drm/amdgpu: Add RAS EEPROM table.
authorAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Thu, 30 May 2019 15:01:26 +0000 (11:01 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Tue, 27 Aug 2019 13:17:14 +0000 (08:17 -0500)
Add RAS EEPROM table manager to eanble RAS errors to be stored
upon appearance and retrived on driver load.

v2: Fix some prints.

v3:
Fix checksum calculation.
Make table record and header structs packed to do correct byte value sum.
Fix record crossing EEPROM page boundry.

v4:
Fix byte sum val calculation for record - look at sizeof(record).
Fix some style comments.

v5: Add description to EEPROM_TABLE_RECORD_SIZE and syntax fixes.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Luben Tuikov <Luben.Tuikov@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/Makefile
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c [new file with mode: 0644]
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h [new file with mode: 0644]

index 28d76bd950517088edfd652fc250a9de9f2d10fb..f016cf118b1b2170d02109c9168f11649f80ab14 100644 (file)
@@ -54,7 +54,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
        amdgpu_gtt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o amdgpu_atomfirmware.o \
        amdgpu_vf_error.o amdgpu_sched.o amdgpu_debugfs.o amdgpu_ids.o \
        amdgpu_gmc.o amdgpu_xgmi.o amdgpu_csa.o amdgpu_ras.o amdgpu_vm_cpu.o \
-       amdgpu_vm_sdma.o amdgpu_discovery.o
+       amdgpu_vm_sdma.o amdgpu_pmu.o amdgpu_discovery.o amdgpu_ras_eeprom.o
 
 amdgpu-$(CONFIG_PERF_EVENTS) += amdgpu_pmu.o
 
index 02a51e3dfa14cad95f80e0d62ffcfcc77243a604..6c76bb2a684325925ca7952d0d0a41c3372096e4 100644 (file)
@@ -29,6 +29,7 @@
 #include "amdgpu.h"
 #include "amdgpu_psp.h"
 #include "ta_ras_if.h"
+#include "amdgpu_ras_eeprom.h"
 
 enum amdgpu_ras_block {
        AMDGPU_RAS_BLOCK__UMC = 0,
@@ -333,6 +334,8 @@ struct amdgpu_ras {
        struct mutex recovery_lock;
 
        uint32_t flags;
+
+       struct amdgpu_ras_eeprom_control eeprom_control;
 };
 
