drm/amd/display: update calculated bounding box logic for NV
authorJun Lei <Jun.Lei@amd.com>
Thu, 9 May 2019 19:32:27 +0000 (15:32 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Sat, 22 Jun 2019 14:34:07 +0000 (09:34 -0500)
[why]
Current calculation of bounding box will cause DML to increase voltage
state due to DPP or DISPCLK, this is unnecessary since from DML perspective
we can max DPP/DISP can be supported at DPM0.  This is because
increasing voltage for DPP/DISP is done separately via actual minimum values
of DISP and DPP CLK

[how]
For each calculated state, DPP, DISP, PHY, and DSC clk should always be set to
maximum.  FCLK, SOCCLK, and DCFCLK should be based of UCLK.

Signed-off-by: Jun Lei <Jun.Lei@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c

index d0c279ab0af14ed51b404ca87c118055668830a1..c4fa0b9e7138c059ed0e0cef32ffdd94e9c1f979 100644 (file)
@@ -2402,54 +2402,47 @@ static void update_bounding_box(struct dc *dc, struct _vcs_dpi_soc_bounding_box_
                struct pp_smu_nv_clock_table *max_clocks, unsigned int *uclk_states, unsigned int num_states)
 {
        struct _vcs_dpi_voltage_scaling_st calculated_states[MAX_CLOCK_LIMIT_STATES] = {0};
-       int i, j;
+       int i;
        int num_calculated_states = 0;
+       int min_dcfclk = 0;
 
        if (num_states == 0)
                return;
 
+       if (dc->bb_overrides.min_dcfclk_mhz > 0)
+               min_dcfclk = dc->bb_overrides.min_dcfclk_mhz;
+
        for (i = 0; i < num_states; i++) {
-               // Find lowest pre-silicon DPM that has equal or higher uCLK
-               for (j = 0; j < bb->num_states; j++) {
-                       if (bb->clock_limits[j].dram_speed_mts * 1000 / 16 >= uclk_states[i])
-                               break;
-               }
+               int min_fclk_required_by_uclk;
+               calculated_states[i].state = i;
+               calculated_states[i].dram_speed_mts = uclk_states[i] * 16 / 1000;
 
-               // If for some reason the available uCLK is higher than all pre-silicon'
-               // DPM targets, then we just use the highest one
-               if (j >= bb->num_states)
-                       j = bb->num_states;
+               min_fclk_required_by_uclk = ((unsigned long long)uclk_states[i]) * 1008 / 1000000;
 
-               // Copy that state
-               memcpy(&calculated_states[num_calculated_states], &bb->clock_limits[j],
-                               sizeof(calculated_states[num_calculated_states]));
+               calculated_states[i].fabricclk_mhz = (min_fclk_required_by_uclk < min_dcfclk) ?
+                               min_dcfclk : min_fclk_required_by_uclk;
 
-               // Cap uClk to actual
-               calculated_states[num_calculated_states].dram_speed_mts = uclk_states[i] * 16 / 1000;
-               // Phy clock can be set to max for all states, since there's nothing to optimize
-               // for spreadsheet and we request voltage for phy clock by frequency anyway
-               calculated_states[num_calculated_states].phyclk_mhz = max_clocks->phyClockInKhz / 1000;
+               calculated_states[i].socclk_mhz = (calculated_states[i].fabricclk_mhz > max_clocks->socClockInKhz / 1000) ?
+                               max_clocks->socClockInKhz / 1000 : calculated_states[i].fabricclk_mhz;
 
-               calculated_states[num_calculated_states].state = num_calculated_states;
+               calculated_states[i].dcfclk_mhz = (calculated_states[i].fabricclk_mhz > max_clocks->dcfClockInKhz / 1000) ?
+                               max_clocks->dcfClockInKhz / 1000 : calculated_states[i].fabricclk_mhz;
 
-               num_calculated_states++;
-       }
+               calculated_states[i].dispclk_mhz = max_clocks->displayClockInKhz / 1000;
+               calculated_states[i].dppclk_mhz = max_clocks->displayClockInKhz / 1000;
+               calculated_states[i].dscclk_mhz = max_clocks->displayClockInKhz / (1000 * 3);
 
-       if (max_clocks->dcfClockInKhz > 0)
-               calculated_states[num_calculated_states - 1].dcfclk_mhz = max_clocks->dcfClockInKhz / 1000;
+               calculated_states[i].phyclk_mhz = max_clocks->phyClockInKhz / 1000;
 
-       if (max_clocks->displayClockInKhz > 0) {
-               calculated_states[num_calculated_states - 1].dispclk_mhz = max_clocks->displayClockInKhz / 1000;
-               calculated_states[num_calculated_states - 1].dppclk_mhz = max_clocks->displayClockInKhz / 1000;
-               // DSC always runs at 1/3 of disp clock
-               calculated_states[num_calculated_states - 1].dscclk_mhz = max_clocks->displayClockInKhz / (1000 * 3);
+               num_calculated_states++;
        }
 
-       if (max_clocks->socClockInKhz > 0)
-               calculated_states[num_calculated_states - 1].socclk_mhz = max_clocks->socClockInKhz / 1000;
-
        memcpy(bb->clock_limits, calculated_states, sizeof(bb->clock_limits));
        bb->num_states = num_calculated_states;
+
+       // Duplicate the last state, DML always an extra state identical to max state to work
+       memcpy(&bb->clock_limits[num_calculated_states], &bb->clock_limits[num_calculated_states - 1], sizeof(struct _vcs_dpi_voltage_scaling_st));
+       bb->clock_limits[num_calculated_states].state = bb->num_states;
 }
 
 static void patch_bounding_box(struct dc *dc, struct _vcs_dpi_soc_bounding_box_st *bb)