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>
28 #include "smu_v11_0_i2c.h"
30 #define EEPROM_I2C_TARGET_ADDR_ARCTURUS 0xA8
31 #define EEPROM_I2C_TARGET_ADDR_VEGA20 0xA0
34 * The 2 macros bellow represent the actual size in bytes that
35 * those entities occupy in the EEPROM memory.
36 * EEPROM_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
37 * uses uint64 to store 6b fields such as retired_page.
39 #define EEPROM_TABLE_HEADER_SIZE 20
40 #define EEPROM_TABLE_RECORD_SIZE 24
42 #define EEPROM_ADDRESS_SIZE 0x2
44 /* Table hdr is 'AMDR' */
45 #define EEPROM_TABLE_HDR_VAL 0x414d4452
46 #define EEPROM_TABLE_VER 0x00010000
48 /* Assume 2 Mbit size */
49 #define EEPROM_SIZE_BYTES 256000
50 #define EEPROM_PAGE__SIZE_BYTES 256
51 #define EEPROM_HDR_START 0
52 #define EEPROM_RECORD_START (EEPROM_HDR_START + EEPROM_TABLE_HEADER_SIZE)
53 #define EEPROM_MAX_RECORD_NUM ((EEPROM_SIZE_BYTES - EEPROM_TABLE_HEADER_SIZE) / EEPROM_TABLE_RECORD_SIZE)
54 #define EEPROM_ADDR_MSB_MASK GENMASK(17, 8)
56 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_ras, eeprom_control))->adev
58 static void __encode_table_header_to_buff(struct amdgpu_ras_eeprom_table_header
*hdr
,
61 uint32_t *pp
= (uint32_t *) buff
;
63 pp
[0] = cpu_to_le32(hdr
->header
);
64 pp
[1] = cpu_to_le32(hdr
->version
);
65 pp
[2] = cpu_to_le32(hdr
->first_rec_offset
);
66 pp
[3] = cpu_to_le32(hdr
->tbl_size
);
67 pp
[4] = cpu_to_le32(hdr
->checksum
);
70 static void __decode_table_header_from_buff(struct amdgpu_ras_eeprom_table_header
*hdr
,
73 uint32_t *pp
= (uint32_t *)buff
;
75 hdr
->header
= le32_to_cpu(pp
[0]);
76 hdr
->version
= le32_to_cpu(pp
[1]);
77 hdr
->first_rec_offset
= le32_to_cpu(pp
[2]);
78 hdr
->tbl_size
= le32_to_cpu(pp
[3]);
79 hdr
->checksum
= le32_to_cpu(pp
[4]);
82 static int __update_table_header(struct amdgpu_ras_eeprom_control
*control
,
86 struct i2c_msg msg
= {
89 .len
= EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_HEADER_SIZE
,
94 *(uint16_t *)buff
= EEPROM_HDR_START
;
95 __encode_table_header_to_buff(&control
->tbl_hdr
, buff
+ EEPROM_ADDRESS_SIZE
);
97 msg
.addr
= control
->i2c_address
;
99 ret
= i2c_transfer(&control
->eeprom_accessor
, &msg
, 1);
101 DRM_ERROR("Failed to write EEPROM table header, ret:%d", ret
);
108 static uint32_t __calc_hdr_byte_sum(struct amdgpu_ras_eeprom_control
*control
)
111 uint32_t tbl_sum
= 0;
113 /* Header checksum, skip checksum field in the calculation */
114 for (i
= 0; i
< sizeof(control
->tbl_hdr
) - sizeof(control
->tbl_hdr
.checksum
); i
++)
115 tbl_sum
+= *(((unsigned char *)&control
->tbl_hdr
) + i
);
120 static uint32_t __calc_recs_byte_sum(struct eeprom_table_record
*records
,
124 uint32_t tbl_sum
= 0;
126 /* Records checksum */
127 for (i
= 0; i
< num
; i
++) {
128 struct eeprom_table_record
*record
= &records
[i
];
130 for (j
= 0; j
< sizeof(*record
); j
++) {
131 tbl_sum
+= *(((unsigned char *)record
) + j
);
138 static inline uint32_t __calc_tbl_byte_sum(struct amdgpu_ras_eeprom_control
*control
,
139 struct eeprom_table_record
*records
, int num
)
141 return __calc_hdr_byte_sum(control
) + __calc_recs_byte_sum(records
, num
);
144 /* Checksum = 256 -((sum of all table entries) mod 256) */
145 static void __update_tbl_checksum(struct amdgpu_ras_eeprom_control
*control
,
146 struct eeprom_table_record
*records
, int num
,
147 uint32_t old_hdr_byte_sum
)
150 * This will update the table sum with new records.
