Commit e4b1edf4 authored by YiPeng Chai's avatar YiPeng Chai Committed by Alex Deucher

drm/amdgpu: add umc ras functions for umc v8_10_0

1. Support query umc ras error counter.
2. Support ras umc ue error address remapping.
Signed-off-by: default avatarYiPeng Chai <YiPeng.Chai@amd.com>
Reviewed-by: default avatarAlexander Deucher <Alexander.Deucher@amd.com>
Reviewed-by: default avatarHawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: default avatarTao Zhou <tao.zhou1@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent 75510fac
......@@ -93,7 +93,7 @@ amdgpu-y += \
# add UMC block
amdgpu-y += \
umc_v6_0.o umc_v6_1.o umc_v6_7.o umc_v8_7.o
umc_v6_0.o umc_v6_1.o umc_v6_7.o umc_v8_7.o umc_v8_10.o
# add IH block
amdgpu-y += \
......
......@@ -41,6 +41,12 @@
#define LOOP_UMC_CH_INST(ch_inst) for ((ch_inst) = 0; (ch_inst) < adev->umc.channel_inst_num; (ch_inst)++)
#define LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) LOOP_UMC_INST((umc_inst)) LOOP_UMC_CH_INST((ch_inst))
#define LOOP_UMC_NODE_INST(node_inst) \
for ((node_inst) = 0; (node_inst) < adev->umc.node_inst_num; (node_inst)++)
#define LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) \
LOOP_UMC_NODE_INST((node_inst)) LOOP_UMC_INST_AND_CH((umc_inst), (ch_inst))
struct amdgpu_umc_ras {
struct amdgpu_ras_block_object ras_block;
void (*err_cnt_init)(struct amdgpu_device *adev);
......@@ -62,6 +68,10 @@ struct amdgpu_umc {
uint32_t channel_inst_num;
/* number of umc instance with memory map register access */
uint32_t umc_inst_num;
/*number of umc node instance with memory map register access*/
uint32_t node_inst_num;
/* UMC regiser per channel offset */
uint32_t channel_offs;
/* channel index table of interleaved memory */
......
......@@ -25,7 +25,7 @@
#include "amdgpu.h"
#include "amdgpu_atomfirmware.h"
#include "gmc_v11_0.h"
#include "umc_v8_7.h"
#include "umc_v8_10.h"
#include "athub/athub_3_0_0_sh_mask.h"
#include "athub/athub_3_0_0_offset.h"
#include "oss/osssys_6_0_0_offset.h"
......@@ -537,11 +537,36 @@ static void gmc_v11_0_set_umc_funcs(struct amdgpu_device *adev)
{
switch (adev->ip_versions[UMC_HWIP][0]) {
case IP_VERSION(8, 10, 0):
adev->umc.channel_inst_num = UMC_V8_10_CHANNEL_INSTANCE_NUM;
adev->umc.umc_inst_num = UMC_V8_10_UMC_INSTANCE_NUM;
adev->umc.node_inst_num = adev->gmc.num_umc;
adev->umc.max_ras_err_cnt_per_query = UMC_V8_10_TOTAL_CHANNEL_NUM(adev);
adev->umc.channel_offs = UMC_V8_10_PER_CHANNEL_OFFSET;
adev->umc.channel_idx_tbl = &umc_v8_10_channel_idx_tbl[0][0][0];
adev->umc.ras = &umc_v8_10_ras;
break;
case IP_VERSION(8, 11, 0):
break;
default:
break;
}
if (adev->umc.ras) {
amdgpu_ras_register_ras_block(adev, &adev->umc.ras->ras_block);
strcpy(adev->umc.ras->ras_block.ras_comm.name, "umc");
adev->umc.ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__UMC;
adev->umc.ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
adev->umc.ras_if = &adev->umc.ras->ras_block.ras_comm;
/* If don't define special ras_late_init function, use default ras_late_init */
if (!adev->umc.ras->ras_block.ras_late_init)
adev->umc.ras->ras_block.ras_late_init = amdgpu_umc_ras_late_init;
/* If not define special ras_cb function, use default ras_cb */
if (!adev->umc.ras->ras_block.ras_cb)
adev->umc.ras->ras_block.ras_cb = amdgpu_umc_process_ras_data_cb;
}
}
......
