xref: /netbsd-src/sys/external/bsd/drm2/dist/drm/amd/amdgpu/amdgpu_ras_eeprom.c (revision 07eb61ce39682b60a0b29b81ad8b6af16527a557)
1 /*	$NetBSD: amdgpu_ras_eeprom.c,v 1.4 2021/12/19 12:31:45 riastradh Exp $	*/
2 
3 /*
4  * Copyright 2019 Advanced Micro Devices, Inc.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  */
25 
26 #include <sys/cdefs.h>
27 __KERNEL_RCSID(0, "$NetBSD: amdgpu_ras_eeprom.c,v 1.4 2021/12/19 12:31:45 riastradh Exp $");
28 
29 #include "amdgpu_ras_eeprom.h"
30 #include "amdgpu.h"
31 #include "amdgpu_ras.h"
32 #include <linux/bits.h>
33 #include "smu_v11_0_i2c.h"
34 
35 #include <linux/nbsd-namespace.h>
36 
37 #define EEPROM_I2C_TARGET_ADDR_ARCTURUS  0xA8
38 #define EEPROM_I2C_TARGET_ADDR_VEGA20    0xA0
39 
40 /*
41  * The 2 macros bellow represent the actual size in bytes that
42  * those entities occupy in the EEPROM memory.
43  * EEPROM_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
44  * uses uint64 to store 6b fields such as retired_page.
45  */
46 #define EEPROM_TABLE_HEADER_SIZE 20
47 #define EEPROM_TABLE_RECORD_SIZE 24
48 
49 #define EEPROM_ADDRESS_SIZE 0x2
50 
51 /* Table hdr is 'AMDR' */
52 #define EEPROM_TABLE_HDR_VAL 0x414d4452
53 #define EEPROM_TABLE_VER 0x00010000
54 
55 /* Assume 2 Mbit size */
56 #define EEPROM_SIZE_BYTES 256000
57 #define EEPROM_PAGE__SIZE_BYTES 256
58 #define EEPROM_HDR_START 0
59 #define EEPROM_RECORD_START (EEPROM_HDR_START + EEPROM_TABLE_HEADER_SIZE)
60 #define EEPROM_MAX_RECORD_NUM ((EEPROM_SIZE_BYTES - EEPROM_TABLE_HEADER_SIZE) / EEPROM_TABLE_RECORD_SIZE)
61 #define EEPROM_ADDR_MSB_MASK GENMASK(17, 8)
62 
63 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_ras, eeprom_control))->adev
64 
__encode_table_header_to_buff(struct amdgpu_ras_eeprom_table_header * hdr,unsigned char * buff)65 static void __encode_table_header_to_buff(struct amdgpu_ras_eeprom_table_header *hdr,
66 					  unsigned char *buff)
67 {
68 	uint32_t *pp = (uint32_t *) buff;
69 
70 	pp[0] = cpu_to_le32(hdr->header);
71 	pp[1] = cpu_to_le32(hdr->version);
72 	pp[2] = cpu_to_le32(hdr->first_rec_offset);
73 	pp[3] = cpu_to_le32(hdr->tbl_size);
74 	pp[4] = cpu_to_le32(hdr->checksum);
75 }
76 
__decode_table_header_from_buff(struct amdgpu_ras_eeprom_table_header * hdr,unsigned char * buff)77 static void __decode_table_header_from_buff(struct amdgpu_ras_eeprom_table_header *hdr,
78 					  unsigned char *buff)
79 {
80 	uint32_t *pp = (uint32_t *)buff;
81 
82 	hdr->header 	      = le32_to_cpu(pp[0]);
83 	hdr->version 	      = le32_to_cpu(pp[1]);
84 	hdr->first_rec_offset = le32_to_cpu(pp[2]);
85 	hdr->tbl_size 	      = le32_to_cpu(pp[3]);
86 	hdr->checksum 	      = le32_to_cpu(pp[4]);
87 }
88 
__update_table_header(struct amdgpu_ras_eeprom_control * control,unsigned char * buff)89 static int __update_table_header(struct amdgpu_ras_eeprom_control *control,
90 				 unsigned char *buff)
91 {
92 	int ret = 0;
93 	struct i2c_msg msg = {
94 			.addr	= 0,
95 			.flags	= 0,
96 			.len	= EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
97 			.buf	= buff,
98 	};
99 
100 
101 	*(uint16_t *)buff = EEPROM_HDR_START;
102 	__encode_table_header_to_buff(&control->tbl_hdr, buff + EEPROM_ADDRESS_SIZE);
103 
104 	msg.addr = control->i2c_address;
105 
106 	ret = i2c_transfer(&control->eeprom_accessor, &msg, 1);
107 	if (ret < 1)
108 		DRM_ERROR("Failed to write EEPROM table header, ret:%d", ret);
109 
110 	return ret;
111 }
112 
113 
114 
__calc_hdr_byte_sum(struct amdgpu_ras_eeprom_control * control)115 static uint32_t  __calc_hdr_byte_sum(struct amdgpu_ras_eeprom_control *control)
116 {
