xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision 570f7e331f5febb30f1384817463c7e42b65ca7d)
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 "amdgpu_ras_eeprom.h"
25 #include "amdgpu.h"
26 #include "amdgpu_ras.h"
27 #include <linux/bits.h>
28 #include "atom.h"
29 #include "amdgpu_eeprom.h"
30 #include "amdgpu_atomfirmware.h"
31 #include <linux/debugfs.h>
32 #include <linux/uaccess.h>
33 
34 #include "amdgpu_reset.h"
35 #include "amdgpu_ras_mgr.h"
36 
37 /* These are memory addresses as would be seen by one or more EEPROM
38  * chips strung on the I2C bus, usually by manipulating pins 1-3 of a
39  * set of EEPROM devices. They form a continuous memory space.
40  *
41  * The I2C device address includes the device type identifier, 1010b,
42  * which is a reserved value and indicates that this is an I2C EEPROM
43  * device. It also includes the top 3 bits of the 19 bit EEPROM memory
44  * address, namely bits 18, 17, and 16. This makes up the 7 bit
45  * address sent on the I2C bus with bit 0 being the direction bit,
46  * which is not represented here, and sent by the hardware directly.
47  *
48  * For instance,
49  *   50h = 1010000b => device type identifier 1010b, bits 18:16 = 000b, address 0.
50  *   54h = 1010100b => --"--, bits 18:16 = 100b, address 40000h.
51  *   56h = 1010110b => --"--, bits 18:16 = 110b, address 60000h.
52  * Depending on the size of the I2C EEPROM device(s), bits 18:16 may
53  * address memory in a device or a device on the I2C bus, depending on
54  * the status of pins 1-3. See top of amdgpu_eeprom.c.
55  *
56  * The RAS table lives either at address 0 or address 40000h of EEPROM.
57  */
58 #define EEPROM_I2C_MADDR_0      0x0
59 #define EEPROM_I2C_MADDR_4      0x40000
60 
61 /*
62  * The 2 macros below represent the actual size in bytes that
63  * those entities occupy in the EEPROM memory.
64  * RAS_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
65  * uses uint64 to store 6b fields such as retired_page.
66  */
67 #define RAS_TABLE_HEADER_SIZE   20
68 #define RAS_TABLE_RECORD_SIZE   24
69 
70 /* Table hdr is 'AMDR' */
71 #define RAS_TABLE_HDR_VAL       0x414d4452
72 
73 /* Bad GPU tag ‘BADG’ */
74 #define RAS_TABLE_HDR_BAD       0x42414447
75 
76 /*
77  * EEPROM Table structure v1
78  * ---------------------------------
79  * |                               |
80  * |     EEPROM TABLE HEADER       |
81  * |      ( size 20 Bytes )        |
82  * |                               |
83  * ---------------------------------
84  * |                               |
85  * |    BAD PAGE RECORD AREA       |
86  * |                               |
87  * ---------------------------------
88  */
89 
90 /* Assume 2-Mbit size EEPROM and take up the whole space. */
91 #define RAS_TBL_SIZE_BYTES      (256 * 1024)
92 #define RAS_TABLE_START         0
93 #define RAS_HDR_START           RAS_TABLE_START
94 #define RAS_RECORD_START        (RAS_HDR_START + RAS_TABLE_HEADER_SIZE)
95 #define RAS_MAX_RECORD_COUNT    ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE) \
96 				 / RAS_TABLE_RECORD_SIZE)
97 
98 /*
99  * EEPROM Table structrue v2.1
100  * ---------------------------------
101  * |                               |
102  * |     EEPROM TABLE HEADER       |
103  * |      ( size 20 Bytes )        |
104  * |                               |
105  * ---------------------------------
106  * |                               |
107  * |     EEPROM TABLE RAS INFO     |
108  * | (available info size 4 Bytes) |
109  * |  ( reserved size 252 Bytes )  |
110  * |                               |
111  * ---------------------------------
112  * |                               |
113  * |     BAD PAGE RECORD AREA      |
114  * |                               |
115  * ---------------------------------
116  */
117 
118 /* EEPROM Table V2_1 */
119 #define RAS_TABLE_V2_1_INFO_SIZE       256
120 #define RAS_TABLE_V2_1_INFO_START      RAS_TABLE_HEADER_SIZE
121 #define RAS_RECORD_START_V2_1          (RAS_HDR_START + RAS_TABLE_HEADER_SIZE + \
122 					RAS_TABLE_V2_1_INFO_SIZE)
123 #define RAS_MAX_RECORD_COUNT_V2_1      ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE - \
124 					RAS_TABLE_V2_1_INFO_SIZE) \
125 					/ RAS_TABLE_RECORD_SIZE)
126 
127 /* Given a zero-based index of an EEPROM RAS record, yields the EEPROM
128  * offset off of RAS_TABLE_START.  That is, this is something you can
129  * add to control->i2c_address, and then tell I2C layer to read
130  * from/write to there. _N is the so called absolute index,
131  * because it starts right after the table header.
132  */
133 #define RAS_INDEX_TO_OFFSET(_C, _N) ((_C)->ras_record_offset + \
134 				     (_N) * RAS_TABLE_RECORD_SIZE)
135 
136 #define RAS_OFFSET_TO_INDEX(_C, _O) (((_O) - \
137 				      (_C)->ras_record_offset) / RAS_TABLE_RECORD_SIZE)
138 
139 /* Given a 0-based relative record index, 0, 1, 2, ..., etc., off
140  * of "fri", return the absolute record index off of the end of
141  * the table header.
142  */
143 #define RAS_RI_TO_AI(_C, _I) (((_I) + (_C)->ras_fri) % \
144 			      (_C)->ras_max_record_count)
145 
146 #define RAS_NUM_RECS(_tbl_hdr) \
147 	(((_tbl_hdr)->tbl_size < RAS_TABLE_HEADER_SIZE) ? 0u : \
148 	 (((_tbl_hdr)->tbl_size - RAS_TABLE_HEADER_SIZE) / RAS_TABLE_RECORD_SIZE))
149 
150 #define RAS_NUM_RECS_V2_1(_tbl_hdr) \
151 	(((_tbl_hdr)->tbl_size < RAS_TABLE_HEADER_SIZE + \
152 	  RAS_TABLE_V2_1_INFO_SIZE) ? 0u : \
153 	 (((_tbl_hdr)->tbl_size - RAS_TABLE_HEADER_SIZE - \
154 	   RAS_TABLE_V2_1_INFO_SIZE) / RAS_TABLE_RECORD_SIZE))
155 
156 #define to_amdgpu_device(x) ((container_of(x, struct amdgpu_ras, eeprom_control))->adev)
157 
158 static bool __is_ras_eeprom_supported(struct amdgpu_device *adev)
159 {
160 	if (amdgpu_sriov_vf(adev))
161 		return false;
162 
163 	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
164 	case IP_VERSION(11, 0, 2): /* VEGA20 and ARCTURUS */
165 	case IP_VERSION(11, 0, 7): /* Sienna cichlid */
166 	case IP_VERSION(13, 0, 0):
167 	case IP_VERSION(13, 0, 2): /* Aldebaran */
168 	case IP_VERSION(13, 0, 10):
169 		return true;
170 	case IP_VERSION(13, 0, 6):
171 	case IP_VERSION(13, 0, 12):
172 	case IP_VERSION(13, 0, 14):
173 		return (adev->gmc.is_app_apu) ? false : true;
174 	default:
175 		return false;
176 	}
177 }
178 
179 static bool __get_eeprom_i2c_addr(struct amdgpu_device *adev,
180 				  struct amdgpu_ras_eeprom_control *control)
181 {
182 	struct atom_context *atom_ctx = adev->mode_info.atom_context;
183 	u8 i2c_addr;
184 
185 	if (!control)
186 		return false;
187 
188 	if (adev->bios && amdgpu_atomfirmware_ras_rom_addr(adev, &i2c_addr)) {
189 		/* The address given by VBIOS is an 8-bit, wire-format
190 		 * address, i.e. the most significant byte.
