xref: /openbsd-src/sys/dev/pci/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision 8dfe214903ce3625c937d5fad2469e8a0d1d4d71)
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 
36 /* These are memory addresses as would be seen by one or more EEPROM
37  * chips strung on the I2C bus, usually by manipulating pins 1-3 of a
38  * set of EEPROM devices. They form a continuous memory space.
39  *
40  * The I2C device address includes the device type identifier, 1010b,
41  * which is a reserved value and indicates that this is an I2C EEPROM
42  * device. It also includes the top 3 bits of the 19 bit EEPROM memory
43  * address, namely bits 18, 17, and 16. This makes up the 7 bit
44  * address sent on the I2C bus with bit 0 being the direction bit,
45  * which is not represented here, and sent by the hardware directly.
46  *
47  * For instance,
48  *   50h = 1010000b => device type identifier 1010b, bits 18:16 = 000b, address 0.
49  *   54h = 1010100b => --"--, bits 18:16 = 100b, address 40000h.
50  *   56h = 1010110b => --"--, bits 18:16 = 110b, address 60000h.
51  * Depending on the size of the I2C EEPROM device(s), bits 18:16 may
52  * address memory in a device or a device on the I2C bus, depending on
53  * the status of pins 1-3. See top of amdgpu_eeprom.c.
54  *
55  * The RAS table lives either at address 0 or address 40000h of EEPROM.
56  */
57 #define EEPROM_I2C_MADDR_0      0x0
58 #define EEPROM_I2C_MADDR_4      0x40000
59 
60 /*
61  * The 2 macros bellow represent the actual size in bytes that
62  * those entities occupy in the EEPROM memory.
63  * RAS_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
64  * uses uint64 to store 6b fields such as retired_page.
65  */
66 #define RAS_TABLE_HEADER_SIZE   20
67 #define RAS_TABLE_RECORD_SIZE   24
68 
69 /* Table hdr is 'AMDR' */
70 #define RAS_TABLE_HDR_VAL       0x414d4452
71 #define RAS_TABLE_VER           0x00010000
72 
73 /* Bad GPU tag ‘BADG’ */
74 #define RAS_TABLE_HDR_BAD       0x42414447
75 
76 /* Assume 2-Mbit size EEPROM and take up the whole space. */
77 #define RAS_TBL_SIZE_BYTES      (256 * 1024)
78 #define RAS_TABLE_START         0
79 #define RAS_HDR_START           RAS_TABLE_START
80 #define RAS_RECORD_START        (RAS_HDR_START + RAS_TABLE_HEADER_SIZE)
81 #define RAS_MAX_RECORD_COUNT    ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE) \
82 				 / RAS_TABLE_RECORD_SIZE)
83 
84 /* Given a zero-based index of an EEPROM RAS record, yields the EEPROM
85  * offset off of RAS_TABLE_START.  That is, this is something you can
86  * add to control->i2c_address, and then tell I2C layer to read
87  * from/write to there. _N is the so called absolute index,
88  * because it starts right after the table header.
89  */
90 #define RAS_INDEX_TO_OFFSET(_C, _N) ((_C)->ras_record_offset + \
91 				     (_N) * RAS_TABLE_RECORD_SIZE)
92 
93 #define RAS_OFFSET_TO_INDEX(_C, _O) (((_O) - \
94 				      (_C)->ras_record_offset) / RAS_TABLE_RECORD_SIZE)
95 
96 /* Given a 0-based relative record index, 0, 1, 2, ..., etc., off
97  * of "fri", return the absolute record index off of the end of
98  * the table header.
99  */
100 #define RAS_RI_TO_AI(_C, _I) (((_I) + (_C)->ras_fri) % \
101 			      (_C)->ras_max_record_count)
102 
103 #define RAS_NUM_RECS(_tbl_hdr)  (((_tbl_hdr)->tbl_size - \
104 				  RAS_TABLE_HEADER_SIZE) / RAS_TABLE_RECORD_SIZE)
105 
106 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_ras, eeprom_control))->adev
107 
108 static bool __is_ras_eeprom_supported(struct amdgpu_device *adev)
109 {
110 	switch (adev->ip_versions[MP1_HWIP][0]) {
111 	case IP_VERSION(11, 0, 2): /* VEGA20 and ARCTURUS */
112 	case IP_VERSION(11, 0, 7): /* Sienna cichlid */
113 	case IP_VERSION(13, 0, 0):
114 	case IP_VERSION(13, 0, 2): /* Aldebaran */
115 	case IP_VERSION(13, 0, 6):
116 	case IP_VERSION(13, 0, 10):
117 		return true;
118 	default:
119 		return false;
120 	}
121 }
122 
123 static bool __get_eeprom_i2c_addr(struct amdgpu_device *adev,
124 				  struct amdgpu_ras_eeprom_control *control)
125 {
126 	struct atom_context *atom_ctx = adev->mode_info.atom_context;
127 	u8 i2c_addr;
128 
129 	if (!control)
130 		return false;
131 
132 	if (amdgpu_atomfirmware_ras_rom_addr(adev, &i2c_addr)) {
133 		/* The address given by VBIOS is an 8-bit, wire-format
134 		 * address, i.e. the most significant byte.