117 	int i;
118 	uint32_t tbl_sum = 0;
119 
120 	/* Header checksum, skip checksum field in the calculation */
121 	for (i = 0; i < sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum); i++)
122 		tbl_sum += *(((unsigned char *)&control->tbl_hdr) + i);
123 
124 	return tbl_sum;
125 }
126 
__calc_recs_byte_sum(struct eeprom_table_record * records,int num)127 static uint32_t  __calc_recs_byte_sum(struct eeprom_table_record *records,
128 				      int num)
129 {
130 	int i, j;
131 	uint32_t tbl_sum = 0;
132 
133 	/* Records checksum */
134 	for (i = 0; i < num; i++) {
135 		struct eeprom_table_record *record = &records[i];
136 
137 		for (j = 0; j < sizeof(*record); j++) {
138 			tbl_sum += *(((unsigned char *)record) + j);
139 		}
140 	}
141 
142 	return tbl_sum;
143 }
144 
__calc_tbl_byte_sum(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * records,int num)145 static inline uint32_t  __calc_tbl_byte_sum(struct amdgpu_ras_eeprom_control *control,
146 				  struct eeprom_table_record *records, int num)
147 {
148 	return __calc_hdr_byte_sum(control) + __calc_recs_byte_sum(records, num);
149 }
150 
151 /* Checksum = 256 -((sum of all table entries) mod 256) */
__update_tbl_checksum(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * records,int num,uint32_t old_hdr_byte_sum)152 static void __update_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
153 				  struct eeprom_table_record *records, int num,
154 				  uint32_t old_hdr_byte_sum)
155 {
156 	/*
157 	 * This will update the table sum with new records.
158 	 *
159 	 * TODO: What happens when the EEPROM table is to be wrapped around
160 	 * and old records from start will get overridden.
161 	 */
162 
163 	/* need to recalculate updated header byte sum */
164 	control->tbl_byte_sum -= old_hdr_byte_sum;
165 	control->tbl_byte_sum += __calc_tbl_byte_sum(control, records, num);
166 
167 	control->tbl_hdr.checksum = 256 - (control->tbl_byte_sum % 256);
168 }
169 
170 /* table sum mod 256 + checksum must equals 256 */
__validate_tbl_checksum(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * records,int num)171 static bool __validate_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
172 			    struct eeprom_table_record *records, int num)
173 {
174 	control->tbl_byte_sum = __calc_tbl_byte_sum(control, records, num);
175 
176 	if (control->tbl_hdr.checksum + (control->tbl_byte_sum % 256) != 256) {
177 		DRM_WARN("Checksum mismatch, checksum: %u ", control->tbl_hdr.checksum);
178 		return false;
179 	}
180 
181 	return true;
182 }
183 
amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control * control)184 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
185 {
186 	unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE] = { 0 };
187 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
188 	int ret = 0;
189 
190 	mutex_lock(&control->tbl_mutex);
191 
192 	hdr->header = EEPROM_TABLE_HDR_VAL;
193 	hdr->version = EEPROM_TABLE_VER;
194 	hdr->first_rec_offset = EEPROM_RECORD_START;
195 	hdr->tbl_size = EEPROM_TABLE_HEADER_SIZE;
196 
197 	control->tbl_byte_sum = 0;
198 	__update_tbl_checksum(control, NULL, 0, 0);
199 	control->next_addr = EEPROM_RECORD_START;
200 
201 	ret = __update_table_header(control, buff);
202 
203 	mutex_unlock(&control->tbl_mutex);
204 
205 	return ret;
206 
207 }
208 
amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control * control)209 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control)
