1 /* 2 * Copyright 2019 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24 #define SWSMU_CODE_LAYER_L2 25 26 #include <linux/firmware.h> 27 #include "amdgpu.h" 28 #include "amdgpu_dpm.h" 29 #include "amdgpu_smu.h" 30 #include "atomfirmware.h" 31 #include "amdgpu_atomfirmware.h" 32 #include "amdgpu_atombios.h" 33 #include "smu_v13_0.h" 34 #include "smu13_driver_if_aldebaran.h" 35 #include "soc15_common.h" 36 #include "atom.h" 37 #include "aldebaran_ppt.h" 38 #include "smu_v13_0_pptable.h" 39 #include "aldebaran_ppsmc.h" 40 #include "nbio/nbio_7_4_offset.h" 41 #include "nbio/nbio_7_4_sh_mask.h" 42 #include "thm/thm_11_0_2_offset.h" 43 #include "thm/thm_11_0_2_sh_mask.h" 44 #include "amdgpu_xgmi.h" 45 #include <linux/pci.h> 46 #include "amdgpu_ras.h" 47 #include "smu_cmn.h" 48 #include "mp/mp_13_0_2_offset.h" 49 50 /* 51 * DO NOT use these for err/warn/info/debug messages. 52 * Use dev_err, dev_warn, dev_info and dev_dbg instead. 53 * They are more MGPU friendly. 54 */ 55 #undef pr_err 56 #undef pr_warn 57 #undef pr_info 58 #undef pr_debug 59 60 #define ALDEBARAN_FEA_MAP(smu_feature, aldebaran_feature) \ 61 [smu_feature] = {1, (aldebaran_feature)} 62 63 #define FEATURE_MASK(feature) (1ULL << feature) 64 #define SMC_DPM_FEATURE ( \ 65 FEATURE_MASK(FEATURE_DATA_CALCULATIONS) | \ 66 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT) | \ 67 FEATURE_MASK(FEATURE_DPM_UCLK_BIT) | \ 68 FEATURE_MASK(FEATURE_DPM_SOCCLK_BIT) | \ 69 FEATURE_MASK(FEATURE_DPM_FCLK_BIT) | \ 70 FEATURE_MASK(FEATURE_DPM_LCLK_BIT) | \ 71 FEATURE_MASK(FEATURE_DPM_XGMI_BIT) | \ 72 FEATURE_MASK(FEATURE_DPM_VCN_BIT)) 73 74 /* possible frequency drift (1Mhz) */ 75 #define EPSILON 1 76 77 #define smnPCIE_ESM_CTRL 0x111003D0 78 79 /* 80 * SMU support ECCTABLE since version 68.42.0, 81 * use this to check ECCTALE feature whether support 82 */ 83 #define SUPPORT_ECCTABLE_SMU_VERSION 0x00442a00 84 85 /* 86 * SMU support BAD CHENNEL info MSG since version 68.51.00, 87 * use this to check ECCTALE feature whether support 88 */ 89 #define SUPPORT_BAD_CHANNEL_INFO_MSG_VERSION 0x00443300 90 91 static const struct smu_temperature_range smu13_thermal_policy[] = 92 { 93 {-273150, 99000, 99000, -273150, 99000, 99000, -273150, 99000, 99000}, 94 { 120000, 120000, 120000, 120000, 120000, 120000, 120000, 120000, 120000}, 95 }; 96 97 static const struct cmn2asic_msg_mapping aldebaran_message_map[SMU_MSG_MAX_COUNT] = { 98 MSG_MAP(TestMessage, PPSMC_MSG_TestMessage, 0), 99 MSG_MAP(GetSmuVersion, PPSMC_MSG_GetSmuVersion, 1), 100 MSG_MAP(GetDriverIfVersion, PPSMC_MSG_GetDriverIfVersion, 1), 101 MSG_MAP(EnableAllSmuFeatures, PPSMC_MSG_EnableAllSmuFeatures, 0), 102 MSG_MAP(DisableAllSmuFeatures, PPSMC_MSG_DisableAllSmuFeatures, 0), 103 MSG_MAP(GetEnabledSmuFeaturesLow, PPSMC_MSG_GetEnabledSmuFeaturesLow, 1), 104 MSG_MAP(GetEnabledSmuFeaturesHigh, PPSMC_MSG_GetEnabledSmuFeaturesHigh, 1), 105 MSG_MAP(SetDriverDramAddrHigh, PPSMC_MSG_SetDriverDramAddrHigh, 1), 106 MSG_MAP(SetDriverDramAddrLow, PPSMC_MSG_SetDriverDramAddrLow, 1), 107 MSG_MAP(SetToolsDramAddrHigh, PPSMC_MSG_SetToolsDramAddrHigh, 0), 108 MSG_MAP(SetToolsDramAddrLow, PPSMC_MSG_SetToolsDramAddrLow, 0), 109 MSG_MAP(TransferTableSmu2Dram, PPSMC_MSG_TransferTableSmu2Dram, 1), 110 MSG_MAP(TransferTableDram2Smu, PPSMC_MSG_TransferTableDram2Smu, 0), 111 MSG_MAP(UseDefaultPPTable, PPSMC_MSG_UseDefaultPPTable, 0), 112 MSG_MAP(SetSystemVirtualDramAddrHigh, PPSMC_MSG_SetSystemVirtualDramAddrHigh, 0), 113 MSG_MAP(SetSystemVirtualDramAddrLow, PPSMC_MSG_SetSystemVirtualDramAddrLow, 0), 114 MSG_MAP(SetSoftMinByFreq, PPSMC_MSG_SetSoftMinByFreq, 0), 115 MSG_MAP(SetSoftMaxByFreq, PPSMC_MSG_SetSoftMaxByFreq, 0), 116 MSG_MAP(SetHardMinByFreq, PPSMC_MSG_SetHardMinByFreq, 0), 117 MSG_MAP(SetHardMaxByFreq, PPSMC_MSG_SetHardMaxByFreq, 0), 118 MSG_MAP(GetMinDpmFreq, PPSMC_MSG_GetMinDpmFreq, 0), 119 MSG_MAP(GetMaxDpmFreq, PPSMC_MSG_GetMaxDpmFreq, 0), 120 MSG_MAP(GetDpmFreqByIndex, PPSMC_MSG_GetDpmFreqByIndex, 1), 121 MSG_MAP(SetWorkloadMask, PPSMC_MSG_SetWorkloadMask, 1), 122 MSG_MAP(GetVoltageByDpm, PPSMC_MSG_GetVoltageByDpm, 0), 123 MSG_MAP(GetVoltageByDpmOverdrive, PPSMC_MSG_GetVoltageByDpmOverdrive, 0), 124 MSG_MAP(SetPptLimit, PPSMC_MSG_SetPptLimit, 0), 125 MSG_MAP(GetPptLimit, PPSMC_MSG_GetPptLimit, 1), 126 MSG_MAP(PrepareMp1ForUnload, PPSMC_MSG_PrepareMp1ForUnload, 0), 127 MSG_MAP(GfxDeviceDriverReset, PPSMC_MSG_GfxDriverReset, 0), 128 MSG_MAP(RunDcBtc, PPSMC_MSG_RunDcBtc, 0), 129 MSG_MAP(DramLogSetDramAddrHigh, PPSMC_MSG_DramLogSetDramAddrHigh, 0), 130 MSG_MAP(DramLogSetDramAddrLow, PPSMC_MSG_DramLogSetDramAddrLow, 0), 131 MSG_MAP(DramLogSetDramSize, PPSMC_MSG_DramLogSetDramSize, 0), 132 MSG_MAP(GetDebugData, PPSMC_MSG_GetDebugData, 0), 133 MSG_MAP(WaflTest, PPSMC_MSG_WaflTest, 0), 134 MSG_MAP(SetMemoryChannelEnable, PPSMC_MSG_SetMemoryChannelEnable, 0), 135 MSG_MAP(SetNumBadHbmPagesRetired, PPSMC_MSG_SetNumBadHbmPagesRetired, 0), 136 MSG_MAP(DFCstateControl, PPSMC_MSG_DFCstateControl, 0), 137 MSG_MAP(GetGmiPwrDnHyst, PPSMC_MSG_GetGmiPwrDnHyst, 0), 138 MSG_MAP(SetGmiPwrDnHyst, PPSMC_MSG_SetGmiPwrDnHyst, 0), 139 MSG_MAP(GmiPwrDnControl, PPSMC_MSG_GmiPwrDnControl, 0), 140 MSG_MAP(EnterGfxoff, PPSMC_MSG_EnterGfxoff, 0), 141 MSG_MAP(ExitGfxoff, PPSMC_MSG_ExitGfxoff, 0), 142 MSG_MAP(SetExecuteDMATest, PPSMC_MSG_SetExecuteDMATest, 0), 143 MSG_MAP(EnableDeterminism, PPSMC_MSG_EnableDeterminism, 0), 144 MSG_MAP(DisableDeterminism, PPSMC_MSG_DisableDeterminism, 0), 145 MSG_MAP(SetUclkDpmMode, PPSMC_MSG_SetUclkDpmMode, 0), 146 MSG_MAP(GfxDriverResetRecovery, PPSMC_MSG_GfxDriverResetRecovery, 0), 147 MSG_MAP(BoardPowerCalibration, PPSMC_MSG_BoardPowerCalibration, 0), 148 MSG_MAP(HeavySBR, PPSMC_MSG_HeavySBR, 0), 149 MSG_MAP(SetBadHBMPagesRetiredFlagsPerChannel, PPSMC_MSG_SetBadHBMPagesRetiredFlagsPerChannel, 0), 150 }; 151 152 static const struct cmn2asic_mapping aldebaran_clk_map[SMU_CLK_COUNT] = { 153 CLK_MAP(GFXCLK, PPCLK_GFXCLK), 154 CLK_MAP(SCLK, PPCLK_GFXCLK), 155 CLK_MAP(SOCCLK, PPCLK_SOCCLK), 156 CLK_MAP(FCLK, PPCLK_FCLK), 157 CLK_MAP(UCLK, PPCLK_UCLK), 158 CLK_MAP(MCLK, PPCLK_UCLK), 159 CLK_MAP(DCLK, PPCLK_DCLK), 160 CLK_MAP(VCLK, PPCLK_VCLK), 161 CLK_MAP(LCLK, PPCLK_LCLK), 162 }; 163 164 static const struct cmn2asic_mapping aldebaran_feature_mask_map[SMU_FEATURE_COUNT] = { 165 ALDEBARAN_FEA_MAP(SMU_FEATURE_DATA_CALCULATIONS_BIT, FEATURE_DATA_CALCULATIONS), 166 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_GFXCLK_BIT, FEATURE_DPM_GFXCLK_BIT), 167 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_UCLK_BIT, FEATURE_DPM_UCLK_BIT), 168 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_SOCCLK_BIT, FEATURE_DPM_SOCCLK_BIT), 169 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_FCLK_BIT, FEATURE_DPM_FCLK_BIT), 170 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_LCLK_BIT, FEATURE_DPM_LCLK_BIT), 171 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_XGMI_BIT, FEATURE_DPM_XGMI_BIT), 172 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_GFXCLK_BIT, FEATURE_DS_GFXCLK_BIT), 173 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_SOCCLK_BIT, FEATURE_DS_SOCCLK_BIT), 174 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_LCLK_BIT, FEATURE_DS_LCLK_BIT), 175 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_FCLK_BIT, FEATURE_DS_FCLK_BIT), 176 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_UCLK_BIT, FEATURE_DS_UCLK_BIT), 177 ALDEBARAN_FEA_MAP(SMU_FEATURE_GFX_SS_BIT, FEATURE_GFX_SS_BIT), 178 ALDEBARAN_FEA_MAP(SMU_FEATURE_VCN_DPM_BIT, FEATURE_DPM_VCN_BIT), 179 ALDEBARAN_FEA_MAP(SMU_FEATURE_RSMU_SMN_CG_BIT, FEATURE_RSMU_SMN_CG_BIT), 180 ALDEBARAN_FEA_MAP(SMU_FEATURE_WAFL_CG_BIT, FEATURE_WAFL_CG_BIT), 181 ALDEBARAN_FEA_MAP(SMU_FEATURE_PPT_BIT, FEATURE_PPT_BIT), 182 ALDEBARAN_FEA_MAP(SMU_FEATURE_TDC_BIT, FEATURE_TDC_BIT), 183 ALDEBARAN_FEA_MAP(SMU_FEATURE_APCC_PLUS_BIT, FEATURE_APCC_PLUS_BIT), 184 ALDEBARAN_FEA_MAP(SMU_FEATURE_APCC_DFLL_BIT, FEATURE_APCC_DFLL_BIT), 185 ALDEBARAN_FEA_MAP(SMU_FEATURE_FUSE_CG_BIT, FEATURE_FUSE_CG_BIT), 186 ALDEBARAN_FEA_MAP(SMU_FEATURE_MP1_CG_BIT, FEATURE_MP1_CG_BIT), 187 ALDEBARAN_FEA_MAP(SMU_FEATURE_SMUIO_CG_BIT, FEATURE_SMUIO_CG_BIT), 188 ALDEBARAN_FEA_MAP(SMU_FEATURE_THM_CG_BIT, FEATURE_THM_CG_BIT), 189 ALDEBARAN_FEA_MAP(SMU_FEATURE_CLK_CG_BIT, FEATURE_CLK_CG_BIT), 190 ALDEBARAN_FEA_MAP(SMU_FEATURE_FW_CTF_BIT, FEATURE_FW_CTF_BIT), 191 ALDEBARAN_FEA_MAP(SMU_FEATURE_THERMAL_BIT, FEATURE_THERMAL_BIT), 192 ALDEBARAN_FEA_MAP(SMU_FEATURE_OUT_OF_BAND_MONITOR_BIT, FEATURE_OUT_OF_BAND_MONITOR_BIT), 193 ALDEBARAN_FEA_MAP(SMU_FEATURE_XGMI_PER_LINK_PWR_DWN_BIT,FEATURE_XGMI_PER_LINK_PWR_DWN), 194 ALDEBARAN_FEA_MAP(SMU_FEATURE_DF_CSTATE_BIT, FEATURE_DF_CSTATE), 195 }; 196 197 static