xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision 5762f9dcf74ae5fb0b0a15afeda9543af38449ef)
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 below 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 
72 /* Bad GPU tag ‘BADG’ */
73 #define RAS_TABLE_HDR_BAD       0x42414447
74 
75 /*
76  * EEPROM Table structure v1
77  * ---------------------------------
78  * |                               |
79  * |     EEPROM TABLE HEADER       |
80  * |      ( size 20 Bytes )        |
81  * |                               |
82  * ---------------------------------
83  * |                               |
84  * |    BAD PAGE RECORD AREA       |
85  * |                               |
86  * ---------------------------------
87  */
88 
89 /* Assume 2-Mbit size EEPROM and take up the whole space. */
90 #define RAS_TBL_SIZE_BYTES      (256 * 1024)
91 #define RAS_TABLE_START         0
92 #define RAS_HDR_START           RAS_TABLE_START
93 #define RAS_RECORD_START        (RAS_HDR_START + RAS_TABLE_HEADER_SIZE)
94 #define RAS_MAX_RECORD_COUNT    ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE) \
95 				 / RAS_TABLE_RECORD_SIZE)
96 
97 /*
98  * EEPROM Table structrue v2.1
99  * ---------------------------------
100  * |                               |
101  * |     EEPROM TABLE HEADER       |
102  * |      ( size 20 Bytes )        |
103  * |                               |
104  * ---------------------------------
105  * |                               |
106  * |     EEPROM TABLE RAS INFO     |
107  * | (available info size 4 Bytes) |
108  * |  ( reserved size 252 Bytes )  |
109  * |                               |
110  * ---------------------------------
111  * |                               |
112  * |     BAD PAGE RECORD AREA      |
113  * |                               |
114  * ---------------------------------
115  */
116 
117 /* EEPROM Table V2_1 */
118 #define RAS_TABLE_V2_1_INFO_SIZE       256
119 #define RAS_TABLE_V2_1_INFO_START      RAS_TABLE_HEADER_SIZE
120 #define RAS_RECORD_START_V2_1          (RAS_HDR_START + RAS_TABLE_HEADER_SIZE + \
121 					RAS_TABLE_V2_1_INFO_SIZE)
122 #define RAS_MAX_RECORD_COUNT_V2_1      ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE - \
123 					RAS_TABLE_V2_1_INFO_SIZE) \
124 					/ RAS_TABLE_RECORD_SIZE)
125 
126 /* Given a zero-based index of an EEPROM RAS record, yields the EEPROM
127  * offset off of RAS_TABLE_START.  That is, this is something you can
128  * add to control->i2c_address, and then tell I2C layer to read
129  * from/write to there. _N is the so called absolute index,
130  * because it starts right after the table header.
131  */
132 #define RAS_INDEX_TO_OFFSET(_C, _N) ((_C)->ras_record_offset + \
133 				     (_N) * RAS_TABLE_RECORD_SIZE)
134 
135 #define RAS_OFFSET_TO_INDEX(_C, _O) (((_O) - \
136 				      (_C)->ras_record_offset) / RAS_TABLE_RECORD_SIZE)
137 
138 /* Given a 0-based relative record index, 0, 1, 2, ..., etc., off
139  * of "fri", return the absolute record index off of the end of
140  * the table header.
141  */
142 #define RAS_RI_TO_AI(_C, _I) (((_I) + (_C)->ras_fri) % \
143 			      (_C)->ras_max_record_count)
144 
145 #define RAS_NUM_RECS(_tbl_hdr)  (((_tbl_hdr)->tbl_size - \
146 				  RAS_TABLE_HEADER_SIZE) / RAS_TABLE_RECORD_SIZE)
147 
148 #define RAS_NUM_RECS_V2_1(_tbl_hdr)  (((_tbl_hdr)->tbl_size - \
149 				       RAS_TABLE_HEADER_SIZE - \
150 				       RAS_TABLE_V2_1_INFO_SIZE) / RAS_TABLE_RECORD_SIZE)
151 
152 #define to_amdgpu_device(x) ((container_of(x, struct amdgpu_ras, eeprom_control))->adev)
153 
154 static bool __is_ras_eeprom_supported(struct amdgpu_device *adev)
155 {
156 	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
157 	case IP_VERSION(11, 0, 2): /* VEGA20 and ARCTURUS */
158 	case IP_VERSION(11, 0, 7): /* Sienna cichlid */
159 	case IP_VERSION(13, 0, 0):
160 	case IP_VERSION(13, 0, 2): /* Aldebaran */
161 	case IP_VERSION(13, 0, 10):
162 		return true;
163 	case IP_VERSION(13, 0, 6):
164 	case IP_VERSION(13, 0, 12):
165 	case IP_VERSION(13, 0, 14):
166 		return (adev->gmc.is_app_apu) ? false : true;
167 	default:
168 		return false;
169 	}
170 }
171 
172 static bool __get_eeprom_i2c_addr(struct amdgpu_device *adev,
173 				  struct amdgpu_ras_eeprom_control *control)
174 {
175 	struct atom_context *atom_ctx = adev->mode_info.atom_context;
176 	u8 i2c_addr;
177 
178 	if (!control)
179 		return false;
180 
181 	if (adev->bios && amdgpu_atomfirmware_ras_rom_addr(adev, &i2c_addr)) {
182 		/* The address given by VBIOS is an 8-bit, wire-format
183 		 * address, i.e. the most significant byte.
184 		 *
185 		 * Normalize it to a 19-bit EEPROM address. Remove the
186 		 * device type identifier and make it a 7-bit address;
187 		 * then make it a 19-bit EEPROM address. See top of
188 		 * amdgpu_eeprom.c.