135 		 *
136 		 * Normalize it to a 19-bit EEPROM address. Remove the
137 		 * device type identifier and make it a 7-bit address;
138 		 * then make it a 19-bit EEPROM address. See top of
139 		 * amdgpu_eeprom.c.
140 		 */
141 		i2c_addr = (i2c_addr & 0x0F) >> 1;
142 		control->i2c_address = ((u32) i2c_addr) << 16;
143 
144 		return true;
145 	}
146 
147 	switch (adev->ip_versions[MP1_HWIP][0]) {
148 	case IP_VERSION(11, 0, 2):
149 		/* VEGA20 and ARCTURUS */
150 		if (adev->asic_type == CHIP_VEGA20)
151 			control->i2c_address = EEPROM_I2C_MADDR_0;
152 		else if (strnstr(atom_ctx->vbios_version,
153 				 "D342",
154 				 sizeof(atom_ctx->vbios_version)))
155 			control->i2c_address = EEPROM_I2C_MADDR_0;
156 		else
157 			control->i2c_address = EEPROM_I2C_MADDR_4;
158 		return true;
159 	case IP_VERSION(11, 0, 7):
160 		control->i2c_address = EEPROM_I2C_MADDR_0;
161 		return true;
162 	case IP_VERSION(13, 0, 2):
163 		if (strnstr(atom_ctx->vbios_version, "D673",
164 			    sizeof(atom_ctx->vbios_version)))
165 			control->i2c_address = EEPROM_I2C_MADDR_4;
166 		else
167 			control->i2c_address = EEPROM_I2C_MADDR_0;
168 		return true;
169 	case IP_VERSION(13, 0, 0):
170 		if (strnstr(atom_ctx->vbios_pn, "D707",
171 			    sizeof(atom_ctx->vbios_pn)))
172 			control->i2c_address = EEPROM_I2C_MADDR_0;
173 		else
174 			control->i2c_address = EEPROM_I2C_MADDR_4;
175 		return true;
176 	case IP_VERSION(13, 0, 6):
177 	case IP_VERSION(13, 0, 10):
178 		control->i2c_address = EEPROM_I2C_MADDR_4;
179 		return true;
180 	default:
181 		return false;
182 	}
183 }
184 
185 static void
186 __encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header *hdr,
187 			     unsigned char *buf)
188 {
189 	u32 *pp = (uint32_t *)buf;
190 
191 	pp[0] = cpu_to_le32(hdr->header);
192 	pp[1] = cpu_to_le32(hdr->version);
193 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
194 	pp[3] = cpu_to_le32(hdr->tbl_size);
195 	pp[4] = cpu_to_le32(hdr->checksum);
196 }
197 
198 static void
199 __decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header *hdr,
200 			       unsigned char *buf)
201 {
202 	u32 *pp = (uint32_t *)buf;
203 
204 	hdr->header	      = le32_to_cpu(pp[0]);
205 	hdr->version	      = le32_to_cpu(pp[1]);
206 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
207 	hdr->tbl_size	      = le32_to_cpu(pp[3]);
208 	hdr->checksum	      = le32_to_cpu(pp[4]);
209 }
210 
211 static int __write_table_header(struct amdgpu_ras_eeprom_control *control)
212 {
213 	u8 buf[RAS_TABLE_HEADER_SIZE];
214 	struct amdgpu_device *adev = to_amdgpu_device(control);
215 	int res;
216 
217 	memset(buf, 0, sizeof(buf));
218 	__encode_table_header_to_buf(&control->tbl_hdr, buf);
219 
220 	/* i2c may be unstable in gpu reset */
221 	down_read(&adev->reset_domain->sem);
222 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
223 				  control->i2c_address +
224 				  control->ras_header_offset,
225 				  buf, RAS_TABLE_HEADER_SIZE);
226 	up_read(&adev->reset_domain->sem);
227 
228 	if (res < 0) {
229 		DRM_ERROR("Failed to write EEPROM table header:%d", res);
230 	} else if (res < RAS_TABLE_HEADER_SIZE) {
231 		DRM_ERROR("Short write:%d out of %d\n",
232 			  res, RAS_TABLE_HEADER_SIZE);
233 		res = -EIO;
234 	} else {
235 		res = 0;
236 	}
237 
238 	return res;
239 }
240 
241 static u8 __calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control *control)
242 {
243 	int ii;
244 	u8  *pp, csum;
245 	size_t sz;
246 
247 	/* Header checksum, skip checksum field in the calculation */
248 	sz = sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum);
249 	pp = (u8 *) &control->tbl_hdr;
250 	csum = 0;
251 	for (ii = 0; ii < sz; ii++, pp++)
252 		csum += *pp;
253 
254 	return csum;
255 }
256 
257 static int amdgpu_ras_eeprom_correct_header_tag(
258 	struct amdgpu_ras_eeprom_control *control,
259 	uint32_t header)
260 {
261 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
262 	u8 *hh;
263 	int res;
264 	u8 csum;
265 
266 	csum = -hdr->checksum;
267 
268 	hh = (void *) &hdr->header;
269 	csum -= (hh[0] + hh[1] + hh[2] + hh[3]);
270 	hh = (void *) &header;
271 	csum += hh[0] + hh[1] + hh[2] + hh[3];
272 	csum = -csum;
273 	mutex_lock(&control->ras_tbl_mutex);
274 	hdr->header = header;
275 	hdr->checksum = csum;
276 	res = __write_table_header(control);
277 	mutex_unlock(&control->ras_tbl_mutex);
278 
279 	return res;
280 }
281 
282 /**
283  * amdgpu_ras_eeprom_reset_table -- Reset the RAS EEPROM table
284  * @control: pointer to control structure
285  *
286  * Reset the contents of the header of the RAS EEPROM table.
