OmniNX/Horizon-OC
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #7 from Lightos1/master

Browse Source 
(hopefully) proper Auto_ADJ implementation, timing fixes, slight configuration tweaks, typos, formating.
This commit is contained in:
Souldbminer
2025-10-01 18:18:42 -04:00
committed by GitHub
parent acf45ffa5e f8c738e5e8
commit 3dfb95f542
6 changed files with 774 additions and 1016 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
README.md
View File

@@ -53,5 +53,5 @@ Run build.bat or cd into folder and run "python -m PyInstaller --onefile --add-d
## Credits ## Credits
meha for Switch-Oc-Suite<br> meha for Switch-Oc-Suite<br>
sys-clk team for sys-clk<br> sys-clk team for sys-clk<br>
b0rd2auth for Ultrahand sys-clk fork<br> b0rd2death for Ultrahand sys-clk fork<br>
Lightos and Sammybigio2010 for early testing<br> Lightos and Sammybigio2011 for early testing<br>

4
Source/Atmosphere/stratosphere/loader/source/oc/customize.cpp
View File

@@ -105,6 +105,8 @@ volatile CustomizeTable C = {
.commonGpuVoltOffset = 0, .commonGpuVoltOffset = 0,
.marikoEmcDvbShift = 0,
.t1_tRCD = 0, .t1_tRCD = 0,
.t2_tRP = 0, .t2_tRP = 0,
.t3_tRAS = 0, .t3_tRAS = 0,
@@ -114,7 +116,7 @@ volatile CustomizeTable C = {
.t7_tWTR = 0, .t7_tWTR = 0,
.t8_tREFI = 0, .t8_tREFI = 0,
.mem_burst_latency = 2, .mem_burst_latency = 0,
// NOTE: These tables should NOT BE USED and are only here as placeholders. Always try and find your own optimal tables. // NOTE: These tables should NOT BE USED and are only here as placeholders. Always try and find your own optimal tables.
// Ensure the voltages actually increase or stay the sameot // Ensure the voltages actually increase or stay the sameot

10
Source/Atmosphere/stratosphere/loader/source/oc/customize.hpp
View File

@@ -30,12 +30,7 @@
#include "mtc_timing_table.hpp" #include "mtc_timing_table.hpp"
enum MtcConfig: u32 { enum MtcConfig: u32 {
AUTO_ADJ_ALL = 0, AUTO_ADJ = 0,
CUSTOM_ADJ_ALL = 1,
NO_ADJ_ALL = 2,
CUSTOMIZED_ALL = 4,
AUTO_ADJ = 5,
}; };
using CustomizeCpuDvfsTable = pcv::cvb_entry_t[pcv::DvfsTableEntryLimit]; using CustomizeCpuDvfsTable = pcv::cvb_entry_t[pcv::DvfsTableEntryLimit];
@@ -49,8 +44,7 @@
typedef struct CustomizeTable { typedef struct CustomizeTable {
u8 cust[4] = {'C', 'U', 'S', 'T'}; u8 cust[4] = {'C', 'U', 'S', 'T'};
u32 custRev = CUST_REV; u32 custRev = CUST_REV;
u32 mtcConfErista = AUTO_ADJ; u32 mtcConf = AUTO_ADJ;
u32 mtcConfMariko = AUTO_ADJ_ALL; // TODO: Fix mariko and merge into mtcConf
u32 commonCpuBoostClock; u32 commonCpuBoostClock;
u32 commonEmcMemVolt; u32 commonEmcMemVolt;
u32 eristaCpuMaxVolt; u32 eristaCpuMaxVolt;

158
Source/Atmosphere/stratosphere/loader/source/oc/mtc_timing_value.hpp
View File

@@ -61,6 +61,11 @@
// Burst Length // Burst Length
const u32 BL = 16; const u32 BL = 16;
// Write Latency
const u32 WL = 14 + C.mem_burst_latency;
// Read Latency
const u32 RL = 32 - C.mem_burst_latency;
// tRFCpb (refresh cycle time per bank) in ns for 8Gb density // tRFCpb (refresh cycle time per bank) in ns for 8Gb density
const u32 tRFCpb = !C.t5_tRFC ? 140 : tRFC_values[C.t5_tRFC-1]; const u32 tRFCpb = !C.t5_tRFC ? 140 : tRFC_values[C.t5_tRFC-1];
@@ -77,23 +82,11 @@
const u32 tRPab = !C.t2_tRP ? 21 : tRPpb + 3; const u32 tRPab = !C.t2_tRP ? 21 : tRPpb + 3;
// tRC (ACTIVATE-ACTIVATE command period same bank) in ns // tRC (ACTIVATE-ACTIVATE command period same bank) in ns
const u32 tRC = tRPpb + tRAS; const u32 tRC = tRPab + tRAS;
const u32 tPPD = 4;
const u32 tRTW = !C.t6_tRTW ? 10 : tWTR_values[C.t6_tRTW-1]; const u32 tRTW = !C.t6_tRTW ? 10 : tWTR_values[C.t6_tRTW-1];
// DQS output access time from CK_t/CK_c
const double tDQSCK_min = 1.5; // TODO: Fix/remove for mariko if needed
// DQS output access time from CK_t/CK_c
const double tDQSCK_max = 3.5; // TODO: Fix/remove for mariko if needed
// Write preamble (tCK)
const double tWPRE = 1.8; // TODO: Fix/remove for mariko if needed
// Read postamble (tCK)
const double tRPST = 0.4; // TODO: Fix/remove for mariko if needed
