xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision 001821b0e79716c4e17c71d8e053a23599a7a508)
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 (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
157 	case IP_VERSION(11, 0, 2): /* VEGA20 and ARCTURUS */
158 	case IP_VERSION(11, 0, 7): /* Sienna cichlid */
159 	case IP_VERSION(13, 0, 0):
160 	case IP_VERSION(13, 0, 2): /* Aldebaran */
161 	case IP_VERSION(13, 0, 10):
162 		return true;
163 	case IP_VERSION(13, 0, 6):
164 		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 (amdgpu_ip_version(adev, 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 		if (strnstr(atom_ctx->vbios_pn, "D707",
218 			    sizeof(atom_ctx->vbios_pn)))
219 			control->i2c_address = EEPROM_I2C_MADDR_0;
220 		else
221 			control->i2c_address = EEPROM_I2C_MADDR_4;
222 		return true;
223 	case IP_VERSION(13, 0, 6):
224 	case IP_VERSION(13, 0, 10):
225 		control->i2c_address = EEPROM_I2C_MADDR_4;
226 		return true;
227 	default:
228 		return false;
229 	}
230 }
231 
232 static void
233 __encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header *hdr,
234 			     unsigned char *buf)
235 {
236 	u32 *pp = (uint32_t *)buf;
237 
238 	pp[0] = cpu_to_le32(hdr->header);
239 	pp[1] = cpu_to_le32(hdr->version);
240 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
241 	pp[3] = cpu_to_le32(hdr->tbl_size);
242 	pp[4] = cpu_to_le32(hdr->checksum);
243 }
244 
245 static void
246 __decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header *hdr,
247 			       unsigned char *buf)
248 {
249 	u32 *pp = (uint32_t *)buf;
250 
251 	hdr->header	      = le32_to_cpu(pp[0]);
252 	hdr->version	      = le32_to_cpu(pp[1]);
253 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
254 	hdr->tbl_size	      = le32_to_cpu(pp[3]);
255 	hdr->checksum	      = le32_to_cpu(pp[4]);
256 }
257 
258 static int __write_table_header(struct amdgpu_ras_eeprom_control *control)
259 {
260 	u8 buf[RAS_TABLE_HEADER_SIZE];
261 	struct amdgpu_device *adev = to_amdgpu_device(control);
262 	int res;
263 
264 	memset(buf, 0, sizeof(buf));
265 	__encode_table_header_to_buf(&control->tbl_hdr, buf);
266 
267 	/* i2c may be unstable in gpu reset */
268 	down_read(&adev->reset_domain->sem);
269 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
270 				  control->i2c_address +
271 				  control->ras_header_offset,
272 				  buf, RAS_TABLE_HEADER_SIZE);
273 	up_read(&adev->reset_domain->sem);
274 
275 	if (res < 0) {
276 		DRM_ERROR("Failed to write EEPROM table header:%d", res);
277 	} else if (res < RAS_TABLE_HEADER_SIZE) {
278 		DRM_ERROR("Short write:%d out of %d\n",
279 			  res, RAS_TABLE_HEADER_SIZE);
280 		res = -EIO;
281 	} else {
282 		res = 0;
283 	}
284 
285 	return res;
286 }
287 
288 static void
289 __encode_table_ras_info_to_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
290 			       unsigned char *buf)
291 {
292 	u32 *pp = (uint32_t *)buf;
293 	u32 tmp;
294 
295 	tmp = ((uint32_t)(rai->rma_status) & 0xFF) |
296 	      (((uint32_t)(rai->health_percent) << 8) & 0xFF00) |
297 	      (((uint32_t)(rai->ecc_page_threshold) << 16) & 0xFFFF0000);
298 	pp[0] = cpu_to_le32(tmp);
299 }
300 
301 static void
302 __decode_table_ras_info_from_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
303 				 unsigned char *buf)
304 {
305 	u32 *pp = (uint32_t *)buf;
306 	u32 tmp;
307 
308 	tmp = le32_to_cpu(pp[0]);
309 	rai->rma_status = tmp & 0xFF;
310 	rai->health_percent = (tmp >> 8) & 0xFF;
311 	rai->ecc_page_threshold = (tmp >> 16) & 0xFFFF;
312 }
313 
314 static int __write_table_ras_info(struct amdgpu_ras_eeprom_control *control)
315 {
316 	struct amdgpu_device *adev = to_amdgpu_device(control);
317 	u8 *buf;
318 	int res;
319 
320 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
321 	if (!buf) {
322 		DRM_ERROR("Failed to alloc buf to write table ras info\n");
323 		return -ENOMEM;
324 	}
325 
326 	__encode_table_ras_info_to_buf(&control->tbl_rai, buf);
327 
328 	/* i2c may be unstable in gpu reset */
329 	down_read(&adev->reset_domain->sem);
330 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
331 				  control->i2c_address +
332 				  control->ras_info_offset,
333 				  buf, RAS_TABLE_V2_1_INFO_SIZE);
334 	up_read(&adev->reset_domain->sem);
335 
336 	if (res < 0) {
337 		DRM_ERROR("Failed to write EEPROM table ras info:%d", res);
338 	} else if (res < RAS_TABLE_V2_1_INFO_SIZE) {
339 		DRM_ERROR("Short write:%d out of %d\n",
340 			  res, RAS_TABLE_V2_1_INFO_SIZE);
341 		res = -EIO;
342 	} else {
343 		res = 0;
344 	}
345 
346 	kfree(buf);
347 
348 	return res;
349 }
350 
351 static u8 __calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control *control)
352 {
353 	int ii;
354 	u8  *pp, csum;
355 	size_t sz;
356 
