xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision 1c9f8dff62d85ce00b0e99f774a84bd783af7cac)
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, 6):
162 	case IP_VERSION(13, 0, 10):
163 		return true;
164 	default:
165 		return false;
166 	}
167 }
168 
169 static bool __get_eeprom_i2c_addr(struct amdgpu_device *adev,
170 				  struct amdgpu_ras_eeprom_control *control)
171 {
172 	struct atom_context *atom_ctx = adev->mode_info.atom_context;
173 	u8 i2c_addr;
174 
175 	if (!control)
176 		return false;
177 
178 	if (amdgpu_atomfirmware_ras_rom_addr(adev, &i2c_addr)) {
179 		/* The address given by VBIOS is an 8-bit, wire-format
180 		 * address, i.e. the most significant byte.
181 		 *
182 		 * Normalize it to a 19-bit EEPROM address. Remove the
183 		 * device type identifier and make it a 7-bit address;
184 		 * then make it a 19-bit EEPROM address. See top of
185 		 * amdgpu_eeprom.c.
186 		 */
187 		i2c_addr = (i2c_addr & 0x0F) >> 1;
188 		control->i2c_address = ((u32) i2c_addr) << 16;
189 
190 		return true;
191 	}
192 
193 	switch (adev->ip_versions[MP1_HWIP][0]) {
194 	case IP_VERSION(11, 0, 2):
195 		/* VEGA20 and ARCTURUS */
196 		if (adev->asic_type == CHIP_VEGA20)
197 			control->i2c_address = EEPROM_I2C_MADDR_0;
198 		else if (strnstr(atom_ctx->vbios_pn,
199 				 "D342",
200 				 sizeof(atom_ctx->vbios_pn)))
201 			control->i2c_address = EEPROM_I2C_MADDR_0;
202 		else
203 			control->i2c_address = EEPROM_I2C_MADDR_4;
204 		return true;
205 	case IP_VERSION(11, 0, 7):
206 		control->i2c_address = EEPROM_I2C_MADDR_0;
207 		return true;
208 	case IP_VERSION(13, 0, 2):
209 		if (strnstr(atom_ctx->vbios_pn, "D673",
210 			    sizeof(atom_ctx->vbios_pn)))
211 			control->i2c_address = EEPROM_I2C_MADDR_4;
212 		else
213 			control->i2c_address = EEPROM_I2C_MADDR_0;
214 		return true;
215 	case IP_VERSION(13, 0, 0):
216 	case IP_VERSION(13, 0, 6):
217 	case IP_VERSION(13, 0, 10):
218 		control->i2c_address = EEPROM_I2C_MADDR_4;
219 		return true;
220 	default:
221 		return false;
222 	}
223 }
224 
225 static void
226 __encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header *hdr,
227 			     unsigned char *buf)
228 {
229 	u32 *pp = (uint32_t *)buf;
230 
231 	pp[0] = cpu_to_le32(hdr->header);
232 	pp[1] = cpu_to_le32(hdr->version);
233 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
234 	pp[3] = cpu_to_le32(hdr->tbl_size);
235 	pp[4] = cpu_to_le32(hdr->checksum);
236 }
237 
238 static void
239 __decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header *hdr,
240 			       unsigned char *buf)
241 {
242 	u32 *pp = (uint32_t *)buf;
243 
244 	hdr->header	      = le32_to_cpu(pp[0]);
245 	hdr->version	      = le32_to_cpu(pp[1]);
246 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
247 	hdr->tbl_size	      = le32_to_cpu(pp[3]);
248 	hdr->checksum	      = le32_to_cpu(pp[4]);
249 }
250 
251 static int __write_table_header(struct amdgpu_ras_eeprom_control *control)
252 {
253 	u8 buf[RAS_TABLE_HEADER_SIZE];
254 	struct amdgpu_device *adev = to_amdgpu_device(control);
255 	int res;
256 
257 	memset(buf, 0, sizeof(buf));
258 	__encode_table_header_to_buf(&control->tbl_hdr, buf);
259 
260 	/* i2c may be unstable in gpu reset */
261 	down_read(&adev->reset_domain->sem);
262 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
263 				  control->i2c_address +
264 				  control->ras_header_offset,
265 				  buf, RAS_TABLE_HEADER_SIZE);
266 	up_read(&adev->reset_domain->sem);
267 
268 	if (res < 0) {
269 		DRM_ERROR("Failed to write EEPROM table header:%d", res);
270 	} else if (res < RAS_TABLE_HEADER_SIZE) {
271 		DRM_ERROR("Short write:%d out of %d\n",
272 			  res, RAS_TABLE_HEADER_SIZE);
273 		res = -EIO;
274 	} else {
275 		res = 0;
276 	}
277 
278 	return res;
279 }
280 
281 static void
282 __encode_table_ras_info_to_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
283 			       unsigned char *buf)
284 {
285 	u32 *pp = (uint32_t *)buf;
286 	u32 tmp;
287 
288 	tmp = ((uint32_t)(rai->rma_status) & 0xFF) |
289 	      (((uint32_t)(rai->health_percent) << 8) & 0xFF00) |
