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 #define EEPROM_I2C_MADDR_VEGA20         0x0
37 #define EEPROM_I2C_MADDR_ARCTURUS       0x40000
38 #define EEPROM_I2C_MADDR_ARCTURUS_D342  0x0
39 #define EEPROM_I2C_MADDR_SIENNA_CICHLID 0x0
40 #define EEPROM_I2C_MADDR_ALDEBARAN      0x0
41 #define EEPROM_I2C_MADDR_SMU_13_0_0     (0x54UL << 16)
42 
43 /*
44  * The 2 macros bellow represent the actual size in bytes that
45  * those entities occupy in the EEPROM memory.
46  * RAS_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
47  * uses uint64 to store 6b fields such as retired_page.
48  */
49 #define RAS_TABLE_HEADER_SIZE   20
50 #define RAS_TABLE_RECORD_SIZE   24
51 
52 /* Table hdr is 'AMDR' */
53 #define RAS_TABLE_HDR_VAL       0x414d4452
54 #define RAS_TABLE_VER           0x00010000
55 
56 /* Bad GPU tag ‘BADG’ */
57 #define RAS_TABLE_HDR_BAD       0x42414447
58 
59 /* Assume 2-Mbit size EEPROM and take up the whole space. */
60 #define RAS_TBL_SIZE_BYTES      (256 * 1024)
61 #define RAS_TABLE_START         0
62 #define RAS_HDR_START           RAS_TABLE_START
63 #define RAS_RECORD_START        (RAS_HDR_START + RAS_TABLE_HEADER_SIZE)
64 #define RAS_MAX_RECORD_COUNT    ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE) \
65 				 / RAS_TABLE_RECORD_SIZE)
66 
67 /* Given a zero-based index of an EEPROM RAS record, yields the EEPROM
68  * offset off of RAS_TABLE_START.  That is, this is something you can
69  * add to control->i2c_address, and then tell I2C layer to read
70  * from/write to there. _N is the so called absolute index,
71  * because it starts right after the table header.
72  */
73 #define RAS_INDEX_TO_OFFSET(_C, _N) ((_C)->ras_record_offset + \
74 				     (_N) * RAS_TABLE_RECORD_SIZE)
75 
76 #define RAS_OFFSET_TO_INDEX(_C, _O) (((_O) - \
77 				      (_C)->ras_record_offset) / RAS_TABLE_RECORD_SIZE)
78 
79 /* Given a 0-based relative record index, 0, 1, 2, ..., etc., off
80  * of "fri", return the absolute record index off of the end of
81  * the table header.
82  */
83 #define RAS_RI_TO_AI(_C, _I) (((_I) + (_C)->ras_fri) % \
84 			      (_C)->ras_max_record_count)
85 
86 #define RAS_NUM_RECS(_tbl_hdr)  (((_tbl_hdr)->tbl_size - \
87 				  RAS_TABLE_HEADER_SIZE) / RAS_TABLE_RECORD_SIZE)
88 
89 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_ras, eeprom_control))->adev
90 
__is_ras_eeprom_supported(struct amdgpu_device * adev)91 static bool __is_ras_eeprom_supported(struct amdgpu_device *adev)
92 {
93 	return  adev->asic_type == CHIP_VEGA20 ||
94 		adev->asic_type == CHIP_ARCTURUS ||
95 		adev->asic_type == CHIP_SIENNA_CICHLID ||
96 		adev->asic_type == CHIP_ALDEBARAN;
97 }
98 
__get_eeprom_i2c_addr_arct(struct amdgpu_device * adev,struct amdgpu_ras_eeprom_control * control)99 static bool __get_eeprom_i2c_addr_arct(struct amdgpu_device *adev,
100 				       struct amdgpu_ras_eeprom_control *control)
101 {
102 	struct atom_context *atom_ctx = adev->mode_info.atom_context;
103 
104 	if (!control || !atom_ctx)
105 		return false;
106 
107 	if (strnstr(atom_ctx->vbios_version,
108 	            "D342",
109 		    sizeof(atom_ctx->vbios_version)))
110 		control->i2c_address = EEPROM_I2C_MADDR_ARCTURUS_D342;
111 	else
112 		control->i2c_address = EEPROM_I2C_MADDR_ARCTURUS;
113 
114 	return true;
115 }
116 
__get_eeprom_i2c_addr(struct amdgpu_device * adev,struct amdgpu_ras_eeprom_control * control)117 static bool __get_eeprom_i2c_addr(struct amdgpu_device *adev,
118 				  struct amdgpu_ras_eeprom_control *control)
119 {
120 	u8 i2c_addr;
121 
122 	if (!control)
123 		return false;
124 
125 	if (amdgpu_atomfirmware_ras_rom_addr(adev, &i2c_addr)) {
126 		/* The address given by VBIOS is an 8-bit, wire-format
127 		 * address, i.e. the most significant byte.
128 		 *
129 		 * Normalize it to a 19-bit EEPROM address. Remove the
130 		 * device type identifier and make it a 7-bit address;
131 		 * then make it a 19-bit EEPROM address. See top of
132 		 * amdgpu_eeprom.c.