357 	/* Header checksum, skip checksum field in the calculation */
358 	sz = sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum);
359 	pp = (u8 *) &control->tbl_hdr;
360 	csum = 0;
361 	for (ii = 0; ii < sz; ii++, pp++)
362 		csum += *pp;
363 
364 	return csum;
365 }
366 
367 static u8 __calc_ras_info_byte_sum(const struct amdgpu_ras_eeprom_control *control)
368 {
369 	int ii;
370 	u8  *pp, csum;
371 	size_t sz;
372 
373 	sz = sizeof(control->tbl_rai);
374 	pp = (u8 *) &control->tbl_rai;
375 	csum = 0;
376 	for (ii = 0; ii < sz; ii++, pp++)
377 		csum += *pp;
378 
379 	return csum;
380 }
381 
382 static int amdgpu_ras_eeprom_correct_header_tag(
383 	struct amdgpu_ras_eeprom_control *control,
384 	uint32_t header)
385 {
386 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
387 	u8 *hh;
388 	int res;
389 	u8 csum;
390 
391 	csum = -hdr->checksum;
392 
393 	hh = (void *) &hdr->header;
394 	csum -= (hh[0] + hh[1] + hh[2] + hh[3]);
395 	hh = (void *) &header;
396 	csum += hh[0] + hh[1] + hh[2] + hh[3];
397 	csum = -csum;
398 	mutex_lock(&control->ras_tbl_mutex);
399 	hdr->header = header;
400 	hdr->checksum = csum;
401 	res = __write_table_header(control);
402 	mutex_unlock(&control->ras_tbl_mutex);
403 
404 	return res;
405 }
406 
407 static void amdgpu_ras_set_eeprom_table_version(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 
412 	switch (amdgpu_ip_version(adev, UMC_HWIP, 0)) {
413 	case IP_VERSION(8, 10, 0):
414 	case IP_VERSION(12, 0, 0):
415 		hdr->version = RAS_TABLE_VER_V2_1;
416 		return;
417 	default:
418 		hdr->version = RAS_TABLE_VER_V1;
419 		return;
420 	}
421 }
422 
423 /**
424  * amdgpu_ras_eeprom_reset_table -- Reset the RAS EEPROM table
425  * @control: pointer to control structure
426  *
427  * Reset the contents of the header of the RAS EEPROM table.
428  * Return 0 on success, -errno on error.
429  */
430 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
431 {
432 	struct amdgpu_device *adev = to_amdgpu_device(control);
433 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
434 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
435 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
436 	u8 csum;
437 	int res;
438 
439 	mutex_lock(&control->ras_tbl_mutex);
440 
441 	hdr->header = RAS_TABLE_HDR_VAL;
442 	amdgpu_ras_set_eeprom_table_version(control);
443 
444 	if (hdr->version == RAS_TABLE_VER_V2_1) {
445 		hdr->first_rec_offset = RAS_RECORD_START_V2_1;
446 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE +
447 				RAS_TABLE_V2_1_INFO_SIZE;
448 		rai->rma_status = GPU_HEALTH_USABLE;
449 		/**
450 		 * GPU health represented as a percentage.
451 		 * 0 means worst health, 100 means fully health.
452 		 */
453 		rai->health_percent = 100;
454 		/* ecc_page_threshold = 0 means disable bad page retirement */
455 		rai->ecc_page_threshold = con->bad_page_cnt_threshold;
456 	} else {
457 		hdr->first_rec_offset = RAS_RECORD_START;
458 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE;
459 	}
460 
461 	csum = __calc_hdr_byte_sum(control);
462 	if (hdr->version == RAS_TABLE_VER_V2_1)
463 		csum += __calc_ras_info_byte_sum(control);
464 	csum = -csum;
465 	hdr->checksum = csum;
466 	res = __write_table_header(control);
467 	if (!res && hdr->version > RAS_TABLE_VER_V1)
468 		res = __write_table_ras_info(control);
469 
470 	control->ras_num_recs = 0;
471 	control->ras_fri = 0;
472 
473 	amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_recs);
474 
475 	control->bad_channel_bitmap = 0;
476 	amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap);
477 	con->update_channel_flag = false;
478 
479 	amdgpu_ras_debugfs_set_ret_size(control);
480 
481 	mutex_unlock(&control->ras_tbl_mutex);
482 
483 	return res;
484 }
485 
486 static void
487 __encode_table_record_to_buf(struct amdgpu_ras_eeprom_control *control,
488 			     struct eeprom_table_record *record,
489 			     unsigned char *buf)
490 {
491 	__le64 tmp = 0;
492 	int i = 0;
493 
494 	/* Next are all record fields according to EEPROM page spec in LE foramt */
495 	buf[i++] = record->err_type;
496 
497 	buf[i++] = record->bank;
498 
499 	tmp = cpu_to_le64(record->ts);
500 	memcpy(buf + i, &tmp, 8);
501 	i += 8;
502 
503 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
504 	memcpy(buf + i, &tmp, 6);
505 	i += 6;
506 
507 	buf[i++] = record->mem_channel;
508 	buf[i++] = record->mcumc_id;
509 
510 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
511 	memcpy(buf + i, &tmp, 6);
512 }
513 
514 static void
515 __decode_table_record_from_buf(struct amdgpu_ras_eeprom_control *control,
516 			       struct eeprom_table_record *record,
517 			       unsigned char *buf)
518 {
519 	__le64 tmp = 0;
520 	int i =  0;
521 
522 	/* Next are all record fields according to EEPROM page spec in LE foramt */
