diff options
Diffstat (limited to 'drivers/gpu/drm/amd/display/dc/dml2')
23 files changed, 750 insertions, 592 deletions
diff --git a/drivers/gpu/drm/amd/display/dc/dml2/Makefile b/drivers/gpu/drm/amd/display/dc/dml2/Makefile index 157ecf008d6c..4c21ce42054c 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/Makefile +++ b/drivers/gpu/drm/amd/display/dc/dml2/Makefile @@ -81,10 +81,11 @@ AMD_DAL_DML2 = $(addprefix $(AMDDALPATH)/dc/dml2/,$(DML2)) AMD_DISPLAY_FILES += $(AMD_DAL_DML2) -CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_interfaces.o := $(dml2_ccflags) -CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_soc15.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4_calcs.o := $(dml2_ccflags) $(frame_warn_flag) +CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_utils.o := $(dml2_ccflags) $(frame_warn_flag) +CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_interfaces.o := $(dml2_ccflags) +CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_soc15.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_factory.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_factory.o := $(dml2_ccflags) @@ -94,17 +95,16 @@ CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_pmo/dml2_pmo_dcn3.o := $(dml2_ccflag CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_pmo/dml2_pmo_dcn4_fams2.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_pmo/dml2_pmo_factory.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_standalone_libraries/lib_float_math.o := $(dml2_ccflags) - - CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/src/dml21_wrapper.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/dml21_translation_helper.o := $(dml2_ccflags) CFLAGS_$(AMDDALPATH)/dc/dml2/dml21/dml21_utils.o := $(dml2_ccflags) -CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_interfaces.o := $(dml2_rcflags) -CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_soc15.o := $(dml2_rcflags) CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4.o := $(dml2_rcflags) CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4_calcs.o := $(dml2_rcflags) CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_factory.o := $(dml2_rcflags) +CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_core/dml2_core_utils.o := $(dml2_rcflags) +CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_interfaces.o := $(dml2_rcflags) +CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_top/dml2_top_soc15.o := $(dml2_rcflags) CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_dcn4.o := $(dml2_rcflags) CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_dpmm/dml2_dpmm_factory.o := $(dml2_rcflags) CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/src/dml2_mcg/dml2_mcg_dcn4.o := $(dml2_rcflags) @@ -120,6 +120,7 @@ CFLAGS_REMOVE_$(AMDDALPATH)/dc/dml2/dml21/dml21_utils.o := $(dml2_rcflags) DML21 := src/dml2_top/dml2_top_interfaces.o DML21 += src/dml2_top/dml2_top_soc15.o DML21 += src/dml2_core/dml2_core_dcn4.o +DML21 += src/dml2_core/dml2_core_utils.o DML21 += src/dml2_core/dml2_core_factory.o DML21 += src/dml2_core/dml2_core_dcn4_calcs.o DML21 += src/dml2_dpmm/dml2_dpmm_dcn4.o diff --git a/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core.c b/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core.c index 7ae9c0ba0c9e..715f9019a33e 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core.c @@ -10189,7 +10189,7 @@ dml_uint_t dml_mode_support_ex(struct dml_mode_support_ex_params_st *in_out_para result = mode_support_pwr_states(&in_out_params->out_lowest_state_idx, in_out_params->mode_lib, in_out_params->in_display_cfg, - 0, + in_out_params->in_start_state_idx, in_out_params->mode_lib->states.num_states - 1); if (result) diff --git a/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core_structs.h b/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core_structs.h index 0670e4dc4fd9..dbeb08466092 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core_structs.h +++ b/drivers/gpu/drm/amd/display/dc/dml2/display_mode_core_structs.h @@ -1917,6 +1917,7 @@ struct display_mode_lib_st { struct dml_mode_support_ex_params_st { struct display_mode_lib_st *mode_lib; const struct dml_display_cfg_st *in_display_cfg; + dml_uint_t in_start_state_idx; dml_uint_t out_lowest_state_idx; struct dml_mode_support_info_st *out_evaluation_info; }; diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.c index d47cacfdb695..a06217a9eef6 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.c @@ -2,7 +2,6 @@ // // Copyright 2024 Advanced Micro Devices, Inc. - #include "dml21_wrapper.h" #include "dml2_core_dcn4_calcs.h" #include "dml2_internal_shared_types.h" @@ -11,277 +10,263 @@ #include "dml21_translation_helper.h" #include "bounding_boxes/dcn4_soc_bb.h" -static void dml21_init_socbb_params(struct dml2_initialize_instance_in_out *dml_init, - const struct dml2_configuration_options *config, - const struct dc *in_dc) -{ - const struct dml2_soc_bb *soc_bb; - const struct dml2_soc_qos_parameters *qos_params; - - switch (in_dc->ctx->dce_version) { - case DCN_VERSION_4_01: - default: - if (config->bb_from_dmub) - soc_bb = config->bb_from_dmub; - else - soc_bb = &dml2_socbb_dcn401; - - qos_params = &dml_dcn4_variant_a_soc_qos_params; - } - - /* patch soc bb */ - memcpy(&dml_init->soc_bb, soc_bb, sizeof(struct dml2_soc_bb)); - - /* patch qos params */ - memcpy(&dml_init->soc_bb.qos_parameters, qos_params, sizeof(struct dml2_soc_qos_parameters)); -} - -static void dml21_external_socbb_params(struct dml2_initialize_instance_in_out *dml_init, - const struct dml2_configuration_options *config) -{ - memcpy(&dml_init->soc_bb, &config->external_socbb_ip_params->soc_bb, sizeof(struct dml2_soc_bb)); -} - -static void dml21_external_ip_params(struct dml2_initialize_instance_in_out *dml_init, +static void dml21_populate_pmo_options(struct dml2_pmo_options *pmo_options, + const struct dc *in_dc, const struct dml2_configuration_options *config) { - memcpy(&dml_init->ip_caps, &config->external_socbb_ip_params->ip_params, sizeof(struct dml2_ip_capabilities)); + bool disable_fams2 = !in_dc->debug.fams2_config.bits.enable; + + /* ODM options */ + pmo_options->disable_dyn_odm = !config->minimize_dispclk_using_odm; + pmo_options->disable_dyn_odm_for_multi_stream = true; + pmo_options->disable_dyn_odm_for_stream_with_svp = true; + + pmo_options->disable_vblank = ((in_dc->debug.dml21_disable_pstate_method_mask >> 1) & 1); + + /* NOTE: DRR and SubVP Require FAMS2 */ + pmo_options->disable_svp = ((in_dc->debug.dml21_disable_pstate_method_mask >> 2) & 1) || + in_dc->debug.force_disable_subvp || + disable_fams2; + pmo_options->disable_drr_clamped = ((in_dc->debug.dml21_disable_pstate_method_mask >> 3) & 1) || + disable_fams2; + pmo_options->disable_drr_var = ((in_dc->debug.dml21_disable_pstate_method_mask >> 4) & 1) || + disable_fams2; + pmo_options->disable_fams2 = disable_fams2; + + pmo_options->disable_drr_var_when_var_active = in_dc->debug.disable_fams_gaming == INGAME_FAMS_DISABLE || + in_dc->debug.disable_fams_gaming == INGAME_FAMS_MULTI_DISP_CLAMPED_ONLY; + pmo_options->disable_drr_clamped_when_var_active = in_dc->debug.disable_fams_gaming == INGAME_FAMS_DISABLE; } -static void dml21_init_ip_params(struct dml2_initialize_instance_in_out *dml_init, +/* + * Populate dml_init based on default static values in soc bb. The default + * values are for reference and support at least minimal operation of current + * SoC and DCN hardware. The values could be modifed by subsequent override + * functions to reflect our true hardware capability. + */ +static void populate_default_dml_init_params(struct dml2_initialize_instance_in_out *dml_init, const struct dml2_configuration_options *config, const struct dc *in_dc) { - const struct dml2_ip_capabilities *ip_caps; - switch (in_dc->ctx->dce_version) { case DCN_VERSION_4_01: + dml_init->options.project_id = dml2_project_dcn4x_stage2_auto_drr_svp; + dml21_populate_pmo_options(&dml_init->options.pmo_options, in_dc, config); + dml_init->soc_bb = dml2_socbb_dcn401; + dml_init->soc_bb.qos_parameters = dml_dcn4_variant_a_soc_qos_params; + dml_init->ip_caps = dml2_dcn401_max_ip_caps; + break; default: - ip_caps = &dml2_dcn401_max_ip_caps; + memset(dml_init, 0, sizeof(*dml_init)); + DC_ERR("unsupported dcn version for DML21!"); + return; } - - memcpy(&dml_init->ip_caps, ip_caps, sizeof(struct dml2_ip_capabilities)); } -void dml21_initialize_soc_bb_params(struct dml2_initialize_instance_in_out *dml_init, +static void override_dml_init_with_values_from_hardware_default(struct dml2_initialize_instance_in_out *dml_init, const struct dml2_configuration_options *config, const struct dc *in_dc) { - if (config->use_native_soc_bb_construction) - dml21_init_socbb_params(dml_init, config, in_dc); - else - dml21_external_socbb_params(dml_init, config); + dml_init->soc_bb.dchub_refclk_mhz = in_dc->res_pool->ref_clocks.dchub_ref_clock_inKhz / 1000; + dml_init->soc_bb.dprefclk_mhz = in_dc->clk_mgr->dprefclk_khz / 1000; + dml_init->soc_bb.dispclk_dppclk_vco_speed_mhz = in_dc->clk_mgr->dentist_vco_freq_khz / 1000.0; } -void dml21_initialize_ip_params(struct dml2_initialize_instance_in_out *dml_init, +/* + * SMU stands for System Management Unit. It is a power management processor. + * It owns the initialization of dc's clock table and programming of clock values + * based on dc's requests. + * Our clock values in base soc bb is a dummy placeholder. The real clock values + * are retrieved from SMU firmware to dc clock table at runtime. + * This function overrides our dummy placeholder values with real values in dc + * clock table. + */ +static void override_dml_init_with_values_from_smu( + struct dml2_initialize_instance_in_out *dml_init, const struct dml2_configuration_options *config, const struct dc *in_dc) { - if (config->use_native_soc_bb_construction) - dml21_init_ip_params(dml_init, config, in_dc); - else - dml21_external_ip_params(dml_init, config); -} - -void dml21_apply_soc_bb_overrides(struct dml2_initialize_instance_in_out *dml_init, - const struct dml2_configuration_options *config, const struct dc *in_dc) -{ int i; - const struct clk_bw_params *dc_bw_params = in_dc->clk_mgr->bw_params; const struct clk_limit_table *dc_clk_table = &dc_bw_params->clk_table; - struct dml2_soc_bb *dml_soc_bb = &dml_init->soc_bb; - struct dml2_soc_state_table *dml_clk_table = &dml_soc_bb->clk_table; - - /* override clocks if smu is present */ - if (in_dc->clk_mgr->funcs->is_smu_present && in_dc->clk_mgr->funcs->is_smu_present(in_dc->clk_mgr)) { - /* dcfclk */ - if (dc_clk_table->num_entries_per_clk.num_dcfclk_levels) { - dml_clk_table->dcfclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dcfclk_levels; - for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { - if (i < dml_clk_table->dcfclk.num_clk_values) { - if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dcfclk_mhz && - dc_clk_table->entries[i].dcfclk_mhz > dc_bw_params->dc_mode_limit.dcfclk_mhz) { - if (i == 0 || dc_clk_table->entries[i-1].dcfclk_mhz < dc_bw_params->dc_mode_limit.dcfclk_mhz) { - dml_clk_table->dcfclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dcfclk_mhz * 1000; - dml_clk_table->dcfclk.num_clk_values = i + 1; - } else { - dml_clk_table->dcfclk.clk_values_khz[i] = 0; - dml_clk_table->dcfclk.num_clk_values = i; - } + struct dml2_soc_state_table *dml_clk_table = &dml_init->soc_bb.clk_table; + + if (!in_dc->clk_mgr->funcs->is_smu_present || + !in_dc->clk_mgr->funcs->is_smu_present(in_dc->clk_mgr)) + /* skip if smu is not present */ + return; + + /* dcfclk */ + if (dc_clk_table->num_entries_per_clk.num_dcfclk_levels) { + dml_clk_table->dcfclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dcfclk_levels; + for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { + if (i < dml_clk_table->dcfclk.num_clk_values) { + if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dcfclk_mhz && + dc_clk_table->entries[i].dcfclk_mhz > dc_bw_params->dc_mode_limit.dcfclk_mhz) { + if (i == 0 || dc_clk_table->entries[i-1].dcfclk_mhz < dc_bw_params->dc_mode_limit.dcfclk_mhz) { + dml_clk_table->dcfclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dcfclk_mhz * 1000; + dml_clk_table->dcfclk.num_clk_values = i + 1; } else { - dml_clk_table->dcfclk.clk_values_khz[i] = dc_clk_table->entries[i].dcfclk_mhz * 1000; + dml_clk_table->dcfclk.clk_values_khz[i] = 0; + dml_clk_table->dcfclk.num_clk_values = i; } } else { - dml_clk_table->dcfclk.clk_values_khz[i] = 0; + dml_clk_table->dcfclk.clk_values_khz[i] = dc_clk_table->entries[i].dcfclk_mhz * 1000; } + } else { + dml_clk_table->dcfclk.clk_values_khz[i] = 0; } } + } - /* fclk */ - if (dc_clk_table->num_entries_per_clk.num_fclk_levels) { - dml_clk_table->fclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_fclk_levels; - for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { - if (i < dml_clk_table->fclk.num_clk_values) { - if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.fclk_mhz && - dc_clk_table->entries[i].fclk_mhz > dc_bw_params->dc_mode_limit.fclk_mhz) { - if (i == 0 || dc_clk_table->entries[i-1].fclk_mhz < dc_bw_params->dc_mode_limit.fclk_mhz) { - dml_clk_table->fclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.fclk_mhz * 1000; - dml_clk_table->fclk.num_clk_values = i + 1; - } else { - dml_clk_table->fclk.clk_values_khz[i] = 0; - dml_clk_table->fclk.num_clk_values = i; - } + /* fclk */ + if (dc_clk_table->num_entries_per_clk.num_fclk_levels) { + dml_clk_table->fclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_fclk_levels; + for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { + if (i < dml_clk_table->fclk.num_clk_values) { + if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.fclk_mhz && + dc_clk_table->entries[i].fclk_mhz > dc_bw_params->dc_mode_limit.fclk_mhz) { + if (i == 0 || dc_clk_table->entries[i-1].fclk_mhz < dc_bw_params->dc_mode_limit.fclk_mhz) { + dml_clk_table->fclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.fclk_mhz * 1000; + dml_clk_table->fclk.num_clk_values = i + 1; } else { - dml_clk_table->fclk.clk_values_khz[i] = dc_clk_table->entries[i].fclk_mhz * 1000; + dml_clk_table->fclk.clk_values_khz[i] = 0; + dml_clk_table->fclk.num_clk_values = i; } } else { - dml_clk_table->fclk.clk_values_khz[i] = 0; + dml_clk_table->fclk.clk_values_khz[i] = dc_clk_table->entries[i].fclk_mhz * 1000; } + } else { + dml_clk_table->fclk.clk_values_khz[i] = 0; } } + } - /* uclk */ - if (dc_clk_table->num_entries_per_clk.num_memclk_levels) { - dml_clk_table->uclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_memclk_levels; - for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { - if (i < dml_clk_table->uclk.num_clk_values) { - if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.memclk_mhz && - dc_clk_table->entries[i].memclk_mhz > dc_bw_params->dc_mode_limit.memclk_mhz) { - if (i == 0 || dc_clk_table->entries[i-1].memclk_mhz < dc_bw_params->dc_mode_limit.memclk_mhz) { - dml_clk_table->uclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.memclk_mhz * 1000; - dml_clk_table->uclk.num_clk_values = i + 1; - } else { - dml_clk_table->uclk.clk_values_khz[i] = 0; - dml_clk_table->uclk.num_clk_values = i; - } + /* uclk */ + if (dc_clk_table->num_entries_per_clk.num_memclk_levels) { + dml_clk_table->uclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_memclk_levels; + for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { + if (i < dml_clk_table->uclk.num_clk_values) { + if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.memclk_mhz && + dc_clk_table->entries[i].memclk_mhz > dc_bw_params->dc_mode_limit.memclk_mhz) { + if (i == 0 || dc_clk_table->entries[i-1].memclk_mhz < dc_bw_params->dc_mode_limit.memclk_mhz) { + dml_clk_table->uclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.memclk_mhz * 1000; + dml_clk_table->uclk.num_clk_values = i + 1; } else { - dml_clk_table->uclk.clk_values_khz[i] = dc_clk_table->entries[i].memclk_mhz * 1000; + dml_clk_table->uclk.clk_values_khz[i] = 0; + dml_clk_table->uclk.num_clk_values = i; } } else { - dml_clk_table->uclk.clk_values_khz[i] = 0; + dml_clk_table->uclk.clk_values_khz[i] = dc_clk_table->entries[i].memclk_mhz * 1000; } + } else { + dml_clk_table->uclk.clk_values_khz[i] = 0; } } + } - /* dispclk */ - if (dc_clk_table->num_entries_per_clk.num_dispclk_levels) { - dml_clk_table->dispclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dispclk_levels; - for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { - if (i < dml_clk_table->dispclk.num_clk_values) { - if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dispclk_mhz && - dc_clk_table->entries[i].dispclk_mhz > dc_bw_params->dc_mode_limit.dispclk_mhz) { - if (i == 0 || dc_clk_table->entries[i-1].dispclk_mhz < dc_bw_params->dc_mode_limit.dispclk_mhz) { - dml_clk_table->dispclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dispclk_mhz * 1000; - dml_clk_table->dispclk.num_clk_values = i + 1; - } else { - dml_clk_table->dispclk.clk_values_khz[i] = 0; - dml_clk_table->dispclk.num_clk_values = i; - } + /* dispclk */ + if (dc_clk_table->num_entries_per_clk.num_dispclk_levels) { + dml_clk_table->dispclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dispclk_levels; + for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { + if (i < dml_clk_table->dispclk.num_clk_values) { + if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dispclk_mhz && + dc_clk_table->entries[i].dispclk_mhz > dc_bw_params->dc_mode_limit.dispclk_mhz) { + if (i == 0 || dc_clk_table->entries[i-1].dispclk_mhz < dc_bw_params->dc_mode_limit.dispclk_mhz) { + dml_clk_table->dispclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dispclk_mhz * 1000; + dml_clk_table->dispclk.num_clk_values = i + 1; } else { - dml_clk_table->dispclk.clk_values_khz[i] = dc_clk_table->entries[i].dispclk_mhz * 1000; + dml_clk_table->dispclk.clk_values_khz[i] = 0; + dml_clk_table->dispclk.num_clk_values = i; } } else { - dml_clk_table->dispclk.clk_values_khz[i] = 0; + dml_clk_table->dispclk.clk_values_khz[i] = dc_clk_table->entries[i].dispclk_mhz * 1000; } + } else { + dml_clk_table->dispclk.clk_values_khz[i] = 0; } } + } - /* dppclk */ - if (dc_clk_table->num_entries_per_clk.num_dppclk_levels) { - dml_clk_table->dppclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dppclk_levels; - for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { - if (i < dml_clk_table->dppclk.num_clk_values) { - if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dppclk_mhz && - dc_clk_table->entries[i].dppclk_mhz > dc_bw_params->dc_mode_limit.dppclk_mhz) { - if (i == 0 || dc_clk_table->entries[i-1].dppclk_mhz < dc_bw_params->dc_mode_limit.dppclk_mhz) { - dml_clk_table->dppclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dppclk_mhz * 1000; - dml_clk_table->dppclk.num_clk_values = i + 1; - } else { - dml_clk_table->dppclk.clk_values_khz[i] = 0; - dml_clk_table->dppclk.num_clk_values = i; - } + /* dppclk */ + if (dc_clk_table->num_entries_per_clk.num_dppclk_levels) { + dml_clk_table->dppclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dppclk_levels; + for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { + if (i < dml_clk_table->dppclk.num_clk_values) { + if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dppclk_mhz && + dc_clk_table->entries[i].dppclk_mhz > dc_bw_params->dc_mode_limit.dppclk_mhz) { + if (i == 0 || dc_clk_table->entries[i-1].dppclk_mhz < dc_bw_params->dc_mode_limit.dppclk_mhz) { + dml_clk_table->dppclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dppclk_mhz * 1000; + dml_clk_table->dppclk.num_clk_values = i + 1; } else { - dml_clk_table->dppclk.clk_values_khz[i] = dc_clk_table->entries[i].dppclk_mhz * 1000; + dml_clk_table->dppclk.clk_values_khz[i] = 0; + dml_clk_table->dppclk.num_clk_values = i; } } else { - dml_clk_table->dppclk.clk_values_khz[i] = 0; + dml_clk_table->dppclk.clk_values_khz[i] = dc_clk_table->entries[i].dppclk_mhz * 1000; } + } else { + dml_clk_table->dppclk.clk_values_khz[i] = 0; } } + } - /* dtbclk */ - if (dc_clk_table->num_entries_per_clk.num_dtbclk_levels) { - dml_clk_table->dtbclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dtbclk_levels; - for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { - if (i < dml_clk_table->dtbclk.num_clk_values) { - if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dtbclk_mhz && - dc_clk_table->entries[i].dtbclk_mhz > dc_bw_params->dc_mode_limit.dtbclk_mhz) { - if (i == 0 || dc_clk_table->entries[i-1].dtbclk_mhz < dc_bw_params->dc_mode_limit.dtbclk_mhz) { - dml_clk_table->dtbclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dtbclk_mhz * 1000; - dml_clk_table->dtbclk.num_clk_values = i + 1; - } else { - dml_clk_table->dtbclk.clk_values_khz[i] = 0; - dml_clk_table->dtbclk.num_clk_values = i; - } + /* dtbclk */ + if (dc_clk_table->num_entries_per_clk.num_dtbclk_levels) { + dml_clk_table->dtbclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_dtbclk_levels; + for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { + if (i < dml_clk_table->dtbclk.num_clk_values) { + if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.dtbclk_mhz && + dc_clk_table->entries[i].dtbclk_mhz > dc_bw_params->dc_mode_limit.dtbclk_mhz) { + if (i == 0 || dc_clk_table->entries[i-1].dtbclk_mhz < dc_bw_params->dc_mode_limit.dtbclk_mhz) { + dml_clk_table->dtbclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.dtbclk_mhz * 1000; + dml_clk_table->dtbclk.num_clk_values = i + 1; } else { - dml_clk_table->dtbclk.clk_values_khz[i] = dc_clk_table->entries[i].dtbclk_mhz * 1000; + dml_clk_table->dtbclk.clk_values_khz[i] = 0; + dml_clk_table->dtbclk.num_clk_values = i; } } else { - dml_clk_table->dtbclk.clk_values_khz[i] = 0; + dml_clk_table->dtbclk.clk_values_khz[i] = dc_clk_table->entries[i].dtbclk_mhz * 1000; } + } else { + dml_clk_table->dtbclk.clk_values_khz[i] = 0; } } + } - /* socclk */ - if (dc_clk_table->num_entries_per_clk.num_socclk_levels) { - dml_clk_table->socclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_socclk_levels; - for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { - if (i < dml_clk_table->socclk.num_clk_values) { - if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.socclk_mhz && - dc_clk_table->entries[i].socclk_mhz > dc_bw_params->dc_mode_limit.socclk_mhz) { - if (i == 0 || dc_clk_table->entries[i-1].socclk_mhz < dc_bw_params->dc_mode_limit.socclk_mhz) { - dml_clk_table->socclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.socclk_mhz * 1000; - dml_clk_table->socclk.num_clk_values = i + 1; - } else { - dml_clk_table->socclk.clk_values_khz[i] = 0; - dml_clk_table->socclk.num_clk_values = i; - } + /* socclk */ + if (dc_clk_table->num_entries_per_clk.num_socclk_levels) { + dml_clk_table->socclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_socclk_levels; + for (i = 0; i < min(DML_MAX_CLK_TABLE_SIZE, MAX_NUM_DPM_LVL); i++) { + if (i < dml_clk_table->socclk.num_clk_values) { + if (config->use_clock_dc_limits && dc_bw_params->dc_mode_limit.socclk_mhz && + dc_clk_table->entries[i].socclk_mhz > dc_bw_params->dc_mode_limit.socclk_mhz) { + if (i == 0 || dc_clk_table->entries[i-1].socclk_mhz < dc_bw_params->dc_mode_limit.socclk_mhz) { + dml_clk_table->socclk.clk_values_khz[i] = dc_bw_params->dc_mode_limit.socclk_mhz * 1000; + dml_clk_table->socclk.num_clk_values = i + 1; } else { - dml_clk_table->socclk.clk_values_khz[i] = dc_clk_table->entries[i].socclk_mhz * 1000; + dml_clk_table->socclk.clk_values_khz[i] = 0; + dml_clk_table->socclk.num_clk_values = i; } } else { - dml_clk_table->socclk.clk_values_khz[i] = 0; + dml_clk_table->socclk.clk_values_khz[i] = dc_clk_table->entries[i].socclk_mhz * 1000; } + } else { + dml_clk_table->socclk.clk_values_khz[i] = 0; } } - - /* do not override phyclks for now */ - /* phyclk */ - // dml_clk_table->phyclk.num_clk_values = dc_clk_table->num_entries_per_clk.num_phyclk_levels; - // for (i = 0; i < DML_MAX_CLK_TABLE_SIZE; i++) { - // dml_clk_table->phyclk.clk_values_khz[i] = dc_clk_table->entries[i].phyclk_mhz * 1000; - // } - - /* phyclk_d18 */ - // dml_clk_table->phyclk_d18.num_clk_values = dc_clk_table->num_entries_per_clk.num_phyclk_d18_levels; - // for (i = 0; i < DML_MAX_CLK_TABLE_SIZE; i++) { - // dml_clk_table->phyclk_d18.clk_values_khz[i] = dc_clk_table->entries[i].phyclk_d18_mhz * 1000; - // } - - /* phyclk_d32 */ - // dml_clk_table->phyclk_d32.num_clk_values = dc_clk_table->num_entries_per_clk.num_phyclk_d32_levels; - // for (i = 0; i < DML_MAX_CLK_TABLE_SIZE; i++) { - // dml_clk_table->phyclk_d32.clk_values_khz[i] = dc_clk_table->entries[i].phyclk_d32_mhz * 1000; - // } } +} - dml_soc_bb->dchub_refclk_mhz = in_dc->res_pool->ref_clocks.dchub_ref_clock_inKhz / 1000; - dml_soc_bb->dprefclk_mhz = in_dc->clk_mgr->dprefclk_khz / 1000; - dml_soc_bb->xtalclk_mhz = in_dc->ctx->dc_bios->fw_info.pll_info.crystal_frequency / 1000; - dml_soc_bb->dispclk_dppclk_vco_speed_mhz = in_dc->clk_mgr->dentist_vco_freq_khz / 1000.0; +static void override_dml_init_with_values_from_vbios( + struct dml2_initialize_instance_in_out *dml_init, + const struct dml2_configuration_options *config, + const struct dc *in_dc) +{ + const struct clk_bw_params *dc_bw_params = in_dc->clk_mgr->bw_params; + struct dml2_soc_bb *dml_soc_bb = &dml_init->soc_bb; + struct dml2_soc_state_table *dml_clk_table = &dml_init->soc_bb.clk_table; - /* override bounding box paramters from VBIOS */ if (in_dc->ctx->dc_bios->bb_info.dram_clock_change_latency_100ns > 0) dml_soc_bb->power_management_parameters.dram_clk_change_blackout_us = (in_dc->ctx->dc_bios->bb_info.dram_clock_change_latency_100ns + 9) / 10; @@ -308,32 +293,120 @@ void dml21_apply_soc_bb_overrides(struct dml2_initialize_instance_in_out *dml_in dml_clk_table->dram_config.channel_width_bytes = in_dc->ctx->dc_bios->vram_info.dram_channel_width_bytes; } - /* override bounding box paramters from DC config */ - if (in_dc->bb_overrides.sr_exit_time_ns) { - dml_soc_bb->power_management_parameters.stutter_exit_latency_us = - in_dc->bb_overrides.sr_exit_time_ns / 1000.0; + dml_init->soc_bb.xtalclk_mhz = in_dc->ctx->dc_bios->fw_info.pll_info.crystal_frequency / 1000; +} + + +static void override_dml_init_with_values_from_dmub(struct dml2_initialize_instance_in_out *dml_init, + const struct dml2_configuration_options *config, + const struct dc *in_dc) +{ + /* + * TODO - There seems to be overlaps between the values overriden from + * dmub and vbios. Investigate and identify the values that DMUB needs + * to own. + */ +// const struct dmub_soc_bb_params *dmub_bb_params = +// (const struct dmub_soc_bb_params *)config->bb_from_dmub; + +// if (dmub_bb_params == NULL) +// return; + +// if (dmub_bb_params->dram_clk_change_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.dram_clk_change_blackout_us = +// (double) dmub_bb_params->dram_clk_change_blackout_ns / 1000.0; +// if (dmub_bb_params->dram_clk_change_read_only_ns > 0) +// dml_init->soc_bb.power_management_parameters.dram_clk_change_read_only_us = +// (double) dmub_bb_params->dram_clk_change_read_only_ns / 1000.0; +// if (dmub_bb_params->dram_clk_change_write_only_ns > 0) +// dml_init->soc_bb.power_management_parameters.dram_clk_change_write_only_us = +// (double) dmub_bb_params->dram_clk_change_write_only_ns / 1000.0; +// if (dmub_bb_params->fclk_change_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.fclk_change_blackout_us = +// (double) dmub_bb_params->fclk_change_blackout_ns / 1000.0; +// if (dmub_bb_params->g7_ppt_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.g7_ppt_blackout_us = +// (double) dmub_bb_params->g7_ppt_blackout_ns / 1000.0; +// if (dmub_bb_params->stutter_enter_plus_exit_latency_ns > 0) +// dml_init->soc_bb.power_management_parameters.stutter_enter_plus_exit_latency_us = +// (double) dmub_bb_params->stutter_enter_plus_exit_latency_ns / 1000.0; +// if (dmub_bb_params->stutter_exit_latency_ns > 0) +// dml_init->soc_bb.power_management_parameters.stutter_exit_latency_us = +// (double) dmub_bb_params->stutter_exit_latency_ns / 1000.0; +// if (dmub_bb_params->z8_stutter_enter_plus_exit_latency_ns > 0) +// dml_init->soc_bb.power_management_parameters.z8_stutter_enter_plus_exit_latency_us = +// (double) dmub_bb_params->z8_stutter_enter_plus_exit_latency_ns / 1000.0; +// if (dmub_bb_params->z8_stutter_exit_latency_ns > 0) +// dml_init->soc_bb.power_management_parameters.z8_stutter_exit_latency_us = +// (double) dmub_bb_params->z8_stutter_exit_latency_ns / 1000.0; +// if (dmub_bb_params->z8_min_idle_time_ns > 0) +// dml_init->soc_bb.power_management_parameters.z8_min_idle_time = +// (double) dmub_bb_params->z8_min_idle_time_ns / 1000.0; +// #ifndef TRIM_DML2_DCN6B_IP_SENSITIVE +// if (dmub_bb_params->type_b_dram_clk_change_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.lpddr5_dram_clk_change_blackout_us = +// (double) dmub_bb_params->type_b_dram_clk_change_blackout_ns / 1000.0; +// if (dmub_bb_params->type_b_ppt_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.lpddr5_ppt_blackout_us = +// (double) dmub_bb_params->type_b_ppt_blackout_ns / 1000.0; +// #else +// if (dmub_bb_params->type_b_dram_clk_change_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.type_b_dram_clk_change_blackout_us = +// (double) dmub_bb_params->type_b_dram_clk_change_blackout_ns / 1000.0; +// if (dmub_bb_params->type_b_ppt_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.type_b_ppt_blackout_us = +// (double) dmub_bb_params->type_b_ppt_blackout_ns / 1000.0; +// #endif +// if (dmub_bb_params->vmin_limit_dispclk_khz > 0) +// dml_init->soc_bb.vmin_limit.dispclk_khz = dmub_bb_params->vmin_limit_dispclk_khz; +// if (dmub_bb_params->vmin_limit_dcfclk_khz > 0) +// dml_init->soc_bb.vmin_limit.dcfclk_khz = dmub_bb_params->vmin_limit_dcfclk_khz; +// if (dmub_bb_params->g7_temperature_read_blackout_ns > 0) +// dml_init->soc_bb.power_management_parameters.g7_temperature_read_blackout_us = +// (double) dmub_bb_params->g7_temperature_read_blackout_ns / 1000.0; +} + +static void override_dml_init_with_values_from_software_policy(struct dml2_initialize_instance_in_out *dml_init, + const struct dml2_configuration_options *config, + const struct dc *in_dc) +{ + if (!config->use_native_soc_bb_construction) { + dml_init->soc_bb = config->external_socbb_ip_params->soc_bb; + dml_init->ip_caps = config->external_socbb_ip_params->ip_params; } - if (in_dc->bb_overrides.sr_enter_plus_exit_time_ns) { - dml_soc_bb->power_management_parameters.stutter_enter_plus_exit_latency_us = + if (in_dc->bb_overrides.sr_exit_time_ns) + dml_init->soc_bb.power_management_parameters.stutter_exit_latency_us = + in_dc->bb_overrides.sr_exit_time_ns / 1000.0; + + if (in_dc->bb_overrides.sr_enter_plus_exit_time_ns) + dml_init->soc_bb.power_management_parameters.stutter_enter_plus_exit_latency_us = in_dc->bb_overrides.sr_enter_plus_exit_time_ns / 1000.0; - } - if (in_dc->bb_overrides.dram_clock_change_latency_ns) { - dml_soc_bb->power_management_parameters.dram_clk_change_blackout_us = + if (in_dc->bb_overrides.dram_clock_change_latency_ns) + dml_init->soc_bb.power_management_parameters.dram_clk_change_blackout_us = in_dc->bb_overrides.dram_clock_change_latency_ns / 1000.0; - } - if (in_dc->bb_overrides.fclk_clock_change_latency_ns) { - dml_soc_bb->power_management_parameters.fclk_change_blackout_us = + if (in_dc->bb_overrides.fclk_clock_change_latency_ns) + dml_init->soc_bb.power_management_parameters.fclk_change_blackout_us = in_dc->bb_overrides.fclk_clock_change_latency_ns / 1000.0; - } +} - //TODO - // if (in_dc->bb_overrides.dummy_clock_change_latency_ns) { - // dml_soc_bb->power_management_parameters.dram_clk_change_blackout_us = - // in_dc->bb_overrides.dram_clock_change_latency_ns / 1000.0; - // } +void dml21_populate_dml_init_params(struct dml2_initialize_instance_in_out *dml_init, + const struct dml2_configuration_options *config, + const struct dc *in_dc) +{ + populate_default_dml_init_params(dml_init, config, in_dc); + + override_dml_init_with_values_from_hardware_default(dml_init, config, in_dc); + + override_dml_init_with_values_from_smu(dml_init, config, in_dc); + + override_dml_init_with_values_from_vbios(dml_init, config, in_dc); + + override_dml_init_with_values_from_dmub(dml_init, config, in_dc); + + override_dml_init_with_values_from_software_policy(dml_init, config, in_dc); } static unsigned int calc_max_hardware_v_total(const struct dc_stream_state *stream) @@ -726,7 +799,6 @@ static void populate_dml21_surface_config_from_plane_state( switch (plane_state->tiling_info.gfxversion) { case DcGfxVersion7: case DcGfxVersion8: - // Placeholder for programming the array_mode break; case DcGfxVersion9: case DcGfxVersion10: @@ -788,6 +860,7 @@ static void populate_dml21_plane_config_from_plane_state(struct dml2_context *dm plane->pixel_format = dml2_420_10; break; case SURFACE_PIXEL_FORMAT_GRPH_ARGB16161616: + case SURFACE_PIXEL_FORMAT_GRPH_ABGR16161616: case SURFACE_PIXEL_FORMAT_GRPH_ARGB16161616F: case SURFACE_PIXEL_FORMAT_GRPH_ABGR16161616F: plane->pixel_format = dml2_444_64; @@ -888,10 +961,8 @@ static void populate_dml21_plane_config_from_plane_state(struct dml2_context *dm case DC_CM2_GPU_MEM_SIZE_171717: plane->tdlut.tdlut_width_mode = dml2_tdlut_width_17_cube; break; - case DC_CM2_GPU_MEM_SIZE_333333: - plane->tdlut.tdlut_width_mode = dml2_tdlut_width_33_cube; - break; case DC_CM2_GPU_MEM_SIZE_TRANSFORMED: + default: //plane->tdlut.tdlut_width_mode = dml2_tdlut_width_flatten; // dml2_tdlut_width_flatten undefined break; } diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.h index 73a013be1e48..9880d3e0398e 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.h +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.h @@ -17,9 +17,7 @@ struct dml2_context; struct dml2_configuration_options; struct dml2_initialize_instance_in_out; -void dml21_apply_soc_bb_overrides(struct dml2_initialize_instance_in_out *dml_init, const struct dml2_configuration_options *config, const struct dc *in_dc); -void dml21_initialize_soc_bb_params(struct dml2_initialize_instance_in_out *dml_init, const struct dml2_configuration_options *config, const struct dc *in_dc); -void dml21_initialize_ip_params(struct dml2_initialize_instance_in_out *dml_init, const struct dml2_configuration_options *config, const struct dc *in_dc); +void dml21_populate_dml_init_params(struct dml2_initialize_instance_in_out *dml_init, const struct dml2_configuration_options *config, const struct dc *in_dc); bool dml21_map_dc_state_into_dml_display_cfg(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml_ctx); void dml21_copy_clocks_to_dc_state(struct dml2_context *in_ctx, struct dc_state *context); void dml21_extract_watermark_sets(const struct dc *in_dc, union dcn_watermark_set *watermarks, struct dml2_context *in_ctx); diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_utils.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_utils.c index 930e86cdb88a..ee721606b883 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_utils.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_utils.c @@ -384,6 +384,7 @@ void dml21_build_fams2_programming(const struct dc *dc, /* reset fams2 data */ memset(&context->bw_ctx.bw.dcn.fams2_stream_base_params, 0, sizeof(union dmub_cmd_fams2_config) * DML2_MAX_PLANES); memset(&context->bw_ctx.bw.dcn.fams2_stream_sub_params, 0, sizeof(union dmub_cmd_fams2_config) * DML2_MAX_PLANES); + memset(&context->bw_ctx.bw.dcn.fams2_stream_sub_params_v2, 0, sizeof(union dmub_fams2_stream_static_sub_state_v2) * DML2_MAX_PLANES); memset(&context->bw_ctx.bw.dcn.fams2_global_config, 0, sizeof(struct dmub_cmd_fams2_global_config)); if (dml_ctx->v21.mode_programming.programming->fams2_required) { @@ -414,9 +415,16 @@ void dml21_build_fams2_programming(const struct dc *dc, memcpy(static_base_state, &dml_ctx->v21.mode_programming.programming->stream_programming[dml_stream_idx].fams2_base_params, sizeof(union dmub_cmd_fams2_config)); - memcpy(static_sub_state, - &dml_ctx->v21.mode_programming.programming->stream_programming[dml_stream_idx].fams2_sub_params, - sizeof(union dmub_cmd_fams2_config)); + + if (dc->debug.fams_version.major == 3) { + memcpy(&context->bw_ctx.bw.dcn.fams2_stream_sub_params_v2[num_fams2_streams], + &dml_ctx->v21.mode_programming.programming->stream_programming[dml_stream_idx].fams2_sub_params_v2, + sizeof(union dmub_fams2_stream_static_sub_state_v2)); + } else { + memcpy(static_sub_state, + &dml_ctx->v21.mode_programming.programming->stream_programming[dml_stream_idx].fams2_sub_params, + sizeof(union dmub_cmd_fams2_config)); + } switch (dc->debug.fams_version.minor) { case 1: diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c index 208d3651b6ba..03de3cf06ae5 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c @@ -2,8 +2,6 @@ // // Copyright 2024 Advanced Micro Devices, Inc. -#include <linux/vmalloc.h> - #include "dml2_internal_types.h" #include "dml_top.h" #include "dml2_core_dcn4_calcs.h" @@ -37,15 +35,11 @@ static bool dml21_allocate_memory(struct dml2_context **dml_ctx) return true; } -static void dml21_apply_debug_options(const struct dc *in_dc, struct dml2_context *dml_ctx, const struct dml2_configuration_options *config) +static void dml21_populate_configuration_options(const struct dc *in_dc, + struct dml2_context *dml_ctx, + const struct dml2_configuration_options *config) { - bool disable_fams2; - struct dml2_pmo_options *pmo_options = &dml_ctx->v21.dml_init.options.pmo_options; - - /* ODM options */ - pmo_options->disable_dyn_odm = !config->minimize_dispclk_using_odm; - pmo_options->disable_dyn_odm_for_multi_stream = true; - pmo_options->disable_dyn_odm_for_stream_with_svp = true; + dml_ctx->config = *config; /* UCLK P-State options */ if (in_dc->debug.dml21_force_pstate_method) { @@ -55,52 +49,20 @@ static void dml21_apply_debug_options(const struct dc *in_dc, struct dml2_contex } else { dml_ctx->config.pmo.force_pstate_method_enable = false; } - - pmo_options->disable_vblank = ((in_dc->debug.dml21_disable_pstate_method_mask >> 1) & 1); - - /* NOTE: DRR and SubVP Require FAMS2 */ - disable_fams2 = !in_dc->debug.fams2_config.bits.enable; - pmo_options->disable_svp = ((in_dc->debug.dml21_disable_pstate_method_mask >> 2) & 1) || - in_dc->debug.force_disable_subvp || - disable_fams2; - pmo_options->disable_drr_clamped = ((in_dc->debug.dml21_disable_pstate_method_mask >> 3) & 1) || - disable_fams2; - pmo_options->disable_drr_var = ((in_dc->debug.dml21_disable_pstate_method_mask >> 4) & 1) || - disable_fams2; - pmo_options->disable_fams2 = disable_fams2; - - pmo_options->disable_drr_var_when_var_active = in_dc->debug.disable_fams_gaming == INGAME_FAMS_DISABLE || - in_dc->debug.disable_fams_gaming == INGAME_FAMS_MULTI_DISP_CLAMPED_ONLY; - pmo_options->disable_drr_clamped_when_var_active = in_dc->debug.disable_fams_gaming == INGAME_FAMS_DISABLE; } -static void dml21_init(const struct dc *in_dc, struct dml2_context **dml_ctx, const struct dml2_configuration_options *config) +static void dml21_init(const struct dc *in_dc, struct dml2_context *dml_ctx, const struct dml2_configuration_options *config) { - switch (in_dc->ctx->dce_version) { - case DCN_VERSION_4_01: - (*dml_ctx)->v21.dml_init.options.project_id = dml2_project_dcn4x_stage2_auto_drr_svp; - break; - default: - (*dml_ctx)->v21.dml_init.options.project_id = dml2_project_invalid; - } - (*dml_ctx)->architecture = dml2_architecture_21; + dml_ctx->architecture = dml2_architecture_21; - /* Store configuration options */ - (*dml_ctx)->config = *config; + dml21_populate_configuration_options(in_dc, dml_ctx, config); DC_FP_START(); - /*Initialize SOCBB and DCNIP params */ - dml21_initialize_soc_bb_params(&(*dml_ctx)->v21.dml_init, config, in_dc); - dml21_initialize_ip_params(&(*dml_ctx)->v21.dml_init, config, in_dc); - dml21_apply_soc_bb_overrides(&(*dml_ctx)->v21.dml_init, config, in_dc); - - /* apply debug overrides */ - dml21_apply_debug_options(in_dc, *dml_ctx, config); + dml21_populate_dml_init_params(&dml_ctx->v21.dml_init, config, in_dc); - /*Initialize DML21 instance */ - dml2_initialize_instance(&(*dml_ctx)->v21.dml_init); + dml2_initialize_instance(&dml_ctx->v21.dml_init); DC_FP_END(); } @@ -111,7 +73,7 @@ bool dml21_create(const struct dc *in_dc, struct dml2_context **dml_ctx, const s if (!dml21_allocate_memory(dml_ctx)) return false; - dml21_init(in_dc, dml_ctx, config); + dml21_init(in_dc, *dml_ctx, config); return true; } @@ -328,12 +290,13 @@ static bool dml21_check_mode_support(const struct dc *in_dc, struct dc_state *co return true; } -bool dml21_validate(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml_ctx, bool fast_validate) +bool dml21_validate(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml_ctx, + enum dc_validate_mode validate_mode) { bool out = false; - /* Use dml_validate_only for fast_validate path */ - if (fast_validate) + /* Use dml21_check_mode_support for DC_VALIDATE_MODE_ONLY and DC_VALIDATE_MODE_AND_STATE_INDEX path */ + if (validate_mode != DC_VALIDATE_MODE_AND_PROGRAMMING) out = dml21_check_mode_support(in_dc, context, dml_ctx); else out = dml21_mode_check_and_programming(in_dc, context, dml_ctx); @@ -496,7 +459,7 @@ bool dml21_create_copy(struct dml2_context **dst_dml_ctx, return true; } -void dml21_reinit(const struct dc *in_dc, struct dml2_context **dml_ctx, const struct dml2_configuration_options *config) +void dml21_reinit(const struct dc *in_dc, struct dml2_context *dml_ctx, const struct dml2_configuration_options *config) { dml21_init(in_dc, dml_ctx, config); } diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.h index 42e715024bc9..15f92029d2e5 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.h +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.h @@ -14,6 +14,7 @@ struct dc; struct dc_state; struct dml2_configuration_options; struct dml2_context; +enum dc_validate_mode; /** * dml2_create - Creates dml21_context. @@ -33,22 +34,23 @@ void dml21_copy(struct dml2_context *dst_dml_ctx, struct dml2_context *src_dml_ctx); bool dml21_create_copy(struct dml2_context **dst_dml_ctx, struct dml2_context *src_dml_ctx); -void dml21_reinit(const struct dc *in_dc, struct dml2_context **dml_ctx, const struct dml2_configuration_options *config); +void dml21_reinit(const struct dc *in_dc, struct dml2_context *dml_ctx, const struct dml2_configuration_options *config); /** * dml21_validate - Determines if a display configuration is supported or not. * @in_dc: dc. * @context: dc_state to be validated. - * @fast_validate: Fast validate will not populate context.res_ctx. + * @validate_mode: DC_VALIDATE_MODE_ONLY and DC_VALIDATE_MODE_AND_STATE_INDEX + * will not populate context.res_ctx. * * Based on fast_validate option internally would call: * - * -dml21_mode_check_and_programming - for non fast_validate option + * -dml21_mode_check_and_programming - for DC_VALIDATE_MODE_AND_PROGRAMMING option * Calculates if dc_state can be supported on the input display * configuration. If supported, generates the necessary HW * programming for the new dc_state. * - * -dml21_check_mode_support - for fast_validate option + * -dml21_check_mode_support - for DC_VALIDATE_MODE_ONLY and DC_VALIDATE_MODE_AND_STATE_INDEX option * Calculates if dc_state can be supported for the input display * config. @@ -56,7 +58,8 @@ void dml21_reinit(const struct dc *in_dc, struct dml2_context **dml_ctx, const s * separate dc_states for validation. * Return: True if mode is supported, false otherwise. */ -bool dml21_validate(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml_ctx, bool fast_validate); +bool dml21_validate(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml_ctx, + enum dc_validate_mode validate_mode); /* Prepare hubp mcache_regs for hubp mcache ID and split coordinate programming */ void dml21_prepare_mcache_programming(struct dc *in_dc, struct dc_state *context, struct dml2_context *dml_ctx); 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_dchub_registers.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_dchub_registers.h index 84c90050668c..b05030926ce8 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_dchub_registers.h +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_dchub_registers.h @@ -46,6 +46,7 @@ struct dml2_display_dlg_regs { uint32_t dst_y_delta_drq_limit; uint32_t refcyc_per_vm_dmdata; uint32_t dmdata_dl_delta; + uint32_t dst_y_svp_drq_limit; // MRQ uint32_t refcyc_per_meta_chunk_vblank_l; diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_display_cfg_types.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_display_cfg_types.h index 255f05de362c..e8dc6471c0be 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_display_cfg_types.h +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/inc/dml_top_display_cfg_types.h @@ -222,6 +222,7 @@ struct dml2_composition_cfg { struct { bool enabled; + bool upsp_enabled; struct { double h_ratio; double v_ratio; @@ -426,6 +427,7 @@ struct dml2_stream_parameters { struct dml2_display_cfg { bool gpuvm_enable; + bool ffbm_enable; bool hostvm_enable; // Allocate DET proportionally between streams based on pixel rate 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..8c9f414aa6bf 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 @@ -93,12 +93,15 @@ struct dml2_soc_power_management_parameters { double dram_clk_change_write_only_us; double fclk_change_blackout_us; double g7_ppt_blackout_us; + double g7_temperature_read_blackout_us; double stutter_enter_plus_exit_latency_us; double stutter_exit_latency_us; double z8_stutter_enter_plus_exit_latency_us; double z8_stutter_exit_latency_us; double z8_min_idle_time; double g6_temp_read_blackout_us[DML_MAX_CLK_TABLE_SIZE]; + double type_b_dram_clk_change_blackout_us; + double type_b_ppt_blackout_us; }; struct dml2_clk_table { @@ -130,6 +133,7 @@ struct dml2_soc_state_table { struct dml2_soc_vmin_clock_limits { unsigned long dispclk_khz; + unsigned long dcfclk_khz; }; struct dml2_soc_bb { @@ -138,6 +142,7 @@ struct dml2_soc_bb { struct dml2_soc_power_management_parameters power_management_parameters; struct dml2_soc_vmin_clock_limits vmin_limit; + double lower_bound_bandwidth_dchub; unsigned int dprefclk_mhz; unsigned int xtalclk_mhz; unsigned int pcie_refclk_mhz; 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..98c0234e2f47 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 @@ -53,7 +53,9 @@ enum dml2_output_type_and_rate__rate { dml2_output_rate_hdmi_rate_6x4 = 9, dml2_output_rate_hdmi_rate_8x4 = 10, dml2_output_rate_hdmi_rate_10x4 = 11, - dml2_output_rate_hdmi_rate_12x4 = 12 + dml2_output_rate_hdmi_rate_12x4 = 12, + dml2_output_rate_hdmi_rate_16x4 = 13, + dml2_output_rate_hdmi_rate_20x4 = 14 }; struct dml2_pmo_options { @@ -279,7 +281,10 @@ 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 +679,14 @@ 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 c4dad7164d31..b9cff2198511 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 @@ -4685,7 +4685,10 @@ static void calculate_tdlut_setting( //the tdlut is fetched during the 2 row times of prefetch. if (p->setup_for_tdlut) { *p->tdlut_groups_per_2row_ub = (unsigned int)math_ceil2((double) *p->tdlut_bytes_per_frame / *p->tdlut_bytes_per_group, 1); - *p->tdlut_opt_time = (*p->tdlut_bytes_per_frame - p->cursor_buffer_size * 1024) / tdlut_drain_rate; + if (*p->tdlut_bytes_per_frame > p->cursor_buffer_size * 1024) + *p->tdlut_opt_time = (*p->tdlut_bytes_per_frame - p->cursor_buffer_size * 1024) / tdlut_drain_rate; + else + *p->tdlut_opt_time = 0; *p->tdlut_drain_time = p->cursor_buffer_size * 1024 / tdlut_drain_rate; *p->tdlut_bytes_to_deliver = (unsigned int) (p->cursor_buffer_size * 1024.0); } @@ -4858,7 +4861,7 @@ static double get_urgent_bandwidth_required( double ReadBandwidthChroma[], double PrefetchBandwidthLuma[], double PrefetchBandwidthChroma[], - double PrefetchBandwidthOto[], + double PrefetchBandwidthMax[], double excess_vactive_fill_bw_l[], double excess_vactive_fill_bw_c[], double cursor_bw[], @@ -4922,9 +4925,9 @@ static double get_urgent_bandwidth_required( l->vm_row_bw = NumberOfDPP[k] * prefetch_vmrow_bw[k]; l->flip_and_active_bw = l->per_plane_flip_bw[k] + ReadBandwidthLuma[k] * l->adj_factor_p0 + ReadBandwidthChroma[k] * l->adj_factor_p1 + cursor_bw[k] * l->adj_factor_cur; l->flip_and_prefetch_bw = l->per_plane_flip_bw[k] + NumberOfDPP[k] * (PrefetchBandwidthLuma[k] * l->adj_factor_p0_pre + PrefetchBandwidthChroma[k] * l->adj_factor_p1_pre) + prefetch_cursor_bw[k] * l->adj_factor_cur_pre; - l->flip_and_prefetch_bw_oto = l->per_plane_flip_bw[k] + NumberOfDPP[k] * (PrefetchBandwidthOto[k] * l->adj_factor_p0_pre + PrefetchBandwidthChroma[k] * l->adj_factor_p1_pre) + prefetch_cursor_bw[k] * l->adj_factor_cur_pre; + l->flip_and_prefetch_bw_max = l->per_plane_flip_bw[k] + NumberOfDPP[k] * (PrefetchBandwidthMax[k] * l->adj_factor_p0_pre + PrefetchBandwidthChroma[k] * l->adj_factor_p1_pre) + prefetch_cursor_bw[k] * l->adj_factor_cur_pre; l->active_and_excess_bw = (ReadBandwidthLuma[k] + excess_vactive_fill_bw_l[k]) * l->tmp_nom_adj_factor_p0 + (ReadBandwidthChroma[k] + excess_vactive_fill_bw_c[k]) * l->tmp_nom_adj_factor_p1 + dpte_row_bw[k] + meta_row_bw[k]; - surface_required_bw[k] = math_max5(l->vm_row_bw, l->flip_and_active_bw, l->flip_and_prefetch_bw, l->active_and_excess_bw, l->flip_and_prefetch_bw_oto); + surface_required_bw[k] = math_max5(l->vm_row_bw, l->flip_and_active_bw, l->flip_and_prefetch_bw, l->active_and_excess_bw, l->flip_and_prefetch_bw_max); /* export peak required bandwidth for the surface */ surface_peak_required_bw[k] = math_max2(surface_required_bw[k], surface_peak_required_bw[k]); @@ -5122,7 +5125,7 @@ static bool CalculatePrefetchSchedule(struct dml2_core_internal_scratch *scratch s->Tsw_est3 = 0.0; s->cursor_prefetch_bytes = 0; *p->prefetch_cursor_bw = 0; - *p->RequiredPrefetchBWOTO = 0.0; + *p->RequiredPrefetchBWMax = 0.0; dcc_mrq_enable = (p->dcc_enable && p->mrq_present); @@ -5353,7 +5356,7 @@ static bool CalculatePrefetchSchedule(struct dml2_core_internal_scratch *scratch * mp will fail if ms decides to use equ schedule and mp decides to use oto schedule * and the required bandwidth increases when going from ms to mp */ - *p->RequiredPrefetchBWOTO = s->prefetch_bw_oto; + *p->RequiredPrefetchBWMax = s->prefetch_bw_oto; #ifdef __DML_VBA_DEBUG__ DML_LOG_VERBOSE("DML::%s: vactive_sw_bw_l = %f\n", __func__, p->vactive_sw_bw_l); @@ -5715,8 +5718,14 @@ static bool CalculatePrefetchSchedule(struct dml2_core_internal_scratch *scratch s->TimeForFetchingVM = s->Tvm_equ; s->TimeForFetchingRowInVBlank = s->Tr0_equ; - *p->dst_y_per_vm_vblank = math_ceil2(4.0 * s->TimeForFetchingVM / s->LineTime, 1.0) / 4.0; - *p->dst_y_per_row_vblank = math_ceil2(4.0 * s->TimeForFetchingRowInVBlank / s->LineTime, 1.0) / 4.0; + *p->dst_y_per_vm_vblank = math_ceil2(4.0 * s->TimeForFetchingVM / s->LineTime, 1.0) / 4.0; + *p->dst_y_per_row_vblank = math_ceil2(4.0 * s->TimeForFetchingRowInVBlank / s->LineTime, 1.0) / 4.0; + + /* equ bw should be propagated so a ceiling of the equ bw is accounted for prior to mode programming. + * Overall bandwidth may be lower when going from mode support to mode programming but final pixel data + * bandwidth may end up higher than what was calculated in mode support. + */ + *p->RequiredPrefetchBWMax = math_max2(s->prefetch_bw_equ, *p->RequiredPrefetchBWMax); #ifdef __DML_VBA_DEBUG__ DML_LOG_VERBOSE("DML::%s: Using equ bw scheduling for prefetch\n", __func__); @@ -6112,7 +6121,7 @@ static void calculate_peak_bandwidth_required( p->surface_read_bandwidth_c, l->zero_array, //PrefetchBandwidthLuma, l->zero_array, //PrefetchBandwidthChroma, - l->zero_array, //PrefetchBWOTO + l->zero_array, //PrefetchBWMax l->zero_array, l->zero_array, l->zero_array, @@ -6149,7 +6158,7 @@ static void calculate_peak_bandwidth_required( p->surface_read_bandwidth_c, l->zero_array, //PrefetchBandwidthLuma, l->zero_array, //PrefetchBandwidthChroma, - l->zero_array, //PrefetchBWOTO + l->zero_array, //PrefetchBWMax p->excess_vactive_fill_bw_l, p->excess_vactive_fill_bw_c, p->cursor_bw, @@ -6186,7 +6195,7 @@ static void calculate_peak_bandwidth_required( p->surface_read_bandwidth_c, p->prefetch_bandwidth_l, p->prefetch_bandwidth_c, - p->prefetch_bandwidth_oto, // to prevent ms/mp mismatch when oto bw > total vactive bw + p->prefetch_bandwidth_max, // to prevent ms/mp mismatches where mp prefetch bw > ms prefetch bw p->excess_vactive_fill_bw_l, p->excess_vactive_fill_bw_c, p->cursor_bw, @@ -6223,7 +6232,7 @@ static void calculate_peak_bandwidth_required( p->surface_read_bandwidth_c, p->prefetch_bandwidth_l, p->prefetch_bandwidth_c, - p->prefetch_bandwidth_oto, // to prevent ms/mp mismatch when oto bw > total vactive bw + p->prefetch_bandwidth_max, // to prevent ms/mp mismatch where mp prefetch bw > ms prefetch bw p->excess_vactive_fill_bw_l, p->excess_vactive_fill_bw_c, p->cursor_bw, @@ -6260,7 +6269,7 @@ static void calculate_peak_bandwidth_required( p->surface_read_bandwidth_c, p->prefetch_bandwidth_l, p->prefetch_bandwidth_c, - p->prefetch_bandwidth_oto, // to prevent ms/mp mismatch when oto bw > total vactive bw + p->prefetch_bandwidth_max, // to prevent ms/mp mismatches where mp prefetch bw > ms prefetch bw p->excess_vactive_fill_bw_l, p->excess_vactive_fill_bw_c, p->cursor_bw, @@ -7487,7 +7496,7 @@ static noinline_for_stack void dml_core_ms_prefetch_check(struct dml2_core_inter CalculatePrefetchSchedule_params->VRatioPrefetchC = &mode_lib->ms.VRatioPreC[k]; CalculatePrefetchSchedule_params->RequiredPrefetchPixelDataBWLuma = &mode_lib->ms.RequiredPrefetchPixelDataBWLuma[k]; // prefetch_sw_bw_l CalculatePrefetchSchedule_params->RequiredPrefetchPixelDataBWChroma = &mode_lib->ms.RequiredPrefetchPixelDataBWChroma[k]; // prefetch_sw_bw_c - CalculatePrefetchSchedule_params->RequiredPrefetchBWOTO = &mode_lib->ms.RequiredPrefetchBWOTO[k]; + CalculatePrefetchSchedule_params->RequiredPrefetchBWMax = &mode_lib->ms.RequiredPrefetchBWMax[k]; CalculatePrefetchSchedule_params->NotEnoughTimeForDynamicMetadata = &mode_lib->ms.NoTimeForDynamicMetadata[k]; CalculatePrefetchSchedule_params->Tno_bw = &mode_lib->ms.Tno_bw[k]; CalculatePrefetchSchedule_params->Tno_bw_flip = &mode_lib->ms.Tno_bw_flip[k]; @@ -7632,7 +7641,7 @@ static noinline_for_stack void dml_core_ms_prefetch_check(struct dml2_core_inter calculate_peak_bandwidth_params->surface_read_bandwidth_c = mode_lib->ms.vactive_sw_bw_c; calculate_peak_bandwidth_params->prefetch_bandwidth_l = mode_lib->ms.RequiredPrefetchPixelDataBWLuma; calculate_peak_bandwidth_params->prefetch_bandwidth_c = mode_lib->ms.RequiredPrefetchPixelDataBWChroma; - calculate_peak_bandwidth_params->prefetch_bandwidth_oto = mode_lib->ms.RequiredPrefetchBWOTO; + calculate_peak_bandwidth_params->prefetch_bandwidth_max = mode_lib->ms.RequiredPrefetchBWMax; calculate_peak_bandwidth_params->excess_vactive_fill_bw_l = mode_lib->ms.excess_vactive_fill_bw_l; calculate_peak_bandwidth_params->excess_vactive_fill_bw_c = mode_lib->ms.excess_vactive_fill_bw_c; calculate_peak_bandwidth_params->cursor_bw = mode_lib->ms.cursor_bw; @@ -7799,7 +7808,7 @@ static noinline_for_stack void dml_core_ms_prefetch_check(struct dml2_core_inter calculate_peak_bandwidth_params->surface_read_bandwidth_c = mode_lib->ms.vactive_sw_bw_c; calculate_peak_bandwidth_params->prefetch_bandwidth_l = mode_lib->ms.RequiredPrefetchPixelDataBWLuma; calculate_peak_bandwidth_params->prefetch_bandwidth_c = mode_lib->ms.RequiredPrefetchPixelDataBWChroma; - calculate_peak_bandwidth_params->prefetch_bandwidth_oto = mode_lib->ms.RequiredPrefetchBWOTO; + calculate_peak_bandwidth_params->prefetch_bandwidth_max = mode_lib->ms.RequiredPrefetchBWMax; calculate_peak_bandwidth_params->excess_vactive_fill_bw_l = mode_lib->ms.excess_vactive_fill_bw_l; calculate_peak_bandwidth_params->excess_vactive_fill_bw_c = mode_lib->ms.excess_vactive_fill_bw_c; calculate_peak_bandwidth_params->cursor_bw = mode_lib->ms.cursor_bw; @@ -7905,6 +7914,7 @@ 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; @@ -11253,7 +11263,7 @@ static bool dml_core_mode_programming(struct dml2_core_calcs_mode_programming_ex CalculatePrefetchSchedule_params->VRatioPrefetchC = &mode_lib->mp.VRatioPrefetchC[k]; CalculatePrefetchSchedule_params->RequiredPrefetchPixelDataBWLuma = &mode_lib->mp.RequiredPrefetchPixelDataBWLuma[k]; CalculatePrefetchSchedule_params->RequiredPrefetchPixelDataBWChroma = &mode_lib->mp.RequiredPrefetchPixelDataBWChroma[k]; - CalculatePrefetchSchedule_params->RequiredPrefetchBWOTO = &s->dummy_single_array[0][k]; + CalculatePrefetchSchedule_params->RequiredPrefetchBWMax = &s->dummy_single_array[0][k]; CalculatePrefetchSchedule_params->NotEnoughTimeForDynamicMetadata = &mode_lib->mp.NotEnoughTimeForDynamicMetadata[k]; CalculatePrefetchSchedule_params->Tno_bw = &mode_lib->mp.Tno_bw[k]; CalculatePrefetchSchedule_params->Tno_bw_flip = &mode_lib->mp.Tno_bw_flip[k]; @@ -11396,7 +11406,7 @@ static bool dml_core_mode_programming(struct dml2_core_calcs_mode_programming_ex calculate_peak_bandwidth_params->surface_read_bandwidth_c = mode_lib->mp.vactive_sw_bw_c; calculate_peak_bandwidth_params->prefetch_bandwidth_l = mode_lib->mp.RequiredPrefetchPixelDataBWLuma; calculate_peak_bandwidth_params->prefetch_bandwidth_c = mode_lib->mp.RequiredPrefetchPixelDataBWChroma; - calculate_peak_bandwidth_params->prefetch_bandwidth_oto = s->dummy_single_array[0]; + calculate_peak_bandwidth_params->prefetch_bandwidth_max = s->dummy_single_array[0]; calculate_peak_bandwidth_params->excess_vactive_fill_bw_l = mode_lib->mp.excess_vactive_fill_bw_l; calculate_peak_bandwidth_params->excess_vactive_fill_bw_c = mode_lib->mp.excess_vactive_fill_bw_c; calculate_peak_bandwidth_params->cursor_bw = mode_lib->mp.cursor_bw; @@ -11536,7 +11546,7 @@ static bool dml_core_mode_programming(struct dml2_core_calcs_mode_programming_ex calculate_peak_bandwidth_params->meta_row_bw = mode_lib->mp.meta_row_bw; calculate_peak_bandwidth_params->prefetch_cursor_bw = mode_lib->mp.prefetch_cursor_bw; calculate_peak_bandwidth_params->prefetch_vmrow_bw = mode_lib->mp.prefetch_vmrow_bw; - calculate_peak_bandwidth_params->prefetch_bandwidth_oto = s->dummy_single_array[0]; + calculate_peak_bandwidth_params->prefetch_bandwidth_max = s->dummy_single_array[0]; calculate_peak_bandwidth_params->flip_bw = mode_lib->mp.final_flip_bw; calculate_peak_bandwidth_params->urgent_burst_factor_l = mode_lib->mp.UrgentBurstFactorLuma; calculate_peak_bandwidth_params->urgent_burst_factor_c = mode_lib->mp.UrgentBurstFactorChroma; @@ -11880,7 +11890,7 @@ 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 +11899,7 @@ 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; } } @@ -13059,6 +13069,10 @@ void dml2_core_calcs_get_informative(const struct dml2_core_internal_display_mod out->informative.mode_support_info.OutputRate[k] = dml2_output_rate_hdmi_rate_10x4; else if (mode_lib->ms.support.OutputRate[k] == dml2_core_internal_output_rate_hdmi_rate_12x4) out->informative.mode_support_info.OutputRate[k] = dml2_output_rate_hdmi_rate_12x4; + else if (mode_lib->ms.support.OutputRate[k] == dml2_core_internal_output_rate_hdmi_rate_16x4) + out->informative.mode_support_info.OutputRate[k] = dml2_output_rate_hdmi_rate_16x4; + else if (mode_lib->ms.support.OutputRate[k] == dml2_core_internal_output_rate_hdmi_rate_20x4) + out->informative.mode_support_info.OutputRate[k] = dml2_output_rate_hdmi_rate_20x4; out->informative.mode_support_info.AlignedYPitch[k] = mode_lib->ms.support.AlignedYPitch[k]; out->informative.mode_support_info.AlignedCPitch[k] = mode_lib->ms.support.AlignedCPitch[k]; @@ -13243,7 +13257,7 @@ 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..28687565ac22 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 @@ -102,6 +102,7 @@ struct dml2_core_internal_DmlPipe { double DCFClkDeepSleep; unsigned int DPPPerSurface; bool ScalerEnabled; + bool UPSPEnabled; enum dml2_rotation_angle RotationAngle; bool mirrored; unsigned int ViewportHeight; @@ -186,7 +187,9 @@ enum dml2_core_internal_output_type_rate { dml2_core_internal_output_rate_hdmi_rate_6x4 = 9, dml2_core_internal_output_rate_hdmi_rate_8x4 = 10, dml2_core_internal_output_rate_hdmi_rate_10x4 = 11, - dml2_core_internal_output_rate_hdmi_rate_12x4 = 12 + dml2_core_internal_output_rate_hdmi_rate_12x4 = 12, + dml2_core_internal_output_rate_hdmi_rate_16x4 = 13, + dml2_core_internal_output_rate_hdmi_rate_20x4 = 14 }; struct dml2_core_internal_watermarks { @@ -260,12 +263,14 @@ struct dml2_core_internal_mode_support_info { bool AvgBandwidthSupport; bool UrgVactiveBandwidthSupport; bool EnoughUrgentLatencyHidingSupport; + bool PrefetchScheduleSupported; bool PrefetchSupported; bool PrefetchBandwidthSupported; bool DynamicMetadataSupported; bool VRatioInPrefetchSupported; bool DISPCLK_DPPCLK_Support; bool TotalAvailablePipesSupport; + bool ODMSupport; bool ModeSupport; bool ViewportSizeSupport; @@ -314,9 +319,7 @@ struct dml2_core_internal_mode_support_info { double non_urg_bandwidth_required[dml2_core_internal_soc_state_max][dml2_core_internal_bw_max]; // same as urg_bandwidth, except not scaled by urg burst factor double non_urg_bandwidth_required_flip[dml2_core_internal_soc_state_max][dml2_core_internal_bw_max]; - bool avg_bandwidth_support_ok[dml2_core_internal_soc_state_max][dml2_core_internal_bw_max]; - double max_urgent_latency_us; double max_non_urgent_latency_us; double avg_non_urgent_latency_us; @@ -329,6 +332,8 @@ struct dml2_core_internal_mode_support_info { bool temp_read_or_ppt_support; struct dml2_core_internal_watermarks watermarks; + bool dcfclk_support; + bool qos_bandwidth_support; }; struct dml2_core_internal_mode_support { @@ -350,9 +355,11 @@ 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 @@ -394,9 +401,13 @@ struct dml2_core_internal_mode_support { double TWait[DML2_MAX_PLANES]; bool UnboundedRequestEnabled; + unsigned int compbuf_reserved_space_64b; + bool hw_debug5; unsigned int CompressedBufferSizeInkByte; double VRatioPreY[DML2_MAX_PLANES]; double VRatioPreC[DML2_MAX_PLANES]; + unsigned int req_per_swath_ub_l[DML2_MAX_PLANES]; + unsigned int req_per_swath_ub_c[DML2_MAX_PLANES]; unsigned int swath_width_luma_ub[DML2_MAX_PLANES]; unsigned int swath_width_chroma_ub[DML2_MAX_PLANES]; unsigned int RequiredSlots[DML2_MAX_PLANES]; @@ -417,8 +428,8 @@ struct dml2_core_internal_mode_support { double dst_y_prefetch[DML2_MAX_PLANES]; double LinesForVM[DML2_MAX_PLANES]; double LinesForDPTERow[DML2_MAX_PLANES]; - double SwathWidthYSingleDPP[DML2_MAX_PLANES]; - double SwathWidthCSingleDPP[DML2_MAX_PLANES]; + unsigned int SwathWidthYSingleDPP[DML2_MAX_PLANES]; + unsigned int SwathWidthCSingleDPP[DML2_MAX_PLANES]; unsigned int BytePerPixelY[DML2_MAX_PLANES]; unsigned int BytePerPixelC[DML2_MAX_PLANES]; double BytePerPixelInDETY[DML2_MAX_PLANES]; @@ -469,13 +480,58 @@ struct dml2_core_internal_mode_support { double mall_prefetch_sdp_overhead_factor[DML2_MAX_PLANES]; // overhead to the imall or phantom pipe double mall_prefetch_dram_overhead_factor[DML2_MAX_PLANES]; + bool is_using_mall_for_ss[DML2_MAX_PLANES]; + unsigned int meta_row_width_chroma[DML2_MAX_PLANES]; + unsigned int PixelPTEReqHeightC[DML2_MAX_PLANES]; + bool PTE_BUFFER_MODE[DML2_MAX_PLANES]; + unsigned int meta_req_height_chroma[DML2_MAX_PLANES]; + unsigned int meta_pte_bytes_per_frame_ub_c[DML2_MAX_PLANES]; + unsigned int dpde0_bytes_per_frame_ub_c[DML2_MAX_PLANES]; + unsigned int dpte_row_width_luma_ub[DML2_MAX_PLANES]; + unsigned int meta_req_width[DML2_MAX_PLANES]; + unsigned int meta_row_width[DML2_MAX_PLANES]; + unsigned int PixelPTEReqWidthY[DML2_MAX_PLANES]; + unsigned int dpte_row_height_linear[DML2_MAX_PLANES]; + unsigned int PTERequestSizeY[DML2_MAX_PLANES]; + unsigned int dpte_row_width_chroma_ub[DML2_MAX_PLANES]; + unsigned int PixelPTEReqWidthC[DML2_MAX_PLANES]; + unsigned int meta_pte_bytes_per_frame_ub_l[DML2_MAX_PLANES]; + unsigned int dpte_row_height_linear_chroma[DML2_MAX_PLANES]; + unsigned int PTERequestSizeC[DML2_MAX_PLANES]; + unsigned int meta_req_height[DML2_MAX_PLANES]; + unsigned int dpde0_bytes_per_frame_ub_l[DML2_MAX_PLANES]; + unsigned int meta_req_width_chroma[DML2_MAX_PLANES]; + unsigned int PixelPTEReqHeightY[DML2_MAX_PLANES]; + unsigned int BIGK_FRAGMENT_SIZE[DML2_MAX_PLANES]; + unsigned int vm_group_bytes[DML2_MAX_PLANES]; + unsigned