// WRITE command to first DQS transition(max) (tCK)
const double tDQSS_max = 1.25; // TODO: Fix/remove for mariko if needed
// DQ-to-DQS offset(max) (ns)
const double tDQS2DQ_max = 0.8; // TODO: Fix/remove for mariko if needed
// DQS_t, DQS_c to DQ skew total, per group, per access (DBI Disabled)
const double tDQSQ = 0.18; // TODO: Fix/remove for mariko if needed
// Write-to-Read delay // Write-to-Read delay
const u32 tWTR = !C.t7_tWTR ? 10 : tWTR_values[C.t7_tWTR-1]; const u32 tWTR = !C.t7_tWTR ? 10 : tWTR_values[C.t7_tWTR-1];
@@ -104,9 +97,6 @@
// write recovery time // write recovery time
const u32 tWR = !TIMING_PRESET_THREE ? 18 : tWR_values[TIMING_PRESET_THREE-1]; const u32 tWR = !TIMING_PRESET_THREE ? 18 : tWR_values[TIMING_PRESET_THREE-1];
// Read to refresh delay
const u32 tR2REF = tRTP + tRPpb;
// tRCD (RAS-CAS delay) in ns // tRCD (RAS-CAS delay) in ns
const u32 tRCD = !C.t1_tRCD ? 18 : tRCD_values[C.t1_tRCD-1]; const u32 tRCD = !C.t1_tRCD ? 18 : tRCD_values[C.t1_tRCD-1];
@@ -115,124 +105,60 @@
// tREFpb (average refresh interval per bank) in ns for 8Gb density // tREFpb (average refresh interval per bank) in ns for 8Gb density
const u32 tREFpb = !C.t8_tREFI ? 488 : tREFpb_values[C.t8_tREFI-1]; const u32 tREFpb = !C.t8_tREFI ? 488 : tREFpb_values[C.t8_tREFI-1];
// tREFab (average refresh interval all 8 banks) in ns for 8Gb density
// const u32 tREFab = tREFpb * 8;
// tPDEX2WR, tPDEX2RD (timing delay from exiting powerdown mode to a write/read command) in ns
// const u32 tPDEX2 = 10;
// Exit power-down to next valid command delay // Exit power-down to next valid command delay
const double tXP = 10; const double tXP = 7.5;
// Delay from valid command to CKE input LOW in ns
const double tCMDCKE = 1.75; // TODO: Fix/remove for mariko if needed
// tACT2PDEN (timing delay from an activate, MRS or EMRS command to power-down entry) in ns
// Valid clock and CS requirement after CKE input LOW after MRW command
const u32 tMRWCKEL = 14; // TODO: Fix/remove for mariko if needed
// Valid CS requirement after CKE input LOW
const double tCKELCS = 5; // TODO: Fix/remove for mariko if needed
// Valid CS requirement before CKE input HIGH
const double tCSCKEH = 1.75; // TODO: Fix/remove for mariko if needed
// tXSR (SELF REFRESH exit to next valid command delay) in ns // tXSR (SELF REFRESH exit to next valid command delay) in ns
const double tXSR = tRFCab + 7.5; const double tXSR = tRFCab + 7.5;
// tCKE (minimum pulse width(HIGH and LOW pulse width)) in ns
const double tCKE = 7.5; // TODO: Fix/remove for mariko if needed
// Minimum self refresh time (entry to exit) // Minimum self refresh time (entry to exit)
const u32 tSR = 15; const u32 tSR = 15;
// tFAW (Four-bank Activate Window) in ns // tFAW (Four-bank Activate Window) in ns
const u32 tFAW = 40;// !TIMING_PRESET_TWO ? 40 : tFAW_values[TIMING_PRESET_TWO-1]; TOGO const u32 tFAW = 40;// !TIMING_PRESET_TWO ? 40 : tFAW_values[TIMING_PRESET_TWO-1]; TOGO
// Valid Clock requirement before CKE Input HIGH in ns
const double tCKCKEH = 1.75; // TODO: Fix/remove for mariko if needed
// p78 The first valid data is available RL × t CK + t DQSCK + t DQSQ
//const u32 QUSE = RL + CEIL(tDQSCK_min/tCK_avg + tDQSQ);
namespace pcv::erista {
// tCK_avg (average clock period) in ns
const double tCK_avg = 1000'000. / C.eristaEmcMaxClock;
// Write Latency
const u32 WL = 14 + C.mem_burst_latency;
// Read Latency
const u32 RL = 32 - C.mem_burst_latency;
// minimum number of cycles from any read command to any write command, irrespective of bank
// const u32 R2W = CEIL (RL + CEIL(tDQSCK_max/tCK_avg) + BL/2 - WL + tWPRE + FLOOR(tRPST)) + 6;
// Delay Time From WRITE-to-READ
// const u32 W2R = WL + BL/2 + 1 + CEIL(tWTR/tCK_avg) - 6;
// write-to-precharge time for commands to the same bank in cycles
// const u32 WTP = WL + BL/2 + 1 + CEIL(tWR/tCK_avg) - 8;
// #_of_rows per die for 16Gb density
const u32 numOfRows = 131072;
// {REFRESH, REFRESH_LO} = max[(tREF/#_of_rows) / (emc_clk_period) - 64, (tREF/#_of_rows) / (emc_clk_period) * 97%]
