2 * Copyright 2023 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.
23 #include "umc_v12_0.h"
24 #include "amdgpu_ras.h"
25 #include "amdgpu_umc.h"
27 #include "umc/umc_12_0_0_offset.h"
28 #include "umc/umc_12_0_0_sh_mask.h"
29 #include "mp/mp_13_0_6_sh_mask.h"
32 umc_v12_0_channel_idx_tbl[]
33 [UMC_V12_0_UMC_INSTANCE_NUM]
34 [UMC_V12_0_CHANNEL_INSTANCE_NUM] = {
35 {{3, 7, 11, 15, 2, 6, 10, 14}, {1, 5, 9, 13, 0, 4, 8, 12},
36 {19, 23, 27, 31, 18, 22, 26, 30}, {17, 21, 25, 29, 16, 20, 24, 28}},
37 {{47, 43, 39, 35, 46, 42, 38, 34}, {45, 41, 37, 33, 44, 40, 36, 32},
38 {63, 59, 55, 51, 62, 58, 54, 50}, {61, 57, 53, 49, 60, 56, 52, 48}},
39 {{79, 75, 71, 67, 78, 74, 70, 66}, {77, 73, 69, 65, 76, 72, 68, 64},
40 {95, 91, 87, 83, 94, 90, 86, 82}, {93, 89, 85, 81, 92, 88, 84, 80}},
41 {{99, 103, 107, 111, 98, 102, 106, 110}, {97, 101, 105, 109, 96, 100, 104, 108},
42 {115, 119, 123, 127, 114, 118, 122, 126}, {113, 117, 121, 125, 112, 116, 120, 124}}
45 /* mapping of MCA error address to normalized address */
46 static const uint32_t umc_v12_0_ma2na_mapping[] = {
47 0, 5, 6, 8, 9, 14, 12, 13,
48 10, 11, 15, 16, 17, 18, 19, 20,
49 21, 22, 23, 24, 25, 26, 27, 28,
53 static inline uint64_t get_umc_v12_0_reg_offset(struct amdgpu_device *adev,
58 uint32_t index = umc_inst * adev->umc.channel_inst_num + ch_inst;
59 uint64_t cross_node_offset = (node_inst == 0) ? 0 : UMC_V12_0_CROSS_NODE_OFFSET;
64 return adev->umc.channel_offs * ch_inst + UMC_V12_0_INST_DIST * umc_inst +
65 UMC_V12_0_NODE_DIST * node_inst + cross_node_offset;
68 static int umc_v12_0_reset_error_count_per_channel(struct amdgpu_device *adev,
69 uint32_t node_inst, uint32_t umc_inst,
70 uint32_t ch_inst, void *data)
72 uint64_t odecc_err_cnt_addr;
73 uint64_t umc_reg_offset =
74 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
77 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_OdEccErrCnt);
79 /* clear error count */
80 WREG32_PCIE_EXT((odecc_err_cnt_addr + umc_reg_offset) * 4,
81 UMC_V12_0_CE_CNT_INIT);
86 static void umc_v12_0_reset_error_count(struct amdgpu_device *adev)
88 amdgpu_umc_loop_channels(adev,
89 umc_v12_0_reset_error_count_per_channel, NULL);
92 bool umc_v12_0_is_uncorrectable_error(struct amdgpu_device *adev, uint64_t mc_umc_status)
94 if (amdgpu_ras_is_poison_mode_supported(adev) &&
95 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
96 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1))
99 return ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
100 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
101 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
102 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1));
105 bool umc_v12_0_is_correctable_error(struct amdgpu_device *adev, uint64_t mc_umc_status)
107 if (amdgpu_ras_is_poison_mode_supported(adev) &&
108 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
109 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1))
112 return (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
113 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1 ||
114 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 &&
115 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 0) ||
116 /* Identify data parity error in replay mode */
117 ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 0x5 ||
118 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 0xb) &&
119 !(umc_v12_0_is_uncorrectable_error(adev, mc_umc_status)))));
