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, 0), 94 MSG_MAP(GetEnabledSmuFeaturesHigh, PPSMC_MSG_GetEnabledSmuFeaturesHigh, 0), 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_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_PG_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 memcpy(&smc_pptable->GfxMaxCurrent, 413 &smc_dpm_table->GfxMaxCurrent, 414 sizeof(*smc_dpm_table) - offsetof(struct atom_smc_dpm_info_v4_10, GfxMaxCurrent)); 415 return 0; 416 } 417 418 static int aldebaran_setup_pptable(struct smu_context *smu) 419 { 420 int ret = 0; 421 422 /* VBIOS pptable is the first choice */ 423 smu->smu_table.boot_values.pp_table_id = 0; 424 425 ret = smu_v13_0_setup_pptable(smu); 426 if (ret) 427 return ret; 428 429 ret = aldebaran_store_powerplay_table(smu); 430 if (ret) 431 return ret; 432 433 ret = aldebaran_append_powerplay_table(smu); 434 if (ret) 435 return ret; 436 437 ret = aldebaran_check_powerplay_table(smu); 438 if (ret) 439 return ret; 440 441 return ret; 442 } 443 444 static bool aldebaran_is_primary(struct smu_context *smu) 445 { 446 struct amdgpu_device *adev = smu->adev; 447 448 if (adev->smuio.funcs && adev->smuio.funcs->get_die_id) 449 return adev->smuio.funcs->get_die_id(adev) == 0; 450 451 return true; 452 } 453 454 static int aldebaran_run_board_btc(struct smu_context *smu) 455 { 456 u32 smu_version; 457 int ret; 458 459 if (!aldebaran_is_primary(smu)) 460 return 0; 461 462 ret = smu_cmn_get_smc_version(smu, NULL, &smu_version); 463 if (ret) { 464 dev_err(smu->adev->dev, "Failed to get smu version!\n"); 465 return ret; 466 } 467 if (smu_version <= 0x00441d00) 468 return 0; 469 470 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_BoardPowerCalibration, NULL); 471 if (ret) 472 dev_err(smu->adev->dev, "Board power calibration failed!\n"); 473 474 return ret; 475 } 476 477 static int aldebaran_run_btc(struct smu_context *smu) 478 { 479 int ret; 480 481 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_RunDcBtc, NULL); 482 if (ret) 483 dev_err(smu->adev->dev, "RunDcBtc failed!\n"); 484 else 485 ret = aldebaran_run_board_btc(smu); 486 487 return ret; 488 } 489 490 static int aldebaran_populate_umd_state_clk(struct smu_context *smu) 491 { 492 struct smu_13_0_dpm_context *dpm_context = 493 smu->smu_dpm.dpm_context; 494 struct smu_13_0_dpm_table *gfx_table = 495 &dpm_context->dpm_tables.gfx_table; 496 struct smu_13_0_dpm_table *mem_table = 497 &dpm_context->dpm_tables.uclk_table; 498 struct smu_13_0_dpm_table *soc_table = 499 &dpm_context->dpm_tables.soc_table; 500 struct smu_umd_pstate_table *pstate_table = 501 &smu->pstate_table; 502 503 pstate_table->gfxclk_pstate.min = gfx_table->min; 504 pstate_table->gfxclk_pstate.peak = gfx_table->max; 505 pstate_table->gfxclk_pstate.curr.min = gfx_table->min; 506 pstate_table->gfxclk_pstate.curr.max = gfx_table->max; 507 508 pstate_table->uclk_pstate.min = mem_table->min; 509 pstate_table->uclk_pstate.peak = mem_table->max; 510 pstate_table->uclk_pstate.curr.min = mem_table->min; 511 pstate_table->uclk_pstate.curr.max = mem_table->max; 512 513 pstate_table->socclk_pstate.min = soc_table->min; 514 pstate_table->socclk_pstate.peak = soc_table->max; 515 pstate_table->socclk_pstate.curr.min = soc_table->min; 516 pstate_table->socclk_pstate.curr.max = soc_table->max; 517 518 if (gfx_table->count > ALDEBARAN_UMD_PSTATE_GFXCLK_LEVEL && 519 mem_table->count > ALDEBARAN_UMD_PSTATE_MCLK_LEVEL && 520 soc_table->count > ALDEBARAN_UMD_PSTATE_SOCCLK_LEVEL) { 521 pstate_table->gfxclk_pstate.standard = 522 gfx_table->dpm_levels[ALDEBARAN_UMD_PSTATE_GFXCLK_LEVEL].value; 523 pstate_table->uclk_pstate.standard = 524 mem_table->dpm_levels[ALDEBARAN_UMD_PSTATE_MCLK_LEVEL].value; 525 pstate_table->socclk_pstate.standard = 526 soc_table->dpm_levels[ALDEBARAN_UMD_PSTATE_SOCCLK_LEVEL].value; 527 } else { 528 pstate_table->gfxclk_pstate.standard = 529 pstate_table->gfxclk_pstate.min; 530 pstate_table->uclk_pstate.standard = 531 pstate_table->uclk_pstate.min; 532 pstate_table->socclk_pstate.standard = 533 pstate_table->socclk_pstate.min; 534 } 535 536 return 0; 537 } 538 539 static int aldebaran_get_clk_table(struct smu_context *smu, 540 struct pp_clock_levels_with_latency *clocks, 541 struct smu_13_0_dpm_table *dpm_table) 542 { 543 int i, count; 544 545 count = (dpm_table->count > MAX_NUM_CLOCKS) ? MAX_NUM_CLOCKS : dpm_table->count; 546 clocks->num_levels = count; 547 548 for (i = 0; i < count; i++) { 549 clocks->data[i].clocks_in_khz = 550 dpm_table->dpm_levels[i].value * 1000; 551 clocks->data[i].latency_in_us = 0; 552 } 553 554 return 0; 555 } 556 557 static int aldebaran_freqs_in_same_level(int32_t frequency1, 558 int32_t frequency2) 559 { 560 return (abs(frequency1 - frequency2) <= EPSILON); 561 } 562 563 static int aldebaran_get_smu_metrics_data(struct smu_context *smu, 564 MetricsMember_t member, 565 uint32_t *value) 566 { 567 struct smu_table_context *smu_table= &smu->smu_table; 568 SmuMetrics_t *metrics = (SmuMetrics_t *)smu_table->metrics_table; 569 int ret = 0; 570 571 mutex_lock(&smu->metrics_lock); 572 573 ret = smu_cmn_get_metrics_table_locked(smu, 574 NULL, 575 false); 576 if (ret) { 577 mutex_unlock(&smu->metrics_lock); 578 return ret; 579 } 580 581 switch (member) { 582 case METRICS_CURR_GFXCLK: 583 *value = metrics->CurrClock[PPCLK_GFXCLK]; 584 break; 585 case METRICS_CURR_SOCCLK: 586 *value = metrics->CurrClock[PPCLK_SOCCLK]; 587 break; 588 case METRICS_CURR_UCLK: 589 *value = metrics->CurrClock[PPCLK_UCLK]; 590 break; 591 case METRICS_CURR_VCLK: 592 *value = metrics->CurrClock[PPCLK_VCLK]; 593 break; 594 case METRICS_CURR_DCLK: 595 *value = metrics->CurrClock[PPCLK_DCLK]; 596 break; 597 case METRICS_CURR_FCLK: 598 *value = metrics->CurrClock[PPCLK_FCLK]; 599 break; 600 case METRICS_AVERAGE_GFXCLK: 601 *value = metrics->AverageGfxclkFrequency; 602 break; 603 case METRICS_AVERAGE_SOCCLK: 604 *value = metrics->AverageSocclkFrequency; 605 break; 606 case METRICS_AVERAGE_UCLK: 607 *value = metrics->AverageUclkFrequency; 608 break; 609 case METRICS_AVERAGE_GFXACTIVITY: 610 *value = metrics->AverageGfxActivity; 611 break; 612 case METRICS_AVERAGE_MEMACTIVITY: 613 *value = metrics->AverageUclkActivity; 614 break; 615 case METRICS_AVERAGE_SOCKETPOWER: 616 /* Valid power data is available only from primary die */ 617 *value = aldebaran_is_primary(smu) ? 