umc_v6_7.c (18717B)
1/* 2 * Copyright 2021 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_7.h" 24#include "amdgpu_ras.h" 25#include "amdgpu_umc.h" 26#include "amdgpu.h" 27 28#include "umc/umc_6_7_0_offset.h" 29#include "umc/umc_6_7_0_sh_mask.h" 30 31const uint32_t 32 umc_v6_7_channel_idx_tbl_second[UMC_V6_7_UMC_INSTANCE_NUM][UMC_V6_7_CHANNEL_INSTANCE_NUM] = { 33 {28, 20, 24, 16, 12, 4, 8, 0}, 34 {6, 30, 2, 26, 22, 14, 18, 10}, 35 {19, 11, 15, 7, 3, 27, 31, 23}, 36 {9, 1, 5, 29, 25, 17, 21, 13} 37}; 38const uint32_t 39 umc_v6_7_channel_idx_tbl_first[UMC_V6_7_UMC_INSTANCE_NUM][UMC_V6_7_CHANNEL_INSTANCE_NUM] = { 40 {19, 11, 15, 7, 3, 27, 31, 23}, 41 {9, 1, 5, 29, 25, 17, 21, 13}, 42 {28, 20, 24, 16, 12, 4, 8, 0}, 43 {6, 30, 2, 26, 22, 14, 18, 10}, 44}; 45 46static inline uint32_t get_umc_v6_7_reg_offset(struct amdgpu_device *adev, 47 uint32_t umc_inst, 48 uint32_t ch_inst) 49{ 50 uint32_t index = umc_inst * adev->umc.channel_inst_num + ch_inst; 51 52 /* adjust umc and channel index offset, 53 * the register address is not linear on each umc instace */ 54 umc_inst = index / 4; 55 ch_inst = index % 4; 56 57 return adev->umc.channel_offs * ch_inst + UMC_V6_7_INST_DIST * umc_inst; 58} 59 60static inline uint32_t get_umc_v6_7_channel_index(struct amdgpu_device *adev, 61 uint32_t umc_inst, 62 uint32_t ch_inst) 63{ 64 return adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst]; 65} 66 67static void umc_v6_7_query_error_status_helper(struct amdgpu_device *adev, 68 uint64_t mc_umc_status, uint32_t umc_reg_offset) 69{ 70 uint32_t mc_umc_addr; 71 uint64_t reg_value; 72 73 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1) 74 dev_info(adev->dev, "Deferred error, no user action is needed.\n"); 75 76 if (mc_umc_status) 77 dev_info(adev->dev, "MCA STATUS 0x%llx, umc_reg_offset 0x%x\n", mc_umc_status, umc_reg_offset); 78 79 /* print IPID registers value */ 80 mc_umc_addr = 81 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_IPIDT0); 82 reg_value = RREG64_PCIE((mc_umc_addr + umc_reg_offset) * 4); 83 if (reg_value) 84 dev_info(adev->dev, "MCA IPID 0x%llx, umc_reg_offset 0x%x\n", reg_value, umc_reg_offset); 85 86 /* print SYND registers value */ 87 mc_umc_addr = 88 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_SYNDT0); 89 reg_value = RREG64_PCIE((mc_umc_addr + umc_reg_offset) * 4); 90 if (reg_value) 91 dev_info(adev->dev, "MCA SYND 0x%llx, umc_reg_offset 0x%x\n", reg_value, umc_reg_offset); 92 93 /* print MISC0 registers value */ 94 mc_umc_addr = 95 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_MISC0T0); 96 reg_value = RREG64_PCIE((mc_umc_addr + umc_reg_offset) * 4); 97 if (reg_value) 98 dev_info(adev->dev, "MCA MISC0 0x%llx, umc_reg_offset 0x%x\n", reg_value, umc_reg_offset); 99} 100 101static void umc_v6_7_ecc_info_query_correctable_error_count(struct amdgpu_device *adev, 102 uint32_t umc_inst, uint32_t ch_inst, 103 unsigned long *error_count) 104{ 105 uint64_t mc_umc_status; 106 uint32_t eccinfo_table_idx; 107 uint32_t umc_reg_offset; 108 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 109 110 umc_reg_offset = get_umc_v6_7_reg_offset(adev, 111 umc_inst, ch_inst); 112 113 eccinfo_table_idx = umc_inst * adev->umc.channel_inst_num + ch_inst; 114 /* check for SRAM correctable error 115 MCUMC_STATUS is a 64 bit register */ 116 mc_umc_status = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_status; 117 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 && 118 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1) { 119 *error_count += 1; 120 121 umc_v6_7_query_error_status_helper(adev, mc_umc_status, umc_reg_offset); 122 } 123} 124 125static void umc_v6_7_ecc_info_querry_uncorrectable_error_count(struct amdgpu_device *adev, 126 uint32_t umc_inst, uint32_t ch_inst, 127 unsigned long *error_count) 128{ 129 uint64_t mc_umc_status; 130 uint32_t eccinfo_table_idx; 131 uint32_t umc_reg_offset; 132 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 133 134 umc_reg_offset = get_umc_v6_7_reg_offset(adev, 135 umc_inst, ch_inst); 136 137 eccinfo_table_idx = umc_inst * adev->umc.channel_inst_num + ch_inst; 138 /* check the MCUMC_STATUS */ 139 mc_umc_status = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_status; 140 if ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) && 141 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 || 142 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 || 143 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 || 144 