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_smu.h" 29 #include "atomfirmware.h" 30 #include "amdgpu_atomfirmware.h" 31 #include "amdgpu_atombios.h" 32 #include "smu_v13_0.h" 33 #include "smu13_driver_if_aldebaran.h" 34 #include "soc15_common.h" 35 #include "atom.h" 36 #include "power_state.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 to_amdgpu_device(x) (container_of(x, struct amdgpu_device, pm.smu_i2c)) 61 62 #define ALDEBARAN_FEA_MAP(smu_feature, aldebaran_feature) \ 63 [smu_feature] = {1, (aldebaran_feature)} 64 65 #define FEATURE_MASK(feature) (1ULL << feature) 66 #define SMC_DPM_FEATURE ( \ 67 FEATURE_MASK(FEATURE_DATA_CALCULATIONS) | \ 68 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT) | \ 69 FEATURE_MASK(FEATURE_DPM_UCLK_BIT) | \ 70 FEATURE_MASK(FEATURE_DPM_SOCCLK_BIT) | \ 71 FEATURE_MASK(FEATURE_DPM_FCLK_BIT) | \ 72 FEATURE_MASK(FEATURE_DPM_LCLK_BIT) | \ 73 FEATURE_MASK(FEATURE_DPM_XGMI_BIT) | \ 74 FEATURE_MASK(FEATURE_DPM_VCN_BIT)) 75 76 /* possible frequency drift (1Mhz) */ 77 #define EPSILON 1 78 79 #define smnPCIE_ESM_CTRL 0x111003D0 80 81 static const struct smu_temperature_range smu13_thermal_policy[] = 82 { 83 {-273150, 99000, 99000, -273150, 99000, 99000, -273150, 99000, 99000}, 84 { 120000, 120000, 120000, 120000, 120000, 120000, 120000, 120000, 120000}, 85 }; 86 87 static const struct cmn2asic_msg_mapping aldebaran_message_map[SMU_MSG_MAX_COUNT] = { 88 MSG_MAP(TestMessage, PPSMC_MSG_TestMessage, 0), 89 MSG_MAP(GetSmuVersion, PPSMC_MSG_GetSmuVersion, 1), 90 MSG_MAP(GetDriverIfVersion, PPSMC_MSG_GetDriverIfVersion, 1), 91 MSG_MAP(EnableAllSmuFeatures, PPSMC_MSG_EnableAllSmuFeatures, 0), 92 MSG_MAP(DisableAllSmuFeatures, PPSMC_MSG_DisableAllSmuFeatures, 0), 93 MSG_MAP(GetEnabledSmuFeaturesLow, PPSMC_MSG_GetEnabledSmuFeaturesLow, 1), 94 MSG_MAP(GetEnabledSmuFeaturesHigh, PPSMC_MSG_GetEnabledSmuFeaturesHigh, 1), 95 MSG_MAP(SetDriverDramAddrHigh, PPSMC_MSG_SetDriverDramAddrHigh, 1), 96 MSG_MAP(SetDriverDramAddrLow, PPSMC_MSG_SetDriverDramAddrLow, 1), 97 MSG_MAP(SetToolsDramAddrHigh, PPSMC_MSG_SetToolsDramAddrHigh, 0), 98 MSG_MAP(SetToolsDramAddrLow, PPSMC_MSG_SetToolsDramAddrLow, 0), 99 MSG_MAP(TransferTableSmu2Dram, PPSMC_MSG_TransferTableSmu2Dram, 1), 100 MSG_MAP(TransferTableDram2Smu, PPSMC_MSG_TransferTableDram2Smu, 0), 101 MSG_MAP(UseDefaultPPTable, PPSMC_MSG_UseDefaultPPTable, 0), 102 MSG_MAP(SetSystemVirtualDramAddrHigh, PPSMC_MSG_SetSystemVirtualDramAddrHigh, 0), 103 MSG_MAP(SetSystemVirtualDramAddrLow, PPSMC_MSG_SetSystemVirtualDramAddrLow, 0), 104 MSG_MAP(SetSoftMinByFreq, PPSMC_MSG_SetSoftMinByFreq, 0), 105 MSG_MAP(SetSoftMaxByFreq, PPSMC_MSG_SetSoftMaxByFreq, 0), 106 MSG_MAP(SetHardMinByFreq, PPSMC_MSG_SetHardMinByFreq, 0), 107 MSG_MAP(SetHardMaxByFreq, PPSMC_MSG_SetHardMaxByFreq, 0), 108 MSG_MAP(GetMinDpmFreq, PPSMC_MSG_GetMinDpmFreq, 0), 109 MSG_MAP(GetMaxDpmFreq, PPSMC_MSG_GetMaxDpmFreq, 0), 110 MSG_MAP(GetDpmFreqByIndex, PPSMC_MSG_GetDpmFreqByIndex, 1), 111 MSG_MAP(SetWorkloadMask, PPSMC_MSG_SetWorkloadMask, 1), 112 MSG_MAP(GetVoltageByDpm, PPSMC_MSG_GetVoltageByDpm, 0), 113 MSG_MAP(GetVoltageByDpmOverdrive, PPSMC_MSG_GetVoltageByDpmOverdrive, 0), 114 MSG_MAP(SetPptLimit, PPSMC_MSG_SetPptLimit, 0), 115 MSG_MAP(GetPptLimit, PPSMC_MSG_GetPptLimit, 1), 116 MSG_MAP(PrepareMp1ForUnload, PPSMC_MSG_PrepareMp1ForUnload, 0), 117 MSG_MAP(GfxDeviceDriverReset, PPSMC_MSG_GfxDriverReset, 0), 118 MSG_MAP(RunDcBtc, PPSMC_MSG_RunDcBtc, 0), 119 MSG_MAP(DramLogSetDramAddrHigh, PPSMC_MSG_DramLogSetDramAddrHigh, 0), 120 MSG_MAP(DramLogSetDramAddrLow, PPSMC_MSG_DramLogSetDramAddrLow, 0), 121 MSG_MAP(DramLogSetDramSize, PPSMC_MSG_DramLogSetDramSize, 0), 122 MSG_MAP(GetDebugData, PPSMC_MSG_GetDebugData, 0), 123 MSG_MAP(WaflTest, PPSMC_MSG_WaflTest, 0), 124 MSG_MAP(SetMemoryChannelEnable, PPSMC_MSG_SetMemoryChannelEnable, 0), 125 MSG_MAP(SetNumBadHbmPagesRetired, PPSMC_MSG_SetNumBadHbmPagesRetired, 0), 126 MSG_MAP(DFCstateControl, PPSMC_MSG_DFCstateControl, 0), 127 MSG_MAP(GetGmiPwrDnHyst, PPSMC_MSG_GetGmiPwrDnHyst, 0), 128 MSG_MAP(SetGmiPwrDnHyst, PPSMC_MSG_SetGmiPwrDnHyst, 0), 129 MSG_MAP(GmiPwrDnControl, PPSMC_MSG_GmiPwrDnControl, 0), 130 MSG_MAP(EnterGfxoff, PPSMC_MSG_EnterGfxoff, 0), 131 MSG_MAP(ExitGfxoff, PPSMC_MSG_ExitGfxoff, 0), 132 MSG_MAP(SetExecuteDMATest, PPSMC_MSG_SetExecuteDMATest, 0), 133 MSG_MAP(EnableDeterminism, PPSMC_MSG_EnableDeterminism, 0), 134 MSG_MAP(DisableDeterminism, PPSMC_MSG_DisableDeterminism, 0), 135 MSG_MAP(SetUclkDpmMode, PPSMC_MSG_SetUclkDpmMode, 0), 136 MSG_MAP(GfxDriverResetRecovery, PPSMC_MSG_GfxDriverResetRecovery, 0), 137 MSG_MAP(BoardPowerCalibration, PPSMC_MSG_BoardPowerCalibration, 0), 138 }; 139 140 static const struct cmn2asic_mapping aldebaran_clk_map[SMU_CLK_COUNT] = { 141 CLK_MAP(GFXCLK, PPCLK_GFXCLK), 142 CLK_MAP(SCLK, PPCLK_GFXCLK), 143 CLK_MAP(SOCCLK, PPCLK_SOCCLK), 144 CLK_MAP(FCLK, PPCLK_FCLK), 145 CLK_MAP(UCLK, PPCLK_UCLK), 146 CLK_MAP(MCLK, PPCLK_UCLK), 147 CLK_MAP(DCLK, PPCLK_DCLK), 148 CLK_MAP(VCLK, PPCLK_VCLK), 149 CLK_MAP(LCLK, PPCLK_LCLK), 150 }; 151 152 static const struct cmn2asic_mapping aldebaran_feature_mask_map[SMU_FEATURE_COUNT] = { 153 ALDEBARAN_FEA_MAP(SMU_FEATURE_DATA_CALCULATIONS_BIT, FEATURE_DATA_CALCULATIONS), 154 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_GFXCLK_BIT, FEATURE_DPM_GFXCLK_BIT), 155 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_UCLK_BIT, FEATURE_DPM_UCLK_BIT), 156 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_SOCCLK_BIT, FEATURE_DPM_SOCCLK_BIT), 157 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_FCLK_BIT, FEATURE_DPM_FCLK_BIT), 158 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_LCLK_BIT, FEATURE_DPM_LCLK_BIT), 159 ALDEBARAN_FEA_MAP(SMU_FEATURE_DPM_XGMI_BIT, FEATURE_DPM_XGMI_BIT), 160 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_GFXCLK_BIT, FEATURE_DS_GFXCLK_BIT), 161 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_SOCCLK_BIT, FEATURE_DS_SOCCLK_BIT), 162 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_LCLK_BIT, FEATURE_DS_LCLK_BIT), 163 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_FCLK_BIT, FEATURE_DS_FCLK_BIT), 164 ALDEBARAN_FEA_MAP(SMU_FEATURE_DS_UCLK_BIT, FEATURE_DS_UCLK_BIT), 165 ALDEBARAN_FEA_MAP(SMU_FEATURE_GFX_SS_BIT, FEATURE_GFX_SS_BIT), 166 ALDEBARAN_FEA_MAP(SMU_FEATURE_VCN_DPM_BIT, FEATURE_DPM_VCN_BIT), 167 ALDEBARAN_FEA_MAP(SMU_FEATURE_RSMU_SMN_CG_BIT, FEATURE_RSMU_SMN_CG_BIT), 168 ALDEBARAN_FEA_MAP(SMU_FEATURE_WAFL_CG_BIT, FEATURE_WAFL_CG_BIT), 169 ALDEBARAN_FEA_MAP(SMU_FEATURE_PPT_BIT, FEATURE_PPT_BIT), 170 ALDEBARAN_FEA_MAP(SMU_FEATURE_TDC_BIT, FEATURE_TDC_BIT), 171 ALDEBARAN_FEA_MAP(SMU_FEATURE_APCC_PLUS_BIT, FEATURE_APCC_PLUS_BIT), 172 ALDEBARAN_FEA_MAP(SMU_FEATURE_APCC_DFLL_BIT, FEATURE_APCC_DFLL_BIT), 173 ALDEBARAN_FEA_MAP(SMU_FEATURE_FUSE_CG_BIT, FEATURE_FUSE_CG_BIT), 174 ALDEBARAN_FEA_MAP(SMU_FEATURE_MP1_CG_BIT, FEATURE_MP1_CG_BIT), 175 ALDEBARAN_FEA_MAP(SMU_FEATURE_SMUIO_CG_BIT, FEATURE_SMUIO_CG_BIT), 176 ALDEBARAN_FEA_MAP(SMU_FEATURE_THM_CG_BIT, FEATURE_THM_CG_BIT), 177 ALDEBARAN_FEA_MAP(SMU_FEATURE_CLK_CG_BIT, FEATURE_CLK_CG_BIT), 178 ALDEBARAN_FEA_MAP(SMU_FEATURE_FW_CTF_BIT, FEATURE_FW_CTF_BIT), 179 ALDEBARAN_FEA_MAP(SMU_FEATURE_THERMAL_BIT, FEATURE_THERMAL_BIT), 180 ALDEBARAN_FEA_MAP(SMU_FEATURE_OUT_OF_BAND_MONITOR_BIT, FEATURE_OUT_OF_BAND_MONITOR_BIT), 181 ALDEBARAN_FEA_MAP(SMU_FEATURE_XGMI_PER_LINK_PWR_DWN_BIT,FEATURE_XGMI_PER_LINK_PWR_DWN), 182 ALDEBARAN_FEA_MAP(SMU_FEATURE_DF_CSTATE_BIT, FEATURE_DF_CSTATE), 183 }; 184 185 static const struct cmn2asic_mapping aldebaran_table_map[SMU_TABLE_COUNT] = { 186 TAB_MAP(PPTABLE), 187 TAB_MAP(AVFS_PSM_DEBUG), 188 TAB_MAP(AVFS_FUSE_OVERRIDE), 189 TAB_MAP(PMSTATUSLOG), 190 TAB_MAP(SMU_METRICS), 191 TAB_MAP(DRIVER_SMU_CONFIG), 192 TAB_MAP(I2C_COMMANDS), 193 }; 194 195 static const uint8_t aldebaran_throttler_map[] = { 196 [THROTTLER_PPT0_BIT] = (SMU_THROTTLER_PPT0_BIT), 197 [THROTTLER_PPT1_BIT] = (SMU_THROTTLER_PPT1_BIT), 