290 	      (((uint32_t)(rai->ecc_page_threshold) << 16) & 0xFFFF0000);
291 	pp[0] = cpu_to_le32(tmp);
292 }
293 
294 static void
295 __decode_table_ras_info_from_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
296 				 unsigned char *buf)
297 {
298 	u32 *pp = (uint32_t *)buf;
299 	u32 tmp;
300 
301 	tmp = le32_to_cpu(pp[0]);
302 	rai->rma_status = tmp & 0xFF;
303 	rai->health_percent = (tmp >> 8) & 0xFF;
304 	rai->ecc_page_threshold = (tmp >> 16) & 0xFFFF;
305 }
306 
307 static int __write_table_ras_info(struct amdgpu_ras_eeprom_control *control)
308 {
309 	struct amdgpu_device *adev = to_amdgpu_device(control);
310 	u8 *buf;
311 	int res;
312 
313 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
314 	if (!buf) {
315 		DRM_ERROR("Failed to alloc buf to write table ras info\n");
316 		return -ENOMEM;
317 	}
318 
319 	__encode_table_ras_info_to_buf(&control->tbl_rai, buf);
320 
321 	/* i2c may be unstable in gpu reset */
322 	down_read(&adev->reset_domain->sem);
323 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
324 				  control->i2c_address +
325 				  control->ras_info_offset,
326 				  buf, RAS_TABLE_V2_1_INFO_SIZE);
327 	up_read(&adev->reset_domain->sem);
328 
329 	if (res < 0) {
330 		DRM_ERROR("Failed to write EEPROM table ras info:%d", res);
331 	} else if (res < RAS_TABLE_V2_1_INFO_SIZE) {
332 		DRM_ERROR("Short write:%d out of %d\n",
333 			  res, RAS_TABLE_V2_1_INFO_SIZE);
334 		res = -EIO;
335 	} else {
336 		res = 0;
337 	}
338 
339 	kfree(buf);
340 
341 	return res;
342 }
343 
344 static u8 __calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control *control)
345 {
346 	int ii;
347 	u8  *pp, csum;
348 	size_t sz;
349 
350 	/* Header checksum, skip checksum field in the calculation */
351 	sz = sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum);
352 	pp = (u8 *) &control->tbl_hdr;
353 	csum = 0;
354 	for (ii = 0; ii < sz; ii++, pp++)
355 		csum += *pp;
356 
357 	return csum;
358 }
359 
360 static u8 __calc_ras_info_byte_sum(const struct amdgpu_ras_eeprom_control *control)
361 {
362 	int ii;
363 	u8  *pp, csum;
364 	size_t sz;
365 
366 	sz = sizeof(control->tbl_rai);
367 	pp = (u8 *) &control->tbl_rai;
368 	csum = 0;
369 	for (ii = 0; ii < sz; ii++, pp++)
370 		csum += *pp;
371 
372 	return csum;
373 }
374 
375 static int amdgpu_ras_eeprom_correct_header_tag(
376 	struct amdgpu_ras_eeprom_control *control,
377 	uint32_t header)
378 {
379 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
380 	u8 *hh;
381 	int res;
382 	u8 csum;
383 
384 	csum = -hdr->checksum;
385 
386 	hh = (void *) &hdr->header;
387 	csum -= (hh[0] + hh[1] + hh[2] + hh[3]);
388 	hh = (void *) &header;
389 	csum += hh[0] + hh[1] + hh[2] + hh[3];
390 	csum = -csum;
391 	mutex_lock(&control->ras_tbl_mutex);
392 	hdr->header = header;
393 	hdr->checksum = csum;
394 	res = __write_table_header(control);
395 	mutex_unlock(&control->ras_tbl_mutex);
396 
397 	return res;
398 }
399 
400 /**
401  * amdgpu_ras_eeprom_reset_table -- Reset the RAS EEPROM table
402  * @control: pointer to control structure
403  *
404  * Reset the contents of the header of the RAS EEPROM table.
405  * Return 0 on success, -errno on error.
406  */
407 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
408 {
409 	struct amdgpu_device *adev = to_amdgpu_device(control);
410 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
411 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
412 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
413 	u8 csum;
414 	int res;
415 
416 	mutex_lock(&control->ras_tbl_mutex);
417 
418 	hdr->header = RAS_TABLE_HDR_VAL;
419 	if (adev->umc.ras &&
420 	    adev->umc.ras->set_eeprom_table_version)
421 		adev->umc.ras->set_eeprom_table_version(hdr);
422 	else
423 		hdr->version = RAS_TABLE_VER_V1;
424 
425 	if (hdr->version == RAS_TABLE_VER_V2_1) {
426 		hdr->first_rec_offset = RAS_RECORD_START_V2_1;
427 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE +
428 				RAS_TABLE_V2_1_INFO_SIZE;
429 		rai->rma_status = GPU_HEALTH_USABLE;
430 		/**
431 		 * GPU health represented as a percentage.
432 		 * 0 means worst health, 100 means fully health.