152 * TODO: What happens when the EEPROM table is to be wrapped around
153 * and old records from start will get overridden.
156 /* need to recalculate updated header byte sum */
157 control
->tbl_byte_sum
-= old_hdr_byte_sum
;
158 control
->tbl_byte_sum
+= __calc_tbl_byte_sum(control
, records
, num
);
160 control
->tbl_hdr
.checksum
= 256 - (control
->tbl_byte_sum
% 256);
163 /* table sum mod 256 + checksum must equals 256 */
164 static bool __validate_tbl_checksum(struct amdgpu_ras_eeprom_control
*control
,
165 struct eeprom_table_record
*records
, int num
)
167 control
->tbl_byte_sum
= __calc_tbl_byte_sum(control
, records
, num
);
169 if (control
->tbl_hdr
.checksum
+ (control
->tbl_byte_sum
% 256) != 256) {
170 DRM_WARN("Checksum mismatch, checksum: %u ", control
->tbl_hdr
.checksum
);
177 int amdgpu_ras_eeprom_reset_table(struct amdgpu_ras_eeprom_control
*control
)
179 unsigned char buff
[EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_HEADER_SIZE
] = { 0 };
180 struct amdgpu_ras_eeprom_table_header
*hdr
= &control
->tbl_hdr
;
183 mutex_lock(&control
->tbl_mutex
);
185 hdr
->header
= EEPROM_TABLE_HDR_VAL
;
186 hdr
->version
= EEPROM_TABLE_VER
;
187 hdr
->first_rec_offset
= EEPROM_RECORD_START
;
188 hdr
->tbl_size
= EEPROM_TABLE_HEADER_SIZE
;
190 control
->tbl_byte_sum
= 0;
191 __update_tbl_checksum(control
, NULL
, 0, 0);
192 control
->next_addr
= EEPROM_RECORD_START
;
194 ret
= __update_table_header(control
, buff
);
196 mutex_unlock(&control
->tbl_mutex
);
202 int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control
*control
)
205 struct amdgpu_device
*adev
= to_amdgpu_device(control
);
206 unsigned char buff
[EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_HEADER_SIZE
] = { 0 };
207 struct amdgpu_ras_eeprom_table_header
*hdr
= &control
->tbl_hdr
;
208 struct i2c_msg msg
= {
211 .len
= EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_HEADER_SIZE
,
215 mutex_init(&control
->tbl_mutex
);
217 switch (adev
->asic_type
) {
219 control
->i2c_address
= EEPROM_I2C_TARGET_ADDR_VEGA20
;
220 ret
= smu_v11_0_i2c_eeprom_control_init(&control
->eeprom_accessor
);
224 control
->i2c_address
= EEPROM_I2C_TARGET_ADDR_ARCTURUS
;
225 ret
= smu_i2c_eeprom_init(&adev
->smu
, &control
->eeprom_accessor
);
233 DRM_ERROR("Failed to init I2C controller, ret:%d", ret
);
237 msg
.addr
= control
->i2c_address
;
239 /* Read/Create table header from EEPROM address 0 */
240 ret
= i2c_transfer(&control
->eeprom_accessor
, &msg
, 1);
242 DRM_ERROR("Failed to read EEPROM table header, ret:%d", ret
);
246 __decode_table_header_from_buff(hdr
, &buff
[2]);
248 if (hdr
->header
== EEPROM_TABLE_HDR_VAL
) {
249 control
->num_recs
= (hdr
->tbl_size
- EEPROM_TABLE_HEADER_SIZE
) /
250 EEPROM_TABLE_RECORD_SIZE
;
251 control
->tbl_byte_sum
= __calc_hdr_byte_sum(control
);
252 control
->next_addr
= EEPROM_RECORD_START
;
254 DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
258 DRM_INFO("Creating new EEPROM table");
260 ret
= amdgpu_ras_eeprom_reset_table(control
);
263 return ret
== 1 ? 0 : -EIO
;
266 void amdgpu_ras_eeprom_fini(struct amdgpu_ras_eeprom_control
*control
)
268 struct amdgpu_device
*adev
= to_amdgpu_device(control
);
270 switch (adev
->asic_type
) {
272 smu_v11_0_i2c_eeprom_control_fini(&control