618 metrics->AverageSocketPower << 8 : 619 0; 620 break; 621 case METRICS_TEMPERATURE_EDGE: 622 *value = metrics->TemperatureEdge * 623 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 624 break; 625 case METRICS_TEMPERATURE_HOTSPOT: 626 *value = metrics->TemperatureHotspot * 627 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 628 break; 629 case METRICS_TEMPERATURE_MEM: 630 *value = metrics->TemperatureHBM * 631 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 632 break; 633 case METRICS_TEMPERATURE_VRGFX: 634 *value = metrics->TemperatureVrGfx * 635 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 636 break; 637 case METRICS_TEMPERATURE_VRSOC: 638 *value = metrics->TemperatureVrSoc * 639 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 640 break; 641 case METRICS_TEMPERATURE_VRMEM: 642 *value = metrics->TemperatureVrMem * 643 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 644 break; 645 case METRICS_THROTTLER_STATUS: 646 *value = metrics->ThrottlerStatus; 647 break; 648 default: 649 *value = UINT_MAX; 650 break; 651 } 652 653 mutex_unlock(&smu->metrics_lock); 654 655 return ret; 656 } 657 658 static int aldebaran_get_current_clk_freq_by_table(struct smu_context *smu, 659 enum smu_clk_type clk_type, 660 uint32_t *value) 661 { 662 MetricsMember_t member_type; 663 int clk_id = 0; 664 665 if (!value) 666 return -EINVAL; 667 668 clk_id = smu_cmn_to_asic_specific_index(smu, 669 CMN2ASIC_MAPPING_CLK, 670 clk_type); 671 if (clk_id < 0) 672 return -EINVAL; 673 674 switch (clk_id) { 675 case PPCLK_GFXCLK: 676 /* 677 * CurrClock[clk_id] can provide accurate 678 * output only when the dpm feature is enabled. 679 * We can use Average_* for dpm disabled case. 680 * But this is available for gfxclk/uclk/socclk/vclk/dclk. 681 */ 682 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT)) 683 member_type = METRICS_CURR_GFXCLK; 684 else 685 member_type = METRICS_AVERAGE_GFXCLK; 686 break; 687 case PPCLK_UCLK: 688 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT)) 689 member_type = METRICS_CURR_UCLK; 690 else 691 member_type = METRICS_AVERAGE_UCLK; 692 break; 693 case PPCLK_SOCCLK: 694 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT)) 695 member_type = METRICS_CURR_SOCCLK; 696 else 697 member_type = METRICS_AVERAGE_SOCCLK; 698 break; 699 case PPCLK_VCLK: 700 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) 701 member_type = METRICS_CURR_VCLK; 702 else 703 member_type = METRICS_AVERAGE_VCLK; 704 break; 705 case PPCLK_DCLK: 706 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_VCN_PG_BIT)) 707 member_type = METRICS_CURR_DCLK; 708 else 709 member_type = METRICS_AVERAGE_DCLK; 710 break; 711 case PPCLK_FCLK: 712 member_type = METRICS_CURR_FCLK; 713 break; 714 default: 715 return -EINVAL; 716 } 717 718 return aldebaran_get_smu_metrics_data(smu, 719 member_type, 720 value); 721 } 722 723 static int aldebaran_print_clk_levels(struct smu_context *smu, 724 enum smu_clk_type type, char *buf) 725 { 726 int i, now, size = 0; 727 int ret = 0; 728 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 729 struct pp_clock_levels_with_latency clocks; 730 struct smu_13_0_dpm_table *single_dpm_table; 731 struct smu_dpm_context *smu_dpm = &smu->smu_dpm; 732 struct smu_13_0_dpm_context *dpm_context = NULL; 733 uint32_t display_levels; 734 uint32_t freq_values[3] = {0}; 735 uint32_t min_clk, max_clk; 736 737 if (amdgpu_ras_intr_triggered()) 738 return snprintf(buf, PAGE_SIZE, "unavailable\n"); 739 740 dpm_context = smu_dpm->dpm_context; 741 742 switch (type) { 743 744 case SMU_OD_SCLK: 745 size = sprintf(buf, "%s:\n", "GFXCLK"); 746 fallthrough; 747 case SMU_SCLK: 748 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_GFXCLK, &now); 749 if (ret) { 750 dev_err(smu->adev->dev, "Attempt to get current gfx clk Failed!"); 751 return ret; 752 } 753 754 single_dpm_table = &(dpm_context->dpm_tables.gfx_table); 755 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 756 if (ret) { 757 dev_err(smu->adev->dev, "Attempt to get gfx clk levels Failed!"); 758 return ret; 759 } 760 761 display_levels = clocks.num_levels; 762 763 min_clk = pstate_table->gfxclk_pstate.curr.min; 764 max_clk = pstate_table->gfxclk_pstate.curr.max; 765 766 freq_values[0] = min_clk; 767 freq_values[1] = max_clk; 768 769 /* fine-grained dpm has only 2 levels */ 770 if (now > min_clk && now < max_clk) { 771 display_levels = clocks.num_levels + 1; 772 freq_values[2] = max_clk; 773 freq_values[1] = now; 774 } 775 776 /* 777 * For DPM disabled case, there will be only one clock level. 778 * And it's safe to assume that is always the current clock. 779 */ 780 if (display_levels == clocks.num_levels) { 781 for (i = 0; i < clocks.num_levels; i++) 782 size += sprintf( 783 buf + size, "%d: %uMhz %s\n", i, 784 freq_values[i], 785 (clocks.num_levels == 1) ? 786 "*" : 787 (aldebaran_freqs_in_same_level( 788 freq_values[i], now) ? 