Loading...
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 "amdgpu_ras.h"
24#include "amdgpu.h"
25#include "amdgpu_mca.h"
26
27#include "umc/umc_6_7_0_offset.h"
28#include "umc/umc_6_7_0_sh_mask.h"
29
30void amdgpu_mca_query_correctable_error_count(struct amdgpu_device *adev,
31 uint64_t mc_status_addr,
32 unsigned long *error_count)
33{
34 uint64_t mc_status = RREG64_PCIE(mc_status_addr);
35
36 if (REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
37 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
38 *error_count += 1;
39}
40
41void amdgpu_mca_query_uncorrectable_error_count(struct amdgpu_device *adev,
42 uint64_t mc_status_addr,
43 unsigned long *error_count)
44{
45 uint64_t mc_status = RREG64_PCIE(mc_status_addr);
46
47 if ((REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
48 (REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
49 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
50 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
51 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
52 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
53 *error_count += 1;
54}
55
56void amdgpu_mca_reset_error_count(struct amdgpu_device *adev,
57 uint64_t mc_status_addr)
58{
59 WREG64_PCIE(mc_status_addr, 0x0ULL);
60}
61
62void amdgpu_mca_query_ras_error_count(struct amdgpu_device *adev,
63 uint64_t mc_status_addr,
64 void *ras_error_status)
65{
66 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
67
68 amdgpu_mca_query_correctable_error_count(adev, mc_status_addr, &(err_data->ce_count));
69 amdgpu_mca_query_uncorrectable_error_count(adev, mc_status_addr, &(err_data->ue_count));
70
71 amdgpu_mca_reset_error_count(adev, mc_status_addr);
72}
73
74int amdgpu_mca_mp0_ras_sw_init(struct amdgpu_device *adev)
75{
76 int err;
77 struct amdgpu_mca_ras_block *ras;
78
79 if (!adev->mca.mp0.ras)
80 return 0;
81
82 ras = adev->mca.mp0.ras;
83
84 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
85 if (err) {
86 dev_err(adev->dev, "Failed to register mca.mp0 ras block!\n");
87 return err;
88 }
89
90 strcpy(ras->ras_block.ras_comm.name, "mca.mp0");
91 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
92 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
93 adev->mca.mp0.ras_if = &ras->ras_block.ras_comm;
94
95 return 0;
96}
97
98int amdgpu_mca_mp1_ras_sw_init(struct amdgpu_device *adev)
99{
100 int err;
101 struct amdgpu_mca_ras_block *ras;
102
103 if (!adev->mca.mp1.ras)
104 return 0;
105
106 ras = adev->mca.mp1.ras;
107
108 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
109 if (err) {
110 dev_err(adev->dev, "Failed to register mca.mp1 ras block!\n");
111 return err;
112 }
113
114 strcpy(ras->ras_block.ras_comm.name, "mca.mp1");
115 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
116 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
117 adev->mca.mp1.ras_if = &ras->ras_block.ras_comm;
118
119 return 0;
120}
121
122int amdgpu_mca_mpio_ras_sw_init(struct amdgpu_device *adev)
123{
124 int err;
125 struct amdgpu_mca_ras_block *ras;
126
127 if (!adev->mca.mpio.ras)
128 return 0;
129
130 ras = adev->mca.mpio.ras;
131
132 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
133 if (err) {
134 dev_err(adev->dev, "Failed to register mca.mpio ras block!\n");
135 return err;
136 }
137
138 strcpy(ras->ras_block.ras_comm.name, "mca.mpio");
139 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
140 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
141 adev->mca.mpio.ras_if = &ras->ras_block.ras_comm;
142
143 return 0;
144}
145
146void amdgpu_mca_bank_set_init(struct mca_bank_set *mca_set)
147{
148 if (!mca_set)
149 return;
150
151 memset(mca_set, 0, sizeof(*mca_set));
152 INIT_LIST_HEAD(&mca_set->list);
153}
154
155int amdgpu_mca_bank_set_add_entry(struct mca_bank_set *mca_set, struct mca_bank_entry *entry)
