2 * Copyright 2019 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
24 #include "amdgpu_ras_eeprom.h"
26 #include "amdgpu_ras.h"
27 #include <linux/bits.h>
29 #include "amdgpu_atomfirmware.h"
31 #define EEPROM_I2C_TARGET_ADDR_VEGA20 0xA0
32 #define EEPROM_I2C_TARGET_ADDR_ARCTURUS 0xA8
33 #define EEPROM_I2C_TARGET_ADDR_ARCTURUS_D342 0xA0
34 #define EEPROM_I2C_TARGET_ADDR_SIENNA_CICHLID 0xA0
35 #define EEPROM_I2C_TARGET_ADDR_ALDEBARAN 0xA0
38 * The 2 macros bellow represent the actual size in bytes that
39 * those entities occupy in the EEPROM memory.
40 * EEPROM_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
41 * uses uint64 to store 6b fields such as retired_page.
43 #define EEPROM_TABLE_HEADER_SIZE 20
44 #define EEPROM_TABLE_RECORD_SIZE 24
46 #define EEPROM_ADDRESS_SIZE 0x2
48 /* Table hdr is 'AMDR' */
49 #define EEPROM_TABLE_HDR_VAL 0x414d4452
50 #define EEPROM_TABLE_VER 0x00010000
52 /* Bad GPU tag ‘BADG’ */
53 #define EEPROM_TABLE_HDR_BAD 0x42414447
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)
63 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_ras, eeprom_control))->adev
65 static bool __is_ras_eeprom_supported(struct amdgpu_device *adev)
67 if ((adev->asic_type == CHIP_VEGA20) ||
68 (adev->asic_type == CHIP_ARCTURUS) ||
69 (adev->asic_type == CHIP_SIENNA_CICHLID) ||
70 (adev->asic_type == CHIP_ALDEBARAN))
76 static bool __get_eeprom_i2c_addr_arct(struct amdgpu_device *adev,
79 struct atom_context *atom_ctx = adev->mode_info.atom_context;
81 if (!i2c_addr || !atom_ctx)
84 if (strnstr(atom_ctx->vbios_version,
86 sizeof(atom_ctx->vbios_version)))
87 *i2c_addr = EEPROM_I2C_TARGET_ADDR_ARCTURUS_D342;
89 *i2c_addr = EEPROM_I2C_TARGET_ADDR_ARCTURUS;
94 static bool __get_eeprom_i2c_addr(struct amdgpu_device *adev,
100 if (amdgpu_atomfirmware_ras_rom_addr(adev, (uint8_t*)i2c_addr))
103 switch (adev->asic_type) {
105 *i2c_addr = EEPROM_I2C_TARGET_ADDR_VEGA20;
109 return __get_eeprom_i2c_addr_arct(adev, i2c_addr);
111 case CHIP_SIENNA_CICHLID:
112 *i2c_addr = EEPROM_I2C_TARGET_ADDR_SIENNA_CICHLID;
116 *i2c_addr = EEPROM_I2C_TARGET_ADDR_ALDEBARAN;
126 static void __encode_table_header_to_buff(struct amdgpu_ras_eeprom_table_header *hdr,
129 uint32_t *pp = (uint32_t *) buff;
131 pp[0] = cpu_to_le32(hdr->header);
132 pp[1] = cpu_to_le32(hdr->version);
133 pp[2] = cpu_to_le32(hdr->first_rec_offset);
134 pp[3] = cpu_to_le32(hdr->tbl_size);
135 pp[4] = cpu_to_le32(hdr->checksum);
138 static void __decode_table_header_from_buff(struct amdgpu_ras_eeprom_table_header *hdr,
141 uint32_t *pp = (uint32_t *)buff;
143 hdr->header = le32_to_cpu(pp[0]);
144 hdr->version = le32_to_cpu(pp[1]);
145 hdr->first_rec_offset = le32_to_cpu(pp[2]);
146 hdr->tbl_size = le32_to_cpu(pp[3]);
147 hdr->checksum = le32_to_cpu(pp[4]);
150 static int __update_table_header(struct amdgpu_ras_eeprom_control *control,
154 struct amdgpu_device *adev = to_amdgpu_device(control);
155 struct i2c_msg msg = {
158 .len = EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
163 *(uint16_t *)buff = EEPROM_HDR_START;
164 __encode_table_header_to_buff(&control->tbl_hdr, buff + EEPROM_ADDRESS_SIZE);
166 msg.addr = control->i2c_address;
168 /* i2c may be unstable in gpu reset */
169 down_read(&adev->reset_sem);
170 ret = i2c_transfer(&adev->pm.smu_i2c, &msg, 1);
171 up_read(&adev->reset_sem);
174 DRM_ERROR("Failed to write EEPROM table header, ret:%d", ret);
179 static uint32_t __calc_hdr_byte_sum(struct amdgpu_ras_eeprom_control *control)
182 uint32_t tbl_sum = 0;
184 /* Header checksum, skip checksum field in the calculation */
185 for (i = 0; i < sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum); i++)
186 tbl_sum += *(((unsigned char *)&control->tbl_hdr) + i);
191 static uint32_t __calc_recs_byte_sum(struct eeprom_table_record *records,
195 uint32_t tbl_sum = 0;
197 /* Records checksum */
198 for (i = 0; i < num; i++) {
199 struct eeprom_table_record *record = &records[i];
201 for (j = 0; j < sizeof(*record); j++) {
202 tbl_sum += *(((unsigned char *)record) + j);
209 static inline uint32_t __calc_tbl_byte_sum(struct amdgpu_ras_eeprom_control *control,
210 struct eeprom_table_record *records, int num)
212 return __calc_hdr_byte_sum(control) + __calc_recs_byte_sum(records, num);
215 /* Checksum = 256 -((sum of all table entries) mod 256) */
216 static void __update_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
217 struct eeprom_table_record *records, int num,
218 uint32_t old_hdr_byte_sum)
221 * This will update the table sum with new records.