 struct ras_fs_data {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
new file mode 100644 (file)
index 0000000..b544e0a
--- /dev/null
@@ -0,0 +1,492 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "amdgpu_ras_eeprom.h"
+#include "amdgpu.h"
+#include "amdgpu_ras.h"
+#include <linux/bits.h>
+
+#define EEPROM_I2C_TARGET_ADDR 0xA0
+
+/*
+ * The 2 macros bellow represent the actual size in bytes that
+ * those entities occupy in the EEPROM memory.
+ * EEPROM_TABLE_RECORD_SIZE is different than sizeof(eeprom_table_record) which
+ * uses uint64 to store 6b fields such as retired_page.
+ */
+#define EEPROM_TABLE_HEADER_SIZE 20
+#define EEPROM_TABLE_RECORD_SIZE 24
+
+#define EEPROM_ADDRESS_SIZE 0x2
+
+/* Table hdr is 'AMDR' */
+#define EEPROM_TABLE_HDR_VAL 0x414d4452
+#define EEPROM_TABLE_VER 0x00010000
+
+/* Assume 2 Mbit size */
+#define EEPROM_SIZE_BYTES 256000
+#define EEPROM_PAGE__SIZE_BYTES 256
+#define EEPROM_HDR_START 0
+#define EEPROM_RECORD_START (EEPROM_HDR_START + EEPROM_TABLE_HEADER_SIZE)
+#define EEPROM_MAX_RECORD_NUM ((EEPROM_SIZE_BYTES - EEPROM_TABLE_HEADER_SIZE) / EEPROM_TABLE_RECORD_SIZE)
+#define EEPROM_ADDR_MSB_MASK GENMASK(17, 8)
+
+#define to_amdgpu_device(x) (container_of(x, struct amdgpu_ras, eeprom_control))->adev
+
+static void __encode_table_header_to_buff(struct amdgpu_ras_eeprom_table_header *hdr,
+                                         unsigned char *buff)
+{
+       uint32_t *pp = (uint32_t *) buff;
+
+       pp[0] = cpu_to_le32(hdr->header);
+       pp[1] = cpu_to_le32(hdr->version);
+       pp[2] = cpu_to_le32(hdr->first_rec_offset);
+       pp[3] = cpu_to_le32(hdr->tbl_size);
+       pp[4] = cpu_to_le32(hdr->checksum);
+}
+
+static void __decode_table_header_from_buff(struct amdgpu_ras_eeprom_table_header *hdr,
+                                         unsigned char *buff)
+{
+       uint32_t *pp = (uint32_t *)buff;
+
+       hdr->header           = le32_to_cpu(pp[0]);
+       hdr->version          = le32_to_cpu(pp[1]);
+       hdr->first_rec_offset = le32_to_cpu(pp[2]);
+       hdr->tbl_size         = le32_to_cpu(pp[3]);
+       hdr->checksum         = le32_to_cpu(pp[4]);
+}
+
+static int __update_table_header(struct amdgpu_ras_eeprom_control *control,
+                                unsigned char *buff)
+{
+       int ret = 0;
+       struct i2c_msg msg = {
+                       .addr   = EEPROM_I2C_TARGET_ADDR,
+                       .flags  = 0,
+                       .len    = EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
+                       .buf    = buff,
+       };
+
+
+       *(uint16_t *)buff = EEPROM_HDR_START;
+       __encode_table_header_to_buff(&control->tbl_hdr, buff + EEPROM_ADDRESS_SIZE);
+
+       ret = i2c_transfer(&control->eeprom_accessor, &msg, 1);
+       if (ret < 1)
+               DRM_ERROR("Failed to write EEPROM table header, ret:%d", ret);
+
+       return ret;
+}
+
+static uint32_t  __calc_hdr_byte_sum(struct amdgpu_ras_eeprom_control *control);
+
+int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control)
+{
+       int ret = 0;
+       struct amdgpu_device *adev = to_amdgpu_device(control);
+       unsigned char buff[EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE] = { 0 };
+       struct amdgpu_ras_eeprom_table_header *hdr = &control->tbl_hdr;
+       struct i2c_msg msg = {
+                       .addr   = EEPROM_I2C_TARGET_ADDR,
+                       .flags  = I2C_M_RD,
+                       .len    = EEPROM_ADDRESS_SIZE + EEPROM_TABLE_HEADER_SIZE,
+                       .buf    = buff,
+       };
+
+       mutex_init(&control->tbl_mutex);
+
+       switch (adev->asic_type) {
+       case CHIP_VEGA20:
+       /*TODO Add MI-60 */
+               break;
+
+       default:
+               return 0;
+       }
+
+       if (ret) {
+               DRM_ERROR("Failed to init I2C controller, ret:%d", ret);
+               return ret;
+       }
+
+       /* Read/Create table header from EEPROM address 0 */
+       ret = i2c_transfer(&control->eeprom_accessor, &msg, 1);
+       if (ret < 1) {
+               DRM_ERROR("Failed to read EEPROM table header, ret:%d", ret);
+               return ret;
+       }
+
+       __decode_table_header_from_buff(hdr, &buff[2]);
+
+       if (hdr->header == EEPROM_TABLE_HDR_VAL) {
+               control->num_recs = (hdr->tbl_size - EEPROM_TABLE_HEADER_SIZE) /
+                                   EEPROM_TABLE_RECORD_SIZE;
+               DRM_DEBUG_DRIVER("Found existing EEPROM table with %d records",
+                                control->num_recs);
+
+       } else {
+               DRM_INFO("Creating new EEPROM table");
+
+               hdr->header = EEPROM_TABLE_HDR_VAL;
+               hdr->version = EEPROM_TABLE_VER;