133 		 */
134 		i2c_addr = (i2c_addr & 0x0F) >> 1;
135 		control->i2c_address = ((u32) i2c_addr) << 16;
136 
137 		return true;
138 	}
139 
140 	switch (adev->asic_type) {
141 	case CHIP_VEGA20:
142 		control->i2c_address = EEPROM_I2C_MADDR_VEGA20;
143 		break;
144 
145 	case CHIP_ARCTURUS:
146 		return __get_eeprom_i2c_addr_arct(adev, control);
147 
148 	case CHIP_SIENNA_CICHLID:
149 		control->i2c_address = EEPROM_I2C_MADDR_SIENNA_CICHLID;
150 		break;
151 
152 	case CHIP_ALDEBARAN:
153 		control->i2c_address = EEPROM_I2C_MADDR_ALDEBARAN;
154 		break;
155 
156 	default:
157 		return false;
158 	}
159 
160 	switch (adev->ip_versions[MP1_HWIP][0]) {
161 	case IP_VERSION(13, 0, 0):
162 		control->i2c_address = EEPROM_I2C_MADDR_SMU_13_0_0;
163 		break;
164 
165 	default:
166 		break;
167 	}
168 
169 	return true;
170 }
171 
172 static void
__encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header * hdr,unsigned char * buf)173 __encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header *hdr,
174 			     unsigned char *buf)
175 {
176 	u32 *pp = (uint32_t *)buf;
177 
178 	pp[0] = cpu_to_le32(hdr->header);
179 	pp[1] = cpu_to_le32(hdr->version);
180 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
181 	pp[3] = cpu_to_le32(hdr->tbl_size);
182 	pp[4] = cpu_to_le32(hdr->checksum);
183 }
184 
185 static void
__decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header * hdr,unsigned char * buf)186 __decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header *hdr,
187 			       unsigned char *buf)
188 {
189 	u32 *pp = (uint32_t *)buf;
190 
191 	hdr->header	      = le32_to_cpu(pp[0]);
192 	hdr->version	      = le32_to_cpu(pp[1]);
193 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
194 	hdr->tbl_size	      = le32_to_cpu(pp[3]);
195 	hdr->checksum	      = le32_to_cpu(pp[4]);
196 }
197 
__write_table_header(struct amdgpu_ras_eeprom_control * control)198 static int __write_table_header(struct amdgpu_ras_eeprom_control *control)
199 {
200 	u8 buf[RAS_TABLE_HEADER_SIZE];
201 	struct amdgpu_device *adev = to_amdgpu_device(control);
202 	int res;
203 
204 	memset(buf, 0, sizeof(buf));
205 	__encode_table_header_to_buf(&control->tbl_hdr, buf);
206 
207 	/* i2c may be unstable in gpu reset */
208 	down_read(&adev->reset_domain->sem);
209 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
210 				  control->i2c_address +
211 				  control->ras_header_offset,
212 				  buf, RAS_TABLE_HEADER_SIZE);
213 	up_read(&adev->reset_domain->sem);
214 
215 	if (res < 0) {
216 		DRM_ERROR("Failed to write EEPROM table header:%d", res);
217 	} else if (res < RAS_TABLE_HEADER_SIZE) {
218 		DRM_ERROR("Short write:%d out of %d\n",
219 			  res, RAS_TABLE_HEADER_SIZE);
220 		res = -EIO;
221 	} else {
222 		res = 0;
223 	}
224 
225 	return res;
226 }
227 
__calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control * control)228 static u8 __calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control *control)
229 {
230 	int ii;
231 	u8  *pp, csum;
232 	size_t sz;
233 
234 	/* Header checksum, skip checksum field in the calculation */
235 	sz = sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum);
236 	pp = (u8 *) &control->tbl_hdr;
237 	csum = 0;
238 	for (ii = 0; ii < sz; ii++, pp++)
239 		csum += *pp;
240 
241 	return csum;
242 }
243 
amdgpu_ras_eeprom_correct_header_tag(struct amdgpu_ras_eeprom_control * control,uint32_t header)244 static int amdgpu_ras_eeprom_correct_header_tag(
245 	struct amdgpu_ras_eeprom_control *control,
246 	uint32_t header)
247 {
248 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
249 	u8 *hh;
250 	int res;
251 	u8 csum;
252 
253 	csum = -hdr->checksum;
254 
255 	hh = (void *) &hdr->header;
256 	csum -= (hh[0] + hh[1] + hh[2] + hh[3]);
257 	hh = (void *) &header;
258 	csum += hh[0] + hh[1] + hh[2] + hh[3];
259 	csum = -csum;
260 	mutex_lock(&control->ras_tbl_mutex);
261 	hdr->header = header;
262 	hdr->checksum = csum;
263 	res = __write_table_header(control);
264 	mutex_unlock(&control->ras_tbl_mutex);
265 
266 	return res;
267 }
268 
269 /**
270  * amdgpu_ras_eeprom_reset_table -- Reset the RAS EEPROM table
271  * @control: pointer to control structure
272  *
273  * Reset the contents of the header of the RAS EEPROM table.
274  * Return 0 on success, -errno on error.