433 		 */
434 		rai->health_percent = 100;
435 		/* ecc_page_threshold = 0 means disable bad page retirement */
436 		rai->ecc_page_threshold = con->bad_page_cnt_threshold;
437 	} else {
438 		hdr->first_rec_offset = RAS_RECORD_START;
439 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE;
440 	}
441 
442 	csum = __calc_hdr_byte_sum(control);
443 	if (hdr->version == RAS_TABLE_VER_V2_1)
444 		csum += __calc_ras_info_byte_sum(control);
445 	csum = -csum;
446 	hdr->checksum = csum;
447 	res = __write_table_header(control);
448 	if (!res && hdr->version > RAS_TABLE_VER_V1)
449 		res = __write_table_ras_info(control);
450 
451 	control->ras_num_recs = 0;
452 	control->ras_fri = 0;
453 
454 	amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_recs);
455 
456 	control->bad_channel_bitmap = 0;
457 	amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap);
458 	con->update_channel_flag = false;
459 
460 	amdgpu_ras_debugfs_set_ret_size(control);
461 
462 	mutex_unlock(&control->ras_tbl_mutex);
463 
464 	return res;
465 }
466 
467 static void
468 __encode_table_record_to_buf(struct amdgpu_ras_eeprom_control *control,
469 			     struct eeprom_table_record *record,
470 			     unsigned char *buf)
471 {
472 	__le64 tmp = 0;
473 	int i = 0;
474 
475 	/* Next are all record fields according to EEPROM page spec in LE foramt */
476 	buf[i++] = record->err_type;
477 
478 	buf[i++] = record->bank;
479 
480 	tmp = cpu_to_le64(record->ts);
481 	memcpy(buf + i, &tmp, 8);
482 	i += 8;
483 
484 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
485 	memcpy(buf + i, &tmp, 6);
486 	i += 6;
487 
488 	buf[i++] = record->mem_channel;
489 	buf[i++] = record->mcumc_id;
490 
491 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
492 	memcpy(buf + i, &tmp, 6);
493 }
494 
495 static void
496 __decode_table_record_from_buf(struct amdgpu_ras_eeprom_control *control,
497 			       struct eeprom_table_record *record,
498 			       unsigned char *buf)
499 {
500 	__le64 tmp = 0;
501 	int i =  0;
502 
503 	/* Next are all record fields according to EEPROM page spec in LE foramt */
504 	record->err_type = buf[i++];
505 
506 	record->bank = buf[i++];
507 
508 	memcpy(&tmp, buf + i, 8);
509 	record->ts = le64_to_cpu(tmp);
510 	i += 8;
511 
512 	memcpy(&tmp, buf + i, 6);
513 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
514 	i += 6;
515 
516 	record->mem_channel = buf[i++];
517 	record->mcumc_id = buf[i++];
518 
519 	memcpy(&tmp, buf + i,  6);
520 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
521 }
522 
523 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
524 {
525 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
526 
527 	if (!__is_ras_eeprom_supported(adev) ||
528 	    !amdgpu_bad_page_threshold)
529 		return false;
530 
531 	/* skip check eeprom table for VEGA20 Gaming */
532 	if (!con)
533 		return false;
534 	else
535 		if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
536 			return false;
537 
538 	if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
539 		if (amdgpu_bad_page_threshold == -1) {
540 			dev_warn(adev->dev, "RAS records:%d exceed threshold:%d",
541 				con->eeprom_control.ras_num_recs, con->bad_page_cnt_threshold);
542 			dev_warn(adev->dev,
543 				"But GPU can be operated due to bad_page_threshold = -1.\n");
544 			return false;
545 		} else {
546 			dev_warn(adev->dev, "This GPU is in BAD status.");
547 			dev_warn(adev->dev, "Please retire it or set a larger "
548 				 "threshold value when reloading driver.\n");
549 			return true;
550 		}
551 	}
552 
553 	return false;
554 }
555 
556 /**
557  * __amdgpu_ras_eeprom_write -- write indexed from buffer to EEPROM
558  * @control: pointer to control structure
559  * @buf: pointer to buffer containing data to write
560  * @fri: start writing at this index
561  * @num: number of records to write
562  *
563  * The caller must hold the table mutex in @control.
564  * Return 0 on success, -errno otherwise.
565  */
566 static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control,
567 				     u8 *buf, const u32 fri, const u32 num)
568 {
569 	struct amdgpu_device *adev = to_amdgpu_device(control);
570 	u32 buf_size;
571 	int res;
572 
573 	/* i2c may be unstable in gpu reset */
574 	down_read(&adev->reset_domain->sem);
575 	buf_size = num * RAS_TABLE_RECORD_SIZE;
576 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
577 				  control->i2c_address +
578 				  RAS_INDEX_TO_OFFSET(control, fri),
579 				  buf, buf_size);
580 	up_read(&adev->reset_domain->sem);
581 	if (res < 0) {
582 		DRM_ERROR("Writing %d EEPROM table records error:%d",
583 			  num, res);
584 	} else if (res < buf_size) {
585 		/* Short write, return error.
586 		 */
587 		DRM_ERROR("Wrote %d records out of %d",
588 			  res / RAS_TABLE_RECORD_SIZE, num);
589 		res = -EIO;
590 	} else {
591 		res = 0;
592 	}
593 
594 	return res;
595 }
596 
597 static int
598 amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
599 			       struct eeprom_table_record *record,
600 			       const u32 num)
601 {
602 	struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
603 	u32 a, b, i;
604 	u8 *buf, *pp;
605 	int res;
606 
607 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
608 	if (!buf)
609 		return -ENOMEM;
610 
611 	/* Encode all of them in one go.
612 	 */
613 	pp = buf;
614 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
615 		__encode_table_record_to_buf(control, &record[i], pp);
616 
617 		/* update bad channel bitmap */
618 		if (!(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
619 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
620 			con->update_channel_flag = true;
621 		}
622 	}
623 
624 	/* a, first record index to write into.
625 	 * b, last record index to write into.
626 	 * a = first index to read (fri) + number of records in the table,
627 	 * b = a + @num - 1.
628 	 * Let N = control->ras_max_num_record_count, then we have,
629 	 * case 0: 0 <= a <= b < N,
630 	 *   just append @num records starting at a;
631 	 * case 1: 0 <= a < N <= b,
632 	 *   append (N - a) records starting at a, and
633 	 *   append the remainder,  b % N + 1, starting at 0.