156{
157 struct mca_bank_node *node;
158
159 if (!entry)
160 return -EINVAL;
161
162 node = kvzalloc(sizeof(*node), GFP_KERNEL);
163 if (!node)
164 return -ENOMEM;
165
166 memcpy(&node->entry, entry, sizeof(*entry));
167
168 INIT_LIST_HEAD(&node->node);
169 list_add_tail(&node->node, &mca_set->list);
170
171 mca_set->nr_entries++;
172
173 return 0;
174}
175
176void amdgpu_mca_bank_set_release(struct mca_bank_set *mca_set)
177{
178 struct mca_bank_node *node, *tmp;
179
180 list_for_each_entry_safe(node, tmp, &mca_set->list, node) {
181 list_del(&node->node);
182 kvfree(node);
183 }
184}
185
186void amdgpu_mca_smu_init_funcs(struct amdgpu_device *adev, const struct amdgpu_mca_smu_funcs *mca_funcs)
187{
188 struct amdgpu_mca *mca = &adev->mca;
189
190 mca->mca_funcs = mca_funcs;
191}
192
193int amdgpu_mca_smu_set_debug_mode(struct amdgpu_device *adev, bool enable)
194{
195 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
196
197 if (mca_funcs && mca_funcs->mca_set_debug_mode)
198 return mca_funcs->mca_set_debug_mode(adev, enable);
199
200 return -EOPNOTSUPP;
201}
202
203static void amdgpu_mca_smu_mca_bank_dump(struct amdgpu_device *adev, int idx, struct mca_bank_entry *entry)
204{
205 dev_info(adev->dev, "[Hardware error] Accelerator Check Architecture events logged\n");
206 dev_info(adev->dev, "[Hardware error] aca entry[%02d].STATUS=0x%016llx\n",
207 idx, entry->regs[MCA_REG_IDX_STATUS]);
208 dev_info(adev->dev, "[Hardware error] aca entry[%02d].ADDR=0x%016llx\n",
209 idx, entry->regs[MCA_REG_IDX_ADDR]);
210 dev_info(adev->dev, "[Hardware error] aca entry[%02d].MISC0=0x%016llx\n",
211 idx, entry->regs[MCA_REG_IDX_MISC0]);
212 dev_info(adev->dev, "[Hardware error] aca entry[%02d].IPID=0x%016llx\n",
213 idx, entry->regs[MCA_REG_IDX_IPID]);
214 dev_info(adev->dev, "[Hardware error] aca entry[%02d].SYND=0x%016llx\n",
215 idx, entry->regs[MCA_REG_IDX_SYND]);
216}
217
218int amdgpu_mca_smu_log_ras_error(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type, struct ras_err_data *err_data)
219{
220 struct amdgpu_smuio_mcm_config_info mcm_info;
221 struct ras_err_addr err_addr = {0};
222 struct mca_bank_set mca_set;
223 struct mca_bank_node *node;
224 struct mca_bank_entry *entry;
225 uint32_t count;
226 int ret, i = 0;
227
228 amdgpu_mca_bank_set_init(&mca_set);
229
230 ret = amdgpu_mca_smu_get_mca_set(adev, blk, type, &mca_set);
231 if (ret)
232 goto out_mca_release;
233
234 list_for_each_entry(node, &mca_set.list, node) {
235 entry = &node->entry;
236
237 amdgpu_mca_smu_mca_bank_dump(adev, i++, entry);
238
239 count = 0;
240 ret = amdgpu_mca_smu_parse_mca_error_count(adev, blk, type, entry, &count);
241 if (ret)
242 goto out_mca_release;
243
244 if (!count)
245 continue;
246
247 mcm_info.socket_id = entry->info.socket_id;
248 mcm_info.die_id = entry->info.aid;
249
250 if (blk == AMDGPU_RAS_BLOCK__UMC) {
251 err_addr.err_status = entry->regs[MCA_REG_IDX_STATUS];
252 err_addr.err_ipid = entry->regs[MCA_REG_IDX_IPID];
253 err_addr.err_addr = entry->regs[MCA_REG_IDX_ADDR];
254 }
255
256 if (type == AMDGPU_MCA_ERROR_TYPE_UE)
257 amdgpu_ras_error_statistic_ue_count(err_data,
258 &mcm_info, &err_addr, (uint64_t)count);
259 else
260 amdgpu_ras_error_statistic_ce_count(err_data,
261 &mcm_info, &err_addr, (uint64_t)count);
262 }
263
264out_mca_release:
265 amdgpu_mca_bank_set_release(&mca_set);
266
267 return ret;
268}
269
270
271int amdgpu_mca_smu_get_valid_mca_count(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, uint32_t *count)
272{
273 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
274
275 if (!count)
276 return -EINVAL;
277
278 if (mca_funcs && mca_funcs->mca_get_valid_mca_count)
279 return mca_funcs->mca_get_valid_mca_count(adev, type, count);
280
281 return -EOPNOTSUPP;
282}
283
284int amdgpu_mca_smu_get_mca_set_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