789 "*" : 790 "")); 791 } else { 792 for (i = 0; i < display_levels; i++) 793 size += sprintf(buf + size, "%d: %uMhz %s\n", i, 794 freq_values[i], i == 1 ? "*" : ""); 795 } 796 797 break; 798 799 case SMU_OD_MCLK: 800 size = sprintf(buf, "%s:\n", "MCLK"); 801 fallthrough; 802 case SMU_MCLK: 803 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_UCLK, &now); 804 if (ret) { 805 dev_err(smu->adev->dev, "Attempt to get current mclk Failed!"); 806 return ret; 807 } 808 809 single_dpm_table = &(dpm_context->dpm_tables.uclk_table); 810 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 811 if (ret) { 812 dev_err(smu->adev->dev, "Attempt to get memory clk levels Failed!"); 813 return ret; 814 } 815 816 for (i = 0; i < clocks.num_levels; i++) 817 size += sprintf(buf + size, "%d: %uMhz %s\n", 818 i, clocks.data[i].clocks_in_khz / 1000, 819 (clocks.num_levels == 1) ? "*" : 820 (aldebaran_freqs_in_same_level( 821 clocks.data[i].clocks_in_khz / 1000, 822 now) ? "*" : "")); 823 break; 824 825 case SMU_SOCCLK: 826 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_SOCCLK, &now); 827 if (ret) { 828 dev_err(smu->adev->dev, "Attempt to get current socclk Failed!"); 829 return ret; 830 } 831 832 single_dpm_table = &(dpm_context->dpm_tables.soc_table); 833 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 834 if (ret) { 835 dev_err(smu->adev->dev, "Attempt to get socclk levels Failed!"); 836 return ret; 837 } 838 839 for (i = 0; i < clocks.num_levels; i++) 840 size += sprintf(buf + size, "%d: %uMhz %s\n", 841 i, clocks.data[i].clocks_in_khz / 1000, 842 (clocks.num_levels == 1) ? "*" : 843 (aldebaran_freqs_in_same_level( 844 clocks.data[i].clocks_in_khz / 1000, 845 now) ? "*" : "")); 846 break; 847 848 case SMU_FCLK: 849 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_FCLK, &now); 850 if (ret) { 851 dev_err(smu->adev->dev, "Attempt to get current fclk Failed!"); 852 return ret; 853 } 854 855 single_dpm_table = &(dpm_context->dpm_tables.fclk_table); 856 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 857 if (ret) { 858 dev_err(smu->adev->dev, "Attempt to get fclk levels Failed!"); 859 return ret; 860 } 861 862 for (i = 0; i < single_dpm_table->count; i++) 863 size += sprintf(buf + size, "%d: %uMhz %s\n", 864 i, single_dpm_table->dpm_levels[i].value, 865 (clocks.num_levels == 1) ? "*" : 866 (aldebaran_freqs_in_same_level( 867 clocks.data[i].clocks_in_khz / 1000, 868 now) ? "*" : "")); 869 break; 870 871 case SMU_VCLK: 872 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_VCLK, &now); 873 if (ret) { 874 dev_err(smu->adev->dev, "Attempt to get current vclk Failed!"); 875 return ret; 876 } 877 878 single_dpm_table = &(dpm_context->dpm_tables.vclk_table); 879 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 880 if (ret) { 881 dev_err(smu->adev->dev, "Attempt to get vclk levels Failed!"); 882 return ret; 883 } 884 885 for (i = 0; i < single_dpm_table->count; i++) 886 size += sprintf(buf + size, "%d: %uMhz %s\n", 887 i, single_dpm_table->dpm_levels[i].value, 888 (clocks.num_levels == 1) ? "*" : 889 (aldebaran_freqs_in_same_level( 890 clocks.data[i].clocks_in_khz / 1000, 891 now) ? "*" : "")); 892 break; 893 894 case SMU_DCLK: 895 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_DCLK, &now); 896 if (ret) { 897 dev_err(smu->adev->dev, "Attempt to get current dclk Failed!"); 898 return ret; 899 } 900 901 single_dpm_table = &(dpm_context->dpm_tables.dclk_table); 902 ret = aldebaran_get_clk_table(smu, &clocks, single_dpm_table); 903 if (ret) { 904 dev_err(smu->adev->dev, "Attempt to get dclk levels Failed!"); 905 return ret; 906 } 907 908 for (i = 0; i < single_dpm_table->count; i++) 909 size += sprintf(buf + size, "%d: %uMhz %s\n", 910 i, single_dpm_table->dpm_levels[i].value, 911 (clocks.num_levels == 1) ? "*" : 912 (aldebaran_freqs_in_same_level( 913 clocks.data[i].clocks_in_khz / 1000, 914 now) ? "*" : "")); 915 break; 916 917 default: 918 break; 919 } 920 921 return size; 922 } 923 924 static int aldebaran_upload_dpm_level(struct smu_context *smu, 925 bool max, 926 uint32_t feature_mask, 927 uint32_t level) 928 { 929 struct smu_13_0_dpm_context *dpm_context = 930 smu->smu_dpm.dpm_context; 931 uint32_t freq; 932 int ret = 0; 933 934 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT) && 935 (feature_mask & FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT))) { 936 freq = dpm_context->dpm_tables.gfx_table.dpm_levels[level].value; 937 ret = smu_cmn_send_smc_msg_with_param(smu, 938 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 939 (PPCLK_GFXCLK << 16) | (freq & 0xffff), 940 NULL); 941 if (ret) { 942 dev_err(smu->adev->dev, "Failed to set soft %s gfxclk !\n", 943 max ? "max" : "min"); 944 return ret; 945 } 946 } 947 948 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT) && 949 (feature_mask & FEATURE_MASK(FEATURE_DPM_UCLK_BIT))) { 950 freq = dpm_context->dpm_tables.uclk_table.dpm_levels[level].value; 951 ret = smu_cmn_send_smc_msg_with_param(smu, 952 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 953 (PPCLK_UCLK << 16) | (freq & 0xffff), 954 NULL); 955 if (ret) { 956 dev_err(smu->adev->dev, "Failed to set soft %s memclk !\n", 957 max ? "max" : "min"); 958 return ret; 959 } 960 } 961 962 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT) && 963 (feature_mask & FEATURE_MASK(FEATURE_DPM_SOCCLK_BIT))) { 964 freq = dpm_context->dpm_tables.soc_table.dpm_levels[level].value; 965 ret = smu_cmn_send_smc_msg_with_param(smu, 966 (max ? SMU_MSG_SetSoftMaxByFreq : SMU_MSG_SetSoftMinByFreq), 967 (PPCLK_SOCCLK << 16) | (freq & 0xffff), 968 NULL); 969 if (ret) { 970 dev_err(smu->adev->dev, "Failed to set soft %s socclk !\n", 971 max ? "max" : "min"); 972 return ret; 973 } 974 } 975 976 return ret; 977 } 978 979 static int aldebaran_force_clk_levels(struct smu_context *smu, 980 enum smu_clk_type type, uint32_t mask) 981 { 982 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context; 983 struct smu_13_0_dpm_table *single_dpm_table = NULL; 984 uint32_t soft_min_level, soft_max_level; 985 int ret = 0; 986 987 soft_min_level = mask ? (ffs(mask) - 1) : 0; 988 soft_max_level = mask ? (fls(mask) - 1) : 0; 989 990 switch (type) { 991 case SMU_SCLK: 992 single_dpm_table = &(dpm_context->dpm_tables.gfx_table); 993 if (soft_max_level >= single_dpm_table->count) { 994 dev_err(smu->adev->dev, "Clock level specified %d is over max allowed %d\n", 995 soft_max_level, single_dpm_table->count - 1); 996 ret = -EINVAL; 997 break; 998 } 999 1000 ret = aldebaran_upload_dpm_level(smu, 1001 false, 1002 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT), 1003 soft_min_level); 1004 if (ret) { 1005 dev_err(smu->adev->dev, "Failed to upload boot level to lowest!