->eeprom_accessor
);
275 smu_i2c_eeprom_fini(&adev
->smu
, &control
->eeprom_accessor
);
283 static void __encode_table_record_to_buff(struct amdgpu_ras_eeprom_control
*control
,
284 struct eeprom_table_record
*record
,
290 /* Next are all record fields according to EEPROM page spec in LE foramt */
291 buff
[i
++] = record
->err_type
;
293 buff
[i
++] = record
->bank
;
295 tmp
= cpu_to_le64(record
->ts
);
296 memcpy(buff
+ i
, &tmp
, 8);
299 tmp
= cpu_to_le64((record
->offset
& 0xffffffffffff));
300 memcpy(buff
+ i
, &tmp
, 6);
303 buff
[i
++] = record
->mem_channel
;
304 buff
[i
++] = record
->mcumc_id
;
306 tmp
= cpu_to_le64((record
->retired_page
& 0xffffffffffff));
307 memcpy(buff
+ i
, &tmp
, 6);
310 static void __decode_table_record_from_buff(struct amdgpu_ras_eeprom_control
*control
,
311 struct eeprom_table_record
*record
,
317 /* Next are all record fields according to EEPROM page spec in LE foramt */
318 record
->err_type
= buff
[i
++];
320 record
->bank
= buff
[i
++];
322 memcpy(&tmp
, buff
+ i
, 8);
323 record
->ts
= le64_to_cpu(tmp
);
326 memcpy(&tmp
, buff
+ i
, 6);
327 record
->offset
= (le64_to_cpu(tmp
) & 0xffffffffffff);
330 record
->mem_channel
= buff
[i
++];
331 record
->mcumc_id
= buff
[i
++];
333 memcpy(&tmp
, buff
+ i
, 6);
334 record
->retired_page
= (le64_to_cpu(tmp
) & 0xffffffffffff);
338 * When reaching end of EEPROM memory jump back to 0 record address
339 * When next record access will go beyond EEPROM page boundary modify bits A17/A8
340 * in I2C selector to go to next page
342 static uint32_t __correct_eeprom_dest_address(uint32_t curr_address
)
344 uint32_t next_address
= curr_address
+ EEPROM_TABLE_RECORD_SIZE
;
346 /* When all EEPROM memory used jump back to 0 address */
347 if (next_address
> EEPROM_SIZE_BYTES
) {
348 DRM_INFO("Reached end of EEPROM memory, jumping to 0 "
349 "and overriding old record");
350 return EEPROM_RECORD_START
;
354 * To check if we overflow page boundary compare next address with
355 * current and see if bits 17/8 of the EEPROM address will change
356 * If they do start from the next 256b page
358 * https://www.st.com/resource/en/datasheet/m24m02-dr.pdf sec. 5.1.2
360 if ((curr_address
& EEPROM_ADDR_MSB_MASK
) != (next_address
& EEPROM_ADDR_MSB_MASK
)) {
361 DRM_DEBUG_DRIVER("Reached end of EEPROM memory page, jumping to next: %lx",
362 (next_address
& EEPROM_ADDR_MSB_MASK
));
364 return (next_address
& EEPROM_ADDR_MSB_MASK
);
370 int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control
*control
,
371 struct eeprom_table_record
*records
,
376 struct i2c_msg
*msgs
, *msg
;
377 unsigned char *buffs
, *buff
;
378 struct eeprom_table_record
*record
;
379 struct amdgpu_device
*adev
= to_amdgpu_device(control
);
381 if (adev
->asic_type
!= CHIP_VEGA20
&& adev
->asic_type
!= CHIP_ARCTURUS
)
384 buffs
= kcalloc(num
, EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_RECORD_SIZE
,
389 mutex_lock(&control
->tbl_mutex
);
391 msgs
= kcalloc(num
, sizeof(*msgs
), GFP_KERNEL
);
397 /* In case of overflow just start from beginning to not lose newest records */
398 if (write
&& (control
->next_addr
+ EEPROM_TABLE_RECORD_SIZE
* num
> EEPROM_SIZE_BYTES
))
399 control
->next_addr
= EEPROM_RECORD_START
;