// emc_clk_period = dram_clk / 2;
// 1600 MHz: 5894, but N' set to 6176 (~4.8% margin)
const u32 REFRESH = MIN((u32)65472, u32(std::ceil((double(tREFpb) * C.eristaEmcMaxClock / numOfRows * 1.048 / 2 - 64))) / 4 * 4);
const u32 REFBW = MIN((u32)65536, REFRESH+64);
// Write With Auto Precharge to to Power-Down Entry
// const u32 WTPDEN = WTP + 1 + CEIL(tDQSS_max/tCK_avg) + CEIL(tDQS2DQ_max/tCK_avg) + 6;
// Additional time after t XP hasexpired until the MRR commandmay be issued
// const double tMRRI = tRCD + 3 * tCK_avg;
// tPDEX2MRR (timing delay from exiting powerdown mode to MRR command) in ns
// const double tPDEX2MRR = tXP + tMRRI;
}
namespace pcv::mariko {
// tCK_avg (average clock period) in ns
const double tCK_avg = 1000'000. / C.marikoEmcMaxClock;
// Write Latency
const u32 WL = 14 + C.mem_burst_latency;
// Read Latency
const u32 RL = 32 - C.mem_burst_latency;
// minimum number of cycles from any read command to any write command, irrespective of bank
const u32 R2W = WL + BL/2 + 1 + CEIL(tRTW/tCK_avg);
// Delay Time From WRITE-to-READ
const u32 W2R = WL + BL/2 + 1 + CEIL(tWTR/tCK_avg);
// write-to-precharge time for commands to the same bank in cycles
const u32 WTP = WL + BL/2 + 1 + CEIL(tWR/tCK_avg);
// Read-To-MRW delay
const u32 RTM = RL + BL/2 + CEIL(tDQSCK_max/tCK_avg) + FLOOR(tRPST) + CEIL(7.5/tCK_avg);
// Write-To-MRW/MRR delay
const u32 WTM = WL + 1 + BL/2 + CEIL(7.5/tCK_avg);
// Read With AP-To-MRW/MRR delay
const u32 RATM = RTM + CEIL(tRTP/tCK_avg) - 8;
// Write With AP-To-MRW/MRR delay
const u32 WATM = WTM + CEIL(tWR/tCK_avg);
// #_of_rows per die for 8Gb density // #_of_rows per die for 8Gb density
const u32 numOfRows = 65536; const u32 numOfRows = 131072;
// {REFRESH, REFRESH_LO} = max[(tREF/#_of_rows) / (emc_clk_period) - 64, (tREF/#_of_rows) / (emc_clk_period) * 97%] // {REFRESH, REFRESH_LO} = max[(tREF/#_of_rows) / (emc_clk_period) - 64, (tREF/#_of_rows) / (emc_clk_period) * 97%]
// emc_clk_period = dram_clk / 2; // emc_clk_period = dram_clk / 2;
// 1600 MHz: 5894, but N' set to 6176 (~4.8% margin) // 1600 MHz: 5894, but N' set to 6176 (~4.8% margin)
const u32 REFRESH = MIN((u32)65472, u32(std::ceil((double(tREFpb) * C.marikoEmcMaxClock / numOfRows * 1.048 / 2 - 64))) / 4 * 4); const u32 REFRESH = MIN((u32)65472, u32(std::ceil((double(tREFpb) * C.marikoEmcMaxClock / numOfRows * 1.048 / 2 - 64))) / 4 * 4);
const u32 REFBW = MIN((u32)65536, REFRESH+64); const u32 REFBW = MIN((u32)65536, REFRESH+64);
// Write With Auto Precharge to to Power-Down Entry // DQS output access time from CK_t/CK_c
const u32 WTPDEN = WTP + 1 + CEIL(tDQSS_max/tCK_avg) + CEIL(tDQS2DQ_max/tCK_avg) + 6; const double tDQSCK_min = 1.5;
const double tDQSCK_max = 3.5;
// Write preamble (tCK)
const double tWPRE = 1.8;
// Read postamble (tCK)
const double tRPST = 0.4;
// Additional time after t XP hasexpired until the MRR commandmay be issued namespace pcv::erista {
const double tMRRI = tRCD + 3 * tCK_avg; // tCK_avg (average clock period) in ns
const double tCK_avg = 1000'000. / C.eristaEmcMaxClock;
// tPDEX2MRR (timing delay from exiting powerdown mode to MRR command) in ns // minimum number of cycles from any read command to any write command, irrespective of bank
const double tPDEX2MRR = tXP + tMRRI; const u32 R2W = CEIL (RL + CEIL(tDQSCK_max/tCK_avg) + BL/2 - WL + tWPRE + FLOOR(tRPST)) + 6;
// Delay Time From WRITE-to-READ
const u32 W2R = WL + BL/2 + 1 + CEIL(tWTR/tCK_avg) - 6;
// write-to-precharge time for commands to the same bank in cycles
const u32 WTP = WL + BL/2 + 1 + CEIL(tWR/tCK_avg) - 8;
}
namespace pcv::mariko {
// tCK_avg (average clock period) in ns
const double tCK_avg = 1000'000. / C.marikoEmcMaxClock;
const u32 R2W = CEIL (RL + CEIL(tDQSCK_max/tCK_avg) + BL/2 - WL + tWPRE + FLOOR(tRPST)) + 6;
// Delay Time From WRITE-to-READ
const u32 W2R = WL + BL/2 + 1 + CEIL(tWTR/tCK_avg) - 6;
// write-to-precharge time for commands to the same bank in cycles