191 		 *
192 		 * Normalize it to a 19-bit EEPROM address. Remove the
193 		 * device type identifier and make it a 7-bit address;
194 		 * then make it a 19-bit EEPROM address. See top of
195 		 * amdgpu_eeprom.c.
196 		 */
197 		i2c_addr = (i2c_addr & 0x0F) >> 1;
198 		control->i2c_address = ((u32) i2c_addr) << 16;
199 
200 		return true;
201 	}
202 
203 	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
204 	case IP_VERSION(11, 0, 2):
205 		/* VEGA20 and ARCTURUS */
206 		if (adev->asic_type == CHIP_VEGA20)
207 			control->i2c_address = EEPROM_I2C_MADDR_0;
208 		else if (strnstr(atom_ctx->vbios_pn,
209 				 "D342",
210 				 sizeof(atom_ctx->vbios_pn)))
211 			control->i2c_address = EEPROM_I2C_MADDR_0;
212 		else
213 			control->i2c_address = EEPROM_I2C_MADDR_4;
214 		return true;
215 	case IP_VERSION(11, 0, 7):
216 		control->i2c_address = EEPROM_I2C_MADDR_0;
217 		return true;
218 	case IP_VERSION(13, 0, 2):
219 		if (strnstr(atom_ctx->vbios_pn, "D673",
220 			    sizeof(atom_ctx->vbios_pn)))
221 			control->i2c_address = EEPROM_I2C_MADDR_4;
222 		else
223 			control->i2c_address = EEPROM_I2C_MADDR_0;
224 		return true;
225 	case IP_VERSION(13, 0, 0):
226 		if (strnstr(atom_ctx->vbios_pn, "D707",
227 			    sizeof(atom_ctx->vbios_pn)))
228 			control->i2c_address = EEPROM_I2C_MADDR_0;
229 		else
230 			control->i2c_address = EEPROM_I2C_MADDR_4;
231 		return true;
232 	case IP_VERSION(13, 0, 6):
233 	case IP_VERSION(13, 0, 10):
234 	case IP_VERSION(13, 0, 12):
235 	case IP_VERSION(13, 0, 14):
236 		control->i2c_address = EEPROM_I2C_MADDR_4;
237 		return true;
238 	default:
239 		return false;
240 	}
241 }
242 
243 static void
244 __encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header *hdr,
245 			     unsigned char *buf)
246 {
247 	u32 *pp = (uint32_t *)buf;
248 
249 	pp[0] = cpu_to_le32(hdr->header);
250 	pp[1] = cpu_to_le32(hdr->version);
251 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
252 	pp[3] = cpu_to_le32(hdr->tbl_size);
253 	pp[4] = cpu_to_le32(hdr->checksum);
254 }
255 
256 static void
257 __decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header *hdr,
258 			       unsigned char *buf)
259 {
260 	u32 *pp = (uint32_t *)buf;
261 
262 	hdr->header	      = le32_to_cpu(pp[0]);
263 	hdr->version	      = le32_to_cpu(pp[1]);
264 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
265 	hdr->tbl_size	      = le32_to_cpu(pp[3]);
266 	hdr->checksum	      = le32_to_cpu(pp[4]);
267 }
268 
269 static int __write_table_header(struct amdgpu_ras_eeprom_control *control)
270 {
271 	u8 buf[RAS_TABLE_HEADER_SIZE];
272 	struct amdgpu_device *adev = to_amdgpu_device(control);
273 	int res;
274 
275 	memset(buf, 0, sizeof(buf));
276 	__encode_table_header_to_buf(&control->tbl_hdr, buf);
277 
278 	/* i2c may be unstable in gpu reset */
279 	down_read(&adev->reset_domain->sem);
280 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
281 				  control->i2c_address +
282 				  control->ras_header_offset,
283 				  buf, RAS_TABLE_HEADER_SIZE);
284 	up_read(&adev->reset_domain->sem);
285 
286 	if (res < 0) {
287 		dev_err(adev->dev, "Failed to write EEPROM table header:%d",
288 			res);
289 	} else if (res < RAS_TABLE_HEADER_SIZE) {
290 		dev_err(adev->dev, "Short write:%d out of %d\n", res,
291 			RAS_TABLE_HEADER_SIZE);
292 		res = -EIO;
293 	} else {
294 		res = 0;
295 	}
296 
297 	return res;
298 }
299 
300 static void
301 __encode_table_ras_info_to_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
302 			       unsigned char *buf)
303 {
304 	u32 *pp = (uint32_t *)buf;
305 	u32 tmp;
306 
307 	tmp = ((uint32_t)(rai->rma_status) & 0xFF) |
308 	      (((uint32_t)(rai->health_percent) << 8) & 0xFF00) |
309 	      (((uint32_t)(rai->ecc_page_threshold) << 16) & 0xFFFF0000);
310 	pp[0] = cpu_to_le32(tmp);
311 }
312 
313 static void
314 __decode_table_ras_info_from_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
315 				 unsigned char *buf)
316 {
317 	u32 *pp = (uint32_t *)buf;
318 	u32 tmp;
319 
320 	tmp = le32_to_cpu(pp[0]);
321 	rai->rma_status = tmp & 0xFF;
322 	rai->health_percent = (tmp >> 8) & 0xFF;
323 	rai->ecc_page_threshold = (tmp >> 16) & 0xFFFF;
324 }
325 
326 static int __write_table_ras_info(struct amdgpu_ras_eeprom_control *control)
327 {
328 	struct amdgpu_device *adev = to_amdgpu_device(control);
329 	u8 *buf;
330 	int res;
331 
332 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
333 	if (!buf) {
334 		dev_err(adev->dev,
335 			"Failed to alloc buf to write table ras info\n");
336 		return -ENOMEM;
337 	}
338 
339 	__encode_table_ras_info_to_buf(&control->tbl_rai, buf);
340 
341 	/* i2c may be unstable in gpu reset */
342 	down_read(&adev->reset_domain->sem);
343 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
344 				  control->i2c_address +
345 				  control->ras_info_offset,
346 				  buf, RAS_TABLE_V2_1_INFO_SIZE);
347 	up_read(&adev->reset_domain->sem);
348 
349 	if (res < 0) {
350 		dev_err(adev->dev, "Failed to write EEPROM table ras info:%d",
351 			res);
352 	} else if (res < RAS_TABLE_V2_1_INFO_SIZE) {
353 		dev_err(adev->dev, "Short write:%d out of %d\n", res,
354 			RAS_TABLE_V2_1_INFO_SIZE);
355 		res = -EIO;
356 	} else {
357 		res = 0;
358 	}
359 
360 	kfree(buf);
361 
362 	return res;
363 }
364 
365 static u8 __calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control *control)
366 {
367 	int ii;
368 	u8  *pp, csum;
369 	size_t sz;
370 
371 	/* Header checksum, skip checksum field in the calculation */
372 	sz = sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum);
373 	pp = (u8 *) &control->tbl_hdr;
374 	csum = 0;
375 	for (ii = 0; ii < sz; ii++, pp++)
376 		csum += *pp;
377 
378 	return csum;
379 }
380 
381 static u8 __calc_ras_info_byte_sum(const struct amdgpu_ras_eeprom_control *control)
382 {
383 	int ii;
384 	u8  *pp, csum;
385 	size_t sz;
386 
387 	sz = sizeof(control->tbl_rai);
388 	pp = (u8 *) &control->tbl_rai;
389 	csum = 0;
390 	for (ii = 0; ii < sz; ii++, pp++)
391 		csum += *pp;
392 
393 	return csum;
394 }
395 
396 static int amdgpu_ras_eeprom_correct_header_tag(
397 	struct amdgpu_ras_eeprom_control *control,
398 	uint32_t header)
399 {
400 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
401 	u8 *hh;
402 	int res;
403 	u8 csum;
404 
405 	csum = -hdr->checksum;
406 
407 	hh = (void *) &hdr->header;
408 	csum -= (hh[0] + hh[1] + hh[2] + hh[3]);
409 	hh = (void *) &header;
410 	csum += hh[0] + hh[1] + hh[2] + hh[3];
411 	csum = -csum;
412 	mutex_lock(&control->ras_tbl_mutex);
413 	hdr->header = header;
414 	hdr->checksum = csum;
415 	res = __write_table_header(control);
416 	mutex_unlock(&control->ras_tbl_mutex);
417 
418 	return res;
419 }
420 
421 static void amdgpu_ras_set_eeprom_table_version(struct amdgpu_ras_eeprom_control *control)
422 {
423 	struct amdgpu_device *adev = to_amdgpu_device(control);
424 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
425 
426 	switch (amdgpu_ip_version(adev, UMC_HWIP, 0)) {
427 	case IP_VERSION(8, 10, 0):
428 		hdr->version = RAS_TABLE_VER_V2_1;
429 		return;
430 	case IP_VERSION(12, 0, 0):
431 	case IP_VERSION(12, 5, 0):
432 		hdr->version = RAS_TABLE_VER_V3;
433 		return;
434 	default:
435 		hdr->version = RAS_TABLE_VER_V1;
436 		return;
437 	}
438 }
439 
440 /**
441  * amdgpu_ras_eeprom_reset_table -- Reset the RAS EEPROM table
442  * @control: pointer to control structure
443  *
444  * Reset the contents of the header of the RAS EEPROM table.
