1 /*
2 * Copyright 2019 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
24 #include <linux/sort.h>
25 #include "amdgpu.h"
26 #include "umc_v6_7.h"
27 #include "amdgpu_ras_mgr.h"
28 #define MAX_UMC_POISON_POLLING_TIME_SYNC 20 //ms
29
30 #define MAX_UMC_HASH_STRING_SIZE 256
31
amdgpu_umc_convert_error_address(struct amdgpu_device * adev,struct ras_err_data * err_data,uint64_t err_addr,uint32_t ch_inst,uint32_t umc_inst)32 static int amdgpu_umc_convert_error_address(struct amdgpu_device *adev,
33 struct ras_err_data *err_data, uint64_t err_addr,
34 uint32_t ch_inst, uint32_t umc_inst)
35 {
36 switch (amdgpu_ip_version(adev, UMC_HWIP, 0)) {
37 case IP_VERSION(6, 7, 0):
38 umc_v6_7_convert_error_address(adev,
39 err_data, err_addr, ch_inst, umc_inst);
40 break;
41 default:
42 dev_warn(adev->dev,
43 "UMC address to Physical address translation is not supported\n");
44 return AMDGPU_RAS_FAIL;
45 }
46
47 return AMDGPU_RAS_SUCCESS;
48 }
49
amdgpu_umc_page_retirement_mca(struct amdgpu_device * adev,uint64_t err_addr,uint32_t ch_inst,uint32_t umc_inst)50 int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
51 uint64_t err_addr, uint32_t ch_inst, uint32_t umc_inst)
52 {
53 struct ras_err_data err_data;
54 int ret;
55
56 ret = amdgpu_ras_error_data_init(&err_data);
57 if (ret)
58 return ret;
59
60 err_data.err_addr =
61 kzalloc_objs(struct eeprom_table_record,
62 adev->umc.max_ras_err_cnt_per_query);
63 if (!err_data.err_addr) {
64 dev_warn(adev->dev,
65 "Failed to alloc memory for umc error record in MCA notifier!\n");
66 ret = AMDGPU_RAS_FAIL;
67 goto out_fini_err_data;
68 }
69
70 err_data.err_addr_len = adev->umc.max_ras_err_cnt_per_query;
71
72 /*
73 * Translate UMC channel address to Physical address
74 */
75 ret = amdgpu_umc_convert_error_address(adev, &err_data, err_addr,
76 ch_inst, umc_inst);
77 if (ret)
78 goto out_free_err_addr;
79
80 if (amdgpu_bad_page_threshold != 0) {
81 amdgpu_ras_add_bad_pages(adev, err_data.err_addr,
82 err_data.err_addr_cnt, false);
83 amdgpu_ras_save_bad_pages(adev, NULL);
84 }
85
86 out_free_err_addr:
87 kfree(err_data.err_addr);
88
89 out_fini_err_data:
90 amdgpu_ras_error_data_fini(&err_data);
91
92 return ret;
93 }
94
amdgpu_umc_handle_bad_pages(struct amdgpu_device * adev,void * ras_error_status)95 void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
96 void *ras_error_status)
97 {
98 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
99 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
100 unsigned int error_query_mode;
101 int ret = 0;
102 unsigned long err_count;
103
104 amdgpu_ras_get_error_query_mode(adev, &error_query_mode);
105
106 err_data->err_addr =
107 kzalloc_objs(struct eeprom_table_record,
108 adev->umc.max_ras_err_cnt_per_query);
109
110 /* still call query_ras_error_address to clear error status
111 * even NOMEM error is encountered
112 */
113 if (!err_data->err_addr)
114 dev_warn(adev->dev,
115 "Failed to alloc memory for umc error address record!\n");
116 else
117 err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query;
118
119 mutex_lock(&con->page_retirement_lock);
120 ret = amdgpu_dpm_get_ecc_info(adev, (void *)&(con->umc_ecc));
121 if (ret == -EOPNOTSUPP &&
122 error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY) {
123 if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops &&
124 adev->umc.ras->ras_block.hw_ops->query_ras_error_count)
125 adev->umc.ras->ras_block.hw_ops->query_ras_error_count(adev,
126 ras_error_status);
127
128 if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops &&
129 adev->umc.ras->ras_block.hw_ops->query_ras_error_address &&
130 adev->umc.max_ras_err_cnt_per_query) {
131 err_data->err_addr =