const u32 WTP = WL + BL/2 + 1 + CEIL(tWR/tCK_avg) - 8;
} }
} }

5
Source/Atmosphere/stratosphere/loader/source/oc/pcv/pcv_erista.cpp
View File

@@ -160,7 +160,7 @@
} }
void MemMtcTableAutoAdjust(EristaMtcTable *table) { void MemMtcTableAutoAdjust(EristaMtcTable *table) {
if (C.mtcConfErista != AUTO_ADJ) if (C.mtcConf != AUTO_ADJ)
return; return;
#define WRITE_PARAM_ALL_REG(TABLE, PARAM, VALUE) \ #define WRITE_PARAM_ALL_REG(TABLE, PARAM, VALUE) \
@@ -177,6 +177,9 @@
WRITE_PARAM_ALL_REG(table, emc_ras, GET_CYCLE_CEIL(tRAS)); WRITE_PARAM_ALL_REG(table, emc_ras, GET_CYCLE_CEIL(tRAS));
WRITE_PARAM_ALL_REG(table, emc_rp, GET_CYCLE_CEIL(tRPpb)); WRITE_PARAM_ALL_REG(table, emc_rp, GET_CYCLE_CEIL(tRPpb));
WRITE_PARAM_ALL_REG(table, emc_r2p, GET_CYCLE_CEIL(tRTP)); WRITE_PARAM_ALL_REG(table, emc_r2p, GET_CYCLE_CEIL(tRTP));
WRITE_PARAM_ALL_REG(table, emc_r2w, R2W);
WRITE_PARAM_ALL_REG(table, emc_w2r, W2R);
WRITE_PARAM_ALL_REG(table, emc_w2p, WTP);
WRITE_PARAM_ALL_REG(table, emc_rd_rcd, GET_CYCLE_CEIL(tRCD)); WRITE_PARAM_ALL_REG(table, emc_rd_rcd, GET_CYCLE_CEIL(tRCD));
WRITE_PARAM_ALL_REG(table, emc_wr_rcd, GET_CYCLE_CEIL(tRCD)); WRITE_PARAM_ALL_REG(table, emc_wr_rcd, GET_CYCLE_CEIL(tRCD));
WRITE_PARAM_ALL_REG(table, emc_rrd, GET_CYCLE_CEIL(tRRD)); WRITE_PARAM_ALL_REG(table, emc_rrd, GET_CYCLE_CEIL(tRRD));

245
Source/Atmosphere/stratosphere/loader/source/oc/pcv/pcv_mariko.cpp
View File

@@ -209,7 +209,25 @@ Result GpuVmax(u32 *ptr)
R_SUCCEED(); R_SUCCEED();
} }
void MemMtcTableAutoAdjust(MarikoMtcTable* table, const MarikoMtcTable* ref) { /* Get RAM vendor data, ty b0rd2death! */
/* Note: I know this is horrible but I don't care atm. */
bool IsMicron() {
u64 packed_version;
splGetConfig((SplConfigItem)2, &packed_version);
switch (packed_version) {
case 11: case 15:
case 25: case 26: case 27:
case 32: case 33: case 34:
/* RAM is Micron. */
return true;
default:
/* Not Micron. */
return false;
}
}
void MemMtcTableAutoAdjust(MarikoMtcTable* table) {
/* Official Tegra X1 TRM, sign up for nvidia developer program (free) to download: /* Official Tegra X1 TRM, sign up for nvidia developer program (free) to download:
* https://developer.nvidia.com/embedded/dlc/tegra-x1-technical-reference-manual * https://developer.nvidia.com/embedded/dlc/tegra-x1-technical-reference-manual
* Section 18.11: MC Registers * Section 18.11: MC Registers
@@ -227,23 +245,9 @@ Result GpuVmax(u32 *ptr)
* you'd better calculate timings yourself rather than relying on following algorithm. * you'd better calculate timings yourself rather than relying on following algorithm.
*/ */
if (C.mtcConfMariko != AUTO_ADJ_ALL) if (C.mtcConf != AUTO_ADJ) {
return; return;
}
// scale with linear interpolation
#define ADJUST_PROP(TARGET, REF) \
(u32)(CEIL((REF + ((C.marikoEmcMaxClock-EmcClkOSAlt)*(TARGET-REF))/(EmcClkOSLimit-EmcClkOSAlt))))
#define ADJUST_PARAM(TARGET, REF) \
TARGET = ADJUST_PROP(TARGET, REF);
#define ADJUST_PARAM_TABLE(TABLE, PARAM, REF) ADJUST_PARAM(TABLE->PARAM, REF->PARAM)
// Burst Register
#define ADJUST_PARAM_ALL_REG(TABLE, PARAM, REF) \
ADJUST_PARAM_TABLE(TABLE, burst_regs.PARAM, REF) \
ADJUST_PARAM_TABLE(TABLE, shadow_regs_ca_train.PARAM, REF) \
ADJUST_PARAM_TABLE(TABLE, shadow_regs_rdwr_train.PARAM, REF)
#define WRITE_PARAM_BURST_REG(TABLE, PARAM, VALUE) TABLE->burst_regs.PARAM = VALUE; #define WRITE_PARAM_BURST_REG(TABLE, PARAM, VALUE) TABLE->burst_regs.PARAM = VALUE;
#define WRITE_PARAM_CA_TRAIN_REG(TABLE, PARAM, VALUE) TABLE->shadow_regs_ca_train.PARAM = VALUE; #define WRITE_PARAM_CA_TRAIN_REG(TABLE, PARAM, VALUE) TABLE->shadow_regs_ca_train.PARAM = VALUE;
@@ -261,219 +265,50 @@ Result GpuVmax(u32 *ptr)
WRITE_PARAM_ALL_REG(table, emc_rfcpb, GET_CYCLE_CEIL(tRFCpb)); WRITE_PARAM_ALL_REG(table, emc_rfcpb, GET_CYCLE_CEIL(tRFCpb));
WRITE_PARAM_ALL_REG(table, emc_ras, GET_CYCLE_CEIL(tRAS)); WRITE_PARAM_ALL_REG(table, emc_ras, GET_CYCLE_CEIL(tRAS));
WRITE_PARAM_ALL_REG(table, emc_rp, GET_CYCLE_CEIL(tRPpb)); WRITE_PARAM_ALL_REG(table, emc_rp, GET_CYCLE_CEIL(tRPpb));