189 		 */
190 		i2c_addr = (i2c_addr & 0x0F) >> 1;
191 		control->i2c_address = ((u32) i2c_addr) << 16;
192 
193 		return true;
194 	}
195 
196 	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
197 	case IP_VERSION(11, 0, 2):
198 		/* VEGA20 and ARCTURUS */
199 		if (adev->asic_type == CHIP_VEGA20)
200 			control->i2c_address = EEPROM_I2C_MADDR_0;
201 		else if (strnstr(atom_ctx->vbios_pn,
202 				 "D342",
203 				 sizeof(atom_ctx->vbios_pn)))
204 			control->i2c_address = EEPROM_I2C_MADDR_0;
205 		else
206 			control->i2c_address = EEPROM_I2C_MADDR_4;
207 		return true;
208 	case IP_VERSION(11, 0, 7):
209 		control->i2c_address = EEPROM_I2C_MADDR_0;
210 		return true;
211 	case IP_VERSION(13, 0, 2):
212 		if (strnstr(atom_ctx->vbios_pn, "D673",
213 			    sizeof(atom_ctx->vbios_pn)))
214 			control->i2c_address = EEPROM_I2C_MADDR_4;
215 		else
216 			control->i2c_address = EEPROM_I2C_MADDR_0;
217 		return true;
218 	case IP_VERSION(13, 0, 0):
219 		if (strnstr(atom_ctx->vbios_pn, "D707",
220 			    sizeof(atom_ctx->vbios_pn)))
221 			control->i2c_address = EEPROM_I2C_MADDR_0;
222 		else
223 			control->i2c_address = EEPROM_I2C_MADDR_4;
224 		return true;
225 	case IP_VERSION(13, 0, 6):
226 	case IP_VERSION(13, 0, 10):
227 	case IP_VERSION(13, 0, 12):
228 	case IP_VERSION(13, 0, 14):
229 		control->i2c_address = EEPROM_I2C_MADDR_4;
230 		return true;
231 	default:
232 		return false;
233 	}
234 }
235 
236 static void
237 __encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header *hdr,
238 			     unsigned char *buf)
239 {
240 	u32 *pp = (uint32_t *)buf;
241 
242 	pp[0] = cpu_to_le32(hdr->header);
243 	pp[1] = cpu_to_le32(hdr->version);
244 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
245 	pp[3] = cpu_to_le32(hdr->tbl_size);
246 	pp[4] = cpu_to_le32(hdr->checksum);
247 }
248 
249 static void
250 __decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header *hdr,
251 			       unsigned char *buf)
252 {
253 	u32 *pp = (uint32_t *)buf;
254 
255 	hdr->header	      = le32_to_cpu(pp[0]);
256 	hdr->version	      = le32_to_cpu(pp[1]);
257 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
258 	hdr->tbl_size	      = le32_to_cpu(pp[3]);
259 	hdr->checksum	      = le32_to_cpu(pp[4]);
260 }
261 
262 static int __write_table_header(struct amdgpu_ras_eeprom_control *control)
263 {
264 	u8 buf[RAS_TABLE_HEADER_SIZE];
265 	struct amdgpu_device *adev = to_amdgpu_device(control);
266 	int res;
267 
268 	memset(buf, 0, sizeof(buf));
269 	__encode_table_header_to_buf(&control->tbl_hdr, buf);
270 
271 	/* i2c may be unstable in gpu reset */
272 	down_read(&adev->reset_domain->sem);
273 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
274 				  control->i2c_address +
275 				  control->ras_header_offset,
276 				  buf, RAS_TABLE_HEADER_SIZE);
277 	up_read(&adev->reset_domain->sem);
278 
279 	if (res < 0) {
280 		DRM_ERROR("Failed to write EEPROM table header:%d", res);
281 	} else if (res < RAS_TABLE_HEADER_SIZE) {
282 		DRM_ERROR("Short write:%d out of %d\n",
283 			  res, RAS_TABLE_HEADER_SIZE);
284 		res = -EIO;
285 	} else {
286 		res = 0;
287 	}
288 
289 	return res;
290 }
291 
292 static void
293 __encode_table_ras_info_to_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
294 			       unsigned char *buf)
295 {
296 	u32 *pp = (uint32_t *)buf;
297 	u32 tmp;
298 
299 	tmp = ((uint32_t)(rai->rma_status) & 0xFF) |
300 	      (((uint32_t)(rai->health_percent) << 8) & 0xFF00) |
301 	      (((uint32_t)(rai->ecc_page_threshold) << 16) & 0xFFFF0000);
302 	pp[0] = cpu_to_le32(tmp);
303 }
304 
305 static void
306 __decode_table_ras_info_from_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
307 				 unsigned char *buf)
308 {
309 	u32 *pp = (uint32_t *)buf;
310 	u32 tmp;
311 
312 	tmp = le32_to_cpu(pp[0]);
313 	rai->rma_status = tmp & 0xFF;
314 	rai->health_percent = (tmp >> 8) & 0xFF;
315 	rai->ecc_page_threshold = (tmp >> 16) & 0xFFFF;
316 }
317 
318 static int __write_table_ras_info(struct amdgpu_ras_eeprom_control *control)
319 {
320 	struct amdgpu_device *adev = to_amdgpu_device(control);
321 	u8 *buf;
322 	int res;
323 
324 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
325 	if (!buf) {
326 		DRM_ERROR("Failed to alloc buf to write table ras info\n");
327 		return -ENOMEM;
328 	}
329 
330 	__encode_table_ras_info_to_buf(&control->tbl_rai, buf);
331 
332 	/* i2c may be unstable in gpu reset */
333 	down_read(&adev->reset_domain->sem);
334 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
335 				  control->i2c_address +
336 				  control->ras_info_offset,
337 				  buf, RAS_TABLE_V2_1_INFO_SIZE);
338 	up_read(&adev->reset_domain->sem);
339 
340 	if (res < 0) {
341 		DRM_ERROR("Failed to write EEPROM table ras info:%d", res);
342 	} else if (res < RAS_TABLE_V2_1_INFO_SIZE) {
343 		DRM_ERROR("Short write:%d out of %d\n",
344 			  res, RAS_TABLE_V2_1_INFO_SIZE);
345 		res = -EIO;
346 	} else {
347 		res = 0;
348 	}
349 
350 	kfree(buf);
351 
352 	return res;
353 }
354 
355 static u8 __calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control *control)
356 {
357 	int ii;
358 	u8  *pp, csum;
359 	size_t sz;
360 
361 	/* Header checksum, skip checksum field in the calculation */
362 	sz = sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum);
363 	pp = (u8 *) &control->tbl_hdr;
364 	csum = 0;
365 	for (ii = 0; ii < sz; ii++, pp++)
366 		csum += *pp;
367 
368 	return csum;
369 }
370 
371 static u8 __calc_ras_info_byte_sum(const struct amdgpu_ras_eeprom_control *control)
372 {
373 	int ii;
374 	u8  *pp, csum;
375 	size_t sz;
376 
377 	sz = sizeof(control->tbl_rai);
378 	pp = (u8 *) &control->tbl_rai;
379 	csum = 0;
380 	for (ii = 0; ii < sz; ii++, pp++)
381 		csum += *pp;
382 
383 	return csum;
384 }
385 
386 static int amdgpu_ras_eeprom_correct_header_tag(
387 	struct amdgpu_ras_eeprom_control *control,
388 	uint32_t header)
389 {
390 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
391 	u8 *hh;
392 	int res;
393 	u8 csum;
394 
395 	csum = -hdr->checksum;
396 
397 	hh = (void *) &hdr->header;
398 	csum -= (hh[0] + hh[1] + hh[2] + hh[3]);
399 	hh = (void *) &header;
400 	csum += hh[0] + hh[1] + hh[2] + hh[3];
401 	csum = -csum;
402 	mutex_lock(&control->ras_tbl_mutex);
403 	hdr->header = header;
404 	hdr->checksum = csum;
405 	res = __write_table_header(control);
406 	mutex_unlock(&control->ras_tbl_mutex);
407 
408 	return res;
409 }
410 
411 static void amdgpu_ras_set_eeprom_table_version(struct amdgpu_ras_eeprom_control *control)
412 {
413 	struct amdgpu_device *adev = to_amdgpu_device(control);
414 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
415 
416 	switch (amdgpu_ip_version(adev, UMC_HWIP, 0)) {
417 	case IP_VERSION(8, 10, 0):
418 		hdr->version = RAS_TABLE_VER_V2_1;
419 		return;
420 	case IP_VERSION(12, 0, 0):
421 		hdr->version = RAS_TABLE_VER_V3;
422 		return;
423 	default:
424 		hdr->version = RAS_TABLE_VER_V1;
425 		return;
426 	}
427 }
428 
429 /**
430  * amdgpu_ras_eeprom_reset_table -- Reset the RAS EEPROM table
431  * @control: pointer to control structure
432  *
433  * Reset the contents of the header of the RAS EEPROM table.
