summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c')
-rw-r--r--drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c157
1 files changed, 120 insertions, 37 deletions
diff --git a/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c b/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c
index 2abe3967f7fb..f94abd124021 100644
--- a/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c
+++ b/drivers/gpu/drm/amd/display/dc/dml/dcn32/dcn32_fpu.c
@@ -256,16 +256,24 @@ int dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(struct dc *dc,
int vlevel)
{
const int max_latency_table_entries = 4;
- const struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
+ struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
int dummy_latency_index = 0;
+ enum clock_change_support temp_clock_change_support = vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
dc_assert_fp_enabled();
while (dummy_latency_index < max_latency_table_entries) {
+ if (temp_clock_change_support != dm_dram_clock_change_unsupported)
+ vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = temp_clock_change_support;
context->bw_ctx.dml.soc.dram_clock_change_latency_us =
dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
+ /* for subvp + DRR case, if subvp pipes are still present we support pstate */
+ if (vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] == dm_dram_clock_change_unsupported &&
+ dcn32_subvp_in_use(dc, context))
+ vba->DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = temp_clock_change_support;
+
if (vlevel < context->bw_ctx.dml.vba.soc.num_states &&
vba->DRAMClockChangeSupport[vlevel][vba->maxMpcComb] != dm_dram_clock_change_unsupported)
break;
@@ -531,9 +539,12 @@ void dcn32_set_phantom_stream_timing(struct dc *dc,
unsigned int i, pipe_idx;
struct pipe_ctx *pipe;
uint32_t phantom_vactive, phantom_bp, pstate_width_fw_delay_lines;
+ unsigned int num_dpp;
unsigned int vlevel = context->bw_ctx.dml.vba.VoltageLevel;
unsigned int dcfclk = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
unsigned int socclk = context->bw_ctx.dml.vba.SOCCLKPerState[vlevel];
+ struct vba_vars_st *vba = &context->bw_ctx.dml.vba;
+ struct dc_stream_state *main_stream = ref_pipe->stream;
dc_assert_fp_enabled();
@@ -569,13 +580,26 @@ void dcn32_set_phantom_stream_timing(struct dc *dc,
phantom_vactive = get_subviewport_lines_needed_in_mall(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx) +
pstate_width_fw_delay_lines + dc->caps.subvp_swath_height_margin_lines;
+ // W/A for DCC corruption with certain high resolution timings.
+ // Determing if pipesplit is used. If so, add meta_row_height to the phantom vactive.
+ num_dpp = vba->NoOfDPP[vba->VoltageLevel][vba->maxMpcComb][vba->pipe_plane[pipe_idx]];
+ phantom_vactive += num_dpp > 1 ? vba->meta_row_height[vba->pipe_plane[pipe_idx]] : 0;
+
+ /* dc->debug.subvp_extra_lines 0 by default*/
+ phantom_vactive += dc->debug.subvp_extra_lines;
+
// For backporch of phantom pipe, use vstartup of the main pipe
phantom_bp = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt, pipe_idx);
phantom_stream->dst.y = 0;
phantom_stream->dst.height = phantom_vactive;
+ /* When scaling, DML provides the end to end required number of lines for MALL.
+ * dst.height is always correct for this case, but src.height is not which causes a
+ * delta between main and phantom pipe scaling outputs. Need to adjust src.height on
+ * phantom for this case.