+               hdr->first_rec_offset = EEPROM_RECORD_START;
+               hdr->tbl_size = EEPROM_TABLE_HEADER_SIZE;
+
+               adev->psp.ras.ras->eeprom_control.tbl_byte_sum =
+                               __calc_hdr_byte_sum(&adev->psp.ras.ras->eeprom_control);
+               ret = __update_table_header(control, buff);
+       }
+
+       /* Start inserting records from here */
+       adev->psp.ras.ras->eeprom_control.next_addr = EEPROM_RECORD_START;
+
+       return ret == 1 ? 0 : -EIO;
+}
+
+void amdgpu_ras_eeprom_fini(struct amdgpu_ras_eeprom_control *control)
+{
+       struct amdgpu_device *adev = to_amdgpu_device(control);
+
+       switch (adev->asic_type) {
+       case CHIP_VEGA20:
+               /*TODO Add MI-60 */
+               break;
+
+       default:
+               return;
+       }
+}
+
+static void __encode_table_record_to_buff(struct amdgpu_ras_eeprom_control *control,
+                                         struct eeprom_table_record *record,
+                                         unsigned char *buff)
+{
+       __le64 tmp = 0;
+       int i = 0;
+
+       /* Next are all record fields according to EEPROM page spec in LE foramt */
+       buff[i++] = record->err_type;
+
+       buff[i++] = record->bank;
+
+       tmp = cpu_to_le64(record->ts);
+       memcpy(buff + i, &tmp, 8);
+       i += 8;
+
+       tmp = cpu_to_le64((record->offset & 0xffffffffffff));
+       memcpy(buff + i, &tmp, 6);
+       i += 6;
+
+       buff[i++] = record->mem_channel;
+       buff[i++] = record->mcumc_id;
+
+       tmp = cpu_to_le64((record->retired_page & 0xffffffffffff));
+       memcpy(buff + i, &tmp, 6);
+}
+
+static void __decode_table_record_from_buff(struct amdgpu_ras_eeprom_control *control,
+                                           struct eeprom_table_record *record,
+                                           unsigned char *buff)
+{
+       __le64 tmp = 0;
+       int i =  0;
+
+       /* Next are all record fields according to EEPROM page spec in LE foramt */
+       record->err_type = buff[i++];
+
+       record->bank = buff[i++];
+
+       memcpy(&tmp, buff + i, 8);
+       record->ts = le64_to_cpu(tmp);
+       i += 8;
+
+       memcpy(&tmp, buff + i, 6);
+       record->offset = (le64_to_cpu(tmp) & 0xffffffffffff);
+       i += 6;
+
+       buff[i++] = record->mem_channel;
+       buff[i++] = record->mcumc_id;
+
+       memcpy(&tmp, buff + i,  6);
+       record->retired_page = (le64_to_cpu(tmp) & 0xffffffffffff);
+}
+
+/*
+ * When reaching end of EEPROM memory jump back to 0 record address
+ * When next record access will go beyond EEPROM page boundary modify bits A17/A8
+ * in I2C selector to go to next page
+ */
+static uint32_t __correct_eeprom_dest_address(uint32_t curr_address)
+{
+       uint32_t next_address = curr_address + EEPROM_TABLE_RECORD_SIZE;
+
+       /* When all EEPROM memory used jump back to 0 address */
+       if (next_address > EEPROM_SIZE_BYTES) {
+               DRM_INFO("Reached end of EEPROM memory, jumping to 0 "
+                        "and overriding old record");
+               return EEPROM_RECORD_START;
+       }
+
+       /*
+        * To check if we overflow page boundary  compare next address with
+        * current and see if bits 17/8 of the EEPROM address will change
+        * If they do start from the next 256b page
+        *
+        * https://www.st.com/resource/en/datasheet/m24m02-dr.pdf sec. 5.1.2
+        */
+       if ((curr_address & EEPROM_ADDR_MSB_MASK) != (next_address & EEPROM_ADDR_MSB_MASK)) {
+               DRM_DEBUG_DRIVER("Reached end of EEPROM memory page, jumpimng to next: %lx",
+                               (next_address & EEPROM_ADDR_MSB_MASK));
+
+               return  (next_address & EEPROM_ADDR_MSB_MASK);
+       }
+
+       return curr_address;
+}
+
+
+static uint32_t  __calc_hdr_byte_sum(struct amdgpu_ras_eeprom_control *control)
+{
+       int i;
+       uint32_t tbl_sum = 0;
+
+       /* Header checksum, skip checksum field in the calculation */
+       for (i = 0; i < sizeof(control->tbl_hdr) - sizeof(control->tbl_hdr.checksum); i++)
+               tbl_sum += *(((unsigned char *)&control->tbl_hdr) + i);
+
+       return tbl_sum;
+}
+
+static uint32_t  __calc_recs_byte_sum(struct eeprom_table_record *records,
+                                     int num)
+{
+       int i, j;
+       uint32_t tbl_sum = 0;
+
+       /* Records checksum */
+       for (i = 0; i < num; i++) {
+               struct eeprom_table_record *record = &records[i];
+
+               for (j = 0; j < sizeof(*record); j++) {
+                       tbl_sum += *(((unsigned char *)record) + j);
+               }
+       }
+
+       return tbl_sum;