WRITE_PARAM_ALL_REG(table, emc_r2w, R2W);
WRITE_PARAM_ALL_REG(table, emc_w2r, W2R);
WRITE_PARAM_ALL_REG(table, emc_r2p, GET_CYCLE_CEIL(tRTP)); WRITE_PARAM_ALL_REG(table, emc_r2p, GET_CYCLE_CEIL(tRTP));
WRITE_PARAM_ALL_REG(table, emc_w2p, WTP);
WRITE_PARAM_ALL_REG(table, emc_trtm, RTM);
WRITE_PARAM_ALL_REG(table, emc_twtm, WTM);
WRITE_PARAM_ALL_REG(table, emc_tratm, RATM);
WRITE_PARAM_ALL_REG(table, emc_twatm, WATM);
//WRITE_PARAM_ALL_REG(table, emc_tr2ref, GET_CYCLE_CEIL(tR2REF));
WRITE_PARAM_ALL_REG(table, emc_rd_rcd, GET_CYCLE_CEIL(tRCD)); WRITE_PARAM_ALL_REG(table, emc_rd_rcd, GET_CYCLE_CEIL(tRCD));
WRITE_PARAM_ALL_REG(table, emc_wr_rcd, GET_CYCLE_CEIL(tRCD)); WRITE_PARAM_ALL_REG(table, emc_wr_rcd, GET_CYCLE_CEIL(tRCD));
WRITE_PARAM_ALL_REG(table, emc_rrd, GET_CYCLE_CEIL(tRRD)); WRITE_PARAM_ALL_REG(table, emc_rrd, GET_CYCLE_CEIL(tRRD));
WRITE_PARAM_ALL_REG(table, emc_rext, 26);
WRITE_PARAM_ALL_REG(table, emc_refresh, REFRESH); WRITE_PARAM_ALL_REG(table, emc_refresh, REFRESH);
WRITE_PARAM_ALL_REG(table, emc_pre_refresh_req_cnt, REFRESH / 4); WRITE_PARAM_ALL_REG(table, emc_pre_refresh_req_cnt, REFRESH / 4);
WRITE_PARAM_ALL_REG(table, emc_r2w, R2W);
WRITE_PARAM_ALL_REG(table, emc_w2r, W2R);
WRITE_PARAM_ALL_REG(table, emc_w2p, WTP);
/* May or may not have to be patched in Micron; let's skip for now. */
if (!IsMicron()) {
WRITE_PARAM_ALL_REG(table, emc_pdex2wr, GET_CYCLE_CEIL(tXP)); WRITE_PARAM_ALL_REG(table, emc_pdex2wr, GET_CYCLE_CEIL(tXP));
WRITE_PARAM_ALL_REG(table, emc_pdex2rd, GET_CYCLE_CEIL(tXP)); WRITE_PARAM_ALL_REG(table, emc_pdex2rd, GET_CYCLE_CEIL(tXP));
WRITE_PARAM_ALL_REG(table, emc_pchg2pden, GET_CYCLE_CEIL(tCMDCKE)); }
WRITE_PARAM_ALL_REG(table, emc_act2pden, GET_CYCLE_CEIL(tMRWCKEL));
WRITE_PARAM_ALL_REG(table, emc_ar2pden, GET_CYCLE_CEIL(tCMDCKE));
WRITE_PARAM_ALL_REG(table, emc_rw2pden, WTPDEN);
WRITE_PARAM_ALL_REG(table, emc_cke2pden, GET_CYCLE_CEIL(tCKELCS));
//WRITE_PARAM_ALL_REG(table, emc_pdex2cke, GET_CYCLE_CEIL(tCSCKEH));
WRITE_PARAM_ALL_REG(table, emc_pdex2mrr, GET_CYCLE_CEIL(tPDEX2MRR));
WRITE_PARAM_ALL_REG(table, emc_txsr, MIN(GET_CYCLE_CEIL(tXSR), (u32)0x3fe)); WRITE_PARAM_ALL_REG(table, emc_txsr, MIN(GET_CYCLE_CEIL(tXSR), (u32)0x3fe));
WRITE_PARAM_ALL_REG(table, emc_txsrdll, MIN(GET_CYCLE_CEIL(tXSR), (u32)0x3fe)); WRITE_PARAM_ALL_REG(table, emc_txsrdll, MIN(GET_CYCLE_CEIL(tXSR), (u32)0x3fe));
WRITE_PARAM_ALL_REG(table, emc_tcke, GET_CYCLE_CEIL(tCKE) + 1);
WRITE_PARAM_ALL_REG(table, emc_tckesr, GET_CYCLE_CEIL(tSR)); WRITE_PARAM_ALL_REG(table, emc_tckesr, GET_CYCLE_CEIL(tSR));
WRITE_PARAM_ALL_REG(table, emc_tpd, GET_CYCLE_CEIL(tCKE));
WRITE_PARAM_ALL_REG(table, emc_tfaw, GET_CYCLE_CEIL(tFAW)); WRITE_PARAM_ALL_REG(table, emc_tfaw, GET_CYCLE_CEIL(tFAW));
WRITE_PARAM_ALL_REG(table, emc_trpab, GET_CYCLE_CEIL(tRPab)); WRITE_PARAM_ALL_REG(table, emc_trpab, GET_CYCLE_CEIL(tRPab));
//WRITE_PARAM_ALL_REG(table, emc_tclkstable, GET_CYCLE_CEIL(tCKCKEH));
WRITE_PARAM_ALL_REG(table, emc_tclkstop, GET_CYCLE_CEIL(tCKE) + 8);
WRITE_PARAM_ALL_REG(table, emc_trefbw, REFBW); WRITE_PARAM_ALL_REG(table, emc_trefbw, REFBW);
ADJUST_PARAM_ALL_REG(table, emc_dyn_self_ref_control, ref); /* Worth replacing with l4t dumps at some point. */
#define CLEAR_BIT(BITS, HIGH, LOW) \
BITS = BITS & ~( ((1u << HIGH) << 1u) - (1u << LOW) );
#define ADJUST(TARGET) (u32)CEIL(TARGET * (C.marikoEmcMaxClock / EmcClkOSLimit))
#define ADJUST_INVERSE(TARGET) (u32)(TARGET * (EmcClkOSLimit / 1000) / (C.marikoEmcMaxClock / 1000))
// Burst MC Regs // Burst MC Regs
#define WRITE_PARAM_BURST_MC_REG(TABLE, PARAM, VALUE) TABLE->burst_mc_regs.PARAM = VALUE; #define WRITE_PARAM_BURST_MC_REG(TABLE, PARAM, VALUE) TABLE->burst_mc_regs.PARAM = VALUE;
constexpr u32 MC_ARB_DIV = 4; constexpr u32 MC_ARB_DIV = 4;
constexpr u32 MC_ARB_SFA = 2; constexpr u32 MC_ARB_SFA = 2;