int VReadyOffsetPix[DML2_MAX_PLANES]; + unsigned int VUpdateOffsetPix[DML2_MAX_PLANES]; + unsigned int VUpdateWidthPix[DML2_MAX_PLANES]; + double TSetup[DML2_MAX_PLANES]; + double Tdmdl_vm_raw[DML2_MAX_PLANES]; + double Tdmdl_raw[DML2_MAX_PLANES]; + unsigned int VStartupMin[DML2_MAX_PLANES]; /// <brief Minimum vstartup to meet the prefetch schedule (i.e. the prefetch solution can be found at this vstartup time); not the actual global sync vstartup pos. + double MaxActiveDRAMClockChangeLatencySupported[DML2_MAX_PLANES]; + double MaxActiveFCLKChangeLatencySupported; + // Backend 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; + unsigned int EstimatedNumberOfDSCSlices[DML2_MAX_PLANES]; // Bandwidth Related Info double BandwidthAvailableForImmediateFlip; @@ -484,8 +540,14 @@ struct dml2_core_internal_mode_support { double WriteBandwidth[DML2_MAX_PLANES][DML2_MAX_WRITEBACK]; double RequiredPrefetchPixelDataBWLuma[DML2_MAX_PLANES]; double RequiredPrefetchPixelDataBWChroma[DML2_MAX_PLANES]; - /* oto bw should also be considered when calculating peak urgent bw to avoid situations oto/equ mismatches between ms and mp */ - double RequiredPrefetchBWOTO[DML2_MAX_PLANES]; + /* Max bandwidth calculated from prefetch schedule should be considered in addition to the pixel data bw to avoid ms/mp mismatches. + * 1. oto bw should also be considered when calculating peak urgent bw to avoid situations oto/equ mismatches between ms and mp + * + * 2. equ bandwidth needs to be considered for calculating peak urgent bw when equ schedule is used in mode support. + * Some slight difference in variables may cause the pixel data bandwidth to be higher + * even though overall equ prefetch bandwidths can be lower going from ms to mp + */ + double RequiredPrefetchBWMax[DML2_MAX_PLANES]; double cursor_bw[DML2_MAX_PLANES]; double prefetch_cursor_bw[DML2_MAX_PLANES]; double prefetch_vmrow_bw[DML2_MAX_PLANES]; @@ -538,7 +600,44 @@ 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]; + 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]; + unsigned int DSTYAfterScaler[DML2_MAX_PLANES]; + unsigned int DSTXAfterScaler[DML2_MAX_PLANES]; + + enum dml2_pstate_method pstate_switch_modes[DML2_MAX_PLANES]; }; /// @brief A mega structure that houses various info for model programming step. @@ -548,6 +647,7 @@ 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]; @@ -599,6 +699,8 @@ struct dml2_core_internal_mode_program { unsigned int MacroTileHeightC[DML2_MAX_PLANES]; unsigned int MacroTileWidthY[DML2_MAX_PLANES]; unsigned int MacroTileWidthC[DML2_MAX_PLANES]; + double MaximumSwathWidthLuma[DML2_MAX_PLANES]; + double MaximumSwathWidthChroma[DML2_MAX_PLANES]; bool surf_linear128_l[DML2_MAX_PLANES]; bool surf_linear128_c[DML2_MAX_PLANES]; @@ -631,6 +733,14 @@ struct dml2_core_internal_mode_program { double UrgentBurstFactorChroma[DML2_MAX_PLANES]; double UrgentBurstFactorChromaPre[DML2_MAX_PLANES]; + double MaximumSwathWidthInLineBufferLuma; + double MaximumSwathWidthInLineBufferChroma; + + 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]; + double meta_row_bw[DML2_MAX_PLANES]; unsigned int meta_row_bytes[DML2_MAX_PLANES]; unsigned int meta_req_width[DML2_MAX_PLANES]; @@ -652,7 +762,9 @@ struct dml2_core_internal_mode_program { unsigned int PTERequestSizeC[DML2_MAX_PLANES]; double TWait[DML2_MAX_PLANES]; + double Tdmdl_vm_raw[DML2_MAX_PLANES]; double Tdmdl_vm[DML2_MAX_PLANES]; + double Tdmdl_raw[DML2_MAX_PLANES]; double Tdmdl[DML2_MAX_PLANES]; double TSetup[DML2_MAX_PLANES]; unsigned int dpde0_bytes_per_frame_ub_l[DML2_MAX_PLANES]; @@ -684,6 +796,38 @@ 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 // ------------------- @@ -694,9 +838,12 @@ struct dml2_core_internal_mode_program { // Support bool UrgVactiveBandwidthSupport; + bool PrefetchScheduleSupported; + bool UrgentBandwidthSupport; bool PrefetchModeSupported; // <brief Is the prefetch mode (bandwidth and latency) supported bool ImmediateFlipSupported; bool ImmediateFlipSupportedForPipe[DML2_MAX_PLANES]; + bool dcfclk_support; // Clock double Dcfclk; @@ -788,7 +935,7 @@ struct dml2_core_internal_mode_program { // RQ registers bool PTE_BUFFER_MODE[DML2_MAX_PLANES]; unsigned int BIGK_FRAGMENT_SIZE[DML2_MAX_PLANES]; - + double VActiveLatencyHidingUs[DML2_MAX_PLANES]; unsigned int SubViewportLinesNeededInMALL[DML2_MAX_PLANES]; bool is_using_mall_for_ss[DML2_MAX_PLANES]; @@ -1001,10 +1148,10 @@ struct dml2_core_calcs_mode_programming_locals { double dummy_bw[dml2_core_internal_soc_state_max][dml2_core_internal_bw_max]; double surface_dummy_bw[dml2_core_internal_soc_state_max][dml2_core_internal_bw_max][DML2_MAX_PLANES]; double surface_dummy_bw0[dml2_core_internal_soc_state_max][dml2_core_internal_bw_max][DML2_MAX_PLANES]; - unsigned int dummy_integer_array[2][DML2_MAX_PLANES]; + unsigned int dummy_integer_array[4][DML2_MAX_PLANES]; enum dml2_output_encoder_class dummy_output_encoder_array[DML2_MAX_PLANES]; double dummy_single_array[2][DML2_MAX_PLANES]; - unsigned int dummy_long_array[4][DML2_MAX_PLANES]; + unsigned int dummy_long_array[8][DML2_MAX_PLANES]; bool dummy_boolean_array[2][DML2_MAX_PLANES]; bool dummy_boolean[2]; double dummy_single[2]; @@ -1028,7 +1175,6 @@ 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; @@ -1153,6 +1299,7 @@ struct dml2_core_calcs_CalculateVMRowAndSwath_params { unsigned int HostVMMinPageSize; unsigned int DCCMetaBufferSizeBytes; bool mrq_present; + enum dml2_pstate_method pstate_switch_modes[DML2_MAX_PLANES]; // Output bool *PTEBufferSizeNotExceeded; @@ -1389,7 +1536,7 @@ struct dml2_core_shared_get_urgent_bandwidth_required_locals { double vm_row_bw; double flip_and_active_bw; double flip_and_prefetch_bw; - double flip_and_prefetch_bw_oto; + double flip_and_prefetch_bw_max; double active_and_excess_bw; }; @@ -1418,6 +1565,7 @@ struct dml2_core_shared_CalculateFlipSchedule_locals { struct dml2_core_shared_rq_dlg_get_dlg_reg_locals { unsigned int plane_idx; + unsigned int stream_idx; enum dml2_source_format_class source_format; const struct dml2_timing_cfg *timing; bool dual_plane; @@ -1625,6 +1773,9 @@ struct dml2_core_calcs_CalculateSwathAndDETConfiguration_params { double *BytePerPixDETC; unsigned int *DPPPerSurface; bool mrq_present; + unsigned int dummy[2][DML2_MAX_PLANES]; + unsigned int swath_width_luma_ub_single_dpp[DML2_MAX_PLANES]; + unsigned int swath_width_chroma_ub_single_dpp[DML2_MAX_PLANES]; // output unsigned int *req_per_swath_ub_l; @@ -1642,6 +1793,8 @@ struct dml2_core_calcs_CalculateSwathAndDETConfiguration_params { unsigned int *DETBufferSizeC; unsigned int *full_swath_bytes_l; unsigned int *full_swath_bytes_c; + unsigned int *full_swath_bytes_single_dpp_l; + unsigned int *full_swath_bytes_single_dpp_c; bool *UnboundedRequestEnabled; unsigned int *compbuf_reserved_space_64b; unsigned int *CompressedBufferSizeInkByte; @@ -1801,7 +1954,7 @@ struct dml2_core_calcs_CalculatePrefetchSchedule_params { double *VRatioPrefetchC; double *RequiredPrefetchPixelDataBWLuma; double *RequiredPrefetchPixelDataBWChroma; - double *RequiredPrefetchBWOTO; + double *RequiredPrefetchBWMax; bool *NotEnoughTimeForDynamicMetadata; double *Tno_bw; double *Tno_bw_flip; @@ -2038,7 +2191,7 @@ struct dml2_core_calcs_calculate_peak_bandwidth_required_params { double *surface_read_bandwidth_c; double *prefetch_bandwidth_l; double *prefetch_bandwidth_c; - double *prefetch_bandwidth_oto; + double *prefetch_bandwidth_max; double *excess_vactive_fill_bw_l; double *excess_vactive_fill_bw_c; double *cursor_bw; diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_utils.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_utils.c index 7a220c0141c2..5f301befed16 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_utils.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_core/dml2_core_utils.c @@ -464,7 +464,7 @@ bool dml2_core_utils_get_segment_horizontal_contiguous(enum dml2_swizzle_mode sw bool dml2_core_utils_is_linear(enum dml2_swizzle_mode sw_mode) { - return (sw_mode == dml2_sw_linear || sw_mode == dml2_sw_linear_256b || sw_mode == dml2_linear_64elements); + return sw_mode == dml2_sw_linear; }; 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_debug.h b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/inc/dml2_debug.h index b226225103c3..611c80f4f1bf 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/inc/dml2_debug.h +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/src/inc/dml2_debug.h @@ -10,15 +10,74 @@ #define DML_LOG_LEVEL_DEFAULT DML_LOG_LEVEL_WARN #define DML_LOG_INTERNAL(fmt, ...) dm_output_to_console(fmt, ## __VA_ARGS__) -/* ASSERT with message output */ -#define DML_ASSERT_MSG(condition, fmt, ...) \ - do { \ - if (!(condition)) { \ - DML_LOG_ERROR("DML ASSERT hit in %s line %d\n", __func__, __LINE__); \ - DML_LOG_ERROR(fmt, ## __VA_ARGS__); \ - DML_ASSERT(condition); \ - } \ - } while (0) +/* private helper macros */ +#define _BOOL_FORMAT(field) "%s", field ? "true" : "false" +#define _UINT_FORMAT(field) "%u", field +#define _INT_FORMAT(field) "%d", field +#define _DOUBLE_FORMAT(field) "%lf", field +#define _ELEMENT_FUNC "function" +#define _ELEMENT_COMP_IF "component_interface" +#define _ELEMENT_TOP_IF "top_interface" +#define _LOG_ENTRY(element) do { \ + DML_LOG_INTERNAL("<"element" name=\""); \ + DML_LOG_INTERNAL(__func__); \ + DML_LOG_INTERNAL("\">\n"); \ +} while (0) +#define _LOG_EXIT(element) DML_LOG_INTERNAL("</"element">\n") +#define _LOG_SCALAR(field, format) do { \ + DML_LOG_INTERNAL(#field" = "format(field)); \ + DML_LOG_INTERNAL("\n"); \ +} while (0) +#define _LOG_ARRAY(field, size, format) do { \ + DML_LOG_INTERNAL(#field " = ["); \ + for (int _i = 0; _i < (int) size; _i++) { \ + DML_LOG_INTERNAL(format(field[_i])); \ + if (_i + 1 == (int) size) \ + DML_LOG_INTERNAL("]\n"); \ + else \ + DML_LOG_INTERNAL(", "); \ +}} while (0) +#define _LOG_2D_ARRAY(field, size0, size1, format) do { \ + DML_LOG_INTERNAL(#field" = ["); \ + for (int _i = 0; _i < (int) size0; _i++) { \ + DML_LOG_INTERNAL("\n\t["); \ + for (int _j = 0; _j < (int) size1; _j++) { \ + DML_LOG_INTERNAL(format(field[_i][_j])); \ + if (_j + 1 == (int) size1) \ + DML_LOG_INTERNAL("]"); \ + else \ + DML_LOG_INTERNAL(", "); \ + } \ + if (_i + 1 == (int) size0) \ + DML_LOG_INTERNAL("]\n"); \ + else \ + DML_LOG_INTERNAL(", "); \ + } \ +} while (0) +#define _LOG_3D_ARRAY(field, size0, size1, size2, format) do { \ + DML_LOG_INTERNAL(#field" = ["); \ + for (int _i = 0; _i < (int) size0; _i++) { \ + DML_LOG_INTERNAL("\n\t["); \ + for (int _j = 0; _j < (int) size1; _j++) { \ + DML_LOG_INTERNAL("["); \ + for (int _k = 0; _k < (int) size2; _k++) { \ + DML_LOG_INTERNAL(format(field[_i][_j][_k])); \ + if (_k + 1 == (int) size2) \ + DML_LOG_INTERNAL("]"); \ + else \ + DML_LOG_INTERNAL(", "); \ + } \ + if (_j + 1 == (int) size1) \ + DML_LOG_INTERNAL("]"); \ + else \ + DML_LOG_INTERNAL(", "); \ + } \ + if (_i + 1 == (int) size0) \ + DML_LOG_INTERNAL("]\n"); \ + else \ + DML_LOG_INTERNAL(", "); \ + } \ +} while (0) /* fatal errors for unrecoverable DML states until a full reset */ #define DML_LOG_LEVEL_FATAL 0 @@ -28,7 +87,7 @@ #define DML_LOG_LEVEL_WARN 2 /* high level tracing of DML interfaces */ #define DML_LOG_LEVEL_INFO 3 -/* detailed tracing of DML internal components */ +/* tracing of DML internal executions */ #define DML_LOG_LEVEL_DEBUG 4 /* detailed tracing of DML calculation procedure */ #define DML_LOG_LEVEL_VERBOSE 5 @@ -37,30 +96,94 @@ #define DML_LOG_LEVEL DML_LOG_LEVEL_DEFAULT #endif /* #ifndef DML_LOG_LEVEL */ +/* public macros for DML_LOG_LEVEL_FATAL and up */ #define DML_LOG_FATAL(fmt, ...) DML_LOG_INTERNAL("[DML FATAL] " fmt, ## __VA_ARGS__) + +/* public macros for DML_LOG_LEVEL_ERROR and up */ #if DML_LOG_LEVEL >= DML_LOG_LEVEL_ERROR #define DML_LOG_ERROR(fmt, ...) DML_LOG_INTERNAL("[DML ERROR] "fmt, ## __VA_ARGS__) +#define DML_ASSERT_MSG(condition, fmt, ...) \ + do { \ + if (!