445  * Return 0 on success, -errno on error.
446  */
447 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
448 {
449 	struct amdgpu_device *adev = to_amdgpu_device(control);
450 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
451 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
452 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
453 	u8 csum;
454 	int res;
455 
456 	mutex_lock(&control->ras_tbl_mutex);
457 
458 	hdr->header = RAS_TABLE_HDR_VAL;
459 	amdgpu_ras_set_eeprom_table_version(control);
460 
461 	if (hdr->version >= RAS_TABLE_VER_V2_1) {
462 		hdr->first_rec_offset = RAS_RECORD_START_V2_1;
463 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE +
464 				RAS_TABLE_V2_1_INFO_SIZE;
465 		rai->rma_status = GPU_HEALTH_USABLE;
466 
467 		control->ras_record_offset = RAS_RECORD_START_V2_1;
468 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT_V2_1;
469 		/**
470 		 * GPU health represented as a percentage.
471 		 * 0 means worst health, 100 means fully health.
472 		 */
473 		rai->health_percent = 100;
474 		/* ecc_page_threshold = 0 means disable bad page retirement */
475 		rai->ecc_page_threshold = con->bad_page_cnt_threshold;
476 	} else {
477 		hdr->first_rec_offset = RAS_RECORD_START;
478 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE;
479 
480 		control->ras_record_offset = RAS_RECORD_START;
481 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT;
482 	}
483 
484 	csum = __calc_hdr_byte_sum(control);
485 	if (hdr->version >= RAS_TABLE_VER_V2_1)
486 		csum += __calc_ras_info_byte_sum(control);
487 	csum = -csum;
488 	hdr->checksum = csum;
489 	res = __write_table_header(control);
490 	if (!res && hdr->version > RAS_TABLE_VER_V1)
491 		res = __write_table_ras_info(control);
492 
493 	control->ras_num_recs = 0;
494 	control->ras_num_bad_pages = 0;
495 	control->ras_num_mca_recs = 0;
496 	control->ras_num_pa_recs = 0;
497 	control->ras_fri = 0;
498 
499 	amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_bad_pages);
500 
501 	control->bad_channel_bitmap = 0;
502 	amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap);
503 	con->update_channel_flag = false;
504 	/* there is no record on eeprom now, clear the counter */
505 	if (con->eh_data)
506 		con->eh_data->count_saved = 0;
507 
508 	amdgpu_ras_debugfs_set_ret_size(control);
509 
510 	mutex_unlock(&control->ras_tbl_mutex);
511 
512 	return res;
513 }
514 
515 static void
516 __encode_table_record_to_buf(struct amdgpu_ras_eeprom_control *control,
517 			     struct eeprom_table_record *record,
518 			     unsigned char *buf)
519 {
520 	__le64 tmp = 0;
521 	int i = 0;
522 
523 	/* Next are all record fields according to EEPROM page spec in LE foramt */
524 	buf[i++] = record->err_type;
525 
526 	buf[i++] = record->bank;
527 
528 	tmp = cpu_to_le64(record->ts);
529 	memcpy(buf + i, &tmp, 8);
530 	i += 8;
531 
532 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
533 	memcpy(buf + i, &tmp, 6);
534 	i += 6;
535 
536 	buf[i++] = record->mem_channel;
537 	buf[i++] = record->mcumc_id;
538 
539 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
540 	memcpy(buf + i, &tmp, 6);
541 }
542 
543 static void
544 __decode_table_record_from_buf(struct amdgpu_ras_eeprom_control *control,
545 			       struct eeprom_table_record *record,
546 			       unsigned char *buf)
547 {
548 	__le64 tmp = 0;
549 	int i =  0;
550 
551 	/* Next are all record fields according to EEPROM page spec in LE foramt */
552 	record->err_type = buf[i++];
553 
554 	record->bank = buf[i++];
555 
556 	memcpy(&tmp, buf + i, 8);
557 	record->ts = le64_to_cpu(tmp);
558 	i += 8;
559 
560 	memcpy(&tmp, buf + i, 6);
561 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
562 	i += 6;
563 
564 	record->mem_channel = buf[i++];
565 	record->mcumc_id = buf[i++];
566 
567 	memcpy(&tmp, buf + i,  6);
568 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
569 }
570 
571 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
572 {
573 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
574 
575 	if (amdgpu_uniras_enabled(adev))
576 		return amdgpu_ras_mgr_check_eeprom_safety_watermark(adev);
577 
578 	if (!__is_ras_eeprom_supported(adev) ||
579 	    !amdgpu_bad_page_threshold)
580 		return false;
581 
582 	/* skip check eeprom table for VEGA20 Gaming */
583 	if (!con)
584 		return false;
585 	else
586 		if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
587 			return false;
588 
589 	if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
590 		if (con->eeprom_control.ras_num_bad_pages > con->bad_page_cnt_threshold)
591 			dev_warn(adev->dev, "RAS records:%d exceed threshold:%d",
592 				 con->eeprom_control.ras_num_bad_pages, con->bad_page_cnt_threshold);
593 		if ((amdgpu_bad_page_threshold == -1) ||
594 		    (amdgpu_bad_page_threshold == -2)) {
595 			dev_warn(adev->dev,
596 				 "Please consult AMD Service Action Guide (SAG) for appropriate service procedures.\n");
597 			return false;
598 		} else {
599 			dev_warn(adev->dev,
600 				 "Please consider adjusting the customized threshold.\n");
601 			return true;
602 		}
603 	}
604 
605 	return false;
606 }
607 
608 /**
609  * __amdgpu_ras_eeprom_write -- write indexed from buffer to EEPROM
610  * @control: pointer to control structure
611  * @buf: pointer to buffer containing data to write
612  * @fri: start writing at this index
613  * @num: number of records to write
614  *
615  * The caller must hold the table mutex in @control.
616  * Return 0 on success, -errno otherwise.
617  */
618 static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control,
619 				     u8 *buf, const u32 fri, const u32 num)
620 {
621 	struct amdgpu_device *adev = to_amdgpu_device(control);
622 	u32 buf_size;
623 	int res;
624 
625 	/* i2c may be unstable in gpu reset */
626 	down_read(&adev->reset_domain->sem);
627 	buf_size = num * RAS_TABLE_RECORD_SIZE;
628 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
629 				  control->i2c_address +
630 				  RAS_INDEX_TO_OFFSET(control, fri),
631 				  buf, buf_size);
632 	up_read(&adev->reset_domain->sem);
633 	if (res < 0) {
634 		dev_err(adev->dev, "Writing %d EEPROM table records error:%d",
635 			num, res);
636 	} else if (res < buf_size) {
637 		/* Short write, return error.