287  * Return 0 on success, -errno on error.
288  */
289 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
290 {
291 	struct amdgpu_device *adev = to_amdgpu_device(control);
292 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
293 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
294 	u8 csum;
295 	int res;
296 
297 	mutex_lock(&control->ras_tbl_mutex);
298 
299 	hdr->header = RAS_TABLE_HDR_VAL;
300 	hdr->version = RAS_TABLE_VER;
301 	hdr->first_rec_offset = RAS_RECORD_START;
302 	hdr->tbl_size = RAS_TABLE_HEADER_SIZE;
303 
304 	csum = __calc_hdr_byte_sum(control);
305 	csum = -csum;
306 	hdr->checksum = csum;
307 	res = __write_table_header(control);
308 
309 	control->ras_num_recs = 0;
310 	control->ras_fri = 0;
311 
312 	amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_recs);
313 
314 	control->bad_channel_bitmap = 0;
315 	amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap);
316 	con->update_channel_flag = false;
317 
318 	amdgpu_ras_debugfs_set_ret_size(control);
319 
320 	mutex_unlock(&control->ras_tbl_mutex);
321 
322 	return res;
323 }
324 
325 static void
326 __encode_table_record_to_buf(struct amdgpu_ras_eeprom_control *control,
327 			     struct eeprom_table_record *record,
328 			     unsigned char *buf)
329 {
330 	__le64 tmp = 0;
331 	int i = 0;
332 
333 	/* Next are all record fields according to EEPROM page spec in LE foramt */
334 	buf[i++] = record->err_type;
335 
336 	buf[i++] = record->bank;
337 
338 	tmp = cpu_to_le64(record->ts);
339 	memcpy(buf + i, &tmp, 8);
340 	i += 8;
341 
342 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
343 	memcpy(buf + i, &tmp, 6);
344 	i += 6;
345 
346 	buf[i++] = record->mem_channel;
347 	buf[i++] = record->mcumc_id;
348 
349 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
350 	memcpy(buf + i, &tmp, 6);
351 }
352 
353 static void
354 __decode_table_record_from_buf(struct amdgpu_ras_eeprom_control *control,
355 			       struct eeprom_table_record *record,
356 			       unsigned char *buf)
357 {
358 	__le64 tmp = 0;
359 	int i =  0;
360 
361 	/* Next are all record fields according to EEPROM page spec in LE foramt */
362 	record->err_type = buf[i++];
363 
364 	record->bank = buf[i++];
365 
366 	memcpy(&tmp, buf + i, 8);
367 	record->ts = le64_to_cpu(tmp);
368 	i += 8;
369 
370 	memcpy(&tmp, buf + i, 6);
371 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
372 	i += 6;
373 
374 	record->mem_channel = buf[i++];
375 	record->mcumc_id = buf[i++];
376 
377 	memcpy(&tmp, buf + i,  6);
378 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
379 }
380 
381 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
382 {
383 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
384 
385 	if (!__is_ras_eeprom_supported(adev))
386 		return false;
387 
388 	/* skip check eeprom table for VEGA20 Gaming */
389 	if (!con)
390 		return false;
391 	else
392 		if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
393 			return false;
394 
395 	if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
396 		dev_warn(adev->dev, "This GPU is in BAD status.");
397 		dev_warn(adev->dev, "Please retire it or set a larger "
398 			 "threshold value when reloading driver.\n");
399 		return true;
400 	}
401 
402 	return false;
403 }
404 
405 /**
406  * __amdgpu_ras_eeprom_write -- write indexed from buffer to EEPROM
407  * @control: pointer to control structure
408  * @buf: pointer to buffer containing data to write
409  * @fri: start writing at this index
410  * @num: number of records to write
411  *
412  * The caller must hold the table mutex in @control.
413  * Return 0 on success, -errno otherwise.
414  */
415 static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control,
416 				     u8 *buf, const u32 fri, const u32 num)
417 {
418 	struct amdgpu_device *adev = to_amdgpu_device(control);
419 	u32 buf_size;
420 	int res;
421 
422 	/* i2c may be unstable in gpu reset */
423 	down_read(&adev->reset_domain->sem);
424 	buf_size = num * RAS_TABLE_RECORD_SIZE;
425 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
426 				  control->i2c_address +
427 				  RAS_INDEX_TO_OFFSET(control, fri),
428 				  buf, buf_size);
429 	up_read(&adev->reset_domain->sem);
430 	if (res < 0) {
431 		DRM_ERROR("Writing %d EEPROM table records error:%d",
432 			  num, res);
433 	} else if (res < buf_size) {
434 		/* Short write, return error.
