drivers/gpu/drm/amd/amdgpu/umc_v6_1.c

   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 #include "umc_v6_1.h"
  24 #include "amdgpu_ras.h"
  25 #include "amdgpu.h"
  26
  27 #include "rsmu/rsmu_0_0_2_offset.h"
  28 #include "rsmu/rsmu_0_0_2_sh_mask.h"
  29 #include "umc/umc_6_1_1_offset.h"
  30 #include "umc/umc_6_1_1_sh_mask.h"
  31 #include "umc/umc_6_1_2_offset.h"
  32
  33 #define UMC_6_INST_DIST                 0x40000
  34
  35 const uint32_t
  36         umc_v6_1_channel_idx_tbl[UMC_V6_1_UMC_INSTANCE_NUM][UMC_V6_1_CHANNEL_INSTANCE_NUM] = {
  37                 {2, 18, 11, 27},        {4, 20, 13, 29},
  38                 {1, 17, 8, 24},         {7, 23, 14, 30},
  39                 {10, 26, 3, 19},        {12, 28, 5, 21},
  40                 {9, 25, 0, 16},         {15, 31, 6, 22}
  41 };
  42
  43 static void umc_v6_1_enable_umc_index_mode(struct amdgpu_device *adev)
  44 {
  45         uint32_t rsmu_umc_addr, rsmu_umc_val;
  46
  47         rsmu_umc_addr = SOC15_REG_OFFSET(RSMU, 0,
  48                         mmRSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU);
  49         rsmu_umc_val = RREG32_PCIE(rsmu_umc_addr * 4);
  50
  51         rsmu_umc_val = REG_SET_FIELD(rsmu_umc_val,
  52                         RSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
  53                         RSMU_UMC_INDEX_MODE_EN, 1);
  54
  55         WREG32_PCIE(rsmu_umc_addr * 4, rsmu_umc_val);
  56 }
  57
  58 static void umc_v6_1_disable_umc_index_mode(struct amdgpu_device *adev)
  59 {
  60         uint32_t rsmu_umc_addr, rsmu_umc_val;
  61
  62         rsmu_umc_addr = SOC15_REG_OFFSET(RSMU, 0,
  63                         mmRSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU);
  64         rsmu_umc_val = RREG32_PCIE(rsmu_umc_addr * 4);
  65
  66         rsmu_umc_val = REG_SET_FIELD(rsmu_umc_val,
  67                         RSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
  68                         RSMU_UMC_INDEX_MODE_EN, 0);
  69
  70         WREG32_PCIE(rsmu_umc_addr * 4, rsmu_umc_val);
  71 }
  72
  73 static uint32_t umc_v6_1_get_umc_index_mode_state(struct amdgpu_device *adev)
  74 {
  75         uint32_t rsmu_umc_addr, rsmu_umc_val;
  76
  77         rsmu_umc_addr = SOC15_REG_OFFSET(RSMU, 0,
  78                         mmRSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU);
  79         rsmu_umc_val = RREG32_PCIE(rsmu_umc_addr * 4);
  80
  81         return REG_GET_FIELD(rsmu_umc_val,
  82                         RSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
  83                         RSMU_UMC_INDEX_MODE_EN);
  84 }
  85
  86 static inline uint32_t get_umc_6_reg_offset(struct amdgpu_device *adev,
  87                                             uint32_t umc_inst,
  88                                             uint32_t ch_inst)
  89 {
  90         return adev->umc.channel_offs*ch_inst + UMC_6_INST_DIST*umc_inst;
  91 }
  92
  93 static void umc_v6_1_clear_error_count_per_channel(struct amdgpu_device *adev,
  94                                         uint32_t umc_reg_offset)
  95 {
  96         uint32_t ecc_err_cnt_addr;
  97         uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
  98
  99         if (adev->asic_type == CHIP_ARCTURUS) {
 100                 /* UMC 6_1_2 registers */
 101                 ecc_err_cnt_sel_addr =
 102                         SOC15_REG_OFFSET(UMC, 0,
 103                                         mmUMCCH0_0_EccErrCntSel_ARCT);
 104                 ecc_err_cnt_addr =
 105                         SOC15_REG_OFFSET(UMC, 0,
 106                                         mmUMCCH0_0_EccErrCnt_ARCT);
 107         } else {
 108                 /* UMC 6_1_1 registers */
 109                 ecc_err_cnt_sel_addr =
 110                         SOC15_REG_OFFSET(UMC, 0,
 111                                         mmUMCCH0_0_EccErrCntSel);
 112                 ecc_err_cnt_addr =
 113                         SOC15_REG_OFFSET(UMC, 0,
 114                                         mmUMCCH0_0_EccErrCnt);
 115         }
 116
 117         /* select the lower chip */
 118         ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr +
 119                                         umc_reg_offset) * 4);
 120         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel,
 121                                         UMCCH0_0_EccErrCntSel,
 122                                         EccErrCntCsSel, 0);
 123         WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4,
 124                         ecc_err_cnt_sel);
 125
 126         /* clear lower chip error count */
 127         WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4,
 128                         UMC_V6_1_CE_CNT_INIT);
 129
 130         /* select the higher chip */
 131         ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr +
 132                                         umc_reg_offset) * 4);
 133         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel,
 134                                         UMCCH0_0_EccErrCntSel,
 135                                         EccErrCntCsSel, 1);
 136         WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4,
 137                         ecc_err_cnt_sel);
 138
 139         /* clear higher chip error count */
