *dc, } } -bool dcn32_predict_pipe_split(struct dc_state *context, display_pipe_params_st pipe, int index) +/** + * ******************************************************************************************* + * dcn32_predict_pipe_split: Predict if pipe split will occur for a given DML pipe + * + * This function takes in a DML pipe (pipe_e2e) and predicts if pipe split is required (both + * ODM and MPC). For pipe split, ODM combine is determined by the ODM mode, and MPC combine is + * determined by DPPClk requirements + * + * This function follows the same policy as DML: + * - Check for ODM combine requirements / policy first + * - MPC combine is only chosen if there is no ODM combine requirements / policy in place, and + * MPC is required + * + * @param [in]: context: New DC state to be programmed + * @param [in]: pipe_e2e: DML pipe end to end context + * + * @return: Number of splits expected (1 for 2:1 split, 3 for 4:1 split, 0 for no splits). + * + * ******************************************************************************************* + */ +uint8_t dcn32_predict_pipe_split(struct dc_state *context, + display_e2e_pipe_params_st *pipe_e2e) { double pscl_throughput; double pscl_throughput_chroma; double dpp_clk_single_dpp, clock; double clk_frequency = 0.0; double vco_speed = context->bw_ctx.dml.soc.dispclk_dppclk_vco_speed_mhz; + bool total_available_pipes_support = false; + uint32_t number_of_dpp = 0; + enum odm_combine_mode odm_mode = dm_odm_combine_mode_disabled; + double req_dispclk_per_surface = 0; + uint8_t num_splits = 0; dc_assert_fp_enabled(); - dml32_CalculateSinglePipeDPPCLKAndSCLThroughput(pipe.scale_ratio_depth.hscl_ratio, - pipe.scale_ratio_depth.hscl_ratio_c, - pipe.scale_ratio_depth.vscl_ratio, - pipe.scale_ratio_depth.vscl_ratio_c, - context->bw_ctx.dml.ip.max_dchub_pscl_bw_pix_per_clk, - context->bw_ctx.dml.ip.max_pscl_lb_bw_pix_per_clk, - pipe.dest.pixel_rate_mhz, - pipe.src.source_format, - pipe.scale_taps.htaps, - pipe.scale_taps.htaps_c, - pipe.scale_taps.vtaps, - pipe.scale_taps.vtaps_c, - /* Output */ - &pscl_throughput, &pscl_throughput_chroma, - &dpp_clk_single_dpp); + dml32_CalculateODMMode(context->bw_ctx.dml.ip.maximum_pixels_per_line_per_dsc_unit, + pipe_e2e->pipe.dest.hactive, + pipe_e2e->dout.output_format, + pipe_e2e->dout.output_type, + pipe_e2e->pipe.dest.odm_combine_policy, + context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dispclk_mhz, + context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dispclk_mhz, + pipe_e2e->dout.dsc_enable != 0, + 0, /* TotalNumberOfActiveDPP can be 0 since we're predicting pipe split requirement */ + context->bw_ctx.dml.ip.max_num_dpp, + pipe_e2e->pipe.dest.pixel_rate_mhz, + context->bw_ctx.dml.soc.dcn_downspread_percent, + context->bw_ctx.dml.ip.dispclk_ramp_margin_percent, + context->bw_ctx.dml.soc.dispclk_dppclk_vco_speed_mhz, + pipe_e2e->dout.dsc_slices, + /* Output */ + &total_available_pipes_support, + &number_of_dpp, + &odm_mode, + &req_dispclk_per_surface); + + dml32_CalculateSinglePipeDPPCLKAndSCLThroughput(pipe_e2e->pipe.scale_ratio_depth.hscl_ratio, + pipe_e2e->pipe.scale_ratio_depth.hscl_ratio_c, + pipe_e2e->pipe.scale_ratio_depth.vscl_ratio, + pipe_e2e->pipe.scale_ratio_depth.vscl_ratio_c, + context->bw_ctx.dml.ip.max_dchub_pscl_bw_pix_per_clk, + context->bw_ctx.dml.ip.max_pscl_lb_bw_pix_per_clk, + pipe_e2e->pipe.dest.pixel_rate_mhz, + pipe_e2e->pipe.src.source_format, + pipe_e2e->pipe.scale_taps.htaps, + pipe_e2e->pipe.scale_taps.htaps_c, + pipe_e2e->pipe.scale_taps.vtaps, + pipe_e2e->pipe.scale_taps.vtaps_c, + /* Output */ + &pscl_throughput, &pscl_throughput_chroma, + &dpp_clk_single_dpp); clock = dpp_clk_single_dpp * (1 + context->bw_ctx.dml.soc.dcn_downspread_percent / 100); if (clock > 0) - clk_frequency = vco_speed * 4.0 / ((int)(vco_speed * 4.0)); + clk_frequency = vco_speed * 4.0 / ((int)(vco_speed * 4.0) / clock); - if (clk_frequency > context->bw_ctx.dml.soc.clock_limits[index].dppclk_mhz) - return true; - else - return false; + if (odm_mode == dm_odm_combine_mode_2to1) + num_splits = 1; + else if (odm_mode == dm_odm_combine_mode_4to1) + num_splits = 3; + else if (clk_frequency > context->bw_ctx.dml.soc.clock_limits[context->bw_ctx.dml.soc.num_states - 1].dppclk_mhz) + num_splits = 1; + + return num_splits; } static float calculate_net_bw_in_kbytes_sec(struct _vcs_dpi_voltage_scaling_st *entry) @@ -560,6 +655,7 @@ static bool dcn32_assign_subvp_pipe(struct dc *dc, bool valid_assignment_found = false; unsigned int free_pipes = dcn32_get_num_free_pipes(dc, context); bool current_assignment_freesync = false; + struct vba_vars_st *vba = &context->bw_ctx.dml.vba; for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) { struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i]; @@ -573,8 +669,16 @@ static bool dcn32_assign_subvp_pipe(struct dc *dc, refresh_rate = (pipe->stream->timing.pix_clk_100hz * 100 + pipe->stream->timing.v_total * pipe->stream->timing.h_total - 1) / (double)(pipe->stream->timing.v_total * pipe->stream->timing.h_total); + /* SubVP pipe candidate requirements: + * - Refresh rate < 120hz + * - Not able to switch in vactive naturally (switching in active means the + * DET provides enough buffer to hide the P-State switch latency -- trying + * to combine this with SubVP can cause issues with the scheduling). + * - Not TMZ surface + */ if (pipe->plane_state && !pipe->top_pipe && - pipe->stream->mall_stream_config.type == SUBVP_NONE && refresh_rate < 120) { + pipe->stream->mall_stream_config.type == SUBVP_NONE && refresh_rate < 120 && !pipe->plane_state->address.tmz_surface && + vba->ActiveDRAMClockChangeLatencyMarginPerState[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]] <= 0) { while (pipe) { num_pipes++; pipe = pipe->bottom_pipe; @@ -998,8 +1102,10 @@ static void dcn32_full_validate_bw_helper(struct dc *dc, *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt); /* This may adjust vlevel and maxMpcComb */ - if (*vlevel < context->bw_ctx.dml.soc.num_states) + if (*vlevel < context->bw_ctx.dml.soc.num_states) { *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge); + vba->VoltageLevel = *vlevel; + } /* Conditions for setting up phantom pipes for SubVP: * 1. Not force disable SubVP @@ -1014,8 +1120,10 @@ static void dcn32_full_validate_bw_helper(struct dc *dc, dc->debug.force_subvp_mclk_switch)) { dcn32_merge_pipes_for_subvp(dc, context); - // to re-initialize viewport after the pipe merge - for (int i = 0; i < dc->res_pool->pipe_count; i++) { + memset(merge, 0, MAX_PIPES * sizeof(bool)); + + /* to re-initialize viewport after the pipe merge */ + for (i = 0; i < dc->res_pool->pipe_count; i++) { struct pipe_ctx *pipe_ctx = &context->res_ctx.pipe_ctx[i]; if (!pipe_ctx->plane_state || !pipe_ctx->stream) @@ -1091,17 +1199,31 @@ static void dcn32_full_validate_bw_helper(struct dc *dc, dc->res_pool->funcs->remove_phantom_pipes(dc, context); vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] = dm_dram_clock_change_unsupported; *pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false); - } else { - // only call dcn20_validate_apply_pipe_split_flags if we found a supported config - memset(split, 0, MAX_PIPES * sizeof(int)); - memset(merge, 0, MAX_PIPES * sizeof(bool)); - *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge); + *vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, *pipe_cnt); + /* This may adjust vlevel and maxMpcComb */ + if (*vlevel < context->bw_ctx.dml.soc.num_states) { + *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge); + vba->VoltageLevel = *vlevel; + } + } else { // Most populate phantom DLG params before programming hardware / timing for phantom pipe DC_FP_START(); dcn32_helper_populate_phantom_dlg_params(dc, context, pipes, *pipe_cnt); DC_FP_END(); + /* Call validate_apply_pipe_split flags after calling DML getters for + * phantom dlg params, or some of the VBA params indicating pipe split + * can be overwritten by the getters. + * + * When setting