435 		 */
436 		DRM_ERROR("Wrote %d records out of %d",
437 			  res / RAS_TABLE_RECORD_SIZE, num);
438 		res = -EIO;
439 	} else {
440 		res = 0;
441 	}
442 
443 	return res;
444 }
445 
446 static int
447 amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
448 			       struct eeprom_table_record *record,
449 			       const u32 num)
450 {
451 	struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
452 	u32 a, b, i;
453 	u8 *buf, *pp;
454 	int res;
455 
456 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
457 	if (!buf)
458 		return -ENOMEM;
459 
460 	/* Encode all of them in one go.
461 	 */
462 	pp = buf;
463 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
464 		__encode_table_record_to_buf(control, &record[i], pp);
465 
466 		/* update bad channel bitmap */
467 		if (!(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
468 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
469 			con->update_channel_flag = true;
470 		}
471 	}
472 
473 	/* a, first record index to write into.
474 	 * b, last record index to write into.
475 	 * a = first index to read (fri) + number of records in the table,
476 	 * b = a + @num - 1.
477 	 * Let N = control->ras_max_num_record_count, then we have,
478 	 * case 0: 0 <= a <= b < N,
479 	 *   just append @num records starting at a;
480 	 * case 1: 0 <= a < N <= b,
481 	 *   append (N - a) records starting at a, and
482 	 *   append the remainder,  b % N + 1, starting at 0.
483 	 * case 2: 0 <= fri < N <= a <= b, then modulo N we get two subcases,
484 	 * case 2a: 0 <= a <= b < N
485 	 *   append num records starting at a; and fix fri if b overwrote it,
486 	 *   and since a <= b, if b overwrote it then a must've also,
487 	 *   and if b didn't overwrite it, then a didn't also.
488 	 * case 2b: 0 <= b < a < N
489 	 *   write num records starting at a, which wraps around 0=N
490 	 *   and overwrite fri unconditionally. Now from case 2a,
491 	 *   this means that b eclipsed fri to overwrite it and wrap
492 	 *   around 0 again, i.e. b = 2N+r pre modulo N, so we unconditionally
493 	 *   set fri = b + 1 (mod N).
494 	 * Now, since fri is updated in every case, except the trivial case 0,
495 	 * the number of records present in the table after writing, is,
496 	 * num_recs - 1 = b - fri (mod N), and we take the positive value,
497 	 * by adding an arbitrary multiple of N before taking the modulo N
498 	 * as shown below.
499 	 */
500 	a = control->ras_fri + control->ras_num_recs;
501 	b = a + num  - 1;
502 	if (b < control->ras_max_record_count) {
503 		res = __amdgpu_ras_eeprom_write(control, buf, a, num);
504 	} else if (a < control->ras_max_record_count) {
505 		u32 g0, g1;
506 
507 		g0 = control->ras_max_record_count - a;
508 		g1 = b % control->ras_max_record_count + 1;
509 		res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
510 		if (res)
511 			goto Out;
512 		res = __amdgpu_ras_eeprom_write(control,
513 						buf + g0 * RAS_TABLE_RECORD_SIZE,
514 						0, g1);
515 		if (res)
516 			goto Out;
517 		if (g1 > control->ras_fri)
518 			control->ras_fri = g1 % control->ras_max_record_count;
519 	} else {
520 		a %= control->ras_max_record_count;
521 		b %= control->ras_max_record_count;
522 
523 		if (a <= b) {
524 			/* Note that, b - a + 1 = num. */
525 			res = __amdgpu_ras_eeprom_write(control, buf, a, num);
526 			if (res)
527 				goto Out;
528 			if (b >= control->ras_fri)
529 				control->ras_fri = (b + 1) % control->ras_max_record_count;
530 		} else {
531 			u32 g0, g1;
532 
533 			/* b < a, which means, we write from
534 			 * a to the end of the table, and from
535 			 * the start of the table to b.
536 			 */
537 			g0 = control->ras_max_record_count - a;
538 			g1 = b + 1;
539 			res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
540 			if (res)
541 				goto Out;
542 			res = __amdgpu_ras_eeprom_write(control,
543 							buf + g0 * RAS_TABLE_RECORD_SIZE,
544 							0, g1);
545 			if (res)
546 				goto Out;
547 			control->ras_fri = g1 % control->ras_max_record_count;
548 		}
549 	}
550 	control->ras_num_recs = 1 + (control->ras_max_record_count + b
551 				     - control->ras_fri)
552 		% control->ras_max_record_count;
553 Out:
554 	kfree(buf);
555 	return res;
556 }
557 
558 static int
559 amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
560 {
561 	struct amdgpu_device *adev = to_amdgpu_device(control);
562 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
563 	u8 *buf, *pp, csum;
564 	u32 buf_size;
565 	int res;
566 
567 	/* Modify the header if it exceeds.