const struct cmn2asic_mapping aldebaran_table_map[SMU_TABLE_COUNT] = { 198 TAB_MAP(PPTABLE), 199 TAB_MAP(AVFS_PSM_DEBUG), 200 TAB_MAP(AVFS_FUSE_OVERRIDE), 201 TAB_MAP(PMSTATUSLOG), 202 TAB_MAP(SMU_METRICS), 203 TAB_MAP(DRIVER_SMU_CONFIG), 204 TAB_MAP(I2C_COMMANDS), 205 TAB_MAP(ECCINFO), 206 }; 207 208 static const uint8_t aldebaran_throttler_map[] = { 209 [THROTTLER_PPT0_BIT] = (SMU_THROTTLER_PPT0_BIT), 210 [THROTTLER_PPT1_BIT] = (SMU_THROTTLER_PPT1_BIT), 211 [THROTTLER_TDC_GFX_BIT] = (SMU_THROTTLER_TDC_GFX_BIT), 212 [THROTTLER_TDC_SOC_BIT] = (SMU_THROTTLER_TDC_SOC_BIT), 213 [THROTTLER_TDC_HBM_BIT] = (SMU_THROTTLER_TDC_MEM_BIT), 214 [THROTTLER_TEMP_GPU_BIT] = (SMU_THROTTLER_TEMP_GPU_BIT), 215 [THROTTLER_TEMP_MEM_BIT] = (SMU_THROTTLER_TEMP_MEM_BIT), 216 [THROTTLER_TEMP_VR_GFX_BIT] = (SMU_THROTTLER_TEMP_VR_GFX_BIT), 217 [THROTTLER_TEMP_VR_SOC_BIT] = (SMU_THROTTLER_TEMP_VR_SOC_BIT), 218 [THROTTLER_TEMP_VR_MEM_BIT] = (SMU_THROTTLER_TEMP_VR_MEM0_BIT), 219 [THROTTLER_APCC_BIT] = (SMU_THROTTLER_APCC_BIT), 220 }; 221 222 static int aldebaran_tables_init(struct smu_context *smu) 223 { 224 struct smu_table_context *smu_table = &smu->smu_table; 225 struct smu_table *tables = smu_table->tables; 226 227 SMU_TABLE_INIT(tables, SMU_TABLE_PPTABLE, sizeof(PPTable_t), 228 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 229 230 SMU_TABLE_INIT(tables, SMU_TABLE_PMSTATUSLOG, SMU13_TOOL_SIZE, 231 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 232 233 SMU_TABLE_INIT(tables, SMU_TABLE_SMU_METRICS, sizeof(SmuMetrics_t), 234 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 235 236 SMU_TABLE_INIT(tables, SMU_TABLE_I2C_COMMANDS, sizeof(SwI2cRequest_t), 237 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 238 239 SMU_TABLE_INIT(tables, SMU_TABLE_ECCINFO, sizeof(EccInfoTable_t), 240 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 241 242 smu_table->metrics_table = kzalloc(sizeof(SmuMetrics_t), GFP_KERNEL); 243 if (!smu_table->metrics_table) 244 return -ENOMEM; 245 smu_table->metrics_time = 0; 246 247 smu_table->gpu_metrics_table_size = sizeof(struct gpu_metrics_v1_3); 248 smu_table->gpu_metrics_table = kzalloc(smu_table->gpu_metrics_table_size, GFP_KERNEL); 249 if (!smu_table->gpu_metrics_table) { 250 kfree(smu_table->metrics_table); 251 return -ENOMEM; 252 } 253 254 smu_table->ecc_table = kzalloc(tables[SMU_TABLE_ECCINFO].size, GFP_KERNEL); 255 if (!smu_table->ecc_table) 256 return -ENOMEM; 257 258 return 0; 259 } 260 261 static int aldebaran_allocate_dpm_context(struct smu_context *smu) 262 { 263 struct smu_dpm_context *smu_dpm = &smu->smu_dpm; 264 265 smu_dpm->dpm_context = kzalloc(sizeof(struct smu_13_0_dpm_context), 266 GFP_KERNEL); 267 if (!smu_dpm->dpm_context) 268 return -ENOMEM; 269 smu_dpm->dpm_context_size = sizeof(struct smu_13_0_dpm_context); 270 271 return 0; 272 } 273 274 static int aldebaran_init_smc_tables(struct smu_context *smu) 275 { 276 int ret = 0; 277 278 ret = aldebaran_tables_init(smu); 279 if (ret) 280 return ret; 281 282 ret = aldebaran_allocate_dpm_context(smu); 283 if (ret) 284 return ret; 285 286 return smu_v13_0_init_smc_tables(smu); 287 } 288 289 static int aldebaran_get_allowed_feature_mask(struct smu_context *smu, 290 uint32_t *feature_mask, uint32_t num) 291 { 292 if (num > 2) 293 return -EINVAL; 294 295 /* pptable will handle the features to enable */ 296 memset(feature_mask, 0xFF, sizeof(uint32_t) * num); 297 298 return 0; 299 } 300 301 static int aldebaran_set_default_dpm_table(struct smu_context *smu) 302 { 303 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context; 304 struct smu_13_0_dpm_table *dpm_table = NULL; 305 PPTable_t *pptable = smu->smu_table.driver_pptable; 306 int ret = 0; 307 308 /* socclk dpm table setup */ 309 dpm_table = &dpm_context->dpm_tables.soc_table; 310 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT)) { 311 ret = smu_v13_0_set_single_dpm_table(smu, 312 SMU_SOCCLK, 313 dpm_table); 314 if (ret) 315 return ret; 316 } else { 317 dpm_table->count = 1; 318 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.socclk / 100; 319 dpm_table->dpm_levels[0].enabled = true; 320 dpm_table->min = dpm_table->dpm_levels[0].value; 321 dpm_table->max = dpm_table->dpm_levels[0].value; 322 } 323 324 /* gfxclk dpm table setup */ 325 dpm_table = &dpm_context->dpm_tables.gfx_table; 326 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT)) { 327 /* in the case of gfxclk, only fine-grained dpm is honored */ 328 dpm_table->count = 2; 329 dpm_table->dpm_levels[0].value = pptable->GfxclkFmin; 330 dpm_table->dpm_levels[0].enabled = true; 331 dpm_table->dpm_levels[1].value = pptable->GfxclkFmax; 332 dpm_table->dpm_levels[1].enabled = true; 333 dpm_table->min = dpm_table->dpm_levels[0].value; 334 dpm_table->max = dpm_table->dpm_levels[1].value; 335 } else { 336 dpm_table->count = 1; 337 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.gfxclk / 100; 338 dpm_table->dpm_levels[0].enabled = true; 339 dpm_table->min = dpm_table->dpm_levels[0].value; 340 dpm_table->max = dpm_table->dpm_levels[0].value; 341 } 342 343 /* memclk dpm table setup */ 344 dpm_table = &dpm_context->dpm_tables.uclk_table; 345 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT)) { 346 ret = smu_v13_0_set_single_dpm_table(smu, 347 SMU_UCLK, 348 dpm_table); 349 if (ret) 350 return ret; 351 } else { 352 dpm_table->count = 1; 353 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.uclk / 100; 354 dpm_table->dpm_levels[0].enabled = true; 355 dpm_table->min = dpm_table->dpm_levels[0].value; 356 dpm_table->max = dpm_table->dpm_levels[0].value; 357 } 358 359 /* fclk dpm table setup */ 360 dpm_table = &dpm_context->dpm_tables.fclk_table; 361 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_FCLK_BIT)) { 362 ret = smu_v13_0_set_single_dpm_table(smu, 363 SMU_FCLK, 364 dpm_table); 365 if (ret) 366 return ret; 367 } else { 368 dpm_table->count = 1; 369 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.fclk / 100; 370 dpm_table->dpm_levels[0].enabled = true; 371 dpm_table->min = dpm_table->dpm_levels[0].value; 372 dpm_table->max = dpm_table->dpm_levels[0].value; 373 } 374 375 return 0; 376 } 377 378 static int aldebaran_check_powerplay_table(struct smu_context *smu) 379 { 380 struct smu_table_context *table_context = &smu->smu_table; 381 struct smu_13_0_powerplay_table *powerplay_table = 382 table_context->power_play_table; 383 384 table_context->thermal_controller_type = 385 powerplay_table->thermal_controller_type; 386 387 return 0; 388 } 389 390 static int aldebaran_store_powerplay_table(struct smu_context *smu) 391 { 392 struct smu_table_context *table_context = &smu->smu_table; 393 struct smu_13_0_powerplay_table *powerplay_table = 394 table_context->power_play_table; 395 memcpy(table_context->driver_pptable, &powerplay_table->smc_pptable, 396 sizeof(PPTable_t)); 397 398 return 0; 399 } 400 401 static int aldebaran_append_powerplay_table(struct smu_context *smu) 402 { 403 struct smu_table_context *table_context = &smu->smu_table; 404 PPTable_t *smc_pptable = table_context->driver_pptable; 405 struct atom_smc_dpm_info_v4_10 *smc_dpm_table; 406 int index, ret; 407 408 index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1, 409 smc_dpm_info); 410 411 ret = amdgpu_atombios_get_data_table(smu->adev, index, NULL, NULL, NULL, 412 (uint8_t **)&smc_dpm_table); 413 if (ret) 414 return ret; 415 416 dev_info(smu->adev->dev, "smc_dpm_info table revision(format.content): %d.%d\n", 417 smc_dpm_table->table_header.format_revision, 418 smc_dpm_table->table_header.content_revision); 419 420 if ((smc_dpm_table->table_header.format_revision == 4) && 421 (smc_dpm_table->table_header.content_revision == 10)) 422 smu_memcpy_trailing(smc_pptable, GfxMaxCurrent, reserved, 423 smc_dpm_table, GfxMaxCurrent); 424 return 0; 425 } 426 427 static int aldebaran_setup_pptable(struct smu_context *smu) 428 { 429 int ret = 0; 430 431 /* VBIOS pptable is the first choice */ 432 smu->smu_table.boot_values.pp_table_id = 0; 433 434 ret = smu_v13_0_setup_pptable(smu); 435 if (ret) 436 return ret; 437 438 ret = aldebaran_store_powerplay_table(smu); 439 if (ret) 440 return ret; 441 442 ret = aldebaran_append_powerplay_table(smu); 443 if (ret) 444 return ret; 445 446 ret = aldebaran_check_powerplay_table(smu); 447 if (ret) 448 return ret; 449 450 return ret; 451 } 452 453 static bool aldebaran_is_primary(struct smu_context *smu) 454 { 455 struct amdgpu_device *adev = smu->adev; 456 457 if (adev->smuio.funcs && adev->smuio.funcs->get_die_id) 458 return adev->smuio.funcs->get_die_id(adev) == 0; 459 460 return true; 461 } 462 463 static int aldebaran_run_board_btc(struct smu_context *smu) 464 { 465 u32 smu_version; 466 int ret; 467 468 if (!aldebaran_is_primary(smu)) 469 return 0; 470 471 ret = smu_cmn_get_smc_version(smu, NULL, &smu_version); 472 if (ret) { 473 dev_err(smu->adev->dev, "Failed to get smu version!