434  * Return 0 on success, -errno on error.
435  */
436 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
437 {
438 	struct amdgpu_device *adev = to_amdgpu_device(control);
439 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
440 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
441 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
442 	u8 csum;
443 	int res;
444 
445 	mutex_lock(&control->ras_tbl_mutex);
446 
447 	hdr->header = RAS_TABLE_HDR_VAL;
448 	amdgpu_ras_set_eeprom_table_version(control);
449 
450 	if (hdr->version >= RAS_TABLE_VER_V2_1) {
451 		hdr->first_rec_offset = RAS_RECORD_START_V2_1;
452 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE +
453 				RAS_TABLE_V2_1_INFO_SIZE;
454 		rai->rma_status = GPU_HEALTH_USABLE;
455 		/**
456 		 * GPU health represented as a percentage.
457 		 * 0 means worst health, 100 means fully health.
458 		 */
459 		rai->health_percent = 100;
460 		/* ecc_page_threshold = 0 means disable bad page retirement */
461 		rai->ecc_page_threshold = con->bad_page_cnt_threshold;
462 	} else {
463 		hdr->first_rec_offset = RAS_RECORD_START;
464 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE;
465 	}
466 
467 	csum = __calc_hdr_byte_sum(control);
468 	if (hdr->version >= RAS_TABLE_VER_V2_1)
469 		csum += __calc_ras_info_byte_sum(control);
470 	csum = -csum;
471 	hdr->checksum = csum;
472 	res = __write_table_header(control);
473 	if (!res && hdr->version > RAS_TABLE_VER_V1)
474 		res = __write_table_ras_info(control);
475 
476 	control->ras_num_recs = 0;
477 	control->ras_num_bad_pages = 0;
478 	control->ras_fri = 0;
479 
480 	amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_bad_pages);
481 
482 	control->bad_channel_bitmap = 0;
483 	amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap);
484 	con->update_channel_flag = false;
485 
486 	amdgpu_ras_debugfs_set_ret_size(control);
487 
488 	mutex_unlock(&control->ras_tbl_mutex);
489 
490 	return res;
491 }
492 
493 static void
494 __encode_table_record_to_buf(struct amdgpu_ras_eeprom_control *control,
495 			     struct eeprom_table_record *record,
496 			     unsigned char *buf)
497 {
498 	__le64 tmp = 0;
499 	int i = 0;
500 
501 	/* Next are all record fields according to EEPROM page spec in LE foramt */
502 	buf[i++] = record->err_type;
503 
504 	buf[i++] = record->bank;
505 
506 	tmp = cpu_to_le64(record->ts);
507 	memcpy(buf + i, &tmp, 8);
508 	i += 8;
509 
510 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
511 	memcpy(buf + i, &tmp, 6);
512 	i += 6;
513 
514 	buf[i++] = record->mem_channel;
515 	buf[i++] = record->mcumc_id;
516 
517 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
518 	memcpy(buf + i, &tmp, 6);
519 }
520 
521 static void
522 __decode_table_record_from_buf(struct amdgpu_ras_eeprom_control *control,
523 			       struct eeprom_table_record *record,
524 			       unsigned char *buf)
525 {
526 	__le64 tmp = 0;
527 	int i =  0;
528 
529 	/* Next are all record fields according to EEPROM page spec in LE foramt */
530 	record->err_type = buf[i++];
531 
532 	record->bank = buf[i++];
533 
534 	memcpy(&tmp, buf + i, 8);
535 	record->ts = le64_to_cpu(tmp);
536 	i += 8;
537 
538 	memcpy(&tmp, buf + i, 6);
539 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
540 	i += 6;
541 
542 	record->mem_channel = buf[i++];
543 	record->mcumc_id = buf[i++];
544 
545 	memcpy(&tmp, buf + i,  6);
546 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
547 }
548 
549 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
550 {
551 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
552 
553 	if (!__is_ras_eeprom_supported(adev) ||
554 	    !amdgpu_bad_page_threshold)
555 		return false;
556 
557 	/* skip check eeprom table for VEGA20 Gaming */
558 	if (!con)
559 		return false;
560 	else
561 		if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
562 			return false;
563 
564 	if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
565 		if (con->eeprom_control.ras_num_bad_pages > con->bad_page_cnt_threshold)
566 			dev_warn(adev->dev, "RAS records:%d exceed threshold:%d",
567 				 con->eeprom_control.ras_num_bad_pages, con->bad_page_cnt_threshold);
568 		if ((amdgpu_bad_page_threshold == -1) ||
569 		    (amdgpu_bad_page_threshold == -2)) {
570 			dev_warn(adev->dev,
571 				 "Please consult AMD Service Action Guide (SAG) for appropriate service procedures.\n");
572 			return false;
573 		} else {
574 			dev_warn(adev->dev,
575 				 "Please consider adjusting the customized threshold.\n");
576 			return true;
577 		}
578 	}
579 
580 	return false;
581 }
582 
583 /**
584  * __amdgpu_ras_eeprom_write -- write indexed from buffer to EEPROM
585  * @control: pointer to control structure
586  * @buf: pointer to buffer containing data to write
587  * @fri: start writing at this index
588  * @num: number of records to write
589  *
590  * The caller must hold the table mutex in @control.