275  */
amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control * control)276 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
277 {
278 	struct amdgpu_device *adev = to_amdgpu_device(control);
279 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
280 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
281 	u8 csum;
282 	int res;
283 
284 	mutex_lock(&control->ras_tbl_mutex);
285 
286 	hdr->header = RAS_TABLE_HDR_VAL;
287 	hdr->version = RAS_TABLE_VER;
288 	hdr->first_rec_offset = RAS_RECORD_START;
289 	hdr->tbl_size = RAS_TABLE_HEADER_SIZE;
290 
291 	csum = __calc_hdr_byte_sum(control);
292 	csum = -csum;
293 	hdr->checksum = csum;
294 	res = __write_table_header(control);
295 
296 	control->ras_num_recs = 0;
297 	control->ras_fri = 0;
298 
299 	amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_recs);
300 
301 	control->bad_channel_bitmap = 0;
302 	amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap);
303 	con->update_channel_flag = false;
304 
305 	amdgpu_ras_debugfs_set_ret_size(control);
306 
307 	mutex_unlock(&control->ras_tbl_mutex);
308 
309 	return res;
310 }
311 
312 static void
__encode_table_record_to_buf(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * record,unsigned char * buf)313 __encode_table_record_to_buf(struct amdgpu_ras_eeprom_control *control,
314 			     struct eeprom_table_record *record,
315 			     unsigned char *buf)
316 {
317 	__le64 tmp = 0;
318 	int i = 0;
319 
320 	/* Next are all record fields according to EEPROM page spec in LE foramt */
321 	buf[i++] = record->err_type;
322 
323 	buf[i++] = record->bank;
324 
325 	tmp = cpu_to_le64(record->ts);
326 	memcpy(buf + i, &tmp, 8);
327 	i += 8;
328 
329 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
330 	memcpy(buf + i, &tmp, 6);
331 	i += 6;
332 
333 	buf[i++] = record->mem_channel;
334 	buf[i++] = record->mcumc_id;
335 
336 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
337 	memcpy(buf + i, &tmp, 6);
338 }
339 
340 static void
__decode_table_record_from_buf(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * record,unsigned char * buf)341 __decode_table_record_from_buf(struct amdgpu_ras_eeprom_control *control,
342 			       struct eeprom_table_record *record,
343 			       unsigned char *buf)
344 {
345 	__le64 tmp = 0;
346 	int i =  0;
347 
348 	/* Next are all record fields according to EEPROM page spec in LE foramt */
349 	record->err_type = buf[i++];
350 
351 	record->bank = buf[i++];
352 
353 	memcpy(&tmp, buf + i, 8);
354 	record->ts = le64_to_cpu(tmp);
355 	i += 8;
356 
357 	memcpy(&tmp, buf + i, 6);
358 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
359 	i += 6;
360 
361 	record->mem_channel = buf[i++];
362 	record->mcumc_id = buf[i++];
363 
364 	memcpy(&tmp, buf + i,  6);
365 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
366 }
367 
amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device * adev)368 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
369 {
370 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
371 
372 	if (!__is_ras_eeprom_supported(adev))
373 		return false;
374 
375 	/* skip check eeprom table for VEGA20 Gaming */
376 	if (!con)
377 		return false;
378 	else
379 		if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
380 			return false;
381 
382 	if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
383 		dev_warn(adev->dev, "This GPU is in BAD status.");
384 		dev_warn(adev->dev, "Please retire it or set a larger "
385 			 "threshold value when reloading driver.\n");
386 		return true;
387 	}
388 
389 	return false;
390 }
391 
392 /**
393  * __amdgpu_ras_eeprom_write -- write indexed from buffer to EEPROM
394  * @control: pointer to control structure
395  * @buf: pointer to buffer containing data to write
396  * @fri: start writing at this index
397  * @num: number of records to write
398  *
399  * The caller must hold the table mutex in @control.
400  * Return 0 on success, -errno otherwise.
401  */
__amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control * control,u8 * buf,const u32 fri,const u32 num)402 static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control,
403 				     u8 *buf, const u32 fri, const u32 num)
404 {
405 	struct amdgpu_device *adev = to_amdgpu_device(control);
406 	u32 buf_size;
407 	int res;
408 
409 	/* i2c may be unstable in gpu reset */
410 	down_read(&adev->reset_domain->sem);
411 	buf_size = num * RAS_TABLE_RECORD_SIZE;
412 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
413 				  control->i2c_address +
414 				  RAS_INDEX_TO_OFFSET(control, fri),
415 				  buf, buf_size);
416 	up_read(&adev->reset_domain->sem);
417 	if (res < 0) {
418 		DRM_ERROR("Writing %d EEPROM table records error:%d",
419 			  num, res);
420 	} else if (res < buf_size) {
421 		/* Short write, return error.
422 		 */
423 		DRM_ERROR("Wrote %d records out of %d",
424 			  res / RAS_TABLE_RECORD_SIZE, num);
425 		res = -EIO;
426 	} else {
427 		res = 0;
428 	}
429 
430 	return res;
431 }
432 
433 static int
amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * record,const u32 num)434 amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
435 			       struct eeprom_table_record *record,
436 			       const u32 num)
437 {
438 	struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
439 	u32 a, b, i;
440 	u8 *buf, *pp;
441 	int res;
442 
443 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
444 	if (!buf)
445 		return -ENOMEM;
446 
447 	/* Encode all of them in one go.