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 || 145 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1)) { 146 *error_count += 1; 147 148 umc_v6_7_query_error_status_helper(adev, mc_umc_status, umc_reg_offset); 149 } 150} 151 152static void umc_v6_7_ecc_info_query_ras_error_count(struct amdgpu_device *adev, 153 void *ras_error_status) 154{ 155 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 156 157 uint32_t umc_inst = 0; 158 uint32_t ch_inst = 0; 159 160 /*TODO: driver needs to toggle DF Cstate to ensure 161 * safe access of UMC registers. Will add the protection */ 162 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) { 163 umc_v6_7_ecc_info_query_correctable_error_count(adev, 164 umc_inst, ch_inst, 165 &(err_data->ce_count)); 166 umc_v6_7_ecc_info_querry_uncorrectable_error_count(adev, 167 umc_inst, ch_inst, 168 &(err_data->ue_count)); 169 } 170} 171 172static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev, 173 struct ras_err_data *err_data, 174 uint32_t ch_inst, 175 uint32_t umc_inst) 176{ 177 uint64_t mc_umc_status, err_addr, soc_pa, retired_page, column; 178 uint32_t channel_index; 179 uint32_t eccinfo_table_idx; 180 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 181 182 eccinfo_table_idx = umc_inst * adev->umc.channel_inst_num + ch_inst; 183 channel_index = 184 adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst]; 185 186 mc_umc_status = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_status; 187 188 if (mc_umc_status == 0) 189 return; 190 191 if (!err_data->err_addr) 192 return; 193 194 /* calculate error address if ue/ce error is detected */ 195 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 && 196 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 || 197 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)) { 198 199 err_addr = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_addr; 200 err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr); 201 202 /* translate umc channel address to soc pa, 3 parts are included */ 203 soc_pa = ADDR_OF_8KB_BLOCK(err_addr) | 204 ADDR_OF_256B_BLOCK(channel_index) | 205 OFFSET_IN_256B_BLOCK(err_addr); 206 207 /* The umc channel bits are not original values, they are hashed */ 208 SET_CHANNEL_HASH(channel_index, soc_pa); 209 210 /* clear [C4 C3 C2] in soc physical address */ 211 soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT); 212 213 /* we only save ue error information currently, ce is skipped */ 214 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) 215 == 1) { 216 /* loop for all possibilities of [C4 C3 C2] */ 217 for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; column++) { 218 retired_page = soc_pa | (column << UMC_V6_7_PA_C2_BIT); 219 dev_info(adev->dev, "Error Address(PA): 0x%llx\n", retired_page); 220 amdgpu_umc_fill_error_record(err_data, err_addr, 221 retired_page, channel_index, umc_inst); 222 223 /* shift R14 bit */ 224 retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT); 225 dev_info(adev->dev, "Error Address(PA): 0x%llx\n", retired_page); 226 amdgpu_umc_fill_error_record(err_data, err_addr, 227 retired_page, channel_index, umc_inst); 228 } 229 } 230 } 231} 232 233static void umc_v6_7_ecc_info_query_ras_error_address(struct amdgpu_device *adev, 234 void *ras_error_status) 235{ 236 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 237 238 uint32_t umc_inst = 0; 239 uint32_t ch_inst = 0; 240 241 /*TODO: driver needs to toggle DF Cstate to ensure 242 * safe access of UMC resgisters. Will add the protection 243 * when firmware interface is ready */ 244 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) { 245 umc_v6_7_ecc_info_query_error_address(adev, 246 err_data, 247 ch_inst, 248 umc_inst); 249 } 250} 251 252static void umc_v6_7_query_correctable_error_count(struct amdgpu_device *adev, 253 uint32_t umc_reg_offset, 254 unsigned long *error_count) 255{ 256 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr; 257 uint32_t ecc_err_cnt, ecc_err_cnt_addr; 258 uint64_t mc_umc_status; 259 uint32_t mc_umc_status_addr; 260 261 /* UMC 6_1_1 registers */ 262 ecc_err_cnt_sel_addr = 263 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_0_EccErrCntSel); 264 ecc_err_cnt_addr = 265 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_0_EccErrCnt); 266 mc_umc_status_addr = 267 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0); 268 269 /* select the lower chip and check the error count */ 270 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4); 271 