\n"); 474 return ret; 475 } 476 if (smu_version <= 0x00441d00) 477 return 0; 478 479 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_BoardPowerCalibration, NULL); 480 if (ret) 481 dev_err(smu->adev->dev, "Board power calibration failed!\n"); 482 483 return ret; 484 } 485 486 static int aldebaran_run_btc(struct smu_context *smu) 487 { 488 int ret; 489 490 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_RunDcBtc, NULL); 491 if (ret) 492 dev_err(smu->adev->dev, "RunDcBtc failed!\n"); 493 else 494 ret = aldebaran_run_board_btc(smu); 495 496 return ret; 497 } 498 499 static int aldebaran_populate_umd_state_clk(struct smu_context *smu) 500 { 501 struct smu_13_0_dpm_context *dpm_context = 502 smu->smu_dpm.dpm_context; 503 struct smu_13_0_dpm_table *gfx_table = 504 &dpm_context->dpm_tables.gfx_table; 505 struct smu_13_0_dpm_table *mem_table = 506 &dpm_context->dpm_tables.uclk_table; 507 struct smu_13_0_dpm_table *soc_table = 508 &dpm_context->dpm_tables.soc_table; 509 struct smu_umd_pstate_table *pstate_table = 510 &smu->pstate_table; 511 512 pstate_table->gfxclk_pstate.min = gfx_table->min; 513 pstate_table->gfxclk_pstate.peak = gfx_table->max; 514 pstate_table->gfxclk_pstate.curr.min = gfx_table->min; 515 pstate_table->gfxclk_pstate.curr.max = gfx_table->max; 516 517 pstate_table->uclk_pstate.min = mem_table->min; 518 pstate_table->uclk_pstate.peak = mem_table->max; 519 pstate_table->uclk_pstate.curr.min = mem_table->min; 520 pstate_table->uclk_pstate.curr.max = mem_table->max; 521 522 pstate_table->socclk_pstate.min = soc_table->min; 523 pstate_table->socclk_pstate.peak = soc_table->max; 524 pstate_table->socclk_pstate.curr.min = soc_table->min; 525 pstate_table->socclk_pstate.curr.max = soc_table->max; 526 527 if (gfx_table->count > ALDEBARAN_UMD_PSTATE_GFXCLK_LEVEL && 528 mem_table->count > ALDEBARAN_UMD_PSTATE_MCLK_LEVEL && 529 soc_table->count > ALDEBARAN_UMD_PSTATE_SOCCLK_LEVEL) { 530 pstate_table->gfxclk_pstate.standard = 531 gfx_table->dpm_levels[ALDEBARAN_UMD_PSTATE_GFXCLK_LEVEL].value; 532 pstate_table->uclk_pstate.standard = 533 mem_table->dpm_levels[ALDEBARAN_UMD_PSTATE_MCLK_LEVEL].value; 534 pstate_table->socclk_pstate.standard = 535 soc_table->dpm_levels[ALDEBARAN_UMD_PSTATE_SOCCLK_LEVEL].value; 536 } else { 537 pstate_table->gfxclk_pstate.standard = 538 pstate_table->gfxclk_pstate.min; 539 pstate_table->uclk_pstate.standard = 540 pstate_table->uclk_pstate.min; 541 pstate_table->socclk_pstate.standard = 542 pstate_table->socclk_pstate.min; 543 } 544 545 return 0; 546 } 547 548 static int aldebaran_get_clk_table(struct smu_context *smu, 549 struct pp_clock_levels_with_latency *clocks, 550 struct smu_13_0_dpm_table *dpm_table) 551 { 552 int i, count; 553 554 count = (dpm_table->count > MAX_NUM_CLOCKS) ? MAX_NUM_CLOCKS : dpm_table->count; 555 clocks->num_levels = count; 556 557 for (i = 0; i < count; i++) { 558 clocks->data[i].clocks_in_khz = 559 dpm_table->dpm_levels[i].value * 1000; 560 clocks->data[i].latency_in_us = 0; 561 } 562 563 return 0; 564 } 565 566 static int aldebaran_freqs_in_same_level(int32_t frequency1, 567 int32_t frequency2) 568 { 569 return (abs(frequency1 - frequency2) <= EPSILON); 570 } 571 572 static int aldebaran_get_smu_metrics_data(struct smu_context *smu, 573 MetricsMember_t member, 574 uint32_t *value) 575 { 576 struct smu_table_context *smu_table= &smu->smu_table; 577 SmuMetrics_t *metrics = (SmuMetrics_t *)smu_table->metrics_table; 578 int ret = 0; 579 580 ret = smu_cmn_get_metrics_table(smu, 581 NULL, 582 false); 583 if (ret) 584 return ret; 585 586 switch (member) { 587 case METRICS_CURR_GFXCLK: 588 *value = metrics->CurrClock[PPCLK_GFXCLK]; 589 break; 590 case METRICS_CURR_SOCCLK: 591 *value = metrics->CurrClock[PPCLK_SOCCLK]; 592 break; 593 case METRICS_CURR_UCLK: 594 *value = metrics->CurrClock[PPCLK_UCLK]; 595 break; 596 case METRICS_CURR_VCLK: 597 *value = metrics->CurrClock[PPCLK_VCLK]; 598 break; 599 case METRICS_CURR_DCLK: 600 *value = metrics->CurrClock[PPCLK_DCLK]; 601 break; 602 case METRICS_CURR_FCLK: 603 *value = metrics->CurrClock[PPCLK_FCLK]; 604 break; 605 case METRICS_AVERAGE_GFXCLK: 606 *value = metrics->AverageGfxclkFrequency; 607 break; 608 case METRICS_AVERAGE_SOCCLK: 609 *value = metrics->AverageSocclkFrequency; 610 break; 611 case METRICS_AVERAGE_UCLK: 612 *value = metrics->AverageUclkFrequency; 613 break; 614 case METRICS_AVERAGE_GFXACTIVITY: 615 *value = metrics->AverageGfxActivity; 616 break; 617 case METRICS_AVERAGE_MEMACTIVITY: 618 *value = metrics->AverageUclkActivity; 619 break; 620 case METRICS_AVERAGE_SOCKETPOWER: 621 /* Valid power data is available only from primary die */ 622 *value = aldebaran_is_primary(smu) ? 623 metrics->AverageSocketPower << 8 : 624 0; 625 break; 626 case METRICS_TEMPERATURE_EDGE: 627 *value = metrics->TemperatureEdge * 628 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 629 break; 630 case METRICS_TEMPERATURE_HOTSPOT: 631 *value = metrics->TemperatureHotspot * 632 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 633 break; 634 case METRICS_TEMPERATURE_MEM: 635 *value = metrics->TemperatureHBM * 636 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 637 break; 638 case METRICS_TEMPERATURE_VRGFX: 639 *value = metrics->TemperatureVrGfx * 640 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 641 break; 642 case METRICS_TEMPERATURE_VRSOC: 643 *value = metrics->TemperatureVrSoc * 644 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 645 break; 646 case METRICS_TEMPERATURE_VRMEM: 647 *value = metrics->TemperatureVrMem * 648 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 649 break; 650 case METRICS_THROTTLER_STATUS: 651 *value = metrics->ThrottlerStatus; 652 break; 653 case METRICS_UNIQUE_ID_UPPER32: 654 *value = metrics->PublicSerialNumUpper32; 655 break; 656 case METRICS_UNIQUE_ID_LOWER32: 657 *value = metrics->PublicSerialNumLower32; 658 break; 659 default: 660 *value = UINT_MAX; 661 break; 662 } 663 664 return ret; 665 } 666 667 static int aldebaran_get_current_clk_freq_by_table(struct smu_context *smu, 668 enum smu_clk_type clk_type, 669 uint32_t *value) 670 { 671 MetricsMember_t member_type; 672 int clk_id = 0; 673 674 if (!value) 675 return -EINVAL; 676 677 clk_id = smu_cmn_to_asic_specific_index(smu, 678 CMN2ASIC_MAPPING_CLK, 679 clk_type); 680 if (clk_id < 0) 681 return -EINVAL; 682 683 switch (clk_id) { 684 case PPCLK_GFXCLK: 685 /* 686 * CurrClock[clk_id] can provide accurate 687 * output only when the dpm feature is enabled. 688 * We can use Average_* for dpm disabled case. 689 * But this is available for gfxclk/uclk/socclk/vclk/dclk. 690 */ 691 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT)) 692 member_type = METRICS_CURR_GFXCLK; 693 else 694 member_type = METRICS_AVERAGE_GFXCLK; 695 break; 696 case PPCLK_UCLK: 697 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT)) 698 member_type = METRICS_CURR_UCLK; 699 else 700 member_type = METRICS_AVERAGE_UCLK; 701 break; 702 case PPCLK_SOCCLK: 703 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT)) 704 member_type = METRICS_CURR_SOCCLK; 705 else 706 member_type = METRICS_AVERAGE_SOCCLK; 707 break; 708 case PPCLK_VCLK: 709 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) 710 member_type = METRICS_CURR_VCLK; 711 else 712 member_type = METRICS_AVERAGE_VCLK; 713 break; 714 case PPCLK_DCLK: 715 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) 716 member_type = METRICS_CURR_DCLK; 717 else 718 member_type = METRICS_AVERAGE_DCLK; 719 break; 720 case PPCLK_FCLK: 721 member_type = METRICS_CURR_FCLK; 722 break; 723 default: 724 return -EINVAL; 725 } 726 727 return aldebaran_get_smu_metrics_data(smu, 728 member_type, 729 value); 730 } 731 732 static int aldebaran_print_clk_levels(struct smu_context *smu, 733 enum smu_clk_type type, char *buf) 734 { 735 int i, now, size = 0; 736 int ret = 0; 737 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 738 struct pp_clock_levels_with_latency clocks; 739 struct smu_13_0_dpm_table *single_dpm_table; 740 struct smu_dpm_context *smu_dpm = &smu->smu_dpm; 741 struct smu_13_0_dpm_context *dpm_context = NULL; 742 uint32_t display_levels; 743 uint32_t freq_values[3] = {0}; 744 uint32_t min_clk, max_clk; 745 746 smu_cmn_get_sysfs_buf(&buf, &size); 747 748 if (amdgpu_ras_intr_triggered()) { 749 size += sysfs_emit_at(buf, size, "unavailable\n"); 750 return size; 751 } 752 753 dpm_context = smu_dpm->dpm_context; 754 755 switch (type) { 756 757 case SMU_OD_SCLK: 758 size += sysfs_emit_at(buf, size, "%s:\n", "GFXCLK"); 759 fallthrough; 760 case SMU_SCLK: 761 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_GFXCLK, &now); 762 if (ret) { 763 dev_err(smu->adev->dev, "Attempt to get current gfx clk Failed!"); 764 return ret; 765 } 766 767 single_dpm_table = &(dpm_context->dpm_tables.gfx_table); 768 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 769 if (ret) { 770 dev_err(smu->adev->dev, "Attempt to get gfx clk levels Failed!"); 771 return ret; 772 } 773 774 display_levels = clocks.num_levels; 775 776 min_clk = pstate_table->gfxclk_pstate.curr.min; 777 max_clk = pstate_table->gfxclk_pstate.curr.max; 778 779 freq_values[0] = min_clk; 780 freq_values[1] = max_clk; 781 782 /* fine-grained dpm has only 2 levels */ 783 if (now > min_clk && now < max_clk) { 784 display_levels = clocks.num_levels + 1; 785 freq_values[2] = max_clk; 786 freq_values[1] = now; 787 } 788 789 /* 790 * For DPM disabled case, there will be only one clock level. 