448 	 */
449 	pp = buf;
450 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
451 		__encode_table_record_to_buf(control, &record[i], pp);
452 
453 		/* update bad channel bitmap */
454 		if (!(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
455 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
456 			con->update_channel_flag = true;
457 		}
458 	}
459 
460 	/* a, first record index to write into.
461 	 * b, last record index to write into.
462 	 * a = first index to read (fri) + number of records in the table,
463 	 * b = a + @num - 1.
464 	 * Let N = control->ras_max_num_record_count, then we have,
465 	 * case 0: 0 <= a <= b < N,
466 	 *   just append @num records starting at a;
467 	 * case 1: 0 <= a < N <= b,
468 	 *   append (N - a) records starting at a, and
469 	 *   append the remainder,  b % N + 1, starting at 0.
470 	 * case 2: 0 <= fri < N <= a <= b, then modulo N we get two subcases,
471 	 * case 2a: 0 <= a <= b < N
472 	 *   append num records starting at a; and fix fri if b overwrote it,
473 	 *   and since a <= b, if b overwrote it then a must've also,
474 	 *   and if b didn't overwrite it, then a didn't also.
475 	 * case 2b: 0 <= b < a < N
476 	 *   write num records starting at a, which wraps around 0=N
477 	 *   and overwrite fri unconditionally. Now from case 2a,
478 	 *   this means that b eclipsed fri to overwrite it and wrap
479 	 *   around 0 again, i.e. b = 2N+r pre modulo N, so we unconditionally
480 	 *   set fri = b + 1 (mod N).
481 	 * Now, since fri is updated in every case, except the trivial case 0,
482 	 * the number of records present in the table after writing, is,
483 	 * num_recs - 1 = b - fri (mod N), and we take the positive value,
484 	 * by adding an arbitrary multiple of N before taking the modulo N
485 	 * as shown below.
486 	 */
487 	a = control->ras_fri + control->ras_num_recs;
488 	b = a + num  - 1;
489 	if (b < control->ras_max_record_count) {
490 		res = __amdgpu_ras_eeprom_write(control, buf, a, num);
491 	} else if (a < control->ras_max_record_count) {
492 		u32 g0, g1;
493 
494 		g0 = control->ras_max_record_count - a;
495 		g1 = b % control->ras_max_record_count + 1;
496 		res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
497 		if (res)
498 			goto Out;
499 		res = __amdgpu_ras_eeprom_write(control,
500 						buf + g0 * RAS_TABLE_RECORD_SIZE,
501 						0, g1);
502 		if (res)
503 			goto Out;
504 		if (g1 > control->ras_fri)
505 			control->ras_fri = g1 % control->ras_max_record_count;
506 	} else {
507 		a %= control->ras_max_record_count;
508 		b %= control->ras_max_record_count;
509 
510 		if (a <= b) {
511 			/* Note that, b - a + 1 = num. */
512 			res = __amdgpu_ras_eeprom_write(control, buf, a, num);
513 			if (res)
514 				goto Out;
515 			if (b >= control->ras_fri)
516 				control->ras_fri = (b + 1) % control->ras_max_record_count;
517 		} else {
518 			u32 g0, g1;
519 
520 			/* b < a, which means, we write from
521 			 * a to the end of the table, and from
522 			 * the start of the table to b.
523 			 */
524 			g0 = control->ras_max_record_count - a;
525 			g1 = b + 1;
526 			res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
527 			if (res)
528 				goto Out;
529 			res = __amdgpu_ras_eeprom_write(control,
530 							buf + g0 * RAS_TABLE_RECORD_SIZE,
531 							0, g1);
532 			if (res)
533 				goto Out;
534 			control->ras_fri = g1 % control->ras_max_record_count;
535 		}
536 	}
537 	control->ras_num_recs = 1 + (control->ras_max_record_count + b
538 				     - control->ras_fri)
539 		% control->ras_max_record_count;
540 Out:
541 	kfree(buf);
542 	return res;
543 }
544 
545 static int
amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control * control)546 amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
547 {
548 	struct amdgpu_device *adev = to_amdgpu_device(control);
549 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
550 	u8 *buf, *pp, csum;
551 	u32 buf_size;
552 	int res;
553 
554 	/* Modify the header if it exceeds.