\n"); 1006 break; 1007 } 1008 1009 ret = aldebaran_upload_dpm_level(smu, 1010 true, 1011 FEATURE_MASK(FEATURE_DPM_GFXCLK_BIT), 1012 soft_max_level); 1013 if (ret) 1014 dev_err(smu->adev->dev, "Failed to upload dpm max level to highest!\n"); 1015 1016 break; 1017 1018 case SMU_MCLK: 1019 case SMU_SOCCLK: 1020 case SMU_FCLK: 1021 /* 1022 * Should not arrive here since aldebaran does not 1023 * support mclk/socclk/fclk softmin/softmax settings 1024 */ 1025 ret = -EINVAL; 1026 break; 1027 1028 default: 1029 break; 1030 } 1031 1032 return ret; 1033 } 1034 1035 static int aldebaran_get_thermal_temperature_range(struct smu_context *smu, 1036 struct smu_temperature_range *range) 1037 { 1038 struct smu_table_context *table_context = &smu->smu_table; 1039 struct smu_13_0_powerplay_table *powerplay_table = 1040 table_context->power_play_table; 1041 PPTable_t *pptable = smu->smu_table.driver_pptable; 1042 1043 if (!range) 1044 return -EINVAL; 1045 1046 memcpy(range, &smu13_thermal_policy[0], sizeof(struct smu_temperature_range)); 1047 1048 range->hotspot_crit_max = pptable->ThotspotLimit * 1049 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1050 range->hotspot_emergency_max = (pptable->ThotspotLimit + CTF_OFFSET_HOTSPOT) * 1051 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1052 range->mem_crit_max = pptable->TmemLimit * 1053 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1054 range->mem_emergency_max = (pptable->TmemLimit + CTF_OFFSET_MEM)* 1055 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES; 1056 range->software_shutdown_temp = powerplay_table->software_shutdown_temp; 1057 1058 return 0; 1059 } 1060 1061 static int aldebaran_get_current_activity_percent(struct smu_context *smu, 1062 enum amd_pp_sensors sensor, 1063 uint32_t *value) 1064 { 1065 int ret = 0; 1066 1067 if (!value) 1068 return -EINVAL; 1069 1070 switch (sensor) { 1071 case AMDGPU_PP_SENSOR_GPU_LOAD: 1072 ret = aldebaran_get_smu_metrics_data(smu, 1073 METRICS_AVERAGE_GFXACTIVITY, 1074 value); 1075 break; 1076 case AMDGPU_PP_SENSOR_MEM_LOAD: 1077 ret = aldebaran_get_smu_metrics_data(smu, 1078 METRICS_AVERAGE_MEMACTIVITY, 1079 value); 1080 break; 1081 default: 1082 dev_err(smu->adev->dev, "Invalid sensor for retrieving clock activity\n"); 1083 return -EINVAL; 1084 } 1085 1086 return ret; 1087 } 1088 1089 static int aldebaran_get_gpu_power(struct smu_context *smu, uint32_t *value) 1090 { 1091 if (!value) 1092 return -EINVAL; 1093 1094 return aldebaran_get_smu_metrics_data(smu, 1095 METRICS_AVERAGE_SOCKETPOWER, 1096 value); 1097 } 1098 1099 static int aldebaran_thermal_get_temperature(struct smu_context *smu, 1100 enum amd_pp_sensors sensor, 1101 uint32_t *value) 1102 { 1103 int ret = 0; 1104 1105 if (!value) 1106 return -EINVAL; 1107 1108 switch (sensor) { 1109 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP: 1110 ret = aldebaran_get_smu_metrics_data(smu, 1111 METRICS_TEMPERATURE_HOTSPOT, 1112 value); 1113 break; 1114 case AMDGPU_PP_SENSOR_EDGE_TEMP: 1115 ret = aldebaran_get_smu_metrics_data(smu, 1116 METRICS_TEMPERATURE_EDGE, 1117 value); 1118 break; 1119 case AMDGPU_PP_SENSOR_MEM_TEMP: 1120 ret = aldebaran_get_smu_metrics_data(smu, 1121 METRICS_TEMPERATURE_MEM, 1122 value); 1123 break; 1124 default: 1125 dev_err(smu->adev->dev, "Invalid sensor for retrieving temp\n"); 1126 return -EINVAL; 1127 } 1128 1129 return ret; 1130 } 1131 1132 static int aldebaran_read_sensor(struct smu_context *smu, 1133 enum amd_pp_sensors sensor, 1134 void *data, uint32_t *size) 1135 { 1136 int ret = 0; 1137 1138 if (amdgpu_ras_intr_triggered()) 1139 return 0; 1140 1141 if (!data || !size) 1142 return -EINVAL; 1143 1144 mutex_lock(&smu->sensor_lock); 1145 switch (sensor) { 1146 case AMDGPU_PP_SENSOR_MEM_LOAD: 1147 case AMDGPU_PP_SENSOR_GPU_LOAD: 1148 ret = aldebaran_get_current_activity_percent(smu, 1149 sensor, 1150 (uint32_t *)data); 1151 *size = 4; 1152 break; 1153 case AMDGPU_PP_SENSOR_GPU_POWER: 1154 ret = aldebaran_get_gpu_power(smu, (uint32_t *)data); 1155 *size = 4; 1156 break; 1157 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP: 1158 case AMDGPU_PP_SENSOR_EDGE_TEMP: 1159 case AMDGPU_PP_SENSOR_MEM_TEMP: 1160 ret = aldebaran_thermal_get_temperature(smu, sensor, 1161 (uint32_t *)data); 1162 *size = 4; 1163 break; 1164 case AMDGPU_PP_SENSOR_GFX_MCLK: 1165 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_UCLK, (uint32_t *)data); 1166 /* the output clock frequency in 10K unit */ 1167 *(uint32_t *)data *= 100; 1168 *size = 4; 1169 break; 1170 case AMDGPU_PP_SENSOR_GFX_SCLK: 1171 ret = aldebaran_get_current_clk_freq_by_table(smu, SMU_GFXCLK, (uint32_t *)data); 1172 *(uint32_t *)data *= 100; 1173 *size = 4; 1174 break; 1175 case AMDGPU_PP_SENSOR_VDDGFX: 1176 ret = smu_v13_0_get_gfx_vdd(smu, (uint32_t *)data); 1177 *size = 4; 1178 break; 1179 default: 1180 ret = -EOPNOTSUPP; 1181 break; 1182 } 1183 mutex_unlock(&smu->sensor_lock); 1184 1185 return ret; 1186 } 1187 1188 static int aldebaran_get_power_limit(struct smu_context *smu, 1189 uint32_t *current_power_limit, 1190 uint32_t *default_power_limit, 1191 uint32_t *max_power_limit) 1192 { 1193 PPTable_t *pptable = smu->smu_table.driver_pptable; 1194 uint32_t power_limit = 0; 1195 int ret; 1196 1197 if (!smu_cmn_feature_is_enabled(smu, SMU_FEATURE_PPT_BIT)) 1198 return -EINVAL; 1199 1200 /* Valid power data is available only from primary die. 1201 * For secondary die show the value as 0. 