/*
* Copyright 2022 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 "umc_v8_10.h"
#include "amdgpu_ras.h"
#include "amdgpu_umc.h"
#include "amdgpu.h"
#include "umc/umc_8_10_0_offset.h"
#include "umc/umc_8_10_0_sh_mask.h"
#define UMC_8_NODE_DIST 0x800000
#define UMC_8_INST_DIST 0x4000
struct channelnum_map_colbit {
uint32_t channel_num;
uint32_t col_bit;
};
const struct channelnum_map_colbit umc_v8_10_channelnum_map_colbit_table[] = {
{24, 13},
{20, 13},
{16, 12},
{14, 12},
{12, 12},
{10, 12},
{6, 11},
};
const uint32_t
umc_v8_10_channel_idx_tbl[]
[UMC_V8_10_UMC_INSTANCE_NUM]
[UMC_V8_10_CHANNEL_INSTANCE_NUM] = {
{{16, 18}, {17, 19}},
{{15, 11}, {3, 7}},
{{1, 5}, {13, 9}},
{{23, 21}, {22, 20}},
{{0, 4}, {12, 8}},
{{14, 10}, {2, 6}}
};
static inline uint32_t get_umc_v8_10_reg_offset(struct amdgpu_device *adev,
uint32_t node_inst,
uint32_t umc_inst,
uint32_t ch_inst)
{
return adev->umc.channel_offs * ch_inst + UMC_8_INST_DIST * umc_inst +
UMC_8_NODE_DIST * node_inst;
}
static void umc_v8_10_clear_error_count_per_channel(struct amdgpu_device *adev,
uint32_t umc_reg_offset)
{
uint32_t ecc_err_cnt_addr;
ecc_err_cnt_addr =
SOC15_REG_OFFSET(UMC, 0, regUMCCH0_0_GeccErrCnt);
/* clear error count */
WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4,
UMC_V8_10_CE_CNT_INIT);
}
static void umc_v8_10_clear_error_count(struct amdgpu_device *adev)
{
uint32_t node_inst = 0;
uint32_t umc_inst = 0;
uint32_t ch_inst = 0;
uint32_t umc_reg_offset = 0;
LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
umc_reg_offset = get_umc_v8_10_reg_offset(adev,
node_inst,
umc_inst,
ch_inst);
umc_v8_10_clear_error_count_per_channel(adev,
umc_reg_offset);
}
}
static void umc_v8_10_query_correctable_error_count(struct amdgpu_device *adev,
uint32_t umc_reg_offset,
unsigned long *error_count)
{
uint32_t ecc_err_cnt, ecc_err_cnt_addr;
uint64_t mc_umc_status;
uint32_t mc_umc_status_addr;
/* UMC 8_10 registers */
ecc_err_cnt_addr =
SOC15_REG_OFFSET(UMC, 0, regUMCCH0_0_GeccErrCnt);
mc_umc_status_addr =
SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
ecc_err_cnt = RREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4);
*error_count +=
(REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_GeccErrCnt, GeccErrCnt) -
UMC_V8_10_CE_CNT_INIT);
/* Check for SRAM correctable error, MCUMC_STATUS is a 64 bit register */
mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
*error_count += 1;
}
static void umc_v8_10_query_uncorrectable_error_count(struct amdgpu_device *adev,
uint32_t umc_reg_offset,
unsigned long *error_count)
{
uint64_t mc_umc_status;
uint32_t mc_umc_status_addr;
mc_umc_status_addr = SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
/* Check the MCUMC_STATUS. */
mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
if ((REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
(REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
*error_count += 1;
}
static void umc_v8_10_query_ras_error_count(struct amdgpu_device *adev,
void *ras_error_status)
{
struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
uint32_t node_inst = 0;
uint32_t umc_inst = 0;
uint32_t ch_inst = 0;
uint32_t umc_reg_offset = 0;
LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
umc_reg_offset = get_umc_v8_10_reg_offset(adev,
node_inst,
umc_inst,
ch_inst);
umc_v8_10_query_correctable_error_count(adev,
umc_reg_offset,
&(err_data->ce_count));
umc_v8_10_query_uncorrectable_error_count(adev,
umc_reg_offset,
&(err_data->ue_count));
}
umc_v8_10_clear_error_count(adev);
}
static uint32_t umc_v8_10_get_col_bit(uint32_t channel_num)
{
uint32_t t = 0;
for (t = 0; t < ARRAY_SIZE(umc_v8_10_channelnum_map_colbit_table); t++)
if (channel_num == umc_v8_10_channelnum_map_colbit_table[t].channel_num)
return umc_v8_10_channelnum_map_colbit_table[t].col_bit;
/* Failed to get col_bit. */
return U32_MAX;
}
/*
* Mapping normal address to soc physical address in swizzle mode.