555 	 */
556 	if (amdgpu_bad_page_threshold != 0 &&
557 	    control->ras_num_recs >= ras->bad_page_cnt_threshold) {
558 		dev_warn(adev->dev,
559 			"Saved bad pages %d reaches threshold value %d\n",
560 			control->ras_num_recs, ras->bad_page_cnt_threshold);
561 		control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
562 	}
563 
564 	control->tbl_hdr.version = RAS_TABLE_VER;
565 	control->tbl_hdr.first_rec_offset = RAS_INDEX_TO_OFFSET(control, control->ras_fri);
566 	control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE + control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
567 	control->tbl_hdr.checksum = 0;
568 
569 	buf_size = control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
570 	buf = kcalloc(control->ras_num_recs, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
571 	if (!buf) {
572 		DRM_ERROR("allocating memory for table of size %d bytes failed\n",
573 			  control->tbl_hdr.tbl_size);
574 		res = -ENOMEM;
575 		goto Out;
576 	}
577 
578 	down_read(&adev->reset_domain->sem);
579 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
580 				 control->i2c_address +
581 				 control->ras_record_offset,
582 				 buf, buf_size);
583 	up_read(&adev->reset_domain->sem);
584 	if (res < 0) {
585 		DRM_ERROR("EEPROM failed reading records:%d\n",
586 			  res);
587 		goto Out;
588 	} else if (res < buf_size) {
589 		DRM_ERROR("EEPROM read %d out of %d bytes\n",
590 			  res, buf_size);
591 		res = -EIO;
592 		goto Out;
593 	}
594 
595 	/* Recalc the checksum.
596 	 */
597 	csum = 0;
598 	for (pp = buf; pp < buf + buf_size; pp++)
599 		csum += *pp;
600 
601 	csum += __calc_hdr_byte_sum(control);
602 	/* avoid sign extension when assigning to "checksum" */
603 	csum = -csum;
604 	control->tbl_hdr.checksum = csum;
605 	res = __write_table_header(control);
606 Out:
607 	kfree(buf);
608 	return res;
609 }
610 
611 /**
612  * amdgpu_ras_eeprom_append -- append records to the EEPROM RAS table
613  * @control: pointer to control structure
614  * @record: array of records to append
615  * @num: number of records in @record array
616  *
617  * Append @num records to the table, calculate the checksum and write
618  * the table back to EEPROM. The maximum number of records that
619  * can be appended is between 1 and control->ras_max_record_count,
620  * regardless of how many records are already stored in the table.
621  *
622  * Return 0 on success or if EEPROM is not supported, -errno on error.
623  */
amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * record,const u32 num)624 int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
625 			     struct eeprom_table_record *record,
626 			     const u32 num)
627 {
628 	struct amdgpu_device *adev = to_amdgpu_device(control);
629 	int res;
630 
631 	if (!__is_ras_eeprom_supported(adev))
632 		return 0;
633 
634 	if (num == 0) {
635 		DRM_ERROR("will not append 0 records\n");
636 		return -EINVAL;
637 	} else if (num > control->ras_max_record_count) {
638 		DRM_ERROR("cannot append %d records than the size of table %d\n",
639 			  num, control->ras_max_record_count);
640 		return -EINVAL;
641 	}
642 
643 	mutex_lock(&control->ras_tbl_mutex);
644 
645 	res = amdgpu_ras_eeprom_append_table(control, record, num);
646 	if (!res)
647 		res = amdgpu_ras_eeprom_update_header(control);
648 	if (!res)
649 		amdgpu_ras_debugfs_set_ret_size(control);
650 
651 	mutex_unlock(&control->ras_tbl_mutex);
652 	return res;
653 }
654 
655 /**
656  * __amdgpu_ras_eeprom_read -- read indexed from EEPROM into buffer
657  * @control: pointer to control structure
658  * @buf: pointer to buffer to read into
659  * @fri: first record index, start reading at this index, absolute index
660  * @num: number of records to read
661  *
662  * The caller must hold the table mutex in @control.
663  * Return 0 on success, -errno otherwise.
664  */
__amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control * control,u8 * buf,const u32 fri,const u32 num)665 static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
666 				    u8 *buf, const u32 fri, const u32 num)
667 {
668 	struct amdgpu_device *adev = to_amdgpu_device(control);
669 	u32 buf_size;
670 	int res;
671 
672 	/* i2c may be unstable in gpu reset */
673 	down_read(&adev->reset_domain->sem);
674 	buf_size = num * RAS_TABLE_RECORD_SIZE;
675 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
676 				 control->i2c_address +
677 				 RAS_INDEX_TO_OFFSET(control, fri),
678 				 buf, buf_size);
679 	up_read(&adev->reset_domain->sem);
680 	if (res < 0) {
681 		DRM_ERROR("Reading %d EEPROM table records error:%d",
682 			  num, res);
683 	} else if (res < buf_size) {
684 		/* Short read, return error.
685 		 */
686 		DRM_ERROR("Read %d records out of %d",
687 			  res / RAS_TABLE_RECORD_SIZE, num);
688 		res = -EIO;
689 	} else {
690 		res = 0;
691 	}
692 
693 	return res;
694 }
695 
696 /**
697  * amdgpu_ras_eeprom_read -- read EEPROM
698  * @control: pointer to control structure
699  * @record: array of records to read into
700  * @num: number of records in @record
701  *
702  * Reads num records from the RAS table in EEPROM and
703  * writes the data into @record array.
704  *
705  * Returns 0 on success, -errno on error.