791 * And it's safe to assume that is always the current clock. 792 */ 793 if (display_levels == clocks.num_levels) { 794 for (i = 0; i < clocks.num_levels; i++) 795 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", i, 796 freq_values[i], 797 (clocks.num_levels == 1) ? 798 "*" : 799 (aldebaran_freqs_in_same_level( 800 freq_values[i], now) ? 801 "*" : 802 "")); 803 } else { 804 for (i = 0; i < display_levels; i++) 805 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", i, 806 freq_values[i], i == 1 ? "*" : ""); 807 } 808 809 break; 810 811 case SMU_OD_MCLK: 812 size += sysfs_emit_at(buf, size, "%s:\n", "MCLK"); 813 fallthrough; 814 case SMU_MCLK: 815 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_UCLK, &now); 816 if (ret) { 817 dev_err(smu->adev->dev, "Attempt to get current mclk Failed!"); 818 return ret; 819 } 820 821 single_dpm_table = &(dpm_context->dpm_tables.uclk_table); 822 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 823 if (ret) { 824 dev_err(smu->adev->dev, "Attempt to get memory clk levels Failed!"); 825 return ret; 826 } 827 828 for (i = 0; i < clocks.num_levels; i++) 829 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 830 i, clocks.data[i].clocks_in_khz / 1000, 831 (clocks.num_levels == 1) ? "*" : 832 (aldebaran_freqs_in_same_level( 833 clocks.data[i].clocks_in_khz / 1000, 834 now) ? "*" : "")); 835 break; 836 837 case SMU_SOCCLK: 838 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_SOCCLK, &now); 839 if (ret) { 840 dev_err(smu->adev->dev, "Attempt to get current socclk Failed!"); 841 return ret; 842 } 843 844 single_dpm_table = &(dpm_context->dpm_tables.soc_table); 845 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 846 if (ret) { 847 dev_err(smu->adev->dev, "Attempt to get socclk levels Failed!"); 848 return ret; 849 } 850 851 for (i = 0; i < clocks.num_levels; i++) 852 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 853 i, clocks.data[i].clocks_in_khz / 1000, 854 (clocks.num_levels == 1) ? "*" : 855 (aldebaran_freqs_in_same_level( 856 clocks.data[i].clocks_in_khz / 1000, 857 now) ? "*" : "")); 858 break; 859 860 case SMU_FCLK: 861 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_FCLK, &now); 862 if (ret) { 863 dev_err(smu->adev->dev, "Attempt to get current fclk Failed!"); 864 return ret; 865 } 866 867 single_dpm_table = &(dpm_context->dpm_tables.fclk_table); 868 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 869 if (ret) { 870 dev_err(smu->adev->dev, "Attempt to get fclk levels Failed!"); 871 return ret; 872 } 873 874 for (i = 0; i < single_dpm_table->count; i++) 875 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 876 i, single_dpm_table->dpm_levels[i].value, 877 (clocks.num_levels == 1) ? "*" : 878 (aldebaran_freqs_in_same_level( 879 clocks.data[i].clocks_in_khz / 1000, 880 now) ? "*" : "")); 881 break; 882 883 case SMU_VCLK: 884 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_VCLK, &now); 885 if (ret) { 886 dev_err(smu->adev->dev, "Attempt to get current vclk Failed!"); 887 return ret; 888 } 889 890 single_dpm_table = &(dpm_context->dpm_tables.vclk_table); 891 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 892 if (ret) { 893 dev_err(smu->adev->dev, "Attempt to get vclk levels Failed!"); 894 return ret; 895 } 896 897 for (i = 0; i < single_dpm_table->count; i++) 898 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 899 i, single_dpm_table->dpm_levels[i].value, 900 (clocks.num_levels == 1) ? "*" : 901 (aldebaran_freqs_in_same_level( 902 clocks.data[i].clocks_in_khz / 1000, 903 now) ? "*" : "")); 904 break; 905 906 case SMU_DCLK: 907 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_DCLK, &now); 908 if (ret) { 909 dev_err(smu->adev->dev, "Attempt to get current dclk Failed!"); 910 return ret; 911 } 912 913 single_dpm_table = &(dpm_context->dpm_tables.dclk_table); 914 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 915 if (ret) { 916 dev_err(smu->adev->dev, "Attempt to get dclk levels Failed!"); 917 return ret; 918 } 919 920 for (i = 0; i < single_dpm_table->count; i++) 921 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 922 i, single_dpm_table->dpm_levels[i].value, 923 (clocks.num_levels == 1) ? "*" : 924 (aldebaran_freqs_in_same_level( 925 clocks.data[i].clocks_in_khz / 1000, 926 now) ? "*" : "")); 927 break; 928 929 default: 930 break; 931 } 932 933 return size; 934 } 935 936 static int aldebaran_upload_dpm_level(struct smu_context *smu, 937 bool max, 938 uint32_t feature_mask, 939 uint32_t level) 940 { 941 struct smu_13_0_dpm_context *dpm_context = 942 smu->smu_dpm.dpm_context; 943 uint32_t freq; 944 int ret = 0; 945 946 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT) && 947 (feature_mask & FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT))) { 948 freq = dpm_context->dpm_tables.gfx_table.dpm_levels[level].value; 949 ret = smu_cmn_send_smc_msg_with_param(smu, 950 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 951 (PPCLK_GFXCLK << 16) | (freq & 0xffff), 952 NULL); 953 if (ret) { 954 dev_err(smu->adev->dev, "Failed to set soft %s gfxclk !\n", 955 max ? "max" : "min"); 956 return ret; 957 } 958 } 959 960 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT) && 961 (feature_mask & FEATURE_MASK(FEATURE_DPM_UCLK_BIT))) { 962 freq = dpm_context->dpm_tables.uclk_table.dpm_levels[level].value; 963 ret = smu_cmn_send_smc_msg_with_param(smu, 964 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 965 (PPCLK_UCLK << 16) | (freq & 0xffff), 966 NULL); 967 if (ret) { 968 dev_err(smu->adev->dev, "Failed to set soft %s memclk !\n", 969 max ? "max" : "min"); 970 return ret; 971 } 972 } 973 974 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT) && 975 (feature_mask & FEATURE_MASK(FEATURE_DPM_SOCCLK_BIT))) { 976 freq = dpm_context->dpm_tables.soc_table.dpm_levels[level].value; 977 ret = smu_cmn_send_smc_msg_with_param(smu, 978 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 979 (PPCLK_SOCCLK << 16) | (freq & 0xffff), 980 NULL); 981 if (ret) { 982 dev_err(smu->adev->dev, "Failed to set soft %s socclk !\n", 983 max ? "max" : "min"); 984 return ret; 985 } 986 } 987 988 return ret; 989 } 990 991 static int aldebaran_force_clk_levels(struct smu_context *smu, 992 enum smu_clk_type type, uint32_t mask) 993 { 994 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context; 995 struct smu_13_0_dpm_table *single_dpm_table = NULL; 996 uint32_t soft_min_level, soft_max_level; 997 int ret = 0; 998 999 soft_min_level = mask ? (ffs(mask) - 1) : 0; 1000 soft_max_level = mask ? (fls(mask) - 1) : 0; 1001 1002 switch (type) { 1003 case SMU_SCLK: 1004 single_dpm_table = &(dpm_context->dpm_tables.gfx_table); 1005 if (soft_max_level >= single_dpm_table->count) { 1006 dev_err(smu->adev->dev, "Clock level specified %d is over max allowed %d\n", 1007 soft_max_level, single_dpm_table->count - 1); 1008 ret = -EINVAL; 1009 break; 1010 } 1011 1012 ret = aldebaran_upload_dpm_level(smu, 1013 false, 1014 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT), 1015 soft_min_level); 1016 if (ret) { 1017 dev_err(smu->adev->dev, "Failed to upload boot level to lowest!\n"); 1018 break; 1019 } 1020 1021 ret = aldebaran_upload_dpm_level(smu, 1022 true, 1023 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT), 1024 soft_max_level); 1025 if (ret) 1026 dev_err(smu->adev->dev, "Failed to upload dpm max level to highest!