122 static void umc_v12_0_query_correctable_error_count(struct amdgpu_device *adev,
123 uint64_t umc_reg_offset,
124 unsigned long *error_count)
126 uint64_t mc_umc_status;
127 uint64_t mc_umc_status_addr;
130 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
132 /* Rely on MCUMC_STATUS for correctable error counter
133 * MCUMC_STATUS is a 64 bit register
136 RREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4);
138 if (umc_v12_0_is_correctable_error(adev, mc_umc_status))
142 static void umc_v12_0_query_uncorrectable_error_count(struct amdgpu_device *adev,
143 uint64_t umc_reg_offset,
144 unsigned long *error_count)
146 uint64_t mc_umc_status;
147 uint64_t mc_umc_status_addr;
150 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
152 /* Check the MCUMC_STATUS. */
154 RREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4);
156 if (umc_v12_0_is_uncorrectable_error(adev, mc_umc_status))
160 static int umc_v12_0_query_error_count(struct amdgpu_device *adev,
161 uint32_t node_inst, uint32_t umc_inst,
162 uint32_t ch_inst, void *data)
164 struct ras_err_data *err_data = (struct ras_err_data *)data;
165 unsigned long ue_count = 0, ce_count = 0;
167 /* NOTE: node_inst is converted by adev->umc.active_mask and the range is [0-3],
168 * which can be used as die ID directly */
169 struct amdgpu_smuio_mcm_config_info mcm_info = {
170 .socket_id = adev->smuio.funcs->get_socket_id(adev),
174 uint64_t umc_reg_offset =
175 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
177 umc_v12_0_query_correctable_error_count(adev, umc_reg_offset, &ce_count);
178 umc_v12_0_query_uncorrectable_error_count(adev, umc_reg_offset, &ue_count);
180 amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, NULL, ue_count);
181 amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, NULL, ce_count);
186 static void umc_v12_0_query_ras_error_count(struct amdgpu_device *adev,
187 void *ras_error_status)
189 amdgpu_umc_loop_channels(adev,
190 umc_v12_0_query_error_count, ras_error_status);
192 umc_v12_0_reset_error_count(adev);
195 static bool umc_v12_0_bit_wise_xor(uint32_t val)
200 for (i = 0; i < 32; i++)
201 result = result ^ ((val >> i) & 0x1);
206 static void umc_v12_0_convert_error_address(struct amdgpu_device *adev,
207 struct ras_err_data *err_data, uint64_t err_addr,
208 uint32_t ch_inst, uint32_t umc_inst,
211 uint32_t channel_index, i;
212 uint64_t soc_pa, na, retired_page, column;
213 uint32_t bank_hash0, bank_hash1, bank_hash2, bank_hash3, col, row, row_xor;
214 uint32_t bank0, bank1, bank2, bank3, bank;
216 bank_hash0 = (err_addr >> UMC_V12_0_MCA_B0_BIT) & 0x1ULL;
217 bank_hash1 = (err_addr >> UMC_V12_0_MCA_B1_BIT) & 0x1ULL;
218 bank_hash2 = (err_addr >> UMC_V12_0_MCA_B2_BIT) & 0x1ULL;
219 bank_hash3 = (err_addr >> UMC_V12_0_MCA_B3_BIT) & 0x1ULL;
220 col = (err_addr >> 1) & 0x1fULL;
221 row = (err_addr >> 10) & 0x3fffULL;
223 /* apply bank hash algorithm */
225 bank_hash0 ^ (UMC_V12_0_XOR_EN0 &
226 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR0) ^
227 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR0))));
229 bank_hash1 ^ (UMC_V12_0_XOR_EN1 &
230 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR1) ^
231 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR1))));
233 bank_hash2 ^ (UMC_V12_0_XOR_EN2 &
234 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR2) ^
235 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR2))));