198 [THROTTLER_TDC_GFX_BIT] = (SMU_THROTTLER_TDC_GFX_BIT), 199 [THROTTLER_TDC_SOC_BIT] = (SMU_THROTTLER_TDC_SOC_BIT), 200 [THROTTLER_TDC_HBM_BIT] = (SMU_THROTTLER_TDC_MEM_BIT), 201 [THROTTLER_TEMP_GPU_BIT] = (SMU_THROTTLER_TEMP_GPU_BIT), 202 [THROTTLER_TEMP_MEM_BIT] = (SMU_THROTTLER_TEMP_MEM_BIT), 203 [THROTTLER_TEMP_VR_GFX_BIT] = (SMU_THROTTLER_TEMP_VR_GFX_BIT), 204 [THROTTLER_TEMP_VR_SOC_BIT] = (SMU_THROTTLER_TEMP_VR_SOC_BIT), 205 [THROTTLER_TEMP_VR_MEM_BIT] = (SMU_THROTTLER_TEMP_VR_MEM0_BIT), 206 [THROTTLER_APCC_BIT] = (SMU_THROTTLER_APCC_BIT), 207 }; 208 209 static int aldebaran_tables_init(struct smu_context *smu) 210 { 211 struct smu_table_context *smu_table = &smu->smu_table; 212 struct smu_table *tables = smu_table->tables; 213 214 SMU_TABLE_INIT(tables, SMU_TABLE_PPTABLE, sizeof(PPTable_t), 215 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 216 217 SMU_TABLE_INIT(tables, SMU_TABLE_PMSTATUSLOG, SMU13_TOOL_SIZE, 218 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 219 220 SMU_TABLE_INIT(tables, SMU_TABLE_SMU_METRICS, sizeof(SmuMetrics_t), 221 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 222 223 SMU_TABLE_INIT(tables, SMU_TABLE_I2C_COMMANDS, sizeof(SwI2cRequest_t), 224 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); 225 226 smu_table->metrics_table = kzalloc(sizeof(SmuMetrics_t), GFP_KERNEL); 227 if (!smu_table->metrics_table) 228 return -ENOMEM; 229 smu_table->metrics_time = 0; 230 231 smu_table->gpu_metrics_table_size = sizeof(struct gpu_metrics_v1_3); 232 smu_table->gpu_metrics_table = kzalloc(smu_table->gpu_metrics_table_size, GFP_KERNEL); 233 if (!smu_table->gpu_metrics_table) { 234 kfree(smu_table->metrics_table); 235 return -ENOMEM; 236 } 237 238 return 0; 239 } 240 241 static int aldebaran_allocate_dpm_context(struct smu_context *smu) 242 { 243 struct smu_dpm_context *smu_dpm = &smu->smu_dpm; 244 245 smu_dpm->dpm_context = kzalloc(sizeof(struct smu_13_0_dpm_context), 246 GFP_KERNEL); 247 if (!smu_dpm->dpm_context) 248 return -ENOMEM; 249 smu_dpm->dpm_context_size = sizeof(struct smu_13_0_dpm_context); 250 251 smu_dpm->dpm_current_power_state = kzalloc(sizeof(struct smu_power_state), 252 GFP_KERNEL); 253 if (!smu_dpm->dpm_current_power_state) 254 return -ENOMEM; 255 256 smu_dpm->dpm_request_power_state = kzalloc(sizeof(struct smu_power_state), 257 GFP_KERNEL); 258 if (!smu_dpm->dpm_request_power_state) 259 return -ENOMEM; 260 261 return 0; 262 } 263 264 static int aldebaran_init_smc_tables(struct smu_context *smu) 265 { 266 int ret = 0; 267 268 ret = aldebaran_tables_init(smu); 269 if (ret) 270 return ret; 271 272 ret = aldebaran_allocate_dpm_context(smu); 273 if (ret) 274 return ret; 275 276 return smu_v13_0_init_smc_tables(smu); 277 } 278 279 static int aldebaran_get_allowed_feature_mask(struct smu_context *smu, 280 uint32_t *feature_mask, uint32_t num) 281 { 282 if (num > 2) 283 return -EINVAL; 284 285 /* pptable will handle the features to enable */ 286 memset(feature_mask, 0xFF, sizeof(uint32_t) * num); 287 288 return 0; 289 } 290 291 static int aldebaran_set_default_dpm_table(struct smu_context *smu) 292 { 293 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context; 294 struct smu_13_0_dpm_table *dpm_table = NULL; 295 PPTable_t *pptable = smu->smu_table.driver_pptable; 296 int ret = 0; 297 298 /* socclk dpm table setup */ 299 dpm_table = &dpm_context->dpm_tables.soc_table; 300 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT)) { 301 ret = smu_v13_0_set_single_dpm_table(smu, 302 SMU_SOCCLK, 303 dpm_table); 304 if (ret) 305 return ret; 306 } else { 307 dpm_table->count = 1; 308 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.socclk / 100; 309 dpm_table->dpm_levels[0].enabled = true; 310 dpm_table->min = dpm_table->dpm_levels[0].value; 311 dpm_table->max = dpm_table->dpm_levels[0].value; 312 } 313 314 /* gfxclk dpm table setup */ 315 dpm_table = &dpm_context->dpm_tables.gfx_table; 316 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT)) { 317 /* in the case of gfxclk, only fine-grained dpm is honored */ 318 dpm_table->count = 2; 319 dpm_table->dpm_levels[0].value = pptable->GfxclkFmin; 320 dpm_table->dpm_levels[0].enabled = true; 321 dpm_table->dpm_levels[1].value = pptable->GfxclkFmax; 322 dpm_table->dpm_levels[1].enabled = true; 323 dpm_table->min = dpm_table->dpm_levels[0].value; 324 dpm_table->max = dpm_table->dpm_levels[1].value; 325 } else { 326 dpm_table->count = 1; 327 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.gfxclk / 100; 328 dpm_table->dpm_levels[0].enabled = true; 329 dpm_table->min = dpm_table->dpm_levels[0].value; 330 dpm_table->max = dpm_table->dpm_levels[0].value; 331 } 332 333 /* memclk dpm table setup */ 334 dpm_table = &dpm_context->dpm_tables.uclk_table; 335 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT)) { 336 ret = smu_v13_0_set_single_dpm_table(smu, 337 SMU_UCLK, 338 dpm_table); 339 if (ret) 340 return ret; 341 } else { 342 dpm_table->count = 1; 343 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.uclk / 100; 344 dpm_table->dpm_levels[0].enabled = true; 345 dpm_table->min = dpm_table->dpm_levels[0].value; 346 dpm_table->max = dpm_table->dpm_levels[0].value; 347 } 348 349 /* fclk dpm table setup */ 350 dpm_table = &dpm_context->dpm_tables.fclk_table; 351 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_FCLK_BIT)) { 352 ret = smu_v13_0_set_single_dpm_table(smu, 353 SMU_FCLK, 354 dpm_table); 355 if (ret) 356 return ret; 357 } else { 358 dpm_table->count = 1; 359 dpm_table->dpm_levels[0].value = smu->smu_table.boot_values.fclk / 100; 360 dpm_table->dpm_levels[0].enabled = true; 361 dpm_table->min = dpm_table->dpm_levels[0].value; 362 dpm_table->max = dpm_table->dpm_levels[0].value; 363 } 364 365 return 0; 366 } 367 368 static int aldebaran_check_powerplay_table(struct smu_context *smu) 369 { 370 struct smu_table_context *table_context = &smu->smu_table; 371 struct smu_13_0_powerplay_table *powerplay_table = 372 table_context->power_play_table; 373 374 table_context->thermal_controller_type = 375 powerplay_table->thermal_controller_type; 376 377 return 0; 378 } 379 380 static int aldebaran_store_powerplay_table(struct smu_context *smu) 381 { 382 struct smu_table_context *table_context = &smu->smu_table; 383 struct smu_13_0_powerplay_table *powerplay_table = 384 table_context->power_play_table; 385 memcpy(table_context->driver_pptable, &powerplay_table->smc_pptable, 386 sizeof(PPTable_t)); 387 388 return 0; 389 } 390 391 static int aldebaran_append_powerplay_table(struct smu_context *smu) 392 { 393 struct smu_table_context *table_context = &smu->smu_table; 394 PPTable_t *smc_pptable = table_context->driver_pptable; 395 struct atom_smc_dpm_info_v4_10 *smc_dpm_table; 396 int index, ret; 397 398 index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1, 399 smc_dpm_info); 400 401 ret = amdgpu_atombios_get_data_table(smu->adev, index, NULL, NULL, NULL, 402 (uint8_t **)&smc_dpm_table); 403 if (ret) 404 return ret; 405 406 dev_info(smu->adev->dev, "smc_dpm_info table revision(format.content): %d.%d\n", 407 smc_dpm_table->table_header.format_revision, 408 smc_dpm_table->table_header.content_revision); 409 410 if ((smc_dpm_table->table_header.format_revision == 4) && 411 (smc_dpm_table->table_header.content_revision == 10)) 412 smu_memcpy_trailing(smc_pptable, GfxMaxCurrent, reserved, 413 smc_dpm_table, GfxMaxCurrent); 414 return 0; 415 } 416 417 static int aldebaran_setup_pptable(struct smu_context *smu) 418 { 419 int ret = 0; 420 421 /* VBIOS pptable is the first choice */ 422 smu->smu_table.boot_values.pp_table_id = 0; 423 424 ret = smu_v13_0_setup_pptable(smu); 425 if (ret) 426 return ret; 427 428 ret = aldebaran_store_powerplay_table(smu); 429 if (ret) 430 return ret; 431 432 ret = aldebaran_append_powerplay_table(smu); 433 if (ret) 434 return ret; 435 436 ret = aldebaran_check_powerplay_table(smu); 437 if (ret) 438 return ret; 439 440 return ret; 441 } 442 443 static bool aldebaran_is_primary(struct smu_context *smu) 444 { 445 struct amdgpu_device *adev = smu->adev; 446 447 if (adev->smuio.funcs && adev->smuio.funcs->get_die_id) 448 return adev->smuio.funcs->get_die_id(adev) == 0; 449 450 return true; 451 } 452 453 static int aldebaran_run_board_btc(struct smu_context *smu) 454 { 455 u32 smu_version; 456 int ret; 457 458 if (!aldebaran_is_primary(smu)) 459 return 0; 460 461 ret = smu_cmn_get_smc_version(smu, NULL, &smu_version); 462 if (ret) { 463 dev_err(smu->adev->dev, "Failed to get smu version!