638 		 */
639 		dev_err(adev->dev, "Wrote %d records out of %d",
640 			res / RAS_TABLE_RECORD_SIZE, num);
641 		res = -EIO;
642 	} else {
643 		res = 0;
644 	}
645 
646 	return res;
647 }
648 
649 static int
650 amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
651 			       struct eeprom_table_record *record,
652 			       const u32 num)
653 {
654 	struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
655 	u32 a, b, i;
656 	u8 *buf, *pp;
657 	int res;
658 
659 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
660 	if (!buf)
661 		return -ENOMEM;
662 
663 	/* Encode all of them in one go.
664 	 */
665 	pp = buf;
666 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
667 		__encode_table_record_to_buf(control, &record[i], pp);
668 
669 		/* update bad channel bitmap */
670 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
671 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
672 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
673 			con->update_channel_flag = true;
674 		}
675 	}
676 
677 	/* a, first record index to write into.
678 	 * b, last record index to write into.
679 	 * a = first index to read (fri) + number of records in the table,
680 	 * b = a + @num - 1.
681 	 * Let N = control->ras_max_num_record_count, then we have,
682 	 * case 0: 0 <= a <= b < N,
683 	 *   just append @num records starting at a;
684 	 * case 1: 0 <= a < N <= b,
685 	 *   append (N - a) records starting at a, and
686 	 *   append the remainder,  b % N + 1, starting at 0.
687 	 * case 2: 0 <= fri < N <= a <= b, then modulo N we get two subcases,
688 	 * case 2a: 0 <= a <= b < N
689 	 *   append num records starting at a; and fix fri if b overwrote it,
690 	 *   and since a <= b, if b overwrote it then a must've also,
691 	 *   and if b didn't overwrite it, then a didn't also.
692 	 * case 2b: 0 <= b < a < N
693 	 *   write num records starting at a, which wraps around 0=N
694 	 *   and overwrite fri unconditionally. Now from case 2a,
695 	 *   this means that b eclipsed fri to overwrite it and wrap
696 	 *   around 0 again, i.e. b = 2N+r pre modulo N, so we unconditionally
697 	 *   set fri = b + 1 (mod N).
698 	 * Now, since fri is updated in every case, except the trivial case 0,
699 	 * the number of records present in the table after writing, is,
700 	 * num_recs - 1 = b - fri (mod N), and we take the positive value,
701 	 * by adding an arbitrary multiple of N before taking the modulo N
702 	 * as shown below.
703 	 */
704 	a = control->ras_fri + control->ras_num_recs;
705 	b = a + num  - 1;
706 	if (b < control->ras_max_record_count) {
707 		res = __amdgpu_ras_eeprom_write(control, buf, a, num);
708 	} else if (a < control->ras_max_record_count) {
709 		u32 g0, g1;
710 
711 		g0 = control->ras_max_record_count - a;
712 		g1 = b % control->ras_max_record_count + 1;
713 		res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
714 		if (res)
715 			goto Out;
716 		res = __amdgpu_ras_eeprom_write(control,
717 						buf + g0 * RAS_TABLE_RECORD_SIZE,
718 						0, g1);
719 		if (res)
720 			goto Out;
721 		if (g1 > control->ras_fri)
722 			control->ras_fri = g1 % control->ras_max_record_count;
723 	} else {
724 		a %= control->ras_max_record_count;
725 		b %= control->ras_max_record_count;
726 
727 		if (a <= b) {
728 			/* Note that, b - a + 1 = num. */
729 			res = __amdgpu_ras_eeprom_write(control, buf, a, num);
730 			if (res)
731 				goto Out;
732 			if (b >= control->ras_fri)
733 				control->ras_fri = (b + 1) % control->ras_max_record_count;
734 		} else {
735 			u32 g0, g1;
736 
737 			/* b < a, which means, we write from
738 			 * a to the end of the table, and from
739 			 * the start of the table to b.
740 			 */
741 			g0 = control->ras_max_record_count - a;
742 			g1 = b + 1;
743 			res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
744 			if (res)
745 				goto Out;
746 			res = __amdgpu_ras_eeprom_write(control,
747 							buf + g0 * RAS_TABLE_RECORD_SIZE,
748 							0, g1);
749 			if (res)
750 				goto Out;
751 			control->ras_fri = g1 % control->ras_max_record_count;
752 		}
753 	}
754 	control->ras_num_recs = 1 + (control->ras_max_record_count + b
755 				     - control->ras_fri)
756 		% control->ras_max_record_count;
757 
758 	/*old asics only save pa to eeprom like before*/
759 	control->ras_num_pa_recs += num;
760 
761 	control->ras_num_bad_pages = con->bad_page_num;
762 Out:
763 	kfree(buf);
764 	return res;
765 }
766 
767 static int
768 amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
769 {
770 	struct amdgpu_device *adev = to_amdgpu_device(control);
771 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
772 	u8 *buf, *pp, csum;
773 	u32 buf_size;
774 	int res;
775 
776 	/* Modify the header if it exceeds.
777 	 */
778 	if (amdgpu_bad_page_threshold != 0 &&
779 	    control->ras_num_bad_pages > ras->bad_page_cnt_threshold) {
780 		dev_warn(adev->dev,
781 			"Saved bad pages %d reaches threshold value %d\n",
782 			control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
783 
784 		if (adev->cper.enabled && !amdgpu_uniras_enabled(adev) &&
785 		    amdgpu_cper_generate_bp_threshold_record(adev))
786 			dev_warn(adev->dev, "fail to generate bad page threshold cper records\n");
787 
788 		if ((amdgpu_bad_page_threshold != -1) &&
789 		    (amdgpu_bad_page_threshold != -2)) {
790 			control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
791 			if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1) {
792 				control->tbl_rai.rma_status = GPU_RETIRED__ECC_REACH_THRESHOLD;
793 				control->tbl_rai.health_percent = 0;
794 			}
795 			ras->is_rma = true;
796 		}
797 
798 		/* ignore the -ENOTSUPP return value */
799 		amdgpu_dpm_send_rma_reason(adev);
800 	}
801 
802 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
803 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
804 					    RAS_TABLE_V2_1_INFO_SIZE +
805 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
806 	else
807 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
808 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
809 	control->tbl_hdr.checksum = 0;
810 
811 	buf_size = control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
812 	buf = kcalloc(control->ras_num_recs, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
813 	if (!buf) {
814 		dev_err(adev->dev,
815 			"allocating memory for table of size %d bytes failed\n",
816 			control->tbl_hdr.tbl_size);
817 		res = -ENOMEM;
818 		goto Out;
819 	}
820 
821 	down_read(&adev->reset_domain->sem);
822 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
823 				 control->i2c_address +
824 				 control->ras_record_offset,
825 				 buf, buf_size);
826 	up_read(&adev->reset_domain->sem);
827 	if (res < 0) {
828 		dev_err(adev->dev, "EEPROM failed reading records:%d\n", res);
829 		goto Out;
830 	} else if (res < buf_size) {
831 		dev_err(adev->dev, "EEPROM read %d out of %d bytes\n", res,
832 			buf_size);
833 		res = -EIO;
834 		goto Out;
835 	}
836 
837 	/**
838 	 * bad page records have been stored in eeprom,
839 	 * now calculate gpu health percent
840 	 */
841 	if (amdgpu_bad_page_threshold != 0 &&
842 	    control->tbl_hdr.version >= RAS_TABLE_VER_V2_1 &&
843 	    control->ras_num_bad_pages <= ras->bad_page_cnt_threshold)
844 		control->tbl_rai.health_percent = ((ras->bad_page_cnt_threshold -
845 						   control->ras_num_bad_pages) * 100) /
846 						   ras->bad_page_cnt_threshold;
847 
848 	/* Recalc the checksum.