523 	record->err_type = buf[i++];
524 
525 	record->bank = buf[i++];
526 
527 	memcpy(&tmp, buf + i, 8);
528 	record->ts = le64_to_cpu(tmp);
529 	i += 8;
530 
531 	memcpy(&tmp, buf + i, 6);
532 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
533 	i += 6;
534 
535 	record->mem_channel = buf[i++];
536 	record->mcumc_id = buf[i++];
537 
538 	memcpy(&tmp, buf + i,  6);
539 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
540 }
541 
542 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
543 {
544 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
545 
546 	if (!__is_ras_eeprom_supported(adev) ||
547 	    !amdgpu_bad_page_threshold)
548 		return false;
549 
550 	/* skip check eeprom table for VEGA20 Gaming */
551 	if (!con)
552 		return false;
553 	else
554 		if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
555 			return false;
556 
557 	if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
558 		if (amdgpu_bad_page_threshold == -1) {
559 			dev_warn(adev->dev, "RAS records:%d exceed threshold:%d",
560 				con->eeprom_control.ras_num_recs, con->bad_page_cnt_threshold);
561 			dev_warn(adev->dev,
562 				"But GPU can be operated due to bad_page_threshold = -1.\n");
563 			return false;
564 		} else {
565 			dev_warn(adev->dev, "This GPU is in BAD status.");
566 			dev_warn(adev->dev, "Please retire it or set a larger "
567 				 "threshold value when reloading driver.\n");
568 			return true;
569 		}
570 	}
571 
572 	return false;
573 }
574 
575 /**
576  * __amdgpu_ras_eeprom_write -- write indexed from buffer to EEPROM
577  * @control: pointer to control structure
578  * @buf: pointer to buffer containing data to write
579  * @fri: start writing at this index
580  * @num: number of records to write
581  *
582  * The caller must hold the table mutex in @control.
583  * Return 0 on success, -errno otherwise.
584  */
585 static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control,
586 				     u8 *buf, const u32 fri, const u32 num)
587 {
588 	struct amdgpu_device *adev = to_amdgpu_device(control);
589 	u32 buf_size;
590 	int res;
591 
592 	/* i2c may be unstable in gpu reset */
593 	down_read(&adev->reset_domain->sem);
594 	buf_size = num * RAS_TABLE_RECORD_SIZE;
595 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
596 				  control->i2c_address +
597 				  RAS_INDEX_TO_OFFSET(control, fri),
598 				  buf, buf_size);
599 	up_read(&adev->reset_domain->sem);
600 	if (res < 0) {
601 		DRM_ERROR("Writing %d EEPROM table records error:%d",
602 			  num, res);
603 	} else if (res < buf_size) {
604 		/* Short write, return error.
605 		 */
606 		DRM_ERROR("Wrote %d records out of %d",
607 			  res / RAS_TABLE_RECORD_SIZE, num);
608 		res = -EIO;
609 	} else {
610 		res = 0;
611 	}
612 
613 	return res;
614 }
615 
616 static int
617 amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
618 			       struct eeprom_table_record *record,
619 			       const u32 num)
620 {
621 	struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
622 	u32 a, b, i;
623 	u8 *buf, *pp;
624 	int res;
625 
626 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
627 	if (!buf)
628 		return -ENOMEM;
629 
630 	/* Encode all of them in one go.
631 	 */
632 	pp = buf;
633 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
634 		__encode_table_record_to_buf(control, &record[i], pp);
635 
636 		/* update bad channel bitmap */
637 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
638 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
639 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
640 			con->update_channel_flag = true;
641 		}
642 	}
643 
644 	/* a, first record index to write into.
645 	 * b, last record index to write into.
646 	 * a = first index to read (fri) + number of records in the table,
647 	 * b = a + @num - 1.
648 	 * Let N = control->ras_max_num_record_count, then we have,
649 	 * case 0: 0 <= a <= b < N,
650 	 *   just append @num records starting at a;
651 	 * case 1: 0 <= a < N <= b,
652 	 *   append (N - a) records starting at a, and
653 	 *   append the remainder,  b % N + 1, starting at 0.
654 	 * case 2: 0 <= fri < N <= a <= b, then modulo N we get two subcases,
655 	 * case 2a: 0 <= a <= b < N
656 	 *   append num records starting at a; and fix fri if b overwrote it,
657 	 *   and since a <= b, if b overwrote it then a must've also,
658 	 *   and if b didn't overwrite it, then a didn't also.
659 	 * case 2b: 0 <= b < a < N
660 	 *   write num records starting at a, which wraps around 0=N
661 	 *   and overwrite fri unconditionally. Now from case 2a,
662 	 *   this means that b eclipsed fri to overwrite it and wrap
663 	 *   around 0 again, i.e. b = 2N+r pre modulo N, so we unconditionally
664 	 *   set fri = b + 1 (mod N).
