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