\n"); 1027 1028 break; 1029 1030 case SMU_MCLK: 1031 case SMU_SOCCLK: 1032 case SMU_FCLK: 1033 /* 1034 * Should not arrive here since aldebaran does not 1035 * support mclk/socclk/fclk softmin/softmax settings 1036 */ 1037 ret = -EINVAL; 1038 break; 1039 1040 default: 1041 break; 1042 } 1043 1044 return ret; 1045 } 1046 1047 static int aldebaran_get_thermal_temperature_range(struct smu_context *smu, 1048 struct smu_temperature_range *range) 1049 { 1050 struct smu_table_context *table_context = &smu->smu_table; 1051 struct smu_13_0_powerplay_table *powerplay_table = 1052 table_context->power_play_table; 1053 PPTable_t *pptable = smu->smu_table.driver_pptable; 1054 1055 if (!range) 1056 return -EINVAL; 1057 1058 memcpy(range, &smu13_thermal_policy[0], sizeof(struct smu_temperature_range)); 1059 1060 range->hotspot_crit_max = pptable->ThotspotLimit * 1061 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1062 range->hotspot_emergency_max = (pptable->ThotspotLimit + CTF_OFFSET_HOTSPOT) * 1063 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1064 range->mem_crit_max = pptable->TmemLimit * 1065 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1066 range->mem_emergency_max = (pptable->TmemLimit + CTF_OFFSET_MEM)* 1067 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1068 range->software_shutdown_temp = powerplay_table->software_shutdown_temp; 1069 1070 return 0; 1071 } 1072 1073 static int aldebaran_get_current_activity_percent(struct smu_context *smu, 1074 enum amd_pp_sensors sensor, 1075 uint32_t *value) 1076 { 1077 int ret = 0; 1078 1079 if (!value) 1080 return -EINVAL; 1081 1082 switch (sensor) { 1083 case AMDGPU_PP_SENSOR_GPU_LOAD: 1084 ret = aldebaran_get_smu_metrics_data(smu, 1085 METRICS_AVERAGE_GFXACTIVITY, 1086 value); 1087 break; 1088 case AMDGPU_PP_SENSOR_MEM_LOAD: 1089 ret = aldebaran_get_smu_metrics_data(smu, 1090 METRICS_AVERAGE_MEMACTIVITY, 1091 value); 1092 break; 1093 default: 1094 dev_err(smu->adev->dev, "Invalid sensor for retrieving clock activity\n"); 1095 return -EINVAL; 1096 } 1097 1098 return ret; 1099 } 1100 1101 static int aldebaran_get_gpu_power(struct smu_context *smu, uint32_t *value) 1102 { 1103 if (!value) 1104 return -EINVAL; 1105 1106 return aldebaran_get_smu_metrics_data(smu, 1107 METRICS_AVERAGE_SOCKETPOWER, 1108 value); 1109 } 1110 1111 static int aldebaran_thermal_get_temperature(struct smu_context *smu, 1112 enum amd_pp_sensors sensor, 1113 uint32_t *value) 1114 { 1115 int ret = 0; 1116 1117 if (!value) 1118 return -EINVAL; 1119 1120 switch (sensor) { 1121 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP: 1122 ret = aldebaran_get_smu_metrics_data(smu, 1123 METRICS_TEMPERATURE_HOTSPOT, 1124 value); 1125 break; 1126 case AMDGPU_PP_SENSOR_EDGE_TEMP: 1127 ret = aldebaran_get_smu_metrics_data(smu, 1128 METRICS_TEMPERATURE_EDGE, 1129 value); 1130 break; 1131 case AMDGPU_PP_SENSOR_MEM_TEMP: 1132 ret = aldebaran_get_smu_metrics_data(smu, 1133 METRICS_TEMPERATURE_MEM, 1134 value); 1135 break; 1136 default: 1137 dev_err(smu->adev->dev, "Invalid sensor for retrieving temp\n"); 1138 return -EINVAL; 1139 } 1140 1141 return ret; 1142 } 1143 1144 static int aldebaran_read_sensor(struct smu_context *smu, 1145 enum amd_pp_sensors sensor, 1146 void *data, uint32_t *size) 1147 { 1148 int ret = 0; 1149 1150 if (amdgpu_ras_intr_triggered()) 1151 return 0; 1152 1153 if (!data || !size) 1154 return -EINVAL; 1155 1156 switch (sensor) { 1157 case AMDGPU_PP_SENSOR_MEM_LOAD: 1158 case AMDGPU_PP_SENSOR_GPU_LOAD: 1159 ret = aldebaran_get_current_activity_percent(smu, 1160 sensor, 1161 (uint32_t *)data); 1162 *size = 4; 1163 break; 1164 case AMDGPU_PP_SENSOR_GPU_POWER: 1165 ret = aldebaran_get_gpu_power(smu, (uint32_t *)data); 1166 *size = 4; 1167 break; 1168 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP: 1169 case AMDGPU_PP_SENSOR_EDGE_TEMP: 1170 case AMDGPU_PP_SENSOR_MEM_TEMP: 1171 ret = aldebaran_thermal_get_temperature(smu, sensor, 1172 (uint32_t *)data); 1173 *size = 4; 1174 break; 1175 case AMDGPU_PP_SENSOR_GFX_MCLK: 1176 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_UCLK, (uint32_t *)data); 1177 /* the output clock frequency in 10K unit */ 1178 *(uint32_t *)data *= 100; 1179 *size = 4; 1180 break; 1181 case AMDGPU_PP_SENSOR_GFX_SCLK: 1182 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_GFXCLK, (uint32_t *)data); 1183 *(uint32_t *)data *= 100; 1184 *size = 4; 1185 break; 1186 case AMDGPU_PP_SENSOR_VDDGFX: 1187 ret = smu_v13_0_get_gfx_vdd(smu, (uint32_t *)data); 1188 *size = 4; 1189 break; 1190 default: 1191 ret = -EOPNOTSUPP; 1192 break; 1193 } 1194 1195 return ret; 1196 } 1197 1198 static int aldebaran_get_power_limit(struct smu_context *smu, 1199 uint32_t *current_power_limit, 1200 uint32_t *default_power_limit, 1201 uint32_t *max_power_limit) 1202 { 1203 PPTable_t *pptable = smu->smu_table.driver_pptable; 1204 uint32_t power_limit = 0; 1205 int ret; 1206 1207 if (!smu_cmn_feature_is_enabled(smu, SMU_FEATURE_PPT_BIT)) { 1208 if (current_power_limit) 1209 *current_power_limit = 0; 1210 if (default_power_limit) 1211 *default_power_limit = 0; 1212 if (max_power_limit) 1213 *max_power_limit = 0; 1214 1215 dev_warn(smu->adev->dev, 1216 "PPT feature is not enabled, power values can't be fetched."); 1217 1218 return 0; 1219 } 1220 1221 /* Valid power data is available only from primary die. 1222 * For secondary die show the value as 0. 1223 */ 1224 if (aldebaran_is_primary(smu)) { 1225 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_GetPptLimit, 1226 &power_limit); 1227 1228 if (ret) { 1229 /* the last hope to figure out the ppt limit */ 1230 if (!pptable) { 1231 dev_err(smu->adev->dev, 1232 "Cannot get PPT limit due to pptable missing!"); 1233 return -EINVAL; 1234 } 1235 power_limit = pptable->PptLimit; 1236 } 1237 } 1238 1239 if (current_power_limit) 1240 *current_power_limit = power_limit; 1241 if (default_power_limit) 1242 *default_power_limit = power_limit; 1243 1244 if (max_power_limit) { 1245 if (pptable) 1246 *max_power_limit = pptable->PptLimit; 1247 } 1248 1249 return 0; 1250 } 1251 1252 static int aldebaran_set_power_limit(struct smu_context *smu, 1253 enum smu_ppt_limit_type limit_type, 1254 uint32_t limit) 1255 { 1256 /* Power limit can be set only through primary die */ 1257 if (aldebaran_is_primary(smu)) 1258 return smu_v13_0_set_power_limit(smu, limit_type, limit); 1259 1260 return -EINVAL; 1261 } 1262 1263 static int aldebaran_system_features_control(struct smu_context *smu, bool enable) 1264 { 1265 int ret; 1266 1267 ret = smu_v13_0_system_features_control(smu, enable); 1268 if (!ret && enable) 1269 ret = aldebaran_run_btc(smu); 1270 1271 return ret; 1272 } 1273 1274 static int aldebaran_set_performance_level(struct smu_context *smu, 1275 enum amd_dpm_forced_level level) 1276 { 1277 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1278 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1279 struct smu_13_0_dpm_table *gfx_table = 1280 &dpm_context->dpm_tables.gfx_table; 1281 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1282 1283 /* Disable determinism if switching to another mode */ 1284 if ((smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) && 1285 (level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) { 1286 smu_cmn_send_smc_msg(smu, SMU_MSG_DisableDeterminism, NULL); 1287 pstate_table->gfxclk_pstate.curr.max = gfx_table->max; 1288 } 1289 1290 switch (level) { 1291 1292 case AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM: 1293 return 0; 1294 1295 case AMD_DPM_FORCED_LEVEL_HIGH: 1296 case AMD_DPM_FORCED_LEVEL_LOW: 1297 case AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD: 1298 case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK: 1299 case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK: 1300 case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK: 1301 default: 1302 break; 1303 } 1304 1305 return smu_v13_0_set_performance_level(smu, level); 1306 } 1307 1308 static int aldebaran_set_soft_freq_limited_range(struct smu_context *smu, 1309 enum smu_clk_type clk_type, 1310 uint32_t min, 1311 uint32_t max) 1312 { 1313 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1314 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1315 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1316 struct amdgpu_device *adev = smu->adev; 1317 uint32_t min_clk; 1318 uint32_t max_clk; 1319 int ret = 0; 1320 1321 if (clk_type != SMU_GFXCLK && clk_type != SMU_SCLK) 1322 return -EINVAL; 1323 1324 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) 1325 && (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) 1326 return -EINVAL; 1327 1328 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_MANUAL) { 1329 if (min >= max) { 1330 dev_err(smu->adev->dev, 1331 "Minimum GFX clk should be less than the maximum allowed clock\n"); 1332 return -EINVAL; 1333 } 1334 1335 if ((min == pstate_table->gfxclk_pstate.curr.min) && 1336 (max == pstate_table->gfxclk_pstate.curr.max)) 1337 return 0; 1338 1339 ret = smu_v13_0_set_soft_freq_limited_range(smu, SMU_GFXCLK, 