568 	 */
569 	if (amdgpu_bad_page_threshold != 0 &&
570 	    control->ras_num_recs >= ras->bad_page_cnt_threshold) {
571 		dev_warn(adev->dev,
572 			"Saved bad pages %d reaches threshold value %d\n",
573 			control->ras_num_recs, ras->bad_page_cnt_threshold);
574 		control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
575 	}
576 
577 	control->tbl_hdr.version = RAS_TABLE_VER;
578 	control->tbl_hdr.first_rec_offset = RAS_INDEX_TO_OFFSET(control, control->ras_fri);
579 	control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE + control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
580 	control->tbl_hdr.checksum = 0;
581 
582 	buf_size = control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
583 	buf = kcalloc(control->ras_num_recs, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
584 	if (!buf) {
585 		DRM_ERROR("allocating memory for table of size %d bytes failed\n",
586 			  control->tbl_hdr.tbl_size);
587 		res = -ENOMEM;
588 		goto Out;
589 	}
590 
591 	down_read(&adev->reset_domain->sem);
592 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
593 				 control->i2c_address +
594 				 control->ras_record_offset,
595 				 buf, buf_size);
596 	up_read(&adev->reset_domain->sem);
597 	if (res < 0) {
598 		DRM_ERROR("EEPROM failed reading records:%d\n",
599 			  res);
600 		goto Out;
601 	} else if (res < buf_size) {
602 		DRM_ERROR("EEPROM read %d out of %d bytes\n",
603 			  res, buf_size);
604 		res = -EIO;
605 		goto Out;
606 	}
607 
608 	/* Recalc the checksum.
609 	 */
610 	csum = 0;
611 	for (pp = buf; pp < buf + buf_size; pp++)
612 		csum += *pp;
613 
614 	csum += __calc_hdr_byte_sum(control);
615 	/* avoid sign extension when assigning to "checksum" */
616 	csum = -csum;
617 	control->tbl_hdr.checksum = csum;
618 	res = __write_table_header(control);
619 Out:
620 	kfree(buf);
621 	return res;
622 }
623 
624 /**
625  * amdgpu_ras_eeprom_append -- append records to the EEPROM RAS table
626  * @control: pointer to control structure
627  * @record: array of records to append
628  * @num: number of records in @record array
629  *
630  * Append @num records to the table, calculate the checksum and write
631  * the table back to EEPROM. The maximum number of records that
632  * can be appended is between 1 and control->ras_max_record_count,
633  * regardless of how many records are already stored in the table.
634  *
635  * Return 0 on success or if EEPROM is not supported, -errno on error.
636  */
637 int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
638 			     struct eeprom_table_record *record,
639 			     const u32 num)
640 {
641 	struct amdgpu_device *adev = to_amdgpu_device(control);
642 	int res;
643 
644 	if (!__is_ras_eeprom_supported(adev))
645 		return 0;
646 
647 	if (num == 0) {
648 		DRM_ERROR("will not append 0 records\n");
649 		return -EINVAL;
650 	} else if (num > control->ras_max_record_count) {
651 		DRM_ERROR("cannot append %d records than the size of table %d\n",
652 			  num, control->ras_max_record_count);
653 		return -EINVAL;
654 	}
655 
656 	mutex_lock(&control->ras_tbl_mutex);
657 
658 	res = amdgpu_ras_eeprom_append_table(control, record, num);
659 	if (!res)
660 		res = amdgpu_ras_eeprom_update_header(control);
661 	if (!res)
662 		amdgpu_ras_debugfs_set_ret_size(control);
663 
664 	mutex_unlock(&control->ras_tbl_mutex);
665 	return res;
666 }
667 
668 /**
669  * __amdgpu_ras_eeprom_read -- read indexed from EEPROM into buffer
670  * @control: pointer to control structure
671  * @buf: pointer to buffer to read into
672  * @fri: first record index, start reading at this index, absolute index
673  * @num: number of records to read
674  *
675  * The caller must hold the table mutex in @control.
676  * Return 0 on success, -errno otherwise.
677  */
678 static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
679 				    u8 *buf, const u32 fri, const u32 num)
680 {
681 	struct amdgpu_device *adev = to_amdgpu_device(control);
682 	u32 buf_size;
683 	int res;
684 
685 	/* i2c may be unstable in gpu reset */
686 	down_read(&adev->reset_domain->sem);
687 	buf_size = num * RAS_TABLE_RECORD_SIZE;
688 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
689 				 control->i2c_address +
690 				 RAS_INDEX_TO_OFFSET(control, fri),
691 				 buf, buf_size);
692 	up_read(&adev->reset_domain->sem);
693 	if (res < 0) {
694 		DRM_ERROR("Reading %d EEPROM table records error:%d",
695 			  num, res);
696 	} else if (res < buf_size) {
697 		/* Short read, return error.
698 		 */
699 		DRM_ERROR("Read %d records out of %d",
700 			  res / RAS_TABLE_RECORD_SIZE, num);
701 		res = -EIO;
702 	} else {
703 		res = 0;
704 	}
705 
706 	return res;
707 }
708 
709 /**
710  * amdgpu_ras_eeprom_read -- read EEPROM
711  * @control: pointer to control structure
712  * @record: array of records to read into
713  * @num: number of records in @record
714  *
715  * Reads num records from the RAS table in EEPROM and
716  * writes the data into @record array.
