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"
31 umc_v12_0_channel_idx_tbl[]
32 [UMC_V12_0_UMC_INSTANCE_NUM]
33 [UMC_V12_0_CHANNEL_INSTANCE_NUM] = {
34 {{3, 7, 11, 15, 2, 6, 10, 14}, {1, 5, 9, 13, 0, 4, 8, 12},
35 {19, 23, 27, 31, 18, 22, 26, 30}, {17, 21, 25, 29, 16, 20, 24, 28}},
36 {{47, 43, 39, 35, 46, 42, 38, 34}, {45, 41, 37, 33, 44, 40, 36, 32},
37 {63, 59, 55, 51, 62, 58, 54, 50}, {61, 57, 53, 49, 60, 56, 52, 48}},
38 {{79, 75, 71, 67, 78, 74, 70, 66}, {77, 73, 69, 65, 76, 72, 68, 64},
39 {95, 91, 87, 83, 94, 90, 86, 82}, {93, 89, 85, 81, 92, 88, 84, 80}},
40 {{99, 103, 107, 111, 98, 102, 106, 110}, {97, 101, 105, 109, 96, 100, 104, 108},
41 {115, 119, 123, 127, 114, 118, 122, 126}, {113, 117, 121, 125, 112, 116, 120, 124}}
44 /* mapping of MCA error address to normalized address */
45 static const uint32_t umc_v12_0_ma2na_mapping[] = {
46 0, 5, 6, 8, 9, 14, 12, 13,
47 10, 11, 15, 16, 17, 18, 19, 20,
48 21, 22, 23, 24, 25, 26, 27, 28,
52 static inline uint64_t get_umc_v12_0_reg_offset(struct amdgpu_device *adev,
57 uint32_t index = umc_inst * adev->umc.channel_inst_num + ch_inst;
58 uint64_t cross_node_offset = (node_inst == 0) ? 0 : UMC_V12_0_CROSS_NODE_OFFSET;
63 return adev->umc.channel_offs * ch_inst + UMC_V12_0_INST_DIST * umc_inst +
64 UMC_V12_0_NODE_DIST * node_inst + cross_node_offset;
67 static int umc_v12_0_reset_error_count_per_channel(struct amdgpu_device *adev,
68 uint32_t node_inst, uint32_t umc_inst,
69 uint32_t ch_inst, void *data)
71 uint64_t odecc_err_cnt_addr;
72 uint64_t umc_reg_offset =
73 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
76 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_OdEccErrCnt);
78 /* clear error count */
79 WREG32_PCIE_EXT((odecc_err_cnt_addr + umc_reg_offset) * 4,
80 UMC_V12_0_CE_CNT_INIT);
85 static void umc_v12_0_reset_error_count(struct amdgpu_device *adev)
87 amdgpu_umc_loop_channels(adev,
88 umc_v12_0_reset_error_count_per_channel, NULL);
91 static bool umc_v12_0_is_uncorrectable_error(uint64_t mc_umc_status)
93 return ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
94 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
95 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
96 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
97 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1));
100 static bool umc_v12_0_is_correctable_error(uint64_t mc_umc_status)
102 return (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
103 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1 ||
104 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 &&
105 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 0) ||
106 /* Identify data parity error in replay mode */
107 ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 0x5 ||
108 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 0xb) &&
109 !(umc_v12_0_is_uncorrectable_error(mc_umc_status)))));
112 static void umc_v12_0_query_correctable_error_count(struct amdgpu_device *adev,
113 uint64_t umc_reg_offset,
114 unsigned long *error_count)
116 uint64_t mc_umc_status;
117 uint64_t mc_umc_status_addr;
120 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
122 /* Rely on MCUMC_STATUS for correctable error counter
123 * MCUMC_STATUS is a 64 bit register
126 RREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4);
128 if (umc_v12_0_is_correctable_error(mc_umc_status))
132 static void umc_v12_0_query_uncorrectable_error_count(struct amdgpu_device *adev,
133 uint64_t umc_reg_offset,
134 unsigned long *error_count)
136 uint64_t mc_umc_status;
137 uint64_t mc_umc_status_addr;
140 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
142 /* Check the MCUMC_STATUS. */
144 RREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4);
146 if (umc_v12_0_is_uncorrectable_error(mc_umc_status))