 140         WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4,
 141                         UMC_V6_1_CE_CNT_INIT);
 142 }
 143
 144 static void umc_v6_1_clear_error_count(struct amdgpu_device *adev)
 145 {
 146         uint32_t umc_inst        = 0;
 147         uint32_t ch_inst         = 0;
 148         uint32_t umc_reg_offset  = 0;
 149         uint32_t rsmu_umc_index_state =
 150                                 umc_v6_1_get_umc_index_mode_state(adev);
 151
 152         if (rsmu_umc_index_state)
 153                 umc_v6_1_disable_umc_index_mode(adev);
 154
 155         LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
 156                 umc_reg_offset = get_umc_6_reg_offset(adev,
 157                                                 umc_inst,
 158                                                 ch_inst);
 159
 160                 umc_v6_1_clear_error_count_per_channel(adev,
 161                                                 umc_reg_offset);
 162         }
 163
 164         if (rsmu_umc_index_state)
 165                 umc_v6_1_enable_umc_index_mode(adev);
 166 }
 167
 168 static void umc_v6_1_query_correctable_error_count(struct amdgpu_device *adev,
 169                                                    uint32_t umc_reg_offset,
 170                                                    unsigned long *error_count)
 171 {
 172         uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
 173         uint32_t ecc_err_cnt, ecc_err_cnt_addr;
 174         uint64_t mc_umc_status;
 175         uint32_t mc_umc_status_addr;
 176
 177         if (adev->asic_type == CHIP_ARCTURUS) {
 178                 /* UMC 6_1_2 registers */
 179                 ecc_err_cnt_sel_addr =
 180                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCntSel_ARCT);
 181                 ecc_err_cnt_addr =
 182                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCnt_ARCT);
 183                 mc_umc_status_addr =
 184                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0_ARCT);
 185         } else {
 186                 /* UMC 6_1_1 registers */
 187                 ecc_err_cnt_sel_addr =
 188                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCntSel);
 189                 ecc_err_cnt_addr =
 190                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCnt);
 191                 mc_umc_status_addr =
 192                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
 193         }
 194
 195         /* select the lower chip and check the error count */
 196         ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4);
 197         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel,
 198                                         EccErrCntCsSel, 0);
 199         WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
 200
 201         ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4);
 202         *error_count +=
 203                 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_EccErrCnt, EccErrCnt) -
 204                  UMC_V6_1_CE_CNT_INIT);
 205
 206         /* select the higher chip and check the err counter */
 207         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel,
 208                                         EccErrCntCsSel, 1);
 209         WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
 210
 211         ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4);
 212         *error_count +=
 213                 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_EccErrCnt, EccErrCnt) -
 214                  UMC_V6_1_CE_CNT_INIT);
 215
 216         /* check for SRAM correctable error
 217           MCUMC_STATUS is a 64 bit register */
 218         mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
 219         if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 6 &&
 220             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
 221             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
 222                 *error_count += 1;
 223 }
 224
 225 static void umc_v6_1_querry_uncorrectable_error_count(struct amdgpu_device *adev,
 226                                                       uint32_t umc_reg_offset,
 227                                                       unsigned long *error_count)
 228 {
 229         uint64_t mc_umc_status;
 230         uint32_t mc_umc_status_addr;
 231
 232         if (adev->asic_type == CHIP_ARCTURUS) {
 233                 /* UMC 6_1_2 registers */
 234                 mc_umc_status_addr =
 235                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0_ARCT);
 236         } else {
 237                 /* UMC 6_1_1 registers */
 238                 mc_umc_status_addr =
 239                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
 240         }
 241
 242         /* check the MCUMC_STATUS */
 243         mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
 244         if ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
 245             (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
 246             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
 247             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
 248             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