285 enum amdgpu_mca_error_type type, uint32_t *total)
286{
287 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
288 struct mca_bank_set mca_set;
289 struct mca_bank_node *node;
290 struct mca_bank_entry *entry;
291 uint32_t count;
292 int ret;
293
294 if (!total)
295 return -EINVAL;
296
297 if (!mca_funcs)
298 return -EOPNOTSUPP;
299
300 if (!mca_funcs->mca_get_ras_mca_set || !mca_funcs->mca_get_valid_mca_count)
301 return -EOPNOTSUPP;
302
303 amdgpu_mca_bank_set_init(&mca_set);
304
305 ret = mca_funcs->mca_get_ras_mca_set(adev, blk, type, &mca_set);
306 if (ret)
307 goto err_mca_set_release;
308
309 *total = 0;
310 list_for_each_entry(node, &mca_set.list, node) {
311 entry = &node->entry;
312
313 count = 0;
314 ret = mca_funcs->mca_parse_mca_error_count(adev, blk, type, entry, &count);
315 if (ret)
316 goto err_mca_set_release;
317
318 *total += count;
319 }
320
321err_mca_set_release:
322 amdgpu_mca_bank_set_release(&mca_set);
323
324 return ret;
325}
326
327int amdgpu_mca_smu_parse_mca_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
328 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count)
329{
330 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
331 if (!count || !entry)
332 return -EINVAL;
333
334 if (!mca_funcs || !mca_funcs->mca_parse_mca_error_count)
335 return -EOPNOTSUPP;
336
337
338 return mca_funcs->mca_parse_mca_error_count(adev, blk, type, entry, count);
339}
340
341int amdgpu_mca_smu_get_mca_set(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
342 enum amdgpu_mca_error_type type, struct mca_bank_set *mca_set)
343{
344 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
345
346 if (!mca_set)
347 return -EINVAL;
348
349 if (!mca_funcs || !mca_funcs->mca_get_ras_mca_set)
350 return -EOPNOTSUPP;
351
352 WARN_ON(!list_empty(&mca_set->list));
353
354 return mca_funcs->mca_get_ras_mca_set(adev, blk, type, mca_set);
355}
356
357int amdgpu_mca_smu_get_mca_entry(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
358 int idx, struct mca_bank_entry *entry)
359{
360 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
361 int count;
362
363 if (!mca_funcs || !mca_funcs->mca_get_mca_entry)
364 return -EOPNOTSUPP;
365
366 switch (type) {
367 case AMDGPU_MCA_ERROR_TYPE_UE:
368 count = mca_funcs->max_ue_count;
369 break;
370 case AMDGPU_MCA_ERROR_TYPE_CE:
371 count = mca_funcs->max_ce_count;
372 break;
373 default:
374 return -EINVAL;
375 }
376
377 if (idx >= count)
378 return -EINVAL;
379
380 return mca_funcs->mca_get_mca_entry(adev, type, idx, entry);
381}
382
383#if defined(CONFIG_DEBUG_FS)
384static int amdgpu_mca_smu_debug_mode_set(void *data, u64 val)
385{
386 struct amdgpu_device *adev = (struct amdgpu_device *)data;
387 int ret;
388
389 ret = amdgpu_ras_set_mca_debug_mode(adev, val ? true : false);
390 if (ret)
391 return ret;
392
393 dev_info(adev->dev, "amdgpu set smu mca debug mode %s success\n", val ? "on" : "off");
394
395 return 0;
396}
397
398static void mca_dump_entry(struct seq_file *m, struct mca_bank_entry *entry)
399{
400 int i, idx = entry->idx;
401 int reg_idx_array[] = {
402 MCA_REG_IDX_STATUS,
403 MCA_REG_IDX_ADDR,
404 MCA_REG_IDX_MISC0,
405 MCA_REG_IDX_IPID,
406 MCA_REG_IDX_SYND,
407 };
408
409 seq_printf(m, "mca entry[%d].type: %s\n", idx, entry->type == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE");
410 seq_printf(m, "mca entry[%d].ip: %d\n", idx, entry->ip);
411 seq_printf(m, "mca entry[%d].info: socketid:%d aid:%d hwid:0x%03x mcatype:0x%04x\n",
412 idx, entry->info.socket_id, entry->info.aid, entry->info.hwid, entry->info.mcatype);
413
414 for (i = 0; i < ARRAY_SIZE(reg_idx_array); i++)
415 seq_printf(m, "mca entry[%d].regs[%d]: 0x%016llx\n", idx, reg_idx_array[i], entry->regs[reg_idx_array[i]]);
416}
417
418static int mca_dump_show(struct seq_file *m, enum amdgpu_mca_error_type type)