1202 */ 1203 if (aldebaran_is_primary(smu)) { 1204 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_GetPptLimit, 1205 &power_limit); 1206 1207 if (ret) { 1208 /* the last hope to figure out the ppt limit */ 1209 if (!pptable) { 1210 dev_err(smu->adev->dev, 1211 "Cannot get PPT limit due to pptable missing!"); 1212 return -EINVAL; 1213 } 1214 power_limit = pptable->PptLimit; 1215 } 1216 } 1217 1218 if (current_power_limit) 1219 *current_power_limit = power_limit; 1220 if (default_power_limit) 1221 *default_power_limit = power_limit; 1222 1223 if (max_power_limit) { 1224 if (pptable) 1225 *max_power_limit = pptable->PptLimit; 1226 } 1227 1228 return 0; 1229 } 1230 1231 static int aldebaran_set_power_limit(struct smu_context *smu, uint32_t n) 1232 { 1233 /* Power limit can be set only through primary die */ 1234 if (aldebaran_is_primary(smu)) 1235 return smu_v13_0_set_power_limit(smu, n); 1236 1237 return -EINVAL; 1238 } 1239 1240 static int aldebaran_system_features_control(struct smu_context *smu, bool enable) 1241 { 1242 int ret; 1243 1244 ret = smu_v13_0_system_features_control(smu, enable); 1245 if (!ret && enable) 1246 ret = aldebaran_run_btc(smu); 1247 1248 return ret; 1249 } 1250 1251 static int aldebaran_set_performance_level(struct smu_context *smu, 1252 enum amd_dpm_forced_level level) 1253 { 1254 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1255 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1256 struct smu_13_0_dpm_table *gfx_table = 1257 &dpm_context->dpm_tables.gfx_table; 1258 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1259 1260 /* Disable determinism if switching to another mode */ 1261 if ((smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) && 1262 (level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) { 1263 smu_cmn_send_smc_msg(smu, SMU_MSG_DisableDeterminism, NULL); 1264 pstate_table->gfxclk_pstate.curr.max = gfx_table->max; 1265 } 1266 1267 switch (level) { 1268 1269 case AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM: 1270 return 0; 1271 1272 case AMD_DPM_FORCED_LEVEL_HIGH: 1273 case AMD_DPM_FORCED_LEVEL_LOW: 1274 case AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD: 1275 case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK: 1276 case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK: 1277 case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK: 1278 default: 1279 break; 1280 } 1281 1282 return smu_v13_0_set_performance_level(smu, level); 1283 } 1284 1285 static int aldebaran_set_soft_freq_limited_range(struct smu_context *smu, 1286 enum smu_clk_type clk_type, 1287 uint32_t min, 1288 uint32_t max) 1289 { 1290 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1291 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1292 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1293 struct amdgpu_device *adev = smu->adev; 1294 uint32_t min_clk; 1295 uint32_t max_clk; 1296 int ret = 0; 1297 1298 if (clk_type != SMU_GFXCLK && clk_type != SMU_SCLK) 1299 return -EINVAL; 1300 1301 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) 1302 && (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) 1303 return -EINVAL; 1304 1305 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_MANUAL) { 1306 if (min >= max) { 1307 dev_err(smu->adev->dev, 1308 "Minimum GFX clk should be less than the maximum allowed clock\n"); 1309 return -EINVAL; 1310 } 1311 1312 if ((min == pstate_table->gfxclk_pstate.curr.min) && 1313 (max == pstate_table->gfxclk_pstate.curr.max)) 1314 return 0; 1315 1316 ret = smu_v13_0_set_soft_freq_limited_range(smu, SMU_GFXCLK, 1317 min, max); 1318 if (!ret) { 1319 pstate_table->gfxclk_pstate.curr.min = min; 1320 pstate_table->gfxclk_pstate.curr.max = max; 1321 } 1322 1323 return ret; 1324 } 1325 1326 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) { 1327 if (!max || (max < dpm_context->dpm_tables.gfx_table.min) || 1328 (max > dpm_context->dpm_tables.gfx_table.max)) { 1329 dev_warn(adev->dev, 1330 "Invalid max frequency %d MHz specified for determinism\n", max); 1331 return -EINVAL; 1332 } 1333 1334 /* Restore default min/max clocks and enable determinism */ 1335 min_clk = dpm_context->dpm_tables.gfx_table.min; 1336 max_clk = dpm_context->dpm_tables.gfx_table.max; 1337 ret = smu_v13_0_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1338 if (!ret) { 1339 usleep_range(500, 1000); 1340 ret = smu_cmn_send_smc_msg_with_param(smu, 1341 SMU_MSG_EnableDeterminism, 1342 max, NULL); 1343 if (ret) { 1344 dev_err(adev->dev, 1345 "Failed to enable determinism at GFX clock %d MHz\n", max); 1346 } else { 1347 pstate_table->gfxclk_pstate.curr.min = min_clk; 1348 pstate_table->gfxclk_pstate.curr.max = max; 1349 } 1350 } 1351 } 1352 1353 return ret; 1354 } 1355 1356 static int aldebaran_usr_edit_dpm_table(struct smu_context *smu, enum PP_OD_DPM_TABLE_COMMAND type, 1357 long input[], uint32_t size) 1358 { 1359 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm); 1360 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context; 1361 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table; 1362 uint32_t min_clk; 1363 uint32_t max_clk; 1364 int ret = 0; 1365 1366 /* Only allowed in manual or determinism mode */ 1367 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) 1368 && (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) 1369 return -EINVAL; 1370 1371 switch (type) { 1372 case PP_OD_EDIT_SCLK_VDDC_TABLE: 1373 if (size != 2) { 1374 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1375 return -EINVAL; 1376 } 1377 1378 if (input[0] == 0) { 1379 if (input[1] < dpm_context->dpm_tables.gfx_table.min) { 1380 dev_warn(smu->adev->dev, "Minimum GFX clk (%ld) MHz specified is less than the minimum allowed (%d) MHz\n", 1381 input[1], dpm_context->dpm_tables.gfx_table.min); 1382 pstate_table->gfxclk_pstate.custom.min = 1383 pstate_table->gfxclk_pstate.curr.min; 1384 return -EINVAL; 1385 } 1386 1387 pstate_table->gfxclk_pstate.custom.min = input[1]; 1388 } else if (input[0] == 1) { 1389 if (input[1] > dpm_context->dpm_tables.gfx_table.max) { 1390 dev_warn(smu->adev->dev, "Maximum GFX clk (%ld) MHz specified is greater than the maximum allowed (%d) MHz\n", 1391 input[1], dpm_context->dpm_tables.gfx_table.max); 1392 pstate_table->gfxclk_pstate.custom.max = 1393 pstate_table->gfxclk_pstate.curr.max; 1394 return -EINVAL; 1395 } 1396 1397 pstate_table->gfxclk_pstate.custom.max = input[1]; 1398 } else { 1399 return -EINVAL; 1400 } 1401 break; 1402 case PP_OD_RESTORE_DEFAULT_TABLE: 1403 if (size != 0) { 1404 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1405 return -EINVAL; 1406 } else { 1407 /* Use the default