849 	 */
850 	csum = 0;
851 	for (pp = buf; pp < buf + buf_size; pp++)
852 		csum += *pp;
853 
854 	csum += __calc_hdr_byte_sum(control);
855 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
856 		csum += __calc_ras_info_byte_sum(control);
857 	/* avoid sign extension when assigning to "checksum" */
858 	csum = -csum;
859 	control->tbl_hdr.checksum = csum;
860 	res = __write_table_header(control);
861 	if (!res && control->tbl_hdr.version > RAS_TABLE_VER_V1)
862 		res = __write_table_ras_info(control);
863 Out:
864 	kfree(buf);
865 	return res;
866 }
867 
868 /**
869  * amdgpu_ras_eeprom_append -- append records to the EEPROM RAS table
870  * @control: pointer to control structure
871  * @record: array of records to append
872  * @num: number of records in @record array
873  *
874  * Append @num records to the table, calculate the checksum and write
875  * the table back to EEPROM. The maximum number of records that
876  * can be appended is between 1 and control->ras_max_record_count,
877  * regardless of how many records are already stored in the table.
878  *
879  * Return 0 on success or if EEPROM is not supported, -errno on error.
880  */
881 int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
882 			     struct eeprom_table_record *record,
883 			     const u32 num)
884 {
885 	struct amdgpu_device *adev = to_amdgpu_device(control);
886 	int res, i;
887 	uint64_t nps = AMDGPU_NPS1_PARTITION_MODE;
888 
889 	if (!__is_ras_eeprom_supported(adev))
890 		return 0;
891 
892 	if (num == 0) {
893 		dev_err(adev->dev, "will not append 0 records\n");
894 		return -EINVAL;
895 	} else if (num > control->ras_max_record_count) {
896 		dev_err(adev->dev,
897 			"cannot append %d records than the size of table %d\n",
898 			num, control->ras_max_record_count);
899 		return -EINVAL;
900 	}
901 
902 	if (adev->gmc.gmc_funcs->query_mem_partition_mode)
903 		nps = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
904 
905 	/* set the new channel index flag */
906 	for (i = 0; i < num; i++)
907 		record[i].retired_page |= (nps << UMC_NPS_SHIFT);
908 
909 	mutex_lock(&control->ras_tbl_mutex);
910 
911 	res = amdgpu_ras_eeprom_append_table(control, record, num);
912 	if (!res)
913 		res = amdgpu_ras_eeprom_update_header(control);
914 	if (!res)
915 		amdgpu_ras_debugfs_set_ret_size(control);
916 
917 	mutex_unlock(&control->ras_tbl_mutex);
918 
919 	/* clear channel index flag, the flag is only saved on eeprom */
920 	for (i = 0; i < num; i++)
921 		record[i].retired_page &= ~(nps << UMC_NPS_SHIFT);
922 
923 	return res;
924 }
925 
926 /**
927  * __amdgpu_ras_eeprom_read -- read indexed from EEPROM into buffer
928  * @control: pointer to control structure
929  * @buf: pointer to buffer to read into
930  * @fri: first record index, start reading at this index, absolute index
931  * @num: number of records to read
932  *
933  * The caller must hold the table mutex in @control.
934  * Return 0 on success, -errno otherwise.
935  */
936 static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
937 				    u8 *buf, const u32 fri, const u32 num)
938 {
939 	struct amdgpu_device *adev = to_amdgpu_device(control);
940 	u32 buf_size;
941 	int res;
942 
943 	/* i2c may be unstable in gpu reset */
944 	down_read(&adev->reset_domain->sem);
945 	buf_size = num * RAS_TABLE_RECORD_SIZE;
946 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
947 				 control->i2c_address +
948 				 RAS_INDEX_TO_OFFSET(control, fri),
949 				 buf, buf_size);
950 	up_read(&adev->reset_domain->sem);
951 	if (res < 0) {
952 		dev_err(adev->dev, "Reading %d EEPROM table records error:%d",
953 			num, res);
954 	} else if (res < buf_size) {
955 		/* Short read, return error.
956 		 */
957 		dev_err(adev->dev, "Read %d records out of %d",
958 			res / RAS_TABLE_RECORD_SIZE, num);
959 		res = -EIO;
960 	} else {
961 		res = 0;
962 	}
963 
964 	return res;
965 }
966 
967 /**
968  * amdgpu_ras_eeprom_read -- read EEPROM
969  * @control: pointer to control structure
970  * @record: array of records to read into
971  * @num: number of records in @record
972  *
973  * Reads num records from the RAS table in EEPROM and
974  * writes the data into @record array.
975  *
976  * Returns 0 on success, -errno on error.
977  */
978 int amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
979 			   struct eeprom_table_record *record,
980 			   const u32 num)
981 {
982 	struct amdgpu_device *adev = to_amdgpu_device(control);
983 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
984 	int i, res;
985 	u8 *buf, *pp;
986 	u32 g0, g1;
987 
988 	if (!__is_ras_eeprom_supported(adev))
989 		return 0;
990 
991 	if (num == 0) {
992 		dev_err(adev->dev, "will not read 0 records\n");
993 		return -EINVAL;
994 	} else if (num > control->ras_num_recs) {
995 		dev_err(adev->dev, "too many records to read:%d available:%d\n",
996 			num, control->ras_num_recs);
997 		return -EINVAL;
998 	}
999 
1000 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
1001 	if (!buf)
1002 		return -ENOMEM;
1003 
1004 	/* Determine how many records to read, from the first record
1005 	 * index, fri, to the end of the table, and from the beginning
1006 	 * of the table, such that the total number of records is
1007 	 * @num, and we handle wrap around when fri > 0 and
1008 	 * fri + num > RAS_MAX_RECORD_COUNT.
1009 	 *
1010 	 * First we compute the index of the last element
1011 	 * which would be fetched from each region,
1012 	 * g0 is in [fri, fri + num - 1], and
1013 	 * g1 is in [0, RAS_MAX_RECORD_COUNT - 1].
1014 	 * Then, if g0 < RAS_MAX_RECORD_COUNT, the index of
1015 	 * the last element to fetch, we set g0 to _the number_
1016 	 * of elements to fetch, @num, since we know that the last
1017 	 * indexed to be fetched does not exceed the table.
1018 	 *
1019 	 * If, however, g0 >= RAS_MAX_RECORD_COUNT, then
1020 	 * we set g0 to the number of elements to read
1021 	 * until the end of the table, and g1 to the number of
1022 	 * elements to read from the beginning of the table.
1023 	 */
1024 	g0 = control->ras_fri + num - 1;
1025 	g1 = g0 % control->ras_max_record_count;
1026 	if (g0 < control->ras_max_record_count) {
1027 		g0 = num;
1028 		g1 = 0;
1029 	} else {
1030 		g0 = control->ras_max_record_count - control->ras_fri;
1031 		g1 += 1;
1032 	}
1033 
1034 	mutex_lock(&control->ras_tbl_mutex);
1035 	res = __amdgpu_ras_eeprom_read(control, buf, control->ras_fri, g0);
1036 	if (res)
1037 		goto Out;
1038 	if (g1) {
1039 		res = __amdgpu_ras_eeprom_read(control,
1040 					       buf + g0 * RAS_TABLE_RECORD_SIZE,
1041 					       0, g1);
1042 		if (res)
1043 			goto Out;
1044 	}
1045 
1046 	res = 0;
1047 
1048 	/* Read up everything? Then transform.