634 	 * case 2: 0 <= fri < N <= a <= b, then modulo N we get two subcases,
635 	 * case 2a: 0 <= a <= b < N
636 	 *   append num records starting at a; and fix fri if b overwrote it,
637 	 *   and since a <= b, if b overwrote it then a must've also,
638 	 *   and if b didn't overwrite it, then a didn't also.
639 	 * case 2b: 0 <= b < a < N
640 	 *   write num records starting at a, which wraps around 0=N
641 	 *   and overwrite fri unconditionally. Now from case 2a,
642 	 *   this means that b eclipsed fri to overwrite it and wrap
643 	 *   around 0 again, i.e. b = 2N+r pre modulo N, so we unconditionally
644 	 *   set fri = b + 1 (mod N).
645 	 * Now, since fri is updated in every case, except the trivial case 0,
646 	 * the number of records present in the table after writing, is,
647 	 * num_recs - 1 = b - fri (mod N), and we take the positive value,
648 	 * by adding an arbitrary multiple of N before taking the modulo N
649 	 * as shown below.
650 	 */
651 	a = control->ras_fri + control->ras_num_recs;
652 	b = a + num  - 1;
653 	if (b < control->ras_max_record_count) {
654 		res = __amdgpu_ras_eeprom_write(control, buf, a, num);
655 	} else if (a < control->ras_max_record_count) {
656 		u32 g0, g1;
657 
658 		g0 = control->ras_max_record_count - a;
659 		g1 = b % control->ras_max_record_count + 1;
660 		res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
661 		if (res)
662 			goto Out;
663 		res = __amdgpu_ras_eeprom_write(control,
664 						buf + g0 * RAS_TABLE_RECORD_SIZE,
665 						0, g1);
666 		if (res)
667 			goto Out;
668 		if (g1 > control->ras_fri)
669 			control->ras_fri = g1 % control->ras_max_record_count;
670 	} else {
671 		a %= control->ras_max_record_count;
672 		b %= control->ras_max_record_count;
673 
674 		if (a <= b) {
675 			/* Note that, b - a + 1 = num. */
676 			res = __amdgpu_ras_eeprom_write(control, buf, a, num);
677 			if (res)
678 				goto Out;
679 			if (b >= control->ras_fri)
680 				control->ras_fri = (b + 1) % control->ras_max_record_count;
681 		} else {
682 			u32 g0, g1;
683 
684 			/* b < a, which means, we write from
685 			 * a to the end of the table, and from
686 			 * the start of the table to b.
687 			 */
688 			g0 = control->ras_max_record_count - a;
689 			g1 = b + 1;
690 			res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
691 			if (res)
692 				goto Out;
693 			res = __amdgpu_ras_eeprom_write(control,
694 							buf + g0 * RAS_TABLE_RECORD_SIZE,
695 							0, g1);
696 			if (res)
697 				goto Out;
698 			control->ras_fri = g1 % control->ras_max_record_count;
699 		}
700 	}
701 	control->ras_num_recs = 1 + (control->ras_max_record_count + b
702 				     - control->ras_fri)
703 		% control->ras_max_record_count;
704 Out:
705 	kfree(buf);
706 	return res;
707 }
708 
709 static int
710 amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
711 {
712 	struct amdgpu_device *adev = to_amdgpu_device(control);
713 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
714 	u8 *buf, *pp, csum;
715 	u32 buf_size;
716 	int res;
717 
718 	/* Modify the header if it exceeds.
719 	 */
720 	if (amdgpu_bad_page_threshold != 0 &&
721 	    control->ras_num_recs >= ras->bad_page_cnt_threshold) {
722 		dev_warn(adev->dev,
723 			"Saved bad pages %d reaches threshold value %d\n",
724 			control->ras_num_recs, ras->bad_page_cnt_threshold);
725 		control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
726 		if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1) {
727 			control->tbl_rai.rma_status = GPU_RETIRED__ECC_REACH_THRESHOLD;
728 			control->tbl_rai.health_percent = 0;
729 		}
730 	}
731 
732 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
733 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
734 					    RAS_TABLE_V2_1_INFO_SIZE +
735 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
736 	else
737 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
738 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
739 	control->tbl_hdr.checksum = 0;
740 
741 	buf_size = control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
742 	buf = kcalloc(control->ras_num_recs, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
743 	if (!buf) {
744 		DRM_ERROR("allocating memory for table of size %d bytes failed\n",
745 			  control->tbl_hdr.tbl_size);
746 		res = -ENOMEM;
747 		goto Out;
748 	}
749 
750 	down_read(&adev->reset_domain->sem);
751 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
752 				 control->i2c_address +
753 				 control->ras_record_offset,
754 				 buf, buf_size);
755 	up_read(&adev->reset_domain->sem);
756 	if (res < 0) {
757 		DRM_ERROR("EEPROM failed reading records:%d\n",
758 			  res);
759 		goto Out;
760 	} else if (res < buf_size) {
761 		DRM_ERROR("EEPROM read %d out of %d bytes\n",
762 			  res, buf_size);
763 		res = -EIO;
764 		goto Out;
765 	}
766 
767 	/**
768 	 * bad page records have been stored in eeprom,
769 	 * now calculate gpu health percent
770 	 */
771 	if (amdgpu_bad_page_threshold != 0 &&
772 	    control->tbl_hdr.version == RAS_TABLE_VER_V2_1 &&
773 	    control->ras_num_recs < ras->bad_page_cnt_threshold)
774 		control->tbl_rai.health_percent = ((ras->bad_page_cnt_threshold -
775 						   control->ras_num_recs) * 100) /
776 						   ras->bad_page_cnt_threshold;
777 
778 	/* Recalc the checksum.