419{
420 struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
421 struct mca_bank_entry *entry;
422 uint32_t count = 0;
423 int i, ret;
424
425 ret = amdgpu_mca_smu_get_valid_mca_count(adev, type, &count);
426 if (ret)
427 return ret;
428
429 seq_printf(m, "amdgpu smu %s valid mca count: %d\n",
430 type == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE", count);
431
432 if (!count)
433 return 0;
434
435 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
436 if (!entry)
437 return -ENOMEM;
438
439 for (i = 0; i < count; i++) {
440 memset(entry, 0, sizeof(*entry));
441
442 ret = amdgpu_mca_smu_get_mca_entry(adev, type, i, entry);
443 if (ret)
444 goto err_free_entry;
445
446 mca_dump_entry(m, entry);
447 }
448
449err_free_entry:
450 kfree(entry);
451
452 return ret;
453}
454
455static int mca_dump_ce_show(struct seq_file *m, void *unused)
456{
457 return mca_dump_show(m, AMDGPU_MCA_ERROR_TYPE_CE);
458}
459
460static int mca_dump_ce_open(struct inode *inode, struct file *file)
461{
462 return single_open(file, mca_dump_ce_show, inode->i_private);
463}
464
465static const struct file_operations mca_ce_dump_debug_fops = {
466 .owner = THIS_MODULE,
467 .open = mca_dump_ce_open,
468 .read = seq_read,
469 .llseek = seq_lseek,
470 .release = single_release,
471};
472
473static int mca_dump_ue_show(struct seq_file *m, void *unused)
474{
475 return mca_dump_show(m, AMDGPU_MCA_ERROR_TYPE_UE);
476}
477
478static int mca_dump_ue_open(struct inode *inode, struct file *file)
479{
480 return single_open(file, mca_dump_ue_show, inode->i_private);
481}
482
483static const struct file_operations mca_ue_dump_debug_fops = {
484 .owner = THIS_MODULE,
485 .open = mca_dump_ue_open,
486 .read = seq_read,
487 .llseek = seq_lseek,
488 .release = single_release,
489};
490
491DEFINE_DEBUGFS_ATTRIBUTE(mca_debug_mode_fops, NULL, amdgpu_mca_smu_debug_mode_set, "%llu\n");
492#endif
493
494void amdgpu_mca_smu_debugfs_init(struct amdgpu_device *adev, struct dentry *root)
495{
496#if defined(CONFIG_DEBUG_FS)
497 if (!root || amdgpu_ip_version(adev, MP1_HWIP, 0) != IP_VERSION(13, 0, 6))
498 return;
499
500 debugfs_create_file("mca_debug_mode", 0200, root, adev, &mca_debug_mode_fops);
501 debugfs_create_file("mca_ue_dump", 0400, root, adev, &mca_ue_dump_debug_fops);
502 debugfs_create_file("mca_ce_dump", 0400, root, adev, &mca_ce_dump_debug_fops);
503#endif
504}
505
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 "amdgpu_ras.h"
24#include "amdgpu.h"
25#include "amdgpu_mca.h"
26
27#include "umc/umc_6_7_0_offset.h"
28#include "umc/umc_6_7_0_sh_mask.h"
29
30static bool amdgpu_mca_is_deferred_error(struct amdgpu_device *adev,
31 uint64_t mc_status)
32{
33 if (adev->umc.ras->check_ecc_err_status)
34 return adev->umc.ras->check_ecc_err_status(adev,
35 AMDGPU_MCA_ERROR_TYPE_DE, &mc_status);
36
37 return false;
38}
39
40void amdgpu_mca_query_correctable_error_count(struct amdgpu_device *adev,
41 uint64_t mc_status_addr,
42 unsigned long *error_count)
43{
44 uint64_t mc_status = RREG64_PCIE(mc_status_addr);
45
46 if (REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
47 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
48 *error_count += 1;
49}
50
51void amdgpu_mca_query_uncorrectable_error_count(struct amdgpu_device *adev,
52 uint64_t mc_status_addr,
53 unsigned long *error_count)
54{
55 uint64_t mc_status = RREG64_PCIE(mc_status_addr);
56
57 if ((REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
58 (REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
59 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
60 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
61 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
62 REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
63 *error_count += 1;
64}
65