\n"); 464 return ret; 465 } 466 if (smu_version <= 0x00441d00) 467 return 0; 468 469 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_BoardPowerCalibration, NULL); 470 if (ret) 471 dev_err(smu->adev->dev, "Board power calibration failed!\n"); 472 473 return ret; 474 } 475 476 static int aldebaran_run_btc(struct smu_context *smu) 477 { 478 int ret; 479 480 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_RunDcBtc, NULL); 481 if (ret) 482 dev_err(smu->adev->dev, "RunDcBtc failed!\n"); 483 else 484 ret = aldebaran_run_board_btc(smu); 485 486 return ret; 487 } 488 489 static int aldebaran_populate_umd_state_clk(struct smu_context *smu) 490 { 491 struct smu_13_0_dpm_context *dpm_context = 492 smu->smu_dpm.dpm_context; 493 struct smu_13_0_dpm_table *gfx_table = 494 &dpm_context->dpm_tables.gfx_table; 495 struct smu_13_0_dpm_table *mem_table = 496 &dpm_context->dpm_tables.uclk_table; 497 struct smu_13_0_dpm_table *soc_table = 498 &dpm_context->dpm_tables.soc_table; 499 struct smu_umd_pstate_table *pstate_table = 500 &smu->pstate_table; 501 502 pstate_table->gfxclk_pstate.min = gfx_table->min; 503 pstate_table->gfxclk_pstate.peak = gfx_table->max; 504 pstate_table->gfxclk_pstate.curr.min = gfx_table->min; 505 pstate_table->gfxclk_pstate.curr.max = gfx_table->max; 506 507 pstate_table->uclk_pstate.min = mem_table->min; 508 pstate_table->uclk_pstate.peak = mem_table->max; 509 pstate_table->uclk_pstate.curr.min = mem_table->min; 510 pstate_table->uclk_pstate.curr.max = mem_table->max; 511 512 pstate_table->socclk_pstate.min = soc_table->min; 513 pstate_table->socclk_pstate.peak = soc_table->max; 514 pstate_table->socclk_pstate.curr.min = soc_table->min; 515 pstate_table->socclk_pstate.curr.max = soc_table->max; 516 517 if (gfx_table->count > ALDEBARAN_UMD_PSTATE_GFXCLK_LEVEL && 518 mem_table->count > ALDEBARAN_UMD_PSTATE_MCLK_LEVEL && 519 soc_table->count > ALDEBARAN_UMD_PSTATE_SOCCLK_LEVEL) { 520 pstate_table->gfxclk_pstate.standard = 521 gfx_table->dpm_levels[ALDEBARAN_UMD_PSTATE_GFXCLK_LEVEL].value; 522 pstate_table->uclk_pstate.standard = 523 mem_table->dpm_levels[ALDEBARAN_UMD_PSTATE_MCLK_LEVEL].value; 524 pstate_table->socclk_pstate.standard = 525 soc_table->dpm_levels[ALDEBARAN_UMD_PSTATE_SOCCLK_LEVEL].value; 526 } else { 527 pstate_table->gfxclk_pstate.standard = 528 pstate_table->gfxclk_pstate.min; 529 pstate_table->uclk_pstate.standard = 530 pstate_table->uclk_pstate.min; 531 pstate_table->socclk_pstate.standard = 532 pstate_table->socclk_pstate.min; 533 } 534 535 return 0; 536 } 537 538 static int aldebaran_get_clk_table(struct smu_context *smu, 539 struct pp_clock_levels_with_latency *clocks, 540 struct smu_13_0_dpm_table *dpm_table) 541 { 542 int i, count; 543 544 count = (dpm_table->count > MAX_NUM_CLOCKS) ? MAX_NUM_CLOCKS : dpm_table->count; 545 clocks->num_levels = count; 546 547 for (i = 0; i < count; i++) { 548 clocks->data[i].clocks_in_khz = 549 dpm_table->dpm_levels[i].value * 1000; 550 clocks->data[i].latency_in_us = 0; 551 } 552 553 return 0; 554 } 555 556 static int aldebaran_freqs_in_same_level(int32_t frequency1, 557 int32_t frequency2) 558 { 559 return (abs(frequency1 - frequency2) <= EPSILON); 560 } 561 562 static int aldebaran_get_smu_metrics_data(struct smu_context *smu, 563 MetricsMember_t member, 564 uint32_t *value) 565 { 566 struct smu_table_context *smu_table= &smu->smu_table; 567 SmuMetrics_t *metrics = (SmuMetrics_t *)smu_table->metrics_table; 568 int ret = 0; 569 570 mutex_lock(&smu->metrics_lock); 571 572 ret = smu_cmn_get_metrics_table_locked(smu, 573 NULL, 574 false); 575 if (ret) { 576 mutex_unlock(&smu->metrics_lock); 577 return ret; 578 } 579 580 switch (member) { 581 case METRICS_CURR_GFXCLK: 582 *value = metrics->CurrClock[PPCLK_GFXCLK]; 583 break; 584 case METRICS_CURR_SOCCLK: 585 *value = metrics->CurrClock[PPCLK_SOCCLK]; 586 break; 587 case METRICS_CURR_UCLK: 588 *value = metrics->CurrClock[PPCLK_UCLK]; 589 break; 590 case METRICS_CURR_VCLK: 591 *value = metrics->CurrClock[PPCLK_VCLK]; 592 break; 593 case METRICS_CURR_DCLK: 594 *value = metrics->CurrClock[PPCLK_DCLK]; 595 break; 596 case METRICS_CURR_FCLK: 597 *value = metrics->CurrClock[PPCLK_FCLK]; 598 break; 599 case METRICS_AVERAGE_GFXCLK: 600 *value = metrics->AverageGfxclkFrequency; 601 break; 602 case METRICS_AVERAGE_SOCCLK: 603 *value = metrics->AverageSocclkFrequency; 604 break; 605 case METRICS_AVERAGE_UCLK: 606 *value = metrics->AverageUclkFrequency; 607 break; 608 case METRICS_AVERAGE_GFXACTIVITY: 609 *value = metrics->AverageGfxActivity; 610 break; 611 case METRICS_AVERAGE_MEMACTIVITY: 612 *value = metrics->AverageUclkActivity; 613 break; 614 case METRICS_AVERAGE_SOCKETPOWER: 615 /* Valid power data is available only from primary die */ 616 *value = aldebaran_is_primary(smu) ? 617 metrics->AverageSocketPower << 8 : 618 0; 619 break; 620 case METRICS_TEMPERATURE_EDGE: 621 *value = metrics->TemperatureEdge * 622 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 623 break; 624 case METRICS_TEMPERATURE_HOTSPOT: 625 *value = metrics->TemperatureHotspot * 626 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 627 break; 628 case METRICS_TEMPERATURE_MEM: 629 *value = metrics->TemperatureHBM * 630 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 631 break; 632 case METRICS_TEMPERATURE_VRGFX: 633 *value = metrics->TemperatureVrGfx * 634 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 635 break; 636 case METRICS_TEMPERATURE_VRSOC: 637 *value = metrics->TemperatureVrSoc * 638 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 639 break; 640 case METRICS_TEMPERATURE_VRMEM: 641 *value = metrics->TemperatureVrMem * 642 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 643 break; 644 case METRICS_THROTTLER_STATUS: 645 *value = metrics->ThrottlerStatus; 646 break; 647 default: 648 *value = UINT_MAX; 649 break; 650 } 651 652 mutex_unlock(&smu->metrics_lock); 653 654 return ret; 655 } 656 657 static int aldebaran_get_current_clk_freq_by_table(struct smu_context *smu, 658 enum smu_clk_type clk_type, 659 uint32_t *value) 660 { 661 MetricsMember_t member_type; 662 int clk_id = 0; 663 664 if (!value) 665 return -EINVAL; 666 667 clk_id = smu_cmn_to_asic_specific_index(smu, 668 CMN2ASIC_MAPPING_CLK, 669 clk_type); 670 if (clk_id < 0) 671 return -EINVAL; 672 673 switch (clk_id) { 674 case PPCLK_GFXCLK: 675 /* 676 * CurrClock[clk_id] can provide accurate 677 * output only when the dpm feature is enabled. 