779 	 */
780 	csum = 0;
781 	for (pp = buf; pp < buf + buf_size; pp++)
782 		csum += *pp;
783 
784 	csum += __calc_hdr_byte_sum(control);
785 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
786 		csum += __calc_ras_info_byte_sum(control);
787 	/* avoid sign extension when assigning to "checksum" */
788 	csum = -csum;
789 	control->tbl_hdr.checksum = csum;
790 	res = __write_table_header(control);
791 	if (!res && control->tbl_hdr.version > RAS_TABLE_VER_V1)
792 		res = __write_table_ras_info(control);
793 Out:
794 	kfree(buf);
795 	return res;
796 }
797 
798 /**
799  * amdgpu_ras_eeprom_append -- append records to the EEPROM RAS table
800  * @control: pointer to control structure
801  * @record: array of records to append
802  * @num: number of records in @record array
803  *
804  * Append @num records to the table, calculate the checksum and write
805  * the table back to EEPROM. The maximum number of records that
806  * can be appended is between 1 and control->ras_max_record_count,
807  * regardless of how many records are already stored in the table.
808  *
809  * Return 0 on success or if EEPROM is not supported, -errno on error.
810  */
811 int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
812 			     struct eeprom_table_record *record,
813 			     const u32 num)
814 {
815 	struct amdgpu_device *adev = to_amdgpu_device(control);
816 	int res;
817 
818 	if (!__is_ras_eeprom_supported(adev))
819 		return 0;
820 
821 	if (num == 0) {
822 		DRM_ERROR("will not append 0 records\n");
823 		return -EINVAL;
824 	} else if (num > control->ras_max_record_count) {
825 		DRM_ERROR("cannot append %d records than the size of table %d\n",
826 			  num, control->ras_max_record_count);
827 		return -EINVAL;
828 	}
829 
830 	mutex_lock(&control->ras_tbl_mutex);
831 
832 	res = amdgpu_ras_eeprom_append_table(control, record, num);
833 	if (!res)
834 		res = amdgpu_ras_eeprom_update_header(control);
835 	if (!res)
836 		amdgpu_ras_debugfs_set_ret_size(control);
837 
838 	mutex_unlock(&control->ras_tbl_mutex);
839 	return res;
840 }
841 
842 /**
843  * __amdgpu_ras_eeprom_read -- read indexed from EEPROM into buffer
844  * @control: pointer to control structure
845  * @buf: pointer to buffer to read into
846  * @fri: first record index, start reading at this index, absolute index
847  * @num: number of records to read
848  *
849  * The caller must hold the table mutex in @control.
850  * Return 0 on success, -errno otherwise.
851  */
852 static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
853 				    u8 *buf, const u32 fri, const u32 num)
854 {
855 	struct amdgpu_device *adev = to_amdgpu_device(control);
856 	u32 buf_size;
857 	int res;
858 
859 	/* i2c may be unstable in gpu reset */
860 	down_read(&adev->reset_domain->sem);
861 	buf_size = num * RAS_TABLE_RECORD_SIZE;
862 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
863 				 control->i2c_address +
864 				 RAS_INDEX_TO_OFFSET(control, fri),
865 				 buf, buf_size);
866 	up_read(&adev->reset_domain->sem);
867 	if (res < 0) {
868 		DRM_ERROR("Reading %d EEPROM table records error:%d",
869 			  num, res);
870 	} else if (res < buf_size) {
871 		/* Short read, return error.
872 		 */
873 		DRM_ERROR("Read %d records out of %d",
874 			  res / RAS_TABLE_RECORD_SIZE, num);
875 		res = -EIO;
876 	} else {
877 		res = 0;
878 	}
879 
880 	return res;
881 }
882 
883 /**
884  * amdgpu_ras_eeprom_read -- read EEPROM
885  * @control: pointer to control structure
886  * @record: array of records to read into
887  * @num: number of records in @record
888  *
889  * Reads num records from the RAS table in EEPROM and
890  * writes the data into @record array.
891  *
892  * Returns 0 on success, -errno on error.
893  */
894 int amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
895 			   struct eeprom_table_record *record,
896 			   const u32 num)
897 {
898 	struct amdgpu_device *adev = to_amdgpu_device(control);
899 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
900 	int i, res;
901 	u8 *buf, *pp;
902 	u32 g0, g1;
903 
904 	if (!__is_ras_eeprom_supported(adev))
905 		return 0;
906 
907 	if (num == 0) {
908 		DRM_ERROR("will not read 0 records\n");
909 		return -EINVAL;
910 	} else if (num > control->ras_num_recs) {
911 		DRM_ERROR("too many records to read:%d available:%d\n",
912 			  num, control->ras_num_recs);
913 		return -EINVAL;
914 	}
915 
916 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
917 	if (!buf)
918 		return -ENOMEM;
919 
920 	/* Determine how many records to read, from the first record
921 	 * index, fri, to the end of the table, and from the beginning
922 	 * of the table, such that the total number of records is
923 	 * @num, and we handle wrap around when fri > 0 and
924 	 * fri + num > RAS_MAX_RECORD_COUNT.
925 	 *
926 	 * First we compute the index of the last element
927 	 * which would be fetched from each region,
928 	 * g0 is in [fri, fri + num - 1], and
929 	 * g1 is in [0, RAS_MAX_RECORD_COUNT - 1].
