2 * Copyright 2020 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
24 #include "amdgpu_ras.h"
27 #include "rsmu/rsmu_0_0_2_offset.h"
28 #include "rsmu/rsmu_0_0_2_sh_mask.h"
29 #include "umc/umc_8_7_0_offset.h"
30 #include "umc/umc_8_7_0_sh_mask.h"
32 #define UMC_8_INST_DIST 0x40000
35 umc_v8_7_channel_idx_tbl[UMC_V8_7_UMC_INSTANCE_NUM][UMC_V8_7_CHANNEL_INSTANCE_NUM] = {
42 static inline uint32_t get_umc_8_reg_offset(struct amdgpu_device *adev,
46 return adev->umc.channel_offs*ch_inst + UMC_8_INST_DIST*umc_inst;
49 static void umc_v8_7_clear_error_count_per_channel(struct amdgpu_device *adev,
50 uint32_t umc_reg_offset)
52 uint32_t ecc_err_cnt_addr;
53 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
55 ecc_err_cnt_sel_addr =
56 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCntSel);
58 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCnt);
60 /* select the lower chip */
61 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr +
63 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel,
64 UMCCH0_0_GeccErrCntSel,
66 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4,
69 /* clear lower chip error count */
70 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4,
71 UMC_V8_7_CE_CNT_INIT);
73 /* select the higher chip */
74 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr +
76 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel,
77 UMCCH0_0_GeccErrCntSel,
79 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4,
82 /* clear higher chip error count */
83 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4,
84 UMC_V8_7_CE_CNT_INIT);
87 static void umc_v8_7_clear_error_count(struct amdgpu_device *adev)
89 uint32_t umc_inst = 0;
91 uint32_t umc_reg_offset = 0;
93 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
94 umc_reg_offset = get_umc_8_reg_offset(adev,
98 umc_v8_7_clear_error_count_per_channel(adev,
103 static void umc_v8_7_query_correctable_error_count(struct amdgpu_device *adev,
104 uint32_t umc_reg_offset,
105 unsigned long *error_count)
107 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
108 uint32_t ecc_err_cnt, ecc_err_cnt_addr;
109 uint64_t mc_umc_status;
110 uint32_t mc_umc_status_addr;
112 /* UMC 8_7_2 registers */
113 ecc_err_cnt_sel_addr =
114 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCntSel);
116 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCnt);
118 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
120 /* select the lower chip and check the error count */
121 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4);
122 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
124 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
126 ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4);
128 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_GeccErrCnt, GeccErrCnt) -
129 UMC_V8_7_CE_CNT_INIT);
131 /* select the higher chip and check the err counter */
132 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
134 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
136 ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4);
138 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_GeccErrCnt, GeccErrCnt) -
139 UMC_V8_7_CE_CNT_INIT);
141 /* check for SRAM correctable error
142 MCUMC_STATUS is a 64 bit register */
143 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
144 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 6 &&
145 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
146 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
150 static void umc_v8_7_querry_uncorrectable_error_count(struct amdgpu_device *adev,
151 uint32_t umc_reg_offset,
152 unsigned long *error_count)
154 uint64_t mc_umc_status;
155 uint32_t mc_umc_status_addr;
157 mc_umc_status_addr = SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
159 /* check the MCUMC_STATUS */
160 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
161 if ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
162 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
163 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
164 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
165 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
166 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
170 static void umc_v8_7_query_ras_error_count(struct amdgpu_device *adev,
171 void *ras_error_status)
173 struct ras_err_data* err_data = (struct ras_err_data*)ras_error_status;
175 uint32_t umc_inst = 0;
176 uint32_t ch_inst = 0;
177 uint32_t umc_reg_offset = 0;
179 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
180 umc_reg_offset = get_umc_8_reg_offset(adev,
184 umc_v8_7_query_correctable_error_count(adev,
186 &(err_data->ce_count));
187 umc_v8_7_querry_uncorrectable_error_count(adev,
189 &(err_data->ue_count));
192 umc_v8_7_clear_error_count(adev);
195 static void umc_v8_7_query_error_address(struct amdgpu_device *adev,
196 struct ras_err_data *err_data,
197 uint32_t umc_reg_offset,
201 uint32_t lsb, mc_umc_status_addr;
202 uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
203 struct eeprom_table_record *err_rec;
204 uint32_t channel_index = adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst];
207 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
209 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_ADDRT0);
211 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
213 if (mc_umc_status == 0)
216 if (!err_data->err_addr) {
217 /* clear umc status */
218 WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
222 err_rec = &err_data->err_addr[err_data->err_addr_cnt];
224 /* calculate error address if ue/ce error is detected */
225 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
226 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
227 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)) {
229 err_addr = RREG64_PCIE((mc_umc_addrt0 + umc_reg_offset) * 4);
230 /* the lowest lsb bits should be ignored */
231 lsb = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, LSB);
232 err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
233 err_addr &= ~((0x1ULL << lsb) - 1);
235 /* translate umc channel address to soc pa, 3 parts are included */
236 retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
237 ADDR_OF_256B_BLOCK(channel_index) |
238 OFFSET_IN_256B_BLOCK(err_addr);
240 /* we only save ue error information currently, ce is skipped */
241 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
243 err_rec->address = err_addr;
244 /* page frame address is saved */
245 err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
246 err_rec->ts = (uint64_t)ktime_get_real_seconds();
247 err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
249 err_rec->mem_channel = channel_index;
250 err_rec->mcumc_id = umc_inst;
252 err_data->err_addr_cnt++;
256 /* clear umc status */
257 WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
260 static void umc_v8_7_query_ras_error_address(struct amdgpu_device *adev,
261 void *ras_error_status)
263 struct ras_err_data* err_data = (struct ras_err_data*)ras_error_status;
265 uint32_t umc_inst = 0;
266 uint32_t ch_inst = 0;
267 uint32_t umc_reg_offset = 0;
269 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
270 umc_reg_offset = get_umc_8_reg_offset(adev,
274 umc_v8_7_query_error_address(adev,
282 static void umc_v8_7_err_cnt_init_per_channel(struct amdgpu_device *adev,
283 uint32_t umc_reg_offset)
285 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
286 uint32_t ecc_err_cnt_addr;
288 ecc_err_cnt_sel_addr =
289 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCntSel);
291 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCnt);
293 /* select the lower chip and check the error count */
294 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4);
295 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
297 /* set ce error interrupt type to APIC based interrupt */
298 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
300 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
301 /* set error count to initial value */
302 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V8_7_CE_CNT_INIT);
304 /* select the higher chip and check the err counter */
305 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
307 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
308 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V8_7_CE_CNT_INIT);
311 static void umc_v8_7_err_cnt_init(struct amdgpu_device *adev)
313 uint32_t umc_inst = 0;
314 uint32_t ch_inst = 0;
315 uint32_t umc_reg_offset = 0;
317 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
318 umc_reg_offset = get_umc_8_reg_offset(adev,
322 umc_v8_7_err_cnt_init_per_channel(adev, umc_reg_offset);
326 const struct amdgpu_umc_funcs umc_v8_7_funcs = {
327 .err_cnt_init = umc_v8_7_err_cnt_init,
328 .ras_late_init = amdgpu_umc_ras_late_init,
329 .query_ras_error_count = umc_v8_7_query_ras_error_count,
330 .query_ras_error_address = umc_v8_7_query_ras_error_address,