(condition)) { \ + DML_LOG_ERROR("ASSERT hit in %s line %d\n", __func__, __LINE__); \ + DML_LOG_ERROR(fmt, ## __VA_ARGS__); \ + DML_ASSERT(condition); \ + } \ + } while (0) #else #define DML_LOG_ERROR(fmt, ...) ((void)0) +#define DML_ASSERT_MSG(condition, fmt, ...) ((void)0) #endif + +/* public macros for DML_LOG_LEVEL_WARN and up */ #if DML_LOG_LEVEL >= DML_LOG_LEVEL_WARN #define DML_LOG_WARN(fmt, ...) DML_LOG_INTERNAL("[DML WARN] "fmt, ## __VA_ARGS__) #else #define DML_LOG_WARN(fmt, ...) ((void)0) #endif + +/* public macros for DML_LOG_LEVEL_INFO and up */ #if DML_LOG_LEVEL >= DML_LOG_LEVEL_INFO #define DML_LOG_INFO(fmt, ...) DML_LOG_INTERNAL("[DML INFO] "fmt, ## __VA_ARGS__) +#define DML_LOG_TOP_IF_ENTER() _LOG_ENTRY(_ELEMENT_TOP_IF) +#define DML_LOG_TOP_IF_EXIT() _LOG_EXIT(_ELEMENT_TOP_IF) #else #define DML_LOG_INFO(fmt, ...) ((void)0) +#define DML_LOG_TOP_IF_ENTER() ((void)0) +#define DML_LOG_TOP_IF_EXIT() ((void)0) #endif + +/* public macros for DML_LOG_LEVEL_DEBUG and up */ #if DML_LOG_LEVEL >= DML_LOG_LEVEL_DEBUG -#define DML_LOG_DEBUG(fmt, ...) DML_LOG_INTERNAL("[DML DEBUG] "fmt, ## __VA_ARGS__) +#define DML_LOG_DEBUG(fmt, ...) DML_LOG_INTERNAL(fmt, ## __VA_ARGS__) +#define DML_LOG_COMP_IF_ENTER() _LOG_ENTRY(_ELEMENT_COMP_IF) +#define DML_LOG_COMP_IF_EXIT() _LOG_EXIT(_ELEMENT_COMP_IF) +#define DML_LOG_FUNC_ENTER() _LOG_ENTRY(_ELEMENT_FUNC) +#define DML_LOG_FUNC_EXIT() _LOG_EXIT(_ELEMENT_FUNC) +#define DML_LOG_DEBUG_BOOL(field) _LOG_SCALAR(field, _BOOL_FORMAT) +#define DML_LOG_DEBUG_UINT(field) _LOG_SCALAR(field, _UINT_FORMAT) +#define DML_LOG_DEBUG_INT(field) _LOG_SCALAR(field, _INT_FORMAT) +#define DML_LOG_DEBUG_DOUBLE(field) _LOG_SCALAR(field, _DOUBLE_FORMAT) +#define DML_LOG_DEBUG_ARRAY_BOOL(field, size) _LOG_ARRAY(field, size, _BOOL_FORMAT) +#define DML_LOG_DEBUG_ARRAY_UINT(field, size) _LOG_ARRAY(field, size, _UINT_FORMAT) +#define DML_LOG_DEBUG_ARRAY_INT(field, size) _LOG_ARRAY(field, size, _INT_FORMAT) +#define DML_LOG_DEBUG_ARRAY_DOUBLE(field, size) _LOG_ARRAY(field, size, _DOUBLE_FORMAT) +#define DML_LOG_DEBUG_2D_ARRAY_BOOL(field, size0, size1) _LOG_2D_ARRAY(field, size0, size1, _BOOL_FORMAT) +#define DML_LOG_DEBUG_2D_ARRAY_UINT(field, size0, size1) _LOG_2D_ARRAY(field, size0, size1, _UINT_FORMAT) +#define DML_LOG_DEBUG_2D_ARRAY_INT(field, size0, size1) _LOG_2D_ARRAY(field, size0, size1, _INT_FORMAT) +#define DML_LOG_DEBUG_2D_ARRAY_DOUBLE(field, size0, size1) _LOG_2D_ARRAY(field, size0, size1, _DOUBLE_FORMAT) +#define DML_LOG_DEBUG_3D_ARRAY_BOOL(field, size0, size1, size2) _LOG_3D_ARRAY(field, size0, size1, size2, _BOOL_FORMAT) +#define DML_LOG_DEBUG_3D_ARRAY_UINT(field, size0, size1, size2) _LOG_3D_ARRAY(field, size0, size1, size2, _UINT_FORMAT) +#define DML_LOG_DEBUG_3D_ARRAY_INT(field, size0, size1, size2) _LOG_3D_ARRAY(field, size0, size1, size2, _INT_FORMAT) +#define DML_LOG_DEBUG_3D_ARRAY_DOUBLE(field, size0, size1, size2) _LOG_3D_ARRAY(field, size0, size1, size2, _DOUBLE_FORMAT) #else #define DML_LOG_DEBUG(fmt, ...) ((void)0) +#define DML_LOG_COMP_IF_ENTER() ((void)0) +#define DML_LOG_COMP_IF_EXIT() ((void)0) +#define DML_LOG_FUNC_ENTER() ((void)0) +#define DML_LOG_FUNC_EXIT() ((void)0) +#define DML_LOG_DEBUG_BOOL(field) ((void)0) +#define DML_LOG_DEBUG_UINT(field) ((void)0) +#define DML_LOG_DEBUG_INT(field) ((void)0) +#define DML_LOG_DEBUG_DOUBLE(field) ((void)0) +#define DML_LOG_DEBUG_ARRAY_BOOL(field, size) ((void)0) +#define DML_LOG_DEBUG_ARRAY_UINT(field, size) ((void)0) +#define DML_LOG_DEBUG_ARRAY_INT(field, size) ((void)0) +#define DML_LOG_DEBUG_ARRAY_DOUBLE(field, size) ((void)0) +#define DML_LOG_DEBUG_2D_ARRAY_BOOL(field, size0, size1) ((void)0) +#define DML_LOG_DEBUG_2D_ARRAY_UINT(field, size0, size1) ((void)0) +#define DML_LOG_DEBUG_2D_ARRAY_INT(field, size0, size1) ((void)0) +#define DML_LOG_DEBUG_2D_ARRAY_DOUBLE(field, size0, size1) ((void)0) +#define DML_LOG_DEBUG_3D_ARRAY_BOOL(field, size0, size1, size2) ((void)0) +#define DML_LOG_DEBUG_3D_ARRAY_UINT(field, size0, size1, size2) ((void)0) +#define DML_LOG_DEBUG_3D_ARRAY_INT(field, size0, size1, size2) ((void)0) +#define DML_LOG_DEBUG_3D_ARRAY_DOUBLE(field, size0, size1, size2) ((void)0) #endif + +/* public macros for DML_LOG_LEVEL_VERBOSE */ #if DML_LOG_LEVEL >= DML_LOG_LEVEL_VERBOSE -#define DML_LOG_VERBOSE(fmt, ...) DML_LOG_INTERNAL("[DML VERBOSE] "fmt, ## __VA_ARGS__) +#define DML_LOG_VERBOSE(fmt, ...) DML_LOG_INTERNAL(fmt, ## __VA_ARGS__) #else #define DML_LOG_VERBOSE(fmt, ...) ((void)0) -#endif +#endif /* #if DML_LOG_LEVEL >= DML_LOG_LEVEL_VERBOSE */ #endif /* __DML2_DEBUG_H__ */ 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/dc/dml2/dml2_mall_phantom.c b/drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c index 6b3b8803e0ae..a56e75cdf712 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml2_mall_phantom.c @@ -868,7 +868,7 @@ bool dml2_svp_remove_all_phantom_pipes(struct dml2_context *ctx, struct dc_state /* Conditions for setting up phantom pipes for SubVP: * 1. Not force disable SubVP - * 2. Full update (i.e. !fast_validate) + * 2. Full update (i.e. DC_VALIDATE_MODE_AND_PROGRAMMING) * 3. Enough pipes are available to support SubVP (TODO: Which pipes will use VACTIVE / VBLANK / SUBVP?) * 4. Display configuration passes validation * 5. (Config doesn't support MCLK in VACTIVE/VBLANK || dc->debug.force_subvp_mclk_switch) diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c index 5de775fd8fce..3b866e876bf4 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml2_translation_helper.c @@ -953,6 +953,7 @@ static void populate_dml_surface_cfg_from_plane_state(enum dml_project_id dml2_p out->SourcePixelFormat[location] = dml_420_10; break; case SURFACE_PIXEL_FORMAT_GRPH_ARGB16161616: + case SURFACE_PIXEL_FORMAT_GRPH_ABGR16161616: case SURFACE_PIXEL_FORMAT_GRPH_ARGB16161616F: case SURFACE_PIXEL_FORMAT_GRPH_ABGR16161616F: out->SourcePixelFormat[location] = dml_444_64; @@ -1188,22 +1189,6 @@ static unsigned int map_plane_to_dml_display_cfg(const struct dml2_context *dml2 return location; } -static void apply_legacy_svp_drr_settings(struct dml2_context *dml2, const struct dc_state *state, struct dml_display_cfg_st *dml_dispcfg) -{ - int i; - - if (state->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) { - ASSERT(state->stream_count == 1); - dml_dispcfg->timing.DRRDisplay[0] = true; - } else if (state->bw_ctx.bw.dcn.legacy_svp_drr_stream_index_valid) { - - for (i = 0; i < dml_dispcfg->num_timings; i++) { - if (dml2->v20.scratch.dml_to_dc_pipe_mapping.disp_cfg_to_stream_id[i] == state->streams[state->bw_ctx.bw.dcn.legacy_svp_drr_stream_index]->stream_id) - dml_dispcfg->timing.DRRDisplay[i] = true; - } - } -} - static void dml2_populate_pipe_to_plane_index_mapping(struct dml2_context *dml2, struct dc_state *state) { unsigned int i; @@ -1436,9 +1421,6 @@ void map_dc_state_into_dml_display_cfg(struct dml2_context *dml2, struct dc_stat } } } - - if (!dml2->config.use_native_pstate_optimization) - apply_legacy_svp_drr_settings(dml2, context, dml_dispcfg); } void dml2_update_pipe_ctx_dchub_regs(struct _vcs_dpi_dml_display_rq_regs_st *rq_regs, diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c index 525b7d04bf84..0318260370ed 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c @@ -24,8 +24,6 @@ * */ -#include <linux/vmalloc.h> - #include "display_mode_core.h" #include "dml2_internal_types.h" #include "dml2_utils.h" @@ -95,12 +93,17 @@ static void map_hw_resources(struct dml2_context *dml2, static unsigned int pack_and_call_dml_mode_support_ex(struct dml2_context *dml2, const struct dml_display_cfg_st *display_cfg, - struct dml_mode_support_info_st *evaluation_info) + struct dml_mode_support_info_st *evaluation_info, + enum dc_validate_mode validate_mode) { struct dml2_wrapper_scratch *s = &dml2->v20.scratch; s->mode_support_params.mode_lib = &dml2->v20.dml_core_ctx; s->mode_support_params.in_display_cfg = display_cfg; + if (validate_mode == DC_VALIDATE_MODE_ONLY) + s->mode_support_params.in_start_state_idx = dml2->v20.dml_core_ctx.states.num_states - 1; + else + s->mode_support_params.in_start_state_idx = 0; s->mode_support_params.out_evaluation_info = evaluation_info; memset(evaluation_info, 0, sizeof(struct dml_mode_support_info_st)); @@ -112,10 +115,8 @@ static unsigned int pack_and_call_dml_mode_support_ex(struct dml2_context *dml2, static bool optimize_configuration(struct dml2_context *dml2, struct dml2_wrapper_optimize_configuration_params *p) { int unused_dpps = p->ip_params->max_num_dpp; - int i, j; - int odms_needed, refresh_rate_hz, dpps_needed, subvp_height, pstate_width_fw_delay_lines, surface_count; - int subvp_timing_to_add, new_timing_index, subvp_surface_to_add, new_surface_index; - float frame_time_sec, max_frame_time_sec; + int i; + int odms_needed; int largest_blend_and_timing = 0; bool optimization_done = false; @@ -130,79 +131,6 @@ static bool optimize_configuration(struct dml2_context *dml2, struct dml2_wrappe if (p->new_display_config != p->cur_display_config) *p->new_display_config = *p->cur_display_config; - // Optimize P-State Support - if (dml2->config.use_native_pstate_optimization) { - if (p->cur_mode_support_info->DRAMClockChangeSupport[0] == dml_dram_clock_change_unsupported) { - // Find a display with < 120Hz refresh rate with maximal refresh rate that's not already subvp - subvp_timing_to_add = -1; - subvp_surface_to_add = -1; - max_frame_time_sec = 0; - surface_count = 0; - for (i = 0; i < (int) p->cur_display_config->num_timings; i++) { - refresh_rate_hz = (int)div_u64((unsigned long long) p->cur_display_config->timing.PixelClock[i] * 1000 * 1000, - (p->cur_display_config->timing.HTotal[i] * p->cur_display_config->timing.VTotal[i])); - if (refresh_rate_hz < 120) { - // Check its upstream surfaces to see if this one could be converted to subvp. - dpps_needed = 0; - for (j = 0; j < (int) p->cur_display_config->num_surfaces; j++) { - if (p->cur_display_config->plane.BlendingAndTiming[j] == i && - p->cur_display_config->plane.UseMALLForPStateChange[j] == dml_use_mall_pstate_change_disable) { - dpps_needed += p->cur_mode_support_info->DPPPerSurface[j]; - subvp_surface_to_add = j; - surface_count++; - } - } - - if (surface_count == 1 && dpps_needed > 0 && dpps_needed <= unused_dpps) { - frame_time_sec = (float)1 / refresh_rate_hz; - if (frame_time_sec > max_frame_time_sec) { - max_frame_time_sec = frame_time_sec; - subvp_timing_to_add = i; - } - } - } - } - if (subvp_timing_to_add >= 0) { - new_timing_index = p->new_display_config->num_timings++; - new_surface_index = p->new_display_config->num_surfaces++; - // Add a phantom pipe reflecting the main pipe's timing - dml2_util_copy_dml_timing(&p->new_display_config->timing, new_timing_index, subvp_timing_to_add); - - pstate_width_fw_delay_lines = (int)(((double)(p->config->svp_pstate.subvp_fw_processing_delay_us + - p->config->svp_pstate.subvp_pstate_allow_width_us) / 1000000) * - (p->new_display_config->timing.PixelClock[subvp_timing_to_add] * 1000 * 1000) / - (double)p->new_display_config->timing.HTotal[subvp_timing_to_add]); - - subvp_height = p->cur_mode_support_info->SubViewportLinesNeededInMALL[subvp_timing_to_add] + pstate_width_fw_delay_lines; - - p->new_display_config->timing.VActive[new_timing_index] = subvp_height; - p->new_display_config->timing.VTotal[new_timing_index] = subvp_height + - p->new_display_config->timing.VTotal[subvp_timing_to_add] - p->new_display_config->timing.VActive[subvp_timing_to_add]; - - p->new_display_config->output.OutputDisabled[new_timing_index] = true; - - p->new_display_config->plane.UseMALLForPStateChange[subvp_surface_to_add] = dml_use_mall_pstate_change_sub_viewport; - - dml2_util_copy_dml_plane(&p->new_display_config->plane, new_surface_index, subvp_surface_to_add); - dml2_util_copy_dml_surface(&p->new_display_config->surface, new_surface_index, subvp_surface_to_add); - - p->new_display_config->plane.ViewportHeight[new_surface_index] = subvp_height; - p->new_display_config->plane.ViewportHeightChroma[new_surface_index] = subvp_height; - p->new_display_config->plane.ViewportStationary[new_surface_index] = false; - - p->new_display_config->plane.UseMALLForStaticScreen[new_surface_index] = dml_use_mall_static_screen_disable; - p->new_display_config->plane.UseMALLForPStateChange[new_surface_index] = dml_use_mall_pstate_change_phantom_pipe; - - p->new_display_config->plane.NumberOfCursors[new_surface_index] = 0; - - p->new_policy->ImmediateFlipRequirement[new_surface_index] = dml_immediate_flip_not_required; - - p->new_display_config->plane.BlendingAndTiming[new_surface_index] = new_timing_index; - - optimization_done = true; - } - } - } // Optimize Clocks if (!optimization_done) { @@ -226,7 +154,8 @@ static bool optimize_configuration(struct dml2_context *dml2, struct dml2_wrappe return optimization_done; } -static int calculate_lowest_supported_state_for_temp_read(struct dml2_context *dml2, struct dc_state *display_state) +static int calculate_lowest_supported_state_for_temp_read(struct dml2_context *dml2, struct dc_state *display_state, + enum dc_validate_mode validate_mode) { struct dml2_calculate_lowest_supported_state_for_temp_read_scratch *s = &dml2->v20.scratch.dml2_calculate_lowest_supported_state_for_temp_read_scratch; struct dml2_wrapper_scratch *s_global = &dml2->v20.scratch; @@ -268,7 +197,8 @@ static int calculate_lowest_supported_state_for_temp_read(struct dml2_context *d dml2->v20.dml_core_ctx.states.state_array[j].dram_clock_change_latency_us = s_global->dummy_pstate_table[i].dummy_pstate_latency_us; } - dml_result = pack_and_call_dml_mode_support_ex(dml2, &s->cur_display_config, &s->evaluation_info); + dml_result = pack_and_call_dml_mode_support_ex(dml2, &s->cur_display_config, &s->evaluation_info, + validate_mode); if (dml_result && s->evaluation_info.DRAMClockChangeSupport[0] == dml_dram_clock_change_vactive) { map_hw_resources(dml2, &s->cur_display_config, &s->evaluation_info); @@ -333,7 +263,8 @@ static bool does_configuration_meet_sw_policies(struct dml2_context *ctx, const } static bool dml_mode_support_wrapper(struct dml2_context *dml2, - struct dc_state *display_state) + struct dc_state *display_state, + enum dc_validate_mode validate_mode) { struct dml2_wrapper_scratch *s = &dml2->v20.scratch; unsigned int result = 0, i; @@ -369,7 +300,8 @@ static bool dml_mode_support_wrapper(struct dml2_context *dml2, result = pack_and_call_dml_mode_support_ex(dml2, &s->cur_display_config, - &s->mode_support_info); + &s->mode_support_info, + validate_mode); if (result) result = does_configuration_meet_sw_policies(dml2, &s->cur_display_config, &s->mode_support_info); @@ -390,7 +322,8 @@ static bool dml_mode_support_wrapper(struct dml2_context *dml2, dml2->v20.dml_core_ctx.policy = s->new_policy; optimized_result = pack_and_call_dml_mode_support_ex(dml2, &s->new_display_config, - &s->mode_support_info); + &s->mode_support_info, + validate_mode); if (optimized_result) optimized_result = does_configuration_meet_sw_policies(dml2, &s->new_display_config, &s->mode_support_info); @@ -409,7 +342,8 @@ static bool dml_mode_support_wrapper(struct dml2_context *dml2, if (!optimized_result) { result = pack_and_call_dml_mode_support_ex(dml2, &s->cur_display_config, - &s->mode_support_info); + &s->mode_support_info, + validate_mode); } } @@ -419,118 +353,7 @@ static bool dml_mode_support_wrapper(struct dml2_context *dml2, return result; } -static int find_drr_eligible_stream(struct dc_state *display_state) -{ - int i; - - for (i = 0; i < display_state->stream_count; i++) { - if (dc_state_get_stream_subvp_type(display_state, display_state->streams[i]) == SUBVP_NONE - && display_state->streams[i]->ignore_msa_timing_param) { - // Use ignore_msa_timing_param flag to identify as DRR - return i; - } - } - - return -1; -} - -static bool optimize_pstate_with_svp_and_drr(struct dml2_context *dml2, struct dc_state *display_state) -{ - struct dml2_wrapper_scratch *s = &dml2->v20.scratch; - bool pstate_optimization_done = false; - bool pstate_optimization_success = false; - bool result = false; - int drr_display_index = 0, non_svp_streams = 0; - bool force_svp = dml2->config.svp_pstate.force_enable_subvp; - - display_state->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false; - display_state->bw_ctx.bw.dcn.legacy_svp_drr_stream_index_valid = false; - - result = dml_mode_support_wrapper(dml2, display_state); - - if (!result) { - pstate_optimization_done = true; - } else if (s->mode_support_info.DRAMClockChangeSupport[0] != dml_dram_clock_change_unsupported && !force_svp) { - pstate_optimization_success = true; - pstate_optimization_done = true; - } - - if (display_state->stream_count == 1 && dml2->config.callbacks.can_support_mclk_switch_using_fw_based_vblank_stretch(dml2->config.callbacks.dc, display_state)) { - display_state->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = true; - - result = dml_mode_support_wrapper(dml2, display_state); - } else { - non_svp_streams = display_state->stream_count; - - while (!pstate_optimization_done) { - result = dml_mode_programming(&dml2->v20.dml_core_ctx, s->mode_support_params.out_lowest_state_idx, &s->cur_display_config, true); - - // Always try adding SVP first - if (result) - result = dml2_svp_add_phantom_pipe_to_dc_state(dml2, display_state, &s->mode_support_info); - else - pstate_optimization_done = true; - - - if (result) { - result = dml_mode_support_wrapper(dml2, display_state); - } else { - pstate_optimization_done = true; - } - - if (result) { - non_svp_streams--; - - if (s->mode_support_info.DRAMClockChangeSupport[0] != dml_dram_clock_change_unsupported) { - if (dml2_svp_validate_static_schedulability(dml2, display_state, s->mode_support_info.DRAMClockChangeSupport[0])) { - pstate_optimization_success = true; - pstate_optimization_done = true; - } else { - pstate_optimization_success = false; - pstate_optimization_done = false; - } - } else { - drr_display_index = find_drr_eligible_stream(display_state); - - // If there is only 1 remaining non SubVP pipe that is DRR, check static - // schedulability for SubVP + DRR. - if (non_svp_streams == 1 && drr_display_index >= 0) { - if (dml2_svp_drr_schedulable(dml2, display_state, &display_state->streams[drr_display_index]->timing)) { - display_state->bw_ctx.bw.dcn.legacy_svp_drr_stream_index_valid = true; - display_state->bw_ctx.bw.dcn.legacy_svp_drr_stream_index = drr_display_index; - result = dml_mode_support_wrapper(dml2, display_state); - } - - if (result && s->mode_support_info.DRAMClockChangeSupport[0] != dml_dram_clock_change_unsupported) { - pstate_optimization_success = true; - pstate_optimization_done = true; - } else { - pstate_optimization_success = false; - pstate_optimization_done = false; - } - } - - if (pstate_optimization_success) { - pstate_optimization_done = true; - } else { - pstate_optimization_done = false; - } - } - } - } - } - - if (!pstate_optimization_success) { - dml2_svp_remove_all_phantom_pipes(dml2, display_state); - display_state->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false; - display_state->bw_ctx.bw.dcn.legacy_svp_drr_stream_index_valid = false; - result = dml_mode_support_wrapper(dml2, display_state); - } - - return result; -} - -static bool call_dml_mode_support_and_programming(struct dc_state *context) +static bool call_dml_mode_support_and_programming(struct dc_state *context, enum dc_validate_mode validate_mode) { unsigned int result = 0; unsigned int min_state = 0; @@ -544,16 +367,13 @@ static bool call_dml_mode_support_and_programming(struct dc_state *context) struct dml2_wrapper_scratch *s = &dml2->v20.scratch; if (!context->streams[0]->sink->link->dc->caps.is_apu) { - min_state_for_g6_temp_read = calculate_lowest_supported_state_for_temp_read(dml2, context); + min_state_for_g6_temp_read = calculate_lowest_supported_state_for_temp_read(dml2, context, + validate_mode); ASSERT(min_state_for_g6_temp_read >= 0); } - if (!dml2->config.use_native_pstate_optimization) { - result = optimize_pstate_with_svp_and_drr(dml2, context); - } else { - result = dml_mode_support_wrapper(dml2, context); - } + result = dml_mode_support_wrapper(dml2, context, validate_mode); /* Upon trying to sett certain frequencies in FRL, min_state_for_g6_temp_read is reported as -1. This leads to an invalid value of min_state causing crashes later on. * Use the default logic for min_state only when min_state_for_g6_temp_read is a valid value. In other cases, use the value calculated by the DML directly. @@ -575,7 +395,8 @@ static bool call_dml_mode_support_and_programming(struct dc_state *context) return result; } -static bool dml2_validate_and_build_resource(const struct dc *in_dc, struct dc_state *context) +static bool dml2_validate_and_build_resource(const struct dc *in_dc, struct dc_state *context, + enum dc_validate_mode validate_mode) { struct dml2_context *dml2 = context->bw_ctx.dml2; struct dml2_wrapper_scratch *s = &dml2->v20.scratch; @@ -611,7 +432,7 @@ static bool dml2_validate_and_build_resource(const struct dc *in_dc, struct dc_s copy_dummy_pstate_table(s->dummy_pstate_table, in_dc->clk_mgr->bw_params->dummy_pstate_table, 4); - result = call_dml_mode_support_and_programming(context); + result = call_dml_mode_support_and_programming(context, validate_mode); /* Call map dc pipes to map the pipes based on the DML output. For correctly determining if recalculation * is required or not, the resource context needs to correctly reflect the number of active pipes. We would * only know the correct number if active pipes after dml2_map_dc_pipes is called. @@ -628,7 +449,7 @@ static bool dml2_validate_and_build_resource(const struct dc *in_dc, struct dc_s need_recalculation = dml2_verify_det_buffer_configuration(dml2, context, &dml2->det_helper_scratch); if (need_recalculation) { /* Engage the DML again if recalculation is required. */ - call_dml_mode_support_and_programming(context); + call_dml_mode_support_and_programming(context, validate_mode); if (!dml2->config.skip_hw_state_mapping) { dml2_map_dc_pipes(dml2, context, &s->cur_display_config, &s->dml_to_dc_pipe_mapping, in_dc->current_state); } @@ -684,7 +505,7 @@ static bool dml2_validate_and_build_resource(const struct dc *in_dc, struct dc_s return result; } -static bool dml2_validate_only(struct dc_state *context) +static bool dml2_validate_only(struct dc_state *context, enum dc_validate_mode validate_mode) { struct dml2_context *dml2; unsigned int result = 0; @@ -708,7 +529,8 @@ static bool dml2_validate_only(struct dc_state *context) result = pack_and_call_dml_mode_support_ex(dml2, &dml2->v20.scratch.cur_display_config, - &dml2->v20.scratch.mode_support_info); + &dml2->v20.scratch.mode_support_info, + validate_mode); if (result) result = does_configuration_meet_sw_policies(dml2, &dml2->v20.scratch.cur_display_config, &dml2->v20.scratch.mode_support_info); @@ -723,7 +545,8 @@ static void dml2_apply_debug_options(const struct dc *dc, struct dml2_context *d } } -bool dml2_validate(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml2, bool fast_validate) +bool dml2_validate(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml2, + enum dc_validate_mode validate_mode) { bool out = false; @@ -733,17 +556,17 @@ bool dml2_validate(const struct dc *in_dc, struct dc_state *context, struct dml2 /* DML2.1 validation path */ if (dml2->architecture == dml2_architecture_21) { - out = dml21_validate(in_dc, context, dml2, fast_validate); + out = dml21_validate(in_dc, context, dml2, validate_mode); return out; } DC_FP_START(); - /* Use dml_validate_only for fast_validate path */ - if (fast_validate) - out = dml2_validate_only(context); + /* Use dml_validate_only for DC_VALIDATE_MODE_ONLY and DC_VALIDATE_MODE_AND_STATE_INDEX path */ + if (validate_mode != DC_VALIDATE_MODE_AND_PROGRAMMING) + out = dml2_validate_only(context, validate_mode); else - out = dml2_validate_and_build_resource(in_dc, context); + out = dml2_validate_and_build_resource(in_dc, context, validate_mode); DC_FP_END(); @@ -757,8 +580,8 @@ static inline struct dml2_context *dml2_allocate_memory(void) static void dml2_init(const struct dc *in_dc, const struct dml2_configuration_options *config, struct dml2_context **dml2) { - if ((in_dc->debug.using_dml21) && (in_dc->ctx->dce_version == DCN_VERSION_4_01)) { - dml21_reinit(in_dc, dml2, config); + if ((in_dc->debug.using_dml21) && (in_dc->ctx->dce_version >= DCN_VERSION_4_01)) { + dml21_reinit(in_dc, *dml2, config); return; } @@ -803,9 +626,7 @@ static void dml2_init(const struct dc *in_dc, const struct dml2_configuration_op bool dml2_create(const struct dc *in_dc, const struct dml2_configuration_options *config, struct dml2_context **dml2) { // TODO : Temporarily add DCN_VERSION_3_2 for N-1 validation. Remove DCN_VERSION_3_2 after N-1 validation phase is complete. - if ((in_dc->debug.using_dml21) - && (in_dc->ctx->dce_version == DCN_VERSION_4_01 - )) + if ((in_dc->debug.using_dml21) && (in_dc->ctx->dce_version >= DCN_VERSION_4_01)) return dml21_create(in_dc, dml2, config); // Allocate Mode Lib Ctx @@ -874,8 +695,8 @@ void dml2_reinit(const struct dc *in_dc, const struct dml2_configuration_options *config, struct dml2_context **dml2) { - if ((in_dc->debug.using_dml21) && (in_dc->ctx->dce_version == DCN_VERSION_4_01)) { - dml21_reinit(in_dc, dml2, config); + if ((in_dc->debug.using_dml21) && (in_dc->ctx->dce_version >= DCN_VERSION_4_01)) { + dml21_reinit(in_dc, *dml2, config); return; } diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.h b/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.h index 5100f269368e..c384e141cebc 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.h +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.h @@ -240,7 +240,7 @@ struct dml2_configuration_options { bool use_clock_dc_limits; bool gpuvm_enable; bool force_tdlut_enable; - struct dml2_soc_bb *bb_from_dmub; + void *bb_from_dmub; }; /* @@ -272,7 +272,7 @@ void dml2_reinit(const struct dc *in_dc, * dml2_validate - Determines if a display configuration is supported or not. * @in_dc: dc. * @context: dc_state to be validated. - * @fast_validate: Fast validate will not populate context.res_ctx. + * @validate_mode: DC_VALIDATE_MODE_ONLY and DC_VALIDATE_MODE_AND_STATE_INDEX will not populate context.res_ctx. * * DML1.0 compatible interface for validation. * @@ -295,7 +295,7 @@ void dml2_reinit(const struct dc *in_dc, bool dml2_validate(const struct dc *in_dc, struct dc_state *context, struct dml2_context *dml2, - bool fast_validate); + enum dc_validate_mode validate_mode); /* * dml2_extract_dram_and_fclk_change_support - Extracts the FCLK and UCLK change support info. |