*/
static int umc_v8_10_swizzle_mode_na_to_pa(struct amdgpu_device *adev,
uint32_t channel_idx,
uint64_t na, uint64_t *soc_pa)
{
uint32_t channel_num = UMC_V8_10_TOTAL_CHANNEL_NUM(adev);
uint32_t col_bit = umc_v8_10_get_col_bit(channel_num);
uint64_t tmp_addr;
if (col_bit == U32_MAX)
return -1;
tmp_addr = SWIZZLE_MODE_TMP_ADDR(na, channel_num, channel_idx);
*soc_pa = SWIZZLE_MODE_ADDR_HI(tmp_addr, col_bit) |
SWIZZLE_MODE_ADDR_MID(na, col_bit) |
SWIZZLE_MODE_ADDR_LOW(tmp_addr, col_bit) |
SWIZZLE_MODE_ADDR_LSB(na);
return 0;
}
static void umc_v8_10_query_error_address(struct amdgpu_device *adev,
struct ras_err_data *err_data,
uint32_t umc_reg_offset,
uint32_t node_inst,
uint32_t ch_inst,
uint32_t umc_inst)
{
uint64_t mc_umc_status_addr;
uint64_t mc_umc_status, err_addr;
uint32_t channel_index;
mc_umc_status_addr =
SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
mc_umc_status = RREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4);
if (mc_umc_status == 0)
return;
if (!err_data->err_addr) {
/* clear umc status */
WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
return;
}
channel_index =
adev->umc.channel_idx_tbl[node_inst * adev->umc.umc_inst_num *
adev->umc.channel_inst_num +
umc_inst * adev->umc.channel_inst_num +
ch_inst];
/* calculate error address if ue/ce error is detected */
if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, AddrV) == 1 &&
(REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)) {
uint32_t addr_lsb;
uint64_t mc_umc_addrt0;
mc_umc_addrt0 = SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_ADDRT0);
err_addr = RREG64_PCIE((mc_umc_addrt0 + umc_reg_offset) * 4);
err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
/* the lowest lsb bits should be ignored */
addr_lsb = REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, AddrLsb);
err_addr &= ~((0x1ULL << addr_lsb) - 1);
/* we only save ue error information currently, ce is skipped */
if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1) {
uint64_t na_err_addr_base = err_addr & ~(0x3ULL << UMC_V8_10_NA_C5_BIT);
uint64_t na_err_addr, retired_page_addr;
uint32_t col = 0;
int ret = 0;
/* loop for all possibilities of [C6 C5] in normal address. */
for (col = 0; col < UMC_V8_10_NA_COL_2BITS_POWER_OF_2_NUM; col++) {
na_err_addr = na_err_addr_base | (col << UMC_V8_10_NA_C5_BIT);
/* Mapping normal error address to retired soc physical address. */
ret = umc_v8_10_swizzle_mode_na_to_pa(adev, channel_index,
na_err_addr, &retired_page_addr);
if (ret) {
dev_err(adev->dev, "Failed to map pa from umc na.\n");
break;
}
dev_info(adev->dev, "Error Address(PA): 0x%llx\n",
retired_page_addr);
amdgpu_umc_fill_error_record(err_data, na_err_addr,
retired_page_addr, channel_index, umc_inst);
}
}
}
/* clear umc status */
WREG64_PCIE((mc_umc_status_addr + umc_reg_offset) * 4, 0x0ULL);
}
static void umc_v8_10_query_ras_error_address(struct amdgpu_device *adev,
void *ras_error_status)
{
struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
uint32_t node_inst = 0;
uint32_t umc_inst = 0;
uint32_t ch_inst = 0;
uint32_t umc_reg_offset = 0;
LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
umc_reg_offset = get_umc_v8_10_reg_offset(adev,
node_inst,
umc_inst,