678 * We can use Average_* for dpm disabled case. 679 * But this is available for gfxclk/uclk/socclk/vclk/dclk. 680 */ 681 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT)) 682 member_type = METRICS_CURR_GFXCLK; 683 else 684 member_type = METRICS_AVERAGE_GFXCLK; 685 break; 686 case PPCLK_UCLK: 687 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT)) 688 member_type = METRICS_CURR_UCLK; 689 else 690 member_type = METRICS_AVERAGE_UCLK; 691 break; 692 case PPCLK_SOCCLK: 693 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT)) 694 member_type = METRICS_CURR_SOCCLK; 695 else 696 member_type = METRICS_AVERAGE_SOCCLK; 697 break; 698 case PPCLK_VCLK: 699 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) 700 member_type = METRICS_CURR_VCLK; 701 else 702 member_type = METRICS_AVERAGE_VCLK; 703 break; 704 case PPCLK_DCLK: 705 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) 706 member_type = METRICS_CURR_DCLK; 707 else 708 member_type = METRICS_AVERAGE_DCLK; 709 break; 710 case PPCLK_FCLK: 711 member_type = METRICS_CURR_FCLK; 712 break; 713 default: 714 return -EINVAL; 715 } 716 717 return aldebaran_get_smu_metrics_data(smu, 718 member_type, 719 value); 720 } 721 722 static int aldebaran_print_clk_levels(struct smu_context *smu, 723 enum smu_clk_type type, char *buf) 724 { 725 int i, now, size = 0; 726 int ret = 0; 727 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 728 struct pp_clock_levels_with_latency clocks; 729 struct smu_13_0_dpm_table *single_dpm_table; 730 struct smu_dpm_context *smu_dpm = &smu->smu_dpm; 731 struct smu_13_0_dpm_context *dpm_context = NULL; 732 uint32_t display_levels; 733 uint32_t freq_values[3] = {0}; 734 uint32_t min_clk, max_clk; 735 736 smu_cmn_get_sysfs_buf(&buf, &size); 737 738 if (amdgpu_ras_intr_triggered()) { 739 size += sysfs_emit_at(buf, size, "unavailable\n"); 740 return size; 741 } 742 743 dpm_context = smu_dpm->dpm_context; 744 745 switch (type) { 746 747 case SMU_OD_SCLK: 748 size += sysfs_emit_at(buf, size, "%s:\n", "GFXCLK"); 749 fallthrough; 750 case SMU_SCLK: 751 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_GFXCLK, &now); 752 if (ret) { 753 dev_err(smu->adev->dev, "Attempt to get current gfx clk Failed!"); 754 return ret; 755 } 756 757 single_dpm_table = &(dpm_context->dpm_tables.gfx_table); 758 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 759 if (ret) { 760 dev_err(smu->adev->dev, "Attempt to get gfx clk levels Failed!"); 761 return ret; 762 } 763 764 display_levels = clocks.num_levels; 765 766 min_clk = pstate_table->gfxclk_pstate.curr.min; 767 max_clk = pstate_table->gfxclk_pstate.curr.max; 768 769 freq_values[0] = min_clk; 770 freq_values[1] = max_clk; 771 772 /* fine-grained dpm has only 2 levels */ 773 if (now > min_clk && now < max_clk) { 774 display_levels = clocks.num_levels + 1; 775 freq_values[2] = max_clk; 776 freq_values[1] = now; 777 } 778 779 /* 780 * For DPM disabled case, there will be only one clock level. 781 * And it's safe to assume that is always the current clock. 782 */ 783 if (display_levels == clocks.num_levels) { 784 for (i = 0; i < clocks.num_levels; i++) 785 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", i, 786 freq_values[i], 787 (clocks.num_levels == 1) ? 788 "*" : 789 (aldebaran_freqs_in_same_level( 790 freq_values[i], now) ? 791 "*" : 792 "")); 793 } else { 794 for (i = 0; i < display_levels; i++) 795 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", i, 796 freq_values[i], i == 1 ? "*" : ""); 797 } 798 799 break; 800 801 case SMU_OD_MCLK: 802 size += sysfs_emit_at(buf, size, "%s:\n", "MCLK"); 803 fallthrough; 804 case SMU_MCLK: 805 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_UCLK, &now); 806 if (ret) { 807 dev_err(smu->adev->dev, "Attempt to get current mclk Failed!"); 808 return ret; 809 } 810 811 single_dpm_table = &(dpm_context->dpm_tables.uclk_table); 812 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 813 if (ret) { 814 dev_err(smu->adev->dev, "Attempt to get memory clk levels Failed!"); 815 return ret; 816 } 817 818 for (i = 0; i < clocks.num_levels; i++) 819 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 820 i, clocks.data[i].clocks_in_khz / 1000, 821 (clocks.num_levels == 1) ? "*" : 822 (aldebaran_freqs_in_same_level( 823 clocks.data[i].clocks_in_khz / 1000, 824 now) ? "*" : "")); 825 break; 826 827 case SMU_SOCCLK: 828 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_SOCCLK, &now); 829 if (ret) { 830 dev_err(smu->adev->dev, "Attempt to get current socclk Failed!"); 831 return ret; 832 } 833 834 single_dpm_table = &(dpm_context->dpm_tables.soc_table); 835 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 836 if (ret) { 837 dev_err(smu->adev->dev, "Attempt to get socclk levels Failed!"); 838 return ret; 839 } 840 841 for (i = 0; i < clocks.num_levels; i++) 842 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 843 i, clocks.data[i].clocks_in_khz / 1000, 844 (clocks.num_levels == 1) ? "*" : 845 (aldebaran_freqs_in_same_level( 846 clocks.data[i].clocks_in_khz / 1000, 847 now) ? "*" : "")); 848 break; 849 850 case SMU_FCLK: 851 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_FCLK, &now); 852 if (ret) { 853 dev_err(smu->adev->dev, "Attempt to get current fclk Failed!"); 854 return ret; 855 } 856 857 single_dpm_table = &(dpm_context->dpm_tables.fclk_table); 858 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 859 if (ret) { 860 dev_err(smu->adev->dev, "Attempt to get fclk levels Failed!"); 861 return ret; 862 } 863 864 for (i = 0; i < single_dpm_table->count; i++) 865 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 866 i, single_dpm_table->dpm_levels[i].value, 867 (clocks.num_levels == 1) ? "*" : 868 (aldebaran_freqs_in_same_level( 869 clocks.data[i].clocks_in_khz / 1000, 870 now) ? "*" : "")); 871 break; 872 873 case SMU_VCLK: 874 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_VCLK, &now); 875 if (ret) { 876 dev_err(smu->adev->dev, "Attempt to get current vclk Failed!"); 877 return ret; 878 } 879 880 single_dpm_table = &(dpm_context->dpm_tables.vclk_table); 881 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 882 if (ret) { 883 dev_err(smu->adev->dev, "Attempt to get vclk levels Failed!"); 884 return ret; 885 } 886 887 for (i = 0; i < single_dpm_table->count; i++) 888 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 889 i, single_dpm_table->dpm_levels[i].value, 890 (clocks.num_levels == 1) ? "*" : 891 (aldebaran_freqs_in_same_level( 892 clocks.data[i].clocks_in_khz / 1000, 893 now) ? "*" : "")); 894 break; 895 896 case SMU_DCLK: 897 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_DCLK, &now); 898 if (ret) { 899 dev_err(smu->adev->dev, "Attempt to get current dclk Failed!"); 900 return ret; 901 } 902 903 single_dpm_table = &(dpm_context->dpm_tables.dclk_table); 904 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 905 if (ret) { 906 dev_err(smu->adev->dev, "Attempt to get dclk levels Failed!"); 907 return ret; 908 } 909 910 for (i = 0; i < single_dpm_table->count; i++) 911 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", 912 i, single_dpm_table->dpm_levels[i].value, 913 (clocks.num_levels == 1) ? "*" : 914 (aldebaran_freqs_in_same_level( 915 clocks.data[i].clocks_in_khz / 1000, 916 now) ? "*" : "")); 917 break; 918 919 default: 920 break; 921 } 922 923 return size; 924 } 925 926 static int aldebaran_upload_dpm_level(struct smu_context *smu, 927 bool max, 928 uint32_t feature_mask, 929 uint32_t level) 930 { 931 struct smu_13_0_dpm_context *dpm_context = 932 smu->smu_dpm.dpm_context; 933 uint32_t freq; 934 int ret = 0; 935 936 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT) && 937 (feature_mask & FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT))) { 938 freq = dpm_context->dpm_tables.gfx_table.dpm_levels[level].value; 939 ret = smu_cmn_send_smc_msg_with_param(smu, 940 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 941 (PPCLK_GFXCLK << 16) | (freq & 0xffff), 942 NULL); 943 if (ret) { 944 dev_err(smu->adev->dev, "Failed to set soft %s gfxclk !\n", 945 max ? "max" : "min"); 946 return ret; 947 } 948 } 949 950 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT) && 951 (feature_mask & FEATURE_MASK(FEATURE_DPM_UCLK_BIT))) { 952 freq = dpm_context->dpm_tables.uclk_table.dpm_levels[level].value; 953 ret = smu_cmn_send_smc_msg_with_param(smu, 954 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 955 (PPCLK_UCLK << 16) | (freq & 0xffff), 956 NULL); 957 if (ret) { 958 dev_err(smu->adev->dev, "Failed to set soft %s memclk !\n", 959 max ? "max" : "min"); 960 return ret; 961 } 962 } 963 964 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT) && 965 (feature_mask & FEATURE_MASK(FEATURE_DPM_SOCCLK_BIT))) { 966 freq = dpm_context->dpm_tables.soc_table.dpm_levels[level].value; 967 ret = smu_cmn_send_smc_msg_with_param(smu, 968 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 969 (PPCLK_SOCCLK << 16) | (freq & 0xffff), 970 NULL); 971 if (ret) { 972 dev_err(smu->adev->dev, "Failed to set soft %s socclk !