+ */
phantom_stream->src.y = 0;
- phantom_stream->src.height = phantom_vactive;
+ phantom_stream->src.height = (double)phantom_vactive * (double)main_stream->src.height / (double)main_stream->dst.height;
phantom_stream->timing.v_addressable = phantom_vactive;
phantom_stream->timing.v_front_porch = 1;
@@ -1128,7 +1152,7 @@ static void dcn32_full_validate_bw_helper(struct dc *dc,
context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final ==
dm_prefetch_support_uclk_fclk_and_stutter) {
context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
- dm_prefetch_support_stutter;
+ dm_prefetch_support_fclk_and_stutter;
/* There are params (such as FabricClock) that need to be recalculated
* after validation fails (otherwise it will be 0). Calculation for
* phantom vactive requires call into DML, so we must ensure all the
@@ -1179,7 +1203,7 @@ static void dcn32_full_validate_bw_helper(struct dc *dc,
// If SubVP pipe config is unsupported (or cannot be used for UCLK switching)
// remove phantom pipes and repopulate dml pipes
if (!found_supported_config) {
- dc->res_pool->funcs->remove_phantom_pipes(dc, context);
+ dc->res_pool->funcs->remove_phantom_pipes(dc, context, false);
vba->DRAMClockChangeSupport[*vlevel][vba->maxMpcComb] = dm_dram_clock_change_unsupported;
*pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, false);
@@ -1191,9 +1215,7 @@ static void dcn32_full_validate_bw_helper(struct dc *dc,
}
} else {
// Most populate phantom DLG params before programming hardware / timing for phantom pipe
- DC_FP_START();
dcn32_helper_populate_phantom_dlg_params(dc, context, pipes, *pipe_cnt);
- DC_FP_END();
/* Call validate_apply_pipe_split flags after calling DML getters for
* phantom dlg params, or some of the VBA params indicating pipe split
@@ -1230,7 +1252,7 @@ static void dcn32_calculate_dlg_params(struct dc *dc, struct dc_state *context,
display_e2e_pipe_params_st *pipes,
int pipe_cnt, int vlevel)
{
- int i, pipe_idx;
+ int i, pipe_idx, active_hubp_count = 0;
bool usr_retraining_support = false;
bool unbounded_req_enabled = false;
@@ -1275,6 +1297,8 @@ static void dcn32_calculate_dlg_params(struct dc *dc, struct dc_state *context,
for (i = 0, pipe_idx = 0; i < dc->res_pool->pipe_count; i++) {
if (!context->res_ctx.pipe_ctx[i].stream)
continue;
+ if (context->res_ctx.pipe_ctx[i].plane_state)
+ active_hubp_count++;
pipes[pipe_idx].pipe.dest.vstartup_start = get_vstartup(&context->bw_ctx.dml, pipes, pipe_cnt,
pipe_idx);
pipes[pipe_idx].pipe.dest.vupdate_offset = get_vupdate_offset(&context->bw_ctx.dml, pipes, pipe_cnt,
@@ -1296,10 +1320,23 @@ static void dcn32_calculate_dlg_params(struct dc *dc, struct dc_state *context,
if (context->bw_ctx.bw.dcn.clk.dppclk_khz < pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000)
context->bw_ctx.bw.dcn.clk.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
- context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
+ if (context->res_ctx.pipe_ctx[i].plane_state)
+ context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = pipes[pipe_idx].clks_cfg.dppclk_mhz * 1000;
+ else
+ context->res_ctx.pipe_ctx[i].plane_res.bw.dppclk_khz = 0;
context->res_ctx.pipe_ctx[i].pipe_dlg_param = pipes[pipe_idx].pipe.dest;
pipe_idx++;
}
+ /* If DCN isn't making memory requests we can allow pstate change and lower clocks */
+ if (!active_hubp_count) {
+ context->bw_ctx.bw.dcn.clk.socclk_khz = 0;
+ context->bw_ctx.bw.dcn.clk.dppclk_khz = 0;
+ context->bw_ctx.bw.dcn.clk.dcfclk_khz = 0;
+ context->bw_ctx.bw.dcn.clk.dcfclk_deep_sleep_khz = 0;
+ context->bw_ctx.bw.dcn.clk.dramclk_khz = 0;
+ context->bw_ctx.bw.dcn.clk.fclk_khz = 0;
+ context->bw_ctx.bw.dcn.clk.p_state_change_support = true;
+ }
/*save a original dppclock copy*/
context->bw_ctx.bw.dcn.clk.bw_dppclk_khz = context->bw_ctx.bw.dcn.clk.dppclk_khz;
context->bw_ctx.bw.dcn.clk.bw_dispclk_khz = context->bw_ctx.bw.dcn.clk.dispclk_khz;
@@ -1481,7 +1518,7 @@ bool dcn32_internal_validate_bw(struct dc *dc,
return false;
// For each full update, remove all existing phantom pipes first
- dc->res_pool->funcs->remove_phantom_pipes(dc, context);
+ dc->res_pool->funcs->remove_phantom_pipes(dc, context, fast_validate);
dc->res_pool->funcs->update_soc_for_wm_a(dc, context);
@@ -1494,11 +1531,8 @@ bool dcn32_internal_validate_bw(struct dc *dc,
dml_log_pipe_params(&context->bw_ctx.dml, pipes, pipe_cnt);
- if (!fast_validate) {
- DC_FP_START();
+ if (!fast_validate)
dcn32_full_validate_bw_helper(dc, context, pipes, &vlevel, split, merge, &pipe_cnt);
- DC_FP_END();
- }
if (fast_validate ||
(dc->debug.dml_disallow_alternate_prefetch_modes &&
@@ -1734,6 +1768,10 @@ bool dcn32_internal_validate_bw(struct dc *dc,
}
if (repopulate_pipes) {
+ int flag_max_mpc_comb = vba->maxMpcComb;
+ int flag_vlevel = vlevel;
+ int i;
+
pipe_cnt = dc->res_pool->funcs->populate_dml_pipes(dc, context, pipes, fast_validate);
/* repopulate_pipes = 1 means the pipes were either split or merged. In this case
@@ -1741,10 +1779,28 @@ bool dcn32_internal_validate_bw(struct dc *dc,
* ensure all the params are calculated correctly. We do not need to run the
* pipe split check again after this call (pipes are already split / merged).