66void amdgpu_mca_reset_error_count(struct amdgpu_device *adev,
67 uint64_t mc_status_addr)
68{
69 WREG64_PCIE(mc_status_addr, 0x0ULL);
70}
71
72void amdgpu_mca_query_ras_error_count(struct amdgpu_device *adev,
73 uint64_t mc_status_addr,
74 void *ras_error_status)
75{
76 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
77
78 amdgpu_mca_query_correctable_error_count(adev, mc_status_addr, &(err_data->ce_count));
79 amdgpu_mca_query_uncorrectable_error_count(adev, mc_status_addr, &(err_data->ue_count));
80
81 amdgpu_mca_reset_error_count(adev, mc_status_addr);
82}
83
84int amdgpu_mca_mp0_ras_sw_init(struct amdgpu_device *adev)
85{
86 int err;
87 struct amdgpu_mca_ras_block *ras;
88
89 if (!adev->mca.mp0.ras)
90 return 0;
91
92 ras = adev->mca.mp0.ras;
93
94 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
95 if (err) {
96 dev_err(adev->dev, "Failed to register mca.mp0 ras block!\n");
97 return err;
98 }
99
100 strcpy(ras->ras_block.ras_comm.name, "mca.mp0");
101 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
102 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
103 adev->mca.mp0.ras_if = &ras->ras_block.ras_comm;
104
105 return 0;
106}
107
108int amdgpu_mca_mp1_ras_sw_init(struct amdgpu_device *adev)
109{
110 int err;
111 struct amdgpu_mca_ras_block *ras;
112
113 if (!adev->mca.mp1.ras)
114 return 0;
115
116 ras = adev->mca.mp1.ras;
117
118 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
119 if (err) {
120 dev_err(adev->dev, "Failed to register mca.mp1 ras block!\n");
121 return err;
122 }
123
124 strcpy(ras->ras_block.ras_comm.name, "mca.mp1");
125 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
126 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
127 adev->mca.mp1.ras_if = &ras->ras_block.ras_comm;
128
129 return 0;
130}
131
132int amdgpu_mca_mpio_ras_sw_init(struct amdgpu_device *adev)
133{
134 int err;
135 struct amdgpu_mca_ras_block *ras;
136
137 if (!adev->mca.mpio.ras)
138 return 0;
139
140 ras = adev->mca.mpio.ras;
141
142 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
143 if (err) {
144 dev_err(adev->dev, "Failed to register mca.mpio ras block!\n");
145 return err;
146 }
147
148 strcpy(ras->ras_block.ras_comm.name, "mca.mpio");
149 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
150 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
151 adev->mca.mpio.ras_if = &ras->ras_block.ras_comm;
152
153 return 0;
154}
155
156static void amdgpu_mca_bank_set_init(struct mca_bank_set *mca_set)
157{
158 if (!mca_set)
159 return;
160
161 memset(mca_set, 0, sizeof(*mca_set));
162 INIT_LIST_HEAD(&mca_set->list);
163}
164
165static int amdgpu_mca_bank_set_add_entry(struct mca_bank_set *mca_set, struct mca_bank_entry *entry)
166{
167 struct mca_bank_node *node;
168
169 if (!entry)
170 return -EINVAL;
171
172 node = kvzalloc(sizeof(*node), GFP_KERNEL);
173 if (!node)
174 return -ENOMEM;
175
176 memcpy(&node->entry, entry, sizeof(*entry));
177
178 INIT_LIST_HEAD(&node->node);
179 list_add_tail(&node->node, &mca_set->list);
180
181 mca_set->nr_entries++;
182
183 return 0;
184}
185
186static int amdgpu_mca_bank_set_merge(struct mca_bank_set *mca_set, struct mca_bank_set *new)
187{
188 struct mca_bank_node *node;
189
190 list_for_each_entry(node, &new->list, node)
191 amdgpu_mca_bank_set_add_entry(mca_set, &node->entry);
192
193 return 0;
194}
195
196static void amdgpu_mca_bank_set_remove_node(struct mca_bank_set *mca_set, struct mca_bank_node *node)
197{
198 if (!node)
199 return;
200
201 list_del(&node->node);
202 kvfree(node);
203
204 mca_set->nr_entries--;
205}
206
207static void amdgpu_mca_bank_set_release(struct mca_bank_set *mca_set)
208{
209 struct mca_bank_node *node, *tmp;
210
211 if (list_empty(&mca_set->list))
212 return;
213
214 list_for_each_entry_safe(node, tmp, &mca_set->list, node)