591  * Return 0 on success, -errno otherwise.
592  */
593 static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control,
594 				     u8 *buf, const u32 fri, const u32 num)
595 {
596 	struct amdgpu_device *adev = to_amdgpu_device(control);
597 	u32 buf_size;
598 	int res;
599 
600 	/* i2c may be unstable in gpu reset */
601 	down_read(&adev->reset_domain->sem);
602 	buf_size = num * RAS_TABLE_RECORD_SIZE;
603 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
604 				  control->i2c_address +
605 				  RAS_INDEX_TO_OFFSET(control, fri),
606 				  buf, buf_size);
607 	up_read(&adev->reset_domain->sem);
608 	if (res < 0) {
609 		DRM_ERROR("Writing %d EEPROM table records error:%d",
610 			  num, res);
611 	} else if (res < buf_size) {
612 		/* Short write, return error.
613 		 */
614 		DRM_ERROR("Wrote %d records out of %d",
615 			  res / RAS_TABLE_RECORD_SIZE, num);
616 		res = -EIO;
617 	} else {
618 		res = 0;
619 	}
620 
621 	return res;
622 }
623 
624 static int
625 amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
626 			       struct eeprom_table_record *record,
627 			       const u32 num)
628 {
629 	struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
630 	struct amdgpu_device *adev = to_amdgpu_device(control);
631 	u32 a, b, i;
632 	u8 *buf, *pp;
633 	int res;
634 
635 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
636 	if (!buf)
637 		return -ENOMEM;
638 
639 	/* Encode all of them in one go.
640 	 */
641 	pp = buf;
642 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
643 		__encode_table_record_to_buf(control, &record[i], pp);
644 
645 		/* update bad channel bitmap */
646 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
647 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
648 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
649 			con->update_channel_flag = true;
650 		}
651 	}
652 
653 	/* a, first record index to write into.
654 	 * b, last record index to write into.
655 	 * a = first index to read (fri) + number of records in the table,
656 	 * b = a + @num - 1.
657 	 * Let N = control->ras_max_num_record_count, then we have,
658 	 * case 0: 0 <= a <= b < N,
659 	 *   just append @num records starting at a;
660 	 * case 1: 0 <= a < N <= b,
661 	 *   append (N - a) records starting at a, and
662 	 *   append the remainder,  b % N + 1, starting at 0.
663 	 * case 2: 0 <= fri < N <= a <= b, then modulo N we get two subcases,
664 	 * case 2a: 0 <= a <= b < N
665 	 *   append num records starting at a; and fix fri if b overwrote it,
666 	 *   and since a <= b, if b overwrote it then a must've also,
667 	 *   and if b didn't overwrite it, then a didn't also.
668 	 * case 2b: 0 <= b < a < N
669 	 *   write num records starting at a, which wraps around 0=N
670 	 *   and overwrite fri unconditionally. Now from case 2a,
671 	 *   this means that b eclipsed fri to overwrite it and wrap
672 	 *   around 0 again, i.e. b = 2N+r pre modulo N, so we unconditionally
673 	 *   set fri = b + 1 (mod N).
674 	 * Now, since fri is updated in every case, except the trivial case 0,
675 	 * the number of records present in the table after writing, is,
676 	 * num_recs - 1 = b - fri (mod N), and we take the positive value,
677 	 * by adding an arbitrary multiple of N before taking the modulo N
678 	 * as shown below.
679 	 */
680 	a = control->ras_fri + control->ras_num_recs;
681 	b = a + num  - 1;
682 	if (b < control->ras_max_record_count) {
683 		res = __amdgpu_ras_eeprom_write(control, buf, a, num);
684 	} else if (a < control->ras_max_record_count) {
685 		u32 g0, g1;
686 
687 		g0 = control->ras_max_record_count - a;
688 		g1 = b % control->ras_max_record_count + 1;
689 		res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
690 		if (res)
691 			goto Out;
692 		res = __amdgpu_ras_eeprom_write(control,
693 						buf + g0 * RAS_TABLE_RECORD_SIZE,
694 						0, g1);
695 		if (res)
696 			goto Out;
697 		if (g1 > control->ras_fri)
698 			control->ras_fri = g1 % control->ras_max_record_count;
699 	} else {
700 		a %= control->ras_max_record_count;
701 		b %= control->ras_max_record_count;
702 
703 		if (a <= b) {
704 			/* Note that, b - a + 1 = num. */
705 			res = __amdgpu_ras_eeprom_write(control, buf, a, num);
706 			if (res)
707 				goto Out;
708 			if (b >= control->ras_fri)
709 				control->ras_fri = (b + 1) % control->ras_max_record_count;
710 		} else {
711 			u32 g0, g1;
712 
713 			/* b < a, which means, we write from
714 			 * a to the end of the table, and from
715 			 * the start of the table to b.
716 			 */
717 			g0 = control->ras_max_record_count - a;
718 			g1 = b + 1;
719 			res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
720 			if (res)
721 				goto Out;
722 			res = __amdgpu_ras_eeprom_write(control,
723 							buf + g0 * RAS_TABLE_RECORD_SIZE,
724 							0, g1);
725 			if (res)
726 				goto Out;
727 			control->ras_fri = g1 % control->ras_max_record_count;
728 		}
729 	}
730 	control->ras_num_recs = 1 + (control->ras_max_record_count + b
731 				     - control->ras_fri)
732 		% control->ras_max_record_count;
733 
734 	/*old asics only save pa to eeprom like before*/
735 	if (IP_VERSION_MAJ(amdgpu_ip_version(adev, UMC_HWIP, 0)) < 12)
736 		control->ras_num_pa_recs += num;
737 	else
738 		control->ras_num_mca_recs += num;
739 
740 	control->ras_num_bad_pages = control->ras_num_pa_recs +
741 				control->ras_num_mca_recs * adev->umc.retire_unit;
742 Out:
743 	kfree(buf);
744 	return res;
745 }
746 
747 static int
748 amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
749 {
750 	struct amdgpu_device *adev = to_amdgpu_device(control);
751 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
752 	u8 *buf, *pp, csum;
753 	u32 buf_size;
754 	int res;
755 
756 	/* Modify the header if it exceeds.
