xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision f94877038770073b465eece8636e221653d2beae)
1 /*
2  * Copyright 2019 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include "amdgpu_ras_eeprom.h"
25 #include "amdgpu.h"
26 #include "amdgpu_ras.h"
27 #include <linux/bits.h>
28 #include "atom.h"
29 #include "amdgpu_eeprom.h"
30 #include "amdgpu_atomfirmware.h"
31 #include <linux/debugfs.h>
32 #include <linux/uaccess.h>
33 
34 #include "amdgpu_reset.h"
35 
36 /* These are memory addresses as would be seen by one or more EEPROM
37  * chips strung on the I2C bus, usually by manipulating pins 1-3 of a
38  * set of EEPROM devices. They form a continuous memory space.
39  *
40  * The I2C device address includes the device type identifier, 1010b,
41  * which is a reserved value and indicates that this is an I2C EEPROM
42  * device. It also includes the top 3 bits of the 19 bit EEPROM memory
43  * address, namely bits 18, 17, and 16. This makes up the 7 bit
44  * address sent on the I2C bus with bit 0 being the direction bit,
45  * which is not represented here, and sent by the hardware directly.
46  *
47  * For instance,
48  *   50h = 1010000b => device type identifier 1010b, bits 18:16 = 000b, address 0.
49  *   54h = 1010100b => --"--, bits 18:16 = 100b, address 40000h.
50  *   56h = 1010110b => --"--, bits 18:16 = 110b, address 60000h.
51  * Depending on the size of the I2C EEPROM device(s), bits 18:16 may
52  * address memory in a device or a device on the I2C bus, depending on
53  * the status of pins 1-3. See top of amdgpu_eeprom.c.
54  *
55  * The RAS table lives either at address 0 or address 40000h of EEPROM.
56  */
57 #define EEPROM_I2C_MADDR_0      0x0
58 #define EEPROM_I2C_MADDR_4      0x40000
59 
60 /*
61  * The 2 macros below represent the actual size in bytes that
62  * those entities occupy in the EEPROM memory.
63  * RAS_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
64  * uses uint64 to store 6b fields such as retired_page.
65  */
66 #define RAS_TABLE_HEADER_SIZE   20
67 #define RAS_TABLE_RECORD_SIZE   24
68 
69 /* Table hdr is 'AMDR' */
70 #define RAS_TABLE_HDR_VAL       0x414d4452
71 
72 /* Bad GPU tag ‘BADG’ */
73 #define RAS_TABLE_HDR_BAD       0x42414447
74 
75 /*
76  * EEPROM Table structure v1
77  * ---------------------------------
78  * |                               |
79  * |     EEPROM TABLE HEADER       |
80  * |      ( size 20 Bytes )        |
81  * |                               |
82  * ---------------------------------
83  * |                               |
84  * |    BAD PAGE RECORD AREA       |
85  * |                               |
86  * ---------------------------------
87  */
88 
89 /* Assume 2-Mbit size EEPROM and take up the whole space. */
90 #define RAS_TBL_SIZE_BYTES      (256 * 1024)
91 #define RAS_TABLE_START         0
92 #define RAS_HDR_START           RAS_TABLE_START
93 #define RAS_RECORD_START        (RAS_HDR_START + RAS_TABLE_HEADER_SIZE)
94 #define RAS_MAX_RECORD_COUNT    ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE) \
95 				 / RAS_TABLE_RECORD_SIZE)
96 
97 /*
98  * EEPROM Table structrue v2.1
99  * ---------------------------------
100  * |                               |
101  * |     EEPROM TABLE HEADER       |
102  * |      ( size 20 Bytes )        |
103  * |                               |
104  * ---------------------------------
105  * |                               |
106  * |     EEPROM TABLE RAS INFO     |
107  * | (available info size 4 Bytes) |
108  * |  ( reserved size 252 Bytes )  |
109  * |                               |
110  * ---------------------------------
111  * |                               |
112  * |     BAD PAGE RECORD AREA      |
113  * |                               |
114  * ---------------------------------
115  */
116 
117 /* EEPROM Table V2_1 */
118 #define RAS_TABLE_V2_1_INFO_SIZE       256
119 #define RAS_TABLE_V2_1_INFO_START      RAS_TABLE_HEADER_SIZE
120 #define RAS_RECORD_START_V2_1          (RAS_HDR_START + RAS_TABLE_HEADER_SIZE + \
121 					RAS_TABLE_V2_1_INFO_SIZE)
122 #define RAS_MAX_RECORD_COUNT_V2_1      ((RAS_TBL_SIZE_BYTES - RAS_TABLE_HEADER_SIZE - \
123 					RAS_TABLE_V2_1_INFO_SIZE) \
124 					/ RAS_TABLE_RECORD_SIZE)
125 
126 /* Given a zero-based index of an EEPROM RAS record, yields the EEPROM
127  * offset off of RAS_TABLE_START.  That is, this is something you can
128  * add to control->i2c_address, and then tell I2C layer to read
129  * from/write to there. _N is the so called absolute index,
130  * because it starts right after the table header.
131  */
132 #define RAS_INDEX_TO_OFFSET(_C, _N) ((_C)->ras_record_offset + \
133 				     (_N) * RAS_TABLE_RECORD_SIZE)
134 
135 #define RAS_OFFSET_TO_INDEX(_C, _O) (((_O) - \
136 				      (_C)->ras_record_offset) / RAS_TABLE_RECORD_SIZE)
137 
138 /* Given a 0-based relative record index, 0, 1, 2, ..., etc., off
139  * of "fri", return the absolute record index off of the end of
140  * the table header.
141  */
142 #define RAS_RI_TO_AI(_C, _I) (((_I) + (_C)->ras_fri) % \
143 			      (_C)->ras_max_record_count)
144 
145 #define RAS_NUM_RECS(_tbl_hdr)  (((_tbl_hdr)->tbl_size - \
146 				  RAS_TABLE_HEADER_SIZE) / RAS_TABLE_RECORD_SIZE)
147 
148 #define RAS_NUM_RECS_V2_1(_tbl_hdr)  (((_tbl_hdr)->tbl_size - \
149 				       RAS_TABLE_HEADER_SIZE - \
150 				       RAS_TABLE_V2_1_INFO_SIZE) / RAS_TABLE_RECORD_SIZE)
151 
152 #define to_amdgpu_device(x) ((container_of(x, struct amdgpu_ras, eeprom_control))->adev)
153 
154 static bool __is_ras_eeprom_supported(struct amdgpu_device *adev)
155 {
156 	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
157 	case IP_VERSION(11, 0, 2): /* VEGA20 and ARCTURUS */
158 	case IP_VERSION(11, 0, 7): /* Sienna cichlid */
159 	case IP_VERSION(13, 0, 0):
160 	case IP_VERSION(13, 0, 2): /* Aldebaran */
161 	case IP_VERSION(13, 0, 10):
162 		return true;
163 	case IP_VERSION(13, 0, 6):
164 	case IP_VERSION(13, 0, 12):
165 	case IP_VERSION(13, 0, 14):
166 		return (adev->gmc.is_app_apu) ? false : true;
167 	default:
168 		return false;
169 	}
170 }
171 
172 static bool __get_eeprom_i2c_addr(struct amdgpu_device *adev,
173 				  struct amdgpu_ras_eeprom_control *control)
174 {
175 	struct atom_context *atom_ctx = adev->mode_info.atom_context;
176 	u8 i2c_addr;
177 
178 	if (!control)
179 		return false;
180 
181 	if (adev->bios && amdgpu_atomfirmware_ras_rom_addr(adev, &i2c_addr)) {
182 		/* The address given by VBIOS is an 8-bit, wire-format
183 		 * address, i.e. the most significant byte.
184 		 *
185 		 * Normalize it to a 19-bit EEPROM address. Remove the
186 		 * device type identifier and make it a 7-bit address;
187 		 * then make it a 19-bit EEPROM address. See top of
188 		 * amdgpu_eeprom.c.
189 		 */
190 		i2c_addr = (i2c_addr & 0x0F) >> 1;
191 		control->i2c_address = ((u32) i2c_addr) << 16;
192 
193 		return true;
194 	}
195 
196 	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
197 	case IP_VERSION(11, 0, 2):
198 		/* VEGA20 and ARCTURUS */
199 		if (adev->asic_type == CHIP_VEGA20)
200 			control->i2c_address = EEPROM_I2C_MADDR_0;
201 		else if (strnstr(atom_ctx->vbios_pn,
202 				 "D342",
203 				 sizeof(atom_ctx->vbios_pn)))
204 			control->i2c_address = EEPROM_I2C_MADDR_0;
205 		else
206 			control->i2c_address = EEPROM_I2C_MADDR_4;
207 		return true;
208 	case IP_VERSION(11, 0, 7):
209 		control->i2c_address = EEPROM_I2C_MADDR_0;
210 		return true;
211 	case IP_VERSION(13, 0, 2):
212 		if (strnstr(atom_ctx->vbios_pn, "D673",
213 			    sizeof(atom_ctx->vbios_pn)))
214 			control->i2c_address = EEPROM_I2C_MADDR_4;
215 		else
216 			control->i2c_address = EEPROM_I2C_MADDR_0;
217 		return true;
218 	case IP_VERSION(13, 0, 0):
219 		if (strnstr(atom_ctx->vbios_pn, "D707",
220 			    sizeof(atom_ctx->vbios_pn)))
221 			control->i2c_address = EEPROM_I2C_MADDR_0;
222 		else
223 			control->i2c_address = EEPROM_I2C_MADDR_4;
224 		return true;
225 	case IP_VERSION(13, 0, 6):
226 	case IP_VERSION(13, 0, 10):
227 	case IP_VERSION(13, 0, 12):
228 	case IP_VERSION(13, 0, 14):
229 		control->i2c_address = EEPROM_I2C_MADDR_4;
230 		return true;
231 	default:
232 		return false;
233 	}
234 }
235 
236 static void
237 __encode_table_header_to_buf(struct amdgpu_ras_eeprom_table_header *hdr,
238 			     unsigned char *buf)
239 {
240 	u32 *pp = (uint32_t *)buf;
241 
242 	pp[0] = cpu_to_le32(hdr->header);
243 	pp[1] = cpu_to_le32(hdr->version);
244 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
245 	pp[3] = cpu_to_le32(hdr->tbl_size);
246 	pp[4] = cpu_to_le32(hdr->checksum);
247 }
248 
249 static void
250 __decode_table_header_from_buf(struct amdgpu_ras_eeprom_table_header *hdr,
251 			       unsigned char *buf)
252 {
253 	u32 *pp = (uint32_t *)buf;
254 
255 	hdr->header	      = le32_to_cpu(pp[0]);
256 	hdr->version	      = le32_to_cpu(pp[1]);
257 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
258 	hdr->tbl_size	      = le32_to_cpu(pp[3]);
259 	hdr->checksum	      = le32_to_cpu(pp[4]);
260 }
261 
262 static int __write_table_header(struct amdgpu_ras_eeprom_control *control)
263 {
264 	u8 buf[RAS_TABLE_HEADER_SIZE];
265 	struct amdgpu_device *adev = to_amdgpu_device(control);
266 	int res;
267 
268 	memset(buf, 0, sizeof(buf));
269 	__encode_table_header_to_buf(&control->tbl_hdr, buf);
270 
271 	/* i2c may be unstable in gpu reset */
272 	down_read(&adev->reset_domain->sem);
273 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
274 				  control->i2c_address +
275 				  control->ras_header_offset,
276 				  buf, RAS_TABLE_HEADER_SIZE);
277 	up_read(&adev->reset_domain->sem);
278 
279 	if (res < 0) {
280 		DRM_ERROR("Failed to write EEPROM table header:%d", res);
281 	} else if (res < RAS_TABLE_HEADER_SIZE) {
282 		DRM_ERROR("Short write:%d out of %d\n",
283 			  res, RAS_TABLE_HEADER_SIZE);
284 		res = -EIO;
285 	} else {
286 		res = 0;
287 	}
288 
289 	return res;
290 }
291 
292 static void
293 __encode_table_ras_info_to_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
294 			       unsigned char *buf)
295 {
296 	u32 *pp = (uint32_t *)buf;
297 	u32 tmp;
298 
299 	tmp = ((uint32_t)(rai->rma_status) & 0xFF) |
300 	      (((uint32_t)(rai->health_percent) << 8) & 0xFF00) |
301 	      (((uint32_t)(rai->ecc_page_threshold) << 16) & 0xFFFF0000);
302 	pp[0] = cpu_to_le32(tmp);
303 }
304 
305 static void
306 __decode_table_ras_info_from_buf(struct amdgpu_ras_eeprom_table_ras_info *rai,
307 				 unsigned char *buf)
308 {
309 	u32 *pp = (uint32_t *)buf;
310 	u32 tmp;
311 
312 	tmp = le32_to_cpu(pp[0]);
313 	rai->rma_status = tmp & 0xFF;
314 	rai->health_percent = (tmp >> 8) & 0xFF;
315 	rai->ecc_page_threshold = (tmp >> 16) & 0xFFFF;
316 }
317 
318 static int __write_table_ras_info(struct amdgpu_ras_eeprom_control *control)
319 {
320 	struct amdgpu_device *adev = to_amdgpu_device(control);
321 	u8 *buf;
322 	int res;
323 
324 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
325 	if (!buf) {
326 		DRM_ERROR("Failed to alloc buf to write table ras info\n");
327 		return -ENOMEM;
328 	}
329 
330 	__encode_table_ras_info_to_buf(&control->tbl_rai, buf);
331 
332 	/* i2c may be unstable in gpu reset */
333 	down_read(&adev->reset_domain->sem);
334 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
335 				  control->i2c_address +
336 				  control->ras_info_offset,
337 				  buf, RAS_TABLE_V2_1_INFO_SIZE);
338 	up_read(&adev->reset_domain->sem);
339 
340 	if (res < 0) {
341 		DRM_ERROR("Failed to write EEPROM table ras info:%d", res);
342 	} else if (res < RAS_TABLE_V2_1_INFO_SIZE) {
343 		DRM_ERROR("Short write:%d out of %d\n",
344 			  res, RAS_TABLE_V2_1_INFO_SIZE);
345 		res = -EIO;
346 	} else {
347 		res = 0;
348 	}
349 
350 	kfree(buf);
351 
352 	return res;
353 }
354 
355 static u8 __calc_hdr_byte_sum(const struct amdgpu_ras_eeprom_control *control)
356 {
357 	int ii;
358 	u8  *pp, csum;
359 	size_t sz;
360 
361 	/* Header checksum, skip checksum field in the calculation */
362 	sz = sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum);
363 	pp = (u8 *) &control->tbl_hdr;
364 	csum = 0;
365 	for (ii = 0; ii < sz; ii++, pp++)
366 		csum += *pp;
367 
368 	return csum;
369 }
370 
371 static u8 __calc_ras_info_byte_sum(const struct amdgpu_ras_eeprom_control *control)
372 {
373 	int ii;
374 	u8  *pp, csum;
375 	size_t sz;
376 
377 	sz = sizeof(control->tbl_rai);
378 	pp = (u8 *) &control->tbl_rai;
379 	csum = 0;
380 	for (ii = 0; ii < sz; ii++, pp++)
381 		csum += *pp;
382 
383 	return csum;
384 }
385 
386 static int amdgpu_ras_eeprom_correct_header_tag(
387 	struct amdgpu_ras_eeprom_control *control,
388 	uint32_t header)
389 {
390 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
391 	u8 *hh;
392 	int res;
393 	u8 csum;
394 
395 	csum = -hdr->checksum;
396 
397 	hh = (void *) &hdr->header;
398 	csum -= (hh[0] + hh[1] + hh[2] + hh[3]);
399 	hh = (void *) &header;
400 	csum += hh[0] + hh[1] + hh[2] + hh[3];
401 	csum = -csum;
402 	mutex_lock(&control->ras_tbl_mutex);
403 	hdr->header = header;
404 	hdr->checksum = csum;
405 	res = __write_table_header(control);
406 	mutex_unlock(&control->ras_tbl_mutex);
407 
408 	return res;
409 }
410 
411 static void amdgpu_ras_set_eeprom_table_version(struct amdgpu_ras_eeprom_control *control)
412 {
413 	struct amdgpu_device *adev = to_amdgpu_device(control);
414 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
415 
416 	switch (amdgpu_ip_version(adev, UMC_HWIP, 0)) {
417 	case IP_VERSION(8, 10, 0):
418 		hdr->version = RAS_TABLE_VER_V2_1;
419 		return;
420 	case IP_VERSION(12, 0, 0):
421 	case IP_VERSION(12, 5, 0):
422 		hdr->version = RAS_TABLE_VER_V3;
423 		return;
424 	default:
425 		hdr->version = RAS_TABLE_VER_V1;
426 		return;
427 	}
428 }
429 
430 /**
431  * amdgpu_ras_eeprom_reset_table -- Reset the RAS EEPROM table
432  * @control: pointer to control structure
433  *
434  * Reset the contents of the header of the RAS EEPROM table.
435  * Return 0 on success, -errno on error.
436  */
437 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
438 {
439 	struct amdgpu_device *adev = to_amdgpu_device(control);
440 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
441 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
442 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
443 	u8 csum;
444 	int res;
445 
446 	mutex_lock(&control->ras_tbl_mutex);
447 
448 	hdr->header = RAS_TABLE_HDR_VAL;
449 	amdgpu_ras_set_eeprom_table_version(control);
450 
451 	if (hdr->version >= RAS_TABLE_VER_V2_1) {
452 		hdr->first_rec_offset = RAS_RECORD_START_V2_1;
453 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE +
454 				RAS_TABLE_V2_1_INFO_SIZE;
455 		rai->rma_status = GPU_HEALTH_USABLE;
456 		/**
457 		 * GPU health represented as a percentage.
458 		 * 0 means worst health, 100 means fully health.
459 		 */
460 		rai->health_percent = 100;
461 		/* ecc_page_threshold = 0 means disable bad page retirement */
462 		rai->ecc_page_threshold = con->bad_page_cnt_threshold;
463 	} else {
464 		hdr->first_rec_offset = RAS_RECORD_START;
465 		hdr->tbl_size = RAS_TABLE_HEADER_SIZE;
466 	}
467 
468 	csum = __calc_hdr_byte_sum(control);
469 	if (hdr->version >= RAS_TABLE_VER_V2_1)
470 		csum += __calc_ras_info_byte_sum(control);
471 	csum = -csum;
472 	hdr->checksum = csum;
473 	res = __write_table_header(control);
474 	if (!res && hdr->version > RAS_TABLE_VER_V1)
475 		res = __write_table_ras_info(control);
476 
477 	control->ras_num_recs = 0;
478 	control->ras_num_bad_pages = 0;
479 	control->ras_num_mca_recs = 0;
480 	control->ras_num_pa_recs = 0;
481 	control->ras_fri = 0;
482 
483 	amdgpu_dpm_send_hbm_bad_pages_num(adev, control->ras_num_bad_pages);
484 
485 	control->bad_channel_bitmap = 0;
486 	amdgpu_dpm_send_hbm_bad_channel_flag(adev, control->bad_channel_bitmap);
487 	con->update_channel_flag = false;
488 
489 	amdgpu_ras_debugfs_set_ret_size(control);
490 
491 	mutex_unlock(&control->ras_tbl_mutex);
492 
493 	return res;
494 }
495 
496 static void
497 __encode_table_record_to_buf(struct amdgpu_ras_eeprom_control *control,
498 			     struct eeprom_table_record *record,
499 			     unsigned char *buf)
500 {
501 	__le64 tmp = 0;
502 	int i = 0;
503 
504 	/* Next are all record fields according to EEPROM page spec in LE foramt */
505 	buf[i++] = record->err_type;
506 
507 	buf[i++] = record->bank;
508 
509 	tmp = cpu_to_le64(record->ts);
510 	memcpy(buf + i, &tmp, 8);
511 	i += 8;
512 
513 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
514 	memcpy(buf + i, &tmp, 6);
515 	i += 6;
516 
517 	buf[i++] = record->mem_channel;
518 	buf[i++] = record->mcumc_id;
519 
520 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
521 	memcpy(buf + i, &tmp, 6);
522 }
523 
524 static void
525 __decode_table_record_from_buf(struct amdgpu_ras_eeprom_control *control,
526 			       struct eeprom_table_record *record,
527 			       unsigned char *buf)
528 {
529 	__le64 tmp = 0;
530 	int i =  0;
531 
532 	/* Next are all record fields according to EEPROM page spec in LE foramt */
533 	record->err_type = buf[i++];
534 
535 	record->bank = buf[i++];
536 
537 	memcpy(&tmp, buf + i, 8);
538 	record->ts = le64_to_cpu(tmp);
539 	i += 8;
540 
541 	memcpy(&tmp, buf + i, 6);
542 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
543 	i += 6;
544 
545 	record->mem_channel = buf[i++];
546 	record->mcumc_id = buf[i++];
547 
548 	memcpy(&tmp, buf + i,  6);
549 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
550 }
551 
552 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
553 {
554 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
555 
556 	if (!__is_ras_eeprom_supported(adev) ||
557 	    !amdgpu_bad_page_threshold)
558 		return false;
559 
560 	/* skip check eeprom table for VEGA20 Gaming */
561 	if (!con)
562 		return false;
563 	else
564 		if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
565 			return false;
566 
567 	if (con->eeprom_control.tbl_hdr.header == RAS_TABLE_HDR_BAD) {
568 		if (con->eeprom_control.ras_num_bad_pages > con->bad_page_cnt_threshold)
569 			dev_warn(adev->dev, "RAS records:%d exceed threshold:%d",
570 				 con->eeprom_control.ras_num_bad_pages, con->bad_page_cnt_threshold);
571 		if ((amdgpu_bad_page_threshold == -1) ||
572 		    (amdgpu_bad_page_threshold == -2)) {
573 			dev_warn(adev->dev,
574 				 "Please consult AMD Service Action Guide (SAG) for appropriate service procedures.\n");
575 			return false;
576 		} else {
577 			dev_warn(adev->dev,
578 				 "Please consider adjusting the customized threshold.\n");
579 			return true;
580 		}
581 	}
582 
583 	return false;
584 }
585 
586 /**
587  * __amdgpu_ras_eeprom_write -- write indexed from buffer to EEPROM
588  * @control: pointer to control structure
589  * @buf: pointer to buffer containing data to write
590  * @fri: start writing at this index
591  * @num: number of records to write
592  *
593  * The caller must hold the table mutex in @control.
594  * Return 0 on success, -errno otherwise.
595  */
596 static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control,
597 				     u8 *buf, const u32 fri, const u32 num)
598 {
599 	struct amdgpu_device *adev = to_amdgpu_device(control);
600 	u32 buf_size;
601 	int res;
602 
603 	/* i2c may be unstable in gpu reset */
604 	down_read(&adev->reset_domain->sem);
605 	buf_size = num * RAS_TABLE_RECORD_SIZE;
606 	res = amdgpu_eeprom_write(adev->pm.ras_eeprom_i2c_bus,
607 				  control->i2c_address +
608 				  RAS_INDEX_TO_OFFSET(control, fri),
609 				  buf, buf_size);
610 	up_read(&adev->reset_domain->sem);
611 	if (res < 0) {
612 		DRM_ERROR("Writing %d EEPROM table records error:%d",
613 			  num, res);
614 	} else if (res < buf_size) {
615 		/* Short write, return error.
616 		 */
617 		DRM_ERROR("Wrote %d records out of %d",
618 			  res / RAS_TABLE_RECORD_SIZE, num);
619 		res = -EIO;
620 	} else {
621 		res = 0;
622 	}
623 
624 	return res;
625 }
626 
627 static int
628 amdgpu_ras_eeprom_append_table(struct amdgpu_ras_eeprom_control *control,
629 			       struct eeprom_table_record *record,
630 			       const u32 num)
631 {
632 	struct amdgpu_ras *con = amdgpu_ras_get_context(to_amdgpu_device(control));
633 	struct amdgpu_device *adev = to_amdgpu_device(control);
634 	u32 a, b, i;
635 	u8 *buf, *pp;
636 	int res;
637 
638 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
639 	if (!buf)
640 		return -ENOMEM;
641 
642 	/* Encode all of them in one go.
643 	 */
644 	pp = buf;
645 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
646 		__encode_table_record_to_buf(control, &record[i], pp);
647 
648 		/* update bad channel bitmap */
649 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
650 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
651 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
652 			con->update_channel_flag = true;
653 		}
654 	}
655 
656 	/* a, first record index to write into.
657 	 * b, last record index to write into.
658 	 * a = first index to read (fri) + number of records in the table,
659 	 * b = a + @num - 1.
660 	 * Let N = control->ras_max_num_record_count, then we have,
661 	 * case 0: 0 <= a <= b < N,
662 	 *   just append @num records starting at a;
663 	 * case 1: 0 <= a < N <= b,
664 	 *   append (N - a) records starting at a, and
665 	 *   append the remainder,  b % N + 1, starting at 0.
666 	 * case 2: 0 <= fri < N <= a <= b, then modulo N we get two subcases,
667 	 * case 2a: 0 <= a <= b < N
668 	 *   append num records starting at a; and fix fri if b overwrote it,
669 	 *   and since a <= b, if b overwrote it then a must've also,
670 	 *   and if b didn't overwrite it, then a didn't also.
671 	 * case 2b: 0 <= b < a < N
672 	 *   write num records starting at a, which wraps around 0=N
673 	 *   and overwrite fri unconditionally. Now from case 2a,
674 	 *   this means that b eclipsed fri to overwrite it and wrap
675 	 *   around 0 again, i.e. b = 2N+r pre modulo N, so we unconditionally
676 	 *   set fri = b + 1 (mod N).
677 	 * Now, since fri is updated in every case, except the trivial case 0,
678 	 * the number of records present in the table after writing, is,
679 	 * num_recs - 1 = b - fri (mod N), and we take the positive value,
680 	 * by adding an arbitrary multiple of N before taking the modulo N
681 	 * as shown below.
682 	 */
683 	a = control->ras_fri + control->ras_num_recs;
684 	b = a + num  - 1;
685 	if (b < control->ras_max_record_count) {
686 		res = __amdgpu_ras_eeprom_write(control, buf, a, num);
687 	} else if (a < control->ras_max_record_count) {
688 		u32 g0, g1;
689 
690 		g0 = control->ras_max_record_count - a;
691 		g1 = b % control->ras_max_record_count + 1;
692 		res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
693 		if (res)
694 			goto Out;
695 		res = __amdgpu_ras_eeprom_write(control,
696 						buf + g0 * RAS_TABLE_RECORD_SIZE,
697 						0, g1);
698 		if (res)
699 			goto Out;
700 		if (g1 > control->ras_fri)
701 			control->ras_fri = g1 % control->ras_max_record_count;
702 	} else {
703 		a %= control->ras_max_record_count;
704 		b %= control->ras_max_record_count;
705 
706 		if (a <= b) {
707 			/* Note that, b - a + 1 = num. */
708 			res = __amdgpu_ras_eeprom_write(control, buf, a, num);
709 			if (res)
710 				goto Out;
711 			if (b >= control->ras_fri)
712 				control->ras_fri = (b + 1) % control->ras_max_record_count;
713 		} else {
714 			u32 g0, g1;
715 
716 			/* b < a, which means, we write from
717 			 * a to the end of the table, and from
718 			 * the start of the table to b.
719 			 */
720 			g0 = control->ras_max_record_count - a;
721 			g1 = b + 1;
722 			res = __amdgpu_ras_eeprom_write(control, buf, a, g0);
723 			if (res)
724 				goto Out;
725 			res = __amdgpu_ras_eeprom_write(control,
726 							buf + g0 * RAS_TABLE_RECORD_SIZE,
727 							0, g1);
728 			if (res)
729 				goto Out;
730 			control->ras_fri = g1 % control->ras_max_record_count;
731 		}
732 	}
733 	control->ras_num_recs = 1 + (control->ras_max_record_count + b
734 				     - control->ras_fri)
735 		% control->ras_max_record_count;
736 
737 	/*old asics only save pa to eeprom like before*/
738 	if (IP_VERSION_MAJ(amdgpu_ip_version(adev, UMC_HWIP, 0)) < 12)
739 		control->ras_num_pa_recs += num;
740 	else
741 		control->ras_num_mca_recs += num;
742 
743 	control->ras_num_bad_pages = control->ras_num_pa_recs +
744 				control->ras_num_mca_recs * adev->umc.retire_unit;
745 Out:
746 	kfree(buf);
747 	return res;
748 }
749 
750 static int
751 amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control)
752 {
753 	struct amdgpu_device *adev = to_amdgpu_device(control);
754 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
755 	u8 *buf, *pp, csum;
756 	u32 buf_size;
757 	int res;
758 
759 	/* Modify the header if it exceeds.
760 	 */
761 	if (amdgpu_bad_page_threshold != 0 &&
762 	    control->ras_num_bad_pages > ras->bad_page_cnt_threshold) {
763 		dev_warn(adev->dev,
764 			"Saved bad pages %d reaches threshold value %d\n",
765 			control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
766 		control->tbl_hdr.header = RAS_TABLE_HDR_BAD;
767 		if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1) {
768 			control->tbl_rai.rma_status = GPU_RETIRED__ECC_REACH_THRESHOLD;
769 			control->tbl_rai.health_percent = 0;
770 		}
771 
772 		if ((amdgpu_bad_page_threshold != -1) &&
773 		    (amdgpu_bad_page_threshold != -2))
774 			ras->is_rma = true;
775 
776 		/* ignore the -ENOTSUPP return value */
777 		amdgpu_dpm_send_rma_reason(adev);
778 	}
779 
780 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
781 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
782 					    RAS_TABLE_V2_1_INFO_SIZE +
783 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
784 	else
785 		control->tbl_hdr.tbl_size = RAS_TABLE_HEADER_SIZE +
786 					    control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
787 	control->tbl_hdr.checksum = 0;
788 
789 	buf_size = control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
790 	buf = kcalloc(control->ras_num_recs, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
791 	if (!buf) {
792 		DRM_ERROR("allocating memory for table of size %d bytes failed\n",
793 			  control->tbl_hdr.tbl_size);
794 		res = -ENOMEM;
795 		goto Out;
796 	}
797 
798 	down_read(&adev->reset_domain->sem);
799 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
800 				 control->i2c_address +
801 				 control->ras_record_offset,
802 				 buf, buf_size);
803 	up_read(&adev->reset_domain->sem);
804 	if (res < 0) {
805 		DRM_ERROR("EEPROM failed reading records:%d\n",
806 			  res);
807 		goto Out;
808 	} else if (res < buf_size) {
809 		DRM_ERROR("EEPROM read %d out of %d bytes\n",
810 			  res, buf_size);
811 		res = -EIO;
812 		goto Out;
813 	}
814 
815 	/**
816 	 * bad page records have been stored in eeprom,
817 	 * now calculate gpu health percent
818 	 */
819 	if (amdgpu_bad_page_threshold != 0 &&
820 	    control->tbl_hdr.version >= RAS_TABLE_VER_V2_1 &&
821 	    control->ras_num_bad_pages <= ras->bad_page_cnt_threshold)
822 		control->tbl_rai.health_percent = ((ras->bad_page_cnt_threshold -
823 						   control->ras_num_bad_pages) * 100) /
824 						   ras->bad_page_cnt_threshold;
825 
826 	/* Recalc the checksum.
827 	 */
828 	csum = 0;
829 	for (pp = buf; pp < buf + buf_size; pp++)
830 		csum += *pp;
831 
832 	csum += __calc_hdr_byte_sum(control);
833 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
834 		csum += __calc_ras_info_byte_sum(control);
835 	/* avoid sign extension when assigning to "checksum" */
836 	csum = -csum;
837 	control->tbl_hdr.checksum = csum;
838 	res = __write_table_header(control);
839 	if (!res && control->tbl_hdr.version > RAS_TABLE_VER_V1)
840 		res = __write_table_ras_info(control);
841 Out:
842 	kfree(buf);
843 	return res;
844 }
845 
846 /**
847  * amdgpu_ras_eeprom_append -- append records to the EEPROM RAS table
848  * @control: pointer to control structure
849  * @record: array of records to append
850  * @num: number of records in @record array
851  *
852  * Append @num records to the table, calculate the checksum and write
853  * the table back to EEPROM. The maximum number of records that
854  * can be appended is between 1 and control->ras_max_record_count,
855  * regardless of how many records are already stored in the table.
856  *
857  * Return 0 on success or if EEPROM is not supported, -errno on error.
858  */
859 int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
860 			     struct eeprom_table_record *record,
861 			     const u32 num)
862 {
863 	struct amdgpu_device *adev = to_amdgpu_device(control);
864 	int res, i;
865 	uint64_t nps = AMDGPU_NPS1_PARTITION_MODE;
866 
867 	if (!__is_ras_eeprom_supported(adev))
868 		return 0;
869 
870 	if (num == 0) {
871 		DRM_ERROR("will not append 0 records\n");
872 		return -EINVAL;
873 	} else if (num > control->ras_max_record_count) {
874 		DRM_ERROR("cannot append %d records than the size of table %d\n",
875 			  num, control->ras_max_record_count);
876 		return -EINVAL;
877 	}
878 
879 	if (adev->gmc.gmc_funcs->query_mem_partition_mode)
880 		nps = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
881 
882 	/* set the new channel index flag */
883 	for (i = 0; i < num; i++)
884 		record[i].retired_page |= (nps << UMC_NPS_SHIFT);
885 
886 	mutex_lock(&control->ras_tbl_mutex);
887 
888 	res = amdgpu_ras_eeprom_append_table(control, record, num);
889 	if (!res)
890 		res = amdgpu_ras_eeprom_update_header(control);
891 	if (!res)
892 		amdgpu_ras_debugfs_set_ret_size(control);
893 
894 	mutex_unlock(&control->ras_tbl_mutex);
895 
896 	/* clear channel index flag, the flag is only saved on eeprom */
897 	for (i = 0; i < num; i++)
898 		record[i].retired_page &= ~(nps << UMC_NPS_SHIFT);
899 
900 	return res;
901 }
902 
903 /**
904  * __amdgpu_ras_eeprom_read -- read indexed from EEPROM into buffer
905  * @control: pointer to control structure
906  * @buf: pointer to buffer to read into
907  * @fri: first record index, start reading at this index, absolute index
908  * @num: number of records to read
909  *
910  * The caller must hold the table mutex in @control.
911  * Return 0 on success, -errno otherwise.
912  */
913 static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
914 				    u8 *buf, const u32 fri, const u32 num)
915 {
916 	struct amdgpu_device *adev = to_amdgpu_device(control);
917 	u32 buf_size;
918 	int res;
919 
920 	/* i2c may be unstable in gpu reset */
921 	down_read(&adev->reset_domain->sem);
922 	buf_size = num * RAS_TABLE_RECORD_SIZE;
923 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
924 				 control->i2c_address +
925 				 RAS_INDEX_TO_OFFSET(control, fri),
926 				 buf, buf_size);
927 	up_read(&adev->reset_domain->sem);
928 	if (res < 0) {
929 		DRM_ERROR("Reading %d EEPROM table records error:%d",
930 			  num, res);
931 	} else if (res < buf_size) {
932 		/* Short read, return error.
933 		 */
934 		DRM_ERROR("Read %d records out of %d",
935 			  res / RAS_TABLE_RECORD_SIZE, num);
936 		res = -EIO;
937 	} else {
938 		res = 0;
939 	}
940 
941 	return res;
942 }
943 
944 /**
945  * amdgpu_ras_eeprom_read -- read EEPROM
946  * @control: pointer to control structure
947  * @record: array of records to read into
948  * @num: number of records in @record
949  *
950  * Reads num records from the RAS table in EEPROM and
951  * writes the data into @record array.
952  *
953  * Returns 0 on success, -errno on error.
954  */
955 int amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control,
956 			   struct eeprom_table_record *record,
957 			   const u32 num)
958 {
959 	struct amdgpu_device *adev = to_amdgpu_device(control);
960 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
961 	int i, res;
962 	u8 *buf, *pp;
963 	u32 g0, g1;
964 
965 	if (!__is_ras_eeprom_supported(adev))
966 		return 0;
967 
968 	if (num == 0) {
969 		DRM_ERROR("will not read 0 records\n");
970 		return -EINVAL;
971 	} else if (num > control->ras_num_recs) {
972 		DRM_ERROR("too many records to read:%d available:%d\n",
973 			  num, control->ras_num_recs);
974 		return -EINVAL;
975 	}
976 
977 	buf = kcalloc(num, RAS_TABLE_RECORD_SIZE, GFP_KERNEL);
978 	if (!buf)
979 		return -ENOMEM;
980 
981 	/* Determine how many records to read, from the first record
982 	 * index, fri, to the end of the table, and from the beginning
983 	 * of the table, such that the total number of records is
984 	 * @num, and we handle wrap around when fri > 0 and
985 	 * fri + num > RAS_MAX_RECORD_COUNT.
986 	 *
987 	 * First we compute the index of the last element
988 	 * which would be fetched from each region,
989 	 * g0 is in [fri, fri + num - 1], and
990 	 * g1 is in [0, RAS_MAX_RECORD_COUNT - 1].
991 	 * Then, if g0 < RAS_MAX_RECORD_COUNT, the index of
992 	 * the last element to fetch, we set g0 to _the number_
993 	 * of elements to fetch, @num, since we know that the last
994 	 * indexed to be fetched does not exceed the table.
995 	 *
996 	 * If, however, g0 >= RAS_MAX_RECORD_COUNT, then
997 	 * we set g0 to the number of elements to read
998 	 * until the end of the table, and g1 to the number of
999 	 * elements to read from the beginning of the table.
1000 	 */
1001 	g0 = control->ras_fri + num - 1;
1002 	g1 = g0 % control->ras_max_record_count;
1003 	if (g0 < control->ras_max_record_count) {
1004 		g0 = num;
1005 		g1 = 0;
1006 	} else {
1007 		g0 = control->ras_max_record_count - control->ras_fri;
1008 		g1 += 1;
1009 	}
1010 
1011 	mutex_lock(&control->ras_tbl_mutex);
1012 	res = __amdgpu_ras_eeprom_read(control, buf, control->ras_fri, g0);
1013 	if (res)
1014 		goto Out;
1015 	if (g1) {
1016 		res = __amdgpu_ras_eeprom_read(control,
1017 					       buf + g0 * RAS_TABLE_RECORD_SIZE,
1018 					       0, g1);
1019 		if (res)
1020 			goto Out;
1021 	}
1022 
1023 	res = 0;
1024 
1025 	/* Read up everything? Then transform.
1026 	 */
1027 	pp = buf;
1028 	for (i = 0; i < num; i++, pp += RAS_TABLE_RECORD_SIZE) {
1029 		__decode_table_record_from_buf(control, &record[i], pp);
1030 
1031 		/* update bad channel bitmap */
1032 		if ((record[i].mem_channel < BITS_PER_TYPE(control->bad_channel_bitmap)) &&
1033 		    !(control->bad_channel_bitmap & (1 << record[i].mem_channel))) {
1034 			control->bad_channel_bitmap |= 1 << record[i].mem_channel;
1035 			con->update_channel_flag = true;
1036 		}
1037 	}
1038 Out:
1039 	kfree(buf);
1040 	mutex_unlock(&control->ras_tbl_mutex);
1041 
1042 	return res;
1043 }
1044 
1045 uint32_t amdgpu_ras_eeprom_max_record_count(struct amdgpu_ras_eeprom_control *control)
1046 {
1047 	/* get available eeprom table version first before eeprom table init */
1048 	amdgpu_ras_set_eeprom_table_version(control);
1049 
1050 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
1051 		return RAS_MAX_RECORD_COUNT_V2_1;
1052 	else
1053 		return RAS_MAX_RECORD_COUNT;
1054 }
1055 
1056 static ssize_t
1057 amdgpu_ras_debugfs_eeprom_size_read(struct file *f, char __user *buf,
1058 				    size_t size, loff_t *pos)
1059 {
1060 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1061 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1062 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1063 	u8 data[50];
1064 	int res;
1065 
1066 	if (!size)
1067 		return size;
1068 
1069 	if (!ras || !control) {
1070 		res = snprintf(data, sizeof(data), "Not supported\n");
1071 	} else {
1072 		res = snprintf(data, sizeof(data), "%d bytes or %d records\n",
1073 			       RAS_TBL_SIZE_BYTES, control->ras_max_record_count);
1074 	}
1075 
1076 	if (*pos >= res)
1077 		return 0;
1078 
1079 	res -= *pos;
1080 	res = min_t(size_t, res, size);
1081 
1082 	if (copy_to_user(buf, &data[*pos], res))
1083 		return -EFAULT;
1084 
1085 	*pos += res;
1086 
1087 	return res;
1088 }
1089 
1090 const struct file_operations amdgpu_ras_debugfs_eeprom_size_ops = {
1091 	.owner = THIS_MODULE,
1092 	.read = amdgpu_ras_debugfs_eeprom_size_read,
1093 	.write = NULL,
1094 	.llseek = default_llseek,
1095 };
1096 
1097 static const char *tbl_hdr_str = " Signature    Version  FirstOffs       Size   Checksum\n";
1098 static const char *tbl_hdr_fmt = "0x%08X 0x%08X 0x%08X 0x%08X 0x%08X\n";
1099 #define tbl_hdr_fmt_size (5 * (2+8) + 4 + 1)
1100 static const char *rec_hdr_str = "Index  Offset ErrType Bank/CU          TimeStamp      Offs/Addr MemChl MCUMCID    RetiredPage\n";
1101 static const char *rec_hdr_fmt = "%5d 0x%05X %7s    0x%02X 0x%016llX 0x%012llX   0x%02X    0x%02X 0x%012llX\n";
1102 #define rec_hdr_fmt_size (5 + 1 + 7 + 1 + 7 + 1 + 7 + 1 + 18 + 1 + 14 + 1 + 6 + 1 + 7 + 1 + 14 + 1)
1103 
1104 static const char *record_err_type_str[AMDGPU_RAS_EEPROM_ERR_COUNT] = {
1105 	"ignore",
1106 	"re",
1107 	"ue",
1108 };
1109 
1110 static loff_t amdgpu_ras_debugfs_table_size(struct amdgpu_ras_eeprom_control *control)
1111 {
1112 	return strlen(tbl_hdr_str) + tbl_hdr_fmt_size +
1113 		strlen(rec_hdr_str) + rec_hdr_fmt_size * control->ras_num_recs;
1114 }
1115 
1116 void amdgpu_ras_debugfs_set_ret_size(struct amdgpu_ras_eeprom_control *control)
1117 {
1118 	struct amdgpu_ras *ras = container_of(control, struct amdgpu_ras,
1119 					      eeprom_control);
1120 	struct dentry *de = ras->de_ras_eeprom_table;
1121 
1122 	if (de)
1123 		d_inode(de)->i_size = amdgpu_ras_debugfs_table_size(control);
1124 }
1125 
1126 static ssize_t amdgpu_ras_debugfs_table_read(struct file *f, char __user *buf,
1127 					     size_t size, loff_t *pos)
1128 {
1129 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1130 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1131 	struct amdgpu_ras_eeprom_control *control = &ras->eeprom_control;
1132 	const size_t orig_size = size;
1133 	int res = -EFAULT;
1134 	size_t data_len;
1135 
1136 	mutex_lock(&control->ras_tbl_mutex);
1137 
1138 	/* We want *pos - data_len > 0, which means there's
1139 	 * bytes to be printed from data.
1140 	 */
1141 	data_len = strlen(tbl_hdr_str);
1142 	if (*pos < data_len) {
1143 		data_len -= *pos;
1144 		data_len = min_t(size_t, data_len, size);
1145 		if (copy_to_user(buf, &tbl_hdr_str[*pos], data_len))
1146 			goto Out;
1147 		buf += data_len;
1148 		size -= data_len;
1149 		*pos += data_len;
1150 	}
1151 
1152 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size;
1153 	if (*pos < data_len && size > 0) {
1154 		u8 data[tbl_hdr_fmt_size + 1];
1155 		loff_t lpos;
1156 
1157 		snprintf(data, sizeof(data), tbl_hdr_fmt,
1158 			 control->tbl_hdr.header,
1159 			 control->tbl_hdr.version,
1160 			 control->tbl_hdr.first_rec_offset,
1161 			 control->tbl_hdr.tbl_size,
1162 			 control->tbl_hdr.checksum);
1163 
1164 		data_len -= *pos;
1165 		data_len = min_t(size_t, data_len, size);
1166 		lpos = *pos - strlen(tbl_hdr_str);
1167 		if (copy_to_user(buf, &data[lpos], data_len))
1168 			goto Out;
1169 		buf += data_len;
1170 		size -= data_len;
1171 		*pos += data_len;
1172 	}
1173 
1174 	data_len = strlen(tbl_hdr_str) + tbl_hdr_fmt_size + strlen(rec_hdr_str);
1175 	if (*pos < data_len && size > 0) {
1176 		loff_t lpos;
1177 
1178 		data_len -= *pos;
1179 		data_len = min_t(size_t, data_len, size);
1180 		lpos = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size;
1181 		if (copy_to_user(buf, &rec_hdr_str[lpos], data_len))
1182 			goto Out;
1183 		buf += data_len;
1184 		size -= data_len;
1185 		*pos += data_len;
1186 	}
1187 
1188 	data_len = amdgpu_ras_debugfs_table_size(control);
1189 	if (*pos < data_len && size > 0) {
1190 		u8 dare[RAS_TABLE_RECORD_SIZE];
1191 		u8 data[rec_hdr_fmt_size + 1];
1192 		struct eeprom_table_record record;
1193 		int s, r;
1194 
1195 		/* Find the starting record index
1196 		 */
1197 		s = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1198 			strlen(rec_hdr_str);
1199 		s = s / rec_hdr_fmt_size;
1200 		r = *pos - strlen(tbl_hdr_str) - tbl_hdr_fmt_size -
1201 			strlen(rec_hdr_str);
1202 		r = r % rec_hdr_fmt_size;
1203 
1204 		for ( ; size > 0 && s < control->ras_num_recs; s++) {
1205 			u32 ai = RAS_RI_TO_AI(control, s);
1206 			/* Read a single record
1207 			 */
1208 			res = __amdgpu_ras_eeprom_read(control, dare, ai, 1);
1209 			if (res)
1210 				goto Out;
1211 			__decode_table_record_from_buf(control, &record, dare);
1212 			snprintf(data, sizeof(data), rec_hdr_fmt,
1213 				 s,
1214 				 RAS_INDEX_TO_OFFSET(control, ai),
1215 				 record_err_type_str[record.err_type],
1216 				 record.bank,
1217 				 record.ts,
1218 				 record.offset,
1219 				 record.mem_channel,
1220 				 record.mcumc_id,
1221 				 record.retired_page);
1222 
1223 			data_len = min_t(size_t, rec_hdr_fmt_size - r, size);
1224 			if (copy_to_user(buf, &data[r], data_len)) {
1225 				res = -EFAULT;
1226 				goto Out;
1227 			}
1228 			buf += data_len;
1229 			size -= data_len;
1230 			*pos += data_len;
1231 			r = 0;
1232 		}
1233 	}
1234 	res = 0;
1235 Out:
1236 	mutex_unlock(&control->ras_tbl_mutex);
1237 	return res < 0 ? res : orig_size - size;
1238 }
1239 
1240 static ssize_t
1241 amdgpu_ras_debugfs_eeprom_table_read(struct file *f, char __user *buf,
1242 				     size_t size, loff_t *pos)
1243 {
1244 	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
1245 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1246 	struct amdgpu_ras_eeprom_control *control = ras ? &ras->eeprom_control : NULL;
1247 	u8 data[81];
1248 	int res;
1249 
1250 	if (!size)
1251 		return size;
1252 
1253 	if (!ras || !control) {
1254 		res = snprintf(data, sizeof(data), "Not supported\n");
1255 		if (*pos >= res)
1256 			return 0;
1257 
1258 		res -= *pos;
1259 		res = min_t(size_t, res, size);
1260 
1261 		if (copy_to_user(buf, &data[*pos], res))
1262 			return -EFAULT;
1263 
1264 		*pos += res;
1265 
1266 		return res;
1267 	} else {
1268 		return amdgpu_ras_debugfs_table_read(f, buf, size, pos);
1269 	}
1270 }
1271 
1272 const struct file_operations amdgpu_ras_debugfs_eeprom_table_ops = {
1273 	.owner = THIS_MODULE,
1274 	.read = amdgpu_ras_debugfs_eeprom_table_read,
1275 	.write = NULL,
1276 	.llseek = default_llseek,
1277 };
1278 
1279 /**
1280  * __verify_ras_table_checksum -- verify the RAS EEPROM table checksum
1281  * @control: pointer to control structure
1282  *
1283  * Check the checksum of the stored in EEPROM RAS table.
1284  *
1285  * Return 0 if the checksum is correct,
1286  * positive if it is not correct, and
1287  * -errno on I/O error.
1288  */
1289 static int __verify_ras_table_checksum(struct amdgpu_ras_eeprom_control *control)
1290 {
1291 	struct amdgpu_device *adev = to_amdgpu_device(control);
1292 	int buf_size, res;
1293 	u8  csum, *buf, *pp;
1294 
1295 	if (control->tbl_hdr.version >= RAS_TABLE_VER_V2_1)
1296 		buf_size = RAS_TABLE_HEADER_SIZE +
1297 			   RAS_TABLE_V2_1_INFO_SIZE +
1298 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1299 	else
1300 		buf_size = RAS_TABLE_HEADER_SIZE +
1301 			   control->ras_num_recs * RAS_TABLE_RECORD_SIZE;
1302 
1303 	buf = kzalloc(buf_size, GFP_KERNEL);
1304 	if (!buf) {
1305 		DRM_ERROR("Out of memory checking RAS table checksum.\n");
1306 		return -ENOMEM;
1307 	}
1308 
1309 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1310 				 control->i2c_address +
1311 				 control->ras_header_offset,
1312 				 buf, buf_size);
1313 	if (res < buf_size) {
1314 		DRM_ERROR("Partial read for checksum, res:%d\n", res);
1315 		/* On partial reads, return -EIO.
1316 		 */
1317 		if (res >= 0)
1318 			res = -EIO;
1319 		goto Out;
1320 	}
1321 
1322 	csum = 0;
1323 	for (pp = buf; pp < buf + buf_size; pp++)
1324 		csum += *pp;
1325 Out:
1326 	kfree(buf);
1327 	return res < 0 ? res : csum;
1328 }
1329 
1330 static int __read_table_ras_info(struct amdgpu_ras_eeprom_control *control)
1331 {
1332 	struct amdgpu_ras_eeprom_table_ras_info *rai = &control->tbl_rai;
1333 	struct amdgpu_device *adev = to_amdgpu_device(control);
1334 	unsigned char *buf;
1335 	int res;
1336 
1337 	buf = kzalloc(RAS_TABLE_V2_1_INFO_SIZE, GFP_KERNEL);
1338 	if (!buf) {
1339 		DRM_ERROR("Failed to alloc buf to read EEPROM table ras info\n");
1340 		return -ENOMEM;
1341 	}
1342 
1343 	/**
1344 	 * EEPROM table V2_1 supports ras info,
1345 	 * read EEPROM table ras info
1346 	 */
1347 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1348 				 control->i2c_address + control->ras_info_offset,
1349 				 buf, RAS_TABLE_V2_1_INFO_SIZE);
1350 	if (res < RAS_TABLE_V2_1_INFO_SIZE) {
1351 		DRM_ERROR("Failed to read EEPROM table ras info, res:%d", res);
1352 		res = res >= 0 ? -EIO : res;
1353 		goto Out;
1354 	}
1355 
1356 	__decode_table_ras_info_from_buf(rai, buf);
1357 
1358 Out:
1359 	kfree(buf);
1360 	return res == RAS_TABLE_V2_1_INFO_SIZE ? 0 : res;
1361 }
1362 
1363 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control)
1364 {
1365 	struct amdgpu_device *adev = to_amdgpu_device(control);
1366 	unsigned char buf[RAS_TABLE_HEADER_SIZE] = { 0 };
1367 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1368 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1369 	int res;
1370 
1371 	ras->is_rma = false;
1372 
1373 	if (!__is_ras_eeprom_supported(adev))
1374 		return 0;
1375 
1376 	/* Verify i2c adapter is initialized */
1377 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1378 		return -ENOENT;
1379 
1380 	if (!__get_eeprom_i2c_addr(adev, control))
1381 		return -EINVAL;
1382 
1383 	control->ras_header_offset = RAS_HDR_START;
1384 	control->ras_info_offset = RAS_TABLE_V2_1_INFO_START;
1385 	mutex_init(&control->ras_tbl_mutex);
1386 
1387 	/* Read the table header from EEPROM address */
1388 	res = amdgpu_eeprom_read(adev->pm.ras_eeprom_i2c_bus,
1389 				 control->i2c_address + control->ras_header_offset,
1390 				 buf, RAS_TABLE_HEADER_SIZE);
1391 	if (res < RAS_TABLE_HEADER_SIZE) {
1392 		DRM_ERROR("Failed to read EEPROM table header, res:%d", res);
1393 		return res >= 0 ? -EIO : res;
1394 	}
1395 
1396 	__decode_table_header_from_buf(hdr, buf);
1397 
1398 	if (hdr->header != RAS_TABLE_HDR_VAL &&
1399 	    hdr->header != RAS_TABLE_HDR_BAD) {
1400 		dev_info(adev->dev, "Creating a new EEPROM table");
1401 		return amdgpu_ras_eeprom_reset_table(control);
1402 	}
1403 
1404 	switch (hdr->version) {
1405 	case RAS_TABLE_VER_V2_1:
1406 	case RAS_TABLE_VER_V3:
1407 		control->ras_num_recs = RAS_NUM_RECS_V2_1(hdr);
1408 		control->ras_record_offset = RAS_RECORD_START_V2_1;
1409 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT_V2_1;
1410 		break;
1411 	case RAS_TABLE_VER_V1:
1412 		control->ras_num_recs = RAS_NUM_RECS(hdr);
1413 		control->ras_record_offset = RAS_RECORD_START;
1414 		control->ras_max_record_count = RAS_MAX_RECORD_COUNT;
1415 		break;
1416 	default:
1417 		dev_err(adev->dev,
1418 			"RAS header invalid, unsupported version: %u",
1419 			hdr->version);
1420 		return -EINVAL;
1421 	}
1422 
1423 	if (control->ras_num_recs > control->ras_max_record_count) {
1424 		dev_err(adev->dev,
1425 			"RAS header invalid, records in header: %u max allowed :%u",
1426 			control->ras_num_recs, control->ras_max_record_count);
1427 		return -EINVAL;
1428 	}
1429 
1430 	control->ras_fri = RAS_OFFSET_TO_INDEX(control, hdr->first_rec_offset);
1431 	control->ras_num_mca_recs = 0;
1432 	control->ras_num_pa_recs = 0;
1433 	return 0;
1434 }
1435 
1436 int amdgpu_ras_eeprom_check(struct amdgpu_ras_eeprom_control *control)
1437 {
1438 	struct amdgpu_device *adev = to_amdgpu_device(control);
1439 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
1440 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
1441 	int res = 0;
1442 
1443 	if (!__is_ras_eeprom_supported(adev))
1444 		return 0;
1445 
1446 	/* Verify i2c adapter is initialized */
1447 	if (!adev->pm.ras_eeprom_i2c_bus || !adev->pm.ras_eeprom_i2c_bus->algo)
1448 		return -ENOENT;
1449 
1450 	if (!__get_eeprom_i2c_addr(adev, control))
1451 		return -EINVAL;
1452 
1453 	control->ras_num_bad_pages = control->ras_num_pa_recs +
1454 			control->ras_num_mca_recs * adev->umc.retire_unit;
1455 
1456 	if (hdr->header == RAS_TABLE_HDR_VAL) {
1457 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
1458 				 control->ras_num_bad_pages);
1459 
1460 		if (hdr->version >= RAS_TABLE_VER_V2_1) {
1461 			res = __read_table_ras_info(control);
1462 			if (res)
1463 				return res;
1464 		}
1465 
1466 		res = __verify_ras_table_checksum(control);
1467 		if (res)
1468 			dev_err(adev->dev,
1469 				"RAS table incorrect checksum or error:%d\n",
1470 				res);
1471 
1472 		/* Warn if we are at 90% of the threshold or above
1473 		 */
1474 		if (10 * control->ras_num_bad_pages >= 9 * ras->bad_page_cnt_threshold)
1475 			dev_warn(adev->dev, "RAS records:%u exceeds 90%% of threshold:%d",
1476 					control->ras_num_bad_pages,
1477 					ras->bad_page_cnt_threshold);
1478 	} else if (hdr->header == RAS_TABLE_HDR_BAD &&
1479 		   amdgpu_bad_page_threshold != 0) {
1480 		if (hdr->version >= RAS_TABLE_VER_V2_1) {
1481 			res = __read_table_ras_info(control);
1482 			if (res)
1483 				return res;
1484 		}
1485 
1486 		res = __verify_ras_table_checksum(control);
1487 		if (res) {
1488 			dev_err(adev->dev,
1489 				"RAS Table incorrect checksum or error:%d\n",
1490 				res);
1491 			return -EINVAL;
1492 		}
1493 		if (ras->bad_page_cnt_threshold >= control->ras_num_bad_pages) {
1494 			/* This means that, the threshold was increased since
1495 			 * the last time the system was booted, and now,
1496 			 * ras->bad_page_cnt_threshold - control->num_recs > 0,
1497 			 * so that at least one more record can be saved,
1498 			 * before the page count threshold is reached.
1499 			 */
1500 			dev_info(adev->dev,
1501 				 "records:%d threshold:%d, resetting "
1502 				 "RAS table header signature",
1503 				 control->ras_num_bad_pages,
1504 				 ras->bad_page_cnt_threshold);
1505 			res = amdgpu_ras_eeprom_correct_header_tag(control,
1506 								   RAS_TABLE_HDR_VAL);
1507 		} else {
1508 			dev_warn(adev->dev,
1509 				"RAS records:%d exceed threshold:%d\n",
1510 				control->ras_num_bad_pages, ras->bad_page_cnt_threshold);
1511 			if ((amdgpu_bad_page_threshold == -1) ||
1512 			    (amdgpu_bad_page_threshold == -2)) {
1513 				res = 0;
1514 				dev_warn(adev->dev,
1515 					 "Please consult AMD Service Action Guide (SAG) for appropriate service procedures\n");
1516 			} else {
1517 				ras->is_rma = true;
1518 				dev_warn(adev->dev,
1519 					 "User defined threshold is set, runtime service will be halt when threshold is reached\n");
1520 			}
1521 		}
1522 	}
1523 
1524 	return res < 0 ? res : 0;
1525 }
1526