132 kzalloc_objs(struct eeprom_table_record,
133 adev->umc.max_ras_err_cnt_per_query);
134
135 /* still call query_ras_error_address to clear error status
136 * even NOMEM error is encountered
137 */
138 if (!err_data->err_addr)
139 dev_warn(adev->dev,
140 "Failed to alloc memory for umc error address record!\n");
141 else
142 err_data->err_addr_len =
143 adev->umc.max_ras_err_cnt_per_query;
144
145 /* umc query_ras_error_address is also responsible for clearing
146 * error status
147 */
148 adev->umc.ras->ras_block.hw_ops->query_ras_error_address(adev,
149 ras_error_status);
150 }
151 } else if (error_query_mode == AMDGPU_RAS_FIRMWARE_ERROR_QUERY ||
152 (!ret && error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY)) {
153 if (adev->umc.ras &&
154 adev->umc.ras->ecc_info_query_ras_error_count)
155 adev->umc.ras->ecc_info_query_ras_error_count(adev,
156 ras_error_status);
157
158 if (adev->umc.ras &&
159 adev->umc.ras->ecc_info_query_ras_error_address &&
160 adev->umc.max_ras_err_cnt_per_query) {
161 err_data->err_addr =
162 kzalloc_objs(struct eeprom_table_record,
163 adev->umc.max_ras_err_cnt_per_query);
164
165 /* still call query_ras_error_address to clear error status
166 * even NOMEM error is encountered
167 */
168 if (!err_data->err_addr)
169 dev_warn(adev->dev,
170 "Failed to alloc memory for umc error address record!\n");
171 else
172 err_data->err_addr_len =
173 adev->umc.max_ras_err_cnt_per_query;
174
175 /* umc query_ras_error_address is also responsible for clearing
176 * error status
177 */
178 adev->umc.ras->ecc_info_query_ras_error_address(adev,
179 ras_error_status);
180 }
181 }
182
183 /* only uncorrectable error needs gpu reset */
184 if (err_data->ue_count || err_data->de_count) {
185 err_count = err_data->ue_count + err_data->de_count;
186 if ((amdgpu_bad_page_threshold != 0) &&
187 err_data->err_addr_cnt) {
188 amdgpu_ras_add_bad_pages(adev, err_data->err_addr,
189 err_data->err_addr_cnt, false);
190 amdgpu_ras_save_bad_pages(adev, &err_count);
191
192 amdgpu_dpm_send_hbm_bad_pages_num(adev,
193 con->eeprom_control.ras_num_bad_pages);
194
195 if (con->update_channel_flag == true) {
196 amdgpu_dpm_send_hbm_bad_channel_flag(adev, con->eeprom_control.bad_channel_bitmap);
197 con->update_channel_flag = false;
198 }
199 }
200 }
201
202 kfree(err_data->err_addr);
203 err_data->err_addr = NULL;
204
205 mutex_unlock(&con->page_retirement_lock);
206 }
207
amdgpu_umc_do_page_retirement(struct amdgpu_device * adev,void * ras_error_status,struct amdgpu_iv_entry * entry,uint32_t reset)208 static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev,
209 void *ras_error_status,
210 struct amdgpu_iv_entry *entry,
211 uint32_t reset)
212 {
213 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
214 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
215
216 kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
217 amdgpu_umc_handle_bad_pages(adev, ras_error_status);
218
219 if ((err_data->ue_count || err_data->de_count) &&
220 (reset || amdgpu_ras_is_rma(adev))) {
221 con->gpu_reset_flags |= reset;
222 amdgpu_ras_reset_gpu(adev);
223 }
224
225 return AMDGPU_RAS_SUCCESS;
226 }
227
amdgpu_umc_pasid_poison_handler(struct amdgpu_device * adev,enum amdgpu_ras_block block,uint16_t pasid,pasid_notify pasid_fn,void * data,uint32_t reset)228 int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev,
229 enum amdgpu_ras_block block, uint16_t pasid,
230 pasid_notify pasid_fn, void *data, uint32_t reset)
231 {
232 int ret = AMDGPU_RAS_SUCCESS;
233
234 if (adev->gmc.xgmi.connected_to_cpu ||
235 adev->gmc.is_app_apu) {
236 if (reset) {
237 /* MCA poison handler is only responsible for GPU reset,
238 * let MCA notifier do page retirement.