210 {
211 	int ret = 0;
212 	struct amdgpu_device *adev = to_amdgpu_device(control);
213 	unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE] = { 0 };
214 	struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
215 	struct i2c_msg msg = {
216 			.addr	= 0,
217 			.flags	= I2C_M_RD,
218 			.len	= EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
219 			.buf	= buff,
220 	};
221 
222 	mutex_init(&control->tbl_mutex);
223 
224 	switch (adev->asic_type) {
225 	case CHIP_VEGA20:
226 		control->i2c_address = EEPROM_I2C_TARGET_ADDR_VEGA20;
227 		ret = smu_v11_0_i2c_eeprom_control_init(&control->eeprom_accessor);
228 		break;
229 
230 	case CHIP_ARCTURUS:
231 		control->i2c_address = EEPROM_I2C_TARGET_ADDR_ARCTURUS;
232 		ret = smu_i2c_eeprom_init(&adev->smu, &control->eeprom_accessor);
233 		break;
234 
235 	default:
236 		return 0;
237 	}
238 
239 	if (ret) {
240 		DRM_ERROR("Failed to init I2C controller, ret:%d", ret);
241 		return ret;
242 	}
243 
244 	msg.addr = control->i2c_address;
245 
246 	/* Read/Create table header from EEPROM address 0 */
247 	ret = i2c_transfer(&control->eeprom_accessor, &msg, 1);
248 	if (ret < 1) {
249 		DRM_ERROR("Failed to read EEPROM table header, ret:%d", ret);
250 		return ret;
251 	}
252 
253 	__decode_table_header_from_buff(hdr, &buff[2]);
254 
255 	if (hdr->header == EEPROM_TABLE_HDR_VAL) {
256 		control->num_recs = (hdr->tbl_size - EEPROM_TABLE_HEADER_SIZE) /
257 				    EEPROM_TABLE_RECORD_SIZE;
258 		control->tbl_byte_sum = __calc_hdr_byte_sum(control);
259 		control->next_addr = EEPROM_RECORD_START;
260 
261 		DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
262 				 control->num_recs);
263 
264 	} else {
265 		DRM_INFO("Creating new EEPROM table");
266 
267 		ret = amdgpu_ras_eeprom_reset_table(control);
268 	}
269 
270 	return ret == 1 ? 0 : -EIO;
271 }
272 
amdgpu_ras_eeprom_fini(struct amdgpu_ras_eeprom_control * control)273 void amdgpu_ras_eeprom_fini(struct amdgpu_ras_eeprom_control *control)
274 {
275 	struct amdgpu_device *adev = to_amdgpu_device(control);
276 
277 	switch (adev->asic_type) {
278 	case CHIP_VEGA20:
279 		smu_v11_0_i2c_eeprom_control_fini(&control->eeprom_accessor);
280 		break;
281 	case CHIP_ARCTURUS:
282 		smu_i2c_eeprom_fini(&adev->smu, &control->eeprom_accessor);
283 		break;
284 
285 	default:
286 		return;
287 	}
288 
289 	mutex_destroy(&control->tbl_mutex);
290 }
291 
__encode_table_record_to_buff(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * record,unsigned char * buff)292 static void __encode_table_record_to_buff(struct amdgpu_ras_eeprom_control *control,
293 					  struct eeprom_table_record *record,
294 					  unsigned char *buff)
295 {
296 	__le64 tmp = 0;
297 	int i = 0;
298 
299 	/* Next are all record fields according to EEPROM page spec in LE foramt */
300 	buff[i++] = record->err_type;
301 
302 	buff[i++] = record->bank;
303 
304 	tmp = cpu_to_le64(record->ts);
305 	memcpy(buff + i, &tmp, 8);
306 	i += 8;
307 
308 	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
309 	memcpy(buff + i, &tmp, 6);
310 	i += 6;
311 
312 	buff[i++] = record->mem_channel;
313 	buff[i++] = record->mcumc_id;
314 
315 	tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
316 	memcpy(buff + i, &tmp, 6);
317 }
318 
__decode_table_record_from_buff(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * record,unsigned char * buff)319 static void __decode_table_record_from_buff(struct amdgpu_ras_eeprom_control *control,
320 					    struct eeprom_table_record *record,