665 	 * Now, since fri is updated in every case, except the trivial case 0,
666 	 * the number of records present in the table after writing, is,
667 	 * num_recs - 1 = b - fri (mod N), and we take the positive value,
668 	 * by adding an arbitrary multiple of N before taking the modulo N
669 	 * as shown below.
670 	 */
671 	a = control->ras_fri + control->ras_num_recs;
672 	b = a + num  - 1;
673 	if (b < control->ras_max_record_count) {
674 		res = __amdgpu_ras_eeprom_write(control, buf, a, num);
675 	} else if (a < control->ras_max_record_count) {
676 		u32 g0, g1;
677 
678 		g0 = control->ras_max_record_count - a;
679 		g1 = b % control->ras_max_record_count + 1;
680 		res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
681 		if (res)
682 			goto Out;
683 		res = __amdgpu_ras_eeprom_write(control,
684 						buf + g0 * RAS_TABLE_RECORD_SIZE,
685 						0, g1);
686 		if (res)
687 			goto Out;
688 		if (g1 > control->ras_fri)
689 			control->ras_fri = g1 % control->ras_max_record_count;
690 	} else {
691 		a %= control->ras_max_record_count;
692 		b %= control->ras_max_record_count;
693 
694 		if (a <= b) {
695 			/* Note that, b - a + 1 = num. */
696 			res = __amdgpu_ras_eeprom_write(control, buf, a, num);
697 			if (res)
698 				goto Out;
699 			if (b >= control->ras_fri)
700 				control->ras_fri = (b + 1) % control->ras_max_record_count;
701 		} else {
702 			u32 g0, g1;
703 
704 			/* b < a, which means, we write from
705 			 * a to the end of the table, and from
706 			 * the start of the table to b.
707 			 */
708 			g0 = control->ras_max_record_count - a;
709 			g1 = b + 1;
710 			res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
711 			if (res)
712 				goto Out;
713 			res = __amdgpu_ras_eeprom_write(control,
714 							buf + g0 * RAS_TABLE_RECORD_SIZE,
715 							0, g1);
716 			if (res)
717 				goto Out;
718 			control->ras_fri = g1 % control->ras_max_record_count;
719 		}
720 	}
721 	control->ras_num_recs = 1 + (control->ras_max_record_count + b
722 				     - control->ras_fri)
723 		% control->ras_max_record_count;
724 Out:
725 	kfree(buf);
726 	return res;
727 }
728 
729 static int
730 amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
731 {
732 	struct amdgpu_device *adev = to_amdgpu_device(control);
733 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
734 	u8 *buf, *pp, csum;
735 	u32 buf_size;
736 	int res;
737 
738 	/* Modify the header if it exceeds.
739 	 */
740 	if (amdgpu_bad_page_threshold != 0 &&
741 	    control->ras_num_recs >= ras->bad_page_cnt_threshold) {
742 		dev_warn(adev->dev,
743 			"Saved bad pages %d reaches threshold value %d\n",
744 			control->ras_num_recs, ras->bad_page_cnt_threshold);
745 		control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
746 		if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1) {
747 			control->tbl_rai.rma_status = GPU_RETIRED__ECC_REACH_THRESHOLD;
748 			control->tbl_rai.health_percent = 0;
749 		}
750 
751 		/* ignore the -ENOTSUPP return value */
752 		amdgpu_dpm_send_rma_reason(adev);
753 	}
754 
755 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
756 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
757 					    RAS_TABLE_V2_1_INFO_SIZE +
758 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
759 	else
760 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
761 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
762 	control->tbl_hdr.checksum = 0;
763 
764 	buf_size = control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
765 	buf = kcalloc(control->ras_num_recs, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
766 	if (!buf) {
767 		DRM_ERROR("allocating memory for table of size %d bytes failed\n",
768 			  control->tbl_hdr.tbl_size);
769 		res = -ENOMEM;
770 		goto Out;
771 	}
772 
773 	down_read(&adev->reset_domain->sem);
774 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
775 				 control->i2c_address +
776 				 control->ras_record_offset,
777 				 buf, buf_size);
778 	up_read(&adev->reset_domain->sem);
779 	if (res < 0) {
780 		DRM_ERROR("EEPROM failed reading records:%d\n",
781 			  res);
782 		goto Out;
783 	} else if (res < buf_size) {
784 		DRM_ERROR("EEPROM read %d out of %d bytes\n",
785 			  res, buf_size);
786 		res = -EIO;
787 		goto Out;
788 	}
789 
790 	/**
791 	 * bad page records have been stored in eeprom,
792 	 * now calculate gpu health percent
793 	 */
794 	if (amdgpu_bad_page_threshold != 0 &&
795 	    control->tbl_hdr.version == RAS_TABLE_VER_V2_1 &&
796 	    control->ras_num_recs < ras->bad_page_cnt_threshold)
797 		control->tbl_rai.health_percent = ((ras->bad_page_cnt_threshold -
798 						   control->ras_num_recs) * 100) /
799 						   ras->bad_page_cnt_threshold;
800 
801 	/* Recalc the checksum.