706  */
amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * record,const u32 num)707 int amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
708 			   struct eeprom_table_record *record,
709 			   const u32 num)
710 {
711 	struct amdgpu_device *adev = to_amdgpu_device(control);
712 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
713 	int i, res;
714 	u8 *buf, *pp;
715 	u32 g0, g1;
716 
717 	if (!__is_ras_eeprom_supported(adev))
718 		return 0;
719 
720 	if (num == 0) {
721 		DRM_ERROR("will not read 0 records\n");
722 		return -EINVAL;
723 	} else if (num > control->ras_num_recs) {
724 		DRM_ERROR("too many records to read:%d available:%d\n",
725 			  num, control->ras_num_recs);
726 		return -EINVAL;
727 	}
728 
729 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
730 	if (!buf)
731 		return -ENOMEM;
732 
733 	/* Determine how many records to read, from the first record
734 	 * index, fri, to the end of the table, and from the beginning
735 	 * of the table, such that the total number of records is
736 	 * @num, and we handle wrap around when fri > 0 and
737 	 * fri + num > RAS_MAX_RECORD_COUNT.
738 	 *
739 	 * First we compute the index of the last element
740 	 * which would be fetched from each region,
741 	 * g0 is in [fri, fri + num - 1], and
742 	 * g1 is in [0, RAS_MAX_RECORD_COUNT - 1].
743 	 * Then, if g0 < RAS_MAX_RECORD_COUNT, the index of
744 	 * the last element to fetch, we set g0 to _the number_
745 	 * of elements to fetch, @num, since we know that the last
746 	 * indexed to be fetched does not exceed the table.
747 	 *
748 	 * If, however, g0 >= RAS_MAX_RECORD_COUNT, then
749 	 * we set g0 to the number of elements to read
750 	 * until the end of the table, and g1 to the number of
751 	 * elements to read from the beginning of the table.
752 	 */
753 	g0 = control->ras_fri + num - 1;
754 	g1 = g0 % control->ras_max_record_count;
755 	if (g0 < control->ras_max_record_count) {
756 		g0 = num;
757 		g1 = 0;
758 	} else {
759 		g0 = control->ras_max_record_count - control->ras_fri;
760 		g1 += 1;
761 	}
762 
763 	mutex_lock(&control->ras_tbl_mutex);
764 	res = __amdgpu_ras_eeprom_read(control, buf, control->ras_fri, g0);
765 	if (res)
766 		goto Out;
767 	if (g1) {
768 		res = __amdgpu_ras_eeprom_read(control,
769 					       buf + g0 * RAS_TABLE_RECORD_SIZE,
770 					       0, g1);
771 		if (res)
772 			goto Out;
773 	}
774 
775 	res = 0;
776 
777 	/* Read up everything? Then transform.
778 	 */
779 	pp = buf;
780 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
781 		__decode_table_record_from_buf(control, &record[i], pp);
782 
783 		/* update bad channel bitmap */
784 		if (!(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
785 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
786 			con->update_channel_flag = true;
787 		}
788 	}
789 Out:
790 	kfree(buf);
791 	mutex_unlock(&control->ras_tbl_mutex);
792 
793 	return res;
794 }
795 
amdgpu_ras_eeprom_max_record_count(void)796 uint32_t amdgpu_ras_eeprom_max_record_count(void)
797 {
798 	return RAS_MAX_RECORD_COUNT;
799 }
800 
801 static ssize_t
amdgpu_ras_debugfs_eeprom_size_read(struct file * f,char __user * buf,size_t size,loff_t * pos)802 amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
803 				    size_t size, loff_t *pos)
804 {
805 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
806 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
807 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
808 	u8 data[50];
809 	int res;
810 
811 	if (!size)
812 		return size;
813 
814 	if (!ras || !control) {
815 		res = snprintf(data, sizeof(data), "Not supported\n");
816 	} else {
817 		res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
818 			       RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
819 	}
820 
821 	if (*pos >= res)
822 		return 0;
823 
824 	res -= *pos;
825 	res = min_t(size_t, res, size);
826 
827 	if (copy_to_user(buf, &data[*pos], res))
828 		return -EFAULT;
829 
830 	*pos += res;
831 
832 	return res;
833 }
834 
835 const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
836 	.owner = THIS_MODULE,
837 	.read = amdgpu_ras_debugfs_eeprom_size_read,
838 	.write = NULL,
839 	.llseek = default_llseek,
840 };
841 
842 static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
843 static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
844 #define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
845 static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
846 static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
847 #define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
848 
849 static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
850 	"ignore",
851 	"re",
852 	"ue",
853 };
854 
amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control * control)855 static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
856 {
857 	return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
858 		strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
859 }
860 
amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control * control)861 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
862 {
863 	struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
864 					      eeprom_control);
865 	struct dentry *de = ras->de_ras_eeprom_table;
866 
867 	if (de)
868 		d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
869 }
870 
amdgpu_ras_debugfs_table_read(struct file * f,char __user * buf,size_t size,loff_t * pos)871 static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
872 					     size_t size, loff_t *pos)
873 {
874 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
875 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
876 	struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
877 	const size_t orig_size = size;
878 	int res = -EFAULT;
879 	size_t data_len;
880 
881 	mutex_lock(&control->ras_tbl_mutex);
882 
883 	/* We want *pos - data_len > 0, which means there's
884 	 * bytes to be printed from data.