930 	 * Then, if g0 < RAS_MAX_RECORD_COUNT, the index of
931 	 * the last element to fetch, we set g0 to _the number_
932 	 * of elements to fetch, @num, since we know that the last
933 	 * indexed to be fetched does not exceed the table.
934 	 *
935 	 * If, however, g0 >= RAS_MAX_RECORD_COUNT, then
936 	 * we set g0 to the number of elements to read
937 	 * until the end of the table, and g1 to the number of
938 	 * elements to read from the beginning of the table.
939 	 */
940 	g0 = control->ras_fri + num - 1;
941 	g1 = g0 % control->ras_max_record_count;
942 	if (g0 < control->ras_max_record_count) {
943 		g0 = num;
944 		g1 = 0;
945 	} else {
946 		g0 = control->ras_max_record_count - control->ras_fri;
947 		g1 += 1;
948 	}
949 
950 	mutex_lock(&control->ras_tbl_mutex);
951 	res = __amdgpu_ras_eeprom_read(control, buf, control->ras_fri, g0);
952 	if (res)
953 		goto Out;
954 	if (g1) {
955 		res = __amdgpu_ras_eeprom_read(control,
956 					       buf + g0 * RAS_TABLE_RECORD_SIZE,
957 					       0, g1);
958 		if (res)
959 			goto Out;
960 	}
961 
962 	res = 0;
963 
964 	/* Read up everything? Then transform.
965 	 */
966 	pp = buf;
967 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
968 		__decode_table_record_from_buf(control, &record[i], pp);
969 
970 		/* update bad channel bitmap */
971 		if (!(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
972 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
973 			con->update_channel_flag = true;
974 		}
975 	}
976 Out:
977 	kfree(buf);
978 	mutex_unlock(&control->ras_tbl_mutex);
979 
980 	return res;
981 }
982 
983 uint32_t amdgpu_ras_eeprom_max_record_count(struct amdgpu_ras_eeprom_control *control)
984 {
985 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
986 		return RAS_MAX_RECORD_COUNT_V2_1;
987 	else
988 		return RAS_MAX_RECORD_COUNT;
989 }
990 
991 static ssize_t
992 amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
993 				    size_t size, loff_t *pos)
994 {
995 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
996 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
997 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
998 	u8 data[50];
999 	int res;
1000 
1001 	if (!size)
1002 		return size;
1003 
1004 	if (!ras || !control) {
1005 		res = snprintf(data, sizeof(data), "Not supported\n");
1006 	} else {
1007 		res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
1008 			       RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
1009 	}
1010 
1011 	if (*pos >= res)
1012 		return 0;
1013 
1014 	res -= *pos;
1015 	res = min_t(size_t, res, size);
1016 
1017 	if (copy_to_user(buf, &data[*pos], res))
1018 		return -EFAULT;
1019 
1020 	*pos += res;
1021 
1022 	return res;
1023 }
1024 
1025 const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
1026 	.owner = THIS_MODULE,
1027 	.read = amdgpu_ras_debugfs_eeprom_size_read,
1028 	.write = NULL,
1029 	.llseek = default_llseek,
1030 };
1031 
1032 static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
1033 static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
1034 #define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
1035 static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
1036 static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
1037 #define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
1038 
1039 static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
1040 	"ignore",
1041 	"re",
1042 	"ue",
1043 };
1044 
1045 static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
1046 {
1047 	return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
1048 		strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
1049 }
1050 
1051 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
1052 {
1053 	struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
1054 					      eeprom_control);
1055 	struct dentry *de = ras->de_ras_eeprom_table;
1056 
1057 	if (de)
1058 		d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
1059 }
1060 
1061 static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
1062 					     size_t size, loff_t *pos)
1063 {
1064 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1065 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1066 	struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
1067 	const size_t orig_size = size;
1068 	int res = -EFAULT;
1069 	size_t data_len;
1070 
1071 	mutex_lock(&control->ras_tbl_mutex);
1072 
1073 	/* We want *pos - data_len > 0, which means there's
1074 	 * bytes to be printed from data.