1049 	 */
1050 	pp = buf;
1051 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
1052 		__decode_table_record_from_buf(control, &record[i], pp);
1053 
1054 		/* update bad channel bitmap */
1055 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
1056 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
1057 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
1058 			con->update_channel_flag = true;
1059 		}
1060 	}
1061 Out:
1062 	kfree(buf);
1063 	mutex_unlock(&control->ras_tbl_mutex);
1064 
1065 	return res;
1066 }
1067 
1068 uint32_t amdgpu_ras_eeprom_max_record_count(struct amdgpu_ras_eeprom_control *control)
1069 {
1070 	/* get available eeprom table version first before eeprom table init */
1071 	amdgpu_ras_set_eeprom_table_version(control);
1072 
1073 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
1074 		return RAS_MAX_RECORD_COUNT_V2_1;
1075 	else
1076 		return RAS_MAX_RECORD_COUNT;
1077 }
1078 
1079 static ssize_t
1080 amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
1081 				    size_t size, loff_t *pos)
1082 {
1083 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1084 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1085 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1086 	u8 data[50];
1087 	int res;
1088 
1089 	if (!size)
1090 		return size;
1091 
1092 	if (!ras || !control) {
1093 		res = snprintf(data, sizeof(data), "Not supported\n");
1094 	} else {
1095 		res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
1096 			       RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
1097 	}
1098 
1099 	if (*pos >= res)
1100 		return 0;
1101 
1102 	res -= *pos;
1103 	res = min_t(size_t, res, size);
1104 
1105 	if (copy_to_user(buf, &data[*pos], res))
1106 		return -EFAULT;
1107 
1108 	*pos += res;
1109 
1110 	return res;
1111 }
1112 
1113 const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
1114 	.owner = THIS_MODULE,
1115 	.read = amdgpu_ras_debugfs_eeprom_size_read,
1116 	.write = NULL,
1117 	.llseek = default_llseek,
1118 };
1119 
1120 static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
1121 static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
1122 #define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
1123 static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
1124 static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
1125 #define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
1126 
1127 static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
1128 	"ignore",
1129 	"re",
1130 	"ue",
1131 };
1132 
1133 static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
1134 {
1135 	return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
1136 		strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
1137 }
1138 
1139 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
1140 {
1141 	struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
1142 					      eeprom_control);
1143 	struct dentry *de = ras->de_ras_eeprom_table;
1144 
1145 	if (de)
1146 		d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
1147 }
1148 
1149 static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
1150 					     size_t size, loff_t *pos)
1151 {
1152 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1153 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1154 	struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
1155 	const size_t orig_size = size;
1156 	int res = -EFAULT;
1157 	size_t data_len;
1158 
1159 	mutex_lock(&control->ras_tbl_mutex);
1160 
1161 	/* We want *pos - data_len > 0, which means there's
1162 	 * bytes to be printed from data.
1163 	 */
1164 	data_len = strlen(tbl_hdr_str);
1165 	if (*pos < data_len) {
1166 		data_len -= *pos;
1167 		data_len = min_t(size_t, data_len, size);
1168 		if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
1169 			goto Out;
1170 		buf += data_len;
1171 		size -= data_len;
1172 		*pos += data_len;
1173 	}
1174 
1175 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
1176 	if (*pos < data_len && size > 0) {
1177 		u8 data[tbl_hdr_fmt_size + 1];
1178 		loff_t lpos;
1179 
1180 		snprintf(data, sizeof(data), tbl_hdr_fmt,
1181 			 control->tbl_hdr.header,
1182 			 control->tbl_hdr.version,
1183 			 control->tbl_hdr.first_rec_offset,
1184 			 control->tbl_hdr.tbl_size,
1185 			 control->tbl_hdr.checksum);
1186 
1187 		data_len -= *pos;
1188 		data_len = min_t(size_t, data_len, size);
1189 		lpos = *pos - strlen(tbl_hdr_str);
1190 		if (copy_to_user(buf, &data[lpos], data_len))
1191 			goto Out;
1192 		buf += data_len;
1193 		size -= data_len;
1194 		*pos += data_len;
1195 	}
1196 
1197 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
1198 	if (*pos < data_len && size > 0) {
1199 		loff_t lpos;
1200 
1201 		data_len -= *pos;
1202 		data_len = min_t(size_t, data_len, size);
1203 		lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
1204 		if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
1205 			goto Out;
1206 		buf += data_len;
1207 		size -= data_len;
1208 		*pos += data_len;
1209 	}
1210 
1211 	data_len = amdgpu_ras_debugfs_table_size(control);
1212 	if (*pos < data_len && size > 0) {
1213 		u8 dare[RAS_TABLE_RECORD_SIZE];
1214 		u8 data[rec_hdr_fmt_size + 1];
1215 		struct eeprom_table_record record;
1216 		int s, r;
1217 
1218 		/* Find the starting record index
1219 		 */
1220 		s = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1221 			strlen(rec_hdr_str);
1222 		s = s / rec_hdr_fmt_size;
1223 		r = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1224 			strlen(rec_hdr_str);
1225 		r = r % rec_hdr_fmt_size;
1226 
1227 		for ( ; size > 0 && s < control->ras_num_recs; s++) {
1228 			u32 ai = RAS_RI_TO_AI(control, s);
1229 			/* Read a single record
1230 			 */
1231 			res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
1232 			if (res)
1233 				goto Out;
1234 			__decode_table_record_from_buf(control, &record, dare);
1235 			snprintf(data, sizeof(data), rec_hdr_fmt,
1236 				 s,
1237 				 RAS_INDEX_TO_OFFSET(control, ai),
1238 				 record_err_type_str[record.err_type],
1239 				 record.bank,
1240 				 record.ts,
1241 				 record.offset,
1242 				 record.mem_channel,
1243 				 record.mcumc_id,
1244 				 record.retired_page);
1245 
1246 			data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
1247 			if (copy_to_user(buf, &data[r], data_len)) {
1248 				res = -EFAULT;
1249 				goto Out;
1250 			}
1251 			buf += data_len;
1252 			size -= data_len;
1253 			*pos += data_len;
1254 			r = 0;
1255 		}
1256 	}
1257 	res = 0;
1258 Out:
1259 	mutex_unlock(&control->ras_tbl_mutex);
1260 	return res < 0 ? res : orig_size - size;
1261 }
1262 
1263 static ssize_t
1264 amdgpu_ras_debugfs_table_read_uniras(struct amdgpu_device *adev,
1265 				     char __user *buf,
1266 				     size_t size, loff_t *pos)
1267 {
1268 	struct amdgpu_ras_mgr *ras_mgr = amdgpu_ras_mgr_get_context(adev);
1269 	struct ras_core_context *ras_core = ras_mgr ? ras_mgr->ras_core : NULL;
1270 	struct eeprom_umc_record *records = NULL;
1271 	struct ras_eeprom_control *control;
1272 	size_t bufsz, len = 0;
1273 	u32 num_recs;
1274 	char *kbuf;
1275 	ssize_t res;
1276 	int i;
1277 
1278 	if (!ras_core)
1279 		return 0;
1280 
1281 	/* pmfw manages eeprom data by itself */
1282 	if (ras_fw_eeprom_supported(ras_core))
1283 		return 0;
1284 
1285 	control = &ras_core->ras_eeprom;
1286 	num_recs = ras_eeprom_get_record_count(ras_core);
1287 
1288 	bufsz = strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
1289 		strlen(rec_hdr_str) + (size_t)rec_hdr_fmt_size * num_recs + 1;
1290 
1291 	kbuf = kvmalloc(bufsz, GFP_KERNEL);
1292 	if (!kbuf)
1293 		return -ENOMEM;
1294 
1295 	if (num_recs) {
1296 		records = kvcalloc(num_recs, sizeof(*records), GFP_KERNEL);
1297 		if (!records) {
1298 			res = -ENOMEM;
1299 			goto out;
1300 		}
1301 
1302 		res = ras_eeprom_read(ras_core, records, num_recs);
1303 		if (res)
1304 			goto out;
1305 	}
1306 
1307 	len += scnprintf(kbuf + len, bufsz - len, "%s", tbl_hdr_str);
1308 	len += scnprintf(kbuf + len, bufsz - len, tbl_hdr_fmt,
1309 				 control->tbl_hdr.header,
1310 				 control->tbl_hdr.version,
1311 				 control->tbl_hdr.first_rec_offset,
1312 				 control->tbl_hdr.tbl_size,
1313 				 control->tbl_hdr.checksum);
1314 	len += scnprintf(kbuf + len, bufsz - len, "%s", rec_hdr_str);
1315 
1316 	for (i = 0; i < num_recs; i++) {
1317 		u32 ai = RAS_RI_TO_AI(control, i);
1318 		int et = records[i].err_type;
1319 		const char *ets = (et >= 0 && et < AMDGPU_RAS_EEPROM_ERR_COUNT) ?