215 amdgpu_mca_bank_set_remove_node(mca_set, node);
216}
217
218void amdgpu_mca_smu_init_funcs(struct amdgpu_device *adev, const struct amdgpu_mca_smu_funcs *mca_funcs)
219{
220 struct amdgpu_mca *mca = &adev->mca;
221
222 mca->mca_funcs = mca_funcs;
223}
224
225int amdgpu_mca_init(struct amdgpu_device *adev)
226{
227 struct amdgpu_mca *mca = &adev->mca;
228 struct mca_bank_cache *mca_cache;
229 int i;
230
231 atomic_set(&mca->ue_update_flag, 0);
232
233 for (i = 0; i < ARRAY_SIZE(mca->mca_caches); i++) {
234 mca_cache = &mca->mca_caches[i];
235 mutex_init(&mca_cache->lock);
236 amdgpu_mca_bank_set_init(&mca_cache->mca_set);
237 }
238
239 return 0;
240}
241
242void amdgpu_mca_fini(struct amdgpu_device *adev)
243{
244 struct amdgpu_mca *mca = &adev->mca;
245 struct mca_bank_cache *mca_cache;
246 int i;
247
248 atomic_set(&mca->ue_update_flag, 0);
249
250 for (i = 0; i < ARRAY_SIZE(mca->mca_caches); i++) {
251 mca_cache = &mca->mca_caches[i];
252 amdgpu_mca_bank_set_release(&mca_cache->mca_set);
253 mutex_destroy(&mca_cache->lock);
254 }
255}
256
257int amdgpu_mca_reset(struct amdgpu_device *adev)
258{
259 amdgpu_mca_fini(adev);
260
261 return amdgpu_mca_init(adev);
262}
263
264int amdgpu_mca_smu_set_debug_mode(struct amdgpu_device *adev, bool enable)
265{
266 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
267
268 if (mca_funcs && mca_funcs->mca_set_debug_mode)
269 return mca_funcs->mca_set_debug_mode(adev, enable);
270
271 return -EOPNOTSUPP;
272}
273
274static void amdgpu_mca_smu_mca_bank_dump(struct amdgpu_device *adev, int idx, struct mca_bank_entry *entry,
275 struct ras_query_context *qctx)
276{
277 u64 event_id = qctx ? qctx->evid.event_id : RAS_EVENT_INVALID_ID;
278
279 RAS_EVENT_LOG(adev, event_id, HW_ERR "Accelerator Check Architecture events logged\n");
280 RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].STATUS=0x%016llx\n",
281 idx, entry->regs[MCA_REG_IDX_STATUS]);
282 RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].ADDR=0x%016llx\n",
283 idx, entry->regs[MCA_REG_IDX_ADDR]);
284 RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].MISC0=0x%016llx\n",
285 idx, entry->regs[MCA_REG_IDX_MISC0]);
286 RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].IPID=0x%016llx\n",
287 idx, entry->regs[MCA_REG_IDX_IPID]);
288 RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].SYND=0x%016llx\n",
289 idx, entry->regs[MCA_REG_IDX_SYND]);
290}
291
292static int amdgpu_mca_smu_get_valid_mca_count(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, uint32_t *count)
293{
294 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
295
296 if (!count)
297 return -EINVAL;
298
299 if (mca_funcs && mca_funcs->mca_get_valid_mca_count)
300 return mca_funcs->mca_get_valid_mca_count(adev, type, count);
301
302 return -EOPNOTSUPP;
303}
304
305static int amdgpu_mca_smu_get_mca_entry(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
306 int idx, struct mca_bank_entry *entry)
307{
308 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
309 int count;
310
311 if (!mca_funcs || !mca_funcs->mca_get_mca_entry)
312 return -EOPNOTSUPP;
313
314 switch (type) {
315 case AMDGPU_MCA_ERROR_TYPE_UE:
316 count = mca_funcs->max_ue_count;
317 break;
318 case AMDGPU_MCA_ERROR_TYPE_CE:
319 count = mca_funcs->max_ce_count;
320 break;
321 default:
322 return -EINVAL;
323 }
324
325 if (idx >= count)
326 return -EINVAL;
327
328 return mca_funcs->mca_get_mca_entry(adev, type, idx, entry);
329}
330
331static bool amdgpu_mca_bank_should_update(struct amdgpu_device *adev, enum amdgpu_mca_error_type type)
332{
333 struct amdgpu_mca *mca = &adev->mca;
334 bool ret = true;
335
336 /*
337 * Because the UE Valid MCA count will only be cleared after reset,
338 * in order to avoid repeated counting of the error count,
339 * the aca bank is only updated once during the gpu recovery stage.