1075 	 */
1076 	data_len = strlen(tbl_hdr_str);
1077 	if (*pos < data_len) {
1078 		data_len -= *pos;
1079 		data_len = min_t(size_t, data_len, size);
1080 		if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
1081 			goto Out;
1082 		buf += data_len;
1083 		size -= data_len;
1084 		*pos += data_len;
1085 	}
1086 
1087 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
1088 	if (*pos < data_len && size > 0) {
1089 		u8 data[tbl_hdr_fmt_size + 1];
1090 		loff_t lpos;
1091 
1092 		snprintf(data, sizeof(data), tbl_hdr_fmt,
1093 			 control->tbl_hdr.header,
1094 			 control->tbl_hdr.version,
1095 			 control->tbl_hdr.first_rec_offset,
1096 			 control->tbl_hdr.tbl_size,
1097 			 control->tbl_hdr.checksum);
1098 
1099 		data_len -= *pos;
1100 		data_len = min_t(size_t, data_len, size);
1101 		lpos = *pos - strlen(tbl_hdr_str);
1102 		if (copy_to_user(buf, &data[lpos], data_len))
1103 			goto Out;
1104 		buf += data_len;
1105 		size -= data_len;
1106 		*pos += data_len;
1107 	}
1108 
1109 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
1110 	if (*pos < data_len && size > 0) {
1111 		loff_t lpos;
1112 
1113 		data_len -= *pos;
1114 		data_len = min_t(size_t, data_len, size);
1115 		lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
1116 		if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
1117 			goto Out;
1118 		buf += data_len;
1119 		size -= data_len;
1120 		*pos += data_len;
1121 	}
1122 
1123 	data_len = amdgpu_ras_debugfs_table_size(control);
1124 	if (*pos < data_len && size > 0) {
1125 		u8 dare[RAS_TABLE_RECORD_SIZE];
1126 		u8 data[rec_hdr_fmt_size + 1];
1127 		struct eeprom_table_record record;
1128 		int s, r;
1129 
1130 		/* Find the starting record index
1131 		 */
1132 		s = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1133 			strlen(rec_hdr_str);
1134 		s = s / rec_hdr_fmt_size;
1135 		r = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1136 			strlen(rec_hdr_str);
1137 		r = r % rec_hdr_fmt_size;
1138 
1139 		for ( ; size > 0 && s < control->ras_num_recs; s++) {
1140 			u32 ai = RAS_RI_TO_AI(control, s);
1141 			/* Read a single record
1142 			 */
1143 			res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
1144 			if (res)
1145 				goto Out;
1146 			__decode_table_record_from_buf(control, &record, dare);
1147 			snprintf(data, sizeof(data), rec_hdr_fmt,
1148 				 s,
1149 				 RAS_INDEX_TO_OFFSET(control, ai),
1150 				 record_err_type_str[record.err_type],
1151 				 record.bank,
1152 				 record.ts,
1153 				 record.offset,
1154 				 record.mem_channel,
1155 				 record.mcumc_id,
1156 				 record.retired_page);
1157 
1158 			data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
1159 			if (copy_to_user(buf, &data[r], data_len)) {
1160 				res = -EFAULT;
1161 				goto Out;
1162 			}
1163 			buf += data_len;
1164 			size -= data_len;
1165 			*pos += data_len;
1166 			r = 0;
1167 		}
1168 	}
1169 	res = 0;
1170 Out:
1171 	mutex_unlock(&control->ras_tbl_mutex);
1172 	return res < 0 ? res : orig_size - size;
1173 }
1174 
1175 static ssize_t
1176 amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
1177 				     size_t size, loff_t *pos)
1178 {
1179 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1180 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1181 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1182 	u8 data[81];
1183 	int res;
1184 
1185 	if (!size)
1186 		return size;
1187 
1188 	if (!ras || !control) {
1189 		res = snprintf(data, sizeof(data), "Not supported\n");
1190 		if (*pos >= res)
1191 			return 0;
1192 
1193 		res -= *pos;
1194 		res = min_t(size_t, res, size);
1195 
1196 		if (copy_to_user(buf, &data[*pos], res))
1197 			return -EFAULT;
1198 
1199 		*pos += res;
1200 
1201 		return res;
1202 	} else {
1203 		return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
1204 	}
1205 }
1206 
1207 const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
1208 	.owner = THIS_MODULE,
1209 	.read = amdgpu_ras_debugfs_eeprom_table_read,
1210 	.write = NULL,
1211 	.llseek = default_llseek,
1212 };
1213 
1214 /**
1215  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
1216  * @control: pointer to control structure
1217  *
1218  * Check the checksum of the stored in EEPROM RAS table.
1219  *
1220  * Return 0 if the checksum is correct,
1221  * positive if it is not correct, and
1222  * -errno on I/O error.
1223  */
1224 static int __verify_ras_table_checksum(struct amdgpu_ras_eeprom_control *control)
1225 {
1226 	struct amdgpu_device *adev = to_amdgpu_device(control);
1227 	int buf_size, res;
1228 	u8  csum, *buf, *pp;
1229 
1230 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
1231 		buf_size = RAS_TABLE_HEADER_SIZE +
1232 			   RAS_TABLE_V2_1_INFO_SIZE +
1233 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1234 	else
1235 		buf_size = RAS_TABLE_HEADER_SIZE +
1236 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1237 
1238 	buf = kzalloc(buf_size, GFP_KERNEL);
1239 	if (!buf) {
1240 		DRM_ERROR("Out of memory checking RAS table checksum.\n");
1241 		return -ENOMEM;
1242 	}
1243 
1244 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1245 				 control->i2c_address +
1246 				 control->ras_header_offset,
1247 				 buf, buf_size);
1248 	if (res < buf_size) {
1249 		DRM_ERROR("Partial read for checksum, res:%d\n", res);
1250 		/* On partial reads, return -EIO.
1251 		 */
1252 		if (res >= 0)
1253 			res = -EIO;
1254 		goto Out;
1255 	}
1256 
1257 	csum = 0;
1258 	for (pp = buf; pp < buf + buf_size; pp++)
1259 		csum += *pp;
1260 Out:
1261 	kfree(buf);
1262 	return res < 0 ? res : csum;
1263 }
1264 
1265 static int __read_table_ras_info(struct amdgpu_ras_eeprom_control *control)
1266 {
1267 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
1268 	struct amdgpu_device *adev = to_amdgpu_device(control);
1269 	unsigned char *buf;
1270 	int res;
1271 
1272 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
1273 	if (!buf) {
1274 		DRM_ERROR("Failed to alloc buf to read EEPROM table ras info\n");
1275 		return -ENOMEM;
1276 	}
1277 
1278 	/**
1279 	 * EEPROM table V2_1 supports ras info,
1280 	 * read EEPROM table ras info
1281 	 */
1282 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1283 				 control->i2c_address + control->ras_info_offset,
1284 				 buf, RAS_TABLE_V2_1_INFO_SIZE);
1285 	if (res < RAS_TABLE_V2_1_INFO_SIZE) {
1286 		DRM_ERROR("Failed to read EEPROM table ras info, res:%d", res);
1287 		res = res >= 0 ? -EIO : res;
1288 		goto Out;
1289 	}
1290 
1291 	__decode_table_ras_info_from_buf(rai, buf);
1292 
1293 Out:
1294 	kfree(buf);
1295 	return res == RAS_TABLE_V2_1_INFO_SIZE ? 0 : res;
1296 }
1297 
1298 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control,
1299 			   bool *exceed_err_limit)
1300 {
1301 	struct amdgpu_device *adev = to_amdgpu_device(control);
1302 	unsigned char buf[RAS_TABLE_HEADER_SIZE] = { 0 };
1303 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1304 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1305 	int res;
1306 
1307 	*exceed_err_limit = false;
1308 
1309 	if (!__is_ras_eeprom_supported(adev))
1310 		return 0;
1311 
1312 	/* Verify i2c adapter is initialized */
1313 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1314 		return -ENOENT;
1315 
1316 	if (!__get_eeprom_i2c_addr(adev, control))
1317 		return -EINVAL;
1318 
1319 	control->ras_header_offset = RAS_HDR_START;
1320 	control->ras_info_offset = RAS_TABLE_V2_1_INFO_START;
1321 	mutex_init(&control->ras_tbl_mutex);
1322 
1323 	/* Read the table header from EEPROM address */
1324 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1325 				 control->i2c_address + control->ras_header_offset,
1326 				 buf, RAS_TABLE_HEADER_SIZE);
1327 	if (res < RAS_TABLE_HEADER_SIZE) {
1328 		DRM_ERROR("Failed to read EEPROM table header, res:%d", res);
1329 		return res >= 0 ? -EIO : res;
1330 	}
1331 
1332 	__decode_table_header_from_buf(hdr, buf);
1333 
1334 	if (hdr->version == RAS_TABLE_VER_V2_1) {
1335 		control->ras_num_recs = RAS_NUM_RECS_V2_1(hdr);
1336 		control->ras_record_offset = RAS_RECORD_START_V2_1;
1337 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT_V2_1;
1338 	} else {
1339 		control->ras_num_recs = RAS_NUM_RECS(hdr);
1340 		control->ras_record_offset = RAS_RECORD_START;
1341 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT;
1342 	}
1343 	control->ras_fri = RAS_OFFSET_TO_INDEX(control, hdr->first_rec_offset);
1344 
1345 	if (hdr->header == RAS_TABLE_HDR_VAL) {
1346 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
1347 				 control->ras_num_recs);
1348 
1349 		if (hdr->version == RAS_TABLE_VER_V2_1) {
1350 			res = __read_table_ras_info(control);
1351 			if (res)
1352 				return res;
1353 		}
1354 
1355 		res = __verify_ras_table_checksum(control);
1356 		if (res)
1357 			DRM_ERROR("RAS table incorrect checksum or error:%d\n",
1358 				  res);
1359 
1360 		/* Warn if we are at 90% of the threshold or above
1361 		 */
1362 		if (10 * control->ras_num_recs >= 9 * ras->bad_page_cnt_threshold)
1363 			dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
1364 					control->ras_num_recs,
1365 					ras->bad_page_cnt_threshold);
1366 	} else if (hdr->header == RAS_TABLE_HDR_BAD &&
1367 		   amdgpu_bad_page_threshold != 0) {
1368 		if (hdr->version == RAS_TABLE_VER_V2_1) {
1369 			res = __read_table_ras_info(control);
1370 			if (res)
1371 				return res;
1372 		}
1373 
1374 		res = __verify_ras_table_checksum(control);
1375 		if (res)
1376 			DRM_ERROR("RAS Table incorrect checksum or error:%d\n",
1377 				  res);
1378 		if (ras->bad_page_cnt_threshold > control->ras_num_recs) {
1379 			/* This means that, the threshold was increased since
1380 			 * the last time the system was booted, and now,
1381 			 * ras->bad_page_cnt_threshold - control->num_recs > 0,
1382 			 * so that at least one more record can be saved,
1383 			 * before the page count threshold is reached.
1384 			 */
1385 			dev_info(adev->dev,
1386 				 "records:%d threshold:%d, resetting "
1387 				 "RAS table header signature",
1388 				 control->ras_num_recs,
1389 				 ras->bad_page_cnt_threshold);
1390 			res = amdgpu_ras_eeprom_correct_header_tag(control,
1391 								   RAS_TABLE_HDR_VAL);
1392 		} else {
1393 			dev_err(adev->dev, "RAS records:%d exceed threshold:%d",
1394 				control->ras_num_recs, ras->bad_page_cnt_threshold);
1395 			if (amdgpu_bad_page_threshold == -1) {
1396 				dev_warn(adev->dev, "GPU will be initialized due to bad_page_threshold = -1.");
1397 				res = 0;
1398 			} else {
1399 				*exceed_err_limit = true;
1400 				dev_err(adev->dev,
1401 					"RAS records:%d exceed threshold:%d, "
1402 					"GPU will not be initialized. Replace this GPU or increase the threshold",
1403 					control->ras_num_recs, ras->bad_page_cnt_threshold);
1404 			}
1405 		}
1406 	} else {
1407 		DRM_INFO("Creating a new EEPROM table");
1408 
1409 		res = amdgpu_ras_eeprom_reset_table(control);
1410 	}
1411 
1412 	return res < 0 ? res : 0;
1413 }
1414