1320 				  record_err_type_str[et] : "na";
1321 
1322 		len += scnprintf(kbuf + len, bufsz - len, rec_hdr_fmt,
1323 				 i,
1324 				 RAS_INDEX_TO_OFFSET(control, ai),
1325 				 ets,
1326 				 records[i].bank,
1327 				 records[i].ts,
1328 				 records[i].offset,
1329 				 records[i].mem_channel,
1330 				 records[i].mcumc_id,
1331 				 records[i].retired_row_pfn);
1332 	}
1333 
1334 	res = simple_read_from_buffer(buf, size, pos, kbuf, len);
1335 
1336 out:
1337 	kvfree(records);
1338 	kvfree(kbuf);
1339 
1340 	return res;
1341 }
1342 
1343 static ssize_t
1344 amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
1345 				     size_t size, loff_t *pos)
1346 {
1347 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1348 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1349 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1350 	u8 data[81];
1351 	int res;
1352 
1353 	if (!size)
1354 		return size;
1355 
1356 	if (amdgpu_uniras_enabled(adev))
1357 		return amdgpu_ras_debugfs_table_read_uniras(adev, buf,
1358 						    size, pos);
1359 
1360 	if (!ras || !control) {
1361 		res = snprintf(data, sizeof(data), "Not supported\n");
1362 		if (*pos >= res)
1363 			return 0;
1364 
1365 		res -= *pos;
1366 		res = min_t(size_t, res, size);
1367 
1368 		if (copy_to_user(buf, &data[*pos], res))
1369 			return -EFAULT;
1370 
1371 		*pos += res;
1372 
1373 		return res;
1374 	} else {
1375 		return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
1376 	}
1377 }
1378 
1379 const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
1380 	.owner = THIS_MODULE,
1381 	.read = amdgpu_ras_debugfs_eeprom_table_read,
1382 	.write = NULL,
1383 	.llseek = default_llseek,
1384 };
1385 
1386 /**
1387  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
1388  * @control: pointer to control structure
1389  *
1390  * Check the checksum of the stored in EEPROM RAS table.
1391  *
1392  * Return 0 if the checksum is correct,
1393  * positive if it is not correct, and
1394  * -errno on I/O error.
1395  */
1396 static int __verify_ras_table_checksum(struct amdgpu_ras_eeprom_control *control)
1397 {
1398 	struct amdgpu_device *adev = to_amdgpu_device(control);
1399 	int buf_size, res;
1400 	u8  csum, *buf, *pp;
1401 
1402 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
1403 		buf_size = RAS_TABLE_HEADER_SIZE +
1404 			   RAS_TABLE_V2_1_INFO_SIZE +
1405 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1406 	else
1407 		buf_size = RAS_TABLE_HEADER_SIZE +
1408 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1409 
1410 	buf = kzalloc(buf_size, GFP_KERNEL);
1411 	if (!buf) {
1412 		dev_err(adev->dev,
1413 			"Out of memory checking RAS table checksum.\n");
1414 		return -ENOMEM;
1415 	}
1416 
1417 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1418 				 control->i2c_address +
1419 				 control->ras_header_offset,
1420 				 buf, buf_size);
1421 	if (res < buf_size) {
1422 		dev_err(adev->dev, "Partial read for checksum, res:%d\n", res);
1423 		/* On partial reads, return -EIO.
1424 		 */
1425 		if (res >= 0)
1426 			res = -EIO;
1427 		goto Out;
1428 	}
1429 
1430 	csum = 0;
1431 	for (pp = buf; pp < buf + buf_size; pp++)
1432 		csum += *pp;
1433 Out:
1434 	kfree(buf);
1435 	return res < 0 ? res : csum;
1436 }
1437 
1438 static int __read_table_ras_info(struct amdgpu_ras_eeprom_control *control)
1439 {
1440 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
1441 	struct amdgpu_device *adev = to_amdgpu_device(control);
1442 	unsigned char *buf;
1443 	int res;
1444 
1445 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
1446 	if (!buf) {
1447 		dev_err(adev->dev,
1448 			"Failed to alloc buf to read EEPROM table ras info\n");
1449 		return -ENOMEM;
1450 	}
1451 
1452 	/**
1453 	 * EEPROM table V2_1 supports ras info,
1454 	 * read EEPROM table ras info
1455 	 */
1456 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1457 				 control->i2c_address + control->ras_info_offset,
1458 				 buf, RAS_TABLE_V2_1_INFO_SIZE);
1459 	if (res < RAS_TABLE_V2_1_INFO_SIZE) {
1460 		dev_err(adev->dev,
1461 			"Failed to read EEPROM table ras info, res:%d", res);
1462 		res = res >= 0 ? -EIO : res;
1463 		goto Out;
1464 	}
1465 
1466 	__decode_table_ras_info_from_buf(rai, buf);
1467 
1468 Out:
1469 	kfree(buf);
1470 	return res == RAS_TABLE_V2_1_INFO_SIZE ? 0 : res;
1471 }
1472 
1473 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control)
1474 {
1475 	struct amdgpu_device *adev = to_amdgpu_device(control);
1476 	unsigned char buf[RAS_TABLE_HEADER_SIZE] = { 0 };
1477 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1478 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1479 	int dev_var = adev->pdev->device & 0xF;
1480 	uint32_t vram_type = adev->gmc.vram_type;
1481 	int res;
1482 
1483 	ras->is_rma = false;
1484 
1485 	if (!__is_ras_eeprom_supported(adev))
1486 		return 0;
1487 
1488 	/* Verify i2c adapter is initialized */
1489 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1490 		return -ENOENT;
1491 
1492 	if (!__get_eeprom_i2c_addr(adev, control))
1493 		return -EINVAL;
1494 
1495 	control->ras_header_offset = RAS_HDR_START;
1496 	control->ras_info_offset = RAS_TABLE_V2_1_INFO_START;
1497 	mutex_init(&control->ras_tbl_mutex);
1498 
1499 	/* Read the table header from EEPROM address */
1500 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1501 				 control->i2c_address + control->ras_header_offset,
1502 				 buf, RAS_TABLE_HEADER_SIZE);
1503 	if (res < RAS_TABLE_HEADER_SIZE) {
1504 		dev_err(adev->dev, "Failed to read EEPROM table header, res:%d",
1505 			res);
1506 		return res >= 0 ? -EIO : res;
1507 	}
1508 
1509 	__decode_table_header_from_buf(hdr, buf);
1510 
1511 	if (hdr->header != RAS_TABLE_HDR_VAL &&
1512 	    hdr->header != RAS_TABLE_HDR_BAD) {
1513 		dev_info(adev->dev, "Creating a new EEPROM table");
1514 		return amdgpu_ras_eeprom_reset_table(control);
1515 	}
1516 
1517 	if (!(adev->flags & AMD_IS_APU) && (dev_var == 0x5) &&
1518 	    (vram_type == AMDGPU_VRAM_TYPE_HBM3E) &&
1519 	    (hdr->version < RAS_TABLE_VER_V3)) {
1520 		return amdgpu_ras_eeprom_reset_table(control);
1521 	}
1522 
1523 	switch (hdr->version) {
1524 	case RAS_TABLE_VER_V2_1:
1525 	case RAS_TABLE_VER_V3:
1526 		if (hdr->tbl_size < RAS_TABLE_HEADER_SIZE + RAS_TABLE_V2_1_INFO_SIZE) {
1527 			dev_err(adev->dev,
1528 				"RAS header invalid, tbl_size %u smaller than minimum %u, resetting table\n",
1529 				hdr->tbl_size,
1530 				RAS_TABLE_HEADER_SIZE + RAS_TABLE_V2_1_INFO_SIZE);
1531 			return amdgpu_ras_eeprom_reset_table(control);
1532 		}