150 static int umc_v12_0_query_error_count(struct amdgpu_device *adev,
151 uint32_t node_inst, uint32_t umc_inst,
152 uint32_t ch_inst, void *data)
154 struct ras_err_data *err_data = (struct ras_err_data *)data;
155 unsigned long ue_count = 0, ce_count = 0;
157 /* NOTE: node_inst is converted by adev->umc.active_mask and the range is [0-3],
158 * which can be used as die ID directly */
159 struct amdgpu_smuio_mcm_config_info mcm_info = {
160 .socket_id = adev->smuio.funcs->get_socket_id(adev),
164 uint64_t umc_reg_offset =
165 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
167 umc_v12_0_query_correctable_error_count(adev, umc_reg_offset, &ce_count);
168 umc_v12_0_query_uncorrectable_error_count(adev, umc_reg_offset, &ue_count);
170 amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count);
171 amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, ce_count);
176 static void umc_v12_0_query_ras_error_count(struct amdgpu_device *adev,
177 void *ras_error_status)
179 amdgpu_umc_loop_channels(adev,
180 umc_v12_0_query_error_count, ras_error_status);
182 umc_v12_0_reset_error_count(adev);
185 static bool umc_v12_0_bit_wise_xor(uint32_t val)
190 for (i = 0; i < 32; i++)
191 result = result ^ ((val >> i) & 0x1);
196 static void umc_v12_0_convert_error_address(struct amdgpu_device *adev,
197 struct ras_err_data *err_data, uint64_t err_addr,
198 uint32_t ch_inst, uint32_t umc_inst,
201 uint32_t channel_index, i;
202 uint64_t soc_pa, na, retired_page, column;
203 uint32_t bank_hash0, bank_hash1, bank_hash2, bank_hash3, col, row, row_xor;
204 uint32_t bank0, bank1, bank2, bank3, bank;
206 bank_hash0 = (err_addr >> UMC_V12_0_MCA_B0_BIT) & 0x1ULL;
207 bank_hash1 = (err_addr >> UMC_V12_0_MCA_B1_BIT) & 0x1ULL;
208 bank_hash2 = (err_addr >> UMC_V12_0_MCA_B2_BIT) & 0x1ULL;
209 bank_hash3 = (err_addr >> UMC_V12_0_MCA_B3_BIT) & 0x1ULL;
210 col = (err_addr >> 1) & 0x1fULL;
211 row = (err_addr >> 10) & 0x3fffULL;
213 /* apply bank hash algorithm */
215 bank_hash0 ^ (UMC_V12_0_XOR_EN0 &
216 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR0) ^
217 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR0))));
219 bank_hash1 ^ (UMC_V12_0_XOR_EN1 &
220 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR1) ^
221 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR1))));
223 bank_hash2 ^ (UMC_V12_0_XOR_EN2 &
224 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR2) ^
225 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR2))));
227 bank_hash3 ^ (UMC_V12_0_XOR_EN3 &
228 (umc_v12_0_bit_wise_xor(col & UMC_V12_0_COL_XOR3) ^
229 (umc_v12_0_bit_wise_xor(row & UMC_V12_0_ROW_XOR3))));
231 bank = bank0 | (bank1 << 1) | (bank2 << 2) | (bank3 << 3);
232 err_addr &= ~0x3c0ULL;
233 err_addr |= (bank << UMC_V12_0_MCA_B0_BIT);
236 /* convert mca error address to normalized address */
237 for (i = 1; i < ARRAY_SIZE(umc_v12_0_ma2na_mapping); i++)
238 na |= ((err_addr >> i) & 0x1ULL) << umc_v12_0_ma2na_mapping[i];
241 adev->umc.channel_idx_tbl[node_inst * adev->umc.umc_inst_num *
242 adev->umc.channel_inst_num +
243 umc_inst * adev->umc.channel_inst_num +
245 /* translate umc channel address to soc pa, 3 parts are included */
246 soc_pa = ADDR_OF_32KB_BLOCK(na) |
247 ADDR_OF_256B_BLOCK(channel_index) |
248 OFFSET_IN_256B_BLOCK(na);
250 /* the umc channel bits are not original values, they are hashed */
251 UMC_V12_0_SET_CHANNEL_HASH(channel_index, soc_pa);
253 /* clear [C3 C2] in soc physical address */
254 soc_pa &= ~(0x3ULL << UMC_V12_0_PA_C2_BIT);
255 /* clear [C4] in soc physical address */
256 soc_pa &= ~(0x1ULL << UMC_V12_0_PA_C4_BIT);
258 row_xor = row ^ (0x1ULL << 13);
259 /* loop for all possibilities of [C4 C3 C2] */
260 for (column = 0; column < UMC_V12_0_NA_MAP_PA_NUM; column++) {
261 retired_page = soc_pa | ((column & 0x3) << UMC_V12_0_PA_C2_BIT);