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_cfg, C.marikoEmcMaxClock / (33.3 * 1000) / MC_ARB_DIV); //CYCLES_PER_UPDATE: The number of mcclk cycles per deadline timer update // WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_cfg, C.marikoEmcMaxClock / (33.3 * 1000) / MC_ARB_DIV); //CYCLES_PER_UPDATE: The number of mcclk cycles per deadline timer update
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rcd, CEIL(GET_CYCLE_CEIL(tRCD) / MC_ARB_DIV) - 2) WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rcd, CEIL(GET_CYCLE_CEIL(tRCD) / MC_ARB_DIV) - 2)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rp, CEIL(GET_CYCLE_CEIL(tRPpb) / MC_ARB_DIV) - 1 + MC_ARB_SFA) WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rp, CEIL(GET_CYCLE_CEIL(tRPpb) / MC_ARB_DIV) - 1 + MC_ARB_SFA)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rc, CEIL(GET_CYCLE_CEIL(tRC) / MC_ARB_DIV) - 1) WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rc, CEIL(GET_CYCLE_CEIL(tRC) / MC_ARB_DIV) - 1)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_ras, CEIL(GET_CYCLE_CEIL(tRAS) / MC_ARB_DIV) - 2) WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_ras, CEIL(GET_CYCLE_CEIL(tRAS) / MC_ARB_DIV) - 2)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_faw, CEIL(GET_CYCLE_CEIL(tFAW) / MC_ARB_DIV) - 1) WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_faw, CEIL(GET_CYCLE_CEIL(tFAW) / MC_ARB_DIV) - 1)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rrd, CEIL(GET_CYCLE_CEIL(tRRD) / MC_ARB_DIV) - 1) WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rrd, CEIL(GET_CYCLE_CEIL(tRRD) / MC_ARB_DIV) - 1)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rap2pre, CEIL(GET_CYCLE_CEIL(tRTP) / MC_ARB_DIV)) // WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rap2pre, CEIL(GET_CYCLE_CEIL(tRTP) / MC_ARB_DIV))
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_wap2pre, CEIL((WTP) / MC_ARB_DIV)) // WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_wap2pre, CEIL((WTP) / MC_ARB_DIV))
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_r2r, CEIL(table->burst_regs.emc_rext / MC_ARB_DIV) - 1 + MC_ARB_SFA) // WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_r2r, CEIL(table->burst_regs.emc_rext / MC_ARB_DIV) - 1 + MC_ARB_SFA)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_r2w, CEIL((R2W) / MC_ARB_DIV) - 1 + MC_ARB_SFA) // WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_r2w, CEIL((R2W) / MC_ARB_DIV) - 1 + MC_ARB_SFA)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_w2r, CEIL((W2R) / MC_ARB_DIV) - 1 + MC_ARB_SFA) // WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_w2r, CEIL((W2R) / MC_ARB_DIV) - 1 + MC_ARB_SFA)
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rfcpb, CEIL(GET_CYCLE_CEIL(tRFCpb) / MC_ARB_DIV)) WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_timing_rfcpb, CEIL(GET_CYCLE_CEIL(tRFCpb) / MC_ARB_DIV))
u32 DA_TURNS = 0;
DA_TURNS |= u8(table->burst_mc_regs.mc_emem_arb_timing_r2w / 2) << 16; //R2W TURN
DA_TURNS |= u8(table->burst_mc_regs.mc_emem_arb_timing_w2r / 2) << 24; //W2R TURN
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_da_turns, DA_TURNS);
u32 DA_COVERS = 0;
u8 R_COVER = (table->burst_mc_regs.mc_emem_arb_timing_rap2pre + table->burst_mc_regs.mc_emem_arb_timing_rp + table->burst_mc_regs.mc_emem_arb_timing_rcd) / 2;
u8 W_COVER = (table->burst_mc_regs.mc_emem_arb_timing_wap2pre + table->burst_mc_regs.mc_emem_arb_timing_rp + table->burst_mc_regs.mc_emem_arb_timing_rcd) / 2;
DA_COVERS |= (u8)(table->burst_mc_regs.mc_emem_arb_timing_rc / 2); //RC COVER
DA_COVERS |= (R_COVER << 8); //RCD_R COVER
DA_COVERS |= (W_COVER << 16); //RCD_W COVER
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_da_covers, DA_COVERS);
CLEAR_BIT(table->burst_mc_regs.mc_emem_arb_misc0, 7, 0);
table->burst_mc_regs.mc_emem_arb_misc0 |= u8(table->burst_mc_regs.mc_emem_arb_timing_rc + 1); //BC2AA_HOLDOFF
CLEAR_BIT(table->burst_mc_regs.mc_emem_arb_misc0, 14, 8);
table->burst_mc_regs.mc_emem_arb_misc0 |= u8((ADJUST(0x24) << 8)); //PRIORITY_INVERSION_THRESHOLD
CLEAR_BIT(table->burst_mc_regs.mc_emem_arb_misc0, 20, 16);