717  *
718  * Returns 0 on success, -errno on error.
719  */
720 int amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
721 			   struct eeprom_table_record *record,
722 			   const u32 num)
723 {
724 	struct amdgpu_device *adev = to_amdgpu_device(control);
725 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
726 	int i, res;
727 	u8 *buf, *pp;
728 	u32 g0, g1;
729 
730 	if (!__is_ras_eeprom_supported(adev))
731 		return 0;
732 
733 	if (num == 0) {
734 		DRM_ERROR("will not read 0 records\n");
735 		return -EINVAL;
736 	} else if (num > control->ras_num_recs) {
737 		DRM_ERROR("too many records to read:%d available:%d\n",
738 			  num, control->ras_num_recs);
739 		return -EINVAL;
740 	}
741 
742 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
743 	if (!buf)
744 		return -ENOMEM;
745 
746 	/* Determine how many records to read, from the first record
747 	 * index, fri, to the end of the table, and from the beginning
748 	 * of the table, such that the total number of records is
749 	 * @num, and we handle wrap around when fri > 0 and
750 	 * fri + num > RAS_MAX_RECORD_COUNT.
751 	 *
752 	 * First we compute the index of the last element
753 	 * which would be fetched from each region,
754 	 * g0 is in [fri, fri + num - 1], and
755 	 * g1 is in [0, RAS_MAX_RECORD_COUNT - 1].
756 	 * Then, if g0 < RAS_MAX_RECORD_COUNT, the index of
757 	 * the last element to fetch, we set g0 to _the number_
758 	 * of elements to fetch, @num, since we know that the last
759 	 * indexed to be fetched does not exceed the table.
760 	 *
761 	 * If, however, g0 >= RAS_MAX_RECORD_COUNT, then
762 	 * we set g0 to the number of elements to read
763 	 * until the end of the table, and g1 to the number of
764 	 * elements to read from the beginning of the table.
765 	 */
766 	g0 = control->ras_fri + num - 1;
767 	g1 = g0 % control->ras_max_record_count;
768 	if (g0 < control->ras_max_record_count) {
769 		g0 = num;
770 		g1 = 0;
771 	} else {
772 		g0 = control->ras_max_record_count - control->ras_fri;
773 		g1 += 1;
774 	}
775 
776 	mutex_lock(&control->ras_tbl_mutex);
777 	res = __amdgpu_ras_eeprom_read(control, buf, control->ras_fri, g0);
778 	if (res)
779 		goto Out;
780 	if (g1) {
781 		res = __amdgpu_ras_eeprom_read(control,
782 					       buf + g0 * RAS_TABLE_RECORD_SIZE,
783 					       0, g1);
784 		if (res)
785 			goto Out;
786 	}
787 
788 	res = 0;
789 
790 	/* Read up everything? Then transform.
791 	 */
792 	pp = buf;
793 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
794 		__decode_table_record_from_buf(control, &record[i], pp);
795 
796 		/* update bad channel bitmap */
797 		if (!(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
798 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
799 			con->update_channel_flag = true;
800 		}
801 	}
802 Out:
803 	kfree(buf);
804 	mutex_unlock(&control->ras_tbl_mutex);
805 
806 	return res;
807 }
808 
809 uint32_t amdgpu_ras_eeprom_max_record_count(void)
810 {
811 	return RAS_MAX_RECORD_COUNT;
812 }
813 
814 #ifdef __linux__
815 static ssize_t
816 amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
817 				    size_t size, loff_t *pos)
818 {
819 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
820 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
821 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
822 	u8 data[50];
823 	int res;
824 
825 	if (!size)
826 		return size;
827 
828 	if (!ras || !control) {
829 		res = snprintf(data, sizeof(data), "Not supported\n");
830 	} else {
831 		res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
832 			       RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
833 	}
834 
835 	if (*pos >= res)
836 		return 0;
837 
838 	res -= *pos;
839 	res = min_t(size_t, res, size);
840 
841 	if (copy_to_user(buf, &data[*pos], res))
842 		return -EFAULT;
843 
844 	*pos += res;
845 
846 	return res;
847 }
848 
849 const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
850 	.owner = THIS_MODULE,
851 	.read = amdgpu_ras_debugfs_eeprom_size_read,
852 	.write = NULL,
853 	.llseek = default_llseek,
854 };
855 
856 static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
857 static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
858 #define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
859 static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
860 static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
861 #define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
862 
863 static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
864 	"ignore",
865 	"re",
866 	"ue",
867 };
868 
869 static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
870 {
871 	return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
872 		strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
873 }
874 
875 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
876 {
877 	struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
878 					      eeprom_control);
879 	struct dentry *de = ras->de_ras_eeprom_table;
880 
881 	if (de)
882 		d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
883 }
884 
885 static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
886 					     size_t size, loff_t *pos)
887 {
888 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
889 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
890 	struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
891 	const size_t orig_size = size;
892 	int res = -EFAULT;
893 	size_t data_len;
894 
895 	mutex_lock(&control->ras_tbl_mutex);
896 
897 	/* We want *pos - data_len > 0, which means there's
898 	 * bytes to be printed from data.