262 retired_page |= (((column & 0x4) >> 2) << UMC_V12_0_PA_C4_BIT);
263 /* include column bit 0 and 1 */
265 col |= (column << 2);
267 "Error Address(PA):0x%-10llx Row:0x%-4x Col:0x%-2x Bank:0x%x Channel:0x%x\n",
268 retired_page, row, col, bank, channel_index);
269 amdgpu_umc_fill_error_record(err_data, err_addr,
270 retired_page, channel_index, umc_inst);
273 retired_page ^= (0x1ULL << UMC_V12_0_PA_R13_BIT);
275 "Error Address(PA):0x%-10llx Row:0x%-4x Col:0x%-2x Bank:0x%x Channel:0x%x\n",
276 retired_page, row_xor, col, bank, channel_index);
277 amdgpu_umc_fill_error_record(err_data, err_addr,
278 retired_page, channel_index, umc_inst);
282 static int umc_v12_0_query_error_address(struct amdgpu_device *adev,
283 uint32_t node_inst, uint32_t umc_inst,
284 uint32_t ch_inst, void *data)
286 uint64_t mc_umc_status_addr;
287 uint64_t mc_umc_status, err_addr;
288 uint64_t mc_umc_addrt0;
289 struct ras_err_data *err_data = (struct ras_err_data *)data;
290 uint64_t umc_reg_offset =
291 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
294 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
296 mc_umc_status = RREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4);
298 if (mc_umc_status == 0)
301 if (!err_data->err_addr) {
302 /* clear umc status */
303 WREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
308 /* calculate error address if ue error is detected */
309 if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
310 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, AddrV) == 1 &&
311 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1) {
314 SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_ADDRT0);
316 err_addr = RREG64_PCIE_EXT((mc_umc_addrt0 + umc_reg_offset) * 4);
318 err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
320 umc_v12_0_convert_error_address(adev, err_data, err_addr,
321 ch_inst, umc_inst, node_inst);
324 /* clear umc status */
325 WREG64_PCIE_EXT((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
330 static void umc_v12_0_query_ras_error_address(struct amdgpu_device *adev,
331 void *ras_error_status)
333 amdgpu_umc_loop_channels(adev,
334 umc_v12_0_query_error_address, ras_error_status);
337 static int umc_v12_0_err_cnt_init_per_channel(struct amdgpu_device *adev,
338 uint32_t node_inst, uint32_t umc_inst,
339 uint32_t ch_inst, void *data)
341 uint32_t odecc_cnt_sel;
342 uint64_t odecc_cnt_sel_addr, odecc_err_cnt_addr;
343 uint64_t umc_reg_offset =
344 get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
347 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_OdEccCntSel);
349 SOC15_REG_OFFSET(UMC, 0, regUMCCH0_OdEccErrCnt);
351 odecc_cnt_sel = RREG32_PCIE_EXT((odecc_cnt_sel_addr + umc_reg_offset) * 4);
353 /* set ce error interrupt type to APIC based interrupt */
354 odecc_cnt_sel = REG_SET_FIELD(odecc_cnt_sel, UMCCH0_OdEccCntSel,
356 WREG32_PCIE_EXT((odecc_cnt_sel_addr + umc_reg_offset) * 4, odecc_cnt_sel);
358 /* set error count to initial value */
359 WREG32_PCIE_EXT((odecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V12_0_CE_CNT_INIT);
364 static void umc_v12_0_err_cnt_init(struct amdgpu_device *adev)
366 amdgpu_umc_loop_channels(adev,
367 umc_v12_0_err_cnt_init_per_channel, NULL);
370 static bool umc_v12_0_query_ras_poison_mode(struct amdgpu_device *adev)
373 * Force return true, because regUMCCH0_EccCtrl
374 * is not accessible from host side
379 const struct amdgpu_ras_block_hw_ops umc_v12_0_ras_hw_ops = {
380 .query_ras_error_count = umc_v12_0_query_ras_error_count,
381 .query_ras_error_address = umc_v12_0_query_ras_error_address,
384 struct amdgpu_umc_ras umc_v12_0_ras = {
386 .hw_ops = &umc_v12_0_ras_hw_ops,
388 .err_cnt_init = umc_v12_0_err_cnt_init,
389 .query_ras_poison_mode = umc_v12_0_query_ras_poison_mode,