802 	 */
803 	csum = 0;
804 	for (pp = buf; pp < buf + buf_size; pp++)
805 		csum += *pp;
806 
807 	csum += __calc_hdr_byte_sum(control);
808 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
809 		csum += __calc_ras_info_byte_sum(control);
810 	/* avoid sign extension when assigning to "checksum" */
811 	csum = -csum;
812 	control->tbl_hdr.checksum = csum;
813 	res = __write_table_header(control);
814 	if (!res && control->tbl_hdr.version > RAS_TABLE_VER_V1)
815 		res = __write_table_ras_info(control);
816 Out:
817 	kfree(buf);
818 	return res;
819 }
820 
821 /**
822  * amdgpu_ras_eeprom_append -- append records to the EEPROM RAS table
823  * @control: pointer to control structure
824  * @record: array of records to append
825  * @num: number of records in @record array
826  *
827  * Append @num records to the table, calculate the checksum and write
828  * the table back to EEPROM. The maximum number of records that
829  * can be appended is between 1 and control->ras_max_record_count,
830  * regardless of how many records are already stored in the table.
831  *
832  * Return 0 on success or if EEPROM is not supported, -errno on error.
833  */
834 int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
835 			     struct eeprom_table_record *record,
836 			     const u32 num)
837 {
838 	struct amdgpu_device *adev = to_amdgpu_device(control);
839 	int res;
840 
841 	if (!__is_ras_eeprom_supported(adev))
842 		return 0;
843 
844 	if (num == 0) {
845 		DRM_ERROR("will not append 0 records\n");
846 		return -EINVAL;
847 	} else if (num > control->ras_max_record_count) {
848 		DRM_ERROR("cannot append %d records than the size of table %d\n",
849 			  num, control->ras_max_record_count);
850 		return -EINVAL;
851 	}
852 
853 	mutex_lock(&control->ras_tbl_mutex);
854 
855 	res = amdgpu_ras_eeprom_append_table(control, record, num);
856 	if (!res)
857 		res = amdgpu_ras_eeprom_update_header(control);
858 	if (!res)
859 		amdgpu_ras_debugfs_set_ret_size(control);
860 
861 	mutex_unlock(&control->ras_tbl_mutex);
862 	return res;
863 }
864 
865 /**
866  * __amdgpu_ras_eeprom_read -- read indexed from EEPROM into buffer
867  * @control: pointer to control structure
868  * @buf: pointer to buffer to read into
869  * @fri: first record index, start reading at this index, absolute index
870  * @num: number of records to read
871  *
872  * The caller must hold the table mutex in @control.
873  * Return 0 on success, -errno otherwise.
874  */
875 static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
876 				    u8 *buf, const u32 fri, const u32 num)
877 {
878 	struct amdgpu_device *adev = to_amdgpu_device(control);
879 	u32 buf_size;
880 	int res;
881 
882 	/* i2c may be unstable in gpu reset */
883 	down_read(&adev->reset_domain->sem);
884 	buf_size = num * RAS_TABLE_RECORD_SIZE;
885 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
886 				 control->i2c_address +
887 				 RAS_INDEX_TO_OFFSET(control, fri),
888 				 buf, buf_size);
889 	up_read(&adev->reset_domain->sem);
890 	if (res < 0) {
891 		DRM_ERROR("Reading %d EEPROM table records error:%d",
892 			  num, res);
893 	} else if (res < buf_size) {
894 		/* Short read, return error.
895 		 */
896 		DRM_ERROR("Read %d records out of %d",
897 			  res / RAS_TABLE_RECORD_SIZE, num);
898 		res = -EIO;
899 	} else {
900 		res = 0;
901 	}
902 
903 	return res;
904 }
905 
906 /**
907  * amdgpu_ras_eeprom_read -- read EEPROM
908  * @control: pointer to control structure
909  * @record: array of records to read into
910  * @num: number of records in @record
911  *
912  * Reads num records from the RAS table in EEPROM and
913  * writes the data into @record array.
914  *
915  * Returns 0 on success, -errno on error.
916  */
917 int amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
918 			   struct eeprom_table_record *record,
919 			   const u32 num)
920 {
921 	struct amdgpu_device *adev = to_amdgpu_device(control);
922 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
923 	int i, res;
924 	u8 *buf, *pp;
925 	u32 g0, g1;
926 
927 	if (!__is_ras_eeprom_supported(adev))
928 		return 0;
929 
930 	if (num == 0) {
931 		DRM_ERROR("will not read 0 records\n");
932 		return -EINVAL;
933 	} else if (num > control->ras_num_recs) {
934 		DRM_ERROR("too many records to read:%d available:%d\n",
935 			  num, control->ras_num_recs);
936 		return -EINVAL;
937 	}
938 
939 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
940 	if (!buf)
941 		return -ENOMEM;
942 
943 	/* Determine how many records to read, from the first record
944 	 * index, fri, to the end of the table, and from the beginning
945 	 * of the table, such that the total number of records is
946 	 * @num, and we handle wrap around when fri > 0 and
947 	 * fri + num > RAS_MAX_RECORD_COUNT.
948 	 *
949 	 * First we compute the index of the last element
950 	 * which would be fetched from each region,
951 	 * g0 is in [fri, fri + num - 1], and
952 	 * g1 is in [0, RAS_MAX_RECORD_COUNT - 1].
953 	 * Then, if g0 < RAS_MAX_RECORD_COUNT, the index of
954 	 * the last element to fetch, we set g0 to _the number_
955 	 * of elements to fetch, @num, since we know that the last
956 	 * indexed to be fetched does not exceed the table.
957 	 *
958 	 * If, however, g0 >= RAS_MAX_RECORD_COUNT, then
959 	 * we set g0 to the number of elements to read
960 	 * until the end of the table, and g1 to the number of
961 	 * elements to read from the beginning of the table.
962 	 */
963 	g0 = control->ras_fri + num - 1;
964 	g1 = g0 % control->ras_max_record_count;
965 	if (g0 < control->ras_max_record_count) {
966 		g0 = num;
967 		g1 = 0;
968 	} else {
969 		g0 = control->ras_max_record_count - control->ras_fri;
970 		g1 += 1;
971 	}
972 
973 	mutex_lock(&control->ras_tbl_mutex);
974 	res = __amdgpu_ras_eeprom_read(control, buf, control->ras_fri, g0);
975 	if (res)
976 		goto Out;
977 	if (g1) {
978 		res = __amdgpu_ras_eeprom_read(control,
979 					       buf + g0 * RAS_TABLE_RECORD_SIZE,
980 					       0, g1);
981 		if (res)
982 			goto Out;
983 	}
984 
985 	res = 0;
986 
987 	/* Read up everything? Then transform.
988 	 */
989 	pp = buf;
990 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
991 		__decode_table_record_from_buf(control, &record[i], pp);
992 
993 		/* update bad channel bitmap */
994 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
995 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
996 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
997 			con->update_channel_flag = true;
998 		}
999 	}
1000 Out:
1001 	kfree(buf);
1002 	mutex_unlock(&control->ras_tbl_mutex);
1003 
1004 	return res;
1005 }
1006 
1007 uint32_t amdgpu_ras_eeprom_max_record_count(struct amdgpu_ras_eeprom_control *control)
1008 {
1009 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
1010 		return RAS_MAX_RECORD_COUNT_V2_1;
1011 	else
1012 		return RAS_MAX_RECORD_COUNT;
1013 }
1014 
1015 static ssize_t
1016 amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
1017 				    size_t size, loff_t *pos)
1018 {
1019 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1020 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1021 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1022 	u8 data[50];
1023 	int res;
1024 
1025 	if (!size)
1026 		return size;
1027 
1028 	if (!ras || !control) {
1029 		res = snprintf(data, sizeof(data), "Not supported\n");
1030 	} else {
1031 		res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
1032 			       RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
1033 	}
1034 
1035 	if (*pos >= res)
1036 		return 0;
1037 
1038 	res -= *pos;
1039 	res = min_t(size_t, res, size);
1040 
1041 	if (copy_to_user(buf, &data[*pos], res))
1042 		return -EFAULT;
1043 
1044 	*pos += res;
1045 
1046 	return res;
1047 }
1048 
1049 const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
1050 	.owner = THIS_MODULE,
1051 	.read = amdgpu_ras_debugfs_eeprom_size_read,
1052 	.write = NULL,
1053 	.llseek = default_llseek,
1054 };
1055 
1056 static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
1057 static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
1058 #define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
1059 static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
1060 static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
1061 #define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
1062 
1063 static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
1064 	"ignore",
1065 	"re",
1066 	"ue",
1067 };
1068 
1069 static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
1070 {
1071 	return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
1072 		strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
1073 }
1074 
1075 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
1076 {
1077 	struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
1078 					      eeprom_control);
1079 	struct dentry *de = ras->de_ras_eeprom_table;
1080 
1081 	if (de)
1082 		d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
1083 }
1084 
1085 static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
1086 					     size_t size, loff_t *pos)
1087 {
1088 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1089 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1090 	struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
1091 	const size_t orig_size = size;
1092 	int res = -EFAULT;
1093 	size_t data_len;
1094 
1095 	mutex_lock(&control->ras_tbl_mutex);
1096 
1097 	/* We want *pos - data_len > 0, which means there's
1098 	 * bytes to be printed from data.
1099 	 */
1100 	data_len = strlen(tbl_hdr_str);
1101 	if (*pos < data_len) {
1102 		data_len -= *pos;
1103 		data_len = min_t(size_t, data_len, size);
1104 		if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
1105 			goto Out;
1106 		buf += data_len;
1107 		size -= data_len;
1108 		*pos += data_len;
1109 	}
1110 
1111 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
1112 	if (*pos < data_len && size > 0) {
1113 		u8 data[tbl_hdr_fmt_size + 1];
1114 		loff_t lpos;
1115 
1116 		snprintf(data, sizeof(data), tbl_hdr_fmt,
1117 			 control->tbl_hdr.header,
1118 			 control->tbl_hdr.version,
1119 			 control->tbl_hdr.first_rec_offset,
1120 			 control->tbl_hdr.tbl_size,
1121 			 control->tbl_hdr.checksum);
1122 
1123 		data_len -= *pos;
1124 		data_len = min_t(size_t, data_len, size);
1125 		lpos = *pos - strlen(tbl_hdr_str);
1126 		if (copy_to_user(buf, &data[lpos], data_len))
1127 			goto Out;
1128 		buf += data_len;
1129 		size -= data_len;
1130 		*pos += data_len;
1131 	}
1132 
1133 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
1134 	if (*pos < data_len && size > 0) {
1135 		loff_t lpos;
1136 
1137 		data_len -= *pos;
1138 		data_len = min_t(size_t, data_len, size);
1139 		lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
1140 		if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
1141 			goto Out;
1142 		buf += data_len;
1143 		size -= data_len;
1144 		*pos += data_len;
1145 	}
1146 
1147 	data_len = amdgpu_ras_debugfs_table_size(control);
1148 	if (*pos < data_len && size > 0) {
1149 		u8 dare[RAS_TABLE_RECORD_SIZE];
1150 		u8 data[rec_hdr_fmt_size + 1];
1151 		struct eeprom_table_record record;
1152 		int s, r;
1153 
1154 		/* Find the starting record index
1155 		 */
1156 		s = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1157 			strlen(rec_hdr_str);
1158 		s = s / rec_hdr_fmt_size;
1159 		r = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1160 			strlen(rec_hdr_str);
1161 		r = r % rec_hdr_fmt_size;
1162 
1163 		for ( ; size > 0 && s < control->ras_num_recs; s++) {
1164 			u32 ai = RAS_RI_TO_AI(control, s);
1165 			/* Read a single record
1166 			 */
1167 			res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
1168 			if (res)
1169 				goto Out;
1170 			__decode_table_record_from_buf(control, &record, dare);
1171 			snprintf(data, sizeof(data), rec_hdr_fmt,
1172 				 s,
1173 				 RAS_INDEX_TO_OFFSET(control, ai),
1174 				 record_err_type_str[record.err_type],
1175 				 record.bank,
1176 				 record.ts,
1177 				 record.offset,
1178 				 record.mem_channel,
1179 				 record.mcumc_id,
1180 				 record.retired_page);
1181 
1182 			data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
1183 			if (copy_to_user(buf, &data[r], data_len)) {
1184 				res = -EFAULT;
1185 				goto Out;
1186 			}
1187 			buf += data_len;
1188 			size -= data_len;
1189 			*pos += data_len;
1190 			r = 0;
1191 		}
1192 	}
1193 	res = 0;
1194 Out:
1195 	mutex_unlock(&control->ras_tbl_mutex);
1196 	return res < 0 ? res : orig_size - size;
1197 }
1198 
1199 static ssize_t
1200 amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
1201 				     size_t size, loff_t *pos)
1202 {
1203 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1204 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1205 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1206 	u8 data[81];
1207 	int res;
1208 
1209 	if (!size)
1210 		return size;
1211 
1212 	if (!ras || !control) {
1213 		res = snprintf(data, sizeof(data), "Not supported\n");
1214 		if (*pos >= res)
1215 			return 0;
1216 
1217 		res -= *pos;
1218 		res = min_t(size_t, res, size);
1219 
1220 		if (copy_to_user(buf, &data[*pos], res))
1221 			return -EFAULT;
1222 
1223 		*pos += res;
1224 
1225 		return res;
1226 	} else {
1227 		return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
1228 	}
1229 }
1230 
1231 const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
1232 	.owner = THIS_MODULE,
1233 	.read = amdgpu_ras_debugfs_eeprom_table_read,
1234 	.write = NULL,
1235 	.llseek = default_llseek,
1236 };
1237 
1238 /**
1239  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
1240  * @control: pointer to control structure
1241  *
1242  * Check the checksum of the stored in EEPROM RAS table.
1243  *
1244  * Return 0 if the checksum is correct,
1245  * positive if it is not correct, and
1246  * -errno on I/O error.
1247  */
1248 static int __verify_ras_table_checksum(struct amdgpu_ras_eeprom_control *control)
1249 {
1250 	struct amdgpu_device *adev = to_amdgpu_device(control);
1251 	int buf_size, res;
1252 	u8  csum, *buf, *pp;
1253 
1254 	if (control->tbl_hdr.version == RAS_TABLE_VER_V2_1)
1255 		buf_size = RAS_TABLE_HEADER_SIZE +
1256 			   RAS_TABLE_V2_1_INFO_SIZE +
1257 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1258 	else
1259 		buf_size = RAS_TABLE_HEADER_SIZE +
1260 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1261 
1262 	buf = kzalloc(buf_size, GFP_KERNEL);
1263 	if (!buf) {
1264 		DRM_ERROR("Out of memory checking RAS table checksum.\n");
1265 		return -ENOMEM;
1266 	}
1267 
1268 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1269 				 control->i2c_address +
1270 				 control->ras_header_offset,
1271 				 buf, buf_size);
1272 	if (res < buf_size) {
1273 		DRM_ERROR("Partial read for checksum, res:%d\n", res);
1274 		/* On partial reads, return -EIO.
1275 		 */
1276 		if (res >= 0)
1277 			res = -EIO;
1278 		goto Out;
1279 	}
1280 
1281 	csum = 0;
1282 	for (pp = buf; pp < buf + buf_size; pp++)
1283 		csum += *pp;
1284 Out:
1285 	kfree(buf);
1286 	return res < 0 ? res : csum;
1287 }
1288 
1289 static int __read_table_ras_info(struct amdgpu_ras_eeprom_control *control)
1290 {
1291 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
1292 	struct amdgpu_device *adev = to_amdgpu_device(control);
1293 	unsigned char *buf;
1294 	int res;
1295 
1296 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
1297 	if (!buf) {
1298 		DRM_ERROR("Failed to alloc buf to read EEPROM table ras info\n");
1299 		return -ENOMEM;
1300 	}
1301 
1302 	/**
1303 	 * EEPROM table V2_1 supports ras info,
1304 	 * read EEPROM table ras info
1305 	 */
1306 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1307 				 control->i2c_address + control->ras_info_offset,
1308 				 buf, RAS_TABLE_V2_1_INFO_SIZE);
1309 	if (res < RAS_TABLE_V2_1_INFO_SIZE) {
1310 		DRM_ERROR("Failed to read EEPROM table ras info, res:%d", res);
1311 		res = res >= 0 ? -EIO : res;
1312 		goto Out;
1313 	}
1314 
1315 	__decode_table_ras_info_from_buf(rai, buf);
1316 
1317 Out:
1318 	kfree(buf);
1319 	return res == RAS_TABLE_V2_1_INFO_SIZE ? 0 : res;
1320 }
1321 
1322 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control,
1323 			   bool *exceed_err_limit)
1324 {
1325 	struct amdgpu_device *adev = to_amdgpu_device(control);
1326 	unsigned char buf[RAS_TABLE_HEADER_SIZE] = { 0 };
1327 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1328 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1329 	int res;
1330 
1331 	*exceed_err_limit = false;
1332 
1333 	if (!__is_ras_eeprom_supported(adev))
1334 		return 0;
1335 
1336 	/* Verify i2c adapter is initialized */
1337 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1338 		return -ENOENT;
1339 
1340 	if (!__get_eeprom_i2c_addr(adev, control))
1341 		return -EINVAL;
1342 
1343 	control->ras_header_offset = RAS_HDR_START;
1344 	control->ras_info_offset = RAS_TABLE_V2_1_INFO_START;
1345 	mutex_init(&control->ras_tbl_mutex);
1346 
1347 	/* Read the table header from EEPROM address */
1348 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1349 				 control->i2c_address + control->ras_header_offset,
1350 				 buf, RAS_TABLE_HEADER_SIZE);
1351 	if (res < RAS_TABLE_HEADER_SIZE) {
1352 		DRM_ERROR("Failed to read EEPROM table header, res:%d", res);
1353 		return res >= 0 ? -EIO : res;
1354 	}
1355 
1356 	__decode_table_header_from_buf(hdr, buf);
1357 
1358 	if (hdr->version == RAS_TABLE_VER_V2_1) {
1359 		control->ras_num_recs = RAS_NUM_RECS_V2_1(hdr);
1360 		control->ras_record_offset = RAS_RECORD_START_V2_1;
1361 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT_V2_1;
1362 	} else {
1363 		control->ras_num_recs = RAS_NUM_RECS(hdr);
1364 		control->ras_record_offset = RAS_RECORD_START;
1365 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT;
1366 	}
1367 	control->ras_fri = RAS_OFFSET_TO_INDEX(control, hdr->first_rec_offset);
1368 
1369 	if (hdr->header == RAS_TABLE_HDR_VAL) {
1370 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
1371 				 control->ras_num_recs);
1372 
1373 		if (hdr->version == RAS_TABLE_VER_V2_1) {
1374 			res = __read_table_ras_info(control);
1375 			if (res)
1376 				return res;
1377 		}
1378 
1379 		res = __verify_ras_table_checksum(control);
1380 		if (res)
1381 			DRM_ERROR("RAS table incorrect checksum or error:%d\n",
1382 				  res);
1383 
1384 		/* Warn if we are at 90% of the threshold or above
1385 		 */
1386 		if (10 * control->ras_num_recs >= 9 * ras->bad_page_cnt_threshold)
1387 			dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
1388 					control->ras_num_recs,
1389 					ras->bad_page_cnt_threshold);
1390 	} else if (hdr->header == RAS_TABLE_HDR_BAD &&
1391 		   amdgpu_bad_page_threshold != 0) {
1392 		if (hdr->version == RAS_TABLE_VER_V2_1) {
1393 			res = __read_table_ras_info(control);
1394 			if (res)
1395 				return res;
1396 		}
1397 
1398 		res = __verify_ras_table_checksum(control);
1399 		if (res)
1400 			DRM_ERROR("RAS Table incorrect checksum or error:%d\n",
1401 				  res);
1402 		if (ras->bad_page_cnt_threshold > control->ras_num_recs) {
1403 			/* This means that, the threshold was increased since
1404 			 * the last time the system was booted, and now,
1405 			 * ras->bad_page_cnt_threshold - control->num_recs > 0,
1406 			 * so that at least one more record can be saved,
1407 			 * before the page count threshold is reached.
1408 			 */
1409 			dev_info(adev->dev,
1410 				 "records:%d threshold:%d, resetting "
1411 				 "RAS table header signature",
1412 				 control->ras_num_recs,
1413 				 ras->bad_page_cnt_threshold);
1414 			res = amdgpu_ras_eeprom_correct_header_tag(control,
1415 								   RAS_TABLE_HDR_VAL);
1416 		} else {
1417 			dev_err(adev->dev, "RAS records:%d exceed threshold:%d",
1418 				control->ras_num_recs, ras->bad_page_cnt_threshold);
1419 			if (amdgpu_bad_page_threshold == -1) {
1420 				dev_warn(adev->dev, "GPU will be initialized due to bad_page_threshold = -1.");
1421 				res = 0;
1422 			} else {
1423 				*exceed_err_limit = true;
1424 				dev_err(adev->dev,
1425 					"RAS records:%d exceed threshold:%d, "
1426 					"GPU will not be initialized. Replace this GPU or increase the threshold",
1427 					control->ras_num_recs, ras->bad_page_cnt_threshold);
1428 			}
1429 		}
1430 	} else {
1431 		DRM_INFO("Creating a new EEPROM table");
1432 
1433 		res = amdgpu_ras_eeprom_reset_table(control);
1434 	}
1435 
1436 	return res < 0 ? res : 0;
1437 }
1438