[PATCH v4 1/4] drm/amdgpu: Add RAS EEPROM table.

Zhou1, Tao Tao.Zhou1 at amd.com
Fri Aug 23 02:16:56 UTC 2019



From: Grodzovsky, Andrey <Andrey.Grodzovsky at amd.com> 
Sent: 2019年8月22日 23:07
To: Zhou1, Tao <Tao.Zhou1 at amd.com>; amd-gfx at lists.freedesktop.org
Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Pan, Xinhui <Xinhui.Pan at amd.com>; Zhang, Hawking <Hawking.Zhang at amd.com>; Tuikov, Luben <Luben.Tuikov at amd.com>; Lazar, Lijo <Lijo.Lazar at amd.com>; Quan, Evan <Evan.Quan at amd.com>; Panariti, David <David.Panariti at amd.com>; Russell, Kent <Kent.Russell at amd.com>
Subject: Re: [PATCH v4 1/4] drm/amdgpu: Add RAS EEPROM table.


On 8/22/19 12:13 AM, Zhou1, Tao wrote:


-----Original Message-----
From: Andrey Grodzovsky mailto:andrey.grodzovsky at amd.com
Sent: 2019年8月22日 4:02
To: mailto:amd-gfx at lists.freedesktop.org
Cc: Deucher, Alexander mailto:Alexander.Deucher at amd.com; Pan, Xinhui
mailto:Xinhui.Pan at amd.com; Zhang, Hawking mailto:Hawking.Zhang at amd.com;
Tuikov, Luben mailto:Luben.Tuikov at amd.com; Lazar, Lijo mailto:Lijo.Lazar at amd.com;
Quan, Evan mailto:Evan.Quan at amd.com; Panariti, David
mailto:David.Panariti at amd.com; Russell, Kent mailto:Kent.Russell at amd.com; Zhou1,
Tao mailto:Tao.Zhou1 at amd.com; Grodzovsky, Andrey
mailto:Andrey.Grodzovsky at amd.com
Subject: [PATCH v4 1/4] drm/amdgpu: Add RAS EEPROM table.

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.

Signed-off-by: Andrey Grodzovsky mailto:andrey.grodzovsky at amd.com
---
 drivers/gpu/drm/amd/amdgpu/Makefile            |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h        |   3 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c | 482
+++++++++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h |  90 +++++
 4 files changed, 576 insertions(+), 1 deletion(-)  create mode 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
 create mode 100644
drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h

diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile
b/drivers/gpu/drm/amd/amdgpu/Makefile
index 28d76bd..f016cf1 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -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

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index 2765f2d..8d5bcd8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -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
index 0000000..bf07515
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
@@ -0,0 +1,482 @@
+/*
+ * 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
+
+#define EEPROM_TABLE_HEADER_SIZE 20
+#define EEPROM_TABLE_RECORD_SIZE 24
[Tao] should we replace fixed value with sizeof for the two macros?

No, as I already explained before the EEPROM_TABLE_HEADER/RECORD_SIZE represent the actual size in bytes that those entities occupy in the EEPROM memory and as defined in the EEPROM table description while struct eeprom_table_record would be larger because for example to store 6b of RetiedPage I have to use uint64 which is 8b. I could of course use some kind of bitvector (bitset) to store them compactly but it seems to me as an overkill. I was told we are not expecting to much records for any given system anyway during it's life time so the dynamic memory saved would be small anyway. 

[Tao] OK, could you add a comment in the code to clarify the two types of SIZE?



+#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;
[Tao] I think sizeof(record->ts) is better

+
+	tmp = cpu_to_le64((record->offset & 0xffffffffffff));
[Tao] (0x1ULL << 49 - 1) is more readable than & 0xffffffffffff, or a macro can be defined, but either way is OK.

+	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); }
[Tao] we can change 256 to EEPROM_PAGE__SIZE_BYTES

This 256 just accidentally has same value as  EEPROM_PAGE__SIZE_BYTES but doesn't relate in any way, it's part of byte value checksum algorithm I was given.
Andrey



+
+/* 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)];
[Tao] space is needed before and after "*"

+			struct eeprom_table_record *record = &records[i];
[Tao] add a space after "*"

+
+			__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;
+}
+
+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"); }
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
index 0000000..41f3fcb
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
@@ -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
--
2.7.4



More information about the amd-gfx mailing list