340 */
341 if (type == AMDGPU_MCA_ERROR_TYPE_UE) {
342 if (amdgpu_ras_intr_triggered())
343 ret = atomic_cmpxchg(&mca->ue_update_flag, 0, 1) == 0;
344 else
345 atomic_set(&mca->ue_update_flag, 0);
346 }
347
348 return ret;
349}
350
351static bool amdgpu_mca_bank_should_dump(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
352 struct mca_bank_entry *entry)
353{
354 bool ret;
355
356 switch (type) {
357 case AMDGPU_MCA_ERROR_TYPE_CE:
358 ret = amdgpu_mca_is_deferred_error(adev, entry->regs[MCA_REG_IDX_STATUS]);
359 break;
360 case AMDGPU_MCA_ERROR_TYPE_UE:
361 default:
362 ret = true;
363 break;
364 }
365
366 return ret;
367}
368
369static int amdgpu_mca_smu_get_mca_set(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, struct mca_bank_set *mca_set,
370 struct ras_query_context *qctx)
371{
372 struct mca_bank_entry entry;
373 uint32_t count = 0, i;
374 int ret;
375
376 if (!mca_set)
377 return -EINVAL;
378
379 if (!amdgpu_mca_bank_should_update(adev, type))
380 return 0;
381
382 ret = amdgpu_mca_smu_get_valid_mca_count(adev, type, &count);
383 if (ret)
384 return ret;
385
386 for (i = 0; i < count; i++) {
387 memset(&entry, 0, sizeof(entry));
388 ret = amdgpu_mca_smu_get_mca_entry(adev, type, i, &entry);
389 if (ret)
390 return ret;
391
392 amdgpu_mca_bank_set_add_entry(mca_set, &entry);
393
394 if (amdgpu_mca_bank_should_dump(adev, type, &entry))
395 amdgpu_mca_smu_mca_bank_dump(adev, i, &entry, qctx);
396 }
397
398 return 0;
399}
400
401static int amdgpu_mca_smu_parse_mca_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
402 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count)
403{
404 const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
405
406 if (!count || !entry)
407 return -EINVAL;
408
409 if (!mca_funcs || !mca_funcs->mca_parse_mca_error_count)
410 return -EOPNOTSUPP;
411
412 return mca_funcs->mca_parse_mca_error_count(adev, blk, type, entry, count);
413}
414
415static int amdgpu_mca_dispatch_mca_set(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type,
416 struct mca_bank_set *mca_set, struct ras_err_data *err_data)
417{
418 struct amdgpu_smuio_mcm_config_info mcm_info;
419 struct mca_bank_node *node, *tmp;
420 struct mca_bank_entry *entry;
421 uint32_t count;
422 int ret;
423
424 if (!mca_set)
425 return -EINVAL;
426
427 if (!mca_set->nr_entries)
428 return 0;
429
430 list_for_each_entry_safe(node, tmp, &mca_set->list, node) {
431 entry = &node->entry;
432
433 count = 0;
434 ret = amdgpu_mca_smu_parse_mca_error_count(adev, blk, type, entry, &count);
435 if (ret && ret != -EOPNOTSUPP)
436 return ret;
437
438 if (!count)
439 continue;
440
441 memset(&mcm_info, 0, sizeof(mcm_info));
442
443 mcm_info.socket_id = entry->info.socket_id;
444 mcm_info.die_id = entry->info.aid;
445
446 if (type == AMDGPU_MCA_ERROR_TYPE_UE) {
447 amdgpu_ras_error_statistic_ue_count(err_data,
448 &mcm_info, (uint64_t)count);
449 } else {
450 if (amdgpu_mca_is_deferred_error(adev, entry->regs[MCA_REG_IDX_STATUS]))
451 amdgpu_ras_error_statistic_de_count(err_data,
452 &mcm_info, (uint64_t)count);
453 else
454 amdgpu_ras_error_statistic_ce_count(err_data,
455 &mcm_info, (uint64_t)count);
456 }
457
458 amdgpu_mca_bank_set_remove_node(mca_set, node);
459 }
460
461 return 0;
462}
463
464static int amdgpu_mca_add_mca_set_to_cache(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, struct mca_bank_set *new)
465{
466 struct mca_bank_cache *mca_cache = &adev->mca.mca_caches[type];
467 int ret;
468
469 mutex_lock(&mca_cache->lock);
470 ret = amdgpu_mca_bank_set_merge(&mca_cache->mca_set, new);
471 mutex_unlock(&mca_cache->lock);
472
473 return ret;
474}
475
476int amdgpu_mca_smu_log_ras_error(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type,
477 struct ras_err_data *err_data, struct ras_query_context *qctx)
478{
479 struct mca_bank_set mca_set;
480 struct mca_bank_cache *mca_cache = &adev->mca.mca_caches[type];
481 int ret;
482
483 amdgpu_mca_bank_set_init(&mca_set);
484
485 ret = amdgpu_mca_smu_get_mca_set(adev, type, &mca_set, qctx);
486 if (ret)
487 goto out_mca_release;
488
489 ret = amdgpu_mca_dispatch_mca_set(adev, blk, type, &mca_set, err_data);
490 if (ret)
491 goto out_mca_release;
492
493 /* add remain mca bank to mca cache */