223 * TODO: What happens when the EEPROM table is to be wrapped around
224 * and old records from start will get overridden.
227 /* need to recalculate updated header byte sum */
228 control->tbl_byte_sum -= old_hdr_byte_sum;
229 control->tbl_byte_sum += __calc_tbl_byte_sum(control, records, num);
231 control->tbl_hdr.checksum = 256 - (control->tbl_byte_sum % 256);
234 /* table sum mod 256 + checksum must equals 256 */
235 static bool __validate_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
236 struct eeprom_table_record *records, int num)
238 control->tbl_byte_sum = __calc_tbl_byte_sum(control, records, num);
240 if (control->tbl_hdr.checksum + (control->tbl_byte_sum % 256) != 256) {
241 DRM_WARN("Checksum mismatch, checksum: %u ", control->tbl_hdr.checksum);
248 static int amdgpu_ras_eeprom_correct_header_tag(
249 struct amdgpu_ras_eeprom_control *control,
252 unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE];
253 struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
256 memset(buff, 0, EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE);
258 mutex_lock(&control->tbl_mutex);
259 hdr->header = header;
260 ret = __update_table_header(control, buff);
261 mutex_unlock(&control->tbl_mutex);
266 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control *control)
268 unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE] = { 0 };
269 struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
272 mutex_lock(&control->tbl_mutex);
274 hdr->header = EEPROM_TABLE_HDR_VAL;
275 hdr->version = EEPROM_TABLE_VER;
276 hdr->first_rec_offset = EEPROM_RECORD_START;
277 hdr->tbl_size = EEPROM_TABLE_HEADER_SIZE;
279 control->tbl_byte_sum = 0;
280 __update_tbl_checksum(control, NULL, 0, 0);
281 control->next_addr = EEPROM_RECORD_START;
283 ret = __update_table_header(control, buff);
285 mutex_unlock(&control->tbl_mutex);
291 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control,
292 bool *exceed_err_limit)
295 struct amdgpu_device *adev = to_amdgpu_device(control);
296 unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE] = { 0 };
297 struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
298 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
299 struct i2c_msg msg = {
302 .len = EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
306 *exceed_err_limit = false;
308 if (!__is_ras_eeprom_supported(adev))
311 /* Verify i2c adapter is initialized */
312 if (!adev->pm.smu_i2c.algo)
315 if (!__get_eeprom_i2c_addr(adev, &control->i2c_address))
318 mutex_init(&control->tbl_mutex);
320 msg.addr = control->i2c_address;
321 /* Read/Create table header from EEPROM address 0 */
322 ret = i2c_transfer(&adev->pm.smu_i2c, &msg, 1);
324 DRM_ERROR("Failed to read EEPROM table header, ret:%d", ret);
328 __decode_table_header_from_buff(hdr, &buff[2]);
330 if (hdr->header == EEPROM_TABLE_HDR_VAL) {
331 control->num_recs = (hdr->tbl_size - EEPROM_TABLE_HEADER_SIZE) /
332 EEPROM_TABLE_RECORD_SIZE;
333 control->tbl_byte_sum = __calc_hdr_byte_sum(control);
334 control->next_addr = EEPROM_RECORD_START;
336 DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
339 } else if ((hdr->header == EEPROM_TABLE_HDR_BAD) &&
340 (amdgpu_bad_page_threshold != 0)) {
341 if (ras->bad_page_cnt_threshold > control->num_recs) {
342 dev_info(adev->dev, "Using one valid bigger bad page "
343 "threshold and correcting eeprom header tag.\n");
344 ret = amdgpu_ras_eeprom_correct_header_tag(control,
345 EEPROM_TABLE_HDR_VAL);
347 *exceed_err_limit = true;
348 dev_err(adev->dev, "Exceeding the bad_page_threshold parameter, "
349 "disabling the GPU.\n");
352 DRM_INFO("Creating new EEPROM table");
354 ret = amdgpu_ras_eeprom_reset_table(control);