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel, 272 EccErrCntCsSel, 0); 273 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel); 274 275 ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4); 276 *error_count += 277 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_EccErrCnt, EccErrCnt) - 278 UMC_V6_7_CE_CNT_INIT); 279 280 /* select the higher chip and check the err counter */ 281 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel, 282 EccErrCntCsSel, 1); 283 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel); 284 285 ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4); 286 *error_count += 287 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_EccErrCnt, EccErrCnt) - 288 UMC_V6_7_CE_CNT_INIT); 289 290 /* check for SRAM correctable error 291 MCUMC_STATUS is a 64 bit register */ 292 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4); 293 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 && 294 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1) { 295 *error_count += 1; 296 297 umc_v6_7_query_error_status_helper(adev, mc_umc_status, umc_reg_offset); 298 } 299} 300 301static void umc_v6_7_querry_uncorrectable_error_count(struct amdgpu_device *adev, 302 uint32_t umc_reg_offset, 303 unsigned long *error_count) 304{ 305 uint64_t mc_umc_status; 306 uint32_t mc_umc_status_addr; 307 308 mc_umc_status_addr = 309 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0); 310 311 /* check the MCUMC_STATUS */ 312 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4); 313 if ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) && 314 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 || 315 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 || 316 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 || 317 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 || 318 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1)) { 319 *error_count += 1; 320 321 umc_v6_7_query_error_status_helper(adev, mc_umc_status, umc_reg_offset); 322 } 323} 324 325static void umc_v6_7_reset_error_count_per_channel(struct amdgpu_device *adev, 326 uint32_t umc_reg_offset) 327{ 328 uint32_t ecc_err_cnt_addr; 329 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr; 330 331 ecc_err_cnt_sel_addr = 332 SOC15_REG_OFFSET(UMC, 0, 333 regUMCCH0_0_EccErrCntSel); 334 ecc_err_cnt_addr = 335 SOC15_REG_OFFSET(UMC, 0, 336 regUMCCH0_0_EccErrCnt); 337 338 /* select the lower chip */ 339 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + 340 umc_reg_offset) * 4); 341 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, 342 UMCCH0_0_EccErrCntSel, 343 EccErrCntCsSel, 0); 344 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, 345 ecc_err_cnt_sel); 346 347 /* clear lower chip error count */ 348 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, 349 UMC_V6_7_CE_CNT_INIT); 350 351 /* select the higher chip */ 352 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + 353 umc_reg_offset) * 4); 354 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, 355 UMCCH0_0_EccErrCntSel, 356 EccErrCntCsSel, 1); 357 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, 358 ecc_err_cnt_sel); 359 360 /* clear higher chip error count */ 361 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, 362 UMC_V6_7_CE_CNT_INIT); 363} 364 365static void umc_v6_7_reset_error_count(struct amdgpu_device *adev) 366{ 367 uint32_t umc_inst = 0; 368 uint32_t ch_inst = 0; 369 uint32_t umc_reg_offset = 0; 370 371 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) { 372 umc_reg_offset = get_umc_v6_7_reg_offset(adev, 373 umc_inst, 374 ch_inst); 375 376 umc_v6_7_reset_error_count_per_channel(adev, 377 umc_reg_offset); 378 } 379} 380 381static void umc_v6_7_query_ras_error_count(struct amdgpu_device *adev, 382 void *ras_error_status) 383{ 384 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 385 386 uint32_t umc_inst = 0; 387 uint32_t ch_inst = 0; 388 uint32_t umc_reg_offset = 0; 389 390 /*TODO: driver needs to toggle DF Cstate to ensure 391 * safe access of UMC registers. Will add the protection */ 392 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) { 393 umc_reg_offset = get_umc_v6_7_reg_offset(adev, 394 umc_inst, 395 ch_inst); 396 umc_v6_7_query_correctable_error_count(adev, 397 umc_reg_offset, 398 &(err_data->ce_count)); 399 umc_v6_7_querry_uncorrectable_error_count(adev, 400 umc_reg_offset, 401 &(err_data->ue_count)); 402 } 403 404 umc_v6_7_reset_error_count(adev); 405} 406 407static void umc_v6_7_query_error_address(struct