237 bank_hash3 ^ (UMC_V12_0_XOR_EN3 &
238 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR3) ^
239 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR3))));
241 bank = bank0 | (bank1 << 1) | (bank2 << 2) | (bank3 << 3);
242 err_addr &= ~0x3c0ULL;
243 err_addr |= (bank << UMC_V12_0_MCA_B0_BIT);
246 /* convert mca error address to normalized address */
247 for (i = 1; i < ARRAY_SIZE(umc_v12_0_ma2na_mapping); i++)
248 na |= ((err_addr >> i) & 0x1ULL) << umc_v12_0_ma2na_mapping[i];
251 adev->umc.channel_idx_tbl[node_inst * adev->umc.umc_inst_num *
252 adev->umc.channel_inst_num +
253 umc_inst * adev->umc.channel_inst_num +
255 /* translate umc channel address to soc pa, 3 parts are included */
256 soc_pa = ADDR_OF_32KB_BLOCK(na) |
257 ADDR_OF_256B_BLOCK(channel_index) |
258 OFFSET_IN_256B_BLOCK(na);
260 /* the umc channel bits are not original values, they are hashed */
261 UMC_V12_0_SET_CHANNEL_HASH(channel_index, soc_pa);
263 /* clear [C3 C2] in soc physical address */
264 soc_pa &= ~(0x3ULL << UMC_V12_0_PA_C2_BIT);
265 /* clear [C4] in soc physical address */
266 soc_pa &= ~(0x1ULL << UMC_V12_0_PA_C4_BIT);
268 row_xor = row ^ (0x1ULL << 13);
269 /* loop for all possibilities of [C4 C3 C2] */
270 for (column = 0; column < UMC_V12_0_NA_MAP_PA_NUM; column++) {
271 retired_page = soc_pa | ((column & 0x3) << UMC_V12_0_PA_C2_BIT);
272 retired_page |= (((column & 0x4) >> 2) << UMC_V12_0_PA_C4_BIT);
273 /* include column bit 0 and 1 */
275 col |= (column << 2);
277 "Error Address(PA):0x%-10llx Row:0x%-4x Col:0x%-2x Bank:0x%x Channel:0x%x\n",
278 retired_page, row, col, bank, channel_index);
279 amdgpu_umc_fill_error_record(err_data, err_addr,
280 retired_page, channel_index, umc_inst);
283 retired_page ^= (0x1ULL << UMC_V12_0_PA_R13_BIT);
285 "Error Address(PA):0x%-10llx Row:0x%-4x Col:0x%-2x Bank:0x%x Channel:0x%x\n",
286 retired_page, row_xor, col, bank, channel_index);
287 amdgpu_umc_fill_error_record(err_data, err_addr,
288 retired_page, channel_index, umc_inst);
292 static int umc_v12_0_query_error_address(struct amdgpu_device *adev,
293 uint32_t node_inst, uint32_t umc_inst,
294 uint32_t ch_inst, void *data)
296 uint64_t mc_umc_status_addr;
297 uint64_t mc_umc_status, err_addr;
298 uint64_t mc_umc_addrt0;
299 struct ras_err_data *err_data = (struct ras_err_data *)data;
300 uint64_t umc_reg_offset =
301 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
304 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
306 mc_umc_status = RREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4);
308 if (mc_umc_status == 0)
311 if (!err_data->err_addr) {
312 /* clear umc status */
313 WREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
318 /* calculate error address if ue error is detected */
319 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
320 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, AddrV) == 1 &&
321 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1) {
324 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_ADDRT0);
326 err_addr = RREG64_PCIE_EXT((mc_umc_addrt0 + umc_reg_offset) * 4);
328 err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
330 umc_v12_0_convert_error_address(adev, err_data, err_addr,
331 ch_inst, umc_inst, node_inst);
334 /* clear umc status */
335 WREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
340 static void umc_v12_0_query_ras_error_address(struct amdgpu_device *adev,
341 void *ras_error_status)