table->burst_mc_regs.mc_emem_arb_misc0 |= u8((ADJUST(12) << 16)); //PRIORITY_INVERSION_ISO_THRESHOLD
// updown registers
#define ADJUST_PARAM_LA_SCALE_REG(TABLE, PARAM) \
TABLE->la_scale_regs.PARAM = ADJUST(TABLE->la_scale_regs.PARAM)
#define ADJUST_PARAM_LA_SCALE_REG_HI(TABLE, PARAM, VALUE) \
CLEAR_BIT(TABLE->la_scale_regs.PARAM, 23, 16) \
TABLE->la_scale_regs.PARAM |= VALUE << 16
#define ADJUST_PARAM_LA_SCALE_REG_LO(TABLE, PARAM, VALUE) \
CLEAR_BIT(TABLE->la_scale_regs.PARAM, 7, 0) \
TABLE->la_scale_regs.PARAM |= VALUE
u8 LA = ADJUST_INVERSE(128); //0x80
ADJUST_PARAM_LA_SCALE_REG(table, mc_mll_mpcorer_ptsa_rate); //208
ADJUST_PARAM_LA_SCALE_REG(table, mc_ptsa_grant_decrement); //4611
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_xusb_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_xusb_1, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_tsec_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_sdmmca_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_sdmmcaa_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_sdmmc_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_sdmmcab_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_sdmmc_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_sdmmcab_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_ppcs_1, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_mpcore_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_avpc_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_gpu_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_gpu2_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_nvenc_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_nvdec_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_vic_0, LA);
ADJUST_PARAM_LA_SCALE_REG_HI(table, mc_latency_allowance_isp2_1, LA);
ADJUST_PARAM_LA_SCALE_REG_LO(table, mc_latency_allowance_hc_0, ADJUST_INVERSE(0x16));
ADJUST_PARAM_LA_SCALE_REG_LO(table, mc_latency_allowance_hc_1, LA);
ADJUST_PARAM_LA_SCALE_REG_LO(table, mc_latency_allowance_gpu_0, ADJUST_INVERSE(0x19));
ADJUST_PARAM_LA_SCALE_REG_LO(table, mc_latency_allowance_gpu2_0, ADJUST_INVERSE(0x19));
ADJUST_PARAM_LA_SCALE_REG_LO(table, mc_latency_allowance_vic_0, ADJUST_INVERSE(0x1d));
ADJUST_PARAM_LA_SCALE_REG_LO(table, mc_latency_allowance_vi2_0, LA);
ADJUST_PARAM_LA_SCALE_REG_LO(table, mc_latency_allowance_isp2_1, LA);
//Spread Spectrum Control
table->pllm_ss_ctrl1 = 0x0b55fe01;
table->pllm_ss_ctrl2 = 0x10170b55;
table->pllmb_ss_ctrl1 = 0x0b55fe01;
table->pllmb_ss_ctrl2 = 0x10170b55;
table->dram_timings.t_rp = tRPpb;
table->dram_timings.t_rfc = tRFCab;
//table->dram_timings.rl = 32;
table->emc_cfg_2 = 0x0011083d;
}
void MemMtcTableCustomAdjust(MarikoMtcTable* table) {
if (C.mtcConfMariko != CUSTOM_ADJ_ALL)
return;
constexpr u32 MC_ARB_DIV = 4;
constexpr u32 MC_ARB_SFA = 2;
WRITE_PARAM_ALL_REG(table, emc_rc, GET_CYCLE_CEIL(tRC));
WRITE_PARAM_ALL_REG(table, emc_ras, GET_CYCLE_CEIL(tRAS));
WRITE_PARAM_ALL_REG(table, emc_rp, GET_CYCLE_CEIL(tRPpb));
WRITE_PARAM_ALL_REG(table, emc_trpab, GET_CYCLE_CEIL(tRPab));
WRITE_PARAM_ALL_REG(table, emc_rd_rcd, GET_CYCLE_CEIL(tRCD));
WRITE_PARAM_ALL_REG(table, emc_wr_rcd, GET_CYCLE_CEIL(tRCD));
WRITE_PARAM_ALL_REG(table, emc_pdex2mrr,GET_CYCLE_CEIL(tPDEX2MRR));
table->burst_mc_regs.mc_emem_arb_timing_rcd = CEIL(GET_CYCLE_CEIL(tRCD) / MC_ARB_DIV) - 2;
table->burst_mc_regs.mc_emem_arb_timing_rc = CEIL(GET_CYCLE_CEIL(tRC) / MC_ARB_DIV) - 1;
table->burst_mc_regs.mc_emem_arb_timing_rp = CEIL(GET_CYCLE_CEIL(tRPpb) / MC_ARB_DIV) - 1 + MC_ARB_SFA;
table->burst_mc_regs.mc_emem_arb_timing_ras = CEIL(GET_CYCLE_CEIL(tRAS) / MC_ARB_DIV) - 2;
WRITE_PARAM_ALL_REG(table, emc_tfaw, GET_CYCLE_CEIL(tFAW));
WRITE_PARAM_ALL_REG(table, emc_rrd, GET_CYCLE_CEIL(tRRD));
table->burst_mc_regs.mc_emem_arb_timing_faw = CEIL(GET_CYCLE_CEIL(tFAW) / MC_ARB_DIV) - 1;
table->burst_mc_regs.mc_emem_arb_timing_rrd = CEIL(GET_CYCLE_CEIL(tRRD) / MC_ARB_DIV) - 1;
WRITE_PARAM_ALL_REG(table, emc_r2p, GET_CYCLE_CEIL(tRTP));
WRITE_PARAM_ALL_REG(table, emc_w2p, WTP);