321 					    unsigned char *buff)
322 {
323 	__le64 tmp = 0;
324 	int i =  0;
325 
326 	/* Next are all record fields according to EEPROM page spec in LE foramt */
327 	record->err_type = buff[i++];
328 
329 	record->bank = buff[i++];
330 
331 	memcpy(&tmp, buff + i, 8);
332 	record->ts = le64_to_cpu(tmp);
333 	i += 8;
334 
335 	memcpy(&tmp, buff + i, 6);
336 	record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
337 	i += 6;
338 
339 	record->mem_channel = buff[i++];
340 	record->mcumc_id = buff[i++];
341 
342 	memcpy(&tmp, buff + i,  6);
343 	record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
344 }
345 
346 /*
347  * When reaching end of EEPROM memory jump back to 0 record address
348  * When next record access will go beyond EEPROM page boundary modify bits A17/A8
349  * in I2C selector to go to next page
350  */
__correct_eeprom_dest_address(uint32_t curr_address)351 static uint32_t __correct_eeprom_dest_address(uint32_t curr_address)
352 {
353 	uint32_t next_address = curr_address + EEPROM_TABLE_RECORD_SIZE;
354 
355 	/* When all EEPROM memory used jump back to 0 address */
356 	if (next_address > EEPROM_SIZE_BYTES) {
357 		DRM_INFO("Reached end of EEPROM memory, jumping to 0 "
358 			 "and overriding old record");
359 		return EEPROM_RECORD_START;
360 	}
361 
362 	/*
363 	 * To check if we overflow page boundary  compare next address with
364 	 * current and see if bits 17/8 of the EEPROM address will change
365 	 * If they do start from the next 256b page
366 	 *
367 	 * https://www.st.com/resource/en/datasheet/m24m02-dr.pdf sec. 5.1.2
368 	 */
369 	if ((curr_address & EEPROM_ADDR_MSB_MASK) != (next_address & EEPROM_ADDR_MSB_MASK)) {
370 		DRM_DEBUG_DRIVER("Reached end of EEPROM memory page, jumping to next: %lx",
371 				(next_address & EEPROM_ADDR_MSB_MASK));
372 
373 		return  (next_address & EEPROM_ADDR_MSB_MASK);
374 	}
375 
376 	return curr_address;
377 }
378 
amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control * control,struct eeprom_table_record * records,bool write,int num)379 int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control *control,
380 					    struct eeprom_table_record *records,
381 					    bool write,
382 					    int num)
383 {
384 	int i, ret = 0;
385 	struct i2c_msg *msgs, *msg;
386 	unsigned char *buffs, *buff;
387 	struct eeprom_table_record *record;
388 	struct amdgpu_device *adev = to_amdgpu_device(control);
389 
390 	if (adev->asic_type != CHIP_VEGA20 && adev->asic_type != CHIP_ARCTURUS)
391 		return 0;
392 
393 	buffs = kcalloc(num, EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE,
394 			 GFP_KERNEL);
395 	if (!buffs)
396 		return -ENOMEM;
397 
398 	mutex_lock(&control->tbl_mutex);
399 
400 	msgs = kcalloc(num, sizeof(*msgs), GFP_KERNEL);
401 	if (!msgs) {
402 		ret = -ENOMEM;
403 		goto free_buff;
404 	}
405 
406 	/* In case of overflow just start from beginning to not lose newest records */
407 	if (write && (control->next_addr + EEPROM_TABLE_RECORD_SIZE * num > EEPROM_SIZE_BYTES))
408 		control->next_addr = EEPROM_RECORD_START;
409 
410 
411 	/*
412 	 * TODO Currently makes EEPROM writes for each record, this creates
413 	 * internal fragmentation. Optimized the code to do full page write of
414 	 * 256b
415 	 */
416 	for (i = 0; i < num; i++) {
417 		buff = &buffs[i * (EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
418 		record = &records[i];
419 		msg = &msgs[i];
420 
421 		control->next_addr = __correct_eeprom_dest_address(control->next_addr);