239 */
240 kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
241 amdgpu_ras_reset_gpu(adev);
242 }
243 return ret;
244 }
245
246 if (!amdgpu_sriov_vf(adev)) {
247 if (amdgpu_ip_version(adev, UMC_HWIP, 0) < IP_VERSION(12, 0, 0)) {
248 struct ras_err_data err_data;
249 struct ras_common_if head = {
250 .block = AMDGPU_RAS_BLOCK__UMC,
251 };
252 struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
253
254 ret = amdgpu_ras_error_data_init(&err_data);
255 if (ret)
256 return ret;
257
258 ret = amdgpu_umc_do_page_retirement(adev, &err_data, NULL, reset);
259
260 if (ret == AMDGPU_RAS_SUCCESS && obj) {
261 obj->err_data.ue_count += err_data.ue_count;
262 obj->err_data.ce_count += err_data.ce_count;
263 }
264
265 amdgpu_ras_error_data_fini(&err_data);
266 } else {
267 struct ras_ih_info ih_info = {0};
268
269 ih_info.block = block;
270 ih_info.pasid = pasid;
271 ih_info.reset = reset;
272 ih_info.pasid_fn = pasid_fn;
273 ih_info.data = data;
274 amdgpu_ras_mgr_handle_consumer_interrupt(adev, &ih_info);
275 }
276 } else {
277 if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
278 adev->virt.ops->ras_poison_handler(adev, block);
279 else
280 dev_warn(adev->dev,
281 "No ras_poison_handler interface in SRIOV!\n");
282 }
283
284 return ret;
285 }
286
amdgpu_umc_poison_handler(struct amdgpu_device * adev,enum amdgpu_ras_block block,uint32_t reset)287 int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
288 enum amdgpu_ras_block block, uint32_t reset)
289 {
290 return amdgpu_umc_pasid_poison_handler(adev,
291 block, 0, NULL, NULL, reset);
292 }
293
amdgpu_umc_process_ras_data_cb(struct amdgpu_device * adev,void * ras_error_status,struct amdgpu_iv_entry * entry)294 int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev,
295 void *ras_error_status,
296 struct amdgpu_iv_entry *entry)
297 {
298 return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry,
299 AMDGPU_RAS_GPU_RESET_MODE1_RESET);
300 }
301
amdgpu_umc_ras_sw_init(struct amdgpu_device * adev)302 int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev)
303 {
304 int err;
305 struct amdgpu_umc_ras *ras;
306
307 if (!adev->umc.ras)
308 return 0;
309
310 ras = adev->umc.ras;
311
312 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
313 if (err) {
314 dev_err(adev->dev, "Failed to register umc ras block!\n");
315 return err;
316 }
317
318 strcpy(adev->umc.ras->ras_block.ras_comm.name, "umc");
319 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__UMC;
320 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
321 adev->umc.ras_if = &ras->ras_block.ras_comm;
322
323 if (!ras->ras_block.ras_late_init)
324 ras->ras_block.ras_late_init = amdgpu_umc_ras_late_init;
325
326 if (!ras->ras_block.ras_cb)
327 ras->ras_block.ras_cb = amdgpu_umc_process_ras_data_cb;
328
329 return 0;
330 }
331
amdgpu_umc_ras_late_init(struct amdgpu_device * adev,struct ras_common_if * ras_block)332 int amdgpu_umc_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
333 {
334 int r;
335
336 r = amdgpu_ras_block_late_init(adev, ras_block);
337 if (r)
338 return r;
339
340 if (amdgpu_sriov_vf(adev))
341 return r;
342
343 if (amdgpu_ras_is_supported(adev, ras_block->block)) {
344 r = amdgpu_irq_get(adev, &adev->gmc.ecc_irq, 0);
345 if (r)
346 goto late_fini;
347 }
348
349 /* ras init of specific umc version */
350 if (adev->umc.ras &&
351 adev->umc.ras->err_cnt_init)
352 adev->umc.ras->err_cnt_init(adev);
353
354 return 0;
355
356 late_fini:
357 amdgpu_ras_block_late_fini(adev, ras_block);
358 return r;
359 }
360
amdgpu_umc_process_ecc_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)361 int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev,
362 struct amdgpu_irq_src *source,
363 struct amdgpu_iv_entry *entry)
364 {
365 struct ras_common_if *ras_if = adev->umc.ras_if;
366 struct ras_dispatch_if ih_data = {