1340 min, max); 1341 if (!ret) { 1342 pstate_table->gfxclk_pstate.curr.min = min; 1343 pstate_table->gfxclk_pstate.curr.max = max; 1344 } 1345 1346 return ret; 1347 } 1348 1349 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) { 1350 if (!max || (max < dpm_context->dpm_tables.gfx_table.min) || 1351 (max > dpm_context->dpm_tables.gfx_table.max)) { 1352 dev_warn(adev->dev, 1353 "Invalid max frequency %d MHz specified for determinism\n", max); 1354 return -EINVAL; 1355 } 1356 1357 /* Restore default min/max clocks and enable determinism */ 1358 min_clk = dpm_context->dpm_tables.gfx_table.min; 1359 max_clk = dpm_context->dpm_tables.gfx_table.max; 1360 ret = smu_v13_0_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1361 if (!ret) { 1362 usleep_range(500, 1000); 1363 ret = smu_cmn_send_smc_msg_with_param(smu, 1364 SMU_MSG_EnableDeterminism, 1365 max, NULL); 1366 if (ret) { 1367 dev_err(adev->dev, 1368 "Failed to enable determinism at GFX clock %d MHz\n", max); 1369 } else { 1370 pstate_table->gfxclk_pstate.curr.min = min_clk; 1371 pstate_table->gfxclk_pstate.curr.max = max; 1372 } 1373 } 1374 } 1375 1376 return ret; 1377 } 1378 1379 static int aldebaran_usr_edit_dpm_table(struct smu_context *smu, enum PP_OD_DPM_TABLE_COMMAND type, 1380 long input[], uint32_t size) 1381 { 1382 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1383 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1384 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1385 uint32_t min_clk; 1386 uint32_t max_clk; 1387 int ret = 0; 1388 1389 /* Only allowed in manual or determinism mode */ 1390 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) 1391 && (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) 1392 return -EINVAL; 1393 1394 switch (type) { 1395 case PP_OD_EDIT_SCLK_VDDC_TABLE: 1396 if (size != 2) { 1397 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1398 return -EINVAL; 1399 } 1400 1401 if (input[0] == 0) { 1402 if (input[1] < dpm_context->dpm_tables.gfx_table.min) { 1403 dev_warn(smu->adev->dev, "Minimum GFX clk (%ld) MHz specified is less than the minimum allowed (%d) MHz\n", 1404 input[1], dpm_context->dpm_tables.gfx_table.min); 1405 pstate_table->gfxclk_pstate.custom.min = 1406 pstate_table->gfxclk_pstate.curr.min; 1407 return -EINVAL; 1408 } 1409 1410 pstate_table->gfxclk_pstate.custom.min = input[1]; 1411 } else if (input[0] == 1) { 1412 if (input[1] > dpm_context->dpm_tables.gfx_table.max) { 1413 dev_warn(smu->adev->dev, "Maximum GFX clk (%ld) MHz specified is greater than the maximum allowed (%d) MHz\n", 1414 input[1], dpm_context->dpm_tables.gfx_table.max); 1415 pstate_table->gfxclk_pstate.custom.max = 1416 pstate_table->gfxclk_pstate.curr.max; 1417 return -EINVAL; 1418 } 1419 1420 pstate_table->gfxclk_pstate.custom.max = input[1]; 1421 } else { 1422 return -EINVAL; 1423 } 1424 break; 1425 case PP_OD_RESTORE_DEFAULT_TABLE: 1426 if (size != 0) { 1427 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1428 return -EINVAL; 1429 } else { 1430 /* Use the default frequencies for manual and determinism mode */ 1431 min_clk = dpm_context->dpm_tables.gfx_table.min; 1432 max_clk = dpm_context->dpm_tables.gfx_table.max; 1433 1434 return aldebaran_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1435 } 1436 break; 1437 case PP_OD_COMMIT_DPM_TABLE: 1438 if (size != 0) { 1439 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1440 return -EINVAL; 1441 } else { 1442 if (!pstate_table->gfxclk_pstate.custom.min) 1443 pstate_table->gfxclk_pstate.custom.min = 1444 pstate_table->gfxclk_pstate.curr.min; 1445 1446 if (!pstate_table->gfxclk_pstate.custom.max) 1447 pstate_table->gfxclk_pstate.custom.max = 1448 pstate_table->gfxclk_pstate.curr.max; 1449 1450 min_clk = pstate_table->gfxclk_pstate.custom.min; 1451 max_clk = pstate_table->gfxclk_pstate.custom.max; 1452 1453 return aldebaran_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1454 } 1455 break; 1456 default: 1457 return -ENOSYS; 1458 } 1459 1460 return ret; 1461 } 1462 1463 static bool aldebaran_is_dpm_running(struct smu_context *smu) 1464 { 1465 int ret; 1466 uint64_t feature_enabled; 1467 1468 ret = smu_cmn_get_enabled_mask(smu, &feature_enabled); 1469 if (ret) 1470 return false; 1471 return !!(feature_enabled & SMC_DPM_FEATURE); 1472 } 1473 1474 static int aldebaran_i2c_xfer(struct i2c_adapter *i2c_adap, 1475 struct i2c_msg *msg, int num_msgs) 1476 { 1477 struct amdgpu_smu_i2c_bus *smu_i2c = i2c_get_adapdata(i2c_adap); 1478 struct amdgpu_device *adev = smu_i2c->adev; 1479 struct smu_context *smu = adev->powerplay.pp_handle; 1480 struct smu_table_context *smu_table = &smu->smu_table; 1481 struct smu_table *table = &smu_table->driver_table; 1482 SwI2cRequest_t *req, *res = (SwI2cRequest_t *)table->cpu_addr; 1483 int i, j, r, c; 1484 u16 dir; 1485 1486 if (!adev->pm.dpm_enabled) 1487 return -EBUSY; 1488 1489 req = kzalloc(sizeof(*req), GFP_KERNEL); 1490 if (!req) 1491 return -ENOMEM; 1492 1493 req->I2CcontrollerPort = smu_i2c->port; 1494 req->I2CSpeed = I2C_SPEED_FAST_400K; 1495 req->SlaveAddress = msg[0].addr << 1; /* wants an 8-bit address */ 1496 dir = msg[0].flags & I2C_M_RD; 1497 1498 for (c = i = 0; i < num_msgs; i++) { 1499 for (j = 0; j < msg[i].len; j++, c++) { 1500 SwI2cCmd_t *cmd = &req->SwI2cCmds[c]; 1501 1502 if (!(msg[i].flags & I2C_M_RD)) { 1503 /* write */ 1504 cmd->CmdConfig |= CMDCONFIG_READWRITE_MASK; 1505 cmd->ReadWriteData = msg[i].buf[j]; 1506 } 1507 1508 if ((dir ^ msg[i].flags) & I2C_M_RD) { 1509 /* The direction changes. 1510 */ 1511 dir = msg[i].flags & I2C_M_RD; 1512 cmd->CmdConfig |= CMDCONFIG_RESTART_MASK; 1513 } 1514 1515 req->NumCmds++; 1516 1517 /* 1518 * Insert STOP if we are at the last byte of either last 1519 * message for the transaction or the client explicitly 1520 * requires a STOP at this particular message. 1521 */ 1522 if ((j == msg[i].len - 1) && 1523 ((i == num_msgs - 1) || (msg[i].flags & I2C_M_STOP))) { 1524 cmd->CmdConfig &= ~CMDCONFIG_RESTART_MASK; 1525 cmd->CmdConfig |= CMDCONFIG_STOP_MASK; 1526 } 1527 } 1528 } 1529 mutex_lock(&adev->pm.mutex); 1530 r = smu_cmn_update_table(smu, SMU_TABLE_I2C_COMMANDS, 0, req, true); 1531 mutex_unlock(&adev->pm.mutex); 1532 if (r) 1533 goto fail; 1534 1535 for (c = i = 0; i < num_msgs; i++) { 1536 if (!(msg[i].flags & I2C_M_RD)) { 1537 c += msg[i].len; 1538 continue; 1539 } 1540 for (j = 0; j < msg[i].len; j++, c++) { 1541 SwI2cCmd_t *cmd = &res->SwI2cCmds[c]; 1542 1543 msg[i].buf[j] = cmd->ReadWriteData; 1544 } 1545 } 1546 r = num_msgs; 1547 fail: 1548 kfree(req); 1549 return r; 1550 } 1551 1552 static u32 aldebaran_i2c_func(struct i2c_adapter *adap) 1553 { 1554 return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL; 1555 } 1556 1557 1558 static const struct i2c_algorithm aldebaran_i2c_algo = { 1559 .master_xfer = aldebaran_i2c_xfer, 1560 .functionality = aldebaran_i2c_func, 1561 }; 1562 1563 static const struct i2c_adapter_quirks aldebaran_i2c_control_quirks = { 1564 .flags = I2C_AQ_COMB | I2C_AQ_COMB_SAME_ADDR | I2C_AQ_NO_ZERO_LEN, 1565 .max_read_len = MAX_SW_I2C_COMMANDS, 1566 .max_write_len = MAX_SW_I2C_COMMANDS, 1567 .max_comb_1st_msg_len = 2, 1568 .max_comb_2nd_msg_len = MAX_SW_I2C_COMMANDS - 2, 1569 }; 1570 1571 static int aldebaran_i2c_control_init(struct smu_context *smu) 1572 { 1573 struct amdgpu_device *adev = smu->adev; 1574 struct amdgpu_smu_i2c_bus *smu_i2c = &adev->pm.smu_i2c[0]; 1575 struct i2c_adapter *control = &smu_i2c->adapter; 1576 int res; 1577 1578 smu_i2c->adev = adev; 1579 smu_i2c->port = 0; 1580 mutex_init(&smu_i2c->mutex); 1581 control->owner = THIS_MODULE; 1582 control->class = I2C_CLASS_SPD; 1583 control->dev.parent = &adev->pdev->dev; 1584 control->algo = &aldebaran_i2c_algo; 1585 snprintf(control->name, sizeof(control->name), "AMDGPU SMU 0"); 1586 control->quirks = &aldebaran_i2c_control_quirks; 1587 i2c_set_adapdata(control, smu_i2c); 1588 1589 res = i2c_add_adapter(control); 1590 if (res) { 1591 DRM_ERROR("Failed to register hw i2c, err: %d\n", res); 1592 goto Out_err; 1593 } 1594 1595 adev->pm.ras_eeprom_i2c_bus = &adev->pm.smu_i2c[0].adapter; 1596 adev->pm.fru_eeprom_i2c_bus = &adev->pm.smu_i2c[0].adapter; 1597 1598 return 0; 1599 Out_err: 1600 i2c_del_adapter(control); 1601 1602 return res; 1603 } 1604 1605 static void aldebaran_i2c_control_fini(struct smu_context *smu) 1606 { 1607 struct amdgpu_device *adev = smu->adev; 1608 int i; 1609 1610 for (i = 0; i < MAX_SMU_I2C_BUSES; i++) { 1611 struct amdgpu_smu_i2c_bus *smu_i2c = &adev->pm.smu_i2c[i]; 1612 struct i2c_adapter *control = &smu_i2c->adapter; 1613 1614 i2c_del_adapter(control); 1615 } 1616 adev->pm.ras_eeprom_i2c_bus = NULL; 1617 adev->pm.fru_eeprom_i2c_bus = NULL; 1618 } 1619 1620 static void aldebaran_get_unique_id(struct smu_context *smu) 1621 { 1622 struct amdgpu_device *adev = smu->adev; 1623 uint32_t upper32 = 0, lower32 = 0; 1624 1625 if (aldebaran_get_smu_metrics_data(smu, METRICS_UNIQUE_ID_UPPER32, &upper32)) 1626 goto out; 1627 if (aldebaran_get_smu_metrics_data(smu, METRICS_UNIQUE_ID_LOWER32, &lower32)) 1628 goto out; 1629 1630 out: 1631 adev->unique_id = ((uint64_t)upper32 << 32) | lower32; 1632 if (adev->serial[0] == '\0') 1633 sprintf(adev->serial, "%016llx", adev->unique_id); 1634 } 1635 1636 static bool aldebaran_is_baco_supported(struct smu_context *smu) 1637 { 1638 /* aldebaran is not support baco */ 1639 1640 return false; 1641 } 1642 1643 static int aldebaran_set_df_cstate(struct smu_context *smu, 1644 enum pp_df_cstate state) 1645 { 1646 return smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_DFCstateControl, state, NULL); 1647 } 1648 1649 static int aldebaran_allow_xgmi_power_down(struct smu_context *smu, bool en) 1650 { 1651 struct amdgpu_device *adev = smu->adev; 1652 1653 /* The message only works on master die and NACK will be sent 1654 back for other dies, only send it on master die */ 1655 if (!adev->smuio.funcs->get_socket_id(adev) && 1656 !adev->smuio.funcs->get_die_id(adev)) 1657 return smu_cmn_send_smc_msg_with_param(smu, 1658 SMU_MSG_GmiPwrDnControl, 1659 en ? 0 : 1, 1660 NULL); 1661 else 1662 return 0; 1663 } 1664 1665 static const struct throttling_logging_label { 1666 uint32_t feature_mask; 1667 const char *label; 1668 } logging_label[] = { 1669 {(1U << THROTTLER_TEMP_MEM_BIT), "HBM"}, 1670 {(1U << THROTTLER_TEMP_VR_GFX_BIT), "VR of GFX rail"}, 1671 {(1U << THROTTLER_TEMP_VR_MEM_BIT), "VR of HBM rail"}, 1672 {(1U << THROTTLER_TEMP_VR_SOC_BIT), "VR of SOC rail"}, 1673 }; 1674 static void aldebaran_log_thermal_throttling_event(struct smu_context *smu) 1675 { 1676 int ret; 1677 int throttler_idx, throtting_events = 0, buf_idx = 0; 1678 struct amdgpu_device *adev = smu->adev; 1679 uint32_t throttler_status; 1680 char log_buf[256]; 1681 1682 ret = aldebaran_get_smu_metrics_data(smu, 1683 METRICS_THROTTLER_STATUS, 1684 &throttler_status); 1685 if (ret) 1686 return; 1687 1688 memset(log_buf, 0, sizeof(log_buf)); 1689 for (throttler_idx = 0; throttler_idx < ARRAY_SIZE(logging_label); 1690 throttler_idx++) { 1691 if (throttler_status & logging_label[throttler_idx].feature_mask) { 1692 throtting_events++; 1693 buf_idx += snprintf(log_buf + buf_idx, 1694 sizeof(log_buf) - buf_idx, 1695 "%s%s", 1696 throtting_events > 1 ? " and " : "", 1697 logging_label[throttler_idx].label); 1698 if (buf_idx >= sizeof(log_buf)) { 1699 dev_err(adev->dev, "buffer overflow!\n"); 1700 log_buf[sizeof(log_buf) - 1] = '\0'; 1701 break; 1702 } 1703 } 1704 } 1705 1706 dev_warn(adev->dev, "WARN: GPU thermal throttling temperature reached, expect performance decrease. %s.\n", 1707 log_buf); 1708 kgd2kfd_smi_event_throttle(smu->adev->kfd.dev, 1709 smu_cmn_get_indep_throttler_status(throttler_status, 1710 aldebaran_throttler_map)); 1711 } 1712 1713 static int aldebaran_get_current_pcie_link_speed(struct smu_context *smu) 1714 { 1715 struct amdgpu_device *adev = smu->adev; 1716 uint32_t esm_ctrl; 1717 1718 /* TODO: confirm this on real target */ 1719 esm_ctrl = RREG32_PCIE(smnPCIE_ESM_CTRL); 1720 if ((esm_ctrl >> 15) & 0x1FFFF) 1721 return (((esm_ctrl >> 8) & 0x3F) + 128); 1722 1723 return smu_v13_0_get_current_pcie_link_speed(smu); 1724 } 1725 1726 static ssize_t aldebaran_get_gpu_metrics(struct smu_context *smu, 1727 void **table) 1728 { 1729 struct smu_table_context *smu_table = &smu->smu_table; 1730 struct gpu_metrics_v1_3 *gpu_metrics = 1731 (struct gpu_metrics_v1_3 *)smu_table->gpu_metrics_table; 1732 SmuMetrics_t metrics; 1733 int i, ret = 0; 1734 1735 ret = smu_cmn_get_metrics_table(smu, 1736 &metrics, 1737 true); 1738 if (ret) 1739 return ret; 1740 1741 smu_cmn_init_soft_gpu_metrics(gpu_metrics, 1, 3); 1742 1743 gpu_metrics->temperature_edge = metrics.TemperatureEdge; 1744 gpu_metrics->temperature_hotspot = metrics.TemperatureHotspot; 1745 gpu_metrics->temperature_mem = metrics.TemperatureHBM; 1746 gpu_metrics->temperature_vrgfx = metrics.TemperatureVrGfx; 1747 gpu_metrics->temperature_vrsoc = metrics.TemperatureVrSoc; 1748 gpu_metrics->temperature_vrmem = metrics.TemperatureVrMem; 1749 1750 gpu_metrics->average_gfx_activity = metrics.AverageGfxActivity; 1751 gpu_metrics->average_umc_activity = metrics.AverageUclkActivity; 1752 gpu_metrics->average_mm_activity = 0; 1753 1754 /* Valid power data is available only from primary die */ 1755 if (aldebaran_is_primary(smu)) { 1756 gpu_metrics->average_socket_power = metrics.AverageSocketPower; 1757 gpu_metrics->energy_accumulator = 1758 (uint64_t)metrics.EnergyAcc64bitHigh << 32 | 1759 metrics.EnergyAcc64bitLow; 1760 } else { 1761 gpu_metrics->average_socket_power = 0; 1762 gpu_metrics->energy_accumulator = 0; 1763 } 1764 1765 gpu_metrics->average_gfxclk_frequency = metrics.AverageGfxclkFrequency; 1766 gpu_metrics->average_socclk_frequency = metrics.AverageSocclkFrequency; 1767 gpu_metrics->average_uclk_frequency = metrics.AverageUclkFrequency; 1768 gpu_metrics->average_vclk0_frequency = 0; 1769 gpu_metrics->average_dclk0_frequency = 0; 1770 1771 gpu_metrics->current_gfxclk = metrics.CurrClock[PPCLK_GFXCLK]; 1772 gpu_metrics->current_socclk = metrics.CurrClock[PPCLK_SOCCLK]; 1773 gpu_metrics->current_uclk = metrics.CurrClock[PPCLK_UCLK]; 1774 gpu_metrics->current_vclk0 = metrics.CurrClock[PPCLK_VCLK]; 1775 gpu_metrics->current_dclk0 = metrics.CurrClock[PPCLK_DCLK]; 1776 1777 gpu_metrics->throttle_status = metrics.ThrottlerStatus; 1778 gpu_metrics->indep_throttle_status = 1779 smu_cmn_get_indep_throttler_status(metrics.ThrottlerStatus, 1780 aldebaran_throttler_map); 1781 1782 gpu_metrics->current_fan_speed = 0; 1783 1784 gpu_metrics->pcie_link_width = 1785 smu_v13_0_get_current_pcie_link_width(smu); 1786 gpu_metrics->pcie_link_speed = 1787 aldebaran_get_current_pcie_link_speed(smu); 1788 1789 gpu_metrics->system_clock_counter = ktime_get_boottime_ns(); 1790 1791 gpu_metrics->gfx_activity_acc = metrics.GfxBusyAcc; 1792 gpu_metrics->mem_activity_acc = metrics.DramBusyAcc; 1793 1794 for (i = 0; i < NUM_HBM_INSTANCES; i++) 1795 gpu_metrics->temperature_hbm[i] = metrics.TemperatureAllHBM[i]; 1796 1797 gpu_metrics->firmware_timestamp = ((uint64_t)metrics.TimeStampHigh << 32) | 1798 metrics.TimeStampLow; 1799 1800 *table = (void *)gpu_metrics; 1801 1802 return sizeof(struct gpu_metrics_v1_3); 1803 } 1804 1805 static int aldebaran_check_ecc_table_support(struct smu_context *smu) 1806 { 1807 uint32_t if_version = 0xff, smu_version = 0xff; 1808 int ret = 0; 1809 1810 ret = smu_cmn_get_smc_version(smu, &if_version, &smu_version); 1811 if (ret) { 1812 /* return not support if failed get smu_version */ 1813 ret = -EOPNOTSUPP; 1814 } 1815 1816 if (smu_version < SUPPORT_ECCTABLE_SMU_VERSION) 1817 ret = -EOPNOTSUPP; 1818 1819 return ret; 1820 } 1821 1822 static ssize_t aldebaran_get_ecc_info(struct smu_context *smu, 1823 void *table) 1824 { 1825 struct smu_table_context *smu_table = &smu->smu_table; 1826 EccInfoTable_t *ecc_table = NULL; 1827 struct ecc_info_per_ch *ecc_info_per_channel = NULL; 1828 int i, ret = 0; 1829 struct umc_ecc_info *eccinfo = (struct umc_ecc_info *)table; 1830 1831 ret = aldebaran_check_ecc_table_support(smu); 1832 if (ret) 1833 return ret; 1834 1835 ret = smu_cmn_update_table(smu, 1836 SMU_TABLE_ECCINFO, 1837 0, 1838 smu_table->ecc_table, 1839 false); 1840 if (ret) { 1841 dev_info(smu->adev->dev, "Failed to export SMU ecc table!