494 if (mca_set.nr_entries) {
495 ret = amdgpu_mca_add_mca_set_to_cache(adev, type, &mca_set);
496 if (ret)
497 goto out_mca_release;
498 }
499
500 /* dispatch mca set again if mca cache has valid data */
501 mutex_lock(&mca_cache->lock);
502 if (mca_cache->mca_set.nr_entries)
503 ret = amdgpu_mca_dispatch_mca_set(adev, blk, type, &mca_cache->mca_set, err_data);
504 mutex_unlock(&mca_cache->lock);
505
506out_mca_release:
507 amdgpu_mca_bank_set_release(&mca_set);
508
509 return ret;
510}
511
512#if defined(CONFIG_DEBUG_FS)
513static int amdgpu_mca_smu_debug_mode_set(void *data, u64 val)
514{
515 struct amdgpu_device *adev = (struct amdgpu_device *)data;
516 int ret;
517
518 ret = amdgpu_ras_set_mca_debug_mode(adev, val ? true : false);
519 if (ret)
520 return ret;
521
522 dev_info(adev->dev, "amdgpu set smu mca debug mode %s success\n", val ? "on" : "off");
523
524 return 0;
525}
526
527static void mca_dump_entry(struct seq_file *m, struct mca_bank_entry *entry)
528{
529 int i, idx = entry->idx;
530 int reg_idx_array[] = {
531 MCA_REG_IDX_STATUS,
532 MCA_REG_IDX_ADDR,
533 MCA_REG_IDX_MISC0,
534 MCA_REG_IDX_IPID,
535 MCA_REG_IDX_SYND,
536 };
537
538 seq_printf(m, "mca entry[%d].type: %s\n", idx, entry->type == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE");
539 seq_printf(m, "mca entry[%d].ip: %d\n", idx, entry->ip);
540 seq_printf(m, "mca entry[%d].info: socketid:%d aid:%d hwid:0x%03x mcatype:0x%04x\n",
541 idx, entry->info.socket_id, entry->info.aid, entry->info.hwid, entry->info.mcatype);
542
543 for (i = 0; i < ARRAY_SIZE(reg_idx_array); i++)
544 seq_printf(m, "mca entry[%d].regs[%d]: 0x%016llx\n", idx, reg_idx_array[i], entry->regs[reg_idx_array[i]]);
545}
546
547static int mca_dump_show(struct seq_file *m, enum amdgpu_mca_error_type type)
548{
549 struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
550 struct mca_bank_node *node;
551 struct mca_bank_set mca_set;
552 struct ras_query_context qctx;
553 int ret;
554
555 amdgpu_mca_bank_set_init(&mca_set);
556
557 qctx.evid.event_id = RAS_EVENT_INVALID_ID;
558 ret = amdgpu_mca_smu_get_mca_set(adev, type, &mca_set, &qctx);
559 if (ret)
560 goto err_free_mca_set;
561
562 seq_printf(m, "amdgpu smu %s valid mca count: %d\n",
563 type == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE", mca_set.nr_entries);
564
565 if (!mca_set.nr_entries)
566 goto err_free_mca_set;
567
568 list_for_each_entry(node, &mca_set.list, node)
569 mca_dump_entry(m, &node->entry);
570
571 /* add mca bank to mca bank cache */
572 ret = amdgpu_mca_add_mca_set_to_cache(adev, type, &mca_set);
573
574err_free_mca_set:
575 amdgpu_mca_bank_set_release(&mca_set);
576
577 return ret;
578}
579
580static int mca_dump_ce_show(struct seq_file *m, void *unused)
581{
582 return mca_dump_show(m, AMDGPU_MCA_ERROR_TYPE_CE);
583}
584
585static int mca_dump_ce_open(struct inode *inode, struct file *file)
586{
587 return single_open(file, mca_dump_ce_show, inode->i_private);
588}
589
590static const struct file_operations mca_ce_dump_debug_fops = {
591 .owner = THIS_MODULE,
592 .open = mca_dump_ce_open,
593 .read = seq_read,
594 .llseek = seq_lseek,
595 .release = single_release,
596};
597
598static int mca_dump_ue_show(struct seq_file *m, void *unused)
599{
600 return mca_dump_show(m, AMDGPU_MCA_ERROR_TYPE_UE);
601}
602
603static int mca_dump_ue_open(struct inode *inode, struct file *file)
604{
605 return single_open(file, mca_dump_ue_show, inode->i_private);
606}
607
608static const struct file_operations mca_ue_dump_debug_fops = {
609 .owner = THIS_MODULE,
610 .open = mca_dump_ue_open,
611 .read = seq_read,
612 .llseek = seq_lseek,
613 .release = single_release,
614};
615
616DEFINE_DEBUGFS_ATTRIBUTE(mca_debug_mode_fops, NULL, amdgpu_mca_smu_debug_mode_set, "%llu\n");
617#endif
618
619void amdgpu_mca_smu_debugfs_init(struct amdgpu_device *adev, struct dentry *root)
620{
621#if defined(CONFIG_DEBUG_FS)
622 if (!root)
623 return;
624
625 debugfs_create_file("mca_debug_mode", 0200, root, adev, &mca_debug_mode_fops);
626 debugfs_create_file("mca_ue_dump", 0400, root, adev, &mca_ue_dump_debug_fops);
627 debugfs_create_file("mca_ce_dump", 0400, root, adev, &mca_ce_dump_debug_fops);
628#endif
629}
630