+}
+
+static inline uint32_t  __calc_tbl_byte_sum(struct amdgpu_ras_eeprom_control *control,
+                                 struct eeprom_table_record *records, int num)
+{
+       return __calc_hdr_byte_sum(control) + __calc_recs_byte_sum(records, num);
+}
+
+/* Checksum = 256 -((sum of all table entries) mod 256) */
+static void __update_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
+                                 struct eeprom_table_record *records, int num,
+                                 uint32_t old_hdr_byte_sum)
+{
+       /*
+        * This will update the table sum with new records.
+        *
+        * TODO: What happens when the EEPROM table is to be wrapped around
+        * and old records from start will get overridden.
+        */
+
+       /* need to recalculate updated header byte sum */
+       control->tbl_byte_sum -= old_hdr_byte_sum;
+       control->tbl_byte_sum += __calc_tbl_byte_sum(control, records, num);
+
+       control->tbl_hdr.checksum = 256 - (control->tbl_byte_sum % 256);
+}
+
+/* table sum mod 256 + checksum must equals 256 */
+static bool __validate_tbl_checksum(struct amdgpu_ras_eeprom_control *control,
+                           struct eeprom_table_record *records, int num)
+{
+       control->tbl_byte_sum = __calc_tbl_byte_sum(control, records, num);
+
+       if (control->tbl_hdr.checksum + (control->tbl_byte_sum % 256) != 256) {
+               DRM_WARN("Checksum mismatch, checksum: %u ", control->tbl_hdr.checksum);
+               return false;
+       }
+
+       return true;
+}
+
+int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control *control,
+                                           struct eeprom_table_record *records,
+                                           bool write,
+                                           int num)
+{
+       int i, ret = 0;
+       struct i2c_msg *msgs;
+       unsigned char *buffs;
+       struct amdgpu_device *adev = to_amdgpu_device(control);
+
+       if (adev->asic_type != CHIP_VEGA20)
+               return 0;
+
+       buffs = kcalloc(num, EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE,
+                        GFP_KERNEL);
+       if (!buffs)
+               return -ENOMEM;
+
+       mutex_lock(&control->tbl_mutex);
+
+       msgs = kcalloc(num, sizeof(*msgs), GFP_KERNEL);
+       if (!msgs) {
+               ret = -ENOMEM;
+               goto free_buff;
+       }
+
+       /* In case of overflow just start from beginning to not lose newest records */
+       if (write && (control->next_addr + EEPROM_TABLE_RECORD_SIZE * num > EEPROM_SIZE_BYTES))
+               control->next_addr = EEPROM_RECORD_START;
+
+
+       /*
+        * TODO Currently makes EEPROM writes for each record, this creates
+        * internal fragmentation. Optimized the code to do full page write of
+        * 256b
+        */
+       for (i = 0; i < num; i++) {
+               unsigned char *buff = &buffs[i * (EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
+               struct eeprom_table_record *record = &records[i];
+               struct i2c_msg *msg = &msgs[i];
+
+               control->next_addr = __correct_eeprom_dest_address(control->next_addr);
+
+               /*
+                * Update bits 16,17 of EEPROM address in I2C address by setting them
+                * to bits 1,2 of Device address byte
+                */
+               msg->addr = EEPROM_I2C_TARGET_ADDR |
+                              ((control->next_addr & EEPROM_ADDR_MSB_MASK) >> 15);
+               msg->flags      = write ? 0 : I2C_M_RD;
+               msg->len        = EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE;
+               msg->buf        = buff;
+
+               /* Insert the EEPROM dest addess, bits 0-15 */
+               buff[0] = ((control->next_addr >> 8) & 0xff);
+               buff[1] = (control->next_addr & 0xff);
+
+               /* EEPROM table content is stored in LE format */
+               if (write)
+                       __encode_table_record_to_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
+
+               /*
+                * The destination EEPROM address might need to be corrected to account
+                * for page or entire memory wrapping
+                */
+               control->next_addr += EEPROM_TABLE_RECORD_SIZE;
+       }
+
+       ret = i2c_transfer(&control->eeprom_accessor, msgs, num);
+       if (ret < 1) {
+               DRM_ERROR("Failed to process EEPROM table records, ret:%d", ret);
+
+               /* TODO Restore prev next EEPROM address ? */
+               goto free_msgs;
+       }
+
+
+       if (!write) {
+               for (i = 0; i < num; i++) {