frequencies for manual and determinism mode */ 1408 min_clk = dpm_context->dpm_tables.gfx_table.min; 1409 max_clk = dpm_context->dpm_tables.gfx_table.max; 1410 1411 return aldebaran_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1412 } 1413 break; 1414 case PP_OD_COMMIT_DPM_TABLE: 1415 if (size != 0) { 1416 dev_err(smu->adev->dev, "Input parameter number not correct\n"); 1417 return -EINVAL; 1418 } else { 1419 if (!pstate_table->gfxclk_pstate.custom.min) 1420 pstate_table->gfxclk_pstate.custom.min = 1421 pstate_table->gfxclk_pstate.curr.min; 1422 1423 if (!pstate_table->gfxclk_pstate.custom.max) 1424 pstate_table->gfxclk_pstate.custom.max = 1425 pstate_table->gfxclk_pstate.curr.max; 1426 1427 min_clk = pstate_table->gfxclk_pstate.custom.min; 1428 max_clk = pstate_table->gfxclk_pstate.custom.max; 1429 1430 return aldebaran_set_soft_freq_limited_range(smu, SMU_GFXCLK, min_clk, max_clk); 1431 } 1432 break; 1433 default: 1434 return -ENOSYS; 1435 } 1436 1437 return ret; 1438 } 1439 1440 static bool aldebaran_is_dpm_running(struct smu_context *smu) 1441 { 1442 int ret; 1443 uint32_t feature_mask[2]; 1444 unsigned long feature_enabled; 1445 1446 ret = smu_cmn_get_enabled_mask(smu, feature_mask, 2); 1447 if (ret) 1448 return false; 1449 feature_enabled = (unsigned long)((uint64_t)feature_mask[0] | 1450 ((uint64_t)feature_mask[1] << 32)); 1451 return !!(feature_enabled & SMC_DPM_FEATURE); 1452 } 1453 1454 static int aldebaran_i2c_xfer(struct i2c_adapter *i2c_adap, 1455 struct i2c_msg *msg, int num_msgs) 1456 { 1457 struct amdgpu_device *adev = to_amdgpu_device(i2c_adap); 1458 struct smu_table_context *smu_table = &adev->smu.smu_table; 1459 struct smu_table *table = &smu_table->driver_table; 1460 SwI2cRequest_t *req, *res = (SwI2cRequest_t *)table->cpu_addr; 1461 int i, j, r, c; 1462 u16 dir; 1463 1464 req = kzalloc(sizeof(*req), GFP_KERNEL); 1465 if (!req) 1466 return -ENOMEM; 1467 1468 req->I2CcontrollerPort = 0; 1469 req->I2CSpeed = I2C_SPEED_FAST_400K; 1470 req->SlaveAddress = msg[0].addr << 1; /* wants an 8-bit address */ 1471 dir = msg[0].flags & I2C_M_RD; 1472 1473 for (c = i = 0; i < num_msgs; i++) { 1474 for (j = 0; j < msg[i].len; j++, c++) { 1475 SwI2cCmd_t *cmd = &req->SwI2cCmds[c]; 1476 1477 if (!(msg[i].flags & I2C_M_RD)) { 1478 /* write */ 1479 cmd->CmdConfig |= CMDCONFIG_READWRITE_MASK; 1480 cmd->ReadWriteData = msg[i].buf[j]; 1481 } 1482 1483 if ((dir ^ msg[i].flags) & I2C_M_RD) { 1484 /* The direction changes. 1485 */ 1486 dir = msg[i].flags & I2C_M_RD; 1487 cmd->CmdConfig |= CMDCONFIG_RESTART_MASK; 1488 } 1489 1490 req->NumCmds++; 1491 1492 /* 1493 * Insert STOP if we are at the last byte of either last 1494 * message for the transaction or the client explicitly 1495 * requires a STOP at this particular message. 1496 */ 1497 if ((j == msg[i].len - 1) && 1498 ((i == num_msgs - 1) || (msg[i].flags & I2C_M_STOP))) { 1499 cmd->CmdConfig &= ~CMDCONFIG_RESTART_MASK; 1500 cmd->CmdConfig |= CMDCONFIG_STOP_MASK; 1501 } 1502 } 1503 } 1504 mutex_lock(&adev->smu.mutex); 1505 r = smu_cmn_update_table(&adev->smu, SMU_TABLE_I2C_COMMANDS, 0, req, true); 1506 mutex_unlock(&adev->smu.mutex); 1507 if (r) 1508 goto fail; 1509 1510 for (c = i = 0; i < num_msgs; i++) { 1511 if (!(msg[i].flags & I2C_M_RD)) { 1512 c += msg[i].len; 1513 continue; 1514 } 1515 for (j = 0; j < msg[i].len; j++, c++) { 1516 SwI2cCmd_t *cmd = &res->SwI2cCmds[c]; 1517 1518 msg[i].buf[j] = cmd->ReadWriteData; 1519 } 1520 } 1521 r = num_msgs; 1522 fail: 1523 kfree(req); 1524 return r; 1525 } 1526 1527 static u32 aldebaran_i2c_func(struct i2c_adapter *adap) 1528 { 1529 return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL; 1530 } 1531 1532 1533 static const struct i2c_algorithm aldebaran_i2c_algo = { 1534 .master_xfer = aldebaran_i2c_xfer, 1535 .functionality = aldebaran_i2c_func, 1536 }; 1537 1538 static const struct i2c_adapter_quirks aldebaran_i2c_control_quirks = { 1539 .flags = I2C_AQ_COMB | I2C_AQ_COMB_SAME_ADDR | I2C_AQ_NO_ZERO_LEN, 1540 .max_read_len = MAX_SW_I2C_COMMANDS, 1541 .max_write_len = MAX_SW_I2C_COMMANDS, 1542 .max_comb_1st_msg_len = 2, 1543 .max_comb_2nd_msg_len = MAX_SW_I2C_COMMANDS - 2, 1544 }; 1545 1546 static int aldebaran_i2c_control_init(struct smu_context *smu, struct i2c_adapter *control) 1547 { 1548 struct amdgpu_device *adev = to_amdgpu_device(control); 1549 int res; 1550 1551 control->owner = THIS_MODULE; 1552 control->class = I2C_CLASS_SPD; 1553 control->dev.parent = &adev->pdev->dev; 1554 control->algo = &aldebaran_i2c_algo; 1555 snprintf(control->name, sizeof(control->name), "AMDGPU SMU"); 1556 control->quirks = &aldebaran_i2c_control_quirks; 1557 1558 res = i2c_add_adapter(control); 1559 if (res) 1560 DRM_ERROR("Failed to register hw i2c, err: %d\n", res); 1561 1562 return res; 1563 } 1564 1565 static void aldebaran_i2c_control_fini(struct smu_context *smu, struct i2c_adapter *control) 1566 { 1567 i2c_del_adapter(control); 1568 } 1569 1570 static void aldebaran_get_unique_id(struct smu_context *smu) 1571 { 1572 struct amdgpu_device *adev = smu->adev; 1573 SmuMetrics_t *metrics = smu->smu_table.metrics_table; 1574 uint32_t upper32 = 0, lower32 = 0; 1575 int ret; 1576 1577 mutex_lock(&smu->metrics_lock); 1578 ret = smu_cmn_get_metrics_table_locked(smu, NULL, false); 1579 if (ret) 1580 goto out_unlock; 1581 1582 upper32 = metrics->PublicSerialNumUpper32; 1583 lower32 = metrics->PublicSerialNumLower32; 1584 1585 out_unlock: 1586 mutex_unlock(&smu->metrics_lock); 1587 1588 adev->unique_id = ((uint64_t)upper32 << 32) | lower32; 1589 sprintf(adev->serial, "%016llx", adev->unique_id); 1590 } 1591 1592 static bool aldebaran_is_baco_supported(struct smu_context *smu) 1593 { 1594 /* aldebaran is not support baco */ 1595 1596 return false; 1597 } 1598 1599 static int aldebaran_set_df_cstate(struct smu_context *smu, 1600 enum pp_df_cstate state) 1601 { 1602 return smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_DFCstateControl, state, NULL); 1603 } 1604 1605 static int aldebaran_allow_xgmi_power_down(struct smu_context *smu, bool en) 1606 { 1607 return smu_cmn_send_smc_msg_with_param(smu, 1608 SMU_MSG_GmiPwrDnControl, 1609 en ? 