885 	 */
886 	data_len = strlen(tbl_hdr_str);
887 	if (*pos < data_len) {
888 		data_len -= *pos;
889 		data_len = min_t(size_t, data_len, size);
890 		if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
891 			goto Out;
892 		buf += data_len;
893 		size -= data_len;
894 		*pos += data_len;
895 	}
896 
897 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
898 	if (*pos < data_len && size > 0) {
899 		u8 data[tbl_hdr_fmt_size + 1];
900 		loff_t lpos;
901 
902 		snprintf(data, sizeof(data), tbl_hdr_fmt,
903 			 control->tbl_hdr.header,
904 			 control->tbl_hdr.version,
905 			 control->tbl_hdr.first_rec_offset,
906 			 control->tbl_hdr.tbl_size,
907 			 control->tbl_hdr.checksum);
908 
909 		data_len -= *pos;
910 		data_len = min_t(size_t, data_len, size);
911 		lpos = *pos - strlen(tbl_hdr_str);
912 		if (copy_to_user(buf, &data[lpos], data_len))
913 			goto Out;
914 		buf += data_len;
915 		size -= data_len;
916 		*pos += data_len;
917 	}
918 
919 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
920 	if (*pos < data_len && size > 0) {
921 		loff_t lpos;
922 
923 		data_len -= *pos;
924 		data_len = min_t(size_t, data_len, size);
925 		lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
926 		if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
927 			goto Out;
928 		buf += data_len;
929 		size -= data_len;
930 		*pos += data_len;
931 	}
932 
933 	data_len = amdgpu_ras_debugfs_table_size(control);
934 	if (*pos < data_len && size > 0) {
935 		u8 dare[RAS_TABLE_RECORD_SIZE];
936 		u8 data[rec_hdr_fmt_size + 1];
937 		struct eeprom_table_record record;
938 		int s, r;
939 
940 		/* Find the starting record index
941 		 */
942 		s = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
943 			strlen(rec_hdr_str);
944 		s = s / rec_hdr_fmt_size;
945 		r = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
946 			strlen(rec_hdr_str);
947 		r = r % rec_hdr_fmt_size;
948 
949 		for ( ; size > 0 && s < control->ras_num_recs; s++) {
950 			u32 ai = RAS_RI_TO_AI(control, s);
951 			/* Read a single record
952 			 */
953 			res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
954 			if (res)
955 				goto Out;
956 			__decode_table_record_from_buf(control, &record, dare);
957 			snprintf(data, sizeof(data), rec_hdr_fmt,
958 				 s,
959 				 RAS_INDEX_TO_OFFSET(control, ai),
960 				 record_err_type_str[record.err_type],
961 				 record.bank,
962 				 record.ts,
963 				 record.offset,
964 				 record.mem_channel,
965 				 record.mcumc_id,
966 				 record.retired_page);
967 
968 			data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
969 			if (copy_to_user(buf, &data[r], data_len)) {
970 				res = -EFAULT;
971 				goto Out;
972 			}
973 			buf += data_len;
974 			size -= data_len;
975 			*pos += data_len;
976 			r = 0;
977 		}
978 	}
979 	res = 0;
980 Out:
981 	mutex_unlock(&control->ras_tbl_mutex);
982 	return res < 0 ? res : orig_size - size;
983 }
984 
985 static ssize_t
amdgpu_ras_debugfs_eeprom_table_read(struct file * f,char __user * buf,size_t size,loff_t * pos)986 amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
987 				     size_t size, loff_t *pos)
988 {
989 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
990 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
991 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
992 	u8 data[81];
993 	int res;
994 
995 	if (!size)
996 		return size;
997 
998 	if (!ras || !control) {
999 		res = snprintf(data, sizeof(data), "Not supported\n");
1000 		if (*pos >= res)
1001 			return 0;
1002 
1003 		res -= *pos;
1004 		res = min_t(size_t, res, size);
1005 
1006 		if (copy_to_user(buf, &data[*pos], res))
1007 			return -EFAULT;
1008 
1009 		*pos += res;
1010 
1011 		return res;
1012 	} else {
1013 		return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
1014 	}
1015 }
1016 
1017 const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
1018 	.owner = THIS_MODULE,
1019 	.read = amdgpu_ras_debugfs_eeprom_table_read,
1020 	.write = NULL,
1021 	.llseek = default_llseek,
1022 };
1023 
1024 /**
1025  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
1026  * @control: pointer to control structure
1027  *
1028  * Check the checksum of the stored in EEPROM RAS table.
1029  *
1030  * Return 0 if the checksum is correct,
1031  * positive if it is not correct, and
1032  * -errno on I/O error.
1033  */
__verify_ras_table_checksum(struct amdgpu_ras_eeprom_control * control)1034 static int __verify_ras_table_checksum(struct amdgpu_ras_eeprom_control *control)
1035 {
1036 	struct amdgpu_device *adev = to_amdgpu_device(control);
1037 	int buf_size, res;
1038 	u8  csum, *buf, *pp;
1039 
1040 	buf_size = RAS_TABLE_HEADER_SIZE +
1041 		control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1042 	buf = kzalloc(buf_size, GFP_KERNEL);
1043 	if (!buf) {
1044 		DRM_ERROR("Out of memory checking RAS table checksum.\n");
1045 		return -ENOMEM;
1046 	}
1047 
1048 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1049 				 control->i2c_address +
1050 				 control->ras_header_offset,
1051 				 buf, buf_size);
1052 	if (res < buf_size) {
1053 		DRM_ERROR("Partial read for checksum, res:%d\n", res);
1054 		/* On partial reads, return -EIO.