422 
423 		/*
424 		 * Update bits 16,17 of EEPROM address in I2C address by setting them
425 		 * to bits 1,2 of Device address byte
426 		 */
427 		msg->addr = control->i2c_address |
428 			        ((control->next_addr & EEPROM_ADDR_MSB_MASK) >> 15);
429 		msg->flags	= write ? 0 : I2C_M_RD;
430 		msg->len	= EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE;
431 		msg->buf	= buff;
432 
433 		/* Insert the EEPROM dest addess, bits 0-15 */
434 		buff[0] = ((control->next_addr >> 8) & 0xff);
435 		buff[1] = (control->next_addr & 0xff);
436 
437 		/* EEPROM table content is stored in LE format */
438 		if (write)
439 			__encode_table_record_to_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
440 
441 		/*
442 		 * The destination EEPROM address might need to be corrected to account
443 		 * for page or entire memory wrapping
444 		 */
445 		control->next_addr += EEPROM_TABLE_RECORD_SIZE;
446 	}
447 
448 	ret = i2c_transfer(&control->eeprom_accessor, msgs, num);
449 	if (ret < 1) {
450 		DRM_ERROR("Failed to process EEPROM table records, ret:%d", ret);
451 
452 		/* TODO Restore prev next EEPROM address ? */
453 		goto free_msgs;
454 	}
455 
456 
457 	if (!write) {
458 		for (i = 0; i < num; i++) {
459 			buff = &buffs[i*(EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
460 			record = &records[i];
461 
462 			__decode_table_record_from_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
463 		}
464 	}
465 
466 	if (write) {
467 		uint32_t old_hdr_byte_sum = __calc_hdr_byte_sum(control);
468 
469 		/*
470 		 * Update table header with size and CRC and account for table
471 		 * wrap around where the assumption is that we treat it as empty
472 		 * table
473 		 *
474 		 * TODO - Check the assumption is correct
475 		 */
476 		control->num_recs += num;
477 		control->num_recs %= EEPROM_MAX_RECORD_NUM;
478 		control->tbl_hdr.tbl_size += EEPROM_TABLE_RECORD_SIZE * num;
479 		if (control->tbl_hdr.tbl_size > EEPROM_SIZE_BYTES)
480 			control->tbl_hdr.tbl_size = EEPROM_TABLE_HEADER_SIZE +
481 			control->num_recs * EEPROM_TABLE_RECORD_SIZE;
482 
483 		__update_tbl_checksum(control, records, num, old_hdr_byte_sum);
484 
485 		__update_table_header(control, buffs);
486 	} else if (!__validate_tbl_checksum(control, records, num)) {
487 		DRM_WARN("EEPROM Table checksum mismatch!");
488 		/* TODO Uncomment when EEPROM read/write is relliable */
489 		/* ret = -EIO; */
490 	}
491 
492 free_msgs:
493 	kfree(msgs);
494 
495 free_buff:
496 	kfree(buffs);
497 
498 	mutex_unlock(&control->tbl_mutex);
499 
500 	return ret == num ? 0 : -EIO;
501 }
502 
503 /* Used for testing if bugs encountered */
504 #if 0
505 void amdgpu_ras_eeprom_test(struct amdgpu_ras_eeprom_control *control)
506 {
507 	int i;
508 	struct eeprom_table_record *recs = kcalloc(1, sizeof(*recs), GFP_KERNEL);
509 
510 	if (!recs)
511 		return;
512 
513 	for (i = 0; i < 1 ; i++) {
514 		recs[i].address = 0xdeadbeef;
515 		recs[i].retired_page = i;
516 	}
517 
518 	if (!amdgpu_ras_eeprom_process_recods(control, recs, true, 1)) {
519 
520 		memset(recs, 0, sizeof(*recs) * 1);
521 
522 		control->next_addr = EEPROM_RECORD_START;
523 
524 		if (!amdgpu_ras_eeprom_process_recods(control, recs, false, 1)) {
525 			for (i = 0; i < 1; i++)
526 				DRM_INFO("rec.address :0x%llx, rec.retired_page :%llu",
527 					 recs[i].address, recs[i].retired_page);
528 		} else
529 			DRM_ERROR("Failed in reading from table");
530 
531 	} else
532 		DRM_ERROR("Failed in writing to table");
533 }
534 #endif
535