\n", 973 max ? "max" : "min"); 974 return ret; 975 } 976 } 977 978 return ret; 979 } 980 981 static int aldebaran_force_clk_levels(struct smu_context *smu, 982 enum smu_clk_type type, uint32_t mask) 983 { 984 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context; 985 struct smu_13_0_dpm_table *single_dpm_table = NULL; 986 uint32_t soft_min_level, soft_max_level; 987 int ret = 0; 988 989 soft_min_level = mask ? (ffs(mask) - 1) : 0; 990 soft_max_level = mask ? (fls(mask) - 1) : 0; 991 992 switch (type) { 993 case SMU_SCLK: 994 single_dpm_table = &(dpm_context->dpm_tables.gfx_table); 995 if (soft_max_level >= single_dpm_table->count) { 996 dev_err(smu->adev->dev, "Clock level specified %d is over max allowed %d\n", 997 soft_max_level, single_dpm_table->count - 1); 998 ret = -EINVAL; 999 break; 1000 } 1001 1002 ret = aldebaran_upload_dpm_level(smu, 1003 false, 1004 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT), 1005 soft_min_level); 1006 if (ret) { 1007 dev_err(smu->adev->dev, "Failed to upload boot level to lowest!\n"); 1008 break; 1009 } 1010 1011 ret = aldebaran_upload_dpm_level(smu, 1012 true, 1013 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT), 1014 soft_max_level); 1015 if (ret) 1016 dev_err(smu->adev->dev, "Failed to upload dpm max level to highest!\n"); 1017 1018 break; 1019 1020 case SMU_MCLK: 1021 case SMU_SOCCLK: 1022 case SMU_FCLK: 1023 /* 1024 * Should not arrive here since aldebaran does not 1025 * support mclk/socclk/fclk softmin/softmax settings 1026 */ 1027 ret = -EINVAL; 1028 break; 1029 1030 default: 1031 break; 1032 } 1033 1034 return ret; 1035 } 1036 1037 static int aldebaran_get_thermal_temperature_range(struct smu_context *smu, 1038 struct smu_temperature_range *range) 1039 { 1040 struct smu_table_context *table_context = &smu->smu_table; 1041 struct smu_13_0_powerplay_table *powerplay_table = 1042 table_context->power_play_table; 1043 PPTable_t *pptable = smu->smu_table.driver_pptable; 1044 1045 if (!range) 1046 return -EINVAL; 1047 1048 memcpy(range, &smu13_thermal_policy[0], sizeof(struct smu_temperature_range)); 1049 1050 range->hotspot_crit_max = pptable->ThotspotLimit * 1051 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1052 range->hotspot_emergency_max = (pptable->ThotspotLimit + CTF_OFFSET_HOTSPOT) * 1053 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1054 range->mem_crit_max = pptable->TmemLimit * 1055 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1056 range->mem_emergency_max = (pptable->TmemLimit + CTF_OFFSET_MEM)* 1057 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1058 range->software_shutdown_temp = powerplay_table->software_shutdown_temp; 1059 1060 return 0; 1061 } 1062 1063 static int aldebaran_get_current_activity_percent(struct smu_context *smu, 1064 enum amd_pp_sensors sensor, 1065 uint32_t *value) 1066 { 1067 int ret = 0; 1068 1069 if (!value) 1070 return -EINVAL; 1071 1072 switch (sensor) { 1073 case AMDGPU_PP_SENSOR_GPU_LOAD: 1074 ret = aldebaran_get_smu_metrics_data(smu, 1075 METRICS_AVERAGE_GFXACTIVITY, 1076 value); 1077 break; 1078 case AMDGPU_PP_SENSOR_MEM_LOAD: 1079 ret = aldebaran_get_smu_metrics_data(smu, 1080 METRICS_AVERAGE_MEMACTIVITY, 1081 value); 1082 break; 1083 default: 1084 dev_err(smu->adev->dev, "Invalid sensor for retrieving clock activity\n"); 1085 return -EINVAL; 1086 } 1087 1088 return ret; 1089 } 1090 1091 static int aldebaran_get_gpu_power(struct smu_context *smu, uint32_t *value) 1092 { 1093 if (!value) 1094 return -EINVAL; 1095 1096 return aldebaran_get_smu_metrics_data(smu, 1097 METRICS_AVERAGE_SOCKETPOWER, 1098 value); 1099 } 1100 1101 static int aldebaran_thermal_get_temperature(struct smu_context *smu, 1102 enum amd_pp_sensors sensor, 1103 uint32_t *value) 1104 { 1105 int ret = 0; 1106 1107 if (!value) 1108 return -EINVAL; 1109 1110 switch (sensor) { 1111 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP: 1112 ret = aldebaran_get_smu_metrics_data(smu, 1113 METRICS_TEMPERATURE_HOTSPOT, 1114 value); 1115 break; 1116 case AMDGPU_PP_SENSOR_EDGE_TEMP: 1117 ret = aldebaran_get_smu_metrics_data(smu, 1118 METRICS_TEMPERATURE_EDGE, 1119 value); 1120 break; 1121 case AMDGPU_PP_SENSOR_MEM_TEMP: 1122 ret = aldebaran_get_smu_metrics_data(smu, 1123 METRICS_TEMPERATURE_MEM, 1124 value); 1125 break; 1126 default: 1127 dev_err(smu->adev->dev, "Invalid sensor for retrieving temp\n"); 1128 return -EINVAL; 1129 } 1130 1131 return ret; 1132 } 1133 1134 static int aldebaran_read_sensor(struct smu_context *smu, 1135 enum amd_pp_sensors sensor, 1136 void *data, uint32_t *size) 1137 { 1138 int ret = 0; 1139 1140 if (amdgpu_ras_intr_triggered()) 1141 return 0; 1142 1143 if (!data || !size) 1144 return -EINVAL; 1145 1146 mutex_lock(&smu->sensor_lock); 1147 switch (sensor) { 1148 case AMDGPU_PP_SENSOR_MEM_LOAD: 1149 case AMDGPU_PP_SENSOR_GPU_LOAD: 1150 ret = aldebaran_get_current_activity_percent(smu, 1151 sensor, 1152 (uint32_t *)data); 1153 *size = 4; 1154 break; 1155 case AMDGPU_PP_SENSOR_GPU_POWER: 1156 ret = aldebaran_get_gpu_power(smu, (uint32_t *)data); 1157 *size = 4; 1158 break; 1159 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP: 1160 case AMDGPU_PP_SENSOR_EDGE_TEMP: 1161 case AMDGPU_PP_SENSOR_MEM_TEMP: 1162 ret = aldebaran_thermal_get_temperature(smu, sensor, 1163 (uint32_t *)data); 1164 *size = 4; 1165 break; 1166 case AMDGPU_PP_SENSOR_GFX_MCLK: 1167 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_UCLK, (uint32_t *)data); 1168 /* the output clock frequency in 10K unit */ 1169 *(uint32_t *)data *= 100; 1170 *size = 4; 1171 break; 1172 case AMDGPU_PP_SENSOR_GFX_SCLK: 1173 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_GFXCLK, (uint32_t *)data); 1174 *(uint32_t *)data *= 100; 1175 *size = 4; 1176 break; 1177 case AMDGPU_PP_SENSOR_VDDGFX: 1178 ret = smu_v13_0_get_gfx_vdd(smu, (uint32_t *)data); 1179 *size = 4; 1180 break; 1181 default: 1182 ret = -EOPNOTSUPP; 1183 break; 1184 } 1185 mutex_unlock(&smu->sensor_lock); 1186 1187 return ret; 1188 } 1189 1190 static int aldebaran_get_power_limit(struct smu_context *smu, 1191 uint32_t *current_power_limit, 1192 uint32_t *default_power_limit, 1193 uint32_t *max_power_limit) 1194 { 1195 PPTable_t *pptable = smu->smu_table.driver_pptable; 1196 uint32_t power_limit = 0; 1197 int ret; 1198 1199 if (!smu_cmn_feature_is_enabled(smu, SMU_FEATURE_PPT_BIT)) { 1200 if (current_power_limit) 1201 *current_power_limit = 0; 1202 if (default_power_limit) 1203 *default_power_limit = 0; 1204 if (max_power_limit) 1205 *max_power_limit = 0; 1206 1207 dev_warn(smu->adev->dev, 1208 "PPT feature is not enabled, power values can't be fetched."); 1209 1210 return 0; 1211 } 1212 1213 /* Valid power data is available only from primary die. 1214 * For secondary die show the value as 0. 