 249             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
 250                 *error_count += 1;
 251 }
 252
 253 static void umc_v6_1_query_ras_error_count(struct amdgpu_device *adev,
 254                                            void *ras_error_status)
 255 {
 256         struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
 257
 258         uint32_t umc_inst        = 0;
 259         uint32_t ch_inst         = 0;
 260         uint32_t umc_reg_offset  = 0;
 261
 262         uint32_t rsmu_umc_index_state = umc_v6_1_get_umc_index_mode_state(adev);
 263
 264         if (rsmu_umc_index_state)
 265                 umc_v6_1_disable_umc_index_mode(adev);
 266
 267         if ((adev->asic_type == CHIP_ARCTURUS) &&
 268                 amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_DISALLOW))
 269                 DRM_WARN("Fail to disable DF-Cstate.\n");
 270
 271         LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
 272                 umc_reg_offset = get_umc_6_reg_offset(adev,
 273                                                       umc_inst,
 274                                                       ch_inst);
 275
 276                 umc_v6_1_query_correctable_error_count(adev,
 277                                                        umc_reg_offset,
 278                                                        &(err_data->ce_count));
 279                 umc_v6_1_querry_uncorrectable_error_count(adev,
 280                                                           umc_reg_offset,
 281                                                           &(err_data->ue_count));
 282         }
 283
 284         if ((adev->asic_type == CHIP_ARCTURUS) &&
 285                 amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_ALLOW))
 286                 DRM_WARN("Fail to enable DF-Cstate\n");
 287
 288         if (rsmu_umc_index_state)
 289                 umc_v6_1_enable_umc_index_mode(adev);
 290
 291         umc_v6_1_clear_error_count(adev);
 292 }
 293
 294 static void umc_v6_1_query_error_address(struct amdgpu_device *adev,
 295                                          struct ras_err_data *err_data,
 296                                          uint32_t umc_reg_offset,
 297                                          uint32_t ch_inst,
 298                                          uint32_t umc_inst)
 299 {
 300         uint32_t lsb, mc_umc_status_addr;
 301         uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
 302         struct eeprom_table_record *err_rec;
 303         uint32_t channel_index = adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst];
 304
 305         if (adev->asic_type == CHIP_ARCTURUS) {
 306                 /* UMC 6_1_2 registers */
 307                 mc_umc_status_addr =
 308                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0_ARCT);
 309                 mc_umc_addrt0 =
 310                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_ADDRT0_ARCT);
 311         } else {
 312                 /* UMC 6_1_1 registers */
 313                 mc_umc_status_addr =
 314                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
 315                 mc_umc_addrt0 =
 316                         SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_ADDRT0);
 317         }
 318
 319         mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
 320
 321         if (mc_umc_status == 0)
 322                 return;
 323
 324         if (!err_data->err_addr) {
 325                 /* clear umc status */
 326                 WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
 327                 return;
 328         }
 329
 330         err_rec = &err_data->err_addr[err_data->err_addr_cnt];
 331
 332         /* calculate error address if ue/ce error is detected */
 333         if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
 334             (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
 335             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)) {
 336
 337                 err_addr = RREG64_PCIE((mc_umc_addrt0 + umc_reg_offset) * 4);
 338                 /* the lowest lsb bits should be ignored */
 339                 lsb = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, LSB);
 340                 err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
 341                 err_addr &= ~((0x1ULL << lsb) - 1);
 342
 343                 /* translate umc channel address to soc pa, 3 parts are included */
 344                 retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
 345                                 ADDR_OF_256B_BLOCK(channel_index) |
 346                                 OFFSET_IN_256B_BLOCK(err_addr);
 347
 348                 /* we only save ue error information currently, ce is skipped */
 349                 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
 350                                 == 1) {
 351                         err_rec->address = err_addr;
 352                         /* page frame address is saved */
 353                         err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
 354                         err_rec->ts = (uint64_t)ktime_get_real_seconds();
 355                         err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