357 return ret == 1 ? 0 : -EIO;
360 static void __encode_table_record_to_buff(struct amdgpu_ras_eeprom_control *control,
361 struct eeprom_table_record *record,
367 /* Next are all record fields according to EEPROM page spec in LE foramt */
368 buff[i++] = record->err_type;
370 buff[i++] = record->bank;
372 tmp = cpu_to_le64(record->ts);
373 memcpy(buff + i, &tmp, 8);
376 tmp = cpu_to_le64((record->offset & 0xffffffffffff));
377 memcpy(buff + i, &tmp, 6);
380 buff[i++] = record->mem_channel;
381 buff[i++] = record->mcumc_id;
383 tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
384 memcpy(buff + i, &tmp, 6);
387 static void __decode_table_record_from_buff(struct amdgpu_ras_eeprom_control *control,
388 struct eeprom_table_record *record,
394 /* Next are all record fields according to EEPROM page spec in LE foramt */
395 record->err_type = buff[i++];
397 record->bank = buff[i++];
399 memcpy(&tmp, buff + i, 8);
400 record->ts = le64_to_cpu(tmp);
403 memcpy(&tmp, buff + i, 6);
404 record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
407 record->mem_channel = buff[i++];
408 record->mcumc_id = buff[i++];
410 memcpy(&tmp, buff + i, 6);
411 record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
415 * When reaching end of EEPROM memory jump back to 0 record address
416 * When next record access will go beyond EEPROM page boundary modify bits A17/A8
417 * in I2C selector to go to next page
419 static uint32_t __correct_eeprom_dest_address(uint32_t curr_address)
421 uint32_t next_address = curr_address + EEPROM_TABLE_RECORD_SIZE;
423 /* When all EEPROM memory used jump back to 0 address */
424 if (next_address > EEPROM_SIZE_BYTES) {
425 DRM_INFO("Reached end of EEPROM memory, jumping to 0 "
426 "and overriding old record");
427 return EEPROM_RECORD_START;
431 * To check if we overflow page boundary compare next address with
432 * current and see if bits 17/8 of the EEPROM address will change
433 * If they do start from the next 256b page
435 * https://www.st.com/resource/en/datasheet/m24m02-dr.pdf sec. 5.1.2
437 if ((curr_address & EEPROM_ADDR_MSB_MASK) != (next_address & EEPROM_ADDR_MSB_MASK)) {
438 DRM_DEBUG_DRIVER("Reached end of EEPROM memory page, jumping to next: %lx",
439 (next_address & EEPROM_ADDR_MSB_MASK));
441 return (next_address & EEPROM_ADDR_MSB_MASK);
447 bool amdgpu_ras_eeprom_check_err_threshold(struct amdgpu_device *adev)
449 struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
451 if (!__is_ras_eeprom_supported(adev))
454 /* skip check eeprom table for VEGA20 Gaming */
458 if (!(con->features & BIT(AMDGPU_RAS_BLOCK__UMC)))
461 if (con->eeprom_control.tbl_hdr.header == EEPROM_TABLE_HDR_BAD) {
462 dev_warn(adev->dev, "This GPU is in BAD status.");
463 dev_warn(adev->dev, "Please retire it or setting one bigger "
464 "threshold value when reloading driver.\n");
471 int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control *control,
472 struct eeprom_table_record *records,
477 struct i2c_msg *msgs, *msg;
478 unsigned char *buffs, *buff;
479 struct eeprom_table_record *record;
480 struct amdgpu_device *adev = to_amdgpu_device(control);
481 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
483 if (!__is_ras_eeprom_supported(adev))
486 buffs = kcalloc(num, EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE,
491 mutex_lock(&control->tbl_mutex);
493 msgs = kcalloc(num, sizeof(*msgs), GFP_KERNEL);
500 * If saved bad pages number exceeds the bad page threshold for
501 * the whole VRAM, update table header to mark the BAD GPU tag
502 * and schedule one ras recovery after eeprom write is done,
503 * this can avoid the missing for latest records.