899 	 */
900 	data_len = strlen(tbl_hdr_str);
901 	if (*pos < data_len) {
902 		data_len -= *pos;
903 		data_len = min_t(size_t, data_len, size);
904 		if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
905 			goto Out;
906 		buf += data_len;
907 		size -= data_len;
908 		*pos += data_len;
909 	}
910 
911 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
912 	if (*pos < data_len && size > 0) {
913 		u8 data[tbl_hdr_fmt_size + 1];
914 		loff_t lpos;
915 
916 		snprintf(data, sizeof(data), tbl_hdr_fmt,
917 			 control->tbl_hdr.header,
918 			 control->tbl_hdr.version,
919 			 control->tbl_hdr.first_rec_offset,
920 			 control->tbl_hdr.tbl_size,
921 			 control->tbl_hdr.checksum);
922 
923 		data_len -= *pos;
924 		data_len = min_t(size_t, data_len, size);
925 		lpos = *pos - strlen(tbl_hdr_str);
926 		if (copy_to_user(buf, &data[lpos], data_len))
927 			goto Out;
928 		buf += data_len;
929 		size -= data_len;
930 		*pos += data_len;
931 	}
932 
933 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
934 	if (*pos < data_len && size > 0) {
935 		loff_t lpos;
936 
937 		data_len -= *pos;
938 		data_len = min_t(size_t, data_len, size);
939 		lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
940 		if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
941 			goto Out;
942 		buf += data_len;
943 		size -= data_len;
944 		*pos += data_len;
945 	}
946 
947 	data_len = amdgpu_ras_debugfs_table_size(control);
948 	if (*pos < data_len && size > 0) {
949 		u8 dare[RAS_TABLE_RECORD_SIZE];
950 		u8 data[rec_hdr_fmt_size + 1];
951 		struct eeprom_table_record record;
952 		int s, r;
953 
954 		/* Find the starting record index
955 		 */
956 		s = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
957 			strlen(rec_hdr_str);
958 		s = s / rec_hdr_fmt_size;
959 		r = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
960 			strlen(rec_hdr_str);
961 		r = r % rec_hdr_fmt_size;
962 
963 		for ( ; size > 0 && s < control->ras_num_recs; s++) {
964 			u32 ai = RAS_RI_TO_AI(control, s);
965 			/* Read a single record
966 			 */
967 			res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
968 			if (res)
969 				goto Out;
970 			__decode_table_record_from_buf(control, &record, dare);
971 			snprintf(data, sizeof(data), rec_hdr_fmt,
972 				 s,
973 				 RAS_INDEX_TO_OFFSET(control, ai),
974 				 record_err_type_str[record.err_type],
975 				 record.bank,
976 				 record.ts,
977 				 record.offset,
978 				 record.mem_channel,
979 				 record.mcumc_id,
980 				 record.retired_page);
981 
982 			data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
983 			if (copy_to_user(buf, &data[r], data_len)) {
984 				res = -EFAULT;
985 				goto Out;
986 			}
987 			buf += data_len;
988 			size -= data_len;
989 			*pos += data_len;
990 			r = 0;
991 		}
992 	}
993 	res = 0;
994 Out:
995 	mutex_unlock(&control->ras_tbl_mutex);
996 	return res < 0 ? res : orig_size - size;
997 }
998 
999 static ssize_t
1000 amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
1001 				     size_t size, loff_t *pos)
1002 {
1003 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1004 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1005 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1006 	u8 data[81];
1007 	int res;
1008 
1009 	if (!size)
1010 		return size;
1011 
1012 	if (!ras || !control) {
1013 		res = snprintf(data, sizeof(data), "Not supported\n");
1014 		if (*pos >= res)
1015 			return 0;
1016 
1017 		res -= *pos;
1018 		res = min_t(size_t, res, size);
1019 
1020 		if (copy_to_user(buf, &data[*pos], res))
1021 			return -EFAULT;
1022 
1023 		*pos += res;
1024 
1025 		return res;
1026 	} else {
1027 		return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
1028 	}
1029 }
1030 
1031 const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
1032 	.owner = THIS_MODULE,
1033 	.read = amdgpu_ras_debugfs_eeprom_table_read,
1034 	.write = NULL,
1035 	.llseek = default_llseek,
1036 };
1037 #else /* !__linux__ */
1038 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
1039 {
1040 }
1041 #endif
1042 
1043 /**
1044  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
1045  * @control: pointer to control structure
1046  *
1047  * Check the checksum of the stored in EEPROM RAS table.
1048  *
1049  * Return 0 if the checksum is correct,
1050  * positive if it is not correct, and
1051  * -errno on I/O error.