+                       unsigned char *buff = &buffs[i*(EEPROM_ADDRESS_SIZE + EEPROM_TABLE_RECORD_SIZE)];
+                       struct eeprom_table_record *record = &records[i];
+
+                       __decode_table_record_from_buff(control, record, buff + EEPROM_ADDRESS_SIZE);
+               }
+       }
+
+       if (write) {
+               uint32_t old_hdr_byte_sum = __calc_hdr_byte_sum(control);
+
+               /*
+                * Update table header with size and CRC and account for table
+                * wrap around where the assumption is that we treat it as empty
+                * table
+                *
+                * TODO - Check the assumption is correct
+                */
+               control->num_recs += num;
+               control->num_recs %= EEPROM_MAX_RECORD_NUM;
+               control->tbl_hdr.tbl_size += EEPROM_TABLE_RECORD_SIZE * num;
+               if (control->tbl_hdr.tbl_size > EEPROM_SIZE_BYTES)
+                       control->tbl_hdr.tbl_size = EEPROM_TABLE_HEADER_SIZE +
+                       control->num_recs * EEPROM_TABLE_RECORD_SIZE;
+
+               __update_tbl_checksum(control, records, num, old_hdr_byte_sum);
+
+               __update_table_header(control, buffs);
+       } else if (!__validate_tbl_checksum(control, records, num)) {
+               DRM_WARN("EEPROM Table checksum mismatch!");
+               /* TODO Uncomment when EEPROM read/write is relliable */
+               /* ret = -EIO; */
+       }
+
+free_msgs:
+       kfree(msgs);
+
+free_buff:
+       kfree(buffs);
+
+       mutex_unlock(&control->tbl_mutex);
+
+       return ret == num ? 0 : -EIO;
+}
+
+/* Used for testing if bugs encountered */
+#if 0
+void amdgpu_ras_eeprom_test(struct amdgpu_ras_eeprom_control *control)
+{
+       int i;
+       struct eeprom_table_record *recs = kcalloc(1, sizeof(*recs), GFP_KERNEL);
+
+       if (!recs)
+               return;
+
+       for (i = 0; i < 1 ; i++) {
+               recs[i].address = 0xdeadbeef;
+               recs[i].retired_page = i;
+       }
+
+       if (!amdgpu_ras_eeprom_process_recods(control, recs, true, 1)) {
+
+               memset(recs, 0, sizeof(*recs) * 1);
+
+               control->next_addr = EEPROM_RECORD_START;
+
+               if (!amdgpu_ras_eeprom_process_recods(control, recs, false, 1)) {
+                       for (i = 0; i < 1; i++)
+                               DRM_INFO("rec.address :0x%llx, rec.retired_page :%llu",
+                                        recs[i].address, recs[i].retired_page);
+               } else
+                       DRM_ERROR("Failed in reading from table");
+
+       } else
+               DRM_ERROR("Failed in writing to table");
+}
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
new file mode 100644 (file)
index 0000000..41f3fcb
--- /dev/null
@@ -0,0 +1,90 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef _AMDGPU_RAS_EEPROM_H
+#define _AMDGPU_RAS_EEPROM_H
+
+#include <linux/i2c.h>
+
+struct amdgpu_device;
+
+enum amdgpu_ras_eeprom_err_type{
+       AMDGPU_RAS_EEPROM_ERR_PLACE_HOLDER,
+       AMDGPU_RAS_EEPROM_ERR_RECOVERABLE,
+       AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE
+};
+
+struct amdgpu_ras_eeprom_table_header {
+       uint32_t header;
+       uint32_t version;
+       uint32_t first_rec_offset;
+       uint32_t tbl_size;
+       uint32_t checksum;
+}__attribute__((__packed__));
+
+struct amdgpu_ras_eeprom_control {
+       struct amdgpu_ras_eeprom_table_header tbl_hdr;
+       struct i2c_adapter eeprom_accessor;
+       uint32_t next_addr;
+       unsigned int num_recs;
+       struct mutex tbl_mutex;
+       bool bus_locked;
+       uint32_t tbl_byte_sum;
+};
+
+/*
+ * Represents single table record. Packed to be easily serialized into byte
+ * stream.
+ */
+struct eeprom_table_record {
+
+       union {
+               uint64_t address;
+               uint64_t offset;
+       };
+
+       uint64_t retired_page;
+       uint64_t ts;
+
+       enum amdgpu_ras_eeprom_err_type err_type;
+
+       union {
+               unsigned char bank;
+               unsigned char cu;
+       };
+
+       unsigned char mem_channel;
+       unsigned char mcumc_id;
+}__attribute__((__packed__));
+
+int amdgpu_ras_eeprom_init(struct amdgpu_ras_eeprom_control *control);
+void amdgpu_ras_eeprom_fini(struct amdgpu_ras_eeprom_control *control);
+
+int amdgpu_ras_eeprom_process_recods(struct amdgpu_ras_eeprom_control *control,
+                                           struct eeprom_table_record *records,
+                                           bool write,
+                                           int num);
+
+void amdgpu_ras_eeprom_test(struct amdgpu_ras_eeprom_control *control);
+
+#endif // _AMDGPU_RAS_EEPROM_H