1215 */ 1216 if (aldebaran_is_primary(smu)) { 1217 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_GetPptLimit, 1218 &power_limit); 1219 1220 if (ret) { 1221 /* the last hope to figure out the ppt limit */ 1222 if (!pptable) { 1223 dev_err(smu->adev->dev, 1224 "Cannot get PPT limit due to pptable missing!"); 1225 return -EINVAL; 1226 } 1227 power_limit = pptable->PptLimit; 1228 } 1229 } 1230 1231 if (current_power_limit) 1232 *current_power_limit = power_limit; 1233 if (default_power_limit) 1234 *default_power_limit = power_limit; 1235 1236 if (max_power_limit) { 1237 if (pptable) 1238 *max_power_limit = pptable->PptLimit; 1239 } 1240 1241 return 0; 1242 } 1243 1244 static int aldebaran_set_power_limit(struct smu_context *smu, 1245 enum smu_ppt_limit_type limit_type, 1246 uint32_t limit) 1247 { 1248 /* Power limit can be set only through primary die */ 1249 if (aldebaran_is_primary(smu)) 1250 return smu_v13_0_set_power_limit(smu, limit_type, limit); 1251 1252 return -EINVAL; 1253 } 1254 1255 static int aldebaran_system_features_control(struct smu_context *smu, bool enable) 1256 { 1257 int ret; 1258 1259 ret = smu_v13_0_system_features_control(smu, enable); 1260 if (!ret && enable) 1261 ret = aldebaran_run_btc(smu); 1262 1263 return ret; 1264 } 1265 1266 static int aldebaran_set_performance_level(struct smu_context *smu, 1267 enum amd_dpm_forced_level level) 1268 { 1269 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1270 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1271 struct smu_13_0_dpm_table *gfx_table = 1272 &dpm_context->dpm_tables.gfx_table; 1273 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1274 1275 /* Disable determinism if switching to another mode */ 1276 if ((smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) && 1277 (level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) { 1278 smu_cmn_send_smc_msg(smu, SMU_MSG_DisableDeterminism, NULL); 1279 pstate_table->gfxclk_pstate.curr.max = gfx_table->max; 1280 } 1281 1282 switch (level) { 1283 1284 case AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM: 1285 return 0; 1286 1287 case AMD_DPM_FORCED_LEVEL_HIGH: 1288 case AMD_DPM_FORCED_LEVEL_LOW: 1289 case AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD: 1290 case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK: 1291 case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK: 1292 case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK: 1293 default: 1294 break; 1295 } 1296 1297 return smu_v13_0_set_performance_level(smu, level); 1298 } 1299 1300 static int aldebaran_set_soft_freq_limited_range(struct smu_context *smu, 1301 enum smu_clk_type clk_type, 1302 uint32_t min, 1303 uint32_t max) 1304 { 1305 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1306 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1307 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1308 struct amdgpu_device *adev = smu->adev; 1309 uint32_t min_clk; 1310 uint32_t max_clk; 1311 int ret = 0; 1312 1313 if (clk_type != SMU_GFXCLK && clk_type != SMU_SCLK) 1314 return -EINVAL; 1315 1316 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) 1317 && (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) 1318 return -EINVAL; 1319 1320 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_MANUAL) { 1321 if (min >= max) { 1322 dev_err(smu->adev->dev, 1323 "Minimum GFX clk should be less than the maximum allowed clock\n"); 1324 return -EINVAL; 1325 } 1326 1327 if ((min == pstate_table->gfxclk_pstate.curr.min) && 1328 (max == pstate_table->gfxclk_pstate.curr.max)) 1329 return 0; 1330 1331 ret = smu_v13_0_set_soft_freq_limited_range(smu, SMU_GFXCLK, 1332 min, max); 1333 if (!ret) { 1334 pstate_table->gfxclk_pstate.curr.min = min; 1335 pstate_table->gfxclk_pstate.curr.max = max; 1336 } 1337 1338 return ret; 1339 } 1340 1341 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) { 1342 if (!max || (max < dpm_context->dpm_tables.gfx_table.min) || 1343 (max > dpm_context->dpm_tables.gfx_table.max)) { 1344 dev_warn(adev->dev, 1345 "Invalid max frequency %d MHz specified for determinism\n", max); 1346 return -EINVAL; 1347 } 1348 1349 /* Restore default min/max clocks and enable determinism */ 1350 min_clk = dpm_context->dpm_tables.gfx_table.min; 1351 max_clk = dpm_context->dpm_tables.gfx_table.max; 1352 ret = smu_v13_0_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1353 if (!ret) { 1354 usleep_range(500, 1000); 1355 ret = smu_cmn_send_smc_msg_with_param(smu, 1356 SMU_MSG_EnableDeterminism, 1357 max, NULL); 1358 if (ret) { 1359 dev_err(adev->dev, 1360 "Failed to enable determinism at GFX clock %d MHz\n", max); 1361 } else { 1362 pstate_table->gfxclk_pstate.curr.min = min_clk; 1363 pstate_table->gfxclk_pstate.curr.max = max; 1364 } 1365 } 1366 } 1367 1368 return ret; 1369 } 1370 1371 static int aldebaran_usr_edit_dpm_table(struct smu_context *smu, enum PP_OD_DPM_TABLE_COMMAND type, 1372 long input[], uint32_t size) 1373 { 1374 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1375 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1376 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1377 uint32_t min_clk; 1378 uint32_t max_clk; 1379 int ret = 0; 1380 1381 /* Only allowed in manual or determinism mode */ 1382 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) 1383 && (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) 1384 return -EINVAL; 1385 1386 switch (type) { 1387 case PP_OD_EDIT_SCLK_VDDC_TABLE: 1388 if (size != 2) { 1389 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1390 return -EINVAL; 1391 } 1392 1393 if (input[0] == 0) { 1394 if (input[1] < dpm_context->dpm_tables.gfx_table.min) { 1395 dev_warn(smu->adev->dev, "Minimum GFX clk (%ld) MHz specified is less than the minimum allowed (%d) MHz\n", 1396 input[1], dpm_context->dpm_tables.gfx_table.min); 1397 pstate_table->gfxclk_pstate.custom.min = 1398 pstate_table->gfxclk_pstate.curr.min; 1399 return -EINVAL; 1400 } 1401 1402 pstate_table->gfxclk_pstate.custom.min = input[1]; 1403 } else if (input[0] == 1) { 1404 if (input[1] > dpm_context->dpm_tables.gfx_table.max) { 1405 dev_warn(smu->adev->dev, "Maximum GFX clk (%ld) MHz specified is greater than the maximum allowed (%d) MHz\n", 1406 input[1], dpm_context->dpm_tables.gfx_table.max); 1407 pstate_table->gfxclk_pstate.custom.max = 1408 pstate_table->gfxclk_pstate.curr.max; 1409 return -EINVAL; 1410 } 1411 1412 pstate_table->gfxclk_pstate.custom.max = input[1]; 1413 } else { 1414 return -EINVAL; 1415 } 1416 break; 1417 case PP_OD_RESTORE_DEFAULT_TABLE: 1418 if (size != 0) { 1419 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1420 return -EINVAL; 1421 } else { 1422 /* Use the default frequencies for manual and determinism mode */ 1423 min_clk = dpm_context->dpm_tables.gfx_table.min; 1424 max_clk = dpm_context->dpm_tables.gfx_table.max; 1425 1426 return aldebaran_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1427 } 1428 break; 1429 case PP_OD_COMMIT_DPM_TABLE: 1430 if (size != 0) { 1431 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1432 return -EINVAL; 1433 } else { 1434 if (!pstate_table->gfxclk_pstate.custom.min) 1435 pstate_table->gfxclk_pstate.custom.min = 1436 pstate_table->gfxclk_pstate.curr.min; 1437 1438 if (!pstate_table->gfxclk_pstate.custom.max) 1439 pstate_table->gfxclk_pstate.custom.max = 1440 pstate_table->gfxclk_pstate.curr.max; 1441 1442 min_clk = pstate_table->gfxclk_pstate.custom.min; 1443 max_clk = pstate_table->gfxclk_pstate.custom.max; 1444 1445 return aldebaran_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1446 } 1447 break; 1448 default: 1449 return -ENOSYS; 1450 } 1451 1452 return ret; 1453 } 1454 1455 static bool aldebaran_is_dpm_running(struct smu_context *smu) 1456 { 1457 int ret; 1458 uint32_t feature_mask[2]; 1459 unsigned long feature_enabled; 1460 1461 ret = smu_cmn_get_enabled_mask(smu, feature_mask, 2); 1462 if (ret) 1463 return false; 1464 feature_enabled = (unsigned long)((uint64_t)feature_mask[0] | 1465 ((uint64_t)feature_mask[1] << 32)); 1466 return !!(feature_enabled & SMC_DPM_FEATURE); 1467 } 1468 1469 static int aldebaran_i2c_xfer(struct i2c_adapter *i2c_adap, 1470 struct i2c_msg *msg, int num_msgs) 1471 { 1472 struct amdgpu_device *adev = to_amdgpu_device(i2c_adap); 1473 struct smu_table_context *smu_table = &adev->smu.smu_table; 1474 struct smu_table *table = &smu_table->driver_table; 1475 SwI2cRequest_t *req, *res = (SwI2cRequest_t *)table->cpu_addr; 1476 int i, j, r, c; 1477 u16 dir; 1478 1479 req = kzalloc(sizeof(*req), GFP_KERNEL); 1480 if (!req) 1481 return -ENOMEM; 1482 1483 req->I2CcontrollerPort = 0; 1484 req->I2CSpeed = I2C_SPEED_FAST_400K; 1485 req->SlaveAddress = msg[0].addr << 1; /* wants an 8-bit address */ 1486 dir = msg[0].flags & I2C_M_RD; 1487 1488 for (c = i = 0; i < num_msgs; i++) { 1489 for (j = 0; j < msg[i].len; j++, c++) { 1490 SwI2cCmd_t *cmd = &req->SwI2cCmds[c]; 1491 1492 if (!