343 amdgpu_umc_loop_channels(adev,
344 umc_v12_0_query_error_address, ras_error_status);
347 static int umc_v12_0_err_cnt_init_per_channel(struct amdgpu_device *adev,
348 uint32_t node_inst, uint32_t umc_inst,
349 uint32_t ch_inst, void *data)
351 uint32_t odecc_cnt_sel;
352 uint64_t odecc_cnt_sel_addr, odecc_err_cnt_addr;
353 uint64_t umc_reg_offset =
354 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
357 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_OdEccCntSel);
359 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_OdEccErrCnt);
361 odecc_cnt_sel = RREG32_PCIE_EXT((odecc_cnt_sel_addr + umc_reg_offset) * 4);
363 /* set ce error interrupt type to APIC based interrupt */
364 odecc_cnt_sel = REG_SET_FIELD(odecc_cnt_sel, UMCCH0_OdEccCntSel,
366 WREG32_PCIE_EXT((odecc_cnt_sel_addr + umc_reg_offset) * 4, odecc_cnt_sel);
368 /* set error count to initial value */
369 WREG32_PCIE_EXT((odecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V12_0_CE_CNT_INIT);
374 static void umc_v12_0_ecc_info_query_ras_error_count(struct amdgpu_device *adev,
375 void *ras_error_status)
377 amdgpu_mca_smu_log_ras_error(adev,
378 AMDGPU_RAS_BLOCK__UMC, AMDGPU_MCA_ERROR_TYPE_CE, ras_error_status);
379 amdgpu_mca_smu_log_ras_error(adev,
380 AMDGPU_RAS_BLOCK__UMC, AMDGPU_MCA_ERROR_TYPE_UE, ras_error_status);
383 static void umc_v12_0_ecc_info_query_ras_error_address(struct amdgpu_device *adev,
384 void *ras_error_status)
386 struct ras_err_node *err_node;
387 uint64_t mc_umc_status;
388 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
390 for_each_ras_error(err_node, err_data) {
391 mc_umc_status = err_node->err_info.err_addr.err_status;
395 if (umc_v12_0_is_uncorrectable_error(adev, mc_umc_status)) {
396 uint64_t mca_addr, err_addr, mca_ipid;
397 uint32_t InstanceIdLo;
398 struct amdgpu_smuio_mcm_config_info *mcm_info;
400 mcm_info = &err_node->err_info.mcm_info;
401 mca_addr = err_node->err_info.err_addr.err_addr;
402 mca_ipid = err_node->err_info.err_addr.err_ipid;
404 err_addr = REG_GET_FIELD(mca_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
405 InstanceIdLo = REG_GET_FIELD(mca_ipid, MCMP1_IPIDT0, InstanceIdLo);
407 dev_info(adev->dev, "UMC:IPID:0x%llx, aid:%d, inst:%d, ch:%d, err_addr:0x%llx\n",
410 MCA_IPID_LO_2_UMC_INST(InstanceIdLo),
411 MCA_IPID_LO_2_UMC_CH(InstanceIdLo),
414 umc_v12_0_convert_error_address(adev,
416 MCA_IPID_LO_2_UMC_CH(InstanceIdLo),
417 MCA_IPID_LO_2_UMC_INST(InstanceIdLo),
420 /* Clear umc error address content */
421 memset(&err_node->err_info.err_addr,
422 0, sizeof(err_node->err_info.err_addr));
427 static void umc_v12_0_err_cnt_init(struct amdgpu_device *adev)
429 amdgpu_umc_loop_channels(adev,
430 umc_v12_0_err_cnt_init_per_channel, NULL);
433 static bool umc_v12_0_query_ras_poison_mode(struct amdgpu_device *adev)
436 * Force return true, because regUMCCH0_EccCtrl
437 * is not accessible from host side
442 const struct amdgpu_ras_block_hw_ops umc_v12_0_ras_hw_ops = {
443 .query_ras_error_count = umc_v12_0_query_ras_error_count,
444 .query_ras_error_address = umc_v12_0_query_ras_error_address,
447 struct amdgpu_umc_ras umc_v12_0_ras = {
449 .hw_ops = &umc_v12_0_ras_hw_ops,
451 .err_cnt_init = umc_v12_0_err_cnt_init,
452 .query_ras_poison_mode = umc_v12_0_query_ras_poison_mode,
453 .ecc_info_query_ras_error_count = umc_v12_0_ecc_info_query_ras_error_count,
454 .ecc_info_query_ras_error_address = umc_v12_0_ecc_info_query_ras_error_address,