757 	 */
758 	if (amdgpu_bad_page_threshold != 0 &&
759 	    control->ras_num_bad_pages > ras->bad_page_cnt_threshold) {
760 		dev_warn(adev->dev,
761 			"Saved bad pages %d reaches threshold value %d\n",
762 			control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
763 		control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
764 		if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1) {
765 			control->tbl_rai.rma_status = GPU_RETIRED__ECC_REACH_THRESHOLD;
766 			control->tbl_rai.health_percent = 0;
767 		}
768 
769 		if ((amdgpu_bad_page_threshold != -1) &&
770 		    (amdgpu_bad_page_threshold != -2))
771 			ras->is_rma = true;
772 
773 		/* ignore the -ENOTSUPP return value */
774 		amdgpu_dpm_send_rma_reason(adev);
775 	}
776 
777 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
778 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
779 					    RAS_TABLE_V2_1_INFO_SIZE +
780 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
781 	else
782 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
783 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
784 	control->tbl_hdr.checksum = 0;
785 
786 	buf_size = control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
787 	buf = kcalloc(control->ras_num_recs, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
788 	if (!buf) {
789 		DRM_ERROR("allocating memory for table of size %d bytes failed\n",
790 			  control->tbl_hdr.tbl_size);
791 		res = -ENOMEM;
792 		goto Out;
793 	}
794 
795 	down_read(&adev->reset_domain->sem);
796 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
797 				 control->i2c_address +
798 				 control->ras_record_offset,
799 				 buf, buf_size);
800 	up_read(&adev->reset_domain->sem);
801 	if (res < 0) {
802 		DRM_ERROR("EEPROM failed reading records:%d\n",
803 			  res);
804 		goto Out;
805 	} else if (res < buf_size) {
806 		DRM_ERROR("EEPROM read %d out of %d bytes\n",
807 			  res, buf_size);
808 		res = -EIO;
809 		goto Out;
810 	}
811 
812 	/**
813 	 * bad page records have been stored in eeprom,
814 	 * now calculate gpu health percent
815 	 */
816 	if (amdgpu_bad_page_threshold != 0 &&
817 	    control->tbl_hdr.version >= RAS_TABLE_VER_V2_1 &&
818 	    control->ras_num_bad_pages <= ras->bad_page_cnt_threshold)
819 		control->tbl_rai.health_percent = ((ras->bad_page_cnt_threshold -
820 						   control->ras_num_bad_pages) * 100) /
821 						   ras->bad_page_cnt_threshold;
822 
823 	/* Recalc the checksum.
824 	 */
825 	csum = 0;
826 	for (pp = buf; pp < buf + buf_size; pp++)
827 		csum += *pp;
828 
829 	csum += __calc_hdr_byte_sum(control);
830 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
831 		csum += __calc_ras_info_byte_sum(control);
832 	/* avoid sign extension when assigning to "checksum" */
833 	csum = -csum;
834 	control->tbl_hdr.checksum = csum;
835 	res = __write_table_header(control);
836 	if (!res && control->tbl_hdr.version > RAS_TABLE_VER_V1)
837 		res = __write_table_ras_info(control);
838 Out:
839 	kfree(buf);
840 	return res;
841 }
842 
843 /**
844  * amdgpu_ras_eeprom_append -- append records to the EEPROM RAS table
845  * @control: pointer to control structure
846  * @record: array of records to append
847  * @num: number of records in @record array
848  *
849  * Append @num records to the table, calculate the checksum and write
850  * the table back to EEPROM. The maximum number of records that
851  * can be appended is between 1 and control->ras_max_record_count,
852  * regardless of how many records are already stored in the table.
853  *
854  * Return 0 on success or if EEPROM is not supported, -errno on error.
855  */
856 int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
857 			     struct eeprom_table_record *record,
858 			     const u32 num)
859 {
860 	struct amdgpu_device *adev = to_amdgpu_device(control);
861 	int res, i;
862 	uint64_t nps = AMDGPU_NPS1_PARTITION_MODE;
863 
864 	if (!__is_ras_eeprom_supported(adev))
865 		return 0;
866 
867 	if (num == 0) {
868 		DRM_ERROR("will not append 0 records\n");
869 		return -EINVAL;
870 	} else if (num > control->ras_max_record_count) {
871 		DRM_ERROR("cannot append %d records than the size of table %d\n",
872 			  num, control->ras_max_record_count);
873 		return -EINVAL;
874 	}
875 
876 	if (adev->gmc.gmc_funcs->query_mem_partition_mode)
877 		nps = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
878 
879 	/* set the new channel index flag */
880 	for (i = 0; i < num; i++)
881 		record[i].retired_page |= (nps << UMC_NPS_SHIFT);
882 
883 	mutex_lock(&control->ras_tbl_mutex);
884 
885 	res = amdgpu_ras_eeprom_append_table(control, record, num);
886 	if (!res)
887 		res = amdgpu_ras_eeprom_update_header(control);
888 	if (!res)
889 		amdgpu_ras_debugfs_set_ret_size(control);
890 
891 	mutex_unlock(&control->ras_tbl_mutex);
892 
893 	/* clear channel index flag, the flag is only saved on eeprom */
894 	for (i = 0; i < num; i++)
895 		record[i].retired_page &= ~(nps << UMC_NPS_SHIFT);
896 
897 	return res;
898 }
899 
900 /**
901  * __amdgpu_ras_eeprom_read -- read indexed from EEPROM into buffer
902  * @control: pointer to control structure
903  * @buf: pointer to buffer to read into
904  * @fri: first record index, start reading at this index, absolute index
905  * @num: number of records to read
906  *
907  * The caller must hold the table mutex in @control.
908  * Return 0 on success, -errno otherwise.
909  */
910 static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
911 				    u8 *buf, const u32 fri, const u32 num)
912 {
913 	struct amdgpu_device *adev = to_amdgpu_device(control);
914 	u32 buf_size;
915 	int res;
916 
917 	/* i2c may be unstable in gpu reset */
918 	down_read(&adev->reset_domain->sem);
919 	buf_size = num * RAS_TABLE_RECORD_SIZE;
920 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
921 				 control->i2c_address +
922 				 RAS_INDEX_TO_OFFSET(control, fri),
923 				 buf, buf_size);
924 	up_read(&adev->reset_domain->sem);
925 	if (res < 0) {
926 		DRM_ERROR("Reading %d EEPROM table records error:%d",
927 			  num, res);
928 	} else if (res < buf_size) {
929 		/* Short read, return error.