(msg[i].flags & I2C_M_RD)) { 1493 /* write */ 1494 cmd->CmdConfig |= CMDCONFIG_READWRITE_MASK; 1495 cmd->ReadWriteData = msg[i].buf[j]; 1496 } 1497 1498 if ((dir ^ msg[i].flags) & I2C_M_RD) { 1499 /* The direction changes. 1500 */ 1501 dir = msg[i].flags & I2C_M_RD; 1502 cmd->CmdConfig |= CMDCONFIG_RESTART_MASK; 1503 } 1504 1505 req->NumCmds++; 1506 1507 /* 1508 * Insert STOP if we are at the last byte of either last 1509 * message for the transaction or the client explicitly 1510 * requires a STOP at this particular message. 1511 */ 1512 if ((j == msg[i].len - 1) && 1513 ((i == num_msgs - 1) || (msg[i].flags & I2C_M_STOP))) { 1514 cmd->CmdConfig &= ~CMDCONFIG_RESTART_MASK; 1515 cmd->CmdConfig |= CMDCONFIG_STOP_MASK; 1516 } 1517 } 1518 } 1519 mutex_lock(&adev->smu.mutex); 1520 r = smu_cmn_update_table(&adev->smu, SMU_TABLE_I2C_COMMANDS, 0, req, true); 1521 mutex_unlock(&adev->smu.mutex); 1522 if (r) 1523 goto fail; 1524 1525 for (c = i = 0; i < num_msgs; i++) { 1526 if (!(msg[i].flags & I2C_M_RD)) { 1527 c += msg[i].len; 1528 continue; 1529 } 1530 for (j = 0; j < msg[i].len; j++, c++) { 1531 SwI2cCmd_t *cmd = &res->SwI2cCmds[c]; 1532 1533 msg[i].buf[j] = cmd->ReadWriteData; 1534 } 1535 } 1536 r = num_msgs; 1537 fail: 1538 kfree(req); 1539 return r; 1540 } 1541 1542 static u32 aldebaran_i2c_func(struct i2c_adapter *adap) 1543 { 1544 return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL; 1545 } 1546 1547 1548 static const struct i2c_algorithm aldebaran_i2c_algo = { 1549 .master_xfer = aldebaran_i2c_xfer, 1550 .functionality = aldebaran_i2c_func, 1551 }; 1552 1553 static const struct i2c_adapter_quirks aldebaran_i2c_control_quirks = { 1554 .flags = I2C_AQ_COMB | I2C_AQ_COMB_SAME_ADDR | I2C_AQ_NO_ZERO_LEN, 1555 .max_read_len = MAX_SW_I2C_COMMANDS, 1556 .max_write_len = MAX_SW_I2C_COMMANDS, 1557 .max_comb_1st_msg_len = 2, 1558 .max_comb_2nd_msg_len = MAX_SW_I2C_COMMANDS - 2, 1559 }; 1560 1561 static int aldebaran_i2c_control_init(struct smu_context *smu, struct i2c_adapter *control) 1562 { 1563 struct amdgpu_device *adev = to_amdgpu_device(control); 1564 int res; 1565 1566 control->owner = THIS_MODULE; 1567 control->class = I2C_CLASS_SPD; 1568 control->dev.parent = &adev->pdev->dev; 1569 control->algo = &aldebaran_i2c_algo; 1570 snprintf(control->name, sizeof(control->name), "AMDGPU SMU"); 1571 control->quirks = &aldebaran_i2c_control_quirks; 1572 1573 res = i2c_add_adapter(control); 1574 if (res) 1575 DRM_ERROR("Failed to register hw i2c, err: %d\n", res); 1576 1577 return res; 1578 } 1579 1580 static void aldebaran_i2c_control_fini(struct smu_context *smu, struct i2c_adapter *control) 1581 { 1582 i2c_del_adapter(control); 1583 } 1584 1585 static void aldebaran_get_unique_id(struct smu_context *smu) 1586 { 1587 struct amdgpu_device *adev = smu->adev; 1588 SmuMetrics_t *metrics = smu->smu_table.metrics_table; 1589 uint32_t upper32 = 0, lower32 = 0; 1590 int ret; 1591 1592 mutex_lock(&smu->metrics_lock); 1593 ret = smu_cmn_get_metrics_table_locked(smu, NULL, false); 1594 if (ret) 1595 goto out_unlock; 1596 1597 upper32 = metrics->PublicSerialNumUpper32; 1598 lower32 = metrics->PublicSerialNumLower32; 1599 1600 out_unlock: 1601 mutex_unlock(&smu->metrics_lock); 1602 1603 adev->unique_id = ((uint64_t)upper32 << 32) | lower32; 1604 sprintf(adev->serial, "%016llx", adev->unique_id); 1605 } 1606 1607 static bool aldebaran_is_baco_supported(struct smu_context *smu) 1608 { 1609 /* aldebaran is not support baco */ 1610 1611 return false; 1612 } 1613 1614 static int aldebaran_set_df_cstate(struct smu_context *smu, 1615 enum pp_df_cstate state) 1616 { 1617 return smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_DFCstateControl, state, NULL); 1618 } 1619 1620 static int aldebaran_allow_xgmi_power_down(struct smu_context *smu, bool en) 1621 { 1622 return smu_cmn_send_smc_msg_with_param(smu, 1623 SMU_MSG_GmiPwrDnControl, 1624 en ? 1 : 0, 1625 NULL); 1626 } 1627 1628 static const struct throttling_logging_label { 1629 uint32_t feature_mask; 1630 const char *label; 1631 } logging_label[] = { 1632 {(1U << THROTTLER_TEMP_MEM_BIT), "HBM"}, 1633 {(1U << THROTTLER_TEMP_VR_GFX_BIT), "VR of GFX rail"}, 1634 {(1U << THROTTLER_TEMP_VR_MEM_BIT), "VR of HBM rail"}, 1635 {(1U << THROTTLER_TEMP_VR_SOC_BIT), "VR of SOC rail"}, 1636 }; 1637 static void aldebaran_log_thermal_throttling_event(struct smu_context *smu) 1638 { 1639 int ret; 1640 int throttler_idx, throtting_events = 0, buf_idx = 0; 1641 struct amdgpu_device *adev = smu->adev; 1642 uint32_t throttler_status; 1643 char log_buf[256]; 1644 1645 ret = aldebaran_get_smu_metrics_data(smu, 1646 METRICS_THROTTLER_STATUS, 1647 &throttler_status); 1648 if (ret) 1649 return; 1650 1651 memset(log_buf, 0, sizeof(log_buf)); 1652 for (throttler_idx = 0; throttler_idx < ARRAY_SIZE(logging_label); 1653 throttler_idx++) { 1654 if (throttler_status & logging_label[throttler_idx].feature_mask) { 1655 throtting_events++; 1656 buf_idx += snprintf(log_buf + buf_idx, 1657 sizeof(log_buf) - buf_idx, 1658 "%s%s", 1659 throtting_events > 1 ? " and " : "", 1660 logging_label[throttler_idx].label); 1661 if (buf_idx >= sizeof(log_buf)) { 1662 dev_err(adev->dev, "buffer overflow!\n"); 1663 log_buf[sizeof(log_buf) - 1] = '\0'; 1664 break; 1665 } 1666 } 1667 } 1668 1669 dev_warn(adev->dev, "WARN: GPU thermal throttling temperature reached, expect performance decrease. %s.\n", 1670 log_buf); 1671 kgd2kfd_smi_event_throttle(smu->adev->kfd.dev, 1672 smu_cmn_get_indep_throttler_status(throttler_status, 1673 aldebaran_throttler_map)); 1674 } 1675 1676 static int aldebaran_get_current_pcie_link_speed(struct smu_context *smu) 1677 { 1678 struct amdgpu_device *adev = smu->adev; 1679 uint32_t esm_ctrl; 1680 1681 /* TODO: confirm this on real target */ 1682 esm_ctrl = RREG32_PCIE(smnPCIE_ESM_CTRL); 1683 if ((esm_ctrl >> 15) & 0x1FFFF) 1684 return (((esm_ctrl >> 8) & 0x3F) + 128); 1685 1686 return smu_v13_0_get_current_pcie_link_speed(smu); 1687 } 1688 1689 static ssize_t aldebaran_get_gpu_metrics(struct smu_context *smu, 1690 void **table) 1691 { 1692 struct smu_table_context *smu_table = &smu->smu_table; 1693 struct gpu_metrics_v1_3 *gpu_metrics = 1694 (struct gpu_metrics_v1_3 *)smu_table->gpu_metrics_table; 1695 SmuMetrics_t metrics; 1696 int i, ret = 0; 1697 1698 ret = smu_cmn_get_metrics_table(smu, 1699 &metrics, 1700 true); 1701 if (ret) 1702 return ret; 1703 1704 smu_cmn_init_soft_gpu_metrics(gpu_metrics, 1, 3); 1705 1706 gpu_metrics->temperature_edge = metrics.TemperatureEdge; 1707 gpu_metrics->temperature_hotspot = metrics.TemperatureHotspot; 1708 gpu_metrics->temperature_mem = metrics.TemperatureHBM; 1709 gpu_metrics->temperature_vrgfx = metrics.TemperatureVrGfx; 1710 gpu_metrics->temperature_vrsoc = metrics.TemperatureVrSoc; 1711 gpu_metrics->temperature_vrmem = metrics.TemperatureVrMem; 1712 1713 gpu_metrics->average_gfx_activity = metrics.AverageGfxActivity; 1714 gpu_metrics->average_umc_activity = metrics.AverageUclkActivity; 1715 gpu_metrics->average_mm_activity = 0; 1716 1717 /* Valid power data is available only from primary die */ 1718 if (aldebaran_is_primary(smu)) { 1719 gpu_metrics->average_socket_power = metrics.AverageSocketPower; 1720 gpu_metrics->energy_accumulator = 1721 (uint64_t)metrics.EnergyAcc64bitHigh << 32 | 1722 metrics.EnergyAcc64bitLow; 1723 } else { 1724 gpu_metrics->average_socket_power = 0; 1725 gpu_metrics->energy_accumulator = 0; 1726 } 1727 1728 gpu_metrics->average_gfxclk_frequency = metrics.AverageGfxclkFrequency; 1729 gpu_metrics->average_socclk_frequency = metrics.AverageSocclkFrequency; 1730 gpu_metrics->average_uclk_frequency = metrics.AverageUclkFrequency; 1731 gpu_metrics->average_vclk0_frequency = 0; 1732 gpu_metrics->average_dclk0_frequency = 0; 1733 1734 gpu_metrics->current_gfxclk = metrics.CurrClock[PPCLK_GFXCLK]; 1735 gpu_metrics->current_socclk = metrics.CurrClock[PPCLK_SOCCLK]; 1736 gpu_metrics->current_uclk = metrics.CurrClock[PPCLK_UCLK]; 1737 gpu_metrics->current_vclk0 = metrics.CurrClock[PPCLK_VCLK]; 1738 gpu_metrics->current_dclk0 = metrics.CurrClock[PPCLK_DCLK]; 1739 1740 gpu_metrics->throttle_status = metrics.ThrottlerStatus; 1741 gpu_metrics->indep_throttle_status = 1742 smu_cmn_get_indep_throttler_status(metrics.ThrottlerStatus, 1743 aldebaran_throttler_map); 1744 1745 gpu_metrics->current_fan_speed = 0; 1746 1747 gpu_metrics->pcie_link_width = 1748 smu_v13_0_get_current_pcie_link_width(smu); 1749 gpu_metrics->pcie_link_speed = 1750 aldebaran_get_current_pcie_link_speed(smu); 1751 1752 gpu_metrics->system_clock_counter = ktime_get_boottime_ns(); 1753 1754 gpu_metrics->gfx_activity_acc = metrics.GfxBusyAcc; 1755 gpu_metrics->mem_activity_acc = metrics.DramBusyAcc; 1756 1757 for (i = 0; i < NUM_HBM_INSTANCES; i++) 1758 gpu_metrics->temperature_hbm[i] = metrics.TemperatureAllHBM[i]; 1759 1760 gpu_metrics->firmware_timestamp = ((uint64_t)metrics.TimeStampHigh << 32) | 1761 metrics.TimeStampLow; 1762 1763 *table = (void *)gpu_metrics; 1764 1765 return sizeof(struct gpu_metrics_v1_3); 1766 } 1767 1768 static int aldebaran_mode2_reset(struct smu_context *smu) 1769 { 1770 u32 smu_version; 1771 int ret = 0, index; 1772 struct amdgpu_device *adev = smu->adev; 1773 int timeout = 10; 1774 1775 smu_cmn_get_smc_version(smu, NULL, &smu_version); 1776 1777 index = smu_cmn_to_asic_specific_index(smu, CMN2ASIC_MAPPING_MSG, 1778 SMU_MSG_GfxDeviceDriverReset); 1779 1780 mutex_lock(&smu->message_lock); 1781 if (smu_version >= 0x00441400) { 1782 ret = smu_cmn_send_msg_without_waiting(smu, (uint16_t)index, SMU_RESET_MODE_2); 1783 /* This is similar to FLR, wait till max FLR timeout */ 1784 msleep(100); 1785 dev_dbg(smu->adev->dev, "restore config space...\n"); 1786 /* Restore the config space saved during init */ 1787 amdgpu_device_load_pci_state(adev->pdev); 1788 1789 dev_dbg(smu->adev->dev, "wait for reset ack\n"); 1790 while (ret == -ETIME && timeout) { 1791 ret = smu_cmn_wait_for_response(smu); 1792 /* Wait a bit more time for getting ACK */ 1793 if (ret == -ETIME) { 1794 --timeout; 1795 usleep_range(500, 1000); 1796 continue; 1797 } 1798 1799 if (ret != 1) { 1800 dev_err(adev->dev, "failed to send mode2 message \tparam: 0x%08x response %#x\n", 1801 SMU_RESET_MODE_2, ret); 1802 goto out; 1803 } 1804 } 1805 1806 } else { 1807 dev_err(adev->dev, "smu fw 0x%x does not support MSG_GfxDeviceDriverReset MSG\n", 1808 smu_version); 1809 } 1810 1811 if (ret == 1) 1812 ret = 0; 1813 out: 1814 mutex_unlock(&smu->message_lock); 1815 1816 return ret; 1817 } 1818 1819 static bool aldebaran_is_mode1_reset_supported(struct smu_context *smu) 1820 { 1821 #if 0 1822 struct amdgpu_device *adev = smu->adev; 1823 u32 smu_version; 1824 uint32_t val; 1825 /** 1826 * PM FW version support mode1 reset from 68.07 1827 */ 1828 smu_cmn_get_smc_version(smu, NULL, &smu_version); 1829 if ((smu_version < 0x00440700)) 1830 return false; 1831 /** 1832 * mode1 reset relies on PSP, so we should check if 1833 * PSP is alive. 1834 */ 1835 val = RREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_81); 1836 1837 return val != 0x0; 1838 #endif 1839 return true; 1840 } 1841 1842 static bool aldebaran_is_mode2_reset_supported(struct smu_context *smu) 1843 { 1844 return true; 1845 } 1846 1847 static int aldebaran_set_mp1_state(struct smu_context *smu, 1848 enum pp_mp1_state mp1_state) 1849 { 1850 switch (mp1_state) { 1851 case PP_MP1_STATE_UNLOAD: 1852 return smu_cmn_set_mp1_state(smu, mp1_state); 1853 default: 1854 return 0; 1855 } 1856 } 1857 1858 static int aldebaran_smu_send_hbm_bad_page_num(struct smu_context *smu, 1859 uint32_t size) 1860 { 1861 int ret = 0; 1862 1863 /* message SMU to update the bad page number on SMUBUS */ 1864 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_SetNumBadHbmPagesRetired, size, NULL); 1865 if (ret) 1866 dev_err(smu->adev->dev, "[%s] failed to message SMU to update HBM bad pages number\n", 1867 __func__); 1868 1869 return ret; 1870 } 1871 1872 static const struct pptable_funcs aldebaran_ppt_funcs = { 1873 /* init dpm */ 1874 .get_allowed_feature_mask = aldebaran_get_allowed_feature_mask, 1875 /* dpm/clk tables */ 1876 .set_default_dpm_table = aldebaran_set_default_dpm_table, 1877 .populate_umd_state_clk = aldebaran_populate_umd_state_clk, 1878 .get_thermal_temperature_range = aldebaran_get_thermal_temperature_range, 1879 .print_clk_levels = aldebaran_print_clk_levels, 1880 .force_clk_levels = aldebaran_force_clk_levels, 1881 .read_sensor = aldebaran_read_sensor, 1882 .set_performance_level = aldebaran_set_performance_level, 1883 .get_power_limit = aldebaran_get_power_limit, 1884 .is_dpm_running = aldebaran_is_dpm_running, 1885 .get_unique_id = aldebaran_get_unique_id, 1886 .init_microcode = smu_v13_0_init_microcode, 1887 .load_microcode = smu_v13_0_load_microcode, 1888 .fini_microcode = smu_v13_0_fini_microcode, 1889 .init_smc_tables = aldebaran_init_smc_tables, 1890 .fini_smc_tables = smu_v13_0_fini_smc_tables, 1891 .init_power = smu_v13_0_init_power, 1892 .fini_power = smu_v13_0_fini_power, 1893 .check_fw_status = smu_v13_0_check_fw_status, 1894 /* pptable related */ 1895 .setup_pptable = aldebaran_setup_pptable, 1896 .get_vbios_bootup_values = smu_v13_0_get_vbios_bootup_values, 1897 .check_fw_version = smu_v13_0_check_fw_version, 1898 .write_pptable = smu_cmn_write_pptable, 1899 .set_driver_table_location = smu_v13_0_set_driver_table_location, 1900 .set_tool_table_location = smu_v13_0_set_tool_table_location, 1901 .notify_memory_pool_location = smu_v13_0_notify_memory_pool_location, 1902 .system_features_control = aldebaran_system_features_control, 1903 .send_smc_msg_with_param = smu_cmn_send_smc_msg_with_param, 1904 .send_smc_msg = smu_cmn_send_smc_msg, 1905 .get_enabled_mask = smu_cmn_get_enabled_mask, 1906 .feature_is_enabled = smu_cmn_feature_is_enabled, 1907 .disable_all_features_with_exception = smu_cmn_disable_all_features_with_exception, 1908 .set_power_limit = aldebaran_set_power_limit, 1909 .init_max_sustainable_clocks = smu_v13_0_init_max_sustainable_clocks, 1910 .enable_thermal_alert = smu_v13_0_enable_thermal_alert, 1911 .disable_thermal_alert = smu_v13_0_disable_thermal_alert, 1912 .set_xgmi_pstate = smu_v13_0_set_xgmi_pstate, 1913 .register_irq_handler = smu_v13_0_register_irq_handler, 1914 .set_azalia_d3_pme = smu_v13_0_set_azalia_d3_pme, 1915 .get_max_sustainable_clocks_by_dc = smu_v13_0_get_max_sustainable_clocks_by_dc, 1916 .baco_is_support= aldebaran_is_baco_supported, 1917 .get_dpm_ultimate_freq = smu_v13_0_get_dpm_ultimate_freq, 1918 .set_soft_freq_limited_range = aldebaran_set_soft_freq_limited_range, 1919 .od_edit_dpm_table = aldebaran_usr_edit_dpm_table, 1920 .set_df_cstate = aldebaran_set_df_cstate, 1921 .allow_xgmi_power_down = aldebaran_allow_xgmi_power_down, 1922 .log_thermal_throttling_event = aldebaran_log_thermal_throttling_event, 1923 .get_pp_feature_mask = smu_cmn_get_pp_feature_mask, 1924 .set_pp_feature_mask = smu_cmn_set_pp_feature_mask, 1925 .get_gpu_metrics = aldebaran_get_gpu_metrics, 1926 .mode1_reset_is_support = aldebaran_is_mode1_reset_supported, 1927 .mode2_reset_is_support = aldebaran_is_mode2_reset_supported, 1928 .mode1_reset = smu_v13_0_mode1_reset, 1929 .set_mp1_state = aldebaran_set_mp1_state, 1930 .mode2_reset = aldebaran_mode2_reset, 1931 .wait_for_event = smu_v13_0_wait_for_event, 1932 .i2c_init = aldebaran_i2c_control_init, 1933 .i2c_fini = aldebaran_i2c_control_fini, 1934 .send_hbm_bad_pages_num = aldebaran_smu_send_hbm_bad_page_num, 1935 }; 1936 1937 void aldebaran_set_ppt_funcs(struct smu_context *smu) 1938 { 1939 smu->ppt_funcs = &aldebaran_ppt_funcs; 1940 smu->message_map = aldebaran_message_map; 1941 smu->clock_map = aldebaran_clk_map; 1942 smu->feature_map = aldebaran_feature_mask_map; 1943 smu->table_map = aldebaran_table_map; 1944 } 1945