WRITE_PARAM_ALL_REG(table, emc_tratm, RATM);
WRITE_PARAM_ALL_REG(table, emc_twatm, WATM);
WRITE_PARAM_ALL_REG(table, emc_rw2pden, WTPDEN);
table->burst_mc_regs.mc_emem_arb_timing_rap2pre = CEIL(GET_CYCLE_CEIL(tRTP) / MC_ARB_DIV);
table->burst_mc_regs.mc_emem_arb_timing_wap2pre = CEIL(WTP / MC_ARB_DIV);
WRITE_PARAM_ALL_REG(table, emc_rfc, GET_CYCLE_CEIL(tRFCab));
WRITE_PARAM_ALL_REG(table, emc_rfcpb, GET_CYCLE_CEIL(tRFCpb));
WRITE_PARAM_ALL_REG(table, emc_txsr, MIN(GET_CYCLE_CEIL(tXSR), (u32)0x3fe));
WRITE_PARAM_ALL_REG(table, emc_txsrdll, MIN(GET_CYCLE_CEIL(tXSR), (u32)0x3fe));
table->burst_mc_regs.mc_emem_arb_timing_rfcpb = CEIL(GET_CYCLE_CEIL(tRFCpb) / MC_ARB_DIV);
WRITE_PARAM_ALL_REG(table, emc_w2r, W2R);
table->burst_mc_regs.mc_emem_arb_timing_w2r = CEIL(W2R / MC_ARB_DIV) - 1 + MC_ARB_SFA;
WRITE_PARAM_ALL_REG(table, emc_refresh, REFRESH);
WRITE_PARAM_ALL_REG(table, emc_pre_refresh_req_cnt, REFRESH / 4);
WRITE_PARAM_ALL_REG(table, emc_trefbw, REFBW);
WRITE_PARAM_ALL_REG(table, emc_r2w, R2W);
WRITE_PARAM_ALL_REG(table, emc_w2r, W2R);
WRITE_PARAM_ALL_REG(table, emc_w2p, WTP);
WRITE_PARAM_ALL_REG(table, emc_trtm, RTM);
WRITE_PARAM_ALL_REG(table, emc_twtm, WTM);
WRITE_PARAM_ALL_REG(table, emc_tratm, RATM);
WRITE_PARAM_ALL_REG(table, emc_twatm, WATM);
WRITE_PARAM_ALL_REG(table, emc_rw2pden, WTPDEN);
table->burst_mc_regs.mc_emem_arb_timing_wap2pre = CEIL(WTP / MC_ARB_DIV);
table->burst_mc_regs.mc_emem_arb_timing_r2w = CEIL(R2W / MC_ARB_DIV) - 1 + MC_ARB_SFA;
table->burst_mc_regs.mc_emem_arb_timing_w2r = CEIL(W2R / MC_ARB_DIV) - 1 + MC_ARB_SFA;
u32 DA_TURNS = 0;
DA_TURNS |= u8(table->burst_mc_regs.mc_emem_arb_timing_r2w / 2) << 16; //R2W TURN
DA_TURNS |= u8(table->burst_mc_regs.mc_emem_arb_timing_w2r / 2) << 24; //W2R TURN
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_da_turns, DA_TURNS);
u32 DA_COVERS = 0;
u8 R_COVER = (table->burst_mc_regs.mc_emem_arb_timing_rap2pre + table->burst_mc_regs.mc_emem_arb_timing_rp + table->burst_mc_regs.mc_emem_arb_timing_rcd) / 2;
u8 W_COVER = (table->burst_mc_regs.mc_emem_arb_timing_wap2pre + table->burst_mc_regs.mc_emem_arb_timing_rp + table->burst_mc_regs.mc_emem_arb_timing_rcd) / 2;
DA_COVERS |= (u8)(table->burst_mc_regs.mc_emem_arb_timing_rc / 2); //RC COVER
DA_COVERS |= (R_COVER << 8); //RCD_R COVER
DA_COVERS |= (W_COVER << 16); //RCD_W COVER
WRITE_PARAM_BURST_MC_REG(table, mc_emem_arb_da_covers, DA_COVERS);
} }
void MemMtcPllmbDivisor(MarikoMtcTable* table) { void MemMtcPllmbDivisor(MarikoMtcTable* table) {
@@ -527,8 +362,7 @@ Result GpuVmax(u32 *ptr)
// Copy unmodified 1600000 table to tmp // Copy unmodified 1600000 table to tmp
std::memcpy(reinterpret_cast<void *>(tmp), reinterpret_cast<void *>(table_max), sizeof(MarikoMtcTable)); std::memcpy(reinterpret_cast<void *>(tmp), reinterpret_cast<void *>(table_max), sizeof(MarikoMtcTable));
// Adjust max freq mtc timing parameters with reference to 1331200 table // Adjust max freq mtc timing parameters with reference to 1331200 table
MemMtcTableAutoAdjust(table_max, table_alt); MemMtcTableAutoAdjust(table_max);
MemMtcTableCustomAdjust(table_max);
MemMtcPllmbDivisor(table_max); MemMtcPllmbDivisor(table_max);
// Overwrite 13312000 table with unmodified 1600000 table copied back // Overwrite 13312000 table with unmodified 1600000 table copied back
std::memcpy(reinterpret_cast<void *>(table_alt), reinterpret_cast<void *>(tmp), sizeof(MarikoMtcTable)); std::memcpy(reinterpret_cast<void *>(table_alt), reinterpret_cast<void *>(tmp), sizeof(MarikoMtcTable));
@@ -672,8 +506,7 @@ Result GpuVmax(u32 *ptr)
for (uintptr_t ptr = mapped_nso; for (uintptr_t ptr = mapped_nso;
ptr <= mapped_nso + nso_size - sizeof(MarikoMtcTable); ptr <= mapped_nso + nso_size - sizeof(MarikoMtcTable);
ptr += sizeof(u32)) ptr += sizeof(u32)) {
{
u32* ptr32 = reinterpret_cast<u32 *>(ptr); u32* ptr32 = reinterpret_cast<u32 *>(ptr);
for (auto& entry : patches) { for (auto& entry : patches) {
if (R_SUCCEEDED(entry.SearchAndApply(ptr32))) if (R_SUCCEEDED(entry.SearchAndApply(ptr32)))