367 .entry = entry,
368 };
369
370 if (!ras_if)
371 return 0;
372
373 ih_data.head = *ras_if;
374
375 amdgpu_ras_interrupt_dispatch(adev, &ih_data);
376 return 0;
377 }
378
amdgpu_umc_uniras_process_ecc_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)379 int amdgpu_umc_uniras_process_ecc_irq(struct amdgpu_device *adev,
380 struct amdgpu_irq_src *source,
381 struct amdgpu_iv_entry *entry)
382 {
383 struct ras_ih_info ih_info = {0};
384
385 ih_info.block = RAS_BLOCK_ID__UMC;
386 amdgpu_ras_mgr_dispatch_interrupt(adev, &ih_info);
387 return 0;
388 }
389
amdgpu_umc_fill_error_record(struct ras_err_data * err_data,uint64_t err_addr,uint64_t retired_page,uint32_t channel_index,uint32_t umc_inst)390 int amdgpu_umc_fill_error_record(struct ras_err_data *err_data,
391 uint64_t err_addr,
392 uint64_t retired_page,
393 uint32_t channel_index,
394 uint32_t umc_inst)
395 {
396 struct eeprom_table_record *err_rec;
397
398 if (!err_data ||
399 !err_data->err_addr ||
400 (err_data->err_addr_cnt >= err_data->err_addr_len))
401 return -EINVAL;
402
403 err_rec = &err_data->err_addr[err_data->err_addr_cnt];
404
405 err_rec->address = err_addr;
406 /* page frame address is saved */
407 err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
408 err_rec->ts = (uint64_t)ktime_get_real_seconds();
409 err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
410 err_rec->cu = 0;
411 err_rec->mem_channel = channel_index;
412 err_rec->mcumc_id = umc_inst;
413
414 err_data->err_addr_cnt++;
415
416 return 0;
417 }
418
amdgpu_umc_loop_all_aid(struct amdgpu_device * adev,umc_func func,void * data)419 static int amdgpu_umc_loop_all_aid(struct amdgpu_device *adev, umc_func func,
420 void *data)
421 {
422 uint32_t umc_node_inst;
423 uint32_t node_inst;
424 uint32_t umc_inst;
425 uint32_t ch_inst;
426 int ret;
427
428 /*
429 * This loop is done based on the following -
430 * umc.active mask = mask of active umc instances across all nodes
431 * umc.umc_inst_num = maximum number of umc instancess per node
432 * umc.node_inst_num = maximum number of node instances
433 * Channel instances are not assumed to be harvested.
434 */
435 dev_dbg(adev->dev, "active umcs :%lx umc_inst per node: %d",
436 adev->umc.active_mask, adev->umc.umc_inst_num);
437 for_each_set_bit(umc_node_inst, &(adev->umc.active_mask),
438 adev->umc.node_inst_num * adev->umc.umc_inst_num) {
439 node_inst = umc_node_inst / adev->umc.umc_inst_num;
440 umc_inst = umc_node_inst % adev->umc.umc_inst_num;
441 LOOP_UMC_CH_INST(ch_inst) {
442 dev_dbg(adev->dev,
443 "node_inst :%d umc_inst: %d ch_inst: %d",
444 node_inst, umc_inst, ch_inst);
445 ret = func(adev, node_inst, umc_inst, ch_inst, data);
446 if (ret) {
447 dev_err(adev->dev,
448 "Node %d umc %d ch %d func returns %d\n",
449 node_inst, umc_inst, ch_inst, ret);
450 return ret;
451 }
452 }
453 }
454
455 return 0;
456 }
457
amdgpu_umc_loop_channels(struct amdgpu_device * adev,umc_func func,void * data)458 int amdgpu_umc_loop_channels(struct amdgpu_device *adev,
459 umc_func func, void *data)
460 {
461 uint32_t node_inst = 0;
462 uint32_t umc_inst = 0;
463 uint32_t ch_inst = 0;
464 int ret = 0;
465
466 if (adev->aid_mask)
467 return amdgpu_umc_loop_all_aid(adev, func, data);
468
469 if (adev->umc.node_inst_num) {
470 LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
471 ret = func(adev, node_inst, umc_inst, ch_inst, data);
472 if (ret) {
473 dev_err(adev->dev, "Node %d umc %d ch %d func returns %d\n",
474 node_inst, umc_inst, ch_inst, ret);
475 return ret;
476 }
477 }
478 } else {
479 LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) {
480 ret = func(adev, 0, umc_inst, ch_inst, data);
481 if (ret) {
482 dev_err(adev->dev, "Umc %d ch %d func returns %d\n",
483 umc_inst, ch_inst, ret);
484 return ret;
485 }
486 }
487 }
488
489 return 0;
490 }
491