1 : 0, 1610 NULL); 1611 } 1612 1613 static const struct throttling_logging_label { 1614 uint32_t feature_mask; 1615 const char *label; 1616 } logging_label[] = { 1617 {(1U << THROTTLER_TEMP_MEM_BIT), "HBM"}, 1618 {(1U << THROTTLER_TEMP_VR_GFX_BIT), "VR of GFX rail"}, 1619 {(1U << THROTTLER_TEMP_VR_MEM_BIT), "VR of HBM rail"}, 1620 {(1U << THROTTLER_TEMP_VR_SOC_BIT), "VR of SOC rail"}, 1621 }; 1622 static void aldebaran_log_thermal_throttling_event(struct smu_context *smu) 1623 { 1624 int ret; 1625 int throttler_idx, throtting_events = 0, buf_idx = 0; 1626 struct amdgpu_device *adev = smu->adev; 1627 uint32_t throttler_status; 1628 char log_buf[256]; 1629 1630 ret = aldebaran_get_smu_metrics_data(smu, 1631 METRICS_THROTTLER_STATUS, 1632 &throttler_status); 1633 if (ret) 1634 return; 1635 1636 memset(log_buf, 0, sizeof(log_buf)); 1637 for (throttler_idx = 0; throttler_idx < ARRAY_SIZE(logging_label); 1638 throttler_idx++) { 1639 if (throttler_status & logging_label[throttler_idx].feature_mask) { 1640 throtting_events++; 1641 buf_idx += snprintf(log_buf + buf_idx, 1642 sizeof(log_buf) - buf_idx, 1643 "%s%s", 1644 throtting_events > 1 ? " and " : "", 1645 logging_label[throttler_idx].label); 1646 if (buf_idx >= sizeof(log_buf)) { 1647 dev_err(adev->dev, "buffer overflow!\n"); 1648 log_buf[sizeof(log_buf) - 1] = '\0'; 1649 break; 1650 } 1651 } 1652 } 1653 1654 dev_warn(adev->dev, "WARN: GPU thermal throttling temperature reached, expect performance decrease. %s.\n", 1655 log_buf); 1656 kgd2kfd_smi_event_throttle(smu->adev->kfd.dev, 1657 smu_cmn_get_indep_throttler_status(throttler_status, 1658 aldebaran_throttler_map)); 1659 } 1660 1661 static int aldebaran_get_current_pcie_link_speed(struct smu_context *smu) 1662 { 1663 struct amdgpu_device *adev = smu->adev; 1664 uint32_t esm_ctrl; 1665 1666 /* TODO: confirm this on real target */ 1667 esm_ctrl = RREG32_PCIE(smnPCIE_ESM_CTRL); 1668 if ((esm_ctrl >> 15) & 0x1FFFF) 1669 return (((esm_ctrl >> 8) & 0x3F) + 128); 1670 1671 return smu_v13_0_get_current_pcie_link_speed(smu); 1672 } 1673 1674 static ssize_t aldebaran_get_gpu_metrics(struct smu_context *smu, 1675 void **table) 1676 { 1677 struct smu_table_context *smu_table = &smu->smu_table; 1678 struct gpu_metrics_v1_3 *gpu_metrics = 1679 (struct gpu_metrics_v1_3 *)smu_table->gpu_metrics_table; 1680 SmuMetrics_t metrics; 1681 int i, ret = 0; 1682 1683 ret = smu_cmn_get_metrics_table(smu, 1684 &metrics, 1685 true); 1686 if (ret) 1687 return ret; 1688 1689 smu_cmn_init_soft_gpu_metrics(gpu_metrics, 1, 3); 1690 1691 gpu_metrics->temperature_edge = metrics.TemperatureEdge; 1692 gpu_metrics->temperature_hotspot = metrics.TemperatureHotspot; 1693 gpu_metrics->temperature_mem = metrics.TemperatureHBM; 1694 gpu_metrics->temperature_vrgfx = metrics.TemperatureVrGfx; 1695 gpu_metrics->temperature_vrsoc = metrics.TemperatureVrSoc; 1696 gpu_metrics->temperature_vrmem = metrics.TemperatureVrMem; 1697 1698 gpu_metrics->average_gfx_activity = metrics.AverageGfxActivity; 1699 gpu_metrics->average_umc_activity = metrics.AverageUclkActivity; 1700 gpu_metrics->average_mm_activity = 0; 1701 1702 /* Valid power data is available only from primary die */ 1703 if (aldebaran_is_primary(smu)) { 1704 gpu_metrics->average_socket_power = metrics.AverageSocketPower; 1705 gpu_metrics->energy_accumulator = 1706 (uint64_t)metrics.EnergyAcc64bitHigh << 32 | 1707 metrics.EnergyAcc64bitLow; 1708 } else { 1709 gpu_metrics->average_socket_power = 0; 1710 gpu_metrics->energy_accumulator = 0; 1711 } 1712 1713 gpu_metrics->average_gfxclk_frequency = metrics.AverageGfxclkFrequency; 1714 gpu_metrics->average_socclk_frequency = metrics.AverageSocclkFrequency; 1715 gpu_metrics->average_uclk_frequency = metrics.AverageUclkFrequency; 1716 gpu_metrics->average_vclk0_frequency = 0; 1717 gpu_metrics->average_dclk0_frequency = 0; 1718 1719 gpu_metrics->current_gfxclk = metrics.CurrClock[PPCLK_GFXCLK]; 1720 gpu_metrics->current_socclk = metrics.CurrClock[PPCLK_SOCCLK]; 1721 gpu_metrics->current_uclk = metrics.CurrClock[PPCLK_UCLK]; 1722 gpu_metrics->current_vclk0 = metrics.CurrClock[PPCLK_VCLK]; 1723 gpu_metrics->current_dclk0 = metrics.CurrClock[PPCLK_DCLK]; 1724 1725 gpu_metrics->throttle_status = metrics.ThrottlerStatus; 1726 gpu_metrics->indep_throttle_status = 1727 smu_cmn_get_indep_throttler_status(metrics.ThrottlerStatus, 1728 aldebaran_throttler_map); 1729 1730 gpu_metrics->current_fan_speed = 0; 1731 1732 gpu_metrics->pcie_link_width = 1733 smu_v13_0_get_current_pcie_link_width(smu); 1734 gpu_metrics->pcie_link_speed = 1735 aldebaran_get_current_pcie_link_speed(smu); 1736 1737 gpu_metrics->system_clock_counter = ktime_get_boottime_ns(); 1738 1739 gpu_metrics->gfx_activity_acc = metrics.GfxBusyAcc; 1740 gpu_metrics->mem_activity_acc = metrics.DramBusyAcc; 1741 1742 for (i = 0; i < NUM_HBM_INSTANCES; i++) 1743 gpu_metrics->temperature_hbm[i] = metrics.TemperatureAllHBM[i]; 1744 1745 gpu_metrics->firmware_timestamp = ((uint64_t)metrics.TimeStampHigh << 32) | 1746 metrics.TimeStampLow; 1747 1748 *table = (void *)gpu_metrics; 1749 1750 return sizeof(struct gpu_metrics_v1_3); 1751 } 1752 1753 static int aldebaran_mode2_reset(struct smu_context *smu) 1754 { 1755 u32 smu_version; 1756 int ret = 0, index; 1757 struct amdgpu_device *adev = smu->adev; 1758 int timeout = 10; 1759 1760 smu_cmn_get_smc_version(smu, NULL, &smu_version); 1761 1762 index = smu_cmn_to_asic_specific_index(smu, CMN2ASIC_MAPPING_MSG, 1763 SMU_MSG_GfxDeviceDriverReset); 1764 1765 mutex_lock(&smu->message_lock); 1766 if (smu_version >= 0x00441400) { 1767 ret = smu_cmn_send_msg_without_waiting(smu, (uint16_t)index, SMU_RESET_MODE_2); 1768 /* This is similar to FLR, wait till max FLR timeout */ 1769 msleep(100); 1770 dev_dbg(smu->adev->dev, "restore config space...