* */
- if (!fast_validate) {
- context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
- dm_prefetch_support_uclk_fclk_and_stutter_if_possible;
- vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
+ context->bw_ctx.dml.soc.allow_for_pstate_or_stutter_in_vblank_final =
+ dm_prefetch_support_uclk_fclk_and_stutter_if_possible;
+ vlevel = dml_get_voltage_level(&context->bw_ctx.dml, pipes, pipe_cnt);
+ if (vlevel == context->bw_ctx.dml.soc.num_states) {
+ /* failed after DET size changes */
+ goto validate_fail;
+ } else if (flag_max_mpc_comb == 0 &&
+ flag_max_mpc_comb != context->bw_ctx.dml.vba.maxMpcComb) {
+ /* check the context constructed with pipe split flags is still valid*/
+ bool flags_valid = false;
+ for (i = flag_vlevel; i < context->bw_ctx.dml.soc.num_states; i++) {
+ if (vba->ModeSupport[i][flag_max_mpc_comb]) {
+ vba->maxMpcComb = flag_max_mpc_comb;
+ vba->VoltageLevel = i;
+ vlevel = i;
+ flags_valid = true;
+ }
+ }
+
+ /* this should never happen */
+ if (!flags_valid)
+ goto validate_fail;
}
}
*vlevel_out = vlevel;
@@ -1775,14 +1831,38 @@ void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context,
unsigned int dummy_latency_index = 0;
int maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
unsigned int min_dram_speed_mts = context->bw_ctx.dml.vba.DRAMSpeed;
+ bool subvp_in_use = dcn32_subvp_in_use(dc, context);
unsigned int min_dram_speed_mts_margin;
+ bool need_fclk_lat_as_dummy = false;
+ bool is_subvp_p_drr = true;
dc_assert_fp_enabled();
- // Override DRAMClockChangeSupport for SubVP + DRR case where the DRR cannot switch without stretching it's VBLANK
- if (!pstate_en && dcn32_subvp_in_use(dc, context)) {
- context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp;
- pstate_en = true;
+ /* need to find dummy latency index for subvp */
+ if (subvp_in_use) {
+ /* Override DRAMClockChangeSupport for SubVP + DRR case where the DRR cannot switch without stretching it's VBLANK */
+ if (!pstate_en) {
+ context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp;
+ pstate_en = true;
+ is_subvp_p_drr = true;
+ }
+ dummy_latency_index = dcn32_find_dummy_latency_index_for_fw_based_mclk_switch(dc,
+ context, pipes, pipe_cnt, vlevel);
+
+ /* For DCN32/321 need to validate with fclk pstate change latency equal to dummy so prefetch is
+ * scheduled correctly to account for dummy pstate.
+ */
+ if (context->bw_ctx.dml.soc.fclk_change_latency_us < dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us) {
+ need_fclk_lat_as_dummy = true;
+ context->bw_ctx.dml.soc.fclk_change_latency_us =
+ dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
+ }
+ context->bw_ctx.dml.soc.dram_clock_change_latency_us =
+ dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.pstate_latency_us;
+ dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
+ if (is_subvp_p_drr) {
+ context->bw_ctx.dml.vba.DRAMClockChangeSupport[vlevel][context->bw_ctx.dml.vba.maxMpcComb] = dm_dram_clock_change_vblank_w_mall_sub_vp;
+ }
}
context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching = false;
@@ -1806,9 +1886,11 @@ void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context,
/* For DCN32/321 need to validate with fclk pstate change latency equal to dummy so
* prefetch is scheduled correctly to account for dummy pstate.