up SubVP config, all pipes are merged before attempting to + * add phantom pipes. If pipe split (ODM / MPC) is required, both the main + * and phantom pipes will be split in the regular pipe splitting sequence. + */ + memset(split, 0, MAX_PIPES * sizeof(int)); + memset(merge, 0, MAX_PIPES * sizeof(bool)); + *vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, *vlevel, split, merge); + vba->VoltageLevel = *vlevel; // Note: We can't apply the phantom pipes to hardware at this time. We have to wait // until driver has acquired the DMCUB lock to do it safely. } @@ -1425,6 +1547,8 @@ bool dcn32_internal_validate_bw(struct dc *dc, memset(split, 0, sizeof(split)); memset(merge, 0, sizeof(merge)); vlevel = dcn20_validate_apply_pipe_split_flags(dc, context, vlevel, split, merge); + // dcn20_validate_apply_pipe_split_flags can modify voltage level outside of DML + vba->VoltageLevel = vlevel; } } @@ -1467,6 +1591,28 @@ bool dcn32_internal_validate_bw(struct dc *dc, if (pipe->next_odm_pipe) pipe->next_odm_pipe->prev_odm_pipe = pipe->prev_odm_pipe; + /*2:1ODM+MPC Split MPO to Single Pipe + MPC Split MPO*/ + if (pipe->bottom_pipe) { + if (pipe->bottom_pipe->prev_odm_pipe || pipe->bottom_pipe->next_odm_pipe) { + /*MPC split rules will handle this case*/ + pipe->bottom_pipe->top_pipe = NULL; + } else { + if (pipe->prev_odm_pipe->bottom_pipe) { + /* 3 plane MPO*/ + pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe->bottom_pipe; + pipe->prev_odm_pipe->bottom_pipe->bottom_pipe = pipe->bottom_pipe; + } else { + /* 2 plane MPO*/ + pipe->bottom_pipe->top_pipe = pipe->prev_odm_pipe; + pipe->prev_odm_pipe->bottom_pipe = pipe->bottom_pipe; + } + } + } + + if (pipe->top_pipe) { + pipe->top_pipe->bottom_pipe = NULL; + } + pipe->bottom_pipe = NULL; pipe->next_odm_pipe = NULL; pipe->plane_state = NULL; @@ -1599,8 +1745,20 @@ bool dcn32_internal_validate_bw(struct dc *dc, goto validate_fail; } - if (repopulate_pipes) + if (repopulate_pipes) { pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate); + + /* repopulate_pipes = 1 means the pipes were either split or merged. In this case + * we have to re-calculate the DET allocation and run through DML once more to + * ensure all the params are calculated correctly. We do not need to run the + * pipe split check again after this call (pipes are already split / merged). + * */ + if (!fast_validate) { + context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final = + dm_prefetch_support_uclk_fclk_and_stutter_if_possible; + vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt); + } + } *vlevel_out = vlevel; *pipe_cnt_out = pipe_cnt; @@ -1646,7 +1804,7 @@ void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context, dcn30_can_support_mclk_switch_using_fw_based_vblank_stretch(dc, context); if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) { - dummy_latency_index = dcn30_find_dummy_latency_index_for_fw_based_mclk_switch(dc, + dummy_latency_index = dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(dc, context, pipes, pipe_cnt, vlevel); /* After calling dcn30_find_dummy_latency_index_for_fw_based_mclk_switch @@ -1785,7 +1943,11 @@ void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context, context->bw_ctx.bw.dcn.watermarks.c.frac_urg_bw_nom = get_fraction_of_urgent_bandwidth(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; context->bw_ctx.bw.dcn.watermarks.c.frac_urg_bw_flip = get_fraction_of_urgent_bandwidth_imm_flip(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; context->bw_ctx.bw.dcn.watermarks.c.urgent_latency_ns = get_urgent_latency(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; - context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.fclk_pstate_change_ns = get_fclk_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; + /* On DCN32/321, PMFW will set PSTATE_CHANGE_TYPE = 1 (FCLK) for UCLK dummy p-state. + * In this case we must program FCLK WM Set C to use the UCLK dummy p-state WM + * value. + */ + context->bw_ctx.bw.dcn.watermarks.c.cstate_pstate.fclk_pstate_change_ns = get_wm_dram_clock_change(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; context->bw_ctx.bw.dcn.watermarks.c.usr_retraining_ns = get_usr_retraining_watermark(&context->bw_ctx.dml, pipes, pipe_cnt) * 1000; if ((!pstate_en) && (dc->clk_mgr->bw_params->wm_table.nv_entries[WM_C].valid)) { @@ -1882,6 +2044,45 @@ static void remove_entry_from_table_at_index(struct _vcs_dpi_voltage_scaling_st memset(&table[--(*num_entries)], 0, sizeof(struct _vcs_dpi_voltage_scaling_st)); } +void dcn32_patch_dpm_table(struct clk_bw_params *bw_params) +{ + int i; + unsigned int max_dcfclk_mhz = 0, max_dispclk_mhz = 0, max_dppclk_mhz = 0, + max_phyclk_mhz = 0, max_dtbclk_mhz = 0, max_fclk_mhz = 0, max_uclk_mhz = 0; + + for (i = 0; i < MAX_NUM_DPM_LVL; i++) { + if (bw_params->clk_table.entries[i].dcfclk_mhz > max_dcfclk_mhz) + max_dcfclk_mhz = bw_params->clk_table.entries[i].dcfclk_mhz; + if (bw_params->clk_table.entries[i].fclk_mhz > max_fclk_mhz) + max_fclk_mhz = bw_params->clk_table.entries[i].fclk_mhz; + if (bw_params->clk_table.entries[i].memclk_mhz > max_uclk_mhz) + max_uclk_mhz = bw_params->clk_table.entries[i].memclk_mhz; + if (bw_params->clk_table.entries[i].dispclk_mhz > max_dispclk_mhz) + max_dispclk_mhz = bw_params->clk_table.entries[i].dispclk_mhz; + if (bw_params->clk_table.entries[i].dppclk_mhz > max_dppclk_mhz) + max_dppclk_mhz = bw_params->clk_table.entries[i].dppclk_mhz; + if (bw_params->clk_table.entries[i].phyclk_mhz > max_phyclk_mhz) + max_phyclk_mhz = bw_params->clk_table.entries[i].phyclk_mhz; + if (bw_params->clk_table.entries[i].dtbclk_mhz > max_dtbclk_mhz) + max_dtbclk_mhz = bw_params->clk_table.entries[i].dtbclk_mhz; + } + + /* Scan through clock values we currently have and if they are 0, + * then populate it with dcn3_2_soc.clock_limits[] value. + * + * Do it for DCFCLK, DISPCLK, DTBCLK and UCLK as any of those being + * 0, will cause it to skip building the clock table. + */ + if (max_dcfclk_mhz == 0) + bw_params->clk_table.entries[0].dcfclk_mhz = dcn3_2_soc.clock_limits[0].dcfclk_mhz; + if (max_dispclk_mhz == 0) + bw_params->clk_table.entries[0].dispclk_mhz = dcn3_2_soc.clock_limits[0].dispclk_mhz; + if (max_dtbclk_mhz == 0) + bw_params->clk_table.entries[0].dtbclk_mhz = dcn3_2_soc.clock_limits[0].dtbclk_mhz; + if (max_uclk_mhz == 0) + bw_params->clk_table.entries[0].memclk_mhz = dcn3_2_soc.clock_limits[0].dram_speed_mts / 16; +} + static int build_synthetic_soc_states(struct clk_bw_params *bw_params, struct _vcs_dpi_voltage_scaling_st *table, unsigned int *num_entries) { @@ -2116,6 +2317,7 @@ void dcn32_update_bw_bounding_box_fpu(struct dc *dc, struct clk_bw_params *bw_pa if ((int)(dcn3_2_soc.urgent_latency_us * 1000) != dc->bb_overrides.urgent_latency_ns && dc->bb_overrides.urgent_latency_ns) { dcn3_2_soc.urgent_latency_us = dc->bb_overrides.urgent_latency_ns / 1000.0; + dcn3_2_soc.urgent_latency_pixel_data_only_us = dc->bb_overrides.urgent_latency_ns / 1000.0; } if ((int)(dcn3_2_soc.dram_clock_change_latency_us * 1000) @@ -2145,13 +2347,16 @@ void dcn32_update_bw_bounding_box_fpu(struct dc *dc, struct clk_bw_params *bw_pa if (dc->ctx->dc_bios->funcs->get_soc_bb_info(dc->ctx->dc_bios, &bb_info) == BP_RESULT_OK) { if (bb_info.dram_clock_change_latency_100ns > 0) - dcn3_2_soc.dram_clock_change_latency_us = bb_info.dram_clock_change_latency_100ns * 10; + dcn3_2_soc.dram_clock_change_latency_us = + bb_info.dram_clock_change_latency_100ns * 10; - if (bb_info.dram_sr_enter_exit_latency_100ns > 0) - dcn3_2_soc.sr_enter_plus_exit_time_us = bb_info.dram_sr_enter_exit_latency_100ns * 10; + if (bb_info.dram_sr_enter_exit_latency_100ns > 0) + dcn3_2_soc.sr_enter_plus_exit_time_us = + bb_info.dram_sr_enter_exit_latency_100ns * 10; - if (bb_info.dram_sr_exit_latency_100ns > 0) - dcn3_2_soc.sr_exit_time_us = bb_info.dram_sr_exit_latency_100ns * 10; + if (bb_info.dram_sr_exit_latency_100ns > 0) + dcn3_2_soc.sr_exit_time_us = + bb_info.dram_sr_exit_latency_100ns * 10; } } |