403 * TODO Currently makes EEPROM writes for each record, this creates
404 * internal fragmentation. Optimized the code to do full page write of
407 for (i
= 0; i
< num
; i
++) {
408 buff
= &buffs
[i
* (EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_RECORD_SIZE
)];
409 record
= &records
[i
];
412 control
->next_addr
= __correct_eeprom_dest_address(control
->next_addr
);
415 * Update bits 16,17 of EEPROM address in I2C address by setting them
416 * to bits 1,2 of Device address byte
418 msg
->addr
= control
->i2c_address
|
419 ((control
->next_addr
& EEPROM_ADDR_MSB_MASK
) >> 15);
420 msg
->flags
= write
? 0 : I2C_M_RD
;
421 msg
->len
= EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_RECORD_SIZE
;
424 /* Insert the EEPROM dest addess, bits 0-15 */
425 buff
[0] = ((control
->next_addr
>> 8) & 0xff);
426 buff
[1] = (control
->next_addr
& 0xff);
428 /* EEPROM table content is stored in LE format */
430 __encode_table_record_to_buff(control
, record
, buff
+ EEPROM_ADDRESS_SIZE
);
433 * The destination EEPROM address might need to be corrected to account
434 * for page or entire memory wrapping
436 control
->next_addr
+= EEPROM_TABLE_RECORD_SIZE
;
439 ret
= i2c_transfer(&control
->eeprom_accessor
, msgs
, num
);
441 DRM_ERROR("Failed to process EEPROM table records, ret:%d", ret
);
443 /* TODO Restore prev next EEPROM address ? */
449 for (i
= 0; i
< num
; i
++) {
450 buff
= &buffs
[i
*(EEPROM_ADDRESS_SIZE
+ EEPROM_TABLE_RECORD_SIZE
)];
451 record
= &records
[i
];
453 __decode_table_record_from_buff(control
, record
, buff
+ EEPROM_ADDRESS_SIZE
);
458 uint32_t old_hdr_byte_sum
= __calc_hdr_byte_sum(control
);
461 * Update table header with size and CRC and account for table
462 * wrap around where the assumption is that we treat it as empty
465 * TODO - Check the assumption is correct
467 control
->num_recs
+= num
;
468 control
->num_recs
%= EEPROM_MAX_RECORD_NUM
;
469 control
->tbl_hdr
.tbl_size
+= EEPROM_TABLE_RECORD_SIZE
* num
;
470 if (control
->tbl_hdr
.tbl_size
> EEPROM_SIZE_BYTES
)
471 control
->tbl_hdr
.tbl_size
= EEPROM_TABLE_HEADER_SIZE
+
472 control
->num_recs
* EEPROM_TABLE_RECORD_SIZE
;
474 __update_tbl_checksum(control
, records
, num
, old_hdr_byte_sum
);
476 __update_table_header(control
, buffs
);
477 } else if (!__validate_tbl_checksum(control
, records
, num
)) {
478 DRM_WARN("EEPROM Table checksum mismatch!");
479 /* TODO Uncomment when EEPROM read/write is relliable */
489 mutex_unlock(&control
->tbl_mutex
);
491 return ret
== num
? 0 : -EIO
;
494 /* Used for testing if bugs encountered */
496 void amdgpu_ras_eeprom_test(struct amdgpu_ras_eeprom_control
*control
)
499 struct eeprom_table_record
*recs
= kcalloc(1, sizeof(*recs
), GFP_KERNEL
);
504 for (i
= 0; i
< 1 ; i
++) {
505 recs
[i
].address
= 0xdeadbeef;
506 recs
[i
].retired_page
= i
;
509 if (!amdgpu_ras_eeprom_process_recods(control
, recs
, true, 1)) {
511 memset(recs
, 0, sizeof(*recs
) * 1);
513 control
->next_addr
= EEPROM_RECORD_START
;
515 if (!amdgpu_ras_eeprom_process_recods(control
, recs
, false, 1)) {
516 for (i
= 0; i
< 1; i
++)
517 DRM_INFO("rec.address :0x%llx, rec.retired_page :%llu",
518 recs
[i
].address
, recs
[i
].retired_page
);
520 DRM_ERROR("Failed in reading from table");
523 DRM_ERROR("Failed in writing to table");