*/
- if (dummy_latency_index == 0)
+ if (context->bw_ctx.dml.soc.fclk_change_latency_us < dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us) {
+ need_fclk_lat_as_dummy = true;
context->bw_ctx.dml.soc.fclk_change_latency_us =
dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
+ }
dcn32_internal_validate_bw(dc, context, pipes, &pipe_cnt, &vlevel, false);
maxMpcComb = context->bw_ctx.dml.vba.maxMpcComb;
dcfclk_from_fw_based_mclk_switching = context->bw_ctx.dml.vba.DCFCLKState[vlevel][context->bw_ctx.dml.vba.maxMpcComb];
@@ -1916,7 +1998,7 @@ void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context,
dc->clk_mgr->bw_params->clk_table.entries[min_dram_speed_mts_offset].memclk_mhz * 16;
}
- if (!context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
+ if (!context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching && !subvp_in_use) {
/* find largest table entry that is lower than dram speed,
* but lower than DPM0 still uses DPM0
*/
@@ -1996,7 +2078,8 @@ void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context,
context->perf_params.stutter_period_us = context->bw_ctx.dml.vba.StutterPeriod;
- if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching && dummy_latency_index == 0)
+ /* for proper prefetch calculations, if dummy lat > fclk lat, use fclk lat = dummy lat */
+ if (need_fclk_lat_as_dummy)
context->bw_ctx.dml.soc.fclk_change_latency_us =
dc->clk_mgr->bw_params->dummy_pstate_table[dummy_latency_index].dummy_pstate_latency_us;
@@ -2009,10 +2092,12 @@ void dcn32_calculate_wm_and_dlg_fpu(struct dc *dc, struct dc_state *context,
if (context->bw_ctx.bw.dcn.clk.fw_based_mclk_switching) {
dcn30_setup_mclk_switch_using_fw_based_vblank_stretch(dc, context);
- if (dummy_latency_index == 0)
- context->bw_ctx.dml.soc.fclk_change_latency_us =
- dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us;
}
+
+ /* revert fclk lat changes if required */
+ if (need_fclk_lat_as_dummy)
+ context->bw_ctx.dml.soc.fclk_change_latency_us =
+ dc->clk_mgr->bw_params->wm_table.nv_entries[WM_A].dml_input.fclk_change_latency_us;
}
static void dcn32_get_optimal_dcfclk_fclk_for_uclk(unsigned int uclk_mts,
@@ -2159,9 +2244,7 @@ static int build_synthetic_soc_states(struct clk_bw_params *bw_params,
entry.fabricclk_mhz = 0;
entry.dram_speed_mts = 0;
- DC_FP_START();
insert_entry_into_table_sorted(table, num_entries, &entry);
- DC_FP_END();
}
// Insert the max DCFCLK
@@ -2169,9 +2252,7 @@ static int build_synthetic_soc_states(struct clk_bw_params *bw_params,
entry.fabricclk_mhz = 0;
entry.dram_speed_mts = 0;
- DC_FP_START();
insert_entry_into_table_sorted(table, num_entries, &entry);
- DC_FP_END();
// Insert the UCLK DPMS
for (i = 0; i < num_uclk_dpms; i++) {
@@ -2179,9 +2260,7 @@ static int build_synthetic_soc_states(struct clk_bw_params *bw_params,
entry.fabricclk_mhz = 0;
entry.dram_speed_mts = bw_params->clk_table.entries[i].memclk_mhz * 16;
- DC_FP_START();
insert_entry_into_table_sorted(table, num_entries, &entry);
- DC_FP_END();
}
// If FCLK is coarse grained, insert individual DPMs.
@@ -2191,9 +2270,7 @@ static int build_synthetic_soc_states(struct clk_bw_params *bw_params,
entry.fabricclk_mhz = bw_params->clk_table.entries[i].fclk_mhz;
entry.dram_speed_mts = 0;
- DC_FP_START();
insert_entry_into_table_sorted(table, num_entries, &entry);
- DC_FP_END();
}
}
// If FCLK fine grained, only insert max
@@ -2202,9 +2279,7 @@ static int build_synthetic_soc_states(struct clk_bw_params *bw_params,
entry.fabricclk_mhz = max_fclk_mhz;
entry.dram_speed_mts = 0;
- DC_FP_START();
insert_entry_into_table_sorted(table, num_entries, &entry);
- DC_FP_END();
}
// At this point, the table contains all "points of interest" based on
@@ -2539,3 +2614,11 @@ void dcn32_update_bw_bounding_box_fpu(struct dc *dc, struct clk_bw_params *bw_pa
}
}
+void dcn32_zero_pipe_dcc_fraction(display_e2e_pipe_params_st *pipes,
+ int pipe_cnt)
+{
+ dc_assert_fp_enabled();
+
+ pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_luma = 0;
+ pipes[pipe_cnt].pipe.src.dcc_fraction_of_zs_req_chroma = 0;
+}
generated by cgit (git 2.34.1) at 2024-05-28 17:12:34 +0000