1052  */
1053 static int __verify_ras_table_checksum(struct amdgpu_ras_eeprom_control *control)
1054 {
1055 	struct amdgpu_device *adev = to_amdgpu_device(control);
1056 	int buf_size, res;
1057 	u8  csum, *buf, *pp;
1058 
1059 	buf_size = RAS_TABLE_HEADER_SIZE +
1060 		control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1061 	buf = kzalloc(buf_size, GFP_KERNEL);
1062 	if (!buf) {
1063 		DRM_ERROR("Out of memory checking RAS table checksum.\n");
1064 		return -ENOMEM;
1065 	}
1066 
1067 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1068 				 control->i2c_address +
1069 				 control->ras_header_offset,
1070 				 buf, buf_size);
1071 	if (res < buf_size) {
1072 		DRM_ERROR("Partial read for checksum, res:%d\n", res);
1073 		/* On partial reads, return -EIO.
1074 		 */
1075 		if (res >= 0)
1076 			res = -EIO;
1077 		goto Out;
1078 	}
1079 
1080 	csum = 0;
1081 	for (pp = buf; pp < buf + buf_size; pp++)
1082 		csum += *pp;
1083 Out:
1084 	kfree(buf);
1085 	return res < 0 ? res : csum;
1086 }
1087 
1088 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control,
1089 			   bool *exceed_err_limit)
1090 {
1091 	struct amdgpu_device *adev = to_amdgpu_device(control);
1092 	unsigned char buf[RAS_TABLE_HEADER_SIZE] = { 0 };
1093 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1094 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1095 	int res;
1096 
1097 	*exceed_err_limit = false;
1098 
1099 	if (!__is_ras_eeprom_supported(adev))
1100 		return 0;
1101 
1102 	/* Verify i2c adapter is initialized */
1103 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1104 		return -ENOENT;
1105 
1106 	if (!__get_eeprom_i2c_addr(adev, control))
1107 		return -EINVAL;
1108 
1109 	control->ras_header_offset = RAS_HDR_START;
1110 	control->ras_record_offset = RAS_RECORD_START;
1111 	control->ras_max_record_count  = RAS_MAX_RECORD_COUNT;
1112 	rw_init(&control->ras_tbl_mutex, "rastbl");
1113 
1114 	/* Read the table header from EEPROM address */
1115 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1116 				 control->i2c_address + control->ras_header_offset,
1117 				 buf, RAS_TABLE_HEADER_SIZE);
1118 	if (res < RAS_TABLE_HEADER_SIZE) {
1119 		DRM_ERROR("Failed to read EEPROM table header, res:%d", res);
1120 		return res >= 0 ? -EIO : res;
1121 	}
1122 
1123 	__decode_table_header_from_buf(hdr, buf);
1124 
1125 	control->ras_num_recs = RAS_NUM_RECS(hdr);
1126 	control->ras_fri = RAS_OFFSET_TO_INDEX(control, hdr->first_rec_offset);
1127 
1128 	if (hdr->header == RAS_TABLE_HDR_VAL) {
1129 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
1130 				 control->ras_num_recs);
1131 		res = __verify_ras_table_checksum(control);
1132 		if (res)
1133 			DRM_ERROR("RAS table incorrect checksum or error:%d\n",
1134 				  res);
1135 
1136 		/* Warn if we are at 90% of the threshold or above
1137 		 */
1138 		if (10 * control->ras_num_recs >= 9 * ras->bad_page_cnt_threshold)
1139 			dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
1140 					control->ras_num_recs,
1141 					ras->bad_page_cnt_threshold);
1142 	} else if (hdr->header == RAS_TABLE_HDR_BAD &&
1143 		   amdgpu_bad_page_threshold != 0) {
1144 		res = __verify_ras_table_checksum(control);
1145 		if (res)
1146 			DRM_ERROR("RAS Table incorrect checksum or error:%d\n",
1147 				  res);
1148 		if (ras->bad_page_cnt_threshold > control->ras_num_recs) {
1149 			/* This means that, the threshold was increased since
1150 			 * the last time the system was booted, and now,
1151 			 * ras->bad_page_cnt_threshold - control->num_recs > 0,
1152 			 * so that at least one more record can be saved,
1153 			 * before the page count threshold is reached.
1154 			 */
1155 			dev_info(adev->dev,
1156 				 "records:%d threshold:%d, resetting "
1157 				 "RAS table header signature",
1158 				 control->ras_num_recs,
1159 				 ras->bad_page_cnt_threshold);
1160 			res = amdgpu_ras_eeprom_correct_header_tag(control,
1161 								   RAS_TABLE_HDR_VAL);
1162 		} else {
1163 			dev_err(adev->dev, "RAS records:%d exceed threshold:%d",
1164 				control->ras_num_recs, ras->bad_page_cnt_threshold);
1165 			if (amdgpu_bad_page_threshold == -2) {
1166 				dev_warn(adev->dev, "GPU will be initialized due to bad_page_threshold = -2.");
1167 				res = 0;
1168 			} else {
1169 				*exceed_err_limit = true;
1170 				dev_err(adev->dev,
1171 					"RAS records:%d exceed threshold:%d, "
1172 					"GPU will not be initialized. Replace this GPU or increase the threshold",
1173 					control->ras_num_recs, ras->bad_page_cnt_threshold);
1174 			}
1175 		}
1176 	} else {
1177 		DRM_INFO("Creating a new EEPROM table");
1178 
1179 		res = amdgpu_ras_eeprom_reset_table(control);
1180 	}
1181 
1182 	return res < 0 ? res : 0;
1183 }
1184