\n"); 1771 /* Restore the config space saved during init */ 1772 amdgpu_device_load_pci_state(adev->pdev); 1773 1774 dev_dbg(smu->adev->dev, "wait for reset ack\n"); 1775 while (ret == -ETIME && timeout) { 1776 ret = smu_cmn_wait_for_response(smu); 1777 /* Wait a bit more time for getting ACK */ 1778 if (ret == -ETIME) { 1779 --timeout; 1780 usleep_range(500, 1000); 1781 continue; 1782 } 1783 1784 if (ret != 1) { 1785 dev_err(adev->dev, "failed to send mode2 message \tparam: 0x%08x response %#x\n", 1786 SMU_RESET_MODE_2, ret); 1787 goto out; 1788 } 1789 } 1790 1791 } else { 1792 dev_err(adev->dev, "smu fw 0x%x does not support MSG_GfxDeviceDriverReset MSG\n", 1793 smu_version); 1794 } 1795 1796 if (ret == 1) 1797 ret = 0; 1798 out: 1799 mutex_unlock(&smu->message_lock); 1800 1801 return ret; 1802 } 1803 1804 static bool aldebaran_is_mode1_reset_supported(struct smu_context *smu) 1805 { 1806 #if 0 1807 struct amdgpu_device *adev = smu->adev; 1808 u32 smu_version; 1809 uint32_t val; 1810 /** 1811 * PM FW version support mode1 reset from 68.07 1812 */ 1813 smu_cmn_get_smc_version(smu, NULL, &smu_version); 1814 if ((smu_version < 0x00440700)) 1815 return false; 1816 /** 1817 * mode1 reset relies on PSP, so we should check if 1818 * PSP is alive. 1819 */ 1820 val = RREG32_SOC15(MP0, 0, regMP0_SMN_C2PMSG_81); 1821 1822 return val != 0x0; 1823 #endif 1824 return true; 1825 } 1826 1827 static bool aldebaran_is_mode2_reset_supported(struct smu_context *smu) 1828 { 1829 return true; 1830 } 1831 1832 static int aldebaran_set_mp1_state(struct smu_context *smu, 1833 enum pp_mp1_state mp1_state) 1834 { 1835 switch (mp1_state) { 1836 case PP_MP1_STATE_UNLOAD: 1837 return smu_cmn_set_mp1_state(smu, mp1_state); 1838 default: 1839 return 0; 1840 } 1841 } 1842 1843 static int aldebaran_smu_send_hbm_bad_page_num(struct smu_context *smu, 1844 uint32_t size) 1845 { 1846 int ret = 0; 1847 1848 /* message SMU to update the bad page number on SMUBUS */ 1849 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_SetNumBadHbmPagesRetired, size, NULL); 1850 if (ret) 1851 dev_err(smu->adev->dev, "[%s] failed to message SMU to update HBM bad pages number\n", 1852 __func__); 1853 1854 return ret; 1855 } 1856 1857 static const struct pptable_funcs aldebaran_ppt_funcs = { 1858 /* init dpm */ 1859 .get_allowed_feature_mask = aldebaran_get_allowed_feature_mask, 1860 /* dpm/clk tables */ 1861 .set_default_dpm_table = aldebaran_set_default_dpm_table, 1862 .populate_umd_state_clk = aldebaran_populate_umd_state_clk, 1863 .get_thermal_temperature_range = aldebaran_get_thermal_temperature_range, 1864 .print_clk_levels = aldebaran_print_clk_levels, 1865 .force_clk_levels = aldebaran_force_clk_levels, 1866 .read_sensor = aldebaran_read_sensor, 1867 .set_performance_level = aldebaran_set_performance_level, 1868 .get_power_limit = aldebaran_get_power_limit, 1869 .is_dpm_running = aldebaran_is_dpm_running, 1870 .get_unique_id = aldebaran_get_unique_id, 1871 .init_microcode = smu_v13_0_init_microcode, 1872 .load_microcode = smu_v13_0_load_microcode, 1873 .fini_microcode = smu_v13_0_fini_microcode, 1874 .init_smc_tables = aldebaran_init_smc_tables, 1875 .fini_smc_tables = smu_v13_0_fini_smc_tables, 1876 .init_power = smu_v13_0_init_power, 1877 .fini_power = smu_v13_0_fini_power, 1878 .check_fw_status = smu_v13_0_check_fw_status, 1879 /* pptable related */ 1880 .setup_pptable = aldebaran_setup_pptable, 1881 .get_vbios_bootup_values = smu_v13_0_get_vbios_bootup_values, 1882 .check_fw_version = smu_v13_0_check_fw_version, 1883 .write_pptable = smu_cmn_write_pptable, 1884 .set_driver_table_location = smu_v13_0_set_driver_table_location, 1885 .set_tool_table_location = smu_v13_0_set_tool_table_location, 1886 .notify_memory_pool_location = smu_v13_0_notify_memory_pool_location, 1887 .system_features_control = aldebaran_system_features_control, 1888 .send_smc_msg_with_param = smu_cmn_send_smc_msg_with_param, 1889 .send_smc_msg = smu_cmn_send_smc_msg, 1890 .get_enabled_mask = smu_cmn_get_enabled_mask, 1891 .feature_is_enabled = smu_cmn_feature_is_enabled, 1892 .disable_all_features_with_exception = smu_cmn_disable_all_features_with_exception, 1893 .set_power_limit = aldebaran_set_power_limit, 1894 .init_max_sustainable_clocks = smu_v13_0_init_max_sustainable_clocks, 1895 .enable_thermal_alert = smu_v13_0_enable_thermal_alert, 1896 .disable_thermal_alert = smu_v13_0_disable_thermal_alert, 1897 .set_xgmi_pstate = smu_v13_0_set_xgmi_pstate, 1898 .register_irq_handler = smu_v13_0_register_irq_handler, 1899 .set_azalia_d3_pme = smu_v13_0_set_azalia_d3_pme, 1900 .get_max_sustainable_clocks_by_dc = smu_v13_0_get_max_sustainable_clocks_by_dc, 1901 .baco_is_support= aldebaran_is_baco_supported, 1902 .get_dpm_ultimate_freq = smu_v13_0_get_dpm_ultimate_freq, 1903 .set_soft_freq_limited_range = aldebaran_set_soft_freq_limited_range, 1904 .od_edit_dpm_table = aldebaran_usr_edit_dpm_table, 1905 .set_df_cstate = aldebaran_set_df_cstate, 1906 .allow_xgmi_power_down = aldebaran_allow_xgmi_power_down, 1907 .log_thermal_throttling_event = aldebaran_log_thermal_throttling_event, 1908 .get_pp_feature_mask = smu_cmn_get_pp_feature_mask, 1909 .set_pp_feature_mask = smu_cmn_set_pp_feature_mask, 1910 .get_gpu_metrics = aldebaran_get_gpu_metrics, 1911 .mode1_reset_is_support = aldebaran_is_mode1_reset_supported, 1912 .mode2_reset_is_support = aldebaran_is_mode2_reset_supported, 1913 .mode1_reset = smu_v13_0_mode1_reset, 1914 .set_mp1_state = aldebaran_set_mp1_state, 1915 .mode2_reset = aldebaran_mode2_reset, 1916 .wait_for_event = smu_v13_0_wait_for_event, 1917 .i2c_init = aldebaran_i2c_control_init, 1918 .i2c_fini = aldebaran_i2c_control_fini, 1919 .send_hbm_bad_pages_num = aldebaran_smu_send_hbm_bad_page_num, 1920 }; 1921 1922 void aldebaran_set_ppt_funcs(struct smu_context *smu) 1923 { 1924 smu->ppt_funcs = &aldebaran_ppt_funcs; 1925 smu->message_map = aldebaran_message_map; 1926 smu->clock_map = aldebaran_clk_map; 1927 smu->feature_map = aldebaran_feature_mask_map; 1928 smu->table_map = aldebaran_table_map; 1929 } 1930