\n"); 1842 return ret; 1843 } 1844 1845 ecc_table = (EccInfoTable_t *)smu_table->ecc_table; 1846 1847 for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) { 1848 ecc_info_per_channel = &(eccinfo->ecc[i]); 1849 ecc_info_per_channel->ce_count_lo_chip = 1850 ecc_table->EccInfo[i].ce_count_lo_chip; 1851 ecc_info_per_channel->ce_count_hi_chip = 1852 ecc_table->EccInfo[i].ce_count_hi_chip; 1853 ecc_info_per_channel->mca_umc_status = 1854 ecc_table->EccInfo[i].mca_umc_status; 1855 ecc_info_per_channel->mca_umc_addr = 1856 ecc_table->EccInfo[i].mca_umc_addr; 1857 } 1858 1859 return ret; 1860 } 1861 1862 static int aldebaran_mode1_reset(struct smu_context *smu) 1863 { 1864 u32 smu_version, fatal_err, param; 1865 int ret = 0; 1866 struct amdgpu_device *adev = smu->adev; 1867 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 1868 1869 fatal_err = 0; 1870 param = SMU_RESET_MODE_1; 1871 1872 /* 1873 * PM FW support SMU_MSG_GfxDeviceDriverReset from 68.07 1874 */ 1875 smu_cmn_get_smc_version(smu, NULL, &smu_version); 1876 if (smu_version < 0x00440700) { 1877 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_Mode1Reset, NULL); 1878 } 1879 else { 1880 /* fatal error triggered by ras, PMFW supports the flag 1881 from 68.44.0 */ 1882 if ((smu_version >= 0x00442c00) && ras && 1883 atomic_read(&ras->in_recovery)) 1884 fatal_err = 1; 1885 1886 param |= (fatal_err << 16); 1887 ret = smu_cmn_send_smc_msg_with_param(smu, 1888 SMU_MSG_GfxDeviceDriverReset, param, NULL); 1889 } 1890 1891 if (!ret) 1892 msleep(SMU13_MODE1_RESET_WAIT_TIME_IN_MS); 1893 1894 return ret; 1895 } 1896 1897 static int aldebaran_mode2_reset(struct smu_context *smu) 1898 { 1899 u32 smu_version; 1900 int ret = 0, index; 1901 struct amdgpu_device *adev = smu->adev; 1902 int timeout = 10; 1903 1904 smu_cmn_get_smc_version(smu, NULL, &smu_version); 1905 1906 index = smu_cmn_to_asic_specific_index(smu, CMN2ASIC_MAPPING_MSG, 1907 SMU_MSG_GfxDeviceDriverReset); 1908 1909 mutex_lock(&smu->message_lock); 1910 if (smu_version >= 0x00441400) { 1911 ret = smu_cmn_send_msg_without_waiting(smu, (uint16_t)index, SMU_RESET_MODE_2); 1912 /* This is similar to FLR, wait till max FLR timeout */ 1913 msleep(100); 1914 dev_dbg(smu->adev->dev, "restore config space...\n"); 1915 /* Restore the config space saved during init */ 1916 amdgpu_device_load_pci_state(adev->pdev); 1917 1918 dev_dbg(smu->adev->dev, "wait for reset ack\n"); 1919 while (ret == -ETIME && timeout) { 1920 ret = smu_cmn_wait_for_response(smu); 1921 /* Wait a bit more time for getting ACK */ 1922 if (ret == -ETIME) { 1923 --timeout; 1924 usleep_range(500, 1000); 1925 continue; 1926 } 1927 1928 if (ret != 1) { 1929 dev_err(adev->dev, "failed to send mode2 message \tparam: 0x%08x response %#x\n", 1930 SMU_RESET_MODE_2, ret); 1931 goto out; 1932 } 1933 } 1934 1935 } else { 1936 dev_err(adev->dev, "smu fw 0x%x does not support MSG_GfxDeviceDriverReset MSG\n", 1937 smu_version); 1938 } 1939 1940 if (ret == 1) 1941 ret = 0; 1942 out: 1943 mutex_unlock(&smu->message_lock); 1944 1945 return ret; 1946 } 1947 1948 static int aldebaran_smu_handle_passthrough_sbr(struct smu_context *smu, bool enable) 1949 { 1950 int ret = 0; 1951 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_HeavySBR, enable ? 1 : 0, NULL); 1952 1953 return ret; 1954 } 1955 1956 static bool aldebaran_is_mode1_reset_supported(struct smu_context *smu) 1957 { 1958 #if 0 1959 struct amdgpu_device *adev = smu->adev; 1960 u32 smu_version; 1961 uint32_t val; 1962 /** 1963 * PM FW version support mode1 reset from 68.07 1964 */ 1965 smu_cmn_get_smc_version(smu, NULL, &smu_version); 1966 if ((smu_version < 0x00440700)) 1967 return false; 1968 /** 1969 * mode1 reset relies on PSP, so we should check if 1970 * PSP is alive. 1971 */ 1972 val = RREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_81); 1973 1974 return val != 0x0; 1975 #endif 1976 return true; 1977 } 1978 1979 static bool aldebaran_is_mode2_reset_supported(struct smu_context *smu) 1980 { 1981 return true; 1982 } 1983 1984 static int aldebaran_set_mp1_state(struct smu_context *smu, 1985 enum pp_mp1_state mp1_state) 1986 { 1987 switch (mp1_state) { 1988 case PP_MP1_STATE_UNLOAD: 1989 return smu_cmn_set_mp1_state(smu, mp1_state); 1990 default: 1991 return 0; 1992 } 1993 } 1994 1995 static int aldebaran_smu_send_hbm_bad_page_num(struct smu_context *smu, 1996 uint32_t size) 1997 { 1998 int ret = 0; 1999 2000 /* message SMU to update the bad page number on SMUBUS */ 2001 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_SetNumBadHbmPagesRetired, size, NULL); 2002 if (ret) 2003 dev_err(smu->adev->dev, "[%s] failed to message SMU to update HBM bad pages number\n", 2004 __func__); 2005 2006 return ret; 2007 } 2008 2009 static int aldebaran_check_bad_channel_info_support(struct smu_context *smu) 2010 { 2011 uint32_t if_version = 0xff, smu_version = 0xff; 2012 int ret = 0; 2013 2014 ret = smu_cmn_get_smc_version(smu, &if_version, &smu_version); 2015 if (ret) { 2016 /* return not support if failed get smu_version */ 2017 ret = -EOPNOTSUPP; 2018 } 2019 2020 if (smu_version < SUPPORT_BAD_CHANNEL_INFO_MSG_VERSION) 2021 ret = -EOPNOTSUPP; 2022 2023 return ret; 2024 } 2025 2026 static int aldebaran_send_hbm_bad_channel_flag(struct smu_context *smu, 2027 uint32_t size) 2028 { 2029 int ret = 0; 2030 2031 ret = aldebaran_check_bad_channel_info_support(smu); 2032 if (ret) 2033 return ret; 2034 2035 /* message SMU to update the bad channel info on SMUBUS */ 2036 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_SetBadHBMPagesRetiredFlagsPerChannel, size, NULL); 2037 if (ret) 2038 dev_err(smu->adev->dev, "[%s] failed to message SMU to update HBM bad channel info\n", 2039 __func__); 2040 2041 return ret; 2042 } 2043 2044 static const struct pptable_funcs aldebaran_ppt_funcs = { 2045 /* init dpm */ 2046 .get_allowed_feature_mask = aldebaran_get_allowed_feature_mask, 2047 /* dpm/clk tables */ 2048 .set_default_dpm_table = aldebaran_set_default_dpm_table, 2049 .populate_umd_state_clk = aldebaran_populate_umd_state_clk, 2050 .get_thermal_temperature_range = aldebaran_get_thermal_temperature_range, 2051 .print_clk_levels = aldebaran_print_clk_levels, 2052 .force_clk_levels = aldebaran_force_clk_levels, 2053 .read_sensor = aldebaran_read_sensor, 2054 .set_performance_level = aldebaran_set_performance_level, 2055 .get_power_limit = aldebaran_get_power_limit, 2056 .is_dpm_running = aldebaran_is_dpm_running, 2057 .get_unique_id = aldebaran_get_unique_id, 2058 .init_microcode = smu_v13_0_init_microcode, 2059 .load_microcode = smu_v13_0_load_microcode, 2060 .fini_microcode = smu_v13_0_fini_microcode, 2061 .init_smc_tables = aldebaran_init_smc_tables, 2062 .fini_smc_tables = smu_v13_0_fini_smc_tables, 2063 .init_power = smu_v13_0_init_power, 2064 .fini_power = smu_v13_0_fini_power, 2065 .check_fw_status = smu_v13_0_check_fw_status, 2066 /* pptable related */ 2067 .setup_pptable = aldebaran_setup_pptable, 2068 .get_vbios_bootup_values = smu_v13_0_get_vbios_bootup_values, 2069 .check_fw_version = smu_v13_0_check_fw_version, 2070 .write_pptable = smu_cmn_write_pptable, 2071 .set_driver_table_location = smu_v13_0_set_driver_table_location, 2072 .set_tool_table_location = smu_v13_0_set_tool_table_location, 2073 .notify_memory_pool_location = smu_v13_0_notify_memory_pool_location, 2074 .system_features_control = aldebaran_system_features_control, 2075 .send_smc_msg_with_param = smu_cmn_send_smc_msg_with_param, 2076 .send_smc_msg = smu_cmn_send_smc_msg, 2077 .get_enabled_mask = smu_cmn_get_enabled_mask, 2078 .feature_is_enabled = smu_cmn_feature_is_enabled, 2079 .disable_all_features_with_exception = smu_cmn_disable_all_features_with_exception, 2080 .set_power_limit = aldebaran_set_power_limit, 2081 .init_max_sustainable_clocks = smu_v13_0_init_max_sustainable_clocks, 2082 .enable_thermal_alert = smu_v13_0_enable_thermal_alert, 2083 .disable_thermal_alert = smu_v13_0_disable_thermal_alert, 2084 .set_xgmi_pstate = smu_v13_0_set_xgmi_pstate, 2085 .register_irq_handler = smu_v13_0_register_irq_handler, 2086 .set_azalia_d3_pme = smu_v13_0_set_azalia_d3_pme, 2087 .get_max_sustainable_clocks_by_dc = smu_v13_0_get_max_sustainable_clocks_by_dc, 2088 .baco_is_support= aldebaran_is_baco_supported, 2089 .get_dpm_ultimate_freq = smu_v13_0_get_dpm_ultimate_freq, 2090 .set_soft_freq_limited_range = aldebaran_set_soft_freq_limited_range, 2091 .od_edit_dpm_table = aldebaran_usr_edit_dpm_table, 2092 .set_df_cstate = aldebaran_set_df_cstate, 2093 .allow_xgmi_power_down = aldebaran_allow_xgmi_power_down, 2094 .log_thermal_throttling_event = aldebaran_log_thermal_throttling_event, 2095 .get_pp_feature_mask = smu_cmn_get_pp_feature_mask, 2096 .set_pp_feature_mask = smu_cmn_set_pp_feature_mask, 2097 .get_gpu_metrics = aldebaran_get_gpu_metrics, 2098 .mode1_reset_is_support = aldebaran_is_mode1_reset_supported, 2099 .mode2_reset_is_support = aldebaran_is_mode2_reset_supported, 2100 .smu_handle_passthrough_sbr = aldebaran_smu_handle_passthrough_sbr, 2101 .mode1_reset = aldebaran_mode1_reset, 2102 .set_mp1_state = aldebaran_set_mp1_state, 2103 .mode2_reset = aldebaran_mode2_reset, 2104 .wait_for_event = smu_v13_0_wait_for_event, 2105 .i2c_init = aldebaran_i2c_control_init, 2106 .i2c_fini = aldebaran_i2c_control_fini, 2107 .send_hbm_bad_pages_num = aldebaran_smu_send_hbm_bad_page_num, 2108 .get_ecc_info = aldebaran_get_ecc_info, 2109 .send_hbm_bad_channel_flag = aldebaran_send_hbm_bad_channel_flag, 2110 }; 2111 2112 void aldebaran_set_ppt_funcs(struct smu_context *smu) 2113 { 2114 smu->ppt_funcs = &aldebaran_ppt_funcs; 2115 smu->message_map = aldebaran_message_map; 2116 smu->clock_map = aldebaran_clk_map; 2117 smu->feature_map = aldebaran_feature_mask_map; 2118 smu->table_map = aldebaran_table_map; 2119 } 2120