ch_inst);
umc_v8_10_query_error_address(adev,
err_data,
umc_reg_offset,
node_inst,
ch_inst,
umc_inst);
}
}
static void umc_v8_10_err_cnt_init_per_channel(struct amdgpu_device *adev,
uint32_t umc_reg_offset)
{
uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
uint32_t ecc_err_cnt_addr;
ecc_err_cnt_sel_addr =
SOC15_REG_OFFSET(UMC, 0, regUMCCH0_0_GeccErrCntSel);
ecc_err_cnt_addr =
SOC15_REG_OFFSET(UMC, 0, regUMCCH0_0_GeccErrCnt);
ecc_err_cnt_sel = RREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4);
/* set ce error interrupt type to APIC based interrupt */
ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_GeccErrCntSel,
GeccErrInt, 0x1);
WREG32_PCIE((ecc_err_cnt_sel_addr + umc_reg_offset) * 4, ecc_err_cnt_sel);
/* set error count to initial value */
WREG32_PCIE((ecc_err_cnt_addr + umc_reg_offset) * 4, UMC_V8_10_CE_CNT_INIT);
}
static void umc_v8_10_err_cnt_init(struct amdgpu_device *adev)
{
uint32_t node_inst = 0;
uint32_t umc_inst = 0;
uint32_t ch_inst = 0;
uint32_t umc_reg_offset = 0;
LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
umc_reg_offset = get_umc_v8_10_reg_offset(adev,
node_inst,
umc_inst,
ch_inst);
umc_v8_10_err_cnt_init_per_channel(adev, umc_reg_offset);
}
}
const struct amdgpu_ras_block_hw_ops umc_v8_10_ras_hw_ops = {
.query_ras_error_count = umc_v8_10_query_ras_error_count,
.query_ras_error_address = umc_v8_10_query_ras_error_address,
};
struct amdgpu_umc_ras umc_v8_10_ras = {
.ras_block = {
.hw_ops = &umc_v8_10_ras_hw_ops,
},
.err_cnt_init = umc_v8_10_err_cnt_init,
};
/*
* Copyright 2022 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 __UMC_V8_10_H__
#define __UMC_V8_10_H__
#include "soc15_common.h"
#include "amdgpu.h"
/* number of umc channel instance with memory map register access */
#define UMC_V8_10_CHANNEL_INSTANCE_NUM 2
/* number of umc instance with memory map register access */
#define UMC_V8_10_UMC_INSTANCE_NUM 2
/* Total channel instances for all umc nodes */
#define UMC_V8_10_TOTAL_CHANNEL_NUM(adev) \
(UMC_V8_10_CHANNEL_INSTANCE_NUM * UMC_V8_10_UMC_INSTANCE_NUM * (adev)->umc.node_inst_num)
/* UMC regiser per channel offset */
#define UMC_V8_10_PER_CHANNEL_OFFSET 0x400
/* EccErrCnt max value */
#define UMC_V8_10_CE_CNT_MAX 0xffff
/* umc ce interrupt threshold */
#define UUMC_V8_10_CE_INT_THRESHOLD 0xffff
/* umc ce count initial value */
#define UMC_V8_10_CE_CNT_INIT (UMC_V8_10_CE_CNT_MAX - UUMC_V8_10_CE_INT_THRESHOLD)
#define UMC_V8_10_NA_COL_2BITS_POWER_OF_2_NUM 4
/* The C5 bit in NA address */
#define UMC_V8_10_NA_C5_BIT 14
/* Map to swizzle mode address */
#define SWIZZLE_MODE_TMP_ADDR(na, ch_num, ch_idx) \
((((na) >> 10) * (ch_num) + (ch_idx)) << 10)
#define SWIZZLE_MODE_ADDR_HI(addr, col_bit) \
(((addr) >> ((col_bit) + 2)) << ((col_bit) + 2))
#define SWIZZLE_MODE_ADDR_MID(na, col_bit) ((((na) >> 8) & 0x3) << (col_bit))
#define SWIZZLE_MODE_ADDR_LOW(addr, col_bit) \
((((addr) >> 10) & ((0x1ULL << (col_bit - 8)) - 1)) << 8)
#define SWIZZLE_MODE_ADDR_LSB(na) ((na) & 0xFF)
extern struct amdgpu_umc_ras umc_v8_10_ras;
extern const uint32_t
umc_v8_10_channel_idx_tbl[]
[UMC_V8_10_UMC_INSTANCE_NUM]
[UMC_V8_10_CHANNEL_INSTANCE_NUM];
#endif
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment