xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision 4c283fdac08abf3211533f70623c90a34f41d08d)
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 "smu_v11_0_i2c.h"
29 
30 #define EEPROM_I2C_TARGET_ADDR 0xA0
31 
32 /*
33  * The 2 macros bellow represent the actual size in bytes that
34  * those entities occupy in the EEPROM memory.
35  * EEPROM_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
36  * uses uint64 to store 6b fields such as retired_page.
37  */
38 #define EEPROM_TABLE_HEADER_SIZE 20
39 #define EEPROM_TABLE_RECORD_SIZE 24
40 
41 #define EEPROM_ADDRESS_SIZE 0x2
42 
43 /* Table hdr is 'AMDR' */
44 #define EEPROM_TABLE_HDR_VAL 0x414d4452
45 #define EEPROM_TABLE_VER 0x00010000
46 
47 /* Assume 2 Mbit size */
48 #define EEPROM_SIZE_BYTES 256000
49 #define EEPROM_PAGE__SIZE_BYTES 256
50 #define EEPROM_HDR_START 0
51 #define EEPROM_RECORD_START (EEPROM_HDR_START + EEPROM_TABLE_HEADER_SIZE)
52 #define EEPROM_MAX_RECORD_NUM ((EEPROM_SIZE_BYTES - EEPROM_TABLE_HEADER_SIZE) / EEPROM_TABLE_RECORD_SIZE)
53 #define EEPROM_ADDR_MSB_MASK GENMASK(17, 8)
54 
55 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_ras, eeprom_control))->adev
56 
57 static void __encode_table_header_to_buff(struct amdgpu_ras_eeprom_table_header *hdr,
58 					  unsigned char *buff)
59 {
60 	uint32_t *pp = (uint32_t *) buff;
61 
62 	pp[0] = cpu_to_le32(hdr->header);
63 	pp[1] = cpu_to_le32(hdr->version);
64 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
65 	pp[3] = cpu_to_le32(hdr->tbl_size);
66 	pp[4] = cpu_to_le32(hdr->checksum);
67 }
68 
69 static void __decode_table_header_from_buff(struct amdgpu_ras_eeprom_table_header *hdr,
70 					  unsigned char *buff)
71 {
72 	uint32_t *pp = (uint32_t *)buff;
73 
74 	hdr->header 	      = le32_to_cpu(pp[0]);
75 	hdr->version 	      = le32_to_cpu(pp[1]);
76 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
77 	hdr->tbl_size 	      = le32_to_cpu(pp[3]);
78 	hdr->checksum 	      = le32_to_cpu(pp[4]);
79 }
80 
81 static int __update_table_header(struct amdgpu_ras_eeprom_control *control,
82 				 unsigned char *buff)
83 {
84 	int ret = 0;
85 	struct i2c_msg msg = {
86 			.addr	= EEPROM_I2C_TARGET_ADDR,
87 			.flags	= 0,
88 			.len	= EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
89 			.buf	= buff,
90 	};
91 
92 
93 	*(uint16_t *)buff = EEPROM_HDR_START;
94 	__encode_table_header_to_buff(&control->tbl_hdr, buff + EEPROM_ADDRESS_SIZE);
95 
96 	ret = i2c_transfer(&control->eeprom_accessor, &msg, 1);
97 	if (ret < 1)
98 		DRM_ERROR("Failed to write EEPROM table header, ret:%d", ret);
99 
100 	return ret;
101 }
102 
103 
104 
105 static uint32_t  __calc_hdr_byte_sum(struct amdgpu_ras_eeprom_control *control)
106 {
107 	int i;
108 	uint32_t tbl_sum = 0;
109 
110 	/* Header checksum, skip checksum field in the calculation */
111 	for (i = 0; i < sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum); i++)
112 		tbl_sum += *(((unsigned char *)&control->tbl_hdr) + i);
113 
114 	return tbl_sum;
115 }
116 
117 static uint32_t  __calc_recs_byte_sum(struct eeprom_table_record *records,
118 				      int num)
119 {
120 	int i, j;
121 	uint32_t tbl_sum = 0;
122 
123 	/* Records checksum */
124 	for (i = 0; i < num; i++) {
125 		struct eeprom_table_record *record = &records[i];
126 
127 		for (j = 0; j < sizeof(*record); j++) {
128 			tbl_sum += *(((unsigned char *)record) + j);
129 		}
130 	}
131 
132 	return tbl_sum;
133 }
134 
135 static inline uint32_t  __calc_tbl_byte_sum(struct amdgpu_ras_eeprom_control *control,
136 				  struct eeprom_table_record *records, int num)
137 {
138 	return __calc_hdr_byte_sum(control) + __calc_recs_byte_sum(records, num);
139 }
140 
141 /* Checksum = 256 -((sum of all table entries) mod 256) */
142 static void __update_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
143 				  struct eeprom_table_record *records, int num,
144 				  uint32_t old_hdr_byte_sum)
145 {
146 	/*
147 	 * This will update the table sum with new records.
148 	 *
149 	 * TODO: What happens when the EEPROM table is to be wrapped around
150 	 * and old records from start will get overridden.
151 	 */
152 
153 	/* need to recalculate updated header byte sum */
154 	control->tbl_byte_sum -= old_hdr_byte_sum;
155 	control->tbl_byte_sum += __calc_tbl_byte_sum(control, records, num);
156 
157 	control->tbl_hdr.checksum = 256 - (control->tbl_byte_sum % 256);
158 }
159 
160 /* table sum mod 256 + checksum must equals 256 */
161 static bool __validate_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
162 			    struct eeprom_table_record *records, int num)
163 {
164 	control->tbl_byte_sum = __calc_tbl_byte_sum(control, records, num);
165 
166 	if (control->tbl_hdr.checksum + (control->tbl_byte_sum % 256) != 256) {
167 		DRM_WARN("Checksum mismatch, checksum: %u ", control->tbl_hdr.checksum);
168 		return false;
169 	}
170 
171 	return true;
172 }
173 
174 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
175 {
176 	unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE] = { 0 };
177 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
178 	int ret = 0;
179 
180 	mutex_lock(&control->tbl_mutex);
181 
182 	hdr->header = EEPROM_TABLE_HDR_VAL;
183 	hdr->version = EEPROM_TABLE_VER;
184 	hdr->first_rec_offset = EEPROM_RECORD_START;
185 	hdr->tbl_size = EEPROM_TABLE_HEADER_SIZE;
186 
187 	control->tbl_byte_sum = 0;
188 	__update_tbl_checksum(control, NULL, 0, 0);
189 	control->next_addr = EEPROM_RECORD_START;
190 
191 	ret = __update_table_header(control, buff);
192 
193 	mutex_unlock(&control->tbl_mutex);
194 
195 	return ret;
196 
197 }
198 
199 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control)
200 {
201 	int ret = 0;
202 	struct amdgpu_device *adev = to_amdgpu_device(control);
203 	unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE] = { 0 };
204 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
205 	struct i2c_msg msg = {
206 			.addr	= EEPROM_I2C_TARGET_ADDR,
207 			.flags	= I2C_M_RD,
208 			.len	= EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
209 			.buf	= buff,
210 	};
211 
212 	mutex_init(&control->tbl_mutex);
213 
214 	switch (adev->asic_type) {
215 	case CHIP_VEGA20:
216 		ret = smu_v11_0_i2c_eeprom_control_init(&control->eeprom_accessor);
217 		break;
218 
219 	default:
220 		return 0;
221 	}
222 
223 	if (ret) {
224 		DRM_ERROR("Failed to init I2C controller, ret:%d", ret);
225 		return ret;
226 	}
227 
228 	/* Read/Create table header from EEPROM address 0 */
229 	ret = i2c_transfer(&control->eeprom_accessor, &msg, 1);
230 	if (ret < 1) {
231 		DRM_ERROR("Failed to read EEPROM table header, ret:%d", ret);
232 		return ret;
233 	}
234 
235 	__decode_table_header_from_buff(hdr, &buff[2]);
236 
237 	if (hdr->header == EEPROM_TABLE_HDR_VAL) {
238 		control->num_recs = (hdr->tbl_size - EEPROM_TABLE_HEADER_SIZE) /
239 				    EEPROM_TABLE_RECORD_SIZE;
240 		control->tbl_byte_sum = __calc_hdr_byte_sum(control);
241 		control->next_addr = EEPROM_RECORD_START;
242 
243 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
244 				 control->num_recs);
245 
246 	} else {
247 		DRM_INFO("Creating new EEPROM table");
248 
249 		ret = amdgpu_ras_eeprom_reset_table(control);
250 	}
251 
252 	return ret == 1 ? 0 : -EIO;
253 }
254 
255 void amdgpu_ras_eeprom_fini(struct amdgpu_ras_eeprom_control *control)
256 {
257 	struct amdgpu_device *adev = to_amdgpu_device(control);
258 
259 	switch (adev->asic_type) {
260 	case CHIP_VEGA20:
261 		smu_v11_0_i2c_eeprom_control_fini(&control->eeprom_accessor);
262 		break;
263 
264 	default:
265 		return;
266 	}
267 }
268 
269 static void __encode_table_record_to_buff(struct amdgpu_ras_eeprom_control *control,
270 					  struct eeprom_table_record *record,
271 					  unsigned char *buff)
272 {
273 	__le64 tmp = 0;
274 	int i = 0;
275 
276 	/* Next are all record fields according to EEPROM page spec in LE foramt */
277 	buff[i++] = record->err_type;
278 
279 	buff[i++] = record->bank;
280 
281 	tmp = cpu_to_le64(record->ts);
282 	memcpy(buff + i, &tmp, 8);
283 	i += 8;
284 
285 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
286 	memcpy(buff + i, &tmp, 6);
287 	i += 6;
288 
289 	buff[i++] = record->mem_channel;
290 	buff[i++] = record->mcumc_id;
291 
292 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
293 	memcpy(buff + i, &tmp, 6);
294 }
295 
296 static void __decode_table_record_from_buff(struct amdgpu_ras_eeprom_control *control,
297 					    struct eeprom_table_record *record,
298 					    unsigned char *buff)
299 {
300 	__le64 tmp = 0;
301 	int i =  0;
302 
303 	/* Next are all record fields according to EEPROM page spec in LE foramt */
304 	record->err_type = buff[i++];
305 
306 	record->bank = buff[i++];
307 
308 	memcpy(&tmp, buff + i, 8);
309 	record->ts = le64_to_cpu(tmp);
310 	i += 8;
311 
312 	memcpy(&tmp, buff + i, 6);
313 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
314 	i += 6;
315 
316 	record->mem_channel = buff[i++];
317 	record->mcumc_id = buff[i++];
318 
319 	memcpy(&tmp, buff + i,  6);
320 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
321 }
322 
323 /*
324  * When reaching end of EEPROM memory jump back to 0 record address
325  * When next record access will go beyond EEPROM page boundary modify bits A17/A8
326  * in I2C selector to go to next page
327  */
328 static uint32_t __correct_eeprom_dest_address(uint32_t curr_address)
329 {
330 	uint32_t next_address = curr_address + EEPROM_TABLE_RECORD_SIZE;
331 
332 	/* When all EEPROM memory used jump back to 0 address */
333 	if (next_address > EEPROM_SIZE_BYTES) {
334 		DRM_INFO("Reached end of EEPROM memory, jumping to 0 "
335 			 "and overriding old record");
336 		return EEPROM_RECORD_START;
337 	}
338 
339 	/*
340 	 * To check if we overflow page boundary  compare next address with
341 	 * current and see if bits 17/8 of the EEPROM address will change
342 	 * If they do start from the next 256b page
343 	 *
344 	 * https://www.st.com/resource/en/datasheet/m24m02-dr.pdf sec. 5.1.2
345 	 */
346 	if ((curr_address & EEPROM_ADDR_MSB_MASK) != (next_address & EEPROM_ADDR_MSB_MASK)) {
347 		DRM_DEBUG_DRIVER("Reached end of EEPROM memory page, jumping to next: %lx",
348 				(next_address & EEPROM_ADDR_MSB_MASK));
349 
350 		return  (next_address & EEPROM_ADDR_MSB_MASK);
351 	}
352 
353 	return curr_address;
354 }
355 
356 int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control *control,
357 					    struct eeprom_table_record *records,
358 					    bool write,
359 					    int num)
360 {
361 	int i, ret = 0;
362 	struct i2c_msg *msgs;
363 	unsigned char *buffs;
364 	struct amdgpu_device *adev = to_amdgpu_device(control);
365 
366 	if (adev->asic_type != CHIP_VEGA20)
367 		return 0;
368 
369 	buffs = kcalloc(num, EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE,
370 			 GFP_KERNEL);
371 	if (!buffs)
372 		return -ENOMEM;
373 
374 	mutex_lock(&control->tbl_mutex);
375 
376 	msgs = kcalloc(num, sizeof(*msgs), GFP_KERNEL);
377 	if (!msgs) {
378 		ret = -ENOMEM;
379 		goto free_buff;
380 	}
381 
382 	/* In case of overflow just start from beginning to not lose newest records */
383 	if (write && (control->next_addr + EEPROM_TABLE_RECORD_SIZE * num > EEPROM_SIZE_BYTES))
384 		control->next_addr = EEPROM_RECORD_START;
385 
386 
387 	/*
388 	 * TODO Currently makes EEPROM writes for each record, this creates
389 	 * internal fragmentation. Optimized the code to do full page write of
390 	 * 256b
391 	 */
392 	for (i = 0; i < num; i++) {
393 		unsigned char *buff = &buffs[i * (EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
394 		struct eeprom_table_record *record = &records[i];
395 		struct i2c_msg *msg = &msgs[i];
396 
397 		control->next_addr = __correct_eeprom_dest_address(control->next_addr);
398 
399 		/*
400 		 * Update bits 16,17 of EEPROM address in I2C address by setting them
401 		 * to bits 1,2 of Device address byte
402 		 */
403 		msg->addr = EEPROM_I2C_TARGET_ADDR |
404 			       ((control->next_addr & EEPROM_ADDR_MSB_MASK) >> 15);
405 		msg->flags	= write ? 0 : I2C_M_RD;
406 		msg->len	= EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE;
407 		msg->buf	= buff;
408 
409 		/* Insert the EEPROM dest addess, bits 0-15 */
410 		buff[0] = ((control->next_addr >> 8) & 0xff);
411 		buff[1] = (control->next_addr & 0xff);
412 
413 		/* EEPROM table content is stored in LE format */
414 		if (write)
415 			__encode_table_record_to_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
416 
417 		/*
418 		 * The destination EEPROM address might need to be corrected to account
419 		 * for page or entire memory wrapping
420 		 */
421 		control->next_addr += EEPROM_TABLE_RECORD_SIZE;
422 	}
423 
424 	ret = i2c_transfer(&control->eeprom_accessor, msgs, num);
425 	if (ret < 1) {
426 		DRM_ERROR("Failed to process EEPROM table records, ret:%d", ret);
427 
428 		/* TODO Restore prev next EEPROM address ? */
429 		goto free_msgs;
430 	}
431 
432 
433 	if (!write) {
434 		for (i = 0; i < num; i++) {
435 			unsigned char *buff = &buffs[i*(EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
436 			struct eeprom_table_record *record = &records[i];
437 
438 			__decode_table_record_from_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
439 		}
440 	}
441 
442 	if (write) {
443 		uint32_t old_hdr_byte_sum = __calc_hdr_byte_sum(control);
444 
445 		/*
446 		 * Update table header with size and CRC and account for table
447 		 * wrap around where the assumption is that we treat it as empty
448 		 * table
449 		 *
450 		 * TODO - Check the assumption is correct
451 		 */
452 		control->num_recs += num;
453 		control->num_recs %= EEPROM_MAX_RECORD_NUM;
454 		control->tbl_hdr.tbl_size += EEPROM_TABLE_RECORD_SIZE * num;
455 		if (control->tbl_hdr.tbl_size > EEPROM_SIZE_BYTES)
456 			control->tbl_hdr.tbl_size = EEPROM_TABLE_HEADER_SIZE +
457 			control->num_recs * EEPROM_TABLE_RECORD_SIZE;
458 
459 		__update_tbl_checksum(control, records, num, old_hdr_byte_sum);
460 
461 		__update_table_header(control, buffs);
462 	} else if (!__validate_tbl_checksum(control, records, num)) {
463 		DRM_WARN("EEPROM Table checksum mismatch!");
464 		/* TODO Uncomment when EEPROM read/write is relliable */
465 		/* ret = -EIO; */
466 	}
467 
468 free_msgs:
469 	kfree(msgs);
470 
471 free_buff:
472 	kfree(buffs);
473 
474 	mutex_unlock(&control->tbl_mutex);
475 
476 	return ret == num ? 0 : -EIO;
477 }
478 
479 /* Used for testing if bugs encountered */
480 #if 0
481 void amdgpu_ras_eeprom_test(struct amdgpu_ras_eeprom_control *control)
482 {
483 	int i;
484 	struct eeprom_table_record *recs = kcalloc(1, sizeof(*recs), GFP_KERNEL);
485 
486 	if (!recs)
487 		return;
488 
489 	for (i = 0; i < 1 ; i++) {
490 		recs[i].address = 0xdeadbeef;
491 		recs[i].retired_page = i;
492 	}
493 
494 	if (!amdgpu_ras_eeprom_process_recods(control, recs, true, 1)) {
495 
496 		memset(recs, 0, sizeof(*recs) * 1);
497 
498 		control->next_addr = EEPROM_RECORD_START;
499 
500 		if (!amdgpu_ras_eeprom_process_recods(control, recs, false, 1)) {
501 			for (i = 0; i < 1; i++)
502 				DRM_INFO("rec.address :0x%llx, rec.retired_page :%llu",
503 					 recs[i].address, recs[i].retired_page);
504 		} else
505 			DRM_ERROR("Failed in reading from table");
506 
507 	} else
508 		DRM_ERROR("Failed in writing to table");
509 }
510 #endif
511