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"
25 #include "amdgpu_umc.h"
28 #include "rsmu/rsmu_0_0_2_offset.h"
29 #include "rsmu/rsmu_0_0_2_sh_mask.h"
30 #include "umc/umc_8_7_0_offset.h"
31 #include "umc/umc_8_7_0_sh_mask.h"
33 #define UMC_8_INST_DIST 0x40000
36 umc_v8_7_channel_idx_tbl[UMC_V8_7_UMC_INSTANCE_NUM][UMC_V8_7_CHANNEL_INSTANCE_NUM] = {
43 static inline uint32_t get_umc_8_reg_offset(struct amdgpu_device *adev,
47 return adev->umc.channel_offs*ch_inst + UMC_8_INST_DIST*umc_inst;
50 static void umc_v8_7_clear_error_count_per_channel(struct amdgpu_device *adev,
51 uint32_t umc_reg_offset)
53 uint32_t ecc_err_cnt_addr;
54 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
56 ecc_err_cnt_sel_addr =
57 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCntSel);
59 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCnt);
61 /* select the lower chip */
62 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr +
64 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel,
65 UMCCH0_0_GeccErrCntSel,
67 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4,
70 /* clear lower chip error count */
71 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4,
72 UMC_V8_7_CE_CNT_INIT);
74 /* select the higher chip */
75 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr +
77 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel,
78 UMCCH0_0_GeccErrCntSel,
80 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4,
83 /* clear higher chip error count */
84 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4,
85 UMC_V8_7_CE_CNT_INIT);
88 static void umc_v8_7_clear_error_count(struct amdgpu_device *adev)
90 uint32_t umc_inst = 0;
92 uint32_t umc_reg_offset = 0;
94 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
95 umc_reg_offset = get_umc_8_reg_offset(adev,
99 umc_v8_7_clear_error_count_per_channel(adev,
104 static void umc_v8_7_query_correctable_error_count(struct amdgpu_device *adev,
105 uint32_t umc_reg_offset,
106 unsigned long *error_count)
108 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
109 uint32_t ecc_err_cnt, ecc_err_cnt_addr;
110 uint64_t mc_umc_status;
111 uint32_t mc_umc_status_addr;
113 /* UMC 8_7_2 registers */
114 ecc_err_cnt_sel_addr =
115 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCntSel);
117 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCnt);
119 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
121 /* select the lower chip and check the error count */
122 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4);
123 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
125 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
127 ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4);
129 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_GeccErrCnt, GeccErrCnt) -
130 UMC_V8_7_CE_CNT_INIT);
132 /* select the higher chip and check the err counter */
133 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
135 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
137 ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4);
139 (REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_GeccErrCnt, GeccErrCnt) -
140 UMC_V8_7_CE_CNT_INIT);
142 /* check for SRAM correctable error
143 MCUMC_STATUS is a 64 bit register */
144 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
145 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 6 &&
146 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
147 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
151 static void umc_v8_7_querry_uncorrectable_error_count(struct amdgpu_device *adev,
152 uint32_t umc_reg_offset,
153 unsigned long *error_count)
155 uint64_t mc_umc_status;
156 uint32_t mc_umc_status_addr;
158 mc_umc_status_addr = SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
160 /* check the MCUMC_STATUS */
161 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
162 if ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
163 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
164 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
165 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
166 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
167 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
171 static void umc_v8_7_query_ras_error_count(struct amdgpu_device *adev,
172 void *ras_error_status)
174 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
176 uint32_t umc_inst = 0;
177 uint32_t ch_inst = 0;
178 uint32_t umc_reg_offset = 0;
180 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
181 umc_reg_offset = get_umc_8_reg_offset(adev,
185 umc_v8_7_query_correctable_error_count(adev,
187 &(err_data->ce_count));
188 umc_v8_7_querry_uncorrectable_error_count(adev,
190 &(err_data->ue_count));
193 umc_v8_7_clear_error_count(adev);
196 static void umc_v8_7_query_error_address(struct amdgpu_device *adev,
197 struct ras_err_data *err_data,
198 uint32_t umc_reg_offset,
202 uint32_t lsb, mc_umc_status_addr;
203 uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
204 struct eeprom_table_record *err_rec;
205 uint32_t channel_index = adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst];
208 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
210 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_ADDRT0);
212 mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
214 if (mc_umc_status == 0)
217 if (!err_data->err_addr) {
218 /* clear umc status */
219 WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
223 err_rec = &err_data->err_addr[err_data->err_addr_cnt];
225 /* calculate error address if ue/ce error is detected */
226 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
227 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
228 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)) {
230 err_addr = RREG64_PCIE((mc_umc_addrt0 + umc_reg_offset) * 4);
231 /* the lowest lsb bits should be ignored */
232 lsb = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, LSB);
233 err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
234 err_addr &= ~((0x1ULL << lsb) - 1);
236 /* translate umc channel address to soc pa, 3 parts are included */
237 retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
238 ADDR_OF_256B_BLOCK(channel_index) |
239 OFFSET_IN_256B_BLOCK(err_addr);
241 /* we only save ue error information currently, ce is skipped */
242 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
244 err_rec->address = err_addr;
245 /* page frame address is saved */
246 err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
247 err_rec->ts = (uint64_t)ktime_get_real_seconds();
248 err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
250 err_rec->mem_channel = channel_index;
251 err_rec->mcumc_id = umc_inst;
253 err_data->err_addr_cnt++;
257 /* clear umc status */
258 WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
261 static void umc_v8_7_query_ras_error_address(struct amdgpu_device *adev,
262 void *ras_error_status)
264 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
266 uint32_t umc_inst = 0;
267 uint32_t ch_inst = 0;
268 uint32_t umc_reg_offset = 0;
270 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
271 umc_reg_offset = get_umc_8_reg_offset(adev,
275 umc_v8_7_query_error_address(adev,
283 static void umc_v8_7_err_cnt_init_per_channel(struct amdgpu_device *adev,
284 uint32_t umc_reg_offset)
286 uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
287 uint32_t ecc_err_cnt_addr;
289 ecc_err_cnt_sel_addr =
290 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCntSel);
292 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_GeccErrCnt);
294 /* select the lower chip and check the error count */
295 ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4);
296 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
298 /* set ce error interrupt type to APIC based interrupt */
299 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
301 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
302 /* set error count to initial value */
303 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V8_7_CE_CNT_INIT);
305 /* select the higher chip and check the err counter */
306 ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
308 WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
309 WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V8_7_CE_CNT_INIT);
312 static void umc_v8_7_err_cnt_init(struct amdgpu_device *adev)
314 uint32_t umc_inst = 0;
315 uint32_t ch_inst = 0;
316 uint32_t umc_reg_offset = 0;
318 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
319 umc_reg_offset = get_umc_8_reg_offset(adev,
323 umc_v8_7_err_cnt_init_per_channel(adev, umc_reg_offset);
327 const struct amdgpu_umc_ras_funcs umc_v8_7_ras_funcs = {
328 .err_cnt_init = umc_v8_7_err_cnt_init,
329 .ras_late_init = amdgpu_umc_ras_late_init,
330 .ras_fini = amdgpu_umc_ras_fini,
331 .query_ras_error_count = umc_v8_7_query_ras_error_count,
332 .query_ras_error_address = umc_v8_7_query_ras_error_address,