930 		 */
931 		DRM_ERROR("Read %d records out of %d",
932 			  res / RAS_TABLE_RECORD_SIZE, num);
933 		res = -EIO;
934 	} else {
935 		res = 0;
936 	}
937 
938 	return res;
939 }
940 
941 /**
942  * amdgpu_ras_eeprom_read -- read EEPROM
943  * @control: pointer to control structure
944  * @record: array of records to read into
945  * @num: number of records in @record
946  *
947  * Reads num records from the RAS table in EEPROM and
948  * writes the data into @record array.
949  *
950  * Returns 0 on success, -errno on error.
951  */
952 int amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
953 			   struct eeprom_table_record *record,
954 			   const u32 num)
955 {
956 	struct amdgpu_device *adev = to_amdgpu_device(control);
957 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
958 	int i, res;
959 	u8 *buf, *pp;
960 	u32 g0, g1;
961 
962 	if (!__is_ras_eeprom_supported(adev))
963 		return 0;
964 
965 	if (num == 0) {
966 		DRM_ERROR("will not read 0 records\n");
967 		return -EINVAL;
968 	} else if (num > control->ras_num_recs) {
969 		DRM_ERROR("too many records to read:%d available:%d\n",
970 			  num, control->ras_num_recs);
971 		return -EINVAL;
972 	}
973 
974 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
975 	if (!buf)
976 		return -ENOMEM;
977 
978 	/* Determine how many records to read, from the first record
979 	 * index, fri, to the end of the table, and from the beginning
980 	 * of the table, such that the total number of records is
981 	 * @num, and we handle wrap around when fri > 0 and
982 	 * fri + num > RAS_MAX_RECORD_COUNT.
983 	 *
984 	 * First we compute the index of the last element
985 	 * which would be fetched from each region,
986 	 * g0 is in [fri, fri + num - 1], and
987 	 * g1 is in [0, RAS_MAX_RECORD_COUNT - 1].
988 	 * Then, if g0 < RAS_MAX_RECORD_COUNT, the index of
989 	 * the last element to fetch, we set g0 to _the number_
990 	 * of elements to fetch, @num, since we know that the last
991 	 * indexed to be fetched does not exceed the table.
992 	 *
993 	 * If, however, g0 >= RAS_MAX_RECORD_COUNT, then
994 	 * we set g0 to the number of elements to read
995 	 * until the end of the table, and g1 to the number of
996 	 * elements to read from the beginning of the table.
997 	 */
998 	g0 = control->ras_fri + num - 1;
999 	g1 = g0 % control->ras_max_record_count;
1000 	if (g0 < control->ras_max_record_count) {
1001 		g0 = num;
1002 		g1 = 0;
1003 	} else {
1004 		g0 = control->ras_max_record_count - control->ras_fri;
1005 		g1 += 1;
1006 	}
1007 
1008 	mutex_lock(&control->ras_tbl_mutex);
1009 	res = __amdgpu_ras_eeprom_read(control, buf, control->ras_fri, g0);
1010 	if (res)
1011 		goto Out;
1012 	if (g1) {
1013 		res = __amdgpu_ras_eeprom_read(control,
1014 					       buf + g0 * RAS_TABLE_RECORD_SIZE,
1015 					       0, g1);
1016 		if (res)
1017 			goto Out;
1018 	}
1019 
1020 	res = 0;
1021 
1022 	/* Read up everything? Then transform.
1023 	 */
1024 	pp = buf;
1025 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
1026 		__decode_table_record_from_buf(control, &record[i], pp);
1027 
1028 		/* update bad channel bitmap */
1029 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
1030 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
1031 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
1032 			con->update_channel_flag = true;
1033 		}
1034 	}
1035 Out:
1036 	kfree(buf);
1037 	mutex_unlock(&control->ras_tbl_mutex);
1038 
1039 	return res;
1040 }
1041 
1042 uint32_t amdgpu_ras_eeprom_max_record_count(struct amdgpu_ras_eeprom_control *control)
1043 {
1044 	/* get available eeprom table version first before eeprom table init */
1045 	amdgpu_ras_set_eeprom_table_version(control);
1046 
1047 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
1048 		return RAS_MAX_RECORD_COUNT_V2_1;
1049 	else
1050 		return RAS_MAX_RECORD_COUNT;
1051 }
1052 
1053 static ssize_t
1054 amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
1055 				    size_t size, loff_t *pos)
1056 {
1057 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1058 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1059 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1060 	u8 data[50];
1061 	int res;
1062 
1063 	if (!size)
1064 		return size;
1065 
1066 	if (!ras || !control) {
1067 		res = snprintf(data, sizeof(data), "Not supported\n");
1068 	} else {
1069 		res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
1070 			       RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
1071 	}
1072 
1073 	if (*pos >= res)
1074 		return 0;
1075 
1076 	res -= *pos;
1077 	res = min_t(size_t, res, size);
1078 
1079 	if (copy_to_user(buf, &data[*pos], res))
1080 		return -EFAULT;
1081 
1082 	*pos += res;
1083 
1084 	return res;
1085 }
1086 
1087 const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
1088 	.owner = THIS_MODULE,
1089 	.read = amdgpu_ras_debugfs_eeprom_size_read,
1090 	.write = NULL,
1091 	.llseek = default_llseek,
1092 };
1093 
1094 static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
1095 static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
1096 #define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
1097 static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
1098 static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
1099 #define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
1100 
1101 static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
1102 	"ignore",
1103 	"re",
1104 	"ue",
1105 };
1106 
1107 static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
1108 {
1109 	return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
1110 		strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
1111 }
1112 
1113 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
1114 {
1115 	struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
1116 					      eeprom_control);
1117 	struct dentry *de = ras->de_ras_eeprom_table;
1118 
1119 	if (de)
1120 		d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
1121 }
1122 
1123 static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
1124 					     size_t size, loff_t *pos)
1125 {
1126 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1127 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1128 	struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
1129 	const size_t orig_size = size;
1130 	int res = -EFAULT;
1131 	size_t data_len;
1132 
1133 	mutex_lock(&control->ras_tbl_mutex);
1134 
1135 	/* We want *pos - data_len > 0, which means there's
1136 	 * bytes to be printed from data.