505 * This new header will be picked up and checked in the bootup
506 * by ras recovery, which may break bootup process to notify
507 * user this GPU is in bad state and to retire such GPU for
510 if (write && (amdgpu_bad_page_threshold != 0) &&
511 ((control->num_recs + num) >= ras->bad_page_cnt_threshold)) {
513 "Saved bad pages(%d) reaches threshold value(%d).\n",
514 control->num_recs + num, ras->bad_page_cnt_threshold);
515 control->tbl_hdr.header = EEPROM_TABLE_HDR_BAD;
518 /* In case of overflow just start from beginning to not lose newest records */
519 if (write && (control->next_addr + EEPROM_TABLE_RECORD_SIZE * num > EEPROM_SIZE_BYTES))
520 control->next_addr = EEPROM_RECORD_START;
523 * TODO Currently makes EEPROM writes for each record, this creates
524 * internal fragmentation. Optimized the code to do full page write of
527 for (i = 0; i < num; i++) {
528 buff = &buffs[i * (EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
529 record = &records[i];
532 control->next_addr = __correct_eeprom_dest_address(control->next_addr);
535 * Update bits 16,17 of EEPROM address in I2C address by setting them
536 * to bits 1,2 of Device address byte
538 msg->addr = control->i2c_address |
539 ((control->next_addr & EEPROM_ADDR_MSB_MASK) >> 15);
540 msg->flags = write ? 0 : I2C_M_RD;
541 msg->len = EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE;
544 /* Insert the EEPROM dest addess, bits 0-15 */
545 buff[0] = ((control->next_addr >> 8) & 0xff);
546 buff[1] = (control->next_addr & 0xff);
548 /* EEPROM table content is stored in LE format */
550 __encode_table_record_to_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
553 * The destination EEPROM address might need to be corrected to account
554 * for page or entire memory wrapping
556 control->next_addr += EEPROM_TABLE_RECORD_SIZE;
559 /* i2c may be unstable in gpu reset */
560 down_read(&adev->reset_sem);
561 ret = i2c_transfer(&adev->pm.smu_i2c, msgs, num);
562 up_read(&adev->reset_sem);
565 DRM_ERROR("Failed to process EEPROM table records, ret:%d", ret);
567 /* TODO Restore prev next EEPROM address ? */
573 for (i = 0; i < num; i++) {
574 buff = &buffs[i*(EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
575 record = &records[i];
577 __decode_table_record_from_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
582 uint32_t old_hdr_byte_sum = __calc_hdr_byte_sum(control);
585 * Update table header with size and CRC and account for table
586 * wrap around where the assumption is that we treat it as empty
589 * TODO - Check the assumption is correct
591 control->num_recs += num;
592 control->num_recs %= EEPROM_MAX_RECORD_NUM;
593 control->tbl_hdr.tbl_size += EEPROM_TABLE_RECORD_SIZE * num;
594 if (control->tbl_hdr.tbl_size > EEPROM_SIZE_BYTES)
595 control->tbl_hdr.tbl_size = EEPROM_TABLE_HEADER_SIZE +
596 control->num_recs * EEPROM_TABLE_RECORD_SIZE;
598 __update_tbl_checksum(control, records, num, old_hdr_byte_sum);
600 __update_table_header(control, buffs);
601 } else if (!__validate_tbl_checksum(control, records, num)) {
602 DRM_WARN("EEPROM Table checksum mismatch!");
603 /* TODO Uncomment when EEPROM read/write is relliable */
613 mutex_unlock(&control->tbl_mutex);
615 return ret == num ? 0 : -EIO;
618 inline uint32_t amdgpu_ras_eeprom_get_record_max_length(void)
620 return EEPROM_MAX_RECORD_NUM;
623 /* Used for testing if bugs encountered */
625 void amdgpu_ras_eeprom_test(struct amdgpu_ras_eeprom_control *control)
628 struct eeprom_table_record *recs = kcalloc(1, sizeof(*recs), GFP_KERNEL);
633 for (i = 0; i < 1 ; i++) {
634 recs[i].address = 0xdeadbeef;
635 recs[i].retired_page = i;
638 if (!amdgpu_ras_eeprom_process_recods(control, recs, true, 1)) {
640 memset(recs, 0, sizeof(*recs) * 1);
642 control->next_addr = EEPROM_RECORD_START;
644 if (!amdgpu_ras_eeprom_process_recods(control, recs, false, 1)) {
645 for (i = 0; i < 1; i++)
646 DRM_INFO("rec.address :0x%llx, rec.retired_page :%llu",
647 recs[i].address, recs[i].retired_page);
649 DRM_ERROR("Failed in reading from table");
652 DRM_ERROR("Failed in writing to table");