 356                         err_rec->cu = 0;
 357                         err_rec->mem_channel = channel_index;
 358                         err_rec->mcumc_id = umc_inst;
 359
 360                         err_data->err_addr_cnt++;
 361                 }
 362         }
 363
 364         /* clear umc status */
 365         WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
 366 }
 367
 368 static void umc_v6_1_query_ras_error_address(struct amdgpu_device *adev,
 369                                              void *ras_error_status)
 370 {
 371         struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
 372
 373         uint32_t umc_inst        = 0;
 374         uint32_t ch_inst         = 0;
 375         uint32_t umc_reg_offset  = 0;
 376
 377         uint32_t rsmu_umc_index_state = umc_v6_1_get_umc_index_mode_state(adev);
 378
 379         if (rsmu_umc_index_state)
 380                 umc_v6_1_disable_umc_index_mode(adev);
 381
 382         if ((adev->asic_type == CHIP_ARCTURUS) &&
 383                 amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_DISALLOW))
 384                 DRM_WARN("Fail to disable DF-Cstate.\n");
 385
 386         LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
 387                 umc_reg_offset = get_umc_6_reg_offset(adev,
 388                                                       umc_inst,
 389                                                       ch_inst);
 390
 391                 umc_v6_1_query_error_address(adev,
 392                                              err_data,
 393                                              umc_reg_offset,
 394                                              ch_inst,
 395                                              umc_inst);
 396         }
 397
 398         if ((adev->asic_type == CHIP_ARCTURUS) &&
 399                 amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_ALLOW))
 400                 DRM_WARN("Fail to enable DF-Cstate\n");
 401
 402         if (rsmu_umc_index_state)
 403                 umc_v6_1_enable_umc_index_mode(adev);
 404 }
 405
 406 static void umc_v6_1_err_cnt_init_per_channel(struct amdgpu_device *adev,
 407                                               uint32_t umc_reg_offset)
 408 {
 409         uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
 410         uint32_t ecc_err_cnt_addr;
 411
 412         if (adev->asic_type == CHIP_ARCTURUS) {
 413                 /* UMC 6_1_2 registers */
 414                 ecc_err_cnt_sel_addr =
 415                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCntSel_ARCT);
 416                 ecc_err_cnt_addr =
 417                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCnt_ARCT);
 418         } else {
 419                 /* UMC 6_1_1 registers */
 420                 ecc_err_cnt_sel_addr =
 421                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCntSel);
 422                 ecc_err_cnt_addr =
 423                         SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCnt);
 424         }
 425
 426         /* select the lower chip and check the error count */
 427         ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4);
 428         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel,
 429                                         EccErrCntCsSel, 0);
 430         /* set ce error interrupt type to APIC based interrupt */
 431         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel,
 432                                         EccErrInt, 0x1);
 433         WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
 434         /* set error count to initial value */
 435         WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V6_1_CE_CNT_INIT);
 436
 437         /* select the higher chip and check the err counter */
 438         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel,
 439                                         EccErrCntCsSel, 1);
 440         WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
 441         WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V6_1_CE_CNT_INIT);
 442 }
 443
 444 static void umc_v6_1_err_cnt_init(struct amdgpu_device *adev)
 445 {
 446         uint32_t umc_inst        = 0;
 447         uint32_t ch_inst         = 0;
 448         uint32_t umc_reg_offset  = 0;
 449
 450         uint32_t rsmu_umc_index_state = umc_v6_1_get_umc_index_mode_state(adev);
 451
 452         if (rsmu_umc_index_state)
 453                 umc_v6_1_disable_umc_index_mode(adev);
 454
 455         LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
 456                 umc_reg_offset = get_umc_6_reg_offset(adev,
 457                                                       umc_inst,
 458                                                       ch_inst);
 459
 460                 umc_v6_1_err_cnt_init_per_channel(adev, umc_reg_offset);
 461         }
 462
 463         if (rsmu_umc_index_state)
 464                 umc_v6_1_enable_umc_index_mode(adev);
 465 }
 466
 467 const struct amdgpu_umc_funcs umc_v6_1_funcs = {
 468         .err_cnt_init = umc_v6_1_err_cnt_init,
 469         .ras_late_init = amdgpu_umc_ras_late_init,
 470         .query_ras_error_count = umc_v6_1_query_ras_error_count,
 471         .query_ras_error_address = umc_v6_1_query_ras_error_address,
 472 };