1055 		 */
1056 		if (res >= 0)
1057 			res = -EIO;
1058 		goto Out;
1059 	}
1060 
1061 	csum = 0;
1062 	for (pp = buf; pp < buf + buf_size; pp++)
1063 		csum += *pp;
1064 Out:
1065 	kfree(buf);
1066 	return res < 0 ? res : csum;
1067 }
1068 
amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control * control,bool * exceed_err_limit)1069 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control,
1070 			   bool *exceed_err_limit)
1071 {
1072 	struct amdgpu_device *adev = to_amdgpu_device(control);
1073 	unsigned char buf[RAS_TABLE_HEADER_SIZE] = { 0 };
1074 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1075 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1076 	int res;
1077 
1078 	*exceed_err_limit = false;
1079 
1080 	if (!__is_ras_eeprom_supported(adev))
1081 		return 0;
1082 
1083 	/* Verify i2c adapter is initialized */
1084 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1085 		return -ENOENT;
1086 
1087 	if (!__get_eeprom_i2c_addr(adev, control))
1088 		return -EINVAL;
1089 
1090 	control->ras_header_offset = RAS_HDR_START;
1091 	control->ras_record_offset = RAS_RECORD_START;
1092 	control->ras_max_record_count  = RAS_MAX_RECORD_COUNT;
1093 	mutex_init(&control->ras_tbl_mutex);
1094 
1095 	/* Read the table header from EEPROM address */
1096 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1097 				 control->i2c_address + control->ras_header_offset,
1098 				 buf, RAS_TABLE_HEADER_SIZE);
1099 	if (res < RAS_TABLE_HEADER_SIZE) {
1100 		DRM_ERROR("Failed to read EEPROM table header, res:%d", res);
1101 		return res >= 0 ? -EIO : res;
1102 	}
1103 
1104 	__decode_table_header_from_buf(hdr, buf);
1105 
1106 	control->ras_num_recs = RAS_NUM_RECS(hdr);
1107 	control->ras_fri = RAS_OFFSET_TO_INDEX(control, hdr->first_rec_offset);
1108 
1109 	if (hdr->header == RAS_TABLE_HDR_VAL) {
1110 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
1111 				 control->ras_num_recs);
1112 		res = __verify_ras_table_checksum(control);
1113 		if (res)
1114 			DRM_ERROR("RAS table incorrect checksum or error:%d\n",
1115 				  res);
1116 
1117 		/* Warn if we are at 90% of the threshold or above
1118 		 */
1119 		if (10 * control->ras_num_recs >= 9 * ras->bad_page_cnt_threshold)
1120 			dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
1121 					control->ras_num_recs,
1122 					ras->bad_page_cnt_threshold);
1123 	} else if (hdr->header == RAS_TABLE_HDR_BAD &&
1124 		   amdgpu_bad_page_threshold != 0) {
1125 		res = __verify_ras_table_checksum(control);
1126 		if (res)
1127 			DRM_ERROR("RAS Table incorrect checksum or error:%d\n",
1128 				  res);
1129 		if (ras->bad_page_cnt_threshold > control->ras_num_recs) {
1130 			/* This means that, the threshold was increased since
1131 			 * the last time the system was booted, and now,
1132 			 * ras->bad_page_cnt_threshold - control->num_recs > 0,
1133 			 * so that at least one more record can be saved,
1134 			 * before the page count threshold is reached.
1135 			 */
1136 			dev_info(adev->dev,
1137 				 "records:%d threshold:%d, resetting "
1138 				 "RAS table header signature",
1139 				 control->ras_num_recs,
1140 				 ras->bad_page_cnt_threshold);
1141 			res = amdgpu_ras_eeprom_correct_header_tag(control,
1142 								   RAS_TABLE_HDR_VAL);
1143 		} else {
1144 			dev_err(adev->dev, "RAS records:%d exceed threshold:%d",
1145 				control->ras_num_recs, ras->bad_page_cnt_threshold);
1146 			if (amdgpu_bad_page_threshold == -2) {
1147 				dev_warn(adev->dev, "GPU will be initialized due to bad_page_threshold = -2.");
1148 				res = 0;
1149 			} else {
1150 				*exceed_err_limit = true;
1151 				dev_err(adev->dev,
1152 					"RAS records:%d exceed threshold:%d, "
1153 					"GPU will not be initialized. Replace this GPU or increase the threshold",
1154 					control->ras_num_recs, ras->bad_page_cnt_threshold);
1155 			}
1156 		}
1157 	} else {
1158 		DRM_INFO("Creating a new EEPROM table");
1159 
1160 		res = amdgpu_ras_eeprom_reset_table(control);
1161 	}
1162 
1163 	return res < 0 ? res : 0;
1164 }
1165