1137 	 */
1138 	data_len = strlen(tbl_hdr_str);
1139 	if (*pos < data_len) {
1140 		data_len -= *pos;
1141 		data_len = min_t(size_t, data_len, size);
1142 		if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
1143 			goto Out;
1144 		buf += data_len;
1145 		size -= data_len;
1146 		*pos += data_len;
1147 	}
1148 
1149 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
1150 	if (*pos < data_len && size > 0) {
1151 		u8 data[tbl_hdr_fmt_size + 1];
1152 		loff_t lpos;
1153 
1154 		snprintf(data, sizeof(data), tbl_hdr_fmt,
1155 			 control->tbl_hdr.header,
1156 			 control->tbl_hdr.version,
1157 			 control->tbl_hdr.first_rec_offset,
1158 			 control->tbl_hdr.tbl_size,
1159 			 control->tbl_hdr.checksum);
1160 
1161 		data_len -= *pos;
1162 		data_len = min_t(size_t, data_len, size);
1163 		lpos = *pos - strlen(tbl_hdr_str);
1164 		if (copy_to_user(buf, &data[lpos], data_len))
1165 			goto Out;
1166 		buf += data_len;
1167 		size -= data_len;
1168 		*pos += data_len;
1169 	}
1170 
1171 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
1172 	if (*pos < data_len && size > 0) {
1173 		loff_t lpos;
1174 
1175 		data_len -= *pos;
1176 		data_len = min_t(size_t, data_len, size);
1177 		lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
1178 		if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
1179 			goto Out;
1180 		buf += data_len;
1181 		size -= data_len;
1182 		*pos += data_len;
1183 	}
1184 
1185 	data_len = amdgpu_ras_debugfs_table_size(control);
1186 	if (*pos < data_len && size > 0) {
1187 		u8 dare[RAS_TABLE_RECORD_SIZE];
1188 		u8 data[rec_hdr_fmt_size + 1];
1189 		struct eeprom_table_record record;
1190 		int s, r;
1191 
1192 		/* Find the starting record index
1193 		 */
1194 		s = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1195 			strlen(rec_hdr_str);
1196 		s = s / rec_hdr_fmt_size;
1197 		r = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1198 			strlen(rec_hdr_str);
1199 		r = r % rec_hdr_fmt_size;
1200 
1201 		for ( ; size > 0 && s < control->ras_num_recs; s++) {
1202 			u32 ai = RAS_RI_TO_AI(control, s);
1203 			/* Read a single record
1204 			 */
1205 			res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
1206 			if (res)
1207 				goto Out;
1208 			__decode_table_record_from_buf(control, &record, dare);
1209 			snprintf(data, sizeof(data), rec_hdr_fmt,
1210 				 s,
1211 				 RAS_INDEX_TO_OFFSET(control, ai),
1212 				 record_err_type_str[record.err_type],
1213 				 record.bank,
1214 				 record.ts,
1215 				 record.offset,
1216 				 record.mem_channel,
1217 				 record.mcumc_id,
1218 				 record.retired_page);
1219 
1220 			data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
1221 			if (copy_to_user(buf, &data[r], data_len)) {
1222 				res = -EFAULT;
1223 				goto Out;
1224 			}
1225 			buf += data_len;
1226 			size -= data_len;
1227 			*pos += data_len;
1228 			r = 0;
1229 		}
1230 	}
1231 	res = 0;
1232 Out:
1233 	mutex_unlock(&control->ras_tbl_mutex);
1234 	return res < 0 ? res : orig_size - size;
1235 }
1236 
1237 static ssize_t
1238 amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
1239 				     size_t size, loff_t *pos)
1240 {
1241 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1242 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1243 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1244 	u8 data[81];
1245 	int res;
1246 
1247 	if (!size)
1248 		return size;
1249 
1250 	if (!ras || !control) {
1251 		res = snprintf(data, sizeof(data), "Not supported\n");
1252 		if (*pos >= res)
1253 			return 0;
1254 
1255 		res -= *pos;
1256 		res = min_t(size_t, res, size);
1257 
1258 		if (copy_to_user(buf, &data[*pos], res))
1259 			return -EFAULT;
1260 
1261 		*pos += res;
1262 
1263 		return res;
1264 	} else {
1265 		return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
1266 	}
1267 }
1268 
1269 const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
1270 	.owner = THIS_MODULE,
1271 	.read = amdgpu_ras_debugfs_eeprom_table_read,
1272 	.write = NULL,
1273 	.llseek = default_llseek,
1274 };
1275 
1276 /**
1277  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
1278  * @control: pointer to control structure
1279  *
1280  * Check the checksum of the stored in EEPROM RAS table.
1281  *
1282  * Return 0 if the checksum is correct,
1283  * positive if it is not correct, and
1284  * -errno on I/O error.
1285  */
1286 static int __verify_ras_table_checksum(struct amdgpu_ras_eeprom_control *control)
1287 {
1288 	struct amdgpu_device *adev = to_amdgpu_device(control);
1289 	int buf_size, res;
1290 	u8  csum, *buf, *pp;
1291 
1292 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
1293 		buf_size = RAS_TABLE_HEADER_SIZE +
1294 			   RAS_TABLE_V2_1_INFO_SIZE +
1295 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1296 	else
1297 		buf_size = RAS_TABLE_HEADER_SIZE +
1298 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1299 
1300 	buf = kzalloc(buf_size, GFP_KERNEL);
1301 	if (!buf) {
1302 		DRM_ERROR("Out of memory checking RAS table checksum.\n");
1303 		return -ENOMEM;
1304 	}
1305 
1306 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1307 				 control->i2c_address +
1308 				 control->ras_header_offset,
1309 				 buf, buf_size);
1310 	if (res < buf_size) {
1311 		DRM_ERROR("Partial read for checksum, res:%d\n", res);
1312 		/* On partial reads, return -EIO.
1313 		 */
1314 		if (res >= 0)
1315 			res = -EIO;
1316 		goto Out;
1317 	}
1318 
1319 	csum = 0;
1320 	for (pp = buf; pp < buf + buf_size; pp++)
1321 		csum += *pp;
1322 Out:
1323 	kfree(buf);
1324 	return res < 0 ? res : csum;
1325 }
1326 
1327 static int __read_table_ras_info(struct amdgpu_ras_eeprom_control *control)
1328 {
1329 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
1330 	struct amdgpu_device *adev = to_amdgpu_device(control);
1331 	unsigned char *buf;
1332 	int res;
1333 
1334 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
1335 	if (!buf) {
1336 		DRM_ERROR("Failed to alloc buf to read EEPROM table ras info\n");
1337 		return -ENOMEM;
1338 	}
1339 
1340 	/**
1341 	 * EEPROM table V2_1 supports ras info,
1342 	 * read EEPROM table ras info
1343 	 */
1344 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1345 				 control->i2c_address + control->ras_info_offset,
1346 				 buf, RAS_TABLE_V2_1_INFO_SIZE);
1347 	if (res < RAS_TABLE_V2_1_INFO_SIZE) {
1348 		DRM_ERROR("Failed to read EEPROM table ras info, res:%d", res);
1349 		res = res >= 0 ? -EIO : res;
1350 		goto Out;
1351 	}
1352 
1353 	__decode_table_ras_info_from_buf(rai, buf);
1354 
1355 Out:
1356 	kfree(buf);
1357 	return res == RAS_TABLE_V2_1_INFO_SIZE ? 0 : res;
1358 }
1359 
1360 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control)
1361 {
1362 	struct amdgpu_device *adev = to_amdgpu_device(control);
1363 	unsigned char buf[RAS_TABLE_HEADER_SIZE] = { 0 };
1364 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1365 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1366 	int res;
1367 
1368 	ras->is_rma = false;
1369 
1370 	if (!__is_ras_eeprom_supported(adev))
1371 		return 0;
1372 
1373 	/* Verify i2c adapter is initialized */
1374 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1375 		return -ENOENT;
1376 
1377 	if (!__get_eeprom_i2c_addr(adev, control))
1378 		return -EINVAL;
1379 
1380 	control->ras_header_offset = RAS_HDR_START;
1381 	control->ras_info_offset = RAS_TABLE_V2_1_INFO_START;
1382 	mutex_init(&control->ras_tbl_mutex);
1383 
1384 	/* Read the table header from EEPROM address */
1385 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1386 				 control->i2c_address + control->ras_header_offset,
1387 				 buf, RAS_TABLE_HEADER_SIZE);
1388 	if (res < RAS_TABLE_HEADER_SIZE) {
1389 		DRM_ERROR("Failed to read EEPROM table header, res:%d", res);
1390 		return res >= 0 ? -EIO : res;
1391 	}
1392 
1393 	__decode_table_header_from_buf(hdr, buf);
1394 
1395 	if (hdr->version >= RAS_TABLE_VER_V2_1) {
1396 		control->ras_num_recs = RAS_NUM_RECS_V2_1(hdr);
1397 		control->ras_record_offset = RAS_RECORD_START_V2_1;
1398 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT_V2_1;
1399 	} else {
1400 		control->ras_num_recs = RAS_NUM_RECS(hdr);
1401 		control->ras_record_offset = RAS_RECORD_START;
1402 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT;
1403 	}
1404 	control->ras_fri = RAS_OFFSET_TO_INDEX(control, hdr->first_rec_offset);
1405 
1406 	control->ras_num_mca_recs = 0;
1407 	control->ras_num_pa_recs = 0;
1408 	return 0;
1409 }
1410 
1411 int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)
1412 {
1413 	struct amdgpu_device *adev = to_amdgpu_device(control);
1414 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1415 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1416 	int res;
1417 
1418 	if (!__is_ras_eeprom_supported(adev))
1419 		return 0;
1420 
1421 	/* Verify i2c adapter is initialized */
1422 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1423 		return -ENOENT;
1424 
1425 	if (!__get_eeprom_i2c_addr(adev, control))
1426 		return -EINVAL;
1427 
1428 	control->ras_num_bad_pages = control->ras_num_pa_recs +
1429 			control->ras_num_mca_recs * adev->umc.retire_unit;
1430 
1431 	if (hdr->header == RAS_TABLE_HDR_VAL) {
1432 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
1433 				 control->ras_num_bad_pages);
1434 
1435 		if (hdr->version >= RAS_TABLE_VER_V2_1) {
1436 			res = __read_table_ras_info(control);
1437 			if (res)
1438 				return res;
1439 		}
1440 
1441 		res = __verify_ras_table_checksum(control);
1442 		if (res)
1443 			dev_err(adev->dev,
1444 				"RAS table incorrect checksum or error:%d\n",
1445 				res);
1446 
1447 		/* Warn if we are at 90% of the threshold or above
1448 		 */
1449 		if (10 * control->ras_num_bad_pages >= 9 * ras->bad_page_cnt_threshold)
1450 			dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
1451 					control->ras_num_bad_pages,
1452 					ras->bad_page_cnt_threshold);
1453 	} else if (hdr->header == RAS_TABLE_HDR_BAD &&
1454 		   amdgpu_bad_page_threshold != 0) {
1455 		if (hdr->version >= RAS_TABLE_VER_V2_1) {
1456 			res = __read_table_ras_info(control);
1457 			if (res)
1458 				return res;
1459 		}
1460 
1461 		res = __verify_ras_table_checksum(control);
1462 		if (res) {
1463 			dev_err(adev->dev,
1464 				"RAS Table incorrect checksum or error:%d\n",
1465 				res);
1466 			return -EINVAL;
1467 		}
1468 		if (ras->bad_page_cnt_threshold >= control->ras_num_bad_pages) {
1469 			/* This means that, the threshold was increased since
1470 			 * the last time the system was booted, and now,
1471 			 * ras->bad_page_cnt_threshold - control->num_recs > 0,
1472 			 * so that at least one more record can be saved,
1473 			 * before the page count threshold is reached.
1474 			 */
1475 			dev_info(adev->dev,
1476 				 "records:%d threshold:%d, resetting "
1477 				 "RAS table header signature",
1478 				 control->ras_num_bad_pages,
1479 				 ras->bad_page_cnt_threshold);
1480 			res = amdgpu_ras_eeprom_correct_header_tag(control,
1481 								   RAS_TABLE_HDR_VAL);
1482 		} else {
1483 			dev_warn(adev->dev,
1484 				"RAS records:%d exceed threshold:%d\n",
1485 				control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
1486 			if ((amdgpu_bad_page_threshold == -1) ||
1487 			    (amdgpu_bad_page_threshold == -2)) {
1488 				res = 0;
1489 				dev_warn(adev->dev,
1490 					 "Please consult AMD Service Action Guide (SAG) for appropriate service procedures\n");
1491 			} else {
1492 				ras->is_rma = true;
1493 				dev_warn(adev->dev,
1494 					 "User defined threshold is set, runtime service will be halt when threshold is reached\n");
1495 			}
1496 		}
1497 	} else {
1498 		DRM_INFO("Creating a new EEPROM table");
1499 
1500 		res = amdgpu_ras_eeprom_reset_table(control);
1501 	}
1502 
1503 	return res < 0 ? res : 0;
1504 }
1505