amdgpu_device *adev, 408 struct ras_err_data *err_data, 409 uint32_t umc_reg_offset, 410 uint32_t ch_inst, 411 uint32_t umc_inst) 412{ 413 uint32_t mc_umc_status_addr; 414 uint32_t channel_index; 415 uint64_t mc_umc_status, mc_umc_addrt0; 416 uint64_t err_addr, soc_pa, retired_page, column; 417 418 mc_umc_status_addr = 419 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0); 420 mc_umc_addrt0 = 421 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_ADDRT0); 422 423 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4); 424 425 if (mc_umc_status == 0) 426 return; 427 428 if (!err_data->err_addr) { 429 /* clear umc status */ 430 WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL); 431 return; 432 } 433 434 channel_index = 435 adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst]; 436 437 /* calculate error address if ue/ce error is detected */ 438 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 && 439 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 || 440 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)) { 441 442 err_addr = RREG64_PCIE((mc_umc_addrt0 + umc_reg_offset) * 4); 443 err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr); 444 445 /* translate umc channel address to soc pa, 3 parts are included */ 446 soc_pa = ADDR_OF_8KB_BLOCK(err_addr) | 447 ADDR_OF_256B_BLOCK(channel_index) | 448 OFFSET_IN_256B_BLOCK(err_addr); 449 450 /* The umc channel bits are not original values, they are hashed */ 451 SET_CHANNEL_HASH(channel_index, soc_pa); 452 453 /* clear [C4 C3 C2] in soc physical address */ 454 soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT); 455 456 /* we only save ue error information currently, ce is skipped */ 457 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) 458 == 1) { 459 /* loop for all possibilities of [C4 C3 C2] */ 460 for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; column++) { 461 retired_page = soc_pa | (column << UMC_V6_7_PA_C2_BIT); 462 dev_info(adev->dev, "Error Address(PA): 0x%llx\n", retired_page); 463 amdgpu_umc_fill_error_record(err_data, err_addr, 464 retired_page, channel_index, umc_inst); 465 466 /* shift R14 bit */ 467 retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT); 468 dev_info(adev->dev, "Error Address(PA): 0x%llx\n", retired_page); 469 amdgpu_umc_fill_error_record(err_data, err_addr, 470 retired_page, channel_index, umc_inst); 471 } 472 } 473 } 474 475 /* clear umc status */ 476 WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL); 477} 478 479static void umc_v6_7_query_ras_error_address(struct amdgpu_device *adev, 480 void *ras_error_status) 481{ 482 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 483 484 uint32_t umc_inst = 0; 485 uint32_t ch_inst = 0; 486 uint32_t umc_reg_offset = 0; 487 488 /*TODO: driver needs to toggle DF Cstate to ensure 489 * safe access of UMC resgisters. Will add the protection 490 * when firmware interface is ready */ 491 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) { 492 umc_reg_offset = get_umc_v6_7_reg_offset(adev, 493 umc_inst, 494 ch_inst); 495 umc_v6_7_query_error_address(adev, 496 err_data, 497 umc_reg_offset, 498 ch_inst, 499 umc_inst); 500 } 501} 502 503static uint32_t umc_v6_7_query_ras_poison_mode_per_channel( 504 struct amdgpu_device *adev, 505 uint32_t umc_reg_offset) 506{ 507 uint32_t ecc_ctrl_addr, ecc_ctrl; 508 509 ecc_ctrl_addr = 510 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_0_EccCtrl); 511 ecc_ctrl = RREG32_PCIE((ecc_ctrl_addr + 512 umc_reg_offset) * 4); 513 514 return REG_GET_FIELD(ecc_ctrl, UMCCH0_0_EccCtrl, UCFatalEn); 515} 516 517static bool umc_v6_7_query_ras_poison_mode(struct amdgpu_device *adev) 518{ 519 uint32_t umc_reg_offset = 0; 520 521 /* Enabling fatal error in umc instance0 channel0 will be 522 * considered as fatal error mode 523 */ 524 umc_reg_offset = get_umc_v6_7_reg_offset(adev, 0, 0); 525 return !umc_v6_7_query_ras_poison_mode_per_channel(adev, umc_reg_offset); 526} 527 528const struct amdgpu_ras_block_hw_ops umc_v6_7_ras_hw_ops = { 529 .query_ras_error_count = umc_v6_7_query_ras_error_count, 530 .query_ras_error_address = umc_v6_7_query_ras_error_address, 531}; 532 533struct amdgpu_umc_ras umc_v6_7_ras = { 534 .ras_block = { 535 .hw_ops = &umc_v6_7_ras_hw_ops, 536 }, 537 .query_ras_poison_mode = umc_v6_7_query_ras_poison_mode, 538 .ecc_info_query_ras_error_count = umc_v6_7_ecc_info_query_ras_error_count, 539 .ecc_info_query_ras_error_address = umc_v6_7_ecc_info_query_ras_error_address, 540};