1533 		control->ras_num_recs = RAS_NUM_RECS_V2_1(hdr);
1534 		control->ras_record_offset = RAS_RECORD_START_V2_1;
1535 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT_V2_1;
1536 		break;
1537 	case RAS_TABLE_VER_V1:
1538 		if (hdr->tbl_size < RAS_TABLE_HEADER_SIZE) {
1539 			dev_err(adev->dev,
1540 				"RAS header invalid, tbl_size %u smaller than minimum %u, resetting table\n",
1541 				hdr->tbl_size, RAS_TABLE_HEADER_SIZE);
1542 			return amdgpu_ras_eeprom_reset_table(control);
1543 		}
1544 		control->ras_num_recs = RAS_NUM_RECS(hdr);
1545 		control->ras_record_offset = RAS_RECORD_START;
1546 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT;
1547 		break;
1548 	default:
1549 		dev_err(adev->dev,
1550 			"RAS header invalid, unsupported version: %u",
1551 			hdr->version);
1552 		return -EINVAL;
1553 	}
1554 
1555 	if (control->ras_num_recs > control->ras_max_record_count) {
1556 		dev_err(adev->dev,
1557 			"RAS header invalid, records in header: %u max allowed :%u",
1558 			control->ras_num_recs, control->ras_max_record_count);
1559 		return -EINVAL;
1560 	}
1561 
1562 	control->ras_fri = RAS_OFFSET_TO_INDEX(control, hdr->first_rec_offset);
1563 	if (hdr->first_rec_offset < control->ras_record_offset ||
1564 	    control->ras_fri >= control->ras_max_record_count) {
1565 		dev_err(adev->dev,
1566 			"RAS header invalid, ras_fri: %u, first_rec_offset:0x%x",
1567 			control->ras_fri, hdr->first_rec_offset);
1568 		return -EINVAL;
1569 	}
1570 
1571 	control->ras_num_mca_recs = 0;
1572 	control->ras_num_pa_recs = 0;
1573 	return 0;
1574 }
1575 
1576 int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)
1577 {
1578 	struct amdgpu_device *adev = to_amdgpu_device(control);
1579 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1580 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1581 	int res = 0;
1582 
1583 	if (!__is_ras_eeprom_supported(adev))
1584 		return 0;
1585 
1586 	/* Verify i2c adapter is initialized */
1587 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1588 		return -ENOENT;
1589 
1590 	if (!__get_eeprom_i2c_addr(adev, control))
1591 		return -EINVAL;
1592 
1593 	control->ras_num_bad_pages = ras->bad_page_num;
1594 
1595 	if (hdr->header == RAS_TABLE_HDR_VAL) {
1596 		dev_dbg(adev->dev,
1597 			"Found existing EEPROM table with %d records",
1598 			control->ras_num_bad_pages);
1599 
1600 		if (hdr->version >= RAS_TABLE_VER_V2_1) {
1601 			res = __read_table_ras_info(control);
1602 			if (res)
1603 				return res;
1604 		}
1605 
1606 		res = __verify_ras_table_checksum(control);
1607 		if (res) {
1608 			dev_err(adev->dev,
1609 				"RAS table incorrect checksum or error:%d\n",
1610 				res);
1611 			return -EINVAL;
1612 		}
1613 
1614 		/* Warn if we are at 90% of the threshold or above
1615 		 */
1616 		if (10 * control->ras_num_bad_pages >= 9 * ras->bad_page_cnt_threshold)
1617 			dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
1618 					control->ras_num_bad_pages,
1619 					ras->bad_page_cnt_threshold);
1620 	} else if (hdr->header == RAS_TABLE_HDR_BAD &&
1621 		   amdgpu_bad_page_threshold != 0) {
1622 		if (hdr->version >= RAS_TABLE_VER_V2_1) {
1623 			res = __read_table_ras_info(control);
1624 			if (res)
1625 				return res;
1626 		}
1627 
1628 		res = __verify_ras_table_checksum(control);
1629 		if (res) {
1630 			dev_err(adev->dev,
1631 				"RAS Table incorrect checksum or error:%d\n",
1632 				res);
1633 			return -EINVAL;
1634 		}
1635 		if (ras->bad_page_cnt_threshold >= control->ras_num_bad_pages) {
1636 			/* This means that, the threshold was increased since
1637 			 * the last time the system was booted, and now,
1638 			 * ras->bad_page_cnt_threshold - control->num_recs > 0,
1639 			 * so that at least one more record can be saved,
1640 			 * before the page count threshold is reached.
1641 			 */
1642 			dev_info(adev->dev,
1643 				 "records:%d threshold:%d, resetting "
1644 				 "RAS table header signature",
1645 				 control->ras_num_bad_pages,
1646 				 ras->bad_page_cnt_threshold);
1647 			res = amdgpu_ras_eeprom_correct_header_tag(control,
1648 								   RAS_TABLE_HDR_VAL);
1649 		} else {
1650 			dev_warn(adev->dev,
1651 				"RAS records:%d exceed threshold:%d\n",
1652 				control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
1653 			if ((amdgpu_bad_page_threshold == -1) ||
1654 			    (amdgpu_bad_page_threshold == -2)) {
1655 				res = 0;
1656 				dev_warn(adev->dev,
1657 					 "Please consult AMD Service Action Guide (SAG) for appropriate service procedures\n");
1658 			} else {
1659 				ras->is_rma = true;
1660 				dev_warn(adev->dev,
1661 					 "User defined threshold is set, runtime service will be halt when threshold is reached\n");
1662 			}
1663 		}
1664 	}
1665 
1666 	return res < 0 ? res : 0;
1667 }
1668 
1669 void amdgpu_ras_eeprom_check_and_recover(struct amdgpu_device *adev)
1670 {
1671 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1672 	struct amdgpu_ras_eeprom_control *control;
1673 	int res;
1674 
1675 	if (!__is_ras_eeprom_supported(adev) || !ras)
1676 		return;
1677 	control = &ras->eeprom_control;
1678 	if (!control->is_eeprom_valid)
1679 		return;
1680 	res = __verify_ras_table_checksum(control);
1681 	if (res) {
1682 		dev_warn(adev->dev,
1683 			"RAS table incorrect checksum or error:%d, try to recover\n",
1684 			res);
1685 		if (!amdgpu_ras_eeprom_reset_table(control))
1686 			if (!amdgpu_ras_save_bad_pages(adev, NULL))
1687 				if (!__verify_ras_table_checksum(control)) {
1688 					dev_info(adev->dev, "RAS table recovery succeed\n");
1689 					return;
1690 				}
1691 		dev_err(adev->dev, "RAS table recovery failed\n");
1692 		control->is_eeprom_valid = false;
1693 	}
1694 	return;
1695 }
1696 
1697 void amdgpu_ras_check_bad_page_status(struct amdgpu_device *adev)
1698 {
1699 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1700 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1701 
1702 	if (!__is_ras_eeprom_supported(adev) || !control || amdgpu_bad_page_threshold == 0)
1703 		return;
1704 
1705 	if (control->ras_num_bad_pages > ras->bad_page_cnt_threshold) {
1706 		if (amdgpu_dpm_send_rma_reason(adev))
1707 			dev_warn(adev->dev, "Unable to send out-of-band RMA CPER");
1708 		else
1709 			dev_dbg(adev->dev, "Sent out-of-band RMA CPER");
1710 
1711 		if (adev->cper.enabled && !amdgpu_uniras_enabled(adev)) {
1712 			if (amdgpu_cper_generate_bp_threshold_record(adev))
1713 				dev_warn(adev->dev, "Unable to send in-band RMA CPER");
1714 			else
1715 				dev_dbg(adev->dev, "Sent in-band RMA CPER");
1716 		}
1717 	}
1718 }
1719